xref: /titanic_51/usr/src/uts/common/fs/nfs/nfs4_srv.c (revision dd51520e127b452179a2ce4ea3bd8dee949f9afe)
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 		if (args->count > clist_len(args->wlist)) {
3187 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3188 			goto out;
3189 		}
3190 		rdma_used = 1;
3191 	}
3192 
3193 	/* use loaned buffers for TCP */
3194 	loaned_buffers = (nfs_loaned_buffers && !rdma_used) ? 1 : 0;
3195 
3196 	va.va_mask = AT_MODE|AT_SIZE|AT_UID;
3197 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3198 
3199 	/*
3200 	 * If we can't get the attributes, then we can't do the
3201 	 * right access checking.  So, we'll fail the request.
3202 	 */
3203 	if (verror) {
3204 		*cs->statusp = resp->status = puterrno4(verror);
3205 		goto out;
3206 	}
3207 
3208 	if (vp->v_type != VREG) {
3209 		*cs->statusp = resp->status =
3210 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
3211 		goto out;
3212 	}
3213 
3214 	if (crgetuid(cs->cr) != va.va_uid &&
3215 	    (error = VOP_ACCESS(vp, VREAD, 0, cs->cr, &ct)) &&
3216 	    (error = VOP_ACCESS(vp, VEXEC, 0, cs->cr, &ct))) {
3217 		*cs->statusp = resp->status = puterrno4(error);
3218 		goto out;
3219 	}
3220 
3221 	if (MANDLOCK(vp, va.va_mode)) { /* XXX - V4 supports mand locking */
3222 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3223 		goto out;
3224 	}
3225 
3226 	offset = args->offset;
3227 	if (offset >= va.va_size) {
3228 		*cs->statusp = resp->status = NFS4_OK;
3229 		resp->eof = TRUE;
3230 		resp->data_len = 0;
3231 		resp->data_val = NULL;
3232 		resp->mblk = NULL;
3233 		/* RDMA */
3234 		resp->wlist = args->wlist;
3235 		resp->wlist_len = resp->data_len;
3236 		*cs->statusp = resp->status = NFS4_OK;
3237 		if (resp->wlist)
3238 			clist_zero_len(resp->wlist);
3239 		goto out;
3240 	}
3241 
3242 	if (args->count == 0) {
3243 		*cs->statusp = resp->status = NFS4_OK;
3244 		resp->eof = FALSE;
3245 		resp->data_len = 0;
3246 		resp->data_val = NULL;
3247 		resp->mblk = NULL;
3248 		/* RDMA */
3249 		resp->wlist = args->wlist;
3250 		resp->wlist_len = resp->data_len;
3251 		if (resp->wlist)
3252 			clist_zero_len(resp->wlist);
3253 		goto out;
3254 	}
3255 
3256 	/*
3257 	 * Do not allocate memory more than maximum allowed
3258 	 * transfer size
3259 	 */
3260 	if (args->count > rfs4_tsize(req))
3261 		args->count = rfs4_tsize(req);
3262 
3263 	if (loaned_buffers) {
3264 		uiop = (uio_t *)rfs_setup_xuio(vp);
3265 		ASSERT(uiop != NULL);
3266 		uiop->uio_segflg = UIO_SYSSPACE;
3267 		uiop->uio_loffset = args->offset;
3268 		uiop->uio_resid = args->count;
3269 
3270 		/* Jump to do the read if successful */
3271 		if (!VOP_REQZCBUF(vp, UIO_READ, (xuio_t *)uiop, cs->cr, &ct)) {
3272 			/*
3273 			 * Need to hold the vnode until after VOP_RETZCBUF()
3274 			 * is called.
3275 			 */
3276 			VN_HOLD(vp);
3277 			goto doio_read;
3278 		}
3279 
3280 		DTRACE_PROBE2(nfss__i__reqzcbuf_failed, int,
3281 		    uiop->uio_loffset, int, uiop->uio_resid);
3282 
3283 		uiop->uio_extflg = 0;
3284 
3285 		/* failure to setup for zero copy */
3286 		rfs_free_xuio((void *)uiop);
3287 		loaned_buffers = 0;
3288 	}
3289 
3290 	/*
3291 	 * If returning data via RDMA Write, then grab the chunk list. If we
3292 	 * aren't returning READ data w/RDMA_WRITE, then grab a mblk.
3293 	 */
3294 	if (rdma_used) {
3295 		mp = NULL;
3296 		(void) rdma_get_wchunk(req, &iov, args->wlist);
3297 	} else {
3298 		/*
3299 		 * mp will contain the data to be sent out in the read reply.
3300 		 * It will be freed after the reply has been sent. Let's
3301 		 * roundup the data to a BYTES_PER_XDR_UNIT multiple, so that
3302 		 * the call to xdrmblk_putmblk() never fails. If the first
3303 		 * alloc of the requested size fails, then decrease the size to
3304 		 * something more reasonable and wait for the allocation to
3305 		 * occur.
3306 		 */
3307 		mp = allocb(RNDUP(args->count), BPRI_MED);
3308 		if (mp == NULL) {
3309 			if (args->count > MAXBSIZE)
3310 				args->count = MAXBSIZE;
3311 			mp = allocb_wait(RNDUP(args->count), BPRI_MED,
3312 			    STR_NOSIG, &alloc_err);
3313 		}
3314 		ASSERT(mp != NULL);
3315 		ASSERT(alloc_err == 0);
3316 
3317 		iov.iov_base = (caddr_t)mp->b_datap->db_base;
3318 		iov.iov_len = args->count;
3319 	}
3320 
3321 	uio.uio_iov = &iov;
3322 	uio.uio_iovcnt = 1;
3323 	uio.uio_segflg = UIO_SYSSPACE;
3324 	uio.uio_extflg = UIO_COPY_CACHED;
3325 	uio.uio_loffset = args->offset;
3326 	uio.uio_resid = args->count;
3327 	uiop = &uio;
3328 
3329 doio_read:
3330 	error = do_io(FREAD, vp, uiop, 0, cs->cr, &ct);
3331 
3332 	va.va_mask = AT_SIZE;
3333 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3334 
3335 	if (error) {
3336 		if (mp)
3337 			freemsg(mp);
3338 		*cs->statusp = resp->status = puterrno4(error);
3339 		goto out;
3340 	}
3341 
3342 	/* make mblk using zc buffers */
3343 	if (loaned_buffers) {
3344 		mp = uio_to_mblk(uiop);
3345 		ASSERT(mp != NULL);
3346 	}
3347 
3348 	*cs->statusp = resp->status = NFS4_OK;
3349 
3350 	ASSERT(uiop->uio_resid >= 0);
3351 	resp->data_len = args->count - uiop->uio_resid;
3352 	if (mp) {
3353 		resp->data_val = (char *)mp->b_datap->db_base;
3354 		rfs_rndup_mblks(mp, resp->data_len, loaned_buffers);
3355 	} else {
3356 		resp->data_val = (caddr_t)iov.iov_base;
3357 	}
3358 
3359 	resp->mblk = mp;
3360 
3361 	if (!verror && offset + resp->data_len == va.va_size)
3362 		resp->eof = TRUE;
3363 	else
3364 		resp->eof = FALSE;
3365 
3366 	if (rdma_used) {
3367 		if (!rdma_setup_read_data4(args, resp)) {
3368 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3369 		}
3370 	} else {
3371 		resp->wlist = NULL;
3372 	}
3373 
3374 out:
3375 	if (in_crit)
3376 		nbl_end_crit(vp);
3377 
3378 	DTRACE_NFSV4_2(op__read__done, struct compound_state *, cs,
3379 	    READ4res *, resp);
3380 }
3381 
3382 static void
3383 rfs4_op_read_free(nfs_resop4 *resop)
3384 {
3385 	READ4res	*resp = &resop->nfs_resop4_u.opread;
3386 
3387 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3388 		freemsg(resp->mblk);
3389 		resp->mblk = NULL;
3390 		resp->data_val = NULL;
3391 		resp->data_len = 0;
3392 	}
3393 }
3394 
3395 static void
3396 rfs4_op_readdir_free(nfs_resop4 * resop)
3397 {
3398 	READDIR4res    *resp = &resop->nfs_resop4_u.opreaddir;
3399 
3400 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3401 		freeb(resp->mblk);
3402 		resp->mblk = NULL;
3403 		resp->data_len = 0;
3404 	}
3405 }
3406 
3407 
3408 /* ARGSUSED */
3409 static void
3410 rfs4_op_putpubfh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
3411     struct compound_state *cs)
3412 {
3413 	PUTPUBFH4res	*resp = &resop->nfs_resop4_u.opputpubfh;
3414 	int		error;
3415 	vnode_t		*vp;
3416 	struct exportinfo *exi, *sav_exi;
3417 	nfs_fh4_fmt_t	*fh_fmtp;
3418 
3419 	DTRACE_NFSV4_1(op__putpubfh__start, struct compound_state *, cs);
3420 
3421 	if (cs->vp) {
3422 		VN_RELE(cs->vp);
3423 		cs->vp = NULL;
3424 	}
3425 
3426 	if (cs->cr)
3427 		crfree(cs->cr);
3428 
3429 	cs->cr = crdup(cs->basecr);
3430 
3431 	vp = exi_public->exi_vp;
3432 	if (vp == NULL) {
3433 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3434 		goto out;
3435 	}
3436 
3437 	error = makefh4(&cs->fh, vp, exi_public);
3438 	if (error != 0) {
3439 		*cs->statusp = resp->status = puterrno4(error);
3440 		goto out;
3441 	}
3442 	sav_exi = cs->exi;
3443 	if (exi_public == exi_root) {
3444 		/*
3445 		 * No filesystem is actually shared public, so we default
3446 		 * to exi_root. In this case, we must check whether root
3447 		 * is exported.
3448 		 */
3449 		fh_fmtp = (nfs_fh4_fmt_t *)cs->fh.nfs_fh4_val;
3450 
3451 		/*
3452 		 * if root filesystem is exported, the exportinfo struct that we
3453 		 * should use is what checkexport4 returns, because root_exi is
3454 		 * actually a mostly empty struct.
3455 		 */
3456 		exi = checkexport4(&fh_fmtp->fh4_fsid,
3457 		    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
3458 		cs->exi = ((exi != NULL) ? exi : exi_public);
3459 	} else {
3460 		/*
3461 		 * it's a properly shared filesystem
3462 		 */
3463 		cs->exi = exi_public;
3464 	}
3465 
3466 	if (is_system_labeled()) {
3467 		bslabel_t *clabel;
3468 
3469 		ASSERT(req->rq_label != NULL);
3470 		clabel = req->rq_label;
3471 		DTRACE_PROBE2(tx__rfs4__log__info__opputpubfh__clabel, char *,
3472 		    "got client label from request(1)",
3473 		    struct svc_req *, req);
3474 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
3475 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
3476 			    cs->exi)) {
3477 				*cs->statusp = resp->status =
3478 				    NFS4ERR_SERVERFAULT;
3479 				goto out;
3480 			}
3481 		}
3482 	}
3483 
3484 	VN_HOLD(vp);
3485 	cs->vp = vp;
3486 
3487 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3488 		VN_RELE(cs->vp);
3489 		cs->vp = NULL;
3490 		cs->exi = sav_exi;
3491 		goto out;
3492 	}
3493 
3494 	*cs->statusp = resp->status = NFS4_OK;
3495 out:
3496 	DTRACE_NFSV4_2(op__putpubfh__done, struct compound_state *, cs,
3497 	    PUTPUBFH4res *, resp);
3498 }
3499 
3500 /*
3501  * XXX - issue with put*fh operations. Suppose /export/home is exported.
3502  * Suppose an NFS client goes to mount /export/home/joe. If /export, home,
3503  * or joe have restrictive search permissions, then we shouldn't let
3504  * the client get a file handle. This is easy to enforce. However, we
3505  * don't know what security flavor should be used until we resolve the
3506  * path name. Another complication is uid mapping. If root is
3507  * the user, then it will be mapped to the anonymous user by default,
3508  * but we won't know that till we've resolved the path name. And we won't
3509  * know what the anonymous user is.
3510  * Luckily, SECINFO is specified to take a full filename.
3511  * So what we will have to in rfs4_op_lookup is check that flavor of
3512  * the target object matches that of the request, and if root was the
3513  * caller, check for the root= and anon= options, and if necessary,
3514  * repeat the lookup using the right cred_t. But that's not done yet.
3515  */
3516 /* ARGSUSED */
3517 static void
3518 rfs4_op_putfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3519     struct compound_state *cs)
3520 {
3521 	PUTFH4args *args = &argop->nfs_argop4_u.opputfh;
3522 	PUTFH4res *resp = &resop->nfs_resop4_u.opputfh;
3523 	nfs_fh4_fmt_t *fh_fmtp;
3524 
3525 	DTRACE_NFSV4_2(op__putfh__start, struct compound_state *, cs,
3526 	    PUTFH4args *, args);
3527 
3528 	if (cs->vp) {
3529 		VN_RELE(cs->vp);
3530 		cs->vp = NULL;
3531 	}
3532 
3533 	if (cs->cr) {
3534 		crfree(cs->cr);
3535 		cs->cr = NULL;
3536 	}
3537 
3538 
3539 	if (args->object.nfs_fh4_len < NFS_FH4_LEN) {
3540 		*cs->statusp = resp->status = NFS4ERR_BADHANDLE;
3541 		goto out;
3542 	}
3543 
3544 	fh_fmtp = (nfs_fh4_fmt_t *)args->object.nfs_fh4_val;
3545 	cs->exi = checkexport4(&fh_fmtp->fh4_fsid, (fid_t *)&fh_fmtp->fh4_xlen,
3546 	    NULL);
3547 
3548 	if (cs->exi == NULL) {
3549 		*cs->statusp = resp->status = NFS4ERR_STALE;
3550 		goto out;
3551 	}
3552 
3553 	cs->cr = crdup(cs->basecr);
3554 
3555 	ASSERT(cs->cr != NULL);
3556 
3557 	if (! (cs->vp = nfs4_fhtovp(&args->object, cs->exi, &resp->status))) {
3558 		*cs->statusp = resp->status;
3559 		goto out;
3560 	}
3561 
3562 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3563 		VN_RELE(cs->vp);
3564 		cs->vp = NULL;
3565 		goto out;
3566 	}
3567 
3568 	nfs_fh4_copy(&args->object, &cs->fh);
3569 	*cs->statusp = resp->status = NFS4_OK;
3570 	cs->deleg = FALSE;
3571 
3572 out:
3573 	DTRACE_NFSV4_2(op__putfh__done, struct compound_state *, cs,
3574 	    PUTFH4res *, resp);
3575 }
3576 
3577 /* ARGSUSED */
3578 static void
3579 rfs4_op_putrootfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3580     struct compound_state *cs)
3581 {
3582 	PUTROOTFH4res *resp = &resop->nfs_resop4_u.opputrootfh;
3583 	int error;
3584 	fid_t fid;
3585 	struct exportinfo *exi, *sav_exi;
3586 
3587 	DTRACE_NFSV4_1(op__putrootfh__start, struct compound_state *, cs);
3588 
3589 	if (cs->vp) {
3590 		VN_RELE(cs->vp);
3591 		cs->vp = NULL;
3592 	}
3593 
3594 	if (cs->cr)
3595 		crfree(cs->cr);
3596 
3597 	cs->cr = crdup(cs->basecr);
3598 
3599 	/*
3600 	 * Using rootdir, the system root vnode,
3601 	 * get its fid.
3602 	 */
3603 	bzero(&fid, sizeof (fid));
3604 	fid.fid_len = MAXFIDSZ;
3605 	error = vop_fid_pseudo(rootdir, &fid);
3606 	if (error != 0) {
3607 		*cs->statusp = resp->status = puterrno4(error);
3608 		goto out;
3609 	}
3610 
3611 	/*
3612 	 * Then use the root fsid & fid it to find out if it's exported
3613 	 *
3614 	 * If the server root isn't exported directly, then
3615 	 * it should at least be a pseudo export based on
3616 	 * one or more exports further down in the server's
3617 	 * file tree.
3618 	 */
3619 	exi = checkexport4(&rootdir->v_vfsp->vfs_fsid, &fid, NULL);
3620 	if (exi == NULL || exi->exi_export.ex_flags & EX_PUBLIC) {
3621 		NFS4_DEBUG(rfs4_debug,
3622 		    (CE_WARN, "rfs4_op_putrootfh: export check failure"));
3623 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3624 		goto out;
3625 	}
3626 
3627 	/*
3628 	 * Now make a filehandle based on the root
3629 	 * export and root vnode.
3630 	 */
3631 	error = makefh4(&cs->fh, rootdir, exi);
3632 	if (error != 0) {
3633 		*cs->statusp = resp->status = puterrno4(error);
3634 		goto out;
3635 	}
3636 
3637 	sav_exi = cs->exi;
3638 	cs->exi = exi;
3639 
3640 	VN_HOLD(rootdir);
3641 	cs->vp = rootdir;
3642 
3643 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3644 		VN_RELE(rootdir);
3645 		cs->vp = NULL;
3646 		cs->exi = sav_exi;
3647 		goto out;
3648 	}
3649 
3650 	*cs->statusp = resp->status = NFS4_OK;
3651 	cs->deleg = FALSE;
3652 out:
3653 	DTRACE_NFSV4_2(op__putrootfh__done, struct compound_state *, cs,
3654 	    PUTROOTFH4res *, resp);
3655 }
3656 
3657 /*
3658  * A directory entry is a valid nfsv4 entry if
3659  * - it has a non-zero ino
3660  * - it is not a dot or dotdot name
3661  * - it is visible in a pseudo export or in a real export that can
3662  *   only have a limited view.
3663  */
3664 static bool_t
3665 valid_nfs4_entry(struct exportinfo *exi, struct dirent64 *dp,
3666     int *expseudo, int check_visible)
3667 {
3668 	if (dp->d_ino == 0 || NFS_IS_DOTNAME(dp->d_name)) {
3669 		*expseudo = 0;
3670 		return (FALSE);
3671 	}
3672 
3673 	if (! check_visible) {
3674 		*expseudo = 0;
3675 		return (TRUE);
3676 	}
3677 
3678 	return (nfs_visible_inode(exi, dp->d_ino, expseudo));
3679 }
3680 
3681 /*
3682  * set_rdattr_params sets up the variables used to manage what information
3683  * to get for each directory entry.
3684  */
3685 static nfsstat4
3686 set_rdattr_params(struct nfs4_svgetit_arg *sargp,
3687     bitmap4 attrs, bool_t *need_to_lookup)
3688 {
3689 	uint_t	va_mask;
3690 	nfsstat4 status;
3691 	bitmap4 objbits;
3692 
3693 	status = bitmap4_to_attrmask(attrs, sargp);
3694 	if (status != NFS4_OK) {
3695 		/*
3696 		 * could not even figure attr mask
3697 		 */
3698 		return (status);
3699 	}
3700 	va_mask = sargp->vap->va_mask;
3701 
3702 	/*
3703 	 * dirent's d_ino is always correct value for mounted_on_fileid.
3704 	 * mntdfid_set is set once here, but mounted_on_fileid is
3705 	 * set in main dirent processing loop for each dirent.
3706 	 * The mntdfid_set is a simple optimization that lets the
3707 	 * server attr code avoid work when caller is readdir.
3708 	 */
3709 	sargp->mntdfid_set = TRUE;
3710 
3711 	/*
3712 	 * Lookup entry only if client asked for any of the following:
3713 	 * a) vattr attrs
3714 	 * b) vfs attrs
3715 	 * c) attrs w/per-object scope requested (change, filehandle, etc)
3716 	 *    other than mounted_on_fileid (which we can take from dirent)
3717 	 */
3718 	objbits = attrs ? attrs & NFS4_VP_ATTR_MASK : 0;
3719 
3720 	if (va_mask || sargp->sbp || (objbits & ~FATTR4_MOUNTED_ON_FILEID_MASK))
3721 		*need_to_lookup = TRUE;
3722 	else
3723 		*need_to_lookup = FALSE;
3724 
3725 	if (sargp->sbp == NULL)
3726 		return (NFS4_OK);
3727 
3728 	/*
3729 	 * If filesystem attrs are requested, get them now from the
3730 	 * directory vp, as most entries will have same filesystem. The only
3731 	 * exception are mounted over entries but we handle
3732 	 * those as we go (XXX mounted over detection not yet implemented).
3733 	 */
3734 	sargp->vap->va_mask = 0;	/* to avoid VOP_GETATTR */
3735 	status = bitmap4_get_sysattrs(sargp);
3736 	sargp->vap->va_mask = va_mask;
3737 
3738 	if ((status != NFS4_OK) && sargp->rdattr_error_req) {
3739 		/*
3740 		 * Failed to get filesystem attributes.
3741 		 * Return a rdattr_error for each entry, but don't fail.
3742 		 * However, don't get any obj-dependent attrs.
3743 		 */
3744 		sargp->rdattr_error = status;	/* for rdattr_error */
3745 		*need_to_lookup = FALSE;
3746 		/*
3747 		 * At least get fileid for regular readdir output
3748 		 */
3749 		sargp->vap->va_mask &= AT_NODEID;
3750 		status = NFS4_OK;
3751 	}
3752 
3753 	return (status);
3754 }
3755 
3756 /*
3757  * readlink: args: CURRENT_FH.
3758  *	res: status. If success - CURRENT_FH unchanged, return linktext.
3759  */
3760 
3761 /* ARGSUSED */
3762 static void
3763 rfs4_op_readlink(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3764     struct compound_state *cs)
3765 {
3766 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3767 	int error;
3768 	vnode_t *vp;
3769 	struct iovec iov;
3770 	struct vattr va;
3771 	struct uio uio;
3772 	char *data;
3773 	struct sockaddr *ca;
3774 	char *name = NULL;
3775 	int is_referral;
3776 
3777 	DTRACE_NFSV4_1(op__readlink__start, struct compound_state *, cs);
3778 
3779 	/* CURRENT_FH: directory */
3780 	vp = cs->vp;
3781 	if (vp == NULL) {
3782 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3783 		goto out;
3784 	}
3785 
3786 	if (cs->access == CS_ACCESS_DENIED) {
3787 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3788 		goto out;
3789 	}
3790 
3791 	/* Is it a referral? */
3792 	if (vn_is_nfs_reparse(vp, cs->cr) && client_is_downrev(req)) {
3793 
3794 		is_referral = 1;
3795 
3796 	} else {
3797 
3798 		is_referral = 0;
3799 
3800 		if (vp->v_type == VDIR) {
3801 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
3802 			goto out;
3803 		}
3804 
3805 		if (vp->v_type != VLNK) {
3806 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3807 			goto out;
3808 		}
3809 
3810 	}
3811 
3812 	va.va_mask = AT_MODE;
3813 	error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
3814 	if (error) {
3815 		*cs->statusp = resp->status = puterrno4(error);
3816 		goto out;
3817 	}
3818 
3819 	if (MANDLOCK(vp, va.va_mode)) {
3820 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3821 		goto out;
3822 	}
3823 
3824 	data = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);
3825 
3826 	if (is_referral) {
3827 		char *s;
3828 		size_t strsz;
3829 
3830 		/* Get an artificial symlink based on a referral */
3831 		s = build_symlink(vp, cs->cr, &strsz);
3832 		global_svstat_ptr[4][NFS_REFERLINKS].value.ui64++;
3833 		DTRACE_PROBE2(nfs4serv__func__referral__reflink,
3834 		    vnode_t *, vp, char *, s);
3835 		if (s == NULL)
3836 			error = EINVAL;
3837 		else {
3838 			error = 0;
3839 			(void) strlcpy(data, s, MAXPATHLEN + 1);
3840 			kmem_free(s, strsz);
3841 		}
3842 
3843 	} else {
3844 
3845 		iov.iov_base = data;
3846 		iov.iov_len = MAXPATHLEN;
3847 		uio.uio_iov = &iov;
3848 		uio.uio_iovcnt = 1;
3849 		uio.uio_segflg = UIO_SYSSPACE;
3850 		uio.uio_extflg = UIO_COPY_CACHED;
3851 		uio.uio_loffset = 0;
3852 		uio.uio_resid = MAXPATHLEN;
3853 
3854 		error = VOP_READLINK(vp, &uio, cs->cr, NULL);
3855 
3856 		if (!error)
3857 			*(data + MAXPATHLEN - uio.uio_resid) = '\0';
3858 	}
3859 
3860 	if (error) {
3861 		kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3862 		*cs->statusp = resp->status = puterrno4(error);
3863 		goto out;
3864 	}
3865 
3866 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
3867 	name = nfscmd_convname(ca, cs->exi, data, NFSCMD_CONV_OUTBOUND,
3868 	    MAXPATHLEN  + 1);
3869 
3870 	if (name == NULL) {
3871 		/*
3872 		 * Even though the conversion failed, we return
3873 		 * something. We just don't translate it.
3874 		 */
3875 		name = data;
3876 	}
3877 
3878 	/*
3879 	 * treat link name as data
3880 	 */
3881 	(void) str_to_utf8(name, &resp->link);
3882 
3883 	if (name != data)
3884 		kmem_free(name, MAXPATHLEN + 1);
3885 	kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3886 	*cs->statusp = resp->status = NFS4_OK;
3887 
3888 out:
3889 	DTRACE_NFSV4_2(op__readlink__done, struct compound_state *, cs,
3890 	    READLINK4res *, resp);
3891 }
3892 
3893 static void
3894 rfs4_op_readlink_free(nfs_resop4 *resop)
3895 {
3896 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3897 	utf8string *symlink = &resp->link;
3898 
3899 	if (symlink->utf8string_val) {
3900 		UTF8STRING_FREE(*symlink)
3901 	}
3902 }
3903 
3904 /*
3905  * release_lockowner:
3906  *	Release any state associated with the supplied
3907  *	lockowner. Note if any lo_state is holding locks we will not
3908  *	rele that lo_state and thus the lockowner will not be destroyed.
3909  *	A client using lock after the lock owner stateid has been released
3910  *	will suffer the consequence of NFS4ERR_BAD_STATEID and would have
3911  *	to reissue the lock with new_lock_owner set to TRUE.
3912  *	args: lock_owner
3913  *	res:  status
3914  */
3915 /* ARGSUSED */
3916 static void
3917 rfs4_op_release_lockowner(nfs_argop4 *argop, nfs_resop4 *resop,
3918     struct svc_req *req, struct compound_state *cs)
3919 {
3920 	RELEASE_LOCKOWNER4args *ap = &argop->nfs_argop4_u.oprelease_lockowner;
3921 	RELEASE_LOCKOWNER4res *resp = &resop->nfs_resop4_u.oprelease_lockowner;
3922 	rfs4_lockowner_t *lo;
3923 	rfs4_openowner_t *oo;
3924 	rfs4_state_t *sp;
3925 	rfs4_lo_state_t *lsp;
3926 	rfs4_client_t *cp;
3927 	bool_t create = FALSE;
3928 	locklist_t *llist;
3929 	sysid_t sysid;
3930 
3931 	DTRACE_NFSV4_2(op__release__lockowner__start, struct compound_state *,
3932 	    cs, RELEASE_LOCKOWNER4args *, ap);
3933 
3934 	/* Make sure there is a clientid around for this request */
3935 	cp = rfs4_findclient_by_id(ap->lock_owner.clientid, FALSE);
3936 
3937 	if (cp == NULL) {
3938 		*cs->statusp = resp->status =
3939 		    rfs4_check_clientid(&ap->lock_owner.clientid, 0);
3940 		goto out;
3941 	}
3942 	rfs4_client_rele(cp);
3943 
3944 	lo = rfs4_findlockowner(&ap->lock_owner, &create);
3945 	if (lo == NULL) {
3946 		*cs->statusp = resp->status = NFS4_OK;
3947 		goto out;
3948 	}
3949 	ASSERT(lo->rl_client != NULL);
3950 
3951 	/*
3952 	 * Check for EXPIRED client. If so will reap state with in a lease
3953 	 * period or on next set_clientid_confirm step
3954 	 */
3955 	if (rfs4_lease_expired(lo->rl_client)) {
3956 		rfs4_lockowner_rele(lo);
3957 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
3958 		goto out;
3959 	}
3960 
3961 	/*
3962 	 * If no sysid has been assigned, then no locks exist; just return.
3963 	 */
3964 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3965 	if (lo->rl_client->rc_sysidt == LM_NOSYSID) {
3966 		rfs4_lockowner_rele(lo);
3967 		rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3968 		goto out;
3969 	}
3970 
3971 	sysid = lo->rl_client->rc_sysidt;
3972 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3973 
3974 	/*
3975 	 * Mark the lockowner invalid.
3976 	 */
3977 	rfs4_dbe_hide(lo->rl_dbe);
3978 
3979 	/*
3980 	 * sysid-pid pair should now not be used since the lockowner is
3981 	 * invalid. If the client were to instantiate the lockowner again
3982 	 * it would be assigned a new pid. Thus we can get the list of
3983 	 * current locks.
3984 	 */
3985 
3986 	llist = flk_get_active_locks(sysid, lo->rl_pid);
3987 	/* If we are still holding locks fail */
3988 	if (llist != NULL) {
3989 
3990 		*cs->statusp = resp->status = NFS4ERR_LOCKS_HELD;
3991 
3992 		flk_free_locklist(llist);
3993 		/*
3994 		 * We need to unhide the lockowner so the client can
3995 		 * try it again. The bad thing here is if the client
3996 		 * has a logic error that took it here in the first place
3997 		 * he probably has lost accounting of the locks that it
3998 		 * is holding. So we may have dangling state until the
3999 		 * open owner state is reaped via close. One scenario
4000 		 * that could possibly occur is that the client has
4001 		 * sent the unlock request(s) in separate threads
4002 		 * and has not waited for the replies before sending the
4003 		 * RELEASE_LOCKOWNER request. Presumably, it would expect
4004 		 * and deal appropriately with NFS4ERR_LOCKS_HELD, by
4005 		 * reissuing the request.
4006 		 */
4007 		rfs4_dbe_unhide(lo->rl_dbe);
4008 		rfs4_lockowner_rele(lo);
4009 		goto out;
4010 	}
4011 
4012 	/*
4013 	 * For the corresponding client we need to check each open
4014 	 * owner for any opens that have lockowner state associated
4015 	 * with this lockowner.
4016 	 */
4017 
4018 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
4019 	for (oo = list_head(&lo->rl_client->rc_openownerlist); oo != NULL;
4020 	    oo = list_next(&lo->rl_client->rc_openownerlist, oo)) {
4021 
4022 		rfs4_dbe_lock(oo->ro_dbe);
4023 		for (sp = list_head(&oo->ro_statelist); sp != NULL;
4024 		    sp = list_next(&oo->ro_statelist, sp)) {
4025 
4026 			rfs4_dbe_lock(sp->rs_dbe);
4027 			for (lsp = list_head(&sp->rs_lostatelist);
4028 			    lsp != NULL;
4029 			    lsp = list_next(&sp->rs_lostatelist, lsp)) {
4030 				if (lsp->rls_locker == lo) {
4031 					rfs4_dbe_lock(lsp->rls_dbe);
4032 					rfs4_dbe_invalidate(lsp->rls_dbe);
4033 					rfs4_dbe_unlock(lsp->rls_dbe);
4034 				}
4035 			}
4036 			rfs4_dbe_unlock(sp->rs_dbe);
4037 		}
4038 		rfs4_dbe_unlock(oo->ro_dbe);
4039 	}
4040 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
4041 
4042 	rfs4_lockowner_rele(lo);
4043 
4044 	*cs->statusp = resp->status = NFS4_OK;
4045 
4046 out:
4047 	DTRACE_NFSV4_2(op__release__lockowner__done, struct compound_state *,
4048 	    cs, RELEASE_LOCKOWNER4res *, resp);
4049 }
4050 
4051 /*
4052  * short utility function to lookup a file and recall the delegation
4053  */
4054 static rfs4_file_t *
4055 rfs4_lookup_and_findfile(vnode_t *dvp, char *nm, vnode_t **vpp,
4056     int *lkup_error, cred_t *cr)
4057 {
4058 	vnode_t *vp;
4059 	rfs4_file_t *fp = NULL;
4060 	bool_t fcreate = FALSE;
4061 	int error;
4062 
4063 	if (vpp)
4064 		*vpp = NULL;
4065 
4066 	if ((error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cr, NULL, NULL,
4067 	    NULL)) == 0) {
4068 		if (vp->v_type == VREG)
4069 			fp = rfs4_findfile(vp, NULL, &fcreate);
4070 		if (vpp)
4071 			*vpp = vp;
4072 		else
4073 			VN_RELE(vp);
4074 	}
4075 
4076 	if (lkup_error)
4077 		*lkup_error = error;
4078 
4079 	return (fp);
4080 }
4081 
4082 /*
4083  * remove: args: CURRENT_FH: directory; name.
4084  *	res: status. If success - CURRENT_FH unchanged, return change_info
4085  *		for directory.
4086  */
4087 /* ARGSUSED */
4088 static void
4089 rfs4_op_remove(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4090     struct compound_state *cs)
4091 {
4092 	REMOVE4args *args = &argop->nfs_argop4_u.opremove;
4093 	REMOVE4res *resp = &resop->nfs_resop4_u.opremove;
4094 	int error;
4095 	vnode_t *dvp, *vp;
4096 	struct vattr bdva, idva, adva;
4097 	char *nm;
4098 	uint_t len;
4099 	rfs4_file_t *fp;
4100 	int in_crit = 0;
4101 	bslabel_t *clabel;
4102 	struct sockaddr *ca;
4103 	char *name = NULL;
4104 
4105 	DTRACE_NFSV4_2(op__remove__start, struct compound_state *, cs,
4106 	    REMOVE4args *, args);
4107 
4108 	/* CURRENT_FH: directory */
4109 	dvp = cs->vp;
4110 	if (dvp == NULL) {
4111 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4112 		goto out;
4113 	}
4114 
4115 	if (cs->access == CS_ACCESS_DENIED) {
4116 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4117 		goto out;
4118 	}
4119 
4120 	/*
4121 	 * If there is an unshared filesystem mounted on this vnode,
4122 	 * Do not allow to remove anything in this directory.
4123 	 */
4124 	if (vn_ismntpt(dvp)) {
4125 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4126 		goto out;
4127 	}
4128 
4129 	if (dvp->v_type != VDIR) {
4130 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4131 		goto out;
4132 	}
4133 
4134 	if (!utf8_dir_verify(&args->target)) {
4135 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4136 		goto out;
4137 	}
4138 
4139 	/*
4140 	 * Lookup the file so that we can check if it's a directory
4141 	 */
4142 	nm = utf8_to_fn(&args->target, &len, NULL);
4143 	if (nm == NULL) {
4144 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4145 		goto out;
4146 	}
4147 
4148 	if (len > MAXNAMELEN) {
4149 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4150 		kmem_free(nm, len);
4151 		goto out;
4152 	}
4153 
4154 	if (rdonly4(cs->exi, cs->vp, req)) {
4155 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4156 		kmem_free(nm, len);
4157 		goto out;
4158 	}
4159 
4160 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4161 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
4162 	    MAXPATHLEN  + 1);
4163 
4164 	if (name == NULL) {
4165 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4166 		kmem_free(nm, len);
4167 		goto out;
4168 	}
4169 
4170 	/*
4171 	 * Lookup the file to determine type and while we are see if
4172 	 * there is a file struct around and check for delegation.
4173 	 * We don't need to acquire va_seq before this lookup, if
4174 	 * it causes an update, cinfo.before will not match, which will
4175 	 * trigger a cache flush even if atomic is TRUE.
4176 	 */
4177 	if (fp = rfs4_lookup_and_findfile(dvp, name, &vp, &error, cs->cr)) {
4178 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4179 		    NULL)) {
4180 			VN_RELE(vp);
4181 			rfs4_file_rele(fp);
4182 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4183 			if (nm != name)
4184 				kmem_free(name, MAXPATHLEN + 1);
4185 			kmem_free(nm, len);
4186 			goto out;
4187 		}
4188 	}
4189 
4190 	/* Didn't find anything to remove */
4191 	if (vp == NULL) {
4192 		*cs->statusp = resp->status = error;
4193 		if (nm != name)
4194 			kmem_free(name, MAXPATHLEN + 1);
4195 		kmem_free(nm, len);
4196 		goto out;
4197 	}
4198 
4199 	if (nbl_need_check(vp)) {
4200 		nbl_start_crit(vp, RW_READER);
4201 		in_crit = 1;
4202 		if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
4203 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4204 			if (nm != name)
4205 				kmem_free(name, MAXPATHLEN + 1);
4206 			kmem_free(nm, len);
4207 			nbl_end_crit(vp);
4208 			VN_RELE(vp);
4209 			if (fp) {
4210 				rfs4_clear_dont_grant(fp);
4211 				rfs4_file_rele(fp);
4212 			}
4213 			goto out;
4214 		}
4215 	}
4216 
4217 	/* check label before allowing removal */
4218 	if (is_system_labeled()) {
4219 		ASSERT(req->rq_label != NULL);
4220 		clabel = req->rq_label;
4221 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
4222 		    "got client label from request(1)",
4223 		    struct svc_req *, req);
4224 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4225 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
4226 			    cs->exi)) {
4227 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4228 				if (name != nm)
4229 					kmem_free(name, MAXPATHLEN + 1);
4230 				kmem_free(nm, len);
4231 				if (in_crit)
4232 					nbl_end_crit(vp);
4233 				VN_RELE(vp);
4234 				if (fp) {
4235 					rfs4_clear_dont_grant(fp);
4236 					rfs4_file_rele(fp);
4237 				}
4238 				goto out;
4239 			}
4240 		}
4241 	}
4242 
4243 	/* Get dir "before" change value */
4244 	bdva.va_mask = AT_CTIME|AT_SEQ;
4245 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
4246 	if (error) {
4247 		*cs->statusp = resp->status = puterrno4(error);
4248 		if (nm != name)
4249 			kmem_free(name, MAXPATHLEN + 1);
4250 		kmem_free(nm, len);
4251 		if (in_crit)
4252 			nbl_end_crit(vp);
4253 		VN_RELE(vp);
4254 		if (fp) {
4255 			rfs4_clear_dont_grant(fp);
4256 			rfs4_file_rele(fp);
4257 		}
4258 		goto out;
4259 	}
4260 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
4261 
4262 	/* Actually do the REMOVE operation */
4263 	if (vp->v_type == VDIR) {
4264 		/*
4265 		 * Can't remove a directory that has a mounted-on filesystem.
4266 		 */
4267 		if (vn_ismntpt(vp)) {
4268 			error = EACCES;
4269 		} else {
4270 			/*
4271 			 * System V defines rmdir to return EEXIST,
4272 			 * not ENOTEMPTY, if the directory is not
4273 			 * empty.  A System V NFS server needs to map
4274 			 * NFS4ERR_EXIST to NFS4ERR_NOTEMPTY to
4275 			 * transmit over the wire.
4276 			 */
4277 			if ((error = VOP_RMDIR(dvp, name, rootdir, cs->cr,
4278 			    NULL, 0)) == EEXIST)
4279 				error = ENOTEMPTY;
4280 		}
4281 	} else {
4282 		if ((error = VOP_REMOVE(dvp, name, cs->cr, NULL, 0)) == 0 &&
4283 		    fp != NULL) {
4284 			struct vattr va;
4285 			vnode_t *tvp;
4286 
4287 			rfs4_dbe_lock(fp->rf_dbe);
4288 			tvp = fp->rf_vp;
4289 			if (tvp)
4290 				VN_HOLD(tvp);
4291 			rfs4_dbe_unlock(fp->rf_dbe);
4292 
4293 			if (tvp) {
4294 				/*
4295 				 * This is va_seq safe because we are not
4296 				 * manipulating dvp.
4297 				 */
4298 				va.va_mask = AT_NLINK;
4299 				if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4300 				    va.va_nlink == 0) {
4301 					/* Remove state on file remove */
4302 					if (in_crit) {
4303 						nbl_end_crit(vp);
4304 						in_crit = 0;
4305 					}
4306 					rfs4_close_all_state(fp);
4307 				}
4308 				VN_RELE(tvp);
4309 			}
4310 		}
4311 	}
4312 
4313 	if (in_crit)
4314 		nbl_end_crit(vp);
4315 	VN_RELE(vp);
4316 
4317 	if (fp) {
4318 		rfs4_clear_dont_grant(fp);
4319 		rfs4_file_rele(fp);
4320 	}
4321 	if (nm != name)
4322 		kmem_free(name, MAXPATHLEN + 1);
4323 	kmem_free(nm, len);
4324 
4325 	if (error) {
4326 		*cs->statusp = resp->status = puterrno4(error);
4327 		goto out;
4328 	}
4329 
4330 	/*
4331 	 * Get the initial "after" sequence number, if it fails, set to zero
4332 	 */
4333 	idva.va_mask = AT_SEQ;
4334 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
4335 		idva.va_seq = 0;
4336 
4337 	/*
4338 	 * Force modified data and metadata out to stable storage.
4339 	 */
4340 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
4341 
4342 	/*
4343 	 * Get "after" change value, if it fails, simply return the
4344 	 * before value.
4345 	 */
4346 	adva.va_mask = AT_CTIME|AT_SEQ;
4347 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
4348 		adva.va_ctime = bdva.va_ctime;
4349 		adva.va_seq = 0;
4350 	}
4351 
4352 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
4353 
4354 	/*
4355 	 * The cinfo.atomic = TRUE only if we have
4356 	 * non-zero va_seq's, and it has incremented by exactly one
4357 	 * during the VOP_REMOVE/RMDIR and it didn't change during
4358 	 * the VOP_FSYNC.
4359 	 */
4360 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
4361 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
4362 		resp->cinfo.atomic = TRUE;
4363 	else
4364 		resp->cinfo.atomic = FALSE;
4365 
4366 	*cs->statusp = resp->status = NFS4_OK;
4367 
4368 out:
4369 	DTRACE_NFSV4_2(op__remove__done, struct compound_state *, cs,
4370 	    REMOVE4res *, resp);
4371 }
4372 
4373 /*
4374  * rename: args: SAVED_FH: from directory, CURRENT_FH: target directory,
4375  *		oldname and newname.
4376  *	res: status. If success - CURRENT_FH unchanged, return change_info
4377  *		for both from and target directories.
4378  */
4379 /* ARGSUSED */
4380 static void
4381 rfs4_op_rename(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4382     struct compound_state *cs)
4383 {
4384 	RENAME4args *args = &argop->nfs_argop4_u.oprename;
4385 	RENAME4res *resp = &resop->nfs_resop4_u.oprename;
4386 	int error;
4387 	vnode_t *odvp;
4388 	vnode_t *ndvp;
4389 	vnode_t *srcvp, *targvp;
4390 	struct vattr obdva, oidva, oadva;
4391 	struct vattr nbdva, nidva, nadva;
4392 	char *onm, *nnm;
4393 	uint_t olen, nlen;
4394 	rfs4_file_t *fp, *sfp;
4395 	int in_crit_src, in_crit_targ;
4396 	int fp_rele_grant_hold, sfp_rele_grant_hold;
4397 	bslabel_t *clabel;
4398 	struct sockaddr *ca;
4399 	char *converted_onm = NULL;
4400 	char *converted_nnm = NULL;
4401 
4402 	DTRACE_NFSV4_2(op__rename__start, struct compound_state *, cs,
4403 	    RENAME4args *, args);
4404 
4405 	fp = sfp = NULL;
4406 	srcvp = targvp = NULL;
4407 	in_crit_src = in_crit_targ = 0;
4408 	fp_rele_grant_hold = sfp_rele_grant_hold = 0;
4409 
4410 	/* CURRENT_FH: target directory */
4411 	ndvp = cs->vp;
4412 	if (ndvp == NULL) {
4413 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4414 		goto out;
4415 	}
4416 
4417 	/* SAVED_FH: from directory */
4418 	odvp = cs->saved_vp;
4419 	if (odvp == NULL) {
4420 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4421 		goto out;
4422 	}
4423 
4424 	if (cs->access == CS_ACCESS_DENIED) {
4425 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4426 		goto out;
4427 	}
4428 
4429 	/*
4430 	 * If there is an unshared filesystem mounted on this vnode,
4431 	 * do not allow to rename objects in this directory.
4432 	 */
4433 	if (vn_ismntpt(odvp)) {
4434 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4435 		goto out;
4436 	}
4437 
4438 	/*
4439 	 * If there is an unshared filesystem mounted on this vnode,
4440 	 * do not allow to rename to this directory.
4441 	 */
4442 	if (vn_ismntpt(ndvp)) {
4443 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4444 		goto out;
4445 	}
4446 
4447 	if (odvp->v_type != VDIR || ndvp->v_type != VDIR) {
4448 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4449 		goto out;
4450 	}
4451 
4452 	if (cs->saved_exi != cs->exi) {
4453 		*cs->statusp = resp->status = NFS4ERR_XDEV;
4454 		goto out;
4455 	}
4456 
4457 	if (!utf8_dir_verify(&args->oldname)) {
4458 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4459 		goto out;
4460 	}
4461 
4462 	if (!utf8_dir_verify(&args->newname)) {
4463 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4464 		goto out;
4465 	}
4466 
4467 	onm = utf8_to_fn(&args->oldname, &olen, NULL);
4468 	if (onm == NULL) {
4469 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4470 		goto out;
4471 	}
4472 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4473 	nlen = MAXPATHLEN + 1;
4474 	converted_onm = nfscmd_convname(ca, cs->exi, onm, NFSCMD_CONV_INBOUND,
4475 	    nlen);
4476 
4477 	if (converted_onm == NULL) {
4478 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4479 		kmem_free(onm, olen);
4480 		goto out;
4481 	}
4482 
4483 	nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4484 	if (nnm == NULL) {
4485 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4486 		if (onm != converted_onm)
4487 			kmem_free(converted_onm, MAXPATHLEN + 1);
4488 		kmem_free(onm, olen);
4489 		goto out;
4490 	}
4491 	converted_nnm = nfscmd_convname(ca, cs->exi, nnm, NFSCMD_CONV_INBOUND,
4492 	    MAXPATHLEN  + 1);
4493 
4494 	if (converted_nnm == NULL) {
4495 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4496 		kmem_free(nnm, nlen);
4497 		nnm = NULL;
4498 		if (onm != converted_onm)
4499 			kmem_free(converted_onm, MAXPATHLEN + 1);
4500 		kmem_free(onm, olen);
4501 		goto out;
4502 	}
4503 
4504 
4505 	if (olen > MAXNAMELEN || nlen > MAXNAMELEN) {
4506 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4507 		kmem_free(onm, olen);
4508 		kmem_free(nnm, nlen);
4509 		goto out;
4510 	}
4511 
4512 
4513 	if (rdonly4(cs->exi, cs->vp, req)) {
4514 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4515 		if (onm != converted_onm)
4516 			kmem_free(converted_onm, MAXPATHLEN + 1);
4517 		kmem_free(onm, olen);
4518 		if (nnm != converted_nnm)
4519 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4520 		kmem_free(nnm, nlen);
4521 		goto out;
4522 	}
4523 
4524 	/* check label of the target dir */
4525 	if (is_system_labeled()) {
4526 		ASSERT(req->rq_label != NULL);
4527 		clabel = req->rq_label;
4528 		DTRACE_PROBE2(tx__rfs4__log__info__oprename__clabel, char *,
4529 		    "got client label from request(1)",
4530 		    struct svc_req *, req);
4531 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4532 			if (!do_rfs_label_check(clabel, ndvp,
4533 			    EQUALITY_CHECK, cs->exi)) {
4534 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4535 				goto err_out;
4536 			}
4537 		}
4538 	}
4539 
4540 	/*
4541 	 * Is the source a file and have a delegation?
4542 	 * We don't need to acquire va_seq before these lookups, if
4543 	 * it causes an update, cinfo.before will not match, which will
4544 	 * trigger a cache flush even if atomic is TRUE.
4545 	 */
4546 	if (sfp = rfs4_lookup_and_findfile(odvp, converted_onm, &srcvp,
4547 	    &error, cs->cr)) {
4548 		if (rfs4_check_delegated_byfp(FWRITE, sfp, TRUE, TRUE, TRUE,
4549 		    NULL)) {
4550 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4551 			goto err_out;
4552 		}
4553 	}
4554 
4555 	if (srcvp == NULL) {
4556 		*cs->statusp = resp->status = puterrno4(error);
4557 		if (onm != converted_onm)
4558 			kmem_free(converted_onm, MAXPATHLEN + 1);
4559 		kmem_free(onm, olen);
4560 		if (nnm != converted_nnm)
4561 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4562 		kmem_free(nnm, nlen);
4563 		goto out;
4564 	}
4565 
4566 	sfp_rele_grant_hold = 1;
4567 
4568 	/* Does the destination exist and a file and have a delegation? */
4569 	if (fp = rfs4_lookup_and_findfile(ndvp, converted_nnm, &targvp,
4570 	    NULL, cs->cr)) {
4571 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4572 		    NULL)) {
4573 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4574 			goto err_out;
4575 		}
4576 	}
4577 	fp_rele_grant_hold = 1;
4578 
4579 
4580 	/* Check for NBMAND lock on both source and target */
4581 	if (nbl_need_check(srcvp)) {
4582 		nbl_start_crit(srcvp, RW_READER);
4583 		in_crit_src = 1;
4584 		if (nbl_conflict(srcvp, NBL_RENAME, 0, 0, 0, NULL)) {
4585 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4586 			goto err_out;
4587 		}
4588 	}
4589 
4590 	if (targvp && nbl_need_check(targvp)) {
4591 		nbl_start_crit(targvp, RW_READER);
4592 		in_crit_targ = 1;
4593 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
4594 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4595 			goto err_out;
4596 		}
4597 	}
4598 
4599 	/* Get source "before" change value */
4600 	obdva.va_mask = AT_CTIME|AT_SEQ;
4601 	error = VOP_GETATTR(odvp, &obdva, 0, cs->cr, NULL);
4602 	if (!error) {
4603 		nbdva.va_mask = AT_CTIME|AT_SEQ;
4604 		error = VOP_GETATTR(ndvp, &nbdva, 0, cs->cr, NULL);
4605 	}
4606 	if (error) {
4607 		*cs->statusp = resp->status = puterrno4(error);
4608 		goto err_out;
4609 	}
4610 
4611 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.before, obdva.va_ctime)
4612 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.before, nbdva.va_ctime)
4613 
4614 	if ((error = VOP_RENAME(odvp, converted_onm, ndvp, converted_nnm,
4615 	    cs->cr, NULL, 0)) == 0 && fp != NULL) {
4616 		struct vattr va;
4617 		vnode_t *tvp;
4618 
4619 		rfs4_dbe_lock(fp->rf_dbe);
4620 		tvp = fp->rf_vp;
4621 		if (tvp)
4622 			VN_HOLD(tvp);
4623 		rfs4_dbe_unlock(fp->rf_dbe);
4624 
4625 		if (tvp) {
4626 			va.va_mask = AT_NLINK;
4627 			if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4628 			    va.va_nlink == 0) {
4629 				/* The file is gone and so should the state */
4630 				if (in_crit_targ) {
4631 					nbl_end_crit(targvp);
4632 					in_crit_targ = 0;
4633 				}
4634 				rfs4_close_all_state(fp);
4635 			}
4636 			VN_RELE(tvp);
4637 		}
4638 	}
4639 	if (error == 0)
4640 		vn_renamepath(ndvp, srcvp, nnm, nlen - 1);
4641 
4642 	if (in_crit_src)
4643 		nbl_end_crit(srcvp);
4644 	if (srcvp)
4645 		VN_RELE(srcvp);
4646 	if (in_crit_targ)
4647 		nbl_end_crit(targvp);
4648 	if (targvp)
4649 		VN_RELE(targvp);
4650 
4651 	if (sfp) {
4652 		rfs4_clear_dont_grant(sfp);
4653 		rfs4_file_rele(sfp);
4654 	}
4655 	if (fp) {
4656 		rfs4_clear_dont_grant(fp);
4657 		rfs4_file_rele(fp);
4658 	}
4659 
4660 	if (converted_onm != onm)
4661 		kmem_free(converted_onm, MAXPATHLEN + 1);
4662 	kmem_free(onm, olen);
4663 	if (converted_nnm != nnm)
4664 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4665 	kmem_free(nnm, nlen);
4666 
4667 	/*
4668 	 * Get the initial "after" sequence number, if it fails, set to zero
4669 	 */
4670 	oidva.va_mask = AT_SEQ;
4671 	if (VOP_GETATTR(odvp, &oidva, 0, cs->cr, NULL))
4672 		oidva.va_seq = 0;
4673 
4674 	nidva.va_mask = AT_SEQ;
4675 	if (VOP_GETATTR(ndvp, &nidva, 0, cs->cr, NULL))
4676 		nidva.va_seq = 0;
4677 
4678 	/*
4679 	 * Force modified data and metadata out to stable storage.
4680 	 */
4681 	(void) VOP_FSYNC(odvp, 0, cs->cr, NULL);
4682 	(void) VOP_FSYNC(ndvp, 0, cs->cr, NULL);
4683 
4684 	if (error) {
4685 		*cs->statusp = resp->status = puterrno4(error);
4686 		goto out;
4687 	}
4688 
4689 	/*
4690 	 * Get "after" change values, if it fails, simply return the
4691 	 * before value.
4692 	 */
4693 	oadva.va_mask = AT_CTIME|AT_SEQ;
4694 	if (VOP_GETATTR(odvp, &oadva, 0, cs->cr, NULL)) {
4695 		oadva.va_ctime = obdva.va_ctime;
4696 		oadva.va_seq = 0;
4697 	}
4698 
4699 	nadva.va_mask = AT_CTIME|AT_SEQ;
4700 	if (VOP_GETATTR(odvp, &nadva, 0, cs->cr, NULL)) {
4701 		nadva.va_ctime = nbdva.va_ctime;
4702 		nadva.va_seq = 0;
4703 	}
4704 
4705 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.after, oadva.va_ctime)
4706 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.after, nadva.va_ctime)
4707 
4708 	/*
4709 	 * The cinfo.atomic = TRUE only if we have
4710 	 * non-zero va_seq's, and it has incremented by exactly one
4711 	 * during the VOP_RENAME and it didn't change during the VOP_FSYNC.
4712 	 */
4713 	if (obdva.va_seq && oidva.va_seq && oadva.va_seq &&
4714 	    oidva.va_seq == (obdva.va_seq + 1) && oidva.va_seq == oadva.va_seq)
4715 		resp->source_cinfo.atomic = TRUE;
4716 	else
4717 		resp->source_cinfo.atomic = FALSE;
4718 
4719 	if (nbdva.va_seq && nidva.va_seq && nadva.va_seq &&
4720 	    nidva.va_seq == (nbdva.va_seq + 1) && nidva.va_seq == nadva.va_seq)
4721 		resp->target_cinfo.atomic = TRUE;
4722 	else
4723 		resp->target_cinfo.atomic = FALSE;
4724 
4725 #ifdef	VOLATILE_FH_TEST
4726 	{
4727 	extern void add_volrnm_fh(struct exportinfo *, vnode_t *);
4728 
4729 	/*
4730 	 * Add the renamed file handle to the volatile rename list
4731 	 */
4732 	if (cs->exi->exi_export.ex_flags & EX_VOLRNM) {
4733 		/* file handles may expire on rename */
4734 		vnode_t *vp;
4735 
4736 		nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4737 		/*
4738 		 * Already know that nnm will be a valid string
4739 		 */
4740 		error = VOP_LOOKUP(ndvp, nnm, &vp, NULL, 0, NULL, cs->cr,
4741 		    NULL, NULL, NULL);
4742 		kmem_free(nnm, nlen);
4743 		if (!error) {
4744 			add_volrnm_fh(cs->exi, vp);
4745 			VN_RELE(vp);
4746 		}
4747 	}
4748 	}
4749 #endif	/* VOLATILE_FH_TEST */
4750 
4751 	*cs->statusp = resp->status = NFS4_OK;
4752 out:
4753 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4754 	    RENAME4res *, resp);
4755 	return;
4756 
4757 err_out:
4758 	if (onm != converted_onm)
4759 		kmem_free(converted_onm, MAXPATHLEN + 1);
4760 	if (onm != NULL)
4761 		kmem_free(onm, olen);
4762 	if (nnm != converted_nnm)
4763 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4764 	if (nnm != NULL)
4765 		kmem_free(nnm, nlen);
4766 
4767 	if (in_crit_src) nbl_end_crit(srcvp);
4768 	if (in_crit_targ) nbl_end_crit(targvp);
4769 	if (targvp) VN_RELE(targvp);
4770 	if (srcvp) VN_RELE(srcvp);
4771 	if (sfp) {
4772 		if (sfp_rele_grant_hold) rfs4_clear_dont_grant(sfp);
4773 		rfs4_file_rele(sfp);
4774 	}
4775 	if (fp) {
4776 		if (fp_rele_grant_hold) rfs4_clear_dont_grant(fp);
4777 		rfs4_file_rele(fp);
4778 	}
4779 
4780 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4781 	    RENAME4res *, resp);
4782 }
4783 
4784 /* ARGSUSED */
4785 static void
4786 rfs4_op_renew(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4787     struct compound_state *cs)
4788 {
4789 	RENEW4args *args = &argop->nfs_argop4_u.oprenew;
4790 	RENEW4res *resp = &resop->nfs_resop4_u.oprenew;
4791 	rfs4_client_t *cp;
4792 
4793 	DTRACE_NFSV4_2(op__renew__start, struct compound_state *, cs,
4794 	    RENEW4args *, args);
4795 
4796 	if ((cp = rfs4_findclient_by_id(args->clientid, FALSE)) == NULL) {
4797 		*cs->statusp = resp->status =
4798 		    rfs4_check_clientid(&args->clientid, 0);
4799 		goto out;
4800 	}
4801 
4802 	if (rfs4_lease_expired(cp)) {
4803 		rfs4_client_rele(cp);
4804 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
4805 		goto out;
4806 	}
4807 
4808 	rfs4_update_lease(cp);
4809 
4810 	mutex_enter(cp->rc_cbinfo.cb_lock);
4811 	if (cp->rc_cbinfo.cb_notified_of_cb_path_down == FALSE) {
4812 		cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
4813 		*cs->statusp = resp->status = NFS4ERR_CB_PATH_DOWN;
4814 	} else {
4815 		*cs->statusp = resp->status = NFS4_OK;
4816 	}
4817 	mutex_exit(cp->rc_cbinfo.cb_lock);
4818 
4819 	rfs4_client_rele(cp);
4820 
4821 out:
4822 	DTRACE_NFSV4_2(op__renew__done, struct compound_state *, cs,
4823 	    RENEW4res *, resp);
4824 }
4825 
4826 /* ARGSUSED */
4827 static void
4828 rfs4_op_restorefh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
4829     struct compound_state *cs)
4830 {
4831 	RESTOREFH4res *resp = &resop->nfs_resop4_u.oprestorefh;
4832 
4833 	DTRACE_NFSV4_1(op__restorefh__start, struct compound_state *, cs);
4834 
4835 	/* No need to check cs->access - we are not accessing any object */
4836 	if ((cs->saved_vp == NULL) || (cs->saved_fh.nfs_fh4_val == NULL)) {
4837 		*cs->statusp = resp->status = NFS4ERR_RESTOREFH;
4838 		goto out;
4839 	}
4840 	if (cs->vp != NULL) {
4841 		VN_RELE(cs->vp);
4842 	}
4843 	cs->vp = cs->saved_vp;
4844 	cs->saved_vp = NULL;
4845 	cs->exi = cs->saved_exi;
4846 	nfs_fh4_copy(&cs->saved_fh, &cs->fh);
4847 	*cs->statusp = resp->status = NFS4_OK;
4848 	cs->deleg = FALSE;
4849 
4850 out:
4851 	DTRACE_NFSV4_2(op__restorefh__done, struct compound_state *, cs,
4852 	    RESTOREFH4res *, resp);
4853 }
4854 
4855 /* ARGSUSED */
4856 static void
4857 rfs4_op_savefh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4858     struct compound_state *cs)
4859 {
4860 	SAVEFH4res *resp = &resop->nfs_resop4_u.opsavefh;
4861 
4862 	DTRACE_NFSV4_1(op__savefh__start, struct compound_state *, cs);
4863 
4864 	/* No need to check cs->access - we are not accessing any object */
4865 	if (cs->vp == NULL) {
4866 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4867 		goto out;
4868 	}
4869 	if (cs->saved_vp != NULL) {
4870 		VN_RELE(cs->saved_vp);
4871 	}
4872 	cs->saved_vp = cs->vp;
4873 	VN_HOLD(cs->saved_vp);
4874 	cs->saved_exi = cs->exi;
4875 	/*
4876 	 * since SAVEFH is fairly rare, don't alloc space for its fh
4877 	 * unless necessary.
4878 	 */
4879 	if (cs->saved_fh.nfs_fh4_val == NULL) {
4880 		cs->saved_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP);
4881 	}
4882 	nfs_fh4_copy(&cs->fh, &cs->saved_fh);
4883 	*cs->statusp = resp->status = NFS4_OK;
4884 
4885 out:
4886 	DTRACE_NFSV4_2(op__savefh__done, struct compound_state *, cs,
4887 	    SAVEFH4res *, resp);
4888 }
4889 
4890 /*
4891  * rfs4_verify_attr is called when nfsv4 Setattr failed, but we wish to
4892  * return the bitmap of attrs that were set successfully. It is also
4893  * called by Verify/Nverify to test the vattr/vfsstat attrs. It should
4894  * always be called only after rfs4_do_set_attrs().
4895  *
4896  * Verify that the attributes are same as the expected ones. sargp->vap
4897  * and sargp->sbp contain the input attributes as translated from fattr4.
4898  *
4899  * This function verifies only the attrs that correspond to a vattr or
4900  * vfsstat struct. That is because of the extra step needed to get the
4901  * corresponding system structs. Other attributes have already been set or
4902  * verified by do_rfs4_set_attrs.
4903  *
4904  * Return 0 if all attrs match, -1 if some don't, error if error processing.
4905  */
4906 static int
4907 rfs4_verify_attr(struct nfs4_svgetit_arg *sargp,
4908     bitmap4 *resp, struct nfs4_ntov_table *ntovp)
4909 {
4910 	int error, ret_error = 0;
4911 	int i, k;
4912 	uint_t sva_mask = sargp->vap->va_mask;
4913 	uint_t vbit;
4914 	union nfs4_attr_u *na;
4915 	uint8_t *amap;
4916 	bool_t getsb = ntovp->vfsstat;
4917 
4918 	if (sva_mask != 0) {
4919 		/*
4920 		 * Okay to overwrite sargp->vap because we verify based
4921 		 * on the incoming values.
4922 		 */
4923 		ret_error = VOP_GETATTR(sargp->cs->vp, sargp->vap, 0,
4924 		    sargp->cs->cr, NULL);
4925 		if (ret_error) {
4926 			if (resp == NULL)
4927 				return (ret_error);
4928 			/*
4929 			 * Must return bitmap of successful attrs
4930 			 */
4931 			sva_mask = 0;	/* to prevent checking vap later */
4932 		} else {
4933 			/*
4934 			 * Some file systems clobber va_mask. it is probably
4935 			 * wrong of them to do so, nonethless we practice
4936 			 * defensive coding.
4937 			 * See bug id 4276830.
4938 			 */
4939 			sargp->vap->va_mask = sva_mask;
4940 		}
4941 	}
4942 
4943 	if (getsb) {
4944 		/*
4945 		 * Now get the superblock and loop on the bitmap, as there is
4946 		 * no simple way of translating from superblock to bitmap4.
4947 		 */
4948 		ret_error = VFS_STATVFS(sargp->cs->vp->v_vfsp, sargp->sbp);
4949 		if (ret_error) {
4950 			if (resp == NULL)
4951 				goto errout;
4952 			getsb = FALSE;
4953 		}
4954 	}
4955 
4956 	/*
4957 	 * Now loop and verify each attribute which getattr returned
4958 	 * whether it's the same as the input.
4959 	 */
4960 	if (resp == NULL && !getsb && (sva_mask == 0))
4961 		goto errout;
4962 
4963 	na = ntovp->na;
4964 	amap = ntovp->amap;
4965 	k = 0;
4966 	for (i = 0; i < ntovp->attrcnt; i++, na++, amap++) {
4967 		k = *amap;
4968 		ASSERT(nfs4_ntov_map[k].nval == k);
4969 		vbit = nfs4_ntov_map[k].vbit;
4970 
4971 		/*
4972 		 * If vattr attribute but VOP_GETATTR failed, or it's
4973 		 * superblock attribute but VFS_STATVFS failed, skip
4974 		 */
4975 		if (vbit) {
4976 			if ((vbit & sva_mask) == 0)
4977 				continue;
4978 		} else if (!(getsb && nfs4_ntov_map[k].vfsstat)) {
4979 			continue;
4980 		}
4981 		error = (*nfs4_ntov_map[k].sv_getit)(NFS4ATTR_VERIT, sargp, na);
4982 		if (resp != NULL) {
4983 			if (error)
4984 				ret_error = -1;	/* not all match */
4985 			else	/* update response bitmap */
4986 				*resp |= nfs4_ntov_map[k].fbit;
4987 			continue;
4988 		}
4989 		if (error) {
4990 			ret_error = -1;	/* not all match */
4991 			break;
4992 		}
4993 	}
4994 errout:
4995 	return (ret_error);
4996 }
4997 
4998 /*
4999  * Decode the attribute to be set/verified. If the attr requires a sys op
5000  * (VOP_GETATTR, VFS_VFSSTAT), and the request is to verify, then don't
5001  * call the sv_getit function for it, because the sys op hasn't yet been done.
5002  * Return 0 for success, error code if failed.
5003  *
5004  * Note: the decoded arg is not freed here but in nfs4_ntov_table_free.
5005  */
5006 static int
5007 decode_fattr4_attr(nfs4_attr_cmd_t cmd, struct nfs4_svgetit_arg *sargp,
5008     int k, XDR *xdrp, bitmap4 *resp_bval, union nfs4_attr_u *nap)
5009 {
5010 	int error = 0;
5011 	bool_t set_later;
5012 
5013 	sargp->vap->va_mask |= nfs4_ntov_map[k].vbit;
5014 
5015 	if ((*nfs4_ntov_map[k].xfunc)(xdrp, nap)) {
5016 		set_later = nfs4_ntov_map[k].vbit || nfs4_ntov_map[k].vfsstat;
5017 		/*
5018 		 * don't verify yet if a vattr or sb dependent attr,
5019 		 * because we don't have their sys values yet.
5020 		 * Will be done later.
5021 		 */
5022 		if (! (set_later && (cmd == NFS4ATTR_VERIT))) {
5023 			/*
5024 			 * ACLs are a special case, since setting the MODE
5025 			 * conflicts with setting the ACL.  We delay setting
5026 			 * the ACL until all other attributes have been set.
5027 			 * The ACL gets set in do_rfs4_op_setattr().
5028 			 */
5029 			if (nfs4_ntov_map[k].fbit != FATTR4_ACL_MASK) {
5030 				error = (*nfs4_ntov_map[k].sv_getit)(cmd,
5031 				    sargp, nap);
5032 				if (error) {
5033 					xdr_free(nfs4_ntov_map[k].xfunc,
5034 					    (caddr_t)nap);
5035 				}
5036 			}
5037 		}
5038 	} else {
5039 #ifdef  DEBUG
5040 		cmn_err(CE_NOTE, "decode_fattr4_attr: error "
5041 		    "decoding attribute %d\n", k);
5042 #endif
5043 		error = EINVAL;
5044 	}
5045 	if (!error && resp_bval && !set_later) {
5046 		*resp_bval |= nfs4_ntov_map[k].fbit;
5047 	}
5048 
5049 	return (error);
5050 }
5051 
5052 /*
5053  * Set vattr based on incoming fattr4 attrs - used by setattr.
5054  * Set response mask. Ignore any values that are not writable vattr attrs.
5055  */
5056 static nfsstat4
5057 do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5058     struct nfs4_svgetit_arg *sargp, struct nfs4_ntov_table *ntovp,
5059     nfs4_attr_cmd_t cmd)
5060 {
5061 	int error = 0;
5062 	int i;
5063 	char *attrs = fattrp->attrlist4;
5064 	uint32_t attrslen = fattrp->attrlist4_len;
5065 	XDR xdr;
5066 	nfsstat4 status = NFS4_OK;
5067 	vnode_t *vp = cs->vp;
5068 	union nfs4_attr_u *na;
5069 	uint8_t *amap;
5070 
5071 #ifndef lint
5072 	/*
5073 	 * Make sure that maximum attribute number can be expressed as an
5074 	 * 8 bit quantity.
5075 	 */
5076 	ASSERT(NFS4_MAXNUM_ATTRS <= (UINT8_MAX + 1));
5077 #endif
5078 
5079 	if (vp == NULL) {
5080 		if (resp)
5081 			*resp = 0;
5082 		return (NFS4ERR_NOFILEHANDLE);
5083 	}
5084 	if (cs->access == CS_ACCESS_DENIED) {
5085 		if (resp)
5086 			*resp = 0;
5087 		return (NFS4ERR_ACCESS);
5088 	}
5089 
5090 	sargp->op = cmd;
5091 	sargp->cs = cs;
5092 	sargp->flag = 0;	/* may be set later */
5093 	sargp->vap->va_mask = 0;
5094 	sargp->rdattr_error = NFS4_OK;
5095 	sargp->rdattr_error_req = FALSE;
5096 	/* sargp->sbp is set by the caller */
5097 
5098 	xdrmem_create(&xdr, attrs, attrslen, XDR_DECODE);
5099 
5100 	na = ntovp->na;
5101 	amap = ntovp->amap;
5102 
5103 	/*
5104 	 * The following loop iterates on the nfs4_ntov_map checking
5105 	 * if the fbit is set in the requested bitmap.
5106 	 * If set then we process the arguments using the
5107 	 * rfs4_fattr4 conversion functions to populate the setattr
5108 	 * vattr and va_mask. Any settable attrs that are not using vattr
5109 	 * will be set in this loop.
5110 	 */
5111 	for (i = 0; i < nfs4_ntov_map_size; i++) {
5112 		if (!(fattrp->attrmask & nfs4_ntov_map[i].fbit)) {
5113 			continue;
5114 		}
5115 		/*
5116 		 * If setattr, must be a writable attr.
5117 		 * If verify/nverify, must be a readable attr.
5118 		 */
5119 		if ((error = (*nfs4_ntov_map[i].sv_getit)(
5120 		    NFS4ATTR_SUPPORTED, sargp, NULL)) != 0) {
5121 			/*
5122 			 * Client tries to set/verify an
5123 			 * unsupported attribute, tries to set
5124 			 * a read only attr or verify a write
5125 			 * only one - error!
5126 			 */
5127 			break;
5128 		}
5129 		/*
5130 		 * Decode the attribute to set/verify
5131 		 */
5132 		error = decode_fattr4_attr(cmd, sargp, nfs4_ntov_map[i].nval,
5133 		    &xdr, resp ? resp : NULL, na);
5134 		if (error)
5135 			break;
5136 		*amap++ = (uint8_t)nfs4_ntov_map[i].nval;
5137 		na++;
5138 		(ntovp->attrcnt)++;
5139 		if (nfs4_ntov_map[i].vfsstat)
5140 			ntovp->vfsstat = TRUE;
5141 	}
5142 
5143 	if (error != 0)
5144 		status = (error == ENOTSUP ? NFS4ERR_ATTRNOTSUPP :
5145 		    puterrno4(error));
5146 	/* xdrmem_destroy(&xdrs); */	/* NO-OP */
5147 	return (status);
5148 }
5149 
5150 static nfsstat4
5151 do_rfs4_op_setattr(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5152     stateid4 *stateid)
5153 {
5154 	int error = 0;
5155 	struct nfs4_svgetit_arg sarg;
5156 	bool_t trunc;
5157 
5158 	nfsstat4 status = NFS4_OK;
5159 	cred_t *cr = cs->cr;
5160 	vnode_t *vp = cs->vp;
5161 	struct nfs4_ntov_table ntov;
5162 	struct statvfs64 sb;
5163 	struct vattr bva;
5164 	struct flock64 bf;
5165 	int in_crit = 0;
5166 	uint_t saved_mask = 0;
5167 	caller_context_t ct;
5168 
5169 	*resp = 0;
5170 	sarg.sbp = &sb;
5171 	sarg.is_referral = B_FALSE;
5172 	nfs4_ntov_table_init(&ntov);
5173 	status = do_rfs4_set_attrs(resp, fattrp, cs, &sarg, &ntov,
5174 	    NFS4ATTR_SETIT);
5175 	if (status != NFS4_OK) {
5176 		/*
5177 		 * failed set attrs
5178 		 */
5179 		goto done;
5180 	}
5181 	if ((sarg.vap->va_mask == 0) &&
5182 	    (! (fattrp->attrmask & FATTR4_ACL_MASK))) {
5183 		/*
5184 		 * no further work to be done
5185 		 */
5186 		goto done;
5187 	}
5188 
5189 	/*
5190 	 * If we got a request to set the ACL and the MODE, only
5191 	 * allow changing VSUID, VSGID, and VSVTX.  Attempting
5192 	 * to change any other bits, along with setting an ACL,
5193 	 * gives NFS4ERR_INVAL.
5194 	 */
5195 	if ((fattrp->attrmask & FATTR4_ACL_MASK) &&
5196 	    (fattrp->attrmask & FATTR4_MODE_MASK)) {
5197 		vattr_t va;
5198 
5199 		va.va_mask = AT_MODE;
5200 		error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
5201 		if (error) {
5202 			status = puterrno4(error);
5203 			goto done;
5204 		}
5205 		if ((sarg.vap->va_mode ^ va.va_mode) &
5206 		    ~(VSUID | VSGID | VSVTX)) {
5207 			status = NFS4ERR_INVAL;
5208 			goto done;
5209 		}
5210 	}
5211 
5212 	/* Check stateid only if size has been set */
5213 	if (sarg.vap->va_mask & AT_SIZE) {
5214 		trunc = (sarg.vap->va_size == 0);
5215 		status = rfs4_check_stateid(FWRITE, cs->vp, stateid,
5216 		    trunc, &cs->deleg, sarg.vap->va_mask & AT_SIZE, &ct);
5217 		if (status != NFS4_OK)
5218 			goto done;
5219 	} else {
5220 		ct.cc_sysid = 0;
5221 		ct.cc_pid = 0;
5222 		ct.cc_caller_id = nfs4_srv_caller_id;
5223 		ct.cc_flags = CC_DONTBLOCK;
5224 	}
5225 
5226 	/* XXX start of possible race with delegations */
5227 
5228 	/*
5229 	 * We need to specially handle size changes because it is
5230 	 * possible for the client to create a file with read-only
5231 	 * modes, but with the file opened for writing. If the client
5232 	 * then tries to set the file size, e.g. ftruncate(3C),
5233 	 * fcntl(F_FREESP), the normal access checking done in
5234 	 * VOP_SETATTR would prevent the client from doing it even though
5235 	 * it should be allowed to do so.  To get around this, we do the
5236 	 * access checking for ourselves and use VOP_SPACE which doesn't
5237 	 * do the access checking.
5238 	 * Also the client should not be allowed to change the file
5239 	 * size if there is a conflicting non-blocking mandatory lock in
5240 	 * the region of the change.
5241 	 */
5242 	if (vp->v_type == VREG && (sarg.vap->va_mask & AT_SIZE)) {
5243 		u_offset_t offset;
5244 		ssize_t length;
5245 
5246 		/*
5247 		 * ufs_setattr clears AT_SIZE from vap->va_mask, but
5248 		 * before returning, sarg.vap->va_mask is used to
5249 		 * generate the setattr reply bitmap.  We also clear
5250 		 * AT_SIZE below before calling VOP_SPACE.  For both
5251 		 * of these cases, the va_mask needs to be saved here
5252 		 * and restored after calling VOP_SETATTR.
5253 		 */
5254 		saved_mask = sarg.vap->va_mask;
5255 
5256 		/*
5257 		 * Check any possible conflict due to NBMAND locks.
5258 		 * Get into critical region before VOP_GETATTR, so the
5259 		 * size attribute is valid when checking conflicts.
5260 		 */
5261 		if (nbl_need_check(vp)) {
5262 			nbl_start_crit(vp, RW_READER);
5263 			in_crit = 1;
5264 		}
5265 
5266 		bva.va_mask = AT_UID|AT_SIZE;
5267 		if (error = VOP_GETATTR(vp, &bva, 0, cr, &ct)) {
5268 			status = puterrno4(error);
5269 			goto done;
5270 		}
5271 
5272 		if (in_crit) {
5273 			if (sarg.vap->va_size < bva.va_size) {
5274 				offset = sarg.vap->va_size;
5275 				length = bva.va_size - sarg.vap->va_size;
5276 			} else {
5277 				offset = bva.va_size;
5278 				length = sarg.vap->va_size - bva.va_size;
5279 			}
5280 			if (nbl_conflict(vp, NBL_WRITE, offset, length, 0,
5281 			    &ct)) {
5282 				status = NFS4ERR_LOCKED;
5283 				goto done;
5284 			}
5285 		}
5286 
5287 		if (crgetuid(cr) == bva.va_uid) {
5288 			sarg.vap->va_mask &= ~AT_SIZE;
5289 			bf.l_type = F_WRLCK;
5290 			bf.l_whence = 0;
5291 			bf.l_start = (off64_t)sarg.vap->va_size;
5292 			bf.l_len = 0;
5293 			bf.l_sysid = 0;
5294 			bf.l_pid = 0;
5295 			error = VOP_SPACE(vp, F_FREESP, &bf, FWRITE,
5296 			    (offset_t)sarg.vap->va_size, cr, &ct);
5297 		}
5298 	}
5299 
5300 	if (!error && sarg.vap->va_mask != 0)
5301 		error = VOP_SETATTR(vp, sarg.vap, sarg.flag, cr, &ct);
5302 
5303 	/* restore va_mask -- ufs_setattr clears AT_SIZE */
5304 	if (saved_mask & AT_SIZE)
5305 		sarg.vap->va_mask |= AT_SIZE;
5306 
5307 	/*
5308 	 * If an ACL was being set, it has been delayed until now,
5309 	 * in order to set the mode (via the VOP_SETATTR() above) first.
5310 	 */
5311 	if ((! error) && (fattrp->attrmask & FATTR4_ACL_MASK)) {
5312 		int i;
5313 
5314 		for (i = 0; i < NFS4_MAXNUM_ATTRS; i++)
5315 			if (ntov.amap[i] == FATTR4_ACL)
5316 				break;
5317 		if (i < NFS4_MAXNUM_ATTRS) {
5318 			error = (*nfs4_ntov_map[FATTR4_ACL].sv_getit)(
5319 			    NFS4ATTR_SETIT, &sarg, &ntov.na[i]);
5320 			if (error == 0) {
5321 				*resp |= FATTR4_ACL_MASK;
5322 			} else if (error == ENOTSUP) {
5323 				(void) rfs4_verify_attr(&sarg, resp, &ntov);
5324 				status = NFS4ERR_ATTRNOTSUPP;
5325 				goto done;
5326 			}
5327 		} else {
5328 			NFS4_DEBUG(rfs4_debug,
5329 			    (CE_NOTE, "do_rfs4_op_setattr: "
5330 			    "unable to find ACL in fattr4"));
5331 			error = EINVAL;
5332 		}
5333 	}
5334 
5335 	if (error) {
5336 		/* check if a monitor detected a delegation conflict */
5337 		if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
5338 			status = NFS4ERR_DELAY;
5339 		else
5340 			status = puterrno4(error);
5341 
5342 		/*
5343 		 * Set the response bitmap when setattr failed.
5344 		 * If VOP_SETATTR partially succeeded, test by doing a
5345 		 * VOP_GETATTR on the object and comparing the data
5346 		 * to the setattr arguments.
5347 		 */
5348 		(void) rfs4_verify_attr(&sarg, resp, &ntov);
5349 	} else {
5350 		/*
5351 		 * Force modified metadata out to stable storage.
5352 		 */
5353 		(void) VOP_FSYNC(vp, FNODSYNC, cr, &ct);
5354 		/*
5355 		 * Set response bitmap
5356 		 */
5357 		nfs4_vmask_to_nmask_set(sarg.vap->va_mask, resp);
5358 	}
5359 
5360 /* Return early and already have a NFSv4 error */
5361 done:
5362 	/*
5363 	 * Except for nfs4_vmask_to_nmask_set(), vattr --> fattr
5364 	 * conversion sets both readable and writeable NFS4 attrs
5365 	 * for AT_MTIME and AT_ATIME.  The line below masks out
5366 	 * unrequested attrs from the setattr result bitmap.  This
5367 	 * is placed after the done: label to catch the ATTRNOTSUP
5368 	 * case.
5369 	 */
5370 	*resp &= fattrp->attrmask;
5371 
5372 	if (in_crit)
5373 		nbl_end_crit(vp);
5374 
5375 	nfs4_ntov_table_free(&ntov, &sarg);
5376 
5377 	return (status);
5378 }
5379 
5380 /* ARGSUSED */
5381 static void
5382 rfs4_op_setattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5383     struct compound_state *cs)
5384 {
5385 	SETATTR4args *args = &argop->nfs_argop4_u.opsetattr;
5386 	SETATTR4res *resp = &resop->nfs_resop4_u.opsetattr;
5387 	bslabel_t *clabel;
5388 
5389 	DTRACE_NFSV4_2(op__setattr__start, struct compound_state *, cs,
5390 	    SETATTR4args *, args);
5391 
5392 	if (cs->vp == NULL) {
5393 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5394 		goto out;
5395 	}
5396 
5397 	/*
5398 	 * If there is an unshared filesystem mounted on this vnode,
5399 	 * do not allow to setattr on this vnode.
5400 	 */
5401 	if (vn_ismntpt(cs->vp)) {
5402 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5403 		goto out;
5404 	}
5405 
5406 	resp->attrsset = 0;
5407 
5408 	if (rdonly4(cs->exi, cs->vp, req)) {
5409 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5410 		goto out;
5411 	}
5412 
5413 	/* check label before setting attributes */
5414 	if (is_system_labeled()) {
5415 		ASSERT(req->rq_label != NULL);
5416 		clabel = req->rq_label;
5417 		DTRACE_PROBE2(tx__rfs4__log__info__opsetattr__clabel, char *,
5418 		    "got client label from request(1)",
5419 		    struct svc_req *, req);
5420 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
5421 			if (!do_rfs_label_check(clabel, cs->vp,
5422 			    EQUALITY_CHECK, cs->exi)) {
5423 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
5424 				goto out;
5425 			}
5426 		}
5427 	}
5428 
5429 	*cs->statusp = resp->status =
5430 	    do_rfs4_op_setattr(&resp->attrsset, &args->obj_attributes, cs,
5431 	    &args->stateid);
5432 
5433 out:
5434 	DTRACE_NFSV4_2(op__setattr__done, struct compound_state *, cs,
5435 	    SETATTR4res *, resp);
5436 }
5437 
5438 /* ARGSUSED */
5439 static void
5440 rfs4_op_verify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5441     struct compound_state *cs)
5442 {
5443 	/*
5444 	 * verify and nverify are exactly the same, except that nverify
5445 	 * succeeds when some argument changed, and verify succeeds when
5446 	 * when none changed.
5447 	 */
5448 
5449 	VERIFY4args  *args = &argop->nfs_argop4_u.opverify;
5450 	VERIFY4res *resp = &resop->nfs_resop4_u.opverify;
5451 
5452 	int error;
5453 	struct nfs4_svgetit_arg sarg;
5454 	struct statvfs64 sb;
5455 	struct nfs4_ntov_table ntov;
5456 
5457 	DTRACE_NFSV4_2(op__verify__start, struct compound_state *, cs,
5458 	    VERIFY4args *, args);
5459 
5460 	if (cs->vp == NULL) {
5461 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5462 		goto out;
5463 	}
5464 
5465 	sarg.sbp = &sb;
5466 	sarg.is_referral = B_FALSE;
5467 	nfs4_ntov_table_init(&ntov);
5468 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5469 	    &sarg, &ntov, NFS4ATTR_VERIT);
5470 	if (resp->status != NFS4_OK) {
5471 		/*
5472 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5473 		 * so could return -1 for "no match".
5474 		 */
5475 		if (resp->status == -1)
5476 			resp->status = NFS4ERR_NOT_SAME;
5477 		goto done;
5478 	}
5479 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5480 	switch (error) {
5481 	case 0:
5482 		resp->status = NFS4_OK;
5483 		break;
5484 	case -1:
5485 		resp->status = NFS4ERR_NOT_SAME;
5486 		break;
5487 	default:
5488 		resp->status = puterrno4(error);
5489 		break;
5490 	}
5491 done:
5492 	*cs->statusp = resp->status;
5493 	nfs4_ntov_table_free(&ntov, &sarg);
5494 out:
5495 	DTRACE_NFSV4_2(op__verify__done, struct compound_state *, cs,
5496 	    VERIFY4res *, resp);
5497 }
5498 
5499 /* ARGSUSED */
5500 static void
5501 rfs4_op_nverify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5502     struct compound_state *cs)
5503 {
5504 	/*
5505 	 * verify and nverify are exactly the same, except that nverify
5506 	 * succeeds when some argument changed, and verify succeeds when
5507 	 * when none changed.
5508 	 */
5509 
5510 	NVERIFY4args  *args = &argop->nfs_argop4_u.opnverify;
5511 	NVERIFY4res *resp = &resop->nfs_resop4_u.opnverify;
5512 
5513 	int error;
5514 	struct nfs4_svgetit_arg sarg;
5515 	struct statvfs64 sb;
5516 	struct nfs4_ntov_table ntov;
5517 
5518 	DTRACE_NFSV4_2(op__nverify__start, struct compound_state *, cs,
5519 	    NVERIFY4args *, args);
5520 
5521 	if (cs->vp == NULL) {
5522 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5523 		DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5524 		    NVERIFY4res *, resp);
5525 		return;
5526 	}
5527 	sarg.sbp = &sb;
5528 	sarg.is_referral = B_FALSE;
5529 	nfs4_ntov_table_init(&ntov);
5530 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5531 	    &sarg, &ntov, NFS4ATTR_VERIT);
5532 	if (resp->status != NFS4_OK) {
5533 		/*
5534 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5535 		 * so could return -1 for "no match".
5536 		 */
5537 		if (resp->status == -1)
5538 			resp->status = NFS4_OK;
5539 		goto done;
5540 	}
5541 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5542 	switch (error) {
5543 	case 0:
5544 		resp->status = NFS4ERR_SAME;
5545 		break;
5546 	case -1:
5547 		resp->status = NFS4_OK;
5548 		break;
5549 	default:
5550 		resp->status = puterrno4(error);
5551 		break;
5552 	}
5553 done:
5554 	*cs->statusp = resp->status;
5555 	nfs4_ntov_table_free(&ntov, &sarg);
5556 
5557 	DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5558 	    NVERIFY4res *, resp);
5559 }
5560 
5561 /*
5562  * XXX - This should live in an NFS header file.
5563  */
5564 #define	MAX_IOVECS	12
5565 
5566 /* ARGSUSED */
5567 static void
5568 rfs4_op_write(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5569     struct compound_state *cs)
5570 {
5571 	WRITE4args *args = &argop->nfs_argop4_u.opwrite;
5572 	WRITE4res *resp = &resop->nfs_resop4_u.opwrite;
5573 	int error;
5574 	vnode_t *vp;
5575 	struct vattr bva;
5576 	u_offset_t rlimit;
5577 	struct uio uio;
5578 	struct iovec iov[MAX_IOVECS];
5579 	struct iovec *iovp;
5580 	int iovcnt;
5581 	int ioflag;
5582 	cred_t *savecred, *cr;
5583 	bool_t *deleg = &cs->deleg;
5584 	nfsstat4 stat;
5585 	int in_crit = 0;
5586 	caller_context_t ct;
5587 
5588 	DTRACE_NFSV4_2(op__write__start, struct compound_state *, cs,
5589 	    WRITE4args *, args);
5590 
5591 	vp = cs->vp;
5592 	if (vp == NULL) {
5593 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5594 		goto out;
5595 	}
5596 	if (cs->access == CS_ACCESS_DENIED) {
5597 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5598 		goto out;
5599 	}
5600 
5601 	cr = cs->cr;
5602 
5603 	if ((stat = rfs4_check_stateid(FWRITE, vp, &args->stateid, FALSE,
5604 	    deleg, TRUE, &ct)) != NFS4_OK) {
5605 		*cs->statusp = resp->status = stat;
5606 		goto out;
5607 	}
5608 
5609 	/*
5610 	 * We have to enter the critical region before calling VOP_RWLOCK
5611 	 * to avoid a deadlock with ufs.
5612 	 */
5613 	if (nbl_need_check(vp)) {
5614 		nbl_start_crit(vp, RW_READER);
5615 		in_crit = 1;
5616 		if (nbl_conflict(vp, NBL_WRITE,
5617 		    args->offset, args->data_len, 0, &ct)) {
5618 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
5619 			goto out;
5620 		}
5621 	}
5622 
5623 	bva.va_mask = AT_MODE | AT_UID;
5624 	error = VOP_GETATTR(vp, &bva, 0, cr, &ct);
5625 
5626 	/*
5627 	 * If we can't get the attributes, then we can't do the
5628 	 * right access checking.  So, we'll fail the request.
5629 	 */
5630 	if (error) {
5631 		*cs->statusp = resp->status = puterrno4(error);
5632 		goto out;
5633 	}
5634 
5635 	if (rdonly4(cs->exi, cs->vp, req)) {
5636 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5637 		goto out;
5638 	}
5639 
5640 	if (vp->v_type != VREG) {
5641 		*cs->statusp = resp->status =
5642 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
5643 		goto out;
5644 	}
5645 
5646 	if (crgetuid(cr) != bva.va_uid &&
5647 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, &ct))) {
5648 		*cs->statusp = resp->status = puterrno4(error);
5649 		goto out;
5650 	}
5651 
5652 	if (MANDLOCK(vp, bva.va_mode)) {
5653 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5654 		goto out;
5655 	}
5656 
5657 	if (args->data_len == 0) {
5658 		*cs->statusp = resp->status = NFS4_OK;
5659 		resp->count = 0;
5660 		resp->committed = args->stable;
5661 		resp->writeverf = Write4verf;
5662 		goto out;
5663 	}
5664 
5665 	if (args->mblk != NULL) {
5666 		mblk_t *m;
5667 		uint_t bytes, round_len;
5668 
5669 		iovcnt = 0;
5670 		bytes = 0;
5671 		round_len = roundup(args->data_len, BYTES_PER_XDR_UNIT);
5672 		for (m = args->mblk;
5673 		    m != NULL && bytes < round_len;
5674 		    m = m->b_cont) {
5675 			iovcnt++;
5676 			bytes += MBLKL(m);
5677 		}
5678 #ifdef DEBUG
5679 		/* should have ended on an mblk boundary */
5680 		if (bytes != round_len) {
5681 			printf("bytes=0x%x, round_len=0x%x, req len=0x%x\n",
5682 			    bytes, round_len, args->data_len);
5683 			printf("args=%p, args->mblk=%p, m=%p", (void *)args,
5684 			    (void *)args->mblk, (void *)m);
5685 			ASSERT(bytes == round_len);
5686 		}
5687 #endif
5688 		if (iovcnt <= MAX_IOVECS) {
5689 			iovp = iov;
5690 		} else {
5691 			iovp = kmem_alloc(sizeof (*iovp) * iovcnt, KM_SLEEP);
5692 		}
5693 		mblk_to_iov(args->mblk, iovcnt, iovp);
5694 	} else if (args->rlist != NULL) {
5695 		iovcnt = 1;
5696 		iovp = iov;
5697 		iovp->iov_base = (char *)((args->rlist)->u.c_daddr3);
5698 		iovp->iov_len = args->data_len;
5699 	} else {
5700 		iovcnt = 1;
5701 		iovp = iov;
5702 		iovp->iov_base = args->data_val;
5703 		iovp->iov_len = args->data_len;
5704 	}
5705 
5706 	uio.uio_iov = iovp;
5707 	uio.uio_iovcnt = iovcnt;
5708 
5709 	uio.uio_segflg = UIO_SYSSPACE;
5710 	uio.uio_extflg = UIO_COPY_DEFAULT;
5711 	uio.uio_loffset = args->offset;
5712 	uio.uio_resid = args->data_len;
5713 	uio.uio_llimit = curproc->p_fsz_ctl;
5714 	rlimit = uio.uio_llimit - args->offset;
5715 	if (rlimit < (u_offset_t)uio.uio_resid)
5716 		uio.uio_resid = (int)rlimit;
5717 
5718 	if (args->stable == UNSTABLE4)
5719 		ioflag = 0;
5720 	else if (args->stable == FILE_SYNC4)
5721 		ioflag = FSYNC;
5722 	else if (args->stable == DATA_SYNC4)
5723 		ioflag = FDSYNC;
5724 	else {
5725 		if (iovp != iov)
5726 			kmem_free(iovp, sizeof (*iovp) * iovcnt);
5727 		*cs->statusp = resp->status = NFS4ERR_INVAL;
5728 		goto out;
5729 	}
5730 
5731 	/*
5732 	 * We're changing creds because VM may fault and we need
5733 	 * the cred of the current thread to be used if quota
5734 	 * checking is enabled.
5735 	 */
5736 	savecred = curthread->t_cred;
5737 	curthread->t_cred = cr;
5738 	error = do_io(FWRITE, vp, &uio, ioflag, cr, &ct);
5739 	curthread->t_cred = savecred;
5740 
5741 	if (iovp != iov)
5742 		kmem_free(iovp, sizeof (*iovp) * iovcnt);
5743 
5744 	if (error) {
5745 		*cs->statusp = resp->status = puterrno4(error);
5746 		goto out;
5747 	}
5748 
5749 	*cs->statusp = resp->status = NFS4_OK;
5750 	resp->count = args->data_len - uio.uio_resid;
5751 
5752 	if (ioflag == 0)
5753 		resp->committed = UNSTABLE4;
5754 	else
5755 		resp->committed = FILE_SYNC4;
5756 
5757 	resp->writeverf = Write4verf;
5758 
5759 out:
5760 	if (in_crit)
5761 		nbl_end_crit(vp);
5762 
5763 	DTRACE_NFSV4_2(op__write__done, struct compound_state *, cs,
5764 	    WRITE4res *, resp);
5765 }
5766 
5767 
5768 /* XXX put in a header file */
5769 extern int	sec_svc_getcred(struct svc_req *, cred_t *,  caddr_t *, int *);
5770 
5771 void
5772 rfs4_compound(COMPOUND4args *args, COMPOUND4res *resp, struct exportinfo *exi,
5773     struct svc_req *req, cred_t *cr, int *rv)
5774 {
5775 	uint_t i;
5776 	struct compound_state cs;
5777 
5778 	if (rv != NULL)
5779 		*rv = 0;
5780 	rfs4_init_compound_state(&cs);
5781 	/*
5782 	 * Form a reply tag by copying over the reqeuest tag.
5783 	 */
5784 	resp->tag.utf8string_val =
5785 	    kmem_alloc(args->tag.utf8string_len, KM_SLEEP);
5786 	resp->tag.utf8string_len = args->tag.utf8string_len;
5787 	bcopy(args->tag.utf8string_val, resp->tag.utf8string_val,
5788 	    resp->tag.utf8string_len);
5789 
5790 	cs.statusp = &resp->status;
5791 	cs.req = req;
5792 
5793 	/*
5794 	 * XXX for now, minorversion should be zero
5795 	 */
5796 	if (args->minorversion != NFS4_MINORVERSION) {
5797 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5798 		    &cs, COMPOUND4args *, args);
5799 		resp->array_len = 0;
5800 		resp->array = NULL;
5801 		resp->status = NFS4ERR_MINOR_VERS_MISMATCH;
5802 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5803 		    &cs, COMPOUND4res *, resp);
5804 		return;
5805 	}
5806 
5807 	ASSERT(exi == NULL);
5808 	ASSERT(cr == NULL);
5809 
5810 	cr = crget();
5811 	ASSERT(cr != NULL);
5812 
5813 	if (sec_svc_getcred(req, cr, &cs.principal, &cs.nfsflavor) == 0) {
5814 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5815 		    &cs, COMPOUND4args *, args);
5816 		crfree(cr);
5817 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5818 		    &cs, COMPOUND4res *, resp);
5819 		svcerr_badcred(req->rq_xprt);
5820 		if (rv != NULL)
5821 			*rv = 1;
5822 		return;
5823 	}
5824 	resp->array_len = args->array_len;
5825 	resp->array = kmem_zalloc(args->array_len * sizeof (nfs_resop4),
5826 	    KM_SLEEP);
5827 
5828 	cs.basecr = cr;
5829 
5830 	DTRACE_NFSV4_2(compound__start, struct compound_state *, &cs,
5831 	    COMPOUND4args *, args);
5832 
5833 	/*
5834 	 * For now, NFS4 compound processing must be protected by
5835 	 * exported_lock because it can access more than one exportinfo
5836 	 * per compound and share/unshare can now change multiple
5837 	 * exinfo structs.  The NFS2/3 code only refs 1 exportinfo
5838 	 * per proc (excluding public exinfo), and exi_count design
5839 	 * is sufficient to protect concurrent execution of NFS2/3
5840 	 * ops along with unexport.  This lock will be removed as
5841 	 * part of the NFSv4 phase 2 namespace redesign work.
5842 	 */
5843 	rw_enter(&exported_lock, RW_READER);
5844 
5845 	/*
5846 	 * If this is the first compound we've seen, we need to start all
5847 	 * new instances' grace periods.
5848 	 */
5849 	if (rfs4_seen_first_compound == 0) {
5850 		rfs4_grace_start_new();
5851 		/*
5852 		 * This must be set after rfs4_grace_start_new(), otherwise
5853 		 * another thread could proceed past here before the former
5854 		 * is finished.
5855 		 */
5856 		rfs4_seen_first_compound = 1;
5857 	}
5858 
5859 	for (i = 0; i < args->array_len && cs.cont; i++) {
5860 		nfs_argop4 *argop;
5861 		nfs_resop4 *resop;
5862 		uint_t op;
5863 
5864 		argop = &args->array[i];
5865 		resop = &resp->array[i];
5866 		resop->resop = argop->argop;
5867 		op = (uint_t)resop->resop;
5868 
5869 		if (op < rfsv4disp_cnt) {
5870 			/*
5871 			 * Count the individual ops here; NULL and COMPOUND
5872 			 * are counted in common_dispatch()
5873 			 */
5874 			rfsproccnt_v4_ptr[op].value.ui64++;
5875 
5876 			NFS4_DEBUG(rfs4_debug > 1,
5877 			    (CE_NOTE, "Executing %s", rfs4_op_string[op]));
5878 			(*rfsv4disptab[op].dis_proc)(argop, resop, req, &cs);
5879 			NFS4_DEBUG(rfs4_debug > 1, (CE_NOTE, "%s returned %d",
5880 			    rfs4_op_string[op], *cs.statusp));
5881 			if (*cs.statusp != NFS4_OK)
5882 				cs.cont = FALSE;
5883 		} else {
5884 			/*
5885 			 * This is effectively dead code since XDR code
5886 			 * will have already returned BADXDR if op doesn't
5887 			 * decode to legal value.  This only done for a
5888 			 * day when XDR code doesn't verify v4 opcodes.
5889 			 */
5890 			op = OP_ILLEGAL;
5891 			rfsproccnt_v4_ptr[OP_ILLEGAL_IDX].value.ui64++;
5892 
5893 			rfs4_op_illegal(argop, resop, req, &cs);
5894 			cs.cont = FALSE;
5895 		}
5896 
5897 		/*
5898 		 * If not at last op, and if we are to stop, then
5899 		 * compact the results array.
5900 		 */
5901 		if ((i + 1) < args->array_len && !cs.cont) {
5902 			nfs_resop4 *new_res = kmem_alloc(
5903 			    (i+1) * sizeof (nfs_resop4), KM_SLEEP);
5904 			bcopy(resp->array,
5905 			    new_res, (i+1) * sizeof (nfs_resop4));
5906 			kmem_free(resp->array,
5907 			    args->array_len * sizeof (nfs_resop4));
5908 
5909 			resp->array_len =  i + 1;
5910 			resp->array = new_res;
5911 		}
5912 	}
5913 
5914 	rw_exit(&exported_lock);
5915 
5916 	DTRACE_NFSV4_2(compound__done, struct compound_state *, &cs,
5917 	    COMPOUND4res *, resp);
5918 
5919 	if (cs.vp)
5920 		VN_RELE(cs.vp);
5921 	if (cs.saved_vp)
5922 		VN_RELE(cs.saved_vp);
5923 	if (cs.saved_fh.nfs_fh4_val)
5924 		kmem_free(cs.saved_fh.nfs_fh4_val, NFS4_FHSIZE);
5925 
5926 	if (cs.basecr)
5927 		crfree(cs.basecr);
5928 	if (cs.cr)
5929 		crfree(cs.cr);
5930 	/*
5931 	 * done with this compound request, free the label
5932 	 */
5933 
5934 	if (req->rq_label != NULL) {
5935 		kmem_free(req->rq_label, sizeof (bslabel_t));
5936 		req->rq_label = NULL;
5937 	}
5938 }
5939 
5940 /*
5941  * XXX because of what appears to be duplicate calls to rfs4_compound_free
5942  * XXX zero out the tag and array values. Need to investigate why the
5943  * XXX calls occur, but at least prevent the panic for now.
5944  */
5945 void
5946 rfs4_compound_free(COMPOUND4res *resp)
5947 {
5948 	uint_t i;
5949 
5950 	if (resp->tag.utf8string_val) {
5951 		UTF8STRING_FREE(resp->tag)
5952 	}
5953 
5954 	for (i = 0; i < resp->array_len; i++) {
5955 		nfs_resop4 *resop;
5956 		uint_t op;
5957 
5958 		resop = &resp->array[i];
5959 		op = (uint_t)resop->resop;
5960 		if (op < rfsv4disp_cnt) {
5961 			(*rfsv4disptab[op].dis_resfree)(resop);
5962 		}
5963 	}
5964 	if (resp->array != NULL) {
5965 		kmem_free(resp->array, resp->array_len * sizeof (nfs_resop4));
5966 	}
5967 }
5968 
5969 /*
5970  * Process the value of the compound request rpc flags, as a bit-AND
5971  * of the individual per-op flags (idempotent, allowork, publicfh_ok)
5972  */
5973 void
5974 rfs4_compound_flagproc(COMPOUND4args *args, int *flagp)
5975 {
5976 	int i;
5977 	int flag = RPC_ALL;
5978 
5979 	for (i = 0; flag && i < args->array_len; i++) {
5980 		uint_t op;
5981 
5982 		op = (uint_t)args->array[i].argop;
5983 
5984 		if (op < rfsv4disp_cnt)
5985 			flag &= rfsv4disptab[op].dis_flags;
5986 		else
5987 			flag = 0;
5988 	}
5989 	*flagp = flag;
5990 }
5991 
5992 nfsstat4
5993 rfs4_client_sysid(rfs4_client_t *cp, sysid_t *sp)
5994 {
5995 	nfsstat4 e;
5996 
5997 	rfs4_dbe_lock(cp->rc_dbe);
5998 
5999 	if (cp->rc_sysidt != LM_NOSYSID) {
6000 		*sp = cp->rc_sysidt;
6001 		e = NFS4_OK;
6002 
6003 	} else if ((cp->rc_sysidt = lm_alloc_sysidt()) != LM_NOSYSID) {
6004 		*sp = cp->rc_sysidt;
6005 		e = NFS4_OK;
6006 
6007 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
6008 		    "rfs4_client_sysid: allocated 0x%x\n", *sp));
6009 	} else
6010 		e = NFS4ERR_DELAY;
6011 
6012 	rfs4_dbe_unlock(cp->rc_dbe);
6013 	return (e);
6014 }
6015 
6016 #if defined(DEBUG) && ! defined(lint)
6017 static void lock_print(char *str, int operation, struct flock64 *flk)
6018 {
6019 	char *op, *type;
6020 
6021 	switch (operation) {
6022 	case F_GETLK: op = "F_GETLK";
6023 		break;
6024 	case F_SETLK: op = "F_SETLK";
6025 		break;
6026 	case F_SETLK_NBMAND: op = "F_SETLK_NBMAND";
6027 		break;
6028 	default: op = "F_UNKNOWN";
6029 		break;
6030 	}
6031 	switch (flk->l_type) {
6032 	case F_UNLCK: type = "F_UNLCK";
6033 		break;
6034 	case F_RDLCK: type = "F_RDLCK";
6035 		break;
6036 	case F_WRLCK: type = "F_WRLCK";
6037 		break;
6038 	default: type = "F_UNKNOWN";
6039 		break;
6040 	}
6041 
6042 	ASSERT(flk->l_whence == 0);
6043 	cmn_err(CE_NOTE, "%s:  %s, type = %s, off = %llx len = %llx pid = %d",
6044 	    str, op, type, (longlong_t)flk->l_start,
6045 	    flk->l_len ? (longlong_t)flk->l_len : ~0LL, flk->l_pid);
6046 }
6047 
6048 #define	LOCK_PRINT(d, s, t, f) if (d) lock_print(s, t, f)
6049 #else
6050 #define	LOCK_PRINT(d, s, t, f)
6051 #endif
6052 
6053 /*ARGSUSED*/
6054 static bool_t
6055 creds_ok(cred_set_t cr_set, struct svc_req *req, struct compound_state *cs)
6056 {
6057 	return (TRUE);
6058 }
6059 
6060 /*
6061  * Look up the pathname using the vp in cs as the directory vnode.
6062  * cs->vp will be the vnode for the file on success
6063  */
6064 
6065 static nfsstat4
6066 rfs4_lookup(component4 *component, struct svc_req *req,
6067     struct compound_state *cs)
6068 {
6069 	char *nm;
6070 	uint32_t len;
6071 	nfsstat4 status;
6072 	struct sockaddr *ca;
6073 	char *name;
6074 
6075 	if (cs->vp == NULL) {
6076 		return (NFS4ERR_NOFILEHANDLE);
6077 	}
6078 	if (cs->vp->v_type != VDIR) {
6079 		return (NFS4ERR_NOTDIR);
6080 	}
6081 
6082 	if (!utf8_dir_verify(component))
6083 		return (NFS4ERR_INVAL);
6084 
6085 	nm = utf8_to_fn(component, &len, NULL);
6086 	if (nm == NULL) {
6087 		return (NFS4ERR_INVAL);
6088 	}
6089 
6090 	if (len > MAXNAMELEN) {
6091 		kmem_free(nm, len);
6092 		return (NFS4ERR_NAMETOOLONG);
6093 	}
6094 
6095 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6096 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6097 	    MAXPATHLEN + 1);
6098 
6099 	if (name == NULL) {
6100 		kmem_free(nm, len);
6101 		return (NFS4ERR_INVAL);
6102 	}
6103 
6104 	status = do_rfs4_op_lookup(name, req, cs);
6105 
6106 	if (name != nm)
6107 		kmem_free(name, MAXPATHLEN + 1);
6108 
6109 	kmem_free(nm, len);
6110 
6111 	return (status);
6112 }
6113 
6114 static nfsstat4
6115 rfs4_lookupfile(component4 *component, struct svc_req *req,
6116     struct compound_state *cs, uint32_t access, change_info4 *cinfo)
6117 {
6118 	nfsstat4 status;
6119 	vnode_t *dvp = cs->vp;
6120 	vattr_t bva, ava, fva;
6121 	int error;
6122 
6123 	/* Get "before" change value */
6124 	bva.va_mask = AT_CTIME|AT_SEQ;
6125 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6126 	if (error)
6127 		return (puterrno4(error));
6128 
6129 	/* rfs4_lookup may VN_RELE directory */
6130 	VN_HOLD(dvp);
6131 
6132 	status = rfs4_lookup(component, req, cs);
6133 	if (status != NFS4_OK) {
6134 		VN_RELE(dvp);
6135 		return (status);
6136 	}
6137 
6138 	/*
6139 	 * Get "after" change value, if it fails, simply return the
6140 	 * before value.
6141 	 */
6142 	ava.va_mask = AT_CTIME|AT_SEQ;
6143 	if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6144 		ava.va_ctime = bva.va_ctime;
6145 		ava.va_seq = 0;
6146 	}
6147 	VN_RELE(dvp);
6148 
6149 	/*
6150 	 * Validate the file is a file
6151 	 */
6152 	fva.va_mask = AT_TYPE|AT_MODE;
6153 	error = VOP_GETATTR(cs->vp, &fva, 0, cs->cr, NULL);
6154 	if (error)
6155 		return (puterrno4(error));
6156 
6157 	if (fva.va_type != VREG) {
6158 		if (fva.va_type == VDIR)
6159 			return (NFS4ERR_ISDIR);
6160 		if (fva.va_type == VLNK)
6161 			return (NFS4ERR_SYMLINK);
6162 		return (NFS4ERR_INVAL);
6163 	}
6164 
6165 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime);
6166 	NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6167 
6168 	/*
6169 	 * It is undefined if VOP_LOOKUP will change va_seq, so
6170 	 * cinfo.atomic = TRUE only if we have
6171 	 * non-zero va_seq's, and they have not changed.
6172 	 */
6173 	if (bva.va_seq && ava.va_seq && ava.va_seq == bva.va_seq)
6174 		cinfo->atomic = TRUE;
6175 	else
6176 		cinfo->atomic = FALSE;
6177 
6178 	/* Check for mandatory locking */
6179 	cs->mandlock = MANDLOCK(cs->vp, fva.va_mode);
6180 	return (check_open_access(access, cs, req));
6181 }
6182 
6183 static nfsstat4
6184 create_vnode(vnode_t *dvp, char *nm,  vattr_t *vap, createmode4 mode,
6185     timespec32_t *mtime, cred_t *cr, vnode_t **vpp, bool_t *created)
6186 {
6187 	int error;
6188 	nfsstat4 status = NFS4_OK;
6189 	vattr_t va;
6190 
6191 tryagain:
6192 
6193 	/*
6194 	 * The file open mode used is VWRITE.  If the client needs
6195 	 * some other semantic, then it should do the access checking
6196 	 * itself.  It would have been nice to have the file open mode
6197 	 * passed as part of the arguments.
6198 	 */
6199 
6200 	*created = TRUE;
6201 	error = VOP_CREATE(dvp, nm, vap, EXCL, VWRITE, vpp, cr, 0, NULL, NULL);
6202 
6203 	if (error) {
6204 		*created = FALSE;
6205 
6206 		/*
6207 		 * If we got something other than file already exists
6208 		 * then just return this error.  Otherwise, we got
6209 		 * EEXIST.  If we were doing a GUARDED create, then
6210 		 * just return this error.  Otherwise, we need to
6211 		 * make sure that this wasn't a duplicate of an
6212 		 * exclusive create request.
6213 		 *
6214 		 * The assumption is made that a non-exclusive create
6215 		 * request will never return EEXIST.
6216 		 */
6217 
6218 		if (error != EEXIST || mode == GUARDED4) {
6219 			status = puterrno4(error);
6220 			return (status);
6221 		}
6222 		error = VOP_LOOKUP(dvp, nm, vpp, NULL, 0, NULL, cr,
6223 		    NULL, NULL, NULL);
6224 
6225 		if (error) {
6226 			/*
6227 			 * We couldn't find the file that we thought that
6228 			 * we just created.  So, we'll just try creating
6229 			 * it again.
6230 			 */
6231 			if (error == ENOENT)
6232 				goto tryagain;
6233 
6234 			status = puterrno4(error);
6235 			return (status);
6236 		}
6237 
6238 		if (mode == UNCHECKED4) {
6239 			/* existing object must be regular file */
6240 			if ((*vpp)->v_type != VREG) {
6241 				if ((*vpp)->v_type == VDIR)
6242 					status = NFS4ERR_ISDIR;
6243 				else if ((*vpp)->v_type == VLNK)
6244 					status = NFS4ERR_SYMLINK;
6245 				else
6246 					status = NFS4ERR_INVAL;
6247 				VN_RELE(*vpp);
6248 				return (status);
6249 			}
6250 
6251 			return (NFS4_OK);
6252 		}
6253 
6254 		/* Check for duplicate request */
6255 		ASSERT(mtime != 0);
6256 		va.va_mask = AT_MTIME;
6257 		error = VOP_GETATTR(*vpp, &va, 0, cr, NULL);
6258 		if (!error) {
6259 			/* We found the file */
6260 			if (va.va_mtime.tv_sec != mtime->tv_sec ||
6261 			    va.va_mtime.tv_nsec != mtime->tv_nsec) {
6262 				/* but its not our creation */
6263 				VN_RELE(*vpp);
6264 				return (NFS4ERR_EXIST);
6265 			}
6266 			*created = TRUE; /* retrans of create == created */
6267 			return (NFS4_OK);
6268 		}
6269 		VN_RELE(*vpp);
6270 		return (NFS4ERR_EXIST);
6271 	}
6272 
6273 	return (NFS4_OK);
6274 }
6275 
6276 static nfsstat4
6277 check_open_access(uint32_t access, struct compound_state *cs,
6278     struct svc_req *req)
6279 {
6280 	int error;
6281 	vnode_t *vp;
6282 	bool_t readonly;
6283 	cred_t *cr = cs->cr;
6284 
6285 	/* For now we don't allow mandatory locking as per V2/V3 */
6286 	if (cs->access == CS_ACCESS_DENIED || cs->mandlock) {
6287 		return (NFS4ERR_ACCESS);
6288 	}
6289 
6290 	vp = cs->vp;
6291 	ASSERT(cr != NULL && vp->v_type == VREG);
6292 
6293 	/*
6294 	 * If the file system is exported read only and we are trying
6295 	 * to open for write, then return NFS4ERR_ROFS
6296 	 */
6297 
6298 	readonly = rdonly4(cs->exi, cs->vp, req);
6299 
6300 	if ((access & OPEN4_SHARE_ACCESS_WRITE) && readonly)
6301 		return (NFS4ERR_ROFS);
6302 
6303 	if (access & OPEN4_SHARE_ACCESS_READ) {
6304 		if ((VOP_ACCESS(vp, VREAD, 0, cr, NULL) != 0) &&
6305 		    (VOP_ACCESS(vp, VEXEC, 0, cr, NULL) != 0)) {
6306 			return (NFS4ERR_ACCESS);
6307 		}
6308 	}
6309 
6310 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
6311 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
6312 		if (error)
6313 			return (NFS4ERR_ACCESS);
6314 	}
6315 
6316 	return (NFS4_OK);
6317 }
6318 
6319 static nfsstat4
6320 rfs4_createfile(OPEN4args *args, struct svc_req *req, struct compound_state *cs,
6321     change_info4 *cinfo, bitmap4 *attrset, clientid4 clientid)
6322 {
6323 	struct nfs4_svgetit_arg sarg;
6324 	struct nfs4_ntov_table ntov;
6325 
6326 	bool_t ntov_table_init = FALSE;
6327 	struct statvfs64 sb;
6328 	nfsstat4 status;
6329 	vnode_t *vp;
6330 	vattr_t bva, ava, iva, cva, *vap;
6331 	vnode_t *dvp;
6332 	timespec32_t *mtime;
6333 	char *nm = NULL;
6334 	uint_t buflen;
6335 	bool_t created;
6336 	bool_t setsize = FALSE;
6337 	len_t reqsize;
6338 	int error;
6339 	bool_t trunc;
6340 	caller_context_t ct;
6341 	component4 *component;
6342 	bslabel_t *clabel;
6343 	struct sockaddr *ca;
6344 	char *name = NULL;
6345 
6346 	sarg.sbp = &sb;
6347 	sarg.is_referral = B_FALSE;
6348 
6349 	dvp = cs->vp;
6350 
6351 	/* Check if the file system is read only */
6352 	if (rdonly4(cs->exi, dvp, req))
6353 		return (NFS4ERR_ROFS);
6354 
6355 	/* check the label of including directory */
6356 	if (is_system_labeled()) {
6357 		ASSERT(req->rq_label != NULL);
6358 		clabel = req->rq_label;
6359 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
6360 		    "got client label from request(1)",
6361 		    struct svc_req *, req);
6362 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
6363 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
6364 			    cs->exi)) {
6365 				return (NFS4ERR_ACCESS);
6366 			}
6367 		}
6368 	}
6369 
6370 	/*
6371 	 * Get the last component of path name in nm. cs will reference
6372 	 * the including directory on success.
6373 	 */
6374 	component = &args->open_claim4_u.file;
6375 	if (!utf8_dir_verify(component))
6376 		return (NFS4ERR_INVAL);
6377 
6378 	nm = utf8_to_fn(component, &buflen, NULL);
6379 
6380 	if (nm == NULL)
6381 		return (NFS4ERR_RESOURCE);
6382 
6383 	if (buflen > MAXNAMELEN) {
6384 		kmem_free(nm, buflen);
6385 		return (NFS4ERR_NAMETOOLONG);
6386 	}
6387 
6388 	bva.va_mask = AT_TYPE|AT_CTIME|AT_SEQ;
6389 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6390 	if (error) {
6391 		kmem_free(nm, buflen);
6392 		return (puterrno4(error));
6393 	}
6394 
6395 	if (bva.va_type != VDIR) {
6396 		kmem_free(nm, buflen);
6397 		return (NFS4ERR_NOTDIR);
6398 	}
6399 
6400 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime)
6401 
6402 	switch (args->mode) {
6403 	case GUARDED4:
6404 		/*FALLTHROUGH*/
6405 	case UNCHECKED4:
6406 		nfs4_ntov_table_init(&ntov);
6407 		ntov_table_init = TRUE;
6408 
6409 		*attrset = 0;
6410 		status = do_rfs4_set_attrs(attrset,
6411 		    &args->createhow4_u.createattrs,
6412 		    cs, &sarg, &ntov, NFS4ATTR_SETIT);
6413 
6414 		if (status == NFS4_OK && (sarg.vap->va_mask & AT_TYPE) &&
6415 		    sarg.vap->va_type != VREG) {
6416 			if (sarg.vap->va_type == VDIR)
6417 				status = NFS4ERR_ISDIR;
6418 			else if (sarg.vap->va_type == VLNK)
6419 				status = NFS4ERR_SYMLINK;
6420 			else
6421 				status = NFS4ERR_INVAL;
6422 		}
6423 
6424 		if (status != NFS4_OK) {
6425 			kmem_free(nm, buflen);
6426 			nfs4_ntov_table_free(&ntov, &sarg);
6427 			*attrset = 0;
6428 			return (status);
6429 		}
6430 
6431 		vap = sarg.vap;
6432 		vap->va_type = VREG;
6433 		vap->va_mask |= AT_TYPE;
6434 
6435 		if ((vap->va_mask & AT_MODE) == 0) {
6436 			vap->va_mask |= AT_MODE;
6437 			vap->va_mode = (mode_t)0600;
6438 		}
6439 
6440 		if (vap->va_mask & AT_SIZE) {
6441 
6442 			/* Disallow create with a non-zero size */
6443 
6444 			if ((reqsize = sarg.vap->va_size) != 0) {
6445 				kmem_free(nm, buflen);
6446 				nfs4_ntov_table_free(&ntov, &sarg);
6447 				*attrset = 0;
6448 				return (NFS4ERR_INVAL);
6449 			}
6450 			setsize = TRUE;
6451 		}
6452 		break;
6453 
6454 	case EXCLUSIVE4:
6455 		/* prohibit EXCL create of named attributes */
6456 		if (dvp->v_flag & V_XATTRDIR) {
6457 			kmem_free(nm, buflen);
6458 			*attrset = 0;
6459 			return (NFS4ERR_INVAL);
6460 		}
6461 
6462 		cva.va_mask = AT_TYPE | AT_MTIME | AT_MODE;
6463 		cva.va_type = VREG;
6464 		/*
6465 		 * Ensure no time overflows. Assumes underlying
6466 		 * filesystem supports at least 32 bits.
6467 		 * Truncate nsec to usec resolution to allow valid
6468 		 * compares even if the underlying filesystem truncates.
6469 		 */
6470 		mtime = (timespec32_t *)&args->createhow4_u.createverf;
6471 		cva.va_mtime.tv_sec = mtime->tv_sec % TIME32_MAX;
6472 		cva.va_mtime.tv_nsec = (mtime->tv_nsec / 1000) * 1000;
6473 		cva.va_mode = (mode_t)0;
6474 		vap = &cva;
6475 
6476 		/*
6477 		 * For EXCL create, attrset is set to the server attr
6478 		 * used to cache the client's verifier.
6479 		 */
6480 		*attrset = FATTR4_TIME_MODIFY_MASK;
6481 		break;
6482 	}
6483 
6484 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6485 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6486 	    MAXPATHLEN  + 1);
6487 
6488 	if (name == NULL) {
6489 		kmem_free(nm, buflen);
6490 		return (NFS4ERR_SERVERFAULT);
6491 	}
6492 
6493 	status = create_vnode(dvp, name, vap, args->mode, mtime,
6494 	    cs->cr, &vp, &created);
6495 	if (nm != name)
6496 		kmem_free(name, MAXPATHLEN + 1);
6497 	kmem_free(nm, buflen);
6498 
6499 	if (status != NFS4_OK) {
6500 		if (ntov_table_init)
6501 			nfs4_ntov_table_free(&ntov, &sarg);
6502 		*attrset = 0;
6503 		return (status);
6504 	}
6505 
6506 	trunc = (setsize && !created);
6507 
6508 	if (args->mode != EXCLUSIVE4) {
6509 		bitmap4 createmask = args->createhow4_u.createattrs.attrmask;
6510 
6511 		/*
6512 		 * True verification that object was created with correct
6513 		 * attrs is impossible.  The attrs could have been changed
6514 		 * immediately after object creation.  If attributes did
6515 		 * not verify, the only recourse for the server is to
6516 		 * destroy the object.  Maybe if some attrs (like gid)
6517 		 * are set incorrectly, the object should be destroyed;
6518 		 * however, seems bad as a default policy.  Do we really
6519 		 * want to destroy an object over one of the times not
6520 		 * verifying correctly?  For these reasons, the server
6521 		 * currently sets bits in attrset for createattrs
6522 		 * that were set; however, no verification is done.
6523 		 *
6524 		 * vmask_to_nmask accounts for vattr bits set on create
6525 		 *	[do_rfs4_set_attrs() only sets resp bits for
6526 		 *	 non-vattr/vfs bits.]
6527 		 * Mask off any bits we set by default so as not to return
6528 		 * more attrset bits than were requested in createattrs
6529 		 */
6530 		if (created) {
6531 			nfs4_vmask_to_nmask(sarg.vap->va_mask, attrset);
6532 			*attrset &= createmask;
6533 		} else {
6534 			/*
6535 			 * We did not create the vnode (we tried but it
6536 			 * already existed).  In this case, the only createattr
6537 			 * that the spec allows the server to set is size,
6538 			 * and even then, it can only be set if it is 0.
6539 			 */
6540 			*attrset = 0;
6541 			if (trunc)
6542 				*attrset = FATTR4_SIZE_MASK;
6543 		}
6544 	}
6545 	if (ntov_table_init)
6546 		nfs4_ntov_table_free(&ntov, &sarg);
6547 
6548 	/*
6549 	 * Get the initial "after" sequence number, if it fails,
6550 	 * set to zero, time to before.
6551 	 */
6552 	iva.va_mask = AT_CTIME|AT_SEQ;
6553 	if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL)) {
6554 		iva.va_seq = 0;
6555 		iva.va_ctime = bva.va_ctime;
6556 	}
6557 
6558 	/*
6559 	 * create_vnode attempts to create the file exclusive,
6560 	 * if it already exists the VOP_CREATE will fail and
6561 	 * may not increase va_seq. It is atomic if
6562 	 * we haven't changed the directory, but if it has changed
6563 	 * we don't know what changed it.
6564 	 */
6565 	if (!created) {
6566 		if (bva.va_seq && iva.va_seq &&
6567 		    bva.va_seq == iva.va_seq)
6568 			cinfo->atomic = TRUE;
6569 		else
6570 			cinfo->atomic = FALSE;
6571 		NFS4_SET_FATTR4_CHANGE(cinfo->after, iva.va_ctime);
6572 	} else {
6573 		/*
6574 		 * The entry was created, we need to sync the
6575 		 * directory metadata.
6576 		 */
6577 		(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
6578 
6579 		/*
6580 		 * Get "after" change value, if it fails, simply return the
6581 		 * before value.
6582 		 */
6583 		ava.va_mask = AT_CTIME|AT_SEQ;
6584 		if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6585 			ava.va_ctime = bva.va_ctime;
6586 			ava.va_seq = 0;
6587 		}
6588 
6589 		NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6590 
6591 		/*
6592 		 * The cinfo->atomic = TRUE only if we have
6593 		 * non-zero va_seq's, and it has incremented by exactly one
6594 		 * during the create_vnode and it didn't
6595 		 * change during the VOP_FSYNC.
6596 		 */
6597 		if (bva.va_seq && iva.va_seq && ava.va_seq &&
6598 		    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
6599 			cinfo->atomic = TRUE;
6600 		else
6601 			cinfo->atomic = FALSE;
6602 	}
6603 
6604 	/* Check for mandatory locking and that the size gets set. */
6605 	cva.va_mask = AT_MODE;
6606 	if (setsize)
6607 		cva.va_mask |= AT_SIZE;
6608 
6609 	/* Assume the worst */
6610 	cs->mandlock = TRUE;
6611 
6612 	if (VOP_GETATTR(vp, &cva, 0, cs->cr, NULL) == 0) {
6613 		cs->mandlock = MANDLOCK(cs->vp, cva.va_mode);
6614 
6615 		/*
6616 		 * Truncate the file if necessary; this would be
6617 		 * the case for create over an existing file.
6618 		 */
6619 
6620 		if (trunc) {
6621 			int in_crit = 0;
6622 			rfs4_file_t *fp;
6623 			bool_t create = FALSE;
6624 
6625 			/*
6626 			 * We are writing over an existing file.
6627 			 * Check to see if we need to recall a delegation.
6628 			 */
6629 			rfs4_hold_deleg_policy();
6630 			if ((fp = rfs4_findfile(vp, NULL, &create)) != NULL) {
6631 				if (rfs4_check_delegated_byfp(FWRITE, fp,
6632 				    (reqsize == 0), FALSE, FALSE, &clientid)) {
6633 					rfs4_file_rele(fp);
6634 					rfs4_rele_deleg_policy();
6635 					VN_RELE(vp);
6636 					*attrset = 0;
6637 					return (NFS4ERR_DELAY);
6638 				}
6639 				rfs4_file_rele(fp);
6640 			}
6641 			rfs4_rele_deleg_policy();
6642 
6643 			if (nbl_need_check(vp)) {
6644 				in_crit = 1;
6645 
6646 				ASSERT(reqsize == 0);
6647 
6648 				nbl_start_crit(vp, RW_READER);
6649 				if (nbl_conflict(vp, NBL_WRITE, 0,
6650 				    cva.va_size, 0, NULL)) {
6651 					in_crit = 0;
6652 					nbl_end_crit(vp);
6653 					VN_RELE(vp);
6654 					*attrset = 0;
6655 					return (NFS4ERR_ACCESS);
6656 				}
6657 			}
6658 			ct.cc_sysid = 0;
6659 			ct.cc_pid = 0;
6660 			ct.cc_caller_id = nfs4_srv_caller_id;
6661 			ct.cc_flags = CC_DONTBLOCK;
6662 
6663 			cva.va_mask = AT_SIZE;
6664 			cva.va_size = reqsize;
6665 			(void) VOP_SETATTR(vp, &cva, 0, cs->cr, &ct);
6666 			if (in_crit)
6667 				nbl_end_crit(vp);
6668 		}
6669 	}
6670 
6671 	error = makefh4(&cs->fh, vp, cs->exi);
6672 
6673 	/*
6674 	 * Force modified data and metadata out to stable storage.
6675 	 */
6676 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
6677 
6678 	if (error) {
6679 		VN_RELE(vp);
6680 		*attrset = 0;
6681 		return (puterrno4(error));
6682 	}
6683 
6684 	/* if parent dir is attrdir, set namedattr fh flag */
6685 	if (dvp->v_flag & V_XATTRDIR)
6686 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
6687 
6688 	if (cs->vp)
6689 		VN_RELE(cs->vp);
6690 
6691 	cs->vp = vp;
6692 
6693 	/*
6694 	 * if we did not create the file, we will need to check
6695 	 * the access bits on the file
6696 	 */
6697 
6698 	if (!created) {
6699 		if (setsize)
6700 			args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
6701 		status = check_open_access(args->share_access, cs, req);
6702 		if (status != NFS4_OK)
6703 			*attrset = 0;
6704 	}
6705 	return (status);
6706 }
6707 
6708 /*ARGSUSED*/
6709 static void
6710 rfs4_do_open(struct compound_state *cs, struct svc_req *req,
6711     rfs4_openowner_t *oo, delegreq_t deleg,
6712     uint32_t access, uint32_t deny,
6713     OPEN4res *resp, int deleg_cur)
6714 {
6715 	/* XXX Currently not using req  */
6716 	rfs4_state_t *sp;
6717 	rfs4_file_t *fp;
6718 	bool_t screate = TRUE;
6719 	bool_t fcreate = TRUE;
6720 	uint32_t open_a, share_a;
6721 	uint32_t open_d, share_d;
6722 	rfs4_deleg_state_t *dsp;
6723 	sysid_t sysid;
6724 	nfsstat4 status;
6725 	caller_context_t ct;
6726 	int fflags = 0;
6727 	int recall = 0;
6728 	int err;
6729 	int first_open;
6730 
6731 	/* get the file struct and hold a lock on it during initial open */
6732 	fp = rfs4_findfile_withlock(cs->vp, &cs->fh, &fcreate);
6733 	if (fp == NULL) {
6734 		resp->status = NFS4ERR_RESOURCE;
6735 		DTRACE_PROBE1(nfss__e__do__open1, nfsstat4, resp->status);
6736 		return;
6737 	}
6738 
6739 	sp = rfs4_findstate_by_owner_file(oo, fp, &screate);
6740 	if (sp == NULL) {
6741 		resp->status = NFS4ERR_RESOURCE;
6742 		DTRACE_PROBE1(nfss__e__do__open2, nfsstat4, resp->status);
6743 		/* No need to keep any reference */
6744 		rw_exit(&fp->rf_file_rwlock);
6745 		rfs4_file_rele(fp);
6746 		return;
6747 	}
6748 
6749 	/* try to get the sysid before continuing */
6750 	if ((status = rfs4_client_sysid(oo->ro_client, &sysid)) != NFS4_OK) {
6751 		resp->status = status;
6752 		rfs4_file_rele(fp);
6753 		/* Not a fully formed open; "close" it */
6754 		if (screate == TRUE)
6755 			rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6756 		rfs4_state_rele(sp);
6757 		return;
6758 	}
6759 
6760 	/* Calculate the fflags for this OPEN. */
6761 	if (access & OPEN4_SHARE_ACCESS_READ)
6762 		fflags |= FREAD;
6763 	if (access & OPEN4_SHARE_ACCESS_WRITE)
6764 		fflags |= FWRITE;
6765 
6766 	rfs4_dbe_lock(sp->rs_dbe);
6767 
6768 	/*
6769 	 * Calculate the new deny and access mode that this open is adding to
6770 	 * the file for this open owner;
6771 	 */
6772 	open_d = (deny & ~sp->rs_open_deny);
6773 	open_a = (access & ~sp->rs_open_access);
6774 
6775 	/*
6776 	 * Calculate the new share access and share deny modes that this open
6777 	 * is adding to the file for this open owner;
6778 	 */
6779 	share_a = (access & ~sp->rs_share_access);
6780 	share_d = (deny & ~sp->rs_share_deny);
6781 
6782 	first_open = (sp->rs_open_access & OPEN4_SHARE_ACCESS_BOTH) == 0;
6783 
6784 	/*
6785 	 * Check to see the client has already sent an open for this
6786 	 * open owner on this file with the same share/deny modes.
6787 	 * If so, we don't need to check for a conflict and we don't
6788 	 * need to add another shrlock.  If not, then we need to
6789 	 * check for conflicts in deny and access before checking for
6790 	 * conflicts in delegation.  We don't want to recall a
6791 	 * delegation based on an open that will eventually fail based
6792 	 * on shares modes.
6793 	 */
6794 
6795 	if (share_a || share_d) {
6796 		if ((err = rfs4_share(sp, access, deny)) != 0) {
6797 			rfs4_dbe_unlock(sp->rs_dbe);
6798 			resp->status = err;
6799 
6800 			rfs4_file_rele(fp);
6801 			/* Not a fully formed open; "close" it */
6802 			if (screate == TRUE)
6803 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6804 			rfs4_state_rele(sp);
6805 			return;
6806 		}
6807 	}
6808 
6809 	rfs4_dbe_lock(fp->rf_dbe);
6810 
6811 	/*
6812 	 * Check to see if this file is delegated and if so, if a
6813 	 * recall needs to be done.
6814 	 */
6815 	if (rfs4_check_recall(sp, access)) {
6816 		rfs4_dbe_unlock(fp->rf_dbe);
6817 		rfs4_dbe_unlock(sp->rs_dbe);
6818 		rfs4_recall_deleg(fp, FALSE, sp->rs_owner->ro_client);
6819 		delay(NFS4_DELEGATION_CONFLICT_DELAY);
6820 		rfs4_dbe_lock(sp->rs_dbe);
6821 
6822 		/* if state closed while lock was dropped */
6823 		if (sp->rs_closed) {
6824 			if (share_a || share_d)
6825 				(void) rfs4_unshare(sp);
6826 			rfs4_dbe_unlock(sp->rs_dbe);
6827 			rfs4_file_rele(fp);
6828 			/* Not a fully formed open; "close" it */
6829 			if (screate == TRUE)
6830 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6831 			rfs4_state_rele(sp);
6832 			resp->status = NFS4ERR_OLD_STATEID;
6833 			return;
6834 		}
6835 
6836 		rfs4_dbe_lock(fp->rf_dbe);
6837 		/* Let's see if the delegation was returned */
6838 		if (rfs4_check_recall(sp, access)) {
6839 			rfs4_dbe_unlock(fp->rf_dbe);
6840 			if (share_a || share_d)
6841 				(void) rfs4_unshare(sp);
6842 			rfs4_dbe_unlock(sp->rs_dbe);
6843 			rfs4_file_rele(fp);
6844 			rfs4_update_lease(sp->rs_owner->ro_client);
6845 
6846 			/* Not a fully formed open; "close" it */
6847 			if (screate == TRUE)
6848 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6849 			rfs4_state_rele(sp);
6850 			resp->status = NFS4ERR_DELAY;
6851 			return;
6852 		}
6853 	}
6854 	/*
6855 	 * the share check passed and any delegation conflict has been
6856 	 * taken care of, now call vop_open.
6857 	 * if this is the first open then call vop_open with fflags.
6858 	 * if not, call vn_open_upgrade with just the upgrade flags.
6859 	 *
6860 	 * if the file has been opened already, it will have the current
6861 	 * access mode in the state struct.  if it has no share access, then
6862 	 * this is a new open.
6863 	 *
6864 	 * However, if this is open with CLAIM_DLEGATE_CUR, then don't
6865 	 * call VOP_OPEN(), just do the open upgrade.
6866 	 */
6867 	if (first_open && !deleg_cur) {
6868 		ct.cc_sysid = sysid;
6869 		ct.cc_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
6870 		ct.cc_caller_id = nfs4_srv_caller_id;
6871 		ct.cc_flags = CC_DONTBLOCK;
6872 		err = VOP_OPEN(&cs->vp, fflags, cs->cr, &ct);
6873 		if (err) {
6874 			rfs4_dbe_unlock(fp->rf_dbe);
6875 			if (share_a || share_d)
6876 				(void) rfs4_unshare(sp);
6877 			rfs4_dbe_unlock(sp->rs_dbe);
6878 			rfs4_file_rele(fp);
6879 
6880 			/* Not a fully formed open; "close" it */
6881 			if (screate == TRUE)
6882 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6883 			rfs4_state_rele(sp);
6884 			/* check if a monitor detected a delegation conflict */
6885 			if (err == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
6886 				resp->status = NFS4ERR_DELAY;
6887 			else
6888 				resp->status = NFS4ERR_SERVERFAULT;
6889 			return;
6890 		}
6891 	} else { /* open upgrade */
6892 		/*
6893 		 * calculate the fflags for the new mode that is being added
6894 		 * by this upgrade.
6895 		 */
6896 		fflags = 0;
6897 		if (open_a & OPEN4_SHARE_ACCESS_READ)
6898 			fflags |= FREAD;
6899 		if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6900 			fflags |= FWRITE;
6901 		vn_open_upgrade(cs->vp, fflags);
6902 	}
6903 	sp->rs_open_access |= access;
6904 	sp->rs_open_deny |= deny;
6905 
6906 	if (open_d & OPEN4_SHARE_DENY_READ)
6907 		fp->rf_deny_read++;
6908 	if (open_d & OPEN4_SHARE_DENY_WRITE)
6909 		fp->rf_deny_write++;
6910 	fp->rf_share_deny |= deny;
6911 
6912 	if (open_a & OPEN4_SHARE_ACCESS_READ)
6913 		fp->rf_access_read++;
6914 	if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6915 		fp->rf_access_write++;
6916 	fp->rf_share_access |= access;
6917 
6918 	/*
6919 	 * Check for delegation here. if the deleg argument is not
6920 	 * DELEG_ANY, then this is a reclaim from a client and
6921 	 * we must honor the delegation requested. If necessary we can
6922 	 * set the recall flag.
6923 	 */
6924 
6925 	dsp = rfs4_grant_delegation(deleg, sp, &recall);
6926 
6927 	cs->deleg = (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE);
6928 
6929 	next_stateid(&sp->rs_stateid);
6930 
6931 	resp->stateid = sp->rs_stateid.stateid;
6932 
6933 	rfs4_dbe_unlock(fp->rf_dbe);
6934 	rfs4_dbe_unlock(sp->rs_dbe);
6935 
6936 	if (dsp) {
6937 		rfs4_set_deleg_response(dsp, &resp->delegation, NULL, recall);
6938 		rfs4_deleg_state_rele(dsp);
6939 	}
6940 
6941 	rfs4_file_rele(fp);
6942 	rfs4_state_rele(sp);
6943 
6944 	resp->status = NFS4_OK;
6945 }
6946 
6947 /*ARGSUSED*/
6948 static void
6949 rfs4_do_opennull(struct compound_state *cs, struct svc_req *req,
6950     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6951 {
6952 	change_info4 *cinfo = &resp->cinfo;
6953 	bitmap4 *attrset = &resp->attrset;
6954 
6955 	if (args->opentype == OPEN4_NOCREATE)
6956 		resp->status = rfs4_lookupfile(&args->open_claim4_u.file,
6957 		    req, cs, args->share_access, cinfo);
6958 	else {
6959 		/* inhibit delegation grants during exclusive create */
6960 
6961 		if (args->mode == EXCLUSIVE4)
6962 			rfs4_disable_delegation();
6963 
6964 		resp->status = rfs4_createfile(args, req, cs, cinfo, attrset,
6965 		    oo->ro_client->rc_clientid);
6966 	}
6967 
6968 	if (resp->status == NFS4_OK) {
6969 
6970 		/* cs->vp cs->fh now reference the desired file */
6971 
6972 		rfs4_do_open(cs, req, oo,
6973 		    oo->ro_need_confirm ? DELEG_NONE : DELEG_ANY,
6974 		    args->share_access, args->share_deny, resp, 0);
6975 
6976 		/*
6977 		 * If rfs4_createfile set attrset, we must
6978 		 * clear this attrset before the response is copied.
6979 		 */
6980 		if (resp->status != NFS4_OK && resp->attrset) {
6981 			resp->attrset = 0;
6982 		}
6983 	}
6984 	else
6985 		*cs->statusp = resp->status;
6986 
6987 	if (args->mode == EXCLUSIVE4)
6988 		rfs4_enable_delegation();
6989 }
6990 
6991 /*ARGSUSED*/
6992 static void
6993 rfs4_do_openprev(struct compound_state *cs, struct svc_req *req,
6994     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6995 {
6996 	change_info4 *cinfo = &resp->cinfo;
6997 	vattr_t va;
6998 	vtype_t v_type = cs->vp->v_type;
6999 	int error = 0;
7000 
7001 	/* Verify that we have a regular file */
7002 	if (v_type != VREG) {
7003 		if (v_type == VDIR)
7004 			resp->status = NFS4ERR_ISDIR;
7005 		else if (v_type == VLNK)
7006 			resp->status = NFS4ERR_SYMLINK;
7007 		else
7008 			resp->status = NFS4ERR_INVAL;
7009 		return;
7010 	}
7011 
7012 	va.va_mask = AT_MODE|AT_UID;
7013 	error = VOP_GETATTR(cs->vp, &va, 0, cs->cr, NULL);
7014 	if (error) {
7015 		resp->status = puterrno4(error);
7016 		return;
7017 	}
7018 
7019 	cs->mandlock = MANDLOCK(cs->vp, va.va_mode);
7020 
7021 	/*
7022 	 * Check if we have access to the file, Note the the file
7023 	 * could have originally been open UNCHECKED or GUARDED
7024 	 * with mode bits that will now fail, but there is nothing
7025 	 * we can really do about that except in the case that the
7026 	 * owner of the file is the one requesting the open.
7027 	 */
7028 	if (crgetuid(cs->cr) != va.va_uid) {
7029 		resp->status = check_open_access(args->share_access, cs, req);
7030 		if (resp->status != NFS4_OK) {
7031 			return;
7032 		}
7033 	}
7034 
7035 	/*
7036 	 * cinfo on a CLAIM_PREVIOUS is undefined, initialize to zero
7037 	 */
7038 	cinfo->before = 0;
7039 	cinfo->after = 0;
7040 	cinfo->atomic = FALSE;
7041 
7042 	rfs4_do_open(cs, req, oo,
7043 	    NFS4_DELEG4TYPE2REQTYPE(args->open_claim4_u.delegate_type),
7044 	    args->share_access, args->share_deny, resp, 0);
7045 }
7046 
7047 static void
7048 rfs4_do_opendelcur(struct compound_state *cs, struct svc_req *req,
7049     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7050 {
7051 	int error;
7052 	nfsstat4 status;
7053 	stateid4 stateid =
7054 	    args->open_claim4_u.delegate_cur_info.delegate_stateid;
7055 	rfs4_deleg_state_t *dsp;
7056 
7057 	/*
7058 	 * Find the state info from the stateid and confirm that the
7059 	 * file is delegated.  If the state openowner is the same as
7060 	 * the supplied openowner we're done. If not, get the file
7061 	 * info from the found state info. Use that file info to
7062 	 * create the state for this lock owner. Note solaris doen't
7063 	 * really need the pathname to find the file. We may want to
7064 	 * lookup the pathname and make sure that the vp exist and
7065 	 * matches the vp in the file structure. However it is
7066 	 * possible that the pathname nolonger exists (local process
7067 	 * unlinks the file), so this may not be that useful.
7068 	 */
7069 
7070 	status = rfs4_get_deleg_state(&stateid, &dsp);
7071 	if (status != NFS4_OK) {
7072 		resp->status = status;
7073 		return;
7074 	}
7075 
7076 	ASSERT(dsp->rds_finfo->rf_dinfo.rd_dtype != OPEN_DELEGATE_NONE);
7077 
7078 	/*
7079 	 * New lock owner, create state. Since this was probably called
7080 	 * in response to a CB_RECALL we set deleg to DELEG_NONE
7081 	 */
7082 
7083 	ASSERT(cs->vp != NULL);
7084 	VN_RELE(cs->vp);
7085 	VN_HOLD(dsp->rds_finfo->rf_vp);
7086 	cs->vp = dsp->rds_finfo->rf_vp;
7087 
7088 	if (error = makefh4(&cs->fh, cs->vp, cs->exi)) {
7089 		rfs4_deleg_state_rele(dsp);
7090 		*cs->statusp = resp->status = puterrno4(error);
7091 		return;
7092 	}
7093 
7094 	/* Mark progress for delegation returns */
7095 	dsp->rds_finfo->rf_dinfo.rd_time_lastwrite = gethrestime_sec();
7096 	rfs4_deleg_state_rele(dsp);
7097 	rfs4_do_open(cs, req, oo, DELEG_NONE,
7098 	    args->share_access, args->share_deny, resp, 1);
7099 }
7100 
7101 /*ARGSUSED*/
7102 static void
7103 rfs4_do_opendelprev(struct compound_state *cs, struct svc_req *req,
7104     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7105 {
7106 	/*
7107 	 * Lookup the pathname, it must already exist since this file
7108 	 * was delegated.
7109 	 *
7110 	 * Find the file and state info for this vp and open owner pair.
7111 	 *	check that they are in fact delegated.
7112 	 *	check that the state access and deny modes are the same.
7113 	 *
7114 	 * Return the delgation possibly seting the recall flag.
7115 	 */
7116 	rfs4_file_t *fp;
7117 	rfs4_state_t *sp;
7118 	bool_t create = FALSE;
7119 	bool_t dcreate = FALSE;
7120 	rfs4_deleg_state_t *dsp;
7121 	nfsace4 *ace;
7122 
7123 	/* Note we ignore oflags */
7124 	resp->status = rfs4_lookupfile(&args->open_claim4_u.file_delegate_prev,
7125 	    req, cs, args->share_access, &resp->cinfo);
7126 
7127 	if (resp->status != NFS4_OK) {
7128 		return;
7129 	}
7130 
7131 	/* get the file struct and hold a lock on it during initial open */
7132 	fp = rfs4_findfile_withlock(cs->vp, NULL, &create);
7133 	if (fp == NULL) {
7134 		resp->status = NFS4ERR_RESOURCE;
7135 		DTRACE_PROBE1(nfss__e__do_opendelprev1, nfsstat4, resp->status);
7136 		return;
7137 	}
7138 
7139 	sp = rfs4_findstate_by_owner_file(oo, fp, &create);
7140 	if (sp == NULL) {
7141 		resp->status = NFS4ERR_SERVERFAULT;
7142 		DTRACE_PROBE1(nfss__e__do_opendelprev2, nfsstat4, resp->status);
7143 		rw_exit(&fp->rf_file_rwlock);
7144 		rfs4_file_rele(fp);
7145 		return;
7146 	}
7147 
7148 	rfs4_dbe_lock(sp->rs_dbe);
7149 	rfs4_dbe_lock(fp->rf_dbe);
7150 	if (args->share_access != sp->rs_share_access ||
7151 	    args->share_deny != sp->rs_share_deny ||
7152 	    sp->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
7153 		NFS4_DEBUG(rfs4_debug,
7154 		    (CE_NOTE, "rfs4_do_opendelprev: state mixup"));
7155 		rfs4_dbe_unlock(fp->rf_dbe);
7156 		rfs4_dbe_unlock(sp->rs_dbe);
7157 		rfs4_file_rele(fp);
7158 		rfs4_state_rele(sp);
7159 		resp->status = NFS4ERR_SERVERFAULT;
7160 		return;
7161 	}
7162 	rfs4_dbe_unlock(fp->rf_dbe);
7163 	rfs4_dbe_unlock(sp->rs_dbe);
7164 
7165 	dsp = rfs4_finddeleg(sp, &dcreate);
7166 	if (dsp == NULL) {
7167 		rfs4_state_rele(sp);
7168 		rfs4_file_rele(fp);
7169 		resp->status = NFS4ERR_SERVERFAULT;
7170 		return;
7171 	}
7172 
7173 	next_stateid(&sp->rs_stateid);
7174 
7175 	resp->stateid = sp->rs_stateid.stateid;
7176 
7177 	resp->delegation.delegation_type = dsp->rds_dtype;
7178 
7179 	if (dsp->rds_dtype == OPEN_DELEGATE_READ) {
7180 		open_read_delegation4 *rv =
7181 		    &resp->delegation.open_delegation4_u.read;
7182 
7183 		rv->stateid = dsp->rds_delegid.stateid;
7184 		rv->recall = FALSE; /* no policy in place to set to TRUE */
7185 		ace = &rv->permissions;
7186 	} else {
7187 		open_write_delegation4 *rv =
7188 		    &resp->delegation.open_delegation4_u.write;
7189 
7190 		rv->stateid = dsp->rds_delegid.stateid;
7191 		rv->recall = FALSE;  /* no policy in place to set to TRUE */
7192 		ace = &rv->permissions;
7193 		rv->space_limit.limitby = NFS_LIMIT_SIZE;
7194 		rv->space_limit.nfs_space_limit4_u.filesize = UINT64_MAX;
7195 	}
7196 
7197 	/* XXX For now */
7198 	ace->type = ACE4_ACCESS_ALLOWED_ACE_TYPE;
7199 	ace->flag = 0;
7200 	ace->access_mask = 0;
7201 	ace->who.utf8string_len = 0;
7202 	ace->who.utf8string_val = 0;
7203 
7204 	rfs4_deleg_state_rele(dsp);
7205 	rfs4_state_rele(sp);
7206 	rfs4_file_rele(fp);
7207 }
7208 
7209 typedef enum {
7210 	NFS4_CHKSEQ_OKAY = 0,
7211 	NFS4_CHKSEQ_REPLAY = 1,
7212 	NFS4_CHKSEQ_BAD = 2
7213 } rfs4_chkseq_t;
7214 
7215 /*
7216  * Generic function for sequence number checks.
7217  */
7218 static rfs4_chkseq_t
7219 rfs4_check_seqid(seqid4 seqid, nfs_resop4 *lastop,
7220     seqid4 rqst_seq, nfs_resop4 *resop, bool_t copyres)
7221 {
7222 	/* Same sequence ids and matching operations? */
7223 	if (seqid == rqst_seq && resop->resop == lastop->resop) {
7224 		if (copyres == TRUE) {
7225 			rfs4_free_reply(resop);
7226 			rfs4_copy_reply(resop, lastop);
7227 		}
7228 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
7229 		    "Replayed SEQID %d\n", seqid));
7230 		return (NFS4_CHKSEQ_REPLAY);
7231 	}
7232 
7233 	/* If the incoming sequence is not the next expected then it is bad */
7234 	if (rqst_seq != seqid + 1) {
7235 		if (rqst_seq == seqid) {
7236 			NFS4_DEBUG(rfs4_debug,
7237 			    (CE_NOTE, "BAD SEQID: Replayed sequence id "
7238 			    "but last op was %d current op is %d\n",
7239 			    lastop->resop, resop->resop));
7240 			return (NFS4_CHKSEQ_BAD);
7241 		}
7242 		NFS4_DEBUG(rfs4_debug,
7243 		    (CE_NOTE, "BAD SEQID: got %u expecting %u\n",
7244 		    rqst_seq, seqid));
7245 		return (NFS4_CHKSEQ_BAD);
7246 	}
7247 
7248 	/* Everything okay -- next expected */
7249 	return (NFS4_CHKSEQ_OKAY);
7250 }
7251 
7252 
7253 static rfs4_chkseq_t
7254 rfs4_check_open_seqid(seqid4 seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7255 {
7256 	rfs4_chkseq_t rc;
7257 
7258 	rfs4_dbe_lock(op->ro_dbe);
7259 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply, seqid, resop,
7260 	    TRUE);
7261 	rfs4_dbe_unlock(op->ro_dbe);
7262 
7263 	if (rc == NFS4_CHKSEQ_OKAY)
7264 		rfs4_update_lease(op->ro_client);
7265 
7266 	return (rc);
7267 }
7268 
7269 static rfs4_chkseq_t
7270 rfs4_check_olo_seqid(seqid4 olo_seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7271 {
7272 	rfs4_chkseq_t rc;
7273 
7274 	rfs4_dbe_lock(op->ro_dbe);
7275 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply,
7276 	    olo_seqid, resop, FALSE);
7277 	rfs4_dbe_unlock(op->ro_dbe);
7278 
7279 	return (rc);
7280 }
7281 
7282 static rfs4_chkseq_t
7283 rfs4_check_lock_seqid(seqid4 seqid, rfs4_lo_state_t *lsp, nfs_resop4 *resop)
7284 {
7285 	rfs4_chkseq_t rc = NFS4_CHKSEQ_OKAY;
7286 
7287 	rfs4_dbe_lock(lsp->rls_dbe);
7288 	if (!lsp->rls_skip_seqid_check)
7289 		rc = rfs4_check_seqid(lsp->rls_seqid, &lsp->rls_reply, seqid,
7290 		    resop, TRUE);
7291 	rfs4_dbe_unlock(lsp->rls_dbe);
7292 
7293 	return (rc);
7294 }
7295 
7296 static void
7297 rfs4_op_open(nfs_argop4 *argop, nfs_resop4 *resop,
7298     struct svc_req *req, struct compound_state *cs)
7299 {
7300 	OPEN4args *args = &argop->nfs_argop4_u.opopen;
7301 	OPEN4res *resp = &resop->nfs_resop4_u.opopen;
7302 	open_owner4 *owner = &args->owner;
7303 	open_claim_type4 claim = args->claim;
7304 	rfs4_client_t *cp;
7305 	rfs4_openowner_t *oo;
7306 	bool_t create;
7307 	bool_t replay = FALSE;
7308 	int can_reclaim;
7309 
7310 	DTRACE_NFSV4_2(op__open__start, struct compound_state *, cs,
7311 	    OPEN4args *, args);
7312 
7313 	if (cs->vp == NULL) {
7314 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7315 		goto end;
7316 	}
7317 
7318 	/*
7319 	 * Need to check clientid and lease expiration first based on
7320 	 * error ordering and incrementing sequence id.
7321 	 */
7322 	cp = rfs4_findclient_by_id(owner->clientid, FALSE);
7323 	if (cp == NULL) {
7324 		*cs->statusp = resp->status =
7325 		    rfs4_check_clientid(&owner->clientid, 0);
7326 		goto end;
7327 	}
7328 
7329 	if (rfs4_lease_expired(cp)) {
7330 		rfs4_client_close(cp);
7331 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7332 		goto end;
7333 	}
7334 	can_reclaim = cp->rc_can_reclaim;
7335 
7336 	/*
7337 	 * Find the open_owner for use from this point forward.  Take
7338 	 * care in updating the sequence id based on the type of error
7339 	 * being returned.
7340 	 */
7341 retry:
7342 	create = TRUE;
7343 	oo = rfs4_findopenowner(owner, &create, args->seqid);
7344 	if (oo == NULL) {
7345 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
7346 		rfs4_client_rele(cp);
7347 		goto end;
7348 	}
7349 
7350 	/* Hold off access to the sequence space while the open is done */
7351 	rfs4_sw_enter(&oo->ro_sw);
7352 
7353 	/*
7354 	 * If the open_owner existed before at the server, then check
7355 	 * the sequence id.
7356 	 */
7357 	if (!create && !oo->ro_postpone_confirm) {
7358 		switch (rfs4_check_open_seqid(args->seqid, oo, resop)) {
7359 		case NFS4_CHKSEQ_BAD:
7360 			if ((args->seqid > oo->ro_open_seqid) &&
7361 			    oo->ro_need_confirm) {
7362 				rfs4_free_opens(oo, TRUE, FALSE);
7363 				rfs4_sw_exit(&oo->ro_sw);
7364 				rfs4_openowner_rele(oo);
7365 				goto retry;
7366 			}
7367 			resp->status = NFS4ERR_BAD_SEQID;
7368 			goto out;
7369 		case NFS4_CHKSEQ_REPLAY: /* replay of previous request */
7370 			replay = TRUE;
7371 			goto out;
7372 		default:
7373 			break;
7374 		}
7375 
7376 		/*
7377 		 * Sequence was ok and open owner exists
7378 		 * check to see if we have yet to see an
7379 		 * open_confirm.
7380 		 */
7381 		if (oo->ro_need_confirm) {
7382 			rfs4_free_opens(oo, TRUE, FALSE);
7383 			rfs4_sw_exit(&oo->ro_sw);
7384 			rfs4_openowner_rele(oo);
7385 			goto retry;
7386 		}
7387 	}
7388 	/* Grace only applies to regular-type OPENs */
7389 	if (rfs4_clnt_in_grace(cp) &&
7390 	    (claim == CLAIM_NULL || claim == CLAIM_DELEGATE_CUR)) {
7391 		*cs->statusp = resp->status = NFS4ERR_GRACE;
7392 		goto out;
7393 	}
7394 
7395 	/*
7396 	 * If previous state at the server existed then can_reclaim
7397 	 * will be set. If not reply NFS4ERR_NO_GRACE to the
7398 	 * client.
7399 	 */
7400 	if (rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS && !can_reclaim) {
7401 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7402 		goto out;
7403 	}
7404 
7405 
7406 	/*
7407 	 * Reject the open if the client has missed the grace period
7408 	 */
7409 	if (!rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS) {
7410 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7411 		goto out;
7412 	}
7413 
7414 	/* Couple of up-front bookkeeping items */
7415 	if (oo->ro_need_confirm) {
7416 		/*
7417 		 * If this is a reclaim OPEN then we should not ask
7418 		 * for a confirmation of the open_owner per the
7419 		 * protocol specification.
7420 		 */
7421 		if (claim == CLAIM_PREVIOUS)
7422 			oo->ro_need_confirm = FALSE;
7423 		else
7424 			resp->rflags |= OPEN4_RESULT_CONFIRM;
7425 	}
7426 	resp->rflags |= OPEN4_RESULT_LOCKTYPE_POSIX;
7427 
7428 	/*
7429 	 * If there is an unshared filesystem mounted on this vnode,
7430 	 * do not allow to open/create in this directory.
7431 	 */
7432 	if (vn_ismntpt(cs->vp)) {
7433 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
7434 		goto out;
7435 	}
7436 
7437 	/*
7438 	 * access must READ, WRITE, or BOTH.  No access is invalid.
7439 	 * deny can be READ, WRITE, BOTH, or NONE.
7440 	 * bits not defined for access/deny are invalid.
7441 	 */
7442 	if (! (args->share_access & OPEN4_SHARE_ACCESS_BOTH) ||
7443 	    (args->share_access & ~OPEN4_SHARE_ACCESS_BOTH) ||
7444 	    (args->share_deny & ~OPEN4_SHARE_DENY_BOTH)) {
7445 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7446 		goto out;
7447 	}
7448 
7449 
7450 	/*
7451 	 * make sure attrset is zero before response is built.
7452 	 */
7453 	resp->attrset = 0;
7454 
7455 	switch (claim) {
7456 	case CLAIM_NULL:
7457 		rfs4_do_opennull(cs, req, args, oo, resp);
7458 		break;
7459 	case CLAIM_PREVIOUS:
7460 		rfs4_do_openprev(cs, req, args, oo, resp);
7461 		break;
7462 	case CLAIM_DELEGATE_CUR:
7463 		rfs4_do_opendelcur(cs, req, args, oo, resp);
7464 		break;
7465 	case CLAIM_DELEGATE_PREV:
7466 		rfs4_do_opendelprev(cs, req, args, oo, resp);
7467 		break;
7468 	default:
7469 		resp->status = NFS4ERR_INVAL;
7470 		break;
7471 	}
7472 
7473 out:
7474 	rfs4_client_rele(cp);
7475 
7476 	/* Catch sequence id handling here to make it a little easier */
7477 	switch (resp->status) {
7478 	case NFS4ERR_BADXDR:
7479 	case NFS4ERR_BAD_SEQID:
7480 	case NFS4ERR_BAD_STATEID:
7481 	case NFS4ERR_NOFILEHANDLE:
7482 	case NFS4ERR_RESOURCE:
7483 	case NFS4ERR_STALE_CLIENTID:
7484 	case NFS4ERR_STALE_STATEID:
7485 		/*
7486 		 * The protocol states that if any of these errors are
7487 		 * being returned, the sequence id should not be
7488 		 * incremented.  Any other return requires an
7489 		 * increment.
7490 		 */
7491 		break;
7492 	default:
7493 		/* Always update the lease in this case */
7494 		rfs4_update_lease(oo->ro_client);
7495 
7496 		/* Regular response - copy the result */
7497 		if (!replay)
7498 			rfs4_update_open_resp(oo, resop, &cs->fh);
7499 
7500 		/*
7501 		 * REPLAY case: Only if the previous response was OK
7502 		 * do we copy the filehandle.  If not OK, no
7503 		 * filehandle to copy.
7504 		 */
7505 		if (replay == TRUE &&
7506 		    resp->status == NFS4_OK &&
7507 		    oo->ro_reply_fh.nfs_fh4_val) {
7508 			/*
7509 			 * If this is a replay, we must restore the
7510 			 * current filehandle/vp to that of what was
7511 			 * returned originally.  Try our best to do
7512 			 * it.
7513 			 */
7514 			nfs_fh4_fmt_t *fh_fmtp =
7515 			    (nfs_fh4_fmt_t *)oo->ro_reply_fh.nfs_fh4_val;
7516 
7517 			cs->exi = checkexport4(&fh_fmtp->fh4_fsid,
7518 			    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
7519 
7520 			if (cs->exi == NULL) {
7521 				resp->status = NFS4ERR_STALE;
7522 				goto finish;
7523 			}
7524 
7525 			VN_RELE(cs->vp);
7526 
7527 			cs->vp = nfs4_fhtovp(&oo->ro_reply_fh, cs->exi,
7528 			    &resp->status);
7529 
7530 			if (cs->vp == NULL)
7531 				goto finish;
7532 
7533 			nfs_fh4_copy(&oo->ro_reply_fh, &cs->fh);
7534 		}
7535 
7536 		/*
7537 		 * If this was a replay, no need to update the
7538 		 * sequence id. If the open_owner was not created on
7539 		 * this pass, then update.  The first use of an
7540 		 * open_owner will not bump the sequence id.
7541 		 */
7542 		if (replay == FALSE && !create)
7543 			rfs4_update_open_sequence(oo);
7544 		/*
7545 		 * If the client is receiving an error and the
7546 		 * open_owner needs to be confirmed, there is no way
7547 		 * to notify the client of this fact ignoring the fact
7548 		 * that the server has no method of returning a
7549 		 * stateid to confirm.  Therefore, the server needs to
7550 		 * mark this open_owner in a way as to avoid the
7551 		 * sequence id checking the next time the client uses
7552 		 * this open_owner.
7553 		 */
7554 		if (resp->status != NFS4_OK && oo->ro_need_confirm)
7555 			oo->ro_postpone_confirm = TRUE;
7556 		/*
7557 		 * If OK response then clear the postpone flag and
7558 		 * reset the sequence id to keep in sync with the
7559 		 * client.
7560 		 */
7561 		if (resp->status == NFS4_OK && oo->ro_postpone_confirm) {
7562 			oo->ro_postpone_confirm = FALSE;
7563 			oo->ro_open_seqid = args->seqid;
7564 		}
7565 		break;
7566 	}
7567 
7568 finish:
7569 	*cs->statusp = resp->status;
7570 
7571 	rfs4_sw_exit(&oo->ro_sw);
7572 	rfs4_openowner_rele(oo);
7573 
7574 end:
7575 	DTRACE_NFSV4_2(op__open__done, struct compound_state *, cs,
7576 	    OPEN4res *, resp);
7577 }
7578 
7579 /*ARGSUSED*/
7580 void
7581 rfs4_op_open_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
7582     struct svc_req *req, struct compound_state *cs)
7583 {
7584 	OPEN_CONFIRM4args *args = &argop->nfs_argop4_u.opopen_confirm;
7585 	OPEN_CONFIRM4res *resp = &resop->nfs_resop4_u.opopen_confirm;
7586 	rfs4_state_t *sp;
7587 	nfsstat4 status;
7588 
7589 	DTRACE_NFSV4_2(op__open__confirm__start, struct compound_state *, cs,
7590 	    OPEN_CONFIRM4args *, args);
7591 
7592 	if (cs->vp == NULL) {
7593 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7594 		goto out;
7595 	}
7596 
7597 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7598 	if (status != NFS4_OK) {
7599 		*cs->statusp = resp->status = status;
7600 		goto out;
7601 	}
7602 
7603 	/* Ensure specified filehandle matches */
7604 	if (cs->vp != sp->rs_finfo->rf_vp) {
7605 		rfs4_state_rele(sp);
7606 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7607 		goto out;
7608 	}
7609 
7610 	/* hold off other access to open_owner while we tinker */
7611 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7612 
7613 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7614 	case NFS4_CHECK_STATEID_OKAY:
7615 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7616 		    resop) != 0) {
7617 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7618 			break;
7619 		}
7620 		/*
7621 		 * If it is the appropriate stateid and determined to
7622 		 * be "OKAY" then this means that the stateid does not
7623 		 * need to be confirmed and the client is in error for
7624 		 * sending an OPEN_CONFIRM.
7625 		 */
7626 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7627 		break;
7628 	case NFS4_CHECK_STATEID_OLD:
7629 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7630 		break;
7631 	case NFS4_CHECK_STATEID_BAD:
7632 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7633 		break;
7634 	case NFS4_CHECK_STATEID_EXPIRED:
7635 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7636 		break;
7637 	case NFS4_CHECK_STATEID_CLOSED:
7638 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7639 		break;
7640 	case NFS4_CHECK_STATEID_REPLAY:
7641 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7642 		    resop)) {
7643 		case NFS4_CHKSEQ_OKAY:
7644 			/*
7645 			 * This is replayed stateid; if seqid matches
7646 			 * next expected, then client is using wrong seqid.
7647 			 */
7648 			/* fall through */
7649 		case NFS4_CHKSEQ_BAD:
7650 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7651 			break;
7652 		case NFS4_CHKSEQ_REPLAY:
7653 			/*
7654 			 * Note this case is the duplicate case so
7655 			 * resp->status is already set.
7656 			 */
7657 			*cs->statusp = resp->status;
7658 			rfs4_update_lease(sp->rs_owner->ro_client);
7659 			break;
7660 		}
7661 		break;
7662 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7663 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7664 		    resop) != NFS4_CHKSEQ_OKAY) {
7665 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7666 			break;
7667 		}
7668 		*cs->statusp = resp->status = NFS4_OK;
7669 
7670 		next_stateid(&sp->rs_stateid);
7671 		resp->open_stateid = sp->rs_stateid.stateid;
7672 		sp->rs_owner->ro_need_confirm = FALSE;
7673 		rfs4_update_lease(sp->rs_owner->ro_client);
7674 		rfs4_update_open_sequence(sp->rs_owner);
7675 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7676 		break;
7677 	default:
7678 		ASSERT(FALSE);
7679 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7680 		break;
7681 	}
7682 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7683 	rfs4_state_rele(sp);
7684 
7685 out:
7686 	DTRACE_NFSV4_2(op__open__confirm__done, struct compound_state *, cs,
7687 	    OPEN_CONFIRM4res *, resp);
7688 }
7689 
7690 /*ARGSUSED*/
7691 void
7692 rfs4_op_open_downgrade(nfs_argop4 *argop, nfs_resop4 *resop,
7693     struct svc_req *req, struct compound_state *cs)
7694 {
7695 	OPEN_DOWNGRADE4args *args = &argop->nfs_argop4_u.opopen_downgrade;
7696 	OPEN_DOWNGRADE4res *resp = &resop->nfs_resop4_u.opopen_downgrade;
7697 	uint32_t access = args->share_access;
7698 	uint32_t deny = args->share_deny;
7699 	nfsstat4 status;
7700 	rfs4_state_t *sp;
7701 	rfs4_file_t *fp;
7702 	int fflags = 0;
7703 
7704 	DTRACE_NFSV4_2(op__open__downgrade__start, struct compound_state *, cs,
7705 	    OPEN_DOWNGRADE4args *, args);
7706 
7707 	if (cs->vp == NULL) {
7708 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7709 		goto out;
7710 	}
7711 
7712 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7713 	if (status != NFS4_OK) {
7714 		*cs->statusp = resp->status = status;
7715 		goto out;
7716 	}
7717 
7718 	/* Ensure specified filehandle matches */
7719 	if (cs->vp != sp->rs_finfo->rf_vp) {
7720 		rfs4_state_rele(sp);
7721 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7722 		goto out;
7723 	}
7724 
7725 	/* hold off other access to open_owner while we tinker */
7726 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7727 
7728 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7729 	case NFS4_CHECK_STATEID_OKAY:
7730 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7731 		    resop) != NFS4_CHKSEQ_OKAY) {
7732 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7733 			goto end;
7734 		}
7735 		break;
7736 	case NFS4_CHECK_STATEID_OLD:
7737 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7738 		goto end;
7739 	case NFS4_CHECK_STATEID_BAD:
7740 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7741 		goto end;
7742 	case NFS4_CHECK_STATEID_EXPIRED:
7743 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7744 		goto end;
7745 	case NFS4_CHECK_STATEID_CLOSED:
7746 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7747 		goto end;
7748 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7749 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7750 		goto end;
7751 	case NFS4_CHECK_STATEID_REPLAY:
7752 		/* Check the sequence id for the open owner */
7753 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7754 		    resop)) {
7755 		case NFS4_CHKSEQ_OKAY:
7756 			/*
7757 			 * This is replayed stateid; if seqid matches
7758 			 * next expected, then client is using wrong seqid.
7759 			 */
7760 			/* fall through */
7761 		case NFS4_CHKSEQ_BAD:
7762 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7763 			goto end;
7764 		case NFS4_CHKSEQ_REPLAY:
7765 			/*
7766 			 * Note this case is the duplicate case so
7767 			 * resp->status is already set.
7768 			 */
7769 			*cs->statusp = resp->status;
7770 			rfs4_update_lease(sp->rs_owner->ro_client);
7771 			goto end;
7772 		}
7773 		break;
7774 	default:
7775 		ASSERT(FALSE);
7776 		break;
7777 	}
7778 
7779 	rfs4_dbe_lock(sp->rs_dbe);
7780 	/*
7781 	 * Check that the new access modes and deny modes are valid.
7782 	 * Check that no invalid bits are set.
7783 	 */
7784 	if ((access & ~(OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) ||
7785 	    (deny & ~(OPEN4_SHARE_DENY_READ | OPEN4_SHARE_DENY_WRITE))) {
7786 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7787 		rfs4_update_open_sequence(sp->rs_owner);
7788 		rfs4_dbe_unlock(sp->rs_dbe);
7789 		goto end;
7790 	}
7791 
7792 	/*
7793 	 * The new modes must be a subset of the current modes and
7794 	 * the access must specify at least one mode. To test that
7795 	 * the new mode is a subset of the current modes we bitwise
7796 	 * AND them together and check that the result equals the new
7797 	 * mode. For example:
7798 	 * New mode, access == R and current mode, sp->rs_open_access  == RW
7799 	 * access & sp->rs_open_access == R == access, so the new access mode
7800 	 * is valid. Consider access == RW, sp->rs_open_access = R
7801 	 * access & sp->rs_open_access == R != access, so the new access mode
7802 	 * is invalid.
7803 	 */
7804 	if ((access & sp->rs_open_access) != access ||
7805 	    (deny & sp->rs_open_deny) != deny ||
7806 	    (access &
7807 	    (OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) == 0) {
7808 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7809 		rfs4_update_open_sequence(sp->rs_owner);
7810 		rfs4_dbe_unlock(sp->rs_dbe);
7811 		goto end;
7812 	}
7813 
7814 	/*
7815 	 * Release any share locks associated with this stateID.
7816 	 * Strictly speaking, this violates the spec because the
7817 	 * spec effectively requires that open downgrade be atomic.
7818 	 * At present, fs_shrlock does not have this capability.
7819 	 */
7820 	(void) rfs4_unshare(sp);
7821 
7822 	status = rfs4_share(sp, access, deny);
7823 	if (status != NFS4_OK) {
7824 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7825 		rfs4_update_open_sequence(sp->rs_owner);
7826 		rfs4_dbe_unlock(sp->rs_dbe);
7827 		goto end;
7828 	}
7829 
7830 	fp = sp->rs_finfo;
7831 	rfs4_dbe_lock(fp->rf_dbe);
7832 
7833 	/*
7834 	 * If the current mode has deny read and the new mode
7835 	 * does not, decrement the number of deny read mode bits
7836 	 * and if it goes to zero turn off the deny read bit
7837 	 * on the file.
7838 	 */
7839 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_READ) &&
7840 	    (deny & OPEN4_SHARE_DENY_READ) == 0) {
7841 		fp->rf_deny_read--;
7842 		if (fp->rf_deny_read == 0)
7843 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
7844 	}
7845 
7846 	/*
7847 	 * If the current mode has deny write and the new mode
7848 	 * does not, decrement the number of deny write mode bits
7849 	 * and if it goes to zero turn off the deny write bit
7850 	 * on the file.
7851 	 */
7852 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) &&
7853 	    (deny & OPEN4_SHARE_DENY_WRITE) == 0) {
7854 		fp->rf_deny_write--;
7855 		if (fp->rf_deny_write == 0)
7856 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
7857 	}
7858 
7859 	/*
7860 	 * If the current mode has access read and the new mode
7861 	 * does not, decrement the number of access read mode bits
7862 	 * and if it goes to zero turn off the access read bit
7863 	 * on the file.  set fflags to FREAD for the call to
7864 	 * vn_open_downgrade().
7865 	 */
7866 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) &&
7867 	    (access & OPEN4_SHARE_ACCESS_READ) == 0) {
7868 		fp->rf_access_read--;
7869 		if (fp->rf_access_read == 0)
7870 			fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
7871 		fflags |= FREAD;
7872 	}
7873 
7874 	/*
7875 	 * If the current mode has access write and the new mode
7876 	 * does not, decrement the number of access write mode bits
7877 	 * and if it goes to zero turn off the access write bit
7878 	 * on the file.  set fflags to FWRITE for the call to
7879 	 * vn_open_downgrade().
7880 	 */
7881 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) &&
7882 	    (access & OPEN4_SHARE_ACCESS_WRITE) == 0) {
7883 		fp->rf_access_write--;
7884 		if (fp->rf_access_write == 0)
7885 			fp->rf_share_deny &= ~OPEN4_SHARE_ACCESS_WRITE;
7886 		fflags |= FWRITE;
7887 	}
7888 
7889 	/* Check that the file is still accessible */
7890 	ASSERT(fp->rf_share_access);
7891 
7892 	rfs4_dbe_unlock(fp->rf_dbe);
7893 
7894 	/* now set the new open access and deny modes */
7895 	sp->rs_open_access = access;
7896 	sp->rs_open_deny = deny;
7897 
7898 	/*
7899 	 * we successfully downgraded the share lock, now we need to downgrade
7900 	 * the open. it is possible that the downgrade was only for a deny
7901 	 * mode and we have nothing else to do.
7902 	 */
7903 	if ((fflags & (FREAD|FWRITE)) != 0)
7904 		vn_open_downgrade(cs->vp, fflags);
7905 
7906 	/* Update the stateid */
7907 	next_stateid(&sp->rs_stateid);
7908 	resp->open_stateid = sp->rs_stateid.stateid;
7909 
7910 	rfs4_dbe_unlock(sp->rs_dbe);
7911 
7912 	*cs->statusp = resp->status = NFS4_OK;
7913 	/* Update the lease */
7914 	rfs4_update_lease(sp->rs_owner->ro_client);
7915 	/* And the sequence */
7916 	rfs4_update_open_sequence(sp->rs_owner);
7917 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7918 
7919 end:
7920 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7921 	rfs4_state_rele(sp);
7922 out:
7923 	DTRACE_NFSV4_2(op__open__downgrade__done, struct compound_state *, cs,
7924 	    OPEN_DOWNGRADE4res *, resp);
7925 }
7926 
7927 /*
7928  * The logic behind this function is detailed in the NFSv4 RFC in the
7929  * SETCLIENTID operation description under IMPLEMENTATION.  Refer to
7930  * that section for explicit guidance to server behavior for
7931  * SETCLIENTID.
7932  */
7933 void
7934 rfs4_op_setclientid(nfs_argop4 *argop, nfs_resop4 *resop,
7935     struct svc_req *req, struct compound_state *cs)
7936 {
7937 	SETCLIENTID4args *args = &argop->nfs_argop4_u.opsetclientid;
7938 	SETCLIENTID4res *res = &resop->nfs_resop4_u.opsetclientid;
7939 	rfs4_client_t *cp, *newcp, *cp_confirmed, *cp_unconfirmed;
7940 	rfs4_clntip_t *ci;
7941 	bool_t create;
7942 	char *addr, *netid;
7943 	int len;
7944 
7945 	DTRACE_NFSV4_2(op__setclientid__start, struct compound_state *, cs,
7946 	    SETCLIENTID4args *, args);
7947 retry:
7948 	newcp = cp_confirmed = cp_unconfirmed = NULL;
7949 
7950 	/*
7951 	 * Save the caller's IP address
7952 	 */
7953 	args->client.cl_addr =
7954 	    (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
7955 
7956 	/*
7957 	 * Record if it is a Solaris client that cannot handle referrals.
7958 	 */
7959 	if (strstr(args->client.id_val, "Solaris") &&
7960 	    !strstr(args->client.id_val, "+referrals")) {
7961 		/* Add a "yes, it's downrev" record */
7962 		create = TRUE;
7963 		ci = rfs4_find_clntip(args->client.cl_addr, &create);
7964 		ASSERT(ci != NULL);
7965 		rfs4_dbe_rele(ci->ri_dbe);
7966 	} else {
7967 		/* Remove any previous record */
7968 		rfs4_invalidate_clntip(args->client.cl_addr);
7969 	}
7970 
7971 	/*
7972 	 * In search of an EXISTING client matching the incoming
7973 	 * request to establish a new client identifier at the server
7974 	 */
7975 	create = TRUE;
7976 	cp = rfs4_findclient(&args->client, &create, NULL);
7977 
7978 	/* Should never happen */
7979 	ASSERT(cp != NULL);
7980 
7981 	if (cp == NULL) {
7982 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
7983 		goto out;
7984 	}
7985 
7986 	/*
7987 	 * Easiest case. Client identifier is newly created and is
7988 	 * unconfirmed.  Also note that for this case, no other
7989 	 * entries exist for the client identifier.  Nothing else to
7990 	 * check.  Just setup the response and respond.
7991 	 */
7992 	if (create) {
7993 		*cs->statusp = res->status = NFS4_OK;
7994 		res->SETCLIENTID4res_u.resok4.clientid = cp->rc_clientid;
7995 		res->SETCLIENTID4res_u.resok4.setclientid_confirm =
7996 		    cp->rc_confirm_verf;
7997 		/* Setup callback information; CB_NULL confirmation later */
7998 		rfs4_client_setcb(cp, &args->callback, args->callback_ident);
7999 
8000 		rfs4_client_rele(cp);
8001 		goto out;
8002 	}
8003 
8004 	/*
8005 	 * An existing, confirmed client may exist but it may not have
8006 	 * been active for at least one lease period.  If so, then
8007 	 * "close" the client and create a new client identifier
8008 	 */
8009 	if (rfs4_lease_expired(cp)) {
8010 		rfs4_client_close(cp);
8011 		goto retry;
8012 	}
8013 
8014 	if (cp->rc_need_confirm == TRUE)
8015 		cp_unconfirmed = cp;
8016 	else
8017 		cp_confirmed = cp;
8018 
8019 	cp = NULL;
8020 
8021 	/*
8022 	 * We have a confirmed client, now check for an
8023 	 * unconfimred entry
8024 	 */
8025 	if (cp_confirmed) {
8026 		/* If creds don't match then client identifier is inuse */
8027 		if (!creds_ok(cp_confirmed->rc_cr_set, req, cs)) {
8028 			rfs4_cbinfo_t *cbp;
8029 			/*
8030 			 * Some one else has established this client
8031 			 * id. Try and say * who they are. We will use
8032 			 * the call back address supplied by * the
8033 			 * first client.
8034 			 */
8035 			*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8036 
8037 			addr = netid = NULL;
8038 
8039 			cbp = &cp_confirmed->rc_cbinfo;
8040 			if (cbp->cb_callback.cb_location.r_addr &&
8041 			    cbp->cb_callback.cb_location.r_netid) {
8042 				cb_client4 *cbcp = &cbp->cb_callback;
8043 
8044 				len = strlen(cbcp->cb_location.r_addr)+1;
8045 				addr = kmem_alloc(len, KM_SLEEP);
8046 				bcopy(cbcp->cb_location.r_addr, addr, len);
8047 				len = strlen(cbcp->cb_location.r_netid)+1;
8048 				netid = kmem_alloc(len, KM_SLEEP);
8049 				bcopy(cbcp->cb_location.r_netid, netid, len);
8050 			}
8051 
8052 			res->SETCLIENTID4res_u.client_using.r_addr = addr;
8053 			res->SETCLIENTID4res_u.client_using.r_netid = netid;
8054 
8055 			rfs4_client_rele(cp_confirmed);
8056 		}
8057 
8058 		/*
8059 		 * Confirmed, creds match, and verifier matches; must
8060 		 * be an update of the callback info
8061 		 */
8062 		if (cp_confirmed->rc_nfs_client.verifier ==
8063 		    args->client.verifier) {
8064 			/* Setup callback information */
8065 			rfs4_client_setcb(cp_confirmed, &args->callback,
8066 			    args->callback_ident);
8067 
8068 			/* everything okay -- move ahead */
8069 			*cs->statusp = res->status = NFS4_OK;
8070 			res->SETCLIENTID4res_u.resok4.clientid =
8071 			    cp_confirmed->rc_clientid;
8072 
8073 			/* update the confirm_verifier and return it */
8074 			rfs4_client_scv_next(cp_confirmed);
8075 			res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8076 			    cp_confirmed->rc_confirm_verf;
8077 
8078 			rfs4_client_rele(cp_confirmed);
8079 			goto out;
8080 		}
8081 
8082 		/*
8083 		 * Creds match but the verifier doesn't.  Must search
8084 		 * for an unconfirmed client that would be replaced by
8085 		 * this request.
8086 		 */
8087 		create = FALSE;
8088 		cp_unconfirmed = rfs4_findclient(&args->client, &create,
8089 		    cp_confirmed);
8090 	}
8091 
8092 	/*
8093 	 * At this point, we have taken care of the brand new client
8094 	 * struct, INUSE case, update of an existing, and confirmed
8095 	 * client struct.
8096 	 */
8097 
8098 	/*
8099 	 * check to see if things have changed while we originally
8100 	 * picked up the client struct.  If they have, then return and
8101 	 * retry the processing of this SETCLIENTID request.
8102 	 */
8103 	if (cp_unconfirmed) {
8104 		rfs4_dbe_lock(cp_unconfirmed->rc_dbe);
8105 		if (!cp_unconfirmed->rc_need_confirm) {
8106 			rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8107 			rfs4_client_rele(cp_unconfirmed);
8108 			if (cp_confirmed)
8109 				rfs4_client_rele(cp_confirmed);
8110 			goto retry;
8111 		}
8112 		/* do away with the old unconfirmed one */
8113 		rfs4_dbe_invalidate(cp_unconfirmed->rc_dbe);
8114 		rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8115 		rfs4_client_rele(cp_unconfirmed);
8116 		cp_unconfirmed = NULL;
8117 	}
8118 
8119 	/*
8120 	 * This search will temporarily hide the confirmed client
8121 	 * struct while a new client struct is created as the
8122 	 * unconfirmed one.
8123 	 */
8124 	create = TRUE;
8125 	newcp = rfs4_findclient(&args->client, &create, cp_confirmed);
8126 
8127 	ASSERT(newcp != NULL);
8128 
8129 	if (newcp == NULL) {
8130 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8131 		rfs4_client_rele(cp_confirmed);
8132 		goto out;
8133 	}
8134 
8135 	/*
8136 	 * If one was not created, then a similar request must be in
8137 	 * process so release and start over with this one
8138 	 */
8139 	if (create != TRUE) {
8140 		rfs4_client_rele(newcp);
8141 		if (cp_confirmed)
8142 			rfs4_client_rele(cp_confirmed);
8143 		goto retry;
8144 	}
8145 
8146 	*cs->statusp = res->status = NFS4_OK;
8147 	res->SETCLIENTID4res_u.resok4.clientid = newcp->rc_clientid;
8148 	res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8149 	    newcp->rc_confirm_verf;
8150 	/* Setup callback information; CB_NULL confirmation later */
8151 	rfs4_client_setcb(newcp, &args->callback, args->callback_ident);
8152 
8153 	newcp->rc_cp_confirmed = cp_confirmed;
8154 
8155 	rfs4_client_rele(newcp);
8156 
8157 out:
8158 	DTRACE_NFSV4_2(op__setclientid__done, struct compound_state *, cs,
8159 	    SETCLIENTID4res *, res);
8160 }
8161 
8162 /*ARGSUSED*/
8163 void
8164 rfs4_op_setclientid_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
8165     struct svc_req *req, struct compound_state *cs)
8166 {
8167 	SETCLIENTID_CONFIRM4args *args =
8168 	    &argop->nfs_argop4_u.opsetclientid_confirm;
8169 	SETCLIENTID_CONFIRM4res *res =
8170 	    &resop->nfs_resop4_u.opsetclientid_confirm;
8171 	rfs4_client_t *cp, *cptoclose = NULL;
8172 
8173 	DTRACE_NFSV4_2(op__setclientid__confirm__start,
8174 	    struct compound_state *, cs,
8175 	    SETCLIENTID_CONFIRM4args *, args);
8176 
8177 	*cs->statusp = res->status = NFS4_OK;
8178 
8179 	cp = rfs4_findclient_by_id(args->clientid, TRUE);
8180 
8181 	if (cp == NULL) {
8182 		*cs->statusp = res->status =
8183 		    rfs4_check_clientid(&args->clientid, 1);
8184 		goto out;
8185 	}
8186 
8187 	if (!creds_ok(cp, req, cs)) {
8188 		*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8189 		rfs4_client_rele(cp);
8190 		goto out;
8191 	}
8192 
8193 	/* If the verifier doesn't match, the record doesn't match */
8194 	if (cp->rc_confirm_verf != args->setclientid_confirm) {
8195 		*cs->statusp = res->status = NFS4ERR_STALE_CLIENTID;
8196 		rfs4_client_rele(cp);
8197 		goto out;
8198 	}
8199 
8200 	rfs4_dbe_lock(cp->rc_dbe);
8201 	cp->rc_need_confirm = FALSE;
8202 	if (cp->rc_cp_confirmed) {
8203 		cptoclose = cp->rc_cp_confirmed;
8204 		cptoclose->rc_ss_remove = 1;
8205 		cp->rc_cp_confirmed = NULL;
8206 	}
8207 
8208 	/*
8209 	 * Update the client's associated server instance, if it's changed
8210 	 * since the client was created.
8211 	 */
8212 	if (rfs4_servinst(cp) != rfs4_cur_servinst)
8213 		rfs4_servinst_assign(cp, rfs4_cur_servinst);
8214 
8215 	/*
8216 	 * Record clientid in stable storage.
8217 	 * Must be done after server instance has been assigned.
8218 	 */
8219 	rfs4_ss_clid(cp);
8220 
8221 	rfs4_dbe_unlock(cp->rc_dbe);
8222 
8223 	if (cptoclose)
8224 		/* don't need to rele, client_close does it */
8225 		rfs4_client_close(cptoclose);
8226 
8227 	/* If needed, initiate CB_NULL call for callback path */
8228 	rfs4_deleg_cb_check(cp);
8229 	rfs4_update_lease(cp);
8230 
8231 	/*
8232 	 * Check to see if client can perform reclaims
8233 	 */
8234 	rfs4_ss_chkclid(cp);
8235 
8236 	rfs4_client_rele(cp);
8237 
8238 out:
8239 	DTRACE_NFSV4_2(op__setclientid__confirm__done,
8240 	    struct compound_state *, cs,
8241 	    SETCLIENTID_CONFIRM4 *, res);
8242 }
8243 
8244 
8245 /*ARGSUSED*/
8246 void
8247 rfs4_op_close(nfs_argop4 *argop, nfs_resop4 *resop,
8248     struct svc_req *req, struct compound_state *cs)
8249 {
8250 	CLOSE4args *args = &argop->nfs_argop4_u.opclose;
8251 	CLOSE4res *resp = &resop->nfs_resop4_u.opclose;
8252 	rfs4_state_t *sp;
8253 	nfsstat4 status;
8254 
8255 	DTRACE_NFSV4_2(op__close__start, struct compound_state *, cs,
8256 	    CLOSE4args *, args);
8257 
8258 	if (cs->vp == NULL) {
8259 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8260 		goto out;
8261 	}
8262 
8263 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_INVALID);
8264 	if (status != NFS4_OK) {
8265 		*cs->statusp = resp->status = status;
8266 		goto out;
8267 	}
8268 
8269 	/* Ensure specified filehandle matches */
8270 	if (cs->vp != sp->rs_finfo->rf_vp) {
8271 		rfs4_state_rele(sp);
8272 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8273 		goto out;
8274 	}
8275 
8276 	/* hold off other access to open_owner while we tinker */
8277 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
8278 
8279 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
8280 	case NFS4_CHECK_STATEID_OKAY:
8281 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8282 		    resop) != NFS4_CHKSEQ_OKAY) {
8283 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8284 			goto end;
8285 		}
8286 		break;
8287 	case NFS4_CHECK_STATEID_OLD:
8288 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8289 		goto end;
8290 	case NFS4_CHECK_STATEID_BAD:
8291 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8292 		goto end;
8293 	case NFS4_CHECK_STATEID_EXPIRED:
8294 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8295 		goto end;
8296 	case NFS4_CHECK_STATEID_CLOSED:
8297 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8298 		goto end;
8299 	case NFS4_CHECK_STATEID_UNCONFIRMED:
8300 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8301 		goto end;
8302 	case NFS4_CHECK_STATEID_REPLAY:
8303 		/* Check the sequence id for the open owner */
8304 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8305 		    resop)) {
8306 		case NFS4_CHKSEQ_OKAY:
8307 			/*
8308 			 * This is replayed stateid; if seqid matches
8309 			 * next expected, then client is using wrong seqid.
8310 			 */
8311 			/* FALL THROUGH */
8312 		case NFS4_CHKSEQ_BAD:
8313 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8314 			goto end;
8315 		case NFS4_CHKSEQ_REPLAY:
8316 			/*
8317 			 * Note this case is the duplicate case so
8318 			 * resp->status is already set.
8319 			 */
8320 			*cs->statusp = resp->status;
8321 			rfs4_update_lease(sp->rs_owner->ro_client);
8322 			goto end;
8323 		}
8324 		break;
8325 	default:
8326 		ASSERT(FALSE);
8327 		break;
8328 	}
8329 
8330 	rfs4_dbe_lock(sp->rs_dbe);
8331 
8332 	/* Update the stateid. */
8333 	next_stateid(&sp->rs_stateid);
8334 	resp->open_stateid = sp->rs_stateid.stateid;
8335 
8336 	rfs4_dbe_unlock(sp->rs_dbe);
8337 
8338 	rfs4_update_lease(sp->rs_owner->ro_client);
8339 	rfs4_update_open_sequence(sp->rs_owner);
8340 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8341 
8342 	rfs4_state_close(sp, FALSE, FALSE, cs->cr);
8343 
8344 	*cs->statusp = resp->status = status;
8345 
8346 end:
8347 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
8348 	rfs4_state_rele(sp);
8349 out:
8350 	DTRACE_NFSV4_2(op__close__done, struct compound_state *, cs,
8351 	    CLOSE4res *, resp);
8352 }
8353 
8354 /*
8355  * Manage the counts on the file struct and close all file locks
8356  */
8357 /*ARGSUSED*/
8358 void
8359 rfs4_release_share_lock_state(rfs4_state_t *sp, cred_t *cr,
8360     bool_t close_of_client)
8361 {
8362 	rfs4_file_t *fp = sp->rs_finfo;
8363 	rfs4_lo_state_t *lsp;
8364 	int fflags = 0;
8365 
8366 	/*
8367 	 * If this call is part of the larger closing down of client
8368 	 * state then it is just easier to release all locks
8369 	 * associated with this client instead of going through each
8370 	 * individual file and cleaning locks there.
8371 	 */
8372 	if (close_of_client) {
8373 		if (sp->rs_owner->ro_client->rc_unlksys_completed == FALSE &&
8374 		    !list_is_empty(&sp->rs_lostatelist) &&
8375 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID) {
8376 			/* Is the PxFS kernel module loaded? */
8377 			if (lm_remove_file_locks != NULL) {
8378 				int new_sysid;
8379 
8380 				/* Encode the cluster nodeid in new sysid */
8381 				new_sysid = sp->rs_owner->ro_client->rc_sysidt;
8382 				lm_set_nlmid_flk(&new_sysid);
8383 
8384 				/*
8385 				 * This PxFS routine removes file locks for a
8386 				 * client over all nodes of a cluster.
8387 				 */
8388 				NFS4_DEBUG(rfs4_debug, (CE_NOTE,
8389 				    "lm_remove_file_locks(sysid=0x%x)\n",
8390 				    new_sysid));
8391 				(*lm_remove_file_locks)(new_sysid);
8392 			} else {
8393 				struct flock64 flk;
8394 
8395 				/* Release all locks for this client */
8396 				flk.l_type = F_UNLKSYS;
8397 				flk.l_whence = 0;
8398 				flk.l_start = 0;
8399 				flk.l_len = 0;
8400 				flk.l_sysid =
8401 				    sp->rs_owner->ro_client->rc_sysidt;
8402 				flk.l_pid = 0;
8403 				(void) VOP_FRLOCK(sp->rs_finfo->rf_vp, F_SETLK,
8404 				    &flk, F_REMOTELOCK | FREAD | FWRITE,
8405 				    (u_offset_t)0, NULL, CRED(), NULL);
8406 			}
8407 
8408 			sp->rs_owner->ro_client->rc_unlksys_completed = TRUE;
8409 		}
8410 	}
8411 
8412 	/*
8413 	 * Release all locks on this file by this lock owner or at
8414 	 * least mark the locks as having been released
8415 	 */
8416 	for (lsp = list_head(&sp->rs_lostatelist); lsp != NULL;
8417 	    lsp = list_next(&sp->rs_lostatelist, lsp)) {
8418 		lsp->rls_locks_cleaned = TRUE;
8419 
8420 		/* Was this already taken care of above? */
8421 		if (!close_of_client &&
8422 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8423 			(void) cleanlocks(sp->rs_finfo->rf_vp,
8424 			    lsp->rls_locker->rl_pid,
8425 			    lsp->rls_locker->rl_client->rc_sysidt);
8426 	}
8427 
8428 	/*
8429 	 * Release any shrlocks associated with this open state ID.
8430 	 * This must be done before the rfs4_state gets marked closed.
8431 	 */
8432 	if (sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8433 		(void) rfs4_unshare(sp);
8434 
8435 	if (sp->rs_open_access) {
8436 		rfs4_dbe_lock(fp->rf_dbe);
8437 
8438 		/*
8439 		 * Decrement the count for each access and deny bit that this
8440 		 * state has contributed to the file.
8441 		 * If the file counts go to zero
8442 		 * clear the appropriate bit in the appropriate mask.
8443 		 */
8444 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) {
8445 			fp->rf_access_read--;
8446 			fflags |= FREAD;
8447 			if (fp->rf_access_read == 0)
8448 				fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
8449 		}
8450 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) {
8451 			fp->rf_access_write--;
8452 			fflags |= FWRITE;
8453 			if (fp->rf_access_write == 0)
8454 				fp->rf_share_access &=
8455 				    ~OPEN4_SHARE_ACCESS_WRITE;
8456 		}
8457 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_READ) {
8458 			fp->rf_deny_read--;
8459 			if (fp->rf_deny_read == 0)
8460 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
8461 		}
8462 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) {
8463 			fp->rf_deny_write--;
8464 			if (fp->rf_deny_write == 0)
8465 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
8466 		}
8467 
8468 		(void) VOP_CLOSE(fp->rf_vp, fflags, 1, (offset_t)0, cr, NULL);
8469 
8470 		rfs4_dbe_unlock(fp->rf_dbe);
8471 
8472 		sp->rs_open_access = 0;
8473 		sp->rs_open_deny = 0;
8474 	}
8475 }
8476 
8477 /*
8478  * lock_denied: Fill in a LOCK4deneid structure given an flock64 structure.
8479  */
8480 static nfsstat4
8481 lock_denied(LOCK4denied *dp, struct flock64 *flk)
8482 {
8483 	rfs4_lockowner_t *lo;
8484 	rfs4_client_t *cp;
8485 	uint32_t len;
8486 
8487 	lo = rfs4_findlockowner_by_pid(flk->l_pid);
8488 	if (lo != NULL) {
8489 		cp = lo->rl_client;
8490 		if (rfs4_lease_expired(cp)) {
8491 			rfs4_lockowner_rele(lo);
8492 			rfs4_dbe_hold(cp->rc_dbe);
8493 			rfs4_client_close(cp);
8494 			return (NFS4ERR_EXPIRED);
8495 		}
8496 		dp->owner.clientid = lo->rl_owner.clientid;
8497 		len = lo->rl_owner.owner_len;
8498 		dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8499 		bcopy(lo->rl_owner.owner_val, dp->owner.owner_val, len);
8500 		dp->owner.owner_len = len;
8501 		rfs4_lockowner_rele(lo);
8502 		goto finish;
8503 	}
8504 
8505 	/*
8506 	 * Its not a NFS4 lock. We take advantage that the upper 32 bits
8507 	 * of the client id contain the boot time for a NFS4 lock. So we
8508 	 * fabricate and identity by setting clientid to the sysid, and
8509 	 * the lock owner to the pid.
8510 	 */
8511 	dp->owner.clientid = flk->l_sysid;
8512 	len = sizeof (pid_t);
8513 	dp->owner.owner_len = len;
8514 	dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8515 	bcopy(&flk->l_pid, dp->owner.owner_val, len);
8516 finish:
8517 	dp->offset = flk->l_start;
8518 	dp->length = flk->l_len;
8519 
8520 	if (flk->l_type == F_RDLCK)
8521 		dp->locktype = READ_LT;
8522 	else if (flk->l_type == F_WRLCK)
8523 		dp->locktype = WRITE_LT;
8524 	else
8525 		return (NFS4ERR_INVAL);	/* no mapping from POSIX ltype to v4 */
8526 
8527 	return (NFS4_OK);
8528 }
8529 
8530 static int
8531 setlock(vnode_t *vp, struct flock64 *flock, int flag, cred_t *cred)
8532 {
8533 	int error;
8534 	struct flock64 flk;
8535 	int i;
8536 	clock_t delaytime;
8537 	int cmd;
8538 
8539 	cmd = nbl_need_check(vp) ? F_SETLK_NBMAND : F_SETLK;
8540 retry:
8541 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
8542 
8543 	for (i = 0; i < rfs4_maxlock_tries; i++) {
8544 		LOCK_PRINT(rfs4_debug, "setlock", cmd, flock);
8545 		error = VOP_FRLOCK(vp, cmd,
8546 		    flock, flag, (u_offset_t)0, NULL, cred, NULL);
8547 
8548 		if (error != EAGAIN && error != EACCES)
8549 			break;
8550 
8551 		if (i < rfs4_maxlock_tries - 1) {
8552 			delay(delaytime);
8553 			delaytime *= 2;
8554 		}
8555 	}
8556 
8557 	if (error == EAGAIN || error == EACCES) {
8558 		/* Get the owner of the lock */
8559 		flk = *flock;
8560 		LOCK_PRINT(rfs4_debug, "setlock", F_GETLK, &flk);
8561 		if (VOP_FRLOCK(vp, F_GETLK, &flk, flag,
8562 		    (u_offset_t)0, NULL, cred, NULL) == 0) {
8563 			if (flk.l_type == F_UNLCK) {
8564 				/* No longer locked, retry */
8565 				goto retry;
8566 			}
8567 			*flock = flk;
8568 			LOCK_PRINT(rfs4_debug, "setlock(blocking lock)",
8569 			    F_GETLK, &flk);
8570 		}
8571 	}
8572 
8573 	return (error);
8574 }
8575 
8576 /*ARGSUSED*/
8577 static nfsstat4
8578 rfs4_do_lock(rfs4_lo_state_t *lsp, nfs_lock_type4 locktype,
8579     offset4 offset, length4 length, cred_t *cred, nfs_resop4 *resop)
8580 {
8581 	nfsstat4 status;
8582 	rfs4_lockowner_t *lo = lsp->rls_locker;
8583 	rfs4_state_t *sp = lsp->rls_state;
8584 	struct flock64 flock;
8585 	int16_t ltype;
8586 	int flag;
8587 	int error;
8588 	sysid_t sysid;
8589 	LOCK4res *lres;
8590 
8591 	if (rfs4_lease_expired(lo->rl_client)) {
8592 		return (NFS4ERR_EXPIRED);
8593 	}
8594 
8595 	if ((status = rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
8596 		return (status);
8597 
8598 	/* Check for zero length. To lock to end of file use all ones for V4 */
8599 	if (length == 0)
8600 		return (NFS4ERR_INVAL);
8601 	else if (length == (length4)(~0))
8602 		length = 0;		/* Posix to end of file  */
8603 
8604 retry:
8605 	rfs4_dbe_lock(sp->rs_dbe);
8606 	if (sp->rs_closed) {
8607 		rfs4_dbe_unlock(sp->rs_dbe);
8608 		return (NFS4ERR_OLD_STATEID);
8609 	}
8610 
8611 	if (resop->resop != OP_LOCKU) {
8612 		switch (locktype) {
8613 		case READ_LT:
8614 		case READW_LT:
8615 			if ((sp->rs_share_access
8616 			    & OPEN4_SHARE_ACCESS_READ) == 0) {
8617 				rfs4_dbe_unlock(sp->rs_dbe);
8618 
8619 				return (NFS4ERR_OPENMODE);
8620 			}
8621 			ltype = F_RDLCK;
8622 			break;
8623 		case WRITE_LT:
8624 		case WRITEW_LT:
8625 			if ((sp->rs_share_access
8626 			    & OPEN4_SHARE_ACCESS_WRITE) == 0) {
8627 				rfs4_dbe_unlock(sp->rs_dbe);
8628 
8629 				return (NFS4ERR_OPENMODE);
8630 			}
8631 			ltype = F_WRLCK;
8632 			break;
8633 		}
8634 	} else
8635 		ltype = F_UNLCK;
8636 
8637 	flock.l_type = ltype;
8638 	flock.l_whence = 0;		/* SEEK_SET */
8639 	flock.l_start = offset;
8640 	flock.l_len = length;
8641 	flock.l_sysid = sysid;
8642 	flock.l_pid = lsp->rls_locker->rl_pid;
8643 
8644 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
8645 	if (flock.l_len < 0 || flock.l_start < 0) {
8646 		rfs4_dbe_unlock(sp->rs_dbe);
8647 		return (NFS4ERR_INVAL);
8648 	}
8649 
8650 	/*
8651 	 * N.B. FREAD has the same value as OPEN4_SHARE_ACCESS_READ and
8652 	 * FWRITE has the same value as OPEN4_SHARE_ACCESS_WRITE.
8653 	 */
8654 	flag = (int)sp->rs_share_access | F_REMOTELOCK;
8655 
8656 	error = setlock(sp->rs_finfo->rf_vp, &flock, flag, cred);
8657 	if (error == 0) {
8658 		rfs4_dbe_lock(lsp->rls_dbe);
8659 		next_stateid(&lsp->rls_lockid);
8660 		rfs4_dbe_unlock(lsp->rls_dbe);
8661 	}
8662 
8663 	rfs4_dbe_unlock(sp->rs_dbe);
8664 
8665 	/*
8666 	 * N.B. We map error values to nfsv4 errors. This is differrent
8667 	 * than puterrno4 routine.
8668 	 */
8669 	switch (error) {
8670 	case 0:
8671 		status = NFS4_OK;
8672 		break;
8673 	case EAGAIN:
8674 	case EACCES:		/* Old value */
8675 		/* Can only get here if op is OP_LOCK */
8676 		ASSERT(resop->resop == OP_LOCK);
8677 		lres = &resop->nfs_resop4_u.oplock;
8678 		status = NFS4ERR_DENIED;
8679 		if (lock_denied(&lres->LOCK4res_u.denied, &flock)
8680 		    == NFS4ERR_EXPIRED)
8681 			goto retry;
8682 		break;
8683 	case ENOLCK:
8684 		status = NFS4ERR_DELAY;
8685 		break;
8686 	case EOVERFLOW:
8687 		status = NFS4ERR_INVAL;
8688 		break;
8689 	case EINVAL:
8690 		status = NFS4ERR_NOTSUPP;
8691 		break;
8692 	default:
8693 		status = NFS4ERR_SERVERFAULT;
8694 		break;
8695 	}
8696 
8697 	return (status);
8698 }
8699 
8700 /*ARGSUSED*/
8701 void
8702 rfs4_op_lock(nfs_argop4 *argop, nfs_resop4 *resop,
8703     struct svc_req *req, struct compound_state *cs)
8704 {
8705 	LOCK4args *args = &argop->nfs_argop4_u.oplock;
8706 	LOCK4res *resp = &resop->nfs_resop4_u.oplock;
8707 	nfsstat4 status;
8708 	stateid4 *stateid;
8709 	rfs4_lockowner_t *lo;
8710 	rfs4_client_t *cp;
8711 	rfs4_state_t *sp = NULL;
8712 	rfs4_lo_state_t *lsp = NULL;
8713 	bool_t ls_sw_held = FALSE;
8714 	bool_t create = TRUE;
8715 	bool_t lcreate = TRUE;
8716 	bool_t dup_lock = FALSE;
8717 	int rc;
8718 
8719 	DTRACE_NFSV4_2(op__lock__start, struct compound_state *, cs,
8720 	    LOCK4args *, args);
8721 
8722 	if (cs->vp == NULL) {
8723 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8724 		DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8725 		    cs, LOCK4res *, resp);
8726 		return;
8727 	}
8728 
8729 	if (args->locker.new_lock_owner) {
8730 		/* Create a new lockowner for this instance */
8731 		open_to_lock_owner4 *olo = &args->locker.locker4_u.open_owner;
8732 
8733 		NFS4_DEBUG(rfs4_debug, (CE_NOTE, "Creating new lock owner"));
8734 
8735 		stateid = &olo->open_stateid;
8736 		status = rfs4_get_state(stateid, &sp, RFS4_DBS_VALID);
8737 		if (status != NFS4_OK) {
8738 			NFS4_DEBUG(rfs4_debug,
8739 			    (CE_NOTE, "Get state failed in lock %d", status));
8740 			*cs->statusp = resp->status = status;
8741 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8742 			    cs, LOCK4res *, resp);
8743 			return;
8744 		}
8745 
8746 		/* Ensure specified filehandle matches */
8747 		if (cs->vp != sp->rs_finfo->rf_vp) {
8748 			rfs4_state_rele(sp);
8749 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8750 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8751 			    cs, LOCK4res *, resp);
8752 			return;
8753 		}
8754 
8755 		/* hold off other access to open_owner while we tinker */
8756 		rfs4_sw_enter(&sp->rs_owner->ro_sw);
8757 
8758 		switch (rc = rfs4_check_stateid_seqid(sp, stateid)) {
8759 		case NFS4_CHECK_STATEID_OLD:
8760 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8761 			goto end;
8762 		case NFS4_CHECK_STATEID_BAD:
8763 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8764 			goto end;
8765 		case NFS4_CHECK_STATEID_EXPIRED:
8766 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8767 			goto end;
8768 		case NFS4_CHECK_STATEID_UNCONFIRMED:
8769 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8770 			goto end;
8771 		case NFS4_CHECK_STATEID_CLOSED:
8772 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8773 			goto end;
8774 		case NFS4_CHECK_STATEID_OKAY:
8775 		case NFS4_CHECK_STATEID_REPLAY:
8776 			switch (rfs4_check_olo_seqid(olo->open_seqid,
8777 			    sp->rs_owner, resop)) {
8778 			case NFS4_CHKSEQ_OKAY:
8779 				if (rc == NFS4_CHECK_STATEID_OKAY)
8780 					break;
8781 				/*
8782 				 * This is replayed stateid; if seqid
8783 				 * matches next expected, then client
8784 				 * is using wrong seqid.
8785 				 */
8786 				/* FALLTHROUGH */
8787 			case NFS4_CHKSEQ_BAD:
8788 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8789 				goto end;
8790 			case NFS4_CHKSEQ_REPLAY:
8791 				/* This is a duplicate LOCK request */
8792 				dup_lock = TRUE;
8793 
8794 				/*
8795 				 * For a duplicate we do not want to
8796 				 * create a new lockowner as it should
8797 				 * already exist.
8798 				 * Turn off the lockowner create flag.
8799 				 */
8800 				lcreate = FALSE;
8801 			}
8802 			break;
8803 		}
8804 
8805 		lo = rfs4_findlockowner(&olo->lock_owner, &lcreate);
8806 		if (lo == NULL) {
8807 			NFS4_DEBUG(rfs4_debug,
8808 			    (CE_NOTE, "rfs4_op_lock: no lock owner"));
8809 			*cs->statusp = resp->status = NFS4ERR_RESOURCE;
8810 			goto end;
8811 		}
8812 
8813 		lsp = rfs4_findlo_state_by_owner(lo, sp, &create);
8814 		if (lsp == NULL) {
8815 			rfs4_update_lease(sp->rs_owner->ro_client);
8816 			/*
8817 			 * Only update theh open_seqid if this is not
8818 			 * a duplicate request
8819 			 */
8820 			if (dup_lock == FALSE) {
8821 				rfs4_update_open_sequence(sp->rs_owner);
8822 			}
8823 
8824 			NFS4_DEBUG(rfs4_debug,
8825 			    (CE_NOTE, "rfs4_op_lock: no state"));
8826 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
8827 			rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8828 			rfs4_lockowner_rele(lo);
8829 			goto end;
8830 		}
8831 
8832 		/*
8833 		 * This is the new_lock_owner branch and the client is
8834 		 * supposed to be associating a new lock_owner with
8835 		 * the open file at this point.  If we find that a
8836 		 * lock_owner/state association already exists and a
8837 		 * successful LOCK request was returned to the client,
8838 		 * an error is returned to the client since this is
8839 		 * not appropriate.  The client should be using the
8840 		 * existing lock_owner branch.
8841 		 */
8842 		if (dup_lock == FALSE && create == FALSE) {
8843 			if (lsp->rls_lock_completed == TRUE) {
8844 				*cs->statusp =
8845 				    resp->status = NFS4ERR_BAD_SEQID;
8846 				rfs4_lockowner_rele(lo);
8847 				goto end;
8848 			}
8849 		}
8850 
8851 		rfs4_update_lease(sp->rs_owner->ro_client);
8852 
8853 		/*
8854 		 * Only update theh open_seqid if this is not
8855 		 * a duplicate request
8856 		 */
8857 		if (dup_lock == FALSE) {
8858 			rfs4_update_open_sequence(sp->rs_owner);
8859 		}
8860 
8861 		/*
8862 		 * If this is a duplicate lock request, just copy the
8863 		 * previously saved reply and return.
8864 		 */
8865 		if (dup_lock == TRUE) {
8866 			/* verify that lock_seqid's match */
8867 			if (lsp->rls_seqid != olo->lock_seqid) {
8868 				NFS4_DEBUG(rfs4_debug,
8869 				    (CE_NOTE, "rfs4_op_lock: Dup-Lock seqid bad"
8870 				    "lsp->seqid=%d old->seqid=%d",
8871 				    lsp->rls_seqid, olo->lock_seqid));
8872 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8873 			} else {
8874 				rfs4_copy_reply(resop, &lsp->rls_reply);
8875 				/*
8876 				 * Make sure to copy the just
8877 				 * retrieved reply status into the
8878 				 * overall compound status
8879 				 */
8880 				*cs->statusp = resp->status;
8881 			}
8882 			rfs4_lockowner_rele(lo);
8883 			goto end;
8884 		}
8885 
8886 		rfs4_dbe_lock(lsp->rls_dbe);
8887 
8888 		/* Make sure to update the lock sequence id */
8889 		lsp->rls_seqid = olo->lock_seqid;
8890 
8891 		NFS4_DEBUG(rfs4_debug,
8892 		    (CE_NOTE, "Lock seqid established as %d", lsp->rls_seqid));
8893 
8894 		/*
8895 		 * This is used to signify the newly created lockowner
8896 		 * stateid and its sequence number.  The checks for
8897 		 * sequence number and increment don't occur on the
8898 		 * very first lock request for a lockowner.
8899 		 */
8900 		lsp->rls_skip_seqid_check = TRUE;
8901 
8902 		/* hold off other access to lsp while we tinker */
8903 		rfs4_sw_enter(&lsp->rls_sw);
8904 		ls_sw_held = TRUE;
8905 
8906 		rfs4_dbe_unlock(lsp->rls_dbe);
8907 
8908 		rfs4_lockowner_rele(lo);
8909 	} else {
8910 		stateid = &args->locker.locker4_u.lock_owner.lock_stateid;
8911 		/* get lsp and hold the lock on the underlying file struct */
8912 		if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE))
8913 		    != NFS4_OK) {
8914 			*cs->statusp = resp->status = status;
8915 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8916 			    cs, LOCK4res *, resp);
8917 			return;
8918 		}
8919 		create = FALSE;	/* We didn't create lsp */
8920 
8921 		/* Ensure specified filehandle matches */
8922 		if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
8923 			rfs4_lo_state_rele(lsp, TRUE);
8924 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8925 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8926 			    cs, LOCK4res *, resp);
8927 			return;
8928 		}
8929 
8930 		/* hold off other access to lsp while we tinker */
8931 		rfs4_sw_enter(&lsp->rls_sw);
8932 		ls_sw_held = TRUE;
8933 
8934 		switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
8935 		/*
8936 		 * The stateid looks like it was okay (expected to be
8937 		 * the next one)
8938 		 */
8939 		case NFS4_CHECK_STATEID_OKAY:
8940 			/*
8941 			 * The sequence id is now checked.  Determine
8942 			 * if this is a replay or if it is in the
8943 			 * expected (next) sequence.  In the case of a
8944 			 * replay, there are two replay conditions
8945 			 * that may occur.  The first is the normal
8946 			 * condition where a LOCK is done with a
8947 			 * NFS4_OK response and the stateid is
8948 			 * updated.  That case is handled below when
8949 			 * the stateid is identified as a REPLAY.  The
8950 			 * second is the case where an error is
8951 			 * returned, like NFS4ERR_DENIED, and the
8952 			 * sequence number is updated but the stateid
8953 			 * is not updated.  This second case is dealt
8954 			 * with here.  So it may seem odd that the
8955 			 * stateid is okay but the sequence id is a
8956 			 * replay but it is okay.
8957 			 */
8958 			switch (rfs4_check_lock_seqid(
8959 			    args->locker.locker4_u.lock_owner.lock_seqid,
8960 			    lsp, resop)) {
8961 			case NFS4_CHKSEQ_REPLAY:
8962 				if (resp->status != NFS4_OK) {
8963 					/*
8964 					 * Here is our replay and need
8965 					 * to verify that the last
8966 					 * response was an error.
8967 					 */
8968 					*cs->statusp = resp->status;
8969 					goto end;
8970 				}
8971 				/*
8972 				 * This is done since the sequence id
8973 				 * looked like a replay but it didn't
8974 				 * pass our check so a BAD_SEQID is
8975 				 * returned as a result.
8976 				 */
8977 				/*FALLTHROUGH*/
8978 			case NFS4_CHKSEQ_BAD:
8979 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8980 				goto end;
8981 			case NFS4_CHKSEQ_OKAY:
8982 				/* Everything looks okay move ahead */
8983 				break;
8984 			}
8985 			break;
8986 		case NFS4_CHECK_STATEID_OLD:
8987 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8988 			goto end;
8989 		case NFS4_CHECK_STATEID_BAD:
8990 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8991 			goto end;
8992 		case NFS4_CHECK_STATEID_EXPIRED:
8993 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8994 			goto end;
8995 		case NFS4_CHECK_STATEID_CLOSED:
8996 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8997 			goto end;
8998 		case NFS4_CHECK_STATEID_REPLAY:
8999 			switch (rfs4_check_lock_seqid(
9000 			    args->locker.locker4_u.lock_owner.lock_seqid,
9001 			    lsp, resop)) {
9002 			case NFS4_CHKSEQ_OKAY:
9003 				/*
9004 				 * This is a replayed stateid; if
9005 				 * seqid matches the next expected,
9006 				 * then client is using wrong seqid.
9007 				 */
9008 			case NFS4_CHKSEQ_BAD:
9009 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9010 				goto end;
9011 			case NFS4_CHKSEQ_REPLAY:
9012 				rfs4_update_lease(lsp->rls_locker->rl_client);
9013 				*cs->statusp = status = resp->status;
9014 				goto end;
9015 			}
9016 			break;
9017 		default:
9018 			ASSERT(FALSE);
9019 			break;
9020 		}
9021 
9022 		rfs4_update_lock_sequence(lsp);
9023 		rfs4_update_lease(lsp->rls_locker->rl_client);
9024 	}
9025 
9026 	/*
9027 	 * NFS4 only allows locking on regular files, so
9028 	 * verify type of object.
9029 	 */
9030 	if (cs->vp->v_type != VREG) {
9031 		if (cs->vp->v_type == VDIR)
9032 			status = NFS4ERR_ISDIR;
9033 		else
9034 			status = NFS4ERR_INVAL;
9035 		goto out;
9036 	}
9037 
9038 	cp = lsp->rls_state->rs_owner->ro_client;
9039 
9040 	if (rfs4_clnt_in_grace(cp) && !args->reclaim) {
9041 		status = NFS4ERR_GRACE;
9042 		goto out;
9043 	}
9044 
9045 	if (rfs4_clnt_in_grace(cp) && args->reclaim && !cp->rc_can_reclaim) {
9046 		status = NFS4ERR_NO_GRACE;
9047 		goto out;
9048 	}
9049 
9050 	if (!rfs4_clnt_in_grace(cp) && args->reclaim) {
9051 		status = NFS4ERR_NO_GRACE;
9052 		goto out;
9053 	}
9054 
9055 	if (lsp->rls_state->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE)
9056 		cs->deleg = TRUE;
9057 
9058 	status = rfs4_do_lock(lsp, args->locktype,
9059 	    args->offset, args->length, cs->cr, resop);
9060 
9061 out:
9062 	lsp->rls_skip_seqid_check = FALSE;
9063 
9064 	*cs->statusp = resp->status = status;
9065 
9066 	if (status == NFS4_OK) {
9067 		resp->LOCK4res_u.lock_stateid = lsp->rls_lockid.stateid;
9068 		lsp->rls_lock_completed = TRUE;
9069 	}
9070 	/*
9071 	 * Only update the "OPEN" response here if this was a new
9072 	 * lock_owner
9073 	 */
9074 	if (sp)
9075 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
9076 
9077 	rfs4_update_lock_resp(lsp, resop);
9078 
9079 end:
9080 	if (lsp) {
9081 		if (ls_sw_held)
9082 			rfs4_sw_exit(&lsp->rls_sw);
9083 		/*
9084 		 * If an sp obtained, then the lsp does not represent
9085 		 * a lock on the file struct.
9086 		 */
9087 		if (sp != NULL)
9088 			rfs4_lo_state_rele(lsp, FALSE);
9089 		else
9090 			rfs4_lo_state_rele(lsp, TRUE);
9091 	}
9092 	if (sp) {
9093 		rfs4_sw_exit(&sp->rs_owner->ro_sw);
9094 		rfs4_state_rele(sp);
9095 	}
9096 
9097 	DTRACE_NFSV4_2(op__lock__done, struct compound_state *, cs,
9098 	    LOCK4res *, resp);
9099 }
9100 
9101 /* free function for LOCK/LOCKT */
9102 static void
9103 lock_denied_free(nfs_resop4 *resop)
9104 {
9105 	LOCK4denied *dp = NULL;
9106 
9107 	switch (resop->resop) {
9108 	case OP_LOCK:
9109 		if (resop->nfs_resop4_u.oplock.status == NFS4ERR_DENIED)
9110 			dp = &resop->nfs_resop4_u.oplock.LOCK4res_u.denied;
9111 		break;
9112 	case OP_LOCKT:
9113 		if (resop->nfs_resop4_u.oplockt.status == NFS4ERR_DENIED)
9114 			dp = &resop->nfs_resop4_u.oplockt.denied;
9115 		break;
9116 	default:
9117 		break;
9118 	}
9119 
9120 	if (dp)
9121 		kmem_free(dp->owner.owner_val, dp->owner.owner_len);
9122 }
9123 
9124 /*ARGSUSED*/
9125 void
9126 rfs4_op_locku(nfs_argop4 *argop, nfs_resop4 *resop,
9127     struct svc_req *req, struct compound_state *cs)
9128 {
9129 	LOCKU4args *args = &argop->nfs_argop4_u.oplocku;
9130 	LOCKU4res *resp = &resop->nfs_resop4_u.oplocku;
9131 	nfsstat4 status;
9132 	stateid4 *stateid = &args->lock_stateid;
9133 	rfs4_lo_state_t *lsp;
9134 
9135 	DTRACE_NFSV4_2(op__locku__start, struct compound_state *, cs,
9136 	    LOCKU4args *, args);
9137 
9138 	if (cs->vp == NULL) {
9139 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9140 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9141 		    LOCKU4res *, resp);
9142 		return;
9143 	}
9144 
9145 	if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE)) != NFS4_OK) {
9146 		*cs->statusp = resp->status = status;
9147 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9148 		    LOCKU4res *, resp);
9149 		return;
9150 	}
9151 
9152 	/* Ensure specified filehandle matches */
9153 	if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9154 		rfs4_lo_state_rele(lsp, TRUE);
9155 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9156 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9157 		    LOCKU4res *, resp);
9158 		return;
9159 	}
9160 
9161 	/* hold off other access to lsp while we tinker */
9162 	rfs4_sw_enter(&lsp->rls_sw);
9163 
9164 	switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9165 	case NFS4_CHECK_STATEID_OKAY:
9166 		if (rfs4_check_lock_seqid(args->seqid, lsp, resop)
9167 		    != NFS4_CHKSEQ_OKAY) {
9168 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9169 			goto end;
9170 		}
9171 		break;
9172 	case NFS4_CHECK_STATEID_OLD:
9173 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9174 		goto end;
9175 	case NFS4_CHECK_STATEID_BAD:
9176 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9177 		goto end;
9178 	case NFS4_CHECK_STATEID_EXPIRED:
9179 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9180 		goto end;
9181 	case NFS4_CHECK_STATEID_CLOSED:
9182 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9183 		goto end;
9184 	case NFS4_CHECK_STATEID_REPLAY:
9185 		switch (rfs4_check_lock_seqid(args->seqid, lsp, resop)) {
9186 		case NFS4_CHKSEQ_OKAY:
9187 				/*
9188 				 * This is a replayed stateid; if
9189 				 * seqid matches the next expected,
9190 				 * then client is using wrong seqid.
9191 				 */
9192 		case NFS4_CHKSEQ_BAD:
9193 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9194 			goto end;
9195 		case NFS4_CHKSEQ_REPLAY:
9196 			rfs4_update_lease(lsp->rls_locker->rl_client);
9197 			*cs->statusp = status = resp->status;
9198 			goto end;
9199 		}
9200 		break;
9201 	default:
9202 		ASSERT(FALSE);
9203 		break;
9204 	}
9205 
9206 	rfs4_update_lock_sequence(lsp);
9207 	rfs4_update_lease(lsp->rls_locker->rl_client);
9208 
9209 	/*
9210 	 * NFS4 only allows locking on regular files, so
9211 	 * verify type of object.
9212 	 */
9213 	if (cs->vp->v_type != VREG) {
9214 		if (cs->vp->v_type == VDIR)
9215 			status = NFS4ERR_ISDIR;
9216 		else
9217 			status = NFS4ERR_INVAL;
9218 		goto out;
9219 	}
9220 
9221 	if (rfs4_clnt_in_grace(lsp->rls_state->rs_owner->ro_client)) {
9222 		status = NFS4ERR_GRACE;
9223 		goto out;
9224 	}
9225 
9226 	status = rfs4_do_lock(lsp, args->locktype,
9227 	    args->offset, args->length, cs->cr, resop);
9228 
9229 out:
9230 	*cs->statusp = resp->status = status;
9231 
9232 	if (status == NFS4_OK)
9233 		resp->lock_stateid = lsp->rls_lockid.stateid;
9234 
9235 	rfs4_update_lock_resp(lsp, resop);
9236 
9237 end:
9238 	rfs4_sw_exit(&lsp->rls_sw);
9239 	rfs4_lo_state_rele(lsp, TRUE);
9240 
9241 	DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9242 	    LOCKU4res *, resp);
9243 }
9244 
9245 /*
9246  * LOCKT is a best effort routine, the client can not be guaranteed that
9247  * the status return is still in effect by the time the reply is received.
9248  * They are numerous race conditions in this routine, but we are not required
9249  * and can not be accurate.
9250  */
9251 /*ARGSUSED*/
9252 void
9253 rfs4_op_lockt(nfs_argop4 *argop, nfs_resop4 *resop,
9254     struct svc_req *req, struct compound_state *cs)
9255 {
9256 	LOCKT4args *args = &argop->nfs_argop4_u.oplockt;
9257 	LOCKT4res *resp = &resop->nfs_resop4_u.oplockt;
9258 	rfs4_lockowner_t *lo;
9259 	rfs4_client_t *cp;
9260 	bool_t create = FALSE;
9261 	struct flock64 flk;
9262 	int error;
9263 	int flag = FREAD | FWRITE;
9264 	int ltype;
9265 	length4 posix_length;
9266 	sysid_t sysid;
9267 	pid_t pid;
9268 
9269 	DTRACE_NFSV4_2(op__lockt__start, struct compound_state *, cs,
9270 	    LOCKT4args *, args);
9271 
9272 	if (cs->vp == NULL) {
9273 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9274 		goto out;
9275 	}
9276 
9277 	/*
9278 	 * NFS4 only allows locking on regular files, so
9279 	 * verify type of object.
9280 	 */
9281 	if (cs->vp->v_type != VREG) {
9282 		if (cs->vp->v_type == VDIR)
9283 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
9284 		else
9285 			*cs->statusp = resp->status =  NFS4ERR_INVAL;
9286 		goto out;
9287 	}
9288 
9289 	/*
9290 	 * Check out the clientid to ensure the server knows about it
9291 	 * so that we correctly inform the client of a server reboot.
9292 	 */
9293 	if ((cp = rfs4_findclient_by_id(args->owner.clientid, FALSE))
9294 	    == NULL) {
9295 		*cs->statusp = resp->status =
9296 		    rfs4_check_clientid(&args->owner.clientid, 0);
9297 		goto out;
9298 	}
9299 	if (rfs4_lease_expired(cp)) {
9300 		rfs4_client_close(cp);
9301 		/*
9302 		 * Protocol doesn't allow returning NFS4ERR_STALE as
9303 		 * other operations do on this check so STALE_CLIENTID
9304 		 * is returned instead
9305 		 */
9306 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
9307 		goto out;
9308 	}
9309 
9310 	if (rfs4_clnt_in_grace(cp) && !(cp->rc_can_reclaim)) {
9311 		*cs->statusp = resp->status = NFS4ERR_GRACE;
9312 		rfs4_client_rele(cp);
9313 		goto out;
9314 	}
9315 	rfs4_client_rele(cp);
9316 
9317 	resp->status = NFS4_OK;
9318 
9319 	switch (args->locktype) {
9320 	case READ_LT:
9321 	case READW_LT:
9322 		ltype = F_RDLCK;
9323 		break;
9324 	case WRITE_LT:
9325 	case WRITEW_LT:
9326 		ltype = F_WRLCK;
9327 		break;
9328 	}
9329 
9330 	posix_length = args->length;
9331 	/* Check for zero length. To lock to end of file use all ones for V4 */
9332 	if (posix_length == 0) {
9333 		*cs->statusp = resp->status = NFS4ERR_INVAL;
9334 		goto out;
9335 	} else if (posix_length == (length4)(~0)) {
9336 		posix_length = 0;	/* Posix to end of file  */
9337 	}
9338 
9339 	/* Find or create a lockowner */
9340 	lo = rfs4_findlockowner(&args->owner, &create);
9341 
9342 	if (lo) {
9343 		pid = lo->rl_pid;
9344 		if ((resp->status =
9345 		    rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
9346 			goto err;
9347 	} else {
9348 		pid = 0;
9349 		sysid = lockt_sysid;
9350 	}
9351 retry:
9352 	flk.l_type = ltype;
9353 	flk.l_whence = 0;		/* SEEK_SET */
9354 	flk.l_start = args->offset;
9355 	flk.l_len = posix_length;
9356 	flk.l_sysid = sysid;
9357 	flk.l_pid = pid;
9358 	flag |= F_REMOTELOCK;
9359 
9360 	LOCK_PRINT(rfs4_debug, "rfs4_op_lockt", F_GETLK, &flk);
9361 
9362 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
9363 	if (flk.l_len < 0 || flk.l_start < 0) {
9364 		resp->status = NFS4ERR_INVAL;
9365 		goto err;
9366 	}
9367 	error = VOP_FRLOCK(cs->vp, F_GETLK, &flk, flag, (u_offset_t)0,
9368 	    NULL, cs->cr, NULL);
9369 
9370 	/*
9371 	 * N.B. We map error values to nfsv4 errors. This is differrent
9372 	 * than puterrno4 routine.
9373 	 */
9374 	switch (error) {
9375 	case 0:
9376 		if (flk.l_type == F_UNLCK)
9377 			resp->status = NFS4_OK;
9378 		else {
9379 			if (lock_denied(&resp->denied, &flk) == NFS4ERR_EXPIRED)
9380 				goto retry;
9381 			resp->status = NFS4ERR_DENIED;
9382 		}
9383 		break;
9384 	case EOVERFLOW:
9385 		resp->status = NFS4ERR_INVAL;
9386 		break;
9387 	case EINVAL:
9388 		resp->status = NFS4ERR_NOTSUPP;
9389 		break;
9390 	default:
9391 		cmn_err(CE_WARN, "rfs4_op_lockt: unexpected errno (%d)",
9392 		    error);
9393 		resp->status = NFS4ERR_SERVERFAULT;
9394 		break;
9395 	}
9396 
9397 err:
9398 	if (lo)
9399 		rfs4_lockowner_rele(lo);
9400 	*cs->statusp = resp->status;
9401 out:
9402 	DTRACE_NFSV4_2(op__lockt__done, struct compound_state *, cs,
9403 	    LOCKT4res *, resp);
9404 }
9405 
9406 int
9407 rfs4_share(rfs4_state_t *sp, uint32_t access, uint32_t deny)
9408 {
9409 	int err;
9410 	int cmd;
9411 	vnode_t *vp;
9412 	struct shrlock shr;
9413 	struct shr_locowner shr_loco;
9414 	int fflags = 0;
9415 
9416 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9417 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9418 
9419 	if (sp->rs_closed)
9420 		return (NFS4ERR_OLD_STATEID);
9421 
9422 	vp = sp->rs_finfo->rf_vp;
9423 	ASSERT(vp);
9424 
9425 	shr.s_access = shr.s_deny = 0;
9426 
9427 	if (access & OPEN4_SHARE_ACCESS_READ) {
9428 		fflags |= FREAD;
9429 		shr.s_access |= F_RDACC;
9430 	}
9431 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
9432 		fflags |= FWRITE;
9433 		shr.s_access |= F_WRACC;
9434 	}
9435 	ASSERT(shr.s_access);
9436 
9437 	if (deny & OPEN4_SHARE_DENY_READ)
9438 		shr.s_deny |= F_RDDNY;
9439 	if (deny & OPEN4_SHARE_DENY_WRITE)
9440 		shr.s_deny |= F_WRDNY;
9441 
9442 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9443 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9444 	shr_loco.sl_pid = shr.s_pid;
9445 	shr_loco.sl_id = shr.s_sysid;
9446 	shr.s_owner = (caddr_t)&shr_loco;
9447 	shr.s_own_len = sizeof (shr_loco);
9448 
9449 	cmd = nbl_need_check(vp) ? F_SHARE_NBMAND : F_SHARE;
9450 
9451 	err = VOP_SHRLOCK(vp, cmd, &shr, fflags, CRED(), NULL);
9452 	if (err != 0) {
9453 		if (err == EAGAIN)
9454 			err = NFS4ERR_SHARE_DENIED;
9455 		else
9456 			err = puterrno4(err);
9457 		return (err);
9458 	}
9459 
9460 	sp->rs_share_access |= access;
9461 	sp->rs_share_deny |= deny;
9462 
9463 	return (0);
9464 }
9465 
9466 int
9467 rfs4_unshare(rfs4_state_t *sp)
9468 {
9469 	int err;
9470 	struct shrlock shr;
9471 	struct shr_locowner shr_loco;
9472 
9473 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9474 
9475 	if (sp->rs_closed || sp->rs_share_access == 0)
9476 		return (0);
9477 
9478 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9479 	ASSERT(sp->rs_finfo->rf_vp);
9480 
9481 	shr.s_access = shr.s_deny = 0;
9482 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9483 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9484 	shr_loco.sl_pid = shr.s_pid;
9485 	shr_loco.sl_id = shr.s_sysid;
9486 	shr.s_owner = (caddr_t)&shr_loco;
9487 	shr.s_own_len = sizeof (shr_loco);
9488 
9489 	err = VOP_SHRLOCK(sp->rs_finfo->rf_vp, F_UNSHARE, &shr, 0, CRED(),
9490 	    NULL);
9491 	if (err != 0) {
9492 		err = puterrno4(err);
9493 		return (err);
9494 	}
9495 
9496 	sp->rs_share_access = 0;
9497 	sp->rs_share_deny = 0;
9498 
9499 	return (0);
9500 
9501 }
9502 
9503 static int
9504 rdma_setup_read_data4(READ4args *args, READ4res *rok)
9505 {
9506 	struct clist	*wcl;
9507 	count4		count = rok->data_len;
9508 	int		wlist_len;
9509 
9510 	wcl = args->wlist;
9511 	if (rdma_setup_read_chunks(wcl, count, &wlist_len) == FALSE) {
9512 		return (FALSE);
9513 	}
9514 	wcl = args->wlist;
9515 	rok->wlist_len = wlist_len;
9516 	rok->wlist = wcl;
9517 	return (TRUE);
9518 }
9519 
9520 /* tunable to disable server referrals */
9521 int rfs4_no_referrals = 0;
9522 
9523 /*
9524  * Find an NFS record in reparse point data.
9525  * Returns 0 for success and <0 or an errno value on failure.
9526  */
9527 int
9528 vn_find_nfs_record(vnode_t *vp, nvlist_t **nvlp, char **svcp, char **datap)
9529 {
9530 	int err;
9531 	char *stype, *val;
9532 	nvlist_t *nvl;
9533 	nvpair_t *curr;
9534 
9535 	if ((nvl = reparse_init()) == NULL)
9536 		return (-1);
9537 
9538 	if ((err = reparse_vnode_parse(vp, nvl)) != 0) {
9539 		reparse_free(nvl);
9540 		return (err);
9541 	}
9542 
9543 	curr = NULL;
9544 	while ((curr = nvlist_next_nvpair(nvl, curr)) != NULL) {
9545 		if ((stype = nvpair_name(curr)) == NULL) {
9546 			reparse_free(nvl);
9547 			return (-2);
9548 		}
9549 		if (strncasecmp(stype, "NFS", 3) == 0)
9550 			break;
9551 	}
9552 
9553 	if ((curr == NULL) ||
9554 	    (nvpair_value_string(curr, &val))) {
9555 		reparse_free(nvl);
9556 		return (-3);
9557 	}
9558 	*nvlp = nvl;
9559 	*svcp = stype;
9560 	*datap = val;
9561 	return (0);
9562 }
9563 
9564 int
9565 vn_is_nfs_reparse(vnode_t *vp, cred_t *cr)
9566 {
9567 	nvlist_t *nvl;
9568 	char *s, *d;
9569 
9570 	if (rfs4_no_referrals != 0)
9571 		return (B_FALSE);
9572 
9573 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9574 		return (B_FALSE);
9575 
9576 	if (vn_find_nfs_record(vp, &nvl, &s, &d) != 0)
9577 		return (B_FALSE);
9578 
9579 	reparse_free(nvl);
9580 
9581 	return (B_TRUE);
9582 }
9583 
9584 /*
9585  * There is a user-level copy of this routine in ref_subr.c.
9586  * Changes should be kept in sync.
9587  */
9588 static int
9589 nfs4_create_components(char *path, component4 *comp4)
9590 {
9591 	int slen, plen, ncomp;
9592 	char *ori_path, *nxtc, buf[MAXNAMELEN];
9593 
9594 	if (path == NULL)
9595 		return (0);
9596 
9597 	plen = strlen(path) + 1;	/* include the terminator */
9598 	ori_path = path;
9599 	ncomp = 0;
9600 
9601 	/* count number of components in the path */
9602 	for (nxtc = path; nxtc < ori_path + plen; nxtc++) {
9603 		if (*nxtc == '/' || *nxtc == '\0' || *nxtc == '\n') {
9604 			if ((slen = nxtc - path) == 0) {
9605 				path = nxtc + 1;
9606 				continue;
9607 			}
9608 
9609 			if (comp4 != NULL) {
9610 				bcopy(path, buf, slen);
9611 				buf[slen] = '\0';
9612 				(void) str_to_utf8(buf, &comp4[ncomp]);
9613 			}
9614 
9615 			ncomp++;	/* 1 valid component */
9616 			path = nxtc + 1;
9617 		}
9618 		if (*nxtc == '\0' || *nxtc == '\n')
9619 			break;
9620 	}
9621 
9622 	return (ncomp);
9623 }
9624 
9625 /*
9626  * There is a user-level copy of this routine in ref_subr.c.
9627  * Changes should be kept in sync.
9628  */
9629 static int
9630 make_pathname4(char *path, pathname4 *pathname)
9631 {
9632 	int ncomp;
9633 	component4 *comp4;
9634 
9635 	if (pathname == NULL)
9636 		return (0);
9637 
9638 	if (path == NULL) {
9639 		pathname->pathname4_val = NULL;
9640 		pathname->pathname4_len = 0;
9641 		return (0);
9642 	}
9643 
9644 	/* count number of components to alloc buffer */
9645 	if ((ncomp = nfs4_create_components(path, NULL)) == 0) {
9646 		pathname->pathname4_val = NULL;
9647 		pathname->pathname4_len = 0;
9648 		return (0);
9649 	}
9650 	comp4 = kmem_zalloc(ncomp * sizeof (component4), KM_SLEEP);
9651 
9652 	/* copy components into allocated buffer */
9653 	ncomp = nfs4_create_components(path, comp4);
9654 
9655 	pathname->pathname4_val = comp4;
9656 	pathname->pathname4_len = ncomp;
9657 
9658 	return (ncomp);
9659 }
9660 
9661 #define	xdr_fs_locations4 xdr_fattr4_fs_locations
9662 
9663 fs_locations4 *
9664 fetch_referral(vnode_t *vp, cred_t *cr)
9665 {
9666 	nvlist_t *nvl;
9667 	char *stype, *sdata;
9668 	fs_locations4 *result;
9669 	char buf[1024];
9670 	size_t bufsize;
9671 	XDR xdr;
9672 	int err;
9673 
9674 	/*
9675 	 * Check attrs to ensure it's a reparse point
9676 	 */
9677 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9678 		return (NULL);
9679 
9680 	/*
9681 	 * Look for an NFS record and get the type and data
9682 	 */
9683 	if (vn_find_nfs_record(vp, &nvl, &stype, &sdata) != 0)
9684 		return (NULL);
9685 
9686 	/*
9687 	 * With the type and data, upcall to get the referral
9688 	 */
9689 	bufsize = sizeof (buf);
9690 	bzero(buf, sizeof (buf));
9691 	err = reparse_kderef((const char *)stype, (const char *)sdata,
9692 	    buf, &bufsize);
9693 	reparse_free(nvl);
9694 
9695 	DTRACE_PROBE4(nfs4serv__func__referral__upcall,
9696 	    char *, stype, char *, sdata, char *, buf, int, err);
9697 	if (err) {
9698 		cmn_err(CE_NOTE,
9699 		    "reparsed daemon not running: unable to get referral (%d)",
9700 		    err);
9701 		return (NULL);
9702 	}
9703 
9704 	/*
9705 	 * We get an XDR'ed record back from the kderef call
9706 	 */
9707 	xdrmem_create(&xdr, buf, bufsize, XDR_DECODE);
9708 	result = kmem_alloc(sizeof (fs_locations4), KM_SLEEP);
9709 	err = xdr_fs_locations4(&xdr, result);
9710 	XDR_DESTROY(&xdr);
9711 	if (err != TRUE) {
9712 		DTRACE_PROBE1(nfs4serv__func__referral__upcall__xdrfail,
9713 		    int, err);
9714 		return (NULL);
9715 	}
9716 
9717 	/*
9718 	 * Look at path to recover fs_root, ignoring the leading '/'
9719 	 */
9720 	(void) make_pathname4(vp->v_path, &result->fs_root);
9721 
9722 	return (result);
9723 }
9724 
9725 char *
9726 build_symlink(vnode_t *vp, cred_t *cr, size_t *strsz)
9727 {
9728 	fs_locations4 *fsl;
9729 	fs_location4 *fs;
9730 	char *server, *path, *symbuf;
9731 	static char *prefix = "/net/";
9732 	int i, size, npaths;
9733 	uint_t len;
9734 
9735 	/* Get the referral */
9736 	if ((fsl = fetch_referral(vp, cr)) == NULL)
9737 		return (NULL);
9738 
9739 	/* Deal with only the first location and first server */
9740 	fs = &fsl->locations_val[0];
9741 	server = utf8_to_str(&fs->server_val[0], &len, NULL);
9742 	if (server == NULL) {
9743 		rfs4_free_fs_locations4(fsl);
9744 		kmem_free(fsl, sizeof (fs_locations4));
9745 		return (NULL);
9746 	}
9747 
9748 	/* Figure out size for "/net/" + host + /path/path/path + NULL */
9749 	size = strlen(prefix) + len;
9750 	for (i = 0; i < fs->rootpath.pathname4_len; i++)
9751 		size += fs->rootpath.pathname4_val[i].utf8string_len + 1;
9752 
9753 	/* Allocate the symlink buffer and fill it */
9754 	symbuf = kmem_zalloc(size, KM_SLEEP);
9755 	(void) strcat(symbuf, prefix);
9756 	(void) strcat(symbuf, server);
9757 	kmem_free(server, len);
9758 
9759 	npaths = 0;
9760 	for (i = 0; i < fs->rootpath.pathname4_len; i++) {
9761 		path = utf8_to_str(&fs->rootpath.pathname4_val[i], &len, NULL);
9762 		if (path == NULL)
9763 			continue;
9764 		(void) strcat(symbuf, "/");
9765 		(void) strcat(symbuf, path);
9766 		npaths++;
9767 		kmem_free(path, len);
9768 	}
9769 
9770 	rfs4_free_fs_locations4(fsl);
9771 	kmem_free(fsl, sizeof (fs_locations4));
9772 
9773 	if (strsz != NULL)
9774 		*strsz = size;
9775 	return (symbuf);
9776 }
9777 
9778 /*
9779  * Check to see if we have a downrev Solaris client, so that we
9780  * can send it a symlink instead of a referral.
9781  */
9782 int
9783 client_is_downrev(struct svc_req *req)
9784 {
9785 	struct sockaddr *ca;
9786 	rfs4_clntip_t *ci;
9787 	bool_t create = FALSE;
9788 	int is_downrev;
9789 
9790 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
9791 	ASSERT(ca);
9792 	ci = rfs4_find_clntip(ca, &create);
9793 	if (ci == NULL)
9794 		return (0);
9795 	is_downrev = ci->ri_no_referrals;
9796 	rfs4_dbe_rele(ci->ri_dbe);
9797 	return (is_downrev);
9798 }
9799