Home | History | Annotate | Line # | Download | only in libperfuse
      1 /*  $NetBSD: ops.c,v 1.92 2023/06/24 05:18:13 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 #define PN_ISDIR(opc) \
    109 	(puffs_pn_getvap((struct puffs_node *)opc)->va_type == VDIR)
    110 
    111 #if 0
    112 static void
    113 print_node(const char *func, puffs_cookie_t opc)
    114 {
    115 	struct puffs_node *pn;
    116 	struct perfuse_node_data *pnd;
    117 	struct vattr *vap;
    118 
    119 	pn = (struct puffs_node *)opc;
    120 	pnd = PERFUSE_NODE_DATA(opc);
    121 	vap = &pn->pn_va;
    122 
    123 	printf("%s: \"%s\", opc = %p, nodeid = 0x%"PRIx64" ino = %"PRIu64"\n",
    124 	       func, pnd->pnd_name, opc, pnd->pnd_nodeid, vap->va_fileid);
    125 
    126 	return;
    127 }
    128 #endif /* PERFUSE_DEBUG */
    129 
    130 int
    131 perfuse_node_close_common(struct puffs_usermount *pu, puffs_cookie_t opc,
    132 	int mode)
    133 {
    134 	struct perfuse_state *ps;
    135 	perfuse_msg_t *pm;
    136 	int op;
    137 	uint64_t fh;
    138 	struct fuse_release_in *fri;
    139 	struct perfuse_node_data *pnd;
    140 	struct puffs_node *pn;
    141 	int error;
    142 
    143 	ps = puffs_getspecific(pu);
    144 	pn = (struct puffs_node *)opc;
    145 	pnd = PERFUSE_NODE_DATA(pn);
    146 
    147 	if (PN_ISDIR(opc)) {
    148 		op = FUSE_RELEASEDIR;
    149 		mode = FREAD;
    150 	} else {
    151 		op = FUSE_RELEASE;
    152 	}
    153 
    154 	/*
    155 	 * Destroy the filehandle before sending the
    156 	 * request to the FUSE filesystem, otherwise
    157 	 * we may get a second close() while we wait
    158 	 * for the reply, and we would end up closing
    159 	 * the same fh twice instead of closng both.
    160 	 */
    161 	fh = perfuse_get_fh(opc, mode);
    162 	perfuse_destroy_fh(pn, fh);
    163 
    164 	/*
    165 	 * release_flags may be set to FUSE_RELEASE_FLUSH
    166 	 * to flush locks. lock_owner must be set in that case
    167 	 *
    168 	 * ps_new_msg() is called with NULL creds, which will
    169 	 * be interpreted as FUSE superuser. We come here from the
    170 	 * inactive method, which provides no creds, but obviously
    171 	 * runs with kernel privilege.
    172 	 */
    173 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*fri), NULL);
    174 	fri = GET_INPAYLOAD(ps, pm, fuse_release_in);
    175 	fri->fh = fh;
    176 	fri->flags = 0;
    177 	fri->release_flags = 0;
    178 	fri->lock_owner = pnd->pnd_lock_owner;
    179 	fri->flags = (fri->lock_owner != 0) ? FUSE_RELEASE_FLUSH : 0;
    180 
    181 #ifdef PERFUSE_DEBUG
    182 	if (perfuse_diagflags & PDF_FH)
    183 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
    184 			 __func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
    185 #endif
    186 
    187 	if ((error = xchg_msg(pu, opc, pm,
    188 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
    189 		DERRX(EX_SOFTWARE, "%s: freed fh = 0x%"PRIx64" but filesystem "
    190 		      "returned error = %d", __func__, fh, error);
    191 
    192 	ps->ps_destroy_msg(pm);
    193 
    194 	return 0;
    195 }
    196 
    197 static int
    198 xchg_msg(struct puffs_usermount *pu, puffs_cookie_t opc, perfuse_msg_t *pm,
    199 	size_t len, enum perfuse_xchg_pb_reply wait)
    200 {
    201 	struct perfuse_state *ps;
    202 	struct perfuse_node_data *pnd;
    203 	struct perfuse_trace *pt = NULL;
    204 	int error;
    205 
    206 	ps = puffs_getspecific(pu);
    207 	pnd = NULL;
    208 	if ((struct puffs_node *)opc != NULL)
    209 		pnd = PERFUSE_NODE_DATA(opc);
    210 
    211 #ifdef PERFUSE_DEBUG
    212 	if ((perfuse_diagflags & PDF_FILENAME) && (opc != 0))
    213 		DPRINTF("file = \"%s\", ino = %"PRIu64" flags = 0x%x\n",
    214 			perfuse_node_path(ps, opc),
    215 			((struct puffs_node *)opc)->pn_va.va_fileid,
    216 			PERFUSE_NODE_DATA(opc)->pnd_flags);
    217 #endif
    218 	ps->ps_xchgcount++;
    219 	if (pnd)
    220 		pnd->pnd_inxchg++;
    221 
    222 	/*
    223 	 * Record FUSE call start if requested
    224 	 */
    225 	if (perfuse_diagflags & PDF_TRACE)
    226 		pt = perfuse_trace_begin(ps, opc, pm);
    227 
    228 	/*
    229 	 * Do actual FUSE exchange
    230 	 */
    231 	if ((error = ps->ps_xchg_msg(pu, pm, len, wait)) != 0)
    232 		ps->ps_destroy_msg(pm);
    233 
    234 	/*
    235 	 * Record FUSE call end if requested
    236 	 */
    237 	if (pt != NULL)
    238 		perfuse_trace_end(ps, pt, error);
    239 
    240 	ps->ps_xchgcount--;
    241 	if (pnd) {
    242 		pnd->pnd_inxchg--;
    243 		(void)dequeue_requests(opc, PCQ_AFTERXCHG, DEQUEUE_ALL);
    244 	}
    245 
    246 	return error;
    247 }
    248 
    249 static int
    250 mode_access(puffs_cookie_t opc, const struct puffs_cred *pcr, mode_t mode)
    251 {
    252 	struct puffs_node *pn;
    253 	struct vattr *va;
    254 
    255 	/*
    256 	 * pcr is NULL for self open through fsync or readdir.
    257 	 * In both case, access control is useless, as it was
    258 	 * done before, at open time.
    259 	 */
    260 	if (pcr == NULL)
    261 		return 0;
    262 
    263 	pn = (struct puffs_node *)opc;
    264 	va = puffs_pn_getvap(pn);
    265 	return puffs_access(va->va_type, va->va_mode,
    266 			    va->va_uid, va->va_gid,
    267 			    mode, pcr);
    268 }
    269 
    270 static int
    271 sticky_access(puffs_cookie_t opc, struct puffs_node *targ,
    272 	      const struct puffs_cred *pcr)
    273 {
    274 	uid_t uid;
    275 	int sticky, owner, parent_owner;
    276 
    277 	/*
    278 	 * This covers the case where the kernel requests a DELETE
    279 	 * or RENAME on its own, and where puffs_cred_getuid would
    280 	 * return -1. While such a situation should not happen,
    281 	 * we allow it here.
    282 	 *
    283 	 * This also allows root to tamper with other users' files
    284 	 * that have the sticky bit.
    285 	 */
    286 	if (puffs_cred_isjuggernaut(pcr))
    287 		return 0;
    288 
    289 	if (puffs_cred_getuid(pcr, &uid) != 0)
    290 		DERRX(EX_SOFTWARE, "puffs_cred_getuid fails in %s", __func__);
    291 
    292 	sticky = puffs_pn_getvap(opc)->va_mode & S_ISTXT;
    293 	owner = puffs_pn_getvap(targ)->va_uid == uid;
    294 	parent_owner = puffs_pn_getvap(opc)->va_uid == uid;
    295 
    296 	if (sticky && !owner && !parent_owner)
    297 		return EPERM;
    298 
    299 	return 0;
    300 }
    301 
    302 
    303 static void
    304 fuse_attr_to_vap(struct perfuse_state *ps, struct vattr *vap,
    305 	struct fuse_attr *fa)
    306 {
    307 	vap->va_type = IFTOVT(fa->mode);
    308 	vap->va_mode = fa->mode & ALLPERMS;
    309 	vap->va_nlink = fa->nlink;
    310 	vap->va_uid = fa->uid;
    311 	vap->va_gid = fa->gid;
    312 	vap->va_fsid = (long)ps->ps_fsid;
    313 	vap->va_fileid = fa->ino;
    314 	vap->va_size = fa->size;
    315 	vap->va_blocksize = fa->blksize;
    316 	vap->va_atime.tv_sec = (time_t)fa->atime;
    317 	vap->va_atime.tv_nsec = (long) fa->atimensec;
    318 	vap->va_mtime.tv_sec = (time_t)fa->mtime;
    319 	vap->va_mtime.tv_nsec = (long)fa->mtimensec;
    320 	vap->va_ctime.tv_sec = (time_t)fa->ctime;
    321 	vap->va_ctime.tv_nsec = (long)fa->ctimensec;
    322 	vap->va_birthtime.tv_sec = 0;
    323 	vap->va_birthtime.tv_nsec = 0;
    324 	vap->va_gen = 0;
    325 	vap->va_flags = 0;
    326 	vap->va_rdev = fa->rdev;
    327 	vap->va_bytes = fa->blocks * S_BLKSIZE;
    328 	vap->va_filerev = (u_quad_t)PUFFS_VNOVAL;
    329 	vap->va_vaflags = 0;
    330 
    331 	if (vap->va_blocksize == 0)
    332 		vap->va_blocksize = DEV_BSIZE;
    333 
    334 	if (vap->va_size == (size_t)PUFFS_VNOVAL) /* XXX */
    335 		vap->va_size = 0;
    336 
    337 	return;
    338 }
    339 
    340 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
    341 static void
    342 perfuse_newinfo_setttl(struct puffs_newinfo *pni,
    343     struct puffs_node *pn, struct fuse_entry_out *feo,
    344     struct fuse_attr_out *fao)
    345 {
    346 #ifdef PERFUSE_DEBUG
    347 	if ((feo == NULL) && (fao == NULL))
    348 		DERRX(EX_SOFTWARE, "%s: feo and fao NULL", __func__);
    349 
    350 	if ((feo != NULL) && (fao != NULL))
    351 		DERRX(EX_SOFTWARE, "%s: feo and fao != NULL", __func__);
    352 #endif /* PERFUSE_DEBUG */
    353 
    354 	if (fao != NULL) {
    355 		struct timespec va_ttl;
    356 
    357 		va_ttl.tv_sec = fao->attr_valid;
    358 		va_ttl.tv_nsec = fao->attr_valid_nsec;
    359 
    360 		puffs_newinfo_setvattl(pni, &va_ttl);
    361 	}
    362 
    363 	if (feo != NULL) {
    364 		struct timespec va_ttl;
    365 		struct timespec cn_ttl;
    366 		struct timespec now;
    367 		struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(pn);
    368 
    369 		va_ttl.tv_sec = feo->attr_valid;
    370 		va_ttl.tv_nsec = feo->attr_valid_nsec;
    371 		cn_ttl.tv_sec = feo->entry_valid;
    372 		cn_ttl.tv_nsec = feo->entry_valid_nsec;
    373 
    374 		puffs_newinfo_setvattl(pni, &va_ttl);
    375 		puffs_newinfo_setcnttl(pni, &cn_ttl);
    376 
    377 		if (clock_gettime(CLOCK_REALTIME, &now) != 0)
    378 			DERR(EX_OSERR, "clock_gettime failed");
    379 
    380                 timespecadd(&now, &cn_ttl, &pnd->pnd_cn_expire);
    381 	}
    382 
    383 	return;
    384 }
    385 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
    386 
    387 static int
    388 node_lookup_common(struct puffs_usermount *pu, puffs_cookie_t opc,
    389 	struct puffs_newinfo *pni, const char *path,
    390 	const struct puffs_cred *pcr, struct puffs_node **pnp)
    391 {
    392 	struct perfuse_state *ps;
    393 	struct perfuse_node_data *oldpnd;
    394 	perfuse_msg_t *pm;
    395 	struct fuse_entry_out *feo;
    396 	struct puffs_node *pn;
    397 	size_t len;
    398 	int error;
    399 
    400 	/*
    401 	 * Prevent further lookups if the parent was removed
    402 	 */
    403 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
    404 		return ESTALE;
    405 
    406 	if (pnp == NULL)
    407 		DERRX(EX_SOFTWARE, "pnp must be != NULL");
    408 
    409 	ps = puffs_getspecific(pu);
    410 
    411 #ifdef PERFUSE_DEBUG
    412 	if (perfuse_diagflags & PDF_FILENAME)
    413 		DPRINTF("%s: opc = %p, file = \"%s\" looking up \"%s\"\n",
    414 			__func__, (void *)opc,
    415 			perfuse_node_path(ps, opc), path);
    416 
    417 	if (strcmp(path, ".") == 0)
    418 		DERRX(EX_SOFTWARE, "unexpected dot-lookup");
    419 
    420 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_RECLAIMED)
    421 		DERRX(EX_SOFTWARE,
    422 		      "looking up reclaimed node opc = %p, name = \"%s\"",
    423 		      opc, path);
    424 
    425 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_INVALID)
    426 		DERRX(EX_SOFTWARE,
    427 		      "looking up freed node opc = %p, name = \"%s\"",
    428 		      opc, path);
    429 #endif /* PERFUSE_DEBUG */
    430 
    431 	len = strlen(path) + 1;
    432 	pm = ps->ps_new_msg(pu, opc, FUSE_LOOKUP, len, pcr);
    433 	(void)strlcpy(_GET_INPAYLOAD(ps, pm, char *), path, len);
    434 
    435 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
    436 		return error;
    437 
    438 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
    439 
    440 	/*
    441 	 * Starting with ABI 7.4, inode number 0 means ENOENT,
    442 	 * with entry_valid / entry_valid_nsec giving negative
    443 	 * cache timeout (which we do not implement yet).
    444 	 */
    445 	if (feo->attr.ino == 0) {
    446 		ps->ps_destroy_msg(pm);
    447 		return ENOENT;
    448 	}
    449 
    450 	/*
    451 	 * Check for a known node, not reclaimed, with another name.
    452 	 * It may have been moved, or we can lookup ../
    453 	 */
    454 	if (((oldpnd = perfuse_node_bynodeid(ps, feo->nodeid)) != NULL) &&
    455 	    !(oldpnd->pnd_flags & PND_RECLAIMED)) {
    456 		/*
    457 		 * Save the new node name if not ..
    458 		 */
    459 		if (strncmp(path, "..", len) != 0)
    460 			(void)strlcpy(oldpnd->pnd_name,
    461 				      path, MAXPATHLEN);
    462 		pn = oldpnd->pnd_pn;
    463 
    464 	} else {
    465 		pn = perfuse_new_pn(pu, path, opc);
    466 		PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
    467 		perfuse_node_cache(ps, pn);
    468 	}
    469 
    470 #ifdef PERFUSE_DEBUG
    471 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_RECLAIMED)
    472 		DERRX(EX_SOFTWARE,
    473 		      "reclaimed in lookup opc = %p, name = \"%s\", ck = %p",
    474 		      opc, path, pn);
    475 
    476 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_INVALID)
    477 		DERRX(EX_SOFTWARE,
    478 		      "freed in lookup opc = %p, name = \"%s\", ck = %p",
    479 		      opc, path, pn);
    480 #endif /* PERFUSE_DEBUG */
    481 
    482 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
    483 	pn->pn_va.va_gen = (u_long)(feo->generation);
    484 	PERFUSE_NODE_DATA(pn)->pnd_fuse_nlookup++;
    485 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
    486 
    487 	*pnp = pn;
    488 
    489 #ifdef PERFUSE_DEBUG
    490 	if (perfuse_diagflags & PDF_FILENAME)
    491 		DPRINTF("%s: opc = %p, looked up opc = %p, "
    492 			"nodeid = 0x%"PRIx64" file = \"%s\"\n", __func__,
    493 			(void *)opc, pn, feo->nodeid, path);
    494 #endif
    495 
    496 	if (pni != NULL) {
    497 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
    498 		puffs_newinfo_setva(pni, &pn->pn_va);
    499 		perfuse_newinfo_setttl(pni, pn, feo, NULL);
    500 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
    501 		puffs_newinfo_setcookie(pni, pn);
    502 		puffs_newinfo_setvtype(pni, pn->pn_va.va_type);
    503 		puffs_newinfo_setsize(pni, (voff_t)pn->pn_va.va_size);
    504 		puffs_newinfo_setrdev(pni, pn->pn_va.va_rdev);
    505 	}
    506 
    507 	ps->ps_destroy_msg(pm);
    508 
    509 	return 0;
    510 }
    511 
    512 
    513 /*
    514  * Common code for methods that create objects:
    515  * perfuse_node_mkdir
    516  * perfuse_node_mknod
    517  * perfuse_node_symlink
    518  */
    519 static int
    520 node_mk_common(struct puffs_usermount *pu, puffs_cookie_t opc,
    521 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
    522 	perfuse_msg_t *pm)
    523 {
    524 	struct perfuse_state *ps;
    525 	struct puffs_node *pn;
    526 	struct fuse_entry_out *feo;
    527 	int error;
    528 
    529 	ps =  puffs_getspecific(pu);
    530 
    531 	if ((error = xchg_msg(pu, opc, pm, sizeof(*feo), wait_reply)) != 0)
    532 		return error;
    533 
    534 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
    535 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
    536 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
    537 
    538 	pn = perfuse_new_pn(pu, pcn->pcn_name, opc);
    539 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
    540 	PERFUSE_NODE_DATA(pn)->pnd_fuse_nlookup++;
    541 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
    542 	perfuse_node_cache(ps, pn);
    543 
    544 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
    545 	pn->pn_va.va_gen = (u_long)(feo->generation);
    546 
    547 	puffs_newinfo_setcookie(pni, pn);
    548 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
    549 	puffs_newinfo_setva(pni, &pn->pn_va);
    550 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
    551 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
    552 
    553 
    554 #ifdef PERFUSE_DEBUG
    555 	if (perfuse_diagflags & PDF_FILENAME)
    556 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
    557 			"nodeid = 0x%"PRIx64"\n",
    558 			__func__, (void *)pn, pcn->pcn_name,
    559 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid);
    560 #endif
    561 	ps->ps_destroy_msg(pm);
    562 
    563 	/* Parents is now dirty */
    564 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
    565 
    566 	return 0;
    567 }
    568 
    569 static uint64_t
    570 readdir_last_cookie(struct fuse_dirent *fd, size_t fd_len)
    571 {
    572 	size_t len;
    573 	size_t seen = 0;
    574 	char *ndp;
    575 
    576 	do {
    577 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
    578 		seen += len;
    579 
    580 		if (seen >= fd_len)
    581 			break;
    582 
    583 		ndp = (char *)(void *)fd + (size_t)len;
    584 		fd = (struct fuse_dirent *)(void *)ndp;
    585 	} while (1 /* CONSTCOND */);
    586 
    587 	return fd->off;
    588 }
    589 
    590 static ssize_t
    591 fuse_to_dirent(struct puffs_usermount *pu, puffs_cookie_t opc,
    592 	struct fuse_dirent *fd, size_t fd_len)
    593 {
    594 	struct dirent *dents;
    595 	size_t dents_len;
    596 	ssize_t written;
    597 	uint64_t fd_offset;
    598 	struct fuse_dirent *fd_base;
    599 	size_t len;
    600 
    601 	fd_base = fd;
    602 	fd_offset = 0;
    603 	written = 0;
    604 	dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
    605 	dents_len = (size_t)PERFUSE_NODE_DATA(opc)->pnd_dirent_len;
    606 
    607 	do {
    608 		char *ndp;
    609 		size_t reclen;
    610 		char name[MAXPATHLEN];
    611 
    612 		reclen = _DIRENT_RECLEN(dents, fd->namelen);
    613 
    614 		/*
    615 		 * Check we do not overflow the output buffer
    616 		 * struct fuse_dirent is bigger than struct dirent,
    617 		 * so we should always use fd_len and never reallocate
    618 		 * later.
    619 		 * If we have to reallocate, try to double the buffer
    620 		 * each time so that we do not have to do it too often.
    621 		 */
    622 		if (written + reclen > dents_len) {
    623 			if (dents_len == 0)
    624 				dents_len = fd_len;
    625 			else
    626 				dents_len =
    627 				   MAX(2 * dents_len, written + reclen);
    628 
    629 			dents = PERFUSE_NODE_DATA(opc)->pnd_dirent;
    630 			if ((dents = realloc(dents, dents_len)) == NULL)
    631 				DERR(EX_OSERR, "%s: malloc failed", __func__);
    632 
    633 			PERFUSE_NODE_DATA(opc)->pnd_dirent = dents;
    634 			PERFUSE_NODE_DATA(opc)->pnd_dirent_len = dents_len;
    635 
    636 			/*
    637 			 * (void *) for delint
    638 			 */
    639 			ndp = (char *)(void *)dents + written;
    640 			dents = (struct dirent *)(void *)ndp;
    641 		}
    642 
    643 		strncpy(name, fd->name, fd->namelen);
    644 		name[fd->namelen] = '\0';
    645 
    646 		/*
    647 		 * Filesystem was mounted without -o use_ino
    648 		 * Perform a lookup to find it.
    649 		 */
    650 		if (fd->ino == PERFUSE_UNKNOWN_INO) {
    651 			struct puffs_node *pn;
    652 			struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
    653 
    654 			if (strcmp(name, "..") == 0) {
    655 				/*
    656 				 * Avoid breaking out of fs
    657 				 * by lookup to .. on root
    658 				 */
    659 				if (pnd->pnd_nodeid == FUSE_ROOT_ID)
    660 					fd->ino = FUSE_ROOT_ID;
    661 				else
    662 					fd->ino = pnd->pnd_parent_nodeid;
    663 			} else if (strcmp(name, ".") == 0 ) {
    664 				fd->ino = pnd->pnd_nodeid;
    665 			} else {
    666 				int error;
    667 
    668 				error = node_lookup_common(pu, opc, NULL,
    669 							   name, NULL, &pn);
    670 				if (error != 0) {
    671 					DWARNX("node_lookup_common %s "
    672 					       "failed: %d", name, error);
    673 				} else {
    674 					fd->ino = pn->pn_va.va_fileid;
    675 					(void)perfuse_node_reclaim2(pu, pn, 1);
    676 				}
    677 			}
    678 		}
    679 
    680 		dents->d_fileno = fd->ino;
    681 		dents->d_reclen = (unsigned short)reclen;
    682 		dents->d_namlen = fd->namelen;
    683 		dents->d_type = fd->type;
    684 		strlcpy(dents->d_name, name, fd->namelen + 1);
    685 
    686 #ifdef PERFUSE_DEBUG
    687 		if (perfuse_diagflags & PDF_READDIR)
    688 			DPRINTF("%s: translated \"%s\" ino = %"PRIu64"\n",
    689 				__func__, dents->d_name, dents->d_fileno);
    690 #endif
    691 
    692 		dents = _DIRENT_NEXT(dents);
    693 		written += reclen;
    694 
    695 		/*
    696 		 * Move to the next record.
    697 		 * fd->off is not the offset, it is an opaque cookie
    698 		 * given by the filesystem to keep state across multiple
    699 		 * readdir() operation.
    700 		 * Use record alignment instead.
    701 		 */
    702 		len = FUSE_DIRENT_ALIGN(sizeof(*fd) + fd->namelen);
    703 #ifdef PERFUSE_DEBUG
    704 		if (perfuse_diagflags & PDF_READDIR)
    705 			DPRINTF("%s: record at %"PRId64"/0x%"PRIx64" "
    706 				"length = %zd/0x%zx. "
    707 				"next record at %"PRId64"/0x%"PRIx64" "
    708 				"max %zd/0x%zx\n",
    709 				__func__, fd_offset, fd_offset, len, len,
    710 				fd_offset + len, fd_offset + len,
    711 				fd_len, fd_len);
    712 #endif
    713 		fd_offset += len;
    714 
    715 		/*
    716 		 * Check if next record is still within the packet
    717 		 * If it is not, we reached the end of the buffer.
    718 		 */
    719 		if (fd_offset >= fd_len)
    720 			break;
    721 
    722 		/*
    723 		 * (void *) for delint
    724 		 */
    725 		ndp = (char *)(void *)fd_base + (size_t)fd_offset;
    726 		fd = (struct fuse_dirent *)(void *)ndp;
    727 
    728 	} while (1 /* CONSTCOND */);
    729 
    730 	/*
    731 	 * Adjust the dirent output length
    732 	 */
    733 	if (written != -1)
    734 		PERFUSE_NODE_DATA(opc)->pnd_dirent_len = written;
    735 
    736 	return written;
    737 }
    738 
    739 static void
    740 readdir_buffered(puffs_cookie_t opc, struct dirent *dent, off_t *readoff,
    741 	size_t *reslen)
    742 {
    743 	struct dirent *fromdent;
    744 	struct perfuse_node_data *pnd;
    745 	char *ndp;
    746 
    747 	pnd = PERFUSE_NODE_DATA(opc);
    748 
    749 	while (*readoff < pnd->pnd_dirent_len) {
    750 		/*
    751 		 * (void *) for delint
    752 		 */
    753 		ndp = (char *)(void *)pnd->pnd_dirent + (size_t)*readoff;
    754 		fromdent = (struct dirent *)(void *)ndp;
    755 
    756 		if (*reslen < _DIRENT_SIZE(fromdent))
    757 			break;
    758 
    759 		memcpy(dent, fromdent, _DIRENT_SIZE(fromdent));
    760 		*readoff += _DIRENT_SIZE(fromdent);
    761 		*reslen -= _DIRENT_SIZE(fromdent);
    762 
    763 		dent = _DIRENT_NEXT(dent);
    764 	}
    765 
    766 #ifdef PERFUSE_DEBUG
    767 	if (perfuse_diagflags & PDF_READDIR)
    768 		DPRINTF("%s: readoff = %"PRId64",  "
    769 			"pnd->pnd_dirent_len = %"PRId64"\n",
    770 			__func__, *readoff, pnd->pnd_dirent_len);
    771 #endif
    772 	if (*readoff >=  pnd->pnd_dirent_len) {
    773 		free(pnd->pnd_dirent);
    774 		pnd->pnd_dirent = NULL;
    775 		pnd->pnd_dirent_len = 0;
    776 	}
    777 
    778 	return;
    779 }
    780 
    781 
    782 static void
    783 node_ref(puffs_cookie_t opc)
    784 {
    785 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
    786 
    787 #ifdef PERFUSE_DEBUG
    788 	if (pnd->pnd_flags & PND_INVALID)
    789 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
    790 #endif /* PERFUSE_DEBUG */
    791 
    792 	pnd->pnd_ref++;
    793 	return;
    794 }
    795 
    796 static void
    797 node_rele(puffs_cookie_t opc)
    798 {
    799 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
    800 
    801 #ifdef PERFUSE_DEBUG
    802 	if (pnd->pnd_flags & PND_INVALID)
    803 		DERRX(EX_SOFTWARE, "Use of freed node opc = %p", opc);
    804 #endif /* PERFUSE_DEBUG */
    805 
    806 	pnd->pnd_ref--;
    807 
    808 	if (pnd->pnd_ref == 0)
    809 		(void)dequeue_requests(opc, PCQ_REF, DEQUEUE_ALL);
    810 
    811 	return;
    812 }
    813 
    814 static void
    815 requeue_request(struct puffs_usermount *pu, puffs_cookie_t opc,
    816 	enum perfuse_qtype type)
    817 {
    818 	struct perfuse_cc_queue pcq;
    819 	struct perfuse_node_data *pnd;
    820 
    821 	pnd = PERFUSE_NODE_DATA(opc);
    822 	pcq.pcq_type = type;
    823 	pcq.pcq_cc = puffs_cc_getcc(pu);
    824 	TAILQ_INSERT_TAIL(&pnd->pnd_pcq, &pcq, pcq_next);
    825 
    826 #ifdef PERFUSE_DEBUG
    827 	if (perfuse_diagflags & PDF_REQUEUE)
    828 		DPRINTF("%s: REQUEUE opc = %p, pcc = %p (%s)\n",
    829 		        __func__, (void *)opc, pcq.pcq_cc,
    830 			perfuse_qtypestr[type]);
    831 #endif
    832 
    833 	puffs_cc_yield(pcq.pcq_cc);
    834 	TAILQ_REMOVE(&pnd->pnd_pcq, &pcq, pcq_next);
    835 
    836 #ifdef PERFUSE_DEBUG
    837 	if (perfuse_diagflags & PDF_REQUEUE)
    838 		DPRINTF("%s: RESUME opc = %p, pcc = %p (%s)\n",
    839 		        __func__, (void *)opc, pcq.pcq_cc,
    840 			perfuse_qtypestr[type]);
    841 #endif
    842 
    843 	return;
    844 }
    845 
    846 static int
    847 dequeue_requests(puffs_cookie_t opc, enum perfuse_qtype type, int max)
    848 {
    849 	struct perfuse_cc_queue *pcq;
    850 	struct perfuse_node_data *pnd;
    851 	int dequeued;
    852 
    853 	pnd = PERFUSE_NODE_DATA(opc);
    854 	dequeued = 0;
    855 	TAILQ_FOREACH(pcq, &pnd->pnd_pcq, pcq_next) {
    856 		if (pcq->pcq_type != type)
    857 			continue;
    858 
    859 #ifdef PERFUSE_DEBUG
    860 		if (perfuse_diagflags & PDF_REQUEUE)
    861 			DPRINTF("%s: SCHEDULE opc = %p, pcc = %p (%s)\n",
    862 				__func__, (void *)opc, pcq->pcq_cc,
    863 				 perfuse_qtypestr[type]);
    864 #endif
    865 		puffs_cc_schedule(pcq->pcq_cc);
    866 
    867 		if (++dequeued == max)
    868 			break;
    869 	}
    870 
    871 #ifdef PERFUSE_DEBUG
    872 	if (perfuse_diagflags & PDF_REQUEUE)
    873 		DPRINTF("%s: DONE  opc = %p\n", __func__, (void *)opc);
    874 #endif
    875 
    876 	return dequeued;
    877 }
    878 
    879 void
    880 perfuse_fs_init(struct puffs_usermount *pu)
    881 {
    882 	struct perfuse_state *ps;
    883 	perfuse_msg_t *pm;
    884 	struct fuse_init_in *fii;
    885 	struct fuse_init_out *fio;
    886 	int error;
    887 
    888 	ps = puffs_getspecific(pu);
    889 
    890         if (puffs_mount(pu, ps->ps_target, ps->ps_mountflags, ps->ps_root) != 0)
    891                 DERR(EX_OSERR, "%s: puffs_mount failed", __func__);
    892 
    893 	/*
    894 	 * Linux 2.6.34.1 sends theses flags:
    895 	 * FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC
    896 	 * FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK
    897 	 *
    898 	 * Linux also sets max_readahead at 32 pages (128 kB)
    899 	 *
    900 	 * ps_new_msg() is called with NULL creds, which will
    901 	 * be interpreted as FUSE superuser.
    902 	 */
    903 	pm = ps->ps_new_msg(pu, 0, FUSE_INIT, sizeof(*fii), NULL);
    904 	fii = GET_INPAYLOAD(ps, pm, fuse_init_in);
    905 	fii->major = FUSE_KERNEL_VERSION;
    906 	fii->minor = FUSE_KERNEL_MINOR_VERSION;
    907 	fii->max_readahead = (unsigned int)(32 * sysconf(_SC_PAGESIZE));
    908 	fii->flags = (FUSE_ASYNC_READ|FUSE_POSIX_LOCKS|FUSE_ATOMIC_O_TRUNC);
    909 
    910 	if ((error = xchg_msg(pu, 0, pm, sizeof(*fio), wait_reply)) != 0)
    911 		DERRX(EX_SOFTWARE, "init message exchange failed (%d)", error);
    912 
    913 	fio = GET_OUTPAYLOAD(ps, pm, fuse_init_out);
    914 	ps->ps_max_readahead = fio->max_readahead;
    915 	ps->ps_max_write = fio->max_write;
    916 
    917 	ps->ps_destroy_msg(pm);
    918 
    919 	return;
    920 }
    921 
    922 int
    923 perfuse_fs_unmount(struct puffs_usermount *pu, int flags)
    924 {
    925 	perfuse_msg_t *pm;
    926 	struct perfuse_state *ps;
    927 	puffs_cookie_t opc;
    928 	int error;
    929 
    930 	ps = puffs_getspecific(pu);
    931 	opc = (puffs_cookie_t)puffs_getroot(pu);
    932 
    933 	/*
    934 	 * ps_new_msg() is called with NULL creds, which will
    935 	 * be interpreted as FUSE superuser.
    936 	 */
    937 	pm = ps->ps_new_msg(pu, opc, FUSE_DESTROY, 0, NULL);
    938 
    939 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0){
    940 		DWARN("unmount %s", ps->ps_target);
    941 		if (!(flags & MNT_FORCE))
    942 			return error;
    943 		else
    944 			error = 0;
    945 	} else {
    946 		ps->ps_destroy_msg(pm);
    947 	}
    948 
    949 	ps->ps_umount(pu);
    950 
    951 	if (perfuse_diagflags & PDF_MISC)
    952 		DPRINTF("%s unmounted, exit\n", ps->ps_target);
    953 
    954 	return 0;
    955 }
    956 
    957 int
    958 perfuse_fs_statvfs(struct puffs_usermount *pu, struct puffs_statvfs *svfsb)
    959 {
    960 	struct perfuse_state *ps;
    961 	perfuse_msg_t *pm;
    962 	puffs_cookie_t opc;
    963 	struct fuse_statfs_out *fso;
    964 	int error;
    965 
    966 	ps = puffs_getspecific(pu);
    967 	opc = (puffs_cookie_t)puffs_getroot(pu);
    968 
    969 	/*
    970 	 * ps_new_msg() is called with NULL creds, which will
    971 	 * be interpreted as FUSE superuser.
    972 	 */
    973 	pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
    974 
    975 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply)) != 0)
    976 		return error;
    977 
    978 	fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
    979 	svfsb->f_flag = ps->ps_mountflags;
    980 	svfsb->f_bsize = fso->st.bsize;
    981 	svfsb->f_frsize = fso->st.frsize;
    982 	svfsb->f_iosize = ((struct puffs_node *)opc)->pn_va.va_blocksize;
    983 	svfsb->f_blocks = fso->st.blocks;
    984 	svfsb->f_bfree = fso->st.bfree;
    985 	svfsb->f_bavail = fso->st.bavail;
    986 	svfsb->f_bresvd = fso->st.bfree - fso->st.bavail;
    987 	svfsb->f_files = fso->st.files;
    988 	svfsb->f_ffree = fso->st.ffree;
    989 	svfsb->f_favail = fso->st.ffree;/* files not reserved for root */
    990 	svfsb->f_fresvd = 0;		/* files reserved for root */
    991 
    992 	svfsb->f_syncreads = ps->ps_syncreads;
    993 	svfsb->f_syncwrites = ps->ps_syncwrites;
    994 
    995 	svfsb->f_asyncreads = ps->ps_asyncreads;
    996 	svfsb->f_asyncwrites = ps->ps_asyncwrites;
    997 
    998 	(void)memcpy(&svfsb->f_fsidx, &ps->ps_fsid, sizeof(ps->ps_fsid));
    999 	svfsb->f_fsid = (unsigned long)ps->ps_fsid;
   1000 	svfsb->f_namemax = MAXPATHLEN;	/* XXX */
   1001 	svfsb->f_owner = ps->ps_owner_uid;
   1002 
   1003 	(void)strlcpy(svfsb->f_mntonname, ps->ps_target, _VFS_NAMELEN);
   1004 
   1005 	if (ps->ps_filesystemtype != NULL)
   1006 		(void)strlcpy(svfsb->f_fstypename,
   1007 			      ps->ps_filesystemtype, _VFS_NAMELEN);
   1008 	else
   1009 		(void)strlcpy(svfsb->f_fstypename, "fuse", _VFS_NAMELEN);
   1010 
   1011 	if (ps->ps_source != NULL)
   1012 		strlcpy(svfsb->f_mntfromname, ps->ps_source, _VFS_NAMELEN);
   1013 	else
   1014 		strlcpy(svfsb->f_mntfromname, _PATH_FUSE, _VFS_NAMELEN);
   1015 
   1016 	ps->ps_destroy_msg(pm);
   1017 
   1018 	return 0;
   1019 }
   1020 
   1021 int
   1022 perfuse_fs_sync(struct puffs_usermount *pu, int waitfor,
   1023 	const struct puffs_cred *pcr)
   1024 {
   1025 	/*
   1026 	 * FUSE does not seem to have a FS sync callback.
   1027 	 * Maybe do not even register this callback
   1028 	 */
   1029 	return puffs_fsnop_sync(pu, waitfor, pcr);
   1030 }
   1031 
   1032 /* ARGSUSED0 */
   1033 int
   1034 perfuse_fs_fhtonode(struct puffs_usermount *pu, void *fid, size_t fidsize,
   1035 	struct puffs_newinfo *pni)
   1036 {
   1037 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
   1038 	return 0;
   1039 }
   1040 
   1041 /* ARGSUSED0 */
   1042 int
   1043 perfuse_fs_nodetofh(struct puffs_usermount *pu, puffs_cookie_t cookie,
   1044 	void *fid, size_t *fidsize)
   1045 {
   1046 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
   1047 	return 0;
   1048 }
   1049 
   1050 #if 0
   1051 /* ARGSUSED0 */
   1052 void
   1053 perfuse_fs_extattrctl(struct puffs_usermount *pu, int cmd,
   1054 	puffs_cookie_t *cookie, int flags, int namespace, const char *attrname)
   1055 {
   1056 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
   1057 	return 0;
   1058 }
   1059 #endif /* 0 */
   1060 
   1061 /* ARGSUSED0 */
   1062 void
   1063 perfuse_fs_suspend(struct puffs_usermount *pu, int status)
   1064 {
   1065 	return;
   1066 }
   1067 
   1068 
   1069 int
   1070 perfuse_node_lookup(struct puffs_usermount *pu, puffs_cookie_t opc,
   1071 	struct puffs_newinfo *pni, const struct puffs_cn *pcn)
   1072 {
   1073 	struct perfuse_state *ps;
   1074 	struct puffs_node *pn;
   1075 	mode_t mode;
   1076 	int error;
   1077 
   1078 	ps = puffs_getspecific(pu);
   1079 	node_ref(opc);
   1080 
   1081 	/*
   1082 	 * Check permissions
   1083 	 */
   1084 	switch(pcn->pcn_nameiop) {
   1085 	case NAMEI_DELETE: /* FALLTHROUGH */
   1086 	case NAMEI_RENAME: /* FALLTHROUGH */
   1087 	case NAMEI_CREATE:
   1088 		if (pcn->pcn_flags & NAMEI_ISLASTCN)
   1089 			mode = PUFFS_VEXEC|PUFFS_VWRITE;
   1090 		else
   1091 			mode = PUFFS_VEXEC;
   1092 		break;
   1093 	case NAMEI_LOOKUP: /* FALLTHROUGH */
   1094 	default:
   1095 		mode = PUFFS_VEXEC;
   1096 		break;
   1097 	}
   1098 
   1099 	if ((error = mode_access(opc, pcn->pcn_cred, mode)) != 0)
   1100 		goto out;
   1101 
   1102 	error = node_lookup_common(pu, (puffs_cookie_t)opc, pni,
   1103 				   pcn->pcn_name, pcn->pcn_cred, &pn);
   1104 
   1105 	if (error != 0)
   1106 		goto out;
   1107 
   1108 	/*
   1109 	 * Kernel would kill us if the filesystem returned the parent
   1110 	 * itself. If we want to live, hide that!
   1111 	 */
   1112 	if ((opc == (puffs_cookie_t)pn) && (strcmp(pcn->pcn_name, ".") != 0)) {
   1113 		DERRX(EX_SOFTWARE, "lookup \"%s\" in \"%s\" returned parent",
   1114 		      pcn->pcn_name, perfuse_node_path(ps, opc));
   1115 		/* NOTREACHED */
   1116 		error = ESTALE;
   1117 		goto out;
   1118 	}
   1119 
   1120 	/*
   1121 	 * Removed node
   1122 	 */
   1123 	if (PERFUSE_NODE_DATA(pn)->pnd_flags & PND_REMOVED) {
   1124 		error = ENOENT;
   1125 		goto out;
   1126 	}
   1127 
   1128 	/*
   1129 	 * Check for sticky bit. Unfortunately there is no way to
   1130 	 * do this before creating the puffs_node, since we require
   1131 	 * this operation to get the node owner.
   1132 	 */
   1133 	switch (pcn->pcn_nameiop) {
   1134 	case NAMEI_DELETE: /* FALLTHROUGH */
   1135 	case NAMEI_RENAME:
   1136 		error = sticky_access(opc, pn, pcn->pcn_cred);
   1137 		if (error != 0) {
   1138 			(void)perfuse_node_reclaim2(pu, pn, 1);
   1139 			goto out;
   1140 		}
   1141 		break;
   1142 	default:
   1143 		break;
   1144 	}
   1145 
   1146 	PERFUSE_NODE_DATA(pn)->pnd_fuse_nlookup++;
   1147 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
   1148 
   1149 	error = 0;
   1150 
   1151 out:
   1152 	node_rele(opc);
   1153 	return error;
   1154 }
   1155 
   1156 int
   1157 perfuse_node_create(struct puffs_usermount *pu, puffs_cookie_t opc,
   1158 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
   1159 	const struct vattr *vap)
   1160 {
   1161 	perfuse_msg_t *pm;
   1162 	struct perfuse_state *ps;
   1163 	struct fuse_create_in *fci;
   1164 	struct fuse_entry_out *feo;
   1165 	struct fuse_open_out *foo;
   1166 	struct puffs_node *pn;
   1167 	const char *name;
   1168 	size_t namelen;
   1169 	size_t len;
   1170 	int error;
   1171 
   1172 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   1173 		return ENOENT;
   1174 
   1175 	node_ref(opc);
   1176 
   1177 	/*
   1178 	 * If create is unimplemented: Check that it does not
   1179 	 * already exists, and if not, do mknod and open
   1180 	 */
   1181 	ps = puffs_getspecific(pu);
   1182 	if (ps->ps_flags & PS_NO_CREAT) {
   1183 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
   1184 					   pcn->pcn_cred, &pn);
   1185 		if (error == 0)	{
   1186 			(void)perfuse_node_reclaim2(pu, pn, 1);
   1187 			error = EEXIST;
   1188 			goto out;
   1189 		}
   1190 
   1191 		error = perfuse_node_mknod(pu, opc, pni, pcn, vap);
   1192 		if (error != 0)
   1193 			goto out;
   1194 
   1195 		error = node_lookup_common(pu, opc, NULL, pcn->pcn_name,
   1196 					   pcn->pcn_cred, &pn);
   1197 		if (error != 0)
   1198 			goto out;
   1199 
   1200 		/*
   1201 		 * FUSE does the open at create time, while
   1202 		 * NetBSD will open in a subsequent operation.
   1203 		 * We need to open now, in order to retain FUSE
   1204 		 * semantics. The calling process will not get
   1205 		 * a file descriptor before the kernel sends
   1206 		 * the open operation.
   1207 		 */
   1208 		error = perfuse_node_open(pu, (puffs_cookie_t)pn,
   1209 					  FWRITE, pcn->pcn_cred);
   1210 		goto out;
   1211 	}
   1212 
   1213 	name = pcn->pcn_name;
   1214 	namelen = pcn->pcn_namelen + 1;
   1215 	len = sizeof(*fci) + namelen;
   1216 
   1217 	/*
   1218 	 * flags should use O_WRONLY instead of O_RDWR, but it
   1219 	 * breaks when the caller tries to read from file.
   1220 	 *
   1221 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
   1222 	 */
   1223 	pm = ps->ps_new_msg(pu, opc, FUSE_CREATE, len, pcn->pcn_cred);
   1224 	fci = GET_INPAYLOAD(ps, pm, fuse_create_in);
   1225 	fci->flags = O_CREAT | O_TRUNC | O_RDWR;
   1226 	fci->mode = vap->va_mode | VTTOIF(vap->va_type);
   1227 	fci->umask = 0; 	/* Seems unused by libfuse */
   1228 	(void)strlcpy((char*)(void *)(fci + 1), name, namelen);
   1229 
   1230 	len = sizeof(*feo) + sizeof(*foo);
   1231 	if ((error = xchg_msg(pu, opc, pm, len, wait_reply)) != 0) {
   1232 		/*
   1233 		 * create is unimplemented, remember it for later,
   1234 		 * and start over using mknod and open instead.
   1235 		 */
   1236 		if (error == ENOSYS) {
   1237 			ps->ps_flags |= PS_NO_CREAT;
   1238 			error = perfuse_node_create(pu, opc, pni, pcn, vap);
   1239 		}
   1240 
   1241 		goto out;
   1242 	}
   1243 
   1244 	feo = GET_OUTPAYLOAD(ps, pm, fuse_entry_out);
   1245 	foo = (struct fuse_open_out *)(void *)(feo + 1);
   1246 	if (feo->nodeid == PERFUSE_UNKNOWN_NODEID)
   1247 		DERRX(EX_SOFTWARE, "%s: no nodeid", __func__);
   1248 
   1249 	/*
   1250 	 * Save the file handle and inode in node private data
   1251 	 * so that we can reuse it later
   1252 	 */
   1253 	pn = perfuse_new_pn(pu, name, opc);
   1254 	perfuse_new_fh((puffs_cookie_t)pn, foo->fh, FWRITE);
   1255 	PERFUSE_NODE_DATA(pn)->pnd_nodeid = feo->nodeid;
   1256 	PERFUSE_NODE_DATA(pn)->pnd_fuse_nlookup++;
   1257 	PERFUSE_NODE_DATA(pn)->pnd_puffs_nlookup++;
   1258 	perfuse_node_cache(ps, pn);
   1259 
   1260 	fuse_attr_to_vap(ps, &pn->pn_va, &feo->attr);
   1261 	pn->pn_va.va_gen = (u_long)(feo->generation);
   1262 
   1263 	puffs_newinfo_setcookie(pni, pn);
   1264 #ifdef PUFFS_KFLAG_CACHE_FS_TTL
   1265 	puffs_newinfo_setva(pni, &pn->pn_va);
   1266 	perfuse_newinfo_setttl(pni, pn, feo, NULL);
   1267 #endif /* PUFFS_KFLAG_CACHE_FS_TTL */
   1268 
   1269 #ifdef PERFUSE_DEBUG
   1270 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
   1271 		DPRINTF("%s: opc = %p, file = \"%s\", flags = 0x%x "
   1272 			"nodeid = 0x%"PRIx64", wfh = 0x%"PRIx64"\n",
   1273 			__func__, (void *)pn, pcn->pcn_name,
   1274 			PERFUSE_NODE_DATA(pn)->pnd_flags, feo->nodeid,
   1275 			foo->fh);
   1276 #endif
   1277 
   1278 	ps->ps_destroy_msg(pm);
   1279 	error = 0;
   1280 
   1281 out:
   1282 	node_rele(opc);
   1283 	return error;
   1284 }
   1285 
   1286 
   1287 int
   1288 perfuse_node_mknod(struct puffs_usermount *pu, puffs_cookie_t opc,
   1289 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
   1290 	const struct vattr *vap)
   1291 {
   1292 	struct perfuse_state *ps;
   1293 	perfuse_msg_t *pm;
   1294 	struct fuse_mknod_in *fmi;
   1295 	const char* path;
   1296 	size_t len;
   1297 	int error;
   1298 
   1299 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   1300 		return ENOENT;
   1301 
   1302 	node_ref(opc);
   1303 
   1304 	/*
   1305 	 * Only superuser can mknod objects other than
   1306 	 * directories, files, socks, fifo and links.
   1307 	 *
   1308 	 * Create an object require -WX permission in the parent directory
   1309 	 */
   1310 	switch (vap->va_type) {
   1311 	case VDIR:	/* FALLTHROUGH */
   1312 	case VREG:	/* FALLTHROUGH */
   1313 	case VFIFO:	/* FALLTHROUGH */
   1314 	case VSOCK:
   1315 		break;
   1316 	default:	/* VNON, VBLK, VCHR, VBAD */
   1317 		if (!puffs_cred_isjuggernaut(pcn->pcn_cred)) {
   1318 			error = EPERM;
   1319 			goto out;
   1320 		}
   1321 		break;
   1322 	}
   1323 
   1324 
   1325 	ps = puffs_getspecific(pu);
   1326 	path = pcn->pcn_name;
   1327 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
   1328 
   1329 	/*
   1330 	 * mode must contain file type (ie: S_IFREG), use VTTOIF(vap->va_type)
   1331 	 */
   1332 	pm = ps->ps_new_msg(pu, opc, FUSE_MKNOD, len, pcn->pcn_cred);
   1333 	fmi = GET_INPAYLOAD(ps, pm, fuse_mknod_in);
   1334 	fmi->mode = vap->va_mode | VTTOIF(vap->va_type);
   1335 	fmi->rdev = (uint32_t)vap->va_rdev;
   1336 	fmi->umask = 0; 	/* Seems unused bu libfuse */
   1337 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
   1338 
   1339 	error = node_mk_common(pu, opc, pni, pcn, pm);
   1340 
   1341 out:
   1342 	node_rele(opc);
   1343 	return error;
   1344 }
   1345 
   1346 
   1347 int
   1348 perfuse_node_open(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
   1349 	const struct puffs_cred *pcr)
   1350 {
   1351 	return perfuse_node_open2(pu, opc, mode, pcr, NULL);
   1352 }
   1353 
   1354 int
   1355 perfuse_node_open2(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
   1356 	const struct puffs_cred *pcr, int *oflags)
   1357 {
   1358 	struct perfuse_state *ps;
   1359 	struct perfuse_node_data *pnd;
   1360 	perfuse_msg_t *pm;
   1361 	mode_t fmode;
   1362 	int op;
   1363 	struct fuse_open_in *foi;
   1364 	struct fuse_open_out *foo;
   1365 	int error;
   1366 
   1367 	ps = puffs_getspecific(pu);
   1368 	pnd = PERFUSE_NODE_DATA(opc);
   1369 	error = 0;
   1370 
   1371 	if (pnd->pnd_flags & PND_REMOVED)
   1372 		return ENOENT;
   1373 
   1374 	node_ref(opc);
   1375 
   1376 	if (PN_ISDIR(opc))
   1377 		op = FUSE_OPENDIR;
   1378 	else
   1379 		op = FUSE_OPEN;
   1380 
   1381 	/*
   1382 	 * libfuse docs says
   1383 	 * - O_CREAT and O_EXCL should never be set.
   1384 	 * - O_TRUNC may be used if mount option atomic_o_trunc is used XXX
   1385 	 *
   1386 	 * O_APPEND makes no sense since FUSE always sends
   1387 	 * the file offset for write operations. If the
   1388 	 * filesystem uses pwrite(), O_APPEND would cause
   1389 	 * the offset to be ignored and cause file corruption.
   1390 	 */
   1391 	mode &= ~(O_CREAT|O_EXCL|O_APPEND);
   1392 
   1393 	/*
   1394 	 * Do not open twice, and do not reopen for reading
   1395 	 * if we already have write handle.
   1396 	 */
   1397 	switch (mode & (FREAD|FWRITE)) {
   1398 	case FREAD:
   1399 		if (pnd->pnd_flags & (PND_RFH|PND_WFH))
   1400 			goto out;
   1401 		break;
   1402 	case FWRITE:
   1403 		if (pnd->pnd_flags & PND_WFH)
   1404 			goto out;
   1405 		break;
   1406 	case FREAD|FWRITE:
   1407 		if (pnd->pnd_flags & PND_WFH)
   1408 			goto out;
   1409 
   1410 		/*
   1411 		 * Corner case: if already open for reading (PND_RFH)
   1412 		 * and re-opening FREAD|FWRITE, we need to reopen,
   1413 		 * but only for writing. Note the change on mode
   1414 		 * will only affect perfuse_new_fh()
   1415 		 */
   1416 		if (pnd->pnd_flags & PND_RFH)
   1417 			mode &= ~FREAD;
   1418 		break;
   1419 	default:
   1420 		DWARNX("open without either FREAD nor FWRITE");
   1421 		error = EPERM;
   1422 		goto out;
   1423 	}
   1424 
   1425 	/*
   1426 	 * Queue open on a node so that we do not open
   1427 	 * twice. This would be better with read and
   1428 	 * write distinguished.
   1429 	 */
   1430 	while (pnd->pnd_flags & PND_INOPEN)
   1431 		requeue_request(pu, opc, PCQ_OPEN);
   1432 	pnd->pnd_flags |= PND_INOPEN;
   1433 
   1434 	/*
   1435 	 * Convert PUFFS mode to FUSE mode: convert FREAD/FWRITE
   1436 	 * to O_RDONLY/O_WRONLY while perserving the other options.
   1437 	 */
   1438 	fmode = mode & ~(FREAD|FWRITE);
   1439 	fmode |= (mode & FWRITE) ? O_RDWR : O_RDONLY;
   1440 
   1441 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*foi), pcr);
   1442 	foi = GET_INPAYLOAD(ps, pm, fuse_open_in);
   1443 	foi->flags = fmode;
   1444 	foi->unused = 0;
   1445 
   1446 	if ((error = xchg_msg(pu, opc, pm, sizeof(*foo), wait_reply)) != 0)
   1447 		goto out;
   1448 
   1449 	foo = GET_OUTPAYLOAD(ps, pm, fuse_open_out);
   1450 
   1451 	/*
   1452 	 * Save the file handle in node private data
   1453 	 * so that we can reuse it later
   1454 	 */
   1455 	perfuse_new_fh(opc, foo->fh, mode);
   1456 
   1457 	/*
   1458 	 * Set direct I/O if the filesystems forces it
   1459 	 */
   1460 	if ((foo->open_flags & FUSE_FOPEN_DIRECT_IO) && (oflags != NULL))
   1461 		*oflags |= PUFFS_OPEN_IO_DIRECT;
   1462 
   1463 #ifdef PERFUSE_DEBUG
   1464 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
   1465 		DPRINTF("%s: opc = %p, file = \"%s\", "
   1466 			"nodeid = 0x%"PRIx64", %s%sfh = 0x%"PRIx64"\n",
   1467 			__func__, (void *)opc, perfuse_node_path(ps, opc),
   1468 			pnd->pnd_nodeid, mode & FREAD ? "r" : "",
   1469 			mode & FWRITE ? "w" : "", foo->fh);
   1470 #endif
   1471 
   1472 	ps->ps_destroy_msg(pm);
   1473 out:
   1474 
   1475 	pnd->pnd_flags &= ~PND_INOPEN;
   1476 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
   1477 
   1478 	node_rele(opc);
   1479 	return error;
   1480 }
   1481 
   1482 /* ARGSUSED0 */
   1483 int
   1484 perfuse_node_close(struct puffs_usermount *pu, puffs_cookie_t opc, int flags,
   1485 	const struct puffs_cred *pcr)
   1486 {
   1487 	struct perfuse_node_data *pnd;
   1488 
   1489 	pnd = PERFUSE_NODE_DATA(opc);
   1490 
   1491 	if (!(pnd->pnd_flags & PND_OPEN))
   1492 		return EBADF;
   1493 
   1494 	/*
   1495 	 * Actual close is postponed at inactive time.
   1496 	 */
   1497 	return 0;
   1498 }
   1499 
   1500 int
   1501 perfuse_node_access(struct puffs_usermount *pu, puffs_cookie_t opc, int mode,
   1502 	const struct puffs_cred *pcr)
   1503 {
   1504 	perfuse_msg_t *pm;
   1505 	struct perfuse_state *ps;
   1506 	struct fuse_access_in *fai;
   1507 	int error;
   1508 
   1509 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   1510 		return ENOENT;
   1511 
   1512 	node_ref(opc);
   1513 
   1514 	/*
   1515 	 * If we previously detected the filesystem does not
   1516 	 * implement access(), short-circuit the call and skip
   1517 	 * to libpuffs access() emulation.
   1518 	 */
   1519 	ps = puffs_getspecific(pu);
   1520 	if (ps->ps_flags & PS_NO_ACCESS) {
   1521 		const struct vattr *vap;
   1522 
   1523 		vap = puffs_pn_getvap((struct puffs_node *)opc);
   1524 
   1525 		error = puffs_access(IFTOVT(vap->va_mode),
   1526 				     vap->va_mode & ACCESSPERMS,
   1527 				     vap->va_uid, vap->va_gid,
   1528 				     (mode_t)mode, pcr);
   1529 		goto out;
   1530 	}
   1531 
   1532 	/*
   1533 	 * Plain access call
   1534 	 */
   1535 	pm = ps->ps_new_msg(pu, opc, FUSE_ACCESS, sizeof(*fai), pcr);
   1536 	fai = GET_INPAYLOAD(ps, pm, fuse_access_in);
   1537 	fai->mask = 0;
   1538 	fai->mask |= (mode & PUFFS_VREAD) ? R_OK : 0;
   1539 	fai->mask |= (mode & PUFFS_VWRITE) ? W_OK : 0;
   1540 	fai->mask |= (mode & PUFFS_VEXEC) ? X_OK : 0;
   1541 
   1542 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
   1543 
   1544 	ps->ps_destroy_msg(pm);
   1545 
   1546 	/*
   1547 	 * If unimplemented, start over with emulation
   1548 	 */
   1549 	if (error == ENOSYS) {
   1550 		ps->ps_flags |= PS_NO_ACCESS;
   1551 		error = perfuse_node_access(pu, opc, mode, pcr);
   1552 	}
   1553 
   1554 out:
   1555 	node_rele(opc);
   1556 	return error;
   1557 }
   1558 
   1559 int
   1560 perfuse_node_getattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   1561 	struct vattr *vap, const struct puffs_cred *pcr)
   1562 {
   1563 	return perfuse_node_getattr_ttl(pu, opc, vap, pcr, NULL);
   1564 }
   1565 
   1566 int
   1567 perfuse_node_getattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
   1568 	struct vattr *vap, const struct puffs_cred *pcr,
   1569 	struct timespec *va_ttl)
   1570 {
   1571 	perfuse_msg_t *pm = NULL;
   1572 	struct perfuse_state *ps;
   1573 	struct perfuse_node_data *pnd = PERFUSE_NODE_DATA(opc);
   1574 	struct fuse_getattr_in *fgi;
   1575 	struct fuse_attr_out *fao;
   1576 	int error = 0;
   1577 
   1578 	if ((pnd->pnd_flags & PND_REMOVED) && !(pnd->pnd_flags & PND_OPEN))
   1579 		return ENOENT;
   1580 
   1581 	node_ref(opc);
   1582 
   1583 	/*
   1584 	 * Serialize size access, see comment in perfuse_node_setattr().
   1585 	 */
   1586 	while (pnd->pnd_flags & PND_INRESIZE)
   1587 		requeue_request(pu, opc, PCQ_RESIZE);
   1588 	pnd->pnd_flags |= PND_INRESIZE;
   1589 
   1590 	ps = puffs_getspecific(pu);
   1591 
   1592 	/*
   1593 	 * FUSE_GETATTR_FH must be set in fgi->flags
   1594 	 * if we use for fgi->fh
   1595 	 */
   1596 	pm = ps->ps_new_msg(pu, opc, FUSE_GETATTR, sizeof(*fgi), pcr);
   1597 	fgi = GET_INPAYLOAD(ps, pm, fuse_getattr_in);
   1598 	fgi->getattr_flags = 0;
   1599 	fgi->dummy = 0;
   1600 	fgi->fh = FUSE_UNKNOWN_FH;
   1601 
   1602 	if (!PN_ISDIR(opc) && PERFUSE_NODE_DATA(opc)->pnd_flags & PND_OPEN) {
   1603 		fgi->fh = perfuse_get_fh(opc, FREAD);
   1604 		fgi->getattr_flags |= FUSE_GETATTR_FH;
   1605 	}
   1606 
   1607 #ifdef PERFUSE_DEBUG
   1608 	if (perfuse_diagflags & PDF_RESIZE)
   1609 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__, (void *)opc,
   1610 		    vap->va_size);
   1611 #endif
   1612 
   1613 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), wait_reply)) != 0)
   1614 		goto out;
   1615 
   1616 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
   1617 
   1618 #ifdef PERFUSE_DEBUG
   1619 	if (perfuse_diagflags & PDF_RESIZE)
   1620 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
   1621 		    (void *)opc, vap->va_size, fao->attr.size);
   1622 #endif
   1623 
   1624 	/*
   1625 	 * We set birthtime, flags, filerev,vaflags to 0.
   1626 	 * This seems the best bet, since the information is
   1627 	 * not available from filesystem.
   1628 	 */
   1629 	fuse_attr_to_vap(ps, vap, &fao->attr);
   1630 
   1631 	if (va_ttl != NULL) {
   1632 		va_ttl->tv_sec = fao->attr_valid;
   1633 		va_ttl->tv_nsec = fao->attr_valid_nsec;
   1634 	}
   1635 
   1636 	ps->ps_destroy_msg(pm);
   1637 	error = 0;
   1638 out:
   1639 
   1640 	pnd->pnd_flags &= ~PND_INRESIZE;
   1641 	(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
   1642 
   1643 	node_rele(opc);
   1644 	return error;
   1645 }
   1646 
   1647 int
   1648 perfuse_node_setattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   1649 	const struct vattr *vap, const struct puffs_cred *pcr)
   1650 {
   1651 	return perfuse_node_setattr_ttl(pu, opc,
   1652 					__UNCONST(vap), pcr, NULL, 0);
   1653 }
   1654 
   1655 int
   1656 perfuse_node_setattr_ttl(struct puffs_usermount *pu, puffs_cookie_t opc,
   1657 	struct vattr *vap, const struct puffs_cred *pcr,
   1658 	struct timespec *va_ttl, int xflag)
   1659 {
   1660 	perfuse_msg_t *pm;
   1661 	uint64_t fh;
   1662 	struct perfuse_state *ps;
   1663 	struct perfuse_node_data *pnd;
   1664 	struct fuse_setattr_in *fsi;
   1665 	struct fuse_attr_out *fao;
   1666 	struct vattr *old_va;
   1667 	enum perfuse_xchg_pb_reply reply;
   1668 	int error;
   1669 #ifdef PERFUSE_DEBUG
   1670 	struct vattr *old_vap;
   1671 	int resize_debug = 0;
   1672 #endif
   1673 	ps = puffs_getspecific(pu);
   1674 	pnd = PERFUSE_NODE_DATA(opc);
   1675 
   1676 	/*
   1677 	 * The only operation we can do once the file is removed
   1678 	 * is to resize it, and we can do it only if it is open.
   1679 	 * Do not even send the operation to the filesystem: the
   1680 	 * file is not there anymore.
   1681 	 */
   1682 	if (pnd->pnd_flags & PND_REMOVED) {
   1683 		if (!(pnd->pnd_flags & PND_OPEN))
   1684 			return ENOENT;
   1685 
   1686 		return 0;
   1687 	}
   1688 
   1689 	old_va = puffs_pn_getvap((struct puffs_node *)opc);
   1690 
   1691 	/*
   1692 	 * Check for permission to change size
   1693 	 * It is always allowed if we already have a write file handle
   1694 	 */
   1695 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
   1696 	    !(pnd->pnd_flags & PND_WFH) &&
   1697 	    (error = mode_access(opc, pcr, PUFFS_VWRITE)) != 0)
   1698 		return error;
   1699 
   1700 	/*
   1701 	 * Check for permission to change dates
   1702 	 */
   1703 	if (((vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) ||
   1704 	     (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL)) &&
   1705 	    (puffs_access_times(old_va->va_uid, old_va->va_gid,
   1706 				old_va->va_mode, 0, pcr) != 0))
   1707 		return EPERM;
   1708 
   1709 	/*
   1710 	 * Check for permission to change owner and group
   1711 	 */
   1712 	if (((vap->va_uid != (uid_t)PUFFS_VNOVAL) ||
   1713 	     (vap->va_gid != (gid_t)PUFFS_VNOVAL)) &&
   1714 	    (puffs_access_chown(old_va->va_uid, old_va->va_gid,
   1715 				vap->va_uid, vap->va_gid, pcr)) != 0)
   1716 		return EPERM;
   1717 
   1718 	/*
   1719 	 * Check for sticky bit on non-directory by non root user
   1720 	 */
   1721 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
   1722 	    (vap->va_mode & S_ISTXT) && (old_va->va_type != VDIR) &&
   1723 	    !puffs_cred_isjuggernaut(pcr))
   1724 		return EFTYPE;
   1725 
   1726 	/*
   1727 	 * Check for permission to change permissions
   1728 	 */
   1729 	if ((vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
   1730 	    (puffs_access_chmod(old_va->va_uid, old_va->va_gid,
   1731 				old_va->va_type, vap->va_mode, pcr)) != 0)
   1732 		return EPERM;
   1733 
   1734 	node_ref(opc);
   1735 
   1736 	if (!PN_ISDIR(opc) && pnd->pnd_flags & PND_WFH)
   1737 		fh = perfuse_get_fh(opc, FWRITE);
   1738 	else
   1739 		fh = FUSE_UNKNOWN_FH;
   1740 
   1741 	/*
   1742 	 * fchmod() sets mode and fh, and it may carry
   1743 	 * a resize as well. That may break if the
   1744 	 * filesystem does chmod then resize, and fails
   1745 	 * because it does not have permission anymore.
   1746 	 * We work this around by splitting into two setattr.
   1747 	 */
   1748 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
   1749 	    (vap->va_mode != (mode_t)PUFFS_VNOVAL) &&
   1750 	    (fh != FUSE_UNKNOWN_FH)) {
   1751 		struct vattr resize_va;
   1752 
   1753 		(void)memcpy(&resize_va, vap, sizeof(resize_va));
   1754 		resize_va.va_mode = (mode_t)PUFFS_VNOVAL;
   1755 		if ((error = perfuse_node_setattr_ttl(pu, opc, &resize_va,
   1756 						      pcr, va_ttl, xflag)) != 0)
   1757 			goto out2;
   1758 
   1759 		vap->va_size = (u_quad_t)PUFFS_VNOVAL;
   1760 	}
   1761 
   1762 	pm = ps->ps_new_msg(pu, opc, FUSE_SETATTR, sizeof(*fsi), pcr);
   1763 	fsi = GET_INPAYLOAD(ps, pm, fuse_setattr_in);
   1764 	fsi->valid = 0;
   1765 
   1766 	/*
   1767 	 * Get a fh if the node is open for writing
   1768 	 */
   1769 	if (fh != FUSE_UNKNOWN_FH) {
   1770 		fsi->fh = fh;
   1771 		fsi->valid |= FUSE_FATTR_FH;
   1772 	}
   1773 
   1774 
   1775 	if (vap->va_size != (u_quad_t)PUFFS_VNOVAL) {
   1776 		fsi->size = vap->va_size;
   1777 		fsi->valid |= FUSE_FATTR_SIZE;
   1778 
   1779 		/*
   1780 		 * Serialize anything that can touch file size
   1781 		 * to avoid reordered GETATTR and SETATTR.
   1782 		 * Out of order SETATTR can report stale size,
   1783 		 * which will cause the kernel to truncate the file.
   1784 		 * XXX Probably useless now we have a lock on GETATTR
   1785 		 */
   1786 		while (pnd->pnd_flags & PND_INRESIZE)
   1787 			requeue_request(pu, opc, PCQ_RESIZE);
   1788 		pnd->pnd_flags |= PND_INRESIZE;
   1789 	}
   1790 
   1791 	/*
   1792  	 * When not sending a time field, still fill with
   1793 	 * current value, as the filesystem may just reset
   1794 	 * the field to Epoch even if fsi->valid bit is
   1795 	 * not set (GlusterFS does that).
   1796  	 */
   1797 	if (vap->va_atime.tv_sec != (time_t)PUFFS_VNOVAL) {
   1798 		fsi->atime = vap->va_atime.tv_sec;
   1799 		fsi->atimensec = (uint32_t)vap->va_atime.tv_nsec;
   1800 		fsi->valid |= FUSE_FATTR_ATIME;
   1801 	} else {
   1802 		fsi->atime = old_va->va_atime.tv_sec;
   1803 		fsi->atimensec = (uint32_t)old_va->va_atime.tv_nsec;
   1804 	}
   1805 
   1806 	if (vap->va_mtime.tv_sec != (time_t)PUFFS_VNOVAL) {
   1807 		fsi->mtime = vap->va_mtime.tv_sec;
   1808 		fsi->mtimensec = (uint32_t)vap->va_mtime.tv_nsec;
   1809 		fsi->valid |= FUSE_FATTR_MTIME;
   1810 	} else {
   1811 		fsi->mtime = old_va->va_mtime.tv_sec;
   1812 		fsi->mtimensec = (uint32_t)old_va->va_mtime.tv_nsec;
   1813 	}
   1814 
   1815 	if (vap->va_mode != (mode_t)PUFFS_VNOVAL) {
   1816 		fsi->mode = vap->va_mode;
   1817 		fsi->valid |= FUSE_FATTR_MODE;
   1818 	}
   1819 
   1820 	if (vap->va_uid != (uid_t)PUFFS_VNOVAL) {
   1821 		fsi->uid = vap->va_uid;
   1822 		fsi->valid |= FUSE_FATTR_UID;
   1823 	}
   1824 
   1825 	if (vap->va_gid != (gid_t)PUFFS_VNOVAL) {
   1826 		fsi->gid = vap->va_gid;
   1827 		fsi->valid |= FUSE_FATTR_GID;
   1828 	}
   1829 
   1830 	if (pnd->pnd_lock_owner != 0) {
   1831 		fsi->lock_owner = pnd->pnd_lock_owner;
   1832 		fsi->valid |= FUSE_FATTR_LOCKOWNER;
   1833 	}
   1834 
   1835 #ifndef PUFFS_KFLAG_NOFLUSH_META
   1836 	/*
   1837 	 * ftruncate() sends only va_size, and metadata cache
   1838 	 * flush adds va_atime and va_mtime. Some FUSE
   1839 	 * filesystems will attempt to detect ftruncate by
   1840 	 * checking for FATTR_SIZE being set without
   1841 	 * FATTR_UID|FATTR_GID|FATTR_ATIME|FATTR_MTIME|FATTR_MODE
   1842 	 *
   1843 	 * Try to adapt and remove FATTR_ATIME|FATTR_MTIME
   1844 	 * if we suspect a ftruncate().
   1845 	 */
   1846 	if ((vap->va_size != (u_quad_t)PUFFS_VNOVAL) &&
   1847 	    ((vap->va_mode == (mode_t)PUFFS_VNOVAL) &&
   1848 	     (vap->va_uid == (uid_t)PUFFS_VNOVAL) &&
   1849 	     (vap->va_gid == (gid_t)PUFFS_VNOVAL))) {
   1850 		fsi->atime = 0;
   1851 		fsi->atimensec = 0;
   1852 		fsi->mtime = 0;
   1853 		fsi->mtimensec = 0;
   1854 		fsi->valid &= ~(FUSE_FATTR_ATIME|FUSE_FATTR_MTIME);
   1855 	}
   1856 
   1857 	/*
   1858 	 * If only atime is changed, discard the operation: it
   1859 	 * happens after read, and in that case the filesystem
   1860 	 * already updated atime. NB: utimes() also change mtime.
   1861 	 */
   1862 	if (fsi->valid == FUSE_FATTR_ATIME)
   1863 		fsi->valid &= ~FUSE_FATTR_ATIME;
   1864 #endif /* PUFFS_KFLAG_NOFLUSH_META */
   1865 
   1866 	/*
   1867 	 * If nothing remain, discard the operation.
   1868 	 */
   1869 	if (!(fsi->valid & (FUSE_FATTR_SIZE|FUSE_FATTR_ATIME|FUSE_FATTR_MTIME|
   1870 			    FUSE_FATTR_MODE|FUSE_FATTR_UID|FUSE_FATTR_GID))) {
   1871 		error = 0;
   1872 		ps->ps_destroy_msg(pm);
   1873 		goto out;
   1874 	}
   1875 
   1876 #ifdef PERFUSE_DEBUG
   1877 	old_vap = puffs_pn_getvap((struct puffs_node *)opc);
   1878 
   1879 	if ((perfuse_diagflags & PDF_RESIZE) &&
   1880 	    (old_vap->va_size != (u_quad_t)PUFFS_VNOVAL)) {
   1881 		resize_debug = 1;
   1882 
   1883 		DPRINTF(">> %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
   1884 		    (void *)opc,
   1885 		    puffs_pn_getvap((struct puffs_node *)opc)->va_size,
   1886 		    fsi->size);
   1887 	}
   1888 #endif
   1889 
   1890 	/*
   1891 	 * Do not honour FAF when changing size. How do
   1892 	 * you want such a thing to work?
   1893 	 */
   1894 	reply = wait_reply;
   1895 #ifdef PUFFS_SETATTR_FAF
   1896 	if ((xflag & PUFFS_SETATTR_FAF) && !(fsi->valid & FUSE_FATTR_SIZE))
   1897 		reply = no_reply;
   1898 #endif
   1899 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fao), reply)) != 0)
   1900 		goto out;
   1901 
   1902 	if (reply == no_reply)
   1903 		goto out;
   1904 
   1905 	/*
   1906 	 * Copy back the new values
   1907 	 */
   1908 	fao = GET_OUTPAYLOAD(ps, pm, fuse_attr_out);
   1909 
   1910 #ifdef PERFUSE_DEBUG
   1911 	if (resize_debug)
   1912 		DPRINTF("<< %s %p %" PRIu64 " -> %" PRIu64 "\n", __func__,
   1913 		    (void *)opc, old_vap->va_size, fao->attr.size);
   1914 #endif
   1915 
   1916 	fuse_attr_to_vap(ps, old_va, &fao->attr);
   1917 
   1918 	if (va_ttl != NULL) {
   1919 		va_ttl->tv_sec = fao->attr_valid;
   1920 		va_ttl->tv_nsec = fao->attr_valid_nsec;
   1921 		(void)memcpy(vap, old_va, sizeof(*vap));
   1922 	}
   1923 
   1924 	ps->ps_destroy_msg(pm);
   1925 	error = 0;
   1926 
   1927 out:
   1928 	if (pnd->pnd_flags & PND_INRESIZE) {
   1929 		pnd->pnd_flags &= ~PND_INRESIZE;
   1930 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
   1931 	}
   1932 
   1933 out2:
   1934 	node_rele(opc);
   1935 	return error;
   1936 }
   1937 
   1938 int
   1939 perfuse_node_poll(struct puffs_usermount *pu, puffs_cookie_t opc, int *events)
   1940 {
   1941 	struct perfuse_state *ps;
   1942 	perfuse_msg_t *pm;
   1943 	struct fuse_poll_in *fpi;
   1944 	struct fuse_poll_out *fpo;
   1945 	int error;
   1946 
   1947 	node_ref(opc);
   1948 	ps = puffs_getspecific(pu);
   1949 	/*
   1950 	 * kh is set if FUSE_POLL_SCHEDULE_NOTIFY is set.
   1951 	 *
   1952 	 * XXX ps_new_msg() is called with NULL creds, which will
   1953 	 * be interpreted as FUSE superuser. We have no way to
   1954 	 * know the requesting process' credential, but since poll
   1955 	 * is supposed to operate on a file that has been open,
   1956 	 * permission should have already been checked at open time.
   1957 	 * That still may breaks on filesystems that provides odd
   1958 	 * semantics.
   1959  	 */
   1960 	pm = ps->ps_new_msg(pu, opc, FUSE_POLL, sizeof(*fpi), NULL);
   1961 	fpi = GET_INPAYLOAD(ps, pm, fuse_poll_in);
   1962 	fpi->fh = PN_ISDIR(opc) ? FUSE_UNKNOWN_FH : perfuse_get_fh(opc, FREAD);
   1963 	fpi->kh = 0;
   1964 	fpi->flags = 0;
   1965 
   1966 #ifdef PERFUSE_DEBUG
   1967 	if (perfuse_diagflags & PDF_FH)
   1968 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
   1969 			"fh = 0x%"PRIx64"\n", __func__, (void *)opc,
   1970 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fpi->fh);
   1971 #endif
   1972 	if ((error = xchg_msg(pu, opc, pm, sizeof(*fpo), wait_reply)) != 0)
   1973 		goto out;
   1974 
   1975 	fpo = GET_OUTPAYLOAD(ps, pm, fuse_poll_out);
   1976 	*events = fpo->revents;
   1977 
   1978 	ps->ps_destroy_msg(pm);
   1979 	error = 0;
   1980 
   1981 out:
   1982 	node_rele(opc);
   1983 	return error;
   1984 }
   1985 
   1986 /* ARGSUSED2 */
   1987 int
   1988 perfuse_node_fsync(struct puffs_usermount *pu, puffs_cookie_t opc,
   1989 	const struct puffs_cred *pcr, int flags, off_t offlo, off_t offhi)
   1990 {
   1991 	int op;
   1992 	perfuse_msg_t *pm;
   1993 	struct perfuse_state *ps;
   1994 	struct perfuse_node_data *pnd;
   1995 	struct fuse_fsync_in *ffi;
   1996 	uint64_t fh;
   1997 	int error = 0;
   1998 
   1999 	pm = NULL;
   2000 	ps = puffs_getspecific(pu);
   2001 	pnd = PERFUSE_NODE_DATA(opc);
   2002 
   2003 	/*
   2004 	 * No need to sync a removed node
   2005 	 */
   2006 	if (pnd->pnd_flags & PND_REMOVED)
   2007 		return 0;
   2008 
   2009 	/*
   2010 	 * We do not sync closed files. They have been
   2011 	 * sync at inactive time already.
   2012 	 */
   2013 	if (!(pnd->pnd_flags & PND_OPEN))
   2014 		return 0;
   2015 
   2016 	node_ref(opc);
   2017 
   2018 	if (PN_ISDIR(opc))
   2019 		op = FUSE_FSYNCDIR;
   2020 	else 		/* VREG but also other types such as VLNK */
   2021 		op = FUSE_FSYNC;
   2022 
   2023 	/*
   2024 	 * Do not sync if there are no change to sync
   2025 	 * XXX remove that test on files if we implement mmap
   2026 	 */
   2027 #ifdef PERFUSE_DEBUG
   2028 	if (perfuse_diagflags & PDF_SYNC)
   2029 		DPRINTF("%s: TEST opc = %p, file = \"%s\" is %sdirty\n",
   2030 			__func__, (void*)opc, perfuse_node_path(ps, opc),
   2031 			pnd->pnd_flags & PND_DIRTY ? "" : "not ");
   2032 #endif
   2033 	if (!(pnd->pnd_flags & PND_DIRTY))
   2034 		goto out;
   2035 
   2036 	/*
   2037 	 * It seems NetBSD can call fsync without open first
   2038 	 * glusterfs complain in such a situation:
   2039 	 * "FSYNC() ERR => -1 (Invalid argument)"
   2040 	 * The file will be closed at inactive time.
   2041 	 *
   2042 	 * We open the directory for reading in order to sync.
   2043 	 * This sounds rather counterintuitive, but it works.
   2044 	 */
   2045 	if (!(pnd->pnd_flags & PND_WFH)) {
   2046 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
   2047 			goto out;
   2048 	}
   2049 
   2050 	if (op == FUSE_FSYNCDIR)
   2051 		fh = perfuse_get_fh(opc, FREAD);
   2052 	else
   2053 		fh = perfuse_get_fh(opc, FWRITE);
   2054 
   2055 	/*
   2056 	 * If fsync_flags  is set, meta data should not be flushed.
   2057 	 */
   2058 	pm = ps->ps_new_msg(pu, opc, op, sizeof(*ffi), pcr);
   2059 	ffi = GET_INPAYLOAD(ps, pm, fuse_fsync_in);
   2060 	ffi->fh = fh;
   2061 	ffi->fsync_flags = (flags & FFILESYNC) ? 0 : 1;
   2062 
   2063 #ifdef PERFUSE_DEBUG
   2064 	if (perfuse_diagflags & PDF_FH)
   2065 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
   2066 			__func__, (void *)opc,
   2067 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, ffi->fh);
   2068 #endif
   2069 
   2070 	if ((error = xchg_msg(pu, opc, pm,
   2071 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
   2072 		goto out;
   2073 
   2074 	/*
   2075 	 * No reply beyond fuse_out_header: nothing to do on success
   2076 	 * just clear the dirty flag
   2077 	 */
   2078 	pnd->pnd_flags &= ~PND_DIRTY;
   2079 
   2080 #ifdef PERFUSE_DEBUG
   2081 	if (perfuse_diagflags & PDF_SYNC)
   2082 		DPRINTF("%s: CLEAR opc = %p, file = \"%s\"\n",
   2083 			__func__, (void*)opc, perfuse_node_path(ps, opc));
   2084 #endif
   2085 
   2086 	ps->ps_destroy_msg(pm);
   2087 	error = 0;
   2088 
   2089 out:
   2090 	/*
   2091 	 * ENOSYS is not returned to kernel,
   2092 	 */
   2093 	if (error == ENOSYS)
   2094 		error = 0;
   2095 
   2096 	node_rele(opc);
   2097 	return error;
   2098 }
   2099 
   2100 int
   2101 perfuse_node_remove(struct puffs_usermount *pu, puffs_cookie_t opc,
   2102 	puffs_cookie_t targ, const struct puffs_cn *pcn)
   2103 {
   2104 	struct perfuse_state *ps;
   2105 	struct perfuse_node_data *pnd;
   2106 	perfuse_msg_t *pm;
   2107 	char *path;
   2108 	const char *name;
   2109 	size_t len;
   2110 	int error;
   2111 
   2112 	pnd = PERFUSE_NODE_DATA(opc);
   2113 
   2114 	if ((pnd->pnd_flags & PND_REMOVED) ||
   2115 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
   2116 		return ENOENT;
   2117 
   2118 #ifdef PERFUSE_DEBUG
   2119 	if (targ == NULL)
   2120 		DERRX(EX_SOFTWARE, "%s: targ is NULL", __func__);
   2121 
   2122 	if (perfuse_diagflags & (PDF_FH|PDF_FILENAME))
   2123 		DPRINTF("%s: opc = %p, remove opc = %p, file = \"%s\"\n",
   2124 			__func__, (void *)opc, (void *)targ, pcn->pcn_name);
   2125 #endif
   2126 	node_ref(opc);
   2127 	node_ref(targ);
   2128 
   2129 	/*
   2130 	 * Await for all operations on the deleted node to drain,
   2131 	 * as the filesystem may be confused to have it deleted
   2132 	 * during a getattr
   2133 	 */
   2134 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
   2135 		requeue_request(pu, targ, PCQ_AFTERXCHG);
   2136 
   2137 	ps = puffs_getspecific(pu);
   2138 	pnd = PERFUSE_NODE_DATA(opc);
   2139 	name = pcn->pcn_name;
   2140 	len = pcn->pcn_namelen + 1;
   2141 
   2142 	pm = ps->ps_new_msg(pu, opc, FUSE_UNLINK, len, pcn->pcn_cred);
   2143 	path = _GET_INPAYLOAD(ps, pm, char *);
   2144 	(void)strlcpy(path, name, len);
   2145 
   2146 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2147 		goto out;
   2148 
   2149 	perfuse_cache_flush(targ);
   2150 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
   2151 
   2152 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
   2153 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
   2154 
   2155 	/*
   2156 	 * The parent directory needs a sync
   2157 	 */
   2158 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
   2159 
   2160 #ifdef PERFUSE_DEBUG
   2161 	if (perfuse_diagflags & PDF_FILENAME)
   2162 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
   2163 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
   2164 			pcn->pcn_name);
   2165 #endif
   2166 	ps->ps_destroy_msg(pm);
   2167 	error = 0;
   2168 
   2169 out:
   2170 	node_rele(opc);
   2171 	node_rele(targ);
   2172 	return error;
   2173 }
   2174 
   2175 int
   2176 perfuse_node_link(struct puffs_usermount *pu, puffs_cookie_t opc,
   2177 	puffs_cookie_t targ, const struct puffs_cn *pcn)
   2178 {
   2179 	struct perfuse_state *ps;
   2180 	perfuse_msg_t *pm;
   2181 	const char *name;
   2182 	size_t len;
   2183 	struct puffs_node *pn;
   2184 	struct fuse_link_in *fli;
   2185 	int error;
   2186 
   2187 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   2188 		return ENOENT;
   2189 
   2190 	node_ref(opc);
   2191 	node_ref(targ);
   2192 	ps = puffs_getspecific(pu);
   2193 	pn = (struct puffs_node *)targ;
   2194 	name = pcn->pcn_name;
   2195 	len =  sizeof(*fli) + pcn->pcn_namelen + 1;
   2196 
   2197 	pm = ps->ps_new_msg(pu, opc, FUSE_LINK, len, pcn->pcn_cred);
   2198 	fli = GET_INPAYLOAD(ps, pm, fuse_link_in);
   2199 	fli->oldnodeid = PERFUSE_NODE_DATA(pn)->pnd_nodeid;
   2200 	(void)strlcpy((char *)(void *)(fli + 1), name, len - sizeof(*fli));
   2201 
   2202 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2203 		goto out;
   2204 
   2205 	ps->ps_destroy_msg(pm);
   2206 	error = 0;
   2207 
   2208 out:
   2209 	node_rele(opc);
   2210 	node_rele(targ);
   2211 	return error;
   2212 }
   2213 
   2214 int
   2215 perfuse_node_rename(struct puffs_usermount *pu, puffs_cookie_t opc,
   2216 	puffs_cookie_t src, const struct puffs_cn *pcn_src,
   2217 	puffs_cookie_t targ_dir, puffs_cookie_t targ,
   2218 	const struct puffs_cn *pcn_targ)
   2219 {
   2220 	struct perfuse_state *ps;
   2221 	struct perfuse_node_data *dstdir_pnd;
   2222 	perfuse_msg_t *pm;
   2223 	struct fuse_rename_in *fri;
   2224 	const char *newname;
   2225 	const char *oldname;
   2226 	char *np;
   2227 	int error;
   2228 	size_t len;
   2229 	size_t newname_len;
   2230 	size_t oldname_len;
   2231 
   2232 	if ((PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED) ||
   2233 	    (PERFUSE_NODE_DATA(src)->pnd_flags & PND_REMOVED) ||
   2234 	    (PERFUSE_NODE_DATA(targ_dir)->pnd_flags & PND_REMOVED))
   2235 		return ENOENT;
   2236 
   2237 	node_ref(opc);
   2238 	node_ref(src);
   2239 
   2240 	/*
   2241 	 * Await for all operations on the deleted node to drain,
   2242 	 * as the filesystem may be confused to have it deleted
   2243 	 * during a getattr
   2244 	 */
   2245 	if ((struct puffs_node *)targ != NULL) {
   2246 		node_ref(targ);
   2247 		while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
   2248 			requeue_request(pu, targ, PCQ_AFTERXCHG);
   2249 	} else {
   2250 		while (PERFUSE_NODE_DATA(src)->pnd_inxchg)
   2251 			requeue_request(pu, src, PCQ_AFTERXCHG);
   2252 	}
   2253 
   2254 	ps = puffs_getspecific(pu);
   2255 	newname =  pcn_targ->pcn_name;
   2256 	newname_len = pcn_targ->pcn_namelen + 1;
   2257 	oldname =  pcn_src->pcn_name;
   2258 	oldname_len = pcn_src->pcn_namelen + 1;
   2259 
   2260 	len = sizeof(*fri) + oldname_len + newname_len;
   2261 	pm = ps->ps_new_msg(pu, opc, FUSE_RENAME, len, pcn_targ->pcn_cred);
   2262 	fri = GET_INPAYLOAD(ps, pm, fuse_rename_in);
   2263 	fri->newdir = PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid;
   2264 	np = (char *)(void *)(fri + 1);
   2265 	(void)strlcpy(np, oldname, oldname_len);
   2266 	np += oldname_len;
   2267 	(void)strlcpy(np, newname, newname_len);
   2268 
   2269 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2270 		goto out;
   2271 
   2272 
   2273 	/*
   2274 	 * Record new parent nodeid
   2275 	 */
   2276 	dstdir_pnd = PERFUSE_NODE_DATA(targ_dir);
   2277 	PERFUSE_NODE_DATA(src)->pnd_parent_nodeid = dstdir_pnd->pnd_nodeid;
   2278 
   2279 	if (opc != targ_dir)
   2280 		dstdir_pnd->pnd_flags |= PND_DIRTY;
   2281 
   2282 	if (strcmp(newname, "..") != 0)
   2283 		(void)strlcpy(PERFUSE_NODE_DATA(src)->pnd_name,
   2284 		    newname, MAXPATHLEN);
   2285 	else
   2286 		PERFUSE_NODE_DATA(src)->pnd_name[0] = 0; /* forget name */
   2287 
   2288 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
   2289 
   2290 	if ((struct puffs_node *)targ != NULL) {
   2291 		perfuse_cache_flush(targ);
   2292 		PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
   2293 	}
   2294 
   2295 #ifdef PERFUSE_DEBUG
   2296 	if (perfuse_diagflags & PDF_FILENAME)
   2297 		DPRINTF("%s: nodeid = 0x%"PRIx64" file = \"%s\" renamed \"%s\" "
   2298 			"nodeid = 0x%"PRIx64" -> nodeid = 0x%"PRIx64" \"%s\"\n",
   2299 	 		__func__, PERFUSE_NODE_DATA(src)->pnd_nodeid,
   2300 			pcn_src->pcn_name, pcn_targ->pcn_name,
   2301 			PERFUSE_NODE_DATA(opc)->pnd_nodeid,
   2302 			PERFUSE_NODE_DATA(targ_dir)->pnd_nodeid,
   2303 			perfuse_node_path(ps, targ_dir));
   2304 #endif
   2305 
   2306 	ps->ps_destroy_msg(pm);
   2307 	error = 0;
   2308 
   2309 out:
   2310 	node_rele(opc);
   2311 	node_rele(src);
   2312 	if ((struct puffs_node *)targ != NULL)
   2313 		node_rele(targ);
   2314 
   2315 	return error;
   2316 }
   2317 
   2318 int
   2319 perfuse_node_mkdir(struct puffs_usermount *pu, puffs_cookie_t opc,
   2320 	struct puffs_newinfo *pni, const struct puffs_cn *pcn,
   2321 	const struct vattr *vap)
   2322 {
   2323 	struct perfuse_state *ps;
   2324 	perfuse_msg_t *pm;
   2325 	struct fuse_mkdir_in *fmi;
   2326 	const char *path;
   2327 	size_t len;
   2328 	int error;
   2329 
   2330 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   2331 		return ENOENT;
   2332 
   2333 	node_ref(opc);
   2334 	ps = puffs_getspecific(pu);
   2335 	path = pcn->pcn_name;
   2336 	len = sizeof(*fmi) + pcn->pcn_namelen + 1;
   2337 
   2338 	pm = ps->ps_new_msg(pu, opc, FUSE_MKDIR, len, pcn->pcn_cred);
   2339 	fmi = GET_INPAYLOAD(ps, pm, fuse_mkdir_in);
   2340 	fmi->mode = vap->va_mode;
   2341 	fmi->umask = 0; 	/* Seems unused by libfuse? */
   2342 	(void)strlcpy((char *)(void *)(fmi + 1), path, len - sizeof(*fmi));
   2343 
   2344 	error = node_mk_common(pu, opc, pni, pcn, pm);
   2345 
   2346 	node_rele(opc);
   2347 	return error;
   2348 }
   2349 
   2350 
   2351 int
   2352 perfuse_node_rmdir(struct puffs_usermount *pu, puffs_cookie_t opc,
   2353 	puffs_cookie_t targ, const struct puffs_cn *pcn)
   2354 {
   2355 	struct perfuse_state *ps;
   2356 	struct perfuse_node_data *pnd;
   2357 	perfuse_msg_t *pm;
   2358 	char *path;
   2359 	const char *name;
   2360 	size_t len;
   2361 	int error;
   2362 
   2363 	pnd = PERFUSE_NODE_DATA(opc);
   2364 
   2365 	if ((pnd->pnd_flags & PND_REMOVED) ||
   2366 	    (PERFUSE_NODE_DATA(targ)->pnd_flags & PND_REMOVED))
   2367 		return ENOENT;
   2368 
   2369 	/*
   2370 	 * Attempt to rmdir dir/.. should raise ENOTEMPTY
   2371 	 */
   2372 	if (PERFUSE_NODE_DATA(targ)->pnd_nodeid == pnd->pnd_parent_nodeid)
   2373 		return ENOTEMPTY;
   2374 
   2375 	node_ref(opc);
   2376 	node_ref(targ);
   2377 
   2378 	/*
   2379 	 * Await for all operations on the deleted node to drain,
   2380 	 * as the filesystem may be confused to have it deleted
   2381 	 * during a getattr
   2382 	 */
   2383 	while (PERFUSE_NODE_DATA(targ)->pnd_inxchg)
   2384 		requeue_request(pu, targ, PCQ_AFTERXCHG);
   2385 
   2386 	ps = puffs_getspecific(pu);
   2387 	name = pcn->pcn_name;
   2388 	len = pcn->pcn_namelen + 1;
   2389 
   2390 	pm = ps->ps_new_msg(pu, opc, FUSE_RMDIR, len, pcn->pcn_cred);
   2391 	path = _GET_INPAYLOAD(ps, pm, char *);
   2392 	(void)strlcpy(path, name, len);
   2393 
   2394 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2395 		goto out;
   2396 
   2397 	perfuse_cache_flush(targ);
   2398 	PERFUSE_NODE_DATA(targ)->pnd_flags |= PND_REMOVED;
   2399 
   2400 	if (!(PERFUSE_NODE_DATA(targ)->pnd_flags & PND_OPEN))
   2401 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N2);
   2402 
   2403 	/*
   2404 	 * The parent directory needs a sync
   2405 	 */
   2406 	PERFUSE_NODE_DATA(opc)->pnd_flags |= PND_DIRTY;
   2407 
   2408 #ifdef PERFUSE_DEBUG
   2409 	if (perfuse_diagflags & PDF_FILENAME)
   2410 		DPRINTF("%s: remove nodeid = 0x%"PRIx64" file = \"%s\"\n",
   2411 			__func__, PERFUSE_NODE_DATA(targ)->pnd_nodeid,
   2412 			perfuse_node_path(ps, targ));
   2413 #endif
   2414 	ps->ps_destroy_msg(pm);
   2415 	error = 0;
   2416 
   2417 out:
   2418 	node_rele(opc);
   2419 	node_rele(targ);
   2420 	return error;
   2421 }
   2422 
   2423 /* vap is unused */
   2424 /* ARGSUSED4 */
   2425 int
   2426 perfuse_node_symlink(struct puffs_usermount *pu, puffs_cookie_t opc,
   2427 	struct puffs_newinfo *pni, const struct puffs_cn *pcn_src,
   2428 	const struct vattr *vap, const char *link_target)
   2429 {
   2430 	struct perfuse_state *ps;
   2431 	perfuse_msg_t *pm;
   2432 	char *np;
   2433 	const char *path;
   2434 	size_t path_len;
   2435 	size_t linkname_len;
   2436 	size_t len;
   2437 	int error;
   2438 
   2439 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   2440 		return ENOENT;
   2441 
   2442 	node_ref(opc);
   2443 	ps = puffs_getspecific(pu);
   2444 	path = pcn_src->pcn_name;
   2445 	path_len = pcn_src->pcn_namelen + 1;
   2446 	linkname_len = strlen(link_target) + 1;
   2447 	len = path_len + linkname_len;
   2448 
   2449 	pm = ps->ps_new_msg(pu, opc, FUSE_SYMLINK, len, pcn_src->pcn_cred);
   2450 	np = _GET_INPAYLOAD(ps, pm, char *);
   2451 	(void)strlcpy(np, path, path_len);
   2452 	np += path_len;
   2453 	(void)strlcpy(np, link_target, linkname_len);
   2454 
   2455 	error = node_mk_common(pu, opc, pni, pcn_src, pm);
   2456 
   2457 	node_rele(opc);
   2458 	return error;
   2459 }
   2460 
   2461 /* ARGSUSED4 */
   2462 int
   2463 perfuse_node_readdir(struct puffs_usermount *pu, puffs_cookie_t opc,
   2464 	struct dirent *dent, off_t *readoff, size_t *reslen,
   2465 	const struct puffs_cred *pcr, int *eofflag, off_t *cookies,
   2466 	size_t *ncookies)
   2467 {
   2468 	perfuse_msg_t *pm;
   2469 	uint64_t fh;
   2470 	struct perfuse_state *ps;
   2471 	struct perfuse_node_data *pnd;
   2472 	struct fuse_read_in *fri;
   2473 	struct fuse_out_header *foh;
   2474 	struct fuse_dirent *fd;
   2475 	size_t foh_len;
   2476 	int error;
   2477 	size_t fd_maxlen;
   2478 
   2479 	error = 0;
   2480 	node_ref(opc);
   2481 	ps = puffs_getspecific(pu);
   2482 
   2483 	/*
   2484 	 * readdir state is kept at node level, and several readdir
   2485 	 * requests can be issued at the same time on the same node.
   2486 	 * We need to queue requests so that only one is in readdir
   2487 	 * code at the same time.
   2488 	 */
   2489 	pnd = PERFUSE_NODE_DATA(opc);
   2490 	while (pnd->pnd_flags & PND_INREADDIR)
   2491 		requeue_request(pu, opc, PCQ_READDIR);
   2492 	pnd->pnd_flags |= PND_INREADDIR;
   2493 
   2494 #ifdef PERFUSE_DEBUG
   2495 	if (perfuse_diagflags & PDF_READDIR)
   2496 		DPRINTF("%s: READDIR opc = %p enter critical section\n",
   2497 			__func__, (void *)opc);
   2498 #endif
   2499 	/*
   2500 	 * Re-initialize pnd->pnd_fd_cookie on the first readdir for a node
   2501 	 */
   2502 	if (*readoff == 0)
   2503 		pnd->pnd_fd_cookie = 0;
   2504 
   2505 	/*
   2506 	 * Do we already have the data buffered?
   2507 	 */
   2508 	if (pnd->pnd_dirent != NULL)
   2509 		goto out;
   2510 	pnd->pnd_dirent_len = 0;
   2511 
   2512 	/*
   2513 	 * It seems NetBSD can call readdir without open first
   2514 	 * libfuse will crash if it is done that way, hence open first.
   2515 	 */
   2516 	if (!(pnd->pnd_flags & PND_OPEN)) {
   2517 		if ((error = perfuse_node_open(pu, opc, FREAD, pcr)) != 0)
   2518 			goto out;
   2519 	}
   2520 
   2521 	fh = perfuse_get_fh(opc, FREAD);
   2522 
   2523 #ifdef PERFUSE_DEBUG
   2524 	if (perfuse_diagflags & PDF_FH)
   2525 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
   2526 			"rfh = 0x%"PRIx64"\n", __func__, (void *)opc,
   2527 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fh);
   2528 #endif
   2529 
   2530 	pnd->pnd_all_fd = NULL;
   2531 	pnd->pnd_all_fd_len = 0;
   2532 	fd_maxlen = ps->ps_max_readahead - sizeof(*foh);
   2533 
   2534 	do {
   2535 		size_t fd_len;
   2536 		char *afdp;
   2537 
   2538 		pm = ps->ps_new_msg(pu, opc, FUSE_READDIR, sizeof(*fri), pcr);
   2539 
   2540 		/*
   2541 		 * read_flags, lock_owner and flags are unused in libfuse
   2542 		 */
   2543 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
   2544 		fri->fh = fh;
   2545 		fri->offset = pnd->pnd_fd_cookie;
   2546 		fri->size = (uint32_t)fd_maxlen;
   2547 		fri->read_flags = 0;
   2548 		fri->lock_owner = 0;
   2549 		fri->flags = 0;
   2550 
   2551 		if ((error = xchg_msg(pu, opc, pm,
   2552 				      UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2553 			goto out;
   2554 
   2555 		/*
   2556 		 * There are many puffs_framebufs calls later,
   2557 		 * therefore foh will not be valid for a long time.
   2558 		 * Just get the length and forget it.
   2559 		 */
   2560 		foh = GET_OUTHDR(ps, pm);
   2561 		foh_len = foh->len;
   2562 
   2563 		/*
   2564 		 * Empty read: we reached the end of the buffer.
   2565 		 */
   2566 		if (foh_len == sizeof(*foh)) {
   2567 			ps->ps_destroy_msg(pm);
   2568 			*eofflag = 1;
   2569 			break;
   2570 		}
   2571 
   2572 		/*
   2573 		 * Check for corrupted message.
   2574 		 */
   2575 		if (foh_len < sizeof(*foh) + sizeof(*fd)) {
   2576 			ps->ps_destroy_msg(pm);
   2577 			DWARNX("readdir reply too short");
   2578 			error = EIO;
   2579 			goto out;
   2580 		}
   2581 
   2582 
   2583 		fd = GET_OUTPAYLOAD(ps, pm, fuse_dirent);
   2584 		fd_len = foh_len - sizeof(*foh);
   2585 
   2586 		pnd->pnd_all_fd = realloc(pnd->pnd_all_fd,
   2587 					  pnd->pnd_all_fd_len + fd_len);
   2588 		if (pnd->pnd_all_fd  == NULL)
   2589 			DERR(EX_OSERR, "%s: malloc failed", __func__);
   2590 
   2591 		afdp = (char *)(void *)pnd->pnd_all_fd + pnd->pnd_all_fd_len;
   2592 		(void)memcpy(afdp, fd, fd_len);
   2593 
   2594 		pnd->pnd_all_fd_len += fd_len;
   2595 
   2596 		/*
   2597 		 * The fd->off field is used as a cookie for
   2598 		 * resuming the next readdir() where this one was left.
   2599 	 	 */
   2600 		pnd->pnd_fd_cookie = readdir_last_cookie(fd, fd_len);
   2601 
   2602 		ps->ps_destroy_msg(pm);
   2603 	} while (1 /* CONSTCOND */);
   2604 
   2605 	if (pnd->pnd_all_fd != NULL) {
   2606 		if (fuse_to_dirent(pu, opc, pnd->pnd_all_fd,
   2607 				   pnd->pnd_all_fd_len) == -1)
   2608 			error = EIO;
   2609 	}
   2610 
   2611 out:
   2612 	if (pnd->pnd_all_fd != NULL) {
   2613 		free(pnd->pnd_all_fd);
   2614 		pnd->pnd_all_fd = NULL;
   2615 		pnd->pnd_all_fd_len = 0;
   2616 	}
   2617 
   2618 	if (error == 0)
   2619 		readdir_buffered(opc, dent, readoff, reslen);
   2620 
   2621 	/*
   2622 	 * Schedule queued readdir requests
   2623 	 */
   2624 	pnd->pnd_flags &= ~PND_INREADDIR;
   2625 	(void)dequeue_requests(opc, PCQ_READDIR, DEQUEUE_ALL);
   2626 
   2627 #ifdef PERFUSE_DEBUG
   2628 	if (perfuse_diagflags & PDF_READDIR)
   2629 		DPRINTF("%s: READDIR opc = %p exit critical section\n",
   2630 			__func__, (void *)opc);
   2631 #endif
   2632 
   2633 	node_rele(opc);
   2634 	return error;
   2635 }
   2636 
   2637 int
   2638 perfuse_node_readlink(struct puffs_usermount *pu, puffs_cookie_t opc,
   2639 	const struct puffs_cred *pcr, char *linkname, size_t *linklen)
   2640 {
   2641 	struct perfuse_state *ps;
   2642 	perfuse_msg_t *pm;
   2643 	int error;
   2644 	size_t len;
   2645 	struct fuse_out_header *foh;
   2646 
   2647 	if (PERFUSE_NODE_DATA(opc)->pnd_flags & PND_REMOVED)
   2648 		return ENOENT;
   2649 
   2650 	node_ref(opc);
   2651 	ps = puffs_getspecific(pu);
   2652 
   2653 	pm = ps->ps_new_msg(pu, opc, FUSE_READLINK, 0, pcr);
   2654 
   2655 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   2656 		goto out;
   2657 
   2658 	foh = GET_OUTHDR(ps, pm);
   2659 	len = foh->len - sizeof(*foh);
   2660 	if (len > *linklen)
   2661 		DERRX(EX_PROTOCOL, "path len = %zd too long", len);
   2662 	if (len == 0)
   2663 		DERRX(EX_PROTOCOL, "path len = %zd too short", len);
   2664 
   2665 	(void)memcpy(linkname, _GET_OUTPAYLOAD(ps, pm, char *), len);
   2666 
   2667 	/*
   2668 	 * FUSE filesystems return a NUL terminated string, we
   2669 	 * do not want the trailing \0
   2670 	 */
   2671 	while (len > 0 && linkname[len - 1] == '\0')
   2672 		len--;
   2673 
   2674 	*linklen = len;
   2675 
   2676 	ps->ps_destroy_msg(pm);
   2677 	error = 0;
   2678 
   2679 out:
   2680 	node_rele(opc);
   2681 	return error;
   2682 }
   2683 
   2684 int
   2685 perfuse_node_reclaim2(struct puffs_usermount *pu,
   2686 		      puffs_cookie_t opc, int nlookup)
   2687 {
   2688 	struct perfuse_state *ps;
   2689 	perfuse_msg_t *pm;
   2690 	struct perfuse_node_data *pnd;
   2691 	struct fuse_forget_in *ffi;
   2692 
   2693 #ifdef PERFUSE_DEBUG
   2694 		if (perfuse_diagflags & PDF_RECLAIM)
   2695 			DPRINTF("%s called with opc = %p, nlookup = %d\n",
   2696 				__func__, (void *)opc, nlookup);
   2697 #endif
   2698 	if (opc == 0 || nlookup == 0) {
   2699 		return 0;
   2700 	}
   2701 
   2702 	ps = puffs_getspecific(pu);
   2703 	pnd = PERFUSE_NODE_DATA(opc);
   2704 
   2705 	/*
   2706 	 * Never forget the root.
   2707 	 */
   2708 	if (pnd->pnd_nodeid == FUSE_ROOT_ID)
   2709 		return 0;
   2710 
   2711 #ifdef PERFUSE_DEBUG
   2712 	if (perfuse_diagflags & PDF_RECLAIM)
   2713 		DPRINTF("%s (nodeid %"PRId64") reclaimed, nlookup = %d/%d\n",
   2714 			perfuse_node_path(ps, opc), pnd->pnd_nodeid,
   2715 			nlookup, pnd->pnd_puffs_nlookup);
   2716 #endif
   2717 	/*
   2718 	 * The kernel tells us how many lookups it made, which allows
   2719 	 * us to detect that we have an uncompleted lookup and that the
   2720 	 * node should not disappear.
   2721 	 */
   2722 	pnd->pnd_puffs_nlookup -= nlookup;
   2723 	if (pnd->pnd_puffs_nlookup > 0)
   2724 		return 0;
   2725 
   2726 	node_ref(opc);
   2727 	pnd->pnd_flags |= PND_RECLAIMED;
   2728 
   2729 #ifdef PERFUSE_DEBUG
   2730 	if (perfuse_diagflags & PDF_RECLAIM)
   2731 		DPRINTF("%s (nodeid %"PRId64") is %sreclaimed, nlookup = %d "
   2732 			"%s%s%s%s, pending ops:%s%s%s\n",
   2733 		        perfuse_node_path(ps, opc), pnd->pnd_nodeid,
   2734 		        pnd->pnd_flags & PND_RECLAIMED ? "" : "not ",
   2735 			pnd->pnd_puffs_nlookup,
   2736 			pnd->pnd_flags & PND_OPEN ? "open " : "not open",
   2737 			pnd->pnd_flags & PND_RFH ? "r" : "",
   2738 			pnd->pnd_flags & PND_WFH ? "w" : "",
   2739 			pnd->pnd_flags & PND_BUSY ? " busy" : "",
   2740 			pnd->pnd_flags & PND_INREADDIR ? " readdir" : "",
   2741 			pnd->pnd_flags & PND_INWRITE ? " write" : "",
   2742 			pnd->pnd_flags & PND_INOPEN ? " open" : "");
   2743 #endif
   2744 	/*
   2745 	 * Make sure it is not looked up again
   2746 	 */
   2747 	if (!(pnd->pnd_flags & PND_REMOVED))
   2748 		perfuse_cache_flush(opc);
   2749 
   2750 	/*
   2751 	 * Purge any activity on the node, while checking
   2752 	 * that it remains eligible for a reclaim.
   2753 	 */
   2754 	while (pnd->pnd_ref > 1)
   2755 		requeue_request(pu, opc, PCQ_REF);
   2756 
   2757 #ifdef PERFUSE_DEBUG
   2758 	if ((pnd->pnd_flags & PND_OPEN) ||
   2759 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
   2760 		DERRX(EX_SOFTWARE, "%s: opc = %p \"%s\": still open",
   2761 		      __func__, opc, pnd->pnd_name);
   2762 
   2763 	if ((pnd->pnd_flags & PND_BUSY) ||
   2764 	       !TAILQ_EMPTY(&pnd->pnd_pcq))
   2765 		DERRX(EX_SOFTWARE, "%s: opc = %p: queued operations",
   2766 		      __func__, opc);
   2767 
   2768 	if (pnd->pnd_inxchg != 0)
   2769 		DERRX(EX_SOFTWARE, "%s: opc = %p: ongoing operations",
   2770 		      __func__, opc);
   2771 #endif
   2772 
   2773 	/*
   2774 	 * Send the FORGET message
   2775 	 *
   2776 	 * ps_new_msg() is called with NULL creds, which will
   2777 	 * be interpreted as FUSE superuser. This is obviously
   2778 	 * fine since we operate with kernel creds here.
   2779 	 */
   2780 	pm = ps->ps_new_msg(pu, opc, FUSE_FORGET,
   2781 		      sizeof(*ffi), NULL);
   2782 	ffi = GET_INPAYLOAD(ps, pm, fuse_forget_in);
   2783 	ffi->nlookup = pnd->pnd_fuse_nlookup;
   2784 
   2785 	/*
   2786 	 * No reply is expected, pm is freed in xchg_msg
   2787 	 */
   2788 	(void)xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, no_reply);
   2789 
   2790 	perfuse_destroy_pn(pu, opc);
   2791 
   2792 	return 0;
   2793 }
   2794 
   2795 int
   2796 perfuse_node_reclaim(struct puffs_usermount *pu, puffs_cookie_t opc)
   2797 {
   2798 #ifdef PERFUSE_DEBUG
   2799 	if (perfuse_diagflags & PDF_RECLAIM)
   2800 		DPRINTF("perfuse_node_reclaim called\n");
   2801 #endif
   2802 	return perfuse_node_reclaim2(pu, opc, 1);
   2803 }
   2804 
   2805 int
   2806 perfuse_node_inactive(struct puffs_usermount *pu, puffs_cookie_t opc)
   2807 {
   2808 	struct perfuse_node_data *pnd;
   2809 	int error;
   2810 
   2811 	if (opc == 0)
   2812 		return 0;
   2813 
   2814 	pnd = PERFUSE_NODE_DATA(opc);
   2815 	if (!(pnd->pnd_flags & (PND_OPEN|PND_REMOVED)))
   2816 		return 0;
   2817 
   2818 	node_ref(opc);
   2819 
   2820 	/*
   2821 	 * Make sure all operation are finished
   2822 	 * There can be an ongoing write. Other
   2823 	 * operation wait for all data before
   2824 	 * the close/inactive.
   2825 	 */
   2826 	while (pnd->pnd_flags & PND_INWRITE)
   2827 		requeue_request(pu, opc, PCQ_AFTERWRITE);
   2828 
   2829 	/*
   2830 	 * The inactive operation may be cancelled,
   2831 	 * If no open is in progress, set PND_INOPEN
   2832 	 * so that a new open will be queued.
   2833 	 */
   2834 	if (pnd->pnd_flags & PND_INOPEN)
   2835 		goto out;
   2836 
   2837 	pnd->pnd_flags |= PND_INOPEN;
   2838 
   2839 	/*
   2840 	 * Sync data
   2841 	 */
   2842 	if (pnd->pnd_flags & PND_DIRTY) {
   2843 		if ((error = perfuse_node_fsync(pu, opc, NULL, 0, 0, 0)) != 0)
   2844 			DWARN("%s: perfuse_node_fsync failed error = %d",
   2845 			      __func__, error);
   2846 	}
   2847 
   2848 
   2849 	/*
   2850 	 * Close handles
   2851 	 */
   2852 	if (pnd->pnd_flags & PND_WFH) {
   2853 		if ((error = perfuse_node_close_common(pu, opc, FWRITE)) != 0)
   2854 			DWARN("%s: close write FH failed error = %d",
   2855 			      __func__, error);
   2856 	}
   2857 
   2858 	if (pnd->pnd_flags & PND_RFH) {
   2859 		if ((error = perfuse_node_close_common(pu, opc, FREAD)) != 0)
   2860 			DWARN("%s: close read FH failed error = %d",
   2861 			      __func__, error);
   2862 	}
   2863 
   2864 	/*
   2865 	 * This will cause a reclaim to be sent
   2866 	 */
   2867 	if (pnd->pnd_flags & PND_REMOVED)
   2868 		puffs_setback(puffs_cc_getcc(pu), PUFFS_SETBACK_NOREF_N1);
   2869 
   2870 	/*
   2871 	 * Schedule awaiting operations
   2872 	 */
   2873 	pnd->pnd_flags &= ~PND_INOPEN;
   2874 	(void)dequeue_requests(opc, PCQ_OPEN, DEQUEUE_ALL);
   2875 
   2876 	/*
   2877 	 * errors are ignored, since the kernel ignores the return code.
   2878 	 */
   2879 out:
   2880 	node_rele(opc);
   2881 	return 0;
   2882 }
   2883 
   2884 
   2885 /* ARGSUSED0 */
   2886 int
   2887 perfuse_node_print(struct puffs_usermount *pu, puffs_cookie_t opc)
   2888 {
   2889 	DERRX(EX_SOFTWARE, "%s: UNIMPLEMENTED (FATAL)", __func__);
   2890 	return 0;
   2891 }
   2892 
   2893 int
   2894 perfuse_node_pathconf(struct puffs_usermount *pu, puffs_cookie_t opc,
   2895 	int name, register_t *retval)
   2896 {
   2897 	perfuse_msg_t *pm;
   2898 	struct perfuse_state *ps;
   2899 	struct fuse_statfs_out *fso;
   2900 	int error = 0;
   2901 
   2902 	/*
   2903 	 * Static values copied from UFS
   2904 	 * in src/sys/ufs/ufs/ufs_vnops.c
   2905 	 */
   2906 	switch (name) {
   2907 	case _PC_LINK_MAX:
   2908 		*retval = LINK_MAX;
   2909 		break;
   2910 	case _PC_PATH_MAX:
   2911 		*retval = PATH_MAX;
   2912 		break;
   2913 	case _PC_PIPE_BUF:
   2914 		*retval = PIPE_BUF;
   2915 		break;
   2916 	case _PC_CHOWN_RESTRICTED:
   2917 		*retval = 1;
   2918 		break;
   2919 	case _PC_NO_TRUNC:
   2920 		*retval = 1;
   2921 		break;
   2922 	case _PC_SYNC_IO:
   2923 		*retval = 1;
   2924 		break;
   2925 	case _PC_FILESIZEBITS:
   2926 		*retval = 42;
   2927 		break;
   2928 	case _PC_SYMLINK_MAX:
   2929 		*retval = MAXPATHLEN;
   2930 		break;
   2931 	case _PC_2_SYMLINKS:
   2932 		*retval = 1;
   2933 		break;
   2934 	case _PC_NAME_MAX:
   2935 		ps = puffs_getspecific(pu);
   2936 		pm = ps->ps_new_msg(pu, opc, FUSE_STATFS, 0, NULL);
   2937 
   2938 		error = xchg_msg(pu, opc, pm, sizeof(*fso), wait_reply);
   2939 		if (error != 0)
   2940 			return error;
   2941 
   2942 		fso = GET_OUTPAYLOAD(ps, pm, fuse_statfs_out);
   2943 		*retval = fso->st.namelen;
   2944 
   2945 		ps->ps_destroy_msg(pm);
   2946 
   2947 		break;
   2948 	default:
   2949 		DWARN("Unimplemented pathconf for name = %d", name);
   2950 		error = ENOSYS;
   2951 		break;
   2952 	}
   2953 
   2954 	return error;
   2955 }
   2956 
   2957 int
   2958 perfuse_node_advlock(struct puffs_usermount *pu, puffs_cookie_t opc,
   2959 	void *id, int op, struct flock *fl, int flags)
   2960 {
   2961 	struct perfuse_state *ps;
   2962 	int fop;
   2963 	perfuse_msg_t *pm;
   2964 	uint64_t fh;
   2965 	struct fuse_lk_in *fli;
   2966 	struct fuse_out_header *foh;
   2967 	struct fuse_lk_out *flo;
   2968 	uint32_t owner;
   2969 	size_t len;
   2970 	int error;
   2971 
   2972 	node_ref(opc);
   2973 
   2974 	/*
   2975 	 * Make sure we do have a filehandle, as the FUSE filesystem
   2976 	 * expect one. E.g.: if we provide none, GlusterFS logs an error
   2977 	 * "0-glusterfs-fuse: xl is NULL"
   2978 	 *
   2979 	 * We need the read file handle if the file is open read only,
   2980 	 * in order to support shared locks on read-only files.
   2981 	 * NB: The kernel always sends advlock for read-only
   2982 	 * files at exit time when the process used lock, see
   2983 	 * sys_exit -> exit1 -> fd_free -> fd_close -> VOP_ADVLOCK
   2984 	 */
   2985 	if ((fh = perfuse_get_fh(opc, FREAD)) == FUSE_UNKNOWN_FH) {
   2986 		error = EBADF;
   2987 		goto out;
   2988 	}
   2989 
   2990 	ps = puffs_getspecific(pu);
   2991 
   2992 	if (op == F_GETLK)
   2993 		fop = FUSE_GETLK;
   2994 	else
   2995 		fop = (flags & F_WAIT) ? FUSE_SETLKW : FUSE_SETLK;
   2996 
   2997 	/*
   2998 	 * XXX ps_new_msg() is called with NULL creds, which will
   2999 	 * be interpreted as FUSE superuser. We have no way to
   3000 	 * know the requesting process' credential, but since advlock()
   3001 	 * is supposed to operate on a file that has been open(),
   3002 	 * permission should have already been checked at open() time.
   3003 	 */
   3004 	pm = ps->ps_new_msg(pu, opc, fop, sizeof(*fli), NULL);
   3005 	fli = GET_INPAYLOAD(ps, pm, fuse_lk_in);
   3006 	fli->fh = fh;
   3007 	fli->owner = (uint64_t)(vaddr_t)id;
   3008 	fli->lk.start = fl->l_start;
   3009 	fli->lk.end = fl->l_start + fl->l_len;
   3010 	fli->lk.type = fl->l_type;
   3011 	fli->lk.pid = fl->l_pid;
   3012 	fli->lk_flags = (flags & F_FLOCK) ? FUSE_LK_FLOCK : 0;
   3013 
   3014 	owner = (uint32_t)(vaddr_t)id;
   3015 
   3016 #ifdef PERFUSE_DEBUG
   3017 	if (perfuse_diagflags & PDF_FH)
   3018 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
   3019 			__func__, (void *)opc,
   3020 			PERFUSE_NODE_DATA(opc)->pnd_nodeid, fli->fh);
   3021 #endif
   3022 
   3023 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   3024 		goto out;
   3025 
   3026 	foh = GET_OUTHDR(ps, pm);
   3027 	len = foh->len - sizeof(*foh);
   3028 
   3029 	/*
   3030 	 * Save or clear the lock
   3031 	 */
   3032 	switch (op) {
   3033 	case F_GETLK:
   3034 		if (len != sizeof(*flo))
   3035 			DERRX(EX_SOFTWARE,
   3036 			      "%s: Unexpected lock reply len %zd",
   3037 			      __func__, len);
   3038 
   3039 		flo = GET_OUTPAYLOAD(ps, pm, fuse_lk_out);
   3040 		fl->l_start = flo->lk.start;
   3041 		fl->l_len = flo->lk.end - flo->lk.start;
   3042 		fl->l_pid = flo->lk.pid;
   3043 		fl->l_type = flo->lk.type;
   3044 		fl->l_whence = SEEK_SET;	/* libfuse hardcodes it */
   3045 
   3046 		PERFUSE_NODE_DATA(opc)->pnd_lock_owner = flo->lk.pid;
   3047 		break;
   3048 	case F_UNLCK:
   3049 		owner = 0;
   3050 		/* FALLTHROUGH */
   3051 	case F_SETLK:
   3052 		/* FALLTHROUGH */
   3053 	case F_SETLKW:
   3054 		if (error != 0)
   3055 			PERFUSE_NODE_DATA(opc)->pnd_lock_owner = owner;
   3056 
   3057 		if (len != 0)
   3058 			DERRX(EX_SOFTWARE,
   3059 			      "%s: Unexpected unlock reply len %zd",
   3060 			      __func__, len);
   3061 
   3062 		break;
   3063 	default:
   3064 		DERRX(EX_SOFTWARE, "%s: Unexpected op %d", __func__, op);
   3065 		break;
   3066 	}
   3067 
   3068 	ps->ps_destroy_msg(pm);
   3069 	error = 0;
   3070 
   3071 out:
   3072 	node_rele(opc);
   3073 	return error;
   3074 }
   3075 
   3076 int
   3077 perfuse_node_read(struct puffs_usermount *pu, puffs_cookie_t opc, uint8_t *buf,
   3078 	off_t offset, size_t *resid, const struct puffs_cred *pcr, int ioflag)
   3079 {
   3080 	struct perfuse_state *ps;
   3081 	struct perfuse_node_data *pnd;
   3082 	const struct vattr *vap;
   3083 	perfuse_msg_t *pm;
   3084 	uint64_t fh;
   3085 	struct fuse_read_in *fri;
   3086 	struct fuse_out_header *foh;
   3087 	size_t readen;
   3088 	int error;
   3089 
   3090 	ps = puffs_getspecific(pu);
   3091 	pnd = PERFUSE_NODE_DATA(opc);
   3092 	vap = puffs_pn_getvap((struct puffs_node *)opc);
   3093 
   3094 	/*
   3095 	 * NetBSD turns that into a getdents(2) output
   3096 	 * We just do a EISDIR as this feature is of little use.
   3097 	 */
   3098 	if (vap->va_type == VDIR)
   3099 		return EISDIR;
   3100 
   3101 	fh =  perfuse_get_fh(opc, FREAD); /* Cannot be VDIR */
   3102 
   3103 	do {
   3104 		size_t max_read;
   3105 
   3106 		max_read = ps->ps_max_readahead - sizeof(*foh);
   3107 		/*
   3108 		 * flags may be set to FUSE_READ_LOCKOWNER
   3109 		 * if lock_owner is provided.
   3110 		 */
   3111 		pm = ps->ps_new_msg(pu, opc, FUSE_READ, sizeof(*fri), pcr);
   3112 		fri = GET_INPAYLOAD(ps, pm, fuse_read_in);
   3113 		fri->fh = fh;
   3114 		fri->offset = offset;
   3115 		fri->size = (uint32_t)MIN(*resid, max_read);
   3116 		fri->read_flags = 0; /* XXX Unused by libfuse? */
   3117 		fri->lock_owner = pnd->pnd_lock_owner;
   3118 		fri->flags = 0;
   3119 		fri->flags |= (fri->lock_owner != 0) ? FUSE_READ_LOCKOWNER : 0;
   3120 
   3121 #ifdef PERFUSE_DEBUG
   3122 	if (perfuse_diagflags & PDF_FH)
   3123 		DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", fh = 0x%"PRIx64"\n",
   3124 			__func__, (void *)opc, pnd->pnd_nodeid, fri->fh);
   3125 #endif
   3126 		error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply);
   3127 		if (error  != 0)
   3128 			return error;
   3129 
   3130 		foh = GET_OUTHDR(ps, pm);
   3131 		readen = foh->len - sizeof(*foh);
   3132 
   3133 #ifdef PERFUSE_DEBUG
   3134 		if (readen > *resid)
   3135 			DERRX(EX_SOFTWARE, "%s: Unexpected big read %zd",
   3136 			      __func__, readen);
   3137 #endif
   3138 
   3139 		(void)memcpy(buf,  _GET_OUTPAYLOAD(ps, pm, char *), readen);
   3140 
   3141 		buf += readen;
   3142 		offset += readen;
   3143 		*resid -= readen;
   3144 
   3145 		ps->ps_destroy_msg(pm);
   3146 	} while ((*resid != 0) && (readen != 0));
   3147 
   3148 	if (ioflag & (IO_SYNC|IO_DSYNC))
   3149 		ps->ps_syncreads++;
   3150 	else
   3151 		ps->ps_asyncreads++;
   3152 
   3153 	return 0;
   3154 }
   3155 
   3156 int
   3157 perfuse_node_write(struct puffs_usermount *pu, puffs_cookie_t opc,
   3158 	uint8_t *buf, off_t offset, size_t *resid,
   3159 	const struct puffs_cred *pcr, int ioflag)
   3160 {
   3161 	return perfuse_node_write2(pu, opc, buf, offset, resid, pcr, ioflag, 0);
   3162 }
   3163 
   3164 /* ARGSUSED7 */
   3165 int
   3166 perfuse_node_write2(struct puffs_usermount *pu, puffs_cookie_t opc,
   3167 	uint8_t *buf, off_t offset, size_t *resid,
   3168 	const struct puffs_cred *pcr, int ioflag, int xflag)
   3169 {
   3170 	struct perfuse_state *ps;
   3171 	struct perfuse_node_data *pnd;
   3172 	struct vattr *vap;
   3173 	perfuse_msg_t *pm;
   3174 	uint64_t fh;
   3175 	struct fuse_write_in *fwi;
   3176 	struct fuse_write_out *fwo;
   3177 	size_t data_len;
   3178 	size_t payload_len;
   3179 	size_t written;
   3180 	int inresize;
   3181 	int error;
   3182 
   3183 	ps = puffs_getspecific(pu);
   3184 	pnd = PERFUSE_NODE_DATA(opc);
   3185 	vap = puffs_pn_getvap((struct puffs_node *)opc);
   3186 	written = 0;
   3187 	inresize = 0;
   3188 	error = 0;
   3189 
   3190 	if (vap->va_type == VDIR)
   3191 		return EISDIR;
   3192 
   3193 	node_ref(opc);
   3194 
   3195 	/*
   3196 	 * We need to queue write requests in order to avoid
   3197 	 * dequeueing PCQ_AFTERWRITE when there are pending writes.
   3198 	 */
   3199 	while (pnd->pnd_flags & PND_INWRITE)
   3200 		requeue_request(pu, opc, PCQ_WRITE);
   3201 	pnd->pnd_flags |= PND_INWRITE;
   3202 
   3203 	/*
   3204 	 * append flag: re-read the file size so that
   3205 	 * we get the latest value.
   3206 	 */
   3207 	if (ioflag & PUFFS_IO_APPEND) {
   3208 		if ((error = perfuse_node_getattr(pu, opc, vap, pcr)) != 0)
   3209 			goto out;
   3210 
   3211 		offset = vap->va_size;
   3212 	}
   3213 
   3214 	/*
   3215 	 * Serialize size access, see comment in perfuse_node_setattr().
   3216 	 */
   3217 	if ((u_quad_t)offset + *resid > vap->va_size) {
   3218 		while (pnd->pnd_flags & PND_INRESIZE)
   3219 			requeue_request(pu, opc, PCQ_RESIZE);
   3220 		pnd->pnd_flags |= PND_INRESIZE;
   3221 		inresize = 1;
   3222 	}
   3223 
   3224 #ifdef PERFUSE_DEBUG
   3225 	if (perfuse_diagflags & PDF_RESIZE)
   3226 		DPRINTF(">> %s %p %" PRIu64 "\n", __func__,
   3227 			(void *)opc, vap->va_size);
   3228 #endif
   3229 
   3230 	fh = perfuse_get_fh(opc, FWRITE); /* Cannot be VDIR */
   3231 
   3232 	do {
   3233 		size_t max_write;
   3234 		/*
   3235 		 * There is a writepage flag when data
   3236 		 * is aligned to page size. Use it for
   3237 		 * everything but the data after the last
   3238 		 * page boundary.
   3239 		 */
   3240 		max_write = ps->ps_max_write - sizeof(*fwi);
   3241 
   3242 		data_len = MIN(*resid, max_write);
   3243 		if (data_len > (size_t)sysconf(_SC_PAGESIZE))
   3244 			data_len = data_len & ~(sysconf(_SC_PAGESIZE) - 1);
   3245 
   3246 		payload_len = data_len + sizeof(*fwi);
   3247 
   3248 		/*
   3249 		 * flags may be set to FUSE_WRITE_CACHE (XXX usage?)
   3250 		 * or FUSE_WRITE_LOCKOWNER, if lock_owner is provided.
   3251 		 * write_flags is set to 1 for writepage.
   3252 		 */
   3253 		pm = ps->ps_new_msg(pu, opc, FUSE_WRITE, payload_len, pcr);
   3254 		fwi = GET_INPAYLOAD(ps, pm, fuse_write_in);
   3255 		fwi->fh = fh;
   3256 		fwi->offset = offset;
   3257 		fwi->size = (uint32_t)data_len;
   3258 		fwi->write_flags = (fwi->size % sysconf(_SC_PAGESIZE)) ? 0 : 1;
   3259 		fwi->lock_owner = pnd->pnd_lock_owner;
   3260 		fwi->flags = 0;
   3261 		fwi->flags |= (fwi->lock_owner != 0) ? FUSE_WRITE_LOCKOWNER : 0;
   3262 		fwi->flags |= (ioflag & IO_DIRECT) ? 0 : FUSE_WRITE_CACHE;
   3263 		(void)memcpy((fwi + 1), buf, data_len);
   3264 
   3265 
   3266 #ifdef PERFUSE_DEBUG
   3267 		if (perfuse_diagflags & PDF_FH)
   3268 			DPRINTF("%s: opc = %p, nodeid = 0x%"PRIx64", "
   3269 				"fh = 0x%"PRIx64"\n", __func__,
   3270 				(void *)opc, pnd->pnd_nodeid, fwi->fh);
   3271 #endif
   3272 		if ((error = xchg_msg(pu, opc, pm,
   3273 				      sizeof(*fwo), wait_reply)) != 0)
   3274 			goto out;
   3275 
   3276 		fwo = GET_OUTPAYLOAD(ps, pm, fuse_write_out);
   3277 		written = fwo->size;
   3278 		ps->ps_destroy_msg(pm);
   3279 
   3280 #ifdef PERFUSE_DEBUG
   3281 		if (written > *resid)
   3282 			DERRX(EX_SOFTWARE, "%s: Unexpected big write %zd",
   3283 			      __func__, written);
   3284 #endif
   3285 		*resid -= written;
   3286 		offset += written;
   3287 		buf += written;
   3288 
   3289 	} while (*resid != 0);
   3290 
   3291 	/*
   3292 	 * puffs_ops(3) says
   3293 	 *  "everything must be written or an error will be generated"
   3294 	 */
   3295 	if (*resid != 0)
   3296 		error = EFBIG;
   3297 
   3298 out:
   3299 #ifdef PERFUSE_DEBUG
   3300 	if (perfuse_diagflags & PDF_RESIZE) {
   3301 		if (offset > (off_t)vap->va_size)
   3302 			DPRINTF("<< %s %p %" PRIu64 " -> %lld\n", __func__,
   3303 				(void *)opc, vap->va_size, (long long)offset);
   3304 		else
   3305 			DPRINTF("<< %s %p \n", __func__, (void *)opc);
   3306 	}
   3307 #endif
   3308 
   3309 	/*
   3310 	 * Update file size if we wrote beyond the end
   3311 	 */
   3312 	if (offset > (off_t)vap->va_size)
   3313 		vap->va_size = offset;
   3314 
   3315 	/*
   3316 	 * Statistics
   3317 	 */
   3318 	if (ioflag & (IO_SYNC|IO_DSYNC))
   3319 		ps->ps_syncwrites++;
   3320 	else
   3321 		ps->ps_asyncwrites++;
   3322 
   3323 	/*
   3324 	 * Remember to sync the file
   3325 	 */
   3326 	pnd->pnd_flags |= PND_DIRTY;
   3327 
   3328 #ifdef PERFUSE_DEBUG
   3329 	if (perfuse_diagflags & PDF_SYNC)
   3330 		DPRINTF("%s: DIRTY opc = %p, file = \"%s\"\n",
   3331 			__func__, (void*)opc, perfuse_node_path(ps, opc));
   3332 #endif
   3333 
   3334 	if (inresize) {
   3335 #ifdef PERFUSE_DEBUG
   3336 		if (!(pnd->pnd_flags & PND_INRESIZE))
   3337 			DERRX(EX_SOFTWARE, "file write grow without resize");
   3338 #endif
   3339 		pnd->pnd_flags &= ~PND_INRESIZE;
   3340 		(void)dequeue_requests(opc, PCQ_RESIZE, DEQUEUE_ALL);
   3341 	}
   3342 
   3343 	/*
   3344 	 * VOP_PUTPAGE causes FAF write where kernel does not
   3345 	 * check operation result. At least warn if it failed.
   3346 	 */
   3347 #ifdef PUFFS_WRITE_FAF
   3348 	if (error && (xflag & PUFFS_WRITE_FAF))
   3349 		DWARN("Data loss caused by FAF write failed on \"%s\"",
   3350 		      pnd->pnd_name);
   3351 #endif /* PUFFS_WRITE_FAF */
   3352 
   3353 	/*
   3354 	 * If there are no more queued write, we can resume
   3355 	 * an operation awaiting write completion.
   3356 	 */
   3357 	pnd->pnd_flags &= ~PND_INWRITE;
   3358 	if (dequeue_requests(opc, PCQ_WRITE, 1) == 0)
   3359 		(void)dequeue_requests(opc, PCQ_AFTERWRITE, DEQUEUE_ALL);
   3360 
   3361 	node_rele(opc);
   3362 	return error;
   3363 }
   3364 
   3365 /* ARGSUSED0 */
   3366 void
   3367 perfuse_cache_write(struct puffs_usermount *pu, puffs_cookie_t opc, size_t size,
   3368 	struct puffs_cacherun *runs)
   3369 {
   3370 	return;
   3371 }
   3372 
   3373 /* ARGSUSED4 */
   3374 int
   3375 perfuse_node_getextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   3376 	int attrns, const char *attrname, size_t *attrsize, uint8_t *attr,
   3377 	size_t *resid, const struct puffs_cred *pcr)
   3378 {
   3379 	struct perfuse_state *ps;
   3380 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
   3381 	perfuse_msg_t *pm;
   3382 	struct fuse_getxattr_in *fgi;
   3383 	struct fuse_getxattr_out *fgo;
   3384 	struct fuse_out_header *foh;
   3385 	size_t attrnamelen;
   3386 	size_t len;
   3387 	char *np;
   3388 	int error;
   3389 
   3390 	/* system namespace attrs are not accessible to non root users */
   3391 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
   3392 		return EPERM;
   3393 
   3394 	node_ref(opc);
   3395 	ps = puffs_getspecific(pu);
   3396 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
   3397 	attrnamelen = strlen(attrname) + 1;
   3398 	len = sizeof(*fgi) + attrnamelen;
   3399 
   3400 	pm = ps->ps_new_msg(pu, opc, FUSE_GETXATTR, len, pcr);
   3401 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
   3402 	fgi->size = (unsigned int)((resid != NULL) ? *resid : 0);
   3403 	np = (char *)(void *)(fgi + 1);
   3404 	(void)strlcpy(np, attrname, attrnamelen);
   3405 
   3406 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   3407 		goto out;
   3408 
   3409 	/*
   3410 	 * We just get fuse_getattr_out with list size if we requested
   3411 	 * a null size.
   3412 	 */
   3413 	if (resid == NULL) {
   3414 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
   3415 
   3416 		if (attrsize != NULL)
   3417 			*attrsize = fgo->size;
   3418 
   3419 		ps->ps_destroy_msg(pm);
   3420 		error = 0;
   3421 		goto out;
   3422 	}
   3423 
   3424 	/*
   3425 	 * And with a non null requested size, we get the list just
   3426 	 * after the header
   3427 	 */
   3428 	foh = GET_OUTHDR(ps, pm);
   3429 	np = (char *)(void *)(foh + 1);
   3430 	len = foh->len - sizeof(*foh);
   3431 
   3432 	if (attrsize != NULL)
   3433 		*attrsize = len;
   3434 
   3435 	if (resid != NULL) {
   3436 		if (*resid < len) {
   3437 			error = ERANGE;
   3438 			ps->ps_destroy_msg(pm);
   3439 			goto out;
   3440 		}
   3441 
   3442 		(void)memcpy(attr, np, len);
   3443 		*resid -= len;
   3444 	}
   3445 
   3446 	ps->ps_destroy_msg(pm);
   3447 	error = 0;
   3448 
   3449 out:
   3450 	node_rele(opc);
   3451 	return error;
   3452 }
   3453 
   3454 int
   3455 perfuse_node_setextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   3456 	int attrns, const char *attrname, uint8_t *attr, size_t *resid,
   3457 	const struct puffs_cred *pcr)
   3458 {
   3459 	struct perfuse_state *ps;
   3460 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
   3461 	perfuse_msg_t *pm;
   3462 	struct fuse_setxattr_in *fsi;
   3463 	size_t attrnamelen;
   3464 	size_t datalen;
   3465 	size_t len;
   3466 	char *np;
   3467 	int error;
   3468 
   3469 	/* system namespace attrs are not accessible to non root users */
   3470 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
   3471 		return EPERM;
   3472 
   3473 	node_ref(opc);
   3474 	ps = puffs_getspecific(pu);
   3475 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
   3476 	attrnamelen = strlen(attrname) + 1;
   3477 
   3478 	datalen = (resid != NULL) ? *resid : 0;
   3479 	len = sizeof(*fsi) + attrnamelen + datalen;
   3480 
   3481 	pm = ps->ps_new_msg(pu, opc, FUSE_SETXATTR, len, pcr);
   3482 	fsi = GET_INPAYLOAD(ps, pm, fuse_setxattr_in);
   3483 	fsi->size = (unsigned int)datalen;
   3484 	fsi->flags = 0;
   3485 	np = (char *)(void *)(fsi + 1);
   3486 	(void)strlcpy(np, attrname, attrnamelen);
   3487 	np += attrnamelen;
   3488 	if (datalen)
   3489 		(void)memcpy(np, (char *)attr, datalen);
   3490 
   3491 	if ((error = xchg_msg(pu, opc, pm,
   3492 			      NO_PAYLOAD_REPLY_LEN, wait_reply)) != 0)
   3493 		goto out;
   3494 
   3495 	ps->ps_destroy_msg(pm);
   3496 	if (resid)
   3497 		*resid = 0;
   3498 	error = 0;
   3499 
   3500 out:
   3501 	node_rele(opc);
   3502 	return error;
   3503 }
   3504 
   3505 /* ARGSUSED2 */
   3506 int
   3507 perfuse_node_listextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   3508 	int attrns, size_t *attrsize, uint8_t *attrs, size_t *resid, int flag,
   3509 	const struct puffs_cred *pcr)
   3510 {
   3511 	struct perfuse_state *ps;
   3512 	perfuse_msg_t *pm;
   3513 	struct fuse_getxattr_in *fgi;
   3514 	struct fuse_getxattr_out *fgo;
   3515 	struct fuse_out_header *foh;
   3516 	char *np;
   3517 	size_t len, puffs_len, i, attrlen, outlen;
   3518 	int error;
   3519 
   3520 	/* system namespace attrs are not accessible to non root users */
   3521 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
   3522 		return EPERM;
   3523 
   3524 	node_ref(opc);
   3525 
   3526 	ps = puffs_getspecific(pu);
   3527 	len = sizeof(*fgi);
   3528 
   3529 	pm = ps->ps_new_msg(pu, opc, FUSE_LISTXATTR, len, pcr);
   3530 	fgi = GET_INPAYLOAD(ps, pm, fuse_getxattr_in);
   3531 	if (resid != NULL)
   3532 		fgi->size = (unsigned int)*resid;
   3533 	else
   3534 		fgi->size = 0;
   3535 
   3536 	if ((error = xchg_msg(pu, opc, pm, UNSPEC_REPLY_LEN, wait_reply)) != 0)
   3537 		goto out;
   3538 
   3539 	/*
   3540 	 * We just get fuse_getattr_out with list size if we requested
   3541 	 * a null size.
   3542 	 */
   3543 	if (resid == NULL) {
   3544 		fgo = GET_OUTPAYLOAD(ps, pm, fuse_getxattr_out);
   3545 
   3546 		if (attrsize != NULL)
   3547 			*attrsize = fgo->size;
   3548 
   3549 		ps->ps_destroy_msg(pm);
   3550 
   3551 		error = 0;
   3552 		goto out;
   3553 	}
   3554 
   3555 	/*
   3556 	 * And with a non null requested size, we get the list just
   3557 	 * after the header
   3558 	 */
   3559 	foh = GET_OUTHDR(ps, pm);
   3560 	np = (char *)(void *)(foh + 1);
   3561 	puffs_len = foh->len - sizeof(*foh);
   3562 
   3563 	if (attrsize != NULL)
   3564 		*attrsize = puffs_len;
   3565 
   3566 	if (attrs != NULL) {
   3567 		if (*resid < puffs_len) {
   3568 			error = ERANGE;
   3569 			ps->ps_destroy_msg(pm);
   3570 			goto out;
   3571 		}
   3572 
   3573 		outlen = 0;
   3574 
   3575 		for (i = 0; i < puffs_len; i += attrlen + 1) {
   3576 			attrlen = strlen(np + i);
   3577 
   3578 			/*
   3579 			 * Filter attributes per namespace
   3580 			 */
   3581 			if (!perfuse_ns_match(attrns, np + i))
   3582 				continue;
   3583 
   3584 #ifdef PUFFS_EXTATTR_LIST_LENPREFIX
   3585 			/*
   3586 			 * Convert the FUSE reply to length prefixed strings
   3587 			 * if this is what the kernel wants.
   3588 			 */
   3589 			if (flag & PUFFS_EXTATTR_LIST_LENPREFIX) {
   3590 				(void)memcpy(attrs + outlen + 1,
   3591 					     np + i, attrlen);
   3592 				*(attrs + outlen) = (uint8_t)attrlen;
   3593 			} else
   3594 #endif /* PUFFS_EXTATTR_LIST_LENPREFIX */
   3595 			(void)memcpy(attrs + outlen, np + i, attrlen + 1);
   3596 			outlen += attrlen + 1;
   3597 		}
   3598 
   3599 		*resid -= outlen;
   3600 	}
   3601 
   3602 	ps->ps_destroy_msg(pm);
   3603 	error = 0;
   3604 
   3605 out:
   3606 	node_rele(opc);
   3607 	return error;
   3608 }
   3609 
   3610 int
   3611 perfuse_node_deleteextattr(struct puffs_usermount *pu, puffs_cookie_t opc,
   3612 	int attrns, const char *attrname, const struct puffs_cred *pcr)
   3613 {
   3614 	struct perfuse_state *ps;
   3615 	char fuse_attrname[LINUX_XATTR_NAME_MAX + 1];
   3616 	perfuse_msg_t *pm;
   3617 	size_t attrnamelen;
   3618 	char *np;
   3619 	int error;
   3620 
   3621 	/* system namespace attrs are not accessible to non root users */
   3622 	if (attrns == EXTATTR_NAMESPACE_SYSTEM && !puffs_cred_isjuggernaut(pcr))
   3623 		return EPERM;
   3624 
   3625 	node_ref(opc);
   3626 
   3627 	ps = puffs_getspecific(pu);
   3628 	attrname = perfuse_native_ns(attrns, attrname, fuse_attrname);
   3629 	attrnamelen = strlen(attrname) + 1;
   3630 
   3631 	pm = ps->ps_new_msg(pu, opc, FUSE_REMOVEXATTR, attrnamelen, pcr);
   3632 	np = _GET_INPAYLOAD(ps, pm, char *);
   3633 	(void)strlcpy(np, attrname, attrnamelen);
   3634 
   3635 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
   3636 	if (error != 0)
   3637 		goto out;
   3638 
   3639 	ps->ps_destroy_msg(pm);
   3640 
   3641 out:
   3642 	node_rele(opc);
   3643 	return error;
   3644 }
   3645 
   3646 int
   3647 perfuse_node_fallocate(struct puffs_usermount *pu, puffs_cookie_t opc,
   3648 	off_t off, off_t len)
   3649 {
   3650 	struct perfuse_state *ps;
   3651 	perfuse_msg_t *pm;
   3652 	struct fuse_fallocate_in *fai;
   3653 	int error;
   3654 
   3655 	ps = puffs_getspecific(pu);
   3656 	if (ps->ps_flags & PS_NO_FALLOCATE)
   3657 		return EOPNOTSUPP;
   3658 
   3659 	node_ref(opc);
   3660 
   3661 	pm = ps->ps_new_msg(pu, opc, FUSE_FALLOCATE, sizeof(*fai), NULL);
   3662 
   3663 	fai = GET_INPAYLOAD(ps, pm, fuse_fallocate_in);
   3664 	fai->fh = PN_ISDIR(opc) ? FUSE_UNKNOWN_FH : perfuse_get_fh(opc, FWRITE);
   3665 	fai->offset = off;
   3666 	fai->length = len;
   3667 	fai->mode = 0;
   3668 
   3669 	error = xchg_msg(pu, opc, pm, NO_PAYLOAD_REPLY_LEN, wait_reply);
   3670 	if (error == EOPNOTSUPP || error == ENOSYS) {
   3671 		ps->ps_flags |= PS_NO_FALLOCATE;
   3672 		error = EOPNOTSUPP;
   3673 	}
   3674 	if (error != 0)
   3675 		goto out;
   3676 
   3677 	ps->ps_destroy_msg(pm);
   3678 
   3679 out:
   3680 	node_rele(opc);
   3681 	return error;
   3682 }
   3683