xref: /titanic_41/usr/src/uts/common/fs/nfs/nfs4_srv.c (revision d583b39bfb4e2571d3e41097c5c357ffe353ad45)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 /*
25  * Copyright 2012 Nexenta Systems, Inc. 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, *tvp;
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 		tvp = vp;
955 		if ((error = traverse(&tvp)) != 0) {
956 			VN_RELE(vp);
957 			return (puterrno4(error));
958 		}
959 		/* remember that we had to traverse mountpoint */
960 		did_traverse = TRUE;
961 		vp = tvp;
962 		different_export = 1;
963 	} else if (vp->v_vfsp != dvp->v_vfsp) {
964 		/*
965 		 * If vp isn't a mountpoint and the vfs ptrs aren't the same,
966 		 * then vp is probably an LOFS object.  We don't need the
967 		 * realvp, we just need to know that we might have crossed
968 		 * a server fs boundary and need to call checkexport4.
969 		 * (LOFS lookup hides server fs mountpoints, and actually calls
970 		 * traverse)
971 		 */
972 		different_export = 1;
973 	}
974 
975 	/*
976 	 * Get the export information for it.
977 	 */
978 	if (different_export) {
979 
980 		bzero(&fid, sizeof (fid));
981 		fid.fid_len = MAXFIDSZ;
982 		error = vop_fid_pseudo(vp, &fid);
983 		if (error) {
984 			VN_RELE(vp);
985 			return (puterrno4(error));
986 		}
987 
988 		if (dotdot)
989 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
990 		else
991 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
992 
993 		if (exi == NULL) {
994 			if (did_traverse == TRUE) {
995 				/*
996 				 * If this vnode is a mounted-on vnode,
997 				 * but the mounted-on file system is not
998 				 * exported, send back the secinfo for
999 				 * the exported node that the mounted-on
1000 				 * vnode lives in.
1001 				 */
1002 				exi = cs->exi;
1003 			} else {
1004 				VN_RELE(vp);
1005 				return (puterrno4(EACCES));
1006 			}
1007 		}
1008 	} else {
1009 		exi = cs->exi;
1010 	}
1011 	ASSERT(exi != NULL);
1012 
1013 
1014 	/*
1015 	 * Create the secinfo result based on the security information
1016 	 * from the exportinfo structure (exi).
1017 	 *
1018 	 * Return all flavors for a pseudo node.
1019 	 * For a real export node, return the flavor that the client
1020 	 * has access with.
1021 	 */
1022 	ASSERT(RW_LOCK_HELD(&exported_lock));
1023 	if (PSEUDO(exi)) {
1024 		count = exi->exi_export.ex_seccnt; /* total sec count */
1025 		resok_val = kmem_alloc(count * sizeof (secinfo4), KM_SLEEP);
1026 		secp = exi->exi_export.ex_secinfo;
1027 
1028 		for (i = 0; i < count; i++) {
1029 			si = &secp[i].s_secinfo;
1030 			resok_val[i].flavor = si->sc_rpcnum;
1031 			if (resok_val[i].flavor == RPCSEC_GSS) {
1032 				rpcsec_gss_info *info;
1033 
1034 				info = &resok_val[i].flavor_info;
1035 				info->qop = si->sc_qop;
1036 				info->service = (rpc_gss_svc_t)si->sc_service;
1037 
1038 				/* get oid opaque data */
1039 				info->oid.sec_oid4_len =
1040 				    si->sc_gss_mech_type->length;
1041 				info->oid.sec_oid4_val = kmem_alloc(
1042 				    si->sc_gss_mech_type->length, KM_SLEEP);
1043 				bcopy(
1044 				    si->sc_gss_mech_type->elements,
1045 				    info->oid.sec_oid4_val,
1046 				    info->oid.sec_oid4_len);
1047 			}
1048 		}
1049 		resp->SECINFO4resok_len = count;
1050 		resp->SECINFO4resok_val = resok_val;
1051 	} else {
1052 		int ret_cnt = 0, k = 0;
1053 		int *flavor_list;
1054 
1055 		count = exi->exi_export.ex_seccnt; /* total sec count */
1056 		secp = exi->exi_export.ex_secinfo;
1057 
1058 		flavor_list = kmem_alloc(count * sizeof (int), KM_SLEEP);
1059 		/* find out which flavors to return */
1060 		for (i = 0; i < count; i ++) {
1061 			int access, flavor, perm;
1062 
1063 			flavor = secp[i].s_secinfo.sc_nfsnum;
1064 			perm = secp[i].s_flags;
1065 
1066 			access = nfsauth4_secinfo_access(exi, cs->req,
1067 			    flavor, perm);
1068 
1069 			if (! (access & NFSAUTH_DENIED) &&
1070 			    ! (access & NFSAUTH_WRONGSEC)) {
1071 				flavor_list[ret_cnt] = flavor;
1072 				ret_cnt++;
1073 			}
1074 		}
1075 
1076 		/* Create the returning SECINFO value */
1077 		resok_val = kmem_alloc(ret_cnt * sizeof (secinfo4), KM_SLEEP);
1078 
1079 		for (i = 0; i < count; i++) {
1080 			/*
1081 			 * If the flavor is in the flavor list,
1082 			 * fill in resok_val.
1083 			 */
1084 			si = &secp[i].s_secinfo;
1085 			if (in_flavor_list(si->sc_nfsnum,
1086 			    flavor_list, ret_cnt)) {
1087 				resok_val[k].flavor = si->sc_rpcnum;
1088 				if (resok_val[k].flavor == RPCSEC_GSS) {
1089 					rpcsec_gss_info *info;
1090 
1091 					info = &resok_val[k].flavor_info;
1092 					info->qop = si->sc_qop;
1093 					info->service = (rpc_gss_svc_t)
1094 					    si->sc_service;
1095 
1096 					/* get oid opaque data */
1097 					info->oid.sec_oid4_len =
1098 					    si->sc_gss_mech_type->length;
1099 					info->oid.sec_oid4_val = kmem_alloc(
1100 					    si->sc_gss_mech_type->length,
1101 					    KM_SLEEP);
1102 					bcopy(si->sc_gss_mech_type->elements,
1103 					    info->oid.sec_oid4_val,
1104 					    info->oid.sec_oid4_len);
1105 				}
1106 				k++;
1107 			}
1108 			if (k >= ret_cnt)
1109 				break;
1110 		}
1111 		resp->SECINFO4resok_len = ret_cnt;
1112 		resp->SECINFO4resok_val = resok_val;
1113 		kmem_free(flavor_list, count * sizeof (int));
1114 	}
1115 
1116 	VN_RELE(vp);
1117 	return (NFS4_OK);
1118 }
1119 
1120 /*
1121  * SECINFO (Operation 33): Obtain required security information on
1122  * the component name in the format of (security-mechanism-oid, qop, service)
1123  * triplets.
1124  */
1125 /* ARGSUSED */
1126 static void
1127 rfs4_op_secinfo(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1128     struct compound_state *cs)
1129 {
1130 	SECINFO4args *args = &argop->nfs_argop4_u.opsecinfo;
1131 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1132 	utf8string *utfnm = &args->name;
1133 	uint_t len;
1134 	char *nm;
1135 	struct sockaddr *ca;
1136 	char *name = NULL;
1137 	nfsstat4 status = NFS4_OK;
1138 
1139 	DTRACE_NFSV4_2(op__secinfo__start, struct compound_state *, cs,
1140 	    SECINFO4args *, args);
1141 
1142 	/*
1143 	 * Current file handle (cfh) should have been set before getting
1144 	 * into this function. If not, return error.
1145 	 */
1146 	if (cs->vp == NULL) {
1147 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1148 		goto out;
1149 	}
1150 
1151 	if (cs->vp->v_type != VDIR) {
1152 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1153 		goto out;
1154 	}
1155 
1156 	/*
1157 	 * Verify the component name. If failed, error out, but
1158 	 * do not error out if the component name is a "..".
1159 	 * SECINFO will return its parents secinfo data for SECINFO "..".
1160 	 */
1161 	status = utf8_dir_verify(utfnm);
1162 	if (status != NFS4_OK) {
1163 		if (utfnm->utf8string_len != 2 ||
1164 		    utfnm->utf8string_val[0] != '.' ||
1165 		    utfnm->utf8string_val[1] != '.') {
1166 			*cs->statusp = resp->status = status;
1167 			goto out;
1168 		}
1169 	}
1170 
1171 	nm = utf8_to_str(utfnm, &len, NULL);
1172 	if (nm == NULL) {
1173 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1174 		goto out;
1175 	}
1176 
1177 	if (len > MAXNAMELEN) {
1178 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1179 		kmem_free(nm, len);
1180 		goto out;
1181 	}
1182 
1183 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1184 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1185 	    MAXPATHLEN  + 1);
1186 
1187 	if (name == NULL) {
1188 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1189 		kmem_free(nm, len);
1190 		goto out;
1191 	}
1192 
1193 
1194 	*cs->statusp = resp->status = do_rfs4_op_secinfo(cs, name, resp);
1195 
1196 	if (name != nm)
1197 		kmem_free(name, MAXPATHLEN + 1);
1198 	kmem_free(nm, len);
1199 
1200 out:
1201 	DTRACE_NFSV4_2(op__secinfo__done, struct compound_state *, cs,
1202 	    SECINFO4res *, resp);
1203 }
1204 
1205 /*
1206  * Free SECINFO result.
1207  */
1208 /* ARGSUSED */
1209 static void
1210 rfs4_op_secinfo_free(nfs_resop4 *resop)
1211 {
1212 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1213 	int count, i;
1214 	secinfo4 *resok_val;
1215 
1216 	/* If this is not an Ok result, nothing to free. */
1217 	if (resp->status != NFS4_OK) {
1218 		return;
1219 	}
1220 
1221 	count = resp->SECINFO4resok_len;
1222 	resok_val = resp->SECINFO4resok_val;
1223 
1224 	for (i = 0; i < count; i++) {
1225 		if (resok_val[i].flavor == RPCSEC_GSS) {
1226 			rpcsec_gss_info *info;
1227 
1228 			info = &resok_val[i].flavor_info;
1229 			kmem_free(info->oid.sec_oid4_val,
1230 			    info->oid.sec_oid4_len);
1231 		}
1232 	}
1233 	kmem_free(resok_val, count * sizeof (secinfo4));
1234 	resp->SECINFO4resok_len = 0;
1235 	resp->SECINFO4resok_val = NULL;
1236 }
1237 
1238 /* ARGSUSED */
1239 static void
1240 rfs4_op_access(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1241     struct compound_state *cs)
1242 {
1243 	ACCESS4args *args = &argop->nfs_argop4_u.opaccess;
1244 	ACCESS4res *resp = &resop->nfs_resop4_u.opaccess;
1245 	int error;
1246 	vnode_t *vp;
1247 	struct vattr va;
1248 	int checkwriteperm;
1249 	cred_t *cr = cs->cr;
1250 	bslabel_t *clabel, *slabel;
1251 	ts_label_t *tslabel;
1252 	boolean_t admin_low_client;
1253 
1254 	DTRACE_NFSV4_2(op__access__start, struct compound_state *, cs,
1255 	    ACCESS4args *, args);
1256 
1257 #if 0	/* XXX allow access even if !cs->access. Eventually only pseudo fs */
1258 	if (cs->access == CS_ACCESS_DENIED) {
1259 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1260 		goto out;
1261 	}
1262 #endif
1263 	if (cs->vp == NULL) {
1264 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1265 		goto out;
1266 	}
1267 
1268 	ASSERT(cr != NULL);
1269 
1270 	vp = cs->vp;
1271 
1272 	/*
1273 	 * If the file system is exported read only, it is not appropriate
1274 	 * to check write permissions for regular files and directories.
1275 	 * Special files are interpreted by the client, so the underlying
1276 	 * permissions are sent back to the client for interpretation.
1277 	 */
1278 	if (rdonly4(cs->exi, cs->vp, req) &&
1279 	    (vp->v_type == VREG || vp->v_type == VDIR))
1280 		checkwriteperm = 0;
1281 	else
1282 		checkwriteperm = 1;
1283 
1284 	/*
1285 	 * XXX
1286 	 * We need the mode so that we can correctly determine access
1287 	 * permissions relative to a mandatory lock file.  Access to
1288 	 * mandatory lock files is denied on the server, so it might
1289 	 * as well be reflected to the server during the open.
1290 	 */
1291 	va.va_mask = AT_MODE;
1292 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1293 	if (error) {
1294 		*cs->statusp = resp->status = puterrno4(error);
1295 		goto out;
1296 	}
1297 	resp->access = 0;
1298 	resp->supported = 0;
1299 
1300 	if (is_system_labeled()) {
1301 		ASSERT(req->rq_label != NULL);
1302 		clabel = req->rq_label;
1303 		DTRACE_PROBE2(tx__rfs4__log__info__opaccess__clabel, char *,
1304 		    "got client label from request(1)",
1305 		    struct svc_req *, req);
1306 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
1307 			if ((tslabel = nfs_getflabel(vp, cs->exi)) == NULL) {
1308 				*cs->statusp = resp->status = puterrno4(EACCES);
1309 				goto out;
1310 			}
1311 			slabel = label2bslabel(tslabel);
1312 			DTRACE_PROBE3(tx__rfs4__log__info__opaccess__slabel,
1313 			    char *, "got server label(1) for vp(2)",
1314 			    bslabel_t *, slabel, vnode_t *, vp);
1315 
1316 			admin_low_client = B_FALSE;
1317 		} else
1318 			admin_low_client = B_TRUE;
1319 	}
1320 
1321 	if (args->access & ACCESS4_READ) {
1322 		error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
1323 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1324 		    (!is_system_labeled() || admin_low_client ||
1325 		    bldominates(clabel, slabel)))
1326 			resp->access |= ACCESS4_READ;
1327 		resp->supported |= ACCESS4_READ;
1328 	}
1329 	if ((args->access & ACCESS4_LOOKUP) && vp->v_type == VDIR) {
1330 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1331 		if (!error && (!is_system_labeled() || admin_low_client ||
1332 		    bldominates(clabel, slabel)))
1333 			resp->access |= ACCESS4_LOOKUP;
1334 		resp->supported |= ACCESS4_LOOKUP;
1335 	}
1336 	if (checkwriteperm &&
1337 	    (args->access & (ACCESS4_MODIFY|ACCESS4_EXTEND))) {
1338 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1339 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1340 		    (!is_system_labeled() || admin_low_client ||
1341 		    blequal(clabel, slabel)))
1342 			resp->access |=
1343 			    (args->access & (ACCESS4_MODIFY | ACCESS4_EXTEND));
1344 		resp->supported |=
1345 		    resp->access & (ACCESS4_MODIFY | ACCESS4_EXTEND);
1346 	}
1347 
1348 	if (checkwriteperm &&
1349 	    (args->access & ACCESS4_DELETE) && vp->v_type == VDIR) {
1350 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1351 		if (!error && (!is_system_labeled() || admin_low_client ||
1352 		    blequal(clabel, slabel)))
1353 			resp->access |= ACCESS4_DELETE;
1354 		resp->supported |= ACCESS4_DELETE;
1355 	}
1356 	if (args->access & ACCESS4_EXECUTE && vp->v_type != VDIR) {
1357 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1358 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1359 		    (!is_system_labeled() || admin_low_client ||
1360 		    bldominates(clabel, slabel)))
1361 			resp->access |= ACCESS4_EXECUTE;
1362 		resp->supported |= ACCESS4_EXECUTE;
1363 	}
1364 
1365 	if (is_system_labeled() && !admin_low_client)
1366 		label_rele(tslabel);
1367 
1368 	*cs->statusp = resp->status = NFS4_OK;
1369 out:
1370 	DTRACE_NFSV4_2(op__access__done, struct compound_state *, cs,
1371 	    ACCESS4res *, resp);
1372 }
1373 
1374 /* ARGSUSED */
1375 static void
1376 rfs4_op_commit(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1377     struct compound_state *cs)
1378 {
1379 	COMMIT4args *args = &argop->nfs_argop4_u.opcommit;
1380 	COMMIT4res *resp = &resop->nfs_resop4_u.opcommit;
1381 	int error;
1382 	vnode_t *vp = cs->vp;
1383 	cred_t *cr = cs->cr;
1384 	vattr_t va;
1385 
1386 	DTRACE_NFSV4_2(op__commit__start, struct compound_state *, cs,
1387 	    COMMIT4args *, args);
1388 
1389 	if (vp == NULL) {
1390 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1391 		goto out;
1392 	}
1393 	if (cs->access == CS_ACCESS_DENIED) {
1394 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1395 		goto out;
1396 	}
1397 
1398 	if (args->offset + args->count < args->offset) {
1399 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1400 		goto out;
1401 	}
1402 
1403 	va.va_mask = AT_UID;
1404 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1405 
1406 	/*
1407 	 * If we can't get the attributes, then we can't do the
1408 	 * right access checking.  So, we'll fail the request.
1409 	 */
1410 	if (error) {
1411 		*cs->statusp = resp->status = puterrno4(error);
1412 		goto out;
1413 	}
1414 	if (rdonly4(cs->exi, cs->vp, req)) {
1415 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1416 		goto out;
1417 	}
1418 
1419 	if (vp->v_type != VREG) {
1420 		if (vp->v_type == VDIR)
1421 			resp->status = NFS4ERR_ISDIR;
1422 		else
1423 			resp->status = NFS4ERR_INVAL;
1424 		*cs->statusp = resp->status;
1425 		goto out;
1426 	}
1427 
1428 	if (crgetuid(cr) != va.va_uid &&
1429 	    (error = VOP_ACCESS(vp, VWRITE, 0, cs->cr, NULL))) {
1430 		*cs->statusp = resp->status = puterrno4(error);
1431 		goto out;
1432 	}
1433 
1434 	error = VOP_FSYNC(vp, FSYNC, cr, NULL);
1435 
1436 	if (error) {
1437 		*cs->statusp = resp->status = puterrno4(error);
1438 		goto out;
1439 	}
1440 
1441 	*cs->statusp = resp->status = NFS4_OK;
1442 	resp->writeverf = Write4verf;
1443 out:
1444 	DTRACE_NFSV4_2(op__commit__done, struct compound_state *, cs,
1445 	    COMMIT4res *, resp);
1446 }
1447 
1448 /*
1449  * rfs4_op_mknod is called from rfs4_op_create after all initial verification
1450  * was completed. It does the nfsv4 create for special files.
1451  */
1452 /* ARGSUSED */
1453 static vnode_t *
1454 do_rfs4_op_mknod(CREATE4args *args, CREATE4res *resp, struct svc_req *req,
1455     struct compound_state *cs, vattr_t *vap, char *nm)
1456 {
1457 	int error;
1458 	cred_t *cr = cs->cr;
1459 	vnode_t *dvp = cs->vp;
1460 	vnode_t *vp = NULL;
1461 	int mode;
1462 	enum vcexcl excl;
1463 
1464 	switch (args->type) {
1465 	case NF4CHR:
1466 	case NF4BLK:
1467 		if (secpolicy_sys_devices(cr) != 0) {
1468 			*cs->statusp = resp->status = NFS4ERR_PERM;
1469 			return (NULL);
1470 		}
1471 		if (args->type == NF4CHR)
1472 			vap->va_type = VCHR;
1473 		else
1474 			vap->va_type = VBLK;
1475 		vap->va_rdev = makedevice(args->ftype4_u.devdata.specdata1,
1476 		    args->ftype4_u.devdata.specdata2);
1477 		vap->va_mask |= AT_RDEV;
1478 		break;
1479 	case NF4SOCK:
1480 		vap->va_type = VSOCK;
1481 		break;
1482 	case NF4FIFO:
1483 		vap->va_type = VFIFO;
1484 		break;
1485 	default:
1486 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1487 		return (NULL);
1488 	}
1489 
1490 	/*
1491 	 * Must specify the mode.
1492 	 */
1493 	if (!(vap->va_mask & AT_MODE)) {
1494 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1495 		return (NULL);
1496 	}
1497 
1498 	excl = EXCL;
1499 
1500 	mode = 0;
1501 
1502 	error = VOP_CREATE(dvp, nm, vap, excl, mode, &vp, cr, 0, NULL, NULL);
1503 	if (error) {
1504 		*cs->statusp = resp->status = puterrno4(error);
1505 		return (NULL);
1506 	}
1507 	return (vp);
1508 }
1509 
1510 /*
1511  * nfsv4 create is used to create non-regular files. For regular files,
1512  * use nfsv4 open.
1513  */
1514 /* ARGSUSED */
1515 static void
1516 rfs4_op_create(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1517     struct compound_state *cs)
1518 {
1519 	CREATE4args *args = &argop->nfs_argop4_u.opcreate;
1520 	CREATE4res *resp = &resop->nfs_resop4_u.opcreate;
1521 	int error;
1522 	struct vattr bva, iva, iva2, ava, *vap;
1523 	cred_t *cr = cs->cr;
1524 	vnode_t *dvp = cs->vp;
1525 	vnode_t *vp = NULL;
1526 	vnode_t *realvp;
1527 	char *nm, *lnm;
1528 	uint_t len, llen;
1529 	int syncval = 0;
1530 	struct nfs4_svgetit_arg sarg;
1531 	struct nfs4_ntov_table ntov;
1532 	struct statvfs64 sb;
1533 	nfsstat4 status;
1534 	struct sockaddr *ca;
1535 	char *name = NULL;
1536 	char *lname = NULL;
1537 
1538 	DTRACE_NFSV4_2(op__create__start, struct compound_state *, cs,
1539 	    CREATE4args *, args);
1540 
1541 	resp->attrset = 0;
1542 
1543 	if (dvp == NULL) {
1544 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1545 		goto out;
1546 	}
1547 
1548 	/*
1549 	 * If there is an unshared filesystem mounted on this vnode,
1550 	 * do not allow to create an object in this directory.
1551 	 */
1552 	if (vn_ismntpt(dvp)) {
1553 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1554 		goto out;
1555 	}
1556 
1557 	/* Verify that type is correct */
1558 	switch (args->type) {
1559 	case NF4LNK:
1560 	case NF4BLK:
1561 	case NF4CHR:
1562 	case NF4SOCK:
1563 	case NF4FIFO:
1564 	case NF4DIR:
1565 		break;
1566 	default:
1567 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1568 		goto out;
1569 	};
1570 
1571 	if (cs->access == CS_ACCESS_DENIED) {
1572 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1573 		goto out;
1574 	}
1575 	if (dvp->v_type != VDIR) {
1576 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1577 		goto out;
1578 	}
1579 	status = utf8_dir_verify(&args->objname);
1580 	if (status != NFS4_OK) {
1581 		*cs->statusp = resp->status = status;
1582 		goto out;
1583 	}
1584 
1585 	if (rdonly4(cs->exi, cs->vp, req)) {
1586 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1587 		goto out;
1588 	}
1589 
1590 	/*
1591 	 * Name of newly created object
1592 	 */
1593 	nm = utf8_to_fn(&args->objname, &len, NULL);
1594 	if (nm == NULL) {
1595 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1596 		goto out;
1597 	}
1598 
1599 	if (len > MAXNAMELEN) {
1600 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1601 		kmem_free(nm, len);
1602 		goto out;
1603 	}
1604 
1605 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1606 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1607 	    MAXPATHLEN  + 1);
1608 
1609 	if (name == NULL) {
1610 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1611 		kmem_free(nm, len);
1612 		goto out;
1613 	}
1614 
1615 	resp->attrset = 0;
1616 
1617 	sarg.sbp = &sb;
1618 	sarg.is_referral = B_FALSE;
1619 	nfs4_ntov_table_init(&ntov);
1620 
1621 	status = do_rfs4_set_attrs(&resp->attrset,
1622 	    &args->createattrs, cs, &sarg, &ntov, NFS4ATTR_SETIT);
1623 
1624 	if (sarg.vap->va_mask == 0 && status == NFS4_OK)
1625 		status = NFS4ERR_INVAL;
1626 
1627 	if (status != NFS4_OK) {
1628 		*cs->statusp = resp->status = status;
1629 		if (name != nm)
1630 			kmem_free(name, MAXPATHLEN + 1);
1631 		kmem_free(nm, len);
1632 		nfs4_ntov_table_free(&ntov, &sarg);
1633 		resp->attrset = 0;
1634 		goto out;
1635 	}
1636 
1637 	/* Get "before" change value */
1638 	bva.va_mask = AT_CTIME|AT_SEQ|AT_MODE;
1639 	error = VOP_GETATTR(dvp, &bva, 0, cr, NULL);
1640 	if (error) {
1641 		*cs->statusp = resp->status = puterrno4(error);
1642 		if (name != nm)
1643 			kmem_free(name, MAXPATHLEN + 1);
1644 		kmem_free(nm, len);
1645 		nfs4_ntov_table_free(&ntov, &sarg);
1646 		resp->attrset = 0;
1647 		goto out;
1648 	}
1649 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bva.va_ctime)
1650 
1651 	vap = sarg.vap;
1652 
1653 	/*
1654 	 * Set the default initial values for attributes when the parent
1655 	 * directory does not have the VSUID/VSGID bit set and they have
1656 	 * not been specified in createattrs.
1657 	 */
1658 	if (!(bva.va_mode & VSUID) && (vap->va_mask & AT_UID) == 0) {
1659 		vap->va_uid = crgetuid(cr);
1660 		vap->va_mask |= AT_UID;
1661 	}
1662 	if (!(bva.va_mode & VSGID) && (vap->va_mask & AT_GID) == 0) {
1663 		vap->va_gid = crgetgid(cr);
1664 		vap->va_mask |= AT_GID;
1665 	}
1666 
1667 	vap->va_mask |= AT_TYPE;
1668 	switch (args->type) {
1669 	case NF4DIR:
1670 		vap->va_type = VDIR;
1671 		if ((vap->va_mask & AT_MODE) == 0) {
1672 			vap->va_mode = 0700;	/* default: owner rwx only */
1673 			vap->va_mask |= AT_MODE;
1674 		}
1675 		error = VOP_MKDIR(dvp, name, vap, &vp, cr, NULL, 0, NULL);
1676 		if (error)
1677 			break;
1678 
1679 		/*
1680 		 * Get the initial "after" sequence number, if it fails,
1681 		 * set to zero
1682 		 */
1683 		iva.va_mask = AT_SEQ;
1684 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1685 			iva.va_seq = 0;
1686 		break;
1687 	case NF4LNK:
1688 		vap->va_type = VLNK;
1689 		if ((vap->va_mask & AT_MODE) == 0) {
1690 			vap->va_mode = 0700;	/* default: owner rwx only */
1691 			vap->va_mask |= AT_MODE;
1692 		}
1693 
1694 		/*
1695 		 * symlink names must be treated as data
1696 		 */
1697 		lnm = utf8_to_str(&args->ftype4_u.linkdata, &llen, NULL);
1698 
1699 		if (lnm == NULL) {
1700 			*cs->statusp = resp->status = NFS4ERR_INVAL;
1701 			if (name != nm)
1702 				kmem_free(name, MAXPATHLEN + 1);
1703 			kmem_free(nm, len);
1704 			nfs4_ntov_table_free(&ntov, &sarg);
1705 			resp->attrset = 0;
1706 			goto out;
1707 		}
1708 
1709 		if (llen > MAXPATHLEN) {
1710 			*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1711 			if (name != nm)
1712 				kmem_free(name, MAXPATHLEN + 1);
1713 			kmem_free(nm, len);
1714 			kmem_free(lnm, llen);
1715 			nfs4_ntov_table_free(&ntov, &sarg);
1716 			resp->attrset = 0;
1717 			goto out;
1718 		}
1719 
1720 		lname = nfscmd_convname(ca, cs->exi, lnm,
1721 		    NFSCMD_CONV_INBOUND, MAXPATHLEN  + 1);
1722 
1723 		if (lname == NULL) {
1724 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
1725 			if (name != nm)
1726 				kmem_free(name, MAXPATHLEN + 1);
1727 			kmem_free(nm, len);
1728 			kmem_free(lnm, llen);
1729 			nfs4_ntov_table_free(&ntov, &sarg);
1730 			resp->attrset = 0;
1731 			goto out;
1732 		}
1733 
1734 		error = VOP_SYMLINK(dvp, name, vap, lname, cr, NULL, 0);
1735 		if (lname != lnm)
1736 			kmem_free(lname, MAXPATHLEN + 1);
1737 		kmem_free(lnm, llen);
1738 		if (error)
1739 			break;
1740 
1741 		/*
1742 		 * Get the initial "after" sequence number, if it fails,
1743 		 * set to zero
1744 		 */
1745 		iva.va_mask = AT_SEQ;
1746 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1747 			iva.va_seq = 0;
1748 
1749 		error = VOP_LOOKUP(dvp, name, &vp, NULL, 0, NULL, cr,
1750 		    NULL, NULL, NULL);
1751 		if (error)
1752 			break;
1753 
1754 		/*
1755 		 * va_seq is not safe over VOP calls, check it again
1756 		 * if it has changed zero out iva to force atomic = FALSE.
1757 		 */
1758 		iva2.va_mask = AT_SEQ;
1759 		if (VOP_GETATTR(dvp, &iva2, 0, cs->cr, NULL) ||
1760 		    iva2.va_seq != iva.va_seq)
1761 			iva.va_seq = 0;
1762 		break;
1763 	default:
1764 		/*
1765 		 * probably a special file.
1766 		 */
1767 		if ((vap->va_mask & AT_MODE) == 0) {
1768 			vap->va_mode = 0600;	/* default: owner rw only */
1769 			vap->va_mask |= AT_MODE;
1770 		}
1771 		syncval = FNODSYNC;
1772 		/*
1773 		 * We know this will only generate one VOP call
1774 		 */
1775 		vp = do_rfs4_op_mknod(args, resp, req, cs, vap, name);
1776 
1777 		if (vp == NULL) {
1778 			if (name != nm)
1779 				kmem_free(name, MAXPATHLEN + 1);
1780 			kmem_free(nm, len);
1781 			nfs4_ntov_table_free(&ntov, &sarg);
1782 			resp->attrset = 0;
1783 			goto out;
1784 		}
1785 
1786 		/*
1787 		 * Get the initial "after" sequence number, if it fails,
1788 		 * set to zero
1789 		 */
1790 		iva.va_mask = AT_SEQ;
1791 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1792 			iva.va_seq = 0;
1793 
1794 		break;
1795 	}
1796 	if (name != nm)
1797 		kmem_free(name, MAXPATHLEN + 1);
1798 	kmem_free(nm, len);
1799 
1800 	if (error) {
1801 		*cs->statusp = resp->status = puterrno4(error);
1802 	}
1803 
1804 	/*
1805 	 * Force modified data and metadata out to stable storage.
1806 	 */
1807 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
1808 
1809 	if (resp->status != NFS4_OK) {
1810 		if (vp != NULL)
1811 			VN_RELE(vp);
1812 		nfs4_ntov_table_free(&ntov, &sarg);
1813 		resp->attrset = 0;
1814 		goto out;
1815 	}
1816 
1817 	/*
1818 	 * Finish setup of cinfo response, "before" value already set.
1819 	 * Get "after" change value, if it fails, simply return the
1820 	 * before value.
1821 	 */
1822 	ava.va_mask = AT_CTIME|AT_SEQ;
1823 	if (VOP_GETATTR(dvp, &ava, 0, cr, NULL)) {
1824 		ava.va_ctime = bva.va_ctime;
1825 		ava.va_seq = 0;
1826 	}
1827 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, ava.va_ctime);
1828 
1829 	/*
1830 	 * True verification that object was created with correct
1831 	 * attrs is impossible.  The attrs could have been changed
1832 	 * immediately after object creation.  If attributes did
1833 	 * not verify, the only recourse for the server is to
1834 	 * destroy the object.  Maybe if some attrs (like gid)
1835 	 * are set incorrectly, the object should be destroyed;
1836 	 * however, seems bad as a default policy.  Do we really
1837 	 * want to destroy an object over one of the times not
1838 	 * verifying correctly?  For these reasons, the server
1839 	 * currently sets bits in attrset for createattrs
1840 	 * that were set; however, no verification is done.
1841 	 *
1842 	 * vmask_to_nmask accounts for vattr bits set on create
1843 	 *	[do_rfs4_set_attrs() only sets resp bits for
1844 	 *	 non-vattr/vfs bits.]
1845 	 * Mask off any bits set by default so as not to return
1846 	 * more attrset bits than were requested in createattrs
1847 	 */
1848 	nfs4_vmask_to_nmask(sarg.vap->va_mask, &resp->attrset);
1849 	resp->attrset &= args->createattrs.attrmask;
1850 	nfs4_ntov_table_free(&ntov, &sarg);
1851 
1852 	error = makefh4(&cs->fh, vp, cs->exi);
1853 	if (error) {
1854 		*cs->statusp = resp->status = puterrno4(error);
1855 	}
1856 
1857 	/*
1858 	 * The cinfo.atomic = TRUE only if we got no errors, we have
1859 	 * non-zero va_seq's, and it has incremented by exactly one
1860 	 * during the creation and it didn't change during the VOP_LOOKUP
1861 	 * or VOP_FSYNC.
1862 	 */
1863 	if (!error && bva.va_seq && iva.va_seq && ava.va_seq &&
1864 	    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
1865 		resp->cinfo.atomic = TRUE;
1866 	else
1867 		resp->cinfo.atomic = FALSE;
1868 
1869 	/*
1870 	 * Force modified metadata out to stable storage.
1871 	 *
1872 	 * if a underlying vp exists, pass it to VOP_FSYNC
1873 	 */
1874 	if (VOP_REALVP(vp, &realvp, NULL) == 0)
1875 		(void) VOP_FSYNC(realvp, syncval, cr, NULL);
1876 	else
1877 		(void) VOP_FSYNC(vp, syncval, cr, NULL);
1878 
1879 	if (resp->status != NFS4_OK) {
1880 		VN_RELE(vp);
1881 		goto out;
1882 	}
1883 	if (cs->vp)
1884 		VN_RELE(cs->vp);
1885 
1886 	cs->vp = vp;
1887 	*cs->statusp = resp->status = NFS4_OK;
1888 out:
1889 	DTRACE_NFSV4_2(op__create__done, struct compound_state *, cs,
1890 	    CREATE4res *, resp);
1891 }
1892 
1893 /*ARGSUSED*/
1894 static void
1895 rfs4_op_delegpurge(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1896     struct compound_state *cs)
1897 {
1898 	DTRACE_NFSV4_2(op__delegpurge__start, struct compound_state *, cs,
1899 	    DELEGPURGE4args *, &argop->nfs_argop4_u.opdelegpurge);
1900 
1901 	rfs4_op_inval(argop, resop, req, cs);
1902 
1903 	DTRACE_NFSV4_2(op__delegpurge__done, struct compound_state *, cs,
1904 	    DELEGPURGE4res *, &resop->nfs_resop4_u.opdelegpurge);
1905 }
1906 
1907 /*ARGSUSED*/
1908 static void
1909 rfs4_op_delegreturn(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1910     struct compound_state *cs)
1911 {
1912 	DELEGRETURN4args *args = &argop->nfs_argop4_u.opdelegreturn;
1913 	DELEGRETURN4res *resp = &resop->nfs_resop4_u.opdelegreturn;
1914 	rfs4_deleg_state_t *dsp;
1915 	nfsstat4 status;
1916 
1917 	DTRACE_NFSV4_2(op__delegreturn__start, struct compound_state *, cs,
1918 	    DELEGRETURN4args *, args);
1919 
1920 	status = rfs4_get_deleg_state(&args->deleg_stateid, &dsp);
1921 	resp->status = *cs->statusp = status;
1922 	if (status != NFS4_OK)
1923 		goto out;
1924 
1925 	/* Ensure specified filehandle matches */
1926 	if (cs->vp != dsp->rds_finfo->rf_vp) {
1927 		resp->status = *cs->statusp = NFS4ERR_BAD_STATEID;
1928 	} else
1929 		rfs4_return_deleg(dsp, FALSE);
1930 
1931 	rfs4_update_lease(dsp->rds_client);
1932 
1933 	rfs4_deleg_state_rele(dsp);
1934 out:
1935 	DTRACE_NFSV4_2(op__delegreturn__done, struct compound_state *, cs,
1936 	    DELEGRETURN4res *, resp);
1937 }
1938 
1939 /*
1940  * Check to see if a given "flavor" is an explicitly shared flavor.
1941  * The assumption of this routine is the "flavor" is already a valid
1942  * flavor in the secinfo list of "exi".
1943  *
1944  *	e.g.
1945  *		# share -o sec=flavor1 /export
1946  *		# share -o sec=flavor2 /export/home
1947  *
1948  *		flavor2 is not an explicitly shared flavor for /export,
1949  *		however it is in the secinfo list for /export thru the
1950  *		server namespace setup.
1951  */
1952 int
1953 is_exported_sec(int flavor, struct exportinfo *exi)
1954 {
1955 	int	i;
1956 	struct secinfo *sp;
1957 
1958 	sp = exi->exi_export.ex_secinfo;
1959 	for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
1960 		if (flavor == sp[i].s_secinfo.sc_nfsnum ||
1961 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE) {
1962 			return (SEC_REF_EXPORTED(&sp[i]));
1963 		}
1964 	}
1965 
1966 	/* Should not reach this point based on the assumption */
1967 	return (0);
1968 }
1969 
1970 /*
1971  * Check if the security flavor used in the request matches what is
1972  * required at the export point or at the root pseudo node (exi_root).
1973  *
1974  * returns 1 if there's a match or if exported with AUTH_NONE; 0 otherwise.
1975  *
1976  */
1977 static int
1978 secinfo_match_or_authnone(struct compound_state *cs)
1979 {
1980 	int	i;
1981 	struct secinfo *sp;
1982 
1983 	/*
1984 	 * Check cs->nfsflavor (from the request) against
1985 	 * the current export data in cs->exi.
1986 	 */
1987 	sp = cs->exi->exi_export.ex_secinfo;
1988 	for (i = 0; i < cs->exi->exi_export.ex_seccnt; i++) {
1989 		if (cs->nfsflavor == sp[i].s_secinfo.sc_nfsnum ||
1990 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE)
1991 			return (1);
1992 	}
1993 
1994 	return (0);
1995 }
1996 
1997 /*
1998  * Check the access authority for the client and return the correct error.
1999  */
2000 nfsstat4
2001 call_checkauth4(struct compound_state *cs, struct svc_req *req)
2002 {
2003 	int	authres;
2004 
2005 	/*
2006 	 * First, check if the security flavor used in the request
2007 	 * are among the flavors set in the server namespace.
2008 	 */
2009 	if (!secinfo_match_or_authnone(cs)) {
2010 		*cs->statusp = NFS4ERR_WRONGSEC;
2011 		return (*cs->statusp);
2012 	}
2013 
2014 	authres = checkauth4(cs, req);
2015 
2016 	if (authres > 0) {
2017 		*cs->statusp = NFS4_OK;
2018 		if (! (cs->access & CS_ACCESS_LIMITED))
2019 			cs->access = CS_ACCESS_OK;
2020 	} else if (authres == 0) {
2021 		*cs->statusp = NFS4ERR_ACCESS;
2022 	} else if (authres == -2) {
2023 		*cs->statusp = NFS4ERR_WRONGSEC;
2024 	} else {
2025 		*cs->statusp = NFS4ERR_DELAY;
2026 	}
2027 	return (*cs->statusp);
2028 }
2029 
2030 /*
2031  * bitmap4_to_attrmask is called by getattr and readdir.
2032  * It sets up the vattr mask and determines whether vfsstat call is needed
2033  * based on the input bitmap.
2034  * Returns nfsv4 status.
2035  */
2036 static nfsstat4
2037 bitmap4_to_attrmask(bitmap4 breq, struct nfs4_svgetit_arg *sargp)
2038 {
2039 	int i;
2040 	uint_t	va_mask;
2041 	struct statvfs64 *sbp = sargp->sbp;
2042 
2043 	sargp->sbp = NULL;
2044 	sargp->flag = 0;
2045 	sargp->rdattr_error = NFS4_OK;
2046 	sargp->mntdfid_set = FALSE;
2047 	if (sargp->cs->vp)
2048 		sargp->xattr = get_fh4_flag(&sargp->cs->fh,
2049 		    FH4_ATTRDIR | FH4_NAMEDATTR);
2050 	else
2051 		sargp->xattr = 0;
2052 
2053 	/*
2054 	 * Set rdattr_error_req to true if return error per
2055 	 * failed entry rather than fail the readdir.
2056 	 */
2057 	if (breq & FATTR4_RDATTR_ERROR_MASK)
2058 		sargp->rdattr_error_req = 1;
2059 	else
2060 		sargp->rdattr_error_req = 0;
2061 
2062 	/*
2063 	 * generate the va_mask
2064 	 * Handle the easy cases first
2065 	 */
2066 	switch (breq) {
2067 	case NFS4_NTOV_ATTR_MASK:
2068 		sargp->vap->va_mask = NFS4_NTOV_ATTR_AT_MASK;
2069 		return (NFS4_OK);
2070 
2071 	case NFS4_FS_ATTR_MASK:
2072 		sargp->vap->va_mask = NFS4_FS_ATTR_AT_MASK;
2073 		sargp->sbp = sbp;
2074 		return (NFS4_OK);
2075 
2076 	case NFS4_NTOV_ATTR_CACHE_MASK:
2077 		sargp->vap->va_mask = NFS4_NTOV_ATTR_CACHE_AT_MASK;
2078 		return (NFS4_OK);
2079 
2080 	case FATTR4_LEASE_TIME_MASK:
2081 		sargp->vap->va_mask = 0;
2082 		return (NFS4_OK);
2083 
2084 	default:
2085 		va_mask = 0;
2086 		for (i = 0; i < nfs4_ntov_map_size; i++) {
2087 			if ((breq & nfs4_ntov_map[i].fbit) &&
2088 			    nfs4_ntov_map[i].vbit)
2089 				va_mask |= nfs4_ntov_map[i].vbit;
2090 		}
2091 
2092 		/*
2093 		 * Check is vfsstat is needed
2094 		 */
2095 		if (breq & NFS4_FS_ATTR_MASK)
2096 			sargp->sbp = sbp;
2097 
2098 		sargp->vap->va_mask = va_mask;
2099 		return (NFS4_OK);
2100 	}
2101 	/* NOTREACHED */
2102 }
2103 
2104 /*
2105  * bitmap4_get_sysattrs is called by getattr and readdir.
2106  * It calls both VOP_GETATTR and VFS_STATVFS calls to get the attrs.
2107  * Returns nfsv4 status.
2108  */
2109 static nfsstat4
2110 bitmap4_get_sysattrs(struct nfs4_svgetit_arg *sargp)
2111 {
2112 	int error;
2113 	struct compound_state *cs = sargp->cs;
2114 	vnode_t *vp = cs->vp;
2115 
2116 	if (sargp->sbp != NULL) {
2117 		if (error = VFS_STATVFS(vp->v_vfsp, sargp->sbp)) {
2118 			sargp->sbp = NULL;	/* to identify error */
2119 			return (puterrno4(error));
2120 		}
2121 	}
2122 
2123 	return (rfs4_vop_getattr(vp, sargp->vap, 0, cs->cr));
2124 }
2125 
2126 static void
2127 nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp)
2128 {
2129 	ntovp->na = kmem_zalloc(sizeof (union nfs4_attr_u) * nfs4_ntov_map_size,
2130 	    KM_SLEEP);
2131 	ntovp->attrcnt = 0;
2132 	ntovp->vfsstat = FALSE;
2133 }
2134 
2135 static void
2136 nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
2137     struct nfs4_svgetit_arg *sargp)
2138 {
2139 	int i;
2140 	union nfs4_attr_u *na;
2141 	uint8_t *amap;
2142 
2143 	/*
2144 	 * XXX Should do the same checks for whether the bit is set
2145 	 */
2146 	for (i = 0, na = ntovp->na, amap = ntovp->amap;
2147 	    i < ntovp->attrcnt; i++, na++, amap++) {
2148 		(void) (*nfs4_ntov_map[*amap].sv_getit)(
2149 		    NFS4ATTR_FREEIT, sargp, na);
2150 	}
2151 	if ((sargp->op == NFS4ATTR_SETIT) || (sargp->op == NFS4ATTR_VERIT)) {
2152 		/*
2153 		 * xdr_free for getattr will be done later
2154 		 */
2155 		for (i = 0, na = ntovp->na, amap = ntovp->amap;
2156 		    i < ntovp->attrcnt; i++, na++, amap++) {
2157 			xdr_free(nfs4_ntov_map[*amap].xfunc, (caddr_t)na);
2158 		}
2159 	}
2160 	kmem_free(ntovp->na, sizeof (union nfs4_attr_u) * nfs4_ntov_map_size);
2161 }
2162 
2163 /*
2164  * do_rfs4_op_getattr gets the system attrs and converts into fattr4.
2165  */
2166 static nfsstat4
2167 do_rfs4_op_getattr(bitmap4 breq, fattr4 *fattrp,
2168     struct nfs4_svgetit_arg *sargp)
2169 {
2170 	int error = 0;
2171 	int i, k;
2172 	struct nfs4_ntov_table ntov;
2173 	XDR xdr;
2174 	ulong_t xdr_size;
2175 	char *xdr_attrs;
2176 	nfsstat4 status = NFS4_OK;
2177 	nfsstat4 prev_rdattr_error = sargp->rdattr_error;
2178 	union nfs4_attr_u *na;
2179 	uint8_t *amap;
2180 
2181 	sargp->op = NFS4ATTR_GETIT;
2182 	sargp->flag = 0;
2183 
2184 	fattrp->attrmask = 0;
2185 	/* if no bits requested, then return empty fattr4 */
2186 	if (breq == 0) {
2187 		fattrp->attrlist4_len = 0;
2188 		fattrp->attrlist4 = NULL;
2189 		return (NFS4_OK);
2190 	}
2191 
2192 	/*
2193 	 * return NFS4ERR_INVAL when client requests write-only attrs
2194 	 */
2195 	if (breq & (FATTR4_TIME_ACCESS_SET_MASK | FATTR4_TIME_MODIFY_SET_MASK))
2196 		return (NFS4ERR_INVAL);
2197 
2198 	nfs4_ntov_table_init(&ntov);
2199 	na = ntov.na;
2200 	amap = ntov.amap;
2201 
2202 	/*
2203 	 * Now loop to get or verify the attrs
2204 	 */
2205 	for (i = 0; i < nfs4_ntov_map_size; i++) {
2206 		if (breq & nfs4_ntov_map[i].fbit) {
2207 			if ((*nfs4_ntov_map[i].sv_getit)(
2208 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0) {
2209 
2210 				error = (*nfs4_ntov_map[i].sv_getit)(
2211 				    NFS4ATTR_GETIT, sargp, na);
2212 
2213 				/*
2214 				 * Possible error values:
2215 				 * >0 if sv_getit failed to
2216 				 * get the attr; 0 if succeeded;
2217 				 * <0 if rdattr_error and the
2218 				 * attribute cannot be returned.
2219 				 */
2220 				if (error && !(sargp->rdattr_error_req))
2221 					goto done;
2222 				/*
2223 				 * If error then just for entry
2224 				 */
2225 				if (error == 0) {
2226 					fattrp->attrmask |=
2227 					    nfs4_ntov_map[i].fbit;
2228 					*amap++ =
2229 					    (uint8_t)nfs4_ntov_map[i].nval;
2230 					na++;
2231 					(ntov.attrcnt)++;
2232 				} else if ((error > 0) &&
2233 				    (sargp->rdattr_error == NFS4_OK)) {
2234 					sargp->rdattr_error = puterrno4(error);
2235 				}
2236 				error = 0;
2237 			}
2238 		}
2239 	}
2240 
2241 	/*
2242 	 * If rdattr_error was set after the return value for it was assigned,
2243 	 * update it.
2244 	 */
2245 	if (prev_rdattr_error != sargp->rdattr_error) {
2246 		na = ntov.na;
2247 		amap = ntov.amap;
2248 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2249 			k = *amap;
2250 			if (k < FATTR4_RDATTR_ERROR) {
2251 				continue;
2252 			}
2253 			if ((k == FATTR4_RDATTR_ERROR) &&
2254 			    ((*nfs4_ntov_map[k].sv_getit)(
2255 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0)) {
2256 
2257 				(void) (*nfs4_ntov_map[k].sv_getit)(
2258 				    NFS4ATTR_GETIT, sargp, na);
2259 			}
2260 			break;
2261 		}
2262 	}
2263 
2264 	xdr_size = 0;
2265 	na = ntov.na;
2266 	amap = ntov.amap;
2267 	for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2268 		xdr_size += xdr_sizeof(nfs4_ntov_map[*amap].xfunc, na);
2269 	}
2270 
2271 	fattrp->attrlist4_len = xdr_size;
2272 	if (xdr_size) {
2273 		/* freed by rfs4_op_getattr_free() */
2274 		fattrp->attrlist4 = xdr_attrs = kmem_zalloc(xdr_size, KM_SLEEP);
2275 
2276 		xdrmem_create(&xdr, xdr_attrs, xdr_size, XDR_ENCODE);
2277 
2278 		na = ntov.na;
2279 		amap = ntov.amap;
2280 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2281 			if (!(*nfs4_ntov_map[*amap].xfunc)(&xdr, na)) {
2282 				DTRACE_PROBE1(nfss__e__getattr4_encfail,
2283 				    int, *amap);
2284 				status = NFS4ERR_SERVERFAULT;
2285 				break;
2286 			}
2287 		}
2288 		/* xdrmem_destroy(&xdrs); */	/* NO-OP */
2289 	} else {
2290 		fattrp->attrlist4 = NULL;
2291 	}
2292 done:
2293 
2294 	nfs4_ntov_table_free(&ntov, sargp);
2295 
2296 	if (error != 0)
2297 		status = puterrno4(error);
2298 
2299 	return (status);
2300 }
2301 
2302 /* ARGSUSED */
2303 static void
2304 rfs4_op_getattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2305     struct compound_state *cs)
2306 {
2307 	GETATTR4args *args = &argop->nfs_argop4_u.opgetattr;
2308 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2309 	struct nfs4_svgetit_arg sarg;
2310 	struct statvfs64 sb;
2311 	nfsstat4 status;
2312 
2313 	DTRACE_NFSV4_2(op__getattr__start, struct compound_state *, cs,
2314 	    GETATTR4args *, args);
2315 
2316 	if (cs->vp == NULL) {
2317 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2318 		goto out;
2319 	}
2320 
2321 	if (cs->access == CS_ACCESS_DENIED) {
2322 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2323 		goto out;
2324 	}
2325 
2326 	sarg.sbp = &sb;
2327 	sarg.cs = cs;
2328 	sarg.is_referral = B_FALSE;
2329 
2330 	status = bitmap4_to_attrmask(args->attr_request, &sarg);
2331 	if (status == NFS4_OK) {
2332 
2333 		status = bitmap4_get_sysattrs(&sarg);
2334 		if (status == NFS4_OK) {
2335 
2336 			/* Is this a referral? */
2337 			if (vn_is_nfs_reparse(cs->vp, cs->cr)) {
2338 				/* Older V4 Solaris client sees a link */
2339 				if (client_is_downrev(req))
2340 					sarg.vap->va_type = VLNK;
2341 				else
2342 					sarg.is_referral = B_TRUE;
2343 			}
2344 
2345 			status = do_rfs4_op_getattr(args->attr_request,
2346 			    &resp->obj_attributes, &sarg);
2347 		}
2348 	}
2349 	*cs->statusp = resp->status = status;
2350 out:
2351 	DTRACE_NFSV4_2(op__getattr__done, struct compound_state *, cs,
2352 	    GETATTR4res *, resp);
2353 }
2354 
2355 static void
2356 rfs4_op_getattr_free(nfs_resop4 *resop)
2357 {
2358 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2359 
2360 	nfs4_fattr4_free(&resp->obj_attributes);
2361 }
2362 
2363 /* ARGSUSED */
2364 static void
2365 rfs4_op_getfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2366     struct compound_state *cs)
2367 {
2368 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2369 
2370 	DTRACE_NFSV4_1(op__getfh__start, struct compound_state *, cs);
2371 
2372 	if (cs->vp == NULL) {
2373 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2374 		goto out;
2375 	}
2376 	if (cs->access == CS_ACCESS_DENIED) {
2377 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2378 		goto out;
2379 	}
2380 
2381 	/* check for reparse point at the share point */
2382 	if (cs->exi->exi_moved || vn_is_nfs_reparse(cs->exi->exi_vp, cs->cr)) {
2383 		/* it's all bad */
2384 		cs->exi->exi_moved = 1;
2385 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2386 		DTRACE_PROBE2(nfs4serv__func__referral__shared__moved,
2387 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2388 		return;
2389 	}
2390 
2391 	/* check for reparse point at vp */
2392 	if (vn_is_nfs_reparse(cs->vp, cs->cr) && !client_is_downrev(req)) {
2393 		/* it's not all bad */
2394 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2395 		DTRACE_PROBE2(nfs4serv__func__referral__moved,
2396 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2397 		return;
2398 	}
2399 
2400 	resp->object.nfs_fh4_val =
2401 	    kmem_alloc(cs->fh.nfs_fh4_len, KM_SLEEP);
2402 	nfs_fh4_copy(&cs->fh, &resp->object);
2403 	*cs->statusp = resp->status = NFS4_OK;
2404 out:
2405 	DTRACE_NFSV4_2(op__getfh__done, struct compound_state *, cs,
2406 	    GETFH4res *, resp);
2407 }
2408 
2409 static void
2410 rfs4_op_getfh_free(nfs_resop4 *resop)
2411 {
2412 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2413 
2414 	if (resp->status == NFS4_OK &&
2415 	    resp->object.nfs_fh4_val != NULL) {
2416 		kmem_free(resp->object.nfs_fh4_val, resp->object.nfs_fh4_len);
2417 		resp->object.nfs_fh4_val = NULL;
2418 		resp->object.nfs_fh4_len = 0;
2419 	}
2420 }
2421 
2422 /*
2423  * illegal: args: void
2424  *	    res : status (NFS4ERR_OP_ILLEGAL)
2425  */
2426 /* ARGSUSED */
2427 static void
2428 rfs4_op_illegal(nfs_argop4 *argop, nfs_resop4 *resop,
2429     struct svc_req *req, struct compound_state *cs)
2430 {
2431 	ILLEGAL4res *resp = &resop->nfs_resop4_u.opillegal;
2432 
2433 	resop->resop = OP_ILLEGAL;
2434 	*cs->statusp = resp->status = NFS4ERR_OP_ILLEGAL;
2435 }
2436 
2437 /*
2438  * link: args: SAVED_FH: file, CURRENT_FH: target directory
2439  *	 res: status. If success - CURRENT_FH unchanged, return change_info
2440  */
2441 /* ARGSUSED */
2442 static void
2443 rfs4_op_link(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2444     struct compound_state *cs)
2445 {
2446 	LINK4args *args = &argop->nfs_argop4_u.oplink;
2447 	LINK4res *resp = &resop->nfs_resop4_u.oplink;
2448 	int error;
2449 	vnode_t *vp;
2450 	vnode_t *dvp;
2451 	struct vattr bdva, idva, adva;
2452 	char *nm;
2453 	uint_t  len;
2454 	struct sockaddr *ca;
2455 	char *name = NULL;
2456 	nfsstat4 status;
2457 
2458 	DTRACE_NFSV4_2(op__link__start, struct compound_state *, cs,
2459 	    LINK4args *, args);
2460 
2461 	/* SAVED_FH: source object */
2462 	vp = cs->saved_vp;
2463 	if (vp == NULL) {
2464 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2465 		goto out;
2466 	}
2467 
2468 	/* CURRENT_FH: target directory */
2469 	dvp = cs->vp;
2470 	if (dvp == NULL) {
2471 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2472 		goto out;
2473 	}
2474 
2475 	/*
2476 	 * If there is a non-shared filesystem mounted on this vnode,
2477 	 * do not allow to link any file in this directory.
2478 	 */
2479 	if (vn_ismntpt(dvp)) {
2480 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2481 		goto out;
2482 	}
2483 
2484 	if (cs->access == CS_ACCESS_DENIED) {
2485 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2486 		goto out;
2487 	}
2488 
2489 	/* Check source object's type validity */
2490 	if (vp->v_type == VDIR) {
2491 		*cs->statusp = resp->status = NFS4ERR_ISDIR;
2492 		goto out;
2493 	}
2494 
2495 	/* Check target directory's type */
2496 	if (dvp->v_type != VDIR) {
2497 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2498 		goto out;
2499 	}
2500 
2501 	if (cs->saved_exi != cs->exi) {
2502 		*cs->statusp = resp->status = NFS4ERR_XDEV;
2503 		goto out;
2504 	}
2505 
2506 	status = utf8_dir_verify(&args->newname);
2507 	if (status != NFS4_OK) {
2508 		*cs->statusp = resp->status = status;
2509 		goto out;
2510 	}
2511 
2512 	nm = utf8_to_fn(&args->newname, &len, NULL);
2513 	if (nm == NULL) {
2514 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2515 		goto out;
2516 	}
2517 
2518 	if (len > MAXNAMELEN) {
2519 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2520 		kmem_free(nm, len);
2521 		goto out;
2522 	}
2523 
2524 	if (rdonly4(cs->exi, cs->vp, req)) {
2525 		*cs->statusp = resp->status = NFS4ERR_ROFS;
2526 		kmem_free(nm, len);
2527 		goto out;
2528 	}
2529 
2530 	/* Get "before" change value */
2531 	bdva.va_mask = AT_CTIME|AT_SEQ;
2532 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
2533 	if (error) {
2534 		*cs->statusp = resp->status = puterrno4(error);
2535 		kmem_free(nm, len);
2536 		goto out;
2537 	}
2538 
2539 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2540 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2541 	    MAXPATHLEN  + 1);
2542 
2543 	if (name == NULL) {
2544 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2545 		kmem_free(nm, len);
2546 		goto out;
2547 	}
2548 
2549 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
2550 
2551 	error = VOP_LINK(dvp, vp, name, cs->cr, NULL, 0);
2552 
2553 	if (nm != name)
2554 		kmem_free(name, MAXPATHLEN + 1);
2555 	kmem_free(nm, len);
2556 
2557 	/*
2558 	 * Get the initial "after" sequence number, if it fails, set to zero
2559 	 */
2560 	idva.va_mask = AT_SEQ;
2561 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
2562 		idva.va_seq = 0;
2563 
2564 	/*
2565 	 * Force modified data and metadata out to stable storage.
2566 	 */
2567 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
2568 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
2569 
2570 	if (error) {
2571 		*cs->statusp = resp->status = puterrno4(error);
2572 		goto out;
2573 	}
2574 
2575 	/*
2576 	 * Get "after" change value, if it fails, simply return the
2577 	 * before value.
2578 	 */
2579 	adva.va_mask = AT_CTIME|AT_SEQ;
2580 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
2581 		adva.va_ctime = bdva.va_ctime;
2582 		adva.va_seq = 0;
2583 	}
2584 
2585 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
2586 
2587 	/*
2588 	 * The cinfo.atomic = TRUE only if we have
2589 	 * non-zero va_seq's, and it has incremented by exactly one
2590 	 * during the VOP_LINK and it didn't change during the VOP_FSYNC.
2591 	 */
2592 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
2593 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
2594 		resp->cinfo.atomic = TRUE;
2595 	else
2596 		resp->cinfo.atomic = FALSE;
2597 
2598 	*cs->statusp = resp->status = NFS4_OK;
2599 out:
2600 	DTRACE_NFSV4_2(op__link__done, struct compound_state *, cs,
2601 	    LINK4res *, resp);
2602 }
2603 
2604 /*
2605  * Used by rfs4_op_lookup and rfs4_op_lookupp to do the actual work.
2606  */
2607 
2608 /* ARGSUSED */
2609 static nfsstat4
2610 do_rfs4_op_lookup(char *nm, struct svc_req *req, struct compound_state *cs)
2611 {
2612 	int error;
2613 	int different_export = 0;
2614 	vnode_t *vp, *tvp, *pre_tvp = NULL, *oldvp = NULL;
2615 	struct exportinfo *exi = NULL, *pre_exi = NULL;
2616 	nfsstat4 stat;
2617 	fid_t fid;
2618 	int attrdir, dotdot, walk;
2619 	bool_t is_newvp = FALSE;
2620 
2621 	if (cs->vp->v_flag & V_XATTRDIR) {
2622 		attrdir = 1;
2623 		ASSERT(get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2624 	} else {
2625 		attrdir = 0;
2626 		ASSERT(! get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2627 	}
2628 
2629 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
2630 
2631 	/*
2632 	 * If dotdotting, then need to check whether it's
2633 	 * above the root of a filesystem, or above an
2634 	 * export point.
2635 	 */
2636 	if (dotdot) {
2637 
2638 		/*
2639 		 * If dotdotting at the root of a filesystem, then
2640 		 * need to traverse back to the mounted-on filesystem
2641 		 * and do the dotdot lookup there.
2642 		 */
2643 		if (cs->vp->v_flag & VROOT) {
2644 
2645 			/*
2646 			 * If at the system root, then can
2647 			 * go up no further.
2648 			 */
2649 			if (VN_CMP(cs->vp, rootdir))
2650 				return (puterrno4(ENOENT));
2651 
2652 			/*
2653 			 * Traverse back to the mounted-on filesystem
2654 			 */
2655 			cs->vp = untraverse(cs->vp);
2656 
2657 			/*
2658 			 * Set the different_export flag so we remember
2659 			 * to pick up a new exportinfo entry for
2660 			 * this new filesystem.
2661 			 */
2662 			different_export = 1;
2663 		} else {
2664 
2665 			/*
2666 			 * If dotdotting above an export point then set
2667 			 * the different_export to get new export info.
2668 			 */
2669 			different_export = nfs_exported(cs->exi, cs->vp);
2670 		}
2671 	}
2672 
2673 	error = VOP_LOOKUP(cs->vp, nm, &vp, NULL, 0, NULL, cs->cr,
2674 	    NULL, NULL, NULL);
2675 	if (error)
2676 		return (puterrno4(error));
2677 
2678 	/*
2679 	 * If the vnode is in a pseudo filesystem, check whether it is visible.
2680 	 *
2681 	 * XXX if the vnode is a symlink and it is not visible in
2682 	 * a pseudo filesystem, return ENOENT (not following symlink).
2683 	 * V4 client can not mount such symlink. This is a regression
2684 	 * from V2/V3.
2685 	 *
2686 	 * In the same exported filesystem, if the security flavor used
2687 	 * is not an explicitly shared flavor, limit the view to the visible
2688 	 * list entries only. This is not a WRONGSEC case because it's already
2689 	 * checked via PUTROOTFH/PUTPUBFH or PUTFH.
2690 	 */
2691 	if (!different_export &&
2692 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
2693 	    cs->access & CS_ACCESS_LIMITED)) {
2694 		if (! nfs_visible(cs->exi, vp, &different_export)) {
2695 			VN_RELE(vp);
2696 			return (puterrno4(ENOENT));
2697 		}
2698 	}
2699 
2700 	/*
2701 	 * If it's a mountpoint, then traverse it.
2702 	 */
2703 	if (vn_ismntpt(vp)) {
2704 		pre_exi = cs->exi;	/* save pre-traversed exportinfo */
2705 		pre_tvp = vp;		/* save pre-traversed vnode	*/
2706 
2707 		/*
2708 		 * hold pre_tvp to counteract rele by traverse.  We will
2709 		 * need pre_tvp below if checkexport4 fails
2710 		 */
2711 		VN_HOLD(pre_tvp);
2712 		tvp = vp;
2713 		if ((error = traverse(&tvp)) != 0) {
2714 			VN_RELE(vp);
2715 			VN_RELE(pre_tvp);
2716 			return (puterrno4(error));
2717 		}
2718 		vp = tvp;
2719 		different_export = 1;
2720 	} else if (vp->v_vfsp != cs->vp->v_vfsp) {
2721 		/*
2722 		 * The vfsp comparison is to handle the case where
2723 		 * a LOFS mount is shared.  lo_lookup traverses mount points,
2724 		 * and NFS is unaware of local fs transistions because
2725 		 * v_vfsmountedhere isn't set.  For this special LOFS case,
2726 		 * the dir and the obj returned by lookup will have different
2727 		 * vfs ptrs.
2728 		 */
2729 		different_export = 1;
2730 	}
2731 
2732 	if (different_export) {
2733 
2734 		bzero(&fid, sizeof (fid));
2735 		fid.fid_len = MAXFIDSZ;
2736 		error = vop_fid_pseudo(vp, &fid);
2737 		if (error) {
2738 			VN_RELE(vp);
2739 			if (pre_tvp)
2740 				VN_RELE(pre_tvp);
2741 			return (puterrno4(error));
2742 		}
2743 
2744 		if (dotdot)
2745 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
2746 		else
2747 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2748 
2749 		if (exi == NULL) {
2750 			if (pre_tvp) {
2751 				/*
2752 				 * If this vnode is a mounted-on vnode,
2753 				 * but the mounted-on file system is not
2754 				 * exported, send back the filehandle for
2755 				 * the mounted-on vnode, not the root of
2756 				 * the mounted-on file system.
2757 				 */
2758 				VN_RELE(vp);
2759 				vp = pre_tvp;
2760 				exi = pre_exi;
2761 			} else {
2762 				VN_RELE(vp);
2763 				return (puterrno4(EACCES));
2764 			}
2765 		} else if (pre_tvp) {
2766 			/* we're done with pre_tvp now. release extra hold */
2767 			VN_RELE(pre_tvp);
2768 		}
2769 
2770 		cs->exi = exi;
2771 
2772 		/*
2773 		 * Now we do a checkauth4. The reason is that
2774 		 * this client/user may not have access to the new
2775 		 * exported file system, and if he does,
2776 		 * the client/user may be mapped to a different uid.
2777 		 *
2778 		 * We start with a new cr, because the checkauth4 done
2779 		 * in the PUT*FH operation over wrote the cred's uid,
2780 		 * gid, etc, and we want the real thing before calling
2781 		 * checkauth4()
2782 		 */
2783 		crfree(cs->cr);
2784 		cs->cr = crdup(cs->basecr);
2785 
2786 		oldvp = cs->vp;
2787 		cs->vp = vp;
2788 		is_newvp = TRUE;
2789 
2790 		stat = call_checkauth4(cs, req);
2791 		if (stat != NFS4_OK) {
2792 			VN_RELE(cs->vp);
2793 			cs->vp = oldvp;
2794 			return (stat);
2795 		}
2796 	}
2797 
2798 	/*
2799 	 * After various NFS checks, do a label check on the path
2800 	 * component. The label on this path should either be the
2801 	 * global zone's label or a zone's label. We are only
2802 	 * interested in the zone's label because exported files
2803 	 * in global zone is accessible (though read-only) to
2804 	 * clients. The exportability/visibility check is already
2805 	 * done before reaching this code.
2806 	 */
2807 	if (is_system_labeled()) {
2808 		bslabel_t *clabel;
2809 
2810 		ASSERT(req->rq_label != NULL);
2811 		clabel = req->rq_label;
2812 		DTRACE_PROBE2(tx__rfs4__log__info__oplookup__clabel, char *,
2813 		    "got client label from request(1)", struct svc_req *, req);
2814 
2815 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
2816 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
2817 			    cs->exi)) {
2818 				error = EACCES;
2819 				goto err_out;
2820 			}
2821 		} else {
2822 			/*
2823 			 * We grant access to admin_low label clients
2824 			 * only if the client is trusted, i.e. also
2825 			 * running Solaris Trusted Extension.
2826 			 */
2827 			struct sockaddr	*ca;
2828 			int		addr_type;
2829 			void		*ipaddr;
2830 			tsol_tpc_t	*tp;
2831 
2832 			ca = (struct sockaddr *)svc_getrpccaller(
2833 			    req->rq_xprt)->buf;
2834 			if (ca->sa_family == AF_INET) {
2835 				addr_type = IPV4_VERSION;
2836 				ipaddr = &((struct sockaddr_in *)ca)->sin_addr;
2837 			} else if (ca->sa_family == AF_INET6) {
2838 				addr_type = IPV6_VERSION;
2839 				ipaddr = &((struct sockaddr_in6 *)
2840 				    ca)->sin6_addr;
2841 			}
2842 			tp = find_tpc(ipaddr, addr_type, B_FALSE);
2843 			if (tp == NULL || tp->tpc_tp.tp_doi !=
2844 			    l_admin_low->tsl_doi || tp->tpc_tp.host_type !=
2845 			    SUN_CIPSO) {
2846 				if (tp != NULL)
2847 					TPC_RELE(tp);
2848 				error = EACCES;
2849 				goto err_out;
2850 			}
2851 			TPC_RELE(tp);
2852 		}
2853 	}
2854 
2855 	error = makefh4(&cs->fh, vp, cs->exi);
2856 
2857 err_out:
2858 	if (error) {
2859 		if (is_newvp) {
2860 			VN_RELE(cs->vp);
2861 			cs->vp = oldvp;
2862 		} else
2863 			VN_RELE(vp);
2864 		return (puterrno4(error));
2865 	}
2866 
2867 	if (!is_newvp) {
2868 		if (cs->vp)
2869 			VN_RELE(cs->vp);
2870 		cs->vp = vp;
2871 	} else if (oldvp)
2872 		VN_RELE(oldvp);
2873 
2874 	/*
2875 	 * if did lookup on attrdir and didn't lookup .., set named
2876 	 * attr fh flag
2877 	 */
2878 	if (attrdir && ! dotdot)
2879 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
2880 
2881 	/* Assume false for now, open proc will set this */
2882 	cs->mandlock = FALSE;
2883 
2884 	return (NFS4_OK);
2885 }
2886 
2887 /* ARGSUSED */
2888 static void
2889 rfs4_op_lookup(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2890     struct compound_state *cs)
2891 {
2892 	LOOKUP4args *args = &argop->nfs_argop4_u.oplookup;
2893 	LOOKUP4res *resp = &resop->nfs_resop4_u.oplookup;
2894 	char *nm;
2895 	uint_t len;
2896 	struct sockaddr *ca;
2897 	char *name = NULL;
2898 	nfsstat4 status;
2899 
2900 	DTRACE_NFSV4_2(op__lookup__start, struct compound_state *, cs,
2901 	    LOOKUP4args *, args);
2902 
2903 	if (cs->vp == NULL) {
2904 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2905 		goto out;
2906 	}
2907 
2908 	if (cs->vp->v_type == VLNK) {
2909 		*cs->statusp = resp->status = NFS4ERR_SYMLINK;
2910 		goto out;
2911 	}
2912 
2913 	if (cs->vp->v_type != VDIR) {
2914 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2915 		goto out;
2916 	}
2917 
2918 	status = utf8_dir_verify(&args->objname);
2919 	if (status != NFS4_OK) {
2920 		*cs->statusp = resp->status = status;
2921 		goto out;
2922 	}
2923 
2924 	nm = utf8_to_str(&args->objname, &len, NULL);
2925 	if (nm == NULL) {
2926 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2927 		goto out;
2928 	}
2929 
2930 	if (len > MAXNAMELEN) {
2931 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2932 		kmem_free(nm, len);
2933 		goto out;
2934 	}
2935 
2936 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2937 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2938 	    MAXPATHLEN  + 1);
2939 
2940 	if (name == NULL) {
2941 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2942 		kmem_free(nm, len);
2943 		goto out;
2944 	}
2945 
2946 	*cs->statusp = resp->status = do_rfs4_op_lookup(name, req, cs);
2947 
2948 	if (name != nm)
2949 		kmem_free(name, MAXPATHLEN + 1);
2950 	kmem_free(nm, len);
2951 
2952 out:
2953 	DTRACE_NFSV4_2(op__lookup__done, struct compound_state *, cs,
2954 	    LOOKUP4res *, resp);
2955 }
2956 
2957 /* ARGSUSED */
2958 static void
2959 rfs4_op_lookupp(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
2960     struct compound_state *cs)
2961 {
2962 	LOOKUPP4res *resp = &resop->nfs_resop4_u.oplookupp;
2963 
2964 	DTRACE_NFSV4_1(op__lookupp__start, struct compound_state *, cs);
2965 
2966 	if (cs->vp == NULL) {
2967 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2968 		goto out;
2969 	}
2970 
2971 	if (cs->vp->v_type != VDIR) {
2972 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2973 		goto out;
2974 	}
2975 
2976 	*cs->statusp = resp->status = do_rfs4_op_lookup("..", req, cs);
2977 
2978 	/*
2979 	 * From NFSV4 Specification, LOOKUPP should not check for
2980 	 * NFS4ERR_WRONGSEC. Retrun NFS4_OK instead.
2981 	 */
2982 	if (resp->status == NFS4ERR_WRONGSEC) {
2983 		*cs->statusp = resp->status = NFS4_OK;
2984 	}
2985 
2986 out:
2987 	DTRACE_NFSV4_2(op__lookupp__done, struct compound_state *, cs,
2988 	    LOOKUPP4res *, resp);
2989 }
2990 
2991 
2992 /*ARGSUSED2*/
2993 static void
2994 rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2995     struct compound_state *cs)
2996 {
2997 	OPENATTR4args	*args = &argop->nfs_argop4_u.opopenattr;
2998 	OPENATTR4res	*resp = &resop->nfs_resop4_u.opopenattr;
2999 	vnode_t		*avp = NULL;
3000 	int		lookup_flags = LOOKUP_XATTR, error;
3001 	int		exp_ro = 0;
3002 
3003 	DTRACE_NFSV4_2(op__openattr__start, struct compound_state *, cs,
3004 	    OPENATTR4args *, args);
3005 
3006 	if (cs->vp == NULL) {
3007 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3008 		goto out;
3009 	}
3010 
3011 	if ((cs->vp->v_vfsp->vfs_flag & VFS_XATTR) == 0 &&
3012 	    !vfs_has_feature(cs->vp->v_vfsp, VFSFT_SYSATTR_VIEWS)) {
3013 		*cs->statusp = resp->status = puterrno4(ENOTSUP);
3014 		goto out;
3015 	}
3016 
3017 	/*
3018 	 * If file system supports passing ACE mask to VOP_ACCESS then
3019 	 * check for ACE_READ_NAMED_ATTRS, otherwise do legacy checks
3020 	 */
3021 
3022 	if (vfs_has_feature(cs->vp->v_vfsp, VFSFT_ACEMASKONACCESS))
3023 		error = VOP_ACCESS(cs->vp, ACE_READ_NAMED_ATTRS,
3024 		    V_ACE_MASK, cs->cr, NULL);
3025 	else
3026 		error = ((VOP_ACCESS(cs->vp, VREAD, 0, cs->cr, NULL) != 0) &&
3027 		    (VOP_ACCESS(cs->vp, VWRITE, 0, cs->cr, NULL) != 0) &&
3028 		    (VOP_ACCESS(cs->vp, VEXEC, 0, cs->cr, NULL) != 0));
3029 
3030 	if (error) {
3031 		*cs->statusp = resp->status = puterrno4(EACCES);
3032 		goto out;
3033 	}
3034 
3035 	/*
3036 	 * The CREATE_XATTR_DIR VOP flag cannot be specified if
3037 	 * the file system is exported read-only -- regardless of
3038 	 * createdir flag.  Otherwise the attrdir would be created
3039 	 * (assuming server fs isn't mounted readonly locally).  If
3040 	 * VOP_LOOKUP returns ENOENT in this case, the error will
3041 	 * be translated into EROFS.  ENOSYS is mapped to ENOTSUP
3042 	 * because specfs has no VOP_LOOKUP op, so the macro would
3043 	 * return ENOSYS.  EINVAL is returned by all (current)
3044 	 * Solaris file system implementations when any of their
3045 	 * restrictions are violated (xattr(dir) can't have xattrdir).
3046 	 * Returning NOTSUPP is more appropriate in this case
3047 	 * because the object will never be able to have an attrdir.
3048 	 */
3049 	if (args->createdir && ! (exp_ro = rdonly4(cs->exi, cs->vp, req)))
3050 		lookup_flags |= CREATE_XATTR_DIR;
3051 
3052 	error = VOP_LOOKUP(cs->vp, "", &avp, NULL, lookup_flags, NULL, cs->cr,
3053 	    NULL, NULL, NULL);
3054 
3055 	if (error) {
3056 		if (error == ENOENT && args->createdir && exp_ro)
3057 			*cs->statusp = resp->status = puterrno4(EROFS);
3058 		else if (error == EINVAL || error == ENOSYS)
3059 			*cs->statusp = resp->status = puterrno4(ENOTSUP);
3060 		else
3061 			*cs->statusp = resp->status = puterrno4(error);
3062 		goto out;
3063 	}
3064 
3065 	ASSERT(avp->v_flag & V_XATTRDIR);
3066 
3067 	error = makefh4(&cs->fh, avp, cs->exi);
3068 
3069 	if (error) {
3070 		VN_RELE(avp);
3071 		*cs->statusp = resp->status = puterrno4(error);
3072 		goto out;
3073 	}
3074 
3075 	VN_RELE(cs->vp);
3076 	cs->vp = avp;
3077 
3078 	/*
3079 	 * There is no requirement for an attrdir fh flag
3080 	 * because the attrdir has a vnode flag to distinguish
3081 	 * it from regular (non-xattr) directories.  The
3082 	 * FH4_ATTRDIR flag is set for future sanity checks.
3083 	 */
3084 	set_fh4_flag(&cs->fh, FH4_ATTRDIR);
3085 	*cs->statusp = resp->status = NFS4_OK;
3086 
3087 out:
3088 	DTRACE_NFSV4_2(op__openattr__done, struct compound_state *, cs,
3089 	    OPENATTR4res *, resp);
3090 }
3091 
3092 static int
3093 do_io(int direction, vnode_t *vp, struct uio *uio, int ioflag, cred_t *cred,
3094     caller_context_t *ct)
3095 {
3096 	int error;
3097 	int i;
3098 	clock_t delaytime;
3099 
3100 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
3101 
3102 	/*
3103 	 * Don't block on mandatory locks. If this routine returns
3104 	 * EAGAIN, the caller should return NFS4ERR_LOCKED.
3105 	 */
3106 	uio->uio_fmode = FNONBLOCK;
3107 
3108 	for (i = 0; i < rfs4_maxlock_tries; i++) {
3109 
3110 
3111 		if (direction == FREAD) {
3112 			(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, ct);
3113 			error = VOP_READ(vp, uio, ioflag, cred, ct);
3114 			VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, ct);
3115 		} else {
3116 			(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, ct);
3117 			error = VOP_WRITE(vp, uio, ioflag, cred, ct);
3118 			VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, ct);
3119 		}
3120 
3121 		if (error != EAGAIN)
3122 			break;
3123 
3124 		if (i < rfs4_maxlock_tries - 1) {
3125 			delay(delaytime);
3126 			delaytime *= 2;
3127 		}
3128 	}
3129 
3130 	return (error);
3131 }
3132 
3133 /* ARGSUSED */
3134 static void
3135 rfs4_op_read(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3136     struct compound_state *cs)
3137 {
3138 	READ4args *args = &argop->nfs_argop4_u.opread;
3139 	READ4res *resp = &resop->nfs_resop4_u.opread;
3140 	int error;
3141 	int verror;
3142 	vnode_t *vp;
3143 	struct vattr va;
3144 	struct iovec iov;
3145 	struct uio uio;
3146 	u_offset_t offset;
3147 	bool_t *deleg = &cs->deleg;
3148 	nfsstat4 stat;
3149 	int in_crit = 0;
3150 	mblk_t *mp = NULL;
3151 	int alloc_err = 0;
3152 	int rdma_used = 0;
3153 	int loaned_buffers;
3154 	caller_context_t ct;
3155 	struct uio *uiop;
3156 
3157 	DTRACE_NFSV4_2(op__read__start, struct compound_state *, cs,
3158 	    READ4args, args);
3159 
3160 	vp = cs->vp;
3161 	if (vp == NULL) {
3162 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3163 		goto out;
3164 	}
3165 	if (cs->access == CS_ACCESS_DENIED) {
3166 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3167 		goto out;
3168 	}
3169 
3170 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3171 	    deleg, TRUE, &ct)) != NFS4_OK) {
3172 		*cs->statusp = resp->status = stat;
3173 		goto out;
3174 	}
3175 
3176 	/*
3177 	 * Enter the critical region before calling VOP_RWLOCK
3178 	 * to avoid a deadlock with write requests.
3179 	 */
3180 	if (nbl_need_check(vp)) {
3181 		nbl_start_crit(vp, RW_READER);
3182 		in_crit = 1;
3183 		if (nbl_conflict(vp, NBL_READ, args->offset, args->count, 0,
3184 		    &ct)) {
3185 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
3186 			goto out;
3187 		}
3188 	}
3189 
3190 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3191 	    deleg, TRUE, &ct)) != NFS4_OK) {
3192 		*cs->statusp = resp->status = stat;
3193 		goto out;
3194 	}
3195 
3196 	if (args->wlist) {
3197 		if (args->count > clist_len(args->wlist)) {
3198 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3199 			goto out;
3200 		}
3201 		rdma_used = 1;
3202 	}
3203 
3204 	/* use loaned buffers for TCP */
3205 	loaned_buffers = (nfs_loaned_buffers && !rdma_used) ? 1 : 0;
3206 
3207 	va.va_mask = AT_MODE|AT_SIZE|AT_UID;
3208 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3209 
3210 	/*
3211 	 * If we can't get the attributes, then we can't do the
3212 	 * right access checking.  So, we'll fail the request.
3213 	 */
3214 	if (verror) {
3215 		*cs->statusp = resp->status = puterrno4(verror);
3216 		goto out;
3217 	}
3218 
3219 	if (vp->v_type != VREG) {
3220 		*cs->statusp = resp->status =
3221 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
3222 		goto out;
3223 	}
3224 
3225 	if (crgetuid(cs->cr) != va.va_uid &&
3226 	    (error = VOP_ACCESS(vp, VREAD, 0, cs->cr, &ct)) &&
3227 	    (error = VOP_ACCESS(vp, VEXEC, 0, cs->cr, &ct))) {
3228 		*cs->statusp = resp->status = puterrno4(error);
3229 		goto out;
3230 	}
3231 
3232 	if (MANDLOCK(vp, va.va_mode)) { /* XXX - V4 supports mand locking */
3233 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3234 		goto out;
3235 	}
3236 
3237 	offset = args->offset;
3238 	if (offset >= va.va_size) {
3239 		*cs->statusp = resp->status = NFS4_OK;
3240 		resp->eof = TRUE;
3241 		resp->data_len = 0;
3242 		resp->data_val = NULL;
3243 		resp->mblk = NULL;
3244 		/* RDMA */
3245 		resp->wlist = args->wlist;
3246 		resp->wlist_len = resp->data_len;
3247 		*cs->statusp = resp->status = NFS4_OK;
3248 		if (resp->wlist)
3249 			clist_zero_len(resp->wlist);
3250 		goto out;
3251 	}
3252 
3253 	if (args->count == 0) {
3254 		*cs->statusp = resp->status = NFS4_OK;
3255 		resp->eof = FALSE;
3256 		resp->data_len = 0;
3257 		resp->data_val = NULL;
3258 		resp->mblk = NULL;
3259 		/* RDMA */
3260 		resp->wlist = args->wlist;
3261 		resp->wlist_len = resp->data_len;
3262 		if (resp->wlist)
3263 			clist_zero_len(resp->wlist);
3264 		goto out;
3265 	}
3266 
3267 	/*
3268 	 * Do not allocate memory more than maximum allowed
3269 	 * transfer size
3270 	 */
3271 	if (args->count > rfs4_tsize(req))
3272 		args->count = rfs4_tsize(req);
3273 
3274 	if (loaned_buffers) {
3275 		uiop = (uio_t *)rfs_setup_xuio(vp);
3276 		ASSERT(uiop != NULL);
3277 		uiop->uio_segflg = UIO_SYSSPACE;
3278 		uiop->uio_loffset = args->offset;
3279 		uiop->uio_resid = args->count;
3280 
3281 		/* Jump to do the read if successful */
3282 		if (!VOP_REQZCBUF(vp, UIO_READ, (xuio_t *)uiop, cs->cr, &ct)) {
3283 			/*
3284 			 * Need to hold the vnode until after VOP_RETZCBUF()
3285 			 * is called.
3286 			 */
3287 			VN_HOLD(vp);
3288 			goto doio_read;
3289 		}
3290 
3291 		DTRACE_PROBE2(nfss__i__reqzcbuf_failed, int,
3292 		    uiop->uio_loffset, int, uiop->uio_resid);
3293 
3294 		uiop->uio_extflg = 0;
3295 
3296 		/* failure to setup for zero copy */
3297 		rfs_free_xuio((void *)uiop);
3298 		loaned_buffers = 0;
3299 	}
3300 
3301 	/*
3302 	 * If returning data via RDMA Write, then grab the chunk list. If we
3303 	 * aren't returning READ data w/RDMA_WRITE, then grab a mblk.
3304 	 */
3305 	if (rdma_used) {
3306 		mp = NULL;
3307 		(void) rdma_get_wchunk(req, &iov, args->wlist);
3308 	} else {
3309 		/*
3310 		 * mp will contain the data to be sent out in the read reply.
3311 		 * It will be freed after the reply has been sent. Let's
3312 		 * roundup the data to a BYTES_PER_XDR_UNIT multiple, so that
3313 		 * the call to xdrmblk_putmblk() never fails. If the first
3314 		 * alloc of the requested size fails, then decrease the size to
3315 		 * something more reasonable and wait for the allocation to
3316 		 * occur.
3317 		 */
3318 		mp = allocb(RNDUP(args->count), BPRI_MED);
3319 		if (mp == NULL) {
3320 			if (args->count > MAXBSIZE)
3321 				args->count = MAXBSIZE;
3322 			mp = allocb_wait(RNDUP(args->count), BPRI_MED,
3323 			    STR_NOSIG, &alloc_err);
3324 		}
3325 		ASSERT(mp != NULL);
3326 		ASSERT(alloc_err == 0);
3327 
3328 		iov.iov_base = (caddr_t)mp->b_datap->db_base;
3329 		iov.iov_len = args->count;
3330 	}
3331 
3332 	uio.uio_iov = &iov;
3333 	uio.uio_iovcnt = 1;
3334 	uio.uio_segflg = UIO_SYSSPACE;
3335 	uio.uio_extflg = UIO_COPY_CACHED;
3336 	uio.uio_loffset = args->offset;
3337 	uio.uio_resid = args->count;
3338 	uiop = &uio;
3339 
3340 doio_read:
3341 	error = do_io(FREAD, vp, uiop, 0, cs->cr, &ct);
3342 
3343 	va.va_mask = AT_SIZE;
3344 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3345 
3346 	if (error) {
3347 		if (mp)
3348 			freemsg(mp);
3349 		*cs->statusp = resp->status = puterrno4(error);
3350 		goto out;
3351 	}
3352 
3353 	/* make mblk using zc buffers */
3354 	if (loaned_buffers) {
3355 		mp = uio_to_mblk(uiop);
3356 		ASSERT(mp != NULL);
3357 	}
3358 
3359 	*cs->statusp = resp->status = NFS4_OK;
3360 
3361 	ASSERT(uiop->uio_resid >= 0);
3362 	resp->data_len = args->count - uiop->uio_resid;
3363 	if (mp) {
3364 		resp->data_val = (char *)mp->b_datap->db_base;
3365 		rfs_rndup_mblks(mp, resp->data_len, loaned_buffers);
3366 	} else {
3367 		resp->data_val = (caddr_t)iov.iov_base;
3368 	}
3369 
3370 	resp->mblk = mp;
3371 
3372 	if (!verror && offset + resp->data_len == va.va_size)
3373 		resp->eof = TRUE;
3374 	else
3375 		resp->eof = FALSE;
3376 
3377 	if (rdma_used) {
3378 		if (!rdma_setup_read_data4(args, resp)) {
3379 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3380 		}
3381 	} else {
3382 		resp->wlist = NULL;
3383 	}
3384 
3385 out:
3386 	if (in_crit)
3387 		nbl_end_crit(vp);
3388 
3389 	DTRACE_NFSV4_2(op__read__done, struct compound_state *, cs,
3390 	    READ4res *, resp);
3391 }
3392 
3393 static void
3394 rfs4_op_read_free(nfs_resop4 *resop)
3395 {
3396 	READ4res	*resp = &resop->nfs_resop4_u.opread;
3397 
3398 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3399 		freemsg(resp->mblk);
3400 		resp->mblk = NULL;
3401 		resp->data_val = NULL;
3402 		resp->data_len = 0;
3403 	}
3404 }
3405 
3406 static void
3407 rfs4_op_readdir_free(nfs_resop4 * resop)
3408 {
3409 	READDIR4res    *resp = &resop->nfs_resop4_u.opreaddir;
3410 
3411 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3412 		freeb(resp->mblk);
3413 		resp->mblk = NULL;
3414 		resp->data_len = 0;
3415 	}
3416 }
3417 
3418 
3419 /* ARGSUSED */
3420 static void
3421 rfs4_op_putpubfh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
3422     struct compound_state *cs)
3423 {
3424 	PUTPUBFH4res	*resp = &resop->nfs_resop4_u.opputpubfh;
3425 	int		error;
3426 	vnode_t		*vp;
3427 	struct exportinfo *exi, *sav_exi;
3428 	nfs_fh4_fmt_t	*fh_fmtp;
3429 
3430 	DTRACE_NFSV4_1(op__putpubfh__start, struct compound_state *, cs);
3431 
3432 	if (cs->vp) {
3433 		VN_RELE(cs->vp);
3434 		cs->vp = NULL;
3435 	}
3436 
3437 	if (cs->cr)
3438 		crfree(cs->cr);
3439 
3440 	cs->cr = crdup(cs->basecr);
3441 
3442 	vp = exi_public->exi_vp;
3443 	if (vp == NULL) {
3444 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3445 		goto out;
3446 	}
3447 
3448 	error = makefh4(&cs->fh, vp, exi_public);
3449 	if (error != 0) {
3450 		*cs->statusp = resp->status = puterrno4(error);
3451 		goto out;
3452 	}
3453 	sav_exi = cs->exi;
3454 	if (exi_public == exi_root) {
3455 		/*
3456 		 * No filesystem is actually shared public, so we default
3457 		 * to exi_root. In this case, we must check whether root
3458 		 * is exported.
3459 		 */
3460 		fh_fmtp = (nfs_fh4_fmt_t *)cs->fh.nfs_fh4_val;
3461 
3462 		/*
3463 		 * if root filesystem is exported, the exportinfo struct that we
3464 		 * should use is what checkexport4 returns, because root_exi is
3465 		 * actually a mostly empty struct.
3466 		 */
3467 		exi = checkexport4(&fh_fmtp->fh4_fsid,
3468 		    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
3469 		cs->exi = ((exi != NULL) ? exi : exi_public);
3470 	} else {
3471 		/*
3472 		 * it's a properly shared filesystem
3473 		 */
3474 		cs->exi = exi_public;
3475 	}
3476 
3477 	if (is_system_labeled()) {
3478 		bslabel_t *clabel;
3479 
3480 		ASSERT(req->rq_label != NULL);
3481 		clabel = req->rq_label;
3482 		DTRACE_PROBE2(tx__rfs4__log__info__opputpubfh__clabel, char *,
3483 		    "got client label from request(1)",
3484 		    struct svc_req *, req);
3485 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
3486 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
3487 			    cs->exi)) {
3488 				*cs->statusp = resp->status =
3489 				    NFS4ERR_SERVERFAULT;
3490 				goto out;
3491 			}
3492 		}
3493 	}
3494 
3495 	VN_HOLD(vp);
3496 	cs->vp = vp;
3497 
3498 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3499 		VN_RELE(cs->vp);
3500 		cs->vp = NULL;
3501 		cs->exi = sav_exi;
3502 		goto out;
3503 	}
3504 
3505 	*cs->statusp = resp->status = NFS4_OK;
3506 out:
3507 	DTRACE_NFSV4_2(op__putpubfh__done, struct compound_state *, cs,
3508 	    PUTPUBFH4res *, resp);
3509 }
3510 
3511 /*
3512  * XXX - issue with put*fh operations. Suppose /export/home is exported.
3513  * Suppose an NFS client goes to mount /export/home/joe. If /export, home,
3514  * or joe have restrictive search permissions, then we shouldn't let
3515  * the client get a file handle. This is easy to enforce. However, we
3516  * don't know what security flavor should be used until we resolve the
3517  * path name. Another complication is uid mapping. If root is
3518  * the user, then it will be mapped to the anonymous user by default,
3519  * but we won't know that till we've resolved the path name. And we won't
3520  * know what the anonymous user is.
3521  * Luckily, SECINFO is specified to take a full filename.
3522  * So what we will have to in rfs4_op_lookup is check that flavor of
3523  * the target object matches that of the request, and if root was the
3524  * caller, check for the root= and anon= options, and if necessary,
3525  * repeat the lookup using the right cred_t. But that's not done yet.
3526  */
3527 /* ARGSUSED */
3528 static void
3529 rfs4_op_putfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3530     struct compound_state *cs)
3531 {
3532 	PUTFH4args *args = &argop->nfs_argop4_u.opputfh;
3533 	PUTFH4res *resp = &resop->nfs_resop4_u.opputfh;
3534 	nfs_fh4_fmt_t *fh_fmtp;
3535 
3536 	DTRACE_NFSV4_2(op__putfh__start, struct compound_state *, cs,
3537 	    PUTFH4args *, args);
3538 
3539 	if (cs->vp) {
3540 		VN_RELE(cs->vp);
3541 		cs->vp = NULL;
3542 	}
3543 
3544 	if (cs->cr) {
3545 		crfree(cs->cr);
3546 		cs->cr = NULL;
3547 	}
3548 
3549 
3550 	if (args->object.nfs_fh4_len < NFS_FH4_LEN) {
3551 		*cs->statusp = resp->status = NFS4ERR_BADHANDLE;
3552 		goto out;
3553 	}
3554 
3555 	fh_fmtp = (nfs_fh4_fmt_t *)args->object.nfs_fh4_val;
3556 	cs->exi = checkexport4(&fh_fmtp->fh4_fsid, (fid_t *)&fh_fmtp->fh4_xlen,
3557 	    NULL);
3558 
3559 	if (cs->exi == NULL) {
3560 		*cs->statusp = resp->status = NFS4ERR_STALE;
3561 		goto out;
3562 	}
3563 
3564 	cs->cr = crdup(cs->basecr);
3565 
3566 	ASSERT(cs->cr != NULL);
3567 
3568 	if (! (cs->vp = nfs4_fhtovp(&args->object, cs->exi, &resp->status))) {
3569 		*cs->statusp = resp->status;
3570 		goto out;
3571 	}
3572 
3573 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3574 		VN_RELE(cs->vp);
3575 		cs->vp = NULL;
3576 		goto out;
3577 	}
3578 
3579 	nfs_fh4_copy(&args->object, &cs->fh);
3580 	*cs->statusp = resp->status = NFS4_OK;
3581 	cs->deleg = FALSE;
3582 
3583 out:
3584 	DTRACE_NFSV4_2(op__putfh__done, struct compound_state *, cs,
3585 	    PUTFH4res *, resp);
3586 }
3587 
3588 /* ARGSUSED */
3589 static void
3590 rfs4_op_putrootfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3591     struct compound_state *cs)
3592 {
3593 	PUTROOTFH4res *resp = &resop->nfs_resop4_u.opputrootfh;
3594 	int error;
3595 	fid_t fid;
3596 	struct exportinfo *exi, *sav_exi;
3597 
3598 	DTRACE_NFSV4_1(op__putrootfh__start, struct compound_state *, cs);
3599 
3600 	if (cs->vp) {
3601 		VN_RELE(cs->vp);
3602 		cs->vp = NULL;
3603 	}
3604 
3605 	if (cs->cr)
3606 		crfree(cs->cr);
3607 
3608 	cs->cr = crdup(cs->basecr);
3609 
3610 	/*
3611 	 * Using rootdir, the system root vnode,
3612 	 * get its fid.
3613 	 */
3614 	bzero(&fid, sizeof (fid));
3615 	fid.fid_len = MAXFIDSZ;
3616 	error = vop_fid_pseudo(rootdir, &fid);
3617 	if (error != 0) {
3618 		*cs->statusp = resp->status = puterrno4(error);
3619 		goto out;
3620 	}
3621 
3622 	/*
3623 	 * Then use the root fsid & fid it to find out if it's exported
3624 	 *
3625 	 * If the server root isn't exported directly, then
3626 	 * it should at least be a pseudo export based on
3627 	 * one or more exports further down in the server's
3628 	 * file tree.
3629 	 */
3630 	exi = checkexport4(&rootdir->v_vfsp->vfs_fsid, &fid, NULL);
3631 	if (exi == NULL || exi->exi_export.ex_flags & EX_PUBLIC) {
3632 		NFS4_DEBUG(rfs4_debug,
3633 		    (CE_WARN, "rfs4_op_putrootfh: export check failure"));
3634 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3635 		goto out;
3636 	}
3637 
3638 	/*
3639 	 * Now make a filehandle based on the root
3640 	 * export and root vnode.
3641 	 */
3642 	error = makefh4(&cs->fh, rootdir, exi);
3643 	if (error != 0) {
3644 		*cs->statusp = resp->status = puterrno4(error);
3645 		goto out;
3646 	}
3647 
3648 	sav_exi = cs->exi;
3649 	cs->exi = exi;
3650 
3651 	VN_HOLD(rootdir);
3652 	cs->vp = rootdir;
3653 
3654 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3655 		VN_RELE(rootdir);
3656 		cs->vp = NULL;
3657 		cs->exi = sav_exi;
3658 		goto out;
3659 	}
3660 
3661 	*cs->statusp = resp->status = NFS4_OK;
3662 	cs->deleg = FALSE;
3663 out:
3664 	DTRACE_NFSV4_2(op__putrootfh__done, struct compound_state *, cs,
3665 	    PUTROOTFH4res *, resp);
3666 }
3667 
3668 /*
3669  * set_rdattr_params sets up the variables used to manage what information
3670  * to get for each directory entry.
3671  */
3672 static nfsstat4
3673 set_rdattr_params(struct nfs4_svgetit_arg *sargp,
3674     bitmap4 attrs, bool_t *need_to_lookup)
3675 {
3676 	uint_t	va_mask;
3677 	nfsstat4 status;
3678 	bitmap4 objbits;
3679 
3680 	status = bitmap4_to_attrmask(attrs, sargp);
3681 	if (status != NFS4_OK) {
3682 		/*
3683 		 * could not even figure attr mask
3684 		 */
3685 		return (status);
3686 	}
3687 	va_mask = sargp->vap->va_mask;
3688 
3689 	/*
3690 	 * dirent's d_ino is always correct value for mounted_on_fileid.
3691 	 * mntdfid_set is set once here, but mounted_on_fileid is
3692 	 * set in main dirent processing loop for each dirent.
3693 	 * The mntdfid_set is a simple optimization that lets the
3694 	 * server attr code avoid work when caller is readdir.
3695 	 */
3696 	sargp->mntdfid_set = TRUE;
3697 
3698 	/*
3699 	 * Lookup entry only if client asked for any of the following:
3700 	 * a) vattr attrs
3701 	 * b) vfs attrs
3702 	 * c) attrs w/per-object scope requested (change, filehandle, etc)
3703 	 *    other than mounted_on_fileid (which we can take from dirent)
3704 	 */
3705 	objbits = attrs ? attrs & NFS4_VP_ATTR_MASK : 0;
3706 
3707 	if (va_mask || sargp->sbp || (objbits & ~FATTR4_MOUNTED_ON_FILEID_MASK))
3708 		*need_to_lookup = TRUE;
3709 	else
3710 		*need_to_lookup = FALSE;
3711 
3712 	if (sargp->sbp == NULL)
3713 		return (NFS4_OK);
3714 
3715 	/*
3716 	 * If filesystem attrs are requested, get them now from the
3717 	 * directory vp, as most entries will have same filesystem. The only
3718 	 * exception are mounted over entries but we handle
3719 	 * those as we go (XXX mounted over detection not yet implemented).
3720 	 */
3721 	sargp->vap->va_mask = 0;	/* to avoid VOP_GETATTR */
3722 	status = bitmap4_get_sysattrs(sargp);
3723 	sargp->vap->va_mask = va_mask;
3724 
3725 	if ((status != NFS4_OK) && sargp->rdattr_error_req) {
3726 		/*
3727 		 * Failed to get filesystem attributes.
3728 		 * Return a rdattr_error for each entry, but don't fail.
3729 		 * However, don't get any obj-dependent attrs.
3730 		 */
3731 		sargp->rdattr_error = status;	/* for rdattr_error */
3732 		*need_to_lookup = FALSE;
3733 		/*
3734 		 * At least get fileid for regular readdir output
3735 		 */
3736 		sargp->vap->va_mask &= AT_NODEID;
3737 		status = NFS4_OK;
3738 	}
3739 
3740 	return (status);
3741 }
3742 
3743 /*
3744  * readlink: args: CURRENT_FH.
3745  *	res: status. If success - CURRENT_FH unchanged, return linktext.
3746  */
3747 
3748 /* ARGSUSED */
3749 static void
3750 rfs4_op_readlink(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3751     struct compound_state *cs)
3752 {
3753 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3754 	int error;
3755 	vnode_t *vp;
3756 	struct iovec iov;
3757 	struct vattr va;
3758 	struct uio uio;
3759 	char *data;
3760 	struct sockaddr *ca;
3761 	char *name = NULL;
3762 	int is_referral;
3763 
3764 	DTRACE_NFSV4_1(op__readlink__start, struct compound_state *, cs);
3765 
3766 	/* CURRENT_FH: directory */
3767 	vp = cs->vp;
3768 	if (vp == NULL) {
3769 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3770 		goto out;
3771 	}
3772 
3773 	if (cs->access == CS_ACCESS_DENIED) {
3774 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3775 		goto out;
3776 	}
3777 
3778 	/* Is it a referral? */
3779 	if (vn_is_nfs_reparse(vp, cs->cr) && client_is_downrev(req)) {
3780 
3781 		is_referral = 1;
3782 
3783 	} else {
3784 
3785 		is_referral = 0;
3786 
3787 		if (vp->v_type == VDIR) {
3788 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
3789 			goto out;
3790 		}
3791 
3792 		if (vp->v_type != VLNK) {
3793 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3794 			goto out;
3795 		}
3796 
3797 	}
3798 
3799 	va.va_mask = AT_MODE;
3800 	error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
3801 	if (error) {
3802 		*cs->statusp = resp->status = puterrno4(error);
3803 		goto out;
3804 	}
3805 
3806 	if (MANDLOCK(vp, va.va_mode)) {
3807 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3808 		goto out;
3809 	}
3810 
3811 	data = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);
3812 
3813 	if (is_referral) {
3814 		char *s;
3815 		size_t strsz;
3816 
3817 		/* Get an artificial symlink based on a referral */
3818 		s = build_symlink(vp, cs->cr, &strsz);
3819 		global_svstat_ptr[4][NFS_REFERLINKS].value.ui64++;
3820 		DTRACE_PROBE2(nfs4serv__func__referral__reflink,
3821 		    vnode_t *, vp, char *, s);
3822 		if (s == NULL)
3823 			error = EINVAL;
3824 		else {
3825 			error = 0;
3826 			(void) strlcpy(data, s, MAXPATHLEN + 1);
3827 			kmem_free(s, strsz);
3828 		}
3829 
3830 	} else {
3831 
3832 		iov.iov_base = data;
3833 		iov.iov_len = MAXPATHLEN;
3834 		uio.uio_iov = &iov;
3835 		uio.uio_iovcnt = 1;
3836 		uio.uio_segflg = UIO_SYSSPACE;
3837 		uio.uio_extflg = UIO_COPY_CACHED;
3838 		uio.uio_loffset = 0;
3839 		uio.uio_resid = MAXPATHLEN;
3840 
3841 		error = VOP_READLINK(vp, &uio, cs->cr, NULL);
3842 
3843 		if (!error)
3844 			*(data + MAXPATHLEN - uio.uio_resid) = '\0';
3845 	}
3846 
3847 	if (error) {
3848 		kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3849 		*cs->statusp = resp->status = puterrno4(error);
3850 		goto out;
3851 	}
3852 
3853 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
3854 	name = nfscmd_convname(ca, cs->exi, data, NFSCMD_CONV_OUTBOUND,
3855 	    MAXPATHLEN  + 1);
3856 
3857 	if (name == NULL) {
3858 		/*
3859 		 * Even though the conversion failed, we return
3860 		 * something. We just don't translate it.
3861 		 */
3862 		name = data;
3863 	}
3864 
3865 	/*
3866 	 * treat link name as data
3867 	 */
3868 	(void) str_to_utf8(name, &resp->link);
3869 
3870 	if (name != data)
3871 		kmem_free(name, MAXPATHLEN + 1);
3872 	kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3873 	*cs->statusp = resp->status = NFS4_OK;
3874 
3875 out:
3876 	DTRACE_NFSV4_2(op__readlink__done, struct compound_state *, cs,
3877 	    READLINK4res *, resp);
3878 }
3879 
3880 static void
3881 rfs4_op_readlink_free(nfs_resop4 *resop)
3882 {
3883 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3884 	utf8string *symlink = &resp->link;
3885 
3886 	if (symlink->utf8string_val) {
3887 		UTF8STRING_FREE(*symlink)
3888 	}
3889 }
3890 
3891 /*
3892  * release_lockowner:
3893  *	Release any state associated with the supplied
3894  *	lockowner. Note if any lo_state is holding locks we will not
3895  *	rele that lo_state and thus the lockowner will not be destroyed.
3896  *	A client using lock after the lock owner stateid has been released
3897  *	will suffer the consequence of NFS4ERR_BAD_STATEID and would have
3898  *	to reissue the lock with new_lock_owner set to TRUE.
3899  *	args: lock_owner
3900  *	res:  status
3901  */
3902 /* ARGSUSED */
3903 static void
3904 rfs4_op_release_lockowner(nfs_argop4 *argop, nfs_resop4 *resop,
3905     struct svc_req *req, struct compound_state *cs)
3906 {
3907 	RELEASE_LOCKOWNER4args *ap = &argop->nfs_argop4_u.oprelease_lockowner;
3908 	RELEASE_LOCKOWNER4res *resp = &resop->nfs_resop4_u.oprelease_lockowner;
3909 	rfs4_lockowner_t *lo;
3910 	rfs4_openowner_t *oo;
3911 	rfs4_state_t *sp;
3912 	rfs4_lo_state_t *lsp;
3913 	rfs4_client_t *cp;
3914 	bool_t create = FALSE;
3915 	locklist_t *llist;
3916 	sysid_t sysid;
3917 
3918 	DTRACE_NFSV4_2(op__release__lockowner__start, struct compound_state *,
3919 	    cs, RELEASE_LOCKOWNER4args *, ap);
3920 
3921 	/* Make sure there is a clientid around for this request */
3922 	cp = rfs4_findclient_by_id(ap->lock_owner.clientid, FALSE);
3923 
3924 	if (cp == NULL) {
3925 		*cs->statusp = resp->status =
3926 		    rfs4_check_clientid(&ap->lock_owner.clientid, 0);
3927 		goto out;
3928 	}
3929 	rfs4_client_rele(cp);
3930 
3931 	lo = rfs4_findlockowner(&ap->lock_owner, &create);
3932 	if (lo == NULL) {
3933 		*cs->statusp = resp->status = NFS4_OK;
3934 		goto out;
3935 	}
3936 	ASSERT(lo->rl_client != NULL);
3937 
3938 	/*
3939 	 * Check for EXPIRED client. If so will reap state with in a lease
3940 	 * period or on next set_clientid_confirm step
3941 	 */
3942 	if (rfs4_lease_expired(lo->rl_client)) {
3943 		rfs4_lockowner_rele(lo);
3944 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
3945 		goto out;
3946 	}
3947 
3948 	/*
3949 	 * If no sysid has been assigned, then no locks exist; just return.
3950 	 */
3951 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3952 	if (lo->rl_client->rc_sysidt == LM_NOSYSID) {
3953 		rfs4_lockowner_rele(lo);
3954 		rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3955 		goto out;
3956 	}
3957 
3958 	sysid = lo->rl_client->rc_sysidt;
3959 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3960 
3961 	/*
3962 	 * Mark the lockowner invalid.
3963 	 */
3964 	rfs4_dbe_hide(lo->rl_dbe);
3965 
3966 	/*
3967 	 * sysid-pid pair should now not be used since the lockowner is
3968 	 * invalid. If the client were to instantiate the lockowner again
3969 	 * it would be assigned a new pid. Thus we can get the list of
3970 	 * current locks.
3971 	 */
3972 
3973 	llist = flk_get_active_locks(sysid, lo->rl_pid);
3974 	/* If we are still holding locks fail */
3975 	if (llist != NULL) {
3976 
3977 		*cs->statusp = resp->status = NFS4ERR_LOCKS_HELD;
3978 
3979 		flk_free_locklist(llist);
3980 		/*
3981 		 * We need to unhide the lockowner so the client can
3982 		 * try it again. The bad thing here is if the client
3983 		 * has a logic error that took it here in the first place
3984 		 * he probably has lost accounting of the locks that it
3985 		 * is holding. So we may have dangling state until the
3986 		 * open owner state is reaped via close. One scenario
3987 		 * that could possibly occur is that the client has
3988 		 * sent the unlock request(s) in separate threads
3989 		 * and has not waited for the replies before sending the
3990 		 * RELEASE_LOCKOWNER request. Presumably, it would expect
3991 		 * and deal appropriately with NFS4ERR_LOCKS_HELD, by
3992 		 * reissuing the request.
3993 		 */
3994 		rfs4_dbe_unhide(lo->rl_dbe);
3995 		rfs4_lockowner_rele(lo);
3996 		goto out;
3997 	}
3998 
3999 	/*
4000 	 * For the corresponding client we need to check each open
4001 	 * owner for any opens that have lockowner state associated
4002 	 * with this lockowner.
4003 	 */
4004 
4005 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
4006 	for (oo = list_head(&lo->rl_client->rc_openownerlist); oo != NULL;
4007 	    oo = list_next(&lo->rl_client->rc_openownerlist, oo)) {
4008 
4009 		rfs4_dbe_lock(oo->ro_dbe);
4010 		for (sp = list_head(&oo->ro_statelist); sp != NULL;
4011 		    sp = list_next(&oo->ro_statelist, sp)) {
4012 
4013 			rfs4_dbe_lock(sp->rs_dbe);
4014 			for (lsp = list_head(&sp->rs_lostatelist);
4015 			    lsp != NULL;
4016 			    lsp = list_next(&sp->rs_lostatelist, lsp)) {
4017 				if (lsp->rls_locker == lo) {
4018 					rfs4_dbe_lock(lsp->rls_dbe);
4019 					rfs4_dbe_invalidate(lsp->rls_dbe);
4020 					rfs4_dbe_unlock(lsp->rls_dbe);
4021 				}
4022 			}
4023 			rfs4_dbe_unlock(sp->rs_dbe);
4024 		}
4025 		rfs4_dbe_unlock(oo->ro_dbe);
4026 	}
4027 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
4028 
4029 	rfs4_lockowner_rele(lo);
4030 
4031 	*cs->statusp = resp->status = NFS4_OK;
4032 
4033 out:
4034 	DTRACE_NFSV4_2(op__release__lockowner__done, struct compound_state *,
4035 	    cs, RELEASE_LOCKOWNER4res *, resp);
4036 }
4037 
4038 /*
4039  * short utility function to lookup a file and recall the delegation
4040  */
4041 static rfs4_file_t *
4042 rfs4_lookup_and_findfile(vnode_t *dvp, char *nm, vnode_t **vpp,
4043     int *lkup_error, cred_t *cr)
4044 {
4045 	vnode_t *vp;
4046 	rfs4_file_t *fp = NULL;
4047 	bool_t fcreate = FALSE;
4048 	int error;
4049 
4050 	if (vpp)
4051 		*vpp = NULL;
4052 
4053 	if ((error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cr, NULL, NULL,
4054 	    NULL)) == 0) {
4055 		if (vp->v_type == VREG)
4056 			fp = rfs4_findfile(vp, NULL, &fcreate);
4057 		if (vpp)
4058 			*vpp = vp;
4059 		else
4060 			VN_RELE(vp);
4061 	}
4062 
4063 	if (lkup_error)
4064 		*lkup_error = error;
4065 
4066 	return (fp);
4067 }
4068 
4069 /*
4070  * remove: args: CURRENT_FH: directory; name.
4071  *	res: status. If success - CURRENT_FH unchanged, return change_info
4072  *		for directory.
4073  */
4074 /* ARGSUSED */
4075 static void
4076 rfs4_op_remove(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4077     struct compound_state *cs)
4078 {
4079 	REMOVE4args *args = &argop->nfs_argop4_u.opremove;
4080 	REMOVE4res *resp = &resop->nfs_resop4_u.opremove;
4081 	int error;
4082 	vnode_t *dvp, *vp;
4083 	struct vattr bdva, idva, adva;
4084 	char *nm;
4085 	uint_t len;
4086 	rfs4_file_t *fp;
4087 	int in_crit = 0;
4088 	bslabel_t *clabel;
4089 	struct sockaddr *ca;
4090 	char *name = NULL;
4091 	nfsstat4 status;
4092 
4093 	DTRACE_NFSV4_2(op__remove__start, struct compound_state *, cs,
4094 	    REMOVE4args *, args);
4095 
4096 	/* CURRENT_FH: directory */
4097 	dvp = cs->vp;
4098 	if (dvp == NULL) {
4099 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4100 		goto out;
4101 	}
4102 
4103 	if (cs->access == CS_ACCESS_DENIED) {
4104 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4105 		goto out;
4106 	}
4107 
4108 	/*
4109 	 * If there is an unshared filesystem mounted on this vnode,
4110 	 * Do not allow to remove anything in this directory.
4111 	 */
4112 	if (vn_ismntpt(dvp)) {
4113 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4114 		goto out;
4115 	}
4116 
4117 	if (dvp->v_type != VDIR) {
4118 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4119 		goto out;
4120 	}
4121 
4122 	status = utf8_dir_verify(&args->target);
4123 	if (status != NFS4_OK) {
4124 		*cs->statusp = resp->status = status;
4125 		goto out;
4126 	}
4127 
4128 	/*
4129 	 * Lookup the file so that we can check if it's a directory
4130 	 */
4131 	nm = utf8_to_fn(&args->target, &len, NULL);
4132 	if (nm == NULL) {
4133 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4134 		goto out;
4135 	}
4136 
4137 	if (len > MAXNAMELEN) {
4138 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4139 		kmem_free(nm, len);
4140 		goto out;
4141 	}
4142 
4143 	if (rdonly4(cs->exi, cs->vp, req)) {
4144 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4145 		kmem_free(nm, len);
4146 		goto out;
4147 	}
4148 
4149 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4150 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
4151 	    MAXPATHLEN  + 1);
4152 
4153 	if (name == NULL) {
4154 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4155 		kmem_free(nm, len);
4156 		goto out;
4157 	}
4158 
4159 	/*
4160 	 * Lookup the file to determine type and while we are see if
4161 	 * there is a file struct around and check for delegation.
4162 	 * We don't need to acquire va_seq before this lookup, if
4163 	 * it causes an update, cinfo.before will not match, which will
4164 	 * trigger a cache flush even if atomic is TRUE.
4165 	 */
4166 	if (fp = rfs4_lookup_and_findfile(dvp, name, &vp, &error, cs->cr)) {
4167 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4168 		    NULL)) {
4169 			VN_RELE(vp);
4170 			rfs4_file_rele(fp);
4171 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4172 			if (nm != name)
4173 				kmem_free(name, MAXPATHLEN + 1);
4174 			kmem_free(nm, len);
4175 			goto out;
4176 		}
4177 	}
4178 
4179 	/* Didn't find anything to remove */
4180 	if (vp == NULL) {
4181 		*cs->statusp = resp->status = error;
4182 		if (nm != name)
4183 			kmem_free(name, MAXPATHLEN + 1);
4184 		kmem_free(nm, len);
4185 		goto out;
4186 	}
4187 
4188 	if (nbl_need_check(vp)) {
4189 		nbl_start_crit(vp, RW_READER);
4190 		in_crit = 1;
4191 		if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
4192 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4193 			if (nm != name)
4194 				kmem_free(name, MAXPATHLEN + 1);
4195 			kmem_free(nm, len);
4196 			nbl_end_crit(vp);
4197 			VN_RELE(vp);
4198 			if (fp) {
4199 				rfs4_clear_dont_grant(fp);
4200 				rfs4_file_rele(fp);
4201 			}
4202 			goto out;
4203 		}
4204 	}
4205 
4206 	/* check label before allowing removal */
4207 	if (is_system_labeled()) {
4208 		ASSERT(req->rq_label != NULL);
4209 		clabel = req->rq_label;
4210 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
4211 		    "got client label from request(1)",
4212 		    struct svc_req *, req);
4213 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4214 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
4215 			    cs->exi)) {
4216 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4217 				if (name != nm)
4218 					kmem_free(name, MAXPATHLEN + 1);
4219 				kmem_free(nm, len);
4220 				if (in_crit)
4221 					nbl_end_crit(vp);
4222 				VN_RELE(vp);
4223 				if (fp) {
4224 					rfs4_clear_dont_grant(fp);
4225 					rfs4_file_rele(fp);
4226 				}
4227 				goto out;
4228 			}
4229 		}
4230 	}
4231 
4232 	/* Get dir "before" change value */
4233 	bdva.va_mask = AT_CTIME|AT_SEQ;
4234 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
4235 	if (error) {
4236 		*cs->statusp = resp->status = puterrno4(error);
4237 		if (nm != name)
4238 			kmem_free(name, MAXPATHLEN + 1);
4239 		kmem_free(nm, len);
4240 		if (in_crit)
4241 			nbl_end_crit(vp);
4242 		VN_RELE(vp);
4243 		if (fp) {
4244 			rfs4_clear_dont_grant(fp);
4245 			rfs4_file_rele(fp);
4246 		}
4247 		goto out;
4248 	}
4249 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
4250 
4251 	/* Actually do the REMOVE operation */
4252 	if (vp->v_type == VDIR) {
4253 		/*
4254 		 * Can't remove a directory that has a mounted-on filesystem.
4255 		 */
4256 		if (vn_ismntpt(vp)) {
4257 			error = EACCES;
4258 		} else {
4259 			/*
4260 			 * System V defines rmdir to return EEXIST,
4261 			 * not ENOTEMPTY, if the directory is not
4262 			 * empty.  A System V NFS server needs to map
4263 			 * NFS4ERR_EXIST to NFS4ERR_NOTEMPTY to
4264 			 * transmit over the wire.
4265 			 */
4266 			if ((error = VOP_RMDIR(dvp, name, rootdir, cs->cr,
4267 			    NULL, 0)) == EEXIST)
4268 				error = ENOTEMPTY;
4269 		}
4270 	} else {
4271 		if ((error = VOP_REMOVE(dvp, name, cs->cr, NULL, 0)) == 0 &&
4272 		    fp != NULL) {
4273 			struct vattr va;
4274 			vnode_t *tvp;
4275 
4276 			rfs4_dbe_lock(fp->rf_dbe);
4277 			tvp = fp->rf_vp;
4278 			if (tvp)
4279 				VN_HOLD(tvp);
4280 			rfs4_dbe_unlock(fp->rf_dbe);
4281 
4282 			if (tvp) {
4283 				/*
4284 				 * This is va_seq safe because we are not
4285 				 * manipulating dvp.
4286 				 */
4287 				va.va_mask = AT_NLINK;
4288 				if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4289 				    va.va_nlink == 0) {
4290 					/* Remove state on file remove */
4291 					if (in_crit) {
4292 						nbl_end_crit(vp);
4293 						in_crit = 0;
4294 					}
4295 					rfs4_close_all_state(fp);
4296 				}
4297 				VN_RELE(tvp);
4298 			}
4299 		}
4300 	}
4301 
4302 	if (in_crit)
4303 		nbl_end_crit(vp);
4304 	VN_RELE(vp);
4305 
4306 	if (fp) {
4307 		rfs4_clear_dont_grant(fp);
4308 		rfs4_file_rele(fp);
4309 	}
4310 	if (nm != name)
4311 		kmem_free(name, MAXPATHLEN + 1);
4312 	kmem_free(nm, len);
4313 
4314 	if (error) {
4315 		*cs->statusp = resp->status = puterrno4(error);
4316 		goto out;
4317 	}
4318 
4319 	/*
4320 	 * Get the initial "after" sequence number, if it fails, set to zero
4321 	 */
4322 	idva.va_mask = AT_SEQ;
4323 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
4324 		idva.va_seq = 0;
4325 
4326 	/*
4327 	 * Force modified data and metadata out to stable storage.
4328 	 */
4329 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
4330 
4331 	/*
4332 	 * Get "after" change value, if it fails, simply return the
4333 	 * before value.
4334 	 */
4335 	adva.va_mask = AT_CTIME|AT_SEQ;
4336 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
4337 		adva.va_ctime = bdva.va_ctime;
4338 		adva.va_seq = 0;
4339 	}
4340 
4341 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
4342 
4343 	/*
4344 	 * The cinfo.atomic = TRUE only if we have
4345 	 * non-zero va_seq's, and it has incremented by exactly one
4346 	 * during the VOP_REMOVE/RMDIR and it didn't change during
4347 	 * the VOP_FSYNC.
4348 	 */
4349 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
4350 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
4351 		resp->cinfo.atomic = TRUE;
4352 	else
4353 		resp->cinfo.atomic = FALSE;
4354 
4355 	*cs->statusp = resp->status = NFS4_OK;
4356 
4357 out:
4358 	DTRACE_NFSV4_2(op__remove__done, struct compound_state *, cs,
4359 	    REMOVE4res *, resp);
4360 }
4361 
4362 /*
4363  * rename: args: SAVED_FH: from directory, CURRENT_FH: target directory,
4364  *		oldname and newname.
4365  *	res: status. If success - CURRENT_FH unchanged, return change_info
4366  *		for both from and target directories.
4367  */
4368 /* ARGSUSED */
4369 static void
4370 rfs4_op_rename(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4371     struct compound_state *cs)
4372 {
4373 	RENAME4args *args = &argop->nfs_argop4_u.oprename;
4374 	RENAME4res *resp = &resop->nfs_resop4_u.oprename;
4375 	int error;
4376 	vnode_t *odvp;
4377 	vnode_t *ndvp;
4378 	vnode_t *srcvp, *targvp;
4379 	struct vattr obdva, oidva, oadva;
4380 	struct vattr nbdva, nidva, nadva;
4381 	char *onm, *nnm;
4382 	uint_t olen, nlen;
4383 	rfs4_file_t *fp, *sfp;
4384 	int in_crit_src, in_crit_targ;
4385 	int fp_rele_grant_hold, sfp_rele_grant_hold;
4386 	bslabel_t *clabel;
4387 	struct sockaddr *ca;
4388 	char *converted_onm = NULL;
4389 	char *converted_nnm = NULL;
4390 	nfsstat4 status;
4391 
4392 	DTRACE_NFSV4_2(op__rename__start, struct compound_state *, cs,
4393 	    RENAME4args *, args);
4394 
4395 	fp = sfp = NULL;
4396 	srcvp = targvp = NULL;
4397 	in_crit_src = in_crit_targ = 0;
4398 	fp_rele_grant_hold = sfp_rele_grant_hold = 0;
4399 
4400 	/* CURRENT_FH: target directory */
4401 	ndvp = cs->vp;
4402 	if (ndvp == NULL) {
4403 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4404 		goto out;
4405 	}
4406 
4407 	/* SAVED_FH: from directory */
4408 	odvp = cs->saved_vp;
4409 	if (odvp == NULL) {
4410 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4411 		goto out;
4412 	}
4413 
4414 	if (cs->access == CS_ACCESS_DENIED) {
4415 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4416 		goto out;
4417 	}
4418 
4419 	/*
4420 	 * If there is an unshared filesystem mounted on this vnode,
4421 	 * do not allow to rename objects in this directory.
4422 	 */
4423 	if (vn_ismntpt(odvp)) {
4424 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4425 		goto out;
4426 	}
4427 
4428 	/*
4429 	 * If there is an unshared filesystem mounted on this vnode,
4430 	 * do not allow to rename to this directory.
4431 	 */
4432 	if (vn_ismntpt(ndvp)) {
4433 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4434 		goto out;
4435 	}
4436 
4437 	if (odvp->v_type != VDIR || ndvp->v_type != VDIR) {
4438 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4439 		goto out;
4440 	}
4441 
4442 	if (cs->saved_exi != cs->exi) {
4443 		*cs->statusp = resp->status = NFS4ERR_XDEV;
4444 		goto out;
4445 	}
4446 
4447 	status = utf8_dir_verify(&args->oldname);
4448 	if (status != NFS4_OK) {
4449 		*cs->statusp = resp->status = status;
4450 		goto out;
4451 	}
4452 
4453 	status = utf8_dir_verify(&args->newname);
4454 	if (status != NFS4_OK) {
4455 		*cs->statusp = resp->status = status;
4456 		goto out;
4457 	}
4458 
4459 	onm = utf8_to_fn(&args->oldname, &olen, NULL);
4460 	if (onm == NULL) {
4461 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4462 		goto out;
4463 	}
4464 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4465 	nlen = MAXPATHLEN + 1;
4466 	converted_onm = nfscmd_convname(ca, cs->exi, onm, NFSCMD_CONV_INBOUND,
4467 	    nlen);
4468 
4469 	if (converted_onm == NULL) {
4470 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4471 		kmem_free(onm, olen);
4472 		goto out;
4473 	}
4474 
4475 	nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4476 	if (nnm == NULL) {
4477 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4478 		if (onm != converted_onm)
4479 			kmem_free(converted_onm, MAXPATHLEN + 1);
4480 		kmem_free(onm, olen);
4481 		goto out;
4482 	}
4483 	converted_nnm = nfscmd_convname(ca, cs->exi, nnm, NFSCMD_CONV_INBOUND,
4484 	    MAXPATHLEN  + 1);
4485 
4486 	if (converted_nnm == NULL) {
4487 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4488 		kmem_free(nnm, nlen);
4489 		nnm = NULL;
4490 		if (onm != converted_onm)
4491 			kmem_free(converted_onm, MAXPATHLEN + 1);
4492 		kmem_free(onm, olen);
4493 		goto out;
4494 	}
4495 
4496 
4497 	if (olen > MAXNAMELEN || nlen > MAXNAMELEN) {
4498 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4499 		kmem_free(onm, olen);
4500 		kmem_free(nnm, nlen);
4501 		goto out;
4502 	}
4503 
4504 
4505 	if (rdonly4(cs->exi, cs->vp, req)) {
4506 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4507 		if (onm != converted_onm)
4508 			kmem_free(converted_onm, MAXPATHLEN + 1);
4509 		kmem_free(onm, olen);
4510 		if (nnm != converted_nnm)
4511 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4512 		kmem_free(nnm, nlen);
4513 		goto out;
4514 	}
4515 
4516 	/* check label of the target dir */
4517 	if (is_system_labeled()) {
4518 		ASSERT(req->rq_label != NULL);
4519 		clabel = req->rq_label;
4520 		DTRACE_PROBE2(tx__rfs4__log__info__oprename__clabel, char *,
4521 		    "got client label from request(1)",
4522 		    struct svc_req *, req);
4523 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4524 			if (!do_rfs_label_check(clabel, ndvp,
4525 			    EQUALITY_CHECK, cs->exi)) {
4526 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4527 				goto err_out;
4528 			}
4529 		}
4530 	}
4531 
4532 	/*
4533 	 * Is the source a file and have a delegation?
4534 	 * We don't need to acquire va_seq before these lookups, if
4535 	 * it causes an update, cinfo.before will not match, which will
4536 	 * trigger a cache flush even if atomic is TRUE.
4537 	 */
4538 	if (sfp = rfs4_lookup_and_findfile(odvp, converted_onm, &srcvp,
4539 	    &error, cs->cr)) {
4540 		if (rfs4_check_delegated_byfp(FWRITE, sfp, TRUE, TRUE, TRUE,
4541 		    NULL)) {
4542 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4543 			goto err_out;
4544 		}
4545 	}
4546 
4547 	if (srcvp == NULL) {
4548 		*cs->statusp = resp->status = puterrno4(error);
4549 		if (onm != converted_onm)
4550 			kmem_free(converted_onm, MAXPATHLEN + 1);
4551 		kmem_free(onm, olen);
4552 		if (nnm != converted_nnm)
4553 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4554 		kmem_free(nnm, nlen);
4555 		goto out;
4556 	}
4557 
4558 	sfp_rele_grant_hold = 1;
4559 
4560 	/* Does the destination exist and a file and have a delegation? */
4561 	if (fp = rfs4_lookup_and_findfile(ndvp, converted_nnm, &targvp,
4562 	    NULL, cs->cr)) {
4563 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4564 		    NULL)) {
4565 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4566 			goto err_out;
4567 		}
4568 	}
4569 	fp_rele_grant_hold = 1;
4570 
4571 
4572 	/* Check for NBMAND lock on both source and target */
4573 	if (nbl_need_check(srcvp)) {
4574 		nbl_start_crit(srcvp, RW_READER);
4575 		in_crit_src = 1;
4576 		if (nbl_conflict(srcvp, NBL_RENAME, 0, 0, 0, NULL)) {
4577 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4578 			goto err_out;
4579 		}
4580 	}
4581 
4582 	if (targvp && nbl_need_check(targvp)) {
4583 		nbl_start_crit(targvp, RW_READER);
4584 		in_crit_targ = 1;
4585 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
4586 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4587 			goto err_out;
4588 		}
4589 	}
4590 
4591 	/* Get source "before" change value */
4592 	obdva.va_mask = AT_CTIME|AT_SEQ;
4593 	error = VOP_GETATTR(odvp, &obdva, 0, cs->cr, NULL);
4594 	if (!error) {
4595 		nbdva.va_mask = AT_CTIME|AT_SEQ;
4596 		error = VOP_GETATTR(ndvp, &nbdva, 0, cs->cr, NULL);
4597 	}
4598 	if (error) {
4599 		*cs->statusp = resp->status = puterrno4(error);
4600 		goto err_out;
4601 	}
4602 
4603 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.before, obdva.va_ctime)
4604 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.before, nbdva.va_ctime)
4605 
4606 	if ((error = VOP_RENAME(odvp, converted_onm, ndvp, converted_nnm,
4607 	    cs->cr, NULL, 0)) == 0 && fp != NULL) {
4608 		struct vattr va;
4609 		vnode_t *tvp;
4610 
4611 		rfs4_dbe_lock(fp->rf_dbe);
4612 		tvp = fp->rf_vp;
4613 		if (tvp)
4614 			VN_HOLD(tvp);
4615 		rfs4_dbe_unlock(fp->rf_dbe);
4616 
4617 		if (tvp) {
4618 			va.va_mask = AT_NLINK;
4619 			if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4620 			    va.va_nlink == 0) {
4621 				/* The file is gone and so should the state */
4622 				if (in_crit_targ) {
4623 					nbl_end_crit(targvp);
4624 					in_crit_targ = 0;
4625 				}
4626 				rfs4_close_all_state(fp);
4627 			}
4628 			VN_RELE(tvp);
4629 		}
4630 	}
4631 	if (error == 0)
4632 		vn_renamepath(ndvp, srcvp, nnm, nlen - 1);
4633 
4634 	if (in_crit_src)
4635 		nbl_end_crit(srcvp);
4636 	if (srcvp)
4637 		VN_RELE(srcvp);
4638 	if (in_crit_targ)
4639 		nbl_end_crit(targvp);
4640 	if (targvp)
4641 		VN_RELE(targvp);
4642 
4643 	if (sfp) {
4644 		rfs4_clear_dont_grant(sfp);
4645 		rfs4_file_rele(sfp);
4646 	}
4647 	if (fp) {
4648 		rfs4_clear_dont_grant(fp);
4649 		rfs4_file_rele(fp);
4650 	}
4651 
4652 	if (converted_onm != onm)
4653 		kmem_free(converted_onm, MAXPATHLEN + 1);
4654 	kmem_free(onm, olen);
4655 	if (converted_nnm != nnm)
4656 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4657 	kmem_free(nnm, nlen);
4658 
4659 	/*
4660 	 * Get the initial "after" sequence number, if it fails, set to zero
4661 	 */
4662 	oidva.va_mask = AT_SEQ;
4663 	if (VOP_GETATTR(odvp, &oidva, 0, cs->cr, NULL))
4664 		oidva.va_seq = 0;
4665 
4666 	nidva.va_mask = AT_SEQ;
4667 	if (VOP_GETATTR(ndvp, &nidva, 0, cs->cr, NULL))
4668 		nidva.va_seq = 0;
4669 
4670 	/*
4671 	 * Force modified data and metadata out to stable storage.
4672 	 */
4673 	(void) VOP_FSYNC(odvp, 0, cs->cr, NULL);
4674 	(void) VOP_FSYNC(ndvp, 0, cs->cr, NULL);
4675 
4676 	if (error) {
4677 		*cs->statusp = resp->status = puterrno4(error);
4678 		goto out;
4679 	}
4680 
4681 	/*
4682 	 * Get "after" change values, if it fails, simply return the
4683 	 * before value.
4684 	 */
4685 	oadva.va_mask = AT_CTIME|AT_SEQ;
4686 	if (VOP_GETATTR(odvp, &oadva, 0, cs->cr, NULL)) {
4687 		oadva.va_ctime = obdva.va_ctime;
4688 		oadva.va_seq = 0;
4689 	}
4690 
4691 	nadva.va_mask = AT_CTIME|AT_SEQ;
4692 	if (VOP_GETATTR(odvp, &nadva, 0, cs->cr, NULL)) {
4693 		nadva.va_ctime = nbdva.va_ctime;
4694 		nadva.va_seq = 0;
4695 	}
4696 
4697 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.after, oadva.va_ctime)
4698 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.after, nadva.va_ctime)
4699 
4700 	/*
4701 	 * The cinfo.atomic = TRUE only if we have
4702 	 * non-zero va_seq's, and it has incremented by exactly one
4703 	 * during the VOP_RENAME and it didn't change during the VOP_FSYNC.
4704 	 */
4705 	if (obdva.va_seq && oidva.va_seq && oadva.va_seq &&
4706 	    oidva.va_seq == (obdva.va_seq + 1) && oidva.va_seq == oadva.va_seq)
4707 		resp->source_cinfo.atomic = TRUE;
4708 	else
4709 		resp->source_cinfo.atomic = FALSE;
4710 
4711 	if (nbdva.va_seq && nidva.va_seq && nadva.va_seq &&
4712 	    nidva.va_seq == (nbdva.va_seq + 1) && nidva.va_seq == nadva.va_seq)
4713 		resp->target_cinfo.atomic = TRUE;
4714 	else
4715 		resp->target_cinfo.atomic = FALSE;
4716 
4717 #ifdef	VOLATILE_FH_TEST
4718 	{
4719 	extern void add_volrnm_fh(struct exportinfo *, vnode_t *);
4720 
4721 	/*
4722 	 * Add the renamed file handle to the volatile rename list
4723 	 */
4724 	if (cs->exi->exi_export.ex_flags & EX_VOLRNM) {
4725 		/* file handles may expire on rename */
4726 		vnode_t *vp;
4727 
4728 		nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4729 		/*
4730 		 * Already know that nnm will be a valid string
4731 		 */
4732 		error = VOP_LOOKUP(ndvp, nnm, &vp, NULL, 0, NULL, cs->cr,
4733 		    NULL, NULL, NULL);
4734 		kmem_free(nnm, nlen);
4735 		if (!error) {
4736 			add_volrnm_fh(cs->exi, vp);
4737 			VN_RELE(vp);
4738 		}
4739 	}
4740 	}
4741 #endif	/* VOLATILE_FH_TEST */
4742 
4743 	*cs->statusp = resp->status = NFS4_OK;
4744 out:
4745 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4746 	    RENAME4res *, resp);
4747 	return;
4748 
4749 err_out:
4750 	if (onm != converted_onm)
4751 		kmem_free(converted_onm, MAXPATHLEN + 1);
4752 	if (onm != NULL)
4753 		kmem_free(onm, olen);
4754 	if (nnm != converted_nnm)
4755 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4756 	if (nnm != NULL)
4757 		kmem_free(nnm, nlen);
4758 
4759 	if (in_crit_src) nbl_end_crit(srcvp);
4760 	if (in_crit_targ) nbl_end_crit(targvp);
4761 	if (targvp) VN_RELE(targvp);
4762 	if (srcvp) VN_RELE(srcvp);
4763 	if (sfp) {
4764 		if (sfp_rele_grant_hold) rfs4_clear_dont_grant(sfp);
4765 		rfs4_file_rele(sfp);
4766 	}
4767 	if (fp) {
4768 		if (fp_rele_grant_hold) rfs4_clear_dont_grant(fp);
4769 		rfs4_file_rele(fp);
4770 	}
4771 
4772 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4773 	    RENAME4res *, resp);
4774 }
4775 
4776 /* ARGSUSED */
4777 static void
4778 rfs4_op_renew(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4779     struct compound_state *cs)
4780 {
4781 	RENEW4args *args = &argop->nfs_argop4_u.oprenew;
4782 	RENEW4res *resp = &resop->nfs_resop4_u.oprenew;
4783 	rfs4_client_t *cp;
4784 
4785 	DTRACE_NFSV4_2(op__renew__start, struct compound_state *, cs,
4786 	    RENEW4args *, args);
4787 
4788 	if ((cp = rfs4_findclient_by_id(args->clientid, FALSE)) == NULL) {
4789 		*cs->statusp = resp->status =
4790 		    rfs4_check_clientid(&args->clientid, 0);
4791 		goto out;
4792 	}
4793 
4794 	if (rfs4_lease_expired(cp)) {
4795 		rfs4_client_rele(cp);
4796 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
4797 		goto out;
4798 	}
4799 
4800 	rfs4_update_lease(cp);
4801 
4802 	mutex_enter(cp->rc_cbinfo.cb_lock);
4803 	if (cp->rc_cbinfo.cb_notified_of_cb_path_down == FALSE) {
4804 		cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
4805 		*cs->statusp = resp->status = NFS4ERR_CB_PATH_DOWN;
4806 	} else {
4807 		*cs->statusp = resp->status = NFS4_OK;
4808 	}
4809 	mutex_exit(cp->rc_cbinfo.cb_lock);
4810 
4811 	rfs4_client_rele(cp);
4812 
4813 out:
4814 	DTRACE_NFSV4_2(op__renew__done, struct compound_state *, cs,
4815 	    RENEW4res *, resp);
4816 }
4817 
4818 /* ARGSUSED */
4819 static void
4820 rfs4_op_restorefh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
4821     struct compound_state *cs)
4822 {
4823 	RESTOREFH4res *resp = &resop->nfs_resop4_u.oprestorefh;
4824 
4825 	DTRACE_NFSV4_1(op__restorefh__start, struct compound_state *, cs);
4826 
4827 	/* No need to check cs->access - we are not accessing any object */
4828 	if ((cs->saved_vp == NULL) || (cs->saved_fh.nfs_fh4_val == NULL)) {
4829 		*cs->statusp = resp->status = NFS4ERR_RESTOREFH;
4830 		goto out;
4831 	}
4832 	if (cs->vp != NULL) {
4833 		VN_RELE(cs->vp);
4834 	}
4835 	cs->vp = cs->saved_vp;
4836 	cs->saved_vp = NULL;
4837 	cs->exi = cs->saved_exi;
4838 	nfs_fh4_copy(&cs->saved_fh, &cs->fh);
4839 	*cs->statusp = resp->status = NFS4_OK;
4840 	cs->deleg = FALSE;
4841 
4842 out:
4843 	DTRACE_NFSV4_2(op__restorefh__done, struct compound_state *, cs,
4844 	    RESTOREFH4res *, resp);
4845 }
4846 
4847 /* ARGSUSED */
4848 static void
4849 rfs4_op_savefh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4850     struct compound_state *cs)
4851 {
4852 	SAVEFH4res *resp = &resop->nfs_resop4_u.opsavefh;
4853 
4854 	DTRACE_NFSV4_1(op__savefh__start, struct compound_state *, cs);
4855 
4856 	/* No need to check cs->access - we are not accessing any object */
4857 	if (cs->vp == NULL) {
4858 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4859 		goto out;
4860 	}
4861 	if (cs->saved_vp != NULL) {
4862 		VN_RELE(cs->saved_vp);
4863 	}
4864 	cs->saved_vp = cs->vp;
4865 	VN_HOLD(cs->saved_vp);
4866 	cs->saved_exi = cs->exi;
4867 	/*
4868 	 * since SAVEFH is fairly rare, don't alloc space for its fh
4869 	 * unless necessary.
4870 	 */
4871 	if (cs->saved_fh.nfs_fh4_val == NULL) {
4872 		cs->saved_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP);
4873 	}
4874 	nfs_fh4_copy(&cs->fh, &cs->saved_fh);
4875 	*cs->statusp = resp->status = NFS4_OK;
4876 
4877 out:
4878 	DTRACE_NFSV4_2(op__savefh__done, struct compound_state *, cs,
4879 	    SAVEFH4res *, resp);
4880 }
4881 
4882 /*
4883  * rfs4_verify_attr is called when nfsv4 Setattr failed, but we wish to
4884  * return the bitmap of attrs that were set successfully. It is also
4885  * called by Verify/Nverify to test the vattr/vfsstat attrs. It should
4886  * always be called only after rfs4_do_set_attrs().
4887  *
4888  * Verify that the attributes are same as the expected ones. sargp->vap
4889  * and sargp->sbp contain the input attributes as translated from fattr4.
4890  *
4891  * This function verifies only the attrs that correspond to a vattr or
4892  * vfsstat struct. That is because of the extra step needed to get the
4893  * corresponding system structs. Other attributes have already been set or
4894  * verified by do_rfs4_set_attrs.
4895  *
4896  * Return 0 if all attrs match, -1 if some don't, error if error processing.
4897  */
4898 static int
4899 rfs4_verify_attr(struct nfs4_svgetit_arg *sargp,
4900     bitmap4 *resp, struct nfs4_ntov_table *ntovp)
4901 {
4902 	int error, ret_error = 0;
4903 	int i, k;
4904 	uint_t sva_mask = sargp->vap->va_mask;
4905 	uint_t vbit;
4906 	union nfs4_attr_u *na;
4907 	uint8_t *amap;
4908 	bool_t getsb = ntovp->vfsstat;
4909 
4910 	if (sva_mask != 0) {
4911 		/*
4912 		 * Okay to overwrite sargp->vap because we verify based
4913 		 * on the incoming values.
4914 		 */
4915 		ret_error = VOP_GETATTR(sargp->cs->vp, sargp->vap, 0,
4916 		    sargp->cs->cr, NULL);
4917 		if (ret_error) {
4918 			if (resp == NULL)
4919 				return (ret_error);
4920 			/*
4921 			 * Must return bitmap of successful attrs
4922 			 */
4923 			sva_mask = 0;	/* to prevent checking vap later */
4924 		} else {
4925 			/*
4926 			 * Some file systems clobber va_mask. it is probably
4927 			 * wrong of them to do so, nonethless we practice
4928 			 * defensive coding.
4929 			 * See bug id 4276830.
4930 			 */
4931 			sargp->vap->va_mask = sva_mask;
4932 		}
4933 	}
4934 
4935 	if (getsb) {
4936 		/*
4937 		 * Now get the superblock and loop on the bitmap, as there is
4938 		 * no simple way of translating from superblock to bitmap4.
4939 		 */
4940 		ret_error = VFS_STATVFS(sargp->cs->vp->v_vfsp, sargp->sbp);
4941 		if (ret_error) {
4942 			if (resp == NULL)
4943 				goto errout;
4944 			getsb = FALSE;
4945 		}
4946 	}
4947 
4948 	/*
4949 	 * Now loop and verify each attribute which getattr returned
4950 	 * whether it's the same as the input.
4951 	 */
4952 	if (resp == NULL && !getsb && (sva_mask == 0))
4953 		goto errout;
4954 
4955 	na = ntovp->na;
4956 	amap = ntovp->amap;
4957 	k = 0;
4958 	for (i = 0; i < ntovp->attrcnt; i++, na++, amap++) {
4959 		k = *amap;
4960 		ASSERT(nfs4_ntov_map[k].nval == k);
4961 		vbit = nfs4_ntov_map[k].vbit;
4962 
4963 		/*
4964 		 * If vattr attribute but VOP_GETATTR failed, or it's
4965 		 * superblock attribute but VFS_STATVFS failed, skip
4966 		 */
4967 		if (vbit) {
4968 			if ((vbit & sva_mask) == 0)
4969 				continue;
4970 		} else if (!(getsb && nfs4_ntov_map[k].vfsstat)) {
4971 			continue;
4972 		}
4973 		error = (*nfs4_ntov_map[k].sv_getit)(NFS4ATTR_VERIT, sargp, na);
4974 		if (resp != NULL) {
4975 			if (error)
4976 				ret_error = -1;	/* not all match */
4977 			else	/* update response bitmap */
4978 				*resp |= nfs4_ntov_map[k].fbit;
4979 			continue;
4980 		}
4981 		if (error) {
4982 			ret_error = -1;	/* not all match */
4983 			break;
4984 		}
4985 	}
4986 errout:
4987 	return (ret_error);
4988 }
4989 
4990 /*
4991  * Decode the attribute to be set/verified. If the attr requires a sys op
4992  * (VOP_GETATTR, VFS_VFSSTAT), and the request is to verify, then don't
4993  * call the sv_getit function for it, because the sys op hasn't yet been done.
4994  * Return 0 for success, error code if failed.
4995  *
4996  * Note: the decoded arg is not freed here but in nfs4_ntov_table_free.
4997  */
4998 static int
4999 decode_fattr4_attr(nfs4_attr_cmd_t cmd, struct nfs4_svgetit_arg *sargp,
5000     int k, XDR *xdrp, bitmap4 *resp_bval, union nfs4_attr_u *nap)
5001 {
5002 	int error = 0;
5003 	bool_t set_later;
5004 
5005 	sargp->vap->va_mask |= nfs4_ntov_map[k].vbit;
5006 
5007 	if ((*nfs4_ntov_map[k].xfunc)(xdrp, nap)) {
5008 		set_later = nfs4_ntov_map[k].vbit || nfs4_ntov_map[k].vfsstat;
5009 		/*
5010 		 * don't verify yet if a vattr or sb dependent attr,
5011 		 * because we don't have their sys values yet.
5012 		 * Will be done later.
5013 		 */
5014 		if (! (set_later && (cmd == NFS4ATTR_VERIT))) {
5015 			/*
5016 			 * ACLs are a special case, since setting the MODE
5017 			 * conflicts with setting the ACL.  We delay setting
5018 			 * the ACL until all other attributes have been set.
5019 			 * The ACL gets set in do_rfs4_op_setattr().
5020 			 */
5021 			if (nfs4_ntov_map[k].fbit != FATTR4_ACL_MASK) {
5022 				error = (*nfs4_ntov_map[k].sv_getit)(cmd,
5023 				    sargp, nap);
5024 				if (error) {
5025 					xdr_free(nfs4_ntov_map[k].xfunc,
5026 					    (caddr_t)nap);
5027 				}
5028 			}
5029 		}
5030 	} else {
5031 #ifdef  DEBUG
5032 		cmn_err(CE_NOTE, "decode_fattr4_attr: error "
5033 		    "decoding attribute %d\n", k);
5034 #endif
5035 		error = EINVAL;
5036 	}
5037 	if (!error && resp_bval && !set_later) {
5038 		*resp_bval |= nfs4_ntov_map[k].fbit;
5039 	}
5040 
5041 	return (error);
5042 }
5043 
5044 /*
5045  * Set vattr based on incoming fattr4 attrs - used by setattr.
5046  * Set response mask. Ignore any values that are not writable vattr attrs.
5047  */
5048 static nfsstat4
5049 do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5050     struct nfs4_svgetit_arg *sargp, struct nfs4_ntov_table *ntovp,
5051     nfs4_attr_cmd_t cmd)
5052 {
5053 	int error = 0;
5054 	int i;
5055 	char *attrs = fattrp->attrlist4;
5056 	uint32_t attrslen = fattrp->attrlist4_len;
5057 	XDR xdr;
5058 	nfsstat4 status = NFS4_OK;
5059 	vnode_t *vp = cs->vp;
5060 	union nfs4_attr_u *na;
5061 	uint8_t *amap;
5062 
5063 #ifndef lint
5064 	/*
5065 	 * Make sure that maximum attribute number can be expressed as an
5066 	 * 8 bit quantity.
5067 	 */
5068 	ASSERT(NFS4_MAXNUM_ATTRS <= (UINT8_MAX + 1));
5069 #endif
5070 
5071 	if (vp == NULL) {
5072 		if (resp)
5073 			*resp = 0;
5074 		return (NFS4ERR_NOFILEHANDLE);
5075 	}
5076 	if (cs->access == CS_ACCESS_DENIED) {
5077 		if (resp)
5078 			*resp = 0;
5079 		return (NFS4ERR_ACCESS);
5080 	}
5081 
5082 	sargp->op = cmd;
5083 	sargp->cs = cs;
5084 	sargp->flag = 0;	/* may be set later */
5085 	sargp->vap->va_mask = 0;
5086 	sargp->rdattr_error = NFS4_OK;
5087 	sargp->rdattr_error_req = FALSE;
5088 	/* sargp->sbp is set by the caller */
5089 
5090 	xdrmem_create(&xdr, attrs, attrslen, XDR_DECODE);
5091 
5092 	na = ntovp->na;
5093 	amap = ntovp->amap;
5094 
5095 	/*
5096 	 * The following loop iterates on the nfs4_ntov_map checking
5097 	 * if the fbit is set in the requested bitmap.
5098 	 * If set then we process the arguments using the
5099 	 * rfs4_fattr4 conversion functions to populate the setattr
5100 	 * vattr and va_mask. Any settable attrs that are not using vattr
5101 	 * will be set in this loop.
5102 	 */
5103 	for (i = 0; i < nfs4_ntov_map_size; i++) {
5104 		if (!(fattrp->attrmask & nfs4_ntov_map[i].fbit)) {
5105 			continue;
5106 		}
5107 		/*
5108 		 * If setattr, must be a writable attr.
5109 		 * If verify/nverify, must be a readable attr.
5110 		 */
5111 		if ((error = (*nfs4_ntov_map[i].sv_getit)(
5112 		    NFS4ATTR_SUPPORTED, sargp, NULL)) != 0) {
5113 			/*
5114 			 * Client tries to set/verify an
5115 			 * unsupported attribute, tries to set
5116 			 * a read only attr or verify a write
5117 			 * only one - error!
5118 			 */
5119 			break;
5120 		}
5121 		/*
5122 		 * Decode the attribute to set/verify
5123 		 */
5124 		error = decode_fattr4_attr(cmd, sargp, nfs4_ntov_map[i].nval,
5125 		    &xdr, resp ? resp : NULL, na);
5126 		if (error)
5127 			break;
5128 		*amap++ = (uint8_t)nfs4_ntov_map[i].nval;
5129 		na++;
5130 		(ntovp->attrcnt)++;
5131 		if (nfs4_ntov_map[i].vfsstat)
5132 			ntovp->vfsstat = TRUE;
5133 	}
5134 
5135 	if (error != 0)
5136 		status = (error == ENOTSUP ? NFS4ERR_ATTRNOTSUPP :
5137 		    puterrno4(error));
5138 	/* xdrmem_destroy(&xdrs); */	/* NO-OP */
5139 	return (status);
5140 }
5141 
5142 static nfsstat4
5143 do_rfs4_op_setattr(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5144     stateid4 *stateid)
5145 {
5146 	int error = 0;
5147 	struct nfs4_svgetit_arg sarg;
5148 	bool_t trunc;
5149 
5150 	nfsstat4 status = NFS4_OK;
5151 	cred_t *cr = cs->cr;
5152 	vnode_t *vp = cs->vp;
5153 	struct nfs4_ntov_table ntov;
5154 	struct statvfs64 sb;
5155 	struct vattr bva;
5156 	struct flock64 bf;
5157 	int in_crit = 0;
5158 	uint_t saved_mask = 0;
5159 	caller_context_t ct;
5160 
5161 	*resp = 0;
5162 	sarg.sbp = &sb;
5163 	sarg.is_referral = B_FALSE;
5164 	nfs4_ntov_table_init(&ntov);
5165 	status = do_rfs4_set_attrs(resp, fattrp, cs, &sarg, &ntov,
5166 	    NFS4ATTR_SETIT);
5167 	if (status != NFS4_OK) {
5168 		/*
5169 		 * failed set attrs
5170 		 */
5171 		goto done;
5172 	}
5173 	if ((sarg.vap->va_mask == 0) &&
5174 	    (! (fattrp->attrmask & FATTR4_ACL_MASK))) {
5175 		/*
5176 		 * no further work to be done
5177 		 */
5178 		goto done;
5179 	}
5180 
5181 	/*
5182 	 * If we got a request to set the ACL and the MODE, only
5183 	 * allow changing VSUID, VSGID, and VSVTX.  Attempting
5184 	 * to change any other bits, along with setting an ACL,
5185 	 * gives NFS4ERR_INVAL.
5186 	 */
5187 	if ((fattrp->attrmask & FATTR4_ACL_MASK) &&
5188 	    (fattrp->attrmask & FATTR4_MODE_MASK)) {
5189 		vattr_t va;
5190 
5191 		va.va_mask = AT_MODE;
5192 		error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
5193 		if (error) {
5194 			status = puterrno4(error);
5195 			goto done;
5196 		}
5197 		if ((sarg.vap->va_mode ^ va.va_mode) &
5198 		    ~(VSUID | VSGID | VSVTX)) {
5199 			status = NFS4ERR_INVAL;
5200 			goto done;
5201 		}
5202 	}
5203 
5204 	/* Check stateid only if size has been set */
5205 	if (sarg.vap->va_mask & AT_SIZE) {
5206 		trunc = (sarg.vap->va_size == 0);
5207 		status = rfs4_check_stateid(FWRITE, cs->vp, stateid,
5208 		    trunc, &cs->deleg, sarg.vap->va_mask & AT_SIZE, &ct);
5209 		if (status != NFS4_OK)
5210 			goto done;
5211 	} else {
5212 		ct.cc_sysid = 0;
5213 		ct.cc_pid = 0;
5214 		ct.cc_caller_id = nfs4_srv_caller_id;
5215 		ct.cc_flags = CC_DONTBLOCK;
5216 	}
5217 
5218 	/* XXX start of possible race with delegations */
5219 
5220 	/*
5221 	 * We need to specially handle size changes because it is
5222 	 * possible for the client to create a file with read-only
5223 	 * modes, but with the file opened for writing. If the client
5224 	 * then tries to set the file size, e.g. ftruncate(3C),
5225 	 * fcntl(F_FREESP), the normal access checking done in
5226 	 * VOP_SETATTR would prevent the client from doing it even though
5227 	 * it should be allowed to do so.  To get around this, we do the
5228 	 * access checking for ourselves and use VOP_SPACE which doesn't
5229 	 * do the access checking.
5230 	 * Also the client should not be allowed to change the file
5231 	 * size if there is a conflicting non-blocking mandatory lock in
5232 	 * the region of the change.
5233 	 */
5234 	if (vp->v_type == VREG && (sarg.vap->va_mask & AT_SIZE)) {
5235 		u_offset_t offset;
5236 		ssize_t length;
5237 
5238 		/*
5239 		 * ufs_setattr clears AT_SIZE from vap->va_mask, but
5240 		 * before returning, sarg.vap->va_mask is used to
5241 		 * generate the setattr reply bitmap.  We also clear
5242 		 * AT_SIZE below before calling VOP_SPACE.  For both
5243 		 * of these cases, the va_mask needs to be saved here
5244 		 * and restored after calling VOP_SETATTR.
5245 		 */
5246 		saved_mask = sarg.vap->va_mask;
5247 
5248 		/*
5249 		 * Check any possible conflict due to NBMAND locks.
5250 		 * Get into critical region before VOP_GETATTR, so the
5251 		 * size attribute is valid when checking conflicts.
5252 		 */
5253 		if (nbl_need_check(vp)) {
5254 			nbl_start_crit(vp, RW_READER);
5255 			in_crit = 1;
5256 		}
5257 
5258 		bva.va_mask = AT_UID|AT_SIZE;
5259 		if (error = VOP_GETATTR(vp, &bva, 0, cr, &ct)) {
5260 			status = puterrno4(error);
5261 			goto done;
5262 		}
5263 
5264 		if (in_crit) {
5265 			if (sarg.vap->va_size < bva.va_size) {
5266 				offset = sarg.vap->va_size;
5267 				length = bva.va_size - sarg.vap->va_size;
5268 			} else {
5269 				offset = bva.va_size;
5270 				length = sarg.vap->va_size - bva.va_size;
5271 			}
5272 			if (nbl_conflict(vp, NBL_WRITE, offset, length, 0,
5273 			    &ct)) {
5274 				status = NFS4ERR_LOCKED;
5275 				goto done;
5276 			}
5277 		}
5278 
5279 		if (crgetuid(cr) == bva.va_uid) {
5280 			sarg.vap->va_mask &= ~AT_SIZE;
5281 			bf.l_type = F_WRLCK;
5282 			bf.l_whence = 0;
5283 			bf.l_start = (off64_t)sarg.vap->va_size;
5284 			bf.l_len = 0;
5285 			bf.l_sysid = 0;
5286 			bf.l_pid = 0;
5287 			error = VOP_SPACE(vp, F_FREESP, &bf, FWRITE,
5288 			    (offset_t)sarg.vap->va_size, cr, &ct);
5289 		}
5290 	}
5291 
5292 	if (!error && sarg.vap->va_mask != 0)
5293 		error = VOP_SETATTR(vp, sarg.vap, sarg.flag, cr, &ct);
5294 
5295 	/* restore va_mask -- ufs_setattr clears AT_SIZE */
5296 	if (saved_mask & AT_SIZE)
5297 		sarg.vap->va_mask |= AT_SIZE;
5298 
5299 	/*
5300 	 * If an ACL was being set, it has been delayed until now,
5301 	 * in order to set the mode (via the VOP_SETATTR() above) first.
5302 	 */
5303 	if ((! error) && (fattrp->attrmask & FATTR4_ACL_MASK)) {
5304 		int i;
5305 
5306 		for (i = 0; i < NFS4_MAXNUM_ATTRS; i++)
5307 			if (ntov.amap[i] == FATTR4_ACL)
5308 				break;
5309 		if (i < NFS4_MAXNUM_ATTRS) {
5310 			error = (*nfs4_ntov_map[FATTR4_ACL].sv_getit)(
5311 			    NFS4ATTR_SETIT, &sarg, &ntov.na[i]);
5312 			if (error == 0) {
5313 				*resp |= FATTR4_ACL_MASK;
5314 			} else if (error == ENOTSUP) {
5315 				(void) rfs4_verify_attr(&sarg, resp, &ntov);
5316 				status = NFS4ERR_ATTRNOTSUPP;
5317 				goto done;
5318 			}
5319 		} else {
5320 			NFS4_DEBUG(rfs4_debug,
5321 			    (CE_NOTE, "do_rfs4_op_setattr: "
5322 			    "unable to find ACL in fattr4"));
5323 			error = EINVAL;
5324 		}
5325 	}
5326 
5327 	if (error) {
5328 		/* check if a monitor detected a delegation conflict */
5329 		if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
5330 			status = NFS4ERR_DELAY;
5331 		else
5332 			status = puterrno4(error);
5333 
5334 		/*
5335 		 * Set the response bitmap when setattr failed.
5336 		 * If VOP_SETATTR partially succeeded, test by doing a
5337 		 * VOP_GETATTR on the object and comparing the data
5338 		 * to the setattr arguments.
5339 		 */
5340 		(void) rfs4_verify_attr(&sarg, resp, &ntov);
5341 	} else {
5342 		/*
5343 		 * Force modified metadata out to stable storage.
5344 		 */
5345 		(void) VOP_FSYNC(vp, FNODSYNC, cr, &ct);
5346 		/*
5347 		 * Set response bitmap
5348 		 */
5349 		nfs4_vmask_to_nmask_set(sarg.vap->va_mask, resp);
5350 	}
5351 
5352 /* Return early and already have a NFSv4 error */
5353 done:
5354 	/*
5355 	 * Except for nfs4_vmask_to_nmask_set(), vattr --> fattr
5356 	 * conversion sets both readable and writeable NFS4 attrs
5357 	 * for AT_MTIME and AT_ATIME.  The line below masks out
5358 	 * unrequested attrs from the setattr result bitmap.  This
5359 	 * is placed after the done: label to catch the ATTRNOTSUP
5360 	 * case.
5361 	 */
5362 	*resp &= fattrp->attrmask;
5363 
5364 	if (in_crit)
5365 		nbl_end_crit(vp);
5366 
5367 	nfs4_ntov_table_free(&ntov, &sarg);
5368 
5369 	return (status);
5370 }
5371 
5372 /* ARGSUSED */
5373 static void
5374 rfs4_op_setattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5375     struct compound_state *cs)
5376 {
5377 	SETATTR4args *args = &argop->nfs_argop4_u.opsetattr;
5378 	SETATTR4res *resp = &resop->nfs_resop4_u.opsetattr;
5379 	bslabel_t *clabel;
5380 
5381 	DTRACE_NFSV4_2(op__setattr__start, struct compound_state *, cs,
5382 	    SETATTR4args *, args);
5383 
5384 	if (cs->vp == NULL) {
5385 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5386 		goto out;
5387 	}
5388 
5389 	/*
5390 	 * If there is an unshared filesystem mounted on this vnode,
5391 	 * do not allow to setattr on this vnode.
5392 	 */
5393 	if (vn_ismntpt(cs->vp)) {
5394 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5395 		goto out;
5396 	}
5397 
5398 	resp->attrsset = 0;
5399 
5400 	if (rdonly4(cs->exi, cs->vp, req)) {
5401 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5402 		goto out;
5403 	}
5404 
5405 	/* check label before setting attributes */
5406 	if (is_system_labeled()) {
5407 		ASSERT(req->rq_label != NULL);
5408 		clabel = req->rq_label;
5409 		DTRACE_PROBE2(tx__rfs4__log__info__opsetattr__clabel, char *,
5410 		    "got client label from request(1)",
5411 		    struct svc_req *, req);
5412 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
5413 			if (!do_rfs_label_check(clabel, cs->vp,
5414 			    EQUALITY_CHECK, cs->exi)) {
5415 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
5416 				goto out;
5417 			}
5418 		}
5419 	}
5420 
5421 	*cs->statusp = resp->status =
5422 	    do_rfs4_op_setattr(&resp->attrsset, &args->obj_attributes, cs,
5423 	    &args->stateid);
5424 
5425 out:
5426 	DTRACE_NFSV4_2(op__setattr__done, struct compound_state *, cs,
5427 	    SETATTR4res *, resp);
5428 }
5429 
5430 /* ARGSUSED */
5431 static void
5432 rfs4_op_verify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5433     struct compound_state *cs)
5434 {
5435 	/*
5436 	 * verify and nverify are exactly the same, except that nverify
5437 	 * succeeds when some argument changed, and verify succeeds when
5438 	 * when none changed.
5439 	 */
5440 
5441 	VERIFY4args  *args = &argop->nfs_argop4_u.opverify;
5442 	VERIFY4res *resp = &resop->nfs_resop4_u.opverify;
5443 
5444 	int error;
5445 	struct nfs4_svgetit_arg sarg;
5446 	struct statvfs64 sb;
5447 	struct nfs4_ntov_table ntov;
5448 
5449 	DTRACE_NFSV4_2(op__verify__start, struct compound_state *, cs,
5450 	    VERIFY4args *, args);
5451 
5452 	if (cs->vp == NULL) {
5453 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5454 		goto out;
5455 	}
5456 
5457 	sarg.sbp = &sb;
5458 	sarg.is_referral = B_FALSE;
5459 	nfs4_ntov_table_init(&ntov);
5460 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5461 	    &sarg, &ntov, NFS4ATTR_VERIT);
5462 	if (resp->status != NFS4_OK) {
5463 		/*
5464 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5465 		 * so could return -1 for "no match".
5466 		 */
5467 		if (resp->status == -1)
5468 			resp->status = NFS4ERR_NOT_SAME;
5469 		goto done;
5470 	}
5471 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5472 	switch (error) {
5473 	case 0:
5474 		resp->status = NFS4_OK;
5475 		break;
5476 	case -1:
5477 		resp->status = NFS4ERR_NOT_SAME;
5478 		break;
5479 	default:
5480 		resp->status = puterrno4(error);
5481 		break;
5482 	}
5483 done:
5484 	*cs->statusp = resp->status;
5485 	nfs4_ntov_table_free(&ntov, &sarg);
5486 out:
5487 	DTRACE_NFSV4_2(op__verify__done, struct compound_state *, cs,
5488 	    VERIFY4res *, resp);
5489 }
5490 
5491 /* ARGSUSED */
5492 static void
5493 rfs4_op_nverify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5494     struct compound_state *cs)
5495 {
5496 	/*
5497 	 * verify and nverify are exactly the same, except that nverify
5498 	 * succeeds when some argument changed, and verify succeeds when
5499 	 * when none changed.
5500 	 */
5501 
5502 	NVERIFY4args  *args = &argop->nfs_argop4_u.opnverify;
5503 	NVERIFY4res *resp = &resop->nfs_resop4_u.opnverify;
5504 
5505 	int error;
5506 	struct nfs4_svgetit_arg sarg;
5507 	struct statvfs64 sb;
5508 	struct nfs4_ntov_table ntov;
5509 
5510 	DTRACE_NFSV4_2(op__nverify__start, struct compound_state *, cs,
5511 	    NVERIFY4args *, args);
5512 
5513 	if (cs->vp == NULL) {
5514 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5515 		DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5516 		    NVERIFY4res *, resp);
5517 		return;
5518 	}
5519 	sarg.sbp = &sb;
5520 	sarg.is_referral = B_FALSE;
5521 	nfs4_ntov_table_init(&ntov);
5522 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5523 	    &sarg, &ntov, NFS4ATTR_VERIT);
5524 	if (resp->status != NFS4_OK) {
5525 		/*
5526 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5527 		 * so could return -1 for "no match".
5528 		 */
5529 		if (resp->status == -1)
5530 			resp->status = NFS4_OK;
5531 		goto done;
5532 	}
5533 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5534 	switch (error) {
5535 	case 0:
5536 		resp->status = NFS4ERR_SAME;
5537 		break;
5538 	case -1:
5539 		resp->status = NFS4_OK;
5540 		break;
5541 	default:
5542 		resp->status = puterrno4(error);
5543 		break;
5544 	}
5545 done:
5546 	*cs->statusp = resp->status;
5547 	nfs4_ntov_table_free(&ntov, &sarg);
5548 
5549 	DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5550 	    NVERIFY4res *, resp);
5551 }
5552 
5553 /*
5554  * XXX - This should live in an NFS header file.
5555  */
5556 #define	MAX_IOVECS	12
5557 
5558 /* ARGSUSED */
5559 static void
5560 rfs4_op_write(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5561     struct compound_state *cs)
5562 {
5563 	WRITE4args *args = &argop->nfs_argop4_u.opwrite;
5564 	WRITE4res *resp = &resop->nfs_resop4_u.opwrite;
5565 	int error;
5566 	vnode_t *vp;
5567 	struct vattr bva;
5568 	u_offset_t rlimit;
5569 	struct uio uio;
5570 	struct iovec iov[MAX_IOVECS];
5571 	struct iovec *iovp;
5572 	int iovcnt;
5573 	int ioflag;
5574 	cred_t *savecred, *cr;
5575 	bool_t *deleg = &cs->deleg;
5576 	nfsstat4 stat;
5577 	int in_crit = 0;
5578 	caller_context_t ct;
5579 
5580 	DTRACE_NFSV4_2(op__write__start, struct compound_state *, cs,
5581 	    WRITE4args *, args);
5582 
5583 	vp = cs->vp;
5584 	if (vp == NULL) {
5585 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5586 		goto out;
5587 	}
5588 	if (cs->access == CS_ACCESS_DENIED) {
5589 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5590 		goto out;
5591 	}
5592 
5593 	cr = cs->cr;
5594 
5595 	if ((stat = rfs4_check_stateid(FWRITE, vp, &args->stateid, FALSE,
5596 	    deleg, TRUE, &ct)) != NFS4_OK) {
5597 		*cs->statusp = resp->status = stat;
5598 		goto out;
5599 	}
5600 
5601 	/*
5602 	 * We have to enter the critical region before calling VOP_RWLOCK
5603 	 * to avoid a deadlock with ufs.
5604 	 */
5605 	if (nbl_need_check(vp)) {
5606 		nbl_start_crit(vp, RW_READER);
5607 		in_crit = 1;
5608 		if (nbl_conflict(vp, NBL_WRITE,
5609 		    args->offset, args->data_len, 0, &ct)) {
5610 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
5611 			goto out;
5612 		}
5613 	}
5614 
5615 	bva.va_mask = AT_MODE | AT_UID;
5616 	error = VOP_GETATTR(vp, &bva, 0, cr, &ct);
5617 
5618 	/*
5619 	 * If we can't get the attributes, then we can't do the
5620 	 * right access checking.  So, we'll fail the request.
5621 	 */
5622 	if (error) {
5623 		*cs->statusp = resp->status = puterrno4(error);
5624 		goto out;
5625 	}
5626 
5627 	if (rdonly4(cs->exi, cs->vp, req)) {
5628 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5629 		goto out;
5630 	}
5631 
5632 	if (vp->v_type != VREG) {
5633 		*cs->statusp = resp->status =
5634 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
5635 		goto out;
5636 	}
5637 
5638 	if (crgetuid(cr) != bva.va_uid &&
5639 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, &ct))) {
5640 		*cs->statusp = resp->status = puterrno4(error);
5641 		goto out;
5642 	}
5643 
5644 	if (MANDLOCK(vp, bva.va_mode)) {
5645 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5646 		goto out;
5647 	}
5648 
5649 	if (args->data_len == 0) {
5650 		*cs->statusp = resp->status = NFS4_OK;
5651 		resp->count = 0;
5652 		resp->committed = args->stable;
5653 		resp->writeverf = Write4verf;
5654 		goto out;
5655 	}
5656 
5657 	if (args->mblk != NULL) {
5658 		mblk_t *m;
5659 		uint_t bytes, round_len;
5660 
5661 		iovcnt = 0;
5662 		bytes = 0;
5663 		round_len = roundup(args->data_len, BYTES_PER_XDR_UNIT);
5664 		for (m = args->mblk;
5665 		    m != NULL && bytes < round_len;
5666 		    m = m->b_cont) {
5667 			iovcnt++;
5668 			bytes += MBLKL(m);
5669 		}
5670 #ifdef DEBUG
5671 		/* should have ended on an mblk boundary */
5672 		if (bytes != round_len) {
5673 			printf("bytes=0x%x, round_len=0x%x, req len=0x%x\n",
5674 			    bytes, round_len, args->data_len);
5675 			printf("args=%p, args->mblk=%p, m=%p", (void *)args,
5676 			    (void *)args->mblk, (void *)m);
5677 			ASSERT(bytes == round_len);
5678 		}
5679 #endif
5680 		if (iovcnt <= MAX_IOVECS) {
5681 			iovp = iov;
5682 		} else {
5683 			iovp = kmem_alloc(sizeof (*iovp) * iovcnt, KM_SLEEP);
5684 		}
5685 		mblk_to_iov(args->mblk, iovcnt, iovp);
5686 	} else if (args->rlist != NULL) {
5687 		iovcnt = 1;
5688 		iovp = iov;
5689 		iovp->iov_base = (char *)((args->rlist)->u.c_daddr3);
5690 		iovp->iov_len = args->data_len;
5691 	} else {
5692 		iovcnt = 1;
5693 		iovp = iov;
5694 		iovp->iov_base = args->data_val;
5695 		iovp->iov_len = args->data_len;
5696 	}
5697 
5698 	uio.uio_iov = iovp;
5699 	uio.uio_iovcnt = iovcnt;
5700 
5701 	uio.uio_segflg = UIO_SYSSPACE;
5702 	uio.uio_extflg = UIO_COPY_DEFAULT;
5703 	uio.uio_loffset = args->offset;
5704 	uio.uio_resid = args->data_len;
5705 	uio.uio_llimit = curproc->p_fsz_ctl;
5706 	rlimit = uio.uio_llimit - args->offset;
5707 	if (rlimit < (u_offset_t)uio.uio_resid)
5708 		uio.uio_resid = (int)rlimit;
5709 
5710 	if (args->stable == UNSTABLE4)
5711 		ioflag = 0;
5712 	else if (args->stable == FILE_SYNC4)
5713 		ioflag = FSYNC;
5714 	else if (args->stable == DATA_SYNC4)
5715 		ioflag = FDSYNC;
5716 	else {
5717 		if (iovp != iov)
5718 			kmem_free(iovp, sizeof (*iovp) * iovcnt);
5719 		*cs->statusp = resp->status = NFS4ERR_INVAL;
5720 		goto out;
5721 	}
5722 
5723 	/*
5724 	 * We're changing creds because VM may fault and we need
5725 	 * the cred of the current thread to be used if quota
5726 	 * checking is enabled.
5727 	 */
5728 	savecred = curthread->t_cred;
5729 	curthread->t_cred = cr;
5730 	error = do_io(FWRITE, vp, &uio, ioflag, cr, &ct);
5731 	curthread->t_cred = savecred;
5732 
5733 	if (iovp != iov)
5734 		kmem_free(iovp, sizeof (*iovp) * iovcnt);
5735 
5736 	if (error) {
5737 		*cs->statusp = resp->status = puterrno4(error);
5738 		goto out;
5739 	}
5740 
5741 	*cs->statusp = resp->status = NFS4_OK;
5742 	resp->count = args->data_len - uio.uio_resid;
5743 
5744 	if (ioflag == 0)
5745 		resp->committed = UNSTABLE4;
5746 	else
5747 		resp->committed = FILE_SYNC4;
5748 
5749 	resp->writeverf = Write4verf;
5750 
5751 out:
5752 	if (in_crit)
5753 		nbl_end_crit(vp);
5754 
5755 	DTRACE_NFSV4_2(op__write__done, struct compound_state *, cs,
5756 	    WRITE4res *, resp);
5757 }
5758 
5759 
5760 /* XXX put in a header file */
5761 extern int	sec_svc_getcred(struct svc_req *, cred_t *,  caddr_t *, int *);
5762 
5763 void
5764 rfs4_compound(COMPOUND4args *args, COMPOUND4res *resp, struct exportinfo *exi,
5765     struct svc_req *req, cred_t *cr, int *rv)
5766 {
5767 	uint_t i;
5768 	struct compound_state cs;
5769 
5770 	if (rv != NULL)
5771 		*rv = 0;
5772 	rfs4_init_compound_state(&cs);
5773 	/*
5774 	 * Form a reply tag by copying over the reqeuest tag.
5775 	 */
5776 	resp->tag.utf8string_val =
5777 	    kmem_alloc(args->tag.utf8string_len, KM_SLEEP);
5778 	resp->tag.utf8string_len = args->tag.utf8string_len;
5779 	bcopy(args->tag.utf8string_val, resp->tag.utf8string_val,
5780 	    resp->tag.utf8string_len);
5781 
5782 	cs.statusp = &resp->status;
5783 	cs.req = req;
5784 	resp->array = NULL;
5785 	resp->array_len = 0;
5786 
5787 	/*
5788 	 * XXX for now, minorversion should be zero
5789 	 */
5790 	if (args->minorversion != NFS4_MINORVERSION) {
5791 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5792 		    &cs, COMPOUND4args *, args);
5793 		resp->status = NFS4ERR_MINOR_VERS_MISMATCH;
5794 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5795 		    &cs, COMPOUND4res *, resp);
5796 		return;
5797 	}
5798 
5799 	if (args->array_len == 0) {
5800 		resp->status = NFS4_OK;
5801 		return;
5802 	}
5803 
5804 	ASSERT(exi == NULL);
5805 	ASSERT(cr == NULL);
5806 
5807 	cr = crget();
5808 	ASSERT(cr != NULL);
5809 
5810 	if (sec_svc_getcred(req, cr, &cs.principal, &cs.nfsflavor) == 0) {
5811 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5812 		    &cs, COMPOUND4args *, args);
5813 		crfree(cr);
5814 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5815 		    &cs, COMPOUND4res *, resp);
5816 		svcerr_badcred(req->rq_xprt);
5817 		if (rv != NULL)
5818 			*rv = 1;
5819 		return;
5820 	}
5821 	resp->array_len = args->array_len;
5822 	resp->array = kmem_zalloc(args->array_len * sizeof (nfs_resop4),
5823 	    KM_SLEEP);
5824 
5825 	cs.basecr = cr;
5826 
5827 	DTRACE_NFSV4_2(compound__start, struct compound_state *, &cs,
5828 	    COMPOUND4args *, args);
5829 
5830 	/*
5831 	 * For now, NFS4 compound processing must be protected by
5832 	 * exported_lock because it can access more than one exportinfo
5833 	 * per compound and share/unshare can now change multiple
5834 	 * exinfo structs.  The NFS2/3 code only refs 1 exportinfo
5835 	 * per proc (excluding public exinfo), and exi_count design
5836 	 * is sufficient to protect concurrent execution of NFS2/3
5837 	 * ops along with unexport.  This lock will be removed as
5838 	 * part of the NFSv4 phase 2 namespace redesign work.
5839 	 */
5840 	rw_enter(&exported_lock, RW_READER);
5841 
5842 	/*
5843 	 * If this is the first compound we've seen, we need to start all
5844 	 * new instances' grace periods.
5845 	 */
5846 	if (rfs4_seen_first_compound == 0) {
5847 		rfs4_grace_start_new();
5848 		/*
5849 		 * This must be set after rfs4_grace_start_new(), otherwise
5850 		 * another thread could proceed past here before the former
5851 		 * is finished.
5852 		 */
5853 		rfs4_seen_first_compound = 1;
5854 	}
5855 
5856 	for (i = 0; i < args->array_len && cs.cont; i++) {
5857 		nfs_argop4 *argop;
5858 		nfs_resop4 *resop;
5859 		uint_t op;
5860 
5861 		argop = &args->array[i];
5862 		resop = &resp->array[i];
5863 		resop->resop = argop->argop;
5864 		op = (uint_t)resop->resop;
5865 
5866 		if (op < rfsv4disp_cnt) {
5867 			/*
5868 			 * Count the individual ops here; NULL and COMPOUND
5869 			 * are counted in common_dispatch()
5870 			 */
5871 			rfsproccnt_v4_ptr[op].value.ui64++;
5872 
5873 			NFS4_DEBUG(rfs4_debug > 1,
5874 			    (CE_NOTE, "Executing %s", rfs4_op_string[op]));
5875 			(*rfsv4disptab[op].dis_proc)(argop, resop, req, &cs);
5876 			NFS4_DEBUG(rfs4_debug > 1, (CE_NOTE, "%s returned %d",
5877 			    rfs4_op_string[op], *cs.statusp));
5878 			if (*cs.statusp != NFS4_OK)
5879 				cs.cont = FALSE;
5880 		} else {
5881 			/*
5882 			 * This is effectively dead code since XDR code
5883 			 * will have already returned BADXDR if op doesn't
5884 			 * decode to legal value.  This only done for a
5885 			 * day when XDR code doesn't verify v4 opcodes.
5886 			 */
5887 			op = OP_ILLEGAL;
5888 			rfsproccnt_v4_ptr[OP_ILLEGAL_IDX].value.ui64++;
5889 
5890 			rfs4_op_illegal(argop, resop, req, &cs);
5891 			cs.cont = FALSE;
5892 		}
5893 
5894 		/*
5895 		 * If not at last op, and if we are to stop, then
5896 		 * compact the results array.
5897 		 */
5898 		if ((i + 1) < args->array_len && !cs.cont) {
5899 			nfs_resop4 *new_res = kmem_alloc(
5900 			    (i+1) * sizeof (nfs_resop4), KM_SLEEP);
5901 			bcopy(resp->array,
5902 			    new_res, (i+1) * sizeof (nfs_resop4));
5903 			kmem_free(resp->array,
5904 			    args->array_len * sizeof (nfs_resop4));
5905 
5906 			resp->array_len =  i + 1;
5907 			resp->array = new_res;
5908 		}
5909 	}
5910 
5911 	rw_exit(&exported_lock);
5912 
5913 	DTRACE_NFSV4_2(compound__done, struct compound_state *, &cs,
5914 	    COMPOUND4res *, resp);
5915 
5916 	if (cs.vp)
5917 		VN_RELE(cs.vp);
5918 	if (cs.saved_vp)
5919 		VN_RELE(cs.saved_vp);
5920 	if (cs.saved_fh.nfs_fh4_val)
5921 		kmem_free(cs.saved_fh.nfs_fh4_val, NFS4_FHSIZE);
5922 
5923 	if (cs.basecr)
5924 		crfree(cs.basecr);
5925 	if (cs.cr)
5926 		crfree(cs.cr);
5927 	/*
5928 	 * done with this compound request, free the label
5929 	 */
5930 
5931 	if (req->rq_label != NULL) {
5932 		kmem_free(req->rq_label, sizeof (bslabel_t));
5933 		req->rq_label = NULL;
5934 	}
5935 }
5936 
5937 /*
5938  * XXX because of what appears to be duplicate calls to rfs4_compound_free
5939  * XXX zero out the tag and array values. Need to investigate why the
5940  * XXX calls occur, but at least prevent the panic for now.
5941  */
5942 void
5943 rfs4_compound_free(COMPOUND4res *resp)
5944 {
5945 	uint_t i;
5946 
5947 	if (resp->tag.utf8string_val) {
5948 		UTF8STRING_FREE(resp->tag)
5949 	}
5950 
5951 	for (i = 0; i < resp->array_len; i++) {
5952 		nfs_resop4 *resop;
5953 		uint_t op;
5954 
5955 		resop = &resp->array[i];
5956 		op = (uint_t)resop->resop;
5957 		if (op < rfsv4disp_cnt) {
5958 			(*rfsv4disptab[op].dis_resfree)(resop);
5959 		}
5960 	}
5961 	if (resp->array != NULL) {
5962 		kmem_free(resp->array, resp->array_len * sizeof (nfs_resop4));
5963 	}
5964 }
5965 
5966 /*
5967  * Process the value of the compound request rpc flags, as a bit-AND
5968  * of the individual per-op flags (idempotent, allowork, publicfh_ok)
5969  */
5970 void
5971 rfs4_compound_flagproc(COMPOUND4args *args, int *flagp)
5972 {
5973 	int i;
5974 	int flag = RPC_ALL;
5975 
5976 	for (i = 0; flag && i < args->array_len; i++) {
5977 		uint_t op;
5978 
5979 		op = (uint_t)args->array[i].argop;
5980 
5981 		if (op < rfsv4disp_cnt)
5982 			flag &= rfsv4disptab[op].dis_flags;
5983 		else
5984 			flag = 0;
5985 	}
5986 	*flagp = flag;
5987 }
5988 
5989 nfsstat4
5990 rfs4_client_sysid(rfs4_client_t *cp, sysid_t *sp)
5991 {
5992 	nfsstat4 e;
5993 
5994 	rfs4_dbe_lock(cp->rc_dbe);
5995 
5996 	if (cp->rc_sysidt != LM_NOSYSID) {
5997 		*sp = cp->rc_sysidt;
5998 		e = NFS4_OK;
5999 
6000 	} else if ((cp->rc_sysidt = lm_alloc_sysidt()) != LM_NOSYSID) {
6001 		*sp = cp->rc_sysidt;
6002 		e = NFS4_OK;
6003 
6004 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
6005 		    "rfs4_client_sysid: allocated 0x%x\n", *sp));
6006 	} else
6007 		e = NFS4ERR_DELAY;
6008 
6009 	rfs4_dbe_unlock(cp->rc_dbe);
6010 	return (e);
6011 }
6012 
6013 #if defined(DEBUG) && ! defined(lint)
6014 static void lock_print(char *str, int operation, struct flock64 *flk)
6015 {
6016 	char *op, *type;
6017 
6018 	switch (operation) {
6019 	case F_GETLK: op = "F_GETLK";
6020 		break;
6021 	case F_SETLK: op = "F_SETLK";
6022 		break;
6023 	case F_SETLK_NBMAND: op = "F_SETLK_NBMAND";
6024 		break;
6025 	default: op = "F_UNKNOWN";
6026 		break;
6027 	}
6028 	switch (flk->l_type) {
6029 	case F_UNLCK: type = "F_UNLCK";
6030 		break;
6031 	case F_RDLCK: type = "F_RDLCK";
6032 		break;
6033 	case F_WRLCK: type = "F_WRLCK";
6034 		break;
6035 	default: type = "F_UNKNOWN";
6036 		break;
6037 	}
6038 
6039 	ASSERT(flk->l_whence == 0);
6040 	cmn_err(CE_NOTE, "%s:  %s, type = %s, off = %llx len = %llx pid = %d",
6041 	    str, op, type, (longlong_t)flk->l_start,
6042 	    flk->l_len ? (longlong_t)flk->l_len : ~0LL, flk->l_pid);
6043 }
6044 
6045 #define	LOCK_PRINT(d, s, t, f) if (d) lock_print(s, t, f)
6046 #else
6047 #define	LOCK_PRINT(d, s, t, f)
6048 #endif
6049 
6050 /*ARGSUSED*/
6051 static bool_t
6052 creds_ok(cred_set_t cr_set, struct svc_req *req, struct compound_state *cs)
6053 {
6054 	return (TRUE);
6055 }
6056 
6057 /*
6058  * Look up the pathname using the vp in cs as the directory vnode.
6059  * cs->vp will be the vnode for the file on success
6060  */
6061 
6062 static nfsstat4
6063 rfs4_lookup(component4 *component, struct svc_req *req,
6064     struct compound_state *cs)
6065 {
6066 	char *nm;
6067 	uint32_t len;
6068 	nfsstat4 status;
6069 	struct sockaddr *ca;
6070 	char *name;
6071 
6072 	if (cs->vp == NULL) {
6073 		return (NFS4ERR_NOFILEHANDLE);
6074 	}
6075 	if (cs->vp->v_type != VDIR) {
6076 		return (NFS4ERR_NOTDIR);
6077 	}
6078 
6079 	status = utf8_dir_verify(component);
6080 	if (status != NFS4_OK)
6081 		return (status);
6082 
6083 	nm = utf8_to_fn(component, &len, NULL);
6084 	if (nm == NULL) {
6085 		return (NFS4ERR_INVAL);
6086 	}
6087 
6088 	if (len > MAXNAMELEN) {
6089 		kmem_free(nm, len);
6090 		return (NFS4ERR_NAMETOOLONG);
6091 	}
6092 
6093 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6094 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6095 	    MAXPATHLEN + 1);
6096 
6097 	if (name == NULL) {
6098 		kmem_free(nm, len);
6099 		return (NFS4ERR_INVAL);
6100 	}
6101 
6102 	status = do_rfs4_op_lookup(name, req, cs);
6103 
6104 	if (name != nm)
6105 		kmem_free(name, MAXPATHLEN + 1);
6106 
6107 	kmem_free(nm, len);
6108 
6109 	return (status);
6110 }
6111 
6112 static nfsstat4
6113 rfs4_lookupfile(component4 *component, struct svc_req *req,
6114     struct compound_state *cs, uint32_t access, change_info4 *cinfo)
6115 {
6116 	nfsstat4 status;
6117 	vnode_t *dvp = cs->vp;
6118 	vattr_t bva, ava, fva;
6119 	int error;
6120 
6121 	/* Get "before" change value */
6122 	bva.va_mask = AT_CTIME|AT_SEQ;
6123 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6124 	if (error)
6125 		return (puterrno4(error));
6126 
6127 	/* rfs4_lookup may VN_RELE directory */
6128 	VN_HOLD(dvp);
6129 
6130 	status = rfs4_lookup(component, req, cs);
6131 	if (status != NFS4_OK) {
6132 		VN_RELE(dvp);
6133 		return (status);
6134 	}
6135 
6136 	/*
6137 	 * Get "after" change value, if it fails, simply return the
6138 	 * before value.
6139 	 */
6140 	ava.va_mask = AT_CTIME|AT_SEQ;
6141 	if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6142 		ava.va_ctime = bva.va_ctime;
6143 		ava.va_seq = 0;
6144 	}
6145 	VN_RELE(dvp);
6146 
6147 	/*
6148 	 * Validate the file is a file
6149 	 */
6150 	fva.va_mask = AT_TYPE|AT_MODE;
6151 	error = VOP_GETATTR(cs->vp, &fva, 0, cs->cr, NULL);
6152 	if (error)
6153 		return (puterrno4(error));
6154 
6155 	if (fva.va_type != VREG) {
6156 		if (fva.va_type == VDIR)
6157 			return (NFS4ERR_ISDIR);
6158 		if (fva.va_type == VLNK)
6159 			return (NFS4ERR_SYMLINK);
6160 		return (NFS4ERR_INVAL);
6161 	}
6162 
6163 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime);
6164 	NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6165 
6166 	/*
6167 	 * It is undefined if VOP_LOOKUP will change va_seq, so
6168 	 * cinfo.atomic = TRUE only if we have
6169 	 * non-zero va_seq's, and they have not changed.
6170 	 */
6171 	if (bva.va_seq && ava.va_seq && ava.va_seq == bva.va_seq)
6172 		cinfo->atomic = TRUE;
6173 	else
6174 		cinfo->atomic = FALSE;
6175 
6176 	/* Check for mandatory locking */
6177 	cs->mandlock = MANDLOCK(cs->vp, fva.va_mode);
6178 	return (check_open_access(access, cs, req));
6179 }
6180 
6181 static nfsstat4
6182 create_vnode(vnode_t *dvp, char *nm,  vattr_t *vap, createmode4 mode,
6183     timespec32_t *mtime, cred_t *cr, vnode_t **vpp, bool_t *created)
6184 {
6185 	int error;
6186 	nfsstat4 status = NFS4_OK;
6187 	vattr_t va;
6188 
6189 tryagain:
6190 
6191 	/*
6192 	 * The file open mode used is VWRITE.  If the client needs
6193 	 * some other semantic, then it should do the access checking
6194 	 * itself.  It would have been nice to have the file open mode
6195 	 * passed as part of the arguments.
6196 	 */
6197 
6198 	*created = TRUE;
6199 	error = VOP_CREATE(dvp, nm, vap, EXCL, VWRITE, vpp, cr, 0, NULL, NULL);
6200 
6201 	if (error) {
6202 		*created = FALSE;
6203 
6204 		/*
6205 		 * If we got something other than file already exists
6206 		 * then just return this error.  Otherwise, we got
6207 		 * EEXIST.  If we were doing a GUARDED create, then
6208 		 * just return this error.  Otherwise, we need to
6209 		 * make sure that this wasn't a duplicate of an
6210 		 * exclusive create request.
6211 		 *
6212 		 * The assumption is made that a non-exclusive create
6213 		 * request will never return EEXIST.
6214 		 */
6215 
6216 		if (error != EEXIST || mode == GUARDED4) {
6217 			status = puterrno4(error);
6218 			return (status);
6219 		}
6220 		error = VOP_LOOKUP(dvp, nm, vpp, NULL, 0, NULL, cr,
6221 		    NULL, NULL, NULL);
6222 
6223 		if (error) {
6224 			/*
6225 			 * We couldn't find the file that we thought that
6226 			 * we just created.  So, we'll just try creating
6227 			 * it again.
6228 			 */
6229 			if (error == ENOENT)
6230 				goto tryagain;
6231 
6232 			status = puterrno4(error);
6233 			return (status);
6234 		}
6235 
6236 		if (mode == UNCHECKED4) {
6237 			/* existing object must be regular file */
6238 			if ((*vpp)->v_type != VREG) {
6239 				if ((*vpp)->v_type == VDIR)
6240 					status = NFS4ERR_ISDIR;
6241 				else if ((*vpp)->v_type == VLNK)
6242 					status = NFS4ERR_SYMLINK;
6243 				else
6244 					status = NFS4ERR_INVAL;
6245 				VN_RELE(*vpp);
6246 				return (status);
6247 			}
6248 
6249 			return (NFS4_OK);
6250 		}
6251 
6252 		/* Check for duplicate request */
6253 		ASSERT(mtime != 0);
6254 		va.va_mask = AT_MTIME;
6255 		error = VOP_GETATTR(*vpp, &va, 0, cr, NULL);
6256 		if (!error) {
6257 			/* We found the file */
6258 			if (va.va_mtime.tv_sec != mtime->tv_sec ||
6259 			    va.va_mtime.tv_nsec != mtime->tv_nsec) {
6260 				/* but its not our creation */
6261 				VN_RELE(*vpp);
6262 				return (NFS4ERR_EXIST);
6263 			}
6264 			*created = TRUE; /* retrans of create == created */
6265 			return (NFS4_OK);
6266 		}
6267 		VN_RELE(*vpp);
6268 		return (NFS4ERR_EXIST);
6269 	}
6270 
6271 	return (NFS4_OK);
6272 }
6273 
6274 static nfsstat4
6275 check_open_access(uint32_t access, struct compound_state *cs,
6276     struct svc_req *req)
6277 {
6278 	int error;
6279 	vnode_t *vp;
6280 	bool_t readonly;
6281 	cred_t *cr = cs->cr;
6282 
6283 	/* For now we don't allow mandatory locking as per V2/V3 */
6284 	if (cs->access == CS_ACCESS_DENIED || cs->mandlock) {
6285 		return (NFS4ERR_ACCESS);
6286 	}
6287 
6288 	vp = cs->vp;
6289 	ASSERT(cr != NULL && vp->v_type == VREG);
6290 
6291 	/*
6292 	 * If the file system is exported read only and we are trying
6293 	 * to open for write, then return NFS4ERR_ROFS
6294 	 */
6295 
6296 	readonly = rdonly4(cs->exi, cs->vp, req);
6297 
6298 	if ((access & OPEN4_SHARE_ACCESS_WRITE) && readonly)
6299 		return (NFS4ERR_ROFS);
6300 
6301 	if (access & OPEN4_SHARE_ACCESS_READ) {
6302 		if ((VOP_ACCESS(vp, VREAD, 0, cr, NULL) != 0) &&
6303 		    (VOP_ACCESS(vp, VEXEC, 0, cr, NULL) != 0)) {
6304 			return (NFS4ERR_ACCESS);
6305 		}
6306 	}
6307 
6308 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
6309 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
6310 		if (error)
6311 			return (NFS4ERR_ACCESS);
6312 	}
6313 
6314 	return (NFS4_OK);
6315 }
6316 
6317 static nfsstat4
6318 rfs4_createfile(OPEN4args *args, struct svc_req *req, struct compound_state *cs,
6319     change_info4 *cinfo, bitmap4 *attrset, clientid4 clientid)
6320 {
6321 	struct nfs4_svgetit_arg sarg;
6322 	struct nfs4_ntov_table ntov;
6323 
6324 	bool_t ntov_table_init = FALSE;
6325 	struct statvfs64 sb;
6326 	nfsstat4 status;
6327 	vnode_t *vp;
6328 	vattr_t bva, ava, iva, cva, *vap;
6329 	vnode_t *dvp;
6330 	timespec32_t *mtime;
6331 	char *nm = NULL;
6332 	uint_t buflen;
6333 	bool_t created;
6334 	bool_t setsize = FALSE;
6335 	len_t reqsize;
6336 	int error;
6337 	bool_t trunc;
6338 	caller_context_t ct;
6339 	component4 *component;
6340 	bslabel_t *clabel;
6341 	struct sockaddr *ca;
6342 	char *name = NULL;
6343 
6344 	sarg.sbp = &sb;
6345 	sarg.is_referral = B_FALSE;
6346 
6347 	dvp = cs->vp;
6348 
6349 	/* Check if the file system is read only */
6350 	if (rdonly4(cs->exi, dvp, req))
6351 		return (NFS4ERR_ROFS);
6352 
6353 	/* check the label of including directory */
6354 	if (is_system_labeled()) {
6355 		ASSERT(req->rq_label != NULL);
6356 		clabel = req->rq_label;
6357 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
6358 		    "got client label from request(1)",
6359 		    struct svc_req *, req);
6360 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
6361 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
6362 			    cs->exi)) {
6363 				return (NFS4ERR_ACCESS);
6364 			}
6365 		}
6366 	}
6367 
6368 	/*
6369 	 * Get the last component of path name in nm. cs will reference
6370 	 * the including directory on success.
6371 	 */
6372 	component = &args->open_claim4_u.file;
6373 	status = utf8_dir_verify(component);
6374 	if (status != NFS4_OK)
6375 		return (status);
6376 
6377 	nm = utf8_to_fn(component, &buflen, NULL);
6378 
6379 	if (nm == NULL)
6380 		return (NFS4ERR_RESOURCE);
6381 
6382 	if (buflen > MAXNAMELEN) {
6383 		kmem_free(nm, buflen);
6384 		return (NFS4ERR_NAMETOOLONG);
6385 	}
6386 
6387 	bva.va_mask = AT_TYPE|AT_CTIME|AT_SEQ;
6388 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6389 	if (error) {
6390 		kmem_free(nm, buflen);
6391 		return (puterrno4(error));
6392 	}
6393 
6394 	if (bva.va_type != VDIR) {
6395 		kmem_free(nm, buflen);
6396 		return (NFS4ERR_NOTDIR);
6397 	}
6398 
6399 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime)
6400 
6401 	switch (args->mode) {
6402 	case GUARDED4:
6403 		/*FALLTHROUGH*/
6404 	case UNCHECKED4:
6405 		nfs4_ntov_table_init(&ntov);
6406 		ntov_table_init = TRUE;
6407 
6408 		*attrset = 0;
6409 		status = do_rfs4_set_attrs(attrset,
6410 		    &args->createhow4_u.createattrs,
6411 		    cs, &sarg, &ntov, NFS4ATTR_SETIT);
6412 
6413 		if (status == NFS4_OK && (sarg.vap->va_mask & AT_TYPE) &&
6414 		    sarg.vap->va_type != VREG) {
6415 			if (sarg.vap->va_type == VDIR)
6416 				status = NFS4ERR_ISDIR;
6417 			else if (sarg.vap->va_type == VLNK)
6418 				status = NFS4ERR_SYMLINK;
6419 			else
6420 				status = NFS4ERR_INVAL;
6421 		}
6422 
6423 		if (status != NFS4_OK) {
6424 			kmem_free(nm, buflen);
6425 			nfs4_ntov_table_free(&ntov, &sarg);
6426 			*attrset = 0;
6427 			return (status);
6428 		}
6429 
6430 		vap = sarg.vap;
6431 		vap->va_type = VREG;
6432 		vap->va_mask |= AT_TYPE;
6433 
6434 		if ((vap->va_mask & AT_MODE) == 0) {
6435 			vap->va_mask |= AT_MODE;
6436 			vap->va_mode = (mode_t)0600;
6437 		}
6438 
6439 		if (vap->va_mask & AT_SIZE) {
6440 
6441 			/* Disallow create with a non-zero size */
6442 
6443 			if ((reqsize = sarg.vap->va_size) != 0) {
6444 				kmem_free(nm, buflen);
6445 				nfs4_ntov_table_free(&ntov, &sarg);
6446 				*attrset = 0;
6447 				return (NFS4ERR_INVAL);
6448 			}
6449 			setsize = TRUE;
6450 		}
6451 		break;
6452 
6453 	case EXCLUSIVE4:
6454 		/* prohibit EXCL create of named attributes */
6455 		if (dvp->v_flag & V_XATTRDIR) {
6456 			kmem_free(nm, buflen);
6457 			*attrset = 0;
6458 			return (NFS4ERR_INVAL);
6459 		}
6460 
6461 		cva.va_mask = AT_TYPE | AT_MTIME | AT_MODE;
6462 		cva.va_type = VREG;
6463 		/*
6464 		 * Ensure no time overflows. Assumes underlying
6465 		 * filesystem supports at least 32 bits.
6466 		 * Truncate nsec to usec resolution to allow valid
6467 		 * compares even if the underlying filesystem truncates.
6468 		 */
6469 		mtime = (timespec32_t *)&args->createhow4_u.createverf;
6470 		cva.va_mtime.tv_sec = mtime->tv_sec % TIME32_MAX;
6471 		cva.va_mtime.tv_nsec = (mtime->tv_nsec / 1000) * 1000;
6472 		cva.va_mode = (mode_t)0;
6473 		vap = &cva;
6474 
6475 		/*
6476 		 * For EXCL create, attrset is set to the server attr
6477 		 * used to cache the client's verifier.
6478 		 */
6479 		*attrset = FATTR4_TIME_MODIFY_MASK;
6480 		break;
6481 	}
6482 
6483 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6484 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6485 	    MAXPATHLEN  + 1);
6486 
6487 	if (name == NULL) {
6488 		kmem_free(nm, buflen);
6489 		return (NFS4ERR_SERVERFAULT);
6490 	}
6491 
6492 	status = create_vnode(dvp, name, vap, args->mode, mtime,
6493 	    cs->cr, &vp, &created);
6494 	if (nm != name)
6495 		kmem_free(name, MAXPATHLEN + 1);
6496 	kmem_free(nm, buflen);
6497 
6498 	if (status != NFS4_OK) {
6499 		if (ntov_table_init)
6500 			nfs4_ntov_table_free(&ntov, &sarg);
6501 		*attrset = 0;
6502 		return (status);
6503 	}
6504 
6505 	trunc = (setsize && !created);
6506 
6507 	if (args->mode != EXCLUSIVE4) {
6508 		bitmap4 createmask = args->createhow4_u.createattrs.attrmask;
6509 
6510 		/*
6511 		 * True verification that object was created with correct
6512 		 * attrs is impossible.  The attrs could have been changed
6513 		 * immediately after object creation.  If attributes did
6514 		 * not verify, the only recourse for the server is to
6515 		 * destroy the object.  Maybe if some attrs (like gid)
6516 		 * are set incorrectly, the object should be destroyed;
6517 		 * however, seems bad as a default policy.  Do we really
6518 		 * want to destroy an object over one of the times not
6519 		 * verifying correctly?  For these reasons, the server
6520 		 * currently sets bits in attrset for createattrs
6521 		 * that were set; however, no verification is done.
6522 		 *
6523 		 * vmask_to_nmask accounts for vattr bits set on create
6524 		 *	[do_rfs4_set_attrs() only sets resp bits for
6525 		 *	 non-vattr/vfs bits.]
6526 		 * Mask off any bits we set by default so as not to return
6527 		 * more attrset bits than were requested in createattrs
6528 		 */
6529 		if (created) {
6530 			nfs4_vmask_to_nmask(sarg.vap->va_mask, attrset);
6531 			*attrset &= createmask;
6532 		} else {
6533 			/*
6534 			 * We did not create the vnode (we tried but it
6535 			 * already existed).  In this case, the only createattr
6536 			 * that the spec allows the server to set is size,
6537 			 * and even then, it can only be set if it is 0.
6538 			 */
6539 			*attrset = 0;
6540 			if (trunc)
6541 				*attrset = FATTR4_SIZE_MASK;
6542 		}
6543 	}
6544 	if (ntov_table_init)
6545 		nfs4_ntov_table_free(&ntov, &sarg);
6546 
6547 	/*
6548 	 * Get the initial "after" sequence number, if it fails,
6549 	 * set to zero, time to before.
6550 	 */
6551 	iva.va_mask = AT_CTIME|AT_SEQ;
6552 	if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL)) {
6553 		iva.va_seq = 0;
6554 		iva.va_ctime = bva.va_ctime;
6555 	}
6556 
6557 	/*
6558 	 * create_vnode attempts to create the file exclusive,
6559 	 * if it already exists the VOP_CREATE will fail and
6560 	 * may not increase va_seq. It is atomic if
6561 	 * we haven't changed the directory, but if it has changed
6562 	 * we don't know what changed it.
6563 	 */
6564 	if (!created) {
6565 		if (bva.va_seq && iva.va_seq &&
6566 		    bva.va_seq == iva.va_seq)
6567 			cinfo->atomic = TRUE;
6568 		else
6569 			cinfo->atomic = FALSE;
6570 		NFS4_SET_FATTR4_CHANGE(cinfo->after, iva.va_ctime);
6571 	} else {
6572 		/*
6573 		 * The entry was created, we need to sync the
6574 		 * directory metadata.
6575 		 */
6576 		(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
6577 
6578 		/*
6579 		 * Get "after" change value, if it fails, simply return the
6580 		 * before value.
6581 		 */
6582 		ava.va_mask = AT_CTIME|AT_SEQ;
6583 		if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6584 			ava.va_ctime = bva.va_ctime;
6585 			ava.va_seq = 0;
6586 		}
6587 
6588 		NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6589 
6590 		/*
6591 		 * The cinfo->atomic = TRUE only if we have
6592 		 * non-zero va_seq's, and it has incremented by exactly one
6593 		 * during the create_vnode and it didn't
6594 		 * change during the VOP_FSYNC.
6595 		 */
6596 		if (bva.va_seq && iva.va_seq && ava.va_seq &&
6597 		    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
6598 			cinfo->atomic = TRUE;
6599 		else
6600 			cinfo->atomic = FALSE;
6601 	}
6602 
6603 	/* Check for mandatory locking and that the size gets set. */
6604 	cva.va_mask = AT_MODE;
6605 	if (setsize)
6606 		cva.va_mask |= AT_SIZE;
6607 
6608 	/* Assume the worst */
6609 	cs->mandlock = TRUE;
6610 
6611 	if (VOP_GETATTR(vp, &cva, 0, cs->cr, NULL) == 0) {
6612 		cs->mandlock = MANDLOCK(cs->vp, cva.va_mode);
6613 
6614 		/*
6615 		 * Truncate the file if necessary; this would be
6616 		 * the case for create over an existing file.
6617 		 */
6618 
6619 		if (trunc) {
6620 			int in_crit = 0;
6621 			rfs4_file_t *fp;
6622 			bool_t create = FALSE;
6623 
6624 			/*
6625 			 * We are writing over an existing file.
6626 			 * Check to see if we need to recall a delegation.
6627 			 */
6628 			rfs4_hold_deleg_policy();
6629 			if ((fp = rfs4_findfile(vp, NULL, &create)) != NULL) {
6630 				if (rfs4_check_delegated_byfp(FWRITE, fp,
6631 				    (reqsize == 0), FALSE, FALSE, &clientid)) {
6632 					rfs4_file_rele(fp);
6633 					rfs4_rele_deleg_policy();
6634 					VN_RELE(vp);
6635 					*attrset = 0;
6636 					return (NFS4ERR_DELAY);
6637 				}
6638 				rfs4_file_rele(fp);
6639 			}
6640 			rfs4_rele_deleg_policy();
6641 
6642 			if (nbl_need_check(vp)) {
6643 				in_crit = 1;
6644 
6645 				ASSERT(reqsize == 0);
6646 
6647 				nbl_start_crit(vp, RW_READER);
6648 				if (nbl_conflict(vp, NBL_WRITE, 0,
6649 				    cva.va_size, 0, NULL)) {
6650 					in_crit = 0;
6651 					nbl_end_crit(vp);
6652 					VN_RELE(vp);
6653 					*attrset = 0;
6654 					return (NFS4ERR_ACCESS);
6655 				}
6656 			}
6657 			ct.cc_sysid = 0;
6658 			ct.cc_pid = 0;
6659 			ct.cc_caller_id = nfs4_srv_caller_id;
6660 			ct.cc_flags = CC_DONTBLOCK;
6661 
6662 			cva.va_mask = AT_SIZE;
6663 			cva.va_size = reqsize;
6664 			(void) VOP_SETATTR(vp, &cva, 0, cs->cr, &ct);
6665 			if (in_crit)
6666 				nbl_end_crit(vp);
6667 		}
6668 	}
6669 
6670 	error = makefh4(&cs->fh, vp, cs->exi);
6671 
6672 	/*
6673 	 * Force modified data and metadata out to stable storage.
6674 	 */
6675 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
6676 
6677 	if (error) {
6678 		VN_RELE(vp);
6679 		*attrset = 0;
6680 		return (puterrno4(error));
6681 	}
6682 
6683 	/* if parent dir is attrdir, set namedattr fh flag */
6684 	if (dvp->v_flag & V_XATTRDIR)
6685 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
6686 
6687 	if (cs->vp)
6688 		VN_RELE(cs->vp);
6689 
6690 	cs->vp = vp;
6691 
6692 	/*
6693 	 * if we did not create the file, we will need to check
6694 	 * the access bits on the file
6695 	 */
6696 
6697 	if (!created) {
6698 		if (setsize)
6699 			args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
6700 		status = check_open_access(args->share_access, cs, req);
6701 		if (status != NFS4_OK)
6702 			*attrset = 0;
6703 	}
6704 	return (status);
6705 }
6706 
6707 /*ARGSUSED*/
6708 static void
6709 rfs4_do_open(struct compound_state *cs, struct svc_req *req,
6710     rfs4_openowner_t *oo, delegreq_t deleg,
6711     uint32_t access, uint32_t deny,
6712     OPEN4res *resp, int deleg_cur)
6713 {
6714 	/* XXX Currently not using req  */
6715 	rfs4_state_t *sp;
6716 	rfs4_file_t *fp;
6717 	bool_t screate = TRUE;
6718 	bool_t fcreate = TRUE;
6719 	uint32_t open_a, share_a;
6720 	uint32_t open_d, share_d;
6721 	rfs4_deleg_state_t *dsp;
6722 	sysid_t sysid;
6723 	nfsstat4 status;
6724 	caller_context_t ct;
6725 	int fflags = 0;
6726 	int recall = 0;
6727 	int err;
6728 	int first_open;
6729 
6730 	/* get the file struct and hold a lock on it during initial open */
6731 	fp = rfs4_findfile_withlock(cs->vp, &cs->fh, &fcreate);
6732 	if (fp == NULL) {
6733 		resp->status = NFS4ERR_RESOURCE;
6734 		DTRACE_PROBE1(nfss__e__do__open1, nfsstat4, resp->status);
6735 		return;
6736 	}
6737 
6738 	sp = rfs4_findstate_by_owner_file(oo, fp, &screate);
6739 	if (sp == NULL) {
6740 		resp->status = NFS4ERR_RESOURCE;
6741 		DTRACE_PROBE1(nfss__e__do__open2, nfsstat4, resp->status);
6742 		/* No need to keep any reference */
6743 		rw_exit(&fp->rf_file_rwlock);
6744 		rfs4_file_rele(fp);
6745 		return;
6746 	}
6747 
6748 	/* try to get the sysid before continuing */
6749 	if ((status = rfs4_client_sysid(oo->ro_client, &sysid)) != NFS4_OK) {
6750 		resp->status = status;
6751 		rfs4_file_rele(fp);
6752 		/* Not a fully formed open; "close" it */
6753 		if (screate == TRUE)
6754 			rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6755 		rfs4_state_rele(sp);
6756 		return;
6757 	}
6758 
6759 	/* Calculate the fflags for this OPEN. */
6760 	if (access & OPEN4_SHARE_ACCESS_READ)
6761 		fflags |= FREAD;
6762 	if (access & OPEN4_SHARE_ACCESS_WRITE)
6763 		fflags |= FWRITE;
6764 
6765 	rfs4_dbe_lock(sp->rs_dbe);
6766 
6767 	/*
6768 	 * Calculate the new deny and access mode that this open is adding to
6769 	 * the file for this open owner;
6770 	 */
6771 	open_d = (deny & ~sp->rs_open_deny);
6772 	open_a = (access & ~sp->rs_open_access);
6773 
6774 	/*
6775 	 * Calculate the new share access and share deny modes that this open
6776 	 * is adding to the file for this open owner;
6777 	 */
6778 	share_a = (access & ~sp->rs_share_access);
6779 	share_d = (deny & ~sp->rs_share_deny);
6780 
6781 	first_open = (sp->rs_open_access & OPEN4_SHARE_ACCESS_BOTH) == 0;
6782 
6783 	/*
6784 	 * Check to see the client has already sent an open for this
6785 	 * open owner on this file with the same share/deny modes.
6786 	 * If so, we don't need to check for a conflict and we don't
6787 	 * need to add another shrlock.  If not, then we need to
6788 	 * check for conflicts in deny and access before checking for
6789 	 * conflicts in delegation.  We don't want to recall a
6790 	 * delegation based on an open that will eventually fail based
6791 	 * on shares modes.
6792 	 */
6793 
6794 	if (share_a || share_d) {
6795 		if ((err = rfs4_share(sp, access, deny)) != 0) {
6796 			rfs4_dbe_unlock(sp->rs_dbe);
6797 			resp->status = err;
6798 
6799 			rfs4_file_rele(fp);
6800 			/* Not a fully formed open; "close" it */
6801 			if (screate == TRUE)
6802 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6803 			rfs4_state_rele(sp);
6804 			return;
6805 		}
6806 	}
6807 
6808 	rfs4_dbe_lock(fp->rf_dbe);
6809 
6810 	/*
6811 	 * Check to see if this file is delegated and if so, if a
6812 	 * recall needs to be done.
6813 	 */
6814 	if (rfs4_check_recall(sp, access)) {
6815 		rfs4_dbe_unlock(fp->rf_dbe);
6816 		rfs4_dbe_unlock(sp->rs_dbe);
6817 		rfs4_recall_deleg(fp, FALSE, sp->rs_owner->ro_client);
6818 		delay(NFS4_DELEGATION_CONFLICT_DELAY);
6819 		rfs4_dbe_lock(sp->rs_dbe);
6820 
6821 		/* if state closed while lock was dropped */
6822 		if (sp->rs_closed) {
6823 			if (share_a || share_d)
6824 				(void) rfs4_unshare(sp);
6825 			rfs4_dbe_unlock(sp->rs_dbe);
6826 			rfs4_file_rele(fp);
6827 			/* Not a fully formed open; "close" it */
6828 			if (screate == TRUE)
6829 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6830 			rfs4_state_rele(sp);
6831 			resp->status = NFS4ERR_OLD_STATEID;
6832 			return;
6833 		}
6834 
6835 		rfs4_dbe_lock(fp->rf_dbe);
6836 		/* Let's see if the delegation was returned */
6837 		if (rfs4_check_recall(sp, access)) {
6838 			rfs4_dbe_unlock(fp->rf_dbe);
6839 			if (share_a || share_d)
6840 				(void) rfs4_unshare(sp);
6841 			rfs4_dbe_unlock(sp->rs_dbe);
6842 			rfs4_file_rele(fp);
6843 			rfs4_update_lease(sp->rs_owner->ro_client);
6844 
6845 			/* Not a fully formed open; "close" it */
6846 			if (screate == TRUE)
6847 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6848 			rfs4_state_rele(sp);
6849 			resp->status = NFS4ERR_DELAY;
6850 			return;
6851 		}
6852 	}
6853 	/*
6854 	 * the share check passed and any delegation conflict has been
6855 	 * taken care of, now call vop_open.
6856 	 * if this is the first open then call vop_open with fflags.
6857 	 * if not, call vn_open_upgrade with just the upgrade flags.
6858 	 *
6859 	 * if the file has been opened already, it will have the current
6860 	 * access mode in the state struct.  if it has no share access, then
6861 	 * this is a new open.
6862 	 *
6863 	 * However, if this is open with CLAIM_DLEGATE_CUR, then don't
6864 	 * call VOP_OPEN(), just do the open upgrade.
6865 	 */
6866 	if (first_open && !deleg_cur) {
6867 		ct.cc_sysid = sysid;
6868 		ct.cc_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
6869 		ct.cc_caller_id = nfs4_srv_caller_id;
6870 		ct.cc_flags = CC_DONTBLOCK;
6871 		err = VOP_OPEN(&cs->vp, fflags, cs->cr, &ct);
6872 		if (err) {
6873 			rfs4_dbe_unlock(fp->rf_dbe);
6874 			if (share_a || share_d)
6875 				(void) rfs4_unshare(sp);
6876 			rfs4_dbe_unlock(sp->rs_dbe);
6877 			rfs4_file_rele(fp);
6878 
6879 			/* Not a fully formed open; "close" it */
6880 			if (screate == TRUE)
6881 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6882 			rfs4_state_rele(sp);
6883 			/* check if a monitor detected a delegation conflict */
6884 			if (err == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
6885 				resp->status = NFS4ERR_DELAY;
6886 			else
6887 				resp->status = NFS4ERR_SERVERFAULT;
6888 			return;
6889 		}
6890 	} else { /* open upgrade */
6891 		/*
6892 		 * calculate the fflags for the new mode that is being added
6893 		 * by this upgrade.
6894 		 */
6895 		fflags = 0;
6896 		if (open_a & OPEN4_SHARE_ACCESS_READ)
6897 			fflags |= FREAD;
6898 		if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6899 			fflags |= FWRITE;
6900 		vn_open_upgrade(cs->vp, fflags);
6901 	}
6902 	sp->rs_open_access |= access;
6903 	sp->rs_open_deny |= deny;
6904 
6905 	if (open_d & OPEN4_SHARE_DENY_READ)
6906 		fp->rf_deny_read++;
6907 	if (open_d & OPEN4_SHARE_DENY_WRITE)
6908 		fp->rf_deny_write++;
6909 	fp->rf_share_deny |= deny;
6910 
6911 	if (open_a & OPEN4_SHARE_ACCESS_READ)
6912 		fp->rf_access_read++;
6913 	if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6914 		fp->rf_access_write++;
6915 	fp->rf_share_access |= access;
6916 
6917 	/*
6918 	 * Check for delegation here. if the deleg argument is not
6919 	 * DELEG_ANY, then this is a reclaim from a client and
6920 	 * we must honor the delegation requested. If necessary we can
6921 	 * set the recall flag.
6922 	 */
6923 
6924 	dsp = rfs4_grant_delegation(deleg, sp, &recall);
6925 
6926 	cs->deleg = (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE);
6927 
6928 	next_stateid(&sp->rs_stateid);
6929 
6930 	resp->stateid = sp->rs_stateid.stateid;
6931 
6932 	rfs4_dbe_unlock(fp->rf_dbe);
6933 	rfs4_dbe_unlock(sp->rs_dbe);
6934 
6935 	if (dsp) {
6936 		rfs4_set_deleg_response(dsp, &resp->delegation, NULL, recall);
6937 		rfs4_deleg_state_rele(dsp);
6938 	}
6939 
6940 	rfs4_file_rele(fp);
6941 	rfs4_state_rele(sp);
6942 
6943 	resp->status = NFS4_OK;
6944 }
6945 
6946 /*ARGSUSED*/
6947 static void
6948 rfs4_do_opennull(struct compound_state *cs, struct svc_req *req,
6949     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6950 {
6951 	change_info4 *cinfo = &resp->cinfo;
6952 	bitmap4 *attrset = &resp->attrset;
6953 
6954 	if (args->opentype == OPEN4_NOCREATE)
6955 		resp->status = rfs4_lookupfile(&args->open_claim4_u.file,
6956 		    req, cs, args->share_access, cinfo);
6957 	else {
6958 		/* inhibit delegation grants during exclusive create */
6959 
6960 		if (args->mode == EXCLUSIVE4)
6961 			rfs4_disable_delegation();
6962 
6963 		resp->status = rfs4_createfile(args, req, cs, cinfo, attrset,
6964 		    oo->ro_client->rc_clientid);
6965 	}
6966 
6967 	if (resp->status == NFS4_OK) {
6968 
6969 		/* cs->vp cs->fh now reference the desired file */
6970 
6971 		rfs4_do_open(cs, req, oo,
6972 		    oo->ro_need_confirm ? DELEG_NONE : DELEG_ANY,
6973 		    args->share_access, args->share_deny, resp, 0);
6974 
6975 		/*
6976 		 * If rfs4_createfile set attrset, we must
6977 		 * clear this attrset before the response is copied.
6978 		 */
6979 		if (resp->status != NFS4_OK && resp->attrset) {
6980 			resp->attrset = 0;
6981 		}
6982 	}
6983 	else
6984 		*cs->statusp = resp->status;
6985 
6986 	if (args->mode == EXCLUSIVE4)
6987 		rfs4_enable_delegation();
6988 }
6989 
6990 /*ARGSUSED*/
6991 static void
6992 rfs4_do_openprev(struct compound_state *cs, struct svc_req *req,
6993     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6994 {
6995 	change_info4 *cinfo = &resp->cinfo;
6996 	vattr_t va;
6997 	vtype_t v_type = cs->vp->v_type;
6998 	int error = 0;
6999 
7000 	/* Verify that we have a regular file */
7001 	if (v_type != VREG) {
7002 		if (v_type == VDIR)
7003 			resp->status = NFS4ERR_ISDIR;
7004 		else if (v_type == VLNK)
7005 			resp->status = NFS4ERR_SYMLINK;
7006 		else
7007 			resp->status = NFS4ERR_INVAL;
7008 		return;
7009 	}
7010 
7011 	va.va_mask = AT_MODE|AT_UID;
7012 	error = VOP_GETATTR(cs->vp, &va, 0, cs->cr, NULL);
7013 	if (error) {
7014 		resp->status = puterrno4(error);
7015 		return;
7016 	}
7017 
7018 	cs->mandlock = MANDLOCK(cs->vp, va.va_mode);
7019 
7020 	/*
7021 	 * Check if we have access to the file, Note the the file
7022 	 * could have originally been open UNCHECKED or GUARDED
7023 	 * with mode bits that will now fail, but there is nothing
7024 	 * we can really do about that except in the case that the
7025 	 * owner of the file is the one requesting the open.
7026 	 */
7027 	if (crgetuid(cs->cr) != va.va_uid) {
7028 		resp->status = check_open_access(args->share_access, cs, req);
7029 		if (resp->status != NFS4_OK) {
7030 			return;
7031 		}
7032 	}
7033 
7034 	/*
7035 	 * cinfo on a CLAIM_PREVIOUS is undefined, initialize to zero
7036 	 */
7037 	cinfo->before = 0;
7038 	cinfo->after = 0;
7039 	cinfo->atomic = FALSE;
7040 
7041 	rfs4_do_open(cs, req, oo,
7042 	    NFS4_DELEG4TYPE2REQTYPE(args->open_claim4_u.delegate_type),
7043 	    args->share_access, args->share_deny, resp, 0);
7044 }
7045 
7046 static void
7047 rfs4_do_opendelcur(struct compound_state *cs, struct svc_req *req,
7048     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7049 {
7050 	int error;
7051 	nfsstat4 status;
7052 	stateid4 stateid =
7053 	    args->open_claim4_u.delegate_cur_info.delegate_stateid;
7054 	rfs4_deleg_state_t *dsp;
7055 
7056 	/*
7057 	 * Find the state info from the stateid and confirm that the
7058 	 * file is delegated.  If the state openowner is the same as
7059 	 * the supplied openowner we're done. If not, get the file
7060 	 * info from the found state info. Use that file info to
7061 	 * create the state for this lock owner. Note solaris doen't
7062 	 * really need the pathname to find the file. We may want to
7063 	 * lookup the pathname and make sure that the vp exist and
7064 	 * matches the vp in the file structure. However it is
7065 	 * possible that the pathname nolonger exists (local process
7066 	 * unlinks the file), so this may not be that useful.
7067 	 */
7068 
7069 	status = rfs4_get_deleg_state(&stateid, &dsp);
7070 	if (status != NFS4_OK) {
7071 		resp->status = status;
7072 		return;
7073 	}
7074 
7075 	ASSERT(dsp->rds_finfo->rf_dinfo.rd_dtype != OPEN_DELEGATE_NONE);
7076 
7077 	/*
7078 	 * New lock owner, create state. Since this was probably called
7079 	 * in response to a CB_RECALL we set deleg to DELEG_NONE
7080 	 */
7081 
7082 	ASSERT(cs->vp != NULL);
7083 	VN_RELE(cs->vp);
7084 	VN_HOLD(dsp->rds_finfo->rf_vp);
7085 	cs->vp = dsp->rds_finfo->rf_vp;
7086 
7087 	if (error = makefh4(&cs->fh, cs->vp, cs->exi)) {
7088 		rfs4_deleg_state_rele(dsp);
7089 		*cs->statusp = resp->status = puterrno4(error);
7090 		return;
7091 	}
7092 
7093 	/* Mark progress for delegation returns */
7094 	dsp->rds_finfo->rf_dinfo.rd_time_lastwrite = gethrestime_sec();
7095 	rfs4_deleg_state_rele(dsp);
7096 	rfs4_do_open(cs, req, oo, DELEG_NONE,
7097 	    args->share_access, args->share_deny, resp, 1);
7098 }
7099 
7100 /*ARGSUSED*/
7101 static void
7102 rfs4_do_opendelprev(struct compound_state *cs, struct svc_req *req,
7103     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7104 {
7105 	/*
7106 	 * Lookup the pathname, it must already exist since this file
7107 	 * was delegated.
7108 	 *
7109 	 * Find the file and state info for this vp and open owner pair.
7110 	 *	check that they are in fact delegated.
7111 	 *	check that the state access and deny modes are the same.
7112 	 *
7113 	 * Return the delgation possibly seting the recall flag.
7114 	 */
7115 	rfs4_file_t *fp;
7116 	rfs4_state_t *sp;
7117 	bool_t create = FALSE;
7118 	bool_t dcreate = FALSE;
7119 	rfs4_deleg_state_t *dsp;
7120 	nfsace4 *ace;
7121 
7122 	/* Note we ignore oflags */
7123 	resp->status = rfs4_lookupfile(&args->open_claim4_u.file_delegate_prev,
7124 	    req, cs, args->share_access, &resp->cinfo);
7125 
7126 	if (resp->status != NFS4_OK) {
7127 		return;
7128 	}
7129 
7130 	/* get the file struct and hold a lock on it during initial open */
7131 	fp = rfs4_findfile_withlock(cs->vp, NULL, &create);
7132 	if (fp == NULL) {
7133 		resp->status = NFS4ERR_RESOURCE;
7134 		DTRACE_PROBE1(nfss__e__do_opendelprev1, nfsstat4, resp->status);
7135 		return;
7136 	}
7137 
7138 	sp = rfs4_findstate_by_owner_file(oo, fp, &create);
7139 	if (sp == NULL) {
7140 		resp->status = NFS4ERR_SERVERFAULT;
7141 		DTRACE_PROBE1(nfss__e__do_opendelprev2, nfsstat4, resp->status);
7142 		rw_exit(&fp->rf_file_rwlock);
7143 		rfs4_file_rele(fp);
7144 		return;
7145 	}
7146 
7147 	rfs4_dbe_lock(sp->rs_dbe);
7148 	rfs4_dbe_lock(fp->rf_dbe);
7149 	if (args->share_access != sp->rs_share_access ||
7150 	    args->share_deny != sp->rs_share_deny ||
7151 	    sp->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
7152 		NFS4_DEBUG(rfs4_debug,
7153 		    (CE_NOTE, "rfs4_do_opendelprev: state mixup"));
7154 		rfs4_dbe_unlock(fp->rf_dbe);
7155 		rfs4_dbe_unlock(sp->rs_dbe);
7156 		rfs4_file_rele(fp);
7157 		rfs4_state_rele(sp);
7158 		resp->status = NFS4ERR_SERVERFAULT;
7159 		return;
7160 	}
7161 	rfs4_dbe_unlock(fp->rf_dbe);
7162 	rfs4_dbe_unlock(sp->rs_dbe);
7163 
7164 	dsp = rfs4_finddeleg(sp, &dcreate);
7165 	if (dsp == NULL) {
7166 		rfs4_state_rele(sp);
7167 		rfs4_file_rele(fp);
7168 		resp->status = NFS4ERR_SERVERFAULT;
7169 		return;
7170 	}
7171 
7172 	next_stateid(&sp->rs_stateid);
7173 
7174 	resp->stateid = sp->rs_stateid.stateid;
7175 
7176 	resp->delegation.delegation_type = dsp->rds_dtype;
7177 
7178 	if (dsp->rds_dtype == OPEN_DELEGATE_READ) {
7179 		open_read_delegation4 *rv =
7180 		    &resp->delegation.open_delegation4_u.read;
7181 
7182 		rv->stateid = dsp->rds_delegid.stateid;
7183 		rv->recall = FALSE; /* no policy in place to set to TRUE */
7184 		ace = &rv->permissions;
7185 	} else {
7186 		open_write_delegation4 *rv =
7187 		    &resp->delegation.open_delegation4_u.write;
7188 
7189 		rv->stateid = dsp->rds_delegid.stateid;
7190 		rv->recall = FALSE;  /* no policy in place to set to TRUE */
7191 		ace = &rv->permissions;
7192 		rv->space_limit.limitby = NFS_LIMIT_SIZE;
7193 		rv->space_limit.nfs_space_limit4_u.filesize = UINT64_MAX;
7194 	}
7195 
7196 	/* XXX For now */
7197 	ace->type = ACE4_ACCESS_ALLOWED_ACE_TYPE;
7198 	ace->flag = 0;
7199 	ace->access_mask = 0;
7200 	ace->who.utf8string_len = 0;
7201 	ace->who.utf8string_val = 0;
7202 
7203 	rfs4_deleg_state_rele(dsp);
7204 	rfs4_state_rele(sp);
7205 	rfs4_file_rele(fp);
7206 }
7207 
7208 typedef enum {
7209 	NFS4_CHKSEQ_OKAY = 0,
7210 	NFS4_CHKSEQ_REPLAY = 1,
7211 	NFS4_CHKSEQ_BAD = 2
7212 } rfs4_chkseq_t;
7213 
7214 /*
7215  * Generic function for sequence number checks.
7216  */
7217 static rfs4_chkseq_t
7218 rfs4_check_seqid(seqid4 seqid, nfs_resop4 *lastop,
7219     seqid4 rqst_seq, nfs_resop4 *resop, bool_t copyres)
7220 {
7221 	/* Same sequence ids and matching operations? */
7222 	if (seqid == rqst_seq && resop->resop == lastop->resop) {
7223 		if (copyres == TRUE) {
7224 			rfs4_free_reply(resop);
7225 			rfs4_copy_reply(resop, lastop);
7226 		}
7227 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
7228 		    "Replayed SEQID %d\n", seqid));
7229 		return (NFS4_CHKSEQ_REPLAY);
7230 	}
7231 
7232 	/* If the incoming sequence is not the next expected then it is bad */
7233 	if (rqst_seq != seqid + 1) {
7234 		if (rqst_seq == seqid) {
7235 			NFS4_DEBUG(rfs4_debug,
7236 			    (CE_NOTE, "BAD SEQID: Replayed sequence id "
7237 			    "but last op was %d current op is %d\n",
7238 			    lastop->resop, resop->resop));
7239 			return (NFS4_CHKSEQ_BAD);
7240 		}
7241 		NFS4_DEBUG(rfs4_debug,
7242 		    (CE_NOTE, "BAD SEQID: got %u expecting %u\n",
7243 		    rqst_seq, seqid));
7244 		return (NFS4_CHKSEQ_BAD);
7245 	}
7246 
7247 	/* Everything okay -- next expected */
7248 	return (NFS4_CHKSEQ_OKAY);
7249 }
7250 
7251 
7252 static rfs4_chkseq_t
7253 rfs4_check_open_seqid(seqid4 seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7254 {
7255 	rfs4_chkseq_t rc;
7256 
7257 	rfs4_dbe_lock(op->ro_dbe);
7258 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply, seqid, resop,
7259 	    TRUE);
7260 	rfs4_dbe_unlock(op->ro_dbe);
7261 
7262 	if (rc == NFS4_CHKSEQ_OKAY)
7263 		rfs4_update_lease(op->ro_client);
7264 
7265 	return (rc);
7266 }
7267 
7268 static rfs4_chkseq_t
7269 rfs4_check_olo_seqid(seqid4 olo_seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7270 {
7271 	rfs4_chkseq_t rc;
7272 
7273 	rfs4_dbe_lock(op->ro_dbe);
7274 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply,
7275 	    olo_seqid, resop, FALSE);
7276 	rfs4_dbe_unlock(op->ro_dbe);
7277 
7278 	return (rc);
7279 }
7280 
7281 static rfs4_chkseq_t
7282 rfs4_check_lock_seqid(seqid4 seqid, rfs4_lo_state_t *lsp, nfs_resop4 *resop)
7283 {
7284 	rfs4_chkseq_t rc = NFS4_CHKSEQ_OKAY;
7285 
7286 	rfs4_dbe_lock(lsp->rls_dbe);
7287 	if (!lsp->rls_skip_seqid_check)
7288 		rc = rfs4_check_seqid(lsp->rls_seqid, &lsp->rls_reply, seqid,
7289 		    resop, TRUE);
7290 	rfs4_dbe_unlock(lsp->rls_dbe);
7291 
7292 	return (rc);
7293 }
7294 
7295 static void
7296 rfs4_op_open(nfs_argop4 *argop, nfs_resop4 *resop,
7297     struct svc_req *req, struct compound_state *cs)
7298 {
7299 	OPEN4args *args = &argop->nfs_argop4_u.opopen;
7300 	OPEN4res *resp = &resop->nfs_resop4_u.opopen;
7301 	open_owner4 *owner = &args->owner;
7302 	open_claim_type4 claim = args->claim;
7303 	rfs4_client_t *cp;
7304 	rfs4_openowner_t *oo;
7305 	bool_t create;
7306 	bool_t replay = FALSE;
7307 	int can_reclaim;
7308 
7309 	DTRACE_NFSV4_2(op__open__start, struct compound_state *, cs,
7310 	    OPEN4args *, args);
7311 
7312 	if (cs->vp == NULL) {
7313 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7314 		goto end;
7315 	}
7316 
7317 	/*
7318 	 * Need to check clientid and lease expiration first based on
7319 	 * error ordering and incrementing sequence id.
7320 	 */
7321 	cp = rfs4_findclient_by_id(owner->clientid, FALSE);
7322 	if (cp == NULL) {
7323 		*cs->statusp = resp->status =
7324 		    rfs4_check_clientid(&owner->clientid, 0);
7325 		goto end;
7326 	}
7327 
7328 	if (rfs4_lease_expired(cp)) {
7329 		rfs4_client_close(cp);
7330 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7331 		goto end;
7332 	}
7333 	can_reclaim = cp->rc_can_reclaim;
7334 
7335 	/*
7336 	 * Find the open_owner for use from this point forward.  Take
7337 	 * care in updating the sequence id based on the type of error
7338 	 * being returned.
7339 	 */
7340 retry:
7341 	create = TRUE;
7342 	oo = rfs4_findopenowner(owner, &create, args->seqid);
7343 	if (oo == NULL) {
7344 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
7345 		rfs4_client_rele(cp);
7346 		goto end;
7347 	}
7348 
7349 	/* Hold off access to the sequence space while the open is done */
7350 	rfs4_sw_enter(&oo->ro_sw);
7351 
7352 	/*
7353 	 * If the open_owner existed before at the server, then check
7354 	 * the sequence id.
7355 	 */
7356 	if (!create && !oo->ro_postpone_confirm) {
7357 		switch (rfs4_check_open_seqid(args->seqid, oo, resop)) {
7358 		case NFS4_CHKSEQ_BAD:
7359 			if ((args->seqid > oo->ro_open_seqid) &&
7360 			    oo->ro_need_confirm) {
7361 				rfs4_free_opens(oo, TRUE, FALSE);
7362 				rfs4_sw_exit(&oo->ro_sw);
7363 				rfs4_openowner_rele(oo);
7364 				goto retry;
7365 			}
7366 			resp->status = NFS4ERR_BAD_SEQID;
7367 			goto out;
7368 		case NFS4_CHKSEQ_REPLAY: /* replay of previous request */
7369 			replay = TRUE;
7370 			goto out;
7371 		default:
7372 			break;
7373 		}
7374 
7375 		/*
7376 		 * Sequence was ok and open owner exists
7377 		 * check to see if we have yet to see an
7378 		 * open_confirm.
7379 		 */
7380 		if (oo->ro_need_confirm) {
7381 			rfs4_free_opens(oo, TRUE, FALSE);
7382 			rfs4_sw_exit(&oo->ro_sw);
7383 			rfs4_openowner_rele(oo);
7384 			goto retry;
7385 		}
7386 	}
7387 	/* Grace only applies to regular-type OPENs */
7388 	if (rfs4_clnt_in_grace(cp) &&
7389 	    (claim == CLAIM_NULL || claim == CLAIM_DELEGATE_CUR)) {
7390 		*cs->statusp = resp->status = NFS4ERR_GRACE;
7391 		goto out;
7392 	}
7393 
7394 	/*
7395 	 * If previous state at the server existed then can_reclaim
7396 	 * will be set. If not reply NFS4ERR_NO_GRACE to the
7397 	 * client.
7398 	 */
7399 	if (rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS && !can_reclaim) {
7400 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7401 		goto out;
7402 	}
7403 
7404 
7405 	/*
7406 	 * Reject the open if the client has missed the grace period
7407 	 */
7408 	if (!rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS) {
7409 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7410 		goto out;
7411 	}
7412 
7413 	/* Couple of up-front bookkeeping items */
7414 	if (oo->ro_need_confirm) {
7415 		/*
7416 		 * If this is a reclaim OPEN then we should not ask
7417 		 * for a confirmation of the open_owner per the
7418 		 * protocol specification.
7419 		 */
7420 		if (claim == CLAIM_PREVIOUS)
7421 			oo->ro_need_confirm = FALSE;
7422 		else
7423 			resp->rflags |= OPEN4_RESULT_CONFIRM;
7424 	}
7425 	resp->rflags |= OPEN4_RESULT_LOCKTYPE_POSIX;
7426 
7427 	/*
7428 	 * If there is an unshared filesystem mounted on this vnode,
7429 	 * do not allow to open/create in this directory.
7430 	 */
7431 	if (vn_ismntpt(cs->vp)) {
7432 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
7433 		goto out;
7434 	}
7435 
7436 	/*
7437 	 * access must READ, WRITE, or BOTH.  No access is invalid.
7438 	 * deny can be READ, WRITE, BOTH, or NONE.
7439 	 * bits not defined for access/deny are invalid.
7440 	 */
7441 	if (! (args->share_access & OPEN4_SHARE_ACCESS_BOTH) ||
7442 	    (args->share_access & ~OPEN4_SHARE_ACCESS_BOTH) ||
7443 	    (args->share_deny & ~OPEN4_SHARE_DENY_BOTH)) {
7444 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7445 		goto out;
7446 	}
7447 
7448 
7449 	/*
7450 	 * make sure attrset is zero before response is built.
7451 	 */
7452 	resp->attrset = 0;
7453 
7454 	switch (claim) {
7455 	case CLAIM_NULL:
7456 		rfs4_do_opennull(cs, req, args, oo, resp);
7457 		break;
7458 	case CLAIM_PREVIOUS:
7459 		rfs4_do_openprev(cs, req, args, oo, resp);
7460 		break;
7461 	case CLAIM_DELEGATE_CUR:
7462 		rfs4_do_opendelcur(cs, req, args, oo, resp);
7463 		break;
7464 	case CLAIM_DELEGATE_PREV:
7465 		rfs4_do_opendelprev(cs, req, args, oo, resp);
7466 		break;
7467 	default:
7468 		resp->status = NFS4ERR_INVAL;
7469 		break;
7470 	}
7471 
7472 out:
7473 	rfs4_client_rele(cp);
7474 
7475 	/* Catch sequence id handling here to make it a little easier */
7476 	switch (resp->status) {
7477 	case NFS4ERR_BADXDR:
7478 	case NFS4ERR_BAD_SEQID:
7479 	case NFS4ERR_BAD_STATEID:
7480 	case NFS4ERR_NOFILEHANDLE:
7481 	case NFS4ERR_RESOURCE:
7482 	case NFS4ERR_STALE_CLIENTID:
7483 	case NFS4ERR_STALE_STATEID:
7484 		/*
7485 		 * The protocol states that if any of these errors are
7486 		 * being returned, the sequence id should not be
7487 		 * incremented.  Any other return requires an
7488 		 * increment.
7489 		 */
7490 		break;
7491 	default:
7492 		/* Always update the lease in this case */
7493 		rfs4_update_lease(oo->ro_client);
7494 
7495 		/* Regular response - copy the result */
7496 		if (!replay)
7497 			rfs4_update_open_resp(oo, resop, &cs->fh);
7498 
7499 		/*
7500 		 * REPLAY case: Only if the previous response was OK
7501 		 * do we copy the filehandle.  If not OK, no
7502 		 * filehandle to copy.
7503 		 */
7504 		if (replay == TRUE &&
7505 		    resp->status == NFS4_OK &&
7506 		    oo->ro_reply_fh.nfs_fh4_val) {
7507 			/*
7508 			 * If this is a replay, we must restore the
7509 			 * current filehandle/vp to that of what was
7510 			 * returned originally.  Try our best to do
7511 			 * it.
7512 			 */
7513 			nfs_fh4_fmt_t *fh_fmtp =
7514 			    (nfs_fh4_fmt_t *)oo->ro_reply_fh.nfs_fh4_val;
7515 
7516 			cs->exi = checkexport4(&fh_fmtp->fh4_fsid,
7517 			    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
7518 
7519 			if (cs->exi == NULL) {
7520 				resp->status = NFS4ERR_STALE;
7521 				goto finish;
7522 			}
7523 
7524 			VN_RELE(cs->vp);
7525 
7526 			cs->vp = nfs4_fhtovp(&oo->ro_reply_fh, cs->exi,
7527 			    &resp->status);
7528 
7529 			if (cs->vp == NULL)
7530 				goto finish;
7531 
7532 			nfs_fh4_copy(&oo->ro_reply_fh, &cs->fh);
7533 		}
7534 
7535 		/*
7536 		 * If this was a replay, no need to update the
7537 		 * sequence id. If the open_owner was not created on
7538 		 * this pass, then update.  The first use of an
7539 		 * open_owner will not bump the sequence id.
7540 		 */
7541 		if (replay == FALSE && !create)
7542 			rfs4_update_open_sequence(oo);
7543 		/*
7544 		 * If the client is receiving an error and the
7545 		 * open_owner needs to be confirmed, there is no way
7546 		 * to notify the client of this fact ignoring the fact
7547 		 * that the server has no method of returning a
7548 		 * stateid to confirm.  Therefore, the server needs to
7549 		 * mark this open_owner in a way as to avoid the
7550 		 * sequence id checking the next time the client uses
7551 		 * this open_owner.
7552 		 */
7553 		if (resp->status != NFS4_OK && oo->ro_need_confirm)
7554 			oo->ro_postpone_confirm = TRUE;
7555 		/*
7556 		 * If OK response then clear the postpone flag and
7557 		 * reset the sequence id to keep in sync with the
7558 		 * client.
7559 		 */
7560 		if (resp->status == NFS4_OK && oo->ro_postpone_confirm) {
7561 			oo->ro_postpone_confirm = FALSE;
7562 			oo->ro_open_seqid = args->seqid;
7563 		}
7564 		break;
7565 	}
7566 
7567 finish:
7568 	*cs->statusp = resp->status;
7569 
7570 	rfs4_sw_exit(&oo->ro_sw);
7571 	rfs4_openowner_rele(oo);
7572 
7573 end:
7574 	DTRACE_NFSV4_2(op__open__done, struct compound_state *, cs,
7575 	    OPEN4res *, resp);
7576 }
7577 
7578 /*ARGSUSED*/
7579 void
7580 rfs4_op_open_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
7581     struct svc_req *req, struct compound_state *cs)
7582 {
7583 	OPEN_CONFIRM4args *args = &argop->nfs_argop4_u.opopen_confirm;
7584 	OPEN_CONFIRM4res *resp = &resop->nfs_resop4_u.opopen_confirm;
7585 	rfs4_state_t *sp;
7586 	nfsstat4 status;
7587 
7588 	DTRACE_NFSV4_2(op__open__confirm__start, struct compound_state *, cs,
7589 	    OPEN_CONFIRM4args *, args);
7590 
7591 	if (cs->vp == NULL) {
7592 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7593 		goto out;
7594 	}
7595 
7596 	if (cs->vp->v_type != VREG) {
7597 		*cs->statusp = resp->status =
7598 		    cs->vp->v_type == VDIR ? NFS4ERR_ISDIR : NFS4ERR_INVAL;
7599 		return;
7600 	}
7601 
7602 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7603 	if (status != NFS4_OK) {
7604 		*cs->statusp = resp->status = status;
7605 		goto out;
7606 	}
7607 
7608 	/* Ensure specified filehandle matches */
7609 	if (cs->vp != sp->rs_finfo->rf_vp) {
7610 		rfs4_state_rele(sp);
7611 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7612 		goto out;
7613 	}
7614 
7615 	/* hold off other access to open_owner while we tinker */
7616 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7617 
7618 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7619 	case NFS4_CHECK_STATEID_OKAY:
7620 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7621 		    resop) != 0) {
7622 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7623 			break;
7624 		}
7625 		/*
7626 		 * If it is the appropriate stateid and determined to
7627 		 * be "OKAY" then this means that the stateid does not
7628 		 * need to be confirmed and the client is in error for
7629 		 * sending an OPEN_CONFIRM.
7630 		 */
7631 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7632 		break;
7633 	case NFS4_CHECK_STATEID_OLD:
7634 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7635 		break;
7636 	case NFS4_CHECK_STATEID_BAD:
7637 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7638 		break;
7639 	case NFS4_CHECK_STATEID_EXPIRED:
7640 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7641 		break;
7642 	case NFS4_CHECK_STATEID_CLOSED:
7643 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7644 		break;
7645 	case NFS4_CHECK_STATEID_REPLAY:
7646 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7647 		    resop)) {
7648 		case NFS4_CHKSEQ_OKAY:
7649 			/*
7650 			 * This is replayed stateid; if seqid matches
7651 			 * next expected, then client is using wrong seqid.
7652 			 */
7653 			/* fall through */
7654 		case NFS4_CHKSEQ_BAD:
7655 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7656 			break;
7657 		case NFS4_CHKSEQ_REPLAY:
7658 			/*
7659 			 * Note this case is the duplicate case so
7660 			 * resp->status is already set.
7661 			 */
7662 			*cs->statusp = resp->status;
7663 			rfs4_update_lease(sp->rs_owner->ro_client);
7664 			break;
7665 		}
7666 		break;
7667 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7668 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7669 		    resop) != NFS4_CHKSEQ_OKAY) {
7670 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7671 			break;
7672 		}
7673 		*cs->statusp = resp->status = NFS4_OK;
7674 
7675 		next_stateid(&sp->rs_stateid);
7676 		resp->open_stateid = sp->rs_stateid.stateid;
7677 		sp->rs_owner->ro_need_confirm = FALSE;
7678 		rfs4_update_lease(sp->rs_owner->ro_client);
7679 		rfs4_update_open_sequence(sp->rs_owner);
7680 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7681 		break;
7682 	default:
7683 		ASSERT(FALSE);
7684 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7685 		break;
7686 	}
7687 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7688 	rfs4_state_rele(sp);
7689 
7690 out:
7691 	DTRACE_NFSV4_2(op__open__confirm__done, struct compound_state *, cs,
7692 	    OPEN_CONFIRM4res *, resp);
7693 }
7694 
7695 /*ARGSUSED*/
7696 void
7697 rfs4_op_open_downgrade(nfs_argop4 *argop, nfs_resop4 *resop,
7698     struct svc_req *req, struct compound_state *cs)
7699 {
7700 	OPEN_DOWNGRADE4args *args = &argop->nfs_argop4_u.opopen_downgrade;
7701 	OPEN_DOWNGRADE4res *resp = &resop->nfs_resop4_u.opopen_downgrade;
7702 	uint32_t access = args->share_access;
7703 	uint32_t deny = args->share_deny;
7704 	nfsstat4 status;
7705 	rfs4_state_t *sp;
7706 	rfs4_file_t *fp;
7707 	int fflags = 0;
7708 
7709 	DTRACE_NFSV4_2(op__open__downgrade__start, struct compound_state *, cs,
7710 	    OPEN_DOWNGRADE4args *, args);
7711 
7712 	if (cs->vp == NULL) {
7713 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7714 		goto out;
7715 	}
7716 
7717 	if (cs->vp->v_type != VREG) {
7718 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7719 		return;
7720 	}
7721 
7722 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7723 	if (status != NFS4_OK) {
7724 		*cs->statusp = resp->status = status;
7725 		goto out;
7726 	}
7727 
7728 	/* Ensure specified filehandle matches */
7729 	if (cs->vp != sp->rs_finfo->rf_vp) {
7730 		rfs4_state_rele(sp);
7731 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7732 		goto out;
7733 	}
7734 
7735 	/* hold off other access to open_owner while we tinker */
7736 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7737 
7738 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7739 	case NFS4_CHECK_STATEID_OKAY:
7740 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7741 		    resop) != NFS4_CHKSEQ_OKAY) {
7742 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7743 			goto end;
7744 		}
7745 		break;
7746 	case NFS4_CHECK_STATEID_OLD:
7747 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7748 		goto end;
7749 	case NFS4_CHECK_STATEID_BAD:
7750 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7751 		goto end;
7752 	case NFS4_CHECK_STATEID_EXPIRED:
7753 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7754 		goto end;
7755 	case NFS4_CHECK_STATEID_CLOSED:
7756 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7757 		goto end;
7758 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7759 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7760 		goto end;
7761 	case NFS4_CHECK_STATEID_REPLAY:
7762 		/* Check the sequence id for the open owner */
7763 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7764 		    resop)) {
7765 		case NFS4_CHKSEQ_OKAY:
7766 			/*
7767 			 * This is replayed stateid; if seqid matches
7768 			 * next expected, then client is using wrong seqid.
7769 			 */
7770 			/* fall through */
7771 		case NFS4_CHKSEQ_BAD:
7772 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7773 			goto end;
7774 		case NFS4_CHKSEQ_REPLAY:
7775 			/*
7776 			 * Note this case is the duplicate case so
7777 			 * resp->status is already set.
7778 			 */
7779 			*cs->statusp = resp->status;
7780 			rfs4_update_lease(sp->rs_owner->ro_client);
7781 			goto end;
7782 		}
7783 		break;
7784 	default:
7785 		ASSERT(FALSE);
7786 		break;
7787 	}
7788 
7789 	rfs4_dbe_lock(sp->rs_dbe);
7790 	/*
7791 	 * Check that the new access modes and deny modes are valid.
7792 	 * Check that no invalid bits are set.
7793 	 */
7794 	if ((access & ~(OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) ||
7795 	    (deny & ~(OPEN4_SHARE_DENY_READ | OPEN4_SHARE_DENY_WRITE))) {
7796 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7797 		rfs4_update_open_sequence(sp->rs_owner);
7798 		rfs4_dbe_unlock(sp->rs_dbe);
7799 		goto end;
7800 	}
7801 
7802 	/*
7803 	 * The new modes must be a subset of the current modes and
7804 	 * the access must specify at least one mode. To test that
7805 	 * the new mode is a subset of the current modes we bitwise
7806 	 * AND them together and check that the result equals the new
7807 	 * mode. For example:
7808 	 * New mode, access == R and current mode, sp->rs_open_access  == RW
7809 	 * access & sp->rs_open_access == R == access, so the new access mode
7810 	 * is valid. Consider access == RW, sp->rs_open_access = R
7811 	 * access & sp->rs_open_access == R != access, so the new access mode
7812 	 * is invalid.
7813 	 */
7814 	if ((access & sp->rs_open_access) != access ||
7815 	    (deny & sp->rs_open_deny) != deny ||
7816 	    (access &
7817 	    (OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) == 0) {
7818 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7819 		rfs4_update_open_sequence(sp->rs_owner);
7820 		rfs4_dbe_unlock(sp->rs_dbe);
7821 		goto end;
7822 	}
7823 
7824 	/*
7825 	 * Release any share locks associated with this stateID.
7826 	 * Strictly speaking, this violates the spec because the
7827 	 * spec effectively requires that open downgrade be atomic.
7828 	 * At present, fs_shrlock does not have this capability.
7829 	 */
7830 	(void) rfs4_unshare(sp);
7831 
7832 	status = rfs4_share(sp, access, deny);
7833 	if (status != NFS4_OK) {
7834 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7835 		rfs4_update_open_sequence(sp->rs_owner);
7836 		rfs4_dbe_unlock(sp->rs_dbe);
7837 		goto end;
7838 	}
7839 
7840 	fp = sp->rs_finfo;
7841 	rfs4_dbe_lock(fp->rf_dbe);
7842 
7843 	/*
7844 	 * If the current mode has deny read and the new mode
7845 	 * does not, decrement the number of deny read mode bits
7846 	 * and if it goes to zero turn off the deny read bit
7847 	 * on the file.
7848 	 */
7849 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_READ) &&
7850 	    (deny & OPEN4_SHARE_DENY_READ) == 0) {
7851 		fp->rf_deny_read--;
7852 		if (fp->rf_deny_read == 0)
7853 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
7854 	}
7855 
7856 	/*
7857 	 * If the current mode has deny write and the new mode
7858 	 * does not, decrement the number of deny write mode bits
7859 	 * and if it goes to zero turn off the deny write bit
7860 	 * on the file.
7861 	 */
7862 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) &&
7863 	    (deny & OPEN4_SHARE_DENY_WRITE) == 0) {
7864 		fp->rf_deny_write--;
7865 		if (fp->rf_deny_write == 0)
7866 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
7867 	}
7868 
7869 	/*
7870 	 * If the current mode has access read and the new mode
7871 	 * does not, decrement the number of access read mode bits
7872 	 * and if it goes to zero turn off the access read bit
7873 	 * on the file.  set fflags to FREAD for the call to
7874 	 * vn_open_downgrade().
7875 	 */
7876 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) &&
7877 	    (access & OPEN4_SHARE_ACCESS_READ) == 0) {
7878 		fp->rf_access_read--;
7879 		if (fp->rf_access_read == 0)
7880 			fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
7881 		fflags |= FREAD;
7882 	}
7883 
7884 	/*
7885 	 * If the current mode has access write and the new mode
7886 	 * does not, decrement the number of access write mode bits
7887 	 * and if it goes to zero turn off the access write bit
7888 	 * on the file.  set fflags to FWRITE for the call to
7889 	 * vn_open_downgrade().
7890 	 */
7891 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) &&
7892 	    (access & OPEN4_SHARE_ACCESS_WRITE) == 0) {
7893 		fp->rf_access_write--;
7894 		if (fp->rf_access_write == 0)
7895 			fp->rf_share_deny &= ~OPEN4_SHARE_ACCESS_WRITE;
7896 		fflags |= FWRITE;
7897 	}
7898 
7899 	/* Check that the file is still accessible */
7900 	ASSERT(fp->rf_share_access);
7901 
7902 	rfs4_dbe_unlock(fp->rf_dbe);
7903 
7904 	/* now set the new open access and deny modes */
7905 	sp->rs_open_access = access;
7906 	sp->rs_open_deny = deny;
7907 
7908 	/*
7909 	 * we successfully downgraded the share lock, now we need to downgrade
7910 	 * the open. it is possible that the downgrade was only for a deny
7911 	 * mode and we have nothing else to do.
7912 	 */
7913 	if ((fflags & (FREAD|FWRITE)) != 0)
7914 		vn_open_downgrade(cs->vp, fflags);
7915 
7916 	/* Update the stateid */
7917 	next_stateid(&sp->rs_stateid);
7918 	resp->open_stateid = sp->rs_stateid.stateid;
7919 
7920 	rfs4_dbe_unlock(sp->rs_dbe);
7921 
7922 	*cs->statusp = resp->status = NFS4_OK;
7923 	/* Update the lease */
7924 	rfs4_update_lease(sp->rs_owner->ro_client);
7925 	/* And the sequence */
7926 	rfs4_update_open_sequence(sp->rs_owner);
7927 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7928 
7929 end:
7930 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7931 	rfs4_state_rele(sp);
7932 out:
7933 	DTRACE_NFSV4_2(op__open__downgrade__done, struct compound_state *, cs,
7934 	    OPEN_DOWNGRADE4res *, resp);
7935 }
7936 
7937 /*
7938  * The logic behind this function is detailed in the NFSv4 RFC in the
7939  * SETCLIENTID operation description under IMPLEMENTATION.  Refer to
7940  * that section for explicit guidance to server behavior for
7941  * SETCLIENTID.
7942  */
7943 void
7944 rfs4_op_setclientid(nfs_argop4 *argop, nfs_resop4 *resop,
7945     struct svc_req *req, struct compound_state *cs)
7946 {
7947 	SETCLIENTID4args *args = &argop->nfs_argop4_u.opsetclientid;
7948 	SETCLIENTID4res *res = &resop->nfs_resop4_u.opsetclientid;
7949 	rfs4_client_t *cp, *newcp, *cp_confirmed, *cp_unconfirmed;
7950 	rfs4_clntip_t *ci;
7951 	bool_t create;
7952 	char *addr, *netid;
7953 	int len;
7954 
7955 	DTRACE_NFSV4_2(op__setclientid__start, struct compound_state *, cs,
7956 	    SETCLIENTID4args *, args);
7957 retry:
7958 	newcp = cp_confirmed = cp_unconfirmed = NULL;
7959 
7960 	/*
7961 	 * Save the caller's IP address
7962 	 */
7963 	args->client.cl_addr =
7964 	    (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
7965 
7966 	/*
7967 	 * Record if it is a Solaris client that cannot handle referrals.
7968 	 */
7969 	if (strstr(args->client.id_val, "Solaris") &&
7970 	    !strstr(args->client.id_val, "+referrals")) {
7971 		/* Add a "yes, it's downrev" record */
7972 		create = TRUE;
7973 		ci = rfs4_find_clntip(args->client.cl_addr, &create);
7974 		ASSERT(ci != NULL);
7975 		rfs4_dbe_rele(ci->ri_dbe);
7976 	} else {
7977 		/* Remove any previous record */
7978 		rfs4_invalidate_clntip(args->client.cl_addr);
7979 	}
7980 
7981 	/*
7982 	 * In search of an EXISTING client matching the incoming
7983 	 * request to establish a new client identifier at the server
7984 	 */
7985 	create = TRUE;
7986 	cp = rfs4_findclient(&args->client, &create, NULL);
7987 
7988 	/* Should never happen */
7989 	ASSERT(cp != NULL);
7990 
7991 	if (cp == NULL) {
7992 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
7993 		goto out;
7994 	}
7995 
7996 	/*
7997 	 * Easiest case. Client identifier is newly created and is
7998 	 * unconfirmed.  Also note that for this case, no other
7999 	 * entries exist for the client identifier.  Nothing else to
8000 	 * check.  Just setup the response and respond.
8001 	 */
8002 	if (create) {
8003 		*cs->statusp = res->status = NFS4_OK;
8004 		res->SETCLIENTID4res_u.resok4.clientid = cp->rc_clientid;
8005 		res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8006 		    cp->rc_confirm_verf;
8007 		/* Setup callback information; CB_NULL confirmation later */
8008 		rfs4_client_setcb(cp, &args->callback, args->callback_ident);
8009 
8010 		rfs4_client_rele(cp);
8011 		goto out;
8012 	}
8013 
8014 	/*
8015 	 * An existing, confirmed client may exist but it may not have
8016 	 * been active for at least one lease period.  If so, then
8017 	 * "close" the client and create a new client identifier
8018 	 */
8019 	if (rfs4_lease_expired(cp)) {
8020 		rfs4_client_close(cp);
8021 		goto retry;
8022 	}
8023 
8024 	if (cp->rc_need_confirm == TRUE)
8025 		cp_unconfirmed = cp;
8026 	else
8027 		cp_confirmed = cp;
8028 
8029 	cp = NULL;
8030 
8031 	/*
8032 	 * We have a confirmed client, now check for an
8033 	 * unconfimred entry
8034 	 */
8035 	if (cp_confirmed) {
8036 		/* If creds don't match then client identifier is inuse */
8037 		if (!creds_ok(cp_confirmed->rc_cr_set, req, cs)) {
8038 			rfs4_cbinfo_t *cbp;
8039 			/*
8040 			 * Some one else has established this client
8041 			 * id. Try and say * who they are. We will use
8042 			 * the call back address supplied by * the
8043 			 * first client.
8044 			 */
8045 			*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8046 
8047 			addr = netid = NULL;
8048 
8049 			cbp = &cp_confirmed->rc_cbinfo;
8050 			if (cbp->cb_callback.cb_location.r_addr &&
8051 			    cbp->cb_callback.cb_location.r_netid) {
8052 				cb_client4 *cbcp = &cbp->cb_callback;
8053 
8054 				len = strlen(cbcp->cb_location.r_addr)+1;
8055 				addr = kmem_alloc(len, KM_SLEEP);
8056 				bcopy(cbcp->cb_location.r_addr, addr, len);
8057 				len = strlen(cbcp->cb_location.r_netid)+1;
8058 				netid = kmem_alloc(len, KM_SLEEP);
8059 				bcopy(cbcp->cb_location.r_netid, netid, len);
8060 			}
8061 
8062 			res->SETCLIENTID4res_u.client_using.r_addr = addr;
8063 			res->SETCLIENTID4res_u.client_using.r_netid = netid;
8064 
8065 			rfs4_client_rele(cp_confirmed);
8066 		}
8067 
8068 		/*
8069 		 * Confirmed, creds match, and verifier matches; must
8070 		 * be an update of the callback info
8071 		 */
8072 		if (cp_confirmed->rc_nfs_client.verifier ==
8073 		    args->client.verifier) {
8074 			/* Setup callback information */
8075 			rfs4_client_setcb(cp_confirmed, &args->callback,
8076 			    args->callback_ident);
8077 
8078 			/* everything okay -- move ahead */
8079 			*cs->statusp = res->status = NFS4_OK;
8080 			res->SETCLIENTID4res_u.resok4.clientid =
8081 			    cp_confirmed->rc_clientid;
8082 
8083 			/* update the confirm_verifier and return it */
8084 			rfs4_client_scv_next(cp_confirmed);
8085 			res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8086 			    cp_confirmed->rc_confirm_verf;
8087 
8088 			rfs4_client_rele(cp_confirmed);
8089 			goto out;
8090 		}
8091 
8092 		/*
8093 		 * Creds match but the verifier doesn't.  Must search
8094 		 * for an unconfirmed client that would be replaced by
8095 		 * this request.
8096 		 */
8097 		create = FALSE;
8098 		cp_unconfirmed = rfs4_findclient(&args->client, &create,
8099 		    cp_confirmed);
8100 	}
8101 
8102 	/*
8103 	 * At this point, we have taken care of the brand new client
8104 	 * struct, INUSE case, update of an existing, and confirmed
8105 	 * client struct.
8106 	 */
8107 
8108 	/*
8109 	 * check to see if things have changed while we originally
8110 	 * picked up the client struct.  If they have, then return and
8111 	 * retry the processing of this SETCLIENTID request.
8112 	 */
8113 	if (cp_unconfirmed) {
8114 		rfs4_dbe_lock(cp_unconfirmed->rc_dbe);
8115 		if (!cp_unconfirmed->rc_need_confirm) {
8116 			rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8117 			rfs4_client_rele(cp_unconfirmed);
8118 			if (cp_confirmed)
8119 				rfs4_client_rele(cp_confirmed);
8120 			goto retry;
8121 		}
8122 		/* do away with the old unconfirmed one */
8123 		rfs4_dbe_invalidate(cp_unconfirmed->rc_dbe);
8124 		rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8125 		rfs4_client_rele(cp_unconfirmed);
8126 		cp_unconfirmed = NULL;
8127 	}
8128 
8129 	/*
8130 	 * This search will temporarily hide the confirmed client
8131 	 * struct while a new client struct is created as the
8132 	 * unconfirmed one.
8133 	 */
8134 	create = TRUE;
8135 	newcp = rfs4_findclient(&args->client, &create, cp_confirmed);
8136 
8137 	ASSERT(newcp != NULL);
8138 
8139 	if (newcp == NULL) {
8140 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8141 		rfs4_client_rele(cp_confirmed);
8142 		goto out;
8143 	}
8144 
8145 	/*
8146 	 * If one was not created, then a similar request must be in
8147 	 * process so release and start over with this one
8148 	 */
8149 	if (create != TRUE) {
8150 		rfs4_client_rele(newcp);
8151 		if (cp_confirmed)
8152 			rfs4_client_rele(cp_confirmed);
8153 		goto retry;
8154 	}
8155 
8156 	*cs->statusp = res->status = NFS4_OK;
8157 	res->SETCLIENTID4res_u.resok4.clientid = newcp->rc_clientid;
8158 	res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8159 	    newcp->rc_confirm_verf;
8160 	/* Setup callback information; CB_NULL confirmation later */
8161 	rfs4_client_setcb(newcp, &args->callback, args->callback_ident);
8162 
8163 	newcp->rc_cp_confirmed = cp_confirmed;
8164 
8165 	rfs4_client_rele(newcp);
8166 
8167 out:
8168 	DTRACE_NFSV4_2(op__setclientid__done, struct compound_state *, cs,
8169 	    SETCLIENTID4res *, res);
8170 }
8171 
8172 /*ARGSUSED*/
8173 void
8174 rfs4_op_setclientid_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
8175     struct svc_req *req, struct compound_state *cs)
8176 {
8177 	SETCLIENTID_CONFIRM4args *args =
8178 	    &argop->nfs_argop4_u.opsetclientid_confirm;
8179 	SETCLIENTID_CONFIRM4res *res =
8180 	    &resop->nfs_resop4_u.opsetclientid_confirm;
8181 	rfs4_client_t *cp, *cptoclose = NULL;
8182 
8183 	DTRACE_NFSV4_2(op__setclientid__confirm__start,
8184 	    struct compound_state *, cs,
8185 	    SETCLIENTID_CONFIRM4args *, args);
8186 
8187 	*cs->statusp = res->status = NFS4_OK;
8188 
8189 	cp = rfs4_findclient_by_id(args->clientid, TRUE);
8190 
8191 	if (cp == NULL) {
8192 		*cs->statusp = res->status =
8193 		    rfs4_check_clientid(&args->clientid, 1);
8194 		goto out;
8195 	}
8196 
8197 	if (!creds_ok(cp, req, cs)) {
8198 		*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8199 		rfs4_client_rele(cp);
8200 		goto out;
8201 	}
8202 
8203 	/* If the verifier doesn't match, the record doesn't match */
8204 	if (cp->rc_confirm_verf != args->setclientid_confirm) {
8205 		*cs->statusp = res->status = NFS4ERR_STALE_CLIENTID;
8206 		rfs4_client_rele(cp);
8207 		goto out;
8208 	}
8209 
8210 	rfs4_dbe_lock(cp->rc_dbe);
8211 	cp->rc_need_confirm = FALSE;
8212 	if (cp->rc_cp_confirmed) {
8213 		cptoclose = cp->rc_cp_confirmed;
8214 		cptoclose->rc_ss_remove = 1;
8215 		cp->rc_cp_confirmed = NULL;
8216 	}
8217 
8218 	/*
8219 	 * Update the client's associated server instance, if it's changed
8220 	 * since the client was created.
8221 	 */
8222 	if (rfs4_servinst(cp) != rfs4_cur_servinst)
8223 		rfs4_servinst_assign(cp, rfs4_cur_servinst);
8224 
8225 	/*
8226 	 * Record clientid in stable storage.
8227 	 * Must be done after server instance has been assigned.
8228 	 */
8229 	rfs4_ss_clid(cp);
8230 
8231 	rfs4_dbe_unlock(cp->rc_dbe);
8232 
8233 	if (cptoclose)
8234 		/* don't need to rele, client_close does it */
8235 		rfs4_client_close(cptoclose);
8236 
8237 	/* If needed, initiate CB_NULL call for callback path */
8238 	rfs4_deleg_cb_check(cp);
8239 	rfs4_update_lease(cp);
8240 
8241 	/*
8242 	 * Check to see if client can perform reclaims
8243 	 */
8244 	rfs4_ss_chkclid(cp);
8245 
8246 	rfs4_client_rele(cp);
8247 
8248 out:
8249 	DTRACE_NFSV4_2(op__setclientid__confirm__done,
8250 	    struct compound_state *, cs,
8251 	    SETCLIENTID_CONFIRM4 *, res);
8252 }
8253 
8254 
8255 /*ARGSUSED*/
8256 void
8257 rfs4_op_close(nfs_argop4 *argop, nfs_resop4 *resop,
8258     struct svc_req *req, struct compound_state *cs)
8259 {
8260 	CLOSE4args *args = &argop->nfs_argop4_u.opclose;
8261 	CLOSE4res *resp = &resop->nfs_resop4_u.opclose;
8262 	rfs4_state_t *sp;
8263 	nfsstat4 status;
8264 
8265 	DTRACE_NFSV4_2(op__close__start, struct compound_state *, cs,
8266 	    CLOSE4args *, args);
8267 
8268 	if (cs->vp == NULL) {
8269 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8270 		goto out;
8271 	}
8272 
8273 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_INVALID);
8274 	if (status != NFS4_OK) {
8275 		*cs->statusp = resp->status = status;
8276 		goto out;
8277 	}
8278 
8279 	/* Ensure specified filehandle matches */
8280 	if (cs->vp != sp->rs_finfo->rf_vp) {
8281 		rfs4_state_rele(sp);
8282 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8283 		goto out;
8284 	}
8285 
8286 	/* hold off other access to open_owner while we tinker */
8287 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
8288 
8289 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
8290 	case NFS4_CHECK_STATEID_OKAY:
8291 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8292 		    resop) != NFS4_CHKSEQ_OKAY) {
8293 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8294 			goto end;
8295 		}
8296 		break;
8297 	case NFS4_CHECK_STATEID_OLD:
8298 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8299 		goto end;
8300 	case NFS4_CHECK_STATEID_BAD:
8301 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8302 		goto end;
8303 	case NFS4_CHECK_STATEID_EXPIRED:
8304 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8305 		goto end;
8306 	case NFS4_CHECK_STATEID_CLOSED:
8307 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8308 		goto end;
8309 	case NFS4_CHECK_STATEID_UNCONFIRMED:
8310 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8311 		goto end;
8312 	case NFS4_CHECK_STATEID_REPLAY:
8313 		/* Check the sequence id for the open owner */
8314 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8315 		    resop)) {
8316 		case NFS4_CHKSEQ_OKAY:
8317 			/*
8318 			 * This is replayed stateid; if seqid matches
8319 			 * next expected, then client is using wrong seqid.
8320 			 */
8321 			/* FALL THROUGH */
8322 		case NFS4_CHKSEQ_BAD:
8323 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8324 			goto end;
8325 		case NFS4_CHKSEQ_REPLAY:
8326 			/*
8327 			 * Note this case is the duplicate case so
8328 			 * resp->status is already set.
8329 			 */
8330 			*cs->statusp = resp->status;
8331 			rfs4_update_lease(sp->rs_owner->ro_client);
8332 			goto end;
8333 		}
8334 		break;
8335 	default:
8336 		ASSERT(FALSE);
8337 		break;
8338 	}
8339 
8340 	rfs4_dbe_lock(sp->rs_dbe);
8341 
8342 	/* Update the stateid. */
8343 	next_stateid(&sp->rs_stateid);
8344 	resp->open_stateid = sp->rs_stateid.stateid;
8345 
8346 	rfs4_dbe_unlock(sp->rs_dbe);
8347 
8348 	rfs4_update_lease(sp->rs_owner->ro_client);
8349 	rfs4_update_open_sequence(sp->rs_owner);
8350 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8351 
8352 	rfs4_state_close(sp, FALSE, FALSE, cs->cr);
8353 
8354 	*cs->statusp = resp->status = status;
8355 
8356 end:
8357 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
8358 	rfs4_state_rele(sp);
8359 out:
8360 	DTRACE_NFSV4_2(op__close__done, struct compound_state *, cs,
8361 	    CLOSE4res *, resp);
8362 }
8363 
8364 /*
8365  * Manage the counts on the file struct and close all file locks
8366  */
8367 /*ARGSUSED*/
8368 void
8369 rfs4_release_share_lock_state(rfs4_state_t *sp, cred_t *cr,
8370     bool_t close_of_client)
8371 {
8372 	rfs4_file_t *fp = sp->rs_finfo;
8373 	rfs4_lo_state_t *lsp;
8374 	int fflags = 0;
8375 
8376 	/*
8377 	 * If this call is part of the larger closing down of client
8378 	 * state then it is just easier to release all locks
8379 	 * associated with this client instead of going through each
8380 	 * individual file and cleaning locks there.
8381 	 */
8382 	if (close_of_client) {
8383 		if (sp->rs_owner->ro_client->rc_unlksys_completed == FALSE &&
8384 		    !list_is_empty(&sp->rs_lostatelist) &&
8385 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID) {
8386 			/* Is the PxFS kernel module loaded? */
8387 			if (lm_remove_file_locks != NULL) {
8388 				int new_sysid;
8389 
8390 				/* Encode the cluster nodeid in new sysid */
8391 				new_sysid = sp->rs_owner->ro_client->rc_sysidt;
8392 				lm_set_nlmid_flk(&new_sysid);
8393 
8394 				/*
8395 				 * This PxFS routine removes file locks for a
8396 				 * client over all nodes of a cluster.
8397 				 */
8398 				NFS4_DEBUG(rfs4_debug, (CE_NOTE,
8399 				    "lm_remove_file_locks(sysid=0x%x)\n",
8400 				    new_sysid));
8401 				(*lm_remove_file_locks)(new_sysid);
8402 			} else {
8403 				struct flock64 flk;
8404 
8405 				/* Release all locks for this client */
8406 				flk.l_type = F_UNLKSYS;
8407 				flk.l_whence = 0;
8408 				flk.l_start = 0;
8409 				flk.l_len = 0;
8410 				flk.l_sysid =
8411 				    sp->rs_owner->ro_client->rc_sysidt;
8412 				flk.l_pid = 0;
8413 				(void) VOP_FRLOCK(sp->rs_finfo->rf_vp, F_SETLK,
8414 				    &flk, F_REMOTELOCK | FREAD | FWRITE,
8415 				    (u_offset_t)0, NULL, CRED(), NULL);
8416 			}
8417 
8418 			sp->rs_owner->ro_client->rc_unlksys_completed = TRUE;
8419 		}
8420 	}
8421 
8422 	/*
8423 	 * Release all locks on this file by this lock owner or at
8424 	 * least mark the locks as having been released
8425 	 */
8426 	for (lsp = list_head(&sp->rs_lostatelist); lsp != NULL;
8427 	    lsp = list_next(&sp->rs_lostatelist, lsp)) {
8428 		lsp->rls_locks_cleaned = TRUE;
8429 
8430 		/* Was this already taken care of above? */
8431 		if (!close_of_client &&
8432 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8433 			(void) cleanlocks(sp->rs_finfo->rf_vp,
8434 			    lsp->rls_locker->rl_pid,
8435 			    lsp->rls_locker->rl_client->rc_sysidt);
8436 	}
8437 
8438 	/*
8439 	 * Release any shrlocks associated with this open state ID.
8440 	 * This must be done before the rfs4_state gets marked closed.
8441 	 */
8442 	if (sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8443 		(void) rfs4_unshare(sp);
8444 
8445 	if (sp->rs_open_access) {
8446 		rfs4_dbe_lock(fp->rf_dbe);
8447 
8448 		/*
8449 		 * Decrement the count for each access and deny bit that this
8450 		 * state has contributed to the file.
8451 		 * If the file counts go to zero
8452 		 * clear the appropriate bit in the appropriate mask.
8453 		 */
8454 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) {
8455 			fp->rf_access_read--;
8456 			fflags |= FREAD;
8457 			if (fp->rf_access_read == 0)
8458 				fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
8459 		}
8460 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) {
8461 			fp->rf_access_write--;
8462 			fflags |= FWRITE;
8463 			if (fp->rf_access_write == 0)
8464 				fp->rf_share_access &=
8465 				    ~OPEN4_SHARE_ACCESS_WRITE;
8466 		}
8467 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_READ) {
8468 			fp->rf_deny_read--;
8469 			if (fp->rf_deny_read == 0)
8470 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
8471 		}
8472 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) {
8473 			fp->rf_deny_write--;
8474 			if (fp->rf_deny_write == 0)
8475 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
8476 		}
8477 
8478 		(void) VOP_CLOSE(fp->rf_vp, fflags, 1, (offset_t)0, cr, NULL);
8479 
8480 		rfs4_dbe_unlock(fp->rf_dbe);
8481 
8482 		sp->rs_open_access = 0;
8483 		sp->rs_open_deny = 0;
8484 	}
8485 }
8486 
8487 /*
8488  * lock_denied: Fill in a LOCK4deneid structure given an flock64 structure.
8489  */
8490 static nfsstat4
8491 lock_denied(LOCK4denied *dp, struct flock64 *flk)
8492 {
8493 	rfs4_lockowner_t *lo;
8494 	rfs4_client_t *cp;
8495 	uint32_t len;
8496 
8497 	lo = rfs4_findlockowner_by_pid(flk->l_pid);
8498 	if (lo != NULL) {
8499 		cp = lo->rl_client;
8500 		if (rfs4_lease_expired(cp)) {
8501 			rfs4_lockowner_rele(lo);
8502 			rfs4_dbe_hold(cp->rc_dbe);
8503 			rfs4_client_close(cp);
8504 			return (NFS4ERR_EXPIRED);
8505 		}
8506 		dp->owner.clientid = lo->rl_owner.clientid;
8507 		len = lo->rl_owner.owner_len;
8508 		dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8509 		bcopy(lo->rl_owner.owner_val, dp->owner.owner_val, len);
8510 		dp->owner.owner_len = len;
8511 		rfs4_lockowner_rele(lo);
8512 		goto finish;
8513 	}
8514 
8515 	/*
8516 	 * Its not a NFS4 lock. We take advantage that the upper 32 bits
8517 	 * of the client id contain the boot time for a NFS4 lock. So we
8518 	 * fabricate and identity by setting clientid to the sysid, and
8519 	 * the lock owner to the pid.
8520 	 */
8521 	dp->owner.clientid = flk->l_sysid;
8522 	len = sizeof (pid_t);
8523 	dp->owner.owner_len = len;
8524 	dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8525 	bcopy(&flk->l_pid, dp->owner.owner_val, len);
8526 finish:
8527 	dp->offset = flk->l_start;
8528 	dp->length = flk->l_len;
8529 
8530 	if (flk->l_type == F_RDLCK)
8531 		dp->locktype = READ_LT;
8532 	else if (flk->l_type == F_WRLCK)
8533 		dp->locktype = WRITE_LT;
8534 	else
8535 		return (NFS4ERR_INVAL);	/* no mapping from POSIX ltype to v4 */
8536 
8537 	return (NFS4_OK);
8538 }
8539 
8540 static int
8541 setlock(vnode_t *vp, struct flock64 *flock, int flag, cred_t *cred)
8542 {
8543 	int error;
8544 	struct flock64 flk;
8545 	int i;
8546 	clock_t delaytime;
8547 	int cmd;
8548 
8549 	cmd = nbl_need_check(vp) ? F_SETLK_NBMAND : F_SETLK;
8550 retry:
8551 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
8552 
8553 	for (i = 0; i < rfs4_maxlock_tries; i++) {
8554 		LOCK_PRINT(rfs4_debug, "setlock", cmd, flock);
8555 		error = VOP_FRLOCK(vp, cmd,
8556 		    flock, flag, (u_offset_t)0, NULL, cred, NULL);
8557 
8558 		if (error != EAGAIN && error != EACCES)
8559 			break;
8560 
8561 		if (i < rfs4_maxlock_tries - 1) {
8562 			delay(delaytime);
8563 			delaytime *= 2;
8564 		}
8565 	}
8566 
8567 	if (error == EAGAIN || error == EACCES) {
8568 		/* Get the owner of the lock */
8569 		flk = *flock;
8570 		LOCK_PRINT(rfs4_debug, "setlock", F_GETLK, &flk);
8571 		if (VOP_FRLOCK(vp, F_GETLK, &flk, flag,
8572 		    (u_offset_t)0, NULL, cred, NULL) == 0) {
8573 			if (flk.l_type == F_UNLCK) {
8574 				/* No longer locked, retry */
8575 				goto retry;
8576 			}
8577 			*flock = flk;
8578 			LOCK_PRINT(rfs4_debug, "setlock(blocking lock)",
8579 			    F_GETLK, &flk);
8580 		}
8581 	}
8582 
8583 	return (error);
8584 }
8585 
8586 /*ARGSUSED*/
8587 static nfsstat4
8588 rfs4_do_lock(rfs4_lo_state_t *lsp, nfs_lock_type4 locktype,
8589     offset4 offset, length4 length, cred_t *cred, nfs_resop4 *resop)
8590 {
8591 	nfsstat4 status;
8592 	rfs4_lockowner_t *lo = lsp->rls_locker;
8593 	rfs4_state_t *sp = lsp->rls_state;
8594 	struct flock64 flock;
8595 	int16_t ltype;
8596 	int flag;
8597 	int error;
8598 	sysid_t sysid;
8599 	LOCK4res *lres;
8600 
8601 	if (rfs4_lease_expired(lo->rl_client)) {
8602 		return (NFS4ERR_EXPIRED);
8603 	}
8604 
8605 	if ((status = rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
8606 		return (status);
8607 
8608 	/* Check for zero length. To lock to end of file use all ones for V4 */
8609 	if (length == 0)
8610 		return (NFS4ERR_INVAL);
8611 	else if (length == (length4)(~0))
8612 		length = 0;		/* Posix to end of file  */
8613 
8614 retry:
8615 	rfs4_dbe_lock(sp->rs_dbe);
8616 	if (sp->rs_closed) {
8617 		rfs4_dbe_unlock(sp->rs_dbe);
8618 		return (NFS4ERR_OLD_STATEID);
8619 	}
8620 
8621 	if (resop->resop != OP_LOCKU) {
8622 		switch (locktype) {
8623 		case READ_LT:
8624 		case READW_LT:
8625 			if ((sp->rs_share_access
8626 			    & OPEN4_SHARE_ACCESS_READ) == 0) {
8627 				rfs4_dbe_unlock(sp->rs_dbe);
8628 
8629 				return (NFS4ERR_OPENMODE);
8630 			}
8631 			ltype = F_RDLCK;
8632 			break;
8633 		case WRITE_LT:
8634 		case WRITEW_LT:
8635 			if ((sp->rs_share_access
8636 			    & OPEN4_SHARE_ACCESS_WRITE) == 0) {
8637 				rfs4_dbe_unlock(sp->rs_dbe);
8638 
8639 				return (NFS4ERR_OPENMODE);
8640 			}
8641 			ltype = F_WRLCK;
8642 			break;
8643 		}
8644 	} else
8645 		ltype = F_UNLCK;
8646 
8647 	flock.l_type = ltype;
8648 	flock.l_whence = 0;		/* SEEK_SET */
8649 	flock.l_start = offset;
8650 	flock.l_len = length;
8651 	flock.l_sysid = sysid;
8652 	flock.l_pid = lsp->rls_locker->rl_pid;
8653 
8654 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
8655 	if (flock.l_len < 0 || flock.l_start < 0) {
8656 		rfs4_dbe_unlock(sp->rs_dbe);
8657 		return (NFS4ERR_INVAL);
8658 	}
8659 
8660 	/*
8661 	 * N.B. FREAD has the same value as OPEN4_SHARE_ACCESS_READ and
8662 	 * FWRITE has the same value as OPEN4_SHARE_ACCESS_WRITE.
8663 	 */
8664 	flag = (int)sp->rs_share_access | F_REMOTELOCK;
8665 
8666 	error = setlock(sp->rs_finfo->rf_vp, &flock, flag, cred);
8667 	if (error == 0) {
8668 		rfs4_dbe_lock(lsp->rls_dbe);
8669 		next_stateid(&lsp->rls_lockid);
8670 		rfs4_dbe_unlock(lsp->rls_dbe);
8671 	}
8672 
8673 	rfs4_dbe_unlock(sp->rs_dbe);
8674 
8675 	/*
8676 	 * N.B. We map error values to nfsv4 errors. This is differrent
8677 	 * than puterrno4 routine.
8678 	 */
8679 	switch (error) {
8680 	case 0:
8681 		status = NFS4_OK;
8682 		break;
8683 	case EAGAIN:
8684 	case EACCES:		/* Old value */
8685 		/* Can only get here if op is OP_LOCK */
8686 		ASSERT(resop->resop == OP_LOCK);
8687 		lres = &resop->nfs_resop4_u.oplock;
8688 		status = NFS4ERR_DENIED;
8689 		if (lock_denied(&lres->LOCK4res_u.denied, &flock)
8690 		    == NFS4ERR_EXPIRED)
8691 			goto retry;
8692 		break;
8693 	case ENOLCK:
8694 		status = NFS4ERR_DELAY;
8695 		break;
8696 	case EOVERFLOW:
8697 		status = NFS4ERR_INVAL;
8698 		break;
8699 	case EINVAL:
8700 		status = NFS4ERR_NOTSUPP;
8701 		break;
8702 	default:
8703 		status = NFS4ERR_SERVERFAULT;
8704 		break;
8705 	}
8706 
8707 	return (status);
8708 }
8709 
8710 /*ARGSUSED*/
8711 void
8712 rfs4_op_lock(nfs_argop4 *argop, nfs_resop4 *resop,
8713     struct svc_req *req, struct compound_state *cs)
8714 {
8715 	LOCK4args *args = &argop->nfs_argop4_u.oplock;
8716 	LOCK4res *resp = &resop->nfs_resop4_u.oplock;
8717 	nfsstat4 status;
8718 	stateid4 *stateid;
8719 	rfs4_lockowner_t *lo;
8720 	rfs4_client_t *cp;
8721 	rfs4_state_t *sp = NULL;
8722 	rfs4_lo_state_t *lsp = NULL;
8723 	bool_t ls_sw_held = FALSE;
8724 	bool_t create = TRUE;
8725 	bool_t lcreate = TRUE;
8726 	bool_t dup_lock = FALSE;
8727 	int rc;
8728 
8729 	DTRACE_NFSV4_2(op__lock__start, struct compound_state *, cs,
8730 	    LOCK4args *, args);
8731 
8732 	if (cs->vp == NULL) {
8733 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8734 		DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8735 		    cs, LOCK4res *, resp);
8736 		return;
8737 	}
8738 
8739 	if (args->locker.new_lock_owner) {
8740 		/* Create a new lockowner for this instance */
8741 		open_to_lock_owner4 *olo = &args->locker.locker4_u.open_owner;
8742 
8743 		NFS4_DEBUG(rfs4_debug, (CE_NOTE, "Creating new lock owner"));
8744 
8745 		stateid = &olo->open_stateid;
8746 		status = rfs4_get_state(stateid, &sp, RFS4_DBS_VALID);
8747 		if (status != NFS4_OK) {
8748 			NFS4_DEBUG(rfs4_debug,
8749 			    (CE_NOTE, "Get state failed in lock %d", status));
8750 			*cs->statusp = resp->status = status;
8751 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8752 			    cs, LOCK4res *, resp);
8753 			return;
8754 		}
8755 
8756 		/* Ensure specified filehandle matches */
8757 		if (cs->vp != sp->rs_finfo->rf_vp) {
8758 			rfs4_state_rele(sp);
8759 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8760 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8761 			    cs, LOCK4res *, resp);
8762 			return;
8763 		}
8764 
8765 		/* hold off other access to open_owner while we tinker */
8766 		rfs4_sw_enter(&sp->rs_owner->ro_sw);
8767 
8768 		switch (rc = rfs4_check_stateid_seqid(sp, stateid)) {
8769 		case NFS4_CHECK_STATEID_OLD:
8770 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8771 			goto end;
8772 		case NFS4_CHECK_STATEID_BAD:
8773 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8774 			goto end;
8775 		case NFS4_CHECK_STATEID_EXPIRED:
8776 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8777 			goto end;
8778 		case NFS4_CHECK_STATEID_UNCONFIRMED:
8779 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8780 			goto end;
8781 		case NFS4_CHECK_STATEID_CLOSED:
8782 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8783 			goto end;
8784 		case NFS4_CHECK_STATEID_OKAY:
8785 		case NFS4_CHECK_STATEID_REPLAY:
8786 			switch (rfs4_check_olo_seqid(olo->open_seqid,
8787 			    sp->rs_owner, resop)) {
8788 			case NFS4_CHKSEQ_OKAY:
8789 				if (rc == NFS4_CHECK_STATEID_OKAY)
8790 					break;
8791 				/*
8792 				 * This is replayed stateid; if seqid
8793 				 * matches next expected, then client
8794 				 * is using wrong seqid.
8795 				 */
8796 				/* FALLTHROUGH */
8797 			case NFS4_CHKSEQ_BAD:
8798 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8799 				goto end;
8800 			case NFS4_CHKSEQ_REPLAY:
8801 				/* This is a duplicate LOCK request */
8802 				dup_lock = TRUE;
8803 
8804 				/*
8805 				 * For a duplicate we do not want to
8806 				 * create a new lockowner as it should
8807 				 * already exist.
8808 				 * Turn off the lockowner create flag.
8809 				 */
8810 				lcreate = FALSE;
8811 			}
8812 			break;
8813 		}
8814 
8815 		lo = rfs4_findlockowner(&olo->lock_owner, &lcreate);
8816 		if (lo == NULL) {
8817 			NFS4_DEBUG(rfs4_debug,
8818 			    (CE_NOTE, "rfs4_op_lock: no lock owner"));
8819 			*cs->statusp = resp->status = NFS4ERR_RESOURCE;
8820 			goto end;
8821 		}
8822 
8823 		lsp = rfs4_findlo_state_by_owner(lo, sp, &create);
8824 		if (lsp == NULL) {
8825 			rfs4_update_lease(sp->rs_owner->ro_client);
8826 			/*
8827 			 * Only update theh open_seqid if this is not
8828 			 * a duplicate request
8829 			 */
8830 			if (dup_lock == FALSE) {
8831 				rfs4_update_open_sequence(sp->rs_owner);
8832 			}
8833 
8834 			NFS4_DEBUG(rfs4_debug,
8835 			    (CE_NOTE, "rfs4_op_lock: no state"));
8836 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
8837 			rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8838 			rfs4_lockowner_rele(lo);
8839 			goto end;
8840 		}
8841 
8842 		/*
8843 		 * This is the new_lock_owner branch and the client is
8844 		 * supposed to be associating a new lock_owner with
8845 		 * the open file at this point.  If we find that a
8846 		 * lock_owner/state association already exists and a
8847 		 * successful LOCK request was returned to the client,
8848 		 * an error is returned to the client since this is
8849 		 * not appropriate.  The client should be using the
8850 		 * existing lock_owner branch.
8851 		 */
8852 		if (dup_lock == FALSE && create == FALSE) {
8853 			if (lsp->rls_lock_completed == TRUE) {
8854 				*cs->statusp =
8855 				    resp->status = NFS4ERR_BAD_SEQID;
8856 				rfs4_lockowner_rele(lo);
8857 				goto end;
8858 			}
8859 		}
8860 
8861 		rfs4_update_lease(sp->rs_owner->ro_client);
8862 
8863 		/*
8864 		 * Only update theh open_seqid if this is not
8865 		 * a duplicate request
8866 		 */
8867 		if (dup_lock == FALSE) {
8868 			rfs4_update_open_sequence(sp->rs_owner);
8869 		}
8870 
8871 		/*
8872 		 * If this is a duplicate lock request, just copy the
8873 		 * previously saved reply and return.
8874 		 */
8875 		if (dup_lock == TRUE) {
8876 			/* verify that lock_seqid's match */
8877 			if (lsp->rls_seqid != olo->lock_seqid) {
8878 				NFS4_DEBUG(rfs4_debug,
8879 				    (CE_NOTE, "rfs4_op_lock: Dup-Lock seqid bad"
8880 				    "lsp->seqid=%d old->seqid=%d",
8881 				    lsp->rls_seqid, olo->lock_seqid));
8882 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8883 			} else {
8884 				rfs4_copy_reply(resop, &lsp->rls_reply);
8885 				/*
8886 				 * Make sure to copy the just
8887 				 * retrieved reply status into the
8888 				 * overall compound status
8889 				 */
8890 				*cs->statusp = resp->status;
8891 			}
8892 			rfs4_lockowner_rele(lo);
8893 			goto end;
8894 		}
8895 
8896 		rfs4_dbe_lock(lsp->rls_dbe);
8897 
8898 		/* Make sure to update the lock sequence id */
8899 		lsp->rls_seqid = olo->lock_seqid;
8900 
8901 		NFS4_DEBUG(rfs4_debug,
8902 		    (CE_NOTE, "Lock seqid established as %d", lsp->rls_seqid));
8903 
8904 		/*
8905 		 * This is used to signify the newly created lockowner
8906 		 * stateid and its sequence number.  The checks for
8907 		 * sequence number and increment don't occur on the
8908 		 * very first lock request for a lockowner.
8909 		 */
8910 		lsp->rls_skip_seqid_check = TRUE;
8911 
8912 		/* hold off other access to lsp while we tinker */
8913 		rfs4_sw_enter(&lsp->rls_sw);
8914 		ls_sw_held = TRUE;
8915 
8916 		rfs4_dbe_unlock(lsp->rls_dbe);
8917 
8918 		rfs4_lockowner_rele(lo);
8919 	} else {
8920 		stateid = &args->locker.locker4_u.lock_owner.lock_stateid;
8921 		/* get lsp and hold the lock on the underlying file struct */
8922 		if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE))
8923 		    != NFS4_OK) {
8924 			*cs->statusp = resp->status = status;
8925 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8926 			    cs, LOCK4res *, resp);
8927 			return;
8928 		}
8929 		create = FALSE;	/* We didn't create lsp */
8930 
8931 		/* Ensure specified filehandle matches */
8932 		if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
8933 			rfs4_lo_state_rele(lsp, TRUE);
8934 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8935 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8936 			    cs, LOCK4res *, resp);
8937 			return;
8938 		}
8939 
8940 		/* hold off other access to lsp while we tinker */
8941 		rfs4_sw_enter(&lsp->rls_sw);
8942 		ls_sw_held = TRUE;
8943 
8944 		switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
8945 		/*
8946 		 * The stateid looks like it was okay (expected to be
8947 		 * the next one)
8948 		 */
8949 		case NFS4_CHECK_STATEID_OKAY:
8950 			/*
8951 			 * The sequence id is now checked.  Determine
8952 			 * if this is a replay or if it is in the
8953 			 * expected (next) sequence.  In the case of a
8954 			 * replay, there are two replay conditions
8955 			 * that may occur.  The first is the normal
8956 			 * condition where a LOCK is done with a
8957 			 * NFS4_OK response and the stateid is
8958 			 * updated.  That case is handled below when
8959 			 * the stateid is identified as a REPLAY.  The
8960 			 * second is the case where an error is
8961 			 * returned, like NFS4ERR_DENIED, and the
8962 			 * sequence number is updated but the stateid
8963 			 * is not updated.  This second case is dealt
8964 			 * with here.  So it may seem odd that the
8965 			 * stateid is okay but the sequence id is a
8966 			 * replay but it is okay.
8967 			 */
8968 			switch (rfs4_check_lock_seqid(
8969 			    args->locker.locker4_u.lock_owner.lock_seqid,
8970 			    lsp, resop)) {
8971 			case NFS4_CHKSEQ_REPLAY:
8972 				if (resp->status != NFS4_OK) {
8973 					/*
8974 					 * Here is our replay and need
8975 					 * to verify that the last
8976 					 * response was an error.
8977 					 */
8978 					*cs->statusp = resp->status;
8979 					goto end;
8980 				}
8981 				/*
8982 				 * This is done since the sequence id
8983 				 * looked like a replay but it didn't
8984 				 * pass our check so a BAD_SEQID is
8985 				 * returned as a result.
8986 				 */
8987 				/*FALLTHROUGH*/
8988 			case NFS4_CHKSEQ_BAD:
8989 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8990 				goto end;
8991 			case NFS4_CHKSEQ_OKAY:
8992 				/* Everything looks okay move ahead */
8993 				break;
8994 			}
8995 			break;
8996 		case NFS4_CHECK_STATEID_OLD:
8997 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8998 			goto end;
8999 		case NFS4_CHECK_STATEID_BAD:
9000 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9001 			goto end;
9002 		case NFS4_CHECK_STATEID_EXPIRED:
9003 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9004 			goto end;
9005 		case NFS4_CHECK_STATEID_CLOSED:
9006 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9007 			goto end;
9008 		case NFS4_CHECK_STATEID_REPLAY:
9009 			switch (rfs4_check_lock_seqid(
9010 			    args->locker.locker4_u.lock_owner.lock_seqid,
9011 			    lsp, resop)) {
9012 			case NFS4_CHKSEQ_OKAY:
9013 				/*
9014 				 * This is a replayed stateid; if
9015 				 * seqid matches the next expected,
9016 				 * then client is using wrong seqid.
9017 				 */
9018 			case NFS4_CHKSEQ_BAD:
9019 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9020 				goto end;
9021 			case NFS4_CHKSEQ_REPLAY:
9022 				rfs4_update_lease(lsp->rls_locker->rl_client);
9023 				*cs->statusp = status = resp->status;
9024 				goto end;
9025 			}
9026 			break;
9027 		default:
9028 			ASSERT(FALSE);
9029 			break;
9030 		}
9031 
9032 		rfs4_update_lock_sequence(lsp);
9033 		rfs4_update_lease(lsp->rls_locker->rl_client);
9034 	}
9035 
9036 	/*
9037 	 * NFS4 only allows locking on regular files, so
9038 	 * verify type of object.
9039 	 */
9040 	if (cs->vp->v_type != VREG) {
9041 		if (cs->vp->v_type == VDIR)
9042 			status = NFS4ERR_ISDIR;
9043 		else
9044 			status = NFS4ERR_INVAL;
9045 		goto out;
9046 	}
9047 
9048 	cp = lsp->rls_state->rs_owner->ro_client;
9049 
9050 	if (rfs4_clnt_in_grace(cp) && !args->reclaim) {
9051 		status = NFS4ERR_GRACE;
9052 		goto out;
9053 	}
9054 
9055 	if (rfs4_clnt_in_grace(cp) && args->reclaim && !cp->rc_can_reclaim) {
9056 		status = NFS4ERR_NO_GRACE;
9057 		goto out;
9058 	}
9059 
9060 	if (!rfs4_clnt_in_grace(cp) && args->reclaim) {
9061 		status = NFS4ERR_NO_GRACE;
9062 		goto out;
9063 	}
9064 
9065 	if (lsp->rls_state->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE)
9066 		cs->deleg = TRUE;
9067 
9068 	status = rfs4_do_lock(lsp, args->locktype,
9069 	    args->offset, args->length, cs->cr, resop);
9070 
9071 out:
9072 	lsp->rls_skip_seqid_check = FALSE;
9073 
9074 	*cs->statusp = resp->status = status;
9075 
9076 	if (status == NFS4_OK) {
9077 		resp->LOCK4res_u.lock_stateid = lsp->rls_lockid.stateid;
9078 		lsp->rls_lock_completed = TRUE;
9079 	}
9080 	/*
9081 	 * Only update the "OPEN" response here if this was a new
9082 	 * lock_owner
9083 	 */
9084 	if (sp)
9085 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
9086 
9087 	rfs4_update_lock_resp(lsp, resop);
9088 
9089 end:
9090 	if (lsp) {
9091 		if (ls_sw_held)
9092 			rfs4_sw_exit(&lsp->rls_sw);
9093 		/*
9094 		 * If an sp obtained, then the lsp does not represent
9095 		 * a lock on the file struct.
9096 		 */
9097 		if (sp != NULL)
9098 			rfs4_lo_state_rele(lsp, FALSE);
9099 		else
9100 			rfs4_lo_state_rele(lsp, TRUE);
9101 	}
9102 	if (sp) {
9103 		rfs4_sw_exit(&sp->rs_owner->ro_sw);
9104 		rfs4_state_rele(sp);
9105 	}
9106 
9107 	DTRACE_NFSV4_2(op__lock__done, struct compound_state *, cs,
9108 	    LOCK4res *, resp);
9109 }
9110 
9111 /* free function for LOCK/LOCKT */
9112 static void
9113 lock_denied_free(nfs_resop4 *resop)
9114 {
9115 	LOCK4denied *dp = NULL;
9116 
9117 	switch (resop->resop) {
9118 	case OP_LOCK:
9119 		if (resop->nfs_resop4_u.oplock.status == NFS4ERR_DENIED)
9120 			dp = &resop->nfs_resop4_u.oplock.LOCK4res_u.denied;
9121 		break;
9122 	case OP_LOCKT:
9123 		if (resop->nfs_resop4_u.oplockt.status == NFS4ERR_DENIED)
9124 			dp = &resop->nfs_resop4_u.oplockt.denied;
9125 		break;
9126 	default:
9127 		break;
9128 	}
9129 
9130 	if (dp)
9131 		kmem_free(dp->owner.owner_val, dp->owner.owner_len);
9132 }
9133 
9134 /*ARGSUSED*/
9135 void
9136 rfs4_op_locku(nfs_argop4 *argop, nfs_resop4 *resop,
9137     struct svc_req *req, struct compound_state *cs)
9138 {
9139 	LOCKU4args *args = &argop->nfs_argop4_u.oplocku;
9140 	LOCKU4res *resp = &resop->nfs_resop4_u.oplocku;
9141 	nfsstat4 status;
9142 	stateid4 *stateid = &args->lock_stateid;
9143 	rfs4_lo_state_t *lsp;
9144 
9145 	DTRACE_NFSV4_2(op__locku__start, struct compound_state *, cs,
9146 	    LOCKU4args *, args);
9147 
9148 	if (cs->vp == NULL) {
9149 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9150 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9151 		    LOCKU4res *, resp);
9152 		return;
9153 	}
9154 
9155 	if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE)) != NFS4_OK) {
9156 		*cs->statusp = resp->status = status;
9157 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9158 		    LOCKU4res *, resp);
9159 		return;
9160 	}
9161 
9162 	/* Ensure specified filehandle matches */
9163 	if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9164 		rfs4_lo_state_rele(lsp, TRUE);
9165 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9166 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9167 		    LOCKU4res *, resp);
9168 		return;
9169 	}
9170 
9171 	/* hold off other access to lsp while we tinker */
9172 	rfs4_sw_enter(&lsp->rls_sw);
9173 
9174 	switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9175 	case NFS4_CHECK_STATEID_OKAY:
9176 		if (rfs4_check_lock_seqid(args->seqid, lsp, resop)
9177 		    != NFS4_CHKSEQ_OKAY) {
9178 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9179 			goto end;
9180 		}
9181 		break;
9182 	case NFS4_CHECK_STATEID_OLD:
9183 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9184 		goto end;
9185 	case NFS4_CHECK_STATEID_BAD:
9186 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9187 		goto end;
9188 	case NFS4_CHECK_STATEID_EXPIRED:
9189 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9190 		goto end;
9191 	case NFS4_CHECK_STATEID_CLOSED:
9192 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9193 		goto end;
9194 	case NFS4_CHECK_STATEID_REPLAY:
9195 		switch (rfs4_check_lock_seqid(args->seqid, lsp, resop)) {
9196 		case NFS4_CHKSEQ_OKAY:
9197 				/*
9198 				 * This is a replayed stateid; if
9199 				 * seqid matches the next expected,
9200 				 * then client is using wrong seqid.
9201 				 */
9202 		case NFS4_CHKSEQ_BAD:
9203 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9204 			goto end;
9205 		case NFS4_CHKSEQ_REPLAY:
9206 			rfs4_update_lease(lsp->rls_locker->rl_client);
9207 			*cs->statusp = status = resp->status;
9208 			goto end;
9209 		}
9210 		break;
9211 	default:
9212 		ASSERT(FALSE);
9213 		break;
9214 	}
9215 
9216 	rfs4_update_lock_sequence(lsp);
9217 	rfs4_update_lease(lsp->rls_locker->rl_client);
9218 
9219 	/*
9220 	 * NFS4 only allows locking on regular files, so
9221 	 * verify type of object.
9222 	 */
9223 	if (cs->vp->v_type != VREG) {
9224 		if (cs->vp->v_type == VDIR)
9225 			status = NFS4ERR_ISDIR;
9226 		else
9227 			status = NFS4ERR_INVAL;
9228 		goto out;
9229 	}
9230 
9231 	if (rfs4_clnt_in_grace(lsp->rls_state->rs_owner->ro_client)) {
9232 		status = NFS4ERR_GRACE;
9233 		goto out;
9234 	}
9235 
9236 	status = rfs4_do_lock(lsp, args->locktype,
9237 	    args->offset, args->length, cs->cr, resop);
9238 
9239 out:
9240 	*cs->statusp = resp->status = status;
9241 
9242 	if (status == NFS4_OK)
9243 		resp->lock_stateid = lsp->rls_lockid.stateid;
9244 
9245 	rfs4_update_lock_resp(lsp, resop);
9246 
9247 end:
9248 	rfs4_sw_exit(&lsp->rls_sw);
9249 	rfs4_lo_state_rele(lsp, TRUE);
9250 
9251 	DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9252 	    LOCKU4res *, resp);
9253 }
9254 
9255 /*
9256  * LOCKT is a best effort routine, the client can not be guaranteed that
9257  * the status return is still in effect by the time the reply is received.
9258  * They are numerous race conditions in this routine, but we are not required
9259  * and can not be accurate.
9260  */
9261 /*ARGSUSED*/
9262 void
9263 rfs4_op_lockt(nfs_argop4 *argop, nfs_resop4 *resop,
9264     struct svc_req *req, struct compound_state *cs)
9265 {
9266 	LOCKT4args *args = &argop->nfs_argop4_u.oplockt;
9267 	LOCKT4res *resp = &resop->nfs_resop4_u.oplockt;
9268 	rfs4_lockowner_t *lo;
9269 	rfs4_client_t *cp;
9270 	bool_t create = FALSE;
9271 	struct flock64 flk;
9272 	int error;
9273 	int flag = FREAD | FWRITE;
9274 	int ltype;
9275 	length4 posix_length;
9276 	sysid_t sysid;
9277 	pid_t pid;
9278 
9279 	DTRACE_NFSV4_2(op__lockt__start, struct compound_state *, cs,
9280 	    LOCKT4args *, args);
9281 
9282 	if (cs->vp == NULL) {
9283 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9284 		goto out;
9285 	}
9286 
9287 	/*
9288 	 * NFS4 only allows locking on regular files, so
9289 	 * verify type of object.
9290 	 */
9291 	if (cs->vp->v_type != VREG) {
9292 		if (cs->vp->v_type == VDIR)
9293 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
9294 		else
9295 			*cs->statusp = resp->status =  NFS4ERR_INVAL;
9296 		goto out;
9297 	}
9298 
9299 	/*
9300 	 * Check out the clientid to ensure the server knows about it
9301 	 * so that we correctly inform the client of a server reboot.
9302 	 */
9303 	if ((cp = rfs4_findclient_by_id(args->owner.clientid, FALSE))
9304 	    == NULL) {
9305 		*cs->statusp = resp->status =
9306 		    rfs4_check_clientid(&args->owner.clientid, 0);
9307 		goto out;
9308 	}
9309 	if (rfs4_lease_expired(cp)) {
9310 		rfs4_client_close(cp);
9311 		/*
9312 		 * Protocol doesn't allow returning NFS4ERR_STALE as
9313 		 * other operations do on this check so STALE_CLIENTID
9314 		 * is returned instead
9315 		 */
9316 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
9317 		goto out;
9318 	}
9319 
9320 	if (rfs4_clnt_in_grace(cp) && !(cp->rc_can_reclaim)) {
9321 		*cs->statusp = resp->status = NFS4ERR_GRACE;
9322 		rfs4_client_rele(cp);
9323 		goto out;
9324 	}
9325 	rfs4_client_rele(cp);
9326 
9327 	resp->status = NFS4_OK;
9328 
9329 	switch (args->locktype) {
9330 	case READ_LT:
9331 	case READW_LT:
9332 		ltype = F_RDLCK;
9333 		break;
9334 	case WRITE_LT:
9335 	case WRITEW_LT:
9336 		ltype = F_WRLCK;
9337 		break;
9338 	}
9339 
9340 	posix_length = args->length;
9341 	/* Check for zero length. To lock to end of file use all ones for V4 */
9342 	if (posix_length == 0) {
9343 		*cs->statusp = resp->status = NFS4ERR_INVAL;
9344 		goto out;
9345 	} else if (posix_length == (length4)(~0)) {
9346 		posix_length = 0;	/* Posix to end of file  */
9347 	}
9348 
9349 	/* Find or create a lockowner */
9350 	lo = rfs4_findlockowner(&args->owner, &create);
9351 
9352 	if (lo) {
9353 		pid = lo->rl_pid;
9354 		if ((resp->status =
9355 		    rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
9356 			goto err;
9357 	} else {
9358 		pid = 0;
9359 		sysid = lockt_sysid;
9360 	}
9361 retry:
9362 	flk.l_type = ltype;
9363 	flk.l_whence = 0;		/* SEEK_SET */
9364 	flk.l_start = args->offset;
9365 	flk.l_len = posix_length;
9366 	flk.l_sysid = sysid;
9367 	flk.l_pid = pid;
9368 	flag |= F_REMOTELOCK;
9369 
9370 	LOCK_PRINT(rfs4_debug, "rfs4_op_lockt", F_GETLK, &flk);
9371 
9372 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
9373 	if (flk.l_len < 0 || flk.l_start < 0) {
9374 		resp->status = NFS4ERR_INVAL;
9375 		goto err;
9376 	}
9377 	error = VOP_FRLOCK(cs->vp, F_GETLK, &flk, flag, (u_offset_t)0,
9378 	    NULL, cs->cr, NULL);
9379 
9380 	/*
9381 	 * N.B. We map error values to nfsv4 errors. This is differrent
9382 	 * than puterrno4 routine.
9383 	 */
9384 	switch (error) {
9385 	case 0:
9386 		if (flk.l_type == F_UNLCK)
9387 			resp->status = NFS4_OK;
9388 		else {
9389 			if (lock_denied(&resp->denied, &flk) == NFS4ERR_EXPIRED)
9390 				goto retry;
9391 			resp->status = NFS4ERR_DENIED;
9392 		}
9393 		break;
9394 	case EOVERFLOW:
9395 		resp->status = NFS4ERR_INVAL;
9396 		break;
9397 	case EINVAL:
9398 		resp->status = NFS4ERR_NOTSUPP;
9399 		break;
9400 	default:
9401 		cmn_err(CE_WARN, "rfs4_op_lockt: unexpected errno (%d)",
9402 		    error);
9403 		resp->status = NFS4ERR_SERVERFAULT;
9404 		break;
9405 	}
9406 
9407 err:
9408 	if (lo)
9409 		rfs4_lockowner_rele(lo);
9410 	*cs->statusp = resp->status;
9411 out:
9412 	DTRACE_NFSV4_2(op__lockt__done, struct compound_state *, cs,
9413 	    LOCKT4res *, resp);
9414 }
9415 
9416 int
9417 rfs4_share(rfs4_state_t *sp, uint32_t access, uint32_t deny)
9418 {
9419 	int err;
9420 	int cmd;
9421 	vnode_t *vp;
9422 	struct shrlock shr;
9423 	struct shr_locowner shr_loco;
9424 	int fflags = 0;
9425 
9426 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9427 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9428 
9429 	if (sp->rs_closed)
9430 		return (NFS4ERR_OLD_STATEID);
9431 
9432 	vp = sp->rs_finfo->rf_vp;
9433 	ASSERT(vp);
9434 
9435 	shr.s_access = shr.s_deny = 0;
9436 
9437 	if (access & OPEN4_SHARE_ACCESS_READ) {
9438 		fflags |= FREAD;
9439 		shr.s_access |= F_RDACC;
9440 	}
9441 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
9442 		fflags |= FWRITE;
9443 		shr.s_access |= F_WRACC;
9444 	}
9445 	ASSERT(shr.s_access);
9446 
9447 	if (deny & OPEN4_SHARE_DENY_READ)
9448 		shr.s_deny |= F_RDDNY;
9449 	if (deny & OPEN4_SHARE_DENY_WRITE)
9450 		shr.s_deny |= F_WRDNY;
9451 
9452 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9453 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9454 	shr_loco.sl_pid = shr.s_pid;
9455 	shr_loco.sl_id = shr.s_sysid;
9456 	shr.s_owner = (caddr_t)&shr_loco;
9457 	shr.s_own_len = sizeof (shr_loco);
9458 
9459 	cmd = nbl_need_check(vp) ? F_SHARE_NBMAND : F_SHARE;
9460 
9461 	err = VOP_SHRLOCK(vp, cmd, &shr, fflags, CRED(), NULL);
9462 	if (err != 0) {
9463 		if (err == EAGAIN)
9464 			err = NFS4ERR_SHARE_DENIED;
9465 		else
9466 			err = puterrno4(err);
9467 		return (err);
9468 	}
9469 
9470 	sp->rs_share_access |= access;
9471 	sp->rs_share_deny |= deny;
9472 
9473 	return (0);
9474 }
9475 
9476 int
9477 rfs4_unshare(rfs4_state_t *sp)
9478 {
9479 	int err;
9480 	struct shrlock shr;
9481 	struct shr_locowner shr_loco;
9482 
9483 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9484 
9485 	if (sp->rs_closed || sp->rs_share_access == 0)
9486 		return (0);
9487 
9488 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9489 	ASSERT(sp->rs_finfo->rf_vp);
9490 
9491 	shr.s_access = shr.s_deny = 0;
9492 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9493 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9494 	shr_loco.sl_pid = shr.s_pid;
9495 	shr_loco.sl_id = shr.s_sysid;
9496 	shr.s_owner = (caddr_t)&shr_loco;
9497 	shr.s_own_len = sizeof (shr_loco);
9498 
9499 	err = VOP_SHRLOCK(sp->rs_finfo->rf_vp, F_UNSHARE, &shr, 0, CRED(),
9500 	    NULL);
9501 	if (err != 0) {
9502 		err = puterrno4(err);
9503 		return (err);
9504 	}
9505 
9506 	sp->rs_share_access = 0;
9507 	sp->rs_share_deny = 0;
9508 
9509 	return (0);
9510 
9511 }
9512 
9513 static int
9514 rdma_setup_read_data4(READ4args *args, READ4res *rok)
9515 {
9516 	struct clist	*wcl;
9517 	count4		count = rok->data_len;
9518 	int		wlist_len;
9519 
9520 	wcl = args->wlist;
9521 	if (rdma_setup_read_chunks(wcl, count, &wlist_len) == FALSE) {
9522 		return (FALSE);
9523 	}
9524 	wcl = args->wlist;
9525 	rok->wlist_len = wlist_len;
9526 	rok->wlist = wcl;
9527 	return (TRUE);
9528 }
9529 
9530 /* tunable to disable server referrals */
9531 int rfs4_no_referrals = 0;
9532 
9533 /*
9534  * Find an NFS record in reparse point data.
9535  * Returns 0 for success and <0 or an errno value on failure.
9536  */
9537 int
9538 vn_find_nfs_record(vnode_t *vp, nvlist_t **nvlp, char **svcp, char **datap)
9539 {
9540 	int err;
9541 	char *stype, *val;
9542 	nvlist_t *nvl;
9543 	nvpair_t *curr;
9544 
9545 	if ((nvl = reparse_init()) == NULL)
9546 		return (-1);
9547 
9548 	if ((err = reparse_vnode_parse(vp, nvl)) != 0) {
9549 		reparse_free(nvl);
9550 		return (err);
9551 	}
9552 
9553 	curr = NULL;
9554 	while ((curr = nvlist_next_nvpair(nvl, curr)) != NULL) {
9555 		if ((stype = nvpair_name(curr)) == NULL) {
9556 			reparse_free(nvl);
9557 			return (-2);
9558 		}
9559 		if (strncasecmp(stype, "NFS", 3) == 0)
9560 			break;
9561 	}
9562 
9563 	if ((curr == NULL) ||
9564 	    (nvpair_value_string(curr, &val))) {
9565 		reparse_free(nvl);
9566 		return (-3);
9567 	}
9568 	*nvlp = nvl;
9569 	*svcp = stype;
9570 	*datap = val;
9571 	return (0);
9572 }
9573 
9574 int
9575 vn_is_nfs_reparse(vnode_t *vp, cred_t *cr)
9576 {
9577 	nvlist_t *nvl;
9578 	char *s, *d;
9579 
9580 	if (rfs4_no_referrals != 0)
9581 		return (B_FALSE);
9582 
9583 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9584 		return (B_FALSE);
9585 
9586 	if (vn_find_nfs_record(vp, &nvl, &s, &d) != 0)
9587 		return (B_FALSE);
9588 
9589 	reparse_free(nvl);
9590 
9591 	return (B_TRUE);
9592 }
9593 
9594 /*
9595  * There is a user-level copy of this routine in ref_subr.c.
9596  * Changes should be kept in sync.
9597  */
9598 static int
9599 nfs4_create_components(char *path, component4 *comp4)
9600 {
9601 	int slen, plen, ncomp;
9602 	char *ori_path, *nxtc, buf[MAXNAMELEN];
9603 
9604 	if (path == NULL)
9605 		return (0);
9606 
9607 	plen = strlen(path) + 1;	/* include the terminator */
9608 	ori_path = path;
9609 	ncomp = 0;
9610 
9611 	/* count number of components in the path */
9612 	for (nxtc = path; nxtc < ori_path + plen; nxtc++) {
9613 		if (*nxtc == '/' || *nxtc == '\0' || *nxtc == '\n') {
9614 			if ((slen = nxtc - path) == 0) {
9615 				path = nxtc + 1;
9616 				continue;
9617 			}
9618 
9619 			if (comp4 != NULL) {
9620 				bcopy(path, buf, slen);
9621 				buf[slen] = '\0';
9622 				(void) str_to_utf8(buf, &comp4[ncomp]);
9623 			}
9624 
9625 			ncomp++;	/* 1 valid component */
9626 			path = nxtc + 1;
9627 		}
9628 		if (*nxtc == '\0' || *nxtc == '\n')
9629 			break;
9630 	}
9631 
9632 	return (ncomp);
9633 }
9634 
9635 /*
9636  * There is a user-level copy of this routine in ref_subr.c.
9637  * Changes should be kept in sync.
9638  */
9639 static int
9640 make_pathname4(char *path, pathname4 *pathname)
9641 {
9642 	int ncomp;
9643 	component4 *comp4;
9644 
9645 	if (pathname == NULL)
9646 		return (0);
9647 
9648 	if (path == NULL) {
9649 		pathname->pathname4_val = NULL;
9650 		pathname->pathname4_len = 0;
9651 		return (0);
9652 	}
9653 
9654 	/* count number of components to alloc buffer */
9655 	if ((ncomp = nfs4_create_components(path, NULL)) == 0) {
9656 		pathname->pathname4_val = NULL;
9657 		pathname->pathname4_len = 0;
9658 		return (0);
9659 	}
9660 	comp4 = kmem_zalloc(ncomp * sizeof (component4), KM_SLEEP);
9661 
9662 	/* copy components into allocated buffer */
9663 	ncomp = nfs4_create_components(path, comp4);
9664 
9665 	pathname->pathname4_val = comp4;
9666 	pathname->pathname4_len = ncomp;
9667 
9668 	return (ncomp);
9669 }
9670 
9671 #define	xdr_fs_locations4 xdr_fattr4_fs_locations
9672 
9673 fs_locations4 *
9674 fetch_referral(vnode_t *vp, cred_t *cr)
9675 {
9676 	nvlist_t *nvl;
9677 	char *stype, *sdata;
9678 	fs_locations4 *result;
9679 	char buf[1024];
9680 	size_t bufsize;
9681 	XDR xdr;
9682 	int err;
9683 
9684 	/*
9685 	 * Check attrs to ensure it's a reparse point
9686 	 */
9687 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9688 		return (NULL);
9689 
9690 	/*
9691 	 * Look for an NFS record and get the type and data
9692 	 */
9693 	if (vn_find_nfs_record(vp, &nvl, &stype, &sdata) != 0)
9694 		return (NULL);
9695 
9696 	/*
9697 	 * With the type and data, upcall to get the referral
9698 	 */
9699 	bufsize = sizeof (buf);
9700 	bzero(buf, sizeof (buf));
9701 	err = reparse_kderef((const char *)stype, (const char *)sdata,
9702 	    buf, &bufsize);
9703 	reparse_free(nvl);
9704 
9705 	DTRACE_PROBE4(nfs4serv__func__referral__upcall,
9706 	    char *, stype, char *, sdata, char *, buf, int, err);
9707 	if (err) {
9708 		cmn_err(CE_NOTE,
9709 		    "reparsed daemon not running: unable to get referral (%d)",
9710 		    err);
9711 		return (NULL);
9712 	}
9713 
9714 	/*
9715 	 * We get an XDR'ed record back from the kderef call
9716 	 */
9717 	xdrmem_create(&xdr, buf, bufsize, XDR_DECODE);
9718 	result = kmem_alloc(sizeof (fs_locations4), KM_SLEEP);
9719 	err = xdr_fs_locations4(&xdr, result);
9720 	XDR_DESTROY(&xdr);
9721 	if (err != TRUE) {
9722 		DTRACE_PROBE1(nfs4serv__func__referral__upcall__xdrfail,
9723 		    int, err);
9724 		return (NULL);
9725 	}
9726 
9727 	/*
9728 	 * Look at path to recover fs_root, ignoring the leading '/'
9729 	 */
9730 	(void) make_pathname4(vp->v_path, &result->fs_root);
9731 
9732 	return (result);
9733 }
9734 
9735 char *
9736 build_symlink(vnode_t *vp, cred_t *cr, size_t *strsz)
9737 {
9738 	fs_locations4 *fsl;
9739 	fs_location4 *fs;
9740 	char *server, *path, *symbuf;
9741 	static char *prefix = "/net/";
9742 	int i, size, npaths;
9743 	uint_t len;
9744 
9745 	/* Get the referral */
9746 	if ((fsl = fetch_referral(vp, cr)) == NULL)
9747 		return (NULL);
9748 
9749 	/* Deal with only the first location and first server */
9750 	fs = &fsl->locations_val[0];
9751 	server = utf8_to_str(&fs->server_val[0], &len, NULL);
9752 	if (server == NULL) {
9753 		rfs4_free_fs_locations4(fsl);
9754 		kmem_free(fsl, sizeof (fs_locations4));
9755 		return (NULL);
9756 	}
9757 
9758 	/* Figure out size for "/net/" + host + /path/path/path + NULL */
9759 	size = strlen(prefix) + len;
9760 	for (i = 0; i < fs->rootpath.pathname4_len; i++)
9761 		size += fs->rootpath.pathname4_val[i].utf8string_len + 1;
9762 
9763 	/* Allocate the symlink buffer and fill it */
9764 	symbuf = kmem_zalloc(size, KM_SLEEP);
9765 	(void) strcat(symbuf, prefix);
9766 	(void) strcat(symbuf, server);
9767 	kmem_free(server, len);
9768 
9769 	npaths = 0;
9770 	for (i = 0; i < fs->rootpath.pathname4_len; i++) {
9771 		path = utf8_to_str(&fs->rootpath.pathname4_val[i], &len, NULL);
9772 		if (path == NULL)
9773 			continue;
9774 		(void) strcat(symbuf, "/");
9775 		(void) strcat(symbuf, path);
9776 		npaths++;
9777 		kmem_free(path, len);
9778 	}
9779 
9780 	rfs4_free_fs_locations4(fsl);
9781 	kmem_free(fsl, sizeof (fs_locations4));
9782 
9783 	if (strsz != NULL)
9784 		*strsz = size;
9785 	return (symbuf);
9786 }
9787 
9788 /*
9789  * Check to see if we have a downrev Solaris client, so that we
9790  * can send it a symlink instead of a referral.
9791  */
9792 int
9793 client_is_downrev(struct svc_req *req)
9794 {
9795 	struct sockaddr *ca;
9796 	rfs4_clntip_t *ci;
9797 	bool_t create = FALSE;
9798 	int is_downrev;
9799 
9800 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
9801 	ASSERT(ca);
9802 	ci = rfs4_find_clntip(ca, &create);
9803 	if (ci == NULL)
9804 		return (0);
9805 	is_downrev = ci->ri_no_referrals;
9806 	rfs4_dbe_rele(ci->ri_dbe);
9807 	return (is_downrev);
9808 }
9809