xref: /titanic_41/usr/src/uts/common/fs/nfs/nfs4_srv.c (revision 5203bc321053fb87d7073c7640548fab73634793)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 /*
26  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
27  *	All Rights Reserved
28  */
29 
30 #include <sys/param.h>
31 #include <sys/types.h>
32 #include <sys/systm.h>
33 #include <sys/cred.h>
34 #include <sys/buf.h>
35 #include <sys/vfs.h>
36 #include <sys/vfs_opreg.h>
37 #include <sys/vnode.h>
38 #include <sys/uio.h>
39 #include <sys/errno.h>
40 #include <sys/sysmacros.h>
41 #include <sys/statvfs.h>
42 #include <sys/kmem.h>
43 #include <sys/dirent.h>
44 #include <sys/cmn_err.h>
45 #include <sys/debug.h>
46 #include <sys/systeminfo.h>
47 #include <sys/flock.h>
48 #include <sys/pathname.h>
49 #include <sys/nbmlock.h>
50 #include <sys/share.h>
51 #include <sys/atomic.h>
52 #include <sys/policy.h>
53 #include <sys/fem.h>
54 #include <sys/sdt.h>
55 #include <sys/ddi.h>
56 #include <sys/zone.h>
57 
58 #include <fs/fs_reparse.h>
59 
60 #include <rpc/types.h>
61 #include <rpc/auth.h>
62 #include <rpc/rpcsec_gss.h>
63 #include <rpc/svc.h>
64 
65 #include <nfs/nfs.h>
66 #include <nfs/export.h>
67 #include <nfs/nfs_cmd.h>
68 #include <nfs/lm.h>
69 #include <nfs/nfs4.h>
70 
71 #include <sys/strsubr.h>
72 #include <sys/strsun.h>
73 
74 #include <inet/common.h>
75 #include <inet/ip.h>
76 #include <inet/ip6.h>
77 
78 #include <sys/tsol/label.h>
79 #include <sys/tsol/tndb.h>
80 
81 #define	RFS4_MAXLOCK_TRIES 4	/* Try to get the lock this many times */
82 static int rfs4_maxlock_tries = RFS4_MAXLOCK_TRIES;
83 #define	RFS4_LOCK_DELAY 10	/* Milliseconds */
84 static clock_t  rfs4_lock_delay = RFS4_LOCK_DELAY;
85 extern struct svc_ops rdma_svc_ops;
86 extern int nfs_loaned_buffers;
87 /* End of Tunables */
88 
89 static int rdma_setup_read_data4(READ4args *, READ4res *);
90 
91 /*
92  * Used to bump the stateid4.seqid value and show changes in the stateid
93  */
94 #define	next_stateid(sp) (++(sp)->bits.chgseq)
95 
96 /*
97  * RFS4_MINLEN_ENTRY4: XDR-encoded size of smallest possible dirent.
98  *	This is used to return NFS4ERR_TOOSMALL when clients specify
99  *	maxcount that isn't large enough to hold the smallest possible
100  *	XDR encoded dirent.
101  *
102  *	    sizeof cookie (8 bytes) +
103  *	    sizeof name_len (4 bytes) +
104  *	    sizeof smallest (padded) name (4 bytes) +
105  *	    sizeof bitmap4_len (12 bytes) +   NOTE: we always encode len=2 bm4
106  *	    sizeof attrlist4_len (4 bytes) +
107  *	    sizeof next boolean (4 bytes)
108  *
109  * RFS4_MINLEN_RDDIR4: XDR-encoded size of READDIR op reply containing
110  * the smallest possible entry4 (assumes no attrs requested).
111  *	sizeof nfsstat4 (4 bytes) +
112  *	sizeof verifier4 (8 bytes) +
113  *	sizeof entry4list bool (4 bytes) +
114  *	sizeof entry4 	(36 bytes) +
115  *	sizeof eof bool  (4 bytes)
116  *
117  * RFS4_MINLEN_RDDIR_BUF: minimum length of buffer server will provide to
118  *	VOP_READDIR.  Its value is the size of the maximum possible dirent
119  *	for solaris.  The DIRENT64_RECLEN macro returns	the size of dirent
120  *	required for a given name length.  MAXNAMELEN is the maximum
121  *	filename length allowed in Solaris.  The first two DIRENT64_RECLEN()
122  *	macros are to allow for . and .. entries -- just a minor tweak to try
123  *	and guarantee that buffer we give to VOP_READDIR will be large enough
124  *	to hold ., .., and the largest possible solaris dirent64.
125  */
126 #define	RFS4_MINLEN_ENTRY4 36
127 #define	RFS4_MINLEN_RDDIR4 (4 + NFS4_VERIFIER_SIZE + 4 + RFS4_MINLEN_ENTRY4 + 4)
128 #define	RFS4_MINLEN_RDDIR_BUF \
129 	(DIRENT64_RECLEN(1) + DIRENT64_RECLEN(2) + DIRENT64_RECLEN(MAXNAMELEN))
130 
131 /*
132  * It would be better to pad to 4 bytes since that's what XDR would do,
133  * but the dirents UFS gives us are already padded to 8, so just take
134  * what we're given.  Dircount is only a hint anyway.  Currently the
135  * solaris kernel is ASCII only, so there's no point in calling the
136  * UTF8 functions.
137  *
138  * dirent64: named padded to provide 8 byte struct alignment
139  *	d_ino(8) + d_off(8) + d_reclen(2) + d_name(namelen + null(1) + pad)
140  *
141  * cookie: uint64_t   +  utf8namelen: uint_t  +   utf8name padded to 8 bytes
142  *
143  */
144 #define	DIRENT64_TO_DIRCOUNT(dp) \
145 	(3 * BYTES_PER_XDR_UNIT + DIRENT64_NAMELEN((dp)->d_reclen))
146 
147 time_t rfs4_start_time;			/* Initialized in rfs4_srvrinit */
148 
149 static sysid_t lockt_sysid;		/* dummy sysid for all LOCKT calls */
150 
151 u_longlong_t	nfs4_srv_caller_id;
152 uint_t		nfs4_srv_vkey = 0;
153 
154 verifier4	Write4verf;
155 verifier4	Readdir4verf;
156 
157 void	rfs4_init_compound_state(struct compound_state *);
158 
159 static void	nullfree(caddr_t);
160 static void	rfs4_op_inval(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
161 			struct compound_state *);
162 static void	rfs4_op_access(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
163 			struct compound_state *);
164 static void	rfs4_op_close(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
165 			struct compound_state *);
166 static void	rfs4_op_commit(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
167 			struct compound_state *);
168 static void	rfs4_op_create(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
169 			struct compound_state *);
170 static void	rfs4_op_create_free(nfs_resop4 *resop);
171 static void	rfs4_op_delegreturn(nfs_argop4 *, nfs_resop4 *,
172 			struct svc_req *, struct compound_state *);
173 static void	rfs4_op_delegpurge(nfs_argop4 *, nfs_resop4 *,
174 			struct svc_req *, struct compound_state *);
175 static void	rfs4_op_getattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
176 			struct compound_state *);
177 static void	rfs4_op_getattr_free(nfs_resop4 *);
178 static void	rfs4_op_getfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
179 			struct compound_state *);
180 static void	rfs4_op_getfh_free(nfs_resop4 *);
181 static void	rfs4_op_illegal(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
182 			struct compound_state *);
183 static void	rfs4_op_link(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
184 			struct compound_state *);
185 static void	rfs4_op_lock(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
186 			struct compound_state *);
187 static void	lock_denied_free(nfs_resop4 *);
188 static void	rfs4_op_locku(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
189 			struct compound_state *);
190 static void	rfs4_op_lockt(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
191 			struct compound_state *);
192 static void	rfs4_op_lookup(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
193 			struct compound_state *);
194 static void	rfs4_op_lookupp(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
195 			struct compound_state *);
196 static void	rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop,
197 				struct svc_req *req, struct compound_state *cs);
198 static void	rfs4_op_nverify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
199 			struct compound_state *);
200 static void	rfs4_op_open(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
201 			struct compound_state *);
202 static void	rfs4_op_open_confirm(nfs_argop4 *, nfs_resop4 *,
203 			struct svc_req *, struct compound_state *);
204 static void	rfs4_op_open_downgrade(nfs_argop4 *, nfs_resop4 *,
205 			struct svc_req *, struct compound_state *);
206 static void	rfs4_op_putfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
207 			struct compound_state *);
208 static void	rfs4_op_putpubfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
209 			struct compound_state *);
210 static void	rfs4_op_putrootfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
211 			struct compound_state *);
212 static void	rfs4_op_read(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
213 			struct compound_state *);
214 static void	rfs4_op_read_free(nfs_resop4 *);
215 static void	rfs4_op_readdir_free(nfs_resop4 *resop);
216 static void	rfs4_op_readlink(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
217 			struct compound_state *);
218 static void	rfs4_op_readlink_free(nfs_resop4 *);
219 static void	rfs4_op_release_lockowner(nfs_argop4 *, nfs_resop4 *,
220 			struct svc_req *, struct compound_state *);
221 static void	rfs4_op_remove(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
222 			struct compound_state *);
223 static void	rfs4_op_rename(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
224 			struct compound_state *);
225 static void	rfs4_op_renew(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
226 			struct compound_state *);
227 static void	rfs4_op_restorefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
228 			struct compound_state *);
229 static void	rfs4_op_savefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
230 			struct compound_state *);
231 static void	rfs4_op_setattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
232 			struct compound_state *);
233 static void	rfs4_op_verify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
234 			struct compound_state *);
235 static void	rfs4_op_write(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
236 			struct compound_state *);
237 static void	rfs4_op_setclientid(nfs_argop4 *, nfs_resop4 *,
238 			struct svc_req *, struct compound_state *);
239 static void	rfs4_op_setclientid_confirm(nfs_argop4 *, nfs_resop4 *,
240 			struct svc_req *req, struct compound_state *);
241 static void	rfs4_op_secinfo(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
242 			struct compound_state *);
243 static void	rfs4_op_secinfo_free(nfs_resop4 *);
244 
245 static nfsstat4 check_open_access(uint32_t,
246 				struct compound_state *, struct svc_req *);
247 nfsstat4 rfs4_client_sysid(rfs4_client_t *, sysid_t *);
248 void rfs4_ss_clid(rfs4_client_t *);
249 
250 /*
251  * translation table for attrs
252  */
253 struct nfs4_ntov_table {
254 	union nfs4_attr_u *na;
255 	uint8_t amap[NFS4_MAXNUM_ATTRS];
256 	int attrcnt;
257 	bool_t vfsstat;
258 };
259 
260 static void	nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp);
261 static void	nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
262 				    struct nfs4_svgetit_arg *sargp);
263 
264 static nfsstat4	do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp,
265 		    struct compound_state *cs, struct nfs4_svgetit_arg *sargp,
266 		    struct nfs4_ntov_table *ntovp, nfs4_attr_cmd_t cmd);
267 
268 fem_t		*deleg_rdops;
269 fem_t		*deleg_wrops;
270 
271 rfs4_servinst_t *rfs4_cur_servinst = NULL;	/* current server instance */
272 kmutex_t	rfs4_servinst_lock;	/* protects linked list */
273 int		rfs4_seen_first_compound;	/* set first time we see one */
274 
275 /*
276  * NFS4 op dispatch table
277  */
278 
279 struct rfsv4disp {
280 	void	(*dis_proc)();		/* proc to call */
281 	void	(*dis_resfree)();	/* frees space allocated by proc */
282 	int	dis_flags;		/* RPC_IDEMPOTENT, etc... */
283 };
284 
285 static struct rfsv4disp rfsv4disptab[] = {
286 	/*
287 	 * NFS VERSION 4
288 	 */
289 
290 	/* RFS_NULL = 0 */
291 	{rfs4_op_illegal, nullfree, 0},
292 
293 	/* UNUSED = 1 */
294 	{rfs4_op_illegal, nullfree, 0},
295 
296 	/* UNUSED = 2 */
297 	{rfs4_op_illegal, nullfree, 0},
298 
299 	/* OP_ACCESS = 3 */
300 	{rfs4_op_access, nullfree, RPC_IDEMPOTENT},
301 
302 	/* OP_CLOSE = 4 */
303 	{rfs4_op_close, nullfree, 0},
304 
305 	/* OP_COMMIT = 5 */
306 	{rfs4_op_commit, nullfree, RPC_IDEMPOTENT},
307 
308 	/* OP_CREATE = 6 */
309 	{rfs4_op_create, nullfree, 0},
310 
311 	/* OP_DELEGPURGE = 7 */
312 	{rfs4_op_delegpurge, nullfree, 0},
313 
314 	/* OP_DELEGRETURN = 8 */
315 	{rfs4_op_delegreturn, nullfree, 0},
316 
317 	/* OP_GETATTR = 9 */
318 	{rfs4_op_getattr, rfs4_op_getattr_free, RPC_IDEMPOTENT},
319 
320 	/* OP_GETFH = 10 */
321 	{rfs4_op_getfh, rfs4_op_getfh_free, RPC_ALL},
322 
323 	/* OP_LINK = 11 */
324 	{rfs4_op_link, nullfree, 0},
325 
326 	/* OP_LOCK = 12 */
327 	{rfs4_op_lock, lock_denied_free, 0},
328 
329 	/* OP_LOCKT = 13 */
330 	{rfs4_op_lockt, lock_denied_free, 0},
331 
332 	/* OP_LOCKU = 14 */
333 	{rfs4_op_locku, nullfree, 0},
334 
335 	/* OP_LOOKUP = 15 */
336 	{rfs4_op_lookup, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
337 
338 	/* OP_LOOKUPP = 16 */
339 	{rfs4_op_lookupp, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
340 
341 	/* OP_NVERIFY = 17 */
342 	{rfs4_op_nverify, nullfree, RPC_IDEMPOTENT},
343 
344 	/* OP_OPEN = 18 */
345 	{rfs4_op_open, rfs4_free_reply, 0},
346 
347 	/* OP_OPENATTR = 19 */
348 	{rfs4_op_openattr, nullfree, 0},
349 
350 	/* OP_OPEN_CONFIRM = 20 */
351 	{rfs4_op_open_confirm, nullfree, 0},
352 
353 	/* OP_OPEN_DOWNGRADE = 21 */
354 	{rfs4_op_open_downgrade, nullfree, 0},
355 
356 	/* OP_OPEN_PUTFH = 22 */
357 	{rfs4_op_putfh, nullfree, RPC_ALL},
358 
359 	/* OP_PUTPUBFH = 23 */
360 	{rfs4_op_putpubfh, nullfree, RPC_ALL},
361 
362 	/* OP_PUTROOTFH = 24 */
363 	{rfs4_op_putrootfh, nullfree, RPC_ALL},
364 
365 	/* OP_READ = 25 */
366 	{rfs4_op_read, rfs4_op_read_free, RPC_IDEMPOTENT},
367 
368 	/* OP_READDIR = 26 */
369 	{rfs4_op_readdir, rfs4_op_readdir_free, RPC_IDEMPOTENT},
370 
371 	/* OP_READLINK = 27 */
372 	{rfs4_op_readlink, rfs4_op_readlink_free, RPC_IDEMPOTENT},
373 
374 	/* OP_REMOVE = 28 */
375 	{rfs4_op_remove, nullfree, 0},
376 
377 	/* OP_RENAME = 29 */
378 	{rfs4_op_rename, nullfree, 0},
379 
380 	/* OP_RENEW = 30 */
381 	{rfs4_op_renew, nullfree, 0},
382 
383 	/* OP_RESTOREFH = 31 */
384 	{rfs4_op_restorefh, nullfree, RPC_ALL},
385 
386 	/* OP_SAVEFH = 32 */
387 	{rfs4_op_savefh, nullfree, RPC_ALL},
388 
389 	/* OP_SECINFO = 33 */
390 	{rfs4_op_secinfo, rfs4_op_secinfo_free, 0},
391 
392 	/* OP_SETATTR = 34 */
393 	{rfs4_op_setattr, nullfree, 0},
394 
395 	/* OP_SETCLIENTID = 35 */
396 	{rfs4_op_setclientid, nullfree, 0},
397 
398 	/* OP_SETCLIENTID_CONFIRM = 36 */
399 	{rfs4_op_setclientid_confirm, nullfree, 0},
400 
401 	/* OP_VERIFY = 37 */
402 	{rfs4_op_verify, nullfree, RPC_IDEMPOTENT},
403 
404 	/* OP_WRITE = 38 */
405 	{rfs4_op_write, nullfree, 0},
406 
407 	/* OP_RELEASE_LOCKOWNER = 39 */
408 	{rfs4_op_release_lockowner, nullfree, 0},
409 };
410 
411 static uint_t rfsv4disp_cnt = sizeof (rfsv4disptab) / sizeof (rfsv4disptab[0]);
412 
413 #define	OP_ILLEGAL_IDX (rfsv4disp_cnt)
414 
415 #ifdef DEBUG
416 
417 int		rfs4_fillone_debug = 0;
418 int		rfs4_no_stub_access = 1;
419 int		rfs4_rddir_debug = 0;
420 
421 static char    *rfs4_op_string[] = {
422 	"rfs4_op_null",
423 	"rfs4_op_1 unused",
424 	"rfs4_op_2 unused",
425 	"rfs4_op_access",
426 	"rfs4_op_close",
427 	"rfs4_op_commit",
428 	"rfs4_op_create",
429 	"rfs4_op_delegpurge",
430 	"rfs4_op_delegreturn",
431 	"rfs4_op_getattr",
432 	"rfs4_op_getfh",
433 	"rfs4_op_link",
434 	"rfs4_op_lock",
435 	"rfs4_op_lockt",
436 	"rfs4_op_locku",
437 	"rfs4_op_lookup",
438 	"rfs4_op_lookupp",
439 	"rfs4_op_nverify",
440 	"rfs4_op_open",
441 	"rfs4_op_openattr",
442 	"rfs4_op_open_confirm",
443 	"rfs4_op_open_downgrade",
444 	"rfs4_op_putfh",
445 	"rfs4_op_putpubfh",
446 	"rfs4_op_putrootfh",
447 	"rfs4_op_read",
448 	"rfs4_op_readdir",
449 	"rfs4_op_readlink",
450 	"rfs4_op_remove",
451 	"rfs4_op_rename",
452 	"rfs4_op_renew",
453 	"rfs4_op_restorefh",
454 	"rfs4_op_savefh",
455 	"rfs4_op_secinfo",
456 	"rfs4_op_setattr",
457 	"rfs4_op_setclientid",
458 	"rfs4_op_setclient_confirm",
459 	"rfs4_op_verify",
460 	"rfs4_op_write",
461 	"rfs4_op_release_lockowner",
462 	"rfs4_op_illegal"
463 };
464 #endif
465 
466 void	rfs4_ss_chkclid(rfs4_client_t *);
467 
468 extern size_t   strlcpy(char *dst, const char *src, size_t dstsize);
469 
470 extern void	rfs4_free_fs_locations4(fs_locations4 *);
471 
472 #ifdef	nextdp
473 #undef nextdp
474 #endif
475 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
476 
477 static const fs_operation_def_t nfs4_rd_deleg_tmpl[] = {
478 	VOPNAME_OPEN,		{ .femop_open = deleg_rd_open },
479 	VOPNAME_WRITE,		{ .femop_write = deleg_rd_write },
480 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_rd_setattr },
481 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_rd_rwlock },
482 	VOPNAME_SPACE,		{ .femop_space = deleg_rd_space },
483 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_rd_setsecattr },
484 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_rd_vnevent },
485 	NULL,			NULL
486 };
487 static const fs_operation_def_t nfs4_wr_deleg_tmpl[] = {
488 	VOPNAME_OPEN,		{ .femop_open = deleg_wr_open },
489 	VOPNAME_READ,		{ .femop_read = deleg_wr_read },
490 	VOPNAME_WRITE,		{ .femop_write = deleg_wr_write },
491 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_wr_setattr },
492 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_wr_rwlock },
493 	VOPNAME_SPACE,		{ .femop_space = deleg_wr_space },
494 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_wr_setsecattr },
495 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_wr_vnevent },
496 	NULL,			NULL
497 };
498 
499 int
500 rfs4_srvrinit(void)
501 {
502 	timespec32_t verf;
503 	int error;
504 	extern void rfs4_attr_init();
505 	extern krwlock_t rfs4_deleg_policy_lock;
506 
507 	/*
508 	 * The following algorithm attempts to find a unique verifier
509 	 * to be used as the write verifier returned from the server
510 	 * to the client.  It is important that this verifier change
511 	 * whenever the server reboots.  Of secondary importance, it
512 	 * is important for the verifier to be unique between two
513 	 * different servers.
514 	 *
515 	 * Thus, an attempt is made to use the system hostid and the
516 	 * current time in seconds when the nfssrv kernel module is
517 	 * loaded.  It is assumed that an NFS server will not be able
518 	 * to boot and then to reboot in less than a second.  If the
519 	 * hostid has not been set, then the current high resolution
520 	 * time is used.  This will ensure different verifiers each
521 	 * time the server reboots and minimize the chances that two
522 	 * different servers will have the same verifier.
523 	 * XXX - this is broken on LP64 kernels.
524 	 */
525 	verf.tv_sec = (time_t)zone_get_hostid(NULL);
526 	if (verf.tv_sec != 0) {
527 		verf.tv_nsec = gethrestime_sec();
528 	} else {
529 		timespec_t tverf;
530 
531 		gethrestime(&tverf);
532 		verf.tv_sec = (time_t)tverf.tv_sec;
533 		verf.tv_nsec = tverf.tv_nsec;
534 	}
535 
536 	Write4verf = *(uint64_t *)&verf;
537 
538 	rfs4_attr_init();
539 	mutex_init(&rfs4_deleg_lock, NULL, MUTEX_DEFAULT, NULL);
540 
541 	/* Used to manage create/destroy of server state */
542 	mutex_init(&rfs4_state_lock, NULL, MUTEX_DEFAULT, NULL);
543 
544 	/* Used to manage access to server instance linked list */
545 	mutex_init(&rfs4_servinst_lock, NULL, MUTEX_DEFAULT, NULL);
546 
547 	/* Used to manage access to rfs4_deleg_policy */
548 	rw_init(&rfs4_deleg_policy_lock, NULL, RW_DEFAULT, NULL);
549 
550 	error = fem_create("deleg_rdops", nfs4_rd_deleg_tmpl, &deleg_rdops);
551 	if (error != 0) {
552 		rfs4_disable_delegation();
553 	} else {
554 		error = fem_create("deleg_wrops", nfs4_wr_deleg_tmpl,
555 		    &deleg_wrops);
556 		if (error != 0) {
557 			rfs4_disable_delegation();
558 			fem_free(deleg_rdops);
559 		}
560 	}
561 
562 	nfs4_srv_caller_id = fs_new_caller_id();
563 
564 	lockt_sysid = lm_alloc_sysidt();
565 
566 	vsd_create(&nfs4_srv_vkey, NULL);
567 
568 	return (0);
569 }
570 
571 void
572 rfs4_srvrfini(void)
573 {
574 	extern krwlock_t rfs4_deleg_policy_lock;
575 
576 	if (lockt_sysid != LM_NOSYSID) {
577 		lm_free_sysidt(lockt_sysid);
578 		lockt_sysid = LM_NOSYSID;
579 	}
580 
581 	mutex_destroy(&rfs4_deleg_lock);
582 	mutex_destroy(&rfs4_state_lock);
583 	rw_destroy(&rfs4_deleg_policy_lock);
584 
585 	fem_free(deleg_rdops);
586 	fem_free(deleg_wrops);
587 }
588 
589 void
590 rfs4_init_compound_state(struct compound_state *cs)
591 {
592 	bzero(cs, sizeof (*cs));
593 	cs->cont = TRUE;
594 	cs->access = CS_ACCESS_DENIED;
595 	cs->deleg = FALSE;
596 	cs->mandlock = FALSE;
597 	cs->fh.nfs_fh4_val = cs->fhbuf;
598 }
599 
600 void
601 rfs4_grace_start(rfs4_servinst_t *sip)
602 {
603 	rw_enter(&sip->rwlock, RW_WRITER);
604 	sip->start_time = (time_t)TICK_TO_SEC(ddi_get_lbolt());
605 	sip->grace_period = rfs4_grace_period;
606 	rw_exit(&sip->rwlock);
607 }
608 
609 /*
610  * returns true if the instance's grace period has never been started
611  */
612 int
613 rfs4_servinst_grace_new(rfs4_servinst_t *sip)
614 {
615 	time_t start_time;
616 
617 	rw_enter(&sip->rwlock, RW_READER);
618 	start_time = sip->start_time;
619 	rw_exit(&sip->rwlock);
620 
621 	return (start_time == 0);
622 }
623 
624 /*
625  * Indicates if server instance is within the
626  * grace period.
627  */
628 int
629 rfs4_servinst_in_grace(rfs4_servinst_t *sip)
630 {
631 	time_t grace_expiry;
632 
633 	rw_enter(&sip->rwlock, RW_READER);
634 	grace_expiry = sip->start_time + sip->grace_period;
635 	rw_exit(&sip->rwlock);
636 
637 	return (((time_t)TICK_TO_SEC(ddi_get_lbolt())) < grace_expiry);
638 }
639 
640 int
641 rfs4_clnt_in_grace(rfs4_client_t *cp)
642 {
643 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
644 
645 	return (rfs4_servinst_in_grace(cp->rc_server_instance));
646 }
647 
648 /*
649  * reset all currently active grace periods
650  */
651 void
652 rfs4_grace_reset_all(void)
653 {
654 	rfs4_servinst_t *sip;
655 
656 	mutex_enter(&rfs4_servinst_lock);
657 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
658 		if (rfs4_servinst_in_grace(sip))
659 			rfs4_grace_start(sip);
660 	mutex_exit(&rfs4_servinst_lock);
661 }
662 
663 /*
664  * start any new instances' grace periods
665  */
666 void
667 rfs4_grace_start_new(void)
668 {
669 	rfs4_servinst_t *sip;
670 
671 	mutex_enter(&rfs4_servinst_lock);
672 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
673 		if (rfs4_servinst_grace_new(sip))
674 			rfs4_grace_start(sip);
675 	mutex_exit(&rfs4_servinst_lock);
676 }
677 
678 static rfs4_dss_path_t *
679 rfs4_dss_newpath(rfs4_servinst_t *sip, char *path, unsigned index)
680 {
681 	size_t len;
682 	rfs4_dss_path_t *dss_path;
683 
684 	dss_path = kmem_alloc(sizeof (rfs4_dss_path_t), KM_SLEEP);
685 
686 	/*
687 	 * Take a copy of the string, since the original may be overwritten.
688 	 * Sadly, no strdup() in the kernel.
689 	 */
690 	/* allow for NUL */
691 	len = strlen(path) + 1;
692 	dss_path->path = kmem_alloc(len, KM_SLEEP);
693 	(void) strlcpy(dss_path->path, path, len);
694 
695 	/* associate with servinst */
696 	dss_path->sip = sip;
697 	dss_path->index = index;
698 
699 	/*
700 	 * Add to list of served paths.
701 	 * No locking required, as we're only ever called at startup.
702 	 */
703 	if (rfs4_dss_pathlist == NULL) {
704 		/* this is the first dss_path_t */
705 
706 		/* needed for insque/remque */
707 		dss_path->next = dss_path->prev = dss_path;
708 
709 		rfs4_dss_pathlist = dss_path;
710 	} else {
711 		insque(dss_path, rfs4_dss_pathlist);
712 	}
713 
714 	return (dss_path);
715 }
716 
717 /*
718  * Create a new server instance, and make it the currently active instance.
719  * Note that starting the grace period too early will reduce the clients'
720  * recovery window.
721  */
722 void
723 rfs4_servinst_create(int start_grace, int dss_npaths, char **dss_paths)
724 {
725 	unsigned i;
726 	rfs4_servinst_t *sip;
727 	rfs4_oldstate_t *oldstate;
728 
729 	sip = kmem_alloc(sizeof (rfs4_servinst_t), KM_SLEEP);
730 	rw_init(&sip->rwlock, NULL, RW_DEFAULT, NULL);
731 
732 	sip->start_time = (time_t)0;
733 	sip->grace_period = (time_t)0;
734 	sip->next = NULL;
735 	sip->prev = NULL;
736 
737 	rw_init(&sip->oldstate_lock, NULL, RW_DEFAULT, NULL);
738 	/*
739 	 * This initial dummy entry is required to setup for insque/remque.
740 	 * It must be skipped over whenever the list is traversed.
741 	 */
742 	oldstate = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
743 	/* insque/remque require initial list entry to be self-terminated */
744 	oldstate->next = oldstate;
745 	oldstate->prev = oldstate;
746 	sip->oldstate = oldstate;
747 
748 
749 	sip->dss_npaths = dss_npaths;
750 	sip->dss_paths = kmem_alloc(dss_npaths *
751 	    sizeof (rfs4_dss_path_t *), KM_SLEEP);
752 
753 	for (i = 0; i < dss_npaths; i++) {
754 		sip->dss_paths[i] = rfs4_dss_newpath(sip, dss_paths[i], i);
755 	}
756 
757 	mutex_enter(&rfs4_servinst_lock);
758 	if (rfs4_cur_servinst != NULL) {
759 		/* add to linked list */
760 		sip->prev = rfs4_cur_servinst;
761 		rfs4_cur_servinst->next = sip;
762 	}
763 	if (start_grace)
764 		rfs4_grace_start(sip);
765 	/* make the new instance "current" */
766 	rfs4_cur_servinst = sip;
767 
768 	mutex_exit(&rfs4_servinst_lock);
769 }
770 
771 /*
772  * In future, we might add a rfs4_servinst_destroy(sip) but, for now, destroy
773  * all instances directly.
774  */
775 void
776 rfs4_servinst_destroy_all(void)
777 {
778 	rfs4_servinst_t *sip, *prev, *current;
779 #ifdef DEBUG
780 	int n = 0;
781 #endif
782 
783 	mutex_enter(&rfs4_servinst_lock);
784 	ASSERT(rfs4_cur_servinst != NULL);
785 	current = rfs4_cur_servinst;
786 	rfs4_cur_servinst = NULL;
787 	for (sip = current; sip != NULL; sip = prev) {
788 		prev = sip->prev;
789 		rw_destroy(&sip->rwlock);
790 		if (sip->oldstate)
791 			kmem_free(sip->oldstate, sizeof (rfs4_oldstate_t));
792 		if (sip->dss_paths)
793 			kmem_free(sip->dss_paths,
794 			    sip->dss_npaths * sizeof (rfs4_dss_path_t *));
795 		kmem_free(sip, sizeof (rfs4_servinst_t));
796 #ifdef DEBUG
797 		n++;
798 #endif
799 	}
800 	mutex_exit(&rfs4_servinst_lock);
801 }
802 
803 /*
804  * Assign the current server instance to a client_t.
805  * Should be called with cp->rc_dbe held.
806  */
807 void
808 rfs4_servinst_assign(rfs4_client_t *cp, rfs4_servinst_t *sip)
809 {
810 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
811 
812 	/*
813 	 * The lock ensures that if the current instance is in the process
814 	 * of changing, we will see the new one.
815 	 */
816 	mutex_enter(&rfs4_servinst_lock);
817 	cp->rc_server_instance = sip;
818 	mutex_exit(&rfs4_servinst_lock);
819 }
820 
821 rfs4_servinst_t *
822 rfs4_servinst(rfs4_client_t *cp)
823 {
824 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
825 
826 	return (cp->rc_server_instance);
827 }
828 
829 /* ARGSUSED */
830 static void
831 nullfree(caddr_t resop)
832 {
833 }
834 
835 /*
836  * This is a fall-through for invalid or not implemented (yet) ops
837  */
838 /* ARGSUSED */
839 static void
840 rfs4_op_inval(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
841 	struct compound_state *cs)
842 {
843 	*cs->statusp = *((nfsstat4 *)&(resop)->nfs_resop4_u) = NFS4ERR_INVAL;
844 }
845 
846 /*
847  * Check if the security flavor, nfsnum, is in the flavor_list.
848  */
849 bool_t
850 in_flavor_list(int nfsnum, int *flavor_list, int count)
851 {
852 	int i;
853 
854 	for (i = 0; i < count; i++) {
855 		if (nfsnum == flavor_list[i])
856 			return (TRUE);
857 	}
858 	return (FALSE);
859 }
860 
861 /*
862  * Used by rfs4_op_secinfo to get the security information from the
863  * export structure associated with the component.
864  */
865 /* ARGSUSED */
866 static nfsstat4
867 do_rfs4_op_secinfo(struct compound_state *cs, char *nm, SECINFO4res *resp)
868 {
869 	int error, different_export = 0;
870 	vnode_t *dvp, *vp, *tvp;
871 	struct exportinfo *exi = NULL;
872 	fid_t fid;
873 	uint_t count, i;
874 	secinfo4 *resok_val;
875 	struct secinfo *secp;
876 	seconfig_t *si;
877 	bool_t did_traverse = FALSE;
878 	int dotdot, walk;
879 
880 	dvp = cs->vp;
881 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
882 
883 	/*
884 	 * If dotdotting, then need to check whether it's above the
885 	 * root of a filesystem, or above an export point.
886 	 */
887 	if (dotdot) {
888 
889 		/*
890 		 * If dotdotting at the root of a filesystem, then
891 		 * need to traverse back to the mounted-on filesystem
892 		 * and do the dotdot lookup there.
893 		 */
894 		if (cs->vp->v_flag & VROOT) {
895 
896 			/*
897 			 * If at the system root, then can
898 			 * go up no further.
899 			 */
900 			if (VN_CMP(dvp, rootdir))
901 				return (puterrno4(ENOENT));
902 
903 			/*
904 			 * Traverse back to the mounted-on filesystem
905 			 */
906 			dvp = untraverse(cs->vp);
907 
908 			/*
909 			 * Set the different_export flag so we remember
910 			 * to pick up a new exportinfo entry for
911 			 * this new filesystem.
912 			 */
913 			different_export = 1;
914 		} else {
915 
916 			/*
917 			 * If dotdotting above an export point then set
918 			 * the different_export to get new export info.
919 			 */
920 			different_export = nfs_exported(cs->exi, cs->vp);
921 		}
922 	}
923 
924 	/*
925 	 * Get the vnode for the component "nm".
926 	 */
927 	error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cs->cr,
928 	    NULL, NULL, NULL);
929 	if (error)
930 		return (puterrno4(error));
931 
932 	/*
933 	 * If the vnode is in a pseudo filesystem, or if the security flavor
934 	 * used in the request is valid but not an explicitly shared flavor,
935 	 * or the access bit indicates that this is a limited access,
936 	 * check whether this vnode is visible.
937 	 */
938 	if (!different_export &&
939 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
940 	    cs->access & CS_ACCESS_LIMITED)) {
941 		if (! nfs_visible(cs->exi, vp, &different_export)) {
942 			VN_RELE(vp);
943 			return (puterrno4(ENOENT));
944 		}
945 	}
946 
947 	/*
948 	 * If it's a mountpoint, then traverse it.
949 	 */
950 	if (vn_ismntpt(vp)) {
951 		tvp = vp;
952 		if ((error = traverse(&tvp)) != 0) {
953 			VN_RELE(vp);
954 			return (puterrno4(error));
955 		}
956 		/* remember that we had to traverse mountpoint */
957 		did_traverse = TRUE;
958 		vp = tvp;
959 		different_export = 1;
960 	} else if (vp->v_vfsp != dvp->v_vfsp) {
961 		/*
962 		 * If vp isn't a mountpoint and the vfs ptrs aren't the same,
963 		 * then vp is probably an LOFS object.  We don't need the
964 		 * realvp, we just need to know that we might have crossed
965 		 * a server fs boundary and need to call checkexport4.
966 		 * (LOFS lookup hides server fs mountpoints, and actually calls
967 		 * traverse)
968 		 */
969 		different_export = 1;
970 	}
971 
972 	/*
973 	 * Get the export information for it.
974 	 */
975 	if (different_export) {
976 
977 		bzero(&fid, sizeof (fid));
978 		fid.fid_len = MAXFIDSZ;
979 		error = vop_fid_pseudo(vp, &fid);
980 		if (error) {
981 			VN_RELE(vp);
982 			return (puterrno4(error));
983 		}
984 
985 		if (dotdot)
986 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
987 		else
988 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
989 
990 		if (exi == NULL) {
991 			if (did_traverse == TRUE) {
992 				/*
993 				 * If this vnode is a mounted-on vnode,
994 				 * but the mounted-on file system is not
995 				 * exported, send back the secinfo for
996 				 * the exported node that the mounted-on
997 				 * vnode lives in.
998 				 */
999 				exi = cs->exi;
1000 			} else {
1001 				VN_RELE(vp);
1002 				return (puterrno4(EACCES));
1003 			}
1004 		}
1005 	} else {
1006 		exi = cs->exi;
1007 	}
1008 	ASSERT(exi != NULL);
1009 
1010 
1011 	/*
1012 	 * Create the secinfo result based on the security information
1013 	 * from the exportinfo structure (exi).
1014 	 *
1015 	 * Return all flavors for a pseudo node.
1016 	 * For a real export node, return the flavor that the client
1017 	 * has access with.
1018 	 */
1019 	ASSERT(RW_LOCK_HELD(&exported_lock));
1020 	if (PSEUDO(exi)) {
1021 		count = exi->exi_export.ex_seccnt; /* total sec count */
1022 		resok_val = kmem_alloc(count * sizeof (secinfo4), KM_SLEEP);
1023 		secp = exi->exi_export.ex_secinfo;
1024 
1025 		for (i = 0; i < count; i++) {
1026 			si = &secp[i].s_secinfo;
1027 			resok_val[i].flavor = si->sc_rpcnum;
1028 			if (resok_val[i].flavor == RPCSEC_GSS) {
1029 				rpcsec_gss_info *info;
1030 
1031 				info = &resok_val[i].flavor_info;
1032 				info->qop = si->sc_qop;
1033 				info->service = (rpc_gss_svc_t)si->sc_service;
1034 
1035 				/* get oid opaque data */
1036 				info->oid.sec_oid4_len =
1037 				    si->sc_gss_mech_type->length;
1038 				info->oid.sec_oid4_val = kmem_alloc(
1039 				    si->sc_gss_mech_type->length, KM_SLEEP);
1040 				bcopy(
1041 				    si->sc_gss_mech_type->elements,
1042 				    info->oid.sec_oid4_val,
1043 				    info->oid.sec_oid4_len);
1044 			}
1045 		}
1046 		resp->SECINFO4resok_len = count;
1047 		resp->SECINFO4resok_val = resok_val;
1048 	} else {
1049 		int ret_cnt = 0, k = 0;
1050 		int *flavor_list;
1051 
1052 		count = exi->exi_export.ex_seccnt; /* total sec count */
1053 		secp = exi->exi_export.ex_secinfo;
1054 
1055 		flavor_list = kmem_alloc(count * sizeof (int), KM_SLEEP);
1056 		/* find out which flavors to return */
1057 		for (i = 0; i < count; i ++) {
1058 			int access, flavor, perm;
1059 
1060 			flavor = secp[i].s_secinfo.sc_nfsnum;
1061 			perm = secp[i].s_flags;
1062 
1063 			access = nfsauth4_secinfo_access(exi, cs->req,
1064 			    flavor, perm);
1065 
1066 			if (! (access & NFSAUTH_DENIED) &&
1067 			    ! (access & NFSAUTH_WRONGSEC)) {
1068 				flavor_list[ret_cnt] = flavor;
1069 				ret_cnt++;
1070 			}
1071 		}
1072 
1073 		/* Create the returning SECINFO value */
1074 		resok_val = kmem_alloc(ret_cnt * sizeof (secinfo4), KM_SLEEP);
1075 
1076 		for (i = 0; i < count; i++) {
1077 			/*
1078 			 * If the flavor is in the flavor list,
1079 			 * fill in resok_val.
1080 			 */
1081 			si = &secp[i].s_secinfo;
1082 			if (in_flavor_list(si->sc_nfsnum,
1083 			    flavor_list, ret_cnt)) {
1084 				resok_val[k].flavor = si->sc_rpcnum;
1085 				if (resok_val[k].flavor == RPCSEC_GSS) {
1086 					rpcsec_gss_info *info;
1087 
1088 					info = &resok_val[k].flavor_info;
1089 					info->qop = si->sc_qop;
1090 					info->service = (rpc_gss_svc_t)
1091 					    si->sc_service;
1092 
1093 					/* get oid opaque data */
1094 					info->oid.sec_oid4_len =
1095 					    si->sc_gss_mech_type->length;
1096 					info->oid.sec_oid4_val = kmem_alloc(
1097 					    si->sc_gss_mech_type->length,
1098 					    KM_SLEEP);
1099 					bcopy(si->sc_gss_mech_type->elements,
1100 					    info->oid.sec_oid4_val,
1101 					    info->oid.sec_oid4_len);
1102 				}
1103 				k++;
1104 			}
1105 			if (k >= ret_cnt)
1106 				break;
1107 		}
1108 		resp->SECINFO4resok_len = ret_cnt;
1109 		resp->SECINFO4resok_val = resok_val;
1110 		kmem_free(flavor_list, count * sizeof (int));
1111 	}
1112 
1113 	VN_RELE(vp);
1114 	return (NFS4_OK);
1115 }
1116 
1117 /*
1118  * SECINFO (Operation 33): Obtain required security information on
1119  * the component name in the format of (security-mechanism-oid, qop, service)
1120  * triplets.
1121  */
1122 /* ARGSUSED */
1123 static void
1124 rfs4_op_secinfo(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1125     struct compound_state *cs)
1126 {
1127 	SECINFO4args *args = &argop->nfs_argop4_u.opsecinfo;
1128 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1129 	utf8string *utfnm = &args->name;
1130 	uint_t len;
1131 	char *nm;
1132 	struct sockaddr *ca;
1133 	char *name = NULL;
1134 
1135 	DTRACE_NFSV4_2(op__secinfo__start, struct compound_state *, cs,
1136 	    SECINFO4args *, args);
1137 
1138 	/*
1139 	 * Current file handle (cfh) should have been set before getting
1140 	 * into this function. If not, return error.
1141 	 */
1142 	if (cs->vp == NULL) {
1143 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1144 		goto out;
1145 	}
1146 
1147 	if (cs->vp->v_type != VDIR) {
1148 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1149 		goto out;
1150 	}
1151 
1152 	/*
1153 	 * Verify the component name. If failed, error out, but
1154 	 * do not error out if the component name is a "..".
1155 	 * SECINFO will return its parents secinfo data for SECINFO "..".
1156 	 */
1157 	if (!utf8_dir_verify(utfnm)) {
1158 		if (utfnm->utf8string_len != 2 ||
1159 		    utfnm->utf8string_val[0] != '.' ||
1160 		    utfnm->utf8string_val[1] != '.') {
1161 			*cs->statusp = resp->status = NFS4ERR_INVAL;
1162 			goto out;
1163 		}
1164 	}
1165 
1166 	nm = utf8_to_str(utfnm, &len, NULL);
1167 	if (nm == NULL) {
1168 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1169 		goto out;
1170 	}
1171 
1172 	if (len > MAXNAMELEN) {
1173 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1174 		kmem_free(nm, len);
1175 		goto out;
1176 	}
1177 
1178 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1179 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1180 	    MAXPATHLEN  + 1);
1181 
1182 	if (name == NULL) {
1183 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1184 		kmem_free(nm, len);
1185 		goto out;
1186 	}
1187 
1188 
1189 	*cs->statusp = resp->status = do_rfs4_op_secinfo(cs, name, resp);
1190 
1191 	if (name != nm)
1192 		kmem_free(name, MAXPATHLEN + 1);
1193 	kmem_free(nm, len);
1194 
1195 out:
1196 	DTRACE_NFSV4_2(op__secinfo__done, struct compound_state *, cs,
1197 	    SECINFO4res *, resp);
1198 }
1199 
1200 /*
1201  * Free SECINFO result.
1202  */
1203 /* ARGSUSED */
1204 static void
1205 rfs4_op_secinfo_free(nfs_resop4 *resop)
1206 {
1207 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1208 	int count, i;
1209 	secinfo4 *resok_val;
1210 
1211 	/* If this is not an Ok result, nothing to free. */
1212 	if (resp->status != NFS4_OK) {
1213 		return;
1214 	}
1215 
1216 	count = resp->SECINFO4resok_len;
1217 	resok_val = resp->SECINFO4resok_val;
1218 
1219 	for (i = 0; i < count; i++) {
1220 		if (resok_val[i].flavor == RPCSEC_GSS) {
1221 			rpcsec_gss_info *info;
1222 
1223 			info = &resok_val[i].flavor_info;
1224 			kmem_free(info->oid.sec_oid4_val,
1225 			    info->oid.sec_oid4_len);
1226 		}
1227 	}
1228 	kmem_free(resok_val, count * sizeof (secinfo4));
1229 	resp->SECINFO4resok_len = 0;
1230 	resp->SECINFO4resok_val = NULL;
1231 }
1232 
1233 /* ARGSUSED */
1234 static void
1235 rfs4_op_access(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1236     struct compound_state *cs)
1237 {
1238 	ACCESS4args *args = &argop->nfs_argop4_u.opaccess;
1239 	ACCESS4res *resp = &resop->nfs_resop4_u.opaccess;
1240 	int error;
1241 	vnode_t *vp;
1242 	struct vattr va;
1243 	int checkwriteperm;
1244 	cred_t *cr = cs->cr;
1245 	bslabel_t *clabel, *slabel;
1246 	ts_label_t *tslabel;
1247 	boolean_t admin_low_client;
1248 
1249 	DTRACE_NFSV4_2(op__access__start, struct compound_state *, cs,
1250 	    ACCESS4args *, args);
1251 
1252 #if 0	/* XXX allow access even if !cs->access. Eventually only pseudo fs */
1253 	if (cs->access == CS_ACCESS_DENIED) {
1254 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1255 		goto out;
1256 	}
1257 #endif
1258 	if (cs->vp == NULL) {
1259 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1260 		goto out;
1261 	}
1262 
1263 	ASSERT(cr != NULL);
1264 
1265 	vp = cs->vp;
1266 
1267 	/*
1268 	 * If the file system is exported read only, it is not appropriate
1269 	 * to check write permissions for regular files and directories.
1270 	 * Special files are interpreted by the client, so the underlying
1271 	 * permissions are sent back to the client for interpretation.
1272 	 */
1273 	if (rdonly4(cs->exi, cs->vp, req) &&
1274 	    (vp->v_type == VREG || vp->v_type == VDIR))
1275 		checkwriteperm = 0;
1276 	else
1277 		checkwriteperm = 1;
1278 
1279 	/*
1280 	 * XXX
1281 	 * We need the mode so that we can correctly determine access
1282 	 * permissions relative to a mandatory lock file.  Access to
1283 	 * mandatory lock files is denied on the server, so it might
1284 	 * as well be reflected to the server during the open.
1285 	 */
1286 	va.va_mask = AT_MODE;
1287 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1288 	if (error) {
1289 		*cs->statusp = resp->status = puterrno4(error);
1290 		goto out;
1291 	}
1292 	resp->access = 0;
1293 	resp->supported = 0;
1294 
1295 	if (is_system_labeled()) {
1296 		ASSERT(req->rq_label != NULL);
1297 		clabel = req->rq_label;
1298 		DTRACE_PROBE2(tx__rfs4__log__info__opaccess__clabel, char *,
1299 		    "got client label from request(1)",
1300 		    struct svc_req *, req);
1301 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
1302 			if ((tslabel = nfs_getflabel(vp, cs->exi)) == NULL) {
1303 				*cs->statusp = resp->status = puterrno4(EACCES);
1304 				goto out;
1305 			}
1306 			slabel = label2bslabel(tslabel);
1307 			DTRACE_PROBE3(tx__rfs4__log__info__opaccess__slabel,
1308 			    char *, "got server label(1) for vp(2)",
1309 			    bslabel_t *, slabel, vnode_t *, vp);
1310 
1311 			admin_low_client = B_FALSE;
1312 		} else
1313 			admin_low_client = B_TRUE;
1314 	}
1315 
1316 	if (args->access & ACCESS4_READ) {
1317 		error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
1318 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1319 		    (!is_system_labeled() || admin_low_client ||
1320 		    bldominates(clabel, slabel)))
1321 			resp->access |= ACCESS4_READ;
1322 		resp->supported |= ACCESS4_READ;
1323 	}
1324 	if ((args->access & ACCESS4_LOOKUP) && vp->v_type == VDIR) {
1325 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1326 		if (!error && (!is_system_labeled() || admin_low_client ||
1327 		    bldominates(clabel, slabel)))
1328 			resp->access |= ACCESS4_LOOKUP;
1329 		resp->supported |= ACCESS4_LOOKUP;
1330 	}
1331 	if (checkwriteperm &&
1332 	    (args->access & (ACCESS4_MODIFY|ACCESS4_EXTEND))) {
1333 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1334 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1335 		    (!is_system_labeled() || admin_low_client ||
1336 		    blequal(clabel, slabel)))
1337 			resp->access |=
1338 			    (args->access & (ACCESS4_MODIFY | ACCESS4_EXTEND));
1339 		resp->supported |= (ACCESS4_MODIFY | ACCESS4_EXTEND);
1340 	}
1341 
1342 	if (checkwriteperm &&
1343 	    (args->access & ACCESS4_DELETE) && vp->v_type == VDIR) {
1344 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1345 		if (!error && (!is_system_labeled() || admin_low_client ||
1346 		    blequal(clabel, slabel)))
1347 			resp->access |= ACCESS4_DELETE;
1348 		resp->supported |= ACCESS4_DELETE;
1349 	}
1350 	if (args->access & ACCESS4_EXECUTE && vp->v_type != VDIR) {
1351 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1352 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1353 		    (!is_system_labeled() || admin_low_client ||
1354 		    bldominates(clabel, slabel)))
1355 			resp->access |= ACCESS4_EXECUTE;
1356 		resp->supported |= ACCESS4_EXECUTE;
1357 	}
1358 
1359 	if (is_system_labeled() && !admin_low_client)
1360 		label_rele(tslabel);
1361 
1362 	*cs->statusp = resp->status = NFS4_OK;
1363 out:
1364 	DTRACE_NFSV4_2(op__access__done, struct compound_state *, cs,
1365 	    ACCESS4res *, resp);
1366 }
1367 
1368 /* ARGSUSED */
1369 static void
1370 rfs4_op_commit(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1371     struct compound_state *cs)
1372 {
1373 	COMMIT4args *args = &argop->nfs_argop4_u.opcommit;
1374 	COMMIT4res *resp = &resop->nfs_resop4_u.opcommit;
1375 	int error;
1376 	vnode_t *vp = cs->vp;
1377 	cred_t *cr = cs->cr;
1378 	vattr_t va;
1379 
1380 	DTRACE_NFSV4_2(op__commit__start, struct compound_state *, cs,
1381 	    COMMIT4args *, args);
1382 
1383 	if (vp == NULL) {
1384 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1385 		goto out;
1386 	}
1387 	if (cs->access == CS_ACCESS_DENIED) {
1388 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1389 		goto out;
1390 	}
1391 
1392 	if (args->offset + args->count < args->offset) {
1393 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1394 		goto out;
1395 	}
1396 
1397 	va.va_mask = AT_UID;
1398 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1399 
1400 	/*
1401 	 * If we can't get the attributes, then we can't do the
1402 	 * right access checking.  So, we'll fail the request.
1403 	 */
1404 	if (error) {
1405 		*cs->statusp = resp->status = puterrno4(error);
1406 		goto out;
1407 	}
1408 	if (rdonly4(cs->exi, cs->vp, req)) {
1409 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1410 		goto out;
1411 	}
1412 
1413 	if (vp->v_type != VREG) {
1414 		if (vp->v_type == VDIR)
1415 			resp->status = NFS4ERR_ISDIR;
1416 		else
1417 			resp->status = NFS4ERR_INVAL;
1418 		*cs->statusp = resp->status;
1419 		goto out;
1420 	}
1421 
1422 	if (crgetuid(cr) != va.va_uid &&
1423 	    (error = VOP_ACCESS(vp, VWRITE, 0, cs->cr, NULL))) {
1424 		*cs->statusp = resp->status = puterrno4(error);
1425 		goto out;
1426 	}
1427 
1428 	error = VOP_FSYNC(vp, FSYNC, cr, NULL);
1429 
1430 	if (error) {
1431 		*cs->statusp = resp->status = puterrno4(error);
1432 		goto out;
1433 	}
1434 
1435 	*cs->statusp = resp->status = NFS4_OK;
1436 	resp->writeverf = Write4verf;
1437 out:
1438 	DTRACE_NFSV4_2(op__commit__done, struct compound_state *, cs,
1439 	    COMMIT4res *, resp);
1440 }
1441 
1442 /*
1443  * rfs4_op_mknod is called from rfs4_op_create after all initial verification
1444  * was completed. It does the nfsv4 create for special files.
1445  */
1446 /* ARGSUSED */
1447 static vnode_t *
1448 do_rfs4_op_mknod(CREATE4args *args, CREATE4res *resp, struct svc_req *req,
1449     struct compound_state *cs, vattr_t *vap, char *nm)
1450 {
1451 	int error;
1452 	cred_t *cr = cs->cr;
1453 	vnode_t *dvp = cs->vp;
1454 	vnode_t *vp = NULL;
1455 	int mode;
1456 	enum vcexcl excl;
1457 
1458 	switch (args->type) {
1459 	case NF4CHR:
1460 	case NF4BLK:
1461 		if (secpolicy_sys_devices(cr) != 0) {
1462 			*cs->statusp = resp->status = NFS4ERR_PERM;
1463 			return (NULL);
1464 		}
1465 		if (args->type == NF4CHR)
1466 			vap->va_type = VCHR;
1467 		else
1468 			vap->va_type = VBLK;
1469 		vap->va_rdev = makedevice(args->ftype4_u.devdata.specdata1,
1470 		    args->ftype4_u.devdata.specdata2);
1471 		vap->va_mask |= AT_RDEV;
1472 		break;
1473 	case NF4SOCK:
1474 		vap->va_type = VSOCK;
1475 		break;
1476 	case NF4FIFO:
1477 		vap->va_type = VFIFO;
1478 		break;
1479 	default:
1480 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1481 		return (NULL);
1482 	}
1483 
1484 	/*
1485 	 * Must specify the mode.
1486 	 */
1487 	if (!(vap->va_mask & AT_MODE)) {
1488 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1489 		return (NULL);
1490 	}
1491 
1492 	excl = EXCL;
1493 
1494 	mode = 0;
1495 
1496 	error = VOP_CREATE(dvp, nm, vap, excl, mode, &vp, cr, 0, NULL, NULL);
1497 	if (error) {
1498 		*cs->statusp = resp->status = puterrno4(error);
1499 		return (NULL);
1500 	}
1501 	return (vp);
1502 }
1503 
1504 /*
1505  * nfsv4 create is used to create non-regular files. For regular files,
1506  * use nfsv4 open.
1507  */
1508 /* ARGSUSED */
1509 static void
1510 rfs4_op_create(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1511     struct compound_state *cs)
1512 {
1513 	CREATE4args *args = &argop->nfs_argop4_u.opcreate;
1514 	CREATE4res *resp = &resop->nfs_resop4_u.opcreate;
1515 	int error;
1516 	struct vattr bva, iva, iva2, ava, *vap;
1517 	cred_t *cr = cs->cr;
1518 	vnode_t *dvp = cs->vp;
1519 	vnode_t *vp = NULL;
1520 	vnode_t *realvp;
1521 	char *nm, *lnm;
1522 	uint_t len, llen;
1523 	int syncval = 0;
1524 	struct nfs4_svgetit_arg sarg;
1525 	struct nfs4_ntov_table ntov;
1526 	struct statvfs64 sb;
1527 	nfsstat4 status;
1528 	struct sockaddr *ca;
1529 	char *name = NULL;
1530 	char *lname = NULL;
1531 
1532 	DTRACE_NFSV4_2(op__create__start, struct compound_state *, cs,
1533 	    CREATE4args *, args);
1534 
1535 	resp->attrset = 0;
1536 
1537 	if (dvp == NULL) {
1538 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1539 		goto out;
1540 	}
1541 
1542 	/*
1543 	 * If there is an unshared filesystem mounted on this vnode,
1544 	 * do not allow to create an object in this directory.
1545 	 */
1546 	if (vn_ismntpt(dvp)) {
1547 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1548 		goto out;
1549 	}
1550 
1551 	/* Verify that type is correct */
1552 	switch (args->type) {
1553 	case NF4LNK:
1554 	case NF4BLK:
1555 	case NF4CHR:
1556 	case NF4SOCK:
1557 	case NF4FIFO:
1558 	case NF4DIR:
1559 		break;
1560 	default:
1561 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1562 		goto out;
1563 	};
1564 
1565 	if (cs->access == CS_ACCESS_DENIED) {
1566 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1567 		goto out;
1568 	}
1569 	if (dvp->v_type != VDIR) {
1570 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1571 		goto out;
1572 	}
1573 	if (!utf8_dir_verify(&args->objname)) {
1574 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1575 		goto out;
1576 	}
1577 
1578 	if (rdonly4(cs->exi, cs->vp, req)) {
1579 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1580 		goto out;
1581 	}
1582 
1583 	/*
1584 	 * Name of newly created object
1585 	 */
1586 	nm = utf8_to_fn(&args->objname, &len, NULL);
1587 	if (nm == NULL) {
1588 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1589 		goto out;
1590 	}
1591 
1592 	if (len > MAXNAMELEN) {
1593 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1594 		kmem_free(nm, len);
1595 		goto out;
1596 	}
1597 
1598 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1599 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1600 	    MAXPATHLEN  + 1);
1601 
1602 	if (name == NULL) {
1603 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1604 		kmem_free(nm, len);
1605 		goto out;
1606 	}
1607 
1608 	resp->attrset = 0;
1609 
1610 	sarg.sbp = &sb;
1611 	sarg.is_referral = B_FALSE;
1612 	nfs4_ntov_table_init(&ntov);
1613 
1614 	status = do_rfs4_set_attrs(&resp->attrset,
1615 	    &args->createattrs, cs, &sarg, &ntov, NFS4ATTR_SETIT);
1616 
1617 	if (sarg.vap->va_mask == 0 && status == NFS4_OK)
1618 		status = NFS4ERR_INVAL;
1619 
1620 	if (status != NFS4_OK) {
1621 		*cs->statusp = resp->status = status;
1622 		if (name != nm)
1623 			kmem_free(name, MAXPATHLEN + 1);
1624 		kmem_free(nm, len);
1625 		nfs4_ntov_table_free(&ntov, &sarg);
1626 		resp->attrset = 0;
1627 		goto out;
1628 	}
1629 
1630 	/* Get "before" change value */
1631 	bva.va_mask = AT_CTIME|AT_SEQ|AT_MODE;
1632 	error = VOP_GETATTR(dvp, &bva, 0, cr, NULL);
1633 	if (error) {
1634 		*cs->statusp = resp->status = puterrno4(error);
1635 		if (name != nm)
1636 			kmem_free(name, MAXPATHLEN + 1);
1637 		kmem_free(nm, len);
1638 		nfs4_ntov_table_free(&ntov, &sarg);
1639 		resp->attrset = 0;
1640 		goto out;
1641 	}
1642 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bva.va_ctime)
1643 
1644 	vap = sarg.vap;
1645 
1646 	/*
1647 	 * Set the default initial values for attributes when the parent
1648 	 * directory does not have the VSUID/VSGID bit set and they have
1649 	 * not been specified in createattrs.
1650 	 */
1651 	if (!(bva.va_mode & VSUID) && (vap->va_mask & AT_UID) == 0) {
1652 		vap->va_uid = crgetuid(cr);
1653 		vap->va_mask |= AT_UID;
1654 	}
1655 	if (!(bva.va_mode & VSGID) && (vap->va_mask & AT_GID) == 0) {
1656 		vap->va_gid = crgetgid(cr);
1657 		vap->va_mask |= AT_GID;
1658 	}
1659 
1660 	vap->va_mask |= AT_TYPE;
1661 	switch (args->type) {
1662 	case NF4DIR:
1663 		vap->va_type = VDIR;
1664 		if ((vap->va_mask & AT_MODE) == 0) {
1665 			vap->va_mode = 0700;	/* default: owner rwx only */
1666 			vap->va_mask |= AT_MODE;
1667 		}
1668 		error = VOP_MKDIR(dvp, name, vap, &vp, cr, NULL, 0, NULL);
1669 		if (error)
1670 			break;
1671 
1672 		/*
1673 		 * Get the initial "after" sequence number, if it fails,
1674 		 * set to zero
1675 		 */
1676 		iva.va_mask = AT_SEQ;
1677 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1678 			iva.va_seq = 0;
1679 		break;
1680 	case NF4LNK:
1681 		vap->va_type = VLNK;
1682 		if ((vap->va_mask & AT_MODE) == 0) {
1683 			vap->va_mode = 0700;	/* default: owner rwx only */
1684 			vap->va_mask |= AT_MODE;
1685 		}
1686 
1687 		/*
1688 		 * symlink names must be treated as data
1689 		 */
1690 		lnm = utf8_to_str(&args->ftype4_u.linkdata, &llen, NULL);
1691 
1692 		if (lnm == NULL) {
1693 			*cs->statusp = resp->status = NFS4ERR_INVAL;
1694 			if (name != nm)
1695 				kmem_free(name, MAXPATHLEN + 1);
1696 			kmem_free(nm, len);
1697 			nfs4_ntov_table_free(&ntov, &sarg);
1698 			resp->attrset = 0;
1699 			goto out;
1700 		}
1701 
1702 		if (llen > MAXPATHLEN) {
1703 			*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1704 			if (name != nm)
1705 				kmem_free(name, MAXPATHLEN + 1);
1706 			kmem_free(nm, len);
1707 			kmem_free(lnm, llen);
1708 			nfs4_ntov_table_free(&ntov, &sarg);
1709 			resp->attrset = 0;
1710 			goto out;
1711 		}
1712 
1713 		lname = nfscmd_convname(ca, cs->exi, lnm,
1714 		    NFSCMD_CONV_INBOUND, MAXPATHLEN  + 1);
1715 
1716 		if (lname == NULL) {
1717 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
1718 			if (name != nm)
1719 				kmem_free(name, MAXPATHLEN + 1);
1720 			kmem_free(nm, len);
1721 			kmem_free(lnm, llen);
1722 			nfs4_ntov_table_free(&ntov, &sarg);
1723 			resp->attrset = 0;
1724 			goto out;
1725 		}
1726 
1727 		error = VOP_SYMLINK(dvp, name, vap, lname, cr, NULL, 0);
1728 		if (lname != lnm)
1729 			kmem_free(lname, MAXPATHLEN + 1);
1730 		kmem_free(lnm, llen);
1731 		if (error)
1732 			break;
1733 
1734 		/*
1735 		 * Get the initial "after" sequence number, if it fails,
1736 		 * set to zero
1737 		 */
1738 		iva.va_mask = AT_SEQ;
1739 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1740 			iva.va_seq = 0;
1741 
1742 		error = VOP_LOOKUP(dvp, name, &vp, NULL, 0, NULL, cr,
1743 		    NULL, NULL, NULL);
1744 		if (error)
1745 			break;
1746 
1747 		/*
1748 		 * va_seq is not safe over VOP calls, check it again
1749 		 * if it has changed zero out iva to force atomic = FALSE.
1750 		 */
1751 		iva2.va_mask = AT_SEQ;
1752 		if (VOP_GETATTR(dvp, &iva2, 0, cs->cr, NULL) ||
1753 		    iva2.va_seq != iva.va_seq)
1754 			iva.va_seq = 0;
1755 		break;
1756 	default:
1757 		/*
1758 		 * probably a special file.
1759 		 */
1760 		if ((vap->va_mask & AT_MODE) == 0) {
1761 			vap->va_mode = 0600;	/* default: owner rw only */
1762 			vap->va_mask |= AT_MODE;
1763 		}
1764 		syncval = FNODSYNC;
1765 		/*
1766 		 * We know this will only generate one VOP call
1767 		 */
1768 		vp = do_rfs4_op_mknod(args, resp, req, cs, vap, name);
1769 
1770 		if (vp == NULL) {
1771 			if (name != nm)
1772 				kmem_free(name, MAXPATHLEN + 1);
1773 			kmem_free(nm, len);
1774 			nfs4_ntov_table_free(&ntov, &sarg);
1775 			resp->attrset = 0;
1776 			goto out;
1777 		}
1778 
1779 		/*
1780 		 * Get the initial "after" sequence number, if it fails,
1781 		 * set to zero
1782 		 */
1783 		iva.va_mask = AT_SEQ;
1784 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1785 			iva.va_seq = 0;
1786 
1787 		break;
1788 	}
1789 	if (name != nm)
1790 		kmem_free(name, MAXPATHLEN + 1);
1791 	kmem_free(nm, len);
1792 
1793 	if (error) {
1794 		*cs->statusp = resp->status = puterrno4(error);
1795 	}
1796 
1797 	/*
1798 	 * Force modified data and metadata out to stable storage.
1799 	 */
1800 	(void) VOP_FSYNC(dvp, 0, cr, NULL);
1801 
1802 	if (resp->status != NFS4_OK) {
1803 		if (vp != NULL)
1804 			VN_RELE(vp);
1805 		nfs4_ntov_table_free(&ntov, &sarg);
1806 		resp->attrset = 0;
1807 		goto out;
1808 	}
1809 
1810 	/*
1811 	 * Finish setup of cinfo response, "before" value already set.
1812 	 * Get "after" change value, if it fails, simply return the
1813 	 * before value.
1814 	 */
1815 	ava.va_mask = AT_CTIME|AT_SEQ;
1816 	if (VOP_GETATTR(dvp, &ava, 0, cr, NULL)) {
1817 		ava.va_ctime = bva.va_ctime;
1818 		ava.va_seq = 0;
1819 	}
1820 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, ava.va_ctime);
1821 
1822 	/*
1823 	 * True verification that object was created with correct
1824 	 * attrs is impossible.  The attrs could have been changed
1825 	 * immediately after object creation.  If attributes did
1826 	 * not verify, the only recourse for the server is to
1827 	 * destroy the object.  Maybe if some attrs (like gid)
1828 	 * are set incorrectly, the object should be destroyed;
1829 	 * however, seems bad as a default policy.  Do we really
1830 	 * want to destroy an object over one of the times not
1831 	 * verifying correctly?  For these reasons, the server
1832 	 * currently sets bits in attrset for createattrs
1833 	 * that were set; however, no verification is done.
1834 	 *
1835 	 * vmask_to_nmask accounts for vattr bits set on create
1836 	 *	[do_rfs4_set_attrs() only sets resp bits for
1837 	 *	 non-vattr/vfs bits.]
1838 	 * Mask off any bits set by default so as not to return
1839 	 * more attrset bits than were requested in createattrs
1840 	 */
1841 	nfs4_vmask_to_nmask(sarg.vap->va_mask, &resp->attrset);
1842 	resp->attrset &= args->createattrs.attrmask;
1843 	nfs4_ntov_table_free(&ntov, &sarg);
1844 
1845 	error = makefh4(&cs->fh, vp, cs->exi);
1846 	if (error) {
1847 		*cs->statusp = resp->status = puterrno4(error);
1848 	}
1849 
1850 	/*
1851 	 * The cinfo.atomic = TRUE only if we got no errors, we have
1852 	 * non-zero va_seq's, and it has incremented by exactly one
1853 	 * during the creation and it didn't change during the VOP_LOOKUP
1854 	 * or VOP_FSYNC.
1855 	 */
1856 	if (!error && bva.va_seq && iva.va_seq && ava.va_seq &&
1857 	    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
1858 		resp->cinfo.atomic = TRUE;
1859 	else
1860 		resp->cinfo.atomic = FALSE;
1861 
1862 	/*
1863 	 * Force modified metadata out to stable storage.
1864 	 *
1865 	 * if a underlying vp exists, pass it to VOP_FSYNC
1866 	 */
1867 	if (VOP_REALVP(vp, &realvp, NULL) == 0)
1868 		(void) VOP_FSYNC(realvp, syncval, cr, NULL);
1869 	else
1870 		(void) VOP_FSYNC(vp, syncval, cr, NULL);
1871 
1872 	if (resp->status != NFS4_OK) {
1873 		VN_RELE(vp);
1874 		goto out;
1875 	}
1876 	if (cs->vp)
1877 		VN_RELE(cs->vp);
1878 
1879 	cs->vp = vp;
1880 	*cs->statusp = resp->status = NFS4_OK;
1881 out:
1882 	DTRACE_NFSV4_2(op__create__done, struct compound_state *, cs,
1883 	    CREATE4res *, resp);
1884 }
1885 
1886 /*ARGSUSED*/
1887 static void
1888 rfs4_op_delegpurge(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1889     struct compound_state *cs)
1890 {
1891 	DTRACE_NFSV4_2(op__delegpurge__start, struct compound_state *, cs,
1892 	    DELEGPURGE4args *, &argop->nfs_argop4_u.opdelegpurge);
1893 
1894 	rfs4_op_inval(argop, resop, req, cs);
1895 
1896 	DTRACE_NFSV4_2(op__delegpurge__done, struct compound_state *, cs,
1897 	    DELEGPURGE4res *, &resop->nfs_resop4_u.opdelegpurge);
1898 }
1899 
1900 /*ARGSUSED*/
1901 static void
1902 rfs4_op_delegreturn(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1903     struct compound_state *cs)
1904 {
1905 	DELEGRETURN4args *args = &argop->nfs_argop4_u.opdelegreturn;
1906 	DELEGRETURN4res *resp = &resop->nfs_resop4_u.opdelegreturn;
1907 	rfs4_deleg_state_t *dsp;
1908 	nfsstat4 status;
1909 
1910 	DTRACE_NFSV4_2(op__delegreturn__start, struct compound_state *, cs,
1911 	    DELEGRETURN4args *, args);
1912 
1913 	status = rfs4_get_deleg_state(&args->deleg_stateid, &dsp);
1914 	resp->status = *cs->statusp = status;
1915 	if (status != NFS4_OK)
1916 		goto out;
1917 
1918 	/* Ensure specified filehandle matches */
1919 	if (cs->vp != dsp->rds_finfo->rf_vp) {
1920 		resp->status = *cs->statusp = NFS4ERR_BAD_STATEID;
1921 	} else
1922 		rfs4_return_deleg(dsp, FALSE);
1923 
1924 	rfs4_update_lease(dsp->rds_client);
1925 
1926 	rfs4_deleg_state_rele(dsp);
1927 out:
1928 	DTRACE_NFSV4_2(op__delegreturn__done, struct compound_state *, cs,
1929 	    DELEGRETURN4res *, resp);
1930 }
1931 
1932 /*
1933  * Check to see if a given "flavor" is an explicitly shared flavor.
1934  * The assumption of this routine is the "flavor" is already a valid
1935  * flavor in the secinfo list of "exi".
1936  *
1937  *	e.g.
1938  *		# share -o sec=flavor1 /export
1939  *		# share -o sec=flavor2 /export/home
1940  *
1941  *		flavor2 is not an explicitly shared flavor for /export,
1942  *		however it is in the secinfo list for /export thru the
1943  *		server namespace setup.
1944  */
1945 int
1946 is_exported_sec(int flavor, struct exportinfo *exi)
1947 {
1948 	int	i;
1949 	struct secinfo *sp;
1950 
1951 	sp = exi->exi_export.ex_secinfo;
1952 	for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
1953 		if (flavor == sp[i].s_secinfo.sc_nfsnum ||
1954 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE) {
1955 			return (SEC_REF_EXPORTED(&sp[i]));
1956 		}
1957 	}
1958 
1959 	/* Should not reach this point based on the assumption */
1960 	return (0);
1961 }
1962 
1963 /*
1964  * Check if the security flavor used in the request matches what is
1965  * required at the export point or at the root pseudo node (exi_root).
1966  *
1967  * returns 1 if there's a match or if exported with AUTH_NONE; 0 otherwise.
1968  *
1969  */
1970 static int
1971 secinfo_match_or_authnone(struct compound_state *cs)
1972 {
1973 	int	i;
1974 	struct secinfo *sp;
1975 
1976 	/*
1977 	 * Check cs->nfsflavor (from the request) against
1978 	 * the current export data in cs->exi.
1979 	 */
1980 	sp = cs->exi->exi_export.ex_secinfo;
1981 	for (i = 0; i < cs->exi->exi_export.ex_seccnt; i++) {
1982 		if (cs->nfsflavor == sp[i].s_secinfo.sc_nfsnum ||
1983 		    sp[i].s_secinfo.sc_nfsnum == AUTH_NONE)
1984 			return (1);
1985 	}
1986 
1987 	return (0);
1988 }
1989 
1990 /*
1991  * Check the access authority for the client and return the correct error.
1992  */
1993 nfsstat4
1994 call_checkauth4(struct compound_state *cs, struct svc_req *req)
1995 {
1996 	int	authres;
1997 
1998 	/*
1999 	 * First, check if the security flavor used in the request
2000 	 * are among the flavors set in the server namespace.
2001 	 */
2002 	if (!secinfo_match_or_authnone(cs)) {
2003 		*cs->statusp = NFS4ERR_WRONGSEC;
2004 		return (*cs->statusp);
2005 	}
2006 
2007 	authres = checkauth4(cs, req);
2008 
2009 	if (authres > 0) {
2010 		*cs->statusp = NFS4_OK;
2011 		if (! (cs->access & CS_ACCESS_LIMITED))
2012 			cs->access = CS_ACCESS_OK;
2013 	} else if (authres == 0) {
2014 		*cs->statusp = NFS4ERR_ACCESS;
2015 	} else if (authres == -2) {
2016 		*cs->statusp = NFS4ERR_WRONGSEC;
2017 	} else {
2018 		*cs->statusp = NFS4ERR_DELAY;
2019 	}
2020 	return (*cs->statusp);
2021 }
2022 
2023 /*
2024  * bitmap4_to_attrmask is called by getattr and readdir.
2025  * It sets up the vattr mask and determines whether vfsstat call is needed
2026  * based on the input bitmap.
2027  * Returns nfsv4 status.
2028  */
2029 static nfsstat4
2030 bitmap4_to_attrmask(bitmap4 breq, struct nfs4_svgetit_arg *sargp)
2031 {
2032 	int i;
2033 	uint_t	va_mask;
2034 	struct statvfs64 *sbp = sargp->sbp;
2035 
2036 	sargp->sbp = NULL;
2037 	sargp->flag = 0;
2038 	sargp->rdattr_error = NFS4_OK;
2039 	sargp->mntdfid_set = FALSE;
2040 	if (sargp->cs->vp)
2041 		sargp->xattr = get_fh4_flag(&sargp->cs->fh,
2042 		    FH4_ATTRDIR | FH4_NAMEDATTR);
2043 	else
2044 		sargp->xattr = 0;
2045 
2046 	/*
2047 	 * Set rdattr_error_req to true if return error per
2048 	 * failed entry rather than fail the readdir.
2049 	 */
2050 	if (breq & FATTR4_RDATTR_ERROR_MASK)
2051 		sargp->rdattr_error_req = 1;
2052 	else
2053 		sargp->rdattr_error_req = 0;
2054 
2055 	/*
2056 	 * generate the va_mask
2057 	 * Handle the easy cases first
2058 	 */
2059 	switch (breq) {
2060 	case NFS4_NTOV_ATTR_MASK:
2061 		sargp->vap->va_mask = NFS4_NTOV_ATTR_AT_MASK;
2062 		return (NFS4_OK);
2063 
2064 	case NFS4_FS_ATTR_MASK:
2065 		sargp->vap->va_mask = NFS4_FS_ATTR_AT_MASK;
2066 		sargp->sbp = sbp;
2067 		return (NFS4_OK);
2068 
2069 	case NFS4_NTOV_ATTR_CACHE_MASK:
2070 		sargp->vap->va_mask = NFS4_NTOV_ATTR_CACHE_AT_MASK;
2071 		return (NFS4_OK);
2072 
2073 	case FATTR4_LEASE_TIME_MASK:
2074 		sargp->vap->va_mask = 0;
2075 		return (NFS4_OK);
2076 
2077 	default:
2078 		va_mask = 0;
2079 		for (i = 0; i < nfs4_ntov_map_size; i++) {
2080 			if ((breq & nfs4_ntov_map[i].fbit) &&
2081 			    nfs4_ntov_map[i].vbit)
2082 				va_mask |= nfs4_ntov_map[i].vbit;
2083 		}
2084 
2085 		/*
2086 		 * Check is vfsstat is needed
2087 		 */
2088 		if (breq & NFS4_FS_ATTR_MASK)
2089 			sargp->sbp = sbp;
2090 
2091 		sargp->vap->va_mask = va_mask;
2092 		return (NFS4_OK);
2093 	}
2094 	/* NOTREACHED */
2095 }
2096 
2097 /*
2098  * bitmap4_get_sysattrs is called by getattr and readdir.
2099  * It calls both VOP_GETATTR and VFS_STATVFS calls to get the attrs.
2100  * Returns nfsv4 status.
2101  */
2102 static nfsstat4
2103 bitmap4_get_sysattrs(struct nfs4_svgetit_arg *sargp)
2104 {
2105 	int error;
2106 	struct compound_state *cs = sargp->cs;
2107 	vnode_t *vp = cs->vp;
2108 
2109 	if (sargp->sbp != NULL) {
2110 		if (error = VFS_STATVFS(vp->v_vfsp, sargp->sbp)) {
2111 			sargp->sbp = NULL;	/* to identify error */
2112 			return (puterrno4(error));
2113 		}
2114 	}
2115 
2116 	return (rfs4_vop_getattr(vp, sargp->vap, 0, cs->cr));
2117 }
2118 
2119 static void
2120 nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp)
2121 {
2122 	ntovp->na = kmem_zalloc(sizeof (union nfs4_attr_u) * nfs4_ntov_map_size,
2123 	    KM_SLEEP);
2124 	ntovp->attrcnt = 0;
2125 	ntovp->vfsstat = FALSE;
2126 }
2127 
2128 static void
2129 nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
2130     struct nfs4_svgetit_arg *sargp)
2131 {
2132 	int i;
2133 	union nfs4_attr_u *na;
2134 	uint8_t *amap;
2135 
2136 	/*
2137 	 * XXX Should do the same checks for whether the bit is set
2138 	 */
2139 	for (i = 0, na = ntovp->na, amap = ntovp->amap;
2140 	    i < ntovp->attrcnt; i++, na++, amap++) {
2141 		(void) (*nfs4_ntov_map[*amap].sv_getit)(
2142 		    NFS4ATTR_FREEIT, sargp, na);
2143 	}
2144 	if ((sargp->op == NFS4ATTR_SETIT) || (sargp->op == NFS4ATTR_VERIT)) {
2145 		/*
2146 		 * xdr_free for getattr will be done later
2147 		 */
2148 		for (i = 0, na = ntovp->na, amap = ntovp->amap;
2149 		    i < ntovp->attrcnt; i++, na++, amap++) {
2150 			xdr_free(nfs4_ntov_map[*amap].xfunc, (caddr_t)na);
2151 		}
2152 	}
2153 	kmem_free(ntovp->na, sizeof (union nfs4_attr_u) * nfs4_ntov_map_size);
2154 }
2155 
2156 /*
2157  * do_rfs4_op_getattr gets the system attrs and converts into fattr4.
2158  */
2159 static nfsstat4
2160 do_rfs4_op_getattr(bitmap4 breq, fattr4 *fattrp,
2161     struct nfs4_svgetit_arg *sargp)
2162 {
2163 	int error = 0;
2164 	int i, k;
2165 	struct nfs4_ntov_table ntov;
2166 	XDR xdr;
2167 	ulong_t xdr_size;
2168 	char *xdr_attrs;
2169 	nfsstat4 status = NFS4_OK;
2170 	nfsstat4 prev_rdattr_error = sargp->rdattr_error;
2171 	union nfs4_attr_u *na;
2172 	uint8_t *amap;
2173 
2174 	sargp->op = NFS4ATTR_GETIT;
2175 	sargp->flag = 0;
2176 
2177 	fattrp->attrmask = 0;
2178 	/* if no bits requested, then return empty fattr4 */
2179 	if (breq == 0) {
2180 		fattrp->attrlist4_len = 0;
2181 		fattrp->attrlist4 = NULL;
2182 		return (NFS4_OK);
2183 	}
2184 
2185 	/*
2186 	 * return NFS4ERR_INVAL when client requests write-only attrs
2187 	 */
2188 	if (breq & (FATTR4_TIME_ACCESS_SET_MASK | FATTR4_TIME_MODIFY_SET_MASK))
2189 		return (NFS4ERR_INVAL);
2190 
2191 	nfs4_ntov_table_init(&ntov);
2192 	na = ntov.na;
2193 	amap = ntov.amap;
2194 
2195 	/*
2196 	 * Now loop to get or verify the attrs
2197 	 */
2198 	for (i = 0; i < nfs4_ntov_map_size; i++) {
2199 		if (breq & nfs4_ntov_map[i].fbit) {
2200 			if ((*nfs4_ntov_map[i].sv_getit)(
2201 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0) {
2202 
2203 				error = (*nfs4_ntov_map[i].sv_getit)(
2204 				    NFS4ATTR_GETIT, sargp, na);
2205 
2206 				/*
2207 				 * Possible error values:
2208 				 * >0 if sv_getit failed to
2209 				 * get the attr; 0 if succeeded;
2210 				 * <0 if rdattr_error and the
2211 				 * attribute cannot be returned.
2212 				 */
2213 				if (error && !(sargp->rdattr_error_req))
2214 					goto done;
2215 				/*
2216 				 * If error then just for entry
2217 				 */
2218 				if (error == 0) {
2219 					fattrp->attrmask |=
2220 					    nfs4_ntov_map[i].fbit;
2221 					*amap++ =
2222 					    (uint8_t)nfs4_ntov_map[i].nval;
2223 					na++;
2224 					(ntov.attrcnt)++;
2225 				} else if ((error > 0) &&
2226 				    (sargp->rdattr_error == NFS4_OK)) {
2227 					sargp->rdattr_error = puterrno4(error);
2228 				}
2229 				error = 0;
2230 			}
2231 		}
2232 	}
2233 
2234 	/*
2235 	 * If rdattr_error was set after the return value for it was assigned,
2236 	 * update it.
2237 	 */
2238 	if (prev_rdattr_error != sargp->rdattr_error) {
2239 		na = ntov.na;
2240 		amap = ntov.amap;
2241 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2242 			k = *amap;
2243 			if (k < FATTR4_RDATTR_ERROR) {
2244 				continue;
2245 			}
2246 			if ((k == FATTR4_RDATTR_ERROR) &&
2247 			    ((*nfs4_ntov_map[k].sv_getit)(
2248 			    NFS4ATTR_SUPPORTED, sargp, NULL) == 0)) {
2249 
2250 				(void) (*nfs4_ntov_map[k].sv_getit)(
2251 				    NFS4ATTR_GETIT, sargp, na);
2252 			}
2253 			break;
2254 		}
2255 	}
2256 
2257 	xdr_size = 0;
2258 	na = ntov.na;
2259 	amap = ntov.amap;
2260 	for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2261 		xdr_size += xdr_sizeof(nfs4_ntov_map[*amap].xfunc, na);
2262 	}
2263 
2264 	fattrp->attrlist4_len = xdr_size;
2265 	if (xdr_size) {
2266 		/* freed by rfs4_op_getattr_free() */
2267 		fattrp->attrlist4 = xdr_attrs = kmem_zalloc(xdr_size, KM_SLEEP);
2268 
2269 		xdrmem_create(&xdr, xdr_attrs, xdr_size, XDR_ENCODE);
2270 
2271 		na = ntov.na;
2272 		amap = ntov.amap;
2273 		for (i = 0; i < ntov.attrcnt; i++, na++, amap++) {
2274 			if (!(*nfs4_ntov_map[*amap].xfunc)(&xdr, na)) {
2275 				DTRACE_PROBE1(nfss__e__getattr4_encfail,
2276 				    int, *amap);
2277 				status = NFS4ERR_SERVERFAULT;
2278 				break;
2279 			}
2280 		}
2281 		/* xdrmem_destroy(&xdrs); */	/* NO-OP */
2282 	} else {
2283 		fattrp->attrlist4 = NULL;
2284 	}
2285 done:
2286 
2287 	nfs4_ntov_table_free(&ntov, sargp);
2288 
2289 	if (error != 0)
2290 		status = puterrno4(error);
2291 
2292 	return (status);
2293 }
2294 
2295 /* ARGSUSED */
2296 static void
2297 rfs4_op_getattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2298     struct compound_state *cs)
2299 {
2300 	GETATTR4args *args = &argop->nfs_argop4_u.opgetattr;
2301 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2302 	struct nfs4_svgetit_arg sarg;
2303 	struct statvfs64 sb;
2304 	nfsstat4 status;
2305 
2306 	DTRACE_NFSV4_2(op__getattr__start, struct compound_state *, cs,
2307 	    GETATTR4args *, args);
2308 
2309 	if (cs->vp == NULL) {
2310 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2311 		goto out;
2312 	}
2313 
2314 	if (cs->access == CS_ACCESS_DENIED) {
2315 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2316 		goto out;
2317 	}
2318 
2319 	sarg.sbp = &sb;
2320 	sarg.cs = cs;
2321 	sarg.is_referral = B_FALSE;
2322 
2323 	status = bitmap4_to_attrmask(args->attr_request, &sarg);
2324 	if (status == NFS4_OK) {
2325 
2326 		status = bitmap4_get_sysattrs(&sarg);
2327 		if (status == NFS4_OK) {
2328 
2329 			/* Is this a referral? */
2330 			if (vn_is_nfs_reparse(cs->vp, cs->cr)) {
2331 				/* Older V4 Solaris client sees a link */
2332 				if (client_is_downrev(req))
2333 					sarg.vap->va_type = VLNK;
2334 				else
2335 					sarg.is_referral = B_TRUE;
2336 			}
2337 
2338 			status = do_rfs4_op_getattr(args->attr_request,
2339 			    &resp->obj_attributes, &sarg);
2340 		}
2341 	}
2342 	*cs->statusp = resp->status = status;
2343 out:
2344 	DTRACE_NFSV4_2(op__getattr__done, struct compound_state *, cs,
2345 	    GETATTR4res *, resp);
2346 }
2347 
2348 static void
2349 rfs4_op_getattr_free(nfs_resop4 *resop)
2350 {
2351 	GETATTR4res *resp = &resop->nfs_resop4_u.opgetattr;
2352 
2353 	nfs4_fattr4_free(&resp->obj_attributes);
2354 }
2355 
2356 /* ARGSUSED */
2357 static void
2358 rfs4_op_getfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2359     struct compound_state *cs)
2360 {
2361 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2362 
2363 	DTRACE_NFSV4_1(op__getfh__start, struct compound_state *, cs);
2364 
2365 	if (cs->vp == NULL) {
2366 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2367 		goto out;
2368 	}
2369 	if (cs->access == CS_ACCESS_DENIED) {
2370 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2371 		goto out;
2372 	}
2373 
2374 	/* check for reparse point at the share point */
2375 	if (cs->exi->exi_moved || vn_is_nfs_reparse(cs->exi->exi_vp, cs->cr)) {
2376 		/* it's all bad */
2377 		cs->exi->exi_moved = 1;
2378 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2379 		DTRACE_PROBE2(nfs4serv__func__referral__shared__moved,
2380 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2381 		return;
2382 	}
2383 
2384 	/* check for reparse point at vp */
2385 	if (vn_is_nfs_reparse(cs->vp, cs->cr) && !client_is_downrev(req)) {
2386 		/* it's not all bad */
2387 		*cs->statusp = resp->status = NFS4ERR_MOVED;
2388 		DTRACE_PROBE2(nfs4serv__func__referral__moved,
2389 		    vnode_t *, cs->vp, char *, "rfs4_op_getfh");
2390 		return;
2391 	}
2392 
2393 	resp->object.nfs_fh4_val =
2394 	    kmem_alloc(cs->fh.nfs_fh4_len, KM_SLEEP);
2395 	nfs_fh4_copy(&cs->fh, &resp->object);
2396 	*cs->statusp = resp->status = NFS4_OK;
2397 out:
2398 	DTRACE_NFSV4_2(op__getfh__done, struct compound_state *, cs,
2399 	    GETFH4res *, resp);
2400 }
2401 
2402 static void
2403 rfs4_op_getfh_free(nfs_resop4 *resop)
2404 {
2405 	GETFH4res *resp = &resop->nfs_resop4_u.opgetfh;
2406 
2407 	if (resp->status == NFS4_OK &&
2408 	    resp->object.nfs_fh4_val != NULL) {
2409 		kmem_free(resp->object.nfs_fh4_val, resp->object.nfs_fh4_len);
2410 		resp->object.nfs_fh4_val = NULL;
2411 		resp->object.nfs_fh4_len = 0;
2412 	}
2413 }
2414 
2415 /*
2416  * illegal: args: void
2417  *	    res : status (NFS4ERR_OP_ILLEGAL)
2418  */
2419 /* ARGSUSED */
2420 static void
2421 rfs4_op_illegal(nfs_argop4 *argop, nfs_resop4 *resop,
2422     struct svc_req *req, struct compound_state *cs)
2423 {
2424 	ILLEGAL4res *resp = &resop->nfs_resop4_u.opillegal;
2425 
2426 	resop->resop = OP_ILLEGAL;
2427 	*cs->statusp = resp->status = NFS4ERR_OP_ILLEGAL;
2428 }
2429 
2430 /*
2431  * link: args: SAVED_FH: file, CURRENT_FH: target directory
2432  *	 res: status. If success - CURRENT_FH unchanged, return change_info
2433  */
2434 /* ARGSUSED */
2435 static void
2436 rfs4_op_link(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2437     struct compound_state *cs)
2438 {
2439 	LINK4args *args = &argop->nfs_argop4_u.oplink;
2440 	LINK4res *resp = &resop->nfs_resop4_u.oplink;
2441 	int error;
2442 	vnode_t *vp;
2443 	vnode_t *dvp;
2444 	struct vattr bdva, idva, adva;
2445 	char *nm;
2446 	uint_t  len;
2447 	struct sockaddr *ca;
2448 	char *name = NULL;
2449 
2450 	DTRACE_NFSV4_2(op__link__start, struct compound_state *, cs,
2451 	    LINK4args *, args);
2452 
2453 	/* SAVED_FH: source object */
2454 	vp = cs->saved_vp;
2455 	if (vp == NULL) {
2456 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2457 		goto out;
2458 	}
2459 
2460 	/* CURRENT_FH: target directory */
2461 	dvp = cs->vp;
2462 	if (dvp == NULL) {
2463 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2464 		goto out;
2465 	}
2466 
2467 	/*
2468 	 * If there is a non-shared filesystem mounted on this vnode,
2469 	 * do not allow to link any file in this directory.
2470 	 */
2471 	if (vn_ismntpt(dvp)) {
2472 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2473 		goto out;
2474 	}
2475 
2476 	if (cs->access == CS_ACCESS_DENIED) {
2477 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
2478 		goto out;
2479 	}
2480 
2481 	/* Check source object's type validity */
2482 	if (vp->v_type == VDIR) {
2483 		*cs->statusp = resp->status = NFS4ERR_ISDIR;
2484 		goto out;
2485 	}
2486 
2487 	/* Check target directory's type */
2488 	if (dvp->v_type != VDIR) {
2489 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2490 		goto out;
2491 	}
2492 
2493 	if (cs->saved_exi != cs->exi) {
2494 		*cs->statusp = resp->status = NFS4ERR_XDEV;
2495 		goto out;
2496 	}
2497 
2498 	if (!utf8_dir_verify(&args->newname)) {
2499 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2500 		goto out;
2501 	}
2502 
2503 	nm = utf8_to_fn(&args->newname, &len, NULL);
2504 	if (nm == NULL) {
2505 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2506 		goto out;
2507 	}
2508 
2509 	if (len > MAXNAMELEN) {
2510 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2511 		kmem_free(nm, len);
2512 		goto out;
2513 	}
2514 
2515 	if (rdonly4(cs->exi, cs->vp, req)) {
2516 		*cs->statusp = resp->status = NFS4ERR_ROFS;
2517 		kmem_free(nm, len);
2518 		goto out;
2519 	}
2520 
2521 	/* Get "before" change value */
2522 	bdva.va_mask = AT_CTIME|AT_SEQ;
2523 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
2524 	if (error) {
2525 		*cs->statusp = resp->status = puterrno4(error);
2526 		kmem_free(nm, len);
2527 		goto out;
2528 	}
2529 
2530 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2531 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2532 	    MAXPATHLEN  + 1);
2533 
2534 	if (name == NULL) {
2535 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2536 		kmem_free(nm, len);
2537 		goto out;
2538 	}
2539 
2540 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
2541 
2542 	error = VOP_LINK(dvp, vp, name, cs->cr, NULL, 0);
2543 
2544 	if (nm != name)
2545 		kmem_free(name, MAXPATHLEN + 1);
2546 	kmem_free(nm, len);
2547 
2548 	/*
2549 	 * Get the initial "after" sequence number, if it fails, set to zero
2550 	 */
2551 	idva.va_mask = AT_SEQ;
2552 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
2553 		idva.va_seq = 0;
2554 
2555 	/*
2556 	 * Force modified data and metadata out to stable storage.
2557 	 */
2558 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
2559 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
2560 
2561 	if (error) {
2562 		*cs->statusp = resp->status = puterrno4(error);
2563 		goto out;
2564 	}
2565 
2566 	/*
2567 	 * Get "after" change value, if it fails, simply return the
2568 	 * before value.
2569 	 */
2570 	adva.va_mask = AT_CTIME|AT_SEQ;
2571 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
2572 		adva.va_ctime = bdva.va_ctime;
2573 		adva.va_seq = 0;
2574 	}
2575 
2576 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
2577 
2578 	/*
2579 	 * The cinfo.atomic = TRUE only if we have
2580 	 * non-zero va_seq's, and it has incremented by exactly one
2581 	 * during the VOP_LINK and it didn't change during the VOP_FSYNC.
2582 	 */
2583 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
2584 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
2585 		resp->cinfo.atomic = TRUE;
2586 	else
2587 		resp->cinfo.atomic = FALSE;
2588 
2589 	*cs->statusp = resp->status = NFS4_OK;
2590 out:
2591 	DTRACE_NFSV4_2(op__link__done, struct compound_state *, cs,
2592 	    LINK4res *, resp);
2593 }
2594 
2595 /*
2596  * Used by rfs4_op_lookup and rfs4_op_lookupp to do the actual work.
2597  */
2598 
2599 /* ARGSUSED */
2600 static nfsstat4
2601 do_rfs4_op_lookup(char *nm, struct svc_req *req, struct compound_state *cs)
2602 {
2603 	int error;
2604 	int different_export = 0;
2605 	vnode_t *vp, *tvp, *pre_tvp = NULL, *oldvp = NULL;
2606 	struct exportinfo *exi = NULL, *pre_exi = NULL;
2607 	nfsstat4 stat;
2608 	fid_t fid;
2609 	int attrdir, dotdot, walk;
2610 	bool_t is_newvp = FALSE;
2611 
2612 	if (cs->vp->v_flag & V_XATTRDIR) {
2613 		attrdir = 1;
2614 		ASSERT(get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2615 	} else {
2616 		attrdir = 0;
2617 		ASSERT(! get_fh4_flag(&cs->fh, FH4_ATTRDIR));
2618 	}
2619 
2620 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
2621 
2622 	/*
2623 	 * If dotdotting, then need to check whether it's
2624 	 * above the root of a filesystem, or above an
2625 	 * export point.
2626 	 */
2627 	if (dotdot) {
2628 
2629 		/*
2630 		 * If dotdotting at the root of a filesystem, then
2631 		 * need to traverse back to the mounted-on filesystem
2632 		 * and do the dotdot lookup there.
2633 		 */
2634 		if (cs->vp->v_flag & VROOT) {
2635 
2636 			/*
2637 			 * If at the system root, then can
2638 			 * go up no further.
2639 			 */
2640 			if (VN_CMP(cs->vp, rootdir))
2641 				return (puterrno4(ENOENT));
2642 
2643 			/*
2644 			 * Traverse back to the mounted-on filesystem
2645 			 */
2646 			cs->vp = untraverse(cs->vp);
2647 
2648 			/*
2649 			 * Set the different_export flag so we remember
2650 			 * to pick up a new exportinfo entry for
2651 			 * this new filesystem.
2652 			 */
2653 			different_export = 1;
2654 		} else {
2655 
2656 			/*
2657 			 * If dotdotting above an export point then set
2658 			 * the different_export to get new export info.
2659 			 */
2660 			different_export = nfs_exported(cs->exi, cs->vp);
2661 		}
2662 	}
2663 
2664 	error = VOP_LOOKUP(cs->vp, nm, &vp, NULL, 0, NULL, cs->cr,
2665 	    NULL, NULL, NULL);
2666 	if (error)
2667 		return (puterrno4(error));
2668 
2669 	/*
2670 	 * If the vnode is in a pseudo filesystem, check whether it is visible.
2671 	 *
2672 	 * XXX if the vnode is a symlink and it is not visible in
2673 	 * a pseudo filesystem, return ENOENT (not following symlink).
2674 	 * V4 client can not mount such symlink. This is a regression
2675 	 * from V2/V3.
2676 	 *
2677 	 * In the same exported filesystem, if the security flavor used
2678 	 * is not an explicitly shared flavor, limit the view to the visible
2679 	 * list entries only. This is not a WRONGSEC case because it's already
2680 	 * checked via PUTROOTFH/PUTPUBFH or PUTFH.
2681 	 */
2682 	if (!different_export &&
2683 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
2684 	    cs->access & CS_ACCESS_LIMITED)) {
2685 		if (! nfs_visible(cs->exi, vp, &different_export)) {
2686 			VN_RELE(vp);
2687 			return (puterrno4(ENOENT));
2688 		}
2689 	}
2690 
2691 	/*
2692 	 * If it's a mountpoint, then traverse it.
2693 	 */
2694 	if (vn_ismntpt(vp)) {
2695 		pre_exi = cs->exi;	/* save pre-traversed exportinfo */
2696 		pre_tvp = vp;		/* save pre-traversed vnode	*/
2697 
2698 		/*
2699 		 * hold pre_tvp to counteract rele by traverse.  We will
2700 		 * need pre_tvp below if checkexport4 fails
2701 		 */
2702 		VN_HOLD(pre_tvp);
2703 		tvp = vp;
2704 		if ((error = traverse(&tvp)) != 0) {
2705 			VN_RELE(vp);
2706 			VN_RELE(pre_tvp);
2707 			return (puterrno4(error));
2708 		}
2709 		vp = tvp;
2710 		different_export = 1;
2711 	} else if (vp->v_vfsp != cs->vp->v_vfsp) {
2712 		/*
2713 		 * The vfsp comparison is to handle the case where
2714 		 * a LOFS mount is shared.  lo_lookup traverses mount points,
2715 		 * and NFS is unaware of local fs transistions because
2716 		 * v_vfsmountedhere isn't set.  For this special LOFS case,
2717 		 * the dir and the obj returned by lookup will have different
2718 		 * vfs ptrs.
2719 		 */
2720 		different_export = 1;
2721 	}
2722 
2723 	if (different_export) {
2724 
2725 		bzero(&fid, sizeof (fid));
2726 		fid.fid_len = MAXFIDSZ;
2727 		error = vop_fid_pseudo(vp, &fid);
2728 		if (error) {
2729 			VN_RELE(vp);
2730 			if (pre_tvp)
2731 				VN_RELE(pre_tvp);
2732 			return (puterrno4(error));
2733 		}
2734 
2735 		if (dotdot)
2736 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
2737 		else
2738 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2739 
2740 		if (exi == NULL) {
2741 			if (pre_tvp) {
2742 				/*
2743 				 * If this vnode is a mounted-on vnode,
2744 				 * but the mounted-on file system is not
2745 				 * exported, send back the filehandle for
2746 				 * the mounted-on vnode, not the root of
2747 				 * the mounted-on file system.
2748 				 */
2749 				VN_RELE(vp);
2750 				vp = pre_tvp;
2751 				exi = pre_exi;
2752 			} else {
2753 				VN_RELE(vp);
2754 				return (puterrno4(EACCES));
2755 			}
2756 		} else if (pre_tvp) {
2757 			/* we're done with pre_tvp now. release extra hold */
2758 			VN_RELE(pre_tvp);
2759 		}
2760 
2761 		cs->exi = exi;
2762 
2763 		/*
2764 		 * Now we do a checkauth4. The reason is that
2765 		 * this client/user may not have access to the new
2766 		 * exported file system, and if he does,
2767 		 * the client/user may be mapped to a different uid.
2768 		 *
2769 		 * We start with a new cr, because the checkauth4 done
2770 		 * in the PUT*FH operation over wrote the cred's uid,
2771 		 * gid, etc, and we want the real thing before calling
2772 		 * checkauth4()
2773 		 */
2774 		crfree(cs->cr);
2775 		cs->cr = crdup(cs->basecr);
2776 
2777 		oldvp = cs->vp;
2778 		cs->vp = vp;
2779 		is_newvp = TRUE;
2780 
2781 		stat = call_checkauth4(cs, req);
2782 		if (stat != NFS4_OK) {
2783 			VN_RELE(cs->vp);
2784 			cs->vp = oldvp;
2785 			return (stat);
2786 		}
2787 	}
2788 
2789 	/*
2790 	 * After various NFS checks, do a label check on the path
2791 	 * component. The label on this path should either be the
2792 	 * global zone's label or a zone's label. We are only
2793 	 * interested in the zone's label because exported files
2794 	 * in global zone is accessible (though read-only) to
2795 	 * clients. The exportability/visibility check is already
2796 	 * done before reaching this code.
2797 	 */
2798 	if (is_system_labeled()) {
2799 		bslabel_t *clabel;
2800 
2801 		ASSERT(req->rq_label != NULL);
2802 		clabel = req->rq_label;
2803 		DTRACE_PROBE2(tx__rfs4__log__info__oplookup__clabel, char *,
2804 		    "got client label from request(1)", struct svc_req *, req);
2805 
2806 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
2807 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
2808 			    cs->exi)) {
2809 				error = EACCES;
2810 				goto err_out;
2811 			}
2812 		} else {
2813 			/*
2814 			 * We grant access to admin_low label clients
2815 			 * only if the client is trusted, i.e. also
2816 			 * running Solaris Trusted Extension.
2817 			 */
2818 			struct sockaddr	*ca;
2819 			int		addr_type;
2820 			void		*ipaddr;
2821 			tsol_tpc_t	*tp;
2822 
2823 			ca = (struct sockaddr *)svc_getrpccaller(
2824 			    req->rq_xprt)->buf;
2825 			if (ca->sa_family == AF_INET) {
2826 				addr_type = IPV4_VERSION;
2827 				ipaddr = &((struct sockaddr_in *)ca)->sin_addr;
2828 			} else if (ca->sa_family == AF_INET6) {
2829 				addr_type = IPV6_VERSION;
2830 				ipaddr = &((struct sockaddr_in6 *)
2831 				    ca)->sin6_addr;
2832 			}
2833 			tp = find_tpc(ipaddr, addr_type, B_FALSE);
2834 			if (tp == NULL || tp->tpc_tp.tp_doi !=
2835 			    l_admin_low->tsl_doi || tp->tpc_tp.host_type !=
2836 			    SUN_CIPSO) {
2837 				if (tp != NULL)
2838 					TPC_RELE(tp);
2839 				error = EACCES;
2840 				goto err_out;
2841 			}
2842 			TPC_RELE(tp);
2843 		}
2844 	}
2845 
2846 	error = makefh4(&cs->fh, vp, cs->exi);
2847 
2848 err_out:
2849 	if (error) {
2850 		if (is_newvp) {
2851 			VN_RELE(cs->vp);
2852 			cs->vp = oldvp;
2853 		} else
2854 			VN_RELE(vp);
2855 		return (puterrno4(error));
2856 	}
2857 
2858 	if (!is_newvp) {
2859 		if (cs->vp)
2860 			VN_RELE(cs->vp);
2861 		cs->vp = vp;
2862 	} else if (oldvp)
2863 		VN_RELE(oldvp);
2864 
2865 	/*
2866 	 * if did lookup on attrdir and didn't lookup .., set named
2867 	 * attr fh flag
2868 	 */
2869 	if (attrdir && ! dotdot)
2870 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
2871 
2872 	/* Assume false for now, open proc will set this */
2873 	cs->mandlock = FALSE;
2874 
2875 	return (NFS4_OK);
2876 }
2877 
2878 /* ARGSUSED */
2879 static void
2880 rfs4_op_lookup(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2881     struct compound_state *cs)
2882 {
2883 	LOOKUP4args *args = &argop->nfs_argop4_u.oplookup;
2884 	LOOKUP4res *resp = &resop->nfs_resop4_u.oplookup;
2885 	char *nm;
2886 	uint_t len;
2887 	struct sockaddr *ca;
2888 	char *name = NULL;
2889 
2890 	DTRACE_NFSV4_2(op__lookup__start, struct compound_state *, cs,
2891 	    LOOKUP4args *, args);
2892 
2893 	if (cs->vp == NULL) {
2894 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2895 		goto out;
2896 	}
2897 
2898 	if (cs->vp->v_type == VLNK) {
2899 		*cs->statusp = resp->status = NFS4ERR_SYMLINK;
2900 		goto out;
2901 	}
2902 
2903 	if (cs->vp->v_type != VDIR) {
2904 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2905 		goto out;
2906 	}
2907 
2908 	if (!utf8_dir_verify(&args->objname)) {
2909 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2910 		goto out;
2911 	}
2912 
2913 	nm = utf8_to_str(&args->objname, &len, NULL);
2914 	if (nm == NULL) {
2915 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2916 		goto out;
2917 	}
2918 
2919 	if (len > MAXNAMELEN) {
2920 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2921 		kmem_free(nm, len);
2922 		goto out;
2923 	}
2924 
2925 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2926 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2927 	    MAXPATHLEN  + 1);
2928 
2929 	if (name == NULL) {
2930 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2931 		kmem_free(nm, len);
2932 		goto out;
2933 	}
2934 
2935 	*cs->statusp = resp->status = do_rfs4_op_lookup(name, req, cs);
2936 
2937 	if (name != nm)
2938 		kmem_free(name, MAXPATHLEN + 1);
2939 	kmem_free(nm, len);
2940 
2941 out:
2942 	DTRACE_NFSV4_2(op__lookup__done, struct compound_state *, cs,
2943 	    LOOKUP4res *, resp);
2944 }
2945 
2946 /* ARGSUSED */
2947 static void
2948 rfs4_op_lookupp(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
2949     struct compound_state *cs)
2950 {
2951 	LOOKUPP4res *resp = &resop->nfs_resop4_u.oplookupp;
2952 
2953 	DTRACE_NFSV4_1(op__lookupp__start, struct compound_state *, cs);
2954 
2955 	if (cs->vp == NULL) {
2956 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2957 		goto out;
2958 	}
2959 
2960 	if (cs->vp->v_type != VDIR) {
2961 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2962 		goto out;
2963 	}
2964 
2965 	*cs->statusp = resp->status = do_rfs4_op_lookup("..", req, cs);
2966 
2967 	/*
2968 	 * From NFSV4 Specification, LOOKUPP should not check for
2969 	 * NFS4ERR_WRONGSEC. Retrun NFS4_OK instead.
2970 	 */
2971 	if (resp->status == NFS4ERR_WRONGSEC) {
2972 		*cs->statusp = resp->status = NFS4_OK;
2973 	}
2974 
2975 out:
2976 	DTRACE_NFSV4_2(op__lookupp__done, struct compound_state *, cs,
2977 	    LOOKUPP4res *, resp);
2978 }
2979 
2980 
2981 /*ARGSUSED2*/
2982 static void
2983 rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2984     struct compound_state *cs)
2985 {
2986 	OPENATTR4args	*args = &argop->nfs_argop4_u.opopenattr;
2987 	OPENATTR4res	*resp = &resop->nfs_resop4_u.opopenattr;
2988 	vnode_t		*avp = NULL;
2989 	int		lookup_flags = LOOKUP_XATTR, error;
2990 	int		exp_ro = 0;
2991 
2992 	DTRACE_NFSV4_2(op__openattr__start, struct compound_state *, cs,
2993 	    OPENATTR4args *, args);
2994 
2995 	if (cs->vp == NULL) {
2996 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2997 		goto out;
2998 	}
2999 
3000 	if ((cs->vp->v_vfsp->vfs_flag & VFS_XATTR) == 0 &&
3001 	    !vfs_has_feature(cs->vp->v_vfsp, VFSFT_SYSATTR_VIEWS)) {
3002 		*cs->statusp = resp->status = puterrno4(ENOTSUP);
3003 		goto out;
3004 	}
3005 
3006 	/*
3007 	 * If file system supports passing ACE mask to VOP_ACCESS then
3008 	 * check for ACE_READ_NAMED_ATTRS, otherwise do legacy checks
3009 	 */
3010 
3011 	if (vfs_has_feature(cs->vp->v_vfsp, VFSFT_ACEMASKONACCESS))
3012 		error = VOP_ACCESS(cs->vp, ACE_READ_NAMED_ATTRS,
3013 		    V_ACE_MASK, cs->cr, NULL);
3014 	else
3015 		error = ((VOP_ACCESS(cs->vp, VREAD, 0, cs->cr, NULL) != 0) &&
3016 		    (VOP_ACCESS(cs->vp, VWRITE, 0, cs->cr, NULL) != 0) &&
3017 		    (VOP_ACCESS(cs->vp, VEXEC, 0, cs->cr, NULL) != 0));
3018 
3019 	if (error) {
3020 		*cs->statusp = resp->status = puterrno4(EACCES);
3021 		goto out;
3022 	}
3023 
3024 	/*
3025 	 * The CREATE_XATTR_DIR VOP flag cannot be specified if
3026 	 * the file system is exported read-only -- regardless of
3027 	 * createdir flag.  Otherwise the attrdir would be created
3028 	 * (assuming server fs isn't mounted readonly locally).  If
3029 	 * VOP_LOOKUP returns ENOENT in this case, the error will
3030 	 * be translated into EROFS.  ENOSYS is mapped to ENOTSUP
3031 	 * because specfs has no VOP_LOOKUP op, so the macro would
3032 	 * return ENOSYS.  EINVAL is returned by all (current)
3033 	 * Solaris file system implementations when any of their
3034 	 * restrictions are violated (xattr(dir) can't have xattrdir).
3035 	 * Returning NOTSUPP is more appropriate in this case
3036 	 * because the object will never be able to have an attrdir.
3037 	 */
3038 	if (args->createdir && ! (exp_ro = rdonly4(cs->exi, cs->vp, req)))
3039 		lookup_flags |= CREATE_XATTR_DIR;
3040 
3041 	error = VOP_LOOKUP(cs->vp, "", &avp, NULL, lookup_flags, NULL, cs->cr,
3042 	    NULL, NULL, NULL);
3043 
3044 	if (error) {
3045 		if (error == ENOENT && args->createdir && exp_ro)
3046 			*cs->statusp = resp->status = puterrno4(EROFS);
3047 		else if (error == EINVAL || error == ENOSYS)
3048 			*cs->statusp = resp->status = puterrno4(ENOTSUP);
3049 		else
3050 			*cs->statusp = resp->status = puterrno4(error);
3051 		goto out;
3052 	}
3053 
3054 	ASSERT(avp->v_flag & V_XATTRDIR);
3055 
3056 	error = makefh4(&cs->fh, avp, cs->exi);
3057 
3058 	if (error) {
3059 		VN_RELE(avp);
3060 		*cs->statusp = resp->status = puterrno4(error);
3061 		goto out;
3062 	}
3063 
3064 	VN_RELE(cs->vp);
3065 	cs->vp = avp;
3066 
3067 	/*
3068 	 * There is no requirement for an attrdir fh flag
3069 	 * because the attrdir has a vnode flag to distinguish
3070 	 * it from regular (non-xattr) directories.  The
3071 	 * FH4_ATTRDIR flag is set for future sanity checks.
3072 	 */
3073 	set_fh4_flag(&cs->fh, FH4_ATTRDIR);
3074 	*cs->statusp = resp->status = NFS4_OK;
3075 
3076 out:
3077 	DTRACE_NFSV4_2(op__openattr__done, struct compound_state *, cs,
3078 	    OPENATTR4res *, resp);
3079 }
3080 
3081 static int
3082 do_io(int direction, vnode_t *vp, struct uio *uio, int ioflag, cred_t *cred,
3083     caller_context_t *ct)
3084 {
3085 	int error;
3086 	int i;
3087 	clock_t delaytime;
3088 
3089 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
3090 
3091 	/*
3092 	 * Don't block on mandatory locks. If this routine returns
3093 	 * EAGAIN, the caller should return NFS4ERR_LOCKED.
3094 	 */
3095 	uio->uio_fmode = FNONBLOCK;
3096 
3097 	for (i = 0; i < rfs4_maxlock_tries; i++) {
3098 
3099 
3100 		if (direction == FREAD) {
3101 			(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, ct);
3102 			error = VOP_READ(vp, uio, ioflag, cred, ct);
3103 			VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, ct);
3104 		} else {
3105 			(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, ct);
3106 			error = VOP_WRITE(vp, uio, ioflag, cred, ct);
3107 			VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, ct);
3108 		}
3109 
3110 		if (error != EAGAIN)
3111 			break;
3112 
3113 		if (i < rfs4_maxlock_tries - 1) {
3114 			delay(delaytime);
3115 			delaytime *= 2;
3116 		}
3117 	}
3118 
3119 	return (error);
3120 }
3121 
3122 /* ARGSUSED */
3123 static void
3124 rfs4_op_read(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3125     struct compound_state *cs)
3126 {
3127 	READ4args *args = &argop->nfs_argop4_u.opread;
3128 	READ4res *resp = &resop->nfs_resop4_u.opread;
3129 	int error;
3130 	int verror;
3131 	vnode_t *vp;
3132 	struct vattr va;
3133 	struct iovec iov;
3134 	struct uio uio;
3135 	u_offset_t offset;
3136 	bool_t *deleg = &cs->deleg;
3137 	nfsstat4 stat;
3138 	int in_crit = 0;
3139 	mblk_t *mp = NULL;
3140 	int alloc_err = 0;
3141 	int rdma_used = 0;
3142 	int loaned_buffers;
3143 	caller_context_t ct;
3144 	struct uio *uiop;
3145 
3146 	DTRACE_NFSV4_2(op__read__start, struct compound_state *, cs,
3147 	    READ4args, args);
3148 
3149 	vp = cs->vp;
3150 	if (vp == NULL) {
3151 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3152 		goto out;
3153 	}
3154 	if (cs->access == CS_ACCESS_DENIED) {
3155 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3156 		goto out;
3157 	}
3158 
3159 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3160 	    deleg, TRUE, &ct)) != NFS4_OK) {
3161 		*cs->statusp = resp->status = stat;
3162 		goto out;
3163 	}
3164 
3165 	/*
3166 	 * Enter the critical region before calling VOP_RWLOCK
3167 	 * to avoid a deadlock with write requests.
3168 	 */
3169 	if (nbl_need_check(vp)) {
3170 		nbl_start_crit(vp, RW_READER);
3171 		in_crit = 1;
3172 		if (nbl_conflict(vp, NBL_READ, args->offset, args->count, 0,
3173 		    &ct)) {
3174 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
3175 			goto out;
3176 		}
3177 	}
3178 
3179 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3180 	    deleg, TRUE, &ct)) != NFS4_OK) {
3181 		*cs->statusp = resp->status = stat;
3182 		goto out;
3183 	}
3184 
3185 	if (args->wlist)
3186 		rdma_used = 1;
3187 
3188 	/* use loaned buffers for TCP */
3189 	loaned_buffers = (nfs_loaned_buffers && !rdma_used) ? 1 : 0;
3190 
3191 	va.va_mask = AT_MODE|AT_SIZE|AT_UID;
3192 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3193 
3194 	/*
3195 	 * If we can't get the attributes, then we can't do the
3196 	 * right access checking.  So, we'll fail the request.
3197 	 */
3198 	if (verror) {
3199 		*cs->statusp = resp->status = puterrno4(verror);
3200 		goto out;
3201 	}
3202 
3203 	if (vp->v_type != VREG) {
3204 		*cs->statusp = resp->status =
3205 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
3206 		goto out;
3207 	}
3208 
3209 	if (crgetuid(cs->cr) != va.va_uid &&
3210 	    (error = VOP_ACCESS(vp, VREAD, 0, cs->cr, &ct)) &&
3211 	    (error = VOP_ACCESS(vp, VEXEC, 0, cs->cr, &ct))) {
3212 		*cs->statusp = resp->status = puterrno4(error);
3213 		goto out;
3214 	}
3215 
3216 	if (MANDLOCK(vp, va.va_mode)) { /* XXX - V4 supports mand locking */
3217 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3218 		goto out;
3219 	}
3220 
3221 	offset = args->offset;
3222 	if (offset >= va.va_size) {
3223 		*cs->statusp = resp->status = NFS4_OK;
3224 		resp->eof = TRUE;
3225 		resp->data_len = 0;
3226 		resp->data_val = NULL;
3227 		resp->mblk = NULL;
3228 		/* RDMA */
3229 		resp->wlist = args->wlist;
3230 		resp->wlist_len = resp->data_len;
3231 		*cs->statusp = resp->status = NFS4_OK;
3232 		if (resp->wlist)
3233 			clist_zero_len(resp->wlist);
3234 		goto out;
3235 	}
3236 
3237 	if (args->count == 0) {
3238 		*cs->statusp = resp->status = NFS4_OK;
3239 		resp->eof = FALSE;
3240 		resp->data_len = 0;
3241 		resp->data_val = NULL;
3242 		resp->mblk = NULL;
3243 		/* RDMA */
3244 		resp->wlist = args->wlist;
3245 		resp->wlist_len = resp->data_len;
3246 		if (resp->wlist)
3247 			clist_zero_len(resp->wlist);
3248 		goto out;
3249 	}
3250 
3251 	/*
3252 	 * Do not allocate memory more than maximum allowed
3253 	 * transfer size
3254 	 */
3255 	if (args->count > rfs4_tsize(req))
3256 		args->count = rfs4_tsize(req);
3257 
3258 	if (loaned_buffers) {
3259 		uiop = (uio_t *)rfs_setup_xuio(vp);
3260 		ASSERT(uiop != NULL);
3261 		uiop->uio_segflg = UIO_SYSSPACE;
3262 		uiop->uio_loffset = args->offset;
3263 		uiop->uio_resid = args->count;
3264 
3265 		/* Jump to do the read if successful */
3266 		if (!VOP_REQZCBUF(vp, UIO_READ, (xuio_t *)uiop, cs->cr, &ct)) {
3267 			/*
3268 			 * Need to hold the vnode until after VOP_RETZCBUF()
3269 			 * is called.
3270 			 */
3271 			VN_HOLD(vp);
3272 			goto doio_read;
3273 		}
3274 
3275 		DTRACE_PROBE2(nfss__i__reqzcbuf_failed, int,
3276 		    uiop->uio_loffset, int, uiop->uio_resid);
3277 
3278 		uiop->uio_extflg = 0;
3279 
3280 		/* failure to setup for zero copy */
3281 		rfs_free_xuio((void *)uiop);
3282 		loaned_buffers = 0;
3283 	}
3284 
3285 	/*
3286 	 * If returning data via RDMA Write, then grab the chunk list. If we
3287 	 * aren't returning READ data w/RDMA_WRITE, then grab a mblk.
3288 	 */
3289 	if (rdma_used) {
3290 		mp = NULL;
3291 		(void) rdma_get_wchunk(req, &iov, args->wlist);
3292 	} else {
3293 		/*
3294 		 * mp will contain the data to be sent out in the read reply.
3295 		 * It will be freed after the reply has been sent. Let's
3296 		 * roundup the data to a BYTES_PER_XDR_UNIT multiple, so that
3297 		 * the call to xdrmblk_putmblk() never fails. If the first
3298 		 * alloc of the requested size fails, then decrease the size to
3299 		 * something more reasonable and wait for the allocation to
3300 		 * occur.
3301 		 */
3302 		mp = allocb(RNDUP(args->count), BPRI_MED);
3303 		if (mp == NULL) {
3304 			if (args->count > MAXBSIZE)
3305 				args->count = MAXBSIZE;
3306 			mp = allocb_wait(RNDUP(args->count), BPRI_MED,
3307 			    STR_NOSIG, &alloc_err);
3308 		}
3309 		ASSERT(mp != NULL);
3310 		ASSERT(alloc_err == 0);
3311 
3312 		iov.iov_base = (caddr_t)mp->b_datap->db_base;
3313 		iov.iov_len = args->count;
3314 	}
3315 
3316 	uio.uio_iov = &iov;
3317 	uio.uio_iovcnt = 1;
3318 	uio.uio_segflg = UIO_SYSSPACE;
3319 	uio.uio_extflg = UIO_COPY_CACHED;
3320 	uio.uio_loffset = args->offset;
3321 	uio.uio_resid = args->count;
3322 	uiop = &uio;
3323 
3324 doio_read:
3325 	error = do_io(FREAD, vp, uiop, 0, cs->cr, &ct);
3326 
3327 	va.va_mask = AT_SIZE;
3328 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3329 
3330 	if (error) {
3331 		if (mp)
3332 			freemsg(mp);
3333 		*cs->statusp = resp->status = puterrno4(error);
3334 		goto out;
3335 	}
3336 
3337 	/* make mblk using zc buffers */
3338 	if (loaned_buffers) {
3339 		mp = uio_to_mblk(uiop);
3340 		ASSERT(mp != NULL);
3341 	}
3342 
3343 	*cs->statusp = resp->status = NFS4_OK;
3344 
3345 	ASSERT(uiop->uio_resid >= 0);
3346 	resp->data_len = args->count - uiop->uio_resid;
3347 	if (mp) {
3348 		resp->data_val = (char *)mp->b_datap->db_base;
3349 		rfs_rndup_mblks(mp, resp->data_len, loaned_buffers);
3350 	} else {
3351 		resp->data_val = (caddr_t)iov.iov_base;
3352 	}
3353 
3354 	resp->mblk = mp;
3355 
3356 	if (!verror && offset + resp->data_len == va.va_size)
3357 		resp->eof = TRUE;
3358 	else
3359 		resp->eof = FALSE;
3360 
3361 	if (rdma_used) {
3362 		if (!rdma_setup_read_data4(args, resp)) {
3363 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3364 		}
3365 	} else {
3366 		resp->wlist = NULL;
3367 	}
3368 
3369 out:
3370 	if (in_crit)
3371 		nbl_end_crit(vp);
3372 
3373 	DTRACE_NFSV4_2(op__read__done, struct compound_state *, cs,
3374 	    READ4res *, resp);
3375 }
3376 
3377 static void
3378 rfs4_op_read_free(nfs_resop4 *resop)
3379 {
3380 	READ4res	*resp = &resop->nfs_resop4_u.opread;
3381 
3382 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3383 		freemsg(resp->mblk);
3384 		resp->mblk = NULL;
3385 		resp->data_val = NULL;
3386 		resp->data_len = 0;
3387 	}
3388 }
3389 
3390 static void
3391 rfs4_op_readdir_free(nfs_resop4 * resop)
3392 {
3393 	READDIR4res    *resp = &resop->nfs_resop4_u.opreaddir;
3394 
3395 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3396 		freeb(resp->mblk);
3397 		resp->mblk = NULL;
3398 		resp->data_len = 0;
3399 	}
3400 }
3401 
3402 
3403 /* ARGSUSED */
3404 static void
3405 rfs4_op_putpubfh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
3406     struct compound_state *cs)
3407 {
3408 	PUTPUBFH4res	*resp = &resop->nfs_resop4_u.opputpubfh;
3409 	int		error;
3410 	vnode_t		*vp;
3411 	struct exportinfo *exi, *sav_exi;
3412 	nfs_fh4_fmt_t	*fh_fmtp;
3413 
3414 	DTRACE_NFSV4_1(op__putpubfh__start, struct compound_state *, cs);
3415 
3416 	if (cs->vp) {
3417 		VN_RELE(cs->vp);
3418 		cs->vp = NULL;
3419 	}
3420 
3421 	if (cs->cr)
3422 		crfree(cs->cr);
3423 
3424 	cs->cr = crdup(cs->basecr);
3425 
3426 	vp = exi_public->exi_vp;
3427 	if (vp == NULL) {
3428 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3429 		goto out;
3430 	}
3431 
3432 	error = makefh4(&cs->fh, vp, exi_public);
3433 	if (error != 0) {
3434 		*cs->statusp = resp->status = puterrno4(error);
3435 		goto out;
3436 	}
3437 	sav_exi = cs->exi;
3438 	if (exi_public == exi_root) {
3439 		/*
3440 		 * No filesystem is actually shared public, so we default
3441 		 * to exi_root. In this case, we must check whether root
3442 		 * is exported.
3443 		 */
3444 		fh_fmtp = (nfs_fh4_fmt_t *)cs->fh.nfs_fh4_val;
3445 
3446 		/*
3447 		 * if root filesystem is exported, the exportinfo struct that we
3448 		 * should use is what checkexport4 returns, because root_exi is
3449 		 * actually a mostly empty struct.
3450 		 */
3451 		exi = checkexport4(&fh_fmtp->fh4_fsid,
3452 		    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
3453 		cs->exi = ((exi != NULL) ? exi : exi_public);
3454 	} else {
3455 		/*
3456 		 * it's a properly shared filesystem
3457 		 */
3458 		cs->exi = exi_public;
3459 	}
3460 
3461 	if (is_system_labeled()) {
3462 		bslabel_t *clabel;
3463 
3464 		ASSERT(req->rq_label != NULL);
3465 		clabel = req->rq_label;
3466 		DTRACE_PROBE2(tx__rfs4__log__info__opputpubfh__clabel, char *,
3467 		    "got client label from request(1)",
3468 		    struct svc_req *, req);
3469 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
3470 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
3471 			    cs->exi)) {
3472 				*cs->statusp = resp->status =
3473 				    NFS4ERR_SERVERFAULT;
3474 				goto out;
3475 			}
3476 		}
3477 	}
3478 
3479 	VN_HOLD(vp);
3480 	cs->vp = vp;
3481 
3482 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3483 		VN_RELE(cs->vp);
3484 		cs->vp = NULL;
3485 		cs->exi = sav_exi;
3486 		goto out;
3487 	}
3488 
3489 	*cs->statusp = resp->status = NFS4_OK;
3490 out:
3491 	DTRACE_NFSV4_2(op__putpubfh__done, struct compound_state *, cs,
3492 	    PUTPUBFH4res *, resp);
3493 }
3494 
3495 /*
3496  * XXX - issue with put*fh operations. Suppose /export/home is exported.
3497  * Suppose an NFS client goes to mount /export/home/joe. If /export, home,
3498  * or joe have restrictive search permissions, then we shouldn't let
3499  * the client get a file handle. This is easy to enforce. However, we
3500  * don't know what security flavor should be used until we resolve the
3501  * path name. Another complication is uid mapping. If root is
3502  * the user, then it will be mapped to the anonymous user by default,
3503  * but we won't know that till we've resolved the path name. And we won't
3504  * know what the anonymous user is.
3505  * Luckily, SECINFO is specified to take a full filename.
3506  * So what we will have to in rfs4_op_lookup is check that flavor of
3507  * the target object matches that of the request, and if root was the
3508  * caller, check for the root= and anon= options, and if necessary,
3509  * repeat the lookup using the right cred_t. But that's not done yet.
3510  */
3511 /* ARGSUSED */
3512 static void
3513 rfs4_op_putfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3514     struct compound_state *cs)
3515 {
3516 	PUTFH4args *args = &argop->nfs_argop4_u.opputfh;
3517 	PUTFH4res *resp = &resop->nfs_resop4_u.opputfh;
3518 	nfs_fh4_fmt_t *fh_fmtp;
3519 
3520 	DTRACE_NFSV4_2(op__putfh__start, struct compound_state *, cs,
3521 	    PUTFH4args *, args);
3522 
3523 	if (cs->vp) {
3524 		VN_RELE(cs->vp);
3525 		cs->vp = NULL;
3526 	}
3527 
3528 	if (cs->cr) {
3529 		crfree(cs->cr);
3530 		cs->cr = NULL;
3531 	}
3532 
3533 
3534 	if (args->object.nfs_fh4_len < NFS_FH4_LEN) {
3535 		*cs->statusp = resp->status = NFS4ERR_BADHANDLE;
3536 		goto out;
3537 	}
3538 
3539 	fh_fmtp = (nfs_fh4_fmt_t *)args->object.nfs_fh4_val;
3540 	cs->exi = checkexport4(&fh_fmtp->fh4_fsid, (fid_t *)&fh_fmtp->fh4_xlen,
3541 	    NULL);
3542 
3543 	if (cs->exi == NULL) {
3544 		*cs->statusp = resp->status = NFS4ERR_STALE;
3545 		goto out;
3546 	}
3547 
3548 	cs->cr = crdup(cs->basecr);
3549 
3550 	ASSERT(cs->cr != NULL);
3551 
3552 	if (! (cs->vp = nfs4_fhtovp(&args->object, cs->exi, &resp->status))) {
3553 		*cs->statusp = resp->status;
3554 		goto out;
3555 	}
3556 
3557 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3558 		VN_RELE(cs->vp);
3559 		cs->vp = NULL;
3560 		goto out;
3561 	}
3562 
3563 	nfs_fh4_copy(&args->object, &cs->fh);
3564 	*cs->statusp = resp->status = NFS4_OK;
3565 	cs->deleg = FALSE;
3566 
3567 out:
3568 	DTRACE_NFSV4_2(op__putfh__done, struct compound_state *, cs,
3569 	    PUTFH4res *, resp);
3570 }
3571 
3572 /* ARGSUSED */
3573 static void
3574 rfs4_op_putrootfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3575     struct compound_state *cs)
3576 {
3577 	PUTROOTFH4res *resp = &resop->nfs_resop4_u.opputrootfh;
3578 	int error;
3579 	fid_t fid;
3580 	struct exportinfo *exi, *sav_exi;
3581 
3582 	DTRACE_NFSV4_1(op__putrootfh__start, struct compound_state *, cs);
3583 
3584 	if (cs->vp) {
3585 		VN_RELE(cs->vp);
3586 		cs->vp = NULL;
3587 	}
3588 
3589 	if (cs->cr)
3590 		crfree(cs->cr);
3591 
3592 	cs->cr = crdup(cs->basecr);
3593 
3594 	/*
3595 	 * Using rootdir, the system root vnode,
3596 	 * get its fid.
3597 	 */
3598 	bzero(&fid, sizeof (fid));
3599 	fid.fid_len = MAXFIDSZ;
3600 	error = vop_fid_pseudo(rootdir, &fid);
3601 	if (error != 0) {
3602 		*cs->statusp = resp->status = puterrno4(error);
3603 		goto out;
3604 	}
3605 
3606 	/*
3607 	 * Then use the root fsid & fid it to find out if it's exported
3608 	 *
3609 	 * If the server root isn't exported directly, then
3610 	 * it should at least be a pseudo export based on
3611 	 * one or more exports further down in the server's
3612 	 * file tree.
3613 	 */
3614 	exi = checkexport4(&rootdir->v_vfsp->vfs_fsid, &fid, NULL);
3615 	if (exi == NULL || exi->exi_export.ex_flags & EX_PUBLIC) {
3616 		NFS4_DEBUG(rfs4_debug,
3617 		    (CE_WARN, "rfs4_op_putrootfh: export check failure"));
3618 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3619 		goto out;
3620 	}
3621 
3622 	/*
3623 	 * Now make a filehandle based on the root
3624 	 * export and root vnode.
3625 	 */
3626 	error = makefh4(&cs->fh, rootdir, exi);
3627 	if (error != 0) {
3628 		*cs->statusp = resp->status = puterrno4(error);
3629 		goto out;
3630 	}
3631 
3632 	sav_exi = cs->exi;
3633 	cs->exi = exi;
3634 
3635 	VN_HOLD(rootdir);
3636 	cs->vp = rootdir;
3637 
3638 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3639 		VN_RELE(rootdir);
3640 		cs->vp = NULL;
3641 		cs->exi = sav_exi;
3642 		goto out;
3643 	}
3644 
3645 	*cs->statusp = resp->status = NFS4_OK;
3646 	cs->deleg = FALSE;
3647 out:
3648 	DTRACE_NFSV4_2(op__putrootfh__done, struct compound_state *, cs,
3649 	    PUTROOTFH4res *, resp);
3650 }
3651 
3652 /*
3653  * A directory entry is a valid nfsv4 entry if
3654  * - it has a non-zero ino
3655  * - it is not a dot or dotdot name
3656  * - it is visible in a pseudo export or in a real export that can
3657  *   only have a limited view.
3658  */
3659 static bool_t
3660 valid_nfs4_entry(struct exportinfo *exi, struct dirent64 *dp,
3661     int *expseudo, int check_visible)
3662 {
3663 	if (dp->d_ino == 0 || NFS_IS_DOTNAME(dp->d_name)) {
3664 		*expseudo = 0;
3665 		return (FALSE);
3666 	}
3667 
3668 	if (! check_visible) {
3669 		*expseudo = 0;
3670 		return (TRUE);
3671 	}
3672 
3673 	return (nfs_visible_inode(exi, dp->d_ino, expseudo));
3674 }
3675 
3676 /*
3677  * set_rdattr_params sets up the variables used to manage what information
3678  * to get for each directory entry.
3679  */
3680 static nfsstat4
3681 set_rdattr_params(struct nfs4_svgetit_arg *sargp,
3682     bitmap4 attrs, bool_t *need_to_lookup)
3683 {
3684 	uint_t	va_mask;
3685 	nfsstat4 status;
3686 	bitmap4 objbits;
3687 
3688 	status = bitmap4_to_attrmask(attrs, sargp);
3689 	if (status != NFS4_OK) {
3690 		/*
3691 		 * could not even figure attr mask
3692 		 */
3693 		return (status);
3694 	}
3695 	va_mask = sargp->vap->va_mask;
3696 
3697 	/*
3698 	 * dirent's d_ino is always correct value for mounted_on_fileid.
3699 	 * mntdfid_set is set once here, but mounted_on_fileid is
3700 	 * set in main dirent processing loop for each dirent.
3701 	 * The mntdfid_set is a simple optimization that lets the
3702 	 * server attr code avoid work when caller is readdir.
3703 	 */
3704 	sargp->mntdfid_set = TRUE;
3705 
3706 	/*
3707 	 * Lookup entry only if client asked for any of the following:
3708 	 * a) vattr attrs
3709 	 * b) vfs attrs
3710 	 * c) attrs w/per-object scope requested (change, filehandle, etc)
3711 	 *    other than mounted_on_fileid (which we can take from dirent)
3712 	 */
3713 	objbits = attrs ? attrs & NFS4_VP_ATTR_MASK : 0;
3714 
3715 	if (va_mask || sargp->sbp || (objbits & ~FATTR4_MOUNTED_ON_FILEID_MASK))
3716 		*need_to_lookup = TRUE;
3717 	else
3718 		*need_to_lookup = FALSE;
3719 
3720 	if (sargp->sbp == NULL)
3721 		return (NFS4_OK);
3722 
3723 	/*
3724 	 * If filesystem attrs are requested, get them now from the
3725 	 * directory vp, as most entries will have same filesystem. The only
3726 	 * exception are mounted over entries but we handle
3727 	 * those as we go (XXX mounted over detection not yet implemented).
3728 	 */
3729 	sargp->vap->va_mask = 0;	/* to avoid VOP_GETATTR */
3730 	status = bitmap4_get_sysattrs(sargp);
3731 	sargp->vap->va_mask = va_mask;
3732 
3733 	if ((status != NFS4_OK) && sargp->rdattr_error_req) {
3734 		/*
3735 		 * Failed to get filesystem attributes.
3736 		 * Return a rdattr_error for each entry, but don't fail.
3737 		 * However, don't get any obj-dependent attrs.
3738 		 */
3739 		sargp->rdattr_error = status;	/* for rdattr_error */
3740 		*need_to_lookup = FALSE;
3741 		/*
3742 		 * At least get fileid for regular readdir output
3743 		 */
3744 		sargp->vap->va_mask &= AT_NODEID;
3745 		status = NFS4_OK;
3746 	}
3747 
3748 	return (status);
3749 }
3750 
3751 /*
3752  * readlink: args: CURRENT_FH.
3753  *	res: status. If success - CURRENT_FH unchanged, return linktext.
3754  */
3755 
3756 /* ARGSUSED */
3757 static void
3758 rfs4_op_readlink(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3759     struct compound_state *cs)
3760 {
3761 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3762 	int error;
3763 	vnode_t *vp;
3764 	struct iovec iov;
3765 	struct vattr va;
3766 	struct uio uio;
3767 	char *data;
3768 	struct sockaddr *ca;
3769 	char *name = NULL;
3770 	int is_referral;
3771 
3772 	DTRACE_NFSV4_1(op__readlink__start, struct compound_state *, cs);
3773 
3774 	/* CURRENT_FH: directory */
3775 	vp = cs->vp;
3776 	if (vp == NULL) {
3777 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3778 		goto out;
3779 	}
3780 
3781 	if (cs->access == CS_ACCESS_DENIED) {
3782 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3783 		goto out;
3784 	}
3785 
3786 	/* Is it a referral? */
3787 	if (vn_is_nfs_reparse(vp, cs->cr) && client_is_downrev(req)) {
3788 
3789 		is_referral = 1;
3790 
3791 	} else {
3792 
3793 		is_referral = 0;
3794 
3795 		if (vp->v_type == VDIR) {
3796 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
3797 			goto out;
3798 		}
3799 
3800 		if (vp->v_type != VLNK) {
3801 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3802 			goto out;
3803 		}
3804 
3805 	}
3806 
3807 	va.va_mask = AT_MODE;
3808 	error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
3809 	if (error) {
3810 		*cs->statusp = resp->status = puterrno4(error);
3811 		goto out;
3812 	}
3813 
3814 	if (MANDLOCK(vp, va.va_mode)) {
3815 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3816 		goto out;
3817 	}
3818 
3819 	data = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);
3820 
3821 	if (is_referral) {
3822 		char *s;
3823 		size_t strsz;
3824 
3825 		/* Get an artificial symlink based on a referral */
3826 		s = build_symlink(vp, cs->cr, &strsz);
3827 		global_svstat_ptr[4][NFS_REFERLINKS].value.ui64++;
3828 		DTRACE_PROBE2(nfs4serv__func__referral__reflink,
3829 		    vnode_t *, vp, char *, s);
3830 		if (s == NULL)
3831 			error = EINVAL;
3832 		else {
3833 			error = 0;
3834 			(void) strlcpy(data, s, MAXPATHLEN + 1);
3835 			kmem_free(s, strsz);
3836 		}
3837 
3838 	} else {
3839 
3840 		iov.iov_base = data;
3841 		iov.iov_len = MAXPATHLEN;
3842 		uio.uio_iov = &iov;
3843 		uio.uio_iovcnt = 1;
3844 		uio.uio_segflg = UIO_SYSSPACE;
3845 		uio.uio_extflg = UIO_COPY_CACHED;
3846 		uio.uio_loffset = 0;
3847 		uio.uio_resid = MAXPATHLEN;
3848 
3849 		error = VOP_READLINK(vp, &uio, cs->cr, NULL);
3850 
3851 		if (!error)
3852 			*(data + MAXPATHLEN - uio.uio_resid) = '\0';
3853 	}
3854 
3855 	if (error) {
3856 		kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3857 		*cs->statusp = resp->status = puterrno4(error);
3858 		goto out;
3859 	}
3860 
3861 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
3862 	name = nfscmd_convname(ca, cs->exi, data, NFSCMD_CONV_OUTBOUND,
3863 	    MAXPATHLEN  + 1);
3864 
3865 	if (name == NULL) {
3866 		/*
3867 		 * Even though the conversion failed, we return
3868 		 * something. We just don't translate it.
3869 		 */
3870 		name = data;
3871 	}
3872 
3873 	/*
3874 	 * treat link name as data
3875 	 */
3876 	(void) str_to_utf8(name, &resp->link);
3877 
3878 	if (name != data)
3879 		kmem_free(name, MAXPATHLEN + 1);
3880 	kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3881 	*cs->statusp = resp->status = NFS4_OK;
3882 
3883 out:
3884 	DTRACE_NFSV4_2(op__readlink__done, struct compound_state *, cs,
3885 	    READLINK4res *, resp);
3886 }
3887 
3888 static void
3889 rfs4_op_readlink_free(nfs_resop4 *resop)
3890 {
3891 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3892 	utf8string *symlink = &resp->link;
3893 
3894 	if (symlink->utf8string_val) {
3895 		UTF8STRING_FREE(*symlink)
3896 	}
3897 }
3898 
3899 /*
3900  * release_lockowner:
3901  *	Release any state associated with the supplied
3902  *	lockowner. Note if any lo_state is holding locks we will not
3903  *	rele that lo_state and thus the lockowner will not be destroyed.
3904  *	A client using lock after the lock owner stateid has been released
3905  *	will suffer the consequence of NFS4ERR_BAD_STATEID and would have
3906  *	to reissue the lock with new_lock_owner set to TRUE.
3907  *	args: lock_owner
3908  *	res:  status
3909  */
3910 /* ARGSUSED */
3911 static void
3912 rfs4_op_release_lockowner(nfs_argop4 *argop, nfs_resop4 *resop,
3913     struct svc_req *req, struct compound_state *cs)
3914 {
3915 	RELEASE_LOCKOWNER4args *ap = &argop->nfs_argop4_u.oprelease_lockowner;
3916 	RELEASE_LOCKOWNER4res *resp = &resop->nfs_resop4_u.oprelease_lockowner;
3917 	rfs4_lockowner_t *lo;
3918 	rfs4_openowner_t *oo;
3919 	rfs4_state_t *sp;
3920 	rfs4_lo_state_t *lsp;
3921 	rfs4_client_t *cp;
3922 	bool_t create = FALSE;
3923 	locklist_t *llist;
3924 	sysid_t sysid;
3925 
3926 	DTRACE_NFSV4_2(op__release__lockowner__start, struct compound_state *,
3927 	    cs, RELEASE_LOCKOWNER4args *, ap);
3928 
3929 	/* Make sure there is a clientid around for this request */
3930 	cp = rfs4_findclient_by_id(ap->lock_owner.clientid, FALSE);
3931 
3932 	if (cp == NULL) {
3933 		*cs->statusp = resp->status =
3934 		    rfs4_check_clientid(&ap->lock_owner.clientid, 0);
3935 		goto out;
3936 	}
3937 	rfs4_client_rele(cp);
3938 
3939 	lo = rfs4_findlockowner(&ap->lock_owner, &create);
3940 	if (lo == NULL) {
3941 		*cs->statusp = resp->status = NFS4_OK;
3942 		goto out;
3943 	}
3944 	ASSERT(lo->rl_client != NULL);
3945 
3946 	/*
3947 	 * Check for EXPIRED client. If so will reap state with in a lease
3948 	 * period or on next set_clientid_confirm step
3949 	 */
3950 	if (rfs4_lease_expired(lo->rl_client)) {
3951 		rfs4_lockowner_rele(lo);
3952 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
3953 		goto out;
3954 	}
3955 
3956 	/*
3957 	 * If no sysid has been assigned, then no locks exist; just return.
3958 	 */
3959 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3960 	if (lo->rl_client->rc_sysidt == LM_NOSYSID) {
3961 		rfs4_lockowner_rele(lo);
3962 		rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3963 		goto out;
3964 	}
3965 
3966 	sysid = lo->rl_client->rc_sysidt;
3967 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3968 
3969 	/*
3970 	 * Mark the lockowner invalid.
3971 	 */
3972 	rfs4_dbe_hide(lo->rl_dbe);
3973 
3974 	/*
3975 	 * sysid-pid pair should now not be used since the lockowner is
3976 	 * invalid. If the client were to instantiate the lockowner again
3977 	 * it would be assigned a new pid. Thus we can get the list of
3978 	 * current locks.
3979 	 */
3980 
3981 	llist = flk_get_active_locks(sysid, lo->rl_pid);
3982 	/* If we are still holding locks fail */
3983 	if (llist != NULL) {
3984 
3985 		*cs->statusp = resp->status = NFS4ERR_LOCKS_HELD;
3986 
3987 		flk_free_locklist(llist);
3988 		/*
3989 		 * We need to unhide the lockowner so the client can
3990 		 * try it again. The bad thing here is if the client
3991 		 * has a logic error that took it here in the first place
3992 		 * he probably has lost accounting of the locks that it
3993 		 * is holding. So we may have dangling state until the
3994 		 * open owner state is reaped via close. One scenario
3995 		 * that could possibly occur is that the client has
3996 		 * sent the unlock request(s) in separate threads
3997 		 * and has not waited for the replies before sending the
3998 		 * RELEASE_LOCKOWNER request. Presumably, it would expect
3999 		 * and deal appropriately with NFS4ERR_LOCKS_HELD, by
4000 		 * reissuing the request.
4001 		 */
4002 		rfs4_dbe_unhide(lo->rl_dbe);
4003 		rfs4_lockowner_rele(lo);
4004 		goto out;
4005 	}
4006 
4007 	/*
4008 	 * For the corresponding client we need to check each open
4009 	 * owner for any opens that have lockowner state associated
4010 	 * with this lockowner.
4011 	 */
4012 
4013 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
4014 	for (oo = list_head(&lo->rl_client->rc_openownerlist); oo != NULL;
4015 	    oo = list_next(&lo->rl_client->rc_openownerlist, oo)) {
4016 
4017 		rfs4_dbe_lock(oo->ro_dbe);
4018 		for (sp = list_head(&oo->ro_statelist); sp != NULL;
4019 		    sp = list_next(&oo->ro_statelist, sp)) {
4020 
4021 			rfs4_dbe_lock(sp->rs_dbe);
4022 			for (lsp = list_head(&sp->rs_lostatelist);
4023 			    lsp != NULL;
4024 			    lsp = list_next(&sp->rs_lostatelist, lsp)) {
4025 				if (lsp->rls_locker == lo) {
4026 					rfs4_dbe_lock(lsp->rls_dbe);
4027 					rfs4_dbe_invalidate(lsp->rls_dbe);
4028 					rfs4_dbe_unlock(lsp->rls_dbe);
4029 				}
4030 			}
4031 			rfs4_dbe_unlock(sp->rs_dbe);
4032 		}
4033 		rfs4_dbe_unlock(oo->ro_dbe);
4034 	}
4035 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
4036 
4037 	rfs4_lockowner_rele(lo);
4038 
4039 	*cs->statusp = resp->status = NFS4_OK;
4040 
4041 out:
4042 	DTRACE_NFSV4_2(op__release__lockowner__done, struct compound_state *,
4043 	    cs, RELEASE_LOCKOWNER4res *, resp);
4044 }
4045 
4046 /*
4047  * short utility function to lookup a file and recall the delegation
4048  */
4049 static rfs4_file_t *
4050 rfs4_lookup_and_findfile(vnode_t *dvp, char *nm, vnode_t **vpp,
4051     int *lkup_error, cred_t *cr)
4052 {
4053 	vnode_t *vp;
4054 	rfs4_file_t *fp = NULL;
4055 	bool_t fcreate = FALSE;
4056 	int error;
4057 
4058 	if (vpp)
4059 		*vpp = NULL;
4060 
4061 	if ((error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cr, NULL, NULL,
4062 	    NULL)) == 0) {
4063 		if (vp->v_type == VREG)
4064 			fp = rfs4_findfile(vp, NULL, &fcreate);
4065 		if (vpp)
4066 			*vpp = vp;
4067 		else
4068 			VN_RELE(vp);
4069 	}
4070 
4071 	if (lkup_error)
4072 		*lkup_error = error;
4073 
4074 	return (fp);
4075 }
4076 
4077 /*
4078  * remove: args: CURRENT_FH: directory; name.
4079  *	res: status. If success - CURRENT_FH unchanged, return change_info
4080  *		for directory.
4081  */
4082 /* ARGSUSED */
4083 static void
4084 rfs4_op_remove(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4085     struct compound_state *cs)
4086 {
4087 	REMOVE4args *args = &argop->nfs_argop4_u.opremove;
4088 	REMOVE4res *resp = &resop->nfs_resop4_u.opremove;
4089 	int error;
4090 	vnode_t *dvp, *vp;
4091 	struct vattr bdva, idva, adva;
4092 	char *nm;
4093 	uint_t len;
4094 	rfs4_file_t *fp;
4095 	int in_crit = 0;
4096 	bslabel_t *clabel;
4097 	struct sockaddr *ca;
4098 	char *name = NULL;
4099 
4100 	DTRACE_NFSV4_2(op__remove__start, struct compound_state *, cs,
4101 	    REMOVE4args *, args);
4102 
4103 	/* CURRENT_FH: directory */
4104 	dvp = cs->vp;
4105 	if (dvp == NULL) {
4106 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4107 		goto out;
4108 	}
4109 
4110 	if (cs->access == CS_ACCESS_DENIED) {
4111 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4112 		goto out;
4113 	}
4114 
4115 	/*
4116 	 * If there is an unshared filesystem mounted on this vnode,
4117 	 * Do not allow to remove anything in this directory.
4118 	 */
4119 	if (vn_ismntpt(dvp)) {
4120 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4121 		goto out;
4122 	}
4123 
4124 	if (dvp->v_type != VDIR) {
4125 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4126 		goto out;
4127 	}
4128 
4129 	if (!utf8_dir_verify(&args->target)) {
4130 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4131 		goto out;
4132 	}
4133 
4134 	/*
4135 	 * Lookup the file so that we can check if it's a directory
4136 	 */
4137 	nm = utf8_to_fn(&args->target, &len, NULL);
4138 	if (nm == NULL) {
4139 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4140 		goto out;
4141 	}
4142 
4143 	if (len > MAXNAMELEN) {
4144 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4145 		kmem_free(nm, len);
4146 		goto out;
4147 	}
4148 
4149 	if (rdonly4(cs->exi, cs->vp, req)) {
4150 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4151 		kmem_free(nm, len);
4152 		goto out;
4153 	}
4154 
4155 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4156 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
4157 	    MAXPATHLEN  + 1);
4158 
4159 	if (name == NULL) {
4160 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4161 		kmem_free(nm, len);
4162 		goto out;
4163 	}
4164 
4165 	/*
4166 	 * Lookup the file to determine type and while we are see if
4167 	 * there is a file struct around and check for delegation.
4168 	 * We don't need to acquire va_seq before this lookup, if
4169 	 * it causes an update, cinfo.before will not match, which will
4170 	 * trigger a cache flush even if atomic is TRUE.
4171 	 */
4172 	if (fp = rfs4_lookup_and_findfile(dvp, name, &vp, &error, cs->cr)) {
4173 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4174 		    NULL)) {
4175 			VN_RELE(vp);
4176 			rfs4_file_rele(fp);
4177 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4178 			if (nm != name)
4179 				kmem_free(name, MAXPATHLEN + 1);
4180 			kmem_free(nm, len);
4181 			goto out;
4182 		}
4183 	}
4184 
4185 	/* Didn't find anything to remove */
4186 	if (vp == NULL) {
4187 		*cs->statusp = resp->status = error;
4188 		if (nm != name)
4189 			kmem_free(name, MAXPATHLEN + 1);
4190 		kmem_free(nm, len);
4191 		goto out;
4192 	}
4193 
4194 	if (nbl_need_check(vp)) {
4195 		nbl_start_crit(vp, RW_READER);
4196 		in_crit = 1;
4197 		if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
4198 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4199 			if (nm != name)
4200 				kmem_free(name, MAXPATHLEN + 1);
4201 			kmem_free(nm, len);
4202 			nbl_end_crit(vp);
4203 			VN_RELE(vp);
4204 			if (fp) {
4205 				rfs4_clear_dont_grant(fp);
4206 				rfs4_file_rele(fp);
4207 			}
4208 			goto out;
4209 		}
4210 	}
4211 
4212 	/* check label before allowing removal */
4213 	if (is_system_labeled()) {
4214 		ASSERT(req->rq_label != NULL);
4215 		clabel = req->rq_label;
4216 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
4217 		    "got client label from request(1)",
4218 		    struct svc_req *, req);
4219 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4220 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
4221 			    cs->exi)) {
4222 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4223 				if (name != nm)
4224 					kmem_free(name, MAXPATHLEN + 1);
4225 				kmem_free(nm, len);
4226 				if (in_crit)
4227 					nbl_end_crit(vp);
4228 				VN_RELE(vp);
4229 				if (fp) {
4230 					rfs4_clear_dont_grant(fp);
4231 					rfs4_file_rele(fp);
4232 				}
4233 				goto out;
4234 			}
4235 		}
4236 	}
4237 
4238 	/* Get dir "before" change value */
4239 	bdva.va_mask = AT_CTIME|AT_SEQ;
4240 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
4241 	if (error) {
4242 		*cs->statusp = resp->status = puterrno4(error);
4243 		if (nm != name)
4244 			kmem_free(name, MAXPATHLEN + 1);
4245 		kmem_free(nm, len);
4246 		if (in_crit)
4247 			nbl_end_crit(vp);
4248 		VN_RELE(vp);
4249 		if (fp) {
4250 			rfs4_clear_dont_grant(fp);
4251 			rfs4_file_rele(fp);
4252 		}
4253 		goto out;
4254 	}
4255 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
4256 
4257 	/* Actually do the REMOVE operation */
4258 	if (vp->v_type == VDIR) {
4259 		/*
4260 		 * Can't remove a directory that has a mounted-on filesystem.
4261 		 */
4262 		if (vn_ismntpt(vp)) {
4263 			error = EACCES;
4264 		} else {
4265 			/*
4266 			 * System V defines rmdir to return EEXIST,
4267 			 * not ENOTEMPTY, if the directory is not
4268 			 * empty.  A System V NFS server needs to map
4269 			 * NFS4ERR_EXIST to NFS4ERR_NOTEMPTY to
4270 			 * transmit over the wire.
4271 			 */
4272 			if ((error = VOP_RMDIR(dvp, name, rootdir, cs->cr,
4273 			    NULL, 0)) == EEXIST)
4274 				error = ENOTEMPTY;
4275 		}
4276 	} else {
4277 		if ((error = VOP_REMOVE(dvp, name, cs->cr, NULL, 0)) == 0 &&
4278 		    fp != NULL) {
4279 			struct vattr va;
4280 			vnode_t *tvp;
4281 
4282 			rfs4_dbe_lock(fp->rf_dbe);
4283 			tvp = fp->rf_vp;
4284 			if (tvp)
4285 				VN_HOLD(tvp);
4286 			rfs4_dbe_unlock(fp->rf_dbe);
4287 
4288 			if (tvp) {
4289 				/*
4290 				 * This is va_seq safe because we are not
4291 				 * manipulating dvp.
4292 				 */
4293 				va.va_mask = AT_NLINK;
4294 				if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4295 				    va.va_nlink == 0) {
4296 					/* Remove state on file remove */
4297 					if (in_crit) {
4298 						nbl_end_crit(vp);
4299 						in_crit = 0;
4300 					}
4301 					rfs4_close_all_state(fp);
4302 				}
4303 				VN_RELE(tvp);
4304 			}
4305 		}
4306 	}
4307 
4308 	if (in_crit)
4309 		nbl_end_crit(vp);
4310 	VN_RELE(vp);
4311 
4312 	if (fp) {
4313 		rfs4_clear_dont_grant(fp);
4314 		rfs4_file_rele(fp);
4315 	}
4316 	if (nm != name)
4317 		kmem_free(name, MAXPATHLEN + 1);
4318 	kmem_free(nm, len);
4319 
4320 	if (error) {
4321 		*cs->statusp = resp->status = puterrno4(error);
4322 		goto out;
4323 	}
4324 
4325 	/*
4326 	 * Get the initial "after" sequence number, if it fails, set to zero
4327 	 */
4328 	idva.va_mask = AT_SEQ;
4329 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
4330 		idva.va_seq = 0;
4331 
4332 	/*
4333 	 * Force modified data and metadata out to stable storage.
4334 	 */
4335 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
4336 
4337 	/*
4338 	 * Get "after" change value, if it fails, simply return the
4339 	 * before value.
4340 	 */
4341 	adva.va_mask = AT_CTIME|AT_SEQ;
4342 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
4343 		adva.va_ctime = bdva.va_ctime;
4344 		adva.va_seq = 0;
4345 	}
4346 
4347 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
4348 
4349 	/*
4350 	 * The cinfo.atomic = TRUE only if we have
4351 	 * non-zero va_seq's, and it has incremented by exactly one
4352 	 * during the VOP_REMOVE/RMDIR and it didn't change during
4353 	 * the VOP_FSYNC.
4354 	 */
4355 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
4356 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
4357 		resp->cinfo.atomic = TRUE;
4358 	else
4359 		resp->cinfo.atomic = FALSE;
4360 
4361 	*cs->statusp = resp->status = NFS4_OK;
4362 
4363 out:
4364 	DTRACE_NFSV4_2(op__remove__done, struct compound_state *, cs,
4365 	    REMOVE4res *, resp);
4366 }
4367 
4368 /*
4369  * rename: args: SAVED_FH: from directory, CURRENT_FH: target directory,
4370  *		oldname and newname.
4371  *	res: status. If success - CURRENT_FH unchanged, return change_info
4372  *		for both from and target directories.
4373  */
4374 /* ARGSUSED */
4375 static void
4376 rfs4_op_rename(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4377     struct compound_state *cs)
4378 {
4379 	RENAME4args *args = &argop->nfs_argop4_u.oprename;
4380 	RENAME4res *resp = &resop->nfs_resop4_u.oprename;
4381 	int error;
4382 	vnode_t *odvp;
4383 	vnode_t *ndvp;
4384 	vnode_t *srcvp, *targvp;
4385 	struct vattr obdva, oidva, oadva;
4386 	struct vattr nbdva, nidva, nadva;
4387 	char *onm, *nnm;
4388 	uint_t olen, nlen;
4389 	rfs4_file_t *fp, *sfp;
4390 	int in_crit_src, in_crit_targ;
4391 	int fp_rele_grant_hold, sfp_rele_grant_hold;
4392 	bslabel_t *clabel;
4393 	struct sockaddr *ca;
4394 	char *converted_onm = NULL;
4395 	char *converted_nnm = NULL;
4396 
4397 	DTRACE_NFSV4_2(op__rename__start, struct compound_state *, cs,
4398 	    RENAME4args *, args);
4399 
4400 	fp = sfp = NULL;
4401 	srcvp = targvp = NULL;
4402 	in_crit_src = in_crit_targ = 0;
4403 	fp_rele_grant_hold = sfp_rele_grant_hold = 0;
4404 
4405 	/* CURRENT_FH: target directory */
4406 	ndvp = cs->vp;
4407 	if (ndvp == NULL) {
4408 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4409 		goto out;
4410 	}
4411 
4412 	/* SAVED_FH: from directory */
4413 	odvp = cs->saved_vp;
4414 	if (odvp == NULL) {
4415 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4416 		goto out;
4417 	}
4418 
4419 	if (cs->access == CS_ACCESS_DENIED) {
4420 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4421 		goto out;
4422 	}
4423 
4424 	/*
4425 	 * If there is an unshared filesystem mounted on this vnode,
4426 	 * do not allow to rename objects in this directory.
4427 	 */
4428 	if (vn_ismntpt(odvp)) {
4429 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4430 		goto out;
4431 	}
4432 
4433 	/*
4434 	 * If there is an unshared filesystem mounted on this vnode,
4435 	 * do not allow to rename to this directory.
4436 	 */
4437 	if (vn_ismntpt(ndvp)) {
4438 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4439 		goto out;
4440 	}
4441 
4442 	if (odvp->v_type != VDIR || ndvp->v_type != VDIR) {
4443 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4444 		goto out;
4445 	}
4446 
4447 	if (cs->saved_exi != cs->exi) {
4448 		*cs->statusp = resp->status = NFS4ERR_XDEV;
4449 		goto out;
4450 	}
4451 
4452 	if (!utf8_dir_verify(&args->oldname)) {
4453 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4454 		goto out;
4455 	}
4456 
4457 	if (!utf8_dir_verify(&args->newname)) {
4458 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4459 		goto out;
4460 	}
4461 
4462 	onm = utf8_to_fn(&args->oldname, &olen, NULL);
4463 	if (onm == NULL) {
4464 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4465 		goto out;
4466 	}
4467 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4468 	nlen = MAXPATHLEN + 1;
4469 	converted_onm = nfscmd_convname(ca, cs->exi, onm, NFSCMD_CONV_INBOUND,
4470 	    nlen);
4471 
4472 	if (converted_onm == NULL) {
4473 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4474 		kmem_free(onm, olen);
4475 		goto out;
4476 	}
4477 
4478 	nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4479 	if (nnm == NULL) {
4480 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4481 		if (onm != converted_onm)
4482 			kmem_free(converted_onm, MAXPATHLEN + 1);
4483 		kmem_free(onm, olen);
4484 		goto out;
4485 	}
4486 	converted_nnm = nfscmd_convname(ca, cs->exi, nnm, NFSCMD_CONV_INBOUND,
4487 	    MAXPATHLEN  + 1);
4488 
4489 	if (converted_nnm == NULL) {
4490 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4491 		kmem_free(nnm, nlen);
4492 		nnm = NULL;
4493 		if (onm != converted_onm)
4494 			kmem_free(converted_onm, MAXPATHLEN + 1);
4495 		kmem_free(onm, olen);
4496 		goto out;
4497 	}
4498 
4499 
4500 	if (olen > MAXNAMELEN || nlen > MAXNAMELEN) {
4501 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4502 		kmem_free(onm, olen);
4503 		kmem_free(nnm, nlen);
4504 		goto out;
4505 	}
4506 
4507 
4508 	if (rdonly4(cs->exi, cs->vp, req)) {
4509 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4510 		if (onm != converted_onm)
4511 			kmem_free(converted_onm, MAXPATHLEN + 1);
4512 		kmem_free(onm, olen);
4513 		if (nnm != converted_nnm)
4514 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4515 		kmem_free(nnm, nlen);
4516 		goto out;
4517 	}
4518 
4519 	/* check label of the target dir */
4520 	if (is_system_labeled()) {
4521 		ASSERT(req->rq_label != NULL);
4522 		clabel = req->rq_label;
4523 		DTRACE_PROBE2(tx__rfs4__log__info__oprename__clabel, char *,
4524 		    "got client label from request(1)",
4525 		    struct svc_req *, req);
4526 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4527 			if (!do_rfs_label_check(clabel, ndvp,
4528 			    EQUALITY_CHECK, cs->exi)) {
4529 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4530 				goto err_out;
4531 			}
4532 		}
4533 	}
4534 
4535 	/*
4536 	 * Is the source a file and have a delegation?
4537 	 * We don't need to acquire va_seq before these lookups, if
4538 	 * it causes an update, cinfo.before will not match, which will
4539 	 * trigger a cache flush even if atomic is TRUE.
4540 	 */
4541 	if (sfp = rfs4_lookup_and_findfile(odvp, converted_onm, &srcvp,
4542 	    &error, cs->cr)) {
4543 		if (rfs4_check_delegated_byfp(FWRITE, sfp, TRUE, TRUE, TRUE,
4544 		    NULL)) {
4545 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4546 			goto err_out;
4547 		}
4548 	}
4549 
4550 	if (srcvp == NULL) {
4551 		*cs->statusp = resp->status = puterrno4(error);
4552 		if (onm != converted_onm)
4553 			kmem_free(converted_onm, MAXPATHLEN + 1);
4554 		kmem_free(onm, olen);
4555 		if (nnm != converted_nnm)
4556 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4557 		kmem_free(nnm, nlen);
4558 		goto out;
4559 	}
4560 
4561 	sfp_rele_grant_hold = 1;
4562 
4563 	/* Does the destination exist and a file and have a delegation? */
4564 	if (fp = rfs4_lookup_and_findfile(ndvp, converted_nnm, &targvp,
4565 	    NULL, cs->cr)) {
4566 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4567 		    NULL)) {
4568 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4569 			goto err_out;
4570 		}
4571 	}
4572 	fp_rele_grant_hold = 1;
4573 
4574 
4575 	/* Check for NBMAND lock on both source and target */
4576 	if (nbl_need_check(srcvp)) {
4577 		nbl_start_crit(srcvp, RW_READER);
4578 		in_crit_src = 1;
4579 		if (nbl_conflict(srcvp, NBL_RENAME, 0, 0, 0, NULL)) {
4580 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4581 			goto err_out;
4582 		}
4583 	}
4584 
4585 	if (targvp && nbl_need_check(targvp)) {
4586 		nbl_start_crit(targvp, RW_READER);
4587 		in_crit_targ = 1;
4588 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
4589 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4590 			goto err_out;
4591 		}
4592 	}
4593 
4594 	/* Get source "before" change value */
4595 	obdva.va_mask = AT_CTIME|AT_SEQ;
4596 	error = VOP_GETATTR(odvp, &obdva, 0, cs->cr, NULL);
4597 	if (!error) {
4598 		nbdva.va_mask = AT_CTIME|AT_SEQ;
4599 		error = VOP_GETATTR(ndvp, &nbdva, 0, cs->cr, NULL);
4600 	}
4601 	if (error) {
4602 		*cs->statusp = resp->status = puterrno4(error);
4603 		goto err_out;
4604 	}
4605 
4606 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.before, obdva.va_ctime)
4607 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.before, nbdva.va_ctime)
4608 
4609 	if ((error = VOP_RENAME(odvp, converted_onm, ndvp, converted_nnm,
4610 	    cs->cr, NULL, 0)) == 0 && fp != NULL) {
4611 		struct vattr va;
4612 		vnode_t *tvp;
4613 
4614 		rfs4_dbe_lock(fp->rf_dbe);
4615 		tvp = fp->rf_vp;
4616 		if (tvp)
4617 			VN_HOLD(tvp);
4618 		rfs4_dbe_unlock(fp->rf_dbe);
4619 
4620 		if (tvp) {
4621 			va.va_mask = AT_NLINK;
4622 			if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4623 			    va.va_nlink == 0) {
4624 				/* The file is gone and so should the state */
4625 				if (in_crit_targ) {
4626 					nbl_end_crit(targvp);
4627 					in_crit_targ = 0;
4628 				}
4629 				rfs4_close_all_state(fp);
4630 			}
4631 			VN_RELE(tvp);
4632 		}
4633 	}
4634 	if (error == 0)
4635 		vn_renamepath(ndvp, srcvp, nnm, nlen - 1);
4636 
4637 	if (in_crit_src)
4638 		nbl_end_crit(srcvp);
4639 	if (srcvp)
4640 		VN_RELE(srcvp);
4641 	if (in_crit_targ)
4642 		nbl_end_crit(targvp);
4643 	if (targvp)
4644 		VN_RELE(targvp);
4645 
4646 	if (sfp) {
4647 		rfs4_clear_dont_grant(sfp);
4648 		rfs4_file_rele(sfp);
4649 	}
4650 	if (fp) {
4651 		rfs4_clear_dont_grant(fp);
4652 		rfs4_file_rele(fp);
4653 	}
4654 
4655 	if (converted_onm != onm)
4656 		kmem_free(converted_onm, MAXPATHLEN + 1);
4657 	kmem_free(onm, olen);
4658 	if (converted_nnm != nnm)
4659 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4660 	kmem_free(nnm, nlen);
4661 
4662 	/*
4663 	 * Get the initial "after" sequence number, if it fails, set to zero
4664 	 */
4665 	oidva.va_mask = AT_SEQ;
4666 	if (VOP_GETATTR(odvp, &oidva, 0, cs->cr, NULL))
4667 		oidva.va_seq = 0;
4668 
4669 	nidva.va_mask = AT_SEQ;
4670 	if (VOP_GETATTR(ndvp, &nidva, 0, cs->cr, NULL))
4671 		nidva.va_seq = 0;
4672 
4673 	/*
4674 	 * Force modified data and metadata out to stable storage.
4675 	 */
4676 	(void) VOP_FSYNC(odvp, 0, cs->cr, NULL);
4677 	(void) VOP_FSYNC(ndvp, 0, cs->cr, NULL);
4678 
4679 	if (error) {
4680 		*cs->statusp = resp->status = puterrno4(error);
4681 		goto out;
4682 	}
4683 
4684 	/*
4685 	 * Get "after" change values, if it fails, simply return the
4686 	 * before value.
4687 	 */
4688 	oadva.va_mask = AT_CTIME|AT_SEQ;
4689 	if (VOP_GETATTR(odvp, &oadva, 0, cs->cr, NULL)) {
4690 		oadva.va_ctime = obdva.va_ctime;
4691 		oadva.va_seq = 0;
4692 	}
4693 
4694 	nadva.va_mask = AT_CTIME|AT_SEQ;
4695 	if (VOP_GETATTR(odvp, &nadva, 0, cs->cr, NULL)) {
4696 		nadva.va_ctime = nbdva.va_ctime;
4697 		nadva.va_seq = 0;
4698 	}
4699 
4700 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.after, oadva.va_ctime)
4701 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.after, nadva.va_ctime)
4702 
4703 	/*
4704 	 * The cinfo.atomic = TRUE only if we have
4705 	 * non-zero va_seq's, and it has incremented by exactly one
4706 	 * during the VOP_RENAME and it didn't change during the VOP_FSYNC.
4707 	 */
4708 	if (obdva.va_seq && oidva.va_seq && oadva.va_seq &&
4709 	    oidva.va_seq == (obdva.va_seq + 1) && oidva.va_seq == oadva.va_seq)
4710 		resp->source_cinfo.atomic = TRUE;
4711 	else
4712 		resp->source_cinfo.atomic = FALSE;
4713 
4714 	if (nbdva.va_seq && nidva.va_seq && nadva.va_seq &&
4715 	    nidva.va_seq == (nbdva.va_seq + 1) && nidva.va_seq == nadva.va_seq)
4716 		resp->target_cinfo.atomic = TRUE;
4717 	else
4718 		resp->target_cinfo.atomic = FALSE;
4719 
4720 #ifdef	VOLATILE_FH_TEST
4721 	{
4722 	extern void add_volrnm_fh(struct exportinfo *, vnode_t *);
4723 
4724 	/*
4725 	 * Add the renamed file handle to the volatile rename list
4726 	 */
4727 	if (cs->exi->exi_export.ex_flags & EX_VOLRNM) {
4728 		/* file handles may expire on rename */
4729 		vnode_t *vp;
4730 
4731 		nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4732 		/*
4733 		 * Already know that nnm will be a valid string
4734 		 */
4735 		error = VOP_LOOKUP(ndvp, nnm, &vp, NULL, 0, NULL, cs->cr,
4736 		    NULL, NULL, NULL);
4737 		kmem_free(nnm, nlen);
4738 		if (!error) {
4739 			add_volrnm_fh(cs->exi, vp);
4740 			VN_RELE(vp);
4741 		}
4742 	}
4743 	}
4744 #endif	/* VOLATILE_FH_TEST */
4745 
4746 	*cs->statusp = resp->status = NFS4_OK;
4747 out:
4748 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4749 	    RENAME4res *, resp);
4750 	return;
4751 
4752 err_out:
4753 	if (onm != converted_onm)
4754 		kmem_free(converted_onm, MAXPATHLEN + 1);
4755 	if (onm != NULL)
4756 		kmem_free(onm, olen);
4757 	if (nnm != converted_nnm)
4758 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4759 	if (nnm != NULL)
4760 		kmem_free(nnm, nlen);
4761 
4762 	if (in_crit_src) nbl_end_crit(srcvp);
4763 	if (in_crit_targ) nbl_end_crit(targvp);
4764 	if (targvp) VN_RELE(targvp);
4765 	if (srcvp) VN_RELE(srcvp);
4766 	if (sfp) {
4767 		if (sfp_rele_grant_hold) rfs4_clear_dont_grant(sfp);
4768 		rfs4_file_rele(sfp);
4769 	}
4770 	if (fp) {
4771 		if (fp_rele_grant_hold) rfs4_clear_dont_grant(fp);
4772 		rfs4_file_rele(fp);
4773 	}
4774 
4775 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4776 	    RENAME4res *, resp);
4777 }
4778 
4779 /* ARGSUSED */
4780 static void
4781 rfs4_op_renew(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4782     struct compound_state *cs)
4783 {
4784 	RENEW4args *args = &argop->nfs_argop4_u.oprenew;
4785 	RENEW4res *resp = &resop->nfs_resop4_u.oprenew;
4786 	rfs4_client_t *cp;
4787 
4788 	DTRACE_NFSV4_2(op__renew__start, struct compound_state *, cs,
4789 	    RENEW4args *, args);
4790 
4791 	if ((cp = rfs4_findclient_by_id(args->clientid, FALSE)) == NULL) {
4792 		*cs->statusp = resp->status =
4793 		    rfs4_check_clientid(&args->clientid, 0);
4794 		goto out;
4795 	}
4796 
4797 	if (rfs4_lease_expired(cp)) {
4798 		rfs4_client_rele(cp);
4799 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
4800 		goto out;
4801 	}
4802 
4803 	rfs4_update_lease(cp);
4804 
4805 	mutex_enter(cp->rc_cbinfo.cb_lock);
4806 	if (cp->rc_cbinfo.cb_notified_of_cb_path_down == FALSE) {
4807 		cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
4808 		*cs->statusp = resp->status = NFS4ERR_CB_PATH_DOWN;
4809 	} else {
4810 		*cs->statusp = resp->status = NFS4_OK;
4811 	}
4812 	mutex_exit(cp->rc_cbinfo.cb_lock);
4813 
4814 	rfs4_client_rele(cp);
4815 
4816 out:
4817 	DTRACE_NFSV4_2(op__renew__done, struct compound_state *, cs,
4818 	    RENEW4res *, resp);
4819 }
4820 
4821 /* ARGSUSED */
4822 static void
4823 rfs4_op_restorefh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
4824     struct compound_state *cs)
4825 {
4826 	RESTOREFH4res *resp = &resop->nfs_resop4_u.oprestorefh;
4827 
4828 	DTRACE_NFSV4_1(op__restorefh__start, struct compound_state *, cs);
4829 
4830 	/* No need to check cs->access - we are not accessing any object */
4831 	if ((cs->saved_vp == NULL) || (cs->saved_fh.nfs_fh4_val == NULL)) {
4832 		*cs->statusp = resp->status = NFS4ERR_RESTOREFH;
4833 		goto out;
4834 	}
4835 	if (cs->vp != NULL) {
4836 		VN_RELE(cs->vp);
4837 	}
4838 	cs->vp = cs->saved_vp;
4839 	cs->saved_vp = NULL;
4840 	cs->exi = cs->saved_exi;
4841 	nfs_fh4_copy(&cs->saved_fh, &cs->fh);
4842 	*cs->statusp = resp->status = NFS4_OK;
4843 	cs->deleg = FALSE;
4844 
4845 out:
4846 	DTRACE_NFSV4_2(op__restorefh__done, struct compound_state *, cs,
4847 	    RESTOREFH4res *, resp);
4848 }
4849 
4850 /* ARGSUSED */
4851 static void
4852 rfs4_op_savefh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4853     struct compound_state *cs)
4854 {
4855 	SAVEFH4res *resp = &resop->nfs_resop4_u.opsavefh;
4856 
4857 	DTRACE_NFSV4_1(op__savefh__start, struct compound_state *, cs);
4858 
4859 	/* No need to check cs->access - we are not accessing any object */
4860 	if (cs->vp == NULL) {
4861 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4862 		goto out;
4863 	}
4864 	if (cs->saved_vp != NULL) {
4865 		VN_RELE(cs->saved_vp);
4866 	}
4867 	cs->saved_vp = cs->vp;
4868 	VN_HOLD(cs->saved_vp);
4869 	cs->saved_exi = cs->exi;
4870 	/*
4871 	 * since SAVEFH is fairly rare, don't alloc space for its fh
4872 	 * unless necessary.
4873 	 */
4874 	if (cs->saved_fh.nfs_fh4_val == NULL) {
4875 		cs->saved_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP);
4876 	}
4877 	nfs_fh4_copy(&cs->fh, &cs->saved_fh);
4878 	*cs->statusp = resp->status = NFS4_OK;
4879 
4880 out:
4881 	DTRACE_NFSV4_2(op__savefh__done, struct compound_state *, cs,
4882 	    SAVEFH4res *, resp);
4883 }
4884 
4885 /*
4886  * rfs4_verify_attr is called when nfsv4 Setattr failed, but we wish to
4887  * return the bitmap of attrs that were set successfully. It is also
4888  * called by Verify/Nverify to test the vattr/vfsstat attrs. It should
4889  * always be called only after rfs4_do_set_attrs().
4890  *
4891  * Verify that the attributes are same as the expected ones. sargp->vap
4892  * and sargp->sbp contain the input attributes as translated from fattr4.
4893  *
4894  * This function verifies only the attrs that correspond to a vattr or
4895  * vfsstat struct. That is because of the extra step needed to get the
4896  * corresponding system structs. Other attributes have already been set or
4897  * verified by do_rfs4_set_attrs.
4898  *
4899  * Return 0 if all attrs match, -1 if some don't, error if error processing.
4900  */
4901 static int
4902 rfs4_verify_attr(struct nfs4_svgetit_arg *sargp,
4903     bitmap4 *resp, struct nfs4_ntov_table *ntovp)
4904 {
4905 	int error, ret_error = 0;
4906 	int i, k;
4907 	uint_t sva_mask = sargp->vap->va_mask;
4908 	uint_t vbit;
4909 	union nfs4_attr_u *na;
4910 	uint8_t *amap;
4911 	bool_t getsb = ntovp->vfsstat;
4912 
4913 	if (sva_mask != 0) {
4914 		/*
4915 		 * Okay to overwrite sargp->vap because we verify based
4916 		 * on the incoming values.
4917 		 */
4918 		ret_error = VOP_GETATTR(sargp->cs->vp, sargp->vap, 0,
4919 		    sargp->cs->cr, NULL);
4920 		if (ret_error) {
4921 			if (resp == NULL)
4922 				return (ret_error);
4923 			/*
4924 			 * Must return bitmap of successful attrs
4925 			 */
4926 			sva_mask = 0;	/* to prevent checking vap later */
4927 		} else {
4928 			/*
4929 			 * Some file systems clobber va_mask. it is probably
4930 			 * wrong of them to do so, nonethless we practice
4931 			 * defensive coding.
4932 			 * See bug id 4276830.
4933 			 */
4934 			sargp->vap->va_mask = sva_mask;
4935 		}
4936 	}
4937 
4938 	if (getsb) {
4939 		/*
4940 		 * Now get the superblock and loop on the bitmap, as there is
4941 		 * no simple way of translating from superblock to bitmap4.
4942 		 */
4943 		ret_error = VFS_STATVFS(sargp->cs->vp->v_vfsp, sargp->sbp);
4944 		if (ret_error) {
4945 			if (resp == NULL)
4946 				goto errout;
4947 			getsb = FALSE;
4948 		}
4949 	}
4950 
4951 	/*
4952 	 * Now loop and verify each attribute which getattr returned
4953 	 * whether it's the same as the input.
4954 	 */
4955 	if (resp == NULL && !getsb && (sva_mask == 0))
4956 		goto errout;
4957 
4958 	na = ntovp->na;
4959 	amap = ntovp->amap;
4960 	k = 0;
4961 	for (i = 0; i < ntovp->attrcnt; i++, na++, amap++) {
4962 		k = *amap;
4963 		ASSERT(nfs4_ntov_map[k].nval == k);
4964 		vbit = nfs4_ntov_map[k].vbit;
4965 
4966 		/*
4967 		 * If vattr attribute but VOP_GETATTR failed, or it's
4968 		 * superblock attribute but VFS_STATVFS failed, skip
4969 		 */
4970 		if (vbit) {
4971 			if ((vbit & sva_mask) == 0)
4972 				continue;
4973 		} else if (!(getsb && nfs4_ntov_map[k].vfsstat)) {
4974 			continue;
4975 		}
4976 		error = (*nfs4_ntov_map[k].sv_getit)(NFS4ATTR_VERIT, sargp, na);
4977 		if (resp != NULL) {
4978 			if (error)
4979 				ret_error = -1;	/* not all match */
4980 			else	/* update response bitmap */
4981 				*resp |= nfs4_ntov_map[k].fbit;
4982 			continue;
4983 		}
4984 		if (error) {
4985 			ret_error = -1;	/* not all match */
4986 			break;
4987 		}
4988 	}
4989 errout:
4990 	return (ret_error);
4991 }
4992 
4993 /*
4994  * Decode the attribute to be set/verified. If the attr requires a sys op
4995  * (VOP_GETATTR, VFS_VFSSTAT), and the request is to verify, then don't
4996  * call the sv_getit function for it, because the sys op hasn't yet been done.
4997  * Return 0 for success, error code if failed.
4998  *
4999  * Note: the decoded arg is not freed here but in nfs4_ntov_table_free.
5000  */
5001 static int
5002 decode_fattr4_attr(nfs4_attr_cmd_t cmd, struct nfs4_svgetit_arg *sargp,
5003     int k, XDR *xdrp, bitmap4 *resp_bval, union nfs4_attr_u *nap)
5004 {
5005 	int error = 0;
5006 	bool_t set_later;
5007 
5008 	sargp->vap->va_mask |= nfs4_ntov_map[k].vbit;
5009 
5010 	if ((*nfs4_ntov_map[k].xfunc)(xdrp, nap)) {
5011 		set_later = nfs4_ntov_map[k].vbit || nfs4_ntov_map[k].vfsstat;
5012 		/*
5013 		 * don't verify yet if a vattr or sb dependent attr,
5014 		 * because we don't have their sys values yet.
5015 		 * Will be done later.
5016 		 */
5017 		if (! (set_later && (cmd == NFS4ATTR_VERIT))) {
5018 			/*
5019 			 * ACLs are a special case, since setting the MODE
5020 			 * conflicts with setting the ACL.  We delay setting
5021 			 * the ACL until all other attributes have been set.
5022 			 * The ACL gets set in do_rfs4_op_setattr().
5023 			 */
5024 			if (nfs4_ntov_map[k].fbit != FATTR4_ACL_MASK) {
5025 				error = (*nfs4_ntov_map[k].sv_getit)(cmd,
5026 				    sargp, nap);
5027 				if (error) {
5028 					xdr_free(nfs4_ntov_map[k].xfunc,
5029 					    (caddr_t)nap);
5030 				}
5031 			}
5032 		}
5033 	} else {
5034 #ifdef  DEBUG
5035 		cmn_err(CE_NOTE, "decode_fattr4_attr: error "
5036 		    "decoding attribute %d\n", k);
5037 #endif
5038 		error = EINVAL;
5039 	}
5040 	if (!error && resp_bval && !set_later) {
5041 		*resp_bval |= nfs4_ntov_map[k].fbit;
5042 	}
5043 
5044 	return (error);
5045 }
5046 
5047 /*
5048  * Set vattr based on incoming fattr4 attrs - used by setattr.
5049  * Set response mask. Ignore any values that are not writable vattr attrs.
5050  */
5051 static nfsstat4
5052 do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5053     struct nfs4_svgetit_arg *sargp, struct nfs4_ntov_table *ntovp,
5054     nfs4_attr_cmd_t cmd)
5055 {
5056 	int error = 0;
5057 	int i;
5058 	char *attrs = fattrp->attrlist4;
5059 	uint32_t attrslen = fattrp->attrlist4_len;
5060 	XDR xdr;
5061 	nfsstat4 status = NFS4_OK;
5062 	vnode_t *vp = cs->vp;
5063 	union nfs4_attr_u *na;
5064 	uint8_t *amap;
5065 
5066 #ifndef lint
5067 	/*
5068 	 * Make sure that maximum attribute number can be expressed as an
5069 	 * 8 bit quantity.
5070 	 */
5071 	ASSERT(NFS4_MAXNUM_ATTRS <= (UINT8_MAX + 1));
5072 #endif
5073 
5074 	if (vp == NULL) {
5075 		if (resp)
5076 			*resp = 0;
5077 		return (NFS4ERR_NOFILEHANDLE);
5078 	}
5079 	if (cs->access == CS_ACCESS_DENIED) {
5080 		if (resp)
5081 			*resp = 0;
5082 		return (NFS4ERR_ACCESS);
5083 	}
5084 
5085 	sargp->op = cmd;
5086 	sargp->cs = cs;
5087 	sargp->flag = 0;	/* may be set later */
5088 	sargp->vap->va_mask = 0;
5089 	sargp->rdattr_error = NFS4_OK;
5090 	sargp->rdattr_error_req = FALSE;
5091 	/* sargp->sbp is set by the caller */
5092 
5093 	xdrmem_create(&xdr, attrs, attrslen, XDR_DECODE);
5094 
5095 	na = ntovp->na;
5096 	amap = ntovp->amap;
5097 
5098 	/*
5099 	 * The following loop iterates on the nfs4_ntov_map checking
5100 	 * if the fbit is set in the requested bitmap.
5101 	 * If set then we process the arguments using the
5102 	 * rfs4_fattr4 conversion functions to populate the setattr
5103 	 * vattr and va_mask. Any settable attrs that are not using vattr
5104 	 * will be set in this loop.
5105 	 */
5106 	for (i = 0; i < nfs4_ntov_map_size; i++) {
5107 		if (!(fattrp->attrmask & nfs4_ntov_map[i].fbit)) {
5108 			continue;
5109 		}
5110 		/*
5111 		 * If setattr, must be a writable attr.
5112 		 * If verify/nverify, must be a readable attr.
5113 		 */
5114 		if ((error = (*nfs4_ntov_map[i].sv_getit)(
5115 		    NFS4ATTR_SUPPORTED, sargp, NULL)) != 0) {
5116 			/*
5117 			 * Client tries to set/verify an
5118 			 * unsupported attribute, tries to set
5119 			 * a read only attr or verify a write
5120 			 * only one - error!
5121 			 */
5122 			break;
5123 		}
5124 		/*
5125 		 * Decode the attribute to set/verify
5126 		 */
5127 		error = decode_fattr4_attr(cmd, sargp, nfs4_ntov_map[i].nval,
5128 		    &xdr, resp ? resp : NULL, na);
5129 		if (error)
5130 			break;
5131 		*amap++ = (uint8_t)nfs4_ntov_map[i].nval;
5132 		na++;
5133 		(ntovp->attrcnt)++;
5134 		if (nfs4_ntov_map[i].vfsstat)
5135 			ntovp->vfsstat = TRUE;
5136 	}
5137 
5138 	if (error != 0)
5139 		status = (error == ENOTSUP ? NFS4ERR_ATTRNOTSUPP :
5140 		    puterrno4(error));
5141 	/* xdrmem_destroy(&xdrs); */	/* NO-OP */
5142 	return (status);
5143 }
5144 
5145 static nfsstat4
5146 do_rfs4_op_setattr(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5147     stateid4 *stateid)
5148 {
5149 	int error = 0;
5150 	struct nfs4_svgetit_arg sarg;
5151 	bool_t trunc;
5152 
5153 	nfsstat4 status = NFS4_OK;
5154 	cred_t *cr = cs->cr;
5155 	vnode_t *vp = cs->vp;
5156 	struct nfs4_ntov_table ntov;
5157 	struct statvfs64 sb;
5158 	struct vattr bva;
5159 	struct flock64 bf;
5160 	int in_crit = 0;
5161 	uint_t saved_mask = 0;
5162 	caller_context_t ct;
5163 
5164 	*resp = 0;
5165 	sarg.sbp = &sb;
5166 	sarg.is_referral = B_FALSE;
5167 	nfs4_ntov_table_init(&ntov);
5168 	status = do_rfs4_set_attrs(resp, fattrp, cs, &sarg, &ntov,
5169 	    NFS4ATTR_SETIT);
5170 	if (status != NFS4_OK) {
5171 		/*
5172 		 * failed set attrs
5173 		 */
5174 		goto done;
5175 	}
5176 	if ((sarg.vap->va_mask == 0) &&
5177 	    (! (fattrp->attrmask & FATTR4_ACL_MASK))) {
5178 		/*
5179 		 * no further work to be done
5180 		 */
5181 		goto done;
5182 	}
5183 
5184 	/*
5185 	 * If we got a request to set the ACL and the MODE, only
5186 	 * allow changing VSUID, VSGID, and VSVTX.  Attempting
5187 	 * to change any other bits, along with setting an ACL,
5188 	 * gives NFS4ERR_INVAL.
5189 	 */
5190 	if ((fattrp->attrmask & FATTR4_ACL_MASK) &&
5191 	    (fattrp->attrmask & FATTR4_MODE_MASK)) {
5192 		vattr_t va;
5193 
5194 		va.va_mask = AT_MODE;
5195 		error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
5196 		if (error) {
5197 			status = puterrno4(error);
5198 			goto done;
5199 		}
5200 		if ((sarg.vap->va_mode ^ va.va_mode) &
5201 		    ~(VSUID | VSGID | VSVTX)) {
5202 			status = NFS4ERR_INVAL;
5203 			goto done;
5204 		}
5205 	}
5206 
5207 	/* Check stateid only if size has been set */
5208 	if (sarg.vap->va_mask & AT_SIZE) {
5209 		trunc = (sarg.vap->va_size == 0);
5210 		status = rfs4_check_stateid(FWRITE, cs->vp, stateid,
5211 		    trunc, &cs->deleg, sarg.vap->va_mask & AT_SIZE, &ct);
5212 		if (status != NFS4_OK)
5213 			goto done;
5214 	} else {
5215 		ct.cc_sysid = 0;
5216 		ct.cc_pid = 0;
5217 		ct.cc_caller_id = nfs4_srv_caller_id;
5218 		ct.cc_flags = CC_DONTBLOCK;
5219 	}
5220 
5221 	/* XXX start of possible race with delegations */
5222 
5223 	/*
5224 	 * We need to specially handle size changes because it is
5225 	 * possible for the client to create a file with read-only
5226 	 * modes, but with the file opened for writing. If the client
5227 	 * then tries to set the file size, e.g. ftruncate(3C),
5228 	 * fcntl(F_FREESP), the normal access checking done in
5229 	 * VOP_SETATTR would prevent the client from doing it even though
5230 	 * it should be allowed to do so.  To get around this, we do the
5231 	 * access checking for ourselves and use VOP_SPACE which doesn't
5232 	 * do the access checking.
5233 	 * Also the client should not be allowed to change the file
5234 	 * size if there is a conflicting non-blocking mandatory lock in
5235 	 * the region of the change.
5236 	 */
5237 	if (vp->v_type == VREG && (sarg.vap->va_mask & AT_SIZE)) {
5238 		u_offset_t offset;
5239 		ssize_t length;
5240 
5241 		/*
5242 		 * ufs_setattr clears AT_SIZE from vap->va_mask, but
5243 		 * before returning, sarg.vap->va_mask is used to
5244 		 * generate the setattr reply bitmap.  We also clear
5245 		 * AT_SIZE below before calling VOP_SPACE.  For both
5246 		 * of these cases, the va_mask needs to be saved here
5247 		 * and restored after calling VOP_SETATTR.
5248 		 */
5249 		saved_mask = sarg.vap->va_mask;
5250 
5251 		/*
5252 		 * Check any possible conflict due to NBMAND locks.
5253 		 * Get into critical region before VOP_GETATTR, so the
5254 		 * size attribute is valid when checking conflicts.
5255 		 */
5256 		if (nbl_need_check(vp)) {
5257 			nbl_start_crit(vp, RW_READER);
5258 			in_crit = 1;
5259 		}
5260 
5261 		bva.va_mask = AT_UID|AT_SIZE;
5262 		if (error = VOP_GETATTR(vp, &bva, 0, cr, &ct)) {
5263 			status = puterrno4(error);
5264 			goto done;
5265 		}
5266 
5267 		if (in_crit) {
5268 			if (sarg.vap->va_size < bva.va_size) {
5269 				offset = sarg.vap->va_size;
5270 				length = bva.va_size - sarg.vap->va_size;
5271 			} else {
5272 				offset = bva.va_size;
5273 				length = sarg.vap->va_size - bva.va_size;
5274 			}
5275 			if (nbl_conflict(vp, NBL_WRITE, offset, length, 0,
5276 			    &ct)) {
5277 				status = NFS4ERR_LOCKED;
5278 				goto done;
5279 			}
5280 		}
5281 
5282 		if (crgetuid(cr) == bva.va_uid) {
5283 			sarg.vap->va_mask &= ~AT_SIZE;
5284 			bf.l_type = F_WRLCK;
5285 			bf.l_whence = 0;
5286 			bf.l_start = (off64_t)sarg.vap->va_size;
5287 			bf.l_len = 0;
5288 			bf.l_sysid = 0;
5289 			bf.l_pid = 0;
5290 			error = VOP_SPACE(vp, F_FREESP, &bf, FWRITE,
5291 			    (offset_t)sarg.vap->va_size, cr, &ct);
5292 		}
5293 	}
5294 
5295 	if (!error && sarg.vap->va_mask != 0)
5296 		error = VOP_SETATTR(vp, sarg.vap, sarg.flag, cr, &ct);
5297 
5298 	/* restore va_mask -- ufs_setattr clears AT_SIZE */
5299 	if (saved_mask & AT_SIZE)
5300 		sarg.vap->va_mask |= AT_SIZE;
5301 
5302 	/*
5303 	 * If an ACL was being set, it has been delayed until now,
5304 	 * in order to set the mode (via the VOP_SETATTR() above) first.
5305 	 */
5306 	if ((! error) && (fattrp->attrmask & FATTR4_ACL_MASK)) {
5307 		int i;
5308 
5309 		for (i = 0; i < NFS4_MAXNUM_ATTRS; i++)
5310 			if (ntov.amap[i] == FATTR4_ACL)
5311 				break;
5312 		if (i < NFS4_MAXNUM_ATTRS) {
5313 			error = (*nfs4_ntov_map[FATTR4_ACL].sv_getit)(
5314 			    NFS4ATTR_SETIT, &sarg, &ntov.na[i]);
5315 			if (error == 0) {
5316 				*resp |= FATTR4_ACL_MASK;
5317 			} else if (error == ENOTSUP) {
5318 				(void) rfs4_verify_attr(&sarg, resp, &ntov);
5319 				status = NFS4ERR_ATTRNOTSUPP;
5320 				goto done;
5321 			}
5322 		} else {
5323 			NFS4_DEBUG(rfs4_debug,
5324 			    (CE_NOTE, "do_rfs4_op_setattr: "
5325 			    "unable to find ACL in fattr4"));
5326 			error = EINVAL;
5327 		}
5328 	}
5329 
5330 	if (error) {
5331 		/* check if a monitor detected a delegation conflict */
5332 		if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
5333 			status = NFS4ERR_DELAY;
5334 		else
5335 			status = puterrno4(error);
5336 
5337 		/*
5338 		 * Set the response bitmap when setattr failed.
5339 		 * If VOP_SETATTR partially succeeded, test by doing a
5340 		 * VOP_GETATTR on the object and comparing the data
5341 		 * to the setattr arguments.
5342 		 */
5343 		(void) rfs4_verify_attr(&sarg, resp, &ntov);
5344 	} else {
5345 		/*
5346 		 * Force modified metadata out to stable storage.
5347 		 */
5348 		(void) VOP_FSYNC(vp, FNODSYNC, cr, &ct);
5349 		/*
5350 		 * Set response bitmap
5351 		 */
5352 		nfs4_vmask_to_nmask_set(sarg.vap->va_mask, resp);
5353 	}
5354 
5355 /* Return early and already have a NFSv4 error */
5356 done:
5357 	/*
5358 	 * Except for nfs4_vmask_to_nmask_set(), vattr --> fattr
5359 	 * conversion sets both readable and writeable NFS4 attrs
5360 	 * for AT_MTIME and AT_ATIME.  The line below masks out
5361 	 * unrequested attrs from the setattr result bitmap.  This
5362 	 * is placed after the done: label to catch the ATTRNOTSUP
5363 	 * case.
5364 	 */
5365 	*resp &= fattrp->attrmask;
5366 
5367 	if (in_crit)
5368 		nbl_end_crit(vp);
5369 
5370 	nfs4_ntov_table_free(&ntov, &sarg);
5371 
5372 	return (status);
5373 }
5374 
5375 /* ARGSUSED */
5376 static void
5377 rfs4_op_setattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5378     struct compound_state *cs)
5379 {
5380 	SETATTR4args *args = &argop->nfs_argop4_u.opsetattr;
5381 	SETATTR4res *resp = &resop->nfs_resop4_u.opsetattr;
5382 	bslabel_t *clabel;
5383 
5384 	DTRACE_NFSV4_2(op__setattr__start, struct compound_state *, cs,
5385 	    SETATTR4args *, args);
5386 
5387 	if (cs->vp == NULL) {
5388 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5389 		goto out;
5390 	}
5391 
5392 	/*
5393 	 * If there is an unshared filesystem mounted on this vnode,
5394 	 * do not allow to setattr on this vnode.
5395 	 */
5396 	if (vn_ismntpt(cs->vp)) {
5397 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5398 		goto out;
5399 	}
5400 
5401 	resp->attrsset = 0;
5402 
5403 	if (rdonly4(cs->exi, cs->vp, req)) {
5404 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5405 		goto out;
5406 	}
5407 
5408 	/* check label before setting attributes */
5409 	if (is_system_labeled()) {
5410 		ASSERT(req->rq_label != NULL);
5411 		clabel = req->rq_label;
5412 		DTRACE_PROBE2(tx__rfs4__log__info__opsetattr__clabel, char *,
5413 		    "got client label from request(1)",
5414 		    struct svc_req *, req);
5415 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
5416 			if (!do_rfs_label_check(clabel, cs->vp,
5417 			    EQUALITY_CHECK, cs->exi)) {
5418 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
5419 				goto out;
5420 			}
5421 		}
5422 	}
5423 
5424 	*cs->statusp = resp->status =
5425 	    do_rfs4_op_setattr(&resp->attrsset, &args->obj_attributes, cs,
5426 	    &args->stateid);
5427 
5428 out:
5429 	DTRACE_NFSV4_2(op__setattr__done, struct compound_state *, cs,
5430 	    SETATTR4res *, resp);
5431 }
5432 
5433 /* ARGSUSED */
5434 static void
5435 rfs4_op_verify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5436     struct compound_state *cs)
5437 {
5438 	/*
5439 	 * verify and nverify are exactly the same, except that nverify
5440 	 * succeeds when some argument changed, and verify succeeds when
5441 	 * when none changed.
5442 	 */
5443 
5444 	VERIFY4args  *args = &argop->nfs_argop4_u.opverify;
5445 	VERIFY4res *resp = &resop->nfs_resop4_u.opverify;
5446 
5447 	int error;
5448 	struct nfs4_svgetit_arg sarg;
5449 	struct statvfs64 sb;
5450 	struct nfs4_ntov_table ntov;
5451 
5452 	DTRACE_NFSV4_2(op__verify__start, struct compound_state *, cs,
5453 	    VERIFY4args *, args);
5454 
5455 	if (cs->vp == NULL) {
5456 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5457 		goto out;
5458 	}
5459 
5460 	sarg.sbp = &sb;
5461 	sarg.is_referral = B_FALSE;
5462 	nfs4_ntov_table_init(&ntov);
5463 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5464 	    &sarg, &ntov, NFS4ATTR_VERIT);
5465 	if (resp->status != NFS4_OK) {
5466 		/*
5467 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5468 		 * so could return -1 for "no match".
5469 		 */
5470 		if (resp->status == -1)
5471 			resp->status = NFS4ERR_NOT_SAME;
5472 		goto done;
5473 	}
5474 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5475 	switch (error) {
5476 	case 0:
5477 		resp->status = NFS4_OK;
5478 		break;
5479 	case -1:
5480 		resp->status = NFS4ERR_NOT_SAME;
5481 		break;
5482 	default:
5483 		resp->status = puterrno4(error);
5484 		break;
5485 	}
5486 done:
5487 	*cs->statusp = resp->status;
5488 	nfs4_ntov_table_free(&ntov, &sarg);
5489 out:
5490 	DTRACE_NFSV4_2(op__verify__done, struct compound_state *, cs,
5491 	    VERIFY4res *, resp);
5492 }
5493 
5494 /* ARGSUSED */
5495 static void
5496 rfs4_op_nverify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5497     struct compound_state *cs)
5498 {
5499 	/*
5500 	 * verify and nverify are exactly the same, except that nverify
5501 	 * succeeds when some argument changed, and verify succeeds when
5502 	 * when none changed.
5503 	 */
5504 
5505 	NVERIFY4args  *args = &argop->nfs_argop4_u.opnverify;
5506 	NVERIFY4res *resp = &resop->nfs_resop4_u.opnverify;
5507 
5508 	int error;
5509 	struct nfs4_svgetit_arg sarg;
5510 	struct statvfs64 sb;
5511 	struct nfs4_ntov_table ntov;
5512 
5513 	DTRACE_NFSV4_2(op__nverify__start, struct compound_state *, cs,
5514 	    NVERIFY4args *, args);
5515 
5516 	if (cs->vp == NULL) {
5517 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5518 		DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5519 		    NVERIFY4res *, resp);
5520 		return;
5521 	}
5522 	sarg.sbp = &sb;
5523 	sarg.is_referral = B_FALSE;
5524 	nfs4_ntov_table_init(&ntov);
5525 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5526 	    &sarg, &ntov, NFS4ATTR_VERIT);
5527 	if (resp->status != NFS4_OK) {
5528 		/*
5529 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5530 		 * so could return -1 for "no match".
5531 		 */
5532 		if (resp->status == -1)
5533 			resp->status = NFS4_OK;
5534 		goto done;
5535 	}
5536 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5537 	switch (error) {
5538 	case 0:
5539 		resp->status = NFS4ERR_SAME;
5540 		break;
5541 	case -1:
5542 		resp->status = NFS4_OK;
5543 		break;
5544 	default:
5545 		resp->status = puterrno4(error);
5546 		break;
5547 	}
5548 done:
5549 	*cs->statusp = resp->status;
5550 	nfs4_ntov_table_free(&ntov, &sarg);
5551 
5552 	DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5553 	    NVERIFY4res *, resp);
5554 }
5555 
5556 /*
5557  * XXX - This should live in an NFS header file.
5558  */
5559 #define	MAX_IOVECS	12
5560 
5561 /* ARGSUSED */
5562 static void
5563 rfs4_op_write(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5564     struct compound_state *cs)
5565 {
5566 	WRITE4args *args = &argop->nfs_argop4_u.opwrite;
5567 	WRITE4res *resp = &resop->nfs_resop4_u.opwrite;
5568 	int error;
5569 	vnode_t *vp;
5570 	struct vattr bva;
5571 	u_offset_t rlimit;
5572 	struct uio uio;
5573 	struct iovec iov[MAX_IOVECS];
5574 	struct iovec *iovp;
5575 	int iovcnt;
5576 	int ioflag;
5577 	cred_t *savecred, *cr;
5578 	bool_t *deleg = &cs->deleg;
5579 	nfsstat4 stat;
5580 	int in_crit = 0;
5581 	caller_context_t ct;
5582 
5583 	DTRACE_NFSV4_2(op__write__start, struct compound_state *, cs,
5584 	    WRITE4args *, args);
5585 
5586 	vp = cs->vp;
5587 	if (vp == NULL) {
5588 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5589 		goto out;
5590 	}
5591 	if (cs->access == CS_ACCESS_DENIED) {
5592 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5593 		goto out;
5594 	}
5595 
5596 	cr = cs->cr;
5597 
5598 	if ((stat = rfs4_check_stateid(FWRITE, vp, &args->stateid, FALSE,
5599 	    deleg, TRUE, &ct)) != NFS4_OK) {
5600 		*cs->statusp = resp->status = stat;
5601 		goto out;
5602 	}
5603 
5604 	/*
5605 	 * We have to enter the critical region before calling VOP_RWLOCK
5606 	 * to avoid a deadlock with ufs.
5607 	 */
5608 	if (nbl_need_check(vp)) {
5609 		nbl_start_crit(vp, RW_READER);
5610 		in_crit = 1;
5611 		if (nbl_conflict(vp, NBL_WRITE,
5612 		    args->offset, args->data_len, 0, &ct)) {
5613 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
5614 			goto out;
5615 		}
5616 	}
5617 
5618 	bva.va_mask = AT_MODE | AT_UID;
5619 	error = VOP_GETATTR(vp, &bva, 0, cr, &ct);
5620 
5621 	/*
5622 	 * If we can't get the attributes, then we can't do the
5623 	 * right access checking.  So, we'll fail the request.
5624 	 */
5625 	if (error) {
5626 		*cs->statusp = resp->status = puterrno4(error);
5627 		goto out;
5628 	}
5629 
5630 	if (rdonly4(cs->exi, cs->vp, req)) {
5631 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5632 		goto out;
5633 	}
5634 
5635 	if (vp->v_type != VREG) {
5636 		*cs->statusp = resp->status =
5637 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
5638 		goto out;
5639 	}
5640 
5641 	if (crgetuid(cr) != bva.va_uid &&
5642 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, &ct))) {
5643 		*cs->statusp = resp->status = puterrno4(error);
5644 		goto out;
5645 	}
5646 
5647 	if (MANDLOCK(vp, bva.va_mode)) {
5648 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5649 		goto out;
5650 	}
5651 
5652 	if (args->data_len == 0) {
5653 		*cs->statusp = resp->status = NFS4_OK;
5654 		resp->count = 0;
5655 		resp->committed = args->stable;
5656 		resp->writeverf = Write4verf;
5657 		goto out;
5658 	}
5659 
5660 	if (args->mblk != NULL) {
5661 		mblk_t *m;
5662 		uint_t bytes, round_len;
5663 
5664 		iovcnt = 0;
5665 		bytes = 0;
5666 		round_len = roundup(args->data_len, BYTES_PER_XDR_UNIT);
5667 		for (m = args->mblk;
5668 		    m != NULL && bytes < round_len;
5669 		    m = m->b_cont) {
5670 			iovcnt++;
5671 			bytes += MBLKL(m);
5672 		}
5673 #ifdef DEBUG
5674 		/* should have ended on an mblk boundary */
5675 		if (bytes != round_len) {
5676 			printf("bytes=0x%x, round_len=0x%x, req len=0x%x\n",
5677 			    bytes, round_len, args->data_len);
5678 			printf("args=%p, args->mblk=%p, m=%p", (void *)args,
5679 			    (void *)args->mblk, (void *)m);
5680 			ASSERT(bytes == round_len);
5681 		}
5682 #endif
5683 		if (iovcnt <= MAX_IOVECS) {
5684 			iovp = iov;
5685 		} else {
5686 			iovp = kmem_alloc(sizeof (*iovp) * iovcnt, KM_SLEEP);
5687 		}
5688 		mblk_to_iov(args->mblk, iovcnt, iovp);
5689 	} else if (args->rlist != NULL) {
5690 		iovcnt = 1;
5691 		iovp = iov;
5692 		iovp->iov_base = (char *)((args->rlist)->u.c_daddr3);
5693 		iovp->iov_len = args->data_len;
5694 	} else {
5695 		iovcnt = 1;
5696 		iovp = iov;
5697 		iovp->iov_base = args->data_val;
5698 		iovp->iov_len = args->data_len;
5699 	}
5700 
5701 	uio.uio_iov = iovp;
5702 	uio.uio_iovcnt = iovcnt;
5703 
5704 	uio.uio_segflg = UIO_SYSSPACE;
5705 	uio.uio_extflg = UIO_COPY_DEFAULT;
5706 	uio.uio_loffset = args->offset;
5707 	uio.uio_resid = args->data_len;
5708 	uio.uio_llimit = curproc->p_fsz_ctl;
5709 	rlimit = uio.uio_llimit - args->offset;
5710 	if (rlimit < (u_offset_t)uio.uio_resid)
5711 		uio.uio_resid = (int)rlimit;
5712 
5713 	if (args->stable == UNSTABLE4)
5714 		ioflag = 0;
5715 	else if (args->stable == FILE_SYNC4)
5716 		ioflag = FSYNC;
5717 	else if (args->stable == DATA_SYNC4)
5718 		ioflag = FDSYNC;
5719 	else {
5720 		if (iovp != iov)
5721 			kmem_free(iovp, sizeof (*iovp) * iovcnt);
5722 		*cs->statusp = resp->status = NFS4ERR_INVAL;
5723 		goto out;
5724 	}
5725 
5726 	/*
5727 	 * We're changing creds because VM may fault and we need
5728 	 * the cred of the current thread to be used if quota
5729 	 * checking is enabled.
5730 	 */
5731 	savecred = curthread->t_cred;
5732 	curthread->t_cred = cr;
5733 	error = do_io(FWRITE, vp, &uio, ioflag, cr, &ct);
5734 	curthread->t_cred = savecred;
5735 
5736 	if (iovp != iov)
5737 		kmem_free(iovp, sizeof (*iovp) * iovcnt);
5738 
5739 	if (error) {
5740 		*cs->statusp = resp->status = puterrno4(error);
5741 		goto out;
5742 	}
5743 
5744 	*cs->statusp = resp->status = NFS4_OK;
5745 	resp->count = args->data_len - uio.uio_resid;
5746 
5747 	if (ioflag == 0)
5748 		resp->committed = UNSTABLE4;
5749 	else
5750 		resp->committed = FILE_SYNC4;
5751 
5752 	resp->writeverf = Write4verf;
5753 
5754 out:
5755 	if (in_crit)
5756 		nbl_end_crit(vp);
5757 
5758 	DTRACE_NFSV4_2(op__write__done, struct compound_state *, cs,
5759 	    WRITE4res *, resp);
5760 }
5761 
5762 
5763 /* XXX put in a header file */
5764 extern int	sec_svc_getcred(struct svc_req *, cred_t *,  caddr_t *, int *);
5765 
5766 void
5767 rfs4_compound(COMPOUND4args *args, COMPOUND4res *resp, struct exportinfo *exi,
5768     struct svc_req *req, cred_t *cr, int *rv)
5769 {
5770 	uint_t i;
5771 	struct compound_state cs;
5772 
5773 	if (rv != NULL)
5774 		*rv = 0;
5775 	rfs4_init_compound_state(&cs);
5776 	/*
5777 	 * Form a reply tag by copying over the reqeuest tag.
5778 	 */
5779 	resp->tag.utf8string_val =
5780 	    kmem_alloc(args->tag.utf8string_len, KM_SLEEP);
5781 	resp->tag.utf8string_len = args->tag.utf8string_len;
5782 	bcopy(args->tag.utf8string_val, resp->tag.utf8string_val,
5783 	    resp->tag.utf8string_len);
5784 
5785 	cs.statusp = &resp->status;
5786 	cs.req = req;
5787 
5788 	/*
5789 	 * XXX for now, minorversion should be zero
5790 	 */
5791 	if (args->minorversion != NFS4_MINORVERSION) {
5792 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5793 		    &cs, COMPOUND4args *, args);
5794 		resp->array_len = 0;
5795 		resp->array = NULL;
5796 		resp->status = NFS4ERR_MINOR_VERS_MISMATCH;
5797 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5798 		    &cs, COMPOUND4res *, resp);
5799 		return;
5800 	}
5801 
5802 	ASSERT(exi == NULL);
5803 	ASSERT(cr == NULL);
5804 
5805 	cr = crget();
5806 	ASSERT(cr != NULL);
5807 
5808 	if (sec_svc_getcred(req, cr, &cs.principal, &cs.nfsflavor) == 0) {
5809 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5810 		    &cs, COMPOUND4args *, args);
5811 		crfree(cr);
5812 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5813 		    &cs, COMPOUND4res *, resp);
5814 		svcerr_badcred(req->rq_xprt);
5815 		if (rv != NULL)
5816 			*rv = 1;
5817 		return;
5818 	}
5819 	resp->array_len = args->array_len;
5820 	resp->array = kmem_zalloc(args->array_len * sizeof (nfs_resop4),
5821 	    KM_SLEEP);
5822 
5823 	cs.basecr = cr;
5824 
5825 	DTRACE_NFSV4_2(compound__start, struct compound_state *, &cs,
5826 	    COMPOUND4args *, args);
5827 
5828 	/*
5829 	 * For now, NFS4 compound processing must be protected by
5830 	 * exported_lock because it can access more than one exportinfo
5831 	 * per compound and share/unshare can now change multiple
5832 	 * exinfo structs.  The NFS2/3 code only refs 1 exportinfo
5833 	 * per proc (excluding public exinfo), and exi_count design
5834 	 * is sufficient to protect concurrent execution of NFS2/3
5835 	 * ops along with unexport.  This lock will be removed as
5836 	 * part of the NFSv4 phase 2 namespace redesign work.
5837 	 */
5838 	rw_enter(&exported_lock, RW_READER);
5839 
5840 	/*
5841 	 * If this is the first compound we've seen, we need to start all
5842 	 * new instances' grace periods.
5843 	 */
5844 	if (rfs4_seen_first_compound == 0) {
5845 		rfs4_grace_start_new();
5846 		/*
5847 		 * This must be set after rfs4_grace_start_new(), otherwise
5848 		 * another thread could proceed past here before the former
5849 		 * is finished.
5850 		 */
5851 		rfs4_seen_first_compound = 1;
5852 	}
5853 
5854 	for (i = 0; i < args->array_len && cs.cont; i++) {
5855 		nfs_argop4 *argop;
5856 		nfs_resop4 *resop;
5857 		uint_t op;
5858 
5859 		argop = &args->array[i];
5860 		resop = &resp->array[i];
5861 		resop->resop = argop->argop;
5862 		op = (uint_t)resop->resop;
5863 
5864 		if (op < rfsv4disp_cnt) {
5865 			/*
5866 			 * Count the individual ops here; NULL and COMPOUND
5867 			 * are counted in common_dispatch()
5868 			 */
5869 			rfsproccnt_v4_ptr[op].value.ui64++;
5870 
5871 			NFS4_DEBUG(rfs4_debug > 1,
5872 			    (CE_NOTE, "Executing %s", rfs4_op_string[op]));
5873 			(*rfsv4disptab[op].dis_proc)(argop, resop, req, &cs);
5874 			NFS4_DEBUG(rfs4_debug > 1, (CE_NOTE, "%s returned %d",
5875 			    rfs4_op_string[op], *cs.statusp));
5876 			if (*cs.statusp != NFS4_OK)
5877 				cs.cont = FALSE;
5878 		} else {
5879 			/*
5880 			 * This is effectively dead code since XDR code
5881 			 * will have already returned BADXDR if op doesn't
5882 			 * decode to legal value.  This only done for a
5883 			 * day when XDR code doesn't verify v4 opcodes.
5884 			 */
5885 			op = OP_ILLEGAL;
5886 			rfsproccnt_v4_ptr[OP_ILLEGAL_IDX].value.ui64++;
5887 
5888 			rfs4_op_illegal(argop, resop, req, &cs);
5889 			cs.cont = FALSE;
5890 		}
5891 
5892 		/*
5893 		 * If not at last op, and if we are to stop, then
5894 		 * compact the results array.
5895 		 */
5896 		if ((i + 1) < args->array_len && !cs.cont) {
5897 			nfs_resop4 *new_res = kmem_alloc(
5898 			    (i+1) * sizeof (nfs_resop4), KM_SLEEP);
5899 			bcopy(resp->array,
5900 			    new_res, (i+1) * sizeof (nfs_resop4));
5901 			kmem_free(resp->array,
5902 			    args->array_len * sizeof (nfs_resop4));
5903 
5904 			resp->array_len =  i + 1;
5905 			resp->array = new_res;
5906 		}
5907 	}
5908 
5909 	rw_exit(&exported_lock);
5910 
5911 	DTRACE_NFSV4_2(compound__done, struct compound_state *, &cs,
5912 	    COMPOUND4res *, resp);
5913 
5914 	if (cs.vp)
5915 		VN_RELE(cs.vp);
5916 	if (cs.saved_vp)
5917 		VN_RELE(cs.saved_vp);
5918 	if (cs.saved_fh.nfs_fh4_val)
5919 		kmem_free(cs.saved_fh.nfs_fh4_val, NFS4_FHSIZE);
5920 
5921 	if (cs.basecr)
5922 		crfree(cs.basecr);
5923 	if (cs.cr)
5924 		crfree(cs.cr);
5925 	/*
5926 	 * done with this compound request, free the label
5927 	 */
5928 
5929 	if (req->rq_label != NULL) {
5930 		kmem_free(req->rq_label, sizeof (bslabel_t));
5931 		req->rq_label = NULL;
5932 	}
5933 }
5934 
5935 /*
5936  * XXX because of what appears to be duplicate calls to rfs4_compound_free
5937  * XXX zero out the tag and array values. Need to investigate why the
5938  * XXX calls occur, but at least prevent the panic for now.
5939  */
5940 void
5941 rfs4_compound_free(COMPOUND4res *resp)
5942 {
5943 	uint_t i;
5944 
5945 	if (resp->tag.utf8string_val) {
5946 		UTF8STRING_FREE(resp->tag)
5947 	}
5948 
5949 	for (i = 0; i < resp->array_len; i++) {
5950 		nfs_resop4 *resop;
5951 		uint_t op;
5952 
5953 		resop = &resp->array[i];
5954 		op = (uint_t)resop->resop;
5955 		if (op < rfsv4disp_cnt) {
5956 			(*rfsv4disptab[op].dis_resfree)(resop);
5957 		}
5958 	}
5959 	if (resp->array != NULL) {
5960 		kmem_free(resp->array, resp->array_len * sizeof (nfs_resop4));
5961 	}
5962 }
5963 
5964 /*
5965  * Process the value of the compound request rpc flags, as a bit-AND
5966  * of the individual per-op flags (idempotent, allowork, publicfh_ok)
5967  */
5968 void
5969 rfs4_compound_flagproc(COMPOUND4args *args, int *flagp)
5970 {
5971 	int i;
5972 	int flag = RPC_ALL;
5973 
5974 	for (i = 0; flag && i < args->array_len; i++) {
5975 		uint_t op;
5976 
5977 		op = (uint_t)args->array[i].argop;
5978 
5979 		if (op < rfsv4disp_cnt)
5980 			flag &= rfsv4disptab[op].dis_flags;
5981 		else
5982 			flag = 0;
5983 	}
5984 	*flagp = flag;
5985 }
5986 
5987 nfsstat4
5988 rfs4_client_sysid(rfs4_client_t *cp, sysid_t *sp)
5989 {
5990 	nfsstat4 e;
5991 
5992 	rfs4_dbe_lock(cp->rc_dbe);
5993 
5994 	if (cp->rc_sysidt != LM_NOSYSID) {
5995 		*sp = cp->rc_sysidt;
5996 		e = NFS4_OK;
5997 
5998 	} else if ((cp->rc_sysidt = lm_alloc_sysidt()) != LM_NOSYSID) {
5999 		*sp = cp->rc_sysidt;
6000 		e = NFS4_OK;
6001 
6002 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
6003 		    "rfs4_client_sysid: allocated 0x%x\n", *sp));
6004 	} else
6005 		e = NFS4ERR_DELAY;
6006 
6007 	rfs4_dbe_unlock(cp->rc_dbe);
6008 	return (e);
6009 }
6010 
6011 #if defined(DEBUG) && ! defined(lint)
6012 static void lock_print(char *str, int operation, struct flock64 *flk)
6013 {
6014 	char *op, *type;
6015 
6016 	switch (operation) {
6017 	case F_GETLK: op = "F_GETLK";
6018 		break;
6019 	case F_SETLK: op = "F_SETLK";
6020 		break;
6021 	case F_SETLK_NBMAND: op = "F_SETLK_NBMAND";
6022 		break;
6023 	default: op = "F_UNKNOWN";
6024 		break;
6025 	}
6026 	switch (flk->l_type) {
6027 	case F_UNLCK: type = "F_UNLCK";
6028 		break;
6029 	case F_RDLCK: type = "F_RDLCK";
6030 		break;
6031 	case F_WRLCK: type = "F_WRLCK";
6032 		break;
6033 	default: type = "F_UNKNOWN";
6034 		break;
6035 	}
6036 
6037 	ASSERT(flk->l_whence == 0);
6038 	cmn_err(CE_NOTE, "%s:  %s, type = %s, off = %llx len = %llx pid = %d",
6039 	    str, op, type, (longlong_t)flk->l_start,
6040 	    flk->l_len ? (longlong_t)flk->l_len : ~0LL, flk->l_pid);
6041 }
6042 
6043 #define	LOCK_PRINT(d, s, t, f) if (d) lock_print(s, t, f)
6044 #else
6045 #define	LOCK_PRINT(d, s, t, f)
6046 #endif
6047 
6048 /*ARGSUSED*/
6049 static bool_t
6050 creds_ok(cred_set_t cr_set, struct svc_req *req, struct compound_state *cs)
6051 {
6052 	return (TRUE);
6053 }
6054 
6055 /*
6056  * Look up the pathname using the vp in cs as the directory vnode.
6057  * cs->vp will be the vnode for the file on success
6058  */
6059 
6060 static nfsstat4
6061 rfs4_lookup(component4 *component, struct svc_req *req,
6062     struct compound_state *cs)
6063 {
6064 	char *nm;
6065 	uint32_t len;
6066 	nfsstat4 status;
6067 	struct sockaddr *ca;
6068 	char *name;
6069 
6070 	if (cs->vp == NULL) {
6071 		return (NFS4ERR_NOFILEHANDLE);
6072 	}
6073 	if (cs->vp->v_type != VDIR) {
6074 		return (NFS4ERR_NOTDIR);
6075 	}
6076 
6077 	if (!utf8_dir_verify(component))
6078 		return (NFS4ERR_INVAL);
6079 
6080 	nm = utf8_to_fn(component, &len, NULL);
6081 	if (nm == NULL) {
6082 		return (NFS4ERR_INVAL);
6083 	}
6084 
6085 	if (len > MAXNAMELEN) {
6086 		kmem_free(nm, len);
6087 		return (NFS4ERR_NAMETOOLONG);
6088 	}
6089 
6090 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6091 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6092 	    MAXPATHLEN + 1);
6093 
6094 	if (name == NULL) {
6095 		kmem_free(nm, len);
6096 		return (NFS4ERR_INVAL);
6097 	}
6098 
6099 	status = do_rfs4_op_lookup(name, req, cs);
6100 
6101 	if (name != nm)
6102 		kmem_free(name, MAXPATHLEN + 1);
6103 
6104 	kmem_free(nm, len);
6105 
6106 	return (status);
6107 }
6108 
6109 static nfsstat4
6110 rfs4_lookupfile(component4 *component, struct svc_req *req,
6111     struct compound_state *cs, uint32_t access, change_info4 *cinfo)
6112 {
6113 	nfsstat4 status;
6114 	vnode_t *dvp = cs->vp;
6115 	vattr_t bva, ava, fva;
6116 	int error;
6117 
6118 	/* Get "before" change value */
6119 	bva.va_mask = AT_CTIME|AT_SEQ;
6120 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6121 	if (error)
6122 		return (puterrno4(error));
6123 
6124 	/* rfs4_lookup may VN_RELE directory */
6125 	VN_HOLD(dvp);
6126 
6127 	status = rfs4_lookup(component, req, cs);
6128 	if (status != NFS4_OK) {
6129 		VN_RELE(dvp);
6130 		return (status);
6131 	}
6132 
6133 	/*
6134 	 * Get "after" change value, if it fails, simply return the
6135 	 * before value.
6136 	 */
6137 	ava.va_mask = AT_CTIME|AT_SEQ;
6138 	if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6139 		ava.va_ctime = bva.va_ctime;
6140 		ava.va_seq = 0;
6141 	}
6142 	VN_RELE(dvp);
6143 
6144 	/*
6145 	 * Validate the file is a file
6146 	 */
6147 	fva.va_mask = AT_TYPE|AT_MODE;
6148 	error = VOP_GETATTR(cs->vp, &fva, 0, cs->cr, NULL);
6149 	if (error)
6150 		return (puterrno4(error));
6151 
6152 	if (fva.va_type != VREG) {
6153 		if (fva.va_type == VDIR)
6154 			return (NFS4ERR_ISDIR);
6155 		if (fva.va_type == VLNK)
6156 			return (NFS4ERR_SYMLINK);
6157 		return (NFS4ERR_INVAL);
6158 	}
6159 
6160 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime);
6161 	NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6162 
6163 	/*
6164 	 * It is undefined if VOP_LOOKUP will change va_seq, so
6165 	 * cinfo.atomic = TRUE only if we have
6166 	 * non-zero va_seq's, and they have not changed.
6167 	 */
6168 	if (bva.va_seq && ava.va_seq && ava.va_seq == bva.va_seq)
6169 		cinfo->atomic = TRUE;
6170 	else
6171 		cinfo->atomic = FALSE;
6172 
6173 	/* Check for mandatory locking */
6174 	cs->mandlock = MANDLOCK(cs->vp, fva.va_mode);
6175 	return (check_open_access(access, cs, req));
6176 }
6177 
6178 static nfsstat4
6179 create_vnode(vnode_t *dvp, char *nm,  vattr_t *vap, createmode4 mode,
6180     timespec32_t *mtime, cred_t *cr, vnode_t **vpp, bool_t *created)
6181 {
6182 	int error;
6183 	nfsstat4 status = NFS4_OK;
6184 	vattr_t va;
6185 
6186 tryagain:
6187 
6188 	/*
6189 	 * The file open mode used is VWRITE.  If the client needs
6190 	 * some other semantic, then it should do the access checking
6191 	 * itself.  It would have been nice to have the file open mode
6192 	 * passed as part of the arguments.
6193 	 */
6194 
6195 	*created = TRUE;
6196 	error = VOP_CREATE(dvp, nm, vap, EXCL, VWRITE, vpp, cr, 0, NULL, NULL);
6197 
6198 	if (error) {
6199 		*created = FALSE;
6200 
6201 		/*
6202 		 * If we got something other than file already exists
6203 		 * then just return this error.  Otherwise, we got
6204 		 * EEXIST.  If we were doing a GUARDED create, then
6205 		 * just return this error.  Otherwise, we need to
6206 		 * make sure that this wasn't a duplicate of an
6207 		 * exclusive create request.
6208 		 *
6209 		 * The assumption is made that a non-exclusive create
6210 		 * request will never return EEXIST.
6211 		 */
6212 
6213 		if (error != EEXIST || mode == GUARDED4) {
6214 			status = puterrno4(error);
6215 			return (status);
6216 		}
6217 		error = VOP_LOOKUP(dvp, nm, vpp, NULL, 0, NULL, cr,
6218 		    NULL, NULL, NULL);
6219 
6220 		if (error) {
6221 			/*
6222 			 * We couldn't find the file that we thought that
6223 			 * we just created.  So, we'll just try creating
6224 			 * it again.
6225 			 */
6226 			if (error == ENOENT)
6227 				goto tryagain;
6228 
6229 			status = puterrno4(error);
6230 			return (status);
6231 		}
6232 
6233 		if (mode == UNCHECKED4) {
6234 			/* existing object must be regular file */
6235 			if ((*vpp)->v_type != VREG) {
6236 				if ((*vpp)->v_type == VDIR)
6237 					status = NFS4ERR_ISDIR;
6238 				else if ((*vpp)->v_type == VLNK)
6239 					status = NFS4ERR_SYMLINK;
6240 				else
6241 					status = NFS4ERR_INVAL;
6242 				VN_RELE(*vpp);
6243 				return (status);
6244 			}
6245 
6246 			return (NFS4_OK);
6247 		}
6248 
6249 		/* Check for duplicate request */
6250 		ASSERT(mtime != 0);
6251 		va.va_mask = AT_MTIME;
6252 		error = VOP_GETATTR(*vpp, &va, 0, cr, NULL);
6253 		if (!error) {
6254 			/* We found the file */
6255 			if (va.va_mtime.tv_sec != mtime->tv_sec ||
6256 			    va.va_mtime.tv_nsec != mtime->tv_nsec) {
6257 				/* but its not our creation */
6258 				VN_RELE(*vpp);
6259 				return (NFS4ERR_EXIST);
6260 			}
6261 			*created = TRUE; /* retrans of create == created */
6262 			return (NFS4_OK);
6263 		}
6264 		VN_RELE(*vpp);
6265 		return (NFS4ERR_EXIST);
6266 	}
6267 
6268 	return (NFS4_OK);
6269 }
6270 
6271 static nfsstat4
6272 check_open_access(uint32_t access, struct compound_state *cs,
6273     struct svc_req *req)
6274 {
6275 	int error;
6276 	vnode_t *vp;
6277 	bool_t readonly;
6278 	cred_t *cr = cs->cr;
6279 
6280 	/* For now we don't allow mandatory locking as per V2/V3 */
6281 	if (cs->access == CS_ACCESS_DENIED || cs->mandlock) {
6282 		return (NFS4ERR_ACCESS);
6283 	}
6284 
6285 	vp = cs->vp;
6286 	ASSERT(cr != NULL && vp->v_type == VREG);
6287 
6288 	/*
6289 	 * If the file system is exported read only and we are trying
6290 	 * to open for write, then return NFS4ERR_ROFS
6291 	 */
6292 
6293 	readonly = rdonly4(cs->exi, cs->vp, req);
6294 
6295 	if ((access & OPEN4_SHARE_ACCESS_WRITE) && readonly)
6296 		return (NFS4ERR_ROFS);
6297 
6298 	if (access & OPEN4_SHARE_ACCESS_READ) {
6299 		if ((VOP_ACCESS(vp, VREAD, 0, cr, NULL) != 0) &&
6300 		    (VOP_ACCESS(vp, VEXEC, 0, cr, NULL) != 0)) {
6301 			return (NFS4ERR_ACCESS);
6302 		}
6303 	}
6304 
6305 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
6306 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
6307 		if (error)
6308 			return (NFS4ERR_ACCESS);
6309 	}
6310 
6311 	return (NFS4_OK);
6312 }
6313 
6314 static nfsstat4
6315 rfs4_createfile(OPEN4args *args, struct svc_req *req, struct compound_state *cs,
6316     change_info4 *cinfo, bitmap4 *attrset, clientid4 clientid)
6317 {
6318 	struct nfs4_svgetit_arg sarg;
6319 	struct nfs4_ntov_table ntov;
6320 
6321 	bool_t ntov_table_init = FALSE;
6322 	struct statvfs64 sb;
6323 	nfsstat4 status;
6324 	vnode_t *vp;
6325 	vattr_t bva, ava, iva, cva, *vap;
6326 	vnode_t *dvp;
6327 	timespec32_t *mtime;
6328 	char *nm = NULL;
6329 	uint_t buflen;
6330 	bool_t created;
6331 	bool_t setsize = FALSE;
6332 	len_t reqsize;
6333 	int error;
6334 	bool_t trunc;
6335 	caller_context_t ct;
6336 	component4 *component;
6337 	bslabel_t *clabel;
6338 	struct sockaddr *ca;
6339 	char *name = NULL;
6340 
6341 	sarg.sbp = &sb;
6342 	sarg.is_referral = B_FALSE;
6343 
6344 	dvp = cs->vp;
6345 
6346 	/* Check if the file system is read only */
6347 	if (rdonly4(cs->exi, dvp, req))
6348 		return (NFS4ERR_ROFS);
6349 
6350 	/* check the label of including directory */
6351 	if (is_system_labeled()) {
6352 		ASSERT(req->rq_label != NULL);
6353 		clabel = req->rq_label;
6354 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
6355 		    "got client label from request(1)",
6356 		    struct svc_req *, req);
6357 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
6358 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
6359 			    cs->exi)) {
6360 				return (NFS4ERR_ACCESS);
6361 			}
6362 		}
6363 	}
6364 
6365 	/*
6366 	 * Get the last component of path name in nm. cs will reference
6367 	 * the including directory on success.
6368 	 */
6369 	component = &args->open_claim4_u.file;
6370 	if (!utf8_dir_verify(component))
6371 		return (NFS4ERR_INVAL);
6372 
6373 	nm = utf8_to_fn(component, &buflen, NULL);
6374 
6375 	if (nm == NULL)
6376 		return (NFS4ERR_RESOURCE);
6377 
6378 	if (buflen > MAXNAMELEN) {
6379 		kmem_free(nm, buflen);
6380 		return (NFS4ERR_NAMETOOLONG);
6381 	}
6382 
6383 	bva.va_mask = AT_TYPE|AT_CTIME|AT_SEQ;
6384 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6385 	if (error) {
6386 		kmem_free(nm, buflen);
6387 		return (puterrno4(error));
6388 	}
6389 
6390 	if (bva.va_type != VDIR) {
6391 		kmem_free(nm, buflen);
6392 		return (NFS4ERR_NOTDIR);
6393 	}
6394 
6395 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime)
6396 
6397 	switch (args->mode) {
6398 	case GUARDED4:
6399 		/*FALLTHROUGH*/
6400 	case UNCHECKED4:
6401 		nfs4_ntov_table_init(&ntov);
6402 		ntov_table_init = TRUE;
6403 
6404 		*attrset = 0;
6405 		status = do_rfs4_set_attrs(attrset,
6406 		    &args->createhow4_u.createattrs,
6407 		    cs, &sarg, &ntov, NFS4ATTR_SETIT);
6408 
6409 		if (status == NFS4_OK && (sarg.vap->va_mask & AT_TYPE) &&
6410 		    sarg.vap->va_type != VREG) {
6411 			if (sarg.vap->va_type == VDIR)
6412 				status = NFS4ERR_ISDIR;
6413 			else if (sarg.vap->va_type == VLNK)
6414 				status = NFS4ERR_SYMLINK;
6415 			else
6416 				status = NFS4ERR_INVAL;
6417 		}
6418 
6419 		if (status != NFS4_OK) {
6420 			kmem_free(nm, buflen);
6421 			nfs4_ntov_table_free(&ntov, &sarg);
6422 			*attrset = 0;
6423 			return (status);
6424 		}
6425 
6426 		vap = sarg.vap;
6427 		vap->va_type = VREG;
6428 		vap->va_mask |= AT_TYPE;
6429 
6430 		if ((vap->va_mask & AT_MODE) == 0) {
6431 			vap->va_mask |= AT_MODE;
6432 			vap->va_mode = (mode_t)0600;
6433 		}
6434 
6435 		if (vap->va_mask & AT_SIZE) {
6436 
6437 			/* Disallow create with a non-zero size */
6438 
6439 			if ((reqsize = sarg.vap->va_size) != 0) {
6440 				kmem_free(nm, buflen);
6441 				nfs4_ntov_table_free(&ntov, &sarg);
6442 				*attrset = 0;
6443 				return (NFS4ERR_INVAL);
6444 			}
6445 			setsize = TRUE;
6446 		}
6447 		break;
6448 
6449 	case EXCLUSIVE4:
6450 		/* prohibit EXCL create of named attributes */
6451 		if (dvp->v_flag & V_XATTRDIR) {
6452 			kmem_free(nm, buflen);
6453 			*attrset = 0;
6454 			return (NFS4ERR_INVAL);
6455 		}
6456 
6457 		cva.va_mask = AT_TYPE | AT_MTIME | AT_MODE;
6458 		cva.va_type = VREG;
6459 		/*
6460 		 * Ensure no time overflows. Assumes underlying
6461 		 * filesystem supports at least 32 bits.
6462 		 * Truncate nsec to usec resolution to allow valid
6463 		 * compares even if the underlying filesystem truncates.
6464 		 */
6465 		mtime = (timespec32_t *)&args->createhow4_u.createverf;
6466 		cva.va_mtime.tv_sec = mtime->tv_sec % TIME32_MAX;
6467 		cva.va_mtime.tv_nsec = (mtime->tv_nsec / 1000) * 1000;
6468 		cva.va_mode = (mode_t)0;
6469 		vap = &cva;
6470 
6471 		/*
6472 		 * For EXCL create, attrset is set to the server attr
6473 		 * used to cache the client's verifier.
6474 		 */
6475 		*attrset = FATTR4_TIME_MODIFY_MASK;
6476 		break;
6477 	}
6478 
6479 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6480 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6481 	    MAXPATHLEN  + 1);
6482 
6483 	if (name == NULL) {
6484 		kmem_free(nm, buflen);
6485 		return (NFS4ERR_SERVERFAULT);
6486 	}
6487 
6488 	status = create_vnode(dvp, name, vap, args->mode, mtime,
6489 	    cs->cr, &vp, &created);
6490 	if (nm != name)
6491 		kmem_free(name, MAXPATHLEN + 1);
6492 	kmem_free(nm, buflen);
6493 
6494 	if (status != NFS4_OK) {
6495 		if (ntov_table_init)
6496 			nfs4_ntov_table_free(&ntov, &sarg);
6497 		*attrset = 0;
6498 		return (status);
6499 	}
6500 
6501 	trunc = (setsize && !created);
6502 
6503 	if (args->mode != EXCLUSIVE4) {
6504 		bitmap4 createmask = args->createhow4_u.createattrs.attrmask;
6505 
6506 		/*
6507 		 * True verification that object was created with correct
6508 		 * attrs is impossible.  The attrs could have been changed
6509 		 * immediately after object creation.  If attributes did
6510 		 * not verify, the only recourse for the server is to
6511 		 * destroy the object.  Maybe if some attrs (like gid)
6512 		 * are set incorrectly, the object should be destroyed;
6513 		 * however, seems bad as a default policy.  Do we really
6514 		 * want to destroy an object over one of the times not
6515 		 * verifying correctly?  For these reasons, the server
6516 		 * currently sets bits in attrset for createattrs
6517 		 * that were set; however, no verification is done.
6518 		 *
6519 		 * vmask_to_nmask accounts for vattr bits set on create
6520 		 *	[do_rfs4_set_attrs() only sets resp bits for
6521 		 *	 non-vattr/vfs bits.]
6522 		 * Mask off any bits we set by default so as not to return
6523 		 * more attrset bits than were requested in createattrs
6524 		 */
6525 		if (created) {
6526 			nfs4_vmask_to_nmask(sarg.vap->va_mask, attrset);
6527 			*attrset &= createmask;
6528 		} else {
6529 			/*
6530 			 * We did not create the vnode (we tried but it
6531 			 * already existed).  In this case, the only createattr
6532 			 * that the spec allows the server to set is size,
6533 			 * and even then, it can only be set if it is 0.
6534 			 */
6535 			*attrset = 0;
6536 			if (trunc)
6537 				*attrset = FATTR4_SIZE_MASK;
6538 		}
6539 	}
6540 	if (ntov_table_init)
6541 		nfs4_ntov_table_free(&ntov, &sarg);
6542 
6543 	/*
6544 	 * Get the initial "after" sequence number, if it fails,
6545 	 * set to zero, time to before.
6546 	 */
6547 	iva.va_mask = AT_CTIME|AT_SEQ;
6548 	if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL)) {
6549 		iva.va_seq = 0;
6550 		iva.va_ctime = bva.va_ctime;
6551 	}
6552 
6553 	/*
6554 	 * create_vnode attempts to create the file exclusive,
6555 	 * if it already exists the VOP_CREATE will fail and
6556 	 * may not increase va_seq. It is atomic if
6557 	 * we haven't changed the directory, but if it has changed
6558 	 * we don't know what changed it.
6559 	 */
6560 	if (!created) {
6561 		if (bva.va_seq && iva.va_seq &&
6562 		    bva.va_seq == iva.va_seq)
6563 			cinfo->atomic = TRUE;
6564 		else
6565 			cinfo->atomic = FALSE;
6566 		NFS4_SET_FATTR4_CHANGE(cinfo->after, iva.va_ctime);
6567 	} else {
6568 		/*
6569 		 * The entry was created, we need to sync the
6570 		 * directory metadata.
6571 		 */
6572 		(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
6573 
6574 		/*
6575 		 * Get "after" change value, if it fails, simply return the
6576 		 * before value.
6577 		 */
6578 		ava.va_mask = AT_CTIME|AT_SEQ;
6579 		if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6580 			ava.va_ctime = bva.va_ctime;
6581 			ava.va_seq = 0;
6582 		}
6583 
6584 		NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6585 
6586 		/*
6587 		 * The cinfo->atomic = TRUE only if we have
6588 		 * non-zero va_seq's, and it has incremented by exactly one
6589 		 * during the create_vnode and it didn't
6590 		 * change during the VOP_FSYNC.
6591 		 */
6592 		if (bva.va_seq && iva.va_seq && ava.va_seq &&
6593 		    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
6594 			cinfo->atomic = TRUE;
6595 		else
6596 			cinfo->atomic = FALSE;
6597 	}
6598 
6599 	/* Check for mandatory locking and that the size gets set. */
6600 	cva.va_mask = AT_MODE;
6601 	if (setsize)
6602 		cva.va_mask |= AT_SIZE;
6603 
6604 	/* Assume the worst */
6605 	cs->mandlock = TRUE;
6606 
6607 	if (VOP_GETATTR(vp, &cva, 0, cs->cr, NULL) == 0) {
6608 		cs->mandlock = MANDLOCK(cs->vp, cva.va_mode);
6609 
6610 		/*
6611 		 * Truncate the file if necessary; this would be
6612 		 * the case for create over an existing file.
6613 		 */
6614 
6615 		if (trunc) {
6616 			int in_crit = 0;
6617 			rfs4_file_t *fp;
6618 			bool_t create = FALSE;
6619 
6620 			/*
6621 			 * We are writing over an existing file.
6622 			 * Check to see if we need to recall a delegation.
6623 			 */
6624 			rfs4_hold_deleg_policy();
6625 			if ((fp = rfs4_findfile(vp, NULL, &create)) != NULL) {
6626 				if (rfs4_check_delegated_byfp(FWRITE, fp,
6627 				    (reqsize == 0), FALSE, FALSE, &clientid)) {
6628 					rfs4_file_rele(fp);
6629 					rfs4_rele_deleg_policy();
6630 					VN_RELE(vp);
6631 					*attrset = 0;
6632 					return (NFS4ERR_DELAY);
6633 				}
6634 				rfs4_file_rele(fp);
6635 			}
6636 			rfs4_rele_deleg_policy();
6637 
6638 			if (nbl_need_check(vp)) {
6639 				in_crit = 1;
6640 
6641 				ASSERT(reqsize == 0);
6642 
6643 				nbl_start_crit(vp, RW_READER);
6644 				if (nbl_conflict(vp, NBL_WRITE, 0,
6645 				    cva.va_size, 0, NULL)) {
6646 					in_crit = 0;
6647 					nbl_end_crit(vp);
6648 					VN_RELE(vp);
6649 					*attrset = 0;
6650 					return (NFS4ERR_ACCESS);
6651 				}
6652 			}
6653 			ct.cc_sysid = 0;
6654 			ct.cc_pid = 0;
6655 			ct.cc_caller_id = nfs4_srv_caller_id;
6656 			ct.cc_flags = CC_DONTBLOCK;
6657 
6658 			cva.va_mask = AT_SIZE;
6659 			cva.va_size = reqsize;
6660 			(void) VOP_SETATTR(vp, &cva, 0, cs->cr, &ct);
6661 			if (in_crit)
6662 				nbl_end_crit(vp);
6663 		}
6664 	}
6665 
6666 	error = makefh4(&cs->fh, vp, cs->exi);
6667 
6668 	/*
6669 	 * Force modified data and metadata out to stable storage.
6670 	 */
6671 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
6672 
6673 	if (error) {
6674 		VN_RELE(vp);
6675 		*attrset = 0;
6676 		return (puterrno4(error));
6677 	}
6678 
6679 	/* if parent dir is attrdir, set namedattr fh flag */
6680 	if (dvp->v_flag & V_XATTRDIR)
6681 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
6682 
6683 	if (cs->vp)
6684 		VN_RELE(cs->vp);
6685 
6686 	cs->vp = vp;
6687 
6688 	/*
6689 	 * if we did not create the file, we will need to check
6690 	 * the access bits on the file
6691 	 */
6692 
6693 	if (!created) {
6694 		if (setsize)
6695 			args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
6696 		status = check_open_access(args->share_access, cs, req);
6697 		if (status != NFS4_OK)
6698 			*attrset = 0;
6699 	}
6700 	return (status);
6701 }
6702 
6703 /*ARGSUSED*/
6704 static void
6705 rfs4_do_open(struct compound_state *cs, struct svc_req *req,
6706     rfs4_openowner_t *oo, delegreq_t deleg,
6707     uint32_t access, uint32_t deny,
6708     OPEN4res *resp, int deleg_cur)
6709 {
6710 	/* XXX Currently not using req  */
6711 	rfs4_state_t *sp;
6712 	rfs4_file_t *fp;
6713 	bool_t screate = TRUE;
6714 	bool_t fcreate = TRUE;
6715 	uint32_t open_a, share_a;
6716 	uint32_t open_d, share_d;
6717 	rfs4_deleg_state_t *dsp;
6718 	sysid_t sysid;
6719 	nfsstat4 status;
6720 	caller_context_t ct;
6721 	int fflags = 0;
6722 	int recall = 0;
6723 	int err;
6724 	int first_open;
6725 
6726 	/* get the file struct and hold a lock on it during initial open */
6727 	fp = rfs4_findfile_withlock(cs->vp, &cs->fh, &fcreate);
6728 	if (fp == NULL) {
6729 		resp->status = NFS4ERR_RESOURCE;
6730 		DTRACE_PROBE1(nfss__e__do__open1, nfsstat4, resp->status);
6731 		return;
6732 	}
6733 
6734 	sp = rfs4_findstate_by_owner_file(oo, fp, &screate);
6735 	if (sp == NULL) {
6736 		resp->status = NFS4ERR_RESOURCE;
6737 		DTRACE_PROBE1(nfss__e__do__open2, nfsstat4, resp->status);
6738 		/* No need to keep any reference */
6739 		rw_exit(&fp->rf_file_rwlock);
6740 		rfs4_file_rele(fp);
6741 		return;
6742 	}
6743 
6744 	/* try to get the sysid before continuing */
6745 	if ((status = rfs4_client_sysid(oo->ro_client, &sysid)) != NFS4_OK) {
6746 		resp->status = status;
6747 		rfs4_file_rele(fp);
6748 		/* Not a fully formed open; "close" it */
6749 		if (screate == TRUE)
6750 			rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6751 		rfs4_state_rele(sp);
6752 		return;
6753 	}
6754 
6755 	/* Calculate the fflags for this OPEN. */
6756 	if (access & OPEN4_SHARE_ACCESS_READ)
6757 		fflags |= FREAD;
6758 	if (access & OPEN4_SHARE_ACCESS_WRITE)
6759 		fflags |= FWRITE;
6760 
6761 	rfs4_dbe_lock(sp->rs_dbe);
6762 
6763 	/*
6764 	 * Calculate the new deny and access mode that this open is adding to
6765 	 * the file for this open owner;
6766 	 */
6767 	open_d = (deny & ~sp->rs_open_deny);
6768 	open_a = (access & ~sp->rs_open_access);
6769 
6770 	/*
6771 	 * Calculate the new share access and share deny modes that this open
6772 	 * is adding to the file for this open owner;
6773 	 */
6774 	share_a = (access & ~sp->rs_share_access);
6775 	share_d = (deny & ~sp->rs_share_deny);
6776 
6777 	first_open = (sp->rs_open_access & OPEN4_SHARE_ACCESS_BOTH) == 0;
6778 
6779 	/*
6780 	 * Check to see the client has already sent an open for this
6781 	 * open owner on this file with the same share/deny modes.
6782 	 * If so, we don't need to check for a conflict and we don't
6783 	 * need to add another shrlock.  If not, then we need to
6784 	 * check for conflicts in deny and access before checking for
6785 	 * conflicts in delegation.  We don't want to recall a
6786 	 * delegation based on an open that will eventually fail based
6787 	 * on shares modes.
6788 	 */
6789 
6790 	if (share_a || share_d) {
6791 		if ((err = rfs4_share(sp, access, deny)) != 0) {
6792 			rfs4_dbe_unlock(sp->rs_dbe);
6793 			resp->status = err;
6794 
6795 			rfs4_file_rele(fp);
6796 			/* Not a fully formed open; "close" it */
6797 			if (screate == TRUE)
6798 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6799 			rfs4_state_rele(sp);
6800 			return;
6801 		}
6802 	}
6803 
6804 	rfs4_dbe_lock(fp->rf_dbe);
6805 
6806 	/*
6807 	 * Check to see if this file is delegated and if so, if a
6808 	 * recall needs to be done.
6809 	 */
6810 	if (rfs4_check_recall(sp, access)) {
6811 		rfs4_dbe_unlock(fp->rf_dbe);
6812 		rfs4_dbe_unlock(sp->rs_dbe);
6813 		rfs4_recall_deleg(fp, FALSE, sp->rs_owner->ro_client);
6814 		delay(NFS4_DELEGATION_CONFLICT_DELAY);
6815 		rfs4_dbe_lock(sp->rs_dbe);
6816 
6817 		/* if state closed while lock was dropped */
6818 		if (sp->rs_closed) {
6819 			if (share_a || share_d)
6820 				(void) rfs4_unshare(sp);
6821 			rfs4_dbe_unlock(sp->rs_dbe);
6822 			rfs4_file_rele(fp);
6823 			/* Not a fully formed open; "close" it */
6824 			if (screate == TRUE)
6825 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6826 			rfs4_state_rele(sp);
6827 			resp->status = NFS4ERR_OLD_STATEID;
6828 			return;
6829 		}
6830 
6831 		rfs4_dbe_lock(fp->rf_dbe);
6832 		/* Let's see if the delegation was returned */
6833 		if (rfs4_check_recall(sp, access)) {
6834 			rfs4_dbe_unlock(fp->rf_dbe);
6835 			if (share_a || share_d)
6836 				(void) rfs4_unshare(sp);
6837 			rfs4_dbe_unlock(sp->rs_dbe);
6838 			rfs4_file_rele(fp);
6839 			rfs4_update_lease(sp->rs_owner->ro_client);
6840 
6841 			/* Not a fully formed open; "close" it */
6842 			if (screate == TRUE)
6843 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6844 			rfs4_state_rele(sp);
6845 			resp->status = NFS4ERR_DELAY;
6846 			return;
6847 		}
6848 	}
6849 	/*
6850 	 * the share check passed and any delegation conflict has been
6851 	 * taken care of, now call vop_open.
6852 	 * if this is the first open then call vop_open with fflags.
6853 	 * if not, call vn_open_upgrade with just the upgrade flags.
6854 	 *
6855 	 * if the file has been opened already, it will have the current
6856 	 * access mode in the state struct.  if it has no share access, then
6857 	 * this is a new open.
6858 	 *
6859 	 * However, if this is open with CLAIM_DLEGATE_CUR, then don't
6860 	 * call VOP_OPEN(), just do the open upgrade.
6861 	 */
6862 	if (first_open && !deleg_cur) {
6863 		ct.cc_sysid = sysid;
6864 		ct.cc_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
6865 		ct.cc_caller_id = nfs4_srv_caller_id;
6866 		ct.cc_flags = CC_DONTBLOCK;
6867 		err = VOP_OPEN(&cs->vp, fflags, cs->cr, &ct);
6868 		if (err) {
6869 			rfs4_dbe_unlock(fp->rf_dbe);
6870 			if (share_a || share_d)
6871 				(void) rfs4_unshare(sp);
6872 			rfs4_dbe_unlock(sp->rs_dbe);
6873 			rfs4_file_rele(fp);
6874 
6875 			/* Not a fully formed open; "close" it */
6876 			if (screate == TRUE)
6877 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6878 			rfs4_state_rele(sp);
6879 			/* check if a monitor detected a delegation conflict */
6880 			if (err == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
6881 				resp->status = NFS4ERR_DELAY;
6882 			else
6883 				resp->status = NFS4ERR_SERVERFAULT;
6884 			return;
6885 		}
6886 	} else { /* open upgrade */
6887 		/*
6888 		 * calculate the fflags for the new mode that is being added
6889 		 * by this upgrade.
6890 		 */
6891 		fflags = 0;
6892 		if (open_a & OPEN4_SHARE_ACCESS_READ)
6893 			fflags |= FREAD;
6894 		if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6895 			fflags |= FWRITE;
6896 		vn_open_upgrade(cs->vp, fflags);
6897 	}
6898 	sp->rs_open_access |= access;
6899 	sp->rs_open_deny |= deny;
6900 
6901 	if (open_d & OPEN4_SHARE_DENY_READ)
6902 		fp->rf_deny_read++;
6903 	if (open_d & OPEN4_SHARE_DENY_WRITE)
6904 		fp->rf_deny_write++;
6905 	fp->rf_share_deny |= deny;
6906 
6907 	if (open_a & OPEN4_SHARE_ACCESS_READ)
6908 		fp->rf_access_read++;
6909 	if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6910 		fp->rf_access_write++;
6911 	fp->rf_share_access |= access;
6912 
6913 	/*
6914 	 * Check for delegation here. if the deleg argument is not
6915 	 * DELEG_ANY, then this is a reclaim from a client and
6916 	 * we must honor the delegation requested. If necessary we can
6917 	 * set the recall flag.
6918 	 */
6919 
6920 	dsp = rfs4_grant_delegation(deleg, sp, &recall);
6921 
6922 	cs->deleg = (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE);
6923 
6924 	next_stateid(&sp->rs_stateid);
6925 
6926 	resp->stateid = sp->rs_stateid.stateid;
6927 
6928 	rfs4_dbe_unlock(fp->rf_dbe);
6929 	rfs4_dbe_unlock(sp->rs_dbe);
6930 
6931 	if (dsp) {
6932 		rfs4_set_deleg_response(dsp, &resp->delegation, NULL, recall);
6933 		rfs4_deleg_state_rele(dsp);
6934 	}
6935 
6936 	rfs4_file_rele(fp);
6937 	rfs4_state_rele(sp);
6938 
6939 	resp->status = NFS4_OK;
6940 }
6941 
6942 /*ARGSUSED*/
6943 static void
6944 rfs4_do_opennull(struct compound_state *cs, struct svc_req *req,
6945     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6946 {
6947 	change_info4 *cinfo = &resp->cinfo;
6948 	bitmap4 *attrset = &resp->attrset;
6949 
6950 	if (args->opentype == OPEN4_NOCREATE)
6951 		resp->status = rfs4_lookupfile(&args->open_claim4_u.file,
6952 		    req, cs, args->share_access, cinfo);
6953 	else {
6954 		/* inhibit delegation grants during exclusive create */
6955 
6956 		if (args->mode == EXCLUSIVE4)
6957 			rfs4_disable_delegation();
6958 
6959 		resp->status = rfs4_createfile(args, req, cs, cinfo, attrset,
6960 		    oo->ro_client->rc_clientid);
6961 	}
6962 
6963 	if (resp->status == NFS4_OK) {
6964 
6965 		/* cs->vp cs->fh now reference the desired file */
6966 
6967 		rfs4_do_open(cs, req, oo,
6968 		    oo->ro_need_confirm ? DELEG_NONE : DELEG_ANY,
6969 		    args->share_access, args->share_deny, resp, 0);
6970 
6971 		/*
6972 		 * If rfs4_createfile set attrset, we must
6973 		 * clear this attrset before the response is copied.
6974 		 */
6975 		if (resp->status != NFS4_OK && resp->attrset) {
6976 			resp->attrset = 0;
6977 		}
6978 	}
6979 	else
6980 		*cs->statusp = resp->status;
6981 
6982 	if (args->mode == EXCLUSIVE4)
6983 		rfs4_enable_delegation();
6984 }
6985 
6986 /*ARGSUSED*/
6987 static void
6988 rfs4_do_openprev(struct compound_state *cs, struct svc_req *req,
6989     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6990 {
6991 	change_info4 *cinfo = &resp->cinfo;
6992 	vattr_t va;
6993 	vtype_t v_type = cs->vp->v_type;
6994 	int error = 0;
6995 
6996 	/* Verify that we have a regular file */
6997 	if (v_type != VREG) {
6998 		if (v_type == VDIR)
6999 			resp->status = NFS4ERR_ISDIR;
7000 		else if (v_type == VLNK)
7001 			resp->status = NFS4ERR_SYMLINK;
7002 		else
7003 			resp->status = NFS4ERR_INVAL;
7004 		return;
7005 	}
7006 
7007 	va.va_mask = AT_MODE|AT_UID;
7008 	error = VOP_GETATTR(cs->vp, &va, 0, cs->cr, NULL);
7009 	if (error) {
7010 		resp->status = puterrno4(error);
7011 		return;
7012 	}
7013 
7014 	cs->mandlock = MANDLOCK(cs->vp, va.va_mode);
7015 
7016 	/*
7017 	 * Check if we have access to the file, Note the the file
7018 	 * could have originally been open UNCHECKED or GUARDED
7019 	 * with mode bits that will now fail, but there is nothing
7020 	 * we can really do about that except in the case that the
7021 	 * owner of the file is the one requesting the open.
7022 	 */
7023 	if (crgetuid(cs->cr) != va.va_uid) {
7024 		resp->status = check_open_access(args->share_access, cs, req);
7025 		if (resp->status != NFS4_OK) {
7026 			return;
7027 		}
7028 	}
7029 
7030 	/*
7031 	 * cinfo on a CLAIM_PREVIOUS is undefined, initialize to zero
7032 	 */
7033 	cinfo->before = 0;
7034 	cinfo->after = 0;
7035 	cinfo->atomic = FALSE;
7036 
7037 	rfs4_do_open(cs, req, oo,
7038 	    NFS4_DELEG4TYPE2REQTYPE(args->open_claim4_u.delegate_type),
7039 	    args->share_access, args->share_deny, resp, 0);
7040 }
7041 
7042 static void
7043 rfs4_do_opendelcur(struct compound_state *cs, struct svc_req *req,
7044     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7045 {
7046 	int error;
7047 	nfsstat4 status;
7048 	stateid4 stateid =
7049 	    args->open_claim4_u.delegate_cur_info.delegate_stateid;
7050 	rfs4_deleg_state_t *dsp;
7051 
7052 	/*
7053 	 * Find the state info from the stateid and confirm that the
7054 	 * file is delegated.  If the state openowner is the same as
7055 	 * the supplied openowner we're done. If not, get the file
7056 	 * info from the found state info. Use that file info to
7057 	 * create the state for this lock owner. Note solaris doen't
7058 	 * really need the pathname to find the file. We may want to
7059 	 * lookup the pathname and make sure that the vp exist and
7060 	 * matches the vp in the file structure. However it is
7061 	 * possible that the pathname nolonger exists (local process
7062 	 * unlinks the file), so this may not be that useful.
7063 	 */
7064 
7065 	status = rfs4_get_deleg_state(&stateid, &dsp);
7066 	if (status != NFS4_OK) {
7067 		resp->status = status;
7068 		return;
7069 	}
7070 
7071 	ASSERT(dsp->rds_finfo->rf_dinfo.rd_dtype != OPEN_DELEGATE_NONE);
7072 
7073 	/*
7074 	 * New lock owner, create state. Since this was probably called
7075 	 * in response to a CB_RECALL we set deleg to DELEG_NONE
7076 	 */
7077 
7078 	ASSERT(cs->vp != NULL);
7079 	VN_RELE(cs->vp);
7080 	VN_HOLD(dsp->rds_finfo->rf_vp);
7081 	cs->vp = dsp->rds_finfo->rf_vp;
7082 
7083 	if (error = makefh4(&cs->fh, cs->vp, cs->exi)) {
7084 		rfs4_deleg_state_rele(dsp);
7085 		*cs->statusp = resp->status = puterrno4(error);
7086 		return;
7087 	}
7088 
7089 	/* Mark progress for delegation returns */
7090 	dsp->rds_finfo->rf_dinfo.rd_time_lastwrite = gethrestime_sec();
7091 	rfs4_deleg_state_rele(dsp);
7092 	rfs4_do_open(cs, req, oo, DELEG_NONE,
7093 	    args->share_access, args->share_deny, resp, 1);
7094 }
7095 
7096 /*ARGSUSED*/
7097 static void
7098 rfs4_do_opendelprev(struct compound_state *cs, struct svc_req *req,
7099     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7100 {
7101 	/*
7102 	 * Lookup the pathname, it must already exist since this file
7103 	 * was delegated.
7104 	 *
7105 	 * Find the file and state info for this vp and open owner pair.
7106 	 *	check that they are in fact delegated.
7107 	 *	check that the state access and deny modes are the same.
7108 	 *
7109 	 * Return the delgation possibly seting the recall flag.
7110 	 */
7111 	rfs4_file_t *fp;
7112 	rfs4_state_t *sp;
7113 	bool_t create = FALSE;
7114 	bool_t dcreate = FALSE;
7115 	rfs4_deleg_state_t *dsp;
7116 	nfsace4 *ace;
7117 
7118 	/* Note we ignore oflags */
7119 	resp->status = rfs4_lookupfile(&args->open_claim4_u.file_delegate_prev,
7120 	    req, cs, args->share_access, &resp->cinfo);
7121 
7122 	if (resp->status != NFS4_OK) {
7123 		return;
7124 	}
7125 
7126 	/* get the file struct and hold a lock on it during initial open */
7127 	fp = rfs4_findfile_withlock(cs->vp, NULL, &create);
7128 	if (fp == NULL) {
7129 		resp->status = NFS4ERR_RESOURCE;
7130 		DTRACE_PROBE1(nfss__e__do_opendelprev1, nfsstat4, resp->status);
7131 		return;
7132 	}
7133 
7134 	sp = rfs4_findstate_by_owner_file(oo, fp, &create);
7135 	if (sp == NULL) {
7136 		resp->status = NFS4ERR_SERVERFAULT;
7137 		DTRACE_PROBE1(nfss__e__do_opendelprev2, nfsstat4, resp->status);
7138 		rw_exit(&fp->rf_file_rwlock);
7139 		rfs4_file_rele(fp);
7140 		return;
7141 	}
7142 
7143 	rfs4_dbe_lock(sp->rs_dbe);
7144 	rfs4_dbe_lock(fp->rf_dbe);
7145 	if (args->share_access != sp->rs_share_access ||
7146 	    args->share_deny != sp->rs_share_deny ||
7147 	    sp->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
7148 		NFS4_DEBUG(rfs4_debug,
7149 		    (CE_NOTE, "rfs4_do_opendelprev: state mixup"));
7150 		rfs4_dbe_unlock(fp->rf_dbe);
7151 		rfs4_dbe_unlock(sp->rs_dbe);
7152 		rfs4_file_rele(fp);
7153 		rfs4_state_rele(sp);
7154 		resp->status = NFS4ERR_SERVERFAULT;
7155 		return;
7156 	}
7157 	rfs4_dbe_unlock(fp->rf_dbe);
7158 	rfs4_dbe_unlock(sp->rs_dbe);
7159 
7160 	dsp = rfs4_finddeleg(sp, &dcreate);
7161 	if (dsp == NULL) {
7162 		rfs4_state_rele(sp);
7163 		rfs4_file_rele(fp);
7164 		resp->status = NFS4ERR_SERVERFAULT;
7165 		return;
7166 	}
7167 
7168 	next_stateid(&sp->rs_stateid);
7169 
7170 	resp->stateid = sp->rs_stateid.stateid;
7171 
7172 	resp->delegation.delegation_type = dsp->rds_dtype;
7173 
7174 	if (dsp->rds_dtype == OPEN_DELEGATE_READ) {
7175 		open_read_delegation4 *rv =
7176 		    &resp->delegation.open_delegation4_u.read;
7177 
7178 		rv->stateid = dsp->rds_delegid.stateid;
7179 		rv->recall = FALSE; /* no policy in place to set to TRUE */
7180 		ace = &rv->permissions;
7181 	} else {
7182 		open_write_delegation4 *rv =
7183 		    &resp->delegation.open_delegation4_u.write;
7184 
7185 		rv->stateid = dsp->rds_delegid.stateid;
7186 		rv->recall = FALSE;  /* no policy in place to set to TRUE */
7187 		ace = &rv->permissions;
7188 		rv->space_limit.limitby = NFS_LIMIT_SIZE;
7189 		rv->space_limit.nfs_space_limit4_u.filesize = UINT64_MAX;
7190 	}
7191 
7192 	/* XXX For now */
7193 	ace->type = ACE4_ACCESS_ALLOWED_ACE_TYPE;
7194 	ace->flag = 0;
7195 	ace->access_mask = 0;
7196 	ace->who.utf8string_len = 0;
7197 	ace->who.utf8string_val = 0;
7198 
7199 	rfs4_deleg_state_rele(dsp);
7200 	rfs4_state_rele(sp);
7201 	rfs4_file_rele(fp);
7202 }
7203 
7204 typedef enum {
7205 	NFS4_CHKSEQ_OKAY = 0,
7206 	NFS4_CHKSEQ_REPLAY = 1,
7207 	NFS4_CHKSEQ_BAD = 2
7208 } rfs4_chkseq_t;
7209 
7210 /*
7211  * Generic function for sequence number checks.
7212  */
7213 static rfs4_chkseq_t
7214 rfs4_check_seqid(seqid4 seqid, nfs_resop4 *lastop,
7215     seqid4 rqst_seq, nfs_resop4 *resop, bool_t copyres)
7216 {
7217 	/* Same sequence ids and matching operations? */
7218 	if (seqid == rqst_seq && resop->resop == lastop->resop) {
7219 		if (copyres == TRUE) {
7220 			rfs4_free_reply(resop);
7221 			rfs4_copy_reply(resop, lastop);
7222 		}
7223 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
7224 		    "Replayed SEQID %d\n", seqid));
7225 		return (NFS4_CHKSEQ_REPLAY);
7226 	}
7227 
7228 	/* If the incoming sequence is not the next expected then it is bad */
7229 	if (rqst_seq != seqid + 1) {
7230 		if (rqst_seq == seqid) {
7231 			NFS4_DEBUG(rfs4_debug,
7232 			    (CE_NOTE, "BAD SEQID: Replayed sequence id "
7233 			    "but last op was %d current op is %d\n",
7234 			    lastop->resop, resop->resop));
7235 			return (NFS4_CHKSEQ_BAD);
7236 		}
7237 		NFS4_DEBUG(rfs4_debug,
7238 		    (CE_NOTE, "BAD SEQID: got %u expecting %u\n",
7239 		    rqst_seq, seqid));
7240 		return (NFS4_CHKSEQ_BAD);
7241 	}
7242 
7243 	/* Everything okay -- next expected */
7244 	return (NFS4_CHKSEQ_OKAY);
7245 }
7246 
7247 
7248 static rfs4_chkseq_t
7249 rfs4_check_open_seqid(seqid4 seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7250 {
7251 	rfs4_chkseq_t rc;
7252 
7253 	rfs4_dbe_lock(op->ro_dbe);
7254 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply, seqid, resop,
7255 	    TRUE);
7256 	rfs4_dbe_unlock(op->ro_dbe);
7257 
7258 	if (rc == NFS4_CHKSEQ_OKAY)
7259 		rfs4_update_lease(op->ro_client);
7260 
7261 	return (rc);
7262 }
7263 
7264 static rfs4_chkseq_t
7265 rfs4_check_olo_seqid(seqid4 olo_seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7266 {
7267 	rfs4_chkseq_t rc;
7268 
7269 	rfs4_dbe_lock(op->ro_dbe);
7270 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply,
7271 	    olo_seqid, resop, FALSE);
7272 	rfs4_dbe_unlock(op->ro_dbe);
7273 
7274 	return (rc);
7275 }
7276 
7277 static rfs4_chkseq_t
7278 rfs4_check_lock_seqid(seqid4 seqid, rfs4_lo_state_t *lsp, nfs_resop4 *resop)
7279 {
7280 	rfs4_chkseq_t rc = NFS4_CHKSEQ_OKAY;
7281 
7282 	rfs4_dbe_lock(lsp->rls_dbe);
7283 	if (!lsp->rls_skip_seqid_check)
7284 		rc = rfs4_check_seqid(lsp->rls_seqid, &lsp->rls_reply, seqid,
7285 		    resop, TRUE);
7286 	rfs4_dbe_unlock(lsp->rls_dbe);
7287 
7288 	return (rc);
7289 }
7290 
7291 static void
7292 rfs4_op_open(nfs_argop4 *argop, nfs_resop4 *resop,
7293     struct svc_req *req, struct compound_state *cs)
7294 {
7295 	OPEN4args *args = &argop->nfs_argop4_u.opopen;
7296 	OPEN4res *resp = &resop->nfs_resop4_u.opopen;
7297 	open_owner4 *owner = &args->owner;
7298 	open_claim_type4 claim = args->claim;
7299 	rfs4_client_t *cp;
7300 	rfs4_openowner_t *oo;
7301 	bool_t create;
7302 	bool_t replay = FALSE;
7303 	int can_reclaim;
7304 
7305 	DTRACE_NFSV4_2(op__open__start, struct compound_state *, cs,
7306 	    OPEN4args *, args);
7307 
7308 	if (cs->vp == NULL) {
7309 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7310 		goto end;
7311 	}
7312 
7313 	/*
7314 	 * Need to check clientid and lease expiration first based on
7315 	 * error ordering and incrementing sequence id.
7316 	 */
7317 	cp = rfs4_findclient_by_id(owner->clientid, FALSE);
7318 	if (cp == NULL) {
7319 		*cs->statusp = resp->status =
7320 		    rfs4_check_clientid(&owner->clientid, 0);
7321 		goto end;
7322 	}
7323 
7324 	if (rfs4_lease_expired(cp)) {
7325 		rfs4_client_close(cp);
7326 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7327 		goto end;
7328 	}
7329 	can_reclaim = cp->rc_can_reclaim;
7330 
7331 	/*
7332 	 * Find the open_owner for use from this point forward.  Take
7333 	 * care in updating the sequence id based on the type of error
7334 	 * being returned.
7335 	 */
7336 retry:
7337 	create = TRUE;
7338 	oo = rfs4_findopenowner(owner, &create, args->seqid);
7339 	if (oo == NULL) {
7340 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
7341 		rfs4_client_rele(cp);
7342 		goto end;
7343 	}
7344 
7345 	/* Hold off access to the sequence space while the open is done */
7346 	rfs4_sw_enter(&oo->ro_sw);
7347 
7348 	/*
7349 	 * If the open_owner existed before at the server, then check
7350 	 * the sequence id.
7351 	 */
7352 	if (!create && !oo->ro_postpone_confirm) {
7353 		switch (rfs4_check_open_seqid(args->seqid, oo, resop)) {
7354 		case NFS4_CHKSEQ_BAD:
7355 			if ((args->seqid > oo->ro_open_seqid) &&
7356 			    oo->ro_need_confirm) {
7357 				rfs4_free_opens(oo, TRUE, FALSE);
7358 				rfs4_sw_exit(&oo->ro_sw);
7359 				rfs4_openowner_rele(oo);
7360 				goto retry;
7361 			}
7362 			resp->status = NFS4ERR_BAD_SEQID;
7363 			goto out;
7364 		case NFS4_CHKSEQ_REPLAY: /* replay of previous request */
7365 			replay = TRUE;
7366 			goto out;
7367 		default:
7368 			break;
7369 		}
7370 
7371 		/*
7372 		 * Sequence was ok and open owner exists
7373 		 * check to see if we have yet to see an
7374 		 * open_confirm.
7375 		 */
7376 		if (oo->ro_need_confirm) {
7377 			rfs4_free_opens(oo, TRUE, FALSE);
7378 			rfs4_sw_exit(&oo->ro_sw);
7379 			rfs4_openowner_rele(oo);
7380 			goto retry;
7381 		}
7382 	}
7383 	/* Grace only applies to regular-type OPENs */
7384 	if (rfs4_clnt_in_grace(cp) &&
7385 	    (claim == CLAIM_NULL || claim == CLAIM_DELEGATE_CUR)) {
7386 		*cs->statusp = resp->status = NFS4ERR_GRACE;
7387 		goto out;
7388 	}
7389 
7390 	/*
7391 	 * If previous state at the server existed then can_reclaim
7392 	 * will be set. If not reply NFS4ERR_NO_GRACE to the
7393 	 * client.
7394 	 */
7395 	if (rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS && !can_reclaim) {
7396 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7397 		goto out;
7398 	}
7399 
7400 
7401 	/*
7402 	 * Reject the open if the client has missed the grace period
7403 	 */
7404 	if (!rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS) {
7405 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7406 		goto out;
7407 	}
7408 
7409 	/* Couple of up-front bookkeeping items */
7410 	if (oo->ro_need_confirm) {
7411 		/*
7412 		 * If this is a reclaim OPEN then we should not ask
7413 		 * for a confirmation of the open_owner per the
7414 		 * protocol specification.
7415 		 */
7416 		if (claim == CLAIM_PREVIOUS)
7417 			oo->ro_need_confirm = FALSE;
7418 		else
7419 			resp->rflags |= OPEN4_RESULT_CONFIRM;
7420 	}
7421 	resp->rflags |= OPEN4_RESULT_LOCKTYPE_POSIX;
7422 
7423 	/*
7424 	 * If there is an unshared filesystem mounted on this vnode,
7425 	 * do not allow to open/create in this directory.
7426 	 */
7427 	if (vn_ismntpt(cs->vp)) {
7428 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
7429 		goto out;
7430 	}
7431 
7432 	/*
7433 	 * access must READ, WRITE, or BOTH.  No access is invalid.
7434 	 * deny can be READ, WRITE, BOTH, or NONE.
7435 	 * bits not defined for access/deny are invalid.
7436 	 */
7437 	if (! (args->share_access & OPEN4_SHARE_ACCESS_BOTH) ||
7438 	    (args->share_access & ~OPEN4_SHARE_ACCESS_BOTH) ||
7439 	    (args->share_deny & ~OPEN4_SHARE_DENY_BOTH)) {
7440 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7441 		goto out;
7442 	}
7443 
7444 
7445 	/*
7446 	 * make sure attrset is zero before response is built.
7447 	 */
7448 	resp->attrset = 0;
7449 
7450 	switch (claim) {
7451 	case CLAIM_NULL:
7452 		rfs4_do_opennull(cs, req, args, oo, resp);
7453 		break;
7454 	case CLAIM_PREVIOUS:
7455 		rfs4_do_openprev(cs, req, args, oo, resp);
7456 		break;
7457 	case CLAIM_DELEGATE_CUR:
7458 		rfs4_do_opendelcur(cs, req, args, oo, resp);
7459 		break;
7460 	case CLAIM_DELEGATE_PREV:
7461 		rfs4_do_opendelprev(cs, req, args, oo, resp);
7462 		break;
7463 	default:
7464 		resp->status = NFS4ERR_INVAL;
7465 		break;
7466 	}
7467 
7468 out:
7469 	rfs4_client_rele(cp);
7470 
7471 	/* Catch sequence id handling here to make it a little easier */
7472 	switch (resp->status) {
7473 	case NFS4ERR_BADXDR:
7474 	case NFS4ERR_BAD_SEQID:
7475 	case NFS4ERR_BAD_STATEID:
7476 	case NFS4ERR_NOFILEHANDLE:
7477 	case NFS4ERR_RESOURCE:
7478 	case NFS4ERR_STALE_CLIENTID:
7479 	case NFS4ERR_STALE_STATEID:
7480 		/*
7481 		 * The protocol states that if any of these errors are
7482 		 * being returned, the sequence id should not be
7483 		 * incremented.  Any other return requires an
7484 		 * increment.
7485 		 */
7486 		break;
7487 	default:
7488 		/* Always update the lease in this case */
7489 		rfs4_update_lease(oo->ro_client);
7490 
7491 		/* Regular response - copy the result */
7492 		if (!replay)
7493 			rfs4_update_open_resp(oo, resop, &cs->fh);
7494 
7495 		/*
7496 		 * REPLAY case: Only if the previous response was OK
7497 		 * do we copy the filehandle.  If not OK, no
7498 		 * filehandle to copy.
7499 		 */
7500 		if (replay == TRUE &&
7501 		    resp->status == NFS4_OK &&
7502 		    oo->ro_reply_fh.nfs_fh4_val) {
7503 			/*
7504 			 * If this is a replay, we must restore the
7505 			 * current filehandle/vp to that of what was
7506 			 * returned originally.  Try our best to do
7507 			 * it.
7508 			 */
7509 			nfs_fh4_fmt_t *fh_fmtp =
7510 			    (nfs_fh4_fmt_t *)oo->ro_reply_fh.nfs_fh4_val;
7511 
7512 			cs->exi = checkexport4(&fh_fmtp->fh4_fsid,
7513 			    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
7514 
7515 			if (cs->exi == NULL) {
7516 				resp->status = NFS4ERR_STALE;
7517 				goto finish;
7518 			}
7519 
7520 			VN_RELE(cs->vp);
7521 
7522 			cs->vp = nfs4_fhtovp(&oo->ro_reply_fh, cs->exi,
7523 			    &resp->status);
7524 
7525 			if (cs->vp == NULL)
7526 				goto finish;
7527 
7528 			nfs_fh4_copy(&oo->ro_reply_fh, &cs->fh);
7529 		}
7530 
7531 		/*
7532 		 * If this was a replay, no need to update the
7533 		 * sequence id. If the open_owner was not created on
7534 		 * this pass, then update.  The first use of an
7535 		 * open_owner will not bump the sequence id.
7536 		 */
7537 		if (replay == FALSE && !create)
7538 			rfs4_update_open_sequence(oo);
7539 		/*
7540 		 * If the client is receiving an error and the
7541 		 * open_owner needs to be confirmed, there is no way
7542 		 * to notify the client of this fact ignoring the fact
7543 		 * that the server has no method of returning a
7544 		 * stateid to confirm.  Therefore, the server needs to
7545 		 * mark this open_owner in a way as to avoid the
7546 		 * sequence id checking the next time the client uses
7547 		 * this open_owner.
7548 		 */
7549 		if (resp->status != NFS4_OK && oo->ro_need_confirm)
7550 			oo->ro_postpone_confirm = TRUE;
7551 		/*
7552 		 * If OK response then clear the postpone flag and
7553 		 * reset the sequence id to keep in sync with the
7554 		 * client.
7555 		 */
7556 		if (resp->status == NFS4_OK && oo->ro_postpone_confirm) {
7557 			oo->ro_postpone_confirm = FALSE;
7558 			oo->ro_open_seqid = args->seqid;
7559 		}
7560 		break;
7561 	}
7562 
7563 finish:
7564 	*cs->statusp = resp->status;
7565 
7566 	rfs4_sw_exit(&oo->ro_sw);
7567 	rfs4_openowner_rele(oo);
7568 
7569 end:
7570 	DTRACE_NFSV4_2(op__open__done, struct compound_state *, cs,
7571 	    OPEN4res *, resp);
7572 }
7573 
7574 /*ARGSUSED*/
7575 void
7576 rfs4_op_open_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
7577     struct svc_req *req, struct compound_state *cs)
7578 {
7579 	OPEN_CONFIRM4args *args = &argop->nfs_argop4_u.opopen_confirm;
7580 	OPEN_CONFIRM4res *resp = &resop->nfs_resop4_u.opopen_confirm;
7581 	rfs4_state_t *sp;
7582 	nfsstat4 status;
7583 
7584 	DTRACE_NFSV4_2(op__open__confirm__start, struct compound_state *, cs,
7585 	    OPEN_CONFIRM4args *, args);
7586 
7587 	if (cs->vp == NULL) {
7588 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7589 		goto out;
7590 	}
7591 
7592 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7593 	if (status != NFS4_OK) {
7594 		*cs->statusp = resp->status = status;
7595 		goto out;
7596 	}
7597 
7598 	/* Ensure specified filehandle matches */
7599 	if (cs->vp != sp->rs_finfo->rf_vp) {
7600 		rfs4_state_rele(sp);
7601 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7602 		goto out;
7603 	}
7604 
7605 	/* hold off other access to open_owner while we tinker */
7606 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7607 
7608 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7609 	case NFS4_CHECK_STATEID_OKAY:
7610 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7611 		    resop) != 0) {
7612 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7613 			break;
7614 		}
7615 		/*
7616 		 * If it is the appropriate stateid and determined to
7617 		 * be "OKAY" then this means that the stateid does not
7618 		 * need to be confirmed and the client is in error for
7619 		 * sending an OPEN_CONFIRM.
7620 		 */
7621 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7622 		break;
7623 	case NFS4_CHECK_STATEID_OLD:
7624 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7625 		break;
7626 	case NFS4_CHECK_STATEID_BAD:
7627 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7628 		break;
7629 	case NFS4_CHECK_STATEID_EXPIRED:
7630 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7631 		break;
7632 	case NFS4_CHECK_STATEID_CLOSED:
7633 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7634 		break;
7635 	case NFS4_CHECK_STATEID_REPLAY:
7636 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7637 		    resop)) {
7638 		case NFS4_CHKSEQ_OKAY:
7639 			/*
7640 			 * This is replayed stateid; if seqid matches
7641 			 * next expected, then client is using wrong seqid.
7642 			 */
7643 			/* fall through */
7644 		case NFS4_CHKSEQ_BAD:
7645 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7646 			break;
7647 		case NFS4_CHKSEQ_REPLAY:
7648 			/*
7649 			 * Note this case is the duplicate case so
7650 			 * resp->status is already set.
7651 			 */
7652 			*cs->statusp = resp->status;
7653 			rfs4_update_lease(sp->rs_owner->ro_client);
7654 			break;
7655 		}
7656 		break;
7657 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7658 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7659 		    resop) != NFS4_CHKSEQ_OKAY) {
7660 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7661 			break;
7662 		}
7663 		*cs->statusp = resp->status = NFS4_OK;
7664 
7665 		next_stateid(&sp->rs_stateid);
7666 		resp->open_stateid = sp->rs_stateid.stateid;
7667 		sp->rs_owner->ro_need_confirm = FALSE;
7668 		rfs4_update_lease(sp->rs_owner->ro_client);
7669 		rfs4_update_open_sequence(sp->rs_owner);
7670 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7671 		break;
7672 	default:
7673 		ASSERT(FALSE);
7674 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7675 		break;
7676 	}
7677 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7678 	rfs4_state_rele(sp);
7679 
7680 out:
7681 	DTRACE_NFSV4_2(op__open__confirm__done, struct compound_state *, cs,
7682 	    OPEN_CONFIRM4res *, resp);
7683 }
7684 
7685 /*ARGSUSED*/
7686 void
7687 rfs4_op_open_downgrade(nfs_argop4 *argop, nfs_resop4 *resop,
7688     struct svc_req *req, struct compound_state *cs)
7689 {
7690 	OPEN_DOWNGRADE4args *args = &argop->nfs_argop4_u.opopen_downgrade;
7691 	OPEN_DOWNGRADE4res *resp = &resop->nfs_resop4_u.opopen_downgrade;
7692 	uint32_t access = args->share_access;
7693 	uint32_t deny = args->share_deny;
7694 	nfsstat4 status;
7695 	rfs4_state_t *sp;
7696 	rfs4_file_t *fp;
7697 	int fflags = 0;
7698 
7699 	DTRACE_NFSV4_2(op__open__downgrade__start, struct compound_state *, cs,
7700 	    OPEN_DOWNGRADE4args *, args);
7701 
7702 	if (cs->vp == NULL) {
7703 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7704 		goto out;
7705 	}
7706 
7707 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7708 	if (status != NFS4_OK) {
7709 		*cs->statusp = resp->status = status;
7710 		goto out;
7711 	}
7712 
7713 	/* Ensure specified filehandle matches */
7714 	if (cs->vp != sp->rs_finfo->rf_vp) {
7715 		rfs4_state_rele(sp);
7716 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7717 		goto out;
7718 	}
7719 
7720 	/* hold off other access to open_owner while we tinker */
7721 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7722 
7723 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7724 	case NFS4_CHECK_STATEID_OKAY:
7725 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7726 		    resop) != NFS4_CHKSEQ_OKAY) {
7727 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7728 			goto end;
7729 		}
7730 		break;
7731 	case NFS4_CHECK_STATEID_OLD:
7732 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7733 		goto end;
7734 	case NFS4_CHECK_STATEID_BAD:
7735 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7736 		goto end;
7737 	case NFS4_CHECK_STATEID_EXPIRED:
7738 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7739 		goto end;
7740 	case NFS4_CHECK_STATEID_CLOSED:
7741 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7742 		goto end;
7743 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7744 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7745 		goto end;
7746 	case NFS4_CHECK_STATEID_REPLAY:
7747 		/* Check the sequence id for the open owner */
7748 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7749 		    resop)) {
7750 		case NFS4_CHKSEQ_OKAY:
7751 			/*
7752 			 * This is replayed stateid; if seqid matches
7753 			 * next expected, then client is using wrong seqid.
7754 			 */
7755 			/* fall through */
7756 		case NFS4_CHKSEQ_BAD:
7757 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7758 			goto end;
7759 		case NFS4_CHKSEQ_REPLAY:
7760 			/*
7761 			 * Note this case is the duplicate case so
7762 			 * resp->status is already set.
7763 			 */
7764 			*cs->statusp = resp->status;
7765 			rfs4_update_lease(sp->rs_owner->ro_client);
7766 			goto end;
7767 		}
7768 		break;
7769 	default:
7770 		ASSERT(FALSE);
7771 		break;
7772 	}
7773 
7774 	rfs4_dbe_lock(sp->rs_dbe);
7775 	/*
7776 	 * Check that the new access modes and deny modes are valid.
7777 	 * Check that no invalid bits are set.
7778 	 */
7779 	if ((access & ~(OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) ||
7780 	    (deny & ~(OPEN4_SHARE_DENY_READ | OPEN4_SHARE_DENY_WRITE))) {
7781 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7782 		rfs4_update_open_sequence(sp->rs_owner);
7783 		rfs4_dbe_unlock(sp->rs_dbe);
7784 		goto end;
7785 	}
7786 
7787 	/*
7788 	 * The new modes must be a subset of the current modes and
7789 	 * the access must specify at least one mode. To test that
7790 	 * the new mode is a subset of the current modes we bitwise
7791 	 * AND them together and check that the result equals the new
7792 	 * mode. For example:
7793 	 * New mode, access == R and current mode, sp->rs_open_access  == RW
7794 	 * access & sp->rs_open_access == R == access, so the new access mode
7795 	 * is valid. Consider access == RW, sp->rs_open_access = R
7796 	 * access & sp->rs_open_access == R != access, so the new access mode
7797 	 * is invalid.
7798 	 */
7799 	if ((access & sp->rs_open_access) != access ||
7800 	    (deny & sp->rs_open_deny) != deny ||
7801 	    (access &
7802 	    (OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) == 0) {
7803 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7804 		rfs4_update_open_sequence(sp->rs_owner);
7805 		rfs4_dbe_unlock(sp->rs_dbe);
7806 		goto end;
7807 	}
7808 
7809 	/*
7810 	 * Release any share locks associated with this stateID.
7811 	 * Strictly speaking, this violates the spec because the
7812 	 * spec effectively requires that open downgrade be atomic.
7813 	 * At present, fs_shrlock does not have this capability.
7814 	 */
7815 	(void) rfs4_unshare(sp);
7816 
7817 	status = rfs4_share(sp, access, deny);
7818 	if (status != NFS4_OK) {
7819 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7820 		rfs4_update_open_sequence(sp->rs_owner);
7821 		rfs4_dbe_unlock(sp->rs_dbe);
7822 		goto end;
7823 	}
7824 
7825 	fp = sp->rs_finfo;
7826 	rfs4_dbe_lock(fp->rf_dbe);
7827 
7828 	/*
7829 	 * If the current mode has deny read and the new mode
7830 	 * does not, decrement the number of deny read mode bits
7831 	 * and if it goes to zero turn off the deny read bit
7832 	 * on the file.
7833 	 */
7834 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_READ) &&
7835 	    (deny & OPEN4_SHARE_DENY_READ) == 0) {
7836 		fp->rf_deny_read--;
7837 		if (fp->rf_deny_read == 0)
7838 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
7839 	}
7840 
7841 	/*
7842 	 * If the current mode has deny write and the new mode
7843 	 * does not, decrement the number of deny write mode bits
7844 	 * and if it goes to zero turn off the deny write bit
7845 	 * on the file.
7846 	 */
7847 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) &&
7848 	    (deny & OPEN4_SHARE_DENY_WRITE) == 0) {
7849 		fp->rf_deny_write--;
7850 		if (fp->rf_deny_write == 0)
7851 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
7852 	}
7853 
7854 	/*
7855 	 * If the current mode has access read and the new mode
7856 	 * does not, decrement the number of access read mode bits
7857 	 * and if it goes to zero turn off the access read bit
7858 	 * on the file.  set fflags to FREAD for the call to
7859 	 * vn_open_downgrade().
7860 	 */
7861 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) &&
7862 	    (access & OPEN4_SHARE_ACCESS_READ) == 0) {
7863 		fp->rf_access_read--;
7864 		if (fp->rf_access_read == 0)
7865 			fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
7866 		fflags |= FREAD;
7867 	}
7868 
7869 	/*
7870 	 * If the current mode has access write and the new mode
7871 	 * does not, decrement the number of access write mode bits
7872 	 * and if it goes to zero turn off the access write bit
7873 	 * on the file.  set fflags to FWRITE for the call to
7874 	 * vn_open_downgrade().
7875 	 */
7876 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) &&
7877 	    (access & OPEN4_SHARE_ACCESS_WRITE) == 0) {
7878 		fp->rf_access_write--;
7879 		if (fp->rf_access_write == 0)
7880 			fp->rf_share_deny &= ~OPEN4_SHARE_ACCESS_WRITE;
7881 		fflags |= FWRITE;
7882 	}
7883 
7884 	/* Check that the file is still accessible */
7885 	ASSERT(fp->rf_share_access);
7886 
7887 	rfs4_dbe_unlock(fp->rf_dbe);
7888 
7889 	/* now set the new open access and deny modes */
7890 	sp->rs_open_access = access;
7891 	sp->rs_open_deny = deny;
7892 
7893 	/*
7894 	 * we successfully downgraded the share lock, now we need to downgrade
7895 	 * the open. it is possible that the downgrade was only for a deny
7896 	 * mode and we have nothing else to do.
7897 	 */
7898 	if ((fflags & (FREAD|FWRITE)) != 0)
7899 		vn_open_downgrade(cs->vp, fflags);
7900 
7901 	/* Update the stateid */
7902 	next_stateid(&sp->rs_stateid);
7903 	resp->open_stateid = sp->rs_stateid.stateid;
7904 
7905 	rfs4_dbe_unlock(sp->rs_dbe);
7906 
7907 	*cs->statusp = resp->status = NFS4_OK;
7908 	/* Update the lease */
7909 	rfs4_update_lease(sp->rs_owner->ro_client);
7910 	/* And the sequence */
7911 	rfs4_update_open_sequence(sp->rs_owner);
7912 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7913 
7914 end:
7915 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7916 	rfs4_state_rele(sp);
7917 out:
7918 	DTRACE_NFSV4_2(op__open__downgrade__done, struct compound_state *, cs,
7919 	    OPEN_DOWNGRADE4res *, resp);
7920 }
7921 
7922 /*
7923  * The logic behind this function is detailed in the NFSv4 RFC in the
7924  * SETCLIENTID operation description under IMPLEMENTATION.  Refer to
7925  * that section for explicit guidance to server behavior for
7926  * SETCLIENTID.
7927  */
7928 void
7929 rfs4_op_setclientid(nfs_argop4 *argop, nfs_resop4 *resop,
7930     struct svc_req *req, struct compound_state *cs)
7931 {
7932 	SETCLIENTID4args *args = &argop->nfs_argop4_u.opsetclientid;
7933 	SETCLIENTID4res *res = &resop->nfs_resop4_u.opsetclientid;
7934 	rfs4_client_t *cp, *newcp, *cp_confirmed, *cp_unconfirmed;
7935 	rfs4_clntip_t *ci;
7936 	bool_t create;
7937 	char *addr, *netid;
7938 	int len;
7939 
7940 	DTRACE_NFSV4_2(op__setclientid__start, struct compound_state *, cs,
7941 	    SETCLIENTID4args *, args);
7942 retry:
7943 	newcp = cp_confirmed = cp_unconfirmed = NULL;
7944 
7945 	/*
7946 	 * Save the caller's IP address
7947 	 */
7948 	args->client.cl_addr =
7949 	    (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
7950 
7951 	/*
7952 	 * Record if it is a Solaris client that cannot handle referrals.
7953 	 */
7954 	if (strstr(args->client.id_val, "Solaris") &&
7955 	    !strstr(args->client.id_val, "+referrals")) {
7956 		/* Add a "yes, it's downrev" record */
7957 		create = TRUE;
7958 		ci = rfs4_find_clntip(args->client.cl_addr, &create);
7959 		ASSERT(ci != NULL);
7960 		rfs4_dbe_rele(ci->ri_dbe);
7961 	} else {
7962 		/* Remove any previous record */
7963 		rfs4_invalidate_clntip(args->client.cl_addr);
7964 	}
7965 
7966 	/*
7967 	 * In search of an EXISTING client matching the incoming
7968 	 * request to establish a new client identifier at the server
7969 	 */
7970 	create = TRUE;
7971 	cp = rfs4_findclient(&args->client, &create, NULL);
7972 
7973 	/* Should never happen */
7974 	ASSERT(cp != NULL);
7975 
7976 	if (cp == NULL) {
7977 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
7978 		goto out;
7979 	}
7980 
7981 	/*
7982 	 * Easiest case. Client identifier is newly created and is
7983 	 * unconfirmed.  Also note that for this case, no other
7984 	 * entries exist for the client identifier.  Nothing else to
7985 	 * check.  Just setup the response and respond.
7986 	 */
7987 	if (create) {
7988 		*cs->statusp = res->status = NFS4_OK;
7989 		res->SETCLIENTID4res_u.resok4.clientid = cp->rc_clientid;
7990 		res->SETCLIENTID4res_u.resok4.setclientid_confirm =
7991 		    cp->rc_confirm_verf;
7992 		/* Setup callback information; CB_NULL confirmation later */
7993 		rfs4_client_setcb(cp, &args->callback, args->callback_ident);
7994 
7995 		rfs4_client_rele(cp);
7996 		goto out;
7997 	}
7998 
7999 	/*
8000 	 * An existing, confirmed client may exist but it may not have
8001 	 * been active for at least one lease period.  If so, then
8002 	 * "close" the client and create a new client identifier
8003 	 */
8004 	if (rfs4_lease_expired(cp)) {
8005 		rfs4_client_close(cp);
8006 		goto retry;
8007 	}
8008 
8009 	if (cp->rc_need_confirm == TRUE)
8010 		cp_unconfirmed = cp;
8011 	else
8012 		cp_confirmed = cp;
8013 
8014 	cp = NULL;
8015 
8016 	/*
8017 	 * We have a confirmed client, now check for an
8018 	 * unconfimred entry
8019 	 */
8020 	if (cp_confirmed) {
8021 		/* If creds don't match then client identifier is inuse */
8022 		if (!creds_ok(cp_confirmed->rc_cr_set, req, cs)) {
8023 			rfs4_cbinfo_t *cbp;
8024 			/*
8025 			 * Some one else has established this client
8026 			 * id. Try and say * who they are. We will use
8027 			 * the call back address supplied by * the
8028 			 * first client.
8029 			 */
8030 			*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8031 
8032 			addr = netid = NULL;
8033 
8034 			cbp = &cp_confirmed->rc_cbinfo;
8035 			if (cbp->cb_callback.cb_location.r_addr &&
8036 			    cbp->cb_callback.cb_location.r_netid) {
8037 				cb_client4 *cbcp = &cbp->cb_callback;
8038 
8039 				len = strlen(cbcp->cb_location.r_addr)+1;
8040 				addr = kmem_alloc(len, KM_SLEEP);
8041 				bcopy(cbcp->cb_location.r_addr, addr, len);
8042 				len = strlen(cbcp->cb_location.r_netid)+1;
8043 				netid = kmem_alloc(len, KM_SLEEP);
8044 				bcopy(cbcp->cb_location.r_netid, netid, len);
8045 			}
8046 
8047 			res->SETCLIENTID4res_u.client_using.r_addr = addr;
8048 			res->SETCLIENTID4res_u.client_using.r_netid = netid;
8049 
8050 			rfs4_client_rele(cp_confirmed);
8051 		}
8052 
8053 		/*
8054 		 * Confirmed, creds match, and verifier matches; must
8055 		 * be an update of the callback info
8056 		 */
8057 		if (cp_confirmed->rc_nfs_client.verifier ==
8058 		    args->client.verifier) {
8059 			/* Setup callback information */
8060 			rfs4_client_setcb(cp_confirmed, &args->callback,
8061 			    args->callback_ident);
8062 
8063 			/* everything okay -- move ahead */
8064 			*cs->statusp = res->status = NFS4_OK;
8065 			res->SETCLIENTID4res_u.resok4.clientid =
8066 			    cp_confirmed->rc_clientid;
8067 
8068 			/* update the confirm_verifier and return it */
8069 			rfs4_client_scv_next(cp_confirmed);
8070 			res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8071 			    cp_confirmed->rc_confirm_verf;
8072 
8073 			rfs4_client_rele(cp_confirmed);
8074 			goto out;
8075 		}
8076 
8077 		/*
8078 		 * Creds match but the verifier doesn't.  Must search
8079 		 * for an unconfirmed client that would be replaced by
8080 		 * this request.
8081 		 */
8082 		create = FALSE;
8083 		cp_unconfirmed = rfs4_findclient(&args->client, &create,
8084 		    cp_confirmed);
8085 	}
8086 
8087 	/*
8088 	 * At this point, we have taken care of the brand new client
8089 	 * struct, INUSE case, update of an existing, and confirmed
8090 	 * client struct.
8091 	 */
8092 
8093 	/*
8094 	 * check to see if things have changed while we originally
8095 	 * picked up the client struct.  If they have, then return and
8096 	 * retry the processing of this SETCLIENTID request.
8097 	 */
8098 	if (cp_unconfirmed) {
8099 		rfs4_dbe_lock(cp_unconfirmed->rc_dbe);
8100 		if (!cp_unconfirmed->rc_need_confirm) {
8101 			rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8102 			rfs4_client_rele(cp_unconfirmed);
8103 			if (cp_confirmed)
8104 				rfs4_client_rele(cp_confirmed);
8105 			goto retry;
8106 		}
8107 		/* do away with the old unconfirmed one */
8108 		rfs4_dbe_invalidate(cp_unconfirmed->rc_dbe);
8109 		rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8110 		rfs4_client_rele(cp_unconfirmed);
8111 		cp_unconfirmed = NULL;
8112 	}
8113 
8114 	/*
8115 	 * This search will temporarily hide the confirmed client
8116 	 * struct while a new client struct is created as the
8117 	 * unconfirmed one.
8118 	 */
8119 	create = TRUE;
8120 	newcp = rfs4_findclient(&args->client, &create, cp_confirmed);
8121 
8122 	ASSERT(newcp != NULL);
8123 
8124 	if (newcp == NULL) {
8125 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8126 		rfs4_client_rele(cp_confirmed);
8127 		goto out;
8128 	}
8129 
8130 	/*
8131 	 * If one was not created, then a similar request must be in
8132 	 * process so release and start over with this one
8133 	 */
8134 	if (create != TRUE) {
8135 		rfs4_client_rele(newcp);
8136 		if (cp_confirmed)
8137 			rfs4_client_rele(cp_confirmed);
8138 		goto retry;
8139 	}
8140 
8141 	*cs->statusp = res->status = NFS4_OK;
8142 	res->SETCLIENTID4res_u.resok4.clientid = newcp->rc_clientid;
8143 	res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8144 	    newcp->rc_confirm_verf;
8145 	/* Setup callback information; CB_NULL confirmation later */
8146 	rfs4_client_setcb(newcp, &args->callback, args->callback_ident);
8147 
8148 	newcp->rc_cp_confirmed = cp_confirmed;
8149 
8150 	rfs4_client_rele(newcp);
8151 
8152 out:
8153 	DTRACE_NFSV4_2(op__setclientid__done, struct compound_state *, cs,
8154 	    SETCLIENTID4res *, res);
8155 }
8156 
8157 /*ARGSUSED*/
8158 void
8159 rfs4_op_setclientid_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
8160     struct svc_req *req, struct compound_state *cs)
8161 {
8162 	SETCLIENTID_CONFIRM4args *args =
8163 	    &argop->nfs_argop4_u.opsetclientid_confirm;
8164 	SETCLIENTID_CONFIRM4res *res =
8165 	    &resop->nfs_resop4_u.opsetclientid_confirm;
8166 	rfs4_client_t *cp, *cptoclose = NULL;
8167 
8168 	DTRACE_NFSV4_2(op__setclientid__confirm__start,
8169 	    struct compound_state *, cs,
8170 	    SETCLIENTID_CONFIRM4args *, args);
8171 
8172 	*cs->statusp = res->status = NFS4_OK;
8173 
8174 	cp = rfs4_findclient_by_id(args->clientid, TRUE);
8175 
8176 	if (cp == NULL) {
8177 		*cs->statusp = res->status =
8178 		    rfs4_check_clientid(&args->clientid, 1);
8179 		goto out;
8180 	}
8181 
8182 	if (!creds_ok(cp, req, cs)) {
8183 		*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8184 		rfs4_client_rele(cp);
8185 		goto out;
8186 	}
8187 
8188 	/* If the verifier doesn't match, the record doesn't match */
8189 	if (cp->rc_confirm_verf != args->setclientid_confirm) {
8190 		*cs->statusp = res->status = NFS4ERR_STALE_CLIENTID;
8191 		rfs4_client_rele(cp);
8192 		goto out;
8193 	}
8194 
8195 	rfs4_dbe_lock(cp->rc_dbe);
8196 	cp->rc_need_confirm = FALSE;
8197 	if (cp->rc_cp_confirmed) {
8198 		cptoclose = cp->rc_cp_confirmed;
8199 		cptoclose->rc_ss_remove = 1;
8200 		cp->rc_cp_confirmed = NULL;
8201 	}
8202 
8203 	/*
8204 	 * Update the client's associated server instance, if it's changed
8205 	 * since the client was created.
8206 	 */
8207 	if (rfs4_servinst(cp) != rfs4_cur_servinst)
8208 		rfs4_servinst_assign(cp, rfs4_cur_servinst);
8209 
8210 	/*
8211 	 * Record clientid in stable storage.
8212 	 * Must be done after server instance has been assigned.
8213 	 */
8214 	rfs4_ss_clid(cp);
8215 
8216 	rfs4_dbe_unlock(cp->rc_dbe);
8217 
8218 	if (cptoclose)
8219 		/* don't need to rele, client_close does it */
8220 		rfs4_client_close(cptoclose);
8221 
8222 	/* If needed, initiate CB_NULL call for callback path */
8223 	rfs4_deleg_cb_check(cp);
8224 	rfs4_update_lease(cp);
8225 
8226 	/*
8227 	 * Check to see if client can perform reclaims
8228 	 */
8229 	rfs4_ss_chkclid(cp);
8230 
8231 	rfs4_client_rele(cp);
8232 
8233 out:
8234 	DTRACE_NFSV4_2(op__setclientid__confirm__done,
8235 	    struct compound_state *, cs,
8236 	    SETCLIENTID_CONFIRM4 *, res);
8237 }
8238 
8239 
8240 /*ARGSUSED*/
8241 void
8242 rfs4_op_close(nfs_argop4 *argop, nfs_resop4 *resop,
8243     struct svc_req *req, struct compound_state *cs)
8244 {
8245 	CLOSE4args *args = &argop->nfs_argop4_u.opclose;
8246 	CLOSE4res *resp = &resop->nfs_resop4_u.opclose;
8247 	rfs4_state_t *sp;
8248 	nfsstat4 status;
8249 
8250 	DTRACE_NFSV4_2(op__close__start, struct compound_state *, cs,
8251 	    CLOSE4args *, args);
8252 
8253 	if (cs->vp == NULL) {
8254 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8255 		goto out;
8256 	}
8257 
8258 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_INVALID);
8259 	if (status != NFS4_OK) {
8260 		*cs->statusp = resp->status = status;
8261 		goto out;
8262 	}
8263 
8264 	/* Ensure specified filehandle matches */
8265 	if (cs->vp != sp->rs_finfo->rf_vp) {
8266 		rfs4_state_rele(sp);
8267 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8268 		goto out;
8269 	}
8270 
8271 	/* hold off other access to open_owner while we tinker */
8272 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
8273 
8274 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
8275 	case NFS4_CHECK_STATEID_OKAY:
8276 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8277 		    resop) != NFS4_CHKSEQ_OKAY) {
8278 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8279 			goto end;
8280 		}
8281 		break;
8282 	case NFS4_CHECK_STATEID_OLD:
8283 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8284 		goto end;
8285 	case NFS4_CHECK_STATEID_BAD:
8286 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8287 		goto end;
8288 	case NFS4_CHECK_STATEID_EXPIRED:
8289 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8290 		goto end;
8291 	case NFS4_CHECK_STATEID_CLOSED:
8292 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8293 		goto end;
8294 	case NFS4_CHECK_STATEID_UNCONFIRMED:
8295 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8296 		goto end;
8297 	case NFS4_CHECK_STATEID_REPLAY:
8298 		/* Check the sequence id for the open owner */
8299 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8300 		    resop)) {
8301 		case NFS4_CHKSEQ_OKAY:
8302 			/*
8303 			 * This is replayed stateid; if seqid matches
8304 			 * next expected, then client is using wrong seqid.
8305 			 */
8306 			/* FALL THROUGH */
8307 		case NFS4_CHKSEQ_BAD:
8308 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8309 			goto end;
8310 		case NFS4_CHKSEQ_REPLAY:
8311 			/*
8312 			 * Note this case is the duplicate case so
8313 			 * resp->status is already set.
8314 			 */
8315 			*cs->statusp = resp->status;
8316 			rfs4_update_lease(sp->rs_owner->ro_client);
8317 			goto end;
8318 		}
8319 		break;
8320 	default:
8321 		ASSERT(FALSE);
8322 		break;
8323 	}
8324 
8325 	rfs4_dbe_lock(sp->rs_dbe);
8326 
8327 	/* Update the stateid. */
8328 	next_stateid(&sp->rs_stateid);
8329 	resp->open_stateid = sp->rs_stateid.stateid;
8330 
8331 	rfs4_dbe_unlock(sp->rs_dbe);
8332 
8333 	rfs4_update_lease(sp->rs_owner->ro_client);
8334 	rfs4_update_open_sequence(sp->rs_owner);
8335 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8336 
8337 	rfs4_state_close(sp, FALSE, FALSE, cs->cr);
8338 
8339 	*cs->statusp = resp->status = status;
8340 
8341 end:
8342 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
8343 	rfs4_state_rele(sp);
8344 out:
8345 	DTRACE_NFSV4_2(op__close__done, struct compound_state *, cs,
8346 	    CLOSE4res *, resp);
8347 }
8348 
8349 /*
8350  * Manage the counts on the file struct and close all file locks
8351  */
8352 /*ARGSUSED*/
8353 void
8354 rfs4_release_share_lock_state(rfs4_state_t *sp, cred_t *cr,
8355     bool_t close_of_client)
8356 {
8357 	rfs4_file_t *fp = sp->rs_finfo;
8358 	rfs4_lo_state_t *lsp;
8359 	int fflags = 0;
8360 
8361 	/*
8362 	 * If this call is part of the larger closing down of client
8363 	 * state then it is just easier to release all locks
8364 	 * associated with this client instead of going through each
8365 	 * individual file and cleaning locks there.
8366 	 */
8367 	if (close_of_client) {
8368 		if (sp->rs_owner->ro_client->rc_unlksys_completed == FALSE &&
8369 		    !list_is_empty(&sp->rs_lostatelist) &&
8370 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID) {
8371 			/* Is the PxFS kernel module loaded? */
8372 			if (lm_remove_file_locks != NULL) {
8373 				int new_sysid;
8374 
8375 				/* Encode the cluster nodeid in new sysid */
8376 				new_sysid = sp->rs_owner->ro_client->rc_sysidt;
8377 				lm_set_nlmid_flk(&new_sysid);
8378 
8379 				/*
8380 				 * This PxFS routine removes file locks for a
8381 				 * client over all nodes of a cluster.
8382 				 */
8383 				NFS4_DEBUG(rfs4_debug, (CE_NOTE,
8384 				    "lm_remove_file_locks(sysid=0x%x)\n",
8385 				    new_sysid));
8386 				(*lm_remove_file_locks)(new_sysid);
8387 			} else {
8388 				struct flock64 flk;
8389 
8390 				/* Release all locks for this client */
8391 				flk.l_type = F_UNLKSYS;
8392 				flk.l_whence = 0;
8393 				flk.l_start = 0;
8394 				flk.l_len = 0;
8395 				flk.l_sysid =
8396 				    sp->rs_owner->ro_client->rc_sysidt;
8397 				flk.l_pid = 0;
8398 				(void) VOP_FRLOCK(sp->rs_finfo->rf_vp, F_SETLK,
8399 				    &flk, F_REMOTELOCK | FREAD | FWRITE,
8400 				    (u_offset_t)0, NULL, CRED(), NULL);
8401 			}
8402 
8403 			sp->rs_owner->ro_client->rc_unlksys_completed = TRUE;
8404 		}
8405 	}
8406 
8407 	/*
8408 	 * Release all locks on this file by this lock owner or at
8409 	 * least mark the locks as having been released
8410 	 */
8411 	for (lsp = list_head(&sp->rs_lostatelist); lsp != NULL;
8412 	    lsp = list_next(&sp->rs_lostatelist, lsp)) {
8413 		lsp->rls_locks_cleaned = TRUE;
8414 
8415 		/* Was this already taken care of above? */
8416 		if (!close_of_client &&
8417 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8418 			(void) cleanlocks(sp->rs_finfo->rf_vp,
8419 			    lsp->rls_locker->rl_pid,
8420 			    lsp->rls_locker->rl_client->rc_sysidt);
8421 	}
8422 
8423 	/*
8424 	 * Release any shrlocks associated with this open state ID.
8425 	 * This must be done before the rfs4_state gets marked closed.
8426 	 */
8427 	if (sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8428 		(void) rfs4_unshare(sp);
8429 
8430 	if (sp->rs_open_access) {
8431 		rfs4_dbe_lock(fp->rf_dbe);
8432 
8433 		/*
8434 		 * Decrement the count for each access and deny bit that this
8435 		 * state has contributed to the file.
8436 		 * If the file counts go to zero
8437 		 * clear the appropriate bit in the appropriate mask.
8438 		 */
8439 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) {
8440 			fp->rf_access_read--;
8441 			fflags |= FREAD;
8442 			if (fp->rf_access_read == 0)
8443 				fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
8444 		}
8445 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) {
8446 			fp->rf_access_write--;
8447 			fflags |= FWRITE;
8448 			if (fp->rf_access_write == 0)
8449 				fp->rf_share_access &=
8450 				    ~OPEN4_SHARE_ACCESS_WRITE;
8451 		}
8452 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_READ) {
8453 			fp->rf_deny_read--;
8454 			if (fp->rf_deny_read == 0)
8455 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
8456 		}
8457 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) {
8458 			fp->rf_deny_write--;
8459 			if (fp->rf_deny_write == 0)
8460 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
8461 		}
8462 
8463 		(void) VOP_CLOSE(fp->rf_vp, fflags, 1, (offset_t)0, cr, NULL);
8464 
8465 		rfs4_dbe_unlock(fp->rf_dbe);
8466 
8467 		sp->rs_open_access = 0;
8468 		sp->rs_open_deny = 0;
8469 	}
8470 }
8471 
8472 /*
8473  * lock_denied: Fill in a LOCK4deneid structure given an flock64 structure.
8474  */
8475 static nfsstat4
8476 lock_denied(LOCK4denied *dp, struct flock64 *flk)
8477 {
8478 	rfs4_lockowner_t *lo;
8479 	rfs4_client_t *cp;
8480 	uint32_t len;
8481 
8482 	lo = rfs4_findlockowner_by_pid(flk->l_pid);
8483 	if (lo != NULL) {
8484 		cp = lo->rl_client;
8485 		if (rfs4_lease_expired(cp)) {
8486 			rfs4_lockowner_rele(lo);
8487 			rfs4_dbe_hold(cp->rc_dbe);
8488 			rfs4_client_close(cp);
8489 			return (NFS4ERR_EXPIRED);
8490 		}
8491 		dp->owner.clientid = lo->rl_owner.clientid;
8492 		len = lo->rl_owner.owner_len;
8493 		dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8494 		bcopy(lo->rl_owner.owner_val, dp->owner.owner_val, len);
8495 		dp->owner.owner_len = len;
8496 		rfs4_lockowner_rele(lo);
8497 		goto finish;
8498 	}
8499 
8500 	/*
8501 	 * Its not a NFS4 lock. We take advantage that the upper 32 bits
8502 	 * of the client id contain the boot time for a NFS4 lock. So we
8503 	 * fabricate and identity by setting clientid to the sysid, and
8504 	 * the lock owner to the pid.
8505 	 */
8506 	dp->owner.clientid = flk->l_sysid;
8507 	len = sizeof (pid_t);
8508 	dp->owner.owner_len = len;
8509 	dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8510 	bcopy(&flk->l_pid, dp->owner.owner_val, len);
8511 finish:
8512 	dp->offset = flk->l_start;
8513 	dp->length = flk->l_len;
8514 
8515 	if (flk->l_type == F_RDLCK)
8516 		dp->locktype = READ_LT;
8517 	else if (flk->l_type == F_WRLCK)
8518 		dp->locktype = WRITE_LT;
8519 	else
8520 		return (NFS4ERR_INVAL);	/* no mapping from POSIX ltype to v4 */
8521 
8522 	return (NFS4_OK);
8523 }
8524 
8525 static int
8526 setlock(vnode_t *vp, struct flock64 *flock, int flag, cred_t *cred)
8527 {
8528 	int error;
8529 	struct flock64 flk;
8530 	int i;
8531 	clock_t delaytime;
8532 	int cmd;
8533 
8534 	cmd = nbl_need_check(vp) ? F_SETLK_NBMAND : F_SETLK;
8535 retry:
8536 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
8537 
8538 	for (i = 0; i < rfs4_maxlock_tries; i++) {
8539 		LOCK_PRINT(rfs4_debug, "setlock", cmd, flock);
8540 		error = VOP_FRLOCK(vp, cmd,
8541 		    flock, flag, (u_offset_t)0, NULL, cred, NULL);
8542 
8543 		if (error != EAGAIN && error != EACCES)
8544 			break;
8545 
8546 		if (i < rfs4_maxlock_tries - 1) {
8547 			delay(delaytime);
8548 			delaytime *= 2;
8549 		}
8550 	}
8551 
8552 	if (error == EAGAIN || error == EACCES) {
8553 		/* Get the owner of the lock */
8554 		flk = *flock;
8555 		LOCK_PRINT(rfs4_debug, "setlock", F_GETLK, &flk);
8556 		if (VOP_FRLOCK(vp, F_GETLK, &flk, flag,
8557 		    (u_offset_t)0, NULL, cred, NULL) == 0) {
8558 			if (flk.l_type == F_UNLCK) {
8559 				/* No longer locked, retry */
8560 				goto retry;
8561 			}
8562 			*flock = flk;
8563 			LOCK_PRINT(rfs4_debug, "setlock(blocking lock)",
8564 			    F_GETLK, &flk);
8565 		}
8566 	}
8567 
8568 	return (error);
8569 }
8570 
8571 /*ARGSUSED*/
8572 static nfsstat4
8573 rfs4_do_lock(rfs4_lo_state_t *lsp, nfs_lock_type4 locktype,
8574     offset4 offset, length4 length, cred_t *cred, nfs_resop4 *resop)
8575 {
8576 	nfsstat4 status;
8577 	rfs4_lockowner_t *lo = lsp->rls_locker;
8578 	rfs4_state_t *sp = lsp->rls_state;
8579 	struct flock64 flock;
8580 	int16_t ltype;
8581 	int flag;
8582 	int error;
8583 	sysid_t sysid;
8584 	LOCK4res *lres;
8585 
8586 	if (rfs4_lease_expired(lo->rl_client)) {
8587 		return (NFS4ERR_EXPIRED);
8588 	}
8589 
8590 	if ((status = rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
8591 		return (status);
8592 
8593 	/* Check for zero length. To lock to end of file use all ones for V4 */
8594 	if (length == 0)
8595 		return (NFS4ERR_INVAL);
8596 	else if (length == (length4)(~0))
8597 		length = 0;		/* Posix to end of file  */
8598 
8599 retry:
8600 	rfs4_dbe_lock(sp->rs_dbe);
8601 	if (sp->rs_closed) {
8602 		rfs4_dbe_unlock(sp->rs_dbe);
8603 		return (NFS4ERR_OLD_STATEID);
8604 	}
8605 
8606 	if (resop->resop != OP_LOCKU) {
8607 		switch (locktype) {
8608 		case READ_LT:
8609 		case READW_LT:
8610 			if ((sp->rs_share_access
8611 			    & OPEN4_SHARE_ACCESS_READ) == 0) {
8612 				rfs4_dbe_unlock(sp->rs_dbe);
8613 
8614 				return (NFS4ERR_OPENMODE);
8615 			}
8616 			ltype = F_RDLCK;
8617 			break;
8618 		case WRITE_LT:
8619 		case WRITEW_LT:
8620 			if ((sp->rs_share_access
8621 			    & OPEN4_SHARE_ACCESS_WRITE) == 0) {
8622 				rfs4_dbe_unlock(sp->rs_dbe);
8623 
8624 				return (NFS4ERR_OPENMODE);
8625 			}
8626 			ltype = F_WRLCK;
8627 			break;
8628 		}
8629 	} else
8630 		ltype = F_UNLCK;
8631 
8632 	flock.l_type = ltype;
8633 	flock.l_whence = 0;		/* SEEK_SET */
8634 	flock.l_start = offset;
8635 	flock.l_len = length;
8636 	flock.l_sysid = sysid;
8637 	flock.l_pid = lsp->rls_locker->rl_pid;
8638 
8639 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
8640 	if (flock.l_len < 0 || flock.l_start < 0) {
8641 		rfs4_dbe_unlock(sp->rs_dbe);
8642 		return (NFS4ERR_INVAL);
8643 	}
8644 
8645 	/*
8646 	 * N.B. FREAD has the same value as OPEN4_SHARE_ACCESS_READ and
8647 	 * FWRITE has the same value as OPEN4_SHARE_ACCESS_WRITE.
8648 	 */
8649 	flag = (int)sp->rs_share_access | F_REMOTELOCK;
8650 
8651 	error = setlock(sp->rs_finfo->rf_vp, &flock, flag, cred);
8652 	if (error == 0) {
8653 		rfs4_dbe_lock(lsp->rls_dbe);
8654 		next_stateid(&lsp->rls_lockid);
8655 		rfs4_dbe_unlock(lsp->rls_dbe);
8656 	}
8657 
8658 	rfs4_dbe_unlock(sp->rs_dbe);
8659 
8660 	/*
8661 	 * N.B. We map error values to nfsv4 errors. This is differrent
8662 	 * than puterrno4 routine.
8663 	 */
8664 	switch (error) {
8665 	case 0:
8666 		status = NFS4_OK;
8667 		break;
8668 	case EAGAIN:
8669 	case EACCES:		/* Old value */
8670 		/* Can only get here if op is OP_LOCK */
8671 		ASSERT(resop->resop == OP_LOCK);
8672 		lres = &resop->nfs_resop4_u.oplock;
8673 		status = NFS4ERR_DENIED;
8674 		if (lock_denied(&lres->LOCK4res_u.denied, &flock)
8675 		    == NFS4ERR_EXPIRED)
8676 			goto retry;
8677 		break;
8678 	case ENOLCK:
8679 		status = NFS4ERR_DELAY;
8680 		break;
8681 	case EOVERFLOW:
8682 		status = NFS4ERR_INVAL;
8683 		break;
8684 	case EINVAL:
8685 		status = NFS4ERR_NOTSUPP;
8686 		break;
8687 	default:
8688 		status = NFS4ERR_SERVERFAULT;
8689 		break;
8690 	}
8691 
8692 	return (status);
8693 }
8694 
8695 /*ARGSUSED*/
8696 void
8697 rfs4_op_lock(nfs_argop4 *argop, nfs_resop4 *resop,
8698     struct svc_req *req, struct compound_state *cs)
8699 {
8700 	LOCK4args *args = &argop->nfs_argop4_u.oplock;
8701 	LOCK4res *resp = &resop->nfs_resop4_u.oplock;
8702 	nfsstat4 status;
8703 	stateid4 *stateid;
8704 	rfs4_lockowner_t *lo;
8705 	rfs4_client_t *cp;
8706 	rfs4_state_t *sp = NULL;
8707 	rfs4_lo_state_t *lsp = NULL;
8708 	bool_t ls_sw_held = FALSE;
8709 	bool_t create = TRUE;
8710 	bool_t lcreate = TRUE;
8711 	bool_t dup_lock = FALSE;
8712 	int rc;
8713 
8714 	DTRACE_NFSV4_2(op__lock__start, struct compound_state *, cs,
8715 	    LOCK4args *, args);
8716 
8717 	if (cs->vp == NULL) {
8718 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8719 		DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8720 		    cs, LOCK4res *, resp);
8721 		return;
8722 	}
8723 
8724 	if (args->locker.new_lock_owner) {
8725 		/* Create a new lockowner for this instance */
8726 		open_to_lock_owner4 *olo = &args->locker.locker4_u.open_owner;
8727 
8728 		NFS4_DEBUG(rfs4_debug, (CE_NOTE, "Creating new lock owner"));
8729 
8730 		stateid = &olo->open_stateid;
8731 		status = rfs4_get_state(stateid, &sp, RFS4_DBS_VALID);
8732 		if (status != NFS4_OK) {
8733 			NFS4_DEBUG(rfs4_debug,
8734 			    (CE_NOTE, "Get state failed in lock %d", status));
8735 			*cs->statusp = resp->status = status;
8736 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8737 			    cs, LOCK4res *, resp);
8738 			return;
8739 		}
8740 
8741 		/* Ensure specified filehandle matches */
8742 		if (cs->vp != sp->rs_finfo->rf_vp) {
8743 			rfs4_state_rele(sp);
8744 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8745 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8746 			    cs, LOCK4res *, resp);
8747 			return;
8748 		}
8749 
8750 		/* hold off other access to open_owner while we tinker */
8751 		rfs4_sw_enter(&sp->rs_owner->ro_sw);
8752 
8753 		switch (rc = rfs4_check_stateid_seqid(sp, stateid)) {
8754 		case NFS4_CHECK_STATEID_OLD:
8755 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8756 			goto end;
8757 		case NFS4_CHECK_STATEID_BAD:
8758 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8759 			goto end;
8760 		case NFS4_CHECK_STATEID_EXPIRED:
8761 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8762 			goto end;
8763 		case NFS4_CHECK_STATEID_UNCONFIRMED:
8764 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8765 			goto end;
8766 		case NFS4_CHECK_STATEID_CLOSED:
8767 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8768 			goto end;
8769 		case NFS4_CHECK_STATEID_OKAY:
8770 		case NFS4_CHECK_STATEID_REPLAY:
8771 			switch (rfs4_check_olo_seqid(olo->open_seqid,
8772 			    sp->rs_owner, resop)) {
8773 			case NFS4_CHKSEQ_OKAY:
8774 				if (rc == NFS4_CHECK_STATEID_OKAY)
8775 					break;
8776 				/*
8777 				 * This is replayed stateid; if seqid
8778 				 * matches next expected, then client
8779 				 * is using wrong seqid.
8780 				 */
8781 				/* FALLTHROUGH */
8782 			case NFS4_CHKSEQ_BAD:
8783 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8784 				goto end;
8785 			case NFS4_CHKSEQ_REPLAY:
8786 				/* This is a duplicate LOCK request */
8787 				dup_lock = TRUE;
8788 
8789 				/*
8790 				 * For a duplicate we do not want to
8791 				 * create a new lockowner as it should
8792 				 * already exist.
8793 				 * Turn off the lockowner create flag.
8794 				 */
8795 				lcreate = FALSE;
8796 			}
8797 			break;
8798 		}
8799 
8800 		lo = rfs4_findlockowner(&olo->lock_owner, &lcreate);
8801 		if (lo == NULL) {
8802 			NFS4_DEBUG(rfs4_debug,
8803 			    (CE_NOTE, "rfs4_op_lock: no lock owner"));
8804 			*cs->statusp = resp->status = NFS4ERR_RESOURCE;
8805 			goto end;
8806 		}
8807 
8808 		lsp = rfs4_findlo_state_by_owner(lo, sp, &create);
8809 		if (lsp == NULL) {
8810 			rfs4_update_lease(sp->rs_owner->ro_client);
8811 			/*
8812 			 * Only update theh open_seqid if this is not
8813 			 * a duplicate request
8814 			 */
8815 			if (dup_lock == FALSE) {
8816 				rfs4_update_open_sequence(sp->rs_owner);
8817 			}
8818 
8819 			NFS4_DEBUG(rfs4_debug,
8820 			    (CE_NOTE, "rfs4_op_lock: no state"));
8821 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
8822 			rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8823 			rfs4_lockowner_rele(lo);
8824 			goto end;
8825 		}
8826 
8827 		/*
8828 		 * This is the new_lock_owner branch and the client is
8829 		 * supposed to be associating a new lock_owner with
8830 		 * the open file at this point.  If we find that a
8831 		 * lock_owner/state association already exists and a
8832 		 * successful LOCK request was returned to the client,
8833 		 * an error is returned to the client since this is
8834 		 * not appropriate.  The client should be using the
8835 		 * existing lock_owner branch.
8836 		 */
8837 		if (dup_lock == FALSE && create == FALSE) {
8838 			if (lsp->rls_lock_completed == TRUE) {
8839 				*cs->statusp =
8840 				    resp->status = NFS4ERR_BAD_SEQID;
8841 				rfs4_lockowner_rele(lo);
8842 				goto end;
8843 			}
8844 		}
8845 
8846 		rfs4_update_lease(sp->rs_owner->ro_client);
8847 
8848 		/*
8849 		 * Only update theh open_seqid if this is not
8850 		 * a duplicate request
8851 		 */
8852 		if (dup_lock == FALSE) {
8853 			rfs4_update_open_sequence(sp->rs_owner);
8854 		}
8855 
8856 		/*
8857 		 * If this is a duplicate lock request, just copy the
8858 		 * previously saved reply and return.
8859 		 */
8860 		if (dup_lock == TRUE) {
8861 			/* verify that lock_seqid's match */
8862 			if (lsp->rls_seqid != olo->lock_seqid) {
8863 				NFS4_DEBUG(rfs4_debug,
8864 				    (CE_NOTE, "rfs4_op_lock: Dup-Lock seqid bad"
8865 				    "lsp->seqid=%d old->seqid=%d",
8866 				    lsp->rls_seqid, olo->lock_seqid));
8867 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8868 			} else {
8869 				rfs4_copy_reply(resop, &lsp->rls_reply);
8870 				/*
8871 				 * Make sure to copy the just
8872 				 * retrieved reply status into the
8873 				 * overall compound status
8874 				 */
8875 				*cs->statusp = resp->status;
8876 			}
8877 			rfs4_lockowner_rele(lo);
8878 			goto end;
8879 		}
8880 
8881 		rfs4_dbe_lock(lsp->rls_dbe);
8882 
8883 		/* Make sure to update the lock sequence id */
8884 		lsp->rls_seqid = olo->lock_seqid;
8885 
8886 		NFS4_DEBUG(rfs4_debug,
8887 		    (CE_NOTE, "Lock seqid established as %d", lsp->rls_seqid));
8888 
8889 		/*
8890 		 * This is used to signify the newly created lockowner
8891 		 * stateid and its sequence number.  The checks for
8892 		 * sequence number and increment don't occur on the
8893 		 * very first lock request for a lockowner.
8894 		 */
8895 		lsp->rls_skip_seqid_check = TRUE;
8896 
8897 		/* hold off other access to lsp while we tinker */
8898 		rfs4_sw_enter(&lsp->rls_sw);
8899 		ls_sw_held = TRUE;
8900 
8901 		rfs4_dbe_unlock(lsp->rls_dbe);
8902 
8903 		rfs4_lockowner_rele(lo);
8904 	} else {
8905 		stateid = &args->locker.locker4_u.lock_owner.lock_stateid;
8906 		/* get lsp and hold the lock on the underlying file struct */
8907 		if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE))
8908 		    != NFS4_OK) {
8909 			*cs->statusp = resp->status = status;
8910 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8911 			    cs, LOCK4res *, resp);
8912 			return;
8913 		}
8914 		create = FALSE;	/* We didn't create lsp */
8915 
8916 		/* Ensure specified filehandle matches */
8917 		if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
8918 			rfs4_lo_state_rele(lsp, TRUE);
8919 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8920 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8921 			    cs, LOCK4res *, resp);
8922 			return;
8923 		}
8924 
8925 		/* hold off other access to lsp while we tinker */
8926 		rfs4_sw_enter(&lsp->rls_sw);
8927 		ls_sw_held = TRUE;
8928 
8929 		switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
8930 		/*
8931 		 * The stateid looks like it was okay (expected to be
8932 		 * the next one)
8933 		 */
8934 		case NFS4_CHECK_STATEID_OKAY:
8935 			/*
8936 			 * The sequence id is now checked.  Determine
8937 			 * if this is a replay or if it is in the
8938 			 * expected (next) sequence.  In the case of a
8939 			 * replay, there are two replay conditions
8940 			 * that may occur.  The first is the normal
8941 			 * condition where a LOCK is done with a
8942 			 * NFS4_OK response and the stateid is
8943 			 * updated.  That case is handled below when
8944 			 * the stateid is identified as a REPLAY.  The
8945 			 * second is the case where an error is
8946 			 * returned, like NFS4ERR_DENIED, and the
8947 			 * sequence number is updated but the stateid
8948 			 * is not updated.  This second case is dealt
8949 			 * with here.  So it may seem odd that the
8950 			 * stateid is okay but the sequence id is a
8951 			 * replay but it is okay.
8952 			 */
8953 			switch (rfs4_check_lock_seqid(
8954 			    args->locker.locker4_u.lock_owner.lock_seqid,
8955 			    lsp, resop)) {
8956 			case NFS4_CHKSEQ_REPLAY:
8957 				if (resp->status != NFS4_OK) {
8958 					/*
8959 					 * Here is our replay and need
8960 					 * to verify that the last
8961 					 * response was an error.
8962 					 */
8963 					*cs->statusp = resp->status;
8964 					goto end;
8965 				}
8966 				/*
8967 				 * This is done since the sequence id
8968 				 * looked like a replay but it didn't
8969 				 * pass our check so a BAD_SEQID is
8970 				 * returned as a result.
8971 				 */
8972 				/*FALLTHROUGH*/
8973 			case NFS4_CHKSEQ_BAD:
8974 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8975 				goto end;
8976 			case NFS4_CHKSEQ_OKAY:
8977 				/* Everything looks okay move ahead */
8978 				break;
8979 			}
8980 			break;
8981 		case NFS4_CHECK_STATEID_OLD:
8982 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8983 			goto end;
8984 		case NFS4_CHECK_STATEID_BAD:
8985 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8986 			goto end;
8987 		case NFS4_CHECK_STATEID_EXPIRED:
8988 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8989 			goto end;
8990 		case NFS4_CHECK_STATEID_CLOSED:
8991 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8992 			goto end;
8993 		case NFS4_CHECK_STATEID_REPLAY:
8994 			switch (rfs4_check_lock_seqid(
8995 			    args->locker.locker4_u.lock_owner.lock_seqid,
8996 			    lsp, resop)) {
8997 			case NFS4_CHKSEQ_OKAY:
8998 				/*
8999 				 * This is a replayed stateid; if
9000 				 * seqid matches the next expected,
9001 				 * then client is using wrong seqid.
9002 				 */
9003 			case NFS4_CHKSEQ_BAD:
9004 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9005 				goto end;
9006 			case NFS4_CHKSEQ_REPLAY:
9007 				rfs4_update_lease(lsp->rls_locker->rl_client);
9008 				*cs->statusp = status = resp->status;
9009 				goto end;
9010 			}
9011 			break;
9012 		default:
9013 			ASSERT(FALSE);
9014 			break;
9015 		}
9016 
9017 		rfs4_update_lock_sequence(lsp);
9018 		rfs4_update_lease(lsp->rls_locker->rl_client);
9019 	}
9020 
9021 	/*
9022 	 * NFS4 only allows locking on regular files, so
9023 	 * verify type of object.
9024 	 */
9025 	if (cs->vp->v_type != VREG) {
9026 		if (cs->vp->v_type == VDIR)
9027 			status = NFS4ERR_ISDIR;
9028 		else
9029 			status = NFS4ERR_INVAL;
9030 		goto out;
9031 	}
9032 
9033 	cp = lsp->rls_state->rs_owner->ro_client;
9034 
9035 	if (rfs4_clnt_in_grace(cp) && !args->reclaim) {
9036 		status = NFS4ERR_GRACE;
9037 		goto out;
9038 	}
9039 
9040 	if (rfs4_clnt_in_grace(cp) && args->reclaim && !cp->rc_can_reclaim) {
9041 		status = NFS4ERR_NO_GRACE;
9042 		goto out;
9043 	}
9044 
9045 	if (!rfs4_clnt_in_grace(cp) && args->reclaim) {
9046 		status = NFS4ERR_NO_GRACE;
9047 		goto out;
9048 	}
9049 
9050 	if (lsp->rls_state->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE)
9051 		cs->deleg = TRUE;
9052 
9053 	status = rfs4_do_lock(lsp, args->locktype,
9054 	    args->offset, args->length, cs->cr, resop);
9055 
9056 out:
9057 	lsp->rls_skip_seqid_check = FALSE;
9058 
9059 	*cs->statusp = resp->status = status;
9060 
9061 	if (status == NFS4_OK) {
9062 		resp->LOCK4res_u.lock_stateid = lsp->rls_lockid.stateid;
9063 		lsp->rls_lock_completed = TRUE;
9064 	}
9065 	/*
9066 	 * Only update the "OPEN" response here if this was a new
9067 	 * lock_owner
9068 	 */
9069 	if (sp)
9070 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
9071 
9072 	rfs4_update_lock_resp(lsp, resop);
9073 
9074 end:
9075 	if (lsp) {
9076 		if (ls_sw_held)
9077 			rfs4_sw_exit(&lsp->rls_sw);
9078 		/*
9079 		 * If an sp obtained, then the lsp does not represent
9080 		 * a lock on the file struct.
9081 		 */
9082 		if (sp != NULL)
9083 			rfs4_lo_state_rele(lsp, FALSE);
9084 		else
9085 			rfs4_lo_state_rele(lsp, TRUE);
9086 	}
9087 	if (sp) {
9088 		rfs4_sw_exit(&sp->rs_owner->ro_sw);
9089 		rfs4_state_rele(sp);
9090 	}
9091 
9092 	DTRACE_NFSV4_2(op__lock__done, struct compound_state *, cs,
9093 	    LOCK4res *, resp);
9094 }
9095 
9096 /* free function for LOCK/LOCKT */
9097 static void
9098 lock_denied_free(nfs_resop4 *resop)
9099 {
9100 	LOCK4denied *dp = NULL;
9101 
9102 	switch (resop->resop) {
9103 	case OP_LOCK:
9104 		if (resop->nfs_resop4_u.oplock.status == NFS4ERR_DENIED)
9105 			dp = &resop->nfs_resop4_u.oplock.LOCK4res_u.denied;
9106 		break;
9107 	case OP_LOCKT:
9108 		if (resop->nfs_resop4_u.oplockt.status == NFS4ERR_DENIED)
9109 			dp = &resop->nfs_resop4_u.oplockt.denied;
9110 		break;
9111 	default:
9112 		break;
9113 	}
9114 
9115 	if (dp)
9116 		kmem_free(dp->owner.owner_val, dp->owner.owner_len);
9117 }
9118 
9119 /*ARGSUSED*/
9120 void
9121 rfs4_op_locku(nfs_argop4 *argop, nfs_resop4 *resop,
9122     struct svc_req *req, struct compound_state *cs)
9123 {
9124 	LOCKU4args *args = &argop->nfs_argop4_u.oplocku;
9125 	LOCKU4res *resp = &resop->nfs_resop4_u.oplocku;
9126 	nfsstat4 status;
9127 	stateid4 *stateid = &args->lock_stateid;
9128 	rfs4_lo_state_t *lsp;
9129 
9130 	DTRACE_NFSV4_2(op__locku__start, struct compound_state *, cs,
9131 	    LOCKU4args *, args);
9132 
9133 	if (cs->vp == NULL) {
9134 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9135 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9136 		    LOCKU4res *, resp);
9137 		return;
9138 	}
9139 
9140 	if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE)) != NFS4_OK) {
9141 		*cs->statusp = resp->status = status;
9142 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9143 		    LOCKU4res *, resp);
9144 		return;
9145 	}
9146 
9147 	/* Ensure specified filehandle matches */
9148 	if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9149 		rfs4_lo_state_rele(lsp, TRUE);
9150 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9151 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9152 		    LOCKU4res *, resp);
9153 		return;
9154 	}
9155 
9156 	/* hold off other access to lsp while we tinker */
9157 	rfs4_sw_enter(&lsp->rls_sw);
9158 
9159 	switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9160 	case NFS4_CHECK_STATEID_OKAY:
9161 		if (rfs4_check_lock_seqid(args->seqid, lsp, resop)
9162 		    != NFS4_CHKSEQ_OKAY) {
9163 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9164 			goto end;
9165 		}
9166 		break;
9167 	case NFS4_CHECK_STATEID_OLD:
9168 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9169 		goto end;
9170 	case NFS4_CHECK_STATEID_BAD:
9171 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9172 		goto end;
9173 	case NFS4_CHECK_STATEID_EXPIRED:
9174 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9175 		goto end;
9176 	case NFS4_CHECK_STATEID_CLOSED:
9177 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9178 		goto end;
9179 	case NFS4_CHECK_STATEID_REPLAY:
9180 		switch (rfs4_check_lock_seqid(args->seqid, lsp, resop)) {
9181 		case NFS4_CHKSEQ_OKAY:
9182 				/*
9183 				 * This is a replayed stateid; if
9184 				 * seqid matches the next expected,
9185 				 * then client is using wrong seqid.
9186 				 */
9187 		case NFS4_CHKSEQ_BAD:
9188 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9189 			goto end;
9190 		case NFS4_CHKSEQ_REPLAY:
9191 			rfs4_update_lease(lsp->rls_locker->rl_client);
9192 			*cs->statusp = status = resp->status;
9193 			goto end;
9194 		}
9195 		break;
9196 	default:
9197 		ASSERT(FALSE);
9198 		break;
9199 	}
9200 
9201 	rfs4_update_lock_sequence(lsp);
9202 	rfs4_update_lease(lsp->rls_locker->rl_client);
9203 
9204 	/*
9205 	 * NFS4 only allows locking on regular files, so
9206 	 * verify type of object.
9207 	 */
9208 	if (cs->vp->v_type != VREG) {
9209 		if (cs->vp->v_type == VDIR)
9210 			status = NFS4ERR_ISDIR;
9211 		else
9212 			status = NFS4ERR_INVAL;
9213 		goto out;
9214 	}
9215 
9216 	if (rfs4_clnt_in_grace(lsp->rls_state->rs_owner->ro_client)) {
9217 		status = NFS4ERR_GRACE;
9218 		goto out;
9219 	}
9220 
9221 	status = rfs4_do_lock(lsp, args->locktype,
9222 	    args->offset, args->length, cs->cr, resop);
9223 
9224 out:
9225 	*cs->statusp = resp->status = status;
9226 
9227 	if (status == NFS4_OK)
9228 		resp->lock_stateid = lsp->rls_lockid.stateid;
9229 
9230 	rfs4_update_lock_resp(lsp, resop);
9231 
9232 end:
9233 	rfs4_sw_exit(&lsp->rls_sw);
9234 	rfs4_lo_state_rele(lsp, TRUE);
9235 
9236 	DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9237 	    LOCKU4res *, resp);
9238 }
9239 
9240 /*
9241  * LOCKT is a best effort routine, the client can not be guaranteed that
9242  * the status return is still in effect by the time the reply is received.
9243  * They are numerous race conditions in this routine, but we are not required
9244  * and can not be accurate.
9245  */
9246 /*ARGSUSED*/
9247 void
9248 rfs4_op_lockt(nfs_argop4 *argop, nfs_resop4 *resop,
9249     struct svc_req *req, struct compound_state *cs)
9250 {
9251 	LOCKT4args *args = &argop->nfs_argop4_u.oplockt;
9252 	LOCKT4res *resp = &resop->nfs_resop4_u.oplockt;
9253 	rfs4_lockowner_t *lo;
9254 	rfs4_client_t *cp;
9255 	bool_t create = FALSE;
9256 	struct flock64 flk;
9257 	int error;
9258 	int flag = FREAD | FWRITE;
9259 	int ltype;
9260 	length4 posix_length;
9261 	sysid_t sysid;
9262 	pid_t pid;
9263 
9264 	DTRACE_NFSV4_2(op__lockt__start, struct compound_state *, cs,
9265 	    LOCKT4args *, args);
9266 
9267 	if (cs->vp == NULL) {
9268 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9269 		goto out;
9270 	}
9271 
9272 	/*
9273 	 * NFS4 only allows locking on regular files, so
9274 	 * verify type of object.
9275 	 */
9276 	if (cs->vp->v_type != VREG) {
9277 		if (cs->vp->v_type == VDIR)
9278 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
9279 		else
9280 			*cs->statusp = resp->status =  NFS4ERR_INVAL;
9281 		goto out;
9282 	}
9283 
9284 	/*
9285 	 * Check out the clientid to ensure the server knows about it
9286 	 * so that we correctly inform the client of a server reboot.
9287 	 */
9288 	if ((cp = rfs4_findclient_by_id(args->owner.clientid, FALSE))
9289 	    == NULL) {
9290 		*cs->statusp = resp->status =
9291 		    rfs4_check_clientid(&args->owner.clientid, 0);
9292 		goto out;
9293 	}
9294 	if (rfs4_lease_expired(cp)) {
9295 		rfs4_client_close(cp);
9296 		/*
9297 		 * Protocol doesn't allow returning NFS4ERR_STALE as
9298 		 * other operations do on this check so STALE_CLIENTID
9299 		 * is returned instead
9300 		 */
9301 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
9302 		goto out;
9303 	}
9304 
9305 	if (rfs4_clnt_in_grace(cp) && !(cp->rc_can_reclaim)) {
9306 		*cs->statusp = resp->status = NFS4ERR_GRACE;
9307 		rfs4_client_rele(cp);
9308 		goto out;
9309 	}
9310 	rfs4_client_rele(cp);
9311 
9312 	resp->status = NFS4_OK;
9313 
9314 	switch (args->locktype) {
9315 	case READ_LT:
9316 	case READW_LT:
9317 		ltype = F_RDLCK;
9318 		break;
9319 	case WRITE_LT:
9320 	case WRITEW_LT:
9321 		ltype = F_WRLCK;
9322 		break;
9323 	}
9324 
9325 	posix_length = args->length;
9326 	/* Check for zero length. To lock to end of file use all ones for V4 */
9327 	if (posix_length == 0) {
9328 		*cs->statusp = resp->status = NFS4ERR_INVAL;
9329 		goto out;
9330 	} else if (posix_length == (length4)(~0)) {
9331 		posix_length = 0;	/* Posix to end of file  */
9332 	}
9333 
9334 	/* Find or create a lockowner */
9335 	lo = rfs4_findlockowner(&args->owner, &create);
9336 
9337 	if (lo) {
9338 		pid = lo->rl_pid;
9339 		if ((resp->status =
9340 		    rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
9341 			goto err;
9342 	} else {
9343 		pid = 0;
9344 		sysid = lockt_sysid;
9345 	}
9346 retry:
9347 	flk.l_type = ltype;
9348 	flk.l_whence = 0;		/* SEEK_SET */
9349 	flk.l_start = args->offset;
9350 	flk.l_len = posix_length;
9351 	flk.l_sysid = sysid;
9352 	flk.l_pid = pid;
9353 	flag |= F_REMOTELOCK;
9354 
9355 	LOCK_PRINT(rfs4_debug, "rfs4_op_lockt", F_GETLK, &flk);
9356 
9357 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
9358 	if (flk.l_len < 0 || flk.l_start < 0) {
9359 		resp->status = NFS4ERR_INVAL;
9360 		goto err;
9361 	}
9362 	error = VOP_FRLOCK(cs->vp, F_GETLK, &flk, flag, (u_offset_t)0,
9363 	    NULL, cs->cr, NULL);
9364 
9365 	/*
9366 	 * N.B. We map error values to nfsv4 errors. This is differrent
9367 	 * than puterrno4 routine.
9368 	 */
9369 	switch (error) {
9370 	case 0:
9371 		if (flk.l_type == F_UNLCK)
9372 			resp->status = NFS4_OK;
9373 		else {
9374 			if (lock_denied(&resp->denied, &flk) == NFS4ERR_EXPIRED)
9375 				goto retry;
9376 			resp->status = NFS4ERR_DENIED;
9377 		}
9378 		break;
9379 	case EOVERFLOW:
9380 		resp->status = NFS4ERR_INVAL;
9381 		break;
9382 	case EINVAL:
9383 		resp->status = NFS4ERR_NOTSUPP;
9384 		break;
9385 	default:
9386 		cmn_err(CE_WARN, "rfs4_op_lockt: unexpected errno (%d)",
9387 		    error);
9388 		resp->status = NFS4ERR_SERVERFAULT;
9389 		break;
9390 	}
9391 
9392 err:
9393 	if (lo)
9394 		rfs4_lockowner_rele(lo);
9395 	*cs->statusp = resp->status;
9396 out:
9397 	DTRACE_NFSV4_2(op__lockt__done, struct compound_state *, cs,
9398 	    LOCKT4res *, resp);
9399 }
9400 
9401 int
9402 rfs4_share(rfs4_state_t *sp, uint32_t access, uint32_t deny)
9403 {
9404 	int err;
9405 	int cmd;
9406 	vnode_t *vp;
9407 	struct shrlock shr;
9408 	struct shr_locowner shr_loco;
9409 	int fflags = 0;
9410 
9411 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9412 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9413 
9414 	if (sp->rs_closed)
9415 		return (NFS4ERR_OLD_STATEID);
9416 
9417 	vp = sp->rs_finfo->rf_vp;
9418 	ASSERT(vp);
9419 
9420 	shr.s_access = shr.s_deny = 0;
9421 
9422 	if (access & OPEN4_SHARE_ACCESS_READ) {
9423 		fflags |= FREAD;
9424 		shr.s_access |= F_RDACC;
9425 	}
9426 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
9427 		fflags |= FWRITE;
9428 		shr.s_access |= F_WRACC;
9429 	}
9430 	ASSERT(shr.s_access);
9431 
9432 	if (deny & OPEN4_SHARE_DENY_READ)
9433 		shr.s_deny |= F_RDDNY;
9434 	if (deny & OPEN4_SHARE_DENY_WRITE)
9435 		shr.s_deny |= F_WRDNY;
9436 
9437 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9438 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9439 	shr_loco.sl_pid = shr.s_pid;
9440 	shr_loco.sl_id = shr.s_sysid;
9441 	shr.s_owner = (caddr_t)&shr_loco;
9442 	shr.s_own_len = sizeof (shr_loco);
9443 
9444 	cmd = nbl_need_check(vp) ? F_SHARE_NBMAND : F_SHARE;
9445 
9446 	err = VOP_SHRLOCK(vp, cmd, &shr, fflags, CRED(), NULL);
9447 	if (err != 0) {
9448 		if (err == EAGAIN)
9449 			err = NFS4ERR_SHARE_DENIED;
9450 		else
9451 			err = puterrno4(err);
9452 		return (err);
9453 	}
9454 
9455 	sp->rs_share_access |= access;
9456 	sp->rs_share_deny |= deny;
9457 
9458 	return (0);
9459 }
9460 
9461 int
9462 rfs4_unshare(rfs4_state_t *sp)
9463 {
9464 	int err;
9465 	struct shrlock shr;
9466 	struct shr_locowner shr_loco;
9467 
9468 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9469 
9470 	if (sp->rs_closed || sp->rs_share_access == 0)
9471 		return (0);
9472 
9473 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9474 	ASSERT(sp->rs_finfo->rf_vp);
9475 
9476 	shr.s_access = shr.s_deny = 0;
9477 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9478 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9479 	shr_loco.sl_pid = shr.s_pid;
9480 	shr_loco.sl_id = shr.s_sysid;
9481 	shr.s_owner = (caddr_t)&shr_loco;
9482 	shr.s_own_len = sizeof (shr_loco);
9483 
9484 	err = VOP_SHRLOCK(sp->rs_finfo->rf_vp, F_UNSHARE, &shr, 0, CRED(),
9485 	    NULL);
9486 	if (err != 0) {
9487 		err = puterrno4(err);
9488 		return (err);
9489 	}
9490 
9491 	sp->rs_share_access = 0;
9492 	sp->rs_share_deny = 0;
9493 
9494 	return (0);
9495 
9496 }
9497 
9498 static int
9499 rdma_setup_read_data4(READ4args *args, READ4res *rok)
9500 {
9501 	struct clist	*wcl;
9502 	count4		count = rok->data_len;
9503 	int		wlist_len;
9504 
9505 	wcl = args->wlist;
9506 	if (rdma_setup_read_chunks(wcl, count, &wlist_len) == FALSE) {
9507 		return (FALSE);
9508 	}
9509 	wcl = args->wlist;
9510 	rok->wlist_len = wlist_len;
9511 	rok->wlist = wcl;
9512 	return (TRUE);
9513 }
9514 
9515 /* tunable to disable server referrals */
9516 int rfs4_no_referrals = 0;
9517 
9518 /*
9519  * Find an NFS record in reparse point data.
9520  * Returns 0 for success and <0 or an errno value on failure.
9521  */
9522 int
9523 vn_find_nfs_record(vnode_t *vp, nvlist_t **nvlp, char **svcp, char **datap)
9524 {
9525 	int err;
9526 	char *stype, *val;
9527 	nvlist_t *nvl;
9528 	nvpair_t *curr;
9529 
9530 	if ((nvl = reparse_init()) == NULL)
9531 		return (-1);
9532 
9533 	if ((err = reparse_vnode_parse(vp, nvl)) != 0) {
9534 		reparse_free(nvl);
9535 		return (err);
9536 	}
9537 
9538 	curr = NULL;
9539 	while ((curr = nvlist_next_nvpair(nvl, curr)) != NULL) {
9540 		if ((stype = nvpair_name(curr)) == NULL) {
9541 			reparse_free(nvl);
9542 			return (-2);
9543 		}
9544 		if (strncasecmp(stype, "NFS", 3) == 0)
9545 			break;
9546 	}
9547 
9548 	if ((curr == NULL) ||
9549 	    (nvpair_value_string(curr, &val))) {
9550 		reparse_free(nvl);
9551 		return (-3);
9552 	}
9553 	*nvlp = nvl;
9554 	*svcp = stype;
9555 	*datap = val;
9556 	return (0);
9557 }
9558 
9559 int
9560 vn_is_nfs_reparse(vnode_t *vp, cred_t *cr)
9561 {
9562 	nvlist_t *nvl;
9563 	char *s, *d;
9564 
9565 	if (rfs4_no_referrals != 0)
9566 		return (B_FALSE);
9567 
9568 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9569 		return (B_FALSE);
9570 
9571 	if (vn_find_nfs_record(vp, &nvl, &s, &d) != 0)
9572 		return (B_FALSE);
9573 
9574 	reparse_free(nvl);
9575 
9576 	return (B_TRUE);
9577 }
9578 
9579 /*
9580  * There is a user-level copy of this routine in ref_subr.c.
9581  * Changes should be kept in sync.
9582  */
9583 static int
9584 nfs4_create_components(char *path, component4 *comp4)
9585 {
9586 	int slen, plen, ncomp;
9587 	char *ori_path, *nxtc, buf[MAXNAMELEN];
9588 
9589 	if (path == NULL)
9590 		return (0);
9591 
9592 	plen = strlen(path) + 1;	/* include the terminator */
9593 	ori_path = path;
9594 	ncomp = 0;
9595 
9596 	/* count number of components in the path */
9597 	for (nxtc = path; nxtc < ori_path + plen; nxtc++) {
9598 		if (*nxtc == '/' || *nxtc == '\0' || *nxtc == '\n') {
9599 			if ((slen = nxtc - path) == 0) {
9600 				path = nxtc + 1;
9601 				continue;
9602 			}
9603 
9604 			if (comp4 != NULL) {
9605 				bcopy(path, buf, slen);
9606 				buf[slen] = '\0';
9607 				(void) str_to_utf8(buf, &comp4[ncomp]);
9608 			}
9609 
9610 			ncomp++;	/* 1 valid component */
9611 			path = nxtc + 1;
9612 		}
9613 		if (*nxtc == '\0' || *nxtc == '\n')
9614 			break;
9615 	}
9616 
9617 	return (ncomp);
9618 }
9619 
9620 /*
9621  * There is a user-level copy of this routine in ref_subr.c.
9622  * Changes should be kept in sync.
9623  */
9624 static int
9625 make_pathname4(char *path, pathname4 *pathname)
9626 {
9627 	int ncomp;
9628 	component4 *comp4;
9629 
9630 	if (pathname == NULL)
9631 		return (0);
9632 
9633 	if (path == NULL) {
9634 		pathname->pathname4_val = NULL;
9635 		pathname->pathname4_len = 0;
9636 		return (0);
9637 	}
9638 
9639 	/* count number of components to alloc buffer */
9640 	if ((ncomp = nfs4_create_components(path, NULL)) == 0) {
9641 		pathname->pathname4_val = NULL;
9642 		pathname->pathname4_len = 0;
9643 		return (0);
9644 	}
9645 	comp4 = kmem_zalloc(ncomp * sizeof (component4), KM_SLEEP);
9646 
9647 	/* copy components into allocated buffer */
9648 	ncomp = nfs4_create_components(path, comp4);
9649 
9650 	pathname->pathname4_val = comp4;
9651 	pathname->pathname4_len = ncomp;
9652 
9653 	return (ncomp);
9654 }
9655 
9656 #define	xdr_fs_locations4 xdr_fattr4_fs_locations
9657 
9658 fs_locations4 *
9659 fetch_referral(vnode_t *vp, cred_t *cr)
9660 {
9661 	nvlist_t *nvl;
9662 	char *stype, *sdata;
9663 	fs_locations4 *result;
9664 	char buf[1024];
9665 	size_t bufsize;
9666 	XDR xdr;
9667 	int err;
9668 
9669 	/*
9670 	 * Check attrs to ensure it's a reparse point
9671 	 */
9672 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9673 		return (NULL);
9674 
9675 	/*
9676 	 * Look for an NFS record and get the type and data
9677 	 */
9678 	if (vn_find_nfs_record(vp, &nvl, &stype, &sdata) != 0)
9679 		return (NULL);
9680 
9681 	/*
9682 	 * With the type and data, upcall to get the referral
9683 	 */
9684 	bufsize = sizeof (buf);
9685 	bzero(buf, sizeof (buf));
9686 	err = reparse_kderef((const char *)stype, (const char *)sdata,
9687 	    buf, &bufsize);
9688 	reparse_free(nvl);
9689 
9690 	DTRACE_PROBE4(nfs4serv__func__referral__upcall,
9691 	    char *, stype, char *, sdata, char *, buf, int, err);
9692 	if (err) {
9693 		cmn_err(CE_NOTE,
9694 		    "reparsed daemon not running: unable to get referral (%d)",
9695 		    err);
9696 		return (NULL);
9697 	}
9698 
9699 	/*
9700 	 * We get an XDR'ed record back from the kderef call
9701 	 */
9702 	xdrmem_create(&xdr, buf, bufsize, XDR_DECODE);
9703 	result = kmem_alloc(sizeof (fs_locations4), KM_SLEEP);
9704 	err = xdr_fs_locations4(&xdr, result);
9705 	XDR_DESTROY(&xdr);
9706 	if (err != TRUE) {
9707 		DTRACE_PROBE1(nfs4serv__func__referral__upcall__xdrfail,
9708 		    int, err);
9709 		return (NULL);
9710 	}
9711 
9712 	/*
9713 	 * Look at path to recover fs_root, ignoring the leading '/'
9714 	 */
9715 	(void) make_pathname4(vp->v_path, &result->fs_root);
9716 
9717 	return (result);
9718 }
9719 
9720 char *
9721 build_symlink(vnode_t *vp, cred_t *cr, size_t *strsz)
9722 {
9723 	fs_locations4 *fsl;
9724 	fs_location4 *fs;
9725 	char *server, *path, *symbuf;
9726 	static char *prefix = "/net/";
9727 	int i, size, npaths;
9728 	uint_t len;
9729 
9730 	/* Get the referral */
9731 	if ((fsl = fetch_referral(vp, cr)) == NULL)
9732 		return (NULL);
9733 
9734 	/* Deal with only the first location and first server */
9735 	fs = &fsl->locations_val[0];
9736 	server = utf8_to_str(&fs->server_val[0], &len, NULL);
9737 	if (server == NULL) {
9738 		rfs4_free_fs_locations4(fsl);
9739 		kmem_free(fsl, sizeof (fs_locations4));
9740 		return (NULL);
9741 	}
9742 
9743 	/* Figure out size for "/net/" + host + /path/path/path + NULL */
9744 	size = strlen(prefix) + len;
9745 	for (i = 0; i < fs->rootpath.pathname4_len; i++)
9746 		size += fs->rootpath.pathname4_val[i].utf8string_len + 1;
9747 
9748 	/* Allocate the symlink buffer and fill it */
9749 	symbuf = kmem_zalloc(size, KM_SLEEP);
9750 	(void) strcat(symbuf, prefix);
9751 	(void) strcat(symbuf, server);
9752 	kmem_free(server, len);
9753 
9754 	npaths = 0;
9755 	for (i = 0; i < fs->rootpath.pathname4_len; i++) {
9756 		path = utf8_to_str(&fs->rootpath.pathname4_val[i], &len, NULL);
9757 		if (path == NULL)
9758 			continue;
9759 		(void) strcat(symbuf, "/");
9760 		(void) strcat(symbuf, path);
9761 		npaths++;
9762 		kmem_free(path, len);
9763 	}
9764 
9765 	rfs4_free_fs_locations4(fsl);
9766 	kmem_free(fsl, sizeof (fs_locations4));
9767 
9768 	if (strsz != NULL)
9769 		*strsz = size;
9770 	return (symbuf);
9771 }
9772 
9773 /*
9774  * Check to see if we have a downrev Solaris client, so that we
9775  * can send it a symlink instead of a referral.
9776  */
9777 int
9778 client_is_downrev(struct svc_req *req)
9779 {
9780 	struct sockaddr *ca;
9781 	rfs4_clntip_t *ci;
9782 	bool_t create = FALSE;
9783 	int is_downrev;
9784 
9785 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
9786 	ASSERT(ca);
9787 	ci = rfs4_find_clntip(ca, &create);
9788 	if (ci == NULL)
9789 		return (0);
9790 	is_downrev = ci->ri_no_referrals;
9791 	rfs4_dbe_rele(ci->ri_dbe);
9792 	return (is_downrev);
9793 }
9794