xref: /titanic_51/usr/src/uts/common/fs/nfs/nfs4_srv.c (revision b60ae21d2303cc238394b46cddb93a2dbcdb2e07)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2016 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2012 by Delphix. All rights reserved.
26  */
27 
28 /*
29  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
30  *	All Rights Reserved
31  */
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <sys/systm.h>
36 #include <sys/cred.h>
37 #include <sys/buf.h>
38 #include <sys/vfs.h>
39 #include <sys/vfs_opreg.h>
40 #include <sys/vnode.h>
41 #include <sys/uio.h>
42 #include <sys/errno.h>
43 #include <sys/sysmacros.h>
44 #include <sys/statvfs.h>
45 #include <sys/kmem.h>
46 #include <sys/dirent.h>
47 #include <sys/cmn_err.h>
48 #include <sys/debug.h>
49 #include <sys/systeminfo.h>
50 #include <sys/flock.h>
51 #include <sys/pathname.h>
52 #include <sys/nbmlock.h>
53 #include <sys/share.h>
54 #include <sys/atomic.h>
55 #include <sys/policy.h>
56 #include <sys/fem.h>
57 #include <sys/sdt.h>
58 #include <sys/ddi.h>
59 #include <sys/zone.h>
60 
61 #include <fs/fs_reparse.h>
62 
63 #include <rpc/types.h>
64 #include <rpc/auth.h>
65 #include <rpc/rpcsec_gss.h>
66 #include <rpc/svc.h>
67 
68 #include <nfs/nfs.h>
69 #include <nfs/export.h>
70 #include <nfs/nfs_cmd.h>
71 #include <nfs/lm.h>
72 #include <nfs/nfs4.h>
73 
74 #include <sys/strsubr.h>
75 #include <sys/strsun.h>
76 
77 #include <inet/common.h>
78 #include <inet/ip.h>
79 #include <inet/ip6.h>
80 
81 #include <sys/tsol/label.h>
82 #include <sys/tsol/tndb.h>
83 
84 #define	RFS4_MAXLOCK_TRIES 4	/* Try to get the lock this many times */
85 static int rfs4_maxlock_tries = RFS4_MAXLOCK_TRIES;
86 #define	RFS4_LOCK_DELAY 10	/* Milliseconds */
87 static clock_t  rfs4_lock_delay = RFS4_LOCK_DELAY;
88 extern struct svc_ops rdma_svc_ops;
89 extern int nfs_loaned_buffers;
90 /* End of Tunables */
91 
92 static int rdma_setup_read_data4(READ4args *, READ4res *);
93 
94 /*
95  * Used to bump the stateid4.seqid value and show changes in the stateid
96  */
97 #define	next_stateid(sp) (++(sp)->bits.chgseq)
98 
99 /*
100  * RFS4_MINLEN_ENTRY4: XDR-encoded size of smallest possible dirent.
101  *	This is used to return NFS4ERR_TOOSMALL when clients specify
102  *	maxcount that isn't large enough to hold the smallest possible
103  *	XDR encoded dirent.
104  *
105  *	    sizeof cookie (8 bytes) +
106  *	    sizeof name_len (4 bytes) +
107  *	    sizeof smallest (padded) name (4 bytes) +
108  *	    sizeof bitmap4_len (12 bytes) +   NOTE: we always encode len=2 bm4
109  *	    sizeof attrlist4_len (4 bytes) +
110  *	    sizeof next boolean (4 bytes)
111  *
112  * RFS4_MINLEN_RDDIR4: XDR-encoded size of READDIR op reply containing
113  * the smallest possible entry4 (assumes no attrs requested).
114  *	sizeof nfsstat4 (4 bytes) +
115  *	sizeof verifier4 (8 bytes) +
116  *	sizeof entry4list bool (4 bytes) +
117  *	sizeof entry4 	(36 bytes) +
118  *	sizeof eof bool  (4 bytes)
119  *
120  * RFS4_MINLEN_RDDIR_BUF: minimum length of buffer server will provide to
121  *	VOP_READDIR.  Its value is the size of the maximum possible dirent
122  *	for solaris.  The DIRENT64_RECLEN macro returns	the size of dirent
123  *	required for a given name length.  MAXNAMELEN is the maximum
124  *	filename length allowed in Solaris.  The first two DIRENT64_RECLEN()
125  *	macros are to allow for . and .. entries -- just a minor tweak to try
126  *	and guarantee that buffer we give to VOP_READDIR will be large enough
127  *	to hold ., .., and the largest possible solaris dirent64.
128  */
129 #define	RFS4_MINLEN_ENTRY4 36
130 #define	RFS4_MINLEN_RDDIR4 (4 + NFS4_VERIFIER_SIZE + 4 + RFS4_MINLEN_ENTRY4 + 4)
131 #define	RFS4_MINLEN_RDDIR_BUF \
132 	(DIRENT64_RECLEN(1) + DIRENT64_RECLEN(2) + DIRENT64_RECLEN(MAXNAMELEN))
133 
134 /*
135  * It would be better to pad to 4 bytes since that's what XDR would do,
136  * but the dirents UFS gives us are already padded to 8, so just take
137  * what we're given.  Dircount is only a hint anyway.  Currently the
138  * solaris kernel is ASCII only, so there's no point in calling the
139  * UTF8 functions.
140  *
141  * dirent64: named padded to provide 8 byte struct alignment
142  *	d_ino(8) + d_off(8) + d_reclen(2) + d_name(namelen + null(1) + pad)
143  *
144  * cookie: uint64_t   +  utf8namelen: uint_t  +   utf8name padded to 8 bytes
145  *
146  */
147 #define	DIRENT64_TO_DIRCOUNT(dp) \
148 	(3 * BYTES_PER_XDR_UNIT + DIRENT64_NAMELEN((dp)->d_reclen))
149 
150 time_t rfs4_start_time;			/* Initialized in rfs4_srvrinit */
151 
152 static sysid_t lockt_sysid;		/* dummy sysid for all LOCKT calls */
153 
154 u_longlong_t	nfs4_srv_caller_id;
155 uint_t		nfs4_srv_vkey = 0;
156 
157 verifier4	Write4verf;
158 verifier4	Readdir4verf;
159 
160 void	rfs4_init_compound_state(struct compound_state *);
161 
162 static void	nullfree(caddr_t);
163 static void	rfs4_op_inval(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
164 			struct compound_state *);
165 static void	rfs4_op_access(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
166 			struct compound_state *);
167 static void	rfs4_op_close(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
168 			struct compound_state *);
169 static void	rfs4_op_commit(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
170 			struct compound_state *);
171 static void	rfs4_op_create(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
172 			struct compound_state *);
173 static void	rfs4_op_create_free(nfs_resop4 *resop);
174 static void	rfs4_op_delegreturn(nfs_argop4 *, nfs_resop4 *,
175 			struct svc_req *, struct compound_state *);
176 static void	rfs4_op_delegpurge(nfs_argop4 *, nfs_resop4 *,
177 			struct svc_req *, struct compound_state *);
178 static void	rfs4_op_getattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
179 			struct compound_state *);
180 static void	rfs4_op_getattr_free(nfs_resop4 *);
181 static void	rfs4_op_getfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
182 			struct compound_state *);
183 static void	rfs4_op_getfh_free(nfs_resop4 *);
184 static void	rfs4_op_illegal(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
185 			struct compound_state *);
186 static void	rfs4_op_link(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
187 			struct compound_state *);
188 static void	rfs4_op_lock(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
189 			struct compound_state *);
190 static void	lock_denied_free(nfs_resop4 *);
191 static void	rfs4_op_locku(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
192 			struct compound_state *);
193 static void	rfs4_op_lockt(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
194 			struct compound_state *);
195 static void	rfs4_op_lookup(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
196 			struct compound_state *);
197 static void	rfs4_op_lookupp(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
198 			struct compound_state *);
199 static void	rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop,
200 				struct svc_req *req, struct compound_state *cs);
201 static void	rfs4_op_nverify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
202 			struct compound_state *);
203 static void	rfs4_op_open(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
204 			struct compound_state *);
205 static void	rfs4_op_open_confirm(nfs_argop4 *, nfs_resop4 *,
206 			struct svc_req *, struct compound_state *);
207 static void	rfs4_op_open_downgrade(nfs_argop4 *, nfs_resop4 *,
208 			struct svc_req *, struct compound_state *);
209 static void	rfs4_op_putfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
210 			struct compound_state *);
211 static void	rfs4_op_putpubfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
212 			struct compound_state *);
213 static void	rfs4_op_putrootfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
214 			struct compound_state *);
215 static void	rfs4_op_read(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
216 			struct compound_state *);
217 static void	rfs4_op_read_free(nfs_resop4 *);
218 static void	rfs4_op_readdir_free(nfs_resop4 *resop);
219 static void	rfs4_op_readlink(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
220 			struct compound_state *);
221 static void	rfs4_op_readlink_free(nfs_resop4 *);
222 static void	rfs4_op_release_lockowner(nfs_argop4 *, nfs_resop4 *,
223 			struct svc_req *, struct compound_state *);
224 static void	rfs4_op_remove(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
225 			struct compound_state *);
226 static void	rfs4_op_rename(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
227 			struct compound_state *);
228 static void	rfs4_op_renew(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
229 			struct compound_state *);
230 static void	rfs4_op_restorefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
231 			struct compound_state *);
232 static void	rfs4_op_savefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
233 			struct compound_state *);
234 static void	rfs4_op_setattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
235 			struct compound_state *);
236 static void	rfs4_op_verify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
237 			struct compound_state *);
238 static void	rfs4_op_write(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
239 			struct compound_state *);
240 static void	rfs4_op_setclientid(nfs_argop4 *, nfs_resop4 *,
241 			struct svc_req *, struct compound_state *);
242 static void	rfs4_op_setclientid_confirm(nfs_argop4 *, nfs_resop4 *,
243 			struct svc_req *req, struct compound_state *);
244 static void	rfs4_op_secinfo(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
245 			struct compound_state *);
246 static void	rfs4_op_secinfo_free(nfs_resop4 *);
247 
248 static nfsstat4 check_open_access(uint32_t,
249 				struct compound_state *, struct svc_req *);
250 nfsstat4 rfs4_client_sysid(rfs4_client_t *, sysid_t *);
251 void rfs4_ss_clid(rfs4_client_t *);
252 
253 /*
254  * translation table for attrs
255  */
256 struct nfs4_ntov_table {
257 	union nfs4_attr_u *na;
258 	uint8_t amap[NFS4_MAXNUM_ATTRS];
259 	int attrcnt;
260 	bool_t vfsstat;
261 };
262 
263 static void	nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp);
264 static void	nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
265 				    struct nfs4_svgetit_arg *sargp);
266 
267 static nfsstat4	do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp,
268 		    struct compound_state *cs, struct nfs4_svgetit_arg *sargp,
269 		    struct nfs4_ntov_table *ntovp, nfs4_attr_cmd_t cmd);
270 
271 fem_t		*deleg_rdops;
272 fem_t		*deleg_wrops;
273 
274 rfs4_servinst_t *rfs4_cur_servinst = NULL;	/* current server instance */
275 kmutex_t	rfs4_servinst_lock;	/* protects linked list */
276 int		rfs4_seen_first_compound;	/* set first time we see one */
277 
278 /*
279  * NFS4 op dispatch table
280  */
281 
282 struct rfsv4disp {
283 	void	(*dis_proc)();		/* proc to call */
284 	void	(*dis_resfree)();	/* frees space allocated by proc */
285 	int	dis_flags;		/* RPC_IDEMPOTENT, etc... */
286 };
287 
288 static struct rfsv4disp rfsv4disptab[] = {
289 	/*
290 	 * NFS VERSION 4
291 	 */
292 
293 	/* RFS_NULL = 0 */
294 	{rfs4_op_illegal, nullfree, 0},
295 
296 	/* UNUSED = 1 */
297 	{rfs4_op_illegal, nullfree, 0},
298 
299 	/* UNUSED = 2 */
300 	{rfs4_op_illegal, nullfree, 0},
301 
302 	/* OP_ACCESS = 3 */
303 	{rfs4_op_access, nullfree, RPC_IDEMPOTENT},
304 
305 	/* OP_CLOSE = 4 */
306 	{rfs4_op_close, nullfree, 0},
307 
308 	/* OP_COMMIT = 5 */
309 	{rfs4_op_commit, nullfree, RPC_IDEMPOTENT},
310 
311 	/* OP_CREATE = 6 */
312 	{rfs4_op_create, nullfree, 0},
313 
314 	/* OP_DELEGPURGE = 7 */
315 	{rfs4_op_delegpurge, nullfree, 0},
316 
317 	/* OP_DELEGRETURN = 8 */
318 	{rfs4_op_delegreturn, nullfree, 0},
319 
320 	/* OP_GETATTR = 9 */
321 	{rfs4_op_getattr, rfs4_op_getattr_free, RPC_IDEMPOTENT},
322 
323 	/* OP_GETFH = 10 */
324 	{rfs4_op_getfh, rfs4_op_getfh_free, RPC_ALL},
325 
326 	/* OP_LINK = 11 */
327 	{rfs4_op_link, nullfree, 0},
328 
329 	/* OP_LOCK = 12 */
330 	{rfs4_op_lock, lock_denied_free, 0},
331 
332 	/* OP_LOCKT = 13 */
333 	{rfs4_op_lockt, lock_denied_free, 0},
334 
335 	/* OP_LOCKU = 14 */
336 	{rfs4_op_locku, nullfree, 0},
337 
338 	/* OP_LOOKUP = 15 */
339 	{rfs4_op_lookup, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
340 
341 	/* OP_LOOKUPP = 16 */
342 	{rfs4_op_lookupp, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
343 
344 	/* OP_NVERIFY = 17 */
345 	{rfs4_op_nverify, nullfree, RPC_IDEMPOTENT},
346 
347 	/* OP_OPEN = 18 */
348 	{rfs4_op_open, rfs4_free_reply, 0},
349 
350 	/* OP_OPENATTR = 19 */
351 	{rfs4_op_openattr, nullfree, 0},
352 
353 	/* OP_OPEN_CONFIRM = 20 */
354 	{rfs4_op_open_confirm, nullfree, 0},
355 
356 	/* OP_OPEN_DOWNGRADE = 21 */
357 	{rfs4_op_open_downgrade, nullfree, 0},
358 
359 	/* OP_OPEN_PUTFH = 22 */
360 	{rfs4_op_putfh, nullfree, RPC_ALL},
361 
362 	/* OP_PUTPUBFH = 23 */
363 	{rfs4_op_putpubfh, nullfree, RPC_ALL},
364 
365 	/* OP_PUTROOTFH = 24 */
366 	{rfs4_op_putrootfh, nullfree, RPC_ALL},
367 
368 	/* OP_READ = 25 */
369 	{rfs4_op_read, rfs4_op_read_free, RPC_IDEMPOTENT},
370 
371 	/* OP_READDIR = 26 */
372 	{rfs4_op_readdir, rfs4_op_readdir_free, RPC_IDEMPOTENT},
373 
374 	/* OP_READLINK = 27 */
375 	{rfs4_op_readlink, rfs4_op_readlink_free, RPC_IDEMPOTENT},
376 
377 	/* OP_REMOVE = 28 */
378 	{rfs4_op_remove, nullfree, 0},
379 
380 	/* OP_RENAME = 29 */
381 	{rfs4_op_rename, nullfree, 0},
382 
383 	/* OP_RENEW = 30 */
384 	{rfs4_op_renew, nullfree, 0},
385 
386 	/* OP_RESTOREFH = 31 */
387 	{rfs4_op_restorefh, nullfree, RPC_ALL},
388 
389 	/* OP_SAVEFH = 32 */
390 	{rfs4_op_savefh, nullfree, RPC_ALL},
391 
392 	/* OP_SECINFO = 33 */
393 	{rfs4_op_secinfo, rfs4_op_secinfo_free, 0},
394 
395 	/* OP_SETATTR = 34 */
396 	{rfs4_op_setattr, nullfree, 0},
397 
398 	/* OP_SETCLIENTID = 35 */
399 	{rfs4_op_setclientid, nullfree, 0},
400 
401 	/* OP_SETCLIENTID_CONFIRM = 36 */
402 	{rfs4_op_setclientid_confirm, nullfree, 0},
403 
404 	/* OP_VERIFY = 37 */
405 	{rfs4_op_verify, nullfree, RPC_IDEMPOTENT},
406 
407 	/* OP_WRITE = 38 */
408 	{rfs4_op_write, nullfree, 0},
409 
410 	/* OP_RELEASE_LOCKOWNER = 39 */
411 	{rfs4_op_release_lockowner, nullfree, 0},
412 };
413 
414 static uint_t rfsv4disp_cnt = sizeof (rfsv4disptab) / sizeof (rfsv4disptab[0]);
415 
416 #define	OP_ILLEGAL_IDX (rfsv4disp_cnt)
417 
418 #ifdef DEBUG
419 
420 int		rfs4_fillone_debug = 0;
421 int		rfs4_no_stub_access = 1;
422 int		rfs4_rddir_debug = 0;
423 
424 static char    *rfs4_op_string[] = {
425 	"rfs4_op_null",
426 	"rfs4_op_1 unused",
427 	"rfs4_op_2 unused",
428 	"rfs4_op_access",
429 	"rfs4_op_close",
430 	"rfs4_op_commit",
431 	"rfs4_op_create",
432 	"rfs4_op_delegpurge",
433 	"rfs4_op_delegreturn",
434 	"rfs4_op_getattr",
435 	"rfs4_op_getfh",
436 	"rfs4_op_link",
437 	"rfs4_op_lock",
438 	"rfs4_op_lockt",
439 	"rfs4_op_locku",
440 	"rfs4_op_lookup",
441 	"rfs4_op_lookupp",
442 	"rfs4_op_nverify",
443 	"rfs4_op_open",
444 	"rfs4_op_openattr",
445 	"rfs4_op_open_confirm",
446 	"rfs4_op_open_downgrade",
447 	"rfs4_op_putfh",
448 	"rfs4_op_putpubfh",
449 	"rfs4_op_putrootfh",
450 	"rfs4_op_read",
451 	"rfs4_op_readdir",
452 	"rfs4_op_readlink",
453 	"rfs4_op_remove",
454 	"rfs4_op_rename",
455 	"rfs4_op_renew",
456 	"rfs4_op_restorefh",
457 	"rfs4_op_savefh",
458 	"rfs4_op_secinfo",
459 	"rfs4_op_setattr",
460 	"rfs4_op_setclientid",
461 	"rfs4_op_setclient_confirm",
462 	"rfs4_op_verify",
463 	"rfs4_op_write",
464 	"rfs4_op_release_lockowner",
465 	"rfs4_op_illegal"
466 };
467 #endif
468 
469 void	rfs4_ss_chkclid(rfs4_client_t *);
470 
471 extern size_t   strlcpy(char *dst, const char *src, size_t dstsize);
472 
473 extern void	rfs4_free_fs_locations4(fs_locations4 *);
474 
475 #ifdef	nextdp
476 #undef nextdp
477 #endif
478 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
479 
480 static const fs_operation_def_t nfs4_rd_deleg_tmpl[] = {
481 	VOPNAME_OPEN,		{ .femop_open = deleg_rd_open },
482 	VOPNAME_WRITE,		{ .femop_write = deleg_rd_write },
483 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_rd_setattr },
484 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_rd_rwlock },
485 	VOPNAME_SPACE,		{ .femop_space = deleg_rd_space },
486 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_rd_setsecattr },
487 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_rd_vnevent },
488 	NULL,			NULL
489 };
490 static const fs_operation_def_t nfs4_wr_deleg_tmpl[] = {
491 	VOPNAME_OPEN,		{ .femop_open = deleg_wr_open },
492 	VOPNAME_READ,		{ .femop_read = deleg_wr_read },
493 	VOPNAME_WRITE,		{ .femop_write = deleg_wr_write },
494 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_wr_setattr },
495 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_wr_rwlock },
496 	VOPNAME_SPACE,		{ .femop_space = deleg_wr_space },
497 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_wr_setsecattr },
498 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_wr_vnevent },
499 	NULL,			NULL
500 };
501 
502 int
503 rfs4_srvrinit(void)
504 {
505 	timespec32_t verf;
506 	int error;
507 	extern void rfs4_attr_init();
508 	extern krwlock_t rfs4_deleg_policy_lock;
509 
510 	/*
511 	 * The following algorithm attempts to find a unique verifier
512 	 * to be used as the write verifier returned from the server
513 	 * to the client.  It is important that this verifier change
514 	 * whenever the server reboots.  Of secondary importance, it
515 	 * is important for the verifier to be unique between two
516 	 * different servers.
517 	 *
518 	 * Thus, an attempt is made to use the system hostid and the
519 	 * current time in seconds when the nfssrv kernel module is
520 	 * loaded.  It is assumed that an NFS server will not be able
521 	 * to boot and then to reboot in less than a second.  If the
522 	 * hostid has not been set, then the current high resolution
523 	 * time is used.  This will ensure different verifiers each
524 	 * time the server reboots and minimize the chances that two
525 	 * different servers will have the same verifier.
526 	 * XXX - this is broken on LP64 kernels.
527 	 */
528 	verf.tv_sec = (time_t)zone_get_hostid(NULL);
529 	if (verf.tv_sec != 0) {
530 		verf.tv_nsec = gethrestime_sec();
531 	} else {
532 		timespec_t tverf;
533 
534 		gethrestime(&tverf);
535 		verf.tv_sec = (time_t)tverf.tv_sec;
536 		verf.tv_nsec = tverf.tv_nsec;
537 	}
538 
539 	Write4verf = *(uint64_t *)&verf;
540 
541 	rfs4_attr_init();
542 	mutex_init(&rfs4_deleg_lock, NULL, MUTEX_DEFAULT, NULL);
543 
544 	/* Used to manage create/destroy of server state */
545 	mutex_init(&rfs4_state_lock, NULL, MUTEX_DEFAULT, NULL);
546 
547 	/* Used to manage access to server instance linked list */
548 	mutex_init(&rfs4_servinst_lock, NULL, MUTEX_DEFAULT, NULL);
549 
550 	/* Used to manage access to rfs4_deleg_policy */
551 	rw_init(&rfs4_deleg_policy_lock, NULL, RW_DEFAULT, NULL);
552 
553 	error = fem_create("deleg_rdops", nfs4_rd_deleg_tmpl, &deleg_rdops);
554 	if (error != 0) {
555 		rfs4_disable_delegation();
556 	} else {
557 		error = fem_create("deleg_wrops", nfs4_wr_deleg_tmpl,
558 		    &deleg_wrops);
559 		if (error != 0) {
560 			rfs4_disable_delegation();
561 			fem_free(deleg_rdops);
562 		}
563 	}
564 
565 	nfs4_srv_caller_id = fs_new_caller_id();
566 
567 	lockt_sysid = lm_alloc_sysidt();
568 
569 	vsd_create(&nfs4_srv_vkey, NULL);
570 
571 	return (0);
572 }
573 
574 void
575 rfs4_srvrfini(void)
576 {
577 	extern krwlock_t rfs4_deleg_policy_lock;
578 
579 	if (lockt_sysid != LM_NOSYSID) {
580 		lm_free_sysidt(lockt_sysid);
581 		lockt_sysid = LM_NOSYSID;
582 	}
583 
584 	mutex_destroy(&rfs4_deleg_lock);
585 	mutex_destroy(&rfs4_state_lock);
586 	rw_destroy(&rfs4_deleg_policy_lock);
587 
588 	fem_free(deleg_rdops);
589 	fem_free(deleg_wrops);
590 }
591 
592 void
593 rfs4_init_compound_state(struct compound_state *cs)
594 {
595 	bzero(cs, sizeof (*cs));
596 	cs->cont = TRUE;
597 	cs->access = CS_ACCESS_DENIED;
598 	cs->deleg = FALSE;
599 	cs->mandlock = FALSE;
600 	cs->fh.nfs_fh4_val = cs->fhbuf;
601 }
602 
603 void
604 rfs4_grace_start(rfs4_servinst_t *sip)
605 {
606 	rw_enter(&sip->rwlock, RW_WRITER);
607 	sip->start_time = (time_t)TICK_TO_SEC(ddi_get_lbolt());
608 	sip->grace_period = rfs4_grace_period;
609 	rw_exit(&sip->rwlock);
610 }
611 
612 /*
613  * returns true if the instance's grace period has never been started
614  */
615 int
616 rfs4_servinst_grace_new(rfs4_servinst_t *sip)
617 {
618 	time_t start_time;
619 
620 	rw_enter(&sip->rwlock, RW_READER);
621 	start_time = sip->start_time;
622 	rw_exit(&sip->rwlock);
623 
624 	return (start_time == 0);
625 }
626 
627 /*
628  * Indicates if server instance is within the
629  * grace period.
630  */
631 int
632 rfs4_servinst_in_grace(rfs4_servinst_t *sip)
633 {
634 	time_t grace_expiry;
635 
636 	rw_enter(&sip->rwlock, RW_READER);
637 	grace_expiry = sip->start_time + sip->grace_period;
638 	rw_exit(&sip->rwlock);
639 
640 	return (((time_t)TICK_TO_SEC(ddi_get_lbolt())) < grace_expiry);
641 }
642 
643 int
644 rfs4_clnt_in_grace(rfs4_client_t *cp)
645 {
646 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
647 
648 	return (rfs4_servinst_in_grace(cp->rc_server_instance));
649 }
650 
651 /*
652  * reset all currently active grace periods
653  */
654 void
655 rfs4_grace_reset_all(void)
656 {
657 	rfs4_servinst_t *sip;
658 
659 	mutex_enter(&rfs4_servinst_lock);
660 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
661 		if (rfs4_servinst_in_grace(sip))
662 			rfs4_grace_start(sip);
663 	mutex_exit(&rfs4_servinst_lock);
664 }
665 
666 /*
667  * start any new instances' grace periods
668  */
669 void
670 rfs4_grace_start_new(void)
671 {
672 	rfs4_servinst_t *sip;
673 
674 	mutex_enter(&rfs4_servinst_lock);
675 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
676 		if (rfs4_servinst_grace_new(sip))
677 			rfs4_grace_start(sip);
678 	mutex_exit(&rfs4_servinst_lock);
679 }
680 
681 static rfs4_dss_path_t *
682 rfs4_dss_newpath(rfs4_servinst_t *sip, char *path, unsigned index)
683 {
684 	size_t len;
685 	rfs4_dss_path_t *dss_path;
686 
687 	dss_path = kmem_alloc(sizeof (rfs4_dss_path_t), KM_SLEEP);
688 
689 	/*
690 	 * Take a copy of the string, since the original may be overwritten.
691 	 * Sadly, no strdup() in the kernel.
692 	 */
693 	/* allow for NUL */
694 	len = strlen(path) + 1;
695 	dss_path->path = kmem_alloc(len, KM_SLEEP);
696 	(void) strlcpy(dss_path->path, path, len);
697 
698 	/* associate with servinst */
699 	dss_path->sip = sip;
700 	dss_path->index = index;
701 
702 	/*
703 	 * Add to list of served paths.
704 	 * No locking required, as we're only ever called at startup.
705 	 */
706 	if (rfs4_dss_pathlist == NULL) {
707 		/* this is the first dss_path_t */
708 
709 		/* needed for insque/remque */
710 		dss_path->next = dss_path->prev = dss_path;
711 
712 		rfs4_dss_pathlist = dss_path;
713 	} else {
714 		insque(dss_path, rfs4_dss_pathlist);
715 	}
716 
717 	return (dss_path);
718 }
719 
720 /*
721  * Create a new server instance, and make it the currently active instance.
722  * Note that starting the grace period too early will reduce the clients'
723  * recovery window.
724  */
725 void
726 rfs4_servinst_create(int start_grace, int dss_npaths, char **dss_paths)
727 {
728 	unsigned i;
729 	rfs4_servinst_t *sip;
730 	rfs4_oldstate_t *oldstate;
731 
732 	sip = kmem_alloc(sizeof (rfs4_servinst_t), KM_SLEEP);
733 	rw_init(&sip->rwlock, NULL, RW_DEFAULT, NULL);
734 
735 	sip->start_time = (time_t)0;
736 	sip->grace_period = (time_t)0;
737 	sip->next = NULL;
738 	sip->prev = NULL;
739 
740 	rw_init(&sip->oldstate_lock, NULL, RW_DEFAULT, NULL);
741 	/*
742 	 * This initial dummy entry is required to setup for insque/remque.
743 	 * It must be skipped over whenever the list is traversed.
744 	 */
745 	oldstate = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
746 	/* insque/remque require initial list entry to be self-terminated */
747 	oldstate->next = oldstate;
748 	oldstate->prev = oldstate;
749 	sip->oldstate = oldstate;
750 
751 
752 	sip->dss_npaths = dss_npaths;
753 	sip->dss_paths = kmem_alloc(dss_npaths *
754 	    sizeof (rfs4_dss_path_t *), KM_SLEEP);
755 
756 	for (i = 0; i < dss_npaths; i++) {
757 		sip->dss_paths[i] = rfs4_dss_newpath(sip, dss_paths[i], i);
758 	}
759 
760 	mutex_enter(&rfs4_servinst_lock);
761 	if (rfs4_cur_servinst != NULL) {
762 		/* add to linked list */
763 		sip->prev = rfs4_cur_servinst;
764 		rfs4_cur_servinst->next = sip;
765 	}
766 	if (start_grace)
767 		rfs4_grace_start(sip);
768 	/* make the new instance "current" */
769 	rfs4_cur_servinst = sip;
770 
771 	mutex_exit(&rfs4_servinst_lock);
772 }
773 
774 /*
775  * In future, we might add a rfs4_servinst_destroy(sip) but, for now, destroy
776  * all instances directly.
777  */
778 void
779 rfs4_servinst_destroy_all(void)
780 {
781 	rfs4_servinst_t *sip, *prev, *current;
782 #ifdef DEBUG
783 	int n = 0;
784 #endif
785 
786 	mutex_enter(&rfs4_servinst_lock);
787 	ASSERT(rfs4_cur_servinst != NULL);
788 	current = rfs4_cur_servinst;
789 	rfs4_cur_servinst = NULL;
790 	for (sip = current; sip != NULL; sip = prev) {
791 		prev = sip->prev;
792 		rw_destroy(&sip->rwlock);
793 		if (sip->oldstate)
794 			kmem_free(sip->oldstate, sizeof (rfs4_oldstate_t));
795 		if (sip->dss_paths)
796 			kmem_free(sip->dss_paths,
797 			    sip->dss_npaths * sizeof (rfs4_dss_path_t *));
798 		kmem_free(sip, sizeof (rfs4_servinst_t));
799 #ifdef DEBUG
800 		n++;
801 #endif
802 	}
803 	mutex_exit(&rfs4_servinst_lock);
804 }
805 
806 /*
807  * Assign the current server instance to a client_t.
808  * Should be called with cp->rc_dbe held.
809  */
810 void
811 rfs4_servinst_assign(rfs4_client_t *cp, rfs4_servinst_t *sip)
812 {
813 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
814 
815 	/*
816 	 * The lock ensures that if the current instance is in the process
817 	 * of changing, we will see the new one.
818 	 */
819 	mutex_enter(&rfs4_servinst_lock);
820 	cp->rc_server_instance = sip;
821 	mutex_exit(&rfs4_servinst_lock);
822 }
823 
824 rfs4_servinst_t *
825 rfs4_servinst(rfs4_client_t *cp)
826 {
827 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
828 
829 	return (cp->rc_server_instance);
830 }
831 
832 /* ARGSUSED */
833 static void
834 nullfree(caddr_t resop)
835 {
836 }
837 
838 /*
839  * This is a fall-through for invalid or not implemented (yet) ops
840  */
841 /* ARGSUSED */
842 static void
843 rfs4_op_inval(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
844     struct compound_state *cs)
845 {
846 	*cs->statusp = *((nfsstat4 *)&(resop)->nfs_resop4_u) = NFS4ERR_INVAL;
847 }
848 
849 /*
850  * Check if the security flavor, nfsnum, is in the flavor_list.
851  */
852 bool_t
853 in_flavor_list(int nfsnum, int *flavor_list, int count)
854 {
855 	int i;
856 
857 	for (i = 0; i < count; i++) {
858 		if (nfsnum == flavor_list[i])
859 			return (TRUE);
860 	}
861 	return (FALSE);
862 }
863 
864 /*
865  * Used by rfs4_op_secinfo to get the security information from the
866  * export structure associated with the component.
867  */
868 /* ARGSUSED */
869 static nfsstat4
870 do_rfs4_op_secinfo(struct compound_state *cs, char *nm, SECINFO4res *resp)
871 {
872 	int error, different_export = 0;
873 	vnode_t *dvp, *vp;
874 	struct exportinfo *exi = NULL;
875 	struct exportinfo *oexi = NULL;
876 	fid_t fid;
877 	uint_t count, i;
878 	secinfo4 *resok_val;
879 	struct secinfo *secp;
880 	seconfig_t *si;
881 	bool_t did_traverse = FALSE;
882 	int dotdot, walk;
883 
884 	dvp = cs->vp;
885 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
886 
887 	/*
888 	 * If dotdotting, then need to check whether it's above the
889 	 * root of a filesystem, or above an export point.
890 	 */
891 	if (dotdot) {
892 
893 		/*
894 		 * If dotdotting at the root of a filesystem, then
895 		 * need to traverse back to the mounted-on filesystem
896 		 * and do the dotdot lookup there.
897 		 */
898 		if (cs->vp->v_flag & VROOT) {
899 
900 			/*
901 			 * If at the system root, then can
902 			 * go up no further.
903 			 */
904 			if (VN_CMP(dvp, rootdir))
905 				return (puterrno4(ENOENT));
906 
907 			/*
908 			 * Traverse back to the mounted-on filesystem
909 			 */
910 			dvp = untraverse(cs->vp);
911 
912 			/*
913 			 * Set the different_export flag so we remember
914 			 * to pick up a new exportinfo entry for
915 			 * this new filesystem.
916 			 */
917 			different_export = 1;
918 		} else {
919 
920 			/*
921 			 * If dotdotting above an export point then set
922 			 * the different_export to get new export info.
923 			 */
924 			different_export = nfs_exported(cs->exi, cs->vp);
925 		}
926 	}
927 
928 	/*
929 	 * Get the vnode for the component "nm".
930 	 */
931 	error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cs->cr,
932 	    NULL, NULL, NULL);
933 	if (error)
934 		return (puterrno4(error));
935 
936 	/*
937 	 * If the vnode is in a pseudo filesystem, or if the security flavor
938 	 * used in the request is valid but not an explicitly shared flavor,
939 	 * or the access bit indicates that this is a limited access,
940 	 * check whether this vnode is visible.
941 	 */
942 	if (!different_export &&
943 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
944 	    cs->access & CS_ACCESS_LIMITED)) {
945 		if (! nfs_visible(cs->exi, vp, &different_export)) {
946 			VN_RELE(vp);
947 			return (puterrno4(ENOENT));
948 		}
949 	}
950 
951 	/*
952 	 * If it's a mountpoint, then traverse it.
953 	 */
954 	if (vn_ismntpt(vp)) {
955 		if ((error = traverse(&vp)) != 0) {
956 			VN_RELE(vp);
957 			return (puterrno4(error));
958 		}
959 		/* remember that we had to traverse mountpoint */
960 		did_traverse = TRUE;
961 		different_export = 1;
962 	} else if (vp->v_vfsp != dvp->v_vfsp) {
963 		/*
964 		 * If vp isn't a mountpoint and the vfs ptrs aren't the same,
965 		 * then vp is probably an LOFS object.  We don't need the
966 		 * realvp, we just need to know that we might have crossed
967 		 * a server fs boundary and need to call checkexport.
968 		 * (LOFS lookup hides server fs mountpoints, and actually calls
969 		 * traverse)
970 		 */
971 		different_export = 1;
972 	}
973 
974 	/*
975 	 * Get the export information for it.
976 	 */
977 	if (different_export) {
978 
979 		bzero(&fid, sizeof (fid));
980 		fid.fid_len = MAXFIDSZ;
981 		error = vop_fid_pseudo(vp, &fid);
982 		if (error) {
983 			VN_RELE(vp);
984 			return (puterrno4(error));
985 		}
986 
987 		if (dotdot)
988 			oexi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
989 		else
990 			oexi = checkexport(&vp->v_vfsp->vfs_fsid, &fid, vp);
991 
992 		if (oexi == NULL) {
993 			if (did_traverse == TRUE) {
994 				/*
995 				 * If this vnode is a mounted-on vnode,
996 				 * but the mounted-on file system is not
997 				 * exported, send back the secinfo for
998 				 * the exported node that the mounted-on
999 				 * vnode lives in.
1000 				 */
1001 				exi = cs->exi;
1002 			} else {
1003 				VN_RELE(vp);
1004 				return (puterrno4(EACCES));
1005 			}
1006 		} else {
1007 			exi = oexi;
1008 		}
1009 	} else {
1010 		exi = cs->exi;
1011 	}
1012 	ASSERT(exi != NULL);
1013 
1014 
1015 	/*
1016 	 * Create the secinfo result based on the security information
1017 	 * from the exportinfo structure (exi).
1018 	 *
1019 	 * Return all flavors for a pseudo node.
1020 	 * For a real export node, return the flavor that the client
1021 	 * has access with.
1022 	 */
1023 	rw_enter(&exported_lock, RW_READER);
1024 	if (PSEUDO(exi)) {
1025 		count = exi->exi_export.ex_seccnt; /* total sec count */
1026 		resok_val = kmem_alloc(count * sizeof (secinfo4), KM_SLEEP);
1027 		secp = exi->exi_export.ex_secinfo;
1028 
1029 		for (i = 0; i < count; i++) {
1030 			si = &secp[i].s_secinfo;
1031 			resok_val[i].flavor = si->sc_rpcnum;
1032 			if (resok_val[i].flavor == RPCSEC_GSS) {
1033 				rpcsec_gss_info *info;
1034 
1035 				info = &resok_val[i].flavor_info;
1036 				info->qop = si->sc_qop;
1037 				info->service = (rpc_gss_svc_t)si->sc_service;
1038 
1039 				/* get oid opaque data */
1040 				info->oid.sec_oid4_len =
1041 				    si->sc_gss_mech_type->length;
1042 				info->oid.sec_oid4_val = kmem_alloc(
1043 				    si->sc_gss_mech_type->length, KM_SLEEP);
1044 				bcopy(
1045 				    si->sc_gss_mech_type->elements,
1046 				    info->oid.sec_oid4_val,
1047 				    info->oid.sec_oid4_len);
1048 			}
1049 		}
1050 		resp->SECINFO4resok_len = count;
1051 		resp->SECINFO4resok_val = resok_val;
1052 	} else {
1053 		int ret_cnt = 0, k = 0;
1054 		int *flavor_list;
1055 
1056 		count = exi->exi_export.ex_seccnt; /* total sec count */
1057 		secp = exi->exi_export.ex_secinfo;
1058 
1059 		flavor_list = kmem_alloc(count * sizeof (int), KM_SLEEP);
1060 		/* find out which flavors to return */
1061 		for (i = 0; i < count; i ++) {
1062 			int access, flavor, perm;
1063 
1064 			flavor = secp[i].s_secinfo.sc_nfsnum;
1065 			perm = secp[i].s_flags;
1066 
1067 			access = nfsauth4_secinfo_access(exi, cs->req,
1068 			    flavor, perm, cs->basecr);
1069 
1070 			if (! (access & NFSAUTH_DENIED) &&
1071 			    ! (access & NFSAUTH_WRONGSEC)) {
1072 				flavor_list[ret_cnt] = flavor;
1073 				ret_cnt++;
1074 			}
1075 		}
1076 
1077 		/* Create the returning SECINFO value */
1078 		resok_val = kmem_alloc(ret_cnt * sizeof (secinfo4), KM_SLEEP);
1079 
1080 		for (i = 0; i < count; i++) {
1081 			/*
1082 			 * If the flavor is in the flavor list,
1083 			 * fill in resok_val.
1084 			 */
1085 			si = &secp[i].s_secinfo;
1086 			if (in_flavor_list(si->sc_nfsnum,
1087 			    flavor_list, ret_cnt)) {
1088 				resok_val[k].flavor = si->sc_rpcnum;
1089 				if (resok_val[k].flavor == RPCSEC_GSS) {
1090 					rpcsec_gss_info *info;
1091 
1092 					info = &resok_val[k].flavor_info;
1093 					info->qop = si->sc_qop;
1094 					info->service = (rpc_gss_svc_t)
1095 					    si->sc_service;
1096 
1097 					/* get oid opaque data */
1098 					info->oid.sec_oid4_len =
1099 					    si->sc_gss_mech_type->length;
1100 					info->oid.sec_oid4_val = kmem_alloc(
1101 					    si->sc_gss_mech_type->length,
1102 					    KM_SLEEP);
1103 					bcopy(si->sc_gss_mech_type->elements,
1104 					    info->oid.sec_oid4_val,
1105 					    info->oid.sec_oid4_len);
1106 				}
1107 				k++;
1108 			}
1109 			if (k >= ret_cnt)
1110 				break;
1111 		}
1112 		resp->SECINFO4resok_len = ret_cnt;
1113 		resp->SECINFO4resok_val = resok_val;
1114 		kmem_free(flavor_list, count * sizeof (int));
1115 	}
1116 	rw_exit(&exported_lock);
1117 	if (oexi)
1118 		exi_rele(oexi);
1119 	VN_RELE(vp);
1120 	return (NFS4_OK);
1121 }
1122 
1123 /*
1124  * SECINFO (Operation 33): Obtain required security information on
1125  * the component name in the format of (security-mechanism-oid, qop, service)
1126  * triplets.
1127  */
1128 /* ARGSUSED */
1129 static void
1130 rfs4_op_secinfo(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1131     struct compound_state *cs)
1132 {
1133 	SECINFO4args *args = &argop->nfs_argop4_u.opsecinfo;
1134 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1135 	utf8string *utfnm = &args->name;
1136 	uint_t len;
1137 	char *nm;
1138 	struct sockaddr *ca;
1139 	char *name = NULL;
1140 	nfsstat4 status = NFS4_OK;
1141 
1142 	DTRACE_NFSV4_2(op__secinfo__start, struct compound_state *, cs,
1143 	    SECINFO4args *, args);
1144 
1145 	/*
1146 	 * Current file handle (cfh) should have been set before getting
1147 	 * into this function. If not, return error.
1148 	 */
1149 	if (cs->vp == NULL) {
1150 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1151 		goto out;
1152 	}
1153 
1154 	if (cs->vp->v_type != VDIR) {
1155 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1156 		goto out;
1157 	}
1158 
1159 	/*
1160 	 * Verify the component name. If failed, error out, but
1161 	 * do not error out if the component name is a "..".
1162 	 * SECINFO will return its parents secinfo data for SECINFO "..".
1163 	 */
1164 	status = utf8_dir_verify(utfnm);
1165 	if (status != NFS4_OK) {
1166 		if (utfnm->utf8string_len != 2 ||
1167 		    utfnm->utf8string_val[0] != '.' ||
1168 		    utfnm->utf8string_val[1] != '.') {
1169 			*cs->statusp = resp->status = status;
1170 			goto out;
1171 		}
1172 	}
1173 
1174 	nm = utf8_to_str(utfnm, &len, NULL);
1175 	if (nm == NULL) {
1176 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1177 		goto out;
1178 	}
1179 
1180 	if (len > MAXNAMELEN) {
1181 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1182 		kmem_free(nm, len);
1183 		goto out;
1184 	}
1185 
1186 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1187 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1188 	    MAXPATHLEN  + 1);
1189 
1190 	if (name == NULL) {
1191 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1192 		kmem_free(nm, len);
1193 		goto out;
1194 	}
1195 
1196 
1197 	*cs->statusp = resp->status = do_rfs4_op_secinfo(cs, name, resp);
1198 
1199 	if (name != nm)
1200 		kmem_free(name, MAXPATHLEN + 1);
1201 	kmem_free(nm, len);
1202 
1203 out:
1204 	DTRACE_NFSV4_2(op__secinfo__done, struct compound_state *, cs,
1205 	    SECINFO4res *, resp);
1206 }
1207 
1208 /*
1209  * Free SECINFO result.
1210  */
1211 /* ARGSUSED */
1212 static void
1213 rfs4_op_secinfo_free(nfs_resop4 *resop)
1214 {
1215 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1216 	int count, i;
1217 	secinfo4 *resok_val;
1218 
1219 	/* If this is not an Ok result, nothing to free. */
1220 	if (resp->status != NFS4_OK) {
1221 		return;
1222 	}
1223 
1224 	count = resp->SECINFO4resok_len;
1225 	resok_val = resp->SECINFO4resok_val;
1226 
1227 	for (i = 0; i < count; i++) {
1228 		if (resok_val[i].flavor == RPCSEC_GSS) {
1229 			rpcsec_gss_info *info;
1230 
1231 			info = &resok_val[i].flavor_info;
1232 			kmem_free(info->oid.sec_oid4_val,
1233 			    info->oid.sec_oid4_len);
1234 		}
1235 	}
1236 	kmem_free(resok_val, count * sizeof (secinfo4));
1237 	resp->SECINFO4resok_len = 0;
1238 	resp->SECINFO4resok_val = NULL;
1239 }
1240 
1241 /* ARGSUSED */
1242 static void
1243 rfs4_op_access(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1244     struct compound_state *cs)
1245 {
1246 	ACCESS4args *args = &argop->nfs_argop4_u.opaccess;
1247 	ACCESS4res *resp = &resop->nfs_resop4_u.opaccess;
1248 	int error;
1249 	vnode_t *vp;
1250 	struct vattr va;
1251 	int checkwriteperm;
1252 	cred_t *cr = cs->cr;
1253 	bslabel_t *clabel, *slabel;
1254 	ts_label_t *tslabel;
1255 	boolean_t admin_low_client;
1256 
1257 	DTRACE_NFSV4_2(op__access__start, struct compound_state *, cs,
1258 	    ACCESS4args *, args);
1259 
1260 #if 0	/* XXX allow access even if !cs->access. Eventually only pseudo fs */
1261 	if (cs->access == CS_ACCESS_DENIED) {
1262 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1263 		goto out;
1264 	}
1265 #endif
1266 	if (cs->vp == NULL) {
1267 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1268 		goto out;
1269 	}
1270 
1271 	ASSERT(cr != NULL);
1272 
1273 	vp = cs->vp;
1274 
1275 	/*
1276 	 * If the file system is exported read only, it is not appropriate
1277 	 * to check write permissions for regular files and directories.
1278 	 * Special files are interpreted by the client, so the underlying
1279 	 * permissions are sent back to the client for interpretation.
1280 	 */
1281 	if (rdonly4(req, cs) &&
1282 	    (vp->v_type == VREG || vp->v_type == VDIR))
1283 		checkwriteperm = 0;
1284 	else
1285 		checkwriteperm = 1;
1286 
1287 	/*
1288 	 * XXX
1289 	 * We need the mode so that we can correctly determine access
1290 	 * permissions relative to a mandatory lock file.  Access to
1291 	 * mandatory lock files is denied on the server, so it might
1292 	 * as well be reflected to the server during the open.
1293 	 */
1294 	va.va_mask = AT_MODE;
1295 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1296 	if (error) {
1297 		*cs->statusp = resp->status = puterrno4(error);
1298 		goto out;
1299 	}
1300 	resp->access = 0;
1301 	resp->supported = 0;
1302 
1303 	if (is_system_labeled()) {
1304 		ASSERT(req->rq_label != NULL);
1305 		clabel = req->rq_label;
1306 		DTRACE_PROBE2(tx__rfs4__log__info__opaccess__clabel, char *,
1307 		    "got client label from request(1)",
1308 		    struct svc_req *, req);
1309 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
1310 			if ((tslabel = nfs_getflabel(vp, cs->exi)) == NULL) {
1311 				*cs->statusp = resp->status = puterrno4(EACCES);
1312 				goto out;
1313 			}
1314 			slabel = label2bslabel(tslabel);
1315 			DTRACE_PROBE3(tx__rfs4__log__info__opaccess__slabel,
1316 			    char *, "got server label(1) for vp(2)",
1317 			    bslabel_t *, slabel, vnode_t *, vp);
1318 
1319 			admin_low_client = B_FALSE;
1320 		} else
1321 			admin_low_client = B_TRUE;
1322 	}
1323 
1324 	if (args->access & ACCESS4_READ) {
1325 		error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
1326 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1327 		    (!is_system_labeled() || admin_low_client ||
1328 		    bldominates(clabel, slabel)))
1329 			resp->access |= ACCESS4_READ;
1330 		resp->supported |= ACCESS4_READ;
1331 	}
1332 	if ((args->access & ACCESS4_LOOKUP) && vp->v_type == VDIR) {
1333 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1334 		if (!error && (!is_system_labeled() || admin_low_client ||
1335 		    bldominates(clabel, slabel)))
1336 			resp->access |= ACCESS4_LOOKUP;
1337 		resp->supported |= ACCESS4_LOOKUP;
1338 	}
1339 	if (checkwriteperm &&
1340 	    (args->access & (ACCESS4_MODIFY|ACCESS4_EXTEND))) {
1341 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1342 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1343 		    (!is_system_labeled() || admin_low_client ||
1344 		    blequal(clabel, slabel)))
1345 			resp->access |=
1346 			    (args->access & (ACCESS4_MODIFY | ACCESS4_EXTEND));
1347 		resp->supported |=
1348 		    resp->access & (ACCESS4_MODIFY | ACCESS4_EXTEND);
1349 	}
1350 
1351 	if (checkwriteperm &&
1352 	    (args->access & ACCESS4_DELETE) && vp->v_type == VDIR) {
1353 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1354 		if (!error && (!is_system_labeled() || admin_low_client ||
1355 		    blequal(clabel, slabel)))
1356 			resp->access |= ACCESS4_DELETE;
1357 		resp->supported |= ACCESS4_DELETE;
1358 	}
1359 	if (args->access & ACCESS4_EXECUTE && vp->v_type != VDIR) {
1360 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1361 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1362 		    (!is_system_labeled() || admin_low_client ||
1363 		    bldominates(clabel, slabel)))
1364 			resp->access |= ACCESS4_EXECUTE;
1365 		resp->supported |= ACCESS4_EXECUTE;
1366 	}
1367 
1368 	if (is_system_labeled() && !admin_low_client)
1369 		label_rele(tslabel);
1370 
1371 	*cs->statusp = resp->status = NFS4_OK;
1372 out:
1373 	DTRACE_NFSV4_2(op__access__done, struct compound_state *, cs,
1374 	    ACCESS4res *, resp);
1375 }
1376 
1377 /* ARGSUSED */
1378 static void
1379 rfs4_op_commit(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1380     struct compound_state *cs)
1381 {
1382 	COMMIT4args *args = &argop->nfs_argop4_u.opcommit;
1383 	COMMIT4res *resp = &resop->nfs_resop4_u.opcommit;
1384 	int error;
1385 	vnode_t *vp = cs->vp;
1386 	cred_t *cr = cs->cr;
1387 	vattr_t va;
1388 
1389 	DTRACE_NFSV4_2(op__commit__start, struct compound_state *, cs,
1390 	    COMMIT4args *, args);
1391 
1392 	if (vp == NULL) {
1393 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1394 		goto out;
1395 	}
1396 	if (cs->access == CS_ACCESS_DENIED) {
1397 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1398 		goto out;
1399 	}
1400 
1401 	if (args->offset + args->count < args->offset) {
1402 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1403 		goto out;
1404 	}
1405 
1406 	va.va_mask = AT_UID;
1407 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1408 
1409 	/*
1410 	 * If we can't get the attributes, then we can't do the
1411 	 * right access checking.  So, we'll fail the request.
1412 	 */
1413 	if (error) {
1414 		*cs->statusp = resp->status = puterrno4(error);
1415 		goto out;
1416 	}
1417 	if (rdonly4(req, cs)) {
1418 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1419 		goto out;
1420 	}
1421 
1422 	if (vp->v_type != VREG) {
1423 		if (vp->v_type == VDIR)
1424 			resp->status = NFS4ERR_ISDIR;
1425 		else
1426 			resp->status = NFS4ERR_INVAL;
1427 		*cs->statusp = resp->status;
1428 		goto out;
1429 	}
1430 
1431 	if (crgetuid(cr) != va.va_uid &&
1432 	    (error = VOP_ACCESS(vp, VWRITE, 0, cs->cr, NULL))) {
1433 		*cs->statusp = resp->status = puterrno4(error);
1434 		goto out;
1435 	}
1436 
1437 	error = VOP_FSYNC(vp, FSYNC, cr, NULL);
1438 
1439 	if (error) {
1440 		*cs->statusp = resp->status = puterrno4(error);
1441 		goto out;
1442 	}
1443 
1444 	*cs->statusp = resp->status = NFS4_OK;
1445 	resp->writeverf = Write4verf;
1446 out:
1447 	DTRACE_NFSV4_2(op__commit__done, struct compound_state *, cs,
1448 	    COMMIT4res *, resp);
1449 }
1450 
1451 /*
1452  * rfs4_op_mknod is called from rfs4_op_create after all initial verification
1453  * was completed. It does the nfsv4 create for special files.
1454  */
1455 /* ARGSUSED */
1456 static vnode_t *
1457 do_rfs4_op_mknod(CREATE4args *args, CREATE4res *resp, struct svc_req *req,
1458     struct compound_state *cs, vattr_t *vap, char *nm)
1459 {
1460 	int error;
1461 	cred_t *cr = cs->cr;
1462 	vnode_t *dvp = cs->vp;
1463 	vnode_t *vp = NULL;
1464 	int mode;
1465 	enum vcexcl excl;
1466 
1467 	switch (args->type) {
1468 	case NF4CHR:
1469 	case NF4BLK:
1470 		if (secpolicy_sys_devices(cr) != 0) {
1471 			*cs->statusp = resp->status = NFS4ERR_PERM;
1472 			return (NULL);
1473 		}
1474 		if (args->type == NF4CHR)
1475 			vap->va_type = VCHR;
1476 		else
1477 			vap->va_type = VBLK;
1478 		vap->va_rdev = makedevice(args->ftype4_u.devdata.specdata1,
1479 		    args->ftype4_u.devdata.specdata2);
1480 		vap->va_mask |= AT_RDEV;
1481 		break;
1482 	case NF4SOCK:
1483 		vap->va_type = VSOCK;
1484 		break;
1485 	case NF4FIFO:
1486 		vap->va_type = VFIFO;
1487 		break;
1488 	default:
1489 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1490 		return (NULL);
1491 	}
1492 
1493 	/*
1494 	 * Must specify the mode.
1495 	 */
1496 	if (!(vap->va_mask & AT_MODE)) {
1497 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1498 		return (NULL);
1499 	}
1500 
1501 	excl = EXCL;
1502 
1503 	mode = 0;
1504 
1505 	error = VOP_CREATE(dvp, nm, vap, excl, mode, &vp, cr, 0, NULL, NULL);
1506 	if (error) {
1507 		*cs->statusp = resp->status = puterrno4(error);
1508 		return (NULL);
1509 	}
1510 	return (vp);
1511 }
1512 
1513 /*
1514  * nfsv4 create is used to create non-regular files. For regular files,
1515  * use nfsv4 open.
1516  */
1517 /* ARGSUSED */
1518 static void
1519 rfs4_op_create(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1520     struct compound_state *cs)
1521 {
1522 	CREATE4args *args = &argop->nfs_argop4_u.opcreate;
1523 	CREATE4res *resp = &resop->nfs_resop4_u.opcreate;
1524 	int error;
1525 	struct vattr bva, iva, iva2, ava, *vap;
1526 	cred_t *cr = cs->cr;
1527 	vnode_t *dvp = cs->vp;
1528 	vnode_t *vp = NULL;
1529 	vnode_t *realvp;
1530 	char *nm, *lnm;
1531 	uint_t len, llen;
1532 	int syncval = 0;
1533 	struct nfs4_svgetit_arg sarg;
1534 	struct nfs4_ntov_table ntov;
1535 	struct statvfs64 sb;
1536 	nfsstat4 status;
1537 	struct sockaddr *ca;
1538 	char *name = NULL;
1539 	char *lname = NULL;
1540 
1541 	DTRACE_NFSV4_2(op__create__start, struct compound_state *, cs,
1542 	    CREATE4args *, args);
1543 
1544 	resp->attrset = 0;
1545 
1546 	if (dvp == NULL) {
1547 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1548 		goto out;
1549 	}
1550 
1551 	/*
1552 	 * If there is an unshared filesystem mounted on this vnode,
1553 	 * do not allow to create an object in this directory.
1554 	 */
1555 	if (vn_ismntpt(dvp)) {
1556 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1557 		goto out;
1558 	}
1559 
1560 	/* Verify that type is correct */
1561 	switch (args->type) {
1562 	case NF4LNK:
1563 	case NF4BLK:
1564 	case NF4CHR:
1565 	case NF4SOCK:
1566 	case NF4FIFO:
1567 	case NF4DIR:
1568 		break;
1569 	default:
1570 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1571 		goto out;
1572 	};
1573 
1574 	if (cs->access == CS_ACCESS_DENIED) {
1575 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1576 		goto out;
1577 	}
1578 	if (dvp->v_type != VDIR) {
1579 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1580 		goto out;
1581 	}
1582 	status = utf8_dir_verify(&args->objname);
1583 	if (status != NFS4_OK) {
1584 		*cs->statusp = resp->status = status;
1585 		goto out;
1586 	}
1587 
1588 	if (rdonly4(req, cs)) {
1589 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1590 		goto out;
1591 	}
1592 
1593 	/*
1594 	 * Name of newly created object
1595 	 */
1596 	nm = utf8_to_fn(&args->objname, &len, NULL);
1597 	if (nm == NULL) {
1598 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1599 		goto out;
1600 	}
1601 
1602 	if (len > MAXNAMELEN) {
1603 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1604 		kmem_free(nm, len);
1605 		goto out;
1606 	}
1607 
1608 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1609 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1610 	    MAXPATHLEN  + 1);
1611 
1612 	if (name == NULL) {
1613 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1614 		kmem_free(nm, len);
1615 		goto out;
1616 	}
1617 
1618 	resp->attrset = 0;
1619 
1620 	sarg.sbp = &sb;
1621 	sarg.is_referral = B_FALSE;
1622 	nfs4_ntov_table_init(&ntov);
1623 
1624 	status = do_rfs4_set_attrs(&resp->attrset,
1625 	    &args->createattrs, cs, &sarg, &ntov, NFS4ATTR_SETIT);
1626 
1627 	if (sarg.vap->va_mask == 0 && status == NFS4_OK)
1628 		status = NFS4ERR_INVAL;
1629 
1630 	if (status != NFS4_OK) {
1631 		*cs->statusp = resp->status = status;
1632 		if (name != nm)
1633 			kmem_free(name, MAXPATHLEN + 1);
1634 		kmem_free(nm, len);
1635 		nfs4_ntov_table_free(&ntov, &sarg);
1636 		resp->attrset = 0;
1637 		goto out;
1638 	}
1639 
1640 	/* Get "before" change value */
1641 	bva.va_mask = AT_CTIME|AT_SEQ|AT_MODE;
1642 	error = VOP_GETATTR(dvp, &bva, 0, cr, NULL);
1643 	if (error) {
1644 		*cs->statusp = resp->status = puterrno4(error);
1645 		if (name != nm)
1646 			kmem_free(name, MAXPATHLEN + 1);
1647 		kmem_free(nm, len);
1648 		nfs4_ntov_table_free(&ntov, &sarg);
1649 		resp->attrset = 0;
1650 		goto out;
1651 	}
1652 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bva.va_ctime)
1653 
1654 	vap = sarg.vap;
1655 
1656 	/*
1657 	 * Set the default initial values for attributes when the parent
1658 	 * directory does not have the VSUID/VSGID bit set and they have
1659 	 * not been specified in createattrs.
1660 	 */
1661 	if (!(bva.va_mode & VSUID) && (vap->va_mask & AT_UID) == 0) {
1662 		vap->va_uid = crgetuid(cr);
1663 		vap->va_mask |= AT_UID;
1664 	}
1665 	if (!(bva.va_mode & VSGID) && (vap->va_mask & AT_GID) == 0) {
1666 		vap->va_gid = crgetgid(cr);
1667 		vap->va_mask |= AT_GID;
1668 	}
1669 
1670 	vap->va_mask |= AT_TYPE;
1671 	switch (args->type) {
1672 	case NF4DIR:
1673 		vap->va_type = VDIR;
1674 		if ((vap->va_mask & AT_MODE) == 0) {
1675 			vap->va_mode = 0700;	/* default: owner rwx only */
1676 			vap->va_mask |= AT_MODE;
1677 		}
1678 		error = VOP_MKDIR(dvp, name, vap, &vp, cr, NULL, 0, NULL);
1679 		if (error)
1680 			break;
1681 
1682 		/*
1683 		 * Get the initial "after" sequence number, if it fails,
1684 		 * set to zero
1685 		 */
1686 		iva.va_mask = AT_SEQ;
1687 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1688 			iva.va_seq = 0;
1689 		break;
1690 	case NF4LNK:
1691 		vap->va_type = VLNK;
1692 		if ((vap->va_mask & AT_MODE) == 0) {
1693 			vap->va_mode = 0700;	/* default: owner rwx only */
1694 			vap->va_mask |= AT_MODE;
1695 		}
1696 
1697 		/*
1698 		 * symlink names must be treated as data
1699 		 */
1700 		lnm = utf8_to_str((utf8string *)&args->ftype4_u.linkdata,
1701 		    &llen, NULL);
1702 
1703 		if (lnm == NULL) {
1704 			*cs->statusp = resp->status = NFS4ERR_INVAL;
1705 			if (name != nm)
1706 				kmem_free(name, MAXPATHLEN + 1);
1707 			kmem_free(nm, len);
1708 			nfs4_ntov_table_free(&ntov, &sarg);
1709 			resp->attrset = 0;
1710 			goto out;
1711 		}
1712 
1713 		if (llen > MAXPATHLEN) {
1714 			*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1715 			if (name != nm)
1716 				kmem_free(name, MAXPATHLEN + 1);
1717 			kmem_free(nm, len);
1718 			kmem_free(lnm, llen);
1719 			nfs4_ntov_table_free(&ntov, &sarg);
1720 			resp->attrset = 0;
1721 			goto out;
1722 		}
1723 
1724 		lname = nfscmd_convname(ca, cs->exi, lnm,
1725 		    NFSCMD_CONV_INBOUND, MAXPATHLEN  + 1);
1726 
1727 		if (lname == NULL) {
1728 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
1729 			if (name != nm)
1730 				kmem_free(name, MAXPATHLEN + 1);
1731 			kmem_free(nm, len);
1732 			kmem_free(lnm, llen);
1733 			nfs4_ntov_table_free(&ntov, &sarg);
1734 			resp->attrset = 0;
1735 			goto out;
1736 		}
1737 
1738 		error = VOP_SYMLINK(dvp, name, vap, lname, cr, NULL, 0);
1739 		if (lname != lnm)
1740 			kmem_free(lname, MAXPATHLEN + 1);
1741 		kmem_free(lnm, llen);
1742 		if (error)
1743 			break;
1744 
1745 		/*
1746 		 * Get the initial "after" sequence number, if it fails,
1747 		 * set to zero
1748 		 */
1749 		iva.va_mask = AT_SEQ;
1750 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1751 			iva.va_seq = 0;
1752 
1753 		error = VOP_LOOKUP(dvp, name, &vp, NULL, 0, NULL, cr,
1754 		    NULL, NULL, NULL);
1755 		if (error)
1756 			break;
1757 
1758 		/*
1759 		 * va_seq is not safe over VOP calls, check it again
1760 		 * if it has changed zero out iva to force atomic = FALSE.
1761 		 */
1762 		iva2.va_mask = AT_SEQ;
1763 		if (VOP_GETATTR(dvp, &iva2, 0, cs->cr, NULL) ||
1764 		    iva2.va_seq != iva.va_seq)
1765 			iva.va_seq = 0;
1766 		break;
1767 	default:
1768 		/*
1769 		 * probably a special file.
1770 		 */
1771 		if ((vap->va_mask & AT_MODE) == 0) {
1772 			vap->va_mode = 0600;	/* default: owner rw only */
1773 			vap->va_mask |= AT_MODE;
1774 		}
1775 		syncval = FNODSYNC;
1776 		/*
1777 		 * We know this will only generate one VOP call
1778 		 */
1779 		vp = do_rfs4_op_mknod(args, resp, req, cs, vap, name);
1780 
1781 		if (vp == NULL) {
1782 			if (name != nm)
1783 				kmem_free(name, MAXPATHLEN + 1);
1784 			kmem_free(nm, len);
1785 			nfs4_ntov_table_free(&ntov, &sarg);
1786 			resp->attrset = 0;
1787 			goto out;
1788 		}
1789 
1790 		/*
1791 		 * Get the initial "after" sequence number, if it fails,
1792 		 * set to zero
1793 		 */
1794 		iva.va_mask = AT_SEQ;
1795 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1796 			iva.va_seq = 0;
1797 
1798 		break;
1799 	}
1800 	if (name != nm)
1801 		kmem_free(name, MAXPATHLEN + 1);
1802 	kmem_free(nm, len);
1803 
1804 	if (error) {
1805 		*cs->statusp = resp->status = puterrno4(error);
1806 	}
1807 
1808 	/*
1809 	 * Force modified data and metadata out to stable storage.
1810 	 */
1811 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
1812 
1813 	if (resp->status != NFS4_OK) {
1814 		if (vp != NULL)
1815 			VN_RELE(vp);
1816 		nfs4_ntov_table_free(&ntov, &sarg);
1817 		resp->attrset = 0;
1818 		goto out;
1819 	}
1820 
1821 	/*
1822 	 * Finish setup of cinfo response, "before" value already set.
1823 	 * Get "after" change value, if it fails, simply return the
1824 	 * before value.
1825 	 */
1826 	ava.va_mask = AT_CTIME|AT_SEQ;
1827 	if (VOP_GETATTR(dvp, &ava, 0, cr, NULL)) {
1828 		ava.va_ctime = bva.va_ctime;
1829 		ava.va_seq = 0;
1830 	}
1831 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, ava.va_ctime);
1832 
1833 	/*
1834 	 * True verification that object was created with correct
1835 	 * attrs is impossible.  The attrs could have been changed
1836 	 * immediately after object creation.  If attributes did
1837 	 * not verify, the only recourse for the server is to
1838 	 * destroy the object.  Maybe if some attrs (like gid)
1839 	 * are set incorrectly, the object should be destroyed;
1840 	 * however, seems bad as a default policy.  Do we really
1841 	 * want to destroy an object over one of the times not
1842 	 * verifying correctly?  For these reasons, the server
1843 	 * currently sets bits in attrset for createattrs
1844 	 * that were set; however, no verification is done.
1845 	 *
1846 	 * vmask_to_nmask accounts for vattr bits set on create
1847 	 *	[do_rfs4_set_attrs() only sets resp bits for
1848 	 *	 non-vattr/vfs bits.]
1849 	 * Mask off any bits set by default so as not to return
1850 	 * more attrset bits than were requested in createattrs
1851 	 */
1852 	nfs4_vmask_to_nmask(sarg.vap->va_mask, &resp->attrset);
1853 	resp->attrset &= args->createattrs.attrmask;
1854 	nfs4_ntov_table_free(&ntov, &sarg);
1855 
1856 	error = makefh4(&cs->fh, vp, cs->exi);
1857 	if (error) {
1858 		*cs->statusp = resp->status = puterrno4(error);
1859 	}
1860 
1861 	/*
1862 	 * The cinfo.atomic = TRUE only if we got no errors, we have
1863 	 * non-zero va_seq's, and it has incremented by exactly one
1864 	 * during the creation and it didn't change during the VOP_LOOKUP
1865 	 * or VOP_FSYNC.
1866 	 */
1867 	if (!error && bva.va_seq && iva.va_seq && ava.va_seq &&
1868 	    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
1869 		resp->cinfo.atomic = TRUE;
1870 	else
1871 		resp->cinfo.atomic = FALSE;
1872 
1873 	/*
1874 	 * Force modified metadata out to stable storage.
1875 	 *
1876 	 * if a underlying vp exists, pass it to VOP_FSYNC
1877 	 */
1878 	if (VOP_REALVP(vp, &realvp, NULL) == 0)
1879 		(void) VOP_FSYNC(realvp, syncval, cr, NULL);
1880 	else
1881 		(void) VOP_FSYNC(vp, syncval, cr, NULL);
1882 
1883 	if (resp->status != NFS4_OK) {
1884 		VN_RELE(vp);
1885 		goto out;
1886 	}
1887 	if (cs->vp)
1888 		VN_RELE(cs->vp);
1889 
1890 	cs->vp = vp;
1891 	*cs->statusp = resp->status = NFS4_OK;
1892 out:
1893 	DTRACE_NFSV4_2(op__create__done, struct compound_state *, cs,
1894 	    CREATE4res *, resp);
1895 }
1896 
1897 /*ARGSUSED*/
1898 static void
1899 rfs4_op_delegpurge(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1900     struct compound_state *cs)
1901 {
1902 	DTRACE_NFSV4_2(op__delegpurge__start, struct compound_state *, cs,
1903 	    DELEGPURGE4args *, &argop->nfs_argop4_u.opdelegpurge);
1904 
1905 	rfs4_op_inval(argop, resop, req, cs);
1906 
1907 	DTRACE_NFSV4_2(op__delegpurge__done, struct compound_state *, cs,
1908 	    DELEGPURGE4res *, &resop->nfs_resop4_u.opdelegpurge);
1909 }
1910 
1911 /*ARGSUSED*/
1912 static void
1913 rfs4_op_delegreturn(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1914     struct compound_state *cs)
1915 {
1916 	DELEGRETURN4args *args = &argop->nfs_argop4_u.opdelegreturn;
1917 	DELEGRETURN4res *resp = &resop->nfs_resop4_u.opdelegreturn;
1918 	rfs4_deleg_state_t *dsp;
1919 	nfsstat4 status;
1920 
1921 	DTRACE_NFSV4_2(op__delegreturn__start, struct compound_state *, cs,
1922 	    DELEGRETURN4args *, args);
1923 
1924 	status = rfs4_get_deleg_state(&args->deleg_stateid, &dsp);
1925 	resp->status = *cs->statusp = status;
1926 	if (status != NFS4_OK)
1927 		goto out;
1928 
1929 	/* Ensure specified filehandle matches */
1930 	if (cs->vp != dsp->rds_finfo->rf_vp) {
1931 		resp->status = *cs->statusp = NFS4ERR_BAD_STATEID;
1932 	} else
1933 		rfs4_return_deleg(dsp, FALSE);
1934 
1935 	rfs4_update_lease(dsp->rds_client);
1936 
1937 	rfs4_deleg_state_rele(dsp);
1938 out:
1939 	DTRACE_NFSV4_2(op__delegreturn__done, struct compound_state *, cs,
1940 	    DELEGRETURN4res *, resp);
1941 }
1942 
1943 /*
1944  * Check to see if a given "flavor" is an explicitly shared flavor.
1945  * The assumption of this routine is the "flavor" is already a valid
1946  * flavor in the secinfo list of "exi".
1947  *
1948  *	e.g.
1949  *		# share -o sec=flavor1 /export
1950  *		# share -o sec=flavor2 /export/home
1951  *
1952  *		flavor2 is not an explicitly shared flavor for /export,
1953  *		however it is in the secinfo list for /export thru the
1954  *		server namespace setup.
1955  */
1956 int
1957 is_exported_sec(int flavor, struct exportinfo *exi)
1958 {
1959 	int	i;
1960 	struct secinfo *sp;
1961 
1962 	sp = exi->exi_export.ex_secinfo;
1963 	for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
1964 		if (flavor == sp[i].s_secinfo.sc_nfsnum ||
1965 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE) {
1966 			return (SEC_REF_EXPORTED(&sp[i]));
1967 		}
1968 	}
1969 
1970 	/* Should not reach this point based on the assumption */
1971 	return (0);
1972 }
1973 
1974 /*
1975  * Check if the security flavor used in the request matches what is
1976  * required at the export point or at the root pseudo node (exi_root).
1977  *
1978  * returns 1 if there's a match or if exported with AUTH_NONE; 0 otherwise.
1979  *
1980  */
1981 static int
1982 secinfo_match_or_authnone(struct compound_state *cs)
1983 {
1984 	int	i;
1985 	struct secinfo *sp;
1986 
1987 	/*
1988 	 * Check cs->nfsflavor (from the request) against
1989 	 * the current export data in cs->exi.
1990 	 */
1991 	sp = cs->exi->exi_export.ex_secinfo;
1992 	for (i = 0; i < cs->exi->exi_export.ex_seccnt; i++) {
1993 		if (cs->nfsflavor == sp[i].s_secinfo.sc_nfsnum ||
1994 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE)
1995 			return (1);
1996 	}
1997 
1998 	return (0);
1999 }
2000 
2001 /*
2002  * Check the access authority for the client and return the correct error.
2003  */
2004 nfsstat4
2005 call_checkauth4(struct compound_state *cs, struct svc_req *req)
2006 {
2007 	int	authres;
2008 
2009 	/*
2010 	 * First, check if the security flavor used in the request
2011 	 * are among the flavors set in the server namespace.
2012 	 */
2013 	if (!secinfo_match_or_authnone(cs)) {
2014 		*cs->statusp = NFS4ERR_WRONGSEC;
2015 		return (*cs->statusp);
2016 	}
2017 
2018 	authres = checkauth4(cs, req);
2019 
2020 	if (authres > 0) {
2021 		*cs->statusp = NFS4_OK;
2022 		if (! (cs->access & CS_ACCESS_LIMITED))
2023 			cs->access = CS_ACCESS_OK;
2024 	} else if (authres == 0) {
2025 		*cs->statusp = NFS4ERR_ACCESS;
2026 	} else if (authres == -2) {
2027 		*cs->statusp = NFS4ERR_WRONGSEC;
2028 	} else {
2029 		*cs->statusp = NFS4ERR_DELAY;
2030 	}
2031 	return (*cs->statusp);
2032 }
2033 
2034 /*
2035  * bitmap4_to_attrmask is called by getattr and readdir.
2036  * It sets up the vattr mask and determines whether vfsstat call is needed
2037  * based on the input bitmap.
2038  * Returns nfsv4 status.
2039  */
2040 static nfsstat4
2041 bitmap4_to_attrmask(bitmap4 breq, struct nfs4_svgetit_arg *sargp)
2042 {
2043 	int i;
2044 	uint_t	va_mask;
2045 	struct statvfs64 *sbp = sargp->sbp;
2046 
2047 	sargp->sbp = NULL;
2048 	sargp->flag = 0;
2049 	sargp->rdattr_error = NFS4_OK;
2050 	sargp->mntdfid_set = FALSE;
2051 	if (sargp->cs->vp)
2052 		sargp->xattr = get_fh4_flag(&sargp->cs->fh,
2053 		    FH4_ATTRDIR | FH4_NAMEDATTR);
2054 	else
2055 		sargp->xattr = 0;
2056 
2057 	/*
2058 	 * Set rdattr_error_req to true if return error per
2059 	 * failed entry rather than fail the readdir.
2060 	 */
2061 	if (breq & FATTR4_RDATTR_ERROR_MASK)
2062 		sargp->rdattr_error_req = 1;
2063 	else
2064 		sargp->rdattr_error_req = 0;
2065 
2066 	/*
2067 	 * generate the va_mask
2068 	 * Handle the easy cases first
2069 	 */
2070 	switch (breq) {
2071 	case NFS4_NTOV_ATTR_MASK:
2072 		sargp->vap->va_mask = NFS4_NTOV_ATTR_AT_MASK;
2073 		return (NFS4_OK);
2074 
2075 	case NFS4_FS_ATTR_MASK:
2076 		sargp->vap->va_mask = NFS4_FS_ATTR_AT_MASK;
2077 		sargp->sbp = sbp;
2078 		return (NFS4_OK);
2079 
2080 	case NFS4_NTOV_ATTR_CACHE_MASK:
2081 		sargp->vap->va_mask = NFS4_NTOV_ATTR_CACHE_AT_MASK;
2082 		return (NFS4_OK);
2083 
2084 	case FATTR4_LEASE_TIME_MASK:
2085 		sargp->vap->va_mask = 0;
2086 		return (NFS4_OK);
2087 
2088 	default:
2089 		va_mask = 0;
2090 		for (i = 0; i < nfs4_ntov_map_size; i++) {
2091 			if ((breq & nfs4_ntov_map[i].fbit) &&
2092 			    nfs4_ntov_map[i].vbit)
2093 				va_mask |= nfs4_ntov_map[i].vbit;
2094 		}
2095 
2096 		/*
2097 		 * Check is vfsstat is needed
2098 		 */
2099 		if (breq & NFS4_FS_ATTR_MASK)
2100 			sargp->sbp = sbp;
2101 
2102 		sargp->vap->va_mask = va_mask;
2103 		return (NFS4_OK);
2104 	}
2105 	/* NOTREACHED */
2106 }
2107 
2108 /*
2109  * bitmap4_get_sysattrs is called by getattr and readdir.
2110  * It calls both VOP_GETATTR and VFS_STATVFS calls to get the attrs.
2111  * Returns nfsv4 status.
2112  */
2113 static nfsstat4
2114 bitmap4_get_sysattrs(struct nfs4_svgetit_arg *sargp)
2115 {
2116 	int error;
2117 	struct compound_state *cs = sargp->cs;
2118 	vnode_t *vp = cs->vp;
2119 
2120 	if (sargp->sbp != NULL) {
2121 		if (error = VFS_STATVFS(vp->v_vfsp, sargp->sbp)) {
2122 			sargp->sbp = NULL;	/* to identify error */
2123 			return (puterrno4(error));
2124 		}
2125 	}
2126 
2127 	return (rfs4_vop_getattr(vp, sargp->vap, 0, cs->cr));
2128 }
2129 
2130 static void
2131 nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp)
2132 {
2133 	ntovp->na = kmem_zalloc(sizeof (union nfs4_attr_u) * nfs4_ntov_map_size,
2134 	    KM_SLEEP);
2135 	ntovp->attrcnt = 0;
2136 	ntovp->vfsstat = FALSE;
2137 }
2138 
2139 static void
2140 nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
2141     struct nfs4_svgetit_arg *sargp)
2142 {
2143 	int i;
2144 	union nfs4_attr_u *na;
2145 	uint8_t *amap;
2146 
2147 	/*
2148 	 * XXX Should do the same checks for whether the bit is set
2149 	 */
2150 	for (i = 0, na = ntovp->na, amap = ntovp->amap;
2151 	    i < ntovp->attrcnt; i++, na++, amap++) {
2152 		(void) (*nfs4_ntov_map[*amap].sv_getit)(
2153 		    NFS4ATTR_FREEIT, sargp, na);
2154 	}
2155 	if ((sargp->op == NFS4ATTR_SETIT) || (sargp->op == NFS4ATTR_VERIT)) {
2156 		/*
2157 		 * xdr_free for getattr will be done later
2158 		 */
2159 		for (i = 0, na = ntovp->na, amap = ntovp->amap;
2160 		    i < ntovp->attrcnt; i++, na++, amap++) {
2161 			xdr_free(nfs4_ntov_map[*amap].xfunc, (caddr_t)na);
2162 		}
2163 	}
2164 	kmem_free(ntovp->na, sizeof (union nfs4_attr_u) * nfs4_ntov_map_size);
2165 }
2166 
2167 /*
2168  * do_rfs4_op_getattr gets the system attrs and converts into fattr4.
2169  */
2170 static nfsstat4
2171 do_rfs4_op_getattr(bitmap4 breq, fattr4 *fattrp,
2172     struct nfs4_svgetit_arg *sargp)
2173 {
2174 	int error = 0;
2175 	int i, k;
2176 	struct nfs4_ntov_table ntov;
2177 	XDR xdr;
2178 	ulong_t xdr_size;
2179 	char *xdr_attrs;
2180 	nfsstat4 status = NFS4_OK;
2181 	nfsstat4 prev_rdattr_error = sargp->rdattr_error;
2182 	union nfs4_attr_u *na;
2183 	uint8_t *amap;
2184 
2185 	sargp->op = NFS4ATTR_GETIT;
2186 	sargp->flag = 0;
2187 
2188 	fattrp->attrmask = 0;
2189 	/* if no bits requested, then return empty fattr4 */
2190 	if (breq == 0) {
2191 		fattrp->attrlist4_len = 0;
2192 		fattrp->attrlist4 = NULL;
2193 		return (NFS4_OK);
2194 	}
2195 
2196 	/*
2197 	 * return NFS4ERR_INVAL when client requests write-only attrs
2198 	 */
2199 	if (breq & (FATTR4_TIME_ACCESS_SET_MASK | FATTR4_TIME_MODIFY_SET_MASK))
2200 		return (NFS4ERR_INVAL);
2201 
2202 	nfs4_ntov_table_init(&ntov);
2203 	na = ntov.na;
2204 	amap = ntov.amap;
2205 
2206 	/*
2207 	 * Now loop to get or verify the attrs
2208 	 */
2209 	for (i = 0; i < nfs4_ntov_map_size; i++) {
2210 		if (breq & nfs4_ntov_map[i].fbit) {
2211 			if ((*nfs4_ntov_map[i].sv_getit)(
2212 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0) {
2213 
2214 				error = (*nfs4_ntov_map[i].sv_getit)(
2215 				    NFS4ATTR_GETIT, sargp, na);
2216 
2217 				/*
2218 				 * Possible error values:
2219 				 * >0 if sv_getit failed to
2220 				 * get the attr; 0 if succeeded;
2221 				 * <0 if rdattr_error and the
2222 				 * attribute cannot be returned.
2223 				 */
2224 				if (error && !(sargp->rdattr_error_req))
2225 					goto done;
2226 				/*
2227 				 * If error then just for entry
2228 				 */
2229 				if (error == 0) {
2230 					fattrp->attrmask |=
2231 					    nfs4_ntov_map[i].fbit;
2232 					*amap++ =
2233 					    (uint8_t)nfs4_ntov_map[i].nval;
2234 					na++;
2235 					(ntov.attrcnt)++;
2236 				} else if ((error > 0) &&
2237 				    (sargp->rdattr_error == NFS4_OK)) {
2238 					sargp->rdattr_error = puterrno4(error);
2239 				}
2240 				error = 0;
2241 			}
2242 		}
2243 	}
2244 
2245 	/*
2246 	 * If rdattr_error was set after the return value for it was assigned,
2247 	 * update it.
2248 	 */
2249 	if (prev_rdattr_error != sargp->rdattr_error) {
2250 		na = ntov.na;
2251 		amap = ntov.amap;
2252 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2253 			k = *amap;
2254 			if (k < FATTR4_RDATTR_ERROR) {
2255 				continue;
2256 			}
2257 			if ((k == FATTR4_RDATTR_ERROR) &&
2258 			    ((*nfs4_ntov_map[k].sv_getit)(
2259 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0)) {
2260 
2261 				(void) (*nfs4_ntov_map[k].sv_getit)(
2262 				    NFS4ATTR_GETIT, sargp, na);
2263 			}
2264 			break;
2265 		}
2266 	}
2267 
2268 	xdr_size = 0;
2269 	na = ntov.na;
2270 	amap = ntov.amap;
2271 	for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2272 		xdr_size += xdr_sizeof(nfs4_ntov_map[*amap].xfunc, na);
2273 	}
2274 
2275 	fattrp->attrlist4_len = xdr_size;
2276 	if (xdr_size) {
2277 		/* freed by rfs4_op_getattr_free() */
2278 		fattrp->attrlist4 = xdr_attrs = kmem_zalloc(xdr_size, KM_SLEEP);
2279 
2280 		xdrmem_create(&xdr, xdr_attrs, xdr_size, XDR_ENCODE);
2281 
2282 		na = ntov.na;
2283 		amap = ntov.amap;
2284 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2285 			if (!(*nfs4_ntov_map[*amap].xfunc)(&xdr, na)) {
2286 				DTRACE_PROBE1(nfss__e__getattr4_encfail,
2287 				    int, *amap);
2288 				status = NFS4ERR_SERVERFAULT;
2289 				break;
2290 			}
2291 		}
2292 		/* xdrmem_destroy(&xdrs); */	/* NO-OP */
2293 	} else {
2294 		fattrp->attrlist4 = NULL;
2295 	}
2296 done:
2297 
2298 	nfs4_ntov_table_free(&ntov, sargp);
2299 
2300 	if (error != 0)
2301 		status = puterrno4(error);
2302 
2303 	return (status);
2304 }
2305 
2306 /* ARGSUSED */
2307 static void
2308 rfs4_op_getattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2309     struct compound_state *cs)
2310 {
2311 	GETATTR4args *args = &argop->nfs_argop4_u.opgetattr;
2312 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2313 	struct nfs4_svgetit_arg sarg;
2314 	struct statvfs64 sb;
2315 	nfsstat4 status;
2316 
2317 	DTRACE_NFSV4_2(op__getattr__start, struct compound_state *, cs,
2318 	    GETATTR4args *, args);
2319 
2320 	if (cs->vp == NULL) {
2321 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2322 		goto out;
2323 	}
2324 
2325 	if (cs->access == CS_ACCESS_DENIED) {
2326 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2327 		goto out;
2328 	}
2329 
2330 	sarg.sbp = &sb;
2331 	sarg.cs = cs;
2332 	sarg.is_referral = B_FALSE;
2333 
2334 	status = bitmap4_to_attrmask(args->attr_request, &sarg);
2335 	if (status == NFS4_OK) {
2336 
2337 		status = bitmap4_get_sysattrs(&sarg);
2338 		if (status == NFS4_OK) {
2339 
2340 			/* Is this a referral? */
2341 			if (vn_is_nfs_reparse(cs->vp, cs->cr)) {
2342 				/* Older V4 Solaris client sees a link */
2343 				if (client_is_downrev(req))
2344 					sarg.vap->va_type = VLNK;
2345 				else
2346 					sarg.is_referral = B_TRUE;
2347 			}
2348 
2349 			status = do_rfs4_op_getattr(args->attr_request,
2350 			    &resp->obj_attributes, &sarg);
2351 		}
2352 	}
2353 	*cs->statusp = resp->status = status;
2354 out:
2355 	DTRACE_NFSV4_2(op__getattr__done, struct compound_state *, cs,
2356 	    GETATTR4res *, resp);
2357 }
2358 
2359 static void
2360 rfs4_op_getattr_free(nfs_resop4 *resop)
2361 {
2362 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2363 
2364 	nfs4_fattr4_free(&resp->obj_attributes);
2365 }
2366 
2367 /* ARGSUSED */
2368 static void
2369 rfs4_op_getfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2370     struct compound_state *cs)
2371 {
2372 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2373 
2374 	DTRACE_NFSV4_1(op__getfh__start, struct compound_state *, cs);
2375 
2376 	if (cs->vp == NULL) {
2377 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2378 		goto out;
2379 	}
2380 	if (cs->access == CS_ACCESS_DENIED) {
2381 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2382 		goto out;
2383 	}
2384 
2385 	/* check for reparse point at the share point */
2386 	if (cs->exi->exi_moved || vn_is_nfs_reparse(cs->exi->exi_vp, cs->cr)) {
2387 		/* it's all bad */
2388 		cs->exi->exi_moved = 1;
2389 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2390 		DTRACE_PROBE2(nfs4serv__func__referral__shared__moved,
2391 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2392 		return;
2393 	}
2394 
2395 	/* check for reparse point at vp */
2396 	if (vn_is_nfs_reparse(cs->vp, cs->cr) && !client_is_downrev(req)) {
2397 		/* it's not all bad */
2398 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2399 		DTRACE_PROBE2(nfs4serv__func__referral__moved,
2400 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2401 		return;
2402 	}
2403 
2404 	resp->object.nfs_fh4_val =
2405 	    kmem_alloc(cs->fh.nfs_fh4_len, KM_SLEEP);
2406 	nfs_fh4_copy(&cs->fh, &resp->object);
2407 	*cs->statusp = resp->status = NFS4_OK;
2408 out:
2409 	DTRACE_NFSV4_2(op__getfh__done, struct compound_state *, cs,
2410 	    GETFH4res *, resp);
2411 }
2412 
2413 static void
2414 rfs4_op_getfh_free(nfs_resop4 *resop)
2415 {
2416 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2417 
2418 	if (resp->status == NFS4_OK &&
2419 	    resp->object.nfs_fh4_val != NULL) {
2420 		kmem_free(resp->object.nfs_fh4_val, resp->object.nfs_fh4_len);
2421 		resp->object.nfs_fh4_val = NULL;
2422 		resp->object.nfs_fh4_len = 0;
2423 	}
2424 }
2425 
2426 /*
2427  * illegal: args: void
2428  *	    res : status (NFS4ERR_OP_ILLEGAL)
2429  */
2430 /* ARGSUSED */
2431 static void
2432 rfs4_op_illegal(nfs_argop4 *argop, nfs_resop4 *resop,
2433     struct svc_req *req, struct compound_state *cs)
2434 {
2435 	ILLEGAL4res *resp = &resop->nfs_resop4_u.opillegal;
2436 
2437 	resop->resop = OP_ILLEGAL;
2438 	*cs->statusp = resp->status = NFS4ERR_OP_ILLEGAL;
2439 }
2440 
2441 /*
2442  * link: args: SAVED_FH: file, CURRENT_FH: target directory
2443  *	 res: status. If success - CURRENT_FH unchanged, return change_info
2444  */
2445 /* ARGSUSED */
2446 static void
2447 rfs4_op_link(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2448     struct compound_state *cs)
2449 {
2450 	LINK4args *args = &argop->nfs_argop4_u.oplink;
2451 	LINK4res *resp = &resop->nfs_resop4_u.oplink;
2452 	int error;
2453 	vnode_t *vp;
2454 	vnode_t *dvp;
2455 	struct vattr bdva, idva, adva;
2456 	char *nm;
2457 	uint_t  len;
2458 	struct sockaddr *ca;
2459 	char *name = NULL;
2460 	nfsstat4 status;
2461 
2462 	DTRACE_NFSV4_2(op__link__start, struct compound_state *, cs,
2463 	    LINK4args *, args);
2464 
2465 	/* SAVED_FH: source object */
2466 	vp = cs->saved_vp;
2467 	if (vp == NULL) {
2468 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2469 		goto out;
2470 	}
2471 
2472 	/* CURRENT_FH: target directory */
2473 	dvp = cs->vp;
2474 	if (dvp == NULL) {
2475 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2476 		goto out;
2477 	}
2478 
2479 	/*
2480 	 * If there is a non-shared filesystem mounted on this vnode,
2481 	 * do not allow to link any file in this directory.
2482 	 */
2483 	if (vn_ismntpt(dvp)) {
2484 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2485 		goto out;
2486 	}
2487 
2488 	if (cs->access == CS_ACCESS_DENIED) {
2489 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2490 		goto out;
2491 	}
2492 
2493 	/* Check source object's type validity */
2494 	if (vp->v_type == VDIR) {
2495 		*cs->statusp = resp->status = NFS4ERR_ISDIR;
2496 		goto out;
2497 	}
2498 
2499 	/* Check target directory's type */
2500 	if (dvp->v_type != VDIR) {
2501 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2502 		goto out;
2503 	}
2504 
2505 	if (cs->saved_exi != cs->exi) {
2506 		*cs->statusp = resp->status = NFS4ERR_XDEV;
2507 		goto out;
2508 	}
2509 
2510 	status = utf8_dir_verify(&args->newname);
2511 	if (status != NFS4_OK) {
2512 		*cs->statusp = resp->status = status;
2513 		goto out;
2514 	}
2515 
2516 	nm = utf8_to_fn(&args->newname, &len, NULL);
2517 	if (nm == NULL) {
2518 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2519 		goto out;
2520 	}
2521 
2522 	if (len > MAXNAMELEN) {
2523 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2524 		kmem_free(nm, len);
2525 		goto out;
2526 	}
2527 
2528 	if (rdonly4(req, cs)) {
2529 		*cs->statusp = resp->status = NFS4ERR_ROFS;
2530 		kmem_free(nm, len);
2531 		goto out;
2532 	}
2533 
2534 	/* Get "before" change value */
2535 	bdva.va_mask = AT_CTIME|AT_SEQ;
2536 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
2537 	if (error) {
2538 		*cs->statusp = resp->status = puterrno4(error);
2539 		kmem_free(nm, len);
2540 		goto out;
2541 	}
2542 
2543 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2544 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2545 	    MAXPATHLEN  + 1);
2546 
2547 	if (name == NULL) {
2548 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2549 		kmem_free(nm, len);
2550 		goto out;
2551 	}
2552 
2553 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
2554 
2555 	error = VOP_LINK(dvp, vp, name, cs->cr, NULL, 0);
2556 
2557 	if (nm != name)
2558 		kmem_free(name, MAXPATHLEN + 1);
2559 	kmem_free(nm, len);
2560 
2561 	/*
2562 	 * Get the initial "after" sequence number, if it fails, set to zero
2563 	 */
2564 	idva.va_mask = AT_SEQ;
2565 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
2566 		idva.va_seq = 0;
2567 
2568 	/*
2569 	 * Force modified data and metadata out to stable storage.
2570 	 */
2571 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
2572 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
2573 
2574 	if (error) {
2575 		*cs->statusp = resp->status = puterrno4(error);
2576 		goto out;
2577 	}
2578 
2579 	/*
2580 	 * Get "after" change value, if it fails, simply return the
2581 	 * before value.
2582 	 */
2583 	adva.va_mask = AT_CTIME|AT_SEQ;
2584 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
2585 		adva.va_ctime = bdva.va_ctime;
2586 		adva.va_seq = 0;
2587 	}
2588 
2589 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
2590 
2591 	/*
2592 	 * The cinfo.atomic = TRUE only if we have
2593 	 * non-zero va_seq's, and it has incremented by exactly one
2594 	 * during the VOP_LINK and it didn't change during the VOP_FSYNC.
2595 	 */
2596 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
2597 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
2598 		resp->cinfo.atomic = TRUE;
2599 	else
2600 		resp->cinfo.atomic = FALSE;
2601 
2602 	*cs->statusp = resp->status = NFS4_OK;
2603 out:
2604 	DTRACE_NFSV4_2(op__link__done, struct compound_state *, cs,
2605 	    LINK4res *, resp);
2606 }
2607 
2608 /*
2609  * Used by rfs4_op_lookup and rfs4_op_lookupp to do the actual work.
2610  */
2611 
2612 /* ARGSUSED */
2613 static nfsstat4
2614 do_rfs4_op_lookup(char *nm, struct svc_req *req, struct compound_state *cs)
2615 {
2616 	int error;
2617 	int different_export = 0;
2618 	vnode_t *vp, *pre_tvp = NULL, *oldvp = NULL;
2619 	struct exportinfo *exi = NULL, *pre_exi = NULL, *oexi = NULL;
2620 	nfsstat4 stat;
2621 	fid_t fid;
2622 	int attrdir, dotdot, walk;
2623 	bool_t is_newvp = FALSE;
2624 
2625 	if (cs->vp->v_flag & V_XATTRDIR) {
2626 		attrdir = 1;
2627 		ASSERT(get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2628 	} else {
2629 		attrdir = 0;
2630 		ASSERT(! get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2631 	}
2632 
2633 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
2634 
2635 	/*
2636 	 * If dotdotting, then need to check whether it's
2637 	 * above the root of a filesystem, or above an
2638 	 * export point.
2639 	 */
2640 	if (dotdot) {
2641 
2642 		/*
2643 		 * If dotdotting at the root of a filesystem, then
2644 		 * need to traverse back to the mounted-on filesystem
2645 		 * and do the dotdot lookup there.
2646 		 */
2647 		if (cs->vp->v_flag & VROOT) {
2648 
2649 			/*
2650 			 * If at the system root, then can
2651 			 * go up no further.
2652 			 */
2653 			if (VN_CMP(cs->vp, rootdir))
2654 				return (puterrno4(ENOENT));
2655 
2656 			/*
2657 			 * Traverse back to the mounted-on filesystem
2658 			 */
2659 			cs->vp = untraverse(cs->vp);
2660 
2661 			/*
2662 			 * Set the different_export flag so we remember
2663 			 * to pick up a new exportinfo entry for
2664 			 * this new filesystem.
2665 			 */
2666 			different_export = 1;
2667 		} else {
2668 
2669 			/*
2670 			 * If dotdotting above an export point then set
2671 			 * the different_export to get new export info.
2672 			 */
2673 			different_export = nfs_exported(cs->exi, cs->vp);
2674 		}
2675 	}
2676 
2677 	error = VOP_LOOKUP(cs->vp, nm, &vp, NULL, 0, NULL, cs->cr,
2678 	    NULL, NULL, NULL);
2679 	if (error)
2680 		return (puterrno4(error));
2681 
2682 	/*
2683 	 * If the vnode is in a pseudo filesystem, check whether it is visible.
2684 	 *
2685 	 * XXX if the vnode is a symlink and it is not visible in
2686 	 * a pseudo filesystem, return ENOENT (not following symlink).
2687 	 * V4 client can not mount such symlink. This is a regression
2688 	 * from V2/V3.
2689 	 *
2690 	 * In the same exported filesystem, if the security flavor used
2691 	 * is not an explicitly shared flavor, limit the view to the visible
2692 	 * list entries only. This is not a WRONGSEC case because it's already
2693 	 * checked via PUTROOTFH/PUTPUBFH or PUTFH.
2694 	 */
2695 	if (!different_export &&
2696 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
2697 	    cs->access & CS_ACCESS_LIMITED)) {
2698 		if (! nfs_visible(cs->exi, vp, &different_export)) {
2699 			VN_RELE(vp);
2700 			return (puterrno4(ENOENT));
2701 		}
2702 	}
2703 
2704 	/*
2705 	 * If it's a mountpoint, then traverse it.
2706 	 */
2707 	if (vn_ismntpt(vp)) {
2708 		pre_exi = cs->exi;	/* save pre-traversed exportinfo */
2709 		pre_tvp = vp;		/* save pre-traversed vnode	*/
2710 
2711 		/*
2712 		 * hold pre_tvp to counteract rele by traverse.  We will
2713 		 * need pre_tvp below if checkexport fails
2714 		 */
2715 		VN_HOLD(pre_tvp);
2716 		if ((error = traverse(&vp)) != 0) {
2717 			VN_RELE(vp);
2718 			VN_RELE(pre_tvp);
2719 			return (puterrno4(error));
2720 		}
2721 		different_export = 1;
2722 	} else if (vp->v_vfsp != cs->vp->v_vfsp) {
2723 		/*
2724 		 * The vfsp comparison is to handle the case where
2725 		 * a LOFS mount is shared.  lo_lookup traverses mount points,
2726 		 * and NFS is unaware of local fs transistions because
2727 		 * v_vfsmountedhere isn't set.  For this special LOFS case,
2728 		 * the dir and the obj returned by lookup will have different
2729 		 * vfs ptrs.
2730 		 */
2731 		different_export = 1;
2732 	}
2733 
2734 	if (different_export) {
2735 
2736 		bzero(&fid, sizeof (fid));
2737 		fid.fid_len = MAXFIDSZ;
2738 		error = vop_fid_pseudo(vp, &fid);
2739 		if (error) {
2740 			VN_RELE(vp);
2741 			if (pre_tvp)
2742 				VN_RELE(pre_tvp);
2743 			return (puterrno4(error));
2744 		}
2745 
2746 		if (dotdot)
2747 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
2748 		else
2749 			exi = checkexport(&vp->v_vfsp->vfs_fsid, &fid, vp);
2750 
2751 		if (exi == NULL) {
2752 			if (pre_tvp) {
2753 				/*
2754 				 * If this vnode is a mounted-on vnode,
2755 				 * but the mounted-on file system is not
2756 				 * exported, send back the filehandle for
2757 				 * the mounted-on vnode, not the root of
2758 				 * the mounted-on file system.
2759 				 */
2760 				VN_RELE(vp);
2761 				vp = pre_tvp;
2762 				exi = pre_exi;
2763 				if (exi)
2764 					exi_hold(exi);
2765 			} else {
2766 				VN_RELE(vp);
2767 				return (puterrno4(EACCES));
2768 			}
2769 		} else if (pre_tvp) {
2770 			/* we're done with pre_tvp now. release extra hold */
2771 			VN_RELE(pre_tvp);
2772 		}
2773 
2774 		if (cs->exi)
2775 			exi_rele(cs->exi);
2776 		cs->exi = exi;
2777 
2778 		/*
2779 		 * Now we do a checkauth4. The reason is that
2780 		 * this client/user may not have access to the new
2781 		 * exported file system, and if he does,
2782 		 * the client/user may be mapped to a different uid.
2783 		 *
2784 		 * We start with a new cr, because the checkauth4 done
2785 		 * in the PUT*FH operation over wrote the cred's uid,
2786 		 * gid, etc, and we want the real thing before calling
2787 		 * checkauth4()
2788 		 */
2789 		crfree(cs->cr);
2790 		cs->cr = crdup(cs->basecr);
2791 
2792 		oldvp = cs->vp;
2793 		cs->vp = vp;
2794 		is_newvp = TRUE;
2795 
2796 		stat = call_checkauth4(cs, req);
2797 		if (stat != NFS4_OK) {
2798 			VN_RELE(cs->vp);
2799 			cs->vp = oldvp;
2800 			return (stat);
2801 		}
2802 	}
2803 
2804 	/*
2805 	 * After various NFS checks, do a label check on the path
2806 	 * component. The label on this path should either be the
2807 	 * global zone's label or a zone's label. We are only
2808 	 * interested in the zone's label because exported files
2809 	 * in global zone is accessible (though read-only) to
2810 	 * clients. The exportability/visibility check is already
2811 	 * done before reaching this code.
2812 	 */
2813 	if (is_system_labeled()) {
2814 		bslabel_t *clabel;
2815 
2816 		ASSERT(req->rq_label != NULL);
2817 		clabel = req->rq_label;
2818 		DTRACE_PROBE2(tx__rfs4__log__info__oplookup__clabel, char *,
2819 		    "got client label from request(1)", struct svc_req *, req);
2820 
2821 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
2822 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
2823 			    cs->exi)) {
2824 				error = EACCES;
2825 				goto err_out;
2826 			}
2827 		} else {
2828 			/*
2829 			 * We grant access to admin_low label clients
2830 			 * only if the client is trusted, i.e. also
2831 			 * running Solaris Trusted Extension.
2832 			 */
2833 			struct sockaddr	*ca;
2834 			int		addr_type;
2835 			void		*ipaddr;
2836 			tsol_tpc_t	*tp;
2837 
2838 			ca = (struct sockaddr *)svc_getrpccaller(
2839 			    req->rq_xprt)->buf;
2840 			if (ca->sa_family == AF_INET) {
2841 				addr_type = IPV4_VERSION;
2842 				ipaddr = &((struct sockaddr_in *)ca)->sin_addr;
2843 			} else if (ca->sa_family == AF_INET6) {
2844 				addr_type = IPV6_VERSION;
2845 				ipaddr = &((struct sockaddr_in6 *)
2846 				    ca)->sin6_addr;
2847 			}
2848 			tp = find_tpc(ipaddr, addr_type, B_FALSE);
2849 			if (tp == NULL || tp->tpc_tp.tp_doi !=
2850 			    l_admin_low->tsl_doi || tp->tpc_tp.host_type !=
2851 			    SUN_CIPSO) {
2852 				if (tp != NULL)
2853 					TPC_RELE(tp);
2854 				error = EACCES;
2855 				goto err_out;
2856 			}
2857 			TPC_RELE(tp);
2858 		}
2859 	}
2860 
2861 	error = makefh4(&cs->fh, vp, cs->exi);
2862 
2863 err_out:
2864 	if (error) {
2865 		if (is_newvp) {
2866 			VN_RELE(cs->vp);
2867 			cs->vp = oldvp;
2868 		} else
2869 			VN_RELE(vp);
2870 		return (puterrno4(error));
2871 	}
2872 
2873 	if (!is_newvp) {
2874 		if (cs->vp)
2875 			VN_RELE(cs->vp);
2876 		cs->vp = vp;
2877 	} else if (oldvp)
2878 		VN_RELE(oldvp);
2879 
2880 	/*
2881 	 * if did lookup on attrdir and didn't lookup .., set named
2882 	 * attr fh flag
2883 	 */
2884 	if (attrdir && ! dotdot)
2885 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
2886 
2887 	/* Assume false for now, open proc will set this */
2888 	cs->mandlock = FALSE;
2889 
2890 	return (NFS4_OK);
2891 }
2892 
2893 /* ARGSUSED */
2894 static void
2895 rfs4_op_lookup(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2896     struct compound_state *cs)
2897 {
2898 	LOOKUP4args *args = &argop->nfs_argop4_u.oplookup;
2899 	LOOKUP4res *resp = &resop->nfs_resop4_u.oplookup;
2900 	char *nm;
2901 	uint_t len;
2902 	struct sockaddr *ca;
2903 	char *name = NULL;
2904 	nfsstat4 status;
2905 
2906 	DTRACE_NFSV4_2(op__lookup__start, struct compound_state *, cs,
2907 	    LOOKUP4args *, args);
2908 
2909 	if (cs->vp == NULL) {
2910 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2911 		goto out;
2912 	}
2913 
2914 	if (cs->vp->v_type == VLNK) {
2915 		*cs->statusp = resp->status = NFS4ERR_SYMLINK;
2916 		goto out;
2917 	}
2918 
2919 	if (cs->vp->v_type != VDIR) {
2920 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2921 		goto out;
2922 	}
2923 
2924 	status = utf8_dir_verify(&args->objname);
2925 	if (status != NFS4_OK) {
2926 		*cs->statusp = resp->status = status;
2927 		goto out;
2928 	}
2929 
2930 	nm = utf8_to_str(&args->objname, &len, NULL);
2931 	if (nm == NULL) {
2932 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2933 		goto out;
2934 	}
2935 
2936 	if (len > MAXNAMELEN) {
2937 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2938 		kmem_free(nm, len);
2939 		goto out;
2940 	}
2941 
2942 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2943 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2944 	    MAXPATHLEN  + 1);
2945 
2946 	if (name == NULL) {
2947 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2948 		kmem_free(nm, len);
2949 		goto out;
2950 	}
2951 
2952 	*cs->statusp = resp->status = do_rfs4_op_lookup(name, req, cs);
2953 
2954 	if (name != nm)
2955 		kmem_free(name, MAXPATHLEN + 1);
2956 	kmem_free(nm, len);
2957 
2958 out:
2959 	DTRACE_NFSV4_2(op__lookup__done, struct compound_state *, cs,
2960 	    LOOKUP4res *, resp);
2961 }
2962 
2963 /* ARGSUSED */
2964 static void
2965 rfs4_op_lookupp(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
2966     struct compound_state *cs)
2967 {
2968 	LOOKUPP4res *resp = &resop->nfs_resop4_u.oplookupp;
2969 
2970 	DTRACE_NFSV4_1(op__lookupp__start, struct compound_state *, cs);
2971 
2972 	if (cs->vp == NULL) {
2973 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2974 		goto out;
2975 	}
2976 
2977 	if (cs->vp->v_type != VDIR) {
2978 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2979 		goto out;
2980 	}
2981 
2982 	*cs->statusp = resp->status = do_rfs4_op_lookup("..", req, cs);
2983 
2984 	/*
2985 	 * From NFSV4 Specification, LOOKUPP should not check for
2986 	 * NFS4ERR_WRONGSEC. Retrun NFS4_OK instead.
2987 	 */
2988 	if (resp->status == NFS4ERR_WRONGSEC) {
2989 		*cs->statusp = resp->status = NFS4_OK;
2990 	}
2991 
2992 out:
2993 	DTRACE_NFSV4_2(op__lookupp__done, struct compound_state *, cs,
2994 	    LOOKUPP4res *, resp);
2995 }
2996 
2997 
2998 /*ARGSUSED2*/
2999 static void
3000 rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3001     struct compound_state *cs)
3002 {
3003 	OPENATTR4args	*args = &argop->nfs_argop4_u.opopenattr;
3004 	OPENATTR4res	*resp = &resop->nfs_resop4_u.opopenattr;
3005 	vnode_t		*avp = NULL;
3006 	int		lookup_flags = LOOKUP_XATTR, error;
3007 	int		exp_ro = 0;
3008 
3009 	DTRACE_NFSV4_2(op__openattr__start, struct compound_state *, cs,
3010 	    OPENATTR4args *, args);
3011 
3012 	if (cs->vp == NULL) {
3013 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3014 		goto out;
3015 	}
3016 
3017 	if ((cs->vp->v_vfsp->vfs_flag & VFS_XATTR) == 0 &&
3018 	    !vfs_has_feature(cs->vp->v_vfsp, VFSFT_SYSATTR_VIEWS)) {
3019 		*cs->statusp = resp->status = puterrno4(ENOTSUP);
3020 		goto out;
3021 	}
3022 
3023 	/*
3024 	 * If file system supports passing ACE mask to VOP_ACCESS then
3025 	 * check for ACE_READ_NAMED_ATTRS, otherwise do legacy checks
3026 	 */
3027 
3028 	if (vfs_has_feature(cs->vp->v_vfsp, VFSFT_ACEMASKONACCESS))
3029 		error = VOP_ACCESS(cs->vp, ACE_READ_NAMED_ATTRS,
3030 		    V_ACE_MASK, cs->cr, NULL);
3031 	else
3032 		error = ((VOP_ACCESS(cs->vp, VREAD, 0, cs->cr, NULL) != 0) &&
3033 		    (VOP_ACCESS(cs->vp, VWRITE, 0, cs->cr, NULL) != 0) &&
3034 		    (VOP_ACCESS(cs->vp, VEXEC, 0, cs->cr, NULL) != 0));
3035 
3036 	if (error) {
3037 		*cs->statusp = resp->status = puterrno4(EACCES);
3038 		goto out;
3039 	}
3040 
3041 	/*
3042 	 * The CREATE_XATTR_DIR VOP flag cannot be specified if
3043 	 * the file system is exported read-only -- regardless of
3044 	 * createdir flag.  Otherwise the attrdir would be created
3045 	 * (assuming server fs isn't mounted readonly locally).  If
3046 	 * VOP_LOOKUP returns ENOENT in this case, the error will
3047 	 * be translated into EROFS.  ENOSYS is mapped to ENOTSUP
3048 	 * because specfs has no VOP_LOOKUP op, so the macro would
3049 	 * return ENOSYS.  EINVAL is returned by all (current)
3050 	 * Solaris file system implementations when any of their
3051 	 * restrictions are violated (xattr(dir) can't have xattrdir).
3052 	 * Returning NOTSUPP is more appropriate in this case
3053 	 * because the object will never be able to have an attrdir.
3054 	 */
3055 	if (args->createdir && ! (exp_ro = rdonly4(req, cs)))
3056 		lookup_flags |= CREATE_XATTR_DIR;
3057 
3058 	error = VOP_LOOKUP(cs->vp, "", &avp, NULL, lookup_flags, NULL, cs->cr,
3059 	    NULL, NULL, NULL);
3060 
3061 	if (error) {
3062 		if (error == ENOENT && args->createdir && exp_ro)
3063 			*cs->statusp = resp->status = puterrno4(EROFS);
3064 		else if (error == EINVAL || error == ENOSYS)
3065 			*cs->statusp = resp->status = puterrno4(ENOTSUP);
3066 		else
3067 			*cs->statusp = resp->status = puterrno4(error);
3068 		goto out;
3069 	}
3070 
3071 	ASSERT(avp->v_flag & V_XATTRDIR);
3072 
3073 	error = makefh4(&cs->fh, avp, cs->exi);
3074 
3075 	if (error) {
3076 		VN_RELE(avp);
3077 		*cs->statusp = resp->status = puterrno4(error);
3078 		goto out;
3079 	}
3080 
3081 	VN_RELE(cs->vp);
3082 	cs->vp = avp;
3083 
3084 	/*
3085 	 * There is no requirement for an attrdir fh flag
3086 	 * because the attrdir has a vnode flag to distinguish
3087 	 * it from regular (non-xattr) directories.  The
3088 	 * FH4_ATTRDIR flag is set for future sanity checks.
3089 	 */
3090 	set_fh4_flag(&cs->fh, FH4_ATTRDIR);
3091 	*cs->statusp = resp->status = NFS4_OK;
3092 
3093 out:
3094 	DTRACE_NFSV4_2(op__openattr__done, struct compound_state *, cs,
3095 	    OPENATTR4res *, resp);
3096 }
3097 
3098 static int
3099 do_io(int direction, vnode_t *vp, struct uio *uio, int ioflag, cred_t *cred,
3100     caller_context_t *ct)
3101 {
3102 	int error;
3103 	int i;
3104 	clock_t delaytime;
3105 
3106 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
3107 
3108 	/*
3109 	 * Don't block on mandatory locks. If this routine returns
3110 	 * EAGAIN, the caller should return NFS4ERR_LOCKED.
3111 	 */
3112 	uio->uio_fmode = FNONBLOCK;
3113 
3114 	for (i = 0; i < rfs4_maxlock_tries; i++) {
3115 
3116 
3117 		if (direction == FREAD) {
3118 			(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, ct);
3119 			error = VOP_READ(vp, uio, ioflag, cred, ct);
3120 			VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, ct);
3121 		} else {
3122 			(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, ct);
3123 			error = VOP_WRITE(vp, uio, ioflag, cred, ct);
3124 			VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, ct);
3125 		}
3126 
3127 		if (error != EAGAIN)
3128 			break;
3129 
3130 		if (i < rfs4_maxlock_tries - 1) {
3131 			delay(delaytime);
3132 			delaytime *= 2;
3133 		}
3134 	}
3135 
3136 	return (error);
3137 }
3138 
3139 /* ARGSUSED */
3140 static void
3141 rfs4_op_read(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3142     struct compound_state *cs)
3143 {
3144 	READ4args *args = &argop->nfs_argop4_u.opread;
3145 	READ4res *resp = &resop->nfs_resop4_u.opread;
3146 	int error;
3147 	int verror;
3148 	vnode_t *vp;
3149 	struct vattr va;
3150 	struct iovec iov, *iovp = NULL;
3151 	int iovcnt;
3152 	struct uio uio;
3153 	u_offset_t offset;
3154 	bool_t *deleg = &cs->deleg;
3155 	nfsstat4 stat;
3156 	int in_crit = 0;
3157 	mblk_t *mp = NULL;
3158 	int alloc_err = 0;
3159 	int rdma_used = 0;
3160 	int loaned_buffers;
3161 	caller_context_t ct;
3162 	struct uio *uiop;
3163 
3164 	DTRACE_NFSV4_2(op__read__start, struct compound_state *, cs,
3165 	    READ4args, args);
3166 
3167 	vp = cs->vp;
3168 	if (vp == NULL) {
3169 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3170 		goto out;
3171 	}
3172 	if (cs->access == CS_ACCESS_DENIED) {
3173 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3174 		goto out;
3175 	}
3176 
3177 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3178 	    deleg, TRUE, &ct)) != NFS4_OK) {
3179 		*cs->statusp = resp->status = stat;
3180 		goto out;
3181 	}
3182 
3183 	/*
3184 	 * Enter the critical region before calling VOP_RWLOCK
3185 	 * to avoid a deadlock with write requests.
3186 	 */
3187 	if (nbl_need_check(vp)) {
3188 		nbl_start_crit(vp, RW_READER);
3189 		in_crit = 1;
3190 		if (nbl_conflict(vp, NBL_READ, args->offset, args->count, 0,
3191 		    &ct)) {
3192 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
3193 			goto out;
3194 		}
3195 	}
3196 
3197 	if (args->wlist) {
3198 		if (args->count > clist_len(args->wlist)) {
3199 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3200 			goto out;
3201 		}
3202 		rdma_used = 1;
3203 	}
3204 
3205 	/* use loaned buffers for TCP */
3206 	loaned_buffers = (nfs_loaned_buffers && !rdma_used) ? 1 : 0;
3207 
3208 	va.va_mask = AT_MODE|AT_SIZE|AT_UID;
3209 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3210 
3211 	/*
3212 	 * If we can't get the attributes, then we can't do the
3213 	 * right access checking.  So, we'll fail the request.
3214 	 */
3215 	if (verror) {
3216 		*cs->statusp = resp->status = puterrno4(verror);
3217 		goto out;
3218 	}
3219 
3220 	if (vp->v_type != VREG) {
3221 		*cs->statusp = resp->status =
3222 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
3223 		goto out;
3224 	}
3225 
3226 	if (crgetuid(cs->cr) != va.va_uid &&
3227 	    (error = VOP_ACCESS(vp, VREAD, 0, cs->cr, &ct)) &&
3228 	    (error = VOP_ACCESS(vp, VEXEC, 0, cs->cr, &ct))) {
3229 		*cs->statusp = resp->status = puterrno4(error);
3230 		goto out;
3231 	}
3232 
3233 	if (MANDLOCK(vp, va.va_mode)) { /* XXX - V4 supports mand locking */
3234 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3235 		goto out;
3236 	}
3237 
3238 	offset = args->offset;
3239 	if (offset >= va.va_size) {
3240 		*cs->statusp = resp->status = NFS4_OK;
3241 		resp->eof = TRUE;
3242 		resp->data_len = 0;
3243 		resp->data_val = NULL;
3244 		resp->mblk = NULL;
3245 		/* RDMA */
3246 		resp->wlist = args->wlist;
3247 		resp->wlist_len = resp->data_len;
3248 		*cs->statusp = resp->status = NFS4_OK;
3249 		if (resp->wlist)
3250 			clist_zero_len(resp->wlist);
3251 		goto out;
3252 	}
3253 
3254 	if (args->count == 0) {
3255 		*cs->statusp = resp->status = NFS4_OK;
3256 		resp->eof = FALSE;
3257 		resp->data_len = 0;
3258 		resp->data_val = NULL;
3259 		resp->mblk = NULL;
3260 		/* RDMA */
3261 		resp->wlist = args->wlist;
3262 		resp->wlist_len = resp->data_len;
3263 		if (resp->wlist)
3264 			clist_zero_len(resp->wlist);
3265 		goto out;
3266 	}
3267 
3268 	/*
3269 	 * Do not allocate memory more than maximum allowed
3270 	 * transfer size
3271 	 */
3272 	if (args->count > rfs4_tsize(req))
3273 		args->count = rfs4_tsize(req);
3274 
3275 	if (loaned_buffers) {
3276 		uiop = (uio_t *)rfs_setup_xuio(vp);
3277 		ASSERT(uiop != NULL);
3278 		uiop->uio_segflg = UIO_SYSSPACE;
3279 		uiop->uio_loffset = args->offset;
3280 		uiop->uio_resid = args->count;
3281 
3282 		/* Jump to do the read if successful */
3283 		if (!VOP_REQZCBUF(vp, UIO_READ, (xuio_t *)uiop, cs->cr, &ct)) {
3284 			/*
3285 			 * Need to hold the vnode until after VOP_RETZCBUF()
3286 			 * is called.
3287 			 */
3288 			VN_HOLD(vp);
3289 			goto doio_read;
3290 		}
3291 
3292 		DTRACE_PROBE2(nfss__i__reqzcbuf_failed, int,
3293 		    uiop->uio_loffset, int, uiop->uio_resid);
3294 
3295 		uiop->uio_extflg = 0;
3296 
3297 		/* failure to setup for zero copy */
3298 		rfs_free_xuio((void *)uiop);
3299 		loaned_buffers = 0;
3300 	}
3301 
3302 	/*
3303 	 * If returning data via RDMA Write, then grab the chunk list. If we
3304 	 * aren't returning READ data w/RDMA_WRITE, then grab a mblk.
3305 	 */
3306 	if (rdma_used) {
3307 		mp = NULL;
3308 		(void) rdma_get_wchunk(req, &iov, args->wlist);
3309 		uio.uio_iov = &iov;
3310 		uio.uio_iovcnt = 1;
3311 	} else {
3312 		/*
3313 		 * mp will contain the data to be sent out in the read reply.
3314 		 * It will be freed after the reply has been sent.
3315 		 */
3316 		mp = rfs_read_alloc(args->count, &iovp, &iovcnt);
3317 		ASSERT(mp != NULL);
3318 		ASSERT(alloc_err == 0);
3319 		uio.uio_iov = iovp;
3320 		uio.uio_iovcnt = iovcnt;
3321 	}
3322 
3323 	uio.uio_segflg = UIO_SYSSPACE;
3324 	uio.uio_extflg = UIO_COPY_CACHED;
3325 	uio.uio_loffset = args->offset;
3326 	uio.uio_resid = args->count;
3327 	uiop = &uio;
3328 
3329 doio_read:
3330 	error = do_io(FREAD, vp, uiop, 0, cs->cr, &ct);
3331 
3332 	va.va_mask = AT_SIZE;
3333 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3334 
3335 	if (error) {
3336 		if (mp)
3337 			freemsg(mp);
3338 		*cs->statusp = resp->status = puterrno4(error);
3339 		goto out;
3340 	}
3341 
3342 	/* make mblk using zc buffers */
3343 	if (loaned_buffers) {
3344 		mp = uio_to_mblk(uiop);
3345 		ASSERT(mp != NULL);
3346 	}
3347 
3348 	*cs->statusp = resp->status = NFS4_OK;
3349 
3350 	ASSERT(uiop->uio_resid >= 0);
3351 	resp->data_len = args->count - uiop->uio_resid;
3352 	if (mp) {
3353 		resp->data_val = (char *)mp->b_datap->db_base;
3354 		rfs_rndup_mblks(mp, resp->data_len, loaned_buffers);
3355 	} else {
3356 		resp->data_val = (caddr_t)iov.iov_base;
3357 	}
3358 
3359 	resp->mblk = mp;
3360 
3361 	if (!verror && offset + resp->data_len == va.va_size)
3362 		resp->eof = TRUE;
3363 	else
3364 		resp->eof = FALSE;
3365 
3366 	if (rdma_used) {
3367 		if (!rdma_setup_read_data4(args, resp)) {
3368 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3369 		}
3370 	} else {
3371 		resp->wlist = NULL;
3372 	}
3373 
3374 out:
3375 	if (in_crit)
3376 		nbl_end_crit(vp);
3377 
3378 	if (iovp != NULL)
3379 		kmem_free(iovp, iovcnt * sizeof (struct iovec));
3380 
3381 	DTRACE_NFSV4_2(op__read__done, struct compound_state *, cs,
3382 	    READ4res *, resp);
3383 }
3384 
3385 static void
3386 rfs4_op_read_free(nfs_resop4 *resop)
3387 {
3388 	READ4res	*resp = &resop->nfs_resop4_u.opread;
3389 
3390 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3391 		freemsg(resp->mblk);
3392 		resp->mblk = NULL;
3393 		resp->data_val = NULL;
3394 		resp->data_len = 0;
3395 	}
3396 }
3397 
3398 static void
3399 rfs4_op_readdir_free(nfs_resop4 * resop)
3400 {
3401 	READDIR4res    *resp = &resop->nfs_resop4_u.opreaddir;
3402 
3403 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3404 		freeb(resp->mblk);
3405 		resp->mblk = NULL;
3406 		resp->data_len = 0;
3407 	}
3408 }
3409 
3410 
3411 /* ARGSUSED */
3412 static void
3413 rfs4_op_putpubfh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
3414     struct compound_state *cs)
3415 {
3416 	PUTPUBFH4res	*resp = &resop->nfs_resop4_u.opputpubfh;
3417 	int		error;
3418 	vnode_t		*vp;
3419 	struct exportinfo *exi, *sav_exi;
3420 	nfs_fh4_fmt_t	*fh_fmtp;
3421 
3422 	DTRACE_NFSV4_1(op__putpubfh__start, struct compound_state *, cs);
3423 
3424 	if (cs->vp) {
3425 		VN_RELE(cs->vp);
3426 		cs->vp = NULL;
3427 	}
3428 
3429 	if (cs->cr)
3430 		crfree(cs->cr);
3431 
3432 	cs->cr = crdup(cs->basecr);
3433 
3434 	vp = exi_public->exi_vp;
3435 	if (vp == NULL) {
3436 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3437 		goto out;
3438 	}
3439 
3440 	error = makefh4(&cs->fh, vp, exi_public);
3441 	if (error != 0) {
3442 		*cs->statusp = resp->status = puterrno4(error);
3443 		goto out;
3444 	}
3445 	sav_exi = cs->exi;
3446 	if (exi_public == exi_root) {
3447 		/*
3448 		 * No filesystem is actually shared public, so we default
3449 		 * to exi_root. In this case, we must check whether root
3450 		 * is exported.
3451 		 */
3452 		fh_fmtp = (nfs_fh4_fmt_t *)cs->fh.nfs_fh4_val;
3453 
3454 		/*
3455 		 * if root filesystem is exported, the exportinfo struct that we
3456 		 * should use is what checkexport returns, because root_exi is
3457 		 * actually a mostly empty struct.
3458 		 */
3459 		exi = checkexport(&fh_fmtp->fh4_fsid,
3460 		    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
3461 		if (exi) {
3462 			cs->exi = exi;
3463 		} else {
3464 			exi_hold(exi_public);
3465 			cs->exi = exi_public;
3466 		}
3467 	} else {
3468 		/*
3469 		 * it's a properly shared filesystem
3470 		 */
3471 		exi_hold(exi_public);
3472 		cs->exi = exi_public;
3473 	}
3474 
3475 	if (is_system_labeled()) {
3476 		bslabel_t *clabel;
3477 
3478 		ASSERT(req->rq_label != NULL);
3479 		clabel = req->rq_label;
3480 		DTRACE_PROBE2(tx__rfs4__log__info__opputpubfh__clabel, char *,
3481 		    "got client label from request(1)",
3482 		    struct svc_req *, req);
3483 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
3484 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
3485 			    cs->exi)) {
3486 				*cs->statusp = resp->status =
3487 				    NFS4ERR_SERVERFAULT;
3488 				if (sav_exi)
3489 					exi_rele(sav_exi);
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 		exi_rele(cs->exi);
3502 		cs->exi = sav_exi;
3503 		goto out;
3504 	}
3505 	if (sav_exi)
3506 		exi_rele(sav_exi);
3507 
3508 	*cs->statusp = resp->status = NFS4_OK;
3509 out:
3510 	DTRACE_NFSV4_2(op__putpubfh__done, struct compound_state *, cs,
3511 	    PUTPUBFH4res *, resp);
3512 }
3513 
3514 /*
3515  * XXX - issue with put*fh operations. Suppose /export/home is exported.
3516  * Suppose an NFS client goes to mount /export/home/joe. If /export, home,
3517  * or joe have restrictive search permissions, then we shouldn't let
3518  * the client get a file handle. This is easy to enforce. However, we
3519  * don't know what security flavor should be used until we resolve the
3520  * path name. Another complication is uid mapping. If root is
3521  * the user, then it will be mapped to the anonymous user by default,
3522  * but we won't know that till we've resolved the path name. And we won't
3523  * know what the anonymous user is.
3524  * Luckily, SECINFO is specified to take a full filename.
3525  * So what we will have to in rfs4_op_lookup is check that flavor of
3526  * the target object matches that of the request, and if root was the
3527  * caller, check for the root= and anon= options, and if necessary,
3528  * repeat the lookup using the right cred_t. But that's not done yet.
3529  */
3530 /* ARGSUSED */
3531 static void
3532 rfs4_op_putfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3533     struct compound_state *cs)
3534 {
3535 	PUTFH4args *args = &argop->nfs_argop4_u.opputfh;
3536 	PUTFH4res *resp = &resop->nfs_resop4_u.opputfh;
3537 	nfs_fh4_fmt_t *fh_fmtp;
3538 
3539 	DTRACE_NFSV4_2(op__putfh__start, struct compound_state *, cs,
3540 	    PUTFH4args *, args);
3541 
3542 	if (cs->vp) {
3543 		VN_RELE(cs->vp);
3544 		cs->vp = NULL;
3545 	}
3546 
3547 	if (cs->cr) {
3548 		crfree(cs->cr);
3549 		cs->cr = NULL;
3550 	}
3551 
3552 
3553 	if (args->object.nfs_fh4_len < NFS_FH4_LEN) {
3554 		*cs->statusp = resp->status = NFS4ERR_BADHANDLE;
3555 		goto out;
3556 	}
3557 
3558 	fh_fmtp = (nfs_fh4_fmt_t *)args->object.nfs_fh4_val;
3559 	if (cs->exi)
3560 		exi_rele(cs->exi);
3561 	cs->exi = checkexport(&fh_fmtp->fh4_fsid, (fid_t *)&fh_fmtp->fh4_xlen,
3562 	    NULL);
3563 
3564 	if (cs->exi == NULL) {
3565 		*cs->statusp = resp->status = NFS4ERR_STALE;
3566 		goto out;
3567 	}
3568 
3569 	cs->cr = crdup(cs->basecr);
3570 
3571 	ASSERT(cs->cr != NULL);
3572 
3573 	if (! (cs->vp = nfs4_fhtovp(&args->object, cs->exi, &resp->status))) {
3574 		*cs->statusp = resp->status;
3575 		goto out;
3576 	}
3577 
3578 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3579 		VN_RELE(cs->vp);
3580 		cs->vp = NULL;
3581 		goto out;
3582 	}
3583 
3584 	nfs_fh4_copy(&args->object, &cs->fh);
3585 	*cs->statusp = resp->status = NFS4_OK;
3586 	cs->deleg = FALSE;
3587 
3588 out:
3589 	DTRACE_NFSV4_2(op__putfh__done, struct compound_state *, cs,
3590 	    PUTFH4res *, resp);
3591 }
3592 
3593 /* ARGSUSED */
3594 static void
3595 rfs4_op_putrootfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3596     struct compound_state *cs)
3597 {
3598 	PUTROOTFH4res *resp = &resop->nfs_resop4_u.opputrootfh;
3599 	int error;
3600 	fid_t fid;
3601 	struct exportinfo *exi, *sav_exi;
3602 
3603 	DTRACE_NFSV4_1(op__putrootfh__start, struct compound_state *, cs);
3604 
3605 	if (cs->vp) {
3606 		VN_RELE(cs->vp);
3607 		cs->vp = NULL;
3608 	}
3609 
3610 	if (cs->cr)
3611 		crfree(cs->cr);
3612 
3613 	cs->cr = crdup(cs->basecr);
3614 
3615 	/*
3616 	 * Using rootdir, the system root vnode,
3617 	 * get its fid.
3618 	 */
3619 	bzero(&fid, sizeof (fid));
3620 	fid.fid_len = MAXFIDSZ;
3621 	error = vop_fid_pseudo(rootdir, &fid);
3622 	if (error != 0) {
3623 		*cs->statusp = resp->status = puterrno4(error);
3624 		goto out;
3625 	}
3626 
3627 	/*
3628 	 * Then use the root fsid & fid it to find out if it's exported
3629 	 *
3630 	 * If the server root isn't exported directly, then
3631 	 * it should at least be a pseudo export based on
3632 	 * one or more exports further down in the server's
3633 	 * file tree.
3634 	 */
3635 	exi = checkexport(&rootdir->v_vfsp->vfs_fsid, &fid, NULL);
3636 	if (exi == NULL || exi->exi_export.ex_flags & EX_PUBLIC) {
3637 		NFS4_DEBUG(rfs4_debug,
3638 		    (CE_WARN, "rfs4_op_putrootfh: export check failure"));
3639 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3640 		if (exi)
3641 			exi_rele(exi);
3642 		goto out;
3643 	}
3644 
3645 	/*
3646 	 * Now make a filehandle based on the root
3647 	 * export and root vnode.
3648 	 */
3649 	error = makefh4(&cs->fh, rootdir, exi);
3650 	if (error != 0) {
3651 		*cs->statusp = resp->status = puterrno4(error);
3652 		exi_rele(exi);
3653 		goto out;
3654 	}
3655 
3656 	sav_exi = cs->exi;
3657 	cs->exi = exi;
3658 
3659 	VN_HOLD(rootdir);
3660 	cs->vp = rootdir;
3661 
3662 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3663 		VN_RELE(rootdir);
3664 		cs->vp = NULL;
3665 		exi_rele(exi);
3666 		cs->exi = sav_exi;
3667 		goto out;
3668 	}
3669 	if (sav_exi)
3670 		exi_rele(sav_exi);
3671 
3672 	*cs->statusp = resp->status = NFS4_OK;
3673 	cs->deleg = FALSE;
3674 out:
3675 	DTRACE_NFSV4_2(op__putrootfh__done, struct compound_state *, cs,
3676 	    PUTROOTFH4res *, resp);
3677 }
3678 
3679 /*
3680  * set_rdattr_params sets up the variables used to manage what information
3681  * to get for each directory entry.
3682  */
3683 static nfsstat4
3684 set_rdattr_params(struct nfs4_svgetit_arg *sargp,
3685     bitmap4 attrs, bool_t *need_to_lookup)
3686 {
3687 	uint_t	va_mask;
3688 	nfsstat4 status;
3689 	bitmap4 objbits;
3690 
3691 	status = bitmap4_to_attrmask(attrs, sargp);
3692 	if (status != NFS4_OK) {
3693 		/*
3694 		 * could not even figure attr mask
3695 		 */
3696 		return (status);
3697 	}
3698 	va_mask = sargp->vap->va_mask;
3699 
3700 	/*
3701 	 * dirent's d_ino is always correct value for mounted_on_fileid.
3702 	 * mntdfid_set is set once here, but mounted_on_fileid is
3703 	 * set in main dirent processing loop for each dirent.
3704 	 * The mntdfid_set is a simple optimization that lets the
3705 	 * server attr code avoid work when caller is readdir.
3706 	 */
3707 	sargp->mntdfid_set = TRUE;
3708 
3709 	/*
3710 	 * Lookup entry only if client asked for any of the following:
3711 	 * a) vattr attrs
3712 	 * b) vfs attrs
3713 	 * c) attrs w/per-object scope requested (change, filehandle, etc)
3714 	 *    other than mounted_on_fileid (which we can take from dirent)
3715 	 */
3716 	objbits = attrs ? attrs & NFS4_VP_ATTR_MASK : 0;
3717 
3718 	if (va_mask || sargp->sbp || (objbits & ~FATTR4_MOUNTED_ON_FILEID_MASK))
3719 		*need_to_lookup = TRUE;
3720 	else
3721 		*need_to_lookup = FALSE;
3722 
3723 	if (sargp->sbp == NULL)
3724 		return (NFS4_OK);
3725 
3726 	/*
3727 	 * If filesystem attrs are requested, get them now from the
3728 	 * directory vp, as most entries will have same filesystem. The only
3729 	 * exception are mounted over entries but we handle
3730 	 * those as we go (XXX mounted over detection not yet implemented).
3731 	 */
3732 	sargp->vap->va_mask = 0;	/* to avoid VOP_GETATTR */
3733 	status = bitmap4_get_sysattrs(sargp);
3734 	sargp->vap->va_mask = va_mask;
3735 
3736 	if ((status != NFS4_OK) && sargp->rdattr_error_req) {
3737 		/*
3738 		 * Failed to get filesystem attributes.
3739 		 * Return a rdattr_error for each entry, but don't fail.
3740 		 * However, don't get any obj-dependent attrs.
3741 		 */
3742 		sargp->rdattr_error = status;	/* for rdattr_error */
3743 		*need_to_lookup = FALSE;
3744 		/*
3745 		 * At least get fileid for regular readdir output
3746 		 */
3747 		sargp->vap->va_mask &= AT_NODEID;
3748 		status = NFS4_OK;
3749 	}
3750 
3751 	return (status);
3752 }
3753 
3754 /*
3755  * readlink: args: CURRENT_FH.
3756  *	res: status. If success - CURRENT_FH unchanged, return linktext.
3757  */
3758 
3759 /* ARGSUSED */
3760 static void
3761 rfs4_op_readlink(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3762     struct compound_state *cs)
3763 {
3764 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3765 	int error;
3766 	vnode_t *vp;
3767 	struct iovec iov;
3768 	struct vattr va;
3769 	struct uio uio;
3770 	char *data;
3771 	struct sockaddr *ca;
3772 	char *name = NULL;
3773 	int is_referral;
3774 
3775 	DTRACE_NFSV4_1(op__readlink__start, struct compound_state *, cs);
3776 
3777 	/* CURRENT_FH: directory */
3778 	vp = cs->vp;
3779 	if (vp == NULL) {
3780 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3781 		goto out;
3782 	}
3783 
3784 	if (cs->access == CS_ACCESS_DENIED) {
3785 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3786 		goto out;
3787 	}
3788 
3789 	/* Is it a referral? */
3790 	if (vn_is_nfs_reparse(vp, cs->cr) && client_is_downrev(req)) {
3791 
3792 		is_referral = 1;
3793 
3794 	} else {
3795 
3796 		is_referral = 0;
3797 
3798 		if (vp->v_type == VDIR) {
3799 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
3800 			goto out;
3801 		}
3802 
3803 		if (vp->v_type != VLNK) {
3804 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3805 			goto out;
3806 		}
3807 
3808 	}
3809 
3810 	va.va_mask = AT_MODE;
3811 	error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
3812 	if (error) {
3813 		*cs->statusp = resp->status = puterrno4(error);
3814 		goto out;
3815 	}
3816 
3817 	if (MANDLOCK(vp, va.va_mode)) {
3818 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3819 		goto out;
3820 	}
3821 
3822 	data = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);
3823 
3824 	if (is_referral) {
3825 		char *s;
3826 		size_t strsz;
3827 
3828 		/* Get an artificial symlink based on a referral */
3829 		s = build_symlink(vp, cs->cr, &strsz);
3830 		global_svstat_ptr[4][NFS_REFERLINKS].value.ui64++;
3831 		DTRACE_PROBE2(nfs4serv__func__referral__reflink,
3832 		    vnode_t *, vp, char *, s);
3833 		if (s == NULL)
3834 			error = EINVAL;
3835 		else {
3836 			error = 0;
3837 			(void) strlcpy(data, s, MAXPATHLEN + 1);
3838 			kmem_free(s, strsz);
3839 		}
3840 
3841 	} else {
3842 
3843 		iov.iov_base = data;
3844 		iov.iov_len = MAXPATHLEN;
3845 		uio.uio_iov = &iov;
3846 		uio.uio_iovcnt = 1;
3847 		uio.uio_segflg = UIO_SYSSPACE;
3848 		uio.uio_extflg = UIO_COPY_CACHED;
3849 		uio.uio_loffset = 0;
3850 		uio.uio_resid = MAXPATHLEN;
3851 
3852 		error = VOP_READLINK(vp, &uio, cs->cr, NULL);
3853 
3854 		if (!error)
3855 			*(data + MAXPATHLEN - uio.uio_resid) = '\0';
3856 	}
3857 
3858 	if (error) {
3859 		kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3860 		*cs->statusp = resp->status = puterrno4(error);
3861 		goto out;
3862 	}
3863 
3864 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
3865 	name = nfscmd_convname(ca, cs->exi, data, NFSCMD_CONV_OUTBOUND,
3866 	    MAXPATHLEN  + 1);
3867 
3868 	if (name == NULL) {
3869 		/*
3870 		 * Even though the conversion failed, we return
3871 		 * something. We just don't translate it.
3872 		 */
3873 		name = data;
3874 	}
3875 
3876 	/*
3877 	 * treat link name as data
3878 	 */
3879 	(void) str_to_utf8(name, (utf8string *)&resp->link);
3880 
3881 	if (name != data)
3882 		kmem_free(name, MAXPATHLEN + 1);
3883 	kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3884 	*cs->statusp = resp->status = NFS4_OK;
3885 
3886 out:
3887 	DTRACE_NFSV4_2(op__readlink__done, struct compound_state *, cs,
3888 	    READLINK4res *, resp);
3889 }
3890 
3891 static void
3892 rfs4_op_readlink_free(nfs_resop4 *resop)
3893 {
3894 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3895 	utf8string *symlink = (utf8string *)&resp->link;
3896 
3897 	if (symlink->utf8string_val) {
3898 		UTF8STRING_FREE(*symlink)
3899 	}
3900 }
3901 
3902 /*
3903  * release_lockowner:
3904  *	Release any state associated with the supplied
3905  *	lockowner. Note if any lo_state is holding locks we will not
3906  *	rele that lo_state and thus the lockowner will not be destroyed.
3907  *	A client using lock after the lock owner stateid has been released
3908  *	will suffer the consequence of NFS4ERR_BAD_STATEID and would have
3909  *	to reissue the lock with new_lock_owner set to TRUE.
3910  *	args: lock_owner
3911  *	res:  status
3912  */
3913 /* ARGSUSED */
3914 static void
3915 rfs4_op_release_lockowner(nfs_argop4 *argop, nfs_resop4 *resop,
3916     struct svc_req *req, struct compound_state *cs)
3917 {
3918 	RELEASE_LOCKOWNER4args *ap = &argop->nfs_argop4_u.oprelease_lockowner;
3919 	RELEASE_LOCKOWNER4res *resp = &resop->nfs_resop4_u.oprelease_lockowner;
3920 	rfs4_lockowner_t *lo;
3921 	rfs4_openowner_t *oo;
3922 	rfs4_state_t *sp;
3923 	rfs4_lo_state_t *lsp;
3924 	rfs4_client_t *cp;
3925 	bool_t create = FALSE;
3926 	locklist_t *llist;
3927 	sysid_t sysid;
3928 
3929 	DTRACE_NFSV4_2(op__release__lockowner__start, struct compound_state *,
3930 	    cs, RELEASE_LOCKOWNER4args *, ap);
3931 
3932 	/* Make sure there is a clientid around for this request */
3933 	cp = rfs4_findclient_by_id(ap->lock_owner.clientid, FALSE);
3934 
3935 	if (cp == NULL) {
3936 		*cs->statusp = resp->status =
3937 		    rfs4_check_clientid(&ap->lock_owner.clientid, 0);
3938 		goto out;
3939 	}
3940 	rfs4_client_rele(cp);
3941 
3942 	lo = rfs4_findlockowner(&ap->lock_owner, &create);
3943 	if (lo == NULL) {
3944 		*cs->statusp = resp->status = NFS4_OK;
3945 		goto out;
3946 	}
3947 	ASSERT(lo->rl_client != NULL);
3948 
3949 	/*
3950 	 * Check for EXPIRED client. If so will reap state with in a lease
3951 	 * period or on next set_clientid_confirm step
3952 	 */
3953 	if (rfs4_lease_expired(lo->rl_client)) {
3954 		rfs4_lockowner_rele(lo);
3955 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
3956 		goto out;
3957 	}
3958 
3959 	/*
3960 	 * If no sysid has been assigned, then no locks exist; just return.
3961 	 */
3962 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3963 	if (lo->rl_client->rc_sysidt == LM_NOSYSID) {
3964 		rfs4_lockowner_rele(lo);
3965 		rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3966 		goto out;
3967 	}
3968 
3969 	sysid = lo->rl_client->rc_sysidt;
3970 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3971 
3972 	/*
3973 	 * Mark the lockowner invalid.
3974 	 */
3975 	rfs4_dbe_hide(lo->rl_dbe);
3976 
3977 	/*
3978 	 * sysid-pid pair should now not be used since the lockowner is
3979 	 * invalid. If the client were to instantiate the lockowner again
3980 	 * it would be assigned a new pid. Thus we can get the list of
3981 	 * current locks.
3982 	 */
3983 
3984 	llist = flk_get_active_locks(sysid, lo->rl_pid);
3985 	/* If we are still holding locks fail */
3986 	if (llist != NULL) {
3987 
3988 		*cs->statusp = resp->status = NFS4ERR_LOCKS_HELD;
3989 
3990 		flk_free_locklist(llist);
3991 		/*
3992 		 * We need to unhide the lockowner so the client can
3993 		 * try it again. The bad thing here is if the client
3994 		 * has a logic error that took it here in the first place
3995 		 * he probably has lost accounting of the locks that it
3996 		 * is holding. So we may have dangling state until the
3997 		 * open owner state is reaped via close. One scenario
3998 		 * that could possibly occur is that the client has
3999 		 * sent the unlock request(s) in separate threads
4000 		 * and has not waited for the replies before sending the
4001 		 * RELEASE_LOCKOWNER request. Presumably, it would expect
4002 		 * and deal appropriately with NFS4ERR_LOCKS_HELD, by
4003 		 * reissuing the request.
4004 		 */
4005 		rfs4_dbe_unhide(lo->rl_dbe);
4006 		rfs4_lockowner_rele(lo);
4007 		goto out;
4008 	}
4009 
4010 	/*
4011 	 * For the corresponding client we need to check each open
4012 	 * owner for any opens that have lockowner state associated
4013 	 * with this lockowner.
4014 	 */
4015 
4016 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
4017 	for (oo = list_head(&lo->rl_client->rc_openownerlist); oo != NULL;
4018 	    oo = list_next(&lo->rl_client->rc_openownerlist, oo)) {
4019 
4020 		rfs4_dbe_lock(oo->ro_dbe);
4021 		for (sp = list_head(&oo->ro_statelist); sp != NULL;
4022 		    sp = list_next(&oo->ro_statelist, sp)) {
4023 
4024 			rfs4_dbe_lock(sp->rs_dbe);
4025 			for (lsp = list_head(&sp->rs_lostatelist);
4026 			    lsp != NULL;
4027 			    lsp = list_next(&sp->rs_lostatelist, lsp)) {
4028 				if (lsp->rls_locker == lo) {
4029 					rfs4_dbe_lock(lsp->rls_dbe);
4030 					rfs4_dbe_invalidate(lsp->rls_dbe);
4031 					rfs4_dbe_unlock(lsp->rls_dbe);
4032 				}
4033 			}
4034 			rfs4_dbe_unlock(sp->rs_dbe);
4035 		}
4036 		rfs4_dbe_unlock(oo->ro_dbe);
4037 	}
4038 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
4039 
4040 	rfs4_lockowner_rele(lo);
4041 
4042 	*cs->statusp = resp->status = NFS4_OK;
4043 
4044 out:
4045 	DTRACE_NFSV4_2(op__release__lockowner__done, struct compound_state *,
4046 	    cs, RELEASE_LOCKOWNER4res *, resp);
4047 }
4048 
4049 /*
4050  * short utility function to lookup a file and recall the delegation
4051  */
4052 static rfs4_file_t *
4053 rfs4_lookup_and_findfile(vnode_t *dvp, char *nm, vnode_t **vpp,
4054     int *lkup_error, cred_t *cr)
4055 {
4056 	vnode_t *vp;
4057 	rfs4_file_t *fp = NULL;
4058 	bool_t fcreate = FALSE;
4059 	int error;
4060 
4061 	if (vpp)
4062 		*vpp = NULL;
4063 
4064 	if ((error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cr, NULL, NULL,
4065 	    NULL)) == 0) {
4066 		if (vp->v_type == VREG)
4067 			fp = rfs4_findfile(vp, NULL, &fcreate);
4068 		if (vpp)
4069 			*vpp = vp;
4070 		else
4071 			VN_RELE(vp);
4072 	}
4073 
4074 	if (lkup_error)
4075 		*lkup_error = error;
4076 
4077 	return (fp);
4078 }
4079 
4080 /*
4081  * remove: args: CURRENT_FH: directory; name.
4082  *	res: status. If success - CURRENT_FH unchanged, return change_info
4083  *		for directory.
4084  */
4085 /* ARGSUSED */
4086 static void
4087 rfs4_op_remove(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4088     struct compound_state *cs)
4089 {
4090 	REMOVE4args *args = &argop->nfs_argop4_u.opremove;
4091 	REMOVE4res *resp = &resop->nfs_resop4_u.opremove;
4092 	int error;
4093 	vnode_t *dvp, *vp;
4094 	struct vattr bdva, idva, adva;
4095 	char *nm;
4096 	uint_t len;
4097 	rfs4_file_t *fp;
4098 	int in_crit = 0;
4099 	bslabel_t *clabel;
4100 	struct sockaddr *ca;
4101 	char *name = NULL;
4102 	nfsstat4 status;
4103 
4104 	DTRACE_NFSV4_2(op__remove__start, struct compound_state *, cs,
4105 	    REMOVE4args *, args);
4106 
4107 	/* CURRENT_FH: directory */
4108 	dvp = cs->vp;
4109 	if (dvp == NULL) {
4110 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4111 		goto out;
4112 	}
4113 
4114 	if (cs->access == CS_ACCESS_DENIED) {
4115 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4116 		goto out;
4117 	}
4118 
4119 	/*
4120 	 * If there is an unshared filesystem mounted on this vnode,
4121 	 * Do not allow to remove anything in this directory.
4122 	 */
4123 	if (vn_ismntpt(dvp)) {
4124 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4125 		goto out;
4126 	}
4127 
4128 	if (dvp->v_type != VDIR) {
4129 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4130 		goto out;
4131 	}
4132 
4133 	status = utf8_dir_verify(&args->target);
4134 	if (status != NFS4_OK) {
4135 		*cs->statusp = resp->status = status;
4136 		goto out;
4137 	}
4138 
4139 	/*
4140 	 * Lookup the file so that we can check if it's a directory
4141 	 */
4142 	nm = utf8_to_fn(&args->target, &len, NULL);
4143 	if (nm == NULL) {
4144 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4145 		goto out;
4146 	}
4147 
4148 	if (len > MAXNAMELEN) {
4149 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4150 		kmem_free(nm, len);
4151 		goto out;
4152 	}
4153 
4154 	if (rdonly4(req, cs)) {
4155 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4156 		kmem_free(nm, len);
4157 		goto out;
4158 	}
4159 
4160 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4161 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
4162 	    MAXPATHLEN  + 1);
4163 
4164 	if (name == NULL) {
4165 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4166 		kmem_free(nm, len);
4167 		goto out;
4168 	}
4169 
4170 	/*
4171 	 * Lookup the file to determine type and while we are see if
4172 	 * there is a file struct around and check for delegation.
4173 	 * We don't need to acquire va_seq before this lookup, if
4174 	 * it causes an update, cinfo.before will not match, which will
4175 	 * trigger a cache flush even if atomic is TRUE.
4176 	 */
4177 	if (fp = rfs4_lookup_and_findfile(dvp, name, &vp, &error, cs->cr)) {
4178 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4179 		    NULL)) {
4180 			VN_RELE(vp);
4181 			rfs4_file_rele(fp);
4182 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4183 			if (nm != name)
4184 				kmem_free(name, MAXPATHLEN + 1);
4185 			kmem_free(nm, len);
4186 			goto out;
4187 		}
4188 	}
4189 
4190 	/* Didn't find anything to remove */
4191 	if (vp == NULL) {
4192 		*cs->statusp = resp->status = error;
4193 		if (nm != name)
4194 			kmem_free(name, MAXPATHLEN + 1);
4195 		kmem_free(nm, len);
4196 		goto out;
4197 	}
4198 
4199 	if (nbl_need_check(vp)) {
4200 		nbl_start_crit(vp, RW_READER);
4201 		in_crit = 1;
4202 		if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
4203 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4204 			if (nm != name)
4205 				kmem_free(name, MAXPATHLEN + 1);
4206 			kmem_free(nm, len);
4207 			nbl_end_crit(vp);
4208 			VN_RELE(vp);
4209 			if (fp) {
4210 				rfs4_clear_dont_grant(fp);
4211 				rfs4_file_rele(fp);
4212 			}
4213 			goto out;
4214 		}
4215 	}
4216 
4217 	/* check label before allowing removal */
4218 	if (is_system_labeled()) {
4219 		ASSERT(req->rq_label != NULL);
4220 		clabel = req->rq_label;
4221 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
4222 		    "got client label from request(1)",
4223 		    struct svc_req *, req);
4224 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4225 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
4226 			    cs->exi)) {
4227 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4228 				if (name != nm)
4229 					kmem_free(name, MAXPATHLEN + 1);
4230 				kmem_free(nm, len);
4231 				if (in_crit)
4232 					nbl_end_crit(vp);
4233 				VN_RELE(vp);
4234 				if (fp) {
4235 					rfs4_clear_dont_grant(fp);
4236 					rfs4_file_rele(fp);
4237 				}
4238 				goto out;
4239 			}
4240 		}
4241 	}
4242 
4243 	/* Get dir "before" change value */
4244 	bdva.va_mask = AT_CTIME|AT_SEQ;
4245 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
4246 	if (error) {
4247 		*cs->statusp = resp->status = puterrno4(error);
4248 		if (nm != name)
4249 			kmem_free(name, MAXPATHLEN + 1);
4250 		kmem_free(nm, len);
4251 		if (in_crit)
4252 			nbl_end_crit(vp);
4253 		VN_RELE(vp);
4254 		if (fp) {
4255 			rfs4_clear_dont_grant(fp);
4256 			rfs4_file_rele(fp);
4257 		}
4258 		goto out;
4259 	}
4260 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
4261 
4262 	/* Actually do the REMOVE operation */
4263 	if (vp->v_type == VDIR) {
4264 		/*
4265 		 * Can't remove a directory that has a mounted-on filesystem.
4266 		 */
4267 		if (vn_ismntpt(vp)) {
4268 			error = EACCES;
4269 		} else {
4270 			/*
4271 			 * System V defines rmdir to return EEXIST,
4272 			 * not ENOTEMPTY, if the directory is not
4273 			 * empty.  A System V NFS server needs to map
4274 			 * NFS4ERR_EXIST to NFS4ERR_NOTEMPTY to
4275 			 * transmit over the wire.
4276 			 */
4277 			if ((error = VOP_RMDIR(dvp, name, rootdir, cs->cr,
4278 			    NULL, 0)) == EEXIST)
4279 				error = ENOTEMPTY;
4280 		}
4281 
4282 		if (in_crit)
4283 			nbl_end_crit(vp);
4284 		VN_RELE(vp);
4285 	} else {
4286 		if (!in_crit)
4287 			VN_RELE(vp);
4288 		if ((error = VOP_REMOVE(dvp, name, cs->cr, NULL, 0)) == 0 &&
4289 		    fp != NULL) {
4290 			struct vattr va;
4291 			vnode_t *tvp;
4292 
4293 			rfs4_dbe_lock(fp->rf_dbe);
4294 			tvp = fp->rf_vp;
4295 			if (tvp)
4296 				VN_HOLD(tvp);
4297 			rfs4_dbe_unlock(fp->rf_dbe);
4298 
4299 			if (tvp) {
4300 				/*
4301 				 * This is va_seq safe because we are not
4302 				 * manipulating dvp.
4303 				 */
4304 				va.va_mask = AT_NLINK;
4305 				if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4306 				    va.va_nlink == 0) {
4307 					/* Remove state on file remove */
4308 					if (in_crit) {
4309 						nbl_end_crit(vp);
4310 						VN_RELE(vp);
4311 						in_crit = 0;
4312 					}
4313 					rfs4_close_all_state(fp);
4314 				}
4315 				VN_RELE(tvp);
4316 			}
4317 		}
4318 
4319 		if (in_crit) {
4320 			nbl_end_crit(vp);
4321 			VN_RELE(vp);
4322 		}
4323 	}
4324 
4325 	if (fp) {
4326 		rfs4_clear_dont_grant(fp);
4327 		rfs4_file_rele(fp);
4328 	}
4329 	if (nm != name)
4330 		kmem_free(name, MAXPATHLEN + 1);
4331 	kmem_free(nm, len);
4332 
4333 	if (error) {
4334 		*cs->statusp = resp->status = puterrno4(error);
4335 		goto out;
4336 	}
4337 
4338 	/*
4339 	 * Get the initial "after" sequence number, if it fails, set to zero
4340 	 */
4341 	idva.va_mask = AT_SEQ;
4342 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
4343 		idva.va_seq = 0;
4344 
4345 	/*
4346 	 * Force modified data and metadata out to stable storage.
4347 	 */
4348 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
4349 
4350 	/*
4351 	 * Get "after" change value, if it fails, simply return the
4352 	 * before value.
4353 	 */
4354 	adva.va_mask = AT_CTIME|AT_SEQ;
4355 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
4356 		adva.va_ctime = bdva.va_ctime;
4357 		adva.va_seq = 0;
4358 	}
4359 
4360 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
4361 
4362 	/*
4363 	 * The cinfo.atomic = TRUE only if we have
4364 	 * non-zero va_seq's, and it has incremented by exactly one
4365 	 * during the VOP_REMOVE/RMDIR and it didn't change during
4366 	 * the VOP_FSYNC.
4367 	 */
4368 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
4369 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
4370 		resp->cinfo.atomic = TRUE;
4371 	else
4372 		resp->cinfo.atomic = FALSE;
4373 
4374 	*cs->statusp = resp->status = NFS4_OK;
4375 
4376 out:
4377 	DTRACE_NFSV4_2(op__remove__done, struct compound_state *, cs,
4378 	    REMOVE4res *, resp);
4379 }
4380 
4381 /*
4382  * rename: args: SAVED_FH: from directory, CURRENT_FH: target directory,
4383  *		oldname and newname.
4384  *	res: status. If success - CURRENT_FH unchanged, return change_info
4385  *		for both from and target directories.
4386  */
4387 /* ARGSUSED */
4388 static void
4389 rfs4_op_rename(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4390     struct compound_state *cs)
4391 {
4392 	RENAME4args *args = &argop->nfs_argop4_u.oprename;
4393 	RENAME4res *resp = &resop->nfs_resop4_u.oprename;
4394 	int error;
4395 	vnode_t *odvp;
4396 	vnode_t *ndvp;
4397 	vnode_t *srcvp, *targvp;
4398 	struct vattr obdva, oidva, oadva;
4399 	struct vattr nbdva, nidva, nadva;
4400 	char *onm, *nnm;
4401 	uint_t olen, nlen;
4402 	rfs4_file_t *fp, *sfp;
4403 	int in_crit_src, in_crit_targ;
4404 	int fp_rele_grant_hold, sfp_rele_grant_hold;
4405 	bslabel_t *clabel;
4406 	struct sockaddr *ca;
4407 	char *converted_onm = NULL;
4408 	char *converted_nnm = NULL;
4409 	nfsstat4 status;
4410 
4411 	DTRACE_NFSV4_2(op__rename__start, struct compound_state *, cs,
4412 	    RENAME4args *, args);
4413 
4414 	fp = sfp = NULL;
4415 	srcvp = targvp = NULL;
4416 	in_crit_src = in_crit_targ = 0;
4417 	fp_rele_grant_hold = sfp_rele_grant_hold = 0;
4418 
4419 	/* CURRENT_FH: target directory */
4420 	ndvp = cs->vp;
4421 	if (ndvp == NULL) {
4422 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4423 		goto out;
4424 	}
4425 
4426 	/* SAVED_FH: from directory */
4427 	odvp = cs->saved_vp;
4428 	if (odvp == NULL) {
4429 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4430 		goto out;
4431 	}
4432 
4433 	if (cs->access == CS_ACCESS_DENIED) {
4434 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4435 		goto out;
4436 	}
4437 
4438 	/*
4439 	 * If there is an unshared filesystem mounted on this vnode,
4440 	 * do not allow to rename objects in this directory.
4441 	 */
4442 	if (vn_ismntpt(odvp)) {
4443 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4444 		goto out;
4445 	}
4446 
4447 	/*
4448 	 * If there is an unshared filesystem mounted on this vnode,
4449 	 * do not allow to rename to this directory.
4450 	 */
4451 	if (vn_ismntpt(ndvp)) {
4452 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4453 		goto out;
4454 	}
4455 
4456 	if (odvp->v_type != VDIR || ndvp->v_type != VDIR) {
4457 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4458 		goto out;
4459 	}
4460 
4461 	if (cs->saved_exi != cs->exi) {
4462 		*cs->statusp = resp->status = NFS4ERR_XDEV;
4463 		goto out;
4464 	}
4465 
4466 	status = utf8_dir_verify(&args->oldname);
4467 	if (status != NFS4_OK) {
4468 		*cs->statusp = resp->status = status;
4469 		goto out;
4470 	}
4471 
4472 	status = utf8_dir_verify(&args->newname);
4473 	if (status != NFS4_OK) {
4474 		*cs->statusp = resp->status = status;
4475 		goto out;
4476 	}
4477 
4478 	onm = utf8_to_fn(&args->oldname, &olen, NULL);
4479 	if (onm == NULL) {
4480 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4481 		goto out;
4482 	}
4483 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4484 	nlen = MAXPATHLEN + 1;
4485 	converted_onm = nfscmd_convname(ca, cs->exi, onm, NFSCMD_CONV_INBOUND,
4486 	    nlen);
4487 
4488 	if (converted_onm == NULL) {
4489 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4490 		kmem_free(onm, olen);
4491 		goto out;
4492 	}
4493 
4494 	nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4495 	if (nnm == NULL) {
4496 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4497 		if (onm != converted_onm)
4498 			kmem_free(converted_onm, MAXPATHLEN + 1);
4499 		kmem_free(onm, olen);
4500 		goto out;
4501 	}
4502 	converted_nnm = nfscmd_convname(ca, cs->exi, nnm, NFSCMD_CONV_INBOUND,
4503 	    MAXPATHLEN  + 1);
4504 
4505 	if (converted_nnm == NULL) {
4506 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4507 		kmem_free(nnm, nlen);
4508 		nnm = NULL;
4509 		if (onm != converted_onm)
4510 			kmem_free(converted_onm, MAXPATHLEN + 1);
4511 		kmem_free(onm, olen);
4512 		goto out;
4513 	}
4514 
4515 
4516 	if (olen > MAXNAMELEN || nlen > MAXNAMELEN) {
4517 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4518 		kmem_free(onm, olen);
4519 		kmem_free(nnm, nlen);
4520 		goto out;
4521 	}
4522 
4523 
4524 	if (rdonly4(req, cs)) {
4525 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4526 		if (onm != converted_onm)
4527 			kmem_free(converted_onm, MAXPATHLEN + 1);
4528 		kmem_free(onm, olen);
4529 		if (nnm != converted_nnm)
4530 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4531 		kmem_free(nnm, nlen);
4532 		goto out;
4533 	}
4534 
4535 	/* check label of the target dir */
4536 	if (is_system_labeled()) {
4537 		ASSERT(req->rq_label != NULL);
4538 		clabel = req->rq_label;
4539 		DTRACE_PROBE2(tx__rfs4__log__info__oprename__clabel, char *,
4540 		    "got client label from request(1)",
4541 		    struct svc_req *, req);
4542 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4543 			if (!do_rfs_label_check(clabel, ndvp,
4544 			    EQUALITY_CHECK, cs->exi)) {
4545 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4546 				goto err_out;
4547 			}
4548 		}
4549 	}
4550 
4551 	/*
4552 	 * Is the source a file and have a delegation?
4553 	 * We don't need to acquire va_seq before these lookups, if
4554 	 * it causes an update, cinfo.before will not match, which will
4555 	 * trigger a cache flush even if atomic is TRUE.
4556 	 */
4557 	if (sfp = rfs4_lookup_and_findfile(odvp, converted_onm, &srcvp,
4558 	    &error, cs->cr)) {
4559 		if (rfs4_check_delegated_byfp(FWRITE, sfp, TRUE, TRUE, TRUE,
4560 		    NULL)) {
4561 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4562 			goto err_out;
4563 		}
4564 	}
4565 
4566 	if (srcvp == NULL) {
4567 		*cs->statusp = resp->status = puterrno4(error);
4568 		if (onm != converted_onm)
4569 			kmem_free(converted_onm, MAXPATHLEN + 1);
4570 		kmem_free(onm, olen);
4571 		if (nnm != converted_nnm)
4572 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4573 		kmem_free(nnm, nlen);
4574 		goto out;
4575 	}
4576 
4577 	sfp_rele_grant_hold = 1;
4578 
4579 	/* Does the destination exist and a file and have a delegation? */
4580 	if (fp = rfs4_lookup_and_findfile(ndvp, converted_nnm, &targvp,
4581 	    NULL, cs->cr)) {
4582 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4583 		    NULL)) {
4584 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4585 			goto err_out;
4586 		}
4587 	}
4588 	fp_rele_grant_hold = 1;
4589 
4590 
4591 	/* Check for NBMAND lock on both source and target */
4592 	if (nbl_need_check(srcvp)) {
4593 		nbl_start_crit(srcvp, RW_READER);
4594 		in_crit_src = 1;
4595 		if (nbl_conflict(srcvp, NBL_RENAME, 0, 0, 0, NULL)) {
4596 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4597 			goto err_out;
4598 		}
4599 	}
4600 
4601 	if (targvp && nbl_need_check(targvp)) {
4602 		nbl_start_crit(targvp, RW_READER);
4603 		in_crit_targ = 1;
4604 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
4605 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4606 			goto err_out;
4607 		}
4608 	}
4609 
4610 	/* Get source "before" change value */
4611 	obdva.va_mask = AT_CTIME|AT_SEQ;
4612 	error = VOP_GETATTR(odvp, &obdva, 0, cs->cr, NULL);
4613 	if (!error) {
4614 		nbdva.va_mask = AT_CTIME|AT_SEQ;
4615 		error = VOP_GETATTR(ndvp, &nbdva, 0, cs->cr, NULL);
4616 	}
4617 	if (error) {
4618 		*cs->statusp = resp->status = puterrno4(error);
4619 		goto err_out;
4620 	}
4621 
4622 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.before, obdva.va_ctime)
4623 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.before, nbdva.va_ctime)
4624 
4625 	if ((error = VOP_RENAME(odvp, converted_onm, ndvp, converted_nnm,
4626 	    cs->cr, NULL, 0)) == 0 && fp != NULL) {
4627 		struct vattr va;
4628 		vnode_t *tvp;
4629 
4630 		rfs4_dbe_lock(fp->rf_dbe);
4631 		tvp = fp->rf_vp;
4632 		if (tvp)
4633 			VN_HOLD(tvp);
4634 		rfs4_dbe_unlock(fp->rf_dbe);
4635 
4636 		if (tvp) {
4637 			va.va_mask = AT_NLINK;
4638 			if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4639 			    va.va_nlink == 0) {
4640 				/* The file is gone and so should the state */
4641 				if (in_crit_targ) {
4642 					nbl_end_crit(targvp);
4643 					in_crit_targ = 0;
4644 				}
4645 				rfs4_close_all_state(fp);
4646 			}
4647 			VN_RELE(tvp);
4648 		}
4649 	}
4650 	if (error == 0)
4651 		vn_renamepath(ndvp, srcvp, nnm, nlen - 1);
4652 
4653 	if (in_crit_src)
4654 		nbl_end_crit(srcvp);
4655 	if (srcvp)
4656 		VN_RELE(srcvp);
4657 	if (in_crit_targ)
4658 		nbl_end_crit(targvp);
4659 	if (targvp)
4660 		VN_RELE(targvp);
4661 
4662 	if (sfp) {
4663 		rfs4_clear_dont_grant(sfp);
4664 		rfs4_file_rele(sfp);
4665 	}
4666 	if (fp) {
4667 		rfs4_clear_dont_grant(fp);
4668 		rfs4_file_rele(fp);
4669 	}
4670 
4671 	if (converted_onm != onm)
4672 		kmem_free(converted_onm, MAXPATHLEN + 1);
4673 	kmem_free(onm, olen);
4674 	if (converted_nnm != nnm)
4675 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4676 	kmem_free(nnm, nlen);
4677 
4678 	/*
4679 	 * Get the initial "after" sequence number, if it fails, set to zero
4680 	 */
4681 	oidva.va_mask = AT_SEQ;
4682 	if (VOP_GETATTR(odvp, &oidva, 0, cs->cr, NULL))
4683 		oidva.va_seq = 0;
4684 
4685 	nidva.va_mask = AT_SEQ;
4686 	if (VOP_GETATTR(ndvp, &nidva, 0, cs->cr, NULL))
4687 		nidva.va_seq = 0;
4688 
4689 	/*
4690 	 * Force modified data and metadata out to stable storage.
4691 	 */
4692 	(void) VOP_FSYNC(odvp, 0, cs->cr, NULL);
4693 	(void) VOP_FSYNC(ndvp, 0, cs->cr, NULL);
4694 
4695 	if (error) {
4696 		*cs->statusp = resp->status = puterrno4(error);
4697 		goto out;
4698 	}
4699 
4700 	/*
4701 	 * Get "after" change values, if it fails, simply return the
4702 	 * before value.
4703 	 */
4704 	oadva.va_mask = AT_CTIME|AT_SEQ;
4705 	if (VOP_GETATTR(odvp, &oadva, 0, cs->cr, NULL)) {
4706 		oadva.va_ctime = obdva.va_ctime;
4707 		oadva.va_seq = 0;
4708 	}
4709 
4710 	nadva.va_mask = AT_CTIME|AT_SEQ;
4711 	if (VOP_GETATTR(odvp, &nadva, 0, cs->cr, NULL)) {
4712 		nadva.va_ctime = nbdva.va_ctime;
4713 		nadva.va_seq = 0;
4714 	}
4715 
4716 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.after, oadva.va_ctime)
4717 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.after, nadva.va_ctime)
4718 
4719 	/*
4720 	 * The cinfo.atomic = TRUE only if we have
4721 	 * non-zero va_seq's, and it has incremented by exactly one
4722 	 * during the VOP_RENAME and it didn't change during the VOP_FSYNC.
4723 	 */
4724 	if (obdva.va_seq && oidva.va_seq && oadva.va_seq &&
4725 	    oidva.va_seq == (obdva.va_seq + 1) && oidva.va_seq == oadva.va_seq)
4726 		resp->source_cinfo.atomic = TRUE;
4727 	else
4728 		resp->source_cinfo.atomic = FALSE;
4729 
4730 	if (nbdva.va_seq && nidva.va_seq && nadva.va_seq &&
4731 	    nidva.va_seq == (nbdva.va_seq + 1) && nidva.va_seq == nadva.va_seq)
4732 		resp->target_cinfo.atomic = TRUE;
4733 	else
4734 		resp->target_cinfo.atomic = FALSE;
4735 
4736 #ifdef	VOLATILE_FH_TEST
4737 	{
4738 	extern void add_volrnm_fh(struct exportinfo *, vnode_t *);
4739 
4740 	/*
4741 	 * Add the renamed file handle to the volatile rename list
4742 	 */
4743 	if (cs->exi->exi_export.ex_flags & EX_VOLRNM) {
4744 		/* file handles may expire on rename */
4745 		vnode_t *vp;
4746 
4747 		nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4748 		/*
4749 		 * Already know that nnm will be a valid string
4750 		 */
4751 		error = VOP_LOOKUP(ndvp, nnm, &vp, NULL, 0, NULL, cs->cr,
4752 		    NULL, NULL, NULL);
4753 		kmem_free(nnm, nlen);
4754 		if (!error) {
4755 			add_volrnm_fh(cs->exi, vp);
4756 			VN_RELE(vp);
4757 		}
4758 	}
4759 	}
4760 #endif	/* VOLATILE_FH_TEST */
4761 
4762 	*cs->statusp = resp->status = NFS4_OK;
4763 out:
4764 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4765 	    RENAME4res *, resp);
4766 	return;
4767 
4768 err_out:
4769 	if (onm != converted_onm)
4770 		kmem_free(converted_onm, MAXPATHLEN + 1);
4771 	if (onm != NULL)
4772 		kmem_free(onm, olen);
4773 	if (nnm != converted_nnm)
4774 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4775 	if (nnm != NULL)
4776 		kmem_free(nnm, nlen);
4777 
4778 	if (in_crit_src) nbl_end_crit(srcvp);
4779 	if (in_crit_targ) nbl_end_crit(targvp);
4780 	if (targvp) VN_RELE(targvp);
4781 	if (srcvp) VN_RELE(srcvp);
4782 	if (sfp) {
4783 		if (sfp_rele_grant_hold) rfs4_clear_dont_grant(sfp);
4784 		rfs4_file_rele(sfp);
4785 	}
4786 	if (fp) {
4787 		if (fp_rele_grant_hold) rfs4_clear_dont_grant(fp);
4788 		rfs4_file_rele(fp);
4789 	}
4790 
4791 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4792 	    RENAME4res *, resp);
4793 }
4794 
4795 /* ARGSUSED */
4796 static void
4797 rfs4_op_renew(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4798     struct compound_state *cs)
4799 {
4800 	RENEW4args *args = &argop->nfs_argop4_u.oprenew;
4801 	RENEW4res *resp = &resop->nfs_resop4_u.oprenew;
4802 	rfs4_client_t *cp;
4803 
4804 	DTRACE_NFSV4_2(op__renew__start, struct compound_state *, cs,
4805 	    RENEW4args *, args);
4806 
4807 	if ((cp = rfs4_findclient_by_id(args->clientid, FALSE)) == NULL) {
4808 		*cs->statusp = resp->status =
4809 		    rfs4_check_clientid(&args->clientid, 0);
4810 		goto out;
4811 	}
4812 
4813 	if (rfs4_lease_expired(cp)) {
4814 		rfs4_client_rele(cp);
4815 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
4816 		goto out;
4817 	}
4818 
4819 	rfs4_update_lease(cp);
4820 
4821 	mutex_enter(cp->rc_cbinfo.cb_lock);
4822 	if (cp->rc_cbinfo.cb_notified_of_cb_path_down == FALSE) {
4823 		cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
4824 		*cs->statusp = resp->status = NFS4ERR_CB_PATH_DOWN;
4825 	} else {
4826 		*cs->statusp = resp->status = NFS4_OK;
4827 	}
4828 	mutex_exit(cp->rc_cbinfo.cb_lock);
4829 
4830 	rfs4_client_rele(cp);
4831 
4832 out:
4833 	DTRACE_NFSV4_2(op__renew__done, struct compound_state *, cs,
4834 	    RENEW4res *, resp);
4835 }
4836 
4837 /* ARGSUSED */
4838 static void
4839 rfs4_op_restorefh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
4840     struct compound_state *cs)
4841 {
4842 	RESTOREFH4res *resp = &resop->nfs_resop4_u.oprestorefh;
4843 
4844 	DTRACE_NFSV4_1(op__restorefh__start, struct compound_state *, cs);
4845 
4846 	/* No need to check cs->access - we are not accessing any object */
4847 	if ((cs->saved_vp == NULL) || (cs->saved_fh.nfs_fh4_val == NULL)) {
4848 		*cs->statusp = resp->status = NFS4ERR_RESTOREFH;
4849 		goto out;
4850 	}
4851 	if (cs->vp != NULL) {
4852 		VN_RELE(cs->vp);
4853 	}
4854 	cs->vp = cs->saved_vp;
4855 	cs->saved_vp = NULL;
4856 	if (cs->exi)
4857 		exi_rele(cs->exi);
4858 	cs->exi = cs->saved_exi;
4859 	if (cs->exi)
4860 		exi_hold(cs->exi);
4861 	nfs_fh4_copy(&cs->saved_fh, &cs->fh);
4862 	*cs->statusp = resp->status = NFS4_OK;
4863 	cs->deleg = FALSE;
4864 
4865 out:
4866 	DTRACE_NFSV4_2(op__restorefh__done, struct compound_state *, cs,
4867 	    RESTOREFH4res *, resp);
4868 }
4869 
4870 /* ARGSUSED */
4871 static void
4872 rfs4_op_savefh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4873     struct compound_state *cs)
4874 {
4875 	SAVEFH4res *resp = &resop->nfs_resop4_u.opsavefh;
4876 
4877 	DTRACE_NFSV4_1(op__savefh__start, struct compound_state *, cs);
4878 
4879 	/* No need to check cs->access - we are not accessing any object */
4880 	if (cs->vp == NULL) {
4881 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4882 		goto out;
4883 	}
4884 	if (cs->saved_vp != NULL) {
4885 		VN_RELE(cs->saved_vp);
4886 	}
4887 	cs->saved_vp = cs->vp;
4888 	VN_HOLD(cs->saved_vp);
4889 	if (cs->saved_exi)
4890 		exi_rele(cs->saved_exi);
4891 	cs->saved_exi = cs->exi;
4892 	if (cs->saved_exi)
4893 		exi_hold(cs->saved_exi);
4894 	/*
4895 	 * since SAVEFH is fairly rare, don't alloc space for its fh
4896 	 * unless necessary.
4897 	 */
4898 	if (cs->saved_fh.nfs_fh4_val == NULL) {
4899 		cs->saved_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP);
4900 	}
4901 	nfs_fh4_copy(&cs->fh, &cs->saved_fh);
4902 	*cs->statusp = resp->status = NFS4_OK;
4903 
4904 out:
4905 	DTRACE_NFSV4_2(op__savefh__done, struct compound_state *, cs,
4906 	    SAVEFH4res *, resp);
4907 }
4908 
4909 /*
4910  * rfs4_verify_attr is called when nfsv4 Setattr failed, but we wish to
4911  * return the bitmap of attrs that were set successfully. It is also
4912  * called by Verify/Nverify to test the vattr/vfsstat attrs. It should
4913  * always be called only after rfs4_do_set_attrs().
4914  *
4915  * Verify that the attributes are same as the expected ones. sargp->vap
4916  * and sargp->sbp contain the input attributes as translated from fattr4.
4917  *
4918  * This function verifies only the attrs that correspond to a vattr or
4919  * vfsstat struct. That is because of the extra step needed to get the
4920  * corresponding system structs. Other attributes have already been set or
4921  * verified by do_rfs4_set_attrs.
4922  *
4923  * Return 0 if all attrs match, -1 if some don't, error if error processing.
4924  */
4925 static int
4926 rfs4_verify_attr(struct nfs4_svgetit_arg *sargp,
4927     bitmap4 *resp, struct nfs4_ntov_table *ntovp)
4928 {
4929 	int error, ret_error = 0;
4930 	int i, k;
4931 	uint_t sva_mask = sargp->vap->va_mask;
4932 	uint_t vbit;
4933 	union nfs4_attr_u *na;
4934 	uint8_t *amap;
4935 	bool_t getsb = ntovp->vfsstat;
4936 
4937 	if (sva_mask != 0) {
4938 		/*
4939 		 * Okay to overwrite sargp->vap because we verify based
4940 		 * on the incoming values.
4941 		 */
4942 		ret_error = VOP_GETATTR(sargp->cs->vp, sargp->vap, 0,
4943 		    sargp->cs->cr, NULL);
4944 		if (ret_error) {
4945 			if (resp == NULL)
4946 				return (ret_error);
4947 			/*
4948 			 * Must return bitmap of successful attrs
4949 			 */
4950 			sva_mask = 0;	/* to prevent checking vap later */
4951 		} else {
4952 			/*
4953 			 * Some file systems clobber va_mask. it is probably
4954 			 * wrong of them to do so, nonethless we practice
4955 			 * defensive coding.
4956 			 * See bug id 4276830.
4957 			 */
4958 			sargp->vap->va_mask = sva_mask;
4959 		}
4960 	}
4961 
4962 	if (getsb) {
4963 		/*
4964 		 * Now get the superblock and loop on the bitmap, as there is
4965 		 * no simple way of translating from superblock to bitmap4.
4966 		 */
4967 		ret_error = VFS_STATVFS(sargp->cs->vp->v_vfsp, sargp->sbp);
4968 		if (ret_error) {
4969 			if (resp == NULL)
4970 				goto errout;
4971 			getsb = FALSE;
4972 		}
4973 	}
4974 
4975 	/*
4976 	 * Now loop and verify each attribute which getattr returned
4977 	 * whether it's the same as the input.
4978 	 */
4979 	if (resp == NULL && !getsb && (sva_mask == 0))
4980 		goto errout;
4981 
4982 	na = ntovp->na;
4983 	amap = ntovp->amap;
4984 	k = 0;
4985 	for (i = 0; i < ntovp->attrcnt; i++, na++, amap++) {
4986 		k = *amap;
4987 		ASSERT(nfs4_ntov_map[k].nval == k);
4988 		vbit = nfs4_ntov_map[k].vbit;
4989 
4990 		/*
4991 		 * If vattr attribute but VOP_GETATTR failed, or it's
4992 		 * superblock attribute but VFS_STATVFS failed, skip
4993 		 */
4994 		if (vbit) {
4995 			if ((vbit & sva_mask) == 0)
4996 				continue;
4997 		} else if (!(getsb && nfs4_ntov_map[k].vfsstat)) {
4998 			continue;
4999 		}
5000 		error = (*nfs4_ntov_map[k].sv_getit)(NFS4ATTR_VERIT, sargp, na);
5001 		if (resp != NULL) {
5002 			if (error)
5003 				ret_error = -1;	/* not all match */
5004 			else	/* update response bitmap */
5005 				*resp |= nfs4_ntov_map[k].fbit;
5006 			continue;
5007 		}
5008 		if (error) {
5009 			ret_error = -1;	/* not all match */
5010 			break;
5011 		}
5012 	}
5013 errout:
5014 	return (ret_error);
5015 }
5016 
5017 /*
5018  * Decode the attribute to be set/verified. If the attr requires a sys op
5019  * (VOP_GETATTR, VFS_VFSSTAT), and the request is to verify, then don't
5020  * call the sv_getit function for it, because the sys op hasn't yet been done.
5021  * Return 0 for success, error code if failed.
5022  *
5023  * Note: the decoded arg is not freed here but in nfs4_ntov_table_free.
5024  */
5025 static int
5026 decode_fattr4_attr(nfs4_attr_cmd_t cmd, struct nfs4_svgetit_arg *sargp,
5027     int k, XDR *xdrp, bitmap4 *resp_bval, union nfs4_attr_u *nap)
5028 {
5029 	int error = 0;
5030 	bool_t set_later;
5031 
5032 	sargp->vap->va_mask |= nfs4_ntov_map[k].vbit;
5033 
5034 	if ((*nfs4_ntov_map[k].xfunc)(xdrp, nap)) {
5035 		set_later = nfs4_ntov_map[k].vbit || nfs4_ntov_map[k].vfsstat;
5036 		/*
5037 		 * don't verify yet if a vattr or sb dependent attr,
5038 		 * because we don't have their sys values yet.
5039 		 * Will be done later.
5040 		 */
5041 		if (! (set_later && (cmd == NFS4ATTR_VERIT))) {
5042 			/*
5043 			 * ACLs are a special case, since setting the MODE
5044 			 * conflicts with setting the ACL.  We delay setting
5045 			 * the ACL until all other attributes have been set.
5046 			 * The ACL gets set in do_rfs4_op_setattr().
5047 			 */
5048 			if (nfs4_ntov_map[k].fbit != FATTR4_ACL_MASK) {
5049 				error = (*nfs4_ntov_map[k].sv_getit)(cmd,
5050 				    sargp, nap);
5051 				if (error) {
5052 					xdr_free(nfs4_ntov_map[k].xfunc,
5053 					    (caddr_t)nap);
5054 				}
5055 			}
5056 		}
5057 	} else {
5058 #ifdef  DEBUG
5059 		cmn_err(CE_NOTE, "decode_fattr4_attr: error "
5060 		    "decoding attribute %d\n", k);
5061 #endif
5062 		error = EINVAL;
5063 	}
5064 	if (!error && resp_bval && !set_later) {
5065 		*resp_bval |= nfs4_ntov_map[k].fbit;
5066 	}
5067 
5068 	return (error);
5069 }
5070 
5071 /*
5072  * Set vattr based on incoming fattr4 attrs - used by setattr.
5073  * Set response mask. Ignore any values that are not writable vattr attrs.
5074  */
5075 static nfsstat4
5076 do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5077     struct nfs4_svgetit_arg *sargp, struct nfs4_ntov_table *ntovp,
5078     nfs4_attr_cmd_t cmd)
5079 {
5080 	int error = 0;
5081 	int i;
5082 	char *attrs = fattrp->attrlist4;
5083 	uint32_t attrslen = fattrp->attrlist4_len;
5084 	XDR xdr;
5085 	nfsstat4 status = NFS4_OK;
5086 	vnode_t *vp = cs->vp;
5087 	union nfs4_attr_u *na;
5088 	uint8_t *amap;
5089 
5090 #ifndef lint
5091 	/*
5092 	 * Make sure that maximum attribute number can be expressed as an
5093 	 * 8 bit quantity.
5094 	 */
5095 	ASSERT(NFS4_MAXNUM_ATTRS <= (UINT8_MAX + 1));
5096 #endif
5097 
5098 	if (vp == NULL) {
5099 		if (resp)
5100 			*resp = 0;
5101 		return (NFS4ERR_NOFILEHANDLE);
5102 	}
5103 	if (cs->access == CS_ACCESS_DENIED) {
5104 		if (resp)
5105 			*resp = 0;
5106 		return (NFS4ERR_ACCESS);
5107 	}
5108 
5109 	sargp->op = cmd;
5110 	sargp->cs = cs;
5111 	sargp->flag = 0;	/* may be set later */
5112 	sargp->vap->va_mask = 0;
5113 	sargp->rdattr_error = NFS4_OK;
5114 	sargp->rdattr_error_req = FALSE;
5115 	/* sargp->sbp is set by the caller */
5116 
5117 	xdrmem_create(&xdr, attrs, attrslen, XDR_DECODE);
5118 
5119 	na = ntovp->na;
5120 	amap = ntovp->amap;
5121 
5122 	/*
5123 	 * The following loop iterates on the nfs4_ntov_map checking
5124 	 * if the fbit is set in the requested bitmap.
5125 	 * If set then we process the arguments using the
5126 	 * rfs4_fattr4 conversion functions to populate the setattr
5127 	 * vattr and va_mask. Any settable attrs that are not using vattr
5128 	 * will be set in this loop.
5129 	 */
5130 	for (i = 0; i < nfs4_ntov_map_size; i++) {
5131 		if (!(fattrp->attrmask & nfs4_ntov_map[i].fbit)) {
5132 			continue;
5133 		}
5134 		/*
5135 		 * If setattr, must be a writable attr.
5136 		 * If verify/nverify, must be a readable attr.
5137 		 */
5138 		if ((error = (*nfs4_ntov_map[i].sv_getit)(
5139 		    NFS4ATTR_SUPPORTED, sargp, NULL)) != 0) {
5140 			/*
5141 			 * Client tries to set/verify an
5142 			 * unsupported attribute, tries to set
5143 			 * a read only attr or verify a write
5144 			 * only one - error!
5145 			 */
5146 			break;
5147 		}
5148 		/*
5149 		 * Decode the attribute to set/verify
5150 		 */
5151 		error = decode_fattr4_attr(cmd, sargp, nfs4_ntov_map[i].nval,
5152 		    &xdr, resp ? resp : NULL, na);
5153 		if (error)
5154 			break;
5155 		*amap++ = (uint8_t)nfs4_ntov_map[i].nval;
5156 		na++;
5157 		(ntovp->attrcnt)++;
5158 		if (nfs4_ntov_map[i].vfsstat)
5159 			ntovp->vfsstat = TRUE;
5160 	}
5161 
5162 	if (error != 0)
5163 		status = (error == ENOTSUP ? NFS4ERR_ATTRNOTSUPP :
5164 		    puterrno4(error));
5165 	/* xdrmem_destroy(&xdrs); */	/* NO-OP */
5166 	return (status);
5167 }
5168 
5169 static nfsstat4
5170 do_rfs4_op_setattr(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5171     stateid4 *stateid)
5172 {
5173 	int error = 0;
5174 	struct nfs4_svgetit_arg sarg;
5175 	bool_t trunc;
5176 
5177 	nfsstat4 status = NFS4_OK;
5178 	cred_t *cr = cs->cr;
5179 	vnode_t *vp = cs->vp;
5180 	struct nfs4_ntov_table ntov;
5181 	struct statvfs64 sb;
5182 	struct vattr bva;
5183 	struct flock64 bf;
5184 	int in_crit = 0;
5185 	uint_t saved_mask = 0;
5186 	caller_context_t ct;
5187 
5188 	*resp = 0;
5189 	sarg.sbp = &sb;
5190 	sarg.is_referral = B_FALSE;
5191 	nfs4_ntov_table_init(&ntov);
5192 	status = do_rfs4_set_attrs(resp, fattrp, cs, &sarg, &ntov,
5193 	    NFS4ATTR_SETIT);
5194 	if (status != NFS4_OK) {
5195 		/*
5196 		 * failed set attrs
5197 		 */
5198 		goto done;
5199 	}
5200 	if ((sarg.vap->va_mask == 0) &&
5201 	    (! (fattrp->attrmask & FATTR4_ACL_MASK))) {
5202 		/*
5203 		 * no further work to be done
5204 		 */
5205 		goto done;
5206 	}
5207 
5208 	/*
5209 	 * If we got a request to set the ACL and the MODE, only
5210 	 * allow changing VSUID, VSGID, and VSVTX.  Attempting
5211 	 * to change any other bits, along with setting an ACL,
5212 	 * gives NFS4ERR_INVAL.
5213 	 */
5214 	if ((fattrp->attrmask & FATTR4_ACL_MASK) &&
5215 	    (fattrp->attrmask & FATTR4_MODE_MASK)) {
5216 		vattr_t va;
5217 
5218 		va.va_mask = AT_MODE;
5219 		error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
5220 		if (error) {
5221 			status = puterrno4(error);
5222 			goto done;
5223 		}
5224 		if ((sarg.vap->va_mode ^ va.va_mode) &
5225 		    ~(VSUID | VSGID | VSVTX)) {
5226 			status = NFS4ERR_INVAL;
5227 			goto done;
5228 		}
5229 	}
5230 
5231 	/* Check stateid only if size has been set */
5232 	if (sarg.vap->va_mask & AT_SIZE) {
5233 		trunc = (sarg.vap->va_size == 0);
5234 		status = rfs4_check_stateid(FWRITE, cs->vp, stateid,
5235 		    trunc, &cs->deleg, sarg.vap->va_mask & AT_SIZE, &ct);
5236 		if (status != NFS4_OK)
5237 			goto done;
5238 	} else {
5239 		ct.cc_sysid = 0;
5240 		ct.cc_pid = 0;
5241 		ct.cc_caller_id = nfs4_srv_caller_id;
5242 		ct.cc_flags = CC_DONTBLOCK;
5243 	}
5244 
5245 	/* XXX start of possible race with delegations */
5246 
5247 	/*
5248 	 * We need to specially handle size changes because it is
5249 	 * possible for the client to create a file with read-only
5250 	 * modes, but with the file opened for writing. If the client
5251 	 * then tries to set the file size, e.g. ftruncate(3C),
5252 	 * fcntl(F_FREESP), the normal access checking done in
5253 	 * VOP_SETATTR would prevent the client from doing it even though
5254 	 * it should be allowed to do so.  To get around this, we do the
5255 	 * access checking for ourselves and use VOP_SPACE which doesn't
5256 	 * do the access checking.
5257 	 * Also the client should not be allowed to change the file
5258 	 * size if there is a conflicting non-blocking mandatory lock in
5259 	 * the region of the change.
5260 	 */
5261 	if (vp->v_type == VREG && (sarg.vap->va_mask & AT_SIZE)) {
5262 		u_offset_t offset;
5263 		ssize_t length;
5264 
5265 		/*
5266 		 * ufs_setattr clears AT_SIZE from vap->va_mask, but
5267 		 * before returning, sarg.vap->va_mask is used to
5268 		 * generate the setattr reply bitmap.  We also clear
5269 		 * AT_SIZE below before calling VOP_SPACE.  For both
5270 		 * of these cases, the va_mask needs to be saved here
5271 		 * and restored after calling VOP_SETATTR.
5272 		 */
5273 		saved_mask = sarg.vap->va_mask;
5274 
5275 		/*
5276 		 * Check any possible conflict due to NBMAND locks.
5277 		 * Get into critical region before VOP_GETATTR, so the
5278 		 * size attribute is valid when checking conflicts.
5279 		 */
5280 		if (nbl_need_check(vp)) {
5281 			nbl_start_crit(vp, RW_READER);
5282 			in_crit = 1;
5283 		}
5284 
5285 		bva.va_mask = AT_UID|AT_SIZE;
5286 		if (error = VOP_GETATTR(vp, &bva, 0, cr, &ct)) {
5287 			status = puterrno4(error);
5288 			goto done;
5289 		}
5290 
5291 		if (in_crit) {
5292 			if (sarg.vap->va_size < bva.va_size) {
5293 				offset = sarg.vap->va_size;
5294 				length = bva.va_size - sarg.vap->va_size;
5295 			} else {
5296 				offset = bva.va_size;
5297 				length = sarg.vap->va_size - bva.va_size;
5298 			}
5299 			if (nbl_conflict(vp, NBL_WRITE, offset, length, 0,
5300 			    &ct)) {
5301 				status = NFS4ERR_LOCKED;
5302 				goto done;
5303 			}
5304 		}
5305 
5306 		if (crgetuid(cr) == bva.va_uid) {
5307 			sarg.vap->va_mask &= ~AT_SIZE;
5308 			bf.l_type = F_WRLCK;
5309 			bf.l_whence = 0;
5310 			bf.l_start = (off64_t)sarg.vap->va_size;
5311 			bf.l_len = 0;
5312 			bf.l_sysid = 0;
5313 			bf.l_pid = 0;
5314 			error = VOP_SPACE(vp, F_FREESP, &bf, FWRITE,
5315 			    (offset_t)sarg.vap->va_size, cr, &ct);
5316 		}
5317 	}
5318 
5319 	if (!error && sarg.vap->va_mask != 0)
5320 		error = VOP_SETATTR(vp, sarg.vap, sarg.flag, cr, &ct);
5321 
5322 	/* restore va_mask -- ufs_setattr clears AT_SIZE */
5323 	if (saved_mask & AT_SIZE)
5324 		sarg.vap->va_mask |= AT_SIZE;
5325 
5326 	/*
5327 	 * If an ACL was being set, it has been delayed until now,
5328 	 * in order to set the mode (via the VOP_SETATTR() above) first.
5329 	 */
5330 	if ((! error) && (fattrp->attrmask & FATTR4_ACL_MASK)) {
5331 		int i;
5332 
5333 		for (i = 0; i < NFS4_MAXNUM_ATTRS; i++)
5334 			if (ntov.amap[i] == FATTR4_ACL)
5335 				break;
5336 		if (i < NFS4_MAXNUM_ATTRS) {
5337 			error = (*nfs4_ntov_map[FATTR4_ACL].sv_getit)(
5338 			    NFS4ATTR_SETIT, &sarg, &ntov.na[i]);
5339 			if (error == 0) {
5340 				*resp |= FATTR4_ACL_MASK;
5341 			} else if (error == ENOTSUP) {
5342 				(void) rfs4_verify_attr(&sarg, resp, &ntov);
5343 				status = NFS4ERR_ATTRNOTSUPP;
5344 				goto done;
5345 			}
5346 		} else {
5347 			NFS4_DEBUG(rfs4_debug,
5348 			    (CE_NOTE, "do_rfs4_op_setattr: "
5349 			    "unable to find ACL in fattr4"));
5350 			error = EINVAL;
5351 		}
5352 	}
5353 
5354 	if (error) {
5355 		/* check if a monitor detected a delegation conflict */
5356 		if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
5357 			status = NFS4ERR_DELAY;
5358 		else
5359 			status = puterrno4(error);
5360 
5361 		/*
5362 		 * Set the response bitmap when setattr failed.
5363 		 * If VOP_SETATTR partially succeeded, test by doing a
5364 		 * VOP_GETATTR on the object and comparing the data
5365 		 * to the setattr arguments.
5366 		 */
5367 		(void) rfs4_verify_attr(&sarg, resp, &ntov);
5368 	} else {
5369 		/*
5370 		 * Force modified metadata out to stable storage.
5371 		 */
5372 		(void) VOP_FSYNC(vp, FNODSYNC, cr, &ct);
5373 		/*
5374 		 * Set response bitmap
5375 		 */
5376 		nfs4_vmask_to_nmask_set(sarg.vap->va_mask, resp);
5377 	}
5378 
5379 /* Return early and already have a NFSv4 error */
5380 done:
5381 	/*
5382 	 * Except for nfs4_vmask_to_nmask_set(), vattr --> fattr
5383 	 * conversion sets both readable and writeable NFS4 attrs
5384 	 * for AT_MTIME and AT_ATIME.  The line below masks out
5385 	 * unrequested attrs from the setattr result bitmap.  This
5386 	 * is placed after the done: label to catch the ATTRNOTSUP
5387 	 * case.
5388 	 */
5389 	*resp &= fattrp->attrmask;
5390 
5391 	if (in_crit)
5392 		nbl_end_crit(vp);
5393 
5394 	nfs4_ntov_table_free(&ntov, &sarg);
5395 
5396 	return (status);
5397 }
5398 
5399 /* ARGSUSED */
5400 static void
5401 rfs4_op_setattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5402     struct compound_state *cs)
5403 {
5404 	SETATTR4args *args = &argop->nfs_argop4_u.opsetattr;
5405 	SETATTR4res *resp = &resop->nfs_resop4_u.opsetattr;
5406 	bslabel_t *clabel;
5407 
5408 	DTRACE_NFSV4_2(op__setattr__start, struct compound_state *, cs,
5409 	    SETATTR4args *, args);
5410 
5411 	if (cs->vp == NULL) {
5412 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5413 		goto out;
5414 	}
5415 
5416 	/*
5417 	 * If there is an unshared filesystem mounted on this vnode,
5418 	 * do not allow to setattr on this vnode.
5419 	 */
5420 	if (vn_ismntpt(cs->vp)) {
5421 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5422 		goto out;
5423 	}
5424 
5425 	resp->attrsset = 0;
5426 
5427 	if (rdonly4(req, cs)) {
5428 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5429 		goto out;
5430 	}
5431 
5432 	/* check label before setting attributes */
5433 	if (is_system_labeled()) {
5434 		ASSERT(req->rq_label != NULL);
5435 		clabel = req->rq_label;
5436 		DTRACE_PROBE2(tx__rfs4__log__info__opsetattr__clabel, char *,
5437 		    "got client label from request(1)",
5438 		    struct svc_req *, req);
5439 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
5440 			if (!do_rfs_label_check(clabel, cs->vp,
5441 			    EQUALITY_CHECK, cs->exi)) {
5442 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
5443 				goto out;
5444 			}
5445 		}
5446 	}
5447 
5448 	*cs->statusp = resp->status =
5449 	    do_rfs4_op_setattr(&resp->attrsset, &args->obj_attributes, cs,
5450 	    &args->stateid);
5451 
5452 out:
5453 	DTRACE_NFSV4_2(op__setattr__done, struct compound_state *, cs,
5454 	    SETATTR4res *, resp);
5455 }
5456 
5457 /* ARGSUSED */
5458 static void
5459 rfs4_op_verify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5460     struct compound_state *cs)
5461 {
5462 	/*
5463 	 * verify and nverify are exactly the same, except that nverify
5464 	 * succeeds when some argument changed, and verify succeeds when
5465 	 * when none changed.
5466 	 */
5467 
5468 	VERIFY4args  *args = &argop->nfs_argop4_u.opverify;
5469 	VERIFY4res *resp = &resop->nfs_resop4_u.opverify;
5470 
5471 	int error;
5472 	struct nfs4_svgetit_arg sarg;
5473 	struct statvfs64 sb;
5474 	struct nfs4_ntov_table ntov;
5475 
5476 	DTRACE_NFSV4_2(op__verify__start, struct compound_state *, cs,
5477 	    VERIFY4args *, args);
5478 
5479 	if (cs->vp == NULL) {
5480 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5481 		goto out;
5482 	}
5483 
5484 	sarg.sbp = &sb;
5485 	sarg.is_referral = B_FALSE;
5486 	nfs4_ntov_table_init(&ntov);
5487 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5488 	    &sarg, &ntov, NFS4ATTR_VERIT);
5489 	if (resp->status != NFS4_OK) {
5490 		/*
5491 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5492 		 * so could return -1 for "no match".
5493 		 */
5494 		if (resp->status == -1)
5495 			resp->status = NFS4ERR_NOT_SAME;
5496 		goto done;
5497 	}
5498 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5499 	switch (error) {
5500 	case 0:
5501 		resp->status = NFS4_OK;
5502 		break;
5503 	case -1:
5504 		resp->status = NFS4ERR_NOT_SAME;
5505 		break;
5506 	default:
5507 		resp->status = puterrno4(error);
5508 		break;
5509 	}
5510 done:
5511 	*cs->statusp = resp->status;
5512 	nfs4_ntov_table_free(&ntov, &sarg);
5513 out:
5514 	DTRACE_NFSV4_2(op__verify__done, struct compound_state *, cs,
5515 	    VERIFY4res *, resp);
5516 }
5517 
5518 /* ARGSUSED */
5519 static void
5520 rfs4_op_nverify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5521     struct compound_state *cs)
5522 {
5523 	/*
5524 	 * verify and nverify are exactly the same, except that nverify
5525 	 * succeeds when some argument changed, and verify succeeds when
5526 	 * when none changed.
5527 	 */
5528 
5529 	NVERIFY4args  *args = &argop->nfs_argop4_u.opnverify;
5530 	NVERIFY4res *resp = &resop->nfs_resop4_u.opnverify;
5531 
5532 	int error;
5533 	struct nfs4_svgetit_arg sarg;
5534 	struct statvfs64 sb;
5535 	struct nfs4_ntov_table ntov;
5536 
5537 	DTRACE_NFSV4_2(op__nverify__start, struct compound_state *, cs,
5538 	    NVERIFY4args *, args);
5539 
5540 	if (cs->vp == NULL) {
5541 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5542 		DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5543 		    NVERIFY4res *, resp);
5544 		return;
5545 	}
5546 	sarg.sbp = &sb;
5547 	sarg.is_referral = B_FALSE;
5548 	nfs4_ntov_table_init(&ntov);
5549 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5550 	    &sarg, &ntov, NFS4ATTR_VERIT);
5551 	if (resp->status != NFS4_OK) {
5552 		/*
5553 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5554 		 * so could return -1 for "no match".
5555 		 */
5556 		if (resp->status == -1)
5557 			resp->status = NFS4_OK;
5558 		goto done;
5559 	}
5560 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5561 	switch (error) {
5562 	case 0:
5563 		resp->status = NFS4ERR_SAME;
5564 		break;
5565 	case -1:
5566 		resp->status = NFS4_OK;
5567 		break;
5568 	default:
5569 		resp->status = puterrno4(error);
5570 		break;
5571 	}
5572 done:
5573 	*cs->statusp = resp->status;
5574 	nfs4_ntov_table_free(&ntov, &sarg);
5575 
5576 	DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5577 	    NVERIFY4res *, resp);
5578 }
5579 
5580 /*
5581  * XXX - This should live in an NFS header file.
5582  */
5583 #define	MAX_IOVECS	12
5584 
5585 /* ARGSUSED */
5586 static void
5587 rfs4_op_write(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5588     struct compound_state *cs)
5589 {
5590 	WRITE4args *args = &argop->nfs_argop4_u.opwrite;
5591 	WRITE4res *resp = &resop->nfs_resop4_u.opwrite;
5592 	int error;
5593 	vnode_t *vp;
5594 	struct vattr bva;
5595 	u_offset_t rlimit;
5596 	struct uio uio;
5597 	struct iovec iov[MAX_IOVECS];
5598 	struct iovec *iovp;
5599 	int iovcnt;
5600 	int ioflag;
5601 	cred_t *savecred, *cr;
5602 	bool_t *deleg = &cs->deleg;
5603 	nfsstat4 stat;
5604 	int in_crit = 0;
5605 	caller_context_t ct;
5606 
5607 	DTRACE_NFSV4_2(op__write__start, struct compound_state *, cs,
5608 	    WRITE4args *, args);
5609 
5610 	vp = cs->vp;
5611 	if (vp == NULL) {
5612 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5613 		goto out;
5614 	}
5615 	if (cs->access == CS_ACCESS_DENIED) {
5616 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5617 		goto out;
5618 	}
5619 
5620 	cr = cs->cr;
5621 
5622 	if ((stat = rfs4_check_stateid(FWRITE, vp, &args->stateid, FALSE,
5623 	    deleg, TRUE, &ct)) != NFS4_OK) {
5624 		*cs->statusp = resp->status = stat;
5625 		goto out;
5626 	}
5627 
5628 	/*
5629 	 * We have to enter the critical region before calling VOP_RWLOCK
5630 	 * to avoid a deadlock with ufs.
5631 	 */
5632 	if (nbl_need_check(vp)) {
5633 		nbl_start_crit(vp, RW_READER);
5634 		in_crit = 1;
5635 		if (nbl_conflict(vp, NBL_WRITE,
5636 		    args->offset, args->data_len, 0, &ct)) {
5637 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
5638 			goto out;
5639 		}
5640 	}
5641 
5642 	bva.va_mask = AT_MODE | AT_UID;
5643 	error = VOP_GETATTR(vp, &bva, 0, cr, &ct);
5644 
5645 	/*
5646 	 * If we can't get the attributes, then we can't do the
5647 	 * right access checking.  So, we'll fail the request.
5648 	 */
5649 	if (error) {
5650 		*cs->statusp = resp->status = puterrno4(error);
5651 		goto out;
5652 	}
5653 
5654 	if (rdonly4(req, cs)) {
5655 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5656 		goto out;
5657 	}
5658 
5659 	if (vp->v_type != VREG) {
5660 		*cs->statusp = resp->status =
5661 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
5662 		goto out;
5663 	}
5664 
5665 	if (crgetuid(cr) != bva.va_uid &&
5666 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, &ct))) {
5667 		*cs->statusp = resp->status = puterrno4(error);
5668 		goto out;
5669 	}
5670 
5671 	if (MANDLOCK(vp, bva.va_mode)) {
5672 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5673 		goto out;
5674 	}
5675 
5676 	if (args->data_len == 0) {
5677 		*cs->statusp = resp->status = NFS4_OK;
5678 		resp->count = 0;
5679 		resp->committed = args->stable;
5680 		resp->writeverf = Write4verf;
5681 		goto out;
5682 	}
5683 
5684 	if (args->mblk != NULL) {
5685 		mblk_t *m;
5686 		uint_t bytes, round_len;
5687 
5688 		iovcnt = 0;
5689 		bytes = 0;
5690 		round_len = roundup(args->data_len, BYTES_PER_XDR_UNIT);
5691 		for (m = args->mblk;
5692 		    m != NULL && bytes < round_len;
5693 		    m = m->b_cont) {
5694 			iovcnt++;
5695 			bytes += MBLKL(m);
5696 		}
5697 #ifdef DEBUG
5698 		/* should have ended on an mblk boundary */
5699 		if (bytes != round_len) {
5700 			printf("bytes=0x%x, round_len=0x%x, req len=0x%x\n",
5701 			    bytes, round_len, args->data_len);
5702 			printf("args=%p, args->mblk=%p, m=%p", (void *)args,
5703 			    (void *)args->mblk, (void *)m);
5704 			ASSERT(bytes == round_len);
5705 		}
5706 #endif
5707 		if (iovcnt <= MAX_IOVECS) {
5708 			iovp = iov;
5709 		} else {
5710 			iovp = kmem_alloc(sizeof (*iovp) * iovcnt, KM_SLEEP);
5711 		}
5712 		mblk_to_iov(args->mblk, iovcnt, iovp);
5713 	} else if (args->rlist != NULL) {
5714 		iovcnt = 1;
5715 		iovp = iov;
5716 		iovp->iov_base = (char *)((args->rlist)->u.c_daddr3);
5717 		iovp->iov_len = args->data_len;
5718 	} else {
5719 		iovcnt = 1;
5720 		iovp = iov;
5721 		iovp->iov_base = args->data_val;
5722 		iovp->iov_len = args->data_len;
5723 	}
5724 
5725 	uio.uio_iov = iovp;
5726 	uio.uio_iovcnt = iovcnt;
5727 
5728 	uio.uio_segflg = UIO_SYSSPACE;
5729 	uio.uio_extflg = UIO_COPY_DEFAULT;
5730 	uio.uio_loffset = args->offset;
5731 	uio.uio_resid = args->data_len;
5732 	uio.uio_llimit = curproc->p_fsz_ctl;
5733 	rlimit = uio.uio_llimit - args->offset;
5734 	if (rlimit < (u_offset_t)uio.uio_resid)
5735 		uio.uio_resid = (int)rlimit;
5736 
5737 	if (args->stable == UNSTABLE4)
5738 		ioflag = 0;
5739 	else if (args->stable == FILE_SYNC4)
5740 		ioflag = FSYNC;
5741 	else if (args->stable == DATA_SYNC4)
5742 		ioflag = FDSYNC;
5743 	else {
5744 		if (iovp != iov)
5745 			kmem_free(iovp, sizeof (*iovp) * iovcnt);
5746 		*cs->statusp = resp->status = NFS4ERR_INVAL;
5747 		goto out;
5748 	}
5749 
5750 	/*
5751 	 * We're changing creds because VM may fault and we need
5752 	 * the cred of the current thread to be used if quota
5753 	 * checking is enabled.
5754 	 */
5755 	savecred = curthread->t_cred;
5756 	curthread->t_cred = cr;
5757 	error = do_io(FWRITE, vp, &uio, ioflag, cr, &ct);
5758 	curthread->t_cred = savecred;
5759 
5760 	if (iovp != iov)
5761 		kmem_free(iovp, sizeof (*iovp) * iovcnt);
5762 
5763 	if (error) {
5764 		*cs->statusp = resp->status = puterrno4(error);
5765 		goto out;
5766 	}
5767 
5768 	*cs->statusp = resp->status = NFS4_OK;
5769 	resp->count = args->data_len - uio.uio_resid;
5770 
5771 	if (ioflag == 0)
5772 		resp->committed = UNSTABLE4;
5773 	else
5774 		resp->committed = FILE_SYNC4;
5775 
5776 	resp->writeverf = Write4verf;
5777 
5778 out:
5779 	if (in_crit)
5780 		nbl_end_crit(vp);
5781 
5782 	DTRACE_NFSV4_2(op__write__done, struct compound_state *, cs,
5783 	    WRITE4res *, resp);
5784 }
5785 
5786 
5787 /* XXX put in a header file */
5788 extern int	sec_svc_getcred(struct svc_req *, cred_t *,  caddr_t *, int *);
5789 
5790 void
5791 rfs4_compound(COMPOUND4args *args, COMPOUND4res *resp, struct exportinfo *exi,
5792     struct svc_req *req, cred_t *cr, int *rv)
5793 {
5794 	uint_t i;
5795 	struct compound_state cs;
5796 
5797 	if (rv != NULL)
5798 		*rv = 0;
5799 	rfs4_init_compound_state(&cs);
5800 	/*
5801 	 * Form a reply tag by copying over the reqeuest tag.
5802 	 */
5803 	resp->tag.utf8string_val =
5804 	    kmem_alloc(args->tag.utf8string_len, KM_SLEEP);
5805 	resp->tag.utf8string_len = args->tag.utf8string_len;
5806 	bcopy(args->tag.utf8string_val, resp->tag.utf8string_val,
5807 	    resp->tag.utf8string_len);
5808 
5809 	cs.statusp = &resp->status;
5810 	cs.req = req;
5811 	resp->array = NULL;
5812 	resp->array_len = 0;
5813 
5814 	/*
5815 	 * XXX for now, minorversion should be zero
5816 	 */
5817 	if (args->minorversion != NFS4_MINORVERSION) {
5818 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5819 		    &cs, COMPOUND4args *, args);
5820 		resp->status = NFS4ERR_MINOR_VERS_MISMATCH;
5821 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5822 		    &cs, COMPOUND4res *, resp);
5823 		return;
5824 	}
5825 
5826 	if (args->array_len == 0) {
5827 		resp->status = NFS4_OK;
5828 		return;
5829 	}
5830 
5831 	ASSERT(exi == NULL);
5832 	ASSERT(cr == NULL);
5833 
5834 	cr = crget();
5835 	ASSERT(cr != NULL);
5836 
5837 	if (sec_svc_getcred(req, cr, &cs.principal, &cs.nfsflavor) == 0) {
5838 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5839 		    &cs, COMPOUND4args *, args);
5840 		crfree(cr);
5841 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5842 		    &cs, COMPOUND4res *, resp);
5843 		svcerr_badcred(req->rq_xprt);
5844 		if (rv != NULL)
5845 			*rv = 1;
5846 		return;
5847 	}
5848 	resp->array_len = args->array_len;
5849 	resp->array = kmem_zalloc(args->array_len * sizeof (nfs_resop4),
5850 	    KM_SLEEP);
5851 
5852 	cs.basecr = cr;
5853 
5854 	DTRACE_NFSV4_2(compound__start, struct compound_state *, &cs,
5855 	    COMPOUND4args *, args);
5856 
5857 	/*
5858 	 * If this is the first compound we've seen, we need to start all
5859 	 * new instances' grace periods.
5860 	 */
5861 	if (rfs4_seen_first_compound == 0) {
5862 		rfs4_grace_start_new();
5863 		/*
5864 		 * This must be set after rfs4_grace_start_new(), otherwise
5865 		 * another thread could proceed past here before the former
5866 		 * is finished.
5867 		 */
5868 		rfs4_seen_first_compound = 1;
5869 	}
5870 
5871 	for (i = 0; i < args->array_len && cs.cont; i++) {
5872 		nfs_argop4 *argop;
5873 		nfs_resop4 *resop;
5874 		uint_t op;
5875 
5876 		argop = &args->array[i];
5877 		resop = &resp->array[i];
5878 		resop->resop = argop->argop;
5879 		op = (uint_t)resop->resop;
5880 
5881 		if (op < rfsv4disp_cnt) {
5882 			/*
5883 			 * Count the individual ops here; NULL and COMPOUND
5884 			 * are counted in common_dispatch()
5885 			 */
5886 			rfsproccnt_v4_ptr[op].value.ui64++;
5887 
5888 			NFS4_DEBUG(rfs4_debug > 1,
5889 			    (CE_NOTE, "Executing %s", rfs4_op_string[op]));
5890 			(*rfsv4disptab[op].dis_proc)(argop, resop, req, &cs);
5891 			NFS4_DEBUG(rfs4_debug > 1, (CE_NOTE, "%s returned %d",
5892 			    rfs4_op_string[op], *cs.statusp));
5893 			if (*cs.statusp != NFS4_OK)
5894 				cs.cont = FALSE;
5895 		} else {
5896 			/*
5897 			 * This is effectively dead code since XDR code
5898 			 * will have already returned BADXDR if op doesn't
5899 			 * decode to legal value.  This only done for a
5900 			 * day when XDR code doesn't verify v4 opcodes.
5901 			 */
5902 			op = OP_ILLEGAL;
5903 			rfsproccnt_v4_ptr[OP_ILLEGAL_IDX].value.ui64++;
5904 
5905 			rfs4_op_illegal(argop, resop, req, &cs);
5906 			cs.cont = FALSE;
5907 		}
5908 
5909 		/*
5910 		 * If not at last op, and if we are to stop, then
5911 		 * compact the results array.
5912 		 */
5913 		if ((i + 1) < args->array_len && !cs.cont) {
5914 			nfs_resop4 *new_res = kmem_alloc(
5915 			    (i+1) * sizeof (nfs_resop4), KM_SLEEP);
5916 			bcopy(resp->array,
5917 			    new_res, (i+1) * sizeof (nfs_resop4));
5918 			kmem_free(resp->array,
5919 			    args->array_len * sizeof (nfs_resop4));
5920 
5921 			resp->array_len =  i + 1;
5922 			resp->array = new_res;
5923 		}
5924 	}
5925 
5926 
5927 	DTRACE_NFSV4_2(compound__done, struct compound_state *, &cs,
5928 	    COMPOUND4res *, resp);
5929 
5930 	if (cs.exi)
5931 		exi_rele(cs.exi);
5932 	if (cs.saved_exi)
5933 		exi_rele(cs.saved_exi);
5934 	if (cs.vp)
5935 		VN_RELE(cs.vp);
5936 	if (cs.saved_vp)
5937 		VN_RELE(cs.saved_vp);
5938 	if (cs.saved_fh.nfs_fh4_val)
5939 		kmem_free(cs.saved_fh.nfs_fh4_val, NFS4_FHSIZE);
5940 
5941 	if (cs.basecr)
5942 		crfree(cs.basecr);
5943 	if (cs.cr)
5944 		crfree(cs.cr);
5945 	/*
5946 	 * done with this compound request, free the label
5947 	 */
5948 
5949 	if (req->rq_label != NULL) {
5950 		kmem_free(req->rq_label, sizeof (bslabel_t));
5951 		req->rq_label = NULL;
5952 	}
5953 }
5954 
5955 /*
5956  * XXX because of what appears to be duplicate calls to rfs4_compound_free
5957  * XXX zero out the tag and array values. Need to investigate why the
5958  * XXX calls occur, but at least prevent the panic for now.
5959  */
5960 void
5961 rfs4_compound_free(COMPOUND4res *resp)
5962 {
5963 	uint_t i;
5964 
5965 	if (resp->tag.utf8string_val) {
5966 		UTF8STRING_FREE(resp->tag)
5967 	}
5968 
5969 	for (i = 0; i < resp->array_len; i++) {
5970 		nfs_resop4 *resop;
5971 		uint_t op;
5972 
5973 		resop = &resp->array[i];
5974 		op = (uint_t)resop->resop;
5975 		if (op < rfsv4disp_cnt) {
5976 			(*rfsv4disptab[op].dis_resfree)(resop);
5977 		}
5978 	}
5979 	if (resp->array != NULL) {
5980 		kmem_free(resp->array, resp->array_len * sizeof (nfs_resop4));
5981 	}
5982 }
5983 
5984 /*
5985  * Process the value of the compound request rpc flags, as a bit-AND
5986  * of the individual per-op flags (idempotent, allowork, publicfh_ok)
5987  */
5988 void
5989 rfs4_compound_flagproc(COMPOUND4args *args, int *flagp)
5990 {
5991 	int i;
5992 	int flag = RPC_ALL;
5993 
5994 	for (i = 0; flag && i < args->array_len; i++) {
5995 		uint_t op;
5996 
5997 		op = (uint_t)args->array[i].argop;
5998 
5999 		if (op < rfsv4disp_cnt)
6000 			flag &= rfsv4disptab[op].dis_flags;
6001 		else
6002 			flag = 0;
6003 	}
6004 	*flagp = flag;
6005 }
6006 
6007 nfsstat4
6008 rfs4_client_sysid(rfs4_client_t *cp, sysid_t *sp)
6009 {
6010 	nfsstat4 e;
6011 
6012 	rfs4_dbe_lock(cp->rc_dbe);
6013 
6014 	if (cp->rc_sysidt != LM_NOSYSID) {
6015 		*sp = cp->rc_sysidt;
6016 		e = NFS4_OK;
6017 
6018 	} else if ((cp->rc_sysidt = lm_alloc_sysidt()) != LM_NOSYSID) {
6019 		*sp = cp->rc_sysidt;
6020 		e = NFS4_OK;
6021 
6022 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
6023 		    "rfs4_client_sysid: allocated 0x%x\n", *sp));
6024 	} else
6025 		e = NFS4ERR_DELAY;
6026 
6027 	rfs4_dbe_unlock(cp->rc_dbe);
6028 	return (e);
6029 }
6030 
6031 #if defined(DEBUG) && ! defined(lint)
6032 static void lock_print(char *str, int operation, struct flock64 *flk)
6033 {
6034 	char *op, *type;
6035 
6036 	switch (operation) {
6037 	case F_GETLK: op = "F_GETLK";
6038 		break;
6039 	case F_SETLK: op = "F_SETLK";
6040 		break;
6041 	case F_SETLK_NBMAND: op = "F_SETLK_NBMAND";
6042 		break;
6043 	default: op = "F_UNKNOWN";
6044 		break;
6045 	}
6046 	switch (flk->l_type) {
6047 	case F_UNLCK: type = "F_UNLCK";
6048 		break;
6049 	case F_RDLCK: type = "F_RDLCK";
6050 		break;
6051 	case F_WRLCK: type = "F_WRLCK";
6052 		break;
6053 	default: type = "F_UNKNOWN";
6054 		break;
6055 	}
6056 
6057 	ASSERT(flk->l_whence == 0);
6058 	cmn_err(CE_NOTE, "%s:  %s, type = %s, off = %llx len = %llx pid = %d",
6059 	    str, op, type, (longlong_t)flk->l_start,
6060 	    flk->l_len ? (longlong_t)flk->l_len : ~0LL, flk->l_pid);
6061 }
6062 
6063 #define	LOCK_PRINT(d, s, t, f) if (d) lock_print(s, t, f)
6064 #else
6065 #define	LOCK_PRINT(d, s, t, f)
6066 #endif
6067 
6068 /*ARGSUSED*/
6069 static bool_t
6070 creds_ok(cred_set_t cr_set, struct svc_req *req, struct compound_state *cs)
6071 {
6072 	return (TRUE);
6073 }
6074 
6075 /*
6076  * Look up the pathname using the vp in cs as the directory vnode.
6077  * cs->vp will be the vnode for the file on success
6078  */
6079 
6080 static nfsstat4
6081 rfs4_lookup(component4 *component, struct svc_req *req,
6082     struct compound_state *cs)
6083 {
6084 	char *nm;
6085 	uint32_t len;
6086 	nfsstat4 status;
6087 	struct sockaddr *ca;
6088 	char *name;
6089 
6090 	if (cs->vp == NULL) {
6091 		return (NFS4ERR_NOFILEHANDLE);
6092 	}
6093 	if (cs->vp->v_type != VDIR) {
6094 		return (NFS4ERR_NOTDIR);
6095 	}
6096 
6097 	status = utf8_dir_verify(component);
6098 	if (status != NFS4_OK)
6099 		return (status);
6100 
6101 	nm = utf8_to_fn(component, &len, NULL);
6102 	if (nm == NULL) {
6103 		return (NFS4ERR_INVAL);
6104 	}
6105 
6106 	if (len > MAXNAMELEN) {
6107 		kmem_free(nm, len);
6108 		return (NFS4ERR_NAMETOOLONG);
6109 	}
6110 
6111 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6112 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6113 	    MAXPATHLEN + 1);
6114 
6115 	if (name == NULL) {
6116 		kmem_free(nm, len);
6117 		return (NFS4ERR_INVAL);
6118 	}
6119 
6120 	status = do_rfs4_op_lookup(name, req, cs);
6121 
6122 	if (name != nm)
6123 		kmem_free(name, MAXPATHLEN + 1);
6124 
6125 	kmem_free(nm, len);
6126 
6127 	return (status);
6128 }
6129 
6130 static nfsstat4
6131 rfs4_lookupfile(component4 *component, struct svc_req *req,
6132     struct compound_state *cs, uint32_t access, change_info4 *cinfo)
6133 {
6134 	nfsstat4 status;
6135 	vnode_t *dvp = cs->vp;
6136 	vattr_t bva, ava, fva;
6137 	int error;
6138 
6139 	/* Get "before" change value */
6140 	bva.va_mask = AT_CTIME|AT_SEQ;
6141 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6142 	if (error)
6143 		return (puterrno4(error));
6144 
6145 	/* rfs4_lookup may VN_RELE directory */
6146 	VN_HOLD(dvp);
6147 
6148 	status = rfs4_lookup(component, req, cs);
6149 	if (status != NFS4_OK) {
6150 		VN_RELE(dvp);
6151 		return (status);
6152 	}
6153 
6154 	/*
6155 	 * Get "after" change value, if it fails, simply return the
6156 	 * before value.
6157 	 */
6158 	ava.va_mask = AT_CTIME|AT_SEQ;
6159 	if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6160 		ava.va_ctime = bva.va_ctime;
6161 		ava.va_seq = 0;
6162 	}
6163 	VN_RELE(dvp);
6164 
6165 	/*
6166 	 * Validate the file is a file
6167 	 */
6168 	fva.va_mask = AT_TYPE|AT_MODE;
6169 	error = VOP_GETATTR(cs->vp, &fva, 0, cs->cr, NULL);
6170 	if (error)
6171 		return (puterrno4(error));
6172 
6173 	if (fva.va_type != VREG) {
6174 		if (fva.va_type == VDIR)
6175 			return (NFS4ERR_ISDIR);
6176 		if (fva.va_type == VLNK)
6177 			return (NFS4ERR_SYMLINK);
6178 		return (NFS4ERR_INVAL);
6179 	}
6180 
6181 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime);
6182 	NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6183 
6184 	/*
6185 	 * It is undefined if VOP_LOOKUP will change va_seq, so
6186 	 * cinfo.atomic = TRUE only if we have
6187 	 * non-zero va_seq's, and they have not changed.
6188 	 */
6189 	if (bva.va_seq && ava.va_seq && ava.va_seq == bva.va_seq)
6190 		cinfo->atomic = TRUE;
6191 	else
6192 		cinfo->atomic = FALSE;
6193 
6194 	/* Check for mandatory locking */
6195 	cs->mandlock = MANDLOCK(cs->vp, fva.va_mode);
6196 	return (check_open_access(access, cs, req));
6197 }
6198 
6199 static nfsstat4
6200 create_vnode(vnode_t *dvp, char *nm,  vattr_t *vap, createmode4 mode,
6201     timespec32_t *mtime, cred_t *cr, vnode_t **vpp, bool_t *created)
6202 {
6203 	int error;
6204 	nfsstat4 status = NFS4_OK;
6205 	vattr_t va;
6206 
6207 tryagain:
6208 
6209 	/*
6210 	 * The file open mode used is VWRITE.  If the client needs
6211 	 * some other semantic, then it should do the access checking
6212 	 * itself.  It would have been nice to have the file open mode
6213 	 * passed as part of the arguments.
6214 	 */
6215 
6216 	*created = TRUE;
6217 	error = VOP_CREATE(dvp, nm, vap, EXCL, VWRITE, vpp, cr, 0, NULL, NULL);
6218 
6219 	if (error) {
6220 		*created = FALSE;
6221 
6222 		/*
6223 		 * If we got something other than file already exists
6224 		 * then just return this error.  Otherwise, we got
6225 		 * EEXIST.  If we were doing a GUARDED create, then
6226 		 * just return this error.  Otherwise, we need to
6227 		 * make sure that this wasn't a duplicate of an
6228 		 * exclusive create request.
6229 		 *
6230 		 * The assumption is made that a non-exclusive create
6231 		 * request will never return EEXIST.
6232 		 */
6233 
6234 		if (error != EEXIST || mode == GUARDED4) {
6235 			status = puterrno4(error);
6236 			return (status);
6237 		}
6238 		error = VOP_LOOKUP(dvp, nm, vpp, NULL, 0, NULL, cr,
6239 		    NULL, NULL, NULL);
6240 
6241 		if (error) {
6242 			/*
6243 			 * We couldn't find the file that we thought that
6244 			 * we just created.  So, we'll just try creating
6245 			 * it again.
6246 			 */
6247 			if (error == ENOENT)
6248 				goto tryagain;
6249 
6250 			status = puterrno4(error);
6251 			return (status);
6252 		}
6253 
6254 		if (mode == UNCHECKED4) {
6255 			/* existing object must be regular file */
6256 			if ((*vpp)->v_type != VREG) {
6257 				if ((*vpp)->v_type == VDIR)
6258 					status = NFS4ERR_ISDIR;
6259 				else if ((*vpp)->v_type == VLNK)
6260 					status = NFS4ERR_SYMLINK;
6261 				else
6262 					status = NFS4ERR_INVAL;
6263 				VN_RELE(*vpp);
6264 				return (status);
6265 			}
6266 
6267 			return (NFS4_OK);
6268 		}
6269 
6270 		/* Check for duplicate request */
6271 		ASSERT(mtime != 0);
6272 		va.va_mask = AT_MTIME;
6273 		error = VOP_GETATTR(*vpp, &va, 0, cr, NULL);
6274 		if (!error) {
6275 			/* We found the file */
6276 			if (va.va_mtime.tv_sec != mtime->tv_sec ||
6277 			    va.va_mtime.tv_nsec != mtime->tv_nsec) {
6278 				/* but its not our creation */
6279 				VN_RELE(*vpp);
6280 				return (NFS4ERR_EXIST);
6281 			}
6282 			*created = TRUE; /* retrans of create == created */
6283 			return (NFS4_OK);
6284 		}
6285 		VN_RELE(*vpp);
6286 		return (NFS4ERR_EXIST);
6287 	}
6288 
6289 	return (NFS4_OK);
6290 }
6291 
6292 static nfsstat4
6293 check_open_access(uint32_t access, struct compound_state *cs,
6294     struct svc_req *req)
6295 {
6296 	int error;
6297 	vnode_t *vp;
6298 	bool_t readonly;
6299 	cred_t *cr = cs->cr;
6300 
6301 	/* For now we don't allow mandatory locking as per V2/V3 */
6302 	if (cs->access == CS_ACCESS_DENIED || cs->mandlock) {
6303 		return (NFS4ERR_ACCESS);
6304 	}
6305 
6306 	vp = cs->vp;
6307 	ASSERT(cr != NULL && vp->v_type == VREG);
6308 
6309 	/*
6310 	 * If the file system is exported read only and we are trying
6311 	 * to open for write, then return NFS4ERR_ROFS
6312 	 */
6313 
6314 	readonly = rdonly4(req, cs);
6315 
6316 	if ((access & OPEN4_SHARE_ACCESS_WRITE) && readonly)
6317 		return (NFS4ERR_ROFS);
6318 
6319 	if (access & OPEN4_SHARE_ACCESS_READ) {
6320 		if ((VOP_ACCESS(vp, VREAD, 0, cr, NULL) != 0) &&
6321 		    (VOP_ACCESS(vp, VEXEC, 0, cr, NULL) != 0)) {
6322 			return (NFS4ERR_ACCESS);
6323 		}
6324 	}
6325 
6326 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
6327 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
6328 		if (error)
6329 			return (NFS4ERR_ACCESS);
6330 	}
6331 
6332 	return (NFS4_OK);
6333 }
6334 
6335 static nfsstat4
6336 rfs4_createfile(OPEN4args *args, struct svc_req *req, struct compound_state *cs,
6337     change_info4 *cinfo, bitmap4 *attrset, clientid4 clientid)
6338 {
6339 	struct nfs4_svgetit_arg sarg;
6340 	struct nfs4_ntov_table ntov;
6341 
6342 	bool_t ntov_table_init = FALSE;
6343 	struct statvfs64 sb;
6344 	nfsstat4 status;
6345 	vnode_t *vp;
6346 	vattr_t bva, ava, iva, cva, *vap;
6347 	vnode_t *dvp;
6348 	timespec32_t *mtime;
6349 	char *nm = NULL;
6350 	uint_t buflen;
6351 	bool_t created;
6352 	bool_t setsize = FALSE;
6353 	len_t reqsize;
6354 	int error;
6355 	bool_t trunc;
6356 	caller_context_t ct;
6357 	component4 *component;
6358 	bslabel_t *clabel;
6359 	struct sockaddr *ca;
6360 	char *name = NULL;
6361 
6362 	sarg.sbp = &sb;
6363 	sarg.is_referral = B_FALSE;
6364 
6365 	dvp = cs->vp;
6366 
6367 	/* Check if the file system is read only */
6368 	if (rdonly4(req, cs))
6369 		return (NFS4ERR_ROFS);
6370 
6371 	/* check the label of including directory */
6372 	if (is_system_labeled()) {
6373 		ASSERT(req->rq_label != NULL);
6374 		clabel = req->rq_label;
6375 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
6376 		    "got client label from request(1)",
6377 		    struct svc_req *, req);
6378 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
6379 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
6380 			    cs->exi)) {
6381 				return (NFS4ERR_ACCESS);
6382 			}
6383 		}
6384 	}
6385 
6386 	/*
6387 	 * Get the last component of path name in nm. cs will reference
6388 	 * the including directory on success.
6389 	 */
6390 	component = &args->open_claim4_u.file;
6391 	status = utf8_dir_verify(component);
6392 	if (status != NFS4_OK)
6393 		return (status);
6394 
6395 	nm = utf8_to_fn(component, &buflen, NULL);
6396 
6397 	if (nm == NULL)
6398 		return (NFS4ERR_RESOURCE);
6399 
6400 	if (buflen > MAXNAMELEN) {
6401 		kmem_free(nm, buflen);
6402 		return (NFS4ERR_NAMETOOLONG);
6403 	}
6404 
6405 	bva.va_mask = AT_TYPE|AT_CTIME|AT_SEQ;
6406 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6407 	if (error) {
6408 		kmem_free(nm, buflen);
6409 		return (puterrno4(error));
6410 	}
6411 
6412 	if (bva.va_type != VDIR) {
6413 		kmem_free(nm, buflen);
6414 		return (NFS4ERR_NOTDIR);
6415 	}
6416 
6417 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime)
6418 
6419 	switch (args->mode) {
6420 	case GUARDED4:
6421 		/*FALLTHROUGH*/
6422 	case UNCHECKED4:
6423 		nfs4_ntov_table_init(&ntov);
6424 		ntov_table_init = TRUE;
6425 
6426 		*attrset = 0;
6427 		status = do_rfs4_set_attrs(attrset,
6428 		    &args->createhow4_u.createattrs,
6429 		    cs, &sarg, &ntov, NFS4ATTR_SETIT);
6430 
6431 		if (status == NFS4_OK && (sarg.vap->va_mask & AT_TYPE) &&
6432 		    sarg.vap->va_type != VREG) {
6433 			if (sarg.vap->va_type == VDIR)
6434 				status = NFS4ERR_ISDIR;
6435 			else if (sarg.vap->va_type == VLNK)
6436 				status = NFS4ERR_SYMLINK;
6437 			else
6438 				status = NFS4ERR_INVAL;
6439 		}
6440 
6441 		if (status != NFS4_OK) {
6442 			kmem_free(nm, buflen);
6443 			nfs4_ntov_table_free(&ntov, &sarg);
6444 			*attrset = 0;
6445 			return (status);
6446 		}
6447 
6448 		vap = sarg.vap;
6449 		vap->va_type = VREG;
6450 		vap->va_mask |= AT_TYPE;
6451 
6452 		if ((vap->va_mask & AT_MODE) == 0) {
6453 			vap->va_mask |= AT_MODE;
6454 			vap->va_mode = (mode_t)0600;
6455 		}
6456 
6457 		if (vap->va_mask & AT_SIZE) {
6458 
6459 			/* Disallow create with a non-zero size */
6460 
6461 			if ((reqsize = sarg.vap->va_size) != 0) {
6462 				kmem_free(nm, buflen);
6463 				nfs4_ntov_table_free(&ntov, &sarg);
6464 				*attrset = 0;
6465 				return (NFS4ERR_INVAL);
6466 			}
6467 			setsize = TRUE;
6468 		}
6469 		break;
6470 
6471 	case EXCLUSIVE4:
6472 		/* prohibit EXCL create of named attributes */
6473 		if (dvp->v_flag & V_XATTRDIR) {
6474 			kmem_free(nm, buflen);
6475 			*attrset = 0;
6476 			return (NFS4ERR_INVAL);
6477 		}
6478 
6479 		cva.va_mask = AT_TYPE | AT_MTIME | AT_MODE;
6480 		cva.va_type = VREG;
6481 		/*
6482 		 * Ensure no time overflows. Assumes underlying
6483 		 * filesystem supports at least 32 bits.
6484 		 * Truncate nsec to usec resolution to allow valid
6485 		 * compares even if the underlying filesystem truncates.
6486 		 */
6487 		mtime = (timespec32_t *)&args->createhow4_u.createverf;
6488 		cva.va_mtime.tv_sec = mtime->tv_sec % TIME32_MAX;
6489 		cva.va_mtime.tv_nsec = (mtime->tv_nsec / 1000) * 1000;
6490 		cva.va_mode = (mode_t)0;
6491 		vap = &cva;
6492 
6493 		/*
6494 		 * For EXCL create, attrset is set to the server attr
6495 		 * used to cache the client's verifier.
6496 		 */
6497 		*attrset = FATTR4_TIME_MODIFY_MASK;
6498 		break;
6499 	}
6500 
6501 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6502 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6503 	    MAXPATHLEN  + 1);
6504 
6505 	if (name == NULL) {
6506 		kmem_free(nm, buflen);
6507 		return (NFS4ERR_SERVERFAULT);
6508 	}
6509 
6510 	status = create_vnode(dvp, name, vap, args->mode, mtime,
6511 	    cs->cr, &vp, &created);
6512 	if (nm != name)
6513 		kmem_free(name, MAXPATHLEN + 1);
6514 	kmem_free(nm, buflen);
6515 
6516 	if (status != NFS4_OK) {
6517 		if (ntov_table_init)
6518 			nfs4_ntov_table_free(&ntov, &sarg);
6519 		*attrset = 0;
6520 		return (status);
6521 	}
6522 
6523 	trunc = (setsize && !created);
6524 
6525 	if (args->mode != EXCLUSIVE4) {
6526 		bitmap4 createmask = args->createhow4_u.createattrs.attrmask;
6527 
6528 		/*
6529 		 * True verification that object was created with correct
6530 		 * attrs is impossible.  The attrs could have been changed
6531 		 * immediately after object creation.  If attributes did
6532 		 * not verify, the only recourse for the server is to
6533 		 * destroy the object.  Maybe if some attrs (like gid)
6534 		 * are set incorrectly, the object should be destroyed;
6535 		 * however, seems bad as a default policy.  Do we really
6536 		 * want to destroy an object over one of the times not
6537 		 * verifying correctly?  For these reasons, the server
6538 		 * currently sets bits in attrset for createattrs
6539 		 * that were set; however, no verification is done.
6540 		 *
6541 		 * vmask_to_nmask accounts for vattr bits set on create
6542 		 *	[do_rfs4_set_attrs() only sets resp bits for
6543 		 *	 non-vattr/vfs bits.]
6544 		 * Mask off any bits we set by default so as not to return
6545 		 * more attrset bits than were requested in createattrs
6546 		 */
6547 		if (created) {
6548 			nfs4_vmask_to_nmask(sarg.vap->va_mask, attrset);
6549 			*attrset &= createmask;
6550 		} else {
6551 			/*
6552 			 * We did not create the vnode (we tried but it
6553 			 * already existed).  In this case, the only createattr
6554 			 * that the spec allows the server to set is size,
6555 			 * and even then, it can only be set if it is 0.
6556 			 */
6557 			*attrset = 0;
6558 			if (trunc)
6559 				*attrset = FATTR4_SIZE_MASK;
6560 		}
6561 	}
6562 	if (ntov_table_init)
6563 		nfs4_ntov_table_free(&ntov, &sarg);
6564 
6565 	/*
6566 	 * Get the initial "after" sequence number, if it fails,
6567 	 * set to zero, time to before.
6568 	 */
6569 	iva.va_mask = AT_CTIME|AT_SEQ;
6570 	if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL)) {
6571 		iva.va_seq = 0;
6572 		iva.va_ctime = bva.va_ctime;
6573 	}
6574 
6575 	/*
6576 	 * create_vnode attempts to create the file exclusive,
6577 	 * if it already exists the VOP_CREATE will fail and
6578 	 * may not increase va_seq. It is atomic if
6579 	 * we haven't changed the directory, but if it has changed
6580 	 * we don't know what changed it.
6581 	 */
6582 	if (!created) {
6583 		if (bva.va_seq && iva.va_seq &&
6584 		    bva.va_seq == iva.va_seq)
6585 			cinfo->atomic = TRUE;
6586 		else
6587 			cinfo->atomic = FALSE;
6588 		NFS4_SET_FATTR4_CHANGE(cinfo->after, iva.va_ctime);
6589 	} else {
6590 		/*
6591 		 * The entry was created, we need to sync the
6592 		 * directory metadata.
6593 		 */
6594 		(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
6595 
6596 		/*
6597 		 * Get "after" change value, if it fails, simply return the
6598 		 * before value.
6599 		 */
6600 		ava.va_mask = AT_CTIME|AT_SEQ;
6601 		if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6602 			ava.va_ctime = bva.va_ctime;
6603 			ava.va_seq = 0;
6604 		}
6605 
6606 		NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6607 
6608 		/*
6609 		 * The cinfo->atomic = TRUE only if we have
6610 		 * non-zero va_seq's, and it has incremented by exactly one
6611 		 * during the create_vnode and it didn't
6612 		 * change during the VOP_FSYNC.
6613 		 */
6614 		if (bva.va_seq && iva.va_seq && ava.va_seq &&
6615 		    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
6616 			cinfo->atomic = TRUE;
6617 		else
6618 			cinfo->atomic = FALSE;
6619 	}
6620 
6621 	/* Check for mandatory locking and that the size gets set. */
6622 	cva.va_mask = AT_MODE;
6623 	if (setsize)
6624 		cva.va_mask |= AT_SIZE;
6625 
6626 	/* Assume the worst */
6627 	cs->mandlock = TRUE;
6628 
6629 	if (VOP_GETATTR(vp, &cva, 0, cs->cr, NULL) == 0) {
6630 		cs->mandlock = MANDLOCK(cs->vp, cva.va_mode);
6631 
6632 		/*
6633 		 * Truncate the file if necessary; this would be
6634 		 * the case for create over an existing file.
6635 		 */
6636 
6637 		if (trunc) {
6638 			int in_crit = 0;
6639 			rfs4_file_t *fp;
6640 			bool_t create = FALSE;
6641 
6642 			/*
6643 			 * We are writing over an existing file.
6644 			 * Check to see if we need to recall a delegation.
6645 			 */
6646 			rfs4_hold_deleg_policy();
6647 			if ((fp = rfs4_findfile(vp, NULL, &create)) != NULL) {
6648 				if (rfs4_check_delegated_byfp(FWRITE, fp,
6649 				    (reqsize == 0), FALSE, FALSE, &clientid)) {
6650 					rfs4_file_rele(fp);
6651 					rfs4_rele_deleg_policy();
6652 					VN_RELE(vp);
6653 					*attrset = 0;
6654 					return (NFS4ERR_DELAY);
6655 				}
6656 				rfs4_file_rele(fp);
6657 			}
6658 			rfs4_rele_deleg_policy();
6659 
6660 			if (nbl_need_check(vp)) {
6661 				in_crit = 1;
6662 
6663 				ASSERT(reqsize == 0);
6664 
6665 				nbl_start_crit(vp, RW_READER);
6666 				if (nbl_conflict(vp, NBL_WRITE, 0,
6667 				    cva.va_size, 0, NULL)) {
6668 					in_crit = 0;
6669 					nbl_end_crit(vp);
6670 					VN_RELE(vp);
6671 					*attrset = 0;
6672 					return (NFS4ERR_ACCESS);
6673 				}
6674 			}
6675 			ct.cc_sysid = 0;
6676 			ct.cc_pid = 0;
6677 			ct.cc_caller_id = nfs4_srv_caller_id;
6678 			ct.cc_flags = CC_DONTBLOCK;
6679 
6680 			cva.va_mask = AT_SIZE;
6681 			cva.va_size = reqsize;
6682 			(void) VOP_SETATTR(vp, &cva, 0, cs->cr, &ct);
6683 			if (in_crit)
6684 				nbl_end_crit(vp);
6685 		}
6686 	}
6687 
6688 	error = makefh4(&cs->fh, vp, cs->exi);
6689 
6690 	/*
6691 	 * Force modified data and metadata out to stable storage.
6692 	 */
6693 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
6694 
6695 	if (error) {
6696 		VN_RELE(vp);
6697 		*attrset = 0;
6698 		return (puterrno4(error));
6699 	}
6700 
6701 	/* if parent dir is attrdir, set namedattr fh flag */
6702 	if (dvp->v_flag & V_XATTRDIR)
6703 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
6704 
6705 	if (cs->vp)
6706 		VN_RELE(cs->vp);
6707 
6708 	cs->vp = vp;
6709 
6710 	/*
6711 	 * if we did not create the file, we will need to check
6712 	 * the access bits on the file
6713 	 */
6714 
6715 	if (!created) {
6716 		if (setsize)
6717 			args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
6718 		status = check_open_access(args->share_access, cs, req);
6719 		if (status != NFS4_OK)
6720 			*attrset = 0;
6721 	}
6722 	return (status);
6723 }
6724 
6725 /*ARGSUSED*/
6726 static void
6727 rfs4_do_open(struct compound_state *cs, struct svc_req *req,
6728     rfs4_openowner_t *oo, delegreq_t deleg,
6729     uint32_t access, uint32_t deny,
6730     OPEN4res *resp, int deleg_cur)
6731 {
6732 	/* XXX Currently not using req  */
6733 	rfs4_state_t *sp;
6734 	rfs4_file_t *fp;
6735 	bool_t screate = TRUE;
6736 	bool_t fcreate = TRUE;
6737 	uint32_t open_a, share_a;
6738 	uint32_t open_d, share_d;
6739 	rfs4_deleg_state_t *dsp;
6740 	sysid_t sysid;
6741 	nfsstat4 status;
6742 	caller_context_t ct;
6743 	int fflags = 0;
6744 	int recall = 0;
6745 	int err;
6746 	int first_open;
6747 
6748 	/* get the file struct and hold a lock on it during initial open */
6749 	fp = rfs4_findfile_withlock(cs->vp, &cs->fh, &fcreate);
6750 	if (fp == NULL) {
6751 		resp->status = NFS4ERR_RESOURCE;
6752 		DTRACE_PROBE1(nfss__e__do__open1, nfsstat4, resp->status);
6753 		return;
6754 	}
6755 
6756 	sp = rfs4_findstate_by_owner_file(oo, fp, &screate);
6757 	if (sp == NULL) {
6758 		resp->status = NFS4ERR_RESOURCE;
6759 		DTRACE_PROBE1(nfss__e__do__open2, nfsstat4, resp->status);
6760 		/* No need to keep any reference */
6761 		rw_exit(&fp->rf_file_rwlock);
6762 		rfs4_file_rele(fp);
6763 		return;
6764 	}
6765 
6766 	/* try to get the sysid before continuing */
6767 	if ((status = rfs4_client_sysid(oo->ro_client, &sysid)) != NFS4_OK) {
6768 		resp->status = status;
6769 		rfs4_file_rele(fp);
6770 		/* Not a fully formed open; "close" it */
6771 		if (screate == TRUE)
6772 			rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6773 		rfs4_state_rele(sp);
6774 		return;
6775 	}
6776 
6777 	/* Calculate the fflags for this OPEN. */
6778 	if (access & OPEN4_SHARE_ACCESS_READ)
6779 		fflags |= FREAD;
6780 	if (access & OPEN4_SHARE_ACCESS_WRITE)
6781 		fflags |= FWRITE;
6782 
6783 	rfs4_dbe_lock(sp->rs_dbe);
6784 
6785 	/*
6786 	 * Calculate the new deny and access mode that this open is adding to
6787 	 * the file for this open owner;
6788 	 */
6789 	open_d = (deny & ~sp->rs_open_deny);
6790 	open_a = (access & ~sp->rs_open_access);
6791 
6792 	/*
6793 	 * Calculate the new share access and share deny modes that this open
6794 	 * is adding to the file for this open owner;
6795 	 */
6796 	share_a = (access & ~sp->rs_share_access);
6797 	share_d = (deny & ~sp->rs_share_deny);
6798 
6799 	first_open = (sp->rs_open_access & OPEN4_SHARE_ACCESS_BOTH) == 0;
6800 
6801 	/*
6802 	 * Check to see the client has already sent an open for this
6803 	 * open owner on this file with the same share/deny modes.
6804 	 * If so, we don't need to check for a conflict and we don't
6805 	 * need to add another shrlock.  If not, then we need to
6806 	 * check for conflicts in deny and access before checking for
6807 	 * conflicts in delegation.  We don't want to recall a
6808 	 * delegation based on an open that will eventually fail based
6809 	 * on shares modes.
6810 	 */
6811 
6812 	if (share_a || share_d) {
6813 		if ((err = rfs4_share(sp, access, deny)) != 0) {
6814 			rfs4_dbe_unlock(sp->rs_dbe);
6815 			resp->status = err;
6816 
6817 			rfs4_file_rele(fp);
6818 			/* Not a fully formed open; "close" it */
6819 			if (screate == TRUE)
6820 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6821 			rfs4_state_rele(sp);
6822 			return;
6823 		}
6824 	}
6825 
6826 	rfs4_dbe_lock(fp->rf_dbe);
6827 
6828 	/*
6829 	 * Check to see if this file is delegated and if so, if a
6830 	 * recall needs to be done.
6831 	 */
6832 	if (rfs4_check_recall(sp, access)) {
6833 		rfs4_dbe_unlock(fp->rf_dbe);
6834 		rfs4_dbe_unlock(sp->rs_dbe);
6835 		rfs4_recall_deleg(fp, FALSE, sp->rs_owner->ro_client);
6836 		delay(NFS4_DELEGATION_CONFLICT_DELAY);
6837 		rfs4_dbe_lock(sp->rs_dbe);
6838 
6839 		/* if state closed while lock was dropped */
6840 		if (sp->rs_closed) {
6841 			if (share_a || share_d)
6842 				(void) rfs4_unshare(sp);
6843 			rfs4_dbe_unlock(sp->rs_dbe);
6844 			rfs4_file_rele(fp);
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_OLD_STATEID;
6850 			return;
6851 		}
6852 
6853 		rfs4_dbe_lock(fp->rf_dbe);
6854 		/* Let's see if the delegation was returned */
6855 		if (rfs4_check_recall(sp, access)) {
6856 			rfs4_dbe_unlock(fp->rf_dbe);
6857 			if (share_a || share_d)
6858 				(void) rfs4_unshare(sp);
6859 			rfs4_dbe_unlock(sp->rs_dbe);
6860 			rfs4_file_rele(fp);
6861 			rfs4_update_lease(sp->rs_owner->ro_client);
6862 
6863 			/* Not a fully formed open; "close" it */
6864 			if (screate == TRUE)
6865 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6866 			rfs4_state_rele(sp);
6867 			resp->status = NFS4ERR_DELAY;
6868 			return;
6869 		}
6870 	}
6871 	/*
6872 	 * the share check passed and any delegation conflict has been
6873 	 * taken care of, now call vop_open.
6874 	 * if this is the first open then call vop_open with fflags.
6875 	 * if not, call vn_open_upgrade with just the upgrade flags.
6876 	 *
6877 	 * if the file has been opened already, it will have the current
6878 	 * access mode in the state struct.  if it has no share access, then
6879 	 * this is a new open.
6880 	 *
6881 	 * However, if this is open with CLAIM_DLEGATE_CUR, then don't
6882 	 * call VOP_OPEN(), just do the open upgrade.
6883 	 */
6884 	if (first_open && !deleg_cur) {
6885 		ct.cc_sysid = sysid;
6886 		ct.cc_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
6887 		ct.cc_caller_id = nfs4_srv_caller_id;
6888 		ct.cc_flags = CC_DONTBLOCK;
6889 		err = VOP_OPEN(&cs->vp, fflags, cs->cr, &ct);
6890 		if (err) {
6891 			rfs4_dbe_unlock(fp->rf_dbe);
6892 			if (share_a || share_d)
6893 				(void) rfs4_unshare(sp);
6894 			rfs4_dbe_unlock(sp->rs_dbe);
6895 			rfs4_file_rele(fp);
6896 
6897 			/* Not a fully formed open; "close" it */
6898 			if (screate == TRUE)
6899 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6900 			rfs4_state_rele(sp);
6901 			/* check if a monitor detected a delegation conflict */
6902 			if (err == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
6903 				resp->status = NFS4ERR_DELAY;
6904 			else
6905 				resp->status = NFS4ERR_SERVERFAULT;
6906 			return;
6907 		}
6908 	} else { /* open upgrade */
6909 		/*
6910 		 * calculate the fflags for the new mode that is being added
6911 		 * by this upgrade.
6912 		 */
6913 		fflags = 0;
6914 		if (open_a & OPEN4_SHARE_ACCESS_READ)
6915 			fflags |= FREAD;
6916 		if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6917 			fflags |= FWRITE;
6918 		vn_open_upgrade(cs->vp, fflags);
6919 	}
6920 	sp->rs_open_access |= access;
6921 	sp->rs_open_deny |= deny;
6922 
6923 	if (open_d & OPEN4_SHARE_DENY_READ)
6924 		fp->rf_deny_read++;
6925 	if (open_d & OPEN4_SHARE_DENY_WRITE)
6926 		fp->rf_deny_write++;
6927 	fp->rf_share_deny |= deny;
6928 
6929 	if (open_a & OPEN4_SHARE_ACCESS_READ)
6930 		fp->rf_access_read++;
6931 	if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6932 		fp->rf_access_write++;
6933 	fp->rf_share_access |= access;
6934 
6935 	/*
6936 	 * Check for delegation here. if the deleg argument is not
6937 	 * DELEG_ANY, then this is a reclaim from a client and
6938 	 * we must honor the delegation requested. If necessary we can
6939 	 * set the recall flag.
6940 	 */
6941 
6942 	dsp = rfs4_grant_delegation(deleg, sp, &recall);
6943 
6944 	cs->deleg = (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE);
6945 
6946 	next_stateid(&sp->rs_stateid);
6947 
6948 	resp->stateid = sp->rs_stateid.stateid;
6949 
6950 	rfs4_dbe_unlock(fp->rf_dbe);
6951 	rfs4_dbe_unlock(sp->rs_dbe);
6952 
6953 	if (dsp) {
6954 		rfs4_set_deleg_response(dsp, &resp->delegation, NULL, recall);
6955 		rfs4_deleg_state_rele(dsp);
6956 	}
6957 
6958 	rfs4_file_rele(fp);
6959 	rfs4_state_rele(sp);
6960 
6961 	resp->status = NFS4_OK;
6962 }
6963 
6964 /*ARGSUSED*/
6965 static void
6966 rfs4_do_opennull(struct compound_state *cs, struct svc_req *req,
6967     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6968 {
6969 	change_info4 *cinfo = &resp->cinfo;
6970 	bitmap4 *attrset = &resp->attrset;
6971 
6972 	if (args->opentype == OPEN4_NOCREATE)
6973 		resp->status = rfs4_lookupfile(&args->open_claim4_u.file,
6974 		    req, cs, args->share_access, cinfo);
6975 	else {
6976 		/* inhibit delegation grants during exclusive create */
6977 
6978 		if (args->mode == EXCLUSIVE4)
6979 			rfs4_disable_delegation();
6980 
6981 		resp->status = rfs4_createfile(args, req, cs, cinfo, attrset,
6982 		    oo->ro_client->rc_clientid);
6983 	}
6984 
6985 	if (resp->status == NFS4_OK) {
6986 
6987 		/* cs->vp cs->fh now reference the desired file */
6988 
6989 		rfs4_do_open(cs, req, oo,
6990 		    oo->ro_need_confirm ? DELEG_NONE : DELEG_ANY,
6991 		    args->share_access, args->share_deny, resp, 0);
6992 
6993 		/*
6994 		 * If rfs4_createfile set attrset, we must
6995 		 * clear this attrset before the response is copied.
6996 		 */
6997 		if (resp->status != NFS4_OK && resp->attrset) {
6998 			resp->attrset = 0;
6999 		}
7000 	}
7001 	else
7002 		*cs->statusp = resp->status;
7003 
7004 	if (args->mode == EXCLUSIVE4)
7005 		rfs4_enable_delegation();
7006 }
7007 
7008 /*ARGSUSED*/
7009 static void
7010 rfs4_do_openprev(struct compound_state *cs, struct svc_req *req,
7011     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7012 {
7013 	change_info4 *cinfo = &resp->cinfo;
7014 	vattr_t va;
7015 	vtype_t v_type = cs->vp->v_type;
7016 	int error = 0;
7017 
7018 	/* Verify that we have a regular file */
7019 	if (v_type != VREG) {
7020 		if (v_type == VDIR)
7021 			resp->status = NFS4ERR_ISDIR;
7022 		else if (v_type == VLNK)
7023 			resp->status = NFS4ERR_SYMLINK;
7024 		else
7025 			resp->status = NFS4ERR_INVAL;
7026 		return;
7027 	}
7028 
7029 	va.va_mask = AT_MODE|AT_UID;
7030 	error = VOP_GETATTR(cs->vp, &va, 0, cs->cr, NULL);
7031 	if (error) {
7032 		resp->status = puterrno4(error);
7033 		return;
7034 	}
7035 
7036 	cs->mandlock = MANDLOCK(cs->vp, va.va_mode);
7037 
7038 	/*
7039 	 * Check if we have access to the file, Note the the file
7040 	 * could have originally been open UNCHECKED or GUARDED
7041 	 * with mode bits that will now fail, but there is nothing
7042 	 * we can really do about that except in the case that the
7043 	 * owner of the file is the one requesting the open.
7044 	 */
7045 	if (crgetuid(cs->cr) != va.va_uid) {
7046 		resp->status = check_open_access(args->share_access, cs, req);
7047 		if (resp->status != NFS4_OK) {
7048 			return;
7049 		}
7050 	}
7051 
7052 	/*
7053 	 * cinfo on a CLAIM_PREVIOUS is undefined, initialize to zero
7054 	 */
7055 	cinfo->before = 0;
7056 	cinfo->after = 0;
7057 	cinfo->atomic = FALSE;
7058 
7059 	rfs4_do_open(cs, req, oo,
7060 	    NFS4_DELEG4TYPE2REQTYPE(args->open_claim4_u.delegate_type),
7061 	    args->share_access, args->share_deny, resp, 0);
7062 }
7063 
7064 static void
7065 rfs4_do_opendelcur(struct compound_state *cs, struct svc_req *req,
7066     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7067 {
7068 	int error;
7069 	nfsstat4 status;
7070 	stateid4 stateid =
7071 	    args->open_claim4_u.delegate_cur_info.delegate_stateid;
7072 	rfs4_deleg_state_t *dsp;
7073 
7074 	/*
7075 	 * Find the state info from the stateid and confirm that the
7076 	 * file is delegated.  If the state openowner is the same as
7077 	 * the supplied openowner we're done. If not, get the file
7078 	 * info from the found state info. Use that file info to
7079 	 * create the state for this lock owner. Note solaris doen't
7080 	 * really need the pathname to find the file. We may want to
7081 	 * lookup the pathname and make sure that the vp exist and
7082 	 * matches the vp in the file structure. However it is
7083 	 * possible that the pathname nolonger exists (local process
7084 	 * unlinks the file), so this may not be that useful.
7085 	 */
7086 
7087 	status = rfs4_get_deleg_state(&stateid, &dsp);
7088 	if (status != NFS4_OK) {
7089 		resp->status = status;
7090 		return;
7091 	}
7092 
7093 	ASSERT(dsp->rds_finfo->rf_dinfo.rd_dtype != OPEN_DELEGATE_NONE);
7094 
7095 	/*
7096 	 * New lock owner, create state. Since this was probably called
7097 	 * in response to a CB_RECALL we set deleg to DELEG_NONE
7098 	 */
7099 
7100 	ASSERT(cs->vp != NULL);
7101 	VN_RELE(cs->vp);
7102 	VN_HOLD(dsp->rds_finfo->rf_vp);
7103 	cs->vp = dsp->rds_finfo->rf_vp;
7104 
7105 	if (error = makefh4(&cs->fh, cs->vp, cs->exi)) {
7106 		rfs4_deleg_state_rele(dsp);
7107 		*cs->statusp = resp->status = puterrno4(error);
7108 		return;
7109 	}
7110 
7111 	/* Mark progress for delegation returns */
7112 	dsp->rds_finfo->rf_dinfo.rd_time_lastwrite = gethrestime_sec();
7113 	rfs4_deleg_state_rele(dsp);
7114 	rfs4_do_open(cs, req, oo, DELEG_NONE,
7115 	    args->share_access, args->share_deny, resp, 1);
7116 }
7117 
7118 /*ARGSUSED*/
7119 static void
7120 rfs4_do_opendelprev(struct compound_state *cs, struct svc_req *req,
7121     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7122 {
7123 	/*
7124 	 * Lookup the pathname, it must already exist since this file
7125 	 * was delegated.
7126 	 *
7127 	 * Find the file and state info for this vp and open owner pair.
7128 	 *	check that they are in fact delegated.
7129 	 *	check that the state access and deny modes are the same.
7130 	 *
7131 	 * Return the delgation possibly seting the recall flag.
7132 	 */
7133 	rfs4_file_t *fp;
7134 	rfs4_state_t *sp;
7135 	bool_t create = FALSE;
7136 	bool_t dcreate = FALSE;
7137 	rfs4_deleg_state_t *dsp;
7138 	nfsace4 *ace;
7139 
7140 	/* Note we ignore oflags */
7141 	resp->status = rfs4_lookupfile(&args->open_claim4_u.file_delegate_prev,
7142 	    req, cs, args->share_access, &resp->cinfo);
7143 
7144 	if (resp->status != NFS4_OK) {
7145 		return;
7146 	}
7147 
7148 	/* get the file struct and hold a lock on it during initial open */
7149 	fp = rfs4_findfile_withlock(cs->vp, NULL, &create);
7150 	if (fp == NULL) {
7151 		resp->status = NFS4ERR_RESOURCE;
7152 		DTRACE_PROBE1(nfss__e__do_opendelprev1, nfsstat4, resp->status);
7153 		return;
7154 	}
7155 
7156 	sp = rfs4_findstate_by_owner_file(oo, fp, &create);
7157 	if (sp == NULL) {
7158 		resp->status = NFS4ERR_SERVERFAULT;
7159 		DTRACE_PROBE1(nfss__e__do_opendelprev2, nfsstat4, resp->status);
7160 		rw_exit(&fp->rf_file_rwlock);
7161 		rfs4_file_rele(fp);
7162 		return;
7163 	}
7164 
7165 	rfs4_dbe_lock(sp->rs_dbe);
7166 	rfs4_dbe_lock(fp->rf_dbe);
7167 	if (args->share_access != sp->rs_share_access ||
7168 	    args->share_deny != sp->rs_share_deny ||
7169 	    sp->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
7170 		NFS4_DEBUG(rfs4_debug,
7171 		    (CE_NOTE, "rfs4_do_opendelprev: state mixup"));
7172 		rfs4_dbe_unlock(fp->rf_dbe);
7173 		rfs4_dbe_unlock(sp->rs_dbe);
7174 		rfs4_file_rele(fp);
7175 		rfs4_state_rele(sp);
7176 		resp->status = NFS4ERR_SERVERFAULT;
7177 		return;
7178 	}
7179 	rfs4_dbe_unlock(fp->rf_dbe);
7180 	rfs4_dbe_unlock(sp->rs_dbe);
7181 
7182 	dsp = rfs4_finddeleg(sp, &dcreate);
7183 	if (dsp == NULL) {
7184 		rfs4_state_rele(sp);
7185 		rfs4_file_rele(fp);
7186 		resp->status = NFS4ERR_SERVERFAULT;
7187 		return;
7188 	}
7189 
7190 	next_stateid(&sp->rs_stateid);
7191 
7192 	resp->stateid = sp->rs_stateid.stateid;
7193 
7194 	resp->delegation.delegation_type = dsp->rds_dtype;
7195 
7196 	if (dsp->rds_dtype == OPEN_DELEGATE_READ) {
7197 		open_read_delegation4 *rv =
7198 		    &resp->delegation.open_delegation4_u.read;
7199 
7200 		rv->stateid = dsp->rds_delegid.stateid;
7201 		rv->recall = FALSE; /* no policy in place to set to TRUE */
7202 		ace = &rv->permissions;
7203 	} else {
7204 		open_write_delegation4 *rv =
7205 		    &resp->delegation.open_delegation4_u.write;
7206 
7207 		rv->stateid = dsp->rds_delegid.stateid;
7208 		rv->recall = FALSE;  /* no policy in place to set to TRUE */
7209 		ace = &rv->permissions;
7210 		rv->space_limit.limitby = NFS_LIMIT_SIZE;
7211 		rv->space_limit.nfs_space_limit4_u.filesize = UINT64_MAX;
7212 	}
7213 
7214 	/* XXX For now */
7215 	ace->type = ACE4_ACCESS_ALLOWED_ACE_TYPE;
7216 	ace->flag = 0;
7217 	ace->access_mask = 0;
7218 	ace->who.utf8string_len = 0;
7219 	ace->who.utf8string_val = 0;
7220 
7221 	rfs4_deleg_state_rele(dsp);
7222 	rfs4_state_rele(sp);
7223 	rfs4_file_rele(fp);
7224 }
7225 
7226 typedef enum {
7227 	NFS4_CHKSEQ_OKAY = 0,
7228 	NFS4_CHKSEQ_REPLAY = 1,
7229 	NFS4_CHKSEQ_BAD = 2
7230 } rfs4_chkseq_t;
7231 
7232 /*
7233  * Generic function for sequence number checks.
7234  */
7235 static rfs4_chkseq_t
7236 rfs4_check_seqid(seqid4 seqid, nfs_resop4 *lastop,
7237     seqid4 rqst_seq, nfs_resop4 *resop, bool_t copyres)
7238 {
7239 	/* Same sequence ids and matching operations? */
7240 	if (seqid == rqst_seq && resop->resop == lastop->resop) {
7241 		if (copyres == TRUE) {
7242 			rfs4_free_reply(resop);
7243 			rfs4_copy_reply(resop, lastop);
7244 		}
7245 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
7246 		    "Replayed SEQID %d\n", seqid));
7247 		return (NFS4_CHKSEQ_REPLAY);
7248 	}
7249 
7250 	/* If the incoming sequence is not the next expected then it is bad */
7251 	if (rqst_seq != seqid + 1) {
7252 		if (rqst_seq == seqid) {
7253 			NFS4_DEBUG(rfs4_debug,
7254 			    (CE_NOTE, "BAD SEQID: Replayed sequence id "
7255 			    "but last op was %d current op is %d\n",
7256 			    lastop->resop, resop->resop));
7257 			return (NFS4_CHKSEQ_BAD);
7258 		}
7259 		NFS4_DEBUG(rfs4_debug,
7260 		    (CE_NOTE, "BAD SEQID: got %u expecting %u\n",
7261 		    rqst_seq, seqid));
7262 		return (NFS4_CHKSEQ_BAD);
7263 	}
7264 
7265 	/* Everything okay -- next expected */
7266 	return (NFS4_CHKSEQ_OKAY);
7267 }
7268 
7269 
7270 static rfs4_chkseq_t
7271 rfs4_check_open_seqid(seqid4 seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7272 {
7273 	rfs4_chkseq_t rc;
7274 
7275 	rfs4_dbe_lock(op->ro_dbe);
7276 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply, seqid, resop,
7277 	    TRUE);
7278 	rfs4_dbe_unlock(op->ro_dbe);
7279 
7280 	if (rc == NFS4_CHKSEQ_OKAY)
7281 		rfs4_update_lease(op->ro_client);
7282 
7283 	return (rc);
7284 }
7285 
7286 static rfs4_chkseq_t
7287 rfs4_check_olo_seqid(seqid4 olo_seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7288 {
7289 	rfs4_chkseq_t rc;
7290 
7291 	rfs4_dbe_lock(op->ro_dbe);
7292 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply,
7293 	    olo_seqid, resop, FALSE);
7294 	rfs4_dbe_unlock(op->ro_dbe);
7295 
7296 	return (rc);
7297 }
7298 
7299 static rfs4_chkseq_t
7300 rfs4_check_lock_seqid(seqid4 seqid, rfs4_lo_state_t *lsp, nfs_resop4 *resop)
7301 {
7302 	rfs4_chkseq_t rc = NFS4_CHKSEQ_OKAY;
7303 
7304 	rfs4_dbe_lock(lsp->rls_dbe);
7305 	if (!lsp->rls_skip_seqid_check)
7306 		rc = rfs4_check_seqid(lsp->rls_seqid, &lsp->rls_reply, seqid,
7307 		    resop, TRUE);
7308 	rfs4_dbe_unlock(lsp->rls_dbe);
7309 
7310 	return (rc);
7311 }
7312 
7313 static void
7314 rfs4_op_open(nfs_argop4 *argop, nfs_resop4 *resop,
7315     struct svc_req *req, struct compound_state *cs)
7316 {
7317 	OPEN4args *args = &argop->nfs_argop4_u.opopen;
7318 	OPEN4res *resp = &resop->nfs_resop4_u.opopen;
7319 	open_owner4 *owner = &args->owner;
7320 	open_claim_type4 claim = args->claim;
7321 	rfs4_client_t *cp;
7322 	rfs4_openowner_t *oo;
7323 	bool_t create;
7324 	bool_t replay = FALSE;
7325 	int can_reclaim;
7326 
7327 	DTRACE_NFSV4_2(op__open__start, struct compound_state *, cs,
7328 	    OPEN4args *, args);
7329 
7330 	if (cs->vp == NULL) {
7331 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7332 		goto end;
7333 	}
7334 
7335 	/*
7336 	 * Need to check clientid and lease expiration first based on
7337 	 * error ordering and incrementing sequence id.
7338 	 */
7339 	cp = rfs4_findclient_by_id(owner->clientid, FALSE);
7340 	if (cp == NULL) {
7341 		*cs->statusp = resp->status =
7342 		    rfs4_check_clientid(&owner->clientid, 0);
7343 		goto end;
7344 	}
7345 
7346 	if (rfs4_lease_expired(cp)) {
7347 		rfs4_client_close(cp);
7348 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7349 		goto end;
7350 	}
7351 	can_reclaim = cp->rc_can_reclaim;
7352 
7353 	/*
7354 	 * Find the open_owner for use from this point forward.  Take
7355 	 * care in updating the sequence id based on the type of error
7356 	 * being returned.
7357 	 */
7358 retry:
7359 	create = TRUE;
7360 	oo = rfs4_findopenowner(owner, &create, args->seqid);
7361 	if (oo == NULL) {
7362 		*cs->statusp = resp->status = NFS4ERR_RESOURCE;
7363 		rfs4_client_rele(cp);
7364 		goto end;
7365 	}
7366 
7367 	/* Hold off access to the sequence space while the open is done */
7368 	rfs4_sw_enter(&oo->ro_sw);
7369 
7370 	/*
7371 	 * If the open_owner existed before at the server, then check
7372 	 * the sequence id.
7373 	 */
7374 	if (!create && !oo->ro_postpone_confirm) {
7375 		switch (rfs4_check_open_seqid(args->seqid, oo, resop)) {
7376 		case NFS4_CHKSEQ_BAD:
7377 			if ((args->seqid > oo->ro_open_seqid) &&
7378 			    oo->ro_need_confirm) {
7379 				rfs4_free_opens(oo, TRUE, FALSE);
7380 				rfs4_sw_exit(&oo->ro_sw);
7381 				rfs4_openowner_rele(oo);
7382 				goto retry;
7383 			}
7384 			resp->status = NFS4ERR_BAD_SEQID;
7385 			goto out;
7386 		case NFS4_CHKSEQ_REPLAY: /* replay of previous request */
7387 			replay = TRUE;
7388 			goto out;
7389 		default:
7390 			break;
7391 		}
7392 
7393 		/*
7394 		 * Sequence was ok and open owner exists
7395 		 * check to see if we have yet to see an
7396 		 * open_confirm.
7397 		 */
7398 		if (oo->ro_need_confirm) {
7399 			rfs4_free_opens(oo, TRUE, FALSE);
7400 			rfs4_sw_exit(&oo->ro_sw);
7401 			rfs4_openowner_rele(oo);
7402 			goto retry;
7403 		}
7404 	}
7405 	/* Grace only applies to regular-type OPENs */
7406 	if (rfs4_clnt_in_grace(cp) &&
7407 	    (claim == CLAIM_NULL || claim == CLAIM_DELEGATE_CUR)) {
7408 		*cs->statusp = resp->status = NFS4ERR_GRACE;
7409 		goto out;
7410 	}
7411 
7412 	/*
7413 	 * If previous state at the server existed then can_reclaim
7414 	 * will be set. If not reply NFS4ERR_NO_GRACE to the
7415 	 * client.
7416 	 */
7417 	if (rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS && !can_reclaim) {
7418 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7419 		goto out;
7420 	}
7421 
7422 
7423 	/*
7424 	 * Reject the open if the client has missed the grace period
7425 	 */
7426 	if (!rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS) {
7427 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7428 		goto out;
7429 	}
7430 
7431 	/* Couple of up-front bookkeeping items */
7432 	if (oo->ro_need_confirm) {
7433 		/*
7434 		 * If this is a reclaim OPEN then we should not ask
7435 		 * for a confirmation of the open_owner per the
7436 		 * protocol specification.
7437 		 */
7438 		if (claim == CLAIM_PREVIOUS)
7439 			oo->ro_need_confirm = FALSE;
7440 		else
7441 			resp->rflags |= OPEN4_RESULT_CONFIRM;
7442 	}
7443 	resp->rflags |= OPEN4_RESULT_LOCKTYPE_POSIX;
7444 
7445 	/*
7446 	 * If there is an unshared filesystem mounted on this vnode,
7447 	 * do not allow to open/create in this directory.
7448 	 */
7449 	if (vn_ismntpt(cs->vp)) {
7450 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
7451 		goto out;
7452 	}
7453 
7454 	/*
7455 	 * access must READ, WRITE, or BOTH.  No access is invalid.
7456 	 * deny can be READ, WRITE, BOTH, or NONE.
7457 	 * bits not defined for access/deny are invalid.
7458 	 */
7459 	if (! (args->share_access & OPEN4_SHARE_ACCESS_BOTH) ||
7460 	    (args->share_access & ~OPEN4_SHARE_ACCESS_BOTH) ||
7461 	    (args->share_deny & ~OPEN4_SHARE_DENY_BOTH)) {
7462 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7463 		goto out;
7464 	}
7465 
7466 
7467 	/*
7468 	 * make sure attrset is zero before response is built.
7469 	 */
7470 	resp->attrset = 0;
7471 
7472 	switch (claim) {
7473 	case CLAIM_NULL:
7474 		rfs4_do_opennull(cs, req, args, oo, resp);
7475 		break;
7476 	case CLAIM_PREVIOUS:
7477 		rfs4_do_openprev(cs, req, args, oo, resp);
7478 		break;
7479 	case CLAIM_DELEGATE_CUR:
7480 		rfs4_do_opendelcur(cs, req, args, oo, resp);
7481 		break;
7482 	case CLAIM_DELEGATE_PREV:
7483 		rfs4_do_opendelprev(cs, req, args, oo, resp);
7484 		break;
7485 	default:
7486 		resp->status = NFS4ERR_INVAL;
7487 		break;
7488 	}
7489 
7490 out:
7491 	rfs4_client_rele(cp);
7492 
7493 	/* Catch sequence id handling here to make it a little easier */
7494 	switch (resp->status) {
7495 	case NFS4ERR_BADXDR:
7496 	case NFS4ERR_BAD_SEQID:
7497 	case NFS4ERR_BAD_STATEID:
7498 	case NFS4ERR_NOFILEHANDLE:
7499 	case NFS4ERR_RESOURCE:
7500 	case NFS4ERR_STALE_CLIENTID:
7501 	case NFS4ERR_STALE_STATEID:
7502 		/*
7503 		 * The protocol states that if any of these errors are
7504 		 * being returned, the sequence id should not be
7505 		 * incremented.  Any other return requires an
7506 		 * increment.
7507 		 */
7508 		break;
7509 	default:
7510 		/* Always update the lease in this case */
7511 		rfs4_update_lease(oo->ro_client);
7512 
7513 		/* Regular response - copy the result */
7514 		if (!replay)
7515 			rfs4_update_open_resp(oo, resop, &cs->fh);
7516 
7517 		/*
7518 		 * REPLAY case: Only if the previous response was OK
7519 		 * do we copy the filehandle.  If not OK, no
7520 		 * filehandle to copy.
7521 		 */
7522 		if (replay == TRUE &&
7523 		    resp->status == NFS4_OK &&
7524 		    oo->ro_reply_fh.nfs_fh4_val) {
7525 			/*
7526 			 * If this is a replay, we must restore the
7527 			 * current filehandle/vp to that of what was
7528 			 * returned originally.  Try our best to do
7529 			 * it.
7530 			 */
7531 			nfs_fh4_fmt_t *fh_fmtp =
7532 			    (nfs_fh4_fmt_t *)oo->ro_reply_fh.nfs_fh4_val;
7533 
7534 			if (cs->exi)
7535 				exi_rele(cs->exi);
7536 			cs->exi = checkexport(&fh_fmtp->fh4_fsid,
7537 			    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
7538 
7539 			if (cs->exi == NULL) {
7540 				resp->status = NFS4ERR_STALE;
7541 				goto finish;
7542 			}
7543 
7544 			VN_RELE(cs->vp);
7545 
7546 			cs->vp = nfs4_fhtovp(&oo->ro_reply_fh, cs->exi,
7547 			    &resp->status);
7548 
7549 			if (cs->vp == NULL)
7550 				goto finish;
7551 
7552 			nfs_fh4_copy(&oo->ro_reply_fh, &cs->fh);
7553 		}
7554 
7555 		/*
7556 		 * If this was a replay, no need to update the
7557 		 * sequence id. If the open_owner was not created on
7558 		 * this pass, then update.  The first use of an
7559 		 * open_owner will not bump the sequence id.
7560 		 */
7561 		if (replay == FALSE && !create)
7562 			rfs4_update_open_sequence(oo);
7563 		/*
7564 		 * If the client is receiving an error and the
7565 		 * open_owner needs to be confirmed, there is no way
7566 		 * to notify the client of this fact ignoring the fact
7567 		 * that the server has no method of returning a
7568 		 * stateid to confirm.  Therefore, the server needs to
7569 		 * mark this open_owner in a way as to avoid the
7570 		 * sequence id checking the next time the client uses
7571 		 * this open_owner.
7572 		 */
7573 		if (resp->status != NFS4_OK && oo->ro_need_confirm)
7574 			oo->ro_postpone_confirm = TRUE;
7575 		/*
7576 		 * If OK response then clear the postpone flag and
7577 		 * reset the sequence id to keep in sync with the
7578 		 * client.
7579 		 */
7580 		if (resp->status == NFS4_OK && oo->ro_postpone_confirm) {
7581 			oo->ro_postpone_confirm = FALSE;
7582 			oo->ro_open_seqid = args->seqid;
7583 		}
7584 		break;
7585 	}
7586 
7587 finish:
7588 	*cs->statusp = resp->status;
7589 
7590 	rfs4_sw_exit(&oo->ro_sw);
7591 	rfs4_openowner_rele(oo);
7592 
7593 end:
7594 	DTRACE_NFSV4_2(op__open__done, struct compound_state *, cs,
7595 	    OPEN4res *, resp);
7596 }
7597 
7598 /*ARGSUSED*/
7599 void
7600 rfs4_op_open_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
7601     struct svc_req *req, struct compound_state *cs)
7602 {
7603 	OPEN_CONFIRM4args *args = &argop->nfs_argop4_u.opopen_confirm;
7604 	OPEN_CONFIRM4res *resp = &resop->nfs_resop4_u.opopen_confirm;
7605 	rfs4_state_t *sp;
7606 	nfsstat4 status;
7607 
7608 	DTRACE_NFSV4_2(op__open__confirm__start, struct compound_state *, cs,
7609 	    OPEN_CONFIRM4args *, args);
7610 
7611 	if (cs->vp == NULL) {
7612 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7613 		goto out;
7614 	}
7615 
7616 	if (cs->vp->v_type != VREG) {
7617 		*cs->statusp = resp->status =
7618 		    cs->vp->v_type == VDIR ? NFS4ERR_ISDIR : NFS4ERR_INVAL;
7619 		return;
7620 	}
7621 
7622 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7623 	if (status != NFS4_OK) {
7624 		*cs->statusp = resp->status = status;
7625 		goto out;
7626 	}
7627 
7628 	/* Ensure specified filehandle matches */
7629 	if (cs->vp != sp->rs_finfo->rf_vp) {
7630 		rfs4_state_rele(sp);
7631 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7632 		goto out;
7633 	}
7634 
7635 	/* hold off other access to open_owner while we tinker */
7636 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7637 
7638 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7639 	case NFS4_CHECK_STATEID_OKAY:
7640 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7641 		    resop) != 0) {
7642 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7643 			break;
7644 		}
7645 		/*
7646 		 * If it is the appropriate stateid and determined to
7647 		 * be "OKAY" then this means that the stateid does not
7648 		 * need to be confirmed and the client is in error for
7649 		 * sending an OPEN_CONFIRM.
7650 		 */
7651 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7652 		break;
7653 	case NFS4_CHECK_STATEID_OLD:
7654 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7655 		break;
7656 	case NFS4_CHECK_STATEID_BAD:
7657 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7658 		break;
7659 	case NFS4_CHECK_STATEID_EXPIRED:
7660 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7661 		break;
7662 	case NFS4_CHECK_STATEID_CLOSED:
7663 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7664 		break;
7665 	case NFS4_CHECK_STATEID_REPLAY:
7666 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7667 		    resop)) {
7668 		case NFS4_CHKSEQ_OKAY:
7669 			/*
7670 			 * This is replayed stateid; if seqid matches
7671 			 * next expected, then client is using wrong seqid.
7672 			 */
7673 			/* fall through */
7674 		case NFS4_CHKSEQ_BAD:
7675 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7676 			break;
7677 		case NFS4_CHKSEQ_REPLAY:
7678 			/*
7679 			 * Note this case is the duplicate case so
7680 			 * resp->status is already set.
7681 			 */
7682 			*cs->statusp = resp->status;
7683 			rfs4_update_lease(sp->rs_owner->ro_client);
7684 			break;
7685 		}
7686 		break;
7687 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7688 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7689 		    resop) != NFS4_CHKSEQ_OKAY) {
7690 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7691 			break;
7692 		}
7693 		*cs->statusp = resp->status = NFS4_OK;
7694 
7695 		next_stateid(&sp->rs_stateid);
7696 		resp->open_stateid = sp->rs_stateid.stateid;
7697 		sp->rs_owner->ro_need_confirm = FALSE;
7698 		rfs4_update_lease(sp->rs_owner->ro_client);
7699 		rfs4_update_open_sequence(sp->rs_owner);
7700 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7701 		break;
7702 	default:
7703 		ASSERT(FALSE);
7704 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7705 		break;
7706 	}
7707 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7708 	rfs4_state_rele(sp);
7709 
7710 out:
7711 	DTRACE_NFSV4_2(op__open__confirm__done, struct compound_state *, cs,
7712 	    OPEN_CONFIRM4res *, resp);
7713 }
7714 
7715 /*ARGSUSED*/
7716 void
7717 rfs4_op_open_downgrade(nfs_argop4 *argop, nfs_resop4 *resop,
7718     struct svc_req *req, struct compound_state *cs)
7719 {
7720 	OPEN_DOWNGRADE4args *args = &argop->nfs_argop4_u.opopen_downgrade;
7721 	OPEN_DOWNGRADE4res *resp = &resop->nfs_resop4_u.opopen_downgrade;
7722 	uint32_t access = args->share_access;
7723 	uint32_t deny = args->share_deny;
7724 	nfsstat4 status;
7725 	rfs4_state_t *sp;
7726 	rfs4_file_t *fp;
7727 	int fflags = 0;
7728 
7729 	DTRACE_NFSV4_2(op__open__downgrade__start, struct compound_state *, cs,
7730 	    OPEN_DOWNGRADE4args *, args);
7731 
7732 	if (cs->vp == NULL) {
7733 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7734 		goto out;
7735 	}
7736 
7737 	if (cs->vp->v_type != VREG) {
7738 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7739 		return;
7740 	}
7741 
7742 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7743 	if (status != NFS4_OK) {
7744 		*cs->statusp = resp->status = status;
7745 		goto out;
7746 	}
7747 
7748 	/* Ensure specified filehandle matches */
7749 	if (cs->vp != sp->rs_finfo->rf_vp) {
7750 		rfs4_state_rele(sp);
7751 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7752 		goto out;
7753 	}
7754 
7755 	/* hold off other access to open_owner while we tinker */
7756 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7757 
7758 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7759 	case NFS4_CHECK_STATEID_OKAY:
7760 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7761 		    resop) != NFS4_CHKSEQ_OKAY) {
7762 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7763 			goto end;
7764 		}
7765 		break;
7766 	case NFS4_CHECK_STATEID_OLD:
7767 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7768 		goto end;
7769 	case NFS4_CHECK_STATEID_BAD:
7770 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7771 		goto end;
7772 	case NFS4_CHECK_STATEID_EXPIRED:
7773 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7774 		goto end;
7775 	case NFS4_CHECK_STATEID_CLOSED:
7776 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7777 		goto end;
7778 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7779 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7780 		goto end;
7781 	case NFS4_CHECK_STATEID_REPLAY:
7782 		/* Check the sequence id for the open owner */
7783 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7784 		    resop)) {
7785 		case NFS4_CHKSEQ_OKAY:
7786 			/*
7787 			 * This is replayed stateid; if seqid matches
7788 			 * next expected, then client is using wrong seqid.
7789 			 */
7790 			/* fall through */
7791 		case NFS4_CHKSEQ_BAD:
7792 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7793 			goto end;
7794 		case NFS4_CHKSEQ_REPLAY:
7795 			/*
7796 			 * Note this case is the duplicate case so
7797 			 * resp->status is already set.
7798 			 */
7799 			*cs->statusp = resp->status;
7800 			rfs4_update_lease(sp->rs_owner->ro_client);
7801 			goto end;
7802 		}
7803 		break;
7804 	default:
7805 		ASSERT(FALSE);
7806 		break;
7807 	}
7808 
7809 	rfs4_dbe_lock(sp->rs_dbe);
7810 	/*
7811 	 * Check that the new access modes and deny modes are valid.
7812 	 * Check that no invalid bits are set.
7813 	 */
7814 	if ((access & ~(OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) ||
7815 	    (deny & ~(OPEN4_SHARE_DENY_READ | OPEN4_SHARE_DENY_WRITE))) {
7816 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7817 		rfs4_update_open_sequence(sp->rs_owner);
7818 		rfs4_dbe_unlock(sp->rs_dbe);
7819 		goto end;
7820 	}
7821 
7822 	/*
7823 	 * The new modes must be a subset of the current modes and
7824 	 * the access must specify at least one mode. To test that
7825 	 * the new mode is a subset of the current modes we bitwise
7826 	 * AND them together and check that the result equals the new
7827 	 * mode. For example:
7828 	 * New mode, access == R and current mode, sp->rs_open_access  == RW
7829 	 * access & sp->rs_open_access == R == access, so the new access mode
7830 	 * is valid. Consider access == RW, sp->rs_open_access = R
7831 	 * access & sp->rs_open_access == R != access, so the new access mode
7832 	 * is invalid.
7833 	 */
7834 	if ((access & sp->rs_open_access) != access ||
7835 	    (deny & sp->rs_open_deny) != deny ||
7836 	    (access &
7837 	    (OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) == 0) {
7838 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7839 		rfs4_update_open_sequence(sp->rs_owner);
7840 		rfs4_dbe_unlock(sp->rs_dbe);
7841 		goto end;
7842 	}
7843 
7844 	/*
7845 	 * Release any share locks associated with this stateID.
7846 	 * Strictly speaking, this violates the spec because the
7847 	 * spec effectively requires that open downgrade be atomic.
7848 	 * At present, fs_shrlock does not have this capability.
7849 	 */
7850 	(void) rfs4_unshare(sp);
7851 
7852 	status = rfs4_share(sp, access, deny);
7853 	if (status != NFS4_OK) {
7854 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7855 		rfs4_update_open_sequence(sp->rs_owner);
7856 		rfs4_dbe_unlock(sp->rs_dbe);
7857 		goto end;
7858 	}
7859 
7860 	fp = sp->rs_finfo;
7861 	rfs4_dbe_lock(fp->rf_dbe);
7862 
7863 	/*
7864 	 * If the current mode has deny read and the new mode
7865 	 * does not, decrement the number of deny read mode bits
7866 	 * and if it goes to zero turn off the deny read bit
7867 	 * on the file.
7868 	 */
7869 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_READ) &&
7870 	    (deny & OPEN4_SHARE_DENY_READ) == 0) {
7871 		fp->rf_deny_read--;
7872 		if (fp->rf_deny_read == 0)
7873 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
7874 	}
7875 
7876 	/*
7877 	 * If the current mode has deny write and the new mode
7878 	 * does not, decrement the number of deny write mode bits
7879 	 * and if it goes to zero turn off the deny write bit
7880 	 * on the file.
7881 	 */
7882 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) &&
7883 	    (deny & OPEN4_SHARE_DENY_WRITE) == 0) {
7884 		fp->rf_deny_write--;
7885 		if (fp->rf_deny_write == 0)
7886 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
7887 	}
7888 
7889 	/*
7890 	 * If the current mode has access read and the new mode
7891 	 * does not, decrement the number of access read mode bits
7892 	 * and if it goes to zero turn off the access read bit
7893 	 * on the file.  set fflags to FREAD for the call to
7894 	 * vn_open_downgrade().
7895 	 */
7896 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) &&
7897 	    (access & OPEN4_SHARE_ACCESS_READ) == 0) {
7898 		fp->rf_access_read--;
7899 		if (fp->rf_access_read == 0)
7900 			fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
7901 		fflags |= FREAD;
7902 	}
7903 
7904 	/*
7905 	 * If the current mode has access write and the new mode
7906 	 * does not, decrement the number of access write mode bits
7907 	 * and if it goes to zero turn off the access write bit
7908 	 * on the file.  set fflags to FWRITE for the call to
7909 	 * vn_open_downgrade().
7910 	 */
7911 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) &&
7912 	    (access & OPEN4_SHARE_ACCESS_WRITE) == 0) {
7913 		fp->rf_access_write--;
7914 		if (fp->rf_access_write == 0)
7915 			fp->rf_share_deny &= ~OPEN4_SHARE_ACCESS_WRITE;
7916 		fflags |= FWRITE;
7917 	}
7918 
7919 	/* Check that the file is still accessible */
7920 	ASSERT(fp->rf_share_access);
7921 
7922 	rfs4_dbe_unlock(fp->rf_dbe);
7923 
7924 	/* now set the new open access and deny modes */
7925 	sp->rs_open_access = access;
7926 	sp->rs_open_deny = deny;
7927 
7928 	/*
7929 	 * we successfully downgraded the share lock, now we need to downgrade
7930 	 * the open. it is possible that the downgrade was only for a deny
7931 	 * mode and we have nothing else to do.
7932 	 */
7933 	if ((fflags & (FREAD|FWRITE)) != 0)
7934 		vn_open_downgrade(cs->vp, fflags);
7935 
7936 	/* Update the stateid */
7937 	next_stateid(&sp->rs_stateid);
7938 	resp->open_stateid = sp->rs_stateid.stateid;
7939 
7940 	rfs4_dbe_unlock(sp->rs_dbe);
7941 
7942 	*cs->statusp = resp->status = NFS4_OK;
7943 	/* Update the lease */
7944 	rfs4_update_lease(sp->rs_owner->ro_client);
7945 	/* And the sequence */
7946 	rfs4_update_open_sequence(sp->rs_owner);
7947 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7948 
7949 end:
7950 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7951 	rfs4_state_rele(sp);
7952 out:
7953 	DTRACE_NFSV4_2(op__open__downgrade__done, struct compound_state *, cs,
7954 	    OPEN_DOWNGRADE4res *, resp);
7955 }
7956 
7957 static void *
7958 memstr(const void *s1, const char *s2, size_t n)
7959 {
7960 	size_t l = strlen(s2);
7961 	char *p = (char *)s1;
7962 
7963 	while (n >= l) {
7964 		if (bcmp(p, s2, l) == 0)
7965 			return (p);
7966 		p++;
7967 		n--;
7968 	}
7969 
7970 	return (NULL);
7971 }
7972 
7973 /*
7974  * The logic behind this function is detailed in the NFSv4 RFC in the
7975  * SETCLIENTID operation description under IMPLEMENTATION.  Refer to
7976  * that section for explicit guidance to server behavior for
7977  * SETCLIENTID.
7978  */
7979 void
7980 rfs4_op_setclientid(nfs_argop4 *argop, nfs_resop4 *resop,
7981     struct svc_req *req, struct compound_state *cs)
7982 {
7983 	SETCLIENTID4args *args = &argop->nfs_argop4_u.opsetclientid;
7984 	SETCLIENTID4res *res = &resop->nfs_resop4_u.opsetclientid;
7985 	rfs4_client_t *cp, *newcp, *cp_confirmed, *cp_unconfirmed;
7986 	rfs4_clntip_t *ci;
7987 	bool_t create;
7988 	char *addr, *netid;
7989 	int len;
7990 
7991 	DTRACE_NFSV4_2(op__setclientid__start, struct compound_state *, cs,
7992 	    SETCLIENTID4args *, args);
7993 retry:
7994 	newcp = cp_confirmed = cp_unconfirmed = NULL;
7995 
7996 	/*
7997 	 * Save the caller's IP address
7998 	 */
7999 	args->client.cl_addr =
8000 	    (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
8001 
8002 	/*
8003 	 * Record if it is a Solaris client that cannot handle referrals.
8004 	 */
8005 	if (memstr(args->client.id_val, "Solaris", args->client.id_len) &&
8006 	    !memstr(args->client.id_val, "+referrals", args->client.id_len)) {
8007 		/* Add a "yes, it's downrev" record */
8008 		create = TRUE;
8009 		ci = rfs4_find_clntip(args->client.cl_addr, &create);
8010 		ASSERT(ci != NULL);
8011 		rfs4_dbe_rele(ci->ri_dbe);
8012 	} else {
8013 		/* Remove any previous record */
8014 		rfs4_invalidate_clntip(args->client.cl_addr);
8015 	}
8016 
8017 	/*
8018 	 * In search of an EXISTING client matching the incoming
8019 	 * request to establish a new client identifier at the server
8020 	 */
8021 	create = TRUE;
8022 	cp = rfs4_findclient(&args->client, &create, NULL);
8023 
8024 	/* Should never happen */
8025 	ASSERT(cp != NULL);
8026 
8027 	if (cp == NULL) {
8028 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8029 		goto out;
8030 	}
8031 
8032 	/*
8033 	 * Easiest case. Client identifier is newly created and is
8034 	 * unconfirmed.  Also note that for this case, no other
8035 	 * entries exist for the client identifier.  Nothing else to
8036 	 * check.  Just setup the response and respond.
8037 	 */
8038 	if (create) {
8039 		*cs->statusp = res->status = NFS4_OK;
8040 		res->SETCLIENTID4res_u.resok4.clientid = cp->rc_clientid;
8041 		res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8042 		    cp->rc_confirm_verf;
8043 		/* Setup callback information; CB_NULL confirmation later */
8044 		rfs4_client_setcb(cp, &args->callback, args->callback_ident);
8045 
8046 		rfs4_client_rele(cp);
8047 		goto out;
8048 	}
8049 
8050 	/*
8051 	 * An existing, confirmed client may exist but it may not have
8052 	 * been active for at least one lease period.  If so, then
8053 	 * "close" the client and create a new client identifier
8054 	 */
8055 	if (rfs4_lease_expired(cp)) {
8056 		rfs4_client_close(cp);
8057 		goto retry;
8058 	}
8059 
8060 	if (cp->rc_need_confirm == TRUE)
8061 		cp_unconfirmed = cp;
8062 	else
8063 		cp_confirmed = cp;
8064 
8065 	cp = NULL;
8066 
8067 	/*
8068 	 * We have a confirmed client, now check for an
8069 	 * unconfimred entry
8070 	 */
8071 	if (cp_confirmed) {
8072 		/* If creds don't match then client identifier is inuse */
8073 		if (!creds_ok(cp_confirmed->rc_cr_set, req, cs)) {
8074 			rfs4_cbinfo_t *cbp;
8075 			/*
8076 			 * Some one else has established this client
8077 			 * id. Try and say * who they are. We will use
8078 			 * the call back address supplied by * the
8079 			 * first client.
8080 			 */
8081 			*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8082 
8083 			addr = netid = NULL;
8084 
8085 			cbp = &cp_confirmed->rc_cbinfo;
8086 			if (cbp->cb_callback.cb_location.r_addr &&
8087 			    cbp->cb_callback.cb_location.r_netid) {
8088 				cb_client4 *cbcp = &cbp->cb_callback;
8089 
8090 				len = strlen(cbcp->cb_location.r_addr)+1;
8091 				addr = kmem_alloc(len, KM_SLEEP);
8092 				bcopy(cbcp->cb_location.r_addr, addr, len);
8093 				len = strlen(cbcp->cb_location.r_netid)+1;
8094 				netid = kmem_alloc(len, KM_SLEEP);
8095 				bcopy(cbcp->cb_location.r_netid, netid, len);
8096 			}
8097 
8098 			res->SETCLIENTID4res_u.client_using.r_addr = addr;
8099 			res->SETCLIENTID4res_u.client_using.r_netid = netid;
8100 
8101 			rfs4_client_rele(cp_confirmed);
8102 		}
8103 
8104 		/*
8105 		 * Confirmed, creds match, and verifier matches; must
8106 		 * be an update of the callback info
8107 		 */
8108 		if (cp_confirmed->rc_nfs_client.verifier ==
8109 		    args->client.verifier) {
8110 			/* Setup callback information */
8111 			rfs4_client_setcb(cp_confirmed, &args->callback,
8112 			    args->callback_ident);
8113 
8114 			/* everything okay -- move ahead */
8115 			*cs->statusp = res->status = NFS4_OK;
8116 			res->SETCLIENTID4res_u.resok4.clientid =
8117 			    cp_confirmed->rc_clientid;
8118 
8119 			/* update the confirm_verifier and return it */
8120 			rfs4_client_scv_next(cp_confirmed);
8121 			res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8122 			    cp_confirmed->rc_confirm_verf;
8123 
8124 			rfs4_client_rele(cp_confirmed);
8125 			goto out;
8126 		}
8127 
8128 		/*
8129 		 * Creds match but the verifier doesn't.  Must search
8130 		 * for an unconfirmed client that would be replaced by
8131 		 * this request.
8132 		 */
8133 		create = FALSE;
8134 		cp_unconfirmed = rfs4_findclient(&args->client, &create,
8135 		    cp_confirmed);
8136 	}
8137 
8138 	/*
8139 	 * At this point, we have taken care of the brand new client
8140 	 * struct, INUSE case, update of an existing, and confirmed
8141 	 * client struct.
8142 	 */
8143 
8144 	/*
8145 	 * check to see if things have changed while we originally
8146 	 * picked up the client struct.  If they have, then return and
8147 	 * retry the processing of this SETCLIENTID request.
8148 	 */
8149 	if (cp_unconfirmed) {
8150 		rfs4_dbe_lock(cp_unconfirmed->rc_dbe);
8151 		if (!cp_unconfirmed->rc_need_confirm) {
8152 			rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8153 			rfs4_client_rele(cp_unconfirmed);
8154 			if (cp_confirmed)
8155 				rfs4_client_rele(cp_confirmed);
8156 			goto retry;
8157 		}
8158 		/* do away with the old unconfirmed one */
8159 		rfs4_dbe_invalidate(cp_unconfirmed->rc_dbe);
8160 		rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8161 		rfs4_client_rele(cp_unconfirmed);
8162 		cp_unconfirmed = NULL;
8163 	}
8164 
8165 	/*
8166 	 * This search will temporarily hide the confirmed client
8167 	 * struct while a new client struct is created as the
8168 	 * unconfirmed one.
8169 	 */
8170 	create = TRUE;
8171 	newcp = rfs4_findclient(&args->client, &create, cp_confirmed);
8172 
8173 	ASSERT(newcp != NULL);
8174 
8175 	if (newcp == NULL) {
8176 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8177 		rfs4_client_rele(cp_confirmed);
8178 		goto out;
8179 	}
8180 
8181 	/*
8182 	 * If one was not created, then a similar request must be in
8183 	 * process so release and start over with this one
8184 	 */
8185 	if (create != TRUE) {
8186 		rfs4_client_rele(newcp);
8187 		if (cp_confirmed)
8188 			rfs4_client_rele(cp_confirmed);
8189 		goto retry;
8190 	}
8191 
8192 	*cs->statusp = res->status = NFS4_OK;
8193 	res->SETCLIENTID4res_u.resok4.clientid = newcp->rc_clientid;
8194 	res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8195 	    newcp->rc_confirm_verf;
8196 	/* Setup callback information; CB_NULL confirmation later */
8197 	rfs4_client_setcb(newcp, &args->callback, args->callback_ident);
8198 
8199 	newcp->rc_cp_confirmed = cp_confirmed;
8200 
8201 	rfs4_client_rele(newcp);
8202 
8203 out:
8204 	DTRACE_NFSV4_2(op__setclientid__done, struct compound_state *, cs,
8205 	    SETCLIENTID4res *, res);
8206 }
8207 
8208 /*ARGSUSED*/
8209 void
8210 rfs4_op_setclientid_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
8211     struct svc_req *req, struct compound_state *cs)
8212 {
8213 	SETCLIENTID_CONFIRM4args *args =
8214 	    &argop->nfs_argop4_u.opsetclientid_confirm;
8215 	SETCLIENTID_CONFIRM4res *res =
8216 	    &resop->nfs_resop4_u.opsetclientid_confirm;
8217 	rfs4_client_t *cp, *cptoclose = NULL;
8218 
8219 	DTRACE_NFSV4_2(op__setclientid__confirm__start,
8220 	    struct compound_state *, cs,
8221 	    SETCLIENTID_CONFIRM4args *, args);
8222 
8223 	*cs->statusp = res->status = NFS4_OK;
8224 
8225 	cp = rfs4_findclient_by_id(args->clientid, TRUE);
8226 
8227 	if (cp == NULL) {
8228 		*cs->statusp = res->status =
8229 		    rfs4_check_clientid(&args->clientid, 1);
8230 		goto out;
8231 	}
8232 
8233 	if (!creds_ok(cp, req, cs)) {
8234 		*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8235 		rfs4_client_rele(cp);
8236 		goto out;
8237 	}
8238 
8239 	/* If the verifier doesn't match, the record doesn't match */
8240 	if (cp->rc_confirm_verf != args->setclientid_confirm) {
8241 		*cs->statusp = res->status = NFS4ERR_STALE_CLIENTID;
8242 		rfs4_client_rele(cp);
8243 		goto out;
8244 	}
8245 
8246 	rfs4_dbe_lock(cp->rc_dbe);
8247 	cp->rc_need_confirm = FALSE;
8248 	if (cp->rc_cp_confirmed) {
8249 		cptoclose = cp->rc_cp_confirmed;
8250 		cptoclose->rc_ss_remove = 1;
8251 		cp->rc_cp_confirmed = NULL;
8252 	}
8253 
8254 	/*
8255 	 * Update the client's associated server instance, if it's changed
8256 	 * since the client was created.
8257 	 */
8258 	if (rfs4_servinst(cp) != rfs4_cur_servinst)
8259 		rfs4_servinst_assign(cp, rfs4_cur_servinst);
8260 
8261 	/*
8262 	 * Record clientid in stable storage.
8263 	 * Must be done after server instance has been assigned.
8264 	 */
8265 	rfs4_ss_clid(cp);
8266 
8267 	rfs4_dbe_unlock(cp->rc_dbe);
8268 
8269 	if (cptoclose)
8270 		/* don't need to rele, client_close does it */
8271 		rfs4_client_close(cptoclose);
8272 
8273 	/* If needed, initiate CB_NULL call for callback path */
8274 	rfs4_deleg_cb_check(cp);
8275 	rfs4_update_lease(cp);
8276 
8277 	/*
8278 	 * Check to see if client can perform reclaims
8279 	 */
8280 	rfs4_ss_chkclid(cp);
8281 
8282 	rfs4_client_rele(cp);
8283 
8284 out:
8285 	DTRACE_NFSV4_2(op__setclientid__confirm__done,
8286 	    struct compound_state *, cs,
8287 	    SETCLIENTID_CONFIRM4 *, res);
8288 }
8289 
8290 
8291 /*ARGSUSED*/
8292 void
8293 rfs4_op_close(nfs_argop4 *argop, nfs_resop4 *resop,
8294     struct svc_req *req, struct compound_state *cs)
8295 {
8296 	CLOSE4args *args = &argop->nfs_argop4_u.opclose;
8297 	CLOSE4res *resp = &resop->nfs_resop4_u.opclose;
8298 	rfs4_state_t *sp;
8299 	nfsstat4 status;
8300 
8301 	DTRACE_NFSV4_2(op__close__start, struct compound_state *, cs,
8302 	    CLOSE4args *, args);
8303 
8304 	if (cs->vp == NULL) {
8305 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8306 		goto out;
8307 	}
8308 
8309 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_INVALID);
8310 	if (status != NFS4_OK) {
8311 		*cs->statusp = resp->status = status;
8312 		goto out;
8313 	}
8314 
8315 	/* Ensure specified filehandle matches */
8316 	if (cs->vp != sp->rs_finfo->rf_vp) {
8317 		rfs4_state_rele(sp);
8318 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8319 		goto out;
8320 	}
8321 
8322 	/* hold off other access to open_owner while we tinker */
8323 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
8324 
8325 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
8326 	case NFS4_CHECK_STATEID_OKAY:
8327 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8328 		    resop) != NFS4_CHKSEQ_OKAY) {
8329 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8330 			goto end;
8331 		}
8332 		break;
8333 	case NFS4_CHECK_STATEID_OLD:
8334 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8335 		goto end;
8336 	case NFS4_CHECK_STATEID_BAD:
8337 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8338 		goto end;
8339 	case NFS4_CHECK_STATEID_EXPIRED:
8340 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8341 		goto end;
8342 	case NFS4_CHECK_STATEID_CLOSED:
8343 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8344 		goto end;
8345 	case NFS4_CHECK_STATEID_UNCONFIRMED:
8346 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8347 		goto end;
8348 	case NFS4_CHECK_STATEID_REPLAY:
8349 		/* Check the sequence id for the open owner */
8350 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8351 		    resop)) {
8352 		case NFS4_CHKSEQ_OKAY:
8353 			/*
8354 			 * This is replayed stateid; if seqid matches
8355 			 * next expected, then client is using wrong seqid.
8356 			 */
8357 			/* FALL THROUGH */
8358 		case NFS4_CHKSEQ_BAD:
8359 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8360 			goto end;
8361 		case NFS4_CHKSEQ_REPLAY:
8362 			/*
8363 			 * Note this case is the duplicate case so
8364 			 * resp->status is already set.
8365 			 */
8366 			*cs->statusp = resp->status;
8367 			rfs4_update_lease(sp->rs_owner->ro_client);
8368 			goto end;
8369 		}
8370 		break;
8371 	default:
8372 		ASSERT(FALSE);
8373 		break;
8374 	}
8375 
8376 	rfs4_dbe_lock(sp->rs_dbe);
8377 
8378 	/* Update the stateid. */
8379 	next_stateid(&sp->rs_stateid);
8380 	resp->open_stateid = sp->rs_stateid.stateid;
8381 
8382 	rfs4_dbe_unlock(sp->rs_dbe);
8383 
8384 	rfs4_update_lease(sp->rs_owner->ro_client);
8385 	rfs4_update_open_sequence(sp->rs_owner);
8386 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8387 
8388 	rfs4_state_close(sp, FALSE, FALSE, cs->cr);
8389 
8390 	*cs->statusp = resp->status = status;
8391 
8392 end:
8393 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
8394 	rfs4_state_rele(sp);
8395 out:
8396 	DTRACE_NFSV4_2(op__close__done, struct compound_state *, cs,
8397 	    CLOSE4res *, resp);
8398 }
8399 
8400 /*
8401  * Manage the counts on the file struct and close all file locks
8402  */
8403 /*ARGSUSED*/
8404 void
8405 rfs4_release_share_lock_state(rfs4_state_t *sp, cred_t *cr,
8406     bool_t close_of_client)
8407 {
8408 	rfs4_file_t *fp = sp->rs_finfo;
8409 	rfs4_lo_state_t *lsp;
8410 	int fflags = 0;
8411 
8412 	/*
8413 	 * If this call is part of the larger closing down of client
8414 	 * state then it is just easier to release all locks
8415 	 * associated with this client instead of going through each
8416 	 * individual file and cleaning locks there.
8417 	 */
8418 	if (close_of_client) {
8419 		if (sp->rs_owner->ro_client->rc_unlksys_completed == FALSE &&
8420 		    !list_is_empty(&sp->rs_lostatelist) &&
8421 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID) {
8422 			/* Is the PxFS kernel module loaded? */
8423 			if (lm_remove_file_locks != NULL) {
8424 				int new_sysid;
8425 
8426 				/* Encode the cluster nodeid in new sysid */
8427 				new_sysid = sp->rs_owner->ro_client->rc_sysidt;
8428 				lm_set_nlmid_flk(&new_sysid);
8429 
8430 				/*
8431 				 * This PxFS routine removes file locks for a
8432 				 * client over all nodes of a cluster.
8433 				 */
8434 				NFS4_DEBUG(rfs4_debug, (CE_NOTE,
8435 				    "lm_remove_file_locks(sysid=0x%x)\n",
8436 				    new_sysid));
8437 				(*lm_remove_file_locks)(new_sysid);
8438 			} else {
8439 				struct flock64 flk;
8440 
8441 				/* Release all locks for this client */
8442 				flk.l_type = F_UNLKSYS;
8443 				flk.l_whence = 0;
8444 				flk.l_start = 0;
8445 				flk.l_len = 0;
8446 				flk.l_sysid =
8447 				    sp->rs_owner->ro_client->rc_sysidt;
8448 				flk.l_pid = 0;
8449 				(void) VOP_FRLOCK(sp->rs_finfo->rf_vp, F_SETLK,
8450 				    &flk, F_REMOTELOCK | FREAD | FWRITE,
8451 				    (u_offset_t)0, NULL, CRED(), NULL);
8452 			}
8453 
8454 			sp->rs_owner->ro_client->rc_unlksys_completed = TRUE;
8455 		}
8456 	}
8457 
8458 	/*
8459 	 * Release all locks on this file by this lock owner or at
8460 	 * least mark the locks as having been released
8461 	 */
8462 	for (lsp = list_head(&sp->rs_lostatelist); lsp != NULL;
8463 	    lsp = list_next(&sp->rs_lostatelist, lsp)) {
8464 		lsp->rls_locks_cleaned = TRUE;
8465 
8466 		/* Was this already taken care of above? */
8467 		if (!close_of_client &&
8468 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8469 			(void) cleanlocks(sp->rs_finfo->rf_vp,
8470 			    lsp->rls_locker->rl_pid,
8471 			    lsp->rls_locker->rl_client->rc_sysidt);
8472 	}
8473 
8474 	/*
8475 	 * Release any shrlocks associated with this open state ID.
8476 	 * This must be done before the rfs4_state gets marked closed.
8477 	 */
8478 	if (sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8479 		(void) rfs4_unshare(sp);
8480 
8481 	if (sp->rs_open_access) {
8482 		rfs4_dbe_lock(fp->rf_dbe);
8483 
8484 		/*
8485 		 * Decrement the count for each access and deny bit that this
8486 		 * state has contributed to the file.
8487 		 * If the file counts go to zero
8488 		 * clear the appropriate bit in the appropriate mask.
8489 		 */
8490 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) {
8491 			fp->rf_access_read--;
8492 			fflags |= FREAD;
8493 			if (fp->rf_access_read == 0)
8494 				fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
8495 		}
8496 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) {
8497 			fp->rf_access_write--;
8498 			fflags |= FWRITE;
8499 			if (fp->rf_access_write == 0)
8500 				fp->rf_share_access &=
8501 				    ~OPEN4_SHARE_ACCESS_WRITE;
8502 		}
8503 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_READ) {
8504 			fp->rf_deny_read--;
8505 			if (fp->rf_deny_read == 0)
8506 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
8507 		}
8508 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) {
8509 			fp->rf_deny_write--;
8510 			if (fp->rf_deny_write == 0)
8511 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
8512 		}
8513 
8514 		(void) VOP_CLOSE(fp->rf_vp, fflags, 1, (offset_t)0, cr, NULL);
8515 
8516 		rfs4_dbe_unlock(fp->rf_dbe);
8517 
8518 		sp->rs_open_access = 0;
8519 		sp->rs_open_deny = 0;
8520 	}
8521 }
8522 
8523 /*
8524  * lock_denied: Fill in a LOCK4deneid structure given an flock64 structure.
8525  */
8526 static nfsstat4
8527 lock_denied(LOCK4denied *dp, struct flock64 *flk)
8528 {
8529 	rfs4_lockowner_t *lo;
8530 	rfs4_client_t *cp;
8531 	uint32_t len;
8532 
8533 	lo = rfs4_findlockowner_by_pid(flk->l_pid);
8534 	if (lo != NULL) {
8535 		cp = lo->rl_client;
8536 		if (rfs4_lease_expired(cp)) {
8537 			rfs4_lockowner_rele(lo);
8538 			rfs4_dbe_hold(cp->rc_dbe);
8539 			rfs4_client_close(cp);
8540 			return (NFS4ERR_EXPIRED);
8541 		}
8542 		dp->owner.clientid = lo->rl_owner.clientid;
8543 		len = lo->rl_owner.owner_len;
8544 		dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8545 		bcopy(lo->rl_owner.owner_val, dp->owner.owner_val, len);
8546 		dp->owner.owner_len = len;
8547 		rfs4_lockowner_rele(lo);
8548 		goto finish;
8549 	}
8550 
8551 	/*
8552 	 * Its not a NFS4 lock. We take advantage that the upper 32 bits
8553 	 * of the client id contain the boot time for a NFS4 lock. So we
8554 	 * fabricate and identity by setting clientid to the sysid, and
8555 	 * the lock owner to the pid.
8556 	 */
8557 	dp->owner.clientid = flk->l_sysid;
8558 	len = sizeof (pid_t);
8559 	dp->owner.owner_len = len;
8560 	dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8561 	bcopy(&flk->l_pid, dp->owner.owner_val, len);
8562 finish:
8563 	dp->offset = flk->l_start;
8564 	dp->length = flk->l_len;
8565 
8566 	if (flk->l_type == F_RDLCK)
8567 		dp->locktype = READ_LT;
8568 	else if (flk->l_type == F_WRLCK)
8569 		dp->locktype = WRITE_LT;
8570 	else
8571 		return (NFS4ERR_INVAL);	/* no mapping from POSIX ltype to v4 */
8572 
8573 	return (NFS4_OK);
8574 }
8575 
8576 /*
8577  * The NFSv4.0 LOCK operation does not support the blocking lock (at the
8578  * NFSv4.0 protocol level) so the client needs to resend the LOCK request in a
8579  * case the lock is denied by the NFSv4.0 server.  NFSv4.0 clients are prepared
8580  * for that (obviously); they are sending the LOCK requests with some delays
8581  * between the attempts.  See nfs4frlock() and nfs4_block_and_wait() for the
8582  * locking and delay implementation at the client side.
8583  *
8584  * To make the life of the clients easier, the NFSv4.0 server tries to do some
8585  * fast retries on its own (the for loop below) in a hope the lock will be
8586  * available soon.  And if not, the client won't need to resend the LOCK
8587  * requests so fast to check the lock availability.  This basically saves some
8588  * network traffic and tries to make sure the client gets the lock ASAP.
8589  */
8590 static int
8591 setlock(vnode_t *vp, struct flock64 *flock, int flag, cred_t *cred)
8592 {
8593 	int error;
8594 	struct flock64 flk;
8595 	int i;
8596 	clock_t delaytime;
8597 	int cmd;
8598 	int spin_cnt = 0;
8599 
8600 	cmd = nbl_need_check(vp) ? F_SETLK_NBMAND : F_SETLK;
8601 retry:
8602 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
8603 
8604 	for (i = 0; i < rfs4_maxlock_tries; i++) {
8605 		LOCK_PRINT(rfs4_debug, "setlock", cmd, flock);
8606 		error = VOP_FRLOCK(vp, cmd,
8607 		    flock, flag, (u_offset_t)0, NULL, cred, NULL);
8608 
8609 		if (error != EAGAIN && error != EACCES)
8610 			break;
8611 
8612 		if (i < rfs4_maxlock_tries - 1) {
8613 			delay(delaytime);
8614 			delaytime *= 2;
8615 		}
8616 	}
8617 
8618 	if (error == EAGAIN || error == EACCES) {
8619 		/* Get the owner of the lock */
8620 		flk = *flock;
8621 		LOCK_PRINT(rfs4_debug, "setlock", F_GETLK, &flk);
8622 		if (VOP_FRLOCK(vp, F_GETLK, &flk, flag, 0, NULL, cred,
8623 		    NULL) == 0) {
8624 			/*
8625 			 * There's a race inherent in the current VOP_FRLOCK
8626 			 * design where:
8627 			 * a: "other guy" takes a lock that conflicts with a
8628 			 * lock we want
8629 			 * b: we attempt to take our lock (non-blocking) and
8630 			 * the attempt fails.
8631 			 * c: "other guy" releases the conflicting lock
8632 			 * d: we ask what lock conflicts with the lock we want,
8633 			 * getting F_UNLCK (no lock blocks us)
8634 			 *
8635 			 * If we retry the non-blocking lock attempt in this
8636 			 * case (restart at step 'b') there's some possibility
8637 			 * that many such attempts might fail.  However a test
8638 			 * designed to actually provoke this race shows that
8639 			 * the vast majority of cases require no retry, and
8640 			 * only a few took as many as three retries.  Here's
8641 			 * the test outcome:
8642 			 *
8643 			 *	   number of retries    how many times we needed
8644 			 *				that many retries
8645 			 *	   0			79461
8646 			 *	   1			  862
8647 			 *	   2			   49
8648 			 *	   3			    5
8649 			 *
8650 			 * Given those empirical results, we arbitrarily limit
8651 			 * the retry count to ten.
8652 			 *
8653 			 * If we actually make to ten retries and give up,
8654 			 * nothing catastrophic happens, but we're unable to
8655 			 * return the information about the conflicting lock to
8656 			 * the NFS client.  That's an acceptable trade off vs.
8657 			 * letting this retry loop run forever.
8658 			 */
8659 			if (flk.l_type == F_UNLCK) {
8660 				if (spin_cnt++ < 10) {
8661 					/* No longer locked, retry */
8662 					goto retry;
8663 				}
8664 			} else {
8665 				*flock = flk;
8666 				LOCK_PRINT(rfs4_debug, "setlock(blocking lock)",
8667 				    F_GETLK, &flk);
8668 			}
8669 		}
8670 	}
8671 
8672 	return (error);
8673 }
8674 
8675 /*ARGSUSED*/
8676 static nfsstat4
8677 rfs4_do_lock(rfs4_lo_state_t *lsp, nfs_lock_type4 locktype,
8678     offset4 offset, length4 length, cred_t *cred, nfs_resop4 *resop)
8679 {
8680 	nfsstat4 status;
8681 	rfs4_lockowner_t *lo = lsp->rls_locker;
8682 	rfs4_state_t *sp = lsp->rls_state;
8683 	struct flock64 flock;
8684 	int16_t ltype;
8685 	int flag;
8686 	int error;
8687 	sysid_t sysid;
8688 	LOCK4res *lres;
8689 	vnode_t *vp;
8690 
8691 	if (rfs4_lease_expired(lo->rl_client)) {
8692 		return (NFS4ERR_EXPIRED);
8693 	}
8694 
8695 	if ((status = rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
8696 		return (status);
8697 
8698 	/* Check for zero length. To lock to end of file use all ones for V4 */
8699 	if (length == 0)
8700 		return (NFS4ERR_INVAL);
8701 	else if (length == (length4)(~0))
8702 		length = 0;		/* Posix to end of file  */
8703 
8704 retry:
8705 	rfs4_dbe_lock(sp->rs_dbe);
8706 	if (sp->rs_closed == TRUE) {
8707 		rfs4_dbe_unlock(sp->rs_dbe);
8708 		return (NFS4ERR_OLD_STATEID);
8709 	}
8710 
8711 	if (resop->resop != OP_LOCKU) {
8712 		switch (locktype) {
8713 		case READ_LT:
8714 		case READW_LT:
8715 			if ((sp->rs_share_access
8716 			    & OPEN4_SHARE_ACCESS_READ) == 0) {
8717 				rfs4_dbe_unlock(sp->rs_dbe);
8718 
8719 				return (NFS4ERR_OPENMODE);
8720 			}
8721 			ltype = F_RDLCK;
8722 			break;
8723 		case WRITE_LT:
8724 		case WRITEW_LT:
8725 			if ((sp->rs_share_access
8726 			    & OPEN4_SHARE_ACCESS_WRITE) == 0) {
8727 				rfs4_dbe_unlock(sp->rs_dbe);
8728 
8729 				return (NFS4ERR_OPENMODE);
8730 			}
8731 			ltype = F_WRLCK;
8732 			break;
8733 		}
8734 	} else
8735 		ltype = F_UNLCK;
8736 
8737 	flock.l_type = ltype;
8738 	flock.l_whence = 0;		/* SEEK_SET */
8739 	flock.l_start = offset;
8740 	flock.l_len = length;
8741 	flock.l_sysid = sysid;
8742 	flock.l_pid = lsp->rls_locker->rl_pid;
8743 
8744 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
8745 	if (flock.l_len < 0 || flock.l_start < 0) {
8746 		rfs4_dbe_unlock(sp->rs_dbe);
8747 		return (NFS4ERR_INVAL);
8748 	}
8749 
8750 	/*
8751 	 * N.B. FREAD has the same value as OPEN4_SHARE_ACCESS_READ and
8752 	 * FWRITE has the same value as OPEN4_SHARE_ACCESS_WRITE.
8753 	 */
8754 	flag = (int)sp->rs_share_access | F_REMOTELOCK;
8755 
8756 	vp = sp->rs_finfo->rf_vp;
8757 	VN_HOLD(vp);
8758 
8759 	/*
8760 	 * We need to unlock sp before we call the underlying filesystem to
8761 	 * acquire the file lock.
8762 	 */
8763 	rfs4_dbe_unlock(sp->rs_dbe);
8764 
8765 	error = setlock(vp, &flock, flag, cred);
8766 
8767 	/*
8768 	 * Make sure the file is still open.  In a case the file was closed in
8769 	 * the meantime, clean the lock we acquired using the setlock() call
8770 	 * above, and return the appropriate error.
8771 	 */
8772 	rfs4_dbe_lock(sp->rs_dbe);
8773 	if (sp->rs_closed == TRUE) {
8774 		cleanlocks(vp, lsp->rls_locker->rl_pid, sysid);
8775 		rfs4_dbe_unlock(sp->rs_dbe);
8776 
8777 		VN_RELE(vp);
8778 
8779 		return (NFS4ERR_OLD_STATEID);
8780 	}
8781 	rfs4_dbe_unlock(sp->rs_dbe);
8782 
8783 	VN_RELE(vp);
8784 
8785 	if (error == 0) {
8786 		rfs4_dbe_lock(lsp->rls_dbe);
8787 		next_stateid(&lsp->rls_lockid);
8788 		rfs4_dbe_unlock(lsp->rls_dbe);
8789 	}
8790 
8791 	/*
8792 	 * N.B. We map error values to nfsv4 errors. This is differrent
8793 	 * than puterrno4 routine.
8794 	 */
8795 	switch (error) {
8796 	case 0:
8797 		status = NFS4_OK;
8798 		break;
8799 	case EAGAIN:
8800 	case EACCES:		/* Old value */
8801 		/* Can only get here if op is OP_LOCK */
8802 		ASSERT(resop->resop == OP_LOCK);
8803 		lres = &resop->nfs_resop4_u.oplock;
8804 		status = NFS4ERR_DENIED;
8805 		if (lock_denied(&lres->LOCK4res_u.denied, &flock)
8806 		    == NFS4ERR_EXPIRED)
8807 			goto retry;
8808 		break;
8809 	case ENOLCK:
8810 		status = NFS4ERR_DELAY;
8811 		break;
8812 	case EOVERFLOW:
8813 		status = NFS4ERR_INVAL;
8814 		break;
8815 	case EINVAL:
8816 		status = NFS4ERR_NOTSUPP;
8817 		break;
8818 	default:
8819 		status = NFS4ERR_SERVERFAULT;
8820 		break;
8821 	}
8822 
8823 	return (status);
8824 }
8825 
8826 /*ARGSUSED*/
8827 void
8828 rfs4_op_lock(nfs_argop4 *argop, nfs_resop4 *resop,
8829     struct svc_req *req, struct compound_state *cs)
8830 {
8831 	LOCK4args *args = &argop->nfs_argop4_u.oplock;
8832 	LOCK4res *resp = &resop->nfs_resop4_u.oplock;
8833 	nfsstat4 status;
8834 	stateid4 *stateid;
8835 	rfs4_lockowner_t *lo;
8836 	rfs4_client_t *cp;
8837 	rfs4_state_t *sp = NULL;
8838 	rfs4_lo_state_t *lsp = NULL;
8839 	bool_t ls_sw_held = FALSE;
8840 	bool_t create = TRUE;
8841 	bool_t lcreate = TRUE;
8842 	bool_t dup_lock = FALSE;
8843 	int rc;
8844 
8845 	DTRACE_NFSV4_2(op__lock__start, struct compound_state *, cs,
8846 	    LOCK4args *, args);
8847 
8848 	if (cs->vp == NULL) {
8849 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8850 		DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8851 		    cs, LOCK4res *, resp);
8852 		return;
8853 	}
8854 
8855 	if (args->locker.new_lock_owner) {
8856 		/* Create a new lockowner for this instance */
8857 		open_to_lock_owner4 *olo = &args->locker.locker4_u.open_owner;
8858 
8859 		NFS4_DEBUG(rfs4_debug, (CE_NOTE, "Creating new lock owner"));
8860 
8861 		stateid = &olo->open_stateid;
8862 		status = rfs4_get_state(stateid, &sp, RFS4_DBS_VALID);
8863 		if (status != NFS4_OK) {
8864 			NFS4_DEBUG(rfs4_debug,
8865 			    (CE_NOTE, "Get state failed in lock %d", status));
8866 			*cs->statusp = resp->status = status;
8867 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8868 			    cs, LOCK4res *, resp);
8869 			return;
8870 		}
8871 
8872 		/* Ensure specified filehandle matches */
8873 		if (cs->vp != sp->rs_finfo->rf_vp) {
8874 			rfs4_state_rele(sp);
8875 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8876 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8877 			    cs, LOCK4res *, resp);
8878 			return;
8879 		}
8880 
8881 		/* hold off other access to open_owner while we tinker */
8882 		rfs4_sw_enter(&sp->rs_owner->ro_sw);
8883 
8884 		switch (rc = rfs4_check_stateid_seqid(sp, stateid)) {
8885 		case NFS4_CHECK_STATEID_OLD:
8886 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8887 			goto end;
8888 		case NFS4_CHECK_STATEID_BAD:
8889 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8890 			goto end;
8891 		case NFS4_CHECK_STATEID_EXPIRED:
8892 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8893 			goto end;
8894 		case NFS4_CHECK_STATEID_UNCONFIRMED:
8895 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8896 			goto end;
8897 		case NFS4_CHECK_STATEID_CLOSED:
8898 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8899 			goto end;
8900 		case NFS4_CHECK_STATEID_OKAY:
8901 		case NFS4_CHECK_STATEID_REPLAY:
8902 			switch (rfs4_check_olo_seqid(olo->open_seqid,
8903 			    sp->rs_owner, resop)) {
8904 			case NFS4_CHKSEQ_OKAY:
8905 				if (rc == NFS4_CHECK_STATEID_OKAY)
8906 					break;
8907 				/*
8908 				 * This is replayed stateid; if seqid
8909 				 * matches next expected, then client
8910 				 * is using wrong seqid.
8911 				 */
8912 				/* FALLTHROUGH */
8913 			case NFS4_CHKSEQ_BAD:
8914 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8915 				goto end;
8916 			case NFS4_CHKSEQ_REPLAY:
8917 				/* This is a duplicate LOCK request */
8918 				dup_lock = TRUE;
8919 
8920 				/*
8921 				 * For a duplicate we do not want to
8922 				 * create a new lockowner as it should
8923 				 * already exist.
8924 				 * Turn off the lockowner create flag.
8925 				 */
8926 				lcreate = FALSE;
8927 			}
8928 			break;
8929 		}
8930 
8931 		lo = rfs4_findlockowner(&olo->lock_owner, &lcreate);
8932 		if (lo == NULL) {
8933 			NFS4_DEBUG(rfs4_debug,
8934 			    (CE_NOTE, "rfs4_op_lock: no lock owner"));
8935 			*cs->statusp = resp->status = NFS4ERR_RESOURCE;
8936 			goto end;
8937 		}
8938 
8939 		lsp = rfs4_findlo_state_by_owner(lo, sp, &create);
8940 		if (lsp == NULL) {
8941 			rfs4_update_lease(sp->rs_owner->ro_client);
8942 			/*
8943 			 * Only update theh open_seqid if this is not
8944 			 * a duplicate request
8945 			 */
8946 			if (dup_lock == FALSE) {
8947 				rfs4_update_open_sequence(sp->rs_owner);
8948 			}
8949 
8950 			NFS4_DEBUG(rfs4_debug,
8951 			    (CE_NOTE, "rfs4_op_lock: no state"));
8952 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
8953 			rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8954 			rfs4_lockowner_rele(lo);
8955 			goto end;
8956 		}
8957 
8958 		/*
8959 		 * This is the new_lock_owner branch and the client is
8960 		 * supposed to be associating a new lock_owner with
8961 		 * the open file at this point.  If we find that a
8962 		 * lock_owner/state association already exists and a
8963 		 * successful LOCK request was returned to the client,
8964 		 * an error is returned to the client since this is
8965 		 * not appropriate.  The client should be using the
8966 		 * existing lock_owner branch.
8967 		 */
8968 		if (dup_lock == FALSE && create == FALSE) {
8969 			if (lsp->rls_lock_completed == TRUE) {
8970 				*cs->statusp =
8971 				    resp->status = NFS4ERR_BAD_SEQID;
8972 				rfs4_lockowner_rele(lo);
8973 				goto end;
8974 			}
8975 		}
8976 
8977 		rfs4_update_lease(sp->rs_owner->ro_client);
8978 
8979 		/*
8980 		 * Only update theh open_seqid if this is not
8981 		 * a duplicate request
8982 		 */
8983 		if (dup_lock == FALSE) {
8984 			rfs4_update_open_sequence(sp->rs_owner);
8985 		}
8986 
8987 		/*
8988 		 * If this is a duplicate lock request, just copy the
8989 		 * previously saved reply and return.
8990 		 */
8991 		if (dup_lock == TRUE) {
8992 			/* verify that lock_seqid's match */
8993 			if (lsp->rls_seqid != olo->lock_seqid) {
8994 				NFS4_DEBUG(rfs4_debug,
8995 				    (CE_NOTE, "rfs4_op_lock: Dup-Lock seqid bad"
8996 				    "lsp->seqid=%d old->seqid=%d",
8997 				    lsp->rls_seqid, olo->lock_seqid));
8998 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8999 			} else {
9000 				rfs4_copy_reply(resop, &lsp->rls_reply);
9001 				/*
9002 				 * Make sure to copy the just
9003 				 * retrieved reply status into the
9004 				 * overall compound status
9005 				 */
9006 				*cs->statusp = resp->status;
9007 			}
9008 			rfs4_lockowner_rele(lo);
9009 			goto end;
9010 		}
9011 
9012 		rfs4_dbe_lock(lsp->rls_dbe);
9013 
9014 		/* Make sure to update the lock sequence id */
9015 		lsp->rls_seqid = olo->lock_seqid;
9016 
9017 		NFS4_DEBUG(rfs4_debug,
9018 		    (CE_NOTE, "Lock seqid established as %d", lsp->rls_seqid));
9019 
9020 		/*
9021 		 * This is used to signify the newly created lockowner
9022 		 * stateid and its sequence number.  The checks for
9023 		 * sequence number and increment don't occur on the
9024 		 * very first lock request for a lockowner.
9025 		 */
9026 		lsp->rls_skip_seqid_check = TRUE;
9027 
9028 		/* hold off other access to lsp while we tinker */
9029 		rfs4_sw_enter(&lsp->rls_sw);
9030 		ls_sw_held = TRUE;
9031 
9032 		rfs4_dbe_unlock(lsp->rls_dbe);
9033 
9034 		rfs4_lockowner_rele(lo);
9035 	} else {
9036 		stateid = &args->locker.locker4_u.lock_owner.lock_stateid;
9037 		/* get lsp and hold the lock on the underlying file struct */
9038 		if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE))
9039 		    != NFS4_OK) {
9040 			*cs->statusp = resp->status = status;
9041 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
9042 			    cs, LOCK4res *, resp);
9043 			return;
9044 		}
9045 		create = FALSE;	/* We didn't create lsp */
9046 
9047 		/* Ensure specified filehandle matches */
9048 		if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9049 			rfs4_lo_state_rele(lsp, TRUE);
9050 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9051 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
9052 			    cs, LOCK4res *, resp);
9053 			return;
9054 		}
9055 
9056 		/* hold off other access to lsp while we tinker */
9057 		rfs4_sw_enter(&lsp->rls_sw);
9058 		ls_sw_held = TRUE;
9059 
9060 		switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9061 		/*
9062 		 * The stateid looks like it was okay (expected to be
9063 		 * the next one)
9064 		 */
9065 		case NFS4_CHECK_STATEID_OKAY:
9066 			/*
9067 			 * The sequence id is now checked.  Determine
9068 			 * if this is a replay or if it is in the
9069 			 * expected (next) sequence.  In the case of a
9070 			 * replay, there are two replay conditions
9071 			 * that may occur.  The first is the normal
9072 			 * condition where a LOCK is done with a
9073 			 * NFS4_OK response and the stateid is
9074 			 * updated.  That case is handled below when
9075 			 * the stateid is identified as a REPLAY.  The
9076 			 * second is the case where an error is
9077 			 * returned, like NFS4ERR_DENIED, and the
9078 			 * sequence number is updated but the stateid
9079 			 * is not updated.  This second case is dealt
9080 			 * with here.  So it may seem odd that the
9081 			 * stateid is okay but the sequence id is a
9082 			 * replay but it is okay.
9083 			 */
9084 			switch (rfs4_check_lock_seqid(
9085 			    args->locker.locker4_u.lock_owner.lock_seqid,
9086 			    lsp, resop)) {
9087 			case NFS4_CHKSEQ_REPLAY:
9088 				if (resp->status != NFS4_OK) {
9089 					/*
9090 					 * Here is our replay and need
9091 					 * to verify that the last
9092 					 * response was an error.
9093 					 */
9094 					*cs->statusp = resp->status;
9095 					goto end;
9096 				}
9097 				/*
9098 				 * This is done since the sequence id
9099 				 * looked like a replay but it didn't
9100 				 * pass our check so a BAD_SEQID is
9101 				 * returned as a result.
9102 				 */
9103 				/*FALLTHROUGH*/
9104 			case NFS4_CHKSEQ_BAD:
9105 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9106 				goto end;
9107 			case NFS4_CHKSEQ_OKAY:
9108 				/* Everything looks okay move ahead */
9109 				break;
9110 			}
9111 			break;
9112 		case NFS4_CHECK_STATEID_OLD:
9113 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9114 			goto end;
9115 		case NFS4_CHECK_STATEID_BAD:
9116 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9117 			goto end;
9118 		case NFS4_CHECK_STATEID_EXPIRED:
9119 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9120 			goto end;
9121 		case NFS4_CHECK_STATEID_CLOSED:
9122 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9123 			goto end;
9124 		case NFS4_CHECK_STATEID_REPLAY:
9125 			switch (rfs4_check_lock_seqid(
9126 			    args->locker.locker4_u.lock_owner.lock_seqid,
9127 			    lsp, resop)) {
9128 			case NFS4_CHKSEQ_OKAY:
9129 				/*
9130 				 * This is a replayed stateid; if
9131 				 * seqid matches the next expected,
9132 				 * then client is using wrong seqid.
9133 				 */
9134 			case NFS4_CHKSEQ_BAD:
9135 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9136 				goto end;
9137 			case NFS4_CHKSEQ_REPLAY:
9138 				rfs4_update_lease(lsp->rls_locker->rl_client);
9139 				*cs->statusp = status = resp->status;
9140 				goto end;
9141 			}
9142 			break;
9143 		default:
9144 			ASSERT(FALSE);
9145 			break;
9146 		}
9147 
9148 		rfs4_update_lock_sequence(lsp);
9149 		rfs4_update_lease(lsp->rls_locker->rl_client);
9150 	}
9151 
9152 	/*
9153 	 * NFS4 only allows locking on regular files, so
9154 	 * verify type of object.
9155 	 */
9156 	if (cs->vp->v_type != VREG) {
9157 		if (cs->vp->v_type == VDIR)
9158 			status = NFS4ERR_ISDIR;
9159 		else
9160 			status = NFS4ERR_INVAL;
9161 		goto out;
9162 	}
9163 
9164 	cp = lsp->rls_state->rs_owner->ro_client;
9165 
9166 	if (rfs4_clnt_in_grace(cp) && !args->reclaim) {
9167 		status = NFS4ERR_GRACE;
9168 		goto out;
9169 	}
9170 
9171 	if (rfs4_clnt_in_grace(cp) && args->reclaim && !cp->rc_can_reclaim) {
9172 		status = NFS4ERR_NO_GRACE;
9173 		goto out;
9174 	}
9175 
9176 	if (!rfs4_clnt_in_grace(cp) && args->reclaim) {
9177 		status = NFS4ERR_NO_GRACE;
9178 		goto out;
9179 	}
9180 
9181 	if (lsp->rls_state->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE)
9182 		cs->deleg = TRUE;
9183 
9184 	status = rfs4_do_lock(lsp, args->locktype,
9185 	    args->offset, args->length, cs->cr, resop);
9186 
9187 out:
9188 	lsp->rls_skip_seqid_check = FALSE;
9189 
9190 	*cs->statusp = resp->status = status;
9191 
9192 	if (status == NFS4_OK) {
9193 		resp->LOCK4res_u.lock_stateid = lsp->rls_lockid.stateid;
9194 		lsp->rls_lock_completed = TRUE;
9195 	}
9196 	/*
9197 	 * Only update the "OPEN" response here if this was a new
9198 	 * lock_owner
9199 	 */
9200 	if (sp)
9201 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
9202 
9203 	rfs4_update_lock_resp(lsp, resop);
9204 
9205 end:
9206 	if (lsp) {
9207 		if (ls_sw_held)
9208 			rfs4_sw_exit(&lsp->rls_sw);
9209 		/*
9210 		 * If an sp obtained, then the lsp does not represent
9211 		 * a lock on the file struct.
9212 		 */
9213 		if (sp != NULL)
9214 			rfs4_lo_state_rele(lsp, FALSE);
9215 		else
9216 			rfs4_lo_state_rele(lsp, TRUE);
9217 	}
9218 	if (sp) {
9219 		rfs4_sw_exit(&sp->rs_owner->ro_sw);
9220 		rfs4_state_rele(sp);
9221 	}
9222 
9223 	DTRACE_NFSV4_2(op__lock__done, struct compound_state *, cs,
9224 	    LOCK4res *, resp);
9225 }
9226 
9227 /* free function for LOCK/LOCKT */
9228 static void
9229 lock_denied_free(nfs_resop4 *resop)
9230 {
9231 	LOCK4denied *dp = NULL;
9232 
9233 	switch (resop->resop) {
9234 	case OP_LOCK:
9235 		if (resop->nfs_resop4_u.oplock.status == NFS4ERR_DENIED)
9236 			dp = &resop->nfs_resop4_u.oplock.LOCK4res_u.denied;
9237 		break;
9238 	case OP_LOCKT:
9239 		if (resop->nfs_resop4_u.oplockt.status == NFS4ERR_DENIED)
9240 			dp = &resop->nfs_resop4_u.oplockt.denied;
9241 		break;
9242 	default:
9243 		break;
9244 	}
9245 
9246 	if (dp)
9247 		kmem_free(dp->owner.owner_val, dp->owner.owner_len);
9248 }
9249 
9250 /*ARGSUSED*/
9251 void
9252 rfs4_op_locku(nfs_argop4 *argop, nfs_resop4 *resop,
9253     struct svc_req *req, struct compound_state *cs)
9254 {
9255 	LOCKU4args *args = &argop->nfs_argop4_u.oplocku;
9256 	LOCKU4res *resp = &resop->nfs_resop4_u.oplocku;
9257 	nfsstat4 status;
9258 	stateid4 *stateid = &args->lock_stateid;
9259 	rfs4_lo_state_t *lsp;
9260 
9261 	DTRACE_NFSV4_2(op__locku__start, struct compound_state *, cs,
9262 	    LOCKU4args *, args);
9263 
9264 	if (cs->vp == NULL) {
9265 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9266 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9267 		    LOCKU4res *, resp);
9268 		return;
9269 	}
9270 
9271 	if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE)) != NFS4_OK) {
9272 		*cs->statusp = resp->status = status;
9273 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9274 		    LOCKU4res *, resp);
9275 		return;
9276 	}
9277 
9278 	/* Ensure specified filehandle matches */
9279 	if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9280 		rfs4_lo_state_rele(lsp, TRUE);
9281 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9282 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9283 		    LOCKU4res *, resp);
9284 		return;
9285 	}
9286 
9287 	/* hold off other access to lsp while we tinker */
9288 	rfs4_sw_enter(&lsp->rls_sw);
9289 
9290 	switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9291 	case NFS4_CHECK_STATEID_OKAY:
9292 		if (rfs4_check_lock_seqid(args->seqid, lsp, resop)
9293 		    != NFS4_CHKSEQ_OKAY) {
9294 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9295 			goto end;
9296 		}
9297 		break;
9298 	case NFS4_CHECK_STATEID_OLD:
9299 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9300 		goto end;
9301 	case NFS4_CHECK_STATEID_BAD:
9302 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9303 		goto end;
9304 	case NFS4_CHECK_STATEID_EXPIRED:
9305 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9306 		goto end;
9307 	case NFS4_CHECK_STATEID_CLOSED:
9308 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9309 		goto end;
9310 	case NFS4_CHECK_STATEID_REPLAY:
9311 		switch (rfs4_check_lock_seqid(args->seqid, lsp, resop)) {
9312 		case NFS4_CHKSEQ_OKAY:
9313 				/*
9314 				 * This is a replayed stateid; if
9315 				 * seqid matches the next expected,
9316 				 * then client is using wrong seqid.
9317 				 */
9318 		case NFS4_CHKSEQ_BAD:
9319 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9320 			goto end;
9321 		case NFS4_CHKSEQ_REPLAY:
9322 			rfs4_update_lease(lsp->rls_locker->rl_client);
9323 			*cs->statusp = status = resp->status;
9324 			goto end;
9325 		}
9326 		break;
9327 	default:
9328 		ASSERT(FALSE);
9329 		break;
9330 	}
9331 
9332 	rfs4_update_lock_sequence(lsp);
9333 	rfs4_update_lease(lsp->rls_locker->rl_client);
9334 
9335 	/*
9336 	 * NFS4 only allows locking on regular files, so
9337 	 * verify type of object.
9338 	 */
9339 	if (cs->vp->v_type != VREG) {
9340 		if (cs->vp->v_type == VDIR)
9341 			status = NFS4ERR_ISDIR;
9342 		else
9343 			status = NFS4ERR_INVAL;
9344 		goto out;
9345 	}
9346 
9347 	if (rfs4_clnt_in_grace(lsp->rls_state->rs_owner->ro_client)) {
9348 		status = NFS4ERR_GRACE;
9349 		goto out;
9350 	}
9351 
9352 	status = rfs4_do_lock(lsp, args->locktype,
9353 	    args->offset, args->length, cs->cr, resop);
9354 
9355 out:
9356 	*cs->statusp = resp->status = status;
9357 
9358 	if (status == NFS4_OK)
9359 		resp->lock_stateid = lsp->rls_lockid.stateid;
9360 
9361 	rfs4_update_lock_resp(lsp, resop);
9362 
9363 end:
9364 	rfs4_sw_exit(&lsp->rls_sw);
9365 	rfs4_lo_state_rele(lsp, TRUE);
9366 
9367 	DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9368 	    LOCKU4res *, resp);
9369 }
9370 
9371 /*
9372  * LOCKT is a best effort routine, the client can not be guaranteed that
9373  * the status return is still in effect by the time the reply is received.
9374  * They are numerous race conditions in this routine, but we are not required
9375  * and can not be accurate.
9376  */
9377 /*ARGSUSED*/
9378 void
9379 rfs4_op_lockt(nfs_argop4 *argop, nfs_resop4 *resop,
9380     struct svc_req *req, struct compound_state *cs)
9381 {
9382 	LOCKT4args *args = &argop->nfs_argop4_u.oplockt;
9383 	LOCKT4res *resp = &resop->nfs_resop4_u.oplockt;
9384 	rfs4_lockowner_t *lo;
9385 	rfs4_client_t *cp;
9386 	bool_t create = FALSE;
9387 	struct flock64 flk;
9388 	int error;
9389 	int flag = FREAD | FWRITE;
9390 	int ltype;
9391 	length4 posix_length;
9392 	sysid_t sysid;
9393 	pid_t pid;
9394 
9395 	DTRACE_NFSV4_2(op__lockt__start, struct compound_state *, cs,
9396 	    LOCKT4args *, args);
9397 
9398 	if (cs->vp == NULL) {
9399 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9400 		goto out;
9401 	}
9402 
9403 	/*
9404 	 * NFS4 only allows locking on regular files, so
9405 	 * verify type of object.
9406 	 */
9407 	if (cs->vp->v_type != VREG) {
9408 		if (cs->vp->v_type == VDIR)
9409 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
9410 		else
9411 			*cs->statusp = resp->status =  NFS4ERR_INVAL;
9412 		goto out;
9413 	}
9414 
9415 	/*
9416 	 * Check out the clientid to ensure the server knows about it
9417 	 * so that we correctly inform the client of a server reboot.
9418 	 */
9419 	if ((cp = rfs4_findclient_by_id(args->owner.clientid, FALSE))
9420 	    == NULL) {
9421 		*cs->statusp = resp->status =
9422 		    rfs4_check_clientid(&args->owner.clientid, 0);
9423 		goto out;
9424 	}
9425 	if (rfs4_lease_expired(cp)) {
9426 		rfs4_client_close(cp);
9427 		/*
9428 		 * Protocol doesn't allow returning NFS4ERR_STALE as
9429 		 * other operations do on this check so STALE_CLIENTID
9430 		 * is returned instead
9431 		 */
9432 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
9433 		goto out;
9434 	}
9435 
9436 	if (rfs4_clnt_in_grace(cp) && !(cp->rc_can_reclaim)) {
9437 		*cs->statusp = resp->status = NFS4ERR_GRACE;
9438 		rfs4_client_rele(cp);
9439 		goto out;
9440 	}
9441 	rfs4_client_rele(cp);
9442 
9443 	resp->status = NFS4_OK;
9444 
9445 	switch (args->locktype) {
9446 	case READ_LT:
9447 	case READW_LT:
9448 		ltype = F_RDLCK;
9449 		break;
9450 	case WRITE_LT:
9451 	case WRITEW_LT:
9452 		ltype = F_WRLCK;
9453 		break;
9454 	}
9455 
9456 	posix_length = args->length;
9457 	/* Check for zero length. To lock to end of file use all ones for V4 */
9458 	if (posix_length == 0) {
9459 		*cs->statusp = resp->status = NFS4ERR_INVAL;
9460 		goto out;
9461 	} else if (posix_length == (length4)(~0)) {
9462 		posix_length = 0;	/* Posix to end of file  */
9463 	}
9464 
9465 	/* Find or create a lockowner */
9466 	lo = rfs4_findlockowner(&args->owner, &create);
9467 
9468 	if (lo) {
9469 		pid = lo->rl_pid;
9470 		if ((resp->status =
9471 		    rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
9472 			goto err;
9473 	} else {
9474 		pid = 0;
9475 		sysid = lockt_sysid;
9476 	}
9477 retry:
9478 	flk.l_type = ltype;
9479 	flk.l_whence = 0;		/* SEEK_SET */
9480 	flk.l_start = args->offset;
9481 	flk.l_len = posix_length;
9482 	flk.l_sysid = sysid;
9483 	flk.l_pid = pid;
9484 	flag |= F_REMOTELOCK;
9485 
9486 	LOCK_PRINT(rfs4_debug, "rfs4_op_lockt", F_GETLK, &flk);
9487 
9488 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
9489 	if (flk.l_len < 0 || flk.l_start < 0) {
9490 		resp->status = NFS4ERR_INVAL;
9491 		goto err;
9492 	}
9493 	error = VOP_FRLOCK(cs->vp, F_GETLK, &flk, flag, (u_offset_t)0,
9494 	    NULL, cs->cr, NULL);
9495 
9496 	/*
9497 	 * N.B. We map error values to nfsv4 errors. This is differrent
9498 	 * than puterrno4 routine.
9499 	 */
9500 	switch (error) {
9501 	case 0:
9502 		if (flk.l_type == F_UNLCK)
9503 			resp->status = NFS4_OK;
9504 		else {
9505 			if (lock_denied(&resp->denied, &flk) == NFS4ERR_EXPIRED)
9506 				goto retry;
9507 			resp->status = NFS4ERR_DENIED;
9508 		}
9509 		break;
9510 	case EOVERFLOW:
9511 		resp->status = NFS4ERR_INVAL;
9512 		break;
9513 	case EINVAL:
9514 		resp->status = NFS4ERR_NOTSUPP;
9515 		break;
9516 	default:
9517 		cmn_err(CE_WARN, "rfs4_op_lockt: unexpected errno (%d)",
9518 		    error);
9519 		resp->status = NFS4ERR_SERVERFAULT;
9520 		break;
9521 	}
9522 
9523 err:
9524 	if (lo)
9525 		rfs4_lockowner_rele(lo);
9526 	*cs->statusp = resp->status;
9527 out:
9528 	DTRACE_NFSV4_2(op__lockt__done, struct compound_state *, cs,
9529 	    LOCKT4res *, resp);
9530 }
9531 
9532 int
9533 rfs4_share(rfs4_state_t *sp, uint32_t access, uint32_t deny)
9534 {
9535 	int err;
9536 	int cmd;
9537 	vnode_t *vp;
9538 	struct shrlock shr;
9539 	struct shr_locowner shr_loco;
9540 	int fflags = 0;
9541 
9542 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9543 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9544 
9545 	if (sp->rs_closed)
9546 		return (NFS4ERR_OLD_STATEID);
9547 
9548 	vp = sp->rs_finfo->rf_vp;
9549 	ASSERT(vp);
9550 
9551 	shr.s_access = shr.s_deny = 0;
9552 
9553 	if (access & OPEN4_SHARE_ACCESS_READ) {
9554 		fflags |= FREAD;
9555 		shr.s_access |= F_RDACC;
9556 	}
9557 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
9558 		fflags |= FWRITE;
9559 		shr.s_access |= F_WRACC;
9560 	}
9561 	ASSERT(shr.s_access);
9562 
9563 	if (deny & OPEN4_SHARE_DENY_READ)
9564 		shr.s_deny |= F_RDDNY;
9565 	if (deny & OPEN4_SHARE_DENY_WRITE)
9566 		shr.s_deny |= F_WRDNY;
9567 
9568 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9569 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9570 	shr_loco.sl_pid = shr.s_pid;
9571 	shr_loco.sl_id = shr.s_sysid;
9572 	shr.s_owner = (caddr_t)&shr_loco;
9573 	shr.s_own_len = sizeof (shr_loco);
9574 
9575 	cmd = nbl_need_check(vp) ? F_SHARE_NBMAND : F_SHARE;
9576 
9577 	err = VOP_SHRLOCK(vp, cmd, &shr, fflags, CRED(), NULL);
9578 	if (err != 0) {
9579 		if (err == EAGAIN)
9580 			err = NFS4ERR_SHARE_DENIED;
9581 		else
9582 			err = puterrno4(err);
9583 		return (err);
9584 	}
9585 
9586 	sp->rs_share_access |= access;
9587 	sp->rs_share_deny |= deny;
9588 
9589 	return (0);
9590 }
9591 
9592 int
9593 rfs4_unshare(rfs4_state_t *sp)
9594 {
9595 	int err;
9596 	struct shrlock shr;
9597 	struct shr_locowner shr_loco;
9598 
9599 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9600 
9601 	if (sp->rs_closed || sp->rs_share_access == 0)
9602 		return (0);
9603 
9604 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9605 	ASSERT(sp->rs_finfo->rf_vp);
9606 
9607 	shr.s_access = shr.s_deny = 0;
9608 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9609 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9610 	shr_loco.sl_pid = shr.s_pid;
9611 	shr_loco.sl_id = shr.s_sysid;
9612 	shr.s_owner = (caddr_t)&shr_loco;
9613 	shr.s_own_len = sizeof (shr_loco);
9614 
9615 	err = VOP_SHRLOCK(sp->rs_finfo->rf_vp, F_UNSHARE, &shr, 0, CRED(),
9616 	    NULL);
9617 	if (err != 0) {
9618 		err = puterrno4(err);
9619 		return (err);
9620 	}
9621 
9622 	sp->rs_share_access = 0;
9623 	sp->rs_share_deny = 0;
9624 
9625 	return (0);
9626 
9627 }
9628 
9629 static int
9630 rdma_setup_read_data4(READ4args *args, READ4res *rok)
9631 {
9632 	struct clist	*wcl;
9633 	count4		count = rok->data_len;
9634 	int		wlist_len;
9635 
9636 	wcl = args->wlist;
9637 	if (rdma_setup_read_chunks(wcl, count, &wlist_len) == FALSE) {
9638 		return (FALSE);
9639 	}
9640 	wcl = args->wlist;
9641 	rok->wlist_len = wlist_len;
9642 	rok->wlist = wcl;
9643 	return (TRUE);
9644 }
9645 
9646 /* tunable to disable server referrals */
9647 int rfs4_no_referrals = 0;
9648 
9649 /*
9650  * Find an NFS record in reparse point data.
9651  * Returns 0 for success and <0 or an errno value on failure.
9652  */
9653 int
9654 vn_find_nfs_record(vnode_t *vp, nvlist_t **nvlp, char **svcp, char **datap)
9655 {
9656 	int err;
9657 	char *stype, *val;
9658 	nvlist_t *nvl;
9659 	nvpair_t *curr;
9660 
9661 	if ((nvl = reparse_init()) == NULL)
9662 		return (-1);
9663 
9664 	if ((err = reparse_vnode_parse(vp, nvl)) != 0) {
9665 		reparse_free(nvl);
9666 		return (err);
9667 	}
9668 
9669 	curr = NULL;
9670 	while ((curr = nvlist_next_nvpair(nvl, curr)) != NULL) {
9671 		if ((stype = nvpair_name(curr)) == NULL) {
9672 			reparse_free(nvl);
9673 			return (-2);
9674 		}
9675 		if (strncasecmp(stype, "NFS", 3) == 0)
9676 			break;
9677 	}
9678 
9679 	if ((curr == NULL) ||
9680 	    (nvpair_value_string(curr, &val))) {
9681 		reparse_free(nvl);
9682 		return (-3);
9683 	}
9684 	*nvlp = nvl;
9685 	*svcp = stype;
9686 	*datap = val;
9687 	return (0);
9688 }
9689 
9690 int
9691 vn_is_nfs_reparse(vnode_t *vp, cred_t *cr)
9692 {
9693 	nvlist_t *nvl;
9694 	char *s, *d;
9695 
9696 	if (rfs4_no_referrals != 0)
9697 		return (B_FALSE);
9698 
9699 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9700 		return (B_FALSE);
9701 
9702 	if (vn_find_nfs_record(vp, &nvl, &s, &d) != 0)
9703 		return (B_FALSE);
9704 
9705 	reparse_free(nvl);
9706 
9707 	return (B_TRUE);
9708 }
9709 
9710 /*
9711  * There is a user-level copy of this routine in ref_subr.c.
9712  * Changes should be kept in sync.
9713  */
9714 static int
9715 nfs4_create_components(char *path, component4 *comp4)
9716 {
9717 	int slen, plen, ncomp;
9718 	char *ori_path, *nxtc, buf[MAXNAMELEN];
9719 
9720 	if (path == NULL)
9721 		return (0);
9722 
9723 	plen = strlen(path) + 1;	/* include the terminator */
9724 	ori_path = path;
9725 	ncomp = 0;
9726 
9727 	/* count number of components in the path */
9728 	for (nxtc = path; nxtc < ori_path + plen; nxtc++) {
9729 		if (*nxtc == '/' || *nxtc == '\0' || *nxtc == '\n') {
9730 			if ((slen = nxtc - path) == 0) {
9731 				path = nxtc + 1;
9732 				continue;
9733 			}
9734 
9735 			if (comp4 != NULL) {
9736 				bcopy(path, buf, slen);
9737 				buf[slen] = '\0';
9738 				(void) str_to_utf8(buf, &comp4[ncomp]);
9739 			}
9740 
9741 			ncomp++;	/* 1 valid component */
9742 			path = nxtc + 1;
9743 		}
9744 		if (*nxtc == '\0' || *nxtc == '\n')
9745 			break;
9746 	}
9747 
9748 	return (ncomp);
9749 }
9750 
9751 /*
9752  * There is a user-level copy of this routine in ref_subr.c.
9753  * Changes should be kept in sync.
9754  */
9755 static int
9756 make_pathname4(char *path, pathname4 *pathname)
9757 {
9758 	int ncomp;
9759 	component4 *comp4;
9760 
9761 	if (pathname == NULL)
9762 		return (0);
9763 
9764 	if (path == NULL) {
9765 		pathname->pathname4_val = NULL;
9766 		pathname->pathname4_len = 0;
9767 		return (0);
9768 	}
9769 
9770 	/* count number of components to alloc buffer */
9771 	if ((ncomp = nfs4_create_components(path, NULL)) == 0) {
9772 		pathname->pathname4_val = NULL;
9773 		pathname->pathname4_len = 0;
9774 		return (0);
9775 	}
9776 	comp4 = kmem_zalloc(ncomp * sizeof (component4), KM_SLEEP);
9777 
9778 	/* copy components into allocated buffer */
9779 	ncomp = nfs4_create_components(path, comp4);
9780 
9781 	pathname->pathname4_val = comp4;
9782 	pathname->pathname4_len = ncomp;
9783 
9784 	return (ncomp);
9785 }
9786 
9787 #define	xdr_fs_locations4 xdr_fattr4_fs_locations
9788 
9789 fs_locations4 *
9790 fetch_referral(vnode_t *vp, cred_t *cr)
9791 {
9792 	nvlist_t *nvl;
9793 	char *stype, *sdata;
9794 	fs_locations4 *result;
9795 	char buf[1024];
9796 	size_t bufsize;
9797 	XDR xdr;
9798 	int err;
9799 
9800 	/*
9801 	 * Check attrs to ensure it's a reparse point
9802 	 */
9803 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9804 		return (NULL);
9805 
9806 	/*
9807 	 * Look for an NFS record and get the type and data
9808 	 */
9809 	if (vn_find_nfs_record(vp, &nvl, &stype, &sdata) != 0)
9810 		return (NULL);
9811 
9812 	/*
9813 	 * With the type and data, upcall to get the referral
9814 	 */
9815 	bufsize = sizeof (buf);
9816 	bzero(buf, sizeof (buf));
9817 	err = reparse_kderef((const char *)stype, (const char *)sdata,
9818 	    buf, &bufsize);
9819 	reparse_free(nvl);
9820 
9821 	DTRACE_PROBE4(nfs4serv__func__referral__upcall,
9822 	    char *, stype, char *, sdata, char *, buf, int, err);
9823 	if (err) {
9824 		cmn_err(CE_NOTE,
9825 		    "reparsed daemon not running: unable to get referral (%d)",
9826 		    err);
9827 		return (NULL);
9828 	}
9829 
9830 	/*
9831 	 * We get an XDR'ed record back from the kderef call
9832 	 */
9833 	xdrmem_create(&xdr, buf, bufsize, XDR_DECODE);
9834 	result = kmem_alloc(sizeof (fs_locations4), KM_SLEEP);
9835 	err = xdr_fs_locations4(&xdr, result);
9836 	XDR_DESTROY(&xdr);
9837 	if (err != TRUE) {
9838 		DTRACE_PROBE1(nfs4serv__func__referral__upcall__xdrfail,
9839 		    int, err);
9840 		return (NULL);
9841 	}
9842 
9843 	/*
9844 	 * Look at path to recover fs_root, ignoring the leading '/'
9845 	 */
9846 	(void) make_pathname4(vp->v_path, &result->fs_root);
9847 
9848 	return (result);
9849 }
9850 
9851 char *
9852 build_symlink(vnode_t *vp, cred_t *cr, size_t *strsz)
9853 {
9854 	fs_locations4 *fsl;
9855 	fs_location4 *fs;
9856 	char *server, *path, *symbuf;
9857 	static char *prefix = "/net/";
9858 	int i, size, npaths;
9859 	uint_t len;
9860 
9861 	/* Get the referral */
9862 	if ((fsl = fetch_referral(vp, cr)) == NULL)
9863 		return (NULL);
9864 
9865 	/* Deal with only the first location and first server */
9866 	fs = &fsl->locations_val[0];
9867 	server = utf8_to_str(&fs->server_val[0], &len, NULL);
9868 	if (server == NULL) {
9869 		rfs4_free_fs_locations4(fsl);
9870 		kmem_free(fsl, sizeof (fs_locations4));
9871 		return (NULL);
9872 	}
9873 
9874 	/* Figure out size for "/net/" + host + /path/path/path + NULL */
9875 	size = strlen(prefix) + len;
9876 	for (i = 0; i < fs->rootpath.pathname4_len; i++)
9877 		size += fs->rootpath.pathname4_val[i].utf8string_len + 1;
9878 
9879 	/* Allocate the symlink buffer and fill it */
9880 	symbuf = kmem_zalloc(size, KM_SLEEP);
9881 	(void) strcat(symbuf, prefix);
9882 	(void) strcat(symbuf, server);
9883 	kmem_free(server, len);
9884 
9885 	npaths = 0;
9886 	for (i = 0; i < fs->rootpath.pathname4_len; i++) {
9887 		path = utf8_to_str(&fs->rootpath.pathname4_val[i], &len, NULL);
9888 		if (path == NULL)
9889 			continue;
9890 		(void) strcat(symbuf, "/");
9891 		(void) strcat(symbuf, path);
9892 		npaths++;
9893 		kmem_free(path, len);
9894 	}
9895 
9896 	rfs4_free_fs_locations4(fsl);
9897 	kmem_free(fsl, sizeof (fs_locations4));
9898 
9899 	if (strsz != NULL)
9900 		*strsz = size;
9901 	return (symbuf);
9902 }
9903 
9904 /*
9905  * Check to see if we have a downrev Solaris client, so that we
9906  * can send it a symlink instead of a referral.
9907  */
9908 int
9909 client_is_downrev(struct svc_req *req)
9910 {
9911 	struct sockaddr *ca;
9912 	rfs4_clntip_t *ci;
9913 	bool_t create = FALSE;
9914 	int is_downrev;
9915 
9916 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
9917 	ASSERT(ca);
9918 	ci = rfs4_find_clntip(ca, &create);
9919 	if (ci == NULL)
9920 		return (0);
9921 	is_downrev = ci->ri_no_referrals;
9922 	rfs4_dbe_rele(ci->ri_dbe);
9923 	return (is_downrev);
9924 }
9925