Home | History | Annotate | Download | only in nfs
      1 /*
      2  * CDDL HEADER START
      3  *
      4  * The contents of this file are subject to the terms of the
      5  * Common Development and Distribution License (the "License").
      6  * You may not use this file except in compliance with the License.
      7  *
      8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
      9  * or http://www.opensolaris.org/os/licensing.
     10  * See the License for the specific language governing permissions
     11  * and limitations under the License.
     12  *
     13  * When distributing Covered Code, include this CDDL HEADER in each
     14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
     15  * If applicable, add the following below this CDDL HEADER, with the
     16  * fields enclosed by brackets "[]" replaced with your own identifying
     17  * information: Portions Copyright [yyyy] [name of copyright owner]
     18  *
     19  * CDDL HEADER END
     20  */
     21 /*
     22  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
     23  * Use is subject to license terms.
     24  */
     25 
     26 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
     27 /* All Rights Reserved */
     28 
     29 #include <sys/param.h>
     30 #include <sys/types.h>
     31 #include <sys/systm.h>
     32 #include <sys/cred.h>
     33 #include <sys/buf.h>
     34 #include <sys/vfs.h>
     35 #include <sys/vnode.h>
     36 #include <sys/uio.h>
     37 #include <sys/errno.h>
     38 #include <sys/sysmacros.h>
     39 #include <sys/statvfs.h>
     40 #include <sys/kmem.h>
     41 #include <sys/dirent.h>
     42 #include <sys/cmn_err.h>
     43 #include <sys/debug.h>
     44 #include <sys/systeminfo.h>
     45 #include <sys/flock.h>
     46 #include <sys/nbmlock.h>
     47 #include <sys/policy.h>
     48 #include <sys/sdt.h>
     49 
     50 #include <rpc/types.h>
     51 #include <rpc/auth.h>
     52 #include <rpc/svc.h>
     53 #include <rpc/rpc_rdma.h>
     54 
     55 #include <nfs/nfs.h>
     56 #include <nfs/export.h>
     57 #include <nfs/nfs_cmd.h>
     58 
     59 #include <sys/strsubr.h>
     60 
     61 #include <sys/tsol/label.h>
     62 #include <sys/tsol/tndb.h>
     63 
     64 #include <sys/zone.h>
     65 
     66 #include <inet/ip.h>
     67 #include <inet/ip6.h>
     68 
     69 /*
     70  * These are the interface routines for the server side of the
     71  * Network File System.  See the NFS version 3 protocol specification
     72  * for a description of this interface.
     73  */
     74 
     75 #ifdef DEBUG
     76 int rfs3_do_pre_op_attr = 1;
     77 int rfs3_do_post_op_attr = 1;
     78 int rfs3_do_post_op_fh3 = 1;
     79 #endif
     80 
     81 static writeverf3 write3verf;
     82 
     83 static int	sattr3_to_vattr(sattr3 *, struct vattr *);
     84 static int	vattr_to_fattr3(struct vattr *, fattr3 *);
     85 static int	vattr_to_wcc_attr(struct vattr *, wcc_attr *);
     86 static void	vattr_to_pre_op_attr(struct vattr *, pre_op_attr *);
     87 static void	vattr_to_wcc_data(struct vattr *, struct vattr *, wcc_data *);
     88 static int	rdma_setup_read_data3(READ3args *, READ3resok *);
     89 
     90 extern int nfs_loaned_buffers;
     91 
     92 u_longlong_t nfs3_srv_caller_id;
     93 
     94 /* ARGSUSED */
     95 void
     96 rfs3_getattr(GETATTR3args *args, GETATTR3res *resp, struct exportinfo *exi,
     97 	struct svc_req *req, cred_t *cr)
     98 {
     99 	int error;
    100 	vnode_t *vp;
    101 	struct vattr va;
    102 
    103 	vp = nfs3_fhtovp(&args->object, exi);
    104 
    105 	DTRACE_NFSV3_4(op__getattr__start, struct svc_req *, req,
    106 	    cred_t *, cr, vnode_t *, vp, GETATTR3args *, args);
    107 
    108 	if (vp == NULL) {
    109 		error = ESTALE;
    110 		goto out;
    111 	}
    112 
    113 	va.va_mask = AT_ALL;
    114 	error = rfs4_delegated_getattr(vp, &va, 0, cr);
    115 
    116 	if (!error) {
    117 		/* Lie about the object type for a referral */
    118 		if (vn_is_nfs_reparse(vp, cr))
    119 			va.va_type = VLNK;
    120 
    121 		/* overflow error if time or size is out of range */
    122 		error = vattr_to_fattr3(&va, &resp->resok.obj_attributes);
    123 		if (error)
    124 			goto out;
    125 		resp->status = NFS3_OK;
    126 
    127 		DTRACE_NFSV3_4(op__getattr__done, struct svc_req *, req,
    128 		    cred_t *, cr, vnode_t *, vp, GETATTR3res *, resp);
    129 
    130 		VN_RELE(vp);
    131 
    132 		return;
    133 	}
    134 
    135 out:
    136 	if (curthread->t_flag & T_WOULDBLOCK) {
    137 		curthread->t_flag &= ~T_WOULDBLOCK;
    138 		resp->status = NFS3ERR_JUKEBOX;
    139 	} else
    140 		resp->status = puterrno3(error);
    141 
    142 	DTRACE_NFSV3_4(op__getattr__done, struct svc_req *, req,
    143 	    cred_t *, cr, vnode_t *, vp, GETATTR3res *, resp);
    144 
    145 	if (vp != NULL)
    146 		VN_RELE(vp);
    147 }
    148 
    149 void *
    150 rfs3_getattr_getfh(GETATTR3args *args)
    151 {
    152 
    153 	return (&args->object);
    154 }
    155 
    156 void
    157 rfs3_setattr(SETATTR3args *args, SETATTR3res *resp, struct exportinfo *exi,
    158 	struct svc_req *req, cred_t *cr)
    159 {
    160 	int error;
    161 	vnode_t *vp;
    162 	struct vattr *bvap;
    163 	struct vattr bva;
    164 	struct vattr *avap;
    165 	struct vattr ava;
    166 	int flag;
    167 	int in_crit = 0;
    168 	struct flock64 bf;
    169 	caller_context_t ct;
    170 
    171 	bvap = NULL;
    172 	avap = NULL;
    173 
    174 	vp = nfs3_fhtovp(&args->object, exi);
    175 
    176 	DTRACE_NFSV3_4(op__setattr__start, struct svc_req *, req,
    177 	    cred_t *, cr, vnode_t *, vp, SETATTR3args *, args);
    178 
    179 	if (vp == NULL) {
    180 		error = ESTALE;
    181 		goto out;
    182 	}
    183 
    184 	error = sattr3_to_vattr(&args->new_attributes, &ava);
    185 	if (error)
    186 		goto out;
    187 
    188 	if (is_system_labeled()) {
    189 		bslabel_t *clabel = req->rq_label;
    190 
    191 		ASSERT(clabel != NULL);
    192 		DTRACE_PROBE2(tx__rfs3__log__info__opsetattr__clabel, char *,
    193 		    "got client label from request(1)", struct svc_req *, req);
    194 
    195 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
    196 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
    197 			    exi)) {
    198 				resp->status = NFS3ERR_ACCES;
    199 				goto out1;
    200 			}
    201 		}
    202 	}
    203 
    204 	/*
    205 	 * We need to specially handle size changes because of
    206 	 * possible conflicting NBMAND locks. Get into critical
    207 	 * region before VOP_GETATTR, so the size attribute is
    208 	 * valid when checking conflicts.
    209 	 *
    210 	 * Also, check to see if the v4 side of the server has
    211 	 * delegated this file.  If so, then we return JUKEBOX to
    212 	 * allow the client to retrasmit its request.
    213 	 */
    214 	if (vp->v_type == VREG && (ava.va_mask & AT_SIZE)) {
    215 		if (nbl_need_check(vp)) {
    216 			nbl_start_crit(vp, RW_READER);
    217 			in_crit = 1;
    218 		}
    219 	}
    220 
    221 	bva.va_mask = AT_ALL;
    222 	error = rfs4_delegated_getattr(vp, &bva, 0, cr);
    223 
    224 	/*
    225 	 * If we can't get the attributes, then we can't do the
    226 	 * right access checking.  So, we'll fail the request.
    227 	 */
    228 	if (error)
    229 		goto out;
    230 
    231 #ifdef DEBUG
    232 	if (rfs3_do_pre_op_attr)
    233 		bvap = &bva;
    234 #else
    235 	bvap = &bva;
    236 #endif
    237 
    238 	if (rdonly(exi, req) || vn_is_readonly(vp)) {
    239 		resp->status = NFS3ERR_ROFS;
    240 		goto out1;
    241 	}
    242 
    243 	if (args->guard.check &&
    244 	    (args->guard.obj_ctime.seconds != bva.va_ctime.tv_sec ||
    245 	    args->guard.obj_ctime.nseconds != bva.va_ctime.tv_nsec)) {
    246 		resp->status = NFS3ERR_NOT_SYNC;
    247 		goto out1;
    248 	}
    249 
    250 	if (args->new_attributes.mtime.set_it == SET_TO_CLIENT_TIME)
    251 		flag = ATTR_UTIME;
    252 	else
    253 		flag = 0;
    254 
    255 	/*
    256 	 * If the filesystem is exported with nosuid, then mask off
    257 	 * the setuid and setgid bits.
    258 	 */
    259 	if ((ava.va_mask & AT_MODE) && vp->v_type == VREG &&
    260 	    (exi->exi_export.ex_flags & EX_NOSUID))
    261 		ava.va_mode &= ~(VSUID | VSGID);
    262 
    263 	ct.cc_sysid = 0;
    264 	ct.cc_pid = 0;
    265 	ct.cc_caller_id = nfs3_srv_caller_id;
    266 	ct.cc_flags = CC_DONTBLOCK;
    267 
    268 	/*
    269 	 * We need to specially handle size changes because it is
    270 	 * possible for the client to create a file with modes
    271 	 * which indicate read-only, but with the file opened for
    272 	 * writing.  If the client then tries to set the size of
    273 	 * the file, then the normal access checking done in
    274 	 * VOP_SETATTR would prevent the client from doing so,
    275 	 * although it should be legal for it to do so.  To get
    276 	 * around this, we do the access checking for ourselves
    277 	 * and then use VOP_SPACE which doesn't do the access
    278 	 * checking which VOP_SETATTR does. VOP_SPACE can only
    279 	 * operate on VREG files, let VOP_SETATTR handle the other
    280 	 * extremely rare cases.
    281 	 * Also the client should not be allowed to change the
    282 	 * size of the file if there is a conflicting non-blocking
    283 	 * mandatory lock in the region the change.
    284 	 */
    285 	if (vp->v_type == VREG && (ava.va_mask & AT_SIZE)) {
    286 		if (in_crit) {
    287 			u_offset_t offset;
    288 			ssize_t length;
    289 
    290 			if (ava.va_size < bva.va_size) {
    291 				offset = ava.va_size;
    292 				length = bva.va_size - ava.va_size;
    293 			} else {
    294 				offset = bva.va_size;
    295 				length = ava.va_size - bva.va_size;
    296 			}
    297 			if (nbl_conflict(vp, NBL_WRITE, offset, length, 0,
    298 			    NULL)) {
    299 				error = EACCES;
    300 				goto out;
    301 			}
    302 		}
    303 
    304 		if (crgetuid(cr) == bva.va_uid && ava.va_size != bva.va_size) {
    305 			ava.va_mask &= ~AT_SIZE;
    306 			bf.l_type = F_WRLCK;
    307 			bf.l_whence = 0;
    308 			bf.l_start = (off64_t)ava.va_size;
    309 			bf.l_len = 0;
    310 			bf.l_sysid = 0;
    311 			bf.l_pid = 0;
    312 			error = VOP_SPACE(vp, F_FREESP, &bf, FWRITE,
    313 			    (offset_t)ava.va_size, cr, &ct);
    314 		}
    315 	}
    316 
    317 	if (!error && ava.va_mask)
    318 		error = VOP_SETATTR(vp, &ava, flag, cr, &ct);
    319 
    320 	/* check if a monitor detected a delegation conflict */
    321 	if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK)) {
    322 		resp->status = NFS3ERR_JUKEBOX;
    323 		goto out1;
    324 	}
    325 
    326 #ifdef DEBUG
    327 	if (rfs3_do_post_op_attr) {
    328 		ava.va_mask = AT_ALL;
    329 		avap = rfs4_delegated_getattr(vp, &ava, 0, cr) ? NULL : &ava;
    330 	} else
    331 		avap = NULL;
    332 #else
    333 	ava.va_mask = AT_ALL;
    334 	avap = rfs4_delegated_getattr(vp, &ava, 0, cr) ? NULL : &ava;
    335 #endif
    336 
    337 	/*
    338 	 * Force modified metadata out to stable storage.
    339 	 */
    340 	(void) VOP_FSYNC(vp, FNODSYNC, cr, &ct);
    341 
    342 	if (error)
    343 		goto out;
    344 
    345 	if (in_crit)
    346 		nbl_end_crit(vp);
    347 
    348 	resp->status = NFS3_OK;
    349 	vattr_to_wcc_data(bvap, avap, &resp->resok.obj_wcc);
    350 
    351 	DTRACE_NFSV3_4(op__setattr__done, struct svc_req *, req,
    352 	    cred_t *, cr, vnode_t *, vp, SETATTR3res *, resp);
    353 
    354 	VN_RELE(vp);
    355 
    356 	return;
    357 
    358 out:
    359 	if (curthread->t_flag & T_WOULDBLOCK) {
    360 		curthread->t_flag &= ~T_WOULDBLOCK;
    361 		resp->status = NFS3ERR_JUKEBOX;
    362 	} else
    363 		resp->status = puterrno3(error);
    364 out1:
    365 	DTRACE_NFSV3_4(op__setattr__done, struct svc_req *, req,
    366 	    cred_t *, cr, vnode_t *, vp, SETATTR3res *, resp);
    367 
    368 	if (vp != NULL) {
    369 		if (in_crit)
    370 			nbl_end_crit(vp);
    371 		VN_RELE(vp);
    372 	}
    373 	vattr_to_wcc_data(bvap, avap, &resp->resfail.obj_wcc);
    374 }
    375 
    376 void *
    377 rfs3_setattr_getfh(SETATTR3args *args)
    378 {
    379 
    380 	return (&args->object);
    381 }
    382 
    383 /* ARGSUSED */
    384 void
    385 rfs3_lookup(LOOKUP3args *args, LOOKUP3res *resp, struct exportinfo *exi,
    386 	struct svc_req *req, cred_t *cr)
    387 {
    388 	int error;
    389 	vnode_t *vp;
    390 	vnode_t *dvp;
    391 	struct vattr *vap;
    392 	struct vattr va;
    393 	struct vattr *dvap;
    394 	struct vattr dva;
    395 	nfs_fh3 *fhp;
    396 	struct sec_ol sec = {0, 0};
    397 	bool_t publicfh_flag = FALSE, auth_weak = FALSE;
    398 	struct sockaddr *ca;
    399 	char *name = NULL;
    400 
    401 	dvap = NULL;
    402 
    403 	/*
    404 	 * Allow lookups from the root - the default
    405 	 * location of the public filehandle.
    406 	 */
    407 	if (exi != NULL && (exi->exi_export.ex_flags & EX_PUBLIC)) {
    408 		dvp = rootdir;
    409 		VN_HOLD(dvp);
    410 
    411 		DTRACE_NFSV3_4(op__lookup__start, struct svc_req *, req,
    412 		    cred_t *, cr, vnode_t *, dvp, LOOKUP3args *, args);
    413 	} else {
    414 		dvp = nfs3_fhtovp(&args->what.dir, exi);
    415 
    416 		DTRACE_NFSV3_4(op__lookup__start, struct svc_req *, req,
    417 		    cred_t *, cr, vnode_t *, dvp, LOOKUP3args *, args);
    418 
    419 		if (dvp == NULL) {
    420 			error = ESTALE;
    421 			goto out;
    422 		}
    423 	}
    424 
    425 #ifdef DEBUG
    426 	if (rfs3_do_pre_op_attr) {
    427 		dva.va_mask = AT_ALL;
    428 		dvap = VOP_GETATTR(dvp, &dva, 0, cr, NULL) ? NULL : &dva;
    429 	}
    430 #else
    431 	dva.va_mask = AT_ALL;
    432 	dvap = VOP_GETATTR(dvp, &dva, 0, cr, NULL) ? NULL : &dva;
    433 #endif
    434 
    435 	if (args->what.name == nfs3nametoolong) {
    436 		resp->status = NFS3ERR_NAMETOOLONG;
    437 		goto out1;
    438 	}
    439 
    440 	if (args->what.name == NULL || *(args->what.name) == '\0') {
    441 		resp->status = NFS3ERR_ACCES;
    442 		goto out1;
    443 	}
    444 
    445 	fhp = &args->what.dir;
    446 	if (strcmp(args->what.name, "..") == 0 &&
    447 	    EQFID(&exi->exi_fid, FH3TOFIDP(fhp))) {
    448 		resp->status = NFS3ERR_NOENT;
    449 		goto out1;
    450 	}
    451 
    452 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
    453 	name = nfscmd_convname(ca, exi, args->what.name,
    454 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
    455 
    456 	if (name == NULL) {
    457 		resp->status = NFS3ERR_ACCES;
    458 		goto out1;
    459 	}
    460 
    461 	/*
    462 	 * If the public filehandle is used then allow
    463 	 * a multi-component lookup
    464 	 */
    465 	if (PUBLIC_FH3(&args->what.dir)) {
    466 		publicfh_flag = TRUE;
    467 		error = rfs_publicfh_mclookup(name, dvp, cr, &vp,
    468 		    &exi, &sec);
    469 		if (error && exi != NULL)
    470 			exi_rele(exi); /* See comment below Re: publicfh_flag */
    471 		/*
    472 		 * Since WebNFS may bypass MOUNT, we need to ensure this
    473 		 * request didn't come from an unlabeled admin_low client.
    474 		 */
    475 		if (is_system_labeled() && error == 0) {
    476 			int		addr_type;
    477 			void		*ipaddr;
    478 			tsol_tpc_t	*tp;
    479 
    480 			if (ca->sa_family == AF_INET) {
    481 				addr_type = IPV4_VERSION;
    482 				ipaddr = &((struct sockaddr_in *)ca)->sin_addr;
    483 			} else if (ca->sa_family == AF_INET6) {
    484 				addr_type = IPV6_VERSION;
    485 				ipaddr = &((struct sockaddr_in6 *)
    486 				    ca)->sin6_addr;
    487 			}
    488 			tp = find_tpc(ipaddr, addr_type, B_FALSE);
    489 			if (tp == NULL || tp->tpc_tp.tp_doi !=
    490 			    l_admin_low->tsl_doi || tp->tpc_tp.host_type !=
    491 			    SUN_CIPSO) {
    492 				if (exi != NULL)
    493 					exi_rele(exi);
    494 				VN_RELE(vp);
    495 				resp->status = NFS3ERR_ACCES;
    496 				error = 1;
    497 			}
    498 			if (tp != NULL)
    499 				TPC_RELE(tp);
    500 		}
    501 	} else {
    502 		error = VOP_LOOKUP(dvp, name, &vp,
    503 		    NULL, 0, NULL, cr, NULL, NULL, NULL);
    504 	}
    505 
    506 	if (name != args->what.name)
    507 		kmem_free(name, MAXPATHLEN + 1);
    508 
    509 	if (is_system_labeled() && error == 0) {
    510 		bslabel_t *clabel = req->rq_label;
    511 
    512 		ASSERT(clabel != NULL);
    513 		DTRACE_PROBE2(tx__rfs3__log__info__oplookup__clabel, char *,
    514 		    "got client label from request(1)", struct svc_req *, req);
    515 
    516 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
    517 			if (!do_rfs_label_check(clabel, dvp,
    518 			    DOMINANCE_CHECK, exi)) {
    519 				if (publicfh_flag && exi != NULL)
    520 					exi_rele(exi);
    521 				VN_RELE(vp);
    522 				resp->status = NFS3ERR_ACCES;
    523 				error = 1;
    524 			}
    525 		}
    526 	}
    527 
    528 #ifdef DEBUG
    529 	if (rfs3_do_post_op_attr) {
    530 		dva.va_mask = AT_ALL;
    531 		dvap = VOP_GETATTR(dvp, &dva, 0, cr, NULL) ? NULL : &dva;
    532 	} else
    533 		dvap = NULL;
    534 #else
    535 	dva.va_mask = AT_ALL;
    536 	dvap = VOP_GETATTR(dvp, &dva, 0, cr, NULL) ? NULL : &dva;
    537 #endif
    538 
    539 	if (error)
    540 		goto out;
    541 
    542 	if (sec.sec_flags & SEC_QUERY) {
    543 		error = makefh3_ol(&resp->resok.object, exi, sec.sec_index);
    544 	} else {
    545 		error = makefh3(&resp->resok.object, vp, exi);
    546 		if (!error && publicfh_flag && !chk_clnt_sec(exi, req))
    547 			auth_weak = TRUE;
    548 	}
    549 
    550 	if (error) {
    551 		VN_RELE(vp);
    552 		goto out;
    553 	}
    554 
    555 	/*
    556 	 * If publicfh_flag is true then we have called rfs_publicfh_mclookup
    557 	 * and have obtained a new exportinfo in exi which needs to be
    558 	 * released. Note the the original exportinfo pointed to by exi
    559 	 * will be released by the caller, common_dispatch.
    560 	 */
    561 	if (publicfh_flag)
    562 		exi_rele(exi);
    563 
    564 #ifdef DEBUG
    565 	if (rfs3_do_post_op_attr) {
    566 		va.va_mask = AT_ALL;
    567 		vap = rfs4_delegated_getattr(vp, &va, 0, cr) ? NULL : &va;
    568 	} else
    569 		vap = NULL;
    570 #else
    571 	va.va_mask = AT_ALL;
    572 	vap = rfs4_delegated_getattr(vp, &va, 0, cr) ? NULL : &va;
    573 #endif
    574 
    575 	VN_RELE(vp);
    576 
    577 	resp->status = NFS3_OK;
    578 	vattr_to_post_op_attr(vap, &resp->resok.obj_attributes);
    579 	vattr_to_post_op_attr(dvap, &resp->resok.dir_attributes);
    580 
    581 	/*
    582 	 * If it's public fh, no 0x81, and client's flavor is
    583 	 * invalid, set WebNFS status to WNFSERR_CLNT_FLAVOR now.
    584 	 * Then set RPC status to AUTH_TOOWEAK in common_dispatch.
    585 	 */
    586 	if (auth_weak)
    587 		resp->status = (enum nfsstat3)WNFSERR_CLNT_FLAVOR;
    588 
    589 	DTRACE_NFSV3_4(op__lookup__done, struct svc_req *, req,
    590 	    cred_t *, cr, vnode_t *, dvp, LOOKUP3res *, resp);
    591 	VN_RELE(dvp);
    592 
    593 	return;
    594 
    595 out:
    596 	if (curthread->t_flag & T_WOULDBLOCK) {
    597 		curthread->t_flag &= ~T_WOULDBLOCK;
    598 		resp->status = NFS3ERR_JUKEBOX;
    599 	} else
    600 		resp->status = puterrno3(error);
    601 out1:
    602 	DTRACE_NFSV3_4(op__lookup__done, struct svc_req *, req,
    603 	    cred_t *, cr, vnode_t *, dvp, LOOKUP3res *, resp);
    604 
    605 	if (dvp != NULL)
    606 		VN_RELE(dvp);
    607 	vattr_to_post_op_attr(dvap, &resp->resfail.dir_attributes);
    608 
    609 }
    610 
    611 void *
    612 rfs3_lookup_getfh(LOOKUP3args *args)
    613 {
    614 
    615 	return (&args->what.dir);
    616 }
    617 
    618 /* ARGSUSED */
    619 void
    620 rfs3_access(ACCESS3args *args, ACCESS3res *resp, struct exportinfo *exi,
    621 	struct svc_req *req, cred_t *cr)
    622 {
    623 	int error;
    624 	vnode_t *vp;
    625 	struct vattr *vap;
    626 	struct vattr va;
    627 	int checkwriteperm;
    628 	boolean_t dominant_label = B_FALSE;
    629 	boolean_t equal_label = B_FALSE;
    630 	boolean_t admin_low_client;
    631 
    632 	vap = NULL;
    633 
    634 	vp = nfs3_fhtovp(&args->object, exi);
    635 
    636 	DTRACE_NFSV3_4(op__access__start, struct svc_req *, req,
    637 	    cred_t *, cr, vnode_t *, vp, ACCESS3args *, args);
    638 
    639 	if (vp == NULL) {
    640 		error = ESTALE;
    641 		goto out;
    642 	}
    643 
    644 	/*
    645 	 * If the file system is exported read only, it is not appropriate
    646 	 * to check write permissions for regular files and directories.
    647 	 * Special files are interpreted by the client, so the underlying
    648 	 * permissions are sent back to the client for interpretation.
    649 	 */
    650 	if (rdonly(exi, req) && (vp->v_type == VREG || vp->v_type == VDIR))
    651 		checkwriteperm = 0;
    652 	else
    653 		checkwriteperm = 1;
    654 
    655 	/*
    656 	 * We need the mode so that we can correctly determine access
    657 	 * permissions relative to a mandatory lock file.  Access to
    658 	 * mandatory lock files is denied on the server, so it might
    659 	 * as well be reflected to the server during the open.
    660 	 */
    661 	va.va_mask = AT_MODE;
    662 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
    663 	if (error)
    664 		goto out;
    665 
    666 #ifdef DEBUG
    667 	if (rfs3_do_post_op_attr)
    668 		vap = &va;
    669 #else
    670 	vap = &va;
    671 #endif
    672 
    673 	resp->resok.access = 0;
    674 
    675 	if (is_system_labeled()) {
    676 		bslabel_t *clabel = req->rq_label;
    677 
    678 		ASSERT(clabel != NULL);
    679 		DTRACE_PROBE2(tx__rfs3__log__info__opaccess__clabel, char *,
    680 		    "got client label from request(1)", struct svc_req *, req);
    681 
    682 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
    683 			if ((equal_label = do_rfs_label_check(clabel, vp,
    684 			    EQUALITY_CHECK, exi)) == B_FALSE) {
    685 				dominant_label = do_rfs_label_check(clabel,
    686 				    vp, DOMINANCE_CHECK, exi);
    687 			} else
    688 				dominant_label = B_TRUE;
    689 			admin_low_client = B_FALSE;
    690 		} else
    691 			admin_low_client = B_TRUE;
    692 	}
    693 
    694 	if (args->access & ACCESS3_READ) {
    695 		error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
    696 		if (error) {
    697 			if (curthread->t_flag & T_WOULDBLOCK)
    698 				goto out;
    699 		} else if (!MANDLOCK(vp, va.va_mode) &&
    700 		    (!is_system_labeled() || admin_low_client ||
    701 		    dominant_label))
    702 			resp->resok.access |= ACCESS3_READ;
    703 	}
    704 	if ((args->access & ACCESS3_LOOKUP) && vp->v_type == VDIR) {
    705 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
    706 		if (error) {
    707 			if (curthread->t_flag & T_WOULDBLOCK)
    708 				goto out;
    709 		} else if (!is_system_labeled() || admin_low_client ||
    710 		    dominant_label)
    711 			resp->resok.access |= ACCESS3_LOOKUP;
    712 	}
    713 	if (checkwriteperm &&
    714 	    (args->access & (ACCESS3_MODIFY|ACCESS3_EXTEND))) {
    715 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
    716 		if (error) {
    717 			if (curthread->t_flag & T_WOULDBLOCK)
    718 				goto out;
    719 		} else if (!MANDLOCK(vp, va.va_mode) &&
    720 		    (!is_system_labeled() || admin_low_client || equal_label)) {
    721 			resp->resok.access |=
    722 			    (args->access & (ACCESS3_MODIFY|ACCESS3_EXTEND));
    723 		}
    724 	}
    725 	if (checkwriteperm &&
    726 	    (args->access & ACCESS3_DELETE) && vp->v_type == VDIR) {
    727 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
    728 		if (error) {
    729 			if (curthread->t_flag & T_WOULDBLOCK)
    730 				goto out;
    731 		} else if (!is_system_labeled() || admin_low_client ||
    732 		    equal_label)
    733 			resp->resok.access |= ACCESS3_DELETE;
    734 	}
    735 	if (args->access & ACCESS3_EXECUTE) {
    736 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
    737 		if (error) {
    738 			if (curthread->t_flag & T_WOULDBLOCK)
    739 				goto out;
    740 		} else if (!MANDLOCK(vp, va.va_mode) &&
    741 		    (!is_system_labeled() || admin_low_client ||
    742 		    dominant_label))
    743 			resp->resok.access |= ACCESS3_EXECUTE;
    744 	}
    745 
    746 #ifdef DEBUG
    747 	if (rfs3_do_post_op_attr) {
    748 		va.va_mask = AT_ALL;
    749 		vap = rfs4_delegated_getattr(vp, &va, 0, cr) ? NULL : &va;
    750 	} else
    751 		vap = NULL;
    752 #else
    753 	va.va_mask = AT_ALL;
    754 	vap = rfs4_delegated_getattr(vp, &va, 0, cr) ? NULL : &va;
    755 #endif
    756 
    757 	resp->status = NFS3_OK;
    758 	vattr_to_post_op_attr(vap, &resp->resok.obj_attributes);
    759 
    760 	DTRACE_NFSV3_4(op__access__done, struct svc_req *, req,
    761 	    cred_t *, cr, vnode_t *, vp, ACCESS3res *, resp);
    762 
    763 	VN_RELE(vp);
    764 
    765 	return;
    766 
    767 out:
    768 	if (curthread->t_flag & T_WOULDBLOCK) {
    769 		curthread->t_flag &= ~T_WOULDBLOCK;
    770 		resp->status = NFS3ERR_JUKEBOX;
    771 	} else
    772 		resp->status = puterrno3(error);
    773 	DTRACE_NFSV3_4(op__access__done, struct svc_req *, req,
    774 	    cred_t *, cr, vnode_t *, vp, ACCESS3res *, resp);
    775 	if (vp != NULL)
    776 		VN_RELE(vp);
    777 	vattr_to_post_op_attr(vap, &resp->resfail.obj_attributes);
    778 }
    779 
    780 void *
    781 rfs3_access_getfh(ACCESS3args *args)
    782 {
    783 
    784 	return (&args->object);
    785 }
    786 
    787 /* ARGSUSED */
    788 void
    789 rfs3_readlink(READLINK3args *args, READLINK3res *resp, struct exportinfo *exi,
    790 	struct svc_req *req, cred_t *cr)
    791 {
    792 	int error;
    793 	vnode_t *vp;
    794 	struct vattr *vap;
    795 	struct vattr va;
    796 	struct iovec iov;
    797 	struct uio uio;
    798 	char *data;
    799 	struct sockaddr *ca;
    800 	char *name = NULL;
    801 	int is_referral = 0;
    802 
    803 	vap = NULL;
    804 
    805 	vp = nfs3_fhtovp(&args->symlink, exi);
    806 
    807 	DTRACE_NFSV3_4(op__readlink__start, struct svc_req *, req,
    808 	    cred_t *, cr, vnode_t *, vp, READLINK3args *, args);
    809 
    810 	if (vp == NULL) {
    811 		error = ESTALE;
    812 		goto out;
    813 	}
    814 
    815 	va.va_mask = AT_ALL;
    816 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
    817 	if (error)
    818 		goto out;
    819 
    820 #ifdef DEBUG
    821 	if (rfs3_do_post_op_attr)
    822 		vap = &va;
    823 #else
    824 	vap = &va;
    825 #endif
    826 
    827 	/* We lied about the object type for a referral */
    828 	if (vn_is_nfs_reparse(vp, cr))
    829 		is_referral = 1;
    830 
    831 	if (vp->v_type != VLNK && !is_referral) {
    832 		resp->status = NFS3ERR_INVAL;
    833 		goto out1;
    834 	}
    835 
    836 	if (MANDLOCK(vp, va.va_mode)) {
    837 		resp->status = NFS3ERR_ACCES;
    838 		goto out1;
    839 	}
    840 
    841 	if (is_system_labeled()) {
    842 		bslabel_t *clabel = req->rq_label;
    843 
    844 		ASSERT(clabel != NULL);
    845 		DTRACE_PROBE2(tx__rfs3__log__info__opreadlink__clabel, char *,
    846 		    "got client label from request(1)", struct svc_req *, req);
    847 
    848 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
    849 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
    850 			    exi)) {
    851 				resp->status = NFS3ERR_ACCES;
    852 				goto out1;
    853 			}
    854 		}
    855 	}
    856 
    857 	data = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);
    858 
    859 	if (is_referral) {
    860 		char *s;
    861 		size_t strsz;
    862 
    863 		/* Get an artificial symlink based on a referral */
    864 		s = build_symlink(vp, cr, &strsz);
    865 		global_svstat_ptr[3][NFS_REFERLINKS].value.ui64++;
    866 		DTRACE_PROBE2(nfs3serv__func__referral__reflink,
    867 		    vnode_t *, vp, char *, s);
    868 		if (s == NULL)
    869 			error = EINVAL;
    870 		else {
    871 			error = 0;
    872 			(void) strlcpy(data, s, MAXPATHLEN + 1);
    873 			kmem_free(s, strsz);
    874 		}
    875 
    876 	} else {
    877 
    878 		iov.iov_base = data;
    879 		iov.iov_len = MAXPATHLEN;
    880 		uio.uio_iov = &iov;
    881 		uio.uio_iovcnt = 1;
    882 		uio.uio_segflg = UIO_SYSSPACE;
    883 		uio.uio_extflg = UIO_COPY_CACHED;
    884 		uio.uio_loffset = 0;
    885 		uio.uio_resid = MAXPATHLEN;
    886 
    887 		error = VOP_READLINK(vp, &uio, cr, NULL);
    888 
    889 		if (!error)
    890 			*(data + MAXPATHLEN - uio.uio_resid) = '\0';
    891 	}
    892 
    893 #ifdef DEBUG
    894 	if (rfs3_do_post_op_attr) {
    895 		va.va_mask = AT_ALL;
    896 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
    897 	} else
    898 		vap = NULL;
    899 #else
    900 	va.va_mask = AT_ALL;
    901 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
    902 #endif
    903 	/* Lie about object type again just to be consistent */
    904 	if (is_referral && vap != NULL)
    905 		vap->va_type = VLNK;
    906 
    907 #if 0 /* notyet */
    908 	/*
    909 	 * Don't do this.  It causes local disk writes when just
    910 	 * reading the file and the overhead is deemed larger
    911 	 * than the benefit.
    912 	 */
    913 	/*
    914 	 * Force modified metadata out to stable storage.
    915 	 */
    916 	(void) VOP_FSYNC(vp, FNODSYNC, cr, NULL);
    917 #endif
    918 
    919 	if (error) {
    920 		kmem_free(data, MAXPATHLEN + 1);
    921 		goto out;
    922 	}
    923 
    924 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
    925 	name = nfscmd_convname(ca, exi, data, NFSCMD_CONV_OUTBOUND,
    926 	    MAXPATHLEN + 1);
    927 
    928 	if (name == NULL) {
    929 		/*
    930 		 * Even though the conversion failed, we return
    931 		 * something. We just don't translate it.
    932 		 */
    933 		name = data;
    934 	}
    935 
    936 	resp->status = NFS3_OK;
    937 	vattr_to_post_op_attr(vap, &resp->resok.symlink_attributes);
    938 	resp->resok.data = name;
    939 
    940 	DTRACE_NFSV3_4(op__readlink__done, struct svc_req *, req,
    941 	    cred_t *, cr, vnode_t *, vp, READLINK3res *, resp);
    942 	VN_RELE(vp);
    943 
    944 	if (name != data)
    945 		kmem_free(data, MAXPATHLEN + 1);
    946 
    947 	return;
    948 
    949 out:
    950 	if (curthread->t_flag & T_WOULDBLOCK) {
    951 		curthread->t_flag &= ~T_WOULDBLOCK;
    952 		resp->status = NFS3ERR_JUKEBOX;
    953 	} else
    954 		resp->status = puterrno3(error);
    955 out1:
    956 	DTRACE_NFSV3_4(op__readlink__done, struct svc_req *, req,
    957 	    cred_t *, cr, vnode_t *, vp, READLINK3res *, resp);
    958 	if (vp != NULL)
    959 		VN_RELE(vp);
    960 	vattr_to_post_op_attr(vap, &resp->resfail.symlink_attributes);
    961 }
    962 
    963 void *
    964 rfs3_readlink_getfh(READLINK3args *args)
    965 {
    966 
    967 	return (&args->symlink);
    968 }
    969 
    970 void
    971 rfs3_readlink_free(READLINK3res *resp)
    972 {
    973 
    974 	if (resp->status == NFS3_OK)
    975 		kmem_free(resp->resok.data, MAXPATHLEN + 1);
    976 }
    977 
    978 /*
    979  * Server routine to handle read
    980  * May handle RDMA data as well as mblks
    981  */
    982 /* ARGSUSED */
    983 void
    984 rfs3_read(READ3args *args, READ3res *resp, struct exportinfo *exi,
    985 	struct svc_req *req, cred_t *cr)
    986 {
    987 	int error;
    988 	vnode_t *vp;
    989 	struct vattr *vap;
    990 	struct vattr va;
    991 	struct iovec iov;
    992 	struct uio uio;
    993 	u_offset_t offset;
    994 	mblk_t *mp = NULL;
    995 	int alloc_err = 0;
    996 	int in_crit = 0;
    997 	int need_rwunlock = 0;
    998 	caller_context_t ct;
    999 	int rdma_used = 0;
   1000 	int loaned_buffers;
   1001 	struct uio *uiop;
   1002 
   1003 	vap = NULL;
   1004 
   1005 	vp = nfs3_fhtovp(&args->file, exi);
   1006 
   1007 	DTRACE_NFSV3_4(op__read__start, struct svc_req *, req,
   1008 	    cred_t *, cr, vnode_t *, vp, READ3args *, args);
   1009 
   1010 	if (vp == NULL) {
   1011 		error = ESTALE;
   1012 		goto out;
   1013 	}
   1014 
   1015 	if (args->wlist)
   1016 		rdma_used = 1;
   1017 
   1018 	/* use loaned buffers for TCP */
   1019 	loaned_buffers = (nfs_loaned_buffers && !rdma_used) ? 1 : 0;
   1020 
   1021 	if (is_system_labeled()) {
   1022 		bslabel_t *clabel = req->rq_label;
   1023 
   1024 		ASSERT(clabel != NULL);
   1025 		DTRACE_PROBE2(tx__rfs3__log__info__opread__clabel, char *,
   1026 		    "got client label from request(1)", struct svc_req *, req);
   1027 
   1028 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   1029 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
   1030 			    exi)) {
   1031 				resp->status = NFS3ERR_ACCES;
   1032 				goto out1;
   1033 			}
   1034 		}
   1035 	}
   1036 
   1037 	ct.cc_sysid = 0;
   1038 	ct.cc_pid = 0;
   1039 	ct.cc_caller_id = nfs3_srv_caller_id;
   1040 	ct.cc_flags = CC_DONTBLOCK;
   1041 
   1042 	/*
   1043 	 * Enter the critical region before calling VOP_RWLOCK
   1044 	 * to avoid a deadlock with write requests.
   1045 	 */
   1046 	if (nbl_need_check(vp)) {
   1047 		nbl_start_crit(vp, RW_READER);
   1048 		in_crit = 1;
   1049 		if (nbl_conflict(vp, NBL_READ, args->offset, args->count, 0,
   1050 		    NULL)) {
   1051 			error = EACCES;
   1052 			goto out;
   1053 		}
   1054 	}
   1055 
   1056 	error = VOP_RWLOCK(vp, V_WRITELOCK_FALSE, &ct);
   1057 
   1058 	/* check if a monitor detected a delegation conflict */
   1059 	if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK)) {
   1060 		resp->status = NFS3ERR_JUKEBOX;
   1061 		goto out1;
   1062 	}
   1063 
   1064 	need_rwunlock = 1;
   1065 
   1066 	va.va_mask = AT_ALL;
   1067 	error = VOP_GETATTR(vp, &va, 0, cr, &ct);
   1068 
   1069 	/*
   1070 	 * If we can't get the attributes, then we can't do the
   1071 	 * right access checking.  So, we'll fail the request.
   1072 	 */
   1073 	if (error)
   1074 		goto out;
   1075 
   1076 #ifdef DEBUG
   1077 	if (rfs3_do_post_op_attr)
   1078 		vap = &va;
   1079 #else
   1080 	vap = &va;
   1081 #endif
   1082 
   1083 	if (vp->v_type != VREG) {
   1084 		resp->status = NFS3ERR_INVAL;
   1085 		goto out1;
   1086 	}
   1087 
   1088 	if (crgetuid(cr) != va.va_uid) {
   1089 		error = VOP_ACCESS(vp, VREAD, 0, cr, &ct);
   1090 		if (error) {
   1091 			if (curthread->t_flag & T_WOULDBLOCK)
   1092 				goto out;
   1093 			error = VOP_ACCESS(vp, VEXEC, 0, cr, &ct);
   1094 			if (error)
   1095 				goto out;
   1096 		}
   1097 	}
   1098 
   1099 	if (MANDLOCK(vp, va.va_mode)) {
   1100 		resp->status = NFS3ERR_ACCES;
   1101 		goto out1;
   1102 	}
   1103 
   1104 	offset = args->offset;
   1105 	if (offset >= va.va_size) {
   1106 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, &ct);
   1107 		if (in_crit)
   1108 			nbl_end_crit(vp);
   1109 		resp->status = NFS3_OK;
   1110 		vattr_to_post_op_attr(vap, &resp->resok.file_attributes);
   1111 		resp->resok.count = 0;
   1112 		resp->resok.eof = TRUE;
   1113 		resp->resok.data.data_len = 0;
   1114 		resp->resok.data.data_val = NULL;
   1115 		resp->resok.data.mp = NULL;
   1116 		/* RDMA */
   1117 		resp->resok.wlist = args->wlist;
   1118 		resp->resok.wlist_len = resp->resok.count;
   1119 		if (resp->resok.wlist)
   1120 			clist_zero_len(resp->resok.wlist);
   1121 		goto done;
   1122 	}
   1123 
   1124 	if (args->count == 0) {
   1125 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, &ct);
   1126 		if (in_crit)
   1127 			nbl_end_crit(vp);
   1128 		resp->status = NFS3_OK;
   1129 		vattr_to_post_op_attr(vap, &resp->resok.file_attributes);
   1130 		resp->resok.count = 0;
   1131 		resp->resok.eof = FALSE;
   1132 		resp->resok.data.data_len = 0;
   1133 		resp->resok.data.data_val = NULL;
   1134 		resp->resok.data.mp = NULL;
   1135 		/* RDMA */
   1136 		resp->resok.wlist = args->wlist;
   1137 		resp->resok.wlist_len = resp->resok.count;
   1138 		if (resp->resok.wlist)
   1139 			clist_zero_len(resp->resok.wlist);
   1140 		goto done;
   1141 	}
   1142 
   1143 	/*
   1144 	 * do not allocate memory more the max. allowed
   1145 	 * transfer size
   1146 	 */
   1147 	if (args->count > rfs3_tsize(req))
   1148 		args->count = rfs3_tsize(req);
   1149 
   1150 	if (loaned_buffers) {
   1151 		uiop = (uio_t *)rfs_setup_xuio(vp);
   1152 		ASSERT(uiop != NULL);
   1153 		uiop->uio_segflg = UIO_SYSSPACE;
   1154 		uiop->uio_loffset = args->offset;
   1155 		uiop->uio_resid = args->count;
   1156 
   1157 		/* Jump to do the read if successful */
   1158 		if (VOP_REQZCBUF(vp, UIO_READ, (xuio_t *)uiop, cr, &ct) == 0) {
   1159 			/*
   1160 			 * Need to hold the vnode until after VOP_RETZCBUF()
   1161 			 * is called.
   1162 			 */
   1163 			VN_HOLD(vp);
   1164 			goto doio_read;
   1165 		}
   1166 
   1167 		DTRACE_PROBE2(nfss__i__reqzcbuf_failed, int,
   1168 		    uiop->uio_loffset, int, uiop->uio_resid);
   1169 
   1170 		uiop->uio_extflg = 0;
   1171 		/* failure to setup for zero copy */
   1172 		rfs_free_xuio((void *)uiop);
   1173 		loaned_buffers = 0;
   1174 	}
   1175 
   1176 	/*
   1177 	 * If returning data via RDMA Write, then grab the chunk list.
   1178 	 * If we aren't returning READ data w/RDMA_WRITE, then grab
   1179 	 * a mblk.
   1180 	 */
   1181 	if (rdma_used) {
   1182 		(void) rdma_get_wchunk(req, &iov, args->wlist);
   1183 	} else {
   1184 		/*
   1185 		 * mp will contain the data to be sent out in the read reply.
   1186 		 * This will be freed after the reply has been sent out (by the
   1187 		 * driver).
   1188 		 * Let's roundup the data to a BYTES_PER_XDR_UNIT multiple, so
   1189 		 * that the call to xdrmblk_putmblk() never fails.
   1190 		 */
   1191 		mp = allocb_wait(RNDUP(args->count), BPRI_MED, STR_NOSIG,
   1192 		    &alloc_err);
   1193 		ASSERT(mp != NULL);
   1194 		ASSERT(alloc_err == 0);
   1195 
   1196 		iov.iov_base = (caddr_t)mp->b_datap->db_base;
   1197 		iov.iov_len = args->count;
   1198 	}
   1199 
   1200 	uio.uio_iov = &iov;
   1201 	uio.uio_iovcnt = 1;
   1202 	uio.uio_segflg = UIO_SYSSPACE;
   1203 	uio.uio_extflg = UIO_COPY_CACHED;
   1204 	uio.uio_loffset = args->offset;
   1205 	uio.uio_resid = args->count;
   1206 	uiop = &uio;
   1207 
   1208 doio_read:
   1209 	error = VOP_READ(vp, uiop, 0, cr, &ct);
   1210 
   1211 	if (error) {
   1212 		if (mp)
   1213 			freemsg(mp);
   1214 		/* check if a monitor detected a delegation conflict */
   1215 		if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK)) {
   1216 			resp->status = NFS3ERR_JUKEBOX;
   1217 			goto out1;
   1218 		}
   1219 		goto out;
   1220 	}
   1221 
   1222 	/* make mblk using zc buffers */
   1223 	if (loaned_buffers) {
   1224 		mp = uio_to_mblk(uiop);
   1225 		ASSERT(mp != NULL);
   1226 	}
   1227 
   1228 	va.va_mask = AT_ALL;
   1229 	error = VOP_GETATTR(vp, &va, 0, cr, &ct);
   1230 
   1231 #ifdef DEBUG
   1232 	if (rfs3_do_post_op_attr) {
   1233 		if (error)
   1234 			vap = NULL;
   1235 		else
   1236 			vap = &va;
   1237 	} else
   1238 		vap = NULL;
   1239 #else
   1240 	if (error)
   1241 		vap = NULL;
   1242 	else
   1243 		vap = &va;
   1244 #endif
   1245 
   1246 	VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, &ct);
   1247 
   1248 	if (in_crit)
   1249 		nbl_end_crit(vp);
   1250 
   1251 	resp->status = NFS3_OK;
   1252 	vattr_to_post_op_attr(vap, &resp->resok.file_attributes);
   1253 	resp->resok.count = args->count - uiop->uio_resid;
   1254 	if (!error && offset + resp->resok.count == va.va_size)
   1255 		resp->resok.eof = TRUE;
   1256 	else
   1257 		resp->resok.eof = FALSE;
   1258 	resp->resok.data.data_len = resp->resok.count;
   1259 
   1260 	if (mp)
   1261 		rfs_rndup_mblks(mp, resp->resok.count, loaned_buffers);
   1262 
   1263 	resp->resok.data.mp = mp;
   1264 	resp->resok.size = (uint_t)args->count;
   1265 
   1266 	if (rdma_used) {
   1267 		resp->resok.data.data_val = (caddr_t)iov.iov_base;
   1268 		if (!rdma_setup_read_data3(args, &(resp->resok))) {
   1269 			resp->status = NFS3ERR_INVAL;
   1270 		}
   1271 	} else {
   1272 		resp->resok.data.data_val = (caddr_t)mp->b_datap->db_base;
   1273 		(resp->resok).wlist = NULL;
   1274 	}
   1275 
   1276 done:
   1277 	DTRACE_NFSV3_4(op__read__done, struct svc_req *, req,
   1278 	    cred_t *, cr, vnode_t *, vp, READ3res *, resp);
   1279 
   1280 	VN_RELE(vp);
   1281 
   1282 	return;
   1283 
   1284 out:
   1285 	if (curthread->t_flag & T_WOULDBLOCK) {
   1286 		curthread->t_flag &= ~T_WOULDBLOCK;
   1287 		resp->status = NFS3ERR_JUKEBOX;
   1288 	} else
   1289 		resp->status = puterrno3(error);
   1290 out1:
   1291 	DTRACE_NFSV3_4(op__read__done, struct svc_req *, req,
   1292 	    cred_t *, cr, vnode_t *, vp, READ3res *, resp);
   1293 
   1294 	if (vp != NULL) {
   1295 		if (need_rwunlock)
   1296 			VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, &ct);
   1297 		if (in_crit)
   1298 			nbl_end_crit(vp);
   1299 		VN_RELE(vp);
   1300 	}
   1301 	vattr_to_post_op_attr(vap, &resp->resfail.file_attributes);
   1302 }
   1303 
   1304 void
   1305 rfs3_read_free(READ3res *resp)
   1306 {
   1307 	mblk_t *mp;
   1308 
   1309 	if (resp->status == NFS3_OK) {
   1310 		mp = resp->resok.data.mp;
   1311 		if (mp != NULL)
   1312 			freemsg(mp);
   1313 	}
   1314 }
   1315 
   1316 void *
   1317 rfs3_read_getfh(READ3args *args)
   1318 {
   1319 
   1320 	return (&args->file);
   1321 }
   1322 
   1323 #define	MAX_IOVECS	12
   1324 
   1325 #ifdef DEBUG
   1326 static int rfs3_write_hits = 0;
   1327 static int rfs3_write_misses = 0;
   1328 #endif
   1329 
   1330 void
   1331 rfs3_write(WRITE3args *args, WRITE3res *resp, struct exportinfo *exi,
   1332 	struct svc_req *req, cred_t *cr)
   1333 {
   1334 	int error;
   1335 	vnode_t *vp;
   1336 	struct vattr *bvap = NULL;
   1337 	struct vattr bva;
   1338 	struct vattr *avap = NULL;
   1339 	struct vattr ava;
   1340 	u_offset_t rlimit;
   1341 	struct uio uio;
   1342 	struct iovec iov[MAX_IOVECS];
   1343 	mblk_t *m;
   1344 	struct iovec *iovp;
   1345 	int iovcnt;
   1346 	int ioflag;
   1347 	cred_t *savecred;
   1348 	int in_crit = 0;
   1349 	int rwlock_ret = -1;
   1350 	caller_context_t ct;
   1351 
   1352 	vp = nfs3_fhtovp(&args->file, exi);
   1353 
   1354 	DTRACE_NFSV3_4(op__write__start, struct svc_req *, req,
   1355 	    cred_t *, cr, vnode_t *, vp, WRITE3args *, args);
   1356 
   1357 	if (vp == NULL) {
   1358 		error = ESTALE;
   1359 		goto err;
   1360 	}
   1361 
   1362 	if (is_system_labeled()) {
   1363 		bslabel_t *clabel = req->rq_label;
   1364 
   1365 		ASSERT(clabel != NULL);
   1366 		DTRACE_PROBE2(tx__rfs3__log__info__opwrite__clabel, char *,
   1367 		    "got client label from request(1)", struct svc_req *, req);
   1368 
   1369 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   1370 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
   1371 			    exi)) {
   1372 				resp->status = NFS3ERR_ACCES;
   1373 				goto err1;
   1374 			}
   1375 		}
   1376 	}
   1377 
   1378 	ct.cc_sysid = 0;
   1379 	ct.cc_pid = 0;
   1380 	ct.cc_caller_id = nfs3_srv_caller_id;
   1381 	ct.cc_flags = CC_DONTBLOCK;
   1382 
   1383 	/*
   1384 	 * We have to enter the critical region before calling VOP_RWLOCK
   1385 	 * to avoid a deadlock with ufs.
   1386 	 */
   1387 	if (nbl_need_check(vp)) {
   1388 		nbl_start_crit(vp, RW_READER);
   1389 		in_crit = 1;
   1390 		if (nbl_conflict(vp, NBL_WRITE, args->offset, args->count, 0,
   1391 		    NULL)) {
   1392 			error = EACCES;
   1393 			goto err;
   1394 		}
   1395 	}
   1396 
   1397 	rwlock_ret = VOP_RWLOCK(vp, V_WRITELOCK_TRUE, &ct);
   1398 
   1399 	/* check if a monitor detected a delegation conflict */
   1400 	if (rwlock_ret == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK)) {
   1401 		resp->status = NFS3ERR_JUKEBOX;
   1402 		rwlock_ret = -1;
   1403 		goto err1;
   1404 	}
   1405 
   1406 
   1407 	bva.va_mask = AT_ALL;
   1408 	error = VOP_GETATTR(vp, &bva, 0, cr, &ct);
   1409 
   1410 	/*
   1411 	 * If we can't get the attributes, then we can't do the
   1412 	 * right access checking.  So, we'll fail the request.
   1413 	 */
   1414 	if (error)
   1415 		goto err;
   1416 
   1417 	bvap = &bva;
   1418 #ifdef DEBUG
   1419 	if (!rfs3_do_pre_op_attr)
   1420 		bvap = NULL;
   1421 #endif
   1422 	avap = bvap;
   1423 
   1424 	if (args->count != args->data.data_len) {
   1425 		resp->status = NFS3ERR_INVAL;
   1426 		goto err1;
   1427 	}
   1428 
   1429 	if (rdonly(exi, req)) {
   1430 		resp->status = NFS3ERR_ROFS;
   1431 		goto err1;
   1432 	}
   1433 
   1434 	if (vp->v_type != VREG) {
   1435 		resp->status = NFS3ERR_INVAL;
   1436 		goto err1;
   1437 	}
   1438 
   1439 	if (crgetuid(cr) != bva.va_uid &&
   1440 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, &ct)))
   1441 		goto err;
   1442 
   1443 	if (MANDLOCK(vp, bva.va_mode)) {
   1444 		resp->status = NFS3ERR_ACCES;
   1445 		goto err1;
   1446 	}
   1447 
   1448 	if (args->count == 0) {
   1449 		resp->status = NFS3_OK;
   1450 		vattr_to_wcc_data(bvap, avap, &resp->resok.file_wcc);
   1451 		resp->resok.count = 0;
   1452 		resp->resok.committed = args->stable;
   1453 		resp->resok.verf = write3verf;
   1454 		goto out;
   1455 	}
   1456 
   1457 	if (args->mblk != NULL) {
   1458 		iovcnt = 0;
   1459 		for (m = args->mblk; m != NULL; m = m->b_cont)
   1460 			iovcnt++;
   1461 		if (iovcnt <= MAX_IOVECS) {
   1462 #ifdef DEBUG
   1463 			rfs3_write_hits++;
   1464 #endif
   1465 			iovp = iov;
   1466 		} else {
   1467 #ifdef DEBUG
   1468 			rfs3_write_misses++;
   1469 #endif
   1470 			iovp = kmem_alloc(sizeof (*iovp) * iovcnt, KM_SLEEP);
   1471 		}
   1472 		mblk_to_iov(args->mblk, iovcnt, iovp);
   1473 
   1474 	} else if (args->rlist != NULL) {
   1475 		iovcnt = 1;
   1476 		iovp = iov;
   1477 		iovp->iov_base = (char *)((args->rlist)->u.c_daddr3);
   1478 		iovp->iov_len = args->count;
   1479 	} else {
   1480 		iovcnt = 1;
   1481 		iovp = iov;
   1482 		iovp->iov_base = args->data.data_val;
   1483 		iovp->iov_len = args->count;
   1484 	}
   1485 
   1486 	uio.uio_iov = iovp;
   1487 	uio.uio_iovcnt = iovcnt;
   1488 
   1489 	uio.uio_segflg = UIO_SYSSPACE;
   1490 	uio.uio_extflg = UIO_COPY_DEFAULT;
   1491 	uio.uio_loffset = args->offset;
   1492 	uio.uio_resid = args->count;
   1493 	uio.uio_llimit = curproc->p_fsz_ctl;
   1494 	rlimit = uio.uio_llimit - args->offset;
   1495 	if (rlimit < (u_offset_t)uio.uio_resid)
   1496 		uio.uio_resid = (int)rlimit;
   1497 
   1498 	if (args->stable == UNSTABLE)
   1499 		ioflag = 0;
   1500 	else if (args->stable == FILE_SYNC)
   1501 		ioflag = FSYNC;
   1502 	else if (args->stable == DATA_SYNC)
   1503 		ioflag = FDSYNC;
   1504 	else {
   1505 		if (iovp != iov)
   1506 			kmem_free(iovp, sizeof (*iovp) * iovcnt);
   1507 		resp->status = NFS3ERR_INVAL;
   1508 		goto err1;
   1509 	}
   1510 
   1511 	/*
   1512 	 * We're changing creds because VM may fault and we need
   1513 	 * the cred of the current thread to be used if quota
   1514 	 * checking is enabled.
   1515 	 */
   1516 	savecred = curthread->t_cred;
   1517 	curthread->t_cred = cr;
   1518 	error = VOP_WRITE(vp, &uio, ioflag, cr, &ct);
   1519 	curthread->t_cred = savecred;
   1520 
   1521 	if (iovp != iov)
   1522 		kmem_free(iovp, sizeof (*iovp) * iovcnt);
   1523 
   1524 	/* check if a monitor detected a delegation conflict */
   1525 	if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK)) {
   1526 		resp->status = NFS3ERR_JUKEBOX;
   1527 		goto err1;
   1528 	}
   1529 
   1530 	ava.va_mask = AT_ALL;
   1531 	avap = VOP_GETATTR(vp, &ava, 0, cr, &ct) ? NULL : &ava;
   1532 
   1533 #ifdef DEBUG
   1534 	if (!rfs3_do_post_op_attr)
   1535 		avap = NULL;
   1536 #endif
   1537 
   1538 	if (error)
   1539 		goto err;
   1540 
   1541 	/*
   1542 	 * If we were unable to get the V_WRITELOCK_TRUE, then we
   1543 	 * may not have accurate after attrs, so check if
   1544 	 * we have both attributes, they have a non-zero va_seq, and
   1545 	 * va_seq has changed by exactly one,
   1546 	 * if not, turn off the before attr.
   1547 	 */
   1548 	if (rwlock_ret != V_WRITELOCK_TRUE) {
   1549 		if (bvap == NULL || avap == NULL ||
   1550 		    bvap->va_seq == 0 || avap->va_seq == 0 ||
   1551 		    avap->va_seq != (bvap->va_seq + 1)) {
   1552 			bvap = NULL;
   1553 		}
   1554 	}
   1555 
   1556 	resp->status = NFS3_OK;
   1557 	vattr_to_wcc_data(bvap, avap, &resp->resok.file_wcc);
   1558 	resp->resok.count = args->count - uio.uio_resid;
   1559 	resp->resok.committed = args->stable;
   1560 	resp->resok.verf = write3verf;
   1561 	goto out;
   1562 
   1563 err:
   1564 	if (curthread->t_flag & T_WOULDBLOCK) {
   1565 		curthread->t_flag &= ~T_WOULDBLOCK;
   1566 		resp->status = NFS3ERR_JUKEBOX;
   1567 	} else
   1568 		resp->status = puterrno3(error);
   1569 err1:
   1570 	vattr_to_wcc_data(bvap, avap, &resp->resfail.file_wcc);
   1571 out:
   1572 	DTRACE_NFSV3_4(op__write__done, struct svc_req *, req,
   1573 	    cred_t *, cr, vnode_t *, vp, WRITE3res *, resp);
   1574 
   1575 	if (vp != NULL) {
   1576 		if (rwlock_ret != -1)
   1577 			VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, &ct);
   1578 		if (in_crit)
   1579 			nbl_end_crit(vp);
   1580 		VN_RELE(vp);
   1581 	}
   1582 }
   1583 
   1584 void *
   1585 rfs3_write_getfh(WRITE3args *args)
   1586 {
   1587 
   1588 	return (&args->file);
   1589 }
   1590 
   1591 void
   1592 rfs3_create(CREATE3args *args, CREATE3res *resp, struct exportinfo *exi,
   1593 	struct svc_req *req, cred_t *cr)
   1594 {
   1595 	int error;
   1596 	int in_crit = 0;
   1597 	vnode_t *vp;
   1598 	vnode_t *tvp = NULL;
   1599 	vnode_t *dvp;
   1600 	struct vattr *vap;
   1601 	struct vattr va;
   1602 	struct vattr *dbvap;
   1603 	struct vattr dbva;
   1604 	struct vattr *davap;
   1605 	struct vattr dava;
   1606 	enum vcexcl excl;
   1607 	nfstime3 *mtime;
   1608 	len_t reqsize;
   1609 	bool_t trunc;
   1610 	struct sockaddr *ca;
   1611 	char *name = NULL;
   1612 
   1613 	dbvap = NULL;
   1614 	davap = NULL;
   1615 
   1616 	dvp = nfs3_fhtovp(&args->where.dir, exi);
   1617 
   1618 	DTRACE_NFSV3_4(op__create__start, struct svc_req *, req,
   1619 	    cred_t *, cr, vnode_t *, dvp, CREATE3args *, args);
   1620 
   1621 	if (dvp == NULL) {
   1622 		error = ESTALE;
   1623 		goto out;
   1624 	}
   1625 
   1626 #ifdef DEBUG
   1627 	if (rfs3_do_pre_op_attr) {
   1628 		dbva.va_mask = AT_ALL;
   1629 		dbvap = VOP_GETATTR(dvp, &dbva, 0, cr, NULL) ? NULL : &dbva;
   1630 	} else
   1631 		dbvap = NULL;
   1632 #else
   1633 	dbva.va_mask = AT_ALL;
   1634 	dbvap = VOP_GETATTR(dvp, &dbva, 0, cr, NULL) ? NULL : &dbva;
   1635 #endif
   1636 	davap = dbvap;
   1637 
   1638 	if (args->where.name == nfs3nametoolong) {
   1639 		resp->status = NFS3ERR_NAMETOOLONG;
   1640 		goto out1;
   1641 	}
   1642 
   1643 	if (args->where.name == NULL || *(args->where.name) == '\0') {
   1644 		resp->status = NFS3ERR_ACCES;
   1645 		goto out1;
   1646 	}
   1647 
   1648 	if (rdonly(exi, req)) {
   1649 		resp->status = NFS3ERR_ROFS;
   1650 		goto out1;
   1651 	}
   1652 
   1653 	if (is_system_labeled()) {
   1654 		bslabel_t *clabel = req->rq_label;
   1655 
   1656 		ASSERT(clabel != NULL);
   1657 		DTRACE_PROBE2(tx__rfs3__log__info__opcreate__clabel, char *,
   1658 		    "got client label from request(1)", struct svc_req *, req);
   1659 
   1660 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   1661 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
   1662 			    exi)) {
   1663 				resp->status = NFS3ERR_ACCES;
   1664 				goto out1;
   1665 			}
   1666 		}
   1667 	}
   1668 
   1669 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
   1670 	name = nfscmd_convname(ca, exi, args->where.name,
   1671 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
   1672 
   1673 	if (name == NULL) {
   1674 		/* This is really a Solaris EILSEQ */
   1675 		resp->status = NFS3ERR_INVAL;
   1676 		goto out1;
   1677 	}
   1678 
   1679 	if (args->how.mode == EXCLUSIVE) {
   1680 		va.va_mask = AT_TYPE | AT_MODE | AT_MTIME;
   1681 		va.va_type = VREG;
   1682 		va.va_mode = (mode_t)0;
   1683 		/*
   1684 		 * Ensure no time overflows and that types match
   1685 		 */
   1686 		mtime = (nfstime3 *)&args->how.createhow3_u.verf;
   1687 		va.va_mtime.tv_sec = mtime->seconds % INT32_MAX;
   1688 		va.va_mtime.tv_nsec = mtime->nseconds;
   1689 		excl = EXCL;
   1690 	} else {
   1691 		error = sattr3_to_vattr(&args->how.createhow3_u.obj_attributes,
   1692 		    &va);
   1693 		if (error)
   1694 			goto out;
   1695 		va.va_mask |= AT_TYPE;
   1696 		va.va_type = VREG;
   1697 		if (args->how.mode == GUARDED)
   1698 			excl = EXCL;
   1699 		else {
   1700 			excl = NONEXCL;
   1701 
   1702 			/*
   1703 			 * During creation of file in non-exclusive mode
   1704 			 * if size of file is being set then make sure
   1705 			 * that if the file already exists that no conflicting
   1706 			 * non-blocking mandatory locks exists in the region
   1707 			 * being modified. If there are conflicting locks fail
   1708 			 * the operation with EACCES.
   1709 			 */
   1710 			if (va.va_mask & AT_SIZE) {
   1711 				struct vattr tva;
   1712 
   1713 				/*
   1714 				 * Does file already exist?
   1715 				 */
   1716 				error = VOP_LOOKUP(dvp, name, &tvp,
   1717 				    NULL, 0, NULL, cr, NULL, NULL, NULL);
   1718 
   1719 				/*
   1720 				 * Check to see if the file has been delegated
   1721 				 * to a v4 client.  If so, then begin recall of
   1722 				 * the delegation and return JUKEBOX to allow
   1723 				 * the client to retrasmit its request.
   1724 				 */
   1725 
   1726 				trunc = va.va_size == 0;
   1727 				if (!error &&
   1728 				    rfs4_check_delegated(FWRITE, tvp, trunc)) {
   1729 					resp->status = NFS3ERR_JUKEBOX;
   1730 					goto out1;
   1731 				}
   1732 
   1733 				/*
   1734 				 * Check for NBMAND lock conflicts
   1735 				 */
   1736 				if (!error && nbl_need_check(tvp)) {
   1737 					u_offset_t offset;
   1738 					ssize_t len;
   1739 
   1740 					nbl_start_crit(tvp, RW_READER);
   1741 					in_crit = 1;
   1742 
   1743 					tva.va_mask = AT_SIZE;
   1744 					error = VOP_GETATTR(tvp, &tva, 0, cr,
   1745 					    NULL);
   1746 					/*
   1747 					 * Can't check for conflicts, so return
   1748 					 * error.
   1749 					 */
   1750 					if (error)
   1751 						goto out;
   1752 
   1753 					offset = tva.va_size < va.va_size ?
   1754 					    tva.va_size : va.va_size;
   1755 					len = tva.va_size < va.va_size ?
   1756 					    va.va_size - tva.va_size :
   1757 					    tva.va_size - va.va_size;
   1758 					if (nbl_conflict(tvp, NBL_WRITE,
   1759 					    offset, len, 0, NULL)) {
   1760 						error = EACCES;
   1761 						goto out;
   1762 					}
   1763 				} else if (tvp) {
   1764 					VN_RELE(tvp);
   1765 					tvp = NULL;
   1766 				}
   1767 			}
   1768 		}
   1769 		if (va.va_mask & AT_SIZE)
   1770 			reqsize = va.va_size;
   1771 	}
   1772 
   1773 	/*
   1774 	 * Must specify the mode.
   1775 	 */
   1776 	if (!(va.va_mask & AT_MODE)) {
   1777 		resp->status = NFS3ERR_INVAL;
   1778 		goto out1;
   1779 	}
   1780 
   1781 	/*
   1782 	 * If the filesystem is exported with nosuid, then mask off
   1783 	 * the setuid and setgid bits.
   1784 	 */
   1785 	if (va.va_type == VREG && (exi->exi_export.ex_flags & EX_NOSUID))
   1786 		va.va_mode &= ~(VSUID | VSGID);
   1787 
   1788 tryagain:
   1789 	/*
   1790 	 * The file open mode used is VWRITE.  If the client needs
   1791 	 * some other semantic, then it should do the access checking
   1792 	 * itself.  It would have been nice to have the file open mode
   1793 	 * passed as part of the arguments.
   1794 	 */
   1795 	error = VOP_CREATE(dvp, name, &va, excl, VWRITE,
   1796 	    &vp, cr, 0, NULL, NULL);
   1797 
   1798 #ifdef DEBUG
   1799 	if (rfs3_do_post_op_attr) {
   1800 		dava.va_mask = AT_ALL;
   1801 		davap = VOP_GETATTR(dvp, &dava, 0, cr, NULL) ? NULL : &dava;
   1802 	} else
   1803 		davap = NULL;
   1804 #else
   1805 	dava.va_mask = AT_ALL;
   1806 	davap = VOP_GETATTR(dvp, &dava, 0, cr, NULL) ? NULL : &dava;
   1807 #endif
   1808 
   1809 	if (error) {
   1810 		/*
   1811 		 * If we got something other than file already exists
   1812 		 * then just return this error.  Otherwise, we got
   1813 		 * EEXIST.  If we were doing a GUARDED create, then
   1814 		 * just return this error.  Otherwise, we need to
   1815 		 * make sure that this wasn't a duplicate of an
   1816 		 * exclusive create request.
   1817 		 *
   1818 		 * The assumption is made that a non-exclusive create
   1819 		 * request will never return EEXIST.
   1820 		 */
   1821 		if (error != EEXIST || args->how.mode == GUARDED)
   1822 			goto out;
   1823 		/*
   1824 		 * Lookup the file so that we can get a vnode for it.
   1825 		 */
   1826 		error = VOP_LOOKUP(dvp, name, &vp, NULL, 0,
   1827 		    NULL, cr, NULL, NULL, NULL);
   1828 		if (error) {
   1829 			/*
   1830 			 * We couldn't find the file that we thought that
   1831 			 * we just created.  So, we'll just try creating
   1832 			 * it again.
   1833 			 */
   1834 			if (error == ENOENT)
   1835 				goto tryagain;
   1836 			goto out;
   1837 		}
   1838 
   1839 		/*
   1840 		 * If the file is delegated to a v4 client, go ahead
   1841 		 * and initiate recall, this create is a hint that a
   1842 		 * conflicting v3 open has occurred.
   1843 		 */
   1844 
   1845 		if (rfs4_check_delegated(FWRITE, vp, FALSE)) {
   1846 			VN_RELE(vp);
   1847 			resp->status = NFS3ERR_JUKEBOX;
   1848 			goto out1;
   1849 		}
   1850 
   1851 		va.va_mask = AT_ALL;
   1852 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   1853 
   1854 		mtime = (nfstime3 *)&args->how.createhow3_u.verf;
   1855 		/* % with INT32_MAX to prevent overflows */
   1856 		if (args->how.mode == EXCLUSIVE && (vap == NULL ||
   1857 		    vap->va_mtime.tv_sec !=
   1858 		    (mtime->seconds % INT32_MAX) ||
   1859 		    vap->va_mtime.tv_nsec != mtime->nseconds)) {
   1860 			VN_RELE(vp);
   1861 			error = EEXIST;
   1862 			goto out;
   1863 		}
   1864 	} else {
   1865 
   1866 		if ((args->how.mode == UNCHECKED ||
   1867 		    args->how.mode == GUARDED) &&
   1868 		    args->how.createhow3_u.obj_attributes.size.set_it &&
   1869 		    va.va_size == 0)
   1870 			trunc = TRUE;
   1871 		else
   1872 			trunc = FALSE;
   1873 
   1874 		if (rfs4_check_delegated(FWRITE, vp, trunc)) {
   1875 			VN_RELE(vp);
   1876 			resp->status = NFS3ERR_JUKEBOX;
   1877 			goto out1;
   1878 		}
   1879 
   1880 		va.va_mask = AT_ALL;
   1881 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   1882 
   1883 		/*
   1884 		 * We need to check to make sure that the file got
   1885 		 * created to the indicated size.  If not, we do a
   1886 		 * setattr to try to change the size, but we don't
   1887 		 * try too hard.  This shouldn't a problem as most
   1888 		 * clients will only specifiy a size of zero which
   1889 		 * local file systems handle.  However, even if
   1890 		 * the client does specify a non-zero size, it can
   1891 		 * still recover by checking the size of the file
   1892 		 * after it has created it and then issue a setattr
   1893 		 * request of its own to set the size of the file.
   1894 		 */
   1895 		if (vap != NULL &&
   1896 		    (args->how.mode == UNCHECKED ||
   1897 		    args->how.mode == GUARDED) &&
   1898 		    args->how.createhow3_u.obj_attributes.size.set_it &&
   1899 		    vap->va_size != reqsize) {
   1900 			va.va_mask = AT_SIZE;
   1901 			va.va_size = reqsize;
   1902 			(void) VOP_SETATTR(vp, &va, 0, cr, NULL);
   1903 			va.va_mask = AT_ALL;
   1904 			vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   1905 		}
   1906 	}
   1907 
   1908 	if (name != args->where.name)
   1909 		kmem_free(name, MAXPATHLEN + 1);
   1910 
   1911 #ifdef DEBUG
   1912 	if (!rfs3_do_post_op_attr)
   1913 		vap = NULL;
   1914 #endif
   1915 
   1916 #ifdef DEBUG
   1917 	if (!rfs3_do_post_op_fh3)
   1918 		resp->resok.obj.handle_follows = FALSE;
   1919 	else {
   1920 #endif
   1921 	error = makefh3(&resp->resok.obj.handle, vp, exi);
   1922 	if (error)
   1923 		resp->resok.obj.handle_follows = FALSE;
   1924 	else
   1925 		resp->resok.obj.handle_follows = TRUE;
   1926 #ifdef DEBUG
   1927 	}
   1928 #endif
   1929 
   1930 	/*
   1931 	 * Force modified data and metadata out to stable storage.
   1932 	 */
   1933 	(void) VOP_FSYNC(vp, FNODSYNC, cr, NULL);
   1934 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
   1935 
   1936 	VN_RELE(vp);
   1937 	if (tvp != NULL) {
   1938 		if (in_crit)
   1939 			nbl_end_crit(tvp);
   1940 		VN_RELE(tvp);
   1941 	}
   1942 
   1943 	resp->status = NFS3_OK;
   1944 	vattr_to_post_op_attr(vap, &resp->resok.obj_attributes);
   1945 	vattr_to_wcc_data(dbvap, davap, &resp->resok.dir_wcc);
   1946 
   1947 	DTRACE_NFSV3_4(op__create__done, struct svc_req *, req,
   1948 	    cred_t *, cr, vnode_t *, dvp, CREATE3res *, resp);
   1949 
   1950 	VN_RELE(dvp);
   1951 	return;
   1952 
   1953 out:
   1954 	if (curthread->t_flag & T_WOULDBLOCK) {
   1955 		curthread->t_flag &= ~T_WOULDBLOCK;
   1956 		resp->status = NFS3ERR_JUKEBOX;
   1957 	} else
   1958 		resp->status = puterrno3(error);
   1959 out1:
   1960 	DTRACE_NFSV3_4(op__create__done, struct svc_req *, req,
   1961 	    cred_t *, cr, vnode_t *, dvp, CREATE3res *, resp);
   1962 
   1963 	if (name != NULL && name != args->where.name)
   1964 		kmem_free(name, MAXPATHLEN + 1);
   1965 
   1966 	if (tvp != NULL) {
   1967 		if (in_crit)
   1968 			nbl_end_crit(tvp);
   1969 		VN_RELE(tvp);
   1970 	}
   1971 	if (dvp != NULL)
   1972 		VN_RELE(dvp);
   1973 	vattr_to_wcc_data(dbvap, davap, &resp->resfail.dir_wcc);
   1974 }
   1975 
   1976 void *
   1977 rfs3_create_getfh(CREATE3args *args)
   1978 {
   1979 
   1980 	return (&args->where.dir);
   1981 }
   1982 
   1983 void
   1984 rfs3_mkdir(MKDIR3args *args, MKDIR3res *resp, struct exportinfo *exi,
   1985 	struct svc_req *req, cred_t *cr)
   1986 {
   1987 	int error;
   1988 	vnode_t *vp = NULL;
   1989 	vnode_t *dvp;
   1990 	struct vattr *vap;
   1991 	struct vattr va;
   1992 	struct vattr *dbvap;
   1993 	struct vattr dbva;
   1994 	struct vattr *davap;
   1995 	struct vattr dava;
   1996 	struct sockaddr *ca;
   1997 	char *name = NULL;
   1998 
   1999 	dbvap = NULL;
   2000 	davap = NULL;
   2001 
   2002 	dvp = nfs3_fhtovp(&args->where.dir, exi);
   2003 
   2004 	DTRACE_NFSV3_4(op__mkdir__start, struct svc_req *, req,
   2005 	    cred_t *, cr, vnode_t *, dvp, MKDIR3args *, args);
   2006 
   2007 	if (dvp == NULL) {
   2008 		error = ESTALE;
   2009 		goto out;
   2010 	}
   2011 
   2012 #ifdef DEBUG
   2013 	if (rfs3_do_pre_op_attr) {
   2014 		dbva.va_mask = AT_ALL;
   2015 		dbvap = VOP_GETATTR(dvp, &dbva, 0, cr, NULL) ? NULL : &dbva;
   2016 	} else
   2017 		dbvap = NULL;
   2018 #else
   2019 	dbva.va_mask = AT_ALL;
   2020 	dbvap = VOP_GETATTR(dvp, &dbva, 0, cr, NULL) ? NULL : &dbva;
   2021 #endif
   2022 	davap = dbvap;
   2023 
   2024 	if (args->where.name == nfs3nametoolong) {
   2025 		resp->status = NFS3ERR_NAMETOOLONG;
   2026 		goto out1;
   2027 	}
   2028 
   2029 	if (args->where.name == NULL || *(args->where.name) == '\0') {
   2030 		resp->status = NFS3ERR_ACCES;
   2031 		goto out1;
   2032 	}
   2033 
   2034 	if (rdonly(exi, req)) {
   2035 		resp->status = NFS3ERR_ROFS;
   2036 		goto out1;
   2037 	}
   2038 
   2039 	if (is_system_labeled()) {
   2040 		bslabel_t *clabel = req->rq_label;
   2041 
   2042 		ASSERT(clabel != NULL);
   2043 		DTRACE_PROBE2(tx__rfs3__log__info__opmkdir__clabel, char *,
   2044 		    "got client label from request(1)", struct svc_req *, req);
   2045 
   2046 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   2047 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
   2048 			    exi)) {
   2049 				resp->status = NFS3ERR_ACCES;
   2050 				goto out1;
   2051 			}
   2052 		}
   2053 	}
   2054 
   2055 	error = sattr3_to_vattr(&args->attributes, &va);
   2056 	if (error)
   2057 		goto out;
   2058 
   2059 	if (!(va.va_mask & AT_MODE)) {
   2060 		resp->status = NFS3ERR_INVAL;
   2061 		goto out1;
   2062 	}
   2063 
   2064 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
   2065 	name = nfscmd_convname(ca, exi, args->where.name,
   2066 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
   2067 
   2068 	if (name == NULL) {
   2069 		resp->status = NFS3ERR_INVAL;
   2070 		goto out1;
   2071 	}
   2072 
   2073 	va.va_mask |= AT_TYPE;
   2074 	va.va_type = VDIR;
   2075 
   2076 	error = VOP_MKDIR(dvp, name, &va, &vp, cr, NULL, 0, NULL);
   2077 
   2078 	if (name != args->where.name)
   2079 		kmem_free(name, MAXPATHLEN + 1);
   2080 
   2081 #ifdef DEBUG
   2082 	if (rfs3_do_post_op_attr) {
   2083 		dava.va_mask = AT_ALL;
   2084 		davap = VOP_GETATTR(dvp, &dava, 0, cr, NULL) ? NULL : &dava;
   2085 	} else
   2086 		davap = NULL;
   2087 #else
   2088 	dava.va_mask = AT_ALL;
   2089 	davap = VOP_GETATTR(dvp, &dava, 0, cr, NULL) ? NULL : &dava;
   2090 #endif
   2091 
   2092 	/*
   2093 	 * Force modified data and metadata out to stable storage.
   2094 	 */
   2095 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
   2096 
   2097 	if (error)
   2098 		goto out;
   2099 
   2100 #ifdef DEBUG
   2101 	if (!rfs3_do_post_op_fh3)
   2102 		resp->resok.obj.handle_follows = FALSE;
   2103 	else {
   2104 #endif
   2105 	error = makefh3(&resp->resok.obj.handle, vp, exi);
   2106 	if (error)
   2107 		resp->resok.obj.handle_follows = FALSE;
   2108 	else
   2109 		resp->resok.obj.handle_follows = TRUE;
   2110 #ifdef DEBUG
   2111 	}
   2112 #endif
   2113 
   2114 #ifdef DEBUG
   2115 	if (rfs3_do_post_op_attr) {
   2116 		va.va_mask = AT_ALL;
   2117 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   2118 	} else
   2119 		vap = NULL;
   2120 #else
   2121 	va.va_mask = AT_ALL;
   2122 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   2123 #endif
   2124 
   2125 	/*
   2126 	 * Force modified data and metadata out to stable storage.
   2127 	 */
   2128 	(void) VOP_FSYNC(vp, 0, cr, NULL);
   2129 
   2130 	VN_RELE(vp);
   2131 
   2132 	resp->status = NFS3_OK;
   2133 	vattr_to_post_op_attr(vap, &resp->resok.obj_attributes);
   2134 	vattr_to_wcc_data(dbvap, davap, &resp->resok.dir_wcc);
   2135 
   2136 	DTRACE_NFSV3_4(op__mkdir__done, struct svc_req *, req,
   2137 	    cred_t *, cr, vnode_t *, dvp, MKDIR3res *, resp);
   2138 	VN_RELE(dvp);
   2139 
   2140 	return;
   2141 
   2142 out:
   2143 	if (curthread->t_flag & T_WOULDBLOCK) {
   2144 		curthread->t_flag &= ~T_WOULDBLOCK;
   2145 		resp->status = NFS3ERR_JUKEBOX;
   2146 	} else
   2147 		resp->status = puterrno3(error);
   2148 out1:
   2149 	DTRACE_NFSV3_4(op__mkdir__done, struct svc_req *, req,
   2150 	    cred_t *, cr, vnode_t *, dvp, MKDIR3res *, resp);
   2151 	if (dvp != NULL)
   2152 		VN_RELE(dvp);
   2153 	vattr_to_wcc_data(dbvap, davap, &resp->resfail.dir_wcc);
   2154 }
   2155 
   2156 void *
   2157 rfs3_mkdir_getfh(MKDIR3args *args)
   2158 {
   2159 
   2160 	return (&args->where.dir);
   2161 }
   2162 
   2163 void
   2164 rfs3_symlink(SYMLINK3args *args, SYMLINK3res *resp, struct exportinfo *exi,
   2165 	struct svc_req *req, cred_t *cr)
   2166 {
   2167 	int error;
   2168 	vnode_t *vp;
   2169 	vnode_t *dvp;
   2170 	struct vattr *vap;
   2171 	struct vattr va;
   2172 	struct vattr *dbvap;
   2173 	struct vattr dbva;
   2174 	struct vattr *davap;
   2175 	struct vattr dava;
   2176 	struct sockaddr *ca;
   2177 	char *name = NULL;
   2178 	char *symdata = NULL;
   2179 
   2180 	dbvap = NULL;
   2181 	davap = NULL;
   2182 
   2183 	dvp = nfs3_fhtovp(&args->where.dir, exi);
   2184 
   2185 	DTRACE_NFSV3_4(op__symlink__start, struct svc_req *, req,
   2186 	    cred_t *, cr, vnode_t *, dvp, SYMLINK3args *, args);
   2187 
   2188 	if (dvp == NULL) {
   2189 		error = ESTALE;
   2190 		goto err;
   2191 	}
   2192 
   2193 #ifdef DEBUG
   2194 	if (rfs3_do_pre_op_attr) {
   2195 		dbva.va_mask = AT_ALL;
   2196 		dbvap = VOP_GETATTR(dvp, &dbva, 0, cr, NULL) ? NULL : &dbva;
   2197 	} else
   2198 		dbvap = NULL;
   2199 #else
   2200 	dbva.va_mask = AT_ALL;
   2201 	dbvap = VOP_GETATTR(dvp, &dbva, 0, cr, NULL) ? NULL : &dbva;
   2202 #endif
   2203 	davap = dbvap;
   2204 
   2205 	if (args->where.name == nfs3nametoolong) {
   2206 		resp->status = NFS3ERR_NAMETOOLONG;
   2207 		goto err1;
   2208 	}
   2209 
   2210 	if (args->where.name == NULL || *(args->where.name) == '\0') {
   2211 		resp->status = NFS3ERR_ACCES;
   2212 		goto err1;
   2213 	}
   2214 
   2215 	if (rdonly(exi, req)) {
   2216 		resp->status = NFS3ERR_ROFS;
   2217 		goto err1;
   2218 	}
   2219 
   2220 	if (is_system_labeled()) {
   2221 		bslabel_t *clabel = req->rq_label;
   2222 
   2223 		ASSERT(clabel != NULL);
   2224 		DTRACE_PROBE2(tx__rfs3__log__info__opsymlink__clabel, char *,
   2225 		    "got client label from request(1)", struct svc_req *, req);
   2226 
   2227 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   2228 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
   2229 			    exi)) {
   2230 				resp->status = NFS3ERR_ACCES;
   2231 				goto err1;
   2232 			}
   2233 		}
   2234 	}
   2235 
   2236 	error = sattr3_to_vattr(&args->symlink.symlink_attributes, &va);
   2237 	if (error)
   2238 		goto err;
   2239 
   2240 	if (!(va.va_mask & AT_MODE)) {
   2241 		resp->status = NFS3ERR_INVAL;
   2242 		goto err1;
   2243 	}
   2244 
   2245 	if (args->symlink.symlink_data == nfs3nametoolong) {
   2246 		resp->status = NFS3ERR_NAMETOOLONG;
   2247 		goto err1;
   2248 	}
   2249 
   2250 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
   2251 	name = nfscmd_convname(ca, exi, args->where.name,
   2252 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
   2253 
   2254 	if (name == NULL) {
   2255 		/* This is really a Solaris EILSEQ */
   2256 		resp->status = NFS3ERR_INVAL;
   2257 		goto err1;
   2258 	}
   2259 
   2260 	symdata = nfscmd_convname(ca, exi, args->symlink.symlink_data,
   2261 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
   2262 	if (symdata == NULL) {
   2263 		/* This is really a Solaris EILSEQ */
   2264 		resp->status = NFS3ERR_INVAL;
   2265 		goto err1;
   2266 	}
   2267 
   2268 
   2269 	va.va_mask |= AT_TYPE;
   2270 	va.va_type = VLNK;
   2271 
   2272 	error = VOP_SYMLINK(dvp, name, &va, symdata, cr, NULL, 0);
   2273 
   2274 #ifdef DEBUG
   2275 	if (rfs3_do_post_op_attr) {
   2276 		dava.va_mask = AT_ALL;
   2277 		davap = VOP_GETATTR(dvp, &dava, 0, cr, NULL) ? NULL : &dava;
   2278 	} else
   2279 		davap = NULL;
   2280 #else
   2281 	dava.va_mask = AT_ALL;
   2282 	davap = VOP_GETATTR(dvp, &dava, 0, cr, NULL) ? NULL : &dava;
   2283 #endif
   2284 
   2285 	if (error)
   2286 		goto err;
   2287 
   2288 	error = VOP_LOOKUP(dvp, name, &vp, NULL, 0, NULL, cr,
   2289 	    NULL, NULL, NULL);
   2290 
   2291 	/*
   2292 	 * Force modified data and metadata out to stable storage.
   2293 	 */
   2294 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
   2295 
   2296 
   2297 	resp->status = NFS3_OK;
   2298 	if (error) {
   2299 		resp->resok.obj.handle_follows = FALSE;
   2300 		vattr_to_post_op_attr(NULL, &resp->resok.obj_attributes);
   2301 		vattr_to_wcc_data(dbvap, davap, &resp->resok.dir_wcc);
   2302 		goto out;
   2303 	}
   2304 
   2305 #ifdef DEBUG
   2306 	if (!rfs3_do_post_op_fh3)
   2307 		resp->resok.obj.handle_follows = FALSE;
   2308 	else {
   2309 #endif
   2310 	error = makefh3(&resp->resok.obj.handle, vp, exi);
   2311 	if (error)
   2312 		resp->resok.obj.handle_follows = FALSE;
   2313 	else
   2314 		resp->resok.obj.handle_follows = TRUE;
   2315 #ifdef DEBUG
   2316 	}
   2317 #endif
   2318 
   2319 #ifdef DEBUG
   2320 	if (rfs3_do_post_op_attr) {
   2321 		va.va_mask = AT_ALL;
   2322 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   2323 	} else
   2324 		vap = NULL;
   2325 #else
   2326 	va.va_mask = AT_ALL;
   2327 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   2328 #endif
   2329 
   2330 	/*
   2331 	 * Force modified data and metadata out to stable storage.
   2332 	 */
   2333 	(void) VOP_FSYNC(vp, 0, cr, NULL);
   2334 
   2335 	VN_RELE(vp);
   2336 
   2337 	vattr_to_post_op_attr(vap, &resp->resok.obj_attributes);
   2338 	vattr_to_wcc_data(dbvap, davap, &resp->resok.dir_wcc);
   2339 	goto out;
   2340 
   2341 err:
   2342 	if (curthread->t_flag & T_WOULDBLOCK) {
   2343 		curthread->t_flag &= ~T_WOULDBLOCK;
   2344 		resp->status = NFS3ERR_JUKEBOX;
   2345 	} else
   2346 		resp->status = puterrno3(error);
   2347 err1:
   2348 	vattr_to_wcc_data(dbvap, davap, &resp->resfail.dir_wcc);
   2349 out:
   2350 	if (name != NULL && name != args->where.name)
   2351 		kmem_free(name, MAXPATHLEN + 1);
   2352 	if (symdata != NULL && symdata != args->symlink.symlink_data)
   2353 		kmem_free(symdata, MAXPATHLEN + 1);
   2354 
   2355 	DTRACE_NFSV3_4(op__symlink__done, struct svc_req *, req,
   2356 	    cred_t *, cr, vnode_t *, dvp, SYMLINK3res *, resp);
   2357 
   2358 	if (dvp != NULL)
   2359 		VN_RELE(dvp);
   2360 }
   2361 
   2362 void *
   2363 rfs3_symlink_getfh(SYMLINK3args *args)
   2364 {
   2365 
   2366 	return (&args->where.dir);
   2367 }
   2368 
   2369 void
   2370 rfs3_mknod(MKNOD3args *args, MKNOD3res *resp, struct exportinfo *exi,
   2371 	struct svc_req *req, cred_t *cr)
   2372 {
   2373 	int error;
   2374 	vnode_t *vp;
   2375 	vnode_t *realvp;
   2376 	vnode_t *dvp;
   2377 	struct vattr *vap;
   2378 	struct vattr va;
   2379 	struct vattr *dbvap;
   2380 	struct vattr dbva;
   2381 	struct vattr *davap;
   2382 	struct vattr dava;
   2383 	int mode;
   2384 	enum vcexcl excl;
   2385 	struct sockaddr *ca;
   2386 	char *name = NULL;
   2387 
   2388 	dbvap = NULL;
   2389 	davap = NULL;
   2390 
   2391 	dvp = nfs3_fhtovp(&args->where.dir, exi);
   2392 
   2393 	DTRACE_NFSV3_4(op__mknod__start, struct svc_req *, req,
   2394 	    cred_t *, cr, vnode_t *, dvp, MKNOD3args *, args);
   2395 
   2396 	if (dvp == NULL) {
   2397 		error = ESTALE;
   2398 		goto out;
   2399 	}
   2400 
   2401 #ifdef DEBUG
   2402 	if (rfs3_do_pre_op_attr) {
   2403 		dbva.va_mask = AT_ALL;
   2404 		dbvap = VOP_GETATTR(dvp, &dbva, 0, cr, NULL) ? NULL : &dbva;
   2405 	} else
   2406 		dbvap = NULL;
   2407 #else
   2408 	dbva.va_mask = AT_ALL;
   2409 	dbvap = VOP_GETATTR(dvp, &dbva, 0, cr, NULL) ? NULL : &dbva;
   2410 #endif
   2411 	davap = dbvap;
   2412 
   2413 	if (args->where.name == nfs3nametoolong) {
   2414 		resp->status = NFS3ERR_NAMETOOLONG;
   2415 		goto out1;
   2416 	}
   2417 
   2418 	if (args->where.name == NULL || *(args->where.name) == '\0') {
   2419 		resp->status = NFS3ERR_ACCES;
   2420 		goto out1;
   2421 	}
   2422 
   2423 	if (rdonly(exi, req)) {
   2424 		resp->status = NFS3ERR_ROFS;
   2425 		goto out1;
   2426 	}
   2427 
   2428 	if (is_system_labeled()) {
   2429 		bslabel_t *clabel = req->rq_label;
   2430 
   2431 		ASSERT(clabel != NULL);
   2432 		DTRACE_PROBE2(tx__rfs3__log__info__opmknod__clabel, char *,
   2433 		    "got client label from request(1)", struct svc_req *, req);
   2434 
   2435 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   2436 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
   2437 			    exi)) {
   2438 				resp->status = NFS3ERR_ACCES;
   2439 				goto out1;
   2440 			}
   2441 		}
   2442 	}
   2443 
   2444 	switch (args->what.type) {
   2445 	case NF3CHR:
   2446 	case NF3BLK:
   2447 		error = sattr3_to_vattr(
   2448 		    &args->what.mknoddata3_u.device.dev_attributes, &va);
   2449 		if (error)
   2450 			goto out;
   2451 		if (secpolicy_sys_devices(cr) != 0) {
   2452 			resp->status = NFS3ERR_PERM;
   2453 			goto out1;
   2454 		}
   2455 		if (args->what.type == NF3CHR)
   2456 			va.va_type = VCHR;
   2457 		else
   2458 			va.va_type = VBLK;
   2459 		va.va_rdev = makedevice(
   2460 		    args->what.mknoddata3_u.device.spec.specdata1,
   2461 		    args->what.mknoddata3_u.device.spec.specdata2);
   2462 		va.va_mask |= AT_TYPE | AT_RDEV;
   2463 		break;
   2464 	case NF3SOCK:
   2465 		error = sattr3_to_vattr(
   2466 		    &args->what.mknoddata3_u.pipe_attributes, &va);
   2467 		if (error)
   2468 			goto out;
   2469 		va.va_type = VSOCK;
   2470 		va.va_mask |= AT_TYPE;
   2471 		break;
   2472 	case NF3FIFO:
   2473 		error = sattr3_to_vattr(
   2474 		    &args->what.mknoddata3_u.pipe_attributes, &va);
   2475 		if (error)
   2476 			goto out;
   2477 		va.va_type = VFIFO;
   2478 		va.va_mask |= AT_TYPE;
   2479 		break;
   2480 	default:
   2481 		resp->status = NFS3ERR_BADTYPE;
   2482 		goto out1;
   2483 	}
   2484 
   2485 	/*
   2486 	 * Must specify the mode.
   2487 	 */
   2488 	if (!(va.va_mask & AT_MODE)) {
   2489 		resp->status = NFS3ERR_INVAL;
   2490 		goto out1;
   2491 	}
   2492 
   2493 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
   2494 	name = nfscmd_convname(ca, exi, args->where.name,
   2495 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
   2496 
   2497 	if (name == NULL) {
   2498 		resp->status = NFS3ERR_INVAL;
   2499 		goto out1;
   2500 	}
   2501 
   2502 	excl = EXCL;
   2503 
   2504 	mode = 0;
   2505 
   2506 	error = VOP_CREATE(dvp, name, &va, excl, mode,
   2507 	    &vp, cr, 0, NULL, NULL);
   2508 
   2509 	if (name != args->where.name)
   2510 		kmem_free(name, MAXPATHLEN + 1);
   2511 
   2512 #ifdef DEBUG
   2513 	if (rfs3_do_post_op_attr) {
   2514 		dava.va_mask = AT_ALL;
   2515 		davap = VOP_GETATTR(dvp, &dava, 0, cr, NULL) ? NULL : &dava;
   2516 	} else
   2517 		davap = NULL;
   2518 #else
   2519 	dava.va_mask = AT_ALL;
   2520 	davap = VOP_GETATTR(dvp, &dava, 0, cr, NULL) ? NULL : &dava;
   2521 #endif
   2522 
   2523 	/*
   2524 	 * Force modified data and metadata out to stable storage.
   2525 	 */
   2526 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
   2527 
   2528 	if (error)
   2529 		goto out;
   2530 
   2531 	resp->status = NFS3_OK;
   2532 
   2533 #ifdef DEBUG
   2534 	if (!rfs3_do_post_op_fh3)
   2535 		resp->resok.obj.handle_follows = FALSE;
   2536 	else {
   2537 #endif
   2538 	error = makefh3(&resp->resok.obj.handle, vp, exi);
   2539 	if (error)
   2540 		resp->resok.obj.handle_follows = FALSE;
   2541 	else
   2542 		resp->resok.obj.handle_follows = TRUE;
   2543 #ifdef DEBUG
   2544 	}
   2545 #endif
   2546 
   2547 #ifdef DEBUG
   2548 	if (rfs3_do_post_op_attr) {
   2549 		va.va_mask = AT_ALL;
   2550 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   2551 	} else
   2552 		vap = NULL;
   2553 #else
   2554 	va.va_mask = AT_ALL;
   2555 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   2556 #endif
   2557 
   2558 	/*
   2559 	 * Force modified metadata out to stable storage.
   2560 	 *
   2561 	 * if a underlying vp exists, pass it to VOP_FSYNC
   2562 	 */
   2563 	if (VOP_REALVP(vp, &realvp, NULL) == 0)
   2564 		(void) VOP_FSYNC(realvp, FNODSYNC, cr, NULL);
   2565 	else
   2566 		(void) VOP_FSYNC(vp, FNODSYNC, cr, NULL);
   2567 
   2568 	VN_RELE(vp);
   2569 
   2570 	vattr_to_post_op_attr(vap, &resp->resok.obj_attributes);
   2571 	vattr_to_wcc_data(dbvap, davap, &resp->resok.dir_wcc);
   2572 	DTRACE_NFSV3_4(op__mknod__done, struct svc_req *, req,
   2573 	    cred_t *, cr, vnode_t *, dvp, MKNOD3res *, resp);
   2574 	VN_RELE(dvp);
   2575 	return;
   2576 
   2577 out:
   2578 	if (curthread->t_flag & T_WOULDBLOCK) {
   2579 		curthread->t_flag &= ~T_WOULDBLOCK;
   2580 		resp->status = NFS3ERR_JUKEBOX;
   2581 	} else
   2582 		resp->status = puterrno3(error);
   2583 out1:
   2584 	DTRACE_NFSV3_4(op__mknod__done, struct svc_req *, req,
   2585 	    cred_t *, cr, vnode_t *, dvp, MKNOD3res *, resp);
   2586 	if (dvp != NULL)
   2587 		VN_RELE(dvp);
   2588 	vattr_to_wcc_data(dbvap, davap, &resp->resfail.dir_wcc);
   2589 }
   2590 
   2591 void *
   2592 rfs3_mknod_getfh(MKNOD3args *args)
   2593 {
   2594 
   2595 	return (&args->where.dir);
   2596 }
   2597 
   2598 void
   2599 rfs3_remove(REMOVE3args *args, REMOVE3res *resp, struct exportinfo *exi,
   2600 	struct svc_req *req, cred_t *cr)
   2601 {
   2602 	int error = 0;
   2603 	vnode_t *vp;
   2604 	struct vattr *bvap;
   2605 	struct vattr bva;
   2606 	struct vattr *avap;
   2607 	struct vattr ava;
   2608 	vnode_t *targvp = NULL;
   2609 	struct sockaddr *ca;
   2610 	char *name = NULL;
   2611 
   2612 	bvap = NULL;
   2613 	avap = NULL;
   2614 
   2615 	vp = nfs3_fhtovp(&args->object.dir, exi);
   2616 
   2617 	DTRACE_NFSV3_4(op__remove__start, struct svc_req *, req,
   2618 	    cred_t *, cr, vnode_t *, vp, REMOVE3args *, args);
   2619 
   2620 	if (vp == NULL) {
   2621 		error = ESTALE;
   2622 		goto err;
   2623 	}
   2624 
   2625 #ifdef DEBUG
   2626 	if (rfs3_do_pre_op_attr) {
   2627 		bva.va_mask = AT_ALL;
   2628 		bvap = VOP_GETATTR(vp, &bva, 0, cr, NULL) ? NULL : &bva;
   2629 	} else
   2630 		bvap = NULL;
   2631 #else
   2632 	bva.va_mask = AT_ALL;
   2633 	bvap = VOP_GETATTR(vp, &bva, 0, cr, NULL) ? NULL : &bva;
   2634 #endif
   2635 	avap = bvap;
   2636 
   2637 	if (vp->v_type != VDIR) {
   2638 		resp->status = NFS3ERR_NOTDIR;
   2639 		goto err1;
   2640 	}
   2641 
   2642 	if (args->object.name == nfs3nametoolong) {
   2643 		resp->status = NFS3ERR_NAMETOOLONG;
   2644 		goto err1;
   2645 	}
   2646 
   2647 	if (args->object.name == NULL || *(args->object.name) == '\0') {
   2648 		resp->status = NFS3ERR_ACCES;
   2649 		goto err1;
   2650 	}
   2651 
   2652 	if (rdonly(exi, req)) {
   2653 		resp->status = NFS3ERR_ROFS;
   2654 		goto err1;
   2655 	}
   2656 
   2657 	if (is_system_labeled()) {
   2658 		bslabel_t *clabel = req->rq_label;
   2659 
   2660 		ASSERT(clabel != NULL);
   2661 		DTRACE_PROBE2(tx__rfs3__log__info__opremove__clabel, char *,
   2662 		    "got client label from request(1)", struct svc_req *, req);
   2663 
   2664 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   2665 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
   2666 			    exi)) {
   2667 				resp->status = NFS3ERR_ACCES;
   2668 				goto err1;
   2669 			}
   2670 		}
   2671 	}
   2672 
   2673 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
   2674 	name = nfscmd_convname(ca, exi, args->object.name,
   2675 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
   2676 
   2677 	if (name == NULL) {
   2678 		resp->status = NFS3ERR_INVAL;
   2679 		goto err1;
   2680 	}
   2681 
   2682 	/*
   2683 	 * Check for a conflict with a non-blocking mandatory share
   2684 	 * reservation and V4 delegations
   2685 	 */
   2686 	error = VOP_LOOKUP(vp, name, &targvp, NULL, 0,
   2687 	    NULL, cr, NULL, NULL, NULL);
   2688 	if (error != 0)
   2689 		goto err;
   2690 
   2691 	if (rfs4_check_delegated(FWRITE, targvp, TRUE)) {
   2692 		resp->status = NFS3ERR_JUKEBOX;
   2693 		goto err1;
   2694 	}
   2695 
   2696 	if (!nbl_need_check(targvp)) {
   2697 		error = VOP_REMOVE(vp, name, cr, NULL, 0);
   2698 	} else {
   2699 		nbl_start_crit(targvp, RW_READER);
   2700 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
   2701 			error = EACCES;
   2702 		} else {
   2703 			error = VOP_REMOVE(vp, name, cr, NULL, 0);
   2704 		}
   2705 		nbl_end_crit(targvp);
   2706 	}
   2707 	VN_RELE(targvp);
   2708 	targvp = NULL;
   2709 
   2710 #ifdef DEBUG
   2711 	if (rfs3_do_post_op_attr) {
   2712 		ava.va_mask = AT_ALL;
   2713 		avap = VOP_GETATTR(vp, &ava, 0, cr, NULL) ? NULL : &ava;
   2714 	} else
   2715 		avap = NULL;
   2716 #else
   2717 	ava.va_mask = AT_ALL;
   2718 	avap = VOP_GETATTR(vp, &ava, 0, cr, NULL) ? NULL : &ava;
   2719 #endif
   2720 
   2721 	/*
   2722 	 * Force modified data and metadata out to stable storage.
   2723 	 */
   2724 	(void) VOP_FSYNC(vp, 0, cr, NULL);
   2725 
   2726 	if (error)
   2727 		goto err;
   2728 
   2729 	resp->status = NFS3_OK;
   2730 	vattr_to_wcc_data(bvap, avap, &resp->resok.dir_wcc);
   2731 	goto out;
   2732 
   2733 err:
   2734 	if (curthread->t_flag & T_WOULDBLOCK) {
   2735 		curthread->t_flag &= ~T_WOULDBLOCK;
   2736 		resp->status = NFS3ERR_JUKEBOX;
   2737 	} else
   2738 		resp->status = puterrno3(error);
   2739 err1:
   2740 	vattr_to_wcc_data(bvap, avap, &resp->resfail.dir_wcc);
   2741 out:
   2742 	DTRACE_NFSV3_4(op__remove__done, struct svc_req *, req,
   2743 	    cred_t *, cr, vnode_t *, vp, REMOVE3res *, resp);
   2744 
   2745 	if (name != NULL && name != args->object.name)
   2746 		kmem_free(name, MAXPATHLEN + 1);
   2747 
   2748 	if (vp != NULL)
   2749 		VN_RELE(vp);
   2750 }
   2751 
   2752 void *
   2753 rfs3_remove_getfh(REMOVE3args *args)
   2754 {
   2755 
   2756 	return (&args->object.dir);
   2757 }
   2758 
   2759 void
   2760 rfs3_rmdir(RMDIR3args *args, RMDIR3res *resp, struct exportinfo *exi,
   2761 	struct svc_req *req, cred_t *cr)
   2762 {
   2763 	int error;
   2764 	vnode_t *vp;
   2765 	struct vattr *bvap;
   2766 	struct vattr bva;
   2767 	struct vattr *avap;
   2768 	struct vattr ava;
   2769 	struct sockaddr *ca;
   2770 	char *name = NULL;
   2771 
   2772 	bvap = NULL;
   2773 	avap = NULL;
   2774 
   2775 	vp = nfs3_fhtovp(&args->object.dir, exi);
   2776 
   2777 	DTRACE_NFSV3_4(op__rmdir__start, struct svc_req *, req,
   2778 	    cred_t *, cr, vnode_t *, vp, RMDIR3args *, args);
   2779 
   2780 	if (vp == NULL) {
   2781 		error = ESTALE;
   2782 		goto err;
   2783 	}
   2784 
   2785 #ifdef DEBUG
   2786 	if (rfs3_do_pre_op_attr) {
   2787 		bva.va_mask = AT_ALL;
   2788 		bvap = VOP_GETATTR(vp, &bva, 0, cr, NULL) ? NULL : &bva;
   2789 	} else
   2790 		bvap = NULL;
   2791 #else
   2792 	bva.va_mask = AT_ALL;
   2793 	bvap = VOP_GETATTR(vp, &bva, 0, cr, NULL) ? NULL : &bva;
   2794 #endif
   2795 	avap = bvap;
   2796 
   2797 	if (vp->v_type != VDIR) {
   2798 		resp->status = NFS3ERR_NOTDIR;
   2799 		goto err1;
   2800 	}
   2801 
   2802 	if (args->object.name == nfs3nametoolong) {
   2803 		resp->status = NFS3ERR_NAMETOOLONG;
   2804 		goto err1;
   2805 	}
   2806 
   2807 	if (args->object.name == NULL || *(args->object.name) == '\0') {
   2808 		resp->status = NFS3ERR_ACCES;
   2809 		goto err1;
   2810 	}
   2811 
   2812 	if (rdonly(exi, req)) {
   2813 		resp->status = NFS3ERR_ROFS;
   2814 		goto err1;
   2815 	}
   2816 
   2817 	if (is_system_labeled()) {
   2818 		bslabel_t *clabel = req->rq_label;
   2819 
   2820 		ASSERT(clabel != NULL);
   2821 		DTRACE_PROBE2(tx__rfs3__log__info__opremovedir__clabel, char *,
   2822 		    "got client label from request(1)", struct svc_req *, req);
   2823 
   2824 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   2825 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
   2826 			    exi)) {
   2827 				resp->status = NFS3ERR_ACCES;
   2828 				goto err1;
   2829 			}
   2830 		}
   2831 	}
   2832 
   2833 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
   2834 	name = nfscmd_convname(ca, exi, args->object.name,
   2835 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
   2836 
   2837 	if (name == NULL) {
   2838 		resp->status = NFS3ERR_INVAL;
   2839 		goto err1;
   2840 	}
   2841 
   2842 	error = VOP_RMDIR(vp, name, rootdir, cr, NULL, 0);
   2843 
   2844 	if (name != args->object.name)
   2845 		kmem_free(name, MAXPATHLEN + 1);
   2846 
   2847 #ifdef DEBUG
   2848 	if (rfs3_do_post_op_attr) {
   2849 		ava.va_mask = AT_ALL;
   2850 		avap = VOP_GETATTR(vp, &ava, 0, cr, NULL) ? NULL : &ava;
   2851 	} else
   2852 		avap = NULL;
   2853 #else
   2854 	ava.va_mask = AT_ALL;
   2855 	avap = VOP_GETATTR(vp, &ava, 0, cr, NULL) ? NULL : &ava;
   2856 #endif
   2857 
   2858 	/*
   2859 	 * Force modified data and metadata out to stable storage.
   2860 	 */
   2861 	(void) VOP_FSYNC(vp, 0, cr, NULL);
   2862 
   2863 	if (error) {
   2864 		/*
   2865 		 * System V defines rmdir to return EEXIST, not ENOTEMPTY,
   2866 		 * if the directory is not empty.  A System V NFS server
   2867 		 * needs to map NFS3ERR_EXIST to NFS3ERR_NOTEMPTY to transmit
   2868 		 * over the wire.
   2869 		 */
   2870 		if (error == EEXIST)
   2871 			error = ENOTEMPTY;
   2872 		goto err;
   2873 	}
   2874 
   2875 	resp->status = NFS3_OK;
   2876 	vattr_to_wcc_data(bvap, avap, &resp->resok.dir_wcc);
   2877 	goto out;
   2878 
   2879 err:
   2880 	if (curthread->t_flag & T_WOULDBLOCK) {
   2881 		curthread->t_flag &= ~T_WOULDBLOCK;
   2882 		resp->status = NFS3ERR_JUKEBOX;
   2883 	} else
   2884 		resp->status = puterrno3(error);
   2885 err1:
   2886 	vattr_to_wcc_data(bvap, avap, &resp->resfail.dir_wcc);
   2887 out:
   2888 	DTRACE_NFSV3_4(op__rmdir__done, struct svc_req *, req,
   2889 	    cred_t *, cr, vnode_t *, vp, RMDIR3res *, resp);
   2890 	if (vp != NULL)
   2891 		VN_RELE(vp);
   2892 
   2893 }
   2894 
   2895 void *
   2896 rfs3_rmdir_getfh(RMDIR3args *args)
   2897 {
   2898 
   2899 	return (&args->object.dir);
   2900 }
   2901 
   2902 void
   2903 rfs3_rename(RENAME3args *args, RENAME3res *resp, struct exportinfo *exi,
   2904 	struct svc_req *req, cred_t *cr)
   2905 {
   2906 	int error = 0;
   2907 	vnode_t *fvp;
   2908 	vnode_t *tvp;
   2909 	vnode_t *targvp;
   2910 	struct vattr *fbvap;
   2911 	struct vattr fbva;
   2912 	struct vattr *favap;
   2913 	struct vattr fava;
   2914 	struct vattr *tbvap;
   2915 	struct vattr tbva;
   2916 	struct vattr *tavap;
   2917 	struct vattr tava;
   2918 	nfs_fh3 *fh3;
   2919 	struct exportinfo *to_exi;
   2920 	vnode_t *srcvp = NULL;
   2921 	bslabel_t *clabel;
   2922 	struct sockaddr *ca;
   2923 	char *name = NULL;
   2924 	char *toname = NULL;
   2925 
   2926 	fbvap = NULL;
   2927 	favap = NULL;
   2928 	tbvap = NULL;
   2929 	tavap = NULL;
   2930 	tvp = NULL;
   2931 
   2932 	fvp = nfs3_fhtovp(&args->from.dir, exi);
   2933 
   2934 	DTRACE_NFSV3_4(op__rename__start, struct svc_req *, req,
   2935 	    cred_t *, cr, vnode_t *, fvp, RENAME3args *, args);
   2936 
   2937 	if (fvp == NULL) {
   2938 		error = ESTALE;
   2939 		goto err;
   2940 	}
   2941 
   2942 	if (is_system_labeled()) {
   2943 		clabel = req->rq_label;
   2944 		ASSERT(clabel != NULL);
   2945 		DTRACE_PROBE2(tx__rfs3__log__info__oprename__clabel, char *,
   2946 		    "got client label from request(1)", struct svc_req *, req);
   2947 
   2948 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   2949 			if (!do_rfs_label_check(clabel, fvp, EQUALITY_CHECK,
   2950 			    exi)) {
   2951 				resp->status = NFS3ERR_ACCES;
   2952 				goto err1;
   2953 			}
   2954 		}
   2955 	}
   2956 
   2957 #ifdef DEBUG
   2958 	if (rfs3_do_pre_op_attr) {
   2959 		fbva.va_mask = AT_ALL;
   2960 		fbvap = VOP_GETATTR(fvp, &fbva, 0, cr, NULL) ? NULL : &fbva;
   2961 	} else
   2962 		fbvap = NULL;
   2963 #else
   2964 	fbva.va_mask = AT_ALL;
   2965 	fbvap = VOP_GETATTR(fvp, &fbva, 0, cr, NULL) ? NULL : &fbva;
   2966 #endif
   2967 	favap = fbvap;
   2968 
   2969 	fh3 = &args->to.dir;
   2970 	to_exi = checkexport(&fh3->fh3_fsid, FH3TOXFIDP(fh3));
   2971 	if (to_exi == NULL) {
   2972 		resp->status = NFS3ERR_ACCES;
   2973 		goto err1;
   2974 	}
   2975 	exi_rele(to_exi);
   2976 
   2977 	if (to_exi != exi) {
   2978 		resp->status = NFS3ERR_XDEV;
   2979 		goto err1;
   2980 	}
   2981 
   2982 	tvp = nfs3_fhtovp(&args->to.dir, exi);
   2983 	if (tvp == NULL) {
   2984 		error = ESTALE;
   2985 		goto err;
   2986 	}
   2987 
   2988 #ifdef DEBUG
   2989 	if (rfs3_do_pre_op_attr) {
   2990 		tbva.va_mask = AT_ALL;
   2991 		tbvap = VOP_GETATTR(tvp, &tbva, 0, cr, NULL) ? NULL : &tbva;
   2992 	} else
   2993 		tbvap = NULL;
   2994 #else
   2995 	tbva.va_mask = AT_ALL;
   2996 	tbvap = VOP_GETATTR(tvp, &tbva, 0, cr, NULL) ? NULL : &tbva;
   2997 #endif
   2998 	tavap = tbvap;
   2999 
   3000 	if (fvp->v_type != VDIR || tvp->v_type != VDIR) {
   3001 		resp->status = NFS3ERR_NOTDIR;
   3002 		goto err1;
   3003 	}
   3004 
   3005 	if (args->from.name == nfs3nametoolong ||
   3006 	    args->to.name == nfs3nametoolong) {
   3007 		resp->status = NFS3ERR_NAMETOOLONG;
   3008 		goto err1;
   3009 	}
   3010 	if (args->from.name == NULL || *(args->from.name) == '\0' ||
   3011 	    args->to.name == NULL || *(args->to.name) == '\0') {
   3012 		resp->status = NFS3ERR_ACCES;
   3013 		goto err1;
   3014 	}
   3015 
   3016 	if (rdonly(exi, req)) {
   3017 		resp->status = NFS3ERR_ROFS;
   3018 		goto err1;
   3019 	}
   3020 
   3021 	if (is_system_labeled()) {
   3022 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   3023 			if (!do_rfs_label_check(clabel, tvp, EQUALITY_CHECK,
   3024 			    exi)) {
   3025 				resp->status = NFS3ERR_ACCES;
   3026 				goto err1;
   3027 			}
   3028 		}
   3029 	}
   3030 
   3031 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
   3032 	name = nfscmd_convname(ca, exi, args->from.name,
   3033 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
   3034 
   3035 	if (name == NULL) {
   3036 		resp->status = NFS3ERR_INVAL;
   3037 		goto err1;
   3038 	}
   3039 
   3040 	toname = nfscmd_convname(ca, exi, args->to.name,
   3041 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
   3042 
   3043 	if (toname == NULL) {
   3044 		resp->status = NFS3ERR_INVAL;
   3045 		goto err1;
   3046 	}
   3047 
   3048 	/*
   3049 	 * Check for a conflict with a non-blocking mandatory share
   3050 	 * reservation or V4 delegations.
   3051 	 */
   3052 	error = VOP_LOOKUP(fvp, name, &srcvp, NULL, 0,
   3053 	    NULL, cr, NULL, NULL, NULL);
   3054 	if (error != 0)
   3055 		goto err;
   3056 
   3057 	/*
   3058 	 * If we rename a delegated file we should recall the
   3059 	 * delegation, since future opens should fail or would
   3060 	 * refer to a new file.
   3061 	 */
   3062 	if (rfs4_check_delegated(FWRITE, srcvp, FALSE)) {
   3063 		resp->status = NFS3ERR_JUKEBOX;
   3064 		goto err1;
   3065 	}
   3066 
   3067 	/*
   3068 	 * Check for renaming over a delegated file.  Check rfs4_deleg_policy
   3069 	 * first to avoid VOP_LOOKUP if possible.
   3070 	 */
   3071 	if (rfs4_deleg_policy != SRV_NEVER_DELEGATE &&
   3072 	    VOP_LOOKUP(tvp, toname, &targvp, NULL, 0, NULL, cr,
   3073 	    NULL, NULL, NULL) == 0) {
   3074 
   3075 		if (rfs4_check_delegated(FWRITE, targvp, TRUE)) {
   3076 			VN_RELE(targvp);
   3077 			resp->status = NFS3ERR_JUKEBOX;
   3078 			goto err1;
   3079 		}
   3080 		VN_RELE(targvp);
   3081 	}
   3082 
   3083 	if (!nbl_need_check(srcvp)) {
   3084 		error = VOP_RENAME(fvp, name, tvp, toname, cr, NULL, 0);
   3085 	} else {
   3086 		nbl_start_crit(srcvp, RW_READER);
   3087 		if (nbl_conflict(srcvp, NBL_RENAME, 0, 0, 0, NULL))
   3088 			error = EACCES;
   3089 		else
   3090 			error = VOP_RENAME(fvp, name, tvp, toname, cr, NULL, 0);
   3091 		nbl_end_crit(srcvp);
   3092 	}
   3093 	if (error == 0)
   3094 		vn_renamepath(tvp, srcvp, args->to.name,
   3095 		    strlen(args->to.name));
   3096 	VN_RELE(srcvp);
   3097 	srcvp = NULL;
   3098 
   3099 #ifdef DEBUG
   3100 	if (rfs3_do_post_op_attr) {
   3101 		fava.va_mask = AT_ALL;
   3102 		favap = VOP_GETATTR(fvp, &fava, 0, cr, NULL) ? NULL : &fava;
   3103 		tava.va_mask = AT_ALL;
   3104 		tavap = VOP_GETATTR(tvp, &tava, 0, cr, NULL) ? NULL : &tava;
   3105 	} else {
   3106 		favap = NULL;
   3107 		tavap = NULL;
   3108 	}
   3109 #else
   3110 	fava.va_mask = AT_ALL;
   3111 	favap = VOP_GETATTR(fvp, &fava, 0, cr, NULL) ? NULL : &fava;
   3112 	tava.va_mask = AT_ALL;
   3113 	tavap = VOP_GETATTR(tvp, &tava, 0, cr, NULL) ? NULL : &tava;
   3114 #endif
   3115 
   3116 	/*
   3117 	 * Force modified data and metadata out to stable storage.
   3118 	 */
   3119 	(void) VOP_FSYNC(fvp, 0, cr, NULL);
   3120 	(void) VOP_FSYNC(tvp, 0, cr, NULL);
   3121 
   3122 	if (error)
   3123 		goto err;
   3124 
   3125 	resp->status = NFS3_OK;
   3126 	vattr_to_wcc_data(fbvap, favap, &resp->resok.fromdir_wcc);
   3127 	vattr_to_wcc_data(tbvap, tavap, &resp->resok.todir_wcc);
   3128 	goto out;
   3129 
   3130 err:
   3131 	if (curthread->t_flag & T_WOULDBLOCK) {
   3132 		curthread->t_flag &= ~T_WOULDBLOCK;
   3133 		resp->status = NFS3ERR_JUKEBOX;
   3134 	} else {
   3135 		resp->status = puterrno3(error);
   3136 	}
   3137 err1:
   3138 	vattr_to_wcc_data(fbvap, favap, &resp->resfail.fromdir_wcc);
   3139 	vattr_to_wcc_data(tbvap, tavap, &resp->resfail.todir_wcc);
   3140 
   3141 out:
   3142 	if (name != NULL && name != args->from.name)
   3143 		kmem_free(name, MAXPATHLEN + 1);
   3144 	if (toname != NULL && toname != args->to.name)
   3145 		kmem_free(toname, MAXPATHLEN + 1);
   3146 
   3147 	DTRACE_NFSV3_4(op__rename__done, struct svc_req *, req,
   3148 	    cred_t *, cr, vnode_t *, fvp, RENAME3res *, resp);
   3149 	if (fvp != NULL)
   3150 		VN_RELE(fvp);
   3151 	if (tvp != NULL)
   3152 		VN_RELE(tvp);
   3153 }
   3154 
   3155 void *
   3156 rfs3_rename_getfh(RENAME3args *args)
   3157 {
   3158 
   3159 	return (&args->from.dir);
   3160 }
   3161 
   3162 void
   3163 rfs3_link(LINK3args *args, LINK3res *resp, struct exportinfo *exi,
   3164 	struct svc_req *req, cred_t *cr)
   3165 {
   3166 	int error;
   3167 	vnode_t *vp;
   3168 	vnode_t *dvp;
   3169 	struct vattr *vap;
   3170 	struct vattr va;
   3171 	struct vattr *bvap;
   3172 	struct vattr bva;
   3173 	struct vattr *avap;
   3174 	struct vattr ava;
   3175 	nfs_fh3	*fh3;
   3176 	struct exportinfo *to_exi;
   3177 	bslabel_t *clabel;
   3178 	struct sockaddr *ca;
   3179 	char *name = NULL;
   3180 
   3181 	vap = NULL;
   3182 	bvap = NULL;
   3183 	avap = NULL;
   3184 	dvp = NULL;
   3185 
   3186 	vp = nfs3_fhtovp(&args->file, exi);
   3187 
   3188 	DTRACE_NFSV3_4(op__link__start, struct svc_req *, req,
   3189 	    cred_t *, cr, vnode_t *, vp, LINK3args *, args);
   3190 
   3191 	if (vp == NULL) {
   3192 		error = ESTALE;
   3193 		goto out;
   3194 	}
   3195 
   3196 #ifdef DEBUG
   3197 	if (rfs3_do_pre_op_attr) {
   3198 		va.va_mask = AT_ALL;
   3199 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3200 	} else
   3201 		vap = NULL;
   3202 #else
   3203 	va.va_mask = AT_ALL;
   3204 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3205 #endif
   3206 
   3207 	fh3 = &args->link.dir;
   3208 	to_exi = checkexport(&fh3->fh3_fsid, FH3TOXFIDP(fh3));
   3209 	if (to_exi == NULL) {
   3210 		resp->status = NFS3ERR_ACCES;
   3211 		goto out1;
   3212 	}
   3213 	exi_rele(to_exi);
   3214 
   3215 	if (to_exi != exi) {
   3216 		resp->status = NFS3ERR_XDEV;
   3217 		goto out1;
   3218 	}
   3219 
   3220 	if (is_system_labeled()) {
   3221 		clabel = req->rq_label;
   3222 
   3223 		ASSERT(clabel != NULL);
   3224 		DTRACE_PROBE2(tx__rfs3__log__info__oplink__clabel, char *,
   3225 		    "got client label from request(1)", struct svc_req *, req);
   3226 
   3227 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   3228 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
   3229 			    exi)) {
   3230 				resp->status = NFS3ERR_ACCES;
   3231 				goto out1;
   3232 			}
   3233 		}
   3234 	}
   3235 
   3236 	dvp = nfs3_fhtovp(&args->link.dir, exi);
   3237 	if (dvp == NULL) {
   3238 		error = ESTALE;
   3239 		goto out;
   3240 	}
   3241 
   3242 #ifdef DEBUG
   3243 	if (rfs3_do_pre_op_attr) {
   3244 		bva.va_mask = AT_ALL;
   3245 		bvap = VOP_GETATTR(dvp, &bva, 0, cr, NULL) ? NULL : &bva;
   3246 	} else
   3247 		bvap = NULL;
   3248 #else
   3249 	bva.va_mask = AT_ALL;
   3250 	bvap = VOP_GETATTR(dvp, &bva, 0, cr, NULL) ? NULL : &bva;
   3251 #endif
   3252 
   3253 	if (dvp->v_type != VDIR) {
   3254 		resp->status = NFS3ERR_NOTDIR;
   3255 		goto out1;
   3256 	}
   3257 
   3258 	if (args->link.name == nfs3nametoolong) {
   3259 		resp->status = NFS3ERR_NAMETOOLONG;
   3260 		goto out1;
   3261 	}
   3262 
   3263 	if (args->link.name == NULL || *(args->link.name) == '\0') {
   3264 		resp->status = NFS3ERR_ACCES;
   3265 		goto out1;
   3266 	}
   3267 
   3268 	if (rdonly(exi, req)) {
   3269 		resp->status = NFS3ERR_ROFS;
   3270 		goto out1;
   3271 	}
   3272 
   3273 	if (is_system_labeled()) {
   3274 		DTRACE_PROBE2(tx__rfs3__log__info__oplinkdir__clabel, char *,
   3275 		    "got client label from request(1)", struct svc_req *, req);
   3276 
   3277 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   3278 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
   3279 			    exi)) {
   3280 				resp->status = NFS3ERR_ACCES;
   3281 				goto out1;
   3282 			}
   3283 		}
   3284 	}
   3285 
   3286 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
   3287 	name = nfscmd_convname(ca, exi, args->link.name,
   3288 	    NFSCMD_CONV_INBOUND, MAXPATHLEN + 1);
   3289 
   3290 	if (name == NULL) {
   3291 		resp->status = NFS3ERR_SERVERFAULT;
   3292 		goto out1;
   3293 	}
   3294 
   3295 	error = VOP_LINK(dvp, vp, name, cr, NULL, 0);
   3296 
   3297 #ifdef DEBUG
   3298 	if (rfs3_do_post_op_attr) {
   3299 		va.va_mask = AT_ALL;
   3300 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3301 		ava.va_mask = AT_ALL;
   3302 		avap = VOP_GETATTR(dvp, &ava, 0, cr, NULL) ? NULL : &ava;
   3303 	} else {
   3304 		vap = NULL;
   3305 		avap = NULL;
   3306 	}
   3307 #else
   3308 	va.va_mask = AT_ALL;
   3309 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3310 	ava.va_mask = AT_ALL;
   3311 	avap = VOP_GETATTR(dvp, &ava, 0, cr, NULL) ? NULL : &ava;
   3312 #endif
   3313 
   3314 	/*
   3315 	 * Force modified data and metadata out to stable storage.
   3316 	 */
   3317 	(void) VOP_FSYNC(vp, FNODSYNC, cr, NULL);
   3318 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
   3319 
   3320 	if (error)
   3321 		goto out;
   3322 
   3323 	VN_RELE(dvp);
   3324 
   3325 	resp->status = NFS3_OK;
   3326 	vattr_to_post_op_attr(vap, &resp->resok.file_attributes);
   3327 	vattr_to_wcc_data(bvap, avap, &resp->resok.linkdir_wcc);
   3328 
   3329 	DTRACE_NFSV3_4(op__link__done, struct svc_req *, req,
   3330 	    cred_t *, cr, vnode_t *, vp, LINK3res *, resp);
   3331 
   3332 	VN_RELE(vp);
   3333 
   3334 	return;
   3335 
   3336 out:
   3337 	if (curthread->t_flag & T_WOULDBLOCK) {
   3338 		curthread->t_flag &= ~T_WOULDBLOCK;
   3339 		resp->status = NFS3ERR_JUKEBOX;
   3340 	} else
   3341 		resp->status = puterrno3(error);
   3342 out1:
   3343 	if (name != NULL && name != args->link.name)
   3344 		kmem_free(name, MAXPATHLEN + 1);
   3345 
   3346 	DTRACE_NFSV3_4(op__link__done, struct svc_req *, req,
   3347 	    cred_t *, cr, vnode_t *, vp, LINK3res *, resp);
   3348 
   3349 	if (vp != NULL)
   3350 		VN_RELE(vp);
   3351 	if (dvp != NULL)
   3352 		VN_RELE(dvp);
   3353 	vattr_to_post_op_attr(vap, &resp->resfail.file_attributes);
   3354 	vattr_to_wcc_data(bvap, avap, &resp->resfail.linkdir_wcc);
   3355 }
   3356 
   3357 void *
   3358 rfs3_link_getfh(LINK3args *args)
   3359 {
   3360 
   3361 	return (&args->file);
   3362 }
   3363 
   3364 /*
   3365  * This macro defines the size of a response which contains attribute
   3366  * information and one directory entry (whose length is specified by
   3367  * the macro parameter).  If the incoming request is larger than this,
   3368  * then we are guaranteed to be able to return at one directory entry
   3369  * if one exists.  Therefore, we do not need to check for
   3370  * NFS3ERR_TOOSMALL if the requested size is larger then this.  If it
   3371  * is not, then we need to check to make sure that this error does not
   3372  * need to be returned.
   3373  *
   3374  * NFS3_READDIR_MIN_COUNT is comprised of following :
   3375  *
   3376  * status - 1 * BYTES_PER_XDR_UNIT
   3377  * attr. flag - 1 * BYTES_PER_XDR_UNIT
   3378  * cookie verifier - 2 * BYTES_PER_XDR_UNIT
   3379  * attributes  - NFS3_SIZEOF_FATTR3 * BYTES_PER_XDR_UNIT
   3380  * boolean - 1 * BYTES_PER_XDR_UNIT
   3381  * file id - 2 * BYTES_PER_XDR_UNIT
   3382  * directory name length - 1 * BYTES_PER_XDR_UNIT
   3383  * cookie - 2 * BYTES_PER_XDR_UNIT
   3384  * end of list - 1 * BYTES_PER_XDR_UNIT
   3385  * end of file - 1 * BYTES_PER_XDR_UNIT
   3386  * Name length of directory to the nearest byte
   3387  */
   3388 
   3389 #define	NFS3_READDIR_MIN_COUNT(length)	\
   3390 	((1 + 1 + 2 + NFS3_SIZEOF_FATTR3 + 1 + 2 + 1 + 2 + 1 + 1) * \
   3391 		BYTES_PER_XDR_UNIT + roundup((length), BYTES_PER_XDR_UNIT))
   3392 
   3393 /* ARGSUSED */
   3394 void
   3395 rfs3_readdir(READDIR3args *args, READDIR3res *resp, struct exportinfo *exi,
   3396 	struct svc_req *req, cred_t *cr)
   3397 {
   3398 	int error;
   3399 	vnode_t *vp;
   3400 	struct vattr *vap;
   3401 	struct vattr va;
   3402 	struct iovec iov;
   3403 	struct uio uio;
   3404 	char *data;
   3405 	int iseof;
   3406 	int bufsize;
   3407 	int namlen;
   3408 	uint_t count;
   3409 	struct sockaddr *ca;
   3410 
   3411 	vap = NULL;
   3412 
   3413 	vp = nfs3_fhtovp(&args->dir, exi);
   3414 
   3415 	DTRACE_NFSV3_4(op__readdir__start, struct svc_req *, req,
   3416 	    cred_t *, cr, vnode_t *, vp, READDIR3args *, args);
   3417 
   3418 	if (vp == NULL) {
   3419 		error = ESTALE;
   3420 		goto out;
   3421 	}
   3422 
   3423 	if (is_system_labeled()) {
   3424 		bslabel_t *clabel = req->rq_label;
   3425 
   3426 		ASSERT(clabel != NULL);
   3427 		DTRACE_PROBE2(tx__rfs3__log__info__opreaddir__clabel, char *,
   3428 		    "got client label from request(1)", struct svc_req *, req);
   3429 
   3430 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   3431 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
   3432 			    exi)) {
   3433 				resp->status = NFS3ERR_ACCES;
   3434 				goto out1;
   3435 			}
   3436 		}
   3437 	}
   3438 
   3439 	(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
   3440 
   3441 #ifdef DEBUG
   3442 	if (rfs3_do_pre_op_attr) {
   3443 		va.va_mask = AT_ALL;
   3444 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3445 	} else
   3446 		vap = NULL;
   3447 #else
   3448 	va.va_mask = AT_ALL;
   3449 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3450 #endif
   3451 
   3452 	if (vp->v_type != VDIR) {
   3453 		resp->status = NFS3ERR_NOTDIR;
   3454 		goto out1;
   3455 	}
   3456 
   3457 	error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
   3458 	if (error)
   3459 		goto out;
   3460 
   3461 	/*
   3462 	 * Now don't allow arbitrary count to alloc;
   3463 	 * allow the maximum not to exceed rfs3_tsize()
   3464 	 */
   3465 	if (args->count > rfs3_tsize(req))
   3466 		args->count = rfs3_tsize(req);
   3467 
   3468 	/*
   3469 	 * Make sure that there is room to read at least one entry
   3470 	 * if any are available.
   3471 	 */
   3472 	if (args->count < DIRENT64_RECLEN(MAXNAMELEN))
   3473 		count = DIRENT64_RECLEN(MAXNAMELEN);
   3474 	else
   3475 		count = args->count;
   3476 
   3477 	data = kmem_alloc(count, KM_SLEEP);
   3478 
   3479 	iov.iov_base = data;
   3480 	iov.iov_len = count;
   3481 	uio.uio_iov = &iov;
   3482 	uio.uio_iovcnt = 1;
   3483 	uio.uio_segflg = UIO_SYSSPACE;
   3484 	uio.uio_extflg = UIO_COPY_CACHED;
   3485 	uio.uio_loffset = (offset_t)args->cookie;
   3486 	uio.uio_resid = count;
   3487 
   3488 	error = VOP_READDIR(vp, &uio, cr, &iseof, NULL, 0);
   3489 
   3490 #ifdef DEBUG
   3491 	if (rfs3_do_post_op_attr) {
   3492 		va.va_mask = AT_ALL;
   3493 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3494 	} else
   3495 		vap = NULL;
   3496 #else
   3497 	va.va_mask = AT_ALL;
   3498 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3499 #endif
   3500 
   3501 	if (error) {
   3502 		kmem_free(data, count);
   3503 		goto out;
   3504 	}
   3505 
   3506 	/*
   3507 	 * If the count was not large enough to be able to guarantee
   3508 	 * to be able to return at least one entry, then need to
   3509 	 * check to see if NFS3ERR_TOOSMALL should be returned.
   3510 	 */
   3511 	if (args->count < NFS3_READDIR_MIN_COUNT(MAXNAMELEN)) {
   3512 		/*
   3513 		 * bufsize is used to keep track of the size of the response.
   3514 		 * It is primed with:
   3515 		 *	1 for the status +
   3516 		 *	1 for the dir_attributes.attributes boolean +
   3517 		 *	2 for the cookie verifier
   3518 		 * all times BYTES_PER_XDR_UNIT to convert from XDR units
   3519 		 * to bytes.  If there are directory attributes to be
   3520 		 * returned, then:
   3521 		 *	NFS3_SIZEOF_FATTR3 for the dir_attributes.attr fattr3
   3522 		 * time BYTES_PER_XDR_UNIT is added to account for them.
   3523 		 */
   3524 		bufsize = (1 + 1 + 2) * BYTES_PER_XDR_UNIT;
   3525 		if (vap != NULL)
   3526 			bufsize += NFS3_SIZEOF_FATTR3 * BYTES_PER_XDR_UNIT;
   3527 		/*
   3528 		 * An entry is composed of:
   3529 		 *	1 for the true/false list indicator +
   3530 		 *	2 for the fileid +
   3531 		 *	1 for the length of the name +
   3532 		 *	2 for the cookie +
   3533 		 * all times BYTES_PER_XDR_UNIT to convert from
   3534 		 * XDR units to bytes, plus the length of the name
   3535 		 * rounded up to the nearest BYTES_PER_XDR_UNIT.
   3536 		 */
   3537 		if (count != uio.uio_resid) {
   3538 			namlen = strlen(((struct dirent64 *)data)->d_name);
   3539 			bufsize += (1 + 2 + 1 + 2) * BYTES_PER_XDR_UNIT +
   3540 			    roundup(namlen, BYTES_PER_XDR_UNIT);
   3541 		}
   3542 		/*
   3543 		 * We need to check to see if the number of bytes left
   3544 		 * to go into the buffer will actually fit into the
   3545 		 * buffer.  This is calculated as the size of this
   3546 		 * entry plus:
   3547 		 *	1 for the true/false list indicator +
   3548 		 *	1 for the eof indicator
   3549 		 * times BYTES_PER_XDR_UNIT to convert from from
   3550 		 * XDR units to bytes.
   3551 		 */
   3552 		bufsize += (1 + 1) * BYTES_PER_XDR_UNIT;
   3553 		if (bufsize > args->count) {
   3554 			kmem_free(data, count);
   3555 			resp->status = NFS3ERR_TOOSMALL;
   3556 			goto out1;
   3557 		}
   3558 	}
   3559 
   3560 	/*
   3561 	 * Have a valid readir buffer for the native character
   3562 	 * set. Need to check if a conversion is necessary and
   3563 	 * potentially rewrite the whole buffer. Note that if the
   3564 	 * conversion expands names enough, the structure may not
   3565 	 * fit. In this case, we need to drop entries until if fits
   3566 	 * and patch the counts in order that the next readdir will
   3567 	 * get the correct entries.
   3568 	 */
   3569 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
   3570 	data = nfscmd_convdirent(ca, exi, data, count, &resp->status);
   3571 
   3572 
   3573 	VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
   3574 
   3575 #if 0 /* notyet */
   3576 	/*
   3577 	 * Don't do this.  It causes local disk writes when just
   3578 	 * reading the file and the overhead is deemed larger
   3579 	 * than the benefit.
   3580 	 */
   3581 	/*
   3582 	 * Force modified metadata out to stable storage.
   3583 	 */
   3584 	(void) VOP_FSYNC(vp, FNODSYNC, cr, NULL);
   3585 #endif
   3586 
   3587 	resp->status = NFS3_OK;
   3588 	vattr_to_post_op_attr(vap, &resp->resok.dir_attributes);
   3589 	resp->resok.cookieverf = 0;
   3590 	resp->resok.reply.entries = (entry3 *)data;
   3591 	resp->resok.reply.eof = iseof;
   3592 	resp->resok.size = count - uio.uio_resid;
   3593 	resp->resok.count = args->count;
   3594 	resp->resok.freecount = count;
   3595 
   3596 	DTRACE_NFSV3_4(op__readdir__done, struct svc_req *, req,
   3597 	    cred_t *, cr, vnode_t *, vp, READDIR3res *, resp);
   3598 
   3599 	VN_RELE(vp);
   3600 
   3601 	return;
   3602 
   3603 out:
   3604 	if (curthread->t_flag & T_WOULDBLOCK) {
   3605 		curthread->t_flag &= ~T_WOULDBLOCK;
   3606 		resp->status = NFS3ERR_JUKEBOX;
   3607 	} else
   3608 		resp->status = puterrno3(error);
   3609 out1:
   3610 	DTRACE_NFSV3_4(op__readdir__done, struct svc_req *, req,
   3611 	    cred_t *, cr, vnode_t *, vp, READDIR3res *, resp);
   3612 
   3613 	if (vp != NULL) {
   3614 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
   3615 		VN_RELE(vp);
   3616 	}
   3617 	vattr_to_post_op_attr(vap, &resp->resfail.dir_attributes);
   3618 }
   3619 
   3620 void *
   3621 rfs3_readdir_getfh(READDIR3args *args)
   3622 {
   3623 
   3624 	return (&args->dir);
   3625 }
   3626 
   3627 void
   3628 rfs3_readdir_free(READDIR3res *resp)
   3629 {
   3630 
   3631 	if (resp->status == NFS3_OK)
   3632 		kmem_free(resp->resok.reply.entries, resp->resok.freecount);
   3633 }
   3634 
   3635 #ifdef nextdp
   3636 #undef nextdp
   3637 #endif
   3638 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
   3639 
   3640 /*
   3641  * This macro computes the size of a response which contains
   3642  * one directory entry including the attributes as well as file handle.
   3643  * If the incoming request is larger than this, then we are guaranteed to be
   3644  * able to return at least one more directory entry if one exists.
   3645  *
   3646  * NFS3_READDIRPLUS_ENTRY is made up of the following:
   3647  *
   3648  * boolean - 1 * BYTES_PER_XDR_UNIT
   3649  * file id - 2 * BYTES_PER_XDR_UNIT
   3650  * directory name length - 1 * BYTES_PER_XDR_UNIT
   3651  * cookie - 2 * BYTES_PER_XDR_UNIT
   3652  * attribute flag - 1 * BYTES_PER_XDR_UNIT
   3653  * attributes - NFS3_SIZEOF_FATTR3 * BYTES_PER_XDR_UNIT
   3654  * status byte for file handle - 1 *  BYTES_PER_XDR_UNIT
   3655  * length of a file handle - 1 * BYTES_PER_XDR_UNIT
   3656  * Maximum length of a file handle (NFS3_MAXFHSIZE)
   3657  * name length of the entry to the nearest bytes
   3658  */
   3659 #define	NFS3_READDIRPLUS_ENTRY(namelen)	\
   3660 	((1 + 2 + 1 + 2 + 1 + NFS3_SIZEOF_FATTR3 + 1 + 1) * \
   3661 		BYTES_PER_XDR_UNIT + \
   3662 	NFS3_MAXFHSIZE + roundup(namelen, BYTES_PER_XDR_UNIT))
   3663 
   3664 static int rfs3_readdir_unit = MAXBSIZE;
   3665 
   3666 /* ARGSUSED */
   3667 void
   3668 rfs3_readdirplus(READDIRPLUS3args *args, READDIRPLUS3res *resp,
   3669 	struct exportinfo *exi, struct svc_req *req, cred_t *cr)
   3670 {
   3671 	int error;
   3672 	vnode_t *vp;
   3673 	struct vattr *vap;
   3674 	struct vattr va;
   3675 	struct iovec iov;
   3676 	struct uio uio;
   3677 	char *data;
   3678 	int iseof;
   3679 	struct dirent64 *dp;
   3680 	vnode_t *nvp;
   3681 	struct vattr *nvap;
   3682 	struct vattr nva;
   3683 	entryplus3_info *infop = NULL;
   3684 	int size = 0;
   3685 	int nents = 0;
   3686 	int bufsize = 0;
   3687 	int entrysize = 0;
   3688 	int tofit = 0;
   3689 	int rd_unit = rfs3_readdir_unit;
   3690 	int prev_len;
   3691 	int space_left;
   3692 	int i;
   3693 	uint_t *namlen = NULL;
   3694 	char *ndata = NULL;
   3695 	struct sockaddr *ca;
   3696 	size_t ret;
   3697 
   3698 	vap = NULL;
   3699 
   3700 	vp = nfs3_fhtovp(&args->dir, exi);
   3701 
   3702 	DTRACE_NFSV3_4(op__readdirplus__start, struct svc_req *, req,
   3703 	    cred_t *, cr, vnode_t *, vp, READDIRPLUS3args *, args);
   3704 
   3705 	if (vp == NULL) {
   3706 		error = ESTALE;
   3707 		goto out;
   3708 	}
   3709 
   3710 	if (is_system_labeled()) {
   3711 		bslabel_t *clabel = req->rq_label;
   3712 
   3713 		ASSERT(clabel != NULL);
   3714 		DTRACE_PROBE2(tx__rfs3__log__info__opreaddirplus__clabel,
   3715 		    char *, "got client label from request(1)",
   3716 		    struct svc_req *, req);
   3717 
   3718 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   3719 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
   3720 			    exi)) {
   3721 				resp->status = NFS3ERR_ACCES;
   3722 				goto out1;
   3723 			}
   3724 		}
   3725 	}
   3726 
   3727 	(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, NULL);
   3728 
   3729 #ifdef DEBUG
   3730 	if (rfs3_do_pre_op_attr) {
   3731 		va.va_mask = AT_ALL;
   3732 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3733 	} else
   3734 		vap = NULL;
   3735 #else
   3736 	va.va_mask = AT_ALL;
   3737 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3738 #endif
   3739 
   3740 	if (vp->v_type != VDIR) {
   3741 		error = ENOTDIR;
   3742 		goto out;
   3743 	}
   3744 
   3745 	error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
   3746 	if (error)
   3747 		goto out;
   3748 
   3749 	/*
   3750 	 * Don't allow arbitrary counts for allocation
   3751 	 */
   3752 	if (args->maxcount > rfs3_tsize(req))
   3753 		args->maxcount = rfs3_tsize(req);
   3754 
   3755 	/*
   3756 	 * Make sure that there is room to read at least one entry
   3757 	 * if any are available
   3758 	 */
   3759 	args->dircount = MIN(args->dircount, args->maxcount);
   3760 
   3761 	if (args->dircount < DIRENT64_RECLEN(MAXNAMELEN))
   3762 		args->dircount = DIRENT64_RECLEN(MAXNAMELEN);
   3763 
   3764 	/*
   3765 	 * This allocation relies on a minimum directory entry
   3766 	 * being roughly 24 bytes.  Therefore, the namlen array
   3767 	 * will have enough space based on the maximum number of
   3768 	 * entries to read.
   3769 	 */
   3770 	namlen = kmem_alloc(args->dircount, KM_SLEEP);
   3771 
   3772 	space_left = args->dircount;
   3773 	data = kmem_alloc(args->dircount, KM_SLEEP);
   3774 	dp = (struct dirent64 *)data;
   3775 	uio.uio_iov = &iov;
   3776 	uio.uio_iovcnt = 1;
   3777 	uio.uio_segflg = UIO_SYSSPACE;
   3778 	uio.uio_extflg = UIO_COPY_CACHED;
   3779 	uio.uio_loffset = (offset_t)args->cookie;
   3780 
   3781 	/*
   3782 	 * bufsize is used to keep track of the size of the response as we
   3783 	 * get post op attributes and filehandles for each entry.  This is
   3784 	 * an optimization as the server may have read more entries than will
   3785 	 * fit in the buffer specified by maxcount.  We stop calculating
   3786 	 * post op attributes and filehandles once we have exceeded maxcount.
   3787 	 * This will minimize the effect of truncation.
   3788 	 *
   3789 	 * It is primed with:
   3790 	 *	1 for the status +
   3791 	 *	1 for the dir_attributes.attributes boolean +
   3792 	 *	2 for the cookie verifier
   3793 	 * all times BYTES_PER_XDR_UNIT to convert from XDR units
   3794 	 * to bytes.  If there are directory attributes to be
   3795 	 * returned, then:
   3796 	 *	NFS3_SIZEOF_FATTR3 for the dir_attributes.attr fattr3
   3797 	 * time BYTES_PER_XDR_UNIT is added to account for them.
   3798 	 */
   3799 	bufsize = (1 + 1 + 2) * BYTES_PER_XDR_UNIT;
   3800 	if (vap != NULL)
   3801 		bufsize += NFS3_SIZEOF_FATTR3 * BYTES_PER_XDR_UNIT;
   3802 
   3803 getmoredents:
   3804 	/*
   3805 	 * Here we make a check so that our read unit is not larger than
   3806 	 * the space left in the buffer.
   3807 	 */
   3808 	rd_unit = MIN(rd_unit, space_left);
   3809 	iov.iov_base = (char *)dp;
   3810 	iov.iov_len = rd_unit;
   3811 	uio.uio_resid = rd_unit;
   3812 	prev_len = rd_unit;
   3813 
   3814 	error = VOP_READDIR(vp, &uio, cr, &iseof, NULL, 0);
   3815 
   3816 	if (error) {
   3817 		kmem_free(data, args->dircount);
   3818 		goto out;
   3819 	}
   3820 
   3821 	if (uio.uio_resid == prev_len && !iseof) {
   3822 		if (nents == 0) {
   3823 			kmem_free(data, args->dircount);
   3824 			resp->status = NFS3ERR_TOOSMALL;
   3825 			goto out1;
   3826 		}
   3827 
   3828 		/*
   3829 		 * We could not get any more entries, so get the attributes
   3830 		 * and filehandle for the entries already obtained.
   3831 		 */
   3832 		goto good;
   3833 	}
   3834 
   3835 	/*
   3836 	 * We estimate the size of the response by assuming the
   3837 	 * entry exists and attributes and filehandle are also valid
   3838 	 */
   3839 	for (size = prev_len - uio.uio_resid;
   3840 	    size > 0;
   3841 	    size -= dp->d_reclen, dp = nextdp(dp)) {
   3842 
   3843 		if (dp->d_ino == 0) {
   3844 			nents++;
   3845 			continue;
   3846 		}
   3847 
   3848 		namlen[nents] = strlen(dp->d_name);
   3849 		entrysize = NFS3_READDIRPLUS_ENTRY(namlen[nents]);
   3850 
   3851 		/*
   3852 		 * We need to check to see if the number of bytes left
   3853 		 * to go into the buffer will actually fit into the
   3854 		 * buffer.  This is calculated as the size of this
   3855 		 * entry plus:
   3856 		 *	1 for the true/false list indicator +
   3857 		 *	1 for the eof indicator
   3858 		 * times BYTES_PER_XDR_UNIT to convert from XDR units
   3859 		 * to bytes.
   3860 		 *
   3861 		 * Also check the dircount limit against the first entry read
   3862 		 *
   3863 		 */
   3864 		tofit = entrysize + (1 + 1) * BYTES_PER_XDR_UNIT;
   3865 		if (bufsize + tofit > args->maxcount) {
   3866 			/*
   3867 			 * We make a check here to see if this was the
   3868 			 * first entry being measured.  If so, then maxcount
   3869 			 * was too small to begin with and so we need to
   3870 			 * return with NFS3ERR_TOOSMALL.
   3871 			 */
   3872 			if (nents == 0) {
   3873 				kmem_free(data, args->dircount);
   3874 				resp->status = NFS3ERR_TOOSMALL;
   3875 				goto out1;
   3876 			}
   3877 			iseof = FALSE;
   3878 			goto good;
   3879 		}
   3880 		bufsize += entrysize;
   3881 		nents++;
   3882 	}
   3883 
   3884 	/*
   3885 	 * If there is enough room to fit at least 1 more entry including
   3886 	 * post op attributes and filehandle in the buffer AND that we haven't
   3887 	 * exceeded dircount then go back and get some more.
   3888 	 */
   3889 	if (!iseof &&
   3890 	    (args->maxcount - bufsize) >= NFS3_READDIRPLUS_ENTRY(MAXNAMELEN)) {
   3891 		space_left -= (prev_len - uio.uio_resid);
   3892 		if (space_left >= DIRENT64_RECLEN(MAXNAMELEN))
   3893 			goto getmoredents;
   3894 
   3895 		/* else, fall through */
   3896 	}
   3897 
   3898 good:
   3899 
   3900 #ifdef DEBUG
   3901 	if (rfs3_do_post_op_attr) {
   3902 		va.va_mask = AT_ALL;
   3903 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3904 	} else
   3905 		vap = NULL;
   3906 #else
   3907 	va.va_mask = AT_ALL;
   3908 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   3909 #endif
   3910 
   3911 	VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
   3912 
   3913 	infop = kmem_alloc(nents * sizeof (struct entryplus3_info), KM_SLEEP);
   3914 	resp->resok.infop = infop;
   3915 
   3916 	dp = (struct dirent64 *)data;
   3917 	for (i = 0; i < nents; i++) {
   3918 
   3919 		if (dp->d_ino == 0) {
   3920 			infop[i].attr.attributes = FALSE;
   3921 			infop[i].fh.handle_follows = FALSE;
   3922 			dp = nextdp(dp);
   3923 			continue;
   3924 		}
   3925 
   3926 		infop[i].namelen = namlen[i];
   3927 
   3928 		error = VOP_LOOKUP(vp, dp->d_name, &nvp, NULL, 0, NULL, cr,
   3929 		    NULL, NULL, NULL);
   3930 		if (error) {
   3931 			infop[i].attr.attributes = FALSE;
   3932 			infop[i].fh.handle_follows = FALSE;
   3933 			dp = nextdp(dp);
   3934 			continue;
   3935 		}
   3936 
   3937 #ifdef DEBUG
   3938 		if (rfs3_do_post_op_attr) {
   3939 			nva.va_mask = AT_ALL;
   3940 			nvap = rfs4_delegated_getattr(nvp, &nva, 0, cr) ?
   3941 			    NULL : &nva;
   3942 		} else
   3943 			nvap = NULL;
   3944 #else
   3945 		nva.va_mask = AT_ALL;
   3946 		nvap = rfs4_delegated_getattr(nvp, &nva, 0, cr) ? NULL : &nva;
   3947 #endif
   3948 		/* Lie about the object type for a referral */
   3949 		if (vn_is_nfs_reparse(nvp, cr))
   3950 			nvap->va_type = VLNK;
   3951 
   3952 		vattr_to_post_op_attr(nvap, &infop[i].attr);
   3953 
   3954 #ifdef DEBUG
   3955 		if (!rfs3_do_post_op_fh3)
   3956 			infop[i].fh.handle_follows = FALSE;
   3957 		else {
   3958 #endif
   3959 		error = makefh3(&infop[i].fh.handle, nvp, exi);
   3960 		if (!error)
   3961 			infop[i].fh.handle_follows = TRUE;
   3962 		else
   3963 			infop[i].fh.handle_follows = FALSE;
   3964 #ifdef DEBUG
   3965 		}
   3966 #endif
   3967 
   3968 		VN_RELE(nvp);
   3969 		dp = nextdp(dp);
   3970 	}
   3971 
   3972 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
   3973 	ret = nfscmd_convdirplus(ca, exi, data, nents, args->dircount, &ndata);
   3974 	if (ndata == NULL)
   3975 		ndata = data;
   3976 
   3977 	if (ret > 0) {
   3978 		/*
   3979 		 * We had to drop one or more entries in order to fit
   3980 		 * during the character conversion.  We need to patch
   3981 		 * up the size and eof info.
   3982 		 */
   3983 		if (iseof)
   3984 			iseof = FALSE;
   3985 
   3986 		ret = nfscmd_dropped_entrysize((struct dirent64 *)data,
   3987 		    nents, ret);
   3988 	}
   3989 
   3990 
   3991 #if 0 /* notyet */
   3992 	/*
   3993 	 * Don't do this.  It causes local disk writes when just
   3994 	 * reading the file and the overhead is deemed larger
   3995 	 * than the benefit.
   3996 	 */
   3997 	/*
   3998 	 * Force modified metadata out to stable storage.
   3999 	 */
   4000 	(void) VOP_FSYNC(vp, FNODSYNC, cr, NULL);
   4001 #endif
   4002 
   4003 	kmem_free(namlen, args->dircount);
   4004 
   4005 	resp->status = NFS3_OK;
   4006 	vattr_to_post_op_attr(vap, &resp->resok.dir_attributes);
   4007 	resp->resok.cookieverf = 0;
   4008 	resp->resok.reply.entries = (entryplus3 *)ndata;
   4009 	resp->resok.reply.eof = iseof;
   4010 	resp->resok.size = nents;
   4011 	resp->resok.count = args->dircount - ret;
   4012 	resp->resok.maxcount = args->maxcount;
   4013 
   4014 	DTRACE_NFSV3_4(op__readdirplus__done, struct svc_req *, req,
   4015 	    cred_t *, cr, vnode_t *, vp, READDIRPLUS3res *, resp);
   4016 	if (ndata != data)
   4017 		kmem_free(data, args->dircount);
   4018 
   4019 
   4020 	VN_RELE(vp);
   4021 
   4022 	return;
   4023 
   4024 out:
   4025 	if (curthread->t_flag & T_WOULDBLOCK) {
   4026 		curthread->t_flag &= ~T_WOULDBLOCK;
   4027 		resp->status = NFS3ERR_JUKEBOX;
   4028 	} else {
   4029 		resp->status = puterrno3(error);
   4030 	}
   4031 out1:
   4032 	DTRACE_NFSV3_4(op__readdirplus__done, struct svc_req *, req,
   4033 	    cred_t *, cr, vnode_t *, vp, READDIRPLUS3res *, resp);
   4034 
   4035 	if (vp != NULL) {
   4036 		VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, NULL);
   4037 		VN_RELE(vp);
   4038 	}
   4039 
   4040 	if (namlen != NULL)
   4041 		kmem_free(namlen, args->dircount);
   4042 
   4043 	vattr_to_post_op_attr(vap, &resp->resfail.dir_attributes);
   4044 }
   4045 
   4046 void *
   4047 rfs3_readdirplus_getfh(READDIRPLUS3args *args)
   4048 {
   4049 
   4050 	return (&args->dir);
   4051 }
   4052 
   4053 void
   4054 rfs3_readdirplus_free(READDIRPLUS3res *resp)
   4055 {
   4056 
   4057 	if (resp->status == NFS3_OK) {
   4058 		kmem_free(resp->resok.reply.entries, resp->resok.count);
   4059 		kmem_free(resp->resok.infop,
   4060 		    resp->resok.size * sizeof (struct entryplus3_info));
   4061 	}
   4062 }
   4063 
   4064 /* ARGSUSED */
   4065 void
   4066 rfs3_fsstat(FSSTAT3args *args, FSSTAT3res *resp, struct exportinfo *exi,
   4067 	struct svc_req *req, cred_t *cr)
   4068 {
   4069 	int error;
   4070 	vnode_t *vp;
   4071 	struct vattr *vap;
   4072 	struct vattr va;
   4073 	struct statvfs64 sb;
   4074 
   4075 	vap = NULL;
   4076 
   4077 	vp = nfs3_fhtovp(&args->fsroot, exi);
   4078 
   4079 	DTRACE_NFSV3_4(op__fsstat__start, struct svc_req *, req,
   4080 	    cred_t *, cr, vnode_t *, vp, FSSTAT3args *, args);
   4081 
   4082 	if (vp == NULL) {
   4083 		error = ESTALE;
   4084 		goto out;
   4085 	}
   4086 
   4087 	if (is_system_labeled()) {
   4088 		bslabel_t *clabel = req->rq_label;
   4089 
   4090 		ASSERT(clabel != NULL);
   4091 		DTRACE_PROBE2(tx__rfs3__log__info__opfsstat__clabel, char *,
   4092 		    "got client label from request(1)", struct svc_req *, req);
   4093 
   4094 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   4095 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
   4096 			    exi)) {
   4097 				resp->status = NFS3ERR_ACCES;
   4098 				goto out1;
   4099 			}
   4100 		}
   4101 	}
   4102 
   4103 	error = VFS_STATVFS(vp->v_vfsp, &sb);
   4104 
   4105 #ifdef DEBUG
   4106 	if (rfs3_do_post_op_attr) {
   4107 		va.va_mask = AT_ALL;
   4108 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   4109 	} else
   4110 		vap = NULL;
   4111 #else
   4112 	va.va_mask = AT_ALL;
   4113 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   4114 #endif
   4115 
   4116 	if (error)
   4117 		goto out;
   4118 
   4119 	resp->status = NFS3_OK;
   4120 	vattr_to_post_op_attr(vap, &resp->resok.obj_attributes);
   4121 	if (sb.f_blocks != (fsblkcnt64_t)-1)
   4122 		resp->resok.tbytes = (size3)sb.f_frsize * (size3)sb.f_blocks;
   4123 	else
   4124 		resp->resok.tbytes = (size3)sb.f_blocks;
   4125 	if (sb.f_bfree != (fsblkcnt64_t)-1)
   4126 		resp->resok.fbytes = (size3)sb.f_frsize * (size3)sb.f_bfree;
   4127 	else
   4128 		resp->resok.fbytes = (size3)sb.f_bfree;
   4129 	if (sb.f_bavail != (fsblkcnt64_t)-1)
   4130 		resp->resok.abytes = (size3)sb.f_frsize * (size3)sb.f_bavail;
   4131 	else
   4132 		resp->resok.abytes = (size3)sb.f_bavail;
   4133 	resp->resok.tfiles = (size3)sb.f_files;
   4134 	resp->resok.ffiles = (size3)sb.f_ffree;
   4135 	resp->resok.afiles = (size3)sb.f_favail;
   4136 	resp->resok.invarsec = 0;
   4137 
   4138 	DTRACE_NFSV3_4(op__fsstat__done, struct svc_req *, req,
   4139 	    cred_t *, cr, vnode_t *, vp, FSSTAT3res *, resp);
   4140 	VN_RELE(vp);
   4141 
   4142 	return;
   4143 
   4144 out:
   4145 	if (curthread->t_flag & T_WOULDBLOCK) {
   4146 		curthread->t_flag &= ~T_WOULDBLOCK;
   4147 		resp->status = NFS3ERR_JUKEBOX;
   4148 	} else
   4149 		resp->status = puterrno3(error);
   4150 out1:
   4151 	DTRACE_NFSV3_4(op__fsstat__done, struct svc_req *, req,
   4152 	    cred_t *, cr, vnode_t *, vp, FSSTAT3res *, resp);
   4153 
   4154 	if (vp != NULL)
   4155 		VN_RELE(vp);
   4156 	vattr_to_post_op_attr(vap, &resp->resfail.obj_attributes);
   4157 }
   4158 
   4159 void *
   4160 rfs3_fsstat_getfh(FSSTAT3args *args)
   4161 {
   4162 
   4163 	return (&args->fsroot);
   4164 }
   4165 
   4166 void
   4167 rfs3_fsinfo(FSINFO3args *args, FSINFO3res *resp, struct exportinfo *exi,
   4168 	struct svc_req *req, cred_t *cr)
   4169 {
   4170 	vnode_t *vp;
   4171 	struct vattr *vap;
   4172 	struct vattr va;
   4173 	uint32_t xfer_size;
   4174 	ulong_t l = 0;
   4175 	int error;
   4176 
   4177 	vp = nfs3_fhtovp(&args->fsroot, exi);
   4178 
   4179 	DTRACE_NFSV3_4(op__fsinfo__start, struct svc_req *, req,
   4180 	    cred_t *, cr, vnode_t *, vp, FSINFO3args *, args);
   4181 
   4182 	if (vp == NULL) {
   4183 		if (curthread->t_flag & T_WOULDBLOCK) {
   4184 			curthread->t_flag &= ~T_WOULDBLOCK;
   4185 			resp->status = NFS3ERR_JUKEBOX;
   4186 		} else
   4187 			resp->status = NFS3ERR_STALE;
   4188 		vattr_to_post_op_attr(NULL, &resp->resfail.obj_attributes);
   4189 		goto out;
   4190 	}
   4191 
   4192 	if (is_system_labeled()) {
   4193 		bslabel_t *clabel = req->rq_label;
   4194 
   4195 		ASSERT(clabel != NULL);
   4196 		DTRACE_PROBE2(tx__rfs3__log__info__opfsinfo__clabel, char *,
   4197 		    "got client label from request(1)", struct svc_req *, req);
   4198 
   4199 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   4200 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
   4201 			    exi)) {
   4202 				resp->status = NFS3ERR_STALE;
   4203 				vattr_to_post_op_attr(NULL,
   4204 				    &resp->resfail.obj_attributes);
   4205 				goto out;
   4206 			}
   4207 		}
   4208 	}
   4209 
   4210 #ifdef DEBUG
   4211 	if (rfs3_do_post_op_attr) {
   4212 		va.va_mask = AT_ALL;
   4213 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   4214 	} else
   4215 		vap = NULL;
   4216 #else
   4217 	va.va_mask = AT_ALL;
   4218 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   4219 #endif
   4220 
   4221 	resp->status = NFS3_OK;
   4222 	vattr_to_post_op_attr(vap, &resp->resok.obj_attributes);
   4223 	xfer_size = rfs3_tsize(req);
   4224 	resp->resok.rtmax = xfer_size;
   4225 	resp->resok.rtpref = xfer_size;
   4226 	resp->resok.rtmult = DEV_BSIZE;
   4227 	resp->resok.wtmax = xfer_size;
   4228 	resp->resok.wtpref = xfer_size;
   4229 	resp->resok.wtmult = DEV_BSIZE;
   4230 	resp->resok.dtpref = MAXBSIZE;
   4231 
   4232 	/*
   4233 	 * Large file spec: want maxfilesize based on limit of
   4234 	 * underlying filesystem.  We can guess 2^31-1 if need be.
   4235 	 */
   4236 	error = VOP_PATHCONF(vp, _PC_FILESIZEBITS, &l, cr, NULL);
   4237 	if (error) {
   4238 		resp->status = puterrno3(error);
   4239 		goto out;
   4240 	}
   4241 
   4242 	/*
   4243 	 * If the underlying file system does not support _PC_FILESIZEBITS,
   4244 	 * return a reasonable default. Note that error code on VOP_PATHCONF
   4245 	 * will be 0, even if the underlying file system does not support
   4246 	 * _PC_FILESIZEBITS.
   4247 	 */
   4248 	if (l == (ulong_t)-1) {
   4249 		resp->resok.maxfilesize = MAXOFF32_T;
   4250 	} else {
   4251 		if (l >= (sizeof (uint64_t) * 8))
   4252 			resp->resok.maxfilesize = INT64_MAX;
   4253 		else
   4254 			resp->resok.maxfilesize = (1LL << (l-1)) - 1;
   4255 	}
   4256 
   4257 	resp->resok.time_delta.seconds = 0;
   4258 	resp->resok.time_delta.nseconds = 1000;
   4259 	resp->resok.properties = FSF3_LINK | FSF3_SYMLINK |
   4260 	    FSF3_HOMOGENEOUS | FSF3_CANSETTIME;
   4261 
   4262 	DTRACE_NFSV3_4(op__fsinfo__done, struct svc_req *, req,
   4263 	    cred_t *, cr, vnode_t *, vp, FSINFO3res *, resp);
   4264 
   4265 	VN_RELE(vp);
   4266 
   4267 	return;
   4268 
   4269 out:
   4270 	DTRACE_NFSV3_4(op__fsinfo__done, struct svc_req *, req,
   4271 	    cred_t *, cr, vnode_t *, NULL, FSINFO3res *, resp);
   4272 	if (vp != NULL)
   4273 		VN_RELE(vp);
   4274 }
   4275 
   4276 void *
   4277 rfs3_fsinfo_getfh(FSINFO3args *args)
   4278 {
   4279 
   4280 	return (&args->fsroot);
   4281 }
   4282 
   4283 /* ARGSUSED */
   4284 void
   4285 rfs3_pathconf(PATHCONF3args *args, PATHCONF3res *resp, struct exportinfo *exi,
   4286 	struct svc_req *req, cred_t *cr)
   4287 {
   4288 	int error;
   4289 	vnode_t *vp;
   4290 	struct vattr *vap;
   4291 	struct vattr va;
   4292 	ulong_t val;
   4293 
   4294 	vap = NULL;
   4295 
   4296 	vp = nfs3_fhtovp(&args->object, exi);
   4297 
   4298 	DTRACE_NFSV3_4(op__pathconf__start, struct svc_req *, req,
   4299 	    cred_t *, cr, vnode_t *, vp, PATHCONF3args *, args);
   4300 
   4301 	if (vp == NULL) {
   4302 		error = ESTALE;
   4303 		goto out;
   4304 	}
   4305 
   4306 	if (is_system_labeled()) {
   4307 		bslabel_t *clabel = req->rq_label;
   4308 
   4309 		ASSERT(clabel != NULL);
   4310 		DTRACE_PROBE2(tx__rfs3__log__info__oppathconf__clabel, char *,
   4311 		    "got client label from request(1)", struct svc_req *, req);
   4312 
   4313 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   4314 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
   4315 			    exi)) {
   4316 				resp->status = NFS3ERR_ACCES;
   4317 				goto out1;
   4318 			}
   4319 		}
   4320 	}
   4321 
   4322 #ifdef DEBUG
   4323 	if (rfs3_do_post_op_attr) {
   4324 		va.va_mask = AT_ALL;
   4325 		vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   4326 	} else
   4327 		vap = NULL;
   4328 #else
   4329 	va.va_mask = AT_ALL;
   4330 	vap = VOP_GETATTR(vp, &va, 0, cr, NULL) ? NULL : &va;
   4331 #endif
   4332 
   4333 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &val, cr, NULL);
   4334 	if (error)
   4335 		goto out;
   4336 	resp->resok.info.link_max = (uint32)val;
   4337 
   4338 	error = VOP_PATHCONF(vp, _PC_NAME_MAX, &val, cr, NULL);
   4339 	if (error)
   4340 		goto out;
   4341 	resp->resok.info.name_max = (uint32)val;
   4342 
   4343 	error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &val, cr, NULL);
   4344 	if (error)
   4345 		goto out;
   4346 	if (val == 1)
   4347 		resp->resok.info.no_trunc = TRUE;
   4348 	else
   4349 		resp->resok.info.no_trunc = FALSE;
   4350 
   4351 	error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &val, cr, NULL);
   4352 	if (error)
   4353 		goto out;
   4354 	if (val == 1)
   4355 		resp->resok.info.chown_restricted = TRUE;
   4356 	else
   4357 		resp->resok.info.chown_restricted = FALSE;
   4358 
   4359 	resp->status = NFS3_OK;
   4360 	vattr_to_post_op_attr(vap, &resp->resok.obj_attributes);
   4361 	resp->resok.info.case_insensitive = FALSE;
   4362 	resp->resok.info.case_preserving = TRUE;
   4363 	DTRACE_NFSV3_4(op__pathconf__done, struct svc_req *, req,
   4364 	    cred_t *, cr, vnode_t *, vp, PATHCONF3res *, resp);
   4365 	VN_RELE(vp);
   4366 	return;
   4367 
   4368 out:
   4369 	if (curthread->t_flag & T_WOULDBLOCK) {
   4370 		curthread->t_flag &= ~T_WOULDBLOCK;
   4371 		resp->status = NFS3ERR_JUKEBOX;
   4372 	} else
   4373 		resp->status = puterrno3(error);
   4374 out1:
   4375 	DTRACE_NFSV3_4(op__pathconf__done, struct svc_req *, req,
   4376 	    cred_t *, cr, vnode_t *, vp, PATHCONF3res *, resp);
   4377 	if (vp != NULL)
   4378 		VN_RELE(vp);
   4379 	vattr_to_post_op_attr(vap, &resp->resfail.obj_attributes);
   4380 }
   4381 
   4382 void *
   4383 rfs3_pathconf_getfh(PATHCONF3args *args)
   4384 {
   4385 
   4386 	return (&args->object);
   4387 }
   4388 
   4389 void
   4390 rfs3_commit(COMMIT3args *args, COMMIT3res *resp, struct exportinfo *exi,
   4391 	struct svc_req *req, cred_t *cr)
   4392 {
   4393 	int error;
   4394 	vnode_t *vp;
   4395 	struct vattr *bvap;
   4396 	struct vattr bva;
   4397 	struct vattr *avap;
   4398 	struct vattr ava;
   4399 
   4400 	bvap = NULL;
   4401 	avap = NULL;
   4402 
   4403 	vp = nfs3_fhtovp(&args->file, exi);
   4404 
   4405 	DTRACE_NFSV3_4(op__commit__start, struct svc_req *, req,
   4406 	    cred_t *, cr, vnode_t *, vp, COMMIT3args *, args);
   4407 
   4408 	if (vp == NULL) {
   4409 		error = ESTALE;
   4410 		goto out;
   4411 	}
   4412 
   4413 	bva.va_mask = AT_ALL;
   4414 	error = VOP_GETATTR(vp, &bva, 0, cr, NULL);
   4415 
   4416 	/*
   4417 	 * If we can't get the attributes, then we can't do the
   4418 	 * right access checking.  So, we'll fail the request.
   4419 	 */
   4420 	if (error)
   4421 		goto out;
   4422 
   4423 #ifdef DEBUG
   4424 	if (rfs3_do_pre_op_attr)
   4425 		bvap = &bva;
   4426 	else
   4427 		bvap = NULL;
   4428 #else
   4429 	bvap = &bva;
   4430 #endif
   4431 
   4432 	if (rdonly(exi, req)) {
   4433 		resp->status = NFS3ERR_ROFS;
   4434 		goto out1;
   4435 	}
   4436 
   4437 	if (vp->v_type != VREG) {
   4438 		resp->status = NFS3ERR_INVAL;
   4439 		goto out1;
   4440 	}
   4441 
   4442 	if (is_system_labeled()) {
   4443 		bslabel_t *clabel = req->rq_label;
   4444 
   4445 		ASSERT(clabel != NULL);
   4446 		DTRACE_PROBE2(tx__rfs3__log__info__opcommit__clabel, char *,
   4447 		    "got client label from request(1)", struct svc_req *, req);
   4448 
   4449 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
   4450 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
   4451 			    exi)) {
   4452 				resp->status = NFS3ERR_ACCES;
   4453 				goto out1;
   4454 			}
   4455 		}
   4456 	}
   4457 
   4458 	if (crgetuid(cr) != bva.va_uid &&
   4459 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL)))
   4460 		goto out;
   4461 
   4462 	error = VOP_PUTPAGE(vp, args->offset, args->count, 0, cr, NULL);
   4463 	if (!error)
   4464 		error = VOP_FSYNC(vp, FNODSYNC, cr, NULL);
   4465 
   4466 #ifdef DEBUG
   4467 	if (rfs3_do_post_op_attr) {
   4468 		ava.va_mask = AT_ALL;
   4469 		avap = VOP_GETATTR(vp, &ava, 0, cr, NULL) ? NULL : &ava;
   4470 	} else
   4471 		avap = NULL;
   4472 #else
   4473 	ava.va_mask = AT_ALL;
   4474 	avap = VOP_GETATTR(vp, &ava, 0, cr, NULL) ? NULL : &ava;
   4475 #endif
   4476 
   4477 	if (error)
   4478 		goto out;
   4479 
   4480 	resp->status = NFS3_OK;
   4481 	vattr_to_wcc_data(bvap, avap, &resp->resok.file_wcc);
   4482 	resp->resok.verf = write3verf;
   4483 
   4484 	DTRACE_NFSV3_4(op__commit__done, struct svc_req *, req,
   4485 	    cred_t *, cr, vnode_t *, vp, COMMIT3res *, resp);
   4486 
   4487 	VN_RELE(vp);
   4488 
   4489 	return;
   4490 
   4491 out:
   4492 	if (curthread->t_flag & T_WOULDBLOCK) {
   4493 		curthread->t_flag &= ~T_WOULDBLOCK;
   4494 		resp->status = NFS3ERR_JUKEBOX;
   4495 	} else
   4496 		resp->status = puterrno3(error);
   4497 out1:
   4498 	DTRACE_NFSV3_4(op__commit__done, struct svc_req *, req,
   4499 	    cred_t *, cr, vnode_t *, vp, COMMIT3res *, resp);
   4500 
   4501 	if (vp != NULL)
   4502 		VN_RELE(vp);
   4503 	vattr_to_wcc_data(bvap, avap, &resp->resfail.file_wcc);
   4504 }
   4505 
   4506 void *
   4507 rfs3_commit_getfh(COMMIT3args *args)
   4508 {
   4509 
   4510 	return (&args->file);
   4511 }
   4512 
   4513 static int
   4514 sattr3_to_vattr(sattr3 *sap, struct vattr *vap)
   4515 {
   4516 
   4517 	vap->va_mask = 0;
   4518 
   4519 	if (sap->mode.set_it) {
   4520 		vap->va_mode = (mode_t)sap->mode.mode;
   4521 		vap->va_mask |= AT_MODE;
   4522 	}
   4523 	if (sap->uid.set_it) {
   4524 		vap->va_uid = (uid_t)sap->uid.uid;
   4525 		vap->va_mask |= AT_UID;
   4526 	}
   4527 	if (sap->gid.set_it) {
   4528 		vap->va_gid = (gid_t)sap->gid.gid;
   4529 		vap->va_mask |= AT_GID;
   4530 	}
   4531 	if (sap->size.set_it) {
   4532 		if (sap->size.size > (size3)((u_longlong_t)-1))
   4533 			return (EINVAL);
   4534 		vap->va_size = sap->size.size;
   4535 		vap->va_mask |= AT_SIZE;
   4536 	}
   4537 	if (sap->atime.set_it == SET_TO_CLIENT_TIME) {
   4538 #ifndef _LP64
   4539 		/* check time validity */
   4540 		if (!NFS3_TIME_OK(sap->atime.atime.seconds))
   4541 			return (EOVERFLOW);
   4542 #endif
   4543 		/*
   4544 		 * nfs protocol defines times as unsigned so don't extend sign,
   4545 		 * unless sysadmin set nfs_allow_preepoch_time.
   4546 		 */
   4547 		NFS_TIME_T_CONVERT(vap->va_atime.tv_sec,
   4548 		    sap->atime.atime.seconds);
   4549 		vap->va_atime.tv_nsec = (uint32_t)sap->atime.atime.nseconds;
   4550 		vap->va_mask |= AT_ATIME;
   4551 	} else if (sap->atime.set_it == SET_TO_SERVER_TIME) {
   4552 		gethrestime(&vap->va_atime);
   4553 		vap->va_mask |= AT_ATIME;
   4554 	}
   4555 	if (sap->mtime.set_it == SET_TO_CLIENT_TIME) {
   4556 #ifndef _LP64
   4557 		/* check time validity */
   4558 		if (!NFS3_TIME_OK(sap->mtime.mtime.seconds))
   4559 			return (EOVERFLOW);
   4560 #endif
   4561 		/*
   4562 		 * nfs protocol defines times as unsigned so don't extend sign,
   4563 		 * unless sysadmin set nfs_allow_preepoch_time.
   4564 		 */
   4565 		NFS_TIME_T_CONVERT(vap->va_mtime.tv_sec,
   4566 		    sap->mtime.mtime.seconds);
   4567 		vap->va_mtime.tv_nsec = (uint32_t)sap->mtime.mtime.nseconds;
   4568 		vap->va_mask |= AT_MTIME;
   4569 	} else if (sap->mtime.set_it == SET_TO_SERVER_TIME) {
   4570 		gethrestime(&vap->va_mtime);
   4571 		vap->va_mask |= AT_MTIME;
   4572 	}
   4573 
   4574 	return (0);
   4575 }
   4576 
   4577 static ftype3 vt_to_nf3[] = {
   4578 	0, NF3REG, NF3DIR, NF3BLK, NF3CHR, NF3LNK, NF3FIFO, 0, 0, NF3SOCK, 0
   4579 };
   4580 
   4581 static int
   4582 vattr_to_fattr3(struct vattr *vap, fattr3 *fap)
   4583 {
   4584 
   4585 	ASSERT(vap->va_type >= VNON && vap->va_type <= VBAD);
   4586 	/* Return error if time or size overflow */
   4587 	if (! (NFS_VAP_TIME_OK(vap) && NFS3_SIZE_OK(vap->va_size))) {
   4588 		return (EOVERFLOW);
   4589 	}
   4590 	fap->type = vt_to_nf3[vap->va_type];
   4591 	fap->mode = (mode3)(vap->va_mode & MODEMASK);
   4592 	fap->nlink = (uint32)vap->va_nlink;
   4593 	if (vap->va_uid == UID_NOBODY)
   4594 		fap->uid = (uid3)NFS_UID_NOBODY;
   4595 	else
   4596 		fap->uid = (uid3)vap->va_uid;
   4597 	if (vap->va_gid == GID_NOBODY)
   4598 		fap->gid = (gid3)NFS_GID_NOBODY;
   4599 	else
   4600 		fap->gid = (gid3)vap->va_gid;
   4601 	fap->size = (size3)vap->va_size;
   4602 	fap->used = (size3)DEV_BSIZE * (size3)vap->va_nblocks;
   4603 	fap->rdev.specdata1 = (uint32)getmajor(vap->va_rdev);
   4604 	fap->rdev.specdata2 = (uint32)getminor(vap->va_rdev);
   4605 	fap->fsid = (uint64)vap->va_fsid;
   4606 	fap->fileid = (fileid3)vap->va_nodeid;
   4607 	fap->atime.seconds = vap->va_atime.tv_sec;
   4608 	fap->atime.nseconds = vap->va_atime.tv_nsec;
   4609 	fap->mtime.seconds = vap->va_mtime.tv_sec;
   4610 	fap->mtime.nseconds = vap->va_mtime.tv_nsec;
   4611 	fap->ctime.seconds = vap->va_ctime.tv_sec;
   4612 	fap->ctime.nseconds = vap->va_ctime.tv_nsec;
   4613 	return (0);
   4614 }
   4615 
   4616 static int
   4617 vattr_to_wcc_attr(struct vattr *vap, wcc_attr *wccap)
   4618 {
   4619 
   4620 	/* Return error if time or size overflow */
   4621 	if (!(NFS_TIME_T_OK(vap->va_mtime.tv_sec) &&
   4622 	    NFS_TIME_T_OK(vap->va_ctime.tv_sec) &&
   4623 	    NFS3_SIZE_OK(vap->va_size))) {
   4624 		return (EOVERFLOW);
   4625 	}
   4626 	wccap->size = (size3)vap->va_size;
   4627 	wccap->mtime.seconds = vap->va_mtime.tv_sec;
   4628 	wccap->mtime.nseconds = vap->va_mtime.tv_nsec;
   4629 	wccap->ctime.seconds = vap->va_ctime.tv_sec;
   4630 	wccap->ctime.nseconds = vap->va_ctime.tv_nsec;
   4631 	return (0);
   4632 }
   4633 
   4634 static void
   4635 vattr_to_pre_op_attr(struct vattr *vap, pre_op_attr *poap)
   4636 {
   4637 
   4638 	/* don't return attrs if time overflow */
   4639 	if ((vap != NULL) && !vattr_to_wcc_attr(vap, &poap->attr)) {
   4640 		poap->attributes = TRUE;
   4641 	} else
   4642 		poap->attributes = FALSE;
   4643 }
   4644 
   4645 void
   4646 vattr_to_post_op_attr(struct vattr *vap, post_op_attr *poap)
   4647 {
   4648 
   4649 	/* don't return attrs if time overflow */
   4650 	if ((vap != NULL) && !vattr_to_fattr3(vap, &poap->attr)) {
   4651 		poap->attributes = TRUE;
   4652 	} else
   4653 		poap->attributes = FALSE;
   4654 }
   4655 
   4656 static void
   4657 vattr_to_wcc_data(struct vattr *bvap, struct vattr *avap, wcc_data *wccp)
   4658 {
   4659 
   4660 	vattr_to_pre_op_attr(bvap, &wccp->before);
   4661 	vattr_to_post_op_attr(avap, &wccp->after);
   4662 }
   4663 
   4664 void
   4665 rfs3_srvrinit(void)
   4666 {
   4667 	struct rfs3_verf_overlay {
   4668 		uint_t id; /* a "unique" identifier */
   4669 		int ts; /* a unique timestamp */
   4670 	} *verfp;
   4671 	timestruc_t now;
   4672 
   4673 	/*
   4674 	 * The following algorithm attempts to find a unique verifier
   4675 	 * to be used as the write verifier returned from the server
   4676 	 * to the client.  It is important that this verifier change
   4677 	 * whenever the server reboots.  Of secondary importance, it
   4678 	 * is important for the verifier to be unique between two
   4679 	 * different servers.
   4680 	 *
   4681 	 * Thus, an attempt is made to use the system hostid and the
   4682 	 * current time in seconds when the nfssrv kernel module is
   4683 	 * loaded.  It is assumed that an NFS server will not be able
   4684 	 * to boot and then to reboot in less than a second.  If the
   4685 	 * hostid has not been set, then the current high resolution
   4686 	 * time is used.  This will ensure different verifiers each
   4687 	 * time the server reboots and minimize the chances that two
   4688 	 * different servers will have the same verifier.
   4689 	 */
   4690 
   4691 #ifndef	lint
   4692 	/*
   4693 	 * We ASSERT that this constant logic expression is
   4694 	 * always true because in the past, it wasn't.
   4695 	 */
   4696 	ASSERT(sizeof (*verfp) <= sizeof (write3verf));
   4697 #endif
   4698 
   4699 	gethrestime(&now);
   4700 	verfp = (struct rfs3_verf_overlay *)&write3verf;
   4701 	verfp->ts = (int)now.tv_sec;
   4702 	verfp->id = zone_get_hostid(NULL);
   4703 
   4704 	if (verfp->id == 0)
   4705 		verfp->id = (uint_t)now.tv_nsec;
   4706 
   4707 	nfs3_srv_caller_id = fs_new_caller_id();
   4708 
   4709 }
   4710 
   4711 static int
   4712 rdma_setup_read_data3(READ3args *args, READ3resok *rok)
   4713 {
   4714 	struct clist	*wcl;
   4715 	int		wlist_len;
   4716 	count3		count = rok->count;
   4717 
   4718 	wcl = args->wlist;
   4719 	if (rdma_setup_read_chunks(wcl, count, &wlist_len) == FALSE) {
   4720 		return (FALSE);
   4721 	}
   4722 
   4723 	wcl = args->wlist;
   4724 	rok->wlist_len = wlist_len;
   4725 	rok->wlist = wcl;
   4726 	return (TRUE);
   4727 }
   4728 
   4729 void
   4730 rfs3_srvrfini(void)
   4731 {
   4732 	/* Nothing to do */
   4733 }
   4734