xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs4_srv.c (revision e8921a52c53ee69f7b65f054d9b2e886139daa59)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2016 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
26  */
27 
28 /*
29  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
30  *	All Rights Reserved
31  */
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <sys/systm.h>
36 #include <sys/cred.h>
37 #include <sys/buf.h>
38 #include <sys/vfs.h>
39 #include <sys/vfs_opreg.h>
40 #include <sys/vnode.h>
41 #include <sys/uio.h>
42 #include <sys/errno.h>
43 #include <sys/sysmacros.h>
44 #include <sys/statvfs.h>
45 #include <sys/kmem.h>
46 #include <sys/dirent.h>
47 #include <sys/cmn_err.h>
48 #include <sys/debug.h>
49 #include <sys/systeminfo.h>
50 #include <sys/flock.h>
51 #include <sys/pathname.h>
52 #include <sys/nbmlock.h>
53 #include <sys/share.h>
54 #include <sys/atomic.h>
55 #include <sys/policy.h>
56 #include <sys/fem.h>
57 #include <sys/sdt.h>
58 #include <sys/ddi.h>
59 #include <sys/zone.h>
60 
61 #include <fs/fs_reparse.h>
62 
63 #include <rpc/types.h>
64 #include <rpc/auth.h>
65 #include <rpc/rpcsec_gss.h>
66 #include <rpc/svc.h>
67 
68 #include <nfs/nfs.h>
69 #include <nfs/export.h>
70 #include <nfs/nfs_cmd.h>
71 #include <nfs/lm.h>
72 #include <nfs/nfs4.h>
73 
74 #include <sys/strsubr.h>
75 #include <sys/strsun.h>
76 
77 #include <inet/common.h>
78 #include <inet/ip.h>
79 #include <inet/ip6.h>
80 
81 #include <sys/tsol/label.h>
82 #include <sys/tsol/tndb.h>
83 
84 #define	RFS4_MAXLOCK_TRIES 4	/* Try to get the lock this many times */
85 static int rfs4_maxlock_tries = RFS4_MAXLOCK_TRIES;
86 #define	RFS4_LOCK_DELAY 10	/* Milliseconds */
87 static clock_t  rfs4_lock_delay = RFS4_LOCK_DELAY;
88 extern struct svc_ops rdma_svc_ops;
89 extern int nfs_loaned_buffers;
90 /* End of Tunables */
91 
92 static int rdma_setup_read_data4(READ4args *, READ4res *);
93 
94 /*
95  * Used to bump the stateid4.seqid value and show changes in the stateid
96  */
97 #define	next_stateid(sp) (++(sp)->bits.chgseq)
98 
99 /*
100  * RFS4_MINLEN_ENTRY4: XDR-encoded size of smallest possible dirent.
101  *	This is used to return NFS4ERR_TOOSMALL when clients specify
102  *	maxcount that isn't large enough to hold the smallest possible
103  *	XDR encoded dirent.
104  *
105  *	    sizeof cookie (8 bytes) +
106  *	    sizeof name_len (4 bytes) +
107  *	    sizeof smallest (padded) name (4 bytes) +
108  *	    sizeof bitmap4_len (12 bytes) +   NOTE: we always encode len=2 bm4
109  *	    sizeof attrlist4_len (4 bytes) +
110  *	    sizeof next boolean (4 bytes)
111  *
112  * RFS4_MINLEN_RDDIR4: XDR-encoded size of READDIR op reply containing
113  * the smallest possible entry4 (assumes no attrs requested).
114  *	sizeof nfsstat4 (4 bytes) +
115  *	sizeof verifier4 (8 bytes) +
116  *	sizeof entry4list bool (4 bytes) +
117  *	sizeof entry4 	(36 bytes) +
118  *	sizeof eof bool  (4 bytes)
119  *
120  * RFS4_MINLEN_RDDIR_BUF: minimum length of buffer server will provide to
121  *	VOP_READDIR.  Its value is the size of the maximum possible dirent
122  *	for solaris.  The DIRENT64_RECLEN macro returns	the size of dirent
123  *	required for a given name length.  MAXNAMELEN is the maximum
124  *	filename length allowed in Solaris.  The first two DIRENT64_RECLEN()
125  *	macros are to allow for . and .. entries -- just a minor tweak to try
126  *	and guarantee that buffer we give to VOP_READDIR will be large enough
127  *	to hold ., .., and the largest possible solaris dirent64.
128  */
129 #define	RFS4_MINLEN_ENTRY4 36
130 #define	RFS4_MINLEN_RDDIR4 (4 + NFS4_VERIFIER_SIZE + 4 + RFS4_MINLEN_ENTRY4 + 4)
131 #define	RFS4_MINLEN_RDDIR_BUF \
132 	(DIRENT64_RECLEN(1) + DIRENT64_RECLEN(2) + DIRENT64_RECLEN(MAXNAMELEN))
133 
134 /*
135  * It would be better to pad to 4 bytes since that's what XDR would do,
136  * but the dirents UFS gives us are already padded to 8, so just take
137  * what we're given.  Dircount is only a hint anyway.  Currently the
138  * solaris kernel is ASCII only, so there's no point in calling the
139  * UTF8 functions.
140  *
141  * dirent64: named padded to provide 8 byte struct alignment
142  *	d_ino(8) + d_off(8) + d_reclen(2) + d_name(namelen + null(1) + pad)
143  *
144  * cookie: uint64_t   +  utf8namelen: uint_t  +   utf8name padded to 8 bytes
145  *
146  */
147 #define	DIRENT64_TO_DIRCOUNT(dp) \
148 	(3 * BYTES_PER_XDR_UNIT + DIRENT64_NAMELEN((dp)->d_reclen))
149 
150 time_t rfs4_start_time;			/* Initialized in rfs4_srvrinit */
151 
152 static sysid_t lockt_sysid;		/* dummy sysid for all LOCKT calls */
153 
154 u_longlong_t	nfs4_srv_caller_id;
155 uint_t		nfs4_srv_vkey = 0;
156 
157 verifier4	Write4verf;
158 verifier4	Readdir4verf;
159 
160 void	rfs4_init_compound_state(struct compound_state *);
161 
162 static void	nullfree(caddr_t);
163 static void	rfs4_op_inval(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
164 			struct compound_state *);
165 static void	rfs4_op_access(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
166 			struct compound_state *);
167 static void	rfs4_op_close(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
168 			struct compound_state *);
169 static void	rfs4_op_commit(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
170 			struct compound_state *);
171 static void	rfs4_op_create(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
172 			struct compound_state *);
173 static void	rfs4_op_create_free(nfs_resop4 *resop);
174 static void	rfs4_op_delegreturn(nfs_argop4 *, nfs_resop4 *,
175 			struct svc_req *, struct compound_state *);
176 static void	rfs4_op_delegpurge(nfs_argop4 *, nfs_resop4 *,
177 			struct svc_req *, struct compound_state *);
178 static void	rfs4_op_getattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
179 			struct compound_state *);
180 static void	rfs4_op_getattr_free(nfs_resop4 *);
181 static void	rfs4_op_getfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
182 			struct compound_state *);
183 static void	rfs4_op_getfh_free(nfs_resop4 *);
184 static void	rfs4_op_illegal(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
185 			struct compound_state *);
186 static void	rfs4_op_link(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
187 			struct compound_state *);
188 static void	rfs4_op_lock(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
189 			struct compound_state *);
190 static void	lock_denied_free(nfs_resop4 *);
191 static void	rfs4_op_locku(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
192 			struct compound_state *);
193 static void	rfs4_op_lockt(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
194 			struct compound_state *);
195 static void	rfs4_op_lookup(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
196 			struct compound_state *);
197 static void	rfs4_op_lookupp(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
198 			struct compound_state *);
199 static void	rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop,
200 				struct svc_req *req, struct compound_state *cs);
201 static void	rfs4_op_nverify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
202 			struct compound_state *);
203 static void	rfs4_op_open(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
204 			struct compound_state *);
205 static void	rfs4_op_open_confirm(nfs_argop4 *, nfs_resop4 *,
206 			struct svc_req *, struct compound_state *);
207 static void	rfs4_op_open_downgrade(nfs_argop4 *, nfs_resop4 *,
208 			struct svc_req *, struct compound_state *);
209 static void	rfs4_op_putfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
210 			struct compound_state *);
211 static void	rfs4_op_putpubfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
212 			struct compound_state *);
213 static void	rfs4_op_putrootfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
214 			struct compound_state *);
215 static void	rfs4_op_read(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
216 			struct compound_state *);
217 static void	rfs4_op_read_free(nfs_resop4 *);
218 static void	rfs4_op_readdir_free(nfs_resop4 *resop);
219 static void	rfs4_op_readlink(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
220 			struct compound_state *);
221 static void	rfs4_op_readlink_free(nfs_resop4 *);
222 static void	rfs4_op_release_lockowner(nfs_argop4 *, nfs_resop4 *,
223 			struct svc_req *, struct compound_state *);
224 static void	rfs4_op_remove(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
225 			struct compound_state *);
226 static void	rfs4_op_rename(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
227 			struct compound_state *);
228 static void	rfs4_op_renew(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
229 			struct compound_state *);
230 static void	rfs4_op_restorefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
231 			struct compound_state *);
232 static void	rfs4_op_savefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
233 			struct compound_state *);
234 static void	rfs4_op_setattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
235 			struct compound_state *);
236 static void	rfs4_op_verify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
237 			struct compound_state *);
238 static void	rfs4_op_write(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
239 			struct compound_state *);
240 static void	rfs4_op_setclientid(nfs_argop4 *, nfs_resop4 *,
241 			struct svc_req *, struct compound_state *);
242 static void	rfs4_op_setclientid_confirm(nfs_argop4 *, nfs_resop4 *,
243 			struct svc_req *req, struct compound_state *);
244 static void	rfs4_op_secinfo(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
245 			struct compound_state *);
246 static void	rfs4_op_secinfo_free(nfs_resop4 *);
247 
248 static nfsstat4 check_open_access(uint32_t,
249 				struct compound_state *, struct svc_req *);
250 nfsstat4 rfs4_client_sysid(rfs4_client_t *, sysid_t *);
251 void rfs4_ss_clid(rfs4_client_t *);
252 
253 /*
254  * translation table for attrs
255  */
256 struct nfs4_ntov_table {
257 	union nfs4_attr_u *na;
258 	uint8_t amap[NFS4_MAXNUM_ATTRS];
259 	int attrcnt;
260 	bool_t vfsstat;
261 };
262 
263 static void	nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp);
264 static void	nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
265 				    struct nfs4_svgetit_arg *sargp);
266 
267 static nfsstat4	do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp,
268 		    struct compound_state *cs, struct nfs4_svgetit_arg *sargp,
269 		    struct nfs4_ntov_table *ntovp, nfs4_attr_cmd_t cmd);
270 
271 fem_t		*deleg_rdops;
272 fem_t		*deleg_wrops;
273 
274 rfs4_servinst_t *rfs4_cur_servinst = NULL;	/* current server instance */
275 kmutex_t	rfs4_servinst_lock;	/* protects linked list */
276 int		rfs4_seen_first_compound;	/* set first time we see one */
277 
278 /*
279  * NFS4 op dispatch table
280  */
281 
282 struct rfsv4disp {
283 	void	(*dis_proc)();		/* proc to call */
284 	void	(*dis_resfree)();	/* frees space allocated by proc */
285 	int	dis_flags;		/* RPC_IDEMPOTENT, etc... */
286 };
287 
288 static struct rfsv4disp rfsv4disptab[] = {
289 	/*
290 	 * NFS VERSION 4
291 	 */
292 
293 	/* RFS_NULL = 0 */
294 	{rfs4_op_illegal, nullfree, 0},
295 
296 	/* UNUSED = 1 */
297 	{rfs4_op_illegal, nullfree, 0},
298 
299 	/* UNUSED = 2 */
300 	{rfs4_op_illegal, nullfree, 0},
301 
302 	/* OP_ACCESS = 3 */
303 	{rfs4_op_access, nullfree, RPC_IDEMPOTENT},
304 
305 	/* OP_CLOSE = 4 */
306 	{rfs4_op_close, nullfree, 0},
307 
308 	/* OP_COMMIT = 5 */
309 	{rfs4_op_commit, nullfree, RPC_IDEMPOTENT},
310 
311 	/* OP_CREATE = 6 */
312 	{rfs4_op_create, nullfree, 0},
313 
314 	/* OP_DELEGPURGE = 7 */
315 	{rfs4_op_delegpurge, nullfree, 0},
316 
317 	/* OP_DELEGRETURN = 8 */
318 	{rfs4_op_delegreturn, nullfree, 0},
319 
320 	/* OP_GETATTR = 9 */
321 	{rfs4_op_getattr, rfs4_op_getattr_free, RPC_IDEMPOTENT},
322 
323 	/* OP_GETFH = 10 */
324 	{rfs4_op_getfh, rfs4_op_getfh_free, RPC_ALL},
325 
326 	/* OP_LINK = 11 */
327 	{rfs4_op_link, nullfree, 0},
328 
329 	/* OP_LOCK = 12 */
330 	{rfs4_op_lock, lock_denied_free, 0},
331 
332 	/* OP_LOCKT = 13 */
333 	{rfs4_op_lockt, lock_denied_free, 0},
334 
335 	/* OP_LOCKU = 14 */
336 	{rfs4_op_locku, nullfree, 0},
337 
338 	/* OP_LOOKUP = 15 */
339 	{rfs4_op_lookup, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
340 
341 	/* OP_LOOKUPP = 16 */
342 	{rfs4_op_lookupp, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
343 
344 	/* OP_NVERIFY = 17 */
345 	{rfs4_op_nverify, nullfree, RPC_IDEMPOTENT},
346 
347 	/* OP_OPEN = 18 */
348 	{rfs4_op_open, rfs4_free_reply, 0},
349 
350 	/* OP_OPENATTR = 19 */
351 	{rfs4_op_openattr, nullfree, 0},
352 
353 	/* OP_OPEN_CONFIRM = 20 */
354 	{rfs4_op_open_confirm, nullfree, 0},
355 
356 	/* OP_OPEN_DOWNGRADE = 21 */
357 	{rfs4_op_open_downgrade, nullfree, 0},
358 
359 	/* OP_OPEN_PUTFH = 22 */
360 	{rfs4_op_putfh, nullfree, RPC_ALL},
361 
362 	/* OP_PUTPUBFH = 23 */
363 	{rfs4_op_putpubfh, nullfree, RPC_ALL},
364 
365 	/* OP_PUTROOTFH = 24 */
366 	{rfs4_op_putrootfh, nullfree, RPC_ALL},
367 
368 	/* OP_READ = 25 */
369 	{rfs4_op_read, rfs4_op_read_free, RPC_IDEMPOTENT},
370 
371 	/* OP_READDIR = 26 */
372 	{rfs4_op_readdir, rfs4_op_readdir_free, RPC_IDEMPOTENT},
373 
374 	/* OP_READLINK = 27 */
375 	{rfs4_op_readlink, rfs4_op_readlink_free, RPC_IDEMPOTENT},
376 
377 	/* OP_REMOVE = 28 */
378 	{rfs4_op_remove, nullfree, 0},
379 
380 	/* OP_RENAME = 29 */
381 	{rfs4_op_rename, nullfree, 0},
382 
383 	/* OP_RENEW = 30 */
384 	{rfs4_op_renew, nullfree, 0},
385 
386 	/* OP_RESTOREFH = 31 */
387 	{rfs4_op_restorefh, nullfree, RPC_ALL},
388 
389 	/* OP_SAVEFH = 32 */
390 	{rfs4_op_savefh, nullfree, RPC_ALL},
391 
392 	/* OP_SECINFO = 33 */
393 	{rfs4_op_secinfo, rfs4_op_secinfo_free, 0},
394 
395 	/* OP_SETATTR = 34 */
396 	{rfs4_op_setattr, nullfree, 0},
397 
398 	/* OP_SETCLIENTID = 35 */
399 	{rfs4_op_setclientid, nullfree, 0},
400 
401 	/* OP_SETCLIENTID_CONFIRM = 36 */
402 	{rfs4_op_setclientid_confirm, nullfree, 0},
403 
404 	/* OP_VERIFY = 37 */
405 	{rfs4_op_verify, nullfree, RPC_IDEMPOTENT},
406 
407 	/* OP_WRITE = 38 */
408 	{rfs4_op_write, nullfree, 0},
409 
410 	/* OP_RELEASE_LOCKOWNER = 39 */
411 	{rfs4_op_release_lockowner, nullfree, 0},
412 };
413 
414 static uint_t rfsv4disp_cnt = sizeof (rfsv4disptab) / sizeof (rfsv4disptab[0]);
415 
416 #define	OP_ILLEGAL_IDX (rfsv4disp_cnt)
417 
418 #ifdef DEBUG
419 
420 int		rfs4_fillone_debug = 0;
421 int		rfs4_no_stub_access = 1;
422 int		rfs4_rddir_debug = 0;
423 
424 static char    *rfs4_op_string[] = {
425 	"rfs4_op_null",
426 	"rfs4_op_1 unused",
427 	"rfs4_op_2 unused",
428 	"rfs4_op_access",
429 	"rfs4_op_close",
430 	"rfs4_op_commit",
431 	"rfs4_op_create",
432 	"rfs4_op_delegpurge",
433 	"rfs4_op_delegreturn",
434 	"rfs4_op_getattr",
435 	"rfs4_op_getfh",
436 	"rfs4_op_link",
437 	"rfs4_op_lock",
438 	"rfs4_op_lockt",
439 	"rfs4_op_locku",
440 	"rfs4_op_lookup",
441 	"rfs4_op_lookupp",
442 	"rfs4_op_nverify",
443 	"rfs4_op_open",
444 	"rfs4_op_openattr",
445 	"rfs4_op_open_confirm",
446 	"rfs4_op_open_downgrade",
447 	"rfs4_op_putfh",
448 	"rfs4_op_putpubfh",
449 	"rfs4_op_putrootfh",
450 	"rfs4_op_read",
451 	"rfs4_op_readdir",
452 	"rfs4_op_readlink",
453 	"rfs4_op_remove",
454 	"rfs4_op_rename",
455 	"rfs4_op_renew",
456 	"rfs4_op_restorefh",
457 	"rfs4_op_savefh",
458 	"rfs4_op_secinfo",
459 	"rfs4_op_setattr",
460 	"rfs4_op_setclientid",
461 	"rfs4_op_setclient_confirm",
462 	"rfs4_op_verify",
463 	"rfs4_op_write",
464 	"rfs4_op_release_lockowner",
465 	"rfs4_op_illegal"
466 };
467 #endif
468 
469 void	rfs4_ss_chkclid(rfs4_client_t *);
470 
471 extern size_t   strlcpy(char *dst, const char *src, size_t dstsize);
472 
473 extern void	rfs4_free_fs_locations4(fs_locations4 *);
474 
475 #ifdef	nextdp
476 #undef nextdp
477 #endif
478 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
479 
480 static const fs_operation_def_t nfs4_rd_deleg_tmpl[] = {
481 	VOPNAME_OPEN,		{ .femop_open = deleg_rd_open },
482 	VOPNAME_WRITE,		{ .femop_write = deleg_rd_write },
483 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_rd_setattr },
484 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_rd_rwlock },
485 	VOPNAME_SPACE,		{ .femop_space = deleg_rd_space },
486 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_rd_setsecattr },
487 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_rd_vnevent },
488 	NULL,			NULL
489 };
490 static const fs_operation_def_t nfs4_wr_deleg_tmpl[] = {
491 	VOPNAME_OPEN,		{ .femop_open = deleg_wr_open },
492 	VOPNAME_READ,		{ .femop_read = deleg_wr_read },
493 	VOPNAME_WRITE,		{ .femop_write = deleg_wr_write },
494 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_wr_setattr },
495 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_wr_rwlock },
496 	VOPNAME_SPACE,		{ .femop_space = deleg_wr_space },
497 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_wr_setsecattr },
498 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_wr_vnevent },
499 	NULL,			NULL
500 };
501 
502 int
503 rfs4_srvrinit(void)
504 {
505 	timespec32_t verf;
506 	int error;
507 	extern void rfs4_attr_init();
508 	extern krwlock_t rfs4_deleg_policy_lock;
509 
510 	/*
511 	 * The following algorithm attempts to find a unique verifier
512 	 * to be used as the write verifier returned from the server
513 	 * to the client.  It is important that this verifier change
514 	 * whenever the server reboots.  Of secondary importance, it
515 	 * is important for the verifier to be unique between two
516 	 * different servers.
517 	 *
518 	 * Thus, an attempt is made to use the system hostid and the
519 	 * current time in seconds when the nfssrv kernel module is
520 	 * loaded.  It is assumed that an NFS server will not be able
521 	 * to boot and then to reboot in less than a second.  If the
522 	 * hostid has not been set, then the current high resolution
523 	 * time is used.  This will ensure different verifiers each
524 	 * time the server reboots and minimize the chances that two
525 	 * different servers will have the same verifier.
526 	 * XXX - this is broken on LP64 kernels.
527 	 */
528 	verf.tv_sec = (time_t)zone_get_hostid(NULL);
529 	if (verf.tv_sec != 0) {
530 		verf.tv_nsec = gethrestime_sec();
531 	} else {
532 		timespec_t tverf;
533 
534 		gethrestime(&tverf);
535 		verf.tv_sec = (time_t)tverf.tv_sec;
536 		verf.tv_nsec = tverf.tv_nsec;
537 	}
538 
539 	Write4verf = *(uint64_t *)&verf;
540 
541 	rfs4_attr_init();
542 	mutex_init(&rfs4_deleg_lock, NULL, MUTEX_DEFAULT, NULL);
543 
544 	/* Used to manage create/destroy of server state */
545 	mutex_init(&rfs4_state_lock, NULL, MUTEX_DEFAULT, NULL);
546 
547 	/* Used to manage access to server instance linked list */
548 	mutex_init(&rfs4_servinst_lock, NULL, MUTEX_DEFAULT, NULL);
549 
550 	/* Used to manage access to rfs4_deleg_policy */
551 	rw_init(&rfs4_deleg_policy_lock, NULL, RW_DEFAULT, NULL);
552 
553 	error = fem_create("deleg_rdops", nfs4_rd_deleg_tmpl, &deleg_rdops);
554 	if (error != 0) {
555 		rfs4_disable_delegation();
556 	} else {
557 		error = fem_create("deleg_wrops", nfs4_wr_deleg_tmpl,
558 		    &deleg_wrops);
559 		if (error != 0) {
560 			rfs4_disable_delegation();
561 			fem_free(deleg_rdops);
562 		}
563 	}
564 
565 	nfs4_srv_caller_id = fs_new_caller_id();
566 
567 	lockt_sysid = lm_alloc_sysidt();
568 
569 	vsd_create(&nfs4_srv_vkey, NULL);
570 
571 	return (0);
572 }
573 
574 void
575 rfs4_srvrfini(void)
576 {
577 	extern krwlock_t rfs4_deleg_policy_lock;
578 
579 	if (lockt_sysid != LM_NOSYSID) {
580 		lm_free_sysidt(lockt_sysid);
581 		lockt_sysid = LM_NOSYSID;
582 	}
583 
584 	mutex_destroy(&rfs4_deleg_lock);
585 	mutex_destroy(&rfs4_state_lock);
586 	rw_destroy(&rfs4_deleg_policy_lock);
587 
588 	fem_free(deleg_rdops);
589 	fem_free(deleg_wrops);
590 }
591 
592 void
593 rfs4_init_compound_state(struct compound_state *cs)
594 {
595 	bzero(cs, sizeof (*cs));
596 	cs->cont = TRUE;
597 	cs->access = CS_ACCESS_DENIED;
598 	cs->deleg = FALSE;
599 	cs->mandlock = FALSE;
600 	cs->fh.nfs_fh4_val = cs->fhbuf;
601 }
602 
603 void
604 rfs4_grace_start(rfs4_servinst_t *sip)
605 {
606 	rw_enter(&sip->rwlock, RW_WRITER);
607 	sip->start_time = (time_t)TICK_TO_SEC(ddi_get_lbolt());
608 	sip->grace_period = rfs4_grace_period;
609 	rw_exit(&sip->rwlock);
610 }
611 
612 /*
613  * returns true if the instance's grace period has never been started
614  */
615 int
616 rfs4_servinst_grace_new(rfs4_servinst_t *sip)
617 {
618 	time_t start_time;
619 
620 	rw_enter(&sip->rwlock, RW_READER);
621 	start_time = sip->start_time;
622 	rw_exit(&sip->rwlock);
623 
624 	return (start_time == 0);
625 }
626 
627 /*
628  * Indicates if server instance is within the
629  * grace period.
630  */
631 int
632 rfs4_servinst_in_grace(rfs4_servinst_t *sip)
633 {
634 	time_t grace_expiry;
635 
636 	rw_enter(&sip->rwlock, RW_READER);
637 	grace_expiry = sip->start_time + sip->grace_period;
638 	rw_exit(&sip->rwlock);
639 
640 	return (((time_t)TICK_TO_SEC(ddi_get_lbolt())) < grace_expiry);
641 }
642 
643 int
644 rfs4_clnt_in_grace(rfs4_client_t *cp)
645 {
646 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
647 
648 	return (rfs4_servinst_in_grace(cp->rc_server_instance));
649 }
650 
651 /*
652  * reset all currently active grace periods
653  */
654 void
655 rfs4_grace_reset_all(void)
656 {
657 	rfs4_servinst_t *sip;
658 
659 	mutex_enter(&rfs4_servinst_lock);
660 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
661 		if (rfs4_servinst_in_grace(sip))
662 			rfs4_grace_start(sip);
663 	mutex_exit(&rfs4_servinst_lock);
664 }
665 
666 /*
667  * start any new instances' grace periods
668  */
669 void
670 rfs4_grace_start_new(void)
671 {
672 	rfs4_servinst_t *sip;
673 
674 	mutex_enter(&rfs4_servinst_lock);
675 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
676 		if (rfs4_servinst_grace_new(sip))
677 			rfs4_grace_start(sip);
678 	mutex_exit(&rfs4_servinst_lock);
679 }
680 
681 static rfs4_dss_path_t *
682 rfs4_dss_newpath(rfs4_servinst_t *sip, char *path, unsigned index)
683 {
684 	size_t len;
685 	rfs4_dss_path_t *dss_path;
686 
687 	dss_path = kmem_alloc(sizeof (rfs4_dss_path_t), KM_SLEEP);
688 
689 	/*
690 	 * Take a copy of the string, since the original may be overwritten.
691 	 * Sadly, no strdup() in the kernel.
692 	 */
693 	/* allow for NUL */
694 	len = strlen(path) + 1;
695 	dss_path->path = kmem_alloc(len, KM_SLEEP);
696 	(void) strlcpy(dss_path->path, path, len);
697 
698 	/* associate with servinst */
699 	dss_path->sip = sip;
700 	dss_path->index = index;
701 
702 	/*
703 	 * Add to list of served paths.
704 	 * No locking required, as we're only ever called at startup.
705 	 */
706 	if (rfs4_dss_pathlist == NULL) {
707 		/* this is the first dss_path_t */
708 
709 		/* needed for insque/remque */
710 		dss_path->next = dss_path->prev = dss_path;
711 
712 		rfs4_dss_pathlist = dss_path;
713 	} else {
714 		insque(dss_path, rfs4_dss_pathlist);
715 	}
716 
717 	return (dss_path);
718 }
719 
720 /*
721  * Create a new server instance, and make it the currently active instance.
722  * Note that starting the grace period too early will reduce the clients'
723  * recovery window.
724  */
725 void
726 rfs4_servinst_create(int start_grace, int dss_npaths, char **dss_paths)
727 {
728 	unsigned i;
729 	rfs4_servinst_t *sip;
730 	rfs4_oldstate_t *oldstate;
731 
732 	sip = kmem_alloc(sizeof (rfs4_servinst_t), KM_SLEEP);
733 	rw_init(&sip->rwlock, NULL, RW_DEFAULT, NULL);
734 
735 	sip->start_time = (time_t)0;
736 	sip->grace_period = (time_t)0;
737 	sip->next = NULL;
738 	sip->prev = NULL;
739 
740 	rw_init(&sip->oldstate_lock, NULL, RW_DEFAULT, NULL);
741 	/*
742 	 * This initial dummy entry is required to setup for insque/remque.
743 	 * It must be skipped over whenever the list is traversed.
744 	 */
745 	oldstate = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
746 	/* insque/remque require initial list entry to be self-terminated */
747 	oldstate->next = oldstate;
748 	oldstate->prev = oldstate;
749 	sip->oldstate = oldstate;
750 
751 
752 	sip->dss_npaths = dss_npaths;
753 	sip->dss_paths = kmem_alloc(dss_npaths *
754 	    sizeof (rfs4_dss_path_t *), KM_SLEEP);
755 
756 	for (i = 0; i < dss_npaths; i++) {
757 		sip->dss_paths[i] = rfs4_dss_newpath(sip, dss_paths[i], i);
758 	}
759 
760 	mutex_enter(&rfs4_servinst_lock);
761 	if (rfs4_cur_servinst != NULL) {
762 		/* add to linked list */
763 		sip->prev = rfs4_cur_servinst;
764 		rfs4_cur_servinst->next = sip;
765 	}
766 	if (start_grace)
767 		rfs4_grace_start(sip);
768 	/* make the new instance "current" */
769 	rfs4_cur_servinst = sip;
770 
771 	mutex_exit(&rfs4_servinst_lock);
772 }
773 
774 /*
775  * In future, we might add a rfs4_servinst_destroy(sip) but, for now, destroy
776  * all instances directly.
777  */
778 void
779 rfs4_servinst_destroy_all(void)
780 {
781 	rfs4_servinst_t *sip, *prev, *current;
782 #ifdef DEBUG
783 	int n = 0;
784 #endif
785 
786 	mutex_enter(&rfs4_servinst_lock);
787 	ASSERT(rfs4_cur_servinst != NULL);
788 	current = rfs4_cur_servinst;
789 	rfs4_cur_servinst = NULL;
790 	for (sip = current; sip != NULL; sip = prev) {
791 		prev = sip->prev;
792 		rw_destroy(&sip->rwlock);
793 		if (sip->oldstate)
794 			kmem_free(sip->oldstate, sizeof (rfs4_oldstate_t));
795 		if (sip->dss_paths)
796 			kmem_free(sip->dss_paths,
797 			    sip->dss_npaths * sizeof (rfs4_dss_path_t *));
798 		kmem_free(sip, sizeof (rfs4_servinst_t));
799 #ifdef DEBUG
800 		n++;
801 #endif
802 	}
803 	mutex_exit(&rfs4_servinst_lock);
804 }
805 
806 /*
807  * Assign the current server instance to a client_t.
808  * Should be called with cp->rc_dbe held.
809  */
810 void
811 rfs4_servinst_assign(rfs4_client_t *cp, rfs4_servinst_t *sip)
812 {
813 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
814 
815 	/*
816 	 * The lock ensures that if the current instance is in the process
817 	 * of changing, we will see the new one.
818 	 */
819 	mutex_enter(&rfs4_servinst_lock);
820 	cp->rc_server_instance = sip;
821 	mutex_exit(&rfs4_servinst_lock);
822 }
823 
824 rfs4_servinst_t *
825 rfs4_servinst(rfs4_client_t *cp)
826 {
827 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
828 
829 	return (cp->rc_server_instance);
830 }
831 
832 /* ARGSUSED */
833 static void
834 nullfree(caddr_t resop)
835 {
836 }
837 
838 /*
839  * This is a fall-through for invalid or not implemented (yet) ops
840  */
841 /* ARGSUSED */
842 static void
843 rfs4_op_inval(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
844     struct compound_state *cs)
845 {
846 	*cs->statusp = *((nfsstat4 *)&(resop)->nfs_resop4_u) = NFS4ERR_INVAL;
847 }
848 
849 /*
850  * Check if the security flavor, nfsnum, is in the flavor_list.
851  */
852 bool_t
853 in_flavor_list(int nfsnum, int *flavor_list, int count)
854 {
855 	int i;
856 
857 	for (i = 0; i < count; i++) {
858 		if (nfsnum == flavor_list[i])
859 			return (TRUE);
860 	}
861 	return (FALSE);
862 }
863 
864 /*
865  * Used by rfs4_op_secinfo to get the security information from the
866  * export structure associated with the component.
867  */
868 /* ARGSUSED */
869 static nfsstat4
870 do_rfs4_op_secinfo(struct compound_state *cs, char *nm, SECINFO4res *resp)
871 {
872 	int error, different_export = 0;
873 	vnode_t *dvp, *vp;
874 	struct exportinfo *exi = NULL;
875 	fid_t fid;
876 	uint_t count, i;
877 	secinfo4 *resok_val;
878 	struct secinfo *secp;
879 	seconfig_t *si;
880 	bool_t did_traverse = FALSE;
881 	int dotdot, walk;
882 
883 	dvp = cs->vp;
884 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
885 
886 	/*
887 	 * If dotdotting, then need to check whether it's above the
888 	 * root of a filesystem, or above an export point.
889 	 */
890 	if (dotdot) {
891 
892 		/*
893 		 * If dotdotting at the root of a filesystem, then
894 		 * need to traverse back to the mounted-on filesystem
895 		 * and do the dotdot lookup there.
896 		 */
897 		if (cs->vp->v_flag & VROOT) {
898 
899 			/*
900 			 * If at the system root, then can
901 			 * go up no further.
902 			 */
903 			if (VN_CMP(dvp, rootdir))
904 				return (puterrno4(ENOENT));
905 
906 			/*
907 			 * Traverse back to the mounted-on filesystem
908 			 */
909 			dvp = untraverse(cs->vp);
910 
911 			/*
912 			 * Set the different_export flag so we remember
913 			 * to pick up a new exportinfo entry for
914 			 * this new filesystem.
915 			 */
916 			different_export = 1;
917 		} else {
918 
919 			/*
920 			 * If dotdotting above an export point then set
921 			 * the different_export to get new export info.
922 			 */
923 			different_export = nfs_exported(cs->exi, cs->vp);
924 		}
925 	}
926 
927 	/*
928 	 * Get the vnode for the component "nm".
929 	 */
930 	error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cs->cr,
931 	    NULL, NULL, NULL);
932 	if (error)
933 		return (puterrno4(error));
934 
935 	/*
936 	 * If the vnode is in a pseudo filesystem, or if the security flavor
937 	 * used in the request is valid but not an explicitly shared flavor,
938 	 * or the access bit indicates that this is a limited access,
939 	 * check whether this vnode is visible.
940 	 */
941 	if (!different_export &&
942 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
943 	    cs->access & CS_ACCESS_LIMITED)) {
944 		if (! nfs_visible(cs->exi, vp, &different_export)) {
945 			VN_RELE(vp);
946 			return (puterrno4(ENOENT));
947 		}
948 	}
949 
950 	/*
951 	 * If it's a mountpoint, then traverse it.
952 	 */
953 	if (vn_ismntpt(vp)) {
954 		if ((error = traverse(&vp)) != 0) {
955 			VN_RELE(vp);
956 			return (puterrno4(error));
957 		}
958 		/* remember that we had to traverse mountpoint */
959 		did_traverse = TRUE;
960 		different_export = 1;
961 	} else if (vp->v_vfsp != dvp->v_vfsp) {
962 		/*
963 		 * If vp isn't a mountpoint and the vfs ptrs aren't the same,
964 		 * then vp is probably an LOFS object.  We don't need the
965 		 * realvp, we just need to know that we might have crossed
966 		 * a server fs boundary and need to call checkexport4.
967 		 * (LOFS lookup hides server fs mountpoints, and actually calls
968 		 * traverse)
969 		 */
970 		different_export = 1;
971 	}
972 
973 	/*
974 	 * Get the export information for it.
975 	 */
976 	if (different_export) {
977 
978 		bzero(&fid, sizeof (fid));
979 		fid.fid_len = MAXFIDSZ;
980 		error = vop_fid_pseudo(vp, &fid);
981 		if (error) {
982 			VN_RELE(vp);
983 			return (puterrno4(error));
984 		}
985 
986 		if (dotdot)
987 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
988 		else
989 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
990 
991 		if (exi == NULL) {
992 			if (did_traverse == TRUE) {
993 				/*
994 				 * If this vnode is a mounted-on vnode,
995 				 * but the mounted-on file system is not
996 				 * exported, send back the secinfo for
997 				 * the exported node that the mounted-on
998 				 * vnode lives in.
999 				 */
1000 				exi = cs->exi;
1001 			} else {
1002 				VN_RELE(vp);
1003 				return (puterrno4(EACCES));
1004 			}
1005 		}
1006 	} else {
1007 		exi = cs->exi;
1008 	}
1009 	ASSERT(exi != NULL);
1010 
1011 
1012 	/*
1013 	 * Create the secinfo result based on the security information
1014 	 * from the exportinfo structure (exi).
1015 	 *
1016 	 * Return all flavors for a pseudo node.
1017 	 * For a real export node, return the flavor that the client
1018 	 * has access with.
1019 	 */
1020 	ASSERT(RW_LOCK_HELD(&exported_lock));
1021 	if (PSEUDO(exi)) {
1022 		count = exi->exi_export.ex_seccnt; /* total sec count */
1023 		resok_val = kmem_alloc(count * sizeof (secinfo4), KM_SLEEP);
1024 		secp = exi->exi_export.ex_secinfo;
1025 
1026 		for (i = 0; i < count; i++) {
1027 			si = &secp[i].s_secinfo;
1028 			resok_val[i].flavor = si->sc_rpcnum;
1029 			if (resok_val[i].flavor == RPCSEC_GSS) {
1030 				rpcsec_gss_info *info;
1031 
1032 				info = &resok_val[i].flavor_info;
1033 				info->qop = si->sc_qop;
1034 				info->service = (rpc_gss_svc_t)si->sc_service;
1035 
1036 				/* get oid opaque data */
1037 				info->oid.sec_oid4_len =
1038 				    si->sc_gss_mech_type->length;
1039 				info->oid.sec_oid4_val = kmem_alloc(
1040 				    si->sc_gss_mech_type->length, KM_SLEEP);
1041 				bcopy(
1042 				    si->sc_gss_mech_type->elements,
1043 				    info->oid.sec_oid4_val,
1044 				    info->oid.sec_oid4_len);
1045 			}
1046 		}
1047 		resp->SECINFO4resok_len = count;
1048 		resp->SECINFO4resok_val = resok_val;
1049 	} else {
1050 		int ret_cnt = 0, k = 0;
1051 		int *flavor_list;
1052 
1053 		count = exi->exi_export.ex_seccnt; /* total sec count */
1054 		secp = exi->exi_export.ex_secinfo;
1055 
1056 		flavor_list = kmem_alloc(count * sizeof (int), KM_SLEEP);
1057 		/* find out which flavors to return */
1058 		for (i = 0; i < count; i ++) {
1059 			int access, flavor, perm;
1060 
1061 			flavor = secp[i].s_secinfo.sc_nfsnum;
1062 			perm = secp[i].s_flags;
1063 
1064 			access = nfsauth4_secinfo_access(exi, cs->req,
1065 			    flavor, perm, cs->basecr);
1066 
1067 			if (! (access & NFSAUTH_DENIED) &&
1068 			    ! (access & NFSAUTH_WRONGSEC)) {
1069 				flavor_list[ret_cnt] = flavor;
1070 				ret_cnt++;
1071 			}
1072 		}
1073 
1074 		/* Create the returning SECINFO value */
1075 		resok_val = kmem_alloc(ret_cnt * sizeof (secinfo4), KM_SLEEP);
1076 
1077 		for (i = 0; i < count; i++) {
1078 			/*
1079 			 * If the flavor is in the flavor list,
1080 			 * fill in resok_val.
1081 			 */
1082 			si = &secp[i].s_secinfo;
1083 			if (in_flavor_list(si->sc_nfsnum,
1084 			    flavor_list, ret_cnt)) {
1085 				resok_val[k].flavor = si->sc_rpcnum;
1086 				if (resok_val[k].flavor == RPCSEC_GSS) {
1087 					rpcsec_gss_info *info;
1088 
1089 					info = &resok_val[k].flavor_info;
1090 					info->qop = si->sc_qop;
1091 					info->service = (rpc_gss_svc_t)
1092 					    si->sc_service;
1093 
1094 					/* get oid opaque data */
1095 					info->oid.sec_oid4_len =
1096 					    si->sc_gss_mech_type->length;
1097 					info->oid.sec_oid4_val = kmem_alloc(
1098 					    si->sc_gss_mech_type->length,
1099 					    KM_SLEEP);
1100 					bcopy(si->sc_gss_mech_type->elements,
1101 					    info->oid.sec_oid4_val,
1102 					    info->oid.sec_oid4_len);
1103 				}
1104 				k++;
1105 			}
1106 			if (k >= ret_cnt)
1107 				break;
1108 		}
1109 		resp->SECINFO4resok_len = ret_cnt;
1110 		resp->SECINFO4resok_val = resok_val;
1111 		kmem_free(flavor_list, count * sizeof (int));
1112 	}
1113 
1114 	VN_RELE(vp);
1115 	return (NFS4_OK);
1116 }
1117 
1118 /*
1119  * SECINFO (Operation 33): Obtain required security information on
1120  * the component name in the format of (security-mechanism-oid, qop, service)
1121  * triplets.
1122  */
1123 /* ARGSUSED */
1124 static void
1125 rfs4_op_secinfo(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1126     struct compound_state *cs)
1127 {
1128 	SECINFO4args *args = &argop->nfs_argop4_u.opsecinfo;
1129 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1130 	utf8string *utfnm = &args->name;
1131 	uint_t len;
1132 	char *nm;
1133 	struct sockaddr *ca;
1134 	char *name = NULL;
1135 	nfsstat4 status = NFS4_OK;
1136 
1137 	DTRACE_NFSV4_2(op__secinfo__start, struct compound_state *, cs,
1138 	    SECINFO4args *, args);
1139 
1140 	/*
1141 	 * Current file handle (cfh) should have been set before getting
1142 	 * into this function. If not, return error.
1143 	 */
1144 	if (cs->vp == NULL) {
1145 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1146 		goto out;
1147 	}
1148 
1149 	if (cs->vp->v_type != VDIR) {
1150 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1151 		goto out;
1152 	}
1153 
1154 	/*
1155 	 * Verify the component name. If failed, error out, but
1156 	 * do not error out if the component name is a "..".
1157 	 * SECINFO will return its parents secinfo data for SECINFO "..".
1158 	 */
1159 	status = utf8_dir_verify(utfnm);
1160 	if (status != NFS4_OK) {
1161 		if (utfnm->utf8string_len != 2 ||
1162 		    utfnm->utf8string_val[0] != '.' ||
1163 		    utfnm->utf8string_val[1] != '.') {
1164 			*cs->statusp = resp->status = status;
1165 			goto out;
1166 		}
1167 	}
1168 
1169 	nm = utf8_to_str(utfnm, &len, NULL);
1170 	if (nm == NULL) {
1171 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1172 		goto out;
1173 	}
1174 
1175 	if (len > MAXNAMELEN) {
1176 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1177 		kmem_free(nm, len);
1178 		goto out;
1179 	}
1180 
1181 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1182 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1183 	    MAXPATHLEN  + 1);
1184 
1185 	if (name == NULL) {
1186 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1187 		kmem_free(nm, len);
1188 		goto out;
1189 	}
1190 
1191 
1192 	*cs->statusp = resp->status = do_rfs4_op_secinfo(cs, name, resp);
1193 
1194 	if (name != nm)
1195 		kmem_free(name, MAXPATHLEN + 1);
1196 	kmem_free(nm, len);
1197 
1198 out:
1199 	DTRACE_NFSV4_2(op__secinfo__done, struct compound_state *, cs,
1200 	    SECINFO4res *, resp);
1201 }
1202 
1203 /*
1204  * Free SECINFO result.
1205  */
1206 /* ARGSUSED */
1207 static void
1208 rfs4_op_secinfo_free(nfs_resop4 *resop)
1209 {
1210 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1211 	int count, i;
1212 	secinfo4 *resok_val;
1213 
1214 	/* If this is not an Ok result, nothing to free. */
1215 	if (resp->status != NFS4_OK) {
1216 		return;
1217 	}
1218 
1219 	count = resp->SECINFO4resok_len;
1220 	resok_val = resp->SECINFO4resok_val;
1221 
1222 	for (i = 0; i < count; i++) {
1223 		if (resok_val[i].flavor == RPCSEC_GSS) {
1224 			rpcsec_gss_info *info;
1225 
1226 			info = &resok_val[i].flavor_info;
1227 			kmem_free(info->oid.sec_oid4_val,
1228 			    info->oid.sec_oid4_len);
1229 		}
1230 	}
1231 	kmem_free(resok_val, count * sizeof (secinfo4));
1232 	resp->SECINFO4resok_len = 0;
1233 	resp->SECINFO4resok_val = NULL;
1234 }
1235 
1236 /* ARGSUSED */
1237 static void
1238 rfs4_op_access(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1239     struct compound_state *cs)
1240 {
1241 	ACCESS4args *args = &argop->nfs_argop4_u.opaccess;
1242 	ACCESS4res *resp = &resop->nfs_resop4_u.opaccess;
1243 	int error;
1244 	vnode_t *vp;
1245 	struct vattr va;
1246 	int checkwriteperm;
1247 	cred_t *cr = cs->cr;
1248 	bslabel_t *clabel, *slabel;
1249 	ts_label_t *tslabel;
1250 	boolean_t admin_low_client;
1251 
1252 	DTRACE_NFSV4_2(op__access__start, struct compound_state *, cs,
1253 	    ACCESS4args *, args);
1254 
1255 #if 0	/* XXX allow access even if !cs->access. Eventually only pseudo fs */
1256 	if (cs->access == CS_ACCESS_DENIED) {
1257 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1258 		goto out;
1259 	}
1260 #endif
1261 	if (cs->vp == NULL) {
1262 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1263 		goto out;
1264 	}
1265 
1266 	ASSERT(cr != NULL);
1267 
1268 	vp = cs->vp;
1269 
1270 	/*
1271 	 * If the file system is exported read only, it is not appropriate
1272 	 * to check write permissions for regular files and directories.
1273 	 * Special files are interpreted by the client, so the underlying
1274 	 * permissions are sent back to the client for interpretation.
1275 	 */
1276 	if (rdonly4(req, cs) &&
1277 	    (vp->v_type == VREG || vp->v_type == VDIR))
1278 		checkwriteperm = 0;
1279 	else
1280 		checkwriteperm = 1;
1281 
1282 	/*
1283 	 * XXX
1284 	 * We need the mode so that we can correctly determine access
1285 	 * permissions relative to a mandatory lock file.  Access to
1286 	 * mandatory lock files is denied on the server, so it might
1287 	 * as well be reflected to the server during the open.
1288 	 */
1289 	va.va_mask = AT_MODE;
1290 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1291 	if (error) {
1292 		*cs->statusp = resp->status = puterrno4(error);
1293 		goto out;
1294 	}
1295 	resp->access = 0;
1296 	resp->supported = 0;
1297 
1298 	if (is_system_labeled()) {
1299 		ASSERT(req->rq_label != NULL);
1300 		clabel = req->rq_label;
1301 		DTRACE_PROBE2(tx__rfs4__log__info__opaccess__clabel, char *,
1302 		    "got client label from request(1)",
1303 		    struct svc_req *, req);
1304 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
1305 			if ((tslabel = nfs_getflabel(vp, cs->exi)) == NULL) {
1306 				*cs->statusp = resp->status = puterrno4(EACCES);
1307 				goto out;
1308 			}
1309 			slabel = label2bslabel(tslabel);
1310 			DTRACE_PROBE3(tx__rfs4__log__info__opaccess__slabel,
1311 			    char *, "got server label(1) for vp(2)",
1312 			    bslabel_t *, slabel, vnode_t *, vp);
1313 
1314 			admin_low_client = B_FALSE;
1315 		} else
1316 			admin_low_client = B_TRUE;
1317 	}
1318 
1319 	if (args->access & ACCESS4_READ) {
1320 		error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
1321 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1322 		    (!is_system_labeled() || admin_low_client ||
1323 		    bldominates(clabel, slabel)))
1324 			resp->access |= ACCESS4_READ;
1325 		resp->supported |= ACCESS4_READ;
1326 	}
1327 	if ((args->access & ACCESS4_LOOKUP) && vp->v_type == VDIR) {
1328 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1329 		if (!error && (!is_system_labeled() || admin_low_client ||
1330 		    bldominates(clabel, slabel)))
1331 			resp->access |= ACCESS4_LOOKUP;
1332 		resp->supported |= ACCESS4_LOOKUP;
1333 	}
1334 	if (checkwriteperm &&
1335 	    (args->access & (ACCESS4_MODIFY|ACCESS4_EXTEND))) {
1336 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1337 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1338 		    (!is_system_labeled() || admin_low_client ||
1339 		    blequal(clabel, slabel)))
1340 			resp->access |=
1341 			    (args->access & (ACCESS4_MODIFY | ACCESS4_EXTEND));
1342 		resp->supported |=
1343 		    resp->access & (ACCESS4_MODIFY | ACCESS4_EXTEND);
1344 	}
1345 
1346 	if (checkwriteperm &&
1347 	    (args->access & ACCESS4_DELETE) && vp->v_type == VDIR) {
1348 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1349 		if (!error && (!is_system_labeled() || admin_low_client ||
1350 		    blequal(clabel, slabel)))
1351 			resp->access |= ACCESS4_DELETE;
1352 		resp->supported |= ACCESS4_DELETE;
1353 	}
1354 	if (args->access & ACCESS4_EXECUTE && vp->v_type != VDIR) {
1355 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1356 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1357 		    (!is_system_labeled() || admin_low_client ||
1358 		    bldominates(clabel, slabel)))
1359 			resp->access |= ACCESS4_EXECUTE;
1360 		resp->supported |= ACCESS4_EXECUTE;
1361 	}
1362 
1363 	if (is_system_labeled() && !admin_low_client)
1364 		label_rele(tslabel);
1365 
1366 	*cs->statusp = resp->status = NFS4_OK;
1367 out:
1368 	DTRACE_NFSV4_2(op__access__done, struct compound_state *, cs,
1369 	    ACCESS4res *, resp);
1370 }
1371 
1372 /* ARGSUSED */
1373 static void
1374 rfs4_op_commit(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1375     struct compound_state *cs)
1376 {
1377 	COMMIT4args *args = &argop->nfs_argop4_u.opcommit;
1378 	COMMIT4res *resp = &resop->nfs_resop4_u.opcommit;
1379 	int error;
1380 	vnode_t *vp = cs->vp;
1381 	cred_t *cr = cs->cr;
1382 	vattr_t va;
1383 
1384 	DTRACE_NFSV4_2(op__commit__start, struct compound_state *, cs,
1385 	    COMMIT4args *, args);
1386 
1387 	if (vp == NULL) {
1388 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1389 		goto out;
1390 	}
1391 	if (cs->access == CS_ACCESS_DENIED) {
1392 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1393 		goto out;
1394 	}
1395 
1396 	if (args->offset + args->count < args->offset) {
1397 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1398 		goto out;
1399 	}
1400 
1401 	va.va_mask = AT_UID;
1402 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1403 
1404 	/*
1405 	 * If we can't get the attributes, then we can't do the
1406 	 * right access checking.  So, we'll fail the request.
1407 	 */
1408 	if (error) {
1409 		*cs->statusp = resp->status = puterrno4(error);
1410 		goto out;
1411 	}
1412 	if (rdonly4(req, cs)) {
1413 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1414 		goto out;
1415 	}
1416 
1417 	if (vp->v_type != VREG) {
1418 		if (vp->v_type == VDIR)
1419 			resp->status = NFS4ERR_ISDIR;
1420 		else
1421 			resp->status = NFS4ERR_INVAL;
1422 		*cs->statusp = resp->status;
1423 		goto out;
1424 	}
1425 
1426 	if (crgetuid(cr) != va.va_uid &&
1427 	    (error = VOP_ACCESS(vp, VWRITE, 0, cs->cr, NULL))) {
1428 		*cs->statusp = resp->status = puterrno4(error);
1429 		goto out;
1430 	}
1431 
1432 	error = VOP_FSYNC(vp, FSYNC, cr, NULL);
1433 
1434 	if (error) {
1435 		*cs->statusp = resp->status = puterrno4(error);
1436 		goto out;
1437 	}
1438 
1439 	*cs->statusp = resp->status = NFS4_OK;
1440 	resp->writeverf = Write4verf;
1441 out:
1442 	DTRACE_NFSV4_2(op__commit__done, struct compound_state *, cs,
1443 	    COMMIT4res *, resp);
1444 }
1445 
1446 /*
1447  * rfs4_op_mknod is called from rfs4_op_create after all initial verification
1448  * was completed. It does the nfsv4 create for special files.
1449  */
1450 /* ARGSUSED */
1451 static vnode_t *
1452 do_rfs4_op_mknod(CREATE4args *args, CREATE4res *resp, struct svc_req *req,
1453     struct compound_state *cs, vattr_t *vap, char *nm)
1454 {
1455 	int error;
1456 	cred_t *cr = cs->cr;
1457 	vnode_t *dvp = cs->vp;
1458 	vnode_t *vp = NULL;
1459 	int mode;
1460 	enum vcexcl excl;
1461 
1462 	switch (args->type) {
1463 	case NF4CHR:
1464 	case NF4BLK:
1465 		if (secpolicy_sys_devices(cr) != 0) {
1466 			*cs->statusp = resp->status = NFS4ERR_PERM;
1467 			return (NULL);
1468 		}
1469 		if (args->type == NF4CHR)
1470 			vap->va_type = VCHR;
1471 		else
1472 			vap->va_type = VBLK;
1473 		vap->va_rdev = makedevice(args->ftype4_u.devdata.specdata1,
1474 		    args->ftype4_u.devdata.specdata2);
1475 		vap->va_mask |= AT_RDEV;
1476 		break;
1477 	case NF4SOCK:
1478 		vap->va_type = VSOCK;
1479 		break;
1480 	case NF4FIFO:
1481 		vap->va_type = VFIFO;
1482 		break;
1483 	default:
1484 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1485 		return (NULL);
1486 	}
1487 
1488 	/*
1489 	 * Must specify the mode.
1490 	 */
1491 	if (!(vap->va_mask & AT_MODE)) {
1492 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1493 		return (NULL);
1494 	}
1495 
1496 	excl = EXCL;
1497 
1498 	mode = 0;
1499 
1500 	error = VOP_CREATE(dvp, nm, vap, excl, mode, &vp, cr, 0, NULL, NULL);
1501 	if (error) {
1502 		*cs->statusp = resp->status = puterrno4(error);
1503 		return (NULL);
1504 	}
1505 	return (vp);
1506 }
1507 
1508 /*
1509  * nfsv4 create is used to create non-regular files. For regular files,
1510  * use nfsv4 open.
1511  */
1512 /* ARGSUSED */
1513 static void
1514 rfs4_op_create(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1515     struct compound_state *cs)
1516 {
1517 	CREATE4args *args = &argop->nfs_argop4_u.opcreate;
1518 	CREATE4res *resp = &resop->nfs_resop4_u.opcreate;
1519 	int error;
1520 	struct vattr bva, iva, iva2, ava, *vap;
1521 	cred_t *cr = cs->cr;
1522 	vnode_t *dvp = cs->vp;
1523 	vnode_t *vp = NULL;
1524 	vnode_t *realvp;
1525 	char *nm, *lnm;
1526 	uint_t len, llen;
1527 	int syncval = 0;
1528 	struct nfs4_svgetit_arg sarg;
1529 	struct nfs4_ntov_table ntov;
1530 	struct statvfs64 sb;
1531 	nfsstat4 status;
1532 	struct sockaddr *ca;
1533 	char *name = NULL;
1534 	char *lname = NULL;
1535 
1536 	DTRACE_NFSV4_2(op__create__start, struct compound_state *, cs,
1537 	    CREATE4args *, args);
1538 
1539 	resp->attrset = 0;
1540 
1541 	if (dvp == NULL) {
1542 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1543 		goto out;
1544 	}
1545 
1546 	/*
1547 	 * If there is an unshared filesystem mounted on this vnode,
1548 	 * do not allow to create an object in this directory.
1549 	 */
1550 	if (vn_ismntpt(dvp)) {
1551 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1552 		goto out;
1553 	}
1554 
1555 	/* Verify that type is correct */
1556 	switch (args->type) {
1557 	case NF4LNK:
1558 	case NF4BLK:
1559 	case NF4CHR:
1560 	case NF4SOCK:
1561 	case NF4FIFO:
1562 	case NF4DIR:
1563 		break;
1564 	default:
1565 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1566 		goto out;
1567 	};
1568 
1569 	if (cs->access == CS_ACCESS_DENIED) {
1570 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1571 		goto out;
1572 	}
1573 	if (dvp->v_type != VDIR) {
1574 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1575 		goto out;
1576 	}
1577 	status = utf8_dir_verify(&args->objname);
1578 	if (status != NFS4_OK) {
1579 		*cs->statusp = resp->status = status;
1580 		goto out;
1581 	}
1582 
1583 	if (rdonly4(req, cs)) {
1584 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1585 		goto out;
1586 	}
1587 
1588 	/*
1589 	 * Name of newly created object
1590 	 */
1591 	nm = utf8_to_fn(&args->objname, &len, NULL);
1592 	if (nm == NULL) {
1593 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1594 		goto out;
1595 	}
1596 
1597 	if (len > MAXNAMELEN) {
1598 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1599 		kmem_free(nm, len);
1600 		goto out;
1601 	}
1602 
1603 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1604 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1605 	    MAXPATHLEN  + 1);
1606 
1607 	if (name == NULL) {
1608 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1609 		kmem_free(nm, len);
1610 		goto out;
1611 	}
1612 
1613 	resp->attrset = 0;
1614 
1615 	sarg.sbp = &sb;
1616 	sarg.is_referral = B_FALSE;
1617 	nfs4_ntov_table_init(&ntov);
1618 
1619 	status = do_rfs4_set_attrs(&resp->attrset,
1620 	    &args->createattrs, cs, &sarg, &ntov, NFS4ATTR_SETIT);
1621 
1622 	if (sarg.vap->va_mask == 0 && status == NFS4_OK)
1623 		status = NFS4ERR_INVAL;
1624 
1625 	if (status != NFS4_OK) {
1626 		*cs->statusp = resp->status = status;
1627 		if (name != nm)
1628 			kmem_free(name, MAXPATHLEN + 1);
1629 		kmem_free(nm, len);
1630 		nfs4_ntov_table_free(&ntov, &sarg);
1631 		resp->attrset = 0;
1632 		goto out;
1633 	}
1634 
1635 	/* Get "before" change value */
1636 	bva.va_mask = AT_CTIME|AT_SEQ|AT_MODE;
1637 	error = VOP_GETATTR(dvp, &bva, 0, cr, NULL);
1638 	if (error) {
1639 		*cs->statusp = resp->status = puterrno4(error);
1640 		if (name != nm)
1641 			kmem_free(name, MAXPATHLEN + 1);
1642 		kmem_free(nm, len);
1643 		nfs4_ntov_table_free(&ntov, &sarg);
1644 		resp->attrset = 0;
1645 		goto out;
1646 	}
1647 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bva.va_ctime)
1648 
1649 	vap = sarg.vap;
1650 
1651 	/*
1652 	 * Set the default initial values for attributes when the parent
1653 	 * directory does not have the VSUID/VSGID bit set and they have
1654 	 * not been specified in createattrs.
1655 	 */
1656 	if (!(bva.va_mode & VSUID) && (vap->va_mask & AT_UID) == 0) {
1657 		vap->va_uid = crgetuid(cr);
1658 		vap->va_mask |= AT_UID;
1659 	}
1660 	if (!(bva.va_mode & VSGID) && (vap->va_mask & AT_GID) == 0) {
1661 		vap->va_gid = crgetgid(cr);
1662 		vap->va_mask |= AT_GID;
1663 	}
1664 
1665 	vap->va_mask |= AT_TYPE;
1666 	switch (args->type) {
1667 	case NF4DIR:
1668 		vap->va_type = VDIR;
1669 		if ((vap->va_mask & AT_MODE) == 0) {
1670 			vap->va_mode = 0700;	/* default: owner rwx only */
1671 			vap->va_mask |= AT_MODE;
1672 		}
1673 		error = VOP_MKDIR(dvp, name, vap, &vp, cr, NULL, 0, NULL);
1674 		if (error)
1675 			break;
1676 
1677 		/*
1678 		 * Get the initial "after" sequence number, if it fails,
1679 		 * set to zero
1680 		 */
1681 		iva.va_mask = AT_SEQ;
1682 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1683 			iva.va_seq = 0;
1684 		break;
1685 	case NF4LNK:
1686 		vap->va_type = VLNK;
1687 		if ((vap->va_mask & AT_MODE) == 0) {
1688 			vap->va_mode = 0700;	/* default: owner rwx only */
1689 			vap->va_mask |= AT_MODE;
1690 		}
1691 
1692 		/*
1693 		 * symlink names must be treated as data
1694 		 */
1695 		lnm = utf8_to_str((utf8string *)&args->ftype4_u.linkdata,
1696 		    &llen, NULL);
1697 
1698 		if (lnm == NULL) {
1699 			*cs->statusp = resp->status = NFS4ERR_INVAL;
1700 			if (name != nm)
1701 				kmem_free(name, MAXPATHLEN + 1);
1702 			kmem_free(nm, len);
1703 			nfs4_ntov_table_free(&ntov, &sarg);
1704 			resp->attrset = 0;
1705 			goto out;
1706 		}
1707 
1708 		if (llen > MAXPATHLEN) {
1709 			*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1710 			if (name != nm)
1711 				kmem_free(name, MAXPATHLEN + 1);
1712 			kmem_free(nm, len);
1713 			kmem_free(lnm, llen);
1714 			nfs4_ntov_table_free(&ntov, &sarg);
1715 			resp->attrset = 0;
1716 			goto out;
1717 		}
1718 
1719 		lname = nfscmd_convname(ca, cs->exi, lnm,
1720 		    NFSCMD_CONV_INBOUND, MAXPATHLEN  + 1);
1721 
1722 		if (lname == NULL) {
1723 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
1724 			if (name != nm)
1725 				kmem_free(name, MAXPATHLEN + 1);
1726 			kmem_free(nm, len);
1727 			kmem_free(lnm, llen);
1728 			nfs4_ntov_table_free(&ntov, &sarg);
1729 			resp->attrset = 0;
1730 			goto out;
1731 		}
1732 
1733 		error = VOP_SYMLINK(dvp, name, vap, lname, cr, NULL, 0);
1734 		if (lname != lnm)
1735 			kmem_free(lname, MAXPATHLEN + 1);
1736 		kmem_free(lnm, llen);
1737 		if (error)
1738 			break;
1739 
1740 		/*
1741 		 * Get the initial "after" sequence number, if it fails,
1742 		 * set to zero
1743 		 */
1744 		iva.va_mask = AT_SEQ;
1745 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1746 			iva.va_seq = 0;
1747 
1748 		error = VOP_LOOKUP(dvp, name, &vp, NULL, 0, NULL, cr,
1749 		    NULL, NULL, NULL);
1750 		if (error)
1751 			break;
1752 
1753 		/*
1754 		 * va_seq is not safe over VOP calls, check it again
1755 		 * if it has changed zero out iva to force atomic = FALSE.
1756 		 */
1757 		iva2.va_mask = AT_SEQ;
1758 		if (VOP_GETATTR(dvp, &iva2, 0, cs->cr, NULL) ||
1759 		    iva2.va_seq != iva.va_seq)
1760 			iva.va_seq = 0;
1761 		break;
1762 	default:
1763 		/*
1764 		 * probably a special file.
1765 		 */
1766 		if ((vap->va_mask & AT_MODE) == 0) {
1767 			vap->va_mode = 0600;	/* default: owner rw only */
1768 			vap->va_mask |= AT_MODE;
1769 		}
1770 		syncval = FNODSYNC;
1771 		/*
1772 		 * We know this will only generate one VOP call
1773 		 */
1774 		vp = do_rfs4_op_mknod(args, resp, req, cs, vap, name);
1775 
1776 		if (vp == NULL) {
1777 			if (name != nm)
1778 				kmem_free(name, MAXPATHLEN + 1);
1779 			kmem_free(nm, len);
1780 			nfs4_ntov_table_free(&ntov, &sarg);
1781 			resp->attrset = 0;
1782 			goto out;
1783 		}
1784 
1785 		/*
1786 		 * Get the initial "after" sequence number, if it fails,
1787 		 * set to zero
1788 		 */
1789 		iva.va_mask = AT_SEQ;
1790 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1791 			iva.va_seq = 0;
1792 
1793 		break;
1794 	}
1795 	if (name != nm)
1796 		kmem_free(name, MAXPATHLEN + 1);
1797 	kmem_free(nm, len);
1798 
1799 	if (error) {
1800 		*cs->statusp = resp->status = puterrno4(error);
1801 	}
1802 
1803 	/*
1804 	 * Force modified data and metadata out to stable storage.
1805 	 */
1806 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
1807 
1808 	if (resp->status != NFS4_OK) {
1809 		if (vp != NULL)
1810 			VN_RELE(vp);
1811 		nfs4_ntov_table_free(&ntov, &sarg);
1812 		resp->attrset = 0;
1813 		goto out;
1814 	}
1815 
1816 	/*
1817 	 * Finish setup of cinfo response, "before" value already set.
1818 	 * Get "after" change value, if it fails, simply return the
1819 	 * before value.
1820 	 */
1821 	ava.va_mask = AT_CTIME|AT_SEQ;
1822 	if (VOP_GETATTR(dvp, &ava, 0, cr, NULL)) {
1823 		ava.va_ctime = bva.va_ctime;
1824 		ava.va_seq = 0;
1825 	}
1826 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, ava.va_ctime);
1827 
1828 	/*
1829 	 * True verification that object was created with correct
1830 	 * attrs is impossible.  The attrs could have been changed
1831 	 * immediately after object creation.  If attributes did
1832 	 * not verify, the only recourse for the server is to
1833 	 * destroy the object.  Maybe if some attrs (like gid)
1834 	 * are set incorrectly, the object should be destroyed;
1835 	 * however, seems bad as a default policy.  Do we really
1836 	 * want to destroy an object over one of the times not
1837 	 * verifying correctly?  For these reasons, the server
1838 	 * currently sets bits in attrset for createattrs
1839 	 * that were set; however, no verification is done.
1840 	 *
1841 	 * vmask_to_nmask accounts for vattr bits set on create
1842 	 *	[do_rfs4_set_attrs() only sets resp bits for
1843 	 *	 non-vattr/vfs bits.]
1844 	 * Mask off any bits set by default so as not to return
1845 	 * more attrset bits than were requested in createattrs
1846 	 */
1847 	nfs4_vmask_to_nmask(sarg.vap->va_mask, &resp->attrset);
1848 	resp->attrset &= args->createattrs.attrmask;
1849 	nfs4_ntov_table_free(&ntov, &sarg);
1850 
1851 	error = makefh4(&cs->fh, vp, cs->exi);
1852 	if (error) {
1853 		*cs->statusp = resp->status = puterrno4(error);
1854 	}
1855 
1856 	/*
1857 	 * The cinfo.atomic = TRUE only if we got no errors, we have
1858 	 * non-zero va_seq's, and it has incremented by exactly one
1859 	 * during the creation and it didn't change during the VOP_LOOKUP
1860 	 * or VOP_FSYNC.
1861 	 */
1862 	if (!error && bva.va_seq && iva.va_seq && ava.va_seq &&
1863 	    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
1864 		resp->cinfo.atomic = TRUE;
1865 	else
1866 		resp->cinfo.atomic = FALSE;
1867 
1868 	/*
1869 	 * Force modified metadata out to stable storage.
1870 	 *
1871 	 * if a underlying vp exists, pass it to VOP_FSYNC
1872 	 */
1873 	if (VOP_REALVP(vp, &realvp, NULL) == 0)
1874 		(void) VOP_FSYNC(realvp, syncval, cr, NULL);
1875 	else
1876 		(void) VOP_FSYNC(vp, syncval, cr, NULL);
1877 
1878 	if (resp->status != NFS4_OK) {
1879 		VN_RELE(vp);
1880 		goto out;
1881 	}
1882 	if (cs->vp)
1883 		VN_RELE(cs->vp);
1884 
1885 	cs->vp = vp;
1886 	*cs->statusp = resp->status = NFS4_OK;
1887 out:
1888 	DTRACE_NFSV4_2(op__create__done, struct compound_state *, cs,
1889 	    CREATE4res *, resp);
1890 }
1891 
1892 /*ARGSUSED*/
1893 static void
1894 rfs4_op_delegpurge(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1895     struct compound_state *cs)
1896 {
1897 	DTRACE_NFSV4_2(op__delegpurge__start, struct compound_state *, cs,
1898 	    DELEGPURGE4args *, &argop->nfs_argop4_u.opdelegpurge);
1899 
1900 	rfs4_op_inval(argop, resop, req, cs);
1901 
1902 	DTRACE_NFSV4_2(op__delegpurge__done, struct compound_state *, cs,
1903 	    DELEGPURGE4res *, &resop->nfs_resop4_u.opdelegpurge);
1904 }
1905 
1906 /*ARGSUSED*/
1907 static void
1908 rfs4_op_delegreturn(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1909     struct compound_state *cs)
1910 {
1911 	DELEGRETURN4args *args = &argop->nfs_argop4_u.opdelegreturn;
1912 	DELEGRETURN4res *resp = &resop->nfs_resop4_u.opdelegreturn;
1913 	rfs4_deleg_state_t *dsp;
1914 	nfsstat4 status;
1915 
1916 	DTRACE_NFSV4_2(op__delegreturn__start, struct compound_state *, cs,
1917 	    DELEGRETURN4args *, args);
1918 
1919 	status = rfs4_get_deleg_state(&args->deleg_stateid, &dsp);
1920 	resp->status = *cs->statusp = status;
1921 	if (status != NFS4_OK)
1922 		goto out;
1923 
1924 	/* Ensure specified filehandle matches */
1925 	if (cs->vp != dsp->rds_finfo->rf_vp) {
1926 		resp->status = *cs->statusp = NFS4ERR_BAD_STATEID;
1927 	} else
1928 		rfs4_return_deleg(dsp, FALSE);
1929 
1930 	rfs4_update_lease(dsp->rds_client);
1931 
1932 	rfs4_deleg_state_rele(dsp);
1933 out:
1934 	DTRACE_NFSV4_2(op__delegreturn__done, struct compound_state *, cs,
1935 	    DELEGRETURN4res *, resp);
1936 }
1937 
1938 /*
1939  * Check to see if a given "flavor" is an explicitly shared flavor.
1940  * The assumption of this routine is the "flavor" is already a valid
1941  * flavor in the secinfo list of "exi".
1942  *
1943  *	e.g.
1944  *		# share -o sec=flavor1 /export
1945  *		# share -o sec=flavor2 /export/home
1946  *
1947  *		flavor2 is not an explicitly shared flavor for /export,
1948  *		however it is in the secinfo list for /export thru the
1949  *		server namespace setup.
1950  */
1951 int
1952 is_exported_sec(int flavor, struct exportinfo *exi)
1953 {
1954 	int	i;
1955 	struct secinfo *sp;
1956 
1957 	sp = exi->exi_export.ex_secinfo;
1958 	for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
1959 		if (flavor == sp[i].s_secinfo.sc_nfsnum ||
1960 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE) {
1961 			return (SEC_REF_EXPORTED(&sp[i]));
1962 		}
1963 	}
1964 
1965 	/* Should not reach this point based on the assumption */
1966 	return (0);
1967 }
1968 
1969 /*
1970  * Check if the security flavor used in the request matches what is
1971  * required at the export point or at the root pseudo node (exi_root).
1972  *
1973  * returns 1 if there's a match or if exported with AUTH_NONE; 0 otherwise.
1974  *
1975  */
1976 static int
1977 secinfo_match_or_authnone(struct compound_state *cs)
1978 {
1979 	int	i;
1980 	struct secinfo *sp;
1981 
1982 	/*
1983 	 * Check cs->nfsflavor (from the request) against
1984 	 * the current export data in cs->exi.
1985 	 */
1986 	sp = cs->exi->exi_export.ex_secinfo;
1987 	for (i = 0; i < cs->exi->exi_export.ex_seccnt; i++) {
1988 		if (cs->nfsflavor == sp[i].s_secinfo.sc_nfsnum ||
1989 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE)
1990 			return (1);
1991 	}
1992 
1993 	return (0);
1994 }
1995 
1996 /*
1997  * Check the access authority for the client and return the correct error.
1998  */
1999 nfsstat4
2000 call_checkauth4(struct compound_state *cs, struct svc_req *req)
2001 {
2002 	int	authres;
2003 
2004 	/*
2005 	 * First, check if the security flavor used in the request
2006 	 * are among the flavors set in the server namespace.
2007 	 */
2008 	if (!secinfo_match_or_authnone(cs)) {
2009 		*cs->statusp = NFS4ERR_WRONGSEC;
2010 		return (*cs->statusp);
2011 	}
2012 
2013 	authres = checkauth4(cs, req);
2014 
2015 	if (authres > 0) {
2016 		*cs->statusp = NFS4_OK;
2017 		if (! (cs->access & CS_ACCESS_LIMITED))
2018 			cs->access = CS_ACCESS_OK;
2019 	} else if (authres == 0) {
2020 		*cs->statusp = NFS4ERR_ACCESS;
2021 	} else if (authres == -2) {
2022 		*cs->statusp = NFS4ERR_WRONGSEC;
2023 	} else {
2024 		*cs->statusp = NFS4ERR_DELAY;
2025 	}
2026 	return (*cs->statusp);
2027 }
2028 
2029 /*
2030  * bitmap4_to_attrmask is called by getattr and readdir.
2031  * It sets up the vattr mask and determines whether vfsstat call is needed
2032  * based on the input bitmap.
2033  * Returns nfsv4 status.
2034  */
2035 static nfsstat4
2036 bitmap4_to_attrmask(bitmap4 breq, struct nfs4_svgetit_arg *sargp)
2037 {
2038 	int i;
2039 	uint_t	va_mask;
2040 	struct statvfs64 *sbp = sargp->sbp;
2041 
2042 	sargp->sbp = NULL;
2043 	sargp->flag = 0;
2044 	sargp->rdattr_error = NFS4_OK;
2045 	sargp->mntdfid_set = FALSE;
2046 	if (sargp->cs->vp)
2047 		sargp->xattr = get_fh4_flag(&sargp->cs->fh,
2048 		    FH4_ATTRDIR | FH4_NAMEDATTR);
2049 	else
2050 		sargp->xattr = 0;
2051 
2052 	/*
2053 	 * Set rdattr_error_req to true if return error per
2054 	 * failed entry rather than fail the readdir.
2055 	 */
2056 	if (breq & FATTR4_RDATTR_ERROR_MASK)
2057 		sargp->rdattr_error_req = 1;
2058 	else
2059 		sargp->rdattr_error_req = 0;
2060 
2061 	/*
2062 	 * generate the va_mask
2063 	 * Handle the easy cases first
2064 	 */
2065 	switch (breq) {
2066 	case NFS4_NTOV_ATTR_MASK:
2067 		sargp->vap->va_mask = NFS4_NTOV_ATTR_AT_MASK;
2068 		return (NFS4_OK);
2069 
2070 	case NFS4_FS_ATTR_MASK:
2071 		sargp->vap->va_mask = NFS4_FS_ATTR_AT_MASK;
2072 		sargp->sbp = sbp;
2073 		return (NFS4_OK);
2074 
2075 	case NFS4_NTOV_ATTR_CACHE_MASK:
2076 		sargp->vap->va_mask = NFS4_NTOV_ATTR_CACHE_AT_MASK;
2077 		return (NFS4_OK);
2078 
2079 	case FATTR4_LEASE_TIME_MASK:
2080 		sargp->vap->va_mask = 0;
2081 		return (NFS4_OK);
2082 
2083 	default:
2084 		va_mask = 0;
2085 		for (i = 0; i < nfs4_ntov_map_size; i++) {
2086 			if ((breq & nfs4_ntov_map[i].fbit) &&
2087 			    nfs4_ntov_map[i].vbit)
2088 				va_mask |= nfs4_ntov_map[i].vbit;
2089 		}
2090 
2091 		/*
2092 		 * Check is vfsstat is needed
2093 		 */
2094 		if (breq & NFS4_FS_ATTR_MASK)
2095 			sargp->sbp = sbp;
2096 
2097 		sargp->vap->va_mask = va_mask;
2098 		return (NFS4_OK);
2099 	}
2100 	/* NOTREACHED */
2101 }
2102 
2103 /*
2104  * bitmap4_get_sysattrs is called by getattr and readdir.
2105  * It calls both VOP_GETATTR and VFS_STATVFS calls to get the attrs.
2106  * Returns nfsv4 status.
2107  */
2108 static nfsstat4
2109 bitmap4_get_sysattrs(struct nfs4_svgetit_arg *sargp)
2110 {
2111 	int error;
2112 	struct compound_state *cs = sargp->cs;
2113 	vnode_t *vp = cs->vp;
2114 
2115 	if (sargp->sbp != NULL) {
2116 		if (error = VFS_STATVFS(vp->v_vfsp, sargp->sbp)) {
2117 			sargp->sbp = NULL;	/* to identify error */
2118 			return (puterrno4(error));
2119 		}
2120 	}
2121 
2122 	return (rfs4_vop_getattr(vp, sargp->vap, 0, cs->cr));
2123 }
2124 
2125 static void
2126 nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp)
2127 {
2128 	ntovp->na = kmem_zalloc(sizeof (union nfs4_attr_u) * nfs4_ntov_map_size,
2129 	    KM_SLEEP);
2130 	ntovp->attrcnt = 0;
2131 	ntovp->vfsstat = FALSE;
2132 }
2133 
2134 static void
2135 nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
2136     struct nfs4_svgetit_arg *sargp)
2137 {
2138 	int i;
2139 	union nfs4_attr_u *na;
2140 	uint8_t *amap;
2141 
2142 	/*
2143 	 * XXX Should do the same checks for whether the bit is set
2144 	 */
2145 	for (i = 0, na = ntovp->na, amap = ntovp->amap;
2146 	    i < ntovp->attrcnt; i++, na++, amap++) {
2147 		(void) (*nfs4_ntov_map[*amap].sv_getit)(
2148 		    NFS4ATTR_FREEIT, sargp, na);
2149 	}
2150 	if ((sargp->op == NFS4ATTR_SETIT) || (sargp->op == NFS4ATTR_VERIT)) {
2151 		/*
2152 		 * xdr_free for getattr will be done later
2153 		 */
2154 		for (i = 0, na = ntovp->na, amap = ntovp->amap;
2155 		    i < ntovp->attrcnt; i++, na++, amap++) {
2156 			xdr_free(nfs4_ntov_map[*amap].xfunc, (caddr_t)na);
2157 		}
2158 	}
2159 	kmem_free(ntovp->na, sizeof (union nfs4_attr_u) * nfs4_ntov_map_size);
2160 }
2161 
2162 /*
2163  * do_rfs4_op_getattr gets the system attrs and converts into fattr4.
2164  */
2165 static nfsstat4
2166 do_rfs4_op_getattr(bitmap4 breq, fattr4 *fattrp,
2167     struct nfs4_svgetit_arg *sargp)
2168 {
2169 	int error = 0;
2170 	int i, k;
2171 	struct nfs4_ntov_table ntov;
2172 	XDR xdr;
2173 	ulong_t xdr_size;
2174 	char *xdr_attrs;
2175 	nfsstat4 status = NFS4_OK;
2176 	nfsstat4 prev_rdattr_error = sargp->rdattr_error;
2177 	union nfs4_attr_u *na;
2178 	uint8_t *amap;
2179 
2180 	sargp->op = NFS4ATTR_GETIT;
2181 	sargp->flag = 0;
2182 
2183 	fattrp->attrmask = 0;
2184 	/* if no bits requested, then return empty fattr4 */
2185 	if (breq == 0) {
2186 		fattrp->attrlist4_len = 0;
2187 		fattrp->attrlist4 = NULL;
2188 		return (NFS4_OK);
2189 	}
2190 
2191 	/*
2192 	 * return NFS4ERR_INVAL when client requests write-only attrs
2193 	 */
2194 	if (breq & (FATTR4_TIME_ACCESS_SET_MASK | FATTR4_TIME_MODIFY_SET_MASK))
2195 		return (NFS4ERR_INVAL);
2196 
2197 	nfs4_ntov_table_init(&ntov);
2198 	na = ntov.na;
2199 	amap = ntov.amap;
2200 
2201 	/*
2202 	 * Now loop to get or verify the attrs
2203 	 */
2204 	for (i = 0; i < nfs4_ntov_map_size; i++) {
2205 		if (breq & nfs4_ntov_map[i].fbit) {
2206 			if ((*nfs4_ntov_map[i].sv_getit)(
2207 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0) {
2208 
2209 				error = (*nfs4_ntov_map[i].sv_getit)(
2210 				    NFS4ATTR_GETIT, sargp, na);
2211 
2212 				/*
2213 				 * Possible error values:
2214 				 * >0 if sv_getit failed to
2215 				 * get the attr; 0 if succeeded;
2216 				 * <0 if rdattr_error and the
2217 				 * attribute cannot be returned.
2218 				 */
2219 				if (error && !(sargp->rdattr_error_req))
2220 					goto done;
2221 				/*
2222 				 * If error then just for entry
2223 				 */
2224 				if (error == 0) {
2225 					fattrp->attrmask |=
2226 					    nfs4_ntov_map[i].fbit;
2227 					*amap++ =
2228 					    (uint8_t)nfs4_ntov_map[i].nval;
2229 					na++;
2230 					(ntov.attrcnt)++;
2231 				} else if ((error > 0) &&
2232 				    (sargp->rdattr_error == NFS4_OK)) {
2233 					sargp->rdattr_error = puterrno4(error);
2234 				}
2235 				error = 0;
2236 			}
2237 		}
2238 	}
2239 
2240 	/*
2241 	 * If rdattr_error was set after the return value for it was assigned,
2242 	 * update it.
2243 	 */
2244 	if (prev_rdattr_error != sargp->rdattr_error) {
2245 		na = ntov.na;
2246 		amap = ntov.amap;
2247 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2248 			k = *amap;
2249 			if (k < FATTR4_RDATTR_ERROR) {
2250 				continue;
2251 			}
2252 			if ((k == FATTR4_RDATTR_ERROR) &&
2253 			    ((*nfs4_ntov_map[k].sv_getit)(
2254 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0)) {
2255 
2256 				(void) (*nfs4_ntov_map[k].sv_getit)(
2257 				    NFS4ATTR_GETIT, sargp, na);
2258 			}
2259 			break;
2260 		}
2261 	}
2262 
2263 	xdr_size = 0;
2264 	na = ntov.na;
2265 	amap = ntov.amap;
2266 	for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2267 		xdr_size += xdr_sizeof(nfs4_ntov_map[*amap].xfunc, na);
2268 	}
2269 
2270 	fattrp->attrlist4_len = xdr_size;
2271 	if (xdr_size) {
2272 		/* freed by rfs4_op_getattr_free() */
2273 		fattrp->attrlist4 = xdr_attrs = kmem_zalloc(xdr_size, KM_SLEEP);
2274 
2275 		xdrmem_create(&xdr, xdr_attrs, xdr_size, XDR_ENCODE);
2276 
2277 		na = ntov.na;
2278 		amap = ntov.amap;
2279 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2280 			if (!(*nfs4_ntov_map[*amap].xfunc)(&xdr, na)) {
2281 				DTRACE_PROBE1(nfss__e__getattr4_encfail,
2282 				    int, *amap);
2283 				status = NFS4ERR_SERVERFAULT;
2284 				break;
2285 			}
2286 		}
2287 		/* xdrmem_destroy(&xdrs); */	/* NO-OP */
2288 	} else {
2289 		fattrp->attrlist4 = NULL;
2290 	}
2291 done:
2292 
2293 	nfs4_ntov_table_free(&ntov, sargp);
2294 
2295 	if (error != 0)
2296 		status = puterrno4(error);
2297 
2298 	return (status);
2299 }
2300 
2301 /* ARGSUSED */
2302 static void
2303 rfs4_op_getattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2304     struct compound_state *cs)
2305 {
2306 	GETATTR4args *args = &argop->nfs_argop4_u.opgetattr;
2307 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2308 	struct nfs4_svgetit_arg sarg;
2309 	struct statvfs64 sb;
2310 	nfsstat4 status;
2311 
2312 	DTRACE_NFSV4_2(op__getattr__start, struct compound_state *, cs,
2313 	    GETATTR4args *, args);
2314 
2315 	if (cs->vp == NULL) {
2316 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2317 		goto out;
2318 	}
2319 
2320 	if (cs->access == CS_ACCESS_DENIED) {
2321 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2322 		goto out;
2323 	}
2324 
2325 	sarg.sbp = &sb;
2326 	sarg.cs = cs;
2327 	sarg.is_referral = B_FALSE;
2328 
2329 	status = bitmap4_to_attrmask(args->attr_request, &sarg);
2330 	if (status == NFS4_OK) {
2331 
2332 		status = bitmap4_get_sysattrs(&sarg);
2333 		if (status == NFS4_OK) {
2334 
2335 			/* Is this a referral? */
2336 			if (vn_is_nfs_reparse(cs->vp, cs->cr)) {
2337 				/* Older V4 Solaris client sees a link */
2338 				if (client_is_downrev(req))
2339 					sarg.vap->va_type = VLNK;
2340 				else
2341 					sarg.is_referral = B_TRUE;
2342 			}
2343 
2344 			status = do_rfs4_op_getattr(args->attr_request,
2345 			    &resp->obj_attributes, &sarg);
2346 		}
2347 	}
2348 	*cs->statusp = resp->status = status;
2349 out:
2350 	DTRACE_NFSV4_2(op__getattr__done, struct compound_state *, cs,
2351 	    GETATTR4res *, resp);
2352 }
2353 
2354 static void
2355 rfs4_op_getattr_free(nfs_resop4 *resop)
2356 {
2357 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2358 
2359 	nfs4_fattr4_free(&resp->obj_attributes);
2360 }
2361 
2362 /* ARGSUSED */
2363 static void
2364 rfs4_op_getfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2365     struct compound_state *cs)
2366 {
2367 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2368 
2369 	DTRACE_NFSV4_1(op__getfh__start, struct compound_state *, cs);
2370 
2371 	if (cs->vp == NULL) {
2372 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2373 		goto out;
2374 	}
2375 	if (cs->access == CS_ACCESS_DENIED) {
2376 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2377 		goto out;
2378 	}
2379 
2380 	/* check for reparse point at the share point */
2381 	if (cs->exi->exi_moved || vn_is_nfs_reparse(cs->exi->exi_vp, cs->cr)) {
2382 		/* it's all bad */
2383 		cs->exi->exi_moved = 1;
2384 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2385 		DTRACE_PROBE2(nfs4serv__func__referral__shared__moved,
2386 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2387 		return;
2388 	}
2389 
2390 	/* check for reparse point at vp */
2391 	if (vn_is_nfs_reparse(cs->vp, cs->cr) && !client_is_downrev(req)) {
2392 		/* it's not all bad */
2393 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2394 		DTRACE_PROBE2(nfs4serv__func__referral__moved,
2395 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2396 		return;
2397 	}
2398 
2399 	resp->object.nfs_fh4_val =
2400 	    kmem_alloc(cs->fh.nfs_fh4_len, KM_SLEEP);
2401 	nfs_fh4_copy(&cs->fh, &resp->object);
2402 	*cs->statusp = resp->status = NFS4_OK;
2403 out:
2404 	DTRACE_NFSV4_2(op__getfh__done, struct compound_state *, cs,
2405 	    GETFH4res *, resp);
2406 }
2407 
2408 static void
2409 rfs4_op_getfh_free(nfs_resop4 *resop)
2410 {
2411 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2412 
2413 	if (resp->status == NFS4_OK &&
2414 	    resp->object.nfs_fh4_val != NULL) {
2415 		kmem_free(resp->object.nfs_fh4_val, resp->object.nfs_fh4_len);
2416 		resp->object.nfs_fh4_val = NULL;
2417 		resp->object.nfs_fh4_len = 0;
2418 	}
2419 }
2420 
2421 /*
2422  * illegal: args: void
2423  *	    res : status (NFS4ERR_OP_ILLEGAL)
2424  */
2425 /* ARGSUSED */
2426 static void
2427 rfs4_op_illegal(nfs_argop4 *argop, nfs_resop4 *resop,
2428     struct svc_req *req, struct compound_state *cs)
2429 {
2430 	ILLEGAL4res *resp = &resop->nfs_resop4_u.opillegal;
2431 
2432 	resop->resop = OP_ILLEGAL;
2433 	*cs->statusp = resp->status = NFS4ERR_OP_ILLEGAL;
2434 }
2435 
2436 /*
2437  * link: args: SAVED_FH: file, CURRENT_FH: target directory
2438  *	 res: status. If success - CURRENT_FH unchanged, return change_info
2439  */
2440 /* ARGSUSED */
2441 static void
2442 rfs4_op_link(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2443     struct compound_state *cs)
2444 {
2445 	LINK4args *args = &argop->nfs_argop4_u.oplink;
2446 	LINK4res *resp = &resop->nfs_resop4_u.oplink;
2447 	int error;
2448 	vnode_t *vp;
2449 	vnode_t *dvp;
2450 	struct vattr bdva, idva, adva;
2451 	char *nm;
2452 	uint_t  len;
2453 	struct sockaddr *ca;
2454 	char *name = NULL;
2455 	nfsstat4 status;
2456 
2457 	DTRACE_NFSV4_2(op__link__start, struct compound_state *, cs,
2458 	    LINK4args *, args);
2459 
2460 	/* SAVED_FH: source object */
2461 	vp = cs->saved_vp;
2462 	if (vp == NULL) {
2463 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2464 		goto out;
2465 	}
2466 
2467 	/* CURRENT_FH: target directory */
2468 	dvp = cs->vp;
2469 	if (dvp == NULL) {
2470 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2471 		goto out;
2472 	}
2473 
2474 	/*
2475 	 * If there is a non-shared filesystem mounted on this vnode,
2476 	 * do not allow to link any file in this directory.
2477 	 */
2478 	if (vn_ismntpt(dvp)) {
2479 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2480 		goto out;
2481 	}
2482 
2483 	if (cs->access == CS_ACCESS_DENIED) {
2484 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2485 		goto out;
2486 	}
2487 
2488 	/* Check source object's type validity */
2489 	if (vp->v_type == VDIR) {
2490 		*cs->statusp = resp->status = NFS4ERR_ISDIR;
2491 		goto out;
2492 	}
2493 
2494 	/* Check target directory's type */
2495 	if (dvp->v_type != VDIR) {
2496 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2497 		goto out;
2498 	}
2499 
2500 	if (cs->saved_exi != cs->exi) {
2501 		*cs->statusp = resp->status = NFS4ERR_XDEV;
2502 		goto out;
2503 	}
2504 
2505 	status = utf8_dir_verify(&args->newname);
2506 	if (status != NFS4_OK) {
2507 		*cs->statusp = resp->status = status;
2508 		goto out;
2509 	}
2510 
2511 	nm = utf8_to_fn(&args->newname, &len, NULL);
2512 	if (nm == NULL) {
2513 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2514 		goto out;
2515 	}
2516 
2517 	if (len > MAXNAMELEN) {
2518 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2519 		kmem_free(nm, len);
2520 		goto out;
2521 	}
2522 
2523 	if (rdonly4(req, cs)) {
2524 		*cs->statusp = resp->status = NFS4ERR_ROFS;
2525 		kmem_free(nm, len);
2526 		goto out;
2527 	}
2528 
2529 	/* Get "before" change value */
2530 	bdva.va_mask = AT_CTIME|AT_SEQ;
2531 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
2532 	if (error) {
2533 		*cs->statusp = resp->status = puterrno4(error);
2534 		kmem_free(nm, len);
2535 		goto out;
2536 	}
2537 
2538 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2539 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2540 	    MAXPATHLEN  + 1);
2541 
2542 	if (name == NULL) {
2543 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2544 		kmem_free(nm, len);
2545 		goto out;
2546 	}
2547 
2548 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
2549 
2550 	error = VOP_LINK(dvp, vp, name, cs->cr, NULL, 0);
2551 
2552 	if (nm != name)
2553 		kmem_free(name, MAXPATHLEN + 1);
2554 	kmem_free(nm, len);
2555 
2556 	/*
2557 	 * Get the initial "after" sequence number, if it fails, set to zero
2558 	 */
2559 	idva.va_mask = AT_SEQ;
2560 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
2561 		idva.va_seq = 0;
2562 
2563 	/*
2564 	 * Force modified data and metadata out to stable storage.
2565 	 */
2566 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
2567 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
2568 
2569 	if (error) {
2570 		*cs->statusp = resp->status = puterrno4(error);
2571 		goto out;
2572 	}
2573 
2574 	/*
2575 	 * Get "after" change value, if it fails, simply return the
2576 	 * before value.
2577 	 */
2578 	adva.va_mask = AT_CTIME|AT_SEQ;
2579 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
2580 		adva.va_ctime = bdva.va_ctime;
2581 		adva.va_seq = 0;
2582 	}
2583 
2584 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
2585 
2586 	/*
2587 	 * The cinfo.atomic = TRUE only if we have
2588 	 * non-zero va_seq's, and it has incremented by exactly one
2589 	 * during the VOP_LINK and it didn't change during the VOP_FSYNC.
2590 	 */
2591 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
2592 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
2593 		resp->cinfo.atomic = TRUE;
2594 	else
2595 		resp->cinfo.atomic = FALSE;
2596 
2597 	*cs->statusp = resp->status = NFS4_OK;
2598 out:
2599 	DTRACE_NFSV4_2(op__link__done, struct compound_state *, cs,
2600 	    LINK4res *, resp);
2601 }
2602 
2603 /*
2604  * Used by rfs4_op_lookup and rfs4_op_lookupp to do the actual work.
2605  */
2606 
2607 /* ARGSUSED */
2608 static nfsstat4
2609 do_rfs4_op_lookup(char *nm, struct svc_req *req, struct compound_state *cs)
2610 {
2611 	int error;
2612 	int different_export = 0;
2613 	vnode_t *vp, *pre_tvp = NULL, *oldvp = NULL;
2614 	struct exportinfo *exi = NULL, *pre_exi = NULL;
2615 	nfsstat4 stat;
2616 	fid_t fid;
2617 	int attrdir, dotdot, walk;
2618 	bool_t is_newvp = FALSE;
2619 
2620 	if (cs->vp->v_flag & V_XATTRDIR) {
2621 		attrdir = 1;
2622 		ASSERT(get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2623 	} else {
2624 		attrdir = 0;
2625 		ASSERT(! get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2626 	}
2627 
2628 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
2629 
2630 	/*
2631 	 * If dotdotting, then need to check whether it's
2632 	 * above the root of a filesystem, or above an
2633 	 * export point.
2634 	 */
2635 	if (dotdot) {
2636 
2637 		/*
2638 		 * If dotdotting at the root of a filesystem, then
2639 		 * need to traverse back to the mounted-on filesystem
2640 		 * and do the dotdot lookup there.
2641 		 */
2642 		if (cs->vp->v_flag & VROOT) {
2643 
2644 			/*
2645 			 * If at the system root, then can
2646 			 * go up no further.
2647 			 */
2648 			if (VN_CMP(cs->vp, rootdir))
2649 				return (puterrno4(ENOENT));
2650 
2651 			/*
2652 			 * Traverse back to the mounted-on filesystem
2653 			 */
2654 			cs->vp = untraverse(cs->vp);
2655 
2656 			/*
2657 			 * Set the different_export flag so we remember
2658 			 * to pick up a new exportinfo entry for
2659 			 * this new filesystem.
2660 			 */
2661 			different_export = 1;
2662 		} else {
2663 
2664 			/*
2665 			 * If dotdotting above an export point then set
2666 			 * the different_export to get new export info.
2667 			 */
2668 			different_export = nfs_exported(cs->exi, cs->vp);
2669 		}
2670 	}
2671 
2672 	error = VOP_LOOKUP(cs->vp, nm, &vp, NULL, 0, NULL, cs->cr,
2673 	    NULL, NULL, NULL);
2674 	if (error)
2675 		return (puterrno4(error));
2676 
2677 	/*
2678 	 * If the vnode is in a pseudo filesystem, check whether it is visible.
2679 	 *
2680 	 * XXX if the vnode is a symlink and it is not visible in
2681 	 * a pseudo filesystem, return ENOENT (not following symlink).
2682 	 * V4 client can not mount such symlink. This is a regression
2683 	 * from V2/V3.
2684 	 *
2685 	 * In the same exported filesystem, if the security flavor used
2686 	 * is not an explicitly shared flavor, limit the view to the visible
2687 	 * list entries only. This is not a WRONGSEC case because it's already
2688 	 * checked via PUTROOTFH/PUTPUBFH or PUTFH.
2689 	 */
2690 	if (!different_export &&
2691 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
2692 	    cs->access & CS_ACCESS_LIMITED)) {
2693 		if (! nfs_visible(cs->exi, vp, &different_export)) {
2694 			VN_RELE(vp);
2695 			return (puterrno4(ENOENT));
2696 		}
2697 	}
2698 
2699 	/*
2700 	 * If it's a mountpoint, then traverse it.
2701 	 */
2702 	if (vn_ismntpt(vp)) {
2703 		pre_exi = cs->exi;	/* save pre-traversed exportinfo */
2704 		pre_tvp = vp;		/* save pre-traversed vnode	*/
2705 
2706 		/*
2707 		 * hold pre_tvp to counteract rele by traverse.  We will
2708 		 * need pre_tvp below if checkexport4 fails
2709 		 */
2710 		VN_HOLD(pre_tvp);
2711 		if ((error = traverse(&vp)) != 0) {
2712 			VN_RELE(vp);
2713 			VN_RELE(pre_tvp);
2714 			return (puterrno4(error));
2715 		}
2716 		different_export = 1;
2717 	} else if (vp->v_vfsp != cs->vp->v_vfsp) {
2718 		/*
2719 		 * The vfsp comparison is to handle the case where
2720 		 * a LOFS mount is shared.  lo_lookup traverses mount points,
2721 		 * and NFS is unaware of local fs transistions because
2722 		 * v_vfsmountedhere isn't set.  For this special LOFS case,
2723 		 * the dir and the obj returned by lookup will have different
2724 		 * vfs ptrs.
2725 		 */
2726 		different_export = 1;
2727 	}
2728 
2729 	if (different_export) {
2730 
2731 		bzero(&fid, sizeof (fid));
2732 		fid.fid_len = MAXFIDSZ;
2733 		error = vop_fid_pseudo(vp, &fid);
2734 		if (error) {
2735 			VN_RELE(vp);
2736 			if (pre_tvp)
2737 				VN_RELE(pre_tvp);
2738 			return (puterrno4(error));
2739 		}
2740 
2741 		if (dotdot)
2742 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
2743 		else
2744 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2745 
2746 		if (exi == NULL) {
2747 			if (pre_tvp) {
2748 				/*
2749 				 * If this vnode is a mounted-on vnode,
2750 				 * but the mounted-on file system is not
2751 				 * exported, send back the filehandle for
2752 				 * the mounted-on vnode, not the root of
2753 				 * the mounted-on file system.
2754 				 */
2755 				VN_RELE(vp);
2756 				vp = pre_tvp;
2757 				exi = pre_exi;
2758 			} else {
2759 				VN_RELE(vp);
2760 				return (puterrno4(EACCES));
2761 			}
2762 		} else if (pre_tvp) {
2763 			/* we're done with pre_tvp now. release extra hold */
2764 			VN_RELE(pre_tvp);
2765 		}
2766 
2767 		cs->exi = exi;
2768 
2769 		/*
2770 		 * Now we do a checkauth4. The reason is that
2771 		 * this client/user may not have access to the new
2772 		 * exported file system, and if they do,
2773 		 * the client/user may be mapped to a different uid.
2774 		 *
2775 		 * We start with a new cr, because the checkauth4 done
2776 		 * in the PUT*FH operation over wrote the cred's uid,
2777 		 * gid, etc, and we want the real thing before calling
2778 		 * checkauth4()
2779 		 */
2780 		crfree(cs->cr);
2781 		cs->cr = crdup(cs->basecr);
2782 
2783 		oldvp = cs->vp;
2784 		cs->vp = vp;
2785 		is_newvp = TRUE;
2786 
2787 		stat = call_checkauth4(cs, req);
2788 		if (stat != NFS4_OK) {
2789 			VN_RELE(cs->vp);
2790 			cs->vp = oldvp;
2791 			return (stat);
2792 		}
2793 	}
2794 
2795 	/*
2796 	 * After various NFS checks, do a label check on the path
2797 	 * component. The label on this path should either be the
2798 	 * global zone's label or a zone's label. We are only
2799 	 * interested in the zone's label because exported files
2800 	 * in global zone is accessible (though read-only) to
2801 	 * clients. The exportability/visibility check is already
2802 	 * done before reaching this code.
2803 	 */
2804 	if (is_system_labeled()) {
2805 		bslabel_t *clabel;
2806 
2807 		ASSERT(req->rq_label != NULL);
2808 		clabel = req->rq_label;
2809 		DTRACE_PROBE2(tx__rfs4__log__info__oplookup__clabel, char *,
2810 		    "got client label from request(1)", struct svc_req *, req);
2811 
2812 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
2813 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
2814 			    cs->exi)) {
2815 				error = EACCES;
2816 				goto err_out;
2817 			}
2818 		} else {
2819 			/*
2820 			 * We grant access to admin_low label clients
2821 			 * only if the client is trusted, i.e. also
2822 			 * running Solaris Trusted Extension.
2823 			 */
2824 			struct sockaddr	*ca;
2825 			int		addr_type;
2826 			void		*ipaddr;
2827 			tsol_tpc_t	*tp;
2828 
2829 			ca = (struct sockaddr *)svc_getrpccaller(
2830 			    req->rq_xprt)->buf;
2831 			if (ca->sa_family == AF_INET) {
2832 				addr_type = IPV4_VERSION;
2833 				ipaddr = &((struct sockaddr_in *)ca)->sin_addr;
2834 			} else if (ca->sa_family == AF_INET6) {
2835 				addr_type = IPV6_VERSION;
2836 				ipaddr = &((struct sockaddr_in6 *)
2837 				    ca)->sin6_addr;
2838 			}
2839 			tp = find_tpc(ipaddr, addr_type, B_FALSE);
2840 			if (tp == NULL || tp->tpc_tp.tp_doi !=
2841 			    l_admin_low->tsl_doi || tp->tpc_tp.host_type !=
2842 			    SUN_CIPSO) {
2843 				if (tp != NULL)
2844 					TPC_RELE(tp);
2845 				error = EACCES;
2846 				goto err_out;
2847 			}
2848 			TPC_RELE(tp);
2849 		}
2850 	}
2851 
2852 	error = makefh4(&cs->fh, vp, cs->exi);
2853 
2854 err_out:
2855 	if (error) {
2856 		if (is_newvp) {
2857 			VN_RELE(cs->vp);
2858 			cs->vp = oldvp;
2859 		} else
2860 			VN_RELE(vp);
2861 		return (puterrno4(error));
2862 	}
2863 
2864 	if (!is_newvp) {
2865 		if (cs->vp)
2866 			VN_RELE(cs->vp);
2867 		cs->vp = vp;
2868 	} else if (oldvp)
2869 		VN_RELE(oldvp);
2870 
2871 	/*
2872 	 * if did lookup on attrdir and didn't lookup .., set named
2873 	 * attr fh flag
2874 	 */
2875 	if (attrdir && ! dotdot)
2876 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
2877 
2878 	/* Assume false for now, open proc will set this */
2879 	cs->mandlock = FALSE;
2880 
2881 	return (NFS4_OK);
2882 }
2883 
2884 /* ARGSUSED */
2885 static void
2886 rfs4_op_lookup(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2887     struct compound_state *cs)
2888 {
2889 	LOOKUP4args *args = &argop->nfs_argop4_u.oplookup;
2890 	LOOKUP4res *resp = &resop->nfs_resop4_u.oplookup;
2891 	char *nm;
2892 	uint_t len;
2893 	struct sockaddr *ca;
2894 	char *name = NULL;
2895 	nfsstat4 status;
2896 
2897 	DTRACE_NFSV4_2(op__lookup__start, struct compound_state *, cs,
2898 	    LOOKUP4args *, args);
2899 
2900 	if (cs->vp == NULL) {
2901 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2902 		goto out;
2903 	}
2904 
2905 	if (cs->vp->v_type == VLNK) {
2906 		*cs->statusp = resp->status = NFS4ERR_SYMLINK;
2907 		goto out;
2908 	}
2909 
2910 	if (cs->vp->v_type != VDIR) {
2911 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2912 		goto out;
2913 	}
2914 
2915 	status = utf8_dir_verify(&args->objname);
2916 	if (status != NFS4_OK) {
2917 		*cs->statusp = resp->status = status;
2918 		goto out;
2919 	}
2920 
2921 	nm = utf8_to_str(&args->objname, &len, NULL);
2922 	if (nm == NULL) {
2923 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2924 		goto out;
2925 	}
2926 
2927 	if (len > MAXNAMELEN) {
2928 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2929 		kmem_free(nm, len);
2930 		goto out;
2931 	}
2932 
2933 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2934 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2935 	    MAXPATHLEN  + 1);
2936 
2937 	if (name == NULL) {
2938 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2939 		kmem_free(nm, len);
2940 		goto out;
2941 	}
2942 
2943 	*cs->statusp = resp->status = do_rfs4_op_lookup(name, req, cs);
2944 
2945 	if (name != nm)
2946 		kmem_free(name, MAXPATHLEN + 1);
2947 	kmem_free(nm, len);
2948 
2949 out:
2950 	DTRACE_NFSV4_2(op__lookup__done, struct compound_state *, cs,
2951 	    LOOKUP4res *, resp);
2952 }
2953 
2954 /* ARGSUSED */
2955 static void
2956 rfs4_op_lookupp(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
2957     struct compound_state *cs)
2958 {
2959 	LOOKUPP4res *resp = &resop->nfs_resop4_u.oplookupp;
2960 
2961 	DTRACE_NFSV4_1(op__lookupp__start, struct compound_state *, cs);
2962 
2963 	if (cs->vp == NULL) {
2964 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2965 		goto out;
2966 	}
2967 
2968 	if (cs->vp->v_type != VDIR) {
2969 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2970 		goto out;
2971 	}
2972 
2973 	*cs->statusp = resp->status = do_rfs4_op_lookup("..", req, cs);
2974 
2975 	/*
2976 	 * From NFSV4 Specification, LOOKUPP should not check for
2977 	 * NFS4ERR_WRONGSEC. Retrun NFS4_OK instead.
2978 	 */
2979 	if (resp->status == NFS4ERR_WRONGSEC) {
2980 		*cs->statusp = resp->status = NFS4_OK;
2981 	}
2982 
2983 out:
2984 	DTRACE_NFSV4_2(op__lookupp__done, struct compound_state *, cs,
2985 	    LOOKUPP4res *, resp);
2986 }
2987 
2988 
2989 /*ARGSUSED2*/
2990 static void
2991 rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2992     struct compound_state *cs)
2993 {
2994 	OPENATTR4args	*args = &argop->nfs_argop4_u.opopenattr;
2995 	OPENATTR4res	*resp = &resop->nfs_resop4_u.opopenattr;
2996 	vnode_t		*avp = NULL;
2997 	int		lookup_flags = LOOKUP_XATTR, error;
2998 	int		exp_ro = 0;
2999 
3000 	DTRACE_NFSV4_2(op__openattr__start, struct compound_state *, cs,
3001 	    OPENATTR4args *, args);
3002 
3003 	if (cs->vp == NULL) {
3004 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3005 		goto out;
3006 	}
3007 
3008 	if ((cs->vp->v_vfsp->vfs_flag & VFS_XATTR) == 0 &&
3009 	    !vfs_has_feature(cs->vp->v_vfsp, VFSFT_SYSATTR_VIEWS)) {
3010 		*cs->statusp = resp->status = puterrno4(ENOTSUP);
3011 		goto out;
3012 	}
3013 
3014 	/*
3015 	 * If file system supports passing ACE mask to VOP_ACCESS then
3016 	 * check for ACE_READ_NAMED_ATTRS, otherwise do legacy checks
3017 	 */
3018 
3019 	if (vfs_has_feature(cs->vp->v_vfsp, VFSFT_ACEMASKONACCESS))
3020 		error = VOP_ACCESS(cs->vp, ACE_READ_NAMED_ATTRS,
3021 		    V_ACE_MASK, cs->cr, NULL);
3022 	else
3023 		error = ((VOP_ACCESS(cs->vp, VREAD, 0, cs->cr, NULL) != 0) &&
3024 		    (VOP_ACCESS(cs->vp, VWRITE, 0, cs->cr, NULL) != 0) &&
3025 		    (VOP_ACCESS(cs->vp, VEXEC, 0, cs->cr, NULL) != 0));
3026 
3027 	if (error) {
3028 		*cs->statusp = resp->status = puterrno4(EACCES);
3029 		goto out;
3030 	}
3031 
3032 	/*
3033 	 * The CREATE_XATTR_DIR VOP flag cannot be specified if
3034 	 * the file system is exported read-only -- regardless of
3035 	 * createdir flag.  Otherwise the attrdir would be created
3036 	 * (assuming server fs isn't mounted readonly locally).  If
3037 	 * VOP_LOOKUP returns ENOENT in this case, the error will
3038 	 * be translated into EROFS.  ENOSYS is mapped to ENOTSUP
3039 	 * because specfs has no VOP_LOOKUP op, so the macro would
3040 	 * return ENOSYS.  EINVAL is returned by all (current)
3041 	 * Solaris file system implementations when any of their
3042 	 * restrictions are violated (xattr(dir) can't have xattrdir).
3043 	 * Returning NOTSUPP is more appropriate in this case
3044 	 * because the object will never be able to have an attrdir.
3045 	 */
3046 	if (args->createdir && ! (exp_ro = rdonly4(req, cs)))
3047 		lookup_flags |= CREATE_XATTR_DIR;
3048 
3049 	error = VOP_LOOKUP(cs->vp, "", &avp, NULL, lookup_flags, NULL, cs->cr,
3050 	    NULL, NULL, NULL);
3051 
3052 	if (error) {
3053 		if (error == ENOENT && args->createdir && exp_ro)
3054 			*cs->statusp = resp->status = puterrno4(EROFS);
3055 		else if (error == EINVAL || error == ENOSYS)
3056 			*cs->statusp = resp->status = puterrno4(ENOTSUP);
3057 		else
3058 			*cs->statusp = resp->status = puterrno4(error);
3059 		goto out;
3060 	}
3061 
3062 	ASSERT(avp->v_flag & V_XATTRDIR);
3063 
3064 	error = makefh4(&cs->fh, avp, cs->exi);
3065 
3066 	if (error) {
3067 		VN_RELE(avp);
3068 		*cs->statusp = resp->status = puterrno4(error);
3069 		goto out;
3070 	}
3071 
3072 	VN_RELE(cs->vp);
3073 	cs->vp = avp;
3074 
3075 	/*
3076 	 * There is no requirement for an attrdir fh flag
3077 	 * because the attrdir has a vnode flag to distinguish
3078 	 * it from regular (non-xattr) directories.  The
3079 	 * FH4_ATTRDIR flag is set for future sanity checks.
3080 	 */
3081 	set_fh4_flag(&cs->fh, FH4_ATTRDIR);
3082 	*cs->statusp = resp->status = NFS4_OK;
3083 
3084 out:
3085 	DTRACE_NFSV4_2(op__openattr__done, struct compound_state *, cs,
3086 	    OPENATTR4res *, resp);
3087 }
3088 
3089 static int
3090 do_io(int direction, vnode_t *vp, struct uio *uio, int ioflag, cred_t *cred,
3091     caller_context_t *ct)
3092 {
3093 	int error;
3094 	int i;
3095 	clock_t delaytime;
3096 
3097 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
3098 
3099 	/*
3100 	 * Don't block on mandatory locks. If this routine returns
3101 	 * EAGAIN, the caller should return NFS4ERR_LOCKED.
3102 	 */
3103 	uio->uio_fmode = FNONBLOCK;
3104 
3105 	for (i = 0; i < rfs4_maxlock_tries; i++) {
3106 
3107 
3108 		if (direction == FREAD) {
3109 			(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, ct);
3110 			error = VOP_READ(vp, uio, ioflag, cred, ct);
3111 			VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, ct);
3112 		} else {
3113 			(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, ct);
3114 			error = VOP_WRITE(vp, uio, ioflag, cred, ct);
3115 			VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, ct);
3116 		}
3117 
3118 		if (error != EAGAIN)
3119 			break;
3120 
3121 		if (i < rfs4_maxlock_tries - 1) {
3122 			delay(delaytime);
3123 			delaytime *= 2;
3124 		}
3125 	}
3126 
3127 	return (error);
3128 }
3129 
3130 /* ARGSUSED */
3131 static void
3132 rfs4_op_read(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3133     struct compound_state *cs)
3134 {
3135 	READ4args *args = &argop->nfs_argop4_u.opread;
3136 	READ4res *resp = &resop->nfs_resop4_u.opread;
3137 	int error;
3138 	int verror;
3139 	vnode_t *vp;
3140 	struct vattr va;
3141 	struct iovec iov, *iovp = NULL;
3142 	int iovcnt;
3143 	struct uio uio;
3144 	u_offset_t offset;
3145 	bool_t *deleg = &cs->deleg;
3146 	nfsstat4 stat;
3147 	int in_crit = 0;
3148 	mblk_t *mp = NULL;
3149 	int alloc_err = 0;
3150 	int rdma_used = 0;
3151 	int loaned_buffers;
3152 	caller_context_t ct;
3153 	struct uio *uiop;
3154 
3155 	DTRACE_NFSV4_2(op__read__start, struct compound_state *, cs,
3156 	    READ4args, args);
3157 
3158 	vp = cs->vp;
3159 	if (vp == NULL) {
3160 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3161 		goto out;
3162 	}
3163 	if (cs->access == CS_ACCESS_DENIED) {
3164 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3165 		goto out;
3166 	}
3167 
3168 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3169 	    deleg, TRUE, &ct)) != NFS4_OK) {
3170 		*cs->statusp = resp->status = stat;
3171 		goto out;
3172 	}
3173 
3174 	/*
3175 	 * Enter the critical region before calling VOP_RWLOCK
3176 	 * to avoid a deadlock with write requests.
3177 	 */
3178 	if (nbl_need_check(vp)) {
3179 		nbl_start_crit(vp, RW_READER);
3180 		in_crit = 1;
3181 		if (nbl_conflict(vp, NBL_READ, args->offset, args->count, 0,
3182 		    &ct)) {
3183 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
3184 			goto out;
3185 		}
3186 	}
3187 
3188 	if (args->wlist) {
3189 		if (args->count > clist_len(args->wlist)) {
3190 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3191 			goto out;
3192 		}
3193 		rdma_used = 1;
3194 	}
3195 
3196 	/* use loaned buffers for TCP */
3197 	loaned_buffers = (nfs_loaned_buffers && !rdma_used) ? 1 : 0;
3198 
3199 	va.va_mask = AT_MODE|AT_SIZE|AT_UID;
3200 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3201 
3202 	/*
3203 	 * If we can't get the attributes, then we can't do the
3204 	 * right access checking.  So, we'll fail the request.
3205 	 */
3206 	if (verror) {
3207 		*cs->statusp = resp->status = puterrno4(verror);
3208 		goto out;
3209 	}
3210 
3211 	if (vp->v_type != VREG) {
3212 		*cs->statusp = resp->status =
3213 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
3214 		goto out;
3215 	}
3216 
3217 	if (crgetuid(cs->cr) != va.va_uid &&
3218 	    (error = VOP_ACCESS(vp, VREAD, 0, cs->cr, &ct)) &&
3219 	    (error = VOP_ACCESS(vp, VEXEC, 0, cs->cr, &ct))) {
3220 		*cs->statusp = resp->status = puterrno4(error);
3221 		goto out;
3222 	}
3223 
3224 	if (MANDLOCK(vp, va.va_mode)) { /* XXX - V4 supports mand locking */
3225 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3226 		goto out;
3227 	}
3228 
3229 	offset = args->offset;
3230 	if (offset >= va.va_size) {
3231 		*cs->statusp = resp->status = NFS4_OK;
3232 		resp->eof = TRUE;
3233 		resp->data_len = 0;
3234 		resp->data_val = NULL;
3235 		resp->mblk = NULL;
3236 		/* RDMA */
3237 		resp->wlist = args->wlist;
3238 		resp->wlist_len = resp->data_len;
3239 		*cs->statusp = resp->status = NFS4_OK;
3240 		if (resp->wlist)
3241 			clist_zero_len(resp->wlist);
3242 		goto out;
3243 	}
3244 
3245 	if (args->count == 0) {
3246 		*cs->statusp = resp->status = NFS4_OK;
3247 		resp->eof = FALSE;
3248 		resp->data_len = 0;
3249 		resp->data_val = NULL;
3250 		resp->mblk = NULL;
3251 		/* RDMA */
3252 		resp->wlist = args->wlist;
3253 		resp->wlist_len = resp->data_len;
3254 		if (resp->wlist)
3255 			clist_zero_len(resp->wlist);
3256 		goto out;
3257 	}
3258 
3259 	/*
3260 	 * Do not allocate memory more than maximum allowed
3261 	 * transfer size
3262 	 */
3263 	if (args->count > rfs4_tsize(req))
3264 		args->count = rfs4_tsize(req);
3265 
3266 	if (loaned_buffers) {
3267 		uiop = (uio_t *)rfs_setup_xuio(vp);
3268 		ASSERT(uiop != NULL);
3269 		uiop->uio_segflg = UIO_SYSSPACE;
3270 		uiop->uio_loffset = args->offset;
3271 		uiop->uio_resid = args->count;
3272 
3273 		/* Jump to do the read if successful */
3274 		if (!VOP_REQZCBUF(vp, UIO_READ, (xuio_t *)uiop, cs->cr, &ct)) {
3275 			/*
3276 			 * Need to hold the vnode until after VOP_RETZCBUF()
3277 			 * is called.
3278 			 */
3279 			VN_HOLD(vp);
3280 			goto doio_read;
3281 		}
3282 
3283 		DTRACE_PROBE2(nfss__i__reqzcbuf_failed, int,
3284 		    uiop->uio_loffset, int, uiop->uio_resid);
3285 
3286 		uiop->uio_extflg = 0;
3287 
3288 		/* failure to setup for zero copy */
3289 		rfs_free_xuio((void *)uiop);
3290 		loaned_buffers = 0;
3291 	}
3292 
3293 	/*
3294 	 * If returning data via RDMA Write, then grab the chunk list. If we
3295 	 * aren't returning READ data w/RDMA_WRITE, then grab a mblk.
3296 	 */
3297 	if (rdma_used) {
3298 		mp = NULL;
3299 		(void) rdma_get_wchunk(req, &iov, args->wlist);
3300 		uio.uio_iov = &iov;
3301 		uio.uio_iovcnt = 1;
3302 	} else {
3303 		/*
3304 		 * mp will contain the data to be sent out in the read reply.
3305 		 * It will be freed after the reply has been sent.
3306 		 */
3307 		mp = rfs_read_alloc(args->count, &iovp, &iovcnt);
3308 		ASSERT(mp != NULL);
3309 		ASSERT(alloc_err == 0);
3310 		uio.uio_iov = iovp;
3311 		uio.uio_iovcnt = iovcnt;
3312 	}
3313 
3314 	uio.uio_segflg = UIO_SYSSPACE;
3315 	uio.uio_extflg = UIO_COPY_CACHED;
3316 	uio.uio_loffset = args->offset;
3317 	uio.uio_resid = args->count;
3318 	uiop = &uio;
3319 
3320 doio_read:
3321 	error = do_io(FREAD, vp, uiop, 0, cs->cr, &ct);
3322 
3323 	va.va_mask = AT_SIZE;
3324 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3325 
3326 	if (error) {
3327 		if (mp)
3328 			freemsg(mp);
3329 		*cs->statusp = resp->status = puterrno4(error);
3330 		goto out;
3331 	}
3332 
3333 	/* make mblk using zc buffers */
3334 	if (loaned_buffers) {
3335 		mp = uio_to_mblk(uiop);
3336 		ASSERT(mp != NULL);
3337 	}
3338 
3339 	*cs->statusp = resp->status = NFS4_OK;
3340 
3341 	ASSERT(uiop->uio_resid >= 0);
3342 	resp->data_len = args->count - uiop->uio_resid;
3343 	if (mp) {
3344 		resp->data_val = (char *)mp->b_datap->db_base;
3345 		rfs_rndup_mblks(mp, resp->data_len, loaned_buffers);
3346 	} else {
3347 		resp->data_val = (caddr_t)iov.iov_base;
3348 	}
3349 
3350 	resp->mblk = mp;
3351 
3352 	if (!verror && offset + resp->data_len == va.va_size)
3353 		resp->eof = TRUE;
3354 	else
3355 		resp->eof = FALSE;
3356 
3357 	if (rdma_used) {
3358 		if (!rdma_setup_read_data4(args, resp)) {
3359 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3360 		}
3361 	} else {
3362 		resp->wlist = NULL;
3363 	}
3364 
3365 out:
3366 	if (in_crit)
3367 		nbl_end_crit(vp);
3368 
3369 	if (iovp != NULL)
3370 		kmem_free(iovp, iovcnt * sizeof (struct iovec));
3371 
3372 	DTRACE_NFSV4_2(op__read__done, struct compound_state *, cs,
3373 	    READ4res *, resp);
3374 }
3375 
3376 static void
3377 rfs4_op_read_free(nfs_resop4 *resop)
3378 {
3379 	READ4res	*resp = &resop->nfs_resop4_u.opread;
3380 
3381 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3382 		freemsg(resp->mblk);
3383 		resp->mblk = NULL;
3384 		resp->data_val = NULL;
3385 		resp->data_len = 0;
3386 	}
3387 }
3388 
3389 static void
3390 rfs4_op_readdir_free(nfs_resop4 * resop)
3391 {
3392 	READDIR4res    *resp = &resop->nfs_resop4_u.opreaddir;
3393 
3394 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3395 		freeb(resp->mblk);
3396 		resp->mblk = NULL;
3397 		resp->data_len = 0;
3398 	}
3399 }
3400 
3401 
3402 /* ARGSUSED */
3403 static void
3404 rfs4_op_putpubfh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
3405     struct compound_state *cs)
3406 {
3407 	PUTPUBFH4res	*resp = &resop->nfs_resop4_u.opputpubfh;
3408 	int		error;
3409 	vnode_t		*vp;
3410 	struct exportinfo *exi, *sav_exi;
3411 	nfs_fh4_fmt_t	*fh_fmtp;
3412 
3413 	DTRACE_NFSV4_1(op__putpubfh__start, struct compound_state *, cs);
3414 
3415 	if (cs->vp) {
3416 		VN_RELE(cs->vp);
3417 		cs->vp = NULL;
3418 	}
3419 
3420 	if (cs->cr)
3421 		crfree(cs->cr);
3422 
3423 	cs->cr = crdup(cs->basecr);
3424 
3425 	vp = exi_public->exi_vp;
3426 	if (vp == NULL) {
3427 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3428 		goto out;
3429 	}
3430 
3431 	error = makefh4(&cs->fh, vp, exi_public);
3432 	if (error != 0) {
3433 		*cs->statusp = resp->status = puterrno4(error);
3434 		goto out;
3435 	}
3436 	sav_exi = cs->exi;
3437 	if (exi_public == exi_root) {
3438 		/*
3439 		 * No filesystem is actually shared public, so we default
3440 		 * to exi_root. In this case, we must check whether root
3441 		 * is exported.
3442 		 */
3443 		fh_fmtp = (nfs_fh4_fmt_t *)cs->fh.nfs_fh4_val;
3444 
3445 		/*
3446 		 * if root filesystem is exported, the exportinfo struct that we
3447 		 * should use is what checkexport4 returns, because root_exi is
3448 		 * actually a mostly empty struct.
3449 		 */
3450 		exi = checkexport4(&fh_fmtp->fh4_fsid,
3451 		    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
3452 		cs->exi = ((exi != NULL) ? exi : exi_public);
3453 	} else {
3454 		/*
3455 		 * it's a properly shared filesystem
3456 		 */
3457 		cs->exi = exi_public;
3458 	}
3459 
3460 	if (is_system_labeled()) {
3461 		bslabel_t *clabel;
3462 
3463 		ASSERT(req->rq_label != NULL);
3464 		clabel = req->rq_label;
3465 		DTRACE_PROBE2(tx__rfs4__log__info__opputpubfh__clabel, char *,
3466 		    "got client label from request(1)",
3467 		    struct svc_req *, req);
3468 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
3469 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
3470 			    cs->exi)) {
3471 				*cs->statusp = resp->status =
3472 				    NFS4ERR_SERVERFAULT;
3473 				goto out;
3474 			}
3475 		}
3476 	}
3477 
3478 	VN_HOLD(vp);
3479 	cs->vp = vp;
3480 
3481 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3482 		VN_RELE(cs->vp);
3483 		cs->vp = NULL;
3484 		cs->exi = sav_exi;
3485 		goto out;
3486 	}
3487 
3488 	*cs->statusp = resp->status = NFS4_OK;
3489 out:
3490 	DTRACE_NFSV4_2(op__putpubfh__done, struct compound_state *, cs,
3491 	    PUTPUBFH4res *, resp);
3492 }
3493 
3494 /*
3495  * XXX - issue with put*fh operations. Suppose /export/home is exported.
3496  * Suppose an NFS client goes to mount /export/home/joe. If /export, home,
3497  * or joe have restrictive search permissions, then we shouldn't let
3498  * the client get a file handle. This is easy to enforce. However, we
3499  * don't know what security flavor should be used until we resolve the
3500  * path name. Another complication is uid mapping. If root is
3501  * the user, then it will be mapped to the anonymous user by default,
3502  * but we won't know that till we've resolved the path name. And we won't
3503  * know what the anonymous user is.
3504  * Luckily, SECINFO is specified to take a full filename.
3505  * So what we will have to in rfs4_op_lookup is check that flavor of
3506  * the target object matches that of the request, and if root was the
3507  * caller, check for the root= and anon= options, and if necessary,
3508  * repeat the lookup using the right cred_t. But that's not done yet.
3509  */
3510 /* ARGSUSED */
3511 static void
3512 rfs4_op_putfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3513     struct compound_state *cs)
3514 {
3515 	PUTFH4args *args = &argop->nfs_argop4_u.opputfh;
3516 	PUTFH4res *resp = &resop->nfs_resop4_u.opputfh;
3517 	nfs_fh4_fmt_t *fh_fmtp;
3518 
3519 	DTRACE_NFSV4_2(op__putfh__start, struct compound_state *, cs,
3520 	    PUTFH4args *, args);
3521 
3522 	if (cs->vp) {
3523 		VN_RELE(cs->vp);
3524 		cs->vp = NULL;
3525 	}
3526 
3527 	if (cs->cr) {
3528 		crfree(cs->cr);
3529 		cs->cr = NULL;
3530 	}
3531 
3532 
3533 	if (args->object.nfs_fh4_len < NFS_FH4_LEN) {
3534 		*cs->statusp = resp->status = NFS4ERR_BADHANDLE;
3535 		goto out;
3536 	}
3537 
3538 	fh_fmtp = (nfs_fh4_fmt_t *)args->object.nfs_fh4_val;
3539 	cs->exi = checkexport4(&fh_fmtp->fh4_fsid, (fid_t *)&fh_fmtp->fh4_xlen,
3540 	    NULL);
3541 
3542 	if (cs->exi == NULL) {
3543 		*cs->statusp = resp->status = NFS4ERR_STALE;
3544 		goto out;
3545 	}
3546 
3547 	cs->cr = crdup(cs->basecr);
3548 
3549 	ASSERT(cs->cr != NULL);
3550 
3551 	if (! (cs->vp = nfs4_fhtovp(&args->object, cs->exi, &resp->status))) {
3552 		*cs->statusp = resp->status;
3553 		goto out;
3554 	}
3555 
3556 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3557 		VN_RELE(cs->vp);
3558 		cs->vp = NULL;
3559 		goto out;
3560 	}
3561 
3562 	nfs_fh4_copy(&args->object, &cs->fh);
3563 	*cs->statusp = resp->status = NFS4_OK;
3564 	cs->deleg = FALSE;
3565 
3566 out:
3567 	DTRACE_NFSV4_2(op__putfh__done, struct compound_state *, cs,
3568 	    PUTFH4res *, resp);
3569 }
3570 
3571 /* ARGSUSED */
3572 static void
3573 rfs4_op_putrootfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3574     struct compound_state *cs)
3575 {
3576 	PUTROOTFH4res *resp = &resop->nfs_resop4_u.opputrootfh;
3577 	int error;
3578 	fid_t fid;
3579 	struct exportinfo *exi, *sav_exi;
3580 
3581 	DTRACE_NFSV4_1(op__putrootfh__start, struct compound_state *, cs);
3582 
3583 	if (cs->vp) {
3584 		VN_RELE(cs->vp);
3585 		cs->vp = NULL;
3586 	}
3587 
3588 	if (cs->cr)
3589 		crfree(cs->cr);
3590 
3591 	cs->cr = crdup(cs->basecr);
3592 
3593 	/*
3594 	 * Using rootdir, the system root vnode,
3595 	 * get its fid.
3596 	 */
3597 	bzero(&fid, sizeof (fid));
3598 	fid.fid_len = MAXFIDSZ;
3599 	error = vop_fid_pseudo(rootdir, &fid);
3600 	if (error != 0) {
3601 		*cs->statusp = resp->status = puterrno4(error);
3602 		goto out;
3603 	}
3604 
3605 	/*
3606 	 * Then use the root fsid & fid it to find out if it's exported
3607 	 *
3608 	 * If the server root isn't exported directly, then
3609 	 * it should at least be a pseudo export based on
3610 	 * one or more exports further down in the server's
3611 	 * file tree.
3612 	 */
3613 	exi = checkexport4(&rootdir->v_vfsp->vfs_fsid, &fid, NULL);
3614 	if (exi == NULL || exi->exi_export.ex_flags & EX_PUBLIC) {
3615 		NFS4_DEBUG(rfs4_debug,
3616 		    (CE_WARN, "rfs4_op_putrootfh: export check failure"));
3617 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3618 		goto out;
3619 	}
3620 
3621 	/*
3622 	 * Now make a filehandle based on the root
3623 	 * export and root vnode.
3624 	 */
3625 	error = makefh4(&cs->fh, rootdir, exi);
3626 	if (error != 0) {
3627 		*cs->statusp = resp->status = puterrno4(error);
3628 		goto out;
3629 	}
3630 
3631 	sav_exi = cs->exi;
3632 	cs->exi = exi;
3633 
3634 	VN_HOLD(rootdir);
3635 	cs->vp = rootdir;
3636 
3637 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3638 		VN_RELE(rootdir);
3639 		cs->vp = NULL;
3640 		cs->exi = sav_exi;
3641 		goto out;
3642 	}
3643 
3644 	*cs->statusp = resp->status = NFS4_OK;
3645 	cs->deleg = FALSE;
3646 out:
3647 	DTRACE_NFSV4_2(op__putrootfh__done, struct compound_state *, cs,
3648 	    PUTROOTFH4res *, resp);
3649 }
3650 
3651 /*
3652  * set_rdattr_params sets up the variables used to manage what information
3653  * to get for each directory entry.
3654  */
3655 static nfsstat4
3656 set_rdattr_params(struct nfs4_svgetit_arg *sargp,
3657     bitmap4 attrs, bool_t *need_to_lookup)
3658 {
3659 	uint_t	va_mask;
3660 	nfsstat4 status;
3661 	bitmap4 objbits;
3662 
3663 	status = bitmap4_to_attrmask(attrs, sargp);
3664 	if (status != NFS4_OK) {
3665 		/*
3666 		 * could not even figure attr mask
3667 		 */
3668 		return (status);
3669 	}
3670 	va_mask = sargp->vap->va_mask;
3671 
3672 	/*
3673 	 * dirent's d_ino is always correct value for mounted_on_fileid.
3674 	 * mntdfid_set is set once here, but mounted_on_fileid is
3675 	 * set in main dirent processing loop for each dirent.
3676 	 * The mntdfid_set is a simple optimization that lets the
3677 	 * server attr code avoid work when caller is readdir.
3678 	 */
3679 	sargp->mntdfid_set = TRUE;
3680 
3681 	/*
3682 	 * Lookup entry only if client asked for any of the following:
3683 	 * a) vattr attrs
3684 	 * b) vfs attrs
3685 	 * c) attrs w/per-object scope requested (change, filehandle, etc)
3686 	 *    other than mounted_on_fileid (which we can take from dirent)
3687 	 */
3688 	objbits = attrs ? attrs & NFS4_VP_ATTR_MASK : 0;
3689 
3690 	if (va_mask || sargp->sbp || (objbits & ~FATTR4_MOUNTED_ON_FILEID_MASK))
3691 		*need_to_lookup = TRUE;
3692 	else
3693 		*need_to_lookup = FALSE;
3694 
3695 	if (sargp->sbp == NULL)
3696 		return (NFS4_OK);
3697 
3698 	/*
3699 	 * If filesystem attrs are requested, get them now from the
3700 	 * directory vp, as most entries will have same filesystem. The only
3701 	 * exception are mounted over entries but we handle
3702 	 * those as we go (XXX mounted over detection not yet implemented).
3703 	 */
3704 	sargp->vap->va_mask = 0;	/* to avoid VOP_GETATTR */
3705 	status = bitmap4_get_sysattrs(sargp);
3706 	sargp->vap->va_mask = va_mask;
3707 
3708 	if ((status != NFS4_OK) && sargp->rdattr_error_req) {
3709 		/*
3710 		 * Failed to get filesystem attributes.
3711 		 * Return a rdattr_error for each entry, but don't fail.
3712 		 * However, don't get any obj-dependent attrs.
3713 		 */
3714 		sargp->rdattr_error = status;	/* for rdattr_error */
3715 		*need_to_lookup = FALSE;
3716 		/*
3717 		 * At least get fileid for regular readdir output
3718 		 */
3719 		sargp->vap->va_mask &= AT_NODEID;
3720 		status = NFS4_OK;
3721 	}
3722 
3723 	return (status);
3724 }
3725 
3726 /*
3727  * readlink: args: CURRENT_FH.
3728  *	res: status. If success - CURRENT_FH unchanged, return linktext.
3729  */
3730 
3731 /* ARGSUSED */
3732 static void
3733 rfs4_op_readlink(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3734     struct compound_state *cs)
3735 {
3736 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3737 	int error;
3738 	vnode_t *vp;
3739 	struct iovec iov;
3740 	struct vattr va;
3741 	struct uio uio;
3742 	char *data;
3743 	struct sockaddr *ca;
3744 	char *name = NULL;
3745 	int is_referral;
3746 
3747 	DTRACE_NFSV4_1(op__readlink__start, struct compound_state *, cs);
3748 
3749 	/* CURRENT_FH: directory */
3750 	vp = cs->vp;
3751 	if (vp == NULL) {
3752 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3753 		goto out;
3754 	}
3755 
3756 	if (cs->access == CS_ACCESS_DENIED) {
3757 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3758 		goto out;
3759 	}
3760 
3761 	/* Is it a referral? */
3762 	if (vn_is_nfs_reparse(vp, cs->cr) && client_is_downrev(req)) {
3763 
3764 		is_referral = 1;
3765 
3766 	} else {
3767 
3768 		is_referral = 0;
3769 
3770 		if (vp->v_type == VDIR) {
3771 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
3772 			goto out;
3773 		}
3774 
3775 		if (vp->v_type != VLNK) {
3776 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3777 			goto out;
3778 		}
3779 
3780 	}
3781 
3782 	va.va_mask = AT_MODE;
3783 	error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
3784 	if (error) {
3785 		*cs->statusp = resp->status = puterrno4(error);
3786 		goto out;
3787 	}
3788 
3789 	if (MANDLOCK(vp, va.va_mode)) {
3790 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3791 		goto out;
3792 	}
3793 
3794 	data = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);
3795 
3796 	if (is_referral) {
3797 		char *s;
3798 		size_t strsz;
3799 
3800 		/* Get an artificial symlink based on a referral */
3801 		s = build_symlink(vp, cs->cr, &strsz);
3802 		global_svstat_ptr[4][NFS_REFERLINKS].value.ui64++;
3803 		DTRACE_PROBE2(nfs4serv__func__referral__reflink,
3804 		    vnode_t *, vp, char *, s);
3805 		if (s == NULL)
3806 			error = EINVAL;
3807 		else {
3808 			error = 0;
3809 			(void) strlcpy(data, s, MAXPATHLEN + 1);
3810 			kmem_free(s, strsz);
3811 		}
3812 
3813 	} else {
3814 
3815 		iov.iov_base = data;
3816 		iov.iov_len = MAXPATHLEN;
3817 		uio.uio_iov = &iov;
3818 		uio.uio_iovcnt = 1;
3819 		uio.uio_segflg = UIO_SYSSPACE;
3820 		uio.uio_extflg = UIO_COPY_CACHED;
3821 		uio.uio_loffset = 0;
3822 		uio.uio_resid = MAXPATHLEN;
3823 
3824 		error = VOP_READLINK(vp, &uio, cs->cr, NULL);
3825 
3826 		if (!error)
3827 			*(data + MAXPATHLEN - uio.uio_resid) = '\0';
3828 	}
3829 
3830 	if (error) {
3831 		kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3832 		*cs->statusp = resp->status = puterrno4(error);
3833 		goto out;
3834 	}
3835 
3836 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
3837 	name = nfscmd_convname(ca, cs->exi, data, NFSCMD_CONV_OUTBOUND,
3838 	    MAXPATHLEN  + 1);
3839 
3840 	if (name == NULL) {
3841 		/*
3842 		 * Even though the conversion failed, we return
3843 		 * something. We just don't translate it.
3844 		 */
3845 		name = data;
3846 	}
3847 
3848 	/*
3849 	 * treat link name as data
3850 	 */
3851 	(void) str_to_utf8(name, (utf8string *)&resp->link);
3852 
3853 	if (name != data)
3854 		kmem_free(name, MAXPATHLEN + 1);
3855 	kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3856 	*cs->statusp = resp->status = NFS4_OK;
3857 
3858 out:
3859 	DTRACE_NFSV4_2(op__readlink__done, struct compound_state *, cs,
3860 	    READLINK4res *, resp);
3861 }
3862 
3863 static void
3864 rfs4_op_readlink_free(nfs_resop4 *resop)
3865 {
3866 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3867 	utf8string *symlink = (utf8string *)&resp->link;
3868 
3869 	if (symlink->utf8string_val) {
3870 		UTF8STRING_FREE(*symlink)
3871 	}
3872 }
3873 
3874 /*
3875  * release_lockowner:
3876  *	Release any state associated with the supplied
3877  *	lockowner. Note if any lo_state is holding locks we will not
3878  *	rele that lo_state and thus the lockowner will not be destroyed.
3879  *	A client using lock after the lock owner stateid has been released
3880  *	will suffer the consequence of NFS4ERR_BAD_STATEID and would have
3881  *	to reissue the lock with new_lock_owner set to TRUE.
3882  *	args: lock_owner
3883  *	res:  status
3884  */
3885 /* ARGSUSED */
3886 static void
3887 rfs4_op_release_lockowner(nfs_argop4 *argop, nfs_resop4 *resop,
3888     struct svc_req *req, struct compound_state *cs)
3889 {
3890 	RELEASE_LOCKOWNER4args *ap = &argop->nfs_argop4_u.oprelease_lockowner;
3891 	RELEASE_LOCKOWNER4res *resp = &resop->nfs_resop4_u.oprelease_lockowner;
3892 	rfs4_lockowner_t *lo;
3893 	rfs4_openowner_t *oo;
3894 	rfs4_state_t *sp;
3895 	rfs4_lo_state_t *lsp;
3896 	rfs4_client_t *cp;
3897 	bool_t create = FALSE;
3898 	locklist_t *llist;
3899 	sysid_t sysid;
3900 
3901 	DTRACE_NFSV4_2(op__release__lockowner__start, struct compound_state *,
3902 	    cs, RELEASE_LOCKOWNER4args *, ap);
3903 
3904 	/* Make sure there is a clientid around for this request */
3905 	cp = rfs4_findclient_by_id(ap->lock_owner.clientid, FALSE);
3906 
3907 	if (cp == NULL) {
3908 		*cs->statusp = resp->status =
3909 		    rfs4_check_clientid(&ap->lock_owner.clientid, 0);
3910 		goto out;
3911 	}
3912 	rfs4_client_rele(cp);
3913 
3914 	lo = rfs4_findlockowner(&ap->lock_owner, &create);
3915 	if (lo == NULL) {
3916 		*cs->statusp = resp->status = NFS4_OK;
3917 		goto out;
3918 	}
3919 	ASSERT(lo->rl_client != NULL);
3920 
3921 	/*
3922 	 * Check for EXPIRED client. If so will reap state with in a lease
3923 	 * period or on next set_clientid_confirm step
3924 	 */
3925 	if (rfs4_lease_expired(lo->rl_client)) {
3926 		rfs4_lockowner_rele(lo);
3927 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
3928 		goto out;
3929 	}
3930 
3931 	/*
3932 	 * If no sysid has been assigned, then no locks exist; just return.
3933 	 */
3934 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3935 	if (lo->rl_client->rc_sysidt == LM_NOSYSID) {
3936 		rfs4_lockowner_rele(lo);
3937 		rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3938 		goto out;
3939 	}
3940 
3941 	sysid = lo->rl_client->rc_sysidt;
3942 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3943 
3944 	/*
3945 	 * Mark the lockowner invalid.
3946 	 */
3947 	rfs4_dbe_hide(lo->rl_dbe);
3948 
3949 	/*
3950 	 * sysid-pid pair should now not be used since the lockowner is
3951 	 * invalid. If the client were to instantiate the lockowner again
3952 	 * it would be assigned a new pid. Thus we can get the list of
3953 	 * current locks.
3954 	 */
3955 
3956 	llist = flk_get_active_locks(sysid, lo->rl_pid);
3957 	/* If we are still holding locks fail */
3958 	if (llist != NULL) {
3959 
3960 		*cs->statusp = resp->status = NFS4ERR_LOCKS_HELD;
3961 
3962 		flk_free_locklist(llist);
3963 		/*
3964 		 * We need to unhide the lockowner so the client can
3965 		 * try it again. The bad thing here is if the client
3966 		 * has a logic error that took it here in the first place
3967 		 * they probably have lost accounting of the locks that it
3968 		 * is holding. So we may have dangling state until the
3969 		 * open owner state is reaped via close. One scenario
3970 		 * that could possibly occur is that the client has
3971 		 * sent the unlock request(s) in separate threads
3972 		 * and has not waited for the replies before sending the
3973 		 * RELEASE_LOCKOWNER request. Presumably, it would expect
3974 		 * and deal appropriately with NFS4ERR_LOCKS_HELD, by
3975 		 * reissuing the request.
3976 		 */
3977 		rfs4_dbe_unhide(lo->rl_dbe);
3978 		rfs4_lockowner_rele(lo);
3979 		goto out;
3980 	}
3981 
3982 	/*
3983 	 * For the corresponding client we need to check each open
3984 	 * owner for any opens that have lockowner state associated
3985 	 * with this lockowner.
3986 	 */
3987 
3988 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3989 	for (oo = list_head(&lo->rl_client->rc_openownerlist); oo != NULL;
3990 	    oo = list_next(&lo->rl_client->rc_openownerlist, oo)) {
3991 
3992 		rfs4_dbe_lock(oo->ro_dbe);
3993 		for (sp = list_head(&oo->ro_statelist); sp != NULL;
3994 		    sp = list_next(&oo->ro_statelist, sp)) {
3995 
3996 			rfs4_dbe_lock(sp->rs_dbe);
3997 			for (lsp = list_head(&sp->rs_lostatelist);
3998 			    lsp != NULL;
3999 			    lsp = list_next(&sp->rs_lostatelist, lsp)) {
4000 				if (lsp->rls_locker == lo) {
4001 					rfs4_dbe_lock(lsp->rls_dbe);
4002 					rfs4_dbe_invalidate(lsp->rls_dbe);
4003 					rfs4_dbe_unlock(lsp->rls_dbe);
4004 				}
4005 			}
4006 			rfs4_dbe_unlock(sp->rs_dbe);
4007 		}
4008 		rfs4_dbe_unlock(oo->ro_dbe);
4009 	}
4010 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
4011 
4012 	rfs4_lockowner_rele(lo);
4013 
4014 	*cs->statusp = resp->status = NFS4_OK;
4015 
4016 out:
4017 	DTRACE_NFSV4_2(op__release__lockowner__done, struct compound_state *,
4018 	    cs, RELEASE_LOCKOWNER4res *, resp);
4019 }
4020 
4021 /*
4022  * short utility function to lookup a file and recall the delegation
4023  */
4024 static rfs4_file_t *
4025 rfs4_lookup_and_findfile(vnode_t *dvp, char *nm, vnode_t **vpp,
4026     int *lkup_error, cred_t *cr)
4027 {
4028 	vnode_t *vp;
4029 	rfs4_file_t *fp = NULL;
4030 	bool_t fcreate = FALSE;
4031 	int error;
4032 
4033 	if (vpp)
4034 		*vpp = NULL;
4035 
4036 	if ((error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cr, NULL, NULL,
4037 	    NULL)) == 0) {
4038 		if (vp->v_type == VREG)
4039 			fp = rfs4_findfile(vp, NULL, &fcreate);
4040 		if (vpp)
4041 			*vpp = vp;
4042 		else
4043 			VN_RELE(vp);
4044 	}
4045 
4046 	if (lkup_error)
4047 		*lkup_error = error;
4048 
4049 	return (fp);
4050 }
4051 
4052 /*
4053  * remove: args: CURRENT_FH: directory; name.
4054  *	res: status. If success - CURRENT_FH unchanged, return change_info
4055  *		for directory.
4056  */
4057 /* ARGSUSED */
4058 static void
4059 rfs4_op_remove(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4060     struct compound_state *cs)
4061 {
4062 	REMOVE4args *args = &argop->nfs_argop4_u.opremove;
4063 	REMOVE4res *resp = &resop->nfs_resop4_u.opremove;
4064 	int error;
4065 	vnode_t *dvp, *vp;
4066 	struct vattr bdva, idva, adva;
4067 	char *nm;
4068 	uint_t len;
4069 	rfs4_file_t *fp;
4070 	int in_crit = 0;
4071 	bslabel_t *clabel;
4072 	struct sockaddr *ca;
4073 	char *name = NULL;
4074 	nfsstat4 status;
4075 
4076 	DTRACE_NFSV4_2(op__remove__start, struct compound_state *, cs,
4077 	    REMOVE4args *, args);
4078 
4079 	/* CURRENT_FH: directory */
4080 	dvp = cs->vp;
4081 	if (dvp == NULL) {
4082 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4083 		goto out;
4084 	}
4085 
4086 	if (cs->access == CS_ACCESS_DENIED) {
4087 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4088 		goto out;
4089 	}
4090 
4091 	/*
4092 	 * If there is an unshared filesystem mounted on this vnode,
4093 	 * Do not allow to remove anything in this directory.
4094 	 */
4095 	if (vn_ismntpt(dvp)) {
4096 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4097 		goto out;
4098 	}
4099 
4100 	if (dvp->v_type != VDIR) {
4101 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4102 		goto out;
4103 	}
4104 
4105 	status = utf8_dir_verify(&args->target);
4106 	if (status != NFS4_OK) {
4107 		*cs->statusp = resp->status = status;
4108 		goto out;
4109 	}
4110 
4111 	/*
4112 	 * Lookup the file so that we can check if it's a directory
4113 	 */
4114 	nm = utf8_to_fn(&args->target, &len, NULL);
4115 	if (nm == NULL) {
4116 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4117 		goto out;
4118 	}
4119 
4120 	if (len > MAXNAMELEN) {
4121 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4122 		kmem_free(nm, len);
4123 		goto out;
4124 	}
4125 
4126 	if (rdonly4(req, cs)) {
4127 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4128 		kmem_free(nm, len);
4129 		goto out;
4130 	}
4131 
4132 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4133 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
4134 	    MAXPATHLEN  + 1);
4135 
4136 	if (name == NULL) {
4137 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4138 		kmem_free(nm, len);
4139 		goto out;
4140 	}
4141 
4142 	/*
4143 	 * Lookup the file to determine type and while we are see if
4144 	 * there is a file struct around and check for delegation.
4145 	 * We don't need to acquire va_seq before this lookup, if
4146 	 * it causes an update, cinfo.before will not match, which will
4147 	 * trigger a cache flush even if atomic is TRUE.
4148 	 */
4149 	if (fp = rfs4_lookup_and_findfile(dvp, name, &vp, &error, cs->cr)) {
4150 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4151 		    NULL)) {
4152 			VN_RELE(vp);
4153 			rfs4_file_rele(fp);
4154 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4155 			if (nm != name)
4156 				kmem_free(name, MAXPATHLEN + 1);
4157 			kmem_free(nm, len);
4158 			goto out;
4159 		}
4160 	}
4161 
4162 	/* Didn't find anything to remove */
4163 	if (vp == NULL) {
4164 		*cs->statusp = resp->status = error;
4165 		if (nm != name)
4166 			kmem_free(name, MAXPATHLEN + 1);
4167 		kmem_free(nm, len);
4168 		goto out;
4169 	}
4170 
4171 	if (nbl_need_check(vp)) {
4172 		nbl_start_crit(vp, RW_READER);
4173 		in_crit = 1;
4174 		if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
4175 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4176 			if (nm != name)
4177 				kmem_free(name, MAXPATHLEN + 1);
4178 			kmem_free(nm, len);
4179 			nbl_end_crit(vp);
4180 			VN_RELE(vp);
4181 			if (fp) {
4182 				rfs4_clear_dont_grant(fp);
4183 				rfs4_file_rele(fp);
4184 			}
4185 			goto out;
4186 		}
4187 	}
4188 
4189 	/* check label before allowing removal */
4190 	if (is_system_labeled()) {
4191 		ASSERT(req->rq_label != NULL);
4192 		clabel = req->rq_label;
4193 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
4194 		    "got client label from request(1)",
4195 		    struct svc_req *, req);
4196 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4197 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
4198 			    cs->exi)) {
4199 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4200 				if (name != nm)
4201 					kmem_free(name, MAXPATHLEN + 1);
4202 				kmem_free(nm, len);
4203 				if (in_crit)
4204 					nbl_end_crit(vp);
4205 				VN_RELE(vp);
4206 				if (fp) {
4207 					rfs4_clear_dont_grant(fp);
4208 					rfs4_file_rele(fp);
4209 				}
4210 				goto out;
4211 			}
4212 		}
4213 	}
4214 
4215 	/* Get dir "before" change value */
4216 	bdva.va_mask = AT_CTIME|AT_SEQ;
4217 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
4218 	if (error) {
4219 		*cs->statusp = resp->status = puterrno4(error);
4220 		if (nm != name)
4221 			kmem_free(name, MAXPATHLEN + 1);
4222 		kmem_free(nm, len);
4223 		if (in_crit)
4224 			nbl_end_crit(vp);
4225 		VN_RELE(vp);
4226 		if (fp) {
4227 			rfs4_clear_dont_grant(fp);
4228 			rfs4_file_rele(fp);
4229 		}
4230 		goto out;
4231 	}
4232 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
4233 
4234 	/* Actually do the REMOVE operation */
4235 	if (vp->v_type == VDIR) {
4236 		/*
4237 		 * Can't remove a directory that has a mounted-on filesystem.
4238 		 */
4239 		if (vn_ismntpt(vp)) {
4240 			error = EACCES;
4241 		} else {
4242 			/*
4243 			 * System V defines rmdir to return EEXIST,
4244 			 * not ENOTEMPTY, if the directory is not
4245 			 * empty.  A System V NFS server needs to map
4246 			 * NFS4ERR_EXIST to NFS4ERR_NOTEMPTY to
4247 			 * transmit over the wire.
4248 			 */
4249 			if ((error = VOP_RMDIR(dvp, name, rootdir, cs->cr,
4250 			    NULL, 0)) == EEXIST)
4251 				error = ENOTEMPTY;
4252 		}
4253 	} else {
4254 		if ((error = VOP_REMOVE(dvp, name, cs->cr, NULL, 0)) == 0 &&
4255 		    fp != NULL) {
4256 			struct vattr va;
4257 			vnode_t *tvp;
4258 
4259 			rfs4_dbe_lock(fp->rf_dbe);
4260 			tvp = fp->rf_vp;
4261 			if (tvp)
4262 				VN_HOLD(tvp);
4263 			rfs4_dbe_unlock(fp->rf_dbe);
4264 
4265 			if (tvp) {
4266 				/*
4267 				 * This is va_seq safe because we are not
4268 				 * manipulating dvp.
4269 				 */
4270 				va.va_mask = AT_NLINK;
4271 				if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4272 				    va.va_nlink == 0) {
4273 					/* Remove state on file remove */
4274 					if (in_crit) {
4275 						nbl_end_crit(vp);
4276 						in_crit = 0;
4277 					}
4278 					rfs4_close_all_state(fp);
4279 				}
4280 				VN_RELE(tvp);
4281 			}
4282 		}
4283 	}
4284 
4285 	if (in_crit)
4286 		nbl_end_crit(vp);
4287 	VN_RELE(vp);
4288 
4289 	if (fp) {
4290 		rfs4_clear_dont_grant(fp);
4291 		rfs4_file_rele(fp);
4292 	}
4293 	if (nm != name)
4294 		kmem_free(name, MAXPATHLEN + 1);
4295 	kmem_free(nm, len);
4296 
4297 	if (error) {
4298 		*cs->statusp = resp->status = puterrno4(error);
4299 		goto out;
4300 	}
4301 
4302 	/*
4303 	 * Get the initial "after" sequence number, if it fails, set to zero
4304 	 */
4305 	idva.va_mask = AT_SEQ;
4306 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
4307 		idva.va_seq = 0;
4308 
4309 	/*
4310 	 * Force modified data and metadata out to stable storage.
4311 	 */
4312 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
4313 
4314 	/*
4315 	 * Get "after" change value, if it fails, simply return the
4316 	 * before value.
4317 	 */
4318 	adva.va_mask = AT_CTIME|AT_SEQ;
4319 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
4320 		adva.va_ctime = bdva.va_ctime;
4321 		adva.va_seq = 0;
4322 	}
4323 
4324 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
4325 
4326 	/*
4327 	 * The cinfo.atomic = TRUE only if we have
4328 	 * non-zero va_seq's, and it has incremented by exactly one
4329 	 * during the VOP_REMOVE/RMDIR and it didn't change during
4330 	 * the VOP_FSYNC.
4331 	 */
4332 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
4333 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
4334 		resp->cinfo.atomic = TRUE;
4335 	else
4336 		resp->cinfo.atomic = FALSE;
4337 
4338 	*cs->statusp = resp->status = NFS4_OK;
4339 
4340 out:
4341 	DTRACE_NFSV4_2(op__remove__done, struct compound_state *, cs,
4342 	    REMOVE4res *, resp);
4343 }
4344 
4345 /*
4346  * rename: args: SAVED_FH: from directory, CURRENT_FH: target directory,
4347  *		oldname and newname.
4348  *	res: status. If success - CURRENT_FH unchanged, return change_info
4349  *		for both from and target directories.
4350  */
4351 /* ARGSUSED */
4352 static void
4353 rfs4_op_rename(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4354     struct compound_state *cs)
4355 {
4356 	RENAME4args *args = &argop->nfs_argop4_u.oprename;
4357 	RENAME4res *resp = &resop->nfs_resop4_u.oprename;
4358 	int error;
4359 	vnode_t *odvp;
4360 	vnode_t *ndvp;
4361 	vnode_t *srcvp, *targvp;
4362 	struct vattr obdva, oidva, oadva;
4363 	struct vattr nbdva, nidva, nadva;
4364 	char *onm, *nnm;
4365 	uint_t olen, nlen;
4366 	rfs4_file_t *fp, *sfp;
4367 	int in_crit_src, in_crit_targ;
4368 	int fp_rele_grant_hold, sfp_rele_grant_hold;
4369 	bslabel_t *clabel;
4370 	struct sockaddr *ca;
4371 	char *converted_onm = NULL;
4372 	char *converted_nnm = NULL;
4373 	nfsstat4 status;
4374 
4375 	DTRACE_NFSV4_2(op__rename__start, struct compound_state *, cs,
4376 	    RENAME4args *, args);
4377 
4378 	fp = sfp = NULL;
4379 	srcvp = targvp = NULL;
4380 	in_crit_src = in_crit_targ = 0;
4381 	fp_rele_grant_hold = sfp_rele_grant_hold = 0;
4382 
4383 	/* CURRENT_FH: target directory */
4384 	ndvp = cs->vp;
4385 	if (ndvp == NULL) {
4386 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4387 		goto out;
4388 	}
4389 
4390 	/* SAVED_FH: from directory */
4391 	odvp = cs->saved_vp;
4392 	if (odvp == NULL) {
4393 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4394 		goto out;
4395 	}
4396 
4397 	if (cs->access == CS_ACCESS_DENIED) {
4398 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4399 		goto out;
4400 	}
4401 
4402 	/*
4403 	 * If there is an unshared filesystem mounted on this vnode,
4404 	 * do not allow to rename objects in this directory.
4405 	 */
4406 	if (vn_ismntpt(odvp)) {
4407 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4408 		goto out;
4409 	}
4410 
4411 	/*
4412 	 * If there is an unshared filesystem mounted on this vnode,
4413 	 * do not allow to rename to this directory.
4414 	 */
4415 	if (vn_ismntpt(ndvp)) {
4416 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4417 		goto out;
4418 	}
4419 
4420 	if (odvp->v_type != VDIR || ndvp->v_type != VDIR) {
4421 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4422 		goto out;
4423 	}
4424 
4425 	if (cs->saved_exi != cs->exi) {
4426 		*cs->statusp = resp->status = NFS4ERR_XDEV;
4427 		goto out;
4428 	}
4429 
4430 	status = utf8_dir_verify(&args->oldname);
4431 	if (status != NFS4_OK) {
4432 		*cs->statusp = resp->status = status;
4433 		goto out;
4434 	}
4435 
4436 	status = utf8_dir_verify(&args->newname);
4437 	if (status != NFS4_OK) {
4438 		*cs->statusp = resp->status = status;
4439 		goto out;
4440 	}
4441 
4442 	onm = utf8_to_fn(&args->oldname, &olen, NULL);
4443 	if (onm == NULL) {
4444 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4445 		goto out;
4446 	}
4447 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4448 	nlen = MAXPATHLEN + 1;
4449 	converted_onm = nfscmd_convname(ca, cs->exi, onm, NFSCMD_CONV_INBOUND,
4450 	    nlen);
4451 
4452 	if (converted_onm == NULL) {
4453 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4454 		kmem_free(onm, olen);
4455 		goto out;
4456 	}
4457 
4458 	nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4459 	if (nnm == NULL) {
4460 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4461 		if (onm != converted_onm)
4462 			kmem_free(converted_onm, MAXPATHLEN + 1);
4463 		kmem_free(onm, olen);
4464 		goto out;
4465 	}
4466 	converted_nnm = nfscmd_convname(ca, cs->exi, nnm, NFSCMD_CONV_INBOUND,
4467 	    MAXPATHLEN  + 1);
4468 
4469 	if (converted_nnm == NULL) {
4470 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4471 		kmem_free(nnm, nlen);
4472 		nnm = NULL;
4473 		if (onm != converted_onm)
4474 			kmem_free(converted_onm, MAXPATHLEN + 1);
4475 		kmem_free(onm, olen);
4476 		goto out;
4477 	}
4478 
4479 
4480 	if (olen > MAXNAMELEN || nlen > MAXNAMELEN) {
4481 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4482 		kmem_free(onm, olen);
4483 		kmem_free(nnm, nlen);
4484 		goto out;
4485 	}
4486 
4487 
4488 	if (rdonly4(req, cs)) {
4489 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4490 		if (onm != converted_onm)
4491 			kmem_free(converted_onm, MAXPATHLEN + 1);
4492 		kmem_free(onm, olen);
4493 		if (nnm != converted_nnm)
4494 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4495 		kmem_free(nnm, nlen);
4496 		goto out;
4497 	}
4498 
4499 	/* check label of the target dir */
4500 	if (is_system_labeled()) {
4501 		ASSERT(req->rq_label != NULL);
4502 		clabel = req->rq_label;
4503 		DTRACE_PROBE2(tx__rfs4__log__info__oprename__clabel, char *,
4504 		    "got client label from request(1)",
4505 		    struct svc_req *, req);
4506 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4507 			if (!do_rfs_label_check(clabel, ndvp,
4508 			    EQUALITY_CHECK, cs->exi)) {
4509 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4510 				goto err_out;
4511 			}
4512 		}
4513 	}
4514 
4515 	/*
4516 	 * Is the source a file and have a delegation?
4517 	 * We don't need to acquire va_seq before these lookups, if
4518 	 * it causes an update, cinfo.before will not match, which will
4519 	 * trigger a cache flush even if atomic is TRUE.
4520 	 */
4521 	if (sfp = rfs4_lookup_and_findfile(odvp, converted_onm, &srcvp,
4522 	    &error, cs->cr)) {
4523 		if (rfs4_check_delegated_byfp(FWRITE, sfp, TRUE, TRUE, TRUE,
4524 		    NULL)) {
4525 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4526 			goto err_out;
4527 		}
4528 	}
4529 
4530 	if (srcvp == NULL) {
4531 		*cs->statusp = resp->status = puterrno4(error);
4532 		if (onm != converted_onm)
4533 			kmem_free(converted_onm, MAXPATHLEN + 1);
4534 		kmem_free(onm, olen);
4535 		if (nnm != converted_nnm)
4536 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4537 		kmem_free(nnm, nlen);
4538 		goto out;
4539 	}
4540 
4541 	sfp_rele_grant_hold = 1;
4542 
4543 	/* Does the destination exist and a file and have a delegation? */
4544 	if (fp = rfs4_lookup_and_findfile(ndvp, converted_nnm, &targvp,
4545 	    NULL, cs->cr)) {
4546 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4547 		    NULL)) {
4548 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4549 			goto err_out;
4550 		}
4551 	}
4552 	fp_rele_grant_hold = 1;
4553 
4554 
4555 	/* Check for NBMAND lock on both source and target */
4556 	if (nbl_need_check(srcvp)) {
4557 		nbl_start_crit(srcvp, RW_READER);
4558 		in_crit_src = 1;
4559 		if (nbl_conflict(srcvp, NBL_RENAME, 0, 0, 0, NULL)) {
4560 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4561 			goto err_out;
4562 		}
4563 	}
4564 
4565 	if (targvp && nbl_need_check(targvp)) {
4566 		nbl_start_crit(targvp, RW_READER);
4567 		in_crit_targ = 1;
4568 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
4569 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4570 			goto err_out;
4571 		}
4572 	}
4573 
4574 	/* Get source "before" change value */
4575 	obdva.va_mask = AT_CTIME|AT_SEQ;
4576 	error = VOP_GETATTR(odvp, &obdva, 0, cs->cr, NULL);
4577 	if (!error) {
4578 		nbdva.va_mask = AT_CTIME|AT_SEQ;
4579 		error = VOP_GETATTR(ndvp, &nbdva, 0, cs->cr, NULL);
4580 	}
4581 	if (error) {
4582 		*cs->statusp = resp->status = puterrno4(error);
4583 		goto err_out;
4584 	}
4585 
4586 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.before, obdva.va_ctime)
4587 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.before, nbdva.va_ctime)
4588 
4589 	if ((error = VOP_RENAME(odvp, converted_onm, ndvp, converted_nnm,
4590 	    cs->cr, NULL, 0)) == 0 && fp != NULL) {
4591 		struct vattr va;
4592 		vnode_t *tvp;
4593 
4594 		rfs4_dbe_lock(fp->rf_dbe);
4595 		tvp = fp->rf_vp;
4596 		if (tvp)
4597 			VN_HOLD(tvp);
4598 		rfs4_dbe_unlock(fp->rf_dbe);
4599 
4600 		if (tvp) {
4601 			va.va_mask = AT_NLINK;
4602 			if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4603 			    va.va_nlink == 0) {
4604 				/* The file is gone and so should the state */
4605 				if (in_crit_targ) {
4606 					nbl_end_crit(targvp);
4607 					in_crit_targ = 0;
4608 				}
4609 				rfs4_close_all_state(fp);
4610 			}
4611 			VN_RELE(tvp);
4612 		}
4613 	}
4614 	if (error == 0)
4615 		vn_renamepath(ndvp, srcvp, nnm, nlen - 1);
4616 
4617 	if (in_crit_src)
4618 		nbl_end_crit(srcvp);
4619 	if (srcvp)
4620 		VN_RELE(srcvp);
4621 	if (in_crit_targ)
4622 		nbl_end_crit(targvp);
4623 	if (targvp)
4624 		VN_RELE(targvp);
4625 
4626 	if (sfp) {
4627 		rfs4_clear_dont_grant(sfp);
4628 		rfs4_file_rele(sfp);
4629 	}
4630 	if (fp) {
4631 		rfs4_clear_dont_grant(fp);
4632 		rfs4_file_rele(fp);
4633 	}
4634 
4635 	if (converted_onm != onm)
4636 		kmem_free(converted_onm, MAXPATHLEN + 1);
4637 	kmem_free(onm, olen);
4638 	if (converted_nnm != nnm)
4639 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4640 	kmem_free(nnm, nlen);
4641 
4642 	/*
4643 	 * Get the initial "after" sequence number, if it fails, set to zero
4644 	 */
4645 	oidva.va_mask = AT_SEQ;
4646 	if (VOP_GETATTR(odvp, &oidva, 0, cs->cr, NULL))
4647 		oidva.va_seq = 0;
4648 
4649 	nidva.va_mask = AT_SEQ;
4650 	if (VOP_GETATTR(ndvp, &nidva, 0, cs->cr, NULL))
4651 		nidva.va_seq = 0;
4652 
4653 	/*
4654 	 * Force modified data and metadata out to stable storage.
4655 	 */
4656 	(void) VOP_FSYNC(odvp, 0, cs->cr, NULL);
4657 	(void) VOP_FSYNC(ndvp, 0, cs->cr, NULL);
4658 
4659 	if (error) {
4660 		*cs->statusp = resp->status = puterrno4(error);
4661 		goto out;
4662 	}
4663 
4664 	/*
4665 	 * Get "after" change values, if it fails, simply return the
4666 	 * before value.
4667 	 */
4668 	oadva.va_mask = AT_CTIME|AT_SEQ;
4669 	if (VOP_GETATTR(odvp, &oadva, 0, cs->cr, NULL)) {
4670 		oadva.va_ctime = obdva.va_ctime;
4671 		oadva.va_seq = 0;
4672 	}
4673 
4674 	nadva.va_mask = AT_CTIME|AT_SEQ;
4675 	if (VOP_GETATTR(odvp, &nadva, 0, cs->cr, NULL)) {
4676 		nadva.va_ctime = nbdva.va_ctime;
4677 		nadva.va_seq = 0;
4678 	}
4679 
4680 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.after, oadva.va_ctime)
4681 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.after, nadva.va_ctime)
4682 
4683 	/*
4684 	 * The cinfo.atomic = TRUE only if we have
4685 	 * non-zero va_seq's, and it has incremented by exactly one
4686 	 * during the VOP_RENAME and it didn't change during the VOP_FSYNC.
4687 	 */
4688 	if (obdva.va_seq && oidva.va_seq && oadva.va_seq &&
4689 	    oidva.va_seq == (obdva.va_seq + 1) && oidva.va_seq == oadva.va_seq)
4690 		resp->source_cinfo.atomic = TRUE;
4691 	else
4692 		resp->source_cinfo.atomic = FALSE;
4693 
4694 	if (nbdva.va_seq && nidva.va_seq && nadva.va_seq &&
4695 	    nidva.va_seq == (nbdva.va_seq + 1) && nidva.va_seq == nadva.va_seq)
4696 		resp->target_cinfo.atomic = TRUE;
4697 	else
4698 		resp->target_cinfo.atomic = FALSE;
4699 
4700 #ifdef	VOLATILE_FH_TEST
4701 	{
4702 	extern void add_volrnm_fh(struct exportinfo *, vnode_t *);
4703 
4704 	/*
4705 	 * Add the renamed file handle to the volatile rename list
4706 	 */
4707 	if (cs->exi->exi_export.ex_flags & EX_VOLRNM) {
4708 		/* file handles may expire on rename */
4709 		vnode_t *vp;
4710 
4711 		nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4712 		/*
4713 		 * Already know that nnm will be a valid string
4714 		 */
4715 		error = VOP_LOOKUP(ndvp, nnm, &vp, NULL, 0, NULL, cs->cr,
4716 		    NULL, NULL, NULL);
4717 		kmem_free(nnm, nlen);
4718 		if (!error) {
4719 			add_volrnm_fh(cs->exi, vp);
4720 			VN_RELE(vp);
4721 		}
4722 	}
4723 	}
4724 #endif	/* VOLATILE_FH_TEST */
4725 
4726 	*cs->statusp = resp->status = NFS4_OK;
4727 out:
4728 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4729 	    RENAME4res *, resp);
4730 	return;
4731 
4732 err_out:
4733 	if (onm != converted_onm)
4734 		kmem_free(converted_onm, MAXPATHLEN + 1);
4735 	if (onm != NULL)
4736 		kmem_free(onm, olen);
4737 	if (nnm != converted_nnm)
4738 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4739 	if (nnm != NULL)
4740 		kmem_free(nnm, nlen);
4741 
4742 	if (in_crit_src) nbl_end_crit(srcvp);
4743 	if (in_crit_targ) nbl_end_crit(targvp);
4744 	if (targvp) VN_RELE(targvp);
4745 	if (srcvp) VN_RELE(srcvp);
4746 	if (sfp) {
4747 		if (sfp_rele_grant_hold) rfs4_clear_dont_grant(sfp);
4748 		rfs4_file_rele(sfp);
4749 	}
4750 	if (fp) {
4751 		if (fp_rele_grant_hold) rfs4_clear_dont_grant(fp);
4752 		rfs4_file_rele(fp);
4753 	}
4754 
4755 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4756 	    RENAME4res *, resp);
4757 }
4758 
4759 /* ARGSUSED */
4760 static void
4761 rfs4_op_renew(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4762     struct compound_state *cs)
4763 {
4764 	RENEW4args *args = &argop->nfs_argop4_u.oprenew;
4765 	RENEW4res *resp = &resop->nfs_resop4_u.oprenew;
4766 	rfs4_client_t *cp;
4767 
4768 	DTRACE_NFSV4_2(op__renew__start, struct compound_state *, cs,
4769 	    RENEW4args *, args);
4770 
4771 	if ((cp = rfs4_findclient_by_id(args->clientid, FALSE)) == NULL) {
4772 		*cs->statusp = resp->status =
4773 		    rfs4_check_clientid(&args->clientid, 0);
4774 		goto out;
4775 	}
4776 
4777 	if (rfs4_lease_expired(cp)) {
4778 		rfs4_client_rele(cp);
4779 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
4780 		goto out;
4781 	}
4782 
4783 	rfs4_update_lease(cp);
4784 
4785 	mutex_enter(cp->rc_cbinfo.cb_lock);
4786 	if (cp->rc_cbinfo.cb_notified_of_cb_path_down == FALSE) {
4787 		cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
4788 		*cs->statusp = resp->status = NFS4ERR_CB_PATH_DOWN;
4789 	} else {
4790 		*cs->statusp = resp->status = NFS4_OK;
4791 	}
4792 	mutex_exit(cp->rc_cbinfo.cb_lock);
4793 
4794 	rfs4_client_rele(cp);
4795 
4796 out:
4797 	DTRACE_NFSV4_2(op__renew__done, struct compound_state *, cs,
4798 	    RENEW4res *, resp);
4799 }
4800 
4801 /* ARGSUSED */
4802 static void
4803 rfs4_op_restorefh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
4804     struct compound_state *cs)
4805 {
4806 	RESTOREFH4res *resp = &resop->nfs_resop4_u.oprestorefh;
4807 
4808 	DTRACE_NFSV4_1(op__restorefh__start, struct compound_state *, cs);
4809 
4810 	/* No need to check cs->access - we are not accessing any object */
4811 	if ((cs->saved_vp == NULL) || (cs->saved_fh.nfs_fh4_val == NULL)) {
4812 		*cs->statusp = resp->status = NFS4ERR_RESTOREFH;
4813 		goto out;
4814 	}
4815 	if (cs->vp != NULL) {
4816 		VN_RELE(cs->vp);
4817 	}
4818 	cs->vp = cs->saved_vp;
4819 	cs->saved_vp = NULL;
4820 	cs->exi = cs->saved_exi;
4821 	nfs_fh4_copy(&cs->saved_fh, &cs->fh);
4822 	*cs->statusp = resp->status = NFS4_OK;
4823 	cs->deleg = FALSE;
4824 
4825 out:
4826 	DTRACE_NFSV4_2(op__restorefh__done, struct compound_state *, cs,
4827 	    RESTOREFH4res *, resp);
4828 }
4829 
4830 /* ARGSUSED */
4831 static void
4832 rfs4_op_savefh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4833     struct compound_state *cs)
4834 {
4835 	SAVEFH4res *resp = &resop->nfs_resop4_u.opsavefh;
4836 
4837 	DTRACE_NFSV4_1(op__savefh__start, struct compound_state *, cs);
4838 
4839 	/* No need to check cs->access - we are not accessing any object */
4840 	if (cs->vp == NULL) {
4841 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4842 		goto out;
4843 	}
4844 	if (cs->saved_vp != NULL) {
4845 		VN_RELE(cs->saved_vp);
4846 	}
4847 	cs->saved_vp = cs->vp;
4848 	VN_HOLD(cs->saved_vp);
4849 	cs->saved_exi = cs->exi;
4850 	/*
4851 	 * since SAVEFH is fairly rare, don't alloc space for its fh
4852 	 * unless necessary.
4853 	 */
4854 	if (cs->saved_fh.nfs_fh4_val == NULL) {
4855 		cs->saved_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP);
4856 	}
4857 	nfs_fh4_copy(&cs->fh, &cs->saved_fh);
4858 	*cs->statusp = resp->status = NFS4_OK;
4859 
4860 out:
4861 	DTRACE_NFSV4_2(op__savefh__done, struct compound_state *, cs,
4862 	    SAVEFH4res *, resp);
4863 }
4864 
4865 /*
4866  * rfs4_verify_attr is called when nfsv4 Setattr failed, but we wish to
4867  * return the bitmap of attrs that were set successfully. It is also
4868  * called by Verify/Nverify to test the vattr/vfsstat attrs. It should
4869  * always be called only after rfs4_do_set_attrs().
4870  *
4871  * Verify that the attributes are same as the expected ones. sargp->vap
4872  * and sargp->sbp contain the input attributes as translated from fattr4.
4873  *
4874  * This function verifies only the attrs that correspond to a vattr or
4875  * vfsstat struct. That is because of the extra step needed to get the
4876  * corresponding system structs. Other attributes have already been set or
4877  * verified by do_rfs4_set_attrs.
4878  *
4879  * Return 0 if all attrs match, -1 if some don't, error if error processing.
4880  */
4881 static int
4882 rfs4_verify_attr(struct nfs4_svgetit_arg *sargp,
4883     bitmap4 *resp, struct nfs4_ntov_table *ntovp)
4884 {
4885 	int error, ret_error = 0;
4886 	int i, k;
4887 	uint_t sva_mask = sargp->vap->va_mask;
4888 	uint_t vbit;
4889 	union nfs4_attr_u *na;
4890 	uint8_t *amap;
4891 	bool_t getsb = ntovp->vfsstat;
4892 
4893 	if (sva_mask != 0) {
4894 		/*
4895 		 * Okay to overwrite sargp->vap because we verify based
4896 		 * on the incoming values.
4897 		 */
4898 		ret_error = VOP_GETATTR(sargp->cs->vp, sargp->vap, 0,
4899 		    sargp->cs->cr, NULL);
4900 		if (ret_error) {
4901 			if (resp == NULL)
4902 				return (ret_error);
4903 			/*
4904 			 * Must return bitmap of successful attrs
4905 			 */
4906 			sva_mask = 0;	/* to prevent checking vap later */
4907 		} else {
4908 			/*
4909 			 * Some file systems clobber va_mask. it is probably
4910 			 * wrong of them to do so, nonethless we practice
4911 			 * defensive coding.
4912 			 * See bug id 4276830.
4913 			 */
4914 			sargp->vap->va_mask = sva_mask;
4915 		}
4916 	}
4917 
4918 	if (getsb) {
4919 		/*
4920 		 * Now get the superblock and loop on the bitmap, as there is
4921 		 * no simple way of translating from superblock to bitmap4.
4922 		 */
4923 		ret_error = VFS_STATVFS(sargp->cs->vp->v_vfsp, sargp->sbp);
4924 		if (ret_error) {
4925 			if (resp == NULL)
4926 				goto errout;
4927 			getsb = FALSE;
4928 		}
4929 	}
4930 
4931 	/*
4932 	 * Now loop and verify each attribute which getattr returned
4933 	 * whether it's the same as the input.
4934 	 */
4935 	if (resp == NULL && !getsb && (sva_mask == 0))
4936 		goto errout;
4937 
4938 	na = ntovp->na;
4939 	amap = ntovp->amap;
4940 	k = 0;
4941 	for (i = 0; i < ntovp->attrcnt; i++, na++, amap++) {
4942 		k = *amap;
4943 		ASSERT(nfs4_ntov_map[k].nval == k);
4944 		vbit = nfs4_ntov_map[k].vbit;
4945 
4946 		/*
4947 		 * If vattr attribute but VOP_GETATTR failed, or it's
4948 		 * superblock attribute but VFS_STATVFS failed, skip
4949 		 */
4950 		if (vbit) {
4951 			if ((vbit & sva_mask) == 0)
4952 				continue;
4953 		} else if (!(getsb && nfs4_ntov_map[k].vfsstat)) {
4954 			continue;
4955 		}
4956 		error = (*nfs4_ntov_map[k].sv_getit)(NFS4ATTR_VERIT, sargp, na);
4957 		if (resp != NULL) {
4958 			if (error)
4959 				ret_error = -1;	/* not all match */
4960 			else	/* update response bitmap */
4961 				*resp |= nfs4_ntov_map[k].fbit;
4962 			continue;
4963 		}
4964 		if (error) {
4965 			ret_error = -1;	/* not all match */
4966 			break;
4967 		}
4968 	}
4969 errout:
4970 	return (ret_error);
4971 }
4972 
4973 /*
4974  * Decode the attribute to be set/verified. If the attr requires a sys op
4975  * (VOP_GETATTR, VFS_VFSSTAT), and the request is to verify, then don't
4976  * call the sv_getit function for it, because the sys op hasn't yet been done.
4977  * Return 0 for success, error code if failed.
4978  *
4979  * Note: the decoded arg is not freed here but in nfs4_ntov_table_free.
4980  */
4981 static int
4982 decode_fattr4_attr(nfs4_attr_cmd_t cmd, struct nfs4_svgetit_arg *sargp,
4983     int k, XDR *xdrp, bitmap4 *resp_bval, union nfs4_attr_u *nap)
4984 {
4985 	int error = 0;
4986 	bool_t set_later;
4987 
4988 	sargp->vap->va_mask |= nfs4_ntov_map[k].vbit;
4989 
4990 	if ((*nfs4_ntov_map[k].xfunc)(xdrp, nap)) {
4991 		set_later = nfs4_ntov_map[k].vbit || nfs4_ntov_map[k].vfsstat;
4992 		/*
4993 		 * don't verify yet if a vattr or sb dependent attr,
4994 		 * because we don't have their sys values yet.
4995 		 * Will be done later.
4996 		 */
4997 		if (! (set_later && (cmd == NFS4ATTR_VERIT))) {
4998 			/*
4999 			 * ACLs are a special case, since setting the MODE
5000 			 * conflicts with setting the ACL.  We delay setting
5001 			 * the ACL until all other attributes have been set.
5002 			 * The ACL gets set in do_rfs4_op_setattr().
5003 			 */
5004 			if (nfs4_ntov_map[k].fbit != FATTR4_ACL_MASK) {
5005 				error = (*nfs4_ntov_map[k].sv_getit)(cmd,
5006 				    sargp, nap);
5007 				if (error) {
5008 					xdr_free(nfs4_ntov_map[k].xfunc,
5009 					    (caddr_t)nap);
5010 				}
5011 			}
5012 		}
5013 	} else {
5014 #ifdef  DEBUG
5015 		cmn_err(CE_NOTE, "decode_fattr4_attr: error "
5016 		    "decoding attribute %d\n", k);
5017 #endif
5018 		error = EINVAL;
5019 	}
5020 	if (!error && resp_bval && !set_later) {
5021 		*resp_bval |= nfs4_ntov_map[k].fbit;
5022 	}
5023 
5024 	return (error);
5025 }
5026 
5027 /*
5028  * Set vattr based on incoming fattr4 attrs - used by setattr.
5029  * Set response mask. Ignore any values that are not writable vattr attrs.
5030  */
5031 static nfsstat4
5032 do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5033     struct nfs4_svgetit_arg *sargp, struct nfs4_ntov_table *ntovp,
5034     nfs4_attr_cmd_t cmd)
5035 {
5036 	int error = 0;
5037 	int i;
5038 	char *attrs = fattrp->attrlist4;
5039 	uint32_t attrslen = fattrp->attrlist4_len;
5040 	XDR xdr;
5041 	nfsstat4 status = NFS4_OK;
5042 	vnode_t *vp = cs->vp;
5043 	union nfs4_attr_u *na;
5044 	uint8_t *amap;
5045 
5046 #ifndef lint
5047 	/*
5048 	 * Make sure that maximum attribute number can be expressed as an
5049 	 * 8 bit quantity.
5050 	 */
5051 	ASSERT(NFS4_MAXNUM_ATTRS <= (UINT8_MAX + 1));
5052 #endif
5053 
5054 	if (vp == NULL) {
5055 		if (resp)
5056 			*resp = 0;
5057 		return (NFS4ERR_NOFILEHANDLE);
5058 	}
5059 	if (cs->access == CS_ACCESS_DENIED) {
5060 		if (resp)
5061 			*resp = 0;
5062 		return (NFS4ERR_ACCESS);
5063 	}
5064 
5065 	sargp->op = cmd;
5066 	sargp->cs = cs;
5067 	sargp->flag = 0;	/* may be set later */
5068 	sargp->vap->va_mask = 0;
5069 	sargp->rdattr_error = NFS4_OK;
5070 	sargp->rdattr_error_req = FALSE;
5071 	/* sargp->sbp is set by the caller */
5072 
5073 	xdrmem_create(&xdr, attrs, attrslen, XDR_DECODE);
5074 
5075 	na = ntovp->na;
5076 	amap = ntovp->amap;
5077 
5078 	/*
5079 	 * The following loop iterates on the nfs4_ntov_map checking
5080 	 * if the fbit is set in the requested bitmap.
5081 	 * If set then we process the arguments using the
5082 	 * rfs4_fattr4 conversion functions to populate the setattr
5083 	 * vattr and va_mask. Any settable attrs that are not using vattr
5084 	 * will be set in this loop.
5085 	 */
5086 	for (i = 0; i < nfs4_ntov_map_size; i++) {
5087 		if (!(fattrp->attrmask & nfs4_ntov_map[i].fbit)) {
5088 			continue;
5089 		}
5090 		/*
5091 		 * If setattr, must be a writable attr.
5092 		 * If verify/nverify, must be a readable attr.
5093 		 */
5094 		if ((error = (*nfs4_ntov_map[i].sv_getit)(
5095 		    NFS4ATTR_SUPPORTED, sargp, NULL)) != 0) {
5096 			/*
5097 			 * Client tries to set/verify an
5098 			 * unsupported attribute, tries to set
5099 			 * a read only attr or verify a write
5100 			 * only one - error!
5101 			 */
5102 			break;
5103 		}
5104 		/*
5105 		 * Decode the attribute to set/verify
5106 		 */
5107 		error = decode_fattr4_attr(cmd, sargp, nfs4_ntov_map[i].nval,
5108 		    &xdr, resp ? resp : NULL, na);
5109 		if (error)
5110 			break;
5111 		*amap++ = (uint8_t)nfs4_ntov_map[i].nval;
5112 		na++;
5113 		(ntovp->attrcnt)++;
5114 		if (nfs4_ntov_map[i].vfsstat)
5115 			ntovp->vfsstat = TRUE;
5116 	}
5117 
5118 	if (error != 0)
5119 		status = (error == ENOTSUP ? NFS4ERR_ATTRNOTSUPP :
5120 		    puterrno4(error));
5121 	/* xdrmem_destroy(&xdrs); */	/* NO-OP */
5122 	return (status);
5123 }
5124 
5125 static nfsstat4
5126 do_rfs4_op_setattr(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5127     stateid4 *stateid)
5128 {
5129 	int error = 0;
5130 	struct nfs4_svgetit_arg sarg;
5131 	bool_t trunc;
5132 
5133 	nfsstat4 status = NFS4_OK;
5134 	cred_t *cr = cs->cr;
5135 	vnode_t *vp = cs->vp;
5136 	struct nfs4_ntov_table ntov;
5137 	struct statvfs64 sb;
5138 	struct vattr bva;
5139 	struct flock64 bf;
5140 	int in_crit = 0;
5141 	uint_t saved_mask = 0;
5142 	caller_context_t ct;
5143 
5144 	*resp = 0;
5145 	sarg.sbp = &sb;
5146 	sarg.is_referral = B_FALSE;
5147 	nfs4_ntov_table_init(&ntov);
5148 	status = do_rfs4_set_attrs(resp, fattrp, cs, &sarg, &ntov,
5149 	    NFS4ATTR_SETIT);
5150 	if (status != NFS4_OK) {
5151 		/*
5152 		 * failed set attrs
5153 		 */
5154 		goto done;
5155 	}
5156 	if ((sarg.vap->va_mask == 0) &&
5157 	    (! (fattrp->attrmask & FATTR4_ACL_MASK))) {
5158 		/*
5159 		 * no further work to be done
5160 		 */
5161 		goto done;
5162 	}
5163 
5164 	/*
5165 	 * If we got a request to set the ACL and the MODE, only
5166 	 * allow changing VSUID, VSGID, and VSVTX.  Attempting
5167 	 * to change any other bits, along with setting an ACL,
5168 	 * gives NFS4ERR_INVAL.
5169 	 */
5170 	if ((fattrp->attrmask & FATTR4_ACL_MASK) &&
5171 	    (fattrp->attrmask & FATTR4_MODE_MASK)) {
5172 		vattr_t va;
5173 
5174 		va.va_mask = AT_MODE;
5175 		error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
5176 		if (error) {
5177 			status = puterrno4(error);
5178 			goto done;
5179 		}
5180 		if ((sarg.vap->va_mode ^ va.va_mode) &
5181 		    ~(VSUID | VSGID | VSVTX)) {
5182 			status = NFS4ERR_INVAL;
5183 			goto done;
5184 		}
5185 	}
5186 
5187 	/* Check stateid only if size has been set */
5188 	if (sarg.vap->va_mask & AT_SIZE) {
5189 		trunc = (sarg.vap->va_size == 0);
5190 		status = rfs4_check_stateid(FWRITE, cs->vp, stateid,
5191 		    trunc, &cs->deleg, sarg.vap->va_mask & AT_SIZE, &ct);
5192 		if (status != NFS4_OK)
5193 			goto done;
5194 	} else {
5195 		ct.cc_sysid = 0;
5196 		ct.cc_pid = 0;
5197 		ct.cc_caller_id = nfs4_srv_caller_id;
5198 		ct.cc_flags = CC_DONTBLOCK;
5199 	}
5200 
5201 	/* XXX start of possible race with delegations */
5202 
5203 	/*
5204 	 * We need to specially handle size changes because it is
5205 	 * possible for the client to create a file with read-only
5206 	 * modes, but with the file opened for writing. If the client
5207 	 * then tries to set the file size, e.g. ftruncate(3C),
5208 	 * fcntl(F_FREESP), the normal access checking done in
5209 	 * VOP_SETATTR would prevent the client from doing it even though
5210 	 * it should be allowed to do so.  To get around this, we do the
5211 	 * access checking for ourselves and use VOP_SPACE which doesn't
5212 	 * do the access checking.
5213 	 * Also the client should not be allowed to change the file
5214 	 * size if there is a conflicting non-blocking mandatory lock in
5215 	 * the region of the change.
5216 	 */
5217 	if (vp->v_type == VREG && (sarg.vap->va_mask & AT_SIZE)) {
5218 		u_offset_t offset;
5219 		ssize_t length;
5220 
5221 		/*
5222 		 * ufs_setattr clears AT_SIZE from vap->va_mask, but
5223 		 * before returning, sarg.vap->va_mask is used to
5224 		 * generate the setattr reply bitmap.  We also clear
5225 		 * AT_SIZE below before calling VOP_SPACE.  For both
5226 		 * of these cases, the va_mask needs to be saved here
5227 		 * and restored after calling VOP_SETATTR.
5228 		 */
5229 		saved_mask = sarg.vap->va_mask;
5230 
5231 		/*
5232 		 * Check any possible conflict due to NBMAND locks.
5233 		 * Get into critical region before VOP_GETATTR, so the
5234 		 * size attribute is valid when checking conflicts.
5235 		 */
5236 		if (nbl_need_check(vp)) {
5237 			nbl_start_crit(vp, RW_READER);
5238 			in_crit = 1;
5239 		}
5240 
5241 		bva.va_mask = AT_UID|AT_SIZE;
5242 		if (error = VOP_GETATTR(vp, &bva, 0, cr, &ct)) {
5243 			status = puterrno4(error);
5244 			goto done;
5245 		}
5246 
5247 		if (in_crit) {
5248 			if (sarg.vap->va_size < bva.va_size) {
5249 				offset = sarg.vap->va_size;
5250 				length = bva.va_size - sarg.vap->va_size;
5251 			} else {
5252 				offset = bva.va_size;
5253 				length = sarg.vap->va_size - bva.va_size;
5254 			}
5255 			if (nbl_conflict(vp, NBL_WRITE, offset, length, 0,
5256 			    &ct)) {
5257 				status = NFS4ERR_LOCKED;
5258 				goto done;
5259 			}
5260 		}
5261 
5262 		if (crgetuid(cr) == bva.va_uid) {
5263 			sarg.vap->va_mask &= ~AT_SIZE;
5264 			bf.l_type = F_WRLCK;
5265 			bf.l_whence = 0;
5266 			bf.l_start = (off64_t)sarg.vap->va_size;
5267 			bf.l_len = 0;
5268 			bf.l_sysid = 0;
5269 			bf.l_pid = 0;
5270 			error = VOP_SPACE(vp, F_FREESP, &bf, FWRITE,
5271 			    (offset_t)sarg.vap->va_size, cr, &ct);
5272 		}
5273 	}
5274 
5275 	if (!error && sarg.vap->va_mask != 0)
5276 		error = VOP_SETATTR(vp, sarg.vap, sarg.flag, cr, &ct);
5277 
5278 	/* restore va_mask -- ufs_setattr clears AT_SIZE */
5279 	if (saved_mask & AT_SIZE)
5280 		sarg.vap->va_mask |= AT_SIZE;
5281 
5282 	/*
5283 	 * If an ACL was being set, it has been delayed until now,
5284 	 * in order to set the mode (via the VOP_SETATTR() above) first.
5285 	 */
5286 	if ((! error) && (fattrp->attrmask & FATTR4_ACL_MASK)) {
5287 		int i;
5288 
5289 		for (i = 0; i < NFS4_MAXNUM_ATTRS; i++)
5290 			if (ntov.amap[i] == FATTR4_ACL)
5291 				break;
5292 		if (i < NFS4_MAXNUM_ATTRS) {
5293 			error = (*nfs4_ntov_map[FATTR4_ACL].sv_getit)(
5294 			    NFS4ATTR_SETIT, &sarg, &ntov.na[i]);
5295 			if (error == 0) {
5296 				*resp |= FATTR4_ACL_MASK;
5297 			} else if (error == ENOTSUP) {
5298 				(void) rfs4_verify_attr(&sarg, resp, &ntov);
5299 				status = NFS4ERR_ATTRNOTSUPP;
5300 				goto done;
5301 			}
5302 		} else {
5303 			NFS4_DEBUG(rfs4_debug,
5304 			    (CE_NOTE, "do_rfs4_op_setattr: "
5305 			    "unable to find ACL in fattr4"));
5306 			error = EINVAL;
5307 		}
5308 	}
5309 
5310 	if (error) {
5311 		/* check if a monitor detected a delegation conflict */
5312 		if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
5313 			status = NFS4ERR_DELAY;
5314 		else
5315 			status = puterrno4(error);
5316 
5317 		/*
5318 		 * Set the response bitmap when setattr failed.
5319 		 * If VOP_SETATTR partially succeeded, test by doing a
5320 		 * VOP_GETATTR on the object and comparing the data
5321 		 * to the setattr arguments.
5322 		 */
5323 		(void) rfs4_verify_attr(&sarg, resp, &ntov);
5324 	} else {
5325 		/*
5326 		 * Force modified metadata out to stable storage.
5327 		 */
5328 		(void) VOP_FSYNC(vp, FNODSYNC, cr, &ct);
5329 		/*
5330 		 * Set response bitmap
5331 		 */
5332 		nfs4_vmask_to_nmask_set(sarg.vap->va_mask, resp);
5333 	}
5334 
5335 /* Return early and already have a NFSv4 error */
5336 done:
5337 	/*
5338 	 * Except for nfs4_vmask_to_nmask_set(), vattr --> fattr
5339 	 * conversion sets both readable and writeable NFS4 attrs
5340 	 * for AT_MTIME and AT_ATIME.  The line below masks out
5341 	 * unrequested attrs from the setattr result bitmap.  This
5342 	 * is placed after the done: label to catch the ATTRNOTSUP
5343 	 * case.
5344 	 */
5345 	*resp &= fattrp->attrmask;
5346 
5347 	if (in_crit)
5348 		nbl_end_crit(vp);
5349 
5350 	nfs4_ntov_table_free(&ntov, &sarg);
5351 
5352 	return (status);
5353 }
5354 
5355 /* ARGSUSED */
5356 static void
5357 rfs4_op_setattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5358     struct compound_state *cs)
5359 {
5360 	SETATTR4args *args = &argop->nfs_argop4_u.opsetattr;
5361 	SETATTR4res *resp = &resop->nfs_resop4_u.opsetattr;
5362 	bslabel_t *clabel;
5363 
5364 	DTRACE_NFSV4_2(op__setattr__start, struct compound_state *, cs,
5365 	    SETATTR4args *, args);
5366 
5367 	if (cs->vp == NULL) {
5368 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5369 		goto out;
5370 	}
5371 
5372 	/*
5373 	 * If there is an unshared filesystem mounted on this vnode,
5374 	 * do not allow to setattr on this vnode.
5375 	 */
5376 	if (vn_ismntpt(cs->vp)) {
5377 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5378 		goto out;
5379 	}
5380 
5381 	resp->attrsset = 0;
5382 
5383 	if (rdonly4(req, cs)) {
5384 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5385 		goto out;
5386 	}
5387 
5388 	/* check label before setting attributes */
5389 	if (is_system_labeled()) {
5390 		ASSERT(req->rq_label != NULL);
5391 		clabel = req->rq_label;
5392 		DTRACE_PROBE2(tx__rfs4__log__info__opsetattr__clabel, char *,
5393 		    "got client label from request(1)",
5394 		    struct svc_req *, req);
5395 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
5396 			if (!do_rfs_label_check(clabel, cs->vp,
5397 			    EQUALITY_CHECK, cs->exi)) {
5398 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
5399 				goto out;
5400 			}
5401 		}
5402 	}
5403 
5404 	*cs->statusp = resp->status =
5405 	    do_rfs4_op_setattr(&resp->attrsset, &args->obj_attributes, cs,
5406 	    &args->stateid);
5407 
5408 out:
5409 	DTRACE_NFSV4_2(op__setattr__done, struct compound_state *, cs,
5410 	    SETATTR4res *, resp);
5411 }
5412 
5413 /* ARGSUSED */
5414 static void
5415 rfs4_op_verify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5416     struct compound_state *cs)
5417 {
5418 	/*
5419 	 * verify and nverify are exactly the same, except that nverify
5420 	 * succeeds when some argument changed, and verify succeeds when
5421 	 * when none changed.
5422 	 */
5423 
5424 	VERIFY4args  *args = &argop->nfs_argop4_u.opverify;
5425 	VERIFY4res *resp = &resop->nfs_resop4_u.opverify;
5426 
5427 	int error;
5428 	struct nfs4_svgetit_arg sarg;
5429 	struct statvfs64 sb;
5430 	struct nfs4_ntov_table ntov;
5431 
5432 	DTRACE_NFSV4_2(op__verify__start, struct compound_state *, cs,
5433 	    VERIFY4args *, args);
5434 
5435 	if (cs->vp == NULL) {
5436 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5437 		goto out;
5438 	}
5439 
5440 	sarg.sbp = &sb;
5441 	sarg.is_referral = B_FALSE;
5442 	nfs4_ntov_table_init(&ntov);
5443 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5444 	    &sarg, &ntov, NFS4ATTR_VERIT);
5445 	if (resp->status != NFS4_OK) {
5446 		/*
5447 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5448 		 * so could return -1 for "no match".
5449 		 */
5450 		if (resp->status == -1)
5451 			resp->status = NFS4ERR_NOT_SAME;
5452 		goto done;
5453 	}
5454 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5455 	switch (error) {
5456 	case 0:
5457 		resp->status = NFS4_OK;
5458 		break;
5459 	case -1:
5460 		resp->status = NFS4ERR_NOT_SAME;
5461 		break;
5462 	default:
5463 		resp->status = puterrno4(error);
5464 		break;
5465 	}
5466 done:
5467 	*cs->statusp = resp->status;
5468 	nfs4_ntov_table_free(&ntov, &sarg);
5469 out:
5470 	DTRACE_NFSV4_2(op__verify__done, struct compound_state *, cs,
5471 	    VERIFY4res *, resp);
5472 }
5473 
5474 /* ARGSUSED */
5475 static void
5476 rfs4_op_nverify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5477     struct compound_state *cs)
5478 {
5479 	/*
5480 	 * verify and nverify are exactly the same, except that nverify
5481 	 * succeeds when some argument changed, and verify succeeds when
5482 	 * when none changed.
5483 	 */
5484 
5485 	NVERIFY4args  *args = &argop->nfs_argop4_u.opnverify;
5486 	NVERIFY4res *resp = &resop->nfs_resop4_u.opnverify;
5487 
5488 	int error;
5489 	struct nfs4_svgetit_arg sarg;
5490 	struct statvfs64 sb;
5491 	struct nfs4_ntov_table ntov;
5492 
5493 	DTRACE_NFSV4_2(op__nverify__start, struct compound_state *, cs,
5494 	    NVERIFY4args *, args);
5495 
5496 	if (cs->vp == NULL) {
5497 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5498 		DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5499 		    NVERIFY4res *, resp);
5500 		return;
5501 	}
5502 	sarg.sbp = &sb;
5503 	sarg.is_referral = B_FALSE;
5504 	nfs4_ntov_table_init(&ntov);
5505 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5506 	    &sarg, &ntov, NFS4ATTR_VERIT);
5507 	if (resp->status != NFS4_OK) {
5508 		/*
5509 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5510 		 * so could return -1 for "no match".
5511 		 */
5512 		if (resp->status == -1)
5513 			resp->status = NFS4_OK;
5514 		goto done;
5515 	}
5516 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5517 	switch (error) {
5518 	case 0:
5519 		resp->status = NFS4ERR_SAME;
5520 		break;
5521 	case -1:
5522 		resp->status = NFS4_OK;
5523 		break;
5524 	default:
5525 		resp->status = puterrno4(error);
5526 		break;
5527 	}
5528 done:
5529 	*cs->statusp = resp->status;
5530 	nfs4_ntov_table_free(&ntov, &sarg);
5531 
5532 	DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5533 	    NVERIFY4res *, resp);
5534 }
5535 
5536 /*
5537  * XXX - This should live in an NFS header file.
5538  */
5539 #define	MAX_IOVECS	12
5540 
5541 /* ARGSUSED */
5542 static void
5543 rfs4_op_write(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5544     struct compound_state *cs)
5545 {
5546 	WRITE4args *args = &argop->nfs_argop4_u.opwrite;
5547 	WRITE4res *resp = &resop->nfs_resop4_u.opwrite;
5548 	int error;
5549 	vnode_t *vp;
5550 	struct vattr bva;
5551 	u_offset_t rlimit;
5552 	struct uio uio;
5553 	struct iovec iov[MAX_IOVECS];
5554 	struct iovec *iovp;
5555 	int iovcnt;
5556 	int ioflag;
5557 	cred_t *savecred, *cr;
5558 	bool_t *deleg = &cs->deleg;
5559 	nfsstat4 stat;
5560 	int in_crit = 0;
5561 	caller_context_t ct;
5562 
5563 	DTRACE_NFSV4_2(op__write__start, struct compound_state *, cs,
5564 	    WRITE4args *, args);
5565 
5566 	vp = cs->vp;
5567 	if (vp == NULL) {
5568 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5569 		goto out;
5570 	}
5571 	if (cs->access == CS_ACCESS_DENIED) {
5572 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5573 		goto out;
5574 	}
5575 
5576 	cr = cs->cr;
5577 
5578 	if ((stat = rfs4_check_stateid(FWRITE, vp, &args->stateid, FALSE,
5579 	    deleg, TRUE, &ct)) != NFS4_OK) {
5580 		*cs->statusp = resp->status = stat;
5581 		goto out;
5582 	}
5583 
5584 	/*
5585 	 * We have to enter the critical region before calling VOP_RWLOCK
5586 	 * to avoid a deadlock with ufs.
5587 	 */
5588 	if (nbl_need_check(vp)) {
5589 		nbl_start_crit(vp, RW_READER);
5590 		in_crit = 1;
5591 		if (nbl_conflict(vp, NBL_WRITE,
5592 		    args->offset, args->data_len, 0, &ct)) {
5593 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
5594 			goto out;
5595 		}
5596 	}
5597 
5598 	bva.va_mask = AT_MODE | AT_UID;
5599 	error = VOP_GETATTR(vp, &bva, 0, cr, &ct);
5600 
5601 	/*
5602 	 * If we can't get the attributes, then we can't do the
5603 	 * right access checking.  So, we'll fail the request.
5604 	 */
5605 	if (error) {
5606 		*cs->statusp = resp->status = puterrno4(error);
5607 		goto out;
5608 	}
5609 
5610 	if (rdonly4(req, cs)) {
5611 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5612 		goto out;
5613 	}
5614 
5615 	if (vp->v_type != VREG) {
5616 		*cs->statusp = resp->status =
5617 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
5618 		goto out;
5619 	}
5620 
5621 	if (crgetuid(cr) != bva.va_uid &&
5622 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, &ct))) {
5623 		*cs->statusp = resp->status = puterrno4(error);
5624 		goto out;
5625 	}
5626 
5627 	if (MANDLOCK(vp, bva.va_mode)) {
5628 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5629 		goto out;
5630 	}
5631 
5632 	if (args->data_len == 0) {
5633 		*cs->statusp = resp->status = NFS4_OK;
5634 		resp->count = 0;
5635 		resp->committed = args->stable;
5636 		resp->writeverf = Write4verf;
5637 		goto out;
5638 	}
5639 
5640 	if (args->mblk != NULL) {
5641 		mblk_t *m;
5642 		uint_t bytes, round_len;
5643 
5644 		iovcnt = 0;
5645 		bytes = 0;
5646 		round_len = roundup(args->data_len, BYTES_PER_XDR_UNIT);
5647 		for (m = args->mblk;
5648 		    m != NULL && bytes < round_len;
5649 		    m = m->b_cont) {
5650 			iovcnt++;
5651 			bytes += MBLKL(m);
5652 		}
5653 #ifdef DEBUG
5654 		/* should have ended on an mblk boundary */
5655 		if (bytes != round_len) {
5656 			printf("bytes=0x%x, round_len=0x%x, req len=0x%x\n",
5657 			    bytes, round_len, args->data_len);
5658 			printf("args=%p, args->mblk=%p, m=%p", (void *)args,
5659 			    (void *)args->mblk, (void *)m);
5660 			ASSERT(bytes == round_len);
5661 		}
5662 #endif
5663 		if (iovcnt <= MAX_IOVECS) {
5664 			iovp = iov;
5665 		} else {
5666 			iovp = kmem_alloc(sizeof (*iovp) * iovcnt, KM_SLEEP);
5667 		}
5668 		mblk_to_iov(args->mblk, iovcnt, iovp);
5669 	} else if (args->rlist != NULL) {
5670 		iovcnt = 1;
5671 		iovp = iov;
5672 		iovp->iov_base = (char *)((args->rlist)->u.c_daddr3);
5673 		iovp->iov_len = args->data_len;
5674 	} else {
5675 		iovcnt = 1;
5676 		iovp = iov;
5677 		iovp->iov_base = args->data_val;
5678 		iovp->iov_len = args->data_len;
5679 	}
5680 
5681 	uio.uio_iov = iovp;
5682 	uio.uio_iovcnt = iovcnt;
5683 
5684 	uio.uio_segflg = UIO_SYSSPACE;
5685 	uio.uio_extflg = UIO_COPY_DEFAULT;
5686 	uio.uio_loffset = args->offset;
5687 	uio.uio_resid = args->data_len;
5688 	uio.uio_llimit = curproc->p_fsz_ctl;
5689 	rlimit = uio.uio_llimit - args->offset;
5690 	if (rlimit < (u_offset_t)uio.uio_resid)
5691 		uio.uio_resid = (int)rlimit;
5692 
5693 	if (args->stable == UNSTABLE4)
5694 		ioflag = 0;
5695 	else if (args->stable == FILE_SYNC4)
5696 		ioflag = FSYNC;
5697 	else if (args->stable == DATA_SYNC4)
5698 		ioflag = FDSYNC;
5699 	else {
5700 		if (iovp != iov)
5701 			kmem_free(iovp, sizeof (*iovp) * iovcnt);
5702 		*cs->statusp = resp->status = NFS4ERR_INVAL;
5703 		goto out;
5704 	}
5705 
5706 	/*
5707 	 * We're changing creds because VM may fault and we need
5708 	 * the cred of the current thread to be used if quota
5709 	 * checking is enabled.
5710 	 */
5711 	savecred = curthread->t_cred;
5712 	curthread->t_cred = cr;
5713 	error = do_io(FWRITE, vp, &uio, ioflag, cr, &ct);
5714 	curthread->t_cred = savecred;
5715 
5716 	if (iovp != iov)
5717 		kmem_free(iovp, sizeof (*iovp) * iovcnt);
5718 
5719 	if (error) {
5720 		*cs->statusp = resp->status = puterrno4(error);
5721 		goto out;
5722 	}
5723 
5724 	*cs->statusp = resp->status = NFS4_OK;
5725 	resp->count = args->data_len - uio.uio_resid;
5726 
5727 	if (ioflag == 0)
5728 		resp->committed = UNSTABLE4;
5729 	else
5730 		resp->committed = FILE_SYNC4;
5731 
5732 	resp->writeverf = Write4verf;
5733 
5734 out:
5735 	if (in_crit)
5736 		nbl_end_crit(vp);
5737 
5738 	DTRACE_NFSV4_2(op__write__done, struct compound_state *, cs,
5739 	    WRITE4res *, resp);
5740 }
5741 
5742 
5743 /* XXX put in a header file */
5744 extern int	sec_svc_getcred(struct svc_req *, cred_t *,  caddr_t *, int *);
5745 
5746 void
5747 rfs4_compound(COMPOUND4args *args, COMPOUND4res *resp, struct exportinfo *exi,
5748     struct svc_req *req, cred_t *cr, int *rv)
5749 {
5750 	uint_t i;
5751 	struct compound_state cs;
5752 
5753 	if (rv != NULL)
5754 		*rv = 0;
5755 	rfs4_init_compound_state(&cs);
5756 	/*
5757 	 * Form a reply tag by copying over the reqeuest tag.
5758 	 */
5759 	resp->tag.utf8string_val =
5760 	    kmem_alloc(args->tag.utf8string_len, KM_SLEEP);
5761 	resp->tag.utf8string_len = args->tag.utf8string_len;
5762 	bcopy(args->tag.utf8string_val, resp->tag.utf8string_val,
5763 	    resp->tag.utf8string_len);
5764 
5765 	cs.statusp = &resp->status;
5766 	cs.req = req;
5767 	resp->array = NULL;
5768 	resp->array_len = 0;
5769 
5770 	/*
5771 	 * XXX for now, minorversion should be zero
5772 	 */
5773 	if (args->minorversion != NFS4_MINORVERSION) {
5774 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5775 		    &cs, COMPOUND4args *, args);
5776 		resp->status = NFS4ERR_MINOR_VERS_MISMATCH;
5777 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5778 		    &cs, COMPOUND4res *, resp);
5779 		return;
5780 	}
5781 
5782 	if (args->array_len == 0) {
5783 		resp->status = NFS4_OK;
5784 		return;
5785 	}
5786 
5787 	ASSERT(exi == NULL);
5788 	ASSERT(cr == NULL);
5789 
5790 	cr = crget();
5791 	ASSERT(cr != NULL);
5792 
5793 	if (sec_svc_getcred(req, cr, &cs.principal, &cs.nfsflavor) == 0) {
5794 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5795 		    &cs, COMPOUND4args *, args);
5796 		crfree(cr);
5797 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5798 		    &cs, COMPOUND4res *, resp);
5799 		svcerr_badcred(req->rq_xprt);
5800 		if (rv != NULL)
5801 			*rv = 1;
5802 		return;
5803 	}
5804 	resp->array_len = args->array_len;
5805 	resp->array = kmem_zalloc(args->array_len * sizeof (nfs_resop4),
5806 	    KM_SLEEP);
5807 
5808 	cs.basecr = cr;
5809 
5810 	DTRACE_NFSV4_2(compound__start, struct compound_state *, &cs,
5811 	    COMPOUND4args *, args);
5812 
5813 	/*
5814 	 * For now, NFS4 compound processing must be protected by
5815 	 * exported_lock because it can access more than one exportinfo
5816 	 * per compound and share/unshare can now change multiple
5817 	 * exinfo structs.  The NFS2/3 code only refs 1 exportinfo
5818 	 * per proc (excluding public exinfo), and exi_count design
5819 	 * is sufficient to protect concurrent execution of NFS2/3
5820 	 * ops along with unexport.  This lock will be removed as
5821 	 * part of the NFSv4 phase 2 namespace redesign work.
5822 	 */
5823 	rw_enter(&exported_lock, RW_READER);
5824 
5825 	/*
5826 	 * If this is the first compound we've seen, we need to start all
5827 	 * new instances' grace periods.
5828 	 */
5829 	if (rfs4_seen_first_compound == 0) {
5830 		rfs4_grace_start_new();
5831 		/*
5832 		 * This must be set after rfs4_grace_start_new(), otherwise
5833 		 * another thread could proceed past here before the former
5834 		 * is finished.
5835 		 */
5836 		rfs4_seen_first_compound = 1;
5837 	}
5838 
5839 	for (i = 0; i < args->array_len && cs.cont; i++) {
5840 		nfs_argop4 *argop;
5841 		nfs_resop4 *resop;
5842 		uint_t op;
5843 
5844 		argop = &args->array[i];
5845 		resop = &resp->array[i];
5846 		resop->resop = argop->argop;
5847 		op = (uint_t)resop->resop;
5848 
5849 		if (op < rfsv4disp_cnt) {
5850 			/*
5851 			 * Count the individual ops here; NULL and COMPOUND
5852 			 * are counted in common_dispatch()
5853 			 */
5854 			rfsproccnt_v4_ptr[op].value.ui64++;
5855 
5856 			NFS4_DEBUG(rfs4_debug > 1,
5857 			    (CE_NOTE, "Executing %s", rfs4_op_string[op]));
5858 			(*rfsv4disptab[op].dis_proc)(argop, resop, req, &cs);
5859 			NFS4_DEBUG(rfs4_debug > 1, (CE_NOTE, "%s returned %d",
5860 			    rfs4_op_string[op], *cs.statusp));
5861 			if (*cs.statusp != NFS4_OK)
5862 				cs.cont = FALSE;
5863 		} else {
5864 			/*
5865 			 * This is effectively dead code since XDR code
5866 			 * will have already returned BADXDR if op doesn't
5867 			 * decode to legal value.  This only done for a
5868 			 * day when XDR code doesn't verify v4 opcodes.
5869 			 */
5870 			op = OP_ILLEGAL;
5871 			rfsproccnt_v4_ptr[OP_ILLEGAL_IDX].value.ui64++;
5872 
5873 			rfs4_op_illegal(argop, resop, req, &cs);
5874 			cs.cont = FALSE;
5875 		}
5876 
5877 		/*
5878 		 * If not at last op, and if we are to stop, then
5879 		 * compact the results array.
5880 		 */
5881 		if ((i + 1) < args->array_len && !cs.cont) {
5882 			nfs_resop4 *new_res = kmem_alloc(
5883 			    (i+1) * sizeof (nfs_resop4), KM_SLEEP);
5884 			bcopy(resp->array,
5885 			    new_res, (i+1) * sizeof (nfs_resop4));
5886 			kmem_free(resp->array,
5887 			    args->array_len * sizeof (nfs_resop4));
5888 
5889 			resp->array_len =  i + 1;
5890 			resp->array = new_res;
5891 		}
5892 	}
5893 
5894 	rw_exit(&exported_lock);
5895 
5896 	DTRACE_NFSV4_2(compound__done, struct compound_state *, &cs,
5897 	    COMPOUND4res *, resp);
5898 
5899 	if (cs.vp)
5900 		VN_RELE(cs.vp);
5901 	if (cs.saved_vp)
5902 		VN_RELE(cs.saved_vp);
5903 	if (cs.saved_fh.nfs_fh4_val)
5904 		kmem_free(cs.saved_fh.nfs_fh4_val, NFS4_FHSIZE);
5905 
5906 	if (cs.basecr)
5907 		crfree(cs.basecr);
5908 	if (cs.cr)
5909 		crfree(cs.cr);
5910 	/*
5911 	 * done with this compound request, free the label
5912 	 */
5913 
5914 	if (req->rq_label != NULL) {
5915 		kmem_free(req->rq_label, sizeof (bslabel_t));
5916 		req->rq_label = NULL;
5917 	}
5918 }
5919 
5920 /*
5921  * XXX because of what appears to be duplicate calls to rfs4_compound_free
5922  * XXX zero out the tag and array values. Need to investigate why the
5923  * XXX calls occur, but at least prevent the panic for now.
5924  */
5925 void
5926 rfs4_compound_free(COMPOUND4res *resp)
5927 {
5928 	uint_t i;
5929 
5930 	if (resp->tag.utf8string_val) {
5931 		UTF8STRING_FREE(resp->tag)
5932 	}
5933 
5934 	for (i = 0; i < resp->array_len; i++) {
5935 		nfs_resop4 *resop;
5936 		uint_t op;
5937 
5938 		resop = &resp->array[i];
5939 		op = (uint_t)resop->resop;
5940 		if (op < rfsv4disp_cnt) {
5941 			(*rfsv4disptab[op].dis_resfree)(resop);
5942 		}
5943 	}
5944 	if (resp->array != NULL) {
5945 		kmem_free(resp->array, resp->array_len * sizeof (nfs_resop4));
5946 	}
5947 }
5948 
5949 /*
5950  * Process the value of the compound request rpc flags, as a bit-AND
5951  * of the individual per-op flags (idempotent, allowork, publicfh_ok)
5952  */
5953 void
5954 rfs4_compound_flagproc(COMPOUND4args *args, int *flagp)
5955 {
5956 	int i;
5957 	int flag = RPC_ALL;
5958 
5959 	for (i = 0; flag && i < args->array_len; i++) {
5960 		uint_t op;
5961 
5962 		op = (uint_t)args->array[i].argop;
5963 
5964 		if (op < rfsv4disp_cnt)
5965 			flag &= rfsv4disptab[op].dis_flags;
5966 		else
5967 			flag = 0;
5968 	}
5969 	*flagp = flag;
5970 }
5971 
5972 nfsstat4
5973 rfs4_client_sysid(rfs4_client_t *cp, sysid_t *sp)
5974 {
5975 	nfsstat4 e;
5976 
5977 	rfs4_dbe_lock(cp->rc_dbe);
5978 
5979 	if (cp->rc_sysidt != LM_NOSYSID) {
5980 		*sp = cp->rc_sysidt;
5981 		e = NFS4_OK;
5982 
5983 	} else if ((cp->rc_sysidt = lm_alloc_sysidt()) != LM_NOSYSID) {
5984 		*sp = cp->rc_sysidt;
5985 		e = NFS4_OK;
5986 
5987 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
5988 		    "rfs4_client_sysid: allocated 0x%x\n", *sp));
5989 	} else
5990 		e = NFS4ERR_DELAY;
5991 
5992 	rfs4_dbe_unlock(cp->rc_dbe);
5993 	return (e);
5994 }
5995 
5996 #if defined(DEBUG) && ! defined(lint)
5997 static void lock_print(char *str, int operation, struct flock64 *flk)
5998 {
5999 	char *op, *type;
6000 
6001 	switch (operation) {
6002 	case F_GETLK: op = "F_GETLK";
6003 		break;
6004 	case F_SETLK: op = "F_SETLK";
6005 		break;
6006 	case F_SETLK_NBMAND: op = "F_SETLK_NBMAND";
6007 		break;
6008 	default: op = "F_UNKNOWN";
6009 		break;
6010 	}
6011 	switch (flk->l_type) {
6012 	case F_UNLCK: type = "F_UNLCK";
6013 		break;
6014 	case F_RDLCK: type = "F_RDLCK";
6015 		break;
6016 	case F_WRLCK: type = "F_WRLCK";
6017 		break;
6018 	default: type = "F_UNKNOWN";
6019 		break;
6020 	}
6021 
6022 	ASSERT(flk->l_whence == 0);
6023 	cmn_err(CE_NOTE, "%s:  %s, type = %s, off = %llx len = %llx pid = %d",
6024 	    str, op, type, (longlong_t)flk->l_start,
6025 	    flk->l_len ? (longlong_t)flk->l_len : ~0LL, flk->l_pid);
6026 }
6027 
6028 #define	LOCK_PRINT(d, s, t, f) if (d) lock_print(s, t, f)
6029 #else
6030 #define	LOCK_PRINT(d, s, t, f)
6031 #endif
6032 
6033 /*ARGSUSED*/
6034 static bool_t
6035 creds_ok(cred_set_t cr_set, struct svc_req *req, struct compound_state *cs)
6036 {
6037 	return (TRUE);
6038 }
6039 
6040 /*
6041  * Look up the pathname using the vp in cs as the directory vnode.
6042  * cs->vp will be the vnode for the file on success
6043  */
6044 
6045 static nfsstat4
6046 rfs4_lookup(component4 *component, struct svc_req *req,
6047     struct compound_state *cs)
6048 {
6049 	char *nm;
6050 	uint32_t len;
6051 	nfsstat4 status;
6052 	struct sockaddr *ca;
6053 	char *name;
6054 
6055 	if (cs->vp == NULL) {
6056 		return (NFS4ERR_NOFILEHANDLE);
6057 	}
6058 	if (cs->vp->v_type != VDIR) {
6059 		return (NFS4ERR_NOTDIR);
6060 	}
6061 
6062 	status = utf8_dir_verify(component);
6063 	if (status != NFS4_OK)
6064 		return (status);
6065 
6066 	nm = utf8_to_fn(component, &len, NULL);
6067 	if (nm == NULL) {
6068 		return (NFS4ERR_INVAL);
6069 	}
6070 
6071 	if (len > MAXNAMELEN) {
6072 		kmem_free(nm, len);
6073 		return (NFS4ERR_NAMETOOLONG);
6074 	}
6075 
6076 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6077 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6078 	    MAXPATHLEN + 1);
6079 
6080 	if (name == NULL) {
6081 		kmem_free(nm, len);
6082 		return (NFS4ERR_INVAL);
6083 	}
6084 
6085 	status = do_rfs4_op_lookup(name, req, cs);
6086 
6087 	if (name != nm)
6088 		kmem_free(name, MAXPATHLEN + 1);
6089 
6090 	kmem_free(nm, len);
6091 
6092 	return (status);
6093 }
6094 
6095 static nfsstat4
6096 rfs4_lookupfile(component4 *component, struct svc_req *req,
6097     struct compound_state *cs, uint32_t access, change_info4 *cinfo)
6098 {
6099 	nfsstat4 status;
6100 	vnode_t *dvp = cs->vp;
6101 	vattr_t bva, ava, fva;
6102 	int error;
6103 
6104 	/* Get "before" change value */
6105 	bva.va_mask = AT_CTIME|AT_SEQ;
6106 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6107 	if (error)
6108 		return (puterrno4(error));
6109 
6110 	/* rfs4_lookup may VN_RELE directory */
6111 	VN_HOLD(dvp);
6112 
6113 	status = rfs4_lookup(component, req, cs);
6114 	if (status != NFS4_OK) {
6115 		VN_RELE(dvp);
6116 		return (status);
6117 	}
6118 
6119 	/*
6120 	 * Get "after" change value, if it fails, simply return the
6121 	 * before value.
6122 	 */
6123 	ava.va_mask = AT_CTIME|AT_SEQ;
6124 	if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6125 		ava.va_ctime = bva.va_ctime;
6126 		ava.va_seq = 0;
6127 	}
6128 	VN_RELE(dvp);
6129 
6130 	/*
6131 	 * Validate the file is a file
6132 	 */
6133 	fva.va_mask = AT_TYPE|AT_MODE;
6134 	error = VOP_GETATTR(cs->vp, &fva, 0, cs->cr, NULL);
6135 	if (error)
6136 		return (puterrno4(error));
6137 
6138 	if (fva.va_type != VREG) {
6139 		if (fva.va_type == VDIR)
6140 			return (NFS4ERR_ISDIR);
6141 		if (fva.va_type == VLNK)
6142 			return (NFS4ERR_SYMLINK);
6143 		return (NFS4ERR_INVAL);
6144 	}
6145 
6146 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime);
6147 	NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6148 
6149 	/*
6150 	 * It is undefined if VOP_LOOKUP will change va_seq, so
6151 	 * cinfo.atomic = TRUE only if we have
6152 	 * non-zero va_seq's, and they have not changed.
6153 	 */
6154 	if (bva.va_seq && ava.va_seq && ava.va_seq == bva.va_seq)
6155 		cinfo->atomic = TRUE;
6156 	else
6157 		cinfo->atomic = FALSE;
6158 
6159 	/* Check for mandatory locking */
6160 	cs->mandlock = MANDLOCK(cs->vp, fva.va_mode);
6161 	return (check_open_access(access, cs, req));
6162 }
6163 
6164 static nfsstat4
6165 create_vnode(vnode_t *dvp, char *nm,  vattr_t *vap, createmode4 mode,
6166     cred_t *cr, vnode_t **vpp, bool_t *created)
6167 {
6168 	int error;
6169 	nfsstat4 status = NFS4_OK;
6170 	vattr_t va;
6171 
6172 tryagain:
6173 
6174 	/*
6175 	 * The file open mode used is VWRITE.  If the client needs
6176 	 * some other semantic, then it should do the access checking
6177 	 * itself.  It would have been nice to have the file open mode
6178 	 * passed as part of the arguments.
6179 	 */
6180 
6181 	*created = TRUE;
6182 	error = VOP_CREATE(dvp, nm, vap, EXCL, VWRITE, vpp, cr, 0, NULL, NULL);
6183 
6184 	if (error) {
6185 		*created = FALSE;
6186 
6187 		/*
6188 		 * If we got something other than file already exists
6189 		 * then just return this error.  Otherwise, we got
6190 		 * EEXIST.  If we were doing a GUARDED create, then
6191 		 * just return this error.  Otherwise, we need to
6192 		 * make sure that this wasn't a duplicate of an
6193 		 * exclusive create request.
6194 		 *
6195 		 * The assumption is made that a non-exclusive create
6196 		 * request will never return EEXIST.
6197 		 */
6198 
6199 		if (error != EEXIST || mode == GUARDED4) {
6200 			status = puterrno4(error);
6201 			return (status);
6202 		}
6203 		error = VOP_LOOKUP(dvp, nm, vpp, NULL, 0, NULL, cr,
6204 		    NULL, NULL, NULL);
6205 
6206 		if (error) {
6207 			/*
6208 			 * We couldn't find the file that we thought that
6209 			 * we just created.  So, we'll just try creating
6210 			 * it again.
6211 			 */
6212 			if (error == ENOENT)
6213 				goto tryagain;
6214 
6215 			status = puterrno4(error);
6216 			return (status);
6217 		}
6218 
6219 		if (mode == UNCHECKED4) {
6220 			/* existing object must be regular file */
6221 			if ((*vpp)->v_type != VREG) {
6222 				if ((*vpp)->v_type == VDIR)
6223 					status = NFS4ERR_ISDIR;
6224 				else if ((*vpp)->v_type == VLNK)
6225 					status = NFS4ERR_SYMLINK;
6226 				else
6227 					status = NFS4ERR_INVAL;
6228 				VN_RELE(*vpp);
6229 				return (status);
6230 			}
6231 
6232 			return (NFS4_OK);
6233 		}
6234 
6235 		/* Check for duplicate request */
6236 		va.va_mask = AT_MTIME;
6237 		error = VOP_GETATTR(*vpp, &va, 0, cr, NULL);
6238 		if (!error) {
6239 			/* We found the file */
6240 			const timestruc_t *mtime = &vap->va_mtime;
6241 
6242 			if (va.va_mtime.tv_sec != mtime->tv_sec ||
6243 			    va.va_mtime.tv_nsec != mtime->tv_nsec) {
6244 				/* but its not our creation */
6245 				VN_RELE(*vpp);
6246 				return (NFS4ERR_EXIST);
6247 			}
6248 			*created = TRUE; /* retrans of create == created */
6249 			return (NFS4_OK);
6250 		}
6251 		VN_RELE(*vpp);
6252 		return (NFS4ERR_EXIST);
6253 	}
6254 
6255 	return (NFS4_OK);
6256 }
6257 
6258 static nfsstat4
6259 check_open_access(uint32_t access, struct compound_state *cs,
6260     struct svc_req *req)
6261 {
6262 	int error;
6263 	vnode_t *vp;
6264 	bool_t readonly;
6265 	cred_t *cr = cs->cr;
6266 
6267 	/* For now we don't allow mandatory locking as per V2/V3 */
6268 	if (cs->access == CS_ACCESS_DENIED || cs->mandlock) {
6269 		return (NFS4ERR_ACCESS);
6270 	}
6271 
6272 	vp = cs->vp;
6273 	ASSERT(cr != NULL && vp->v_type == VREG);
6274 
6275 	/*
6276 	 * If the file system is exported read only and we are trying
6277 	 * to open for write, then return NFS4ERR_ROFS
6278 	 */
6279 
6280 	readonly = rdonly4(req, cs);
6281 
6282 	if ((access & OPEN4_SHARE_ACCESS_WRITE) && readonly)
6283 		return (NFS4ERR_ROFS);
6284 
6285 	if (access & OPEN4_SHARE_ACCESS_READ) {
6286 		if ((VOP_ACCESS(vp, VREAD, 0, cr, NULL) != 0) &&
6287 		    (VOP_ACCESS(vp, VEXEC, 0, cr, NULL) != 0)) {
6288 			return (NFS4ERR_ACCESS);
6289 		}
6290 	}
6291 
6292 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
6293 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
6294 		if (error)
6295 			return (NFS4ERR_ACCESS);
6296 	}
6297 
6298 	return (NFS4_OK);
6299 }
6300 
6301 static nfsstat4
6302 rfs4_createfile(OPEN4args *args, struct svc_req *req, struct compound_state *cs,
6303     change_info4 *cinfo, bitmap4 *attrset, clientid4 clientid)
6304 {
6305 	struct nfs4_svgetit_arg sarg;
6306 	struct nfs4_ntov_table ntov;
6307 
6308 	bool_t ntov_table_init = FALSE;
6309 	struct statvfs64 sb;
6310 	nfsstat4 status;
6311 	vnode_t *vp;
6312 	vattr_t bva, ava, iva, cva, *vap;
6313 	vnode_t *dvp;
6314 	timespec32_t *mtime;
6315 	char *nm = NULL;
6316 	uint_t buflen;
6317 	bool_t created;
6318 	bool_t setsize = FALSE;
6319 	len_t reqsize;
6320 	int error;
6321 	bool_t trunc;
6322 	caller_context_t ct;
6323 	component4 *component;
6324 	bslabel_t *clabel;
6325 	struct sockaddr *ca;
6326 	char *name = NULL;
6327 
6328 	sarg.sbp = &sb;
6329 	sarg.is_referral = B_FALSE;
6330 
6331 	dvp = cs->vp;
6332 
6333 	/* Check if the file system is read only */
6334 	if (rdonly4(req, cs))
6335 		return (NFS4ERR_ROFS);
6336 
6337 	/* check the label of including directory */
6338 	if (is_system_labeled()) {
6339 		ASSERT(req->rq_label != NULL);
6340 		clabel = req->rq_label;
6341 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
6342 		    "got client label from request(1)",
6343 		    struct svc_req *, req);
6344 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
6345 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
6346 			    cs->exi)) {
6347 				return (NFS4ERR_ACCESS);
6348 			}
6349 		}
6350 	}
6351 
6352 	/*
6353 	 * Get the last component of path name in nm. cs will reference
6354 	 * the including directory on success.
6355 	 */
6356 	component = &args->open_claim4_u.file;
6357 	status = utf8_dir_verify(component);
6358 	if (status != NFS4_OK)
6359 		return (status);
6360 
6361 	nm = utf8_to_fn(component, &buflen, NULL);
6362 
6363 	if (nm == NULL)
6364 		return (NFS4ERR_RESOURCE);
6365 
6366 	if (buflen > MAXNAMELEN) {
6367 		kmem_free(nm, buflen);
6368 		return (NFS4ERR_NAMETOOLONG);
6369 	}
6370 
6371 	bva.va_mask = AT_TYPE|AT_CTIME|AT_SEQ;
6372 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6373 	if (error) {
6374 		kmem_free(nm, buflen);
6375 		return (puterrno4(error));
6376 	}
6377 
6378 	if (bva.va_type != VDIR) {
6379 		kmem_free(nm, buflen);
6380 		return (NFS4ERR_NOTDIR);
6381 	}
6382 
6383 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime)
6384 
6385 	switch (args->mode) {
6386 	case GUARDED4:
6387 		/*FALLTHROUGH*/
6388 	case UNCHECKED4:
6389 		nfs4_ntov_table_init(&ntov);
6390 		ntov_table_init = TRUE;
6391 
6392 		*attrset = 0;
6393 		status = do_rfs4_set_attrs(attrset,
6394 		    &args->createhow4_u.createattrs,
6395 		    cs, &sarg, &ntov, NFS4ATTR_SETIT);
6396 
6397 		if (status == NFS4_OK && (sarg.vap->va_mask & AT_TYPE) &&
6398 		    sarg.vap->va_type != VREG) {
6399 			if (sarg.vap->va_type == VDIR)
6400 				status = NFS4ERR_ISDIR;
6401 			else if (sarg.vap->va_type == VLNK)
6402 				status = NFS4ERR_SYMLINK;
6403 			else
6404 				status = NFS4ERR_INVAL;
6405 		}
6406 
6407 		if (status != NFS4_OK) {
6408 			kmem_free(nm, buflen);
6409 			nfs4_ntov_table_free(&ntov, &sarg);
6410 			*attrset = 0;
6411 			return (status);
6412 		}
6413 
6414 		vap = sarg.vap;
6415 		vap->va_type = VREG;
6416 		vap->va_mask |= AT_TYPE;
6417 
6418 		if ((vap->va_mask & AT_MODE) == 0) {
6419 			vap->va_mask |= AT_MODE;
6420 			vap->va_mode = (mode_t)0600;
6421 		}
6422 
6423 		if (vap->va_mask & AT_SIZE) {
6424 
6425 			/* Disallow create with a non-zero size */
6426 
6427 			if ((reqsize = sarg.vap->va_size) != 0) {
6428 				kmem_free(nm, buflen);
6429 				nfs4_ntov_table_free(&ntov, &sarg);
6430 				*attrset = 0;
6431 				return (NFS4ERR_INVAL);
6432 			}
6433 			setsize = TRUE;
6434 		}
6435 		break;
6436 
6437 	case EXCLUSIVE4:
6438 		/* prohibit EXCL create of named attributes */
6439 		if (dvp->v_flag & V_XATTRDIR) {
6440 			kmem_free(nm, buflen);
6441 			*attrset = 0;
6442 			return (NFS4ERR_INVAL);
6443 		}
6444 
6445 		cva.va_mask = AT_TYPE | AT_MTIME | AT_MODE;
6446 		cva.va_type = VREG;
6447 		/*
6448 		 * Ensure no time overflows. Assumes underlying
6449 		 * filesystem supports at least 32 bits.
6450 		 * Truncate nsec to usec resolution to allow valid
6451 		 * compares even if the underlying filesystem truncates.
6452 		 */
6453 		mtime = (timespec32_t *)&args->createhow4_u.createverf;
6454 		cva.va_mtime.tv_sec = mtime->tv_sec % TIME32_MAX;
6455 		cva.va_mtime.tv_nsec = (mtime->tv_nsec / 1000) * 1000;
6456 		cva.va_mode = (mode_t)0;
6457 		vap = &cva;
6458 
6459 		/*
6460 		 * For EXCL create, attrset is set to the server attr
6461 		 * used to cache the client's verifier.
6462 		 */
6463 		*attrset = FATTR4_TIME_MODIFY_MASK;
6464 		break;
6465 	}
6466 
6467 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6468 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6469 	    MAXPATHLEN  + 1);
6470 
6471 	if (name == NULL) {
6472 		kmem_free(nm, buflen);
6473 		return (NFS4ERR_SERVERFAULT);
6474 	}
6475 
6476 	status = create_vnode(dvp, name, vap, args->mode,
6477 	    cs->cr, &vp, &created);
6478 	if (nm != name)
6479 		kmem_free(name, MAXPATHLEN + 1);
6480 	kmem_free(nm, buflen);
6481 
6482 	if (status != NFS4_OK) {
6483 		if (ntov_table_init)
6484 			nfs4_ntov_table_free(&ntov, &sarg);
6485 		*attrset = 0;
6486 		return (status);
6487 	}
6488 
6489 	trunc = (setsize && !created);
6490 
6491 	if (args->mode != EXCLUSIVE4) {
6492 		bitmap4 createmask = args->createhow4_u.createattrs.attrmask;
6493 
6494 		/*
6495 		 * True verification that object was created with correct
6496 		 * attrs is impossible.  The attrs could have been changed
6497 		 * immediately after object creation.  If attributes did
6498 		 * not verify, the only recourse for the server is to
6499 		 * destroy the object.  Maybe if some attrs (like gid)
6500 		 * are set incorrectly, the object should be destroyed;
6501 		 * however, seems bad as a default policy.  Do we really
6502 		 * want to destroy an object over one of the times not
6503 		 * verifying correctly?  For these reasons, the server
6504 		 * currently sets bits in attrset for createattrs
6505 		 * that were set; however, no verification is done.
6506 		 *
6507 		 * vmask_to_nmask accounts for vattr bits set on create
6508 		 *	[do_rfs4_set_attrs() only sets resp bits for
6509 		 *	 non-vattr/vfs bits.]
6510 		 * Mask off any bits we set by default so as not to return
6511 		 * more attrset bits than were requested in createattrs
6512 		 */
6513 		if (created) {
6514 			nfs4_vmask_to_nmask(sarg.vap->va_mask, attrset);
6515 			*attrset &= createmask;
6516 		} else {
6517 			/*
6518 			 * We did not create the vnode (we tried but it
6519 			 * already existed).  In this case, the only createattr
6520 			 * that the spec allows the server to set is size,
6521 			 * and even then, it can only be set if it is 0.
6522 			 */
6523 			*attrset = 0;
6524 			if (trunc)
6525 				*attrset = FATTR4_SIZE_MASK;
6526 		}
6527 	}
6528 	if (ntov_table_init)
6529 		nfs4_ntov_table_free(&ntov, &sarg);
6530 
6531 	/*
6532 	 * Get the initial "after" sequence number, if it fails,
6533 	 * set to zero, time to before.
6534 	 */
6535 	iva.va_mask = AT_CTIME|AT_SEQ;
6536 	if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL)) {
6537 		iva.va_seq = 0;
6538 		iva.va_ctime = bva.va_ctime;
6539 	}
6540 
6541 	/*
6542 	 * create_vnode attempts to create the file exclusive,
6543 	 * if it already exists the VOP_CREATE will fail and
6544 	 * may not increase va_seq. It is atomic if
6545 	 * we haven't changed the directory, but if it has changed
6546 	 * we don't know what changed it.
6547 	 */
6548 	if (!created) {
6549 		if (bva.va_seq && iva.va_seq &&
6550 		    bva.va_seq == iva.va_seq)
6551 			cinfo->atomic = TRUE;
6552 		else
6553 			cinfo->atomic = FALSE;
6554 		NFS4_SET_FATTR4_CHANGE(cinfo->after, iva.va_ctime);
6555 	} else {
6556 		/*
6557 		 * The entry was created, we need to sync the
6558 		 * directory metadata.
6559 		 */
6560 		(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
6561 
6562 		/*
6563 		 * Get "after" change value, if it fails, simply return the
6564 		 * before value.
6565 		 */
6566 		ava.va_mask = AT_CTIME|AT_SEQ;
6567 		if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6568 			ava.va_ctime = bva.va_ctime;
6569 			ava.va_seq = 0;
6570 		}
6571 
6572 		NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6573 
6574 		/*
6575 		 * The cinfo->atomic = TRUE only if we have
6576 		 * non-zero va_seq's, and it has incremented by exactly one
6577 		 * during the create_vnode and it didn't
6578 		 * change during the VOP_FSYNC.
6579 		 */
6580 		if (bva.va_seq && iva.va_seq && ava.va_seq &&
6581 		    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
6582 			cinfo->atomic = TRUE;
6583 		else
6584 			cinfo->atomic = FALSE;
6585 	}
6586 
6587 	/* Check for mandatory locking and that the size gets set. */
6588 	cva.va_mask = AT_MODE;
6589 	if (setsize)
6590 		cva.va_mask |= AT_SIZE;
6591 
6592 	/* Assume the worst */
6593 	cs->mandlock = TRUE;
6594 
6595 	if (VOP_GETATTR(vp, &cva, 0, cs->cr, NULL) == 0) {
6596 		cs->mandlock = MANDLOCK(cs->vp, cva.va_mode);
6597 
6598 		/*
6599 		 * Truncate the file if necessary; this would be
6600 		 * the case for create over an existing file.
6601 		 */
6602 
6603 		if (trunc) {
6604 			int in_crit = 0;
6605 			rfs4_file_t *fp;
6606 			bool_t create = FALSE;
6607 
6608 			/*
6609 			 * We are writing over an existing file.
6610 			 * Check to see if we need to recall a delegation.
6611 			 */
6612 			rfs4_hold_deleg_policy();
6613 			if ((fp = rfs4_findfile(vp, NULL, &create)) != NULL) {
6614 				if (rfs4_check_delegated_byfp(FWRITE, fp,
6615 				    (reqsize == 0), FALSE, FALSE, &clientid)) {
6616 					rfs4_file_rele(fp);
6617 					rfs4_rele_deleg_policy();
6618 					VN_RELE(vp);
6619 					*attrset = 0;
6620 					return (NFS4ERR_DELAY);
6621 				}
6622 				rfs4_file_rele(fp);
6623 			}
6624 			rfs4_rele_deleg_policy();
6625 
6626 			if (nbl_need_check(vp)) {
6627 				in_crit = 1;
6628 
6629 				ASSERT(reqsize == 0);
6630 
6631 				nbl_start_crit(vp, RW_READER);
6632 				if (nbl_conflict(vp, NBL_WRITE, 0,
6633 				    cva.va_size, 0, NULL)) {
6634 					in_crit = 0;
6635 					nbl_end_crit(vp);
6636 					VN_RELE(vp);
6637 					*attrset = 0;
6638 					return (NFS4ERR_ACCESS);
6639 				}
6640 			}
6641 			ct.cc_sysid = 0;
6642 			ct.cc_pid = 0;
6643 			ct.cc_caller_id = nfs4_srv_caller_id;
6644 			ct.cc_flags = CC_DONTBLOCK;
6645 
6646 			cva.va_mask = AT_SIZE;
6647 			cva.va_size = reqsize;
6648 			(void) VOP_SETATTR(vp, &cva, 0, cs->cr, &ct);
6649 			if (in_crit)
6650 				nbl_end_crit(vp);
6651 		}
6652 	}
6653 
6654 	error = makefh4(&cs->fh, vp, cs->exi);
6655 
6656 	/*
6657 	 * Force modified data and metadata out to stable storage.
6658 	 */
6659 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
6660 
6661 	if (error) {
6662 		VN_RELE(vp);
6663 		*attrset = 0;
6664 		return (puterrno4(error));
6665 	}
6666 
6667 	/* if parent dir is attrdir, set namedattr fh flag */
6668 	if (dvp->v_flag & V_XATTRDIR)
6669 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
6670 
6671 	if (cs->vp)
6672 		VN_RELE(cs->vp);
6673 
6674 	cs->vp = vp;
6675 
6676 	/*
6677 	 * if we did not create the file, we will need to check
6678 	 * the access bits on the file
6679 	 */
6680 
6681 	if (!created) {
6682 		if (setsize)
6683 			args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
6684 		status = check_open_access(args->share_access, cs, req);
6685 		if (status != NFS4_OK)
6686 			*attrset = 0;
6687 	}
6688 	return (status);
6689 }
6690 
6691 /*ARGSUSED*/
6692 static void
6693 rfs4_do_open(struct compound_state *cs, struct svc_req *req,
6694     rfs4_openowner_t *oo, delegreq_t deleg,
6695     uint32_t access, uint32_t deny,
6696     OPEN4res *resp, int deleg_cur)
6697 {
6698 	/* XXX Currently not using req  */
6699 	rfs4_state_t *sp;
6700 	rfs4_file_t *fp;
6701 	bool_t screate = TRUE;
6702 	bool_t fcreate = TRUE;
6703 	uint32_t open_a, share_a;
6704 	uint32_t open_d, share_d;
6705 	rfs4_deleg_state_t *dsp;
6706 	sysid_t sysid;
6707 	nfsstat4 status;
6708 	caller_context_t ct;
6709 	int fflags = 0;
6710 	int recall = 0;
6711 	int err;
6712 	int first_open;
6713 
6714 	/* get the file struct and hold a lock on it during initial open */
6715 	fp = rfs4_findfile_withlock(cs->vp, &cs->fh, &fcreate);
6716 	if (fp == NULL) {
6717 		resp->status = NFS4ERR_RESOURCE;
6718 		DTRACE_PROBE1(nfss__e__do__open1, nfsstat4, resp->status);
6719 		return;
6720 	}
6721 
6722 	sp = rfs4_findstate_by_owner_file(oo, fp, &screate);
6723 	if (sp == NULL) {
6724 		resp->status = NFS4ERR_RESOURCE;
6725 		DTRACE_PROBE1(nfss__e__do__open2, nfsstat4, resp->status);
6726 		/* No need to keep any reference */
6727 		rw_exit(&fp->rf_file_rwlock);
6728 		rfs4_file_rele(fp);
6729 		return;
6730 	}
6731 
6732 	/* try to get the sysid before continuing */
6733 	if ((status = rfs4_client_sysid(oo->ro_client, &sysid)) != NFS4_OK) {
6734 		resp->status = status;
6735 		rfs4_file_rele(fp);
6736 		/* Not a fully formed open; "close" it */
6737 		if (screate == TRUE)
6738 			rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6739 		rfs4_state_rele(sp);
6740 		return;
6741 	}
6742 
6743 	/* Calculate the fflags for this OPEN. */
6744 	if (access & OPEN4_SHARE_ACCESS_READ)
6745 		fflags |= FREAD;
6746 	if (access & OPEN4_SHARE_ACCESS_WRITE)
6747 		fflags |= FWRITE;
6748 
6749 	rfs4_dbe_lock(sp->rs_dbe);
6750 
6751 	/*
6752 	 * Calculate the new deny and access mode that this open is adding to
6753 	 * the file for this open owner;
6754 	 */
6755 	open_d = (deny & ~sp->rs_open_deny);
6756 	open_a = (access & ~sp->rs_open_access);
6757 
6758 	/*
6759 	 * Calculate the new share access and share deny modes that this open
6760 	 * is adding to the file for this open owner;
6761 	 */
6762 	share_a = (access & ~sp->rs_share_access);
6763 	share_d = (deny & ~sp->rs_share_deny);
6764 
6765 	first_open = (sp->rs_open_access & OPEN4_SHARE_ACCESS_BOTH) == 0;
6766 
6767 	/*
6768 	 * Check to see the client has already sent an open for this
6769 	 * open owner on this file with the same share/deny modes.
6770 	 * If so, we don't need to check for a conflict and we don't
6771 	 * need to add another shrlock.  If not, then we need to
6772 	 * check for conflicts in deny and access before checking for
6773 	 * conflicts in delegation.  We don't want to recall a
6774 	 * delegation based on an open that will eventually fail based
6775 	 * on shares modes.
6776 	 */
6777 
6778 	if (share_a || share_d) {
6779 		if ((err = rfs4_share(sp, access, deny)) != 0) {
6780 			rfs4_dbe_unlock(sp->rs_dbe);
6781 			resp->status = err;
6782 
6783 			rfs4_file_rele(fp);
6784 			/* Not a fully formed open; "close" it */
6785 			if (screate == TRUE)
6786 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6787 			rfs4_state_rele(sp);
6788 			return;
6789 		}
6790 	}
6791 
6792 	rfs4_dbe_lock(fp->rf_dbe);
6793 
6794 	/*
6795 	 * Check to see if this file is delegated and if so, if a
6796 	 * recall needs to be done.
6797 	 */
6798 	if (rfs4_check_recall(sp, access)) {
6799 		rfs4_dbe_unlock(fp->rf_dbe);
6800 		rfs4_dbe_unlock(sp->rs_dbe);
6801 		rfs4_recall_deleg(fp, FALSE, sp->rs_owner->ro_client);
6802 		delay(NFS4_DELEGATION_CONFLICT_DELAY);
6803 		rfs4_dbe_lock(sp->rs_dbe);
6804 
6805 		/* if state closed while lock was dropped */
6806 		if (sp->rs_closed) {
6807 			if (share_a || share_d)
6808 				(void) rfs4_unshare(sp);
6809 			rfs4_dbe_unlock(sp->rs_dbe);
6810 			rfs4_file_rele(fp);
6811 			/* Not a fully formed open; "close" it */
6812 			if (screate == TRUE)
6813 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6814 			rfs4_state_rele(sp);
6815 			resp->status = NFS4ERR_OLD_STATEID;
6816 			return;
6817 		}
6818 
6819 		rfs4_dbe_lock(fp->rf_dbe);
6820 		/* Let's see if the delegation was returned */
6821 		if (rfs4_check_recall(sp, access)) {
6822 			rfs4_dbe_unlock(fp->rf_dbe);
6823 			if (share_a || share_d)
6824 				(void) rfs4_unshare(sp);
6825 			rfs4_dbe_unlock(sp->rs_dbe);
6826 			rfs4_file_rele(fp);
6827 			rfs4_update_lease(sp->rs_owner->ro_client);
6828 
6829 			/* Not a fully formed open; "close" it */
6830 			if (screate == TRUE)
6831 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6832 			rfs4_state_rele(sp);
6833 			resp->status = NFS4ERR_DELAY;
6834 			return;
6835 		}
6836 	}
6837 	/*
6838 	 * the share check passed and any delegation conflict has been
6839 	 * taken care of, now call vop_open.
6840 	 * if this is the first open then call vop_open with fflags.
6841 	 * if not, call vn_open_upgrade with just the upgrade flags.
6842 	 *
6843 	 * if the file has been opened already, it will have the current
6844 	 * access mode in the state struct.  if it has no share access, then
6845 	 * this is a new open.
6846 	 *
6847 	 * However, if this is open with CLAIM_DLEGATE_CUR, then don't
6848 	 * call VOP_OPEN(), just do the open upgrade.
6849 	 */
6850 	if (first_open && !deleg_cur) {
6851 		ct.cc_sysid = sysid;
6852 		ct.cc_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
6853 		ct.cc_caller_id = nfs4_srv_caller_id;
6854 		ct.cc_flags = CC_DONTBLOCK;
6855 		err = VOP_OPEN(&cs->vp, fflags, cs->cr, &ct);
6856 		if (err) {
6857 			rfs4_dbe_unlock(fp->rf_dbe);
6858 			if (share_a || share_d)
6859 				(void) rfs4_unshare(sp);
6860 			rfs4_dbe_unlock(sp->rs_dbe);
6861 			rfs4_file_rele(fp);
6862 
6863 			/* Not a fully formed open; "close" it */
6864 			if (screate == TRUE)
6865 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6866 			rfs4_state_rele(sp);
6867 			/* check if a monitor detected a delegation conflict */
6868 			if (err == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
6869 				resp->status = NFS4ERR_DELAY;
6870 			else
6871 				resp->status = NFS4ERR_SERVERFAULT;
6872 			return;
6873 		}
6874 	} else { /* open upgrade */
6875 		/*
6876 		 * calculate the fflags for the new mode that is being added
6877 		 * by this upgrade.
6878 		 */
6879 		fflags = 0;
6880 		if (open_a & OPEN4_SHARE_ACCESS_READ)
6881 			fflags |= FREAD;
6882 		if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6883 			fflags |= FWRITE;
6884 		vn_open_upgrade(cs->vp, fflags);
6885 	}
6886 	sp->rs_open_access |= access;
6887 	sp->rs_open_deny |= deny;
6888 
6889 	if (open_d & OPEN4_SHARE_DENY_READ)
6890 		fp->rf_deny_read++;
6891 	if (open_d & OPEN4_SHARE_DENY_WRITE)
6892 		fp->rf_deny_write++;
6893 	fp->rf_share_deny |= deny;
6894 
6895 	if (open_a & OPEN4_SHARE_ACCESS_READ)
6896 		fp->rf_access_read++;
6897 	if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6898 		fp->rf_access_write++;
6899 	fp->rf_share_access |= access;
6900 
6901 	/*
6902 	 * Check for delegation here. if the deleg argument is not
6903 	 * DELEG_ANY, then this is a reclaim from a client and
6904 	 * we must honor the delegation requested. If necessary we can
6905 	 * set the recall flag.
6906 	 */
6907 
6908 	dsp = rfs4_grant_delegation(deleg, sp, &recall);
6909 
6910 	cs->deleg = (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE);
6911 
6912 	next_stateid(&sp->rs_stateid);
6913 
6914 	resp->stateid = sp->rs_stateid.stateid;
6915 
6916 	rfs4_dbe_unlock(fp->rf_dbe);
6917 	rfs4_dbe_unlock(sp->rs_dbe);
6918 
6919 	if (dsp) {
6920 		rfs4_set_deleg_response(dsp, &resp->delegation, NULL, recall);
6921 		rfs4_deleg_state_rele(dsp);
6922 	}
6923 
6924 	rfs4_file_rele(fp);
6925 	rfs4_state_rele(sp);
6926 
6927 	resp->status = NFS4_OK;
6928 }
6929 
6930 /*ARGSUSED*/
6931 static void
6932 rfs4_do_opennull(struct compound_state *cs, struct svc_req *req,
6933     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6934 {
6935 	change_info4 *cinfo = &resp->cinfo;
6936 	bitmap4 *attrset = &resp->attrset;
6937 
6938 	if (args->opentype == OPEN4_NOCREATE)
6939 		resp->status = rfs4_lookupfile(&args->open_claim4_u.file,
6940 		    req, cs, args->share_access, cinfo);
6941 	else {
6942 		/* inhibit delegation grants during exclusive create */
6943 
6944 		if (args->mode == EXCLUSIVE4)
6945 			rfs4_disable_delegation();
6946 
6947 		resp->status = rfs4_createfile(args, req, cs, cinfo, attrset,
6948 		    oo->ro_client->rc_clientid);
6949 	}
6950 
6951 	if (resp->status == NFS4_OK) {
6952 
6953 		/* cs->vp cs->fh now reference the desired file */
6954 
6955 		rfs4_do_open(cs, req, oo,
6956 		    oo->ro_need_confirm ? DELEG_NONE : DELEG_ANY,
6957 		    args->share_access, args->share_deny, resp, 0);
6958 
6959 		/*
6960 		 * If rfs4_createfile set attrset, we must
6961 		 * clear this attrset before the response is copied.
6962 		 */
6963 		if (resp->status != NFS4_OK && resp->attrset) {
6964 			resp->attrset = 0;
6965 		}
6966 	}
6967 	else
6968 		*cs->statusp = resp->status;
6969 
6970 	if (args->mode == EXCLUSIVE4)
6971 		rfs4_enable_delegation();
6972 }
6973 
6974 /*ARGSUSED*/
6975 static void
6976 rfs4_do_openprev(struct compound_state *cs, struct svc_req *req,
6977     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6978 {
6979 	change_info4 *cinfo = &resp->cinfo;
6980 	vattr_t va;
6981 	vtype_t v_type = cs->vp->v_type;
6982 	int error = 0;
6983 
6984 	/* Verify that we have a regular file */
6985 	if (v_type != VREG) {
6986 		if (v_type == VDIR)
6987 			resp->status = NFS4ERR_ISDIR;
6988 		else if (v_type == VLNK)
6989 			resp->status = NFS4ERR_SYMLINK;
6990 		else
6991 			resp->status = NFS4ERR_INVAL;
6992 		return;
6993 	}
6994 
6995 	va.va_mask = AT_MODE|AT_UID;
6996 	error = VOP_GETATTR(cs->vp, &va, 0, cs->cr, NULL);
6997 	if (error) {
6998 		resp->status = puterrno4(error);
6999 		return;
7000 	}
7001 
7002 	cs->mandlock = MANDLOCK(cs->vp, va.va_mode);
7003 
7004 	/*
7005 	 * Check if we have access to the file, Note the the file
7006 	 * could have originally been open UNCHECKED or GUARDED
7007 	 * with mode bits that will now fail, but there is nothing
7008 	 * we can really do about that except in the case that the
7009 	 * owner of the file is the one requesting the open.
7010 	 */
7011 	if (crgetuid(cs->cr) != va.va_uid) {
7012 		resp->status = check_open_access(args->share_access, cs, req);
7013 		if (resp->status != NFS4_OK) {
7014 			return;
7015 		}
7016 	}
7017 
7018 	/*
7019 	 * cinfo on a CLAIM_PREVIOUS is undefined, initialize to zero
7020 	 */
7021 	cinfo->before = 0;
7022 	cinfo->after = 0;
7023 	cinfo->atomic = FALSE;
7024 
7025 	rfs4_do_open(cs, req, oo,
7026 	    NFS4_DELEG4TYPE2REQTYPE(args->open_claim4_u.delegate_type),
7027 	    args->share_access, args->share_deny, resp, 0);
7028 }
7029 
7030 static void
7031 rfs4_do_opendelcur(struct compound_state *cs, struct svc_req *req,
7032     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7033 {
7034 	int error;
7035 	nfsstat4 status;
7036 	stateid4 stateid =
7037 	    args->open_claim4_u.delegate_cur_info.delegate_stateid;
7038 	rfs4_deleg_state_t *dsp;
7039 
7040 	/*
7041 	 * Find the state info from the stateid and confirm that the
7042 	 * file is delegated.  If the state openowner is the same as
7043 	 * the supplied openowner we're done. If not, get the file
7044 	 * info from the found state info. Use that file info to
7045 	 * create the state for this lock owner. Note solaris doen't
7046 	 * really need the pathname to find the file. We may want to
7047 	 * lookup the pathname and make sure that the vp exist and
7048 	 * matches the vp in the file structure. However it is
7049 	 * possible that the pathname nolonger exists (local process
7050 	 * unlinks the file), so this may not be that useful.
7051 	 */
7052 
7053 	status = rfs4_get_deleg_state(&stateid, &dsp);
7054 	if (status != NFS4_OK) {
7055 		resp->status = status;
7056 		return;
7057 	}
7058 
7059 	ASSERT(dsp->rds_finfo->rf_dinfo.rd_dtype != OPEN_DELEGATE_NONE);
7060 
7061 	/*
7062 	 * New lock owner, create state. Since this was probably called
7063 	 * in response to a CB_RECALL we set deleg to DELEG_NONE
7064 	 */
7065 
7066 	ASSERT(cs->vp != NULL);
7067 	VN_RELE(cs->vp);
7068 	VN_HOLD(dsp->rds_finfo->rf_vp);
7069 	cs->vp = dsp->rds_finfo->rf_vp;
7070 
7071 	if (error = makefh4(&cs->fh, cs->vp, cs->exi)) {
7072 		rfs4_deleg_state_rele(dsp);
7073 		*cs->statusp = resp->status = puterrno4(error);
7074 		return;
7075 	}
7076 
7077 	/* Mark progress for delegation returns */
7078 	dsp->rds_finfo->rf_dinfo.rd_time_lastwrite = gethrestime_sec();
7079 	rfs4_deleg_state_rele(dsp);
7080 	rfs4_do_open(cs, req, oo, DELEG_NONE,
7081 	    args->share_access, args->share_deny, resp, 1);
7082 }
7083 
7084 /*ARGSUSED*/
7085 static void
7086 rfs4_do_opendelprev(struct compound_state *cs, struct svc_req *req,
7087     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7088 {
7089 	/*
7090 	 * Lookup the pathname, it must already exist since this file
7091 	 * was delegated.
7092 	 *
7093 	 * Find the file and state info for this vp and open owner pair.
7094 	 *	check that they are in fact delegated.
7095 	 *	check that the state access and deny modes are the same.
7096 	 *
7097 	 * Return the delgation possibly seting the recall flag.
7098 	 */
7099 	rfs4_file_t *fp;
7100 	rfs4_state_t *sp;
7101 	bool_t create = FALSE;
7102 	bool_t dcreate = FALSE;
7103 	rfs4_deleg_state_t *dsp;
7104 	nfsace4 *ace;
7105 
7106 	/* Note we ignore oflags */
7107 	resp->status = rfs4_lookupfile(&args->open_claim4_u.file_delegate_prev,
7108 	    req, cs, args->share_access, &resp->cinfo);
7109 
7110 	if (resp->status != NFS4_OK) {
7111 		return;
7112 	}
7113 
7114 	/* get the file struct and hold a lock on it during initial open */
7115 	fp = rfs4_findfile_withlock(cs->vp, NULL, &create);
7116 	if (fp == NULL) {
7117 		resp->status = NFS4ERR_RESOURCE;
7118 		DTRACE_PROBE1(nfss__e__do_opendelprev1, nfsstat4, resp->status);
7119 		return;
7120 	}
7121 
7122 	sp = rfs4_findstate_by_owner_file(oo, fp, &create);
7123 	if (sp == NULL) {
7124 		resp->status = NFS4ERR_SERVERFAULT;
7125 		DTRACE_PROBE1(nfss__e__do_opendelprev2, nfsstat4, resp->status);
7126 		rw_exit(&fp->rf_file_rwlock);
7127 		rfs4_file_rele(fp);
7128 		return;
7129 	}
7130 
7131 	rfs4_dbe_lock(sp->rs_dbe);
7132 	rfs4_dbe_lock(fp->rf_dbe);
7133 	if (args->share_access != sp->rs_share_access ||
7134 	    args->share_deny != sp->rs_share_deny ||
7135 	    sp->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
7136 		NFS4_DEBUG(rfs4_debug,
7137 		    (CE_NOTE, "rfs4_do_opendelprev: state mixup"));
7138 		rfs4_dbe_unlock(fp->rf_dbe);
7139 		rfs4_dbe_unlock(sp->rs_dbe);
7140 		rfs4_file_rele(fp);
7141 		rfs4_state_rele(sp);
7142 		resp->status = NFS4ERR_SERVERFAULT;
7143 		return;
7144 	}
7145 	rfs4_dbe_unlock(fp->rf_dbe);
7146 	rfs4_dbe_unlock(sp->rs_dbe);
7147 
7148 	dsp = rfs4_finddeleg(sp, &dcreate);
7149 	if (dsp == NULL) {
7150 		rfs4_state_rele(sp);
7151 		rfs4_file_rele(fp);
7152 		resp->status = NFS4ERR_SERVERFAULT;
7153 		return;
7154 	}
7155 
7156 	next_stateid(&sp->rs_stateid);
7157 
7158 	resp->stateid = sp->rs_stateid.stateid;
7159 
7160 	resp->delegation.delegation_type = dsp->rds_dtype;
7161 
7162 	if (dsp->rds_dtype == OPEN_DELEGATE_READ) {
7163 		open_read_delegation4 *rv =
7164 		    &resp->delegation.open_delegation4_u.read;
7165 
7166 		rv->stateid = dsp->rds_delegid.stateid;
7167 		rv->recall = FALSE; /* no policy in place to set to TRUE */
7168 		ace = &rv->permissions;
7169 	} else {
7170 		open_write_delegation4 *rv =
7171 		    &resp->delegation.open_delegation4_u.write;
7172 
7173 		rv->stateid = dsp->rds_delegid.stateid;
7174 		rv->recall = FALSE;  /* no policy in place to set to TRUE */
7175 		ace = &rv->permissions;
7176 		rv->space_limit.limitby = NFS_LIMIT_SIZE;
7177 		rv->space_limit.nfs_space_limit4_u.filesize = UINT64_MAX;
7178 	}
7179 
7180 	/* XXX For now */
7181 	ace->type = ACE4_ACCESS_ALLOWED_ACE_TYPE;
7182 	ace->flag = 0;
7183 	ace->access_mask = 0;
7184 	ace->who.utf8string_len = 0;
7185 	ace->who.utf8string_val = 0;
7186 
7187 	rfs4_deleg_state_rele(dsp);
7188 	rfs4_state_rele(sp);
7189 	rfs4_file_rele(fp);
7190 }
7191 
7192 typedef enum {
7193 	NFS4_CHKSEQ_OKAY = 0,
7194 	NFS4_CHKSEQ_REPLAY = 1,
7195 	NFS4_CHKSEQ_BAD = 2
7196 } rfs4_chkseq_t;
7197 
7198 /*
7199  * Generic function for sequence number checks.
7200  */
7201 static rfs4_chkseq_t
7202 rfs4_check_seqid(seqid4 seqid, nfs_resop4 *lastop,
7203     seqid4 rqst_seq, nfs_resop4 *resop, bool_t copyres)
7204 {
7205 	/* Same sequence ids and matching operations? */
7206 	if (seqid == rqst_seq && resop->resop == lastop->resop) {
7207 		if (copyres == TRUE) {
7208 			rfs4_free_reply(resop);
7209 			rfs4_copy_reply(resop, lastop);
7210 		}
7211 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
7212 		    "Replayed SEQID %d\n", seqid));
7213 		return (NFS4_CHKSEQ_REPLAY);
7214 	}
7215 
7216 	/* If the incoming sequence is not the next expected then it is bad */
7217 	if (rqst_seq != seqid + 1) {
7218 		if (rqst_seq == seqid) {
7219 			NFS4_DEBUG(rfs4_debug,
7220 			    (CE_NOTE, "BAD SEQID: Replayed sequence id "
7221 			    "but last op was %d current op is %d\n",
7222 			    lastop->resop, resop->resop));
7223 			return (NFS4_CHKSEQ_BAD);
7224 		}
7225 		NFS4_DEBUG(rfs4_debug,
7226 		    (CE_NOTE, "BAD SEQID: got %u expecting %u\n",
7227 		    rqst_seq, seqid));
7228 		return (NFS4_CHKSEQ_BAD);
7229 	}
7230 
7231 	/* Everything okay -- next expected */
7232 	return (NFS4_CHKSEQ_OKAY);
7233 }
7234 
7235 
7236 static rfs4_chkseq_t
7237 rfs4_check_open_seqid(seqid4 seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7238 {
7239 	rfs4_chkseq_t rc;
7240 
7241 	rfs4_dbe_lock(op->ro_dbe);
7242 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply, seqid, resop,
7243 	    TRUE);
7244 	rfs4_dbe_unlock(op->ro_dbe);
7245 
7246 	if (rc == NFS4_CHKSEQ_OKAY)
7247 		rfs4_update_lease(op->ro_client);
7248 
7249 	return (rc);
7250 }
7251 
7252 static rfs4_chkseq_t
7253 rfs4_check_olo_seqid(seqid4 olo_seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7254 {
7255 	rfs4_chkseq_t rc;
7256 
7257 	rfs4_dbe_lock(op->ro_dbe);
7258 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply,
7259 	    olo_seqid, resop, FALSE);
7260 	rfs4_dbe_unlock(op->ro_dbe);
7261 
7262 	return (rc);
7263 }
7264 
7265 static rfs4_chkseq_t
7266 rfs4_check_lock_seqid(seqid4 seqid, rfs4_lo_state_t *lsp, nfs_resop4 *resop)
7267 {
7268 	rfs4_chkseq_t rc = NFS4_CHKSEQ_OKAY;
7269 
7270 	rfs4_dbe_lock(lsp->rls_dbe);
7271 	if (!lsp->rls_skip_seqid_check)
7272 		rc = rfs4_check_seqid(lsp->rls_seqid, &lsp->rls_reply, seqid,
7273 		    resop, TRUE);
7274 	rfs4_dbe_unlock(lsp->rls_dbe);
7275 
7276 	return (rc);
7277 }
7278 
7279 static void
7280 rfs4_op_open(nfs_argop4 *argop, nfs_resop4 *resop,
7281     struct svc_req *req, struct compound_state *cs)
7282 {
7283 	OPEN4args *args = &argop->nfs_argop4_u.opopen;
7284 	OPEN4res *resp = &resop->nfs_resop4_u.opopen;
7285 	open_owner4 *owner = &args->owner;
7286 	open_claim_type4 claim = args->claim;
7287 	rfs4_client_t *cp;
7288 	rfs4_openowner_t *oo;
7289 	bool_t create;
7290 	bool_t replay = FALSE;
7291 	int can_reclaim;
7292 
7293 	DTRACE_NFSV4_2(op__open__start, struct compound_state *, cs,
7294 	    OPEN4args *, args);
7295 
7296 	if (cs->vp == NULL) {
7297 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7298 		goto end;
7299 	}
7300 
7301 	/*
7302 	 * Need to check clientid and lease expiration first based on
7303 	 * error ordering and incrementing sequence id.
7304 	 */
7305 	cp = rfs4_findclient_by_id(owner->clientid, FALSE);
7306 	if (cp == NULL) {
7307 		*cs->statusp = resp->status =
7308 		    rfs4_check_clientid(&owner->clientid, 0);
7309 		goto end;
7310 	}
7311 
7312 	if (rfs4_lease_expired(cp)) {
7313 		rfs4_client_close(cp);
7314 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7315 		goto end;
7316 	}
7317 	can_reclaim = cp->rc_can_reclaim;
7318 
7319 	/*
7320 	 * Find the open_owner for use from this point forward.  Take
7321 	 * care in updating the sequence id based on the type of error
7322 	 * being returned.
7323 	 */
7324 retry:
7325 	create = TRUE;
7326 	oo = rfs4_findopenowner(owner, &create, args->seqid);
7327 	if (oo == NULL) {
7328 		*cs->statusp = resp->status = NFS4ERR_RESOURCE;
7329 		rfs4_client_rele(cp);
7330 		goto end;
7331 	}
7332 
7333 	/* Hold off access to the sequence space while the open is done */
7334 	rfs4_sw_enter(&oo->ro_sw);
7335 
7336 	/*
7337 	 * If the open_owner existed before at the server, then check
7338 	 * the sequence id.
7339 	 */
7340 	if (!create && !oo->ro_postpone_confirm) {
7341 		switch (rfs4_check_open_seqid(args->seqid, oo, resop)) {
7342 		case NFS4_CHKSEQ_BAD:
7343 			if ((args->seqid > oo->ro_open_seqid) &&
7344 			    oo->ro_need_confirm) {
7345 				rfs4_free_opens(oo, TRUE, FALSE);
7346 				rfs4_sw_exit(&oo->ro_sw);
7347 				rfs4_openowner_rele(oo);
7348 				goto retry;
7349 			}
7350 			resp->status = NFS4ERR_BAD_SEQID;
7351 			goto out;
7352 		case NFS4_CHKSEQ_REPLAY: /* replay of previous request */
7353 			replay = TRUE;
7354 			goto out;
7355 		default:
7356 			break;
7357 		}
7358 
7359 		/*
7360 		 * Sequence was ok and open owner exists
7361 		 * check to see if we have yet to see an
7362 		 * open_confirm.
7363 		 */
7364 		if (oo->ro_need_confirm) {
7365 			rfs4_free_opens(oo, TRUE, FALSE);
7366 			rfs4_sw_exit(&oo->ro_sw);
7367 			rfs4_openowner_rele(oo);
7368 			goto retry;
7369 		}
7370 	}
7371 	/* Grace only applies to regular-type OPENs */
7372 	if (rfs4_clnt_in_grace(cp) &&
7373 	    (claim == CLAIM_NULL || claim == CLAIM_DELEGATE_CUR)) {
7374 		*cs->statusp = resp->status = NFS4ERR_GRACE;
7375 		goto out;
7376 	}
7377 
7378 	/*
7379 	 * If previous state at the server existed then can_reclaim
7380 	 * will be set. If not reply NFS4ERR_NO_GRACE to the
7381 	 * client.
7382 	 */
7383 	if (rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS && !can_reclaim) {
7384 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7385 		goto out;
7386 	}
7387 
7388 
7389 	/*
7390 	 * Reject the open if the client has missed the grace period
7391 	 */
7392 	if (!rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS) {
7393 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7394 		goto out;
7395 	}
7396 
7397 	/* Couple of up-front bookkeeping items */
7398 	if (oo->ro_need_confirm) {
7399 		/*
7400 		 * If this is a reclaim OPEN then we should not ask
7401 		 * for a confirmation of the open_owner per the
7402 		 * protocol specification.
7403 		 */
7404 		if (claim == CLAIM_PREVIOUS)
7405 			oo->ro_need_confirm = FALSE;
7406 		else
7407 			resp->rflags |= OPEN4_RESULT_CONFIRM;
7408 	}
7409 	resp->rflags |= OPEN4_RESULT_LOCKTYPE_POSIX;
7410 
7411 	/*
7412 	 * If there is an unshared filesystem mounted on this vnode,
7413 	 * do not allow to open/create in this directory.
7414 	 */
7415 	if (vn_ismntpt(cs->vp)) {
7416 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
7417 		goto out;
7418 	}
7419 
7420 	/*
7421 	 * access must READ, WRITE, or BOTH.  No access is invalid.
7422 	 * deny can be READ, WRITE, BOTH, or NONE.
7423 	 * bits not defined for access/deny are invalid.
7424 	 */
7425 	if (! (args->share_access & OPEN4_SHARE_ACCESS_BOTH) ||
7426 	    (args->share_access & ~OPEN4_SHARE_ACCESS_BOTH) ||
7427 	    (args->share_deny & ~OPEN4_SHARE_DENY_BOTH)) {
7428 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7429 		goto out;
7430 	}
7431 
7432 
7433 	/*
7434 	 * make sure attrset is zero before response is built.
7435 	 */
7436 	resp->attrset = 0;
7437 
7438 	switch (claim) {
7439 	case CLAIM_NULL:
7440 		rfs4_do_opennull(cs, req, args, oo, resp);
7441 		break;
7442 	case CLAIM_PREVIOUS:
7443 		rfs4_do_openprev(cs, req, args, oo, resp);
7444 		break;
7445 	case CLAIM_DELEGATE_CUR:
7446 		rfs4_do_opendelcur(cs, req, args, oo, resp);
7447 		break;
7448 	case CLAIM_DELEGATE_PREV:
7449 		rfs4_do_opendelprev(cs, req, args, oo, resp);
7450 		break;
7451 	default:
7452 		resp->status = NFS4ERR_INVAL;
7453 		break;
7454 	}
7455 
7456 out:
7457 	rfs4_client_rele(cp);
7458 
7459 	/* Catch sequence id handling here to make it a little easier */
7460 	switch (resp->status) {
7461 	case NFS4ERR_BADXDR:
7462 	case NFS4ERR_BAD_SEQID:
7463 	case NFS4ERR_BAD_STATEID:
7464 	case NFS4ERR_NOFILEHANDLE:
7465 	case NFS4ERR_RESOURCE:
7466 	case NFS4ERR_STALE_CLIENTID:
7467 	case NFS4ERR_STALE_STATEID:
7468 		/*
7469 		 * The protocol states that if any of these errors are
7470 		 * being returned, the sequence id should not be
7471 		 * incremented.  Any other return requires an
7472 		 * increment.
7473 		 */
7474 		break;
7475 	default:
7476 		/* Always update the lease in this case */
7477 		rfs4_update_lease(oo->ro_client);
7478 
7479 		/* Regular response - copy the result */
7480 		if (!replay)
7481 			rfs4_update_open_resp(oo, resop, &cs->fh);
7482 
7483 		/*
7484 		 * REPLAY case: Only if the previous response was OK
7485 		 * do we copy the filehandle.  If not OK, no
7486 		 * filehandle to copy.
7487 		 */
7488 		if (replay == TRUE &&
7489 		    resp->status == NFS4_OK &&
7490 		    oo->ro_reply_fh.nfs_fh4_val) {
7491 			/*
7492 			 * If this is a replay, we must restore the
7493 			 * current filehandle/vp to that of what was
7494 			 * returned originally.  Try our best to do
7495 			 * it.
7496 			 */
7497 			nfs_fh4_fmt_t *fh_fmtp =
7498 			    (nfs_fh4_fmt_t *)oo->ro_reply_fh.nfs_fh4_val;
7499 
7500 			cs->exi = checkexport4(&fh_fmtp->fh4_fsid,
7501 			    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
7502 
7503 			if (cs->exi == NULL) {
7504 				resp->status = NFS4ERR_STALE;
7505 				goto finish;
7506 			}
7507 
7508 			VN_RELE(cs->vp);
7509 
7510 			cs->vp = nfs4_fhtovp(&oo->ro_reply_fh, cs->exi,
7511 			    &resp->status);
7512 
7513 			if (cs->vp == NULL)
7514 				goto finish;
7515 
7516 			nfs_fh4_copy(&oo->ro_reply_fh, &cs->fh);
7517 		}
7518 
7519 		/*
7520 		 * If this was a replay, no need to update the
7521 		 * sequence id. If the open_owner was not created on
7522 		 * this pass, then update.  The first use of an
7523 		 * open_owner will not bump the sequence id.
7524 		 */
7525 		if (replay == FALSE && !create)
7526 			rfs4_update_open_sequence(oo);
7527 		/*
7528 		 * If the client is receiving an error and the
7529 		 * open_owner needs to be confirmed, there is no way
7530 		 * to notify the client of this fact ignoring the fact
7531 		 * that the server has no method of returning a
7532 		 * stateid to confirm.  Therefore, the server needs to
7533 		 * mark this open_owner in a way as to avoid the
7534 		 * sequence id checking the next time the client uses
7535 		 * this open_owner.
7536 		 */
7537 		if (resp->status != NFS4_OK && oo->ro_need_confirm)
7538 			oo->ro_postpone_confirm = TRUE;
7539 		/*
7540 		 * If OK response then clear the postpone flag and
7541 		 * reset the sequence id to keep in sync with the
7542 		 * client.
7543 		 */
7544 		if (resp->status == NFS4_OK && oo->ro_postpone_confirm) {
7545 			oo->ro_postpone_confirm = FALSE;
7546 			oo->ro_open_seqid = args->seqid;
7547 		}
7548 		break;
7549 	}
7550 
7551 finish:
7552 	*cs->statusp = resp->status;
7553 
7554 	rfs4_sw_exit(&oo->ro_sw);
7555 	rfs4_openowner_rele(oo);
7556 
7557 end:
7558 	DTRACE_NFSV4_2(op__open__done, struct compound_state *, cs,
7559 	    OPEN4res *, resp);
7560 }
7561 
7562 /*ARGSUSED*/
7563 void
7564 rfs4_op_open_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
7565     struct svc_req *req, struct compound_state *cs)
7566 {
7567 	OPEN_CONFIRM4args *args = &argop->nfs_argop4_u.opopen_confirm;
7568 	OPEN_CONFIRM4res *resp = &resop->nfs_resop4_u.opopen_confirm;
7569 	rfs4_state_t *sp;
7570 	nfsstat4 status;
7571 
7572 	DTRACE_NFSV4_2(op__open__confirm__start, struct compound_state *, cs,
7573 	    OPEN_CONFIRM4args *, args);
7574 
7575 	if (cs->vp == NULL) {
7576 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7577 		goto out;
7578 	}
7579 
7580 	if (cs->vp->v_type != VREG) {
7581 		*cs->statusp = resp->status =
7582 		    cs->vp->v_type == VDIR ? NFS4ERR_ISDIR : NFS4ERR_INVAL;
7583 		return;
7584 	}
7585 
7586 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7587 	if (status != NFS4_OK) {
7588 		*cs->statusp = resp->status = status;
7589 		goto out;
7590 	}
7591 
7592 	/* Ensure specified filehandle matches */
7593 	if (cs->vp != sp->rs_finfo->rf_vp) {
7594 		rfs4_state_rele(sp);
7595 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7596 		goto out;
7597 	}
7598 
7599 	/* hold off other access to open_owner while we tinker */
7600 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7601 
7602 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7603 	case NFS4_CHECK_STATEID_OKAY:
7604 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7605 		    resop) != 0) {
7606 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7607 			break;
7608 		}
7609 		/*
7610 		 * If it is the appropriate stateid and determined to
7611 		 * be "OKAY" then this means that the stateid does not
7612 		 * need to be confirmed and the client is in error for
7613 		 * sending an OPEN_CONFIRM.
7614 		 */
7615 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7616 		break;
7617 	case NFS4_CHECK_STATEID_OLD:
7618 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7619 		break;
7620 	case NFS4_CHECK_STATEID_BAD:
7621 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7622 		break;
7623 	case NFS4_CHECK_STATEID_EXPIRED:
7624 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7625 		break;
7626 	case NFS4_CHECK_STATEID_CLOSED:
7627 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7628 		break;
7629 	case NFS4_CHECK_STATEID_REPLAY:
7630 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7631 		    resop)) {
7632 		case NFS4_CHKSEQ_OKAY:
7633 			/*
7634 			 * This is replayed stateid; if seqid matches
7635 			 * next expected, then client is using wrong seqid.
7636 			 */
7637 			/* fall through */
7638 		case NFS4_CHKSEQ_BAD:
7639 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7640 			break;
7641 		case NFS4_CHKSEQ_REPLAY:
7642 			/*
7643 			 * Note this case is the duplicate case so
7644 			 * resp->status is already set.
7645 			 */
7646 			*cs->statusp = resp->status;
7647 			rfs4_update_lease(sp->rs_owner->ro_client);
7648 			break;
7649 		}
7650 		break;
7651 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7652 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7653 		    resop) != NFS4_CHKSEQ_OKAY) {
7654 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7655 			break;
7656 		}
7657 		*cs->statusp = resp->status = NFS4_OK;
7658 
7659 		next_stateid(&sp->rs_stateid);
7660 		resp->open_stateid = sp->rs_stateid.stateid;
7661 		sp->rs_owner->ro_need_confirm = FALSE;
7662 		rfs4_update_lease(sp->rs_owner->ro_client);
7663 		rfs4_update_open_sequence(sp->rs_owner);
7664 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7665 		break;
7666 	default:
7667 		ASSERT(FALSE);
7668 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7669 		break;
7670 	}
7671 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7672 	rfs4_state_rele(sp);
7673 
7674 out:
7675 	DTRACE_NFSV4_2(op__open__confirm__done, struct compound_state *, cs,
7676 	    OPEN_CONFIRM4res *, resp);
7677 }
7678 
7679 /*ARGSUSED*/
7680 void
7681 rfs4_op_open_downgrade(nfs_argop4 *argop, nfs_resop4 *resop,
7682     struct svc_req *req, struct compound_state *cs)
7683 {
7684 	OPEN_DOWNGRADE4args *args = &argop->nfs_argop4_u.opopen_downgrade;
7685 	OPEN_DOWNGRADE4res *resp = &resop->nfs_resop4_u.opopen_downgrade;
7686 	uint32_t access = args->share_access;
7687 	uint32_t deny = args->share_deny;
7688 	nfsstat4 status;
7689 	rfs4_state_t *sp;
7690 	rfs4_file_t *fp;
7691 	int fflags = 0;
7692 
7693 	DTRACE_NFSV4_2(op__open__downgrade__start, struct compound_state *, cs,
7694 	    OPEN_DOWNGRADE4args *, args);
7695 
7696 	if (cs->vp == NULL) {
7697 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7698 		goto out;
7699 	}
7700 
7701 	if (cs->vp->v_type != VREG) {
7702 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7703 		return;
7704 	}
7705 
7706 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7707 	if (status != NFS4_OK) {
7708 		*cs->statusp = resp->status = status;
7709 		goto out;
7710 	}
7711 
7712 	/* Ensure specified filehandle matches */
7713 	if (cs->vp != sp->rs_finfo->rf_vp) {
7714 		rfs4_state_rele(sp);
7715 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7716 		goto out;
7717 	}
7718 
7719 	/* hold off other access to open_owner while we tinker */
7720 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7721 
7722 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7723 	case NFS4_CHECK_STATEID_OKAY:
7724 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7725 		    resop) != NFS4_CHKSEQ_OKAY) {
7726 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7727 			goto end;
7728 		}
7729 		break;
7730 	case NFS4_CHECK_STATEID_OLD:
7731 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7732 		goto end;
7733 	case NFS4_CHECK_STATEID_BAD:
7734 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7735 		goto end;
7736 	case NFS4_CHECK_STATEID_EXPIRED:
7737 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7738 		goto end;
7739 	case NFS4_CHECK_STATEID_CLOSED:
7740 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7741 		goto end;
7742 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7743 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7744 		goto end;
7745 	case NFS4_CHECK_STATEID_REPLAY:
7746 		/* Check the sequence id for the open owner */
7747 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7748 		    resop)) {
7749 		case NFS4_CHKSEQ_OKAY:
7750 			/*
7751 			 * This is replayed stateid; if seqid matches
7752 			 * next expected, then client is using wrong seqid.
7753 			 */
7754 			/* fall through */
7755 		case NFS4_CHKSEQ_BAD:
7756 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7757 			goto end;
7758 		case NFS4_CHKSEQ_REPLAY:
7759 			/*
7760 			 * Note this case is the duplicate case so
7761 			 * resp->status is already set.
7762 			 */
7763 			*cs->statusp = resp->status;
7764 			rfs4_update_lease(sp->rs_owner->ro_client);
7765 			goto end;
7766 		}
7767 		break;
7768 	default:
7769 		ASSERT(FALSE);
7770 		break;
7771 	}
7772 
7773 	rfs4_dbe_lock(sp->rs_dbe);
7774 	/*
7775 	 * Check that the new access modes and deny modes are valid.
7776 	 * Check that no invalid bits are set.
7777 	 */
7778 	if ((access & ~(OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) ||
7779 	    (deny & ~(OPEN4_SHARE_DENY_READ | OPEN4_SHARE_DENY_WRITE))) {
7780 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7781 		rfs4_update_open_sequence(sp->rs_owner);
7782 		rfs4_dbe_unlock(sp->rs_dbe);
7783 		goto end;
7784 	}
7785 
7786 	/*
7787 	 * The new modes must be a subset of the current modes and
7788 	 * the access must specify at least one mode. To test that
7789 	 * the new mode is a subset of the current modes we bitwise
7790 	 * AND them together and check that the result equals the new
7791 	 * mode. For example:
7792 	 * New mode, access == R and current mode, sp->rs_open_access  == RW
7793 	 * access & sp->rs_open_access == R == access, so the new access mode
7794 	 * is valid. Consider access == RW, sp->rs_open_access = R
7795 	 * access & sp->rs_open_access == R != access, so the new access mode
7796 	 * is invalid.
7797 	 */
7798 	if ((access & sp->rs_open_access) != access ||
7799 	    (deny & sp->rs_open_deny) != deny ||
7800 	    (access &
7801 	    (OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) == 0) {
7802 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7803 		rfs4_update_open_sequence(sp->rs_owner);
7804 		rfs4_dbe_unlock(sp->rs_dbe);
7805 		goto end;
7806 	}
7807 
7808 	/*
7809 	 * Release any share locks associated with this stateID.
7810 	 * Strictly speaking, this violates the spec because the
7811 	 * spec effectively requires that open downgrade be atomic.
7812 	 * At present, fs_shrlock does not have this capability.
7813 	 */
7814 	(void) rfs4_unshare(sp);
7815 
7816 	status = rfs4_share(sp, access, deny);
7817 	if (status != NFS4_OK) {
7818 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7819 		rfs4_update_open_sequence(sp->rs_owner);
7820 		rfs4_dbe_unlock(sp->rs_dbe);
7821 		goto end;
7822 	}
7823 
7824 	fp = sp->rs_finfo;
7825 	rfs4_dbe_lock(fp->rf_dbe);
7826 
7827 	/*
7828 	 * If the current mode has deny read and the new mode
7829 	 * does not, decrement the number of deny read mode bits
7830 	 * and if it goes to zero turn off the deny read bit
7831 	 * on the file.
7832 	 */
7833 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_READ) &&
7834 	    (deny & OPEN4_SHARE_DENY_READ) == 0) {
7835 		fp->rf_deny_read--;
7836 		if (fp->rf_deny_read == 0)
7837 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
7838 	}
7839 
7840 	/*
7841 	 * If the current mode has deny write and the new mode
7842 	 * does not, decrement the number of deny write mode bits
7843 	 * and if it goes to zero turn off the deny write bit
7844 	 * on the file.
7845 	 */
7846 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) &&
7847 	    (deny & OPEN4_SHARE_DENY_WRITE) == 0) {
7848 		fp->rf_deny_write--;
7849 		if (fp->rf_deny_write == 0)
7850 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
7851 	}
7852 
7853 	/*
7854 	 * If the current mode has access read and the new mode
7855 	 * does not, decrement the number of access read mode bits
7856 	 * and if it goes to zero turn off the access read bit
7857 	 * on the file.  set fflags to FREAD for the call to
7858 	 * vn_open_downgrade().
7859 	 */
7860 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) &&
7861 	    (access & OPEN4_SHARE_ACCESS_READ) == 0) {
7862 		fp->rf_access_read--;
7863 		if (fp->rf_access_read == 0)
7864 			fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
7865 		fflags |= FREAD;
7866 	}
7867 
7868 	/*
7869 	 * If the current mode has access write and the new mode
7870 	 * does not, decrement the number of access write mode bits
7871 	 * and if it goes to zero turn off the access write bit
7872 	 * on the file.  set fflags to FWRITE for the call to
7873 	 * vn_open_downgrade().
7874 	 */
7875 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) &&
7876 	    (access & OPEN4_SHARE_ACCESS_WRITE) == 0) {
7877 		fp->rf_access_write--;
7878 		if (fp->rf_access_write == 0)
7879 			fp->rf_share_deny &= ~OPEN4_SHARE_ACCESS_WRITE;
7880 		fflags |= FWRITE;
7881 	}
7882 
7883 	/* Check that the file is still accessible */
7884 	ASSERT(fp->rf_share_access);
7885 
7886 	rfs4_dbe_unlock(fp->rf_dbe);
7887 
7888 	/* now set the new open access and deny modes */
7889 	sp->rs_open_access = access;
7890 	sp->rs_open_deny = deny;
7891 
7892 	/*
7893 	 * we successfully downgraded the share lock, now we need to downgrade
7894 	 * the open. it is possible that the downgrade was only for a deny
7895 	 * mode and we have nothing else to do.
7896 	 */
7897 	if ((fflags & (FREAD|FWRITE)) != 0)
7898 		vn_open_downgrade(cs->vp, fflags);
7899 
7900 	/* Update the stateid */
7901 	next_stateid(&sp->rs_stateid);
7902 	resp->open_stateid = sp->rs_stateid.stateid;
7903 
7904 	rfs4_dbe_unlock(sp->rs_dbe);
7905 
7906 	*cs->statusp = resp->status = NFS4_OK;
7907 	/* Update the lease */
7908 	rfs4_update_lease(sp->rs_owner->ro_client);
7909 	/* And the sequence */
7910 	rfs4_update_open_sequence(sp->rs_owner);
7911 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7912 
7913 end:
7914 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7915 	rfs4_state_rele(sp);
7916 out:
7917 	DTRACE_NFSV4_2(op__open__downgrade__done, struct compound_state *, cs,
7918 	    OPEN_DOWNGRADE4res *, resp);
7919 }
7920 
7921 static void *
7922 memstr(const void *s1, const char *s2, size_t n)
7923 {
7924 	size_t l = strlen(s2);
7925 	char *p = (char *)s1;
7926 
7927 	while (n >= l) {
7928 		if (bcmp(p, s2, l) == 0)
7929 			return (p);
7930 		p++;
7931 		n--;
7932 	}
7933 
7934 	return (NULL);
7935 }
7936 
7937 /*
7938  * The logic behind this function is detailed in the NFSv4 RFC in the
7939  * SETCLIENTID operation description under IMPLEMENTATION.  Refer to
7940  * that section for explicit guidance to server behavior for
7941  * SETCLIENTID.
7942  */
7943 void
7944 rfs4_op_setclientid(nfs_argop4 *argop, nfs_resop4 *resop,
7945     struct svc_req *req, struct compound_state *cs)
7946 {
7947 	SETCLIENTID4args *args = &argop->nfs_argop4_u.opsetclientid;
7948 	SETCLIENTID4res *res = &resop->nfs_resop4_u.opsetclientid;
7949 	rfs4_client_t *cp, *newcp, *cp_confirmed, *cp_unconfirmed;
7950 	rfs4_clntip_t *ci;
7951 	bool_t create;
7952 	char *addr, *netid;
7953 	int len;
7954 
7955 	DTRACE_NFSV4_2(op__setclientid__start, struct compound_state *, cs,
7956 	    SETCLIENTID4args *, args);
7957 retry:
7958 	newcp = cp_confirmed = cp_unconfirmed = NULL;
7959 
7960 	/*
7961 	 * Save the caller's IP address
7962 	 */
7963 	args->client.cl_addr =
7964 	    (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
7965 
7966 	/*
7967 	 * Record if it is a Solaris client that cannot handle referrals.
7968 	 */
7969 	if (memstr(args->client.id_val, "Solaris", args->client.id_len) &&
7970 	    !memstr(args->client.id_val, "+referrals", args->client.id_len)) {
7971 		/* Add a "yes, it's downrev" record */
7972 		create = TRUE;
7973 		ci = rfs4_find_clntip(args->client.cl_addr, &create);
7974 		ASSERT(ci != NULL);
7975 		rfs4_dbe_rele(ci->ri_dbe);
7976 	} else {
7977 		/* Remove any previous record */
7978 		rfs4_invalidate_clntip(args->client.cl_addr);
7979 	}
7980 
7981 	/*
7982 	 * In search of an EXISTING client matching the incoming
7983 	 * request to establish a new client identifier at the server
7984 	 */
7985 	create = TRUE;
7986 	cp = rfs4_findclient(&args->client, &create, NULL);
7987 
7988 	/* Should never happen */
7989 	ASSERT(cp != NULL);
7990 
7991 	if (cp == NULL) {
7992 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
7993 		goto out;
7994 	}
7995 
7996 	/*
7997 	 * Easiest case. Client identifier is newly created and is
7998 	 * unconfirmed.  Also note that for this case, no other
7999 	 * entries exist for the client identifier.  Nothing else to
8000 	 * check.  Just setup the response and respond.
8001 	 */
8002 	if (create) {
8003 		*cs->statusp = res->status = NFS4_OK;
8004 		res->SETCLIENTID4res_u.resok4.clientid = cp->rc_clientid;
8005 		res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8006 		    cp->rc_confirm_verf;
8007 		/* Setup callback information; CB_NULL confirmation later */
8008 		rfs4_client_setcb(cp, &args->callback, args->callback_ident);
8009 
8010 		rfs4_client_rele(cp);
8011 		goto out;
8012 	}
8013 
8014 	/*
8015 	 * An existing, confirmed client may exist but it may not have
8016 	 * been active for at least one lease period.  If so, then
8017 	 * "close" the client and create a new client identifier
8018 	 */
8019 	if (rfs4_lease_expired(cp)) {
8020 		rfs4_client_close(cp);
8021 		goto retry;
8022 	}
8023 
8024 	if (cp->rc_need_confirm == TRUE)
8025 		cp_unconfirmed = cp;
8026 	else
8027 		cp_confirmed = cp;
8028 
8029 	cp = NULL;
8030 
8031 	/*
8032 	 * We have a confirmed client, now check for an
8033 	 * unconfimred entry
8034 	 */
8035 	if (cp_confirmed) {
8036 		/* If creds don't match then client identifier is inuse */
8037 		if (!creds_ok(cp_confirmed->rc_cr_set, req, cs)) {
8038 			rfs4_cbinfo_t *cbp;
8039 			/*
8040 			 * Some one else has established this client
8041 			 * id. Try and say * who they are. We will use
8042 			 * the call back address supplied by * the
8043 			 * first client.
8044 			 */
8045 			*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8046 
8047 			addr = netid = NULL;
8048 
8049 			cbp = &cp_confirmed->rc_cbinfo;
8050 			if (cbp->cb_callback.cb_location.r_addr &&
8051 			    cbp->cb_callback.cb_location.r_netid) {
8052 				cb_client4 *cbcp = &cbp->cb_callback;
8053 
8054 				len = strlen(cbcp->cb_location.r_addr)+1;
8055 				addr = kmem_alloc(len, KM_SLEEP);
8056 				bcopy(cbcp->cb_location.r_addr, addr, len);
8057 				len = strlen(cbcp->cb_location.r_netid)+1;
8058 				netid = kmem_alloc(len, KM_SLEEP);
8059 				bcopy(cbcp->cb_location.r_netid, netid, len);
8060 			}
8061 
8062 			res->SETCLIENTID4res_u.client_using.r_addr = addr;
8063 			res->SETCLIENTID4res_u.client_using.r_netid = netid;
8064 
8065 			rfs4_client_rele(cp_confirmed);
8066 		}
8067 
8068 		/*
8069 		 * Confirmed, creds match, and verifier matches; must
8070 		 * be an update of the callback info
8071 		 */
8072 		if (cp_confirmed->rc_nfs_client.verifier ==
8073 		    args->client.verifier) {
8074 			/* Setup callback information */
8075 			rfs4_client_setcb(cp_confirmed, &args->callback,
8076 			    args->callback_ident);
8077 
8078 			/* everything okay -- move ahead */
8079 			*cs->statusp = res->status = NFS4_OK;
8080 			res->SETCLIENTID4res_u.resok4.clientid =
8081 			    cp_confirmed->rc_clientid;
8082 
8083 			/* update the confirm_verifier and return it */
8084 			rfs4_client_scv_next(cp_confirmed);
8085 			res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8086 			    cp_confirmed->rc_confirm_verf;
8087 
8088 			rfs4_client_rele(cp_confirmed);
8089 			goto out;
8090 		}
8091 
8092 		/*
8093 		 * Creds match but the verifier doesn't.  Must search
8094 		 * for an unconfirmed client that would be replaced by
8095 		 * this request.
8096 		 */
8097 		create = FALSE;
8098 		cp_unconfirmed = rfs4_findclient(&args->client, &create,
8099 		    cp_confirmed);
8100 	}
8101 
8102 	/*
8103 	 * At this point, we have taken care of the brand new client
8104 	 * struct, INUSE case, update of an existing, and confirmed
8105 	 * client struct.
8106 	 */
8107 
8108 	/*
8109 	 * check to see if things have changed while we originally
8110 	 * picked up the client struct.  If they have, then return and
8111 	 * retry the processing of this SETCLIENTID request.
8112 	 */
8113 	if (cp_unconfirmed) {
8114 		rfs4_dbe_lock(cp_unconfirmed->rc_dbe);
8115 		if (!cp_unconfirmed->rc_need_confirm) {
8116 			rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8117 			rfs4_client_rele(cp_unconfirmed);
8118 			if (cp_confirmed)
8119 				rfs4_client_rele(cp_confirmed);
8120 			goto retry;
8121 		}
8122 		/* do away with the old unconfirmed one */
8123 		rfs4_dbe_invalidate(cp_unconfirmed->rc_dbe);
8124 		rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8125 		rfs4_client_rele(cp_unconfirmed);
8126 		cp_unconfirmed = NULL;
8127 	}
8128 
8129 	/*
8130 	 * This search will temporarily hide the confirmed client
8131 	 * struct while a new client struct is created as the
8132 	 * unconfirmed one.
8133 	 */
8134 	create = TRUE;
8135 	newcp = rfs4_findclient(&args->client, &create, cp_confirmed);
8136 
8137 	ASSERT(newcp != NULL);
8138 
8139 	if (newcp == NULL) {
8140 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8141 		rfs4_client_rele(cp_confirmed);
8142 		goto out;
8143 	}
8144 
8145 	/*
8146 	 * If one was not created, then a similar request must be in
8147 	 * process so release and start over with this one
8148 	 */
8149 	if (create != TRUE) {
8150 		rfs4_client_rele(newcp);
8151 		if (cp_confirmed)
8152 			rfs4_client_rele(cp_confirmed);
8153 		goto retry;
8154 	}
8155 
8156 	*cs->statusp = res->status = NFS4_OK;
8157 	res->SETCLIENTID4res_u.resok4.clientid = newcp->rc_clientid;
8158 	res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8159 	    newcp->rc_confirm_verf;
8160 	/* Setup callback information; CB_NULL confirmation later */
8161 	rfs4_client_setcb(newcp, &args->callback, args->callback_ident);
8162 
8163 	newcp->rc_cp_confirmed = cp_confirmed;
8164 
8165 	rfs4_client_rele(newcp);
8166 
8167 out:
8168 	DTRACE_NFSV4_2(op__setclientid__done, struct compound_state *, cs,
8169 	    SETCLIENTID4res *, res);
8170 }
8171 
8172 /*ARGSUSED*/
8173 void
8174 rfs4_op_setclientid_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
8175     struct svc_req *req, struct compound_state *cs)
8176 {
8177 	SETCLIENTID_CONFIRM4args *args =
8178 	    &argop->nfs_argop4_u.opsetclientid_confirm;
8179 	SETCLIENTID_CONFIRM4res *res =
8180 	    &resop->nfs_resop4_u.opsetclientid_confirm;
8181 	rfs4_client_t *cp, *cptoclose = NULL;
8182 
8183 	DTRACE_NFSV4_2(op__setclientid__confirm__start,
8184 	    struct compound_state *, cs,
8185 	    SETCLIENTID_CONFIRM4args *, args);
8186 
8187 	*cs->statusp = res->status = NFS4_OK;
8188 
8189 	cp = rfs4_findclient_by_id(args->clientid, TRUE);
8190 
8191 	if (cp == NULL) {
8192 		*cs->statusp = res->status =
8193 		    rfs4_check_clientid(&args->clientid, 1);
8194 		goto out;
8195 	}
8196 
8197 	if (!creds_ok(cp, req, cs)) {
8198 		*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8199 		rfs4_client_rele(cp);
8200 		goto out;
8201 	}
8202 
8203 	/* If the verifier doesn't match, the record doesn't match */
8204 	if (cp->rc_confirm_verf != args->setclientid_confirm) {
8205 		*cs->statusp = res->status = NFS4ERR_STALE_CLIENTID;
8206 		rfs4_client_rele(cp);
8207 		goto out;
8208 	}
8209 
8210 	rfs4_dbe_lock(cp->rc_dbe);
8211 	cp->rc_need_confirm = FALSE;
8212 	if (cp->rc_cp_confirmed) {
8213 		cptoclose = cp->rc_cp_confirmed;
8214 		cptoclose->rc_ss_remove = 1;
8215 		cp->rc_cp_confirmed = NULL;
8216 	}
8217 
8218 	/*
8219 	 * Update the client's associated server instance, if it's changed
8220 	 * since the client was created.
8221 	 */
8222 	if (rfs4_servinst(cp) != rfs4_cur_servinst)
8223 		rfs4_servinst_assign(cp, rfs4_cur_servinst);
8224 
8225 	/*
8226 	 * Record clientid in stable storage.
8227 	 * Must be done after server instance has been assigned.
8228 	 */
8229 	rfs4_ss_clid(cp);
8230 
8231 	rfs4_dbe_unlock(cp->rc_dbe);
8232 
8233 	if (cptoclose)
8234 		/* don't need to rele, client_close does it */
8235 		rfs4_client_close(cptoclose);
8236 
8237 	/* If needed, initiate CB_NULL call for callback path */
8238 	rfs4_deleg_cb_check(cp);
8239 	rfs4_update_lease(cp);
8240 
8241 	/*
8242 	 * Check to see if client can perform reclaims
8243 	 */
8244 	rfs4_ss_chkclid(cp);
8245 
8246 	rfs4_client_rele(cp);
8247 
8248 out:
8249 	DTRACE_NFSV4_2(op__setclientid__confirm__done,
8250 	    struct compound_state *, cs,
8251 	    SETCLIENTID_CONFIRM4 *, res);
8252 }
8253 
8254 
8255 /*ARGSUSED*/
8256 void
8257 rfs4_op_close(nfs_argop4 *argop, nfs_resop4 *resop,
8258     struct svc_req *req, struct compound_state *cs)
8259 {
8260 	CLOSE4args *args = &argop->nfs_argop4_u.opclose;
8261 	CLOSE4res *resp = &resop->nfs_resop4_u.opclose;
8262 	rfs4_state_t *sp;
8263 	nfsstat4 status;
8264 
8265 	DTRACE_NFSV4_2(op__close__start, struct compound_state *, cs,
8266 	    CLOSE4args *, args);
8267 
8268 	if (cs->vp == NULL) {
8269 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8270 		goto out;
8271 	}
8272 
8273 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_INVALID);
8274 	if (status != NFS4_OK) {
8275 		*cs->statusp = resp->status = status;
8276 		goto out;
8277 	}
8278 
8279 	/* Ensure specified filehandle matches */
8280 	if (cs->vp != sp->rs_finfo->rf_vp) {
8281 		rfs4_state_rele(sp);
8282 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8283 		goto out;
8284 	}
8285 
8286 	/* hold off other access to open_owner while we tinker */
8287 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
8288 
8289 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
8290 	case NFS4_CHECK_STATEID_OKAY:
8291 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8292 		    resop) != NFS4_CHKSEQ_OKAY) {
8293 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8294 			goto end;
8295 		}
8296 		break;
8297 	case NFS4_CHECK_STATEID_OLD:
8298 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8299 		goto end;
8300 	case NFS4_CHECK_STATEID_BAD:
8301 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8302 		goto end;
8303 	case NFS4_CHECK_STATEID_EXPIRED:
8304 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8305 		goto end;
8306 	case NFS4_CHECK_STATEID_CLOSED:
8307 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8308 		goto end;
8309 	case NFS4_CHECK_STATEID_UNCONFIRMED:
8310 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8311 		goto end;
8312 	case NFS4_CHECK_STATEID_REPLAY:
8313 		/* Check the sequence id for the open owner */
8314 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8315 		    resop)) {
8316 		case NFS4_CHKSEQ_OKAY:
8317 			/*
8318 			 * This is replayed stateid; if seqid matches
8319 			 * next expected, then client is using wrong seqid.
8320 			 */
8321 			/* FALL THROUGH */
8322 		case NFS4_CHKSEQ_BAD:
8323 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8324 			goto end;
8325 		case NFS4_CHKSEQ_REPLAY:
8326 			/*
8327 			 * Note this case is the duplicate case so
8328 			 * resp->status is already set.
8329 			 */
8330 			*cs->statusp = resp->status;
8331 			rfs4_update_lease(sp->rs_owner->ro_client);
8332 			goto end;
8333 		}
8334 		break;
8335 	default:
8336 		ASSERT(FALSE);
8337 		break;
8338 	}
8339 
8340 	rfs4_dbe_lock(sp->rs_dbe);
8341 
8342 	/* Update the stateid. */
8343 	next_stateid(&sp->rs_stateid);
8344 	resp->open_stateid = sp->rs_stateid.stateid;
8345 
8346 	rfs4_dbe_unlock(sp->rs_dbe);
8347 
8348 	rfs4_update_lease(sp->rs_owner->ro_client);
8349 	rfs4_update_open_sequence(sp->rs_owner);
8350 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8351 
8352 	rfs4_state_close(sp, FALSE, FALSE, cs->cr);
8353 
8354 	*cs->statusp = resp->status = status;
8355 
8356 end:
8357 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
8358 	rfs4_state_rele(sp);
8359 out:
8360 	DTRACE_NFSV4_2(op__close__done, struct compound_state *, cs,
8361 	    CLOSE4res *, resp);
8362 }
8363 
8364 /*
8365  * Manage the counts on the file struct and close all file locks
8366  */
8367 /*ARGSUSED*/
8368 void
8369 rfs4_release_share_lock_state(rfs4_state_t *sp, cred_t *cr,
8370     bool_t close_of_client)
8371 {
8372 	rfs4_file_t *fp = sp->rs_finfo;
8373 	rfs4_lo_state_t *lsp;
8374 	int fflags = 0;
8375 
8376 	/*
8377 	 * If this call is part of the larger closing down of client
8378 	 * state then it is just easier to release all locks
8379 	 * associated with this client instead of going through each
8380 	 * individual file and cleaning locks there.
8381 	 */
8382 	if (close_of_client) {
8383 		if (sp->rs_owner->ro_client->rc_unlksys_completed == FALSE &&
8384 		    !list_is_empty(&sp->rs_lostatelist) &&
8385 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID) {
8386 			/* Is the PxFS kernel module loaded? */
8387 			if (lm_remove_file_locks != NULL) {
8388 				int new_sysid;
8389 
8390 				/* Encode the cluster nodeid in new sysid */
8391 				new_sysid = sp->rs_owner->ro_client->rc_sysidt;
8392 				lm_set_nlmid_flk(&new_sysid);
8393 
8394 				/*
8395 				 * This PxFS routine removes file locks for a
8396 				 * client over all nodes of a cluster.
8397 				 */
8398 				NFS4_DEBUG(rfs4_debug, (CE_NOTE,
8399 				    "lm_remove_file_locks(sysid=0x%x)\n",
8400 				    new_sysid));
8401 				(*lm_remove_file_locks)(new_sysid);
8402 			} else {
8403 				struct flock64 flk;
8404 
8405 				/* Release all locks for this client */
8406 				flk.l_type = F_UNLKSYS;
8407 				flk.l_whence = 0;
8408 				flk.l_start = 0;
8409 				flk.l_len = 0;
8410 				flk.l_sysid =
8411 				    sp->rs_owner->ro_client->rc_sysidt;
8412 				flk.l_pid = 0;
8413 				(void) VOP_FRLOCK(sp->rs_finfo->rf_vp, F_SETLK,
8414 				    &flk, F_REMOTELOCK | FREAD | FWRITE,
8415 				    (u_offset_t)0, NULL, CRED(), NULL);
8416 			}
8417 
8418 			sp->rs_owner->ro_client->rc_unlksys_completed = TRUE;
8419 		}
8420 	}
8421 
8422 	/*
8423 	 * Release all locks on this file by this lock owner or at
8424 	 * least mark the locks as having been released
8425 	 */
8426 	for (lsp = list_head(&sp->rs_lostatelist); lsp != NULL;
8427 	    lsp = list_next(&sp->rs_lostatelist, lsp)) {
8428 		lsp->rls_locks_cleaned = TRUE;
8429 
8430 		/* Was this already taken care of above? */
8431 		if (!close_of_client &&
8432 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8433 			(void) cleanlocks(sp->rs_finfo->rf_vp,
8434 			    lsp->rls_locker->rl_pid,
8435 			    lsp->rls_locker->rl_client->rc_sysidt);
8436 	}
8437 
8438 	/*
8439 	 * Release any shrlocks associated with this open state ID.
8440 	 * This must be done before the rfs4_state gets marked closed.
8441 	 */
8442 	if (sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8443 		(void) rfs4_unshare(sp);
8444 
8445 	if (sp->rs_open_access) {
8446 		rfs4_dbe_lock(fp->rf_dbe);
8447 
8448 		/*
8449 		 * Decrement the count for each access and deny bit that this
8450 		 * state has contributed to the file.
8451 		 * If the file counts go to zero
8452 		 * clear the appropriate bit in the appropriate mask.
8453 		 */
8454 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) {
8455 			fp->rf_access_read--;
8456 			fflags |= FREAD;
8457 			if (fp->rf_access_read == 0)
8458 				fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
8459 		}
8460 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) {
8461 			fp->rf_access_write--;
8462 			fflags |= FWRITE;
8463 			if (fp->rf_access_write == 0)
8464 				fp->rf_share_access &=
8465 				    ~OPEN4_SHARE_ACCESS_WRITE;
8466 		}
8467 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_READ) {
8468 			fp->rf_deny_read--;
8469 			if (fp->rf_deny_read == 0)
8470 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
8471 		}
8472 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) {
8473 			fp->rf_deny_write--;
8474 			if (fp->rf_deny_write == 0)
8475 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
8476 		}
8477 
8478 		(void) VOP_CLOSE(fp->rf_vp, fflags, 1, (offset_t)0, cr, NULL);
8479 
8480 		rfs4_dbe_unlock(fp->rf_dbe);
8481 
8482 		sp->rs_open_access = 0;
8483 		sp->rs_open_deny = 0;
8484 	}
8485 }
8486 
8487 /*
8488  * lock_denied: Fill in a LOCK4deneid structure given an flock64 structure.
8489  */
8490 static nfsstat4
8491 lock_denied(LOCK4denied *dp, struct flock64 *flk)
8492 {
8493 	rfs4_lockowner_t *lo;
8494 	rfs4_client_t *cp;
8495 	uint32_t len;
8496 
8497 	lo = rfs4_findlockowner_by_pid(flk->l_pid);
8498 	if (lo != NULL) {
8499 		cp = lo->rl_client;
8500 		if (rfs4_lease_expired(cp)) {
8501 			rfs4_lockowner_rele(lo);
8502 			rfs4_dbe_hold(cp->rc_dbe);
8503 			rfs4_client_close(cp);
8504 			return (NFS4ERR_EXPIRED);
8505 		}
8506 		dp->owner.clientid = lo->rl_owner.clientid;
8507 		len = lo->rl_owner.owner_len;
8508 		dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8509 		bcopy(lo->rl_owner.owner_val, dp->owner.owner_val, len);
8510 		dp->owner.owner_len = len;
8511 		rfs4_lockowner_rele(lo);
8512 		goto finish;
8513 	}
8514 
8515 	/*
8516 	 * Its not a NFS4 lock. We take advantage that the upper 32 bits
8517 	 * of the client id contain the boot time for a NFS4 lock. So we
8518 	 * fabricate and identity by setting clientid to the sysid, and
8519 	 * the lock owner to the pid.
8520 	 */
8521 	dp->owner.clientid = flk->l_sysid;
8522 	len = sizeof (pid_t);
8523 	dp->owner.owner_len = len;
8524 	dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8525 	bcopy(&flk->l_pid, dp->owner.owner_val, len);
8526 finish:
8527 	dp->offset = flk->l_start;
8528 	dp->length = flk->l_len;
8529 
8530 	if (flk->l_type == F_RDLCK)
8531 		dp->locktype = READ_LT;
8532 	else if (flk->l_type == F_WRLCK)
8533 		dp->locktype = WRITE_LT;
8534 	else
8535 		return (NFS4ERR_INVAL);	/* no mapping from POSIX ltype to v4 */
8536 
8537 	return (NFS4_OK);
8538 }
8539 
8540 /*
8541  * The NFSv4.0 LOCK operation does not support the blocking lock (at the
8542  * NFSv4.0 protocol level) so the client needs to resend the LOCK request in a
8543  * case the lock is denied by the NFSv4.0 server.  NFSv4.0 clients are prepared
8544  * for that (obviously); they are sending the LOCK requests with some delays
8545  * between the attempts.  See nfs4frlock() and nfs4_block_and_wait() for the
8546  * locking and delay implementation at the client side.
8547  *
8548  * To make the life of the clients easier, the NFSv4.0 server tries to do some
8549  * fast retries on its own (the for loop below) in a hope the lock will be
8550  * available soon.  And if not, the client won't need to resend the LOCK
8551  * requests so fast to check the lock availability.  This basically saves some
8552  * network traffic and tries to make sure the client gets the lock ASAP.
8553  */
8554 static int
8555 setlock(vnode_t *vp, struct flock64 *flock, int flag, cred_t *cred)
8556 {
8557 	int error;
8558 	struct flock64 flk;
8559 	int i;
8560 	clock_t delaytime;
8561 	int cmd;
8562 	int spin_cnt = 0;
8563 
8564 	cmd = nbl_need_check(vp) ? F_SETLK_NBMAND : F_SETLK;
8565 retry:
8566 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
8567 
8568 	for (i = 0; i < rfs4_maxlock_tries; i++) {
8569 		LOCK_PRINT(rfs4_debug, "setlock", cmd, flock);
8570 		error = VOP_FRLOCK(vp, cmd,
8571 		    flock, flag, (u_offset_t)0, NULL, cred, NULL);
8572 
8573 		if (error != EAGAIN && error != EACCES)
8574 			break;
8575 
8576 		if (i < rfs4_maxlock_tries - 1) {
8577 			delay(delaytime);
8578 			delaytime *= 2;
8579 		}
8580 	}
8581 
8582 	if (error == EAGAIN || error == EACCES) {
8583 		/* Get the owner of the lock */
8584 		flk = *flock;
8585 		LOCK_PRINT(rfs4_debug, "setlock", F_GETLK, &flk);
8586 		if (VOP_FRLOCK(vp, F_GETLK, &flk, flag, 0, NULL, cred,
8587 		    NULL) == 0) {
8588 			/*
8589 			 * There's a race inherent in the current VOP_FRLOCK
8590 			 * design where:
8591 			 * a: "other guy" takes a lock that conflicts with a
8592 			 * lock we want
8593 			 * b: we attempt to take our lock (non-blocking) and
8594 			 * the attempt fails.
8595 			 * c: "other guy" releases the conflicting lock
8596 			 * d: we ask what lock conflicts with the lock we want,
8597 			 * getting F_UNLCK (no lock blocks us)
8598 			 *
8599 			 * If we retry the non-blocking lock attempt in this
8600 			 * case (restart at step 'b') there's some possibility
8601 			 * that many such attempts might fail.  However a test
8602 			 * designed to actually provoke this race shows that
8603 			 * the vast majority of cases require no retry, and
8604 			 * only a few took as many as three retries.  Here's
8605 			 * the test outcome:
8606 			 *
8607 			 *	   number of retries    how many times we needed
8608 			 *				that many retries
8609 			 *	   0			79461
8610 			 *	   1			  862
8611 			 *	   2			   49
8612 			 *	   3			    5
8613 			 *
8614 			 * Given those empirical results, we arbitrarily limit
8615 			 * the retry count to ten.
8616 			 *
8617 			 * If we actually make to ten retries and give up,
8618 			 * nothing catastrophic happens, but we're unable to
8619 			 * return the information about the conflicting lock to
8620 			 * the NFS client.  That's an acceptable trade off vs.
8621 			 * letting this retry loop run forever.
8622 			 */
8623 			if (flk.l_type == F_UNLCK) {
8624 				if (spin_cnt++ < 10) {
8625 					/* No longer locked, retry */
8626 					goto retry;
8627 				}
8628 			} else {
8629 				*flock = flk;
8630 				LOCK_PRINT(rfs4_debug, "setlock(blocking lock)",
8631 				    F_GETLK, &flk);
8632 			}
8633 		}
8634 	}
8635 
8636 	return (error);
8637 }
8638 
8639 /*ARGSUSED*/
8640 static nfsstat4
8641 rfs4_do_lock(rfs4_lo_state_t *lsp, nfs_lock_type4 locktype,
8642     offset4 offset, length4 length, cred_t *cred, nfs_resop4 *resop)
8643 {
8644 	nfsstat4 status;
8645 	rfs4_lockowner_t *lo = lsp->rls_locker;
8646 	rfs4_state_t *sp = lsp->rls_state;
8647 	struct flock64 flock;
8648 	int16_t ltype;
8649 	int flag;
8650 	int error;
8651 	sysid_t sysid;
8652 	LOCK4res *lres;
8653 	vnode_t *vp;
8654 
8655 	if (rfs4_lease_expired(lo->rl_client)) {
8656 		return (NFS4ERR_EXPIRED);
8657 	}
8658 
8659 	if ((status = rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
8660 		return (status);
8661 
8662 	/* Check for zero length. To lock to end of file use all ones for V4 */
8663 	if (length == 0)
8664 		return (NFS4ERR_INVAL);
8665 	else if (length == (length4)(~0))
8666 		length = 0;		/* Posix to end of file  */
8667 
8668 retry:
8669 	rfs4_dbe_lock(sp->rs_dbe);
8670 	if (sp->rs_closed == TRUE) {
8671 		rfs4_dbe_unlock(sp->rs_dbe);
8672 		return (NFS4ERR_OLD_STATEID);
8673 	}
8674 
8675 	if (resop->resop != OP_LOCKU) {
8676 		switch (locktype) {
8677 		case READ_LT:
8678 		case READW_LT:
8679 			if ((sp->rs_share_access
8680 			    & OPEN4_SHARE_ACCESS_READ) == 0) {
8681 				rfs4_dbe_unlock(sp->rs_dbe);
8682 
8683 				return (NFS4ERR_OPENMODE);
8684 			}
8685 			ltype = F_RDLCK;
8686 			break;
8687 		case WRITE_LT:
8688 		case WRITEW_LT:
8689 			if ((sp->rs_share_access
8690 			    & OPEN4_SHARE_ACCESS_WRITE) == 0) {
8691 				rfs4_dbe_unlock(sp->rs_dbe);
8692 
8693 				return (NFS4ERR_OPENMODE);
8694 			}
8695 			ltype = F_WRLCK;
8696 			break;
8697 		}
8698 	} else
8699 		ltype = F_UNLCK;
8700 
8701 	flock.l_type = ltype;
8702 	flock.l_whence = 0;		/* SEEK_SET */
8703 	flock.l_start = offset;
8704 	flock.l_len = length;
8705 	flock.l_sysid = sysid;
8706 	flock.l_pid = lsp->rls_locker->rl_pid;
8707 
8708 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
8709 	if (flock.l_len < 0 || flock.l_start < 0) {
8710 		rfs4_dbe_unlock(sp->rs_dbe);
8711 		return (NFS4ERR_INVAL);
8712 	}
8713 
8714 	/*
8715 	 * N.B. FREAD has the same value as OPEN4_SHARE_ACCESS_READ and
8716 	 * FWRITE has the same value as OPEN4_SHARE_ACCESS_WRITE.
8717 	 */
8718 	flag = (int)sp->rs_share_access | F_REMOTELOCK;
8719 
8720 	vp = sp->rs_finfo->rf_vp;
8721 	VN_HOLD(vp);
8722 
8723 	/*
8724 	 * We need to unlock sp before we call the underlying filesystem to
8725 	 * acquire the file lock.
8726 	 */
8727 	rfs4_dbe_unlock(sp->rs_dbe);
8728 
8729 	error = setlock(vp, &flock, flag, cred);
8730 
8731 	/*
8732 	 * Make sure the file is still open.  In a case the file was closed in
8733 	 * the meantime, clean the lock we acquired using the setlock() call
8734 	 * above, and return the appropriate error.
8735 	 */
8736 	rfs4_dbe_lock(sp->rs_dbe);
8737 	if (sp->rs_closed == TRUE) {
8738 		cleanlocks(vp, lsp->rls_locker->rl_pid, sysid);
8739 		rfs4_dbe_unlock(sp->rs_dbe);
8740 
8741 		VN_RELE(vp);
8742 
8743 		return (NFS4ERR_OLD_STATEID);
8744 	}
8745 	rfs4_dbe_unlock(sp->rs_dbe);
8746 
8747 	VN_RELE(vp);
8748 
8749 	if (error == 0) {
8750 		rfs4_dbe_lock(lsp->rls_dbe);
8751 		next_stateid(&lsp->rls_lockid);
8752 		rfs4_dbe_unlock(lsp->rls_dbe);
8753 	}
8754 
8755 	/*
8756 	 * N.B. We map error values to nfsv4 errors. This is differrent
8757 	 * than puterrno4 routine.
8758 	 */
8759 	switch (error) {
8760 	case 0:
8761 		status = NFS4_OK;
8762 		break;
8763 	case EAGAIN:
8764 	case EACCES:		/* Old value */
8765 		/* Can only get here if op is OP_LOCK */
8766 		ASSERT(resop->resop == OP_LOCK);
8767 		lres = &resop->nfs_resop4_u.oplock;
8768 		status = NFS4ERR_DENIED;
8769 		if (lock_denied(&lres->LOCK4res_u.denied, &flock)
8770 		    == NFS4ERR_EXPIRED)
8771 			goto retry;
8772 		break;
8773 	case ENOLCK:
8774 		status = NFS4ERR_DELAY;
8775 		break;
8776 	case EOVERFLOW:
8777 		status = NFS4ERR_INVAL;
8778 		break;
8779 	case EINVAL:
8780 		status = NFS4ERR_NOTSUPP;
8781 		break;
8782 	default:
8783 		status = NFS4ERR_SERVERFAULT;
8784 		break;
8785 	}
8786 
8787 	return (status);
8788 }
8789 
8790 /*ARGSUSED*/
8791 void
8792 rfs4_op_lock(nfs_argop4 *argop, nfs_resop4 *resop,
8793     struct svc_req *req, struct compound_state *cs)
8794 {
8795 	LOCK4args *args = &argop->nfs_argop4_u.oplock;
8796 	LOCK4res *resp = &resop->nfs_resop4_u.oplock;
8797 	nfsstat4 status;
8798 	stateid4 *stateid;
8799 	rfs4_lockowner_t *lo;
8800 	rfs4_client_t *cp;
8801 	rfs4_state_t *sp = NULL;
8802 	rfs4_lo_state_t *lsp = NULL;
8803 	bool_t ls_sw_held = FALSE;
8804 	bool_t create = TRUE;
8805 	bool_t lcreate = TRUE;
8806 	bool_t dup_lock = FALSE;
8807 	int rc;
8808 
8809 	DTRACE_NFSV4_2(op__lock__start, struct compound_state *, cs,
8810 	    LOCK4args *, args);
8811 
8812 	if (cs->vp == NULL) {
8813 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8814 		DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8815 		    cs, LOCK4res *, resp);
8816 		return;
8817 	}
8818 
8819 	if (args->locker.new_lock_owner) {
8820 		/* Create a new lockowner for this instance */
8821 		open_to_lock_owner4 *olo = &args->locker.locker4_u.open_owner;
8822 
8823 		NFS4_DEBUG(rfs4_debug, (CE_NOTE, "Creating new lock owner"));
8824 
8825 		stateid = &olo->open_stateid;
8826 		status = rfs4_get_state(stateid, &sp, RFS4_DBS_VALID);
8827 		if (status != NFS4_OK) {
8828 			NFS4_DEBUG(rfs4_debug,
8829 			    (CE_NOTE, "Get state failed in lock %d", status));
8830 			*cs->statusp = resp->status = status;
8831 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8832 			    cs, LOCK4res *, resp);
8833 			return;
8834 		}
8835 
8836 		/* Ensure specified filehandle matches */
8837 		if (cs->vp != sp->rs_finfo->rf_vp) {
8838 			rfs4_state_rele(sp);
8839 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8840 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8841 			    cs, LOCK4res *, resp);
8842 			return;
8843 		}
8844 
8845 		/* hold off other access to open_owner while we tinker */
8846 		rfs4_sw_enter(&sp->rs_owner->ro_sw);
8847 
8848 		switch (rc = rfs4_check_stateid_seqid(sp, stateid)) {
8849 		case NFS4_CHECK_STATEID_OLD:
8850 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8851 			goto end;
8852 		case NFS4_CHECK_STATEID_BAD:
8853 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8854 			goto end;
8855 		case NFS4_CHECK_STATEID_EXPIRED:
8856 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8857 			goto end;
8858 		case NFS4_CHECK_STATEID_UNCONFIRMED:
8859 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8860 			goto end;
8861 		case NFS4_CHECK_STATEID_CLOSED:
8862 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8863 			goto end;
8864 		case NFS4_CHECK_STATEID_OKAY:
8865 		case NFS4_CHECK_STATEID_REPLAY:
8866 			switch (rfs4_check_olo_seqid(olo->open_seqid,
8867 			    sp->rs_owner, resop)) {
8868 			case NFS4_CHKSEQ_OKAY:
8869 				if (rc == NFS4_CHECK_STATEID_OKAY)
8870 					break;
8871 				/*
8872 				 * This is replayed stateid; if seqid
8873 				 * matches next expected, then client
8874 				 * is using wrong seqid.
8875 				 */
8876 				/* FALLTHROUGH */
8877 			case NFS4_CHKSEQ_BAD:
8878 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8879 				goto end;
8880 			case NFS4_CHKSEQ_REPLAY:
8881 				/* This is a duplicate LOCK request */
8882 				dup_lock = TRUE;
8883 
8884 				/*
8885 				 * For a duplicate we do not want to
8886 				 * create a new lockowner as it should
8887 				 * already exist.
8888 				 * Turn off the lockowner create flag.
8889 				 */
8890 				lcreate = FALSE;
8891 			}
8892 			break;
8893 		}
8894 
8895 		lo = rfs4_findlockowner(&olo->lock_owner, &lcreate);
8896 		if (lo == NULL) {
8897 			NFS4_DEBUG(rfs4_debug,
8898 			    (CE_NOTE, "rfs4_op_lock: no lock owner"));
8899 			*cs->statusp = resp->status = NFS4ERR_RESOURCE;
8900 			goto end;
8901 		}
8902 
8903 		lsp = rfs4_findlo_state_by_owner(lo, sp, &create);
8904 		if (lsp == NULL) {
8905 			rfs4_update_lease(sp->rs_owner->ro_client);
8906 			/*
8907 			 * Only update theh open_seqid if this is not
8908 			 * a duplicate request
8909 			 */
8910 			if (dup_lock == FALSE) {
8911 				rfs4_update_open_sequence(sp->rs_owner);
8912 			}
8913 
8914 			NFS4_DEBUG(rfs4_debug,
8915 			    (CE_NOTE, "rfs4_op_lock: no state"));
8916 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
8917 			rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8918 			rfs4_lockowner_rele(lo);
8919 			goto end;
8920 		}
8921 
8922 		/*
8923 		 * This is the new_lock_owner branch and the client is
8924 		 * supposed to be associating a new lock_owner with
8925 		 * the open file at this point.  If we find that a
8926 		 * lock_owner/state association already exists and a
8927 		 * successful LOCK request was returned to the client,
8928 		 * an error is returned to the client since this is
8929 		 * not appropriate.  The client should be using the
8930 		 * existing lock_owner branch.
8931 		 */
8932 		if (dup_lock == FALSE && create == FALSE) {
8933 			if (lsp->rls_lock_completed == TRUE) {
8934 				*cs->statusp =
8935 				    resp->status = NFS4ERR_BAD_SEQID;
8936 				rfs4_lockowner_rele(lo);
8937 				goto end;
8938 			}
8939 		}
8940 
8941 		rfs4_update_lease(sp->rs_owner->ro_client);
8942 
8943 		/*
8944 		 * Only update theh open_seqid if this is not
8945 		 * a duplicate request
8946 		 */
8947 		if (dup_lock == FALSE) {
8948 			rfs4_update_open_sequence(sp->rs_owner);
8949 		}
8950 
8951 		/*
8952 		 * If this is a duplicate lock request, just copy the
8953 		 * previously saved reply and return.
8954 		 */
8955 		if (dup_lock == TRUE) {
8956 			/* verify that lock_seqid's match */
8957 			if (lsp->rls_seqid != olo->lock_seqid) {
8958 				NFS4_DEBUG(rfs4_debug,
8959 				    (CE_NOTE, "rfs4_op_lock: Dup-Lock seqid bad"
8960 				    "lsp->seqid=%d old->seqid=%d",
8961 				    lsp->rls_seqid, olo->lock_seqid));
8962 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8963 			} else {
8964 				rfs4_copy_reply(resop, &lsp->rls_reply);
8965 				/*
8966 				 * Make sure to copy the just
8967 				 * retrieved reply status into the
8968 				 * overall compound status
8969 				 */
8970 				*cs->statusp = resp->status;
8971 			}
8972 			rfs4_lockowner_rele(lo);
8973 			goto end;
8974 		}
8975 
8976 		rfs4_dbe_lock(lsp->rls_dbe);
8977 
8978 		/* Make sure to update the lock sequence id */
8979 		lsp->rls_seqid = olo->lock_seqid;
8980 
8981 		NFS4_DEBUG(rfs4_debug,
8982 		    (CE_NOTE, "Lock seqid established as %d", lsp->rls_seqid));
8983 
8984 		/*
8985 		 * This is used to signify the newly created lockowner
8986 		 * stateid and its sequence number.  The checks for
8987 		 * sequence number and increment don't occur on the
8988 		 * very first lock request for a lockowner.
8989 		 */
8990 		lsp->rls_skip_seqid_check = TRUE;
8991 
8992 		/* hold off other access to lsp while we tinker */
8993 		rfs4_sw_enter(&lsp->rls_sw);
8994 		ls_sw_held = TRUE;
8995 
8996 		rfs4_dbe_unlock(lsp->rls_dbe);
8997 
8998 		rfs4_lockowner_rele(lo);
8999 	} else {
9000 		stateid = &args->locker.locker4_u.lock_owner.lock_stateid;
9001 		/* get lsp and hold the lock on the underlying file struct */
9002 		if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE))
9003 		    != NFS4_OK) {
9004 			*cs->statusp = resp->status = status;
9005 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
9006 			    cs, LOCK4res *, resp);
9007 			return;
9008 		}
9009 		create = FALSE;	/* We didn't create lsp */
9010 
9011 		/* Ensure specified filehandle matches */
9012 		if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9013 			rfs4_lo_state_rele(lsp, TRUE);
9014 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9015 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
9016 			    cs, LOCK4res *, resp);
9017 			return;
9018 		}
9019 
9020 		/* hold off other access to lsp while we tinker */
9021 		rfs4_sw_enter(&lsp->rls_sw);
9022 		ls_sw_held = TRUE;
9023 
9024 		switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9025 		/*
9026 		 * The stateid looks like it was okay (expected to be
9027 		 * the next one)
9028 		 */
9029 		case NFS4_CHECK_STATEID_OKAY:
9030 			/*
9031 			 * The sequence id is now checked.  Determine
9032 			 * if this is a replay or if it is in the
9033 			 * expected (next) sequence.  In the case of a
9034 			 * replay, there are two replay conditions
9035 			 * that may occur.  The first is the normal
9036 			 * condition where a LOCK is done with a
9037 			 * NFS4_OK response and the stateid is
9038 			 * updated.  That case is handled below when
9039 			 * the stateid is identified as a REPLAY.  The
9040 			 * second is the case where an error is
9041 			 * returned, like NFS4ERR_DENIED, and the
9042 			 * sequence number is updated but the stateid
9043 			 * is not updated.  This second case is dealt
9044 			 * with here.  So it may seem odd that the
9045 			 * stateid is okay but the sequence id is a
9046 			 * replay but it is okay.
9047 			 */
9048 			switch (rfs4_check_lock_seqid(
9049 			    args->locker.locker4_u.lock_owner.lock_seqid,
9050 			    lsp, resop)) {
9051 			case NFS4_CHKSEQ_REPLAY:
9052 				if (resp->status != NFS4_OK) {
9053 					/*
9054 					 * Here is our replay and need
9055 					 * to verify that the last
9056 					 * response was an error.
9057 					 */
9058 					*cs->statusp = resp->status;
9059 					goto end;
9060 				}
9061 				/*
9062 				 * This is done since the sequence id
9063 				 * looked like a replay but it didn't
9064 				 * pass our check so a BAD_SEQID is
9065 				 * returned as a result.
9066 				 */
9067 				/*FALLTHROUGH*/
9068 			case NFS4_CHKSEQ_BAD:
9069 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9070 				goto end;
9071 			case NFS4_CHKSEQ_OKAY:
9072 				/* Everything looks okay move ahead */
9073 				break;
9074 			}
9075 			break;
9076 		case NFS4_CHECK_STATEID_OLD:
9077 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9078 			goto end;
9079 		case NFS4_CHECK_STATEID_BAD:
9080 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9081 			goto end;
9082 		case NFS4_CHECK_STATEID_EXPIRED:
9083 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9084 			goto end;
9085 		case NFS4_CHECK_STATEID_CLOSED:
9086 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9087 			goto end;
9088 		case NFS4_CHECK_STATEID_REPLAY:
9089 			switch (rfs4_check_lock_seqid(
9090 			    args->locker.locker4_u.lock_owner.lock_seqid,
9091 			    lsp, resop)) {
9092 			case NFS4_CHKSEQ_OKAY:
9093 				/*
9094 				 * This is a replayed stateid; if
9095 				 * seqid matches the next expected,
9096 				 * then client is using wrong seqid.
9097 				 */
9098 			case NFS4_CHKSEQ_BAD:
9099 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9100 				goto end;
9101 			case NFS4_CHKSEQ_REPLAY:
9102 				rfs4_update_lease(lsp->rls_locker->rl_client);
9103 				*cs->statusp = status = resp->status;
9104 				goto end;
9105 			}
9106 			break;
9107 		default:
9108 			ASSERT(FALSE);
9109 			break;
9110 		}
9111 
9112 		rfs4_update_lock_sequence(lsp);
9113 		rfs4_update_lease(lsp->rls_locker->rl_client);
9114 	}
9115 
9116 	/*
9117 	 * NFS4 only allows locking on regular files, so
9118 	 * verify type of object.
9119 	 */
9120 	if (cs->vp->v_type != VREG) {
9121 		if (cs->vp->v_type == VDIR)
9122 			status = NFS4ERR_ISDIR;
9123 		else
9124 			status = NFS4ERR_INVAL;
9125 		goto out;
9126 	}
9127 
9128 	cp = lsp->rls_state->rs_owner->ro_client;
9129 
9130 	if (rfs4_clnt_in_grace(cp) && !args->reclaim) {
9131 		status = NFS4ERR_GRACE;
9132 		goto out;
9133 	}
9134 
9135 	if (rfs4_clnt_in_grace(cp) && args->reclaim && !cp->rc_can_reclaim) {
9136 		status = NFS4ERR_NO_GRACE;
9137 		goto out;
9138 	}
9139 
9140 	if (!rfs4_clnt_in_grace(cp) && args->reclaim) {
9141 		status = NFS4ERR_NO_GRACE;
9142 		goto out;
9143 	}
9144 
9145 	if (lsp->rls_state->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE)
9146 		cs->deleg = TRUE;
9147 
9148 	status = rfs4_do_lock(lsp, args->locktype,
9149 	    args->offset, args->length, cs->cr, resop);
9150 
9151 out:
9152 	lsp->rls_skip_seqid_check = FALSE;
9153 
9154 	*cs->statusp = resp->status = status;
9155 
9156 	if (status == NFS4_OK) {
9157 		resp->LOCK4res_u.lock_stateid = lsp->rls_lockid.stateid;
9158 		lsp->rls_lock_completed = TRUE;
9159 	}
9160 	/*
9161 	 * Only update the "OPEN" response here if this was a new
9162 	 * lock_owner
9163 	 */
9164 	if (sp)
9165 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
9166 
9167 	rfs4_update_lock_resp(lsp, resop);
9168 
9169 end:
9170 	if (lsp) {
9171 		if (ls_sw_held)
9172 			rfs4_sw_exit(&lsp->rls_sw);
9173 		/*
9174 		 * If an sp obtained, then the lsp does not represent
9175 		 * a lock on the file struct.
9176 		 */
9177 		if (sp != NULL)
9178 			rfs4_lo_state_rele(lsp, FALSE);
9179 		else
9180 			rfs4_lo_state_rele(lsp, TRUE);
9181 	}
9182 	if (sp) {
9183 		rfs4_sw_exit(&sp->rs_owner->ro_sw);
9184 		rfs4_state_rele(sp);
9185 	}
9186 
9187 	DTRACE_NFSV4_2(op__lock__done, struct compound_state *, cs,
9188 	    LOCK4res *, resp);
9189 }
9190 
9191 /* free function for LOCK/LOCKT */
9192 static void
9193 lock_denied_free(nfs_resop4 *resop)
9194 {
9195 	LOCK4denied *dp = NULL;
9196 
9197 	switch (resop->resop) {
9198 	case OP_LOCK:
9199 		if (resop->nfs_resop4_u.oplock.status == NFS4ERR_DENIED)
9200 			dp = &resop->nfs_resop4_u.oplock.LOCK4res_u.denied;
9201 		break;
9202 	case OP_LOCKT:
9203 		if (resop->nfs_resop4_u.oplockt.status == NFS4ERR_DENIED)
9204 			dp = &resop->nfs_resop4_u.oplockt.denied;
9205 		break;
9206 	default:
9207 		break;
9208 	}
9209 
9210 	if (dp)
9211 		kmem_free(dp->owner.owner_val, dp->owner.owner_len);
9212 }
9213 
9214 /*ARGSUSED*/
9215 void
9216 rfs4_op_locku(nfs_argop4 *argop, nfs_resop4 *resop,
9217     struct svc_req *req, struct compound_state *cs)
9218 {
9219 	LOCKU4args *args = &argop->nfs_argop4_u.oplocku;
9220 	LOCKU4res *resp = &resop->nfs_resop4_u.oplocku;
9221 	nfsstat4 status;
9222 	stateid4 *stateid = &args->lock_stateid;
9223 	rfs4_lo_state_t *lsp;
9224 
9225 	DTRACE_NFSV4_2(op__locku__start, struct compound_state *, cs,
9226 	    LOCKU4args *, args);
9227 
9228 	if (cs->vp == NULL) {
9229 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9230 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9231 		    LOCKU4res *, resp);
9232 		return;
9233 	}
9234 
9235 	if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE)) != NFS4_OK) {
9236 		*cs->statusp = resp->status = status;
9237 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9238 		    LOCKU4res *, resp);
9239 		return;
9240 	}
9241 
9242 	/* Ensure specified filehandle matches */
9243 	if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9244 		rfs4_lo_state_rele(lsp, TRUE);
9245 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9246 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9247 		    LOCKU4res *, resp);
9248 		return;
9249 	}
9250 
9251 	/* hold off other access to lsp while we tinker */
9252 	rfs4_sw_enter(&lsp->rls_sw);
9253 
9254 	switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9255 	case NFS4_CHECK_STATEID_OKAY:
9256 		if (rfs4_check_lock_seqid(args->seqid, lsp, resop)
9257 		    != NFS4_CHKSEQ_OKAY) {
9258 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9259 			goto end;
9260 		}
9261 		break;
9262 	case NFS4_CHECK_STATEID_OLD:
9263 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9264 		goto end;
9265 	case NFS4_CHECK_STATEID_BAD:
9266 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9267 		goto end;
9268 	case NFS4_CHECK_STATEID_EXPIRED:
9269 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9270 		goto end;
9271 	case NFS4_CHECK_STATEID_CLOSED:
9272 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9273 		goto end;
9274 	case NFS4_CHECK_STATEID_REPLAY:
9275 		switch (rfs4_check_lock_seqid(args->seqid, lsp, resop)) {
9276 		case NFS4_CHKSEQ_OKAY:
9277 				/*
9278 				 * This is a replayed stateid; if
9279 				 * seqid matches the next expected,
9280 				 * then client is using wrong seqid.
9281 				 */
9282 		case NFS4_CHKSEQ_BAD:
9283 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9284 			goto end;
9285 		case NFS4_CHKSEQ_REPLAY:
9286 			rfs4_update_lease(lsp->rls_locker->rl_client);
9287 			*cs->statusp = status = resp->status;
9288 			goto end;
9289 		}
9290 		break;
9291 	default:
9292 		ASSERT(FALSE);
9293 		break;
9294 	}
9295 
9296 	rfs4_update_lock_sequence(lsp);
9297 	rfs4_update_lease(lsp->rls_locker->rl_client);
9298 
9299 	/*
9300 	 * NFS4 only allows locking on regular files, so
9301 	 * verify type of object.
9302 	 */
9303 	if (cs->vp->v_type != VREG) {
9304 		if (cs->vp->v_type == VDIR)
9305 			status = NFS4ERR_ISDIR;
9306 		else
9307 			status = NFS4ERR_INVAL;
9308 		goto out;
9309 	}
9310 
9311 	if (rfs4_clnt_in_grace(lsp->rls_state->rs_owner->ro_client)) {
9312 		status = NFS4ERR_GRACE;
9313 		goto out;
9314 	}
9315 
9316 	status = rfs4_do_lock(lsp, args->locktype,
9317 	    args->offset, args->length, cs->cr, resop);
9318 
9319 out:
9320 	*cs->statusp = resp->status = status;
9321 
9322 	if (status == NFS4_OK)
9323 		resp->lock_stateid = lsp->rls_lockid.stateid;
9324 
9325 	rfs4_update_lock_resp(lsp, resop);
9326 
9327 end:
9328 	rfs4_sw_exit(&lsp->rls_sw);
9329 	rfs4_lo_state_rele(lsp, TRUE);
9330 
9331 	DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9332 	    LOCKU4res *, resp);
9333 }
9334 
9335 /*
9336  * LOCKT is a best effort routine, the client can not be guaranteed that
9337  * the status return is still in effect by the time the reply is received.
9338  * They are numerous race conditions in this routine, but we are not required
9339  * and can not be accurate.
9340  */
9341 /*ARGSUSED*/
9342 void
9343 rfs4_op_lockt(nfs_argop4 *argop, nfs_resop4 *resop,
9344     struct svc_req *req, struct compound_state *cs)
9345 {
9346 	LOCKT4args *args = &argop->nfs_argop4_u.oplockt;
9347 	LOCKT4res *resp = &resop->nfs_resop4_u.oplockt;
9348 	rfs4_lockowner_t *lo;
9349 	rfs4_client_t *cp;
9350 	bool_t create = FALSE;
9351 	struct flock64 flk;
9352 	int error;
9353 	int flag = FREAD | FWRITE;
9354 	int ltype;
9355 	length4 posix_length;
9356 	sysid_t sysid;
9357 	pid_t pid;
9358 
9359 	DTRACE_NFSV4_2(op__lockt__start, struct compound_state *, cs,
9360 	    LOCKT4args *, args);
9361 
9362 	if (cs->vp == NULL) {
9363 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9364 		goto out;
9365 	}
9366 
9367 	/*
9368 	 * NFS4 only allows locking on regular files, so
9369 	 * verify type of object.
9370 	 */
9371 	if (cs->vp->v_type != VREG) {
9372 		if (cs->vp->v_type == VDIR)
9373 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
9374 		else
9375 			*cs->statusp = resp->status =  NFS4ERR_INVAL;
9376 		goto out;
9377 	}
9378 
9379 	/*
9380 	 * Check out the clientid to ensure the server knows about it
9381 	 * so that we correctly inform the client of a server reboot.
9382 	 */
9383 	if ((cp = rfs4_findclient_by_id(args->owner.clientid, FALSE))
9384 	    == NULL) {
9385 		*cs->statusp = resp->status =
9386 		    rfs4_check_clientid(&args->owner.clientid, 0);
9387 		goto out;
9388 	}
9389 	if (rfs4_lease_expired(cp)) {
9390 		rfs4_client_close(cp);
9391 		/*
9392 		 * Protocol doesn't allow returning NFS4ERR_STALE as
9393 		 * other operations do on this check so STALE_CLIENTID
9394 		 * is returned instead
9395 		 */
9396 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
9397 		goto out;
9398 	}
9399 
9400 	if (rfs4_clnt_in_grace(cp) && !(cp->rc_can_reclaim)) {
9401 		*cs->statusp = resp->status = NFS4ERR_GRACE;
9402 		rfs4_client_rele(cp);
9403 		goto out;
9404 	}
9405 	rfs4_client_rele(cp);
9406 
9407 	resp->status = NFS4_OK;
9408 
9409 	switch (args->locktype) {
9410 	case READ_LT:
9411 	case READW_LT:
9412 		ltype = F_RDLCK;
9413 		break;
9414 	case WRITE_LT:
9415 	case WRITEW_LT:
9416 		ltype = F_WRLCK;
9417 		break;
9418 	}
9419 
9420 	posix_length = args->length;
9421 	/* Check for zero length. To lock to end of file use all ones for V4 */
9422 	if (posix_length == 0) {
9423 		*cs->statusp = resp->status = NFS4ERR_INVAL;
9424 		goto out;
9425 	} else if (posix_length == (length4)(~0)) {
9426 		posix_length = 0;	/* Posix to end of file  */
9427 	}
9428 
9429 	/* Find or create a lockowner */
9430 	lo = rfs4_findlockowner(&args->owner, &create);
9431 
9432 	if (lo) {
9433 		pid = lo->rl_pid;
9434 		if ((resp->status =
9435 		    rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
9436 			goto err;
9437 	} else {
9438 		pid = 0;
9439 		sysid = lockt_sysid;
9440 	}
9441 retry:
9442 	flk.l_type = ltype;
9443 	flk.l_whence = 0;		/* SEEK_SET */
9444 	flk.l_start = args->offset;
9445 	flk.l_len = posix_length;
9446 	flk.l_sysid = sysid;
9447 	flk.l_pid = pid;
9448 	flag |= F_REMOTELOCK;
9449 
9450 	LOCK_PRINT(rfs4_debug, "rfs4_op_lockt", F_GETLK, &flk);
9451 
9452 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
9453 	if (flk.l_len < 0 || flk.l_start < 0) {
9454 		resp->status = NFS4ERR_INVAL;
9455 		goto err;
9456 	}
9457 	error = VOP_FRLOCK(cs->vp, F_GETLK, &flk, flag, (u_offset_t)0,
9458 	    NULL, cs->cr, NULL);
9459 
9460 	/*
9461 	 * N.B. We map error values to nfsv4 errors. This is differrent
9462 	 * than puterrno4 routine.
9463 	 */
9464 	switch (error) {
9465 	case 0:
9466 		if (flk.l_type == F_UNLCK)
9467 			resp->status = NFS4_OK;
9468 		else {
9469 			if (lock_denied(&resp->denied, &flk) == NFS4ERR_EXPIRED)
9470 				goto retry;
9471 			resp->status = NFS4ERR_DENIED;
9472 		}
9473 		break;
9474 	case EOVERFLOW:
9475 		resp->status = NFS4ERR_INVAL;
9476 		break;
9477 	case EINVAL:
9478 		resp->status = NFS4ERR_NOTSUPP;
9479 		break;
9480 	default:
9481 		cmn_err(CE_WARN, "rfs4_op_lockt: unexpected errno (%d)",
9482 		    error);
9483 		resp->status = NFS4ERR_SERVERFAULT;
9484 		break;
9485 	}
9486 
9487 err:
9488 	if (lo)
9489 		rfs4_lockowner_rele(lo);
9490 	*cs->statusp = resp->status;
9491 out:
9492 	DTRACE_NFSV4_2(op__lockt__done, struct compound_state *, cs,
9493 	    LOCKT4res *, resp);
9494 }
9495 
9496 int
9497 rfs4_share(rfs4_state_t *sp, uint32_t access, uint32_t deny)
9498 {
9499 	int err;
9500 	int cmd;
9501 	vnode_t *vp;
9502 	struct shrlock shr;
9503 	struct shr_locowner shr_loco;
9504 	int fflags = 0;
9505 
9506 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9507 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9508 
9509 	if (sp->rs_closed)
9510 		return (NFS4ERR_OLD_STATEID);
9511 
9512 	vp = sp->rs_finfo->rf_vp;
9513 	ASSERT(vp);
9514 
9515 	shr.s_access = shr.s_deny = 0;
9516 
9517 	if (access & OPEN4_SHARE_ACCESS_READ) {
9518 		fflags |= FREAD;
9519 		shr.s_access |= F_RDACC;
9520 	}
9521 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
9522 		fflags |= FWRITE;
9523 		shr.s_access |= F_WRACC;
9524 	}
9525 	ASSERT(shr.s_access);
9526 
9527 	if (deny & OPEN4_SHARE_DENY_READ)
9528 		shr.s_deny |= F_RDDNY;
9529 	if (deny & OPEN4_SHARE_DENY_WRITE)
9530 		shr.s_deny |= F_WRDNY;
9531 
9532 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9533 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9534 	shr_loco.sl_pid = shr.s_pid;
9535 	shr_loco.sl_id = shr.s_sysid;
9536 	shr.s_owner = (caddr_t)&shr_loco;
9537 	shr.s_own_len = sizeof (shr_loco);
9538 
9539 	cmd = nbl_need_check(vp) ? F_SHARE_NBMAND : F_SHARE;
9540 
9541 	err = VOP_SHRLOCK(vp, cmd, &shr, fflags, CRED(), NULL);
9542 	if (err != 0) {
9543 		if (err == EAGAIN)
9544 			err = NFS4ERR_SHARE_DENIED;
9545 		else
9546 			err = puterrno4(err);
9547 		return (err);
9548 	}
9549 
9550 	sp->rs_share_access |= access;
9551 	sp->rs_share_deny |= deny;
9552 
9553 	return (0);
9554 }
9555 
9556 int
9557 rfs4_unshare(rfs4_state_t *sp)
9558 {
9559 	int err;
9560 	struct shrlock shr;
9561 	struct shr_locowner shr_loco;
9562 
9563 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9564 
9565 	if (sp->rs_closed || sp->rs_share_access == 0)
9566 		return (0);
9567 
9568 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9569 	ASSERT(sp->rs_finfo->rf_vp);
9570 
9571 	shr.s_access = shr.s_deny = 0;
9572 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9573 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9574 	shr_loco.sl_pid = shr.s_pid;
9575 	shr_loco.sl_id = shr.s_sysid;
9576 	shr.s_owner = (caddr_t)&shr_loco;
9577 	shr.s_own_len = sizeof (shr_loco);
9578 
9579 	err = VOP_SHRLOCK(sp->rs_finfo->rf_vp, F_UNSHARE, &shr, 0, CRED(),
9580 	    NULL);
9581 	if (err != 0) {
9582 		err = puterrno4(err);
9583 		return (err);
9584 	}
9585 
9586 	sp->rs_share_access = 0;
9587 	sp->rs_share_deny = 0;
9588 
9589 	return (0);
9590 
9591 }
9592 
9593 static int
9594 rdma_setup_read_data4(READ4args *args, READ4res *rok)
9595 {
9596 	struct clist	*wcl;
9597 	count4		count = rok->data_len;
9598 	int		wlist_len;
9599 
9600 	wcl = args->wlist;
9601 	if (rdma_setup_read_chunks(wcl, count, &wlist_len) == FALSE) {
9602 		return (FALSE);
9603 	}
9604 	wcl = args->wlist;
9605 	rok->wlist_len = wlist_len;
9606 	rok->wlist = wcl;
9607 	return (TRUE);
9608 }
9609 
9610 /* tunable to disable server referrals */
9611 int rfs4_no_referrals = 0;
9612 
9613 /*
9614  * Find an NFS record in reparse point data.
9615  * Returns 0 for success and <0 or an errno value on failure.
9616  */
9617 int
9618 vn_find_nfs_record(vnode_t *vp, nvlist_t **nvlp, char **svcp, char **datap)
9619 {
9620 	int err;
9621 	char *stype, *val;
9622 	nvlist_t *nvl;
9623 	nvpair_t *curr;
9624 
9625 	if ((nvl = reparse_init()) == NULL)
9626 		return (-1);
9627 
9628 	if ((err = reparse_vnode_parse(vp, nvl)) != 0) {
9629 		reparse_free(nvl);
9630 		return (err);
9631 	}
9632 
9633 	curr = NULL;
9634 	while ((curr = nvlist_next_nvpair(nvl, curr)) != NULL) {
9635 		if ((stype = nvpair_name(curr)) == NULL) {
9636 			reparse_free(nvl);
9637 			return (-2);
9638 		}
9639 		if (strncasecmp(stype, "NFS", 3) == 0)
9640 			break;
9641 	}
9642 
9643 	if ((curr == NULL) ||
9644 	    (nvpair_value_string(curr, &val))) {
9645 		reparse_free(nvl);
9646 		return (-3);
9647 	}
9648 	*nvlp = nvl;
9649 	*svcp = stype;
9650 	*datap = val;
9651 	return (0);
9652 }
9653 
9654 int
9655 vn_is_nfs_reparse(vnode_t *vp, cred_t *cr)
9656 {
9657 	nvlist_t *nvl;
9658 	char *s, *d;
9659 
9660 	if (rfs4_no_referrals != 0)
9661 		return (B_FALSE);
9662 
9663 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9664 		return (B_FALSE);
9665 
9666 	if (vn_find_nfs_record(vp, &nvl, &s, &d) != 0)
9667 		return (B_FALSE);
9668 
9669 	reparse_free(nvl);
9670 
9671 	return (B_TRUE);
9672 }
9673 
9674 /*
9675  * There is a user-level copy of this routine in ref_subr.c.
9676  * Changes should be kept in sync.
9677  */
9678 static int
9679 nfs4_create_components(char *path, component4 *comp4)
9680 {
9681 	int slen, plen, ncomp;
9682 	char *ori_path, *nxtc, buf[MAXNAMELEN];
9683 
9684 	if (path == NULL)
9685 		return (0);
9686 
9687 	plen = strlen(path) + 1;	/* include the terminator */
9688 	ori_path = path;
9689 	ncomp = 0;
9690 
9691 	/* count number of components in the path */
9692 	for (nxtc = path; nxtc < ori_path + plen; nxtc++) {
9693 		if (*nxtc == '/' || *nxtc == '\0' || *nxtc == '\n') {
9694 			if ((slen = nxtc - path) == 0) {
9695 				path = nxtc + 1;
9696 				continue;
9697 			}
9698 
9699 			if (comp4 != NULL) {
9700 				bcopy(path, buf, slen);
9701 				buf[slen] = '\0';
9702 				(void) str_to_utf8(buf, &comp4[ncomp]);
9703 			}
9704 
9705 			ncomp++;	/* 1 valid component */
9706 			path = nxtc + 1;
9707 		}
9708 		if (*nxtc == '\0' || *nxtc == '\n')
9709 			break;
9710 	}
9711 
9712 	return (ncomp);
9713 }
9714 
9715 /*
9716  * There is a user-level copy of this routine in ref_subr.c.
9717  * Changes should be kept in sync.
9718  */
9719 static int
9720 make_pathname4(char *path, pathname4 *pathname)
9721 {
9722 	int ncomp;
9723 	component4 *comp4;
9724 
9725 	if (pathname == NULL)
9726 		return (0);
9727 
9728 	if (path == NULL) {
9729 		pathname->pathname4_val = NULL;
9730 		pathname->pathname4_len = 0;
9731 		return (0);
9732 	}
9733 
9734 	/* count number of components to alloc buffer */
9735 	if ((ncomp = nfs4_create_components(path, NULL)) == 0) {
9736 		pathname->pathname4_val = NULL;
9737 		pathname->pathname4_len = 0;
9738 		return (0);
9739 	}
9740 	comp4 = kmem_zalloc(ncomp * sizeof (component4), KM_SLEEP);
9741 
9742 	/* copy components into allocated buffer */
9743 	ncomp = nfs4_create_components(path, comp4);
9744 
9745 	pathname->pathname4_val = comp4;
9746 	pathname->pathname4_len = ncomp;
9747 
9748 	return (ncomp);
9749 }
9750 
9751 #define	xdr_fs_locations4 xdr_fattr4_fs_locations
9752 
9753 fs_locations4 *
9754 fetch_referral(vnode_t *vp, cred_t *cr)
9755 {
9756 	nvlist_t *nvl;
9757 	char *stype, *sdata;
9758 	fs_locations4 *result;
9759 	char buf[1024];
9760 	size_t bufsize;
9761 	XDR xdr;
9762 	int err;
9763 
9764 	/*
9765 	 * Check attrs to ensure it's a reparse point
9766 	 */
9767 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9768 		return (NULL);
9769 
9770 	/*
9771 	 * Look for an NFS record and get the type and data
9772 	 */
9773 	if (vn_find_nfs_record(vp, &nvl, &stype, &sdata) != 0)
9774 		return (NULL);
9775 
9776 	/*
9777 	 * With the type and data, upcall to get the referral
9778 	 */
9779 	bufsize = sizeof (buf);
9780 	bzero(buf, sizeof (buf));
9781 	err = reparse_kderef((const char *)stype, (const char *)sdata,
9782 	    buf, &bufsize);
9783 	reparse_free(nvl);
9784 
9785 	DTRACE_PROBE4(nfs4serv__func__referral__upcall,
9786 	    char *, stype, char *, sdata, char *, buf, int, err);
9787 	if (err) {
9788 		cmn_err(CE_NOTE,
9789 		    "reparsed daemon not running: unable to get referral (%d)",
9790 		    err);
9791 		return (NULL);
9792 	}
9793 
9794 	/*
9795 	 * We get an XDR'ed record back from the kderef call
9796 	 */
9797 	xdrmem_create(&xdr, buf, bufsize, XDR_DECODE);
9798 	result = kmem_alloc(sizeof (fs_locations4), KM_SLEEP);
9799 	err = xdr_fs_locations4(&xdr, result);
9800 	XDR_DESTROY(&xdr);
9801 	if (err != TRUE) {
9802 		DTRACE_PROBE1(nfs4serv__func__referral__upcall__xdrfail,
9803 		    int, err);
9804 		return (NULL);
9805 	}
9806 
9807 	/*
9808 	 * Look at path to recover fs_root, ignoring the leading '/'
9809 	 */
9810 	(void) make_pathname4(vp->v_path, &result->fs_root);
9811 
9812 	return (result);
9813 }
9814 
9815 char *
9816 build_symlink(vnode_t *vp, cred_t *cr, size_t *strsz)
9817 {
9818 	fs_locations4 *fsl;
9819 	fs_location4 *fs;
9820 	char *server, *path, *symbuf;
9821 	static char *prefix = "/net/";
9822 	int i, size, npaths;
9823 	uint_t len;
9824 
9825 	/* Get the referral */
9826 	if ((fsl = fetch_referral(vp, cr)) == NULL)
9827 		return (NULL);
9828 
9829 	/* Deal with only the first location and first server */
9830 	fs = &fsl->locations_val[0];
9831 	server = utf8_to_str(&fs->server_val[0], &len, NULL);
9832 	if (server == NULL) {
9833 		rfs4_free_fs_locations4(fsl);
9834 		kmem_free(fsl, sizeof (fs_locations4));
9835 		return (NULL);
9836 	}
9837 
9838 	/* Figure out size for "/net/" + host + /path/path/path + NULL */
9839 	size = strlen(prefix) + len;
9840 	for (i = 0; i < fs->rootpath.pathname4_len; i++)
9841 		size += fs->rootpath.pathname4_val[i].utf8string_len + 1;
9842 
9843 	/* Allocate the symlink buffer and fill it */
9844 	symbuf = kmem_zalloc(size, KM_SLEEP);
9845 	(void) strcat(symbuf, prefix);
9846 	(void) strcat(symbuf, server);
9847 	kmem_free(server, len);
9848 
9849 	npaths = 0;
9850 	for (i = 0; i < fs->rootpath.pathname4_len; i++) {
9851 		path = utf8_to_str(&fs->rootpath.pathname4_val[i], &len, NULL);
9852 		if (path == NULL)
9853 			continue;
9854 		(void) strcat(symbuf, "/");
9855 		(void) strcat(symbuf, path);
9856 		npaths++;
9857 		kmem_free(path, len);
9858 	}
9859 
9860 	rfs4_free_fs_locations4(fsl);
9861 	kmem_free(fsl, sizeof (fs_locations4));
9862 
9863 	if (strsz != NULL)
9864 		*strsz = size;
9865 	return (symbuf);
9866 }
9867 
9868 /*
9869  * Check to see if we have a downrev Solaris client, so that we
9870  * can send it a symlink instead of a referral.
9871  */
9872 int
9873 client_is_downrev(struct svc_req *req)
9874 {
9875 	struct sockaddr *ca;
9876 	rfs4_clntip_t *ci;
9877 	bool_t create = FALSE;
9878 	int is_downrev;
9879 
9880 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
9881 	ASSERT(ca);
9882 	ci = rfs4_find_clntip(ca, &create);
9883 	if (ci == NULL)
9884 		return (0);
9885 	is_downrev = ci->ri_no_referrals;
9886 	rfs4_dbe_rele(ci->ri_dbe);
9887 	return (is_downrev);
9888 }
9889