xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs4_srv.c (revision 03270635d68df6a0392fb8f4b7c04acad764648b)
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 2016 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2012, 2016 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 	fid_t fid;
876 	uint_t count, i;
877 	secinfo4 *resok_val;
878 	struct secinfo *secp;
879 	seconfig_t *si;
880 	bool_t did_traverse = FALSE;
881 	int dotdot, walk;
882 
883 	dvp = cs->vp;
884 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
885 
886 	/*
887 	 * If dotdotting, then need to check whether it's above the
888 	 * root of a filesystem, or above an export point.
889 	 */
890 	if (dotdot) {
891 
892 		/*
893 		 * If dotdotting at the root of a filesystem, then
894 		 * need to traverse back to the mounted-on filesystem
895 		 * and do the dotdot lookup there.
896 		 */
897 		if (cs->vp->v_flag & VROOT) {
898 
899 			/*
900 			 * If at the system root, then can
901 			 * go up no further.
902 			 */
903 			if (VN_CMP(dvp, rootdir))
904 				return (puterrno4(ENOENT));
905 
906 			/*
907 			 * Traverse back to the mounted-on filesystem
908 			 */
909 			dvp = untraverse(cs->vp);
910 
911 			/*
912 			 * Set the different_export flag so we remember
913 			 * to pick up a new exportinfo entry for
914 			 * this new filesystem.
915 			 */
916 			different_export = 1;
917 		} else {
918 
919 			/*
920 			 * If dotdotting above an export point then set
921 			 * the different_export to get new export info.
922 			 */
923 			different_export = nfs_exported(cs->exi, cs->vp);
924 		}
925 	}
926 
927 	/*
928 	 * Get the vnode for the component "nm".
929 	 */
930 	error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cs->cr,
931 	    NULL, NULL, NULL);
932 	if (error)
933 		return (puterrno4(error));
934 
935 	/*
936 	 * If the vnode is in a pseudo filesystem, or if the security flavor
937 	 * used in the request is valid but not an explicitly shared flavor,
938 	 * or the access bit indicates that this is a limited access,
939 	 * check whether this vnode is visible.
940 	 */
941 	if (!different_export &&
942 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
943 	    cs->access & CS_ACCESS_LIMITED)) {
944 		if (! nfs_visible(cs->exi, vp, &different_export)) {
945 			VN_RELE(vp);
946 			return (puterrno4(ENOENT));
947 		}
948 	}
949 
950 	/*
951 	 * If it's a mountpoint, then traverse it.
952 	 */
953 	if (vn_ismntpt(vp)) {
954 		if ((error = traverse(&vp)) != 0) {
955 			VN_RELE(vp);
956 			return (puterrno4(error));
957 		}
958 		/* remember that we had to traverse mountpoint */
959 		did_traverse = TRUE;
960 		different_export = 1;
961 	} else if (vp->v_vfsp != dvp->v_vfsp) {
962 		/*
963 		 * If vp isn't a mountpoint and the vfs ptrs aren't the same,
964 		 * then vp is probably an LOFS object.  We don't need the
965 		 * realvp, we just need to know that we might have crossed
966 		 * a server fs boundary and need to call checkexport4.
967 		 * (LOFS lookup hides server fs mountpoints, and actually calls
968 		 * traverse)
969 		 */
970 		different_export = 1;
971 	}
972 
973 	/*
974 	 * Get the export information for it.
975 	 */
976 	if (different_export) {
977 
978 		bzero(&fid, sizeof (fid));
979 		fid.fid_len = MAXFIDSZ;
980 		error = vop_fid_pseudo(vp, &fid);
981 		if (error) {
982 			VN_RELE(vp);
983 			return (puterrno4(error));
984 		}
985 
986 		if (dotdot)
987 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
988 		else
989 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
990 
991 		if (exi == NULL) {
992 			if (did_traverse == TRUE) {
993 				/*
994 				 * If this vnode is a mounted-on vnode,
995 				 * but the mounted-on file system is not
996 				 * exported, send back the secinfo for
997 				 * the exported node that the mounted-on
998 				 * vnode lives in.
999 				 */
1000 				exi = cs->exi;
1001 			} else {
1002 				VN_RELE(vp);
1003 				return (puterrno4(EACCES));
1004 			}
1005 		}
1006 	} else {
1007 		exi = cs->exi;
1008 	}
1009 	ASSERT(exi != NULL);
1010 
1011 
1012 	/*
1013 	 * Create the secinfo result based on the security information
1014 	 * from the exportinfo structure (exi).
1015 	 *
1016 	 * Return all flavors for a pseudo node.
1017 	 * For a real export node, return the flavor that the client
1018 	 * has access with.
1019 	 */
1020 	ASSERT(RW_LOCK_HELD(&exported_lock));
1021 	if (PSEUDO(exi)) {
1022 		count = exi->exi_export.ex_seccnt; /* total sec count */
1023 		resok_val = kmem_alloc(count * sizeof (secinfo4), KM_SLEEP);
1024 		secp = exi->exi_export.ex_secinfo;
1025 
1026 		for (i = 0; i < count; i++) {
1027 			si = &secp[i].s_secinfo;
1028 			resok_val[i].flavor = si->sc_rpcnum;
1029 			if (resok_val[i].flavor == RPCSEC_GSS) {
1030 				rpcsec_gss_info *info;
1031 
1032 				info = &resok_val[i].flavor_info;
1033 				info->qop = si->sc_qop;
1034 				info->service = (rpc_gss_svc_t)si->sc_service;
1035 
1036 				/* get oid opaque data */
1037 				info->oid.sec_oid4_len =
1038 				    si->sc_gss_mech_type->length;
1039 				info->oid.sec_oid4_val = kmem_alloc(
1040 				    si->sc_gss_mech_type->length, KM_SLEEP);
1041 				bcopy(
1042 				    si->sc_gss_mech_type->elements,
1043 				    info->oid.sec_oid4_val,
1044 				    info->oid.sec_oid4_len);
1045 			}
1046 		}
1047 		resp->SECINFO4resok_len = count;
1048 		resp->SECINFO4resok_val = resok_val;
1049 	} else {
1050 		int ret_cnt = 0, k = 0;
1051 		int *flavor_list;
1052 
1053 		count = exi->exi_export.ex_seccnt; /* total sec count */
1054 		secp = exi->exi_export.ex_secinfo;
1055 
1056 		flavor_list = kmem_alloc(count * sizeof (int), KM_SLEEP);
1057 		/* find out which flavors to return */
1058 		for (i = 0; i < count; i ++) {
1059 			int access, flavor, perm;
1060 
1061 			flavor = secp[i].s_secinfo.sc_nfsnum;
1062 			perm = secp[i].s_flags;
1063 
1064 			access = nfsauth4_secinfo_access(exi, cs->req,
1065 			    flavor, perm, cs->basecr);
1066 
1067 			if (! (access & NFSAUTH_DENIED) &&
1068 			    ! (access & NFSAUTH_WRONGSEC)) {
1069 				flavor_list[ret_cnt] = flavor;
1070 				ret_cnt++;
1071 			}
1072 		}
1073 
1074 		/* Create the returning SECINFO value */
1075 		resok_val = kmem_alloc(ret_cnt * sizeof (secinfo4), KM_SLEEP);
1076 
1077 		for (i = 0; i < count; i++) {
1078 			/*
1079 			 * If the flavor is in the flavor list,
1080 			 * fill in resok_val.
1081 			 */
1082 			si = &secp[i].s_secinfo;
1083 			if (in_flavor_list(si->sc_nfsnum,
1084 			    flavor_list, ret_cnt)) {
1085 				resok_val[k].flavor = si->sc_rpcnum;
1086 				if (resok_val[k].flavor == RPCSEC_GSS) {
1087 					rpcsec_gss_info *info;
1088 
1089 					info = &resok_val[k].flavor_info;
1090 					info->qop = si->sc_qop;
1091 					info->service = (rpc_gss_svc_t)
1092 					    si->sc_service;
1093 
1094 					/* get oid opaque data */
1095 					info->oid.sec_oid4_len =
1096 					    si->sc_gss_mech_type->length;
1097 					info->oid.sec_oid4_val = kmem_alloc(
1098 					    si->sc_gss_mech_type->length,
1099 					    KM_SLEEP);
1100 					bcopy(si->sc_gss_mech_type->elements,
1101 					    info->oid.sec_oid4_val,
1102 					    info->oid.sec_oid4_len);
1103 				}
1104 				k++;
1105 			}
1106 			if (k >= ret_cnt)
1107 				break;
1108 		}
1109 		resp->SECINFO4resok_len = ret_cnt;
1110 		resp->SECINFO4resok_val = resok_val;
1111 		kmem_free(flavor_list, count * sizeof (int));
1112 	}
1113 
1114 	VN_RELE(vp);
1115 	return (NFS4_OK);
1116 }
1117 
1118 /*
1119  * SECINFO (Operation 33): Obtain required security information on
1120  * the component name in the format of (security-mechanism-oid, qop, service)
1121  * triplets.
1122  */
1123 /* ARGSUSED */
1124 static void
1125 rfs4_op_secinfo(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1126     struct compound_state *cs)
1127 {
1128 	SECINFO4args *args = &argop->nfs_argop4_u.opsecinfo;
1129 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1130 	utf8string *utfnm = &args->name;
1131 	uint_t len;
1132 	char *nm;
1133 	struct sockaddr *ca;
1134 	char *name = NULL;
1135 	nfsstat4 status = NFS4_OK;
1136 
1137 	DTRACE_NFSV4_2(op__secinfo__start, struct compound_state *, cs,
1138 	    SECINFO4args *, args);
1139 
1140 	/*
1141 	 * Current file handle (cfh) should have been set before getting
1142 	 * into this function. If not, return error.
1143 	 */
1144 	if (cs->vp == NULL) {
1145 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1146 		goto out;
1147 	}
1148 
1149 	if (cs->vp->v_type != VDIR) {
1150 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1151 		goto out;
1152 	}
1153 
1154 	/*
1155 	 * Verify the component name. If failed, error out, but
1156 	 * do not error out if the component name is a "..".
1157 	 * SECINFO will return its parents secinfo data for SECINFO "..".
1158 	 */
1159 	status = utf8_dir_verify(utfnm);
1160 	if (status != NFS4_OK) {
1161 		if (utfnm->utf8string_len != 2 ||
1162 		    utfnm->utf8string_val[0] != '.' ||
1163 		    utfnm->utf8string_val[1] != '.') {
1164 			*cs->statusp = resp->status = status;
1165 			goto out;
1166 		}
1167 	}
1168 
1169 	nm = utf8_to_str(utfnm, &len, NULL);
1170 	if (nm == NULL) {
1171 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1172 		goto out;
1173 	}
1174 
1175 	if (len > MAXNAMELEN) {
1176 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1177 		kmem_free(nm, len);
1178 		goto out;
1179 	}
1180 
1181 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1182 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1183 	    MAXPATHLEN  + 1);
1184 
1185 	if (name == NULL) {
1186 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1187 		kmem_free(nm, len);
1188 		goto out;
1189 	}
1190 
1191 
1192 	*cs->statusp = resp->status = do_rfs4_op_secinfo(cs, name, resp);
1193 
1194 	if (name != nm)
1195 		kmem_free(name, MAXPATHLEN + 1);
1196 	kmem_free(nm, len);
1197 
1198 out:
1199 	DTRACE_NFSV4_2(op__secinfo__done, struct compound_state *, cs,
1200 	    SECINFO4res *, resp);
1201 }
1202 
1203 /*
1204  * Free SECINFO result.
1205  */
1206 /* ARGSUSED */
1207 static void
1208 rfs4_op_secinfo_free(nfs_resop4 *resop)
1209 {
1210 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1211 	int count, i;
1212 	secinfo4 *resok_val;
1213 
1214 	/* If this is not an Ok result, nothing to free. */
1215 	if (resp->status != NFS4_OK) {
1216 		return;
1217 	}
1218 
1219 	count = resp->SECINFO4resok_len;
1220 	resok_val = resp->SECINFO4resok_val;
1221 
1222 	for (i = 0; i < count; i++) {
1223 		if (resok_val[i].flavor == RPCSEC_GSS) {
1224 			rpcsec_gss_info *info;
1225 
1226 			info = &resok_val[i].flavor_info;
1227 			kmem_free(info->oid.sec_oid4_val,
1228 			    info->oid.sec_oid4_len);
1229 		}
1230 	}
1231 	kmem_free(resok_val, count * sizeof (secinfo4));
1232 	resp->SECINFO4resok_len = 0;
1233 	resp->SECINFO4resok_val = NULL;
1234 }
1235 
1236 /* ARGSUSED */
1237 static void
1238 rfs4_op_access(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1239     struct compound_state *cs)
1240 {
1241 	ACCESS4args *args = &argop->nfs_argop4_u.opaccess;
1242 	ACCESS4res *resp = &resop->nfs_resop4_u.opaccess;
1243 	int error;
1244 	vnode_t *vp;
1245 	struct vattr va;
1246 	int checkwriteperm;
1247 	cred_t *cr = cs->cr;
1248 	bslabel_t *clabel, *slabel;
1249 	ts_label_t *tslabel;
1250 	boolean_t admin_low_client;
1251 
1252 	DTRACE_NFSV4_2(op__access__start, struct compound_state *, cs,
1253 	    ACCESS4args *, args);
1254 
1255 #if 0	/* XXX allow access even if !cs->access. Eventually only pseudo fs */
1256 	if (cs->access == CS_ACCESS_DENIED) {
1257 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1258 		goto out;
1259 	}
1260 #endif
1261 	if (cs->vp == NULL) {
1262 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1263 		goto out;
1264 	}
1265 
1266 	ASSERT(cr != NULL);
1267 
1268 	vp = cs->vp;
1269 
1270 	/*
1271 	 * If the file system is exported read only, it is not appropriate
1272 	 * to check write permissions for regular files and directories.
1273 	 * Special files are interpreted by the client, so the underlying
1274 	 * permissions are sent back to the client for interpretation.
1275 	 */
1276 	if (rdonly4(req, cs) &&
1277 	    (vp->v_type == VREG || vp->v_type == VDIR))
1278 		checkwriteperm = 0;
1279 	else
1280 		checkwriteperm = 1;
1281 
1282 	/*
1283 	 * XXX
1284 	 * We need the mode so that we can correctly determine access
1285 	 * permissions relative to a mandatory lock file.  Access to
1286 	 * mandatory lock files is denied on the server, so it might
1287 	 * as well be reflected to the server during the open.
1288 	 */
1289 	va.va_mask = AT_MODE;
1290 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1291 	if (error) {
1292 		*cs->statusp = resp->status = puterrno4(error);
1293 		goto out;
1294 	}
1295 	resp->access = 0;
1296 	resp->supported = 0;
1297 
1298 	if (is_system_labeled()) {
1299 		ASSERT(req->rq_label != NULL);
1300 		clabel = req->rq_label;
1301 		DTRACE_PROBE2(tx__rfs4__log__info__opaccess__clabel, char *,
1302 		    "got client label from request(1)",
1303 		    struct svc_req *, req);
1304 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
1305 			if ((tslabel = nfs_getflabel(vp, cs->exi)) == NULL) {
1306 				*cs->statusp = resp->status = puterrno4(EACCES);
1307 				goto out;
1308 			}
1309 			slabel = label2bslabel(tslabel);
1310 			DTRACE_PROBE3(tx__rfs4__log__info__opaccess__slabel,
1311 			    char *, "got server label(1) for vp(2)",
1312 			    bslabel_t *, slabel, vnode_t *, vp);
1313 
1314 			admin_low_client = B_FALSE;
1315 		} else
1316 			admin_low_client = B_TRUE;
1317 	}
1318 
1319 	if (args->access & ACCESS4_READ) {
1320 		error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
1321 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1322 		    (!is_system_labeled() || admin_low_client ||
1323 		    bldominates(clabel, slabel)))
1324 			resp->access |= ACCESS4_READ;
1325 		resp->supported |= ACCESS4_READ;
1326 	}
1327 	if ((args->access & ACCESS4_LOOKUP) && vp->v_type == VDIR) {
1328 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1329 		if (!error && (!is_system_labeled() || admin_low_client ||
1330 		    bldominates(clabel, slabel)))
1331 			resp->access |= ACCESS4_LOOKUP;
1332 		resp->supported |= ACCESS4_LOOKUP;
1333 	}
1334 	if (checkwriteperm &&
1335 	    (args->access & (ACCESS4_MODIFY|ACCESS4_EXTEND))) {
1336 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1337 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1338 		    (!is_system_labeled() || admin_low_client ||
1339 		    blequal(clabel, slabel)))
1340 			resp->access |=
1341 			    (args->access & (ACCESS4_MODIFY | ACCESS4_EXTEND));
1342 		resp->supported |=
1343 		    resp->access & (ACCESS4_MODIFY | ACCESS4_EXTEND);
1344 	}
1345 
1346 	if (checkwriteperm &&
1347 	    (args->access & ACCESS4_DELETE) && vp->v_type == VDIR) {
1348 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1349 		if (!error && (!is_system_labeled() || admin_low_client ||
1350 		    blequal(clabel, slabel)))
1351 			resp->access |= ACCESS4_DELETE;
1352 		resp->supported |= ACCESS4_DELETE;
1353 	}
1354 	if (args->access & ACCESS4_EXECUTE && vp->v_type != VDIR) {
1355 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1356 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1357 		    (!is_system_labeled() || admin_low_client ||
1358 		    bldominates(clabel, slabel)))
1359 			resp->access |= ACCESS4_EXECUTE;
1360 		resp->supported |= ACCESS4_EXECUTE;
1361 	}
1362 
1363 	if (is_system_labeled() && !admin_low_client)
1364 		label_rele(tslabel);
1365 
1366 	*cs->statusp = resp->status = NFS4_OK;
1367 out:
1368 	DTRACE_NFSV4_2(op__access__done, struct compound_state *, cs,
1369 	    ACCESS4res *, resp);
1370 }
1371 
1372 /* ARGSUSED */
1373 static void
1374 rfs4_op_commit(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1375     struct compound_state *cs)
1376 {
1377 	COMMIT4args *args = &argop->nfs_argop4_u.opcommit;
1378 	COMMIT4res *resp = &resop->nfs_resop4_u.opcommit;
1379 	int error;
1380 	vnode_t *vp = cs->vp;
1381 	cred_t *cr = cs->cr;
1382 	vattr_t va;
1383 
1384 	DTRACE_NFSV4_2(op__commit__start, struct compound_state *, cs,
1385 	    COMMIT4args *, args);
1386 
1387 	if (vp == NULL) {
1388 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1389 		goto out;
1390 	}
1391 	if (cs->access == CS_ACCESS_DENIED) {
1392 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1393 		goto out;
1394 	}
1395 
1396 	if (args->offset + args->count < args->offset) {
1397 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1398 		goto out;
1399 	}
1400 
1401 	va.va_mask = AT_UID;
1402 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1403 
1404 	/*
1405 	 * If we can't get the attributes, then we can't do the
1406 	 * right access checking.  So, we'll fail the request.
1407 	 */
1408 	if (error) {
1409 		*cs->statusp = resp->status = puterrno4(error);
1410 		goto out;
1411 	}
1412 	if (rdonly4(req, cs)) {
1413 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1414 		goto out;
1415 	}
1416 
1417 	if (vp->v_type != VREG) {
1418 		if (vp->v_type == VDIR)
1419 			resp->status = NFS4ERR_ISDIR;
1420 		else
1421 			resp->status = NFS4ERR_INVAL;
1422 		*cs->statusp = resp->status;
1423 		goto out;
1424 	}
1425 
1426 	if (crgetuid(cr) != va.va_uid &&
1427 	    (error = VOP_ACCESS(vp, VWRITE, 0, cs->cr, NULL))) {
1428 		*cs->statusp = resp->status = puterrno4(error);
1429 		goto out;
1430 	}
1431 
1432 	error = VOP_FSYNC(vp, FSYNC, cr, NULL);
1433 
1434 	if (error) {
1435 		*cs->statusp = resp->status = puterrno4(error);
1436 		goto out;
1437 	}
1438 
1439 	*cs->statusp = resp->status = NFS4_OK;
1440 	resp->writeverf = Write4verf;
1441 out:
1442 	DTRACE_NFSV4_2(op__commit__done, struct compound_state *, cs,
1443 	    COMMIT4res *, resp);
1444 }
1445 
1446 /*
1447  * rfs4_op_mknod is called from rfs4_op_create after all initial verification
1448  * was completed. It does the nfsv4 create for special files.
1449  */
1450 /* ARGSUSED */
1451 static vnode_t *
1452 do_rfs4_op_mknod(CREATE4args *args, CREATE4res *resp, struct svc_req *req,
1453     struct compound_state *cs, vattr_t *vap, char *nm)
1454 {
1455 	int error;
1456 	cred_t *cr = cs->cr;
1457 	vnode_t *dvp = cs->vp;
1458 	vnode_t *vp = NULL;
1459 	int mode;
1460 	enum vcexcl excl;
1461 
1462 	switch (args->type) {
1463 	case NF4CHR:
1464 	case NF4BLK:
1465 		if (secpolicy_sys_devices(cr) != 0) {
1466 			*cs->statusp = resp->status = NFS4ERR_PERM;
1467 			return (NULL);
1468 		}
1469 		if (args->type == NF4CHR)
1470 			vap->va_type = VCHR;
1471 		else
1472 			vap->va_type = VBLK;
1473 		vap->va_rdev = makedevice(args->ftype4_u.devdata.specdata1,
1474 		    args->ftype4_u.devdata.specdata2);
1475 		vap->va_mask |= AT_RDEV;
1476 		break;
1477 	case NF4SOCK:
1478 		vap->va_type = VSOCK;
1479 		break;
1480 	case NF4FIFO:
1481 		vap->va_type = VFIFO;
1482 		break;
1483 	default:
1484 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1485 		return (NULL);
1486 	}
1487 
1488 	/*
1489 	 * Must specify the mode.
1490 	 */
1491 	if (!(vap->va_mask & AT_MODE)) {
1492 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1493 		return (NULL);
1494 	}
1495 
1496 	excl = EXCL;
1497 
1498 	mode = 0;
1499 
1500 	error = VOP_CREATE(dvp, nm, vap, excl, mode, &vp, cr, 0, NULL, NULL);
1501 	if (error) {
1502 		*cs->statusp = resp->status = puterrno4(error);
1503 		return (NULL);
1504 	}
1505 	return (vp);
1506 }
1507 
1508 /*
1509  * nfsv4 create is used to create non-regular files. For regular files,
1510  * use nfsv4 open.
1511  */
1512 /* ARGSUSED */
1513 static void
1514 rfs4_op_create(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1515     struct compound_state *cs)
1516 {
1517 	CREATE4args *args = &argop->nfs_argop4_u.opcreate;
1518 	CREATE4res *resp = &resop->nfs_resop4_u.opcreate;
1519 	int error;
1520 	struct vattr bva, iva, iva2, ava, *vap;
1521 	cred_t *cr = cs->cr;
1522 	vnode_t *dvp = cs->vp;
1523 	vnode_t *vp = NULL;
1524 	vnode_t *realvp;
1525 	char *nm, *lnm;
1526 	uint_t len, llen;
1527 	int syncval = 0;
1528 	struct nfs4_svgetit_arg sarg;
1529 	struct nfs4_ntov_table ntov;
1530 	struct statvfs64 sb;
1531 	nfsstat4 status;
1532 	struct sockaddr *ca;
1533 	char *name = NULL;
1534 	char *lname = NULL;
1535 
1536 	DTRACE_NFSV4_2(op__create__start, struct compound_state *, cs,
1537 	    CREATE4args *, args);
1538 
1539 	resp->attrset = 0;
1540 
1541 	if (dvp == NULL) {
1542 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1543 		goto out;
1544 	}
1545 
1546 	/*
1547 	 * If there is an unshared filesystem mounted on this vnode,
1548 	 * do not allow to create an object in this directory.
1549 	 */
1550 	if (vn_ismntpt(dvp)) {
1551 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1552 		goto out;
1553 	}
1554 
1555 	/* Verify that type is correct */
1556 	switch (args->type) {
1557 	case NF4LNK:
1558 	case NF4BLK:
1559 	case NF4CHR:
1560 	case NF4SOCK:
1561 	case NF4FIFO:
1562 	case NF4DIR:
1563 		break;
1564 	default:
1565 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1566 		goto out;
1567 	};
1568 
1569 	if (cs->access == CS_ACCESS_DENIED) {
1570 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1571 		goto out;
1572 	}
1573 	if (dvp->v_type != VDIR) {
1574 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1575 		goto out;
1576 	}
1577 	status = utf8_dir_verify(&args->objname);
1578 	if (status != NFS4_OK) {
1579 		*cs->statusp = resp->status = status;
1580 		goto out;
1581 	}
1582 
1583 	if (rdonly4(req, cs)) {
1584 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1585 		goto out;
1586 	}
1587 
1588 	/*
1589 	 * Name of newly created object
1590 	 */
1591 	nm = utf8_to_fn(&args->objname, &len, NULL);
1592 	if (nm == NULL) {
1593 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1594 		goto out;
1595 	}
1596 
1597 	if (len > MAXNAMELEN) {
1598 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1599 		kmem_free(nm, len);
1600 		goto out;
1601 	}
1602 
1603 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1604 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1605 	    MAXPATHLEN  + 1);
1606 
1607 	if (name == NULL) {
1608 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1609 		kmem_free(nm, len);
1610 		goto out;
1611 	}
1612 
1613 	resp->attrset = 0;
1614 
1615 	sarg.sbp = &sb;
1616 	sarg.is_referral = B_FALSE;
1617 	nfs4_ntov_table_init(&ntov);
1618 
1619 	status = do_rfs4_set_attrs(&resp->attrset,
1620 	    &args->createattrs, cs, &sarg, &ntov, NFS4ATTR_SETIT);
1621 
1622 	if (sarg.vap->va_mask == 0 && status == NFS4_OK)
1623 		status = NFS4ERR_INVAL;
1624 
1625 	if (status != NFS4_OK) {
1626 		*cs->statusp = resp->status = status;
1627 		if (name != nm)
1628 			kmem_free(name, MAXPATHLEN + 1);
1629 		kmem_free(nm, len);
1630 		nfs4_ntov_table_free(&ntov, &sarg);
1631 		resp->attrset = 0;
1632 		goto out;
1633 	}
1634 
1635 	/* Get "before" change value */
1636 	bva.va_mask = AT_CTIME|AT_SEQ|AT_MODE;
1637 	error = VOP_GETATTR(dvp, &bva, 0, cr, NULL);
1638 	if (error) {
1639 		*cs->statusp = resp->status = puterrno4(error);
1640 		if (name != nm)
1641 			kmem_free(name, MAXPATHLEN + 1);
1642 		kmem_free(nm, len);
1643 		nfs4_ntov_table_free(&ntov, &sarg);
1644 		resp->attrset = 0;
1645 		goto out;
1646 	}
1647 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bva.va_ctime)
1648 
1649 	vap = sarg.vap;
1650 
1651 	/*
1652 	 * Set the default initial values for attributes when the parent
1653 	 * directory does not have the VSUID/VSGID bit set and they have
1654 	 * not been specified in createattrs.
1655 	 */
1656 	if (!(bva.va_mode & VSUID) && (vap->va_mask & AT_UID) == 0) {
1657 		vap->va_uid = crgetuid(cr);
1658 		vap->va_mask |= AT_UID;
1659 	}
1660 	if (!(bva.va_mode & VSGID) && (vap->va_mask & AT_GID) == 0) {
1661 		vap->va_gid = crgetgid(cr);
1662 		vap->va_mask |= AT_GID;
1663 	}
1664 
1665 	vap->va_mask |= AT_TYPE;
1666 	switch (args->type) {
1667 	case NF4DIR:
1668 		vap->va_type = VDIR;
1669 		if ((vap->va_mask & AT_MODE) == 0) {
1670 			vap->va_mode = 0700;	/* default: owner rwx only */
1671 			vap->va_mask |= AT_MODE;
1672 		}
1673 		error = VOP_MKDIR(dvp, name, vap, &vp, cr, NULL, 0, NULL);
1674 		if (error)
1675 			break;
1676 
1677 		/*
1678 		 * Get the initial "after" sequence number, if it fails,
1679 		 * set to zero
1680 		 */
1681 		iva.va_mask = AT_SEQ;
1682 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1683 			iva.va_seq = 0;
1684 		break;
1685 	case NF4LNK:
1686 		vap->va_type = VLNK;
1687 		if ((vap->va_mask & AT_MODE) == 0) {
1688 			vap->va_mode = 0700;	/* default: owner rwx only */
1689 			vap->va_mask |= AT_MODE;
1690 		}
1691 
1692 		/*
1693 		 * symlink names must be treated as data
1694 		 */
1695 		lnm = utf8_to_str((utf8string *)&args->ftype4_u.linkdata,
1696 		    &llen, NULL);
1697 
1698 		if (lnm == NULL) {
1699 			*cs->statusp = resp->status = NFS4ERR_INVAL;
1700 			if (name != nm)
1701 				kmem_free(name, MAXPATHLEN + 1);
1702 			kmem_free(nm, len);
1703 			nfs4_ntov_table_free(&ntov, &sarg);
1704 			resp->attrset = 0;
1705 			goto out;
1706 		}
1707 
1708 		if (llen > MAXPATHLEN) {
1709 			*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1710 			if (name != nm)
1711 				kmem_free(name, MAXPATHLEN + 1);
1712 			kmem_free(nm, len);
1713 			kmem_free(lnm, llen);
1714 			nfs4_ntov_table_free(&ntov, &sarg);
1715 			resp->attrset = 0;
1716 			goto out;
1717 		}
1718 
1719 		lname = nfscmd_convname(ca, cs->exi, lnm,
1720 		    NFSCMD_CONV_INBOUND, MAXPATHLEN  + 1);
1721 
1722 		if (lname == NULL) {
1723 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
1724 			if (name != nm)
1725 				kmem_free(name, MAXPATHLEN + 1);
1726 			kmem_free(nm, len);
1727 			kmem_free(lnm, llen);
1728 			nfs4_ntov_table_free(&ntov, &sarg);
1729 			resp->attrset = 0;
1730 			goto out;
1731 		}
1732 
1733 		error = VOP_SYMLINK(dvp, name, vap, lname, cr, NULL, 0);
1734 		if (lname != lnm)
1735 			kmem_free(lname, MAXPATHLEN + 1);
1736 		kmem_free(lnm, llen);
1737 		if (error)
1738 			break;
1739 
1740 		/*
1741 		 * Get the initial "after" sequence number, if it fails,
1742 		 * set to zero
1743 		 */
1744 		iva.va_mask = AT_SEQ;
1745 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1746 			iva.va_seq = 0;
1747 
1748 		error = VOP_LOOKUP(dvp, name, &vp, NULL, 0, NULL, cr,
1749 		    NULL, NULL, NULL);
1750 		if (error)
1751 			break;
1752 
1753 		/*
1754 		 * va_seq is not safe over VOP calls, check it again
1755 		 * if it has changed zero out iva to force atomic = FALSE.
1756 		 */
1757 		iva2.va_mask = AT_SEQ;
1758 		if (VOP_GETATTR(dvp, &iva2, 0, cs->cr, NULL) ||
1759 		    iva2.va_seq != iva.va_seq)
1760 			iva.va_seq = 0;
1761 		break;
1762 	default:
1763 		/*
1764 		 * probably a special file.
1765 		 */
1766 		if ((vap->va_mask & AT_MODE) == 0) {
1767 			vap->va_mode = 0600;	/* default: owner rw only */
1768 			vap->va_mask |= AT_MODE;
1769 		}
1770 		syncval = FNODSYNC;
1771 		/*
1772 		 * We know this will only generate one VOP call
1773 		 */
1774 		vp = do_rfs4_op_mknod(args, resp, req, cs, vap, name);
1775 
1776 		if (vp == NULL) {
1777 			if (name != nm)
1778 				kmem_free(name, MAXPATHLEN + 1);
1779 			kmem_free(nm, len);
1780 			nfs4_ntov_table_free(&ntov, &sarg);
1781 			resp->attrset = 0;
1782 			goto out;
1783 		}
1784 
1785 		/*
1786 		 * Get the initial "after" sequence number, if it fails,
1787 		 * set to zero
1788 		 */
1789 		iva.va_mask = AT_SEQ;
1790 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1791 			iva.va_seq = 0;
1792 
1793 		break;
1794 	}
1795 	if (name != nm)
1796 		kmem_free(name, MAXPATHLEN + 1);
1797 	kmem_free(nm, len);
1798 
1799 	if (error) {
1800 		*cs->statusp = resp->status = puterrno4(error);
1801 	}
1802 
1803 	/*
1804 	 * Force modified data and metadata out to stable storage.
1805 	 */
1806 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
1807 
1808 	if (resp->status != NFS4_OK) {
1809 		if (vp != NULL)
1810 			VN_RELE(vp);
1811 		nfs4_ntov_table_free(&ntov, &sarg);
1812 		resp->attrset = 0;
1813 		goto out;
1814 	}
1815 
1816 	/*
1817 	 * Finish setup of cinfo response, "before" value already set.
1818 	 * Get "after" change value, if it fails, simply return the
1819 	 * before value.
1820 	 */
1821 	ava.va_mask = AT_CTIME|AT_SEQ;
1822 	if (VOP_GETATTR(dvp, &ava, 0, cr, NULL)) {
1823 		ava.va_ctime = bva.va_ctime;
1824 		ava.va_seq = 0;
1825 	}
1826 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, ava.va_ctime);
1827 
1828 	/*
1829 	 * True verification that object was created with correct
1830 	 * attrs is impossible.  The attrs could have been changed
1831 	 * immediately after object creation.  If attributes did
1832 	 * not verify, the only recourse for the server is to
1833 	 * destroy the object.  Maybe if some attrs (like gid)
1834 	 * are set incorrectly, the object should be destroyed;
1835 	 * however, seems bad as a default policy.  Do we really
1836 	 * want to destroy an object over one of the times not
1837 	 * verifying correctly?  For these reasons, the server
1838 	 * currently sets bits in attrset for createattrs
1839 	 * that were set; however, no verification is done.
1840 	 *
1841 	 * vmask_to_nmask accounts for vattr bits set on create
1842 	 *	[do_rfs4_set_attrs() only sets resp bits for
1843 	 *	 non-vattr/vfs bits.]
1844 	 * Mask off any bits set by default so as not to return
1845 	 * more attrset bits than were requested in createattrs
1846 	 */
1847 	nfs4_vmask_to_nmask(sarg.vap->va_mask, &resp->attrset);
1848 	resp->attrset &= args->createattrs.attrmask;
1849 	nfs4_ntov_table_free(&ntov, &sarg);
1850 
1851 	error = makefh4(&cs->fh, vp, cs->exi);
1852 	if (error) {
1853 		*cs->statusp = resp->status = puterrno4(error);
1854 	}
1855 
1856 	/*
1857 	 * The cinfo.atomic = TRUE only if we got no errors, we have
1858 	 * non-zero va_seq's, and it has incremented by exactly one
1859 	 * during the creation and it didn't change during the VOP_LOOKUP
1860 	 * or VOP_FSYNC.
1861 	 */
1862 	if (!error && bva.va_seq && iva.va_seq && ava.va_seq &&
1863 	    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
1864 		resp->cinfo.atomic = TRUE;
1865 	else
1866 		resp->cinfo.atomic = FALSE;
1867 
1868 	/*
1869 	 * Force modified metadata out to stable storage.
1870 	 *
1871 	 * if a underlying vp exists, pass it to VOP_FSYNC
1872 	 */
1873 	if (VOP_REALVP(vp, &realvp, NULL) == 0)
1874 		(void) VOP_FSYNC(realvp, syncval, cr, NULL);
1875 	else
1876 		(void) VOP_FSYNC(vp, syncval, cr, NULL);
1877 
1878 	if (resp->status != NFS4_OK) {
1879 		VN_RELE(vp);
1880 		goto out;
1881 	}
1882 	if (cs->vp)
1883 		VN_RELE(cs->vp);
1884 
1885 	cs->vp = vp;
1886 	*cs->statusp = resp->status = NFS4_OK;
1887 out:
1888 	DTRACE_NFSV4_2(op__create__done, struct compound_state *, cs,
1889 	    CREATE4res *, resp);
1890 }
1891 
1892 /*ARGSUSED*/
1893 static void
1894 rfs4_op_delegpurge(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1895     struct compound_state *cs)
1896 {
1897 	DTRACE_NFSV4_2(op__delegpurge__start, struct compound_state *, cs,
1898 	    DELEGPURGE4args *, &argop->nfs_argop4_u.opdelegpurge);
1899 
1900 	rfs4_op_inval(argop, resop, req, cs);
1901 
1902 	DTRACE_NFSV4_2(op__delegpurge__done, struct compound_state *, cs,
1903 	    DELEGPURGE4res *, &resop->nfs_resop4_u.opdelegpurge);
1904 }
1905 
1906 /*ARGSUSED*/
1907 static void
1908 rfs4_op_delegreturn(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1909     struct compound_state *cs)
1910 {
1911 	DELEGRETURN4args *args = &argop->nfs_argop4_u.opdelegreturn;
1912 	DELEGRETURN4res *resp = &resop->nfs_resop4_u.opdelegreturn;
1913 	rfs4_deleg_state_t *dsp;
1914 	nfsstat4 status;
1915 
1916 	DTRACE_NFSV4_2(op__delegreturn__start, struct compound_state *, cs,
1917 	    DELEGRETURN4args *, args);
1918 
1919 	status = rfs4_get_deleg_state(&args->deleg_stateid, &dsp);
1920 	resp->status = *cs->statusp = status;
1921 	if (status != NFS4_OK)
1922 		goto out;
1923 
1924 	/* Ensure specified filehandle matches */
1925 	if (cs->vp != dsp->rds_finfo->rf_vp) {
1926 		resp->status = *cs->statusp = NFS4ERR_BAD_STATEID;
1927 	} else
1928 		rfs4_return_deleg(dsp, FALSE);
1929 
1930 	rfs4_update_lease(dsp->rds_client);
1931 
1932 	rfs4_deleg_state_rele(dsp);
1933 out:
1934 	DTRACE_NFSV4_2(op__delegreturn__done, struct compound_state *, cs,
1935 	    DELEGRETURN4res *, resp);
1936 }
1937 
1938 /*
1939  * Check to see if a given "flavor" is an explicitly shared flavor.
1940  * The assumption of this routine is the "flavor" is already a valid
1941  * flavor in the secinfo list of "exi".
1942  *
1943  *	e.g.
1944  *		# share -o sec=flavor1 /export
1945  *		# share -o sec=flavor2 /export/home
1946  *
1947  *		flavor2 is not an explicitly shared flavor for /export,
1948  *		however it is in the secinfo list for /export thru the
1949  *		server namespace setup.
1950  */
1951 int
1952 is_exported_sec(int flavor, struct exportinfo *exi)
1953 {
1954 	int	i;
1955 	struct secinfo *sp;
1956 
1957 	sp = exi->exi_export.ex_secinfo;
1958 	for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
1959 		if (flavor == sp[i].s_secinfo.sc_nfsnum ||
1960 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE) {
1961 			return (SEC_REF_EXPORTED(&sp[i]));
1962 		}
1963 	}
1964 
1965 	/* Should not reach this point based on the assumption */
1966 	return (0);
1967 }
1968 
1969 /*
1970  * Check if the security flavor used in the request matches what is
1971  * required at the export point or at the root pseudo node (exi_root).
1972  *
1973  * returns 1 if there's a match or if exported with AUTH_NONE; 0 otherwise.
1974  *
1975  */
1976 static int
1977 secinfo_match_or_authnone(struct compound_state *cs)
1978 {
1979 	int	i;
1980 	struct secinfo *sp;
1981 
1982 	/*
1983 	 * Check cs->nfsflavor (from the request) against
1984 	 * the current export data in cs->exi.
1985 	 */
1986 	sp = cs->exi->exi_export.ex_secinfo;
1987 	for (i = 0; i < cs->exi->exi_export.ex_seccnt; i++) {
1988 		if (cs->nfsflavor == sp[i].s_secinfo.sc_nfsnum ||
1989 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE)
1990 			return (1);
1991 	}
1992 
1993 	return (0);
1994 }
1995 
1996 /*
1997  * Check the access authority for the client and return the correct error.
1998  */
1999 nfsstat4
2000 call_checkauth4(struct compound_state *cs, struct svc_req *req)
2001 {
2002 	int	authres;
2003 
2004 	/*
2005 	 * First, check if the security flavor used in the request
2006 	 * are among the flavors set in the server namespace.
2007 	 */
2008 	if (!secinfo_match_or_authnone(cs)) {
2009 		*cs->statusp = NFS4ERR_WRONGSEC;
2010 		return (*cs->statusp);
2011 	}
2012 
2013 	authres = checkauth4(cs, req);
2014 
2015 	if (authres > 0) {
2016 		*cs->statusp = NFS4_OK;
2017 		if (! (cs->access & CS_ACCESS_LIMITED))
2018 			cs->access = CS_ACCESS_OK;
2019 	} else if (authres == 0) {
2020 		*cs->statusp = NFS4ERR_ACCESS;
2021 	} else if (authres == -2) {
2022 		*cs->statusp = NFS4ERR_WRONGSEC;
2023 	} else {
2024 		*cs->statusp = NFS4ERR_DELAY;
2025 	}
2026 	return (*cs->statusp);
2027 }
2028 
2029 /*
2030  * bitmap4_to_attrmask is called by getattr and readdir.
2031  * It sets up the vattr mask and determines whether vfsstat call is needed
2032  * based on the input bitmap.
2033  * Returns nfsv4 status.
2034  */
2035 static nfsstat4
2036 bitmap4_to_attrmask(bitmap4 breq, struct nfs4_svgetit_arg *sargp)
2037 {
2038 	int i;
2039 	uint_t	va_mask;
2040 	struct statvfs64 *sbp = sargp->sbp;
2041 
2042 	sargp->sbp = NULL;
2043 	sargp->flag = 0;
2044 	sargp->rdattr_error = NFS4_OK;
2045 	sargp->mntdfid_set = FALSE;
2046 	if (sargp->cs->vp)
2047 		sargp->xattr = get_fh4_flag(&sargp->cs->fh,
2048 		    FH4_ATTRDIR | FH4_NAMEDATTR);
2049 	else
2050 		sargp->xattr = 0;
2051 
2052 	/*
2053 	 * Set rdattr_error_req to true if return error per
2054 	 * failed entry rather than fail the readdir.
2055 	 */
2056 	if (breq & FATTR4_RDATTR_ERROR_MASK)
2057 		sargp->rdattr_error_req = 1;
2058 	else
2059 		sargp->rdattr_error_req = 0;
2060 
2061 	/*
2062 	 * generate the va_mask
2063 	 * Handle the easy cases first
2064 	 */
2065 	switch (breq) {
2066 	case NFS4_NTOV_ATTR_MASK:
2067 		sargp->vap->va_mask = NFS4_NTOV_ATTR_AT_MASK;
2068 		return (NFS4_OK);
2069 
2070 	case NFS4_FS_ATTR_MASK:
2071 		sargp->vap->va_mask = NFS4_FS_ATTR_AT_MASK;
2072 		sargp->sbp = sbp;
2073 		return (NFS4_OK);
2074 
2075 	case NFS4_NTOV_ATTR_CACHE_MASK:
2076 		sargp->vap->va_mask = NFS4_NTOV_ATTR_CACHE_AT_MASK;
2077 		return (NFS4_OK);
2078 
2079 	case FATTR4_LEASE_TIME_MASK:
2080 		sargp->vap->va_mask = 0;
2081 		return (NFS4_OK);
2082 
2083 	default:
2084 		va_mask = 0;
2085 		for (i = 0; i < nfs4_ntov_map_size; i++) {
2086 			if ((breq & nfs4_ntov_map[i].fbit) &&
2087 			    nfs4_ntov_map[i].vbit)
2088 				va_mask |= nfs4_ntov_map[i].vbit;
2089 		}
2090 
2091 		/*
2092 		 * Check is vfsstat is needed
2093 		 */
2094 		if (breq & NFS4_FS_ATTR_MASK)
2095 			sargp->sbp = sbp;
2096 
2097 		sargp->vap->va_mask = va_mask;
2098 		return (NFS4_OK);
2099 	}
2100 	/* NOTREACHED */
2101 }
2102 
2103 /*
2104  * bitmap4_get_sysattrs is called by getattr and readdir.
2105  * It calls both VOP_GETATTR and VFS_STATVFS calls to get the attrs.
2106  * Returns nfsv4 status.
2107  */
2108 static nfsstat4
2109 bitmap4_get_sysattrs(struct nfs4_svgetit_arg *sargp)
2110 {
2111 	int error;
2112 	struct compound_state *cs = sargp->cs;
2113 	vnode_t *vp = cs->vp;
2114 
2115 	if (sargp->sbp != NULL) {
2116 		if (error = VFS_STATVFS(vp->v_vfsp, sargp->sbp)) {
2117 			sargp->sbp = NULL;	/* to identify error */
2118 			return (puterrno4(error));
2119 		}
2120 	}
2121 
2122 	return (rfs4_vop_getattr(vp, sargp->vap, 0, cs->cr));
2123 }
2124 
2125 static void
2126 nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp)
2127 {
2128 	ntovp->na = kmem_zalloc(sizeof (union nfs4_attr_u) * nfs4_ntov_map_size,
2129 	    KM_SLEEP);
2130 	ntovp->attrcnt = 0;
2131 	ntovp->vfsstat = FALSE;
2132 }
2133 
2134 static void
2135 nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
2136     struct nfs4_svgetit_arg *sargp)
2137 {
2138 	int i;
2139 	union nfs4_attr_u *na;
2140 	uint8_t *amap;
2141 
2142 	/*
2143 	 * XXX Should do the same checks for whether the bit is set
2144 	 */
2145 	for (i = 0, na = ntovp->na, amap = ntovp->amap;
2146 	    i < ntovp->attrcnt; i++, na++, amap++) {
2147 		(void) (*nfs4_ntov_map[*amap].sv_getit)(
2148 		    NFS4ATTR_FREEIT, sargp, na);
2149 	}
2150 	if ((sargp->op == NFS4ATTR_SETIT) || (sargp->op == NFS4ATTR_VERIT)) {
2151 		/*
2152 		 * xdr_free for getattr will be done later
2153 		 */
2154 		for (i = 0, na = ntovp->na, amap = ntovp->amap;
2155 		    i < ntovp->attrcnt; i++, na++, amap++) {
2156 			xdr_free(nfs4_ntov_map[*amap].xfunc, (caddr_t)na);
2157 		}
2158 	}
2159 	kmem_free(ntovp->na, sizeof (union nfs4_attr_u) * nfs4_ntov_map_size);
2160 }
2161 
2162 /*
2163  * do_rfs4_op_getattr gets the system attrs and converts into fattr4.
2164  */
2165 static nfsstat4
2166 do_rfs4_op_getattr(bitmap4 breq, fattr4 *fattrp,
2167     struct nfs4_svgetit_arg *sargp)
2168 {
2169 	int error = 0;
2170 	int i, k;
2171 	struct nfs4_ntov_table ntov;
2172 	XDR xdr;
2173 	ulong_t xdr_size;
2174 	char *xdr_attrs;
2175 	nfsstat4 status = NFS4_OK;
2176 	nfsstat4 prev_rdattr_error = sargp->rdattr_error;
2177 	union nfs4_attr_u *na;
2178 	uint8_t *amap;
2179 
2180 	sargp->op = NFS4ATTR_GETIT;
2181 	sargp->flag = 0;
2182 
2183 	fattrp->attrmask = 0;
2184 	/* if no bits requested, then return empty fattr4 */
2185 	if (breq == 0) {
2186 		fattrp->attrlist4_len = 0;
2187 		fattrp->attrlist4 = NULL;
2188 		return (NFS4_OK);
2189 	}
2190 
2191 	/*
2192 	 * return NFS4ERR_INVAL when client requests write-only attrs
2193 	 */
2194 	if (breq & (FATTR4_TIME_ACCESS_SET_MASK | FATTR4_TIME_MODIFY_SET_MASK))
2195 		return (NFS4ERR_INVAL);
2196 
2197 	nfs4_ntov_table_init(&ntov);
2198 	na = ntov.na;
2199 	amap = ntov.amap;
2200 
2201 	/*
2202 	 * Now loop to get or verify the attrs
2203 	 */
2204 	for (i = 0; i < nfs4_ntov_map_size; i++) {
2205 		if (breq & nfs4_ntov_map[i].fbit) {
2206 			if ((*nfs4_ntov_map[i].sv_getit)(
2207 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0) {
2208 
2209 				error = (*nfs4_ntov_map[i].sv_getit)(
2210 				    NFS4ATTR_GETIT, sargp, na);
2211 
2212 				/*
2213 				 * Possible error values:
2214 				 * >0 if sv_getit failed to
2215 				 * get the attr; 0 if succeeded;
2216 				 * <0 if rdattr_error and the
2217 				 * attribute cannot be returned.
2218 				 */
2219 				if (error && !(sargp->rdattr_error_req))
2220 					goto done;
2221 				/*
2222 				 * If error then just for entry
2223 				 */
2224 				if (error == 0) {
2225 					fattrp->attrmask |=
2226 					    nfs4_ntov_map[i].fbit;
2227 					*amap++ =
2228 					    (uint8_t)nfs4_ntov_map[i].nval;
2229 					na++;
2230 					(ntov.attrcnt)++;
2231 				} else if ((error > 0) &&
2232 				    (sargp->rdattr_error == NFS4_OK)) {
2233 					sargp->rdattr_error = puterrno4(error);
2234 				}
2235 				error = 0;
2236 			}
2237 		}
2238 	}
2239 
2240 	/*
2241 	 * If rdattr_error was set after the return value for it was assigned,
2242 	 * update it.
2243 	 */
2244 	if (prev_rdattr_error != sargp->rdattr_error) {
2245 		na = ntov.na;
2246 		amap = ntov.amap;
2247 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2248 			k = *amap;
2249 			if (k < FATTR4_RDATTR_ERROR) {
2250 				continue;
2251 			}
2252 			if ((k == FATTR4_RDATTR_ERROR) &&
2253 			    ((*nfs4_ntov_map[k].sv_getit)(
2254 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0)) {
2255 
2256 				(void) (*nfs4_ntov_map[k].sv_getit)(
2257 				    NFS4ATTR_GETIT, sargp, na);
2258 			}
2259 			break;
2260 		}
2261 	}
2262 
2263 	xdr_size = 0;
2264 	na = ntov.na;
2265 	amap = ntov.amap;
2266 	for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2267 		xdr_size += xdr_sizeof(nfs4_ntov_map[*amap].xfunc, na);
2268 	}
2269 
2270 	fattrp->attrlist4_len = xdr_size;
2271 	if (xdr_size) {
2272 		/* freed by rfs4_op_getattr_free() */
2273 		fattrp->attrlist4 = xdr_attrs = kmem_zalloc(xdr_size, KM_SLEEP);
2274 
2275 		xdrmem_create(&xdr, xdr_attrs, xdr_size, XDR_ENCODE);
2276 
2277 		na = ntov.na;
2278 		amap = ntov.amap;
2279 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2280 			if (!(*nfs4_ntov_map[*amap].xfunc)(&xdr, na)) {
2281 				DTRACE_PROBE1(nfss__e__getattr4_encfail,
2282 				    int, *amap);
2283 				status = NFS4ERR_SERVERFAULT;
2284 				break;
2285 			}
2286 		}
2287 		/* xdrmem_destroy(&xdrs); */	/* NO-OP */
2288 	} else {
2289 		fattrp->attrlist4 = NULL;
2290 	}
2291 done:
2292 
2293 	nfs4_ntov_table_free(&ntov, sargp);
2294 
2295 	if (error != 0)
2296 		status = puterrno4(error);
2297 
2298 	return (status);
2299 }
2300 
2301 /* ARGSUSED */
2302 static void
2303 rfs4_op_getattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2304     struct compound_state *cs)
2305 {
2306 	GETATTR4args *args = &argop->nfs_argop4_u.opgetattr;
2307 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2308 	struct nfs4_svgetit_arg sarg;
2309 	struct statvfs64 sb;
2310 	nfsstat4 status;
2311 
2312 	DTRACE_NFSV4_2(op__getattr__start, struct compound_state *, cs,
2313 	    GETATTR4args *, args);
2314 
2315 	if (cs->vp == NULL) {
2316 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2317 		goto out;
2318 	}
2319 
2320 	if (cs->access == CS_ACCESS_DENIED) {
2321 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2322 		goto out;
2323 	}
2324 
2325 	sarg.sbp = &sb;
2326 	sarg.cs = cs;
2327 	sarg.is_referral = B_FALSE;
2328 
2329 	status = bitmap4_to_attrmask(args->attr_request, &sarg);
2330 	if (status == NFS4_OK) {
2331 
2332 		status = bitmap4_get_sysattrs(&sarg);
2333 		if (status == NFS4_OK) {
2334 
2335 			/* Is this a referral? */
2336 			if (vn_is_nfs_reparse(cs->vp, cs->cr)) {
2337 				/* Older V4 Solaris client sees a link */
2338 				if (client_is_downrev(req))
2339 					sarg.vap->va_type = VLNK;
2340 				else
2341 					sarg.is_referral = B_TRUE;
2342 			}
2343 
2344 			status = do_rfs4_op_getattr(args->attr_request,
2345 			    &resp->obj_attributes, &sarg);
2346 		}
2347 	}
2348 	*cs->statusp = resp->status = status;
2349 out:
2350 	DTRACE_NFSV4_2(op__getattr__done, struct compound_state *, cs,
2351 	    GETATTR4res *, resp);
2352 }
2353 
2354 static void
2355 rfs4_op_getattr_free(nfs_resop4 *resop)
2356 {
2357 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2358 
2359 	nfs4_fattr4_free(&resp->obj_attributes);
2360 }
2361 
2362 /* ARGSUSED */
2363 static void
2364 rfs4_op_getfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2365     struct compound_state *cs)
2366 {
2367 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2368 
2369 	DTRACE_NFSV4_1(op__getfh__start, struct compound_state *, cs);
2370 
2371 	if (cs->vp == NULL) {
2372 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2373 		goto out;
2374 	}
2375 	if (cs->access == CS_ACCESS_DENIED) {
2376 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2377 		goto out;
2378 	}
2379 
2380 	/* check for reparse point at the share point */
2381 	if (cs->exi->exi_moved || vn_is_nfs_reparse(cs->exi->exi_vp, cs->cr)) {
2382 		/* it's all bad */
2383 		cs->exi->exi_moved = 1;
2384 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2385 		DTRACE_PROBE2(nfs4serv__func__referral__shared__moved,
2386 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2387 		return;
2388 	}
2389 
2390 	/* check for reparse point at vp */
2391 	if (vn_is_nfs_reparse(cs->vp, cs->cr) && !client_is_downrev(req)) {
2392 		/* it's not all bad */
2393 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2394 		DTRACE_PROBE2(nfs4serv__func__referral__moved,
2395 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2396 		return;
2397 	}
2398 
2399 	resp->object.nfs_fh4_val =
2400 	    kmem_alloc(cs->fh.nfs_fh4_len, KM_SLEEP);
2401 	nfs_fh4_copy(&cs->fh, &resp->object);
2402 	*cs->statusp = resp->status = NFS4_OK;
2403 out:
2404 	DTRACE_NFSV4_2(op__getfh__done, struct compound_state *, cs,
2405 	    GETFH4res *, resp);
2406 }
2407 
2408 static void
2409 rfs4_op_getfh_free(nfs_resop4 *resop)
2410 {
2411 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2412 
2413 	if (resp->status == NFS4_OK &&
2414 	    resp->object.nfs_fh4_val != NULL) {
2415 		kmem_free(resp->object.nfs_fh4_val, resp->object.nfs_fh4_len);
2416 		resp->object.nfs_fh4_val = NULL;
2417 		resp->object.nfs_fh4_len = 0;
2418 	}
2419 }
2420 
2421 /*
2422  * illegal: args: void
2423  *	    res : status (NFS4ERR_OP_ILLEGAL)
2424  */
2425 /* ARGSUSED */
2426 static void
2427 rfs4_op_illegal(nfs_argop4 *argop, nfs_resop4 *resop,
2428     struct svc_req *req, struct compound_state *cs)
2429 {
2430 	ILLEGAL4res *resp = &resop->nfs_resop4_u.opillegal;
2431 
2432 	resop->resop = OP_ILLEGAL;
2433 	*cs->statusp = resp->status = NFS4ERR_OP_ILLEGAL;
2434 }
2435 
2436 /*
2437  * link: args: SAVED_FH: file, CURRENT_FH: target directory
2438  *	 res: status. If success - CURRENT_FH unchanged, return change_info
2439  */
2440 /* ARGSUSED */
2441 static void
2442 rfs4_op_link(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2443     struct compound_state *cs)
2444 {
2445 	LINK4args *args = &argop->nfs_argop4_u.oplink;
2446 	LINK4res *resp = &resop->nfs_resop4_u.oplink;
2447 	int error;
2448 	vnode_t *vp;
2449 	vnode_t *dvp;
2450 	struct vattr bdva, idva, adva;
2451 	char *nm;
2452 	uint_t  len;
2453 	struct sockaddr *ca;
2454 	char *name = NULL;
2455 	nfsstat4 status;
2456 
2457 	DTRACE_NFSV4_2(op__link__start, struct compound_state *, cs,
2458 	    LINK4args *, args);
2459 
2460 	/* SAVED_FH: source object */
2461 	vp = cs->saved_vp;
2462 	if (vp == NULL) {
2463 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2464 		goto out;
2465 	}
2466 
2467 	/* CURRENT_FH: target directory */
2468 	dvp = cs->vp;
2469 	if (dvp == NULL) {
2470 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2471 		goto out;
2472 	}
2473 
2474 	/*
2475 	 * If there is a non-shared filesystem mounted on this vnode,
2476 	 * do not allow to link any file in this directory.
2477 	 */
2478 	if (vn_ismntpt(dvp)) {
2479 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2480 		goto out;
2481 	}
2482 
2483 	if (cs->access == CS_ACCESS_DENIED) {
2484 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2485 		goto out;
2486 	}
2487 
2488 	/* Check source object's type validity */
2489 	if (vp->v_type == VDIR) {
2490 		*cs->statusp = resp->status = NFS4ERR_ISDIR;
2491 		goto out;
2492 	}
2493 
2494 	/* Check target directory's type */
2495 	if (dvp->v_type != VDIR) {
2496 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2497 		goto out;
2498 	}
2499 
2500 	if (cs->saved_exi != cs->exi) {
2501 		*cs->statusp = resp->status = NFS4ERR_XDEV;
2502 		goto out;
2503 	}
2504 
2505 	status = utf8_dir_verify(&args->newname);
2506 	if (status != NFS4_OK) {
2507 		*cs->statusp = resp->status = status;
2508 		goto out;
2509 	}
2510 
2511 	nm = utf8_to_fn(&args->newname, &len, NULL);
2512 	if (nm == NULL) {
2513 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2514 		goto out;
2515 	}
2516 
2517 	if (len > MAXNAMELEN) {
2518 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2519 		kmem_free(nm, len);
2520 		goto out;
2521 	}
2522 
2523 	if (rdonly4(req, cs)) {
2524 		*cs->statusp = resp->status = NFS4ERR_ROFS;
2525 		kmem_free(nm, len);
2526 		goto out;
2527 	}
2528 
2529 	/* Get "before" change value */
2530 	bdva.va_mask = AT_CTIME|AT_SEQ;
2531 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
2532 	if (error) {
2533 		*cs->statusp = resp->status = puterrno4(error);
2534 		kmem_free(nm, len);
2535 		goto out;
2536 	}
2537 
2538 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2539 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2540 	    MAXPATHLEN  + 1);
2541 
2542 	if (name == NULL) {
2543 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2544 		kmem_free(nm, len);
2545 		goto out;
2546 	}
2547 
2548 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
2549 
2550 	error = VOP_LINK(dvp, vp, name, cs->cr, NULL, 0);
2551 
2552 	if (nm != name)
2553 		kmem_free(name, MAXPATHLEN + 1);
2554 	kmem_free(nm, len);
2555 
2556 	/*
2557 	 * Get the initial "after" sequence number, if it fails, set to zero
2558 	 */
2559 	idva.va_mask = AT_SEQ;
2560 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
2561 		idva.va_seq = 0;
2562 
2563 	/*
2564 	 * Force modified data and metadata out to stable storage.
2565 	 */
2566 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
2567 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
2568 
2569 	if (error) {
2570 		*cs->statusp = resp->status = puterrno4(error);
2571 		goto out;
2572 	}
2573 
2574 	/*
2575 	 * Get "after" change value, if it fails, simply return the
2576 	 * before value.
2577 	 */
2578 	adva.va_mask = AT_CTIME|AT_SEQ;
2579 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
2580 		adva.va_ctime = bdva.va_ctime;
2581 		adva.va_seq = 0;
2582 	}
2583 
2584 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
2585 
2586 	/*
2587 	 * The cinfo.atomic = TRUE only if we have
2588 	 * non-zero va_seq's, and it has incremented by exactly one
2589 	 * during the VOP_LINK and it didn't change during the VOP_FSYNC.
2590 	 */
2591 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
2592 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
2593 		resp->cinfo.atomic = TRUE;
2594 	else
2595 		resp->cinfo.atomic = FALSE;
2596 
2597 	*cs->statusp = resp->status = NFS4_OK;
2598 out:
2599 	DTRACE_NFSV4_2(op__link__done, struct compound_state *, cs,
2600 	    LINK4res *, resp);
2601 }
2602 
2603 /*
2604  * Used by rfs4_op_lookup and rfs4_op_lookupp to do the actual work.
2605  */
2606 
2607 /* ARGSUSED */
2608 static nfsstat4
2609 do_rfs4_op_lookup(char *nm, struct svc_req *req, struct compound_state *cs)
2610 {
2611 	int error;
2612 	int different_export = 0;
2613 	vnode_t *vp, *pre_tvp = NULL, *oldvp = NULL;
2614 	struct exportinfo *exi = NULL, *pre_exi = NULL;
2615 	nfsstat4 stat;
2616 	fid_t fid;
2617 	int attrdir, dotdot, walk;
2618 	bool_t is_newvp = FALSE;
2619 
2620 	if (cs->vp->v_flag & V_XATTRDIR) {
2621 		attrdir = 1;
2622 		ASSERT(get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2623 	} else {
2624 		attrdir = 0;
2625 		ASSERT(! get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2626 	}
2627 
2628 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
2629 
2630 	/*
2631 	 * If dotdotting, then need to check whether it's
2632 	 * above the root of a filesystem, or above an
2633 	 * export point.
2634 	 */
2635 	if (dotdot) {
2636 
2637 		/*
2638 		 * If dotdotting at the root of a filesystem, then
2639 		 * need to traverse back to the mounted-on filesystem
2640 		 * and do the dotdot lookup there.
2641 		 */
2642 		if (cs->vp->v_flag & VROOT) {
2643 
2644 			/*
2645 			 * If at the system root, then can
2646 			 * go up no further.
2647 			 */
2648 			if (VN_CMP(cs->vp, rootdir))
2649 				return (puterrno4(ENOENT));
2650 
2651 			/*
2652 			 * Traverse back to the mounted-on filesystem
2653 			 */
2654 			cs->vp = untraverse(cs->vp);
2655 
2656 			/*
2657 			 * Set the different_export flag so we remember
2658 			 * to pick up a new exportinfo entry for
2659 			 * this new filesystem.
2660 			 */
2661 			different_export = 1;
2662 		} else {
2663 
2664 			/*
2665 			 * If dotdotting above an export point then set
2666 			 * the different_export to get new export info.
2667 			 */
2668 			different_export = nfs_exported(cs->exi, cs->vp);
2669 		}
2670 	}
2671 
2672 	error = VOP_LOOKUP(cs->vp, nm, &vp, NULL, 0, NULL, cs->cr,
2673 	    NULL, NULL, NULL);
2674 	if (error)
2675 		return (puterrno4(error));
2676 
2677 	/*
2678 	 * If the vnode is in a pseudo filesystem, check whether it is visible.
2679 	 *
2680 	 * XXX if the vnode is a symlink and it is not visible in
2681 	 * a pseudo filesystem, return ENOENT (not following symlink).
2682 	 * V4 client can not mount such symlink. This is a regression
2683 	 * from V2/V3.
2684 	 *
2685 	 * In the same exported filesystem, if the security flavor used
2686 	 * is not an explicitly shared flavor, limit the view to the visible
2687 	 * list entries only. This is not a WRONGSEC case because it's already
2688 	 * checked via PUTROOTFH/PUTPUBFH or PUTFH.
2689 	 */
2690 	if (!different_export &&
2691 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
2692 	    cs->access & CS_ACCESS_LIMITED)) {
2693 		if (! nfs_visible(cs->exi, vp, &different_export)) {
2694 			VN_RELE(vp);
2695 			return (puterrno4(ENOENT));
2696 		}
2697 	}
2698 
2699 	/*
2700 	 * If it's a mountpoint, then traverse it.
2701 	 */
2702 	if (vn_ismntpt(vp)) {
2703 		pre_exi = cs->exi;	/* save pre-traversed exportinfo */
2704 		pre_tvp = vp;		/* save pre-traversed vnode	*/
2705 
2706 		/*
2707 		 * hold pre_tvp to counteract rele by traverse.  We will
2708 		 * need pre_tvp below if checkexport4 fails
2709 		 */
2710 		VN_HOLD(pre_tvp);
2711 		if ((error = traverse(&vp)) != 0) {
2712 			VN_RELE(vp);
2713 			VN_RELE(pre_tvp);
2714 			return (puterrno4(error));
2715 		}
2716 		different_export = 1;
2717 	} else if (vp->v_vfsp != cs->vp->v_vfsp) {
2718 		/*
2719 		 * The vfsp comparison is to handle the case where
2720 		 * a LOFS mount is shared.  lo_lookup traverses mount points,
2721 		 * and NFS is unaware of local fs transistions because
2722 		 * v_vfsmountedhere isn't set.  For this special LOFS case,
2723 		 * the dir and the obj returned by lookup will have different
2724 		 * vfs ptrs.
2725 		 */
2726 		different_export = 1;
2727 	}
2728 
2729 	if (different_export) {
2730 
2731 		bzero(&fid, sizeof (fid));
2732 		fid.fid_len = MAXFIDSZ;
2733 		error = vop_fid_pseudo(vp, &fid);
2734 		if (error) {
2735 			VN_RELE(vp);
2736 			if (pre_tvp)
2737 				VN_RELE(pre_tvp);
2738 			return (puterrno4(error));
2739 		}
2740 
2741 		if (dotdot)
2742 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
2743 		else
2744 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2745 
2746 		if (exi == NULL) {
2747 			if (pre_tvp) {
2748 				/*
2749 				 * If this vnode is a mounted-on vnode,
2750 				 * but the mounted-on file system is not
2751 				 * exported, send back the filehandle for
2752 				 * the mounted-on vnode, not the root of
2753 				 * the mounted-on file system.
2754 				 */
2755 				VN_RELE(vp);
2756 				vp = pre_tvp;
2757 				exi = pre_exi;
2758 			} else {
2759 				VN_RELE(vp);
2760 				return (puterrno4(EACCES));
2761 			}
2762 		} else if (pre_tvp) {
2763 			/* we're done with pre_tvp now. release extra hold */
2764 			VN_RELE(pre_tvp);
2765 		}
2766 
2767 		cs->exi = exi;
2768 
2769 		/*
2770 		 * Now we do a checkauth4. The reason is that
2771 		 * this client/user may not have access to the new
2772 		 * exported file system, and if they do,
2773 		 * the client/user may be mapped to a different uid.
2774 		 *
2775 		 * We start with a new cr, because the checkauth4 done
2776 		 * in the PUT*FH operation over wrote the cred's uid,
2777 		 * gid, etc, and we want the real thing before calling
2778 		 * checkauth4()
2779 		 */
2780 		crfree(cs->cr);
2781 		cs->cr = crdup(cs->basecr);
2782 
2783 		oldvp = cs->vp;
2784 		cs->vp = vp;
2785 		is_newvp = TRUE;
2786 
2787 		stat = call_checkauth4(cs, req);
2788 		if (stat != NFS4_OK) {
2789 			VN_RELE(cs->vp);
2790 			cs->vp = oldvp;
2791 			return (stat);
2792 		}
2793 	}
2794 
2795 	/*
2796 	 * After various NFS checks, do a label check on the path
2797 	 * component. The label on this path should either be the
2798 	 * global zone's label or a zone's label. We are only
2799 	 * interested in the zone's label because exported files
2800 	 * in global zone is accessible (though read-only) to
2801 	 * clients. The exportability/visibility check is already
2802 	 * done before reaching this code.
2803 	 */
2804 	if (is_system_labeled()) {
2805 		bslabel_t *clabel;
2806 
2807 		ASSERT(req->rq_label != NULL);
2808 		clabel = req->rq_label;
2809 		DTRACE_PROBE2(tx__rfs4__log__info__oplookup__clabel, char *,
2810 		    "got client label from request(1)", struct svc_req *, req);
2811 
2812 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
2813 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
2814 			    cs->exi)) {
2815 				error = EACCES;
2816 				goto err_out;
2817 			}
2818 		} else {
2819 			/*
2820 			 * We grant access to admin_low label clients
2821 			 * only if the client is trusted, i.e. also
2822 			 * running Solaris Trusted Extension.
2823 			 */
2824 			struct sockaddr	*ca;
2825 			int		addr_type;
2826 			void		*ipaddr;
2827 			tsol_tpc_t	*tp;
2828 
2829 			ca = (struct sockaddr *)svc_getrpccaller(
2830 			    req->rq_xprt)->buf;
2831 			if (ca->sa_family == AF_INET) {
2832 				addr_type = IPV4_VERSION;
2833 				ipaddr = &((struct sockaddr_in *)ca)->sin_addr;
2834 			} else if (ca->sa_family == AF_INET6) {
2835 				addr_type = IPV6_VERSION;
2836 				ipaddr = &((struct sockaddr_in6 *)
2837 				    ca)->sin6_addr;
2838 			}
2839 			tp = find_tpc(ipaddr, addr_type, B_FALSE);
2840 			if (tp == NULL || tp->tpc_tp.tp_doi !=
2841 			    l_admin_low->tsl_doi || tp->tpc_tp.host_type !=
2842 			    SUN_CIPSO) {
2843 				if (tp != NULL)
2844 					TPC_RELE(tp);
2845 				error = EACCES;
2846 				goto err_out;
2847 			}
2848 			TPC_RELE(tp);
2849 		}
2850 	}
2851 
2852 	error = makefh4(&cs->fh, vp, cs->exi);
2853 
2854 err_out:
2855 	if (error) {
2856 		if (is_newvp) {
2857 			VN_RELE(cs->vp);
2858 			cs->vp = oldvp;
2859 		} else
2860 			VN_RELE(vp);
2861 		return (puterrno4(error));
2862 	}
2863 
2864 	if (!is_newvp) {
2865 		if (cs->vp)
2866 			VN_RELE(cs->vp);
2867 		cs->vp = vp;
2868 	} else if (oldvp)
2869 		VN_RELE(oldvp);
2870 
2871 	/*
2872 	 * if did lookup on attrdir and didn't lookup .., set named
2873 	 * attr fh flag
2874 	 */
2875 	if (attrdir && ! dotdot)
2876 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
2877 
2878 	/* Assume false for now, open proc will set this */
2879 	cs->mandlock = FALSE;
2880 
2881 	return (NFS4_OK);
2882 }
2883 
2884 /* ARGSUSED */
2885 static void
2886 rfs4_op_lookup(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2887     struct compound_state *cs)
2888 {
2889 	LOOKUP4args *args = &argop->nfs_argop4_u.oplookup;
2890 	LOOKUP4res *resp = &resop->nfs_resop4_u.oplookup;
2891 	char *nm;
2892 	uint_t len;
2893 	struct sockaddr *ca;
2894 	char *name = NULL;
2895 	nfsstat4 status;
2896 
2897 	DTRACE_NFSV4_2(op__lookup__start, struct compound_state *, cs,
2898 	    LOOKUP4args *, args);
2899 
2900 	if (cs->vp == NULL) {
2901 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2902 		goto out;
2903 	}
2904 
2905 	if (cs->vp->v_type == VLNK) {
2906 		*cs->statusp = resp->status = NFS4ERR_SYMLINK;
2907 		goto out;
2908 	}
2909 
2910 	if (cs->vp->v_type != VDIR) {
2911 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2912 		goto out;
2913 	}
2914 
2915 	status = utf8_dir_verify(&args->objname);
2916 	if (status != NFS4_OK) {
2917 		*cs->statusp = resp->status = status;
2918 		goto out;
2919 	}
2920 
2921 	nm = utf8_to_str(&args->objname, &len, NULL);
2922 	if (nm == NULL) {
2923 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2924 		goto out;
2925 	}
2926 
2927 	if (len > MAXNAMELEN) {
2928 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2929 		kmem_free(nm, len);
2930 		goto out;
2931 	}
2932 
2933 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2934 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2935 	    MAXPATHLEN  + 1);
2936 
2937 	if (name == NULL) {
2938 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2939 		kmem_free(nm, len);
2940 		goto out;
2941 	}
2942 
2943 	*cs->statusp = resp->status = do_rfs4_op_lookup(name, req, cs);
2944 
2945 	if (name != nm)
2946 		kmem_free(name, MAXPATHLEN + 1);
2947 	kmem_free(nm, len);
2948 
2949 out:
2950 	DTRACE_NFSV4_2(op__lookup__done, struct compound_state *, cs,
2951 	    LOOKUP4res *, resp);
2952 }
2953 
2954 /* ARGSUSED */
2955 static void
2956 rfs4_op_lookupp(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
2957     struct compound_state *cs)
2958 {
2959 	LOOKUPP4res *resp = &resop->nfs_resop4_u.oplookupp;
2960 
2961 	DTRACE_NFSV4_1(op__lookupp__start, struct compound_state *, cs);
2962 
2963 	if (cs->vp == NULL) {
2964 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2965 		goto out;
2966 	}
2967 
2968 	if (cs->vp->v_type != VDIR) {
2969 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2970 		goto out;
2971 	}
2972 
2973 	*cs->statusp = resp->status = do_rfs4_op_lookup("..", req, cs);
2974 
2975 	/*
2976 	 * From NFSV4 Specification, LOOKUPP should not check for
2977 	 * NFS4ERR_WRONGSEC. Retrun NFS4_OK instead.
2978 	 */
2979 	if (resp->status == NFS4ERR_WRONGSEC) {
2980 		*cs->statusp = resp->status = NFS4_OK;
2981 	}
2982 
2983 out:
2984 	DTRACE_NFSV4_2(op__lookupp__done, struct compound_state *, cs,
2985 	    LOOKUPP4res *, resp);
2986 }
2987 
2988 
2989 /*ARGSUSED2*/
2990 static void
2991 rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2992     struct compound_state *cs)
2993 {
2994 	OPENATTR4args	*args = &argop->nfs_argop4_u.opopenattr;
2995 	OPENATTR4res	*resp = &resop->nfs_resop4_u.opopenattr;
2996 	vnode_t		*avp = NULL;
2997 	int		lookup_flags = LOOKUP_XATTR, error;
2998 	int		exp_ro = 0;
2999 
3000 	DTRACE_NFSV4_2(op__openattr__start, struct compound_state *, cs,
3001 	    OPENATTR4args *, args);
3002 
3003 	if (cs->vp == NULL) {
3004 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3005 		goto out;
3006 	}
3007 
3008 	if ((cs->vp->v_vfsp->vfs_flag & VFS_XATTR) == 0 &&
3009 	    !vfs_has_feature(cs->vp->v_vfsp, VFSFT_SYSATTR_VIEWS)) {
3010 		*cs->statusp = resp->status = puterrno4(ENOTSUP);
3011 		goto out;
3012 	}
3013 
3014 	/*
3015 	 * If file system supports passing ACE mask to VOP_ACCESS then
3016 	 * check for ACE_READ_NAMED_ATTRS, otherwise do legacy checks
3017 	 */
3018 
3019 	if (vfs_has_feature(cs->vp->v_vfsp, VFSFT_ACEMASKONACCESS))
3020 		error = VOP_ACCESS(cs->vp, ACE_READ_NAMED_ATTRS,
3021 		    V_ACE_MASK, cs->cr, NULL);
3022 	else
3023 		error = ((VOP_ACCESS(cs->vp, VREAD, 0, cs->cr, NULL) != 0) &&
3024 		    (VOP_ACCESS(cs->vp, VWRITE, 0, cs->cr, NULL) != 0) &&
3025 		    (VOP_ACCESS(cs->vp, VEXEC, 0, cs->cr, NULL) != 0));
3026 
3027 	if (error) {
3028 		*cs->statusp = resp->status = puterrno4(EACCES);
3029 		goto out;
3030 	}
3031 
3032 	/*
3033 	 * The CREATE_XATTR_DIR VOP flag cannot be specified if
3034 	 * the file system is exported read-only -- regardless of
3035 	 * createdir flag.  Otherwise the attrdir would be created
3036 	 * (assuming server fs isn't mounted readonly locally).  If
3037 	 * VOP_LOOKUP returns ENOENT in this case, the error will
3038 	 * be translated into EROFS.  ENOSYS is mapped to ENOTSUP
3039 	 * because specfs has no VOP_LOOKUP op, so the macro would
3040 	 * return ENOSYS.  EINVAL is returned by all (current)
3041 	 * Solaris file system implementations when any of their
3042 	 * restrictions are violated (xattr(dir) can't have xattrdir).
3043 	 * Returning NOTSUPP is more appropriate in this case
3044 	 * because the object will never be able to have an attrdir.
3045 	 */
3046 	if (args->createdir && ! (exp_ro = rdonly4(req, cs)))
3047 		lookup_flags |= CREATE_XATTR_DIR;
3048 
3049 	error = VOP_LOOKUP(cs->vp, "", &avp, NULL, lookup_flags, NULL, cs->cr,
3050 	    NULL, NULL, NULL);
3051 
3052 	if (error) {
3053 		if (error == ENOENT && args->createdir && exp_ro)
3054 			*cs->statusp = resp->status = puterrno4(EROFS);
3055 		else if (error == EINVAL || error == ENOSYS)
3056 			*cs->statusp = resp->status = puterrno4(ENOTSUP);
3057 		else
3058 			*cs->statusp = resp->status = puterrno4(error);
3059 		goto out;
3060 	}
3061 
3062 	ASSERT(avp->v_flag & V_XATTRDIR);
3063 
3064 	error = makefh4(&cs->fh, avp, cs->exi);
3065 
3066 	if (error) {
3067 		VN_RELE(avp);
3068 		*cs->statusp = resp->status = puterrno4(error);
3069 		goto out;
3070 	}
3071 
3072 	VN_RELE(cs->vp);
3073 	cs->vp = avp;
3074 
3075 	/*
3076 	 * There is no requirement for an attrdir fh flag
3077 	 * because the attrdir has a vnode flag to distinguish
3078 	 * it from regular (non-xattr) directories.  The
3079 	 * FH4_ATTRDIR flag is set for future sanity checks.
3080 	 */
3081 	set_fh4_flag(&cs->fh, FH4_ATTRDIR);
3082 	*cs->statusp = resp->status = NFS4_OK;
3083 
3084 out:
3085 	DTRACE_NFSV4_2(op__openattr__done, struct compound_state *, cs,
3086 	    OPENATTR4res *, resp);
3087 }
3088 
3089 static int
3090 do_io(int direction, vnode_t *vp, struct uio *uio, int ioflag, cred_t *cred,
3091     caller_context_t *ct)
3092 {
3093 	int error;
3094 	int i;
3095 	clock_t delaytime;
3096 
3097 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
3098 
3099 	/*
3100 	 * Don't block on mandatory locks. If this routine returns
3101 	 * EAGAIN, the caller should return NFS4ERR_LOCKED.
3102 	 */
3103 	uio->uio_fmode = FNONBLOCK;
3104 
3105 	for (i = 0; i < rfs4_maxlock_tries; i++) {
3106 
3107 
3108 		if (direction == FREAD) {
3109 			(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, ct);
3110 			error = VOP_READ(vp, uio, ioflag, cred, ct);
3111 			VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, ct);
3112 		} else {
3113 			(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, ct);
3114 			error = VOP_WRITE(vp, uio, ioflag, cred, ct);
3115 			VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, ct);
3116 		}
3117 
3118 		if (error != EAGAIN)
3119 			break;
3120 
3121 		if (i < rfs4_maxlock_tries - 1) {
3122 			delay(delaytime);
3123 			delaytime *= 2;
3124 		}
3125 	}
3126 
3127 	return (error);
3128 }
3129 
3130 /* ARGSUSED */
3131 static void
3132 rfs4_op_read(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3133     struct compound_state *cs)
3134 {
3135 	READ4args *args = &argop->nfs_argop4_u.opread;
3136 	READ4res *resp = &resop->nfs_resop4_u.opread;
3137 	int error;
3138 	int verror;
3139 	vnode_t *vp;
3140 	struct vattr va;
3141 	struct iovec iov, *iovp = NULL;
3142 	int iovcnt;
3143 	struct uio uio;
3144 	u_offset_t offset;
3145 	bool_t *deleg = &cs->deleg;
3146 	nfsstat4 stat;
3147 	int in_crit = 0;
3148 	mblk_t *mp = NULL;
3149 	int alloc_err = 0;
3150 	int rdma_used = 0;
3151 	int loaned_buffers;
3152 	caller_context_t ct;
3153 	struct uio *uiop;
3154 
3155 	DTRACE_NFSV4_2(op__read__start, struct compound_state *, cs,
3156 	    READ4args, args);
3157 
3158 	vp = cs->vp;
3159 	if (vp == NULL) {
3160 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3161 		goto out;
3162 	}
3163 	if (cs->access == CS_ACCESS_DENIED) {
3164 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3165 		goto out;
3166 	}
3167 
3168 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3169 	    deleg, TRUE, &ct)) != NFS4_OK) {
3170 		*cs->statusp = resp->status = stat;
3171 		goto out;
3172 	}
3173 
3174 	/*
3175 	 * Enter the critical region before calling VOP_RWLOCK
3176 	 * to avoid a deadlock with write requests.
3177 	 */
3178 	if (nbl_need_check(vp)) {
3179 		nbl_start_crit(vp, RW_READER);
3180 		in_crit = 1;
3181 		if (nbl_conflict(vp, NBL_READ, args->offset, args->count, 0,
3182 		    &ct)) {
3183 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
3184 			goto out;
3185 		}
3186 	}
3187 
3188 	if (args->wlist) {
3189 		if (args->count > clist_len(args->wlist)) {
3190 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3191 			goto out;
3192 		}
3193 		rdma_used = 1;
3194 	}
3195 
3196 	/* use loaned buffers for TCP */
3197 	loaned_buffers = (nfs_loaned_buffers && !rdma_used) ? 1 : 0;
3198 
3199 	va.va_mask = AT_MODE|AT_SIZE|AT_UID;
3200 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3201 
3202 	/*
3203 	 * If we can't get the attributes, then we can't do the
3204 	 * right access checking.  So, we'll fail the request.
3205 	 */
3206 	if (verror) {
3207 		*cs->statusp = resp->status = puterrno4(verror);
3208 		goto out;
3209 	}
3210 
3211 	if (vp->v_type != VREG) {
3212 		*cs->statusp = resp->status =
3213 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
3214 		goto out;
3215 	}
3216 
3217 	if (crgetuid(cs->cr) != va.va_uid &&
3218 	    (error = VOP_ACCESS(vp, VREAD, 0, cs->cr, &ct)) &&
3219 	    (error = VOP_ACCESS(vp, VEXEC, 0, cs->cr, &ct))) {
3220 		*cs->statusp = resp->status = puterrno4(error);
3221 		goto out;
3222 	}
3223 
3224 	if (MANDLOCK(vp, va.va_mode)) { /* XXX - V4 supports mand locking */
3225 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3226 		goto out;
3227 	}
3228 
3229 	offset = args->offset;
3230 	if (offset >= va.va_size) {
3231 		*cs->statusp = resp->status = NFS4_OK;
3232 		resp->eof = TRUE;
3233 		resp->data_len = 0;
3234 		resp->data_val = NULL;
3235 		resp->mblk = NULL;
3236 		/* RDMA */
3237 		resp->wlist = args->wlist;
3238 		resp->wlist_len = resp->data_len;
3239 		*cs->statusp = resp->status = NFS4_OK;
3240 		if (resp->wlist)
3241 			clist_zero_len(resp->wlist);
3242 		goto out;
3243 	}
3244 
3245 	if (args->count == 0) {
3246 		*cs->statusp = resp->status = NFS4_OK;
3247 		resp->eof = FALSE;
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 		if (resp->wlist)
3255 			clist_zero_len(resp->wlist);
3256 		goto out;
3257 	}
3258 
3259 	/*
3260 	 * Do not allocate memory more than maximum allowed
3261 	 * transfer size
3262 	 */
3263 	if (args->count > rfs4_tsize(req))
3264 		args->count = rfs4_tsize(req);
3265 
3266 	if (loaned_buffers) {
3267 		uiop = (uio_t *)rfs_setup_xuio(vp);
3268 		ASSERT(uiop != NULL);
3269 		uiop->uio_segflg = UIO_SYSSPACE;
3270 		uiop->uio_loffset = args->offset;
3271 		uiop->uio_resid = args->count;
3272 
3273 		/* Jump to do the read if successful */
3274 		if (!VOP_REQZCBUF(vp, UIO_READ, (xuio_t *)uiop, cs->cr, &ct)) {
3275 			/*
3276 			 * Need to hold the vnode until after VOP_RETZCBUF()
3277 			 * is called.
3278 			 */
3279 			VN_HOLD(vp);
3280 			goto doio_read;
3281 		}
3282 
3283 		DTRACE_PROBE2(nfss__i__reqzcbuf_failed, int,
3284 		    uiop->uio_loffset, int, uiop->uio_resid);
3285 
3286 		uiop->uio_extflg = 0;
3287 
3288 		/* failure to setup for zero copy */
3289 		rfs_free_xuio((void *)uiop);
3290 		loaned_buffers = 0;
3291 	}
3292 
3293 	/*
3294 	 * If returning data via RDMA Write, then grab the chunk list. If we
3295 	 * aren't returning READ data w/RDMA_WRITE, then grab a mblk.
3296 	 */
3297 	if (rdma_used) {
3298 		mp = NULL;
3299 		(void) rdma_get_wchunk(req, &iov, args->wlist);
3300 		uio.uio_iov = &iov;
3301 		uio.uio_iovcnt = 1;
3302 	} else {
3303 		/*
3304 		 * mp will contain the data to be sent out in the read reply.
3305 		 * It will be freed after the reply has been sent.
3306 		 */
3307 		mp = rfs_read_alloc(args->count, &iovp, &iovcnt);
3308 		ASSERT(mp != NULL);
3309 		ASSERT(alloc_err == 0);
3310 		uio.uio_iov = iovp;
3311 		uio.uio_iovcnt = iovcnt;
3312 	}
3313 
3314 	uio.uio_segflg = UIO_SYSSPACE;
3315 	uio.uio_extflg = UIO_COPY_CACHED;
3316 	uio.uio_loffset = args->offset;
3317 	uio.uio_resid = args->count;
3318 	uiop = &uio;
3319 
3320 doio_read:
3321 	error = do_io(FREAD, vp, uiop, 0, cs->cr, &ct);
3322 
3323 	va.va_mask = AT_SIZE;
3324 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3325 
3326 	if (error) {
3327 		if (mp)
3328 			freemsg(mp);
3329 		*cs->statusp = resp->status = puterrno4(error);
3330 		goto out;
3331 	}
3332 
3333 	/* make mblk using zc buffers */
3334 	if (loaned_buffers) {
3335 		mp = uio_to_mblk(uiop);
3336 		ASSERT(mp != NULL);
3337 	}
3338 
3339 	*cs->statusp = resp->status = NFS4_OK;
3340 
3341 	ASSERT(uiop->uio_resid >= 0);
3342 	resp->data_len = args->count - uiop->uio_resid;
3343 	if (mp) {
3344 		resp->data_val = (char *)mp->b_datap->db_base;
3345 		rfs_rndup_mblks(mp, resp->data_len, loaned_buffers);
3346 	} else {
3347 		resp->data_val = (caddr_t)iov.iov_base;
3348 	}
3349 
3350 	resp->mblk = mp;
3351 
3352 	if (!verror && offset + resp->data_len == va.va_size)
3353 		resp->eof = TRUE;
3354 	else
3355 		resp->eof = FALSE;
3356 
3357 	if (rdma_used) {
3358 		if (!rdma_setup_read_data4(args, resp)) {
3359 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3360 		}
3361 	} else {
3362 		resp->wlist = NULL;
3363 	}
3364 
3365 out:
3366 	if (in_crit)
3367 		nbl_end_crit(vp);
3368 
3369 	if (iovp != NULL)
3370 		kmem_free(iovp, iovcnt * sizeof (struct iovec));
3371 
3372 	DTRACE_NFSV4_2(op__read__done, struct compound_state *, cs,
3373 	    READ4res *, resp);
3374 }
3375 
3376 static void
3377 rfs4_op_read_free(nfs_resop4 *resop)
3378 {
3379 	READ4res	*resp = &resop->nfs_resop4_u.opread;
3380 
3381 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3382 		freemsg(resp->mblk);
3383 		resp->mblk = NULL;
3384 		resp->data_val = NULL;
3385 		resp->data_len = 0;
3386 	}
3387 }
3388 
3389 static void
3390 rfs4_op_readdir_free(nfs_resop4 * resop)
3391 {
3392 	READDIR4res    *resp = &resop->nfs_resop4_u.opreaddir;
3393 
3394 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3395 		freeb(resp->mblk);
3396 		resp->mblk = NULL;
3397 		resp->data_len = 0;
3398 	}
3399 }
3400 
3401 
3402 /* ARGSUSED */
3403 static void
3404 rfs4_op_putpubfh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
3405     struct compound_state *cs)
3406 {
3407 	PUTPUBFH4res	*resp = &resop->nfs_resop4_u.opputpubfh;
3408 	int		error;
3409 	vnode_t		*vp;
3410 	struct exportinfo *exi, *sav_exi;
3411 	nfs_fh4_fmt_t	*fh_fmtp;
3412 
3413 	DTRACE_NFSV4_1(op__putpubfh__start, struct compound_state *, cs);
3414 
3415 	if (cs->vp) {
3416 		VN_RELE(cs->vp);
3417 		cs->vp = NULL;
3418 	}
3419 
3420 	if (cs->cr)
3421 		crfree(cs->cr);
3422 
3423 	cs->cr = crdup(cs->basecr);
3424 
3425 	vp = exi_public->exi_vp;
3426 	if (vp == NULL) {
3427 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3428 		goto out;
3429 	}
3430 
3431 	error = makefh4(&cs->fh, vp, exi_public);
3432 	if (error != 0) {
3433 		*cs->statusp = resp->status = puterrno4(error);
3434 		goto out;
3435 	}
3436 	sav_exi = cs->exi;
3437 	if (exi_public == exi_root) {
3438 		/*
3439 		 * No filesystem is actually shared public, so we default
3440 		 * to exi_root. In this case, we must check whether root
3441 		 * is exported.
3442 		 */
3443 		fh_fmtp = (nfs_fh4_fmt_t *)cs->fh.nfs_fh4_val;
3444 
3445 		/*
3446 		 * if root filesystem is exported, the exportinfo struct that we
3447 		 * should use is what checkexport4 returns, because root_exi is
3448 		 * actually a mostly empty struct.
3449 		 */
3450 		exi = checkexport4(&fh_fmtp->fh4_fsid,
3451 		    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
3452 		cs->exi = ((exi != NULL) ? exi : exi_public);
3453 	} else {
3454 		/*
3455 		 * it's a properly shared filesystem
3456 		 */
3457 		cs->exi = exi_public;
3458 	}
3459 
3460 	if (is_system_labeled()) {
3461 		bslabel_t *clabel;
3462 
3463 		ASSERT(req->rq_label != NULL);
3464 		clabel = req->rq_label;
3465 		DTRACE_PROBE2(tx__rfs4__log__info__opputpubfh__clabel, char *,
3466 		    "got client label from request(1)",
3467 		    struct svc_req *, req);
3468 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
3469 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
3470 			    cs->exi)) {
3471 				*cs->statusp = resp->status =
3472 				    NFS4ERR_SERVERFAULT;
3473 				goto out;
3474 			}
3475 		}
3476 	}
3477 
3478 	VN_HOLD(vp);
3479 	cs->vp = vp;
3480 
3481 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3482 		VN_RELE(cs->vp);
3483 		cs->vp = NULL;
3484 		cs->exi = sav_exi;
3485 		goto out;
3486 	}
3487 
3488 	*cs->statusp = resp->status = NFS4_OK;
3489 out:
3490 	DTRACE_NFSV4_2(op__putpubfh__done, struct compound_state *, cs,
3491 	    PUTPUBFH4res *, resp);
3492 }
3493 
3494 /*
3495  * XXX - issue with put*fh operations. Suppose /export/home is exported.
3496  * Suppose an NFS client goes to mount /export/home/joe. If /export, home,
3497  * or joe have restrictive search permissions, then we shouldn't let
3498  * the client get a file handle. This is easy to enforce. However, we
3499  * don't know what security flavor should be used until we resolve the
3500  * path name. Another complication is uid mapping. If root is
3501  * the user, then it will be mapped to the anonymous user by default,
3502  * but we won't know that till we've resolved the path name. And we won't
3503  * know what the anonymous user is.
3504  * Luckily, SECINFO is specified to take a full filename.
3505  * So what we will have to in rfs4_op_lookup is check that flavor of
3506  * the target object matches that of the request, and if root was the
3507  * caller, check for the root= and anon= options, and if necessary,
3508  * repeat the lookup using the right cred_t. But that's not done yet.
3509  */
3510 /* ARGSUSED */
3511 static void
3512 rfs4_op_putfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3513     struct compound_state *cs)
3514 {
3515 	PUTFH4args *args = &argop->nfs_argop4_u.opputfh;
3516 	PUTFH4res *resp = &resop->nfs_resop4_u.opputfh;
3517 	nfs_fh4_fmt_t *fh_fmtp;
3518 
3519 	DTRACE_NFSV4_2(op__putfh__start, struct compound_state *, cs,
3520 	    PUTFH4args *, args);
3521 
3522 	if (cs->vp) {
3523 		VN_RELE(cs->vp);
3524 		cs->vp = NULL;
3525 	}
3526 
3527 	if (cs->cr) {
3528 		crfree(cs->cr);
3529 		cs->cr = NULL;
3530 	}
3531 
3532 
3533 	if (args->object.nfs_fh4_len < NFS_FH4_LEN) {
3534 		*cs->statusp = resp->status = NFS4ERR_BADHANDLE;
3535 		goto out;
3536 	}
3537 
3538 	fh_fmtp = (nfs_fh4_fmt_t *)args->object.nfs_fh4_val;
3539 	cs->exi = checkexport4(&fh_fmtp->fh4_fsid, (fid_t *)&fh_fmtp->fh4_xlen,
3540 	    NULL);
3541 
3542 	if (cs->exi == NULL) {
3543 		*cs->statusp = resp->status = NFS4ERR_STALE;
3544 		goto out;
3545 	}
3546 
3547 	cs->cr = crdup(cs->basecr);
3548 
3549 	ASSERT(cs->cr != NULL);
3550 
3551 	if (! (cs->vp = nfs4_fhtovp(&args->object, cs->exi, &resp->status))) {
3552 		*cs->statusp = resp->status;
3553 		goto out;
3554 	}
3555 
3556 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3557 		VN_RELE(cs->vp);
3558 		cs->vp = NULL;
3559 		goto out;
3560 	}
3561 
3562 	nfs_fh4_copy(&args->object, &cs->fh);
3563 	*cs->statusp = resp->status = NFS4_OK;
3564 	cs->deleg = FALSE;
3565 
3566 out:
3567 	DTRACE_NFSV4_2(op__putfh__done, struct compound_state *, cs,
3568 	    PUTFH4res *, resp);
3569 }
3570 
3571 /* ARGSUSED */
3572 static void
3573 rfs4_op_putrootfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3574     struct compound_state *cs)
3575 {
3576 	PUTROOTFH4res *resp = &resop->nfs_resop4_u.opputrootfh;
3577 	int error;
3578 	fid_t fid;
3579 	struct exportinfo *exi, *sav_exi;
3580 
3581 	DTRACE_NFSV4_1(op__putrootfh__start, struct compound_state *, cs);
3582 
3583 	if (cs->vp) {
3584 		VN_RELE(cs->vp);
3585 		cs->vp = NULL;
3586 	}
3587 
3588 	if (cs->cr)
3589 		crfree(cs->cr);
3590 
3591 	cs->cr = crdup(cs->basecr);
3592 
3593 	/*
3594 	 * Using rootdir, the system root vnode,
3595 	 * get its fid.
3596 	 */
3597 	bzero(&fid, sizeof (fid));
3598 	fid.fid_len = MAXFIDSZ;
3599 	error = vop_fid_pseudo(rootdir, &fid);
3600 	if (error != 0) {
3601 		*cs->statusp = resp->status = puterrno4(error);
3602 		goto out;
3603 	}
3604 
3605 	/*
3606 	 * Then use the root fsid & fid it to find out if it's exported
3607 	 *
3608 	 * If the server root isn't exported directly, then
3609 	 * it should at least be a pseudo export based on
3610 	 * one or more exports further down in the server's
3611 	 * file tree.
3612 	 */
3613 	exi = checkexport4(&rootdir->v_vfsp->vfs_fsid, &fid, NULL);
3614 	if (exi == NULL || exi->exi_export.ex_flags & EX_PUBLIC) {
3615 		NFS4_DEBUG(rfs4_debug,
3616 		    (CE_WARN, "rfs4_op_putrootfh: export check failure"));
3617 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3618 		goto out;
3619 	}
3620 
3621 	/*
3622 	 * Now make a filehandle based on the root
3623 	 * export and root vnode.
3624 	 */
3625 	error = makefh4(&cs->fh, rootdir, exi);
3626 	if (error != 0) {
3627 		*cs->statusp = resp->status = puterrno4(error);
3628 		goto out;
3629 	}
3630 
3631 	sav_exi = cs->exi;
3632 	cs->exi = exi;
3633 
3634 	VN_HOLD(rootdir);
3635 	cs->vp = rootdir;
3636 
3637 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3638 		VN_RELE(rootdir);
3639 		cs->vp = NULL;
3640 		cs->exi = sav_exi;
3641 		goto out;
3642 	}
3643 
3644 	*cs->statusp = resp->status = NFS4_OK;
3645 	cs->deleg = FALSE;
3646 out:
3647 	DTRACE_NFSV4_2(op__putrootfh__done, struct compound_state *, cs,
3648 	    PUTROOTFH4res *, resp);
3649 }
3650 
3651 /*
3652  * readlink: args: CURRENT_FH.
3653  *	res: status. If success - CURRENT_FH unchanged, return linktext.
3654  */
3655 
3656 /* ARGSUSED */
3657 static void
3658 rfs4_op_readlink(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3659     struct compound_state *cs)
3660 {
3661 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3662 	int error;
3663 	vnode_t *vp;
3664 	struct iovec iov;
3665 	struct vattr va;
3666 	struct uio uio;
3667 	char *data;
3668 	struct sockaddr *ca;
3669 	char *name = NULL;
3670 	int is_referral;
3671 
3672 	DTRACE_NFSV4_1(op__readlink__start, struct compound_state *, cs);
3673 
3674 	/* CURRENT_FH: directory */
3675 	vp = cs->vp;
3676 	if (vp == NULL) {
3677 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3678 		goto out;
3679 	}
3680 
3681 	if (cs->access == CS_ACCESS_DENIED) {
3682 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3683 		goto out;
3684 	}
3685 
3686 	/* Is it a referral? */
3687 	if (vn_is_nfs_reparse(vp, cs->cr) && client_is_downrev(req)) {
3688 
3689 		is_referral = 1;
3690 
3691 	} else {
3692 
3693 		is_referral = 0;
3694 
3695 		if (vp->v_type == VDIR) {
3696 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
3697 			goto out;
3698 		}
3699 
3700 		if (vp->v_type != VLNK) {
3701 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3702 			goto out;
3703 		}
3704 
3705 	}
3706 
3707 	va.va_mask = AT_MODE;
3708 	error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
3709 	if (error) {
3710 		*cs->statusp = resp->status = puterrno4(error);
3711 		goto out;
3712 	}
3713 
3714 	if (MANDLOCK(vp, va.va_mode)) {
3715 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3716 		goto out;
3717 	}
3718 
3719 	data = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);
3720 
3721 	if (is_referral) {
3722 		char *s;
3723 		size_t strsz;
3724 
3725 		/* Get an artificial symlink based on a referral */
3726 		s = build_symlink(vp, cs->cr, &strsz);
3727 		global_svstat_ptr[4][NFS_REFERLINKS].value.ui64++;
3728 		DTRACE_PROBE2(nfs4serv__func__referral__reflink,
3729 		    vnode_t *, vp, char *, s);
3730 		if (s == NULL)
3731 			error = EINVAL;
3732 		else {
3733 			error = 0;
3734 			(void) strlcpy(data, s, MAXPATHLEN + 1);
3735 			kmem_free(s, strsz);
3736 		}
3737 
3738 	} else {
3739 
3740 		iov.iov_base = data;
3741 		iov.iov_len = MAXPATHLEN;
3742 		uio.uio_iov = &iov;
3743 		uio.uio_iovcnt = 1;
3744 		uio.uio_segflg = UIO_SYSSPACE;
3745 		uio.uio_extflg = UIO_COPY_CACHED;
3746 		uio.uio_loffset = 0;
3747 		uio.uio_resid = MAXPATHLEN;
3748 
3749 		error = VOP_READLINK(vp, &uio, cs->cr, NULL);
3750 
3751 		if (!error)
3752 			*(data + MAXPATHLEN - uio.uio_resid) = '\0';
3753 	}
3754 
3755 	if (error) {
3756 		kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3757 		*cs->statusp = resp->status = puterrno4(error);
3758 		goto out;
3759 	}
3760 
3761 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
3762 	name = nfscmd_convname(ca, cs->exi, data, NFSCMD_CONV_OUTBOUND,
3763 	    MAXPATHLEN  + 1);
3764 
3765 	if (name == NULL) {
3766 		/*
3767 		 * Even though the conversion failed, we return
3768 		 * something. We just don't translate it.
3769 		 */
3770 		name = data;
3771 	}
3772 
3773 	/*
3774 	 * treat link name as data
3775 	 */
3776 	(void) str_to_utf8(name, (utf8string *)&resp->link);
3777 
3778 	if (name != data)
3779 		kmem_free(name, MAXPATHLEN + 1);
3780 	kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3781 	*cs->statusp = resp->status = NFS4_OK;
3782 
3783 out:
3784 	DTRACE_NFSV4_2(op__readlink__done, struct compound_state *, cs,
3785 	    READLINK4res *, resp);
3786 }
3787 
3788 static void
3789 rfs4_op_readlink_free(nfs_resop4 *resop)
3790 {
3791 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3792 	utf8string *symlink = (utf8string *)&resp->link;
3793 
3794 	if (symlink->utf8string_val) {
3795 		UTF8STRING_FREE(*symlink)
3796 	}
3797 }
3798 
3799 /*
3800  * release_lockowner:
3801  *	Release any state associated with the supplied
3802  *	lockowner. Note if any lo_state is holding locks we will not
3803  *	rele that lo_state and thus the lockowner will not be destroyed.
3804  *	A client using lock after the lock owner stateid has been released
3805  *	will suffer the consequence of NFS4ERR_BAD_STATEID and would have
3806  *	to reissue the lock with new_lock_owner set to TRUE.
3807  *	args: lock_owner
3808  *	res:  status
3809  */
3810 /* ARGSUSED */
3811 static void
3812 rfs4_op_release_lockowner(nfs_argop4 *argop, nfs_resop4 *resop,
3813     struct svc_req *req, struct compound_state *cs)
3814 {
3815 	RELEASE_LOCKOWNER4args *ap = &argop->nfs_argop4_u.oprelease_lockowner;
3816 	RELEASE_LOCKOWNER4res *resp = &resop->nfs_resop4_u.oprelease_lockowner;
3817 	rfs4_lockowner_t *lo;
3818 	rfs4_openowner_t *oo;
3819 	rfs4_state_t *sp;
3820 	rfs4_lo_state_t *lsp;
3821 	rfs4_client_t *cp;
3822 	bool_t create = FALSE;
3823 	locklist_t *llist;
3824 	sysid_t sysid;
3825 
3826 	DTRACE_NFSV4_2(op__release__lockowner__start, struct compound_state *,
3827 	    cs, RELEASE_LOCKOWNER4args *, ap);
3828 
3829 	/* Make sure there is a clientid around for this request */
3830 	cp = rfs4_findclient_by_id(ap->lock_owner.clientid, FALSE);
3831 
3832 	if (cp == NULL) {
3833 		*cs->statusp = resp->status =
3834 		    rfs4_check_clientid(&ap->lock_owner.clientid, 0);
3835 		goto out;
3836 	}
3837 	rfs4_client_rele(cp);
3838 
3839 	lo = rfs4_findlockowner(&ap->lock_owner, &create);
3840 	if (lo == NULL) {
3841 		*cs->statusp = resp->status = NFS4_OK;
3842 		goto out;
3843 	}
3844 	ASSERT(lo->rl_client != NULL);
3845 
3846 	/*
3847 	 * Check for EXPIRED client. If so will reap state with in a lease
3848 	 * period or on next set_clientid_confirm step
3849 	 */
3850 	if (rfs4_lease_expired(lo->rl_client)) {
3851 		rfs4_lockowner_rele(lo);
3852 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
3853 		goto out;
3854 	}
3855 
3856 	/*
3857 	 * If no sysid has been assigned, then no locks exist; just return.
3858 	 */
3859 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3860 	if (lo->rl_client->rc_sysidt == LM_NOSYSID) {
3861 		rfs4_lockowner_rele(lo);
3862 		rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3863 		goto out;
3864 	}
3865 
3866 	sysid = lo->rl_client->rc_sysidt;
3867 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3868 
3869 	/*
3870 	 * Mark the lockowner invalid.
3871 	 */
3872 	rfs4_dbe_hide(lo->rl_dbe);
3873 
3874 	/*
3875 	 * sysid-pid pair should now not be used since the lockowner is
3876 	 * invalid. If the client were to instantiate the lockowner again
3877 	 * it would be assigned a new pid. Thus we can get the list of
3878 	 * current locks.
3879 	 */
3880 
3881 	llist = flk_get_active_locks(sysid, lo->rl_pid);
3882 	/* If we are still holding locks fail */
3883 	if (llist != NULL) {
3884 
3885 		*cs->statusp = resp->status = NFS4ERR_LOCKS_HELD;
3886 
3887 		flk_free_locklist(llist);
3888 		/*
3889 		 * We need to unhide the lockowner so the client can
3890 		 * try it again. The bad thing here is if the client
3891 		 * has a logic error that took it here in the first place
3892 		 * they probably have lost accounting of the locks that it
3893 		 * is holding. So we may have dangling state until the
3894 		 * open owner state is reaped via close. One scenario
3895 		 * that could possibly occur is that the client has
3896 		 * sent the unlock request(s) in separate threads
3897 		 * and has not waited for the replies before sending the
3898 		 * RELEASE_LOCKOWNER request. Presumably, it would expect
3899 		 * and deal appropriately with NFS4ERR_LOCKS_HELD, by
3900 		 * reissuing the request.
3901 		 */
3902 		rfs4_dbe_unhide(lo->rl_dbe);
3903 		rfs4_lockowner_rele(lo);
3904 		goto out;
3905 	}
3906 
3907 	/*
3908 	 * For the corresponding client we need to check each open
3909 	 * owner for any opens that have lockowner state associated
3910 	 * with this lockowner.
3911 	 */
3912 
3913 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3914 	for (oo = list_head(&lo->rl_client->rc_openownerlist); oo != NULL;
3915 	    oo = list_next(&lo->rl_client->rc_openownerlist, oo)) {
3916 
3917 		rfs4_dbe_lock(oo->ro_dbe);
3918 		for (sp = list_head(&oo->ro_statelist); sp != NULL;
3919 		    sp = list_next(&oo->ro_statelist, sp)) {
3920 
3921 			rfs4_dbe_lock(sp->rs_dbe);
3922 			for (lsp = list_head(&sp->rs_lostatelist);
3923 			    lsp != NULL;
3924 			    lsp = list_next(&sp->rs_lostatelist, lsp)) {
3925 				if (lsp->rls_locker == lo) {
3926 					rfs4_dbe_lock(lsp->rls_dbe);
3927 					rfs4_dbe_invalidate(lsp->rls_dbe);
3928 					rfs4_dbe_unlock(lsp->rls_dbe);
3929 				}
3930 			}
3931 			rfs4_dbe_unlock(sp->rs_dbe);
3932 		}
3933 		rfs4_dbe_unlock(oo->ro_dbe);
3934 	}
3935 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3936 
3937 	rfs4_lockowner_rele(lo);
3938 
3939 	*cs->statusp = resp->status = NFS4_OK;
3940 
3941 out:
3942 	DTRACE_NFSV4_2(op__release__lockowner__done, struct compound_state *,
3943 	    cs, RELEASE_LOCKOWNER4res *, resp);
3944 }
3945 
3946 /*
3947  * short utility function to lookup a file and recall the delegation
3948  */
3949 static rfs4_file_t *
3950 rfs4_lookup_and_findfile(vnode_t *dvp, char *nm, vnode_t **vpp,
3951     int *lkup_error, cred_t *cr)
3952 {
3953 	vnode_t *vp;
3954 	rfs4_file_t *fp = NULL;
3955 	bool_t fcreate = FALSE;
3956 	int error;
3957 
3958 	if (vpp)
3959 		*vpp = NULL;
3960 
3961 	if ((error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cr, NULL, NULL,
3962 	    NULL)) == 0) {
3963 		if (vp->v_type == VREG)
3964 			fp = rfs4_findfile(vp, NULL, &fcreate);
3965 		if (vpp)
3966 			*vpp = vp;
3967 		else
3968 			VN_RELE(vp);
3969 	}
3970 
3971 	if (lkup_error)
3972 		*lkup_error = error;
3973 
3974 	return (fp);
3975 }
3976 
3977 /*
3978  * remove: args: CURRENT_FH: directory; name.
3979  *	res: status. If success - CURRENT_FH unchanged, return change_info
3980  *		for directory.
3981  */
3982 /* ARGSUSED */
3983 static void
3984 rfs4_op_remove(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3985     struct compound_state *cs)
3986 {
3987 	REMOVE4args *args = &argop->nfs_argop4_u.opremove;
3988 	REMOVE4res *resp = &resop->nfs_resop4_u.opremove;
3989 	int error;
3990 	vnode_t *dvp, *vp;
3991 	struct vattr bdva, idva, adva;
3992 	char *nm;
3993 	uint_t len;
3994 	rfs4_file_t *fp;
3995 	int in_crit = 0;
3996 	bslabel_t *clabel;
3997 	struct sockaddr *ca;
3998 	char *name = NULL;
3999 	nfsstat4 status;
4000 
4001 	DTRACE_NFSV4_2(op__remove__start, struct compound_state *, cs,
4002 	    REMOVE4args *, args);
4003 
4004 	/* CURRENT_FH: directory */
4005 	dvp = cs->vp;
4006 	if (dvp == NULL) {
4007 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4008 		goto out;
4009 	}
4010 
4011 	if (cs->access == CS_ACCESS_DENIED) {
4012 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4013 		goto out;
4014 	}
4015 
4016 	/*
4017 	 * If there is an unshared filesystem mounted on this vnode,
4018 	 * Do not allow to remove anything in this directory.
4019 	 */
4020 	if (vn_ismntpt(dvp)) {
4021 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4022 		goto out;
4023 	}
4024 
4025 	if (dvp->v_type != VDIR) {
4026 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4027 		goto out;
4028 	}
4029 
4030 	status = utf8_dir_verify(&args->target);
4031 	if (status != NFS4_OK) {
4032 		*cs->statusp = resp->status = status;
4033 		goto out;
4034 	}
4035 
4036 	/*
4037 	 * Lookup the file so that we can check if it's a directory
4038 	 */
4039 	nm = utf8_to_fn(&args->target, &len, NULL);
4040 	if (nm == NULL) {
4041 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4042 		goto out;
4043 	}
4044 
4045 	if (len > MAXNAMELEN) {
4046 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4047 		kmem_free(nm, len);
4048 		goto out;
4049 	}
4050 
4051 	if (rdonly4(req, cs)) {
4052 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4053 		kmem_free(nm, len);
4054 		goto out;
4055 	}
4056 
4057 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4058 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
4059 	    MAXPATHLEN  + 1);
4060 
4061 	if (name == NULL) {
4062 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4063 		kmem_free(nm, len);
4064 		goto out;
4065 	}
4066 
4067 	/*
4068 	 * Lookup the file to determine type and while we are see if
4069 	 * there is a file struct around and check for delegation.
4070 	 * We don't need to acquire va_seq before this lookup, if
4071 	 * it causes an update, cinfo.before will not match, which will
4072 	 * trigger a cache flush even if atomic is TRUE.
4073 	 */
4074 	if (fp = rfs4_lookup_and_findfile(dvp, name, &vp, &error, cs->cr)) {
4075 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4076 		    NULL)) {
4077 			VN_RELE(vp);
4078 			rfs4_file_rele(fp);
4079 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4080 			if (nm != name)
4081 				kmem_free(name, MAXPATHLEN + 1);
4082 			kmem_free(nm, len);
4083 			goto out;
4084 		}
4085 	}
4086 
4087 	/* Didn't find anything to remove */
4088 	if (vp == NULL) {
4089 		*cs->statusp = resp->status = error;
4090 		if (nm != name)
4091 			kmem_free(name, MAXPATHLEN + 1);
4092 		kmem_free(nm, len);
4093 		goto out;
4094 	}
4095 
4096 	if (nbl_need_check(vp)) {
4097 		nbl_start_crit(vp, RW_READER);
4098 		in_crit = 1;
4099 		if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
4100 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4101 			if (nm != name)
4102 				kmem_free(name, MAXPATHLEN + 1);
4103 			kmem_free(nm, len);
4104 			nbl_end_crit(vp);
4105 			VN_RELE(vp);
4106 			if (fp) {
4107 				rfs4_clear_dont_grant(fp);
4108 				rfs4_file_rele(fp);
4109 			}
4110 			goto out;
4111 		}
4112 	}
4113 
4114 	/* check label before allowing removal */
4115 	if (is_system_labeled()) {
4116 		ASSERT(req->rq_label != NULL);
4117 		clabel = req->rq_label;
4118 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
4119 		    "got client label from request(1)",
4120 		    struct svc_req *, req);
4121 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4122 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
4123 			    cs->exi)) {
4124 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4125 				if (name != nm)
4126 					kmem_free(name, MAXPATHLEN + 1);
4127 				kmem_free(nm, len);
4128 				if (in_crit)
4129 					nbl_end_crit(vp);
4130 				VN_RELE(vp);
4131 				if (fp) {
4132 					rfs4_clear_dont_grant(fp);
4133 					rfs4_file_rele(fp);
4134 				}
4135 				goto out;
4136 			}
4137 		}
4138 	}
4139 
4140 	/* Get dir "before" change value */
4141 	bdva.va_mask = AT_CTIME|AT_SEQ;
4142 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
4143 	if (error) {
4144 		*cs->statusp = resp->status = puterrno4(error);
4145 		if (nm != name)
4146 			kmem_free(name, MAXPATHLEN + 1);
4147 		kmem_free(nm, len);
4148 		if (in_crit)
4149 			nbl_end_crit(vp);
4150 		VN_RELE(vp);
4151 		if (fp) {
4152 			rfs4_clear_dont_grant(fp);
4153 			rfs4_file_rele(fp);
4154 		}
4155 		goto out;
4156 	}
4157 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
4158 
4159 	/* Actually do the REMOVE operation */
4160 	if (vp->v_type == VDIR) {
4161 		/*
4162 		 * Can't remove a directory that has a mounted-on filesystem.
4163 		 */
4164 		if (vn_ismntpt(vp)) {
4165 			error = EACCES;
4166 		} else {
4167 			/*
4168 			 * System V defines rmdir to return EEXIST,
4169 			 * not ENOTEMPTY, if the directory is not
4170 			 * empty.  A System V NFS server needs to map
4171 			 * NFS4ERR_EXIST to NFS4ERR_NOTEMPTY to
4172 			 * transmit over the wire.
4173 			 */
4174 			if ((error = VOP_RMDIR(dvp, name, rootdir, cs->cr,
4175 			    NULL, 0)) == EEXIST)
4176 				error = ENOTEMPTY;
4177 		}
4178 	} else {
4179 		if ((error = VOP_REMOVE(dvp, name, cs->cr, NULL, 0)) == 0 &&
4180 		    fp != NULL) {
4181 			struct vattr va;
4182 			vnode_t *tvp;
4183 
4184 			rfs4_dbe_lock(fp->rf_dbe);
4185 			tvp = fp->rf_vp;
4186 			if (tvp)
4187 				VN_HOLD(tvp);
4188 			rfs4_dbe_unlock(fp->rf_dbe);
4189 
4190 			if (tvp) {
4191 				/*
4192 				 * This is va_seq safe because we are not
4193 				 * manipulating dvp.
4194 				 */
4195 				va.va_mask = AT_NLINK;
4196 				if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4197 				    va.va_nlink == 0) {
4198 					/* Remove state on file remove */
4199 					if (in_crit) {
4200 						nbl_end_crit(vp);
4201 						in_crit = 0;
4202 					}
4203 					rfs4_close_all_state(fp);
4204 				}
4205 				VN_RELE(tvp);
4206 			}
4207 		}
4208 	}
4209 
4210 	if (in_crit)
4211 		nbl_end_crit(vp);
4212 	VN_RELE(vp);
4213 
4214 	if (fp) {
4215 		rfs4_clear_dont_grant(fp);
4216 		rfs4_file_rele(fp);
4217 	}
4218 	if (nm != name)
4219 		kmem_free(name, MAXPATHLEN + 1);
4220 	kmem_free(nm, len);
4221 
4222 	if (error) {
4223 		*cs->statusp = resp->status = puterrno4(error);
4224 		goto out;
4225 	}
4226 
4227 	/*
4228 	 * Get the initial "after" sequence number, if it fails, set to zero
4229 	 */
4230 	idva.va_mask = AT_SEQ;
4231 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
4232 		idva.va_seq = 0;
4233 
4234 	/*
4235 	 * Force modified data and metadata out to stable storage.
4236 	 */
4237 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
4238 
4239 	/*
4240 	 * Get "after" change value, if it fails, simply return the
4241 	 * before value.
4242 	 */
4243 	adva.va_mask = AT_CTIME|AT_SEQ;
4244 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
4245 		adva.va_ctime = bdva.va_ctime;
4246 		adva.va_seq = 0;
4247 	}
4248 
4249 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
4250 
4251 	/*
4252 	 * The cinfo.atomic = TRUE only if we have
4253 	 * non-zero va_seq's, and it has incremented by exactly one
4254 	 * during the VOP_REMOVE/RMDIR and it didn't change during
4255 	 * the VOP_FSYNC.
4256 	 */
4257 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
4258 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
4259 		resp->cinfo.atomic = TRUE;
4260 	else
4261 		resp->cinfo.atomic = FALSE;
4262 
4263 	*cs->statusp = resp->status = NFS4_OK;
4264 
4265 out:
4266 	DTRACE_NFSV4_2(op__remove__done, struct compound_state *, cs,
4267 	    REMOVE4res *, resp);
4268 }
4269 
4270 /*
4271  * rename: args: SAVED_FH: from directory, CURRENT_FH: target directory,
4272  *		oldname and newname.
4273  *	res: status. If success - CURRENT_FH unchanged, return change_info
4274  *		for both from and target directories.
4275  */
4276 /* ARGSUSED */
4277 static void
4278 rfs4_op_rename(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4279     struct compound_state *cs)
4280 {
4281 	RENAME4args *args = &argop->nfs_argop4_u.oprename;
4282 	RENAME4res *resp = &resop->nfs_resop4_u.oprename;
4283 	int error;
4284 	vnode_t *odvp;
4285 	vnode_t *ndvp;
4286 	vnode_t *srcvp, *targvp;
4287 	struct vattr obdva, oidva, oadva;
4288 	struct vattr nbdva, nidva, nadva;
4289 	char *onm, *nnm;
4290 	uint_t olen, nlen;
4291 	rfs4_file_t *fp, *sfp;
4292 	int in_crit_src, in_crit_targ;
4293 	int fp_rele_grant_hold, sfp_rele_grant_hold;
4294 	bslabel_t *clabel;
4295 	struct sockaddr *ca;
4296 	char *converted_onm = NULL;
4297 	char *converted_nnm = NULL;
4298 	nfsstat4 status;
4299 
4300 	DTRACE_NFSV4_2(op__rename__start, struct compound_state *, cs,
4301 	    RENAME4args *, args);
4302 
4303 	fp = sfp = NULL;
4304 	srcvp = targvp = NULL;
4305 	in_crit_src = in_crit_targ = 0;
4306 	fp_rele_grant_hold = sfp_rele_grant_hold = 0;
4307 
4308 	/* CURRENT_FH: target directory */
4309 	ndvp = cs->vp;
4310 	if (ndvp == NULL) {
4311 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4312 		goto out;
4313 	}
4314 
4315 	/* SAVED_FH: from directory */
4316 	odvp = cs->saved_vp;
4317 	if (odvp == NULL) {
4318 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4319 		goto out;
4320 	}
4321 
4322 	if (cs->access == CS_ACCESS_DENIED) {
4323 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4324 		goto out;
4325 	}
4326 
4327 	/*
4328 	 * If there is an unshared filesystem mounted on this vnode,
4329 	 * do not allow to rename objects in this directory.
4330 	 */
4331 	if (vn_ismntpt(odvp)) {
4332 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4333 		goto out;
4334 	}
4335 
4336 	/*
4337 	 * If there is an unshared filesystem mounted on this vnode,
4338 	 * do not allow to rename to this directory.
4339 	 */
4340 	if (vn_ismntpt(ndvp)) {
4341 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4342 		goto out;
4343 	}
4344 
4345 	if (odvp->v_type != VDIR || ndvp->v_type != VDIR) {
4346 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4347 		goto out;
4348 	}
4349 
4350 	if (cs->saved_exi != cs->exi) {
4351 		*cs->statusp = resp->status = NFS4ERR_XDEV;
4352 		goto out;
4353 	}
4354 
4355 	status = utf8_dir_verify(&args->oldname);
4356 	if (status != NFS4_OK) {
4357 		*cs->statusp = resp->status = status;
4358 		goto out;
4359 	}
4360 
4361 	status = utf8_dir_verify(&args->newname);
4362 	if (status != NFS4_OK) {
4363 		*cs->statusp = resp->status = status;
4364 		goto out;
4365 	}
4366 
4367 	onm = utf8_to_fn(&args->oldname, &olen, NULL);
4368 	if (onm == NULL) {
4369 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4370 		goto out;
4371 	}
4372 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4373 	nlen = MAXPATHLEN + 1;
4374 	converted_onm = nfscmd_convname(ca, cs->exi, onm, NFSCMD_CONV_INBOUND,
4375 	    nlen);
4376 
4377 	if (converted_onm == NULL) {
4378 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4379 		kmem_free(onm, olen);
4380 		goto out;
4381 	}
4382 
4383 	nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4384 	if (nnm == NULL) {
4385 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4386 		if (onm != converted_onm)
4387 			kmem_free(converted_onm, MAXPATHLEN + 1);
4388 		kmem_free(onm, olen);
4389 		goto out;
4390 	}
4391 	converted_nnm = nfscmd_convname(ca, cs->exi, nnm, NFSCMD_CONV_INBOUND,
4392 	    MAXPATHLEN  + 1);
4393 
4394 	if (converted_nnm == NULL) {
4395 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4396 		kmem_free(nnm, nlen);
4397 		nnm = NULL;
4398 		if (onm != converted_onm)
4399 			kmem_free(converted_onm, MAXPATHLEN + 1);
4400 		kmem_free(onm, olen);
4401 		goto out;
4402 	}
4403 
4404 
4405 	if (olen > MAXNAMELEN || nlen > MAXNAMELEN) {
4406 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4407 		kmem_free(onm, olen);
4408 		kmem_free(nnm, nlen);
4409 		goto out;
4410 	}
4411 
4412 
4413 	if (rdonly4(req, cs)) {
4414 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4415 		if (onm != converted_onm)
4416 			kmem_free(converted_onm, MAXPATHLEN + 1);
4417 		kmem_free(onm, olen);
4418 		if (nnm != converted_nnm)
4419 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4420 		kmem_free(nnm, nlen);
4421 		goto out;
4422 	}
4423 
4424 	/* check label of the target dir */
4425 	if (is_system_labeled()) {
4426 		ASSERT(req->rq_label != NULL);
4427 		clabel = req->rq_label;
4428 		DTRACE_PROBE2(tx__rfs4__log__info__oprename__clabel, char *,
4429 		    "got client label from request(1)",
4430 		    struct svc_req *, req);
4431 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4432 			if (!do_rfs_label_check(clabel, ndvp,
4433 			    EQUALITY_CHECK, cs->exi)) {
4434 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4435 				goto err_out;
4436 			}
4437 		}
4438 	}
4439 
4440 	/*
4441 	 * Is the source a file and have a delegation?
4442 	 * We don't need to acquire va_seq before these lookups, if
4443 	 * it causes an update, cinfo.before will not match, which will
4444 	 * trigger a cache flush even if atomic is TRUE.
4445 	 */
4446 	if (sfp = rfs4_lookup_and_findfile(odvp, converted_onm, &srcvp,
4447 	    &error, cs->cr)) {
4448 		if (rfs4_check_delegated_byfp(FWRITE, sfp, TRUE, TRUE, TRUE,
4449 		    NULL)) {
4450 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4451 			goto err_out;
4452 		}
4453 	}
4454 
4455 	if (srcvp == NULL) {
4456 		*cs->statusp = resp->status = puterrno4(error);
4457 		if (onm != converted_onm)
4458 			kmem_free(converted_onm, MAXPATHLEN + 1);
4459 		kmem_free(onm, olen);
4460 		if (nnm != converted_nnm)
4461 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4462 		kmem_free(nnm, nlen);
4463 		goto out;
4464 	}
4465 
4466 	sfp_rele_grant_hold = 1;
4467 
4468 	/* Does the destination exist and a file and have a delegation? */
4469 	if (fp = rfs4_lookup_and_findfile(ndvp, converted_nnm, &targvp,
4470 	    NULL, cs->cr)) {
4471 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4472 		    NULL)) {
4473 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4474 			goto err_out;
4475 		}
4476 	}
4477 	fp_rele_grant_hold = 1;
4478 
4479 
4480 	/* Check for NBMAND lock on both source and target */
4481 	if (nbl_need_check(srcvp)) {
4482 		nbl_start_crit(srcvp, RW_READER);
4483 		in_crit_src = 1;
4484 		if (nbl_conflict(srcvp, NBL_RENAME, 0, 0, 0, NULL)) {
4485 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4486 			goto err_out;
4487 		}
4488 	}
4489 
4490 	if (targvp && nbl_need_check(targvp)) {
4491 		nbl_start_crit(targvp, RW_READER);
4492 		in_crit_targ = 1;
4493 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
4494 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4495 			goto err_out;
4496 		}
4497 	}
4498 
4499 	/* Get source "before" change value */
4500 	obdva.va_mask = AT_CTIME|AT_SEQ;
4501 	error = VOP_GETATTR(odvp, &obdva, 0, cs->cr, NULL);
4502 	if (!error) {
4503 		nbdva.va_mask = AT_CTIME|AT_SEQ;
4504 		error = VOP_GETATTR(ndvp, &nbdva, 0, cs->cr, NULL);
4505 	}
4506 	if (error) {
4507 		*cs->statusp = resp->status = puterrno4(error);
4508 		goto err_out;
4509 	}
4510 
4511 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.before, obdva.va_ctime)
4512 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.before, nbdva.va_ctime)
4513 
4514 	if ((error = VOP_RENAME(odvp, converted_onm, ndvp, converted_nnm,
4515 	    cs->cr, NULL, 0)) == 0 && fp != NULL) {
4516 		struct vattr va;
4517 		vnode_t *tvp;
4518 
4519 		rfs4_dbe_lock(fp->rf_dbe);
4520 		tvp = fp->rf_vp;
4521 		if (tvp)
4522 			VN_HOLD(tvp);
4523 		rfs4_dbe_unlock(fp->rf_dbe);
4524 
4525 		if (tvp) {
4526 			va.va_mask = AT_NLINK;
4527 			if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4528 			    va.va_nlink == 0) {
4529 				/* The file is gone and so should the state */
4530 				if (in_crit_targ) {
4531 					nbl_end_crit(targvp);
4532 					in_crit_targ = 0;
4533 				}
4534 				rfs4_close_all_state(fp);
4535 			}
4536 			VN_RELE(tvp);
4537 		}
4538 	}
4539 	if (error == 0)
4540 		vn_renamepath(ndvp, srcvp, nnm, nlen - 1);
4541 
4542 	if (in_crit_src)
4543 		nbl_end_crit(srcvp);
4544 	if (srcvp)
4545 		VN_RELE(srcvp);
4546 	if (in_crit_targ)
4547 		nbl_end_crit(targvp);
4548 	if (targvp)
4549 		VN_RELE(targvp);
4550 
4551 	if (sfp) {
4552 		rfs4_clear_dont_grant(sfp);
4553 		rfs4_file_rele(sfp);
4554 	}
4555 	if (fp) {
4556 		rfs4_clear_dont_grant(fp);
4557 		rfs4_file_rele(fp);
4558 	}
4559 
4560 	if (converted_onm != onm)
4561 		kmem_free(converted_onm, MAXPATHLEN + 1);
4562 	kmem_free(onm, olen);
4563 	if (converted_nnm != nnm)
4564 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4565 	kmem_free(nnm, nlen);
4566 
4567 	/*
4568 	 * Get the initial "after" sequence number, if it fails, set to zero
4569 	 */
4570 	oidva.va_mask = AT_SEQ;
4571 	if (VOP_GETATTR(odvp, &oidva, 0, cs->cr, NULL))
4572 		oidva.va_seq = 0;
4573 
4574 	nidva.va_mask = AT_SEQ;
4575 	if (VOP_GETATTR(ndvp, &nidva, 0, cs->cr, NULL))
4576 		nidva.va_seq = 0;
4577 
4578 	/*
4579 	 * Force modified data and metadata out to stable storage.
4580 	 */
4581 	(void) VOP_FSYNC(odvp, 0, cs->cr, NULL);
4582 	(void) VOP_FSYNC(ndvp, 0, cs->cr, NULL);
4583 
4584 	if (error) {
4585 		*cs->statusp = resp->status = puterrno4(error);
4586 		goto out;
4587 	}
4588 
4589 	/*
4590 	 * Get "after" change values, if it fails, simply return the
4591 	 * before value.
4592 	 */
4593 	oadva.va_mask = AT_CTIME|AT_SEQ;
4594 	if (VOP_GETATTR(odvp, &oadva, 0, cs->cr, NULL)) {
4595 		oadva.va_ctime = obdva.va_ctime;
4596 		oadva.va_seq = 0;
4597 	}
4598 
4599 	nadva.va_mask = AT_CTIME|AT_SEQ;
4600 	if (VOP_GETATTR(odvp, &nadva, 0, cs->cr, NULL)) {
4601 		nadva.va_ctime = nbdva.va_ctime;
4602 		nadva.va_seq = 0;
4603 	}
4604 
4605 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.after, oadva.va_ctime)
4606 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.after, nadva.va_ctime)
4607 
4608 	/*
4609 	 * The cinfo.atomic = TRUE only if we have
4610 	 * non-zero va_seq's, and it has incremented by exactly one
4611 	 * during the VOP_RENAME and it didn't change during the VOP_FSYNC.
4612 	 */
4613 	if (obdva.va_seq && oidva.va_seq && oadva.va_seq &&
4614 	    oidva.va_seq == (obdva.va_seq + 1) && oidva.va_seq == oadva.va_seq)
4615 		resp->source_cinfo.atomic = TRUE;
4616 	else
4617 		resp->source_cinfo.atomic = FALSE;
4618 
4619 	if (nbdva.va_seq && nidva.va_seq && nadva.va_seq &&
4620 	    nidva.va_seq == (nbdva.va_seq + 1) && nidva.va_seq == nadva.va_seq)
4621 		resp->target_cinfo.atomic = TRUE;
4622 	else
4623 		resp->target_cinfo.atomic = FALSE;
4624 
4625 #ifdef	VOLATILE_FH_TEST
4626 	{
4627 	extern void add_volrnm_fh(struct exportinfo *, vnode_t *);
4628 
4629 	/*
4630 	 * Add the renamed file handle to the volatile rename list
4631 	 */
4632 	if (cs->exi->exi_export.ex_flags & EX_VOLRNM) {
4633 		/* file handles may expire on rename */
4634 		vnode_t *vp;
4635 
4636 		nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4637 		/*
4638 		 * Already know that nnm will be a valid string
4639 		 */
4640 		error = VOP_LOOKUP(ndvp, nnm, &vp, NULL, 0, NULL, cs->cr,
4641 		    NULL, NULL, NULL);
4642 		kmem_free(nnm, nlen);
4643 		if (!error) {
4644 			add_volrnm_fh(cs->exi, vp);
4645 			VN_RELE(vp);
4646 		}
4647 	}
4648 	}
4649 #endif	/* VOLATILE_FH_TEST */
4650 
4651 	*cs->statusp = resp->status = NFS4_OK;
4652 out:
4653 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4654 	    RENAME4res *, resp);
4655 	return;
4656 
4657 err_out:
4658 	if (onm != converted_onm)
4659 		kmem_free(converted_onm, MAXPATHLEN + 1);
4660 	if (onm != NULL)
4661 		kmem_free(onm, olen);
4662 	if (nnm != converted_nnm)
4663 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4664 	if (nnm != NULL)
4665 		kmem_free(nnm, nlen);
4666 
4667 	if (in_crit_src) nbl_end_crit(srcvp);
4668 	if (in_crit_targ) nbl_end_crit(targvp);
4669 	if (targvp) VN_RELE(targvp);
4670 	if (srcvp) VN_RELE(srcvp);
4671 	if (sfp) {
4672 		if (sfp_rele_grant_hold) rfs4_clear_dont_grant(sfp);
4673 		rfs4_file_rele(sfp);
4674 	}
4675 	if (fp) {
4676 		if (fp_rele_grant_hold) rfs4_clear_dont_grant(fp);
4677 		rfs4_file_rele(fp);
4678 	}
4679 
4680 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4681 	    RENAME4res *, resp);
4682 }
4683 
4684 /* ARGSUSED */
4685 static void
4686 rfs4_op_renew(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4687     struct compound_state *cs)
4688 {
4689 	RENEW4args *args = &argop->nfs_argop4_u.oprenew;
4690 	RENEW4res *resp = &resop->nfs_resop4_u.oprenew;
4691 	rfs4_client_t *cp;
4692 
4693 	DTRACE_NFSV4_2(op__renew__start, struct compound_state *, cs,
4694 	    RENEW4args *, args);
4695 
4696 	if ((cp = rfs4_findclient_by_id(args->clientid, FALSE)) == NULL) {
4697 		*cs->statusp = resp->status =
4698 		    rfs4_check_clientid(&args->clientid, 0);
4699 		goto out;
4700 	}
4701 
4702 	if (rfs4_lease_expired(cp)) {
4703 		rfs4_client_rele(cp);
4704 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
4705 		goto out;
4706 	}
4707 
4708 	rfs4_update_lease(cp);
4709 
4710 	mutex_enter(cp->rc_cbinfo.cb_lock);
4711 	if (cp->rc_cbinfo.cb_notified_of_cb_path_down == FALSE) {
4712 		cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
4713 		*cs->statusp = resp->status = NFS4ERR_CB_PATH_DOWN;
4714 	} else {
4715 		*cs->statusp = resp->status = NFS4_OK;
4716 	}
4717 	mutex_exit(cp->rc_cbinfo.cb_lock);
4718 
4719 	rfs4_client_rele(cp);
4720 
4721 out:
4722 	DTRACE_NFSV4_2(op__renew__done, struct compound_state *, cs,
4723 	    RENEW4res *, resp);
4724 }
4725 
4726 /* ARGSUSED */
4727 static void
4728 rfs4_op_restorefh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
4729     struct compound_state *cs)
4730 {
4731 	RESTOREFH4res *resp = &resop->nfs_resop4_u.oprestorefh;
4732 
4733 	DTRACE_NFSV4_1(op__restorefh__start, struct compound_state *, cs);
4734 
4735 	/* No need to check cs->access - we are not accessing any object */
4736 	if ((cs->saved_vp == NULL) || (cs->saved_fh.nfs_fh4_val == NULL)) {
4737 		*cs->statusp = resp->status = NFS4ERR_RESTOREFH;
4738 		goto out;
4739 	}
4740 	if (cs->vp != NULL) {
4741 		VN_RELE(cs->vp);
4742 	}
4743 	cs->vp = cs->saved_vp;
4744 	cs->saved_vp = NULL;
4745 	cs->exi = cs->saved_exi;
4746 	nfs_fh4_copy(&cs->saved_fh, &cs->fh);
4747 	*cs->statusp = resp->status = NFS4_OK;
4748 	cs->deleg = FALSE;
4749 
4750 out:
4751 	DTRACE_NFSV4_2(op__restorefh__done, struct compound_state *, cs,
4752 	    RESTOREFH4res *, resp);
4753 }
4754 
4755 /* ARGSUSED */
4756 static void
4757 rfs4_op_savefh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4758     struct compound_state *cs)
4759 {
4760 	SAVEFH4res *resp = &resop->nfs_resop4_u.opsavefh;
4761 
4762 	DTRACE_NFSV4_1(op__savefh__start, struct compound_state *, cs);
4763 
4764 	/* No need to check cs->access - we are not accessing any object */
4765 	if (cs->vp == NULL) {
4766 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4767 		goto out;
4768 	}
4769 	if (cs->saved_vp != NULL) {
4770 		VN_RELE(cs->saved_vp);
4771 	}
4772 	cs->saved_vp = cs->vp;
4773 	VN_HOLD(cs->saved_vp);
4774 	cs->saved_exi = cs->exi;
4775 	/*
4776 	 * since SAVEFH is fairly rare, don't alloc space for its fh
4777 	 * unless necessary.
4778 	 */
4779 	if (cs->saved_fh.nfs_fh4_val == NULL) {
4780 		cs->saved_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP);
4781 	}
4782 	nfs_fh4_copy(&cs->fh, &cs->saved_fh);
4783 	*cs->statusp = resp->status = NFS4_OK;
4784 
4785 out:
4786 	DTRACE_NFSV4_2(op__savefh__done, struct compound_state *, cs,
4787 	    SAVEFH4res *, resp);
4788 }
4789 
4790 /*
4791  * rfs4_verify_attr is called when nfsv4 Setattr failed, but we wish to
4792  * return the bitmap of attrs that were set successfully. It is also
4793  * called by Verify/Nverify to test the vattr/vfsstat attrs. It should
4794  * always be called only after rfs4_do_set_attrs().
4795  *
4796  * Verify that the attributes are same as the expected ones. sargp->vap
4797  * and sargp->sbp contain the input attributes as translated from fattr4.
4798  *
4799  * This function verifies only the attrs that correspond to a vattr or
4800  * vfsstat struct. That is because of the extra step needed to get the
4801  * corresponding system structs. Other attributes have already been set or
4802  * verified by do_rfs4_set_attrs.
4803  *
4804  * Return 0 if all attrs match, -1 if some don't, error if error processing.
4805  */
4806 static int
4807 rfs4_verify_attr(struct nfs4_svgetit_arg *sargp,
4808     bitmap4 *resp, struct nfs4_ntov_table *ntovp)
4809 {
4810 	int error, ret_error = 0;
4811 	int i, k;
4812 	uint_t sva_mask = sargp->vap->va_mask;
4813 	uint_t vbit;
4814 	union nfs4_attr_u *na;
4815 	uint8_t *amap;
4816 	bool_t getsb = ntovp->vfsstat;
4817 
4818 	if (sva_mask != 0) {
4819 		/*
4820 		 * Okay to overwrite sargp->vap because we verify based
4821 		 * on the incoming values.
4822 		 */
4823 		ret_error = VOP_GETATTR(sargp->cs->vp, sargp->vap, 0,
4824 		    sargp->cs->cr, NULL);
4825 		if (ret_error) {
4826 			if (resp == NULL)
4827 				return (ret_error);
4828 			/*
4829 			 * Must return bitmap of successful attrs
4830 			 */
4831 			sva_mask = 0;	/* to prevent checking vap later */
4832 		} else {
4833 			/*
4834 			 * Some file systems clobber va_mask. it is probably
4835 			 * wrong of them to do so, nonethless we practice
4836 			 * defensive coding.
4837 			 * See bug id 4276830.
4838 			 */
4839 			sargp->vap->va_mask = sva_mask;
4840 		}
4841 	}
4842 
4843 	if (getsb) {
4844 		/*
4845 		 * Now get the superblock and loop on the bitmap, as there is
4846 		 * no simple way of translating from superblock to bitmap4.
4847 		 */
4848 		ret_error = VFS_STATVFS(sargp->cs->vp->v_vfsp, sargp->sbp);
4849 		if (ret_error) {
4850 			if (resp == NULL)
4851 				goto errout;
4852 			getsb = FALSE;
4853 		}
4854 	}
4855 
4856 	/*
4857 	 * Now loop and verify each attribute which getattr returned
4858 	 * whether it's the same as the input.
4859 	 */
4860 	if (resp == NULL && !getsb && (sva_mask == 0))
4861 		goto errout;
4862 
4863 	na = ntovp->na;
4864 	amap = ntovp->amap;
4865 	k = 0;
4866 	for (i = 0; i < ntovp->attrcnt; i++, na++, amap++) {
4867 		k = *amap;
4868 		ASSERT(nfs4_ntov_map[k].nval == k);
4869 		vbit = nfs4_ntov_map[k].vbit;
4870 
4871 		/*
4872 		 * If vattr attribute but VOP_GETATTR failed, or it's
4873 		 * superblock attribute but VFS_STATVFS failed, skip
4874 		 */
4875 		if (vbit) {
4876 			if ((vbit & sva_mask) == 0)
4877 				continue;
4878 		} else if (!(getsb && nfs4_ntov_map[k].vfsstat)) {
4879 			continue;
4880 		}
4881 		error = (*nfs4_ntov_map[k].sv_getit)(NFS4ATTR_VERIT, sargp, na);
4882 		if (resp != NULL) {
4883 			if (error)
4884 				ret_error = -1;	/* not all match */
4885 			else	/* update response bitmap */
4886 				*resp |= nfs4_ntov_map[k].fbit;
4887 			continue;
4888 		}
4889 		if (error) {
4890 			ret_error = -1;	/* not all match */
4891 			break;
4892 		}
4893 	}
4894 errout:
4895 	return (ret_error);
4896 }
4897 
4898 /*
4899  * Decode the attribute to be set/verified. If the attr requires a sys op
4900  * (VOP_GETATTR, VFS_VFSSTAT), and the request is to verify, then don't
4901  * call the sv_getit function for it, because the sys op hasn't yet been done.
4902  * Return 0 for success, error code if failed.
4903  *
4904  * Note: the decoded arg is not freed here but in nfs4_ntov_table_free.
4905  */
4906 static int
4907 decode_fattr4_attr(nfs4_attr_cmd_t cmd, struct nfs4_svgetit_arg *sargp,
4908     int k, XDR *xdrp, bitmap4 *resp_bval, union nfs4_attr_u *nap)
4909 {
4910 	int error = 0;
4911 	bool_t set_later;
4912 
4913 	sargp->vap->va_mask |= nfs4_ntov_map[k].vbit;
4914 
4915 	if ((*nfs4_ntov_map[k].xfunc)(xdrp, nap)) {
4916 		set_later = nfs4_ntov_map[k].vbit || nfs4_ntov_map[k].vfsstat;
4917 		/*
4918 		 * don't verify yet if a vattr or sb dependent attr,
4919 		 * because we don't have their sys values yet.
4920 		 * Will be done later.
4921 		 */
4922 		if (! (set_later && (cmd == NFS4ATTR_VERIT))) {
4923 			/*
4924 			 * ACLs are a special case, since setting the MODE
4925 			 * conflicts with setting the ACL.  We delay setting
4926 			 * the ACL until all other attributes have been set.
4927 			 * The ACL gets set in do_rfs4_op_setattr().
4928 			 */
4929 			if (nfs4_ntov_map[k].fbit != FATTR4_ACL_MASK) {
4930 				error = (*nfs4_ntov_map[k].sv_getit)(cmd,
4931 				    sargp, nap);
4932 				if (error) {
4933 					xdr_free(nfs4_ntov_map[k].xfunc,
4934 					    (caddr_t)nap);
4935 				}
4936 			}
4937 		}
4938 	} else {
4939 #ifdef  DEBUG
4940 		cmn_err(CE_NOTE, "decode_fattr4_attr: error "
4941 		    "decoding attribute %d\n", k);
4942 #endif
4943 		error = EINVAL;
4944 	}
4945 	if (!error && resp_bval && !set_later) {
4946 		*resp_bval |= nfs4_ntov_map[k].fbit;
4947 	}
4948 
4949 	return (error);
4950 }
4951 
4952 /*
4953  * Set vattr based on incoming fattr4 attrs - used by setattr.
4954  * Set response mask. Ignore any values that are not writable vattr attrs.
4955  */
4956 static nfsstat4
4957 do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
4958     struct nfs4_svgetit_arg *sargp, struct nfs4_ntov_table *ntovp,
4959     nfs4_attr_cmd_t cmd)
4960 {
4961 	int error = 0;
4962 	int i;
4963 	char *attrs = fattrp->attrlist4;
4964 	uint32_t attrslen = fattrp->attrlist4_len;
4965 	XDR xdr;
4966 	nfsstat4 status = NFS4_OK;
4967 	vnode_t *vp = cs->vp;
4968 	union nfs4_attr_u *na;
4969 	uint8_t *amap;
4970 
4971 #ifndef lint
4972 	/*
4973 	 * Make sure that maximum attribute number can be expressed as an
4974 	 * 8 bit quantity.
4975 	 */
4976 	ASSERT(NFS4_MAXNUM_ATTRS <= (UINT8_MAX + 1));
4977 #endif
4978 
4979 	if (vp == NULL) {
4980 		if (resp)
4981 			*resp = 0;
4982 		return (NFS4ERR_NOFILEHANDLE);
4983 	}
4984 	if (cs->access == CS_ACCESS_DENIED) {
4985 		if (resp)
4986 			*resp = 0;
4987 		return (NFS4ERR_ACCESS);
4988 	}
4989 
4990 	sargp->op = cmd;
4991 	sargp->cs = cs;
4992 	sargp->flag = 0;	/* may be set later */
4993 	sargp->vap->va_mask = 0;
4994 	sargp->rdattr_error = NFS4_OK;
4995 	sargp->rdattr_error_req = FALSE;
4996 	/* sargp->sbp is set by the caller */
4997 
4998 	xdrmem_create(&xdr, attrs, attrslen, XDR_DECODE);
4999 
5000 	na = ntovp->na;
5001 	amap = ntovp->amap;
5002 
5003 	/*
5004 	 * The following loop iterates on the nfs4_ntov_map checking
5005 	 * if the fbit is set in the requested bitmap.
5006 	 * If set then we process the arguments using the
5007 	 * rfs4_fattr4 conversion functions to populate the setattr
5008 	 * vattr and va_mask. Any settable attrs that are not using vattr
5009 	 * will be set in this loop.
5010 	 */
5011 	for (i = 0; i < nfs4_ntov_map_size; i++) {
5012 		if (!(fattrp->attrmask & nfs4_ntov_map[i].fbit)) {
5013 			continue;
5014 		}
5015 		/*
5016 		 * If setattr, must be a writable attr.
5017 		 * If verify/nverify, must be a readable attr.
5018 		 */
5019 		if ((error = (*nfs4_ntov_map[i].sv_getit)(
5020 		    NFS4ATTR_SUPPORTED, sargp, NULL)) != 0) {
5021 			/*
5022 			 * Client tries to set/verify an
5023 			 * unsupported attribute, tries to set
5024 			 * a read only attr or verify a write
5025 			 * only one - error!
5026 			 */
5027 			break;
5028 		}
5029 		/*
5030 		 * Decode the attribute to set/verify
5031 		 */
5032 		error = decode_fattr4_attr(cmd, sargp, nfs4_ntov_map[i].nval,
5033 		    &xdr, resp ? resp : NULL, na);
5034 		if (error)
5035 			break;
5036 		*amap++ = (uint8_t)nfs4_ntov_map[i].nval;
5037 		na++;
5038 		(ntovp->attrcnt)++;
5039 		if (nfs4_ntov_map[i].vfsstat)
5040 			ntovp->vfsstat = TRUE;
5041 	}
5042 
5043 	if (error != 0)
5044 		status = (error == ENOTSUP ? NFS4ERR_ATTRNOTSUPP :
5045 		    puterrno4(error));
5046 	/* xdrmem_destroy(&xdrs); */	/* NO-OP */
5047 	return (status);
5048 }
5049 
5050 static nfsstat4
5051 do_rfs4_op_setattr(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5052     stateid4 *stateid)
5053 {
5054 	int error = 0;
5055 	struct nfs4_svgetit_arg sarg;
5056 	bool_t trunc;
5057 
5058 	nfsstat4 status = NFS4_OK;
5059 	cred_t *cr = cs->cr;
5060 	vnode_t *vp = cs->vp;
5061 	struct nfs4_ntov_table ntov;
5062 	struct statvfs64 sb;
5063 	struct vattr bva;
5064 	struct flock64 bf;
5065 	int in_crit = 0;
5066 	uint_t saved_mask = 0;
5067 	caller_context_t ct;
5068 
5069 	*resp = 0;
5070 	sarg.sbp = &sb;
5071 	sarg.is_referral = B_FALSE;
5072 	nfs4_ntov_table_init(&ntov);
5073 	status = do_rfs4_set_attrs(resp, fattrp, cs, &sarg, &ntov,
5074 	    NFS4ATTR_SETIT);
5075 	if (status != NFS4_OK) {
5076 		/*
5077 		 * failed set attrs
5078 		 */
5079 		goto done;
5080 	}
5081 	if ((sarg.vap->va_mask == 0) &&
5082 	    (! (fattrp->attrmask & FATTR4_ACL_MASK))) {
5083 		/*
5084 		 * no further work to be done
5085 		 */
5086 		goto done;
5087 	}
5088 
5089 	/*
5090 	 * If we got a request to set the ACL and the MODE, only
5091 	 * allow changing VSUID, VSGID, and VSVTX.  Attempting
5092 	 * to change any other bits, along with setting an ACL,
5093 	 * gives NFS4ERR_INVAL.
5094 	 */
5095 	if ((fattrp->attrmask & FATTR4_ACL_MASK) &&
5096 	    (fattrp->attrmask & FATTR4_MODE_MASK)) {
5097 		vattr_t va;
5098 
5099 		va.va_mask = AT_MODE;
5100 		error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
5101 		if (error) {
5102 			status = puterrno4(error);
5103 			goto done;
5104 		}
5105 		if ((sarg.vap->va_mode ^ va.va_mode) &
5106 		    ~(VSUID | VSGID | VSVTX)) {
5107 			status = NFS4ERR_INVAL;
5108 			goto done;
5109 		}
5110 	}
5111 
5112 	/* Check stateid only if size has been set */
5113 	if (sarg.vap->va_mask & AT_SIZE) {
5114 		trunc = (sarg.vap->va_size == 0);
5115 		status = rfs4_check_stateid(FWRITE, cs->vp, stateid,
5116 		    trunc, &cs->deleg, sarg.vap->va_mask & AT_SIZE, &ct);
5117 		if (status != NFS4_OK)
5118 			goto done;
5119 	} else {
5120 		ct.cc_sysid = 0;
5121 		ct.cc_pid = 0;
5122 		ct.cc_caller_id = nfs4_srv_caller_id;
5123 		ct.cc_flags = CC_DONTBLOCK;
5124 	}
5125 
5126 	/* XXX start of possible race with delegations */
5127 
5128 	/*
5129 	 * We need to specially handle size changes because it is
5130 	 * possible for the client to create a file with read-only
5131 	 * modes, but with the file opened for writing. If the client
5132 	 * then tries to set the file size, e.g. ftruncate(3C),
5133 	 * fcntl(F_FREESP), the normal access checking done in
5134 	 * VOP_SETATTR would prevent the client from doing it even though
5135 	 * it should be allowed to do so.  To get around this, we do the
5136 	 * access checking for ourselves and use VOP_SPACE which doesn't
5137 	 * do the access checking.
5138 	 * Also the client should not be allowed to change the file
5139 	 * size if there is a conflicting non-blocking mandatory lock in
5140 	 * the region of the change.
5141 	 */
5142 	if (vp->v_type == VREG && (sarg.vap->va_mask & AT_SIZE)) {
5143 		u_offset_t offset;
5144 		ssize_t length;
5145 
5146 		/*
5147 		 * ufs_setattr clears AT_SIZE from vap->va_mask, but
5148 		 * before returning, sarg.vap->va_mask is used to
5149 		 * generate the setattr reply bitmap.  We also clear
5150 		 * AT_SIZE below before calling VOP_SPACE.  For both
5151 		 * of these cases, the va_mask needs to be saved here
5152 		 * and restored after calling VOP_SETATTR.
5153 		 */
5154 		saved_mask = sarg.vap->va_mask;
5155 
5156 		/*
5157 		 * Check any possible conflict due to NBMAND locks.
5158 		 * Get into critical region before VOP_GETATTR, so the
5159 		 * size attribute is valid when checking conflicts.
5160 		 */
5161 		if (nbl_need_check(vp)) {
5162 			nbl_start_crit(vp, RW_READER);
5163 			in_crit = 1;
5164 		}
5165 
5166 		bva.va_mask = AT_UID|AT_SIZE;
5167 		if (error = VOP_GETATTR(vp, &bva, 0, cr, &ct)) {
5168 			status = puterrno4(error);
5169 			goto done;
5170 		}
5171 
5172 		if (in_crit) {
5173 			if (sarg.vap->va_size < bva.va_size) {
5174 				offset = sarg.vap->va_size;
5175 				length = bva.va_size - sarg.vap->va_size;
5176 			} else {
5177 				offset = bva.va_size;
5178 				length = sarg.vap->va_size - bva.va_size;
5179 			}
5180 			if (nbl_conflict(vp, NBL_WRITE, offset, length, 0,
5181 			    &ct)) {
5182 				status = NFS4ERR_LOCKED;
5183 				goto done;
5184 			}
5185 		}
5186 
5187 		if (crgetuid(cr) == bva.va_uid) {
5188 			sarg.vap->va_mask &= ~AT_SIZE;
5189 			bf.l_type = F_WRLCK;
5190 			bf.l_whence = 0;
5191 			bf.l_start = (off64_t)sarg.vap->va_size;
5192 			bf.l_len = 0;
5193 			bf.l_sysid = 0;
5194 			bf.l_pid = 0;
5195 			error = VOP_SPACE(vp, F_FREESP, &bf, FWRITE,
5196 			    (offset_t)sarg.vap->va_size, cr, &ct);
5197 		}
5198 	}
5199 
5200 	if (!error && sarg.vap->va_mask != 0)
5201 		error = VOP_SETATTR(vp, sarg.vap, sarg.flag, cr, &ct);
5202 
5203 	/* restore va_mask -- ufs_setattr clears AT_SIZE */
5204 	if (saved_mask & AT_SIZE)
5205 		sarg.vap->va_mask |= AT_SIZE;
5206 
5207 	/*
5208 	 * If an ACL was being set, it has been delayed until now,
5209 	 * in order to set the mode (via the VOP_SETATTR() above) first.
5210 	 */
5211 	if ((! error) && (fattrp->attrmask & FATTR4_ACL_MASK)) {
5212 		int i;
5213 
5214 		for (i = 0; i < NFS4_MAXNUM_ATTRS; i++)
5215 			if (ntov.amap[i] == FATTR4_ACL)
5216 				break;
5217 		if (i < NFS4_MAXNUM_ATTRS) {
5218 			error = (*nfs4_ntov_map[FATTR4_ACL].sv_getit)(
5219 			    NFS4ATTR_SETIT, &sarg, &ntov.na[i]);
5220 			if (error == 0) {
5221 				*resp |= FATTR4_ACL_MASK;
5222 			} else if (error == ENOTSUP) {
5223 				(void) rfs4_verify_attr(&sarg, resp, &ntov);
5224 				status = NFS4ERR_ATTRNOTSUPP;
5225 				goto done;
5226 			}
5227 		} else {
5228 			NFS4_DEBUG(rfs4_debug,
5229 			    (CE_NOTE, "do_rfs4_op_setattr: "
5230 			    "unable to find ACL in fattr4"));
5231 			error = EINVAL;
5232 		}
5233 	}
5234 
5235 	if (error) {
5236 		/* check if a monitor detected a delegation conflict */
5237 		if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
5238 			status = NFS4ERR_DELAY;
5239 		else
5240 			status = puterrno4(error);
5241 
5242 		/*
5243 		 * Set the response bitmap when setattr failed.
5244 		 * If VOP_SETATTR partially succeeded, test by doing a
5245 		 * VOP_GETATTR on the object and comparing the data
5246 		 * to the setattr arguments.
5247 		 */
5248 		(void) rfs4_verify_attr(&sarg, resp, &ntov);
5249 	} else {
5250 		/*
5251 		 * Force modified metadata out to stable storage.
5252 		 */
5253 		(void) VOP_FSYNC(vp, FNODSYNC, cr, &ct);
5254 		/*
5255 		 * Set response bitmap
5256 		 */
5257 		nfs4_vmask_to_nmask_set(sarg.vap->va_mask, resp);
5258 	}
5259 
5260 /* Return early and already have a NFSv4 error */
5261 done:
5262 	/*
5263 	 * Except for nfs4_vmask_to_nmask_set(), vattr --> fattr
5264 	 * conversion sets both readable and writeable NFS4 attrs
5265 	 * for AT_MTIME and AT_ATIME.  The line below masks out
5266 	 * unrequested attrs from the setattr result bitmap.  This
5267 	 * is placed after the done: label to catch the ATTRNOTSUP
5268 	 * case.
5269 	 */
5270 	*resp &= fattrp->attrmask;
5271 
5272 	if (in_crit)
5273 		nbl_end_crit(vp);
5274 
5275 	nfs4_ntov_table_free(&ntov, &sarg);
5276 
5277 	return (status);
5278 }
5279 
5280 /* ARGSUSED */
5281 static void
5282 rfs4_op_setattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5283     struct compound_state *cs)
5284 {
5285 	SETATTR4args *args = &argop->nfs_argop4_u.opsetattr;
5286 	SETATTR4res *resp = &resop->nfs_resop4_u.opsetattr;
5287 	bslabel_t *clabel;
5288 
5289 	DTRACE_NFSV4_2(op__setattr__start, struct compound_state *, cs,
5290 	    SETATTR4args *, args);
5291 
5292 	if (cs->vp == NULL) {
5293 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5294 		goto out;
5295 	}
5296 
5297 	/*
5298 	 * If there is an unshared filesystem mounted on this vnode,
5299 	 * do not allow to setattr on this vnode.
5300 	 */
5301 	if (vn_ismntpt(cs->vp)) {
5302 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5303 		goto out;
5304 	}
5305 
5306 	resp->attrsset = 0;
5307 
5308 	if (rdonly4(req, cs)) {
5309 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5310 		goto out;
5311 	}
5312 
5313 	/* check label before setting attributes */
5314 	if (is_system_labeled()) {
5315 		ASSERT(req->rq_label != NULL);
5316 		clabel = req->rq_label;
5317 		DTRACE_PROBE2(tx__rfs4__log__info__opsetattr__clabel, char *,
5318 		    "got client label from request(1)",
5319 		    struct svc_req *, req);
5320 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
5321 			if (!do_rfs_label_check(clabel, cs->vp,
5322 			    EQUALITY_CHECK, cs->exi)) {
5323 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
5324 				goto out;
5325 			}
5326 		}
5327 	}
5328 
5329 	*cs->statusp = resp->status =
5330 	    do_rfs4_op_setattr(&resp->attrsset, &args->obj_attributes, cs,
5331 	    &args->stateid);
5332 
5333 out:
5334 	DTRACE_NFSV4_2(op__setattr__done, struct compound_state *, cs,
5335 	    SETATTR4res *, resp);
5336 }
5337 
5338 /* ARGSUSED */
5339 static void
5340 rfs4_op_verify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5341     struct compound_state *cs)
5342 {
5343 	/*
5344 	 * verify and nverify are exactly the same, except that nverify
5345 	 * succeeds when some argument changed, and verify succeeds when
5346 	 * when none changed.
5347 	 */
5348 
5349 	VERIFY4args  *args = &argop->nfs_argop4_u.opverify;
5350 	VERIFY4res *resp = &resop->nfs_resop4_u.opverify;
5351 
5352 	int error;
5353 	struct nfs4_svgetit_arg sarg;
5354 	struct statvfs64 sb;
5355 	struct nfs4_ntov_table ntov;
5356 
5357 	DTRACE_NFSV4_2(op__verify__start, struct compound_state *, cs,
5358 	    VERIFY4args *, args);
5359 
5360 	if (cs->vp == NULL) {
5361 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5362 		goto out;
5363 	}
5364 
5365 	sarg.sbp = &sb;
5366 	sarg.is_referral = B_FALSE;
5367 	nfs4_ntov_table_init(&ntov);
5368 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5369 	    &sarg, &ntov, NFS4ATTR_VERIT);
5370 	if (resp->status != NFS4_OK) {
5371 		/*
5372 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5373 		 * so could return -1 for "no match".
5374 		 */
5375 		if (resp->status == -1)
5376 			resp->status = NFS4ERR_NOT_SAME;
5377 		goto done;
5378 	}
5379 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5380 	switch (error) {
5381 	case 0:
5382 		resp->status = NFS4_OK;
5383 		break;
5384 	case -1:
5385 		resp->status = NFS4ERR_NOT_SAME;
5386 		break;
5387 	default:
5388 		resp->status = puterrno4(error);
5389 		break;
5390 	}
5391 done:
5392 	*cs->statusp = resp->status;
5393 	nfs4_ntov_table_free(&ntov, &sarg);
5394 out:
5395 	DTRACE_NFSV4_2(op__verify__done, struct compound_state *, cs,
5396 	    VERIFY4res *, resp);
5397 }
5398 
5399 /* ARGSUSED */
5400 static void
5401 rfs4_op_nverify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5402     struct compound_state *cs)
5403 {
5404 	/*
5405 	 * verify and nverify are exactly the same, except that nverify
5406 	 * succeeds when some argument changed, and verify succeeds when
5407 	 * when none changed.
5408 	 */
5409 
5410 	NVERIFY4args  *args = &argop->nfs_argop4_u.opnverify;
5411 	NVERIFY4res *resp = &resop->nfs_resop4_u.opnverify;
5412 
5413 	int error;
5414 	struct nfs4_svgetit_arg sarg;
5415 	struct statvfs64 sb;
5416 	struct nfs4_ntov_table ntov;
5417 
5418 	DTRACE_NFSV4_2(op__nverify__start, struct compound_state *, cs,
5419 	    NVERIFY4args *, args);
5420 
5421 	if (cs->vp == NULL) {
5422 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5423 		DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5424 		    NVERIFY4res *, resp);
5425 		return;
5426 	}
5427 	sarg.sbp = &sb;
5428 	sarg.is_referral = B_FALSE;
5429 	nfs4_ntov_table_init(&ntov);
5430 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5431 	    &sarg, &ntov, NFS4ATTR_VERIT);
5432 	if (resp->status != NFS4_OK) {
5433 		/*
5434 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5435 		 * so could return -1 for "no match".
5436 		 */
5437 		if (resp->status == -1)
5438 			resp->status = NFS4_OK;
5439 		goto done;
5440 	}
5441 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5442 	switch (error) {
5443 	case 0:
5444 		resp->status = NFS4ERR_SAME;
5445 		break;
5446 	case -1:
5447 		resp->status = NFS4_OK;
5448 		break;
5449 	default:
5450 		resp->status = puterrno4(error);
5451 		break;
5452 	}
5453 done:
5454 	*cs->statusp = resp->status;
5455 	nfs4_ntov_table_free(&ntov, &sarg);
5456 
5457 	DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5458 	    NVERIFY4res *, resp);
5459 }
5460 
5461 /*
5462  * XXX - This should live in an NFS header file.
5463  */
5464 #define	MAX_IOVECS	12
5465 
5466 /* ARGSUSED */
5467 static void
5468 rfs4_op_write(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5469     struct compound_state *cs)
5470 {
5471 	WRITE4args *args = &argop->nfs_argop4_u.opwrite;
5472 	WRITE4res *resp = &resop->nfs_resop4_u.opwrite;
5473 	int error;
5474 	vnode_t *vp;
5475 	struct vattr bva;
5476 	u_offset_t rlimit;
5477 	struct uio uio;
5478 	struct iovec iov[MAX_IOVECS];
5479 	struct iovec *iovp;
5480 	int iovcnt;
5481 	int ioflag;
5482 	cred_t *savecred, *cr;
5483 	bool_t *deleg = &cs->deleg;
5484 	nfsstat4 stat;
5485 	int in_crit = 0;
5486 	caller_context_t ct;
5487 
5488 	DTRACE_NFSV4_2(op__write__start, struct compound_state *, cs,
5489 	    WRITE4args *, args);
5490 
5491 	vp = cs->vp;
5492 	if (vp == NULL) {
5493 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5494 		goto out;
5495 	}
5496 	if (cs->access == CS_ACCESS_DENIED) {
5497 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5498 		goto out;
5499 	}
5500 
5501 	cr = cs->cr;
5502 
5503 	if ((stat = rfs4_check_stateid(FWRITE, vp, &args->stateid, FALSE,
5504 	    deleg, TRUE, &ct)) != NFS4_OK) {
5505 		*cs->statusp = resp->status = stat;
5506 		goto out;
5507 	}
5508 
5509 	/*
5510 	 * We have to enter the critical region before calling VOP_RWLOCK
5511 	 * to avoid a deadlock with ufs.
5512 	 */
5513 	if (nbl_need_check(vp)) {
5514 		nbl_start_crit(vp, RW_READER);
5515 		in_crit = 1;
5516 		if (nbl_conflict(vp, NBL_WRITE,
5517 		    args->offset, args->data_len, 0, &ct)) {
5518 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
5519 			goto out;
5520 		}
5521 	}
5522 
5523 	bva.va_mask = AT_MODE | AT_UID;
5524 	error = VOP_GETATTR(vp, &bva, 0, cr, &ct);
5525 
5526 	/*
5527 	 * If we can't get the attributes, then we can't do the
5528 	 * right access checking.  So, we'll fail the request.
5529 	 */
5530 	if (error) {
5531 		*cs->statusp = resp->status = puterrno4(error);
5532 		goto out;
5533 	}
5534 
5535 	if (rdonly4(req, cs)) {
5536 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5537 		goto out;
5538 	}
5539 
5540 	if (vp->v_type != VREG) {
5541 		*cs->statusp = resp->status =
5542 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
5543 		goto out;
5544 	}
5545 
5546 	if (crgetuid(cr) != bva.va_uid &&
5547 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, &ct))) {
5548 		*cs->statusp = resp->status = puterrno4(error);
5549 		goto out;
5550 	}
5551 
5552 	if (MANDLOCK(vp, bva.va_mode)) {
5553 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5554 		goto out;
5555 	}
5556 
5557 	if (args->data_len == 0) {
5558 		*cs->statusp = resp->status = NFS4_OK;
5559 		resp->count = 0;
5560 		resp->committed = args->stable;
5561 		resp->writeverf = Write4verf;
5562 		goto out;
5563 	}
5564 
5565 	if (args->mblk != NULL) {
5566 		mblk_t *m;
5567 		uint_t bytes, round_len;
5568 
5569 		iovcnt = 0;
5570 		bytes = 0;
5571 		round_len = roundup(args->data_len, BYTES_PER_XDR_UNIT);
5572 		for (m = args->mblk;
5573 		    m != NULL && bytes < round_len;
5574 		    m = m->b_cont) {
5575 			iovcnt++;
5576 			bytes += MBLKL(m);
5577 		}
5578 #ifdef DEBUG
5579 		/* should have ended on an mblk boundary */
5580 		if (bytes != round_len) {
5581 			printf("bytes=0x%x, round_len=0x%x, req len=0x%x\n",
5582 			    bytes, round_len, args->data_len);
5583 			printf("args=%p, args->mblk=%p, m=%p", (void *)args,
5584 			    (void *)args->mblk, (void *)m);
5585 			ASSERT(bytes == round_len);
5586 		}
5587 #endif
5588 		if (iovcnt <= MAX_IOVECS) {
5589 			iovp = iov;
5590 		} else {
5591 			iovp = kmem_alloc(sizeof (*iovp) * iovcnt, KM_SLEEP);
5592 		}
5593 		mblk_to_iov(args->mblk, iovcnt, iovp);
5594 	} else if (args->rlist != NULL) {
5595 		iovcnt = 1;
5596 		iovp = iov;
5597 		iovp->iov_base = (char *)((args->rlist)->u.c_daddr3);
5598 		iovp->iov_len = args->data_len;
5599 	} else {
5600 		iovcnt = 1;
5601 		iovp = iov;
5602 		iovp->iov_base = args->data_val;
5603 		iovp->iov_len = args->data_len;
5604 	}
5605 
5606 	uio.uio_iov = iovp;
5607 	uio.uio_iovcnt = iovcnt;
5608 
5609 	uio.uio_segflg = UIO_SYSSPACE;
5610 	uio.uio_extflg = UIO_COPY_DEFAULT;
5611 	uio.uio_loffset = args->offset;
5612 	uio.uio_resid = args->data_len;
5613 	uio.uio_llimit = curproc->p_fsz_ctl;
5614 	rlimit = uio.uio_llimit - args->offset;
5615 	if (rlimit < (u_offset_t)uio.uio_resid)
5616 		uio.uio_resid = (int)rlimit;
5617 
5618 	if (args->stable == UNSTABLE4)
5619 		ioflag = 0;
5620 	else if (args->stable == FILE_SYNC4)
5621 		ioflag = FSYNC;
5622 	else if (args->stable == DATA_SYNC4)
5623 		ioflag = FDSYNC;
5624 	else {
5625 		if (iovp != iov)
5626 			kmem_free(iovp, sizeof (*iovp) * iovcnt);
5627 		*cs->statusp = resp->status = NFS4ERR_INVAL;
5628 		goto out;
5629 	}
5630 
5631 	/*
5632 	 * We're changing creds because VM may fault and we need
5633 	 * the cred of the current thread to be used if quota
5634 	 * checking is enabled.
5635 	 */
5636 	savecred = curthread->t_cred;
5637 	curthread->t_cred = cr;
5638 	error = do_io(FWRITE, vp, &uio, ioflag, cr, &ct);
5639 	curthread->t_cred = savecred;
5640 
5641 	if (iovp != iov)
5642 		kmem_free(iovp, sizeof (*iovp) * iovcnt);
5643 
5644 	if (error) {
5645 		*cs->statusp = resp->status = puterrno4(error);
5646 		goto out;
5647 	}
5648 
5649 	*cs->statusp = resp->status = NFS4_OK;
5650 	resp->count = args->data_len - uio.uio_resid;
5651 
5652 	if (ioflag == 0)
5653 		resp->committed = UNSTABLE4;
5654 	else
5655 		resp->committed = FILE_SYNC4;
5656 
5657 	resp->writeverf = Write4verf;
5658 
5659 out:
5660 	if (in_crit)
5661 		nbl_end_crit(vp);
5662 
5663 	DTRACE_NFSV4_2(op__write__done, struct compound_state *, cs,
5664 	    WRITE4res *, resp);
5665 }
5666 
5667 
5668 /* XXX put in a header file */
5669 extern int	sec_svc_getcred(struct svc_req *, cred_t *,  caddr_t *, int *);
5670 
5671 void
5672 rfs4_compound(COMPOUND4args *args, COMPOUND4res *resp, struct exportinfo *exi,
5673     struct svc_req *req, cred_t *cr, int *rv)
5674 {
5675 	uint_t i;
5676 	struct compound_state cs;
5677 
5678 	if (rv != NULL)
5679 		*rv = 0;
5680 	rfs4_init_compound_state(&cs);
5681 	/*
5682 	 * Form a reply tag by copying over the reqeuest tag.
5683 	 */
5684 	resp->tag.utf8string_val =
5685 	    kmem_alloc(args->tag.utf8string_len, KM_SLEEP);
5686 	resp->tag.utf8string_len = args->tag.utf8string_len;
5687 	bcopy(args->tag.utf8string_val, resp->tag.utf8string_val,
5688 	    resp->tag.utf8string_len);
5689 
5690 	cs.statusp = &resp->status;
5691 	cs.req = req;
5692 	resp->array = NULL;
5693 	resp->array_len = 0;
5694 
5695 	/*
5696 	 * XXX for now, minorversion should be zero
5697 	 */
5698 	if (args->minorversion != NFS4_MINORVERSION) {
5699 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5700 		    &cs, COMPOUND4args *, args);
5701 		resp->status = NFS4ERR_MINOR_VERS_MISMATCH;
5702 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5703 		    &cs, COMPOUND4res *, resp);
5704 		return;
5705 	}
5706 
5707 	if (args->array_len == 0) {
5708 		resp->status = NFS4_OK;
5709 		return;
5710 	}
5711 
5712 	ASSERT(exi == NULL);
5713 	ASSERT(cr == NULL);
5714 
5715 	cr = crget();
5716 	ASSERT(cr != NULL);
5717 
5718 	if (sec_svc_getcred(req, cr, &cs.principal, &cs.nfsflavor) == 0) {
5719 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5720 		    &cs, COMPOUND4args *, args);
5721 		crfree(cr);
5722 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5723 		    &cs, COMPOUND4res *, resp);
5724 		svcerr_badcred(req->rq_xprt);
5725 		if (rv != NULL)
5726 			*rv = 1;
5727 		return;
5728 	}
5729 	resp->array_len = args->array_len;
5730 	resp->array = kmem_zalloc(args->array_len * sizeof (nfs_resop4),
5731 	    KM_SLEEP);
5732 
5733 	cs.basecr = cr;
5734 
5735 	DTRACE_NFSV4_2(compound__start, struct compound_state *, &cs,
5736 	    COMPOUND4args *, args);
5737 
5738 	/*
5739 	 * For now, NFS4 compound processing must be protected by
5740 	 * exported_lock because it can access more than one exportinfo
5741 	 * per compound and share/unshare can now change multiple
5742 	 * exinfo structs.  The NFS2/3 code only refs 1 exportinfo
5743 	 * per proc (excluding public exinfo), and exi_count design
5744 	 * is sufficient to protect concurrent execution of NFS2/3
5745 	 * ops along with unexport.  This lock will be removed as
5746 	 * part of the NFSv4 phase 2 namespace redesign work.
5747 	 */
5748 	rw_enter(&exported_lock, RW_READER);
5749 
5750 	/*
5751 	 * If this is the first compound we've seen, we need to start all
5752 	 * new instances' grace periods.
5753 	 */
5754 	if (rfs4_seen_first_compound == 0) {
5755 		rfs4_grace_start_new();
5756 		/*
5757 		 * This must be set after rfs4_grace_start_new(), otherwise
5758 		 * another thread could proceed past here before the former
5759 		 * is finished.
5760 		 */
5761 		rfs4_seen_first_compound = 1;
5762 	}
5763 
5764 	for (i = 0; i < args->array_len && cs.cont; i++) {
5765 		nfs_argop4 *argop;
5766 		nfs_resop4 *resop;
5767 		uint_t op;
5768 
5769 		argop = &args->array[i];
5770 		resop = &resp->array[i];
5771 		resop->resop = argop->argop;
5772 		op = (uint_t)resop->resop;
5773 
5774 		if (op < rfsv4disp_cnt) {
5775 			/*
5776 			 * Count the individual ops here; NULL and COMPOUND
5777 			 * are counted in common_dispatch()
5778 			 */
5779 			rfsproccnt_v4_ptr[op].value.ui64++;
5780 
5781 			NFS4_DEBUG(rfs4_debug > 1,
5782 			    (CE_NOTE, "Executing %s", rfs4_op_string[op]));
5783 			(*rfsv4disptab[op].dis_proc)(argop, resop, req, &cs);
5784 			NFS4_DEBUG(rfs4_debug > 1, (CE_NOTE, "%s returned %d",
5785 			    rfs4_op_string[op], *cs.statusp));
5786 			if (*cs.statusp != NFS4_OK)
5787 				cs.cont = FALSE;
5788 		} else {
5789 			/*
5790 			 * This is effectively dead code since XDR code
5791 			 * will have already returned BADXDR if op doesn't
5792 			 * decode to legal value.  This only done for a
5793 			 * day when XDR code doesn't verify v4 opcodes.
5794 			 */
5795 			op = OP_ILLEGAL;
5796 			rfsproccnt_v4_ptr[OP_ILLEGAL_IDX].value.ui64++;
5797 
5798 			rfs4_op_illegal(argop, resop, req, &cs);
5799 			cs.cont = FALSE;
5800 		}
5801 
5802 		/*
5803 		 * If not at last op, and if we are to stop, then
5804 		 * compact the results array.
5805 		 */
5806 		if ((i + 1) < args->array_len && !cs.cont) {
5807 			nfs_resop4 *new_res = kmem_alloc(
5808 			    (i+1) * sizeof (nfs_resop4), KM_SLEEP);
5809 			bcopy(resp->array,
5810 			    new_res, (i+1) * sizeof (nfs_resop4));
5811 			kmem_free(resp->array,
5812 			    args->array_len * sizeof (nfs_resop4));
5813 
5814 			resp->array_len =  i + 1;
5815 			resp->array = new_res;
5816 		}
5817 	}
5818 
5819 	rw_exit(&exported_lock);
5820 
5821 	DTRACE_NFSV4_2(compound__done, struct compound_state *, &cs,
5822 	    COMPOUND4res *, resp);
5823 
5824 	if (cs.vp)
5825 		VN_RELE(cs.vp);
5826 	if (cs.saved_vp)
5827 		VN_RELE(cs.saved_vp);
5828 	if (cs.saved_fh.nfs_fh4_val)
5829 		kmem_free(cs.saved_fh.nfs_fh4_val, NFS4_FHSIZE);
5830 
5831 	if (cs.basecr)
5832 		crfree(cs.basecr);
5833 	if (cs.cr)
5834 		crfree(cs.cr);
5835 	/*
5836 	 * done with this compound request, free the label
5837 	 */
5838 
5839 	if (req->rq_label != NULL) {
5840 		kmem_free(req->rq_label, sizeof (bslabel_t));
5841 		req->rq_label = NULL;
5842 	}
5843 }
5844 
5845 /*
5846  * XXX because of what appears to be duplicate calls to rfs4_compound_free
5847  * XXX zero out the tag and array values. Need to investigate why the
5848  * XXX calls occur, but at least prevent the panic for now.
5849  */
5850 void
5851 rfs4_compound_free(COMPOUND4res *resp)
5852 {
5853 	uint_t i;
5854 
5855 	if (resp->tag.utf8string_val) {
5856 		UTF8STRING_FREE(resp->tag)
5857 	}
5858 
5859 	for (i = 0; i < resp->array_len; i++) {
5860 		nfs_resop4 *resop;
5861 		uint_t op;
5862 
5863 		resop = &resp->array[i];
5864 		op = (uint_t)resop->resop;
5865 		if (op < rfsv4disp_cnt) {
5866 			(*rfsv4disptab[op].dis_resfree)(resop);
5867 		}
5868 	}
5869 	if (resp->array != NULL) {
5870 		kmem_free(resp->array, resp->array_len * sizeof (nfs_resop4));
5871 	}
5872 }
5873 
5874 /*
5875  * Process the value of the compound request rpc flags, as a bit-AND
5876  * of the individual per-op flags (idempotent, allowork, publicfh_ok)
5877  */
5878 void
5879 rfs4_compound_flagproc(COMPOUND4args *args, int *flagp)
5880 {
5881 	int i;
5882 	int flag = RPC_ALL;
5883 
5884 	for (i = 0; flag && i < args->array_len; i++) {
5885 		uint_t op;
5886 
5887 		op = (uint_t)args->array[i].argop;
5888 
5889 		if (op < rfsv4disp_cnt)
5890 			flag &= rfsv4disptab[op].dis_flags;
5891 		else
5892 			flag = 0;
5893 	}
5894 	*flagp = flag;
5895 }
5896 
5897 nfsstat4
5898 rfs4_client_sysid(rfs4_client_t *cp, sysid_t *sp)
5899 {
5900 	nfsstat4 e;
5901 
5902 	rfs4_dbe_lock(cp->rc_dbe);
5903 
5904 	if (cp->rc_sysidt != LM_NOSYSID) {
5905 		*sp = cp->rc_sysidt;
5906 		e = NFS4_OK;
5907 
5908 	} else if ((cp->rc_sysidt = lm_alloc_sysidt()) != LM_NOSYSID) {
5909 		*sp = cp->rc_sysidt;
5910 		e = NFS4_OK;
5911 
5912 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
5913 		    "rfs4_client_sysid: allocated 0x%x\n", *sp));
5914 	} else
5915 		e = NFS4ERR_DELAY;
5916 
5917 	rfs4_dbe_unlock(cp->rc_dbe);
5918 	return (e);
5919 }
5920 
5921 #if defined(DEBUG) && ! defined(lint)
5922 static void lock_print(char *str, int operation, struct flock64 *flk)
5923 {
5924 	char *op, *type;
5925 
5926 	switch (operation) {
5927 	case F_GETLK: op = "F_GETLK";
5928 		break;
5929 	case F_SETLK: op = "F_SETLK";
5930 		break;
5931 	case F_SETLK_NBMAND: op = "F_SETLK_NBMAND";
5932 		break;
5933 	default: op = "F_UNKNOWN";
5934 		break;
5935 	}
5936 	switch (flk->l_type) {
5937 	case F_UNLCK: type = "F_UNLCK";
5938 		break;
5939 	case F_RDLCK: type = "F_RDLCK";
5940 		break;
5941 	case F_WRLCK: type = "F_WRLCK";
5942 		break;
5943 	default: type = "F_UNKNOWN";
5944 		break;
5945 	}
5946 
5947 	ASSERT(flk->l_whence == 0);
5948 	cmn_err(CE_NOTE, "%s:  %s, type = %s, off = %llx len = %llx pid = %d",
5949 	    str, op, type, (longlong_t)flk->l_start,
5950 	    flk->l_len ? (longlong_t)flk->l_len : ~0LL, flk->l_pid);
5951 }
5952 
5953 #define	LOCK_PRINT(d, s, t, f) if (d) lock_print(s, t, f)
5954 #else
5955 #define	LOCK_PRINT(d, s, t, f)
5956 #endif
5957 
5958 /*ARGSUSED*/
5959 static bool_t
5960 creds_ok(cred_set_t cr_set, struct svc_req *req, struct compound_state *cs)
5961 {
5962 	return (TRUE);
5963 }
5964 
5965 /*
5966  * Look up the pathname using the vp in cs as the directory vnode.
5967  * cs->vp will be the vnode for the file on success
5968  */
5969 
5970 static nfsstat4
5971 rfs4_lookup(component4 *component, struct svc_req *req,
5972     struct compound_state *cs)
5973 {
5974 	char *nm;
5975 	uint32_t len;
5976 	nfsstat4 status;
5977 	struct sockaddr *ca;
5978 	char *name;
5979 
5980 	if (cs->vp == NULL) {
5981 		return (NFS4ERR_NOFILEHANDLE);
5982 	}
5983 	if (cs->vp->v_type != VDIR) {
5984 		return (NFS4ERR_NOTDIR);
5985 	}
5986 
5987 	status = utf8_dir_verify(component);
5988 	if (status != NFS4_OK)
5989 		return (status);
5990 
5991 	nm = utf8_to_fn(component, &len, NULL);
5992 	if (nm == NULL) {
5993 		return (NFS4ERR_INVAL);
5994 	}
5995 
5996 	if (len > MAXNAMELEN) {
5997 		kmem_free(nm, len);
5998 		return (NFS4ERR_NAMETOOLONG);
5999 	}
6000 
6001 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6002 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6003 	    MAXPATHLEN + 1);
6004 
6005 	if (name == NULL) {
6006 		kmem_free(nm, len);
6007 		return (NFS4ERR_INVAL);
6008 	}
6009 
6010 	status = do_rfs4_op_lookup(name, req, cs);
6011 
6012 	if (name != nm)
6013 		kmem_free(name, MAXPATHLEN + 1);
6014 
6015 	kmem_free(nm, len);
6016 
6017 	return (status);
6018 }
6019 
6020 static nfsstat4
6021 rfs4_lookupfile(component4 *component, struct svc_req *req,
6022     struct compound_state *cs, uint32_t access, change_info4 *cinfo)
6023 {
6024 	nfsstat4 status;
6025 	vnode_t *dvp = cs->vp;
6026 	vattr_t bva, ava, fva;
6027 	int error;
6028 
6029 	/* Get "before" change value */
6030 	bva.va_mask = AT_CTIME|AT_SEQ;
6031 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6032 	if (error)
6033 		return (puterrno4(error));
6034 
6035 	/* rfs4_lookup may VN_RELE directory */
6036 	VN_HOLD(dvp);
6037 
6038 	status = rfs4_lookup(component, req, cs);
6039 	if (status != NFS4_OK) {
6040 		VN_RELE(dvp);
6041 		return (status);
6042 	}
6043 
6044 	/*
6045 	 * Get "after" change value, if it fails, simply return the
6046 	 * before value.
6047 	 */
6048 	ava.va_mask = AT_CTIME|AT_SEQ;
6049 	if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6050 		ava.va_ctime = bva.va_ctime;
6051 		ava.va_seq = 0;
6052 	}
6053 	VN_RELE(dvp);
6054 
6055 	/*
6056 	 * Validate the file is a file
6057 	 */
6058 	fva.va_mask = AT_TYPE|AT_MODE;
6059 	error = VOP_GETATTR(cs->vp, &fva, 0, cs->cr, NULL);
6060 	if (error)
6061 		return (puterrno4(error));
6062 
6063 	if (fva.va_type != VREG) {
6064 		if (fva.va_type == VDIR)
6065 			return (NFS4ERR_ISDIR);
6066 		if (fva.va_type == VLNK)
6067 			return (NFS4ERR_SYMLINK);
6068 		return (NFS4ERR_INVAL);
6069 	}
6070 
6071 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime);
6072 	NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6073 
6074 	/*
6075 	 * It is undefined if VOP_LOOKUP will change va_seq, so
6076 	 * cinfo.atomic = TRUE only if we have
6077 	 * non-zero va_seq's, and they have not changed.
6078 	 */
6079 	if (bva.va_seq && ava.va_seq && ava.va_seq == bva.va_seq)
6080 		cinfo->atomic = TRUE;
6081 	else
6082 		cinfo->atomic = FALSE;
6083 
6084 	/* Check for mandatory locking */
6085 	cs->mandlock = MANDLOCK(cs->vp, fva.va_mode);
6086 	return (check_open_access(access, cs, req));
6087 }
6088 
6089 static nfsstat4
6090 create_vnode(vnode_t *dvp, char *nm,  vattr_t *vap, createmode4 mode,
6091     cred_t *cr, vnode_t **vpp, bool_t *created)
6092 {
6093 	int error;
6094 	nfsstat4 status = NFS4_OK;
6095 	vattr_t va;
6096 
6097 tryagain:
6098 
6099 	/*
6100 	 * The file open mode used is VWRITE.  If the client needs
6101 	 * some other semantic, then it should do the access checking
6102 	 * itself.  It would have been nice to have the file open mode
6103 	 * passed as part of the arguments.
6104 	 */
6105 
6106 	*created = TRUE;
6107 	error = VOP_CREATE(dvp, nm, vap, EXCL, VWRITE, vpp, cr, 0, NULL, NULL);
6108 
6109 	if (error) {
6110 		*created = FALSE;
6111 
6112 		/*
6113 		 * If we got something other than file already exists
6114 		 * then just return this error.  Otherwise, we got
6115 		 * EEXIST.  If we were doing a GUARDED create, then
6116 		 * just return this error.  Otherwise, we need to
6117 		 * make sure that this wasn't a duplicate of an
6118 		 * exclusive create request.
6119 		 *
6120 		 * The assumption is made that a non-exclusive create
6121 		 * request will never return EEXIST.
6122 		 */
6123 
6124 		if (error != EEXIST || mode == GUARDED4) {
6125 			status = puterrno4(error);
6126 			return (status);
6127 		}
6128 		error = VOP_LOOKUP(dvp, nm, vpp, NULL, 0, NULL, cr,
6129 		    NULL, NULL, NULL);
6130 
6131 		if (error) {
6132 			/*
6133 			 * We couldn't find the file that we thought that
6134 			 * we just created.  So, we'll just try creating
6135 			 * it again.
6136 			 */
6137 			if (error == ENOENT)
6138 				goto tryagain;
6139 
6140 			status = puterrno4(error);
6141 			return (status);
6142 		}
6143 
6144 		if (mode == UNCHECKED4) {
6145 			/* existing object must be regular file */
6146 			if ((*vpp)->v_type != VREG) {
6147 				if ((*vpp)->v_type == VDIR)
6148 					status = NFS4ERR_ISDIR;
6149 				else if ((*vpp)->v_type == VLNK)
6150 					status = NFS4ERR_SYMLINK;
6151 				else
6152 					status = NFS4ERR_INVAL;
6153 				VN_RELE(*vpp);
6154 				return (status);
6155 			}
6156 
6157 			return (NFS4_OK);
6158 		}
6159 
6160 		/* Check for duplicate request */
6161 		va.va_mask = AT_MTIME;
6162 		error = VOP_GETATTR(*vpp, &va, 0, cr, NULL);
6163 		if (!error) {
6164 			/* We found the file */
6165 			const timestruc_t *mtime = &vap->va_mtime;
6166 
6167 			if (va.va_mtime.tv_sec != mtime->tv_sec ||
6168 			    va.va_mtime.tv_nsec != mtime->tv_nsec) {
6169 				/* but its not our creation */
6170 				VN_RELE(*vpp);
6171 				return (NFS4ERR_EXIST);
6172 			}
6173 			*created = TRUE; /* retrans of create == created */
6174 			return (NFS4_OK);
6175 		}
6176 		VN_RELE(*vpp);
6177 		return (NFS4ERR_EXIST);
6178 	}
6179 
6180 	return (NFS4_OK);
6181 }
6182 
6183 static nfsstat4
6184 check_open_access(uint32_t access, struct compound_state *cs,
6185     struct svc_req *req)
6186 {
6187 	int error;
6188 	vnode_t *vp;
6189 	bool_t readonly;
6190 	cred_t *cr = cs->cr;
6191 
6192 	/* For now we don't allow mandatory locking as per V2/V3 */
6193 	if (cs->access == CS_ACCESS_DENIED || cs->mandlock) {
6194 		return (NFS4ERR_ACCESS);
6195 	}
6196 
6197 	vp = cs->vp;
6198 	ASSERT(cr != NULL && vp->v_type == VREG);
6199 
6200 	/*
6201 	 * If the file system is exported read only and we are trying
6202 	 * to open for write, then return NFS4ERR_ROFS
6203 	 */
6204 
6205 	readonly = rdonly4(req, cs);
6206 
6207 	if ((access & OPEN4_SHARE_ACCESS_WRITE) && readonly)
6208 		return (NFS4ERR_ROFS);
6209 
6210 	if (access & OPEN4_SHARE_ACCESS_READ) {
6211 		if ((VOP_ACCESS(vp, VREAD, 0, cr, NULL) != 0) &&
6212 		    (VOP_ACCESS(vp, VEXEC, 0, cr, NULL) != 0)) {
6213 			return (NFS4ERR_ACCESS);
6214 		}
6215 	}
6216 
6217 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
6218 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
6219 		if (error)
6220 			return (NFS4ERR_ACCESS);
6221 	}
6222 
6223 	return (NFS4_OK);
6224 }
6225 
6226 static nfsstat4
6227 rfs4_createfile(OPEN4args *args, struct svc_req *req, struct compound_state *cs,
6228     change_info4 *cinfo, bitmap4 *attrset, clientid4 clientid)
6229 {
6230 	struct nfs4_svgetit_arg sarg;
6231 	struct nfs4_ntov_table ntov;
6232 
6233 	bool_t ntov_table_init = FALSE;
6234 	struct statvfs64 sb;
6235 	nfsstat4 status;
6236 	vnode_t *vp;
6237 	vattr_t bva, ava, iva, cva, *vap;
6238 	vnode_t *dvp;
6239 	timespec32_t *mtime;
6240 	char *nm = NULL;
6241 	uint_t buflen;
6242 	bool_t created;
6243 	bool_t setsize = FALSE;
6244 	len_t reqsize;
6245 	int error;
6246 	bool_t trunc;
6247 	caller_context_t ct;
6248 	component4 *component;
6249 	bslabel_t *clabel;
6250 	struct sockaddr *ca;
6251 	char *name = NULL;
6252 
6253 	sarg.sbp = &sb;
6254 	sarg.is_referral = B_FALSE;
6255 
6256 	dvp = cs->vp;
6257 
6258 	/* Check if the file system is read only */
6259 	if (rdonly4(req, cs))
6260 		return (NFS4ERR_ROFS);
6261 
6262 	/* check the label of including directory */
6263 	if (is_system_labeled()) {
6264 		ASSERT(req->rq_label != NULL);
6265 		clabel = req->rq_label;
6266 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
6267 		    "got client label from request(1)",
6268 		    struct svc_req *, req);
6269 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
6270 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
6271 			    cs->exi)) {
6272 				return (NFS4ERR_ACCESS);
6273 			}
6274 		}
6275 	}
6276 
6277 	/*
6278 	 * Get the last component of path name in nm. cs will reference
6279 	 * the including directory on success.
6280 	 */
6281 	component = &args->open_claim4_u.file;
6282 	status = utf8_dir_verify(component);
6283 	if (status != NFS4_OK)
6284 		return (status);
6285 
6286 	nm = utf8_to_fn(component, &buflen, NULL);
6287 
6288 	if (nm == NULL)
6289 		return (NFS4ERR_RESOURCE);
6290 
6291 	if (buflen > MAXNAMELEN) {
6292 		kmem_free(nm, buflen);
6293 		return (NFS4ERR_NAMETOOLONG);
6294 	}
6295 
6296 	bva.va_mask = AT_TYPE|AT_CTIME|AT_SEQ;
6297 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6298 	if (error) {
6299 		kmem_free(nm, buflen);
6300 		return (puterrno4(error));
6301 	}
6302 
6303 	if (bva.va_type != VDIR) {
6304 		kmem_free(nm, buflen);
6305 		return (NFS4ERR_NOTDIR);
6306 	}
6307 
6308 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime)
6309 
6310 	switch (args->mode) {
6311 	case GUARDED4:
6312 		/*FALLTHROUGH*/
6313 	case UNCHECKED4:
6314 		nfs4_ntov_table_init(&ntov);
6315 		ntov_table_init = TRUE;
6316 
6317 		*attrset = 0;
6318 		status = do_rfs4_set_attrs(attrset,
6319 		    &args->createhow4_u.createattrs,
6320 		    cs, &sarg, &ntov, NFS4ATTR_SETIT);
6321 
6322 		if (status == NFS4_OK && (sarg.vap->va_mask & AT_TYPE) &&
6323 		    sarg.vap->va_type != VREG) {
6324 			if (sarg.vap->va_type == VDIR)
6325 				status = NFS4ERR_ISDIR;
6326 			else if (sarg.vap->va_type == VLNK)
6327 				status = NFS4ERR_SYMLINK;
6328 			else
6329 				status = NFS4ERR_INVAL;
6330 		}
6331 
6332 		if (status != NFS4_OK) {
6333 			kmem_free(nm, buflen);
6334 			nfs4_ntov_table_free(&ntov, &sarg);
6335 			*attrset = 0;
6336 			return (status);
6337 		}
6338 
6339 		vap = sarg.vap;
6340 		vap->va_type = VREG;
6341 		vap->va_mask |= AT_TYPE;
6342 
6343 		if ((vap->va_mask & AT_MODE) == 0) {
6344 			vap->va_mask |= AT_MODE;
6345 			vap->va_mode = (mode_t)0600;
6346 		}
6347 
6348 		if (vap->va_mask & AT_SIZE) {
6349 
6350 			/* Disallow create with a non-zero size */
6351 
6352 			if ((reqsize = sarg.vap->va_size) != 0) {
6353 				kmem_free(nm, buflen);
6354 				nfs4_ntov_table_free(&ntov, &sarg);
6355 				*attrset = 0;
6356 				return (NFS4ERR_INVAL);
6357 			}
6358 			setsize = TRUE;
6359 		}
6360 		break;
6361 
6362 	case EXCLUSIVE4:
6363 		/* prohibit EXCL create of named attributes */
6364 		if (dvp->v_flag & V_XATTRDIR) {
6365 			kmem_free(nm, buflen);
6366 			*attrset = 0;
6367 			return (NFS4ERR_INVAL);
6368 		}
6369 
6370 		cva.va_mask = AT_TYPE | AT_MTIME | AT_MODE;
6371 		cva.va_type = VREG;
6372 		/*
6373 		 * Ensure no time overflows. Assumes underlying
6374 		 * filesystem supports at least 32 bits.
6375 		 * Truncate nsec to usec resolution to allow valid
6376 		 * compares even if the underlying filesystem truncates.
6377 		 */
6378 		mtime = (timespec32_t *)&args->createhow4_u.createverf;
6379 		cva.va_mtime.tv_sec = mtime->tv_sec % TIME32_MAX;
6380 		cva.va_mtime.tv_nsec = (mtime->tv_nsec / 1000) * 1000;
6381 		cva.va_mode = (mode_t)0;
6382 		vap = &cva;
6383 
6384 		/*
6385 		 * For EXCL create, attrset is set to the server attr
6386 		 * used to cache the client's verifier.
6387 		 */
6388 		*attrset = FATTR4_TIME_MODIFY_MASK;
6389 		break;
6390 	}
6391 
6392 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6393 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6394 	    MAXPATHLEN  + 1);
6395 
6396 	if (name == NULL) {
6397 		kmem_free(nm, buflen);
6398 		return (NFS4ERR_SERVERFAULT);
6399 	}
6400 
6401 	status = create_vnode(dvp, name, vap, args->mode,
6402 	    cs->cr, &vp, &created);
6403 	if (nm != name)
6404 		kmem_free(name, MAXPATHLEN + 1);
6405 	kmem_free(nm, buflen);
6406 
6407 	if (status != NFS4_OK) {
6408 		if (ntov_table_init)
6409 			nfs4_ntov_table_free(&ntov, &sarg);
6410 		*attrset = 0;
6411 		return (status);
6412 	}
6413 
6414 	trunc = (setsize && !created);
6415 
6416 	if (args->mode != EXCLUSIVE4) {
6417 		bitmap4 createmask = args->createhow4_u.createattrs.attrmask;
6418 
6419 		/*
6420 		 * True verification that object was created with correct
6421 		 * attrs is impossible.  The attrs could have been changed
6422 		 * immediately after object creation.  If attributes did
6423 		 * not verify, the only recourse for the server is to
6424 		 * destroy the object.  Maybe if some attrs (like gid)
6425 		 * are set incorrectly, the object should be destroyed;
6426 		 * however, seems bad as a default policy.  Do we really
6427 		 * want to destroy an object over one of the times not
6428 		 * verifying correctly?  For these reasons, the server
6429 		 * currently sets bits in attrset for createattrs
6430 		 * that were set; however, no verification is done.
6431 		 *
6432 		 * vmask_to_nmask accounts for vattr bits set on create
6433 		 *	[do_rfs4_set_attrs() only sets resp bits for
6434 		 *	 non-vattr/vfs bits.]
6435 		 * Mask off any bits we set by default so as not to return
6436 		 * more attrset bits than were requested in createattrs
6437 		 */
6438 		if (created) {
6439 			nfs4_vmask_to_nmask(sarg.vap->va_mask, attrset);
6440 			*attrset &= createmask;
6441 		} else {
6442 			/*
6443 			 * We did not create the vnode (we tried but it
6444 			 * already existed).  In this case, the only createattr
6445 			 * that the spec allows the server to set is size,
6446 			 * and even then, it can only be set if it is 0.
6447 			 */
6448 			*attrset = 0;
6449 			if (trunc)
6450 				*attrset = FATTR4_SIZE_MASK;
6451 		}
6452 	}
6453 	if (ntov_table_init)
6454 		nfs4_ntov_table_free(&ntov, &sarg);
6455 
6456 	/*
6457 	 * Get the initial "after" sequence number, if it fails,
6458 	 * set to zero, time to before.
6459 	 */
6460 	iva.va_mask = AT_CTIME|AT_SEQ;
6461 	if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL)) {
6462 		iva.va_seq = 0;
6463 		iva.va_ctime = bva.va_ctime;
6464 	}
6465 
6466 	/*
6467 	 * create_vnode attempts to create the file exclusive,
6468 	 * if it already exists the VOP_CREATE will fail and
6469 	 * may not increase va_seq. It is atomic if
6470 	 * we haven't changed the directory, but if it has changed
6471 	 * we don't know what changed it.
6472 	 */
6473 	if (!created) {
6474 		if (bva.va_seq && iva.va_seq &&
6475 		    bva.va_seq == iva.va_seq)
6476 			cinfo->atomic = TRUE;
6477 		else
6478 			cinfo->atomic = FALSE;
6479 		NFS4_SET_FATTR4_CHANGE(cinfo->after, iva.va_ctime);
6480 	} else {
6481 		/*
6482 		 * The entry was created, we need to sync the
6483 		 * directory metadata.
6484 		 */
6485 		(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
6486 
6487 		/*
6488 		 * Get "after" change value, if it fails, simply return the
6489 		 * before value.
6490 		 */
6491 		ava.va_mask = AT_CTIME|AT_SEQ;
6492 		if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6493 			ava.va_ctime = bva.va_ctime;
6494 			ava.va_seq = 0;
6495 		}
6496 
6497 		NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6498 
6499 		/*
6500 		 * The cinfo->atomic = TRUE only if we have
6501 		 * non-zero va_seq's, and it has incremented by exactly one
6502 		 * during the create_vnode and it didn't
6503 		 * change during the VOP_FSYNC.
6504 		 */
6505 		if (bva.va_seq && iva.va_seq && ava.va_seq &&
6506 		    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
6507 			cinfo->atomic = TRUE;
6508 		else
6509 			cinfo->atomic = FALSE;
6510 	}
6511 
6512 	/* Check for mandatory locking and that the size gets set. */
6513 	cva.va_mask = AT_MODE;
6514 	if (setsize)
6515 		cva.va_mask |= AT_SIZE;
6516 
6517 	/* Assume the worst */
6518 	cs->mandlock = TRUE;
6519 
6520 	if (VOP_GETATTR(vp, &cva, 0, cs->cr, NULL) == 0) {
6521 		cs->mandlock = MANDLOCK(cs->vp, cva.va_mode);
6522 
6523 		/*
6524 		 * Truncate the file if necessary; this would be
6525 		 * the case for create over an existing file.
6526 		 */
6527 
6528 		if (trunc) {
6529 			int in_crit = 0;
6530 			rfs4_file_t *fp;
6531 			bool_t create = FALSE;
6532 
6533 			/*
6534 			 * We are writing over an existing file.
6535 			 * Check to see if we need to recall a delegation.
6536 			 */
6537 			rfs4_hold_deleg_policy();
6538 			if ((fp = rfs4_findfile(vp, NULL, &create)) != NULL) {
6539 				if (rfs4_check_delegated_byfp(FWRITE, fp,
6540 				    (reqsize == 0), FALSE, FALSE, &clientid)) {
6541 					rfs4_file_rele(fp);
6542 					rfs4_rele_deleg_policy();
6543 					VN_RELE(vp);
6544 					*attrset = 0;
6545 					return (NFS4ERR_DELAY);
6546 				}
6547 				rfs4_file_rele(fp);
6548 			}
6549 			rfs4_rele_deleg_policy();
6550 
6551 			if (nbl_need_check(vp)) {
6552 				in_crit = 1;
6553 
6554 				ASSERT(reqsize == 0);
6555 
6556 				nbl_start_crit(vp, RW_READER);
6557 				if (nbl_conflict(vp, NBL_WRITE, 0,
6558 				    cva.va_size, 0, NULL)) {
6559 					in_crit = 0;
6560 					nbl_end_crit(vp);
6561 					VN_RELE(vp);
6562 					*attrset = 0;
6563 					return (NFS4ERR_ACCESS);
6564 				}
6565 			}
6566 			ct.cc_sysid = 0;
6567 			ct.cc_pid = 0;
6568 			ct.cc_caller_id = nfs4_srv_caller_id;
6569 			ct.cc_flags = CC_DONTBLOCK;
6570 
6571 			cva.va_mask = AT_SIZE;
6572 			cva.va_size = reqsize;
6573 			(void) VOP_SETATTR(vp, &cva, 0, cs->cr, &ct);
6574 			if (in_crit)
6575 				nbl_end_crit(vp);
6576 		}
6577 	}
6578 
6579 	error = makefh4(&cs->fh, vp, cs->exi);
6580 
6581 	/*
6582 	 * Force modified data and metadata out to stable storage.
6583 	 */
6584 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
6585 
6586 	if (error) {
6587 		VN_RELE(vp);
6588 		*attrset = 0;
6589 		return (puterrno4(error));
6590 	}
6591 
6592 	/* if parent dir is attrdir, set namedattr fh flag */
6593 	if (dvp->v_flag & V_XATTRDIR)
6594 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
6595 
6596 	if (cs->vp)
6597 		VN_RELE(cs->vp);
6598 
6599 	cs->vp = vp;
6600 
6601 	/*
6602 	 * if we did not create the file, we will need to check
6603 	 * the access bits on the file
6604 	 */
6605 
6606 	if (!created) {
6607 		if (setsize)
6608 			args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
6609 		status = check_open_access(args->share_access, cs, req);
6610 		if (status != NFS4_OK)
6611 			*attrset = 0;
6612 	}
6613 	return (status);
6614 }
6615 
6616 /*ARGSUSED*/
6617 static void
6618 rfs4_do_open(struct compound_state *cs, struct svc_req *req,
6619     rfs4_openowner_t *oo, delegreq_t deleg,
6620     uint32_t access, uint32_t deny,
6621     OPEN4res *resp, int deleg_cur)
6622 {
6623 	/* XXX Currently not using req  */
6624 	rfs4_state_t *sp;
6625 	rfs4_file_t *fp;
6626 	bool_t screate = TRUE;
6627 	bool_t fcreate = TRUE;
6628 	uint32_t open_a, share_a;
6629 	uint32_t open_d, share_d;
6630 	rfs4_deleg_state_t *dsp;
6631 	sysid_t sysid;
6632 	nfsstat4 status;
6633 	caller_context_t ct;
6634 	int fflags = 0;
6635 	int recall = 0;
6636 	int err;
6637 	int first_open;
6638 
6639 	/* get the file struct and hold a lock on it during initial open */
6640 	fp = rfs4_findfile_withlock(cs->vp, &cs->fh, &fcreate);
6641 	if (fp == NULL) {
6642 		resp->status = NFS4ERR_RESOURCE;
6643 		DTRACE_PROBE1(nfss__e__do__open1, nfsstat4, resp->status);
6644 		return;
6645 	}
6646 
6647 	sp = rfs4_findstate_by_owner_file(oo, fp, &screate);
6648 	if (sp == NULL) {
6649 		resp->status = NFS4ERR_RESOURCE;
6650 		DTRACE_PROBE1(nfss__e__do__open2, nfsstat4, resp->status);
6651 		/* No need to keep any reference */
6652 		rw_exit(&fp->rf_file_rwlock);
6653 		rfs4_file_rele(fp);
6654 		return;
6655 	}
6656 
6657 	/* try to get the sysid before continuing */
6658 	if ((status = rfs4_client_sysid(oo->ro_client, &sysid)) != NFS4_OK) {
6659 		resp->status = status;
6660 		rfs4_file_rele(fp);
6661 		/* Not a fully formed open; "close" it */
6662 		if (screate == TRUE)
6663 			rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6664 		rfs4_state_rele(sp);
6665 		return;
6666 	}
6667 
6668 	/* Calculate the fflags for this OPEN. */
6669 	if (access & OPEN4_SHARE_ACCESS_READ)
6670 		fflags |= FREAD;
6671 	if (access & OPEN4_SHARE_ACCESS_WRITE)
6672 		fflags |= FWRITE;
6673 
6674 	rfs4_dbe_lock(sp->rs_dbe);
6675 
6676 	/*
6677 	 * Calculate the new deny and access mode that this open is adding to
6678 	 * the file for this open owner;
6679 	 */
6680 	open_d = (deny & ~sp->rs_open_deny);
6681 	open_a = (access & ~sp->rs_open_access);
6682 
6683 	/*
6684 	 * Calculate the new share access and share deny modes that this open
6685 	 * is adding to the file for this open owner;
6686 	 */
6687 	share_a = (access & ~sp->rs_share_access);
6688 	share_d = (deny & ~sp->rs_share_deny);
6689 
6690 	first_open = (sp->rs_open_access & OPEN4_SHARE_ACCESS_BOTH) == 0;
6691 
6692 	/*
6693 	 * Check to see the client has already sent an open for this
6694 	 * open owner on this file with the same share/deny modes.
6695 	 * If so, we don't need to check for a conflict and we don't
6696 	 * need to add another shrlock.  If not, then we need to
6697 	 * check for conflicts in deny and access before checking for
6698 	 * conflicts in delegation.  We don't want to recall a
6699 	 * delegation based on an open that will eventually fail based
6700 	 * on shares modes.
6701 	 */
6702 
6703 	if (share_a || share_d) {
6704 		if ((err = rfs4_share(sp, access, deny)) != 0) {
6705 			rfs4_dbe_unlock(sp->rs_dbe);
6706 			resp->status = err;
6707 
6708 			rfs4_file_rele(fp);
6709 			/* Not a fully formed open; "close" it */
6710 			if (screate == TRUE)
6711 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6712 			rfs4_state_rele(sp);
6713 			return;
6714 		}
6715 	}
6716 
6717 	rfs4_dbe_lock(fp->rf_dbe);
6718 
6719 	/*
6720 	 * Check to see if this file is delegated and if so, if a
6721 	 * recall needs to be done.
6722 	 */
6723 	if (rfs4_check_recall(sp, access)) {
6724 		rfs4_dbe_unlock(fp->rf_dbe);
6725 		rfs4_dbe_unlock(sp->rs_dbe);
6726 		rfs4_recall_deleg(fp, FALSE, sp->rs_owner->ro_client);
6727 		delay(NFS4_DELEGATION_CONFLICT_DELAY);
6728 		rfs4_dbe_lock(sp->rs_dbe);
6729 
6730 		/* if state closed while lock was dropped */
6731 		if (sp->rs_closed) {
6732 			if (share_a || share_d)
6733 				(void) rfs4_unshare(sp);
6734 			rfs4_dbe_unlock(sp->rs_dbe);
6735 			rfs4_file_rele(fp);
6736 			/* Not a fully formed open; "close" it */
6737 			if (screate == TRUE)
6738 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6739 			rfs4_state_rele(sp);
6740 			resp->status = NFS4ERR_OLD_STATEID;
6741 			return;
6742 		}
6743 
6744 		rfs4_dbe_lock(fp->rf_dbe);
6745 		/* Let's see if the delegation was returned */
6746 		if (rfs4_check_recall(sp, access)) {
6747 			rfs4_dbe_unlock(fp->rf_dbe);
6748 			if (share_a || share_d)
6749 				(void) rfs4_unshare(sp);
6750 			rfs4_dbe_unlock(sp->rs_dbe);
6751 			rfs4_file_rele(fp);
6752 			rfs4_update_lease(sp->rs_owner->ro_client);
6753 
6754 			/* Not a fully formed open; "close" it */
6755 			if (screate == TRUE)
6756 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6757 			rfs4_state_rele(sp);
6758 			resp->status = NFS4ERR_DELAY;
6759 			return;
6760 		}
6761 	}
6762 	/*
6763 	 * the share check passed and any delegation conflict has been
6764 	 * taken care of, now call vop_open.
6765 	 * if this is the first open then call vop_open with fflags.
6766 	 * if not, call vn_open_upgrade with just the upgrade flags.
6767 	 *
6768 	 * if the file has been opened already, it will have the current
6769 	 * access mode in the state struct.  if it has no share access, then
6770 	 * this is a new open.
6771 	 *
6772 	 * However, if this is open with CLAIM_DLEGATE_CUR, then don't
6773 	 * call VOP_OPEN(), just do the open upgrade.
6774 	 */
6775 	if (first_open && !deleg_cur) {
6776 		ct.cc_sysid = sysid;
6777 		ct.cc_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
6778 		ct.cc_caller_id = nfs4_srv_caller_id;
6779 		ct.cc_flags = CC_DONTBLOCK;
6780 		err = VOP_OPEN(&cs->vp, fflags, cs->cr, &ct);
6781 		if (err) {
6782 			rfs4_dbe_unlock(fp->rf_dbe);
6783 			if (share_a || share_d)
6784 				(void) rfs4_unshare(sp);
6785 			rfs4_dbe_unlock(sp->rs_dbe);
6786 			rfs4_file_rele(fp);
6787 
6788 			/* Not a fully formed open; "close" it */
6789 			if (screate == TRUE)
6790 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6791 			rfs4_state_rele(sp);
6792 			/* check if a monitor detected a delegation conflict */
6793 			if (err == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
6794 				resp->status = NFS4ERR_DELAY;
6795 			else
6796 				resp->status = NFS4ERR_SERVERFAULT;
6797 			return;
6798 		}
6799 	} else { /* open upgrade */
6800 		/*
6801 		 * calculate the fflags for the new mode that is being added
6802 		 * by this upgrade.
6803 		 */
6804 		fflags = 0;
6805 		if (open_a & OPEN4_SHARE_ACCESS_READ)
6806 			fflags |= FREAD;
6807 		if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6808 			fflags |= FWRITE;
6809 		vn_open_upgrade(cs->vp, fflags);
6810 	}
6811 	sp->rs_open_access |= access;
6812 	sp->rs_open_deny |= deny;
6813 
6814 	if (open_d & OPEN4_SHARE_DENY_READ)
6815 		fp->rf_deny_read++;
6816 	if (open_d & OPEN4_SHARE_DENY_WRITE)
6817 		fp->rf_deny_write++;
6818 	fp->rf_share_deny |= deny;
6819 
6820 	if (open_a & OPEN4_SHARE_ACCESS_READ)
6821 		fp->rf_access_read++;
6822 	if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6823 		fp->rf_access_write++;
6824 	fp->rf_share_access |= access;
6825 
6826 	/*
6827 	 * Check for delegation here. if the deleg argument is not
6828 	 * DELEG_ANY, then this is a reclaim from a client and
6829 	 * we must honor the delegation requested. If necessary we can
6830 	 * set the recall flag.
6831 	 */
6832 
6833 	dsp = rfs4_grant_delegation(deleg, sp, &recall);
6834 
6835 	cs->deleg = (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE);
6836 
6837 	next_stateid(&sp->rs_stateid);
6838 
6839 	resp->stateid = sp->rs_stateid.stateid;
6840 
6841 	rfs4_dbe_unlock(fp->rf_dbe);
6842 	rfs4_dbe_unlock(sp->rs_dbe);
6843 
6844 	if (dsp) {
6845 		rfs4_set_deleg_response(dsp, &resp->delegation, NULL, recall);
6846 		rfs4_deleg_state_rele(dsp);
6847 	}
6848 
6849 	rfs4_file_rele(fp);
6850 	rfs4_state_rele(sp);
6851 
6852 	resp->status = NFS4_OK;
6853 }
6854 
6855 /*ARGSUSED*/
6856 static void
6857 rfs4_do_opennull(struct compound_state *cs, struct svc_req *req,
6858     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6859 {
6860 	change_info4 *cinfo = &resp->cinfo;
6861 	bitmap4 *attrset = &resp->attrset;
6862 
6863 	if (args->opentype == OPEN4_NOCREATE)
6864 		resp->status = rfs4_lookupfile(&args->open_claim4_u.file,
6865 		    req, cs, args->share_access, cinfo);
6866 	else {
6867 		/* inhibit delegation grants during exclusive create */
6868 
6869 		if (args->mode == EXCLUSIVE4)
6870 			rfs4_disable_delegation();
6871 
6872 		resp->status = rfs4_createfile(args, req, cs, cinfo, attrset,
6873 		    oo->ro_client->rc_clientid);
6874 	}
6875 
6876 	if (resp->status == NFS4_OK) {
6877 
6878 		/* cs->vp cs->fh now reference the desired file */
6879 
6880 		rfs4_do_open(cs, req, oo,
6881 		    oo->ro_need_confirm ? DELEG_NONE : DELEG_ANY,
6882 		    args->share_access, args->share_deny, resp, 0);
6883 
6884 		/*
6885 		 * If rfs4_createfile set attrset, we must
6886 		 * clear this attrset before the response is copied.
6887 		 */
6888 		if (resp->status != NFS4_OK && resp->attrset) {
6889 			resp->attrset = 0;
6890 		}
6891 	}
6892 	else
6893 		*cs->statusp = resp->status;
6894 
6895 	if (args->mode == EXCLUSIVE4)
6896 		rfs4_enable_delegation();
6897 }
6898 
6899 /*ARGSUSED*/
6900 static void
6901 rfs4_do_openprev(struct compound_state *cs, struct svc_req *req,
6902     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6903 {
6904 	change_info4 *cinfo = &resp->cinfo;
6905 	vattr_t va;
6906 	vtype_t v_type = cs->vp->v_type;
6907 	int error = 0;
6908 
6909 	/* Verify that we have a regular file */
6910 	if (v_type != VREG) {
6911 		if (v_type == VDIR)
6912 			resp->status = NFS4ERR_ISDIR;
6913 		else if (v_type == VLNK)
6914 			resp->status = NFS4ERR_SYMLINK;
6915 		else
6916 			resp->status = NFS4ERR_INVAL;
6917 		return;
6918 	}
6919 
6920 	va.va_mask = AT_MODE|AT_UID;
6921 	error = VOP_GETATTR(cs->vp, &va, 0, cs->cr, NULL);
6922 	if (error) {
6923 		resp->status = puterrno4(error);
6924 		return;
6925 	}
6926 
6927 	cs->mandlock = MANDLOCK(cs->vp, va.va_mode);
6928 
6929 	/*
6930 	 * Check if we have access to the file, Note the the file
6931 	 * could have originally been open UNCHECKED or GUARDED
6932 	 * with mode bits that will now fail, but there is nothing
6933 	 * we can really do about that except in the case that the
6934 	 * owner of the file is the one requesting the open.
6935 	 */
6936 	if (crgetuid(cs->cr) != va.va_uid) {
6937 		resp->status = check_open_access(args->share_access, cs, req);
6938 		if (resp->status != NFS4_OK) {
6939 			return;
6940 		}
6941 	}
6942 
6943 	/*
6944 	 * cinfo on a CLAIM_PREVIOUS is undefined, initialize to zero
6945 	 */
6946 	cinfo->before = 0;
6947 	cinfo->after = 0;
6948 	cinfo->atomic = FALSE;
6949 
6950 	rfs4_do_open(cs, req, oo,
6951 	    NFS4_DELEG4TYPE2REQTYPE(args->open_claim4_u.delegate_type),
6952 	    args->share_access, args->share_deny, resp, 0);
6953 }
6954 
6955 static void
6956 rfs4_do_opendelcur(struct compound_state *cs, struct svc_req *req,
6957     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6958 {
6959 	int error;
6960 	nfsstat4 status;
6961 	stateid4 stateid =
6962 	    args->open_claim4_u.delegate_cur_info.delegate_stateid;
6963 	rfs4_deleg_state_t *dsp;
6964 
6965 	/*
6966 	 * Find the state info from the stateid and confirm that the
6967 	 * file is delegated.  If the state openowner is the same as
6968 	 * the supplied openowner we're done. If not, get the file
6969 	 * info from the found state info. Use that file info to
6970 	 * create the state for this lock owner. Note solaris doen't
6971 	 * really need the pathname to find the file. We may want to
6972 	 * lookup the pathname and make sure that the vp exist and
6973 	 * matches the vp in the file structure. However it is
6974 	 * possible that the pathname nolonger exists (local process
6975 	 * unlinks the file), so this may not be that useful.
6976 	 */
6977 
6978 	status = rfs4_get_deleg_state(&stateid, &dsp);
6979 	if (status != NFS4_OK) {
6980 		resp->status = status;
6981 		return;
6982 	}
6983 
6984 	ASSERT(dsp->rds_finfo->rf_dinfo.rd_dtype != OPEN_DELEGATE_NONE);
6985 
6986 	/*
6987 	 * New lock owner, create state. Since this was probably called
6988 	 * in response to a CB_RECALL we set deleg to DELEG_NONE
6989 	 */
6990 
6991 	ASSERT(cs->vp != NULL);
6992 	VN_RELE(cs->vp);
6993 	VN_HOLD(dsp->rds_finfo->rf_vp);
6994 	cs->vp = dsp->rds_finfo->rf_vp;
6995 
6996 	if (error = makefh4(&cs->fh, cs->vp, cs->exi)) {
6997 		rfs4_deleg_state_rele(dsp);
6998 		*cs->statusp = resp->status = puterrno4(error);
6999 		return;
7000 	}
7001 
7002 	/* Mark progress for delegation returns */
7003 	dsp->rds_finfo->rf_dinfo.rd_time_lastwrite = gethrestime_sec();
7004 	rfs4_deleg_state_rele(dsp);
7005 	rfs4_do_open(cs, req, oo, DELEG_NONE,
7006 	    args->share_access, args->share_deny, resp, 1);
7007 }
7008 
7009 /*ARGSUSED*/
7010 static void
7011 rfs4_do_opendelprev(struct compound_state *cs, struct svc_req *req,
7012     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7013 {
7014 	/*
7015 	 * Lookup the pathname, it must already exist since this file
7016 	 * was delegated.
7017 	 *
7018 	 * Find the file and state info for this vp and open owner pair.
7019 	 *	check that they are in fact delegated.
7020 	 *	check that the state access and deny modes are the same.
7021 	 *
7022 	 * Return the delgation possibly seting the recall flag.
7023 	 */
7024 	rfs4_file_t *fp;
7025 	rfs4_state_t *sp;
7026 	bool_t create = FALSE;
7027 	bool_t dcreate = FALSE;
7028 	rfs4_deleg_state_t *dsp;
7029 	nfsace4 *ace;
7030 
7031 	/* Note we ignore oflags */
7032 	resp->status = rfs4_lookupfile(&args->open_claim4_u.file_delegate_prev,
7033 	    req, cs, args->share_access, &resp->cinfo);
7034 
7035 	if (resp->status != NFS4_OK) {
7036 		return;
7037 	}
7038 
7039 	/* get the file struct and hold a lock on it during initial open */
7040 	fp = rfs4_findfile_withlock(cs->vp, NULL, &create);
7041 	if (fp == NULL) {
7042 		resp->status = NFS4ERR_RESOURCE;
7043 		DTRACE_PROBE1(nfss__e__do_opendelprev1, nfsstat4, resp->status);
7044 		return;
7045 	}
7046 
7047 	sp = rfs4_findstate_by_owner_file(oo, fp, &create);
7048 	if (sp == NULL) {
7049 		resp->status = NFS4ERR_SERVERFAULT;
7050 		DTRACE_PROBE1(nfss__e__do_opendelprev2, nfsstat4, resp->status);
7051 		rw_exit(&fp->rf_file_rwlock);
7052 		rfs4_file_rele(fp);
7053 		return;
7054 	}
7055 
7056 	rfs4_dbe_lock(sp->rs_dbe);
7057 	rfs4_dbe_lock(fp->rf_dbe);
7058 	if (args->share_access != sp->rs_share_access ||
7059 	    args->share_deny != sp->rs_share_deny ||
7060 	    sp->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
7061 		NFS4_DEBUG(rfs4_debug,
7062 		    (CE_NOTE, "rfs4_do_opendelprev: state mixup"));
7063 		rfs4_dbe_unlock(fp->rf_dbe);
7064 		rfs4_dbe_unlock(sp->rs_dbe);
7065 		rfs4_file_rele(fp);
7066 		rfs4_state_rele(sp);
7067 		resp->status = NFS4ERR_SERVERFAULT;
7068 		return;
7069 	}
7070 	rfs4_dbe_unlock(fp->rf_dbe);
7071 	rfs4_dbe_unlock(sp->rs_dbe);
7072 
7073 	dsp = rfs4_finddeleg(sp, &dcreate);
7074 	if (dsp == NULL) {
7075 		rfs4_state_rele(sp);
7076 		rfs4_file_rele(fp);
7077 		resp->status = NFS4ERR_SERVERFAULT;
7078 		return;
7079 	}
7080 
7081 	next_stateid(&sp->rs_stateid);
7082 
7083 	resp->stateid = sp->rs_stateid.stateid;
7084 
7085 	resp->delegation.delegation_type = dsp->rds_dtype;
7086 
7087 	if (dsp->rds_dtype == OPEN_DELEGATE_READ) {
7088 		open_read_delegation4 *rv =
7089 		    &resp->delegation.open_delegation4_u.read;
7090 
7091 		rv->stateid = dsp->rds_delegid.stateid;
7092 		rv->recall = FALSE; /* no policy in place to set to TRUE */
7093 		ace = &rv->permissions;
7094 	} else {
7095 		open_write_delegation4 *rv =
7096 		    &resp->delegation.open_delegation4_u.write;
7097 
7098 		rv->stateid = dsp->rds_delegid.stateid;
7099 		rv->recall = FALSE;  /* no policy in place to set to TRUE */
7100 		ace = &rv->permissions;
7101 		rv->space_limit.limitby = NFS_LIMIT_SIZE;
7102 		rv->space_limit.nfs_space_limit4_u.filesize = UINT64_MAX;
7103 	}
7104 
7105 	/* XXX For now */
7106 	ace->type = ACE4_ACCESS_ALLOWED_ACE_TYPE;
7107 	ace->flag = 0;
7108 	ace->access_mask = 0;
7109 	ace->who.utf8string_len = 0;
7110 	ace->who.utf8string_val = 0;
7111 
7112 	rfs4_deleg_state_rele(dsp);
7113 	rfs4_state_rele(sp);
7114 	rfs4_file_rele(fp);
7115 }
7116 
7117 typedef enum {
7118 	NFS4_CHKSEQ_OKAY = 0,
7119 	NFS4_CHKSEQ_REPLAY = 1,
7120 	NFS4_CHKSEQ_BAD = 2
7121 } rfs4_chkseq_t;
7122 
7123 /*
7124  * Generic function for sequence number checks.
7125  */
7126 static rfs4_chkseq_t
7127 rfs4_check_seqid(seqid4 seqid, nfs_resop4 *lastop,
7128     seqid4 rqst_seq, nfs_resop4 *resop, bool_t copyres)
7129 {
7130 	/* Same sequence ids and matching operations? */
7131 	if (seqid == rqst_seq && resop->resop == lastop->resop) {
7132 		if (copyres == TRUE) {
7133 			rfs4_free_reply(resop);
7134 			rfs4_copy_reply(resop, lastop);
7135 		}
7136 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
7137 		    "Replayed SEQID %d\n", seqid));
7138 		return (NFS4_CHKSEQ_REPLAY);
7139 	}
7140 
7141 	/* If the incoming sequence is not the next expected then it is bad */
7142 	if (rqst_seq != seqid + 1) {
7143 		if (rqst_seq == seqid) {
7144 			NFS4_DEBUG(rfs4_debug,
7145 			    (CE_NOTE, "BAD SEQID: Replayed sequence id "
7146 			    "but last op was %d current op is %d\n",
7147 			    lastop->resop, resop->resop));
7148 			return (NFS4_CHKSEQ_BAD);
7149 		}
7150 		NFS4_DEBUG(rfs4_debug,
7151 		    (CE_NOTE, "BAD SEQID: got %u expecting %u\n",
7152 		    rqst_seq, seqid));
7153 		return (NFS4_CHKSEQ_BAD);
7154 	}
7155 
7156 	/* Everything okay -- next expected */
7157 	return (NFS4_CHKSEQ_OKAY);
7158 }
7159 
7160 
7161 static rfs4_chkseq_t
7162 rfs4_check_open_seqid(seqid4 seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7163 {
7164 	rfs4_chkseq_t rc;
7165 
7166 	rfs4_dbe_lock(op->ro_dbe);
7167 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply, seqid, resop,
7168 	    TRUE);
7169 	rfs4_dbe_unlock(op->ro_dbe);
7170 
7171 	if (rc == NFS4_CHKSEQ_OKAY)
7172 		rfs4_update_lease(op->ro_client);
7173 
7174 	return (rc);
7175 }
7176 
7177 static rfs4_chkseq_t
7178 rfs4_check_olo_seqid(seqid4 olo_seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7179 {
7180 	rfs4_chkseq_t rc;
7181 
7182 	rfs4_dbe_lock(op->ro_dbe);
7183 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply,
7184 	    olo_seqid, resop, FALSE);
7185 	rfs4_dbe_unlock(op->ro_dbe);
7186 
7187 	return (rc);
7188 }
7189 
7190 static rfs4_chkseq_t
7191 rfs4_check_lock_seqid(seqid4 seqid, rfs4_lo_state_t *lsp, nfs_resop4 *resop)
7192 {
7193 	rfs4_chkseq_t rc = NFS4_CHKSEQ_OKAY;
7194 
7195 	rfs4_dbe_lock(lsp->rls_dbe);
7196 	if (!lsp->rls_skip_seqid_check)
7197 		rc = rfs4_check_seqid(lsp->rls_seqid, &lsp->rls_reply, seqid,
7198 		    resop, TRUE);
7199 	rfs4_dbe_unlock(lsp->rls_dbe);
7200 
7201 	return (rc);
7202 }
7203 
7204 static void
7205 rfs4_op_open(nfs_argop4 *argop, nfs_resop4 *resop,
7206     struct svc_req *req, struct compound_state *cs)
7207 {
7208 	OPEN4args *args = &argop->nfs_argop4_u.opopen;
7209 	OPEN4res *resp = &resop->nfs_resop4_u.opopen;
7210 	open_owner4 *owner = &args->owner;
7211 	open_claim_type4 claim = args->claim;
7212 	rfs4_client_t *cp;
7213 	rfs4_openowner_t *oo;
7214 	bool_t create;
7215 	bool_t replay = FALSE;
7216 	int can_reclaim;
7217 
7218 	DTRACE_NFSV4_2(op__open__start, struct compound_state *, cs,
7219 	    OPEN4args *, args);
7220 
7221 	if (cs->vp == NULL) {
7222 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7223 		goto end;
7224 	}
7225 
7226 	/*
7227 	 * Need to check clientid and lease expiration first based on
7228 	 * error ordering and incrementing sequence id.
7229 	 */
7230 	cp = rfs4_findclient_by_id(owner->clientid, FALSE);
7231 	if (cp == NULL) {
7232 		*cs->statusp = resp->status =
7233 		    rfs4_check_clientid(&owner->clientid, 0);
7234 		goto end;
7235 	}
7236 
7237 	if (rfs4_lease_expired(cp)) {
7238 		rfs4_client_close(cp);
7239 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7240 		goto end;
7241 	}
7242 	can_reclaim = cp->rc_can_reclaim;
7243 
7244 	/*
7245 	 * Find the open_owner for use from this point forward.  Take
7246 	 * care in updating the sequence id based on the type of error
7247 	 * being returned.
7248 	 */
7249 retry:
7250 	create = TRUE;
7251 	oo = rfs4_findopenowner(owner, &create, args->seqid);
7252 	if (oo == NULL) {
7253 		*cs->statusp = resp->status = NFS4ERR_RESOURCE;
7254 		rfs4_client_rele(cp);
7255 		goto end;
7256 	}
7257 
7258 	/* Hold off access to the sequence space while the open is done */
7259 	rfs4_sw_enter(&oo->ro_sw);
7260 
7261 	/*
7262 	 * If the open_owner existed before at the server, then check
7263 	 * the sequence id.
7264 	 */
7265 	if (!create && !oo->ro_postpone_confirm) {
7266 		switch (rfs4_check_open_seqid(args->seqid, oo, resop)) {
7267 		case NFS4_CHKSEQ_BAD:
7268 			if ((args->seqid > oo->ro_open_seqid) &&
7269 			    oo->ro_need_confirm) {
7270 				rfs4_free_opens(oo, TRUE, FALSE);
7271 				rfs4_sw_exit(&oo->ro_sw);
7272 				rfs4_openowner_rele(oo);
7273 				goto retry;
7274 			}
7275 			resp->status = NFS4ERR_BAD_SEQID;
7276 			goto out;
7277 		case NFS4_CHKSEQ_REPLAY: /* replay of previous request */
7278 			replay = TRUE;
7279 			goto out;
7280 		default:
7281 			break;
7282 		}
7283 
7284 		/*
7285 		 * Sequence was ok and open owner exists
7286 		 * check to see if we have yet to see an
7287 		 * open_confirm.
7288 		 */
7289 		if (oo->ro_need_confirm) {
7290 			rfs4_free_opens(oo, TRUE, FALSE);
7291 			rfs4_sw_exit(&oo->ro_sw);
7292 			rfs4_openowner_rele(oo);
7293 			goto retry;
7294 		}
7295 	}
7296 	/* Grace only applies to regular-type OPENs */
7297 	if (rfs4_clnt_in_grace(cp) &&
7298 	    (claim == CLAIM_NULL || claim == CLAIM_DELEGATE_CUR)) {
7299 		*cs->statusp = resp->status = NFS4ERR_GRACE;
7300 		goto out;
7301 	}
7302 
7303 	/*
7304 	 * If previous state at the server existed then can_reclaim
7305 	 * will be set. If not reply NFS4ERR_NO_GRACE to the
7306 	 * client.
7307 	 */
7308 	if (rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS && !can_reclaim) {
7309 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7310 		goto out;
7311 	}
7312 
7313 
7314 	/*
7315 	 * Reject the open if the client has missed the grace period
7316 	 */
7317 	if (!rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS) {
7318 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7319 		goto out;
7320 	}
7321 
7322 	/* Couple of up-front bookkeeping items */
7323 	if (oo->ro_need_confirm) {
7324 		/*
7325 		 * If this is a reclaim OPEN then we should not ask
7326 		 * for a confirmation of the open_owner per the
7327 		 * protocol specification.
7328 		 */
7329 		if (claim == CLAIM_PREVIOUS)
7330 			oo->ro_need_confirm = FALSE;
7331 		else
7332 			resp->rflags |= OPEN4_RESULT_CONFIRM;
7333 	}
7334 	resp->rflags |= OPEN4_RESULT_LOCKTYPE_POSIX;
7335 
7336 	/*
7337 	 * If there is an unshared filesystem mounted on this vnode,
7338 	 * do not allow to open/create in this directory.
7339 	 */
7340 	if (vn_ismntpt(cs->vp)) {
7341 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
7342 		goto out;
7343 	}
7344 
7345 	/*
7346 	 * access must READ, WRITE, or BOTH.  No access is invalid.
7347 	 * deny can be READ, WRITE, BOTH, or NONE.
7348 	 * bits not defined for access/deny are invalid.
7349 	 */
7350 	if (! (args->share_access & OPEN4_SHARE_ACCESS_BOTH) ||
7351 	    (args->share_access & ~OPEN4_SHARE_ACCESS_BOTH) ||
7352 	    (args->share_deny & ~OPEN4_SHARE_DENY_BOTH)) {
7353 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7354 		goto out;
7355 	}
7356 
7357 
7358 	/*
7359 	 * make sure attrset is zero before response is built.
7360 	 */
7361 	resp->attrset = 0;
7362 
7363 	switch (claim) {
7364 	case CLAIM_NULL:
7365 		rfs4_do_opennull(cs, req, args, oo, resp);
7366 		break;
7367 	case CLAIM_PREVIOUS:
7368 		rfs4_do_openprev(cs, req, args, oo, resp);
7369 		break;
7370 	case CLAIM_DELEGATE_CUR:
7371 		rfs4_do_opendelcur(cs, req, args, oo, resp);
7372 		break;
7373 	case CLAIM_DELEGATE_PREV:
7374 		rfs4_do_opendelprev(cs, req, args, oo, resp);
7375 		break;
7376 	default:
7377 		resp->status = NFS4ERR_INVAL;
7378 		break;
7379 	}
7380 
7381 out:
7382 	rfs4_client_rele(cp);
7383 
7384 	/* Catch sequence id handling here to make it a little easier */
7385 	switch (resp->status) {
7386 	case NFS4ERR_BADXDR:
7387 	case NFS4ERR_BAD_SEQID:
7388 	case NFS4ERR_BAD_STATEID:
7389 	case NFS4ERR_NOFILEHANDLE:
7390 	case NFS4ERR_RESOURCE:
7391 	case NFS4ERR_STALE_CLIENTID:
7392 	case NFS4ERR_STALE_STATEID:
7393 		/*
7394 		 * The protocol states that if any of these errors are
7395 		 * being returned, the sequence id should not be
7396 		 * incremented.  Any other return requires an
7397 		 * increment.
7398 		 */
7399 		break;
7400 	default:
7401 		/* Always update the lease in this case */
7402 		rfs4_update_lease(oo->ro_client);
7403 
7404 		/* Regular response - copy the result */
7405 		if (!replay)
7406 			rfs4_update_open_resp(oo, resop, &cs->fh);
7407 
7408 		/*
7409 		 * REPLAY case: Only if the previous response was OK
7410 		 * do we copy the filehandle.  If not OK, no
7411 		 * filehandle to copy.
7412 		 */
7413 		if (replay == TRUE &&
7414 		    resp->status == NFS4_OK &&
7415 		    oo->ro_reply_fh.nfs_fh4_val) {
7416 			/*
7417 			 * If this is a replay, we must restore the
7418 			 * current filehandle/vp to that of what was
7419 			 * returned originally.  Try our best to do
7420 			 * it.
7421 			 */
7422 			nfs_fh4_fmt_t *fh_fmtp =
7423 			    (nfs_fh4_fmt_t *)oo->ro_reply_fh.nfs_fh4_val;
7424 
7425 			cs->exi = checkexport4(&fh_fmtp->fh4_fsid,
7426 			    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
7427 
7428 			if (cs->exi == NULL) {
7429 				resp->status = NFS4ERR_STALE;
7430 				goto finish;
7431 			}
7432 
7433 			VN_RELE(cs->vp);
7434 
7435 			cs->vp = nfs4_fhtovp(&oo->ro_reply_fh, cs->exi,
7436 			    &resp->status);
7437 
7438 			if (cs->vp == NULL)
7439 				goto finish;
7440 
7441 			nfs_fh4_copy(&oo->ro_reply_fh, &cs->fh);
7442 		}
7443 
7444 		/*
7445 		 * If this was a replay, no need to update the
7446 		 * sequence id. If the open_owner was not created on
7447 		 * this pass, then update.  The first use of an
7448 		 * open_owner will not bump the sequence id.
7449 		 */
7450 		if (replay == FALSE && !create)
7451 			rfs4_update_open_sequence(oo);
7452 		/*
7453 		 * If the client is receiving an error and the
7454 		 * open_owner needs to be confirmed, there is no way
7455 		 * to notify the client of this fact ignoring the fact
7456 		 * that the server has no method of returning a
7457 		 * stateid to confirm.  Therefore, the server needs to
7458 		 * mark this open_owner in a way as to avoid the
7459 		 * sequence id checking the next time the client uses
7460 		 * this open_owner.
7461 		 */
7462 		if (resp->status != NFS4_OK && oo->ro_need_confirm)
7463 			oo->ro_postpone_confirm = TRUE;
7464 		/*
7465 		 * If OK response then clear the postpone flag and
7466 		 * reset the sequence id to keep in sync with the
7467 		 * client.
7468 		 */
7469 		if (resp->status == NFS4_OK && oo->ro_postpone_confirm) {
7470 			oo->ro_postpone_confirm = FALSE;
7471 			oo->ro_open_seqid = args->seqid;
7472 		}
7473 		break;
7474 	}
7475 
7476 finish:
7477 	*cs->statusp = resp->status;
7478 
7479 	rfs4_sw_exit(&oo->ro_sw);
7480 	rfs4_openowner_rele(oo);
7481 
7482 end:
7483 	DTRACE_NFSV4_2(op__open__done, struct compound_state *, cs,
7484 	    OPEN4res *, resp);
7485 }
7486 
7487 /*ARGSUSED*/
7488 void
7489 rfs4_op_open_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
7490     struct svc_req *req, struct compound_state *cs)
7491 {
7492 	OPEN_CONFIRM4args *args = &argop->nfs_argop4_u.opopen_confirm;
7493 	OPEN_CONFIRM4res *resp = &resop->nfs_resop4_u.opopen_confirm;
7494 	rfs4_state_t *sp;
7495 	nfsstat4 status;
7496 
7497 	DTRACE_NFSV4_2(op__open__confirm__start, struct compound_state *, cs,
7498 	    OPEN_CONFIRM4args *, args);
7499 
7500 	if (cs->vp == NULL) {
7501 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7502 		goto out;
7503 	}
7504 
7505 	if (cs->vp->v_type != VREG) {
7506 		*cs->statusp = resp->status =
7507 		    cs->vp->v_type == VDIR ? NFS4ERR_ISDIR : NFS4ERR_INVAL;
7508 		return;
7509 	}
7510 
7511 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7512 	if (status != NFS4_OK) {
7513 		*cs->statusp = resp->status = status;
7514 		goto out;
7515 	}
7516 
7517 	/* Ensure specified filehandle matches */
7518 	if (cs->vp != sp->rs_finfo->rf_vp) {
7519 		rfs4_state_rele(sp);
7520 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7521 		goto out;
7522 	}
7523 
7524 	/* hold off other access to open_owner while we tinker */
7525 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7526 
7527 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7528 	case NFS4_CHECK_STATEID_OKAY:
7529 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7530 		    resop) != 0) {
7531 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7532 			break;
7533 		}
7534 		/*
7535 		 * If it is the appropriate stateid and determined to
7536 		 * be "OKAY" then this means that the stateid does not
7537 		 * need to be confirmed and the client is in error for
7538 		 * sending an OPEN_CONFIRM.
7539 		 */
7540 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7541 		break;
7542 	case NFS4_CHECK_STATEID_OLD:
7543 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7544 		break;
7545 	case NFS4_CHECK_STATEID_BAD:
7546 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7547 		break;
7548 	case NFS4_CHECK_STATEID_EXPIRED:
7549 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7550 		break;
7551 	case NFS4_CHECK_STATEID_CLOSED:
7552 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7553 		break;
7554 	case NFS4_CHECK_STATEID_REPLAY:
7555 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7556 		    resop)) {
7557 		case NFS4_CHKSEQ_OKAY:
7558 			/*
7559 			 * This is replayed stateid; if seqid matches
7560 			 * next expected, then client is using wrong seqid.
7561 			 */
7562 			/* fall through */
7563 		case NFS4_CHKSEQ_BAD:
7564 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7565 			break;
7566 		case NFS4_CHKSEQ_REPLAY:
7567 			/*
7568 			 * Note this case is the duplicate case so
7569 			 * resp->status is already set.
7570 			 */
7571 			*cs->statusp = resp->status;
7572 			rfs4_update_lease(sp->rs_owner->ro_client);
7573 			break;
7574 		}
7575 		break;
7576 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7577 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7578 		    resop) != NFS4_CHKSEQ_OKAY) {
7579 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7580 			break;
7581 		}
7582 		*cs->statusp = resp->status = NFS4_OK;
7583 
7584 		next_stateid(&sp->rs_stateid);
7585 		resp->open_stateid = sp->rs_stateid.stateid;
7586 		sp->rs_owner->ro_need_confirm = FALSE;
7587 		rfs4_update_lease(sp->rs_owner->ro_client);
7588 		rfs4_update_open_sequence(sp->rs_owner);
7589 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7590 		break;
7591 	default:
7592 		ASSERT(FALSE);
7593 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7594 		break;
7595 	}
7596 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7597 	rfs4_state_rele(sp);
7598 
7599 out:
7600 	DTRACE_NFSV4_2(op__open__confirm__done, struct compound_state *, cs,
7601 	    OPEN_CONFIRM4res *, resp);
7602 }
7603 
7604 /*ARGSUSED*/
7605 void
7606 rfs4_op_open_downgrade(nfs_argop4 *argop, nfs_resop4 *resop,
7607     struct svc_req *req, struct compound_state *cs)
7608 {
7609 	OPEN_DOWNGRADE4args *args = &argop->nfs_argop4_u.opopen_downgrade;
7610 	OPEN_DOWNGRADE4res *resp = &resop->nfs_resop4_u.opopen_downgrade;
7611 	uint32_t access = args->share_access;
7612 	uint32_t deny = args->share_deny;
7613 	nfsstat4 status;
7614 	rfs4_state_t *sp;
7615 	rfs4_file_t *fp;
7616 	int fflags = 0;
7617 
7618 	DTRACE_NFSV4_2(op__open__downgrade__start, struct compound_state *, cs,
7619 	    OPEN_DOWNGRADE4args *, args);
7620 
7621 	if (cs->vp == NULL) {
7622 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7623 		goto out;
7624 	}
7625 
7626 	if (cs->vp->v_type != VREG) {
7627 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7628 		return;
7629 	}
7630 
7631 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7632 	if (status != NFS4_OK) {
7633 		*cs->statusp = resp->status = status;
7634 		goto out;
7635 	}
7636 
7637 	/* Ensure specified filehandle matches */
7638 	if (cs->vp != sp->rs_finfo->rf_vp) {
7639 		rfs4_state_rele(sp);
7640 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7641 		goto out;
7642 	}
7643 
7644 	/* hold off other access to open_owner while we tinker */
7645 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7646 
7647 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7648 	case NFS4_CHECK_STATEID_OKAY:
7649 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7650 		    resop) != NFS4_CHKSEQ_OKAY) {
7651 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7652 			goto end;
7653 		}
7654 		break;
7655 	case NFS4_CHECK_STATEID_OLD:
7656 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7657 		goto end;
7658 	case NFS4_CHECK_STATEID_BAD:
7659 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7660 		goto end;
7661 	case NFS4_CHECK_STATEID_EXPIRED:
7662 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7663 		goto end;
7664 	case NFS4_CHECK_STATEID_CLOSED:
7665 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7666 		goto end;
7667 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7668 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7669 		goto end;
7670 	case NFS4_CHECK_STATEID_REPLAY:
7671 		/* Check the sequence id for the open owner */
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 			goto end;
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 			goto end;
7691 		}
7692 		break;
7693 	default:
7694 		ASSERT(FALSE);
7695 		break;
7696 	}
7697 
7698 	rfs4_dbe_lock(sp->rs_dbe);
7699 	/*
7700 	 * Check that the new access modes and deny modes are valid.
7701 	 * Check that no invalid bits are set.
7702 	 */
7703 	if ((access & ~(OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) ||
7704 	    (deny & ~(OPEN4_SHARE_DENY_READ | OPEN4_SHARE_DENY_WRITE))) {
7705 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7706 		rfs4_update_open_sequence(sp->rs_owner);
7707 		rfs4_dbe_unlock(sp->rs_dbe);
7708 		goto end;
7709 	}
7710 
7711 	/*
7712 	 * The new modes must be a subset of the current modes and
7713 	 * the access must specify at least one mode. To test that
7714 	 * the new mode is a subset of the current modes we bitwise
7715 	 * AND them together and check that the result equals the new
7716 	 * mode. For example:
7717 	 * New mode, access == R and current mode, sp->rs_open_access  == RW
7718 	 * access & sp->rs_open_access == R == access, so the new access mode
7719 	 * is valid. Consider access == RW, sp->rs_open_access = R
7720 	 * access & sp->rs_open_access == R != access, so the new access mode
7721 	 * is invalid.
7722 	 */
7723 	if ((access & sp->rs_open_access) != access ||
7724 	    (deny & sp->rs_open_deny) != deny ||
7725 	    (access &
7726 	    (OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) == 0) {
7727 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7728 		rfs4_update_open_sequence(sp->rs_owner);
7729 		rfs4_dbe_unlock(sp->rs_dbe);
7730 		goto end;
7731 	}
7732 
7733 	/*
7734 	 * Release any share locks associated with this stateID.
7735 	 * Strictly speaking, this violates the spec because the
7736 	 * spec effectively requires that open downgrade be atomic.
7737 	 * At present, fs_shrlock does not have this capability.
7738 	 */
7739 	(void) rfs4_unshare(sp);
7740 
7741 	status = rfs4_share(sp, access, deny);
7742 	if (status != NFS4_OK) {
7743 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7744 		rfs4_update_open_sequence(sp->rs_owner);
7745 		rfs4_dbe_unlock(sp->rs_dbe);
7746 		goto end;
7747 	}
7748 
7749 	fp = sp->rs_finfo;
7750 	rfs4_dbe_lock(fp->rf_dbe);
7751 
7752 	/*
7753 	 * If the current mode has deny read and the new mode
7754 	 * does not, decrement the number of deny read mode bits
7755 	 * and if it goes to zero turn off the deny read bit
7756 	 * on the file.
7757 	 */
7758 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_READ) &&
7759 	    (deny & OPEN4_SHARE_DENY_READ) == 0) {
7760 		fp->rf_deny_read--;
7761 		if (fp->rf_deny_read == 0)
7762 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
7763 	}
7764 
7765 	/*
7766 	 * If the current mode has deny write and the new mode
7767 	 * does not, decrement the number of deny write mode bits
7768 	 * and if it goes to zero turn off the deny write bit
7769 	 * on the file.
7770 	 */
7771 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) &&
7772 	    (deny & OPEN4_SHARE_DENY_WRITE) == 0) {
7773 		fp->rf_deny_write--;
7774 		if (fp->rf_deny_write == 0)
7775 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
7776 	}
7777 
7778 	/*
7779 	 * If the current mode has access read and the new mode
7780 	 * does not, decrement the number of access read mode bits
7781 	 * and if it goes to zero turn off the access read bit
7782 	 * on the file.  set fflags to FREAD for the call to
7783 	 * vn_open_downgrade().
7784 	 */
7785 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) &&
7786 	    (access & OPEN4_SHARE_ACCESS_READ) == 0) {
7787 		fp->rf_access_read--;
7788 		if (fp->rf_access_read == 0)
7789 			fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
7790 		fflags |= FREAD;
7791 	}
7792 
7793 	/*
7794 	 * If the current mode has access write and the new mode
7795 	 * does not, decrement the number of access write mode bits
7796 	 * and if it goes to zero turn off the access write bit
7797 	 * on the file.  set fflags to FWRITE for the call to
7798 	 * vn_open_downgrade().
7799 	 */
7800 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) &&
7801 	    (access & OPEN4_SHARE_ACCESS_WRITE) == 0) {
7802 		fp->rf_access_write--;
7803 		if (fp->rf_access_write == 0)
7804 			fp->rf_share_deny &= ~OPEN4_SHARE_ACCESS_WRITE;
7805 		fflags |= FWRITE;
7806 	}
7807 
7808 	/* Check that the file is still accessible */
7809 	ASSERT(fp->rf_share_access);
7810 
7811 	rfs4_dbe_unlock(fp->rf_dbe);
7812 
7813 	/* now set the new open access and deny modes */
7814 	sp->rs_open_access = access;
7815 	sp->rs_open_deny = deny;
7816 
7817 	/*
7818 	 * we successfully downgraded the share lock, now we need to downgrade
7819 	 * the open. it is possible that the downgrade was only for a deny
7820 	 * mode and we have nothing else to do.
7821 	 */
7822 	if ((fflags & (FREAD|FWRITE)) != 0)
7823 		vn_open_downgrade(cs->vp, fflags);
7824 
7825 	/* Update the stateid */
7826 	next_stateid(&sp->rs_stateid);
7827 	resp->open_stateid = sp->rs_stateid.stateid;
7828 
7829 	rfs4_dbe_unlock(sp->rs_dbe);
7830 
7831 	*cs->statusp = resp->status = NFS4_OK;
7832 	/* Update the lease */
7833 	rfs4_update_lease(sp->rs_owner->ro_client);
7834 	/* And the sequence */
7835 	rfs4_update_open_sequence(sp->rs_owner);
7836 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7837 
7838 end:
7839 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7840 	rfs4_state_rele(sp);
7841 out:
7842 	DTRACE_NFSV4_2(op__open__downgrade__done, struct compound_state *, cs,
7843 	    OPEN_DOWNGRADE4res *, resp);
7844 }
7845 
7846 static void *
7847 memstr(const void *s1, const char *s2, size_t n)
7848 {
7849 	size_t l = strlen(s2);
7850 	char *p = (char *)s1;
7851 
7852 	while (n >= l) {
7853 		if (bcmp(p, s2, l) == 0)
7854 			return (p);
7855 		p++;
7856 		n--;
7857 	}
7858 
7859 	return (NULL);
7860 }
7861 
7862 /*
7863  * The logic behind this function is detailed in the NFSv4 RFC in the
7864  * SETCLIENTID operation description under IMPLEMENTATION.  Refer to
7865  * that section for explicit guidance to server behavior for
7866  * SETCLIENTID.
7867  */
7868 void
7869 rfs4_op_setclientid(nfs_argop4 *argop, nfs_resop4 *resop,
7870     struct svc_req *req, struct compound_state *cs)
7871 {
7872 	SETCLIENTID4args *args = &argop->nfs_argop4_u.opsetclientid;
7873 	SETCLIENTID4res *res = &resop->nfs_resop4_u.opsetclientid;
7874 	rfs4_client_t *cp, *newcp, *cp_confirmed, *cp_unconfirmed;
7875 	rfs4_clntip_t *ci;
7876 	bool_t create;
7877 	char *addr, *netid;
7878 	int len;
7879 
7880 	DTRACE_NFSV4_2(op__setclientid__start, struct compound_state *, cs,
7881 	    SETCLIENTID4args *, args);
7882 retry:
7883 	newcp = cp_confirmed = cp_unconfirmed = NULL;
7884 
7885 	/*
7886 	 * Save the caller's IP address
7887 	 */
7888 	args->client.cl_addr =
7889 	    (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
7890 
7891 	/*
7892 	 * Record if it is a Solaris client that cannot handle referrals.
7893 	 */
7894 	if (memstr(args->client.id_val, "Solaris", args->client.id_len) &&
7895 	    !memstr(args->client.id_val, "+referrals", args->client.id_len)) {
7896 		/* Add a "yes, it's downrev" record */
7897 		create = TRUE;
7898 		ci = rfs4_find_clntip(args->client.cl_addr, &create);
7899 		ASSERT(ci != NULL);
7900 		rfs4_dbe_rele(ci->ri_dbe);
7901 	} else {
7902 		/* Remove any previous record */
7903 		rfs4_invalidate_clntip(args->client.cl_addr);
7904 	}
7905 
7906 	/*
7907 	 * In search of an EXISTING client matching the incoming
7908 	 * request to establish a new client identifier at the server
7909 	 */
7910 	create = TRUE;
7911 	cp = rfs4_findclient(&args->client, &create, NULL);
7912 
7913 	/* Should never happen */
7914 	ASSERT(cp != NULL);
7915 
7916 	if (cp == NULL) {
7917 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
7918 		goto out;
7919 	}
7920 
7921 	/*
7922 	 * Easiest case. Client identifier is newly created and is
7923 	 * unconfirmed.  Also note that for this case, no other
7924 	 * entries exist for the client identifier.  Nothing else to
7925 	 * check.  Just setup the response and respond.
7926 	 */
7927 	if (create) {
7928 		*cs->statusp = res->status = NFS4_OK;
7929 		res->SETCLIENTID4res_u.resok4.clientid = cp->rc_clientid;
7930 		res->SETCLIENTID4res_u.resok4.setclientid_confirm =
7931 		    cp->rc_confirm_verf;
7932 		/* Setup callback information; CB_NULL confirmation later */
7933 		rfs4_client_setcb(cp, &args->callback, args->callback_ident);
7934 
7935 		rfs4_client_rele(cp);
7936 		goto out;
7937 	}
7938 
7939 	/*
7940 	 * An existing, confirmed client may exist but it may not have
7941 	 * been active for at least one lease period.  If so, then
7942 	 * "close" the client and create a new client identifier
7943 	 */
7944 	if (rfs4_lease_expired(cp)) {
7945 		rfs4_client_close(cp);
7946 		goto retry;
7947 	}
7948 
7949 	if (cp->rc_need_confirm == TRUE)
7950 		cp_unconfirmed = cp;
7951 	else
7952 		cp_confirmed = cp;
7953 
7954 	cp = NULL;
7955 
7956 	/*
7957 	 * We have a confirmed client, now check for an
7958 	 * unconfimred entry
7959 	 */
7960 	if (cp_confirmed) {
7961 		/* If creds don't match then client identifier is inuse */
7962 		if (!creds_ok(cp_confirmed->rc_cr_set, req, cs)) {
7963 			rfs4_cbinfo_t *cbp;
7964 			/*
7965 			 * Some one else has established this client
7966 			 * id. Try and say * who they are. We will use
7967 			 * the call back address supplied by * the
7968 			 * first client.
7969 			 */
7970 			*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
7971 
7972 			addr = netid = NULL;
7973 
7974 			cbp = &cp_confirmed->rc_cbinfo;
7975 			if (cbp->cb_callback.cb_location.r_addr &&
7976 			    cbp->cb_callback.cb_location.r_netid) {
7977 				cb_client4 *cbcp = &cbp->cb_callback;
7978 
7979 				len = strlen(cbcp->cb_location.r_addr)+1;
7980 				addr = kmem_alloc(len, KM_SLEEP);
7981 				bcopy(cbcp->cb_location.r_addr, addr, len);
7982 				len = strlen(cbcp->cb_location.r_netid)+1;
7983 				netid = kmem_alloc(len, KM_SLEEP);
7984 				bcopy(cbcp->cb_location.r_netid, netid, len);
7985 			}
7986 
7987 			res->SETCLIENTID4res_u.client_using.r_addr = addr;
7988 			res->SETCLIENTID4res_u.client_using.r_netid = netid;
7989 
7990 			rfs4_client_rele(cp_confirmed);
7991 		}
7992 
7993 		/*
7994 		 * Confirmed, creds match, and verifier matches; must
7995 		 * be an update of the callback info
7996 		 */
7997 		if (cp_confirmed->rc_nfs_client.verifier ==
7998 		    args->client.verifier) {
7999 			/* Setup callback information */
8000 			rfs4_client_setcb(cp_confirmed, &args->callback,
8001 			    args->callback_ident);
8002 
8003 			/* everything okay -- move ahead */
8004 			*cs->statusp = res->status = NFS4_OK;
8005 			res->SETCLIENTID4res_u.resok4.clientid =
8006 			    cp_confirmed->rc_clientid;
8007 
8008 			/* update the confirm_verifier and return it */
8009 			rfs4_client_scv_next(cp_confirmed);
8010 			res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8011 			    cp_confirmed->rc_confirm_verf;
8012 
8013 			rfs4_client_rele(cp_confirmed);
8014 			goto out;
8015 		}
8016 
8017 		/*
8018 		 * Creds match but the verifier doesn't.  Must search
8019 		 * for an unconfirmed client that would be replaced by
8020 		 * this request.
8021 		 */
8022 		create = FALSE;
8023 		cp_unconfirmed = rfs4_findclient(&args->client, &create,
8024 		    cp_confirmed);
8025 	}
8026 
8027 	/*
8028 	 * At this point, we have taken care of the brand new client
8029 	 * struct, INUSE case, update of an existing, and confirmed
8030 	 * client struct.
8031 	 */
8032 
8033 	/*
8034 	 * check to see if things have changed while we originally
8035 	 * picked up the client struct.  If they have, then return and
8036 	 * retry the processing of this SETCLIENTID request.
8037 	 */
8038 	if (cp_unconfirmed) {
8039 		rfs4_dbe_lock(cp_unconfirmed->rc_dbe);
8040 		if (!cp_unconfirmed->rc_need_confirm) {
8041 			rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8042 			rfs4_client_rele(cp_unconfirmed);
8043 			if (cp_confirmed)
8044 				rfs4_client_rele(cp_confirmed);
8045 			goto retry;
8046 		}
8047 		/* do away with the old unconfirmed one */
8048 		rfs4_dbe_invalidate(cp_unconfirmed->rc_dbe);
8049 		rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8050 		rfs4_client_rele(cp_unconfirmed);
8051 		cp_unconfirmed = NULL;
8052 	}
8053 
8054 	/*
8055 	 * This search will temporarily hide the confirmed client
8056 	 * struct while a new client struct is created as the
8057 	 * unconfirmed one.
8058 	 */
8059 	create = TRUE;
8060 	newcp = rfs4_findclient(&args->client, &create, cp_confirmed);
8061 
8062 	ASSERT(newcp != NULL);
8063 
8064 	if (newcp == NULL) {
8065 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8066 		rfs4_client_rele(cp_confirmed);
8067 		goto out;
8068 	}
8069 
8070 	/*
8071 	 * If one was not created, then a similar request must be in
8072 	 * process so release and start over with this one
8073 	 */
8074 	if (create != TRUE) {
8075 		rfs4_client_rele(newcp);
8076 		if (cp_confirmed)
8077 			rfs4_client_rele(cp_confirmed);
8078 		goto retry;
8079 	}
8080 
8081 	*cs->statusp = res->status = NFS4_OK;
8082 	res->SETCLIENTID4res_u.resok4.clientid = newcp->rc_clientid;
8083 	res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8084 	    newcp->rc_confirm_verf;
8085 	/* Setup callback information; CB_NULL confirmation later */
8086 	rfs4_client_setcb(newcp, &args->callback, args->callback_ident);
8087 
8088 	newcp->rc_cp_confirmed = cp_confirmed;
8089 
8090 	rfs4_client_rele(newcp);
8091 
8092 out:
8093 	DTRACE_NFSV4_2(op__setclientid__done, struct compound_state *, cs,
8094 	    SETCLIENTID4res *, res);
8095 }
8096 
8097 /*ARGSUSED*/
8098 void
8099 rfs4_op_setclientid_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
8100     struct svc_req *req, struct compound_state *cs)
8101 {
8102 	SETCLIENTID_CONFIRM4args *args =
8103 	    &argop->nfs_argop4_u.opsetclientid_confirm;
8104 	SETCLIENTID_CONFIRM4res *res =
8105 	    &resop->nfs_resop4_u.opsetclientid_confirm;
8106 	rfs4_client_t *cp, *cptoclose = NULL;
8107 
8108 	DTRACE_NFSV4_2(op__setclientid__confirm__start,
8109 	    struct compound_state *, cs,
8110 	    SETCLIENTID_CONFIRM4args *, args);
8111 
8112 	*cs->statusp = res->status = NFS4_OK;
8113 
8114 	cp = rfs4_findclient_by_id(args->clientid, TRUE);
8115 
8116 	if (cp == NULL) {
8117 		*cs->statusp = res->status =
8118 		    rfs4_check_clientid(&args->clientid, 1);
8119 		goto out;
8120 	}
8121 
8122 	if (!creds_ok(cp, req, cs)) {
8123 		*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8124 		rfs4_client_rele(cp);
8125 		goto out;
8126 	}
8127 
8128 	/* If the verifier doesn't match, the record doesn't match */
8129 	if (cp->rc_confirm_verf != args->setclientid_confirm) {
8130 		*cs->statusp = res->status = NFS4ERR_STALE_CLIENTID;
8131 		rfs4_client_rele(cp);
8132 		goto out;
8133 	}
8134 
8135 	rfs4_dbe_lock(cp->rc_dbe);
8136 	cp->rc_need_confirm = FALSE;
8137 	if (cp->rc_cp_confirmed) {
8138 		cptoclose = cp->rc_cp_confirmed;
8139 		cptoclose->rc_ss_remove = 1;
8140 		cp->rc_cp_confirmed = NULL;
8141 	}
8142 
8143 	/*
8144 	 * Update the client's associated server instance, if it's changed
8145 	 * since the client was created.
8146 	 */
8147 	if (rfs4_servinst(cp) != rfs4_cur_servinst)
8148 		rfs4_servinst_assign(cp, rfs4_cur_servinst);
8149 
8150 	/*
8151 	 * Record clientid in stable storage.
8152 	 * Must be done after server instance has been assigned.
8153 	 */
8154 	rfs4_ss_clid(cp);
8155 
8156 	rfs4_dbe_unlock(cp->rc_dbe);
8157 
8158 	if (cptoclose)
8159 		/* don't need to rele, client_close does it */
8160 		rfs4_client_close(cptoclose);
8161 
8162 	/* If needed, initiate CB_NULL call for callback path */
8163 	rfs4_deleg_cb_check(cp);
8164 	rfs4_update_lease(cp);
8165 
8166 	/*
8167 	 * Check to see if client can perform reclaims
8168 	 */
8169 	rfs4_ss_chkclid(cp);
8170 
8171 	rfs4_client_rele(cp);
8172 
8173 out:
8174 	DTRACE_NFSV4_2(op__setclientid__confirm__done,
8175 	    struct compound_state *, cs,
8176 	    SETCLIENTID_CONFIRM4 *, res);
8177 }
8178 
8179 
8180 /*ARGSUSED*/
8181 void
8182 rfs4_op_close(nfs_argop4 *argop, nfs_resop4 *resop,
8183     struct svc_req *req, struct compound_state *cs)
8184 {
8185 	CLOSE4args *args = &argop->nfs_argop4_u.opclose;
8186 	CLOSE4res *resp = &resop->nfs_resop4_u.opclose;
8187 	rfs4_state_t *sp;
8188 	nfsstat4 status;
8189 
8190 	DTRACE_NFSV4_2(op__close__start, struct compound_state *, cs,
8191 	    CLOSE4args *, args);
8192 
8193 	if (cs->vp == NULL) {
8194 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8195 		goto out;
8196 	}
8197 
8198 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_INVALID);
8199 	if (status != NFS4_OK) {
8200 		*cs->statusp = resp->status = status;
8201 		goto out;
8202 	}
8203 
8204 	/* Ensure specified filehandle matches */
8205 	if (cs->vp != sp->rs_finfo->rf_vp) {
8206 		rfs4_state_rele(sp);
8207 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8208 		goto out;
8209 	}
8210 
8211 	/* hold off other access to open_owner while we tinker */
8212 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
8213 
8214 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
8215 	case NFS4_CHECK_STATEID_OKAY:
8216 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8217 		    resop) != NFS4_CHKSEQ_OKAY) {
8218 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8219 			goto end;
8220 		}
8221 		break;
8222 	case NFS4_CHECK_STATEID_OLD:
8223 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8224 		goto end;
8225 	case NFS4_CHECK_STATEID_BAD:
8226 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8227 		goto end;
8228 	case NFS4_CHECK_STATEID_EXPIRED:
8229 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8230 		goto end;
8231 	case NFS4_CHECK_STATEID_CLOSED:
8232 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8233 		goto end;
8234 	case NFS4_CHECK_STATEID_UNCONFIRMED:
8235 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8236 		goto end;
8237 	case NFS4_CHECK_STATEID_REPLAY:
8238 		/* Check the sequence id for the open owner */
8239 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8240 		    resop)) {
8241 		case NFS4_CHKSEQ_OKAY:
8242 			/*
8243 			 * This is replayed stateid; if seqid matches
8244 			 * next expected, then client is using wrong seqid.
8245 			 */
8246 			/* FALL THROUGH */
8247 		case NFS4_CHKSEQ_BAD:
8248 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8249 			goto end;
8250 		case NFS4_CHKSEQ_REPLAY:
8251 			/*
8252 			 * Note this case is the duplicate case so
8253 			 * resp->status is already set.
8254 			 */
8255 			*cs->statusp = resp->status;
8256 			rfs4_update_lease(sp->rs_owner->ro_client);
8257 			goto end;
8258 		}
8259 		break;
8260 	default:
8261 		ASSERT(FALSE);
8262 		break;
8263 	}
8264 
8265 	rfs4_dbe_lock(sp->rs_dbe);
8266 
8267 	/* Update the stateid. */
8268 	next_stateid(&sp->rs_stateid);
8269 	resp->open_stateid = sp->rs_stateid.stateid;
8270 
8271 	rfs4_dbe_unlock(sp->rs_dbe);
8272 
8273 	rfs4_update_lease(sp->rs_owner->ro_client);
8274 	rfs4_update_open_sequence(sp->rs_owner);
8275 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8276 
8277 	rfs4_state_close(sp, FALSE, FALSE, cs->cr);
8278 
8279 	*cs->statusp = resp->status = status;
8280 
8281 end:
8282 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
8283 	rfs4_state_rele(sp);
8284 out:
8285 	DTRACE_NFSV4_2(op__close__done, struct compound_state *, cs,
8286 	    CLOSE4res *, resp);
8287 }
8288 
8289 /*
8290  * Manage the counts on the file struct and close all file locks
8291  */
8292 /*ARGSUSED*/
8293 void
8294 rfs4_release_share_lock_state(rfs4_state_t *sp, cred_t *cr,
8295     bool_t close_of_client)
8296 {
8297 	rfs4_file_t *fp = sp->rs_finfo;
8298 	rfs4_lo_state_t *lsp;
8299 	int fflags = 0;
8300 
8301 	/*
8302 	 * If this call is part of the larger closing down of client
8303 	 * state then it is just easier to release all locks
8304 	 * associated with this client instead of going through each
8305 	 * individual file and cleaning locks there.
8306 	 */
8307 	if (close_of_client) {
8308 		if (sp->rs_owner->ro_client->rc_unlksys_completed == FALSE &&
8309 		    !list_is_empty(&sp->rs_lostatelist) &&
8310 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID) {
8311 			/* Is the PxFS kernel module loaded? */
8312 			if (lm_remove_file_locks != NULL) {
8313 				int new_sysid;
8314 
8315 				/* Encode the cluster nodeid in new sysid */
8316 				new_sysid = sp->rs_owner->ro_client->rc_sysidt;
8317 				lm_set_nlmid_flk(&new_sysid);
8318 
8319 				/*
8320 				 * This PxFS routine removes file locks for a
8321 				 * client over all nodes of a cluster.
8322 				 */
8323 				NFS4_DEBUG(rfs4_debug, (CE_NOTE,
8324 				    "lm_remove_file_locks(sysid=0x%x)\n",
8325 				    new_sysid));
8326 				(*lm_remove_file_locks)(new_sysid);
8327 			} else {
8328 				struct flock64 flk;
8329 
8330 				/* Release all locks for this client */
8331 				flk.l_type = F_UNLKSYS;
8332 				flk.l_whence = 0;
8333 				flk.l_start = 0;
8334 				flk.l_len = 0;
8335 				flk.l_sysid =
8336 				    sp->rs_owner->ro_client->rc_sysidt;
8337 				flk.l_pid = 0;
8338 				(void) VOP_FRLOCK(sp->rs_finfo->rf_vp, F_SETLK,
8339 				    &flk, F_REMOTELOCK | FREAD | FWRITE,
8340 				    (u_offset_t)0, NULL, CRED(), NULL);
8341 			}
8342 
8343 			sp->rs_owner->ro_client->rc_unlksys_completed = TRUE;
8344 		}
8345 	}
8346 
8347 	/*
8348 	 * Release all locks on this file by this lock owner or at
8349 	 * least mark the locks as having been released
8350 	 */
8351 	for (lsp = list_head(&sp->rs_lostatelist); lsp != NULL;
8352 	    lsp = list_next(&sp->rs_lostatelist, lsp)) {
8353 		lsp->rls_locks_cleaned = TRUE;
8354 
8355 		/* Was this already taken care of above? */
8356 		if (!close_of_client &&
8357 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8358 			(void) cleanlocks(sp->rs_finfo->rf_vp,
8359 			    lsp->rls_locker->rl_pid,
8360 			    lsp->rls_locker->rl_client->rc_sysidt);
8361 	}
8362 
8363 	/*
8364 	 * Release any shrlocks associated with this open state ID.
8365 	 * This must be done before the rfs4_state gets marked closed.
8366 	 */
8367 	if (sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8368 		(void) rfs4_unshare(sp);
8369 
8370 	if (sp->rs_open_access) {
8371 		rfs4_dbe_lock(fp->rf_dbe);
8372 
8373 		/*
8374 		 * Decrement the count for each access and deny bit that this
8375 		 * state has contributed to the file.
8376 		 * If the file counts go to zero
8377 		 * clear the appropriate bit in the appropriate mask.
8378 		 */
8379 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) {
8380 			fp->rf_access_read--;
8381 			fflags |= FREAD;
8382 			if (fp->rf_access_read == 0)
8383 				fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
8384 		}
8385 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) {
8386 			fp->rf_access_write--;
8387 			fflags |= FWRITE;
8388 			if (fp->rf_access_write == 0)
8389 				fp->rf_share_access &=
8390 				    ~OPEN4_SHARE_ACCESS_WRITE;
8391 		}
8392 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_READ) {
8393 			fp->rf_deny_read--;
8394 			if (fp->rf_deny_read == 0)
8395 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
8396 		}
8397 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) {
8398 			fp->rf_deny_write--;
8399 			if (fp->rf_deny_write == 0)
8400 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
8401 		}
8402 
8403 		(void) VOP_CLOSE(fp->rf_vp, fflags, 1, (offset_t)0, cr, NULL);
8404 
8405 		rfs4_dbe_unlock(fp->rf_dbe);
8406 
8407 		sp->rs_open_access = 0;
8408 		sp->rs_open_deny = 0;
8409 	}
8410 }
8411 
8412 /*
8413  * lock_denied: Fill in a LOCK4deneid structure given an flock64 structure.
8414  */
8415 static nfsstat4
8416 lock_denied(LOCK4denied *dp, struct flock64 *flk)
8417 {
8418 	rfs4_lockowner_t *lo;
8419 	rfs4_client_t *cp;
8420 	uint32_t len;
8421 
8422 	lo = rfs4_findlockowner_by_pid(flk->l_pid);
8423 	if (lo != NULL) {
8424 		cp = lo->rl_client;
8425 		if (rfs4_lease_expired(cp)) {
8426 			rfs4_lockowner_rele(lo);
8427 			rfs4_dbe_hold(cp->rc_dbe);
8428 			rfs4_client_close(cp);
8429 			return (NFS4ERR_EXPIRED);
8430 		}
8431 		dp->owner.clientid = lo->rl_owner.clientid;
8432 		len = lo->rl_owner.owner_len;
8433 		dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8434 		bcopy(lo->rl_owner.owner_val, dp->owner.owner_val, len);
8435 		dp->owner.owner_len = len;
8436 		rfs4_lockowner_rele(lo);
8437 		goto finish;
8438 	}
8439 
8440 	/*
8441 	 * Its not a NFS4 lock. We take advantage that the upper 32 bits
8442 	 * of the client id contain the boot time for a NFS4 lock. So we
8443 	 * fabricate and identity by setting clientid to the sysid, and
8444 	 * the lock owner to the pid.
8445 	 */
8446 	dp->owner.clientid = flk->l_sysid;
8447 	len = sizeof (pid_t);
8448 	dp->owner.owner_len = len;
8449 	dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8450 	bcopy(&flk->l_pid, dp->owner.owner_val, len);
8451 finish:
8452 	dp->offset = flk->l_start;
8453 	dp->length = flk->l_len;
8454 
8455 	if (flk->l_type == F_RDLCK)
8456 		dp->locktype = READ_LT;
8457 	else if (flk->l_type == F_WRLCK)
8458 		dp->locktype = WRITE_LT;
8459 	else
8460 		return (NFS4ERR_INVAL);	/* no mapping from POSIX ltype to v4 */
8461 
8462 	return (NFS4_OK);
8463 }
8464 
8465 /*
8466  * The NFSv4.0 LOCK operation does not support the blocking lock (at the
8467  * NFSv4.0 protocol level) so the client needs to resend the LOCK request in a
8468  * case the lock is denied by the NFSv4.0 server.  NFSv4.0 clients are prepared
8469  * for that (obviously); they are sending the LOCK requests with some delays
8470  * between the attempts.  See nfs4frlock() and nfs4_block_and_wait() for the
8471  * locking and delay implementation at the client side.
8472  *
8473  * To make the life of the clients easier, the NFSv4.0 server tries to do some
8474  * fast retries on its own (the for loop below) in a hope the lock will be
8475  * available soon.  And if not, the client won't need to resend the LOCK
8476  * requests so fast to check the lock availability.  This basically saves some
8477  * network traffic and tries to make sure the client gets the lock ASAP.
8478  */
8479 static int
8480 setlock(vnode_t *vp, struct flock64 *flock, int flag, cred_t *cred)
8481 {
8482 	int error;
8483 	struct flock64 flk;
8484 	int i;
8485 	clock_t delaytime;
8486 	int cmd;
8487 	int spin_cnt = 0;
8488 
8489 	cmd = nbl_need_check(vp) ? F_SETLK_NBMAND : F_SETLK;
8490 retry:
8491 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
8492 
8493 	for (i = 0; i < rfs4_maxlock_tries; i++) {
8494 		LOCK_PRINT(rfs4_debug, "setlock", cmd, flock);
8495 		error = VOP_FRLOCK(vp, cmd,
8496 		    flock, flag, (u_offset_t)0, NULL, cred, NULL);
8497 
8498 		if (error != EAGAIN && error != EACCES)
8499 			break;
8500 
8501 		if (i < rfs4_maxlock_tries - 1) {
8502 			delay(delaytime);
8503 			delaytime *= 2;
8504 		}
8505 	}
8506 
8507 	if (error == EAGAIN || error == EACCES) {
8508 		/* Get the owner of the lock */
8509 		flk = *flock;
8510 		LOCK_PRINT(rfs4_debug, "setlock", F_GETLK, &flk);
8511 		if (VOP_FRLOCK(vp, F_GETLK, &flk, flag, 0, NULL, cred,
8512 		    NULL) == 0) {
8513 			/*
8514 			 * There's a race inherent in the current VOP_FRLOCK
8515 			 * design where:
8516 			 * a: "other guy" takes a lock that conflicts with a
8517 			 * lock we want
8518 			 * b: we attempt to take our lock (non-blocking) and
8519 			 * the attempt fails.
8520 			 * c: "other guy" releases the conflicting lock
8521 			 * d: we ask what lock conflicts with the lock we want,
8522 			 * getting F_UNLCK (no lock blocks us)
8523 			 *
8524 			 * If we retry the non-blocking lock attempt in this
8525 			 * case (restart at step 'b') there's some possibility
8526 			 * that many such attempts might fail.  However a test
8527 			 * designed to actually provoke this race shows that
8528 			 * the vast majority of cases require no retry, and
8529 			 * only a few took as many as three retries.  Here's
8530 			 * the test outcome:
8531 			 *
8532 			 *	   number of retries    how many times we needed
8533 			 *				that many retries
8534 			 *	   0			79461
8535 			 *	   1			  862
8536 			 *	   2			   49
8537 			 *	   3			    5
8538 			 *
8539 			 * Given those empirical results, we arbitrarily limit
8540 			 * the retry count to ten.
8541 			 *
8542 			 * If we actually make to ten retries and give up,
8543 			 * nothing catastrophic happens, but we're unable to
8544 			 * return the information about the conflicting lock to
8545 			 * the NFS client.  That's an acceptable trade off vs.
8546 			 * letting this retry loop run forever.
8547 			 */
8548 			if (flk.l_type == F_UNLCK) {
8549 				if (spin_cnt++ < 10) {
8550 					/* No longer locked, retry */
8551 					goto retry;
8552 				}
8553 			} else {
8554 				*flock = flk;
8555 				LOCK_PRINT(rfs4_debug, "setlock(blocking lock)",
8556 				    F_GETLK, &flk);
8557 			}
8558 		}
8559 	}
8560 
8561 	return (error);
8562 }
8563 
8564 /*ARGSUSED*/
8565 static nfsstat4
8566 rfs4_do_lock(rfs4_lo_state_t *lsp, nfs_lock_type4 locktype,
8567     offset4 offset, length4 length, cred_t *cred, nfs_resop4 *resop)
8568 {
8569 	nfsstat4 status;
8570 	rfs4_lockowner_t *lo = lsp->rls_locker;
8571 	rfs4_state_t *sp = lsp->rls_state;
8572 	struct flock64 flock;
8573 	int16_t ltype;
8574 	int flag;
8575 	int error;
8576 	sysid_t sysid;
8577 	LOCK4res *lres;
8578 	vnode_t *vp;
8579 
8580 	if (rfs4_lease_expired(lo->rl_client)) {
8581 		return (NFS4ERR_EXPIRED);
8582 	}
8583 
8584 	if ((status = rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
8585 		return (status);
8586 
8587 	/* Check for zero length. To lock to end of file use all ones for V4 */
8588 	if (length == 0)
8589 		return (NFS4ERR_INVAL);
8590 	else if (length == (length4)(~0))
8591 		length = 0;		/* Posix to end of file  */
8592 
8593 retry:
8594 	rfs4_dbe_lock(sp->rs_dbe);
8595 	if (sp->rs_closed == TRUE) {
8596 		rfs4_dbe_unlock(sp->rs_dbe);
8597 		return (NFS4ERR_OLD_STATEID);
8598 	}
8599 
8600 	if (resop->resop != OP_LOCKU) {
8601 		switch (locktype) {
8602 		case READ_LT:
8603 		case READW_LT:
8604 			if ((sp->rs_share_access
8605 			    & OPEN4_SHARE_ACCESS_READ) == 0) {
8606 				rfs4_dbe_unlock(sp->rs_dbe);
8607 
8608 				return (NFS4ERR_OPENMODE);
8609 			}
8610 			ltype = F_RDLCK;
8611 			break;
8612 		case WRITE_LT:
8613 		case WRITEW_LT:
8614 			if ((sp->rs_share_access
8615 			    & OPEN4_SHARE_ACCESS_WRITE) == 0) {
8616 				rfs4_dbe_unlock(sp->rs_dbe);
8617 
8618 				return (NFS4ERR_OPENMODE);
8619 			}
8620 			ltype = F_WRLCK;
8621 			break;
8622 		}
8623 	} else
8624 		ltype = F_UNLCK;
8625 
8626 	flock.l_type = ltype;
8627 	flock.l_whence = 0;		/* SEEK_SET */
8628 	flock.l_start = offset;
8629 	flock.l_len = length;
8630 	flock.l_sysid = sysid;
8631 	flock.l_pid = lsp->rls_locker->rl_pid;
8632 
8633 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
8634 	if (flock.l_len < 0 || flock.l_start < 0) {
8635 		rfs4_dbe_unlock(sp->rs_dbe);
8636 		return (NFS4ERR_INVAL);
8637 	}
8638 
8639 	/*
8640 	 * N.B. FREAD has the same value as OPEN4_SHARE_ACCESS_READ and
8641 	 * FWRITE has the same value as OPEN4_SHARE_ACCESS_WRITE.
8642 	 */
8643 	flag = (int)sp->rs_share_access | F_REMOTELOCK;
8644 
8645 	vp = sp->rs_finfo->rf_vp;
8646 	VN_HOLD(vp);
8647 
8648 	/*
8649 	 * We need to unlock sp before we call the underlying filesystem to
8650 	 * acquire the file lock.
8651 	 */
8652 	rfs4_dbe_unlock(sp->rs_dbe);
8653 
8654 	error = setlock(vp, &flock, flag, cred);
8655 
8656 	/*
8657 	 * Make sure the file is still open.  In a case the file was closed in
8658 	 * the meantime, clean the lock we acquired using the setlock() call
8659 	 * above, and return the appropriate error.
8660 	 */
8661 	rfs4_dbe_lock(sp->rs_dbe);
8662 	if (sp->rs_closed == TRUE) {
8663 		cleanlocks(vp, lsp->rls_locker->rl_pid, sysid);
8664 		rfs4_dbe_unlock(sp->rs_dbe);
8665 
8666 		VN_RELE(vp);
8667 
8668 		return (NFS4ERR_OLD_STATEID);
8669 	}
8670 	rfs4_dbe_unlock(sp->rs_dbe);
8671 
8672 	VN_RELE(vp);
8673 
8674 	if (error == 0) {
8675 		rfs4_dbe_lock(lsp->rls_dbe);
8676 		next_stateid(&lsp->rls_lockid);
8677 		rfs4_dbe_unlock(lsp->rls_dbe);
8678 	}
8679 
8680 	/*
8681 	 * N.B. We map error values to nfsv4 errors. This is differrent
8682 	 * than puterrno4 routine.
8683 	 */
8684 	switch (error) {
8685 	case 0:
8686 		status = NFS4_OK;
8687 		break;
8688 	case EAGAIN:
8689 	case EACCES:		/* Old value */
8690 		/* Can only get here if op is OP_LOCK */
8691 		ASSERT(resop->resop == OP_LOCK);
8692 		lres = &resop->nfs_resop4_u.oplock;
8693 		status = NFS4ERR_DENIED;
8694 		if (lock_denied(&lres->LOCK4res_u.denied, &flock)
8695 		    == NFS4ERR_EXPIRED)
8696 			goto retry;
8697 		break;
8698 	case ENOLCK:
8699 		status = NFS4ERR_DELAY;
8700 		break;
8701 	case EOVERFLOW:
8702 		status = NFS4ERR_INVAL;
8703 		break;
8704 	case EINVAL:
8705 		status = NFS4ERR_NOTSUPP;
8706 		break;
8707 	default:
8708 		status = NFS4ERR_SERVERFAULT;
8709 		break;
8710 	}
8711 
8712 	return (status);
8713 }
8714 
8715 /*ARGSUSED*/
8716 void
8717 rfs4_op_lock(nfs_argop4 *argop, nfs_resop4 *resop,
8718     struct svc_req *req, struct compound_state *cs)
8719 {
8720 	LOCK4args *args = &argop->nfs_argop4_u.oplock;
8721 	LOCK4res *resp = &resop->nfs_resop4_u.oplock;
8722 	nfsstat4 status;
8723 	stateid4 *stateid;
8724 	rfs4_lockowner_t *lo;
8725 	rfs4_client_t *cp;
8726 	rfs4_state_t *sp = NULL;
8727 	rfs4_lo_state_t *lsp = NULL;
8728 	bool_t ls_sw_held = FALSE;
8729 	bool_t create = TRUE;
8730 	bool_t lcreate = TRUE;
8731 	bool_t dup_lock = FALSE;
8732 	int rc;
8733 
8734 	DTRACE_NFSV4_2(op__lock__start, struct compound_state *, cs,
8735 	    LOCK4args *, args);
8736 
8737 	if (cs->vp == NULL) {
8738 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8739 		DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8740 		    cs, LOCK4res *, resp);
8741 		return;
8742 	}
8743 
8744 	if (args->locker.new_lock_owner) {
8745 		/* Create a new lockowner for this instance */
8746 		open_to_lock_owner4 *olo = &args->locker.locker4_u.open_owner;
8747 
8748 		NFS4_DEBUG(rfs4_debug, (CE_NOTE, "Creating new lock owner"));
8749 
8750 		stateid = &olo->open_stateid;
8751 		status = rfs4_get_state(stateid, &sp, RFS4_DBS_VALID);
8752 		if (status != NFS4_OK) {
8753 			NFS4_DEBUG(rfs4_debug,
8754 			    (CE_NOTE, "Get state failed in lock %d", status));
8755 			*cs->statusp = resp->status = status;
8756 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8757 			    cs, LOCK4res *, resp);
8758 			return;
8759 		}
8760 
8761 		/* Ensure specified filehandle matches */
8762 		if (cs->vp != sp->rs_finfo->rf_vp) {
8763 			rfs4_state_rele(sp);
8764 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8765 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8766 			    cs, LOCK4res *, resp);
8767 			return;
8768 		}
8769 
8770 		/* hold off other access to open_owner while we tinker */
8771 		rfs4_sw_enter(&sp->rs_owner->ro_sw);
8772 
8773 		switch (rc = rfs4_check_stateid_seqid(sp, stateid)) {
8774 		case NFS4_CHECK_STATEID_OLD:
8775 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8776 			goto end;
8777 		case NFS4_CHECK_STATEID_BAD:
8778 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8779 			goto end;
8780 		case NFS4_CHECK_STATEID_EXPIRED:
8781 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8782 			goto end;
8783 		case NFS4_CHECK_STATEID_UNCONFIRMED:
8784 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8785 			goto end;
8786 		case NFS4_CHECK_STATEID_CLOSED:
8787 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8788 			goto end;
8789 		case NFS4_CHECK_STATEID_OKAY:
8790 		case NFS4_CHECK_STATEID_REPLAY:
8791 			switch (rfs4_check_olo_seqid(olo->open_seqid,
8792 			    sp->rs_owner, resop)) {
8793 			case NFS4_CHKSEQ_OKAY:
8794 				if (rc == NFS4_CHECK_STATEID_OKAY)
8795 					break;
8796 				/*
8797 				 * This is replayed stateid; if seqid
8798 				 * matches next expected, then client
8799 				 * is using wrong seqid.
8800 				 */
8801 				/* FALLTHROUGH */
8802 			case NFS4_CHKSEQ_BAD:
8803 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8804 				goto end;
8805 			case NFS4_CHKSEQ_REPLAY:
8806 				/* This is a duplicate LOCK request */
8807 				dup_lock = TRUE;
8808 
8809 				/*
8810 				 * For a duplicate we do not want to
8811 				 * create a new lockowner as it should
8812 				 * already exist.
8813 				 * Turn off the lockowner create flag.
8814 				 */
8815 				lcreate = FALSE;
8816 			}
8817 			break;
8818 		}
8819 
8820 		lo = rfs4_findlockowner(&olo->lock_owner, &lcreate);
8821 		if (lo == NULL) {
8822 			NFS4_DEBUG(rfs4_debug,
8823 			    (CE_NOTE, "rfs4_op_lock: no lock owner"));
8824 			*cs->statusp = resp->status = NFS4ERR_RESOURCE;
8825 			goto end;
8826 		}
8827 
8828 		lsp = rfs4_findlo_state_by_owner(lo, sp, &create);
8829 		if (lsp == NULL) {
8830 			rfs4_update_lease(sp->rs_owner->ro_client);
8831 			/*
8832 			 * Only update theh open_seqid if this is not
8833 			 * a duplicate request
8834 			 */
8835 			if (dup_lock == FALSE) {
8836 				rfs4_update_open_sequence(sp->rs_owner);
8837 			}
8838 
8839 			NFS4_DEBUG(rfs4_debug,
8840 			    (CE_NOTE, "rfs4_op_lock: no state"));
8841 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
8842 			rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8843 			rfs4_lockowner_rele(lo);
8844 			goto end;
8845 		}
8846 
8847 		/*
8848 		 * This is the new_lock_owner branch and the client is
8849 		 * supposed to be associating a new lock_owner with
8850 		 * the open file at this point.  If we find that a
8851 		 * lock_owner/state association already exists and a
8852 		 * successful LOCK request was returned to the client,
8853 		 * an error is returned to the client since this is
8854 		 * not appropriate.  The client should be using the
8855 		 * existing lock_owner branch.
8856 		 */
8857 		if (dup_lock == FALSE && create == FALSE) {
8858 			if (lsp->rls_lock_completed == TRUE) {
8859 				*cs->statusp =
8860 				    resp->status = NFS4ERR_BAD_SEQID;
8861 				rfs4_lockowner_rele(lo);
8862 				goto end;
8863 			}
8864 		}
8865 
8866 		rfs4_update_lease(sp->rs_owner->ro_client);
8867 
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 		/*
8877 		 * If this is a duplicate lock request, just copy the
8878 		 * previously saved reply and return.
8879 		 */
8880 		if (dup_lock == TRUE) {
8881 			/* verify that lock_seqid's match */
8882 			if (lsp->rls_seqid != olo->lock_seqid) {
8883 				NFS4_DEBUG(rfs4_debug,
8884 				    (CE_NOTE, "rfs4_op_lock: Dup-Lock seqid bad"
8885 				    "lsp->seqid=%d old->seqid=%d",
8886 				    lsp->rls_seqid, olo->lock_seqid));
8887 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8888 			} else {
8889 				rfs4_copy_reply(resop, &lsp->rls_reply);
8890 				/*
8891 				 * Make sure to copy the just
8892 				 * retrieved reply status into the
8893 				 * overall compound status
8894 				 */
8895 				*cs->statusp = resp->status;
8896 			}
8897 			rfs4_lockowner_rele(lo);
8898 			goto end;
8899 		}
8900 
8901 		rfs4_dbe_lock(lsp->rls_dbe);
8902 
8903 		/* Make sure to update the lock sequence id */
8904 		lsp->rls_seqid = olo->lock_seqid;
8905 
8906 		NFS4_DEBUG(rfs4_debug,
8907 		    (CE_NOTE, "Lock seqid established as %d", lsp->rls_seqid));
8908 
8909 		/*
8910 		 * This is used to signify the newly created lockowner
8911 		 * stateid and its sequence number.  The checks for
8912 		 * sequence number and increment don't occur on the
8913 		 * very first lock request for a lockowner.
8914 		 */
8915 		lsp->rls_skip_seqid_check = TRUE;
8916 
8917 		/* hold off other access to lsp while we tinker */
8918 		rfs4_sw_enter(&lsp->rls_sw);
8919 		ls_sw_held = TRUE;
8920 
8921 		rfs4_dbe_unlock(lsp->rls_dbe);
8922 
8923 		rfs4_lockowner_rele(lo);
8924 	} else {
8925 		stateid = &args->locker.locker4_u.lock_owner.lock_stateid;
8926 		/* get lsp and hold the lock on the underlying file struct */
8927 		if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE))
8928 		    != NFS4_OK) {
8929 			*cs->statusp = resp->status = status;
8930 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8931 			    cs, LOCK4res *, resp);
8932 			return;
8933 		}
8934 		create = FALSE;	/* We didn't create lsp */
8935 
8936 		/* Ensure specified filehandle matches */
8937 		if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
8938 			rfs4_lo_state_rele(lsp, TRUE);
8939 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8940 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8941 			    cs, LOCK4res *, resp);
8942 			return;
8943 		}
8944 
8945 		/* hold off other access to lsp while we tinker */
8946 		rfs4_sw_enter(&lsp->rls_sw);
8947 		ls_sw_held = TRUE;
8948 
8949 		switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
8950 		/*
8951 		 * The stateid looks like it was okay (expected to be
8952 		 * the next one)
8953 		 */
8954 		case NFS4_CHECK_STATEID_OKAY:
8955 			/*
8956 			 * The sequence id is now checked.  Determine
8957 			 * if this is a replay or if it is in the
8958 			 * expected (next) sequence.  In the case of a
8959 			 * replay, there are two replay conditions
8960 			 * that may occur.  The first is the normal
8961 			 * condition where a LOCK is done with a
8962 			 * NFS4_OK response and the stateid is
8963 			 * updated.  That case is handled below when
8964 			 * the stateid is identified as a REPLAY.  The
8965 			 * second is the case where an error is
8966 			 * returned, like NFS4ERR_DENIED, and the
8967 			 * sequence number is updated but the stateid
8968 			 * is not updated.  This second case is dealt
8969 			 * with here.  So it may seem odd that the
8970 			 * stateid is okay but the sequence id is a
8971 			 * replay but it is okay.
8972 			 */
8973 			switch (rfs4_check_lock_seqid(
8974 			    args->locker.locker4_u.lock_owner.lock_seqid,
8975 			    lsp, resop)) {
8976 			case NFS4_CHKSEQ_REPLAY:
8977 				if (resp->status != NFS4_OK) {
8978 					/*
8979 					 * Here is our replay and need
8980 					 * to verify that the last
8981 					 * response was an error.
8982 					 */
8983 					*cs->statusp = resp->status;
8984 					goto end;
8985 				}
8986 				/*
8987 				 * This is done since the sequence id
8988 				 * looked like a replay but it didn't
8989 				 * pass our check so a BAD_SEQID is
8990 				 * returned as a result.
8991 				 */
8992 				/*FALLTHROUGH*/
8993 			case NFS4_CHKSEQ_BAD:
8994 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8995 				goto end;
8996 			case NFS4_CHKSEQ_OKAY:
8997 				/* Everything looks okay move ahead */
8998 				break;
8999 			}
9000 			break;
9001 		case NFS4_CHECK_STATEID_OLD:
9002 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9003 			goto end;
9004 		case NFS4_CHECK_STATEID_BAD:
9005 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9006 			goto end;
9007 		case NFS4_CHECK_STATEID_EXPIRED:
9008 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9009 			goto end;
9010 		case NFS4_CHECK_STATEID_CLOSED:
9011 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9012 			goto end;
9013 		case NFS4_CHECK_STATEID_REPLAY:
9014 			switch (rfs4_check_lock_seqid(
9015 			    args->locker.locker4_u.lock_owner.lock_seqid,
9016 			    lsp, resop)) {
9017 			case NFS4_CHKSEQ_OKAY:
9018 				/*
9019 				 * This is a replayed stateid; if
9020 				 * seqid matches the next expected,
9021 				 * then client is using wrong seqid.
9022 				 */
9023 			case NFS4_CHKSEQ_BAD:
9024 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9025 				goto end;
9026 			case NFS4_CHKSEQ_REPLAY:
9027 				rfs4_update_lease(lsp->rls_locker->rl_client);
9028 				*cs->statusp = status = resp->status;
9029 				goto end;
9030 			}
9031 			break;
9032 		default:
9033 			ASSERT(FALSE);
9034 			break;
9035 		}
9036 
9037 		rfs4_update_lock_sequence(lsp);
9038 		rfs4_update_lease(lsp->rls_locker->rl_client);
9039 	}
9040 
9041 	/*
9042 	 * NFS4 only allows locking on regular files, so
9043 	 * verify type of object.
9044 	 */
9045 	if (cs->vp->v_type != VREG) {
9046 		if (cs->vp->v_type == VDIR)
9047 			status = NFS4ERR_ISDIR;
9048 		else
9049 			status = NFS4ERR_INVAL;
9050 		goto out;
9051 	}
9052 
9053 	cp = lsp->rls_state->rs_owner->ro_client;
9054 
9055 	if (rfs4_clnt_in_grace(cp) && !args->reclaim) {
9056 		status = NFS4ERR_GRACE;
9057 		goto out;
9058 	}
9059 
9060 	if (rfs4_clnt_in_grace(cp) && args->reclaim && !cp->rc_can_reclaim) {
9061 		status = NFS4ERR_NO_GRACE;
9062 		goto out;
9063 	}
9064 
9065 	if (!rfs4_clnt_in_grace(cp) && args->reclaim) {
9066 		status = NFS4ERR_NO_GRACE;
9067 		goto out;
9068 	}
9069 
9070 	if (lsp->rls_state->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE)
9071 		cs->deleg = TRUE;
9072 
9073 	status = rfs4_do_lock(lsp, args->locktype,
9074 	    args->offset, args->length, cs->cr, resop);
9075 
9076 out:
9077 	lsp->rls_skip_seqid_check = FALSE;
9078 
9079 	*cs->statusp = resp->status = status;
9080 
9081 	if (status == NFS4_OK) {
9082 		resp->LOCK4res_u.lock_stateid = lsp->rls_lockid.stateid;
9083 		lsp->rls_lock_completed = TRUE;
9084 	}
9085 	/*
9086 	 * Only update the "OPEN" response here if this was a new
9087 	 * lock_owner
9088 	 */
9089 	if (sp)
9090 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
9091 
9092 	rfs4_update_lock_resp(lsp, resop);
9093 
9094 end:
9095 	if (lsp) {
9096 		if (ls_sw_held)
9097 			rfs4_sw_exit(&lsp->rls_sw);
9098 		/*
9099 		 * If an sp obtained, then the lsp does not represent
9100 		 * a lock on the file struct.
9101 		 */
9102 		if (sp != NULL)
9103 			rfs4_lo_state_rele(lsp, FALSE);
9104 		else
9105 			rfs4_lo_state_rele(lsp, TRUE);
9106 	}
9107 	if (sp) {
9108 		rfs4_sw_exit(&sp->rs_owner->ro_sw);
9109 		rfs4_state_rele(sp);
9110 	}
9111 
9112 	DTRACE_NFSV4_2(op__lock__done, struct compound_state *, cs,
9113 	    LOCK4res *, resp);
9114 }
9115 
9116 /* free function for LOCK/LOCKT */
9117 static void
9118 lock_denied_free(nfs_resop4 *resop)
9119 {
9120 	LOCK4denied *dp = NULL;
9121 
9122 	switch (resop->resop) {
9123 	case OP_LOCK:
9124 		if (resop->nfs_resop4_u.oplock.status == NFS4ERR_DENIED)
9125 			dp = &resop->nfs_resop4_u.oplock.LOCK4res_u.denied;
9126 		break;
9127 	case OP_LOCKT:
9128 		if (resop->nfs_resop4_u.oplockt.status == NFS4ERR_DENIED)
9129 			dp = &resop->nfs_resop4_u.oplockt.denied;
9130 		break;
9131 	default:
9132 		break;
9133 	}
9134 
9135 	if (dp)
9136 		kmem_free(dp->owner.owner_val, dp->owner.owner_len);
9137 }
9138 
9139 /*ARGSUSED*/
9140 void
9141 rfs4_op_locku(nfs_argop4 *argop, nfs_resop4 *resop,
9142     struct svc_req *req, struct compound_state *cs)
9143 {
9144 	LOCKU4args *args = &argop->nfs_argop4_u.oplocku;
9145 	LOCKU4res *resp = &resop->nfs_resop4_u.oplocku;
9146 	nfsstat4 status;
9147 	stateid4 *stateid = &args->lock_stateid;
9148 	rfs4_lo_state_t *lsp;
9149 
9150 	DTRACE_NFSV4_2(op__locku__start, struct compound_state *, cs,
9151 	    LOCKU4args *, args);
9152 
9153 	if (cs->vp == NULL) {
9154 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9155 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9156 		    LOCKU4res *, resp);
9157 		return;
9158 	}
9159 
9160 	if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE)) != NFS4_OK) {
9161 		*cs->statusp = resp->status = status;
9162 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9163 		    LOCKU4res *, resp);
9164 		return;
9165 	}
9166 
9167 	/* Ensure specified filehandle matches */
9168 	if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9169 		rfs4_lo_state_rele(lsp, TRUE);
9170 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9171 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9172 		    LOCKU4res *, resp);
9173 		return;
9174 	}
9175 
9176 	/* hold off other access to lsp while we tinker */
9177 	rfs4_sw_enter(&lsp->rls_sw);
9178 
9179 	switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9180 	case NFS4_CHECK_STATEID_OKAY:
9181 		if (rfs4_check_lock_seqid(args->seqid, lsp, resop)
9182 		    != NFS4_CHKSEQ_OKAY) {
9183 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9184 			goto end;
9185 		}
9186 		break;
9187 	case NFS4_CHECK_STATEID_OLD:
9188 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9189 		goto end;
9190 	case NFS4_CHECK_STATEID_BAD:
9191 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9192 		goto end;
9193 	case NFS4_CHECK_STATEID_EXPIRED:
9194 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9195 		goto end;
9196 	case NFS4_CHECK_STATEID_CLOSED:
9197 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9198 		goto end;
9199 	case NFS4_CHECK_STATEID_REPLAY:
9200 		switch (rfs4_check_lock_seqid(args->seqid, lsp, resop)) {
9201 		case NFS4_CHKSEQ_OKAY:
9202 				/*
9203 				 * This is a replayed stateid; if
9204 				 * seqid matches the next expected,
9205 				 * then client is using wrong seqid.
9206 				 */
9207 		case NFS4_CHKSEQ_BAD:
9208 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9209 			goto end;
9210 		case NFS4_CHKSEQ_REPLAY:
9211 			rfs4_update_lease(lsp->rls_locker->rl_client);
9212 			*cs->statusp = status = resp->status;
9213 			goto end;
9214 		}
9215 		break;
9216 	default:
9217 		ASSERT(FALSE);
9218 		break;
9219 	}
9220 
9221 	rfs4_update_lock_sequence(lsp);
9222 	rfs4_update_lease(lsp->rls_locker->rl_client);
9223 
9224 	/*
9225 	 * NFS4 only allows locking on regular files, so
9226 	 * verify type of object.
9227 	 */
9228 	if (cs->vp->v_type != VREG) {
9229 		if (cs->vp->v_type == VDIR)
9230 			status = NFS4ERR_ISDIR;
9231 		else
9232 			status = NFS4ERR_INVAL;
9233 		goto out;
9234 	}
9235 
9236 	if (rfs4_clnt_in_grace(lsp->rls_state->rs_owner->ro_client)) {
9237 		status = NFS4ERR_GRACE;
9238 		goto out;
9239 	}
9240 
9241 	status = rfs4_do_lock(lsp, args->locktype,
9242 	    args->offset, args->length, cs->cr, resop);
9243 
9244 out:
9245 	*cs->statusp = resp->status = status;
9246 
9247 	if (status == NFS4_OK)
9248 		resp->lock_stateid = lsp->rls_lockid.stateid;
9249 
9250 	rfs4_update_lock_resp(lsp, resop);
9251 
9252 end:
9253 	rfs4_sw_exit(&lsp->rls_sw);
9254 	rfs4_lo_state_rele(lsp, TRUE);
9255 
9256 	DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9257 	    LOCKU4res *, resp);
9258 }
9259 
9260 /*
9261  * LOCKT is a best effort routine, the client can not be guaranteed that
9262  * the status return is still in effect by the time the reply is received.
9263  * They are numerous race conditions in this routine, but we are not required
9264  * and can not be accurate.
9265  */
9266 /*ARGSUSED*/
9267 void
9268 rfs4_op_lockt(nfs_argop4 *argop, nfs_resop4 *resop,
9269     struct svc_req *req, struct compound_state *cs)
9270 {
9271 	LOCKT4args *args = &argop->nfs_argop4_u.oplockt;
9272 	LOCKT4res *resp = &resop->nfs_resop4_u.oplockt;
9273 	rfs4_lockowner_t *lo;
9274 	rfs4_client_t *cp;
9275 	bool_t create = FALSE;
9276 	struct flock64 flk;
9277 	int error;
9278 	int flag = FREAD | FWRITE;
9279 	int ltype;
9280 	length4 posix_length;
9281 	sysid_t sysid;
9282 	pid_t pid;
9283 
9284 	DTRACE_NFSV4_2(op__lockt__start, struct compound_state *, cs,
9285 	    LOCKT4args *, args);
9286 
9287 	if (cs->vp == NULL) {
9288 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9289 		goto out;
9290 	}
9291 
9292 	/*
9293 	 * NFS4 only allows locking on regular files, so
9294 	 * verify type of object.
9295 	 */
9296 	if (cs->vp->v_type != VREG) {
9297 		if (cs->vp->v_type == VDIR)
9298 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
9299 		else
9300 			*cs->statusp = resp->status =  NFS4ERR_INVAL;
9301 		goto out;
9302 	}
9303 
9304 	/*
9305 	 * Check out the clientid to ensure the server knows about it
9306 	 * so that we correctly inform the client of a server reboot.
9307 	 */
9308 	if ((cp = rfs4_findclient_by_id(args->owner.clientid, FALSE))
9309 	    == NULL) {
9310 		*cs->statusp = resp->status =
9311 		    rfs4_check_clientid(&args->owner.clientid, 0);
9312 		goto out;
9313 	}
9314 	if (rfs4_lease_expired(cp)) {
9315 		rfs4_client_close(cp);
9316 		/*
9317 		 * Protocol doesn't allow returning NFS4ERR_STALE as
9318 		 * other operations do on this check so STALE_CLIENTID
9319 		 * is returned instead
9320 		 */
9321 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
9322 		goto out;
9323 	}
9324 
9325 	if (rfs4_clnt_in_grace(cp) && !(cp->rc_can_reclaim)) {
9326 		*cs->statusp = resp->status = NFS4ERR_GRACE;
9327 		rfs4_client_rele(cp);
9328 		goto out;
9329 	}
9330 	rfs4_client_rele(cp);
9331 
9332 	resp->status = NFS4_OK;
9333 
9334 	switch (args->locktype) {
9335 	case READ_LT:
9336 	case READW_LT:
9337 		ltype = F_RDLCK;
9338 		break;
9339 	case WRITE_LT:
9340 	case WRITEW_LT:
9341 		ltype = F_WRLCK;
9342 		break;
9343 	}
9344 
9345 	posix_length = args->length;
9346 	/* Check for zero length. To lock to end of file use all ones for V4 */
9347 	if (posix_length == 0) {
9348 		*cs->statusp = resp->status = NFS4ERR_INVAL;
9349 		goto out;
9350 	} else if (posix_length == (length4)(~0)) {
9351 		posix_length = 0;	/* Posix to end of file  */
9352 	}
9353 
9354 	/* Find or create a lockowner */
9355 	lo = rfs4_findlockowner(&args->owner, &create);
9356 
9357 	if (lo) {
9358 		pid = lo->rl_pid;
9359 		if ((resp->status =
9360 		    rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
9361 			goto err;
9362 	} else {
9363 		pid = 0;
9364 		sysid = lockt_sysid;
9365 	}
9366 retry:
9367 	flk.l_type = ltype;
9368 	flk.l_whence = 0;		/* SEEK_SET */
9369 	flk.l_start = args->offset;
9370 	flk.l_len = posix_length;
9371 	flk.l_sysid = sysid;
9372 	flk.l_pid = pid;
9373 	flag |= F_REMOTELOCK;
9374 
9375 	LOCK_PRINT(rfs4_debug, "rfs4_op_lockt", F_GETLK, &flk);
9376 
9377 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
9378 	if (flk.l_len < 0 || flk.l_start < 0) {
9379 		resp->status = NFS4ERR_INVAL;
9380 		goto err;
9381 	}
9382 	error = VOP_FRLOCK(cs->vp, F_GETLK, &flk, flag, (u_offset_t)0,
9383 	    NULL, cs->cr, NULL);
9384 
9385 	/*
9386 	 * N.B. We map error values to nfsv4 errors. This is differrent
9387 	 * than puterrno4 routine.
9388 	 */
9389 	switch (error) {
9390 	case 0:
9391 		if (flk.l_type == F_UNLCK)
9392 			resp->status = NFS4_OK;
9393 		else {
9394 			if (lock_denied(&resp->denied, &flk) == NFS4ERR_EXPIRED)
9395 				goto retry;
9396 			resp->status = NFS4ERR_DENIED;
9397 		}
9398 		break;
9399 	case EOVERFLOW:
9400 		resp->status = NFS4ERR_INVAL;
9401 		break;
9402 	case EINVAL:
9403 		resp->status = NFS4ERR_NOTSUPP;
9404 		break;
9405 	default:
9406 		cmn_err(CE_WARN, "rfs4_op_lockt: unexpected errno (%d)",
9407 		    error);
9408 		resp->status = NFS4ERR_SERVERFAULT;
9409 		break;
9410 	}
9411 
9412 err:
9413 	if (lo)
9414 		rfs4_lockowner_rele(lo);
9415 	*cs->statusp = resp->status;
9416 out:
9417 	DTRACE_NFSV4_2(op__lockt__done, struct compound_state *, cs,
9418 	    LOCKT4res *, resp);
9419 }
9420 
9421 int
9422 rfs4_share(rfs4_state_t *sp, uint32_t access, uint32_t deny)
9423 {
9424 	int err;
9425 	int cmd;
9426 	vnode_t *vp;
9427 	struct shrlock shr;
9428 	struct shr_locowner shr_loco;
9429 	int fflags = 0;
9430 
9431 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9432 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9433 
9434 	if (sp->rs_closed)
9435 		return (NFS4ERR_OLD_STATEID);
9436 
9437 	vp = sp->rs_finfo->rf_vp;
9438 	ASSERT(vp);
9439 
9440 	shr.s_access = shr.s_deny = 0;
9441 
9442 	if (access & OPEN4_SHARE_ACCESS_READ) {
9443 		fflags |= FREAD;
9444 		shr.s_access |= F_RDACC;
9445 	}
9446 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
9447 		fflags |= FWRITE;
9448 		shr.s_access |= F_WRACC;
9449 	}
9450 	ASSERT(shr.s_access);
9451 
9452 	if (deny & OPEN4_SHARE_DENY_READ)
9453 		shr.s_deny |= F_RDDNY;
9454 	if (deny & OPEN4_SHARE_DENY_WRITE)
9455 		shr.s_deny |= F_WRDNY;
9456 
9457 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9458 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9459 	shr_loco.sl_pid = shr.s_pid;
9460 	shr_loco.sl_id = shr.s_sysid;
9461 	shr.s_owner = (caddr_t)&shr_loco;
9462 	shr.s_own_len = sizeof (shr_loco);
9463 
9464 	cmd = nbl_need_check(vp) ? F_SHARE_NBMAND : F_SHARE;
9465 
9466 	err = VOP_SHRLOCK(vp, cmd, &shr, fflags, CRED(), NULL);
9467 	if (err != 0) {
9468 		if (err == EAGAIN)
9469 			err = NFS4ERR_SHARE_DENIED;
9470 		else
9471 			err = puterrno4(err);
9472 		return (err);
9473 	}
9474 
9475 	sp->rs_share_access |= access;
9476 	sp->rs_share_deny |= deny;
9477 
9478 	return (0);
9479 }
9480 
9481 int
9482 rfs4_unshare(rfs4_state_t *sp)
9483 {
9484 	int err;
9485 	struct shrlock shr;
9486 	struct shr_locowner shr_loco;
9487 
9488 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9489 
9490 	if (sp->rs_closed || sp->rs_share_access == 0)
9491 		return (0);
9492 
9493 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9494 	ASSERT(sp->rs_finfo->rf_vp);
9495 
9496 	shr.s_access = shr.s_deny = 0;
9497 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9498 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9499 	shr_loco.sl_pid = shr.s_pid;
9500 	shr_loco.sl_id = shr.s_sysid;
9501 	shr.s_owner = (caddr_t)&shr_loco;
9502 	shr.s_own_len = sizeof (shr_loco);
9503 
9504 	err = VOP_SHRLOCK(sp->rs_finfo->rf_vp, F_UNSHARE, &shr, 0, CRED(),
9505 	    NULL);
9506 	if (err != 0) {
9507 		err = puterrno4(err);
9508 		return (err);
9509 	}
9510 
9511 	sp->rs_share_access = 0;
9512 	sp->rs_share_deny = 0;
9513 
9514 	return (0);
9515 
9516 }
9517 
9518 static int
9519 rdma_setup_read_data4(READ4args *args, READ4res *rok)
9520 {
9521 	struct clist	*wcl;
9522 	count4		count = rok->data_len;
9523 	int		wlist_len;
9524 
9525 	wcl = args->wlist;
9526 	if (rdma_setup_read_chunks(wcl, count, &wlist_len) == FALSE) {
9527 		return (FALSE);
9528 	}
9529 	wcl = args->wlist;
9530 	rok->wlist_len = wlist_len;
9531 	rok->wlist = wcl;
9532 	return (TRUE);
9533 }
9534 
9535 /* tunable to disable server referrals */
9536 int rfs4_no_referrals = 0;
9537 
9538 /*
9539  * Find an NFS record in reparse point data.
9540  * Returns 0 for success and <0 or an errno value on failure.
9541  */
9542 int
9543 vn_find_nfs_record(vnode_t *vp, nvlist_t **nvlp, char **svcp, char **datap)
9544 {
9545 	int err;
9546 	char *stype, *val;
9547 	nvlist_t *nvl;
9548 	nvpair_t *curr;
9549 
9550 	if ((nvl = reparse_init()) == NULL)
9551 		return (-1);
9552 
9553 	if ((err = reparse_vnode_parse(vp, nvl)) != 0) {
9554 		reparse_free(nvl);
9555 		return (err);
9556 	}
9557 
9558 	curr = NULL;
9559 	while ((curr = nvlist_next_nvpair(nvl, curr)) != NULL) {
9560 		if ((stype = nvpair_name(curr)) == NULL) {
9561 			reparse_free(nvl);
9562 			return (-2);
9563 		}
9564 		if (strncasecmp(stype, "NFS", 3) == 0)
9565 			break;
9566 	}
9567 
9568 	if ((curr == NULL) ||
9569 	    (nvpair_value_string(curr, &val))) {
9570 		reparse_free(nvl);
9571 		return (-3);
9572 	}
9573 	*nvlp = nvl;
9574 	*svcp = stype;
9575 	*datap = val;
9576 	return (0);
9577 }
9578 
9579 int
9580 vn_is_nfs_reparse(vnode_t *vp, cred_t *cr)
9581 {
9582 	nvlist_t *nvl;
9583 	char *s, *d;
9584 
9585 	if (rfs4_no_referrals != 0)
9586 		return (B_FALSE);
9587 
9588 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9589 		return (B_FALSE);
9590 
9591 	if (vn_find_nfs_record(vp, &nvl, &s, &d) != 0)
9592 		return (B_FALSE);
9593 
9594 	reparse_free(nvl);
9595 
9596 	return (B_TRUE);
9597 }
9598 
9599 /*
9600  * There is a user-level copy of this routine in ref_subr.c.
9601  * Changes should be kept in sync.
9602  */
9603 static int
9604 nfs4_create_components(char *path, component4 *comp4)
9605 {
9606 	int slen, plen, ncomp;
9607 	char *ori_path, *nxtc, buf[MAXNAMELEN];
9608 
9609 	if (path == NULL)
9610 		return (0);
9611 
9612 	plen = strlen(path) + 1;	/* include the terminator */
9613 	ori_path = path;
9614 	ncomp = 0;
9615 
9616 	/* count number of components in the path */
9617 	for (nxtc = path; nxtc < ori_path + plen; nxtc++) {
9618 		if (*nxtc == '/' || *nxtc == '\0' || *nxtc == '\n') {
9619 			if ((slen = nxtc - path) == 0) {
9620 				path = nxtc + 1;
9621 				continue;
9622 			}
9623 
9624 			if (comp4 != NULL) {
9625 				bcopy(path, buf, slen);
9626 				buf[slen] = '\0';
9627 				(void) str_to_utf8(buf, &comp4[ncomp]);
9628 			}
9629 
9630 			ncomp++;	/* 1 valid component */
9631 			path = nxtc + 1;
9632 		}
9633 		if (*nxtc == '\0' || *nxtc == '\n')
9634 			break;
9635 	}
9636 
9637 	return (ncomp);
9638 }
9639 
9640 /*
9641  * There is a user-level copy of this routine in ref_subr.c.
9642  * Changes should be kept in sync.
9643  */
9644 static int
9645 make_pathname4(char *path, pathname4 *pathname)
9646 {
9647 	int ncomp;
9648 	component4 *comp4;
9649 
9650 	if (pathname == NULL)
9651 		return (0);
9652 
9653 	if (path == NULL) {
9654 		pathname->pathname4_val = NULL;
9655 		pathname->pathname4_len = 0;
9656 		return (0);
9657 	}
9658 
9659 	/* count number of components to alloc buffer */
9660 	if ((ncomp = nfs4_create_components(path, NULL)) == 0) {
9661 		pathname->pathname4_val = NULL;
9662 		pathname->pathname4_len = 0;
9663 		return (0);
9664 	}
9665 	comp4 = kmem_zalloc(ncomp * sizeof (component4), KM_SLEEP);
9666 
9667 	/* copy components into allocated buffer */
9668 	ncomp = nfs4_create_components(path, comp4);
9669 
9670 	pathname->pathname4_val = comp4;
9671 	pathname->pathname4_len = ncomp;
9672 
9673 	return (ncomp);
9674 }
9675 
9676 #define	xdr_fs_locations4 xdr_fattr4_fs_locations
9677 
9678 fs_locations4 *
9679 fetch_referral(vnode_t *vp, cred_t *cr)
9680 {
9681 	nvlist_t *nvl;
9682 	char *stype, *sdata;
9683 	fs_locations4 *result;
9684 	char buf[1024];
9685 	size_t bufsize;
9686 	XDR xdr;
9687 	int err;
9688 
9689 	/*
9690 	 * Check attrs to ensure it's a reparse point
9691 	 */
9692 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9693 		return (NULL);
9694 
9695 	/*
9696 	 * Look for an NFS record and get the type and data
9697 	 */
9698 	if (vn_find_nfs_record(vp, &nvl, &stype, &sdata) != 0)
9699 		return (NULL);
9700 
9701 	/*
9702 	 * With the type and data, upcall to get the referral
9703 	 */
9704 	bufsize = sizeof (buf);
9705 	bzero(buf, sizeof (buf));
9706 	err = reparse_kderef((const char *)stype, (const char *)sdata,
9707 	    buf, &bufsize);
9708 	reparse_free(nvl);
9709 
9710 	DTRACE_PROBE4(nfs4serv__func__referral__upcall,
9711 	    char *, stype, char *, sdata, char *, buf, int, err);
9712 	if (err) {
9713 		cmn_err(CE_NOTE,
9714 		    "reparsed daemon not running: unable to get referral (%d)",
9715 		    err);
9716 		return (NULL);
9717 	}
9718 
9719 	/*
9720 	 * We get an XDR'ed record back from the kderef call
9721 	 */
9722 	xdrmem_create(&xdr, buf, bufsize, XDR_DECODE);
9723 	result = kmem_alloc(sizeof (fs_locations4), KM_SLEEP);
9724 	err = xdr_fs_locations4(&xdr, result);
9725 	XDR_DESTROY(&xdr);
9726 	if (err != TRUE) {
9727 		DTRACE_PROBE1(nfs4serv__func__referral__upcall__xdrfail,
9728 		    int, err);
9729 		return (NULL);
9730 	}
9731 
9732 	/*
9733 	 * Look at path to recover fs_root, ignoring the leading '/'
9734 	 */
9735 	(void) make_pathname4(vp->v_path, &result->fs_root);
9736 
9737 	return (result);
9738 }
9739 
9740 char *
9741 build_symlink(vnode_t *vp, cred_t *cr, size_t *strsz)
9742 {
9743 	fs_locations4 *fsl;
9744 	fs_location4 *fs;
9745 	char *server, *path, *symbuf;
9746 	static char *prefix = "/net/";
9747 	int i, size, npaths;
9748 	uint_t len;
9749 
9750 	/* Get the referral */
9751 	if ((fsl = fetch_referral(vp, cr)) == NULL)
9752 		return (NULL);
9753 
9754 	/* Deal with only the first location and first server */
9755 	fs = &fsl->locations_val[0];
9756 	server = utf8_to_str(&fs->server_val[0], &len, NULL);
9757 	if (server == NULL) {
9758 		rfs4_free_fs_locations4(fsl);
9759 		kmem_free(fsl, sizeof (fs_locations4));
9760 		return (NULL);
9761 	}
9762 
9763 	/* Figure out size for "/net/" + host + /path/path/path + NULL */
9764 	size = strlen(prefix) + len;
9765 	for (i = 0; i < fs->rootpath.pathname4_len; i++)
9766 		size += fs->rootpath.pathname4_val[i].utf8string_len + 1;
9767 
9768 	/* Allocate the symlink buffer and fill it */
9769 	symbuf = kmem_zalloc(size, KM_SLEEP);
9770 	(void) strcat(symbuf, prefix);
9771 	(void) strcat(symbuf, server);
9772 	kmem_free(server, len);
9773 
9774 	npaths = 0;
9775 	for (i = 0; i < fs->rootpath.pathname4_len; i++) {
9776 		path = utf8_to_str(&fs->rootpath.pathname4_val[i], &len, NULL);
9777 		if (path == NULL)
9778 			continue;
9779 		(void) strcat(symbuf, "/");
9780 		(void) strcat(symbuf, path);
9781 		npaths++;
9782 		kmem_free(path, len);
9783 	}
9784 
9785 	rfs4_free_fs_locations4(fsl);
9786 	kmem_free(fsl, sizeof (fs_locations4));
9787 
9788 	if (strsz != NULL)
9789 		*strsz = size;
9790 	return (symbuf);
9791 }
9792 
9793 /*
9794  * Check to see if we have a downrev Solaris client, so that we
9795  * can send it a symlink instead of a referral.
9796  */
9797 int
9798 client_is_downrev(struct svc_req *req)
9799 {
9800 	struct sockaddr *ca;
9801 	rfs4_clntip_t *ci;
9802 	bool_t create = FALSE;
9803 	int is_downrev;
9804 
9805 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
9806 	ASSERT(ca);
9807 	ci = rfs4_find_clntip(ca, &create);
9808 	if (ci == NULL)
9809 		return (0);
9810 	is_downrev = ci->ri_no_referrals;
9811 	rfs4_dbe_rele(ci->ri_dbe);
9812 	return (is_downrev);
9813 }
9814