Home | History | Annotate | Line # | Download | only in libperfuse
ops.c revision 1.50.2.10
      1 /*  $NetBSD: ops.c,v 1.50.2.10 2014/11/03 19:37:58 msaitoh Exp $ */
      2 
      3 /*-
      4  *  Copyright (c) 2010-2011 Emmanuel Dreyfus. All rights reserved.
      5  *
      6  *  Redistribution and use in source and binary forms, with or without
      7  *  modification, are permitted provided that the following conditions
      8  *  are met:
      9  *  1. Redistributions of source code must retain the above copyright
     10  *     notice, this list of conditions and the following disclaimer.
     11  *  2. Redistributions in binary form must reproduce the above copyright
     12  *     notice, this list of conditions and the following disclaimer in the
     13  *     documentation and/or other materials provided with the distribution.
     14  *
     15  *  THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     16  *  ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     17  *  TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     18  *  PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     19  *  BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     20  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     21  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     22  *  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     23  *  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     24  *  ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     25  *  POSSIBILITY OF SUCH DAMAGE.
     26  */
     27 
     28 #include <stdio.h>
     29 #include <unistd.h>
     30 #include <stdlib.h>
     31 #include <libgen.h>
     32 #include <errno.h>
     33 #include <err.h>
     34 #include <sysexits.h>
     35 #include <syslog.h>
     36 #include <puffs.h>
     37 #include <sys/socket.h>
     38 #include <sys/socket.h>
     39 #include <sys/extattr.h>
     40 #include <sys/time.h>
     41 #include <machine/vmparam.h>
     42 
     43 #include "perfuse_priv.h"
     44 #include "fuse.h"
     45 
     46 extern int perfuse_diagflags;
     47 
     48 #if 0
     49 static void print_node(const char *, puffs_cookie_t);
     50 #endif
     51 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
     52 static void perfuse_newinfo_setttl(struct puffs_newinfo *,
     53     struct puffs_node *, struct fuse_entry_out *, struct fuse_attr_out *);
     54 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
     55 static int xchg_msg(struct puffs_usermount *, puffs_cookie_t,
     56     perfuse_msg_t *, size_t, enum perfuse_xchg_pb_reply);
     57 static int mode_access(puffs_cookie_t, const struct puffs_cred *, mode_t);
     58 static int sticky_access(puffs_cookie_t, struct puffs_node *,
     59     const struct puffs_cred *);
     60 static void fuse_attr_to_vap(struct perfuse_state *,
     61     struct vattr *, struct fuse_attr *);
     62 static int node_lookup_common(struct puffs_usermount *, puffs_cookie_t,
     63     struct puffs_newinfo *, const char *, const struct puffs_cred *,
     64     struct puffs_node **);
     65 static int node_mk_common(struct puffs_usermount *, puffs_cookie_t,
     66     struct puffs_newinfo *, const struct puffs_cn *pcn, perfuse_msg_t *);
     67 static uint64_t readdir_last_cookie(struct fuse_dirent *, size_t);
     68 static ssize_t fuse_to_dirent(struct puffs_usermount *, puffs_cookie_t,
     69     struct fuse_dirent *, size_t);
     70 static void readdir_buffered(puffs_cookie_t, struct dirent *, off_t *,
     71     size_t *);
     72 static void node_ref(puffs_cookie_t);
     73 static void node_rele(puffs_cookie_t);
     74 static void requeue_request(struct puffs_usermount *,
     75     puffs_cookie_t opc, enum perfuse_qtype);
     76 static int dequeue_requests(puffs_cookie_t opc, enum perfuse_qtype, int);
     77 #define DEQUEUE_ALL 0
     78 
     79 /*
     80  *  From <sys/vnode>, inside #ifdef _KERNEL section
     81  */
     82 #define IO_SYNC		(0x40|IO_DSYNC)
     83 #define IO_DSYNC	0x00200
     84 #define IO_DIRECT	0x02000
     85 
     86 /*
     87  *  From <fcntl>, inside #ifdef _KERNEL section
     88  */
     89 #define F_WAIT		0x010
     90 #define F_FLOCK		0x020
     91 #define OFLAGS(fflags)  ((fflags) - 1)
     92 
     93 /*
     94  * Borrowed from src/sys/kern/vfs_subr.c and src/sys/sys/vnode.h
     95  */
     96 const enum vtype iftovt_tab[16] = {
     97 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
     98         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
     99 };
    100 const int vttoif_tab[9] = {
    101 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
    102         S_IFSOCK, S_IFIFO, S_IFMT,
    103 };
    104 
    105 #define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12])
    106 #define VTTOIF(indx) (vttoif_tab[(int)(indx)])
    107 
    108 #if 0
    109 static void
    110 print_node(const char *func, puffs_cookie_t opc)
    111 {
    112 	struct puffs_node *pn;
    113 	struct perfuse_node_data *pnd;
    114 	struct vattr *vap;
    115 
    116 	pn = (struct puffs_node *)opc;
    117 	pnd = PERFUSE_NODE_DATA(opc);
    118 	vap = &pn->pn_va;
    119 
    120 	printf("%s: \"%s\", opc = %p, nodeid = 0x%"PRIx64" ino = %"PRIu64"\n",
    121 	       func, pnd->pnd_name, opc, pnd->pnd_nodeid, vap->va_fileid);
    122 
    123 	return;
    124 }
    125 #endif /* PERFUSE_DEBUG */
    126 
    127 int
    128 perfuse_node_close_common(struct puffs_usermount *pu, puffs_cookie_t opc,
    129 	int mode)
    130 {
    131 	struct perfuse_state *ps;
    132 	perfuse_msg_t *pm;
    133 	int op;
    134 	uint64_t fh;
    135 	struct fuse_release_in *fri;
    136 	struct perfuse_node_data *pnd;
    137 	struct puffs_node *pn;
    138 	int error;
    139 
    140 	ps = puffs_getspecific(pu);
    141 	pn = (struct puffs_node *)opc;
    142 	pnd = PERFUSE_NODE_DATA(pn);
    143 
    144 	if (puffs_pn_getvap(pn)->va_type == VDIR) {
    145 		op = FUSE_RELEASEDIR;
    146 		mode = FREAD;
    147 	} else {
    148 		op = FUSE_RELEASE;
    149 	}
    150 
    151 	/*
    152 	 * Destroy the filehandle before sending the
    153 	 * request to the FUSE filesystem, otherwise
    154 	 * we may get a second close() while we wait
    155 	 * for the reply, and we would end up closing
    156 	 * the same fh twice instead of closng both.
    157 	 */
    158 	fh = perfuse_get_fh(opc, mode);
    159 	perfuse_destroy_fh(pn, fh);
    160 
    161 	/*
    162 	 * release_flags may be set to FUSE_RELEASE_FLUSH
    163 	 * to flush locks. lock_owner must be set in that case
    164 	 *
    165 	 * ps_new_msg() is called with NULL creds, which will
    166 	 * be interpreted as FUSE superuser. We come here from the
    167 	 * inactive method, which provides no creds, but obviously
    168 	 * runs with kernel privilege.
    169 	 */
    170 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*fri), NULL);
    171 	fri = GET_INPAYLOAD(ps, pm, fuse_release_in);
    172 	fri->fh = fh;
    173 	fri->flags = 0;
    174 	fri->release_flags = 0;
    175 	fri->lock_owner = pnd->pnd_lock_owner;
    176 	fri->flags = (fri->lock_owner != 0) ? FUSE_RELEASE_FLUSH : 0;
    177 
    178 #ifdef PERFUSE_DEBUG
    179 	if (perfuse_diagflags & PDF_FH)
    180 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
    181 			 __func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
    182 #endif
    183 
    184 	if ((error = xchg_msg(pu, opc, pm,
    185 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
    186 		DERRX(EX_SOFTWARE, "%s: freed fh = 0x%"PRIx64" but filesystem "
    187 		      "returned error = %d", __func__, fh, error);
    188 
    189 	ps->ps_destroy_msg(pm);
    190 
    191 	return 0;
    192 }
    193 
    194 static int
    195 xchg_msg(struct puffs_usermount *pu, puffs_cookie_t opc, perfuse_msg_t *pm,
    196 	size_t len, enum perfuse_xchg_pb_reply wait)
    197 {
    198 	struct perfuse_state *ps;
    199 	struct perfuse_node_data *pnd;
    200 	struct perfuse_trace *pt = NULL;
    201 	int error;
    202 
    203 	ps = puffs_getspecific(pu);
    204 	pnd = NULL;
    205 	if ((struct puffs_node *)opc != NULL)
    206 		pnd = PERFUSE_NODE_DATA(opc);
    207 
    208 #ifdef PERFUSE_DEBUG
    209 	if ((perfuse_diagflags & PDF_FILENAME) && (opc != 0))
    210 		DPRINTF("file = \"%s\", ino = %"PRIu64" flags = 0x%x\n",
    211 			perfuse_node_path(ps, opc),
    212 			((struct puffs_node *)opc)->pn_va.va_fileid,
    213 			PERFUSE_NODE_DATA(opc)->pnd_flags);
    214 #endif
    215 	ps->ps_xchgcount++;
    216 	if (pnd)
    217 		pnd->pnd_inxchg++;
    218 
    219 	/*
    220 	 * Record FUSE call start if requested
    221 	 */
    222 	if (perfuse_diagflags & PDF_TRACE)
    223 		pt = perfuse_trace_begin(ps, opc, pm);
    224 
    225 	/*
    226 	 * Do actual FUSE exchange
    227 	 */
    228 	if ((error = ps->ps_xchg_msg(pu, pm, len, wait)) != 0)
    229 		ps->ps_destroy_msg(pm);
    230 
    231 	/*
    232 	 * Record FUSE call end if requested
    233 	 */
    234 	if (pt != NULL)
    235 		perfuse_trace_end(ps, pt, error);
    236 
    237 	ps->ps_xchgcount--;
    238 	if (pnd) {
    239 		pnd->pnd_inxchg--;
    240 		(void)dequeue_requests(opc, PCQ_AFTERXCHG, DEQUEUE_ALL);
    241 	}
    242 
    243 	return error;
    244 }
    245 
    246 static int
    247 mode_access(puffs_cookie_t opc, const struct puffs_cred *pcr, mode_t mode)
    248 {
    249 	struct puffs_node *pn;
    250 	struct vattr *va;
    251 
    252 	/*
    253 	 * pcr is NULL for self open through fsync or readdir.
    254 	 * In both case, access control is useless, as it was
    255 	 * done before, at open time.
    256 	 */
    257 	if (pcr == NULL)
    258 		return 0;
    259 
    260 	pn = (struct puffs_node *)opc;
    261 	va = puffs_pn_getvap(pn);
    262 	return puffs_access(va->va_type, va->va_mode,
    263 			    va->va_uid, va->va_gid,
    264 			    mode, pcr);
    265 }
    266 
    267 static int
    268 sticky_access(puffs_cookie_t opc, struct puffs_node *targ,
    269 	      const struct puffs_cred *pcr)
    270 {
    271 	uid_t uid;
    272 	int sticky, owner, parent_owner;
    273 
    274 	/*
    275 	 * This covers the case where the kernel requests a DELETE
    276 	 * or RENAME on its own, and where puffs_cred_getuid would
    277 	 * return -1. While such a situation should not happen,
    278 	 * we allow it here.
    279 	 *
    280 	 * This also allows root to tamper with other users' files
    281 	 * that have the sticky bit.
    282 	 */
    283 	if (puffs_cred_isjuggernaut(pcr))
    284 		return 0;
    285 
    286 	if (puffs_cred_getuid(pcr, &uid) != 0)
    287 		DERRX(EX_SOFTWARE, "puffs_cred_getuid fails in %s", __func__);
    288 
    289 	sticky = puffs_pn_getvap(opc)->va_mode & S_ISTXT;
    290 	owner = puffs_pn_getvap(targ)->va_uid == uid;
    291 	parent_owner = puffs_pn_getvap(opc)->va_uid == uid;
    292 
    293 	if (sticky && !owner && !parent_owner)
    294 		return EPERM;
    295 
    296 	return 0;
    297 }
    298 
    299 
    300 static void
    301 fuse_attr_to_vap(struct perfuse_state *ps, struct vattr *vap,
    302 	struct fuse_attr *fa)
    303 {
    304 	vap->va_type = IFTOVT(fa->mode);
    305 	vap->va_mode = fa->mode & ALLPERMS;
    306 	vap->va_nlink = fa->nlink;
    307 	vap->va_uid = fa->uid;
    308 	vap->va_gid = fa->gid;
    309 	vap->va_fsid = (long)ps->ps_fsid;
    310 	vap->va_fileid = fa->ino;
    311 	vap->va_size = fa->size;
    312 	vap->va_blocksize = fa->blksize;
    313 	vap->va_atime.tv_sec = (time_t)fa->atime;
    314 	vap->va_atime.tv_nsec = (long) fa->atimensec;
    315 	vap->va_mtime.tv_sec = (time_t)fa->mtime;
    316 	vap->va_mtime.tv_nsec = (long)fa->mtimensec;
    317 	vap->va_ctime.tv_sec = (time_t)fa->ctime;
    318 	vap->va_ctime.tv_nsec = (long)fa->ctimensec;
    319 	vap->va_birthtime.tv_sec = 0;
    320 	vap->va_birthtime.tv_nsec = 0;
    321 	vap->va_gen = 0;
    322 	vap->va_flags = 0;
    323 	vap->va_rdev = fa->rdev;
    324 	vap->va_bytes = fa->size;
    325 	vap->va_filerev = (u_quad_t)PUFFS_VNOVAL;
    326 	vap->va_vaflags = 0;
    327 
    328 	if (vap->va_blocksize == 0)
    329 		vap->va_blocksize = DEV_BSIZE;
    330 
    331 	if (vap->va_size == (size_t)PUFFS_VNOVAL) /* XXX */
    332 		vap->va_size = 0;
    333 
    334 	return;
    335 }
    336 
    337 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
    338 static void
    339 perfuse_newinfo_setttl(struct puffs_newinfo *pni,
    340     struct puffs_node *pn, struct fuse_entry_out *feo,
    341     struct fuse_attr_out *fao)
    342 {
    343 #ifdef PERFUSE_DEBUG
    344 	if ((feo == NULL) && (fao == NULL))
    345 		DERRX(EX_SOFTWARE, "%s: feo and fao NULL", __func__);
    346 
    347 	if ((feo != NULL) && (fao != NULL))
    348 		DERRX(EX_SOFTWARE, "%s: feo and fao != NULL", __func__);
    349 #endif /* PERFUSE_DEBUG */
    350 
    351 	if (fao != NULL) {
    352 		struct timespec va_ttl;
    353 
    354 		va_ttl.tv_sec = fao->attr_valid;
    355 		va_ttl.tv_nsec = fao->attr_valid_nsec;
    356 
    357 		puffs_newinfo_setvattl(pni, &va_ttl);
    358 	}
    359 
    360 	if (feo != NULL) {
    361 		struct timespec va_ttl;
    362 		struct timespec cn_ttl;
    363 		struct timespec now;
    364 		struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(pn);
    365 
    366 		va_ttl.tv_sec = feo->attr_valid;
    367 		va_ttl.tv_nsec = feo->attr_valid_nsec;
    368 		cn_ttl.tv_sec = feo->entry_valid;
    369 		cn_ttl.tv_nsec = feo->entry_valid_nsec;
    370 
    371 		puffs_newinfo_setvattl(pni, &va_ttl);
    372 		puffs_newinfo_setcnttl(pni, &cn_ttl);
    373 
    374 		if (clock_gettime(CLOCK_REALTIME, &now) != 0)
    375 			DERR(EX_OSERR, "clock_gettime failed");
    376 
    377                 timespecadd(&now, &cn_ttl, &pnd->pnd_cn_expire);
    378 	}
    379 
    380 	return;
    381 }
    382 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
    383 
    384 static int
    385 node_lookup_common(struct puffs_usermount *pu, puffs_cookie_t opc,
    386 	struct puffs_newinfo *pni, const char *path,
    387 	const struct puffs_cred *pcr, struct puffs_node **pnp)
    388 {
    389 	struct perfuse_state *ps;
    390 	struct perfuse_node_data *oldpnd;
    391 	perfuse_msg_t *pm;
    392 	struct fuse_entry_out *feo;
    393 	struct puffs_node *pn;
    394 	size_t len;
    395 	int error;
    396 
    397 	/*
    398 	 * Prevent further lookups if the parent was removed
    399 	 */
    400 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
    401 		return ESTALE;
    402 
    403 	if (pnp == NULL)
    404 		DERRX(EX_SOFTWARE, "pnp must be != NULL");
    405 
    406 	ps = puffs_getspecific(pu);
    407 
    408 #ifdef PERFUSE_DEBUG
    409 	if (perfuse_diagflags & PDF_FILENAME)
    410 		DPRINTF("%s: opc = %p, file = \"%s\" looking up \"%s\"\n",
    411 			__func__, (void *)opc,
    412 			perfuse_node_path(ps, opc), path);
    413 
    414 	if (strcmp(path, ".") == 0)
    415 		DERRX(EX_SOFTWARE, "unexpected dot-lookup");
    416 
    417 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_RECLAIMED)
    418 		DERRX(EX_SOFTWARE,
    419 		      "looking up reclaimed node opc = %p, name = \"%s\"",
    420 		      opc, path);
    421 
    422 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_INVALID)
    423 		DERRX(EX_SOFTWARE,
    424 		      "looking up freed node opc = %p, name = \"%s\"",
    425 		      opc, path);
    426 #endif /* PERFUSE_DEBUG */
    427 
    428 	len = strlen(path) + 1;
    429 	pm = ps->ps_new_msg(pu, opc, FUSE_LOOKUP, len, pcr);
    430 	(void)strlcpy(_GET_INPAYLOAD(ps, pm, char *), path, len);
    431 
    432 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
    433 		return error;
    434 
    435 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
    436 
    437 	/*
    438 	 * Check for a known node, not reclaimed, with another name.
    439 	 * It may have been moved, or we can lookup ../
    440 	 */
    441 	if (((oldpnd = perfuse_node_bynodeid(ps, feo->nodeid)) != NULL) &&
    442 	    !(oldpnd->pnd_flags & PND_RECLAIMED)) {
    443 		/*
    444 		 * Save the new node name if not ..
    445 		 */
    446 		if (strncmp(path, "..", len) != 0)
    447 			(void)strlcpy(oldpnd->pnd_name,
    448 				      path, MAXPATHLEN);
    449 		pn = oldpnd->pnd_pn;
    450 
    451 	} else {
    452 		pn = perfuse_new_pn(pu, path, opc);
    453 		PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
    454 		perfuse_node_cache(ps, pn);
    455 	}
    456 
    457 #ifdef PERFUSE_DEBUG
    458 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_RECLAIMED)
    459 		DERRX(EX_SOFTWARE,
    460 		      "reclaimed in lookup opc = %p, name = \"%s\", ck = %p",
    461 		      opc, path, pn);
    462 
    463 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_INVALID)
    464 		DERRX(EX_SOFTWARE,
    465 		      "freed in lookup opc = %p, name = \"%s\", ck = %p",
    466 		      opc, path, pn);
    467 #endif /* PERFUSE_DEBUG */
    468 
    469 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
    470 	pn->pn_va.va_gen = (u_long)(feo->generation);
    471 	PERFUSE_NODE_DATA(pn)->pnd_fuse_nlookup++;
    472 
    473 	*pnp = pn;
    474 
    475 #ifdef PERFUSE_DEBUG
    476 	if (perfuse_diagflags & PDF_FILENAME)
    477 		DPRINTF("%s: opc = %p, looked up opc = %p, "
    478 			"nodeid = 0x%"PRIx64" file = \"%s\"\n", __func__,
    479 			(void *)opc, pn, feo->nodeid, path);
    480 #endif
    481 
    482 	if (pni != NULL) {
    483 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
    484 		puffs_newinfo_setva(pni, &pn->pn_va);
    485 		perfuse_newinfo_setttl(pni, pn, feo, NULL);
    486 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
    487 		puffs_newinfo_setcookie(pni, pn);
    488 		puffs_newinfo_setvtype(pni, pn->pn_va.va_type);
    489 		puffs_newinfo_setsize(pni, (voff_t)pn->pn_va.va_size);
    490 		puffs_newinfo_setrdev(pni, pn->pn_va.va_rdev);
    491 	}
    492 
    493 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_NODELEAK) {
    494 		PERFUSE_NODE_DATA(pn)->pnd_flags &= ~PND_NODELEAK;
    495 		ps->ps_nodeleakcount--;
    496 	}
    497 
    498 	ps->ps_destroy_msg(pm);
    499 
    500 	return 0;
    501 }
    502 
    503 
    504 /*
    505  * Common code for methods that create objects:
    506  * perfuse_node_mkdir
    507  * perfuse_node_mknod
    508  * perfuse_node_symlink
    509  */
    510 static int
    511 node_mk_common(struct puffs_usermount *pu, puffs_cookie_t opc,
    512 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
    513 	perfuse_msg_t *pm)
    514 {
    515 	struct perfuse_state *ps;
    516 	struct puffs_node *pn;
    517 	struct fuse_entry_out *feo;
    518 	int error;
    519 
    520 	ps =  puffs_getspecific(pu);
    521 
    522 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
    523 		return error;
    524 
    525 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
    526 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
    527 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
    528 
    529 	pn = perfuse_new_pn(pu, pcn->pcn_name, opc);
    530 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
    531 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
    532 	perfuse_node_cache(ps, pn);
    533 
    534 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
    535 	pn->pn_va.va_gen = (u_long)(feo->generation);
    536 
    537 	puffs_newinfo_setcookie(pni, pn);
    538 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
    539 	puffs_newinfo_setva(pni, &pn->pn_va);
    540 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
    541 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
    542 
    543 
    544 #ifdef PERFUSE_DEBUG
    545 	if (perfuse_diagflags & PDF_FILENAME)
    546 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
    547 			"nodeid = 0x%"PRIx64"\n",
    548 			__func__, (void *)pn, pcn->pcn_name,
    549 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid);
    550 #endif
    551 	ps->ps_destroy_msg(pm);
    552 
    553 	/* Parents is now dirty */
    554 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
    555 
    556 	return 0;
    557 }
    558 
    559 static uint64_t
    560 readdir_last_cookie(struct fuse_dirent *fd, size_t fd_len)
    561 {
    562 	size_t len;
    563 	size_t seen = 0;
    564 	char *ndp;
    565 
    566 	do {
    567 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
    568 		seen += len;
    569 
    570 		if (seen >= fd_len)
    571 			break;
    572 
    573 		ndp = (char *)(void *)fd + (size_t)len;
    574 		fd = (struct fuse_dirent *)(void *)ndp;
    575 	} while (1 /* CONSTCOND */);
    576 
    577 	return fd->off;
    578 }
    579 
    580 static ssize_t
    581 fuse_to_dirent(struct puffs_usermount *pu, puffs_cookie_t opc,
    582 	struct fuse_dirent *fd, size_t fd_len)
    583 {
    584 	struct dirent *dents;
    585 	size_t dents_len;
    586 	ssize_t written;
    587 	uint64_t fd_offset;
    588 	struct fuse_dirent *fd_base;
    589 	size_t len;
    590 
    591 	fd_base = fd;
    592 	fd_offset = 0;
    593 	written = 0;
    594 	dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
    595 	dents_len = (size_t)PERFUSE_NODE_DATA(opc)->pnd_dirent_len;
    596 
    597 	do {
    598 		char *ndp;
    599 		size_t reclen;
    600 
    601 		reclen = _DIRENT_RECLEN(dents, fd->namelen);
    602 
    603 		/*
    604 		 * Check we do not overflow the output buffer
    605 		 * struct fuse_dirent is bigger than struct dirent,
    606 		 * so we should always use fd_len and never reallocate
    607 		 * later.
    608 		 * If we have to reallocate,try to double the buffer
    609 		 * each time so that we do not have to do it too often.
    610 		 */
    611 		if (written + reclen > dents_len) {
    612 			if (dents_len == 0)
    613 				dents_len = fd_len;
    614 			else
    615 				dents_len =
    616 				   MAX(2 * dents_len, written + reclen);
    617 
    618 			dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
    619 			if ((dents = realloc(dents, dents_len)) == NULL)
    620 				DERR(EX_OSERR, "%s: malloc failed", __func__);
    621 
    622 			PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
    623 			PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
    624 
    625 			/*
    626 			 * (void *) for delint
    627 			 */
    628 			ndp = (char *)(void *)dents + written;
    629 			dents = (struct dirent *)(void *)ndp;
    630 		}
    631 
    632 		/*
    633 		 * Filesystem was mounted without -o use_ino
    634 		 * Perform a lookup to find it.
    635 		 */
    636 		if (fd->ino == PERFUSE_UNKNOWN_INO) {
    637 			struct puffs_node *pn;
    638 			struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
    639 
    640 			/*
    641 			 * Avoid breaking out of fs
    642 			 * by lookup to .. on root
    643 			 */
    644 			if ((strcmp(fd->name, "..") == 0) &&
    645 			    (pnd->pnd_nodeid == FUSE_ROOT_ID)) {
    646 				fd->ino = FUSE_ROOT_ID;
    647 			} else {
    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 
    658 		dents->d_fileno = fd->ino;
    659 		dents->d_reclen = (unsigned short)reclen;
    660 		dents->d_namlen = fd->namelen;
    661 		dents->d_type = fd->type;
    662 		strlcpy(dents->d_name, fd->name, fd->namelen + 1);
    663 
    664 #ifdef PERFUSE_DEBUG
    665 		if (perfuse_diagflags & PDF_READDIR)
    666 			DPRINTF("%s: translated \"%s\" ino = %"PRIu64"\n",
    667 				__func__, dents->d_name, dents->d_fileno);
    668 #endif
    669 
    670 		dents = _DIRENT_NEXT(dents);
    671 		written += reclen;
    672 
    673 		/*
    674 		 * Move to the next record.
    675 		 * fd->off is not the offset, it is an opaque cookie
    676 		 * given by the filesystem to keep state across multiple
    677 		 * readdir() operation.
    678 		 * Use record alignement instead.
    679 		 */
    680 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
    681 #ifdef PERFUSE_DEBUG
    682 		if (perfuse_diagflags & PDF_READDIR)
    683 			DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
    684 				"length = %zd/0x%zx. "
    685 				"next record at %"PRId64"/0x%"PRIx64" "
    686 				"max %zd/0x%zx\n",
    687 				__func__, fd_offset, fd_offset, len, len,
    688 				fd_offset + len, fd_offset + len,
    689 				fd_len, fd_len);
    690 #endif
    691 		fd_offset += len;
    692 
    693 		/*
    694 		 * Check if next record is still within the packet
    695 		 * If it is not, we reached the end of the buffer.
    696 		 */
    697 		if (fd_offset >= fd_len)
    698 			break;
    699 
    700 		/*
    701 		 * (void *) for delint
    702 		 */
    703 		ndp = (char *)(void *)fd_base + (size_t)fd_offset;
    704 		fd = (struct fuse_dirent *)(void *)ndp;
    705 
    706 	} while (1 /* CONSTCOND */);
    707 
    708 	/*
    709 	 * Adjust the dirent output length
    710 	 */
    711 	if (written != -1)
    712 		PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
    713 
    714 	return written;
    715 }
    716 
    717 static void
    718 readdir_buffered(puffs_cookie_t opc, struct dirent *dent, off_t *readoff,
    719 	size_t *reslen)
    720 {
    721 	struct dirent *fromdent;
    722 	struct perfuse_node_data *pnd;
    723 	char *ndp;
    724 
    725 	pnd = PERFUSE_NODE_DATA(opc);
    726 
    727 	while (*readoff < pnd->pnd_dirent_len) {
    728 		/*
    729 		 * (void *) for delint
    730 		 */
    731 		ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
    732 		fromdent = (struct dirent *)(void *)ndp;
    733 
    734 		if (*reslen < _DIRENT_SIZE(fromdent))
    735 			break;
    736 
    737 		memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
    738 		*readoff += _DIRENT_SIZE(fromdent);
    739 		*reslen -= _DIRENT_SIZE(fromdent);
    740 
    741 		dent = _DIRENT_NEXT(dent);
    742 	}
    743 
    744 #ifdef PERFUSE_DEBUG
    745 	if (perfuse_diagflags & PDF_READDIR)
    746 		DPRINTF("%s: readoff = %"PRId64",  "
    747 			"pnd->pnd_dirent_len = %"PRId64"\n",
    748 			__func__, *readoff, pnd->pnd_dirent_len);
    749 #endif
    750 	if (*readoff >=  pnd->pnd_dirent_len) {
    751 		free(pnd->pnd_dirent);
    752 		pnd->pnd_dirent = NULL;
    753 		pnd->pnd_dirent_len = 0;
    754 	}
    755 
    756 	return;
    757 }
    758 
    759 
    760 static void
    761 node_ref(puffs_cookie_t opc)
    762 {
    763 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
    764 
    765 #ifdef PERFUSE_DEBUG
    766 	if (pnd->pnd_flags & PND_INVALID)
    767 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
    768 #endif /* PERFUSE_DEBUG */
    769 
    770 	pnd->pnd_ref++;
    771 	return;
    772 }
    773 
    774 static void
    775 node_rele(puffs_cookie_t opc)
    776 {
    777 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
    778 
    779 #ifdef PERFUSE_DEBUG
    780 	if (pnd->pnd_flags & PND_INVALID)
    781 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
    782 #endif /* PERFUSE_DEBUG */
    783 
    784 	pnd->pnd_ref--;
    785 
    786 	if (pnd->pnd_ref == 0)
    787 		(void)dequeue_requests(opc, PCQ_REF, DEQUEUE_ALL);
    788 
    789 	return;
    790 }
    791 
    792 static void
    793 requeue_request(struct puffs_usermount *pu, puffs_cookie_t opc,
    794 	enum perfuse_qtype type)
    795 {
    796 	struct perfuse_cc_queue pcq;
    797 	struct perfuse_node_data *pnd;
    798 #ifdef PERFUSE_DEBUG
    799 	struct perfuse_state *ps;
    800 
    801 	ps = perfuse_getspecific(pu);
    802 #endif
    803 
    804 	pnd = PERFUSE_NODE_DATA(opc);
    805 	pcq.pcq_type = type;
    806 	pcq.pcq_cc = puffs_cc_getcc(pu);
    807 	TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
    808 
    809 #ifdef PERFUSE_DEBUG
    810 	if (perfuse_diagflags & PDF_REQUEUE)
    811 		DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
    812 		        __func__, (void *)opc, pcq.pcq_cc,
    813 			perfuse_qtypestr[type]);
    814 #endif
    815 
    816 	puffs_cc_yield(pcq.pcq_cc);
    817 	TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
    818 
    819 #ifdef PERFUSE_DEBUG
    820 	if (perfuse_diagflags & PDF_REQUEUE)
    821 		DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
    822 		        __func__, (void *)opc, pcq.pcq_cc,
    823 			perfuse_qtypestr[type]);
    824 #endif
    825 
    826 	return;
    827 }
    828 
    829 static int
    830 dequeue_requests(puffs_cookie_t opc, enum perfuse_qtype type, int max)
    831 {
    832 	struct perfuse_cc_queue *pcq;
    833 	struct perfuse_node_data *pnd;
    834 	int dequeued;
    835 
    836 	pnd = PERFUSE_NODE_DATA(opc);
    837 	dequeued = 0;
    838 	TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
    839 		if (pcq->pcq_type != type)
    840 			continue;
    841 
    842 #ifdef PERFUSE_DEBUG
    843 		if (perfuse_diagflags & PDF_REQUEUE)
    844 			DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
    845 				__func__, (void *)opc, pcq->pcq_cc,
    846 				 perfuse_qtypestr[type]);
    847 #endif
    848 		puffs_cc_schedule(pcq->pcq_cc);
    849 
    850 		if (++dequeued == max)
    851 			break;
    852 	}
    853 
    854 #ifdef PERFUSE_DEBUG
    855 	if (perfuse_diagflags & PDF_REQUEUE)
    856 		DPRINTF("%s: DONE  opc = %p\n", __func__, (void *)opc);
    857 #endif
    858 
    859 	return dequeued;
    860 }
    861 
    862 void
    863 perfuse_fs_init(struct puffs_usermount *pu)
    864 {
    865 	struct perfuse_state *ps;
    866 	perfuse_msg_t *pm;
    867 	struct fuse_init_in *fii;
    868 	struct fuse_init_out *fio;
    869 	int error;
    870 
    871 	ps = puffs_getspecific(pu);
    872 
    873         if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
    874                 DERR(EX_OSERR, "%s: puffs_mount failed", __func__);
    875 
    876 	/*
    877 	 * Linux 2.6.34.1 sends theses flags:
    878 	 * FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
    879 	 * FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
    880 	 *
    881 	 * Linux also sets max_readahead at 32 pages (128 kB)
    882 	 *
    883 	 * ps_new_msg() is called with NULL creds, which will
    884 	 * be interpreted as FUSE superuser.
    885 	 */
    886 	pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
    887 	fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
    888 	fii->major = FUSE_KERNEL_VERSION;
    889 	fii->minor = FUSE_KERNEL_MINOR_VERSION;
    890 	fii->max_readahead = (unsigned int)(32 * sysconf(_SC_PAGESIZE));
    891 	fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
    892 
    893 	if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
    894 		DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
    895 
    896 	fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
    897 	ps->ps_max_readahead = fio->max_readahead;
    898 	ps->ps_max_write = fio->max_write;
    899 
    900 	ps->ps_destroy_msg(pm);
    901 
    902 	return;
    903 }
    904 
    905 int
    906 perfuse_fs_unmount(struct puffs_usermount *pu, int flags)
    907 {
    908 	perfuse_msg_t *pm;
    909 	struct perfuse_state *ps;
    910 	puffs_cookie_t opc;
    911 	int error;
    912 
    913 	ps = puffs_getspecific(pu);
    914 	opc = (puffs_cookie_t)puffs_getroot(pu);
    915 
    916 	/*
    917 	 * ps_new_msg() is called with NULL creds, which will
    918 	 * be interpreted as FUSE superuser.
    919 	 */
    920 	pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
    921 
    922 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
    923 		DWARN("unmount %s", ps->ps_target);
    924 		if (!(flags & MNT_FORCE))
    925 			return error;
    926 		else
    927 			error = 0;
    928 	} else {
    929 		ps->ps_destroy_msg(pm);
    930 	}
    931 
    932 	ps->ps_umount(pu);
    933 
    934 	if (perfuse_diagflags & PDF_MISC)
    935 		DPRINTF("%s unmounted, exit\n", ps->ps_target);
    936 
    937 	return 0;
    938 }
    939 
    940 int
    941 perfuse_fs_statvfs(struct puffs_usermount *pu, struct statvfs *svfsb)
    942 {
    943 	struct perfuse_state *ps;
    944 	perfuse_msg_t *pm;
    945 	puffs_cookie_t opc;
    946 	struct fuse_statfs_out *fso;
    947 	int error;
    948 
    949 	ps = puffs_getspecific(pu);
    950 	opc = (puffs_cookie_t)puffs_getroot(pu);
    951 
    952 	/*
    953 	 * ps_new_msg() is called with NULL creds, which will
    954 	 * be interpreted as FUSE superuser.
    955 	 */
    956 	pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
    957 
    958 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
    959 		return error;
    960 
    961 	fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
    962 	svfsb->f_flag = ps->ps_mountflags;
    963 	svfsb->f_bsize = fso->st.bsize;
    964 	svfsb->f_frsize = fso->st.frsize;
    965 	svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
    966 	svfsb->f_blocks = fso->st.blocks;
    967 	svfsb->f_bfree = fso->st.bfree;
    968 	svfsb->f_bavail = fso->st.bavail;
    969 	svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
    970 	svfsb->f_files = fso->st.files;
    971 	svfsb->f_ffree = fso->st.ffree;
    972 	svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
    973 	svfsb->f_fresvd = 0;		/* files reserved for root */
    974 
    975 	svfsb->f_syncreads = ps->ps_syncreads;
    976 	svfsb->f_syncwrites = ps->ps_syncwrites;
    977 
    978 	svfsb->f_asyncreads = ps->ps_asyncreads;
    979 	svfsb->f_asyncwrites = ps->ps_asyncwrites;
    980 
    981 	(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
    982 	svfsb->f_fsid = (unsigned long)ps->ps_fsid;
    983 	svfsb->f_namemax = MAXPATHLEN;	/* XXX */
    984 	svfsb->f_owner = ps->ps_owner_uid;
    985 
    986 	(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
    987 
    988 	if (ps->ps_filesystemtype != NULL)
    989 		(void)strlcpy(svfsb->f_fstypename,
    990 			      ps->ps_filesystemtype, _VFS_NAMELEN);
    991 	else
    992 		(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
    993 
    994 	if (ps->ps_source != NULL)
    995 		strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
    996 	else
    997 		strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
    998 
    999 	ps->ps_destroy_msg(pm);
   1000 
   1001 	return 0;
   1002 }
   1003 
   1004 int
   1005 perfuse_fs_sync(struct puffs_usermount *pu, int waitfor,
   1006 	const struct puffs_cred *pcr)
   1007 {
   1008 	/*
   1009 	 * FUSE does not seem to have a FS sync callback.
   1010 	 * Maybe do not even register this callback
   1011 	 */
   1012 	return puffs_fsnop_sync(pu, waitfor, pcr);
   1013 }
   1014 
   1015 /* ARGSUSED0 */
   1016 int
   1017 perfuse_fs_fhtonode(struct puffs_usermount *pu, void *fid, size_t fidsize,
   1018 	struct puffs_newinfo *pni)
   1019 {
   1020 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
   1021 	return 0;
   1022 }
   1023 
   1024 /* ARGSUSED0 */
   1025 int
   1026 perfuse_fs_nodetofh(struct puffs_usermount *pu, puffs_cookie_t cookie,
   1027 	void *fid, size_t *fidsize)
   1028 {
   1029 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
   1030 	return 0;
   1031 }
   1032 
   1033 #if 0
   1034 /* ARGSUSED0 */
   1035 void
   1036 perfuse_fs_extattrctl(struct puffs_usermount *pu, int cmd,
   1037 	puffs_cookie_t *cookie, int flags, int namespace, const char *attrname)
   1038 {
   1039 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
   1040 	return 0;
   1041 }
   1042 #endif /* 0 */
   1043 
   1044 /* ARGSUSED0 */
   1045 void
   1046 perfuse_fs_suspend(struct puffs_usermount *pu, int status)
   1047 {
   1048 	return;
   1049 }
   1050 
   1051 
   1052 int
   1053 perfuse_node_lookup(struct puffs_usermount *pu, puffs_cookie_t opc,
   1054 	struct puffs_newinfo *pni, const struct puffs_cn *pcn)
   1055 {
   1056 	struct perfuse_state *ps;
   1057 	struct puffs_node *pn;
   1058 	mode_t mode;
   1059 	int error;
   1060 
   1061 	ps = puffs_getspecific(pu);
   1062 	node_ref(opc);
   1063 
   1064 	/*
   1065 	 * Check permissions
   1066 	 */
   1067 	switch(pcn->pcn_nameiop) {
   1068 	case NAMEI_DELETE: /* FALLTHROUGH */
   1069 	case NAMEI_RENAME: /* FALLTHROUGH */
   1070 	case NAMEI_CREATE:
   1071 		if (pcn->pcn_flags & NAMEI_ISLASTCN)
   1072 			mode = PUFFS_VEXEC|PUFFS_VWRITE;
   1073 		else
   1074 			mode = PUFFS_VEXEC;
   1075 		break;
   1076 	case NAMEI_LOOKUP: /* FALLTHROUGH */
   1077 	default:
   1078 		mode = PUFFS_VEXEC;
   1079 		break;
   1080 	}
   1081 
   1082 	if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
   1083 		goto out;
   1084 
   1085 	error = node_lookup_common(pu, (puffs_cookie_t)opc, pni,
   1086 				   pcn->pcn_name, pcn->pcn_cred, &pn);
   1087 
   1088 	if (error != 0)
   1089 		goto out;
   1090 
   1091 	/*
   1092 	 * Kernel would kill us if the filesystem returned the parent
   1093 	 * itself. If we want to live, hide that!
   1094 	 */
   1095 	if ((opc == (puffs_cookie_t)pn) && (strcmp(pcn->pcn_name, ".") != 0)) {
   1096 		DERRX(EX_SOFTWARE, "lookup \"%s\" in \"%s\" returned parent",
   1097 		      pcn->pcn_name, perfuse_node_path(ps, opc));
   1098 		/* NOTREACHED */
   1099 		error = ESTALE;
   1100 		goto out;
   1101 	}
   1102 
   1103 	/*
   1104 	 * Removed node
   1105 	 */
   1106 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED) {
   1107 		error = ENOENT;
   1108 		goto out;
   1109 	}
   1110 
   1111 	/*
   1112 	 * Check for sticky bit. Unfortunately there is no way to
   1113 	 * do this before creating the puffs_node, since we require
   1114 	 * this operation to get the node owner.
   1115 	 */
   1116 	switch (pcn->pcn_nameiop) {
   1117 	case NAMEI_DELETE: /* FALLTHROUGH */
   1118 	case NAMEI_RENAME:
   1119 		error = sticky_access(opc, pn, pcn->pcn_cred);
   1120 		if (error != 0) {
   1121 			(void)perfuse_node_reclaim(pu, pn);
   1122 			goto out;
   1123 		}
   1124 		break;
   1125 	default:
   1126 		break;
   1127 	}
   1128 
   1129 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
   1130 
   1131 	error = 0;
   1132 
   1133 out:
   1134 	node_rele(opc);
   1135 	return error;
   1136 }
   1137 
   1138 int
   1139 perfuse_node_create(struct puffs_usermount *pu, puffs_cookie_t opc,
   1140 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
   1141 	const struct vattr *vap)
   1142 {
   1143 	perfuse_msg_t *pm;
   1144 	struct perfuse_state *ps;
   1145 	struct fuse_create_in *fci;
   1146 	struct fuse_entry_out *feo;
   1147 	struct fuse_open_out *foo;
   1148 	struct puffs_node *pn;
   1149 	const char *name;
   1150 	size_t namelen;
   1151 	size_t len;
   1152 	int error;
   1153 
   1154 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   1155 		return ENOENT;
   1156 
   1157 	node_ref(opc);
   1158 
   1159 	/*
   1160 	 * If create is unimplemented: Check that it does not
   1161 	 * already exists, and if not, do mknod and open
   1162 	 */
   1163 	ps = puffs_getspecific(pu);
   1164 	if (ps->ps_flags & PS_NO_CREAT) {
   1165 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
   1166 					   pcn->pcn_cred, &pn);
   1167 		if (error == 0)	{
   1168 			(void)perfuse_node_reclaim(pu, pn);
   1169 			error = EEXIST;
   1170 			goto out;
   1171 		}
   1172 
   1173 		error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
   1174 		if (error != 0)
   1175 			goto out;
   1176 
   1177 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
   1178 					   pcn->pcn_cred, &pn);
   1179 		if (error != 0)
   1180 			goto out;
   1181 
   1182 		/*
   1183 		 * FUSE does the open at create time, while
   1184 		 * NetBSD will open in a subsequent operation.
   1185 		 * We need to open now, in order to retain FUSE
   1186 		 * semantics. The calling process will not get
   1187 		 * a file descriptor before the kernel sends
   1188 		 * the open operation.
   1189 		 */
   1190 		error = perfuse_node_open(pu, (puffs_cookie_t)pn,
   1191 					  FWRITE, pcn->pcn_cred);
   1192 		goto out;
   1193 	}
   1194 
   1195 	name = pcn->pcn_name;
   1196 	namelen = pcn->pcn_namelen + 1;
   1197 	len = sizeof(*fci) + namelen;
   1198 
   1199 	/*
   1200 	 * flags should use O_WRONLY instead of O_RDWR, but it
   1201 	 * breaks when the caller tries to read from file.
   1202 	 *
   1203 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
   1204 	 */
   1205 	pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, pcn->pcn_cred);
   1206 	fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
   1207 	fci->flags = O_CREAT | O_TRUNC | O_RDWR;
   1208 	fci->mode = vap->va_mode | VTTOIF(vap->va_type);
   1209 	fci->umask = 0; 	/* Seems unused by libfuse */
   1210 	(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
   1211 
   1212 	len = sizeof(*feo) + sizeof(*foo);
   1213 	if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0) {
   1214 		/*
   1215 		 * create is unimplmented, remember it for later,
   1216 		 * and start over using mknod and open instead.
   1217 		 */
   1218 		if (error == ENOSYS) {
   1219 			ps->ps_flags |= PS_NO_CREAT;
   1220 			error = perfuse_node_create(pu, opc, pni, pcn, vap);
   1221 		}
   1222 
   1223 		goto out;
   1224 	}
   1225 
   1226 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
   1227 	foo = (struct fuse_open_out *)(void *)(feo + 1);
   1228 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
   1229 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
   1230 
   1231 	/*
   1232 	 * Save the file handle and inode in node private data
   1233 	 * so that we can reuse it later
   1234 	 */
   1235 	pn = perfuse_new_pn(pu, name, opc);
   1236 	perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
   1237 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
   1238 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
   1239 	perfuse_node_cache(ps, pn);
   1240 
   1241 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
   1242 	pn->pn_va.va_gen = (u_long)(feo->generation);
   1243 
   1244 	puffs_newinfo_setcookie(pni, pn);
   1245 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
   1246 	puffs_newinfo_setva(pni, &pn->pn_va);
   1247 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
   1248 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
   1249 
   1250 #ifdef PERFUSE_DEBUG
   1251 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
   1252 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
   1253 			"nodeid = 0x%"PRIx64", wfh = 0x%"PRIx64"\n",
   1254 			__func__, (void *)pn, pcn->pcn_name,
   1255 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid,
   1256 			foo->fh);
   1257 #endif
   1258 
   1259 	ps->ps_destroy_msg(pm);
   1260 	error = 0;
   1261 
   1262 out:
   1263 	node_rele(opc);
   1264 	return error;
   1265 }
   1266 
   1267 
   1268 int
   1269 perfuse_node_mknod(struct puffs_usermount *pu, puffs_cookie_t opc,
   1270 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
   1271 	const struct vattr *vap)
   1272 {
   1273 	struct perfuse_state *ps;
   1274 	perfuse_msg_t *pm;
   1275 	struct fuse_mknod_in *fmi;
   1276 	const char* path;
   1277 	size_t len;
   1278 	int error;
   1279 
   1280 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   1281 		return ENOENT;
   1282 
   1283 	node_ref(opc);
   1284 
   1285 	/*
   1286 	 * Only superuser can mknod objects other than
   1287 	 * directories, files, socks, fifo and links.
   1288 	 *
   1289 	 * Create an object require -WX permission in the parent directory
   1290 	 */
   1291 	switch (vap->va_type) {
   1292 	case VDIR:	/* FALLTHROUGH */
   1293 	case VREG:	/* FALLTHROUGH */
   1294 	case VFIFO:	/* FALLTHROUGH */
   1295 	case VSOCK:
   1296 		break;
   1297 	default:	/* VNON, VBLK, VCHR, VBAD */
   1298 		if (!puffs_cred_isjuggernaut(pcn->pcn_cred)) {
   1299 			error = EPERM;
   1300 			goto out;
   1301 		}
   1302 		break;
   1303 	}
   1304 
   1305 
   1306 	ps = puffs_getspecific(pu);
   1307 	path = pcn->pcn_name;
   1308 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
   1309 
   1310 	/*
   1311 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
   1312 	 */
   1313 	pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, pcn->pcn_cred);
   1314 	fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
   1315 	fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
   1316 	fmi->rdev = (uint32_t)vap->va_rdev;
   1317 	fmi->umask = 0; 	/* Seems unused bu libfuse */
   1318 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
   1319 
   1320 	error = node_mk_common(pu, opc, pni, pcn, pm);
   1321 
   1322 out:
   1323 	node_rele(opc);
   1324 	return error;
   1325 }
   1326 
   1327 
   1328 int
   1329 perfuse_node_open(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
   1330 	const struct puffs_cred *pcr)
   1331 {
   1332 	return perfuse_node_open2(pu, opc, mode, pcr, NULL);
   1333 }
   1334 
   1335 int
   1336 perfuse_node_open2(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
   1337 	const struct puffs_cred *pcr, int *oflags)
   1338 {
   1339 	struct perfuse_state *ps;
   1340 	struct perfuse_node_data *pnd;
   1341 	perfuse_msg_t *pm;
   1342 	mode_t fmode;
   1343 	int op;
   1344 	struct fuse_open_in *foi;
   1345 	struct fuse_open_out *foo;
   1346 	struct puffs_node *pn;
   1347 	int error;
   1348 
   1349 	ps = puffs_getspecific(pu);
   1350 	pn = (struct puffs_node *)opc;
   1351 	pnd = PERFUSE_NODE_DATA(opc);
   1352 	error = 0;
   1353 
   1354 	if (pnd->pnd_flags & PND_REMOVED)
   1355 		return ENOENT;
   1356 
   1357 	node_ref(opc);
   1358 
   1359 	if (puffs_pn_getvap(pn)->va_type == VDIR)
   1360 		op = FUSE_OPENDIR;
   1361 	else
   1362 		op = FUSE_OPEN;
   1363 
   1364 	/*
   1365 	 * libfuse docs says
   1366 	 * - O_CREAT and O_EXCL should never be set.
   1367 	 * - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
   1368 	 *
   1369 	 * O_APPEND makes no sense since FUSE always sends
   1370 	 * the file offset for write operations. If the
   1371 	 * filesystem uses pwrite(), O_APPEND would cause
   1372 	 * the offset to be ignored and cause file corruption.
   1373 	 */
   1374 	mode &= ~(O_CREAT|O_EXCL|O_APPEND);
   1375 
   1376 	/*
   1377 	 * Do not open twice, and do not reopen for reading
   1378 	 * if we already have write handle.
   1379 	 */
   1380 	switch (mode & (FREAD|FWRITE)) {
   1381 	case FREAD:
   1382 		if (pnd->pnd_flags & (PND_RFH|PND_WFH))
   1383 			goto out;
   1384 		break;
   1385 	case FWRITE:
   1386 		if (pnd->pnd_flags & PND_WFH)
   1387 			goto out;
   1388 		break;
   1389 	case FREAD|FWRITE:
   1390 		if (pnd->pnd_flags & PND_WFH)
   1391 			goto out;
   1392 
   1393 		/*
   1394 		 * Corner case: if already open for reading (PND_RFH)
   1395 		 * and re-opening FREAD|FWRITE, we need to reopen,
   1396 		 * but only for writing. Note the change on mode
   1397 		 * will only affect perfuse_new_fh()
   1398 		 */
   1399 		if (pnd->pnd_flags & PND_RFH)
   1400 			mode &= ~FREAD;
   1401 		break;
   1402 	default:
   1403 		DWARNX("open without either FREAD nor FWRITE");
   1404 		error = EPERM;
   1405 		goto out;
   1406 	}
   1407 
   1408 	/*
   1409 	 * Queue open on a node so that we do not open
   1410 	 * twice. This would be better with read and
   1411 	 * write distinguished.
   1412 	 */
   1413 	while (pnd->pnd_flags & PND_INOPEN)
   1414 		requeue_request(pu, opc, PCQ_OPEN);
   1415 	pnd->pnd_flags |= PND_INOPEN;
   1416 
   1417 	/*
   1418 	 * Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
   1419 	 * to O_RDONLY/O_WRONLY while perserving the other options.
   1420 	 */
   1421 	fmode = mode & ~(FREAD|FWRITE);
   1422 	fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
   1423 
   1424 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
   1425 	foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
   1426 	foi->flags = fmode;
   1427 	foi->unused = 0;
   1428 
   1429 	if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
   1430 		goto out;
   1431 
   1432 	foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
   1433 
   1434 	/*
   1435 	 * Save the file handle in node private data
   1436 	 * so that we can reuse it later
   1437 	 */
   1438 	perfuse_new_fh(opc, foo->fh, mode);
   1439 
   1440 	/*
   1441 	 * Set direct I/O if the filesystems forces it
   1442 	 */
   1443 	if ((foo->open_flags & FUSE_FOPEN_DIRECT_IO) && (oflags != NULL))
   1444 		*oflags |= PUFFS_OPEN_IO_DIRECT;
   1445 
   1446 #ifdef PERFUSE_DEBUG
   1447 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
   1448 		DPRINTF("%s: opc = %p, file = \"%s\", "
   1449 			"nodeid = 0x%"PRIx64", %s%sfh = 0x%"PRIx64"\n",
   1450 			__func__, (void *)opc, perfuse_node_path(ps, opc),
   1451 			pnd->pnd_nodeid, mode & FREAD ? "r" : "",
   1452 			mode & FWRITE ? "w" : "", foo->fh);
   1453 #endif
   1454 
   1455 	ps->ps_destroy_msg(pm);
   1456 out:
   1457 
   1458 	pnd->pnd_flags &= ~PND_INOPEN;
   1459 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
   1460 
   1461 	node_rele(opc);
   1462 	return error;
   1463 }
   1464 
   1465 /* ARGSUSED0 */
   1466 int
   1467 perfuse_node_close(struct puffs_usermount *pu, puffs_cookie_t opc, int flags,
   1468 	const struct puffs_cred *pcr)
   1469 {
   1470 	struct perfuse_node_data *pnd;
   1471 
   1472 	pnd = PERFUSE_NODE_DATA(opc);
   1473 
   1474 	if (!(pnd->pnd_flags & PND_OPEN))
   1475 		return EBADF;
   1476 
   1477 	/*
   1478 	 * Actual close is postponed at inactive time.
   1479 	 */
   1480 	return 0;
   1481 }
   1482 
   1483 int
   1484 perfuse_node_access(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
   1485 	const struct puffs_cred *pcr)
   1486 {
   1487 	perfuse_msg_t *pm;
   1488 	struct perfuse_state *ps;
   1489 	struct fuse_access_in *fai;
   1490 	int error;
   1491 
   1492 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   1493 		return ENOENT;
   1494 
   1495 	node_ref(opc);
   1496 
   1497 	/*
   1498 	 * If we previously detected the filesystem does not
   1499 	 * implement access(), short-circuit the call and skip
   1500 	 * to libpuffs access() emulation.
   1501 	 */
   1502 	ps = puffs_getspecific(pu);
   1503 	if (ps->ps_flags & PS_NO_ACCESS) {
   1504 		const struct vattr *vap;
   1505 
   1506 		vap = puffs_pn_getvap((struct puffs_node *)opc);
   1507 
   1508 		error = puffs_access(IFTOVT(vap->va_mode),
   1509 				     vap->va_mode & ACCESSPERMS,
   1510 				     vap->va_uid, vap->va_gid,
   1511 				     (mode_t)mode, pcr);
   1512 		goto out;
   1513 	}
   1514 
   1515 	/*
   1516 	 * Plain access call
   1517 	 */
   1518 	pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
   1519 	fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
   1520 	fai->mask = 0;
   1521 	fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
   1522 	fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
   1523 	fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
   1524 
   1525 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
   1526 
   1527 	ps->ps_destroy_msg(pm);
   1528 
   1529 	/*
   1530 	 * If unimplemented, start over with emulation
   1531 	 */
   1532 	if (error == ENOSYS) {
   1533 		ps->ps_flags |= PS_NO_ACCESS;
   1534 		error = perfuse_node_access(pu, opc, mode, pcr);
   1535 	}
   1536 
   1537 out:
   1538 	node_rele(opc);
   1539 	return error;
   1540 }
   1541 
   1542 int
   1543 perfuse_node_getattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   1544 	struct vattr *vap, const struct puffs_cred *pcr)
   1545 {
   1546 	return perfuse_node_getattr_ttl(pu, opc, vap, pcr, NULL);
   1547 }
   1548 
   1549 int
   1550 perfuse_node_getattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
   1551 	struct vattr *vap, const struct puffs_cred *pcr,
   1552 	struct timespec *va_ttl)
   1553 {
   1554 	perfuse_msg_t *pm = NULL;
   1555 	struct perfuse_state *ps;
   1556 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
   1557 	struct fuse_getattr_in *fgi;
   1558 	struct fuse_attr_out *fao;
   1559 	int error = 0;
   1560 
   1561 	if ((pnd->pnd_flags & PND_REMOVED) && !(pnd->pnd_flags & PND_OPEN))
   1562 		return ENOENT;
   1563 
   1564 	node_ref(opc);
   1565 
   1566 	/*
   1567 	 * Serialize size access, see comment in perfuse_node_setattr().
   1568 	 */
   1569 	while (pnd->pnd_flags & PND_INRESIZE)
   1570 		requeue_request(pu, opc, PCQ_RESIZE);
   1571 	pnd->pnd_flags |= PND_INRESIZE;
   1572 
   1573 	ps = puffs_getspecific(pu);
   1574 
   1575 	/*
   1576 	 * FUSE_GETATTR_FH must be set in fgi->flags
   1577 	 * if we use for fgi->fh
   1578 	 */
   1579 	pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
   1580 	fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
   1581 	fgi->getattr_flags = 0;
   1582 	fgi->dummy = 0;
   1583 	fgi->fh = 0;
   1584 
   1585 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
   1586 		fgi->fh = perfuse_get_fh(opc, FREAD);
   1587 		fgi->getattr_flags |= FUSE_GETATTR_FH;
   1588 	}
   1589 
   1590 #ifdef PERFUSE_DEBUG
   1591 	if (perfuse_diagflags & PDF_RESIZE)
   1592 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__, (void *)opc,
   1593 		    vap->va_size);
   1594 #endif
   1595 
   1596 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
   1597 		goto out;
   1598 
   1599 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
   1600 
   1601 #ifdef PERFUSE_DEBUG
   1602 	if (perfuse_diagflags & PDF_RESIZE)
   1603 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
   1604 		    (void *)opc, vap->va_size, fao->attr.size);
   1605 #endif
   1606 
   1607 	/*
   1608 	 * We set birthtime, flags, filerev,vaflags to 0.
   1609 	 * This seems the best bet, since the information is
   1610 	 * not available from filesystem.
   1611 	 */
   1612 	fuse_attr_to_vap(ps, vap, &fao->attr);
   1613 
   1614 	if (va_ttl != NULL) {
   1615 		va_ttl->tv_sec = fao->attr_valid;
   1616 		va_ttl->tv_nsec = fao->attr_valid_nsec;
   1617 	}
   1618 
   1619 	ps->ps_destroy_msg(pm);
   1620 	error = 0;
   1621 out:
   1622 
   1623 	pnd->pnd_flags &= ~PND_INRESIZE;
   1624 	(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
   1625 
   1626 	node_rele(opc);
   1627 	return error;
   1628 }
   1629 
   1630 int
   1631 perfuse_node_setattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   1632 	const struct vattr *vap, const struct puffs_cred *pcr)
   1633 {
   1634 	return perfuse_node_setattr_ttl(pu, opc,
   1635 					__UNCONST(vap), pcr, NULL, 0);
   1636 }
   1637 
   1638 int
   1639 perfuse_node_setattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
   1640 	struct vattr *vap, const struct puffs_cred *pcr,
   1641 	struct timespec *va_ttl, int xflag)
   1642 {
   1643 	perfuse_msg_t *pm;
   1644 	uint64_t fh;
   1645 	struct perfuse_state *ps;
   1646 	struct perfuse_node_data *pnd;
   1647 	struct fuse_setattr_in *fsi;
   1648 	struct fuse_attr_out *fao;
   1649 	struct vattr *old_va;
   1650 	enum perfuse_xchg_pb_reply reply;
   1651 	int error;
   1652 #ifdef PERFUSE_DEBUG
   1653 	struct vattr *old_vap;
   1654 	int resize_debug = 0;
   1655 #endif
   1656 	ps = puffs_getspecific(pu);
   1657 	pnd = PERFUSE_NODE_DATA(opc);
   1658 
   1659 	/*
   1660 	 * The only operation we can do once the file is removed
   1661 	 * is to resize it, and we can do it only if it is open.
   1662 	 * Do not even send the operation to the filesystem: the
   1663 	 * file is not there anymore.
   1664 	 */
   1665 	if (pnd->pnd_flags & PND_REMOVED) {
   1666 		if (!(pnd->pnd_flags & PND_OPEN))
   1667 			return ENOENT;
   1668 
   1669 		return 0;
   1670 	}
   1671 
   1672 	old_va = puffs_pn_getvap((struct puffs_node *)opc);
   1673 
   1674 	/*
   1675 	 * Check for permission to change size
   1676 	 * It is always allowed if we already have a write file handle
   1677 	 */
   1678 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
   1679 	    !(pnd->pnd_flags & PND_WFH) &&
   1680 	    (error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
   1681 		return error;
   1682 
   1683 	/*
   1684 	 * Check for permission to change dates
   1685 	 */
   1686 	if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
   1687 	     (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
   1688 	    (puffs_access_times(old_va->va_uid, old_va->va_gid,
   1689 				old_va->va_mode, 0, pcr) != 0))
   1690 		return EPERM;
   1691 
   1692 	/*
   1693 	 * Check for permission to change owner and group
   1694 	 */
   1695 	if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
   1696 	     (vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
   1697 	    (puffs_access_chown(old_va->va_uid, old_va->va_gid,
   1698 				vap->va_uid, vap->va_gid, pcr)) != 0)
   1699 		return EPERM;
   1700 
   1701 	/*
   1702 	 * Check for sticky bit on non-directory by non root user
   1703 	 */
   1704 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
   1705 	    (vap->va_mode & S_ISTXT) && (old_va->va_type != VDIR) &&
   1706 	    !puffs_cred_isjuggernaut(pcr))
   1707 		return EFTYPE;
   1708 
   1709 	/*
   1710 	 * Check for permission to change permissions
   1711 	 */
   1712 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
   1713 	    (puffs_access_chmod(old_va->va_uid, old_va->va_gid,
   1714 				old_va->va_type, vap->va_mode, pcr)) != 0)
   1715 		return EPERM;
   1716 
   1717 	node_ref(opc);
   1718 
   1719 	if (pnd->pnd_flags & PND_WFH)
   1720 		fh = perfuse_get_fh(opc, FWRITE);
   1721 	else
   1722 		fh = FUSE_UNKNOWN_FH;
   1723 
   1724 	/*
   1725 	 * fchmod() sets mode and fh, and it may carry
   1726 	 * a resize as well. That may break if the
   1727 	 * filesystem does chmod then resize, and fails
   1728 	 * because it does not have permission anymore.
   1729 	 * We work this around by splitting into two setattr.
   1730 	 */
   1731 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
   1732 	    (vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
   1733 	    (fh != FUSE_UNKNOWN_FH)) {
   1734 		struct vattr resize_va;
   1735 
   1736 		(void)memcpy(&resize_va, vap, sizeof(resize_va));
   1737 		resize_va.va_mode = (mode_t)PUFFS_VNOVAL;
   1738 		if ((error = perfuse_node_setattr_ttl(pu, opc, &resize_va,
   1739 						      pcr, va_ttl, xflag)) != 0)
   1740 			goto out2;
   1741 
   1742 		vap->va_size = (u_quad_t)PUFFS_VNOVAL;
   1743 	}
   1744 
   1745 	pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
   1746 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
   1747 	fsi->valid = 0;
   1748 
   1749 	/*
   1750 	 * Get a fh if the node is open for writing
   1751 	 */
   1752 	if (fh != FUSE_UNKNOWN_FH) {
   1753 		fsi->fh = fh;
   1754 		fsi->valid |= FUSE_FATTR_FH;
   1755 	}
   1756 
   1757 
   1758 	if (vap->va_size != (u_quad_t)PUFFS_VNOVAL) {
   1759 		fsi->size = vap->va_size;
   1760 		fsi->valid |= FUSE_FATTR_SIZE;
   1761 
   1762 		/*
   1763 		 * Serialize anything that can touch file size
   1764 		 * to avoid reordered GETATTR and SETATTR.
   1765 		 * Out of order SETATTR can report stale size,
   1766 		 * which will cause the kernel to truncate the file.
   1767 		 * XXX Probably useless now we have a lock on GETATTR
   1768 		 */
   1769 		while (pnd->pnd_flags & PND_INRESIZE)
   1770 			requeue_request(pu, opc, PCQ_RESIZE);
   1771 		pnd->pnd_flags |= PND_INRESIZE;
   1772 	}
   1773 
   1774 	/*
   1775  	 * Setting mtime without atime or vice versa leads to
   1776 	 * dates being reset to Epoch on glusterfs. If one
   1777 	 * is missing, use the old value.
   1778  	 */
   1779 	if ((vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) ||
   1780 	    (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL)) {
   1781 
   1782 		if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
   1783 			fsi->atime = vap->va_atime.tv_sec;
   1784 			fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
   1785 		} else {
   1786 			fsi->atime = old_va->va_atime.tv_sec;
   1787 			fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
   1788 		}
   1789 
   1790 		if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
   1791 			fsi->mtime = vap->va_mtime.tv_sec;
   1792 			fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
   1793 		} else {
   1794 			fsi->mtime = old_va->va_mtime.tv_sec;
   1795 			fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
   1796 		}
   1797 
   1798 		fsi->valid |= (FUSE_FATTR_MTIME|FUSE_FATTR_ATIME);
   1799 	}
   1800 
   1801 	if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
   1802 		fsi->mode = vap->va_mode;
   1803 		fsi->valid |= FUSE_FATTR_MODE;
   1804 	}
   1805 
   1806 	if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
   1807 		fsi->uid = vap->va_uid;
   1808 		fsi->valid |= FUSE_FATTR_UID;
   1809 	}
   1810 
   1811 	if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
   1812 		fsi->gid = vap->va_gid;
   1813 		fsi->valid |= FUSE_FATTR_GID;
   1814 	}
   1815 
   1816 	if (pnd->pnd_lock_owner != 0) {
   1817 		fsi->lock_owner = pnd->pnd_lock_owner;
   1818 		fsi->valid |= FUSE_FATTR_LOCKOWNER;
   1819 	}
   1820 
   1821 	/*
   1822 	 * ftruncate() sends only va_size, and metadata cache
   1823 	 * flush adds va_atime and va_mtime. Some FUSE
   1824 	 * filesystems will attempt to detect ftruncate by
   1825 	 * checking for FATTR_SIZE being set without
   1826 	 * FATTR_UID|FATTR_GID|FATTR_ATIME|FATTR_MTIME|FATTR_MODE
   1827 	 *
   1828 	 * Try to adapt and remove FATTR_ATIME|FATTR_MTIME
   1829 	 * if we suspect a ftruncate().
   1830 	 */
   1831 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
   1832 	    ((vap->va_mode == (mode_t)PUFFS_VNOVAL) &&
   1833 	     (vap->va_uid == (uid_t)PUFFS_VNOVAL) &&
   1834 	     (vap->va_gid == (gid_t)PUFFS_VNOVAL))) {
   1835 		fsi->atime = 0;
   1836 		fsi->atimensec = 0;
   1837 		fsi->mtime = 0;
   1838 		fsi->mtimensec = 0;
   1839 		fsi->valid &= ~(FUSE_FATTR_ATIME|FUSE_FATTR_MTIME);
   1840 	}
   1841 
   1842 	/*
   1843 	 * If nothing remain, discard the operation.
   1844 	 */
   1845 	if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
   1846 			    FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
   1847 		error = 0;
   1848 		ps->ps_destroy_msg(pm);
   1849 		goto out;
   1850 	}
   1851 
   1852 #ifdef PERFUSE_DEBUG
   1853 	old_vap = puffs_pn_getvap((struct puffs_node *)opc);
   1854 
   1855 	if ((perfuse_diagflags & PDF_RESIZE) &&
   1856 	    (old_vap->va_size != (u_quad_t)PUFFS_VNOVAL)) {
   1857 		resize_debug = 1;
   1858 
   1859 		DPRINTF(">> %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
   1860 		    (void *)opc,
   1861 		    puffs_pn_getvap((struct puffs_node *)opc)->va_size,
   1862 		    fsi->size);
   1863 	}
   1864 #endif
   1865 
   1866 	/*
   1867 	 * Do not honour FAF when changing size. How do
   1868 	 * you want such a thing to work?
   1869 	 */
   1870 	reply = wait_reply;
   1871 #ifdef PUFFS_SETATTR_FAF
   1872 	if ((xflag & PUFFS_SETATTR_FAF) && !(fsi->valid & FUSE_FATTR_SIZE))
   1873 		reply = no_reply;
   1874 #endif
   1875 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), reply)) != 0)
   1876 		goto out;
   1877 
   1878 	if (reply == no_reply)
   1879 		goto out;
   1880 
   1881 	/*
   1882 	 * Copy back the new values
   1883 	 */
   1884 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
   1885 
   1886 #ifdef PERFUSE_DEBUG
   1887 	if (resize_debug)
   1888 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
   1889 		    (void *)opc, old_vap->va_size, fao->attr.size);
   1890 #endif
   1891 
   1892 	fuse_attr_to_vap(ps, old_va, &fao->attr);
   1893 
   1894 	if (va_ttl != NULL) {
   1895 		va_ttl->tv_sec = fao->attr_valid;
   1896 		va_ttl->tv_nsec = fao->attr_valid_nsec;
   1897 		(void)memcpy(vap, old_va, sizeof(*vap));
   1898 	}
   1899 
   1900 	ps->ps_destroy_msg(pm);
   1901 	error = 0;
   1902 
   1903 out:
   1904 	if (pnd->pnd_flags & PND_INRESIZE) {
   1905 		pnd->pnd_flags &= ~PND_INRESIZE;
   1906 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
   1907 	}
   1908 
   1909 out2:
   1910 	node_rele(opc);
   1911 	return error;
   1912 }
   1913 
   1914 int
   1915 perfuse_node_poll(struct puffs_usermount *pu, puffs_cookie_t opc, int *events)
   1916 {
   1917 	struct perfuse_state *ps;
   1918 	perfuse_msg_t *pm;
   1919 	struct fuse_poll_in *fpi;
   1920 	struct fuse_poll_out *fpo;
   1921 	int error;
   1922 
   1923 	node_ref(opc);
   1924 	ps = puffs_getspecific(pu);
   1925 	/*
   1926 	 * kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
   1927 	 *
   1928 	 * XXX ps_new_msg() is called with NULL creds, which will
   1929 	 * be interpreted as FUSE superuser. We have no way to
   1930 	 * know the requesting process' credential, but since poll
   1931 	 * is supposed to operate on a file that has been open,
   1932 	 * permission should have already been checked at open time.
   1933 	 * That still may breaks on filesystems that provides odd
   1934 	 * semantics.
   1935  	 */
   1936 	pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
   1937 	fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
   1938 	fpi->fh = perfuse_get_fh(opc, FREAD);
   1939 	fpi->kh = 0;
   1940 	fpi->flags = 0;
   1941 
   1942 #ifdef PERFUSE_DEBUG
   1943 	if (perfuse_diagflags & PDF_FH)
   1944 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
   1945 			"fh = 0x%"PRIx64"\n", __func__, (void *)opc,
   1946 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fpi->fh);
   1947 #endif
   1948 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
   1949 		goto out;
   1950 
   1951 	fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
   1952 	*events = fpo->revents;
   1953 
   1954 	ps->ps_destroy_msg(pm);
   1955 	error = 0;
   1956 
   1957 out:
   1958 	node_rele(opc);
   1959 	return error;
   1960 }
   1961 
   1962 /* ARGSUSED2 */
   1963 int
   1964 perfuse_node_fsync(struct puffs_usermount *pu, puffs_cookie_t opc,
   1965 	const struct puffs_cred *pcr, int flags, off_t offlo, off_t offhi)
   1966 {
   1967 	int op;
   1968 	perfuse_msg_t *pm;
   1969 	struct perfuse_state *ps;
   1970 	struct perfuse_node_data *pnd;
   1971 	struct fuse_fsync_in *ffi;
   1972 	uint64_t fh;
   1973 	int error = 0;
   1974 
   1975 	pm = NULL;
   1976 	ps = puffs_getspecific(pu);
   1977 	pnd = PERFUSE_NODE_DATA(opc);
   1978 
   1979 	/*
   1980 	 * No need to sync a removed node
   1981 	 */
   1982 	if (pnd->pnd_flags & PND_REMOVED)
   1983 		return 0;
   1984 
   1985 	/*
   1986 	 * We do not sync closed files. They have been
   1987 	 * sync at inactive time already.
   1988 	 */
   1989 	if (!(pnd->pnd_flags & PND_OPEN))
   1990 		return 0;
   1991 
   1992 	node_ref(opc);
   1993 
   1994 	if (puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
   1995 		op = FUSE_FSYNCDIR;
   1996 	else 		/* VREG but also other types such as VLNK */
   1997 		op = FUSE_FSYNC;
   1998 
   1999 	/*
   2000 	 * Do not sync if there are no change to sync
   2001 	 * XXX remove that test on files if we implement mmap
   2002 	 */
   2003 #ifdef PERFUSE_DEBUG
   2004 	if (perfuse_diagflags & PDF_SYNC)
   2005 		DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
   2006 			__func__, (void*)opc, perfuse_node_path(ps, opc),
   2007 			pnd->pnd_flags & PND_DIRTY ? "" : "not ");
   2008 #endif
   2009 	if (!(pnd->pnd_flags & PND_DIRTY))
   2010 		goto out;
   2011 
   2012 	/*
   2013 	 * It seems NetBSD can call fsync without open first
   2014 	 * glusterfs complain in such a situation:
   2015 	 * "FSYNC() ERR => -1 (Invalid argument)"
   2016 	 * The file will be closed at inactive time.
   2017 	 *
   2018 	 * We open the directory for reading in order to sync.
   2019 	 * This sounds rather counterintuitive, but it works.
   2020 	 */
   2021 	if (!(pnd->pnd_flags & PND_WFH)) {
   2022 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
   2023 			goto out;
   2024 	}
   2025 
   2026 	if (op == FUSE_FSYNCDIR)
   2027 		fh = perfuse_get_fh(opc, FREAD);
   2028 	else
   2029 		fh = perfuse_get_fh(opc, FWRITE);
   2030 
   2031 	/*
   2032 	 * If fsync_flags  is set, meta data should not be flushed.
   2033 	 */
   2034 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), pcr);
   2035 	ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
   2036 	ffi->fh = fh;
   2037 	ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
   2038 
   2039 #ifdef PERFUSE_DEBUG
   2040 	if (perfuse_diagflags & PDF_FH)
   2041 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
   2042 			__func__, (void *)opc,
   2043 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, ffi->fh);
   2044 #endif
   2045 
   2046 	if ((error = xchg_msg(pu, opc, pm,
   2047 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
   2048 		goto out;
   2049 
   2050 	/*
   2051 	 * No reply beyond fuse_out_header: nothing to do on success
   2052 	 * just clear the dirty flag
   2053 	 */
   2054 	pnd->pnd_flags &= ~PND_DIRTY;
   2055 
   2056 #ifdef PERFUSE_DEBUG
   2057 	if (perfuse_diagflags & PDF_SYNC)
   2058 		DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
   2059 			__func__, (void*)opc, perfuse_node_path(ps, opc));
   2060 #endif
   2061 
   2062 	ps->ps_destroy_msg(pm);
   2063 	error = 0;
   2064 
   2065 out:
   2066 	/*
   2067 	 * ENOSYS is not returned to kernel,
   2068 	 */
   2069 	if (error == ENOSYS)
   2070 		error = 0;
   2071 
   2072 	node_rele(opc);
   2073 	return error;
   2074 }
   2075 
   2076 int
   2077 perfuse_node_remove(struct puffs_usermount *pu, puffs_cookie_t opc,
   2078 	puffs_cookie_t targ, const struct puffs_cn *pcn)
   2079 {
   2080 	struct perfuse_state *ps;
   2081 	struct perfuse_node_data *pnd;
   2082 	perfuse_msg_t *pm;
   2083 	char *path;
   2084 	const char *name;
   2085 	size_t len;
   2086 	int error;
   2087 
   2088 	pnd = PERFUSE_NODE_DATA(opc);
   2089 
   2090 	if ((pnd->pnd_flags & PND_REMOVED) ||
   2091 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
   2092 		return ENOENT;
   2093 
   2094 #ifdef PERFUSE_DEBUG
   2095 	if (targ == NULL)
   2096 		DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
   2097 
   2098 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
   2099 		DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
   2100 			__func__, (void *)opc, (void *)targ, pcn->pcn_name);
   2101 #endif
   2102 	node_ref(opc);
   2103 	node_ref(targ);
   2104 
   2105 	/*
   2106 	 * Await for all operations on the deleted node to drain,
   2107 	 * as the filesystem may be confused to have it deleted
   2108 	 * during a getattr
   2109 	 */
   2110 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
   2111 		requeue_request(pu, targ, PCQ_AFTERXCHG);
   2112 
   2113 	ps = puffs_getspecific(pu);
   2114 	pnd = PERFUSE_NODE_DATA(opc);
   2115 	name = pcn->pcn_name;
   2116 	len = pcn->pcn_namelen + 1;
   2117 
   2118 	pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, pcn->pcn_cred);
   2119 	path = _GET_INPAYLOAD(ps, pm, char *);
   2120 	(void)strlcpy(path, name, len);
   2121 
   2122 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2123 		goto out;
   2124 
   2125 	perfuse_cache_flush(targ);
   2126 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
   2127 
   2128 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
   2129 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
   2130 
   2131 	/*
   2132 	 * The parent directory needs a sync
   2133 	 */
   2134 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
   2135 
   2136 #ifdef PERFUSE_DEBUG
   2137 	if (perfuse_diagflags & PDF_FILENAME)
   2138 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
   2139 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
   2140 			pcn->pcn_name);
   2141 #endif
   2142 	ps->ps_destroy_msg(pm);
   2143 	error = 0;
   2144 
   2145 out:
   2146 	node_rele(opc);
   2147 	node_rele(targ);
   2148 	return error;
   2149 }
   2150 
   2151 int
   2152 perfuse_node_link(struct puffs_usermount *pu, puffs_cookie_t opc,
   2153 	puffs_cookie_t targ, const struct puffs_cn *pcn)
   2154 {
   2155 	struct perfuse_state *ps;
   2156 	perfuse_msg_t *pm;
   2157 	const char *name;
   2158 	size_t len;
   2159 	struct puffs_node *pn;
   2160 	struct fuse_link_in *fli;
   2161 	int error;
   2162 
   2163 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   2164 		return ENOENT;
   2165 
   2166 	node_ref(opc);
   2167 	node_ref(targ);
   2168 	ps = puffs_getspecific(pu);
   2169 	pn = (struct puffs_node *)targ;
   2170 	name = pcn->pcn_name;
   2171 	len =  sizeof(*fli) + pcn->pcn_namelen + 1;
   2172 
   2173 	pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, pcn->pcn_cred);
   2174 	fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
   2175 	fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_nodeid;
   2176 	(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
   2177 
   2178 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2179 		goto out;
   2180 
   2181 	ps->ps_destroy_msg(pm);
   2182 	error = 0;
   2183 
   2184 out:
   2185 	node_rele(opc);
   2186 	node_rele(targ);
   2187 	return error;
   2188 }
   2189 
   2190 int
   2191 perfuse_node_rename(struct puffs_usermount *pu, puffs_cookie_t opc,
   2192 	puffs_cookie_t src, const struct puffs_cn *pcn_src,
   2193 	puffs_cookie_t targ_dir, puffs_cookie_t targ,
   2194 	const struct puffs_cn *pcn_targ)
   2195 {
   2196 	struct perfuse_state *ps;
   2197 	struct perfuse_node_data *dstdir_pnd;
   2198 	perfuse_msg_t *pm;
   2199 	struct fuse_rename_in *fri;
   2200 	const char *newname;
   2201 	const char *oldname;
   2202 	char *np;
   2203 	int error;
   2204 	size_t len;
   2205 	size_t newname_len;
   2206 	size_t oldname_len;
   2207 
   2208 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
   2209 	    (PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
   2210 	    (PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
   2211 		return ENOENT;
   2212 
   2213 	node_ref(opc);
   2214 	node_ref(src);
   2215 
   2216 	/*
   2217 	 * Await for all operations on the deleted node to drain,
   2218 	 * as the filesystem may be confused to have it deleted
   2219 	 * during a getattr
   2220 	 */
   2221 	if ((struct puffs_node *)targ != NULL) {
   2222 		node_ref(targ);
   2223 		while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
   2224 			requeue_request(pu, targ, PCQ_AFTERXCHG);
   2225 	} else {
   2226 		while (PERFUSE_NODE_DATA(src)->pnd_inxchg)
   2227 			requeue_request(pu, src, PCQ_AFTERXCHG);
   2228 	}
   2229 
   2230 	ps = puffs_getspecific(pu);
   2231 	newname =  pcn_targ->pcn_name;
   2232 	newname_len = pcn_targ->pcn_namelen + 1;
   2233 	oldname =  pcn_src->pcn_name;
   2234 	oldname_len = pcn_src->pcn_namelen + 1;
   2235 
   2236 	len = sizeof(*fri) + oldname_len + newname_len;
   2237 	pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, pcn_targ->pcn_cred);
   2238 	fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
   2239 	fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid;
   2240 	np = (char *)(void *)(fri + 1);
   2241 	(void)strlcpy(np, oldname, oldname_len);
   2242 	np += oldname_len;
   2243 	(void)strlcpy(np, newname, newname_len);
   2244 
   2245 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2246 		goto out;
   2247 
   2248 
   2249 	/*
   2250 	 * Record new parent nodeid
   2251 	 */
   2252 	dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
   2253 	PERFUSE_NODE_DATA(src)->pnd_parent_nodeid = dstdir_pnd->pnd_nodeid;
   2254 
   2255 	if (opc != targ_dir)
   2256 		dstdir_pnd->pnd_flags |= PND_DIRTY;
   2257 
   2258 	if (strcmp(newname, "..") != 0)
   2259 		(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name,
   2260 		    newname, MAXPATHLEN);
   2261 	else
   2262 		PERFUSE_NODE_DATA(src)->pnd_name[0] = 0; /* forget name */
   2263 
   2264 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
   2265 
   2266 	if ((struct puffs_node *)targ != NULL) {
   2267 		perfuse_cache_flush(targ);
   2268 		PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
   2269 	}
   2270 
   2271 #ifdef PERFUSE_DEBUG
   2272 	if (perfuse_diagflags & PDF_FILENAME)
   2273 		DPRINTF("%s: nodeid = 0x%"PRIx64" file = \"%s\" renamed \"%s\" "
   2274 			"nodeid = 0x%"PRIx64" -> nodeid = 0x%"PRIx64" \"%s\"\n",
   2275 	 		__func__, PERFUSE_NODE_DATA(src)->pnd_nodeid,
   2276 			pcn_src->pcn_name, pcn_targ->pcn_name,
   2277 			PERFUSE_NODE_DATA(opc)->pnd_nodeid,
   2278 			PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid,
   2279 			perfuse_node_path(ps, targ_dir));
   2280 #endif
   2281 
   2282 	ps->ps_destroy_msg(pm);
   2283 	error = 0;
   2284 
   2285 out:
   2286 	node_rele(opc);
   2287 	node_rele(src);
   2288 	if ((struct puffs_node *)targ != NULL)
   2289 		node_rele(targ);
   2290 
   2291 	return error;
   2292 }
   2293 
   2294 int
   2295 perfuse_node_mkdir(struct puffs_usermount *pu, puffs_cookie_t opc,
   2296 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
   2297 	const struct vattr *vap)
   2298 {
   2299 	struct perfuse_state *ps;
   2300 	perfuse_msg_t *pm;
   2301 	struct fuse_mkdir_in *fmi;
   2302 	const char *path;
   2303 	size_t len;
   2304 	int error;
   2305 
   2306 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   2307 		return ENOENT;
   2308 
   2309 	node_ref(opc);
   2310 	ps = puffs_getspecific(pu);
   2311 	path = pcn->pcn_name;
   2312 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
   2313 
   2314 	pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, pcn->pcn_cred);
   2315 	fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
   2316 	fmi->mode = vap->va_mode;
   2317 	fmi->umask = 0; 	/* Seems unused by libfuse? */
   2318 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
   2319 
   2320 	error = node_mk_common(pu, opc, pni, pcn, pm);
   2321 
   2322 	node_rele(opc);
   2323 	return error;
   2324 }
   2325 
   2326 
   2327 int
   2328 perfuse_node_rmdir(struct puffs_usermount *pu, puffs_cookie_t opc,
   2329 	puffs_cookie_t targ, const struct puffs_cn *pcn)
   2330 {
   2331 	struct perfuse_state *ps;
   2332 	struct perfuse_node_data *pnd;
   2333 	perfuse_msg_t *pm;
   2334 	char *path;
   2335 	const char *name;
   2336 	size_t len;
   2337 	int error;
   2338 
   2339 	pnd = PERFUSE_NODE_DATA(opc);
   2340 
   2341 	if ((pnd->pnd_flags & PND_REMOVED) ||
   2342 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
   2343 		return ENOENT;
   2344 
   2345 	node_ref(opc);
   2346 	node_ref(targ);
   2347 
   2348 	/*
   2349 	 * Await for all operations on the deleted node to drain,
   2350 	 * as the filesystem may be confused to have it deleted
   2351 	 * during a getattr
   2352 	 */
   2353 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
   2354 		requeue_request(pu, targ, PCQ_AFTERXCHG);
   2355 
   2356 	ps = puffs_getspecific(pu);
   2357 	name = pcn->pcn_name;
   2358 	len = pcn->pcn_namelen + 1;
   2359 
   2360 	pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, pcn->pcn_cred);
   2361 	path = _GET_INPAYLOAD(ps, pm, char *);
   2362 	(void)strlcpy(path, name, len);
   2363 
   2364 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2365 		goto out;
   2366 
   2367 	perfuse_cache_flush(targ);
   2368 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
   2369 
   2370 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
   2371 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
   2372 
   2373 	/*
   2374 	 * The parent directory needs a sync
   2375 	 */
   2376 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
   2377 
   2378 #ifdef PERFUSE_DEBUG
   2379 	if (perfuse_diagflags & PDF_FILENAME)
   2380 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
   2381 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
   2382 			perfuse_node_path(ps, targ));
   2383 #endif
   2384 	ps->ps_destroy_msg(pm);
   2385 	error = 0;
   2386 
   2387 out:
   2388 	node_rele(opc);
   2389 	node_rele(targ);
   2390 	return error;
   2391 }
   2392 
   2393 /* vap is unused */
   2394 /* ARGSUSED4 */
   2395 int
   2396 perfuse_node_symlink(struct puffs_usermount *pu, puffs_cookie_t opc,
   2397 	struct puffs_newinfo *pni, const struct puffs_cn *pcn_src,
   2398 	const struct vattr *vap, const char *link_target)
   2399 {
   2400 	struct perfuse_state *ps;
   2401 	perfuse_msg_t *pm;
   2402 	char *np;
   2403 	const char *path;
   2404 	size_t path_len;
   2405 	size_t linkname_len;
   2406 	size_t len;
   2407 	int error;
   2408 
   2409 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   2410 		return ENOENT;
   2411 
   2412 	node_ref(opc);
   2413 	ps = puffs_getspecific(pu);
   2414 	path = pcn_src->pcn_name;
   2415 	path_len = pcn_src->pcn_namelen + 1;
   2416 	linkname_len = strlen(link_target) + 1;
   2417 	len = path_len + linkname_len;
   2418 
   2419 	pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, pcn_src->pcn_cred);
   2420 	np = _GET_INPAYLOAD(ps, pm, char *);
   2421 	(void)strlcpy(np, path, path_len);
   2422 	np += path_len;
   2423 	(void)strlcpy(np, link_target, linkname_len);
   2424 
   2425 	error = node_mk_common(pu, opc, pni, pcn_src, pm);
   2426 
   2427 	node_rele(opc);
   2428 	return error;
   2429 }
   2430 
   2431 /* ARGSUSED4 */
   2432 int
   2433 perfuse_node_readdir(struct puffs_usermount *pu, puffs_cookie_t opc,
   2434 	struct dirent *dent, off_t *readoff, size_t *reslen,
   2435 	const struct puffs_cred *pcr, int *eofflag, off_t *cookies,
   2436 	size_t *ncookies)
   2437 {
   2438 	perfuse_msg_t *pm;
   2439 	uint64_t fh;
   2440 	struct perfuse_state *ps;
   2441 	struct perfuse_node_data *pnd;
   2442 	struct fuse_read_in *fri;
   2443 	struct fuse_out_header *foh;
   2444 	struct fuse_dirent *fd;
   2445 	size_t foh_len;
   2446 	int error;
   2447 	size_t fd_maxlen;
   2448 
   2449 	error = 0;
   2450 	node_ref(opc);
   2451 	ps = puffs_getspecific(pu);
   2452 
   2453 	/*
   2454 	 * readdir state is kept at node level, and several readdir
   2455 	 * requests can be issued at the same time on the same node.
   2456 	 * We need to queue requests so that only one is in readdir
   2457 	 * code at the same time.
   2458 	 */
   2459 	pnd = PERFUSE_NODE_DATA(opc);
   2460 	while (pnd->pnd_flags & PND_INREADDIR)
   2461 		requeue_request(pu, opc, PCQ_READDIR);
   2462 	pnd->pnd_flags |= PND_INREADDIR;
   2463 
   2464 #ifdef PERFUSE_DEBUG
   2465 	if (perfuse_diagflags & PDF_READDIR)
   2466 		DPRINTF("%s: READDIR opc = %p enter critical section\n",
   2467 			__func__, (void *)opc);
   2468 #endif
   2469 	/*
   2470 	 * Re-initialize pnd->pnd_fd_cookie on the first readdir for a node
   2471 	 */
   2472 	if (*readoff == 0)
   2473 		pnd->pnd_fd_cookie = 0;
   2474 
   2475 	/*
   2476 	 * Do we already have the data bufered?
   2477 	 */
   2478 	if (pnd->pnd_dirent != NULL)
   2479 		goto out;
   2480 	pnd->pnd_dirent_len = 0;
   2481 
   2482 	/*
   2483 	 * It seems NetBSD can call readdir without open first
   2484 	 * libfuse will crash if it is done that way, hence open first.
   2485 	 */
   2486 	if (!(pnd->pnd_flags & PND_OPEN)) {
   2487 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
   2488 			goto out;
   2489 	}
   2490 
   2491 	fh = perfuse_get_fh(opc, FREAD);
   2492 
   2493 #ifdef PERFUSE_DEBUG
   2494 	if (perfuse_diagflags & PDF_FH)
   2495 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
   2496 			"rfh = 0x%"PRIx64"\n", __func__, (void *)opc,
   2497 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fh);
   2498 #endif
   2499 
   2500 	pnd->pnd_all_fd = NULL;
   2501 	pnd->pnd_all_fd_len = 0;
   2502 	fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
   2503 
   2504 	do {
   2505 		size_t fd_len;
   2506 		char *afdp;
   2507 
   2508 		pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
   2509 
   2510 		/*
   2511 		 * read_flags, lock_owner and flags are unused in libfuse
   2512 		 */
   2513 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
   2514 		fri->fh = fh;
   2515 		fri->offset = pnd->pnd_fd_cookie;
   2516 		fri->size = (uint32_t)fd_maxlen;
   2517 		fri->read_flags = 0;
   2518 		fri->lock_owner = 0;
   2519 		fri->flags = 0;
   2520 
   2521 		if ((error = xchg_msg(pu, opc, pm,
   2522 				      UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2523 			goto out;
   2524 
   2525 		/*
   2526 		 * There are many puffs_framebufs calls later,
   2527 		 * therefore foh will not be valid for a long time.
   2528 		 * Just get the length and forget it.
   2529 		 */
   2530 		foh = GET_OUTHDR(ps, pm);
   2531 		foh_len = foh->len;
   2532 
   2533 		/*
   2534 		 * Empty read: we reached the end of the buffer.
   2535 		 */
   2536 		if (foh_len == sizeof(*foh)) {
   2537 			ps->ps_destroy_msg(pm);
   2538 			*eofflag = 1;
   2539 			break;
   2540 		}
   2541 
   2542 		/*
   2543 		 * Check for corrupted message.
   2544 		 */
   2545 		if (foh_len < sizeof(*foh) + sizeof(*fd)) {
   2546 			ps->ps_destroy_msg(pm);
   2547 			DWARNX("readdir reply too short");
   2548 			error = EIO;
   2549 			goto out;
   2550 		}
   2551 
   2552 
   2553 		fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
   2554 		fd_len = foh_len - sizeof(*foh);
   2555 
   2556 		pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
   2557 					  pnd->pnd_all_fd_len + fd_len);
   2558 		if (pnd->pnd_all_fd  == NULL)
   2559 			DERR(EX_OSERR, "%s: malloc failed", __func__);
   2560 
   2561 		afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
   2562 		(void)memcpy(afdp, fd, fd_len);
   2563 
   2564 		pnd->pnd_all_fd_len += fd_len;
   2565 
   2566 		/*
   2567 		 * The fd->off field is used as a cookie for
   2568 		 * resuming the next readdir() where this one was left.
   2569 	 	 */
   2570 		pnd->pnd_fd_cookie = readdir_last_cookie(fd, fd_len);
   2571 
   2572 		ps->ps_destroy_msg(pm);
   2573 	} while (1 /* CONSTCOND */);
   2574 
   2575 	if (pnd->pnd_all_fd != NULL) {
   2576 		if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd,
   2577 				   pnd->pnd_all_fd_len) == -1)
   2578 			error = EIO;
   2579 	}
   2580 
   2581 out:
   2582 	if (pnd->pnd_all_fd != NULL) {
   2583 		free(pnd->pnd_all_fd);
   2584 		pnd->pnd_all_fd = NULL;
   2585 		pnd->pnd_all_fd_len = 0;
   2586 	}
   2587 
   2588 	if (error == 0)
   2589 		readdir_buffered(opc, dent, readoff, reslen);
   2590 
   2591 	/*
   2592 	 * Schedule queued readdir requests
   2593 	 */
   2594 	pnd->pnd_flags &= ~PND_INREADDIR;
   2595 	(void)dequeue_requests(opc, PCQ_READDIR, DEQUEUE_ALL);
   2596 
   2597 #ifdef PERFUSE_DEBUG
   2598 	if (perfuse_diagflags & PDF_READDIR)
   2599 		DPRINTF("%s: READDIR opc = %p exit critical section\n",
   2600 			__func__, (void *)opc);
   2601 #endif
   2602 
   2603 	node_rele(opc);
   2604 	return error;
   2605 }
   2606 
   2607 int
   2608 perfuse_node_readlink(struct puffs_usermount *pu, puffs_cookie_t opc,
   2609 	const struct puffs_cred *pcr, char *linkname, size_t *linklen)
   2610 {
   2611 	struct perfuse_state *ps;
   2612 	perfuse_msg_t *pm;
   2613 	int error;
   2614 	size_t len;
   2615 	struct fuse_out_header *foh;
   2616 
   2617 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   2618 		return ENOENT;
   2619 
   2620 	node_ref(opc);
   2621 	ps = puffs_getspecific(pu);
   2622 
   2623 	pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
   2624 
   2625 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2626 		goto out;
   2627 
   2628 	foh = GET_OUTHDR(ps, pm);
   2629 	len = foh->len - sizeof(*foh);
   2630 	if (len > *linklen)
   2631 		DERRX(EX_PROTOCOL, "path len = %zd too long", len);
   2632 	if (len == 0)
   2633 		DERRX(EX_PROTOCOL, "path len = %zd too short", len);
   2634 
   2635 	(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
   2636 
   2637 	/*
   2638 	 * FUSE filesystems return a NUL terminated string, we
   2639 	 * do not want the trailing \0
   2640 	 */
   2641 	while (len > 0 && linkname[len - 1] == '\0')
   2642 		len--;
   2643 
   2644 	*linklen = len;
   2645 
   2646 	ps->ps_destroy_msg(pm);
   2647 	error = 0;
   2648 
   2649 out:
   2650 	node_rele(opc);
   2651 	return error;
   2652 }
   2653 
   2654 int
   2655 perfuse_node_reclaim(struct puffs_usermount *pu, puffs_cookie_t opc)
   2656 {
   2657 	struct perfuse_state *ps;
   2658 	perfuse_msg_t *pm;
   2659 	struct perfuse_node_data *pnd;
   2660 	struct fuse_forget_in *ffi;
   2661 	int nlookup;
   2662 	struct timespec now;
   2663 
   2664 	if (opc == 0)
   2665 		return 0;
   2666 
   2667 	ps = puffs_getspecific(pu);
   2668 	pnd = PERFUSE_NODE_DATA(opc);
   2669 
   2670 	/*
   2671 	 * Never forget the root.
   2672 	 */
   2673 	if (pnd->pnd_nodeid == FUSE_ROOT_ID)
   2674 		return 0;
   2675 
   2676 	/*
   2677 	 * There is a race condition between reclaim and lookup.
   2678 	 * When looking up an already known node, the kernel cannot
   2679 	 * hold a reference on the result until it gets the PUFFS
   2680 	 * reply. It mayy therefore reclaim the node after the
   2681 	 * userland looked it up, and before it gets the reply.
   2682 	 * On rely, the kernel re-creates the node, but at that
   2683 	 * time the node has been reclaimed in userland.
   2684 	 *
   2685 	 * In order to avoid this, we refuse reclaiming nodes that
   2686 	 * are too young since the last lookup - and that we do
   2687 	 * not have removed on our own, of course.
   2688 	 */
   2689 	if (clock_gettime(CLOCK_REALTIME, &now) != 0)
   2690 		DERR(EX_OSERR, "clock_gettime failed");
   2691 
   2692 	if (timespeccmp(&pnd->pnd_cn_expire, &now, >) &&
   2693 	    !(pnd->pnd_flags & PND_REMOVED)) {
   2694 		if (!(pnd->pnd_flags & PND_NODELEAK)) {
   2695 			ps->ps_nodeleakcount++;
   2696 			pnd->pnd_flags |= PND_NODELEAK;
   2697 		}
   2698 		DWARNX("possible leaked node:: opc = %p \"%s\"",
   2699 		       opc, pnd->pnd_name);
   2700 		return 0;
   2701 	}
   2702 
   2703 	node_ref(opc);
   2704 	pnd->pnd_flags |= PND_RECLAIMED;
   2705 	pnd->pnd_puffs_nlookup--;
   2706 	nlookup = pnd->pnd_puffs_nlookup;
   2707 
   2708 #ifdef PERFUSE_DEBUG
   2709 	if (perfuse_diagflags & PDF_RECLAIM)
   2710 		DPRINTF("%s (nodeid %"PRId64") reclaimed\n",
   2711 			perfuse_node_path(ps, opc), pnd->pnd_nodeid);
   2712 #endif
   2713 
   2714 #ifdef PERFUSE_DEBUG
   2715 	if (perfuse_diagflags & PDF_RECLAIM)
   2716 		DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, nlookup = %d "
   2717 			"%s%s%s%s, pending ops:%s%s%s\n",
   2718 		        perfuse_node_path(ps, opc), pnd->pnd_nodeid,
   2719 		        pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
   2720 			pnd->pnd_puffs_nlookup,
   2721 			pnd->pnd_flags & PND_OPEN ? "open " : "not open",
   2722 			pnd->pnd_flags & PND_RFH ? "r" : "",
   2723 			pnd->pnd_flags & PND_WFH ? "w" : "",
   2724 			pnd->pnd_flags & PND_BUSY ? "" : " none",
   2725 			pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
   2726 			pnd->pnd_flags & PND_INWRITE ? " write" : "",
   2727 			pnd->pnd_flags & PND_INOPEN ? " open" : "");
   2728 #endif
   2729 	/*
   2730 	 * Make sure it is not looked up again
   2731 	 */
   2732 	if (!(pnd->pnd_flags & PND_REMOVED))
   2733 		perfuse_cache_flush(opc);
   2734 
   2735 	/*
   2736 	 * Purge any activity on the node, while checking
   2737 	 * that it remains eligible for a reclaim.
   2738 	 */
   2739 	while (pnd->pnd_ref > 1)
   2740 		requeue_request(pu, opc, PCQ_REF);
   2741 
   2742 	/*
   2743 	 * reclaim cancel?
   2744 	 */
   2745 	if (pnd->pnd_puffs_nlookup > nlookup) {
   2746 		pnd->pnd_flags &= ~PND_RECLAIMED;
   2747 		perfuse_node_cache(ps, opc);
   2748 		node_rele(opc);
   2749 		return 0;
   2750 	}
   2751 
   2752 
   2753 #ifdef PERFUSE_DEBUG
   2754 	if ((pnd->pnd_flags & PND_OPEN) ||
   2755 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
   2756 		DERRX(EX_SOFTWARE, "%s: opc = %p \"%s\": still open",
   2757 		      __func__, opc, pnd->pnd_name);
   2758 
   2759 	if ((pnd->pnd_flags & PND_BUSY) ||
   2760 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
   2761 		DERRX(EX_SOFTWARE, "%s: opc = %p: queued operations",
   2762 		      __func__, opc);
   2763 
   2764 	if (pnd->pnd_inxchg != 0)
   2765 		DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
   2766 		      __func__, opc);
   2767 #endif
   2768 
   2769 	/*
   2770 	 * Send the FORGET message
   2771 	 *
   2772 	 * ps_new_msg() is called with NULL creds, which will
   2773 	 * be interpreted as FUSE superuser. This is obviously
   2774 	 * fine since we operate with kernel creds here.
   2775 	 */
   2776 	pm = ps->ps_new_msg(pu, opc, FUSE_FORGET,
   2777 		      sizeof(*ffi), NULL);
   2778 	ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
   2779 	ffi->nlookup = pnd->pnd_fuse_nlookup;
   2780 
   2781 	/*
   2782 	 * No reply is expected, pm is freed in xchg_msg
   2783 	 */
   2784 	(void)xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, no_reply);
   2785 
   2786 	perfuse_destroy_pn(pu, opc);
   2787 
   2788 	return 0;
   2789 }
   2790 
   2791 int
   2792 perfuse_node_inactive(struct puffs_usermount *pu, puffs_cookie_t opc)
   2793 {
   2794 	struct perfuse_node_data *pnd;
   2795 	int error;
   2796 
   2797 	if (opc == 0)
   2798 		return 0;
   2799 
   2800 	node_ref(opc);
   2801 	pnd = PERFUSE_NODE_DATA(opc);
   2802 
   2803 	if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
   2804 		goto out;
   2805 
   2806 	/*
   2807 	 * Make sure all operation are finished
   2808 	 * There can be an ongoing write. Other
   2809 	 * operation wait for all data before
   2810 	 * the close/inactive.
   2811 	 */
   2812 	while (pnd->pnd_flags & PND_INWRITE)
   2813 		requeue_request(pu, opc, PCQ_AFTERWRITE);
   2814 
   2815 	/*
   2816 	 * The inactive operation may be cancelled,
   2817 	 * If no open is in progress, set PND_INOPEN
   2818 	 * so that a new open will be queued.
   2819 	 */
   2820 	if (pnd->pnd_flags & PND_INOPEN)
   2821 		goto out;
   2822 
   2823 	pnd->pnd_flags |= PND_INOPEN;
   2824 
   2825 	/*
   2826 	 * Sync data
   2827 	 */
   2828 	if (pnd->pnd_flags & PND_DIRTY) {
   2829 		if ((error = perfuse_node_fsync(pu, opc, NULL, 0, 0, 0)) != 0)
   2830 			DWARN("%s: perfuse_node_fsync failed error = %d",
   2831 			      __func__, error);
   2832 	}
   2833 
   2834 
   2835 	/*
   2836 	 * Close handles
   2837 	 */
   2838 	if (pnd->pnd_flags & PND_WFH) {
   2839 		if ((error = perfuse_node_close_common(pu, opc, FWRITE)) != 0)
   2840 			DWARN("%s: close write FH failed error = %d",
   2841 			      __func__, error);
   2842 	}
   2843 
   2844 	if (pnd->pnd_flags & PND_RFH) {
   2845 		if ((error = perfuse_node_close_common(pu, opc, FREAD)) != 0)
   2846 			DWARN("%s: close read FH failed error = %d",
   2847 			      __func__, error);
   2848 	}
   2849 
   2850 	/*
   2851 	 * This will cause a reclaim to be sent
   2852 	 */
   2853 	if (pnd->pnd_flags & PND_REMOVED)
   2854 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
   2855 
   2856 	/*
   2857 	 * Schedule awaiting operations
   2858 	 */
   2859 	pnd->pnd_flags &= ~PND_INOPEN;
   2860 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
   2861 
   2862 	/*
   2863 	 * errors are ignored, since the kernel ignores the return code.
   2864 	 */
   2865 out:
   2866 	node_rele(opc);
   2867 	return 0;
   2868 }
   2869 
   2870 
   2871 /* ARGSUSED0 */
   2872 int
   2873 perfuse_node_print(struct puffs_usermount *pu, puffs_cookie_t opc)
   2874 {
   2875 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
   2876 	return 0;
   2877 }
   2878 
   2879 int
   2880 perfuse_node_pathconf(struct puffs_usermount *pu, puffs_cookie_t opc,
   2881 	int name, int *retval)
   2882 {
   2883 	perfuse_msg_t *pm;
   2884 	struct perfuse_state *ps;
   2885 	struct fuse_statfs_out *fso;
   2886 	int error = 0;
   2887 
   2888 	/*
   2889 	 * Static values copied from UFS
   2890 	 * in src/sys/ufs/ufs/ufs_vnops.c
   2891 	 */
   2892 	switch (name) {
   2893 	case _PC_LINK_MAX:
   2894 		*retval = LINK_MAX;
   2895 		break;
   2896 	case _PC_PATH_MAX:
   2897 		*retval = PATH_MAX;
   2898 		break;
   2899 	case _PC_PIPE_BUF:
   2900 		*retval = PIPE_BUF;
   2901 		break;
   2902 	case _PC_CHOWN_RESTRICTED:
   2903 		*retval = 1;
   2904 		break;
   2905 	case _PC_NO_TRUNC:
   2906 		*retval = 1;
   2907 		break;
   2908 	case _PC_SYNC_IO:
   2909 		*retval = 1;
   2910 		break;
   2911 	case _PC_FILESIZEBITS:
   2912 		*retval = 42;
   2913 		break;
   2914 	case _PC_SYMLINK_MAX:
   2915 		*retval = MAXPATHLEN;
   2916 		break;
   2917 	case _PC_2_SYMLINKS:
   2918 		*retval = 1;
   2919 		break;
   2920 	case _PC_NAME_MAX:
   2921 		ps = puffs_getspecific(pu);
   2922 		pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
   2923 
   2924 		error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply);
   2925 		if (error != 0)
   2926 			return error;
   2927 
   2928 		fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
   2929 		*retval = fso->st.namelen;
   2930 
   2931 		ps->ps_destroy_msg(pm);
   2932 
   2933 		break;
   2934 	default:
   2935 		DWARN("Unimplemented pathconf for name = %d", name);
   2936 		error = ENOSYS;
   2937 		break;
   2938 	}
   2939 
   2940 	return error;
   2941 }
   2942 
   2943 int
   2944 perfuse_node_advlock(struct puffs_usermount *pu, puffs_cookie_t opc,
   2945 	void *id, int op, struct flock *fl, int flags)
   2946 {
   2947 	struct perfuse_state *ps;
   2948 	int fop;
   2949 	perfuse_msg_t *pm;
   2950 	uint64_t fh;
   2951 	struct fuse_lk_in *fli;
   2952 	struct fuse_out_header *foh;
   2953 	struct fuse_lk_out *flo;
   2954 	uint32_t owner;
   2955 	size_t len;
   2956 	int error;
   2957 
   2958 	node_ref(opc);
   2959 
   2960 	/*
   2961 	 * Make sure we do have a filehandle, as the FUSE filesystem
   2962 	 * expect one. E.g.: if we provide none, GlusterFS logs an error
   2963 	 * "0-glusterfs-fuse: xl is NULL"
   2964 	 *
   2965 	 * We need the read file handle if the file is open read only,
   2966 	 * in order to support shared locks on read-only files.
   2967 	 * NB: The kernel always sends advlock for read-only
   2968 	 * files at exit time when the process used lock, see
   2969 	 * sys_exit -> exit1 -> fd_free -> fd_close -> VOP_ADVLOCK
   2970 	 */
   2971 	if ((fh = perfuse_get_fh(opc, FREAD)) == FUSE_UNKNOWN_FH) {
   2972 		error = EBADF;
   2973 		goto out;
   2974 	}
   2975 
   2976 	ps = puffs_getspecific(pu);
   2977 
   2978 	if (op == F_GETLK)
   2979 		fop = FUSE_GETLK;
   2980 	else
   2981 		fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
   2982 
   2983 	/*
   2984 	 * XXX ps_new_msg() is called with NULL creds, which will
   2985 	 * be interpreted as FUSE superuser. We have no way to
   2986 	 * know the requesting process' credential, but since advlock()
   2987 	 * is supposed to operate on a file that has been open(),
   2988 	 * permission should have already been checked at open() time.
   2989 	 */
   2990 	pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
   2991 	fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
   2992 	fli->fh = fh;
   2993 	fli->owner = (uint64_t)(vaddr_t)id;
   2994 	fli->lk.start = fl->l_start;
   2995 	fli->lk.end = fl->l_start + fl->l_len;
   2996 	fli->lk.type = fl->l_type;
   2997 	fli->lk.pid = fl->l_pid;
   2998 	fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
   2999 
   3000 	owner = (uint32_t)(vaddr_t)id;
   3001 
   3002 #ifdef PERFUSE_DEBUG
   3003 	if (perfuse_diagflags & PDF_FH)
   3004 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
   3005 			__func__, (void *)opc,
   3006 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fli->fh);
   3007 #endif
   3008 
   3009 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   3010 		goto out;
   3011 
   3012 	foh = GET_OUTHDR(ps, pm);
   3013 	len = foh->len - sizeof(*foh);
   3014 
   3015 	/*
   3016 	 * Save or clear the lock
   3017 	 */
   3018 	switch (op) {
   3019 	case F_GETLK:
   3020 		if (len != sizeof(*flo))
   3021 			DERRX(EX_SOFTWARE,
   3022 			      "%s: Unexpected lock reply len %zd",
   3023 			      __func__, len);
   3024 
   3025 		flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
   3026 		fl->l_start = flo->lk.start;
   3027 		fl->l_len = flo->lk.end - flo->lk.start;
   3028 		fl->l_pid = flo->lk.pid;
   3029 		fl->l_type = flo->lk.type;
   3030 		fl->l_whence = SEEK_SET;	/* libfuse hardcodes it */
   3031 
   3032 		PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
   3033 		break;
   3034 	case F_UNLCK:
   3035 		owner = 0;
   3036 		/* FALLTHROUGH */
   3037 	case F_SETLK:
   3038 		/* FALLTHROUGH */
   3039 	case F_SETLKW:
   3040 		if (error != 0)
   3041 			PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
   3042 
   3043 		if (len != 0)
   3044 			DERRX(EX_SOFTWARE,
   3045 			      "%s: Unexpected unlock reply len %zd",
   3046 			      __func__, len);
   3047 
   3048 		break;
   3049 	default:
   3050 		DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
   3051 		break;
   3052 	}
   3053 
   3054 	ps->ps_destroy_msg(pm);
   3055 	error = 0;
   3056 
   3057 out:
   3058 	node_rele(opc);
   3059 	return error;
   3060 }
   3061 
   3062 int
   3063 perfuse_node_read(struct puffs_usermount *pu, puffs_cookie_t opc, uint8_t *buf,
   3064 	off_t offset, size_t *resid, const struct puffs_cred *pcr, int ioflag)
   3065 {
   3066 	struct perfuse_state *ps;
   3067 	struct perfuse_node_data *pnd;
   3068 	const struct vattr *vap;
   3069 	perfuse_msg_t *pm;
   3070 	struct fuse_read_in *fri;
   3071 	struct fuse_out_header *foh;
   3072 	size_t readen;
   3073 	int error;
   3074 
   3075 	ps = puffs_getspecific(pu);
   3076 	pnd = PERFUSE_NODE_DATA(opc);
   3077 	vap = puffs_pn_getvap((struct puffs_node *)opc);
   3078 
   3079 	/*
   3080 	 * NetBSD turns that into a getdents(2) output
   3081 	 * We just do a EISDIR as this feature is of little use.
   3082 	 */
   3083 	if (vap->va_type == VDIR)
   3084 		return EISDIR;
   3085 
   3086 	do {
   3087 		size_t max_read;
   3088 
   3089 		max_read = ps->ps_max_readahead - sizeof(*foh);
   3090 		/*
   3091 		 * flags may be set to FUSE_READ_LOCKOWNER
   3092 		 * if lock_owner is provided.
   3093 		 */
   3094 		pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
   3095 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
   3096 		fri->fh = perfuse_get_fh(opc, FREAD);
   3097 		fri->offset = offset;
   3098 		fri->size = (uint32_t)MIN(*resid, max_read);
   3099 		fri->read_flags = 0; /* XXX Unused by libfuse? */
   3100 		fri->lock_owner = pnd->pnd_lock_owner;
   3101 		fri->flags = 0;
   3102 		fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
   3103 
   3104 #ifdef PERFUSE_DEBUG
   3105 	if (perfuse_diagflags & PDF_FH)
   3106 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
   3107 			__func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
   3108 #endif
   3109 		error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
   3110 		if (error  != 0)
   3111 			return error;
   3112 
   3113 		foh = GET_OUTHDR(ps, pm);
   3114 		readen = foh->len - sizeof(*foh);
   3115 
   3116 #ifdef PERFUSE_DEBUG
   3117 		if (readen > *resid)
   3118 			DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
   3119 			      __func__, readen);
   3120 #endif
   3121 
   3122 		(void)memcpy(buf,  _GET_OUTPAYLOAD(ps, pm, char *), readen);
   3123 
   3124 		buf += readen;
   3125 		offset += readen;
   3126 		*resid -= readen;
   3127 
   3128 		ps->ps_destroy_msg(pm);
   3129 	} while ((*resid != 0) && (readen != 0));
   3130 
   3131 	if (ioflag & (IO_SYNC|IO_DSYNC))
   3132 		ps->ps_syncreads++;
   3133 	else
   3134 		ps->ps_asyncreads++;
   3135 
   3136 	return 0;
   3137 }
   3138 
   3139 int
   3140 perfuse_node_write(struct puffs_usermount *pu, puffs_cookie_t opc,
   3141 	uint8_t *buf, off_t offset, size_t *resid,
   3142 	const struct puffs_cred *pcr, int ioflag)
   3143 {
   3144 	return perfuse_node_write2(pu, opc, buf, offset, resid, pcr, ioflag, 0);
   3145 }
   3146 
   3147 /* ARGSUSED7 */
   3148 int
   3149 perfuse_node_write2(struct puffs_usermount *pu, puffs_cookie_t opc,
   3150 	uint8_t *buf, off_t offset, size_t *resid,
   3151 	const struct puffs_cred *pcr, int ioflag, int xflag)
   3152 {
   3153 	struct perfuse_state *ps;
   3154 	struct perfuse_node_data *pnd;
   3155 	struct vattr *vap;
   3156 	perfuse_msg_t *pm;
   3157 	struct fuse_write_in *fwi;
   3158 	struct fuse_write_out *fwo;
   3159 	size_t data_len;
   3160 	size_t payload_len;
   3161 	size_t written;
   3162 	int inresize;
   3163 	int error;
   3164 
   3165 	ps = puffs_getspecific(pu);
   3166 	pnd = PERFUSE_NODE_DATA(opc);
   3167 	vap = puffs_pn_getvap((struct puffs_node *)opc);
   3168 	written = 0;
   3169 	inresize = 0;
   3170 	error = 0;
   3171 
   3172 	if (vap->va_type == VDIR)
   3173 		return EISDIR;
   3174 
   3175 	node_ref(opc);
   3176 
   3177 	/*
   3178 	 * We need to queue write requests in order to avoid
   3179 	 * dequeueing PCQ_AFTERWRITE when there are pending writes.
   3180 	 */
   3181 	while (pnd->pnd_flags & PND_INWRITE)
   3182 		requeue_request(pu, opc, PCQ_WRITE);
   3183 	pnd->pnd_flags |= PND_INWRITE;
   3184 
   3185 	/*
   3186 	 * Serialize size access, see comment in perfuse_node_setattr().
   3187 	 */
   3188 	if ((u_quad_t)offset + *resid > vap->va_size) {
   3189 		while (pnd->pnd_flags & PND_INRESIZE)
   3190 			requeue_request(pu, opc, PCQ_RESIZE);
   3191 		pnd->pnd_flags |= PND_INRESIZE;
   3192 		inresize = 1;
   3193 	}
   3194 
   3195 	/*
   3196 	 * append flag: re-read the file size so that
   3197 	 * we get the latest value.
   3198 	 */
   3199 	if (ioflag & PUFFS_IO_APPEND) {
   3200 		if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
   3201 			goto out;
   3202 
   3203 		offset = vap->va_size;
   3204 	}
   3205 
   3206 #ifdef PERFUSE_DEBUG
   3207 	if (perfuse_diagflags & PDF_RESIZE)
   3208 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__,
   3209 			(void *)opc, vap->va_size);
   3210 #endif
   3211 
   3212 	do {
   3213 		size_t max_write;
   3214 		/*
   3215 		 * There is a writepage flag when data
   3216 		 * is aligned to page size. Use it for
   3217 		 * everything but the data after the last
   3218 		 * page boundary.
   3219 		 */
   3220 		max_write = ps->ps_max_write - sizeof(*fwi);
   3221 
   3222 		data_len = MIN(*resid, max_write);
   3223 		if (data_len > (size_t)sysconf(_SC_PAGESIZE))
   3224 			data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
   3225 
   3226 		payload_len = data_len + sizeof(*fwi);
   3227 
   3228 		/*
   3229 		 * flags may be set to FUSE_WRITE_CACHE (XXX usage?)
   3230 		 * or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
   3231 		 * write_flags is set to 1 for writepage.
   3232 		 */
   3233 		pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
   3234 		fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
   3235 		fwi->fh = perfuse_get_fh(opc, FWRITE);
   3236 		fwi->offset = offset;
   3237 		fwi->size = (uint32_t)data_len;
   3238 		fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
   3239 		fwi->lock_owner = pnd->pnd_lock_owner;
   3240 		fwi->flags = 0;
   3241 		fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
   3242 		fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
   3243 		(void)memcpy((fwi + 1), buf, data_len);
   3244 
   3245 
   3246 #ifdef PERFUSE_DEBUG
   3247 		if (perfuse_diagflags & PDF_FH)
   3248 			DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
   3249 				"fh = 0x%"PRIx64"\n", __func__,
   3250 				(void *)opc, pnd->pnd_nodeid, fwi->fh);
   3251 #endif
   3252 		if ((error = xchg_msg(pu, opc, pm,
   3253 				      sizeof(*fwo), wait_reply)) != 0)
   3254 			goto out;
   3255 
   3256 		fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
   3257 		written = fwo->size;
   3258 		ps->ps_destroy_msg(pm);
   3259 
   3260 #ifdef PERFUSE_DEBUG
   3261 		if (written > *resid)
   3262 			DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
   3263 			      __func__, written);
   3264 #endif
   3265 		*resid -= written;
   3266 		offset += written;
   3267 		buf += written;
   3268 
   3269 	} while (*resid != 0);
   3270 
   3271 	/*
   3272 	 * puffs_ops(3) says
   3273 	 *  "everything must be written or an error will be generated"
   3274 	 */
   3275 	if (*resid != 0)
   3276 		error = EFBIG;
   3277 
   3278 #ifdef PERFUSE_DEBUG
   3279 	if (perfuse_diagflags & PDF_RESIZE) {
   3280 		if (offset > (off_t)vap->va_size)
   3281 			DPRINTF("<< %s %p %" PRIu64 " -> %lld\n", __func__,
   3282 				(void *)opc, vap->va_size, (long long)offset);
   3283 		else
   3284 			DPRINTF("<< %s %p \n", __func__, (void *)opc);
   3285 	}
   3286 #endif
   3287 
   3288 	/*
   3289 	 * Update file size if we wrote beyond the end
   3290 	 */
   3291 	if (offset > (off_t)vap->va_size)
   3292 		vap->va_size = offset;
   3293 
   3294 	if (inresize) {
   3295 #ifdef PERFUSE_DEBUG
   3296 		if (!(pnd->pnd_flags & PND_INRESIZE))
   3297 			DERRX(EX_SOFTWARE, "file write grow without resize");
   3298 #endif
   3299 		pnd->pnd_flags &= ~PND_INRESIZE;
   3300 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
   3301 	}
   3302 
   3303 
   3304 	/*
   3305 	 * Statistics
   3306 	 */
   3307 	if (ioflag & (IO_SYNC|IO_DSYNC))
   3308 		ps->ps_syncwrites++;
   3309 	else
   3310 		ps->ps_asyncwrites++;
   3311 
   3312 	/*
   3313 	 * Remember to sync the file
   3314 	 */
   3315 	pnd->pnd_flags |= PND_DIRTY;
   3316 
   3317 #ifdef PERFUSE_DEBUG
   3318 	if (perfuse_diagflags & PDF_SYNC)
   3319 		DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
   3320 			__func__, (void*)opc, perfuse_node_path(ps, opc));
   3321 #endif
   3322 
   3323 out:
   3324 	/*
   3325 	 * VOP_PUTPAGE causes FAF write where kernel does not
   3326 	 * check operation result. At least warn if it failed.
   3327 	 */
   3328 #ifdef PUFFS_WRITE_FAF
   3329 	if (error && (xflag & PUFFS_WRITE_FAF))
   3330 		DWARN("Data loss caused by FAF write failed on \"%s\"",
   3331 		      pnd->pnd_name);
   3332 #endif /* PUFFS_WRITE_FAF */
   3333 
   3334 	/*
   3335 	 * If there are no more queued write, we can resume
   3336 	 * an operation awaiting write completion.
   3337 	 */
   3338 	pnd->pnd_flags &= ~PND_INWRITE;
   3339 	if (dequeue_requests(opc, PCQ_WRITE, 1) == 0)
   3340 		(void)dequeue_requests(opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
   3341 
   3342 	node_rele(opc);
   3343 	return error;
   3344 }
   3345 
   3346 /* ARGSUSED0 */
   3347 void
   3348 perfuse_cache_write(struct puffs_usermount *pu, puffs_cookie_t opc, size_t size,
   3349 	struct puffs_cacherun *runs)
   3350 {
   3351 	return;
   3352 }
   3353 
   3354 /* ARGSUSED4 */
   3355 int
   3356 perfuse_node_getextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   3357 	int attrns, const char *attrname, size_t *attrsize, uint8_t *attr,
   3358 	size_t *resid, const struct puffs_cred *pcr)
   3359 {
   3360 	struct perfuse_state *ps;
   3361 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
   3362 	perfuse_msg_t *pm;
   3363 	struct fuse_getxattr_in *fgi;
   3364 	struct fuse_getxattr_out *fgo;
   3365 	struct fuse_out_header *foh;
   3366 	size_t attrnamelen;
   3367 	size_t len;
   3368 	char *np;
   3369 	int error;
   3370 
   3371 	/* system namespace attrs are not accessible to non root users */
   3372 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
   3373 		return EPERM;
   3374 
   3375 	node_ref(opc);
   3376 	ps = puffs_getspecific(pu);
   3377 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
   3378 	attrnamelen = strlen(attrname) + 1;
   3379 	len = sizeof(*fgi) + attrnamelen;
   3380 
   3381 	pm = ps->ps_new_msg(pu, opc, FUSE_GETXATTR, len, pcr);
   3382 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
   3383 	fgi->size = (unsigned int)((resid != NULL) ? *resid : 0);
   3384 	np = (char *)(void *)(fgi + 1);
   3385 	(void)strlcpy(np, attrname, attrnamelen);
   3386 
   3387 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   3388 		goto out;
   3389 
   3390 	/*
   3391 	 * We just get fuse_getattr_out with list size if we requested
   3392 	 * a null size.
   3393 	 */
   3394 	if (resid == NULL) {
   3395 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
   3396 
   3397 		if (attrsize != NULL)
   3398 			*attrsize = fgo->size;
   3399 
   3400 		ps->ps_destroy_msg(pm);
   3401 		error = 0;
   3402 		goto out;
   3403 	}
   3404 
   3405 	/*
   3406 	 * And with a non null requested size, we get the list just
   3407 	 * after the header
   3408 	 */
   3409 	foh = GET_OUTHDR(ps, pm);
   3410 	np = (char *)(void *)(foh + 1);
   3411 	len = foh->len - sizeof(*foh);
   3412 
   3413 	if (attrsize != NULL)
   3414 		*attrsize = len;
   3415 
   3416 	if (resid != NULL) {
   3417 		if (*resid < len) {
   3418 			error = ERANGE;
   3419 			ps->ps_destroy_msg(pm);
   3420 			goto out;
   3421 		}
   3422 
   3423 		(void)memcpy(attr, np, len);
   3424 		*resid -= len;
   3425 	}
   3426 
   3427 	ps->ps_destroy_msg(pm);
   3428 	error = 0;
   3429 
   3430 out:
   3431 	node_rele(opc);
   3432 	return error;
   3433 }
   3434 
   3435 int
   3436 perfuse_node_setextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   3437 	int attrns, const char *attrname, uint8_t *attr, size_t *resid,
   3438 	const struct puffs_cred *pcr)
   3439 {
   3440 	struct perfuse_state *ps;
   3441 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
   3442 	perfuse_msg_t *pm;
   3443 	struct fuse_setxattr_in *fsi;
   3444 	size_t attrnamelen;
   3445 	size_t len;
   3446 	char *np;
   3447 	int error;
   3448 
   3449 	/* system namespace attrs are not accessible to non root users */
   3450 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
   3451 		return EPERM;
   3452 
   3453 	node_ref(opc);
   3454 	ps = puffs_getspecific(pu);
   3455 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
   3456 	attrnamelen = strlen(attrname) + 1;
   3457 	len = sizeof(*fsi) + attrnamelen + *resid;
   3458 
   3459 	pm = ps->ps_new_msg(pu, opc, FUSE_SETXATTR, len, pcr);
   3460 	fsi = GET_INPAYLOAD(ps, pm, fuse_setxattr_in);
   3461 	fsi->size = (unsigned int)*resid;
   3462 	fsi->flags = 0;
   3463 	np = (char *)(void *)(fsi + 1);
   3464 	(void)strlcpy(np, attrname, attrnamelen);
   3465 	np += attrnamelen;
   3466 	(void)memcpy(np, (char *)attr, *resid);
   3467 
   3468 	if ((error = xchg_msg(pu, opc, pm,
   3469 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
   3470 		goto out;
   3471 
   3472 	ps->ps_destroy_msg(pm);
   3473 	*resid = 0;
   3474 	error = 0;
   3475 
   3476 out:
   3477 	node_rele(opc);
   3478 	return error;
   3479 }
   3480 
   3481 /* ARGSUSED2 */
   3482 int
   3483 perfuse_node_listextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   3484 	int attrns, size_t *attrsize, uint8_t *attrs, size_t *resid, int flag,
   3485 	const struct puffs_cred *pcr)
   3486 {
   3487 	struct perfuse_state *ps;
   3488 	perfuse_msg_t *pm;
   3489 	struct fuse_getxattr_in *fgi;
   3490 	struct fuse_getxattr_out *fgo;
   3491 	struct fuse_out_header *foh;
   3492 	char *np;
   3493 	size_t len, puffs_len, i, attrlen, outlen;
   3494 	int error;
   3495 
   3496 	/* system namespace attrs are not accessible to non root users */
   3497 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
   3498 		return EPERM;
   3499 
   3500 	node_ref(opc);
   3501 
   3502 	ps = puffs_getspecific(pu);
   3503 	len = sizeof(*fgi);
   3504 
   3505 	pm = ps->ps_new_msg(pu, opc, FUSE_LISTXATTR, len, pcr);
   3506 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
   3507 	if (resid != NULL)
   3508 		fgi->size = (unsigned int)*resid;
   3509 	else
   3510 		fgi->size = 0;
   3511 
   3512 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   3513 		goto out;
   3514 
   3515 	/*
   3516 	 * We just get fuse_getattr_out with list size if we requested
   3517 	 * a null size.
   3518 	 */
   3519 	if (resid == NULL) {
   3520 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
   3521 
   3522 		if (attrsize != NULL)
   3523 			*attrsize = fgo->size;
   3524 
   3525 		ps->ps_destroy_msg(pm);
   3526 
   3527 		error = 0;
   3528 		goto out;
   3529 	}
   3530 
   3531 	/*
   3532 	 * And with a non null requested size, we get the list just
   3533 	 * after the header
   3534 	 */
   3535 	foh = GET_OUTHDR(ps, pm);
   3536 	np = (char *)(void *)(foh + 1);
   3537 	puffs_len = foh->len - sizeof(*foh);
   3538 
   3539 	if (attrsize != NULL)
   3540 		*attrsize = puffs_len;
   3541 
   3542 	if (attrs != NULL) {
   3543 		if (*resid < puffs_len) {
   3544 			error = ERANGE;
   3545 			ps->ps_destroy_msg(pm);
   3546 			goto out;
   3547 		}
   3548 
   3549 		outlen = 0;
   3550 
   3551 		for (i = 0; i < puffs_len; i += attrlen + 1) {
   3552 			attrlen = strlen(np + i);
   3553 
   3554 			/*
   3555 			 * Filter attributes per namespace
   3556 			 */
   3557 			if (!perfuse_ns_match(attrns, np + i))
   3558 				continue;
   3559 
   3560 #ifdef PUFFS_EXTATTR_LIST_LENPREFIX
   3561 			/*
   3562 			 * Convert the FUSE reply to length prefixed strings
   3563 			 * if this is what the kernel wants.
   3564 			 */
   3565 			if (flag & PUFFS_EXTATTR_LIST_LENPREFIX) {
   3566 				(void)memcpy(attrs + outlen + 1,
   3567 					     np + i, attrlen);
   3568 				*(attrs + outlen) = (uint8_t)attrlen;
   3569 			} else
   3570 #endif /* PUFFS_EXTATTR_LIST_LENPREFIX */
   3571 			(void)memcpy(attrs + outlen, np + i, attrlen + 1);
   3572 			outlen += attrlen + 1;
   3573 		}
   3574 
   3575 		*resid -= outlen;
   3576 	}
   3577 
   3578 	ps->ps_destroy_msg(pm);
   3579 	error = 0;
   3580 
   3581 out:
   3582 	node_rele(opc);
   3583 	return error;
   3584 }
   3585 
   3586 int
   3587 perfuse_node_deleteextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   3588 	int attrns, const char *attrname, const struct puffs_cred *pcr)
   3589 {
   3590 	struct perfuse_state *ps;
   3591 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
   3592 	perfuse_msg_t *pm;
   3593 	size_t attrnamelen;
   3594 	char *np;
   3595 	int error;
   3596 
   3597 	/* system namespace attrs are not accessible to non root users */
   3598 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
   3599 		return EPERM;
   3600 
   3601 	node_ref(opc);
   3602 
   3603 	ps = puffs_getspecific(pu);
   3604 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
   3605 	attrnamelen = strlen(attrname) + 1;
   3606 
   3607 	pm = ps->ps_new_msg(pu, opc, FUSE_REMOVEXATTR, attrnamelen, pcr);
   3608 	np = _GET_INPAYLOAD(ps, pm, char *);
   3609 	(void)strlcpy(np, attrname, attrnamelen);
   3610 
   3611 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
   3612 
   3613 	ps->ps_destroy_msg(pm);
   3614 
   3615 	node_rele(opc);
   3616 	return error;
   3617 }
   3618