xref: /titanic_41/usr/src/uts/common/fs/nfs/nfs4_srv.c (revision 35e863ed502128a206ca8df4bc9705be8ee059f6)
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 /*
23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2012 by Delphix. All rights reserved.
26  */
27 
28 /*
29  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
30  *	All Rights Reserved
31  */
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <sys/systm.h>
36 #include <sys/cred.h>
37 #include <sys/buf.h>
38 #include <sys/vfs.h>
39 #include <sys/vfs_opreg.h>
40 #include <sys/vnode.h>
41 #include <sys/uio.h>
42 #include <sys/errno.h>
43 #include <sys/sysmacros.h>
44 #include <sys/statvfs.h>
45 #include <sys/kmem.h>
46 #include <sys/dirent.h>
47 #include <sys/cmn_err.h>
48 #include <sys/debug.h>
49 #include <sys/systeminfo.h>
50 #include <sys/flock.h>
51 #include <sys/pathname.h>
52 #include <sys/nbmlock.h>
53 #include <sys/share.h>
54 #include <sys/atomic.h>
55 #include <sys/policy.h>
56 #include <sys/fem.h>
57 #include <sys/sdt.h>
58 #include <sys/ddi.h>
59 #include <sys/zone.h>
60 
61 #include <fs/fs_reparse.h>
62 
63 #include <rpc/types.h>
64 #include <rpc/auth.h>
65 #include <rpc/rpcsec_gss.h>
66 #include <rpc/svc.h>
67 
68 #include <nfs/nfs.h>
69 #include <nfs/export.h>
70 #include <nfs/nfs_cmd.h>
71 #include <nfs/lm.h>
72 #include <nfs/nfs4.h>
73 
74 #include <sys/strsubr.h>
75 #include <sys/strsun.h>
76 
77 #include <inet/common.h>
78 #include <inet/ip.h>
79 #include <inet/ip6.h>
80 
81 #include <sys/tsol/label.h>
82 #include <sys/tsol/tndb.h>
83 
84 #define	RFS4_MAXLOCK_TRIES 4	/* Try to get the lock this many times */
85 static int rfs4_maxlock_tries = RFS4_MAXLOCK_TRIES;
86 #define	RFS4_LOCK_DELAY 10	/* Milliseconds */
87 static clock_t  rfs4_lock_delay = RFS4_LOCK_DELAY;
88 extern struct svc_ops rdma_svc_ops;
89 extern int nfs_loaned_buffers;
90 /* End of Tunables */
91 
92 static int rdma_setup_read_data4(READ4args *, READ4res *);
93 
94 /*
95  * Used to bump the stateid4.seqid value and show changes in the stateid
96  */
97 #define	next_stateid(sp) (++(sp)->bits.chgseq)
98 
99 /*
100  * RFS4_MINLEN_ENTRY4: XDR-encoded size of smallest possible dirent.
101  *	This is used to return NFS4ERR_TOOSMALL when clients specify
102  *	maxcount that isn't large enough to hold the smallest possible
103  *	XDR encoded dirent.
104  *
105  *	    sizeof cookie (8 bytes) +
106  *	    sizeof name_len (4 bytes) +
107  *	    sizeof smallest (padded) name (4 bytes) +
108  *	    sizeof bitmap4_len (12 bytes) +   NOTE: we always encode len=2 bm4
109  *	    sizeof attrlist4_len (4 bytes) +
110  *	    sizeof next boolean (4 bytes)
111  *
112  * RFS4_MINLEN_RDDIR4: XDR-encoded size of READDIR op reply containing
113  * the smallest possible entry4 (assumes no attrs requested).
114  *	sizeof nfsstat4 (4 bytes) +
115  *	sizeof verifier4 (8 bytes) +
116  *	sizeof entry4list bool (4 bytes) +
117  *	sizeof entry4 	(36 bytes) +
118  *	sizeof eof bool  (4 bytes)
119  *
120  * RFS4_MINLEN_RDDIR_BUF: minimum length of buffer server will provide to
121  *	VOP_READDIR.  Its value is the size of the maximum possible dirent
122  *	for solaris.  The DIRENT64_RECLEN macro returns	the size of dirent
123  *	required for a given name length.  MAXNAMELEN is the maximum
124  *	filename length allowed in Solaris.  The first two DIRENT64_RECLEN()
125  *	macros are to allow for . and .. entries -- just a minor tweak to try
126  *	and guarantee that buffer we give to VOP_READDIR will be large enough
127  *	to hold ., .., and the largest possible solaris dirent64.
128  */
129 #define	RFS4_MINLEN_ENTRY4 36
130 #define	RFS4_MINLEN_RDDIR4 (4 + NFS4_VERIFIER_SIZE + 4 + RFS4_MINLEN_ENTRY4 + 4)
131 #define	RFS4_MINLEN_RDDIR_BUF \
132 	(DIRENT64_RECLEN(1) + DIRENT64_RECLEN(2) + DIRENT64_RECLEN(MAXNAMELEN))
133 
134 /*
135  * It would be better to pad to 4 bytes since that's what XDR would do,
136  * but the dirents UFS gives us are already padded to 8, so just take
137  * what we're given.  Dircount is only a hint anyway.  Currently the
138  * solaris kernel is ASCII only, so there's no point in calling the
139  * UTF8 functions.
140  *
141  * dirent64: named padded to provide 8 byte struct alignment
142  *	d_ino(8) + d_off(8) + d_reclen(2) + d_name(namelen + null(1) + pad)
143  *
144  * cookie: uint64_t   +  utf8namelen: uint_t  +   utf8name padded to 8 bytes
145  *
146  */
147 #define	DIRENT64_TO_DIRCOUNT(dp) \
148 	(3 * BYTES_PER_XDR_UNIT + DIRENT64_NAMELEN((dp)->d_reclen))
149 
150 time_t rfs4_start_time;			/* Initialized in rfs4_srvrinit */
151 
152 static sysid_t lockt_sysid;		/* dummy sysid for all LOCKT calls */
153 
154 u_longlong_t	nfs4_srv_caller_id;
155 uint_t		nfs4_srv_vkey = 0;
156 
157 verifier4	Write4verf;
158 verifier4	Readdir4verf;
159 
160 void	rfs4_init_compound_state(struct compound_state *);
161 
162 static void	nullfree(caddr_t);
163 static void	rfs4_op_inval(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
164 			struct compound_state *);
165 static void	rfs4_op_access(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
166 			struct compound_state *);
167 static void	rfs4_op_close(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
168 			struct compound_state *);
169 static void	rfs4_op_commit(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
170 			struct compound_state *);
171 static void	rfs4_op_create(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
172 			struct compound_state *);
173 static void	rfs4_op_create_free(nfs_resop4 *resop);
174 static void	rfs4_op_delegreturn(nfs_argop4 *, nfs_resop4 *,
175 			struct svc_req *, struct compound_state *);
176 static void	rfs4_op_delegpurge(nfs_argop4 *, nfs_resop4 *,
177 			struct svc_req *, struct compound_state *);
178 static void	rfs4_op_getattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
179 			struct compound_state *);
180 static void	rfs4_op_getattr_free(nfs_resop4 *);
181 static void	rfs4_op_getfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
182 			struct compound_state *);
183 static void	rfs4_op_getfh_free(nfs_resop4 *);
184 static void	rfs4_op_illegal(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
185 			struct compound_state *);
186 static void	rfs4_op_link(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
187 			struct compound_state *);
188 static void	rfs4_op_lock(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
189 			struct compound_state *);
190 static void	lock_denied_free(nfs_resop4 *);
191 static void	rfs4_op_locku(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
192 			struct compound_state *);
193 static void	rfs4_op_lockt(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
194 			struct compound_state *);
195 static void	rfs4_op_lookup(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
196 			struct compound_state *);
197 static void	rfs4_op_lookupp(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
198 			struct compound_state *);
199 static void	rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop,
200 				struct svc_req *req, struct compound_state *cs);
201 static void	rfs4_op_nverify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
202 			struct compound_state *);
203 static void	rfs4_op_open(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
204 			struct compound_state *);
205 static void	rfs4_op_open_confirm(nfs_argop4 *, nfs_resop4 *,
206 			struct svc_req *, struct compound_state *);
207 static void	rfs4_op_open_downgrade(nfs_argop4 *, nfs_resop4 *,
208 			struct svc_req *, struct compound_state *);
209 static void	rfs4_op_putfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
210 			struct compound_state *);
211 static void	rfs4_op_putpubfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
212 			struct compound_state *);
213 static void	rfs4_op_putrootfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
214 			struct compound_state *);
215 static void	rfs4_op_read(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
216 			struct compound_state *);
217 static void	rfs4_op_read_free(nfs_resop4 *);
218 static void	rfs4_op_readdir_free(nfs_resop4 *resop);
219 static void	rfs4_op_readlink(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
220 			struct compound_state *);
221 static void	rfs4_op_readlink_free(nfs_resop4 *);
222 static void	rfs4_op_release_lockowner(nfs_argop4 *, nfs_resop4 *,
223 			struct svc_req *, struct compound_state *);
224 static void	rfs4_op_remove(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
225 			struct compound_state *);
226 static void	rfs4_op_rename(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
227 			struct compound_state *);
228 static void	rfs4_op_renew(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
229 			struct compound_state *);
230 static void	rfs4_op_restorefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
231 			struct compound_state *);
232 static void	rfs4_op_savefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
233 			struct compound_state *);
234 static void	rfs4_op_setattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
235 			struct compound_state *);
236 static void	rfs4_op_verify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
237 			struct compound_state *);
238 static void	rfs4_op_write(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
239 			struct compound_state *);
240 static void	rfs4_op_setclientid(nfs_argop4 *, nfs_resop4 *,
241 			struct svc_req *, struct compound_state *);
242 static void	rfs4_op_setclientid_confirm(nfs_argop4 *, nfs_resop4 *,
243 			struct svc_req *req, struct compound_state *);
244 static void	rfs4_op_secinfo(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
245 			struct compound_state *);
246 static void	rfs4_op_secinfo_free(nfs_resop4 *);
247 
248 static nfsstat4 check_open_access(uint32_t,
249 				struct compound_state *, struct svc_req *);
250 nfsstat4 rfs4_client_sysid(rfs4_client_t *, sysid_t *);
251 void rfs4_ss_clid(rfs4_client_t *);
252 
253 /*
254  * translation table for attrs
255  */
256 struct nfs4_ntov_table {
257 	union nfs4_attr_u *na;
258 	uint8_t amap[NFS4_MAXNUM_ATTRS];
259 	int attrcnt;
260 	bool_t vfsstat;
261 };
262 
263 static void	nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp);
264 static void	nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
265 				    struct nfs4_svgetit_arg *sargp);
266 
267 static nfsstat4	do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp,
268 		    struct compound_state *cs, struct nfs4_svgetit_arg *sargp,
269 		    struct nfs4_ntov_table *ntovp, nfs4_attr_cmd_t cmd);
270 
271 fem_t		*deleg_rdops;
272 fem_t		*deleg_wrops;
273 
274 rfs4_servinst_t *rfs4_cur_servinst = NULL;	/* current server instance */
275 kmutex_t	rfs4_servinst_lock;	/* protects linked list */
276 int		rfs4_seen_first_compound;	/* set first time we see one */
277 
278 /*
279  * NFS4 op dispatch table
280  */
281 
282 struct rfsv4disp {
283 	void	(*dis_proc)();		/* proc to call */
284 	void	(*dis_resfree)();	/* frees space allocated by proc */
285 	int	dis_flags;		/* RPC_IDEMPOTENT, etc... */
286 };
287 
288 static struct rfsv4disp rfsv4disptab[] = {
289 	/*
290 	 * NFS VERSION 4
291 	 */
292 
293 	/* RFS_NULL = 0 */
294 	{rfs4_op_illegal, nullfree, 0},
295 
296 	/* UNUSED = 1 */
297 	{rfs4_op_illegal, nullfree, 0},
298 
299 	/* UNUSED = 2 */
300 	{rfs4_op_illegal, nullfree, 0},
301 
302 	/* OP_ACCESS = 3 */
303 	{rfs4_op_access, nullfree, RPC_IDEMPOTENT},
304 
305 	/* OP_CLOSE = 4 */
306 	{rfs4_op_close, nullfree, 0},
307 
308 	/* OP_COMMIT = 5 */
309 	{rfs4_op_commit, nullfree, RPC_IDEMPOTENT},
310 
311 	/* OP_CREATE = 6 */
312 	{rfs4_op_create, nullfree, 0},
313 
314 	/* OP_DELEGPURGE = 7 */
315 	{rfs4_op_delegpurge, nullfree, 0},
316 
317 	/* OP_DELEGRETURN = 8 */
318 	{rfs4_op_delegreturn, nullfree, 0},
319 
320 	/* OP_GETATTR = 9 */
321 	{rfs4_op_getattr, rfs4_op_getattr_free, RPC_IDEMPOTENT},
322 
323 	/* OP_GETFH = 10 */
324 	{rfs4_op_getfh, rfs4_op_getfh_free, RPC_ALL},
325 
326 	/* OP_LINK = 11 */
327 	{rfs4_op_link, nullfree, 0},
328 
329 	/* OP_LOCK = 12 */
330 	{rfs4_op_lock, lock_denied_free, 0},
331 
332 	/* OP_LOCKT = 13 */
333 	{rfs4_op_lockt, lock_denied_free, 0},
334 
335 	/* OP_LOCKU = 14 */
336 	{rfs4_op_locku, nullfree, 0},
337 
338 	/* OP_LOOKUP = 15 */
339 	{rfs4_op_lookup, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
340 
341 	/* OP_LOOKUPP = 16 */
342 	{rfs4_op_lookupp, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
343 
344 	/* OP_NVERIFY = 17 */
345 	{rfs4_op_nverify, nullfree, RPC_IDEMPOTENT},
346 
347 	/* OP_OPEN = 18 */
348 	{rfs4_op_open, rfs4_free_reply, 0},
349 
350 	/* OP_OPENATTR = 19 */
351 	{rfs4_op_openattr, nullfree, 0},
352 
353 	/* OP_OPEN_CONFIRM = 20 */
354 	{rfs4_op_open_confirm, nullfree, 0},
355 
356 	/* OP_OPEN_DOWNGRADE = 21 */
357 	{rfs4_op_open_downgrade, nullfree, 0},
358 
359 	/* OP_OPEN_PUTFH = 22 */
360 	{rfs4_op_putfh, nullfree, RPC_ALL},
361 
362 	/* OP_PUTPUBFH = 23 */
363 	{rfs4_op_putpubfh, nullfree, RPC_ALL},
364 
365 	/* OP_PUTROOTFH = 24 */
366 	{rfs4_op_putrootfh, nullfree, RPC_ALL},
367 
368 	/* OP_READ = 25 */
369 	{rfs4_op_read, rfs4_op_read_free, RPC_IDEMPOTENT},
370 
371 	/* OP_READDIR = 26 */
372 	{rfs4_op_readdir, rfs4_op_readdir_free, RPC_IDEMPOTENT},
373 
374 	/* OP_READLINK = 27 */
375 	{rfs4_op_readlink, rfs4_op_readlink_free, RPC_IDEMPOTENT},
376 
377 	/* OP_REMOVE = 28 */
378 	{rfs4_op_remove, nullfree, 0},
379 
380 	/* OP_RENAME = 29 */
381 	{rfs4_op_rename, nullfree, 0},
382 
383 	/* OP_RENEW = 30 */
384 	{rfs4_op_renew, nullfree, 0},
385 
386 	/* OP_RESTOREFH = 31 */
387 	{rfs4_op_restorefh, nullfree, RPC_ALL},
388 
389 	/* OP_SAVEFH = 32 */
390 	{rfs4_op_savefh, nullfree, RPC_ALL},
391 
392 	/* OP_SECINFO = 33 */
393 	{rfs4_op_secinfo, rfs4_op_secinfo_free, 0},
394 
395 	/* OP_SETATTR = 34 */
396 	{rfs4_op_setattr, nullfree, 0},
397 
398 	/* OP_SETCLIENTID = 35 */
399 	{rfs4_op_setclientid, nullfree, 0},
400 
401 	/* OP_SETCLIENTID_CONFIRM = 36 */
402 	{rfs4_op_setclientid_confirm, nullfree, 0},
403 
404 	/* OP_VERIFY = 37 */
405 	{rfs4_op_verify, nullfree, RPC_IDEMPOTENT},
406 
407 	/* OP_WRITE = 38 */
408 	{rfs4_op_write, nullfree, 0},
409 
410 	/* OP_RELEASE_LOCKOWNER = 39 */
411 	{rfs4_op_release_lockowner, nullfree, 0},
412 };
413 
414 static uint_t rfsv4disp_cnt = sizeof (rfsv4disptab) / sizeof (rfsv4disptab[0]);
415 
416 #define	OP_ILLEGAL_IDX (rfsv4disp_cnt)
417 
418 #ifdef DEBUG
419 
420 int		rfs4_fillone_debug = 0;
421 int		rfs4_no_stub_access = 1;
422 int		rfs4_rddir_debug = 0;
423 
424 static char    *rfs4_op_string[] = {
425 	"rfs4_op_null",
426 	"rfs4_op_1 unused",
427 	"rfs4_op_2 unused",
428 	"rfs4_op_access",
429 	"rfs4_op_close",
430 	"rfs4_op_commit",
431 	"rfs4_op_create",
432 	"rfs4_op_delegpurge",
433 	"rfs4_op_delegreturn",
434 	"rfs4_op_getattr",
435 	"rfs4_op_getfh",
436 	"rfs4_op_link",
437 	"rfs4_op_lock",
438 	"rfs4_op_lockt",
439 	"rfs4_op_locku",
440 	"rfs4_op_lookup",
441 	"rfs4_op_lookupp",
442 	"rfs4_op_nverify",
443 	"rfs4_op_open",
444 	"rfs4_op_openattr",
445 	"rfs4_op_open_confirm",
446 	"rfs4_op_open_downgrade",
447 	"rfs4_op_putfh",
448 	"rfs4_op_putpubfh",
449 	"rfs4_op_putrootfh",
450 	"rfs4_op_read",
451 	"rfs4_op_readdir",
452 	"rfs4_op_readlink",
453 	"rfs4_op_remove",
454 	"rfs4_op_rename",
455 	"rfs4_op_renew",
456 	"rfs4_op_restorefh",
457 	"rfs4_op_savefh",
458 	"rfs4_op_secinfo",
459 	"rfs4_op_setattr",
460 	"rfs4_op_setclientid",
461 	"rfs4_op_setclient_confirm",
462 	"rfs4_op_verify",
463 	"rfs4_op_write",
464 	"rfs4_op_release_lockowner",
465 	"rfs4_op_illegal"
466 };
467 #endif
468 
469 void	rfs4_ss_chkclid(rfs4_client_t *);
470 
471 extern size_t   strlcpy(char *dst, const char *src, size_t dstsize);
472 
473 extern void	rfs4_free_fs_locations4(fs_locations4 *);
474 
475 #ifdef	nextdp
476 #undef nextdp
477 #endif
478 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
479 
480 static const fs_operation_def_t nfs4_rd_deleg_tmpl[] = {
481 	VOPNAME_OPEN,		{ .femop_open = deleg_rd_open },
482 	VOPNAME_WRITE,		{ .femop_write = deleg_rd_write },
483 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_rd_setattr },
484 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_rd_rwlock },
485 	VOPNAME_SPACE,		{ .femop_space = deleg_rd_space },
486 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_rd_setsecattr },
487 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_rd_vnevent },
488 	NULL,			NULL
489 };
490 static const fs_operation_def_t nfs4_wr_deleg_tmpl[] = {
491 	VOPNAME_OPEN,		{ .femop_open = deleg_wr_open },
492 	VOPNAME_READ,		{ .femop_read = deleg_wr_read },
493 	VOPNAME_WRITE,		{ .femop_write = deleg_wr_write },
494 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_wr_setattr },
495 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_wr_rwlock },
496 	VOPNAME_SPACE,		{ .femop_space = deleg_wr_space },
497 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_wr_setsecattr },
498 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_wr_vnevent },
499 	NULL,			NULL
500 };
501 
502 int
503 rfs4_srvrinit(void)
504 {
505 	timespec32_t verf;
506 	int error;
507 	extern void rfs4_attr_init();
508 	extern krwlock_t rfs4_deleg_policy_lock;
509 
510 	/*
511 	 * The following algorithm attempts to find a unique verifier
512 	 * to be used as the write verifier returned from the server
513 	 * to the client.  It is important that this verifier change
514 	 * whenever the server reboots.  Of secondary importance, it
515 	 * is important for the verifier to be unique between two
516 	 * different servers.
517 	 *
518 	 * Thus, an attempt is made to use the system hostid and the
519 	 * current time in seconds when the nfssrv kernel module is
520 	 * loaded.  It is assumed that an NFS server will not be able
521 	 * to boot and then to reboot in less than a second.  If the
522 	 * hostid has not been set, then the current high resolution
523 	 * time is used.  This will ensure different verifiers each
524 	 * time the server reboots and minimize the chances that two
525 	 * different servers will have the same verifier.
526 	 * XXX - this is broken on LP64 kernels.
527 	 */
528 	verf.tv_sec = (time_t)zone_get_hostid(NULL);
529 	if (verf.tv_sec != 0) {
530 		verf.tv_nsec = gethrestime_sec();
531 	} else {
532 		timespec_t tverf;
533 
534 		gethrestime(&tverf);
535 		verf.tv_sec = (time_t)tverf.tv_sec;
536 		verf.tv_nsec = tverf.tv_nsec;
537 	}
538 
539 	Write4verf = *(uint64_t *)&verf;
540 
541 	rfs4_attr_init();
542 	mutex_init(&rfs4_deleg_lock, NULL, MUTEX_DEFAULT, NULL);
543 
544 	/* Used to manage create/destroy of server state */
545 	mutex_init(&rfs4_state_lock, NULL, MUTEX_DEFAULT, NULL);
546 
547 	/* Used to manage access to server instance linked list */
548 	mutex_init(&rfs4_servinst_lock, NULL, MUTEX_DEFAULT, NULL);
549 
550 	/* Used to manage access to rfs4_deleg_policy */
551 	rw_init(&rfs4_deleg_policy_lock, NULL, RW_DEFAULT, NULL);
552 
553 	error = fem_create("deleg_rdops", nfs4_rd_deleg_tmpl, &deleg_rdops);
554 	if (error != 0) {
555 		rfs4_disable_delegation();
556 	} else {
557 		error = fem_create("deleg_wrops", nfs4_wr_deleg_tmpl,
558 		    &deleg_wrops);
559 		if (error != 0) {
560 			rfs4_disable_delegation();
561 			fem_free(deleg_rdops);
562 		}
563 	}
564 
565 	nfs4_srv_caller_id = fs_new_caller_id();
566 
567 	lockt_sysid = lm_alloc_sysidt();
568 
569 	vsd_create(&nfs4_srv_vkey, NULL);
570 
571 	return (0);
572 }
573 
574 void
575 rfs4_srvrfini(void)
576 {
577 	extern krwlock_t rfs4_deleg_policy_lock;
578 
579 	if (lockt_sysid != LM_NOSYSID) {
580 		lm_free_sysidt(lockt_sysid);
581 		lockt_sysid = LM_NOSYSID;
582 	}
583 
584 	mutex_destroy(&rfs4_deleg_lock);
585 	mutex_destroy(&rfs4_state_lock);
586 	rw_destroy(&rfs4_deleg_policy_lock);
587 
588 	fem_free(deleg_rdops);
589 	fem_free(deleg_wrops);
590 }
591 
592 void
593 rfs4_init_compound_state(struct compound_state *cs)
594 {
595 	bzero(cs, sizeof (*cs));
596 	cs->cont = TRUE;
597 	cs->access = CS_ACCESS_DENIED;
598 	cs->deleg = FALSE;
599 	cs->mandlock = FALSE;
600 	cs->fh.nfs_fh4_val = cs->fhbuf;
601 }
602 
603 void
604 rfs4_grace_start(rfs4_servinst_t *sip)
605 {
606 	rw_enter(&sip->rwlock, RW_WRITER);
607 	sip->start_time = (time_t)TICK_TO_SEC(ddi_get_lbolt());
608 	sip->grace_period = rfs4_grace_period;
609 	rw_exit(&sip->rwlock);
610 }
611 
612 /*
613  * returns true if the instance's grace period has never been started
614  */
615 int
616 rfs4_servinst_grace_new(rfs4_servinst_t *sip)
617 {
618 	time_t start_time;
619 
620 	rw_enter(&sip->rwlock, RW_READER);
621 	start_time = sip->start_time;
622 	rw_exit(&sip->rwlock);
623 
624 	return (start_time == 0);
625 }
626 
627 /*
628  * Indicates if server instance is within the
629  * grace period.
630  */
631 int
632 rfs4_servinst_in_grace(rfs4_servinst_t *sip)
633 {
634 	time_t grace_expiry;
635 
636 	rw_enter(&sip->rwlock, RW_READER);
637 	grace_expiry = sip->start_time + sip->grace_period;
638 	rw_exit(&sip->rwlock);
639 
640 	return (((time_t)TICK_TO_SEC(ddi_get_lbolt())) < grace_expiry);
641 }
642 
643 int
644 rfs4_clnt_in_grace(rfs4_client_t *cp)
645 {
646 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
647 
648 	return (rfs4_servinst_in_grace(cp->rc_server_instance));
649 }
650 
651 /*
652  * reset all currently active grace periods
653  */
654 void
655 rfs4_grace_reset_all(void)
656 {
657 	rfs4_servinst_t *sip;
658 
659 	mutex_enter(&rfs4_servinst_lock);
660 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
661 		if (rfs4_servinst_in_grace(sip))
662 			rfs4_grace_start(sip);
663 	mutex_exit(&rfs4_servinst_lock);
664 }
665 
666 /*
667  * start any new instances' grace periods
668  */
669 void
670 rfs4_grace_start_new(void)
671 {
672 	rfs4_servinst_t *sip;
673 
674 	mutex_enter(&rfs4_servinst_lock);
675 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
676 		if (rfs4_servinst_grace_new(sip))
677 			rfs4_grace_start(sip);
678 	mutex_exit(&rfs4_servinst_lock);
679 }
680 
681 static rfs4_dss_path_t *
682 rfs4_dss_newpath(rfs4_servinst_t *sip, char *path, unsigned index)
683 {
684 	size_t len;
685 	rfs4_dss_path_t *dss_path;
686 
687 	dss_path = kmem_alloc(sizeof (rfs4_dss_path_t), KM_SLEEP);
688 
689 	/*
690 	 * Take a copy of the string, since the original may be overwritten.
691 	 * Sadly, no strdup() in the kernel.
692 	 */
693 	/* allow for NUL */
694 	len = strlen(path) + 1;
695 	dss_path->path = kmem_alloc(len, KM_SLEEP);
696 	(void) strlcpy(dss_path->path, path, len);
697 
698 	/* associate with servinst */
699 	dss_path->sip = sip;
700 	dss_path->index = index;
701 
702 	/*
703 	 * Add to list of served paths.
704 	 * No locking required, as we're only ever called at startup.
705 	 */
706 	if (rfs4_dss_pathlist == NULL) {
707 		/* this is the first dss_path_t */
708 
709 		/* needed for insque/remque */
710 		dss_path->next = dss_path->prev = dss_path;
711 
712 		rfs4_dss_pathlist = dss_path;
713 	} else {
714 		insque(dss_path, rfs4_dss_pathlist);
715 	}
716 
717 	return (dss_path);
718 }
719 
720 /*
721  * Create a new server instance, and make it the currently active instance.
722  * Note that starting the grace period too early will reduce the clients'
723  * recovery window.
724  */
725 void
726 rfs4_servinst_create(int start_grace, int dss_npaths, char **dss_paths)
727 {
728 	unsigned i;
729 	rfs4_servinst_t *sip;
730 	rfs4_oldstate_t *oldstate;
731 
732 	sip = kmem_alloc(sizeof (rfs4_servinst_t), KM_SLEEP);
733 	rw_init(&sip->rwlock, NULL, RW_DEFAULT, NULL);
734 
735 	sip->start_time = (time_t)0;
736 	sip->grace_period = (time_t)0;
737 	sip->next = NULL;
738 	sip->prev = NULL;
739 
740 	rw_init(&sip->oldstate_lock, NULL, RW_DEFAULT, NULL);
741 	/*
742 	 * This initial dummy entry is required to setup for insque/remque.
743 	 * It must be skipped over whenever the list is traversed.
744 	 */
745 	oldstate = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
746 	/* insque/remque require initial list entry to be self-terminated */
747 	oldstate->next = oldstate;
748 	oldstate->prev = oldstate;
749 	sip->oldstate = oldstate;
750 
751 
752 	sip->dss_npaths = dss_npaths;
753 	sip->dss_paths = kmem_alloc(dss_npaths *
754 	    sizeof (rfs4_dss_path_t *), KM_SLEEP);
755 
756 	for (i = 0; i < dss_npaths; i++) {
757 		sip->dss_paths[i] = rfs4_dss_newpath(sip, dss_paths[i], i);
758 	}
759 
760 	mutex_enter(&rfs4_servinst_lock);
761 	if (rfs4_cur_servinst != NULL) {
762 		/* add to linked list */
763 		sip->prev = rfs4_cur_servinst;
764 		rfs4_cur_servinst->next = sip;
765 	}
766 	if (start_grace)
767 		rfs4_grace_start(sip);
768 	/* make the new instance "current" */
769 	rfs4_cur_servinst = sip;
770 
771 	mutex_exit(&rfs4_servinst_lock);
772 }
773 
774 /*
775  * In future, we might add a rfs4_servinst_destroy(sip) but, for now, destroy
776  * all instances directly.
777  */
778 void
779 rfs4_servinst_destroy_all(void)
780 {
781 	rfs4_servinst_t *sip, *prev, *current;
782 #ifdef DEBUG
783 	int n = 0;
784 #endif
785 
786 	mutex_enter(&rfs4_servinst_lock);
787 	ASSERT(rfs4_cur_servinst != NULL);
788 	current = rfs4_cur_servinst;
789 	rfs4_cur_servinst = NULL;
790 	for (sip = current; sip != NULL; sip = prev) {
791 		prev = sip->prev;
792 		rw_destroy(&sip->rwlock);
793 		if (sip->oldstate)
794 			kmem_free(sip->oldstate, sizeof (rfs4_oldstate_t));
795 		if (sip->dss_paths)
796 			kmem_free(sip->dss_paths,
797 			    sip->dss_npaths * sizeof (rfs4_dss_path_t *));
798 		kmem_free(sip, sizeof (rfs4_servinst_t));
799 #ifdef DEBUG
800 		n++;
801 #endif
802 	}
803 	mutex_exit(&rfs4_servinst_lock);
804 }
805 
806 /*
807  * Assign the current server instance to a client_t.
808  * Should be called with cp->rc_dbe held.
809  */
810 void
811 rfs4_servinst_assign(rfs4_client_t *cp, rfs4_servinst_t *sip)
812 {
813 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
814 
815 	/*
816 	 * The lock ensures that if the current instance is in the process
817 	 * of changing, we will see the new one.
818 	 */
819 	mutex_enter(&rfs4_servinst_lock);
820 	cp->rc_server_instance = sip;
821 	mutex_exit(&rfs4_servinst_lock);
822 }
823 
824 rfs4_servinst_t *
825 rfs4_servinst(rfs4_client_t *cp)
826 {
827 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
828 
829 	return (cp->rc_server_instance);
830 }
831 
832 /* ARGSUSED */
833 static void
834 nullfree(caddr_t resop)
835 {
836 }
837 
838 /*
839  * This is a fall-through for invalid or not implemented (yet) ops
840  */
841 /* ARGSUSED */
842 static void
843 rfs4_op_inval(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
844 	struct compound_state *cs)
845 {
846 	*cs->statusp = *((nfsstat4 *)&(resop)->nfs_resop4_u) = NFS4ERR_INVAL;
847 }
848 
849 /*
850  * Check if the security flavor, nfsnum, is in the flavor_list.
851  */
852 bool_t
853 in_flavor_list(int nfsnum, int *flavor_list, int count)
854 {
855 	int i;
856 
857 	for (i = 0; i < count; i++) {
858 		if (nfsnum == flavor_list[i])
859 			return (TRUE);
860 	}
861 	return (FALSE);
862 }
863 
864 /*
865  * Used by rfs4_op_secinfo to get the security information from the
866  * export structure associated with the component.
867  */
868 /* ARGSUSED */
869 static nfsstat4
870 do_rfs4_op_secinfo(struct compound_state *cs, char *nm, SECINFO4res *resp)
871 {
872 	int error, different_export = 0;
873 	vnode_t *dvp, *vp;
874 	struct exportinfo *exi = NULL;
875 	struct exportinfo *oexi = NULL;
876 	fid_t fid;
877 	uint_t count, i;
878 	secinfo4 *resok_val;
879 	struct secinfo *secp;
880 	seconfig_t *si;
881 	bool_t did_traverse = FALSE;
882 	int dotdot, walk;
883 
884 	dvp = cs->vp;
885 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
886 
887 	/*
888 	 * If dotdotting, then need to check whether it's above the
889 	 * root of a filesystem, or above an export point.
890 	 */
891 	if (dotdot) {
892 
893 		/*
894 		 * If dotdotting at the root of a filesystem, then
895 		 * need to traverse back to the mounted-on filesystem
896 		 * and do the dotdot lookup there.
897 		 */
898 		if (cs->vp->v_flag & VROOT) {
899 
900 			/*
901 			 * If at the system root, then can
902 			 * go up no further.
903 			 */
904 			if (VN_CMP(dvp, rootdir))
905 				return (puterrno4(ENOENT));
906 
907 			/*
908 			 * Traverse back to the mounted-on filesystem
909 			 */
910 			dvp = untraverse(cs->vp);
911 
912 			/*
913 			 * Set the different_export flag so we remember
914 			 * to pick up a new exportinfo entry for
915 			 * this new filesystem.
916 			 */
917 			different_export = 1;
918 		} else {
919 
920 			/*
921 			 * If dotdotting above an export point then set
922 			 * the different_export to get new export info.
923 			 */
924 			different_export = nfs_exported(cs->exi, cs->vp);
925 		}
926 	}
927 
928 	/*
929 	 * Get the vnode for the component "nm".
930 	 */
931 	error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cs->cr,
932 	    NULL, NULL, NULL);
933 	if (error)
934 		return (puterrno4(error));
935 
936 	/*
937 	 * If the vnode is in a pseudo filesystem, or if the security flavor
938 	 * used in the request is valid but not an explicitly shared flavor,
939 	 * or the access bit indicates that this is a limited access,
940 	 * check whether this vnode is visible.
941 	 */
942 	if (!different_export &&
943 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
944 	    cs->access & CS_ACCESS_LIMITED)) {
945 		if (! nfs_visible(cs->exi, vp, &different_export)) {
946 			VN_RELE(vp);
947 			return (puterrno4(ENOENT));
948 		}
949 	}
950 
951 	/*
952 	 * If it's a mountpoint, then traverse it.
953 	 */
954 	if (vn_ismntpt(vp)) {
955 		if ((error = traverse(&vp)) != 0) {
956 			VN_RELE(vp);
957 			return (puterrno4(error));
958 		}
959 		/* remember that we had to traverse mountpoint */
960 		did_traverse = TRUE;
961 		different_export = 1;
962 	} else if (vp->v_vfsp != dvp->v_vfsp) {
963 		/*
964 		 * If vp isn't a mountpoint and the vfs ptrs aren't the same,
965 		 * then vp is probably an LOFS object.  We don't need the
966 		 * realvp, we just need to know that we might have crossed
967 		 * a server fs boundary and need to call checkexport.
968 		 * (LOFS lookup hides server fs mountpoints, and actually calls
969 		 * traverse)
970 		 */
971 		different_export = 1;
972 	}
973 
974 	/*
975 	 * Get the export information for it.
976 	 */
977 	if (different_export) {
978 
979 		bzero(&fid, sizeof (fid));
980 		fid.fid_len = MAXFIDSZ;
981 		error = vop_fid_pseudo(vp, &fid);
982 		if (error) {
983 			VN_RELE(vp);
984 			return (puterrno4(error));
985 		}
986 
987 		if (dotdot)
988 			oexi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
989 		else
990 			oexi = checkexport(&vp->v_vfsp->vfs_fsid, &fid, vp);
991 
992 		if (oexi == NULL) {
993 			if (did_traverse == TRUE) {
994 				/*
995 				 * If this vnode is a mounted-on vnode,
996 				 * but the mounted-on file system is not
997 				 * exported, send back the secinfo for
998 				 * the exported node that the mounted-on
999 				 * vnode lives in.
1000 				 */
1001 				exi = cs->exi;
1002 			} else {
1003 				VN_RELE(vp);
1004 				return (puterrno4(EACCES));
1005 			}
1006 		} else {
1007 			exi = oexi;
1008 		}
1009 	} else {
1010 		exi = cs->exi;
1011 	}
1012 	ASSERT(exi != NULL);
1013 
1014 
1015 	/*
1016 	 * Create the secinfo result based on the security information
1017 	 * from the exportinfo structure (exi).
1018 	 *
1019 	 * Return all flavors for a pseudo node.
1020 	 * For a real export node, return the flavor that the client
1021 	 * has access with.
1022 	 */
1023 	rw_enter(&exported_lock, RW_READER);
1024 	if (PSEUDO(exi)) {
1025 		count = exi->exi_export.ex_seccnt; /* total sec count */
1026 		resok_val = kmem_alloc(count * sizeof (secinfo4), KM_SLEEP);
1027 		secp = exi->exi_export.ex_secinfo;
1028 
1029 		for (i = 0; i < count; i++) {
1030 			si = &secp[i].s_secinfo;
1031 			resok_val[i].flavor = si->sc_rpcnum;
1032 			if (resok_val[i].flavor == RPCSEC_GSS) {
1033 				rpcsec_gss_info *info;
1034 
1035 				info = &resok_val[i].flavor_info;
1036 				info->qop = si->sc_qop;
1037 				info->service = (rpc_gss_svc_t)si->sc_service;
1038 
1039 				/* get oid opaque data */
1040 				info->oid.sec_oid4_len =
1041 				    si->sc_gss_mech_type->length;
1042 				info->oid.sec_oid4_val = kmem_alloc(
1043 				    si->sc_gss_mech_type->length, KM_SLEEP);
1044 				bcopy(
1045 				    si->sc_gss_mech_type->elements,
1046 				    info->oid.sec_oid4_val,
1047 				    info->oid.sec_oid4_len);
1048 			}
1049 		}
1050 		resp->SECINFO4resok_len = count;
1051 		resp->SECINFO4resok_val = resok_val;
1052 	} else {
1053 		int ret_cnt = 0, k = 0;
1054 		int *flavor_list;
1055 
1056 		count = exi->exi_export.ex_seccnt; /* total sec count */
1057 		secp = exi->exi_export.ex_secinfo;
1058 
1059 		flavor_list = kmem_alloc(count * sizeof (int), KM_SLEEP);
1060 		/* find out which flavors to return */
1061 		for (i = 0; i < count; i ++) {
1062 			int access, flavor, perm;
1063 
1064 			flavor = secp[i].s_secinfo.sc_nfsnum;
1065 			perm = secp[i].s_flags;
1066 
1067 			access = nfsauth4_secinfo_access(exi, cs->req,
1068 			    flavor, perm, cs->basecr);
1069 
1070 			if (! (access & NFSAUTH_DENIED) &&
1071 			    ! (access & NFSAUTH_WRONGSEC)) {
1072 				flavor_list[ret_cnt] = flavor;
1073 				ret_cnt++;
1074 			}
1075 		}
1076 
1077 		/* Create the returning SECINFO value */
1078 		resok_val = kmem_alloc(ret_cnt * sizeof (secinfo4), KM_SLEEP);
1079 
1080 		for (i = 0; i < count; i++) {
1081 			/*
1082 			 * If the flavor is in the flavor list,
1083 			 * fill in resok_val.
1084 			 */
1085 			si = &secp[i].s_secinfo;
1086 			if (in_flavor_list(si->sc_nfsnum,
1087 			    flavor_list, ret_cnt)) {
1088 				resok_val[k].flavor = si->sc_rpcnum;
1089 				if (resok_val[k].flavor == RPCSEC_GSS) {
1090 					rpcsec_gss_info *info;
1091 
1092 					info = &resok_val[k].flavor_info;
1093 					info->qop = si->sc_qop;
1094 					info->service = (rpc_gss_svc_t)
1095 					    si->sc_service;
1096 
1097 					/* get oid opaque data */
1098 					info->oid.sec_oid4_len =
1099 					    si->sc_gss_mech_type->length;
1100 					info->oid.sec_oid4_val = kmem_alloc(
1101 					    si->sc_gss_mech_type->length,
1102 					    KM_SLEEP);
1103 					bcopy(si->sc_gss_mech_type->elements,
1104 					    info->oid.sec_oid4_val,
1105 					    info->oid.sec_oid4_len);
1106 				}
1107 				k++;
1108 			}
1109 			if (k >= ret_cnt)
1110 				break;
1111 		}
1112 		resp->SECINFO4resok_len = ret_cnt;
1113 		resp->SECINFO4resok_val = resok_val;
1114 		kmem_free(flavor_list, count * sizeof (int));
1115 	}
1116 	rw_exit(&exported_lock);
1117 	if (oexi)
1118 		exi_rele(oexi);
1119 	VN_RELE(vp);
1120 	return (NFS4_OK);
1121 }
1122 
1123 /*
1124  * SECINFO (Operation 33): Obtain required security information on
1125  * the component name in the format of (security-mechanism-oid, qop, service)
1126  * triplets.
1127  */
1128 /* ARGSUSED */
1129 static void
1130 rfs4_op_secinfo(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1131     struct compound_state *cs)
1132 {
1133 	SECINFO4args *args = &argop->nfs_argop4_u.opsecinfo;
1134 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1135 	utf8string *utfnm = &args->name;
1136 	uint_t len;
1137 	char *nm;
1138 	struct sockaddr *ca;
1139 	char *name = NULL;
1140 	nfsstat4 status = NFS4_OK;
1141 
1142 	DTRACE_NFSV4_2(op__secinfo__start, struct compound_state *, cs,
1143 	    SECINFO4args *, args);
1144 
1145 	/*
1146 	 * Current file handle (cfh) should have been set before getting
1147 	 * into this function. If not, return error.
1148 	 */
1149 	if (cs->vp == NULL) {
1150 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1151 		goto out;
1152 	}
1153 
1154 	if (cs->vp->v_type != VDIR) {
1155 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1156 		goto out;
1157 	}
1158 
1159 	/*
1160 	 * Verify the component name. If failed, error out, but
1161 	 * do not error out if the component name is a "..".
1162 	 * SECINFO will return its parents secinfo data for SECINFO "..".
1163 	 */
1164 	status = utf8_dir_verify(utfnm);
1165 	if (status != NFS4_OK) {
1166 		if (utfnm->utf8string_len != 2 ||
1167 		    utfnm->utf8string_val[0] != '.' ||
1168 		    utfnm->utf8string_val[1] != '.') {
1169 			*cs->statusp = resp->status = status;
1170 			goto out;
1171 		}
1172 	}
1173 
1174 	nm = utf8_to_str(utfnm, &len, NULL);
1175 	if (nm == NULL) {
1176 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1177 		goto out;
1178 	}
1179 
1180 	if (len > MAXNAMELEN) {
1181 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1182 		kmem_free(nm, len);
1183 		goto out;
1184 	}
1185 
1186 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1187 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1188 	    MAXPATHLEN  + 1);
1189 
1190 	if (name == NULL) {
1191 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1192 		kmem_free(nm, len);
1193 		goto out;
1194 	}
1195 
1196 
1197 	*cs->statusp = resp->status = do_rfs4_op_secinfo(cs, name, resp);
1198 
1199 	if (name != nm)
1200 		kmem_free(name, MAXPATHLEN + 1);
1201 	kmem_free(nm, len);
1202 
1203 out:
1204 	DTRACE_NFSV4_2(op__secinfo__done, struct compound_state *, cs,
1205 	    SECINFO4res *, resp);
1206 }
1207 
1208 /*
1209  * Free SECINFO result.
1210  */
1211 /* ARGSUSED */
1212 static void
1213 rfs4_op_secinfo_free(nfs_resop4 *resop)
1214 {
1215 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1216 	int count, i;
1217 	secinfo4 *resok_val;
1218 
1219 	/* If this is not an Ok result, nothing to free. */
1220 	if (resp->status != NFS4_OK) {
1221 		return;
1222 	}
1223 
1224 	count = resp->SECINFO4resok_len;
1225 	resok_val = resp->SECINFO4resok_val;
1226 
1227 	for (i = 0; i < count; i++) {
1228 		if (resok_val[i].flavor == RPCSEC_GSS) {
1229 			rpcsec_gss_info *info;
1230 
1231 			info = &resok_val[i].flavor_info;
1232 			kmem_free(info->oid.sec_oid4_val,
1233 			    info->oid.sec_oid4_len);
1234 		}
1235 	}
1236 	kmem_free(resok_val, count * sizeof (secinfo4));
1237 	resp->SECINFO4resok_len = 0;
1238 	resp->SECINFO4resok_val = NULL;
1239 }
1240 
1241 /* ARGSUSED */
1242 static void
1243 rfs4_op_access(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1244     struct compound_state *cs)
1245 {
1246 	ACCESS4args *args = &argop->nfs_argop4_u.opaccess;
1247 	ACCESS4res *resp = &resop->nfs_resop4_u.opaccess;
1248 	int error;
1249 	vnode_t *vp;
1250 	struct vattr va;
1251 	int checkwriteperm;
1252 	cred_t *cr = cs->cr;
1253 	bslabel_t *clabel, *slabel;
1254 	ts_label_t *tslabel;
1255 	boolean_t admin_low_client;
1256 
1257 	DTRACE_NFSV4_2(op__access__start, struct compound_state *, cs,
1258 	    ACCESS4args *, args);
1259 
1260 #if 0	/* XXX allow access even if !cs->access. Eventually only pseudo fs */
1261 	if (cs->access == CS_ACCESS_DENIED) {
1262 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1263 		goto out;
1264 	}
1265 #endif
1266 	if (cs->vp == NULL) {
1267 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1268 		goto out;
1269 	}
1270 
1271 	ASSERT(cr != NULL);
1272 
1273 	vp = cs->vp;
1274 
1275 	/*
1276 	 * If the file system is exported read only, it is not appropriate
1277 	 * to check write permissions for regular files and directories.
1278 	 * Special files are interpreted by the client, so the underlying
1279 	 * permissions are sent back to the client for interpretation.
1280 	 */
1281 	if (rdonly4(req, cs) &&
1282 	    (vp->v_type == VREG || vp->v_type == VDIR))
1283 		checkwriteperm = 0;
1284 	else
1285 		checkwriteperm = 1;
1286 
1287 	/*
1288 	 * XXX
1289 	 * We need the mode so that we can correctly determine access
1290 	 * permissions relative to a mandatory lock file.  Access to
1291 	 * mandatory lock files is denied on the server, so it might
1292 	 * as well be reflected to the server during the open.
1293 	 */
1294 	va.va_mask = AT_MODE;
1295 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1296 	if (error) {
1297 		*cs->statusp = resp->status = puterrno4(error);
1298 		goto out;
1299 	}
1300 	resp->access = 0;
1301 	resp->supported = 0;
1302 
1303 	if (is_system_labeled()) {
1304 		ASSERT(req->rq_label != NULL);
1305 		clabel = req->rq_label;
1306 		DTRACE_PROBE2(tx__rfs4__log__info__opaccess__clabel, char *,
1307 		    "got client label from request(1)",
1308 		    struct svc_req *, req);
1309 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
1310 			if ((tslabel = nfs_getflabel(vp, cs->exi)) == NULL) {
1311 				*cs->statusp = resp->status = puterrno4(EACCES);
1312 				goto out;
1313 			}
1314 			slabel = label2bslabel(tslabel);
1315 			DTRACE_PROBE3(tx__rfs4__log__info__opaccess__slabel,
1316 			    char *, "got server label(1) for vp(2)",
1317 			    bslabel_t *, slabel, vnode_t *, vp);
1318 
1319 			admin_low_client = B_FALSE;
1320 		} else
1321 			admin_low_client = B_TRUE;
1322 	}
1323 
1324 	if (args->access & ACCESS4_READ) {
1325 		error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
1326 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1327 		    (!is_system_labeled() || admin_low_client ||
1328 		    bldominates(clabel, slabel)))
1329 			resp->access |= ACCESS4_READ;
1330 		resp->supported |= ACCESS4_READ;
1331 	}
1332 	if ((args->access & ACCESS4_LOOKUP) && vp->v_type == VDIR) {
1333 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1334 		if (!error && (!is_system_labeled() || admin_low_client ||
1335 		    bldominates(clabel, slabel)))
1336 			resp->access |= ACCESS4_LOOKUP;
1337 		resp->supported |= ACCESS4_LOOKUP;
1338 	}
1339 	if (checkwriteperm &&
1340 	    (args->access & (ACCESS4_MODIFY|ACCESS4_EXTEND))) {
1341 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1342 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1343 		    (!is_system_labeled() || admin_low_client ||
1344 		    blequal(clabel, slabel)))
1345 			resp->access |=
1346 			    (args->access & (ACCESS4_MODIFY | ACCESS4_EXTEND));
1347 		resp->supported |=
1348 		    resp->access & (ACCESS4_MODIFY | ACCESS4_EXTEND);
1349 	}
1350 
1351 	if (checkwriteperm &&
1352 	    (args->access & ACCESS4_DELETE) && vp->v_type == VDIR) {
1353 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1354 		if (!error && (!is_system_labeled() || admin_low_client ||
1355 		    blequal(clabel, slabel)))
1356 			resp->access |= ACCESS4_DELETE;
1357 		resp->supported |= ACCESS4_DELETE;
1358 	}
1359 	if (args->access & ACCESS4_EXECUTE && vp->v_type != VDIR) {
1360 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1361 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1362 		    (!is_system_labeled() || admin_low_client ||
1363 		    bldominates(clabel, slabel)))
1364 			resp->access |= ACCESS4_EXECUTE;
1365 		resp->supported |= ACCESS4_EXECUTE;
1366 	}
1367 
1368 	if (is_system_labeled() && !admin_low_client)
1369 		label_rele(tslabel);
1370 
1371 	*cs->statusp = resp->status = NFS4_OK;
1372 out:
1373 	DTRACE_NFSV4_2(op__access__done, struct compound_state *, cs,
1374 	    ACCESS4res *, resp);
1375 }
1376 
1377 /* ARGSUSED */
1378 static void
1379 rfs4_op_commit(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1380     struct compound_state *cs)
1381 {
1382 	COMMIT4args *args = &argop->nfs_argop4_u.opcommit;
1383 	COMMIT4res *resp = &resop->nfs_resop4_u.opcommit;
1384 	int error;
1385 	vnode_t *vp = cs->vp;
1386 	cred_t *cr = cs->cr;
1387 	vattr_t va;
1388 
1389 	DTRACE_NFSV4_2(op__commit__start, struct compound_state *, cs,
1390 	    COMMIT4args *, args);
1391 
1392 	if (vp == NULL) {
1393 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1394 		goto out;
1395 	}
1396 	if (cs->access == CS_ACCESS_DENIED) {
1397 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1398 		goto out;
1399 	}
1400 
1401 	if (args->offset + args->count < args->offset) {
1402 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1403 		goto out;
1404 	}
1405 
1406 	va.va_mask = AT_UID;
1407 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1408 
1409 	/*
1410 	 * If we can't get the attributes, then we can't do the
1411 	 * right access checking.  So, we'll fail the request.
1412 	 */
1413 	if (error) {
1414 		*cs->statusp = resp->status = puterrno4(error);
1415 		goto out;
1416 	}
1417 	if (rdonly4(req, cs)) {
1418 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1419 		goto out;
1420 	}
1421 
1422 	if (vp->v_type != VREG) {
1423 		if (vp->v_type == VDIR)
1424 			resp->status = NFS4ERR_ISDIR;
1425 		else
1426 			resp->status = NFS4ERR_INVAL;
1427 		*cs->statusp = resp->status;
1428 		goto out;
1429 	}
1430 
1431 	if (crgetuid(cr) != va.va_uid &&
1432 	    (error = VOP_ACCESS(vp, VWRITE, 0, cs->cr, NULL))) {
1433 		*cs->statusp = resp->status = puterrno4(error);
1434 		goto out;
1435 	}
1436 
1437 	error = VOP_FSYNC(vp, FSYNC, cr, NULL);
1438 
1439 	if (error) {
1440 		*cs->statusp = resp->status = puterrno4(error);
1441 		goto out;
1442 	}
1443 
1444 	*cs->statusp = resp->status = NFS4_OK;
1445 	resp->writeverf = Write4verf;
1446 out:
1447 	DTRACE_NFSV4_2(op__commit__done, struct compound_state *, cs,
1448 	    COMMIT4res *, resp);
1449 }
1450 
1451 /*
1452  * rfs4_op_mknod is called from rfs4_op_create after all initial verification
1453  * was completed. It does the nfsv4 create for special files.
1454  */
1455 /* ARGSUSED */
1456 static vnode_t *
1457 do_rfs4_op_mknod(CREATE4args *args, CREATE4res *resp, struct svc_req *req,
1458     struct compound_state *cs, vattr_t *vap, char *nm)
1459 {
1460 	int error;
1461 	cred_t *cr = cs->cr;
1462 	vnode_t *dvp = cs->vp;
1463 	vnode_t *vp = NULL;
1464 	int mode;
1465 	enum vcexcl excl;
1466 
1467 	switch (args->type) {
1468 	case NF4CHR:
1469 	case NF4BLK:
1470 		if (secpolicy_sys_devices(cr) != 0) {
1471 			*cs->statusp = resp->status = NFS4ERR_PERM;
1472 			return (NULL);
1473 		}
1474 		if (args->type == NF4CHR)
1475 			vap->va_type = VCHR;
1476 		else
1477 			vap->va_type = VBLK;
1478 		vap->va_rdev = makedevice(args->ftype4_u.devdata.specdata1,
1479 		    args->ftype4_u.devdata.specdata2);
1480 		vap->va_mask |= AT_RDEV;
1481 		break;
1482 	case NF4SOCK:
1483 		vap->va_type = VSOCK;
1484 		break;
1485 	case NF4FIFO:
1486 		vap->va_type = VFIFO;
1487 		break;
1488 	default:
1489 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1490 		return (NULL);
1491 	}
1492 
1493 	/*
1494 	 * Must specify the mode.
1495 	 */
1496 	if (!(vap->va_mask & AT_MODE)) {
1497 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1498 		return (NULL);
1499 	}
1500 
1501 	excl = EXCL;
1502 
1503 	mode = 0;
1504 
1505 	error = VOP_CREATE(dvp, nm, vap, excl, mode, &vp, cr, 0, NULL, NULL);
1506 	if (error) {
1507 		*cs->statusp = resp->status = puterrno4(error);
1508 		return (NULL);
1509 	}
1510 	return (vp);
1511 }
1512 
1513 /*
1514  * nfsv4 create is used to create non-regular files. For regular files,
1515  * use nfsv4 open.
1516  */
1517 /* ARGSUSED */
1518 static void
1519 rfs4_op_create(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1520     struct compound_state *cs)
1521 {
1522 	CREATE4args *args = &argop->nfs_argop4_u.opcreate;
1523 	CREATE4res *resp = &resop->nfs_resop4_u.opcreate;
1524 	int error;
1525 	struct vattr bva, iva, iva2, ava, *vap;
1526 	cred_t *cr = cs->cr;
1527 	vnode_t *dvp = cs->vp;
1528 	vnode_t *vp = NULL;
1529 	vnode_t *realvp;
1530 	char *nm, *lnm;
1531 	uint_t len, llen;
1532 	int syncval = 0;
1533 	struct nfs4_svgetit_arg sarg;
1534 	struct nfs4_ntov_table ntov;
1535 	struct statvfs64 sb;
1536 	nfsstat4 status;
1537 	struct sockaddr *ca;
1538 	char *name = NULL;
1539 	char *lname = NULL;
1540 
1541 	DTRACE_NFSV4_2(op__create__start, struct compound_state *, cs,
1542 	    CREATE4args *, args);
1543 
1544 	resp->attrset = 0;
1545 
1546 	if (dvp == NULL) {
1547 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1548 		goto out;
1549 	}
1550 
1551 	/*
1552 	 * If there is an unshared filesystem mounted on this vnode,
1553 	 * do not allow to create an object in this directory.
1554 	 */
1555 	if (vn_ismntpt(dvp)) {
1556 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1557 		goto out;
1558 	}
1559 
1560 	/* Verify that type is correct */
1561 	switch (args->type) {
1562 	case NF4LNK:
1563 	case NF4BLK:
1564 	case NF4CHR:
1565 	case NF4SOCK:
1566 	case NF4FIFO:
1567 	case NF4DIR:
1568 		break;
1569 	default:
1570 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1571 		goto out;
1572 	};
1573 
1574 	if (cs->access == CS_ACCESS_DENIED) {
1575 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1576 		goto out;
1577 	}
1578 	if (dvp->v_type != VDIR) {
1579 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1580 		goto out;
1581 	}
1582 	status = utf8_dir_verify(&args->objname);
1583 	if (status != NFS4_OK) {
1584 		*cs->statusp = resp->status = status;
1585 		goto out;
1586 	}
1587 
1588 	if (rdonly4(req, cs)) {
1589 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1590 		goto out;
1591 	}
1592 
1593 	/*
1594 	 * Name of newly created object
1595 	 */
1596 	nm = utf8_to_fn(&args->objname, &len, NULL);
1597 	if (nm == NULL) {
1598 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1599 		goto out;
1600 	}
1601 
1602 	if (len > MAXNAMELEN) {
1603 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1604 		kmem_free(nm, len);
1605 		goto out;
1606 	}
1607 
1608 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1609 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1610 	    MAXPATHLEN  + 1);
1611 
1612 	if (name == NULL) {
1613 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1614 		kmem_free(nm, len);
1615 		goto out;
1616 	}
1617 
1618 	resp->attrset = 0;
1619 
1620 	sarg.sbp = &sb;
1621 	sarg.is_referral = B_FALSE;
1622 	nfs4_ntov_table_init(&ntov);
1623 
1624 	status = do_rfs4_set_attrs(&resp->attrset,
1625 	    &args->createattrs, cs, &sarg, &ntov, NFS4ATTR_SETIT);
1626 
1627 	if (sarg.vap->va_mask == 0 && status == NFS4_OK)
1628 		status = NFS4ERR_INVAL;
1629 
1630 	if (status != NFS4_OK) {
1631 		*cs->statusp = resp->status = status;
1632 		if (name != nm)
1633 			kmem_free(name, MAXPATHLEN + 1);
1634 		kmem_free(nm, len);
1635 		nfs4_ntov_table_free(&ntov, &sarg);
1636 		resp->attrset = 0;
1637 		goto out;
1638 	}
1639 
1640 	/* Get "before" change value */
1641 	bva.va_mask = AT_CTIME|AT_SEQ|AT_MODE;
1642 	error = VOP_GETATTR(dvp, &bva, 0, cr, NULL);
1643 	if (error) {
1644 		*cs->statusp = resp->status = puterrno4(error);
1645 		if (name != nm)
1646 			kmem_free(name, MAXPATHLEN + 1);
1647 		kmem_free(nm, len);
1648 		nfs4_ntov_table_free(&ntov, &sarg);
1649 		resp->attrset = 0;
1650 		goto out;
1651 	}
1652 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bva.va_ctime)
1653 
1654 	vap = sarg.vap;
1655 
1656 	/*
1657 	 * Set the default initial values for attributes when the parent
1658 	 * directory does not have the VSUID/VSGID bit set and they have
1659 	 * not been specified in createattrs.
1660 	 */
1661 	if (!(bva.va_mode & VSUID) && (vap->va_mask & AT_UID) == 0) {
1662 		vap->va_uid = crgetuid(cr);
1663 		vap->va_mask |= AT_UID;
1664 	}
1665 	if (!(bva.va_mode & VSGID) && (vap->va_mask & AT_GID) == 0) {
1666 		vap->va_gid = crgetgid(cr);
1667 		vap->va_mask |= AT_GID;
1668 	}
1669 
1670 	vap->va_mask |= AT_TYPE;
1671 	switch (args->type) {
1672 	case NF4DIR:
1673 		vap->va_type = VDIR;
1674 		if ((vap->va_mask & AT_MODE) == 0) {
1675 			vap->va_mode = 0700;	/* default: owner rwx only */
1676 			vap->va_mask |= AT_MODE;
1677 		}
1678 		error = VOP_MKDIR(dvp, name, vap, &vp, cr, NULL, 0, NULL);
1679 		if (error)
1680 			break;
1681 
1682 		/*
1683 		 * Get the initial "after" sequence number, if it fails,
1684 		 * set to zero
1685 		 */
1686 		iva.va_mask = AT_SEQ;
1687 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1688 			iva.va_seq = 0;
1689 		break;
1690 	case NF4LNK:
1691 		vap->va_type = VLNK;
1692 		if ((vap->va_mask & AT_MODE) == 0) {
1693 			vap->va_mode = 0700;	/* default: owner rwx only */
1694 			vap->va_mask |= AT_MODE;
1695 		}
1696 
1697 		/*
1698 		 * symlink names must be treated as data
1699 		 */
1700 		lnm = utf8_to_str((utf8string *)&args->ftype4_u.linkdata,
1701 		    &llen, NULL);
1702 
1703 		if (lnm == NULL) {
1704 			*cs->statusp = resp->status = NFS4ERR_INVAL;
1705 			if (name != nm)
1706 				kmem_free(name, MAXPATHLEN + 1);
1707 			kmem_free(nm, len);
1708 			nfs4_ntov_table_free(&ntov, &sarg);
1709 			resp->attrset = 0;
1710 			goto out;
1711 		}
1712 
1713 		if (llen > MAXPATHLEN) {
1714 			*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1715 			if (name != nm)
1716 				kmem_free(name, MAXPATHLEN + 1);
1717 			kmem_free(nm, len);
1718 			kmem_free(lnm, llen);
1719 			nfs4_ntov_table_free(&ntov, &sarg);
1720 			resp->attrset = 0;
1721 			goto out;
1722 		}
1723 
1724 		lname = nfscmd_convname(ca, cs->exi, lnm,
1725 		    NFSCMD_CONV_INBOUND, MAXPATHLEN  + 1);
1726 
1727 		if (lname == NULL) {
1728 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
1729 			if (name != nm)
1730 				kmem_free(name, MAXPATHLEN + 1);
1731 			kmem_free(nm, len);
1732 			kmem_free(lnm, llen);
1733 			nfs4_ntov_table_free(&ntov, &sarg);
1734 			resp->attrset = 0;
1735 			goto out;
1736 		}
1737 
1738 		error = VOP_SYMLINK(dvp, name, vap, lname, cr, NULL, 0);
1739 		if (lname != lnm)
1740 			kmem_free(lname, MAXPATHLEN + 1);
1741 		kmem_free(lnm, llen);
1742 		if (error)
1743 			break;
1744 
1745 		/*
1746 		 * Get the initial "after" sequence number, if it fails,
1747 		 * set to zero
1748 		 */
1749 		iva.va_mask = AT_SEQ;
1750 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1751 			iva.va_seq = 0;
1752 
1753 		error = VOP_LOOKUP(dvp, name, &vp, NULL, 0, NULL, cr,
1754 		    NULL, NULL, NULL);
1755 		if (error)
1756 			break;
1757 
1758 		/*
1759 		 * va_seq is not safe over VOP calls, check it again
1760 		 * if it has changed zero out iva to force atomic = FALSE.
1761 		 */
1762 		iva2.va_mask = AT_SEQ;
1763 		if (VOP_GETATTR(dvp, &iva2, 0, cs->cr, NULL) ||
1764 		    iva2.va_seq != iva.va_seq)
1765 			iva.va_seq = 0;
1766 		break;
1767 	default:
1768 		/*
1769 		 * probably a special file.
1770 		 */
1771 		if ((vap->va_mask & AT_MODE) == 0) {
1772 			vap->va_mode = 0600;	/* default: owner rw only */
1773 			vap->va_mask |= AT_MODE;
1774 		}
1775 		syncval = FNODSYNC;
1776 		/*
1777 		 * We know this will only generate one VOP call
1778 		 */
1779 		vp = do_rfs4_op_mknod(args, resp, req, cs, vap, name);
1780 
1781 		if (vp == NULL) {
1782 			if (name != nm)
1783 				kmem_free(name, MAXPATHLEN + 1);
1784 			kmem_free(nm, len);
1785 			nfs4_ntov_table_free(&ntov, &sarg);
1786 			resp->attrset = 0;
1787 			goto out;
1788 		}
1789 
1790 		/*
1791 		 * Get the initial "after" sequence number, if it fails,
1792 		 * set to zero
1793 		 */
1794 		iva.va_mask = AT_SEQ;
1795 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1796 			iva.va_seq = 0;
1797 
1798 		break;
1799 	}
1800 	if (name != nm)
1801 		kmem_free(name, MAXPATHLEN + 1);
1802 	kmem_free(nm, len);
1803 
1804 	if (error) {
1805 		*cs->statusp = resp->status = puterrno4(error);
1806 	}
1807 
1808 	/*
1809 	 * Force modified data and metadata out to stable storage.
1810 	 */
1811 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
1812 
1813 	if (resp->status != NFS4_OK) {
1814 		if (vp != NULL)
1815 			VN_RELE(vp);
1816 		nfs4_ntov_table_free(&ntov, &sarg);
1817 		resp->attrset = 0;
1818 		goto out;
1819 	}
1820 
1821 	/*
1822 	 * Finish setup of cinfo response, "before" value already set.
1823 	 * Get "after" change value, if it fails, simply return the
1824 	 * before value.
1825 	 */
1826 	ava.va_mask = AT_CTIME|AT_SEQ;
1827 	if (VOP_GETATTR(dvp, &ava, 0, cr, NULL)) {
1828 		ava.va_ctime = bva.va_ctime;
1829 		ava.va_seq = 0;
1830 	}
1831 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, ava.va_ctime);
1832 
1833 	/*
1834 	 * True verification that object was created with correct
1835 	 * attrs is impossible.  The attrs could have been changed
1836 	 * immediately after object creation.  If attributes did
1837 	 * not verify, the only recourse for the server is to
1838 	 * destroy the object.  Maybe if some attrs (like gid)
1839 	 * are set incorrectly, the object should be destroyed;
1840 	 * however, seems bad as a default policy.  Do we really
1841 	 * want to destroy an object over one of the times not
1842 	 * verifying correctly?  For these reasons, the server
1843 	 * currently sets bits in attrset for createattrs
1844 	 * that were set; however, no verification is done.
1845 	 *
1846 	 * vmask_to_nmask accounts for vattr bits set on create
1847 	 *	[do_rfs4_set_attrs() only sets resp bits for
1848 	 *	 non-vattr/vfs bits.]
1849 	 * Mask off any bits set by default so as not to return
1850 	 * more attrset bits than were requested in createattrs
1851 	 */
1852 	nfs4_vmask_to_nmask(sarg.vap->va_mask, &resp->attrset);
1853 	resp->attrset &= args->createattrs.attrmask;
1854 	nfs4_ntov_table_free(&ntov, &sarg);
1855 
1856 	error = makefh4(&cs->fh, vp, cs->exi);
1857 	if (error) {
1858 		*cs->statusp = resp->status = puterrno4(error);
1859 	}
1860 
1861 	/*
1862 	 * The cinfo.atomic = TRUE only if we got no errors, we have
1863 	 * non-zero va_seq's, and it has incremented by exactly one
1864 	 * during the creation and it didn't change during the VOP_LOOKUP
1865 	 * or VOP_FSYNC.
1866 	 */
1867 	if (!error && bva.va_seq && iva.va_seq && ava.va_seq &&
1868 	    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
1869 		resp->cinfo.atomic = TRUE;
1870 	else
1871 		resp->cinfo.atomic = FALSE;
1872 
1873 	/*
1874 	 * Force modified metadata out to stable storage.
1875 	 *
1876 	 * if a underlying vp exists, pass it to VOP_FSYNC
1877 	 */
1878 	if (VOP_REALVP(vp, &realvp, NULL) == 0)
1879 		(void) VOP_FSYNC(realvp, syncval, cr, NULL);
1880 	else
1881 		(void) VOP_FSYNC(vp, syncval, cr, NULL);
1882 
1883 	if (resp->status != NFS4_OK) {
1884 		VN_RELE(vp);
1885 		goto out;
1886 	}
1887 	if (cs->vp)
1888 		VN_RELE(cs->vp);
1889 
1890 	cs->vp = vp;
1891 	*cs->statusp = resp->status = NFS4_OK;
1892 out:
1893 	DTRACE_NFSV4_2(op__create__done, struct compound_state *, cs,
1894 	    CREATE4res *, resp);
1895 }
1896 
1897 /*ARGSUSED*/
1898 static void
1899 rfs4_op_delegpurge(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1900     struct compound_state *cs)
1901 {
1902 	DTRACE_NFSV4_2(op__delegpurge__start, struct compound_state *, cs,
1903 	    DELEGPURGE4args *, &argop->nfs_argop4_u.opdelegpurge);
1904 
1905 	rfs4_op_inval(argop, resop, req, cs);
1906 
1907 	DTRACE_NFSV4_2(op__delegpurge__done, struct compound_state *, cs,
1908 	    DELEGPURGE4res *, &resop->nfs_resop4_u.opdelegpurge);
1909 }
1910 
1911 /*ARGSUSED*/
1912 static void
1913 rfs4_op_delegreturn(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1914     struct compound_state *cs)
1915 {
1916 	DELEGRETURN4args *args = &argop->nfs_argop4_u.opdelegreturn;
1917 	DELEGRETURN4res *resp = &resop->nfs_resop4_u.opdelegreturn;
1918 	rfs4_deleg_state_t *dsp;
1919 	nfsstat4 status;
1920 
1921 	DTRACE_NFSV4_2(op__delegreturn__start, struct compound_state *, cs,
1922 	    DELEGRETURN4args *, args);
1923 
1924 	status = rfs4_get_deleg_state(&args->deleg_stateid, &dsp);
1925 	resp->status = *cs->statusp = status;
1926 	if (status != NFS4_OK)
1927 		goto out;
1928 
1929 	/* Ensure specified filehandle matches */
1930 	if (cs->vp != dsp->rds_finfo->rf_vp) {
1931 		resp->status = *cs->statusp = NFS4ERR_BAD_STATEID;
1932 	} else
1933 		rfs4_return_deleg(dsp, FALSE);
1934 
1935 	rfs4_update_lease(dsp->rds_client);
1936 
1937 	rfs4_deleg_state_rele(dsp);
1938 out:
1939 	DTRACE_NFSV4_2(op__delegreturn__done, struct compound_state *, cs,
1940 	    DELEGRETURN4res *, resp);
1941 }
1942 
1943 /*
1944  * Check to see if a given "flavor" is an explicitly shared flavor.
1945  * The assumption of this routine is the "flavor" is already a valid
1946  * flavor in the secinfo list of "exi".
1947  *
1948  *	e.g.
1949  *		# share -o sec=flavor1 /export
1950  *		# share -o sec=flavor2 /export/home
1951  *
1952  *		flavor2 is not an explicitly shared flavor for /export,
1953  *		however it is in the secinfo list for /export thru the
1954  *		server namespace setup.
1955  */
1956 int
1957 is_exported_sec(int flavor, struct exportinfo *exi)
1958 {
1959 	int	i;
1960 	struct secinfo *sp;
1961 
1962 	sp = exi->exi_export.ex_secinfo;
1963 	for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
1964 		if (flavor == sp[i].s_secinfo.sc_nfsnum ||
1965 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE) {
1966 			return (SEC_REF_EXPORTED(&sp[i]));
1967 		}
1968 	}
1969 
1970 	/* Should not reach this point based on the assumption */
1971 	return (0);
1972 }
1973 
1974 /*
1975  * Check if the security flavor used in the request matches what is
1976  * required at the export point or at the root pseudo node (exi_root).
1977  *
1978  * returns 1 if there's a match or if exported with AUTH_NONE; 0 otherwise.
1979  *
1980  */
1981 static int
1982 secinfo_match_or_authnone(struct compound_state *cs)
1983 {
1984 	int	i;
1985 	struct secinfo *sp;
1986 
1987 	/*
1988 	 * Check cs->nfsflavor (from the request) against
1989 	 * the current export data in cs->exi.
1990 	 */
1991 	sp = cs->exi->exi_export.ex_secinfo;
1992 	for (i = 0; i < cs->exi->exi_export.ex_seccnt; i++) {
1993 		if (cs->nfsflavor == sp[i].s_secinfo.sc_nfsnum ||
1994 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE)
1995 			return (1);
1996 	}
1997 
1998 	return (0);
1999 }
2000 
2001 /*
2002  * Check the access authority for the client and return the correct error.
2003  */
2004 nfsstat4
2005 call_checkauth4(struct compound_state *cs, struct svc_req *req)
2006 {
2007 	int	authres;
2008 
2009 	/*
2010 	 * First, check if the security flavor used in the request
2011 	 * are among the flavors set in the server namespace.
2012 	 */
2013 	if (!secinfo_match_or_authnone(cs)) {
2014 		*cs->statusp = NFS4ERR_WRONGSEC;
2015 		return (*cs->statusp);
2016 	}
2017 
2018 	authres = checkauth4(cs, req);
2019 
2020 	if (authres > 0) {
2021 		*cs->statusp = NFS4_OK;
2022 		if (! (cs->access & CS_ACCESS_LIMITED))
2023 			cs->access = CS_ACCESS_OK;
2024 	} else if (authres == 0) {
2025 		*cs->statusp = NFS4ERR_ACCESS;
2026 	} else if (authres == -2) {
2027 		*cs->statusp = NFS4ERR_WRONGSEC;
2028 	} else {
2029 		*cs->statusp = NFS4ERR_DELAY;
2030 	}
2031 	return (*cs->statusp);
2032 }
2033 
2034 /*
2035  * bitmap4_to_attrmask is called by getattr and readdir.
2036  * It sets up the vattr mask and determines whether vfsstat call is needed
2037  * based on the input bitmap.
2038  * Returns nfsv4 status.
2039  */
2040 static nfsstat4
2041 bitmap4_to_attrmask(bitmap4 breq, struct nfs4_svgetit_arg *sargp)
2042 {
2043 	int i;
2044 	uint_t	va_mask;
2045 	struct statvfs64 *sbp = sargp->sbp;
2046 
2047 	sargp->sbp = NULL;
2048 	sargp->flag = 0;
2049 	sargp->rdattr_error = NFS4_OK;
2050 	sargp->mntdfid_set = FALSE;
2051 	if (sargp->cs->vp)
2052 		sargp->xattr = get_fh4_flag(&sargp->cs->fh,
2053 		    FH4_ATTRDIR | FH4_NAMEDATTR);
2054 	else
2055 		sargp->xattr = 0;
2056 
2057 	/*
2058 	 * Set rdattr_error_req to true if return error per
2059 	 * failed entry rather than fail the readdir.
2060 	 */
2061 	if (breq & FATTR4_RDATTR_ERROR_MASK)
2062 		sargp->rdattr_error_req = 1;
2063 	else
2064 		sargp->rdattr_error_req = 0;
2065 
2066 	/*
2067 	 * generate the va_mask
2068 	 * Handle the easy cases first
2069 	 */
2070 	switch (breq) {
2071 	case NFS4_NTOV_ATTR_MASK:
2072 		sargp->vap->va_mask = NFS4_NTOV_ATTR_AT_MASK;
2073 		return (NFS4_OK);
2074 
2075 	case NFS4_FS_ATTR_MASK:
2076 		sargp->vap->va_mask = NFS4_FS_ATTR_AT_MASK;
2077 		sargp->sbp = sbp;
2078 		return (NFS4_OK);
2079 
2080 	case NFS4_NTOV_ATTR_CACHE_MASK:
2081 		sargp->vap->va_mask = NFS4_NTOV_ATTR_CACHE_AT_MASK;
2082 		return (NFS4_OK);
2083 
2084 	case FATTR4_LEASE_TIME_MASK:
2085 		sargp->vap->va_mask = 0;
2086 		return (NFS4_OK);
2087 
2088 	default:
2089 		va_mask = 0;
2090 		for (i = 0; i < nfs4_ntov_map_size; i++) {
2091 			if ((breq & nfs4_ntov_map[i].fbit) &&
2092 			    nfs4_ntov_map[i].vbit)
2093 				va_mask |= nfs4_ntov_map[i].vbit;
2094 		}
2095 
2096 		/*
2097 		 * Check is vfsstat is needed
2098 		 */
2099 		if (breq & NFS4_FS_ATTR_MASK)
2100 			sargp->sbp = sbp;
2101 
2102 		sargp->vap->va_mask = va_mask;
2103 		return (NFS4_OK);
2104 	}
2105 	/* NOTREACHED */
2106 }
2107 
2108 /*
2109  * bitmap4_get_sysattrs is called by getattr and readdir.
2110  * It calls both VOP_GETATTR and VFS_STATVFS calls to get the attrs.
2111  * Returns nfsv4 status.
2112  */
2113 static nfsstat4
2114 bitmap4_get_sysattrs(struct nfs4_svgetit_arg *sargp)
2115 {
2116 	int error;
2117 	struct compound_state *cs = sargp->cs;
2118 	vnode_t *vp = cs->vp;
2119 
2120 	if (sargp->sbp != NULL) {
2121 		if (error = VFS_STATVFS(vp->v_vfsp, sargp->sbp)) {
2122 			sargp->sbp = NULL;	/* to identify error */
2123 			return (puterrno4(error));
2124 		}
2125 	}
2126 
2127 	return (rfs4_vop_getattr(vp, sargp->vap, 0, cs->cr));
2128 }
2129 
2130 static void
2131 nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp)
2132 {
2133 	ntovp->na = kmem_zalloc(sizeof (union nfs4_attr_u) * nfs4_ntov_map_size,
2134 	    KM_SLEEP);
2135 	ntovp->attrcnt = 0;
2136 	ntovp->vfsstat = FALSE;
2137 }
2138 
2139 static void
2140 nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
2141     struct nfs4_svgetit_arg *sargp)
2142 {
2143 	int i;
2144 	union nfs4_attr_u *na;
2145 	uint8_t *amap;
2146 
2147 	/*
2148 	 * XXX Should do the same checks for whether the bit is set
2149 	 */
2150 	for (i = 0, na = ntovp->na, amap = ntovp->amap;
2151 	    i < ntovp->attrcnt; i++, na++, amap++) {
2152 		(void) (*nfs4_ntov_map[*amap].sv_getit)(
2153 		    NFS4ATTR_FREEIT, sargp, na);
2154 	}
2155 	if ((sargp->op == NFS4ATTR_SETIT) || (sargp->op == NFS4ATTR_VERIT)) {
2156 		/*
2157 		 * xdr_free for getattr will be done later
2158 		 */
2159 		for (i = 0, na = ntovp->na, amap = ntovp->amap;
2160 		    i < ntovp->attrcnt; i++, na++, amap++) {
2161 			xdr_free(nfs4_ntov_map[*amap].xfunc, (caddr_t)na);
2162 		}
2163 	}
2164 	kmem_free(ntovp->na, sizeof (union nfs4_attr_u) * nfs4_ntov_map_size);
2165 }
2166 
2167 /*
2168  * do_rfs4_op_getattr gets the system attrs and converts into fattr4.
2169  */
2170 static nfsstat4
2171 do_rfs4_op_getattr(bitmap4 breq, fattr4 *fattrp,
2172     struct nfs4_svgetit_arg *sargp)
2173 {
2174 	int error = 0;
2175 	int i, k;
2176 	struct nfs4_ntov_table ntov;
2177 	XDR xdr;
2178 	ulong_t xdr_size;
2179 	char *xdr_attrs;
2180 	nfsstat4 status = NFS4_OK;
2181 	nfsstat4 prev_rdattr_error = sargp->rdattr_error;
2182 	union nfs4_attr_u *na;
2183 	uint8_t *amap;
2184 
2185 	sargp->op = NFS4ATTR_GETIT;
2186 	sargp->flag = 0;
2187 
2188 	fattrp->attrmask = 0;
2189 	/* if no bits requested, then return empty fattr4 */
2190 	if (breq == 0) {
2191 		fattrp->attrlist4_len = 0;
2192 		fattrp->attrlist4 = NULL;
2193 		return (NFS4_OK);
2194 	}
2195 
2196 	/*
2197 	 * return NFS4ERR_INVAL when client requests write-only attrs
2198 	 */
2199 	if (breq & (FATTR4_TIME_ACCESS_SET_MASK | FATTR4_TIME_MODIFY_SET_MASK))
2200 		return (NFS4ERR_INVAL);
2201 
2202 	nfs4_ntov_table_init(&ntov);
2203 	na = ntov.na;
2204 	amap = ntov.amap;
2205 
2206 	/*
2207 	 * Now loop to get or verify the attrs
2208 	 */
2209 	for (i = 0; i < nfs4_ntov_map_size; i++) {
2210 		if (breq & nfs4_ntov_map[i].fbit) {
2211 			if ((*nfs4_ntov_map[i].sv_getit)(
2212 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0) {
2213 
2214 				error = (*nfs4_ntov_map[i].sv_getit)(
2215 				    NFS4ATTR_GETIT, sargp, na);
2216 
2217 				/*
2218 				 * Possible error values:
2219 				 * >0 if sv_getit failed to
2220 				 * get the attr; 0 if succeeded;
2221 				 * <0 if rdattr_error and the
2222 				 * attribute cannot be returned.
2223 				 */
2224 				if (error && !(sargp->rdattr_error_req))
2225 					goto done;
2226 				/*
2227 				 * If error then just for entry
2228 				 */
2229 				if (error == 0) {
2230 					fattrp->attrmask |=
2231 					    nfs4_ntov_map[i].fbit;
2232 					*amap++ =
2233 					    (uint8_t)nfs4_ntov_map[i].nval;
2234 					na++;
2235 					(ntov.attrcnt)++;
2236 				} else if ((error > 0) &&
2237 				    (sargp->rdattr_error == NFS4_OK)) {
2238 					sargp->rdattr_error = puterrno4(error);
2239 				}
2240 				error = 0;
2241 			}
2242 		}
2243 	}
2244 
2245 	/*
2246 	 * If rdattr_error was set after the return value for it was assigned,
2247 	 * update it.
2248 	 */
2249 	if (prev_rdattr_error != sargp->rdattr_error) {
2250 		na = ntov.na;
2251 		amap = ntov.amap;
2252 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2253 			k = *amap;
2254 			if (k < FATTR4_RDATTR_ERROR) {
2255 				continue;
2256 			}
2257 			if ((k == FATTR4_RDATTR_ERROR) &&
2258 			    ((*nfs4_ntov_map[k].sv_getit)(
2259 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0)) {
2260 
2261 				(void) (*nfs4_ntov_map[k].sv_getit)(
2262 				    NFS4ATTR_GETIT, sargp, na);
2263 			}
2264 			break;
2265 		}
2266 	}
2267 
2268 	xdr_size = 0;
2269 	na = ntov.na;
2270 	amap = ntov.amap;
2271 	for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2272 		xdr_size += xdr_sizeof(nfs4_ntov_map[*amap].xfunc, na);
2273 	}
2274 
2275 	fattrp->attrlist4_len = xdr_size;
2276 	if (xdr_size) {
2277 		/* freed by rfs4_op_getattr_free() */
2278 		fattrp->attrlist4 = xdr_attrs = kmem_zalloc(xdr_size, KM_SLEEP);
2279 
2280 		xdrmem_create(&xdr, xdr_attrs, xdr_size, XDR_ENCODE);
2281 
2282 		na = ntov.na;
2283 		amap = ntov.amap;
2284 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2285 			if (!(*nfs4_ntov_map[*amap].xfunc)(&xdr, na)) {
2286 				DTRACE_PROBE1(nfss__e__getattr4_encfail,
2287 				    int, *amap);
2288 				status = NFS4ERR_SERVERFAULT;
2289 				break;
2290 			}
2291 		}
2292 		/* xdrmem_destroy(&xdrs); */	/* NO-OP */
2293 	} else {
2294 		fattrp->attrlist4 = NULL;
2295 	}
2296 done:
2297 
2298 	nfs4_ntov_table_free(&ntov, sargp);
2299 
2300 	if (error != 0)
2301 		status = puterrno4(error);
2302 
2303 	return (status);
2304 }
2305 
2306 /* ARGSUSED */
2307 static void
2308 rfs4_op_getattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2309     struct compound_state *cs)
2310 {
2311 	GETATTR4args *args = &argop->nfs_argop4_u.opgetattr;
2312 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2313 	struct nfs4_svgetit_arg sarg;
2314 	struct statvfs64 sb;
2315 	nfsstat4 status;
2316 
2317 	DTRACE_NFSV4_2(op__getattr__start, struct compound_state *, cs,
2318 	    GETATTR4args *, args);
2319 
2320 	if (cs->vp == NULL) {
2321 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2322 		goto out;
2323 	}
2324 
2325 	if (cs->access == CS_ACCESS_DENIED) {
2326 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2327 		goto out;
2328 	}
2329 
2330 	sarg.sbp = &sb;
2331 	sarg.cs = cs;
2332 	sarg.is_referral = B_FALSE;
2333 
2334 	status = bitmap4_to_attrmask(args->attr_request, &sarg);
2335 	if (status == NFS4_OK) {
2336 
2337 		status = bitmap4_get_sysattrs(&sarg);
2338 		if (status == NFS4_OK) {
2339 
2340 			/* Is this a referral? */
2341 			if (vn_is_nfs_reparse(cs->vp, cs->cr)) {
2342 				/* Older V4 Solaris client sees a link */
2343 				if (client_is_downrev(req))
2344 					sarg.vap->va_type = VLNK;
2345 				else
2346 					sarg.is_referral = B_TRUE;
2347 			}
2348 
2349 			status = do_rfs4_op_getattr(args->attr_request,
2350 			    &resp->obj_attributes, &sarg);
2351 		}
2352 	}
2353 	*cs->statusp = resp->status = status;
2354 out:
2355 	DTRACE_NFSV4_2(op__getattr__done, struct compound_state *, cs,
2356 	    GETATTR4res *, resp);
2357 }
2358 
2359 static void
2360 rfs4_op_getattr_free(nfs_resop4 *resop)
2361 {
2362 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2363 
2364 	nfs4_fattr4_free(&resp->obj_attributes);
2365 }
2366 
2367 /* ARGSUSED */
2368 static void
2369 rfs4_op_getfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2370     struct compound_state *cs)
2371 {
2372 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2373 
2374 	DTRACE_NFSV4_1(op__getfh__start, struct compound_state *, cs);
2375 
2376 	if (cs->vp == NULL) {
2377 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2378 		goto out;
2379 	}
2380 	if (cs->access == CS_ACCESS_DENIED) {
2381 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2382 		goto out;
2383 	}
2384 
2385 	/* check for reparse point at the share point */
2386 	if (cs->exi->exi_moved || vn_is_nfs_reparse(cs->exi->exi_vp, cs->cr)) {
2387 		/* it's all bad */
2388 		cs->exi->exi_moved = 1;
2389 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2390 		DTRACE_PROBE2(nfs4serv__func__referral__shared__moved,
2391 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2392 		return;
2393 	}
2394 
2395 	/* check for reparse point at vp */
2396 	if (vn_is_nfs_reparse(cs->vp, cs->cr) && !client_is_downrev(req)) {
2397 		/* it's not all bad */
2398 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2399 		DTRACE_PROBE2(nfs4serv__func__referral__moved,
2400 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2401 		return;
2402 	}
2403 
2404 	resp->object.nfs_fh4_val =
2405 	    kmem_alloc(cs->fh.nfs_fh4_len, KM_SLEEP);
2406 	nfs_fh4_copy(&cs->fh, &resp->object);
2407 	*cs->statusp = resp->status = NFS4_OK;
2408 out:
2409 	DTRACE_NFSV4_2(op__getfh__done, struct compound_state *, cs,
2410 	    GETFH4res *, resp);
2411 }
2412 
2413 static void
2414 rfs4_op_getfh_free(nfs_resop4 *resop)
2415 {
2416 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2417 
2418 	if (resp->status == NFS4_OK &&
2419 	    resp->object.nfs_fh4_val != NULL) {
2420 		kmem_free(resp->object.nfs_fh4_val, resp->object.nfs_fh4_len);
2421 		resp->object.nfs_fh4_val = NULL;
2422 		resp->object.nfs_fh4_len = 0;
2423 	}
2424 }
2425 
2426 /*
2427  * illegal: args: void
2428  *	    res : status (NFS4ERR_OP_ILLEGAL)
2429  */
2430 /* ARGSUSED */
2431 static void
2432 rfs4_op_illegal(nfs_argop4 *argop, nfs_resop4 *resop,
2433     struct svc_req *req, struct compound_state *cs)
2434 {
2435 	ILLEGAL4res *resp = &resop->nfs_resop4_u.opillegal;
2436 
2437 	resop->resop = OP_ILLEGAL;
2438 	*cs->statusp = resp->status = NFS4ERR_OP_ILLEGAL;
2439 }
2440 
2441 /*
2442  * link: args: SAVED_FH: file, CURRENT_FH: target directory
2443  *	 res: status. If success - CURRENT_FH unchanged, return change_info
2444  */
2445 /* ARGSUSED */
2446 static void
2447 rfs4_op_link(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2448     struct compound_state *cs)
2449 {
2450 	LINK4args *args = &argop->nfs_argop4_u.oplink;
2451 	LINK4res *resp = &resop->nfs_resop4_u.oplink;
2452 	int error;
2453 	vnode_t *vp;
2454 	vnode_t *dvp;
2455 	struct vattr bdva, idva, adva;
2456 	char *nm;
2457 	uint_t  len;
2458 	struct sockaddr *ca;
2459 	char *name = NULL;
2460 	nfsstat4 status;
2461 
2462 	DTRACE_NFSV4_2(op__link__start, struct compound_state *, cs,
2463 	    LINK4args *, args);
2464 
2465 	/* SAVED_FH: source object */
2466 	vp = cs->saved_vp;
2467 	if (vp == NULL) {
2468 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2469 		goto out;
2470 	}
2471 
2472 	/* CURRENT_FH: target directory */
2473 	dvp = cs->vp;
2474 	if (dvp == NULL) {
2475 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2476 		goto out;
2477 	}
2478 
2479 	/*
2480 	 * If there is a non-shared filesystem mounted on this vnode,
2481 	 * do not allow to link any file in this directory.
2482 	 */
2483 	if (vn_ismntpt(dvp)) {
2484 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2485 		goto out;
2486 	}
2487 
2488 	if (cs->access == CS_ACCESS_DENIED) {
2489 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2490 		goto out;
2491 	}
2492 
2493 	/* Check source object's type validity */
2494 	if (vp->v_type == VDIR) {
2495 		*cs->statusp = resp->status = NFS4ERR_ISDIR;
2496 		goto out;
2497 	}
2498 
2499 	/* Check target directory's type */
2500 	if (dvp->v_type != VDIR) {
2501 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2502 		goto out;
2503 	}
2504 
2505 	if (cs->saved_exi != cs->exi) {
2506 		*cs->statusp = resp->status = NFS4ERR_XDEV;
2507 		goto out;
2508 	}
2509 
2510 	status = utf8_dir_verify(&args->newname);
2511 	if (status != NFS4_OK) {
2512 		*cs->statusp = resp->status = status;
2513 		goto out;
2514 	}
2515 
2516 	nm = utf8_to_fn(&args->newname, &len, NULL);
2517 	if (nm == NULL) {
2518 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2519 		goto out;
2520 	}
2521 
2522 	if (len > MAXNAMELEN) {
2523 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2524 		kmem_free(nm, len);
2525 		goto out;
2526 	}
2527 
2528 	if (rdonly4(req, cs)) {
2529 		*cs->statusp = resp->status = NFS4ERR_ROFS;
2530 		kmem_free(nm, len);
2531 		goto out;
2532 	}
2533 
2534 	/* Get "before" change value */
2535 	bdva.va_mask = AT_CTIME|AT_SEQ;
2536 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
2537 	if (error) {
2538 		*cs->statusp = resp->status = puterrno4(error);
2539 		kmem_free(nm, len);
2540 		goto out;
2541 	}
2542 
2543 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2544 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2545 	    MAXPATHLEN  + 1);
2546 
2547 	if (name == NULL) {
2548 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2549 		kmem_free(nm, len);
2550 		goto out;
2551 	}
2552 
2553 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
2554 
2555 	error = VOP_LINK(dvp, vp, name, cs->cr, NULL, 0);
2556 
2557 	if (nm != name)
2558 		kmem_free(name, MAXPATHLEN + 1);
2559 	kmem_free(nm, len);
2560 
2561 	/*
2562 	 * Get the initial "after" sequence number, if it fails, set to zero
2563 	 */
2564 	idva.va_mask = AT_SEQ;
2565 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
2566 		idva.va_seq = 0;
2567 
2568 	/*
2569 	 * Force modified data and metadata out to stable storage.
2570 	 */
2571 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
2572 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
2573 
2574 	if (error) {
2575 		*cs->statusp = resp->status = puterrno4(error);
2576 		goto out;
2577 	}
2578 
2579 	/*
2580 	 * Get "after" change value, if it fails, simply return the
2581 	 * before value.
2582 	 */
2583 	adva.va_mask = AT_CTIME|AT_SEQ;
2584 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
2585 		adva.va_ctime = bdva.va_ctime;
2586 		adva.va_seq = 0;
2587 	}
2588 
2589 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
2590 
2591 	/*
2592 	 * The cinfo.atomic = TRUE only if we have
2593 	 * non-zero va_seq's, and it has incremented by exactly one
2594 	 * during the VOP_LINK and it didn't change during the VOP_FSYNC.
2595 	 */
2596 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
2597 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
2598 		resp->cinfo.atomic = TRUE;
2599 	else
2600 		resp->cinfo.atomic = FALSE;
2601 
2602 	*cs->statusp = resp->status = NFS4_OK;
2603 out:
2604 	DTRACE_NFSV4_2(op__link__done, struct compound_state *, cs,
2605 	    LINK4res *, resp);
2606 }
2607 
2608 /*
2609  * Used by rfs4_op_lookup and rfs4_op_lookupp to do the actual work.
2610  */
2611 
2612 /* ARGSUSED */
2613 static nfsstat4
2614 do_rfs4_op_lookup(char *nm, struct svc_req *req, struct compound_state *cs)
2615 {
2616 	int error;
2617 	int different_export = 0;
2618 	vnode_t *vp, *pre_tvp = NULL, *oldvp = NULL;
2619 	struct exportinfo *exi = NULL, *pre_exi = NULL, *oexi = NULL;
2620 	nfsstat4 stat;
2621 	fid_t fid;
2622 	int attrdir, dotdot, walk;
2623 	bool_t is_newvp = FALSE;
2624 
2625 	if (cs->vp->v_flag & V_XATTRDIR) {
2626 		attrdir = 1;
2627 		ASSERT(get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2628 	} else {
2629 		attrdir = 0;
2630 		ASSERT(! get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2631 	}
2632 
2633 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
2634 
2635 	/*
2636 	 * If dotdotting, then need to check whether it's
2637 	 * above the root of a filesystem, or above an
2638 	 * export point.
2639 	 */
2640 	if (dotdot) {
2641 
2642 		/*
2643 		 * If dotdotting at the root of a filesystem, then
2644 		 * need to traverse back to the mounted-on filesystem
2645 		 * and do the dotdot lookup there.
2646 		 */
2647 		if (cs->vp->v_flag & VROOT) {
2648 
2649 			/*
2650 			 * If at the system root, then can
2651 			 * go up no further.
2652 			 */
2653 			if (VN_CMP(cs->vp, rootdir))
2654 				return (puterrno4(ENOENT));
2655 
2656 			/*
2657 			 * Traverse back to the mounted-on filesystem
2658 			 */
2659 			cs->vp = untraverse(cs->vp);
2660 
2661 			/*
2662 			 * Set the different_export flag so we remember
2663 			 * to pick up a new exportinfo entry for
2664 			 * this new filesystem.
2665 			 */
2666 			different_export = 1;
2667 		} else {
2668 
2669 			/*
2670 			 * If dotdotting above an export point then set
2671 			 * the different_export to get new export info.
2672 			 */
2673 			different_export = nfs_exported(cs->exi, cs->vp);
2674 		}
2675 	}
2676 
2677 	error = VOP_LOOKUP(cs->vp, nm, &vp, NULL, 0, NULL, cs->cr,
2678 	    NULL, NULL, NULL);
2679 	if (error)
2680 		return (puterrno4(error));
2681 
2682 	/*
2683 	 * If the vnode is in a pseudo filesystem, check whether it is visible.
2684 	 *
2685 	 * XXX if the vnode is a symlink and it is not visible in
2686 	 * a pseudo filesystem, return ENOENT (not following symlink).
2687 	 * V4 client can not mount such symlink. This is a regression
2688 	 * from V2/V3.
2689 	 *
2690 	 * In the same exported filesystem, if the security flavor used
2691 	 * is not an explicitly shared flavor, limit the view to the visible
2692 	 * list entries only. This is not a WRONGSEC case because it's already
2693 	 * checked via PUTROOTFH/PUTPUBFH or PUTFH.
2694 	 */
2695 	if (!different_export &&
2696 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
2697 	    cs->access & CS_ACCESS_LIMITED)) {
2698 		if (! nfs_visible(cs->exi, vp, &different_export)) {
2699 			VN_RELE(vp);
2700 			return (puterrno4(ENOENT));
2701 		}
2702 	}
2703 
2704 	/*
2705 	 * If it's a mountpoint, then traverse it.
2706 	 */
2707 	if (vn_ismntpt(vp)) {
2708 		pre_exi = cs->exi;	/* save pre-traversed exportinfo */
2709 		pre_tvp = vp;		/* save pre-traversed vnode	*/
2710 
2711 		/*
2712 		 * hold pre_tvp to counteract rele by traverse.  We will
2713 		 * need pre_tvp below if checkexport fails
2714 		 */
2715 		VN_HOLD(pre_tvp);
2716 		if ((error = traverse(&vp)) != 0) {
2717 			VN_RELE(vp);
2718 			VN_RELE(pre_tvp);
2719 			return (puterrno4(error));
2720 		}
2721 		different_export = 1;
2722 	} else if (vp->v_vfsp != cs->vp->v_vfsp) {
2723 		/*
2724 		 * The vfsp comparison is to handle the case where
2725 		 * a LOFS mount is shared.  lo_lookup traverses mount points,
2726 		 * and NFS is unaware of local fs transistions because
2727 		 * v_vfsmountedhere isn't set.  For this special LOFS case,
2728 		 * the dir and the obj returned by lookup will have different
2729 		 * vfs ptrs.
2730 		 */
2731 		different_export = 1;
2732 	}
2733 
2734 	if (different_export) {
2735 
2736 		bzero(&fid, sizeof (fid));
2737 		fid.fid_len = MAXFIDSZ;
2738 		error = vop_fid_pseudo(vp, &fid);
2739 		if (error) {
2740 			VN_RELE(vp);
2741 			if (pre_tvp)
2742 				VN_RELE(pre_tvp);
2743 			return (puterrno4(error));
2744 		}
2745 
2746 		if (dotdot)
2747 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
2748 		else
2749 			exi = checkexport(&vp->v_vfsp->vfs_fsid, &fid, vp);
2750 
2751 		if (exi == NULL) {
2752 			if (pre_tvp) {
2753 				/*
2754 				 * If this vnode is a mounted-on vnode,
2755 				 * but the mounted-on file system is not
2756 				 * exported, send back the filehandle for
2757 				 * the mounted-on vnode, not the root of
2758 				 * the mounted-on file system.
2759 				 */
2760 				VN_RELE(vp);
2761 				vp = pre_tvp;
2762 				exi = pre_exi;
2763 				if (exi)
2764 					exi_hold(exi);
2765 			} else {
2766 				VN_RELE(vp);
2767 				return (puterrno4(EACCES));
2768 			}
2769 		} else if (pre_tvp) {
2770 			/* we're done with pre_tvp now. release extra hold */
2771 			VN_RELE(pre_tvp);
2772 		}
2773 
2774 		if (cs->exi)
2775 			exi_rele(cs->exi);
2776 		cs->exi = exi;
2777 
2778 		/*
2779 		 * Now we do a checkauth4. The reason is that
2780 		 * this client/user may not have access to the new
2781 		 * exported file system, and if he does,
2782 		 * the client/user may be mapped to a different uid.
2783 		 *
2784 		 * We start with a new cr, because the checkauth4 done
2785 		 * in the PUT*FH operation over wrote the cred's uid,
2786 		 * gid, etc, and we want the real thing before calling
2787 		 * checkauth4()
2788 		 */
2789 		crfree(cs->cr);
2790 		cs->cr = crdup(cs->basecr);
2791 
2792 		oldvp = cs->vp;
2793 		cs->vp = vp;
2794 		is_newvp = TRUE;
2795 
2796 		stat = call_checkauth4(cs, req);
2797 		if (stat != NFS4_OK) {
2798 			VN_RELE(cs->vp);
2799 			cs->vp = oldvp;
2800 			return (stat);
2801 		}
2802 	}
2803 
2804 	/*
2805 	 * After various NFS checks, do a label check on the path
2806 	 * component. The label on this path should either be the
2807 	 * global zone's label or a zone's label. We are only
2808 	 * interested in the zone's label because exported files
2809 	 * in global zone is accessible (though read-only) to
2810 	 * clients. The exportability/visibility check is already
2811 	 * done before reaching this code.
2812 	 */
2813 	if (is_system_labeled()) {
2814 		bslabel_t *clabel;
2815 
2816 		ASSERT(req->rq_label != NULL);
2817 		clabel = req->rq_label;
2818 		DTRACE_PROBE2(tx__rfs4__log__info__oplookup__clabel, char *,
2819 		    "got client label from request(1)", struct svc_req *, req);
2820 
2821 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
2822 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
2823 			    cs->exi)) {
2824 				error = EACCES;
2825 				goto err_out;
2826 			}
2827 		} else {
2828 			/*
2829 			 * We grant access to admin_low label clients
2830 			 * only if the client is trusted, i.e. also
2831 			 * running Solaris Trusted Extension.
2832 			 */
2833 			struct sockaddr	*ca;
2834 			int		addr_type;
2835 			void		*ipaddr;
2836 			tsol_tpc_t	*tp;
2837 
2838 			ca = (struct sockaddr *)svc_getrpccaller(
2839 			    req->rq_xprt)->buf;
2840 			if (ca->sa_family == AF_INET) {
2841 				addr_type = IPV4_VERSION;
2842 				ipaddr = &((struct sockaddr_in *)ca)->sin_addr;
2843 			} else if (ca->sa_family == AF_INET6) {
2844 				addr_type = IPV6_VERSION;
2845 				ipaddr = &((struct sockaddr_in6 *)
2846 				    ca)->sin6_addr;
2847 			}
2848 			tp = find_tpc(ipaddr, addr_type, B_FALSE);
2849 			if (tp == NULL || tp->tpc_tp.tp_doi !=
2850 			    l_admin_low->tsl_doi || tp->tpc_tp.host_type !=
2851 			    SUN_CIPSO) {
2852 				if (tp != NULL)
2853 					TPC_RELE(tp);
2854 				error = EACCES;
2855 				goto err_out;
2856 			}
2857 			TPC_RELE(tp);
2858 		}
2859 	}
2860 
2861 	error = makefh4(&cs->fh, vp, cs->exi);
2862 
2863 err_out:
2864 	if (error) {
2865 		if (is_newvp) {
2866 			VN_RELE(cs->vp);
2867 			cs->vp = oldvp;
2868 		} else
2869 			VN_RELE(vp);
2870 		return (puterrno4(error));
2871 	}
2872 
2873 	if (!is_newvp) {
2874 		if (cs->vp)
2875 			VN_RELE(cs->vp);
2876 		cs->vp = vp;
2877 	} else if (oldvp)
2878 		VN_RELE(oldvp);
2879 
2880 	/*
2881 	 * if did lookup on attrdir and didn't lookup .., set named
2882 	 * attr fh flag
2883 	 */
2884 	if (attrdir && ! dotdot)
2885 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
2886 
2887 	/* Assume false for now, open proc will set this */
2888 	cs->mandlock = FALSE;
2889 
2890 	return (NFS4_OK);
2891 }
2892 
2893 /* ARGSUSED */
2894 static void
2895 rfs4_op_lookup(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2896     struct compound_state *cs)
2897 {
2898 	LOOKUP4args *args = &argop->nfs_argop4_u.oplookup;
2899 	LOOKUP4res *resp = &resop->nfs_resop4_u.oplookup;
2900 	char *nm;
2901 	uint_t len;
2902 	struct sockaddr *ca;
2903 	char *name = NULL;
2904 	nfsstat4 status;
2905 
2906 	DTRACE_NFSV4_2(op__lookup__start, struct compound_state *, cs,
2907 	    LOOKUP4args *, args);
2908 
2909 	if (cs->vp == NULL) {
2910 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2911 		goto out;
2912 	}
2913 
2914 	if (cs->vp->v_type == VLNK) {
2915 		*cs->statusp = resp->status = NFS4ERR_SYMLINK;
2916 		goto out;
2917 	}
2918 
2919 	if (cs->vp->v_type != VDIR) {
2920 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2921 		goto out;
2922 	}
2923 
2924 	status = utf8_dir_verify(&args->objname);
2925 	if (status != NFS4_OK) {
2926 		*cs->statusp = resp->status = status;
2927 		goto out;
2928 	}
2929 
2930 	nm = utf8_to_str(&args->objname, &len, NULL);
2931 	if (nm == NULL) {
2932 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2933 		goto out;
2934 	}
2935 
2936 	if (len > MAXNAMELEN) {
2937 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2938 		kmem_free(nm, len);
2939 		goto out;
2940 	}
2941 
2942 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2943 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2944 	    MAXPATHLEN  + 1);
2945 
2946 	if (name == NULL) {
2947 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2948 		kmem_free(nm, len);
2949 		goto out;
2950 	}
2951 
2952 	*cs->statusp = resp->status = do_rfs4_op_lookup(name, req, cs);
2953 
2954 	if (name != nm)
2955 		kmem_free(name, MAXPATHLEN + 1);
2956 	kmem_free(nm, len);
2957 
2958 out:
2959 	DTRACE_NFSV4_2(op__lookup__done, struct compound_state *, cs,
2960 	    LOOKUP4res *, resp);
2961 }
2962 
2963 /* ARGSUSED */
2964 static void
2965 rfs4_op_lookupp(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
2966     struct compound_state *cs)
2967 {
2968 	LOOKUPP4res *resp = &resop->nfs_resop4_u.oplookupp;
2969 
2970 	DTRACE_NFSV4_1(op__lookupp__start, struct compound_state *, cs);
2971 
2972 	if (cs->vp == NULL) {
2973 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2974 		goto out;
2975 	}
2976 
2977 	if (cs->vp->v_type != VDIR) {
2978 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2979 		goto out;
2980 	}
2981 
2982 	*cs->statusp = resp->status = do_rfs4_op_lookup("..", req, cs);
2983 
2984 	/*
2985 	 * From NFSV4 Specification, LOOKUPP should not check for
2986 	 * NFS4ERR_WRONGSEC. Retrun NFS4_OK instead.
2987 	 */
2988 	if (resp->status == NFS4ERR_WRONGSEC) {
2989 		*cs->statusp = resp->status = NFS4_OK;
2990 	}
2991 
2992 out:
2993 	DTRACE_NFSV4_2(op__lookupp__done, struct compound_state *, cs,
2994 	    LOOKUPP4res *, resp);
2995 }
2996 
2997 
2998 /*ARGSUSED2*/
2999 static void
3000 rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3001     struct compound_state *cs)
3002 {
3003 	OPENATTR4args	*args = &argop->nfs_argop4_u.opopenattr;
3004 	OPENATTR4res	*resp = &resop->nfs_resop4_u.opopenattr;
3005 	vnode_t		*avp = NULL;
3006 	int		lookup_flags = LOOKUP_XATTR, error;
3007 	int		exp_ro = 0;
3008 
3009 	DTRACE_NFSV4_2(op__openattr__start, struct compound_state *, cs,
3010 	    OPENATTR4args *, args);
3011 
3012 	if (cs->vp == NULL) {
3013 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3014 		goto out;
3015 	}
3016 
3017 	if ((cs->vp->v_vfsp->vfs_flag & VFS_XATTR) == 0 &&
3018 	    !vfs_has_feature(cs->vp->v_vfsp, VFSFT_SYSATTR_VIEWS)) {
3019 		*cs->statusp = resp->status = puterrno4(ENOTSUP);
3020 		goto out;
3021 	}
3022 
3023 	/*
3024 	 * If file system supports passing ACE mask to VOP_ACCESS then
3025 	 * check for ACE_READ_NAMED_ATTRS, otherwise do legacy checks
3026 	 */
3027 
3028 	if (vfs_has_feature(cs->vp->v_vfsp, VFSFT_ACEMASKONACCESS))
3029 		error = VOP_ACCESS(cs->vp, ACE_READ_NAMED_ATTRS,
3030 		    V_ACE_MASK, cs->cr, NULL);
3031 	else
3032 		error = ((VOP_ACCESS(cs->vp, VREAD, 0, cs->cr, NULL) != 0) &&
3033 		    (VOP_ACCESS(cs->vp, VWRITE, 0, cs->cr, NULL) != 0) &&
3034 		    (VOP_ACCESS(cs->vp, VEXEC, 0, cs->cr, NULL) != 0));
3035 
3036 	if (error) {
3037 		*cs->statusp = resp->status = puterrno4(EACCES);
3038 		goto out;
3039 	}
3040 
3041 	/*
3042 	 * The CREATE_XATTR_DIR VOP flag cannot be specified if
3043 	 * the file system is exported read-only -- regardless of
3044 	 * createdir flag.  Otherwise the attrdir would be created
3045 	 * (assuming server fs isn't mounted readonly locally).  If
3046 	 * VOP_LOOKUP returns ENOENT in this case, the error will
3047 	 * be translated into EROFS.  ENOSYS is mapped to ENOTSUP
3048 	 * because specfs has no VOP_LOOKUP op, so the macro would
3049 	 * return ENOSYS.  EINVAL is returned by all (current)
3050 	 * Solaris file system implementations when any of their
3051 	 * restrictions are violated (xattr(dir) can't have xattrdir).
3052 	 * Returning NOTSUPP is more appropriate in this case
3053 	 * because the object will never be able to have an attrdir.
3054 	 */
3055 	if (args->createdir && ! (exp_ro = rdonly4(req, cs)))
3056 		lookup_flags |= CREATE_XATTR_DIR;
3057 
3058 	error = VOP_LOOKUP(cs->vp, "", &avp, NULL, lookup_flags, NULL, cs->cr,
3059 	    NULL, NULL, NULL);
3060 
3061 	if (error) {
3062 		if (error == ENOENT && args->createdir && exp_ro)
3063 			*cs->statusp = resp->status = puterrno4(EROFS);
3064 		else if (error == EINVAL || error == ENOSYS)
3065 			*cs->statusp = resp->status = puterrno4(ENOTSUP);
3066 		else
3067 			*cs->statusp = resp->status = puterrno4(error);
3068 		goto out;
3069 	}
3070 
3071 	ASSERT(avp->v_flag & V_XATTRDIR);
3072 
3073 	error = makefh4(&cs->fh, avp, cs->exi);
3074 
3075 	if (error) {
3076 		VN_RELE(avp);
3077 		*cs->statusp = resp->status = puterrno4(error);
3078 		goto out;
3079 	}
3080 
3081 	VN_RELE(cs->vp);
3082 	cs->vp = avp;
3083 
3084 	/*
3085 	 * There is no requirement for an attrdir fh flag
3086 	 * because the attrdir has a vnode flag to distinguish
3087 	 * it from regular (non-xattr) directories.  The
3088 	 * FH4_ATTRDIR flag is set for future sanity checks.
3089 	 */
3090 	set_fh4_flag(&cs->fh, FH4_ATTRDIR);
3091 	*cs->statusp = resp->status = NFS4_OK;
3092 
3093 out:
3094 	DTRACE_NFSV4_2(op__openattr__done, struct compound_state *, cs,
3095 	    OPENATTR4res *, resp);
3096 }
3097 
3098 static int
3099 do_io(int direction, vnode_t *vp, struct uio *uio, int ioflag, cred_t *cred,
3100     caller_context_t *ct)
3101 {
3102 	int error;
3103 	int i;
3104 	clock_t delaytime;
3105 
3106 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
3107 
3108 	/*
3109 	 * Don't block on mandatory locks. If this routine returns
3110 	 * EAGAIN, the caller should return NFS4ERR_LOCKED.
3111 	 */
3112 	uio->uio_fmode = FNONBLOCK;
3113 
3114 	for (i = 0; i < rfs4_maxlock_tries; i++) {
3115 
3116 
3117 		if (direction == FREAD) {
3118 			(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, ct);
3119 			error = VOP_READ(vp, uio, ioflag, cred, ct);
3120 			VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, ct);
3121 		} else {
3122 			(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, ct);
3123 			error = VOP_WRITE(vp, uio, ioflag, cred, ct);
3124 			VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, ct);
3125 		}
3126 
3127 		if (error != EAGAIN)
3128 			break;
3129 
3130 		if (i < rfs4_maxlock_tries - 1) {
3131 			delay(delaytime);
3132 			delaytime *= 2;
3133 		}
3134 	}
3135 
3136 	return (error);
3137 }
3138 
3139 /* ARGSUSED */
3140 static void
3141 rfs4_op_read(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3142     struct compound_state *cs)
3143 {
3144 	READ4args *args = &argop->nfs_argop4_u.opread;
3145 	READ4res *resp = &resop->nfs_resop4_u.opread;
3146 	int error;
3147 	int verror;
3148 	vnode_t *vp;
3149 	struct vattr va;
3150 	struct iovec iov, *iovp = NULL;
3151 	int iovcnt;
3152 	struct uio uio;
3153 	u_offset_t offset;
3154 	bool_t *deleg = &cs->deleg;
3155 	nfsstat4 stat;
3156 	int in_crit = 0;
3157 	mblk_t *mp = NULL;
3158 	int alloc_err = 0;
3159 	int rdma_used = 0;
3160 	int loaned_buffers;
3161 	caller_context_t ct;
3162 	struct uio *uiop;
3163 
3164 	DTRACE_NFSV4_2(op__read__start, struct compound_state *, cs,
3165 	    READ4args, args);
3166 
3167 	vp = cs->vp;
3168 	if (vp == NULL) {
3169 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3170 		goto out;
3171 	}
3172 	if (cs->access == CS_ACCESS_DENIED) {
3173 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3174 		goto out;
3175 	}
3176 
3177 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3178 	    deleg, TRUE, &ct)) != NFS4_OK) {
3179 		*cs->statusp = resp->status = stat;
3180 		goto out;
3181 	}
3182 
3183 	/*
3184 	 * Enter the critical region before calling VOP_RWLOCK
3185 	 * to avoid a deadlock with write requests.
3186 	 */
3187 	if (nbl_need_check(vp)) {
3188 		nbl_start_crit(vp, RW_READER);
3189 		in_crit = 1;
3190 		if (nbl_conflict(vp, NBL_READ, args->offset, args->count, 0,
3191 		    &ct)) {
3192 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
3193 			goto out;
3194 		}
3195 	}
3196 
3197 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3198 	    deleg, TRUE, &ct)) != NFS4_OK) {
3199 		*cs->statusp = resp->status = stat;
3200 		goto out;
3201 	}
3202 
3203 	if (args->wlist) {
3204 		if (args->count > clist_len(args->wlist)) {
3205 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3206 			goto out;
3207 		}
3208 		rdma_used = 1;
3209 	}
3210 
3211 	/* use loaned buffers for TCP */
3212 	loaned_buffers = (nfs_loaned_buffers && !rdma_used) ? 1 : 0;
3213 
3214 	va.va_mask = AT_MODE|AT_SIZE|AT_UID;
3215 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3216 
3217 	/*
3218 	 * If we can't get the attributes, then we can't do the
3219 	 * right access checking.  So, we'll fail the request.
3220 	 */
3221 	if (verror) {
3222 		*cs->statusp = resp->status = puterrno4(verror);
3223 		goto out;
3224 	}
3225 
3226 	if (vp->v_type != VREG) {
3227 		*cs->statusp = resp->status =
3228 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
3229 		goto out;
3230 	}
3231 
3232 	if (crgetuid(cs->cr) != va.va_uid &&
3233 	    (error = VOP_ACCESS(vp, VREAD, 0, cs->cr, &ct)) &&
3234 	    (error = VOP_ACCESS(vp, VEXEC, 0, cs->cr, &ct))) {
3235 		*cs->statusp = resp->status = puterrno4(error);
3236 		goto out;
3237 	}
3238 
3239 	if (MANDLOCK(vp, va.va_mode)) { /* XXX - V4 supports mand locking */
3240 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3241 		goto out;
3242 	}
3243 
3244 	offset = args->offset;
3245 	if (offset >= va.va_size) {
3246 		*cs->statusp = resp->status = NFS4_OK;
3247 		resp->eof = TRUE;
3248 		resp->data_len = 0;
3249 		resp->data_val = NULL;
3250 		resp->mblk = NULL;
3251 		/* RDMA */
3252 		resp->wlist = args->wlist;
3253 		resp->wlist_len = resp->data_len;
3254 		*cs->statusp = resp->status = NFS4_OK;
3255 		if (resp->wlist)
3256 			clist_zero_len(resp->wlist);
3257 		goto out;
3258 	}
3259 
3260 	if (args->count == 0) {
3261 		*cs->statusp = resp->status = NFS4_OK;
3262 		resp->eof = FALSE;
3263 		resp->data_len = 0;
3264 		resp->data_val = NULL;
3265 		resp->mblk = NULL;
3266 		/* RDMA */
3267 		resp->wlist = args->wlist;
3268 		resp->wlist_len = resp->data_len;
3269 		if (resp->wlist)
3270 			clist_zero_len(resp->wlist);
3271 		goto out;
3272 	}
3273 
3274 	/*
3275 	 * Do not allocate memory more than maximum allowed
3276 	 * transfer size
3277 	 */
3278 	if (args->count > rfs4_tsize(req))
3279 		args->count = rfs4_tsize(req);
3280 
3281 	if (loaned_buffers) {
3282 		uiop = (uio_t *)rfs_setup_xuio(vp);
3283 		ASSERT(uiop != NULL);
3284 		uiop->uio_segflg = UIO_SYSSPACE;
3285 		uiop->uio_loffset = args->offset;
3286 		uiop->uio_resid = args->count;
3287 
3288 		/* Jump to do the read if successful */
3289 		if (!VOP_REQZCBUF(vp, UIO_READ, (xuio_t *)uiop, cs->cr, &ct)) {
3290 			/*
3291 			 * Need to hold the vnode until after VOP_RETZCBUF()
3292 			 * is called.
3293 			 */
3294 			VN_HOLD(vp);
3295 			goto doio_read;
3296 		}
3297 
3298 		DTRACE_PROBE2(nfss__i__reqzcbuf_failed, int,
3299 		    uiop->uio_loffset, int, uiop->uio_resid);
3300 
3301 		uiop->uio_extflg = 0;
3302 
3303 		/* failure to setup for zero copy */
3304 		rfs_free_xuio((void *)uiop);
3305 		loaned_buffers = 0;
3306 	}
3307 
3308 	/*
3309 	 * If returning data via RDMA Write, then grab the chunk list. If we
3310 	 * aren't returning READ data w/RDMA_WRITE, then grab a mblk.
3311 	 */
3312 	if (rdma_used) {
3313 		mp = NULL;
3314 		(void) rdma_get_wchunk(req, &iov, args->wlist);
3315 		uio.uio_iov = &iov;
3316 		uio.uio_iovcnt = 1;
3317 	} else {
3318 		/*
3319 		 * mp will contain the data to be sent out in the read reply.
3320 		 * It will be freed after the reply has been sent.
3321 		 */
3322 		mp = rfs_read_alloc(args->count, &iovp, &iovcnt);
3323 		ASSERT(mp != NULL);
3324 		ASSERT(alloc_err == 0);
3325 		uio.uio_iov = iovp;
3326 		uio.uio_iovcnt = iovcnt;
3327 	}
3328 
3329 	uio.uio_segflg = UIO_SYSSPACE;
3330 	uio.uio_extflg = UIO_COPY_CACHED;
3331 	uio.uio_loffset = args->offset;
3332 	uio.uio_resid = args->count;
3333 	uiop = &uio;
3334 
3335 doio_read:
3336 	error = do_io(FREAD, vp, uiop, 0, cs->cr, &ct);
3337 
3338 	va.va_mask = AT_SIZE;
3339 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3340 
3341 	if (error) {
3342 		if (mp)
3343 			freemsg(mp);
3344 		*cs->statusp = resp->status = puterrno4(error);
3345 		goto out;
3346 	}
3347 
3348 	/* make mblk using zc buffers */
3349 	if (loaned_buffers) {
3350 		mp = uio_to_mblk(uiop);
3351 		ASSERT(mp != NULL);
3352 	}
3353 
3354 	*cs->statusp = resp->status = NFS4_OK;
3355 
3356 	ASSERT(uiop->uio_resid >= 0);
3357 	resp->data_len = args->count - uiop->uio_resid;
3358 	if (mp) {
3359 		resp->data_val = (char *)mp->b_datap->db_base;
3360 		rfs_rndup_mblks(mp, resp->data_len, loaned_buffers);
3361 	} else {
3362 		resp->data_val = (caddr_t)iov.iov_base;
3363 	}
3364 
3365 	resp->mblk = mp;
3366 
3367 	if (!verror && offset + resp->data_len == va.va_size)
3368 		resp->eof = TRUE;
3369 	else
3370 		resp->eof = FALSE;
3371 
3372 	if (rdma_used) {
3373 		if (!rdma_setup_read_data4(args, resp)) {
3374 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3375 		}
3376 	} else {
3377 		resp->wlist = NULL;
3378 	}
3379 
3380 out:
3381 	if (in_crit)
3382 		nbl_end_crit(vp);
3383 
3384 	if (iovp != NULL)
3385 		kmem_free(iovp, iovcnt * sizeof (struct iovec));
3386 
3387 	DTRACE_NFSV4_2(op__read__done, struct compound_state *, cs,
3388 	    READ4res *, resp);
3389 }
3390 
3391 static void
3392 rfs4_op_read_free(nfs_resop4 *resop)
3393 {
3394 	READ4res	*resp = &resop->nfs_resop4_u.opread;
3395 
3396 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3397 		freemsg(resp->mblk);
3398 		resp->mblk = NULL;
3399 		resp->data_val = NULL;
3400 		resp->data_len = 0;
3401 	}
3402 }
3403 
3404 static void
3405 rfs4_op_readdir_free(nfs_resop4 * resop)
3406 {
3407 	READDIR4res    *resp = &resop->nfs_resop4_u.opreaddir;
3408 
3409 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3410 		freeb(resp->mblk);
3411 		resp->mblk = NULL;
3412 		resp->data_len = 0;
3413 	}
3414 }
3415 
3416 
3417 /* ARGSUSED */
3418 static void
3419 rfs4_op_putpubfh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
3420     struct compound_state *cs)
3421 {
3422 	PUTPUBFH4res	*resp = &resop->nfs_resop4_u.opputpubfh;
3423 	int		error;
3424 	vnode_t		*vp;
3425 	struct exportinfo *exi, *sav_exi;
3426 	nfs_fh4_fmt_t	*fh_fmtp;
3427 
3428 	DTRACE_NFSV4_1(op__putpubfh__start, struct compound_state *, cs);
3429 
3430 	if (cs->vp) {
3431 		VN_RELE(cs->vp);
3432 		cs->vp = NULL;
3433 	}
3434 
3435 	if (cs->cr)
3436 		crfree(cs->cr);
3437 
3438 	cs->cr = crdup(cs->basecr);
3439 
3440 	vp = exi_public->exi_vp;
3441 	if (vp == NULL) {
3442 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3443 		goto out;
3444 	}
3445 
3446 	error = makefh4(&cs->fh, vp, exi_public);
3447 	if (error != 0) {
3448 		*cs->statusp = resp->status = puterrno4(error);
3449 		goto out;
3450 	}
3451 	sav_exi = cs->exi;
3452 	if (exi_public == exi_root) {
3453 		/*
3454 		 * No filesystem is actually shared public, so we default
3455 		 * to exi_root. In this case, we must check whether root
3456 		 * is exported.
3457 		 */
3458 		fh_fmtp = (nfs_fh4_fmt_t *)cs->fh.nfs_fh4_val;
3459 
3460 		/*
3461 		 * if root filesystem is exported, the exportinfo struct that we
3462 		 * should use is what checkexport returns, because root_exi is
3463 		 * actually a mostly empty struct.
3464 		 */
3465 		exi = checkexport(&fh_fmtp->fh4_fsid,
3466 		    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
3467 		if (exi) {
3468 			cs->exi = exi;
3469 		} else {
3470 			exi_hold(exi_public);
3471 			cs->exi = exi_public;
3472 		}
3473 	} else {
3474 		/*
3475 		 * it's a properly shared filesystem
3476 		 */
3477 		exi_hold(exi_public);
3478 		cs->exi = exi_public;
3479 	}
3480 
3481 	if (is_system_labeled()) {
3482 		bslabel_t *clabel;
3483 
3484 		ASSERT(req->rq_label != NULL);
3485 		clabel = req->rq_label;
3486 		DTRACE_PROBE2(tx__rfs4__log__info__opputpubfh__clabel, char *,
3487 		    "got client label from request(1)",
3488 		    struct svc_req *, req);
3489 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
3490 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
3491 			    cs->exi)) {
3492 				*cs->statusp = resp->status =
3493 				    NFS4ERR_SERVERFAULT;
3494 				if (sav_exi)
3495 					exi_rele(sav_exi);
3496 				goto out;
3497 			}
3498 		}
3499 	}
3500 
3501 	VN_HOLD(vp);
3502 	cs->vp = vp;
3503 
3504 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3505 		VN_RELE(cs->vp);
3506 		cs->vp = NULL;
3507 		exi_rele(cs->exi);
3508 		cs->exi = sav_exi;
3509 		goto out;
3510 	}
3511 	if (sav_exi)
3512 		exi_rele(sav_exi);
3513 
3514 	*cs->statusp = resp->status = NFS4_OK;
3515 out:
3516 	DTRACE_NFSV4_2(op__putpubfh__done, struct compound_state *, cs,
3517 	    PUTPUBFH4res *, resp);
3518 }
3519 
3520 /*
3521  * XXX - issue with put*fh operations. Suppose /export/home is exported.
3522  * Suppose an NFS client goes to mount /export/home/joe. If /export, home,
3523  * or joe have restrictive search permissions, then we shouldn't let
3524  * the client get a file handle. This is easy to enforce. However, we
3525  * don't know what security flavor should be used until we resolve the
3526  * path name. Another complication is uid mapping. If root is
3527  * the user, then it will be mapped to the anonymous user by default,
3528  * but we won't know that till we've resolved the path name. And we won't
3529  * know what the anonymous user is.
3530  * Luckily, SECINFO is specified to take a full filename.
3531  * So what we will have to in rfs4_op_lookup is check that flavor of
3532  * the target object matches that of the request, and if root was the
3533  * caller, check for the root= and anon= options, and if necessary,
3534  * repeat the lookup using the right cred_t. But that's not done yet.
3535  */
3536 /* ARGSUSED */
3537 static void
3538 rfs4_op_putfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3539     struct compound_state *cs)
3540 {
3541 	PUTFH4args *args = &argop->nfs_argop4_u.opputfh;
3542 	PUTFH4res *resp = &resop->nfs_resop4_u.opputfh;
3543 	nfs_fh4_fmt_t *fh_fmtp;
3544 
3545 	DTRACE_NFSV4_2(op__putfh__start, struct compound_state *, cs,
3546 	    PUTFH4args *, args);
3547 
3548 	if (cs->vp) {
3549 		VN_RELE(cs->vp);
3550 		cs->vp = NULL;
3551 	}
3552 
3553 	if (cs->cr) {
3554 		crfree(cs->cr);
3555 		cs->cr = NULL;
3556 	}
3557 
3558 
3559 	if (args->object.nfs_fh4_len < NFS_FH4_LEN) {
3560 		*cs->statusp = resp->status = NFS4ERR_BADHANDLE;
3561 		goto out;
3562 	}
3563 
3564 	fh_fmtp = (nfs_fh4_fmt_t *)args->object.nfs_fh4_val;
3565 	if (cs->exi)
3566 		exi_rele(cs->exi);
3567 	cs->exi = checkexport(&fh_fmtp->fh4_fsid, (fid_t *)&fh_fmtp->fh4_xlen,
3568 	    NULL);
3569 
3570 	if (cs->exi == NULL) {
3571 		*cs->statusp = resp->status = NFS4ERR_STALE;
3572 		goto out;
3573 	}
3574 
3575 	cs->cr = crdup(cs->basecr);
3576 
3577 	ASSERT(cs->cr != NULL);
3578 
3579 	if (! (cs->vp = nfs4_fhtovp(&args->object, cs->exi, &resp->status))) {
3580 		*cs->statusp = resp->status;
3581 		goto out;
3582 	}
3583 
3584 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3585 		VN_RELE(cs->vp);
3586 		cs->vp = NULL;
3587 		goto out;
3588 	}
3589 
3590 	nfs_fh4_copy(&args->object, &cs->fh);
3591 	*cs->statusp = resp->status = NFS4_OK;
3592 	cs->deleg = FALSE;
3593 
3594 out:
3595 	DTRACE_NFSV4_2(op__putfh__done, struct compound_state *, cs,
3596 	    PUTFH4res *, resp);
3597 }
3598 
3599 /* ARGSUSED */
3600 static void
3601 rfs4_op_putrootfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3602     struct compound_state *cs)
3603 {
3604 	PUTROOTFH4res *resp = &resop->nfs_resop4_u.opputrootfh;
3605 	int error;
3606 	fid_t fid;
3607 	struct exportinfo *exi, *sav_exi;
3608 
3609 	DTRACE_NFSV4_1(op__putrootfh__start, struct compound_state *, cs);
3610 
3611 	if (cs->vp) {
3612 		VN_RELE(cs->vp);
3613 		cs->vp = NULL;
3614 	}
3615 
3616 	if (cs->cr)
3617 		crfree(cs->cr);
3618 
3619 	cs->cr = crdup(cs->basecr);
3620 
3621 	/*
3622 	 * Using rootdir, the system root vnode,
3623 	 * get its fid.
3624 	 */
3625 	bzero(&fid, sizeof (fid));
3626 	fid.fid_len = MAXFIDSZ;
3627 	error = vop_fid_pseudo(rootdir, &fid);
3628 	if (error != 0) {
3629 		*cs->statusp = resp->status = puterrno4(error);
3630 		goto out;
3631 	}
3632 
3633 	/*
3634 	 * Then use the root fsid & fid it to find out if it's exported
3635 	 *
3636 	 * If the server root isn't exported directly, then
3637 	 * it should at least be a pseudo export based on
3638 	 * one or more exports further down in the server's
3639 	 * file tree.
3640 	 */
3641 	exi = checkexport(&rootdir->v_vfsp->vfs_fsid, &fid, NULL);
3642 	if (exi == NULL || exi->exi_export.ex_flags & EX_PUBLIC) {
3643 		NFS4_DEBUG(rfs4_debug,
3644 		    (CE_WARN, "rfs4_op_putrootfh: export check failure"));
3645 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3646 		if (exi)
3647 			exi_rele(exi);
3648 		goto out;
3649 	}
3650 
3651 	/*
3652 	 * Now make a filehandle based on the root
3653 	 * export and root vnode.
3654 	 */
3655 	error = makefh4(&cs->fh, rootdir, exi);
3656 	if (error != 0) {
3657 		*cs->statusp = resp->status = puterrno4(error);
3658 		exi_rele(exi);
3659 		goto out;
3660 	}
3661 
3662 	sav_exi = cs->exi;
3663 	cs->exi = exi;
3664 
3665 	VN_HOLD(rootdir);
3666 	cs->vp = rootdir;
3667 
3668 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3669 		VN_RELE(rootdir);
3670 		cs->vp = NULL;
3671 		exi_rele(exi);
3672 		cs->exi = sav_exi;
3673 		goto out;
3674 	}
3675 	if (sav_exi)
3676 		exi_rele(sav_exi);
3677 
3678 	*cs->statusp = resp->status = NFS4_OK;
3679 	cs->deleg = FALSE;
3680 out:
3681 	DTRACE_NFSV4_2(op__putrootfh__done, struct compound_state *, cs,
3682 	    PUTROOTFH4res *, resp);
3683 }
3684 
3685 /*
3686  * set_rdattr_params sets up the variables used to manage what information
3687  * to get for each directory entry.
3688  */
3689 static nfsstat4
3690 set_rdattr_params(struct nfs4_svgetit_arg *sargp,
3691     bitmap4 attrs, bool_t *need_to_lookup)
3692 {
3693 	uint_t	va_mask;
3694 	nfsstat4 status;
3695 	bitmap4 objbits;
3696 
3697 	status = bitmap4_to_attrmask(attrs, sargp);
3698 	if (status != NFS4_OK) {
3699 		/*
3700 		 * could not even figure attr mask
3701 		 */
3702 		return (status);
3703 	}
3704 	va_mask = sargp->vap->va_mask;
3705 
3706 	/*
3707 	 * dirent's d_ino is always correct value for mounted_on_fileid.
3708 	 * mntdfid_set is set once here, but mounted_on_fileid is
3709 	 * set in main dirent processing loop for each dirent.
3710 	 * The mntdfid_set is a simple optimization that lets the
3711 	 * server attr code avoid work when caller is readdir.
3712 	 */
3713 	sargp->mntdfid_set = TRUE;
3714 
3715 	/*
3716 	 * Lookup entry only if client asked for any of the following:
3717 	 * a) vattr attrs
3718 	 * b) vfs attrs
3719 	 * c) attrs w/per-object scope requested (change, filehandle, etc)
3720 	 *    other than mounted_on_fileid (which we can take from dirent)
3721 	 */
3722 	objbits = attrs ? attrs & NFS4_VP_ATTR_MASK : 0;
3723 
3724 	if (va_mask || sargp->sbp || (objbits & ~FATTR4_MOUNTED_ON_FILEID_MASK))
3725 		*need_to_lookup = TRUE;
3726 	else
3727 		*need_to_lookup = FALSE;
3728 
3729 	if (sargp->sbp == NULL)
3730 		return (NFS4_OK);
3731 
3732 	/*
3733 	 * If filesystem attrs are requested, get them now from the
3734 	 * directory vp, as most entries will have same filesystem. The only
3735 	 * exception are mounted over entries but we handle
3736 	 * those as we go (XXX mounted over detection not yet implemented).
3737 	 */
3738 	sargp->vap->va_mask = 0;	/* to avoid VOP_GETATTR */
3739 	status = bitmap4_get_sysattrs(sargp);
3740 	sargp->vap->va_mask = va_mask;
3741 
3742 	if ((status != NFS4_OK) && sargp->rdattr_error_req) {
3743 		/*
3744 		 * Failed to get filesystem attributes.
3745 		 * Return a rdattr_error for each entry, but don't fail.
3746 		 * However, don't get any obj-dependent attrs.
3747 		 */
3748 		sargp->rdattr_error = status;	/* for rdattr_error */
3749 		*need_to_lookup = FALSE;
3750 		/*
3751 		 * At least get fileid for regular readdir output
3752 		 */
3753 		sargp->vap->va_mask &= AT_NODEID;
3754 		status = NFS4_OK;
3755 	}
3756 
3757 	return (status);
3758 }
3759 
3760 /*
3761  * readlink: args: CURRENT_FH.
3762  *	res: status. If success - CURRENT_FH unchanged, return linktext.
3763  */
3764 
3765 /* ARGSUSED */
3766 static void
3767 rfs4_op_readlink(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3768     struct compound_state *cs)
3769 {
3770 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3771 	int error;
3772 	vnode_t *vp;
3773 	struct iovec iov;
3774 	struct vattr va;
3775 	struct uio uio;
3776 	char *data;
3777 	struct sockaddr *ca;
3778 	char *name = NULL;
3779 	int is_referral;
3780 
3781 	DTRACE_NFSV4_1(op__readlink__start, struct compound_state *, cs);
3782 
3783 	/* CURRENT_FH: directory */
3784 	vp = cs->vp;
3785 	if (vp == NULL) {
3786 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3787 		goto out;
3788 	}
3789 
3790 	if (cs->access == CS_ACCESS_DENIED) {
3791 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3792 		goto out;
3793 	}
3794 
3795 	/* Is it a referral? */
3796 	if (vn_is_nfs_reparse(vp, cs->cr) && client_is_downrev(req)) {
3797 
3798 		is_referral = 1;
3799 
3800 	} else {
3801 
3802 		is_referral = 0;
3803 
3804 		if (vp->v_type == VDIR) {
3805 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
3806 			goto out;
3807 		}
3808 
3809 		if (vp->v_type != VLNK) {
3810 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3811 			goto out;
3812 		}
3813 
3814 	}
3815 
3816 	va.va_mask = AT_MODE;
3817 	error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
3818 	if (error) {
3819 		*cs->statusp = resp->status = puterrno4(error);
3820 		goto out;
3821 	}
3822 
3823 	if (MANDLOCK(vp, va.va_mode)) {
3824 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3825 		goto out;
3826 	}
3827 
3828 	data = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);
3829 
3830 	if (is_referral) {
3831 		char *s;
3832 		size_t strsz;
3833 
3834 		/* Get an artificial symlink based on a referral */
3835 		s = build_symlink(vp, cs->cr, &strsz);
3836 		global_svstat_ptr[4][NFS_REFERLINKS].value.ui64++;
3837 		DTRACE_PROBE2(nfs4serv__func__referral__reflink,
3838 		    vnode_t *, vp, char *, s);
3839 		if (s == NULL)
3840 			error = EINVAL;
3841 		else {
3842 			error = 0;
3843 			(void) strlcpy(data, s, MAXPATHLEN + 1);
3844 			kmem_free(s, strsz);
3845 		}
3846 
3847 	} else {
3848 
3849 		iov.iov_base = data;
3850 		iov.iov_len = MAXPATHLEN;
3851 		uio.uio_iov = &iov;
3852 		uio.uio_iovcnt = 1;
3853 		uio.uio_segflg = UIO_SYSSPACE;
3854 		uio.uio_extflg = UIO_COPY_CACHED;
3855 		uio.uio_loffset = 0;
3856 		uio.uio_resid = MAXPATHLEN;
3857 
3858 		error = VOP_READLINK(vp, &uio, cs->cr, NULL);
3859 
3860 		if (!error)
3861 			*(data + MAXPATHLEN - uio.uio_resid) = '\0';
3862 	}
3863 
3864 	if (error) {
3865 		kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3866 		*cs->statusp = resp->status = puterrno4(error);
3867 		goto out;
3868 	}
3869 
3870 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
3871 	name = nfscmd_convname(ca, cs->exi, data, NFSCMD_CONV_OUTBOUND,
3872 	    MAXPATHLEN  + 1);
3873 
3874 	if (name == NULL) {
3875 		/*
3876 		 * Even though the conversion failed, we return
3877 		 * something. We just don't translate it.
3878 		 */
3879 		name = data;
3880 	}
3881 
3882 	/*
3883 	 * treat link name as data
3884 	 */
3885 	(void) str_to_utf8(name, (utf8string *)&resp->link);
3886 
3887 	if (name != data)
3888 		kmem_free(name, MAXPATHLEN + 1);
3889 	kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3890 	*cs->statusp = resp->status = NFS4_OK;
3891 
3892 out:
3893 	DTRACE_NFSV4_2(op__readlink__done, struct compound_state *, cs,
3894 	    READLINK4res *, resp);
3895 }
3896 
3897 static void
3898 rfs4_op_readlink_free(nfs_resop4 *resop)
3899 {
3900 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3901 	utf8string *symlink = (utf8string *)&resp->link;
3902 
3903 	if (symlink->utf8string_val) {
3904 		UTF8STRING_FREE(*symlink)
3905 	}
3906 }
3907 
3908 /*
3909  * release_lockowner:
3910  *	Release any state associated with the supplied
3911  *	lockowner. Note if any lo_state is holding locks we will not
3912  *	rele that lo_state and thus the lockowner will not be destroyed.
3913  *	A client using lock after the lock owner stateid has been released
3914  *	will suffer the consequence of NFS4ERR_BAD_STATEID and would have
3915  *	to reissue the lock with new_lock_owner set to TRUE.
3916  *	args: lock_owner
3917  *	res:  status
3918  */
3919 /* ARGSUSED */
3920 static void
3921 rfs4_op_release_lockowner(nfs_argop4 *argop, nfs_resop4 *resop,
3922     struct svc_req *req, struct compound_state *cs)
3923 {
3924 	RELEASE_LOCKOWNER4args *ap = &argop->nfs_argop4_u.oprelease_lockowner;
3925 	RELEASE_LOCKOWNER4res *resp = &resop->nfs_resop4_u.oprelease_lockowner;
3926 	rfs4_lockowner_t *lo;
3927 	rfs4_openowner_t *oo;
3928 	rfs4_state_t *sp;
3929 	rfs4_lo_state_t *lsp;
3930 	rfs4_client_t *cp;
3931 	bool_t create = FALSE;
3932 	locklist_t *llist;
3933 	sysid_t sysid;
3934 
3935 	DTRACE_NFSV4_2(op__release__lockowner__start, struct compound_state *,
3936 	    cs, RELEASE_LOCKOWNER4args *, ap);
3937 
3938 	/* Make sure there is a clientid around for this request */
3939 	cp = rfs4_findclient_by_id(ap->lock_owner.clientid, FALSE);
3940 
3941 	if (cp == NULL) {
3942 		*cs->statusp = resp->status =
3943 		    rfs4_check_clientid(&ap->lock_owner.clientid, 0);
3944 		goto out;
3945 	}
3946 	rfs4_client_rele(cp);
3947 
3948 	lo = rfs4_findlockowner(&ap->lock_owner, &create);
3949 	if (lo == NULL) {
3950 		*cs->statusp = resp->status = NFS4_OK;
3951 		goto out;
3952 	}
3953 	ASSERT(lo->rl_client != NULL);
3954 
3955 	/*
3956 	 * Check for EXPIRED client. If so will reap state with in a lease
3957 	 * period or on next set_clientid_confirm step
3958 	 */
3959 	if (rfs4_lease_expired(lo->rl_client)) {
3960 		rfs4_lockowner_rele(lo);
3961 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
3962 		goto out;
3963 	}
3964 
3965 	/*
3966 	 * If no sysid has been assigned, then no locks exist; just return.
3967 	 */
3968 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3969 	if (lo->rl_client->rc_sysidt == LM_NOSYSID) {
3970 		rfs4_lockowner_rele(lo);
3971 		rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3972 		goto out;
3973 	}
3974 
3975 	sysid = lo->rl_client->rc_sysidt;
3976 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3977 
3978 	/*
3979 	 * Mark the lockowner invalid.
3980 	 */
3981 	rfs4_dbe_hide(lo->rl_dbe);
3982 
3983 	/*
3984 	 * sysid-pid pair should now not be used since the lockowner is
3985 	 * invalid. If the client were to instantiate the lockowner again
3986 	 * it would be assigned a new pid. Thus we can get the list of
3987 	 * current locks.
3988 	 */
3989 
3990 	llist = flk_get_active_locks(sysid, lo->rl_pid);
3991 	/* If we are still holding locks fail */
3992 	if (llist != NULL) {
3993 
3994 		*cs->statusp = resp->status = NFS4ERR_LOCKS_HELD;
3995 
3996 		flk_free_locklist(llist);
3997 		/*
3998 		 * We need to unhide the lockowner so the client can
3999 		 * try it again. The bad thing here is if the client
4000 		 * has a logic error that took it here in the first place
4001 		 * he probably has lost accounting of the locks that it
4002 		 * is holding. So we may have dangling state until the
4003 		 * open owner state is reaped via close. One scenario
4004 		 * that could possibly occur is that the client has
4005 		 * sent the unlock request(s) in separate threads
4006 		 * and has not waited for the replies before sending the
4007 		 * RELEASE_LOCKOWNER request. Presumably, it would expect
4008 		 * and deal appropriately with NFS4ERR_LOCKS_HELD, by
4009 		 * reissuing the request.
4010 		 */
4011 		rfs4_dbe_unhide(lo->rl_dbe);
4012 		rfs4_lockowner_rele(lo);
4013 		goto out;
4014 	}
4015 
4016 	/*
4017 	 * For the corresponding client we need to check each open
4018 	 * owner for any opens that have lockowner state associated
4019 	 * with this lockowner.
4020 	 */
4021 
4022 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
4023 	for (oo = list_head(&lo->rl_client->rc_openownerlist); oo != NULL;
4024 	    oo = list_next(&lo->rl_client->rc_openownerlist, oo)) {
4025 
4026 		rfs4_dbe_lock(oo->ro_dbe);
4027 		for (sp = list_head(&oo->ro_statelist); sp != NULL;
4028 		    sp = list_next(&oo->ro_statelist, sp)) {
4029 
4030 			rfs4_dbe_lock(sp->rs_dbe);
4031 			for (lsp = list_head(&sp->rs_lostatelist);
4032 			    lsp != NULL;
4033 			    lsp = list_next(&sp->rs_lostatelist, lsp)) {
4034 				if (lsp->rls_locker == lo) {
4035 					rfs4_dbe_lock(lsp->rls_dbe);
4036 					rfs4_dbe_invalidate(lsp->rls_dbe);
4037 					rfs4_dbe_unlock(lsp->rls_dbe);
4038 				}
4039 			}
4040 			rfs4_dbe_unlock(sp->rs_dbe);
4041 		}
4042 		rfs4_dbe_unlock(oo->ro_dbe);
4043 	}
4044 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
4045 
4046 	rfs4_lockowner_rele(lo);
4047 
4048 	*cs->statusp = resp->status = NFS4_OK;
4049 
4050 out:
4051 	DTRACE_NFSV4_2(op__release__lockowner__done, struct compound_state *,
4052 	    cs, RELEASE_LOCKOWNER4res *, resp);
4053 }
4054 
4055 /*
4056  * short utility function to lookup a file and recall the delegation
4057  */
4058 static rfs4_file_t *
4059 rfs4_lookup_and_findfile(vnode_t *dvp, char *nm, vnode_t **vpp,
4060     int *lkup_error, cred_t *cr)
4061 {
4062 	vnode_t *vp;
4063 	rfs4_file_t *fp = NULL;
4064 	bool_t fcreate = FALSE;
4065 	int error;
4066 
4067 	if (vpp)
4068 		*vpp = NULL;
4069 
4070 	if ((error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cr, NULL, NULL,
4071 	    NULL)) == 0) {
4072 		if (vp->v_type == VREG)
4073 			fp = rfs4_findfile(vp, NULL, &fcreate);
4074 		if (vpp)
4075 			*vpp = vp;
4076 		else
4077 			VN_RELE(vp);
4078 	}
4079 
4080 	if (lkup_error)
4081 		*lkup_error = error;
4082 
4083 	return (fp);
4084 }
4085 
4086 /*
4087  * remove: args: CURRENT_FH: directory; name.
4088  *	res: status. If success - CURRENT_FH unchanged, return change_info
4089  *		for directory.
4090  */
4091 /* ARGSUSED */
4092 static void
4093 rfs4_op_remove(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4094     struct compound_state *cs)
4095 {
4096 	REMOVE4args *args = &argop->nfs_argop4_u.opremove;
4097 	REMOVE4res *resp = &resop->nfs_resop4_u.opremove;
4098 	int error;
4099 	vnode_t *dvp, *vp;
4100 	struct vattr bdva, idva, adva;
4101 	char *nm;
4102 	uint_t len;
4103 	rfs4_file_t *fp;
4104 	int in_crit = 0;
4105 	bslabel_t *clabel;
4106 	struct sockaddr *ca;
4107 	char *name = NULL;
4108 	nfsstat4 status;
4109 
4110 	DTRACE_NFSV4_2(op__remove__start, struct compound_state *, cs,
4111 	    REMOVE4args *, args);
4112 
4113 	/* CURRENT_FH: directory */
4114 	dvp = cs->vp;
4115 	if (dvp == NULL) {
4116 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4117 		goto out;
4118 	}
4119 
4120 	if (cs->access == CS_ACCESS_DENIED) {
4121 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4122 		goto out;
4123 	}
4124 
4125 	/*
4126 	 * If there is an unshared filesystem mounted on this vnode,
4127 	 * Do not allow to remove anything in this directory.
4128 	 */
4129 	if (vn_ismntpt(dvp)) {
4130 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4131 		goto out;
4132 	}
4133 
4134 	if (dvp->v_type != VDIR) {
4135 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4136 		goto out;
4137 	}
4138 
4139 	status = utf8_dir_verify(&args->target);
4140 	if (status != NFS4_OK) {
4141 		*cs->statusp = resp->status = status;
4142 		goto out;
4143 	}
4144 
4145 	/*
4146 	 * Lookup the file so that we can check if it's a directory
4147 	 */
4148 	nm = utf8_to_fn(&args->target, &len, NULL);
4149 	if (nm == NULL) {
4150 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4151 		goto out;
4152 	}
4153 
4154 	if (len > MAXNAMELEN) {
4155 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4156 		kmem_free(nm, len);
4157 		goto out;
4158 	}
4159 
4160 	if (rdonly4(req, cs)) {
4161 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4162 		kmem_free(nm, len);
4163 		goto out;
4164 	}
4165 
4166 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4167 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
4168 	    MAXPATHLEN  + 1);
4169 
4170 	if (name == NULL) {
4171 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4172 		kmem_free(nm, len);
4173 		goto out;
4174 	}
4175 
4176 	/*
4177 	 * Lookup the file to determine type and while we are see if
4178 	 * there is a file struct around and check for delegation.
4179 	 * We don't need to acquire va_seq before this lookup, if
4180 	 * it causes an update, cinfo.before will not match, which will
4181 	 * trigger a cache flush even if atomic is TRUE.
4182 	 */
4183 	if (fp = rfs4_lookup_and_findfile(dvp, name, &vp, &error, cs->cr)) {
4184 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4185 		    NULL)) {
4186 			VN_RELE(vp);
4187 			rfs4_file_rele(fp);
4188 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4189 			if (nm != name)
4190 				kmem_free(name, MAXPATHLEN + 1);
4191 			kmem_free(nm, len);
4192 			goto out;
4193 		}
4194 	}
4195 
4196 	/* Didn't find anything to remove */
4197 	if (vp == NULL) {
4198 		*cs->statusp = resp->status = error;
4199 		if (nm != name)
4200 			kmem_free(name, MAXPATHLEN + 1);
4201 		kmem_free(nm, len);
4202 		goto out;
4203 	}
4204 
4205 	if (nbl_need_check(vp)) {
4206 		nbl_start_crit(vp, RW_READER);
4207 		in_crit = 1;
4208 		if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
4209 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4210 			if (nm != name)
4211 				kmem_free(name, MAXPATHLEN + 1);
4212 			kmem_free(nm, len);
4213 			nbl_end_crit(vp);
4214 			VN_RELE(vp);
4215 			if (fp) {
4216 				rfs4_clear_dont_grant(fp);
4217 				rfs4_file_rele(fp);
4218 			}
4219 			goto out;
4220 		}
4221 	}
4222 
4223 	/* check label before allowing removal */
4224 	if (is_system_labeled()) {
4225 		ASSERT(req->rq_label != NULL);
4226 		clabel = req->rq_label;
4227 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
4228 		    "got client label from request(1)",
4229 		    struct svc_req *, req);
4230 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4231 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
4232 			    cs->exi)) {
4233 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4234 				if (name != nm)
4235 					kmem_free(name, MAXPATHLEN + 1);
4236 				kmem_free(nm, len);
4237 				if (in_crit)
4238 					nbl_end_crit(vp);
4239 				VN_RELE(vp);
4240 				if (fp) {
4241 					rfs4_clear_dont_grant(fp);
4242 					rfs4_file_rele(fp);
4243 				}
4244 				goto out;
4245 			}
4246 		}
4247 	}
4248 
4249 	/* Get dir "before" change value */
4250 	bdva.va_mask = AT_CTIME|AT_SEQ;
4251 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
4252 	if (error) {
4253 		*cs->statusp = resp->status = puterrno4(error);
4254 		if (nm != name)
4255 			kmem_free(name, MAXPATHLEN + 1);
4256 		kmem_free(nm, len);
4257 		if (in_crit)
4258 			nbl_end_crit(vp);
4259 		VN_RELE(vp);
4260 		if (fp) {
4261 			rfs4_clear_dont_grant(fp);
4262 			rfs4_file_rele(fp);
4263 		}
4264 		goto out;
4265 	}
4266 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
4267 
4268 	/* Actually do the REMOVE operation */
4269 	if (vp->v_type == VDIR) {
4270 		/*
4271 		 * Can't remove a directory that has a mounted-on filesystem.
4272 		 */
4273 		if (vn_ismntpt(vp)) {
4274 			error = EACCES;
4275 		} else {
4276 			/*
4277 			 * System V defines rmdir to return EEXIST,
4278 			 * not ENOTEMPTY, if the directory is not
4279 			 * empty.  A System V NFS server needs to map
4280 			 * NFS4ERR_EXIST to NFS4ERR_NOTEMPTY to
4281 			 * transmit over the wire.
4282 			 */
4283 			if ((error = VOP_RMDIR(dvp, name, rootdir, cs->cr,
4284 			    NULL, 0)) == EEXIST)
4285 				error = ENOTEMPTY;
4286 		}
4287 
4288 		if (in_crit)
4289 			nbl_end_crit(vp);
4290 		VN_RELE(vp);
4291 	} else {
4292 		if (!in_crit)
4293 			VN_RELE(vp);
4294 		if ((error = VOP_REMOVE(dvp, name, cs->cr, NULL, 0)) == 0 &&
4295 		    fp != NULL) {
4296 			struct vattr va;
4297 			vnode_t *tvp;
4298 
4299 			rfs4_dbe_lock(fp->rf_dbe);
4300 			tvp = fp->rf_vp;
4301 			if (tvp)
4302 				VN_HOLD(tvp);
4303 			rfs4_dbe_unlock(fp->rf_dbe);
4304 
4305 			if (tvp) {
4306 				/*
4307 				 * This is va_seq safe because we are not
4308 				 * manipulating dvp.
4309 				 */
4310 				va.va_mask = AT_NLINK;
4311 				if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4312 				    va.va_nlink == 0) {
4313 					/* Remove state on file remove */
4314 					if (in_crit) {
4315 						nbl_end_crit(vp);
4316 						VN_RELE(vp);
4317 						in_crit = 0;
4318 					}
4319 					rfs4_close_all_state(fp);
4320 				}
4321 				VN_RELE(tvp);
4322 			}
4323 		}
4324 
4325 		if (in_crit) {
4326 			nbl_end_crit(vp);
4327 			VN_RELE(vp);
4328 		}
4329 	}
4330 
4331 	if (fp) {
4332 		rfs4_clear_dont_grant(fp);
4333 		rfs4_file_rele(fp);
4334 	}
4335 	if (nm != name)
4336 		kmem_free(name, MAXPATHLEN + 1);
4337 	kmem_free(nm, len);
4338 
4339 	if (error) {
4340 		*cs->statusp = resp->status = puterrno4(error);
4341 		goto out;
4342 	}
4343 
4344 	/*
4345 	 * Get the initial "after" sequence number, if it fails, set to zero
4346 	 */
4347 	idva.va_mask = AT_SEQ;
4348 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
4349 		idva.va_seq = 0;
4350 
4351 	/*
4352 	 * Force modified data and metadata out to stable storage.
4353 	 */
4354 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
4355 
4356 	/*
4357 	 * Get "after" change value, if it fails, simply return the
4358 	 * before value.
4359 	 */
4360 	adva.va_mask = AT_CTIME|AT_SEQ;
4361 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
4362 		adva.va_ctime = bdva.va_ctime;
4363 		adva.va_seq = 0;
4364 	}
4365 
4366 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
4367 
4368 	/*
4369 	 * The cinfo.atomic = TRUE only if we have
4370 	 * non-zero va_seq's, and it has incremented by exactly one
4371 	 * during the VOP_REMOVE/RMDIR and it didn't change during
4372 	 * the VOP_FSYNC.
4373 	 */
4374 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
4375 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
4376 		resp->cinfo.atomic = TRUE;
4377 	else
4378 		resp->cinfo.atomic = FALSE;
4379 
4380 	*cs->statusp = resp->status = NFS4_OK;
4381 
4382 out:
4383 	DTRACE_NFSV4_2(op__remove__done, struct compound_state *, cs,
4384 	    REMOVE4res *, resp);
4385 }
4386 
4387 /*
4388  * rename: args: SAVED_FH: from directory, CURRENT_FH: target directory,
4389  *		oldname and newname.
4390  *	res: status. If success - CURRENT_FH unchanged, return change_info
4391  *		for both from and target directories.
4392  */
4393 /* ARGSUSED */
4394 static void
4395 rfs4_op_rename(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4396     struct compound_state *cs)
4397 {
4398 	RENAME4args *args = &argop->nfs_argop4_u.oprename;
4399 	RENAME4res *resp = &resop->nfs_resop4_u.oprename;
4400 	int error;
4401 	vnode_t *odvp;
4402 	vnode_t *ndvp;
4403 	vnode_t *srcvp, *targvp;
4404 	struct vattr obdva, oidva, oadva;
4405 	struct vattr nbdva, nidva, nadva;
4406 	char *onm, *nnm;
4407 	uint_t olen, nlen;
4408 	rfs4_file_t *fp, *sfp;
4409 	int in_crit_src, in_crit_targ;
4410 	int fp_rele_grant_hold, sfp_rele_grant_hold;
4411 	bslabel_t *clabel;
4412 	struct sockaddr *ca;
4413 	char *converted_onm = NULL;
4414 	char *converted_nnm = NULL;
4415 	nfsstat4 status;
4416 
4417 	DTRACE_NFSV4_2(op__rename__start, struct compound_state *, cs,
4418 	    RENAME4args *, args);
4419 
4420 	fp = sfp = NULL;
4421 	srcvp = targvp = NULL;
4422 	in_crit_src = in_crit_targ = 0;
4423 	fp_rele_grant_hold = sfp_rele_grant_hold = 0;
4424 
4425 	/* CURRENT_FH: target directory */
4426 	ndvp = cs->vp;
4427 	if (ndvp == NULL) {
4428 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4429 		goto out;
4430 	}
4431 
4432 	/* SAVED_FH: from directory */
4433 	odvp = cs->saved_vp;
4434 	if (odvp == NULL) {
4435 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4436 		goto out;
4437 	}
4438 
4439 	if (cs->access == CS_ACCESS_DENIED) {
4440 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4441 		goto out;
4442 	}
4443 
4444 	/*
4445 	 * If there is an unshared filesystem mounted on this vnode,
4446 	 * do not allow to rename objects in this directory.
4447 	 */
4448 	if (vn_ismntpt(odvp)) {
4449 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4450 		goto out;
4451 	}
4452 
4453 	/*
4454 	 * If there is an unshared filesystem mounted on this vnode,
4455 	 * do not allow to rename to this directory.
4456 	 */
4457 	if (vn_ismntpt(ndvp)) {
4458 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4459 		goto out;
4460 	}
4461 
4462 	if (odvp->v_type != VDIR || ndvp->v_type != VDIR) {
4463 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4464 		goto out;
4465 	}
4466 
4467 	if (cs->saved_exi != cs->exi) {
4468 		*cs->statusp = resp->status = NFS4ERR_XDEV;
4469 		goto out;
4470 	}
4471 
4472 	status = utf8_dir_verify(&args->oldname);
4473 	if (status != NFS4_OK) {
4474 		*cs->statusp = resp->status = status;
4475 		goto out;
4476 	}
4477 
4478 	status = utf8_dir_verify(&args->newname);
4479 	if (status != NFS4_OK) {
4480 		*cs->statusp = resp->status = status;
4481 		goto out;
4482 	}
4483 
4484 	onm = utf8_to_fn(&args->oldname, &olen, NULL);
4485 	if (onm == NULL) {
4486 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4487 		goto out;
4488 	}
4489 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4490 	nlen = MAXPATHLEN + 1;
4491 	converted_onm = nfscmd_convname(ca, cs->exi, onm, NFSCMD_CONV_INBOUND,
4492 	    nlen);
4493 
4494 	if (converted_onm == NULL) {
4495 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4496 		kmem_free(onm, olen);
4497 		goto out;
4498 	}
4499 
4500 	nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4501 	if (nnm == NULL) {
4502 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4503 		if (onm != converted_onm)
4504 			kmem_free(converted_onm, MAXPATHLEN + 1);
4505 		kmem_free(onm, olen);
4506 		goto out;
4507 	}
4508 	converted_nnm = nfscmd_convname(ca, cs->exi, nnm, NFSCMD_CONV_INBOUND,
4509 	    MAXPATHLEN  + 1);
4510 
4511 	if (converted_nnm == NULL) {
4512 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4513 		kmem_free(nnm, nlen);
4514 		nnm = NULL;
4515 		if (onm != converted_onm)
4516 			kmem_free(converted_onm, MAXPATHLEN + 1);
4517 		kmem_free(onm, olen);
4518 		goto out;
4519 	}
4520 
4521 
4522 	if (olen > MAXNAMELEN || nlen > MAXNAMELEN) {
4523 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4524 		kmem_free(onm, olen);
4525 		kmem_free(nnm, nlen);
4526 		goto out;
4527 	}
4528 
4529 
4530 	if (rdonly4(req, cs)) {
4531 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4532 		if (onm != converted_onm)
4533 			kmem_free(converted_onm, MAXPATHLEN + 1);
4534 		kmem_free(onm, olen);
4535 		if (nnm != converted_nnm)
4536 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4537 		kmem_free(nnm, nlen);
4538 		goto out;
4539 	}
4540 
4541 	/* check label of the target dir */
4542 	if (is_system_labeled()) {
4543 		ASSERT(req->rq_label != NULL);
4544 		clabel = req->rq_label;
4545 		DTRACE_PROBE2(tx__rfs4__log__info__oprename__clabel, char *,
4546 		    "got client label from request(1)",
4547 		    struct svc_req *, req);
4548 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4549 			if (!do_rfs_label_check(clabel, ndvp,
4550 			    EQUALITY_CHECK, cs->exi)) {
4551 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4552 				goto err_out;
4553 			}
4554 		}
4555 	}
4556 
4557 	/*
4558 	 * Is the source a file and have a delegation?
4559 	 * We don't need to acquire va_seq before these lookups, if
4560 	 * it causes an update, cinfo.before will not match, which will
4561 	 * trigger a cache flush even if atomic is TRUE.
4562 	 */
4563 	if (sfp = rfs4_lookup_and_findfile(odvp, converted_onm, &srcvp,
4564 	    &error, cs->cr)) {
4565 		if (rfs4_check_delegated_byfp(FWRITE, sfp, TRUE, TRUE, TRUE,
4566 		    NULL)) {
4567 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4568 			goto err_out;
4569 		}
4570 	}
4571 
4572 	if (srcvp == NULL) {
4573 		*cs->statusp = resp->status = puterrno4(error);
4574 		if (onm != converted_onm)
4575 			kmem_free(converted_onm, MAXPATHLEN + 1);
4576 		kmem_free(onm, olen);
4577 		if (nnm != converted_nnm)
4578 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4579 		kmem_free(nnm, nlen);
4580 		goto out;
4581 	}
4582 
4583 	sfp_rele_grant_hold = 1;
4584 
4585 	/* Does the destination exist and a file and have a delegation? */
4586 	if (fp = rfs4_lookup_and_findfile(ndvp, converted_nnm, &targvp,
4587 	    NULL, cs->cr)) {
4588 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4589 		    NULL)) {
4590 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4591 			goto err_out;
4592 		}
4593 	}
4594 	fp_rele_grant_hold = 1;
4595 
4596 
4597 	/* Check for NBMAND lock on both source and target */
4598 	if (nbl_need_check(srcvp)) {
4599 		nbl_start_crit(srcvp, RW_READER);
4600 		in_crit_src = 1;
4601 		if (nbl_conflict(srcvp, NBL_RENAME, 0, 0, 0, NULL)) {
4602 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4603 			goto err_out;
4604 		}
4605 	}
4606 
4607 	if (targvp && nbl_need_check(targvp)) {
4608 		nbl_start_crit(targvp, RW_READER);
4609 		in_crit_targ = 1;
4610 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
4611 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4612 			goto err_out;
4613 		}
4614 	}
4615 
4616 	/* Get source "before" change value */
4617 	obdva.va_mask = AT_CTIME|AT_SEQ;
4618 	error = VOP_GETATTR(odvp, &obdva, 0, cs->cr, NULL);
4619 	if (!error) {
4620 		nbdva.va_mask = AT_CTIME|AT_SEQ;
4621 		error = VOP_GETATTR(ndvp, &nbdva, 0, cs->cr, NULL);
4622 	}
4623 	if (error) {
4624 		*cs->statusp = resp->status = puterrno4(error);
4625 		goto err_out;
4626 	}
4627 
4628 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.before, obdva.va_ctime)
4629 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.before, nbdva.va_ctime)
4630 
4631 	if ((error = VOP_RENAME(odvp, converted_onm, ndvp, converted_nnm,
4632 	    cs->cr, NULL, 0)) == 0 && fp != NULL) {
4633 		struct vattr va;
4634 		vnode_t *tvp;
4635 
4636 		rfs4_dbe_lock(fp->rf_dbe);
4637 		tvp = fp->rf_vp;
4638 		if (tvp)
4639 			VN_HOLD(tvp);
4640 		rfs4_dbe_unlock(fp->rf_dbe);
4641 
4642 		if (tvp) {
4643 			va.va_mask = AT_NLINK;
4644 			if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4645 			    va.va_nlink == 0) {
4646 				/* The file is gone and so should the state */
4647 				if (in_crit_targ) {
4648 					nbl_end_crit(targvp);
4649 					in_crit_targ = 0;
4650 				}
4651 				rfs4_close_all_state(fp);
4652 			}
4653 			VN_RELE(tvp);
4654 		}
4655 	}
4656 	if (error == 0)
4657 		vn_renamepath(ndvp, srcvp, nnm, nlen - 1);
4658 
4659 	if (in_crit_src)
4660 		nbl_end_crit(srcvp);
4661 	if (srcvp)
4662 		VN_RELE(srcvp);
4663 	if (in_crit_targ)
4664 		nbl_end_crit(targvp);
4665 	if (targvp)
4666 		VN_RELE(targvp);
4667 
4668 	if (sfp) {
4669 		rfs4_clear_dont_grant(sfp);
4670 		rfs4_file_rele(sfp);
4671 	}
4672 	if (fp) {
4673 		rfs4_clear_dont_grant(fp);
4674 		rfs4_file_rele(fp);
4675 	}
4676 
4677 	if (converted_onm != onm)
4678 		kmem_free(converted_onm, MAXPATHLEN + 1);
4679 	kmem_free(onm, olen);
4680 	if (converted_nnm != nnm)
4681 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4682 	kmem_free(nnm, nlen);
4683 
4684 	/*
4685 	 * Get the initial "after" sequence number, if it fails, set to zero
4686 	 */
4687 	oidva.va_mask = AT_SEQ;
4688 	if (VOP_GETATTR(odvp, &oidva, 0, cs->cr, NULL))
4689 		oidva.va_seq = 0;
4690 
4691 	nidva.va_mask = AT_SEQ;
4692 	if (VOP_GETATTR(ndvp, &nidva, 0, cs->cr, NULL))
4693 		nidva.va_seq = 0;
4694 
4695 	/*
4696 	 * Force modified data and metadata out to stable storage.
4697 	 */
4698 	(void) VOP_FSYNC(odvp, 0, cs->cr, NULL);
4699 	(void) VOP_FSYNC(ndvp, 0, cs->cr, NULL);
4700 
4701 	if (error) {
4702 		*cs->statusp = resp->status = puterrno4(error);
4703 		goto out;
4704 	}
4705 
4706 	/*
4707 	 * Get "after" change values, if it fails, simply return the
4708 	 * before value.
4709 	 */
4710 	oadva.va_mask = AT_CTIME|AT_SEQ;
4711 	if (VOP_GETATTR(odvp, &oadva, 0, cs->cr, NULL)) {
4712 		oadva.va_ctime = obdva.va_ctime;
4713 		oadva.va_seq = 0;
4714 	}
4715 
4716 	nadva.va_mask = AT_CTIME|AT_SEQ;
4717 	if (VOP_GETATTR(odvp, &nadva, 0, cs->cr, NULL)) {
4718 		nadva.va_ctime = nbdva.va_ctime;
4719 		nadva.va_seq = 0;
4720 	}
4721 
4722 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.after, oadva.va_ctime)
4723 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.after, nadva.va_ctime)
4724 
4725 	/*
4726 	 * The cinfo.atomic = TRUE only if we have
4727 	 * non-zero va_seq's, and it has incremented by exactly one
4728 	 * during the VOP_RENAME and it didn't change during the VOP_FSYNC.
4729 	 */
4730 	if (obdva.va_seq && oidva.va_seq && oadva.va_seq &&
4731 	    oidva.va_seq == (obdva.va_seq + 1) && oidva.va_seq == oadva.va_seq)
4732 		resp->source_cinfo.atomic = TRUE;
4733 	else
4734 		resp->source_cinfo.atomic = FALSE;
4735 
4736 	if (nbdva.va_seq && nidva.va_seq && nadva.va_seq &&
4737 	    nidva.va_seq == (nbdva.va_seq + 1) && nidva.va_seq == nadva.va_seq)
4738 		resp->target_cinfo.atomic = TRUE;
4739 	else
4740 		resp->target_cinfo.atomic = FALSE;
4741 
4742 #ifdef	VOLATILE_FH_TEST
4743 	{
4744 	extern void add_volrnm_fh(struct exportinfo *, vnode_t *);
4745 
4746 	/*
4747 	 * Add the renamed file handle to the volatile rename list
4748 	 */
4749 	if (cs->exi->exi_export.ex_flags & EX_VOLRNM) {
4750 		/* file handles may expire on rename */
4751 		vnode_t *vp;
4752 
4753 		nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4754 		/*
4755 		 * Already know that nnm will be a valid string
4756 		 */
4757 		error = VOP_LOOKUP(ndvp, nnm, &vp, NULL, 0, NULL, cs->cr,
4758 		    NULL, NULL, NULL);
4759 		kmem_free(nnm, nlen);
4760 		if (!error) {
4761 			add_volrnm_fh(cs->exi, vp);
4762 			VN_RELE(vp);
4763 		}
4764 	}
4765 	}
4766 #endif	/* VOLATILE_FH_TEST */
4767 
4768 	*cs->statusp = resp->status = NFS4_OK;
4769 out:
4770 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4771 	    RENAME4res *, resp);
4772 	return;
4773 
4774 err_out:
4775 	if (onm != converted_onm)
4776 		kmem_free(converted_onm, MAXPATHLEN + 1);
4777 	if (onm != NULL)
4778 		kmem_free(onm, olen);
4779 	if (nnm != converted_nnm)
4780 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4781 	if (nnm != NULL)
4782 		kmem_free(nnm, nlen);
4783 
4784 	if (in_crit_src) nbl_end_crit(srcvp);
4785 	if (in_crit_targ) nbl_end_crit(targvp);
4786 	if (targvp) VN_RELE(targvp);
4787 	if (srcvp) VN_RELE(srcvp);
4788 	if (sfp) {
4789 		if (sfp_rele_grant_hold) rfs4_clear_dont_grant(sfp);
4790 		rfs4_file_rele(sfp);
4791 	}
4792 	if (fp) {
4793 		if (fp_rele_grant_hold) rfs4_clear_dont_grant(fp);
4794 		rfs4_file_rele(fp);
4795 	}
4796 
4797 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4798 	    RENAME4res *, resp);
4799 }
4800 
4801 /* ARGSUSED */
4802 static void
4803 rfs4_op_renew(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4804     struct compound_state *cs)
4805 {
4806 	RENEW4args *args = &argop->nfs_argop4_u.oprenew;
4807 	RENEW4res *resp = &resop->nfs_resop4_u.oprenew;
4808 	rfs4_client_t *cp;
4809 
4810 	DTRACE_NFSV4_2(op__renew__start, struct compound_state *, cs,
4811 	    RENEW4args *, args);
4812 
4813 	if ((cp = rfs4_findclient_by_id(args->clientid, FALSE)) == NULL) {
4814 		*cs->statusp = resp->status =
4815 		    rfs4_check_clientid(&args->clientid, 0);
4816 		goto out;
4817 	}
4818 
4819 	if (rfs4_lease_expired(cp)) {
4820 		rfs4_client_rele(cp);
4821 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
4822 		goto out;
4823 	}
4824 
4825 	rfs4_update_lease(cp);
4826 
4827 	mutex_enter(cp->rc_cbinfo.cb_lock);
4828 	if (cp->rc_cbinfo.cb_notified_of_cb_path_down == FALSE) {
4829 		cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
4830 		*cs->statusp = resp->status = NFS4ERR_CB_PATH_DOWN;
4831 	} else {
4832 		*cs->statusp = resp->status = NFS4_OK;
4833 	}
4834 	mutex_exit(cp->rc_cbinfo.cb_lock);
4835 
4836 	rfs4_client_rele(cp);
4837 
4838 out:
4839 	DTRACE_NFSV4_2(op__renew__done, struct compound_state *, cs,
4840 	    RENEW4res *, resp);
4841 }
4842 
4843 /* ARGSUSED */
4844 static void
4845 rfs4_op_restorefh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
4846     struct compound_state *cs)
4847 {
4848 	RESTOREFH4res *resp = &resop->nfs_resop4_u.oprestorefh;
4849 
4850 	DTRACE_NFSV4_1(op__restorefh__start, struct compound_state *, cs);
4851 
4852 	/* No need to check cs->access - we are not accessing any object */
4853 	if ((cs->saved_vp == NULL) || (cs->saved_fh.nfs_fh4_val == NULL)) {
4854 		*cs->statusp = resp->status = NFS4ERR_RESTOREFH;
4855 		goto out;
4856 	}
4857 	if (cs->vp != NULL) {
4858 		VN_RELE(cs->vp);
4859 	}
4860 	cs->vp = cs->saved_vp;
4861 	cs->saved_vp = NULL;
4862 	if (cs->exi)
4863 		exi_rele(cs->exi);
4864 	cs->exi = cs->saved_exi;
4865 	if (cs->exi)
4866 		exi_hold(cs->exi);
4867 	nfs_fh4_copy(&cs->saved_fh, &cs->fh);
4868 	*cs->statusp = resp->status = NFS4_OK;
4869 	cs->deleg = FALSE;
4870 
4871 out:
4872 	DTRACE_NFSV4_2(op__restorefh__done, struct compound_state *, cs,
4873 	    RESTOREFH4res *, resp);
4874 }
4875 
4876 /* ARGSUSED */
4877 static void
4878 rfs4_op_savefh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4879     struct compound_state *cs)
4880 {
4881 	SAVEFH4res *resp = &resop->nfs_resop4_u.opsavefh;
4882 
4883 	DTRACE_NFSV4_1(op__savefh__start, struct compound_state *, cs);
4884 
4885 	/* No need to check cs->access - we are not accessing any object */
4886 	if (cs->vp == NULL) {
4887 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4888 		goto out;
4889 	}
4890 	if (cs->saved_vp != NULL) {
4891 		VN_RELE(cs->saved_vp);
4892 	}
4893 	cs->saved_vp = cs->vp;
4894 	VN_HOLD(cs->saved_vp);
4895 	if (cs->saved_exi)
4896 		exi_rele(cs->saved_exi);
4897 	cs->saved_exi = cs->exi;
4898 	if (cs->saved_exi)
4899 		exi_hold(cs->saved_exi);
4900 	/*
4901 	 * since SAVEFH is fairly rare, don't alloc space for its fh
4902 	 * unless necessary.
4903 	 */
4904 	if (cs->saved_fh.nfs_fh4_val == NULL) {
4905 		cs->saved_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP);
4906 	}
4907 	nfs_fh4_copy(&cs->fh, &cs->saved_fh);
4908 	*cs->statusp = resp->status = NFS4_OK;
4909 
4910 out:
4911 	DTRACE_NFSV4_2(op__savefh__done, struct compound_state *, cs,
4912 	    SAVEFH4res *, resp);
4913 }
4914 
4915 /*
4916  * rfs4_verify_attr is called when nfsv4 Setattr failed, but we wish to
4917  * return the bitmap of attrs that were set successfully. It is also
4918  * called by Verify/Nverify to test the vattr/vfsstat attrs. It should
4919  * always be called only after rfs4_do_set_attrs().
4920  *
4921  * Verify that the attributes are same as the expected ones. sargp->vap
4922  * and sargp->sbp contain the input attributes as translated from fattr4.
4923  *
4924  * This function verifies only the attrs that correspond to a vattr or
4925  * vfsstat struct. That is because of the extra step needed to get the
4926  * corresponding system structs. Other attributes have already been set or
4927  * verified by do_rfs4_set_attrs.
4928  *
4929  * Return 0 if all attrs match, -1 if some don't, error if error processing.
4930  */
4931 static int
4932 rfs4_verify_attr(struct nfs4_svgetit_arg *sargp,
4933     bitmap4 *resp, struct nfs4_ntov_table *ntovp)
4934 {
4935 	int error, ret_error = 0;
4936 	int i, k;
4937 	uint_t sva_mask = sargp->vap->va_mask;
4938 	uint_t vbit;
4939 	union nfs4_attr_u *na;
4940 	uint8_t *amap;
4941 	bool_t getsb = ntovp->vfsstat;
4942 
4943 	if (sva_mask != 0) {
4944 		/*
4945 		 * Okay to overwrite sargp->vap because we verify based
4946 		 * on the incoming values.
4947 		 */
4948 		ret_error = VOP_GETATTR(sargp->cs->vp, sargp->vap, 0,
4949 		    sargp->cs->cr, NULL);
4950 		if (ret_error) {
4951 			if (resp == NULL)
4952 				return (ret_error);
4953 			/*
4954 			 * Must return bitmap of successful attrs
4955 			 */
4956 			sva_mask = 0;	/* to prevent checking vap later */
4957 		} else {
4958 			/*
4959 			 * Some file systems clobber va_mask. it is probably
4960 			 * wrong of them to do so, nonethless we practice
4961 			 * defensive coding.
4962 			 * See bug id 4276830.
4963 			 */
4964 			sargp->vap->va_mask = sva_mask;
4965 		}
4966 	}
4967 
4968 	if (getsb) {
4969 		/*
4970 		 * Now get the superblock and loop on the bitmap, as there is
4971 		 * no simple way of translating from superblock to bitmap4.
4972 		 */
4973 		ret_error = VFS_STATVFS(sargp->cs->vp->v_vfsp, sargp->sbp);
4974 		if (ret_error) {
4975 			if (resp == NULL)
4976 				goto errout;
4977 			getsb = FALSE;
4978 		}
4979 	}
4980 
4981 	/*
4982 	 * Now loop and verify each attribute which getattr returned
4983 	 * whether it's the same as the input.
4984 	 */
4985 	if (resp == NULL && !getsb && (sva_mask == 0))
4986 		goto errout;
4987 
4988 	na = ntovp->na;
4989 	amap = ntovp->amap;
4990 	k = 0;
4991 	for (i = 0; i < ntovp->attrcnt; i++, na++, amap++) {
4992 		k = *amap;
4993 		ASSERT(nfs4_ntov_map[k].nval == k);
4994 		vbit = nfs4_ntov_map[k].vbit;
4995 
4996 		/*
4997 		 * If vattr attribute but VOP_GETATTR failed, or it's
4998 		 * superblock attribute but VFS_STATVFS failed, skip
4999 		 */
5000 		if (vbit) {
5001 			if ((vbit & sva_mask) == 0)
5002 				continue;
5003 		} else if (!(getsb && nfs4_ntov_map[k].vfsstat)) {
5004 			continue;
5005 		}
5006 		error = (*nfs4_ntov_map[k].sv_getit)(NFS4ATTR_VERIT, sargp, na);
5007 		if (resp != NULL) {
5008 			if (error)
5009 				ret_error = -1;	/* not all match */
5010 			else	/* update response bitmap */
5011 				*resp |= nfs4_ntov_map[k].fbit;
5012 			continue;
5013 		}
5014 		if (error) {
5015 			ret_error = -1;	/* not all match */
5016 			break;
5017 		}
5018 	}
5019 errout:
5020 	return (ret_error);
5021 }
5022 
5023 /*
5024  * Decode the attribute to be set/verified. If the attr requires a sys op
5025  * (VOP_GETATTR, VFS_VFSSTAT), and the request is to verify, then don't
5026  * call the sv_getit function for it, because the sys op hasn't yet been done.
5027  * Return 0 for success, error code if failed.
5028  *
5029  * Note: the decoded arg is not freed here but in nfs4_ntov_table_free.
5030  */
5031 static int
5032 decode_fattr4_attr(nfs4_attr_cmd_t cmd, struct nfs4_svgetit_arg *sargp,
5033     int k, XDR *xdrp, bitmap4 *resp_bval, union nfs4_attr_u *nap)
5034 {
5035 	int error = 0;
5036 	bool_t set_later;
5037 
5038 	sargp->vap->va_mask |= nfs4_ntov_map[k].vbit;
5039 
5040 	if ((*nfs4_ntov_map[k].xfunc)(xdrp, nap)) {
5041 		set_later = nfs4_ntov_map[k].vbit || nfs4_ntov_map[k].vfsstat;
5042 		/*
5043 		 * don't verify yet if a vattr or sb dependent attr,
5044 		 * because we don't have their sys values yet.
5045 		 * Will be done later.
5046 		 */
5047 		if (! (set_later && (cmd == NFS4ATTR_VERIT))) {
5048 			/*
5049 			 * ACLs are a special case, since setting the MODE
5050 			 * conflicts with setting the ACL.  We delay setting
5051 			 * the ACL until all other attributes have been set.
5052 			 * The ACL gets set in do_rfs4_op_setattr().
5053 			 */
5054 			if (nfs4_ntov_map[k].fbit != FATTR4_ACL_MASK) {
5055 				error = (*nfs4_ntov_map[k].sv_getit)(cmd,
5056 				    sargp, nap);
5057 				if (error) {
5058 					xdr_free(nfs4_ntov_map[k].xfunc,
5059 					    (caddr_t)nap);
5060 				}
5061 			}
5062 		}
5063 	} else {
5064 #ifdef  DEBUG
5065 		cmn_err(CE_NOTE, "decode_fattr4_attr: error "
5066 		    "decoding attribute %d\n", k);
5067 #endif
5068 		error = EINVAL;
5069 	}
5070 	if (!error && resp_bval && !set_later) {
5071 		*resp_bval |= nfs4_ntov_map[k].fbit;
5072 	}
5073 
5074 	return (error);
5075 }
5076 
5077 /*
5078  * Set vattr based on incoming fattr4 attrs - used by setattr.
5079  * Set response mask. Ignore any values that are not writable vattr attrs.
5080  */
5081 static nfsstat4
5082 do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5083     struct nfs4_svgetit_arg *sargp, struct nfs4_ntov_table *ntovp,
5084     nfs4_attr_cmd_t cmd)
5085 {
5086 	int error = 0;
5087 	int i;
5088 	char *attrs = fattrp->attrlist4;
5089 	uint32_t attrslen = fattrp->attrlist4_len;
5090 	XDR xdr;
5091 	nfsstat4 status = NFS4_OK;
5092 	vnode_t *vp = cs->vp;
5093 	union nfs4_attr_u *na;
5094 	uint8_t *amap;
5095 
5096 #ifndef lint
5097 	/*
5098 	 * Make sure that maximum attribute number can be expressed as an
5099 	 * 8 bit quantity.
5100 	 */
5101 	ASSERT(NFS4_MAXNUM_ATTRS <= (UINT8_MAX + 1));
5102 #endif
5103 
5104 	if (vp == NULL) {
5105 		if (resp)
5106 			*resp = 0;
5107 		return (NFS4ERR_NOFILEHANDLE);
5108 	}
5109 	if (cs->access == CS_ACCESS_DENIED) {
5110 		if (resp)
5111 			*resp = 0;
5112 		return (NFS4ERR_ACCESS);
5113 	}
5114 
5115 	sargp->op = cmd;
5116 	sargp->cs = cs;
5117 	sargp->flag = 0;	/* may be set later */
5118 	sargp->vap->va_mask = 0;
5119 	sargp->rdattr_error = NFS4_OK;
5120 	sargp->rdattr_error_req = FALSE;
5121 	/* sargp->sbp is set by the caller */
5122 
5123 	xdrmem_create(&xdr, attrs, attrslen, XDR_DECODE);
5124 
5125 	na = ntovp->na;
5126 	amap = ntovp->amap;
5127 
5128 	/*
5129 	 * The following loop iterates on the nfs4_ntov_map checking
5130 	 * if the fbit is set in the requested bitmap.
5131 	 * If set then we process the arguments using the
5132 	 * rfs4_fattr4 conversion functions to populate the setattr
5133 	 * vattr and va_mask. Any settable attrs that are not using vattr
5134 	 * will be set in this loop.
5135 	 */
5136 	for (i = 0; i < nfs4_ntov_map_size; i++) {
5137 		if (!(fattrp->attrmask & nfs4_ntov_map[i].fbit)) {
5138 			continue;
5139 		}
5140 		/*
5141 		 * If setattr, must be a writable attr.
5142 		 * If verify/nverify, must be a readable attr.
5143 		 */
5144 		if ((error = (*nfs4_ntov_map[i].sv_getit)(
5145 		    NFS4ATTR_SUPPORTED, sargp, NULL)) != 0) {
5146 			/*
5147 			 * Client tries to set/verify an
5148 			 * unsupported attribute, tries to set
5149 			 * a read only attr or verify a write
5150 			 * only one - error!
5151 			 */
5152 			break;
5153 		}
5154 		/*
5155 		 * Decode the attribute to set/verify
5156 		 */
5157 		error = decode_fattr4_attr(cmd, sargp, nfs4_ntov_map[i].nval,
5158 		    &xdr, resp ? resp : NULL, na);
5159 		if (error)
5160 			break;
5161 		*amap++ = (uint8_t)nfs4_ntov_map[i].nval;
5162 		na++;
5163 		(ntovp->attrcnt)++;
5164 		if (nfs4_ntov_map[i].vfsstat)
5165 			ntovp->vfsstat = TRUE;
5166 	}
5167 
5168 	if (error != 0)
5169 		status = (error == ENOTSUP ? NFS4ERR_ATTRNOTSUPP :
5170 		    puterrno4(error));
5171 	/* xdrmem_destroy(&xdrs); */	/* NO-OP */
5172 	return (status);
5173 }
5174 
5175 static nfsstat4
5176 do_rfs4_op_setattr(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5177     stateid4 *stateid)
5178 {
5179 	int error = 0;
5180 	struct nfs4_svgetit_arg sarg;
5181 	bool_t trunc;
5182 
5183 	nfsstat4 status = NFS4_OK;
5184 	cred_t *cr = cs->cr;
5185 	vnode_t *vp = cs->vp;
5186 	struct nfs4_ntov_table ntov;
5187 	struct statvfs64 sb;
5188 	struct vattr bva;
5189 	struct flock64 bf;
5190 	int in_crit = 0;
5191 	uint_t saved_mask = 0;
5192 	caller_context_t ct;
5193 
5194 	*resp = 0;
5195 	sarg.sbp = &sb;
5196 	sarg.is_referral = B_FALSE;
5197 	nfs4_ntov_table_init(&ntov);
5198 	status = do_rfs4_set_attrs(resp, fattrp, cs, &sarg, &ntov,
5199 	    NFS4ATTR_SETIT);
5200 	if (status != NFS4_OK) {
5201 		/*
5202 		 * failed set attrs
5203 		 */
5204 		goto done;
5205 	}
5206 	if ((sarg.vap->va_mask == 0) &&
5207 	    (! (fattrp->attrmask & FATTR4_ACL_MASK))) {
5208 		/*
5209 		 * no further work to be done
5210 		 */
5211 		goto done;
5212 	}
5213 
5214 	/*
5215 	 * If we got a request to set the ACL and the MODE, only
5216 	 * allow changing VSUID, VSGID, and VSVTX.  Attempting
5217 	 * to change any other bits, along with setting an ACL,
5218 	 * gives NFS4ERR_INVAL.
5219 	 */
5220 	if ((fattrp->attrmask & FATTR4_ACL_MASK) &&
5221 	    (fattrp->attrmask & FATTR4_MODE_MASK)) {
5222 		vattr_t va;
5223 
5224 		va.va_mask = AT_MODE;
5225 		error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
5226 		if (error) {
5227 			status = puterrno4(error);
5228 			goto done;
5229 		}
5230 		if ((sarg.vap->va_mode ^ va.va_mode) &
5231 		    ~(VSUID | VSGID | VSVTX)) {
5232 			status = NFS4ERR_INVAL;
5233 			goto done;
5234 		}
5235 	}
5236 
5237 	/* Check stateid only if size has been set */
5238 	if (sarg.vap->va_mask & AT_SIZE) {
5239 		trunc = (sarg.vap->va_size == 0);
5240 		status = rfs4_check_stateid(FWRITE, cs->vp, stateid,
5241 		    trunc, &cs->deleg, sarg.vap->va_mask & AT_SIZE, &ct);
5242 		if (status != NFS4_OK)
5243 			goto done;
5244 	} else {
5245 		ct.cc_sysid = 0;
5246 		ct.cc_pid = 0;
5247 		ct.cc_caller_id = nfs4_srv_caller_id;
5248 		ct.cc_flags = CC_DONTBLOCK;
5249 	}
5250 
5251 	/* XXX start of possible race with delegations */
5252 
5253 	/*
5254 	 * We need to specially handle size changes because it is
5255 	 * possible for the client to create a file with read-only
5256 	 * modes, but with the file opened for writing. If the client
5257 	 * then tries to set the file size, e.g. ftruncate(3C),
5258 	 * fcntl(F_FREESP), the normal access checking done in
5259 	 * VOP_SETATTR would prevent the client from doing it even though
5260 	 * it should be allowed to do so.  To get around this, we do the
5261 	 * access checking for ourselves and use VOP_SPACE which doesn't
5262 	 * do the access checking.
5263 	 * Also the client should not be allowed to change the file
5264 	 * size if there is a conflicting non-blocking mandatory lock in
5265 	 * the region of the change.
5266 	 */
5267 	if (vp->v_type == VREG && (sarg.vap->va_mask & AT_SIZE)) {
5268 		u_offset_t offset;
5269 		ssize_t length;
5270 
5271 		/*
5272 		 * ufs_setattr clears AT_SIZE from vap->va_mask, but
5273 		 * before returning, sarg.vap->va_mask is used to
5274 		 * generate the setattr reply bitmap.  We also clear
5275 		 * AT_SIZE below before calling VOP_SPACE.  For both
5276 		 * of these cases, the va_mask needs to be saved here
5277 		 * and restored after calling VOP_SETATTR.
5278 		 */
5279 		saved_mask = sarg.vap->va_mask;
5280 
5281 		/*
5282 		 * Check any possible conflict due to NBMAND locks.
5283 		 * Get into critical region before VOP_GETATTR, so the
5284 		 * size attribute is valid when checking conflicts.
5285 		 */
5286 		if (nbl_need_check(vp)) {
5287 			nbl_start_crit(vp, RW_READER);
5288 			in_crit = 1;
5289 		}
5290 
5291 		bva.va_mask = AT_UID|AT_SIZE;
5292 		if (error = VOP_GETATTR(vp, &bva, 0, cr, &ct)) {
5293 			status = puterrno4(error);
5294 			goto done;
5295 		}
5296 
5297 		if (in_crit) {
5298 			if (sarg.vap->va_size < bva.va_size) {
5299 				offset = sarg.vap->va_size;
5300 				length = bva.va_size - sarg.vap->va_size;
5301 			} else {
5302 				offset = bva.va_size;
5303 				length = sarg.vap->va_size - bva.va_size;
5304 			}
5305 			if (nbl_conflict(vp, NBL_WRITE, offset, length, 0,
5306 			    &ct)) {
5307 				status = NFS4ERR_LOCKED;
5308 				goto done;
5309 			}
5310 		}
5311 
5312 		if (crgetuid(cr) == bva.va_uid) {
5313 			sarg.vap->va_mask &= ~AT_SIZE;
5314 			bf.l_type = F_WRLCK;
5315 			bf.l_whence = 0;
5316 			bf.l_start = (off64_t)sarg.vap->va_size;
5317 			bf.l_len = 0;
5318 			bf.l_sysid = 0;
5319 			bf.l_pid = 0;
5320 			error = VOP_SPACE(vp, F_FREESP, &bf, FWRITE,
5321 			    (offset_t)sarg.vap->va_size, cr, &ct);
5322 		}
5323 	}
5324 
5325 	if (!error && sarg.vap->va_mask != 0)
5326 		error = VOP_SETATTR(vp, sarg.vap, sarg.flag, cr, &ct);
5327 
5328 	/* restore va_mask -- ufs_setattr clears AT_SIZE */
5329 	if (saved_mask & AT_SIZE)
5330 		sarg.vap->va_mask |= AT_SIZE;
5331 
5332 	/*
5333 	 * If an ACL was being set, it has been delayed until now,
5334 	 * in order to set the mode (via the VOP_SETATTR() above) first.
5335 	 */
5336 	if ((! error) && (fattrp->attrmask & FATTR4_ACL_MASK)) {
5337 		int i;
5338 
5339 		for (i = 0; i < NFS4_MAXNUM_ATTRS; i++)
5340 			if (ntov.amap[i] == FATTR4_ACL)
5341 				break;
5342 		if (i < NFS4_MAXNUM_ATTRS) {
5343 			error = (*nfs4_ntov_map[FATTR4_ACL].sv_getit)(
5344 			    NFS4ATTR_SETIT, &sarg, &ntov.na[i]);
5345 			if (error == 0) {
5346 				*resp |= FATTR4_ACL_MASK;
5347 			} else if (error == ENOTSUP) {
5348 				(void) rfs4_verify_attr(&sarg, resp, &ntov);
5349 				status = NFS4ERR_ATTRNOTSUPP;
5350 				goto done;
5351 			}
5352 		} else {
5353 			NFS4_DEBUG(rfs4_debug,
5354 			    (CE_NOTE, "do_rfs4_op_setattr: "
5355 			    "unable to find ACL in fattr4"));
5356 			error = EINVAL;
5357 		}
5358 	}
5359 
5360 	if (error) {
5361 		/* check if a monitor detected a delegation conflict */
5362 		if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
5363 			status = NFS4ERR_DELAY;
5364 		else
5365 			status = puterrno4(error);
5366 
5367 		/*
5368 		 * Set the response bitmap when setattr failed.
5369 		 * If VOP_SETATTR partially succeeded, test by doing a
5370 		 * VOP_GETATTR on the object and comparing the data
5371 		 * to the setattr arguments.
5372 		 */
5373 		(void) rfs4_verify_attr(&sarg, resp, &ntov);
5374 	} else {
5375 		/*
5376 		 * Force modified metadata out to stable storage.
5377 		 */
5378 		(void) VOP_FSYNC(vp, FNODSYNC, cr, &ct);
5379 		/*
5380 		 * Set response bitmap
5381 		 */
5382 		nfs4_vmask_to_nmask_set(sarg.vap->va_mask, resp);
5383 	}
5384 
5385 /* Return early and already have a NFSv4 error */
5386 done:
5387 	/*
5388 	 * Except for nfs4_vmask_to_nmask_set(), vattr --> fattr
5389 	 * conversion sets both readable and writeable NFS4 attrs
5390 	 * for AT_MTIME and AT_ATIME.  The line below masks out
5391 	 * unrequested attrs from the setattr result bitmap.  This
5392 	 * is placed after the done: label to catch the ATTRNOTSUP
5393 	 * case.
5394 	 */
5395 	*resp &= fattrp->attrmask;
5396 
5397 	if (in_crit)
5398 		nbl_end_crit(vp);
5399 
5400 	nfs4_ntov_table_free(&ntov, &sarg);
5401 
5402 	return (status);
5403 }
5404 
5405 /* ARGSUSED */
5406 static void
5407 rfs4_op_setattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5408     struct compound_state *cs)
5409 {
5410 	SETATTR4args *args = &argop->nfs_argop4_u.opsetattr;
5411 	SETATTR4res *resp = &resop->nfs_resop4_u.opsetattr;
5412 	bslabel_t *clabel;
5413 
5414 	DTRACE_NFSV4_2(op__setattr__start, struct compound_state *, cs,
5415 	    SETATTR4args *, args);
5416 
5417 	if (cs->vp == NULL) {
5418 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5419 		goto out;
5420 	}
5421 
5422 	/*
5423 	 * If there is an unshared filesystem mounted on this vnode,
5424 	 * do not allow to setattr on this vnode.
5425 	 */
5426 	if (vn_ismntpt(cs->vp)) {
5427 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5428 		goto out;
5429 	}
5430 
5431 	resp->attrsset = 0;
5432 
5433 	if (rdonly4(req, cs)) {
5434 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5435 		goto out;
5436 	}
5437 
5438 	/* check label before setting attributes */
5439 	if (is_system_labeled()) {
5440 		ASSERT(req->rq_label != NULL);
5441 		clabel = req->rq_label;
5442 		DTRACE_PROBE2(tx__rfs4__log__info__opsetattr__clabel, char *,
5443 		    "got client label from request(1)",
5444 		    struct svc_req *, req);
5445 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
5446 			if (!do_rfs_label_check(clabel, cs->vp,
5447 			    EQUALITY_CHECK, cs->exi)) {
5448 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
5449 				goto out;
5450 			}
5451 		}
5452 	}
5453 
5454 	*cs->statusp = resp->status =
5455 	    do_rfs4_op_setattr(&resp->attrsset, &args->obj_attributes, cs,
5456 	    &args->stateid);
5457 
5458 out:
5459 	DTRACE_NFSV4_2(op__setattr__done, struct compound_state *, cs,
5460 	    SETATTR4res *, resp);
5461 }
5462 
5463 /* ARGSUSED */
5464 static void
5465 rfs4_op_verify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5466     struct compound_state *cs)
5467 {
5468 	/*
5469 	 * verify and nverify are exactly the same, except that nverify
5470 	 * succeeds when some argument changed, and verify succeeds when
5471 	 * when none changed.
5472 	 */
5473 
5474 	VERIFY4args  *args = &argop->nfs_argop4_u.opverify;
5475 	VERIFY4res *resp = &resop->nfs_resop4_u.opverify;
5476 
5477 	int error;
5478 	struct nfs4_svgetit_arg sarg;
5479 	struct statvfs64 sb;
5480 	struct nfs4_ntov_table ntov;
5481 
5482 	DTRACE_NFSV4_2(op__verify__start, struct compound_state *, cs,
5483 	    VERIFY4args *, args);
5484 
5485 	if (cs->vp == NULL) {
5486 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5487 		goto out;
5488 	}
5489 
5490 	sarg.sbp = &sb;
5491 	sarg.is_referral = B_FALSE;
5492 	nfs4_ntov_table_init(&ntov);
5493 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5494 	    &sarg, &ntov, NFS4ATTR_VERIT);
5495 	if (resp->status != NFS4_OK) {
5496 		/*
5497 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5498 		 * so could return -1 for "no match".
5499 		 */
5500 		if (resp->status == -1)
5501 			resp->status = NFS4ERR_NOT_SAME;
5502 		goto done;
5503 	}
5504 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5505 	switch (error) {
5506 	case 0:
5507 		resp->status = NFS4_OK;
5508 		break;
5509 	case -1:
5510 		resp->status = NFS4ERR_NOT_SAME;
5511 		break;
5512 	default:
5513 		resp->status = puterrno4(error);
5514 		break;
5515 	}
5516 done:
5517 	*cs->statusp = resp->status;
5518 	nfs4_ntov_table_free(&ntov, &sarg);
5519 out:
5520 	DTRACE_NFSV4_2(op__verify__done, struct compound_state *, cs,
5521 	    VERIFY4res *, resp);
5522 }
5523 
5524 /* ARGSUSED */
5525 static void
5526 rfs4_op_nverify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5527     struct compound_state *cs)
5528 {
5529 	/*
5530 	 * verify and nverify are exactly the same, except that nverify
5531 	 * succeeds when some argument changed, and verify succeeds when
5532 	 * when none changed.
5533 	 */
5534 
5535 	NVERIFY4args  *args = &argop->nfs_argop4_u.opnverify;
5536 	NVERIFY4res *resp = &resop->nfs_resop4_u.opnverify;
5537 
5538 	int error;
5539 	struct nfs4_svgetit_arg sarg;
5540 	struct statvfs64 sb;
5541 	struct nfs4_ntov_table ntov;
5542 
5543 	DTRACE_NFSV4_2(op__nverify__start, struct compound_state *, cs,
5544 	    NVERIFY4args *, args);
5545 
5546 	if (cs->vp == NULL) {
5547 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5548 		DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5549 		    NVERIFY4res *, resp);
5550 		return;
5551 	}
5552 	sarg.sbp = &sb;
5553 	sarg.is_referral = B_FALSE;
5554 	nfs4_ntov_table_init(&ntov);
5555 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5556 	    &sarg, &ntov, NFS4ATTR_VERIT);
5557 	if (resp->status != NFS4_OK) {
5558 		/*
5559 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5560 		 * so could return -1 for "no match".
5561 		 */
5562 		if (resp->status == -1)
5563 			resp->status = NFS4_OK;
5564 		goto done;
5565 	}
5566 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5567 	switch (error) {
5568 	case 0:
5569 		resp->status = NFS4ERR_SAME;
5570 		break;
5571 	case -1:
5572 		resp->status = NFS4_OK;
5573 		break;
5574 	default:
5575 		resp->status = puterrno4(error);
5576 		break;
5577 	}
5578 done:
5579 	*cs->statusp = resp->status;
5580 	nfs4_ntov_table_free(&ntov, &sarg);
5581 
5582 	DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5583 	    NVERIFY4res *, resp);
5584 }
5585 
5586 /*
5587  * XXX - This should live in an NFS header file.
5588  */
5589 #define	MAX_IOVECS	12
5590 
5591 /* ARGSUSED */
5592 static void
5593 rfs4_op_write(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5594     struct compound_state *cs)
5595 {
5596 	WRITE4args *args = &argop->nfs_argop4_u.opwrite;
5597 	WRITE4res *resp = &resop->nfs_resop4_u.opwrite;
5598 	int error;
5599 	vnode_t *vp;
5600 	struct vattr bva;
5601 	u_offset_t rlimit;
5602 	struct uio uio;
5603 	struct iovec iov[MAX_IOVECS];
5604 	struct iovec *iovp;
5605 	int iovcnt;
5606 	int ioflag;
5607 	cred_t *savecred, *cr;
5608 	bool_t *deleg = &cs->deleg;
5609 	nfsstat4 stat;
5610 	int in_crit = 0;
5611 	caller_context_t ct;
5612 
5613 	DTRACE_NFSV4_2(op__write__start, struct compound_state *, cs,
5614 	    WRITE4args *, args);
5615 
5616 	vp = cs->vp;
5617 	if (vp == NULL) {
5618 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5619 		goto out;
5620 	}
5621 	if (cs->access == CS_ACCESS_DENIED) {
5622 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5623 		goto out;
5624 	}
5625 
5626 	cr = cs->cr;
5627 
5628 	if ((stat = rfs4_check_stateid(FWRITE, vp, &args->stateid, FALSE,
5629 	    deleg, TRUE, &ct)) != NFS4_OK) {
5630 		*cs->statusp = resp->status = stat;
5631 		goto out;
5632 	}
5633 
5634 	/*
5635 	 * We have to enter the critical region before calling VOP_RWLOCK
5636 	 * to avoid a deadlock with ufs.
5637 	 */
5638 	if (nbl_need_check(vp)) {
5639 		nbl_start_crit(vp, RW_READER);
5640 		in_crit = 1;
5641 		if (nbl_conflict(vp, NBL_WRITE,
5642 		    args->offset, args->data_len, 0, &ct)) {
5643 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
5644 			goto out;
5645 		}
5646 	}
5647 
5648 	bva.va_mask = AT_MODE | AT_UID;
5649 	error = VOP_GETATTR(vp, &bva, 0, cr, &ct);
5650 
5651 	/*
5652 	 * If we can't get the attributes, then we can't do the
5653 	 * right access checking.  So, we'll fail the request.
5654 	 */
5655 	if (error) {
5656 		*cs->statusp = resp->status = puterrno4(error);
5657 		goto out;
5658 	}
5659 
5660 	if (rdonly4(req, cs)) {
5661 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5662 		goto out;
5663 	}
5664 
5665 	if (vp->v_type != VREG) {
5666 		*cs->statusp = resp->status =
5667 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
5668 		goto out;
5669 	}
5670 
5671 	if (crgetuid(cr) != bva.va_uid &&
5672 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, &ct))) {
5673 		*cs->statusp = resp->status = puterrno4(error);
5674 		goto out;
5675 	}
5676 
5677 	if (MANDLOCK(vp, bva.va_mode)) {
5678 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5679 		goto out;
5680 	}
5681 
5682 	if (args->data_len == 0) {
5683 		*cs->statusp = resp->status = NFS4_OK;
5684 		resp->count = 0;
5685 		resp->committed = args->stable;
5686 		resp->writeverf = Write4verf;
5687 		goto out;
5688 	}
5689 
5690 	if (args->mblk != NULL) {
5691 		mblk_t *m;
5692 		uint_t bytes, round_len;
5693 
5694 		iovcnt = 0;
5695 		bytes = 0;
5696 		round_len = roundup(args->data_len, BYTES_PER_XDR_UNIT);
5697 		for (m = args->mblk;
5698 		    m != NULL && bytes < round_len;
5699 		    m = m->b_cont) {
5700 			iovcnt++;
5701 			bytes += MBLKL(m);
5702 		}
5703 #ifdef DEBUG
5704 		/* should have ended on an mblk boundary */
5705 		if (bytes != round_len) {
5706 			printf("bytes=0x%x, round_len=0x%x, req len=0x%x\n",
5707 			    bytes, round_len, args->data_len);
5708 			printf("args=%p, args->mblk=%p, m=%p", (void *)args,
5709 			    (void *)args->mblk, (void *)m);
5710 			ASSERT(bytes == round_len);
5711 		}
5712 #endif
5713 		if (iovcnt <= MAX_IOVECS) {
5714 			iovp = iov;
5715 		} else {
5716 			iovp = kmem_alloc(sizeof (*iovp) * iovcnt, KM_SLEEP);
5717 		}
5718 		mblk_to_iov(args->mblk, iovcnt, iovp);
5719 	} else if (args->rlist != NULL) {
5720 		iovcnt = 1;
5721 		iovp = iov;
5722 		iovp->iov_base = (char *)((args->rlist)->u.c_daddr3);
5723 		iovp->iov_len = args->data_len;
5724 	} else {
5725 		iovcnt = 1;
5726 		iovp = iov;
5727 		iovp->iov_base = args->data_val;
5728 		iovp->iov_len = args->data_len;
5729 	}
5730 
5731 	uio.uio_iov = iovp;
5732 	uio.uio_iovcnt = iovcnt;
5733 
5734 	uio.uio_segflg = UIO_SYSSPACE;
5735 	uio.uio_extflg = UIO_COPY_DEFAULT;
5736 	uio.uio_loffset = args->offset;
5737 	uio.uio_resid = args->data_len;
5738 	uio.uio_llimit = curproc->p_fsz_ctl;
5739 	rlimit = uio.uio_llimit - args->offset;
5740 	if (rlimit < (u_offset_t)uio.uio_resid)
5741 		uio.uio_resid = (int)rlimit;
5742 
5743 	if (args->stable == UNSTABLE4)
5744 		ioflag = 0;
5745 	else if (args->stable == FILE_SYNC4)
5746 		ioflag = FSYNC;
5747 	else if (args->stable == DATA_SYNC4)
5748 		ioflag = FDSYNC;
5749 	else {
5750 		if (iovp != iov)
5751 			kmem_free(iovp, sizeof (*iovp) * iovcnt);
5752 		*cs->statusp = resp->status = NFS4ERR_INVAL;
5753 		goto out;
5754 	}
5755 
5756 	/*
5757 	 * We're changing creds because VM may fault and we need
5758 	 * the cred of the current thread to be used if quota
5759 	 * checking is enabled.
5760 	 */
5761 	savecred = curthread->t_cred;
5762 	curthread->t_cred = cr;
5763 	error = do_io(FWRITE, vp, &uio, ioflag, cr, &ct);
5764 	curthread->t_cred = savecred;
5765 
5766 	if (iovp != iov)
5767 		kmem_free(iovp, sizeof (*iovp) * iovcnt);
5768 
5769 	if (error) {
5770 		*cs->statusp = resp->status = puterrno4(error);
5771 		goto out;
5772 	}
5773 
5774 	*cs->statusp = resp->status = NFS4_OK;
5775 	resp->count = args->data_len - uio.uio_resid;
5776 
5777 	if (ioflag == 0)
5778 		resp->committed = UNSTABLE4;
5779 	else
5780 		resp->committed = FILE_SYNC4;
5781 
5782 	resp->writeverf = Write4verf;
5783 
5784 out:
5785 	if (in_crit)
5786 		nbl_end_crit(vp);
5787 
5788 	DTRACE_NFSV4_2(op__write__done, struct compound_state *, cs,
5789 	    WRITE4res *, resp);
5790 }
5791 
5792 
5793 /* XXX put in a header file */
5794 extern int	sec_svc_getcred(struct svc_req *, cred_t *,  caddr_t *, int *);
5795 
5796 void
5797 rfs4_compound(COMPOUND4args *args, COMPOUND4res *resp, struct exportinfo *exi,
5798     struct svc_req *req, cred_t *cr, int *rv)
5799 {
5800 	uint_t i;
5801 	struct compound_state cs;
5802 
5803 	if (rv != NULL)
5804 		*rv = 0;
5805 	rfs4_init_compound_state(&cs);
5806 	/*
5807 	 * Form a reply tag by copying over the reqeuest tag.
5808 	 */
5809 	resp->tag.utf8string_val =
5810 	    kmem_alloc(args->tag.utf8string_len, KM_SLEEP);
5811 	resp->tag.utf8string_len = args->tag.utf8string_len;
5812 	bcopy(args->tag.utf8string_val, resp->tag.utf8string_val,
5813 	    resp->tag.utf8string_len);
5814 
5815 	cs.statusp = &resp->status;
5816 	cs.req = req;
5817 	resp->array = NULL;
5818 	resp->array_len = 0;
5819 
5820 	/*
5821 	 * XXX for now, minorversion should be zero
5822 	 */
5823 	if (args->minorversion != NFS4_MINORVERSION) {
5824 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5825 		    &cs, COMPOUND4args *, args);
5826 		resp->status = NFS4ERR_MINOR_VERS_MISMATCH;
5827 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5828 		    &cs, COMPOUND4res *, resp);
5829 		return;
5830 	}
5831 
5832 	if (args->array_len == 0) {
5833 		resp->status = NFS4_OK;
5834 		return;
5835 	}
5836 
5837 	ASSERT(exi == NULL);
5838 	ASSERT(cr == NULL);
5839 
5840 	cr = crget();
5841 	ASSERT(cr != NULL);
5842 
5843 	if (sec_svc_getcred(req, cr, &cs.principal, &cs.nfsflavor) == 0) {
5844 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5845 		    &cs, COMPOUND4args *, args);
5846 		crfree(cr);
5847 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5848 		    &cs, COMPOUND4res *, resp);
5849 		svcerr_badcred(req->rq_xprt);
5850 		if (rv != NULL)
5851 			*rv = 1;
5852 		return;
5853 	}
5854 	resp->array_len = args->array_len;
5855 	resp->array = kmem_zalloc(args->array_len * sizeof (nfs_resop4),
5856 	    KM_SLEEP);
5857 
5858 	cs.basecr = cr;
5859 
5860 	DTRACE_NFSV4_2(compound__start, struct compound_state *, &cs,
5861 	    COMPOUND4args *, args);
5862 
5863 	/*
5864 	 * If this is the first compound we've seen, we need to start all
5865 	 * new instances' grace periods.
5866 	 */
5867 	if (rfs4_seen_first_compound == 0) {
5868 		rfs4_grace_start_new();
5869 		/*
5870 		 * This must be set after rfs4_grace_start_new(), otherwise
5871 		 * another thread could proceed past here before the former
5872 		 * is finished.
5873 		 */
5874 		rfs4_seen_first_compound = 1;
5875 	}
5876 
5877 	for (i = 0; i < args->array_len && cs.cont; i++) {
5878 		nfs_argop4 *argop;
5879 		nfs_resop4 *resop;
5880 		uint_t op;
5881 
5882 		argop = &args->array[i];
5883 		resop = &resp->array[i];
5884 		resop->resop = argop->argop;
5885 		op = (uint_t)resop->resop;
5886 
5887 		if (op < rfsv4disp_cnt) {
5888 			/*
5889 			 * Count the individual ops here; NULL and COMPOUND
5890 			 * are counted in common_dispatch()
5891 			 */
5892 			rfsproccnt_v4_ptr[op].value.ui64++;
5893 
5894 			NFS4_DEBUG(rfs4_debug > 1,
5895 			    (CE_NOTE, "Executing %s", rfs4_op_string[op]));
5896 			(*rfsv4disptab[op].dis_proc)(argop, resop, req, &cs);
5897 			NFS4_DEBUG(rfs4_debug > 1, (CE_NOTE, "%s returned %d",
5898 			    rfs4_op_string[op], *cs.statusp));
5899 			if (*cs.statusp != NFS4_OK)
5900 				cs.cont = FALSE;
5901 		} else {
5902 			/*
5903 			 * This is effectively dead code since XDR code
5904 			 * will have already returned BADXDR if op doesn't
5905 			 * decode to legal value.  This only done for a
5906 			 * day when XDR code doesn't verify v4 opcodes.
5907 			 */
5908 			op = OP_ILLEGAL;
5909 			rfsproccnt_v4_ptr[OP_ILLEGAL_IDX].value.ui64++;
5910 
5911 			rfs4_op_illegal(argop, resop, req, &cs);
5912 			cs.cont = FALSE;
5913 		}
5914 
5915 		/*
5916 		 * If not at last op, and if we are to stop, then
5917 		 * compact the results array.
5918 		 */
5919 		if ((i + 1) < args->array_len && !cs.cont) {
5920 			nfs_resop4 *new_res = kmem_alloc(
5921 			    (i+1) * sizeof (nfs_resop4), KM_SLEEP);
5922 			bcopy(resp->array,
5923 			    new_res, (i+1) * sizeof (nfs_resop4));
5924 			kmem_free(resp->array,
5925 			    args->array_len * sizeof (nfs_resop4));
5926 
5927 			resp->array_len =  i + 1;
5928 			resp->array = new_res;
5929 		}
5930 	}
5931 
5932 
5933 	DTRACE_NFSV4_2(compound__done, struct compound_state *, &cs,
5934 	    COMPOUND4res *, resp);
5935 
5936 	if (cs.exi)
5937 		exi_rele(cs.exi);
5938 	if (cs.saved_exi)
5939 		exi_rele(cs.saved_exi);
5940 	if (cs.vp)
5941 		VN_RELE(cs.vp);
5942 	if (cs.saved_vp)
5943 		VN_RELE(cs.saved_vp);
5944 	if (cs.saved_fh.nfs_fh4_val)
5945 		kmem_free(cs.saved_fh.nfs_fh4_val, NFS4_FHSIZE);
5946 
5947 	if (cs.basecr)
5948 		crfree(cs.basecr);
5949 	if (cs.cr)
5950 		crfree(cs.cr);
5951 	/*
5952 	 * done with this compound request, free the label
5953 	 */
5954 
5955 	if (req->rq_label != NULL) {
5956 		kmem_free(req->rq_label, sizeof (bslabel_t));
5957 		req->rq_label = NULL;
5958 	}
5959 }
5960 
5961 /*
5962  * XXX because of what appears to be duplicate calls to rfs4_compound_free
5963  * XXX zero out the tag and array values. Need to investigate why the
5964  * XXX calls occur, but at least prevent the panic for now.
5965  */
5966 void
5967 rfs4_compound_free(COMPOUND4res *resp)
5968 {
5969 	uint_t i;
5970 
5971 	if (resp->tag.utf8string_val) {
5972 		UTF8STRING_FREE(resp->tag)
5973 	}
5974 
5975 	for (i = 0; i < resp->array_len; i++) {
5976 		nfs_resop4 *resop;
5977 		uint_t op;
5978 
5979 		resop = &resp->array[i];
5980 		op = (uint_t)resop->resop;
5981 		if (op < rfsv4disp_cnt) {
5982 			(*rfsv4disptab[op].dis_resfree)(resop);
5983 		}
5984 	}
5985 	if (resp->array != NULL) {
5986 		kmem_free(resp->array, resp->array_len * sizeof (nfs_resop4));
5987 	}
5988 }
5989 
5990 /*
5991  * Process the value of the compound request rpc flags, as a bit-AND
5992  * of the individual per-op flags (idempotent, allowork, publicfh_ok)
5993  */
5994 void
5995 rfs4_compound_flagproc(COMPOUND4args *args, int *flagp)
5996 {
5997 	int i;
5998 	int flag = RPC_ALL;
5999 
6000 	for (i = 0; flag && i < args->array_len; i++) {
6001 		uint_t op;
6002 
6003 		op = (uint_t)args->array[i].argop;
6004 
6005 		if (op < rfsv4disp_cnt)
6006 			flag &= rfsv4disptab[op].dis_flags;
6007 		else
6008 			flag = 0;
6009 	}
6010 	*flagp = flag;
6011 }
6012 
6013 nfsstat4
6014 rfs4_client_sysid(rfs4_client_t *cp, sysid_t *sp)
6015 {
6016 	nfsstat4 e;
6017 
6018 	rfs4_dbe_lock(cp->rc_dbe);
6019 
6020 	if (cp->rc_sysidt != LM_NOSYSID) {
6021 		*sp = cp->rc_sysidt;
6022 		e = NFS4_OK;
6023 
6024 	} else if ((cp->rc_sysidt = lm_alloc_sysidt()) != LM_NOSYSID) {
6025 		*sp = cp->rc_sysidt;
6026 		e = NFS4_OK;
6027 
6028 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
6029 		    "rfs4_client_sysid: allocated 0x%x\n", *sp));
6030 	} else
6031 		e = NFS4ERR_DELAY;
6032 
6033 	rfs4_dbe_unlock(cp->rc_dbe);
6034 	return (e);
6035 }
6036 
6037 #if defined(DEBUG) && ! defined(lint)
6038 static void lock_print(char *str, int operation, struct flock64 *flk)
6039 {
6040 	char *op, *type;
6041 
6042 	switch (operation) {
6043 	case F_GETLK: op = "F_GETLK";
6044 		break;
6045 	case F_SETLK: op = "F_SETLK";
6046 		break;
6047 	case F_SETLK_NBMAND: op = "F_SETLK_NBMAND";
6048 		break;
6049 	default: op = "F_UNKNOWN";
6050 		break;
6051 	}
6052 	switch (flk->l_type) {
6053 	case F_UNLCK: type = "F_UNLCK";
6054 		break;
6055 	case F_RDLCK: type = "F_RDLCK";
6056 		break;
6057 	case F_WRLCK: type = "F_WRLCK";
6058 		break;
6059 	default: type = "F_UNKNOWN";
6060 		break;
6061 	}
6062 
6063 	ASSERT(flk->l_whence == 0);
6064 	cmn_err(CE_NOTE, "%s:  %s, type = %s, off = %llx len = %llx pid = %d",
6065 	    str, op, type, (longlong_t)flk->l_start,
6066 	    flk->l_len ? (longlong_t)flk->l_len : ~0LL, flk->l_pid);
6067 }
6068 
6069 #define	LOCK_PRINT(d, s, t, f) if (d) lock_print(s, t, f)
6070 #else
6071 #define	LOCK_PRINT(d, s, t, f)
6072 #endif
6073 
6074 /*ARGSUSED*/
6075 static bool_t
6076 creds_ok(cred_set_t cr_set, struct svc_req *req, struct compound_state *cs)
6077 {
6078 	return (TRUE);
6079 }
6080 
6081 /*
6082  * Look up the pathname using the vp in cs as the directory vnode.
6083  * cs->vp will be the vnode for the file on success
6084  */
6085 
6086 static nfsstat4
6087 rfs4_lookup(component4 *component, struct svc_req *req,
6088     struct compound_state *cs)
6089 {
6090 	char *nm;
6091 	uint32_t len;
6092 	nfsstat4 status;
6093 	struct sockaddr *ca;
6094 	char *name;
6095 
6096 	if (cs->vp == NULL) {
6097 		return (NFS4ERR_NOFILEHANDLE);
6098 	}
6099 	if (cs->vp->v_type != VDIR) {
6100 		return (NFS4ERR_NOTDIR);
6101 	}
6102 
6103 	status = utf8_dir_verify(component);
6104 	if (status != NFS4_OK)
6105 		return (status);
6106 
6107 	nm = utf8_to_fn(component, &len, NULL);
6108 	if (nm == NULL) {
6109 		return (NFS4ERR_INVAL);
6110 	}
6111 
6112 	if (len > MAXNAMELEN) {
6113 		kmem_free(nm, len);
6114 		return (NFS4ERR_NAMETOOLONG);
6115 	}
6116 
6117 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6118 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6119 	    MAXPATHLEN + 1);
6120 
6121 	if (name == NULL) {
6122 		kmem_free(nm, len);
6123 		return (NFS4ERR_INVAL);
6124 	}
6125 
6126 	status = do_rfs4_op_lookup(name, req, cs);
6127 
6128 	if (name != nm)
6129 		kmem_free(name, MAXPATHLEN + 1);
6130 
6131 	kmem_free(nm, len);
6132 
6133 	return (status);
6134 }
6135 
6136 static nfsstat4
6137 rfs4_lookupfile(component4 *component, struct svc_req *req,
6138     struct compound_state *cs, uint32_t access, change_info4 *cinfo)
6139 {
6140 	nfsstat4 status;
6141 	vnode_t *dvp = cs->vp;
6142 	vattr_t bva, ava, fva;
6143 	int error;
6144 
6145 	/* Get "before" change value */
6146 	bva.va_mask = AT_CTIME|AT_SEQ;
6147 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6148 	if (error)
6149 		return (puterrno4(error));
6150 
6151 	/* rfs4_lookup may VN_RELE directory */
6152 	VN_HOLD(dvp);
6153 
6154 	status = rfs4_lookup(component, req, cs);
6155 	if (status != NFS4_OK) {
6156 		VN_RELE(dvp);
6157 		return (status);
6158 	}
6159 
6160 	/*
6161 	 * Get "after" change value, if it fails, simply return the
6162 	 * before value.
6163 	 */
6164 	ava.va_mask = AT_CTIME|AT_SEQ;
6165 	if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6166 		ava.va_ctime = bva.va_ctime;
6167 		ava.va_seq = 0;
6168 	}
6169 	VN_RELE(dvp);
6170 
6171 	/*
6172 	 * Validate the file is a file
6173 	 */
6174 	fva.va_mask = AT_TYPE|AT_MODE;
6175 	error = VOP_GETATTR(cs->vp, &fva, 0, cs->cr, NULL);
6176 	if (error)
6177 		return (puterrno4(error));
6178 
6179 	if (fva.va_type != VREG) {
6180 		if (fva.va_type == VDIR)
6181 			return (NFS4ERR_ISDIR);
6182 		if (fva.va_type == VLNK)
6183 			return (NFS4ERR_SYMLINK);
6184 		return (NFS4ERR_INVAL);
6185 	}
6186 
6187 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime);
6188 	NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6189 
6190 	/*
6191 	 * It is undefined if VOP_LOOKUP will change va_seq, so
6192 	 * cinfo.atomic = TRUE only if we have
6193 	 * non-zero va_seq's, and they have not changed.
6194 	 */
6195 	if (bva.va_seq && ava.va_seq && ava.va_seq == bva.va_seq)
6196 		cinfo->atomic = TRUE;
6197 	else
6198 		cinfo->atomic = FALSE;
6199 
6200 	/* Check for mandatory locking */
6201 	cs->mandlock = MANDLOCK(cs->vp, fva.va_mode);
6202 	return (check_open_access(access, cs, req));
6203 }
6204 
6205 static nfsstat4
6206 create_vnode(vnode_t *dvp, char *nm,  vattr_t *vap, createmode4 mode,
6207     timespec32_t *mtime, cred_t *cr, vnode_t **vpp, bool_t *created)
6208 {
6209 	int error;
6210 	nfsstat4 status = NFS4_OK;
6211 	vattr_t va;
6212 
6213 tryagain:
6214 
6215 	/*
6216 	 * The file open mode used is VWRITE.  If the client needs
6217 	 * some other semantic, then it should do the access checking
6218 	 * itself.  It would have been nice to have the file open mode
6219 	 * passed as part of the arguments.
6220 	 */
6221 
6222 	*created = TRUE;
6223 	error = VOP_CREATE(dvp, nm, vap, EXCL, VWRITE, vpp, cr, 0, NULL, NULL);
6224 
6225 	if (error) {
6226 		*created = FALSE;
6227 
6228 		/*
6229 		 * If we got something other than file already exists
6230 		 * then just return this error.  Otherwise, we got
6231 		 * EEXIST.  If we were doing a GUARDED create, then
6232 		 * just return this error.  Otherwise, we need to
6233 		 * make sure that this wasn't a duplicate of an
6234 		 * exclusive create request.
6235 		 *
6236 		 * The assumption is made that a non-exclusive create
6237 		 * request will never return EEXIST.
6238 		 */
6239 
6240 		if (error != EEXIST || mode == GUARDED4) {
6241 			status = puterrno4(error);
6242 			return (status);
6243 		}
6244 		error = VOP_LOOKUP(dvp, nm, vpp, NULL, 0, NULL, cr,
6245 		    NULL, NULL, NULL);
6246 
6247 		if (error) {
6248 			/*
6249 			 * We couldn't find the file that we thought that
6250 			 * we just created.  So, we'll just try creating
6251 			 * it again.
6252 			 */
6253 			if (error == ENOENT)
6254 				goto tryagain;
6255 
6256 			status = puterrno4(error);
6257 			return (status);
6258 		}
6259 
6260 		if (mode == UNCHECKED4) {
6261 			/* existing object must be regular file */
6262 			if ((*vpp)->v_type != VREG) {
6263 				if ((*vpp)->v_type == VDIR)
6264 					status = NFS4ERR_ISDIR;
6265 				else if ((*vpp)->v_type == VLNK)
6266 					status = NFS4ERR_SYMLINK;
6267 				else
6268 					status = NFS4ERR_INVAL;
6269 				VN_RELE(*vpp);
6270 				return (status);
6271 			}
6272 
6273 			return (NFS4_OK);
6274 		}
6275 
6276 		/* Check for duplicate request */
6277 		ASSERT(mtime != 0);
6278 		va.va_mask = AT_MTIME;
6279 		error = VOP_GETATTR(*vpp, &va, 0, cr, NULL);
6280 		if (!error) {
6281 			/* We found the file */
6282 			if (va.va_mtime.tv_sec != mtime->tv_sec ||
6283 			    va.va_mtime.tv_nsec != mtime->tv_nsec) {
6284 				/* but its not our creation */
6285 				VN_RELE(*vpp);
6286 				return (NFS4ERR_EXIST);
6287 			}
6288 			*created = TRUE; /* retrans of create == created */
6289 			return (NFS4_OK);
6290 		}
6291 		VN_RELE(*vpp);
6292 		return (NFS4ERR_EXIST);
6293 	}
6294 
6295 	return (NFS4_OK);
6296 }
6297 
6298 static nfsstat4
6299 check_open_access(uint32_t access, struct compound_state *cs,
6300     struct svc_req *req)
6301 {
6302 	int error;
6303 	vnode_t *vp;
6304 	bool_t readonly;
6305 	cred_t *cr = cs->cr;
6306 
6307 	/* For now we don't allow mandatory locking as per V2/V3 */
6308 	if (cs->access == CS_ACCESS_DENIED || cs->mandlock) {
6309 		return (NFS4ERR_ACCESS);
6310 	}
6311 
6312 	vp = cs->vp;
6313 	ASSERT(cr != NULL && vp->v_type == VREG);
6314 
6315 	/*
6316 	 * If the file system is exported read only and we are trying
6317 	 * to open for write, then return NFS4ERR_ROFS
6318 	 */
6319 
6320 	readonly = rdonly4(req, cs);
6321 
6322 	if ((access & OPEN4_SHARE_ACCESS_WRITE) && readonly)
6323 		return (NFS4ERR_ROFS);
6324 
6325 	if (access & OPEN4_SHARE_ACCESS_READ) {
6326 		if ((VOP_ACCESS(vp, VREAD, 0, cr, NULL) != 0) &&
6327 		    (VOP_ACCESS(vp, VEXEC, 0, cr, NULL) != 0)) {
6328 			return (NFS4ERR_ACCESS);
6329 		}
6330 	}
6331 
6332 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
6333 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
6334 		if (error)
6335 			return (NFS4ERR_ACCESS);
6336 	}
6337 
6338 	return (NFS4_OK);
6339 }
6340 
6341 static nfsstat4
6342 rfs4_createfile(OPEN4args *args, struct svc_req *req, struct compound_state *cs,
6343     change_info4 *cinfo, bitmap4 *attrset, clientid4 clientid)
6344 {
6345 	struct nfs4_svgetit_arg sarg;
6346 	struct nfs4_ntov_table ntov;
6347 
6348 	bool_t ntov_table_init = FALSE;
6349 	struct statvfs64 sb;
6350 	nfsstat4 status;
6351 	vnode_t *vp;
6352 	vattr_t bva, ava, iva, cva, *vap;
6353 	vnode_t *dvp;
6354 	timespec32_t *mtime;
6355 	char *nm = NULL;
6356 	uint_t buflen;
6357 	bool_t created;
6358 	bool_t setsize = FALSE;
6359 	len_t reqsize;
6360 	int error;
6361 	bool_t trunc;
6362 	caller_context_t ct;
6363 	component4 *component;
6364 	bslabel_t *clabel;
6365 	struct sockaddr *ca;
6366 	char *name = NULL;
6367 
6368 	sarg.sbp = &sb;
6369 	sarg.is_referral = B_FALSE;
6370 
6371 	dvp = cs->vp;
6372 
6373 	/* Check if the file system is read only */
6374 	if (rdonly4(req, cs))
6375 		return (NFS4ERR_ROFS);
6376 
6377 	/* check the label of including directory */
6378 	if (is_system_labeled()) {
6379 		ASSERT(req->rq_label != NULL);
6380 		clabel = req->rq_label;
6381 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
6382 		    "got client label from request(1)",
6383 		    struct svc_req *, req);
6384 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
6385 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
6386 			    cs->exi)) {
6387 				return (NFS4ERR_ACCESS);
6388 			}
6389 		}
6390 	}
6391 
6392 	/*
6393 	 * Get the last component of path name in nm. cs will reference
6394 	 * the including directory on success.
6395 	 */
6396 	component = &args->open_claim4_u.file;
6397 	status = utf8_dir_verify(component);
6398 	if (status != NFS4_OK)
6399 		return (status);
6400 
6401 	nm = utf8_to_fn(component, &buflen, NULL);
6402 
6403 	if (nm == NULL)
6404 		return (NFS4ERR_RESOURCE);
6405 
6406 	if (buflen > MAXNAMELEN) {
6407 		kmem_free(nm, buflen);
6408 		return (NFS4ERR_NAMETOOLONG);
6409 	}
6410 
6411 	bva.va_mask = AT_TYPE|AT_CTIME|AT_SEQ;
6412 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6413 	if (error) {
6414 		kmem_free(nm, buflen);
6415 		return (puterrno4(error));
6416 	}
6417 
6418 	if (bva.va_type != VDIR) {
6419 		kmem_free(nm, buflen);
6420 		return (NFS4ERR_NOTDIR);
6421 	}
6422 
6423 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime)
6424 
6425 	switch (args->mode) {
6426 	case GUARDED4:
6427 		/*FALLTHROUGH*/
6428 	case UNCHECKED4:
6429 		nfs4_ntov_table_init(&ntov);
6430 		ntov_table_init = TRUE;
6431 
6432 		*attrset = 0;
6433 		status = do_rfs4_set_attrs(attrset,
6434 		    &args->createhow4_u.createattrs,
6435 		    cs, &sarg, &ntov, NFS4ATTR_SETIT);
6436 
6437 		if (status == NFS4_OK && (sarg.vap->va_mask & AT_TYPE) &&
6438 		    sarg.vap->va_type != VREG) {
6439 			if (sarg.vap->va_type == VDIR)
6440 				status = NFS4ERR_ISDIR;
6441 			else if (sarg.vap->va_type == VLNK)
6442 				status = NFS4ERR_SYMLINK;
6443 			else
6444 				status = NFS4ERR_INVAL;
6445 		}
6446 
6447 		if (status != NFS4_OK) {
6448 			kmem_free(nm, buflen);
6449 			nfs4_ntov_table_free(&ntov, &sarg);
6450 			*attrset = 0;
6451 			return (status);
6452 		}
6453 
6454 		vap = sarg.vap;
6455 		vap->va_type = VREG;
6456 		vap->va_mask |= AT_TYPE;
6457 
6458 		if ((vap->va_mask & AT_MODE) == 0) {
6459 			vap->va_mask |= AT_MODE;
6460 			vap->va_mode = (mode_t)0600;
6461 		}
6462 
6463 		if (vap->va_mask & AT_SIZE) {
6464 
6465 			/* Disallow create with a non-zero size */
6466 
6467 			if ((reqsize = sarg.vap->va_size) != 0) {
6468 				kmem_free(nm, buflen);
6469 				nfs4_ntov_table_free(&ntov, &sarg);
6470 				*attrset = 0;
6471 				return (NFS4ERR_INVAL);
6472 			}
6473 			setsize = TRUE;
6474 		}
6475 		break;
6476 
6477 	case EXCLUSIVE4:
6478 		/* prohibit EXCL create of named attributes */
6479 		if (dvp->v_flag & V_XATTRDIR) {
6480 			kmem_free(nm, buflen);
6481 			*attrset = 0;
6482 			return (NFS4ERR_INVAL);
6483 		}
6484 
6485 		cva.va_mask = AT_TYPE | AT_MTIME | AT_MODE;
6486 		cva.va_type = VREG;
6487 		/*
6488 		 * Ensure no time overflows. Assumes underlying
6489 		 * filesystem supports at least 32 bits.
6490 		 * Truncate nsec to usec resolution to allow valid
6491 		 * compares even if the underlying filesystem truncates.
6492 		 */
6493 		mtime = (timespec32_t *)&args->createhow4_u.createverf;
6494 		cva.va_mtime.tv_sec = mtime->tv_sec % TIME32_MAX;
6495 		cva.va_mtime.tv_nsec = (mtime->tv_nsec / 1000) * 1000;
6496 		cva.va_mode = (mode_t)0;
6497 		vap = &cva;
6498 
6499 		/*
6500 		 * For EXCL create, attrset is set to the server attr
6501 		 * used to cache the client's verifier.
6502 		 */
6503 		*attrset = FATTR4_TIME_MODIFY_MASK;
6504 		break;
6505 	}
6506 
6507 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6508 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6509 	    MAXPATHLEN  + 1);
6510 
6511 	if (name == NULL) {
6512 		kmem_free(nm, buflen);
6513 		return (NFS4ERR_SERVERFAULT);
6514 	}
6515 
6516 	status = create_vnode(dvp, name, vap, args->mode, mtime,
6517 	    cs->cr, &vp, &created);
6518 	if (nm != name)
6519 		kmem_free(name, MAXPATHLEN + 1);
6520 	kmem_free(nm, buflen);
6521 
6522 	if (status != NFS4_OK) {
6523 		if (ntov_table_init)
6524 			nfs4_ntov_table_free(&ntov, &sarg);
6525 		*attrset = 0;
6526 		return (status);
6527 	}
6528 
6529 	trunc = (setsize && !created);
6530 
6531 	if (args->mode != EXCLUSIVE4) {
6532 		bitmap4 createmask = args->createhow4_u.createattrs.attrmask;
6533 
6534 		/*
6535 		 * True verification that object was created with correct
6536 		 * attrs is impossible.  The attrs could have been changed
6537 		 * immediately after object creation.  If attributes did
6538 		 * not verify, the only recourse for the server is to
6539 		 * destroy the object.  Maybe if some attrs (like gid)
6540 		 * are set incorrectly, the object should be destroyed;
6541 		 * however, seems bad as a default policy.  Do we really
6542 		 * want to destroy an object over one of the times not
6543 		 * verifying correctly?  For these reasons, the server
6544 		 * currently sets bits in attrset for createattrs
6545 		 * that were set; however, no verification is done.
6546 		 *
6547 		 * vmask_to_nmask accounts for vattr bits set on create
6548 		 *	[do_rfs4_set_attrs() only sets resp bits for
6549 		 *	 non-vattr/vfs bits.]
6550 		 * Mask off any bits we set by default so as not to return
6551 		 * more attrset bits than were requested in createattrs
6552 		 */
6553 		if (created) {
6554 			nfs4_vmask_to_nmask(sarg.vap->va_mask, attrset);
6555 			*attrset &= createmask;
6556 		} else {
6557 			/*
6558 			 * We did not create the vnode (we tried but it
6559 			 * already existed).  In this case, the only createattr
6560 			 * that the spec allows the server to set is size,
6561 			 * and even then, it can only be set if it is 0.
6562 			 */
6563 			*attrset = 0;
6564 			if (trunc)
6565 				*attrset = FATTR4_SIZE_MASK;
6566 		}
6567 	}
6568 	if (ntov_table_init)
6569 		nfs4_ntov_table_free(&ntov, &sarg);
6570 
6571 	/*
6572 	 * Get the initial "after" sequence number, if it fails,
6573 	 * set to zero, time to before.
6574 	 */
6575 	iva.va_mask = AT_CTIME|AT_SEQ;
6576 	if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL)) {
6577 		iva.va_seq = 0;
6578 		iva.va_ctime = bva.va_ctime;
6579 	}
6580 
6581 	/*
6582 	 * create_vnode attempts to create the file exclusive,
6583 	 * if it already exists the VOP_CREATE will fail and
6584 	 * may not increase va_seq. It is atomic if
6585 	 * we haven't changed the directory, but if it has changed
6586 	 * we don't know what changed it.
6587 	 */
6588 	if (!created) {
6589 		if (bva.va_seq && iva.va_seq &&
6590 		    bva.va_seq == iva.va_seq)
6591 			cinfo->atomic = TRUE;
6592 		else
6593 			cinfo->atomic = FALSE;
6594 		NFS4_SET_FATTR4_CHANGE(cinfo->after, iva.va_ctime);
6595 	} else {
6596 		/*
6597 		 * The entry was created, we need to sync the
6598 		 * directory metadata.
6599 		 */
6600 		(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
6601 
6602 		/*
6603 		 * Get "after" change value, if it fails, simply return the
6604 		 * before value.
6605 		 */
6606 		ava.va_mask = AT_CTIME|AT_SEQ;
6607 		if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6608 			ava.va_ctime = bva.va_ctime;
6609 			ava.va_seq = 0;
6610 		}
6611 
6612 		NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6613 
6614 		/*
6615 		 * The cinfo->atomic = TRUE only if we have
6616 		 * non-zero va_seq's, and it has incremented by exactly one
6617 		 * during the create_vnode and it didn't
6618 		 * change during the VOP_FSYNC.
6619 		 */
6620 		if (bva.va_seq && iva.va_seq && ava.va_seq &&
6621 		    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
6622 			cinfo->atomic = TRUE;
6623 		else
6624 			cinfo->atomic = FALSE;
6625 	}
6626 
6627 	/* Check for mandatory locking and that the size gets set. */
6628 	cva.va_mask = AT_MODE;
6629 	if (setsize)
6630 		cva.va_mask |= AT_SIZE;
6631 
6632 	/* Assume the worst */
6633 	cs->mandlock = TRUE;
6634 
6635 	if (VOP_GETATTR(vp, &cva, 0, cs->cr, NULL) == 0) {
6636 		cs->mandlock = MANDLOCK(cs->vp, cva.va_mode);
6637 
6638 		/*
6639 		 * Truncate the file if necessary; this would be
6640 		 * the case for create over an existing file.
6641 		 */
6642 
6643 		if (trunc) {
6644 			int in_crit = 0;
6645 			rfs4_file_t *fp;
6646 			bool_t create = FALSE;
6647 
6648 			/*
6649 			 * We are writing over an existing file.
6650 			 * Check to see if we need to recall a delegation.
6651 			 */
6652 			rfs4_hold_deleg_policy();
6653 			if ((fp = rfs4_findfile(vp, NULL, &create)) != NULL) {
6654 				if (rfs4_check_delegated_byfp(FWRITE, fp,
6655 				    (reqsize == 0), FALSE, FALSE, &clientid)) {
6656 					rfs4_file_rele(fp);
6657 					rfs4_rele_deleg_policy();
6658 					VN_RELE(vp);
6659 					*attrset = 0;
6660 					return (NFS4ERR_DELAY);
6661 				}
6662 				rfs4_file_rele(fp);
6663 			}
6664 			rfs4_rele_deleg_policy();
6665 
6666 			if (nbl_need_check(vp)) {
6667 				in_crit = 1;
6668 
6669 				ASSERT(reqsize == 0);
6670 
6671 				nbl_start_crit(vp, RW_READER);
6672 				if (nbl_conflict(vp, NBL_WRITE, 0,
6673 				    cva.va_size, 0, NULL)) {
6674 					in_crit = 0;
6675 					nbl_end_crit(vp);
6676 					VN_RELE(vp);
6677 					*attrset = 0;
6678 					return (NFS4ERR_ACCESS);
6679 				}
6680 			}
6681 			ct.cc_sysid = 0;
6682 			ct.cc_pid = 0;
6683 			ct.cc_caller_id = nfs4_srv_caller_id;
6684 			ct.cc_flags = CC_DONTBLOCK;
6685 
6686 			cva.va_mask = AT_SIZE;
6687 			cva.va_size = reqsize;
6688 			(void) VOP_SETATTR(vp, &cva, 0, cs->cr, &ct);
6689 			if (in_crit)
6690 				nbl_end_crit(vp);
6691 		}
6692 	}
6693 
6694 	error = makefh4(&cs->fh, vp, cs->exi);
6695 
6696 	/*
6697 	 * Force modified data and metadata out to stable storage.
6698 	 */
6699 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
6700 
6701 	if (error) {
6702 		VN_RELE(vp);
6703 		*attrset = 0;
6704 		return (puterrno4(error));
6705 	}
6706 
6707 	/* if parent dir is attrdir, set namedattr fh flag */
6708 	if (dvp->v_flag & V_XATTRDIR)
6709 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
6710 
6711 	if (cs->vp)
6712 		VN_RELE(cs->vp);
6713 
6714 	cs->vp = vp;
6715 
6716 	/*
6717 	 * if we did not create the file, we will need to check
6718 	 * the access bits on the file
6719 	 */
6720 
6721 	if (!created) {
6722 		if (setsize)
6723 			args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
6724 		status = check_open_access(args->share_access, cs, req);
6725 		if (status != NFS4_OK)
6726 			*attrset = 0;
6727 	}
6728 	return (status);
6729 }
6730 
6731 /*ARGSUSED*/
6732 static void
6733 rfs4_do_open(struct compound_state *cs, struct svc_req *req,
6734     rfs4_openowner_t *oo, delegreq_t deleg,
6735     uint32_t access, uint32_t deny,
6736     OPEN4res *resp, int deleg_cur)
6737 {
6738 	/* XXX Currently not using req  */
6739 	rfs4_state_t *sp;
6740 	rfs4_file_t *fp;
6741 	bool_t screate = TRUE;
6742 	bool_t fcreate = TRUE;
6743 	uint32_t open_a, share_a;
6744 	uint32_t open_d, share_d;
6745 	rfs4_deleg_state_t *dsp;
6746 	sysid_t sysid;
6747 	nfsstat4 status;
6748 	caller_context_t ct;
6749 	int fflags = 0;
6750 	int recall = 0;
6751 	int err;
6752 	int first_open;
6753 
6754 	/* get the file struct and hold a lock on it during initial open */
6755 	fp = rfs4_findfile_withlock(cs->vp, &cs->fh, &fcreate);
6756 	if (fp == NULL) {
6757 		resp->status = NFS4ERR_RESOURCE;
6758 		DTRACE_PROBE1(nfss__e__do__open1, nfsstat4, resp->status);
6759 		return;
6760 	}
6761 
6762 	sp = rfs4_findstate_by_owner_file(oo, fp, &screate);
6763 	if (sp == NULL) {
6764 		resp->status = NFS4ERR_RESOURCE;
6765 		DTRACE_PROBE1(nfss__e__do__open2, nfsstat4, resp->status);
6766 		/* No need to keep any reference */
6767 		rw_exit(&fp->rf_file_rwlock);
6768 		rfs4_file_rele(fp);
6769 		return;
6770 	}
6771 
6772 	/* try to get the sysid before continuing */
6773 	if ((status = rfs4_client_sysid(oo->ro_client, &sysid)) != NFS4_OK) {
6774 		resp->status = status;
6775 		rfs4_file_rele(fp);
6776 		/* Not a fully formed open; "close" it */
6777 		if (screate == TRUE)
6778 			rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6779 		rfs4_state_rele(sp);
6780 		return;
6781 	}
6782 
6783 	/* Calculate the fflags for this OPEN. */
6784 	if (access & OPEN4_SHARE_ACCESS_READ)
6785 		fflags |= FREAD;
6786 	if (access & OPEN4_SHARE_ACCESS_WRITE)
6787 		fflags |= FWRITE;
6788 
6789 	rfs4_dbe_lock(sp->rs_dbe);
6790 
6791 	/*
6792 	 * Calculate the new deny and access mode that this open is adding to
6793 	 * the file for this open owner;
6794 	 */
6795 	open_d = (deny & ~sp->rs_open_deny);
6796 	open_a = (access & ~sp->rs_open_access);
6797 
6798 	/*
6799 	 * Calculate the new share access and share deny modes that this open
6800 	 * is adding to the file for this open owner;
6801 	 */
6802 	share_a = (access & ~sp->rs_share_access);
6803 	share_d = (deny & ~sp->rs_share_deny);
6804 
6805 	first_open = (sp->rs_open_access & OPEN4_SHARE_ACCESS_BOTH) == 0;
6806 
6807 	/*
6808 	 * Check to see the client has already sent an open for this
6809 	 * open owner on this file with the same share/deny modes.
6810 	 * If so, we don't need to check for a conflict and we don't
6811 	 * need to add another shrlock.  If not, then we need to
6812 	 * check for conflicts in deny and access before checking for
6813 	 * conflicts in delegation.  We don't want to recall a
6814 	 * delegation based on an open that will eventually fail based
6815 	 * on shares modes.
6816 	 */
6817 
6818 	if (share_a || share_d) {
6819 		if ((err = rfs4_share(sp, access, deny)) != 0) {
6820 			rfs4_dbe_unlock(sp->rs_dbe);
6821 			resp->status = err;
6822 
6823 			rfs4_file_rele(fp);
6824 			/* Not a fully formed open; "close" it */
6825 			if (screate == TRUE)
6826 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6827 			rfs4_state_rele(sp);
6828 			return;
6829 		}
6830 	}
6831 
6832 	rfs4_dbe_lock(fp->rf_dbe);
6833 
6834 	/*
6835 	 * Check to see if this file is delegated and if so, if a
6836 	 * recall needs to be done.
6837 	 */
6838 	if (rfs4_check_recall(sp, access)) {
6839 		rfs4_dbe_unlock(fp->rf_dbe);
6840 		rfs4_dbe_unlock(sp->rs_dbe);
6841 		rfs4_recall_deleg(fp, FALSE, sp->rs_owner->ro_client);
6842 		delay(NFS4_DELEGATION_CONFLICT_DELAY);
6843 		rfs4_dbe_lock(sp->rs_dbe);
6844 
6845 		/* if state closed while lock was dropped */
6846 		if (sp->rs_closed) {
6847 			if (share_a || share_d)
6848 				(void) rfs4_unshare(sp);
6849 			rfs4_dbe_unlock(sp->rs_dbe);
6850 			rfs4_file_rele(fp);
6851 			/* Not a fully formed open; "close" it */
6852 			if (screate == TRUE)
6853 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6854 			rfs4_state_rele(sp);
6855 			resp->status = NFS4ERR_OLD_STATEID;
6856 			return;
6857 		}
6858 
6859 		rfs4_dbe_lock(fp->rf_dbe);
6860 		/* Let's see if the delegation was returned */
6861 		if (rfs4_check_recall(sp, access)) {
6862 			rfs4_dbe_unlock(fp->rf_dbe);
6863 			if (share_a || share_d)
6864 				(void) rfs4_unshare(sp);
6865 			rfs4_dbe_unlock(sp->rs_dbe);
6866 			rfs4_file_rele(fp);
6867 			rfs4_update_lease(sp->rs_owner->ro_client);
6868 
6869 			/* Not a fully formed open; "close" it */
6870 			if (screate == TRUE)
6871 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6872 			rfs4_state_rele(sp);
6873 			resp->status = NFS4ERR_DELAY;
6874 			return;
6875 		}
6876 	}
6877 	/*
6878 	 * the share check passed and any delegation conflict has been
6879 	 * taken care of, now call vop_open.
6880 	 * if this is the first open then call vop_open with fflags.
6881 	 * if not, call vn_open_upgrade with just the upgrade flags.
6882 	 *
6883 	 * if the file has been opened already, it will have the current
6884 	 * access mode in the state struct.  if it has no share access, then
6885 	 * this is a new open.
6886 	 *
6887 	 * However, if this is open with CLAIM_DLEGATE_CUR, then don't
6888 	 * call VOP_OPEN(), just do the open upgrade.
6889 	 */
6890 	if (first_open && !deleg_cur) {
6891 		ct.cc_sysid = sysid;
6892 		ct.cc_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
6893 		ct.cc_caller_id = nfs4_srv_caller_id;
6894 		ct.cc_flags = CC_DONTBLOCK;
6895 		err = VOP_OPEN(&cs->vp, fflags, cs->cr, &ct);
6896 		if (err) {
6897 			rfs4_dbe_unlock(fp->rf_dbe);
6898 			if (share_a || share_d)
6899 				(void) rfs4_unshare(sp);
6900 			rfs4_dbe_unlock(sp->rs_dbe);
6901 			rfs4_file_rele(fp);
6902 
6903 			/* Not a fully formed open; "close" it */
6904 			if (screate == TRUE)
6905 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6906 			rfs4_state_rele(sp);
6907 			/* check if a monitor detected a delegation conflict */
6908 			if (err == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
6909 				resp->status = NFS4ERR_DELAY;
6910 			else
6911 				resp->status = NFS4ERR_SERVERFAULT;
6912 			return;
6913 		}
6914 	} else { /* open upgrade */
6915 		/*
6916 		 * calculate the fflags for the new mode that is being added
6917 		 * by this upgrade.
6918 		 */
6919 		fflags = 0;
6920 		if (open_a & OPEN4_SHARE_ACCESS_READ)
6921 			fflags |= FREAD;
6922 		if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6923 			fflags |= FWRITE;
6924 		vn_open_upgrade(cs->vp, fflags);
6925 	}
6926 	sp->rs_open_access |= access;
6927 	sp->rs_open_deny |= deny;
6928 
6929 	if (open_d & OPEN4_SHARE_DENY_READ)
6930 		fp->rf_deny_read++;
6931 	if (open_d & OPEN4_SHARE_DENY_WRITE)
6932 		fp->rf_deny_write++;
6933 	fp->rf_share_deny |= deny;
6934 
6935 	if (open_a & OPEN4_SHARE_ACCESS_READ)
6936 		fp->rf_access_read++;
6937 	if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6938 		fp->rf_access_write++;
6939 	fp->rf_share_access |= access;
6940 
6941 	/*
6942 	 * Check for delegation here. if the deleg argument is not
6943 	 * DELEG_ANY, then this is a reclaim from a client and
6944 	 * we must honor the delegation requested. If necessary we can
6945 	 * set the recall flag.
6946 	 */
6947 
6948 	dsp = rfs4_grant_delegation(deleg, sp, &recall);
6949 
6950 	cs->deleg = (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE);
6951 
6952 	next_stateid(&sp->rs_stateid);
6953 
6954 	resp->stateid = sp->rs_stateid.stateid;
6955 
6956 	rfs4_dbe_unlock(fp->rf_dbe);
6957 	rfs4_dbe_unlock(sp->rs_dbe);
6958 
6959 	if (dsp) {
6960 		rfs4_set_deleg_response(dsp, &resp->delegation, NULL, recall);
6961 		rfs4_deleg_state_rele(dsp);
6962 	}
6963 
6964 	rfs4_file_rele(fp);
6965 	rfs4_state_rele(sp);
6966 
6967 	resp->status = NFS4_OK;
6968 }
6969 
6970 /*ARGSUSED*/
6971 static void
6972 rfs4_do_opennull(struct compound_state *cs, struct svc_req *req,
6973     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6974 {
6975 	change_info4 *cinfo = &resp->cinfo;
6976 	bitmap4 *attrset = &resp->attrset;
6977 
6978 	if (args->opentype == OPEN4_NOCREATE)
6979 		resp->status = rfs4_lookupfile(&args->open_claim4_u.file,
6980 		    req, cs, args->share_access, cinfo);
6981 	else {
6982 		/* inhibit delegation grants during exclusive create */
6983 
6984 		if (args->mode == EXCLUSIVE4)
6985 			rfs4_disable_delegation();
6986 
6987 		resp->status = rfs4_createfile(args, req, cs, cinfo, attrset,
6988 		    oo->ro_client->rc_clientid);
6989 	}
6990 
6991 	if (resp->status == NFS4_OK) {
6992 
6993 		/* cs->vp cs->fh now reference the desired file */
6994 
6995 		rfs4_do_open(cs, req, oo,
6996 		    oo->ro_need_confirm ? DELEG_NONE : DELEG_ANY,
6997 		    args->share_access, args->share_deny, resp, 0);
6998 
6999 		/*
7000 		 * If rfs4_createfile set attrset, we must
7001 		 * clear this attrset before the response is copied.
7002 		 */
7003 		if (resp->status != NFS4_OK && resp->attrset) {
7004 			resp->attrset = 0;
7005 		}
7006 	}
7007 	else
7008 		*cs->statusp = resp->status;
7009 
7010 	if (args->mode == EXCLUSIVE4)
7011 		rfs4_enable_delegation();
7012 }
7013 
7014 /*ARGSUSED*/
7015 static void
7016 rfs4_do_openprev(struct compound_state *cs, struct svc_req *req,
7017     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7018 {
7019 	change_info4 *cinfo = &resp->cinfo;
7020 	vattr_t va;
7021 	vtype_t v_type = cs->vp->v_type;
7022 	int error = 0;
7023 
7024 	/* Verify that we have a regular file */
7025 	if (v_type != VREG) {
7026 		if (v_type == VDIR)
7027 			resp->status = NFS4ERR_ISDIR;
7028 		else if (v_type == VLNK)
7029 			resp->status = NFS4ERR_SYMLINK;
7030 		else
7031 			resp->status = NFS4ERR_INVAL;
7032 		return;
7033 	}
7034 
7035 	va.va_mask = AT_MODE|AT_UID;
7036 	error = VOP_GETATTR(cs->vp, &va, 0, cs->cr, NULL);
7037 	if (error) {
7038 		resp->status = puterrno4(error);
7039 		return;
7040 	}
7041 
7042 	cs->mandlock = MANDLOCK(cs->vp, va.va_mode);
7043 
7044 	/*
7045 	 * Check if we have access to the file, Note the the file
7046 	 * could have originally been open UNCHECKED or GUARDED
7047 	 * with mode bits that will now fail, but there is nothing
7048 	 * we can really do about that except in the case that the
7049 	 * owner of the file is the one requesting the open.
7050 	 */
7051 	if (crgetuid(cs->cr) != va.va_uid) {
7052 		resp->status = check_open_access(args->share_access, cs, req);
7053 		if (resp->status != NFS4_OK) {
7054 			return;
7055 		}
7056 	}
7057 
7058 	/*
7059 	 * cinfo on a CLAIM_PREVIOUS is undefined, initialize to zero
7060 	 */
7061 	cinfo->before = 0;
7062 	cinfo->after = 0;
7063 	cinfo->atomic = FALSE;
7064 
7065 	rfs4_do_open(cs, req, oo,
7066 	    NFS4_DELEG4TYPE2REQTYPE(args->open_claim4_u.delegate_type),
7067 	    args->share_access, args->share_deny, resp, 0);
7068 }
7069 
7070 static void
7071 rfs4_do_opendelcur(struct compound_state *cs, struct svc_req *req,
7072     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7073 {
7074 	int error;
7075 	nfsstat4 status;
7076 	stateid4 stateid =
7077 	    args->open_claim4_u.delegate_cur_info.delegate_stateid;
7078 	rfs4_deleg_state_t *dsp;
7079 
7080 	/*
7081 	 * Find the state info from the stateid and confirm that the
7082 	 * file is delegated.  If the state openowner is the same as
7083 	 * the supplied openowner we're done. If not, get the file
7084 	 * info from the found state info. Use that file info to
7085 	 * create the state for this lock owner. Note solaris doen't
7086 	 * really need the pathname to find the file. We may want to
7087 	 * lookup the pathname and make sure that the vp exist and
7088 	 * matches the vp in the file structure. However it is
7089 	 * possible that the pathname nolonger exists (local process
7090 	 * unlinks the file), so this may not be that useful.
7091 	 */
7092 
7093 	status = rfs4_get_deleg_state(&stateid, &dsp);
7094 	if (status != NFS4_OK) {
7095 		resp->status = status;
7096 		return;
7097 	}
7098 
7099 	ASSERT(dsp->rds_finfo->rf_dinfo.rd_dtype != OPEN_DELEGATE_NONE);
7100 
7101 	/*
7102 	 * New lock owner, create state. Since this was probably called
7103 	 * in response to a CB_RECALL we set deleg to DELEG_NONE
7104 	 */
7105 
7106 	ASSERT(cs->vp != NULL);
7107 	VN_RELE(cs->vp);
7108 	VN_HOLD(dsp->rds_finfo->rf_vp);
7109 	cs->vp = dsp->rds_finfo->rf_vp;
7110 
7111 	if (error = makefh4(&cs->fh, cs->vp, cs->exi)) {
7112 		rfs4_deleg_state_rele(dsp);
7113 		*cs->statusp = resp->status = puterrno4(error);
7114 		return;
7115 	}
7116 
7117 	/* Mark progress for delegation returns */
7118 	dsp->rds_finfo->rf_dinfo.rd_time_lastwrite = gethrestime_sec();
7119 	rfs4_deleg_state_rele(dsp);
7120 	rfs4_do_open(cs, req, oo, DELEG_NONE,
7121 	    args->share_access, args->share_deny, resp, 1);
7122 }
7123 
7124 /*ARGSUSED*/
7125 static void
7126 rfs4_do_opendelprev(struct compound_state *cs, struct svc_req *req,
7127     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7128 {
7129 	/*
7130 	 * Lookup the pathname, it must already exist since this file
7131 	 * was delegated.
7132 	 *
7133 	 * Find the file and state info for this vp and open owner pair.
7134 	 *	check that they are in fact delegated.
7135 	 *	check that the state access and deny modes are the same.
7136 	 *
7137 	 * Return the delgation possibly seting the recall flag.
7138 	 */
7139 	rfs4_file_t *fp;
7140 	rfs4_state_t *sp;
7141 	bool_t create = FALSE;
7142 	bool_t dcreate = FALSE;
7143 	rfs4_deleg_state_t *dsp;
7144 	nfsace4 *ace;
7145 
7146 	/* Note we ignore oflags */
7147 	resp->status = rfs4_lookupfile(&args->open_claim4_u.file_delegate_prev,
7148 	    req, cs, args->share_access, &resp->cinfo);
7149 
7150 	if (resp->status != NFS4_OK) {
7151 		return;
7152 	}
7153 
7154 	/* get the file struct and hold a lock on it during initial open */
7155 	fp = rfs4_findfile_withlock(cs->vp, NULL, &create);
7156 	if (fp == NULL) {
7157 		resp->status = NFS4ERR_RESOURCE;
7158 		DTRACE_PROBE1(nfss__e__do_opendelprev1, nfsstat4, resp->status);
7159 		return;
7160 	}
7161 
7162 	sp = rfs4_findstate_by_owner_file(oo, fp, &create);
7163 	if (sp == NULL) {
7164 		resp->status = NFS4ERR_SERVERFAULT;
7165 		DTRACE_PROBE1(nfss__e__do_opendelprev2, nfsstat4, resp->status);
7166 		rw_exit(&fp->rf_file_rwlock);
7167 		rfs4_file_rele(fp);
7168 		return;
7169 	}
7170 
7171 	rfs4_dbe_lock(sp->rs_dbe);
7172 	rfs4_dbe_lock(fp->rf_dbe);
7173 	if (args->share_access != sp->rs_share_access ||
7174 	    args->share_deny != sp->rs_share_deny ||
7175 	    sp->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
7176 		NFS4_DEBUG(rfs4_debug,
7177 		    (CE_NOTE, "rfs4_do_opendelprev: state mixup"));
7178 		rfs4_dbe_unlock(fp->rf_dbe);
7179 		rfs4_dbe_unlock(sp->rs_dbe);
7180 		rfs4_file_rele(fp);
7181 		rfs4_state_rele(sp);
7182 		resp->status = NFS4ERR_SERVERFAULT;
7183 		return;
7184 	}
7185 	rfs4_dbe_unlock(fp->rf_dbe);
7186 	rfs4_dbe_unlock(sp->rs_dbe);
7187 
7188 	dsp = rfs4_finddeleg(sp, &dcreate);
7189 	if (dsp == NULL) {
7190 		rfs4_state_rele(sp);
7191 		rfs4_file_rele(fp);
7192 		resp->status = NFS4ERR_SERVERFAULT;
7193 		return;
7194 	}
7195 
7196 	next_stateid(&sp->rs_stateid);
7197 
7198 	resp->stateid = sp->rs_stateid.stateid;
7199 
7200 	resp->delegation.delegation_type = dsp->rds_dtype;
7201 
7202 	if (dsp->rds_dtype == OPEN_DELEGATE_READ) {
7203 		open_read_delegation4 *rv =
7204 		    &resp->delegation.open_delegation4_u.read;
7205 
7206 		rv->stateid = dsp->rds_delegid.stateid;
7207 		rv->recall = FALSE; /* no policy in place to set to TRUE */
7208 		ace = &rv->permissions;
7209 	} else {
7210 		open_write_delegation4 *rv =
7211 		    &resp->delegation.open_delegation4_u.write;
7212 
7213 		rv->stateid = dsp->rds_delegid.stateid;
7214 		rv->recall = FALSE;  /* no policy in place to set to TRUE */
7215 		ace = &rv->permissions;
7216 		rv->space_limit.limitby = NFS_LIMIT_SIZE;
7217 		rv->space_limit.nfs_space_limit4_u.filesize = UINT64_MAX;
7218 	}
7219 
7220 	/* XXX For now */
7221 	ace->type = ACE4_ACCESS_ALLOWED_ACE_TYPE;
7222 	ace->flag = 0;
7223 	ace->access_mask = 0;
7224 	ace->who.utf8string_len = 0;
7225 	ace->who.utf8string_val = 0;
7226 
7227 	rfs4_deleg_state_rele(dsp);
7228 	rfs4_state_rele(sp);
7229 	rfs4_file_rele(fp);
7230 }
7231 
7232 typedef enum {
7233 	NFS4_CHKSEQ_OKAY = 0,
7234 	NFS4_CHKSEQ_REPLAY = 1,
7235 	NFS4_CHKSEQ_BAD = 2
7236 } rfs4_chkseq_t;
7237 
7238 /*
7239  * Generic function for sequence number checks.
7240  */
7241 static rfs4_chkseq_t
7242 rfs4_check_seqid(seqid4 seqid, nfs_resop4 *lastop,
7243     seqid4 rqst_seq, nfs_resop4 *resop, bool_t copyres)
7244 {
7245 	/* Same sequence ids and matching operations? */
7246 	if (seqid == rqst_seq && resop->resop == lastop->resop) {
7247 		if (copyres == TRUE) {
7248 			rfs4_free_reply(resop);
7249 			rfs4_copy_reply(resop, lastop);
7250 		}
7251 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
7252 		    "Replayed SEQID %d\n", seqid));
7253 		return (NFS4_CHKSEQ_REPLAY);
7254 	}
7255 
7256 	/* If the incoming sequence is not the next expected then it is bad */
7257 	if (rqst_seq != seqid + 1) {
7258 		if (rqst_seq == seqid) {
7259 			NFS4_DEBUG(rfs4_debug,
7260 			    (CE_NOTE, "BAD SEQID: Replayed sequence id "
7261 			    "but last op was %d current op is %d\n",
7262 			    lastop->resop, resop->resop));
7263 			return (NFS4_CHKSEQ_BAD);
7264 		}
7265 		NFS4_DEBUG(rfs4_debug,
7266 		    (CE_NOTE, "BAD SEQID: got %u expecting %u\n",
7267 		    rqst_seq, seqid));
7268 		return (NFS4_CHKSEQ_BAD);
7269 	}
7270 
7271 	/* Everything okay -- next expected */
7272 	return (NFS4_CHKSEQ_OKAY);
7273 }
7274 
7275 
7276 static rfs4_chkseq_t
7277 rfs4_check_open_seqid(seqid4 seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7278 {
7279 	rfs4_chkseq_t rc;
7280 
7281 	rfs4_dbe_lock(op->ro_dbe);
7282 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply, seqid, resop,
7283 	    TRUE);
7284 	rfs4_dbe_unlock(op->ro_dbe);
7285 
7286 	if (rc == NFS4_CHKSEQ_OKAY)
7287 		rfs4_update_lease(op->ro_client);
7288 
7289 	return (rc);
7290 }
7291 
7292 static rfs4_chkseq_t
7293 rfs4_check_olo_seqid(seqid4 olo_seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7294 {
7295 	rfs4_chkseq_t rc;
7296 
7297 	rfs4_dbe_lock(op->ro_dbe);
7298 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply,
7299 	    olo_seqid, resop, FALSE);
7300 	rfs4_dbe_unlock(op->ro_dbe);
7301 
7302 	return (rc);
7303 }
7304 
7305 static rfs4_chkseq_t
7306 rfs4_check_lock_seqid(seqid4 seqid, rfs4_lo_state_t *lsp, nfs_resop4 *resop)
7307 {
7308 	rfs4_chkseq_t rc = NFS4_CHKSEQ_OKAY;
7309 
7310 	rfs4_dbe_lock(lsp->rls_dbe);
7311 	if (!lsp->rls_skip_seqid_check)
7312 		rc = rfs4_check_seqid(lsp->rls_seqid, &lsp->rls_reply, seqid,
7313 		    resop, TRUE);
7314 	rfs4_dbe_unlock(lsp->rls_dbe);
7315 
7316 	return (rc);
7317 }
7318 
7319 static void
7320 rfs4_op_open(nfs_argop4 *argop, nfs_resop4 *resop,
7321     struct svc_req *req, struct compound_state *cs)
7322 {
7323 	OPEN4args *args = &argop->nfs_argop4_u.opopen;
7324 	OPEN4res *resp = &resop->nfs_resop4_u.opopen;
7325 	open_owner4 *owner = &args->owner;
7326 	open_claim_type4 claim = args->claim;
7327 	rfs4_client_t *cp;
7328 	rfs4_openowner_t *oo;
7329 	bool_t create;
7330 	bool_t replay = FALSE;
7331 	int can_reclaim;
7332 
7333 	DTRACE_NFSV4_2(op__open__start, struct compound_state *, cs,
7334 	    OPEN4args *, args);
7335 
7336 	if (cs->vp == NULL) {
7337 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7338 		goto end;
7339 	}
7340 
7341 	/*
7342 	 * Need to check clientid and lease expiration first based on
7343 	 * error ordering and incrementing sequence id.
7344 	 */
7345 	cp = rfs4_findclient_by_id(owner->clientid, FALSE);
7346 	if (cp == NULL) {
7347 		*cs->statusp = resp->status =
7348 		    rfs4_check_clientid(&owner->clientid, 0);
7349 		goto end;
7350 	}
7351 
7352 	if (rfs4_lease_expired(cp)) {
7353 		rfs4_client_close(cp);
7354 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7355 		goto end;
7356 	}
7357 	can_reclaim = cp->rc_can_reclaim;
7358 
7359 	/*
7360 	 * Find the open_owner for use from this point forward.  Take
7361 	 * care in updating the sequence id based on the type of error
7362 	 * being returned.
7363 	 */
7364 retry:
7365 	create = TRUE;
7366 	oo = rfs4_findopenowner(owner, &create, args->seqid);
7367 	if (oo == NULL) {
7368 		*cs->statusp = resp->status = NFS4ERR_RESOURCE;
7369 		rfs4_client_rele(cp);
7370 		goto end;
7371 	}
7372 
7373 	/* Hold off access to the sequence space while the open is done */
7374 	rfs4_sw_enter(&oo->ro_sw);
7375 
7376 	/*
7377 	 * If the open_owner existed before at the server, then check
7378 	 * the sequence id.
7379 	 */
7380 	if (!create && !oo->ro_postpone_confirm) {
7381 		switch (rfs4_check_open_seqid(args->seqid, oo, resop)) {
7382 		case NFS4_CHKSEQ_BAD:
7383 			if ((args->seqid > oo->ro_open_seqid) &&
7384 			    oo->ro_need_confirm) {
7385 				rfs4_free_opens(oo, TRUE, FALSE);
7386 				rfs4_sw_exit(&oo->ro_sw);
7387 				rfs4_openowner_rele(oo);
7388 				goto retry;
7389 			}
7390 			resp->status = NFS4ERR_BAD_SEQID;
7391 			goto out;
7392 		case NFS4_CHKSEQ_REPLAY: /* replay of previous request */
7393 			replay = TRUE;
7394 			goto out;
7395 		default:
7396 			break;
7397 		}
7398 
7399 		/*
7400 		 * Sequence was ok and open owner exists
7401 		 * check to see if we have yet to see an
7402 		 * open_confirm.
7403 		 */
7404 		if (oo->ro_need_confirm) {
7405 			rfs4_free_opens(oo, TRUE, FALSE);
7406 			rfs4_sw_exit(&oo->ro_sw);
7407 			rfs4_openowner_rele(oo);
7408 			goto retry;
7409 		}
7410 	}
7411 	/* Grace only applies to regular-type OPENs */
7412 	if (rfs4_clnt_in_grace(cp) &&
7413 	    (claim == CLAIM_NULL || claim == CLAIM_DELEGATE_CUR)) {
7414 		*cs->statusp = resp->status = NFS4ERR_GRACE;
7415 		goto out;
7416 	}
7417 
7418 	/*
7419 	 * If previous state at the server existed then can_reclaim
7420 	 * will be set. If not reply NFS4ERR_NO_GRACE to the
7421 	 * client.
7422 	 */
7423 	if (rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS && !can_reclaim) {
7424 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7425 		goto out;
7426 	}
7427 
7428 
7429 	/*
7430 	 * Reject the open if the client has missed the grace period
7431 	 */
7432 	if (!rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS) {
7433 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7434 		goto out;
7435 	}
7436 
7437 	/* Couple of up-front bookkeeping items */
7438 	if (oo->ro_need_confirm) {
7439 		/*
7440 		 * If this is a reclaim OPEN then we should not ask
7441 		 * for a confirmation of the open_owner per the
7442 		 * protocol specification.
7443 		 */
7444 		if (claim == CLAIM_PREVIOUS)
7445 			oo->ro_need_confirm = FALSE;
7446 		else
7447 			resp->rflags |= OPEN4_RESULT_CONFIRM;
7448 	}
7449 	resp->rflags |= OPEN4_RESULT_LOCKTYPE_POSIX;
7450 
7451 	/*
7452 	 * If there is an unshared filesystem mounted on this vnode,
7453 	 * do not allow to open/create in this directory.
7454 	 */
7455 	if (vn_ismntpt(cs->vp)) {
7456 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
7457 		goto out;
7458 	}
7459 
7460 	/*
7461 	 * access must READ, WRITE, or BOTH.  No access is invalid.
7462 	 * deny can be READ, WRITE, BOTH, or NONE.
7463 	 * bits not defined for access/deny are invalid.
7464 	 */
7465 	if (! (args->share_access & OPEN4_SHARE_ACCESS_BOTH) ||
7466 	    (args->share_access & ~OPEN4_SHARE_ACCESS_BOTH) ||
7467 	    (args->share_deny & ~OPEN4_SHARE_DENY_BOTH)) {
7468 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7469 		goto out;
7470 	}
7471 
7472 
7473 	/*
7474 	 * make sure attrset is zero before response is built.
7475 	 */
7476 	resp->attrset = 0;
7477 
7478 	switch (claim) {
7479 	case CLAIM_NULL:
7480 		rfs4_do_opennull(cs, req, args, oo, resp);
7481 		break;
7482 	case CLAIM_PREVIOUS:
7483 		rfs4_do_openprev(cs, req, args, oo, resp);
7484 		break;
7485 	case CLAIM_DELEGATE_CUR:
7486 		rfs4_do_opendelcur(cs, req, args, oo, resp);
7487 		break;
7488 	case CLAIM_DELEGATE_PREV:
7489 		rfs4_do_opendelprev(cs, req, args, oo, resp);
7490 		break;
7491 	default:
7492 		resp->status = NFS4ERR_INVAL;
7493 		break;
7494 	}
7495 
7496 out:
7497 	rfs4_client_rele(cp);
7498 
7499 	/* Catch sequence id handling here to make it a little easier */
7500 	switch (resp->status) {
7501 	case NFS4ERR_BADXDR:
7502 	case NFS4ERR_BAD_SEQID:
7503 	case NFS4ERR_BAD_STATEID:
7504 	case NFS4ERR_NOFILEHANDLE:
7505 	case NFS4ERR_RESOURCE:
7506 	case NFS4ERR_STALE_CLIENTID:
7507 	case NFS4ERR_STALE_STATEID:
7508 		/*
7509 		 * The protocol states that if any of these errors are
7510 		 * being returned, the sequence id should not be
7511 		 * incremented.  Any other return requires an
7512 		 * increment.
7513 		 */
7514 		break;
7515 	default:
7516 		/* Always update the lease in this case */
7517 		rfs4_update_lease(oo->ro_client);
7518 
7519 		/* Regular response - copy the result */
7520 		if (!replay)
7521 			rfs4_update_open_resp(oo, resop, &cs->fh);
7522 
7523 		/*
7524 		 * REPLAY case: Only if the previous response was OK
7525 		 * do we copy the filehandle.  If not OK, no
7526 		 * filehandle to copy.
7527 		 */
7528 		if (replay == TRUE &&
7529 		    resp->status == NFS4_OK &&
7530 		    oo->ro_reply_fh.nfs_fh4_val) {
7531 			/*
7532 			 * If this is a replay, we must restore the
7533 			 * current filehandle/vp to that of what was
7534 			 * returned originally.  Try our best to do
7535 			 * it.
7536 			 */
7537 			nfs_fh4_fmt_t *fh_fmtp =
7538 			    (nfs_fh4_fmt_t *)oo->ro_reply_fh.nfs_fh4_val;
7539 
7540 			if (cs->exi)
7541 				exi_rele(cs->exi);
7542 			cs->exi = checkexport(&fh_fmtp->fh4_fsid,
7543 			    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
7544 
7545 			if (cs->exi == NULL) {
7546 				resp->status = NFS4ERR_STALE;
7547 				goto finish;
7548 			}
7549 
7550 			VN_RELE(cs->vp);
7551 
7552 			cs->vp = nfs4_fhtovp(&oo->ro_reply_fh, cs->exi,
7553 			    &resp->status);
7554 
7555 			if (cs->vp == NULL)
7556 				goto finish;
7557 
7558 			nfs_fh4_copy(&oo->ro_reply_fh, &cs->fh);
7559 		}
7560 
7561 		/*
7562 		 * If this was a replay, no need to update the
7563 		 * sequence id. If the open_owner was not created on
7564 		 * this pass, then update.  The first use of an
7565 		 * open_owner will not bump the sequence id.
7566 		 */
7567 		if (replay == FALSE && !create)
7568 			rfs4_update_open_sequence(oo);
7569 		/*
7570 		 * If the client is receiving an error and the
7571 		 * open_owner needs to be confirmed, there is no way
7572 		 * to notify the client of this fact ignoring the fact
7573 		 * that the server has no method of returning a
7574 		 * stateid to confirm.  Therefore, the server needs to
7575 		 * mark this open_owner in a way as to avoid the
7576 		 * sequence id checking the next time the client uses
7577 		 * this open_owner.
7578 		 */
7579 		if (resp->status != NFS4_OK && oo->ro_need_confirm)
7580 			oo->ro_postpone_confirm = TRUE;
7581 		/*
7582 		 * If OK response then clear the postpone flag and
7583 		 * reset the sequence id to keep in sync with the
7584 		 * client.
7585 		 */
7586 		if (resp->status == NFS4_OK && oo->ro_postpone_confirm) {
7587 			oo->ro_postpone_confirm = FALSE;
7588 			oo->ro_open_seqid = args->seqid;
7589 		}
7590 		break;
7591 	}
7592 
7593 finish:
7594 	*cs->statusp = resp->status;
7595 
7596 	rfs4_sw_exit(&oo->ro_sw);
7597 	rfs4_openowner_rele(oo);
7598 
7599 end:
7600 	DTRACE_NFSV4_2(op__open__done, struct compound_state *, cs,
7601 	    OPEN4res *, resp);
7602 }
7603 
7604 /*ARGSUSED*/
7605 void
7606 rfs4_op_open_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
7607     struct svc_req *req, struct compound_state *cs)
7608 {
7609 	OPEN_CONFIRM4args *args = &argop->nfs_argop4_u.opopen_confirm;
7610 	OPEN_CONFIRM4res *resp = &resop->nfs_resop4_u.opopen_confirm;
7611 	rfs4_state_t *sp;
7612 	nfsstat4 status;
7613 
7614 	DTRACE_NFSV4_2(op__open__confirm__start, struct compound_state *, cs,
7615 	    OPEN_CONFIRM4args *, args);
7616 
7617 	if (cs->vp == NULL) {
7618 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7619 		goto out;
7620 	}
7621 
7622 	if (cs->vp->v_type != VREG) {
7623 		*cs->statusp = resp->status =
7624 		    cs->vp->v_type == VDIR ? NFS4ERR_ISDIR : NFS4ERR_INVAL;
7625 		return;
7626 	}
7627 
7628 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7629 	if (status != NFS4_OK) {
7630 		*cs->statusp = resp->status = status;
7631 		goto out;
7632 	}
7633 
7634 	/* Ensure specified filehandle matches */
7635 	if (cs->vp != sp->rs_finfo->rf_vp) {
7636 		rfs4_state_rele(sp);
7637 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7638 		goto out;
7639 	}
7640 
7641 	/* hold off other access to open_owner while we tinker */
7642 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7643 
7644 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7645 	case NFS4_CHECK_STATEID_OKAY:
7646 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7647 		    resop) != 0) {
7648 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7649 			break;
7650 		}
7651 		/*
7652 		 * If it is the appropriate stateid and determined to
7653 		 * be "OKAY" then this means that the stateid does not
7654 		 * need to be confirmed and the client is in error for
7655 		 * sending an OPEN_CONFIRM.
7656 		 */
7657 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7658 		break;
7659 	case NFS4_CHECK_STATEID_OLD:
7660 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7661 		break;
7662 	case NFS4_CHECK_STATEID_BAD:
7663 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7664 		break;
7665 	case NFS4_CHECK_STATEID_EXPIRED:
7666 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7667 		break;
7668 	case NFS4_CHECK_STATEID_CLOSED:
7669 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7670 		break;
7671 	case NFS4_CHECK_STATEID_REPLAY:
7672 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7673 		    resop)) {
7674 		case NFS4_CHKSEQ_OKAY:
7675 			/*
7676 			 * This is replayed stateid; if seqid matches
7677 			 * next expected, then client is using wrong seqid.
7678 			 */
7679 			/* fall through */
7680 		case NFS4_CHKSEQ_BAD:
7681 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7682 			break;
7683 		case NFS4_CHKSEQ_REPLAY:
7684 			/*
7685 			 * Note this case is the duplicate case so
7686 			 * resp->status is already set.
7687 			 */
7688 			*cs->statusp = resp->status;
7689 			rfs4_update_lease(sp->rs_owner->ro_client);
7690 			break;
7691 		}
7692 		break;
7693 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7694 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7695 		    resop) != NFS4_CHKSEQ_OKAY) {
7696 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7697 			break;
7698 		}
7699 		*cs->statusp = resp->status = NFS4_OK;
7700 
7701 		next_stateid(&sp->rs_stateid);
7702 		resp->open_stateid = sp->rs_stateid.stateid;
7703 		sp->rs_owner->ro_need_confirm = FALSE;
7704 		rfs4_update_lease(sp->rs_owner->ro_client);
7705 		rfs4_update_open_sequence(sp->rs_owner);
7706 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7707 		break;
7708 	default:
7709 		ASSERT(FALSE);
7710 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7711 		break;
7712 	}
7713 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7714 	rfs4_state_rele(sp);
7715 
7716 out:
7717 	DTRACE_NFSV4_2(op__open__confirm__done, struct compound_state *, cs,
7718 	    OPEN_CONFIRM4res *, resp);
7719 }
7720 
7721 /*ARGSUSED*/
7722 void
7723 rfs4_op_open_downgrade(nfs_argop4 *argop, nfs_resop4 *resop,
7724     struct svc_req *req, struct compound_state *cs)
7725 {
7726 	OPEN_DOWNGRADE4args *args = &argop->nfs_argop4_u.opopen_downgrade;
7727 	OPEN_DOWNGRADE4res *resp = &resop->nfs_resop4_u.opopen_downgrade;
7728 	uint32_t access = args->share_access;
7729 	uint32_t deny = args->share_deny;
7730 	nfsstat4 status;
7731 	rfs4_state_t *sp;
7732 	rfs4_file_t *fp;
7733 	int fflags = 0;
7734 
7735 	DTRACE_NFSV4_2(op__open__downgrade__start, struct compound_state *, cs,
7736 	    OPEN_DOWNGRADE4args *, args);
7737 
7738 	if (cs->vp == NULL) {
7739 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7740 		goto out;
7741 	}
7742 
7743 	if (cs->vp->v_type != VREG) {
7744 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7745 		return;
7746 	}
7747 
7748 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7749 	if (status != NFS4_OK) {
7750 		*cs->statusp = resp->status = status;
7751 		goto out;
7752 	}
7753 
7754 	/* Ensure specified filehandle matches */
7755 	if (cs->vp != sp->rs_finfo->rf_vp) {
7756 		rfs4_state_rele(sp);
7757 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7758 		goto out;
7759 	}
7760 
7761 	/* hold off other access to open_owner while we tinker */
7762 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7763 
7764 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7765 	case NFS4_CHECK_STATEID_OKAY:
7766 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7767 		    resop) != NFS4_CHKSEQ_OKAY) {
7768 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7769 			goto end;
7770 		}
7771 		break;
7772 	case NFS4_CHECK_STATEID_OLD:
7773 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7774 		goto end;
7775 	case NFS4_CHECK_STATEID_BAD:
7776 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7777 		goto end;
7778 	case NFS4_CHECK_STATEID_EXPIRED:
7779 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7780 		goto end;
7781 	case NFS4_CHECK_STATEID_CLOSED:
7782 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7783 		goto end;
7784 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7785 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7786 		goto end;
7787 	case NFS4_CHECK_STATEID_REPLAY:
7788 		/* Check the sequence id for the open owner */
7789 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7790 		    resop)) {
7791 		case NFS4_CHKSEQ_OKAY:
7792 			/*
7793 			 * This is replayed stateid; if seqid matches
7794 			 * next expected, then client is using wrong seqid.
7795 			 */
7796 			/* fall through */
7797 		case NFS4_CHKSEQ_BAD:
7798 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7799 			goto end;
7800 		case NFS4_CHKSEQ_REPLAY:
7801 			/*
7802 			 * Note this case is the duplicate case so
7803 			 * resp->status is already set.
7804 			 */
7805 			*cs->statusp = resp->status;
7806 			rfs4_update_lease(sp->rs_owner->ro_client);
7807 			goto end;
7808 		}
7809 		break;
7810 	default:
7811 		ASSERT(FALSE);
7812 		break;
7813 	}
7814 
7815 	rfs4_dbe_lock(sp->rs_dbe);
7816 	/*
7817 	 * Check that the new access modes and deny modes are valid.
7818 	 * Check that no invalid bits are set.
7819 	 */
7820 	if ((access & ~(OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) ||
7821 	    (deny & ~(OPEN4_SHARE_DENY_READ | OPEN4_SHARE_DENY_WRITE))) {
7822 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7823 		rfs4_update_open_sequence(sp->rs_owner);
7824 		rfs4_dbe_unlock(sp->rs_dbe);
7825 		goto end;
7826 	}
7827 
7828 	/*
7829 	 * The new modes must be a subset of the current modes and
7830 	 * the access must specify at least one mode. To test that
7831 	 * the new mode is a subset of the current modes we bitwise
7832 	 * AND them together and check that the result equals the new
7833 	 * mode. For example:
7834 	 * New mode, access == R and current mode, sp->rs_open_access  == RW
7835 	 * access & sp->rs_open_access == R == access, so the new access mode
7836 	 * is valid. Consider access == RW, sp->rs_open_access = R
7837 	 * access & sp->rs_open_access == R != access, so the new access mode
7838 	 * is invalid.
7839 	 */
7840 	if ((access & sp->rs_open_access) != access ||
7841 	    (deny & sp->rs_open_deny) != deny ||
7842 	    (access &
7843 	    (OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) == 0) {
7844 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7845 		rfs4_update_open_sequence(sp->rs_owner);
7846 		rfs4_dbe_unlock(sp->rs_dbe);
7847 		goto end;
7848 	}
7849 
7850 	/*
7851 	 * Release any share locks associated with this stateID.
7852 	 * Strictly speaking, this violates the spec because the
7853 	 * spec effectively requires that open downgrade be atomic.
7854 	 * At present, fs_shrlock does not have this capability.
7855 	 */
7856 	(void) rfs4_unshare(sp);
7857 
7858 	status = rfs4_share(sp, access, deny);
7859 	if (status != NFS4_OK) {
7860 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7861 		rfs4_update_open_sequence(sp->rs_owner);
7862 		rfs4_dbe_unlock(sp->rs_dbe);
7863 		goto end;
7864 	}
7865 
7866 	fp = sp->rs_finfo;
7867 	rfs4_dbe_lock(fp->rf_dbe);
7868 
7869 	/*
7870 	 * If the current mode has deny read and the new mode
7871 	 * does not, decrement the number of deny read mode bits
7872 	 * and if it goes to zero turn off the deny read bit
7873 	 * on the file.
7874 	 */
7875 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_READ) &&
7876 	    (deny & OPEN4_SHARE_DENY_READ) == 0) {
7877 		fp->rf_deny_read--;
7878 		if (fp->rf_deny_read == 0)
7879 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
7880 	}
7881 
7882 	/*
7883 	 * If the current mode has deny write and the new mode
7884 	 * does not, decrement the number of deny write mode bits
7885 	 * and if it goes to zero turn off the deny write bit
7886 	 * on the file.
7887 	 */
7888 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) &&
7889 	    (deny & OPEN4_SHARE_DENY_WRITE) == 0) {
7890 		fp->rf_deny_write--;
7891 		if (fp->rf_deny_write == 0)
7892 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
7893 	}
7894 
7895 	/*
7896 	 * If the current mode has access read and the new mode
7897 	 * does not, decrement the number of access read mode bits
7898 	 * and if it goes to zero turn off the access read bit
7899 	 * on the file.  set fflags to FREAD for the call to
7900 	 * vn_open_downgrade().
7901 	 */
7902 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) &&
7903 	    (access & OPEN4_SHARE_ACCESS_READ) == 0) {
7904 		fp->rf_access_read--;
7905 		if (fp->rf_access_read == 0)
7906 			fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
7907 		fflags |= FREAD;
7908 	}
7909 
7910 	/*
7911 	 * If the current mode has access write and the new mode
7912 	 * does not, decrement the number of access write mode bits
7913 	 * and if it goes to zero turn off the access write bit
7914 	 * on the file.  set fflags to FWRITE for the call to
7915 	 * vn_open_downgrade().
7916 	 */
7917 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) &&
7918 	    (access & OPEN4_SHARE_ACCESS_WRITE) == 0) {
7919 		fp->rf_access_write--;
7920 		if (fp->rf_access_write == 0)
7921 			fp->rf_share_deny &= ~OPEN4_SHARE_ACCESS_WRITE;
7922 		fflags |= FWRITE;
7923 	}
7924 
7925 	/* Check that the file is still accessible */
7926 	ASSERT(fp->rf_share_access);
7927 
7928 	rfs4_dbe_unlock(fp->rf_dbe);
7929 
7930 	/* now set the new open access and deny modes */
7931 	sp->rs_open_access = access;
7932 	sp->rs_open_deny = deny;
7933 
7934 	/*
7935 	 * we successfully downgraded the share lock, now we need to downgrade
7936 	 * the open. it is possible that the downgrade was only for a deny
7937 	 * mode and we have nothing else to do.
7938 	 */
7939 	if ((fflags & (FREAD|FWRITE)) != 0)
7940 		vn_open_downgrade(cs->vp, fflags);
7941 
7942 	/* Update the stateid */
7943 	next_stateid(&sp->rs_stateid);
7944 	resp->open_stateid = sp->rs_stateid.stateid;
7945 
7946 	rfs4_dbe_unlock(sp->rs_dbe);
7947 
7948 	*cs->statusp = resp->status = NFS4_OK;
7949 	/* Update the lease */
7950 	rfs4_update_lease(sp->rs_owner->ro_client);
7951 	/* And the sequence */
7952 	rfs4_update_open_sequence(sp->rs_owner);
7953 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7954 
7955 end:
7956 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7957 	rfs4_state_rele(sp);
7958 out:
7959 	DTRACE_NFSV4_2(op__open__downgrade__done, struct compound_state *, cs,
7960 	    OPEN_DOWNGRADE4res *, resp);
7961 }
7962 
7963 static void *
7964 memstr(const void *s1, const char *s2, size_t n)
7965 {
7966 	size_t l = strlen(s2);
7967 	char *p = (char *)s1;
7968 
7969 	while (n >= l) {
7970 		if (bcmp(p, s2, l) == 0)
7971 			return (p);
7972 		p++;
7973 		n--;
7974 	}
7975 
7976 	return (NULL);
7977 }
7978 
7979 /*
7980  * The logic behind this function is detailed in the NFSv4 RFC in the
7981  * SETCLIENTID operation description under IMPLEMENTATION.  Refer to
7982  * that section for explicit guidance to server behavior for
7983  * SETCLIENTID.
7984  */
7985 void
7986 rfs4_op_setclientid(nfs_argop4 *argop, nfs_resop4 *resop,
7987     struct svc_req *req, struct compound_state *cs)
7988 {
7989 	SETCLIENTID4args *args = &argop->nfs_argop4_u.opsetclientid;
7990 	SETCLIENTID4res *res = &resop->nfs_resop4_u.opsetclientid;
7991 	rfs4_client_t *cp, *newcp, *cp_confirmed, *cp_unconfirmed;
7992 	rfs4_clntip_t *ci;
7993 	bool_t create;
7994 	char *addr, *netid;
7995 	int len;
7996 
7997 	DTRACE_NFSV4_2(op__setclientid__start, struct compound_state *, cs,
7998 	    SETCLIENTID4args *, args);
7999 retry:
8000 	newcp = cp_confirmed = cp_unconfirmed = NULL;
8001 
8002 	/*
8003 	 * Save the caller's IP address
8004 	 */
8005 	args->client.cl_addr =
8006 	    (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
8007 
8008 	/*
8009 	 * Record if it is a Solaris client that cannot handle referrals.
8010 	 */
8011 	if (memstr(args->client.id_val, "Solaris", args->client.id_len) &&
8012 	    !memstr(args->client.id_val, "+referrals", args->client.id_len)) {
8013 		/* Add a "yes, it's downrev" record */
8014 		create = TRUE;
8015 		ci = rfs4_find_clntip(args->client.cl_addr, &create);
8016 		ASSERT(ci != NULL);
8017 		rfs4_dbe_rele(ci->ri_dbe);
8018 	} else {
8019 		/* Remove any previous record */
8020 		rfs4_invalidate_clntip(args->client.cl_addr);
8021 	}
8022 
8023 	/*
8024 	 * In search of an EXISTING client matching the incoming
8025 	 * request to establish a new client identifier at the server
8026 	 */
8027 	create = TRUE;
8028 	cp = rfs4_findclient(&args->client, &create, NULL);
8029 
8030 	/* Should never happen */
8031 	ASSERT(cp != NULL);
8032 
8033 	if (cp == NULL) {
8034 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8035 		goto out;
8036 	}
8037 
8038 	/*
8039 	 * Easiest case. Client identifier is newly created and is
8040 	 * unconfirmed.  Also note that for this case, no other
8041 	 * entries exist for the client identifier.  Nothing else to
8042 	 * check.  Just setup the response and respond.
8043 	 */
8044 	if (create) {
8045 		*cs->statusp = res->status = NFS4_OK;
8046 		res->SETCLIENTID4res_u.resok4.clientid = cp->rc_clientid;
8047 		res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8048 		    cp->rc_confirm_verf;
8049 		/* Setup callback information; CB_NULL confirmation later */
8050 		rfs4_client_setcb(cp, &args->callback, args->callback_ident);
8051 
8052 		rfs4_client_rele(cp);
8053 		goto out;
8054 	}
8055 
8056 	/*
8057 	 * An existing, confirmed client may exist but it may not have
8058 	 * been active for at least one lease period.  If so, then
8059 	 * "close" the client and create a new client identifier
8060 	 */
8061 	if (rfs4_lease_expired(cp)) {
8062 		rfs4_client_close(cp);
8063 		goto retry;
8064 	}
8065 
8066 	if (cp->rc_need_confirm == TRUE)
8067 		cp_unconfirmed = cp;
8068 	else
8069 		cp_confirmed = cp;
8070 
8071 	cp = NULL;
8072 
8073 	/*
8074 	 * We have a confirmed client, now check for an
8075 	 * unconfimred entry
8076 	 */
8077 	if (cp_confirmed) {
8078 		/* If creds don't match then client identifier is inuse */
8079 		if (!creds_ok(cp_confirmed->rc_cr_set, req, cs)) {
8080 			rfs4_cbinfo_t *cbp;
8081 			/*
8082 			 * Some one else has established this client
8083 			 * id. Try and say * who they are. We will use
8084 			 * the call back address supplied by * the
8085 			 * first client.
8086 			 */
8087 			*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8088 
8089 			addr = netid = NULL;
8090 
8091 			cbp = &cp_confirmed->rc_cbinfo;
8092 			if (cbp->cb_callback.cb_location.r_addr &&
8093 			    cbp->cb_callback.cb_location.r_netid) {
8094 				cb_client4 *cbcp = &cbp->cb_callback;
8095 
8096 				len = strlen(cbcp->cb_location.r_addr)+1;
8097 				addr = kmem_alloc(len, KM_SLEEP);
8098 				bcopy(cbcp->cb_location.r_addr, addr, len);
8099 				len = strlen(cbcp->cb_location.r_netid)+1;
8100 				netid = kmem_alloc(len, KM_SLEEP);
8101 				bcopy(cbcp->cb_location.r_netid, netid, len);
8102 			}
8103 
8104 			res->SETCLIENTID4res_u.client_using.r_addr = addr;
8105 			res->SETCLIENTID4res_u.client_using.r_netid = netid;
8106 
8107 			rfs4_client_rele(cp_confirmed);
8108 		}
8109 
8110 		/*
8111 		 * Confirmed, creds match, and verifier matches; must
8112 		 * be an update of the callback info
8113 		 */
8114 		if (cp_confirmed->rc_nfs_client.verifier ==
8115 		    args->client.verifier) {
8116 			/* Setup callback information */
8117 			rfs4_client_setcb(cp_confirmed, &args->callback,
8118 			    args->callback_ident);
8119 
8120 			/* everything okay -- move ahead */
8121 			*cs->statusp = res->status = NFS4_OK;
8122 			res->SETCLIENTID4res_u.resok4.clientid =
8123 			    cp_confirmed->rc_clientid;
8124 
8125 			/* update the confirm_verifier and return it */
8126 			rfs4_client_scv_next(cp_confirmed);
8127 			res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8128 			    cp_confirmed->rc_confirm_verf;
8129 
8130 			rfs4_client_rele(cp_confirmed);
8131 			goto out;
8132 		}
8133 
8134 		/*
8135 		 * Creds match but the verifier doesn't.  Must search
8136 		 * for an unconfirmed client that would be replaced by
8137 		 * this request.
8138 		 */
8139 		create = FALSE;
8140 		cp_unconfirmed = rfs4_findclient(&args->client, &create,
8141 		    cp_confirmed);
8142 	}
8143 
8144 	/*
8145 	 * At this point, we have taken care of the brand new client
8146 	 * struct, INUSE case, update of an existing, and confirmed
8147 	 * client struct.
8148 	 */
8149 
8150 	/*
8151 	 * check to see if things have changed while we originally
8152 	 * picked up the client struct.  If they have, then return and
8153 	 * retry the processing of this SETCLIENTID request.
8154 	 */
8155 	if (cp_unconfirmed) {
8156 		rfs4_dbe_lock(cp_unconfirmed->rc_dbe);
8157 		if (!cp_unconfirmed->rc_need_confirm) {
8158 			rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8159 			rfs4_client_rele(cp_unconfirmed);
8160 			if (cp_confirmed)
8161 				rfs4_client_rele(cp_confirmed);
8162 			goto retry;
8163 		}
8164 		/* do away with the old unconfirmed one */
8165 		rfs4_dbe_invalidate(cp_unconfirmed->rc_dbe);
8166 		rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8167 		rfs4_client_rele(cp_unconfirmed);
8168 		cp_unconfirmed = NULL;
8169 	}
8170 
8171 	/*
8172 	 * This search will temporarily hide the confirmed client
8173 	 * struct while a new client struct is created as the
8174 	 * unconfirmed one.
8175 	 */
8176 	create = TRUE;
8177 	newcp = rfs4_findclient(&args->client, &create, cp_confirmed);
8178 
8179 	ASSERT(newcp != NULL);
8180 
8181 	if (newcp == NULL) {
8182 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8183 		rfs4_client_rele(cp_confirmed);
8184 		goto out;
8185 	}
8186 
8187 	/*
8188 	 * If one was not created, then a similar request must be in
8189 	 * process so release and start over with this one
8190 	 */
8191 	if (create != TRUE) {
8192 		rfs4_client_rele(newcp);
8193 		if (cp_confirmed)
8194 			rfs4_client_rele(cp_confirmed);
8195 		goto retry;
8196 	}
8197 
8198 	*cs->statusp = res->status = NFS4_OK;
8199 	res->SETCLIENTID4res_u.resok4.clientid = newcp->rc_clientid;
8200 	res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8201 	    newcp->rc_confirm_verf;
8202 	/* Setup callback information; CB_NULL confirmation later */
8203 	rfs4_client_setcb(newcp, &args->callback, args->callback_ident);
8204 
8205 	newcp->rc_cp_confirmed = cp_confirmed;
8206 
8207 	rfs4_client_rele(newcp);
8208 
8209 out:
8210 	DTRACE_NFSV4_2(op__setclientid__done, struct compound_state *, cs,
8211 	    SETCLIENTID4res *, res);
8212 }
8213 
8214 /*ARGSUSED*/
8215 void
8216 rfs4_op_setclientid_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
8217     struct svc_req *req, struct compound_state *cs)
8218 {
8219 	SETCLIENTID_CONFIRM4args *args =
8220 	    &argop->nfs_argop4_u.opsetclientid_confirm;
8221 	SETCLIENTID_CONFIRM4res *res =
8222 	    &resop->nfs_resop4_u.opsetclientid_confirm;
8223 	rfs4_client_t *cp, *cptoclose = NULL;
8224 
8225 	DTRACE_NFSV4_2(op__setclientid__confirm__start,
8226 	    struct compound_state *, cs,
8227 	    SETCLIENTID_CONFIRM4args *, args);
8228 
8229 	*cs->statusp = res->status = NFS4_OK;
8230 
8231 	cp = rfs4_findclient_by_id(args->clientid, TRUE);
8232 
8233 	if (cp == NULL) {
8234 		*cs->statusp = res->status =
8235 		    rfs4_check_clientid(&args->clientid, 1);
8236 		goto out;
8237 	}
8238 
8239 	if (!creds_ok(cp, req, cs)) {
8240 		*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8241 		rfs4_client_rele(cp);
8242 		goto out;
8243 	}
8244 
8245 	/* If the verifier doesn't match, the record doesn't match */
8246 	if (cp->rc_confirm_verf != args->setclientid_confirm) {
8247 		*cs->statusp = res->status = NFS4ERR_STALE_CLIENTID;
8248 		rfs4_client_rele(cp);
8249 		goto out;
8250 	}
8251 
8252 	rfs4_dbe_lock(cp->rc_dbe);
8253 	cp->rc_need_confirm = FALSE;
8254 	if (cp->rc_cp_confirmed) {
8255 		cptoclose = cp->rc_cp_confirmed;
8256 		cptoclose->rc_ss_remove = 1;
8257 		cp->rc_cp_confirmed = NULL;
8258 	}
8259 
8260 	/*
8261 	 * Update the client's associated server instance, if it's changed
8262 	 * since the client was created.
8263 	 */
8264 	if (rfs4_servinst(cp) != rfs4_cur_servinst)
8265 		rfs4_servinst_assign(cp, rfs4_cur_servinst);
8266 
8267 	/*
8268 	 * Record clientid in stable storage.
8269 	 * Must be done after server instance has been assigned.
8270 	 */
8271 	rfs4_ss_clid(cp);
8272 
8273 	rfs4_dbe_unlock(cp->rc_dbe);
8274 
8275 	if (cptoclose)
8276 		/* don't need to rele, client_close does it */
8277 		rfs4_client_close(cptoclose);
8278 
8279 	/* If needed, initiate CB_NULL call for callback path */
8280 	rfs4_deleg_cb_check(cp);
8281 	rfs4_update_lease(cp);
8282 
8283 	/*
8284 	 * Check to see if client can perform reclaims
8285 	 */
8286 	rfs4_ss_chkclid(cp);
8287 
8288 	rfs4_client_rele(cp);
8289 
8290 out:
8291 	DTRACE_NFSV4_2(op__setclientid__confirm__done,
8292 	    struct compound_state *, cs,
8293 	    SETCLIENTID_CONFIRM4 *, res);
8294 }
8295 
8296 
8297 /*ARGSUSED*/
8298 void
8299 rfs4_op_close(nfs_argop4 *argop, nfs_resop4 *resop,
8300     struct svc_req *req, struct compound_state *cs)
8301 {
8302 	CLOSE4args *args = &argop->nfs_argop4_u.opclose;
8303 	CLOSE4res *resp = &resop->nfs_resop4_u.opclose;
8304 	rfs4_state_t *sp;
8305 	nfsstat4 status;
8306 
8307 	DTRACE_NFSV4_2(op__close__start, struct compound_state *, cs,
8308 	    CLOSE4args *, args);
8309 
8310 	if (cs->vp == NULL) {
8311 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8312 		goto out;
8313 	}
8314 
8315 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_INVALID);
8316 	if (status != NFS4_OK) {
8317 		*cs->statusp = resp->status = status;
8318 		goto out;
8319 	}
8320 
8321 	/* Ensure specified filehandle matches */
8322 	if (cs->vp != sp->rs_finfo->rf_vp) {
8323 		rfs4_state_rele(sp);
8324 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8325 		goto out;
8326 	}
8327 
8328 	/* hold off other access to open_owner while we tinker */
8329 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
8330 
8331 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
8332 	case NFS4_CHECK_STATEID_OKAY:
8333 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8334 		    resop) != NFS4_CHKSEQ_OKAY) {
8335 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8336 			goto end;
8337 		}
8338 		break;
8339 	case NFS4_CHECK_STATEID_OLD:
8340 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8341 		goto end;
8342 	case NFS4_CHECK_STATEID_BAD:
8343 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8344 		goto end;
8345 	case NFS4_CHECK_STATEID_EXPIRED:
8346 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8347 		goto end;
8348 	case NFS4_CHECK_STATEID_CLOSED:
8349 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8350 		goto end;
8351 	case NFS4_CHECK_STATEID_UNCONFIRMED:
8352 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8353 		goto end;
8354 	case NFS4_CHECK_STATEID_REPLAY:
8355 		/* Check the sequence id for the open owner */
8356 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8357 		    resop)) {
8358 		case NFS4_CHKSEQ_OKAY:
8359 			/*
8360 			 * This is replayed stateid; if seqid matches
8361 			 * next expected, then client is using wrong seqid.
8362 			 */
8363 			/* FALL THROUGH */
8364 		case NFS4_CHKSEQ_BAD:
8365 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8366 			goto end;
8367 		case NFS4_CHKSEQ_REPLAY:
8368 			/*
8369 			 * Note this case is the duplicate case so
8370 			 * resp->status is already set.
8371 			 */
8372 			*cs->statusp = resp->status;
8373 			rfs4_update_lease(sp->rs_owner->ro_client);
8374 			goto end;
8375 		}
8376 		break;
8377 	default:
8378 		ASSERT(FALSE);
8379 		break;
8380 	}
8381 
8382 	rfs4_dbe_lock(sp->rs_dbe);
8383 
8384 	/* Update the stateid. */
8385 	next_stateid(&sp->rs_stateid);
8386 	resp->open_stateid = sp->rs_stateid.stateid;
8387 
8388 	rfs4_dbe_unlock(sp->rs_dbe);
8389 
8390 	rfs4_update_lease(sp->rs_owner->ro_client);
8391 	rfs4_update_open_sequence(sp->rs_owner);
8392 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8393 
8394 	rfs4_state_close(sp, FALSE, FALSE, cs->cr);
8395 
8396 	*cs->statusp = resp->status = status;
8397 
8398 end:
8399 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
8400 	rfs4_state_rele(sp);
8401 out:
8402 	DTRACE_NFSV4_2(op__close__done, struct compound_state *, cs,
8403 	    CLOSE4res *, resp);
8404 }
8405 
8406 /*
8407  * Manage the counts on the file struct and close all file locks
8408  */
8409 /*ARGSUSED*/
8410 void
8411 rfs4_release_share_lock_state(rfs4_state_t *sp, cred_t *cr,
8412     bool_t close_of_client)
8413 {
8414 	rfs4_file_t *fp = sp->rs_finfo;
8415 	rfs4_lo_state_t *lsp;
8416 	int fflags = 0;
8417 
8418 	/*
8419 	 * If this call is part of the larger closing down of client
8420 	 * state then it is just easier to release all locks
8421 	 * associated with this client instead of going through each
8422 	 * individual file and cleaning locks there.
8423 	 */
8424 	if (close_of_client) {
8425 		if (sp->rs_owner->ro_client->rc_unlksys_completed == FALSE &&
8426 		    !list_is_empty(&sp->rs_lostatelist) &&
8427 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID) {
8428 			/* Is the PxFS kernel module loaded? */
8429 			if (lm_remove_file_locks != NULL) {
8430 				int new_sysid;
8431 
8432 				/* Encode the cluster nodeid in new sysid */
8433 				new_sysid = sp->rs_owner->ro_client->rc_sysidt;
8434 				lm_set_nlmid_flk(&new_sysid);
8435 
8436 				/*
8437 				 * This PxFS routine removes file locks for a
8438 				 * client over all nodes of a cluster.
8439 				 */
8440 				NFS4_DEBUG(rfs4_debug, (CE_NOTE,
8441 				    "lm_remove_file_locks(sysid=0x%x)\n",
8442 				    new_sysid));
8443 				(*lm_remove_file_locks)(new_sysid);
8444 			} else {
8445 				struct flock64 flk;
8446 
8447 				/* Release all locks for this client */
8448 				flk.l_type = F_UNLKSYS;
8449 				flk.l_whence = 0;
8450 				flk.l_start = 0;
8451 				flk.l_len = 0;
8452 				flk.l_sysid =
8453 				    sp->rs_owner->ro_client->rc_sysidt;
8454 				flk.l_pid = 0;
8455 				(void) VOP_FRLOCK(sp->rs_finfo->rf_vp, F_SETLK,
8456 				    &flk, F_REMOTELOCK | FREAD | FWRITE,
8457 				    (u_offset_t)0, NULL, CRED(), NULL);
8458 			}
8459 
8460 			sp->rs_owner->ro_client->rc_unlksys_completed = TRUE;
8461 		}
8462 	}
8463 
8464 	/*
8465 	 * Release all locks on this file by this lock owner or at
8466 	 * least mark the locks as having been released
8467 	 */
8468 	for (lsp = list_head(&sp->rs_lostatelist); lsp != NULL;
8469 	    lsp = list_next(&sp->rs_lostatelist, lsp)) {
8470 		lsp->rls_locks_cleaned = TRUE;
8471 
8472 		/* Was this already taken care of above? */
8473 		if (!close_of_client &&
8474 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8475 			(void) cleanlocks(sp->rs_finfo->rf_vp,
8476 			    lsp->rls_locker->rl_pid,
8477 			    lsp->rls_locker->rl_client->rc_sysidt);
8478 	}
8479 
8480 	/*
8481 	 * Release any shrlocks associated with this open state ID.
8482 	 * This must be done before the rfs4_state gets marked closed.
8483 	 */
8484 	if (sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8485 		(void) rfs4_unshare(sp);
8486 
8487 	if (sp->rs_open_access) {
8488 		rfs4_dbe_lock(fp->rf_dbe);
8489 
8490 		/*
8491 		 * Decrement the count for each access and deny bit that this
8492 		 * state has contributed to the file.
8493 		 * If the file counts go to zero
8494 		 * clear the appropriate bit in the appropriate mask.
8495 		 */
8496 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) {
8497 			fp->rf_access_read--;
8498 			fflags |= FREAD;
8499 			if (fp->rf_access_read == 0)
8500 				fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
8501 		}
8502 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) {
8503 			fp->rf_access_write--;
8504 			fflags |= FWRITE;
8505 			if (fp->rf_access_write == 0)
8506 				fp->rf_share_access &=
8507 				    ~OPEN4_SHARE_ACCESS_WRITE;
8508 		}
8509 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_READ) {
8510 			fp->rf_deny_read--;
8511 			if (fp->rf_deny_read == 0)
8512 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
8513 		}
8514 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) {
8515 			fp->rf_deny_write--;
8516 			if (fp->rf_deny_write == 0)
8517 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
8518 		}
8519 
8520 		(void) VOP_CLOSE(fp->rf_vp, fflags, 1, (offset_t)0, cr, NULL);
8521 
8522 		rfs4_dbe_unlock(fp->rf_dbe);
8523 
8524 		sp->rs_open_access = 0;
8525 		sp->rs_open_deny = 0;
8526 	}
8527 }
8528 
8529 /*
8530  * lock_denied: Fill in a LOCK4deneid structure given an flock64 structure.
8531  */
8532 static nfsstat4
8533 lock_denied(LOCK4denied *dp, struct flock64 *flk)
8534 {
8535 	rfs4_lockowner_t *lo;
8536 	rfs4_client_t *cp;
8537 	uint32_t len;
8538 
8539 	lo = rfs4_findlockowner_by_pid(flk->l_pid);
8540 	if (lo != NULL) {
8541 		cp = lo->rl_client;
8542 		if (rfs4_lease_expired(cp)) {
8543 			rfs4_lockowner_rele(lo);
8544 			rfs4_dbe_hold(cp->rc_dbe);
8545 			rfs4_client_close(cp);
8546 			return (NFS4ERR_EXPIRED);
8547 		}
8548 		dp->owner.clientid = lo->rl_owner.clientid;
8549 		len = lo->rl_owner.owner_len;
8550 		dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8551 		bcopy(lo->rl_owner.owner_val, dp->owner.owner_val, len);
8552 		dp->owner.owner_len = len;
8553 		rfs4_lockowner_rele(lo);
8554 		goto finish;
8555 	}
8556 
8557 	/*
8558 	 * Its not a NFS4 lock. We take advantage that the upper 32 bits
8559 	 * of the client id contain the boot time for a NFS4 lock. So we
8560 	 * fabricate and identity by setting clientid to the sysid, and
8561 	 * the lock owner to the pid.
8562 	 */
8563 	dp->owner.clientid = flk->l_sysid;
8564 	len = sizeof (pid_t);
8565 	dp->owner.owner_len = len;
8566 	dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8567 	bcopy(&flk->l_pid, dp->owner.owner_val, len);
8568 finish:
8569 	dp->offset = flk->l_start;
8570 	dp->length = flk->l_len;
8571 
8572 	if (flk->l_type == F_RDLCK)
8573 		dp->locktype = READ_LT;
8574 	else if (flk->l_type == F_WRLCK)
8575 		dp->locktype = WRITE_LT;
8576 	else
8577 		return (NFS4ERR_INVAL);	/* no mapping from POSIX ltype to v4 */
8578 
8579 	return (NFS4_OK);
8580 }
8581 
8582 static int
8583 setlock(vnode_t *vp, struct flock64 *flock, int flag, cred_t *cred)
8584 {
8585 	int error;
8586 	struct flock64 flk;
8587 	int i;
8588 	clock_t delaytime;
8589 	int cmd;
8590 
8591 	cmd = nbl_need_check(vp) ? F_SETLK_NBMAND : F_SETLK;
8592 retry:
8593 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
8594 
8595 	for (i = 0; i < rfs4_maxlock_tries; i++) {
8596 		LOCK_PRINT(rfs4_debug, "setlock", cmd, flock);
8597 		error = VOP_FRLOCK(vp, cmd,
8598 		    flock, flag, (u_offset_t)0, NULL, cred, NULL);
8599 
8600 		if (error != EAGAIN && error != EACCES)
8601 			break;
8602 
8603 		if (i < rfs4_maxlock_tries - 1) {
8604 			delay(delaytime);
8605 			delaytime *= 2;
8606 		}
8607 	}
8608 
8609 	if (error == EAGAIN || error == EACCES) {
8610 		/* Get the owner of the lock */
8611 		flk = *flock;
8612 		LOCK_PRINT(rfs4_debug, "setlock", F_GETLK, &flk);
8613 		if (VOP_FRLOCK(vp, F_GETLK, &flk, flag,
8614 		    (u_offset_t)0, NULL, cred, NULL) == 0) {
8615 			if (flk.l_type == F_UNLCK) {
8616 				/* No longer locked, retry */
8617 				goto retry;
8618 			}
8619 			*flock = flk;
8620 			LOCK_PRINT(rfs4_debug, "setlock(blocking lock)",
8621 			    F_GETLK, &flk);
8622 		}
8623 	}
8624 
8625 	return (error);
8626 }
8627 
8628 /*ARGSUSED*/
8629 static nfsstat4
8630 rfs4_do_lock(rfs4_lo_state_t *lsp, nfs_lock_type4 locktype,
8631     offset4 offset, length4 length, cred_t *cred, nfs_resop4 *resop)
8632 {
8633 	nfsstat4 status;
8634 	rfs4_lockowner_t *lo = lsp->rls_locker;
8635 	rfs4_state_t *sp = lsp->rls_state;
8636 	struct flock64 flock;
8637 	int16_t ltype;
8638 	int flag;
8639 	int error;
8640 	sysid_t sysid;
8641 	LOCK4res *lres;
8642 
8643 	if (rfs4_lease_expired(lo->rl_client)) {
8644 		return (NFS4ERR_EXPIRED);
8645 	}
8646 
8647 	if ((status = rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
8648 		return (status);
8649 
8650 	/* Check for zero length. To lock to end of file use all ones for V4 */
8651 	if (length == 0)
8652 		return (NFS4ERR_INVAL);
8653 	else if (length == (length4)(~0))
8654 		length = 0;		/* Posix to end of file  */
8655 
8656 retry:
8657 	rfs4_dbe_lock(sp->rs_dbe);
8658 	if (sp->rs_closed) {
8659 		rfs4_dbe_unlock(sp->rs_dbe);
8660 		return (NFS4ERR_OLD_STATEID);
8661 	}
8662 
8663 	if (resop->resop != OP_LOCKU) {
8664 		switch (locktype) {
8665 		case READ_LT:
8666 		case READW_LT:
8667 			if ((sp->rs_share_access
8668 			    & OPEN4_SHARE_ACCESS_READ) == 0) {
8669 				rfs4_dbe_unlock(sp->rs_dbe);
8670 
8671 				return (NFS4ERR_OPENMODE);
8672 			}
8673 			ltype = F_RDLCK;
8674 			break;
8675 		case WRITE_LT:
8676 		case WRITEW_LT:
8677 			if ((sp->rs_share_access
8678 			    & OPEN4_SHARE_ACCESS_WRITE) == 0) {
8679 				rfs4_dbe_unlock(sp->rs_dbe);
8680 
8681 				return (NFS4ERR_OPENMODE);
8682 			}
8683 			ltype = F_WRLCK;
8684 			break;
8685 		}
8686 	} else
8687 		ltype = F_UNLCK;
8688 
8689 	flock.l_type = ltype;
8690 	flock.l_whence = 0;		/* SEEK_SET */
8691 	flock.l_start = offset;
8692 	flock.l_len = length;
8693 	flock.l_sysid = sysid;
8694 	flock.l_pid = lsp->rls_locker->rl_pid;
8695 
8696 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
8697 	if (flock.l_len < 0 || flock.l_start < 0) {
8698 		rfs4_dbe_unlock(sp->rs_dbe);
8699 		return (NFS4ERR_INVAL);
8700 	}
8701 
8702 	/*
8703 	 * N.B. FREAD has the same value as OPEN4_SHARE_ACCESS_READ and
8704 	 * FWRITE has the same value as OPEN4_SHARE_ACCESS_WRITE.
8705 	 */
8706 	flag = (int)sp->rs_share_access | F_REMOTELOCK;
8707 
8708 	error = setlock(sp->rs_finfo->rf_vp, &flock, flag, cred);
8709 	if (error == 0) {
8710 		rfs4_dbe_lock(lsp->rls_dbe);
8711 		next_stateid(&lsp->rls_lockid);
8712 		rfs4_dbe_unlock(lsp->rls_dbe);
8713 	}
8714 
8715 	rfs4_dbe_unlock(sp->rs_dbe);
8716 
8717 	/*
8718 	 * N.B. We map error values to nfsv4 errors. This is differrent
8719 	 * than puterrno4 routine.
8720 	 */
8721 	switch (error) {
8722 	case 0:
8723 		status = NFS4_OK;
8724 		break;
8725 	case EAGAIN:
8726 	case EACCES:		/* Old value */
8727 		/* Can only get here if op is OP_LOCK */
8728 		ASSERT(resop->resop == OP_LOCK);
8729 		lres = &resop->nfs_resop4_u.oplock;
8730 		status = NFS4ERR_DENIED;
8731 		if (lock_denied(&lres->LOCK4res_u.denied, &flock)
8732 		    == NFS4ERR_EXPIRED)
8733 			goto retry;
8734 		break;
8735 	case ENOLCK:
8736 		status = NFS4ERR_DELAY;
8737 		break;
8738 	case EOVERFLOW:
8739 		status = NFS4ERR_INVAL;
8740 		break;
8741 	case EINVAL:
8742 		status = NFS4ERR_NOTSUPP;
8743 		break;
8744 	default:
8745 		status = NFS4ERR_SERVERFAULT;
8746 		break;
8747 	}
8748 
8749 	return (status);
8750 }
8751 
8752 /*ARGSUSED*/
8753 void
8754 rfs4_op_lock(nfs_argop4 *argop, nfs_resop4 *resop,
8755     struct svc_req *req, struct compound_state *cs)
8756 {
8757 	LOCK4args *args = &argop->nfs_argop4_u.oplock;
8758 	LOCK4res *resp = &resop->nfs_resop4_u.oplock;
8759 	nfsstat4 status;
8760 	stateid4 *stateid;
8761 	rfs4_lockowner_t *lo;
8762 	rfs4_client_t *cp;
8763 	rfs4_state_t *sp = NULL;
8764 	rfs4_lo_state_t *lsp = NULL;
8765 	bool_t ls_sw_held = FALSE;
8766 	bool_t create = TRUE;
8767 	bool_t lcreate = TRUE;
8768 	bool_t dup_lock = FALSE;
8769 	int rc;
8770 
8771 	DTRACE_NFSV4_2(op__lock__start, struct compound_state *, cs,
8772 	    LOCK4args *, args);
8773 
8774 	if (cs->vp == NULL) {
8775 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8776 		DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8777 		    cs, LOCK4res *, resp);
8778 		return;
8779 	}
8780 
8781 	if (args->locker.new_lock_owner) {
8782 		/* Create a new lockowner for this instance */
8783 		open_to_lock_owner4 *olo = &args->locker.locker4_u.open_owner;
8784 
8785 		NFS4_DEBUG(rfs4_debug, (CE_NOTE, "Creating new lock owner"));
8786 
8787 		stateid = &olo->open_stateid;
8788 		status = rfs4_get_state(stateid, &sp, RFS4_DBS_VALID);
8789 		if (status != NFS4_OK) {
8790 			NFS4_DEBUG(rfs4_debug,
8791 			    (CE_NOTE, "Get state failed in lock %d", status));
8792 			*cs->statusp = resp->status = status;
8793 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8794 			    cs, LOCK4res *, resp);
8795 			return;
8796 		}
8797 
8798 		/* Ensure specified filehandle matches */
8799 		if (cs->vp != sp->rs_finfo->rf_vp) {
8800 			rfs4_state_rele(sp);
8801 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8802 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8803 			    cs, LOCK4res *, resp);
8804 			return;
8805 		}
8806 
8807 		/* hold off other access to open_owner while we tinker */
8808 		rfs4_sw_enter(&sp->rs_owner->ro_sw);
8809 
8810 		switch (rc = rfs4_check_stateid_seqid(sp, stateid)) {
8811 		case NFS4_CHECK_STATEID_OLD:
8812 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8813 			goto end;
8814 		case NFS4_CHECK_STATEID_BAD:
8815 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8816 			goto end;
8817 		case NFS4_CHECK_STATEID_EXPIRED:
8818 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8819 			goto end;
8820 		case NFS4_CHECK_STATEID_UNCONFIRMED:
8821 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8822 			goto end;
8823 		case NFS4_CHECK_STATEID_CLOSED:
8824 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8825 			goto end;
8826 		case NFS4_CHECK_STATEID_OKAY:
8827 		case NFS4_CHECK_STATEID_REPLAY:
8828 			switch (rfs4_check_olo_seqid(olo->open_seqid,
8829 			    sp->rs_owner, resop)) {
8830 			case NFS4_CHKSEQ_OKAY:
8831 				if (rc == NFS4_CHECK_STATEID_OKAY)
8832 					break;
8833 				/*
8834 				 * This is replayed stateid; if seqid
8835 				 * matches next expected, then client
8836 				 * is using wrong seqid.
8837 				 */
8838 				/* FALLTHROUGH */
8839 			case NFS4_CHKSEQ_BAD:
8840 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8841 				goto end;
8842 			case NFS4_CHKSEQ_REPLAY:
8843 				/* This is a duplicate LOCK request */
8844 				dup_lock = TRUE;
8845 
8846 				/*
8847 				 * For a duplicate we do not want to
8848 				 * create a new lockowner as it should
8849 				 * already exist.
8850 				 * Turn off the lockowner create flag.
8851 				 */
8852 				lcreate = FALSE;
8853 			}
8854 			break;
8855 		}
8856 
8857 		lo = rfs4_findlockowner(&olo->lock_owner, &lcreate);
8858 		if (lo == NULL) {
8859 			NFS4_DEBUG(rfs4_debug,
8860 			    (CE_NOTE, "rfs4_op_lock: no lock owner"));
8861 			*cs->statusp = resp->status = NFS4ERR_RESOURCE;
8862 			goto end;
8863 		}
8864 
8865 		lsp = rfs4_findlo_state_by_owner(lo, sp, &create);
8866 		if (lsp == NULL) {
8867 			rfs4_update_lease(sp->rs_owner->ro_client);
8868 			/*
8869 			 * Only update theh open_seqid if this is not
8870 			 * a duplicate request
8871 			 */
8872 			if (dup_lock == FALSE) {
8873 				rfs4_update_open_sequence(sp->rs_owner);
8874 			}
8875 
8876 			NFS4_DEBUG(rfs4_debug,
8877 			    (CE_NOTE, "rfs4_op_lock: no state"));
8878 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
8879 			rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8880 			rfs4_lockowner_rele(lo);
8881 			goto end;
8882 		}
8883 
8884 		/*
8885 		 * This is the new_lock_owner branch and the client is
8886 		 * supposed to be associating a new lock_owner with
8887 		 * the open file at this point.  If we find that a
8888 		 * lock_owner/state association already exists and a
8889 		 * successful LOCK request was returned to the client,
8890 		 * an error is returned to the client since this is
8891 		 * not appropriate.  The client should be using the
8892 		 * existing lock_owner branch.
8893 		 */
8894 		if (dup_lock == FALSE && create == FALSE) {
8895 			if (lsp->rls_lock_completed == TRUE) {
8896 				*cs->statusp =
8897 				    resp->status = NFS4ERR_BAD_SEQID;
8898 				rfs4_lockowner_rele(lo);
8899 				goto end;
8900 			}
8901 		}
8902 
8903 		rfs4_update_lease(sp->rs_owner->ro_client);
8904 
8905 		/*
8906 		 * Only update theh open_seqid if this is not
8907 		 * a duplicate request
8908 		 */
8909 		if (dup_lock == FALSE) {
8910 			rfs4_update_open_sequence(sp->rs_owner);
8911 		}
8912 
8913 		/*
8914 		 * If this is a duplicate lock request, just copy the
8915 		 * previously saved reply and return.
8916 		 */
8917 		if (dup_lock == TRUE) {
8918 			/* verify that lock_seqid's match */
8919 			if (lsp->rls_seqid != olo->lock_seqid) {
8920 				NFS4_DEBUG(rfs4_debug,
8921 				    (CE_NOTE, "rfs4_op_lock: Dup-Lock seqid bad"
8922 				    "lsp->seqid=%d old->seqid=%d",
8923 				    lsp->rls_seqid, olo->lock_seqid));
8924 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8925 			} else {
8926 				rfs4_copy_reply(resop, &lsp->rls_reply);
8927 				/*
8928 				 * Make sure to copy the just
8929 				 * retrieved reply status into the
8930 				 * overall compound status
8931 				 */
8932 				*cs->statusp = resp->status;
8933 			}
8934 			rfs4_lockowner_rele(lo);
8935 			goto end;
8936 		}
8937 
8938 		rfs4_dbe_lock(lsp->rls_dbe);
8939 
8940 		/* Make sure to update the lock sequence id */
8941 		lsp->rls_seqid = olo->lock_seqid;
8942 
8943 		NFS4_DEBUG(rfs4_debug,
8944 		    (CE_NOTE, "Lock seqid established as %d", lsp->rls_seqid));
8945 
8946 		/*
8947 		 * This is used to signify the newly created lockowner
8948 		 * stateid and its sequence number.  The checks for
8949 		 * sequence number and increment don't occur on the
8950 		 * very first lock request for a lockowner.
8951 		 */
8952 		lsp->rls_skip_seqid_check = TRUE;
8953 
8954 		/* hold off other access to lsp while we tinker */
8955 		rfs4_sw_enter(&lsp->rls_sw);
8956 		ls_sw_held = TRUE;
8957 
8958 		rfs4_dbe_unlock(lsp->rls_dbe);
8959 
8960 		rfs4_lockowner_rele(lo);
8961 	} else {
8962 		stateid = &args->locker.locker4_u.lock_owner.lock_stateid;
8963 		/* get lsp and hold the lock on the underlying file struct */
8964 		if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE))
8965 		    != NFS4_OK) {
8966 			*cs->statusp = resp->status = status;
8967 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8968 			    cs, LOCK4res *, resp);
8969 			return;
8970 		}
8971 		create = FALSE;	/* We didn't create lsp */
8972 
8973 		/* Ensure specified filehandle matches */
8974 		if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
8975 			rfs4_lo_state_rele(lsp, TRUE);
8976 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8977 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8978 			    cs, LOCK4res *, resp);
8979 			return;
8980 		}
8981 
8982 		/* hold off other access to lsp while we tinker */
8983 		rfs4_sw_enter(&lsp->rls_sw);
8984 		ls_sw_held = TRUE;
8985 
8986 		switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
8987 		/*
8988 		 * The stateid looks like it was okay (expected to be
8989 		 * the next one)
8990 		 */
8991 		case NFS4_CHECK_STATEID_OKAY:
8992 			/*
8993 			 * The sequence id is now checked.  Determine
8994 			 * if this is a replay or if it is in the
8995 			 * expected (next) sequence.  In the case of a
8996 			 * replay, there are two replay conditions
8997 			 * that may occur.  The first is the normal
8998 			 * condition where a LOCK is done with a
8999 			 * NFS4_OK response and the stateid is
9000 			 * updated.  That case is handled below when
9001 			 * the stateid is identified as a REPLAY.  The
9002 			 * second is the case where an error is
9003 			 * returned, like NFS4ERR_DENIED, and the
9004 			 * sequence number is updated but the stateid
9005 			 * is not updated.  This second case is dealt
9006 			 * with here.  So it may seem odd that the
9007 			 * stateid is okay but the sequence id is a
9008 			 * replay but it is okay.
9009 			 */
9010 			switch (rfs4_check_lock_seqid(
9011 			    args->locker.locker4_u.lock_owner.lock_seqid,
9012 			    lsp, resop)) {
9013 			case NFS4_CHKSEQ_REPLAY:
9014 				if (resp->status != NFS4_OK) {
9015 					/*
9016 					 * Here is our replay and need
9017 					 * to verify that the last
9018 					 * response was an error.
9019 					 */
9020 					*cs->statusp = resp->status;
9021 					goto end;
9022 				}
9023 				/*
9024 				 * This is done since the sequence id
9025 				 * looked like a replay but it didn't
9026 				 * pass our check so a BAD_SEQID is
9027 				 * returned as a result.
9028 				 */
9029 				/*FALLTHROUGH*/
9030 			case NFS4_CHKSEQ_BAD:
9031 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9032 				goto end;
9033 			case NFS4_CHKSEQ_OKAY:
9034 				/* Everything looks okay move ahead */
9035 				break;
9036 			}
9037 			break;
9038 		case NFS4_CHECK_STATEID_OLD:
9039 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9040 			goto end;
9041 		case NFS4_CHECK_STATEID_BAD:
9042 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9043 			goto end;
9044 		case NFS4_CHECK_STATEID_EXPIRED:
9045 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9046 			goto end;
9047 		case NFS4_CHECK_STATEID_CLOSED:
9048 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9049 			goto end;
9050 		case NFS4_CHECK_STATEID_REPLAY:
9051 			switch (rfs4_check_lock_seqid(
9052 			    args->locker.locker4_u.lock_owner.lock_seqid,
9053 			    lsp, resop)) {
9054 			case NFS4_CHKSEQ_OKAY:
9055 				/*
9056 				 * This is a replayed stateid; if
9057 				 * seqid matches the next expected,
9058 				 * then client is using wrong seqid.
9059 				 */
9060 			case NFS4_CHKSEQ_BAD:
9061 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9062 				goto end;
9063 			case NFS4_CHKSEQ_REPLAY:
9064 				rfs4_update_lease(lsp->rls_locker->rl_client);
9065 				*cs->statusp = status = resp->status;
9066 				goto end;
9067 			}
9068 			break;
9069 		default:
9070 			ASSERT(FALSE);
9071 			break;
9072 		}
9073 
9074 		rfs4_update_lock_sequence(lsp);
9075 		rfs4_update_lease(lsp->rls_locker->rl_client);
9076 	}
9077 
9078 	/*
9079 	 * NFS4 only allows locking on regular files, so
9080 	 * verify type of object.
9081 	 */
9082 	if (cs->vp->v_type != VREG) {
9083 		if (cs->vp->v_type == VDIR)
9084 			status = NFS4ERR_ISDIR;
9085 		else
9086 			status = NFS4ERR_INVAL;
9087 		goto out;
9088 	}
9089 
9090 	cp = lsp->rls_state->rs_owner->ro_client;
9091 
9092 	if (rfs4_clnt_in_grace(cp) && !args->reclaim) {
9093 		status = NFS4ERR_GRACE;
9094 		goto out;
9095 	}
9096 
9097 	if (rfs4_clnt_in_grace(cp) && args->reclaim && !cp->rc_can_reclaim) {
9098 		status = NFS4ERR_NO_GRACE;
9099 		goto out;
9100 	}
9101 
9102 	if (!rfs4_clnt_in_grace(cp) && args->reclaim) {
9103 		status = NFS4ERR_NO_GRACE;
9104 		goto out;
9105 	}
9106 
9107 	if (lsp->rls_state->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE)
9108 		cs->deleg = TRUE;
9109 
9110 	status = rfs4_do_lock(lsp, args->locktype,
9111 	    args->offset, args->length, cs->cr, resop);
9112 
9113 out:
9114 	lsp->rls_skip_seqid_check = FALSE;
9115 
9116 	*cs->statusp = resp->status = status;
9117 
9118 	if (status == NFS4_OK) {
9119 		resp->LOCK4res_u.lock_stateid = lsp->rls_lockid.stateid;
9120 		lsp->rls_lock_completed = TRUE;
9121 	}
9122 	/*
9123 	 * Only update the "OPEN" response here if this was a new
9124 	 * lock_owner
9125 	 */
9126 	if (sp)
9127 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
9128 
9129 	rfs4_update_lock_resp(lsp, resop);
9130 
9131 end:
9132 	if (lsp) {
9133 		if (ls_sw_held)
9134 			rfs4_sw_exit(&lsp->rls_sw);
9135 		/*
9136 		 * If an sp obtained, then the lsp does not represent
9137 		 * a lock on the file struct.
9138 		 */
9139 		if (sp != NULL)
9140 			rfs4_lo_state_rele(lsp, FALSE);
9141 		else
9142 			rfs4_lo_state_rele(lsp, TRUE);
9143 	}
9144 	if (sp) {
9145 		rfs4_sw_exit(&sp->rs_owner->ro_sw);
9146 		rfs4_state_rele(sp);
9147 	}
9148 
9149 	DTRACE_NFSV4_2(op__lock__done, struct compound_state *, cs,
9150 	    LOCK4res *, resp);
9151 }
9152 
9153 /* free function for LOCK/LOCKT */
9154 static void
9155 lock_denied_free(nfs_resop4 *resop)
9156 {
9157 	LOCK4denied *dp = NULL;
9158 
9159 	switch (resop->resop) {
9160 	case OP_LOCK:
9161 		if (resop->nfs_resop4_u.oplock.status == NFS4ERR_DENIED)
9162 			dp = &resop->nfs_resop4_u.oplock.LOCK4res_u.denied;
9163 		break;
9164 	case OP_LOCKT:
9165 		if (resop->nfs_resop4_u.oplockt.status == NFS4ERR_DENIED)
9166 			dp = &resop->nfs_resop4_u.oplockt.denied;
9167 		break;
9168 	default:
9169 		break;
9170 	}
9171 
9172 	if (dp)
9173 		kmem_free(dp->owner.owner_val, dp->owner.owner_len);
9174 }
9175 
9176 /*ARGSUSED*/
9177 void
9178 rfs4_op_locku(nfs_argop4 *argop, nfs_resop4 *resop,
9179     struct svc_req *req, struct compound_state *cs)
9180 {
9181 	LOCKU4args *args = &argop->nfs_argop4_u.oplocku;
9182 	LOCKU4res *resp = &resop->nfs_resop4_u.oplocku;
9183 	nfsstat4 status;
9184 	stateid4 *stateid = &args->lock_stateid;
9185 	rfs4_lo_state_t *lsp;
9186 
9187 	DTRACE_NFSV4_2(op__locku__start, struct compound_state *, cs,
9188 	    LOCKU4args *, args);
9189 
9190 	if (cs->vp == NULL) {
9191 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9192 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9193 		    LOCKU4res *, resp);
9194 		return;
9195 	}
9196 
9197 	if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE)) != NFS4_OK) {
9198 		*cs->statusp = resp->status = status;
9199 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9200 		    LOCKU4res *, resp);
9201 		return;
9202 	}
9203 
9204 	/* Ensure specified filehandle matches */
9205 	if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9206 		rfs4_lo_state_rele(lsp, TRUE);
9207 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9208 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9209 		    LOCKU4res *, resp);
9210 		return;
9211 	}
9212 
9213 	/* hold off other access to lsp while we tinker */
9214 	rfs4_sw_enter(&lsp->rls_sw);
9215 
9216 	switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9217 	case NFS4_CHECK_STATEID_OKAY:
9218 		if (rfs4_check_lock_seqid(args->seqid, lsp, resop)
9219 		    != NFS4_CHKSEQ_OKAY) {
9220 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9221 			goto end;
9222 		}
9223 		break;
9224 	case NFS4_CHECK_STATEID_OLD:
9225 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9226 		goto end;
9227 	case NFS4_CHECK_STATEID_BAD:
9228 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9229 		goto end;
9230 	case NFS4_CHECK_STATEID_EXPIRED:
9231 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9232 		goto end;
9233 	case NFS4_CHECK_STATEID_CLOSED:
9234 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9235 		goto end;
9236 	case NFS4_CHECK_STATEID_REPLAY:
9237 		switch (rfs4_check_lock_seqid(args->seqid, lsp, resop)) {
9238 		case NFS4_CHKSEQ_OKAY:
9239 				/*
9240 				 * This is a replayed stateid; if
9241 				 * seqid matches the next expected,
9242 				 * then client is using wrong seqid.
9243 				 */
9244 		case NFS4_CHKSEQ_BAD:
9245 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9246 			goto end;
9247 		case NFS4_CHKSEQ_REPLAY:
9248 			rfs4_update_lease(lsp->rls_locker->rl_client);
9249 			*cs->statusp = status = resp->status;
9250 			goto end;
9251 		}
9252 		break;
9253 	default:
9254 		ASSERT(FALSE);
9255 		break;
9256 	}
9257 
9258 	rfs4_update_lock_sequence(lsp);
9259 	rfs4_update_lease(lsp->rls_locker->rl_client);
9260 
9261 	/*
9262 	 * NFS4 only allows locking on regular files, so
9263 	 * verify type of object.
9264 	 */
9265 	if (cs->vp->v_type != VREG) {
9266 		if (cs->vp->v_type == VDIR)
9267 			status = NFS4ERR_ISDIR;
9268 		else
9269 			status = NFS4ERR_INVAL;
9270 		goto out;
9271 	}
9272 
9273 	if (rfs4_clnt_in_grace(lsp->rls_state->rs_owner->ro_client)) {
9274 		status = NFS4ERR_GRACE;
9275 		goto out;
9276 	}
9277 
9278 	status = rfs4_do_lock(lsp, args->locktype,
9279 	    args->offset, args->length, cs->cr, resop);
9280 
9281 out:
9282 	*cs->statusp = resp->status = status;
9283 
9284 	if (status == NFS4_OK)
9285 		resp->lock_stateid = lsp->rls_lockid.stateid;
9286 
9287 	rfs4_update_lock_resp(lsp, resop);
9288 
9289 end:
9290 	rfs4_sw_exit(&lsp->rls_sw);
9291 	rfs4_lo_state_rele(lsp, TRUE);
9292 
9293 	DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9294 	    LOCKU4res *, resp);
9295 }
9296 
9297 /*
9298  * LOCKT is a best effort routine, the client can not be guaranteed that
9299  * the status return is still in effect by the time the reply is received.
9300  * They are numerous race conditions in this routine, but we are not required
9301  * and can not be accurate.
9302  */
9303 /*ARGSUSED*/
9304 void
9305 rfs4_op_lockt(nfs_argop4 *argop, nfs_resop4 *resop,
9306     struct svc_req *req, struct compound_state *cs)
9307 {
9308 	LOCKT4args *args = &argop->nfs_argop4_u.oplockt;
9309 	LOCKT4res *resp = &resop->nfs_resop4_u.oplockt;
9310 	rfs4_lockowner_t *lo;
9311 	rfs4_client_t *cp;
9312 	bool_t create = FALSE;
9313 	struct flock64 flk;
9314 	int error;
9315 	int flag = FREAD | FWRITE;
9316 	int ltype;
9317 	length4 posix_length;
9318 	sysid_t sysid;
9319 	pid_t pid;
9320 
9321 	DTRACE_NFSV4_2(op__lockt__start, struct compound_state *, cs,
9322 	    LOCKT4args *, args);
9323 
9324 	if (cs->vp == NULL) {
9325 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9326 		goto out;
9327 	}
9328 
9329 	/*
9330 	 * NFS4 only allows locking on regular files, so
9331 	 * verify type of object.
9332 	 */
9333 	if (cs->vp->v_type != VREG) {
9334 		if (cs->vp->v_type == VDIR)
9335 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
9336 		else
9337 			*cs->statusp = resp->status =  NFS4ERR_INVAL;
9338 		goto out;
9339 	}
9340 
9341 	/*
9342 	 * Check out the clientid to ensure the server knows about it
9343 	 * so that we correctly inform the client of a server reboot.
9344 	 */
9345 	if ((cp = rfs4_findclient_by_id(args->owner.clientid, FALSE))
9346 	    == NULL) {
9347 		*cs->statusp = resp->status =
9348 		    rfs4_check_clientid(&args->owner.clientid, 0);
9349 		goto out;
9350 	}
9351 	if (rfs4_lease_expired(cp)) {
9352 		rfs4_client_close(cp);
9353 		/*
9354 		 * Protocol doesn't allow returning NFS4ERR_STALE as
9355 		 * other operations do on this check so STALE_CLIENTID
9356 		 * is returned instead
9357 		 */
9358 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
9359 		goto out;
9360 	}
9361 
9362 	if (rfs4_clnt_in_grace(cp) && !(cp->rc_can_reclaim)) {
9363 		*cs->statusp = resp->status = NFS4ERR_GRACE;
9364 		rfs4_client_rele(cp);
9365 		goto out;
9366 	}
9367 	rfs4_client_rele(cp);
9368 
9369 	resp->status = NFS4_OK;
9370 
9371 	switch (args->locktype) {
9372 	case READ_LT:
9373 	case READW_LT:
9374 		ltype = F_RDLCK;
9375 		break;
9376 	case WRITE_LT:
9377 	case WRITEW_LT:
9378 		ltype = F_WRLCK;
9379 		break;
9380 	}
9381 
9382 	posix_length = args->length;
9383 	/* Check for zero length. To lock to end of file use all ones for V4 */
9384 	if (posix_length == 0) {
9385 		*cs->statusp = resp->status = NFS4ERR_INVAL;
9386 		goto out;
9387 	} else if (posix_length == (length4)(~0)) {
9388 		posix_length = 0;	/* Posix to end of file  */
9389 	}
9390 
9391 	/* Find or create a lockowner */
9392 	lo = rfs4_findlockowner(&args->owner, &create);
9393 
9394 	if (lo) {
9395 		pid = lo->rl_pid;
9396 		if ((resp->status =
9397 		    rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
9398 			goto err;
9399 	} else {
9400 		pid = 0;
9401 		sysid = lockt_sysid;
9402 	}
9403 retry:
9404 	flk.l_type = ltype;
9405 	flk.l_whence = 0;		/* SEEK_SET */
9406 	flk.l_start = args->offset;
9407 	flk.l_len = posix_length;
9408 	flk.l_sysid = sysid;
9409 	flk.l_pid = pid;
9410 	flag |= F_REMOTELOCK;
9411 
9412 	LOCK_PRINT(rfs4_debug, "rfs4_op_lockt", F_GETLK, &flk);
9413 
9414 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
9415 	if (flk.l_len < 0 || flk.l_start < 0) {
9416 		resp->status = NFS4ERR_INVAL;
9417 		goto err;
9418 	}
9419 	error = VOP_FRLOCK(cs->vp, F_GETLK, &flk, flag, (u_offset_t)0,
9420 	    NULL, cs->cr, NULL);
9421 
9422 	/*
9423 	 * N.B. We map error values to nfsv4 errors. This is differrent
9424 	 * than puterrno4 routine.
9425 	 */
9426 	switch (error) {
9427 	case 0:
9428 		if (flk.l_type == F_UNLCK)
9429 			resp->status = NFS4_OK;
9430 		else {
9431 			if (lock_denied(&resp->denied, &flk) == NFS4ERR_EXPIRED)
9432 				goto retry;
9433 			resp->status = NFS4ERR_DENIED;
9434 		}
9435 		break;
9436 	case EOVERFLOW:
9437 		resp->status = NFS4ERR_INVAL;
9438 		break;
9439 	case EINVAL:
9440 		resp->status = NFS4ERR_NOTSUPP;
9441 		break;
9442 	default:
9443 		cmn_err(CE_WARN, "rfs4_op_lockt: unexpected errno (%d)",
9444 		    error);
9445 		resp->status = NFS4ERR_SERVERFAULT;
9446 		break;
9447 	}
9448 
9449 err:
9450 	if (lo)
9451 		rfs4_lockowner_rele(lo);
9452 	*cs->statusp = resp->status;
9453 out:
9454 	DTRACE_NFSV4_2(op__lockt__done, struct compound_state *, cs,
9455 	    LOCKT4res *, resp);
9456 }
9457 
9458 int
9459 rfs4_share(rfs4_state_t *sp, uint32_t access, uint32_t deny)
9460 {
9461 	int err;
9462 	int cmd;
9463 	vnode_t *vp;
9464 	struct shrlock shr;
9465 	struct shr_locowner shr_loco;
9466 	int fflags = 0;
9467 
9468 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9469 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9470 
9471 	if (sp->rs_closed)
9472 		return (NFS4ERR_OLD_STATEID);
9473 
9474 	vp = sp->rs_finfo->rf_vp;
9475 	ASSERT(vp);
9476 
9477 	shr.s_access = shr.s_deny = 0;
9478 
9479 	if (access & OPEN4_SHARE_ACCESS_READ) {
9480 		fflags |= FREAD;
9481 		shr.s_access |= F_RDACC;
9482 	}
9483 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
9484 		fflags |= FWRITE;
9485 		shr.s_access |= F_WRACC;
9486 	}
9487 	ASSERT(shr.s_access);
9488 
9489 	if (deny & OPEN4_SHARE_DENY_READ)
9490 		shr.s_deny |= F_RDDNY;
9491 	if (deny & OPEN4_SHARE_DENY_WRITE)
9492 		shr.s_deny |= F_WRDNY;
9493 
9494 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9495 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9496 	shr_loco.sl_pid = shr.s_pid;
9497 	shr_loco.sl_id = shr.s_sysid;
9498 	shr.s_owner = (caddr_t)&shr_loco;
9499 	shr.s_own_len = sizeof (shr_loco);
9500 
9501 	cmd = nbl_need_check(vp) ? F_SHARE_NBMAND : F_SHARE;
9502 
9503 	err = VOP_SHRLOCK(vp, cmd, &shr, fflags, CRED(), NULL);
9504 	if (err != 0) {
9505 		if (err == EAGAIN)
9506 			err = NFS4ERR_SHARE_DENIED;
9507 		else
9508 			err = puterrno4(err);
9509 		return (err);
9510 	}
9511 
9512 	sp->rs_share_access |= access;
9513 	sp->rs_share_deny |= deny;
9514 
9515 	return (0);
9516 }
9517 
9518 int
9519 rfs4_unshare(rfs4_state_t *sp)
9520 {
9521 	int err;
9522 	struct shrlock shr;
9523 	struct shr_locowner shr_loco;
9524 
9525 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9526 
9527 	if (sp->rs_closed || sp->rs_share_access == 0)
9528 		return (0);
9529 
9530 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9531 	ASSERT(sp->rs_finfo->rf_vp);
9532 
9533 	shr.s_access = shr.s_deny = 0;
9534 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9535 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9536 	shr_loco.sl_pid = shr.s_pid;
9537 	shr_loco.sl_id = shr.s_sysid;
9538 	shr.s_owner = (caddr_t)&shr_loco;
9539 	shr.s_own_len = sizeof (shr_loco);
9540 
9541 	err = VOP_SHRLOCK(sp->rs_finfo->rf_vp, F_UNSHARE, &shr, 0, CRED(),
9542 	    NULL);
9543 	if (err != 0) {
9544 		err = puterrno4(err);
9545 		return (err);
9546 	}
9547 
9548 	sp->rs_share_access = 0;
9549 	sp->rs_share_deny = 0;
9550 
9551 	return (0);
9552 
9553 }
9554 
9555 static int
9556 rdma_setup_read_data4(READ4args *args, READ4res *rok)
9557 {
9558 	struct clist	*wcl;
9559 	count4		count = rok->data_len;
9560 	int		wlist_len;
9561 
9562 	wcl = args->wlist;
9563 	if (rdma_setup_read_chunks(wcl, count, &wlist_len) == FALSE) {
9564 		return (FALSE);
9565 	}
9566 	wcl = args->wlist;
9567 	rok->wlist_len = wlist_len;
9568 	rok->wlist = wcl;
9569 	return (TRUE);
9570 }
9571 
9572 /* tunable to disable server referrals */
9573 int rfs4_no_referrals = 0;
9574 
9575 /*
9576  * Find an NFS record in reparse point data.
9577  * Returns 0 for success and <0 or an errno value on failure.
9578  */
9579 int
9580 vn_find_nfs_record(vnode_t *vp, nvlist_t **nvlp, char **svcp, char **datap)
9581 {
9582 	int err;
9583 	char *stype, *val;
9584 	nvlist_t *nvl;
9585 	nvpair_t *curr;
9586 
9587 	if ((nvl = reparse_init()) == NULL)
9588 		return (-1);
9589 
9590 	if ((err = reparse_vnode_parse(vp, nvl)) != 0) {
9591 		reparse_free(nvl);
9592 		return (err);
9593 	}
9594 
9595 	curr = NULL;
9596 	while ((curr = nvlist_next_nvpair(nvl, curr)) != NULL) {
9597 		if ((stype = nvpair_name(curr)) == NULL) {
9598 			reparse_free(nvl);
9599 			return (-2);
9600 		}
9601 		if (strncasecmp(stype, "NFS", 3) == 0)
9602 			break;
9603 	}
9604 
9605 	if ((curr == NULL) ||
9606 	    (nvpair_value_string(curr, &val))) {
9607 		reparse_free(nvl);
9608 		return (-3);
9609 	}
9610 	*nvlp = nvl;
9611 	*svcp = stype;
9612 	*datap = val;
9613 	return (0);
9614 }
9615 
9616 int
9617 vn_is_nfs_reparse(vnode_t *vp, cred_t *cr)
9618 {
9619 	nvlist_t *nvl;
9620 	char *s, *d;
9621 
9622 	if (rfs4_no_referrals != 0)
9623 		return (B_FALSE);
9624 
9625 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9626 		return (B_FALSE);
9627 
9628 	if (vn_find_nfs_record(vp, &nvl, &s, &d) != 0)
9629 		return (B_FALSE);
9630 
9631 	reparse_free(nvl);
9632 
9633 	return (B_TRUE);
9634 }
9635 
9636 /*
9637  * There is a user-level copy of this routine in ref_subr.c.
9638  * Changes should be kept in sync.
9639  */
9640 static int
9641 nfs4_create_components(char *path, component4 *comp4)
9642 {
9643 	int slen, plen, ncomp;
9644 	char *ori_path, *nxtc, buf[MAXNAMELEN];
9645 
9646 	if (path == NULL)
9647 		return (0);
9648 
9649 	plen = strlen(path) + 1;	/* include the terminator */
9650 	ori_path = path;
9651 	ncomp = 0;
9652 
9653 	/* count number of components in the path */
9654 	for (nxtc = path; nxtc < ori_path + plen; nxtc++) {
9655 		if (*nxtc == '/' || *nxtc == '\0' || *nxtc == '\n') {
9656 			if ((slen = nxtc - path) == 0) {
9657 				path = nxtc + 1;
9658 				continue;
9659 			}
9660 
9661 			if (comp4 != NULL) {
9662 				bcopy(path, buf, slen);
9663 				buf[slen] = '\0';
9664 				(void) str_to_utf8(buf, &comp4[ncomp]);
9665 			}
9666 
9667 			ncomp++;	/* 1 valid component */
9668 			path = nxtc + 1;
9669 		}
9670 		if (*nxtc == '\0' || *nxtc == '\n')
9671 			break;
9672 	}
9673 
9674 	return (ncomp);
9675 }
9676 
9677 /*
9678  * There is a user-level copy of this routine in ref_subr.c.
9679  * Changes should be kept in sync.
9680  */
9681 static int
9682 make_pathname4(char *path, pathname4 *pathname)
9683 {
9684 	int ncomp;
9685 	component4 *comp4;
9686 
9687 	if (pathname == NULL)
9688 		return (0);
9689 
9690 	if (path == NULL) {
9691 		pathname->pathname4_val = NULL;
9692 		pathname->pathname4_len = 0;
9693 		return (0);
9694 	}
9695 
9696 	/* count number of components to alloc buffer */
9697 	if ((ncomp = nfs4_create_components(path, NULL)) == 0) {
9698 		pathname->pathname4_val = NULL;
9699 		pathname->pathname4_len = 0;
9700 		return (0);
9701 	}
9702 	comp4 = kmem_zalloc(ncomp * sizeof (component4), KM_SLEEP);
9703 
9704 	/* copy components into allocated buffer */
9705 	ncomp = nfs4_create_components(path, comp4);
9706 
9707 	pathname->pathname4_val = comp4;
9708 	pathname->pathname4_len = ncomp;
9709 
9710 	return (ncomp);
9711 }
9712 
9713 #define	xdr_fs_locations4 xdr_fattr4_fs_locations
9714 
9715 fs_locations4 *
9716 fetch_referral(vnode_t *vp, cred_t *cr)
9717 {
9718 	nvlist_t *nvl;
9719 	char *stype, *sdata;
9720 	fs_locations4 *result;
9721 	char buf[1024];
9722 	size_t bufsize;
9723 	XDR xdr;
9724 	int err;
9725 
9726 	/*
9727 	 * Check attrs to ensure it's a reparse point
9728 	 */
9729 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9730 		return (NULL);
9731 
9732 	/*
9733 	 * Look for an NFS record and get the type and data
9734 	 */
9735 	if (vn_find_nfs_record(vp, &nvl, &stype, &sdata) != 0)
9736 		return (NULL);
9737 
9738 	/*
9739 	 * With the type and data, upcall to get the referral
9740 	 */
9741 	bufsize = sizeof (buf);
9742 	bzero(buf, sizeof (buf));
9743 	err = reparse_kderef((const char *)stype, (const char *)sdata,
9744 	    buf, &bufsize);
9745 	reparse_free(nvl);
9746 
9747 	DTRACE_PROBE4(nfs4serv__func__referral__upcall,
9748 	    char *, stype, char *, sdata, char *, buf, int, err);
9749 	if (err) {
9750 		cmn_err(CE_NOTE,
9751 		    "reparsed daemon not running: unable to get referral (%d)",
9752 		    err);
9753 		return (NULL);
9754 	}
9755 
9756 	/*
9757 	 * We get an XDR'ed record back from the kderef call
9758 	 */
9759 	xdrmem_create(&xdr, buf, bufsize, XDR_DECODE);
9760 	result = kmem_alloc(sizeof (fs_locations4), KM_SLEEP);
9761 	err = xdr_fs_locations4(&xdr, result);
9762 	XDR_DESTROY(&xdr);
9763 	if (err != TRUE) {
9764 		DTRACE_PROBE1(nfs4serv__func__referral__upcall__xdrfail,
9765 		    int, err);
9766 		return (NULL);
9767 	}
9768 
9769 	/*
9770 	 * Look at path to recover fs_root, ignoring the leading '/'
9771 	 */
9772 	(void) make_pathname4(vp->v_path, &result->fs_root);
9773 
9774 	return (result);
9775 }
9776 
9777 char *
9778 build_symlink(vnode_t *vp, cred_t *cr, size_t *strsz)
9779 {
9780 	fs_locations4 *fsl;
9781 	fs_location4 *fs;
9782 	char *server, *path, *symbuf;
9783 	static char *prefix = "/net/";
9784 	int i, size, npaths;
9785 	uint_t len;
9786 
9787 	/* Get the referral */
9788 	if ((fsl = fetch_referral(vp, cr)) == NULL)
9789 		return (NULL);
9790 
9791 	/* Deal with only the first location and first server */
9792 	fs = &fsl->locations_val[0];
9793 	server = utf8_to_str(&fs->server_val[0], &len, NULL);
9794 	if (server == NULL) {
9795 		rfs4_free_fs_locations4(fsl);
9796 		kmem_free(fsl, sizeof (fs_locations4));
9797 		return (NULL);
9798 	}
9799 
9800 	/* Figure out size for "/net/" + host + /path/path/path + NULL */
9801 	size = strlen(prefix) + len;
9802 	for (i = 0; i < fs->rootpath.pathname4_len; i++)
9803 		size += fs->rootpath.pathname4_val[i].utf8string_len + 1;
9804 
9805 	/* Allocate the symlink buffer and fill it */
9806 	symbuf = kmem_zalloc(size, KM_SLEEP);
9807 	(void) strcat(symbuf, prefix);
9808 	(void) strcat(symbuf, server);
9809 	kmem_free(server, len);
9810 
9811 	npaths = 0;
9812 	for (i = 0; i < fs->rootpath.pathname4_len; i++) {
9813 		path = utf8_to_str(&fs->rootpath.pathname4_val[i], &len, NULL);
9814 		if (path == NULL)
9815 			continue;
9816 		(void) strcat(symbuf, "/");
9817 		(void) strcat(symbuf, path);
9818 		npaths++;
9819 		kmem_free(path, len);
9820 	}
9821 
9822 	rfs4_free_fs_locations4(fsl);
9823 	kmem_free(fsl, sizeof (fs_locations4));
9824 
9825 	if (strsz != NULL)
9826 		*strsz = size;
9827 	return (symbuf);
9828 }
9829 
9830 /*
9831  * Check to see if we have a downrev Solaris client, so that we
9832  * can send it a symlink instead of a referral.
9833  */
9834 int
9835 client_is_downrev(struct svc_req *req)
9836 {
9837 	struct sockaddr *ca;
9838 	rfs4_clntip_t *ci;
9839 	bool_t create = FALSE;
9840 	int is_downrev;
9841 
9842 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
9843 	ASSERT(ca);
9844 	ci = rfs4_find_clntip(ca, &create);
9845 	if (ci == NULL)
9846 		return (0);
9847 	is_downrev = ci->ri_no_referrals;
9848 	rfs4_dbe_rele(ci->ri_dbe);
9849 	return (is_downrev);
9850 }
9851