xref: /titanic_41/usr/src/uts/common/fs/nfs/nfs4_srv.c (revision 18ba57035593bcafd166b6540f5881f7ecfca174)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2012 by Delphix. All rights reserved.
26  */
27 
28 /*
29  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
30  *	All Rights Reserved
31  */
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <sys/systm.h>
36 #include <sys/cred.h>
37 #include <sys/buf.h>
38 #include <sys/vfs.h>
39 #include <sys/vfs_opreg.h>
40 #include <sys/vnode.h>
41 #include <sys/uio.h>
42 #include <sys/errno.h>
43 #include <sys/sysmacros.h>
44 #include <sys/statvfs.h>
45 #include <sys/kmem.h>
46 #include <sys/dirent.h>
47 #include <sys/cmn_err.h>
48 #include <sys/debug.h>
49 #include <sys/systeminfo.h>
50 #include <sys/flock.h>
51 #include <sys/pathname.h>
52 #include <sys/nbmlock.h>
53 #include <sys/share.h>
54 #include <sys/atomic.h>
55 #include <sys/policy.h>
56 #include <sys/fem.h>
57 #include <sys/sdt.h>
58 #include <sys/ddi.h>
59 #include <sys/zone.h>
60 
61 #include <fs/fs_reparse.h>
62 
63 #include <rpc/types.h>
64 #include <rpc/auth.h>
65 #include <rpc/rpcsec_gss.h>
66 #include <rpc/svc.h>
67 
68 #include <nfs/nfs.h>
69 #include <nfs/export.h>
70 #include <nfs/nfs_cmd.h>
71 #include <nfs/lm.h>
72 #include <nfs/nfs4.h>
73 
74 #include <sys/strsubr.h>
75 #include <sys/strsun.h>
76 
77 #include <inet/common.h>
78 #include <inet/ip.h>
79 #include <inet/ip6.h>
80 
81 #include <sys/tsol/label.h>
82 #include <sys/tsol/tndb.h>
83 
84 #define	RFS4_MAXLOCK_TRIES 4	/* Try to get the lock this many times */
85 static int rfs4_maxlock_tries = RFS4_MAXLOCK_TRIES;
86 #define	RFS4_LOCK_DELAY 10	/* Milliseconds */
87 static clock_t  rfs4_lock_delay = RFS4_LOCK_DELAY;
88 extern struct svc_ops rdma_svc_ops;
89 extern int nfs_loaned_buffers;
90 /* End of Tunables */
91 
92 static int rdma_setup_read_data4(READ4args *, READ4res *);
93 
94 /*
95  * Used to bump the stateid4.seqid value and show changes in the stateid
96  */
97 #define	next_stateid(sp) (++(sp)->bits.chgseq)
98 
99 /*
100  * RFS4_MINLEN_ENTRY4: XDR-encoded size of smallest possible dirent.
101  *	This is used to return NFS4ERR_TOOSMALL when clients specify
102  *	maxcount that isn't large enough to hold the smallest possible
103  *	XDR encoded dirent.
104  *
105  *	    sizeof cookie (8 bytes) +
106  *	    sizeof name_len (4 bytes) +
107  *	    sizeof smallest (padded) name (4 bytes) +
108  *	    sizeof bitmap4_len (12 bytes) +   NOTE: we always encode len=2 bm4
109  *	    sizeof attrlist4_len (4 bytes) +
110  *	    sizeof next boolean (4 bytes)
111  *
112  * RFS4_MINLEN_RDDIR4: XDR-encoded size of READDIR op reply containing
113  * the smallest possible entry4 (assumes no attrs requested).
114  *	sizeof nfsstat4 (4 bytes) +
115  *	sizeof verifier4 (8 bytes) +
116  *	sizeof entry4list bool (4 bytes) +
117  *	sizeof entry4 	(36 bytes) +
118  *	sizeof eof bool  (4 bytes)
119  *
120  * RFS4_MINLEN_RDDIR_BUF: minimum length of buffer server will provide to
121  *	VOP_READDIR.  Its value is the size of the maximum possible dirent
122  *	for solaris.  The DIRENT64_RECLEN macro returns	the size of dirent
123  *	required for a given name length.  MAXNAMELEN is the maximum
124  *	filename length allowed in Solaris.  The first two DIRENT64_RECLEN()
125  *	macros are to allow for . and .. entries -- just a minor tweak to try
126  *	and guarantee that buffer we give to VOP_READDIR will be large enough
127  *	to hold ., .., and the largest possible solaris dirent64.
128  */
129 #define	RFS4_MINLEN_ENTRY4 36
130 #define	RFS4_MINLEN_RDDIR4 (4 + NFS4_VERIFIER_SIZE + 4 + RFS4_MINLEN_ENTRY4 + 4)
131 #define	RFS4_MINLEN_RDDIR_BUF \
132 	(DIRENT64_RECLEN(1) + DIRENT64_RECLEN(2) + DIRENT64_RECLEN(MAXNAMELEN))
133 
134 /*
135  * It would be better to pad to 4 bytes since that's what XDR would do,
136  * but the dirents UFS gives us are already padded to 8, so just take
137  * what we're given.  Dircount is only a hint anyway.  Currently the
138  * solaris kernel is ASCII only, so there's no point in calling the
139  * UTF8 functions.
140  *
141  * dirent64: named padded to provide 8 byte struct alignment
142  *	d_ino(8) + d_off(8) + d_reclen(2) + d_name(namelen + null(1) + pad)
143  *
144  * cookie: uint64_t   +  utf8namelen: uint_t  +   utf8name padded to 8 bytes
145  *
146  */
147 #define	DIRENT64_TO_DIRCOUNT(dp) \
148 	(3 * BYTES_PER_XDR_UNIT + DIRENT64_NAMELEN((dp)->d_reclen))
149 
150 time_t rfs4_start_time;			/* Initialized in rfs4_srvrinit */
151 
152 static sysid_t lockt_sysid;		/* dummy sysid for all LOCKT calls */
153 
154 u_longlong_t	nfs4_srv_caller_id;
155 uint_t		nfs4_srv_vkey = 0;
156 
157 verifier4	Write4verf;
158 verifier4	Readdir4verf;
159 
160 void	rfs4_init_compound_state(struct compound_state *);
161 
162 static void	nullfree(caddr_t);
163 static void	rfs4_op_inval(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
164 			struct compound_state *);
165 static void	rfs4_op_access(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
166 			struct compound_state *);
167 static void	rfs4_op_close(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
168 			struct compound_state *);
169 static void	rfs4_op_commit(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
170 			struct compound_state *);
171 static void	rfs4_op_create(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
172 			struct compound_state *);
173 static void	rfs4_op_create_free(nfs_resop4 *resop);
174 static void	rfs4_op_delegreturn(nfs_argop4 *, nfs_resop4 *,
175 			struct svc_req *, struct compound_state *);
176 static void	rfs4_op_delegpurge(nfs_argop4 *, nfs_resop4 *,
177 			struct svc_req *, struct compound_state *);
178 static void	rfs4_op_getattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
179 			struct compound_state *);
180 static void	rfs4_op_getattr_free(nfs_resop4 *);
181 static void	rfs4_op_getfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
182 			struct compound_state *);
183 static void	rfs4_op_getfh_free(nfs_resop4 *);
184 static void	rfs4_op_illegal(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
185 			struct compound_state *);
186 static void	rfs4_op_link(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
187 			struct compound_state *);
188 static void	rfs4_op_lock(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
189 			struct compound_state *);
190 static void	lock_denied_free(nfs_resop4 *);
191 static void	rfs4_op_locku(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
192 			struct compound_state *);
193 static void	rfs4_op_lockt(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
194 			struct compound_state *);
195 static void	rfs4_op_lookup(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
196 			struct compound_state *);
197 static void	rfs4_op_lookupp(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
198 			struct compound_state *);
199 static void	rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop,
200 				struct svc_req *req, struct compound_state *cs);
201 static void	rfs4_op_nverify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
202 			struct compound_state *);
203 static void	rfs4_op_open(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
204 			struct compound_state *);
205 static void	rfs4_op_open_confirm(nfs_argop4 *, nfs_resop4 *,
206 			struct svc_req *, struct compound_state *);
207 static void	rfs4_op_open_downgrade(nfs_argop4 *, nfs_resop4 *,
208 			struct svc_req *, struct compound_state *);
209 static void	rfs4_op_putfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
210 			struct compound_state *);
211 static void	rfs4_op_putpubfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
212 			struct compound_state *);
213 static void	rfs4_op_putrootfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
214 			struct compound_state *);
215 static void	rfs4_op_read(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
216 			struct compound_state *);
217 static void	rfs4_op_read_free(nfs_resop4 *);
218 static void	rfs4_op_readdir_free(nfs_resop4 *resop);
219 static void	rfs4_op_readlink(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
220 			struct compound_state *);
221 static void	rfs4_op_readlink_free(nfs_resop4 *);
222 static void	rfs4_op_release_lockowner(nfs_argop4 *, nfs_resop4 *,
223 			struct svc_req *, struct compound_state *);
224 static void	rfs4_op_remove(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
225 			struct compound_state *);
226 static void	rfs4_op_rename(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
227 			struct compound_state *);
228 static void	rfs4_op_renew(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
229 			struct compound_state *);
230 static void	rfs4_op_restorefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
231 			struct compound_state *);
232 static void	rfs4_op_savefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
233 			struct compound_state *);
234 static void	rfs4_op_setattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
235 			struct compound_state *);
236 static void	rfs4_op_verify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
237 			struct compound_state *);
238 static void	rfs4_op_write(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
239 			struct compound_state *);
240 static void	rfs4_op_setclientid(nfs_argop4 *, nfs_resop4 *,
241 			struct svc_req *, struct compound_state *);
242 static void	rfs4_op_setclientid_confirm(nfs_argop4 *, nfs_resop4 *,
243 			struct svc_req *req, struct compound_state *);
244 static void	rfs4_op_secinfo(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
245 			struct compound_state *);
246 static void	rfs4_op_secinfo_free(nfs_resop4 *);
247 
248 static nfsstat4 check_open_access(uint32_t,
249 				struct compound_state *, struct svc_req *);
250 nfsstat4 rfs4_client_sysid(rfs4_client_t *, sysid_t *);
251 void rfs4_ss_clid(rfs4_client_t *);
252 
253 /*
254  * translation table for attrs
255  */
256 struct nfs4_ntov_table {
257 	union nfs4_attr_u *na;
258 	uint8_t amap[NFS4_MAXNUM_ATTRS];
259 	int attrcnt;
260 	bool_t vfsstat;
261 };
262 
263 static void	nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp);
264 static void	nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
265 				    struct nfs4_svgetit_arg *sargp);
266 
267 static nfsstat4	do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp,
268 		    struct compound_state *cs, struct nfs4_svgetit_arg *sargp,
269 		    struct nfs4_ntov_table *ntovp, nfs4_attr_cmd_t cmd);
270 
271 fem_t		*deleg_rdops;
272 fem_t		*deleg_wrops;
273 
274 rfs4_servinst_t *rfs4_cur_servinst = NULL;	/* current server instance */
275 kmutex_t	rfs4_servinst_lock;	/* protects linked list */
276 int		rfs4_seen_first_compound;	/* set first time we see one */
277 
278 /*
279  * NFS4 op dispatch table
280  */
281 
282 struct rfsv4disp {
283 	void	(*dis_proc)();		/* proc to call */
284 	void	(*dis_resfree)();	/* frees space allocated by proc */
285 	int	dis_flags;		/* RPC_IDEMPOTENT, etc... */
286 };
287 
288 static struct rfsv4disp rfsv4disptab[] = {
289 	/*
290 	 * NFS VERSION 4
291 	 */
292 
293 	/* RFS_NULL = 0 */
294 	{rfs4_op_illegal, nullfree, 0},
295 
296 	/* UNUSED = 1 */
297 	{rfs4_op_illegal, nullfree, 0},
298 
299 	/* UNUSED = 2 */
300 	{rfs4_op_illegal, nullfree, 0},
301 
302 	/* OP_ACCESS = 3 */
303 	{rfs4_op_access, nullfree, RPC_IDEMPOTENT},
304 
305 	/* OP_CLOSE = 4 */
306 	{rfs4_op_close, nullfree, 0},
307 
308 	/* OP_COMMIT = 5 */
309 	{rfs4_op_commit, nullfree, RPC_IDEMPOTENT},
310 
311 	/* OP_CREATE = 6 */
312 	{rfs4_op_create, nullfree, 0},
313 
314 	/* OP_DELEGPURGE = 7 */
315 	{rfs4_op_delegpurge, nullfree, 0},
316 
317 	/* OP_DELEGRETURN = 8 */
318 	{rfs4_op_delegreturn, nullfree, 0},
319 
320 	/* OP_GETATTR = 9 */
321 	{rfs4_op_getattr, rfs4_op_getattr_free, RPC_IDEMPOTENT},
322 
323 	/* OP_GETFH = 10 */
324 	{rfs4_op_getfh, rfs4_op_getfh_free, RPC_ALL},
325 
326 	/* OP_LINK = 11 */
327 	{rfs4_op_link, nullfree, 0},
328 
329 	/* OP_LOCK = 12 */
330 	{rfs4_op_lock, lock_denied_free, 0},
331 
332 	/* OP_LOCKT = 13 */
333 	{rfs4_op_lockt, lock_denied_free, 0},
334 
335 	/* OP_LOCKU = 14 */
336 	{rfs4_op_locku, nullfree, 0},
337 
338 	/* OP_LOOKUP = 15 */
339 	{rfs4_op_lookup, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
340 
341 	/* OP_LOOKUPP = 16 */
342 	{rfs4_op_lookupp, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
343 
344 	/* OP_NVERIFY = 17 */
345 	{rfs4_op_nverify, nullfree, RPC_IDEMPOTENT},
346 
347 	/* OP_OPEN = 18 */
348 	{rfs4_op_open, rfs4_free_reply, 0},
349 
350 	/* OP_OPENATTR = 19 */
351 	{rfs4_op_openattr, nullfree, 0},
352 
353 	/* OP_OPEN_CONFIRM = 20 */
354 	{rfs4_op_open_confirm, nullfree, 0},
355 
356 	/* OP_OPEN_DOWNGRADE = 21 */
357 	{rfs4_op_open_downgrade, nullfree, 0},
358 
359 	/* OP_OPEN_PUTFH = 22 */
360 	{rfs4_op_putfh, nullfree, RPC_ALL},
361 
362 	/* OP_PUTPUBFH = 23 */
363 	{rfs4_op_putpubfh, nullfree, RPC_ALL},
364 
365 	/* OP_PUTROOTFH = 24 */
366 	{rfs4_op_putrootfh, nullfree, RPC_ALL},
367 
368 	/* OP_READ = 25 */
369 	{rfs4_op_read, rfs4_op_read_free, RPC_IDEMPOTENT},
370 
371 	/* OP_READDIR = 26 */
372 	{rfs4_op_readdir, rfs4_op_readdir_free, RPC_IDEMPOTENT},
373 
374 	/* OP_READLINK = 27 */
375 	{rfs4_op_readlink, rfs4_op_readlink_free, RPC_IDEMPOTENT},
376 
377 	/* OP_REMOVE = 28 */
378 	{rfs4_op_remove, nullfree, 0},
379 
380 	/* OP_RENAME = 29 */
381 	{rfs4_op_rename, nullfree, 0},
382 
383 	/* OP_RENEW = 30 */
384 	{rfs4_op_renew, nullfree, 0},
385 
386 	/* OP_RESTOREFH = 31 */
387 	{rfs4_op_restorefh, nullfree, RPC_ALL},
388 
389 	/* OP_SAVEFH = 32 */
390 	{rfs4_op_savefh, nullfree, RPC_ALL},
391 
392 	/* OP_SECINFO = 33 */
393 	{rfs4_op_secinfo, rfs4_op_secinfo_free, 0},
394 
395 	/* OP_SETATTR = 34 */
396 	{rfs4_op_setattr, nullfree, 0},
397 
398 	/* OP_SETCLIENTID = 35 */
399 	{rfs4_op_setclientid, nullfree, 0},
400 
401 	/* OP_SETCLIENTID_CONFIRM = 36 */
402 	{rfs4_op_setclientid_confirm, nullfree, 0},
403 
404 	/* OP_VERIFY = 37 */
405 	{rfs4_op_verify, nullfree, RPC_IDEMPOTENT},
406 
407 	/* OP_WRITE = 38 */
408 	{rfs4_op_write, nullfree, 0},
409 
410 	/* OP_RELEASE_LOCKOWNER = 39 */
411 	{rfs4_op_release_lockowner, nullfree, 0},
412 };
413 
414 static uint_t rfsv4disp_cnt = sizeof (rfsv4disptab) / sizeof (rfsv4disptab[0]);
415 
416 #define	OP_ILLEGAL_IDX (rfsv4disp_cnt)
417 
418 #ifdef DEBUG
419 
420 int		rfs4_fillone_debug = 0;
421 int		rfs4_no_stub_access = 1;
422 int		rfs4_rddir_debug = 0;
423 
424 static char    *rfs4_op_string[] = {
425 	"rfs4_op_null",
426 	"rfs4_op_1 unused",
427 	"rfs4_op_2 unused",
428 	"rfs4_op_access",
429 	"rfs4_op_close",
430 	"rfs4_op_commit",
431 	"rfs4_op_create",
432 	"rfs4_op_delegpurge",
433 	"rfs4_op_delegreturn",
434 	"rfs4_op_getattr",
435 	"rfs4_op_getfh",
436 	"rfs4_op_link",
437 	"rfs4_op_lock",
438 	"rfs4_op_lockt",
439 	"rfs4_op_locku",
440 	"rfs4_op_lookup",
441 	"rfs4_op_lookupp",
442 	"rfs4_op_nverify",
443 	"rfs4_op_open",
444 	"rfs4_op_openattr",
445 	"rfs4_op_open_confirm",
446 	"rfs4_op_open_downgrade",
447 	"rfs4_op_putfh",
448 	"rfs4_op_putpubfh",
449 	"rfs4_op_putrootfh",
450 	"rfs4_op_read",
451 	"rfs4_op_readdir",
452 	"rfs4_op_readlink",
453 	"rfs4_op_remove",
454 	"rfs4_op_rename",
455 	"rfs4_op_renew",
456 	"rfs4_op_restorefh",
457 	"rfs4_op_savefh",
458 	"rfs4_op_secinfo",
459 	"rfs4_op_setattr",
460 	"rfs4_op_setclientid",
461 	"rfs4_op_setclient_confirm",
462 	"rfs4_op_verify",
463 	"rfs4_op_write",
464 	"rfs4_op_release_lockowner",
465 	"rfs4_op_illegal"
466 };
467 #endif
468 
469 void	rfs4_ss_chkclid(rfs4_client_t *);
470 
471 extern size_t   strlcpy(char *dst, const char *src, size_t dstsize);
472 
473 extern void	rfs4_free_fs_locations4(fs_locations4 *);
474 
475 #ifdef	nextdp
476 #undef nextdp
477 #endif
478 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
479 
480 static const fs_operation_def_t nfs4_rd_deleg_tmpl[] = {
481 	VOPNAME_OPEN,		{ .femop_open = deleg_rd_open },
482 	VOPNAME_WRITE,		{ .femop_write = deleg_rd_write },
483 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_rd_setattr },
484 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_rd_rwlock },
485 	VOPNAME_SPACE,		{ .femop_space = deleg_rd_space },
486 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_rd_setsecattr },
487 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_rd_vnevent },
488 	NULL,			NULL
489 };
490 static const fs_operation_def_t nfs4_wr_deleg_tmpl[] = {
491 	VOPNAME_OPEN,		{ .femop_open = deleg_wr_open },
492 	VOPNAME_READ,		{ .femop_read = deleg_wr_read },
493 	VOPNAME_WRITE,		{ .femop_write = deleg_wr_write },
494 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_wr_setattr },
495 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_wr_rwlock },
496 	VOPNAME_SPACE,		{ .femop_space = deleg_wr_space },
497 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_wr_setsecattr },
498 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_wr_vnevent },
499 	NULL,			NULL
500 };
501 
502 int
503 rfs4_srvrinit(void)
504 {
505 	timespec32_t verf;
506 	int error;
507 	extern void rfs4_attr_init();
508 	extern krwlock_t rfs4_deleg_policy_lock;
509 
510 	/*
511 	 * The following algorithm attempts to find a unique verifier
512 	 * to be used as the write verifier returned from the server
513 	 * to the client.  It is important that this verifier change
514 	 * whenever the server reboots.  Of secondary importance, it
515 	 * is important for the verifier to be unique between two
516 	 * different servers.
517 	 *
518 	 * Thus, an attempt is made to use the system hostid and the
519 	 * current time in seconds when the nfssrv kernel module is
520 	 * loaded.  It is assumed that an NFS server will not be able
521 	 * to boot and then to reboot in less than a second.  If the
522 	 * hostid has not been set, then the current high resolution
523 	 * time is used.  This will ensure different verifiers each
524 	 * time the server reboots and minimize the chances that two
525 	 * different servers will have the same verifier.
526 	 * XXX - this is broken on LP64 kernels.
527 	 */
528 	verf.tv_sec = (time_t)zone_get_hostid(NULL);
529 	if (verf.tv_sec != 0) {
530 		verf.tv_nsec = gethrestime_sec();
531 	} else {
532 		timespec_t tverf;
533 
534 		gethrestime(&tverf);
535 		verf.tv_sec = (time_t)tverf.tv_sec;
536 		verf.tv_nsec = tverf.tv_nsec;
537 	}
538 
539 	Write4verf = *(uint64_t *)&verf;
540 
541 	rfs4_attr_init();
542 	mutex_init(&rfs4_deleg_lock, NULL, MUTEX_DEFAULT, NULL);
543 
544 	/* Used to manage create/destroy of server state */
545 	mutex_init(&rfs4_state_lock, NULL, MUTEX_DEFAULT, NULL);
546 
547 	/* Used to manage access to server instance linked list */
548 	mutex_init(&rfs4_servinst_lock, NULL, MUTEX_DEFAULT, NULL);
549 
550 	/* Used to manage access to rfs4_deleg_policy */
551 	rw_init(&rfs4_deleg_policy_lock, NULL, RW_DEFAULT, NULL);
552 
553 	error = fem_create("deleg_rdops", nfs4_rd_deleg_tmpl, &deleg_rdops);
554 	if (error != 0) {
555 		rfs4_disable_delegation();
556 	} else {
557 		error = fem_create("deleg_wrops", nfs4_wr_deleg_tmpl,
558 		    &deleg_wrops);
559 		if (error != 0) {
560 			rfs4_disable_delegation();
561 			fem_free(deleg_rdops);
562 		}
563 	}
564 
565 	nfs4_srv_caller_id = fs_new_caller_id();
566 
567 	lockt_sysid = lm_alloc_sysidt();
568 
569 	vsd_create(&nfs4_srv_vkey, NULL);
570 
571 	return (0);
572 }
573 
574 void
575 rfs4_srvrfini(void)
576 {
577 	extern krwlock_t rfs4_deleg_policy_lock;
578 
579 	if (lockt_sysid != LM_NOSYSID) {
580 		lm_free_sysidt(lockt_sysid);
581 		lockt_sysid = LM_NOSYSID;
582 	}
583 
584 	mutex_destroy(&rfs4_deleg_lock);
585 	mutex_destroy(&rfs4_state_lock);
586 	rw_destroy(&rfs4_deleg_policy_lock);
587 
588 	fem_free(deleg_rdops);
589 	fem_free(deleg_wrops);
590 }
591 
592 void
593 rfs4_init_compound_state(struct compound_state *cs)
594 {
595 	bzero(cs, sizeof (*cs));
596 	cs->cont = TRUE;
597 	cs->access = CS_ACCESS_DENIED;
598 	cs->deleg = FALSE;
599 	cs->mandlock = FALSE;
600 	cs->fh.nfs_fh4_val = cs->fhbuf;
601 }
602 
603 void
604 rfs4_grace_start(rfs4_servinst_t *sip)
605 {
606 	rw_enter(&sip->rwlock, RW_WRITER);
607 	sip->start_time = (time_t)TICK_TO_SEC(ddi_get_lbolt());
608 	sip->grace_period = rfs4_grace_period;
609 	rw_exit(&sip->rwlock);
610 }
611 
612 /*
613  * returns true if the instance's grace period has never been started
614  */
615 int
616 rfs4_servinst_grace_new(rfs4_servinst_t *sip)
617 {
618 	time_t start_time;
619 
620 	rw_enter(&sip->rwlock, RW_READER);
621 	start_time = sip->start_time;
622 	rw_exit(&sip->rwlock);
623 
624 	return (start_time == 0);
625 }
626 
627 /*
628  * Indicates if server instance is within the
629  * grace period.
630  */
631 int
632 rfs4_servinst_in_grace(rfs4_servinst_t *sip)
633 {
634 	time_t grace_expiry;
635 
636 	rw_enter(&sip->rwlock, RW_READER);
637 	grace_expiry = sip->start_time + sip->grace_period;
638 	rw_exit(&sip->rwlock);
639 
640 	return (((time_t)TICK_TO_SEC(ddi_get_lbolt())) < grace_expiry);
641 }
642 
643 int
644 rfs4_clnt_in_grace(rfs4_client_t *cp)
645 {
646 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
647 
648 	return (rfs4_servinst_in_grace(cp->rc_server_instance));
649 }
650 
651 /*
652  * reset all currently active grace periods
653  */
654 void
655 rfs4_grace_reset_all(void)
656 {
657 	rfs4_servinst_t *sip;
658 
659 	mutex_enter(&rfs4_servinst_lock);
660 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
661 		if (rfs4_servinst_in_grace(sip))
662 			rfs4_grace_start(sip);
663 	mutex_exit(&rfs4_servinst_lock);
664 }
665 
666 /*
667  * start any new instances' grace periods
668  */
669 void
670 rfs4_grace_start_new(void)
671 {
672 	rfs4_servinst_t *sip;
673 
674 	mutex_enter(&rfs4_servinst_lock);
675 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
676 		if (rfs4_servinst_grace_new(sip))
677 			rfs4_grace_start(sip);
678 	mutex_exit(&rfs4_servinst_lock);
679 }
680 
681 static rfs4_dss_path_t *
682 rfs4_dss_newpath(rfs4_servinst_t *sip, char *path, unsigned index)
683 {
684 	size_t len;
685 	rfs4_dss_path_t *dss_path;
686 
687 	dss_path = kmem_alloc(sizeof (rfs4_dss_path_t), KM_SLEEP);
688 
689 	/*
690 	 * Take a copy of the string, since the original may be overwritten.
691 	 * Sadly, no strdup() in the kernel.
692 	 */
693 	/* allow for NUL */
694 	len = strlen(path) + 1;
695 	dss_path->path = kmem_alloc(len, KM_SLEEP);
696 	(void) strlcpy(dss_path->path, path, len);
697 
698 	/* associate with servinst */
699 	dss_path->sip = sip;
700 	dss_path->index = index;
701 
702 	/*
703 	 * Add to list of served paths.
704 	 * No locking required, as we're only ever called at startup.
705 	 */
706 	if (rfs4_dss_pathlist == NULL) {
707 		/* this is the first dss_path_t */
708 
709 		/* needed for insque/remque */
710 		dss_path->next = dss_path->prev = dss_path;
711 
712 		rfs4_dss_pathlist = dss_path;
713 	} else {
714 		insque(dss_path, rfs4_dss_pathlist);
715 	}
716 
717 	return (dss_path);
718 }
719 
720 /*
721  * Create a new server instance, and make it the currently active instance.
722  * Note that starting the grace period too early will reduce the clients'
723  * recovery window.
724  */
725 void
726 rfs4_servinst_create(int start_grace, int dss_npaths, char **dss_paths)
727 {
728 	unsigned i;
729 	rfs4_servinst_t *sip;
730 	rfs4_oldstate_t *oldstate;
731 
732 	sip = kmem_alloc(sizeof (rfs4_servinst_t), KM_SLEEP);
733 	rw_init(&sip->rwlock, NULL, RW_DEFAULT, NULL);
734 
735 	sip->start_time = (time_t)0;
736 	sip->grace_period = (time_t)0;
737 	sip->next = NULL;
738 	sip->prev = NULL;
739 
740 	rw_init(&sip->oldstate_lock, NULL, RW_DEFAULT, NULL);
741 	/*
742 	 * This initial dummy entry is required to setup for insque/remque.
743 	 * It must be skipped over whenever the list is traversed.
744 	 */
745 	oldstate = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
746 	/* insque/remque require initial list entry to be self-terminated */
747 	oldstate->next = oldstate;
748 	oldstate->prev = oldstate;
749 	sip->oldstate = oldstate;
750 
751 
752 	sip->dss_npaths = dss_npaths;
753 	sip->dss_paths = kmem_alloc(dss_npaths *
754 	    sizeof (rfs4_dss_path_t *), KM_SLEEP);
755 
756 	for (i = 0; i < dss_npaths; i++) {
757 		sip->dss_paths[i] = rfs4_dss_newpath(sip, dss_paths[i], i);
758 	}
759 
760 	mutex_enter(&rfs4_servinst_lock);
761 	if (rfs4_cur_servinst != NULL) {
762 		/* add to linked list */
763 		sip->prev = rfs4_cur_servinst;
764 		rfs4_cur_servinst->next = sip;
765 	}
766 	if (start_grace)
767 		rfs4_grace_start(sip);
768 	/* make the new instance "current" */
769 	rfs4_cur_servinst = sip;
770 
771 	mutex_exit(&rfs4_servinst_lock);
772 }
773 
774 /*
775  * In future, we might add a rfs4_servinst_destroy(sip) but, for now, destroy
776  * all instances directly.
777  */
778 void
779 rfs4_servinst_destroy_all(void)
780 {
781 	rfs4_servinst_t *sip, *prev, *current;
782 #ifdef DEBUG
783 	int n = 0;
784 #endif
785 
786 	mutex_enter(&rfs4_servinst_lock);
787 	ASSERT(rfs4_cur_servinst != NULL);
788 	current = rfs4_cur_servinst;
789 	rfs4_cur_servinst = NULL;
790 	for (sip = current; sip != NULL; sip = prev) {
791 		prev = sip->prev;
792 		rw_destroy(&sip->rwlock);
793 		if (sip->oldstate)
794 			kmem_free(sip->oldstate, sizeof (rfs4_oldstate_t));
795 		if (sip->dss_paths)
796 			kmem_free(sip->dss_paths,
797 			    sip->dss_npaths * sizeof (rfs4_dss_path_t *));
798 		kmem_free(sip, sizeof (rfs4_servinst_t));
799 #ifdef DEBUG
800 		n++;
801 #endif
802 	}
803 	mutex_exit(&rfs4_servinst_lock);
804 }
805 
806 /*
807  * Assign the current server instance to a client_t.
808  * Should be called with cp->rc_dbe held.
809  */
810 void
811 rfs4_servinst_assign(rfs4_client_t *cp, rfs4_servinst_t *sip)
812 {
813 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
814 
815 	/*
816 	 * The lock ensures that if the current instance is in the process
817 	 * of changing, we will see the new one.
818 	 */
819 	mutex_enter(&rfs4_servinst_lock);
820 	cp->rc_server_instance = sip;
821 	mutex_exit(&rfs4_servinst_lock);
822 }
823 
824 rfs4_servinst_t *
825 rfs4_servinst(rfs4_client_t *cp)
826 {
827 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
828 
829 	return (cp->rc_server_instance);
830 }
831 
832 /* ARGSUSED */
833 static void
834 nullfree(caddr_t resop)
835 {
836 }
837 
838 /*
839  * This is a fall-through for invalid or not implemented (yet) ops
840  */
841 /* ARGSUSED */
842 static void
843 rfs4_op_inval(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
844 	struct compound_state *cs)
845 {
846 	*cs->statusp = *((nfsstat4 *)&(resop)->nfs_resop4_u) = NFS4ERR_INVAL;
847 }
848 
849 /*
850  * Check if the security flavor, nfsnum, is in the flavor_list.
851  */
852 bool_t
853 in_flavor_list(int nfsnum, int *flavor_list, int count)
854 {
855 	int i;
856 
857 	for (i = 0; i < count; i++) {
858 		if (nfsnum == flavor_list[i])
859 			return (TRUE);
860 	}
861 	return (FALSE);
862 }
863 
864 /*
865  * Used by rfs4_op_secinfo to get the security information from the
866  * export structure associated with the component.
867  */
868 /* ARGSUSED */
869 static nfsstat4
870 do_rfs4_op_secinfo(struct compound_state *cs, char *nm, SECINFO4res *resp)
871 {
872 	int error, different_export = 0;
873 	vnode_t *dvp, *vp;
874 	struct exportinfo *exi = NULL;
875 	fid_t fid;
876 	uint_t count, i;
877 	secinfo4 *resok_val;
878 	struct secinfo *secp;
879 	seconfig_t *si;
880 	bool_t did_traverse = FALSE;
881 	int dotdot, walk;
882 
883 	dvp = cs->vp;
884 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
885 
886 	/*
887 	 * If dotdotting, then need to check whether it's above the
888 	 * root of a filesystem, or above an export point.
889 	 */
890 	if (dotdot) {
891 
892 		/*
893 		 * If dotdotting at the root of a filesystem, then
894 		 * need to traverse back to the mounted-on filesystem
895 		 * and do the dotdot lookup there.
896 		 */
897 		if (cs->vp->v_flag & VROOT) {
898 
899 			/*
900 			 * If at the system root, then can
901 			 * go up no further.
902 			 */
903 			if (VN_CMP(dvp, rootdir))
904 				return (puterrno4(ENOENT));
905 
906 			/*
907 			 * Traverse back to the mounted-on filesystem
908 			 */
909 			dvp = untraverse(cs->vp);
910 
911 			/*
912 			 * Set the different_export flag so we remember
913 			 * to pick up a new exportinfo entry for
914 			 * this new filesystem.
915 			 */
916 			different_export = 1;
917 		} else {
918 
919 			/*
920 			 * If dotdotting above an export point then set
921 			 * the different_export to get new export info.
922 			 */
923 			different_export = nfs_exported(cs->exi, cs->vp);
924 		}
925 	}
926 
927 	/*
928 	 * Get the vnode for the component "nm".
929 	 */
930 	error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cs->cr,
931 	    NULL, NULL, NULL);
932 	if (error)
933 		return (puterrno4(error));
934 
935 	/*
936 	 * If the vnode is in a pseudo filesystem, or if the security flavor
937 	 * used in the request is valid but not an explicitly shared flavor,
938 	 * or the access bit indicates that this is a limited access,
939 	 * check whether this vnode is visible.
940 	 */
941 	if (!different_export &&
942 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
943 	    cs->access & CS_ACCESS_LIMITED)) {
944 		if (! nfs_visible(cs->exi, vp, &different_export)) {
945 			VN_RELE(vp);
946 			return (puterrno4(ENOENT));
947 		}
948 	}
949 
950 	/*
951 	 * If it's a mountpoint, then traverse it.
952 	 */
953 	if (vn_ismntpt(vp)) {
954 		if ((error = traverse(&vp)) != 0) {
955 			VN_RELE(vp);
956 			return (puterrno4(error));
957 		}
958 		/* remember that we had to traverse mountpoint */
959 		did_traverse = TRUE;
960 		different_export = 1;
961 	} else if (vp->v_vfsp != dvp->v_vfsp) {
962 		/*
963 		 * If vp isn't a mountpoint and the vfs ptrs aren't the same,
964 		 * then vp is probably an LOFS object.  We don't need the
965 		 * realvp, we just need to know that we might have crossed
966 		 * a server fs boundary and need to call checkexport4.
967 		 * (LOFS lookup hides server fs mountpoints, and actually calls
968 		 * traverse)
969 		 */
970 		different_export = 1;
971 	}
972 
973 	/*
974 	 * Get the export information for it.
975 	 */
976 	if (different_export) {
977 
978 		bzero(&fid, sizeof (fid));
979 		fid.fid_len = MAXFIDSZ;
980 		error = vop_fid_pseudo(vp, &fid);
981 		if (error) {
982 			VN_RELE(vp);
983 			return (puterrno4(error));
984 		}
985 
986 		if (dotdot)
987 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
988 		else
989 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
990 
991 		if (exi == NULL) {
992 			if (did_traverse == TRUE) {
993 				/*
994 				 * If this vnode is a mounted-on vnode,
995 				 * but the mounted-on file system is not
996 				 * exported, send back the secinfo for
997 				 * the exported node that the mounted-on
998 				 * vnode lives in.
999 				 */
1000 				exi = cs->exi;
1001 			} else {
1002 				VN_RELE(vp);
1003 				return (puterrno4(EACCES));
1004 			}
1005 		}
1006 	} else {
1007 		exi = cs->exi;
1008 	}
1009 	ASSERT(exi != NULL);
1010 
1011 
1012 	/*
1013 	 * Create the secinfo result based on the security information
1014 	 * from the exportinfo structure (exi).
1015 	 *
1016 	 * Return all flavors for a pseudo node.
1017 	 * For a real export node, return the flavor that the client
1018 	 * has access with.
1019 	 */
1020 	ASSERT(RW_LOCK_HELD(&exported_lock));
1021 	if (PSEUDO(exi)) {
1022 		count = exi->exi_export.ex_seccnt; /* total sec count */
1023 		resok_val = kmem_alloc(count * sizeof (secinfo4), KM_SLEEP);
1024 		secp = exi->exi_export.ex_secinfo;
1025 
1026 		for (i = 0; i < count; i++) {
1027 			si = &secp[i].s_secinfo;
1028 			resok_val[i].flavor = si->sc_rpcnum;
1029 			if (resok_val[i].flavor == RPCSEC_GSS) {
1030 				rpcsec_gss_info *info;
1031 
1032 				info = &resok_val[i].flavor_info;
1033 				info->qop = si->sc_qop;
1034 				info->service = (rpc_gss_svc_t)si->sc_service;
1035 
1036 				/* get oid opaque data */
1037 				info->oid.sec_oid4_len =
1038 				    si->sc_gss_mech_type->length;
1039 				info->oid.sec_oid4_val = kmem_alloc(
1040 				    si->sc_gss_mech_type->length, KM_SLEEP);
1041 				bcopy(
1042 				    si->sc_gss_mech_type->elements,
1043 				    info->oid.sec_oid4_val,
1044 				    info->oid.sec_oid4_len);
1045 			}
1046 		}
1047 		resp->SECINFO4resok_len = count;
1048 		resp->SECINFO4resok_val = resok_val;
1049 	} else {
1050 		int ret_cnt = 0, k = 0;
1051 		int *flavor_list;
1052 
1053 		count = exi->exi_export.ex_seccnt; /* total sec count */
1054 		secp = exi->exi_export.ex_secinfo;
1055 
1056 		flavor_list = kmem_alloc(count * sizeof (int), KM_SLEEP);
1057 		/* find out which flavors to return */
1058 		for (i = 0; i < count; i ++) {
1059 			int access, flavor, perm;
1060 
1061 			flavor = secp[i].s_secinfo.sc_nfsnum;
1062 			perm = secp[i].s_flags;
1063 
1064 			access = nfsauth4_secinfo_access(exi, cs->req,
1065 			    flavor, perm, cs->basecr);
1066 
1067 			if (! (access & NFSAUTH_DENIED) &&
1068 			    ! (access & NFSAUTH_WRONGSEC)) {
1069 				flavor_list[ret_cnt] = flavor;
1070 				ret_cnt++;
1071 			}
1072 		}
1073 
1074 		/* Create the returning SECINFO value */
1075 		resok_val = kmem_alloc(ret_cnt * sizeof (secinfo4), KM_SLEEP);
1076 
1077 		for (i = 0; i < count; i++) {
1078 			/*
1079 			 * If the flavor is in the flavor list,
1080 			 * fill in resok_val.
1081 			 */
1082 			si = &secp[i].s_secinfo;
1083 			if (in_flavor_list(si->sc_nfsnum,
1084 			    flavor_list, ret_cnt)) {
1085 				resok_val[k].flavor = si->sc_rpcnum;
1086 				if (resok_val[k].flavor == RPCSEC_GSS) {
1087 					rpcsec_gss_info *info;
1088 
1089 					info = &resok_val[k].flavor_info;
1090 					info->qop = si->sc_qop;
1091 					info->service = (rpc_gss_svc_t)
1092 					    si->sc_service;
1093 
1094 					/* get oid opaque data */
1095 					info->oid.sec_oid4_len =
1096 					    si->sc_gss_mech_type->length;
1097 					info->oid.sec_oid4_val = kmem_alloc(
1098 					    si->sc_gss_mech_type->length,
1099 					    KM_SLEEP);
1100 					bcopy(si->sc_gss_mech_type->elements,
1101 					    info->oid.sec_oid4_val,
1102 					    info->oid.sec_oid4_len);
1103 				}
1104 				k++;
1105 			}
1106 			if (k >= ret_cnt)
1107 				break;
1108 		}
1109 		resp->SECINFO4resok_len = ret_cnt;
1110 		resp->SECINFO4resok_val = resok_val;
1111 		kmem_free(flavor_list, count * sizeof (int));
1112 	}
1113 
1114 	VN_RELE(vp);
1115 	return (NFS4_OK);
1116 }
1117 
1118 /*
1119  * SECINFO (Operation 33): Obtain required security information on
1120  * the component name in the format of (security-mechanism-oid, qop, service)
1121  * triplets.
1122  */
1123 /* ARGSUSED */
1124 static void
1125 rfs4_op_secinfo(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1126     struct compound_state *cs)
1127 {
1128 	SECINFO4args *args = &argop->nfs_argop4_u.opsecinfo;
1129 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1130 	utf8string *utfnm = &args->name;
1131 	uint_t len;
1132 	char *nm;
1133 	struct sockaddr *ca;
1134 	char *name = NULL;
1135 	nfsstat4 status = NFS4_OK;
1136 
1137 	DTRACE_NFSV4_2(op__secinfo__start, struct compound_state *, cs,
1138 	    SECINFO4args *, args);
1139 
1140 	/*
1141 	 * Current file handle (cfh) should have been set before getting
1142 	 * into this function. If not, return error.
1143 	 */
1144 	if (cs->vp == NULL) {
1145 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1146 		goto out;
1147 	}
1148 
1149 	if (cs->vp->v_type != VDIR) {
1150 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1151 		goto out;
1152 	}
1153 
1154 	/*
1155 	 * Verify the component name. If failed, error out, but
1156 	 * do not error out if the component name is a "..".
1157 	 * SECINFO will return its parents secinfo data for SECINFO "..".
1158 	 */
1159 	status = utf8_dir_verify(utfnm);
1160 	if (status != NFS4_OK) {
1161 		if (utfnm->utf8string_len != 2 ||
1162 		    utfnm->utf8string_val[0] != '.' ||
1163 		    utfnm->utf8string_val[1] != '.') {
1164 			*cs->statusp = resp->status = status;
1165 			goto out;
1166 		}
1167 	}
1168 
1169 	nm = utf8_to_str(utfnm, &len, NULL);
1170 	if (nm == NULL) {
1171 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1172 		goto out;
1173 	}
1174 
1175 	if (len > MAXNAMELEN) {
1176 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1177 		kmem_free(nm, len);
1178 		goto out;
1179 	}
1180 
1181 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1182 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1183 	    MAXPATHLEN  + 1);
1184 
1185 	if (name == NULL) {
1186 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1187 		kmem_free(nm, len);
1188 		goto out;
1189 	}
1190 
1191 
1192 	*cs->statusp = resp->status = do_rfs4_op_secinfo(cs, name, resp);
1193 
1194 	if (name != nm)
1195 		kmem_free(name, MAXPATHLEN + 1);
1196 	kmem_free(nm, len);
1197 
1198 out:
1199 	DTRACE_NFSV4_2(op__secinfo__done, struct compound_state *, cs,
1200 	    SECINFO4res *, resp);
1201 }
1202 
1203 /*
1204  * Free SECINFO result.
1205  */
1206 /* ARGSUSED */
1207 static void
1208 rfs4_op_secinfo_free(nfs_resop4 *resop)
1209 {
1210 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1211 	int count, i;
1212 	secinfo4 *resok_val;
1213 
1214 	/* If this is not an Ok result, nothing to free. */
1215 	if (resp->status != NFS4_OK) {
1216 		return;
1217 	}
1218 
1219 	count = resp->SECINFO4resok_len;
1220 	resok_val = resp->SECINFO4resok_val;
1221 
1222 	for (i = 0; i < count; i++) {
1223 		if (resok_val[i].flavor == RPCSEC_GSS) {
1224 			rpcsec_gss_info *info;
1225 
1226 			info = &resok_val[i].flavor_info;
1227 			kmem_free(info->oid.sec_oid4_val,
1228 			    info->oid.sec_oid4_len);
1229 		}
1230 	}
1231 	kmem_free(resok_val, count * sizeof (secinfo4));
1232 	resp->SECINFO4resok_len = 0;
1233 	resp->SECINFO4resok_val = NULL;
1234 }
1235 
1236 /* ARGSUSED */
1237 static void
1238 rfs4_op_access(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1239     struct compound_state *cs)
1240 {
1241 	ACCESS4args *args = &argop->nfs_argop4_u.opaccess;
1242 	ACCESS4res *resp = &resop->nfs_resop4_u.opaccess;
1243 	int error;
1244 	vnode_t *vp;
1245 	struct vattr va;
1246 	int checkwriteperm;
1247 	cred_t *cr = cs->cr;
1248 	bslabel_t *clabel, *slabel;
1249 	ts_label_t *tslabel;
1250 	boolean_t admin_low_client;
1251 
1252 	DTRACE_NFSV4_2(op__access__start, struct compound_state *, cs,
1253 	    ACCESS4args *, args);
1254 
1255 #if 0	/* XXX allow access even if !cs->access. Eventually only pseudo fs */
1256 	if (cs->access == CS_ACCESS_DENIED) {
1257 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1258 		goto out;
1259 	}
1260 #endif
1261 	if (cs->vp == NULL) {
1262 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1263 		goto out;
1264 	}
1265 
1266 	ASSERT(cr != NULL);
1267 
1268 	vp = cs->vp;
1269 
1270 	/*
1271 	 * If the file system is exported read only, it is not appropriate
1272 	 * to check write permissions for regular files and directories.
1273 	 * Special files are interpreted by the client, so the underlying
1274 	 * permissions are sent back to the client for interpretation.
1275 	 */
1276 	if (rdonly4(req, cs) &&
1277 	    (vp->v_type == VREG || vp->v_type == VDIR))
1278 		checkwriteperm = 0;
1279 	else
1280 		checkwriteperm = 1;
1281 
1282 	/*
1283 	 * XXX
1284 	 * We need the mode so that we can correctly determine access
1285 	 * permissions relative to a mandatory lock file.  Access to
1286 	 * mandatory lock files is denied on the server, so it might
1287 	 * as well be reflected to the server during the open.
1288 	 */
1289 	va.va_mask = AT_MODE;
1290 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1291 	if (error) {
1292 		*cs->statusp = resp->status = puterrno4(error);
1293 		goto out;
1294 	}
1295 	resp->access = 0;
1296 	resp->supported = 0;
1297 
1298 	if (is_system_labeled()) {
1299 		ASSERT(req->rq_label != NULL);
1300 		clabel = req->rq_label;
1301 		DTRACE_PROBE2(tx__rfs4__log__info__opaccess__clabel, char *,
1302 		    "got client label from request(1)",
1303 		    struct svc_req *, req);
1304 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
1305 			if ((tslabel = nfs_getflabel(vp, cs->exi)) == NULL) {
1306 				*cs->statusp = resp->status = puterrno4(EACCES);
1307 				goto out;
1308 			}
1309 			slabel = label2bslabel(tslabel);
1310 			DTRACE_PROBE3(tx__rfs4__log__info__opaccess__slabel,
1311 			    char *, "got server label(1) for vp(2)",
1312 			    bslabel_t *, slabel, vnode_t *, vp);
1313 
1314 			admin_low_client = B_FALSE;
1315 		} else
1316 			admin_low_client = B_TRUE;
1317 	}
1318 
1319 	if (args->access & ACCESS4_READ) {
1320 		error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
1321 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1322 		    (!is_system_labeled() || admin_low_client ||
1323 		    bldominates(clabel, slabel)))
1324 			resp->access |= ACCESS4_READ;
1325 		resp->supported |= ACCESS4_READ;
1326 	}
1327 	if ((args->access & ACCESS4_LOOKUP) && vp->v_type == VDIR) {
1328 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1329 		if (!error && (!is_system_labeled() || admin_low_client ||
1330 		    bldominates(clabel, slabel)))
1331 			resp->access |= ACCESS4_LOOKUP;
1332 		resp->supported |= ACCESS4_LOOKUP;
1333 	}
1334 	if (checkwriteperm &&
1335 	    (args->access & (ACCESS4_MODIFY|ACCESS4_EXTEND))) {
1336 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1337 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1338 		    (!is_system_labeled() || admin_low_client ||
1339 		    blequal(clabel, slabel)))
1340 			resp->access |=
1341 			    (args->access & (ACCESS4_MODIFY | ACCESS4_EXTEND));
1342 		resp->supported |=
1343 		    resp->access & (ACCESS4_MODIFY | ACCESS4_EXTEND);
1344 	}
1345 
1346 	if (checkwriteperm &&
1347 	    (args->access & ACCESS4_DELETE) && vp->v_type == VDIR) {
1348 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1349 		if (!error && (!is_system_labeled() || admin_low_client ||
1350 		    blequal(clabel, slabel)))
1351 			resp->access |= ACCESS4_DELETE;
1352 		resp->supported |= ACCESS4_DELETE;
1353 	}
1354 	if (args->access & ACCESS4_EXECUTE && vp->v_type != VDIR) {
1355 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1356 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1357 		    (!is_system_labeled() || admin_low_client ||
1358 		    bldominates(clabel, slabel)))
1359 			resp->access |= ACCESS4_EXECUTE;
1360 		resp->supported |= ACCESS4_EXECUTE;
1361 	}
1362 
1363 	if (is_system_labeled() && !admin_low_client)
1364 		label_rele(tslabel);
1365 
1366 	*cs->statusp = resp->status = NFS4_OK;
1367 out:
1368 	DTRACE_NFSV4_2(op__access__done, struct compound_state *, cs,
1369 	    ACCESS4res *, resp);
1370 }
1371 
1372 /* ARGSUSED */
1373 static void
1374 rfs4_op_commit(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1375     struct compound_state *cs)
1376 {
1377 	COMMIT4args *args = &argop->nfs_argop4_u.opcommit;
1378 	COMMIT4res *resp = &resop->nfs_resop4_u.opcommit;
1379 	int error;
1380 	vnode_t *vp = cs->vp;
1381 	cred_t *cr = cs->cr;
1382 	vattr_t va;
1383 
1384 	DTRACE_NFSV4_2(op__commit__start, struct compound_state *, cs,
1385 	    COMMIT4args *, args);
1386 
1387 	if (vp == NULL) {
1388 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1389 		goto out;
1390 	}
1391 	if (cs->access == CS_ACCESS_DENIED) {
1392 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1393 		goto out;
1394 	}
1395 
1396 	if (args->offset + args->count < args->offset) {
1397 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1398 		goto out;
1399 	}
1400 
1401 	va.va_mask = AT_UID;
1402 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1403 
1404 	/*
1405 	 * If we can't get the attributes, then we can't do the
1406 	 * right access checking.  So, we'll fail the request.
1407 	 */
1408 	if (error) {
1409 		*cs->statusp = resp->status = puterrno4(error);
1410 		goto out;
1411 	}
1412 	if (rdonly4(req, cs)) {
1413 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1414 		goto out;
1415 	}
1416 
1417 	if (vp->v_type != VREG) {
1418 		if (vp->v_type == VDIR)
1419 			resp->status = NFS4ERR_ISDIR;
1420 		else
1421 			resp->status = NFS4ERR_INVAL;
1422 		*cs->statusp = resp->status;
1423 		goto out;
1424 	}
1425 
1426 	if (crgetuid(cr) != va.va_uid &&
1427 	    (error = VOP_ACCESS(vp, VWRITE, 0, cs->cr, NULL))) {
1428 		*cs->statusp = resp->status = puterrno4(error);
1429 		goto out;
1430 	}
1431 
1432 	error = VOP_FSYNC(vp, FSYNC, cr, NULL);
1433 
1434 	if (error) {
1435 		*cs->statusp = resp->status = puterrno4(error);
1436 		goto out;
1437 	}
1438 
1439 	*cs->statusp = resp->status = NFS4_OK;
1440 	resp->writeverf = Write4verf;
1441 out:
1442 	DTRACE_NFSV4_2(op__commit__done, struct compound_state *, cs,
1443 	    COMMIT4res *, resp);
1444 }
1445 
1446 /*
1447  * rfs4_op_mknod is called from rfs4_op_create after all initial verification
1448  * was completed. It does the nfsv4 create for special files.
1449  */
1450 /* ARGSUSED */
1451 static vnode_t *
1452 do_rfs4_op_mknod(CREATE4args *args, CREATE4res *resp, struct svc_req *req,
1453     struct compound_state *cs, vattr_t *vap, char *nm)
1454 {
1455 	int error;
1456 	cred_t *cr = cs->cr;
1457 	vnode_t *dvp = cs->vp;
1458 	vnode_t *vp = NULL;
1459 	int mode;
1460 	enum vcexcl excl;
1461 
1462 	switch (args->type) {
1463 	case NF4CHR:
1464 	case NF4BLK:
1465 		if (secpolicy_sys_devices(cr) != 0) {
1466 			*cs->statusp = resp->status = NFS4ERR_PERM;
1467 			return (NULL);
1468 		}
1469 		if (args->type == NF4CHR)
1470 			vap->va_type = VCHR;
1471 		else
1472 			vap->va_type = VBLK;
1473 		vap->va_rdev = makedevice(args->ftype4_u.devdata.specdata1,
1474 		    args->ftype4_u.devdata.specdata2);
1475 		vap->va_mask |= AT_RDEV;
1476 		break;
1477 	case NF4SOCK:
1478 		vap->va_type = VSOCK;
1479 		break;
1480 	case NF4FIFO:
1481 		vap->va_type = VFIFO;
1482 		break;
1483 	default:
1484 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1485 		return (NULL);
1486 	}
1487 
1488 	/*
1489 	 * Must specify the mode.
1490 	 */
1491 	if (!(vap->va_mask & AT_MODE)) {
1492 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1493 		return (NULL);
1494 	}
1495 
1496 	excl = EXCL;
1497 
1498 	mode = 0;
1499 
1500 	error = VOP_CREATE(dvp, nm, vap, excl, mode, &vp, cr, 0, NULL, NULL);
1501 	if (error) {
1502 		*cs->statusp = resp->status = puterrno4(error);
1503 		return (NULL);
1504 	}
1505 	return (vp);
1506 }
1507 
1508 /*
1509  * nfsv4 create is used to create non-regular files. For regular files,
1510  * use nfsv4 open.
1511  */
1512 /* ARGSUSED */
1513 static void
1514 rfs4_op_create(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1515     struct compound_state *cs)
1516 {
1517 	CREATE4args *args = &argop->nfs_argop4_u.opcreate;
1518 	CREATE4res *resp = &resop->nfs_resop4_u.opcreate;
1519 	int error;
1520 	struct vattr bva, iva, iva2, ava, *vap;
1521 	cred_t *cr = cs->cr;
1522 	vnode_t *dvp = cs->vp;
1523 	vnode_t *vp = NULL;
1524 	vnode_t *realvp;
1525 	char *nm, *lnm;
1526 	uint_t len, llen;
1527 	int syncval = 0;
1528 	struct nfs4_svgetit_arg sarg;
1529 	struct nfs4_ntov_table ntov;
1530 	struct statvfs64 sb;
1531 	nfsstat4 status;
1532 	struct sockaddr *ca;
1533 	char *name = NULL;
1534 	char *lname = NULL;
1535 
1536 	DTRACE_NFSV4_2(op__create__start, struct compound_state *, cs,
1537 	    CREATE4args *, args);
1538 
1539 	resp->attrset = 0;
1540 
1541 	if (dvp == NULL) {
1542 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1543 		goto out;
1544 	}
1545 
1546 	/*
1547 	 * If there is an unshared filesystem mounted on this vnode,
1548 	 * do not allow to create an object in this directory.
1549 	 */
1550 	if (vn_ismntpt(dvp)) {
1551 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1552 		goto out;
1553 	}
1554 
1555 	/* Verify that type is correct */
1556 	switch (args->type) {
1557 	case NF4LNK:
1558 	case NF4BLK:
1559 	case NF4CHR:
1560 	case NF4SOCK:
1561 	case NF4FIFO:
1562 	case NF4DIR:
1563 		break;
1564 	default:
1565 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1566 		goto out;
1567 	};
1568 
1569 	if (cs->access == CS_ACCESS_DENIED) {
1570 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1571 		goto out;
1572 	}
1573 	if (dvp->v_type != VDIR) {
1574 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1575 		goto out;
1576 	}
1577 	status = utf8_dir_verify(&args->objname);
1578 	if (status != NFS4_OK) {
1579 		*cs->statusp = resp->status = status;
1580 		goto out;
1581 	}
1582 
1583 	if (rdonly4(req, cs)) {
1584 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1585 		goto out;
1586 	}
1587 
1588 	/*
1589 	 * Name of newly created object
1590 	 */
1591 	nm = utf8_to_fn(&args->objname, &len, NULL);
1592 	if (nm == NULL) {
1593 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1594 		goto out;
1595 	}
1596 
1597 	if (len > MAXNAMELEN) {
1598 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1599 		kmem_free(nm, len);
1600 		goto out;
1601 	}
1602 
1603 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1604 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1605 	    MAXPATHLEN  + 1);
1606 
1607 	if (name == NULL) {
1608 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1609 		kmem_free(nm, len);
1610 		goto out;
1611 	}
1612 
1613 	resp->attrset = 0;
1614 
1615 	sarg.sbp = &sb;
1616 	sarg.is_referral = B_FALSE;
1617 	nfs4_ntov_table_init(&ntov);
1618 
1619 	status = do_rfs4_set_attrs(&resp->attrset,
1620 	    &args->createattrs, cs, &sarg, &ntov, NFS4ATTR_SETIT);
1621 
1622 	if (sarg.vap->va_mask == 0 && status == NFS4_OK)
1623 		status = NFS4ERR_INVAL;
1624 
1625 	if (status != NFS4_OK) {
1626 		*cs->statusp = resp->status = status;
1627 		if (name != nm)
1628 			kmem_free(name, MAXPATHLEN + 1);
1629 		kmem_free(nm, len);
1630 		nfs4_ntov_table_free(&ntov, &sarg);
1631 		resp->attrset = 0;
1632 		goto out;
1633 	}
1634 
1635 	/* Get "before" change value */
1636 	bva.va_mask = AT_CTIME|AT_SEQ|AT_MODE;
1637 	error = VOP_GETATTR(dvp, &bva, 0, cr, NULL);
1638 	if (error) {
1639 		*cs->statusp = resp->status = puterrno4(error);
1640 		if (name != nm)
1641 			kmem_free(name, MAXPATHLEN + 1);
1642 		kmem_free(nm, len);
1643 		nfs4_ntov_table_free(&ntov, &sarg);
1644 		resp->attrset = 0;
1645 		goto out;
1646 	}
1647 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bva.va_ctime)
1648 
1649 	vap = sarg.vap;
1650 
1651 	/*
1652 	 * Set the default initial values for attributes when the parent
1653 	 * directory does not have the VSUID/VSGID bit set and they have
1654 	 * not been specified in createattrs.
1655 	 */
1656 	if (!(bva.va_mode & VSUID) && (vap->va_mask & AT_UID) == 0) {
1657 		vap->va_uid = crgetuid(cr);
1658 		vap->va_mask |= AT_UID;
1659 	}
1660 	if (!(bva.va_mode & VSGID) && (vap->va_mask & AT_GID) == 0) {
1661 		vap->va_gid = crgetgid(cr);
1662 		vap->va_mask |= AT_GID;
1663 	}
1664 
1665 	vap->va_mask |= AT_TYPE;
1666 	switch (args->type) {
1667 	case NF4DIR:
1668 		vap->va_type = VDIR;
1669 		if ((vap->va_mask & AT_MODE) == 0) {
1670 			vap->va_mode = 0700;	/* default: owner rwx only */
1671 			vap->va_mask |= AT_MODE;
1672 		}
1673 		error = VOP_MKDIR(dvp, name, vap, &vp, cr, NULL, 0, NULL);
1674 		if (error)
1675 			break;
1676 
1677 		/*
1678 		 * Get the initial "after" sequence number, if it fails,
1679 		 * set to zero
1680 		 */
1681 		iva.va_mask = AT_SEQ;
1682 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1683 			iva.va_seq = 0;
1684 		break;
1685 	case NF4LNK:
1686 		vap->va_type = VLNK;
1687 		if ((vap->va_mask & AT_MODE) == 0) {
1688 			vap->va_mode = 0700;	/* default: owner rwx only */
1689 			vap->va_mask |= AT_MODE;
1690 		}
1691 
1692 		/*
1693 		 * symlink names must be treated as data
1694 		 */
1695 		lnm = utf8_to_str((utf8string *)&args->ftype4_u.linkdata,
1696 		    &llen, NULL);
1697 
1698 		if (lnm == NULL) {
1699 			*cs->statusp = resp->status = NFS4ERR_INVAL;
1700 			if (name != nm)
1701 				kmem_free(name, MAXPATHLEN + 1);
1702 			kmem_free(nm, len);
1703 			nfs4_ntov_table_free(&ntov, &sarg);
1704 			resp->attrset = 0;
1705 			goto out;
1706 		}
1707 
1708 		if (llen > MAXPATHLEN) {
1709 			*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1710 			if (name != nm)
1711 				kmem_free(name, MAXPATHLEN + 1);
1712 			kmem_free(nm, len);
1713 			kmem_free(lnm, llen);
1714 			nfs4_ntov_table_free(&ntov, &sarg);
1715 			resp->attrset = 0;
1716 			goto out;
1717 		}
1718 
1719 		lname = nfscmd_convname(ca, cs->exi, lnm,
1720 		    NFSCMD_CONV_INBOUND, MAXPATHLEN  + 1);
1721 
1722 		if (lname == NULL) {
1723 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
1724 			if (name != nm)
1725 				kmem_free(name, MAXPATHLEN + 1);
1726 			kmem_free(nm, len);
1727 			kmem_free(lnm, llen);
1728 			nfs4_ntov_table_free(&ntov, &sarg);
1729 			resp->attrset = 0;
1730 			goto out;
1731 		}
1732 
1733 		error = VOP_SYMLINK(dvp, name, vap, lname, cr, NULL, 0);
1734 		if (lname != lnm)
1735 			kmem_free(lname, MAXPATHLEN + 1);
1736 		kmem_free(lnm, llen);
1737 		if (error)
1738 			break;
1739 
1740 		/*
1741 		 * Get the initial "after" sequence number, if it fails,
1742 		 * set to zero
1743 		 */
1744 		iva.va_mask = AT_SEQ;
1745 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1746 			iva.va_seq = 0;
1747 
1748 		error = VOP_LOOKUP(dvp, name, &vp, NULL, 0, NULL, cr,
1749 		    NULL, NULL, NULL);
1750 		if (error)
1751 			break;
1752 
1753 		/*
1754 		 * va_seq is not safe over VOP calls, check it again
1755 		 * if it has changed zero out iva to force atomic = FALSE.
1756 		 */
1757 		iva2.va_mask = AT_SEQ;
1758 		if (VOP_GETATTR(dvp, &iva2, 0, cs->cr, NULL) ||
1759 		    iva2.va_seq != iva.va_seq)
1760 			iva.va_seq = 0;
1761 		break;
1762 	default:
1763 		/*
1764 		 * probably a special file.
1765 		 */
1766 		if ((vap->va_mask & AT_MODE) == 0) {
1767 			vap->va_mode = 0600;	/* default: owner rw only */
1768 			vap->va_mask |= AT_MODE;
1769 		}
1770 		syncval = FNODSYNC;
1771 		/*
1772 		 * We know this will only generate one VOP call
1773 		 */
1774 		vp = do_rfs4_op_mknod(args, resp, req, cs, vap, name);
1775 
1776 		if (vp == NULL) {
1777 			if (name != nm)
1778 				kmem_free(name, MAXPATHLEN + 1);
1779 			kmem_free(nm, len);
1780 			nfs4_ntov_table_free(&ntov, &sarg);
1781 			resp->attrset = 0;
1782 			goto out;
1783 		}
1784 
1785 		/*
1786 		 * Get the initial "after" sequence number, if it fails,
1787 		 * set to zero
1788 		 */
1789 		iva.va_mask = AT_SEQ;
1790 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1791 			iva.va_seq = 0;
1792 
1793 		break;
1794 	}
1795 	if (name != nm)
1796 		kmem_free(name, MAXPATHLEN + 1);
1797 	kmem_free(nm, len);
1798 
1799 	if (error) {
1800 		*cs->statusp = resp->status = puterrno4(error);
1801 	}
1802 
1803 	/*
1804 	 * Force modified data and metadata out to stable storage.
1805 	 */
1806 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
1807 
1808 	if (resp->status != NFS4_OK) {
1809 		if (vp != NULL)
1810 			VN_RELE(vp);
1811 		nfs4_ntov_table_free(&ntov, &sarg);
1812 		resp->attrset = 0;
1813 		goto out;
1814 	}
1815 
1816 	/*
1817 	 * Finish setup of cinfo response, "before" value already set.
1818 	 * Get "after" change value, if it fails, simply return the
1819 	 * before value.
1820 	 */
1821 	ava.va_mask = AT_CTIME|AT_SEQ;
1822 	if (VOP_GETATTR(dvp, &ava, 0, cr, NULL)) {
1823 		ava.va_ctime = bva.va_ctime;
1824 		ava.va_seq = 0;
1825 	}
1826 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, ava.va_ctime);
1827 
1828 	/*
1829 	 * True verification that object was created with correct
1830 	 * attrs is impossible.  The attrs could have been changed
1831 	 * immediately after object creation.  If attributes did
1832 	 * not verify, the only recourse for the server is to
1833 	 * destroy the object.  Maybe if some attrs (like gid)
1834 	 * are set incorrectly, the object should be destroyed;
1835 	 * however, seems bad as a default policy.  Do we really
1836 	 * want to destroy an object over one of the times not
1837 	 * verifying correctly?  For these reasons, the server
1838 	 * currently sets bits in attrset for createattrs
1839 	 * that were set; however, no verification is done.
1840 	 *
1841 	 * vmask_to_nmask accounts for vattr bits set on create
1842 	 *	[do_rfs4_set_attrs() only sets resp bits for
1843 	 *	 non-vattr/vfs bits.]
1844 	 * Mask off any bits set by default so as not to return
1845 	 * more attrset bits than were requested in createattrs
1846 	 */
1847 	nfs4_vmask_to_nmask(sarg.vap->va_mask, &resp->attrset);
1848 	resp->attrset &= args->createattrs.attrmask;
1849 	nfs4_ntov_table_free(&ntov, &sarg);
1850 
1851 	error = makefh4(&cs->fh, vp, cs->exi);
1852 	if (error) {
1853 		*cs->statusp = resp->status = puterrno4(error);
1854 	}
1855 
1856 	/*
1857 	 * The cinfo.atomic = TRUE only if we got no errors, we have
1858 	 * non-zero va_seq's, and it has incremented by exactly one
1859 	 * during the creation and it didn't change during the VOP_LOOKUP
1860 	 * or VOP_FSYNC.
1861 	 */
1862 	if (!error && bva.va_seq && iva.va_seq && ava.va_seq &&
1863 	    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
1864 		resp->cinfo.atomic = TRUE;
1865 	else
1866 		resp->cinfo.atomic = FALSE;
1867 
1868 	/*
1869 	 * Force modified metadata out to stable storage.
1870 	 *
1871 	 * if a underlying vp exists, pass it to VOP_FSYNC
1872 	 */
1873 	if (VOP_REALVP(vp, &realvp, NULL) == 0)
1874 		(void) VOP_FSYNC(realvp, syncval, cr, NULL);
1875 	else
1876 		(void) VOP_FSYNC(vp, syncval, cr, NULL);
1877 
1878 	if (resp->status != NFS4_OK) {
1879 		VN_RELE(vp);
1880 		goto out;
1881 	}
1882 	if (cs->vp)
1883 		VN_RELE(cs->vp);
1884 
1885 	cs->vp = vp;
1886 	*cs->statusp = resp->status = NFS4_OK;
1887 out:
1888 	DTRACE_NFSV4_2(op__create__done, struct compound_state *, cs,
1889 	    CREATE4res *, resp);
1890 }
1891 
1892 /*ARGSUSED*/
1893 static void
1894 rfs4_op_delegpurge(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1895     struct compound_state *cs)
1896 {
1897 	DTRACE_NFSV4_2(op__delegpurge__start, struct compound_state *, cs,
1898 	    DELEGPURGE4args *, &argop->nfs_argop4_u.opdelegpurge);
1899 
1900 	rfs4_op_inval(argop, resop, req, cs);
1901 
1902 	DTRACE_NFSV4_2(op__delegpurge__done, struct compound_state *, cs,
1903 	    DELEGPURGE4res *, &resop->nfs_resop4_u.opdelegpurge);
1904 }
1905 
1906 /*ARGSUSED*/
1907 static void
1908 rfs4_op_delegreturn(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1909     struct compound_state *cs)
1910 {
1911 	DELEGRETURN4args *args = &argop->nfs_argop4_u.opdelegreturn;
1912 	DELEGRETURN4res *resp = &resop->nfs_resop4_u.opdelegreturn;
1913 	rfs4_deleg_state_t *dsp;
1914 	nfsstat4 status;
1915 
1916 	DTRACE_NFSV4_2(op__delegreturn__start, struct compound_state *, cs,
1917 	    DELEGRETURN4args *, args);
1918 
1919 	status = rfs4_get_deleg_state(&args->deleg_stateid, &dsp);
1920 	resp->status = *cs->statusp = status;
1921 	if (status != NFS4_OK)
1922 		goto out;
1923 
1924 	/* Ensure specified filehandle matches */
1925 	if (cs->vp != dsp->rds_finfo->rf_vp) {
1926 		resp->status = *cs->statusp = NFS4ERR_BAD_STATEID;
1927 	} else
1928 		rfs4_return_deleg(dsp, FALSE);
1929 
1930 	rfs4_update_lease(dsp->rds_client);
1931 
1932 	rfs4_deleg_state_rele(dsp);
1933 out:
1934 	DTRACE_NFSV4_2(op__delegreturn__done, struct compound_state *, cs,
1935 	    DELEGRETURN4res *, resp);
1936 }
1937 
1938 /*
1939  * Check to see if a given "flavor" is an explicitly shared flavor.
1940  * The assumption of this routine is the "flavor" is already a valid
1941  * flavor in the secinfo list of "exi".
1942  *
1943  *	e.g.
1944  *		# share -o sec=flavor1 /export
1945  *		# share -o sec=flavor2 /export/home
1946  *
1947  *		flavor2 is not an explicitly shared flavor for /export,
1948  *		however it is in the secinfo list for /export thru the
1949  *		server namespace setup.
1950  */
1951 int
1952 is_exported_sec(int flavor, struct exportinfo *exi)
1953 {
1954 	int	i;
1955 	struct secinfo *sp;
1956 
1957 	sp = exi->exi_export.ex_secinfo;
1958 	for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
1959 		if (flavor == sp[i].s_secinfo.sc_nfsnum ||
1960 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE) {
1961 			return (SEC_REF_EXPORTED(&sp[i]));
1962 		}
1963 	}
1964 
1965 	/* Should not reach this point based on the assumption */
1966 	return (0);
1967 }
1968 
1969 /*
1970  * Check if the security flavor used in the request matches what is
1971  * required at the export point or at the root pseudo node (exi_root).
1972  *
1973  * returns 1 if there's a match or if exported with AUTH_NONE; 0 otherwise.
1974  *
1975  */
1976 static int
1977 secinfo_match_or_authnone(struct compound_state *cs)
1978 {
1979 	int	i;
1980 	struct secinfo *sp;
1981 
1982 	/*
1983 	 * Check cs->nfsflavor (from the request) against
1984 	 * the current export data in cs->exi.
1985 	 */
1986 	sp = cs->exi->exi_export.ex_secinfo;
1987 	for (i = 0; i < cs->exi->exi_export.ex_seccnt; i++) {
1988 		if (cs->nfsflavor == sp[i].s_secinfo.sc_nfsnum ||
1989 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE)
1990 			return (1);
1991 	}
1992 
1993 	return (0);
1994 }
1995 
1996 /*
1997  * Check the access authority for the client and return the correct error.
1998  */
1999 nfsstat4
2000 call_checkauth4(struct compound_state *cs, struct svc_req *req)
2001 {
2002 	int	authres;
2003 
2004 	/*
2005 	 * First, check if the security flavor used in the request
2006 	 * are among the flavors set in the server namespace.
2007 	 */
2008 	if (!secinfo_match_or_authnone(cs)) {
2009 		*cs->statusp = NFS4ERR_WRONGSEC;
2010 		return (*cs->statusp);
2011 	}
2012 
2013 	authres = checkauth4(cs, req);
2014 
2015 	if (authres > 0) {
2016 		*cs->statusp = NFS4_OK;
2017 		if (! (cs->access & CS_ACCESS_LIMITED))
2018 			cs->access = CS_ACCESS_OK;
2019 	} else if (authres == 0) {
2020 		*cs->statusp = NFS4ERR_ACCESS;
2021 	} else if (authres == -2) {
2022 		*cs->statusp = NFS4ERR_WRONGSEC;
2023 	} else {
2024 		*cs->statusp = NFS4ERR_DELAY;
2025 	}
2026 	return (*cs->statusp);
2027 }
2028 
2029 /*
2030  * bitmap4_to_attrmask is called by getattr and readdir.
2031  * It sets up the vattr mask and determines whether vfsstat call is needed
2032  * based on the input bitmap.
2033  * Returns nfsv4 status.
2034  */
2035 static nfsstat4
2036 bitmap4_to_attrmask(bitmap4 breq, struct nfs4_svgetit_arg *sargp)
2037 {
2038 	int i;
2039 	uint_t	va_mask;
2040 	struct statvfs64 *sbp = sargp->sbp;
2041 
2042 	sargp->sbp = NULL;
2043 	sargp->flag = 0;
2044 	sargp->rdattr_error = NFS4_OK;
2045 	sargp->mntdfid_set = FALSE;
2046 	if (sargp->cs->vp)
2047 		sargp->xattr = get_fh4_flag(&sargp->cs->fh,
2048 		    FH4_ATTRDIR | FH4_NAMEDATTR);
2049 	else
2050 		sargp->xattr = 0;
2051 
2052 	/*
2053 	 * Set rdattr_error_req to true if return error per
2054 	 * failed entry rather than fail the readdir.
2055 	 */
2056 	if (breq & FATTR4_RDATTR_ERROR_MASK)
2057 		sargp->rdattr_error_req = 1;
2058 	else
2059 		sargp->rdattr_error_req = 0;
2060 
2061 	/*
2062 	 * generate the va_mask
2063 	 * Handle the easy cases first
2064 	 */
2065 	switch (breq) {
2066 	case NFS4_NTOV_ATTR_MASK:
2067 		sargp->vap->va_mask = NFS4_NTOV_ATTR_AT_MASK;
2068 		return (NFS4_OK);
2069 
2070 	case NFS4_FS_ATTR_MASK:
2071 		sargp->vap->va_mask = NFS4_FS_ATTR_AT_MASK;
2072 		sargp->sbp = sbp;
2073 		return (NFS4_OK);
2074 
2075 	case NFS4_NTOV_ATTR_CACHE_MASK:
2076 		sargp->vap->va_mask = NFS4_NTOV_ATTR_CACHE_AT_MASK;
2077 		return (NFS4_OK);
2078 
2079 	case FATTR4_LEASE_TIME_MASK:
2080 		sargp->vap->va_mask = 0;
2081 		return (NFS4_OK);
2082 
2083 	default:
2084 		va_mask = 0;
2085 		for (i = 0; i < nfs4_ntov_map_size; i++) {
2086 			if ((breq & nfs4_ntov_map[i].fbit) &&
2087 			    nfs4_ntov_map[i].vbit)
2088 				va_mask |= nfs4_ntov_map[i].vbit;
2089 		}
2090 
2091 		/*
2092 		 * Check is vfsstat is needed
2093 		 */
2094 		if (breq & NFS4_FS_ATTR_MASK)
2095 			sargp->sbp = sbp;
2096 
2097 		sargp->vap->va_mask = va_mask;
2098 		return (NFS4_OK);
2099 	}
2100 	/* NOTREACHED */
2101 }
2102 
2103 /*
2104  * bitmap4_get_sysattrs is called by getattr and readdir.
2105  * It calls both VOP_GETATTR and VFS_STATVFS calls to get the attrs.
2106  * Returns nfsv4 status.
2107  */
2108 static nfsstat4
2109 bitmap4_get_sysattrs(struct nfs4_svgetit_arg *sargp)
2110 {
2111 	int error;
2112 	struct compound_state *cs = sargp->cs;
2113 	vnode_t *vp = cs->vp;
2114 
2115 	if (sargp->sbp != NULL) {
2116 		if (error = VFS_STATVFS(vp->v_vfsp, sargp->sbp)) {
2117 			sargp->sbp = NULL;	/* to identify error */
2118 			return (puterrno4(error));
2119 		}
2120 	}
2121 
2122 	return (rfs4_vop_getattr(vp, sargp->vap, 0, cs->cr));
2123 }
2124 
2125 static void
2126 nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp)
2127 {
2128 	ntovp->na = kmem_zalloc(sizeof (union nfs4_attr_u) * nfs4_ntov_map_size,
2129 	    KM_SLEEP);
2130 	ntovp->attrcnt = 0;
2131 	ntovp->vfsstat = FALSE;
2132 }
2133 
2134 static void
2135 nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
2136     struct nfs4_svgetit_arg *sargp)
2137 {
2138 	int i;
2139 	union nfs4_attr_u *na;
2140 	uint8_t *amap;
2141 
2142 	/*
2143 	 * XXX Should do the same checks for whether the bit is set
2144 	 */
2145 	for (i = 0, na = ntovp->na, amap = ntovp->amap;
2146 	    i < ntovp->attrcnt; i++, na++, amap++) {
2147 		(void) (*nfs4_ntov_map[*amap].sv_getit)(
2148 		    NFS4ATTR_FREEIT, sargp, na);
2149 	}
2150 	if ((sargp->op == NFS4ATTR_SETIT) || (sargp->op == NFS4ATTR_VERIT)) {
2151 		/*
2152 		 * xdr_free for getattr will be done later
2153 		 */
2154 		for (i = 0, na = ntovp->na, amap = ntovp->amap;
2155 		    i < ntovp->attrcnt; i++, na++, amap++) {
2156 			xdr_free(nfs4_ntov_map[*amap].xfunc, (caddr_t)na);
2157 		}
2158 	}
2159 	kmem_free(ntovp->na, sizeof (union nfs4_attr_u) * nfs4_ntov_map_size);
2160 }
2161 
2162 /*
2163  * do_rfs4_op_getattr gets the system attrs and converts into fattr4.
2164  */
2165 static nfsstat4
2166 do_rfs4_op_getattr(bitmap4 breq, fattr4 *fattrp,
2167     struct nfs4_svgetit_arg *sargp)
2168 {
2169 	int error = 0;
2170 	int i, k;
2171 	struct nfs4_ntov_table ntov;
2172 	XDR xdr;
2173 	ulong_t xdr_size;
2174 	char *xdr_attrs;
2175 	nfsstat4 status = NFS4_OK;
2176 	nfsstat4 prev_rdattr_error = sargp->rdattr_error;
2177 	union nfs4_attr_u *na;
2178 	uint8_t *amap;
2179 
2180 	sargp->op = NFS4ATTR_GETIT;
2181 	sargp->flag = 0;
2182 
2183 	fattrp->attrmask = 0;
2184 	/* if no bits requested, then return empty fattr4 */
2185 	if (breq == 0) {
2186 		fattrp->attrlist4_len = 0;
2187 		fattrp->attrlist4 = NULL;
2188 		return (NFS4_OK);
2189 	}
2190 
2191 	/*
2192 	 * return NFS4ERR_INVAL when client requests write-only attrs
2193 	 */
2194 	if (breq & (FATTR4_TIME_ACCESS_SET_MASK | FATTR4_TIME_MODIFY_SET_MASK))
2195 		return (NFS4ERR_INVAL);
2196 
2197 	nfs4_ntov_table_init(&ntov);
2198 	na = ntov.na;
2199 	amap = ntov.amap;
2200 
2201 	/*
2202 	 * Now loop to get or verify the attrs
2203 	 */
2204 	for (i = 0; i < nfs4_ntov_map_size; i++) {
2205 		if (breq & nfs4_ntov_map[i].fbit) {
2206 			if ((*nfs4_ntov_map[i].sv_getit)(
2207 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0) {
2208 
2209 				error = (*nfs4_ntov_map[i].sv_getit)(
2210 				    NFS4ATTR_GETIT, sargp, na);
2211 
2212 				/*
2213 				 * Possible error values:
2214 				 * >0 if sv_getit failed to
2215 				 * get the attr; 0 if succeeded;
2216 				 * <0 if rdattr_error and the
2217 				 * attribute cannot be returned.
2218 				 */
2219 				if (error && !(sargp->rdattr_error_req))
2220 					goto done;
2221 				/*
2222 				 * If error then just for entry
2223 				 */
2224 				if (error == 0) {
2225 					fattrp->attrmask |=
2226 					    nfs4_ntov_map[i].fbit;
2227 					*amap++ =
2228 					    (uint8_t)nfs4_ntov_map[i].nval;
2229 					na++;
2230 					(ntov.attrcnt)++;
2231 				} else if ((error > 0) &&
2232 				    (sargp->rdattr_error == NFS4_OK)) {
2233 					sargp->rdattr_error = puterrno4(error);
2234 				}
2235 				error = 0;
2236 			}
2237 		}
2238 	}
2239 
2240 	/*
2241 	 * If rdattr_error was set after the return value for it was assigned,
2242 	 * update it.
2243 	 */
2244 	if (prev_rdattr_error != sargp->rdattr_error) {
2245 		na = ntov.na;
2246 		amap = ntov.amap;
2247 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2248 			k = *amap;
2249 			if (k < FATTR4_RDATTR_ERROR) {
2250 				continue;
2251 			}
2252 			if ((k == FATTR4_RDATTR_ERROR) &&
2253 			    ((*nfs4_ntov_map[k].sv_getit)(
2254 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0)) {
2255 
2256 				(void) (*nfs4_ntov_map[k].sv_getit)(
2257 				    NFS4ATTR_GETIT, sargp, na);
2258 			}
2259 			break;
2260 		}
2261 	}
2262 
2263 	xdr_size = 0;
2264 	na = ntov.na;
2265 	amap = ntov.amap;
2266 	for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2267 		xdr_size += xdr_sizeof(nfs4_ntov_map[*amap].xfunc, na);
2268 	}
2269 
2270 	fattrp->attrlist4_len = xdr_size;
2271 	if (xdr_size) {
2272 		/* freed by rfs4_op_getattr_free() */
2273 		fattrp->attrlist4 = xdr_attrs = kmem_zalloc(xdr_size, KM_SLEEP);
2274 
2275 		xdrmem_create(&xdr, xdr_attrs, xdr_size, XDR_ENCODE);
2276 
2277 		na = ntov.na;
2278 		amap = ntov.amap;
2279 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2280 			if (!(*nfs4_ntov_map[*amap].xfunc)(&xdr, na)) {
2281 				DTRACE_PROBE1(nfss__e__getattr4_encfail,
2282 				    int, *amap);
2283 				status = NFS4ERR_SERVERFAULT;
2284 				break;
2285 			}
2286 		}
2287 		/* xdrmem_destroy(&xdrs); */	/* NO-OP */
2288 	} else {
2289 		fattrp->attrlist4 = NULL;
2290 	}
2291 done:
2292 
2293 	nfs4_ntov_table_free(&ntov, sargp);
2294 
2295 	if (error != 0)
2296 		status = puterrno4(error);
2297 
2298 	return (status);
2299 }
2300 
2301 /* ARGSUSED */
2302 static void
2303 rfs4_op_getattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2304     struct compound_state *cs)
2305 {
2306 	GETATTR4args *args = &argop->nfs_argop4_u.opgetattr;
2307 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2308 	struct nfs4_svgetit_arg sarg;
2309 	struct statvfs64 sb;
2310 	nfsstat4 status;
2311 
2312 	DTRACE_NFSV4_2(op__getattr__start, struct compound_state *, cs,
2313 	    GETATTR4args *, args);
2314 
2315 	if (cs->vp == NULL) {
2316 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2317 		goto out;
2318 	}
2319 
2320 	if (cs->access == CS_ACCESS_DENIED) {
2321 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2322 		goto out;
2323 	}
2324 
2325 	sarg.sbp = &sb;
2326 	sarg.cs = cs;
2327 	sarg.is_referral = B_FALSE;
2328 
2329 	status = bitmap4_to_attrmask(args->attr_request, &sarg);
2330 	if (status == NFS4_OK) {
2331 
2332 		status = bitmap4_get_sysattrs(&sarg);
2333 		if (status == NFS4_OK) {
2334 
2335 			/* Is this a referral? */
2336 			if (vn_is_nfs_reparse(cs->vp, cs->cr)) {
2337 				/* Older V4 Solaris client sees a link */
2338 				if (client_is_downrev(req))
2339 					sarg.vap->va_type = VLNK;
2340 				else
2341 					sarg.is_referral = B_TRUE;
2342 			}
2343 
2344 			status = do_rfs4_op_getattr(args->attr_request,
2345 			    &resp->obj_attributes, &sarg);
2346 		}
2347 	}
2348 	*cs->statusp = resp->status = status;
2349 out:
2350 	DTRACE_NFSV4_2(op__getattr__done, struct compound_state *, cs,
2351 	    GETATTR4res *, resp);
2352 }
2353 
2354 static void
2355 rfs4_op_getattr_free(nfs_resop4 *resop)
2356 {
2357 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2358 
2359 	nfs4_fattr4_free(&resp->obj_attributes);
2360 }
2361 
2362 /* ARGSUSED */
2363 static void
2364 rfs4_op_getfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2365     struct compound_state *cs)
2366 {
2367 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2368 
2369 	DTRACE_NFSV4_1(op__getfh__start, struct compound_state *, cs);
2370 
2371 	if (cs->vp == NULL) {
2372 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2373 		goto out;
2374 	}
2375 	if (cs->access == CS_ACCESS_DENIED) {
2376 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2377 		goto out;
2378 	}
2379 
2380 	/* check for reparse point at the share point */
2381 	if (cs->exi->exi_moved || vn_is_nfs_reparse(cs->exi->exi_vp, cs->cr)) {
2382 		/* it's all bad */
2383 		cs->exi->exi_moved = 1;
2384 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2385 		DTRACE_PROBE2(nfs4serv__func__referral__shared__moved,
2386 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2387 		return;
2388 	}
2389 
2390 	/* check for reparse point at vp */
2391 	if (vn_is_nfs_reparse(cs->vp, cs->cr) && !client_is_downrev(req)) {
2392 		/* it's not all bad */
2393 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2394 		DTRACE_PROBE2(nfs4serv__func__referral__moved,
2395 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2396 		return;
2397 	}
2398 
2399 	resp->object.nfs_fh4_val =
2400 	    kmem_alloc(cs->fh.nfs_fh4_len, KM_SLEEP);
2401 	nfs_fh4_copy(&cs->fh, &resp->object);
2402 	*cs->statusp = resp->status = NFS4_OK;
2403 out:
2404 	DTRACE_NFSV4_2(op__getfh__done, struct compound_state *, cs,
2405 	    GETFH4res *, resp);
2406 }
2407 
2408 static void
2409 rfs4_op_getfh_free(nfs_resop4 *resop)
2410 {
2411 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2412 
2413 	if (resp->status == NFS4_OK &&
2414 	    resp->object.nfs_fh4_val != NULL) {
2415 		kmem_free(resp->object.nfs_fh4_val, resp->object.nfs_fh4_len);
2416 		resp->object.nfs_fh4_val = NULL;
2417 		resp->object.nfs_fh4_len = 0;
2418 	}
2419 }
2420 
2421 /*
2422  * illegal: args: void
2423  *	    res : status (NFS4ERR_OP_ILLEGAL)
2424  */
2425 /* ARGSUSED */
2426 static void
2427 rfs4_op_illegal(nfs_argop4 *argop, nfs_resop4 *resop,
2428     struct svc_req *req, struct compound_state *cs)
2429 {
2430 	ILLEGAL4res *resp = &resop->nfs_resop4_u.opillegal;
2431 
2432 	resop->resop = OP_ILLEGAL;
2433 	*cs->statusp = resp->status = NFS4ERR_OP_ILLEGAL;
2434 }
2435 
2436 /*
2437  * link: args: SAVED_FH: file, CURRENT_FH: target directory
2438  *	 res: status. If success - CURRENT_FH unchanged, return change_info
2439  */
2440 /* ARGSUSED */
2441 static void
2442 rfs4_op_link(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2443     struct compound_state *cs)
2444 {
2445 	LINK4args *args = &argop->nfs_argop4_u.oplink;
2446 	LINK4res *resp = &resop->nfs_resop4_u.oplink;
2447 	int error;
2448 	vnode_t *vp;
2449 	vnode_t *dvp;
2450 	struct vattr bdva, idva, adva;
2451 	char *nm;
2452 	uint_t  len;
2453 	struct sockaddr *ca;
2454 	char *name = NULL;
2455 	nfsstat4 status;
2456 
2457 	DTRACE_NFSV4_2(op__link__start, struct compound_state *, cs,
2458 	    LINK4args *, args);
2459 
2460 	/* SAVED_FH: source object */
2461 	vp = cs->saved_vp;
2462 	if (vp == NULL) {
2463 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2464 		goto out;
2465 	}
2466 
2467 	/* CURRENT_FH: target directory */
2468 	dvp = cs->vp;
2469 	if (dvp == NULL) {
2470 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2471 		goto out;
2472 	}
2473 
2474 	/*
2475 	 * If there is a non-shared filesystem mounted on this vnode,
2476 	 * do not allow to link any file in this directory.
2477 	 */
2478 	if (vn_ismntpt(dvp)) {
2479 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2480 		goto out;
2481 	}
2482 
2483 	if (cs->access == CS_ACCESS_DENIED) {
2484 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2485 		goto out;
2486 	}
2487 
2488 	/* Check source object's type validity */
2489 	if (vp->v_type == VDIR) {
2490 		*cs->statusp = resp->status = NFS4ERR_ISDIR;
2491 		goto out;
2492 	}
2493 
2494 	/* Check target directory's type */
2495 	if (dvp->v_type != VDIR) {
2496 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2497 		goto out;
2498 	}
2499 
2500 	if (cs->saved_exi != cs->exi) {
2501 		*cs->statusp = resp->status = NFS4ERR_XDEV;
2502 		goto out;
2503 	}
2504 
2505 	status = utf8_dir_verify(&args->newname);
2506 	if (status != NFS4_OK) {
2507 		*cs->statusp = resp->status = status;
2508 		goto out;
2509 	}
2510 
2511 	nm = utf8_to_fn(&args->newname, &len, NULL);
2512 	if (nm == NULL) {
2513 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2514 		goto out;
2515 	}
2516 
2517 	if (len > MAXNAMELEN) {
2518 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2519 		kmem_free(nm, len);
2520 		goto out;
2521 	}
2522 
2523 	if (rdonly4(req, cs)) {
2524 		*cs->statusp = resp->status = NFS4ERR_ROFS;
2525 		kmem_free(nm, len);
2526 		goto out;
2527 	}
2528 
2529 	/* Get "before" change value */
2530 	bdva.va_mask = AT_CTIME|AT_SEQ;
2531 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
2532 	if (error) {
2533 		*cs->statusp = resp->status = puterrno4(error);
2534 		kmem_free(nm, len);
2535 		goto out;
2536 	}
2537 
2538 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2539 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2540 	    MAXPATHLEN  + 1);
2541 
2542 	if (name == NULL) {
2543 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2544 		kmem_free(nm, len);
2545 		goto out;
2546 	}
2547 
2548 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
2549 
2550 	error = VOP_LINK(dvp, vp, name, cs->cr, NULL, 0);
2551 
2552 	if (nm != name)
2553 		kmem_free(name, MAXPATHLEN + 1);
2554 	kmem_free(nm, len);
2555 
2556 	/*
2557 	 * Get the initial "after" sequence number, if it fails, set to zero
2558 	 */
2559 	idva.va_mask = AT_SEQ;
2560 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
2561 		idva.va_seq = 0;
2562 
2563 	/*
2564 	 * Force modified data and metadata out to stable storage.
2565 	 */
2566 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
2567 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
2568 
2569 	if (error) {
2570 		*cs->statusp = resp->status = puterrno4(error);
2571 		goto out;
2572 	}
2573 
2574 	/*
2575 	 * Get "after" change value, if it fails, simply return the
2576 	 * before value.
2577 	 */
2578 	adva.va_mask = AT_CTIME|AT_SEQ;
2579 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
2580 		adva.va_ctime = bdva.va_ctime;
2581 		adva.va_seq = 0;
2582 	}
2583 
2584 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
2585 
2586 	/*
2587 	 * The cinfo.atomic = TRUE only if we have
2588 	 * non-zero va_seq's, and it has incremented by exactly one
2589 	 * during the VOP_LINK and it didn't change during the VOP_FSYNC.
2590 	 */
2591 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
2592 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
2593 		resp->cinfo.atomic = TRUE;
2594 	else
2595 		resp->cinfo.atomic = FALSE;
2596 
2597 	*cs->statusp = resp->status = NFS4_OK;
2598 out:
2599 	DTRACE_NFSV4_2(op__link__done, struct compound_state *, cs,
2600 	    LINK4res *, resp);
2601 }
2602 
2603 /*
2604  * Used by rfs4_op_lookup and rfs4_op_lookupp to do the actual work.
2605  */
2606 
2607 /* ARGSUSED */
2608 static nfsstat4
2609 do_rfs4_op_lookup(char *nm, struct svc_req *req, struct compound_state *cs)
2610 {
2611 	int error;
2612 	int different_export = 0;
2613 	vnode_t *vp, *pre_tvp = NULL, *oldvp = NULL;
2614 	struct exportinfo *exi = NULL, *pre_exi = NULL;
2615 	nfsstat4 stat;
2616 	fid_t fid;
2617 	int attrdir, dotdot, walk;
2618 	bool_t is_newvp = FALSE;
2619 
2620 	if (cs->vp->v_flag & V_XATTRDIR) {
2621 		attrdir = 1;
2622 		ASSERT(get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2623 	} else {
2624 		attrdir = 0;
2625 		ASSERT(! get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2626 	}
2627 
2628 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
2629 
2630 	/*
2631 	 * If dotdotting, then need to check whether it's
2632 	 * above the root of a filesystem, or above an
2633 	 * export point.
2634 	 */
2635 	if (dotdot) {
2636 
2637 		/*
2638 		 * If dotdotting at the root of a filesystem, then
2639 		 * need to traverse back to the mounted-on filesystem
2640 		 * and do the dotdot lookup there.
2641 		 */
2642 		if (cs->vp->v_flag & VROOT) {
2643 
2644 			/*
2645 			 * If at the system root, then can
2646 			 * go up no further.
2647 			 */
2648 			if (VN_CMP(cs->vp, rootdir))
2649 				return (puterrno4(ENOENT));
2650 
2651 			/*
2652 			 * Traverse back to the mounted-on filesystem
2653 			 */
2654 			cs->vp = untraverse(cs->vp);
2655 
2656 			/*
2657 			 * Set the different_export flag so we remember
2658 			 * to pick up a new exportinfo entry for
2659 			 * this new filesystem.
2660 			 */
2661 			different_export = 1;
2662 		} else {
2663 
2664 			/*
2665 			 * If dotdotting above an export point then set
2666 			 * the different_export to get new export info.
2667 			 */
2668 			different_export = nfs_exported(cs->exi, cs->vp);
2669 		}
2670 	}
2671 
2672 	error = VOP_LOOKUP(cs->vp, nm, &vp, NULL, 0, NULL, cs->cr,
2673 	    NULL, NULL, NULL);
2674 	if (error)
2675 		return (puterrno4(error));
2676 
2677 	/*
2678 	 * If the vnode is in a pseudo filesystem, check whether it is visible.
2679 	 *
2680 	 * XXX if the vnode is a symlink and it is not visible in
2681 	 * a pseudo filesystem, return ENOENT (not following symlink).
2682 	 * V4 client can not mount such symlink. This is a regression
2683 	 * from V2/V3.
2684 	 *
2685 	 * In the same exported filesystem, if the security flavor used
2686 	 * is not an explicitly shared flavor, limit the view to the visible
2687 	 * list entries only. This is not a WRONGSEC case because it's already
2688 	 * checked via PUTROOTFH/PUTPUBFH or PUTFH.
2689 	 */
2690 	if (!different_export &&
2691 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
2692 	    cs->access & CS_ACCESS_LIMITED)) {
2693 		if (! nfs_visible(cs->exi, vp, &different_export)) {
2694 			VN_RELE(vp);
2695 			return (puterrno4(ENOENT));
2696 		}
2697 	}
2698 
2699 	/*
2700 	 * If it's a mountpoint, then traverse it.
2701 	 */
2702 	if (vn_ismntpt(vp)) {
2703 		pre_exi = cs->exi;	/* save pre-traversed exportinfo */
2704 		pre_tvp = vp;		/* save pre-traversed vnode	*/
2705 
2706 		/*
2707 		 * hold pre_tvp to counteract rele by traverse.  We will
2708 		 * need pre_tvp below if checkexport4 fails
2709 		 */
2710 		VN_HOLD(pre_tvp);
2711 		if ((error = traverse(&vp)) != 0) {
2712 			VN_RELE(vp);
2713 			VN_RELE(pre_tvp);
2714 			return (puterrno4(error));
2715 		}
2716 		different_export = 1;
2717 	} else if (vp->v_vfsp != cs->vp->v_vfsp) {
2718 		/*
2719 		 * The vfsp comparison is to handle the case where
2720 		 * a LOFS mount is shared.  lo_lookup traverses mount points,
2721 		 * and NFS is unaware of local fs transistions because
2722 		 * v_vfsmountedhere isn't set.  For this special LOFS case,
2723 		 * the dir and the obj returned by lookup will have different
2724 		 * vfs ptrs.
2725 		 */
2726 		different_export = 1;
2727 	}
2728 
2729 	if (different_export) {
2730 
2731 		bzero(&fid, sizeof (fid));
2732 		fid.fid_len = MAXFIDSZ;
2733 		error = vop_fid_pseudo(vp, &fid);
2734 		if (error) {
2735 			VN_RELE(vp);
2736 			if (pre_tvp)
2737 				VN_RELE(pre_tvp);
2738 			return (puterrno4(error));
2739 		}
2740 
2741 		if (dotdot)
2742 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
2743 		else
2744 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2745 
2746 		if (exi == NULL) {
2747 			if (pre_tvp) {
2748 				/*
2749 				 * If this vnode is a mounted-on vnode,
2750 				 * but the mounted-on file system is not
2751 				 * exported, send back the filehandle for
2752 				 * the mounted-on vnode, not the root of
2753 				 * the mounted-on file system.
2754 				 */
2755 				VN_RELE(vp);
2756 				vp = pre_tvp;
2757 				exi = pre_exi;
2758 			} else {
2759 				VN_RELE(vp);
2760 				return (puterrno4(EACCES));
2761 			}
2762 		} else if (pre_tvp) {
2763 			/* we're done with pre_tvp now. release extra hold */
2764 			VN_RELE(pre_tvp);
2765 		}
2766 
2767 		cs->exi = exi;
2768 
2769 		/*
2770 		 * Now we do a checkauth4. The reason is that
2771 		 * this client/user may not have access to the new
2772 		 * exported file system, and if he does,
2773 		 * the client/user may be mapped to a different uid.
2774 		 *
2775 		 * We start with a new cr, because the checkauth4 done
2776 		 * in the PUT*FH operation over wrote the cred's uid,
2777 		 * gid, etc, and we want the real thing before calling
2778 		 * checkauth4()
2779 		 */
2780 		crfree(cs->cr);
2781 		cs->cr = crdup(cs->basecr);
2782 
2783 		oldvp = cs->vp;
2784 		cs->vp = vp;
2785 		is_newvp = TRUE;
2786 
2787 		stat = call_checkauth4(cs, req);
2788 		if (stat != NFS4_OK) {
2789 			VN_RELE(cs->vp);
2790 			cs->vp = oldvp;
2791 			return (stat);
2792 		}
2793 	}
2794 
2795 	/*
2796 	 * After various NFS checks, do a label check on the path
2797 	 * component. The label on this path should either be the
2798 	 * global zone's label or a zone's label. We are only
2799 	 * interested in the zone's label because exported files
2800 	 * in global zone is accessible (though read-only) to
2801 	 * clients. The exportability/visibility check is already
2802 	 * done before reaching this code.
2803 	 */
2804 	if (is_system_labeled()) {
2805 		bslabel_t *clabel;
2806 
2807 		ASSERT(req->rq_label != NULL);
2808 		clabel = req->rq_label;
2809 		DTRACE_PROBE2(tx__rfs4__log__info__oplookup__clabel, char *,
2810 		    "got client label from request(1)", struct svc_req *, req);
2811 
2812 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
2813 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
2814 			    cs->exi)) {
2815 				error = EACCES;
2816 				goto err_out;
2817 			}
2818 		} else {
2819 			/*
2820 			 * We grant access to admin_low label clients
2821 			 * only if the client is trusted, i.e. also
2822 			 * running Solaris Trusted Extension.
2823 			 */
2824 			struct sockaddr	*ca;
2825 			int		addr_type;
2826 			void		*ipaddr;
2827 			tsol_tpc_t	*tp;
2828 
2829 			ca = (struct sockaddr *)svc_getrpccaller(
2830 			    req->rq_xprt)->buf;
2831 			if (ca->sa_family == AF_INET) {
2832 				addr_type = IPV4_VERSION;
2833 				ipaddr = &((struct sockaddr_in *)ca)->sin_addr;
2834 			} else if (ca->sa_family == AF_INET6) {
2835 				addr_type = IPV6_VERSION;
2836 				ipaddr = &((struct sockaddr_in6 *)
2837 				    ca)->sin6_addr;
2838 			}
2839 			tp = find_tpc(ipaddr, addr_type, B_FALSE);
2840 			if (tp == NULL || tp->tpc_tp.tp_doi !=
2841 			    l_admin_low->tsl_doi || tp->tpc_tp.host_type !=
2842 			    SUN_CIPSO) {
2843 				if (tp != NULL)
2844 					TPC_RELE(tp);
2845 				error = EACCES;
2846 				goto err_out;
2847 			}
2848 			TPC_RELE(tp);
2849 		}
2850 	}
2851 
2852 	error = makefh4(&cs->fh, vp, cs->exi);
2853 
2854 err_out:
2855 	if (error) {
2856 		if (is_newvp) {
2857 			VN_RELE(cs->vp);
2858 			cs->vp = oldvp;
2859 		} else
2860 			VN_RELE(vp);
2861 		return (puterrno4(error));
2862 	}
2863 
2864 	if (!is_newvp) {
2865 		if (cs->vp)
2866 			VN_RELE(cs->vp);
2867 		cs->vp = vp;
2868 	} else if (oldvp)
2869 		VN_RELE(oldvp);
2870 
2871 	/*
2872 	 * if did lookup on attrdir and didn't lookup .., set named
2873 	 * attr fh flag
2874 	 */
2875 	if (attrdir && ! dotdot)
2876 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
2877 
2878 	/* Assume false for now, open proc will set this */
2879 	cs->mandlock = FALSE;
2880 
2881 	return (NFS4_OK);
2882 }
2883 
2884 /* ARGSUSED */
2885 static void
2886 rfs4_op_lookup(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2887     struct compound_state *cs)
2888 {
2889 	LOOKUP4args *args = &argop->nfs_argop4_u.oplookup;
2890 	LOOKUP4res *resp = &resop->nfs_resop4_u.oplookup;
2891 	char *nm;
2892 	uint_t len;
2893 	struct sockaddr *ca;
2894 	char *name = NULL;
2895 	nfsstat4 status;
2896 
2897 	DTRACE_NFSV4_2(op__lookup__start, struct compound_state *, cs,
2898 	    LOOKUP4args *, args);
2899 
2900 	if (cs->vp == NULL) {
2901 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2902 		goto out;
2903 	}
2904 
2905 	if (cs->vp->v_type == VLNK) {
2906 		*cs->statusp = resp->status = NFS4ERR_SYMLINK;
2907 		goto out;
2908 	}
2909 
2910 	if (cs->vp->v_type != VDIR) {
2911 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2912 		goto out;
2913 	}
2914 
2915 	status = utf8_dir_verify(&args->objname);
2916 	if (status != NFS4_OK) {
2917 		*cs->statusp = resp->status = status;
2918 		goto out;
2919 	}
2920 
2921 	nm = utf8_to_str(&args->objname, &len, NULL);
2922 	if (nm == NULL) {
2923 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2924 		goto out;
2925 	}
2926 
2927 	if (len > MAXNAMELEN) {
2928 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2929 		kmem_free(nm, len);
2930 		goto out;
2931 	}
2932 
2933 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2934 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2935 	    MAXPATHLEN  + 1);
2936 
2937 	if (name == NULL) {
2938 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2939 		kmem_free(nm, len);
2940 		goto out;
2941 	}
2942 
2943 	*cs->statusp = resp->status = do_rfs4_op_lookup(name, req, cs);
2944 
2945 	if (name != nm)
2946 		kmem_free(name, MAXPATHLEN + 1);
2947 	kmem_free(nm, len);
2948 
2949 out:
2950 	DTRACE_NFSV4_2(op__lookup__done, struct compound_state *, cs,
2951 	    LOOKUP4res *, resp);
2952 }
2953 
2954 /* ARGSUSED */
2955 static void
2956 rfs4_op_lookupp(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
2957     struct compound_state *cs)
2958 {
2959 	LOOKUPP4res *resp = &resop->nfs_resop4_u.oplookupp;
2960 
2961 	DTRACE_NFSV4_1(op__lookupp__start, struct compound_state *, cs);
2962 
2963 	if (cs->vp == NULL) {
2964 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2965 		goto out;
2966 	}
2967 
2968 	if (cs->vp->v_type != VDIR) {
2969 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2970 		goto out;
2971 	}
2972 
2973 	*cs->statusp = resp->status = do_rfs4_op_lookup("..", req, cs);
2974 
2975 	/*
2976 	 * From NFSV4 Specification, LOOKUPP should not check for
2977 	 * NFS4ERR_WRONGSEC. Retrun NFS4_OK instead.
2978 	 */
2979 	if (resp->status == NFS4ERR_WRONGSEC) {
2980 		*cs->statusp = resp->status = NFS4_OK;
2981 	}
2982 
2983 out:
2984 	DTRACE_NFSV4_2(op__lookupp__done, struct compound_state *, cs,
2985 	    LOOKUPP4res *, resp);
2986 }
2987 
2988 
2989 /*ARGSUSED2*/
2990 static void
2991 rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2992     struct compound_state *cs)
2993 {
2994 	OPENATTR4args	*args = &argop->nfs_argop4_u.opopenattr;
2995 	OPENATTR4res	*resp = &resop->nfs_resop4_u.opopenattr;
2996 	vnode_t		*avp = NULL;
2997 	int		lookup_flags = LOOKUP_XATTR, error;
2998 	int		exp_ro = 0;
2999 
3000 	DTRACE_NFSV4_2(op__openattr__start, struct compound_state *, cs,
3001 	    OPENATTR4args *, args);
3002 
3003 	if (cs->vp == NULL) {
3004 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3005 		goto out;
3006 	}
3007 
3008 	if ((cs->vp->v_vfsp->vfs_flag & VFS_XATTR) == 0 &&
3009 	    !vfs_has_feature(cs->vp->v_vfsp, VFSFT_SYSATTR_VIEWS)) {
3010 		*cs->statusp = resp->status = puterrno4(ENOTSUP);
3011 		goto out;
3012 	}
3013 
3014 	/*
3015 	 * If file system supports passing ACE mask to VOP_ACCESS then
3016 	 * check for ACE_READ_NAMED_ATTRS, otherwise do legacy checks
3017 	 */
3018 
3019 	if (vfs_has_feature(cs->vp->v_vfsp, VFSFT_ACEMASKONACCESS))
3020 		error = VOP_ACCESS(cs->vp, ACE_READ_NAMED_ATTRS,
3021 		    V_ACE_MASK, cs->cr, NULL);
3022 	else
3023 		error = ((VOP_ACCESS(cs->vp, VREAD, 0, cs->cr, NULL) != 0) &&
3024 		    (VOP_ACCESS(cs->vp, VWRITE, 0, cs->cr, NULL) != 0) &&
3025 		    (VOP_ACCESS(cs->vp, VEXEC, 0, cs->cr, NULL) != 0));
3026 
3027 	if (error) {
3028 		*cs->statusp = resp->status = puterrno4(EACCES);
3029 		goto out;
3030 	}
3031 
3032 	/*
3033 	 * The CREATE_XATTR_DIR VOP flag cannot be specified if
3034 	 * the file system is exported read-only -- regardless of
3035 	 * createdir flag.  Otherwise the attrdir would be created
3036 	 * (assuming server fs isn't mounted readonly locally).  If
3037 	 * VOP_LOOKUP returns ENOENT in this case, the error will
3038 	 * be translated into EROFS.  ENOSYS is mapped to ENOTSUP
3039 	 * because specfs has no VOP_LOOKUP op, so the macro would
3040 	 * return ENOSYS.  EINVAL is returned by all (current)
3041 	 * Solaris file system implementations when any of their
3042 	 * restrictions are violated (xattr(dir) can't have xattrdir).
3043 	 * Returning NOTSUPP is more appropriate in this case
3044 	 * because the object will never be able to have an attrdir.
3045 	 */
3046 	if (args->createdir && ! (exp_ro = rdonly4(req, cs)))
3047 		lookup_flags |= CREATE_XATTR_DIR;
3048 
3049 	error = VOP_LOOKUP(cs->vp, "", &avp, NULL, lookup_flags, NULL, cs->cr,
3050 	    NULL, NULL, NULL);
3051 
3052 	if (error) {
3053 		if (error == ENOENT && args->createdir && exp_ro)
3054 			*cs->statusp = resp->status = puterrno4(EROFS);
3055 		else if (error == EINVAL || error == ENOSYS)
3056 			*cs->statusp = resp->status = puterrno4(ENOTSUP);
3057 		else
3058 			*cs->statusp = resp->status = puterrno4(error);
3059 		goto out;
3060 	}
3061 
3062 	ASSERT(avp->v_flag & V_XATTRDIR);
3063 
3064 	error = makefh4(&cs->fh, avp, cs->exi);
3065 
3066 	if (error) {
3067 		VN_RELE(avp);
3068 		*cs->statusp = resp->status = puterrno4(error);
3069 		goto out;
3070 	}
3071 
3072 	VN_RELE(cs->vp);
3073 	cs->vp = avp;
3074 
3075 	/*
3076 	 * There is no requirement for an attrdir fh flag
3077 	 * because the attrdir has a vnode flag to distinguish
3078 	 * it from regular (non-xattr) directories.  The
3079 	 * FH4_ATTRDIR flag is set for future sanity checks.
3080 	 */
3081 	set_fh4_flag(&cs->fh, FH4_ATTRDIR);
3082 	*cs->statusp = resp->status = NFS4_OK;
3083 
3084 out:
3085 	DTRACE_NFSV4_2(op__openattr__done, struct compound_state *, cs,
3086 	    OPENATTR4res *, resp);
3087 }
3088 
3089 static int
3090 do_io(int direction, vnode_t *vp, struct uio *uio, int ioflag, cred_t *cred,
3091     caller_context_t *ct)
3092 {
3093 	int error;
3094 	int i;
3095 	clock_t delaytime;
3096 
3097 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
3098 
3099 	/*
3100 	 * Don't block on mandatory locks. If this routine returns
3101 	 * EAGAIN, the caller should return NFS4ERR_LOCKED.
3102 	 */
3103 	uio->uio_fmode = FNONBLOCK;
3104 
3105 	for (i = 0; i < rfs4_maxlock_tries; i++) {
3106 
3107 
3108 		if (direction == FREAD) {
3109 			(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, ct);
3110 			error = VOP_READ(vp, uio, ioflag, cred, ct);
3111 			VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, ct);
3112 		} else {
3113 			(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, ct);
3114 			error = VOP_WRITE(vp, uio, ioflag, cred, ct);
3115 			VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, ct);
3116 		}
3117 
3118 		if (error != EAGAIN)
3119 			break;
3120 
3121 		if (i < rfs4_maxlock_tries - 1) {
3122 			delay(delaytime);
3123 			delaytime *= 2;
3124 		}
3125 	}
3126 
3127 	return (error);
3128 }
3129 
3130 /* ARGSUSED */
3131 static void
3132 rfs4_op_read(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3133     struct compound_state *cs)
3134 {
3135 	READ4args *args = &argop->nfs_argop4_u.opread;
3136 	READ4res *resp = &resop->nfs_resop4_u.opread;
3137 	int error;
3138 	int verror;
3139 	vnode_t *vp;
3140 	struct vattr va;
3141 	struct iovec iov, *iovp = NULL;
3142 	int iovcnt;
3143 	struct uio uio;
3144 	u_offset_t offset;
3145 	bool_t *deleg = &cs->deleg;
3146 	nfsstat4 stat;
3147 	int in_crit = 0;
3148 	mblk_t *mp = NULL;
3149 	int alloc_err = 0;
3150 	int rdma_used = 0;
3151 	int loaned_buffers;
3152 	caller_context_t ct;
3153 	struct uio *uiop;
3154 
3155 	DTRACE_NFSV4_2(op__read__start, struct compound_state *, cs,
3156 	    READ4args, args);
3157 
3158 	vp = cs->vp;
3159 	if (vp == NULL) {
3160 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3161 		goto out;
3162 	}
3163 	if (cs->access == CS_ACCESS_DENIED) {
3164 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3165 		goto out;
3166 	}
3167 
3168 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3169 	    deleg, TRUE, &ct)) != NFS4_OK) {
3170 		*cs->statusp = resp->status = stat;
3171 		goto out;
3172 	}
3173 
3174 	/*
3175 	 * Enter the critical region before calling VOP_RWLOCK
3176 	 * to avoid a deadlock with write requests.
3177 	 */
3178 	if (nbl_need_check(vp)) {
3179 		nbl_start_crit(vp, RW_READER);
3180 		in_crit = 1;
3181 		if (nbl_conflict(vp, NBL_READ, args->offset, args->count, 0,
3182 		    &ct)) {
3183 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
3184 			goto out;
3185 		}
3186 	}
3187 
3188 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3189 	    deleg, TRUE, &ct)) != NFS4_OK) {
3190 		*cs->statusp = resp->status = stat;
3191 		goto out;
3192 	}
3193 
3194 	if (args->wlist) {
3195 		if (args->count > clist_len(args->wlist)) {
3196 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3197 			goto out;
3198 		}
3199 		rdma_used = 1;
3200 	}
3201 
3202 	/* use loaned buffers for TCP */
3203 	loaned_buffers = (nfs_loaned_buffers && !rdma_used) ? 1 : 0;
3204 
3205 	va.va_mask = AT_MODE|AT_SIZE|AT_UID;
3206 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3207 
3208 	/*
3209 	 * If we can't get the attributes, then we can't do the
3210 	 * right access checking.  So, we'll fail the request.
3211 	 */
3212 	if (verror) {
3213 		*cs->statusp = resp->status = puterrno4(verror);
3214 		goto out;
3215 	}
3216 
3217 	if (vp->v_type != VREG) {
3218 		*cs->statusp = resp->status =
3219 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
3220 		goto out;
3221 	}
3222 
3223 	if (crgetuid(cs->cr) != va.va_uid &&
3224 	    (error = VOP_ACCESS(vp, VREAD, 0, cs->cr, &ct)) &&
3225 	    (error = VOP_ACCESS(vp, VEXEC, 0, cs->cr, &ct))) {
3226 		*cs->statusp = resp->status = puterrno4(error);
3227 		goto out;
3228 	}
3229 
3230 	if (MANDLOCK(vp, va.va_mode)) { /* XXX - V4 supports mand locking */
3231 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3232 		goto out;
3233 	}
3234 
3235 	offset = args->offset;
3236 	if (offset >= va.va_size) {
3237 		*cs->statusp = resp->status = NFS4_OK;
3238 		resp->eof = TRUE;
3239 		resp->data_len = 0;
3240 		resp->data_val = NULL;
3241 		resp->mblk = NULL;
3242 		/* RDMA */
3243 		resp->wlist = args->wlist;
3244 		resp->wlist_len = resp->data_len;
3245 		*cs->statusp = resp->status = NFS4_OK;
3246 		if (resp->wlist)
3247 			clist_zero_len(resp->wlist);
3248 		goto out;
3249 	}
3250 
3251 	if (args->count == 0) {
3252 		*cs->statusp = resp->status = NFS4_OK;
3253 		resp->eof = FALSE;
3254 		resp->data_len = 0;
3255 		resp->data_val = NULL;
3256 		resp->mblk = NULL;
3257 		/* RDMA */
3258 		resp->wlist = args->wlist;
3259 		resp->wlist_len = resp->data_len;
3260 		if (resp->wlist)
3261 			clist_zero_len(resp->wlist);
3262 		goto out;
3263 	}
3264 
3265 	/*
3266 	 * Do not allocate memory more than maximum allowed
3267 	 * transfer size
3268 	 */
3269 	if (args->count > rfs4_tsize(req))
3270 		args->count = rfs4_tsize(req);
3271 
3272 	if (loaned_buffers) {
3273 		uiop = (uio_t *)rfs_setup_xuio(vp);
3274 		ASSERT(uiop != NULL);
3275 		uiop->uio_segflg = UIO_SYSSPACE;
3276 		uiop->uio_loffset = args->offset;
3277 		uiop->uio_resid = args->count;
3278 
3279 		/* Jump to do the read if successful */
3280 		if (!VOP_REQZCBUF(vp, UIO_READ, (xuio_t *)uiop, cs->cr, &ct)) {
3281 			/*
3282 			 * Need to hold the vnode until after VOP_RETZCBUF()
3283 			 * is called.
3284 			 */
3285 			VN_HOLD(vp);
3286 			goto doio_read;
3287 		}
3288 
3289 		DTRACE_PROBE2(nfss__i__reqzcbuf_failed, int,
3290 		    uiop->uio_loffset, int, uiop->uio_resid);
3291 
3292 		uiop->uio_extflg = 0;
3293 
3294 		/* failure to setup for zero copy */
3295 		rfs_free_xuio((void *)uiop);
3296 		loaned_buffers = 0;
3297 	}
3298 
3299 	/*
3300 	 * If returning data via RDMA Write, then grab the chunk list. If we
3301 	 * aren't returning READ data w/RDMA_WRITE, then grab a mblk.
3302 	 */
3303 	if (rdma_used) {
3304 		mp = NULL;
3305 		(void) rdma_get_wchunk(req, &iov, args->wlist);
3306 		uio.uio_iov = &iov;
3307 		uio.uio_iovcnt = 1;
3308 	} else {
3309 		/*
3310 		 * mp will contain the data to be sent out in the read reply.
3311 		 * It will be freed after the reply has been sent.
3312 		 */
3313 		mp = rfs_read_alloc(args->count, &iovp, &iovcnt);
3314 		ASSERT(mp != NULL);
3315 		ASSERT(alloc_err == 0);
3316 		uio.uio_iov = iovp;
3317 		uio.uio_iovcnt = iovcnt;
3318 	}
3319 
3320 	uio.uio_segflg = UIO_SYSSPACE;
3321 	uio.uio_extflg = UIO_COPY_CACHED;
3322 	uio.uio_loffset = args->offset;
3323 	uio.uio_resid = args->count;
3324 	uiop = &uio;
3325 
3326 doio_read:
3327 	error = do_io(FREAD, vp, uiop, 0, cs->cr, &ct);
3328 
3329 	va.va_mask = AT_SIZE;
3330 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3331 
3332 	if (error) {
3333 		if (mp)
3334 			freemsg(mp);
3335 		*cs->statusp = resp->status = puterrno4(error);
3336 		goto out;
3337 	}
3338 
3339 	/* make mblk using zc buffers */
3340 	if (loaned_buffers) {
3341 		mp = uio_to_mblk(uiop);
3342 		ASSERT(mp != NULL);
3343 	}
3344 
3345 	*cs->statusp = resp->status = NFS4_OK;
3346 
3347 	ASSERT(uiop->uio_resid >= 0);
3348 	resp->data_len = args->count - uiop->uio_resid;
3349 	if (mp) {
3350 		resp->data_val = (char *)mp->b_datap->db_base;
3351 		rfs_rndup_mblks(mp, resp->data_len, loaned_buffers);
3352 	} else {
3353 		resp->data_val = (caddr_t)iov.iov_base;
3354 	}
3355 
3356 	resp->mblk = mp;
3357 
3358 	if (!verror && offset + resp->data_len == va.va_size)
3359 		resp->eof = TRUE;
3360 	else
3361 		resp->eof = FALSE;
3362 
3363 	if (rdma_used) {
3364 		if (!rdma_setup_read_data4(args, resp)) {
3365 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3366 		}
3367 	} else {
3368 		resp->wlist = NULL;
3369 	}
3370 
3371 out:
3372 	if (in_crit)
3373 		nbl_end_crit(vp);
3374 
3375 	if (iovp != NULL)
3376 		kmem_free(iovp, iovcnt * sizeof (struct iovec));
3377 
3378 	DTRACE_NFSV4_2(op__read__done, struct compound_state *, cs,
3379 	    READ4res *, resp);
3380 }
3381 
3382 static void
3383 rfs4_op_read_free(nfs_resop4 *resop)
3384 {
3385 	READ4res	*resp = &resop->nfs_resop4_u.opread;
3386 
3387 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3388 		freemsg(resp->mblk);
3389 		resp->mblk = NULL;
3390 		resp->data_val = NULL;
3391 		resp->data_len = 0;
3392 	}
3393 }
3394 
3395 static void
3396 rfs4_op_readdir_free(nfs_resop4 * resop)
3397 {
3398 	READDIR4res    *resp = &resop->nfs_resop4_u.opreaddir;
3399 
3400 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3401 		freeb(resp->mblk);
3402 		resp->mblk = NULL;
3403 		resp->data_len = 0;
3404 	}
3405 }
3406 
3407 
3408 /* ARGSUSED */
3409 static void
3410 rfs4_op_putpubfh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
3411     struct compound_state *cs)
3412 {
3413 	PUTPUBFH4res	*resp = &resop->nfs_resop4_u.opputpubfh;
3414 	int		error;
3415 	vnode_t		*vp;
3416 	struct exportinfo *exi, *sav_exi;
3417 	nfs_fh4_fmt_t	*fh_fmtp;
3418 
3419 	DTRACE_NFSV4_1(op__putpubfh__start, struct compound_state *, cs);
3420 
3421 	if (cs->vp) {
3422 		VN_RELE(cs->vp);
3423 		cs->vp = NULL;
3424 	}
3425 
3426 	if (cs->cr)
3427 		crfree(cs->cr);
3428 
3429 	cs->cr = crdup(cs->basecr);
3430 
3431 	vp = exi_public->exi_vp;
3432 	if (vp == NULL) {
3433 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3434 		goto out;
3435 	}
3436 
3437 	error = makefh4(&cs->fh, vp, exi_public);
3438 	if (error != 0) {
3439 		*cs->statusp = resp->status = puterrno4(error);
3440 		goto out;
3441 	}
3442 	sav_exi = cs->exi;
3443 	if (exi_public == exi_root) {
3444 		/*
3445 		 * No filesystem is actually shared public, so we default
3446 		 * to exi_root. In this case, we must check whether root
3447 		 * is exported.
3448 		 */
3449 		fh_fmtp = (nfs_fh4_fmt_t *)cs->fh.nfs_fh4_val;
3450 
3451 		/*
3452 		 * if root filesystem is exported, the exportinfo struct that we
3453 		 * should use is what checkexport4 returns, because root_exi is
3454 		 * actually a mostly empty struct.
3455 		 */
3456 		exi = checkexport4(&fh_fmtp->fh4_fsid,
3457 		    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
3458 		cs->exi = ((exi != NULL) ? exi : exi_public);
3459 	} else {
3460 		/*
3461 		 * it's a properly shared filesystem
3462 		 */
3463 		cs->exi = exi_public;
3464 	}
3465 
3466 	if (is_system_labeled()) {
3467 		bslabel_t *clabel;
3468 
3469 		ASSERT(req->rq_label != NULL);
3470 		clabel = req->rq_label;
3471 		DTRACE_PROBE2(tx__rfs4__log__info__opputpubfh__clabel, char *,
3472 		    "got client label from request(1)",
3473 		    struct svc_req *, req);
3474 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
3475 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
3476 			    cs->exi)) {
3477 				*cs->statusp = resp->status =
3478 				    NFS4ERR_SERVERFAULT;
3479 				goto out;
3480 			}
3481 		}
3482 	}
3483 
3484 	VN_HOLD(vp);
3485 	cs->vp = vp;
3486 
3487 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3488 		VN_RELE(cs->vp);
3489 		cs->vp = NULL;
3490 		cs->exi = sav_exi;
3491 		goto out;
3492 	}
3493 
3494 	*cs->statusp = resp->status = NFS4_OK;
3495 out:
3496 	DTRACE_NFSV4_2(op__putpubfh__done, struct compound_state *, cs,
3497 	    PUTPUBFH4res *, resp);
3498 }
3499 
3500 /*
3501  * XXX - issue with put*fh operations. Suppose /export/home is exported.
3502  * Suppose an NFS client goes to mount /export/home/joe. If /export, home,
3503  * or joe have restrictive search permissions, then we shouldn't let
3504  * the client get a file handle. This is easy to enforce. However, we
3505  * don't know what security flavor should be used until we resolve the
3506  * path name. Another complication is uid mapping. If root is
3507  * the user, then it will be mapped to the anonymous user by default,
3508  * but we won't know that till we've resolved the path name. And we won't
3509  * know what the anonymous user is.
3510  * Luckily, SECINFO is specified to take a full filename.
3511  * So what we will have to in rfs4_op_lookup is check that flavor of
3512  * the target object matches that of the request, and if root was the
3513  * caller, check for the root= and anon= options, and if necessary,
3514  * repeat the lookup using the right cred_t. But that's not done yet.
3515  */
3516 /* ARGSUSED */
3517 static void
3518 rfs4_op_putfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3519     struct compound_state *cs)
3520 {
3521 	PUTFH4args *args = &argop->nfs_argop4_u.opputfh;
3522 	PUTFH4res *resp = &resop->nfs_resop4_u.opputfh;
3523 	nfs_fh4_fmt_t *fh_fmtp;
3524 
3525 	DTRACE_NFSV4_2(op__putfh__start, struct compound_state *, cs,
3526 	    PUTFH4args *, args);
3527 
3528 	if (cs->vp) {
3529 		VN_RELE(cs->vp);
3530 		cs->vp = NULL;
3531 	}
3532 
3533 	if (cs->cr) {
3534 		crfree(cs->cr);
3535 		cs->cr = NULL;
3536 	}
3537 
3538 
3539 	if (args->object.nfs_fh4_len < NFS_FH4_LEN) {
3540 		*cs->statusp = resp->status = NFS4ERR_BADHANDLE;
3541 		goto out;
3542 	}
3543 
3544 	fh_fmtp = (nfs_fh4_fmt_t *)args->object.nfs_fh4_val;
3545 	cs->exi = checkexport4(&fh_fmtp->fh4_fsid, (fid_t *)&fh_fmtp->fh4_xlen,
3546 	    NULL);
3547 
3548 	if (cs->exi == NULL) {
3549 		*cs->statusp = resp->status = NFS4ERR_STALE;
3550 		goto out;
3551 	}
3552 
3553 	cs->cr = crdup(cs->basecr);
3554 
3555 	ASSERT(cs->cr != NULL);
3556 
3557 	if (! (cs->vp = nfs4_fhtovp(&args->object, cs->exi, &resp->status))) {
3558 		*cs->statusp = resp->status;
3559 		goto out;
3560 	}
3561 
3562 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3563 		VN_RELE(cs->vp);
3564 		cs->vp = NULL;
3565 		goto out;
3566 	}
3567 
3568 	nfs_fh4_copy(&args->object, &cs->fh);
3569 	*cs->statusp = resp->status = NFS4_OK;
3570 	cs->deleg = FALSE;
3571 
3572 out:
3573 	DTRACE_NFSV4_2(op__putfh__done, struct compound_state *, cs,
3574 	    PUTFH4res *, resp);
3575 }
3576 
3577 /* ARGSUSED */
3578 static void
3579 rfs4_op_putrootfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3580     struct compound_state *cs)
3581 {
3582 	PUTROOTFH4res *resp = &resop->nfs_resop4_u.opputrootfh;
3583 	int error;
3584 	fid_t fid;
3585 	struct exportinfo *exi, *sav_exi;
3586 
3587 	DTRACE_NFSV4_1(op__putrootfh__start, struct compound_state *, cs);
3588 
3589 	if (cs->vp) {
3590 		VN_RELE(cs->vp);
3591 		cs->vp = NULL;
3592 	}
3593 
3594 	if (cs->cr)
3595 		crfree(cs->cr);
3596 
3597 	cs->cr = crdup(cs->basecr);
3598 
3599 	/*
3600 	 * Using rootdir, the system root vnode,
3601 	 * get its fid.
3602 	 */
3603 	bzero(&fid, sizeof (fid));
3604 	fid.fid_len = MAXFIDSZ;
3605 	error = vop_fid_pseudo(rootdir, &fid);
3606 	if (error != 0) {
3607 		*cs->statusp = resp->status = puterrno4(error);
3608 		goto out;
3609 	}
3610 
3611 	/*
3612 	 * Then use the root fsid & fid it to find out if it's exported
3613 	 *
3614 	 * If the server root isn't exported directly, then
3615 	 * it should at least be a pseudo export based on
3616 	 * one or more exports further down in the server's
3617 	 * file tree.
3618 	 */
3619 	exi = checkexport4(&rootdir->v_vfsp->vfs_fsid, &fid, NULL);
3620 	if (exi == NULL || exi->exi_export.ex_flags & EX_PUBLIC) {
3621 		NFS4_DEBUG(rfs4_debug,
3622 		    (CE_WARN, "rfs4_op_putrootfh: export check failure"));
3623 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3624 		goto out;
3625 	}
3626 
3627 	/*
3628 	 * Now make a filehandle based on the root
3629 	 * export and root vnode.
3630 	 */
3631 	error = makefh4(&cs->fh, rootdir, exi);
3632 	if (error != 0) {
3633 		*cs->statusp = resp->status = puterrno4(error);
3634 		goto out;
3635 	}
3636 
3637 	sav_exi = cs->exi;
3638 	cs->exi = exi;
3639 
3640 	VN_HOLD(rootdir);
3641 	cs->vp = rootdir;
3642 
3643 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3644 		VN_RELE(rootdir);
3645 		cs->vp = NULL;
3646 		cs->exi = sav_exi;
3647 		goto out;
3648 	}
3649 
3650 	*cs->statusp = resp->status = NFS4_OK;
3651 	cs->deleg = FALSE;
3652 out:
3653 	DTRACE_NFSV4_2(op__putrootfh__done, struct compound_state *, cs,
3654 	    PUTROOTFH4res *, resp);
3655 }
3656 
3657 /*
3658  * set_rdattr_params sets up the variables used to manage what information
3659  * to get for each directory entry.
3660  */
3661 static nfsstat4
3662 set_rdattr_params(struct nfs4_svgetit_arg *sargp,
3663     bitmap4 attrs, bool_t *need_to_lookup)
3664 {
3665 	uint_t	va_mask;
3666 	nfsstat4 status;
3667 	bitmap4 objbits;
3668 
3669 	status = bitmap4_to_attrmask(attrs, sargp);
3670 	if (status != NFS4_OK) {
3671 		/*
3672 		 * could not even figure attr mask
3673 		 */
3674 		return (status);
3675 	}
3676 	va_mask = sargp->vap->va_mask;
3677 
3678 	/*
3679 	 * dirent's d_ino is always correct value for mounted_on_fileid.
3680 	 * mntdfid_set is set once here, but mounted_on_fileid is
3681 	 * set in main dirent processing loop for each dirent.
3682 	 * The mntdfid_set is a simple optimization that lets the
3683 	 * server attr code avoid work when caller is readdir.
3684 	 */
3685 	sargp->mntdfid_set = TRUE;
3686 
3687 	/*
3688 	 * Lookup entry only if client asked for any of the following:
3689 	 * a) vattr attrs
3690 	 * b) vfs attrs
3691 	 * c) attrs w/per-object scope requested (change, filehandle, etc)
3692 	 *    other than mounted_on_fileid (which we can take from dirent)
3693 	 */
3694 	objbits = attrs ? attrs & NFS4_VP_ATTR_MASK : 0;
3695 
3696 	if (va_mask || sargp->sbp || (objbits & ~FATTR4_MOUNTED_ON_FILEID_MASK))
3697 		*need_to_lookup = TRUE;
3698 	else
3699 		*need_to_lookup = FALSE;
3700 
3701 	if (sargp->sbp == NULL)
3702 		return (NFS4_OK);
3703 
3704 	/*
3705 	 * If filesystem attrs are requested, get them now from the
3706 	 * directory vp, as most entries will have same filesystem. The only
3707 	 * exception are mounted over entries but we handle
3708 	 * those as we go (XXX mounted over detection not yet implemented).
3709 	 */
3710 	sargp->vap->va_mask = 0;	/* to avoid VOP_GETATTR */
3711 	status = bitmap4_get_sysattrs(sargp);
3712 	sargp->vap->va_mask = va_mask;
3713 
3714 	if ((status != NFS4_OK) && sargp->rdattr_error_req) {
3715 		/*
3716 		 * Failed to get filesystem attributes.
3717 		 * Return a rdattr_error for each entry, but don't fail.
3718 		 * However, don't get any obj-dependent attrs.
3719 		 */
3720 		sargp->rdattr_error = status;	/* for rdattr_error */
3721 		*need_to_lookup = FALSE;
3722 		/*
3723 		 * At least get fileid for regular readdir output
3724 		 */
3725 		sargp->vap->va_mask &= AT_NODEID;
3726 		status = NFS4_OK;
3727 	}
3728 
3729 	return (status);
3730 }
3731 
3732 /*
3733  * readlink: args: CURRENT_FH.
3734  *	res: status. If success - CURRENT_FH unchanged, return linktext.
3735  */
3736 
3737 /* ARGSUSED */
3738 static void
3739 rfs4_op_readlink(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3740     struct compound_state *cs)
3741 {
3742 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3743 	int error;
3744 	vnode_t *vp;
3745 	struct iovec iov;
3746 	struct vattr va;
3747 	struct uio uio;
3748 	char *data;
3749 	struct sockaddr *ca;
3750 	char *name = NULL;
3751 	int is_referral;
3752 
3753 	DTRACE_NFSV4_1(op__readlink__start, struct compound_state *, cs);
3754 
3755 	/* CURRENT_FH: directory */
3756 	vp = cs->vp;
3757 	if (vp == NULL) {
3758 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3759 		goto out;
3760 	}
3761 
3762 	if (cs->access == CS_ACCESS_DENIED) {
3763 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3764 		goto out;
3765 	}
3766 
3767 	/* Is it a referral? */
3768 	if (vn_is_nfs_reparse(vp, cs->cr) && client_is_downrev(req)) {
3769 
3770 		is_referral = 1;
3771 
3772 	} else {
3773 
3774 		is_referral = 0;
3775 
3776 		if (vp->v_type == VDIR) {
3777 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
3778 			goto out;
3779 		}
3780 
3781 		if (vp->v_type != VLNK) {
3782 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3783 			goto out;
3784 		}
3785 
3786 	}
3787 
3788 	va.va_mask = AT_MODE;
3789 	error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
3790 	if (error) {
3791 		*cs->statusp = resp->status = puterrno4(error);
3792 		goto out;
3793 	}
3794 
3795 	if (MANDLOCK(vp, va.va_mode)) {
3796 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3797 		goto out;
3798 	}
3799 
3800 	data = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);
3801 
3802 	if (is_referral) {
3803 		char *s;
3804 		size_t strsz;
3805 
3806 		/* Get an artificial symlink based on a referral */
3807 		s = build_symlink(vp, cs->cr, &strsz);
3808 		global_svstat_ptr[4][NFS_REFERLINKS].value.ui64++;
3809 		DTRACE_PROBE2(nfs4serv__func__referral__reflink,
3810 		    vnode_t *, vp, char *, s);
3811 		if (s == NULL)
3812 			error = EINVAL;
3813 		else {
3814 			error = 0;
3815 			(void) strlcpy(data, s, MAXPATHLEN + 1);
3816 			kmem_free(s, strsz);
3817 		}
3818 
3819 	} else {
3820 
3821 		iov.iov_base = data;
3822 		iov.iov_len = MAXPATHLEN;
3823 		uio.uio_iov = &iov;
3824 		uio.uio_iovcnt = 1;
3825 		uio.uio_segflg = UIO_SYSSPACE;
3826 		uio.uio_extflg = UIO_COPY_CACHED;
3827 		uio.uio_loffset = 0;
3828 		uio.uio_resid = MAXPATHLEN;
3829 
3830 		error = VOP_READLINK(vp, &uio, cs->cr, NULL);
3831 
3832 		if (!error)
3833 			*(data + MAXPATHLEN - uio.uio_resid) = '\0';
3834 	}
3835 
3836 	if (error) {
3837 		kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3838 		*cs->statusp = resp->status = puterrno4(error);
3839 		goto out;
3840 	}
3841 
3842 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
3843 	name = nfscmd_convname(ca, cs->exi, data, NFSCMD_CONV_OUTBOUND,
3844 	    MAXPATHLEN  + 1);
3845 
3846 	if (name == NULL) {
3847 		/*
3848 		 * Even though the conversion failed, we return
3849 		 * something. We just don't translate it.
3850 		 */
3851 		name = data;
3852 	}
3853 
3854 	/*
3855 	 * treat link name as data
3856 	 */
3857 	(void) str_to_utf8(name, (utf8string *)&resp->link);
3858 
3859 	if (name != data)
3860 		kmem_free(name, MAXPATHLEN + 1);
3861 	kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3862 	*cs->statusp = resp->status = NFS4_OK;
3863 
3864 out:
3865 	DTRACE_NFSV4_2(op__readlink__done, struct compound_state *, cs,
3866 	    READLINK4res *, resp);
3867 }
3868 
3869 static void
3870 rfs4_op_readlink_free(nfs_resop4 *resop)
3871 {
3872 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3873 	utf8string *symlink = (utf8string *)&resp->link;
3874 
3875 	if (symlink->utf8string_val) {
3876 		UTF8STRING_FREE(*symlink)
3877 	}
3878 }
3879 
3880 /*
3881  * release_lockowner:
3882  *	Release any state associated with the supplied
3883  *	lockowner. Note if any lo_state is holding locks we will not
3884  *	rele that lo_state and thus the lockowner will not be destroyed.
3885  *	A client using lock after the lock owner stateid has been released
3886  *	will suffer the consequence of NFS4ERR_BAD_STATEID and would have
3887  *	to reissue the lock with new_lock_owner set to TRUE.
3888  *	args: lock_owner
3889  *	res:  status
3890  */
3891 /* ARGSUSED */
3892 static void
3893 rfs4_op_release_lockowner(nfs_argop4 *argop, nfs_resop4 *resop,
3894     struct svc_req *req, struct compound_state *cs)
3895 {
3896 	RELEASE_LOCKOWNER4args *ap = &argop->nfs_argop4_u.oprelease_lockowner;
3897 	RELEASE_LOCKOWNER4res *resp = &resop->nfs_resop4_u.oprelease_lockowner;
3898 	rfs4_lockowner_t *lo;
3899 	rfs4_openowner_t *oo;
3900 	rfs4_state_t *sp;
3901 	rfs4_lo_state_t *lsp;
3902 	rfs4_client_t *cp;
3903 	bool_t create = FALSE;
3904 	locklist_t *llist;
3905 	sysid_t sysid;
3906 
3907 	DTRACE_NFSV4_2(op__release__lockowner__start, struct compound_state *,
3908 	    cs, RELEASE_LOCKOWNER4args *, ap);
3909 
3910 	/* Make sure there is a clientid around for this request */
3911 	cp = rfs4_findclient_by_id(ap->lock_owner.clientid, FALSE);
3912 
3913 	if (cp == NULL) {
3914 		*cs->statusp = resp->status =
3915 		    rfs4_check_clientid(&ap->lock_owner.clientid, 0);
3916 		goto out;
3917 	}
3918 	rfs4_client_rele(cp);
3919 
3920 	lo = rfs4_findlockowner(&ap->lock_owner, &create);
3921 	if (lo == NULL) {
3922 		*cs->statusp = resp->status = NFS4_OK;
3923 		goto out;
3924 	}
3925 	ASSERT(lo->rl_client != NULL);
3926 
3927 	/*
3928 	 * Check for EXPIRED client. If so will reap state with in a lease
3929 	 * period or on next set_clientid_confirm step
3930 	 */
3931 	if (rfs4_lease_expired(lo->rl_client)) {
3932 		rfs4_lockowner_rele(lo);
3933 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
3934 		goto out;
3935 	}
3936 
3937 	/*
3938 	 * If no sysid has been assigned, then no locks exist; just return.
3939 	 */
3940 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3941 	if (lo->rl_client->rc_sysidt == LM_NOSYSID) {
3942 		rfs4_lockowner_rele(lo);
3943 		rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3944 		goto out;
3945 	}
3946 
3947 	sysid = lo->rl_client->rc_sysidt;
3948 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3949 
3950 	/*
3951 	 * Mark the lockowner invalid.
3952 	 */
3953 	rfs4_dbe_hide(lo->rl_dbe);
3954 
3955 	/*
3956 	 * sysid-pid pair should now not be used since the lockowner is
3957 	 * invalid. If the client were to instantiate the lockowner again
3958 	 * it would be assigned a new pid. Thus we can get the list of
3959 	 * current locks.
3960 	 */
3961 
3962 	llist = flk_get_active_locks(sysid, lo->rl_pid);
3963 	/* If we are still holding locks fail */
3964 	if (llist != NULL) {
3965 
3966 		*cs->statusp = resp->status = NFS4ERR_LOCKS_HELD;
3967 
3968 		flk_free_locklist(llist);
3969 		/*
3970 		 * We need to unhide the lockowner so the client can
3971 		 * try it again. The bad thing here is if the client
3972 		 * has a logic error that took it here in the first place
3973 		 * he probably has lost accounting of the locks that it
3974 		 * is holding. So we may have dangling state until the
3975 		 * open owner state is reaped via close. One scenario
3976 		 * that could possibly occur is that the client has
3977 		 * sent the unlock request(s) in separate threads
3978 		 * and has not waited for the replies before sending the
3979 		 * RELEASE_LOCKOWNER request. Presumably, it would expect
3980 		 * and deal appropriately with NFS4ERR_LOCKS_HELD, by
3981 		 * reissuing the request.
3982 		 */
3983 		rfs4_dbe_unhide(lo->rl_dbe);
3984 		rfs4_lockowner_rele(lo);
3985 		goto out;
3986 	}
3987 
3988 	/*
3989 	 * For the corresponding client we need to check each open
3990 	 * owner for any opens that have lockowner state associated
3991 	 * with this lockowner.
3992 	 */
3993 
3994 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3995 	for (oo = list_head(&lo->rl_client->rc_openownerlist); oo != NULL;
3996 	    oo = list_next(&lo->rl_client->rc_openownerlist, oo)) {
3997 
3998 		rfs4_dbe_lock(oo->ro_dbe);
3999 		for (sp = list_head(&oo->ro_statelist); sp != NULL;
4000 		    sp = list_next(&oo->ro_statelist, sp)) {
4001 
4002 			rfs4_dbe_lock(sp->rs_dbe);
4003 			for (lsp = list_head(&sp->rs_lostatelist);
4004 			    lsp != NULL;
4005 			    lsp = list_next(&sp->rs_lostatelist, lsp)) {
4006 				if (lsp->rls_locker == lo) {
4007 					rfs4_dbe_lock(lsp->rls_dbe);
4008 					rfs4_dbe_invalidate(lsp->rls_dbe);
4009 					rfs4_dbe_unlock(lsp->rls_dbe);
4010 				}
4011 			}
4012 			rfs4_dbe_unlock(sp->rs_dbe);
4013 		}
4014 		rfs4_dbe_unlock(oo->ro_dbe);
4015 	}
4016 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
4017 
4018 	rfs4_lockowner_rele(lo);
4019 
4020 	*cs->statusp = resp->status = NFS4_OK;
4021 
4022 out:
4023 	DTRACE_NFSV4_2(op__release__lockowner__done, struct compound_state *,
4024 	    cs, RELEASE_LOCKOWNER4res *, resp);
4025 }
4026 
4027 /*
4028  * short utility function to lookup a file and recall the delegation
4029  */
4030 static rfs4_file_t *
4031 rfs4_lookup_and_findfile(vnode_t *dvp, char *nm, vnode_t **vpp,
4032     int *lkup_error, cred_t *cr)
4033 {
4034 	vnode_t *vp;
4035 	rfs4_file_t *fp = NULL;
4036 	bool_t fcreate = FALSE;
4037 	int error;
4038 
4039 	if (vpp)
4040 		*vpp = NULL;
4041 
4042 	if ((error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cr, NULL, NULL,
4043 	    NULL)) == 0) {
4044 		if (vp->v_type == VREG)
4045 			fp = rfs4_findfile(vp, NULL, &fcreate);
4046 		if (vpp)
4047 			*vpp = vp;
4048 		else
4049 			VN_RELE(vp);
4050 	}
4051 
4052 	if (lkup_error)
4053 		*lkup_error = error;
4054 
4055 	return (fp);
4056 }
4057 
4058 /*
4059  * remove: args: CURRENT_FH: directory; name.
4060  *	res: status. If success - CURRENT_FH unchanged, return change_info
4061  *		for directory.
4062  */
4063 /* ARGSUSED */
4064 static void
4065 rfs4_op_remove(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4066     struct compound_state *cs)
4067 {
4068 	REMOVE4args *args = &argop->nfs_argop4_u.opremove;
4069 	REMOVE4res *resp = &resop->nfs_resop4_u.opremove;
4070 	int error;
4071 	vnode_t *dvp, *vp;
4072 	struct vattr bdva, idva, adva;
4073 	char *nm;
4074 	uint_t len;
4075 	rfs4_file_t *fp;
4076 	int in_crit = 0;
4077 	bslabel_t *clabel;
4078 	struct sockaddr *ca;
4079 	char *name = NULL;
4080 	nfsstat4 status;
4081 
4082 	DTRACE_NFSV4_2(op__remove__start, struct compound_state *, cs,
4083 	    REMOVE4args *, args);
4084 
4085 	/* CURRENT_FH: directory */
4086 	dvp = cs->vp;
4087 	if (dvp == NULL) {
4088 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4089 		goto out;
4090 	}
4091 
4092 	if (cs->access == CS_ACCESS_DENIED) {
4093 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4094 		goto out;
4095 	}
4096 
4097 	/*
4098 	 * If there is an unshared filesystem mounted on this vnode,
4099 	 * Do not allow to remove anything in this directory.
4100 	 */
4101 	if (vn_ismntpt(dvp)) {
4102 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4103 		goto out;
4104 	}
4105 
4106 	if (dvp->v_type != VDIR) {
4107 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4108 		goto out;
4109 	}
4110 
4111 	status = utf8_dir_verify(&args->target);
4112 	if (status != NFS4_OK) {
4113 		*cs->statusp = resp->status = status;
4114 		goto out;
4115 	}
4116 
4117 	/*
4118 	 * Lookup the file so that we can check if it's a directory
4119 	 */
4120 	nm = utf8_to_fn(&args->target, &len, NULL);
4121 	if (nm == NULL) {
4122 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4123 		goto out;
4124 	}
4125 
4126 	if (len > MAXNAMELEN) {
4127 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4128 		kmem_free(nm, len);
4129 		goto out;
4130 	}
4131 
4132 	if (rdonly4(req, cs)) {
4133 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4134 		kmem_free(nm, len);
4135 		goto out;
4136 	}
4137 
4138 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4139 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
4140 	    MAXPATHLEN  + 1);
4141 
4142 	if (name == NULL) {
4143 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4144 		kmem_free(nm, len);
4145 		goto out;
4146 	}
4147 
4148 	/*
4149 	 * Lookup the file to determine type and while we are see if
4150 	 * there is a file struct around and check for delegation.
4151 	 * We don't need to acquire va_seq before this lookup, if
4152 	 * it causes an update, cinfo.before will not match, which will
4153 	 * trigger a cache flush even if atomic is TRUE.
4154 	 */
4155 	if (fp = rfs4_lookup_and_findfile(dvp, name, &vp, &error, cs->cr)) {
4156 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4157 		    NULL)) {
4158 			VN_RELE(vp);
4159 			rfs4_file_rele(fp);
4160 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4161 			if (nm != name)
4162 				kmem_free(name, MAXPATHLEN + 1);
4163 			kmem_free(nm, len);
4164 			goto out;
4165 		}
4166 	}
4167 
4168 	/* Didn't find anything to remove */
4169 	if (vp == NULL) {
4170 		*cs->statusp = resp->status = error;
4171 		if (nm != name)
4172 			kmem_free(name, MAXPATHLEN + 1);
4173 		kmem_free(nm, len);
4174 		goto out;
4175 	}
4176 
4177 	if (nbl_need_check(vp)) {
4178 		nbl_start_crit(vp, RW_READER);
4179 		in_crit = 1;
4180 		if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
4181 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4182 			if (nm != name)
4183 				kmem_free(name, MAXPATHLEN + 1);
4184 			kmem_free(nm, len);
4185 			nbl_end_crit(vp);
4186 			VN_RELE(vp);
4187 			if (fp) {
4188 				rfs4_clear_dont_grant(fp);
4189 				rfs4_file_rele(fp);
4190 			}
4191 			goto out;
4192 		}
4193 	}
4194 
4195 	/* check label before allowing removal */
4196 	if (is_system_labeled()) {
4197 		ASSERT(req->rq_label != NULL);
4198 		clabel = req->rq_label;
4199 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
4200 		    "got client label from request(1)",
4201 		    struct svc_req *, req);
4202 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4203 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
4204 			    cs->exi)) {
4205 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4206 				if (name != nm)
4207 					kmem_free(name, MAXPATHLEN + 1);
4208 				kmem_free(nm, len);
4209 				if (in_crit)
4210 					nbl_end_crit(vp);
4211 				VN_RELE(vp);
4212 				if (fp) {
4213 					rfs4_clear_dont_grant(fp);
4214 					rfs4_file_rele(fp);
4215 				}
4216 				goto out;
4217 			}
4218 		}
4219 	}
4220 
4221 	/* Get dir "before" change value */
4222 	bdva.va_mask = AT_CTIME|AT_SEQ;
4223 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
4224 	if (error) {
4225 		*cs->statusp = resp->status = puterrno4(error);
4226 		if (nm != name)
4227 			kmem_free(name, MAXPATHLEN + 1);
4228 		kmem_free(nm, len);
4229 		if (in_crit)
4230 			nbl_end_crit(vp);
4231 		VN_RELE(vp);
4232 		if (fp) {
4233 			rfs4_clear_dont_grant(fp);
4234 			rfs4_file_rele(fp);
4235 		}
4236 		goto out;
4237 	}
4238 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
4239 
4240 	/* Actually do the REMOVE operation */
4241 	if (vp->v_type == VDIR) {
4242 		/*
4243 		 * Can't remove a directory that has a mounted-on filesystem.
4244 		 */
4245 		if (vn_ismntpt(vp)) {
4246 			error = EACCES;
4247 		} else {
4248 			/*
4249 			 * System V defines rmdir to return EEXIST,
4250 			 * not ENOTEMPTY, if the directory is not
4251 			 * empty.  A System V NFS server needs to map
4252 			 * NFS4ERR_EXIST to NFS4ERR_NOTEMPTY to
4253 			 * transmit over the wire.
4254 			 */
4255 			if ((error = VOP_RMDIR(dvp, name, rootdir, cs->cr,
4256 			    NULL, 0)) == EEXIST)
4257 				error = ENOTEMPTY;
4258 		}
4259 	} else {
4260 		if ((error = VOP_REMOVE(dvp, name, cs->cr, NULL, 0)) == 0 &&
4261 		    fp != NULL) {
4262 			struct vattr va;
4263 			vnode_t *tvp;
4264 
4265 			rfs4_dbe_lock(fp->rf_dbe);
4266 			tvp = fp->rf_vp;
4267 			if (tvp)
4268 				VN_HOLD(tvp);
4269 			rfs4_dbe_unlock(fp->rf_dbe);
4270 
4271 			if (tvp) {
4272 				/*
4273 				 * This is va_seq safe because we are not
4274 				 * manipulating dvp.
4275 				 */
4276 				va.va_mask = AT_NLINK;
4277 				if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4278 				    va.va_nlink == 0) {
4279 					/* Remove state on file remove */
4280 					if (in_crit) {
4281 						nbl_end_crit(vp);
4282 						in_crit = 0;
4283 					}
4284 					rfs4_close_all_state(fp);
4285 				}
4286 				VN_RELE(tvp);
4287 			}
4288 		}
4289 	}
4290 
4291 	if (in_crit)
4292 		nbl_end_crit(vp);
4293 	VN_RELE(vp);
4294 
4295 	if (fp) {
4296 		rfs4_clear_dont_grant(fp);
4297 		rfs4_file_rele(fp);
4298 	}
4299 	if (nm != name)
4300 		kmem_free(name, MAXPATHLEN + 1);
4301 	kmem_free(nm, len);
4302 
4303 	if (error) {
4304 		*cs->statusp = resp->status = puterrno4(error);
4305 		goto out;
4306 	}
4307 
4308 	/*
4309 	 * Get the initial "after" sequence number, if it fails, set to zero
4310 	 */
4311 	idva.va_mask = AT_SEQ;
4312 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
4313 		idva.va_seq = 0;
4314 
4315 	/*
4316 	 * Force modified data and metadata out to stable storage.
4317 	 */
4318 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
4319 
4320 	/*
4321 	 * Get "after" change value, if it fails, simply return the
4322 	 * before value.
4323 	 */
4324 	adva.va_mask = AT_CTIME|AT_SEQ;
4325 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
4326 		adva.va_ctime = bdva.va_ctime;
4327 		adva.va_seq = 0;
4328 	}
4329 
4330 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
4331 
4332 	/*
4333 	 * The cinfo.atomic = TRUE only if we have
4334 	 * non-zero va_seq's, and it has incremented by exactly one
4335 	 * during the VOP_REMOVE/RMDIR and it didn't change during
4336 	 * the VOP_FSYNC.
4337 	 */
4338 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
4339 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
4340 		resp->cinfo.atomic = TRUE;
4341 	else
4342 		resp->cinfo.atomic = FALSE;
4343 
4344 	*cs->statusp = resp->status = NFS4_OK;
4345 
4346 out:
4347 	DTRACE_NFSV4_2(op__remove__done, struct compound_state *, cs,
4348 	    REMOVE4res *, resp);
4349 }
4350 
4351 /*
4352  * rename: args: SAVED_FH: from directory, CURRENT_FH: target directory,
4353  *		oldname and newname.
4354  *	res: status. If success - CURRENT_FH unchanged, return change_info
4355  *		for both from and target directories.
4356  */
4357 /* ARGSUSED */
4358 static void
4359 rfs4_op_rename(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4360     struct compound_state *cs)
4361 {
4362 	RENAME4args *args = &argop->nfs_argop4_u.oprename;
4363 	RENAME4res *resp = &resop->nfs_resop4_u.oprename;
4364 	int error;
4365 	vnode_t *odvp;
4366 	vnode_t *ndvp;
4367 	vnode_t *srcvp, *targvp;
4368 	struct vattr obdva, oidva, oadva;
4369 	struct vattr nbdva, nidva, nadva;
4370 	char *onm, *nnm;
4371 	uint_t olen, nlen;
4372 	rfs4_file_t *fp, *sfp;
4373 	int in_crit_src, in_crit_targ;
4374 	int fp_rele_grant_hold, sfp_rele_grant_hold;
4375 	bslabel_t *clabel;
4376 	struct sockaddr *ca;
4377 	char *converted_onm = NULL;
4378 	char *converted_nnm = NULL;
4379 	nfsstat4 status;
4380 
4381 	DTRACE_NFSV4_2(op__rename__start, struct compound_state *, cs,
4382 	    RENAME4args *, args);
4383 
4384 	fp = sfp = NULL;
4385 	srcvp = targvp = NULL;
4386 	in_crit_src = in_crit_targ = 0;
4387 	fp_rele_grant_hold = sfp_rele_grant_hold = 0;
4388 
4389 	/* CURRENT_FH: target directory */
4390 	ndvp = cs->vp;
4391 	if (ndvp == NULL) {
4392 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4393 		goto out;
4394 	}
4395 
4396 	/* SAVED_FH: from directory */
4397 	odvp = cs->saved_vp;
4398 	if (odvp == NULL) {
4399 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4400 		goto out;
4401 	}
4402 
4403 	if (cs->access == CS_ACCESS_DENIED) {
4404 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4405 		goto out;
4406 	}
4407 
4408 	/*
4409 	 * If there is an unshared filesystem mounted on this vnode,
4410 	 * do not allow to rename objects in this directory.
4411 	 */
4412 	if (vn_ismntpt(odvp)) {
4413 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4414 		goto out;
4415 	}
4416 
4417 	/*
4418 	 * If there is an unshared filesystem mounted on this vnode,
4419 	 * do not allow to rename to this directory.
4420 	 */
4421 	if (vn_ismntpt(ndvp)) {
4422 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4423 		goto out;
4424 	}
4425 
4426 	if (odvp->v_type != VDIR || ndvp->v_type != VDIR) {
4427 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4428 		goto out;
4429 	}
4430 
4431 	if (cs->saved_exi != cs->exi) {
4432 		*cs->statusp = resp->status = NFS4ERR_XDEV;
4433 		goto out;
4434 	}
4435 
4436 	status = utf8_dir_verify(&args->oldname);
4437 	if (status != NFS4_OK) {
4438 		*cs->statusp = resp->status = status;
4439 		goto out;
4440 	}
4441 
4442 	status = utf8_dir_verify(&args->newname);
4443 	if (status != NFS4_OK) {
4444 		*cs->statusp = resp->status = status;
4445 		goto out;
4446 	}
4447 
4448 	onm = utf8_to_fn(&args->oldname, &olen, NULL);
4449 	if (onm == NULL) {
4450 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4451 		goto out;
4452 	}
4453 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4454 	nlen = MAXPATHLEN + 1;
4455 	converted_onm = nfscmd_convname(ca, cs->exi, onm, NFSCMD_CONV_INBOUND,
4456 	    nlen);
4457 
4458 	if (converted_onm == NULL) {
4459 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4460 		kmem_free(onm, olen);
4461 		goto out;
4462 	}
4463 
4464 	nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4465 	if (nnm == NULL) {
4466 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4467 		if (onm != converted_onm)
4468 			kmem_free(converted_onm, MAXPATHLEN + 1);
4469 		kmem_free(onm, olen);
4470 		goto out;
4471 	}
4472 	converted_nnm = nfscmd_convname(ca, cs->exi, nnm, NFSCMD_CONV_INBOUND,
4473 	    MAXPATHLEN  + 1);
4474 
4475 	if (converted_nnm == NULL) {
4476 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4477 		kmem_free(nnm, nlen);
4478 		nnm = NULL;
4479 		if (onm != converted_onm)
4480 			kmem_free(converted_onm, MAXPATHLEN + 1);
4481 		kmem_free(onm, olen);
4482 		goto out;
4483 	}
4484 
4485 
4486 	if (olen > MAXNAMELEN || nlen > MAXNAMELEN) {
4487 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4488 		kmem_free(onm, olen);
4489 		kmem_free(nnm, nlen);
4490 		goto out;
4491 	}
4492 
4493 
4494 	if (rdonly4(req, cs)) {
4495 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4496 		if (onm != converted_onm)
4497 			kmem_free(converted_onm, MAXPATHLEN + 1);
4498 		kmem_free(onm, olen);
4499 		if (nnm != converted_nnm)
4500 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4501 		kmem_free(nnm, nlen);
4502 		goto out;
4503 	}
4504 
4505 	/* check label of the target dir */
4506 	if (is_system_labeled()) {
4507 		ASSERT(req->rq_label != NULL);
4508 		clabel = req->rq_label;
4509 		DTRACE_PROBE2(tx__rfs4__log__info__oprename__clabel, char *,
4510 		    "got client label from request(1)",
4511 		    struct svc_req *, req);
4512 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4513 			if (!do_rfs_label_check(clabel, ndvp,
4514 			    EQUALITY_CHECK, cs->exi)) {
4515 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4516 				goto err_out;
4517 			}
4518 		}
4519 	}
4520 
4521 	/*
4522 	 * Is the source a file and have a delegation?
4523 	 * We don't need to acquire va_seq before these lookups, if
4524 	 * it causes an update, cinfo.before will not match, which will
4525 	 * trigger a cache flush even if atomic is TRUE.
4526 	 */
4527 	if (sfp = rfs4_lookup_and_findfile(odvp, converted_onm, &srcvp,
4528 	    &error, cs->cr)) {
4529 		if (rfs4_check_delegated_byfp(FWRITE, sfp, TRUE, TRUE, TRUE,
4530 		    NULL)) {
4531 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4532 			goto err_out;
4533 		}
4534 	}
4535 
4536 	if (srcvp == NULL) {
4537 		*cs->statusp = resp->status = puterrno4(error);
4538 		if (onm != converted_onm)
4539 			kmem_free(converted_onm, MAXPATHLEN + 1);
4540 		kmem_free(onm, olen);
4541 		if (nnm != converted_nnm)
4542 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4543 		kmem_free(nnm, nlen);
4544 		goto out;
4545 	}
4546 
4547 	sfp_rele_grant_hold = 1;
4548 
4549 	/* Does the destination exist and a file and have a delegation? */
4550 	if (fp = rfs4_lookup_and_findfile(ndvp, converted_nnm, &targvp,
4551 	    NULL, cs->cr)) {
4552 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4553 		    NULL)) {
4554 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4555 			goto err_out;
4556 		}
4557 	}
4558 	fp_rele_grant_hold = 1;
4559 
4560 
4561 	/* Check for NBMAND lock on both source and target */
4562 	if (nbl_need_check(srcvp)) {
4563 		nbl_start_crit(srcvp, RW_READER);
4564 		in_crit_src = 1;
4565 		if (nbl_conflict(srcvp, NBL_RENAME, 0, 0, 0, NULL)) {
4566 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4567 			goto err_out;
4568 		}
4569 	}
4570 
4571 	if (targvp && nbl_need_check(targvp)) {
4572 		nbl_start_crit(targvp, RW_READER);
4573 		in_crit_targ = 1;
4574 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
4575 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4576 			goto err_out;
4577 		}
4578 	}
4579 
4580 	/* Get source "before" change value */
4581 	obdva.va_mask = AT_CTIME|AT_SEQ;
4582 	error = VOP_GETATTR(odvp, &obdva, 0, cs->cr, NULL);
4583 	if (!error) {
4584 		nbdva.va_mask = AT_CTIME|AT_SEQ;
4585 		error = VOP_GETATTR(ndvp, &nbdva, 0, cs->cr, NULL);
4586 	}
4587 	if (error) {
4588 		*cs->statusp = resp->status = puterrno4(error);
4589 		goto err_out;
4590 	}
4591 
4592 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.before, obdva.va_ctime)
4593 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.before, nbdva.va_ctime)
4594 
4595 	if ((error = VOP_RENAME(odvp, converted_onm, ndvp, converted_nnm,
4596 	    cs->cr, NULL, 0)) == 0 && fp != NULL) {
4597 		struct vattr va;
4598 		vnode_t *tvp;
4599 
4600 		rfs4_dbe_lock(fp->rf_dbe);
4601 		tvp = fp->rf_vp;
4602 		if (tvp)
4603 			VN_HOLD(tvp);
4604 		rfs4_dbe_unlock(fp->rf_dbe);
4605 
4606 		if (tvp) {
4607 			va.va_mask = AT_NLINK;
4608 			if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4609 			    va.va_nlink == 0) {
4610 				/* The file is gone and so should the state */
4611 				if (in_crit_targ) {
4612 					nbl_end_crit(targvp);
4613 					in_crit_targ = 0;
4614 				}
4615 				rfs4_close_all_state(fp);
4616 			}
4617 			VN_RELE(tvp);
4618 		}
4619 	}
4620 	if (error == 0)
4621 		vn_renamepath(ndvp, srcvp, nnm, nlen - 1);
4622 
4623 	if (in_crit_src)
4624 		nbl_end_crit(srcvp);
4625 	if (srcvp)
4626 		VN_RELE(srcvp);
4627 	if (in_crit_targ)
4628 		nbl_end_crit(targvp);
4629 	if (targvp)
4630 		VN_RELE(targvp);
4631 
4632 	if (sfp) {
4633 		rfs4_clear_dont_grant(sfp);
4634 		rfs4_file_rele(sfp);
4635 	}
4636 	if (fp) {
4637 		rfs4_clear_dont_grant(fp);
4638 		rfs4_file_rele(fp);
4639 	}
4640 
4641 	if (converted_onm != onm)
4642 		kmem_free(converted_onm, MAXPATHLEN + 1);
4643 	kmem_free(onm, olen);
4644 	if (converted_nnm != nnm)
4645 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4646 	kmem_free(nnm, nlen);
4647 
4648 	/*
4649 	 * Get the initial "after" sequence number, if it fails, set to zero
4650 	 */
4651 	oidva.va_mask = AT_SEQ;
4652 	if (VOP_GETATTR(odvp, &oidva, 0, cs->cr, NULL))
4653 		oidva.va_seq = 0;
4654 
4655 	nidva.va_mask = AT_SEQ;
4656 	if (VOP_GETATTR(ndvp, &nidva, 0, cs->cr, NULL))
4657 		nidva.va_seq = 0;
4658 
4659 	/*
4660 	 * Force modified data and metadata out to stable storage.
4661 	 */
4662 	(void) VOP_FSYNC(odvp, 0, cs->cr, NULL);
4663 	(void) VOP_FSYNC(ndvp, 0, cs->cr, NULL);
4664 
4665 	if (error) {
4666 		*cs->statusp = resp->status = puterrno4(error);
4667 		goto out;
4668 	}
4669 
4670 	/*
4671 	 * Get "after" change values, if it fails, simply return the
4672 	 * before value.
4673 	 */
4674 	oadva.va_mask = AT_CTIME|AT_SEQ;
4675 	if (VOP_GETATTR(odvp, &oadva, 0, cs->cr, NULL)) {
4676 		oadva.va_ctime = obdva.va_ctime;
4677 		oadva.va_seq = 0;
4678 	}
4679 
4680 	nadva.va_mask = AT_CTIME|AT_SEQ;
4681 	if (VOP_GETATTR(odvp, &nadva, 0, cs->cr, NULL)) {
4682 		nadva.va_ctime = nbdva.va_ctime;
4683 		nadva.va_seq = 0;
4684 	}
4685 
4686 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.after, oadva.va_ctime)
4687 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.after, nadva.va_ctime)
4688 
4689 	/*
4690 	 * The cinfo.atomic = TRUE only if we have
4691 	 * non-zero va_seq's, and it has incremented by exactly one
4692 	 * during the VOP_RENAME and it didn't change during the VOP_FSYNC.
4693 	 */
4694 	if (obdva.va_seq && oidva.va_seq && oadva.va_seq &&
4695 	    oidva.va_seq == (obdva.va_seq + 1) && oidva.va_seq == oadva.va_seq)
4696 		resp->source_cinfo.atomic = TRUE;
4697 	else
4698 		resp->source_cinfo.atomic = FALSE;
4699 
4700 	if (nbdva.va_seq && nidva.va_seq && nadva.va_seq &&
4701 	    nidva.va_seq == (nbdva.va_seq + 1) && nidva.va_seq == nadva.va_seq)
4702 		resp->target_cinfo.atomic = TRUE;
4703 	else
4704 		resp->target_cinfo.atomic = FALSE;
4705 
4706 #ifdef	VOLATILE_FH_TEST
4707 	{
4708 	extern void add_volrnm_fh(struct exportinfo *, vnode_t *);
4709 
4710 	/*
4711 	 * Add the renamed file handle to the volatile rename list
4712 	 */
4713 	if (cs->exi->exi_export.ex_flags & EX_VOLRNM) {
4714 		/* file handles may expire on rename */
4715 		vnode_t *vp;
4716 
4717 		nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4718 		/*
4719 		 * Already know that nnm will be a valid string
4720 		 */
4721 		error = VOP_LOOKUP(ndvp, nnm, &vp, NULL, 0, NULL, cs->cr,
4722 		    NULL, NULL, NULL);
4723 		kmem_free(nnm, nlen);
4724 		if (!error) {
4725 			add_volrnm_fh(cs->exi, vp);
4726 			VN_RELE(vp);
4727 		}
4728 	}
4729 	}
4730 #endif	/* VOLATILE_FH_TEST */
4731 
4732 	*cs->statusp = resp->status = NFS4_OK;
4733 out:
4734 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4735 	    RENAME4res *, resp);
4736 	return;
4737 
4738 err_out:
4739 	if (onm != converted_onm)
4740 		kmem_free(converted_onm, MAXPATHLEN + 1);
4741 	if (onm != NULL)
4742 		kmem_free(onm, olen);
4743 	if (nnm != converted_nnm)
4744 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4745 	if (nnm != NULL)
4746 		kmem_free(nnm, nlen);
4747 
4748 	if (in_crit_src) nbl_end_crit(srcvp);
4749 	if (in_crit_targ) nbl_end_crit(targvp);
4750 	if (targvp) VN_RELE(targvp);
4751 	if (srcvp) VN_RELE(srcvp);
4752 	if (sfp) {
4753 		if (sfp_rele_grant_hold) rfs4_clear_dont_grant(sfp);
4754 		rfs4_file_rele(sfp);
4755 	}
4756 	if (fp) {
4757 		if (fp_rele_grant_hold) rfs4_clear_dont_grant(fp);
4758 		rfs4_file_rele(fp);
4759 	}
4760 
4761 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4762 	    RENAME4res *, resp);
4763 }
4764 
4765 /* ARGSUSED */
4766 static void
4767 rfs4_op_renew(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4768     struct compound_state *cs)
4769 {
4770 	RENEW4args *args = &argop->nfs_argop4_u.oprenew;
4771 	RENEW4res *resp = &resop->nfs_resop4_u.oprenew;
4772 	rfs4_client_t *cp;
4773 
4774 	DTRACE_NFSV4_2(op__renew__start, struct compound_state *, cs,
4775 	    RENEW4args *, args);
4776 
4777 	if ((cp = rfs4_findclient_by_id(args->clientid, FALSE)) == NULL) {
4778 		*cs->statusp = resp->status =
4779 		    rfs4_check_clientid(&args->clientid, 0);
4780 		goto out;
4781 	}
4782 
4783 	if (rfs4_lease_expired(cp)) {
4784 		rfs4_client_rele(cp);
4785 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
4786 		goto out;
4787 	}
4788 
4789 	rfs4_update_lease(cp);
4790 
4791 	mutex_enter(cp->rc_cbinfo.cb_lock);
4792 	if (cp->rc_cbinfo.cb_notified_of_cb_path_down == FALSE) {
4793 		cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
4794 		*cs->statusp = resp->status = NFS4ERR_CB_PATH_DOWN;
4795 	} else {
4796 		*cs->statusp = resp->status = NFS4_OK;
4797 	}
4798 	mutex_exit(cp->rc_cbinfo.cb_lock);
4799 
4800 	rfs4_client_rele(cp);
4801 
4802 out:
4803 	DTRACE_NFSV4_2(op__renew__done, struct compound_state *, cs,
4804 	    RENEW4res *, resp);
4805 }
4806 
4807 /* ARGSUSED */
4808 static void
4809 rfs4_op_restorefh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
4810     struct compound_state *cs)
4811 {
4812 	RESTOREFH4res *resp = &resop->nfs_resop4_u.oprestorefh;
4813 
4814 	DTRACE_NFSV4_1(op__restorefh__start, struct compound_state *, cs);
4815 
4816 	/* No need to check cs->access - we are not accessing any object */
4817 	if ((cs->saved_vp == NULL) || (cs->saved_fh.nfs_fh4_val == NULL)) {
4818 		*cs->statusp = resp->status = NFS4ERR_RESTOREFH;
4819 		goto out;
4820 	}
4821 	if (cs->vp != NULL) {
4822 		VN_RELE(cs->vp);
4823 	}
4824 	cs->vp = cs->saved_vp;
4825 	cs->saved_vp = NULL;
4826 	cs->exi = cs->saved_exi;
4827 	nfs_fh4_copy(&cs->saved_fh, &cs->fh);
4828 	*cs->statusp = resp->status = NFS4_OK;
4829 	cs->deleg = FALSE;
4830 
4831 out:
4832 	DTRACE_NFSV4_2(op__restorefh__done, struct compound_state *, cs,
4833 	    RESTOREFH4res *, resp);
4834 }
4835 
4836 /* ARGSUSED */
4837 static void
4838 rfs4_op_savefh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4839     struct compound_state *cs)
4840 {
4841 	SAVEFH4res *resp = &resop->nfs_resop4_u.opsavefh;
4842 
4843 	DTRACE_NFSV4_1(op__savefh__start, struct compound_state *, cs);
4844 
4845 	/* No need to check cs->access - we are not accessing any object */
4846 	if (cs->vp == NULL) {
4847 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4848 		goto out;
4849 	}
4850 	if (cs->saved_vp != NULL) {
4851 		VN_RELE(cs->saved_vp);
4852 	}
4853 	cs->saved_vp = cs->vp;
4854 	VN_HOLD(cs->saved_vp);
4855 	cs->saved_exi = cs->exi;
4856 	/*
4857 	 * since SAVEFH is fairly rare, don't alloc space for its fh
4858 	 * unless necessary.
4859 	 */
4860 	if (cs->saved_fh.nfs_fh4_val == NULL) {
4861 		cs->saved_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP);
4862 	}
4863 	nfs_fh4_copy(&cs->fh, &cs->saved_fh);
4864 	*cs->statusp = resp->status = NFS4_OK;
4865 
4866 out:
4867 	DTRACE_NFSV4_2(op__savefh__done, struct compound_state *, cs,
4868 	    SAVEFH4res *, resp);
4869 }
4870 
4871 /*
4872  * rfs4_verify_attr is called when nfsv4 Setattr failed, but we wish to
4873  * return the bitmap of attrs that were set successfully. It is also
4874  * called by Verify/Nverify to test the vattr/vfsstat attrs. It should
4875  * always be called only after rfs4_do_set_attrs().
4876  *
4877  * Verify that the attributes are same as the expected ones. sargp->vap
4878  * and sargp->sbp contain the input attributes as translated from fattr4.
4879  *
4880  * This function verifies only the attrs that correspond to a vattr or
4881  * vfsstat struct. That is because of the extra step needed to get the
4882  * corresponding system structs. Other attributes have already been set or
4883  * verified by do_rfs4_set_attrs.
4884  *
4885  * Return 0 if all attrs match, -1 if some don't, error if error processing.
4886  */
4887 static int
4888 rfs4_verify_attr(struct nfs4_svgetit_arg *sargp,
4889     bitmap4 *resp, struct nfs4_ntov_table *ntovp)
4890 {
4891 	int error, ret_error = 0;
4892 	int i, k;
4893 	uint_t sva_mask = sargp->vap->va_mask;
4894 	uint_t vbit;
4895 	union nfs4_attr_u *na;
4896 	uint8_t *amap;
4897 	bool_t getsb = ntovp->vfsstat;
4898 
4899 	if (sva_mask != 0) {
4900 		/*
4901 		 * Okay to overwrite sargp->vap because we verify based
4902 		 * on the incoming values.
4903 		 */
4904 		ret_error = VOP_GETATTR(sargp->cs->vp, sargp->vap, 0,
4905 		    sargp->cs->cr, NULL);
4906 		if (ret_error) {
4907 			if (resp == NULL)
4908 				return (ret_error);
4909 			/*
4910 			 * Must return bitmap of successful attrs
4911 			 */
4912 			sva_mask = 0;	/* to prevent checking vap later */
4913 		} else {
4914 			/*
4915 			 * Some file systems clobber va_mask. it is probably
4916 			 * wrong of them to do so, nonethless we practice
4917 			 * defensive coding.
4918 			 * See bug id 4276830.
4919 			 */
4920 			sargp->vap->va_mask = sva_mask;
4921 		}
4922 	}
4923 
4924 	if (getsb) {
4925 		/*
4926 		 * Now get the superblock and loop on the bitmap, as there is
4927 		 * no simple way of translating from superblock to bitmap4.
4928 		 */
4929 		ret_error = VFS_STATVFS(sargp->cs->vp->v_vfsp, sargp->sbp);
4930 		if (ret_error) {
4931 			if (resp == NULL)
4932 				goto errout;
4933 			getsb = FALSE;
4934 		}
4935 	}
4936 
4937 	/*
4938 	 * Now loop and verify each attribute which getattr returned
4939 	 * whether it's the same as the input.
4940 	 */
4941 	if (resp == NULL && !getsb && (sva_mask == 0))
4942 		goto errout;
4943 
4944 	na = ntovp->na;
4945 	amap = ntovp->amap;
4946 	k = 0;
4947 	for (i = 0; i < ntovp->attrcnt; i++, na++, amap++) {
4948 		k = *amap;
4949 		ASSERT(nfs4_ntov_map[k].nval == k);
4950 		vbit = nfs4_ntov_map[k].vbit;
4951 
4952 		/*
4953 		 * If vattr attribute but VOP_GETATTR failed, or it's
4954 		 * superblock attribute but VFS_STATVFS failed, skip
4955 		 */
4956 		if (vbit) {
4957 			if ((vbit & sva_mask) == 0)
4958 				continue;
4959 		} else if (!(getsb && nfs4_ntov_map[k].vfsstat)) {
4960 			continue;
4961 		}
4962 		error = (*nfs4_ntov_map[k].sv_getit)(NFS4ATTR_VERIT, sargp, na);
4963 		if (resp != NULL) {
4964 			if (error)
4965 				ret_error = -1;	/* not all match */
4966 			else	/* update response bitmap */
4967 				*resp |= nfs4_ntov_map[k].fbit;
4968 			continue;
4969 		}
4970 		if (error) {
4971 			ret_error = -1;	/* not all match */
4972 			break;
4973 		}
4974 	}
4975 errout:
4976 	return (ret_error);
4977 }
4978 
4979 /*
4980  * Decode the attribute to be set/verified. If the attr requires a sys op
4981  * (VOP_GETATTR, VFS_VFSSTAT), and the request is to verify, then don't
4982  * call the sv_getit function for it, because the sys op hasn't yet been done.
4983  * Return 0 for success, error code if failed.
4984  *
4985  * Note: the decoded arg is not freed here but in nfs4_ntov_table_free.
4986  */
4987 static int
4988 decode_fattr4_attr(nfs4_attr_cmd_t cmd, struct nfs4_svgetit_arg *sargp,
4989     int k, XDR *xdrp, bitmap4 *resp_bval, union nfs4_attr_u *nap)
4990 {
4991 	int error = 0;
4992 	bool_t set_later;
4993 
4994 	sargp->vap->va_mask |= nfs4_ntov_map[k].vbit;
4995 
4996 	if ((*nfs4_ntov_map[k].xfunc)(xdrp, nap)) {
4997 		set_later = nfs4_ntov_map[k].vbit || nfs4_ntov_map[k].vfsstat;
4998 		/*
4999 		 * don't verify yet if a vattr or sb dependent attr,
5000 		 * because we don't have their sys values yet.
5001 		 * Will be done later.
5002 		 */
5003 		if (! (set_later && (cmd == NFS4ATTR_VERIT))) {
5004 			/*
5005 			 * ACLs are a special case, since setting the MODE
5006 			 * conflicts with setting the ACL.  We delay setting
5007 			 * the ACL until all other attributes have been set.
5008 			 * The ACL gets set in do_rfs4_op_setattr().
5009 			 */
5010 			if (nfs4_ntov_map[k].fbit != FATTR4_ACL_MASK) {
5011 				error = (*nfs4_ntov_map[k].sv_getit)(cmd,
5012 				    sargp, nap);
5013 				if (error) {
5014 					xdr_free(nfs4_ntov_map[k].xfunc,
5015 					    (caddr_t)nap);
5016 				}
5017 			}
5018 		}
5019 	} else {
5020 #ifdef  DEBUG
5021 		cmn_err(CE_NOTE, "decode_fattr4_attr: error "
5022 		    "decoding attribute %d\n", k);
5023 #endif
5024 		error = EINVAL;
5025 	}
5026 	if (!error && resp_bval && !set_later) {
5027 		*resp_bval |= nfs4_ntov_map[k].fbit;
5028 	}
5029 
5030 	return (error);
5031 }
5032 
5033 /*
5034  * Set vattr based on incoming fattr4 attrs - used by setattr.
5035  * Set response mask. Ignore any values that are not writable vattr attrs.
5036  */
5037 static nfsstat4
5038 do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5039     struct nfs4_svgetit_arg *sargp, struct nfs4_ntov_table *ntovp,
5040     nfs4_attr_cmd_t cmd)
5041 {
5042 	int error = 0;
5043 	int i;
5044 	char *attrs = fattrp->attrlist4;
5045 	uint32_t attrslen = fattrp->attrlist4_len;
5046 	XDR xdr;
5047 	nfsstat4 status = NFS4_OK;
5048 	vnode_t *vp = cs->vp;
5049 	union nfs4_attr_u *na;
5050 	uint8_t *amap;
5051 
5052 #ifndef lint
5053 	/*
5054 	 * Make sure that maximum attribute number can be expressed as an
5055 	 * 8 bit quantity.
5056 	 */
5057 	ASSERT(NFS4_MAXNUM_ATTRS <= (UINT8_MAX + 1));
5058 #endif
5059 
5060 	if (vp == NULL) {
5061 		if (resp)
5062 			*resp = 0;
5063 		return (NFS4ERR_NOFILEHANDLE);
5064 	}
5065 	if (cs->access == CS_ACCESS_DENIED) {
5066 		if (resp)
5067 			*resp = 0;
5068 		return (NFS4ERR_ACCESS);
5069 	}
5070 
5071 	sargp->op = cmd;
5072 	sargp->cs = cs;
5073 	sargp->flag = 0;	/* may be set later */
5074 	sargp->vap->va_mask = 0;
5075 	sargp->rdattr_error = NFS4_OK;
5076 	sargp->rdattr_error_req = FALSE;
5077 	/* sargp->sbp is set by the caller */
5078 
5079 	xdrmem_create(&xdr, attrs, attrslen, XDR_DECODE);
5080 
5081 	na = ntovp->na;
5082 	amap = ntovp->amap;
5083 
5084 	/*
5085 	 * The following loop iterates on the nfs4_ntov_map checking
5086 	 * if the fbit is set in the requested bitmap.
5087 	 * If set then we process the arguments using the
5088 	 * rfs4_fattr4 conversion functions to populate the setattr
5089 	 * vattr and va_mask. Any settable attrs that are not using vattr
5090 	 * will be set in this loop.
5091 	 */
5092 	for (i = 0; i < nfs4_ntov_map_size; i++) {
5093 		if (!(fattrp->attrmask & nfs4_ntov_map[i].fbit)) {
5094 			continue;
5095 		}
5096 		/*
5097 		 * If setattr, must be a writable attr.
5098 		 * If verify/nverify, must be a readable attr.
5099 		 */
5100 		if ((error = (*nfs4_ntov_map[i].sv_getit)(
5101 		    NFS4ATTR_SUPPORTED, sargp, NULL)) != 0) {
5102 			/*
5103 			 * Client tries to set/verify an
5104 			 * unsupported attribute, tries to set
5105 			 * a read only attr or verify a write
5106 			 * only one - error!
5107 			 */
5108 			break;
5109 		}
5110 		/*
5111 		 * Decode the attribute to set/verify
5112 		 */
5113 		error = decode_fattr4_attr(cmd, sargp, nfs4_ntov_map[i].nval,
5114 		    &xdr, resp ? resp : NULL, na);
5115 		if (error)
5116 			break;
5117 		*amap++ = (uint8_t)nfs4_ntov_map[i].nval;
5118 		na++;
5119 		(ntovp->attrcnt)++;
5120 		if (nfs4_ntov_map[i].vfsstat)
5121 			ntovp->vfsstat = TRUE;
5122 	}
5123 
5124 	if (error != 0)
5125 		status = (error == ENOTSUP ? NFS4ERR_ATTRNOTSUPP :
5126 		    puterrno4(error));
5127 	/* xdrmem_destroy(&xdrs); */	/* NO-OP */
5128 	return (status);
5129 }
5130 
5131 static nfsstat4
5132 do_rfs4_op_setattr(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5133     stateid4 *stateid)
5134 {
5135 	int error = 0;
5136 	struct nfs4_svgetit_arg sarg;
5137 	bool_t trunc;
5138 
5139 	nfsstat4 status = NFS4_OK;
5140 	cred_t *cr = cs->cr;
5141 	vnode_t *vp = cs->vp;
5142 	struct nfs4_ntov_table ntov;
5143 	struct statvfs64 sb;
5144 	struct vattr bva;
5145 	struct flock64 bf;
5146 	int in_crit = 0;
5147 	uint_t saved_mask = 0;
5148 	caller_context_t ct;
5149 
5150 	*resp = 0;
5151 	sarg.sbp = &sb;
5152 	sarg.is_referral = B_FALSE;
5153 	nfs4_ntov_table_init(&ntov);
5154 	status = do_rfs4_set_attrs(resp, fattrp, cs, &sarg, &ntov,
5155 	    NFS4ATTR_SETIT);
5156 	if (status != NFS4_OK) {
5157 		/*
5158 		 * failed set attrs
5159 		 */
5160 		goto done;
5161 	}
5162 	if ((sarg.vap->va_mask == 0) &&
5163 	    (! (fattrp->attrmask & FATTR4_ACL_MASK))) {
5164 		/*
5165 		 * no further work to be done
5166 		 */
5167 		goto done;
5168 	}
5169 
5170 	/*
5171 	 * If we got a request to set the ACL and the MODE, only
5172 	 * allow changing VSUID, VSGID, and VSVTX.  Attempting
5173 	 * to change any other bits, along with setting an ACL,
5174 	 * gives NFS4ERR_INVAL.
5175 	 */
5176 	if ((fattrp->attrmask & FATTR4_ACL_MASK) &&
5177 	    (fattrp->attrmask & FATTR4_MODE_MASK)) {
5178 		vattr_t va;
5179 
5180 		va.va_mask = AT_MODE;
5181 		error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
5182 		if (error) {
5183 			status = puterrno4(error);
5184 			goto done;
5185 		}
5186 		if ((sarg.vap->va_mode ^ va.va_mode) &
5187 		    ~(VSUID | VSGID | VSVTX)) {
5188 			status = NFS4ERR_INVAL;
5189 			goto done;
5190 		}
5191 	}
5192 
5193 	/* Check stateid only if size has been set */
5194 	if (sarg.vap->va_mask & AT_SIZE) {
5195 		trunc = (sarg.vap->va_size == 0);
5196 		status = rfs4_check_stateid(FWRITE, cs->vp, stateid,
5197 		    trunc, &cs->deleg, sarg.vap->va_mask & AT_SIZE, &ct);
5198 		if (status != NFS4_OK)
5199 			goto done;
5200 	} else {
5201 		ct.cc_sysid = 0;
5202 		ct.cc_pid = 0;
5203 		ct.cc_caller_id = nfs4_srv_caller_id;
5204 		ct.cc_flags = CC_DONTBLOCK;
5205 	}
5206 
5207 	/* XXX start of possible race with delegations */
5208 
5209 	/*
5210 	 * We need to specially handle size changes because it is
5211 	 * possible for the client to create a file with read-only
5212 	 * modes, but with the file opened for writing. If the client
5213 	 * then tries to set the file size, e.g. ftruncate(3C),
5214 	 * fcntl(F_FREESP), the normal access checking done in
5215 	 * VOP_SETATTR would prevent the client from doing it even though
5216 	 * it should be allowed to do so.  To get around this, we do the
5217 	 * access checking for ourselves and use VOP_SPACE which doesn't
5218 	 * do the access checking.
5219 	 * Also the client should not be allowed to change the file
5220 	 * size if there is a conflicting non-blocking mandatory lock in
5221 	 * the region of the change.
5222 	 */
5223 	if (vp->v_type == VREG && (sarg.vap->va_mask & AT_SIZE)) {
5224 		u_offset_t offset;
5225 		ssize_t length;
5226 
5227 		/*
5228 		 * ufs_setattr clears AT_SIZE from vap->va_mask, but
5229 		 * before returning, sarg.vap->va_mask is used to
5230 		 * generate the setattr reply bitmap.  We also clear
5231 		 * AT_SIZE below before calling VOP_SPACE.  For both
5232 		 * of these cases, the va_mask needs to be saved here
5233 		 * and restored after calling VOP_SETATTR.
5234 		 */
5235 		saved_mask = sarg.vap->va_mask;
5236 
5237 		/*
5238 		 * Check any possible conflict due to NBMAND locks.
5239 		 * Get into critical region before VOP_GETATTR, so the
5240 		 * size attribute is valid when checking conflicts.
5241 		 */
5242 		if (nbl_need_check(vp)) {
5243 			nbl_start_crit(vp, RW_READER);
5244 			in_crit = 1;
5245 		}
5246 
5247 		bva.va_mask = AT_UID|AT_SIZE;
5248 		if (error = VOP_GETATTR(vp, &bva, 0, cr, &ct)) {
5249 			status = puterrno4(error);
5250 			goto done;
5251 		}
5252 
5253 		if (in_crit) {
5254 			if (sarg.vap->va_size < bva.va_size) {
5255 				offset = sarg.vap->va_size;
5256 				length = bva.va_size - sarg.vap->va_size;
5257 			} else {
5258 				offset = bva.va_size;
5259 				length = sarg.vap->va_size - bva.va_size;
5260 			}
5261 			if (nbl_conflict(vp, NBL_WRITE, offset, length, 0,
5262 			    &ct)) {
5263 				status = NFS4ERR_LOCKED;
5264 				goto done;
5265 			}
5266 		}
5267 
5268 		if (crgetuid(cr) == bva.va_uid) {
5269 			sarg.vap->va_mask &= ~AT_SIZE;
5270 			bf.l_type = F_WRLCK;
5271 			bf.l_whence = 0;
5272 			bf.l_start = (off64_t)sarg.vap->va_size;
5273 			bf.l_len = 0;
5274 			bf.l_sysid = 0;
5275 			bf.l_pid = 0;
5276 			error = VOP_SPACE(vp, F_FREESP, &bf, FWRITE,
5277 			    (offset_t)sarg.vap->va_size, cr, &ct);
5278 		}
5279 	}
5280 
5281 	if (!error && sarg.vap->va_mask != 0)
5282 		error = VOP_SETATTR(vp, sarg.vap, sarg.flag, cr, &ct);
5283 
5284 	/* restore va_mask -- ufs_setattr clears AT_SIZE */
5285 	if (saved_mask & AT_SIZE)
5286 		sarg.vap->va_mask |= AT_SIZE;
5287 
5288 	/*
5289 	 * If an ACL was being set, it has been delayed until now,
5290 	 * in order to set the mode (via the VOP_SETATTR() above) first.
5291 	 */
5292 	if ((! error) && (fattrp->attrmask & FATTR4_ACL_MASK)) {
5293 		int i;
5294 
5295 		for (i = 0; i < NFS4_MAXNUM_ATTRS; i++)
5296 			if (ntov.amap[i] == FATTR4_ACL)
5297 				break;
5298 		if (i < NFS4_MAXNUM_ATTRS) {
5299 			error = (*nfs4_ntov_map[FATTR4_ACL].sv_getit)(
5300 			    NFS4ATTR_SETIT, &sarg, &ntov.na[i]);
5301 			if (error == 0) {
5302 				*resp |= FATTR4_ACL_MASK;
5303 			} else if (error == ENOTSUP) {
5304 				(void) rfs4_verify_attr(&sarg, resp, &ntov);
5305 				status = NFS4ERR_ATTRNOTSUPP;
5306 				goto done;
5307 			}
5308 		} else {
5309 			NFS4_DEBUG(rfs4_debug,
5310 			    (CE_NOTE, "do_rfs4_op_setattr: "
5311 			    "unable to find ACL in fattr4"));
5312 			error = EINVAL;
5313 		}
5314 	}
5315 
5316 	if (error) {
5317 		/* check if a monitor detected a delegation conflict */
5318 		if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
5319 			status = NFS4ERR_DELAY;
5320 		else
5321 			status = puterrno4(error);
5322 
5323 		/*
5324 		 * Set the response bitmap when setattr failed.
5325 		 * If VOP_SETATTR partially succeeded, test by doing a
5326 		 * VOP_GETATTR on the object and comparing the data
5327 		 * to the setattr arguments.
5328 		 */
5329 		(void) rfs4_verify_attr(&sarg, resp, &ntov);
5330 	} else {
5331 		/*
5332 		 * Force modified metadata out to stable storage.
5333 		 */
5334 		(void) VOP_FSYNC(vp, FNODSYNC, cr, &ct);
5335 		/*
5336 		 * Set response bitmap
5337 		 */
5338 		nfs4_vmask_to_nmask_set(sarg.vap->va_mask, resp);
5339 	}
5340 
5341 /* Return early and already have a NFSv4 error */
5342 done:
5343 	/*
5344 	 * Except for nfs4_vmask_to_nmask_set(), vattr --> fattr
5345 	 * conversion sets both readable and writeable NFS4 attrs
5346 	 * for AT_MTIME and AT_ATIME.  The line below masks out
5347 	 * unrequested attrs from the setattr result bitmap.  This
5348 	 * is placed after the done: label to catch the ATTRNOTSUP
5349 	 * case.
5350 	 */
5351 	*resp &= fattrp->attrmask;
5352 
5353 	if (in_crit)
5354 		nbl_end_crit(vp);
5355 
5356 	nfs4_ntov_table_free(&ntov, &sarg);
5357 
5358 	return (status);
5359 }
5360 
5361 /* ARGSUSED */
5362 static void
5363 rfs4_op_setattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5364     struct compound_state *cs)
5365 {
5366 	SETATTR4args *args = &argop->nfs_argop4_u.opsetattr;
5367 	SETATTR4res *resp = &resop->nfs_resop4_u.opsetattr;
5368 	bslabel_t *clabel;
5369 
5370 	DTRACE_NFSV4_2(op__setattr__start, struct compound_state *, cs,
5371 	    SETATTR4args *, args);
5372 
5373 	if (cs->vp == NULL) {
5374 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5375 		goto out;
5376 	}
5377 
5378 	/*
5379 	 * If there is an unshared filesystem mounted on this vnode,
5380 	 * do not allow to setattr on this vnode.
5381 	 */
5382 	if (vn_ismntpt(cs->vp)) {
5383 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5384 		goto out;
5385 	}
5386 
5387 	resp->attrsset = 0;
5388 
5389 	if (rdonly4(req, cs)) {
5390 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5391 		goto out;
5392 	}
5393 
5394 	/* check label before setting attributes */
5395 	if (is_system_labeled()) {
5396 		ASSERT(req->rq_label != NULL);
5397 		clabel = req->rq_label;
5398 		DTRACE_PROBE2(tx__rfs4__log__info__opsetattr__clabel, char *,
5399 		    "got client label from request(1)",
5400 		    struct svc_req *, req);
5401 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
5402 			if (!do_rfs_label_check(clabel, cs->vp,
5403 			    EQUALITY_CHECK, cs->exi)) {
5404 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
5405 				goto out;
5406 			}
5407 		}
5408 	}
5409 
5410 	*cs->statusp = resp->status =
5411 	    do_rfs4_op_setattr(&resp->attrsset, &args->obj_attributes, cs,
5412 	    &args->stateid);
5413 
5414 out:
5415 	DTRACE_NFSV4_2(op__setattr__done, struct compound_state *, cs,
5416 	    SETATTR4res *, resp);
5417 }
5418 
5419 /* ARGSUSED */
5420 static void
5421 rfs4_op_verify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5422     struct compound_state *cs)
5423 {
5424 	/*
5425 	 * verify and nverify are exactly the same, except that nverify
5426 	 * succeeds when some argument changed, and verify succeeds when
5427 	 * when none changed.
5428 	 */
5429 
5430 	VERIFY4args  *args = &argop->nfs_argop4_u.opverify;
5431 	VERIFY4res *resp = &resop->nfs_resop4_u.opverify;
5432 
5433 	int error;
5434 	struct nfs4_svgetit_arg sarg;
5435 	struct statvfs64 sb;
5436 	struct nfs4_ntov_table ntov;
5437 
5438 	DTRACE_NFSV4_2(op__verify__start, struct compound_state *, cs,
5439 	    VERIFY4args *, args);
5440 
5441 	if (cs->vp == NULL) {
5442 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5443 		goto out;
5444 	}
5445 
5446 	sarg.sbp = &sb;
5447 	sarg.is_referral = B_FALSE;
5448 	nfs4_ntov_table_init(&ntov);
5449 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5450 	    &sarg, &ntov, NFS4ATTR_VERIT);
5451 	if (resp->status != NFS4_OK) {
5452 		/*
5453 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5454 		 * so could return -1 for "no match".
5455 		 */
5456 		if (resp->status == -1)
5457 			resp->status = NFS4ERR_NOT_SAME;
5458 		goto done;
5459 	}
5460 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5461 	switch (error) {
5462 	case 0:
5463 		resp->status = NFS4_OK;
5464 		break;
5465 	case -1:
5466 		resp->status = NFS4ERR_NOT_SAME;
5467 		break;
5468 	default:
5469 		resp->status = puterrno4(error);
5470 		break;
5471 	}
5472 done:
5473 	*cs->statusp = resp->status;
5474 	nfs4_ntov_table_free(&ntov, &sarg);
5475 out:
5476 	DTRACE_NFSV4_2(op__verify__done, struct compound_state *, cs,
5477 	    VERIFY4res *, resp);
5478 }
5479 
5480 /* ARGSUSED */
5481 static void
5482 rfs4_op_nverify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5483     struct compound_state *cs)
5484 {
5485 	/*
5486 	 * verify and nverify are exactly the same, except that nverify
5487 	 * succeeds when some argument changed, and verify succeeds when
5488 	 * when none changed.
5489 	 */
5490 
5491 	NVERIFY4args  *args = &argop->nfs_argop4_u.opnverify;
5492 	NVERIFY4res *resp = &resop->nfs_resop4_u.opnverify;
5493 
5494 	int error;
5495 	struct nfs4_svgetit_arg sarg;
5496 	struct statvfs64 sb;
5497 	struct nfs4_ntov_table ntov;
5498 
5499 	DTRACE_NFSV4_2(op__nverify__start, struct compound_state *, cs,
5500 	    NVERIFY4args *, args);
5501 
5502 	if (cs->vp == NULL) {
5503 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5504 		DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5505 		    NVERIFY4res *, resp);
5506 		return;
5507 	}
5508 	sarg.sbp = &sb;
5509 	sarg.is_referral = B_FALSE;
5510 	nfs4_ntov_table_init(&ntov);
5511 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5512 	    &sarg, &ntov, NFS4ATTR_VERIT);
5513 	if (resp->status != NFS4_OK) {
5514 		/*
5515 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5516 		 * so could return -1 for "no match".
5517 		 */
5518 		if (resp->status == -1)
5519 			resp->status = NFS4_OK;
5520 		goto done;
5521 	}
5522 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5523 	switch (error) {
5524 	case 0:
5525 		resp->status = NFS4ERR_SAME;
5526 		break;
5527 	case -1:
5528 		resp->status = NFS4_OK;
5529 		break;
5530 	default:
5531 		resp->status = puterrno4(error);
5532 		break;
5533 	}
5534 done:
5535 	*cs->statusp = resp->status;
5536 	nfs4_ntov_table_free(&ntov, &sarg);
5537 
5538 	DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5539 	    NVERIFY4res *, resp);
5540 }
5541 
5542 /*
5543  * XXX - This should live in an NFS header file.
5544  */
5545 #define	MAX_IOVECS	12
5546 
5547 /* ARGSUSED */
5548 static void
5549 rfs4_op_write(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5550     struct compound_state *cs)
5551 {
5552 	WRITE4args *args = &argop->nfs_argop4_u.opwrite;
5553 	WRITE4res *resp = &resop->nfs_resop4_u.opwrite;
5554 	int error;
5555 	vnode_t *vp;
5556 	struct vattr bva;
5557 	u_offset_t rlimit;
5558 	struct uio uio;
5559 	struct iovec iov[MAX_IOVECS];
5560 	struct iovec *iovp;
5561 	int iovcnt;
5562 	int ioflag;
5563 	cred_t *savecred, *cr;
5564 	bool_t *deleg = &cs->deleg;
5565 	nfsstat4 stat;
5566 	int in_crit = 0;
5567 	caller_context_t ct;
5568 
5569 	DTRACE_NFSV4_2(op__write__start, struct compound_state *, cs,
5570 	    WRITE4args *, args);
5571 
5572 	vp = cs->vp;
5573 	if (vp == NULL) {
5574 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5575 		goto out;
5576 	}
5577 	if (cs->access == CS_ACCESS_DENIED) {
5578 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5579 		goto out;
5580 	}
5581 
5582 	cr = cs->cr;
5583 
5584 	if ((stat = rfs4_check_stateid(FWRITE, vp, &args->stateid, FALSE,
5585 	    deleg, TRUE, &ct)) != NFS4_OK) {
5586 		*cs->statusp = resp->status = stat;
5587 		goto out;
5588 	}
5589 
5590 	/*
5591 	 * We have to enter the critical region before calling VOP_RWLOCK
5592 	 * to avoid a deadlock with ufs.
5593 	 */
5594 	if (nbl_need_check(vp)) {
5595 		nbl_start_crit(vp, RW_READER);
5596 		in_crit = 1;
5597 		if (nbl_conflict(vp, NBL_WRITE,
5598 		    args->offset, args->data_len, 0, &ct)) {
5599 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
5600 			goto out;
5601 		}
5602 	}
5603 
5604 	bva.va_mask = AT_MODE | AT_UID;
5605 	error = VOP_GETATTR(vp, &bva, 0, cr, &ct);
5606 
5607 	/*
5608 	 * If we can't get the attributes, then we can't do the
5609 	 * right access checking.  So, we'll fail the request.
5610 	 */
5611 	if (error) {
5612 		*cs->statusp = resp->status = puterrno4(error);
5613 		goto out;
5614 	}
5615 
5616 	if (rdonly4(req, cs)) {
5617 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5618 		goto out;
5619 	}
5620 
5621 	if (vp->v_type != VREG) {
5622 		*cs->statusp = resp->status =
5623 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
5624 		goto out;
5625 	}
5626 
5627 	if (crgetuid(cr) != bva.va_uid &&
5628 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, &ct))) {
5629 		*cs->statusp = resp->status = puterrno4(error);
5630 		goto out;
5631 	}
5632 
5633 	if (MANDLOCK(vp, bva.va_mode)) {
5634 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5635 		goto out;
5636 	}
5637 
5638 	if (args->data_len == 0) {
5639 		*cs->statusp = resp->status = NFS4_OK;
5640 		resp->count = 0;
5641 		resp->committed = args->stable;
5642 		resp->writeverf = Write4verf;
5643 		goto out;
5644 	}
5645 
5646 	if (args->mblk != NULL) {
5647 		mblk_t *m;
5648 		uint_t bytes, round_len;
5649 
5650 		iovcnt = 0;
5651 		bytes = 0;
5652 		round_len = roundup(args->data_len, BYTES_PER_XDR_UNIT);
5653 		for (m = args->mblk;
5654 		    m != NULL && bytes < round_len;
5655 		    m = m->b_cont) {
5656 			iovcnt++;
5657 			bytes += MBLKL(m);
5658 		}
5659 #ifdef DEBUG
5660 		/* should have ended on an mblk boundary */
5661 		if (bytes != round_len) {
5662 			printf("bytes=0x%x, round_len=0x%x, req len=0x%x\n",
5663 			    bytes, round_len, args->data_len);
5664 			printf("args=%p, args->mblk=%p, m=%p", (void *)args,
5665 			    (void *)args->mblk, (void *)m);
5666 			ASSERT(bytes == round_len);
5667 		}
5668 #endif
5669 		if (iovcnt <= MAX_IOVECS) {
5670 			iovp = iov;
5671 		} else {
5672 			iovp = kmem_alloc(sizeof (*iovp) * iovcnt, KM_SLEEP);
5673 		}
5674 		mblk_to_iov(args->mblk, iovcnt, iovp);
5675 	} else if (args->rlist != NULL) {
5676 		iovcnt = 1;
5677 		iovp = iov;
5678 		iovp->iov_base = (char *)((args->rlist)->u.c_daddr3);
5679 		iovp->iov_len = args->data_len;
5680 	} else {
5681 		iovcnt = 1;
5682 		iovp = iov;
5683 		iovp->iov_base = args->data_val;
5684 		iovp->iov_len = args->data_len;
5685 	}
5686 
5687 	uio.uio_iov = iovp;
5688 	uio.uio_iovcnt = iovcnt;
5689 
5690 	uio.uio_segflg = UIO_SYSSPACE;
5691 	uio.uio_extflg = UIO_COPY_DEFAULT;
5692 	uio.uio_loffset = args->offset;
5693 	uio.uio_resid = args->data_len;
5694 	uio.uio_llimit = curproc->p_fsz_ctl;
5695 	rlimit = uio.uio_llimit - args->offset;
5696 	if (rlimit < (u_offset_t)uio.uio_resid)
5697 		uio.uio_resid = (int)rlimit;
5698 
5699 	if (args->stable == UNSTABLE4)
5700 		ioflag = 0;
5701 	else if (args->stable == FILE_SYNC4)
5702 		ioflag = FSYNC;
5703 	else if (args->stable == DATA_SYNC4)
5704 		ioflag = FDSYNC;
5705 	else {
5706 		if (iovp != iov)
5707 			kmem_free(iovp, sizeof (*iovp) * iovcnt);
5708 		*cs->statusp = resp->status = NFS4ERR_INVAL;
5709 		goto out;
5710 	}
5711 
5712 	/*
5713 	 * We're changing creds because VM may fault and we need
5714 	 * the cred of the current thread to be used if quota
5715 	 * checking is enabled.
5716 	 */
5717 	savecred = curthread->t_cred;
5718 	curthread->t_cred = cr;
5719 	error = do_io(FWRITE, vp, &uio, ioflag, cr, &ct);
5720 	curthread->t_cred = savecred;
5721 
5722 	if (iovp != iov)
5723 		kmem_free(iovp, sizeof (*iovp) * iovcnt);
5724 
5725 	if (error) {
5726 		*cs->statusp = resp->status = puterrno4(error);
5727 		goto out;
5728 	}
5729 
5730 	*cs->statusp = resp->status = NFS4_OK;
5731 	resp->count = args->data_len - uio.uio_resid;
5732 
5733 	if (ioflag == 0)
5734 		resp->committed = UNSTABLE4;
5735 	else
5736 		resp->committed = FILE_SYNC4;
5737 
5738 	resp->writeverf = Write4verf;
5739 
5740 out:
5741 	if (in_crit)
5742 		nbl_end_crit(vp);
5743 
5744 	DTRACE_NFSV4_2(op__write__done, struct compound_state *, cs,
5745 	    WRITE4res *, resp);
5746 }
5747 
5748 
5749 /* XXX put in a header file */
5750 extern int	sec_svc_getcred(struct svc_req *, cred_t *,  caddr_t *, int *);
5751 
5752 void
5753 rfs4_compound(COMPOUND4args *args, COMPOUND4res *resp, struct exportinfo *exi,
5754     struct svc_req *req, cred_t *cr, int *rv)
5755 {
5756 	uint_t i;
5757 	struct compound_state cs;
5758 
5759 	if (rv != NULL)
5760 		*rv = 0;
5761 	rfs4_init_compound_state(&cs);
5762 	/*
5763 	 * Form a reply tag by copying over the reqeuest tag.
5764 	 */
5765 	resp->tag.utf8string_val =
5766 	    kmem_alloc(args->tag.utf8string_len, KM_SLEEP);
5767 	resp->tag.utf8string_len = args->tag.utf8string_len;
5768 	bcopy(args->tag.utf8string_val, resp->tag.utf8string_val,
5769 	    resp->tag.utf8string_len);
5770 
5771 	cs.statusp = &resp->status;
5772 	cs.req = req;
5773 	resp->array = NULL;
5774 	resp->array_len = 0;
5775 
5776 	/*
5777 	 * XXX for now, minorversion should be zero
5778 	 */
5779 	if (args->minorversion != NFS4_MINORVERSION) {
5780 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5781 		    &cs, COMPOUND4args *, args);
5782 		resp->status = NFS4ERR_MINOR_VERS_MISMATCH;
5783 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5784 		    &cs, COMPOUND4res *, resp);
5785 		return;
5786 	}
5787 
5788 	if (args->array_len == 0) {
5789 		resp->status = NFS4_OK;
5790 		return;
5791 	}
5792 
5793 	ASSERT(exi == NULL);
5794 	ASSERT(cr == NULL);
5795 
5796 	cr = crget();
5797 	ASSERT(cr != NULL);
5798 
5799 	if (sec_svc_getcred(req, cr, &cs.principal, &cs.nfsflavor) == 0) {
5800 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5801 		    &cs, COMPOUND4args *, args);
5802 		crfree(cr);
5803 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5804 		    &cs, COMPOUND4res *, resp);
5805 		svcerr_badcred(req->rq_xprt);
5806 		if (rv != NULL)
5807 			*rv = 1;
5808 		return;
5809 	}
5810 	resp->array_len = args->array_len;
5811 	resp->array = kmem_zalloc(args->array_len * sizeof (nfs_resop4),
5812 	    KM_SLEEP);
5813 
5814 	cs.basecr = cr;
5815 
5816 	DTRACE_NFSV4_2(compound__start, struct compound_state *, &cs,
5817 	    COMPOUND4args *, args);
5818 
5819 	/*
5820 	 * For now, NFS4 compound processing must be protected by
5821 	 * exported_lock because it can access more than one exportinfo
5822 	 * per compound and share/unshare can now change multiple
5823 	 * exinfo structs.  The NFS2/3 code only refs 1 exportinfo
5824 	 * per proc (excluding public exinfo), and exi_count design
5825 	 * is sufficient to protect concurrent execution of NFS2/3
5826 	 * ops along with unexport.  This lock will be removed as
5827 	 * part of the NFSv4 phase 2 namespace redesign work.
5828 	 */
5829 	rw_enter(&exported_lock, RW_READER);
5830 
5831 	/*
5832 	 * If this is the first compound we've seen, we need to start all
5833 	 * new instances' grace periods.
5834 	 */
5835 	if (rfs4_seen_first_compound == 0) {
5836 		rfs4_grace_start_new();
5837 		/*
5838 		 * This must be set after rfs4_grace_start_new(), otherwise
5839 		 * another thread could proceed past here before the former
5840 		 * is finished.
5841 		 */
5842 		rfs4_seen_first_compound = 1;
5843 	}
5844 
5845 	for (i = 0; i < args->array_len && cs.cont; i++) {
5846 		nfs_argop4 *argop;
5847 		nfs_resop4 *resop;
5848 		uint_t op;
5849 
5850 		argop = &args->array[i];
5851 		resop = &resp->array[i];
5852 		resop->resop = argop->argop;
5853 		op = (uint_t)resop->resop;
5854 
5855 		if (op < rfsv4disp_cnt) {
5856 			/*
5857 			 * Count the individual ops here; NULL and COMPOUND
5858 			 * are counted in common_dispatch()
5859 			 */
5860 			rfsproccnt_v4_ptr[op].value.ui64++;
5861 
5862 			NFS4_DEBUG(rfs4_debug > 1,
5863 			    (CE_NOTE, "Executing %s", rfs4_op_string[op]));
5864 			(*rfsv4disptab[op].dis_proc)(argop, resop, req, &cs);
5865 			NFS4_DEBUG(rfs4_debug > 1, (CE_NOTE, "%s returned %d",
5866 			    rfs4_op_string[op], *cs.statusp));
5867 			if (*cs.statusp != NFS4_OK)
5868 				cs.cont = FALSE;
5869 		} else {
5870 			/*
5871 			 * This is effectively dead code since XDR code
5872 			 * will have already returned BADXDR if op doesn't
5873 			 * decode to legal value.  This only done for a
5874 			 * day when XDR code doesn't verify v4 opcodes.
5875 			 */
5876 			op = OP_ILLEGAL;
5877 			rfsproccnt_v4_ptr[OP_ILLEGAL_IDX].value.ui64++;
5878 
5879 			rfs4_op_illegal(argop, resop, req, &cs);
5880 			cs.cont = FALSE;
5881 		}
5882 
5883 		/*
5884 		 * If not at last op, and if we are to stop, then
5885 		 * compact the results array.
5886 		 */
5887 		if ((i + 1) < args->array_len && !cs.cont) {
5888 			nfs_resop4 *new_res = kmem_alloc(
5889 			    (i+1) * sizeof (nfs_resop4), KM_SLEEP);
5890 			bcopy(resp->array,
5891 			    new_res, (i+1) * sizeof (nfs_resop4));
5892 			kmem_free(resp->array,
5893 			    args->array_len * sizeof (nfs_resop4));
5894 
5895 			resp->array_len =  i + 1;
5896 			resp->array = new_res;
5897 		}
5898 	}
5899 
5900 	rw_exit(&exported_lock);
5901 
5902 	DTRACE_NFSV4_2(compound__done, struct compound_state *, &cs,
5903 	    COMPOUND4res *, resp);
5904 
5905 	if (cs.vp)
5906 		VN_RELE(cs.vp);
5907 	if (cs.saved_vp)
5908 		VN_RELE(cs.saved_vp);
5909 	if (cs.saved_fh.nfs_fh4_val)
5910 		kmem_free(cs.saved_fh.nfs_fh4_val, NFS4_FHSIZE);
5911 
5912 	if (cs.basecr)
5913 		crfree(cs.basecr);
5914 	if (cs.cr)
5915 		crfree(cs.cr);
5916 	/*
5917 	 * done with this compound request, free the label
5918 	 */
5919 
5920 	if (req->rq_label != NULL) {
5921 		kmem_free(req->rq_label, sizeof (bslabel_t));
5922 		req->rq_label = NULL;
5923 	}
5924 }
5925 
5926 /*
5927  * XXX because of what appears to be duplicate calls to rfs4_compound_free
5928  * XXX zero out the tag and array values. Need to investigate why the
5929  * XXX calls occur, but at least prevent the panic for now.
5930  */
5931 void
5932 rfs4_compound_free(COMPOUND4res *resp)
5933 {
5934 	uint_t i;
5935 
5936 	if (resp->tag.utf8string_val) {
5937 		UTF8STRING_FREE(resp->tag)
5938 	}
5939 
5940 	for (i = 0; i < resp->array_len; i++) {
5941 		nfs_resop4 *resop;
5942 		uint_t op;
5943 
5944 		resop = &resp->array[i];
5945 		op = (uint_t)resop->resop;
5946 		if (op < rfsv4disp_cnt) {
5947 			(*rfsv4disptab[op].dis_resfree)(resop);
5948 		}
5949 	}
5950 	if (resp->array != NULL) {
5951 		kmem_free(resp->array, resp->array_len * sizeof (nfs_resop4));
5952 	}
5953 }
5954 
5955 /*
5956  * Process the value of the compound request rpc flags, as a bit-AND
5957  * of the individual per-op flags (idempotent, allowork, publicfh_ok)
5958  */
5959 void
5960 rfs4_compound_flagproc(COMPOUND4args *args, int *flagp)
5961 {
5962 	int i;
5963 	int flag = RPC_ALL;
5964 
5965 	for (i = 0; flag && i < args->array_len; i++) {
5966 		uint_t op;
5967 
5968 		op = (uint_t)args->array[i].argop;
5969 
5970 		if (op < rfsv4disp_cnt)
5971 			flag &= rfsv4disptab[op].dis_flags;
5972 		else
5973 			flag = 0;
5974 	}
5975 	*flagp = flag;
5976 }
5977 
5978 nfsstat4
5979 rfs4_client_sysid(rfs4_client_t *cp, sysid_t *sp)
5980 {
5981 	nfsstat4 e;
5982 
5983 	rfs4_dbe_lock(cp->rc_dbe);
5984 
5985 	if (cp->rc_sysidt != LM_NOSYSID) {
5986 		*sp = cp->rc_sysidt;
5987 		e = NFS4_OK;
5988 
5989 	} else if ((cp->rc_sysidt = lm_alloc_sysidt()) != LM_NOSYSID) {
5990 		*sp = cp->rc_sysidt;
5991 		e = NFS4_OK;
5992 
5993 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
5994 		    "rfs4_client_sysid: allocated 0x%x\n", *sp));
5995 	} else
5996 		e = NFS4ERR_DELAY;
5997 
5998 	rfs4_dbe_unlock(cp->rc_dbe);
5999 	return (e);
6000 }
6001 
6002 #if defined(DEBUG) && ! defined(lint)
6003 static void lock_print(char *str, int operation, struct flock64 *flk)
6004 {
6005 	char *op, *type;
6006 
6007 	switch (operation) {
6008 	case F_GETLK: op = "F_GETLK";
6009 		break;
6010 	case F_SETLK: op = "F_SETLK";
6011 		break;
6012 	case F_SETLK_NBMAND: op = "F_SETLK_NBMAND";
6013 		break;
6014 	default: op = "F_UNKNOWN";
6015 		break;
6016 	}
6017 	switch (flk->l_type) {
6018 	case F_UNLCK: type = "F_UNLCK";
6019 		break;
6020 	case F_RDLCK: type = "F_RDLCK";
6021 		break;
6022 	case F_WRLCK: type = "F_WRLCK";
6023 		break;
6024 	default: type = "F_UNKNOWN";
6025 		break;
6026 	}
6027 
6028 	ASSERT(flk->l_whence == 0);
6029 	cmn_err(CE_NOTE, "%s:  %s, type = %s, off = %llx len = %llx pid = %d",
6030 	    str, op, type, (longlong_t)flk->l_start,
6031 	    flk->l_len ? (longlong_t)flk->l_len : ~0LL, flk->l_pid);
6032 }
6033 
6034 #define	LOCK_PRINT(d, s, t, f) if (d) lock_print(s, t, f)
6035 #else
6036 #define	LOCK_PRINT(d, s, t, f)
6037 #endif
6038 
6039 /*ARGSUSED*/
6040 static bool_t
6041 creds_ok(cred_set_t cr_set, struct svc_req *req, struct compound_state *cs)
6042 {
6043 	return (TRUE);
6044 }
6045 
6046 /*
6047  * Look up the pathname using the vp in cs as the directory vnode.
6048  * cs->vp will be the vnode for the file on success
6049  */
6050 
6051 static nfsstat4
6052 rfs4_lookup(component4 *component, struct svc_req *req,
6053     struct compound_state *cs)
6054 {
6055 	char *nm;
6056 	uint32_t len;
6057 	nfsstat4 status;
6058 	struct sockaddr *ca;
6059 	char *name;
6060 
6061 	if (cs->vp == NULL) {
6062 		return (NFS4ERR_NOFILEHANDLE);
6063 	}
6064 	if (cs->vp->v_type != VDIR) {
6065 		return (NFS4ERR_NOTDIR);
6066 	}
6067 
6068 	status = utf8_dir_verify(component);
6069 	if (status != NFS4_OK)
6070 		return (status);
6071 
6072 	nm = utf8_to_fn(component, &len, NULL);
6073 	if (nm == NULL) {
6074 		return (NFS4ERR_INVAL);
6075 	}
6076 
6077 	if (len > MAXNAMELEN) {
6078 		kmem_free(nm, len);
6079 		return (NFS4ERR_NAMETOOLONG);
6080 	}
6081 
6082 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6083 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6084 	    MAXPATHLEN + 1);
6085 
6086 	if (name == NULL) {
6087 		kmem_free(nm, len);
6088 		return (NFS4ERR_INVAL);
6089 	}
6090 
6091 	status = do_rfs4_op_lookup(name, req, cs);
6092 
6093 	if (name != nm)
6094 		kmem_free(name, MAXPATHLEN + 1);
6095 
6096 	kmem_free(nm, len);
6097 
6098 	return (status);
6099 }
6100 
6101 static nfsstat4
6102 rfs4_lookupfile(component4 *component, struct svc_req *req,
6103     struct compound_state *cs, uint32_t access, change_info4 *cinfo)
6104 {
6105 	nfsstat4 status;
6106 	vnode_t *dvp = cs->vp;
6107 	vattr_t bva, ava, fva;
6108 	int error;
6109 
6110 	/* Get "before" change value */
6111 	bva.va_mask = AT_CTIME|AT_SEQ;
6112 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6113 	if (error)
6114 		return (puterrno4(error));
6115 
6116 	/* rfs4_lookup may VN_RELE directory */
6117 	VN_HOLD(dvp);
6118 
6119 	status = rfs4_lookup(component, req, cs);
6120 	if (status != NFS4_OK) {
6121 		VN_RELE(dvp);
6122 		return (status);
6123 	}
6124 
6125 	/*
6126 	 * Get "after" change value, if it fails, simply return the
6127 	 * before value.
6128 	 */
6129 	ava.va_mask = AT_CTIME|AT_SEQ;
6130 	if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6131 		ava.va_ctime = bva.va_ctime;
6132 		ava.va_seq = 0;
6133 	}
6134 	VN_RELE(dvp);
6135 
6136 	/*
6137 	 * Validate the file is a file
6138 	 */
6139 	fva.va_mask = AT_TYPE|AT_MODE;
6140 	error = VOP_GETATTR(cs->vp, &fva, 0, cs->cr, NULL);
6141 	if (error)
6142 		return (puterrno4(error));
6143 
6144 	if (fva.va_type != VREG) {
6145 		if (fva.va_type == VDIR)
6146 			return (NFS4ERR_ISDIR);
6147 		if (fva.va_type == VLNK)
6148 			return (NFS4ERR_SYMLINK);
6149 		return (NFS4ERR_INVAL);
6150 	}
6151 
6152 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime);
6153 	NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6154 
6155 	/*
6156 	 * It is undefined if VOP_LOOKUP will change va_seq, so
6157 	 * cinfo.atomic = TRUE only if we have
6158 	 * non-zero va_seq's, and they have not changed.
6159 	 */
6160 	if (bva.va_seq && ava.va_seq && ava.va_seq == bva.va_seq)
6161 		cinfo->atomic = TRUE;
6162 	else
6163 		cinfo->atomic = FALSE;
6164 
6165 	/* Check for mandatory locking */
6166 	cs->mandlock = MANDLOCK(cs->vp, fva.va_mode);
6167 	return (check_open_access(access, cs, req));
6168 }
6169 
6170 static nfsstat4
6171 create_vnode(vnode_t *dvp, char *nm,  vattr_t *vap, createmode4 mode,
6172     timespec32_t *mtime, cred_t *cr, vnode_t **vpp, bool_t *created)
6173 {
6174 	int error;
6175 	nfsstat4 status = NFS4_OK;
6176 	vattr_t va;
6177 
6178 tryagain:
6179 
6180 	/*
6181 	 * The file open mode used is VWRITE.  If the client needs
6182 	 * some other semantic, then it should do the access checking
6183 	 * itself.  It would have been nice to have the file open mode
6184 	 * passed as part of the arguments.
6185 	 */
6186 
6187 	*created = TRUE;
6188 	error = VOP_CREATE(dvp, nm, vap, EXCL, VWRITE, vpp, cr, 0, NULL, NULL);
6189 
6190 	if (error) {
6191 		*created = FALSE;
6192 
6193 		/*
6194 		 * If we got something other than file already exists
6195 		 * then just return this error.  Otherwise, we got
6196 		 * EEXIST.  If we were doing a GUARDED create, then
6197 		 * just return this error.  Otherwise, we need to
6198 		 * make sure that this wasn't a duplicate of an
6199 		 * exclusive create request.
6200 		 *
6201 		 * The assumption is made that a non-exclusive create
6202 		 * request will never return EEXIST.
6203 		 */
6204 
6205 		if (error != EEXIST || mode == GUARDED4) {
6206 			status = puterrno4(error);
6207 			return (status);
6208 		}
6209 		error = VOP_LOOKUP(dvp, nm, vpp, NULL, 0, NULL, cr,
6210 		    NULL, NULL, NULL);
6211 
6212 		if (error) {
6213 			/*
6214 			 * We couldn't find the file that we thought that
6215 			 * we just created.  So, we'll just try creating
6216 			 * it again.
6217 			 */
6218 			if (error == ENOENT)
6219 				goto tryagain;
6220 
6221 			status = puterrno4(error);
6222 			return (status);
6223 		}
6224 
6225 		if (mode == UNCHECKED4) {
6226 			/* existing object must be regular file */
6227 			if ((*vpp)->v_type != VREG) {
6228 				if ((*vpp)->v_type == VDIR)
6229 					status = NFS4ERR_ISDIR;
6230 				else if ((*vpp)->v_type == VLNK)
6231 					status = NFS4ERR_SYMLINK;
6232 				else
6233 					status = NFS4ERR_INVAL;
6234 				VN_RELE(*vpp);
6235 				return (status);
6236 			}
6237 
6238 			return (NFS4_OK);
6239 		}
6240 
6241 		/* Check for duplicate request */
6242 		ASSERT(mtime != 0);
6243 		va.va_mask = AT_MTIME;
6244 		error = VOP_GETATTR(*vpp, &va, 0, cr, NULL);
6245 		if (!error) {
6246 			/* We found the file */
6247 			if (va.va_mtime.tv_sec != mtime->tv_sec ||
6248 			    va.va_mtime.tv_nsec != mtime->tv_nsec) {
6249 				/* but its not our creation */
6250 				VN_RELE(*vpp);
6251 				return (NFS4ERR_EXIST);
6252 			}
6253 			*created = TRUE; /* retrans of create == created */
6254 			return (NFS4_OK);
6255 		}
6256 		VN_RELE(*vpp);
6257 		return (NFS4ERR_EXIST);
6258 	}
6259 
6260 	return (NFS4_OK);
6261 }
6262 
6263 static nfsstat4
6264 check_open_access(uint32_t access, struct compound_state *cs,
6265     struct svc_req *req)
6266 {
6267 	int error;
6268 	vnode_t *vp;
6269 	bool_t readonly;
6270 	cred_t *cr = cs->cr;
6271 
6272 	/* For now we don't allow mandatory locking as per V2/V3 */
6273 	if (cs->access == CS_ACCESS_DENIED || cs->mandlock) {
6274 		return (NFS4ERR_ACCESS);
6275 	}
6276 
6277 	vp = cs->vp;
6278 	ASSERT(cr != NULL && vp->v_type == VREG);
6279 
6280 	/*
6281 	 * If the file system is exported read only and we are trying
6282 	 * to open for write, then return NFS4ERR_ROFS
6283 	 */
6284 
6285 	readonly = rdonly4(req, cs);
6286 
6287 	if ((access & OPEN4_SHARE_ACCESS_WRITE) && readonly)
6288 		return (NFS4ERR_ROFS);
6289 
6290 	if (access & OPEN4_SHARE_ACCESS_READ) {
6291 		if ((VOP_ACCESS(vp, VREAD, 0, cr, NULL) != 0) &&
6292 		    (VOP_ACCESS(vp, VEXEC, 0, cr, NULL) != 0)) {
6293 			return (NFS4ERR_ACCESS);
6294 		}
6295 	}
6296 
6297 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
6298 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
6299 		if (error)
6300 			return (NFS4ERR_ACCESS);
6301 	}
6302 
6303 	return (NFS4_OK);
6304 }
6305 
6306 static nfsstat4
6307 rfs4_createfile(OPEN4args *args, struct svc_req *req, struct compound_state *cs,
6308     change_info4 *cinfo, bitmap4 *attrset, clientid4 clientid)
6309 {
6310 	struct nfs4_svgetit_arg sarg;
6311 	struct nfs4_ntov_table ntov;
6312 
6313 	bool_t ntov_table_init = FALSE;
6314 	struct statvfs64 sb;
6315 	nfsstat4 status;
6316 	vnode_t *vp;
6317 	vattr_t bva, ava, iva, cva, *vap;
6318 	vnode_t *dvp;
6319 	timespec32_t *mtime;
6320 	char *nm = NULL;
6321 	uint_t buflen;
6322 	bool_t created;
6323 	bool_t setsize = FALSE;
6324 	len_t reqsize;
6325 	int error;
6326 	bool_t trunc;
6327 	caller_context_t ct;
6328 	component4 *component;
6329 	bslabel_t *clabel;
6330 	struct sockaddr *ca;
6331 	char *name = NULL;
6332 
6333 	sarg.sbp = &sb;
6334 	sarg.is_referral = B_FALSE;
6335 
6336 	dvp = cs->vp;
6337 
6338 	/* Check if the file system is read only */
6339 	if (rdonly4(req, cs))
6340 		return (NFS4ERR_ROFS);
6341 
6342 	/* check the label of including directory */
6343 	if (is_system_labeled()) {
6344 		ASSERT(req->rq_label != NULL);
6345 		clabel = req->rq_label;
6346 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
6347 		    "got client label from request(1)",
6348 		    struct svc_req *, req);
6349 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
6350 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
6351 			    cs->exi)) {
6352 				return (NFS4ERR_ACCESS);
6353 			}
6354 		}
6355 	}
6356 
6357 	/*
6358 	 * Get the last component of path name in nm. cs will reference
6359 	 * the including directory on success.
6360 	 */
6361 	component = &args->open_claim4_u.file;
6362 	status = utf8_dir_verify(component);
6363 	if (status != NFS4_OK)
6364 		return (status);
6365 
6366 	nm = utf8_to_fn(component, &buflen, NULL);
6367 
6368 	if (nm == NULL)
6369 		return (NFS4ERR_RESOURCE);
6370 
6371 	if (buflen > MAXNAMELEN) {
6372 		kmem_free(nm, buflen);
6373 		return (NFS4ERR_NAMETOOLONG);
6374 	}
6375 
6376 	bva.va_mask = AT_TYPE|AT_CTIME|AT_SEQ;
6377 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6378 	if (error) {
6379 		kmem_free(nm, buflen);
6380 		return (puterrno4(error));
6381 	}
6382 
6383 	if (bva.va_type != VDIR) {
6384 		kmem_free(nm, buflen);
6385 		return (NFS4ERR_NOTDIR);
6386 	}
6387 
6388 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime)
6389 
6390 	switch (args->mode) {
6391 	case GUARDED4:
6392 		/*FALLTHROUGH*/
6393 	case UNCHECKED4:
6394 		nfs4_ntov_table_init(&ntov);
6395 		ntov_table_init = TRUE;
6396 
6397 		*attrset = 0;
6398 		status = do_rfs4_set_attrs(attrset,
6399 		    &args->createhow4_u.createattrs,
6400 		    cs, &sarg, &ntov, NFS4ATTR_SETIT);
6401 
6402 		if (status == NFS4_OK && (sarg.vap->va_mask & AT_TYPE) &&
6403 		    sarg.vap->va_type != VREG) {
6404 			if (sarg.vap->va_type == VDIR)
6405 				status = NFS4ERR_ISDIR;
6406 			else if (sarg.vap->va_type == VLNK)
6407 				status = NFS4ERR_SYMLINK;
6408 			else
6409 				status = NFS4ERR_INVAL;
6410 		}
6411 
6412 		if (status != NFS4_OK) {
6413 			kmem_free(nm, buflen);
6414 			nfs4_ntov_table_free(&ntov, &sarg);
6415 			*attrset = 0;
6416 			return (status);
6417 		}
6418 
6419 		vap = sarg.vap;
6420 		vap->va_type = VREG;
6421 		vap->va_mask |= AT_TYPE;
6422 
6423 		if ((vap->va_mask & AT_MODE) == 0) {
6424 			vap->va_mask |= AT_MODE;
6425 			vap->va_mode = (mode_t)0600;
6426 		}
6427 
6428 		if (vap->va_mask & AT_SIZE) {
6429 
6430 			/* Disallow create with a non-zero size */
6431 
6432 			if ((reqsize = sarg.vap->va_size) != 0) {
6433 				kmem_free(nm, buflen);
6434 				nfs4_ntov_table_free(&ntov, &sarg);
6435 				*attrset = 0;
6436 				return (NFS4ERR_INVAL);
6437 			}
6438 			setsize = TRUE;
6439 		}
6440 		break;
6441 
6442 	case EXCLUSIVE4:
6443 		/* prohibit EXCL create of named attributes */
6444 		if (dvp->v_flag & V_XATTRDIR) {
6445 			kmem_free(nm, buflen);
6446 			*attrset = 0;
6447 			return (NFS4ERR_INVAL);
6448 		}
6449 
6450 		cva.va_mask = AT_TYPE | AT_MTIME | AT_MODE;
6451 		cva.va_type = VREG;
6452 		/*
6453 		 * Ensure no time overflows. Assumes underlying
6454 		 * filesystem supports at least 32 bits.
6455 		 * Truncate nsec to usec resolution to allow valid
6456 		 * compares even if the underlying filesystem truncates.
6457 		 */
6458 		mtime = (timespec32_t *)&args->createhow4_u.createverf;
6459 		cva.va_mtime.tv_sec = mtime->tv_sec % TIME32_MAX;
6460 		cva.va_mtime.tv_nsec = (mtime->tv_nsec / 1000) * 1000;
6461 		cva.va_mode = (mode_t)0;
6462 		vap = &cva;
6463 
6464 		/*
6465 		 * For EXCL create, attrset is set to the server attr
6466 		 * used to cache the client's verifier.
6467 		 */
6468 		*attrset = FATTR4_TIME_MODIFY_MASK;
6469 		break;
6470 	}
6471 
6472 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6473 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6474 	    MAXPATHLEN  + 1);
6475 
6476 	if (name == NULL) {
6477 		kmem_free(nm, buflen);
6478 		return (NFS4ERR_SERVERFAULT);
6479 	}
6480 
6481 	status = create_vnode(dvp, name, vap, args->mode, mtime,
6482 	    cs->cr, &vp, &created);
6483 	if (nm != name)
6484 		kmem_free(name, MAXPATHLEN + 1);
6485 	kmem_free(nm, buflen);
6486 
6487 	if (status != NFS4_OK) {
6488 		if (ntov_table_init)
6489 			nfs4_ntov_table_free(&ntov, &sarg);
6490 		*attrset = 0;
6491 		return (status);
6492 	}
6493 
6494 	trunc = (setsize && !created);
6495 
6496 	if (args->mode != EXCLUSIVE4) {
6497 		bitmap4 createmask = args->createhow4_u.createattrs.attrmask;
6498 
6499 		/*
6500 		 * True verification that object was created with correct
6501 		 * attrs is impossible.  The attrs could have been changed
6502 		 * immediately after object creation.  If attributes did
6503 		 * not verify, the only recourse for the server is to
6504 		 * destroy the object.  Maybe if some attrs (like gid)
6505 		 * are set incorrectly, the object should be destroyed;
6506 		 * however, seems bad as a default policy.  Do we really
6507 		 * want to destroy an object over one of the times not
6508 		 * verifying correctly?  For these reasons, the server
6509 		 * currently sets bits in attrset for createattrs
6510 		 * that were set; however, no verification is done.
6511 		 *
6512 		 * vmask_to_nmask accounts for vattr bits set on create
6513 		 *	[do_rfs4_set_attrs() only sets resp bits for
6514 		 *	 non-vattr/vfs bits.]
6515 		 * Mask off any bits we set by default so as not to return
6516 		 * more attrset bits than were requested in createattrs
6517 		 */
6518 		if (created) {
6519 			nfs4_vmask_to_nmask(sarg.vap->va_mask, attrset);
6520 			*attrset &= createmask;
6521 		} else {
6522 			/*
6523 			 * We did not create the vnode (we tried but it
6524 			 * already existed).  In this case, the only createattr
6525 			 * that the spec allows the server to set is size,
6526 			 * and even then, it can only be set if it is 0.
6527 			 */
6528 			*attrset = 0;
6529 			if (trunc)
6530 				*attrset = FATTR4_SIZE_MASK;
6531 		}
6532 	}
6533 	if (ntov_table_init)
6534 		nfs4_ntov_table_free(&ntov, &sarg);
6535 
6536 	/*
6537 	 * Get the initial "after" sequence number, if it fails,
6538 	 * set to zero, time to before.
6539 	 */
6540 	iva.va_mask = AT_CTIME|AT_SEQ;
6541 	if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL)) {
6542 		iva.va_seq = 0;
6543 		iva.va_ctime = bva.va_ctime;
6544 	}
6545 
6546 	/*
6547 	 * create_vnode attempts to create the file exclusive,
6548 	 * if it already exists the VOP_CREATE will fail and
6549 	 * may not increase va_seq. It is atomic if
6550 	 * we haven't changed the directory, but if it has changed
6551 	 * we don't know what changed it.
6552 	 */
6553 	if (!created) {
6554 		if (bva.va_seq && iva.va_seq &&
6555 		    bva.va_seq == iva.va_seq)
6556 			cinfo->atomic = TRUE;
6557 		else
6558 			cinfo->atomic = FALSE;
6559 		NFS4_SET_FATTR4_CHANGE(cinfo->after, iva.va_ctime);
6560 	} else {
6561 		/*
6562 		 * The entry was created, we need to sync the
6563 		 * directory metadata.
6564 		 */
6565 		(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
6566 
6567 		/*
6568 		 * Get "after" change value, if it fails, simply return the
6569 		 * before value.
6570 		 */
6571 		ava.va_mask = AT_CTIME|AT_SEQ;
6572 		if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6573 			ava.va_ctime = bva.va_ctime;
6574 			ava.va_seq = 0;
6575 		}
6576 
6577 		NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6578 
6579 		/*
6580 		 * The cinfo->atomic = TRUE only if we have
6581 		 * non-zero va_seq's, and it has incremented by exactly one
6582 		 * during the create_vnode and it didn't
6583 		 * change during the VOP_FSYNC.
6584 		 */
6585 		if (bva.va_seq && iva.va_seq && ava.va_seq &&
6586 		    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
6587 			cinfo->atomic = TRUE;
6588 		else
6589 			cinfo->atomic = FALSE;
6590 	}
6591 
6592 	/* Check for mandatory locking and that the size gets set. */
6593 	cva.va_mask = AT_MODE;
6594 	if (setsize)
6595 		cva.va_mask |= AT_SIZE;
6596 
6597 	/* Assume the worst */
6598 	cs->mandlock = TRUE;
6599 
6600 	if (VOP_GETATTR(vp, &cva, 0, cs->cr, NULL) == 0) {
6601 		cs->mandlock = MANDLOCK(cs->vp, cva.va_mode);
6602 
6603 		/*
6604 		 * Truncate the file if necessary; this would be
6605 		 * the case for create over an existing file.
6606 		 */
6607 
6608 		if (trunc) {
6609 			int in_crit = 0;
6610 			rfs4_file_t *fp;
6611 			bool_t create = FALSE;
6612 
6613 			/*
6614 			 * We are writing over an existing file.
6615 			 * Check to see if we need to recall a delegation.
6616 			 */
6617 			rfs4_hold_deleg_policy();
6618 			if ((fp = rfs4_findfile(vp, NULL, &create)) != NULL) {
6619 				if (rfs4_check_delegated_byfp(FWRITE, fp,
6620 				    (reqsize == 0), FALSE, FALSE, &clientid)) {
6621 					rfs4_file_rele(fp);
6622 					rfs4_rele_deleg_policy();
6623 					VN_RELE(vp);
6624 					*attrset = 0;
6625 					return (NFS4ERR_DELAY);
6626 				}
6627 				rfs4_file_rele(fp);
6628 			}
6629 			rfs4_rele_deleg_policy();
6630 
6631 			if (nbl_need_check(vp)) {
6632 				in_crit = 1;
6633 
6634 				ASSERT(reqsize == 0);
6635 
6636 				nbl_start_crit(vp, RW_READER);
6637 				if (nbl_conflict(vp, NBL_WRITE, 0,
6638 				    cva.va_size, 0, NULL)) {
6639 					in_crit = 0;
6640 					nbl_end_crit(vp);
6641 					VN_RELE(vp);
6642 					*attrset = 0;
6643 					return (NFS4ERR_ACCESS);
6644 				}
6645 			}
6646 			ct.cc_sysid = 0;
6647 			ct.cc_pid = 0;
6648 			ct.cc_caller_id = nfs4_srv_caller_id;
6649 			ct.cc_flags = CC_DONTBLOCK;
6650 
6651 			cva.va_mask = AT_SIZE;
6652 			cva.va_size = reqsize;
6653 			(void) VOP_SETATTR(vp, &cva, 0, cs->cr, &ct);
6654 			if (in_crit)
6655 				nbl_end_crit(vp);
6656 		}
6657 	}
6658 
6659 	error = makefh4(&cs->fh, vp, cs->exi);
6660 
6661 	/*
6662 	 * Force modified data and metadata out to stable storage.
6663 	 */
6664 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
6665 
6666 	if (error) {
6667 		VN_RELE(vp);
6668 		*attrset = 0;
6669 		return (puterrno4(error));
6670 	}
6671 
6672 	/* if parent dir is attrdir, set namedattr fh flag */
6673 	if (dvp->v_flag & V_XATTRDIR)
6674 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
6675 
6676 	if (cs->vp)
6677 		VN_RELE(cs->vp);
6678 
6679 	cs->vp = vp;
6680 
6681 	/*
6682 	 * if we did not create the file, we will need to check
6683 	 * the access bits on the file
6684 	 */
6685 
6686 	if (!created) {
6687 		if (setsize)
6688 			args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
6689 		status = check_open_access(args->share_access, cs, req);
6690 		if (status != NFS4_OK)
6691 			*attrset = 0;
6692 	}
6693 	return (status);
6694 }
6695 
6696 /*ARGSUSED*/
6697 static void
6698 rfs4_do_open(struct compound_state *cs, struct svc_req *req,
6699     rfs4_openowner_t *oo, delegreq_t deleg,
6700     uint32_t access, uint32_t deny,
6701     OPEN4res *resp, int deleg_cur)
6702 {
6703 	/* XXX Currently not using req  */
6704 	rfs4_state_t *sp;
6705 	rfs4_file_t *fp;
6706 	bool_t screate = TRUE;
6707 	bool_t fcreate = TRUE;
6708 	uint32_t open_a, share_a;
6709 	uint32_t open_d, share_d;
6710 	rfs4_deleg_state_t *dsp;
6711 	sysid_t sysid;
6712 	nfsstat4 status;
6713 	caller_context_t ct;
6714 	int fflags = 0;
6715 	int recall = 0;
6716 	int err;
6717 	int first_open;
6718 
6719 	/* get the file struct and hold a lock on it during initial open */
6720 	fp = rfs4_findfile_withlock(cs->vp, &cs->fh, &fcreate);
6721 	if (fp == NULL) {
6722 		resp->status = NFS4ERR_RESOURCE;
6723 		DTRACE_PROBE1(nfss__e__do__open1, nfsstat4, resp->status);
6724 		return;
6725 	}
6726 
6727 	sp = rfs4_findstate_by_owner_file(oo, fp, &screate);
6728 	if (sp == NULL) {
6729 		resp->status = NFS4ERR_RESOURCE;
6730 		DTRACE_PROBE1(nfss__e__do__open2, nfsstat4, resp->status);
6731 		/* No need to keep any reference */
6732 		rw_exit(&fp->rf_file_rwlock);
6733 		rfs4_file_rele(fp);
6734 		return;
6735 	}
6736 
6737 	/* try to get the sysid before continuing */
6738 	if ((status = rfs4_client_sysid(oo->ro_client, &sysid)) != NFS4_OK) {
6739 		resp->status = status;
6740 		rfs4_file_rele(fp);
6741 		/* Not a fully formed open; "close" it */
6742 		if (screate == TRUE)
6743 			rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6744 		rfs4_state_rele(sp);
6745 		return;
6746 	}
6747 
6748 	/* Calculate the fflags for this OPEN. */
6749 	if (access & OPEN4_SHARE_ACCESS_READ)
6750 		fflags |= FREAD;
6751 	if (access & OPEN4_SHARE_ACCESS_WRITE)
6752 		fflags |= FWRITE;
6753 
6754 	rfs4_dbe_lock(sp->rs_dbe);
6755 
6756 	/*
6757 	 * Calculate the new deny and access mode that this open is adding to
6758 	 * the file for this open owner;
6759 	 */
6760 	open_d = (deny & ~sp->rs_open_deny);
6761 	open_a = (access & ~sp->rs_open_access);
6762 
6763 	/*
6764 	 * Calculate the new share access and share deny modes that this open
6765 	 * is adding to the file for this open owner;
6766 	 */
6767 	share_a = (access & ~sp->rs_share_access);
6768 	share_d = (deny & ~sp->rs_share_deny);
6769 
6770 	first_open = (sp->rs_open_access & OPEN4_SHARE_ACCESS_BOTH) == 0;
6771 
6772 	/*
6773 	 * Check to see the client has already sent an open for this
6774 	 * open owner on this file with the same share/deny modes.
6775 	 * If so, we don't need to check for a conflict and we don't
6776 	 * need to add another shrlock.  If not, then we need to
6777 	 * check for conflicts in deny and access before checking for
6778 	 * conflicts in delegation.  We don't want to recall a
6779 	 * delegation based on an open that will eventually fail based
6780 	 * on shares modes.
6781 	 */
6782 
6783 	if (share_a || share_d) {
6784 		if ((err = rfs4_share(sp, access, deny)) != 0) {
6785 			rfs4_dbe_unlock(sp->rs_dbe);
6786 			resp->status = err;
6787 
6788 			rfs4_file_rele(fp);
6789 			/* Not a fully formed open; "close" it */
6790 			if (screate == TRUE)
6791 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6792 			rfs4_state_rele(sp);
6793 			return;
6794 		}
6795 	}
6796 
6797 	rfs4_dbe_lock(fp->rf_dbe);
6798 
6799 	/*
6800 	 * Check to see if this file is delegated and if so, if a
6801 	 * recall needs to be done.
6802 	 */
6803 	if (rfs4_check_recall(sp, access)) {
6804 		rfs4_dbe_unlock(fp->rf_dbe);
6805 		rfs4_dbe_unlock(sp->rs_dbe);
6806 		rfs4_recall_deleg(fp, FALSE, sp->rs_owner->ro_client);
6807 		delay(NFS4_DELEGATION_CONFLICT_DELAY);
6808 		rfs4_dbe_lock(sp->rs_dbe);
6809 
6810 		/* if state closed while lock was dropped */
6811 		if (sp->rs_closed) {
6812 			if (share_a || share_d)
6813 				(void) rfs4_unshare(sp);
6814 			rfs4_dbe_unlock(sp->rs_dbe);
6815 			rfs4_file_rele(fp);
6816 			/* Not a fully formed open; "close" it */
6817 			if (screate == TRUE)
6818 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6819 			rfs4_state_rele(sp);
6820 			resp->status = NFS4ERR_OLD_STATEID;
6821 			return;
6822 		}
6823 
6824 		rfs4_dbe_lock(fp->rf_dbe);
6825 		/* Let's see if the delegation was returned */
6826 		if (rfs4_check_recall(sp, access)) {
6827 			rfs4_dbe_unlock(fp->rf_dbe);
6828 			if (share_a || share_d)
6829 				(void) rfs4_unshare(sp);
6830 			rfs4_dbe_unlock(sp->rs_dbe);
6831 			rfs4_file_rele(fp);
6832 			rfs4_update_lease(sp->rs_owner->ro_client);
6833 
6834 			/* Not a fully formed open; "close" it */
6835 			if (screate == TRUE)
6836 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6837 			rfs4_state_rele(sp);
6838 			resp->status = NFS4ERR_DELAY;
6839 			return;
6840 		}
6841 	}
6842 	/*
6843 	 * the share check passed and any delegation conflict has been
6844 	 * taken care of, now call vop_open.
6845 	 * if this is the first open then call vop_open with fflags.
6846 	 * if not, call vn_open_upgrade with just the upgrade flags.
6847 	 *
6848 	 * if the file has been opened already, it will have the current
6849 	 * access mode in the state struct.  if it has no share access, then
6850 	 * this is a new open.
6851 	 *
6852 	 * However, if this is open with CLAIM_DLEGATE_CUR, then don't
6853 	 * call VOP_OPEN(), just do the open upgrade.
6854 	 */
6855 	if (first_open && !deleg_cur) {
6856 		ct.cc_sysid = sysid;
6857 		ct.cc_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
6858 		ct.cc_caller_id = nfs4_srv_caller_id;
6859 		ct.cc_flags = CC_DONTBLOCK;
6860 		err = VOP_OPEN(&cs->vp, fflags, cs->cr, &ct);
6861 		if (err) {
6862 			rfs4_dbe_unlock(fp->rf_dbe);
6863 			if (share_a || share_d)
6864 				(void) rfs4_unshare(sp);
6865 			rfs4_dbe_unlock(sp->rs_dbe);
6866 			rfs4_file_rele(fp);
6867 
6868 			/* Not a fully formed open; "close" it */
6869 			if (screate == TRUE)
6870 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6871 			rfs4_state_rele(sp);
6872 			/* check if a monitor detected a delegation conflict */
6873 			if (err == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
6874 				resp->status = NFS4ERR_DELAY;
6875 			else
6876 				resp->status = NFS4ERR_SERVERFAULT;
6877 			return;
6878 		}
6879 	} else { /* open upgrade */
6880 		/*
6881 		 * calculate the fflags for the new mode that is being added
6882 		 * by this upgrade.
6883 		 */
6884 		fflags = 0;
6885 		if (open_a & OPEN4_SHARE_ACCESS_READ)
6886 			fflags |= FREAD;
6887 		if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6888 			fflags |= FWRITE;
6889 		vn_open_upgrade(cs->vp, fflags);
6890 	}
6891 	sp->rs_open_access |= access;
6892 	sp->rs_open_deny |= deny;
6893 
6894 	if (open_d & OPEN4_SHARE_DENY_READ)
6895 		fp->rf_deny_read++;
6896 	if (open_d & OPEN4_SHARE_DENY_WRITE)
6897 		fp->rf_deny_write++;
6898 	fp->rf_share_deny |= deny;
6899 
6900 	if (open_a & OPEN4_SHARE_ACCESS_READ)
6901 		fp->rf_access_read++;
6902 	if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6903 		fp->rf_access_write++;
6904 	fp->rf_share_access |= access;
6905 
6906 	/*
6907 	 * Check for delegation here. if the deleg argument is not
6908 	 * DELEG_ANY, then this is a reclaim from a client and
6909 	 * we must honor the delegation requested. If necessary we can
6910 	 * set the recall flag.
6911 	 */
6912 
6913 	dsp = rfs4_grant_delegation(deleg, sp, &recall);
6914 
6915 	cs->deleg = (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE);
6916 
6917 	next_stateid(&sp->rs_stateid);
6918 
6919 	resp->stateid = sp->rs_stateid.stateid;
6920 
6921 	rfs4_dbe_unlock(fp->rf_dbe);
6922 	rfs4_dbe_unlock(sp->rs_dbe);
6923 
6924 	if (dsp) {
6925 		rfs4_set_deleg_response(dsp, &resp->delegation, NULL, recall);
6926 		rfs4_deleg_state_rele(dsp);
6927 	}
6928 
6929 	rfs4_file_rele(fp);
6930 	rfs4_state_rele(sp);
6931 
6932 	resp->status = NFS4_OK;
6933 }
6934 
6935 /*ARGSUSED*/
6936 static void
6937 rfs4_do_opennull(struct compound_state *cs, struct svc_req *req,
6938     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6939 {
6940 	change_info4 *cinfo = &resp->cinfo;
6941 	bitmap4 *attrset = &resp->attrset;
6942 
6943 	if (args->opentype == OPEN4_NOCREATE)
6944 		resp->status = rfs4_lookupfile(&args->open_claim4_u.file,
6945 		    req, cs, args->share_access, cinfo);
6946 	else {
6947 		/* inhibit delegation grants during exclusive create */
6948 
6949 		if (args->mode == EXCLUSIVE4)
6950 			rfs4_disable_delegation();
6951 
6952 		resp->status = rfs4_createfile(args, req, cs, cinfo, attrset,
6953 		    oo->ro_client->rc_clientid);
6954 	}
6955 
6956 	if (resp->status == NFS4_OK) {
6957 
6958 		/* cs->vp cs->fh now reference the desired file */
6959 
6960 		rfs4_do_open(cs, req, oo,
6961 		    oo->ro_need_confirm ? DELEG_NONE : DELEG_ANY,
6962 		    args->share_access, args->share_deny, resp, 0);
6963 
6964 		/*
6965 		 * If rfs4_createfile set attrset, we must
6966 		 * clear this attrset before the response is copied.
6967 		 */
6968 		if (resp->status != NFS4_OK && resp->attrset) {
6969 			resp->attrset = 0;
6970 		}
6971 	}
6972 	else
6973 		*cs->statusp = resp->status;
6974 
6975 	if (args->mode == EXCLUSIVE4)
6976 		rfs4_enable_delegation();
6977 }
6978 
6979 /*ARGSUSED*/
6980 static void
6981 rfs4_do_openprev(struct compound_state *cs, struct svc_req *req,
6982     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6983 {
6984 	change_info4 *cinfo = &resp->cinfo;
6985 	vattr_t va;
6986 	vtype_t v_type = cs->vp->v_type;
6987 	int error = 0;
6988 
6989 	/* Verify that we have a regular file */
6990 	if (v_type != VREG) {
6991 		if (v_type == VDIR)
6992 			resp->status = NFS4ERR_ISDIR;
6993 		else if (v_type == VLNK)
6994 			resp->status = NFS4ERR_SYMLINK;
6995 		else
6996 			resp->status = NFS4ERR_INVAL;
6997 		return;
6998 	}
6999 
7000 	va.va_mask = AT_MODE|AT_UID;
7001 	error = VOP_GETATTR(cs->vp, &va, 0, cs->cr, NULL);
7002 	if (error) {
7003 		resp->status = puterrno4(error);
7004 		return;
7005 	}
7006 
7007 	cs->mandlock = MANDLOCK(cs->vp, va.va_mode);
7008 
7009 	/*
7010 	 * Check if we have access to the file, Note the the file
7011 	 * could have originally been open UNCHECKED or GUARDED
7012 	 * with mode bits that will now fail, but there is nothing
7013 	 * we can really do about that except in the case that the
7014 	 * owner of the file is the one requesting the open.
7015 	 */
7016 	if (crgetuid(cs->cr) != va.va_uid) {
7017 		resp->status = check_open_access(args->share_access, cs, req);
7018 		if (resp->status != NFS4_OK) {
7019 			return;
7020 		}
7021 	}
7022 
7023 	/*
7024 	 * cinfo on a CLAIM_PREVIOUS is undefined, initialize to zero
7025 	 */
7026 	cinfo->before = 0;
7027 	cinfo->after = 0;
7028 	cinfo->atomic = FALSE;
7029 
7030 	rfs4_do_open(cs, req, oo,
7031 	    NFS4_DELEG4TYPE2REQTYPE(args->open_claim4_u.delegate_type),
7032 	    args->share_access, args->share_deny, resp, 0);
7033 }
7034 
7035 static void
7036 rfs4_do_opendelcur(struct compound_state *cs, struct svc_req *req,
7037     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7038 {
7039 	int error;
7040 	nfsstat4 status;
7041 	stateid4 stateid =
7042 	    args->open_claim4_u.delegate_cur_info.delegate_stateid;
7043 	rfs4_deleg_state_t *dsp;
7044 
7045 	/*
7046 	 * Find the state info from the stateid and confirm that the
7047 	 * file is delegated.  If the state openowner is the same as
7048 	 * the supplied openowner we're done. If not, get the file
7049 	 * info from the found state info. Use that file info to
7050 	 * create the state for this lock owner. Note solaris doen't
7051 	 * really need the pathname to find the file. We may want to
7052 	 * lookup the pathname and make sure that the vp exist and
7053 	 * matches the vp in the file structure. However it is
7054 	 * possible that the pathname nolonger exists (local process
7055 	 * unlinks the file), so this may not be that useful.
7056 	 */
7057 
7058 	status = rfs4_get_deleg_state(&stateid, &dsp);
7059 	if (status != NFS4_OK) {
7060 		resp->status = status;
7061 		return;
7062 	}
7063 
7064 	ASSERT(dsp->rds_finfo->rf_dinfo.rd_dtype != OPEN_DELEGATE_NONE);
7065 
7066 	/*
7067 	 * New lock owner, create state. Since this was probably called
7068 	 * in response to a CB_RECALL we set deleg to DELEG_NONE
7069 	 */
7070 
7071 	ASSERT(cs->vp != NULL);
7072 	VN_RELE(cs->vp);
7073 	VN_HOLD(dsp->rds_finfo->rf_vp);
7074 	cs->vp = dsp->rds_finfo->rf_vp;
7075 
7076 	if (error = makefh4(&cs->fh, cs->vp, cs->exi)) {
7077 		rfs4_deleg_state_rele(dsp);
7078 		*cs->statusp = resp->status = puterrno4(error);
7079 		return;
7080 	}
7081 
7082 	/* Mark progress for delegation returns */
7083 	dsp->rds_finfo->rf_dinfo.rd_time_lastwrite = gethrestime_sec();
7084 	rfs4_deleg_state_rele(dsp);
7085 	rfs4_do_open(cs, req, oo, DELEG_NONE,
7086 	    args->share_access, args->share_deny, resp, 1);
7087 }
7088 
7089 /*ARGSUSED*/
7090 static void
7091 rfs4_do_opendelprev(struct compound_state *cs, struct svc_req *req,
7092     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7093 {
7094 	/*
7095 	 * Lookup the pathname, it must already exist since this file
7096 	 * was delegated.
7097 	 *
7098 	 * Find the file and state info for this vp and open owner pair.
7099 	 *	check that they are in fact delegated.
7100 	 *	check that the state access and deny modes are the same.
7101 	 *
7102 	 * Return the delgation possibly seting the recall flag.
7103 	 */
7104 	rfs4_file_t *fp;
7105 	rfs4_state_t *sp;
7106 	bool_t create = FALSE;
7107 	bool_t dcreate = FALSE;
7108 	rfs4_deleg_state_t *dsp;
7109 	nfsace4 *ace;
7110 
7111 	/* Note we ignore oflags */
7112 	resp->status = rfs4_lookupfile(&args->open_claim4_u.file_delegate_prev,
7113 	    req, cs, args->share_access, &resp->cinfo);
7114 
7115 	if (resp->status != NFS4_OK) {
7116 		return;
7117 	}
7118 
7119 	/* get the file struct and hold a lock on it during initial open */
7120 	fp = rfs4_findfile_withlock(cs->vp, NULL, &create);
7121 	if (fp == NULL) {
7122 		resp->status = NFS4ERR_RESOURCE;
7123 		DTRACE_PROBE1(nfss__e__do_opendelprev1, nfsstat4, resp->status);
7124 		return;
7125 	}
7126 
7127 	sp = rfs4_findstate_by_owner_file(oo, fp, &create);
7128 	if (sp == NULL) {
7129 		resp->status = NFS4ERR_SERVERFAULT;
7130 		DTRACE_PROBE1(nfss__e__do_opendelprev2, nfsstat4, resp->status);
7131 		rw_exit(&fp->rf_file_rwlock);
7132 		rfs4_file_rele(fp);
7133 		return;
7134 	}
7135 
7136 	rfs4_dbe_lock(sp->rs_dbe);
7137 	rfs4_dbe_lock(fp->rf_dbe);
7138 	if (args->share_access != sp->rs_share_access ||
7139 	    args->share_deny != sp->rs_share_deny ||
7140 	    sp->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
7141 		NFS4_DEBUG(rfs4_debug,
7142 		    (CE_NOTE, "rfs4_do_opendelprev: state mixup"));
7143 		rfs4_dbe_unlock(fp->rf_dbe);
7144 		rfs4_dbe_unlock(sp->rs_dbe);
7145 		rfs4_file_rele(fp);
7146 		rfs4_state_rele(sp);
7147 		resp->status = NFS4ERR_SERVERFAULT;
7148 		return;
7149 	}
7150 	rfs4_dbe_unlock(fp->rf_dbe);
7151 	rfs4_dbe_unlock(sp->rs_dbe);
7152 
7153 	dsp = rfs4_finddeleg(sp, &dcreate);
7154 	if (dsp == NULL) {
7155 		rfs4_state_rele(sp);
7156 		rfs4_file_rele(fp);
7157 		resp->status = NFS4ERR_SERVERFAULT;
7158 		return;
7159 	}
7160 
7161 	next_stateid(&sp->rs_stateid);
7162 
7163 	resp->stateid = sp->rs_stateid.stateid;
7164 
7165 	resp->delegation.delegation_type = dsp->rds_dtype;
7166 
7167 	if (dsp->rds_dtype == OPEN_DELEGATE_READ) {
7168 		open_read_delegation4 *rv =
7169 		    &resp->delegation.open_delegation4_u.read;
7170 
7171 		rv->stateid = dsp->rds_delegid.stateid;
7172 		rv->recall = FALSE; /* no policy in place to set to TRUE */
7173 		ace = &rv->permissions;
7174 	} else {
7175 		open_write_delegation4 *rv =
7176 		    &resp->delegation.open_delegation4_u.write;
7177 
7178 		rv->stateid = dsp->rds_delegid.stateid;
7179 		rv->recall = FALSE;  /* no policy in place to set to TRUE */
7180 		ace = &rv->permissions;
7181 		rv->space_limit.limitby = NFS_LIMIT_SIZE;
7182 		rv->space_limit.nfs_space_limit4_u.filesize = UINT64_MAX;
7183 	}
7184 
7185 	/* XXX For now */
7186 	ace->type = ACE4_ACCESS_ALLOWED_ACE_TYPE;
7187 	ace->flag = 0;
7188 	ace->access_mask = 0;
7189 	ace->who.utf8string_len = 0;
7190 	ace->who.utf8string_val = 0;
7191 
7192 	rfs4_deleg_state_rele(dsp);
7193 	rfs4_state_rele(sp);
7194 	rfs4_file_rele(fp);
7195 }
7196 
7197 typedef enum {
7198 	NFS4_CHKSEQ_OKAY = 0,
7199 	NFS4_CHKSEQ_REPLAY = 1,
7200 	NFS4_CHKSEQ_BAD = 2
7201 } rfs4_chkseq_t;
7202 
7203 /*
7204  * Generic function for sequence number checks.
7205  */
7206 static rfs4_chkseq_t
7207 rfs4_check_seqid(seqid4 seqid, nfs_resop4 *lastop,
7208     seqid4 rqst_seq, nfs_resop4 *resop, bool_t copyres)
7209 {
7210 	/* Same sequence ids and matching operations? */
7211 	if (seqid == rqst_seq && resop->resop == lastop->resop) {
7212 		if (copyres == TRUE) {
7213 			rfs4_free_reply(resop);
7214 			rfs4_copy_reply(resop, lastop);
7215 		}
7216 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
7217 		    "Replayed SEQID %d\n", seqid));
7218 		return (NFS4_CHKSEQ_REPLAY);
7219 	}
7220 
7221 	/* If the incoming sequence is not the next expected then it is bad */
7222 	if (rqst_seq != seqid + 1) {
7223 		if (rqst_seq == seqid) {
7224 			NFS4_DEBUG(rfs4_debug,
7225 			    (CE_NOTE, "BAD SEQID: Replayed sequence id "
7226 			    "but last op was %d current op is %d\n",
7227 			    lastop->resop, resop->resop));
7228 			return (NFS4_CHKSEQ_BAD);
7229 		}
7230 		NFS4_DEBUG(rfs4_debug,
7231 		    (CE_NOTE, "BAD SEQID: got %u expecting %u\n",
7232 		    rqst_seq, seqid));
7233 		return (NFS4_CHKSEQ_BAD);
7234 	}
7235 
7236 	/* Everything okay -- next expected */
7237 	return (NFS4_CHKSEQ_OKAY);
7238 }
7239 
7240 
7241 static rfs4_chkseq_t
7242 rfs4_check_open_seqid(seqid4 seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7243 {
7244 	rfs4_chkseq_t rc;
7245 
7246 	rfs4_dbe_lock(op->ro_dbe);
7247 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply, seqid, resop,
7248 	    TRUE);
7249 	rfs4_dbe_unlock(op->ro_dbe);
7250 
7251 	if (rc == NFS4_CHKSEQ_OKAY)
7252 		rfs4_update_lease(op->ro_client);
7253 
7254 	return (rc);
7255 }
7256 
7257 static rfs4_chkseq_t
7258 rfs4_check_olo_seqid(seqid4 olo_seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7259 {
7260 	rfs4_chkseq_t rc;
7261 
7262 	rfs4_dbe_lock(op->ro_dbe);
7263 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply,
7264 	    olo_seqid, resop, FALSE);
7265 	rfs4_dbe_unlock(op->ro_dbe);
7266 
7267 	return (rc);
7268 }
7269 
7270 static rfs4_chkseq_t
7271 rfs4_check_lock_seqid(seqid4 seqid, rfs4_lo_state_t *lsp, nfs_resop4 *resop)
7272 {
7273 	rfs4_chkseq_t rc = NFS4_CHKSEQ_OKAY;
7274 
7275 	rfs4_dbe_lock(lsp->rls_dbe);
7276 	if (!lsp->rls_skip_seqid_check)
7277 		rc = rfs4_check_seqid(lsp->rls_seqid, &lsp->rls_reply, seqid,
7278 		    resop, TRUE);
7279 	rfs4_dbe_unlock(lsp->rls_dbe);
7280 
7281 	return (rc);
7282 }
7283 
7284 static void
7285 rfs4_op_open(nfs_argop4 *argop, nfs_resop4 *resop,
7286     struct svc_req *req, struct compound_state *cs)
7287 {
7288 	OPEN4args *args = &argop->nfs_argop4_u.opopen;
7289 	OPEN4res *resp = &resop->nfs_resop4_u.opopen;
7290 	open_owner4 *owner = &args->owner;
7291 	open_claim_type4 claim = args->claim;
7292 	rfs4_client_t *cp;
7293 	rfs4_openowner_t *oo;
7294 	bool_t create;
7295 	bool_t replay = FALSE;
7296 	int can_reclaim;
7297 
7298 	DTRACE_NFSV4_2(op__open__start, struct compound_state *, cs,
7299 	    OPEN4args *, args);
7300 
7301 	if (cs->vp == NULL) {
7302 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7303 		goto end;
7304 	}
7305 
7306 	/*
7307 	 * Need to check clientid and lease expiration first based on
7308 	 * error ordering and incrementing sequence id.
7309 	 */
7310 	cp = rfs4_findclient_by_id(owner->clientid, FALSE);
7311 	if (cp == NULL) {
7312 		*cs->statusp = resp->status =
7313 		    rfs4_check_clientid(&owner->clientid, 0);
7314 		goto end;
7315 	}
7316 
7317 	if (rfs4_lease_expired(cp)) {
7318 		rfs4_client_close(cp);
7319 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7320 		goto end;
7321 	}
7322 	can_reclaim = cp->rc_can_reclaim;
7323 
7324 	/*
7325 	 * Find the open_owner for use from this point forward.  Take
7326 	 * care in updating the sequence id based on the type of error
7327 	 * being returned.
7328 	 */
7329 retry:
7330 	create = TRUE;
7331 	oo = rfs4_findopenowner(owner, &create, args->seqid);
7332 	if (oo == NULL) {
7333 		*cs->statusp = resp->status = NFS4ERR_RESOURCE;
7334 		rfs4_client_rele(cp);
7335 		goto end;
7336 	}
7337 
7338 	/* Hold off access to the sequence space while the open is done */
7339 	rfs4_sw_enter(&oo->ro_sw);
7340 
7341 	/*
7342 	 * If the open_owner existed before at the server, then check
7343 	 * the sequence id.
7344 	 */
7345 	if (!create && !oo->ro_postpone_confirm) {
7346 		switch (rfs4_check_open_seqid(args->seqid, oo, resop)) {
7347 		case NFS4_CHKSEQ_BAD:
7348 			if ((args->seqid > oo->ro_open_seqid) &&
7349 			    oo->ro_need_confirm) {
7350 				rfs4_free_opens(oo, TRUE, FALSE);
7351 				rfs4_sw_exit(&oo->ro_sw);
7352 				rfs4_openowner_rele(oo);
7353 				goto retry;
7354 			}
7355 			resp->status = NFS4ERR_BAD_SEQID;
7356 			goto out;
7357 		case NFS4_CHKSEQ_REPLAY: /* replay of previous request */
7358 			replay = TRUE;
7359 			goto out;
7360 		default:
7361 			break;
7362 		}
7363 
7364 		/*
7365 		 * Sequence was ok and open owner exists
7366 		 * check to see if we have yet to see an
7367 		 * open_confirm.
7368 		 */
7369 		if (oo->ro_need_confirm) {
7370 			rfs4_free_opens(oo, TRUE, FALSE);
7371 			rfs4_sw_exit(&oo->ro_sw);
7372 			rfs4_openowner_rele(oo);
7373 			goto retry;
7374 		}
7375 	}
7376 	/* Grace only applies to regular-type OPENs */
7377 	if (rfs4_clnt_in_grace(cp) &&
7378 	    (claim == CLAIM_NULL || claim == CLAIM_DELEGATE_CUR)) {
7379 		*cs->statusp = resp->status = NFS4ERR_GRACE;
7380 		goto out;
7381 	}
7382 
7383 	/*
7384 	 * If previous state at the server existed then can_reclaim
7385 	 * will be set. If not reply NFS4ERR_NO_GRACE to the
7386 	 * client.
7387 	 */
7388 	if (rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS && !can_reclaim) {
7389 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7390 		goto out;
7391 	}
7392 
7393 
7394 	/*
7395 	 * Reject the open if the client has missed the grace period
7396 	 */
7397 	if (!rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS) {
7398 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7399 		goto out;
7400 	}
7401 
7402 	/* Couple of up-front bookkeeping items */
7403 	if (oo->ro_need_confirm) {
7404 		/*
7405 		 * If this is a reclaim OPEN then we should not ask
7406 		 * for a confirmation of the open_owner per the
7407 		 * protocol specification.
7408 		 */
7409 		if (claim == CLAIM_PREVIOUS)
7410 			oo->ro_need_confirm = FALSE;
7411 		else
7412 			resp->rflags |= OPEN4_RESULT_CONFIRM;
7413 	}
7414 	resp->rflags |= OPEN4_RESULT_LOCKTYPE_POSIX;
7415 
7416 	/*
7417 	 * If there is an unshared filesystem mounted on this vnode,
7418 	 * do not allow to open/create in this directory.
7419 	 */
7420 	if (vn_ismntpt(cs->vp)) {
7421 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
7422 		goto out;
7423 	}
7424 
7425 	/*
7426 	 * access must READ, WRITE, or BOTH.  No access is invalid.
7427 	 * deny can be READ, WRITE, BOTH, or NONE.
7428 	 * bits not defined for access/deny are invalid.
7429 	 */
7430 	if (! (args->share_access & OPEN4_SHARE_ACCESS_BOTH) ||
7431 	    (args->share_access & ~OPEN4_SHARE_ACCESS_BOTH) ||
7432 	    (args->share_deny & ~OPEN4_SHARE_DENY_BOTH)) {
7433 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7434 		goto out;
7435 	}
7436 
7437 
7438 	/*
7439 	 * make sure attrset is zero before response is built.
7440 	 */
7441 	resp->attrset = 0;
7442 
7443 	switch (claim) {
7444 	case CLAIM_NULL:
7445 		rfs4_do_opennull(cs, req, args, oo, resp);
7446 		break;
7447 	case CLAIM_PREVIOUS:
7448 		rfs4_do_openprev(cs, req, args, oo, resp);
7449 		break;
7450 	case CLAIM_DELEGATE_CUR:
7451 		rfs4_do_opendelcur(cs, req, args, oo, resp);
7452 		break;
7453 	case CLAIM_DELEGATE_PREV:
7454 		rfs4_do_opendelprev(cs, req, args, oo, resp);
7455 		break;
7456 	default:
7457 		resp->status = NFS4ERR_INVAL;
7458 		break;
7459 	}
7460 
7461 out:
7462 	rfs4_client_rele(cp);
7463 
7464 	/* Catch sequence id handling here to make it a little easier */
7465 	switch (resp->status) {
7466 	case NFS4ERR_BADXDR:
7467 	case NFS4ERR_BAD_SEQID:
7468 	case NFS4ERR_BAD_STATEID:
7469 	case NFS4ERR_NOFILEHANDLE:
7470 	case NFS4ERR_RESOURCE:
7471 	case NFS4ERR_STALE_CLIENTID:
7472 	case NFS4ERR_STALE_STATEID:
7473 		/*
7474 		 * The protocol states that if any of these errors are
7475 		 * being returned, the sequence id should not be
7476 		 * incremented.  Any other return requires an
7477 		 * increment.
7478 		 */
7479 		break;
7480 	default:
7481 		/* Always update the lease in this case */
7482 		rfs4_update_lease(oo->ro_client);
7483 
7484 		/* Regular response - copy the result */
7485 		if (!replay)
7486 			rfs4_update_open_resp(oo, resop, &cs->fh);
7487 
7488 		/*
7489 		 * REPLAY case: Only if the previous response was OK
7490 		 * do we copy the filehandle.  If not OK, no
7491 		 * filehandle to copy.
7492 		 */
7493 		if (replay == TRUE &&
7494 		    resp->status == NFS4_OK &&
7495 		    oo->ro_reply_fh.nfs_fh4_val) {
7496 			/*
7497 			 * If this is a replay, we must restore the
7498 			 * current filehandle/vp to that of what was
7499 			 * returned originally.  Try our best to do
7500 			 * it.
7501 			 */
7502 			nfs_fh4_fmt_t *fh_fmtp =
7503 			    (nfs_fh4_fmt_t *)oo->ro_reply_fh.nfs_fh4_val;
7504 
7505 			cs->exi = checkexport4(&fh_fmtp->fh4_fsid,
7506 			    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
7507 
7508 			if (cs->exi == NULL) {
7509 				resp->status = NFS4ERR_STALE;
7510 				goto finish;
7511 			}
7512 
7513 			VN_RELE(cs->vp);
7514 
7515 			cs->vp = nfs4_fhtovp(&oo->ro_reply_fh, cs->exi,
7516 			    &resp->status);
7517 
7518 			if (cs->vp == NULL)
7519 				goto finish;
7520 
7521 			nfs_fh4_copy(&oo->ro_reply_fh, &cs->fh);
7522 		}
7523 
7524 		/*
7525 		 * If this was a replay, no need to update the
7526 		 * sequence id. If the open_owner was not created on
7527 		 * this pass, then update.  The first use of an
7528 		 * open_owner will not bump the sequence id.
7529 		 */
7530 		if (replay == FALSE && !create)
7531 			rfs4_update_open_sequence(oo);
7532 		/*
7533 		 * If the client is receiving an error and the
7534 		 * open_owner needs to be confirmed, there is no way
7535 		 * to notify the client of this fact ignoring the fact
7536 		 * that the server has no method of returning a
7537 		 * stateid to confirm.  Therefore, the server needs to
7538 		 * mark this open_owner in a way as to avoid the
7539 		 * sequence id checking the next time the client uses
7540 		 * this open_owner.
7541 		 */
7542 		if (resp->status != NFS4_OK && oo->ro_need_confirm)
7543 			oo->ro_postpone_confirm = TRUE;
7544 		/*
7545 		 * If OK response then clear the postpone flag and
7546 		 * reset the sequence id to keep in sync with the
7547 		 * client.
7548 		 */
7549 		if (resp->status == NFS4_OK && oo->ro_postpone_confirm) {
7550 			oo->ro_postpone_confirm = FALSE;
7551 			oo->ro_open_seqid = args->seqid;
7552 		}
7553 		break;
7554 	}
7555 
7556 finish:
7557 	*cs->statusp = resp->status;
7558 
7559 	rfs4_sw_exit(&oo->ro_sw);
7560 	rfs4_openowner_rele(oo);
7561 
7562 end:
7563 	DTRACE_NFSV4_2(op__open__done, struct compound_state *, cs,
7564 	    OPEN4res *, resp);
7565 }
7566 
7567 /*ARGSUSED*/
7568 void
7569 rfs4_op_open_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
7570     struct svc_req *req, struct compound_state *cs)
7571 {
7572 	OPEN_CONFIRM4args *args = &argop->nfs_argop4_u.opopen_confirm;
7573 	OPEN_CONFIRM4res *resp = &resop->nfs_resop4_u.opopen_confirm;
7574 	rfs4_state_t *sp;
7575 	nfsstat4 status;
7576 
7577 	DTRACE_NFSV4_2(op__open__confirm__start, struct compound_state *, cs,
7578 	    OPEN_CONFIRM4args *, args);
7579 
7580 	if (cs->vp == NULL) {
7581 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7582 		goto out;
7583 	}
7584 
7585 	if (cs->vp->v_type != VREG) {
7586 		*cs->statusp = resp->status =
7587 		    cs->vp->v_type == VDIR ? NFS4ERR_ISDIR : NFS4ERR_INVAL;
7588 		return;
7589 	}
7590 
7591 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7592 	if (status != NFS4_OK) {
7593 		*cs->statusp = resp->status = status;
7594 		goto out;
7595 	}
7596 
7597 	/* Ensure specified filehandle matches */
7598 	if (cs->vp != sp->rs_finfo->rf_vp) {
7599 		rfs4_state_rele(sp);
7600 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7601 		goto out;
7602 	}
7603 
7604 	/* hold off other access to open_owner while we tinker */
7605 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7606 
7607 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7608 	case NFS4_CHECK_STATEID_OKAY:
7609 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7610 		    resop) != 0) {
7611 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7612 			break;
7613 		}
7614 		/*
7615 		 * If it is the appropriate stateid and determined to
7616 		 * be "OKAY" then this means that the stateid does not
7617 		 * need to be confirmed and the client is in error for
7618 		 * sending an OPEN_CONFIRM.
7619 		 */
7620 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7621 		break;
7622 	case NFS4_CHECK_STATEID_OLD:
7623 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7624 		break;
7625 	case NFS4_CHECK_STATEID_BAD:
7626 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7627 		break;
7628 	case NFS4_CHECK_STATEID_EXPIRED:
7629 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7630 		break;
7631 	case NFS4_CHECK_STATEID_CLOSED:
7632 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7633 		break;
7634 	case NFS4_CHECK_STATEID_REPLAY:
7635 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7636 		    resop)) {
7637 		case NFS4_CHKSEQ_OKAY:
7638 			/*
7639 			 * This is replayed stateid; if seqid matches
7640 			 * next expected, then client is using wrong seqid.
7641 			 */
7642 			/* fall through */
7643 		case NFS4_CHKSEQ_BAD:
7644 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7645 			break;
7646 		case NFS4_CHKSEQ_REPLAY:
7647 			/*
7648 			 * Note this case is the duplicate case so
7649 			 * resp->status is already set.
7650 			 */
7651 			*cs->statusp = resp->status;
7652 			rfs4_update_lease(sp->rs_owner->ro_client);
7653 			break;
7654 		}
7655 		break;
7656 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7657 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7658 		    resop) != NFS4_CHKSEQ_OKAY) {
7659 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7660 			break;
7661 		}
7662 		*cs->statusp = resp->status = NFS4_OK;
7663 
7664 		next_stateid(&sp->rs_stateid);
7665 		resp->open_stateid = sp->rs_stateid.stateid;
7666 		sp->rs_owner->ro_need_confirm = FALSE;
7667 		rfs4_update_lease(sp->rs_owner->ro_client);
7668 		rfs4_update_open_sequence(sp->rs_owner);
7669 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7670 		break;
7671 	default:
7672 		ASSERT(FALSE);
7673 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7674 		break;
7675 	}
7676 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7677 	rfs4_state_rele(sp);
7678 
7679 out:
7680 	DTRACE_NFSV4_2(op__open__confirm__done, struct compound_state *, cs,
7681 	    OPEN_CONFIRM4res *, resp);
7682 }
7683 
7684 /*ARGSUSED*/
7685 void
7686 rfs4_op_open_downgrade(nfs_argop4 *argop, nfs_resop4 *resop,
7687     struct svc_req *req, struct compound_state *cs)
7688 {
7689 	OPEN_DOWNGRADE4args *args = &argop->nfs_argop4_u.opopen_downgrade;
7690 	OPEN_DOWNGRADE4res *resp = &resop->nfs_resop4_u.opopen_downgrade;
7691 	uint32_t access = args->share_access;
7692 	uint32_t deny = args->share_deny;
7693 	nfsstat4 status;
7694 	rfs4_state_t *sp;
7695 	rfs4_file_t *fp;
7696 	int fflags = 0;
7697 
7698 	DTRACE_NFSV4_2(op__open__downgrade__start, struct compound_state *, cs,
7699 	    OPEN_DOWNGRADE4args *, args);
7700 
7701 	if (cs->vp == NULL) {
7702 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7703 		goto out;
7704 	}
7705 
7706 	if (cs->vp->v_type != VREG) {
7707 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7708 		return;
7709 	}
7710 
7711 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7712 	if (status != NFS4_OK) {
7713 		*cs->statusp = resp->status = status;
7714 		goto out;
7715 	}
7716 
7717 	/* Ensure specified filehandle matches */
7718 	if (cs->vp != sp->rs_finfo->rf_vp) {
7719 		rfs4_state_rele(sp);
7720 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7721 		goto out;
7722 	}
7723 
7724 	/* hold off other access to open_owner while we tinker */
7725 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7726 
7727 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7728 	case NFS4_CHECK_STATEID_OKAY:
7729 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7730 		    resop) != NFS4_CHKSEQ_OKAY) {
7731 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7732 			goto end;
7733 		}
7734 		break;
7735 	case NFS4_CHECK_STATEID_OLD:
7736 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7737 		goto end;
7738 	case NFS4_CHECK_STATEID_BAD:
7739 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7740 		goto end;
7741 	case NFS4_CHECK_STATEID_EXPIRED:
7742 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7743 		goto end;
7744 	case NFS4_CHECK_STATEID_CLOSED:
7745 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7746 		goto end;
7747 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7748 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7749 		goto end;
7750 	case NFS4_CHECK_STATEID_REPLAY:
7751 		/* Check the sequence id for the open owner */
7752 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7753 		    resop)) {
7754 		case NFS4_CHKSEQ_OKAY:
7755 			/*
7756 			 * This is replayed stateid; if seqid matches
7757 			 * next expected, then client is using wrong seqid.
7758 			 */
7759 			/* fall through */
7760 		case NFS4_CHKSEQ_BAD:
7761 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7762 			goto end;
7763 		case NFS4_CHKSEQ_REPLAY:
7764 			/*
7765 			 * Note this case is the duplicate case so
7766 			 * resp->status is already set.
7767 			 */
7768 			*cs->statusp = resp->status;
7769 			rfs4_update_lease(sp->rs_owner->ro_client);
7770 			goto end;
7771 		}
7772 		break;
7773 	default:
7774 		ASSERT(FALSE);
7775 		break;
7776 	}
7777 
7778 	rfs4_dbe_lock(sp->rs_dbe);
7779 	/*
7780 	 * Check that the new access modes and deny modes are valid.
7781 	 * Check that no invalid bits are set.
7782 	 */
7783 	if ((access & ~(OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) ||
7784 	    (deny & ~(OPEN4_SHARE_DENY_READ | OPEN4_SHARE_DENY_WRITE))) {
7785 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7786 		rfs4_update_open_sequence(sp->rs_owner);
7787 		rfs4_dbe_unlock(sp->rs_dbe);
7788 		goto end;
7789 	}
7790 
7791 	/*
7792 	 * The new modes must be a subset of the current modes and
7793 	 * the access must specify at least one mode. To test that
7794 	 * the new mode is a subset of the current modes we bitwise
7795 	 * AND them together and check that the result equals the new
7796 	 * mode. For example:
7797 	 * New mode, access == R and current mode, sp->rs_open_access  == RW
7798 	 * access & sp->rs_open_access == R == access, so the new access mode
7799 	 * is valid. Consider access == RW, sp->rs_open_access = R
7800 	 * access & sp->rs_open_access == R != access, so the new access mode
7801 	 * is invalid.
7802 	 */
7803 	if ((access & sp->rs_open_access) != access ||
7804 	    (deny & sp->rs_open_deny) != deny ||
7805 	    (access &
7806 	    (OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) == 0) {
7807 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7808 		rfs4_update_open_sequence(sp->rs_owner);
7809 		rfs4_dbe_unlock(sp->rs_dbe);
7810 		goto end;
7811 	}
7812 
7813 	/*
7814 	 * Release any share locks associated with this stateID.
7815 	 * Strictly speaking, this violates the spec because the
7816 	 * spec effectively requires that open downgrade be atomic.
7817 	 * At present, fs_shrlock does not have this capability.
7818 	 */
7819 	(void) rfs4_unshare(sp);
7820 
7821 	status = rfs4_share(sp, access, deny);
7822 	if (status != NFS4_OK) {
7823 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7824 		rfs4_update_open_sequence(sp->rs_owner);
7825 		rfs4_dbe_unlock(sp->rs_dbe);
7826 		goto end;
7827 	}
7828 
7829 	fp = sp->rs_finfo;
7830 	rfs4_dbe_lock(fp->rf_dbe);
7831 
7832 	/*
7833 	 * If the current mode has deny read and the new mode
7834 	 * does not, decrement the number of deny read mode bits
7835 	 * and if it goes to zero turn off the deny read bit
7836 	 * on the file.
7837 	 */
7838 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_READ) &&
7839 	    (deny & OPEN4_SHARE_DENY_READ) == 0) {
7840 		fp->rf_deny_read--;
7841 		if (fp->rf_deny_read == 0)
7842 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
7843 	}
7844 
7845 	/*
7846 	 * If the current mode has deny write and the new mode
7847 	 * does not, decrement the number of deny write mode bits
7848 	 * and if it goes to zero turn off the deny write bit
7849 	 * on the file.
7850 	 */
7851 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) &&
7852 	    (deny & OPEN4_SHARE_DENY_WRITE) == 0) {
7853 		fp->rf_deny_write--;
7854 		if (fp->rf_deny_write == 0)
7855 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
7856 	}
7857 
7858 	/*
7859 	 * If the current mode has access read and the new mode
7860 	 * does not, decrement the number of access read mode bits
7861 	 * and if it goes to zero turn off the access read bit
7862 	 * on the file.  set fflags to FREAD for the call to
7863 	 * vn_open_downgrade().
7864 	 */
7865 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) &&
7866 	    (access & OPEN4_SHARE_ACCESS_READ) == 0) {
7867 		fp->rf_access_read--;
7868 		if (fp->rf_access_read == 0)
7869 			fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
7870 		fflags |= FREAD;
7871 	}
7872 
7873 	/*
7874 	 * If the current mode has access write and the new mode
7875 	 * does not, decrement the number of access write mode bits
7876 	 * and if it goes to zero turn off the access write bit
7877 	 * on the file.  set fflags to FWRITE for the call to
7878 	 * vn_open_downgrade().
7879 	 */
7880 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) &&
7881 	    (access & OPEN4_SHARE_ACCESS_WRITE) == 0) {
7882 		fp->rf_access_write--;
7883 		if (fp->rf_access_write == 0)
7884 			fp->rf_share_deny &= ~OPEN4_SHARE_ACCESS_WRITE;
7885 		fflags |= FWRITE;
7886 	}
7887 
7888 	/* Check that the file is still accessible */
7889 	ASSERT(fp->rf_share_access);
7890 
7891 	rfs4_dbe_unlock(fp->rf_dbe);
7892 
7893 	/* now set the new open access and deny modes */
7894 	sp->rs_open_access = access;
7895 	sp->rs_open_deny = deny;
7896 
7897 	/*
7898 	 * we successfully downgraded the share lock, now we need to downgrade
7899 	 * the open. it is possible that the downgrade was only for a deny
7900 	 * mode and we have nothing else to do.
7901 	 */
7902 	if ((fflags & (FREAD|FWRITE)) != 0)
7903 		vn_open_downgrade(cs->vp, fflags);
7904 
7905 	/* Update the stateid */
7906 	next_stateid(&sp->rs_stateid);
7907 	resp->open_stateid = sp->rs_stateid.stateid;
7908 
7909 	rfs4_dbe_unlock(sp->rs_dbe);
7910 
7911 	*cs->statusp = resp->status = NFS4_OK;
7912 	/* Update the lease */
7913 	rfs4_update_lease(sp->rs_owner->ro_client);
7914 	/* And the sequence */
7915 	rfs4_update_open_sequence(sp->rs_owner);
7916 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7917 
7918 end:
7919 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7920 	rfs4_state_rele(sp);
7921 out:
7922 	DTRACE_NFSV4_2(op__open__downgrade__done, struct compound_state *, cs,
7923 	    OPEN_DOWNGRADE4res *, resp);
7924 }
7925 
7926 static void *
7927 memstr(const void *s1, const char *s2, size_t n)
7928 {
7929 	size_t l = strlen(s2);
7930 	char *p = (char *)s1;
7931 
7932 	while (n >= l) {
7933 		if (bcmp(p, s2, l) == 0)
7934 			return (p);
7935 		p++;
7936 		n--;
7937 	}
7938 
7939 	return (NULL);
7940 }
7941 
7942 /*
7943  * The logic behind this function is detailed in the NFSv4 RFC in the
7944  * SETCLIENTID operation description under IMPLEMENTATION.  Refer to
7945  * that section for explicit guidance to server behavior for
7946  * SETCLIENTID.
7947  */
7948 void
7949 rfs4_op_setclientid(nfs_argop4 *argop, nfs_resop4 *resop,
7950     struct svc_req *req, struct compound_state *cs)
7951 {
7952 	SETCLIENTID4args *args = &argop->nfs_argop4_u.opsetclientid;
7953 	SETCLIENTID4res *res = &resop->nfs_resop4_u.opsetclientid;
7954 	rfs4_client_t *cp, *newcp, *cp_confirmed, *cp_unconfirmed;
7955 	rfs4_clntip_t *ci;
7956 	bool_t create;
7957 	char *addr, *netid;
7958 	int len;
7959 
7960 	DTRACE_NFSV4_2(op__setclientid__start, struct compound_state *, cs,
7961 	    SETCLIENTID4args *, args);
7962 retry:
7963 	newcp = cp_confirmed = cp_unconfirmed = NULL;
7964 
7965 	/*
7966 	 * Save the caller's IP address
7967 	 */
7968 	args->client.cl_addr =
7969 	    (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
7970 
7971 	/*
7972 	 * Record if it is a Solaris client that cannot handle referrals.
7973 	 */
7974 	if (memstr(args->client.id_val, "Solaris", args->client.id_len) &&
7975 	    !memstr(args->client.id_val, "+referrals", args->client.id_len)) {
7976 		/* Add a "yes, it's downrev" record */
7977 		create = TRUE;
7978 		ci = rfs4_find_clntip(args->client.cl_addr, &create);
7979 		ASSERT(ci != NULL);
7980 		rfs4_dbe_rele(ci->ri_dbe);
7981 	} else {
7982 		/* Remove any previous record */
7983 		rfs4_invalidate_clntip(args->client.cl_addr);
7984 	}
7985 
7986 	/*
7987 	 * In search of an EXISTING client matching the incoming
7988 	 * request to establish a new client identifier at the server
7989 	 */
7990 	create = TRUE;
7991 	cp = rfs4_findclient(&args->client, &create, NULL);
7992 
7993 	/* Should never happen */
7994 	ASSERT(cp != NULL);
7995 
7996 	if (cp == NULL) {
7997 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
7998 		goto out;
7999 	}
8000 
8001 	/*
8002 	 * Easiest case. Client identifier is newly created and is
8003 	 * unconfirmed.  Also note that for this case, no other
8004 	 * entries exist for the client identifier.  Nothing else to
8005 	 * check.  Just setup the response and respond.
8006 	 */
8007 	if (create) {
8008 		*cs->statusp = res->status = NFS4_OK;
8009 		res->SETCLIENTID4res_u.resok4.clientid = cp->rc_clientid;
8010 		res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8011 		    cp->rc_confirm_verf;
8012 		/* Setup callback information; CB_NULL confirmation later */
8013 		rfs4_client_setcb(cp, &args->callback, args->callback_ident);
8014 
8015 		rfs4_client_rele(cp);
8016 		goto out;
8017 	}
8018 
8019 	/*
8020 	 * An existing, confirmed client may exist but it may not have
8021 	 * been active for at least one lease period.  If so, then
8022 	 * "close" the client and create a new client identifier
8023 	 */
8024 	if (rfs4_lease_expired(cp)) {
8025 		rfs4_client_close(cp);
8026 		goto retry;
8027 	}
8028 
8029 	if (cp->rc_need_confirm == TRUE)
8030 		cp_unconfirmed = cp;
8031 	else
8032 		cp_confirmed = cp;
8033 
8034 	cp = NULL;
8035 
8036 	/*
8037 	 * We have a confirmed client, now check for an
8038 	 * unconfimred entry
8039 	 */
8040 	if (cp_confirmed) {
8041 		/* If creds don't match then client identifier is inuse */
8042 		if (!creds_ok(cp_confirmed->rc_cr_set, req, cs)) {
8043 			rfs4_cbinfo_t *cbp;
8044 			/*
8045 			 * Some one else has established this client
8046 			 * id. Try and say * who they are. We will use
8047 			 * the call back address supplied by * the
8048 			 * first client.
8049 			 */
8050 			*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8051 
8052 			addr = netid = NULL;
8053 
8054 			cbp = &cp_confirmed->rc_cbinfo;
8055 			if (cbp->cb_callback.cb_location.r_addr &&
8056 			    cbp->cb_callback.cb_location.r_netid) {
8057 				cb_client4 *cbcp = &cbp->cb_callback;
8058 
8059 				len = strlen(cbcp->cb_location.r_addr)+1;
8060 				addr = kmem_alloc(len, KM_SLEEP);
8061 				bcopy(cbcp->cb_location.r_addr, addr, len);
8062 				len = strlen(cbcp->cb_location.r_netid)+1;
8063 				netid = kmem_alloc(len, KM_SLEEP);
8064 				bcopy(cbcp->cb_location.r_netid, netid, len);
8065 			}
8066 
8067 			res->SETCLIENTID4res_u.client_using.r_addr = addr;
8068 			res->SETCLIENTID4res_u.client_using.r_netid = netid;
8069 
8070 			rfs4_client_rele(cp_confirmed);
8071 		}
8072 
8073 		/*
8074 		 * Confirmed, creds match, and verifier matches; must
8075 		 * be an update of the callback info
8076 		 */
8077 		if (cp_confirmed->rc_nfs_client.verifier ==
8078 		    args->client.verifier) {
8079 			/* Setup callback information */
8080 			rfs4_client_setcb(cp_confirmed, &args->callback,
8081 			    args->callback_ident);
8082 
8083 			/* everything okay -- move ahead */
8084 			*cs->statusp = res->status = NFS4_OK;
8085 			res->SETCLIENTID4res_u.resok4.clientid =
8086 			    cp_confirmed->rc_clientid;
8087 
8088 			/* update the confirm_verifier and return it */
8089 			rfs4_client_scv_next(cp_confirmed);
8090 			res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8091 			    cp_confirmed->rc_confirm_verf;
8092 
8093 			rfs4_client_rele(cp_confirmed);
8094 			goto out;
8095 		}
8096 
8097 		/*
8098 		 * Creds match but the verifier doesn't.  Must search
8099 		 * for an unconfirmed client that would be replaced by
8100 		 * this request.
8101 		 */
8102 		create = FALSE;
8103 		cp_unconfirmed = rfs4_findclient(&args->client, &create,
8104 		    cp_confirmed);
8105 	}
8106 
8107 	/*
8108 	 * At this point, we have taken care of the brand new client
8109 	 * struct, INUSE case, update of an existing, and confirmed
8110 	 * client struct.
8111 	 */
8112 
8113 	/*
8114 	 * check to see if things have changed while we originally
8115 	 * picked up the client struct.  If they have, then return and
8116 	 * retry the processing of this SETCLIENTID request.
8117 	 */
8118 	if (cp_unconfirmed) {
8119 		rfs4_dbe_lock(cp_unconfirmed->rc_dbe);
8120 		if (!cp_unconfirmed->rc_need_confirm) {
8121 			rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8122 			rfs4_client_rele(cp_unconfirmed);
8123 			if (cp_confirmed)
8124 				rfs4_client_rele(cp_confirmed);
8125 			goto retry;
8126 		}
8127 		/* do away with the old unconfirmed one */
8128 		rfs4_dbe_invalidate(cp_unconfirmed->rc_dbe);
8129 		rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8130 		rfs4_client_rele(cp_unconfirmed);
8131 		cp_unconfirmed = NULL;
8132 	}
8133 
8134 	/*
8135 	 * This search will temporarily hide the confirmed client
8136 	 * struct while a new client struct is created as the
8137 	 * unconfirmed one.
8138 	 */
8139 	create = TRUE;
8140 	newcp = rfs4_findclient(&args->client, &create, cp_confirmed);
8141 
8142 	ASSERT(newcp != NULL);
8143 
8144 	if (newcp == NULL) {
8145 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8146 		rfs4_client_rele(cp_confirmed);
8147 		goto out;
8148 	}
8149 
8150 	/*
8151 	 * If one was not created, then a similar request must be in
8152 	 * process so release and start over with this one
8153 	 */
8154 	if (create != TRUE) {
8155 		rfs4_client_rele(newcp);
8156 		if (cp_confirmed)
8157 			rfs4_client_rele(cp_confirmed);
8158 		goto retry;
8159 	}
8160 
8161 	*cs->statusp = res->status = NFS4_OK;
8162 	res->SETCLIENTID4res_u.resok4.clientid = newcp->rc_clientid;
8163 	res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8164 	    newcp->rc_confirm_verf;
8165 	/* Setup callback information; CB_NULL confirmation later */
8166 	rfs4_client_setcb(newcp, &args->callback, args->callback_ident);
8167 
8168 	newcp->rc_cp_confirmed = cp_confirmed;
8169 
8170 	rfs4_client_rele(newcp);
8171 
8172 out:
8173 	DTRACE_NFSV4_2(op__setclientid__done, struct compound_state *, cs,
8174 	    SETCLIENTID4res *, res);
8175 }
8176 
8177 /*ARGSUSED*/
8178 void
8179 rfs4_op_setclientid_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
8180     struct svc_req *req, struct compound_state *cs)
8181 {
8182 	SETCLIENTID_CONFIRM4args *args =
8183 	    &argop->nfs_argop4_u.opsetclientid_confirm;
8184 	SETCLIENTID_CONFIRM4res *res =
8185 	    &resop->nfs_resop4_u.opsetclientid_confirm;
8186 	rfs4_client_t *cp, *cptoclose = NULL;
8187 
8188 	DTRACE_NFSV4_2(op__setclientid__confirm__start,
8189 	    struct compound_state *, cs,
8190 	    SETCLIENTID_CONFIRM4args *, args);
8191 
8192 	*cs->statusp = res->status = NFS4_OK;
8193 
8194 	cp = rfs4_findclient_by_id(args->clientid, TRUE);
8195 
8196 	if (cp == NULL) {
8197 		*cs->statusp = res->status =
8198 		    rfs4_check_clientid(&args->clientid, 1);
8199 		goto out;
8200 	}
8201 
8202 	if (!creds_ok(cp, req, cs)) {
8203 		*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8204 		rfs4_client_rele(cp);
8205 		goto out;
8206 	}
8207 
8208 	/* If the verifier doesn't match, the record doesn't match */
8209 	if (cp->rc_confirm_verf != args->setclientid_confirm) {
8210 		*cs->statusp = res->status = NFS4ERR_STALE_CLIENTID;
8211 		rfs4_client_rele(cp);
8212 		goto out;
8213 	}
8214 
8215 	rfs4_dbe_lock(cp->rc_dbe);
8216 	cp->rc_need_confirm = FALSE;
8217 	if (cp->rc_cp_confirmed) {
8218 		cptoclose = cp->rc_cp_confirmed;
8219 		cptoclose->rc_ss_remove = 1;
8220 		cp->rc_cp_confirmed = NULL;
8221 	}
8222 
8223 	/*
8224 	 * Update the client's associated server instance, if it's changed
8225 	 * since the client was created.
8226 	 */
8227 	if (rfs4_servinst(cp) != rfs4_cur_servinst)
8228 		rfs4_servinst_assign(cp, rfs4_cur_servinst);
8229 
8230 	/*
8231 	 * Record clientid in stable storage.
8232 	 * Must be done after server instance has been assigned.
8233 	 */
8234 	rfs4_ss_clid(cp);
8235 
8236 	rfs4_dbe_unlock(cp->rc_dbe);
8237 
8238 	if (cptoclose)
8239 		/* don't need to rele, client_close does it */
8240 		rfs4_client_close(cptoclose);
8241 
8242 	/* If needed, initiate CB_NULL call for callback path */
8243 	rfs4_deleg_cb_check(cp);
8244 	rfs4_update_lease(cp);
8245 
8246 	/*
8247 	 * Check to see if client can perform reclaims
8248 	 */
8249 	rfs4_ss_chkclid(cp);
8250 
8251 	rfs4_client_rele(cp);
8252 
8253 out:
8254 	DTRACE_NFSV4_2(op__setclientid__confirm__done,
8255 	    struct compound_state *, cs,
8256 	    SETCLIENTID_CONFIRM4 *, res);
8257 }
8258 
8259 
8260 /*ARGSUSED*/
8261 void
8262 rfs4_op_close(nfs_argop4 *argop, nfs_resop4 *resop,
8263     struct svc_req *req, struct compound_state *cs)
8264 {
8265 	CLOSE4args *args = &argop->nfs_argop4_u.opclose;
8266 	CLOSE4res *resp = &resop->nfs_resop4_u.opclose;
8267 	rfs4_state_t *sp;
8268 	nfsstat4 status;
8269 
8270 	DTRACE_NFSV4_2(op__close__start, struct compound_state *, cs,
8271 	    CLOSE4args *, args);
8272 
8273 	if (cs->vp == NULL) {
8274 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8275 		goto out;
8276 	}
8277 
8278 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_INVALID);
8279 	if (status != NFS4_OK) {
8280 		*cs->statusp = resp->status = status;
8281 		goto out;
8282 	}
8283 
8284 	/* Ensure specified filehandle matches */
8285 	if (cs->vp != sp->rs_finfo->rf_vp) {
8286 		rfs4_state_rele(sp);
8287 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8288 		goto out;
8289 	}
8290 
8291 	/* hold off other access to open_owner while we tinker */
8292 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
8293 
8294 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
8295 	case NFS4_CHECK_STATEID_OKAY:
8296 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8297 		    resop) != NFS4_CHKSEQ_OKAY) {
8298 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8299 			goto end;
8300 		}
8301 		break;
8302 	case NFS4_CHECK_STATEID_OLD:
8303 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8304 		goto end;
8305 	case NFS4_CHECK_STATEID_BAD:
8306 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8307 		goto end;
8308 	case NFS4_CHECK_STATEID_EXPIRED:
8309 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8310 		goto end;
8311 	case NFS4_CHECK_STATEID_CLOSED:
8312 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8313 		goto end;
8314 	case NFS4_CHECK_STATEID_UNCONFIRMED:
8315 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8316 		goto end;
8317 	case NFS4_CHECK_STATEID_REPLAY:
8318 		/* Check the sequence id for the open owner */
8319 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8320 		    resop)) {
8321 		case NFS4_CHKSEQ_OKAY:
8322 			/*
8323 			 * This is replayed stateid; if seqid matches
8324 			 * next expected, then client is using wrong seqid.
8325 			 */
8326 			/* FALL THROUGH */
8327 		case NFS4_CHKSEQ_BAD:
8328 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8329 			goto end;
8330 		case NFS4_CHKSEQ_REPLAY:
8331 			/*
8332 			 * Note this case is the duplicate case so
8333 			 * resp->status is already set.
8334 			 */
8335 			*cs->statusp = resp->status;
8336 			rfs4_update_lease(sp->rs_owner->ro_client);
8337 			goto end;
8338 		}
8339 		break;
8340 	default:
8341 		ASSERT(FALSE);
8342 		break;
8343 	}
8344 
8345 	rfs4_dbe_lock(sp->rs_dbe);
8346 
8347 	/* Update the stateid. */
8348 	next_stateid(&sp->rs_stateid);
8349 	resp->open_stateid = sp->rs_stateid.stateid;
8350 
8351 	rfs4_dbe_unlock(sp->rs_dbe);
8352 
8353 	rfs4_update_lease(sp->rs_owner->ro_client);
8354 	rfs4_update_open_sequence(sp->rs_owner);
8355 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8356 
8357 	rfs4_state_close(sp, FALSE, FALSE, cs->cr);
8358 
8359 	*cs->statusp = resp->status = status;
8360 
8361 end:
8362 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
8363 	rfs4_state_rele(sp);
8364 out:
8365 	DTRACE_NFSV4_2(op__close__done, struct compound_state *, cs,
8366 	    CLOSE4res *, resp);
8367 }
8368 
8369 /*
8370  * Manage the counts on the file struct and close all file locks
8371  */
8372 /*ARGSUSED*/
8373 void
8374 rfs4_release_share_lock_state(rfs4_state_t *sp, cred_t *cr,
8375     bool_t close_of_client)
8376 {
8377 	rfs4_file_t *fp = sp->rs_finfo;
8378 	rfs4_lo_state_t *lsp;
8379 	int fflags = 0;
8380 
8381 	/*
8382 	 * If this call is part of the larger closing down of client
8383 	 * state then it is just easier to release all locks
8384 	 * associated with this client instead of going through each
8385 	 * individual file and cleaning locks there.
8386 	 */
8387 	if (close_of_client) {
8388 		if (sp->rs_owner->ro_client->rc_unlksys_completed == FALSE &&
8389 		    !list_is_empty(&sp->rs_lostatelist) &&
8390 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID) {
8391 			/* Is the PxFS kernel module loaded? */
8392 			if (lm_remove_file_locks != NULL) {
8393 				int new_sysid;
8394 
8395 				/* Encode the cluster nodeid in new sysid */
8396 				new_sysid = sp->rs_owner->ro_client->rc_sysidt;
8397 				lm_set_nlmid_flk(&new_sysid);
8398 
8399 				/*
8400 				 * This PxFS routine removes file locks for a
8401 				 * client over all nodes of a cluster.
8402 				 */
8403 				NFS4_DEBUG(rfs4_debug, (CE_NOTE,
8404 				    "lm_remove_file_locks(sysid=0x%x)\n",
8405 				    new_sysid));
8406 				(*lm_remove_file_locks)(new_sysid);
8407 			} else {
8408 				struct flock64 flk;
8409 
8410 				/* Release all locks for this client */
8411 				flk.l_type = F_UNLKSYS;
8412 				flk.l_whence = 0;
8413 				flk.l_start = 0;
8414 				flk.l_len = 0;
8415 				flk.l_sysid =
8416 				    sp->rs_owner->ro_client->rc_sysidt;
8417 				flk.l_pid = 0;
8418 				(void) VOP_FRLOCK(sp->rs_finfo->rf_vp, F_SETLK,
8419 				    &flk, F_REMOTELOCK | FREAD | FWRITE,
8420 				    (u_offset_t)0, NULL, CRED(), NULL);
8421 			}
8422 
8423 			sp->rs_owner->ro_client->rc_unlksys_completed = TRUE;
8424 		}
8425 	}
8426 
8427 	/*
8428 	 * Release all locks on this file by this lock owner or at
8429 	 * least mark the locks as having been released
8430 	 */
8431 	for (lsp = list_head(&sp->rs_lostatelist); lsp != NULL;
8432 	    lsp = list_next(&sp->rs_lostatelist, lsp)) {
8433 		lsp->rls_locks_cleaned = TRUE;
8434 
8435 		/* Was this already taken care of above? */
8436 		if (!close_of_client &&
8437 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8438 			(void) cleanlocks(sp->rs_finfo->rf_vp,
8439 			    lsp->rls_locker->rl_pid,
8440 			    lsp->rls_locker->rl_client->rc_sysidt);
8441 	}
8442 
8443 	/*
8444 	 * Release any shrlocks associated with this open state ID.
8445 	 * This must be done before the rfs4_state gets marked closed.
8446 	 */
8447 	if (sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8448 		(void) rfs4_unshare(sp);
8449 
8450 	if (sp->rs_open_access) {
8451 		rfs4_dbe_lock(fp->rf_dbe);
8452 
8453 		/*
8454 		 * Decrement the count for each access and deny bit that this
8455 		 * state has contributed to the file.
8456 		 * If the file counts go to zero
8457 		 * clear the appropriate bit in the appropriate mask.
8458 		 */
8459 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) {
8460 			fp->rf_access_read--;
8461 			fflags |= FREAD;
8462 			if (fp->rf_access_read == 0)
8463 				fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
8464 		}
8465 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) {
8466 			fp->rf_access_write--;
8467 			fflags |= FWRITE;
8468 			if (fp->rf_access_write == 0)
8469 				fp->rf_share_access &=
8470 				    ~OPEN4_SHARE_ACCESS_WRITE;
8471 		}
8472 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_READ) {
8473 			fp->rf_deny_read--;
8474 			if (fp->rf_deny_read == 0)
8475 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
8476 		}
8477 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) {
8478 			fp->rf_deny_write--;
8479 			if (fp->rf_deny_write == 0)
8480 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
8481 		}
8482 
8483 		(void) VOP_CLOSE(fp->rf_vp, fflags, 1, (offset_t)0, cr, NULL);
8484 
8485 		rfs4_dbe_unlock(fp->rf_dbe);
8486 
8487 		sp->rs_open_access = 0;
8488 		sp->rs_open_deny = 0;
8489 	}
8490 }
8491 
8492 /*
8493  * lock_denied: Fill in a LOCK4deneid structure given an flock64 structure.
8494  */
8495 static nfsstat4
8496 lock_denied(LOCK4denied *dp, struct flock64 *flk)
8497 {
8498 	rfs4_lockowner_t *lo;
8499 	rfs4_client_t *cp;
8500 	uint32_t len;
8501 
8502 	lo = rfs4_findlockowner_by_pid(flk->l_pid);
8503 	if (lo != NULL) {
8504 		cp = lo->rl_client;
8505 		if (rfs4_lease_expired(cp)) {
8506 			rfs4_lockowner_rele(lo);
8507 			rfs4_dbe_hold(cp->rc_dbe);
8508 			rfs4_client_close(cp);
8509 			return (NFS4ERR_EXPIRED);
8510 		}
8511 		dp->owner.clientid = lo->rl_owner.clientid;
8512 		len = lo->rl_owner.owner_len;
8513 		dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8514 		bcopy(lo->rl_owner.owner_val, dp->owner.owner_val, len);
8515 		dp->owner.owner_len = len;
8516 		rfs4_lockowner_rele(lo);
8517 		goto finish;
8518 	}
8519 
8520 	/*
8521 	 * Its not a NFS4 lock. We take advantage that the upper 32 bits
8522 	 * of the client id contain the boot time for a NFS4 lock. So we
8523 	 * fabricate and identity by setting clientid to the sysid, and
8524 	 * the lock owner to the pid.
8525 	 */
8526 	dp->owner.clientid = flk->l_sysid;
8527 	len = sizeof (pid_t);
8528 	dp->owner.owner_len = len;
8529 	dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8530 	bcopy(&flk->l_pid, dp->owner.owner_val, len);
8531 finish:
8532 	dp->offset = flk->l_start;
8533 	dp->length = flk->l_len;
8534 
8535 	if (flk->l_type == F_RDLCK)
8536 		dp->locktype = READ_LT;
8537 	else if (flk->l_type == F_WRLCK)
8538 		dp->locktype = WRITE_LT;
8539 	else
8540 		return (NFS4ERR_INVAL);	/* no mapping from POSIX ltype to v4 */
8541 
8542 	return (NFS4_OK);
8543 }
8544 
8545 static int
8546 setlock(vnode_t *vp, struct flock64 *flock, int flag, cred_t *cred)
8547 {
8548 	int error;
8549 	struct flock64 flk;
8550 	int i;
8551 	clock_t delaytime;
8552 	int cmd;
8553 
8554 	cmd = nbl_need_check(vp) ? F_SETLK_NBMAND : F_SETLK;
8555 retry:
8556 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
8557 
8558 	for (i = 0; i < rfs4_maxlock_tries; i++) {
8559 		LOCK_PRINT(rfs4_debug, "setlock", cmd, flock);
8560 		error = VOP_FRLOCK(vp, cmd,
8561 		    flock, flag, (u_offset_t)0, NULL, cred, NULL);
8562 
8563 		if (error != EAGAIN && error != EACCES)
8564 			break;
8565 
8566 		if (i < rfs4_maxlock_tries - 1) {
8567 			delay(delaytime);
8568 			delaytime *= 2;
8569 		}
8570 	}
8571 
8572 	if (error == EAGAIN || error == EACCES) {
8573 		/* Get the owner of the lock */
8574 		flk = *flock;
8575 		LOCK_PRINT(rfs4_debug, "setlock", F_GETLK, &flk);
8576 		if (VOP_FRLOCK(vp, F_GETLK, &flk, flag,
8577 		    (u_offset_t)0, NULL, cred, NULL) == 0) {
8578 			if (flk.l_type == F_UNLCK) {
8579 				/* No longer locked, retry */
8580 				goto retry;
8581 			}
8582 			*flock = flk;
8583 			LOCK_PRINT(rfs4_debug, "setlock(blocking lock)",
8584 			    F_GETLK, &flk);
8585 		}
8586 	}
8587 
8588 	return (error);
8589 }
8590 
8591 /*ARGSUSED*/
8592 static nfsstat4
8593 rfs4_do_lock(rfs4_lo_state_t *lsp, nfs_lock_type4 locktype,
8594     offset4 offset, length4 length, cred_t *cred, nfs_resop4 *resop)
8595 {
8596 	nfsstat4 status;
8597 	rfs4_lockowner_t *lo = lsp->rls_locker;
8598 	rfs4_state_t *sp = lsp->rls_state;
8599 	struct flock64 flock;
8600 	int16_t ltype;
8601 	int flag;
8602 	int error;
8603 	sysid_t sysid;
8604 	LOCK4res *lres;
8605 
8606 	if (rfs4_lease_expired(lo->rl_client)) {
8607 		return (NFS4ERR_EXPIRED);
8608 	}
8609 
8610 	if ((status = rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
8611 		return (status);
8612 
8613 	/* Check for zero length. To lock to end of file use all ones for V4 */
8614 	if (length == 0)
8615 		return (NFS4ERR_INVAL);
8616 	else if (length == (length4)(~0))
8617 		length = 0;		/* Posix to end of file  */
8618 
8619 retry:
8620 	rfs4_dbe_lock(sp->rs_dbe);
8621 	if (sp->rs_closed) {
8622 		rfs4_dbe_unlock(sp->rs_dbe);
8623 		return (NFS4ERR_OLD_STATEID);
8624 	}
8625 
8626 	if (resop->resop != OP_LOCKU) {
8627 		switch (locktype) {
8628 		case READ_LT:
8629 		case READW_LT:
8630 			if ((sp->rs_share_access
8631 			    & OPEN4_SHARE_ACCESS_READ) == 0) {
8632 				rfs4_dbe_unlock(sp->rs_dbe);
8633 
8634 				return (NFS4ERR_OPENMODE);
8635 			}
8636 			ltype = F_RDLCK;
8637 			break;
8638 		case WRITE_LT:
8639 		case WRITEW_LT:
8640 			if ((sp->rs_share_access
8641 			    & OPEN4_SHARE_ACCESS_WRITE) == 0) {
8642 				rfs4_dbe_unlock(sp->rs_dbe);
8643 
8644 				return (NFS4ERR_OPENMODE);
8645 			}
8646 			ltype = F_WRLCK;
8647 			break;
8648 		}
8649 	} else
8650 		ltype = F_UNLCK;
8651 
8652 	flock.l_type = ltype;
8653 	flock.l_whence = 0;		/* SEEK_SET */
8654 	flock.l_start = offset;
8655 	flock.l_len = length;
8656 	flock.l_sysid = sysid;
8657 	flock.l_pid = lsp->rls_locker->rl_pid;
8658 
8659 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
8660 	if (flock.l_len < 0 || flock.l_start < 0) {
8661 		rfs4_dbe_unlock(sp->rs_dbe);
8662 		return (NFS4ERR_INVAL);
8663 	}
8664 
8665 	/*
8666 	 * N.B. FREAD has the same value as OPEN4_SHARE_ACCESS_READ and
8667 	 * FWRITE has the same value as OPEN4_SHARE_ACCESS_WRITE.
8668 	 */
8669 	flag = (int)sp->rs_share_access | F_REMOTELOCK;
8670 
8671 	error = setlock(sp->rs_finfo->rf_vp, &flock, flag, cred);
8672 	if (error == 0) {
8673 		rfs4_dbe_lock(lsp->rls_dbe);
8674 		next_stateid(&lsp->rls_lockid);
8675 		rfs4_dbe_unlock(lsp->rls_dbe);
8676 	}
8677 
8678 	rfs4_dbe_unlock(sp->rs_dbe);
8679 
8680 	/*
8681 	 * N.B. We map error values to nfsv4 errors. This is differrent
8682 	 * than puterrno4 routine.
8683 	 */
8684 	switch (error) {
8685 	case 0:
8686 		status = NFS4_OK;
8687 		break;
8688 	case EAGAIN:
8689 	case EACCES:		/* Old value */
8690 		/* Can only get here if op is OP_LOCK */
8691 		ASSERT(resop->resop == OP_LOCK);
8692 		lres = &resop->nfs_resop4_u.oplock;
8693 		status = NFS4ERR_DENIED;
8694 		if (lock_denied(&lres->LOCK4res_u.denied, &flock)
8695 		    == NFS4ERR_EXPIRED)
8696 			goto retry;
8697 		break;
8698 	case ENOLCK:
8699 		status = NFS4ERR_DELAY;
8700 		break;
8701 	case EOVERFLOW:
8702 		status = NFS4ERR_INVAL;
8703 		break;
8704 	case EINVAL:
8705 		status = NFS4ERR_NOTSUPP;
8706 		break;
8707 	default:
8708 		status = NFS4ERR_SERVERFAULT;
8709 		break;
8710 	}
8711 
8712 	return (status);
8713 }
8714 
8715 /*ARGSUSED*/
8716 void
8717 rfs4_op_lock(nfs_argop4 *argop, nfs_resop4 *resop,
8718     struct svc_req *req, struct compound_state *cs)
8719 {
8720 	LOCK4args *args = &argop->nfs_argop4_u.oplock;
8721 	LOCK4res *resp = &resop->nfs_resop4_u.oplock;
8722 	nfsstat4 status;
8723 	stateid4 *stateid;
8724 	rfs4_lockowner_t *lo;
8725 	rfs4_client_t *cp;
8726 	rfs4_state_t *sp = NULL;
8727 	rfs4_lo_state_t *lsp = NULL;
8728 	bool_t ls_sw_held = FALSE;
8729 	bool_t create = TRUE;
8730 	bool_t lcreate = TRUE;
8731 	bool_t dup_lock = FALSE;
8732 	int rc;
8733 
8734 	DTRACE_NFSV4_2(op__lock__start, struct compound_state *, cs,
8735 	    LOCK4args *, args);
8736 
8737 	if (cs->vp == NULL) {
8738 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8739 		DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8740 		    cs, LOCK4res *, resp);
8741 		return;
8742 	}
8743 
8744 	if (args->locker.new_lock_owner) {
8745 		/* Create a new lockowner for this instance */
8746 		open_to_lock_owner4 *olo = &args->locker.locker4_u.open_owner;
8747 
8748 		NFS4_DEBUG(rfs4_debug, (CE_NOTE, "Creating new lock owner"));
8749 
8750 		stateid = &olo->open_stateid;
8751 		status = rfs4_get_state(stateid, &sp, RFS4_DBS_VALID);
8752 		if (status != NFS4_OK) {
8753 			NFS4_DEBUG(rfs4_debug,
8754 			    (CE_NOTE, "Get state failed in lock %d", status));
8755 			*cs->statusp = resp->status = status;
8756 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8757 			    cs, LOCK4res *, resp);
8758 			return;
8759 		}
8760 
8761 		/* Ensure specified filehandle matches */
8762 		if (cs->vp != sp->rs_finfo->rf_vp) {
8763 			rfs4_state_rele(sp);
8764 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8765 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8766 			    cs, LOCK4res *, resp);
8767 			return;
8768 		}
8769 
8770 		/* hold off other access to open_owner while we tinker */
8771 		rfs4_sw_enter(&sp->rs_owner->ro_sw);
8772 
8773 		switch (rc = rfs4_check_stateid_seqid(sp, stateid)) {
8774 		case NFS4_CHECK_STATEID_OLD:
8775 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8776 			goto end;
8777 		case NFS4_CHECK_STATEID_BAD:
8778 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8779 			goto end;
8780 		case NFS4_CHECK_STATEID_EXPIRED:
8781 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8782 			goto end;
8783 		case NFS4_CHECK_STATEID_UNCONFIRMED:
8784 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8785 			goto end;
8786 		case NFS4_CHECK_STATEID_CLOSED:
8787 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8788 			goto end;
8789 		case NFS4_CHECK_STATEID_OKAY:
8790 		case NFS4_CHECK_STATEID_REPLAY:
8791 			switch (rfs4_check_olo_seqid(olo->open_seqid,
8792 			    sp->rs_owner, resop)) {
8793 			case NFS4_CHKSEQ_OKAY:
8794 				if (rc == NFS4_CHECK_STATEID_OKAY)
8795 					break;
8796 				/*
8797 				 * This is replayed stateid; if seqid
8798 				 * matches next expected, then client
8799 				 * is using wrong seqid.
8800 				 */
8801 				/* FALLTHROUGH */
8802 			case NFS4_CHKSEQ_BAD:
8803 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8804 				goto end;
8805 			case NFS4_CHKSEQ_REPLAY:
8806 				/* This is a duplicate LOCK request */
8807 				dup_lock = TRUE;
8808 
8809 				/*
8810 				 * For a duplicate we do not want to
8811 				 * create a new lockowner as it should
8812 				 * already exist.
8813 				 * Turn off the lockowner create flag.
8814 				 */
8815 				lcreate = FALSE;
8816 			}
8817 			break;
8818 		}
8819 
8820 		lo = rfs4_findlockowner(&olo->lock_owner, &lcreate);
8821 		if (lo == NULL) {
8822 			NFS4_DEBUG(rfs4_debug,
8823 			    (CE_NOTE, "rfs4_op_lock: no lock owner"));
8824 			*cs->statusp = resp->status = NFS4ERR_RESOURCE;
8825 			goto end;
8826 		}
8827 
8828 		lsp = rfs4_findlo_state_by_owner(lo, sp, &create);
8829 		if (lsp == NULL) {
8830 			rfs4_update_lease(sp->rs_owner->ro_client);
8831 			/*
8832 			 * Only update theh open_seqid if this is not
8833 			 * a duplicate request
8834 			 */
8835 			if (dup_lock == FALSE) {
8836 				rfs4_update_open_sequence(sp->rs_owner);
8837 			}
8838 
8839 			NFS4_DEBUG(rfs4_debug,
8840 			    (CE_NOTE, "rfs4_op_lock: no state"));
8841 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
8842 			rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8843 			rfs4_lockowner_rele(lo);
8844 			goto end;
8845 		}
8846 
8847 		/*
8848 		 * This is the new_lock_owner branch and the client is
8849 		 * supposed to be associating a new lock_owner with
8850 		 * the open file at this point.  If we find that a
8851 		 * lock_owner/state association already exists and a
8852 		 * successful LOCK request was returned to the client,
8853 		 * an error is returned to the client since this is
8854 		 * not appropriate.  The client should be using the
8855 		 * existing lock_owner branch.
8856 		 */
8857 		if (dup_lock == FALSE && create == FALSE) {
8858 			if (lsp->rls_lock_completed == TRUE) {
8859 				*cs->statusp =
8860 				    resp->status = NFS4ERR_BAD_SEQID;
8861 				rfs4_lockowner_rele(lo);
8862 				goto end;
8863 			}
8864 		}
8865 
8866 		rfs4_update_lease(sp->rs_owner->ro_client);
8867 
8868 		/*
8869 		 * Only update theh open_seqid if this is not
8870 		 * a duplicate request
8871 		 */
8872 		if (dup_lock == FALSE) {
8873 			rfs4_update_open_sequence(sp->rs_owner);
8874 		}
8875 
8876 		/*
8877 		 * If this is a duplicate lock request, just copy the
8878 		 * previously saved reply and return.
8879 		 */
8880 		if (dup_lock == TRUE) {
8881 			/* verify that lock_seqid's match */
8882 			if (lsp->rls_seqid != olo->lock_seqid) {
8883 				NFS4_DEBUG(rfs4_debug,
8884 				    (CE_NOTE, "rfs4_op_lock: Dup-Lock seqid bad"
8885 				    "lsp->seqid=%d old->seqid=%d",
8886 				    lsp->rls_seqid, olo->lock_seqid));
8887 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8888 			} else {
8889 				rfs4_copy_reply(resop, &lsp->rls_reply);
8890 				/*
8891 				 * Make sure to copy the just
8892 				 * retrieved reply status into the
8893 				 * overall compound status
8894 				 */
8895 				*cs->statusp = resp->status;
8896 			}
8897 			rfs4_lockowner_rele(lo);
8898 			goto end;
8899 		}
8900 
8901 		rfs4_dbe_lock(lsp->rls_dbe);
8902 
8903 		/* Make sure to update the lock sequence id */
8904 		lsp->rls_seqid = olo->lock_seqid;
8905 
8906 		NFS4_DEBUG(rfs4_debug,
8907 		    (CE_NOTE, "Lock seqid established as %d", lsp->rls_seqid));
8908 
8909 		/*
8910 		 * This is used to signify the newly created lockowner
8911 		 * stateid and its sequence number.  The checks for
8912 		 * sequence number and increment don't occur on the
8913 		 * very first lock request for a lockowner.
8914 		 */
8915 		lsp->rls_skip_seqid_check = TRUE;
8916 
8917 		/* hold off other access to lsp while we tinker */
8918 		rfs4_sw_enter(&lsp->rls_sw);
8919 		ls_sw_held = TRUE;
8920 
8921 		rfs4_dbe_unlock(lsp->rls_dbe);
8922 
8923 		rfs4_lockowner_rele(lo);
8924 	} else {
8925 		stateid = &args->locker.locker4_u.lock_owner.lock_stateid;
8926 		/* get lsp and hold the lock on the underlying file struct */
8927 		if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE))
8928 		    != NFS4_OK) {
8929 			*cs->statusp = resp->status = status;
8930 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8931 			    cs, LOCK4res *, resp);
8932 			return;
8933 		}
8934 		create = FALSE;	/* We didn't create lsp */
8935 
8936 		/* Ensure specified filehandle matches */
8937 		if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
8938 			rfs4_lo_state_rele(lsp, TRUE);
8939 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8940 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8941 			    cs, LOCK4res *, resp);
8942 			return;
8943 		}
8944 
8945 		/* hold off other access to lsp while we tinker */
8946 		rfs4_sw_enter(&lsp->rls_sw);
8947 		ls_sw_held = TRUE;
8948 
8949 		switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
8950 		/*
8951 		 * The stateid looks like it was okay (expected to be
8952 		 * the next one)
8953 		 */
8954 		case NFS4_CHECK_STATEID_OKAY:
8955 			/*
8956 			 * The sequence id is now checked.  Determine
8957 			 * if this is a replay or if it is in the
8958 			 * expected (next) sequence.  In the case of a
8959 			 * replay, there are two replay conditions
8960 			 * that may occur.  The first is the normal
8961 			 * condition where a LOCK is done with a
8962 			 * NFS4_OK response and the stateid is
8963 			 * updated.  That case is handled below when
8964 			 * the stateid is identified as a REPLAY.  The
8965 			 * second is the case where an error is
8966 			 * returned, like NFS4ERR_DENIED, and the
8967 			 * sequence number is updated but the stateid
8968 			 * is not updated.  This second case is dealt
8969 			 * with here.  So it may seem odd that the
8970 			 * stateid is okay but the sequence id is a
8971 			 * replay but it is okay.
8972 			 */
8973 			switch (rfs4_check_lock_seqid(
8974 			    args->locker.locker4_u.lock_owner.lock_seqid,
8975 			    lsp, resop)) {
8976 			case NFS4_CHKSEQ_REPLAY:
8977 				if (resp->status != NFS4_OK) {
8978 					/*
8979 					 * Here is our replay and need
8980 					 * to verify that the last
8981 					 * response was an error.
8982 					 */
8983 					*cs->statusp = resp->status;
8984 					goto end;
8985 				}
8986 				/*
8987 				 * This is done since the sequence id
8988 				 * looked like a replay but it didn't
8989 				 * pass our check so a BAD_SEQID is
8990 				 * returned as a result.
8991 				 */
8992 				/*FALLTHROUGH*/
8993 			case NFS4_CHKSEQ_BAD:
8994 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8995 				goto end;
8996 			case NFS4_CHKSEQ_OKAY:
8997 				/* Everything looks okay move ahead */
8998 				break;
8999 			}
9000 			break;
9001 		case NFS4_CHECK_STATEID_OLD:
9002 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9003 			goto end;
9004 		case NFS4_CHECK_STATEID_BAD:
9005 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9006 			goto end;
9007 		case NFS4_CHECK_STATEID_EXPIRED:
9008 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9009 			goto end;
9010 		case NFS4_CHECK_STATEID_CLOSED:
9011 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9012 			goto end;
9013 		case NFS4_CHECK_STATEID_REPLAY:
9014 			switch (rfs4_check_lock_seqid(
9015 			    args->locker.locker4_u.lock_owner.lock_seqid,
9016 			    lsp, resop)) {
9017 			case NFS4_CHKSEQ_OKAY:
9018 				/*
9019 				 * This is a replayed stateid; if
9020 				 * seqid matches the next expected,
9021 				 * then client is using wrong seqid.
9022 				 */
9023 			case NFS4_CHKSEQ_BAD:
9024 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9025 				goto end;
9026 			case NFS4_CHKSEQ_REPLAY:
9027 				rfs4_update_lease(lsp->rls_locker->rl_client);
9028 				*cs->statusp = status = resp->status;
9029 				goto end;
9030 			}
9031 			break;
9032 		default:
9033 			ASSERT(FALSE);
9034 			break;
9035 		}
9036 
9037 		rfs4_update_lock_sequence(lsp);
9038 		rfs4_update_lease(lsp->rls_locker->rl_client);
9039 	}
9040 
9041 	/*
9042 	 * NFS4 only allows locking on regular files, so
9043 	 * verify type of object.
9044 	 */
9045 	if (cs->vp->v_type != VREG) {
9046 		if (cs->vp->v_type == VDIR)
9047 			status = NFS4ERR_ISDIR;
9048 		else
9049 			status = NFS4ERR_INVAL;
9050 		goto out;
9051 	}
9052 
9053 	cp = lsp->rls_state->rs_owner->ro_client;
9054 
9055 	if (rfs4_clnt_in_grace(cp) && !args->reclaim) {
9056 		status = NFS4ERR_GRACE;
9057 		goto out;
9058 	}
9059 
9060 	if (rfs4_clnt_in_grace(cp) && args->reclaim && !cp->rc_can_reclaim) {
9061 		status = NFS4ERR_NO_GRACE;
9062 		goto out;
9063 	}
9064 
9065 	if (!rfs4_clnt_in_grace(cp) && args->reclaim) {
9066 		status = NFS4ERR_NO_GRACE;
9067 		goto out;
9068 	}
9069 
9070 	if (lsp->rls_state->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE)
9071 		cs->deleg = TRUE;
9072 
9073 	status = rfs4_do_lock(lsp, args->locktype,
9074 	    args->offset, args->length, cs->cr, resop);
9075 
9076 out:
9077 	lsp->rls_skip_seqid_check = FALSE;
9078 
9079 	*cs->statusp = resp->status = status;
9080 
9081 	if (status == NFS4_OK) {
9082 		resp->LOCK4res_u.lock_stateid = lsp->rls_lockid.stateid;
9083 		lsp->rls_lock_completed = TRUE;
9084 	}
9085 	/*
9086 	 * Only update the "OPEN" response here if this was a new
9087 	 * lock_owner
9088 	 */
9089 	if (sp)
9090 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
9091 
9092 	rfs4_update_lock_resp(lsp, resop);
9093 
9094 end:
9095 	if (lsp) {
9096 		if (ls_sw_held)
9097 			rfs4_sw_exit(&lsp->rls_sw);
9098 		/*
9099 		 * If an sp obtained, then the lsp does not represent
9100 		 * a lock on the file struct.
9101 		 */
9102 		if (sp != NULL)
9103 			rfs4_lo_state_rele(lsp, FALSE);
9104 		else
9105 			rfs4_lo_state_rele(lsp, TRUE);
9106 	}
9107 	if (sp) {
9108 		rfs4_sw_exit(&sp->rs_owner->ro_sw);
9109 		rfs4_state_rele(sp);
9110 	}
9111 
9112 	DTRACE_NFSV4_2(op__lock__done, struct compound_state *, cs,
9113 	    LOCK4res *, resp);
9114 }
9115 
9116 /* free function for LOCK/LOCKT */
9117 static void
9118 lock_denied_free(nfs_resop4 *resop)
9119 {
9120 	LOCK4denied *dp = NULL;
9121 
9122 	switch (resop->resop) {
9123 	case OP_LOCK:
9124 		if (resop->nfs_resop4_u.oplock.status == NFS4ERR_DENIED)
9125 			dp = &resop->nfs_resop4_u.oplock.LOCK4res_u.denied;
9126 		break;
9127 	case OP_LOCKT:
9128 		if (resop->nfs_resop4_u.oplockt.status == NFS4ERR_DENIED)
9129 			dp = &resop->nfs_resop4_u.oplockt.denied;
9130 		break;
9131 	default:
9132 		break;
9133 	}
9134 
9135 	if (dp)
9136 		kmem_free(dp->owner.owner_val, dp->owner.owner_len);
9137 }
9138 
9139 /*ARGSUSED*/
9140 void
9141 rfs4_op_locku(nfs_argop4 *argop, nfs_resop4 *resop,
9142     struct svc_req *req, struct compound_state *cs)
9143 {
9144 	LOCKU4args *args = &argop->nfs_argop4_u.oplocku;
9145 	LOCKU4res *resp = &resop->nfs_resop4_u.oplocku;
9146 	nfsstat4 status;
9147 	stateid4 *stateid = &args->lock_stateid;
9148 	rfs4_lo_state_t *lsp;
9149 
9150 	DTRACE_NFSV4_2(op__locku__start, struct compound_state *, cs,
9151 	    LOCKU4args *, args);
9152 
9153 	if (cs->vp == NULL) {
9154 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9155 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9156 		    LOCKU4res *, resp);
9157 		return;
9158 	}
9159 
9160 	if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE)) != NFS4_OK) {
9161 		*cs->statusp = resp->status = status;
9162 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9163 		    LOCKU4res *, resp);
9164 		return;
9165 	}
9166 
9167 	/* Ensure specified filehandle matches */
9168 	if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9169 		rfs4_lo_state_rele(lsp, TRUE);
9170 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9171 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9172 		    LOCKU4res *, resp);
9173 		return;
9174 	}
9175 
9176 	/* hold off other access to lsp while we tinker */
9177 	rfs4_sw_enter(&lsp->rls_sw);
9178 
9179 	switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9180 	case NFS4_CHECK_STATEID_OKAY:
9181 		if (rfs4_check_lock_seqid(args->seqid, lsp, resop)
9182 		    != NFS4_CHKSEQ_OKAY) {
9183 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9184 			goto end;
9185 		}
9186 		break;
9187 	case NFS4_CHECK_STATEID_OLD:
9188 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9189 		goto end;
9190 	case NFS4_CHECK_STATEID_BAD:
9191 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9192 		goto end;
9193 	case NFS4_CHECK_STATEID_EXPIRED:
9194 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9195 		goto end;
9196 	case NFS4_CHECK_STATEID_CLOSED:
9197 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9198 		goto end;
9199 	case NFS4_CHECK_STATEID_REPLAY:
9200 		switch (rfs4_check_lock_seqid(args->seqid, lsp, resop)) {
9201 		case NFS4_CHKSEQ_OKAY:
9202 				/*
9203 				 * This is a replayed stateid; if
9204 				 * seqid matches the next expected,
9205 				 * then client is using wrong seqid.
9206 				 */
9207 		case NFS4_CHKSEQ_BAD:
9208 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9209 			goto end;
9210 		case NFS4_CHKSEQ_REPLAY:
9211 			rfs4_update_lease(lsp->rls_locker->rl_client);
9212 			*cs->statusp = status = resp->status;
9213 			goto end;
9214 		}
9215 		break;
9216 	default:
9217 		ASSERT(FALSE);
9218 		break;
9219 	}
9220 
9221 	rfs4_update_lock_sequence(lsp);
9222 	rfs4_update_lease(lsp->rls_locker->rl_client);
9223 
9224 	/*
9225 	 * NFS4 only allows locking on regular files, so
9226 	 * verify type of object.
9227 	 */
9228 	if (cs->vp->v_type != VREG) {
9229 		if (cs->vp->v_type == VDIR)
9230 			status = NFS4ERR_ISDIR;
9231 		else
9232 			status = NFS4ERR_INVAL;
9233 		goto out;
9234 	}
9235 
9236 	if (rfs4_clnt_in_grace(lsp->rls_state->rs_owner->ro_client)) {
9237 		status = NFS4ERR_GRACE;
9238 		goto out;
9239 	}
9240 
9241 	status = rfs4_do_lock(lsp, args->locktype,
9242 	    args->offset, args->length, cs->cr, resop);
9243 
9244 out:
9245 	*cs->statusp = resp->status = status;
9246 
9247 	if (status == NFS4_OK)
9248 		resp->lock_stateid = lsp->rls_lockid.stateid;
9249 
9250 	rfs4_update_lock_resp(lsp, resop);
9251 
9252 end:
9253 	rfs4_sw_exit(&lsp->rls_sw);
9254 	rfs4_lo_state_rele(lsp, TRUE);
9255 
9256 	DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9257 	    LOCKU4res *, resp);
9258 }
9259 
9260 /*
9261  * LOCKT is a best effort routine, the client can not be guaranteed that
9262  * the status return is still in effect by the time the reply is received.
9263  * They are numerous race conditions in this routine, but we are not required
9264  * and can not be accurate.
9265  */
9266 /*ARGSUSED*/
9267 void
9268 rfs4_op_lockt(nfs_argop4 *argop, nfs_resop4 *resop,
9269     struct svc_req *req, struct compound_state *cs)
9270 {
9271 	LOCKT4args *args = &argop->nfs_argop4_u.oplockt;
9272 	LOCKT4res *resp = &resop->nfs_resop4_u.oplockt;
9273 	rfs4_lockowner_t *lo;
9274 	rfs4_client_t *cp;
9275 	bool_t create = FALSE;
9276 	struct flock64 flk;
9277 	int error;
9278 	int flag = FREAD | FWRITE;
9279 	int ltype;
9280 	length4 posix_length;
9281 	sysid_t sysid;
9282 	pid_t pid;
9283 
9284 	DTRACE_NFSV4_2(op__lockt__start, struct compound_state *, cs,
9285 	    LOCKT4args *, args);
9286 
9287 	if (cs->vp == NULL) {
9288 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9289 		goto out;
9290 	}
9291 
9292 	/*
9293 	 * NFS4 only allows locking on regular files, so
9294 	 * verify type of object.
9295 	 */
9296 	if (cs->vp->v_type != VREG) {
9297 		if (cs->vp->v_type == VDIR)
9298 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
9299 		else
9300 			*cs->statusp = resp->status =  NFS4ERR_INVAL;
9301 		goto out;
9302 	}
9303 
9304 	/*
9305 	 * Check out the clientid to ensure the server knows about it
9306 	 * so that we correctly inform the client of a server reboot.
9307 	 */
9308 	if ((cp = rfs4_findclient_by_id(args->owner.clientid, FALSE))
9309 	    == NULL) {
9310 		*cs->statusp = resp->status =
9311 		    rfs4_check_clientid(&args->owner.clientid, 0);
9312 		goto out;
9313 	}
9314 	if (rfs4_lease_expired(cp)) {
9315 		rfs4_client_close(cp);
9316 		/*
9317 		 * Protocol doesn't allow returning NFS4ERR_STALE as
9318 		 * other operations do on this check so STALE_CLIENTID
9319 		 * is returned instead
9320 		 */
9321 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
9322 		goto out;
9323 	}
9324 
9325 	if (rfs4_clnt_in_grace(cp) && !(cp->rc_can_reclaim)) {
9326 		*cs->statusp = resp->status = NFS4ERR_GRACE;
9327 		rfs4_client_rele(cp);
9328 		goto out;
9329 	}
9330 	rfs4_client_rele(cp);
9331 
9332 	resp->status = NFS4_OK;
9333 
9334 	switch (args->locktype) {
9335 	case READ_LT:
9336 	case READW_LT:
9337 		ltype = F_RDLCK;
9338 		break;
9339 	case WRITE_LT:
9340 	case WRITEW_LT:
9341 		ltype = F_WRLCK;
9342 		break;
9343 	}
9344 
9345 	posix_length = args->length;
9346 	/* Check for zero length. To lock to end of file use all ones for V4 */
9347 	if (posix_length == 0) {
9348 		*cs->statusp = resp->status = NFS4ERR_INVAL;
9349 		goto out;
9350 	} else if (posix_length == (length4)(~0)) {
9351 		posix_length = 0;	/* Posix to end of file  */
9352 	}
9353 
9354 	/* Find or create a lockowner */
9355 	lo = rfs4_findlockowner(&args->owner, &create);
9356 
9357 	if (lo) {
9358 		pid = lo->rl_pid;
9359 		if ((resp->status =
9360 		    rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
9361 			goto err;
9362 	} else {
9363 		pid = 0;
9364 		sysid = lockt_sysid;
9365 	}
9366 retry:
9367 	flk.l_type = ltype;
9368 	flk.l_whence = 0;		/* SEEK_SET */
9369 	flk.l_start = args->offset;
9370 	flk.l_len = posix_length;
9371 	flk.l_sysid = sysid;
9372 	flk.l_pid = pid;
9373 	flag |= F_REMOTELOCK;
9374 
9375 	LOCK_PRINT(rfs4_debug, "rfs4_op_lockt", F_GETLK, &flk);
9376 
9377 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
9378 	if (flk.l_len < 0 || flk.l_start < 0) {
9379 		resp->status = NFS4ERR_INVAL;
9380 		goto err;
9381 	}
9382 	error = VOP_FRLOCK(cs->vp, F_GETLK, &flk, flag, (u_offset_t)0,
9383 	    NULL, cs->cr, NULL);
9384 
9385 	/*
9386 	 * N.B. We map error values to nfsv4 errors. This is differrent
9387 	 * than puterrno4 routine.
9388 	 */
9389 	switch (error) {
9390 	case 0:
9391 		if (flk.l_type == F_UNLCK)
9392 			resp->status = NFS4_OK;
9393 		else {
9394 			if (lock_denied(&resp->denied, &flk) == NFS4ERR_EXPIRED)
9395 				goto retry;
9396 			resp->status = NFS4ERR_DENIED;
9397 		}
9398 		break;
9399 	case EOVERFLOW:
9400 		resp->status = NFS4ERR_INVAL;
9401 		break;
9402 	case EINVAL:
9403 		resp->status = NFS4ERR_NOTSUPP;
9404 		break;
9405 	default:
9406 		cmn_err(CE_WARN, "rfs4_op_lockt: unexpected errno (%d)",
9407 		    error);
9408 		resp->status = NFS4ERR_SERVERFAULT;
9409 		break;
9410 	}
9411 
9412 err:
9413 	if (lo)
9414 		rfs4_lockowner_rele(lo);
9415 	*cs->statusp = resp->status;
9416 out:
9417 	DTRACE_NFSV4_2(op__lockt__done, struct compound_state *, cs,
9418 	    LOCKT4res *, resp);
9419 }
9420 
9421 int
9422 rfs4_share(rfs4_state_t *sp, uint32_t access, uint32_t deny)
9423 {
9424 	int err;
9425 	int cmd;
9426 	vnode_t *vp;
9427 	struct shrlock shr;
9428 	struct shr_locowner shr_loco;
9429 	int fflags = 0;
9430 
9431 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9432 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9433 
9434 	if (sp->rs_closed)
9435 		return (NFS4ERR_OLD_STATEID);
9436 
9437 	vp = sp->rs_finfo->rf_vp;
9438 	ASSERT(vp);
9439 
9440 	shr.s_access = shr.s_deny = 0;
9441 
9442 	if (access & OPEN4_SHARE_ACCESS_READ) {
9443 		fflags |= FREAD;
9444 		shr.s_access |= F_RDACC;
9445 	}
9446 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
9447 		fflags |= FWRITE;
9448 		shr.s_access |= F_WRACC;
9449 	}
9450 	ASSERT(shr.s_access);
9451 
9452 	if (deny & OPEN4_SHARE_DENY_READ)
9453 		shr.s_deny |= F_RDDNY;
9454 	if (deny & OPEN4_SHARE_DENY_WRITE)
9455 		shr.s_deny |= F_WRDNY;
9456 
9457 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9458 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9459 	shr_loco.sl_pid = shr.s_pid;
9460 	shr_loco.sl_id = shr.s_sysid;
9461 	shr.s_owner = (caddr_t)&shr_loco;
9462 	shr.s_own_len = sizeof (shr_loco);
9463 
9464 	cmd = nbl_need_check(vp) ? F_SHARE_NBMAND : F_SHARE;
9465 
9466 	err = VOP_SHRLOCK(vp, cmd, &shr, fflags, CRED(), NULL);
9467 	if (err != 0) {
9468 		if (err == EAGAIN)
9469 			err = NFS4ERR_SHARE_DENIED;
9470 		else
9471 			err = puterrno4(err);
9472 		return (err);
9473 	}
9474 
9475 	sp->rs_share_access |= access;
9476 	sp->rs_share_deny |= deny;
9477 
9478 	return (0);
9479 }
9480 
9481 int
9482 rfs4_unshare(rfs4_state_t *sp)
9483 {
9484 	int err;
9485 	struct shrlock shr;
9486 	struct shr_locowner shr_loco;
9487 
9488 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9489 
9490 	if (sp->rs_closed || sp->rs_share_access == 0)
9491 		return (0);
9492 
9493 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9494 	ASSERT(sp->rs_finfo->rf_vp);
9495 
9496 	shr.s_access = shr.s_deny = 0;
9497 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9498 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9499 	shr_loco.sl_pid = shr.s_pid;
9500 	shr_loco.sl_id = shr.s_sysid;
9501 	shr.s_owner = (caddr_t)&shr_loco;
9502 	shr.s_own_len = sizeof (shr_loco);
9503 
9504 	err = VOP_SHRLOCK(sp->rs_finfo->rf_vp, F_UNSHARE, &shr, 0, CRED(),
9505 	    NULL);
9506 	if (err != 0) {
9507 		err = puterrno4(err);
9508 		return (err);
9509 	}
9510 
9511 	sp->rs_share_access = 0;
9512 	sp->rs_share_deny = 0;
9513 
9514 	return (0);
9515 
9516 }
9517 
9518 static int
9519 rdma_setup_read_data4(READ4args *args, READ4res *rok)
9520 {
9521 	struct clist	*wcl;
9522 	count4		count = rok->data_len;
9523 	int		wlist_len;
9524 
9525 	wcl = args->wlist;
9526 	if (rdma_setup_read_chunks(wcl, count, &wlist_len) == FALSE) {
9527 		return (FALSE);
9528 	}
9529 	wcl = args->wlist;
9530 	rok->wlist_len = wlist_len;
9531 	rok->wlist = wcl;
9532 	return (TRUE);
9533 }
9534 
9535 /* tunable to disable server referrals */
9536 int rfs4_no_referrals = 0;
9537 
9538 /*
9539  * Find an NFS record in reparse point data.
9540  * Returns 0 for success and <0 or an errno value on failure.
9541  */
9542 int
9543 vn_find_nfs_record(vnode_t *vp, nvlist_t **nvlp, char **svcp, char **datap)
9544 {
9545 	int err;
9546 	char *stype, *val;
9547 	nvlist_t *nvl;
9548 	nvpair_t *curr;
9549 
9550 	if ((nvl = reparse_init()) == NULL)
9551 		return (-1);
9552 
9553 	if ((err = reparse_vnode_parse(vp, nvl)) != 0) {
9554 		reparse_free(nvl);
9555 		return (err);
9556 	}
9557 
9558 	curr = NULL;
9559 	while ((curr = nvlist_next_nvpair(nvl, curr)) != NULL) {
9560 		if ((stype = nvpair_name(curr)) == NULL) {
9561 			reparse_free(nvl);
9562 			return (-2);
9563 		}
9564 		if (strncasecmp(stype, "NFS", 3) == 0)
9565 			break;
9566 	}
9567 
9568 	if ((curr == NULL) ||
9569 	    (nvpair_value_string(curr, &val))) {
9570 		reparse_free(nvl);
9571 		return (-3);
9572 	}
9573 	*nvlp = nvl;
9574 	*svcp = stype;
9575 	*datap = val;
9576 	return (0);
9577 }
9578 
9579 int
9580 vn_is_nfs_reparse(vnode_t *vp, cred_t *cr)
9581 {
9582 	nvlist_t *nvl;
9583 	char *s, *d;
9584 
9585 	if (rfs4_no_referrals != 0)
9586 		return (B_FALSE);
9587 
9588 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9589 		return (B_FALSE);
9590 
9591 	if (vn_find_nfs_record(vp, &nvl, &s, &d) != 0)
9592 		return (B_FALSE);
9593 
9594 	reparse_free(nvl);
9595 
9596 	return (B_TRUE);
9597 }
9598 
9599 /*
9600  * There is a user-level copy of this routine in ref_subr.c.
9601  * Changes should be kept in sync.
9602  */
9603 static int
9604 nfs4_create_components(char *path, component4 *comp4)
9605 {
9606 	int slen, plen, ncomp;
9607 	char *ori_path, *nxtc, buf[MAXNAMELEN];
9608 
9609 	if (path == NULL)
9610 		return (0);
9611 
9612 	plen = strlen(path) + 1;	/* include the terminator */
9613 	ori_path = path;
9614 	ncomp = 0;
9615 
9616 	/* count number of components in the path */
9617 	for (nxtc = path; nxtc < ori_path + plen; nxtc++) {
9618 		if (*nxtc == '/' || *nxtc == '\0' || *nxtc == '\n') {
9619 			if ((slen = nxtc - path) == 0) {
9620 				path = nxtc + 1;
9621 				continue;
9622 			}
9623 
9624 			if (comp4 != NULL) {
9625 				bcopy(path, buf, slen);
9626 				buf[slen] = '\0';
9627 				(void) str_to_utf8(buf, &comp4[ncomp]);
9628 			}
9629 
9630 			ncomp++;	/* 1 valid component */
9631 			path = nxtc + 1;
9632 		}
9633 		if (*nxtc == '\0' || *nxtc == '\n')
9634 			break;
9635 	}
9636 
9637 	return (ncomp);
9638 }
9639 
9640 /*
9641  * There is a user-level copy of this routine in ref_subr.c.
9642  * Changes should be kept in sync.
9643  */
9644 static int
9645 make_pathname4(char *path, pathname4 *pathname)
9646 {
9647 	int ncomp;
9648 	component4 *comp4;
9649 
9650 	if (pathname == NULL)
9651 		return (0);
9652 
9653 	if (path == NULL) {
9654 		pathname->pathname4_val = NULL;
9655 		pathname->pathname4_len = 0;
9656 		return (0);
9657 	}
9658 
9659 	/* count number of components to alloc buffer */
9660 	if ((ncomp = nfs4_create_components(path, NULL)) == 0) {
9661 		pathname->pathname4_val = NULL;
9662 		pathname->pathname4_len = 0;
9663 		return (0);
9664 	}
9665 	comp4 = kmem_zalloc(ncomp * sizeof (component4), KM_SLEEP);
9666 
9667 	/* copy components into allocated buffer */
9668 	ncomp = nfs4_create_components(path, comp4);
9669 
9670 	pathname->pathname4_val = comp4;
9671 	pathname->pathname4_len = ncomp;
9672 
9673 	return (ncomp);
9674 }
9675 
9676 #define	xdr_fs_locations4 xdr_fattr4_fs_locations
9677 
9678 fs_locations4 *
9679 fetch_referral(vnode_t *vp, cred_t *cr)
9680 {
9681 	nvlist_t *nvl;
9682 	char *stype, *sdata;
9683 	fs_locations4 *result;
9684 	char buf[1024];
9685 	size_t bufsize;
9686 	XDR xdr;
9687 	int err;
9688 
9689 	/*
9690 	 * Check attrs to ensure it's a reparse point
9691 	 */
9692 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9693 		return (NULL);
9694 
9695 	/*
9696 	 * Look for an NFS record and get the type and data
9697 	 */
9698 	if (vn_find_nfs_record(vp, &nvl, &stype, &sdata) != 0)
9699 		return (NULL);
9700 
9701 	/*
9702 	 * With the type and data, upcall to get the referral
9703 	 */
9704 	bufsize = sizeof (buf);
9705 	bzero(buf, sizeof (buf));
9706 	err = reparse_kderef((const char *)stype, (const char *)sdata,
9707 	    buf, &bufsize);
9708 	reparse_free(nvl);
9709 
9710 	DTRACE_PROBE4(nfs4serv__func__referral__upcall,
9711 	    char *, stype, char *, sdata, char *, buf, int, err);
9712 	if (err) {
9713 		cmn_err(CE_NOTE,
9714 		    "reparsed daemon not running: unable to get referral (%d)",
9715 		    err);
9716 		return (NULL);
9717 	}
9718 
9719 	/*
9720 	 * We get an XDR'ed record back from the kderef call
9721 	 */
9722 	xdrmem_create(&xdr, buf, bufsize, XDR_DECODE);
9723 	result = kmem_alloc(sizeof (fs_locations4), KM_SLEEP);
9724 	err = xdr_fs_locations4(&xdr, result);
9725 	XDR_DESTROY(&xdr);
9726 	if (err != TRUE) {
9727 		DTRACE_PROBE1(nfs4serv__func__referral__upcall__xdrfail,
9728 		    int, err);
9729 		return (NULL);
9730 	}
9731 
9732 	/*
9733 	 * Look at path to recover fs_root, ignoring the leading '/'
9734 	 */
9735 	(void) make_pathname4(vp->v_path, &result->fs_root);
9736 
9737 	return (result);
9738 }
9739 
9740 char *
9741 build_symlink(vnode_t *vp, cred_t *cr, size_t *strsz)
9742 {
9743 	fs_locations4 *fsl;
9744 	fs_location4 *fs;
9745 	char *server, *path, *symbuf;
9746 	static char *prefix = "/net/";
9747 	int i, size, npaths;
9748 	uint_t len;
9749 
9750 	/* Get the referral */
9751 	if ((fsl = fetch_referral(vp, cr)) == NULL)
9752 		return (NULL);
9753 
9754 	/* Deal with only the first location and first server */
9755 	fs = &fsl->locations_val[0];
9756 	server = utf8_to_str(&fs->server_val[0], &len, NULL);
9757 	if (server == NULL) {
9758 		rfs4_free_fs_locations4(fsl);
9759 		kmem_free(fsl, sizeof (fs_locations4));
9760 		return (NULL);
9761 	}
9762 
9763 	/* Figure out size for "/net/" + host + /path/path/path + NULL */
9764 	size = strlen(prefix) + len;
9765 	for (i = 0; i < fs->rootpath.pathname4_len; i++)
9766 		size += fs->rootpath.pathname4_val[i].utf8string_len + 1;
9767 
9768 	/* Allocate the symlink buffer and fill it */
9769 	symbuf = kmem_zalloc(size, KM_SLEEP);
9770 	(void) strcat(symbuf, prefix);
9771 	(void) strcat(symbuf, server);
9772 	kmem_free(server, len);
9773 
9774 	npaths = 0;
9775 	for (i = 0; i < fs->rootpath.pathname4_len; i++) {
9776 		path = utf8_to_str(&fs->rootpath.pathname4_val[i], &len, NULL);
9777 		if (path == NULL)
9778 			continue;
9779 		(void) strcat(symbuf, "/");
9780 		(void) strcat(symbuf, path);
9781 		npaths++;
9782 		kmem_free(path, len);
9783 	}
9784 
9785 	rfs4_free_fs_locations4(fsl);
9786 	kmem_free(fsl, sizeof (fs_locations4));
9787 
9788 	if (strsz != NULL)
9789 		*strsz = size;
9790 	return (symbuf);
9791 }
9792 
9793 /*
9794  * Check to see if we have a downrev Solaris client, so that we
9795  * can send it a symlink instead of a referral.
9796  */
9797 int
9798 client_is_downrev(struct svc_req *req)
9799 {
9800 	struct sockaddr *ca;
9801 	rfs4_clntip_t *ci;
9802 	bool_t create = FALSE;
9803 	int is_downrev;
9804 
9805 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
9806 	ASSERT(ca);
9807 	ci = rfs4_find_clntip(ca, &create);
9808 	if (ci == NULL)
9809 		return (0);
9810 	is_downrev = ci->ri_no_referrals;
9811 	rfs4_dbe_rele(ci->ri_dbe);
9812 	return (is_downrev);
9813 }
9814