xref: /titanic_51/usr/src/uts/common/fs/nfs/nfs4_srv.c (revision 8fc99e42676a23421c75e76660640f9765d693b1)
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 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
28  *	All Rights Reserved
29  */
30 
31 #include <sys/param.h>
32 #include <sys/types.h>
33 #include <sys/systm.h>
34 #include <sys/cred.h>
35 #include <sys/buf.h>
36 #include <sys/vfs.h>
37 #include <sys/vfs_opreg.h>
38 #include <sys/vnode.h>
39 #include <sys/uio.h>
40 #include <sys/errno.h>
41 #include <sys/sysmacros.h>
42 #include <sys/statvfs.h>
43 #include <sys/kmem.h>
44 #include <sys/dirent.h>
45 #include <sys/cmn_err.h>
46 #include <sys/debug.h>
47 #include <sys/systeminfo.h>
48 #include <sys/flock.h>
49 #include <sys/pathname.h>
50 #include <sys/nbmlock.h>
51 #include <sys/share.h>
52 #include <sys/atomic.h>
53 #include <sys/policy.h>
54 #include <sys/fem.h>
55 #include <sys/sdt.h>
56 #include <sys/ddi.h>
57 #include <sys/zone.h>
58 
59 #include <fs/fs_reparse.h>
60 
61 #include <rpc/types.h>
62 #include <rpc/auth.h>
63 #include <rpc/rpcsec_gss.h>
64 #include <rpc/svc.h>
65 
66 #include <nfs/nfs.h>
67 #include <nfs/export.h>
68 #include <nfs/nfs_cmd.h>
69 #include <nfs/lm.h>
70 #include <nfs/nfs4.h>
71 
72 #include <sys/strsubr.h>
73 #include <sys/strsun.h>
74 
75 #include <inet/common.h>
76 #include <inet/ip.h>
77 #include <inet/ip6.h>
78 
79 #include <sys/tsol/label.h>
80 #include <sys/tsol/tndb.h>
81 
82 #define	RFS4_MAXLOCK_TRIES 4	/* Try to get the lock this many times */
83 static int rfs4_maxlock_tries = RFS4_MAXLOCK_TRIES;
84 #define	RFS4_LOCK_DELAY 10	/* Milliseconds */
85 static clock_t  rfs4_lock_delay = RFS4_LOCK_DELAY;
86 extern struct svc_ops rdma_svc_ops;
87 extern int nfs_loaned_buffers;
88 /* End of Tunables */
89 
90 static int rdma_setup_read_data4(READ4args *, READ4res *);
91 
92 /*
93  * Used to bump the stateid4.seqid value and show changes in the stateid
94  */
95 #define	next_stateid(sp) (++(sp)->bits.chgseq)
96 
97 /*
98  * RFS4_MINLEN_ENTRY4: XDR-encoded size of smallest possible dirent.
99  *	This is used to return NFS4ERR_TOOSMALL when clients specify
100  *	maxcount that isn't large enough to hold the smallest possible
101  *	XDR encoded dirent.
102  *
103  *	    sizeof cookie (8 bytes) +
104  *	    sizeof name_len (4 bytes) +
105  *	    sizeof smallest (padded) name (4 bytes) +
106  *	    sizeof bitmap4_len (12 bytes) +   NOTE: we always encode len=2 bm4
107  *	    sizeof attrlist4_len (4 bytes) +
108  *	    sizeof next boolean (4 bytes)
109  *
110  * RFS4_MINLEN_RDDIR4: XDR-encoded size of READDIR op reply containing
111  * the smallest possible entry4 (assumes no attrs requested).
112  *	sizeof nfsstat4 (4 bytes) +
113  *	sizeof verifier4 (8 bytes) +
114  *	sizeof entry4list bool (4 bytes) +
115  *	sizeof entry4 	(36 bytes) +
116  *	sizeof eof bool  (4 bytes)
117  *
118  * RFS4_MINLEN_RDDIR_BUF: minimum length of buffer server will provide to
119  *	VOP_READDIR.  Its value is the size of the maximum possible dirent
120  *	for solaris.  The DIRENT64_RECLEN macro returns	the size of dirent
121  *	required for a given name length.  MAXNAMELEN is the maximum
122  *	filename length allowed in Solaris.  The first two DIRENT64_RECLEN()
123  *	macros are to allow for . and .. entries -- just a minor tweak to try
124  *	and guarantee that buffer we give to VOP_READDIR will be large enough
125  *	to hold ., .., and the largest possible solaris dirent64.
126  */
127 #define	RFS4_MINLEN_ENTRY4 36
128 #define	RFS4_MINLEN_RDDIR4 (4 + NFS4_VERIFIER_SIZE + 4 + RFS4_MINLEN_ENTRY4 + 4)
129 #define	RFS4_MINLEN_RDDIR_BUF \
130 	(DIRENT64_RECLEN(1) + DIRENT64_RECLEN(2) + DIRENT64_RECLEN(MAXNAMELEN))
131 
132 /*
133  * It would be better to pad to 4 bytes since that's what XDR would do,
134  * but the dirents UFS gives us are already padded to 8, so just take
135  * what we're given.  Dircount is only a hint anyway.  Currently the
136  * solaris kernel is ASCII only, so there's no point in calling the
137  * UTF8 functions.
138  *
139  * dirent64: named padded to provide 8 byte struct alignment
140  *	d_ino(8) + d_off(8) + d_reclen(2) + d_name(namelen + null(1) + pad)
141  *
142  * cookie: uint64_t   +  utf8namelen: uint_t  +   utf8name padded to 8 bytes
143  *
144  */
145 #define	DIRENT64_TO_DIRCOUNT(dp) \
146 	(3 * BYTES_PER_XDR_UNIT + DIRENT64_NAMELEN((dp)->d_reclen))
147 
148 time_t rfs4_start_time;			/* Initialized in rfs4_srvrinit */
149 
150 static sysid_t lockt_sysid;		/* dummy sysid for all LOCKT calls */
151 
152 u_longlong_t	nfs4_srv_caller_id;
153 uint_t		nfs4_srv_vkey = 0;
154 
155 verifier4	Write4verf;
156 verifier4	Readdir4verf;
157 
158 void	rfs4_init_compound_state(struct compound_state *);
159 
160 static void	nullfree(caddr_t);
161 static void	rfs4_op_inval(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
162 			struct compound_state *);
163 static void	rfs4_op_access(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
164 			struct compound_state *);
165 static void	rfs4_op_close(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
166 			struct compound_state *);
167 static void	rfs4_op_commit(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
168 			struct compound_state *);
169 static void	rfs4_op_create(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
170 			struct compound_state *);
171 static void	rfs4_op_create_free(nfs_resop4 *resop);
172 static void	rfs4_op_delegreturn(nfs_argop4 *, nfs_resop4 *,
173 			struct svc_req *, struct compound_state *);
174 static void	rfs4_op_delegpurge(nfs_argop4 *, nfs_resop4 *,
175 			struct svc_req *, struct compound_state *);
176 static void	rfs4_op_getattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
177 			struct compound_state *);
178 static void	rfs4_op_getattr_free(nfs_resop4 *);
179 static void	rfs4_op_getfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
180 			struct compound_state *);
181 static void	rfs4_op_getfh_free(nfs_resop4 *);
182 static void	rfs4_op_illegal(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
183 			struct compound_state *);
184 static void	rfs4_op_link(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
185 			struct compound_state *);
186 static void	rfs4_op_lock(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
187 			struct compound_state *);
188 static void	lock_denied_free(nfs_resop4 *);
189 static void	rfs4_op_locku(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
190 			struct compound_state *);
191 static void	rfs4_op_lockt(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
192 			struct compound_state *);
193 static void	rfs4_op_lookup(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
194 			struct compound_state *);
195 static void	rfs4_op_lookupp(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
196 			struct compound_state *);
197 static void	rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop,
198 				struct svc_req *req, struct compound_state *cs);
199 static void	rfs4_op_nverify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
200 			struct compound_state *);
201 static void	rfs4_op_open(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
202 			struct compound_state *);
203 static void	rfs4_op_open_confirm(nfs_argop4 *, nfs_resop4 *,
204 			struct svc_req *, struct compound_state *);
205 static void	rfs4_op_open_downgrade(nfs_argop4 *, nfs_resop4 *,
206 			struct svc_req *, struct compound_state *);
207 static void	rfs4_op_putfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
208 			struct compound_state *);
209 static void	rfs4_op_putpubfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
210 			struct compound_state *);
211 static void	rfs4_op_putrootfh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
212 			struct compound_state *);
213 static void	rfs4_op_read(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
214 			struct compound_state *);
215 static void	rfs4_op_read_free(nfs_resop4 *);
216 static void	rfs4_op_readdir_free(nfs_resop4 *resop);
217 static void	rfs4_op_readlink(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
218 			struct compound_state *);
219 static void	rfs4_op_readlink_free(nfs_resop4 *);
220 static void	rfs4_op_release_lockowner(nfs_argop4 *, nfs_resop4 *,
221 			struct svc_req *, struct compound_state *);
222 static void	rfs4_op_remove(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
223 			struct compound_state *);
224 static void	rfs4_op_rename(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
225 			struct compound_state *);
226 static void	rfs4_op_renew(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
227 			struct compound_state *);
228 static void	rfs4_op_restorefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
229 			struct compound_state *);
230 static void	rfs4_op_savefh(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
231 			struct compound_state *);
232 static void	rfs4_op_setattr(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
233 			struct compound_state *);
234 static void	rfs4_op_verify(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
235 			struct compound_state *);
236 static void	rfs4_op_write(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
237 			struct compound_state *);
238 static void	rfs4_op_setclientid(nfs_argop4 *, nfs_resop4 *,
239 			struct svc_req *, struct compound_state *);
240 static void	rfs4_op_setclientid_confirm(nfs_argop4 *, nfs_resop4 *,
241 			struct svc_req *req, struct compound_state *);
242 static void	rfs4_op_secinfo(nfs_argop4 *, nfs_resop4 *, struct svc_req *,
243 			struct compound_state *);
244 static void	rfs4_op_secinfo_free(nfs_resop4 *);
245 
246 static nfsstat4 check_open_access(uint32_t,
247 				struct compound_state *, struct svc_req *);
248 nfsstat4 rfs4_client_sysid(rfs4_client_t *, sysid_t *);
249 void rfs4_ss_clid(rfs4_client_t *);
250 
251 /*
252  * translation table for attrs
253  */
254 struct nfs4_ntov_table {
255 	union nfs4_attr_u *na;
256 	uint8_t amap[NFS4_MAXNUM_ATTRS];
257 	int attrcnt;
258 	bool_t vfsstat;
259 };
260 
261 static void	nfs4_ntov_table_init(struct nfs4_ntov_table *ntovp);
262 static void	nfs4_ntov_table_free(struct nfs4_ntov_table *ntovp,
263 				    struct nfs4_svgetit_arg *sargp);
264 
265 static nfsstat4	do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp,
266 		    struct compound_state *cs, struct nfs4_svgetit_arg *sargp,
267 		    struct nfs4_ntov_table *ntovp, nfs4_attr_cmd_t cmd);
268 
269 fem_t		*deleg_rdops;
270 fem_t		*deleg_wrops;
271 
272 rfs4_servinst_t *rfs4_cur_servinst = NULL;	/* current server instance */
273 kmutex_t	rfs4_servinst_lock;	/* protects linked list */
274 int		rfs4_seen_first_compound;	/* set first time we see one */
275 
276 /*
277  * NFS4 op dispatch table
278  */
279 
280 struct rfsv4disp {
281 	void	(*dis_proc)();		/* proc to call */
282 	void	(*dis_resfree)();	/* frees space allocated by proc */
283 	int	dis_flags;		/* RPC_IDEMPOTENT, etc... */
284 };
285 
286 static struct rfsv4disp rfsv4disptab[] = {
287 	/*
288 	 * NFS VERSION 4
289 	 */
290 
291 	/* RFS_NULL = 0 */
292 	{rfs4_op_illegal, nullfree, 0},
293 
294 	/* UNUSED = 1 */
295 	{rfs4_op_illegal, nullfree, 0},
296 
297 	/* UNUSED = 2 */
298 	{rfs4_op_illegal, nullfree, 0},
299 
300 	/* OP_ACCESS = 3 */
301 	{rfs4_op_access, nullfree, RPC_IDEMPOTENT},
302 
303 	/* OP_CLOSE = 4 */
304 	{rfs4_op_close, nullfree, 0},
305 
306 	/* OP_COMMIT = 5 */
307 	{rfs4_op_commit, nullfree, RPC_IDEMPOTENT},
308 
309 	/* OP_CREATE = 6 */
310 	{rfs4_op_create, nullfree, 0},
311 
312 	/* OP_DELEGPURGE = 7 */
313 	{rfs4_op_delegpurge, nullfree, 0},
314 
315 	/* OP_DELEGRETURN = 8 */
316 	{rfs4_op_delegreturn, nullfree, 0},
317 
318 	/* OP_GETATTR = 9 */
319 	{rfs4_op_getattr, rfs4_op_getattr_free, RPC_IDEMPOTENT},
320 
321 	/* OP_GETFH = 10 */
322 	{rfs4_op_getfh, rfs4_op_getfh_free, RPC_ALL},
323 
324 	/* OP_LINK = 11 */
325 	{rfs4_op_link, nullfree, 0},
326 
327 	/* OP_LOCK = 12 */
328 	{rfs4_op_lock, lock_denied_free, 0},
329 
330 	/* OP_LOCKT = 13 */
331 	{rfs4_op_lockt, lock_denied_free, 0},
332 
333 	/* OP_LOCKU = 14 */
334 	{rfs4_op_locku, nullfree, 0},
335 
336 	/* OP_LOOKUP = 15 */
337 	{rfs4_op_lookup, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
338 
339 	/* OP_LOOKUPP = 16 */
340 	{rfs4_op_lookupp, nullfree, (RPC_IDEMPOTENT | RPC_PUBLICFH_OK)},
341 
342 	/* OP_NVERIFY = 17 */
343 	{rfs4_op_nverify, nullfree, RPC_IDEMPOTENT},
344 
345 	/* OP_OPEN = 18 */
346 	{rfs4_op_open, rfs4_free_reply, 0},
347 
348 	/* OP_OPENATTR = 19 */
349 	{rfs4_op_openattr, nullfree, 0},
350 
351 	/* OP_OPEN_CONFIRM = 20 */
352 	{rfs4_op_open_confirm, nullfree, 0},
353 
354 	/* OP_OPEN_DOWNGRADE = 21 */
355 	{rfs4_op_open_downgrade, nullfree, 0},
356 
357 	/* OP_OPEN_PUTFH = 22 */
358 	{rfs4_op_putfh, nullfree, RPC_ALL},
359 
360 	/* OP_PUTPUBFH = 23 */
361 	{rfs4_op_putpubfh, nullfree, RPC_ALL},
362 
363 	/* OP_PUTROOTFH = 24 */
364 	{rfs4_op_putrootfh, nullfree, RPC_ALL},
365 
366 	/* OP_READ = 25 */
367 	{rfs4_op_read, rfs4_op_read_free, RPC_IDEMPOTENT},
368 
369 	/* OP_READDIR = 26 */
370 	{rfs4_op_readdir, rfs4_op_readdir_free, RPC_IDEMPOTENT},
371 
372 	/* OP_READLINK = 27 */
373 	{rfs4_op_readlink, rfs4_op_readlink_free, RPC_IDEMPOTENT},
374 
375 	/* OP_REMOVE = 28 */
376 	{rfs4_op_remove, nullfree, 0},
377 
378 	/* OP_RENAME = 29 */
379 	{rfs4_op_rename, nullfree, 0},
380 
381 	/* OP_RENEW = 30 */
382 	{rfs4_op_renew, nullfree, 0},
383 
384 	/* OP_RESTOREFH = 31 */
385 	{rfs4_op_restorefh, nullfree, RPC_ALL},
386 
387 	/* OP_SAVEFH = 32 */
388 	{rfs4_op_savefh, nullfree, RPC_ALL},
389 
390 	/* OP_SECINFO = 33 */
391 	{rfs4_op_secinfo, rfs4_op_secinfo_free, 0},
392 
393 	/* OP_SETATTR = 34 */
394 	{rfs4_op_setattr, nullfree, 0},
395 
396 	/* OP_SETCLIENTID = 35 */
397 	{rfs4_op_setclientid, nullfree, 0},
398 
399 	/* OP_SETCLIENTID_CONFIRM = 36 */
400 	{rfs4_op_setclientid_confirm, nullfree, 0},
401 
402 	/* OP_VERIFY = 37 */
403 	{rfs4_op_verify, nullfree, RPC_IDEMPOTENT},
404 
405 	/* OP_WRITE = 38 */
406 	{rfs4_op_write, nullfree, 0},
407 
408 	/* OP_RELEASE_LOCKOWNER = 39 */
409 	{rfs4_op_release_lockowner, nullfree, 0},
410 };
411 
412 static uint_t rfsv4disp_cnt = sizeof (rfsv4disptab) / sizeof (rfsv4disptab[0]);
413 
414 #define	OP_ILLEGAL_IDX (rfsv4disp_cnt)
415 
416 #ifdef DEBUG
417 
418 int		rfs4_fillone_debug = 0;
419 int		rfs4_no_stub_access = 1;
420 int		rfs4_rddir_debug = 0;
421 
422 static char    *rfs4_op_string[] = {
423 	"rfs4_op_null",
424 	"rfs4_op_1 unused",
425 	"rfs4_op_2 unused",
426 	"rfs4_op_access",
427 	"rfs4_op_close",
428 	"rfs4_op_commit",
429 	"rfs4_op_create",
430 	"rfs4_op_delegpurge",
431 	"rfs4_op_delegreturn",
432 	"rfs4_op_getattr",
433 	"rfs4_op_getfh",
434 	"rfs4_op_link",
435 	"rfs4_op_lock",
436 	"rfs4_op_lockt",
437 	"rfs4_op_locku",
438 	"rfs4_op_lookup",
439 	"rfs4_op_lookupp",
440 	"rfs4_op_nverify",
441 	"rfs4_op_open",
442 	"rfs4_op_openattr",
443 	"rfs4_op_open_confirm",
444 	"rfs4_op_open_downgrade",
445 	"rfs4_op_putfh",
446 	"rfs4_op_putpubfh",
447 	"rfs4_op_putrootfh",
448 	"rfs4_op_read",
449 	"rfs4_op_readdir",
450 	"rfs4_op_readlink",
451 	"rfs4_op_remove",
452 	"rfs4_op_rename",
453 	"rfs4_op_renew",
454 	"rfs4_op_restorefh",
455 	"rfs4_op_savefh",
456 	"rfs4_op_secinfo",
457 	"rfs4_op_setattr",
458 	"rfs4_op_setclientid",
459 	"rfs4_op_setclient_confirm",
460 	"rfs4_op_verify",
461 	"rfs4_op_write",
462 	"rfs4_op_release_lockowner",
463 	"rfs4_op_illegal"
464 };
465 #endif
466 
467 void	rfs4_ss_chkclid(rfs4_client_t *);
468 
469 extern size_t   strlcpy(char *dst, const char *src, size_t dstsize);
470 
471 extern void	rfs4_free_fs_locations4(fs_locations4 *);
472 
473 #ifdef	nextdp
474 #undef nextdp
475 #endif
476 #define	nextdp(dp)	((struct dirent64 *)((char *)(dp) + (dp)->d_reclen))
477 
478 static const fs_operation_def_t nfs4_rd_deleg_tmpl[] = {
479 	VOPNAME_OPEN,		{ .femop_open = deleg_rd_open },
480 	VOPNAME_WRITE,		{ .femop_write = deleg_rd_write },
481 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_rd_setattr },
482 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_rd_rwlock },
483 	VOPNAME_SPACE,		{ .femop_space = deleg_rd_space },
484 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_rd_setsecattr },
485 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_rd_vnevent },
486 	NULL,			NULL
487 };
488 static const fs_operation_def_t nfs4_wr_deleg_tmpl[] = {
489 	VOPNAME_OPEN,		{ .femop_open = deleg_wr_open },
490 	VOPNAME_READ,		{ .femop_read = deleg_wr_read },
491 	VOPNAME_WRITE,		{ .femop_write = deleg_wr_write },
492 	VOPNAME_SETATTR,	{ .femop_setattr = deleg_wr_setattr },
493 	VOPNAME_RWLOCK,		{ .femop_rwlock = deleg_wr_rwlock },
494 	VOPNAME_SPACE,		{ .femop_space = deleg_wr_space },
495 	VOPNAME_SETSECATTR,	{ .femop_setsecattr = deleg_wr_setsecattr },
496 	VOPNAME_VNEVENT,	{ .femop_vnevent = deleg_wr_vnevent },
497 	NULL,			NULL
498 };
499 
500 int
501 rfs4_srvrinit(void)
502 {
503 	timespec32_t verf;
504 	int error;
505 	extern void rfs4_attr_init();
506 	extern krwlock_t rfs4_deleg_policy_lock;
507 
508 	/*
509 	 * The following algorithm attempts to find a unique verifier
510 	 * to be used as the write verifier returned from the server
511 	 * to the client.  It is important that this verifier change
512 	 * whenever the server reboots.  Of secondary importance, it
513 	 * is important for the verifier to be unique between two
514 	 * different servers.
515 	 *
516 	 * Thus, an attempt is made to use the system hostid and the
517 	 * current time in seconds when the nfssrv kernel module is
518 	 * loaded.  It is assumed that an NFS server will not be able
519 	 * to boot and then to reboot in less than a second.  If the
520 	 * hostid has not been set, then the current high resolution
521 	 * time is used.  This will ensure different verifiers each
522 	 * time the server reboots and minimize the chances that two
523 	 * different servers will have the same verifier.
524 	 * XXX - this is broken on LP64 kernels.
525 	 */
526 	verf.tv_sec = (time_t)zone_get_hostid(NULL);
527 	if (verf.tv_sec != 0) {
528 		verf.tv_nsec = gethrestime_sec();
529 	} else {
530 		timespec_t tverf;
531 
532 		gethrestime(&tverf);
533 		verf.tv_sec = (time_t)tverf.tv_sec;
534 		verf.tv_nsec = tverf.tv_nsec;
535 	}
536 
537 	Write4verf = *(uint64_t *)&verf;
538 
539 	rfs4_attr_init();
540 	mutex_init(&rfs4_deleg_lock, NULL, MUTEX_DEFAULT, NULL);
541 
542 	/* Used to manage create/destroy of server state */
543 	mutex_init(&rfs4_state_lock, NULL, MUTEX_DEFAULT, NULL);
544 
545 	/* Used to manage access to server instance linked list */
546 	mutex_init(&rfs4_servinst_lock, NULL, MUTEX_DEFAULT, NULL);
547 
548 	/* Used to manage access to rfs4_deleg_policy */
549 	rw_init(&rfs4_deleg_policy_lock, NULL, RW_DEFAULT, NULL);
550 
551 	error = fem_create("deleg_rdops", nfs4_rd_deleg_tmpl, &deleg_rdops);
552 	if (error != 0) {
553 		rfs4_disable_delegation();
554 	} else {
555 		error = fem_create("deleg_wrops", nfs4_wr_deleg_tmpl,
556 		    &deleg_wrops);
557 		if (error != 0) {
558 			rfs4_disable_delegation();
559 			fem_free(deleg_rdops);
560 		}
561 	}
562 
563 	nfs4_srv_caller_id = fs_new_caller_id();
564 
565 	lockt_sysid = lm_alloc_sysidt();
566 
567 	vsd_create(&nfs4_srv_vkey, NULL);
568 
569 	return (0);
570 }
571 
572 void
573 rfs4_srvrfini(void)
574 {
575 	extern krwlock_t rfs4_deleg_policy_lock;
576 
577 	if (lockt_sysid != LM_NOSYSID) {
578 		lm_free_sysidt(lockt_sysid);
579 		lockt_sysid = LM_NOSYSID;
580 	}
581 
582 	mutex_destroy(&rfs4_deleg_lock);
583 	mutex_destroy(&rfs4_state_lock);
584 	rw_destroy(&rfs4_deleg_policy_lock);
585 
586 	fem_free(deleg_rdops);
587 	fem_free(deleg_wrops);
588 }
589 
590 void
591 rfs4_init_compound_state(struct compound_state *cs)
592 {
593 	bzero(cs, sizeof (*cs));
594 	cs->cont = TRUE;
595 	cs->access = CS_ACCESS_DENIED;
596 	cs->deleg = FALSE;
597 	cs->mandlock = FALSE;
598 	cs->fh.nfs_fh4_val = cs->fhbuf;
599 }
600 
601 void
602 rfs4_grace_start(rfs4_servinst_t *sip)
603 {
604 	rw_enter(&sip->rwlock, RW_WRITER);
605 	sip->start_time = (time_t)TICK_TO_SEC(ddi_get_lbolt());
606 	sip->grace_period = rfs4_grace_period;
607 	rw_exit(&sip->rwlock);
608 }
609 
610 /*
611  * returns true if the instance's grace period has never been started
612  */
613 int
614 rfs4_servinst_grace_new(rfs4_servinst_t *sip)
615 {
616 	time_t start_time;
617 
618 	rw_enter(&sip->rwlock, RW_READER);
619 	start_time = sip->start_time;
620 	rw_exit(&sip->rwlock);
621 
622 	return (start_time == 0);
623 }
624 
625 /*
626  * Indicates if server instance is within the
627  * grace period.
628  */
629 int
630 rfs4_servinst_in_grace(rfs4_servinst_t *sip)
631 {
632 	time_t grace_expiry;
633 
634 	rw_enter(&sip->rwlock, RW_READER);
635 	grace_expiry = sip->start_time + sip->grace_period;
636 	rw_exit(&sip->rwlock);
637 
638 	return (((time_t)TICK_TO_SEC(ddi_get_lbolt())) < grace_expiry);
639 }
640 
641 int
642 rfs4_clnt_in_grace(rfs4_client_t *cp)
643 {
644 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
645 
646 	return (rfs4_servinst_in_grace(cp->rc_server_instance));
647 }
648 
649 /*
650  * reset all currently active grace periods
651  */
652 void
653 rfs4_grace_reset_all(void)
654 {
655 	rfs4_servinst_t *sip;
656 
657 	mutex_enter(&rfs4_servinst_lock);
658 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
659 		if (rfs4_servinst_in_grace(sip))
660 			rfs4_grace_start(sip);
661 	mutex_exit(&rfs4_servinst_lock);
662 }
663 
664 /*
665  * start any new instances' grace periods
666  */
667 void
668 rfs4_grace_start_new(void)
669 {
670 	rfs4_servinst_t *sip;
671 
672 	mutex_enter(&rfs4_servinst_lock);
673 	for (sip = rfs4_cur_servinst; sip != NULL; sip = sip->prev)
674 		if (rfs4_servinst_grace_new(sip))
675 			rfs4_grace_start(sip);
676 	mutex_exit(&rfs4_servinst_lock);
677 }
678 
679 static rfs4_dss_path_t *
680 rfs4_dss_newpath(rfs4_servinst_t *sip, char *path, unsigned index)
681 {
682 	size_t len;
683 	rfs4_dss_path_t *dss_path;
684 
685 	dss_path = kmem_alloc(sizeof (rfs4_dss_path_t), KM_SLEEP);
686 
687 	/*
688 	 * Take a copy of the string, since the original may be overwritten.
689 	 * Sadly, no strdup() in the kernel.
690 	 */
691 	/* allow for NUL */
692 	len = strlen(path) + 1;
693 	dss_path->path = kmem_alloc(len, KM_SLEEP);
694 	(void) strlcpy(dss_path->path, path, len);
695 
696 	/* associate with servinst */
697 	dss_path->sip = sip;
698 	dss_path->index = index;
699 
700 	/*
701 	 * Add to list of served paths.
702 	 * No locking required, as we're only ever called at startup.
703 	 */
704 	if (rfs4_dss_pathlist == NULL) {
705 		/* this is the first dss_path_t */
706 
707 		/* needed for insque/remque */
708 		dss_path->next = dss_path->prev = dss_path;
709 
710 		rfs4_dss_pathlist = dss_path;
711 	} else {
712 		insque(dss_path, rfs4_dss_pathlist);
713 	}
714 
715 	return (dss_path);
716 }
717 
718 /*
719  * Create a new server instance, and make it the currently active instance.
720  * Note that starting the grace period too early will reduce the clients'
721  * recovery window.
722  */
723 void
724 rfs4_servinst_create(int start_grace, int dss_npaths, char **dss_paths)
725 {
726 	unsigned i;
727 	rfs4_servinst_t *sip;
728 	rfs4_oldstate_t *oldstate;
729 
730 	sip = kmem_alloc(sizeof (rfs4_servinst_t), KM_SLEEP);
731 	rw_init(&sip->rwlock, NULL, RW_DEFAULT, NULL);
732 
733 	sip->start_time = (time_t)0;
734 	sip->grace_period = (time_t)0;
735 	sip->next = NULL;
736 	sip->prev = NULL;
737 
738 	rw_init(&sip->oldstate_lock, NULL, RW_DEFAULT, NULL);
739 	/*
740 	 * This initial dummy entry is required to setup for insque/remque.
741 	 * It must be skipped over whenever the list is traversed.
742 	 */
743 	oldstate = kmem_alloc(sizeof (rfs4_oldstate_t), KM_SLEEP);
744 	/* insque/remque require initial list entry to be self-terminated */
745 	oldstate->next = oldstate;
746 	oldstate->prev = oldstate;
747 	sip->oldstate = oldstate;
748 
749 
750 	sip->dss_npaths = dss_npaths;
751 	sip->dss_paths = kmem_alloc(dss_npaths *
752 	    sizeof (rfs4_dss_path_t *), KM_SLEEP);
753 
754 	for (i = 0; i < dss_npaths; i++) {
755 		sip->dss_paths[i] = rfs4_dss_newpath(sip, dss_paths[i], i);
756 	}
757 
758 	mutex_enter(&rfs4_servinst_lock);
759 	if (rfs4_cur_servinst != NULL) {
760 		/* add to linked list */
761 		sip->prev = rfs4_cur_servinst;
762 		rfs4_cur_servinst->next = sip;
763 	}
764 	if (start_grace)
765 		rfs4_grace_start(sip);
766 	/* make the new instance "current" */
767 	rfs4_cur_servinst = sip;
768 
769 	mutex_exit(&rfs4_servinst_lock);
770 }
771 
772 /*
773  * In future, we might add a rfs4_servinst_destroy(sip) but, for now, destroy
774  * all instances directly.
775  */
776 void
777 rfs4_servinst_destroy_all(void)
778 {
779 	rfs4_servinst_t *sip, *prev, *current;
780 #ifdef DEBUG
781 	int n = 0;
782 #endif
783 
784 	mutex_enter(&rfs4_servinst_lock);
785 	ASSERT(rfs4_cur_servinst != NULL);
786 	current = rfs4_cur_servinst;
787 	rfs4_cur_servinst = NULL;
788 	for (sip = current; sip != NULL; sip = prev) {
789 		prev = sip->prev;
790 		rw_destroy(&sip->rwlock);
791 		if (sip->oldstate)
792 			kmem_free(sip->oldstate, sizeof (rfs4_oldstate_t));
793 		if (sip->dss_paths)
794 			kmem_free(sip->dss_paths,
795 			    sip->dss_npaths * sizeof (rfs4_dss_path_t *));
796 		kmem_free(sip, sizeof (rfs4_servinst_t));
797 #ifdef DEBUG
798 		n++;
799 #endif
800 	}
801 	mutex_exit(&rfs4_servinst_lock);
802 }
803 
804 /*
805  * Assign the current server instance to a client_t.
806  * Should be called with cp->rc_dbe held.
807  */
808 void
809 rfs4_servinst_assign(rfs4_client_t *cp, rfs4_servinst_t *sip)
810 {
811 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
812 
813 	/*
814 	 * The lock ensures that if the current instance is in the process
815 	 * of changing, we will see the new one.
816 	 */
817 	mutex_enter(&rfs4_servinst_lock);
818 	cp->rc_server_instance = sip;
819 	mutex_exit(&rfs4_servinst_lock);
820 }
821 
822 rfs4_servinst_t *
823 rfs4_servinst(rfs4_client_t *cp)
824 {
825 	ASSERT(rfs4_dbe_refcnt(cp->rc_dbe) > 0);
826 
827 	return (cp->rc_server_instance);
828 }
829 
830 /* ARGSUSED */
831 static void
832 nullfree(caddr_t resop)
833 {
834 }
835 
836 /*
837  * This is a fall-through for invalid or not implemented (yet) ops
838  */
839 /* ARGSUSED */
840 static void
841 rfs4_op_inval(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
842 	struct compound_state *cs)
843 {
844 	*cs->statusp = *((nfsstat4 *)&(resop)->nfs_resop4_u) = NFS4ERR_INVAL;
845 }
846 
847 /*
848  * Check if the security flavor, nfsnum, is in the flavor_list.
849  */
850 bool_t
851 in_flavor_list(int nfsnum, int *flavor_list, int count)
852 {
853 	int i;
854 
855 	for (i = 0; i < count; i++) {
856 		if (nfsnum == flavor_list[i])
857 			return (TRUE);
858 	}
859 	return (FALSE);
860 }
861 
862 /*
863  * Used by rfs4_op_secinfo to get the security information from the
864  * export structure associated with the component.
865  */
866 /* ARGSUSED */
867 static nfsstat4
868 do_rfs4_op_secinfo(struct compound_state *cs, char *nm, SECINFO4res *resp)
869 {
870 	int error, different_export = 0;
871 	vnode_t *dvp, *vp, *tvp;
872 	struct exportinfo *exi = NULL;
873 	fid_t fid;
874 	uint_t count, i;
875 	secinfo4 *resok_val;
876 	struct secinfo *secp;
877 	seconfig_t *si;
878 	bool_t did_traverse;
879 	int dotdot, walk;
880 
881 	dvp = cs->vp;
882 	dotdot = (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0');
883 
884 	/*
885 	 * If dotdotting, then need to check whether it's above the
886 	 * root of a filesystem, or above an export point.
887 	 */
888 	if (dotdot) {
889 
890 		/*
891 		 * If dotdotting at the root of a filesystem, then
892 		 * need to traverse back to the mounted-on filesystem
893 		 * and do the dotdot lookup there.
894 		 */
895 		if (cs->vp->v_flag & VROOT) {
896 
897 			/*
898 			 * If at the system root, then can
899 			 * go up no further.
900 			 */
901 			if (VN_CMP(dvp, rootdir))
902 				return (puterrno4(ENOENT));
903 
904 			/*
905 			 * Traverse back to the mounted-on filesystem
906 			 */
907 			dvp = untraverse(cs->vp);
908 
909 			/*
910 			 * Set the different_export flag so we remember
911 			 * to pick up a new exportinfo entry for
912 			 * this new filesystem.
913 			 */
914 			different_export = 1;
915 		} else {
916 
917 			/*
918 			 * If dotdotting above an export point then set
919 			 * the different_export to get new export info.
920 			 */
921 			different_export = nfs_exported(cs->exi, cs->vp);
922 		}
923 	}
924 
925 	/*
926 	 * Get the vnode for the component "nm".
927 	 */
928 	error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cs->cr,
929 	    NULL, NULL, NULL);
930 	if (error)
931 		return (puterrno4(error));
932 
933 	/*
934 	 * If the vnode is in a pseudo filesystem, or if the security flavor
935 	 * used in the request is valid but not an explicitly shared flavor,
936 	 * or the access bit indicates that this is a limited access,
937 	 * check whether this vnode is visible.
938 	 */
939 	if (!different_export &&
940 	    (PSEUDO(cs->exi) || ! is_exported_sec(cs->nfsflavor, cs->exi) ||
941 	    cs->access & CS_ACCESS_LIMITED)) {
942 		if (! nfs_visible(cs->exi, vp, &different_export)) {
943 			VN_RELE(vp);
944 			return (puterrno4(ENOENT));
945 		}
946 	}
947 
948 	/*
949 	 * If it's a mountpoint, then traverse it.
950 	 */
951 	if (vn_ismntpt(vp)) {
952 		tvp = vp;
953 		if ((error = traverse(&tvp)) != 0) {
954 			VN_RELE(vp);
955 			return (puterrno4(error));
956 		}
957 		/* remember that we had to traverse mountpoint */
958 		did_traverse = TRUE;
959 		vp = tvp;
960 		different_export = 1;
961 	} else if (vp->v_vfsp != dvp->v_vfsp) {
962 		/*
963 		 * If vp isn't a mountpoint and the vfs ptrs aren't the same,
964 		 * then vp is probably an LOFS object.  We don't need the
965 		 * realvp, we just need to know that we might have crossed
966 		 * a server fs boundary and need to call checkexport4.
967 		 * (LOFS lookup hides server fs mountpoints, and actually calls
968 		 * traverse)
969 		 */
970 		different_export = 1;
971 		did_traverse = FALSE;
972 	}
973 
974 	/*
975 	 * Get the export information for it.
976 	 */
977 	if (different_export) {
978 
979 		bzero(&fid, sizeof (fid));
980 		fid.fid_len = MAXFIDSZ;
981 		error = vop_fid_pseudo(vp, &fid);
982 		if (error) {
983 			VN_RELE(vp);
984 			return (puterrno4(error));
985 		}
986 
987 		if (dotdot)
988 			exi = nfs_vptoexi(NULL, vp, cs->cr, &walk, NULL, TRUE);
989 		else
990 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
991 
992 		if (exi == NULL) {
993 			if (did_traverse == TRUE) {
994 				/*
995 				 * If this vnode is a mounted-on vnode,
996 				 * but the mounted-on file system is not
997 				 * exported, send back the secinfo for
998 				 * the exported node that the mounted-on
999 				 * vnode lives in.
1000 				 */
1001 				exi = cs->exi;
1002 			} else {
1003 				VN_RELE(vp);
1004 				return (puterrno4(EACCES));
1005 			}
1006 		}
1007 	} else {
1008 		exi = cs->exi;
1009 	}
1010 	ASSERT(exi != NULL);
1011 
1012 
1013 	/*
1014 	 * Create the secinfo result based on the security information
1015 	 * from the exportinfo structure (exi).
1016 	 *
1017 	 * Return all flavors for a pseudo node.
1018 	 * For a real export node, return the flavor that the client
1019 	 * has access with.
1020 	 */
1021 	ASSERT(RW_LOCK_HELD(&exported_lock));
1022 	if (PSEUDO(exi)) {
1023 		count = exi->exi_export.ex_seccnt; /* total sec count */
1024 		resok_val = kmem_alloc(count * sizeof (secinfo4), KM_SLEEP);
1025 		secp = exi->exi_export.ex_secinfo;
1026 
1027 		for (i = 0; i < count; i++) {
1028 			si = &secp[i].s_secinfo;
1029 			resok_val[i].flavor = si->sc_rpcnum;
1030 			if (resok_val[i].flavor == RPCSEC_GSS) {
1031 				rpcsec_gss_info *info;
1032 
1033 				info = &resok_val[i].flavor_info;
1034 				info->qop = si->sc_qop;
1035 				info->service = (rpc_gss_svc_t)si->sc_service;
1036 
1037 				/* get oid opaque data */
1038 				info->oid.sec_oid4_len =
1039 				    si->sc_gss_mech_type->length;
1040 				info->oid.sec_oid4_val = kmem_alloc(
1041 				    si->sc_gss_mech_type->length, KM_SLEEP);
1042 				bcopy(
1043 				    si->sc_gss_mech_type->elements,
1044 				    info->oid.sec_oid4_val,
1045 				    info->oid.sec_oid4_len);
1046 			}
1047 		}
1048 		resp->SECINFO4resok_len = count;
1049 		resp->SECINFO4resok_val = resok_val;
1050 	} else {
1051 		int ret_cnt = 0, k = 0;
1052 		int *flavor_list;
1053 
1054 		count = exi->exi_export.ex_seccnt; /* total sec count */
1055 		secp = exi->exi_export.ex_secinfo;
1056 
1057 		flavor_list = kmem_alloc(count * sizeof (int), KM_SLEEP);
1058 		/* find out which flavors to return */
1059 		for (i = 0; i < count; i ++) {
1060 			int access, flavor, perm;
1061 
1062 			flavor = secp[i].s_secinfo.sc_nfsnum;
1063 			perm = secp[i].s_flags;
1064 
1065 			access = nfsauth4_secinfo_access(exi, cs->req,
1066 			    flavor, perm);
1067 
1068 			if (! (access & NFSAUTH_DENIED) &&
1069 			    ! (access & NFSAUTH_WRONGSEC)) {
1070 				flavor_list[ret_cnt] = flavor;
1071 				ret_cnt++;
1072 			}
1073 		}
1074 
1075 		/* Create the returning SECINFO value */
1076 		resok_val = kmem_alloc(ret_cnt * sizeof (secinfo4), KM_SLEEP);
1077 
1078 		for (i = 0; i < count; i++) {
1079 			/*
1080 			 * If the flavor is in the flavor list,
1081 			 * fill in resok_val.
1082 			 */
1083 			si = &secp[i].s_secinfo;
1084 			if (in_flavor_list(si->sc_nfsnum,
1085 			    flavor_list, ret_cnt)) {
1086 				resok_val[k].flavor = si->sc_rpcnum;
1087 				if (resok_val[k].flavor == RPCSEC_GSS) {
1088 					rpcsec_gss_info *info;
1089 
1090 					info = &resok_val[k].flavor_info;
1091 					info->qop = si->sc_qop;
1092 					info->service = (rpc_gss_svc_t)
1093 					    si->sc_service;
1094 
1095 					/* get oid opaque data */
1096 					info->oid.sec_oid4_len =
1097 					    si->sc_gss_mech_type->length;
1098 					info->oid.sec_oid4_val = kmem_alloc(
1099 					    si->sc_gss_mech_type->length,
1100 					    KM_SLEEP);
1101 					bcopy(si->sc_gss_mech_type->elements,
1102 					    info->oid.sec_oid4_val,
1103 					    info->oid.sec_oid4_len);
1104 				}
1105 				k++;
1106 			}
1107 			if (k >= ret_cnt)
1108 				break;
1109 		}
1110 		resp->SECINFO4resok_len = ret_cnt;
1111 		resp->SECINFO4resok_val = resok_val;
1112 		kmem_free(flavor_list, count * sizeof (int));
1113 	}
1114 
1115 	VN_RELE(vp);
1116 	return (NFS4_OK);
1117 }
1118 
1119 /*
1120  * SECINFO (Operation 33): Obtain required security information on
1121  * the component name in the format of (security-mechanism-oid, qop, service)
1122  * triplets.
1123  */
1124 /* ARGSUSED */
1125 static void
1126 rfs4_op_secinfo(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1127     struct compound_state *cs)
1128 {
1129 	SECINFO4args *args = &argop->nfs_argop4_u.opsecinfo;
1130 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1131 	utf8string *utfnm = &args->name;
1132 	uint_t len;
1133 	char *nm;
1134 	struct sockaddr *ca;
1135 	char *name = NULL;
1136 
1137 	DTRACE_NFSV4_2(op__secinfo__start, struct compound_state *, cs,
1138 	    SECINFO4args *, args);
1139 
1140 	/*
1141 	 * Current file handle (cfh) should have been set before getting
1142 	 * into this function. If not, return error.
1143 	 */
1144 	if (cs->vp == NULL) {
1145 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1146 		goto out;
1147 	}
1148 
1149 	if (cs->vp->v_type != VDIR) {
1150 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1151 		goto out;
1152 	}
1153 
1154 	/*
1155 	 * Verify the component name. If failed, error out, but
1156 	 * do not error out if the component name is a "..".
1157 	 * SECINFO will return its parents secinfo data for SECINFO "..".
1158 	 */
1159 	if (!utf8_dir_verify(utfnm)) {
1160 		if (utfnm->utf8string_len != 2 ||
1161 		    utfnm->utf8string_val[0] != '.' ||
1162 		    utfnm->utf8string_val[1] != '.') {
1163 			*cs->statusp = resp->status = NFS4ERR_INVAL;
1164 			goto out;
1165 		}
1166 	}
1167 
1168 	nm = utf8_to_str(utfnm, &len, NULL);
1169 	if (nm == NULL) {
1170 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1171 		goto out;
1172 	}
1173 
1174 	if (len > MAXNAMELEN) {
1175 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1176 		kmem_free(nm, len);
1177 		goto out;
1178 	}
1179 
1180 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1181 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1182 	    MAXPATHLEN  + 1);
1183 
1184 	if (name == NULL) {
1185 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1186 		kmem_free(nm, len);
1187 		goto out;
1188 	}
1189 
1190 
1191 	*cs->statusp = resp->status = do_rfs4_op_secinfo(cs, name, resp);
1192 
1193 	if (name != nm)
1194 		kmem_free(name, MAXPATHLEN + 1);
1195 	kmem_free(nm, len);
1196 
1197 out:
1198 	DTRACE_NFSV4_2(op__secinfo__done, struct compound_state *, cs,
1199 	    SECINFO4res *, resp);
1200 }
1201 
1202 /*
1203  * Free SECINFO result.
1204  */
1205 /* ARGSUSED */
1206 static void
1207 rfs4_op_secinfo_free(nfs_resop4 *resop)
1208 {
1209 	SECINFO4res *resp = &resop->nfs_resop4_u.opsecinfo;
1210 	int count, i;
1211 	secinfo4 *resok_val;
1212 
1213 	/* If this is not an Ok result, nothing to free. */
1214 	if (resp->status != NFS4_OK) {
1215 		return;
1216 	}
1217 
1218 	count = resp->SECINFO4resok_len;
1219 	resok_val = resp->SECINFO4resok_val;
1220 
1221 	for (i = 0; i < count; i++) {
1222 		if (resok_val[i].flavor == RPCSEC_GSS) {
1223 			rpcsec_gss_info *info;
1224 
1225 			info = &resok_val[i].flavor_info;
1226 			kmem_free(info->oid.sec_oid4_val,
1227 			    info->oid.sec_oid4_len);
1228 		}
1229 	}
1230 	kmem_free(resok_val, count * sizeof (secinfo4));
1231 	resp->SECINFO4resok_len = 0;
1232 	resp->SECINFO4resok_val = NULL;
1233 }
1234 
1235 /* ARGSUSED */
1236 static void
1237 rfs4_op_access(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1238     struct compound_state *cs)
1239 {
1240 	ACCESS4args *args = &argop->nfs_argop4_u.opaccess;
1241 	ACCESS4res *resp = &resop->nfs_resop4_u.opaccess;
1242 	int error;
1243 	vnode_t *vp;
1244 	struct vattr va;
1245 	int checkwriteperm;
1246 	cred_t *cr = cs->cr;
1247 	bslabel_t *clabel, *slabel;
1248 	ts_label_t *tslabel;
1249 	boolean_t admin_low_client;
1250 
1251 	DTRACE_NFSV4_2(op__access__start, struct compound_state *, cs,
1252 	    ACCESS4args *, args);
1253 
1254 #if 0	/* XXX allow access even if !cs->access. Eventually only pseudo fs */
1255 	if (cs->access == CS_ACCESS_DENIED) {
1256 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1257 		goto out;
1258 	}
1259 #endif
1260 	if (cs->vp == NULL) {
1261 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1262 		goto out;
1263 	}
1264 
1265 	ASSERT(cr != NULL);
1266 
1267 	vp = cs->vp;
1268 
1269 	/*
1270 	 * If the file system is exported read only, it is not appropriate
1271 	 * to check write permissions for regular files and directories.
1272 	 * Special files are interpreted by the client, so the underlying
1273 	 * permissions are sent back to the client for interpretation.
1274 	 */
1275 	if (rdonly4(cs->exi, cs->vp, req) &&
1276 	    (vp->v_type == VREG || vp->v_type == VDIR))
1277 		checkwriteperm = 0;
1278 	else
1279 		checkwriteperm = 1;
1280 
1281 	/*
1282 	 * XXX
1283 	 * We need the mode so that we can correctly determine access
1284 	 * permissions relative to a mandatory lock file.  Access to
1285 	 * mandatory lock files is denied on the server, so it might
1286 	 * as well be reflected to the server during the open.
1287 	 */
1288 	va.va_mask = AT_MODE;
1289 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1290 	if (error) {
1291 		*cs->statusp = resp->status = puterrno4(error);
1292 		goto out;
1293 	}
1294 	resp->access = 0;
1295 	resp->supported = 0;
1296 
1297 	if (is_system_labeled()) {
1298 		ASSERT(req->rq_label != NULL);
1299 		clabel = req->rq_label;
1300 		DTRACE_PROBE2(tx__rfs4__log__info__opaccess__clabel, char *,
1301 		    "got client label from request(1)",
1302 		    struct svc_req *, req);
1303 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
1304 			if ((tslabel = nfs_getflabel(vp, cs->exi)) == NULL) {
1305 				*cs->statusp = resp->status = puterrno4(EACCES);
1306 				goto out;
1307 			}
1308 			slabel = label2bslabel(tslabel);
1309 			DTRACE_PROBE3(tx__rfs4__log__info__opaccess__slabel,
1310 			    char *, "got server label(1) for vp(2)",
1311 			    bslabel_t *, slabel, vnode_t *, vp);
1312 
1313 			admin_low_client = B_FALSE;
1314 		} else
1315 			admin_low_client = B_TRUE;
1316 	}
1317 
1318 	if (args->access & ACCESS4_READ) {
1319 		error = VOP_ACCESS(vp, VREAD, 0, cr, NULL);
1320 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1321 		    (!is_system_labeled() || admin_low_client ||
1322 		    bldominates(clabel, slabel)))
1323 			resp->access |= ACCESS4_READ;
1324 		resp->supported |= ACCESS4_READ;
1325 	}
1326 	if ((args->access & ACCESS4_LOOKUP) && vp->v_type == VDIR) {
1327 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1328 		if (!error && (!is_system_labeled() || admin_low_client ||
1329 		    bldominates(clabel, slabel)))
1330 			resp->access |= ACCESS4_LOOKUP;
1331 		resp->supported |= ACCESS4_LOOKUP;
1332 	}
1333 	if (checkwriteperm &&
1334 	    (args->access & (ACCESS4_MODIFY|ACCESS4_EXTEND))) {
1335 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1336 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1337 		    (!is_system_labeled() || admin_low_client ||
1338 		    blequal(clabel, slabel)))
1339 			resp->access |=
1340 			    (args->access & (ACCESS4_MODIFY | ACCESS4_EXTEND));
1341 		resp->supported |= (ACCESS4_MODIFY | ACCESS4_EXTEND);
1342 	}
1343 
1344 	if (checkwriteperm &&
1345 	    (args->access & ACCESS4_DELETE) && vp->v_type == VDIR) {
1346 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
1347 		if (!error && (!is_system_labeled() || admin_low_client ||
1348 		    blequal(clabel, slabel)))
1349 			resp->access |= ACCESS4_DELETE;
1350 		resp->supported |= ACCESS4_DELETE;
1351 	}
1352 	if (args->access & ACCESS4_EXECUTE && vp->v_type != VDIR) {
1353 		error = VOP_ACCESS(vp, VEXEC, 0, cr, NULL);
1354 		if (!error && !MANDLOCK(vp, va.va_mode) &&
1355 		    (!is_system_labeled() || admin_low_client ||
1356 		    bldominates(clabel, slabel)))
1357 			resp->access |= ACCESS4_EXECUTE;
1358 		resp->supported |= ACCESS4_EXECUTE;
1359 	}
1360 
1361 	if (is_system_labeled() && !admin_low_client)
1362 		label_rele(tslabel);
1363 
1364 	*cs->statusp = resp->status = NFS4_OK;
1365 out:
1366 	DTRACE_NFSV4_2(op__access__done, struct compound_state *, cs,
1367 	    ACCESS4res *, resp);
1368 }
1369 
1370 /* ARGSUSED */
1371 static void
1372 rfs4_op_commit(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1373     struct compound_state *cs)
1374 {
1375 	COMMIT4args *args = &argop->nfs_argop4_u.opcommit;
1376 	COMMIT4res *resp = &resop->nfs_resop4_u.opcommit;
1377 	int error;
1378 	vnode_t *vp = cs->vp;
1379 	cred_t *cr = cs->cr;
1380 	vattr_t va;
1381 
1382 	DTRACE_NFSV4_2(op__commit__start, struct compound_state *, cs,
1383 	    COMMIT4args *, args);
1384 
1385 	if (vp == NULL) {
1386 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1387 		goto out;
1388 	}
1389 	if (cs->access == CS_ACCESS_DENIED) {
1390 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1391 		goto out;
1392 	}
1393 
1394 	if (args->offset + args->count < args->offset) {
1395 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1396 		goto out;
1397 	}
1398 
1399 	va.va_mask = AT_UID;
1400 	error = VOP_GETATTR(vp, &va, 0, cr, NULL);
1401 
1402 	/*
1403 	 * If we can't get the attributes, then we can't do the
1404 	 * right access checking.  So, we'll fail the request.
1405 	 */
1406 	if (error) {
1407 		*cs->statusp = resp->status = puterrno4(error);
1408 		goto out;
1409 	}
1410 	if (rdonly4(cs->exi, cs->vp, req)) {
1411 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1412 		goto out;
1413 	}
1414 
1415 	if (vp->v_type != VREG) {
1416 		if (vp->v_type == VDIR)
1417 			resp->status = NFS4ERR_ISDIR;
1418 		else
1419 			resp->status = NFS4ERR_INVAL;
1420 		*cs->statusp = resp->status;
1421 		goto out;
1422 	}
1423 
1424 	if (crgetuid(cr) != va.va_uid &&
1425 	    (error = VOP_ACCESS(vp, VWRITE, 0, cs->cr, NULL))) {
1426 		*cs->statusp = resp->status = puterrno4(error);
1427 		goto out;
1428 	}
1429 
1430 	error = VOP_PUTPAGE(vp, args->offset, args->count, 0, cr, NULL);
1431 	if (!error)
1432 		error = VOP_FSYNC(vp, FNODSYNC, cr, NULL);
1433 
1434 	if (error) {
1435 		*cs->statusp = resp->status = puterrno4(error);
1436 		goto out;
1437 	}
1438 
1439 	*cs->statusp = resp->status = NFS4_OK;
1440 	resp->writeverf = Write4verf;
1441 out:
1442 	DTRACE_NFSV4_2(op__commit__done, struct compound_state *, cs,
1443 	    COMMIT4res *, resp);
1444 }
1445 
1446 /*
1447  * rfs4_op_mknod is called from rfs4_op_create after all initial verification
1448  * was completed. It does the nfsv4 create for special files.
1449  */
1450 /* ARGSUSED */
1451 static vnode_t *
1452 do_rfs4_op_mknod(CREATE4args *args, CREATE4res *resp, struct svc_req *req,
1453     struct compound_state *cs, vattr_t *vap, char *nm)
1454 {
1455 	int error;
1456 	cred_t *cr = cs->cr;
1457 	vnode_t *dvp = cs->vp;
1458 	vnode_t *vp = NULL;
1459 	int mode;
1460 	enum vcexcl excl;
1461 
1462 	switch (args->type) {
1463 	case NF4CHR:
1464 	case NF4BLK:
1465 		if (secpolicy_sys_devices(cr) != 0) {
1466 			*cs->statusp = resp->status = NFS4ERR_PERM;
1467 			return (NULL);
1468 		}
1469 		if (args->type == NF4CHR)
1470 			vap->va_type = VCHR;
1471 		else
1472 			vap->va_type = VBLK;
1473 		vap->va_rdev = makedevice(args->ftype4_u.devdata.specdata1,
1474 		    args->ftype4_u.devdata.specdata2);
1475 		vap->va_mask |= AT_RDEV;
1476 		break;
1477 	case NF4SOCK:
1478 		vap->va_type = VSOCK;
1479 		break;
1480 	case NF4FIFO:
1481 		vap->va_type = VFIFO;
1482 		break;
1483 	default:
1484 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1485 		return (NULL);
1486 	}
1487 
1488 	/*
1489 	 * Must specify the mode.
1490 	 */
1491 	if (!(vap->va_mask & AT_MODE)) {
1492 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1493 		return (NULL);
1494 	}
1495 
1496 	excl = EXCL;
1497 
1498 	mode = 0;
1499 
1500 	error = VOP_CREATE(dvp, nm, vap, excl, mode, &vp, cr, 0, NULL, NULL);
1501 	if (error) {
1502 		*cs->statusp = resp->status = puterrno4(error);
1503 		return (NULL);
1504 	}
1505 	return (vp);
1506 }
1507 
1508 /*
1509  * nfsv4 create is used to create non-regular files. For regular files,
1510  * use nfsv4 open.
1511  */
1512 /* ARGSUSED */
1513 static void
1514 rfs4_op_create(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
1515     struct compound_state *cs)
1516 {
1517 	CREATE4args *args = &argop->nfs_argop4_u.opcreate;
1518 	CREATE4res *resp = &resop->nfs_resop4_u.opcreate;
1519 	int error;
1520 	struct vattr bva, iva, iva2, ava, *vap;
1521 	cred_t *cr = cs->cr;
1522 	vnode_t *dvp = cs->vp;
1523 	vnode_t *vp = NULL;
1524 	vnode_t *realvp;
1525 	char *nm, *lnm;
1526 	uint_t len, llen;
1527 	int syncval = 0;
1528 	struct nfs4_svgetit_arg sarg;
1529 	struct nfs4_ntov_table ntov;
1530 	struct statvfs64 sb;
1531 	nfsstat4 status;
1532 	struct sockaddr *ca;
1533 	char *name = NULL;
1534 	char *lname = NULL;
1535 
1536 	DTRACE_NFSV4_2(op__create__start, struct compound_state *, cs,
1537 	    CREATE4args *, args);
1538 
1539 	resp->attrset = 0;
1540 
1541 	if (dvp == NULL) {
1542 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
1543 		goto out;
1544 	}
1545 
1546 	/*
1547 	 * If there is an unshared filesystem mounted on this vnode,
1548 	 * do not allow to create an object in this directory.
1549 	 */
1550 	if (vn_ismntpt(dvp)) {
1551 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1552 		goto out;
1553 	}
1554 
1555 	/* Verify that type is correct */
1556 	switch (args->type) {
1557 	case NF4LNK:
1558 	case NF4BLK:
1559 	case NF4CHR:
1560 	case NF4SOCK:
1561 	case NF4FIFO:
1562 	case NF4DIR:
1563 		break;
1564 	default:
1565 		*cs->statusp = resp->status = NFS4ERR_BADTYPE;
1566 		goto out;
1567 	};
1568 
1569 	if (cs->access == CS_ACCESS_DENIED) {
1570 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
1571 		goto out;
1572 	}
1573 	if (dvp->v_type != VDIR) {
1574 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
1575 		goto out;
1576 	}
1577 	if (!utf8_dir_verify(&args->objname)) {
1578 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1579 		goto out;
1580 	}
1581 
1582 	if (rdonly4(cs->exi, cs->vp, req)) {
1583 		*cs->statusp = resp->status = NFS4ERR_ROFS;
1584 		goto out;
1585 	}
1586 
1587 	/*
1588 	 * Name of newly created object
1589 	 */
1590 	nm = utf8_to_fn(&args->objname, &len, NULL);
1591 	if (nm == NULL) {
1592 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1593 		goto out;
1594 	}
1595 
1596 	if (len > MAXNAMELEN) {
1597 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1598 		kmem_free(nm, len);
1599 		goto out;
1600 	}
1601 
1602 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
1603 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
1604 	    MAXPATHLEN  + 1);
1605 
1606 	if (name == NULL) {
1607 		*cs->statusp = resp->status = NFS4ERR_INVAL;
1608 		kmem_free(nm, len);
1609 		goto out;
1610 	}
1611 
1612 	resp->attrset = 0;
1613 
1614 	sarg.sbp = &sb;
1615 	sarg.is_referral = B_FALSE;
1616 	nfs4_ntov_table_init(&ntov);
1617 
1618 	status = do_rfs4_set_attrs(&resp->attrset,
1619 	    &args->createattrs, cs, &sarg, &ntov, NFS4ATTR_SETIT);
1620 
1621 	if (sarg.vap->va_mask == 0 && status == NFS4_OK)
1622 		status = NFS4ERR_INVAL;
1623 
1624 	if (status != NFS4_OK) {
1625 		*cs->statusp = resp->status = status;
1626 		kmem_free(nm, len);
1627 		nfs4_ntov_table_free(&ntov, &sarg);
1628 		resp->attrset = 0;
1629 		goto out;
1630 	}
1631 
1632 	/* Get "before" change value */
1633 	bva.va_mask = AT_CTIME|AT_SEQ;
1634 	error = VOP_GETATTR(dvp, &bva, 0, cr, NULL);
1635 	if (error) {
1636 		*cs->statusp = resp->status = puterrno4(error);
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 default initial values for attributes when not specified
1648 	 * in createattrs.
1649 	 */
1650 	if ((vap->va_mask & AT_UID) == 0) {
1651 		vap->va_uid = crgetuid(cr);
1652 		vap->va_mask |= AT_UID;
1653 	}
1654 	if ((vap->va_mask & AT_GID) == 0) {
1655 		vap->va_gid = crgetgid(cr);
1656 		vap->va_mask |= AT_GID;
1657 	}
1658 
1659 	vap->va_mask |= AT_TYPE;
1660 	switch (args->type) {
1661 	case NF4DIR:
1662 		vap->va_type = VDIR;
1663 		if ((vap->va_mask & AT_MODE) == 0) {
1664 			vap->va_mode = 0700;	/* default: owner rwx only */
1665 			vap->va_mask |= AT_MODE;
1666 		}
1667 		error = VOP_MKDIR(dvp, nm, vap, &vp, cr, NULL, 0, NULL);
1668 		if (error)
1669 			break;
1670 
1671 		/*
1672 		 * Get the initial "after" sequence number, if it fails,
1673 		 * set to zero
1674 		 */
1675 		iva.va_mask = AT_SEQ;
1676 		if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL))
1677 			iva.va_seq = 0;
1678 		break;
1679 	case NF4LNK:
1680 		vap->va_type = VLNK;
1681 		if ((vap->va_mask & AT_MODE) == 0) {
1682 			vap->va_mode = 0700;	/* default: owner rwx only */
1683 			vap->va_mask |= AT_MODE;
1684 		}
1685 
1686 		/*
1687 		 * symlink names must be treated as data
1688 		 */
1689 		lnm = utf8_to_str(&args->ftype4_u.linkdata, &llen, NULL);
1690 
1691 		if (lnm == NULL) {
1692 			*cs->statusp = resp->status = NFS4ERR_INVAL;
1693 			if (name != nm)
1694 				kmem_free(name, MAXPATHLEN + 1);
1695 			kmem_free(nm, len);
1696 			nfs4_ntov_table_free(&ntov, &sarg);
1697 			resp->attrset = 0;
1698 			goto out;
1699 		}
1700 
1701 		if (llen > MAXPATHLEN) {
1702 			*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
1703 			if (name != nm)
1704 				kmem_free(name, MAXPATHLEN + 1);
1705 			kmem_free(nm, len);
1706 			kmem_free(lnm, llen);
1707 			nfs4_ntov_table_free(&ntov, &sarg);
1708 			resp->attrset = 0;
1709 			goto out;
1710 		}
1711 
1712 		lname = nfscmd_convname(ca, cs->exi, lnm,
1713 		    NFSCMD_CONV_INBOUND, MAXPATHLEN  + 1);
1714 
1715 		if (lname == NULL) {
1716 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
1717 			if (name != nm)
1718 				kmem_free(name, MAXPATHLEN + 1);
1719 			kmem_free(nm, len);
1720 			kmem_free(lnm, llen);
1721 			nfs4_ntov_table_free(&ntov, &sarg);
1722 			resp->attrset = 0;
1723 			goto out;
1724 		}
1725 
1726 		error = VOP_SYMLINK(dvp, nm, vap, lnm, cr, NULL, 0);
1727 		if (lname != lnm)
1728 			kmem_free(lname, MAXPATHLEN + 1);
1729 		if (lnm != NULL)
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, nm, &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, nm);
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 		if (cs->vp)
2778 			oldvp = cs->vp;
2779 		cs->vp = vp;
2780 		is_newvp = TRUE;
2781 
2782 		stat = call_checkauth4(cs, req);
2783 		if (stat != NFS4_OK) {
2784 			VN_RELE(cs->vp);
2785 			cs->vp = oldvp;
2786 			return (stat);
2787 		}
2788 	}
2789 
2790 	/*
2791 	 * After various NFS checks, do a label check on the path
2792 	 * component. The label on this path should either be the
2793 	 * global zone's label or a zone's label. We are only
2794 	 * interested in the zone's label because exported files
2795 	 * in global zone is accessible (though read-only) to
2796 	 * clients. The exportability/visibility check is already
2797 	 * done before reaching this code.
2798 	 */
2799 	if (is_system_labeled()) {
2800 		bslabel_t *clabel;
2801 
2802 		ASSERT(req->rq_label != NULL);
2803 		clabel = req->rq_label;
2804 		DTRACE_PROBE2(tx__rfs4__log__info__oplookup__clabel, char *,
2805 		    "got client label from request(1)", struct svc_req *, req);
2806 
2807 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
2808 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
2809 			    cs->exi)) {
2810 				error = EACCES;
2811 				goto err_out;
2812 			}
2813 		} else {
2814 			/*
2815 			 * We grant access to admin_low label clients
2816 			 * only if the client is trusted, i.e. also
2817 			 * running Solaris Trusted Extension.
2818 			 */
2819 			struct sockaddr	*ca;
2820 			int		addr_type;
2821 			void		*ipaddr;
2822 			tsol_tpc_t	*tp;
2823 
2824 			ca = (struct sockaddr *)svc_getrpccaller(
2825 			    req->rq_xprt)->buf;
2826 			if (ca->sa_family == AF_INET) {
2827 				addr_type = IPV4_VERSION;
2828 				ipaddr = &((struct sockaddr_in *)ca)->sin_addr;
2829 			} else if (ca->sa_family == AF_INET6) {
2830 				addr_type = IPV6_VERSION;
2831 				ipaddr = &((struct sockaddr_in6 *)
2832 				    ca)->sin6_addr;
2833 			}
2834 			tp = find_tpc(ipaddr, addr_type, B_FALSE);
2835 			if (tp == NULL || tp->tpc_tp.tp_doi !=
2836 			    l_admin_low->tsl_doi || tp->tpc_tp.host_type !=
2837 			    SUN_CIPSO) {
2838 				if (tp != NULL)
2839 					TPC_RELE(tp);
2840 				error = EACCES;
2841 				goto err_out;
2842 			}
2843 			TPC_RELE(tp);
2844 		}
2845 	}
2846 
2847 	error = makefh4(&cs->fh, vp, cs->exi);
2848 
2849 err_out:
2850 	if (error) {
2851 		if (is_newvp) {
2852 			VN_RELE(cs->vp);
2853 			cs->vp = oldvp;
2854 		} else
2855 			VN_RELE(vp);
2856 		return (puterrno4(error));
2857 	}
2858 
2859 	if (!is_newvp) {
2860 		if (cs->vp)
2861 			VN_RELE(cs->vp);
2862 		cs->vp = vp;
2863 	} else if (oldvp)
2864 		VN_RELE(oldvp);
2865 
2866 	/*
2867 	 * if did lookup on attrdir and didn't lookup .., set named
2868 	 * attr fh flag
2869 	 */
2870 	if (attrdir && ! dotdot)
2871 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
2872 
2873 	/* Assume false for now, open proc will set this */
2874 	cs->mandlock = FALSE;
2875 
2876 	return (NFS4_OK);
2877 }
2878 
2879 /* ARGSUSED */
2880 static void
2881 rfs4_op_lookup(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2882     struct compound_state *cs)
2883 {
2884 	LOOKUP4args *args = &argop->nfs_argop4_u.oplookup;
2885 	LOOKUP4res *resp = &resop->nfs_resop4_u.oplookup;
2886 	char *nm;
2887 	uint_t len;
2888 	struct sockaddr *ca;
2889 	char *name = NULL;
2890 
2891 	DTRACE_NFSV4_2(op__lookup__start, struct compound_state *, cs,
2892 	    LOOKUP4args *, args);
2893 
2894 	if (cs->vp == NULL) {
2895 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2896 		goto out;
2897 	}
2898 
2899 	if (cs->vp->v_type == VLNK) {
2900 		*cs->statusp = resp->status = NFS4ERR_SYMLINK;
2901 		goto out;
2902 	}
2903 
2904 	if (cs->vp->v_type != VDIR) {
2905 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2906 		goto out;
2907 	}
2908 
2909 	if (!utf8_dir_verify(&args->objname)) {
2910 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2911 		goto out;
2912 	}
2913 
2914 	nm = utf8_to_str(&args->objname, &len, NULL);
2915 	if (nm == NULL) {
2916 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2917 		goto out;
2918 	}
2919 
2920 	if (len > MAXNAMELEN) {
2921 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
2922 		kmem_free(nm, len);
2923 		goto out;
2924 	}
2925 
2926 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
2927 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
2928 	    MAXPATHLEN  + 1);
2929 
2930 	if (name == NULL) {
2931 		*cs->statusp = resp->status = NFS4ERR_INVAL;
2932 		kmem_free(nm, len);
2933 		goto out;
2934 	}
2935 
2936 	*cs->statusp = resp->status = do_rfs4_op_lookup(name, req, cs);
2937 
2938 	if (name != nm)
2939 		kmem_free(name, MAXPATHLEN + 1);
2940 	kmem_free(nm, len);
2941 
2942 out:
2943 	DTRACE_NFSV4_2(op__lookup__done, struct compound_state *, cs,
2944 	    LOOKUP4res *, resp);
2945 }
2946 
2947 /* ARGSUSED */
2948 static void
2949 rfs4_op_lookupp(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
2950     struct compound_state *cs)
2951 {
2952 	LOOKUPP4res *resp = &resop->nfs_resop4_u.oplookupp;
2953 
2954 	DTRACE_NFSV4_1(op__lookupp__start, struct compound_state *, cs);
2955 
2956 	if (cs->vp == NULL) {
2957 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2958 		goto out;
2959 	}
2960 
2961 	if (cs->vp->v_type != VDIR) {
2962 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
2963 		goto out;
2964 	}
2965 
2966 	*cs->statusp = resp->status = do_rfs4_op_lookup("..", req, cs);
2967 
2968 	/*
2969 	 * From NFSV4 Specification, LOOKUPP should not check for
2970 	 * NFS4ERR_WRONGSEC. Retrun NFS4_OK instead.
2971 	 */
2972 	if (resp->status == NFS4ERR_WRONGSEC) {
2973 		*cs->statusp = resp->status = NFS4_OK;
2974 	}
2975 
2976 out:
2977 	DTRACE_NFSV4_2(op__lookupp__done, struct compound_state *, cs,
2978 	    LOOKUPP4res *, resp);
2979 }
2980 
2981 
2982 /*ARGSUSED2*/
2983 static void
2984 rfs4_op_openattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
2985     struct compound_state *cs)
2986 {
2987 	OPENATTR4args	*args = &argop->nfs_argop4_u.opopenattr;
2988 	OPENATTR4res	*resp = &resop->nfs_resop4_u.opopenattr;
2989 	vnode_t		*avp = NULL;
2990 	int		lookup_flags = LOOKUP_XATTR, error;
2991 	int		exp_ro = 0;
2992 
2993 	DTRACE_NFSV4_2(op__openattr__start, struct compound_state *, cs,
2994 	    OPENATTR4args *, args);
2995 
2996 	if (cs->vp == NULL) {
2997 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
2998 		goto out;
2999 	}
3000 
3001 	if ((cs->vp->v_vfsp->vfs_flag & VFS_XATTR) == 0 &&
3002 	    !vfs_has_feature(cs->vp->v_vfsp, VFSFT_SYSATTR_VIEWS)) {
3003 		*cs->statusp = resp->status = puterrno4(ENOTSUP);
3004 		goto out;
3005 	}
3006 
3007 	/*
3008 	 * If file system supports passing ACE mask to VOP_ACCESS then
3009 	 * check for ACE_READ_NAMED_ATTRS, otherwise do legacy checks
3010 	 */
3011 
3012 	if (vfs_has_feature(cs->vp->v_vfsp, VFSFT_ACEMASKONACCESS))
3013 		error = VOP_ACCESS(cs->vp, ACE_READ_NAMED_ATTRS,
3014 		    V_ACE_MASK, cs->cr, NULL);
3015 	else
3016 		error = ((VOP_ACCESS(cs->vp, VREAD, 0, cs->cr, NULL) != 0) &&
3017 		    (VOP_ACCESS(cs->vp, VWRITE, 0, cs->cr, NULL) != 0) &&
3018 		    (VOP_ACCESS(cs->vp, VEXEC, 0, cs->cr, NULL) != 0));
3019 
3020 	if (error) {
3021 		*cs->statusp = resp->status = puterrno4(EACCES);
3022 		goto out;
3023 	}
3024 
3025 	/*
3026 	 * The CREATE_XATTR_DIR VOP flag cannot be specified if
3027 	 * the file system is exported read-only -- regardless of
3028 	 * createdir flag.  Otherwise the attrdir would be created
3029 	 * (assuming server fs isn't mounted readonly locally).  If
3030 	 * VOP_LOOKUP returns ENOENT in this case, the error will
3031 	 * be translated into EROFS.  ENOSYS is mapped to ENOTSUP
3032 	 * because specfs has no VOP_LOOKUP op, so the macro would
3033 	 * return ENOSYS.  EINVAL is returned by all (current)
3034 	 * Solaris file system implementations when any of their
3035 	 * restrictions are violated (xattr(dir) can't have xattrdir).
3036 	 * Returning NOTSUPP is more appropriate in this case
3037 	 * because the object will never be able to have an attrdir.
3038 	 */
3039 	if (args->createdir && ! (exp_ro = rdonly4(cs->exi, cs->vp, req)))
3040 		lookup_flags |= CREATE_XATTR_DIR;
3041 
3042 	error = VOP_LOOKUP(cs->vp, "", &avp, NULL, lookup_flags, NULL, cs->cr,
3043 	    NULL, NULL, NULL);
3044 
3045 	if (error) {
3046 		if (error == ENOENT && args->createdir && exp_ro)
3047 			*cs->statusp = resp->status = puterrno4(EROFS);
3048 		else if (error == EINVAL || error == ENOSYS)
3049 			*cs->statusp = resp->status = puterrno4(ENOTSUP);
3050 		else
3051 			*cs->statusp = resp->status = puterrno4(error);
3052 		goto out;
3053 	}
3054 
3055 	ASSERT(avp->v_flag & V_XATTRDIR);
3056 
3057 	error = makefh4(&cs->fh, avp, cs->exi);
3058 
3059 	if (error) {
3060 		VN_RELE(avp);
3061 		*cs->statusp = resp->status = puterrno4(error);
3062 		goto out;
3063 	}
3064 
3065 	VN_RELE(cs->vp);
3066 	cs->vp = avp;
3067 
3068 	/*
3069 	 * There is no requirement for an attrdir fh flag
3070 	 * because the attrdir has a vnode flag to distinguish
3071 	 * it from regular (non-xattr) directories.  The
3072 	 * FH4_ATTRDIR flag is set for future sanity checks.
3073 	 */
3074 	set_fh4_flag(&cs->fh, FH4_ATTRDIR);
3075 	*cs->statusp = resp->status = NFS4_OK;
3076 
3077 out:
3078 	DTRACE_NFSV4_2(op__openattr__done, struct compound_state *, cs,
3079 	    OPENATTR4res *, resp);
3080 }
3081 
3082 static int
3083 do_io(int direction, vnode_t *vp, struct uio *uio, int ioflag, cred_t *cred,
3084     caller_context_t *ct)
3085 {
3086 	int error;
3087 	int i;
3088 	clock_t delaytime;
3089 
3090 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
3091 
3092 	/*
3093 	 * Don't block on mandatory locks. If this routine returns
3094 	 * EAGAIN, the caller should return NFS4ERR_LOCKED.
3095 	 */
3096 	uio->uio_fmode = FNONBLOCK;
3097 
3098 	for (i = 0; i < rfs4_maxlock_tries; i++) {
3099 
3100 
3101 		if (direction == FREAD) {
3102 			(void) VOP_RWLOCK(vp, V_WRITELOCK_FALSE, ct);
3103 			error = VOP_READ(vp, uio, ioflag, cred, ct);
3104 			VOP_RWUNLOCK(vp, V_WRITELOCK_FALSE, ct);
3105 		} else {
3106 			(void) VOP_RWLOCK(vp, V_WRITELOCK_TRUE, ct);
3107 			error = VOP_WRITE(vp, uio, ioflag, cred, ct);
3108 			VOP_RWUNLOCK(vp, V_WRITELOCK_TRUE, ct);
3109 		}
3110 
3111 		if (error != EAGAIN)
3112 			break;
3113 
3114 		if (i < rfs4_maxlock_tries - 1) {
3115 			delay(delaytime);
3116 			delaytime *= 2;
3117 		}
3118 	}
3119 
3120 	return (error);
3121 }
3122 
3123 /* ARGSUSED */
3124 static void
3125 rfs4_op_read(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3126     struct compound_state *cs)
3127 {
3128 	READ4args *args = &argop->nfs_argop4_u.opread;
3129 	READ4res *resp = &resop->nfs_resop4_u.opread;
3130 	int error;
3131 	int verror;
3132 	vnode_t *vp;
3133 	struct vattr va;
3134 	struct iovec iov;
3135 	struct uio uio;
3136 	u_offset_t offset;
3137 	bool_t *deleg = &cs->deleg;
3138 	nfsstat4 stat;
3139 	int in_crit = 0;
3140 	mblk_t *mp = NULL;
3141 	int alloc_err = 0;
3142 	int rdma_used = 0;
3143 	int loaned_buffers;
3144 	caller_context_t ct;
3145 	struct uio *uiop;
3146 
3147 	DTRACE_NFSV4_2(op__read__start, struct compound_state *, cs,
3148 	    READ4args, args);
3149 
3150 	vp = cs->vp;
3151 	if (vp == NULL) {
3152 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3153 		goto out;
3154 	}
3155 	if (cs->access == CS_ACCESS_DENIED) {
3156 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3157 		goto out;
3158 	}
3159 
3160 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3161 	    deleg, TRUE, &ct)) != NFS4_OK) {
3162 		*cs->statusp = resp->status = stat;
3163 		goto out;
3164 	}
3165 
3166 	/*
3167 	 * Enter the critical region before calling VOP_RWLOCK
3168 	 * to avoid a deadlock with write requests.
3169 	 */
3170 	if (nbl_need_check(vp)) {
3171 		nbl_start_crit(vp, RW_READER);
3172 		in_crit = 1;
3173 		if (nbl_conflict(vp, NBL_READ, args->offset, args->count, 0,
3174 		    &ct)) {
3175 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
3176 			goto out;
3177 		}
3178 	}
3179 
3180 	if ((stat = rfs4_check_stateid(FREAD, vp, &args->stateid, FALSE,
3181 	    deleg, TRUE, &ct)) != NFS4_OK) {
3182 		*cs->statusp = resp->status = stat;
3183 		goto out;
3184 	}
3185 
3186 	if (args->wlist)
3187 		rdma_used = 1;
3188 
3189 	/* use loaned buffers for TCP */
3190 	loaned_buffers = (nfs_loaned_buffers && !rdma_used) ? 1 : 0;
3191 
3192 	va.va_mask = AT_MODE|AT_SIZE|AT_UID;
3193 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3194 
3195 	/*
3196 	 * If we can't get the attributes, then we can't do the
3197 	 * right access checking.  So, we'll fail the request.
3198 	 */
3199 	if (verror) {
3200 		*cs->statusp = resp->status = puterrno4(verror);
3201 		goto out;
3202 	}
3203 
3204 	if (vp->v_type != VREG) {
3205 		*cs->statusp = resp->status =
3206 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
3207 		goto out;
3208 	}
3209 
3210 	if (crgetuid(cs->cr) != va.va_uid &&
3211 	    (error = VOP_ACCESS(vp, VREAD, 0, cs->cr, &ct)) &&
3212 	    (error = VOP_ACCESS(vp, VEXEC, 0, cs->cr, &ct))) {
3213 		*cs->statusp = resp->status = puterrno4(error);
3214 		goto out;
3215 	}
3216 
3217 	if (MANDLOCK(vp, va.va_mode)) { /* XXX - V4 supports mand locking */
3218 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3219 		goto out;
3220 	}
3221 
3222 	offset = args->offset;
3223 	if (offset >= va.va_size) {
3224 		*cs->statusp = resp->status = NFS4_OK;
3225 		resp->eof = TRUE;
3226 		resp->data_len = 0;
3227 		resp->data_val = NULL;
3228 		resp->mblk = NULL;
3229 		/* RDMA */
3230 		resp->wlist = args->wlist;
3231 		resp->wlist_len = resp->data_len;
3232 		*cs->statusp = resp->status = NFS4_OK;
3233 		if (resp->wlist)
3234 			clist_zero_len(resp->wlist);
3235 		goto out;
3236 	}
3237 
3238 	if (args->count == 0) {
3239 		*cs->statusp = resp->status = NFS4_OK;
3240 		resp->eof = FALSE;
3241 		resp->data_len = 0;
3242 		resp->data_val = NULL;
3243 		resp->mblk = NULL;
3244 		/* RDMA */
3245 		resp->wlist = args->wlist;
3246 		resp->wlist_len = resp->data_len;
3247 		if (resp->wlist)
3248 			clist_zero_len(resp->wlist);
3249 		goto out;
3250 	}
3251 
3252 	/*
3253 	 * Do not allocate memory more than maximum allowed
3254 	 * transfer size
3255 	 */
3256 	if (args->count > rfs4_tsize(req))
3257 		args->count = rfs4_tsize(req);
3258 
3259 	if (loaned_buffers) {
3260 		uiop = (uio_t *)rfs_setup_xuio(vp);
3261 		ASSERT(uiop != NULL);
3262 		uiop->uio_segflg = UIO_SYSSPACE;
3263 		uiop->uio_loffset = args->offset;
3264 		uiop->uio_resid = args->count;
3265 
3266 		/* Jump to do the read if successful */
3267 		if (!VOP_REQZCBUF(vp, UIO_READ, (xuio_t *)uiop, cs->cr, &ct)) {
3268 			/*
3269 			 * Need to hold the vnode until after VOP_RETZCBUF()
3270 			 * is called.
3271 			 */
3272 			VN_HOLD(vp);
3273 			goto doio_read;
3274 		}
3275 
3276 		DTRACE_PROBE2(nfss__i__reqzcbuf_failed, int,
3277 		    uiop->uio_loffset, int, uiop->uio_resid);
3278 
3279 		uiop->uio_extflg = 0;
3280 
3281 		/* failure to setup for zero copy */
3282 		rfs_free_xuio((void *)uiop);
3283 		loaned_buffers = 0;
3284 	}
3285 
3286 	/*
3287 	 * If returning data via RDMA Write, then grab the chunk list. If we
3288 	 * aren't returning READ data w/RDMA_WRITE, then grab a mblk.
3289 	 */
3290 	if (rdma_used) {
3291 		mp = NULL;
3292 		(void) rdma_get_wchunk(req, &iov, args->wlist);
3293 	} else {
3294 		/*
3295 		 * mp will contain the data to be sent out in the read reply.
3296 		 * It will be freed after the reply has been sent. Let's
3297 		 * roundup the data to a BYTES_PER_XDR_UNIT multiple, so that
3298 		 * the call to xdrmblk_putmblk() never fails. If the first
3299 		 * alloc of the requested size fails, then decrease the size to
3300 		 * something more reasonable and wait for the allocation to
3301 		 * occur.
3302 		 */
3303 		mp = allocb(RNDUP(args->count), BPRI_MED);
3304 		if (mp == NULL) {
3305 			if (args->count > MAXBSIZE)
3306 				args->count = MAXBSIZE;
3307 			mp = allocb_wait(RNDUP(args->count), BPRI_MED,
3308 			    STR_NOSIG, &alloc_err);
3309 		}
3310 		ASSERT(mp != NULL);
3311 		ASSERT(alloc_err == 0);
3312 
3313 		iov.iov_base = (caddr_t)mp->b_datap->db_base;
3314 		iov.iov_len = args->count;
3315 	}
3316 
3317 	uio.uio_iov = &iov;
3318 	uio.uio_iovcnt = 1;
3319 	uio.uio_segflg = UIO_SYSSPACE;
3320 	uio.uio_extflg = UIO_COPY_CACHED;
3321 	uio.uio_loffset = args->offset;
3322 	uio.uio_resid = args->count;
3323 	uiop = &uio;
3324 
3325 doio_read:
3326 	error = do_io(FREAD, vp, uiop, 0, cs->cr, &ct);
3327 
3328 	va.va_mask = AT_SIZE;
3329 	verror = VOP_GETATTR(vp, &va, 0, cs->cr, &ct);
3330 
3331 	if (error) {
3332 		if (mp)
3333 			freemsg(mp);
3334 		*cs->statusp = resp->status = puterrno4(error);
3335 		goto out;
3336 	}
3337 
3338 	/* make mblk using zc buffers */
3339 	if (loaned_buffers) {
3340 		mp = uio_to_mblk(uiop);
3341 		ASSERT(mp != NULL);
3342 	}
3343 
3344 	*cs->statusp = resp->status = NFS4_OK;
3345 
3346 	ASSERT(uiop->uio_resid >= 0);
3347 	resp->data_len = args->count - uiop->uio_resid;
3348 	if (mp) {
3349 		resp->data_val = (char *)mp->b_datap->db_base;
3350 		rfs_rndup_mblks(mp, resp->data_len, loaned_buffers);
3351 	} else {
3352 		resp->data_val = (caddr_t)iov.iov_base;
3353 	}
3354 
3355 	resp->mblk = mp;
3356 
3357 	if (!verror && offset + resp->data_len == va.va_size)
3358 		resp->eof = TRUE;
3359 	else
3360 		resp->eof = FALSE;
3361 
3362 	if (rdma_used) {
3363 		if (!rdma_setup_read_data4(args, resp)) {
3364 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3365 		}
3366 	} else {
3367 		resp->wlist = NULL;
3368 	}
3369 
3370 out:
3371 	if (in_crit)
3372 		nbl_end_crit(vp);
3373 
3374 	DTRACE_NFSV4_2(op__read__done, struct compound_state *, cs,
3375 	    READ4res *, resp);
3376 }
3377 
3378 static void
3379 rfs4_op_read_free(nfs_resop4 *resop)
3380 {
3381 	READ4res	*resp = &resop->nfs_resop4_u.opread;
3382 
3383 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3384 		freemsg(resp->mblk);
3385 		resp->mblk = NULL;
3386 		resp->data_val = NULL;
3387 		resp->data_len = 0;
3388 	}
3389 }
3390 
3391 static void
3392 rfs4_op_readdir_free(nfs_resop4 * resop)
3393 {
3394 	READDIR4res    *resp = &resop->nfs_resop4_u.opreaddir;
3395 
3396 	if (resp->status == NFS4_OK && resp->mblk != NULL) {
3397 		freeb(resp->mblk);
3398 		resp->mblk = NULL;
3399 		resp->data_len = 0;
3400 	}
3401 }
3402 
3403 
3404 /* ARGSUSED */
3405 static void
3406 rfs4_op_putpubfh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
3407     struct compound_state *cs)
3408 {
3409 	PUTPUBFH4res	*resp = &resop->nfs_resop4_u.opputpubfh;
3410 	int		error;
3411 	vnode_t		*vp;
3412 	struct exportinfo *exi, *sav_exi;
3413 	nfs_fh4_fmt_t	*fh_fmtp;
3414 
3415 	DTRACE_NFSV4_1(op__putpubfh__start, struct compound_state *, cs);
3416 
3417 	if (cs->vp) {
3418 		VN_RELE(cs->vp);
3419 		cs->vp = NULL;
3420 	}
3421 
3422 	if (cs->cr)
3423 		crfree(cs->cr);
3424 
3425 	cs->cr = crdup(cs->basecr);
3426 
3427 	vp = exi_public->exi_vp;
3428 	if (vp == NULL) {
3429 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3430 		goto out;
3431 	}
3432 
3433 	error = makefh4(&cs->fh, vp, exi_public);
3434 	if (error != 0) {
3435 		*cs->statusp = resp->status = puterrno4(error);
3436 		goto out;
3437 	}
3438 	sav_exi = cs->exi;
3439 	if (exi_public == exi_root) {
3440 		/*
3441 		 * No filesystem is actually shared public, so we default
3442 		 * to exi_root. In this case, we must check whether root
3443 		 * is exported.
3444 		 */
3445 		fh_fmtp = (nfs_fh4_fmt_t *)cs->fh.nfs_fh4_val;
3446 
3447 		/*
3448 		 * if root filesystem is exported, the exportinfo struct that we
3449 		 * should use is what checkexport4 returns, because root_exi is
3450 		 * actually a mostly empty struct.
3451 		 */
3452 		exi = checkexport4(&fh_fmtp->fh4_fsid,
3453 		    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
3454 		cs->exi = ((exi != NULL) ? exi : exi_public);
3455 	} else {
3456 		/*
3457 		 * it's a properly shared filesystem
3458 		 */
3459 		cs->exi = exi_public;
3460 	}
3461 
3462 	if (is_system_labeled()) {
3463 		bslabel_t *clabel;
3464 
3465 		ASSERT(req->rq_label != NULL);
3466 		clabel = req->rq_label;
3467 		DTRACE_PROBE2(tx__rfs4__log__info__opputpubfh__clabel, char *,
3468 		    "got client label from request(1)",
3469 		    struct svc_req *, req);
3470 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
3471 			if (!do_rfs_label_check(clabel, vp, DOMINANCE_CHECK,
3472 			    cs->exi)) {
3473 				*cs->statusp = resp->status =
3474 				    NFS4ERR_SERVERFAULT;
3475 				goto out;
3476 			}
3477 		}
3478 	}
3479 
3480 	VN_HOLD(vp);
3481 	cs->vp = vp;
3482 
3483 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3484 		VN_RELE(cs->vp);
3485 		cs->vp = NULL;
3486 		cs->exi = sav_exi;
3487 		goto out;
3488 	}
3489 
3490 	*cs->statusp = resp->status = NFS4_OK;
3491 out:
3492 	DTRACE_NFSV4_2(op__putpubfh__done, struct compound_state *, cs,
3493 	    PUTPUBFH4res *, resp);
3494 }
3495 
3496 /*
3497  * XXX - issue with put*fh operations. Suppose /export/home is exported.
3498  * Suppose an NFS client goes to mount /export/home/joe. If /export, home,
3499  * or joe have restrictive search permissions, then we shouldn't let
3500  * the client get a file handle. This is easy to enforce. However, we
3501  * don't know what security flavor should be used until we resolve the
3502  * path name. Another complication is uid mapping. If root is
3503  * the user, then it will be mapped to the anonymous user by default,
3504  * but we won't know that till we've resolved the path name. And we won't
3505  * know what the anonymous user is.
3506  * Luckily, SECINFO is specified to take a full filename.
3507  * So what we will have to in rfs4_op_lookup is check that flavor of
3508  * the target object matches that of the request, and if root was the
3509  * caller, check for the root= and anon= options, and if necessary,
3510  * repeat the lookup using the right cred_t. But that's not done yet.
3511  */
3512 /* ARGSUSED */
3513 static void
3514 rfs4_op_putfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3515     struct compound_state *cs)
3516 {
3517 	PUTFH4args *args = &argop->nfs_argop4_u.opputfh;
3518 	PUTFH4res *resp = &resop->nfs_resop4_u.opputfh;
3519 	nfs_fh4_fmt_t *fh_fmtp;
3520 
3521 	DTRACE_NFSV4_2(op__putfh__start, struct compound_state *, cs,
3522 	    PUTFH4args *, args);
3523 
3524 	if (cs->vp) {
3525 		VN_RELE(cs->vp);
3526 		cs->vp = NULL;
3527 	}
3528 
3529 	if (cs->cr) {
3530 		crfree(cs->cr);
3531 		cs->cr = NULL;
3532 	}
3533 
3534 
3535 	if (args->object.nfs_fh4_len < NFS_FH4_LEN) {
3536 		*cs->statusp = resp->status = NFS4ERR_BADHANDLE;
3537 		goto out;
3538 	}
3539 
3540 	fh_fmtp = (nfs_fh4_fmt_t *)args->object.nfs_fh4_val;
3541 	cs->exi = checkexport4(&fh_fmtp->fh4_fsid, (fid_t *)&fh_fmtp->fh4_xlen,
3542 	    NULL);
3543 
3544 	if (cs->exi == NULL) {
3545 		*cs->statusp = resp->status = NFS4ERR_STALE;
3546 		goto out;
3547 	}
3548 
3549 	cs->cr = crdup(cs->basecr);
3550 
3551 	ASSERT(cs->cr != NULL);
3552 
3553 	if (! (cs->vp = nfs4_fhtovp(&args->object, cs->exi, &resp->status))) {
3554 		*cs->statusp = resp->status;
3555 		goto out;
3556 	}
3557 
3558 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3559 		VN_RELE(cs->vp);
3560 		cs->vp = NULL;
3561 		goto out;
3562 	}
3563 
3564 	nfs_fh4_copy(&args->object, &cs->fh);
3565 	*cs->statusp = resp->status = NFS4_OK;
3566 	cs->deleg = FALSE;
3567 
3568 out:
3569 	DTRACE_NFSV4_2(op__putfh__done, struct compound_state *, cs,
3570 	    PUTFH4res *, resp);
3571 }
3572 
3573 /* ARGSUSED */
3574 static void
3575 rfs4_op_putrootfh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3576     struct compound_state *cs)
3577 {
3578 	PUTROOTFH4res *resp = &resop->nfs_resop4_u.opputrootfh;
3579 	int error;
3580 	fid_t fid;
3581 	struct exportinfo *exi, *sav_exi;
3582 
3583 	DTRACE_NFSV4_1(op__putrootfh__start, struct compound_state *, cs);
3584 
3585 	if (cs->vp) {
3586 		VN_RELE(cs->vp);
3587 		cs->vp = NULL;
3588 	}
3589 
3590 	if (cs->cr)
3591 		crfree(cs->cr);
3592 
3593 	cs->cr = crdup(cs->basecr);
3594 
3595 	/*
3596 	 * Using rootdir, the system root vnode,
3597 	 * get its fid.
3598 	 */
3599 	bzero(&fid, sizeof (fid));
3600 	fid.fid_len = MAXFIDSZ;
3601 	error = vop_fid_pseudo(rootdir, &fid);
3602 	if (error != 0) {
3603 		*cs->statusp = resp->status = puterrno4(error);
3604 		goto out;
3605 	}
3606 
3607 	/*
3608 	 * Then use the root fsid & fid it to find out if it's exported
3609 	 *
3610 	 * If the server root isn't exported directly, then
3611 	 * it should at least be a pseudo export based on
3612 	 * one or more exports further down in the server's
3613 	 * file tree.
3614 	 */
3615 	exi = checkexport4(&rootdir->v_vfsp->vfs_fsid, &fid, NULL);
3616 	if (exi == NULL || exi->exi_export.ex_flags & EX_PUBLIC) {
3617 		NFS4_DEBUG(rfs4_debug,
3618 		    (CE_WARN, "rfs4_op_putrootfh: export check failure"));
3619 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
3620 		goto out;
3621 	}
3622 
3623 	/*
3624 	 * Now make a filehandle based on the root
3625 	 * export and root vnode.
3626 	 */
3627 	error = makefh4(&cs->fh, rootdir, exi);
3628 	if (error != 0) {
3629 		*cs->statusp = resp->status = puterrno4(error);
3630 		goto out;
3631 	}
3632 
3633 	sav_exi = cs->exi;
3634 	cs->exi = exi;
3635 
3636 	VN_HOLD(rootdir);
3637 	cs->vp = rootdir;
3638 
3639 	if ((resp->status = call_checkauth4(cs, req)) != NFS4_OK) {
3640 		VN_RELE(rootdir);
3641 		cs->vp = NULL;
3642 		cs->exi = sav_exi;
3643 		goto out;
3644 	}
3645 
3646 	*cs->statusp = resp->status = NFS4_OK;
3647 	cs->deleg = FALSE;
3648 out:
3649 	DTRACE_NFSV4_2(op__putrootfh__done, struct compound_state *, cs,
3650 	    PUTROOTFH4res *, resp);
3651 }
3652 
3653 /*
3654  * A directory entry is a valid nfsv4 entry if
3655  * - it has a non-zero ino
3656  * - it is not a dot or dotdot name
3657  * - it is visible in a pseudo export or in a real export that can
3658  *   only have a limited view.
3659  */
3660 static bool_t
3661 valid_nfs4_entry(struct exportinfo *exi, struct dirent64 *dp,
3662     int *expseudo, int check_visible)
3663 {
3664 	if (dp->d_ino == 0 || NFS_IS_DOTNAME(dp->d_name)) {
3665 		*expseudo = 0;
3666 		return (FALSE);
3667 	}
3668 
3669 	if (! check_visible) {
3670 		*expseudo = 0;
3671 		return (TRUE);
3672 	}
3673 
3674 	return (nfs_visible_inode(exi, dp->d_ino, expseudo));
3675 }
3676 
3677 /*
3678  * set_rdattr_params sets up the variables used to manage what information
3679  * to get for each directory entry.
3680  */
3681 static nfsstat4
3682 set_rdattr_params(struct nfs4_svgetit_arg *sargp,
3683     bitmap4 attrs, bool_t *need_to_lookup)
3684 {
3685 	uint_t	va_mask;
3686 	nfsstat4 status;
3687 	bitmap4 objbits;
3688 
3689 	status = bitmap4_to_attrmask(attrs, sargp);
3690 	if (status != NFS4_OK) {
3691 		/*
3692 		 * could not even figure attr mask
3693 		 */
3694 		return (status);
3695 	}
3696 	va_mask = sargp->vap->va_mask;
3697 
3698 	/*
3699 	 * dirent's d_ino is always correct value for mounted_on_fileid.
3700 	 * mntdfid_set is set once here, but mounted_on_fileid is
3701 	 * set in main dirent processing loop for each dirent.
3702 	 * The mntdfid_set is a simple optimization that lets the
3703 	 * server attr code avoid work when caller is readdir.
3704 	 */
3705 	sargp->mntdfid_set = TRUE;
3706 
3707 	/*
3708 	 * Lookup entry only if client asked for any of the following:
3709 	 * a) vattr attrs
3710 	 * b) vfs attrs
3711 	 * c) attrs w/per-object scope requested (change, filehandle, etc)
3712 	 *    other than mounted_on_fileid (which we can take from dirent)
3713 	 */
3714 	objbits = attrs ? attrs & NFS4_VP_ATTR_MASK : 0;
3715 
3716 	if (va_mask || sargp->sbp || (objbits & ~FATTR4_MOUNTED_ON_FILEID_MASK))
3717 		*need_to_lookup = TRUE;
3718 	else
3719 		*need_to_lookup = FALSE;
3720 
3721 	if (sargp->sbp == NULL)
3722 		return (NFS4_OK);
3723 
3724 	/*
3725 	 * If filesystem attrs are requested, get them now from the
3726 	 * directory vp, as most entries will have same filesystem. The only
3727 	 * exception are mounted over entries but we handle
3728 	 * those as we go (XXX mounted over detection not yet implemented).
3729 	 */
3730 	sargp->vap->va_mask = 0;	/* to avoid VOP_GETATTR */
3731 	status = bitmap4_get_sysattrs(sargp);
3732 	sargp->vap->va_mask = va_mask;
3733 
3734 	if ((status != NFS4_OK) && sargp->rdattr_error_req) {
3735 		/*
3736 		 * Failed to get filesystem attributes.
3737 		 * Return a rdattr_error for each entry, but don't fail.
3738 		 * However, don't get any obj-dependent attrs.
3739 		 */
3740 		sargp->rdattr_error = status;	/* for rdattr_error */
3741 		*need_to_lookup = FALSE;
3742 		/*
3743 		 * At least get fileid for regular readdir output
3744 		 */
3745 		sargp->vap->va_mask &= AT_NODEID;
3746 		status = NFS4_OK;
3747 	}
3748 
3749 	return (status);
3750 }
3751 
3752 /*
3753  * readlink: args: CURRENT_FH.
3754  *	res: status. If success - CURRENT_FH unchanged, return linktext.
3755  */
3756 
3757 /* ARGSUSED */
3758 static void
3759 rfs4_op_readlink(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
3760     struct compound_state *cs)
3761 {
3762 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3763 	int error;
3764 	vnode_t *vp;
3765 	struct iovec iov;
3766 	struct vattr va;
3767 	struct uio uio;
3768 	char *data;
3769 	struct sockaddr *ca;
3770 	char *name = NULL;
3771 	int is_referral;
3772 
3773 	DTRACE_NFSV4_1(op__readlink__start, struct compound_state *, cs);
3774 
3775 	/* CURRENT_FH: directory */
3776 	vp = cs->vp;
3777 	if (vp == NULL) {
3778 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
3779 		goto out;
3780 	}
3781 
3782 	if (cs->access == CS_ACCESS_DENIED) {
3783 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3784 		goto out;
3785 	}
3786 
3787 	/* Is it a referral? */
3788 	if (vn_is_nfs_reparse(vp, cs->cr) && client_is_downrev(req)) {
3789 
3790 		is_referral = 1;
3791 
3792 	} else {
3793 
3794 		is_referral = 0;
3795 
3796 		if (vp->v_type == VDIR) {
3797 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
3798 			goto out;
3799 		}
3800 
3801 		if (vp->v_type != VLNK) {
3802 			*cs->statusp = resp->status = NFS4ERR_INVAL;
3803 			goto out;
3804 		}
3805 
3806 	}
3807 
3808 	va.va_mask = AT_MODE;
3809 	error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
3810 	if (error) {
3811 		*cs->statusp = resp->status = puterrno4(error);
3812 		goto out;
3813 	}
3814 
3815 	if (MANDLOCK(vp, va.va_mode)) {
3816 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
3817 		goto out;
3818 	}
3819 
3820 	data = kmem_alloc(MAXPATHLEN + 1, KM_SLEEP);
3821 
3822 	if (is_referral) {
3823 		char *s;
3824 		size_t strsz;
3825 
3826 		/* Get an artificial symlink based on a referral */
3827 		s = build_symlink(vp, cs->cr, &strsz);
3828 		global_svstat_ptr[4][NFS_REFERLINKS].value.ui64++;
3829 		DTRACE_PROBE2(nfs4serv__func__referral__reflink,
3830 		    vnode_t *, vp, char *, s);
3831 		if (s == NULL)
3832 			error = EINVAL;
3833 		else {
3834 			error = 0;
3835 			(void) strlcpy(data, s, MAXPATHLEN + 1);
3836 			kmem_free(s, strsz);
3837 		}
3838 
3839 	} else {
3840 
3841 		iov.iov_base = data;
3842 		iov.iov_len = MAXPATHLEN;
3843 		uio.uio_iov = &iov;
3844 		uio.uio_iovcnt = 1;
3845 		uio.uio_segflg = UIO_SYSSPACE;
3846 		uio.uio_extflg = UIO_COPY_CACHED;
3847 		uio.uio_loffset = 0;
3848 		uio.uio_resid = MAXPATHLEN;
3849 
3850 		error = VOP_READLINK(vp, &uio, cs->cr, NULL);
3851 
3852 		if (!error)
3853 			*(data + MAXPATHLEN - uio.uio_resid) = '\0';
3854 	}
3855 
3856 	if (error) {
3857 		kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3858 		*cs->statusp = resp->status = puterrno4(error);
3859 		goto out;
3860 	}
3861 
3862 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
3863 	name = nfscmd_convname(ca, cs->exi, data, NFSCMD_CONV_OUTBOUND,
3864 	    MAXPATHLEN  + 1);
3865 
3866 	if (name == NULL) {
3867 		/*
3868 		 * Even though the conversion failed, we return
3869 		 * something. We just don't translate it.
3870 		 */
3871 		name = data;
3872 	}
3873 
3874 	/*
3875 	 * treat link name as data
3876 	 */
3877 	(void) str_to_utf8(name, &resp->link);
3878 
3879 	if (name != data)
3880 		kmem_free(name, MAXPATHLEN + 1);
3881 	kmem_free((caddr_t)data, (uint_t)MAXPATHLEN + 1);
3882 	*cs->statusp = resp->status = NFS4_OK;
3883 
3884 out:
3885 	DTRACE_NFSV4_2(op__readlink__done, struct compound_state *, cs,
3886 	    READLINK4res *, resp);
3887 }
3888 
3889 static void
3890 rfs4_op_readlink_free(nfs_resop4 *resop)
3891 {
3892 	READLINK4res *resp = &resop->nfs_resop4_u.opreadlink;
3893 	utf8string *symlink = &resp->link;
3894 
3895 	if (symlink->utf8string_val) {
3896 		UTF8STRING_FREE(*symlink)
3897 	}
3898 }
3899 
3900 /*
3901  * release_lockowner:
3902  *	Release any state associated with the supplied
3903  *	lockowner. Note if any lo_state is holding locks we will not
3904  *	rele that lo_state and thus the lockowner will not be destroyed.
3905  *	A client using lock after the lock owner stateid has been released
3906  *	will suffer the consequence of NFS4ERR_BAD_STATEID and would have
3907  *	to reissue the lock with new_lock_owner set to TRUE.
3908  *	args: lock_owner
3909  *	res:  status
3910  */
3911 /* ARGSUSED */
3912 static void
3913 rfs4_op_release_lockowner(nfs_argop4 *argop, nfs_resop4 *resop,
3914     struct svc_req *req, struct compound_state *cs)
3915 {
3916 	RELEASE_LOCKOWNER4args *ap = &argop->nfs_argop4_u.oprelease_lockowner;
3917 	RELEASE_LOCKOWNER4res *resp = &resop->nfs_resop4_u.oprelease_lockowner;
3918 	rfs4_lockowner_t *lo;
3919 	rfs4_openowner_t *oo;
3920 	rfs4_state_t *sp;
3921 	rfs4_lo_state_t *lsp;
3922 	rfs4_client_t *cp;
3923 	bool_t create = FALSE;
3924 	locklist_t *llist;
3925 	sysid_t sysid;
3926 
3927 	DTRACE_NFSV4_2(op__release__lockowner__start, struct compound_state *,
3928 	    cs, RELEASE_LOCKOWNER4args *, ap);
3929 
3930 	/* Make sure there is a clientid around for this request */
3931 	cp = rfs4_findclient_by_id(ap->lock_owner.clientid, FALSE);
3932 
3933 	if (cp == NULL) {
3934 		*cs->statusp = resp->status =
3935 		    rfs4_check_clientid(&ap->lock_owner.clientid, 0);
3936 		goto out;
3937 	}
3938 	rfs4_client_rele(cp);
3939 
3940 	lo = rfs4_findlockowner(&ap->lock_owner, &create);
3941 	if (lo == NULL) {
3942 		*cs->statusp = resp->status = NFS4_OK;
3943 		goto out;
3944 	}
3945 	ASSERT(lo->rl_client != NULL);
3946 
3947 	/*
3948 	 * Check for EXPIRED client. If so will reap state with in a lease
3949 	 * period or on next set_clientid_confirm step
3950 	 */
3951 	if (rfs4_lease_expired(lo->rl_client)) {
3952 		rfs4_lockowner_rele(lo);
3953 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
3954 		goto out;
3955 	}
3956 
3957 	/*
3958 	 * If no sysid has been assigned, then no locks exist; just return.
3959 	 */
3960 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
3961 	if (lo->rl_client->rc_sysidt == LM_NOSYSID) {
3962 		rfs4_lockowner_rele(lo);
3963 		rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3964 		goto out;
3965 	}
3966 
3967 	sysid = lo->rl_client->rc_sysidt;
3968 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
3969 
3970 	/*
3971 	 * Mark the lockowner invalid.
3972 	 */
3973 	rfs4_dbe_hide(lo->rl_dbe);
3974 
3975 	/*
3976 	 * sysid-pid pair should now not be used since the lockowner is
3977 	 * invalid. If the client were to instantiate the lockowner again
3978 	 * it would be assigned a new pid. Thus we can get the list of
3979 	 * current locks.
3980 	 */
3981 
3982 	llist = flk_get_active_locks(sysid, lo->rl_pid);
3983 	/* If we are still holding locks fail */
3984 	if (llist != NULL) {
3985 
3986 		*cs->statusp = resp->status = NFS4ERR_LOCKS_HELD;
3987 
3988 		flk_free_locklist(llist);
3989 		/*
3990 		 * We need to unhide the lockowner so the client can
3991 		 * try it again. The bad thing here is if the client
3992 		 * has a logic error that took it here in the first place
3993 		 * he probably has lost accounting of the locks that it
3994 		 * is holding. So we may have dangling state until the
3995 		 * open owner state is reaped via close. One scenario
3996 		 * that could possibly occur is that the client has
3997 		 * sent the unlock request(s) in separate threads
3998 		 * and has not waited for the replies before sending the
3999 		 * RELEASE_LOCKOWNER request. Presumably, it would expect
4000 		 * and deal appropriately with NFS4ERR_LOCKS_HELD, by
4001 		 * reissuing the request.
4002 		 */
4003 		rfs4_dbe_unhide(lo->rl_dbe);
4004 		rfs4_lockowner_rele(lo);
4005 		goto out;
4006 	}
4007 
4008 	/*
4009 	 * For the corresponding client we need to check each open
4010 	 * owner for any opens that have lockowner state associated
4011 	 * with this lockowner.
4012 	 */
4013 
4014 	rfs4_dbe_lock(lo->rl_client->rc_dbe);
4015 	for (oo = list_head(&lo->rl_client->rc_openownerlist); oo != NULL;
4016 	    oo = list_next(&lo->rl_client->rc_openownerlist, oo)) {
4017 
4018 		rfs4_dbe_lock(oo->ro_dbe);
4019 		for (sp = list_head(&oo->ro_statelist); sp != NULL;
4020 		    sp = list_next(&oo->ro_statelist, sp)) {
4021 
4022 			rfs4_dbe_lock(sp->rs_dbe);
4023 			for (lsp = list_head(&sp->rs_lostatelist);
4024 			    lsp != NULL;
4025 			    lsp = list_next(&sp->rs_lostatelist, lsp)) {
4026 				if (lsp->rls_locker == lo) {
4027 					rfs4_dbe_lock(lsp->rls_dbe);
4028 					rfs4_dbe_invalidate(lsp->rls_dbe);
4029 					rfs4_dbe_unlock(lsp->rls_dbe);
4030 				}
4031 			}
4032 			rfs4_dbe_unlock(sp->rs_dbe);
4033 		}
4034 		rfs4_dbe_unlock(oo->ro_dbe);
4035 	}
4036 	rfs4_dbe_unlock(lo->rl_client->rc_dbe);
4037 
4038 	rfs4_lockowner_rele(lo);
4039 
4040 	*cs->statusp = resp->status = NFS4_OK;
4041 
4042 out:
4043 	DTRACE_NFSV4_2(op__release__lockowner__done, struct compound_state *,
4044 	    cs, RELEASE_LOCKOWNER4res *, resp);
4045 }
4046 
4047 /*
4048  * short utility function to lookup a file and recall the delegation
4049  */
4050 static rfs4_file_t *
4051 rfs4_lookup_and_findfile(vnode_t *dvp, char *nm, vnode_t **vpp,
4052     int *lkup_error, cred_t *cr)
4053 {
4054 	vnode_t *vp;
4055 	rfs4_file_t *fp = NULL;
4056 	bool_t fcreate = FALSE;
4057 	int error;
4058 
4059 	if (vpp)
4060 		*vpp = NULL;
4061 
4062 	if ((error = VOP_LOOKUP(dvp, nm, &vp, NULL, 0, NULL, cr, NULL, NULL,
4063 	    NULL)) == 0) {
4064 		if (vp->v_type == VREG)
4065 			fp = rfs4_findfile(vp, NULL, &fcreate);
4066 		if (vpp)
4067 			*vpp = vp;
4068 		else
4069 			VN_RELE(vp);
4070 	}
4071 
4072 	if (lkup_error)
4073 		*lkup_error = error;
4074 
4075 	return (fp);
4076 }
4077 
4078 /*
4079  * remove: args: CURRENT_FH: directory; name.
4080  *	res: status. If success - CURRENT_FH unchanged, return change_info
4081  *		for directory.
4082  */
4083 /* ARGSUSED */
4084 static void
4085 rfs4_op_remove(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4086     struct compound_state *cs)
4087 {
4088 	REMOVE4args *args = &argop->nfs_argop4_u.opremove;
4089 	REMOVE4res *resp = &resop->nfs_resop4_u.opremove;
4090 	int error;
4091 	vnode_t *dvp, *vp;
4092 	struct vattr bdva, idva, adva;
4093 	char *nm;
4094 	uint_t len;
4095 	rfs4_file_t *fp;
4096 	int in_crit = 0;
4097 	bslabel_t *clabel;
4098 	struct sockaddr *ca;
4099 	char *name = NULL;
4100 
4101 	DTRACE_NFSV4_2(op__remove__start, struct compound_state *, cs,
4102 	    REMOVE4args *, args);
4103 
4104 	/* CURRENT_FH: directory */
4105 	dvp = cs->vp;
4106 	if (dvp == NULL) {
4107 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4108 		goto out;
4109 	}
4110 
4111 	if (cs->access == CS_ACCESS_DENIED) {
4112 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4113 		goto out;
4114 	}
4115 
4116 	/*
4117 	 * If there is an unshared filesystem mounted on this vnode,
4118 	 * Do not allow to remove anything in this directory.
4119 	 */
4120 	if (vn_ismntpt(dvp)) {
4121 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4122 		goto out;
4123 	}
4124 
4125 	if (dvp->v_type != VDIR) {
4126 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4127 		goto out;
4128 	}
4129 
4130 	if (!utf8_dir_verify(&args->target)) {
4131 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4132 		goto out;
4133 	}
4134 
4135 	/*
4136 	 * Lookup the file so that we can check if it's a directory
4137 	 */
4138 	nm = utf8_to_fn(&args->target, &len, NULL);
4139 	if (nm == NULL) {
4140 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4141 		goto out;
4142 	}
4143 
4144 	if (len > MAXNAMELEN) {
4145 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4146 		kmem_free(nm, len);
4147 		goto out;
4148 	}
4149 
4150 	if (rdonly4(cs->exi, cs->vp, req)) {
4151 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4152 		kmem_free(nm, len);
4153 		goto out;
4154 	}
4155 
4156 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4157 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
4158 	    MAXPATHLEN  + 1);
4159 
4160 	if (name == NULL) {
4161 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4162 		kmem_free(nm, len);
4163 		goto out;
4164 	}
4165 
4166 	/*
4167 	 * Lookup the file to determine type and while we are see if
4168 	 * there is a file struct around and check for delegation.
4169 	 * We don't need to acquire va_seq before this lookup, if
4170 	 * it causes an update, cinfo.before will not match, which will
4171 	 * trigger a cache flush even if atomic is TRUE.
4172 	 */
4173 	if (fp = rfs4_lookup_and_findfile(dvp, name, &vp, &error, cs->cr)) {
4174 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4175 		    NULL)) {
4176 			VN_RELE(vp);
4177 			rfs4_file_rele(fp);
4178 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4179 			if (nm != name)
4180 				kmem_free(name, MAXPATHLEN + 1);
4181 			kmem_free(nm, len);
4182 			goto out;
4183 		}
4184 	}
4185 
4186 	/* Didn't find anything to remove */
4187 	if (vp == NULL) {
4188 		*cs->statusp = resp->status = error;
4189 		if (nm != name)
4190 			kmem_free(name, MAXPATHLEN + 1);
4191 		kmem_free(nm, len);
4192 		goto out;
4193 	}
4194 
4195 	if (nbl_need_check(vp)) {
4196 		nbl_start_crit(vp, RW_READER);
4197 		in_crit = 1;
4198 		if (nbl_conflict(vp, NBL_REMOVE, 0, 0, 0, NULL)) {
4199 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4200 			if (nm != name)
4201 				kmem_free(name, MAXPATHLEN + 1);
4202 			kmem_free(nm, len);
4203 			nbl_end_crit(vp);
4204 			VN_RELE(vp);
4205 			if (fp) {
4206 				rfs4_clear_dont_grant(fp);
4207 				rfs4_file_rele(fp);
4208 			}
4209 			goto out;
4210 		}
4211 	}
4212 
4213 	/* check label before allowing removal */
4214 	if (is_system_labeled()) {
4215 		ASSERT(req->rq_label != NULL);
4216 		clabel = req->rq_label;
4217 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
4218 		    "got client label from request(1)",
4219 		    struct svc_req *, req);
4220 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4221 			if (!do_rfs_label_check(clabel, vp, EQUALITY_CHECK,
4222 			    cs->exi)) {
4223 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4224 				if (name != nm)
4225 					kmem_free(name, MAXPATHLEN + 1);
4226 				kmem_free(nm, len);
4227 				if (in_crit)
4228 					nbl_end_crit(vp);
4229 				VN_RELE(vp);
4230 				if (fp) {
4231 					rfs4_clear_dont_grant(fp);
4232 					rfs4_file_rele(fp);
4233 				}
4234 				goto out;
4235 			}
4236 		}
4237 	}
4238 
4239 	/* Get dir "before" change value */
4240 	bdva.va_mask = AT_CTIME|AT_SEQ;
4241 	error = VOP_GETATTR(dvp, &bdva, 0, cs->cr, NULL);
4242 	if (error) {
4243 		*cs->statusp = resp->status = puterrno4(error);
4244 		if (nm != name)
4245 			kmem_free(name, MAXPATHLEN + 1);
4246 		kmem_free(nm, len);
4247 		if (in_crit)
4248 			nbl_end_crit(vp);
4249 		VN_RELE(vp);
4250 		if (fp) {
4251 			rfs4_clear_dont_grant(fp);
4252 			rfs4_file_rele(fp);
4253 		}
4254 		goto out;
4255 	}
4256 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.before, bdva.va_ctime)
4257 
4258 	/* Actually do the REMOVE operation */
4259 	if (vp->v_type == VDIR) {
4260 		/*
4261 		 * Can't remove a directory that has a mounted-on filesystem.
4262 		 */
4263 		if (vn_ismntpt(vp)) {
4264 			error = EACCES;
4265 		} else {
4266 			/*
4267 			 * System V defines rmdir to return EEXIST,
4268 			 * not * ENOTEMPTY, if the directory is not
4269 			 * empty.  A System V NFS server needs to map
4270 			 * NFS4ERR_EXIST to NFS4ERR_NOTEMPTY to
4271 			 * transmit over the wire.
4272 			 */
4273 			if ((error = VOP_RMDIR(dvp, nm, rootdir, cs->cr,
4274 			    NULL, 0)) == EEXIST)
4275 				error = ENOTEMPTY;
4276 		}
4277 	} else {
4278 		if ((error = VOP_REMOVE(dvp, name, cs->cr, NULL, 0)) == 0 &&
4279 		    fp != NULL) {
4280 			struct vattr va;
4281 			vnode_t *tvp;
4282 
4283 			rfs4_dbe_lock(fp->rf_dbe);
4284 			tvp = fp->rf_vp;
4285 			if (tvp)
4286 				VN_HOLD(tvp);
4287 			rfs4_dbe_unlock(fp->rf_dbe);
4288 
4289 			if (tvp) {
4290 				/*
4291 				 * This is va_seq safe because we are not
4292 				 * manipulating dvp.
4293 				 */
4294 				va.va_mask = AT_NLINK;
4295 				if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4296 				    va.va_nlink == 0) {
4297 					/* Remove state on file remove */
4298 					if (in_crit) {
4299 						nbl_end_crit(vp);
4300 						in_crit = 0;
4301 					}
4302 					rfs4_close_all_state(fp);
4303 				}
4304 				VN_RELE(tvp);
4305 			}
4306 		}
4307 	}
4308 
4309 	if (in_crit)
4310 		nbl_end_crit(vp);
4311 	VN_RELE(vp);
4312 
4313 	if (fp) {
4314 		rfs4_clear_dont_grant(fp);
4315 		rfs4_file_rele(fp);
4316 	}
4317 	if (nm != name)
4318 		kmem_free(name, MAXPATHLEN + 1);
4319 	kmem_free(nm, len);
4320 
4321 	if (error) {
4322 		*cs->statusp = resp->status = puterrno4(error);
4323 		goto out;
4324 	}
4325 
4326 	/*
4327 	 * Get the initial "after" sequence number, if it fails, set to zero
4328 	 */
4329 	idva.va_mask = AT_SEQ;
4330 	if (VOP_GETATTR(dvp, &idva, 0, cs->cr, NULL))
4331 		idva.va_seq = 0;
4332 
4333 	/*
4334 	 * Force modified data and metadata out to stable storage.
4335 	 */
4336 	(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
4337 
4338 	/*
4339 	 * Get "after" change value, if it fails, simply return the
4340 	 * before value.
4341 	 */
4342 	adva.va_mask = AT_CTIME|AT_SEQ;
4343 	if (VOP_GETATTR(dvp, &adva, 0, cs->cr, NULL)) {
4344 		adva.va_ctime = bdva.va_ctime;
4345 		adva.va_seq = 0;
4346 	}
4347 
4348 	NFS4_SET_FATTR4_CHANGE(resp->cinfo.after, adva.va_ctime)
4349 
4350 	/*
4351 	 * The cinfo.atomic = TRUE only if we have
4352 	 * non-zero va_seq's, and it has incremented by exactly one
4353 	 * during the VOP_REMOVE/RMDIR and it didn't change during
4354 	 * the VOP_FSYNC.
4355 	 */
4356 	if (bdva.va_seq && idva.va_seq && adva.va_seq &&
4357 	    idva.va_seq == (bdva.va_seq + 1) && idva.va_seq == adva.va_seq)
4358 		resp->cinfo.atomic = TRUE;
4359 	else
4360 		resp->cinfo.atomic = FALSE;
4361 
4362 	*cs->statusp = resp->status = NFS4_OK;
4363 
4364 out:
4365 	DTRACE_NFSV4_2(op__remove__done, struct compound_state *, cs,
4366 	    REMOVE4res *, resp);
4367 }
4368 
4369 /*
4370  * rename: args: SAVED_FH: from directory, CURRENT_FH: target directory,
4371  *		oldname and newname.
4372  *	res: status. If success - CURRENT_FH unchanged, return change_info
4373  *		for both from and target directories.
4374  */
4375 /* ARGSUSED */
4376 static void
4377 rfs4_op_rename(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4378     struct compound_state *cs)
4379 {
4380 	RENAME4args *args = &argop->nfs_argop4_u.oprename;
4381 	RENAME4res *resp = &resop->nfs_resop4_u.oprename;
4382 	int error;
4383 	vnode_t *odvp;
4384 	vnode_t *ndvp;
4385 	vnode_t *srcvp, *targvp;
4386 	struct vattr obdva, oidva, oadva;
4387 	struct vattr nbdva, nidva, nadva;
4388 	char *onm, *nnm;
4389 	uint_t olen, nlen;
4390 	rfs4_file_t *fp, *sfp;
4391 	int in_crit_src, in_crit_targ;
4392 	int fp_rele_grant_hold, sfp_rele_grant_hold;
4393 	bslabel_t *clabel;
4394 	struct sockaddr *ca;
4395 	char *converted_onm = NULL;
4396 	char *converted_nnm = NULL;
4397 
4398 	DTRACE_NFSV4_2(op__rename__start, struct compound_state *, cs,
4399 	    RENAME4args *, args);
4400 
4401 	fp = sfp = NULL;
4402 	srcvp = targvp = NULL;
4403 	in_crit_src = in_crit_targ = 0;
4404 	fp_rele_grant_hold = sfp_rele_grant_hold = 0;
4405 
4406 	/* CURRENT_FH: target directory */
4407 	ndvp = cs->vp;
4408 	if (ndvp == NULL) {
4409 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4410 		goto out;
4411 	}
4412 
4413 	/* SAVED_FH: from directory */
4414 	odvp = cs->saved_vp;
4415 	if (odvp == NULL) {
4416 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4417 		goto out;
4418 	}
4419 
4420 	if (cs->access == CS_ACCESS_DENIED) {
4421 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4422 		goto out;
4423 	}
4424 
4425 	/*
4426 	 * If there is an unshared filesystem mounted on this vnode,
4427 	 * do not allow to rename objects in this directory.
4428 	 */
4429 	if (vn_ismntpt(odvp)) {
4430 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4431 		goto out;
4432 	}
4433 
4434 	/*
4435 	 * If there is an unshared filesystem mounted on this vnode,
4436 	 * do not allow to rename to this directory.
4437 	 */
4438 	if (vn_ismntpt(ndvp)) {
4439 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
4440 		goto out;
4441 	}
4442 
4443 	if (odvp->v_type != VDIR || ndvp->v_type != VDIR) {
4444 		*cs->statusp = resp->status = NFS4ERR_NOTDIR;
4445 		goto out;
4446 	}
4447 
4448 	if (cs->saved_exi != cs->exi) {
4449 		*cs->statusp = resp->status = NFS4ERR_XDEV;
4450 		goto out;
4451 	}
4452 
4453 	if (!utf8_dir_verify(&args->oldname)) {
4454 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4455 		goto out;
4456 	}
4457 
4458 	if (!utf8_dir_verify(&args->newname)) {
4459 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4460 		goto out;
4461 	}
4462 
4463 	onm = utf8_to_fn(&args->oldname, &olen, NULL);
4464 	if (onm == NULL) {
4465 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4466 		goto out;
4467 	}
4468 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
4469 	nlen = MAXPATHLEN + 1;
4470 	converted_onm = nfscmd_convname(ca, cs->exi, onm, NFSCMD_CONV_INBOUND,
4471 	    nlen);
4472 
4473 	if (converted_onm == NULL) {
4474 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4475 		kmem_free(onm, olen);
4476 		goto out;
4477 	}
4478 
4479 	nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4480 	if (nnm == NULL) {
4481 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4482 		if (onm != converted_onm)
4483 			kmem_free(converted_onm, MAXPATHLEN + 1);
4484 		kmem_free(onm, olen);
4485 		goto out;
4486 	}
4487 	converted_nnm = nfscmd_convname(ca, cs->exi, nnm, NFSCMD_CONV_INBOUND,
4488 	    MAXPATHLEN  + 1);
4489 
4490 	if (converted_nnm == NULL) {
4491 		*cs->statusp = resp->status = NFS4ERR_INVAL;
4492 		kmem_free(nnm, nlen);
4493 		nnm = NULL;
4494 		if (onm != converted_onm)
4495 			kmem_free(converted_onm, MAXPATHLEN + 1);
4496 		kmem_free(onm, olen);
4497 		goto out;
4498 	}
4499 
4500 
4501 	if (olen > MAXNAMELEN || nlen > MAXNAMELEN) {
4502 		*cs->statusp = resp->status = NFS4ERR_NAMETOOLONG;
4503 		kmem_free(onm, olen);
4504 		kmem_free(nnm, nlen);
4505 		goto out;
4506 	}
4507 
4508 
4509 	if (rdonly4(cs->exi, cs->vp, req)) {
4510 		*cs->statusp = resp->status = NFS4ERR_ROFS;
4511 		if (onm != converted_onm)
4512 			kmem_free(converted_onm, MAXPATHLEN + 1);
4513 		kmem_free(onm, olen);
4514 		if (nnm != converted_nnm)
4515 			kmem_free(converted_nnm, MAXPATHLEN + 1);
4516 		kmem_free(nnm, nlen);
4517 		goto out;
4518 	}
4519 
4520 	/* check label of the target dir */
4521 	if (is_system_labeled()) {
4522 		ASSERT(req->rq_label != NULL);
4523 		clabel = req->rq_label;
4524 		DTRACE_PROBE2(tx__rfs4__log__info__oprename__clabel, char *,
4525 		    "got client label from request(1)",
4526 		    struct svc_req *, req);
4527 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
4528 			if (!do_rfs_label_check(clabel, ndvp,
4529 			    EQUALITY_CHECK, cs->exi)) {
4530 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
4531 				goto err_out;
4532 			}
4533 		}
4534 	}
4535 
4536 	/*
4537 	 * Is the source a file and have a delegation?
4538 	 * We don't need to acquire va_seq before these lookups, if
4539 	 * it causes an update, cinfo.before will not match, which will
4540 	 * trigger a cache flush even if atomic is TRUE.
4541 	 */
4542 	if (sfp = rfs4_lookup_and_findfile(odvp, converted_onm, &srcvp,
4543 	    &error, cs->cr)) {
4544 		if (rfs4_check_delegated_byfp(FWRITE, sfp, TRUE, TRUE, TRUE,
4545 		    NULL)) {
4546 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4547 			goto err_out;
4548 		}
4549 	}
4550 
4551 	if (srcvp == NULL) {
4552 		*cs->statusp = resp->status = puterrno4(error);
4553 		if (onm != converted_onm)
4554 			kmem_free(converted_onm, MAXPATHLEN + 1);
4555 		kmem_free(onm, olen);
4556 		if (nnm != converted_nnm)
4557 			kmem_free(converted_onm, MAXPATHLEN + 1);
4558 		kmem_free(nnm, nlen);
4559 		goto out;
4560 	}
4561 
4562 	sfp_rele_grant_hold = 1;
4563 
4564 	/* Does the destination exist and a file and have a delegation? */
4565 	if (fp = rfs4_lookup_and_findfile(ndvp, converted_nnm, &targvp,
4566 	    NULL, cs->cr)) {
4567 		if (rfs4_check_delegated_byfp(FWRITE, fp, TRUE, TRUE, TRUE,
4568 		    NULL)) {
4569 			*cs->statusp = resp->status = NFS4ERR_DELAY;
4570 			goto err_out;
4571 		}
4572 	}
4573 	fp_rele_grant_hold = 1;
4574 
4575 
4576 	/* Check for NBMAND lock on both source and target */
4577 	if (nbl_need_check(srcvp)) {
4578 		nbl_start_crit(srcvp, RW_READER);
4579 		in_crit_src = 1;
4580 		if (nbl_conflict(srcvp, NBL_RENAME, 0, 0, 0, NULL)) {
4581 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4582 			goto err_out;
4583 		}
4584 	}
4585 
4586 	if (targvp && nbl_need_check(targvp)) {
4587 		nbl_start_crit(targvp, RW_READER);
4588 		in_crit_targ = 1;
4589 		if (nbl_conflict(targvp, NBL_REMOVE, 0, 0, 0, NULL)) {
4590 			*cs->statusp = resp->status = NFS4ERR_FILE_OPEN;
4591 			goto err_out;
4592 		}
4593 	}
4594 
4595 	/* Get source "before" change value */
4596 	obdva.va_mask = AT_CTIME|AT_SEQ;
4597 	error = VOP_GETATTR(odvp, &obdva, 0, cs->cr, NULL);
4598 	if (!error) {
4599 		nbdva.va_mask = AT_CTIME|AT_SEQ;
4600 		error = VOP_GETATTR(ndvp, &nbdva, 0, cs->cr, NULL);
4601 	}
4602 	if (error) {
4603 		*cs->statusp = resp->status = puterrno4(error);
4604 		goto err_out;
4605 	}
4606 
4607 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.before, obdva.va_ctime)
4608 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.before, nbdva.va_ctime)
4609 
4610 	if ((error = VOP_RENAME(odvp, converted_onm, ndvp, converted_nnm,
4611 	    cs->cr, NULL, 0)) == 0 && fp != NULL) {
4612 		struct vattr va;
4613 		vnode_t *tvp;
4614 
4615 		rfs4_dbe_lock(fp->rf_dbe);
4616 		tvp = fp->rf_vp;
4617 		if (tvp)
4618 			VN_HOLD(tvp);
4619 		rfs4_dbe_unlock(fp->rf_dbe);
4620 
4621 		if (tvp) {
4622 			va.va_mask = AT_NLINK;
4623 			if (!VOP_GETATTR(tvp, &va, 0, cs->cr, NULL) &&
4624 			    va.va_nlink == 0) {
4625 				/* The file is gone and so should the state */
4626 				if (in_crit_targ) {
4627 					nbl_end_crit(targvp);
4628 					in_crit_targ = 0;
4629 				}
4630 				rfs4_close_all_state(fp);
4631 			}
4632 			VN_RELE(tvp);
4633 		}
4634 	}
4635 	if (error == 0)
4636 		vn_renamepath(ndvp, srcvp, nnm, nlen - 1);
4637 
4638 	if (in_crit_src)
4639 		nbl_end_crit(srcvp);
4640 	if (srcvp)
4641 		VN_RELE(srcvp);
4642 	if (in_crit_targ)
4643 		nbl_end_crit(targvp);
4644 	if (targvp)
4645 		VN_RELE(targvp);
4646 
4647 	if (sfp) {
4648 		rfs4_clear_dont_grant(sfp);
4649 		rfs4_file_rele(sfp);
4650 	}
4651 	if (fp) {
4652 		rfs4_clear_dont_grant(fp);
4653 		rfs4_file_rele(fp);
4654 	}
4655 
4656 	if (converted_onm != onm)
4657 		kmem_free(converted_onm, MAXPATHLEN + 1);
4658 	kmem_free(onm, olen);
4659 	if (converted_nnm != nnm)
4660 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4661 	kmem_free(nnm, nlen);
4662 
4663 	/*
4664 	 * Get the initial "after" sequence number, if it fails, set to zero
4665 	 */
4666 	oidva.va_mask = AT_SEQ;
4667 	if (VOP_GETATTR(odvp, &oidva, 0, cs->cr, NULL))
4668 		oidva.va_seq = 0;
4669 
4670 	nidva.va_mask = AT_SEQ;
4671 	if (VOP_GETATTR(ndvp, &nidva, 0, cs->cr, NULL))
4672 		nidva.va_seq = 0;
4673 
4674 	/*
4675 	 * Force modified data and metadata out to stable storage.
4676 	 */
4677 	(void) VOP_FSYNC(odvp, 0, cs->cr, NULL);
4678 	(void) VOP_FSYNC(ndvp, 0, cs->cr, NULL);
4679 
4680 	if (error) {
4681 		*cs->statusp = resp->status = puterrno4(error);
4682 		goto out;
4683 	}
4684 
4685 	/*
4686 	 * Get "after" change values, if it fails, simply return the
4687 	 * before value.
4688 	 */
4689 	oadva.va_mask = AT_CTIME|AT_SEQ;
4690 	if (VOP_GETATTR(odvp, &oadva, 0, cs->cr, NULL)) {
4691 		oadva.va_ctime = obdva.va_ctime;
4692 		oadva.va_seq = 0;
4693 	}
4694 
4695 	nadva.va_mask = AT_CTIME|AT_SEQ;
4696 	if (VOP_GETATTR(odvp, &nadva, 0, cs->cr, NULL)) {
4697 		nadva.va_ctime = nbdva.va_ctime;
4698 		nadva.va_seq = 0;
4699 	}
4700 
4701 	NFS4_SET_FATTR4_CHANGE(resp->source_cinfo.after, oadva.va_ctime)
4702 	NFS4_SET_FATTR4_CHANGE(resp->target_cinfo.after, nadva.va_ctime)
4703 
4704 	/*
4705 	 * The cinfo.atomic = TRUE only if we have
4706 	 * non-zero va_seq's, and it has incremented by exactly one
4707 	 * during the VOP_RENAME and it didn't change during the VOP_FSYNC.
4708 	 */
4709 	if (obdva.va_seq && oidva.va_seq && oadva.va_seq &&
4710 	    oidva.va_seq == (obdva.va_seq + 1) && oidva.va_seq == oadva.va_seq)
4711 		resp->source_cinfo.atomic = TRUE;
4712 	else
4713 		resp->source_cinfo.atomic = FALSE;
4714 
4715 	if (nbdva.va_seq && nidva.va_seq && nadva.va_seq &&
4716 	    nidva.va_seq == (nbdva.va_seq + 1) && nidva.va_seq == nadva.va_seq)
4717 		resp->target_cinfo.atomic = TRUE;
4718 	else
4719 		resp->target_cinfo.atomic = FALSE;
4720 
4721 #ifdef	VOLATILE_FH_TEST
4722 	{
4723 	extern void add_volrnm_fh(struct exportinfo *, vnode_t *);
4724 
4725 	/*
4726 	 * Add the renamed file handle to the volatile rename list
4727 	 */
4728 	if (cs->exi->exi_export.ex_flags & EX_VOLRNM) {
4729 		/* file handles may expire on rename */
4730 		vnode_t *vp;
4731 
4732 		nnm = utf8_to_fn(&args->newname, &nlen, NULL);
4733 		/*
4734 		 * Already know that nnm will be a valid string
4735 		 */
4736 		error = VOP_LOOKUP(ndvp, nnm, &vp, NULL, 0, NULL, cs->cr,
4737 		    NULL, NULL, NULL);
4738 		kmem_free(nnm, nlen);
4739 		if (!error) {
4740 			add_volrnm_fh(cs->exi, vp);
4741 			VN_RELE(vp);
4742 		}
4743 	}
4744 	}
4745 #endif	/* VOLATILE_FH_TEST */
4746 
4747 	*cs->statusp = resp->status = NFS4_OK;
4748 out:
4749 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4750 	    RENAME4res *, resp);
4751 	return;
4752 
4753 err_out:
4754 	if (onm != converted_onm)
4755 		kmem_free(converted_onm, MAXPATHLEN + 1);
4756 	if (onm != NULL)
4757 		kmem_free(onm, olen);
4758 	if (nnm != converted_nnm)
4759 		kmem_free(converted_nnm, MAXPATHLEN + 1);
4760 	if (nnm != NULL)
4761 		kmem_free(nnm, nlen);
4762 
4763 	if (in_crit_src) nbl_end_crit(srcvp);
4764 	if (in_crit_targ) nbl_end_crit(targvp);
4765 	if (targvp) VN_RELE(targvp);
4766 	if (srcvp) VN_RELE(srcvp);
4767 	if (sfp) {
4768 		if (sfp_rele_grant_hold) rfs4_clear_dont_grant(sfp);
4769 		rfs4_file_rele(sfp);
4770 	}
4771 	if (fp) {
4772 		if (fp_rele_grant_hold) rfs4_clear_dont_grant(fp);
4773 		rfs4_file_rele(fp);
4774 	}
4775 
4776 	DTRACE_NFSV4_2(op__rename__done, struct compound_state *, cs,
4777 	    RENAME4res *, resp);
4778 }
4779 
4780 /* ARGSUSED */
4781 static void
4782 rfs4_op_renew(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4783     struct compound_state *cs)
4784 {
4785 	RENEW4args *args = &argop->nfs_argop4_u.oprenew;
4786 	RENEW4res *resp = &resop->nfs_resop4_u.oprenew;
4787 	rfs4_client_t *cp;
4788 
4789 	DTRACE_NFSV4_2(op__renew__start, struct compound_state *, cs,
4790 	    RENEW4args *, args);
4791 
4792 	if ((cp = rfs4_findclient_by_id(args->clientid, FALSE)) == NULL) {
4793 		*cs->statusp = resp->status =
4794 		    rfs4_check_clientid(&args->clientid, 0);
4795 		goto out;
4796 	}
4797 
4798 	if (rfs4_lease_expired(cp)) {
4799 		rfs4_client_rele(cp);
4800 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
4801 		goto out;
4802 	}
4803 
4804 	rfs4_update_lease(cp);
4805 
4806 	mutex_enter(cp->rc_cbinfo.cb_lock);
4807 	if (cp->rc_cbinfo.cb_notified_of_cb_path_down == FALSE) {
4808 		cp->rc_cbinfo.cb_notified_of_cb_path_down = TRUE;
4809 		*cs->statusp = resp->status = NFS4ERR_CB_PATH_DOWN;
4810 	} else {
4811 		*cs->statusp = resp->status = NFS4_OK;
4812 	}
4813 	mutex_exit(cp->rc_cbinfo.cb_lock);
4814 
4815 	rfs4_client_rele(cp);
4816 
4817 out:
4818 	DTRACE_NFSV4_2(op__renew__done, struct compound_state *, cs,
4819 	    RENEW4res *, resp);
4820 }
4821 
4822 /* ARGSUSED */
4823 static void
4824 rfs4_op_restorefh(nfs_argop4 *args, nfs_resop4 *resop, struct svc_req *req,
4825     struct compound_state *cs)
4826 {
4827 	RESTOREFH4res *resp = &resop->nfs_resop4_u.oprestorefh;
4828 
4829 	DTRACE_NFSV4_1(op__restorefh__start, struct compound_state *, cs);
4830 
4831 	/* No need to check cs->access - we are not accessing any object */
4832 	if ((cs->saved_vp == NULL) || (cs->saved_fh.nfs_fh4_val == NULL)) {
4833 		*cs->statusp = resp->status = NFS4ERR_RESTOREFH;
4834 		goto out;
4835 	}
4836 	if (cs->vp != NULL) {
4837 		VN_RELE(cs->vp);
4838 	}
4839 	cs->vp = cs->saved_vp;
4840 	cs->saved_vp = NULL;
4841 	cs->exi = cs->saved_exi;
4842 	nfs_fh4_copy(&cs->saved_fh, &cs->fh);
4843 	*cs->statusp = resp->status = NFS4_OK;
4844 	cs->deleg = FALSE;
4845 
4846 out:
4847 	DTRACE_NFSV4_2(op__restorefh__done, struct compound_state *, cs,
4848 	    RESTOREFH4res *, resp);
4849 }
4850 
4851 /* ARGSUSED */
4852 static void
4853 rfs4_op_savefh(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
4854     struct compound_state *cs)
4855 {
4856 	SAVEFH4res *resp = &resop->nfs_resop4_u.opsavefh;
4857 
4858 	DTRACE_NFSV4_1(op__savefh__start, struct compound_state *, cs);
4859 
4860 	/* No need to check cs->access - we are not accessing any object */
4861 	if (cs->vp == NULL) {
4862 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
4863 		goto out;
4864 	}
4865 	if (cs->saved_vp != NULL) {
4866 		VN_RELE(cs->saved_vp);
4867 	}
4868 	cs->saved_vp = cs->vp;
4869 	VN_HOLD(cs->saved_vp);
4870 	cs->saved_exi = cs->exi;
4871 	/*
4872 	 * since SAVEFH is fairly rare, don't alloc space for its fh
4873 	 * unless necessary.
4874 	 */
4875 	if (cs->saved_fh.nfs_fh4_val == NULL) {
4876 		cs->saved_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP);
4877 	}
4878 	nfs_fh4_copy(&cs->fh, &cs->saved_fh);
4879 	*cs->statusp = resp->status = NFS4_OK;
4880 
4881 out:
4882 	DTRACE_NFSV4_2(op__savefh__done, struct compound_state *, cs,
4883 	    SAVEFH4res *, resp);
4884 }
4885 
4886 /*
4887  * rfs4_verify_attr is called when nfsv4 Setattr failed, but we wish to
4888  * return the bitmap of attrs that were set successfully. It is also
4889  * called by Verify/Nverify to test the vattr/vfsstat attrs. It should
4890  * always be called only after rfs4_do_set_attrs().
4891  *
4892  * Verify that the attributes are same as the expected ones. sargp->vap
4893  * and sargp->sbp contain the input attributes as translated from fattr4.
4894  *
4895  * This function verifies only the attrs that correspond to a vattr or
4896  * vfsstat struct. That is because of the extra step needed to get the
4897  * corresponding system structs. Other attributes have already been set or
4898  * verified by do_rfs4_set_attrs.
4899  *
4900  * Return 0 if all attrs match, -1 if some don't, error if error processing.
4901  */
4902 static int
4903 rfs4_verify_attr(struct nfs4_svgetit_arg *sargp,
4904     bitmap4 *resp, struct nfs4_ntov_table *ntovp)
4905 {
4906 	int error, ret_error = 0;
4907 	int i, k;
4908 	uint_t sva_mask = sargp->vap->va_mask;
4909 	uint_t vbit;
4910 	union nfs4_attr_u *na;
4911 	uint8_t *amap;
4912 	bool_t getsb = ntovp->vfsstat;
4913 
4914 	if (sva_mask != 0) {
4915 		/*
4916 		 * Okay to overwrite sargp->vap because we verify based
4917 		 * on the incoming values.
4918 		 */
4919 		ret_error = VOP_GETATTR(sargp->cs->vp, sargp->vap, 0,
4920 		    sargp->cs->cr, NULL);
4921 		if (ret_error) {
4922 			if (resp == NULL)
4923 				return (ret_error);
4924 			/*
4925 			 * Must return bitmap of successful attrs
4926 			 */
4927 			sva_mask = 0;	/* to prevent checking vap later */
4928 		} else {
4929 			/*
4930 			 * Some file systems clobber va_mask. it is probably
4931 			 * wrong of them to do so, nonethless we practice
4932 			 * defensive coding.
4933 			 * See bug id 4276830.
4934 			 */
4935 			sargp->vap->va_mask = sva_mask;
4936 		}
4937 	}
4938 
4939 	if (getsb) {
4940 		/*
4941 		 * Now get the superblock and loop on the bitmap, as there is
4942 		 * no simple way of translating from superblock to bitmap4.
4943 		 */
4944 		ret_error = VFS_STATVFS(sargp->cs->vp->v_vfsp, sargp->sbp);
4945 		if (ret_error) {
4946 			if (resp == NULL)
4947 				goto errout;
4948 			getsb = FALSE;
4949 		}
4950 	}
4951 
4952 	/*
4953 	 * Now loop and verify each attribute which getattr returned
4954 	 * whether it's the same as the input.
4955 	 */
4956 	if (resp == NULL && !getsb && (sva_mask == 0))
4957 		goto errout;
4958 
4959 	na = ntovp->na;
4960 	amap = ntovp->amap;
4961 	k = 0;
4962 	for (i = 0; i < ntovp->attrcnt; i++, na++, amap++) {
4963 		k = *amap;
4964 		ASSERT(nfs4_ntov_map[k].nval == k);
4965 		vbit = nfs4_ntov_map[k].vbit;
4966 
4967 		/*
4968 		 * If vattr attribute but VOP_GETATTR failed, or it's
4969 		 * superblock attribute but VFS_STATVFS failed, skip
4970 		 */
4971 		if (vbit) {
4972 			if ((vbit & sva_mask) == 0)
4973 				continue;
4974 		} else if (!(getsb && nfs4_ntov_map[k].vfsstat)) {
4975 			continue;
4976 		}
4977 		error = (*nfs4_ntov_map[k].sv_getit)(NFS4ATTR_VERIT, sargp, na);
4978 		if (resp != NULL) {
4979 			if (error)
4980 				ret_error = -1;	/* not all match */
4981 			else	/* update response bitmap */
4982 				*resp |= nfs4_ntov_map[k].fbit;
4983 			continue;
4984 		}
4985 		if (error) {
4986 			ret_error = -1;	/* not all match */
4987 			break;
4988 		}
4989 	}
4990 errout:
4991 	return (ret_error);
4992 }
4993 
4994 /*
4995  * Decode the attribute to be set/verified. If the attr requires a sys op
4996  * (VOP_GETATTR, VFS_VFSSTAT), and the request is to verify, then don't
4997  * call the sv_getit function for it, because the sys op hasn't yet been done.
4998  * Return 0 for success, error code if failed.
4999  *
5000  * Note: the decoded arg is not freed here but in nfs4_ntov_table_free.
5001  */
5002 static int
5003 decode_fattr4_attr(nfs4_attr_cmd_t cmd, struct nfs4_svgetit_arg *sargp,
5004     int k, XDR *xdrp, bitmap4 *resp_bval, union nfs4_attr_u *nap)
5005 {
5006 	int error = 0;
5007 	bool_t set_later;
5008 
5009 	sargp->vap->va_mask |= nfs4_ntov_map[k].vbit;
5010 
5011 	if ((*nfs4_ntov_map[k].xfunc)(xdrp, nap)) {
5012 		set_later = nfs4_ntov_map[k].vbit || nfs4_ntov_map[k].vfsstat;
5013 		/*
5014 		 * don't verify yet if a vattr or sb dependent attr,
5015 		 * because we don't have their sys values yet.
5016 		 * Will be done later.
5017 		 */
5018 		if (! (set_later && (cmd == NFS4ATTR_VERIT))) {
5019 			/*
5020 			 * ACLs are a special case, since setting the MODE
5021 			 * conflicts with setting the ACL.  We delay setting
5022 			 * the ACL until all other attributes have been set.
5023 			 * The ACL gets set in do_rfs4_op_setattr().
5024 			 */
5025 			if (nfs4_ntov_map[k].fbit != FATTR4_ACL_MASK) {
5026 				error = (*nfs4_ntov_map[k].sv_getit)(cmd,
5027 				    sargp, nap);
5028 				if (error) {
5029 					xdr_free(nfs4_ntov_map[k].xfunc,
5030 					    (caddr_t)nap);
5031 				}
5032 			}
5033 		}
5034 	} else {
5035 #ifdef  DEBUG
5036 		cmn_err(CE_NOTE, "decode_fattr4_attr: error "
5037 		    "decoding attribute %d\n", k);
5038 #endif
5039 		error = EINVAL;
5040 	}
5041 	if (!error && resp_bval && !set_later) {
5042 		*resp_bval |= nfs4_ntov_map[k].fbit;
5043 	}
5044 
5045 	return (error);
5046 }
5047 
5048 /*
5049  * Set vattr based on incoming fattr4 attrs - used by setattr.
5050  * Set response mask. Ignore any values that are not writable vattr attrs.
5051  */
5052 static nfsstat4
5053 do_rfs4_set_attrs(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5054     struct nfs4_svgetit_arg *sargp, struct nfs4_ntov_table *ntovp,
5055     nfs4_attr_cmd_t cmd)
5056 {
5057 	int error = 0;
5058 	int i;
5059 	char *attrs = fattrp->attrlist4;
5060 	uint32_t attrslen = fattrp->attrlist4_len;
5061 	XDR xdr;
5062 	nfsstat4 status = NFS4_OK;
5063 	vnode_t *vp = cs->vp;
5064 	union nfs4_attr_u *na;
5065 	uint8_t *amap;
5066 
5067 #ifndef lint
5068 	/*
5069 	 * Make sure that maximum attribute number can be expressed as an
5070 	 * 8 bit quantity.
5071 	 */
5072 	ASSERT(NFS4_MAXNUM_ATTRS <= (UINT8_MAX + 1));
5073 #endif
5074 
5075 	if (vp == NULL) {
5076 		if (resp)
5077 			*resp = 0;
5078 		return (NFS4ERR_NOFILEHANDLE);
5079 	}
5080 	if (cs->access == CS_ACCESS_DENIED) {
5081 		if (resp)
5082 			*resp = 0;
5083 		return (NFS4ERR_ACCESS);
5084 	}
5085 
5086 	sargp->op = cmd;
5087 	sargp->cs = cs;
5088 	sargp->flag = 0;	/* may be set later */
5089 	sargp->vap->va_mask = 0;
5090 	sargp->rdattr_error = NFS4_OK;
5091 	sargp->rdattr_error_req = FALSE;
5092 	/* sargp->sbp is set by the caller */
5093 
5094 	xdrmem_create(&xdr, attrs, attrslen, XDR_DECODE);
5095 
5096 	na = ntovp->na;
5097 	amap = ntovp->amap;
5098 
5099 	/*
5100 	 * The following loop iterates on the nfs4_ntov_map checking
5101 	 * if the fbit is set in the requested bitmap.
5102 	 * If set then we process the arguments using the
5103 	 * rfs4_fattr4 conversion functions to populate the setattr
5104 	 * vattr and va_mask. Any settable attrs that are not using vattr
5105 	 * will be set in this loop.
5106 	 */
5107 	for (i = 0; i < nfs4_ntov_map_size; i++) {
5108 		if (!(fattrp->attrmask & nfs4_ntov_map[i].fbit)) {
5109 			continue;
5110 		}
5111 		/*
5112 		 * If setattr, must be a writable attr.
5113 		 * If verify/nverify, must be a readable attr.
5114 		 */
5115 		if ((error = (*nfs4_ntov_map[i].sv_getit)(
5116 		    NFS4ATTR_SUPPORTED, sargp, NULL)) != 0) {
5117 			/*
5118 			 * Client tries to set/verify an
5119 			 * unsupported attribute, tries to set
5120 			 * a read only attr or verify a write
5121 			 * only one - error!
5122 			 */
5123 			break;
5124 		}
5125 		/*
5126 		 * Decode the attribute to set/verify
5127 		 */
5128 		error = decode_fattr4_attr(cmd, sargp, nfs4_ntov_map[i].nval,
5129 		    &xdr, resp ? resp : NULL, na);
5130 		if (error)
5131 			break;
5132 		*amap++ = (uint8_t)nfs4_ntov_map[i].nval;
5133 		na++;
5134 		(ntovp->attrcnt)++;
5135 		if (nfs4_ntov_map[i].vfsstat)
5136 			ntovp->vfsstat = TRUE;
5137 	}
5138 
5139 	if (error != 0)
5140 		status = (error == ENOTSUP ? NFS4ERR_ATTRNOTSUPP :
5141 		    puterrno4(error));
5142 	/* xdrmem_destroy(&xdrs); */	/* NO-OP */
5143 	return (status);
5144 }
5145 
5146 static nfsstat4
5147 do_rfs4_op_setattr(bitmap4 *resp, fattr4 *fattrp, struct compound_state *cs,
5148     stateid4 *stateid)
5149 {
5150 	int error = 0;
5151 	struct nfs4_svgetit_arg sarg;
5152 	bool_t trunc;
5153 
5154 	nfsstat4 status = NFS4_OK;
5155 	cred_t *cr = cs->cr;
5156 	vnode_t *vp = cs->vp;
5157 	struct nfs4_ntov_table ntov;
5158 	struct statvfs64 sb;
5159 	struct vattr bva;
5160 	struct flock64 bf;
5161 	int in_crit = 0;
5162 	uint_t saved_mask = 0;
5163 	caller_context_t ct;
5164 
5165 	*resp = 0;
5166 	sarg.sbp = &sb;
5167 	sarg.is_referral = B_FALSE;
5168 	nfs4_ntov_table_init(&ntov);
5169 	status = do_rfs4_set_attrs(resp, fattrp, cs, &sarg, &ntov,
5170 	    NFS4ATTR_SETIT);
5171 	if (status != NFS4_OK) {
5172 		/*
5173 		 * failed set attrs
5174 		 */
5175 		goto done;
5176 	}
5177 	if ((sarg.vap->va_mask == 0) &&
5178 	    (! (fattrp->attrmask & FATTR4_ACL_MASK))) {
5179 		/*
5180 		 * no further work to be done
5181 		 */
5182 		goto done;
5183 	}
5184 
5185 	/*
5186 	 * If we got a request to set the ACL and the MODE, only
5187 	 * allow changing VSUID, VSGID, and VSVTX.  Attempting
5188 	 * to change any other bits, along with setting an ACL,
5189 	 * gives NFS4ERR_INVAL.
5190 	 */
5191 	if ((fattrp->attrmask & FATTR4_ACL_MASK) &&
5192 	    (fattrp->attrmask & FATTR4_MODE_MASK)) {
5193 		vattr_t va;
5194 
5195 		va.va_mask = AT_MODE;
5196 		error = VOP_GETATTR(vp, &va, 0, cs->cr, NULL);
5197 		if (error) {
5198 			status = puterrno4(error);
5199 			goto done;
5200 		}
5201 		if ((sarg.vap->va_mode ^ va.va_mode) &
5202 		    ~(VSUID | VSGID | VSVTX)) {
5203 			status = NFS4ERR_INVAL;
5204 			goto done;
5205 		}
5206 	}
5207 
5208 	/* Check stateid only if size has been set */
5209 	if (sarg.vap->va_mask & AT_SIZE) {
5210 		trunc = (sarg.vap->va_size == 0);
5211 		status = rfs4_check_stateid(FWRITE, cs->vp, stateid,
5212 		    trunc, &cs->deleg, sarg.vap->va_mask & AT_SIZE, &ct);
5213 		if (status != NFS4_OK)
5214 			goto done;
5215 	} else {
5216 		ct.cc_sysid = 0;
5217 		ct.cc_pid = 0;
5218 		ct.cc_caller_id = nfs4_srv_caller_id;
5219 		ct.cc_flags = CC_DONTBLOCK;
5220 	}
5221 
5222 	/* XXX start of possible race with delegations */
5223 
5224 	/*
5225 	 * We need to specially handle size changes because it is
5226 	 * possible for the client to create a file with read-only
5227 	 * modes, but with the file opened for writing. If the client
5228 	 * then tries to set the file size, e.g. ftruncate(3C),
5229 	 * fcntl(F_FREESP), the normal access checking done in
5230 	 * VOP_SETATTR would prevent the client from doing it even though
5231 	 * it should be allowed to do so.  To get around this, we do the
5232 	 * access checking for ourselves and use VOP_SPACE which doesn't
5233 	 * do the access checking.
5234 	 * Also the client should not be allowed to change the file
5235 	 * size if there is a conflicting non-blocking mandatory lock in
5236 	 * the region of the change.
5237 	 */
5238 	if (vp->v_type == VREG && (sarg.vap->va_mask & AT_SIZE)) {
5239 		u_offset_t offset;
5240 		ssize_t length;
5241 
5242 		/*
5243 		 * ufs_setattr clears AT_SIZE from vap->va_mask, but
5244 		 * before returning, sarg.vap->va_mask is used to
5245 		 * generate the setattr reply bitmap.  We also clear
5246 		 * AT_SIZE below before calling VOP_SPACE.  For both
5247 		 * of these cases, the va_mask needs to be saved here
5248 		 * and restored after calling VOP_SETATTR.
5249 		 */
5250 		saved_mask = sarg.vap->va_mask;
5251 
5252 		/*
5253 		 * Check any possible conflict due to NBMAND locks.
5254 		 * Get into critical region before VOP_GETATTR, so the
5255 		 * size attribute is valid when checking conflicts.
5256 		 */
5257 		if (nbl_need_check(vp)) {
5258 			nbl_start_crit(vp, RW_READER);
5259 			in_crit = 1;
5260 		}
5261 
5262 		bva.va_mask = AT_UID|AT_SIZE;
5263 		if (error = VOP_GETATTR(vp, &bva, 0, cr, &ct)) {
5264 			status = puterrno4(error);
5265 			goto done;
5266 		}
5267 
5268 		if (in_crit) {
5269 			if (sarg.vap->va_size < bva.va_size) {
5270 				offset = sarg.vap->va_size;
5271 				length = bva.va_size - sarg.vap->va_size;
5272 			} else {
5273 				offset = bva.va_size;
5274 				length = sarg.vap->va_size - bva.va_size;
5275 			}
5276 			if (nbl_conflict(vp, NBL_WRITE, offset, length, 0,
5277 			    &ct)) {
5278 				status = NFS4ERR_LOCKED;
5279 				goto done;
5280 			}
5281 		}
5282 
5283 		if (crgetuid(cr) == bva.va_uid) {
5284 			sarg.vap->va_mask &= ~AT_SIZE;
5285 			bf.l_type = F_WRLCK;
5286 			bf.l_whence = 0;
5287 			bf.l_start = (off64_t)sarg.vap->va_size;
5288 			bf.l_len = 0;
5289 			bf.l_sysid = 0;
5290 			bf.l_pid = 0;
5291 			error = VOP_SPACE(vp, F_FREESP, &bf, FWRITE,
5292 			    (offset_t)sarg.vap->va_size, cr, &ct);
5293 		}
5294 	}
5295 
5296 	if (!error && sarg.vap->va_mask != 0)
5297 		error = VOP_SETATTR(vp, sarg.vap, sarg.flag, cr, &ct);
5298 
5299 	/* restore va_mask -- ufs_setattr clears AT_SIZE */
5300 	if (saved_mask & AT_SIZE)
5301 		sarg.vap->va_mask |= AT_SIZE;
5302 
5303 	/*
5304 	 * If an ACL was being set, it has been delayed until now,
5305 	 * in order to set the mode (via the VOP_SETATTR() above) first.
5306 	 */
5307 	if ((! error) && (fattrp->attrmask & FATTR4_ACL_MASK)) {
5308 		int i;
5309 
5310 		for (i = 0; i < NFS4_MAXNUM_ATTRS; i++)
5311 			if (ntov.amap[i] == FATTR4_ACL)
5312 				break;
5313 		if (i < NFS4_MAXNUM_ATTRS) {
5314 			error = (*nfs4_ntov_map[FATTR4_ACL].sv_getit)(
5315 			    NFS4ATTR_SETIT, &sarg, &ntov.na[i]);
5316 			if (error == 0) {
5317 				*resp |= FATTR4_ACL_MASK;
5318 			} else if (error == ENOTSUP) {
5319 				(void) rfs4_verify_attr(&sarg, resp, &ntov);
5320 				status = NFS4ERR_ATTRNOTSUPP;
5321 				goto done;
5322 			}
5323 		} else {
5324 			NFS4_DEBUG(rfs4_debug,
5325 			    (CE_NOTE, "do_rfs4_op_setattr: "
5326 			    "unable to find ACL in fattr4"));
5327 			error = EINVAL;
5328 		}
5329 	}
5330 
5331 	if (error) {
5332 		/* check if a monitor detected a delegation conflict */
5333 		if (error == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
5334 			status = NFS4ERR_DELAY;
5335 		else
5336 			status = puterrno4(error);
5337 
5338 		/*
5339 		 * Set the response bitmap when setattr failed.
5340 		 * If VOP_SETATTR partially succeeded, test by doing a
5341 		 * VOP_GETATTR on the object and comparing the data
5342 		 * to the setattr arguments.
5343 		 */
5344 		(void) rfs4_verify_attr(&sarg, resp, &ntov);
5345 	} else {
5346 		/*
5347 		 * Force modified metadata out to stable storage.
5348 		 */
5349 		(void) VOP_FSYNC(vp, FNODSYNC, cr, &ct);
5350 		/*
5351 		 * Set response bitmap
5352 		 */
5353 		nfs4_vmask_to_nmask_set(sarg.vap->va_mask, resp);
5354 	}
5355 
5356 /* Return early and already have a NFSv4 error */
5357 done:
5358 	/*
5359 	 * Except for nfs4_vmask_to_nmask_set(), vattr --> fattr
5360 	 * conversion sets both readable and writeable NFS4 attrs
5361 	 * for AT_MTIME and AT_ATIME.  The line below masks out
5362 	 * unrequested attrs from the setattr result bitmap.  This
5363 	 * is placed after the done: label to catch the ATTRNOTSUP
5364 	 * case.
5365 	 */
5366 	*resp &= fattrp->attrmask;
5367 
5368 	if (in_crit)
5369 		nbl_end_crit(vp);
5370 
5371 	nfs4_ntov_table_free(&ntov, &sarg);
5372 
5373 	return (status);
5374 }
5375 
5376 /* ARGSUSED */
5377 static void
5378 rfs4_op_setattr(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5379     struct compound_state *cs)
5380 {
5381 	SETATTR4args *args = &argop->nfs_argop4_u.opsetattr;
5382 	SETATTR4res *resp = &resop->nfs_resop4_u.opsetattr;
5383 	bslabel_t *clabel;
5384 
5385 	DTRACE_NFSV4_2(op__setattr__start, struct compound_state *, cs,
5386 	    SETATTR4args *, args);
5387 
5388 	if (cs->vp == NULL) {
5389 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5390 		goto out;
5391 	}
5392 
5393 	/*
5394 	 * If there is an unshared filesystem mounted on this vnode,
5395 	 * do not allow to setattr on this vnode.
5396 	 */
5397 	if (vn_ismntpt(cs->vp)) {
5398 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5399 		goto out;
5400 	}
5401 
5402 	resp->attrsset = 0;
5403 
5404 	if (rdonly4(cs->exi, cs->vp, req)) {
5405 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5406 		goto out;
5407 	}
5408 
5409 	/* check label before setting attributes */
5410 	if (is_system_labeled()) {
5411 		ASSERT(req->rq_label != NULL);
5412 		clabel = req->rq_label;
5413 		DTRACE_PROBE2(tx__rfs4__log__info__opsetattr__clabel, char *,
5414 		    "got client label from request(1)",
5415 		    struct svc_req *, req);
5416 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
5417 			if (!do_rfs_label_check(clabel, cs->vp,
5418 			    EQUALITY_CHECK, cs->exi)) {
5419 				*cs->statusp = resp->status = NFS4ERR_ACCESS;
5420 				goto out;
5421 			}
5422 		}
5423 	}
5424 
5425 	*cs->statusp = resp->status =
5426 	    do_rfs4_op_setattr(&resp->attrsset, &args->obj_attributes, cs,
5427 	    &args->stateid);
5428 
5429 out:
5430 	DTRACE_NFSV4_2(op__setattr__done, struct compound_state *, cs,
5431 	    SETATTR4res *, resp);
5432 }
5433 
5434 /* ARGSUSED */
5435 static void
5436 rfs4_op_verify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5437     struct compound_state *cs)
5438 {
5439 	/*
5440 	 * verify and nverify are exactly the same, except that nverify
5441 	 * succeeds when some argument changed, and verify succeeds when
5442 	 * when none changed.
5443 	 */
5444 
5445 	VERIFY4args  *args = &argop->nfs_argop4_u.opverify;
5446 	VERIFY4res *resp = &resop->nfs_resop4_u.opverify;
5447 
5448 	int error;
5449 	struct nfs4_svgetit_arg sarg;
5450 	struct statvfs64 sb;
5451 	struct nfs4_ntov_table ntov;
5452 
5453 	DTRACE_NFSV4_2(op__verify__start, struct compound_state *, cs,
5454 	    VERIFY4args *, args);
5455 
5456 	if (cs->vp == NULL) {
5457 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5458 		goto out;
5459 	}
5460 
5461 	sarg.sbp = &sb;
5462 	sarg.is_referral = B_FALSE;
5463 	nfs4_ntov_table_init(&ntov);
5464 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5465 	    &sarg, &ntov, NFS4ATTR_VERIT);
5466 	if (resp->status != NFS4_OK) {
5467 		/*
5468 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5469 		 * so could return -1 for "no match".
5470 		 */
5471 		if (resp->status == -1)
5472 			resp->status = NFS4ERR_NOT_SAME;
5473 		goto done;
5474 	}
5475 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5476 	switch (error) {
5477 	case 0:
5478 		resp->status = NFS4_OK;
5479 		break;
5480 	case -1:
5481 		resp->status = NFS4ERR_NOT_SAME;
5482 		break;
5483 	default:
5484 		resp->status = puterrno4(error);
5485 		break;
5486 	}
5487 done:
5488 	*cs->statusp = resp->status;
5489 	nfs4_ntov_table_free(&ntov, &sarg);
5490 out:
5491 	DTRACE_NFSV4_2(op__verify__done, struct compound_state *, cs,
5492 	    VERIFY4res *, resp);
5493 }
5494 
5495 /* ARGSUSED */
5496 static void
5497 rfs4_op_nverify(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5498     struct compound_state *cs)
5499 {
5500 	/*
5501 	 * verify and nverify are exactly the same, except that nverify
5502 	 * succeeds when some argument changed, and verify succeeds when
5503 	 * when none changed.
5504 	 */
5505 
5506 	NVERIFY4args  *args = &argop->nfs_argop4_u.opnverify;
5507 	NVERIFY4res *resp = &resop->nfs_resop4_u.opnverify;
5508 
5509 	int error;
5510 	struct nfs4_svgetit_arg sarg;
5511 	struct statvfs64 sb;
5512 	struct nfs4_ntov_table ntov;
5513 
5514 	DTRACE_NFSV4_2(op__nverify__start, struct compound_state *, cs,
5515 	    NVERIFY4args *, args);
5516 
5517 	if (cs->vp == NULL) {
5518 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5519 		DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5520 		    NVERIFY4res *, resp);
5521 		return;
5522 	}
5523 	sarg.sbp = &sb;
5524 	sarg.is_referral = B_FALSE;
5525 	nfs4_ntov_table_init(&ntov);
5526 	resp->status = do_rfs4_set_attrs(NULL, &args->obj_attributes, cs,
5527 	    &sarg, &ntov, NFS4ATTR_VERIT);
5528 	if (resp->status != NFS4_OK) {
5529 		/*
5530 		 * do_rfs4_set_attrs will try to verify systemwide attrs,
5531 		 * so could return -1 for "no match".
5532 		 */
5533 		if (resp->status == -1)
5534 			resp->status = NFS4_OK;
5535 		goto done;
5536 	}
5537 	error = rfs4_verify_attr(&sarg, NULL, &ntov);
5538 	switch (error) {
5539 	case 0:
5540 		resp->status = NFS4ERR_SAME;
5541 		break;
5542 	case -1:
5543 		resp->status = NFS4_OK;
5544 		break;
5545 	default:
5546 		resp->status = puterrno4(error);
5547 		break;
5548 	}
5549 done:
5550 	*cs->statusp = resp->status;
5551 	nfs4_ntov_table_free(&ntov, &sarg);
5552 
5553 	DTRACE_NFSV4_2(op__nverify__done, struct compound_state *, cs,
5554 	    NVERIFY4res *, resp);
5555 }
5556 
5557 /*
5558  * XXX - This should live in an NFS header file.
5559  */
5560 #define	MAX_IOVECS	12
5561 
5562 /* ARGSUSED */
5563 static void
5564 rfs4_op_write(nfs_argop4 *argop, nfs_resop4 *resop, struct svc_req *req,
5565     struct compound_state *cs)
5566 {
5567 	WRITE4args *args = &argop->nfs_argop4_u.opwrite;
5568 	WRITE4res *resp = &resop->nfs_resop4_u.opwrite;
5569 	int error;
5570 	vnode_t *vp;
5571 	struct vattr bva;
5572 	u_offset_t rlimit;
5573 	struct uio uio;
5574 	struct iovec iov[MAX_IOVECS];
5575 	struct iovec *iovp;
5576 	int iovcnt;
5577 	int ioflag;
5578 	cred_t *savecred, *cr;
5579 	bool_t *deleg = &cs->deleg;
5580 	nfsstat4 stat;
5581 	int in_crit = 0;
5582 	caller_context_t ct;
5583 
5584 	DTRACE_NFSV4_2(op__write__start, struct compound_state *, cs,
5585 	    WRITE4args *, args);
5586 
5587 	vp = cs->vp;
5588 	if (vp == NULL) {
5589 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
5590 		goto out;
5591 	}
5592 	if (cs->access == CS_ACCESS_DENIED) {
5593 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5594 		goto out;
5595 	}
5596 
5597 	cr = cs->cr;
5598 
5599 	if ((stat = rfs4_check_stateid(FWRITE, vp, &args->stateid, FALSE,
5600 	    deleg, TRUE, &ct)) != NFS4_OK) {
5601 		*cs->statusp = resp->status = stat;
5602 		goto out;
5603 	}
5604 
5605 	/*
5606 	 * We have to enter the critical region before calling VOP_RWLOCK
5607 	 * to avoid a deadlock with ufs.
5608 	 */
5609 	if (nbl_need_check(vp)) {
5610 		nbl_start_crit(vp, RW_READER);
5611 		in_crit = 1;
5612 		if (nbl_conflict(vp, NBL_WRITE,
5613 		    args->offset, args->data_len, 0, &ct)) {
5614 			*cs->statusp = resp->status = NFS4ERR_LOCKED;
5615 			goto out;
5616 		}
5617 	}
5618 
5619 	bva.va_mask = AT_MODE | AT_UID;
5620 	error = VOP_GETATTR(vp, &bva, 0, cr, &ct);
5621 
5622 	/*
5623 	 * If we can't get the attributes, then we can't do the
5624 	 * right access checking.  So, we'll fail the request.
5625 	 */
5626 	if (error) {
5627 		*cs->statusp = resp->status = puterrno4(error);
5628 		goto out;
5629 	}
5630 
5631 	if (rdonly4(cs->exi, cs->vp, req)) {
5632 		*cs->statusp = resp->status = NFS4ERR_ROFS;
5633 		goto out;
5634 	}
5635 
5636 	if (vp->v_type != VREG) {
5637 		*cs->statusp = resp->status =
5638 		    ((vp->v_type == VDIR) ? NFS4ERR_ISDIR : NFS4ERR_INVAL);
5639 		goto out;
5640 	}
5641 
5642 	if (crgetuid(cr) != bva.va_uid &&
5643 	    (error = VOP_ACCESS(vp, VWRITE, 0, cr, &ct))) {
5644 		*cs->statusp = resp->status = puterrno4(error);
5645 		goto out;
5646 	}
5647 
5648 	if (MANDLOCK(vp, bva.va_mode)) {
5649 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
5650 		goto out;
5651 	}
5652 
5653 	if (args->data_len == 0) {
5654 		*cs->statusp = resp->status = NFS4_OK;
5655 		resp->count = 0;
5656 		resp->committed = args->stable;
5657 		resp->writeverf = Write4verf;
5658 		goto out;
5659 	}
5660 
5661 	if (args->mblk != NULL) {
5662 		mblk_t *m;
5663 		uint_t bytes, round_len;
5664 
5665 		iovcnt = 0;
5666 		bytes = 0;
5667 		round_len = roundup(args->data_len, BYTES_PER_XDR_UNIT);
5668 		for (m = args->mblk;
5669 		    m != NULL && bytes < round_len;
5670 		    m = m->b_cont) {
5671 			iovcnt++;
5672 			bytes += MBLKL(m);
5673 		}
5674 #ifdef DEBUG
5675 		/* should have ended on an mblk boundary */
5676 		if (bytes != round_len) {
5677 			printf("bytes=0x%x, round_len=0x%x, req len=0x%x\n",
5678 			    bytes, round_len, args->data_len);
5679 			printf("args=%p, args->mblk=%p, m=%p", (void *)args,
5680 			    (void *)args->mblk, (void *)m);
5681 			ASSERT(bytes == round_len);
5682 		}
5683 #endif
5684 		if (iovcnt <= MAX_IOVECS) {
5685 			iovp = iov;
5686 		} else {
5687 			iovp = kmem_alloc(sizeof (*iovp) * iovcnt, KM_SLEEP);
5688 		}
5689 		mblk_to_iov(args->mblk, iovcnt, iovp);
5690 	} else if (args->rlist != NULL) {
5691 		iovcnt = 1;
5692 		iovp = iov;
5693 		iovp->iov_base = (char *)((args->rlist)->u.c_daddr3);
5694 		iovp->iov_len = args->data_len;
5695 	} else {
5696 		iovcnt = 1;
5697 		iovp = iov;
5698 		iovp->iov_base = args->data_val;
5699 		iovp->iov_len = args->data_len;
5700 	}
5701 
5702 	uio.uio_iov = iovp;
5703 	uio.uio_iovcnt = iovcnt;
5704 
5705 	uio.uio_segflg = UIO_SYSSPACE;
5706 	uio.uio_extflg = UIO_COPY_DEFAULT;
5707 	uio.uio_loffset = args->offset;
5708 	uio.uio_resid = args->data_len;
5709 	uio.uio_llimit = curproc->p_fsz_ctl;
5710 	rlimit = uio.uio_llimit - args->offset;
5711 	if (rlimit < (u_offset_t)uio.uio_resid)
5712 		uio.uio_resid = (int)rlimit;
5713 
5714 	if (args->stable == UNSTABLE4)
5715 		ioflag = 0;
5716 	else if (args->stable == FILE_SYNC4)
5717 		ioflag = FSYNC;
5718 	else if (args->stable == DATA_SYNC4)
5719 		ioflag = FDSYNC;
5720 	else {
5721 		if (iovp != iov)
5722 			kmem_free(iovp, sizeof (*iovp) * iovcnt);
5723 		*cs->statusp = resp->status = NFS4ERR_INVAL;
5724 		goto out;
5725 	}
5726 
5727 	/*
5728 	 * We're changing creds because VM may fault and we need
5729 	 * the cred of the current thread to be used if quota
5730 	 * checking is enabled.
5731 	 */
5732 	savecred = curthread->t_cred;
5733 	curthread->t_cred = cr;
5734 	error = do_io(FWRITE, vp, &uio, ioflag, cr, &ct);
5735 	curthread->t_cred = savecred;
5736 
5737 	if (iovp != iov)
5738 		kmem_free(iovp, sizeof (*iovp) * iovcnt);
5739 
5740 	if (error) {
5741 		*cs->statusp = resp->status = puterrno4(error);
5742 		goto out;
5743 	}
5744 
5745 	*cs->statusp = resp->status = NFS4_OK;
5746 	resp->count = args->data_len - uio.uio_resid;
5747 
5748 	if (ioflag == 0)
5749 		resp->committed = UNSTABLE4;
5750 	else
5751 		resp->committed = FILE_SYNC4;
5752 
5753 	resp->writeverf = Write4verf;
5754 
5755 out:
5756 	if (in_crit)
5757 		nbl_end_crit(vp);
5758 
5759 	DTRACE_NFSV4_2(op__write__done, struct compound_state *, cs,
5760 	    WRITE4res *, resp);
5761 }
5762 
5763 
5764 /* XXX put in a header file */
5765 extern int	sec_svc_getcred(struct svc_req *, cred_t *,  caddr_t *, int *);
5766 
5767 void
5768 rfs4_compound(COMPOUND4args *args, COMPOUND4res *resp, struct exportinfo *exi,
5769     struct svc_req *req, cred_t *cr, int *rv)
5770 {
5771 	uint_t i;
5772 	struct compound_state cs;
5773 
5774 	if (rv != NULL)
5775 		*rv = 0;
5776 	rfs4_init_compound_state(&cs);
5777 	/*
5778 	 * Form a reply tag by copying over the reqeuest tag.
5779 	 */
5780 	resp->tag.utf8string_val =
5781 	    kmem_alloc(args->tag.utf8string_len, KM_SLEEP);
5782 	resp->tag.utf8string_len = args->tag.utf8string_len;
5783 	bcopy(args->tag.utf8string_val, resp->tag.utf8string_val,
5784 	    resp->tag.utf8string_len);
5785 
5786 	cs.statusp = &resp->status;
5787 	cs.req = req;
5788 
5789 	/*
5790 	 * XXX for now, minorversion should be zero
5791 	 */
5792 	if (args->minorversion != NFS4_MINORVERSION) {
5793 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5794 		    &cs, COMPOUND4args *, args);
5795 		resp->array_len = 0;
5796 		resp->array = NULL;
5797 		resp->status = NFS4ERR_MINOR_VERS_MISMATCH;
5798 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5799 		    &cs, COMPOUND4res *, resp);
5800 		return;
5801 	}
5802 
5803 	ASSERT(exi == NULL);
5804 	ASSERT(cr == NULL);
5805 
5806 	cr = crget();
5807 	ASSERT(cr != NULL);
5808 
5809 	if (sec_svc_getcred(req, cr, &cs.principal, &cs.nfsflavor) == 0) {
5810 		DTRACE_NFSV4_2(compound__start, struct compound_state *,
5811 		    &cs, COMPOUND4args *, args);
5812 		crfree(cr);
5813 		DTRACE_NFSV4_2(compound__done, struct compound_state *,
5814 		    &cs, COMPOUND4res *, resp);
5815 		svcerr_badcred(req->rq_xprt);
5816 		if (rv != NULL)
5817 			*rv = 1;
5818 		return;
5819 	}
5820 	resp->array_len = args->array_len;
5821 	resp->array = kmem_zalloc(args->array_len * sizeof (nfs_resop4),
5822 	    KM_SLEEP);
5823 
5824 	cs.basecr = cr;
5825 
5826 	DTRACE_NFSV4_2(compound__start, struct compound_state *, &cs,
5827 	    COMPOUND4args *, args);
5828 
5829 	/*
5830 	 * For now, NFS4 compound processing must be protected by
5831 	 * exported_lock because it can access more than one exportinfo
5832 	 * per compound and share/unshare can now change multiple
5833 	 * exinfo structs.  The NFS2/3 code only refs 1 exportinfo
5834 	 * per proc (excluding public exinfo), and exi_count design
5835 	 * is sufficient to protect concurrent execution of NFS2/3
5836 	 * ops along with unexport.  This lock will be removed as
5837 	 * part of the NFSv4 phase 2 namespace redesign work.
5838 	 */
5839 	rw_enter(&exported_lock, RW_READER);
5840 
5841 	/*
5842 	 * If this is the first compound we've seen, we need to start all
5843 	 * new instances' grace periods.
5844 	 */
5845 	if (rfs4_seen_first_compound == 0) {
5846 		rfs4_grace_start_new();
5847 		/*
5848 		 * This must be set after rfs4_grace_start_new(), otherwise
5849 		 * another thread could proceed past here before the former
5850 		 * is finished.
5851 		 */
5852 		rfs4_seen_first_compound = 1;
5853 	}
5854 
5855 	for (i = 0; i < args->array_len && cs.cont; i++) {
5856 		nfs_argop4 *argop;
5857 		nfs_resop4 *resop;
5858 		uint_t op;
5859 
5860 		argop = &args->array[i];
5861 		resop = &resp->array[i];
5862 		resop->resop = argop->argop;
5863 		op = (uint_t)resop->resop;
5864 
5865 		if (op < rfsv4disp_cnt) {
5866 			/*
5867 			 * Count the individual ops here; NULL and COMPOUND
5868 			 * are counted in common_dispatch()
5869 			 */
5870 			rfsproccnt_v4_ptr[op].value.ui64++;
5871 
5872 			NFS4_DEBUG(rfs4_debug > 1,
5873 			    (CE_NOTE, "Executing %s", rfs4_op_string[op]));
5874 			(*rfsv4disptab[op].dis_proc)(argop, resop, req, &cs);
5875 			NFS4_DEBUG(rfs4_debug > 1, (CE_NOTE, "%s returned %d",
5876 			    rfs4_op_string[op], *cs.statusp));
5877 			if (*cs.statusp != NFS4_OK)
5878 				cs.cont = FALSE;
5879 		} else {
5880 			/*
5881 			 * This is effectively dead code since XDR code
5882 			 * will have already returned BADXDR if op doesn't
5883 			 * decode to legal value.  This only done for a
5884 			 * day when XDR code doesn't verify v4 opcodes.
5885 			 */
5886 			op = OP_ILLEGAL;
5887 			rfsproccnt_v4_ptr[OP_ILLEGAL_IDX].value.ui64++;
5888 
5889 			rfs4_op_illegal(argop, resop, req, &cs);
5890 			cs.cont = FALSE;
5891 		}
5892 
5893 		/*
5894 		 * If not at last op, and if we are to stop, then
5895 		 * compact the results array.
5896 		 */
5897 		if ((i + 1) < args->array_len && !cs.cont) {
5898 			nfs_resop4 *new_res = kmem_alloc(
5899 			    (i+1) * sizeof (nfs_resop4), KM_SLEEP);
5900 			bcopy(resp->array,
5901 			    new_res, (i+1) * sizeof (nfs_resop4));
5902 			kmem_free(resp->array,
5903 			    args->array_len * sizeof (nfs_resop4));
5904 
5905 			resp->array_len =  i + 1;
5906 			resp->array = new_res;
5907 		}
5908 	}
5909 
5910 	rw_exit(&exported_lock);
5911 
5912 	DTRACE_NFSV4_2(compound__done, struct compound_state *, &cs,
5913 	    COMPOUND4res *, resp);
5914 
5915 	if (cs.vp)
5916 		VN_RELE(cs.vp);
5917 	if (cs.saved_vp)
5918 		VN_RELE(cs.saved_vp);
5919 	if (cs.saved_fh.nfs_fh4_val)
5920 		kmem_free(cs.saved_fh.nfs_fh4_val, NFS4_FHSIZE);
5921 
5922 	if (cs.basecr)
5923 		crfree(cs.basecr);
5924 	if (cs.cr)
5925 		crfree(cs.cr);
5926 	/*
5927 	 * done with this compound request, free the label
5928 	 */
5929 
5930 	if (req->rq_label != NULL) {
5931 		kmem_free(req->rq_label, sizeof (bslabel_t));
5932 		req->rq_label = NULL;
5933 	}
5934 }
5935 
5936 /*
5937  * XXX because of what appears to be duplicate calls to rfs4_compound_free
5938  * XXX zero out the tag and array values. Need to investigate why the
5939  * XXX calls occur, but at least prevent the panic for now.
5940  */
5941 void
5942 rfs4_compound_free(COMPOUND4res *resp)
5943 {
5944 	uint_t i;
5945 
5946 	if (resp->tag.utf8string_val) {
5947 		UTF8STRING_FREE(resp->tag)
5948 	}
5949 
5950 	for (i = 0; i < resp->array_len; i++) {
5951 		nfs_resop4 *resop;
5952 		uint_t op;
5953 
5954 		resop = &resp->array[i];
5955 		op = (uint_t)resop->resop;
5956 		if (op < rfsv4disp_cnt) {
5957 			(*rfsv4disptab[op].dis_resfree)(resop);
5958 		}
5959 	}
5960 	if (resp->array != NULL) {
5961 		kmem_free(resp->array, resp->array_len * sizeof (nfs_resop4));
5962 	}
5963 }
5964 
5965 /*
5966  * Process the value of the compound request rpc flags, as a bit-AND
5967  * of the individual per-op flags (idempotent, allowork, publicfh_ok)
5968  */
5969 void
5970 rfs4_compound_flagproc(COMPOUND4args *args, int *flagp)
5971 {
5972 	int i;
5973 	int flag = RPC_ALL;
5974 
5975 	for (i = 0; flag && i < args->array_len; i++) {
5976 		uint_t op;
5977 
5978 		op = (uint_t)args->array[i].argop;
5979 
5980 		if (op < rfsv4disp_cnt)
5981 			flag &= rfsv4disptab[op].dis_flags;
5982 		else
5983 			flag = 0;
5984 	}
5985 	*flagp = flag;
5986 }
5987 
5988 nfsstat4
5989 rfs4_client_sysid(rfs4_client_t *cp, sysid_t *sp)
5990 {
5991 	nfsstat4 e;
5992 
5993 	rfs4_dbe_lock(cp->rc_dbe);
5994 
5995 	if (cp->rc_sysidt != LM_NOSYSID) {
5996 		*sp = cp->rc_sysidt;
5997 		e = NFS4_OK;
5998 
5999 	} else if ((cp->rc_sysidt = lm_alloc_sysidt()) != LM_NOSYSID) {
6000 		*sp = cp->rc_sysidt;
6001 		e = NFS4_OK;
6002 
6003 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
6004 		    "rfs4_client_sysid: allocated 0x%x\n", *sp));
6005 	} else
6006 		e = NFS4ERR_DELAY;
6007 
6008 	rfs4_dbe_unlock(cp->rc_dbe);
6009 	return (e);
6010 }
6011 
6012 #if defined(DEBUG) && ! defined(lint)
6013 static void lock_print(char *str, int operation, struct flock64 *flk)
6014 {
6015 	char *op, *type;
6016 
6017 	switch (operation) {
6018 	case F_GETLK: op = "F_GETLK";
6019 		break;
6020 	case F_SETLK: op = "F_SETLK";
6021 		break;
6022 	case F_SETLK_NBMAND: op = "F_SETLK_NBMAND";
6023 		break;
6024 	default: op = "F_UNKNOWN";
6025 		break;
6026 	}
6027 	switch (flk->l_type) {
6028 	case F_UNLCK: type = "F_UNLCK";
6029 		break;
6030 	case F_RDLCK: type = "F_RDLCK";
6031 		break;
6032 	case F_WRLCK: type = "F_WRLCK";
6033 		break;
6034 	default: type = "F_UNKNOWN";
6035 		break;
6036 	}
6037 
6038 	ASSERT(flk->l_whence == 0);
6039 	cmn_err(CE_NOTE, "%s:  %s, type = %s, off = %llx len = %llx pid = %d",
6040 	    str, op, type, (longlong_t)flk->l_start,
6041 	    flk->l_len ? (longlong_t)flk->l_len : ~0LL, flk->l_pid);
6042 }
6043 
6044 #define	LOCK_PRINT(d, s, t, f) if (d) lock_print(s, t, f)
6045 #else
6046 #define	LOCK_PRINT(d, s, t, f)
6047 #endif
6048 
6049 /*ARGSUSED*/
6050 static bool_t
6051 creds_ok(cred_set_t cr_set, struct svc_req *req, struct compound_state *cs)
6052 {
6053 	return (TRUE);
6054 }
6055 
6056 /*
6057  * Look up the pathname using the vp in cs as the directory vnode.
6058  * cs->vp will be the vnode for the file on success
6059  */
6060 
6061 static nfsstat4
6062 rfs4_lookup(component4 *component, struct svc_req *req,
6063     struct compound_state *cs)
6064 {
6065 	char *nm;
6066 	uint32_t len;
6067 	nfsstat4 status;
6068 	struct sockaddr *ca;
6069 	char *name;
6070 
6071 	if (cs->vp == NULL) {
6072 		return (NFS4ERR_NOFILEHANDLE);
6073 	}
6074 	if (cs->vp->v_type != VDIR) {
6075 		return (NFS4ERR_NOTDIR);
6076 	}
6077 
6078 	if (!utf8_dir_verify(component))
6079 		return (NFS4ERR_INVAL);
6080 
6081 	nm = utf8_to_fn(component, &len, NULL);
6082 	if (nm == NULL) {
6083 		return (NFS4ERR_INVAL);
6084 	}
6085 
6086 	if (len > MAXNAMELEN) {
6087 		kmem_free(nm, len);
6088 		return (NFS4ERR_NAMETOOLONG);
6089 	}
6090 
6091 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6092 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6093 	    MAXPATHLEN + 1);
6094 
6095 	if (name == NULL) {
6096 		kmem_free(nm, len);
6097 		return (NFS4ERR_INVAL);
6098 	}
6099 
6100 	status = do_rfs4_op_lookup(name, req, cs);
6101 
6102 	if (name != nm)
6103 		kmem_free(name, MAXPATHLEN + 1);
6104 
6105 	kmem_free(nm, len);
6106 
6107 	return (status);
6108 }
6109 
6110 static nfsstat4
6111 rfs4_lookupfile(component4 *component, struct svc_req *req,
6112     struct compound_state *cs, uint32_t access, change_info4 *cinfo)
6113 {
6114 	nfsstat4 status;
6115 	vnode_t *dvp = cs->vp;
6116 	vattr_t bva, ava, fva;
6117 	int error;
6118 
6119 	/* Get "before" change value */
6120 	bva.va_mask = AT_CTIME|AT_SEQ;
6121 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6122 	if (error)
6123 		return (puterrno4(error));
6124 
6125 	/* rfs4_lookup may VN_RELE directory */
6126 	VN_HOLD(dvp);
6127 
6128 	status = rfs4_lookup(component, req, cs);
6129 	if (status != NFS4_OK) {
6130 		VN_RELE(dvp);
6131 		return (status);
6132 	}
6133 
6134 	/*
6135 	 * Get "after" change value, if it fails, simply return the
6136 	 * before value.
6137 	 */
6138 	ava.va_mask = AT_CTIME|AT_SEQ;
6139 	if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6140 		ava.va_ctime = bva.va_ctime;
6141 		ava.va_seq = 0;
6142 	}
6143 	VN_RELE(dvp);
6144 
6145 	/*
6146 	 * Validate the file is a file
6147 	 */
6148 	fva.va_mask = AT_TYPE|AT_MODE;
6149 	error = VOP_GETATTR(cs->vp, &fva, 0, cs->cr, NULL);
6150 	if (error)
6151 		return (puterrno4(error));
6152 
6153 	if (fva.va_type != VREG) {
6154 		if (fva.va_type == VDIR)
6155 			return (NFS4ERR_ISDIR);
6156 		if (fva.va_type == VLNK)
6157 			return (NFS4ERR_SYMLINK);
6158 		return (NFS4ERR_INVAL);
6159 	}
6160 
6161 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime);
6162 	NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6163 
6164 	/*
6165 	 * It is undefined if VOP_LOOKUP will change va_seq, so
6166 	 * cinfo.atomic = TRUE only if we have
6167 	 * non-zero va_seq's, and they have not changed.
6168 	 */
6169 	if (bva.va_seq && ava.va_seq && ava.va_seq == bva.va_seq)
6170 		cinfo->atomic = TRUE;
6171 	else
6172 		cinfo->atomic = FALSE;
6173 
6174 	/* Check for mandatory locking */
6175 	cs->mandlock = MANDLOCK(cs->vp, fva.va_mode);
6176 	return (check_open_access(access, cs, req));
6177 }
6178 
6179 static nfsstat4
6180 create_vnode(vnode_t *dvp, char *nm,  vattr_t *vap, createmode4 mode,
6181     timespec32_t *mtime, cred_t *cr, vnode_t **vpp, bool_t *created)
6182 {
6183 	int error;
6184 	nfsstat4 status = NFS4_OK;
6185 	vattr_t va;
6186 
6187 tryagain:
6188 
6189 	/*
6190 	 * The file open mode used is VWRITE.  If the client needs
6191 	 * some other semantic, then it should do the access checking
6192 	 * itself.  It would have been nice to have the file open mode
6193 	 * passed as part of the arguments.
6194 	 */
6195 
6196 	*created = TRUE;
6197 	error = VOP_CREATE(dvp, nm, vap, EXCL, VWRITE, vpp, cr, 0, NULL, NULL);
6198 
6199 	if (error) {
6200 		*created = FALSE;
6201 
6202 		/*
6203 		 * If we got something other than file already exists
6204 		 * then just return this error.  Otherwise, we got
6205 		 * EEXIST.  If we were doing a GUARDED create, then
6206 		 * just return this error.  Otherwise, we need to
6207 		 * make sure that this wasn't a duplicate of an
6208 		 * exclusive create request.
6209 		 *
6210 		 * The assumption is made that a non-exclusive create
6211 		 * request will never return EEXIST.
6212 		 */
6213 
6214 		if (error != EEXIST || mode == GUARDED4) {
6215 			status = puterrno4(error);
6216 			return (status);
6217 		}
6218 		error = VOP_LOOKUP(dvp, nm, vpp, NULL, 0, NULL, cr,
6219 		    NULL, NULL, NULL);
6220 
6221 		if (error) {
6222 			/*
6223 			 * We couldn't find the file that we thought that
6224 			 * we just created.  So, we'll just try creating
6225 			 * it again.
6226 			 */
6227 			if (error == ENOENT)
6228 				goto tryagain;
6229 
6230 			status = puterrno4(error);
6231 			return (status);
6232 		}
6233 
6234 		if (mode == UNCHECKED4) {
6235 			/* existing object must be regular file */
6236 			if ((*vpp)->v_type != VREG) {
6237 				if ((*vpp)->v_type == VDIR)
6238 					status = NFS4ERR_ISDIR;
6239 				else if ((*vpp)->v_type == VLNK)
6240 					status = NFS4ERR_SYMLINK;
6241 				else
6242 					status = NFS4ERR_INVAL;
6243 				VN_RELE(*vpp);
6244 				return (status);
6245 			}
6246 
6247 			return (NFS4_OK);
6248 		}
6249 
6250 		/* Check for duplicate request */
6251 		ASSERT(mtime != 0);
6252 		va.va_mask = AT_MTIME;
6253 		error = VOP_GETATTR(*vpp, &va, 0, cr, NULL);
6254 		if (!error) {
6255 			/* We found the file */
6256 			if (va.va_mtime.tv_sec != mtime->tv_sec ||
6257 			    va.va_mtime.tv_nsec != mtime->tv_nsec) {
6258 				/* but its not our creation */
6259 				VN_RELE(*vpp);
6260 				return (NFS4ERR_EXIST);
6261 			}
6262 			*created = TRUE; /* retrans of create == created */
6263 			return (NFS4_OK);
6264 		}
6265 		VN_RELE(*vpp);
6266 		return (NFS4ERR_EXIST);
6267 	}
6268 
6269 	return (NFS4_OK);
6270 }
6271 
6272 static nfsstat4
6273 check_open_access(uint32_t access, struct compound_state *cs,
6274     struct svc_req *req)
6275 {
6276 	int error;
6277 	vnode_t *vp;
6278 	bool_t readonly;
6279 	cred_t *cr = cs->cr;
6280 
6281 	/* For now we don't allow mandatory locking as per V2/V3 */
6282 	if (cs->access == CS_ACCESS_DENIED || cs->mandlock) {
6283 		return (NFS4ERR_ACCESS);
6284 	}
6285 
6286 	vp = cs->vp;
6287 	ASSERT(cr != NULL && vp->v_type == VREG);
6288 
6289 	/*
6290 	 * If the file system is exported read only and we are trying
6291 	 * to open for write, then return NFS4ERR_ROFS
6292 	 */
6293 
6294 	readonly = rdonly4(cs->exi, cs->vp, req);
6295 
6296 	if ((access & OPEN4_SHARE_ACCESS_WRITE) && readonly)
6297 		return (NFS4ERR_ROFS);
6298 
6299 	if (access & OPEN4_SHARE_ACCESS_READ) {
6300 		if ((VOP_ACCESS(vp, VREAD, 0, cr, NULL) != 0) &&
6301 		    (VOP_ACCESS(vp, VEXEC, 0, cr, NULL) != 0)) {
6302 			return (NFS4ERR_ACCESS);
6303 		}
6304 	}
6305 
6306 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
6307 		error = VOP_ACCESS(vp, VWRITE, 0, cr, NULL);
6308 		if (error)
6309 			return (NFS4ERR_ACCESS);
6310 	}
6311 
6312 	return (NFS4_OK);
6313 }
6314 
6315 static nfsstat4
6316 rfs4_createfile(OPEN4args *args, struct svc_req *req, struct compound_state *cs,
6317     change_info4 *cinfo, bitmap4 *attrset, clientid4 clientid)
6318 {
6319 	struct nfs4_svgetit_arg sarg;
6320 	struct nfs4_ntov_table ntov;
6321 
6322 	bool_t ntov_table_init = FALSE;
6323 	struct statvfs64 sb;
6324 	nfsstat4 status;
6325 	vnode_t *vp;
6326 	vattr_t bva, ava, iva, cva, *vap;
6327 	vnode_t *dvp;
6328 	timespec32_t *mtime;
6329 	char *nm = NULL;
6330 	uint_t buflen;
6331 	bool_t created;
6332 	bool_t setsize = FALSE;
6333 	len_t reqsize;
6334 	int error;
6335 	bool_t trunc;
6336 	caller_context_t ct;
6337 	component4 *component;
6338 	bslabel_t *clabel;
6339 	struct sockaddr *ca;
6340 	char *name = NULL;
6341 
6342 	sarg.sbp = &sb;
6343 	sarg.is_referral = B_FALSE;
6344 
6345 	dvp = cs->vp;
6346 
6347 	/* Check if the file system is read only */
6348 	if (rdonly4(cs->exi, dvp, req))
6349 		return (NFS4ERR_ROFS);
6350 
6351 	/* check the label of including directory */
6352 	if (is_system_labeled()) {
6353 		ASSERT(req->rq_label != NULL);
6354 		clabel = req->rq_label;
6355 		DTRACE_PROBE2(tx__rfs4__log__info__opremove__clabel, char *,
6356 		    "got client label from request(1)",
6357 		    struct svc_req *, req);
6358 		if (!blequal(&l_admin_low->tsl_label, clabel)) {
6359 			if (!do_rfs_label_check(clabel, dvp, EQUALITY_CHECK,
6360 			    cs->exi)) {
6361 				return (NFS4ERR_ACCESS);
6362 			}
6363 		}
6364 	}
6365 
6366 	/*
6367 	 * Get the last component of path name in nm. cs will reference
6368 	 * the including directory on success.
6369 	 */
6370 	component = &args->open_claim4_u.file;
6371 	if (!utf8_dir_verify(component))
6372 		return (NFS4ERR_INVAL);
6373 
6374 	nm = utf8_to_fn(component, &buflen, NULL);
6375 
6376 	if (nm == NULL)
6377 		return (NFS4ERR_RESOURCE);
6378 
6379 	if (buflen > MAXNAMELEN) {
6380 		kmem_free(nm, buflen);
6381 		return (NFS4ERR_NAMETOOLONG);
6382 	}
6383 
6384 	bva.va_mask = AT_TYPE|AT_CTIME|AT_SEQ;
6385 	error = VOP_GETATTR(dvp, &bva, 0, cs->cr, NULL);
6386 	if (error) {
6387 		kmem_free(nm, buflen);
6388 		return (puterrno4(error));
6389 	}
6390 
6391 	if (bva.va_type != VDIR) {
6392 		kmem_free(nm, buflen);
6393 		return (NFS4ERR_NOTDIR);
6394 	}
6395 
6396 	NFS4_SET_FATTR4_CHANGE(cinfo->before, bva.va_ctime)
6397 
6398 	switch (args->mode) {
6399 	case GUARDED4:
6400 		/*FALLTHROUGH*/
6401 	case UNCHECKED4:
6402 		nfs4_ntov_table_init(&ntov);
6403 		ntov_table_init = TRUE;
6404 
6405 		*attrset = 0;
6406 		status = do_rfs4_set_attrs(attrset,
6407 		    &args->createhow4_u.createattrs,
6408 		    cs, &sarg, &ntov, NFS4ATTR_SETIT);
6409 
6410 		if (status == NFS4_OK && (sarg.vap->va_mask & AT_TYPE) &&
6411 		    sarg.vap->va_type != VREG) {
6412 			if (sarg.vap->va_type == VDIR)
6413 				status = NFS4ERR_ISDIR;
6414 			else if (sarg.vap->va_type == VLNK)
6415 				status = NFS4ERR_SYMLINK;
6416 			else
6417 				status = NFS4ERR_INVAL;
6418 		}
6419 
6420 		if (status != NFS4_OK) {
6421 			kmem_free(nm, buflen);
6422 			nfs4_ntov_table_free(&ntov, &sarg);
6423 			*attrset = 0;
6424 			return (status);
6425 		}
6426 
6427 		vap = sarg.vap;
6428 		vap->va_type = VREG;
6429 		vap->va_mask |= AT_TYPE;
6430 
6431 		if ((vap->va_mask & AT_MODE) == 0) {
6432 			vap->va_mask |= AT_MODE;
6433 			vap->va_mode = (mode_t)0600;
6434 		}
6435 
6436 		if (vap->va_mask & AT_SIZE) {
6437 
6438 			/* Disallow create with a non-zero size */
6439 
6440 			if ((reqsize = sarg.vap->va_size) != 0) {
6441 				kmem_free(nm, buflen);
6442 				nfs4_ntov_table_free(&ntov, &sarg);
6443 				*attrset = 0;
6444 				return (NFS4ERR_INVAL);
6445 			}
6446 			setsize = TRUE;
6447 		}
6448 		break;
6449 
6450 	case EXCLUSIVE4:
6451 		/* prohibit EXCL create of named attributes */
6452 		if (dvp->v_flag & V_XATTRDIR) {
6453 			kmem_free(nm, buflen);
6454 			*attrset = 0;
6455 			return (NFS4ERR_INVAL);
6456 		}
6457 
6458 		cva.va_mask = AT_TYPE | AT_MTIME | AT_MODE;
6459 		cva.va_type = VREG;
6460 		/*
6461 		 * Ensure no time overflows. Assumes underlying
6462 		 * filesystem supports at least 32 bits.
6463 		 * Truncate nsec to usec resolution to allow valid
6464 		 * compares even if the underlying filesystem truncates.
6465 		 */
6466 		mtime = (timespec32_t *)&args->createhow4_u.createverf;
6467 		cva.va_mtime.tv_sec = mtime->tv_sec % TIME32_MAX;
6468 		cva.va_mtime.tv_nsec = (mtime->tv_nsec / 1000) * 1000;
6469 		cva.va_mode = (mode_t)0;
6470 		vap = &cva;
6471 
6472 		/*
6473 		 * For EXCL create, attrset is set to the server attr
6474 		 * used to cache the client's verifier.
6475 		 */
6476 		*attrset = FATTR4_TIME_MODIFY_MASK;
6477 		break;
6478 	}
6479 
6480 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
6481 	name = nfscmd_convname(ca, cs->exi, nm, NFSCMD_CONV_INBOUND,
6482 	    MAXPATHLEN  + 1);
6483 
6484 	if (name == NULL) {
6485 		kmem_free(nm, buflen);
6486 		return (NFS4ERR_SERVERFAULT);
6487 	}
6488 
6489 	status = create_vnode(dvp, name, vap, args->mode, mtime,
6490 	    cs->cr, &vp, &created);
6491 	if (nm != name)
6492 		kmem_free(name, MAXPATHLEN + 1);
6493 	kmem_free(nm, buflen);
6494 
6495 	if (status != NFS4_OK) {
6496 		if (ntov_table_init)
6497 			nfs4_ntov_table_free(&ntov, &sarg);
6498 		*attrset = 0;
6499 		return (status);
6500 	}
6501 
6502 	trunc = (setsize && !created);
6503 
6504 	if (args->mode != EXCLUSIVE4) {
6505 		bitmap4 createmask = args->createhow4_u.createattrs.attrmask;
6506 
6507 		/*
6508 		 * True verification that object was created with correct
6509 		 * attrs is impossible.  The attrs could have been changed
6510 		 * immediately after object creation.  If attributes did
6511 		 * not verify, the only recourse for the server is to
6512 		 * destroy the object.  Maybe if some attrs (like gid)
6513 		 * are set incorrectly, the object should be destroyed;
6514 		 * however, seems bad as a default policy.  Do we really
6515 		 * want to destroy an object over one of the times not
6516 		 * verifying correctly?  For these reasons, the server
6517 		 * currently sets bits in attrset for createattrs
6518 		 * that were set; however, no verification is done.
6519 		 *
6520 		 * vmask_to_nmask accounts for vattr bits set on create
6521 		 *	[do_rfs4_set_attrs() only sets resp bits for
6522 		 *	 non-vattr/vfs bits.]
6523 		 * Mask off any bits we set by default so as not to return
6524 		 * more attrset bits than were requested in createattrs
6525 		 */
6526 		if (created) {
6527 			nfs4_vmask_to_nmask(sarg.vap->va_mask, attrset);
6528 			*attrset &= createmask;
6529 		} else {
6530 			/*
6531 			 * We did not create the vnode (we tried but it
6532 			 * already existed).  In this case, the only createattr
6533 			 * that the spec allows the server to set is size,
6534 			 * and even then, it can only be set if it is 0.
6535 			 */
6536 			*attrset = 0;
6537 			if (trunc)
6538 				*attrset = FATTR4_SIZE_MASK;
6539 		}
6540 	}
6541 	if (ntov_table_init)
6542 		nfs4_ntov_table_free(&ntov, &sarg);
6543 
6544 	/*
6545 	 * Get the initial "after" sequence number, if it fails,
6546 	 * set to zero, time to before.
6547 	 */
6548 	iva.va_mask = AT_CTIME|AT_SEQ;
6549 	if (VOP_GETATTR(dvp, &iva, 0, cs->cr, NULL)) {
6550 		iva.va_seq = 0;
6551 		iva.va_ctime = bva.va_ctime;
6552 	}
6553 
6554 	/*
6555 	 * create_vnode attempts to create the file exclusive,
6556 	 * if it already exists the VOP_CREATE will fail and
6557 	 * may not increase va_seq. It is atomic if
6558 	 * we haven't changed the directory, but if it has changed
6559 	 * we don't know what changed it.
6560 	 */
6561 	if (!created) {
6562 		if (bva.va_seq && iva.va_seq &&
6563 		    bva.va_seq == iva.va_seq)
6564 			cinfo->atomic = TRUE;
6565 		else
6566 			cinfo->atomic = FALSE;
6567 		NFS4_SET_FATTR4_CHANGE(cinfo->after, iva.va_ctime);
6568 	} else {
6569 		/*
6570 		 * The entry was created, we need to sync the
6571 		 * directory metadata.
6572 		 */
6573 		(void) VOP_FSYNC(dvp, 0, cs->cr, NULL);
6574 
6575 		/*
6576 		 * Get "after" change value, if it fails, simply return the
6577 		 * before value.
6578 		 */
6579 		ava.va_mask = AT_CTIME|AT_SEQ;
6580 		if (VOP_GETATTR(dvp, &ava, 0, cs->cr, NULL)) {
6581 			ava.va_ctime = bva.va_ctime;
6582 			ava.va_seq = 0;
6583 		}
6584 
6585 		NFS4_SET_FATTR4_CHANGE(cinfo->after, ava.va_ctime);
6586 
6587 		/*
6588 		 * The cinfo->atomic = TRUE only if we have
6589 		 * non-zero va_seq's, and it has incremented by exactly one
6590 		 * during the create_vnode and it didn't
6591 		 * change during the VOP_FSYNC.
6592 		 */
6593 		if (bva.va_seq && iva.va_seq && ava.va_seq &&
6594 		    iva.va_seq == (bva.va_seq + 1) && iva.va_seq == ava.va_seq)
6595 			cinfo->atomic = TRUE;
6596 		else
6597 			cinfo->atomic = FALSE;
6598 	}
6599 
6600 	/* Check for mandatory locking and that the size gets set. */
6601 	cva.va_mask = AT_MODE;
6602 	if (setsize)
6603 		cva.va_mask |= AT_SIZE;
6604 
6605 	/* Assume the worst */
6606 	cs->mandlock = TRUE;
6607 
6608 	if (VOP_GETATTR(vp, &cva, 0, cs->cr, NULL) == 0) {
6609 		cs->mandlock = MANDLOCK(cs->vp, cva.va_mode);
6610 
6611 		/*
6612 		 * Truncate the file if necessary; this would be
6613 		 * the case for create over an existing file.
6614 		 */
6615 
6616 		if (trunc) {
6617 			int in_crit = 0;
6618 			rfs4_file_t *fp;
6619 			bool_t create = FALSE;
6620 
6621 			/*
6622 			 * We are writing over an existing file.
6623 			 * Check to see if we need to recall a delegation.
6624 			 */
6625 			rfs4_hold_deleg_policy();
6626 			if ((fp = rfs4_findfile(vp, NULL, &create)) != NULL) {
6627 				if (rfs4_check_delegated_byfp(FWRITE, fp,
6628 				    (reqsize == 0), FALSE, FALSE, &clientid)) {
6629 					rfs4_file_rele(fp);
6630 					rfs4_rele_deleg_policy();
6631 					VN_RELE(vp);
6632 					*attrset = 0;
6633 					return (NFS4ERR_DELAY);
6634 				}
6635 				rfs4_file_rele(fp);
6636 			}
6637 			rfs4_rele_deleg_policy();
6638 
6639 			if (nbl_need_check(vp)) {
6640 				in_crit = 1;
6641 
6642 				ASSERT(reqsize == 0);
6643 
6644 				nbl_start_crit(vp, RW_READER);
6645 				if (nbl_conflict(vp, NBL_WRITE, 0,
6646 				    cva.va_size, 0, NULL)) {
6647 					in_crit = 0;
6648 					nbl_end_crit(vp);
6649 					VN_RELE(vp);
6650 					*attrset = 0;
6651 					return (NFS4ERR_ACCESS);
6652 				}
6653 			}
6654 			ct.cc_sysid = 0;
6655 			ct.cc_pid = 0;
6656 			ct.cc_caller_id = nfs4_srv_caller_id;
6657 			ct.cc_flags = CC_DONTBLOCK;
6658 
6659 			cva.va_mask = AT_SIZE;
6660 			cva.va_size = reqsize;
6661 			(void) VOP_SETATTR(vp, &cva, 0, cs->cr, &ct);
6662 			if (in_crit)
6663 				nbl_end_crit(vp);
6664 		}
6665 	}
6666 
6667 	error = makefh4(&cs->fh, vp, cs->exi);
6668 
6669 	/*
6670 	 * Force modified data and metadata out to stable storage.
6671 	 */
6672 	(void) VOP_FSYNC(vp, FNODSYNC, cs->cr, NULL);
6673 
6674 	if (error) {
6675 		VN_RELE(vp);
6676 		*attrset = 0;
6677 		return (puterrno4(error));
6678 	}
6679 
6680 	/* if parent dir is attrdir, set namedattr fh flag */
6681 	if (dvp->v_flag & V_XATTRDIR)
6682 		set_fh4_flag(&cs->fh, FH4_NAMEDATTR);
6683 
6684 	if (cs->vp)
6685 		VN_RELE(cs->vp);
6686 
6687 	cs->vp = vp;
6688 
6689 	/*
6690 	 * if we did not create the file, we will need to check
6691 	 * the access bits on the file
6692 	 */
6693 
6694 	if (!created) {
6695 		if (setsize)
6696 			args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
6697 		status = check_open_access(args->share_access, cs, req);
6698 		if (status != NFS4_OK)
6699 			*attrset = 0;
6700 	}
6701 	return (status);
6702 }
6703 
6704 /*ARGSUSED*/
6705 static void
6706 rfs4_do_open(struct compound_state *cs, struct svc_req *req,
6707     rfs4_openowner_t *oo, delegreq_t deleg,
6708     uint32_t access, uint32_t deny,
6709     OPEN4res *resp, int deleg_cur)
6710 {
6711 	/* XXX Currently not using req  */
6712 	rfs4_state_t *sp;
6713 	rfs4_file_t *fp;
6714 	bool_t screate = TRUE;
6715 	bool_t fcreate = TRUE;
6716 	uint32_t open_a, share_a;
6717 	uint32_t open_d, share_d;
6718 	rfs4_deleg_state_t *dsp;
6719 	sysid_t sysid;
6720 	nfsstat4 status;
6721 	caller_context_t ct;
6722 	int fflags = 0;
6723 	int recall = 0;
6724 	int err;
6725 	int first_open;
6726 
6727 	/* get the file struct and hold a lock on it during initial open */
6728 	fp = rfs4_findfile_withlock(cs->vp, &cs->fh, &fcreate);
6729 	if (fp == NULL) {
6730 		resp->status = NFS4ERR_RESOURCE;
6731 		DTRACE_PROBE1(nfss__e__do__open1, nfsstat4, resp->status);
6732 		return;
6733 	}
6734 
6735 	sp = rfs4_findstate_by_owner_file(oo, fp, &screate);
6736 	if (sp == NULL) {
6737 		resp->status = NFS4ERR_RESOURCE;
6738 		DTRACE_PROBE1(nfss__e__do__open2, nfsstat4, resp->status);
6739 		/* No need to keep any reference */
6740 		rw_exit(&fp->rf_file_rwlock);
6741 		rfs4_file_rele(fp);
6742 		return;
6743 	}
6744 
6745 	/* try to get the sysid before continuing */
6746 	if ((status = rfs4_client_sysid(oo->ro_client, &sysid)) != NFS4_OK) {
6747 		resp->status = status;
6748 		rfs4_file_rele(fp);
6749 		/* Not a fully formed open; "close" it */
6750 		if (screate == TRUE)
6751 			rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6752 		rfs4_state_rele(sp);
6753 		return;
6754 	}
6755 
6756 	/* Calculate the fflags for this OPEN. */
6757 	if (access & OPEN4_SHARE_ACCESS_READ)
6758 		fflags |= FREAD;
6759 	if (access & OPEN4_SHARE_ACCESS_WRITE)
6760 		fflags |= FWRITE;
6761 
6762 	rfs4_dbe_lock(sp->rs_dbe);
6763 
6764 	/*
6765 	 * Calculate the new deny and access mode that this open is adding to
6766 	 * the file for this open owner;
6767 	 */
6768 	open_d = (deny & ~sp->rs_open_deny);
6769 	open_a = (access & ~sp->rs_open_access);
6770 
6771 	/*
6772 	 * Calculate the new share access and share deny modes that this open
6773 	 * is adding to the file for this open owner;
6774 	 */
6775 	share_a = (access & ~sp->rs_share_access);
6776 	share_d = (deny & ~sp->rs_share_deny);
6777 
6778 	first_open = (sp->rs_open_access & OPEN4_SHARE_ACCESS_BOTH) == 0;
6779 
6780 	/*
6781 	 * Check to see the client has already sent an open for this
6782 	 * open owner on this file with the same share/deny modes.
6783 	 * If so, we don't need to check for a conflict and we don't
6784 	 * need to add another shrlock.  If not, then we need to
6785 	 * check for conflicts in deny and access before checking for
6786 	 * conflicts in delegation.  We don't want to recall a
6787 	 * delegation based on an open that will eventually fail based
6788 	 * on shares modes.
6789 	 */
6790 
6791 	if (share_a || share_d) {
6792 		if ((err = rfs4_share(sp, access, deny)) != 0) {
6793 			rfs4_dbe_unlock(sp->rs_dbe);
6794 			resp->status = err;
6795 
6796 			rfs4_file_rele(fp);
6797 			/* Not a fully formed open; "close" it */
6798 			if (screate == TRUE)
6799 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6800 			rfs4_state_rele(sp);
6801 			return;
6802 		}
6803 	}
6804 
6805 	rfs4_dbe_lock(fp->rf_dbe);
6806 
6807 	/*
6808 	 * Check to see if this file is delegated and if so, if a
6809 	 * recall needs to be done.
6810 	 */
6811 	if (rfs4_check_recall(sp, access)) {
6812 		rfs4_dbe_unlock(fp->rf_dbe);
6813 		rfs4_dbe_unlock(sp->rs_dbe);
6814 		rfs4_recall_deleg(fp, FALSE, sp->rs_owner->ro_client);
6815 		delay(NFS4_DELEGATION_CONFLICT_DELAY);
6816 		rfs4_dbe_lock(sp->rs_dbe);
6817 
6818 		/* if state closed while lock was dropped */
6819 		if (sp->rs_closed) {
6820 			if (share_a || share_d)
6821 				(void) rfs4_unshare(sp);
6822 			rfs4_dbe_unlock(sp->rs_dbe);
6823 			rfs4_file_rele(fp);
6824 			/* Not a fully formed open; "close" it */
6825 			if (screate == TRUE)
6826 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6827 			rfs4_state_rele(sp);
6828 			resp->status = NFS4ERR_OLD_STATEID;
6829 			return;
6830 		}
6831 
6832 		rfs4_dbe_lock(fp->rf_dbe);
6833 		/* Let's see if the delegation was returned */
6834 		if (rfs4_check_recall(sp, access)) {
6835 			rfs4_dbe_unlock(fp->rf_dbe);
6836 			if (share_a || share_d)
6837 				(void) rfs4_unshare(sp);
6838 			rfs4_dbe_unlock(sp->rs_dbe);
6839 			rfs4_file_rele(fp);
6840 			rfs4_update_lease(sp->rs_owner->ro_client);
6841 
6842 			/* Not a fully formed open; "close" it */
6843 			if (screate == TRUE)
6844 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6845 			rfs4_state_rele(sp);
6846 			resp->status = NFS4ERR_DELAY;
6847 			return;
6848 		}
6849 	}
6850 	/*
6851 	 * the share check passed and any delegation conflict has been
6852 	 * taken care of, now call vop_open.
6853 	 * if this is the first open then call vop_open with fflags.
6854 	 * if not, call vn_open_upgrade with just the upgrade flags.
6855 	 *
6856 	 * if the file has been opened already, it will have the current
6857 	 * access mode in the state struct.  if it has no share access, then
6858 	 * this is a new open.
6859 	 *
6860 	 * However, if this is open with CLAIM_DLEGATE_CUR, then don't
6861 	 * call VOP_OPEN(), just do the open upgrade.
6862 	 */
6863 	if (first_open && !deleg_cur) {
6864 		ct.cc_sysid = sysid;
6865 		ct.cc_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
6866 		ct.cc_caller_id = nfs4_srv_caller_id;
6867 		ct.cc_flags = CC_DONTBLOCK;
6868 		err = VOP_OPEN(&cs->vp, fflags, cs->cr, &ct);
6869 		if (err) {
6870 			rfs4_dbe_unlock(fp->rf_dbe);
6871 			if (share_a || share_d)
6872 				(void) rfs4_unshare(sp);
6873 			rfs4_dbe_unlock(sp->rs_dbe);
6874 			rfs4_file_rele(fp);
6875 
6876 			/* Not a fully formed open; "close" it */
6877 			if (screate == TRUE)
6878 				rfs4_state_close(sp, FALSE, FALSE, cs->cr);
6879 			rfs4_state_rele(sp);
6880 			/* check if a monitor detected a delegation conflict */
6881 			if (err == EAGAIN && (ct.cc_flags & CC_WOULDBLOCK))
6882 				resp->status = NFS4ERR_DELAY;
6883 			else
6884 				resp->status = NFS4ERR_SERVERFAULT;
6885 			return;
6886 		}
6887 	} else { /* open upgrade */
6888 		/*
6889 		 * calculate the fflags for the new mode that is being added
6890 		 * by this upgrade.
6891 		 */
6892 		fflags = 0;
6893 		if (open_a & OPEN4_SHARE_ACCESS_READ)
6894 			fflags |= FREAD;
6895 		if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6896 			fflags |= FWRITE;
6897 		vn_open_upgrade(cs->vp, fflags);
6898 	}
6899 	sp->rs_open_access |= access;
6900 	sp->rs_open_deny |= deny;
6901 
6902 	if (open_d & OPEN4_SHARE_DENY_READ)
6903 		fp->rf_deny_read++;
6904 	if (open_d & OPEN4_SHARE_DENY_WRITE)
6905 		fp->rf_deny_write++;
6906 	fp->rf_share_deny |= deny;
6907 
6908 	if (open_a & OPEN4_SHARE_ACCESS_READ)
6909 		fp->rf_access_read++;
6910 	if (open_a & OPEN4_SHARE_ACCESS_WRITE)
6911 		fp->rf_access_write++;
6912 	fp->rf_share_access |= access;
6913 
6914 	/*
6915 	 * Check for delegation here. if the deleg argument is not
6916 	 * DELEG_ANY, then this is a reclaim from a client and
6917 	 * we must honor the delegation requested. If necessary we can
6918 	 * set the recall flag.
6919 	 */
6920 
6921 	dsp = rfs4_grant_delegation(deleg, sp, &recall);
6922 
6923 	cs->deleg = (fp->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE);
6924 
6925 	next_stateid(&sp->rs_stateid);
6926 
6927 	resp->stateid = sp->rs_stateid.stateid;
6928 
6929 	rfs4_dbe_unlock(fp->rf_dbe);
6930 	rfs4_dbe_unlock(sp->rs_dbe);
6931 
6932 	if (dsp) {
6933 		rfs4_set_deleg_response(dsp, &resp->delegation, NULL, recall);
6934 		rfs4_deleg_state_rele(dsp);
6935 	}
6936 
6937 	rfs4_file_rele(fp);
6938 	rfs4_state_rele(sp);
6939 
6940 	resp->status = NFS4_OK;
6941 }
6942 
6943 /*ARGSUSED*/
6944 static void
6945 rfs4_do_opennull(struct compound_state *cs, struct svc_req *req,
6946     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6947 {
6948 	change_info4 *cinfo = &resp->cinfo;
6949 	bitmap4 *attrset = &resp->attrset;
6950 
6951 	if (args->opentype == OPEN4_NOCREATE)
6952 		resp->status = rfs4_lookupfile(&args->open_claim4_u.file,
6953 		    req, cs, args->share_access, cinfo);
6954 	else {
6955 		/* inhibit delegation grants during exclusive create */
6956 
6957 		if (args->mode == EXCLUSIVE4)
6958 			rfs4_disable_delegation();
6959 
6960 		resp->status = rfs4_createfile(args, req, cs, cinfo, attrset,
6961 		    oo->ro_client->rc_clientid);
6962 	}
6963 
6964 	if (resp->status == NFS4_OK) {
6965 
6966 		/* cs->vp cs->fh now reference the desired file */
6967 
6968 		rfs4_do_open(cs, req, oo,
6969 		    oo->ro_need_confirm ? DELEG_NONE : DELEG_ANY,
6970 		    args->share_access, args->share_deny, resp, 0);
6971 
6972 		/*
6973 		 * If rfs4_createfile set attrset, we must
6974 		 * clear this attrset before the response is copied.
6975 		 */
6976 		if (resp->status != NFS4_OK && resp->attrset) {
6977 			resp->attrset = 0;
6978 		}
6979 	}
6980 	else
6981 		*cs->statusp = resp->status;
6982 
6983 	if (args->mode == EXCLUSIVE4)
6984 		rfs4_enable_delegation();
6985 }
6986 
6987 /*ARGSUSED*/
6988 static void
6989 rfs4_do_openprev(struct compound_state *cs, struct svc_req *req,
6990     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
6991 {
6992 	change_info4 *cinfo = &resp->cinfo;
6993 	vattr_t va;
6994 	vtype_t v_type = cs->vp->v_type;
6995 	int error = 0;
6996 
6997 	/* Verify that we have a regular file */
6998 	if (v_type != VREG) {
6999 		if (v_type == VDIR)
7000 			resp->status = NFS4ERR_ISDIR;
7001 		else if (v_type == VLNK)
7002 			resp->status = NFS4ERR_SYMLINK;
7003 		else
7004 			resp->status = NFS4ERR_INVAL;
7005 		return;
7006 	}
7007 
7008 	va.va_mask = AT_MODE|AT_UID;
7009 	error = VOP_GETATTR(cs->vp, &va, 0, cs->cr, NULL);
7010 	if (error) {
7011 		resp->status = puterrno4(error);
7012 		return;
7013 	}
7014 
7015 	cs->mandlock = MANDLOCK(cs->vp, va.va_mode);
7016 
7017 	/*
7018 	 * Check if we have access to the file, Note the the file
7019 	 * could have originally been open UNCHECKED or GUARDED
7020 	 * with mode bits that will now fail, but there is nothing
7021 	 * we can really do about that except in the case that the
7022 	 * owner of the file is the one requesting the open.
7023 	 */
7024 	if (crgetuid(cs->cr) != va.va_uid) {
7025 		resp->status = check_open_access(args->share_access, cs, req);
7026 		if (resp->status != NFS4_OK) {
7027 			return;
7028 		}
7029 	}
7030 
7031 	/*
7032 	 * cinfo on a CLAIM_PREVIOUS is undefined, initialize to zero
7033 	 */
7034 	cinfo->before = 0;
7035 	cinfo->after = 0;
7036 	cinfo->atomic = FALSE;
7037 
7038 	rfs4_do_open(cs, req, oo,
7039 	    NFS4_DELEG4TYPE2REQTYPE(args->open_claim4_u.delegate_type),
7040 	    args->share_access, args->share_deny, resp, 0);
7041 }
7042 
7043 static void
7044 rfs4_do_opendelcur(struct compound_state *cs, struct svc_req *req,
7045     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7046 {
7047 	int error;
7048 	nfsstat4 status;
7049 	stateid4 stateid =
7050 	    args->open_claim4_u.delegate_cur_info.delegate_stateid;
7051 	rfs4_deleg_state_t *dsp;
7052 
7053 	/*
7054 	 * Find the state info from the stateid and confirm that the
7055 	 * file is delegated.  If the state openowner is the same as
7056 	 * the supplied openowner we're done. If not, get the file
7057 	 * info from the found state info. Use that file info to
7058 	 * create the state for this lock owner. Note solaris doen't
7059 	 * really need the pathname to find the file. We may want to
7060 	 * lookup the pathname and make sure that the vp exist and
7061 	 * matches the vp in the file structure. However it is
7062 	 * possible that the pathname nolonger exists (local process
7063 	 * unlinks the file), so this may not be that useful.
7064 	 */
7065 
7066 	status = rfs4_get_deleg_state(&stateid, &dsp);
7067 	if (status != NFS4_OK) {
7068 		resp->status = status;
7069 		return;
7070 	}
7071 
7072 	ASSERT(dsp->rds_finfo->rf_dinfo.rd_dtype != OPEN_DELEGATE_NONE);
7073 
7074 	/*
7075 	 * New lock owner, create state. Since this was probably called
7076 	 * in response to a CB_RECALL we set deleg to DELEG_NONE
7077 	 */
7078 
7079 	ASSERT(cs->vp != NULL);
7080 	VN_RELE(cs->vp);
7081 	VN_HOLD(dsp->rds_finfo->rf_vp);
7082 	cs->vp = dsp->rds_finfo->rf_vp;
7083 
7084 	if (error = makefh4(&cs->fh, cs->vp, cs->exi)) {
7085 		rfs4_deleg_state_rele(dsp);
7086 		*cs->statusp = resp->status = puterrno4(error);
7087 		return;
7088 	}
7089 
7090 	/* Mark progress for delegation returns */
7091 	dsp->rds_finfo->rf_dinfo.rd_time_lastwrite = gethrestime_sec();
7092 	rfs4_deleg_state_rele(dsp);
7093 	rfs4_do_open(cs, req, oo, DELEG_NONE,
7094 	    args->share_access, args->share_deny, resp, 1);
7095 }
7096 
7097 /*ARGSUSED*/
7098 static void
7099 rfs4_do_opendelprev(struct compound_state *cs, struct svc_req *req,
7100     OPEN4args *args, rfs4_openowner_t *oo, OPEN4res *resp)
7101 {
7102 	/*
7103 	 * Lookup the pathname, it must already exist since this file
7104 	 * was delegated.
7105 	 *
7106 	 * Find the file and state info for this vp and open owner pair.
7107 	 *	check that they are in fact delegated.
7108 	 *	check that the state access and deny modes are the same.
7109 	 *
7110 	 * Return the delgation possibly seting the recall flag.
7111 	 */
7112 	rfs4_file_t *fp;
7113 	rfs4_state_t *sp;
7114 	bool_t create = FALSE;
7115 	bool_t dcreate = FALSE;
7116 	rfs4_deleg_state_t *dsp;
7117 	nfsace4 *ace;
7118 
7119 	/* Note we ignore oflags */
7120 	resp->status = rfs4_lookupfile(&args->open_claim4_u.file_delegate_prev,
7121 	    req, cs, args->share_access, &resp->cinfo);
7122 
7123 	if (resp->status != NFS4_OK) {
7124 		return;
7125 	}
7126 
7127 	/* get the file struct and hold a lock on it during initial open */
7128 	fp = rfs4_findfile_withlock(cs->vp, NULL, &create);
7129 	if (fp == NULL) {
7130 		resp->status = NFS4ERR_RESOURCE;
7131 		DTRACE_PROBE1(nfss__e__do_opendelprev1, nfsstat4, resp->status);
7132 		return;
7133 	}
7134 
7135 	sp = rfs4_findstate_by_owner_file(oo, fp, &create);
7136 	if (sp == NULL) {
7137 		resp->status = NFS4ERR_SERVERFAULT;
7138 		DTRACE_PROBE1(nfss__e__do_opendelprev2, nfsstat4, resp->status);
7139 		rw_exit(&fp->rf_file_rwlock);
7140 		rfs4_file_rele(fp);
7141 		return;
7142 	}
7143 
7144 	rfs4_dbe_lock(sp->rs_dbe);
7145 	rfs4_dbe_lock(fp->rf_dbe);
7146 	if (args->share_access != sp->rs_share_access ||
7147 	    args->share_deny != sp->rs_share_deny ||
7148 	    sp->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_NONE) {
7149 		NFS4_DEBUG(rfs4_debug,
7150 		    (CE_NOTE, "rfs4_do_opendelprev: state mixup"));
7151 		rfs4_dbe_unlock(fp->rf_dbe);
7152 		rfs4_dbe_unlock(sp->rs_dbe);
7153 		rfs4_file_rele(fp);
7154 		rfs4_state_rele(sp);
7155 		resp->status = NFS4ERR_SERVERFAULT;
7156 		return;
7157 	}
7158 	rfs4_dbe_unlock(fp->rf_dbe);
7159 	rfs4_dbe_unlock(sp->rs_dbe);
7160 
7161 	dsp = rfs4_finddeleg(sp, &dcreate);
7162 	if (dsp == NULL) {
7163 		rfs4_state_rele(sp);
7164 		rfs4_file_rele(fp);
7165 		resp->status = NFS4ERR_SERVERFAULT;
7166 		return;
7167 	}
7168 
7169 	next_stateid(&sp->rs_stateid);
7170 
7171 	resp->stateid = sp->rs_stateid.stateid;
7172 
7173 	resp->delegation.delegation_type = dsp->rds_dtype;
7174 
7175 	if (dsp->rds_dtype == OPEN_DELEGATE_READ) {
7176 		open_read_delegation4 *rv =
7177 		    &resp->delegation.open_delegation4_u.read;
7178 
7179 		rv->stateid = dsp->rds_delegid.stateid;
7180 		rv->recall = FALSE; /* no policy in place to set to TRUE */
7181 		ace = &rv->permissions;
7182 	} else {
7183 		open_write_delegation4 *rv =
7184 		    &resp->delegation.open_delegation4_u.write;
7185 
7186 		rv->stateid = dsp->rds_delegid.stateid;
7187 		rv->recall = FALSE;  /* no policy in place to set to TRUE */
7188 		ace = &rv->permissions;
7189 		rv->space_limit.limitby = NFS_LIMIT_SIZE;
7190 		rv->space_limit.nfs_space_limit4_u.filesize = UINT64_MAX;
7191 	}
7192 
7193 	/* XXX For now */
7194 	ace->type = ACE4_ACCESS_ALLOWED_ACE_TYPE;
7195 	ace->flag = 0;
7196 	ace->access_mask = 0;
7197 	ace->who.utf8string_len = 0;
7198 	ace->who.utf8string_val = 0;
7199 
7200 	rfs4_deleg_state_rele(dsp);
7201 	rfs4_state_rele(sp);
7202 	rfs4_file_rele(fp);
7203 }
7204 
7205 typedef enum {
7206 	NFS4_CHKSEQ_OKAY = 0,
7207 	NFS4_CHKSEQ_REPLAY = 1,
7208 	NFS4_CHKSEQ_BAD = 2
7209 } rfs4_chkseq_t;
7210 
7211 /*
7212  * Generic function for sequence number checks.
7213  */
7214 static rfs4_chkseq_t
7215 rfs4_check_seqid(seqid4 seqid, nfs_resop4 *lastop,
7216     seqid4 rqst_seq, nfs_resop4 *resop, bool_t copyres)
7217 {
7218 	/* Same sequence ids and matching operations? */
7219 	if (seqid == rqst_seq && resop->resop == lastop->resop) {
7220 		if (copyres == TRUE) {
7221 			rfs4_free_reply(resop);
7222 			rfs4_copy_reply(resop, lastop);
7223 		}
7224 		NFS4_DEBUG(rfs4_debug, (CE_NOTE,
7225 		    "Replayed SEQID %d\n", seqid));
7226 		return (NFS4_CHKSEQ_REPLAY);
7227 	}
7228 
7229 	/* If the incoming sequence is not the next expected then it is bad */
7230 	if (rqst_seq != seqid + 1) {
7231 		if (rqst_seq == seqid) {
7232 			NFS4_DEBUG(rfs4_debug,
7233 			    (CE_NOTE, "BAD SEQID: Replayed sequence id "
7234 			    "but last op was %d current op is %d\n",
7235 			    lastop->resop, resop->resop));
7236 			return (NFS4_CHKSEQ_BAD);
7237 		}
7238 		NFS4_DEBUG(rfs4_debug,
7239 		    (CE_NOTE, "BAD SEQID: got %u expecting %u\n",
7240 		    rqst_seq, seqid));
7241 		return (NFS4_CHKSEQ_BAD);
7242 	}
7243 
7244 	/* Everything okay -- next expected */
7245 	return (NFS4_CHKSEQ_OKAY);
7246 }
7247 
7248 
7249 static rfs4_chkseq_t
7250 rfs4_check_open_seqid(seqid4 seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7251 {
7252 	rfs4_chkseq_t rc;
7253 
7254 	rfs4_dbe_lock(op->ro_dbe);
7255 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply, seqid, resop,
7256 	    TRUE);
7257 	rfs4_dbe_unlock(op->ro_dbe);
7258 
7259 	if (rc == NFS4_CHKSEQ_OKAY)
7260 		rfs4_update_lease(op->ro_client);
7261 
7262 	return (rc);
7263 }
7264 
7265 static rfs4_chkseq_t
7266 rfs4_check_olo_seqid(seqid4 olo_seqid, rfs4_openowner_t *op, nfs_resop4 *resop)
7267 {
7268 	rfs4_chkseq_t rc;
7269 
7270 	rfs4_dbe_lock(op->ro_dbe);
7271 	rc = rfs4_check_seqid(op->ro_open_seqid, &op->ro_reply,
7272 	    olo_seqid, resop, FALSE);
7273 	rfs4_dbe_unlock(op->ro_dbe);
7274 
7275 	return (rc);
7276 }
7277 
7278 static rfs4_chkseq_t
7279 rfs4_check_lock_seqid(seqid4 seqid, rfs4_lo_state_t *lsp, nfs_resop4 *resop)
7280 {
7281 	rfs4_chkseq_t rc = NFS4_CHKSEQ_OKAY;
7282 
7283 	rfs4_dbe_lock(lsp->rls_dbe);
7284 	if (!lsp->rls_skip_seqid_check)
7285 		rc = rfs4_check_seqid(lsp->rls_seqid, &lsp->rls_reply, seqid,
7286 		    resop, TRUE);
7287 	rfs4_dbe_unlock(lsp->rls_dbe);
7288 
7289 	return (rc);
7290 }
7291 
7292 static void
7293 rfs4_op_open(nfs_argop4 *argop, nfs_resop4 *resop,
7294     struct svc_req *req, struct compound_state *cs)
7295 {
7296 	OPEN4args *args = &argop->nfs_argop4_u.opopen;
7297 	OPEN4res *resp = &resop->nfs_resop4_u.opopen;
7298 	open_owner4 *owner = &args->owner;
7299 	open_claim_type4 claim = args->claim;
7300 	rfs4_client_t *cp;
7301 	rfs4_openowner_t *oo;
7302 	bool_t create;
7303 	bool_t replay = FALSE;
7304 	int can_reclaim;
7305 
7306 	DTRACE_NFSV4_2(op__open__start, struct compound_state *, cs,
7307 	    OPEN4args *, args);
7308 
7309 	if (cs->vp == NULL) {
7310 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7311 		goto end;
7312 	}
7313 
7314 	/*
7315 	 * Need to check clientid and lease expiration first based on
7316 	 * error ordering and incrementing sequence id.
7317 	 */
7318 	cp = rfs4_findclient_by_id(owner->clientid, FALSE);
7319 	if (cp == NULL) {
7320 		*cs->statusp = resp->status =
7321 		    rfs4_check_clientid(&owner->clientid, 0);
7322 		goto end;
7323 	}
7324 
7325 	if (rfs4_lease_expired(cp)) {
7326 		rfs4_client_close(cp);
7327 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7328 		goto end;
7329 	}
7330 	can_reclaim = cp->rc_can_reclaim;
7331 
7332 	/*
7333 	 * Find the open_owner for use from this point forward.  Take
7334 	 * care in updating the sequence id based on the type of error
7335 	 * being returned.
7336 	 */
7337 retry:
7338 	create = TRUE;
7339 	oo = rfs4_findopenowner(owner, &create, args->seqid);
7340 	if (oo == NULL) {
7341 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
7342 		rfs4_client_rele(cp);
7343 		goto end;
7344 	}
7345 
7346 	/* Hold off access to the sequence space while the open is done */
7347 	rfs4_sw_enter(&oo->ro_sw);
7348 
7349 	/*
7350 	 * If the open_owner existed before at the server, then check
7351 	 * the sequence id.
7352 	 */
7353 	if (!create && !oo->ro_postpone_confirm) {
7354 		switch (rfs4_check_open_seqid(args->seqid, oo, resop)) {
7355 		case NFS4_CHKSEQ_BAD:
7356 			if ((args->seqid > oo->ro_open_seqid) &&
7357 			    oo->ro_need_confirm) {
7358 				rfs4_free_opens(oo, TRUE, FALSE);
7359 				rfs4_sw_exit(&oo->ro_sw);
7360 				rfs4_openowner_rele(oo);
7361 				goto retry;
7362 			}
7363 			resp->status = NFS4ERR_BAD_SEQID;
7364 			goto out;
7365 		case NFS4_CHKSEQ_REPLAY: /* replay of previous request */
7366 			replay = TRUE;
7367 			goto out;
7368 		default:
7369 			break;
7370 		}
7371 
7372 		/*
7373 		 * Sequence was ok and open owner exists
7374 		 * check to see if we have yet to see an
7375 		 * open_confirm.
7376 		 */
7377 		if (oo->ro_need_confirm) {
7378 			rfs4_free_opens(oo, TRUE, FALSE);
7379 			rfs4_sw_exit(&oo->ro_sw);
7380 			rfs4_openowner_rele(oo);
7381 			goto retry;
7382 		}
7383 	}
7384 	/* Grace only applies to regular-type OPENs */
7385 	if (rfs4_clnt_in_grace(cp) &&
7386 	    (claim == CLAIM_NULL || claim == CLAIM_DELEGATE_CUR)) {
7387 		*cs->statusp = resp->status = NFS4ERR_GRACE;
7388 		goto out;
7389 	}
7390 
7391 	/*
7392 	 * If previous state at the server existed then can_reclaim
7393 	 * will be set. If not reply NFS4ERR_NO_GRACE to the
7394 	 * client.
7395 	 */
7396 	if (rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS && !can_reclaim) {
7397 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7398 		goto out;
7399 	}
7400 
7401 
7402 	/*
7403 	 * Reject the open if the client has missed the grace period
7404 	 */
7405 	if (!rfs4_clnt_in_grace(cp) && claim == CLAIM_PREVIOUS) {
7406 		*cs->statusp = resp->status = NFS4ERR_NO_GRACE;
7407 		goto out;
7408 	}
7409 
7410 	/* Couple of up-front bookkeeping items */
7411 	if (oo->ro_need_confirm) {
7412 		/*
7413 		 * If this is a reclaim OPEN then we should not ask
7414 		 * for a confirmation of the open_owner per the
7415 		 * protocol specification.
7416 		 */
7417 		if (claim == CLAIM_PREVIOUS)
7418 			oo->ro_need_confirm = FALSE;
7419 		else
7420 			resp->rflags |= OPEN4_RESULT_CONFIRM;
7421 	}
7422 	resp->rflags |= OPEN4_RESULT_LOCKTYPE_POSIX;
7423 
7424 	/*
7425 	 * If there is an unshared filesystem mounted on this vnode,
7426 	 * do not allow to open/create in this directory.
7427 	 */
7428 	if (vn_ismntpt(cs->vp)) {
7429 		*cs->statusp = resp->status = NFS4ERR_ACCESS;
7430 		goto out;
7431 	}
7432 
7433 	/*
7434 	 * access must READ, WRITE, or BOTH.  No access is invalid.
7435 	 * deny can be READ, WRITE, BOTH, or NONE.
7436 	 * bits not defined for access/deny are invalid.
7437 	 */
7438 	if (! (args->share_access & OPEN4_SHARE_ACCESS_BOTH) ||
7439 	    (args->share_access & ~OPEN4_SHARE_ACCESS_BOTH) ||
7440 	    (args->share_deny & ~OPEN4_SHARE_DENY_BOTH)) {
7441 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7442 		goto out;
7443 	}
7444 
7445 
7446 	/*
7447 	 * make sure attrset is zero before response is built.
7448 	 */
7449 	resp->attrset = 0;
7450 
7451 	switch (claim) {
7452 	case CLAIM_NULL:
7453 		rfs4_do_opennull(cs, req, args, oo, resp);
7454 		break;
7455 	case CLAIM_PREVIOUS:
7456 		rfs4_do_openprev(cs, req, args, oo, resp);
7457 		break;
7458 	case CLAIM_DELEGATE_CUR:
7459 		rfs4_do_opendelcur(cs, req, args, oo, resp);
7460 		break;
7461 	case CLAIM_DELEGATE_PREV:
7462 		rfs4_do_opendelprev(cs, req, args, oo, resp);
7463 		break;
7464 	default:
7465 		resp->status = NFS4ERR_INVAL;
7466 		break;
7467 	}
7468 
7469 out:
7470 	rfs4_client_rele(cp);
7471 
7472 	/* Catch sequence id handling here to make it a little easier */
7473 	switch (resp->status) {
7474 	case NFS4ERR_BADXDR:
7475 	case NFS4ERR_BAD_SEQID:
7476 	case NFS4ERR_BAD_STATEID:
7477 	case NFS4ERR_NOFILEHANDLE:
7478 	case NFS4ERR_RESOURCE:
7479 	case NFS4ERR_STALE_CLIENTID:
7480 	case NFS4ERR_STALE_STATEID:
7481 		/*
7482 		 * The protocol states that if any of these errors are
7483 		 * being returned, the sequence id should not be
7484 		 * incremented.  Any other return requires an
7485 		 * increment.
7486 		 */
7487 		break;
7488 	default:
7489 		/* Always update the lease in this case */
7490 		rfs4_update_lease(oo->ro_client);
7491 
7492 		/* Regular response - copy the result */
7493 		if (!replay)
7494 			rfs4_update_open_resp(oo, resop, &cs->fh);
7495 
7496 		/*
7497 		 * REPLAY case: Only if the previous response was OK
7498 		 * do we copy the filehandle.  If not OK, no
7499 		 * filehandle to copy.
7500 		 */
7501 		if (replay == TRUE &&
7502 		    resp->status == NFS4_OK &&
7503 		    oo->ro_reply_fh.nfs_fh4_val) {
7504 			/*
7505 			 * If this is a replay, we must restore the
7506 			 * current filehandle/vp to that of what was
7507 			 * returned originally.  Try our best to do
7508 			 * it.
7509 			 */
7510 			nfs_fh4_fmt_t *fh_fmtp =
7511 			    (nfs_fh4_fmt_t *)oo->ro_reply_fh.nfs_fh4_val;
7512 
7513 			cs->exi = checkexport4(&fh_fmtp->fh4_fsid,
7514 			    (fid_t *)&fh_fmtp->fh4_xlen, NULL);
7515 
7516 			if (cs->exi == NULL) {
7517 				resp->status = NFS4ERR_STALE;
7518 				goto finish;
7519 			}
7520 
7521 			VN_RELE(cs->vp);
7522 
7523 			cs->vp = nfs4_fhtovp(&oo->ro_reply_fh, cs->exi,
7524 			    &resp->status);
7525 
7526 			if (cs->vp == NULL)
7527 				goto finish;
7528 
7529 			nfs_fh4_copy(&oo->ro_reply_fh, &cs->fh);
7530 		}
7531 
7532 		/*
7533 		 * If this was a replay, no need to update the
7534 		 * sequence id. If the open_owner was not created on
7535 		 * this pass, then update.  The first use of an
7536 		 * open_owner will not bump the sequence id.
7537 		 */
7538 		if (replay == FALSE && !create)
7539 			rfs4_update_open_sequence(oo);
7540 		/*
7541 		 * If the client is receiving an error and the
7542 		 * open_owner needs to be confirmed, there is no way
7543 		 * to notify the client of this fact ignoring the fact
7544 		 * that the server has no method of returning a
7545 		 * stateid to confirm.  Therefore, the server needs to
7546 		 * mark this open_owner in a way as to avoid the
7547 		 * sequence id checking the next time the client uses
7548 		 * this open_owner.
7549 		 */
7550 		if (resp->status != NFS4_OK && oo->ro_need_confirm)
7551 			oo->ro_postpone_confirm = TRUE;
7552 		/*
7553 		 * If OK response then clear the postpone flag and
7554 		 * reset the sequence id to keep in sync with the
7555 		 * client.
7556 		 */
7557 		if (resp->status == NFS4_OK && oo->ro_postpone_confirm) {
7558 			oo->ro_postpone_confirm = FALSE;
7559 			oo->ro_open_seqid = args->seqid;
7560 		}
7561 		break;
7562 	}
7563 
7564 finish:
7565 	*cs->statusp = resp->status;
7566 
7567 	rfs4_sw_exit(&oo->ro_sw);
7568 	rfs4_openowner_rele(oo);
7569 
7570 end:
7571 	DTRACE_NFSV4_2(op__open__done, struct compound_state *, cs,
7572 	    OPEN4res *, resp);
7573 }
7574 
7575 /*ARGSUSED*/
7576 void
7577 rfs4_op_open_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
7578     struct svc_req *req, struct compound_state *cs)
7579 {
7580 	OPEN_CONFIRM4args *args = &argop->nfs_argop4_u.opopen_confirm;
7581 	OPEN_CONFIRM4res *resp = &resop->nfs_resop4_u.opopen_confirm;
7582 	rfs4_state_t *sp;
7583 	nfsstat4 status;
7584 
7585 	DTRACE_NFSV4_2(op__open__confirm__start, struct compound_state *, cs,
7586 	    OPEN_CONFIRM4args *, args);
7587 
7588 	if (cs->vp == NULL) {
7589 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7590 		goto out;
7591 	}
7592 
7593 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7594 	if (status != NFS4_OK) {
7595 		*cs->statusp = resp->status = status;
7596 		goto out;
7597 	}
7598 
7599 	/* Ensure specified filehandle matches */
7600 	if (cs->vp != sp->rs_finfo->rf_vp) {
7601 		rfs4_state_rele(sp);
7602 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7603 		goto out;
7604 	}
7605 
7606 	/* hold off other access to open_owner while we tinker */
7607 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7608 
7609 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7610 	case NFS4_CHECK_STATEID_OKAY:
7611 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7612 		    resop) != 0) {
7613 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7614 			break;
7615 		}
7616 		/*
7617 		 * If it is the appropriate stateid and determined to
7618 		 * be "OKAY" then this means that the stateid does not
7619 		 * need to be confirmed and the client is in error for
7620 		 * sending an OPEN_CONFIRM.
7621 		 */
7622 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7623 		break;
7624 	case NFS4_CHECK_STATEID_OLD:
7625 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7626 		break;
7627 	case NFS4_CHECK_STATEID_BAD:
7628 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7629 		break;
7630 	case NFS4_CHECK_STATEID_EXPIRED:
7631 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7632 		break;
7633 	case NFS4_CHECK_STATEID_CLOSED:
7634 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7635 		break;
7636 	case NFS4_CHECK_STATEID_REPLAY:
7637 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7638 		    resop)) {
7639 		case NFS4_CHKSEQ_OKAY:
7640 			/*
7641 			 * This is replayed stateid; if seqid matches
7642 			 * next expected, then client is using wrong seqid.
7643 			 */
7644 			/* fall through */
7645 		case NFS4_CHKSEQ_BAD:
7646 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7647 			break;
7648 		case NFS4_CHKSEQ_REPLAY:
7649 			/*
7650 			 * Note this case is the duplicate case so
7651 			 * resp->status is already set.
7652 			 */
7653 			*cs->statusp = resp->status;
7654 			rfs4_update_lease(sp->rs_owner->ro_client);
7655 			break;
7656 		}
7657 		break;
7658 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7659 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7660 		    resop) != NFS4_CHKSEQ_OKAY) {
7661 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7662 			break;
7663 		}
7664 		*cs->statusp = resp->status = NFS4_OK;
7665 
7666 		next_stateid(&sp->rs_stateid);
7667 		resp->open_stateid = sp->rs_stateid.stateid;
7668 		sp->rs_owner->ro_need_confirm = FALSE;
7669 		rfs4_update_lease(sp->rs_owner->ro_client);
7670 		rfs4_update_open_sequence(sp->rs_owner);
7671 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7672 		break;
7673 	default:
7674 		ASSERT(FALSE);
7675 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7676 		break;
7677 	}
7678 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7679 	rfs4_state_rele(sp);
7680 
7681 out:
7682 	DTRACE_NFSV4_2(op__open__confirm__done, struct compound_state *, cs,
7683 	    OPEN_CONFIRM4res *, resp);
7684 }
7685 
7686 /*ARGSUSED*/
7687 void
7688 rfs4_op_open_downgrade(nfs_argop4 *argop, nfs_resop4 *resop,
7689     struct svc_req *req, struct compound_state *cs)
7690 {
7691 	OPEN_DOWNGRADE4args *args = &argop->nfs_argop4_u.opopen_downgrade;
7692 	OPEN_DOWNGRADE4res *resp = &resop->nfs_resop4_u.opopen_downgrade;
7693 	uint32_t access = args->share_access;
7694 	uint32_t deny = args->share_deny;
7695 	nfsstat4 status;
7696 	rfs4_state_t *sp;
7697 	rfs4_file_t *fp;
7698 	int fflags = 0;
7699 
7700 	DTRACE_NFSV4_2(op__open__downgrade__start, struct compound_state *, cs,
7701 	    OPEN_DOWNGRADE4args *, args);
7702 
7703 	if (cs->vp == NULL) {
7704 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
7705 		goto out;
7706 	}
7707 
7708 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_VALID);
7709 	if (status != NFS4_OK) {
7710 		*cs->statusp = resp->status = status;
7711 		goto out;
7712 	}
7713 
7714 	/* Ensure specified filehandle matches */
7715 	if (cs->vp != sp->rs_finfo->rf_vp) {
7716 		rfs4_state_rele(sp);
7717 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7718 		goto out;
7719 	}
7720 
7721 	/* hold off other access to open_owner while we tinker */
7722 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
7723 
7724 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
7725 	case NFS4_CHECK_STATEID_OKAY:
7726 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7727 		    resop) != NFS4_CHKSEQ_OKAY) {
7728 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7729 			goto end;
7730 		}
7731 		break;
7732 	case NFS4_CHECK_STATEID_OLD:
7733 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7734 		goto end;
7735 	case NFS4_CHECK_STATEID_BAD:
7736 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7737 		goto end;
7738 	case NFS4_CHECK_STATEID_EXPIRED:
7739 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
7740 		goto end;
7741 	case NFS4_CHECK_STATEID_CLOSED:
7742 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
7743 		goto end;
7744 	case NFS4_CHECK_STATEID_UNCONFIRMED:
7745 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
7746 		goto end;
7747 	case NFS4_CHECK_STATEID_REPLAY:
7748 		/* Check the sequence id for the open owner */
7749 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
7750 		    resop)) {
7751 		case NFS4_CHKSEQ_OKAY:
7752 			/*
7753 			 * This is replayed stateid; if seqid matches
7754 			 * next expected, then client is using wrong seqid.
7755 			 */
7756 			/* fall through */
7757 		case NFS4_CHKSEQ_BAD:
7758 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
7759 			goto end;
7760 		case NFS4_CHKSEQ_REPLAY:
7761 			/*
7762 			 * Note this case is the duplicate case so
7763 			 * resp->status is already set.
7764 			 */
7765 			*cs->statusp = resp->status;
7766 			rfs4_update_lease(sp->rs_owner->ro_client);
7767 			goto end;
7768 		}
7769 		break;
7770 	default:
7771 		ASSERT(FALSE);
7772 		break;
7773 	}
7774 
7775 	rfs4_dbe_lock(sp->rs_dbe);
7776 	/*
7777 	 * Check that the new access modes and deny modes are valid.
7778 	 * Check that no invalid bits are set.
7779 	 */
7780 	if ((access & ~(OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) ||
7781 	    (deny & ~(OPEN4_SHARE_DENY_READ | OPEN4_SHARE_DENY_WRITE))) {
7782 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7783 		rfs4_update_open_sequence(sp->rs_owner);
7784 		rfs4_dbe_unlock(sp->rs_dbe);
7785 		goto end;
7786 	}
7787 
7788 	/*
7789 	 * The new modes must be a subset of the current modes and
7790 	 * the access must specify at least one mode. To test that
7791 	 * the new mode is a subset of the current modes we bitwise
7792 	 * AND them together and check that the result equals the new
7793 	 * mode. For example:
7794 	 * New mode, access == R and current mode, sp->rs_open_access  == RW
7795 	 * access & sp->rs_open_access == R == access, so the new access mode
7796 	 * is valid. Consider access == RW, sp->rs_open_access = R
7797 	 * access & sp->rs_open_access == R != access, so the new access mode
7798 	 * is invalid.
7799 	 */
7800 	if ((access & sp->rs_open_access) != access ||
7801 	    (deny & sp->rs_open_deny) != deny ||
7802 	    (access &
7803 	    (OPEN4_SHARE_ACCESS_READ | OPEN4_SHARE_ACCESS_WRITE)) == 0) {
7804 		*cs->statusp = resp->status = NFS4ERR_INVAL;
7805 		rfs4_update_open_sequence(sp->rs_owner);
7806 		rfs4_dbe_unlock(sp->rs_dbe);
7807 		goto end;
7808 	}
7809 
7810 	/*
7811 	 * Release any share locks associated with this stateID.
7812 	 * Strictly speaking, this violates the spec because the
7813 	 * spec effectively requires that open downgrade be atomic.
7814 	 * At present, fs_shrlock does not have this capability.
7815 	 */
7816 	(void) rfs4_unshare(sp);
7817 
7818 	status = rfs4_share(sp, access, deny);
7819 	if (status != NFS4_OK) {
7820 		*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
7821 		rfs4_update_open_sequence(sp->rs_owner);
7822 		rfs4_dbe_unlock(sp->rs_dbe);
7823 		goto end;
7824 	}
7825 
7826 	fp = sp->rs_finfo;
7827 	rfs4_dbe_lock(fp->rf_dbe);
7828 
7829 	/*
7830 	 * If the current mode has deny read and the new mode
7831 	 * does not, decrement the number of deny read mode bits
7832 	 * and if it goes to zero turn off the deny read bit
7833 	 * on the file.
7834 	 */
7835 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_READ) &&
7836 	    (deny & OPEN4_SHARE_DENY_READ) == 0) {
7837 		fp->rf_deny_read--;
7838 		if (fp->rf_deny_read == 0)
7839 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
7840 	}
7841 
7842 	/*
7843 	 * If the current mode has deny write and the new mode
7844 	 * does not, decrement the number of deny write mode bits
7845 	 * and if it goes to zero turn off the deny write bit
7846 	 * on the file.
7847 	 */
7848 	if ((sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) &&
7849 	    (deny & OPEN4_SHARE_DENY_WRITE) == 0) {
7850 		fp->rf_deny_write--;
7851 		if (fp->rf_deny_write == 0)
7852 			fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
7853 	}
7854 
7855 	/*
7856 	 * If the current mode has access read and the new mode
7857 	 * does not, decrement the number of access read mode bits
7858 	 * and if it goes to zero turn off the access read bit
7859 	 * on the file.  set fflags to FREAD for the call to
7860 	 * vn_open_downgrade().
7861 	 */
7862 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) &&
7863 	    (access & OPEN4_SHARE_ACCESS_READ) == 0) {
7864 		fp->rf_access_read--;
7865 		if (fp->rf_access_read == 0)
7866 			fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
7867 		fflags |= FREAD;
7868 	}
7869 
7870 	/*
7871 	 * If the current mode has access write and the new mode
7872 	 * does not, decrement the number of access write mode bits
7873 	 * and if it goes to zero turn off the access write bit
7874 	 * on the file.  set fflags to FWRITE for the call to
7875 	 * vn_open_downgrade().
7876 	 */
7877 	if ((sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) &&
7878 	    (access & OPEN4_SHARE_ACCESS_WRITE) == 0) {
7879 		fp->rf_access_write--;
7880 		if (fp->rf_access_write == 0)
7881 			fp->rf_share_deny &= ~OPEN4_SHARE_ACCESS_WRITE;
7882 		fflags |= FWRITE;
7883 	}
7884 
7885 	/* Check that the file is still accessible */
7886 	ASSERT(fp->rf_share_access);
7887 
7888 	rfs4_dbe_unlock(fp->rf_dbe);
7889 
7890 	/* now set the new open access and deny modes */
7891 	sp->rs_open_access = access;
7892 	sp->rs_open_deny = deny;
7893 
7894 	/*
7895 	 * we successfully downgraded the share lock, now we need to downgrade
7896 	 * the open. it is possible that the downgrade was only for a deny
7897 	 * mode and we have nothing else to do.
7898 	 */
7899 	if ((fflags & (FREAD|FWRITE)) != 0)
7900 		vn_open_downgrade(cs->vp, fflags);
7901 
7902 	/* Update the stateid */
7903 	next_stateid(&sp->rs_stateid);
7904 	resp->open_stateid = sp->rs_stateid.stateid;
7905 
7906 	rfs4_dbe_unlock(sp->rs_dbe);
7907 
7908 	*cs->statusp = resp->status = NFS4_OK;
7909 	/* Update the lease */
7910 	rfs4_update_lease(sp->rs_owner->ro_client);
7911 	/* And the sequence */
7912 	rfs4_update_open_sequence(sp->rs_owner);
7913 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
7914 
7915 end:
7916 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
7917 	rfs4_state_rele(sp);
7918 out:
7919 	DTRACE_NFSV4_2(op__open__downgrade__done, struct compound_state *, cs,
7920 	    OPEN_DOWNGRADE4res *, resp);
7921 }
7922 
7923 /*
7924  * The logic behind this function is detailed in the NFSv4 RFC in the
7925  * SETCLIENTID operation description under IMPLEMENTATION.  Refer to
7926  * that section for explicit guidance to server behavior for
7927  * SETCLIENTID.
7928  */
7929 void
7930 rfs4_op_setclientid(nfs_argop4 *argop, nfs_resop4 *resop,
7931     struct svc_req *req, struct compound_state *cs)
7932 {
7933 	SETCLIENTID4args *args = &argop->nfs_argop4_u.opsetclientid;
7934 	SETCLIENTID4res *res = &resop->nfs_resop4_u.opsetclientid;
7935 	rfs4_client_t *cp, *newcp, *cp_confirmed, *cp_unconfirmed;
7936 	rfs4_clntip_t *ci;
7937 	bool_t create;
7938 	char *addr, *netid;
7939 	int len;
7940 
7941 	DTRACE_NFSV4_2(op__setclientid__start, struct compound_state *, cs,
7942 	    SETCLIENTID4args *, args);
7943 retry:
7944 	newcp = cp_confirmed = cp_unconfirmed = NULL;
7945 
7946 	/*
7947 	 * Save the caller's IP address
7948 	 */
7949 	args->client.cl_addr =
7950 	    (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
7951 
7952 	/*
7953 	 * Record if it is a Solaris client that cannot handle referrals.
7954 	 */
7955 	if (strstr(args->client.id_val, "Solaris") &&
7956 	    !strstr(args->client.id_val, "+referrals")) {
7957 		/* Add a "yes, it's downrev" record */
7958 		create = TRUE;
7959 		ci = rfs4_find_clntip(args->client.cl_addr, &create);
7960 		ASSERT(ci != NULL);
7961 		rfs4_dbe_rele(ci->ri_dbe);
7962 	} else {
7963 		/* Remove any previous record */
7964 		rfs4_invalidate_clntip(args->client.cl_addr);
7965 	}
7966 
7967 	/*
7968 	 * In search of an EXISTING client matching the incoming
7969 	 * request to establish a new client identifier at the server
7970 	 */
7971 	create = TRUE;
7972 	cp = rfs4_findclient(&args->client, &create, NULL);
7973 
7974 	/* Should never happen */
7975 	ASSERT(cp != NULL);
7976 
7977 	if (cp == NULL) {
7978 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
7979 		goto out;
7980 	}
7981 
7982 	/*
7983 	 * Easiest case. Client identifier is newly created and is
7984 	 * unconfirmed.  Also note that for this case, no other
7985 	 * entries exist for the client identifier.  Nothing else to
7986 	 * check.  Just setup the response and respond.
7987 	 */
7988 	if (create) {
7989 		*cs->statusp = res->status = NFS4_OK;
7990 		res->SETCLIENTID4res_u.resok4.clientid = cp->rc_clientid;
7991 		res->SETCLIENTID4res_u.resok4.setclientid_confirm =
7992 		    cp->rc_confirm_verf;
7993 		/* Setup callback information; CB_NULL confirmation later */
7994 		rfs4_client_setcb(cp, &args->callback, args->callback_ident);
7995 
7996 		rfs4_client_rele(cp);
7997 		goto out;
7998 	}
7999 
8000 	/*
8001 	 * An existing, confirmed client may exist but it may not have
8002 	 * been active for at least one lease period.  If so, then
8003 	 * "close" the client and create a new client identifier
8004 	 */
8005 	if (rfs4_lease_expired(cp)) {
8006 		rfs4_client_close(cp);
8007 		goto retry;
8008 	}
8009 
8010 	if (cp->rc_need_confirm == TRUE)
8011 		cp_unconfirmed = cp;
8012 	else
8013 		cp_confirmed = cp;
8014 
8015 	cp = NULL;
8016 
8017 	/*
8018 	 * We have a confirmed client, now check for an
8019 	 * unconfimred entry
8020 	 */
8021 	if (cp_confirmed) {
8022 		/* If creds don't match then client identifier is inuse */
8023 		if (!creds_ok(cp_confirmed->rc_cr_set, req, cs)) {
8024 			rfs4_cbinfo_t *cbp;
8025 			/*
8026 			 * Some one else has established this client
8027 			 * id. Try and say * who they are. We will use
8028 			 * the call back address supplied by * the
8029 			 * first client.
8030 			 */
8031 			*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8032 
8033 			addr = netid = NULL;
8034 
8035 			cbp = &cp_confirmed->rc_cbinfo;
8036 			if (cbp->cb_callback.cb_location.r_addr &&
8037 			    cbp->cb_callback.cb_location.r_netid) {
8038 				cb_client4 *cbcp = &cbp->cb_callback;
8039 
8040 				len = strlen(cbcp->cb_location.r_addr)+1;
8041 				addr = kmem_alloc(len, KM_SLEEP);
8042 				bcopy(cbcp->cb_location.r_addr, addr, len);
8043 				len = strlen(cbcp->cb_location.r_netid)+1;
8044 				netid = kmem_alloc(len, KM_SLEEP);
8045 				bcopy(cbcp->cb_location.r_netid, netid, len);
8046 			}
8047 
8048 			res->SETCLIENTID4res_u.client_using.r_addr = addr;
8049 			res->SETCLIENTID4res_u.client_using.r_netid = netid;
8050 
8051 			rfs4_client_rele(cp_confirmed);
8052 		}
8053 
8054 		/*
8055 		 * Confirmed, creds match, and verifier matches; must
8056 		 * be an update of the callback info
8057 		 */
8058 		if (cp_confirmed->rc_nfs_client.verifier ==
8059 		    args->client.verifier) {
8060 			/* Setup callback information */
8061 			rfs4_client_setcb(cp_confirmed, &args->callback,
8062 			    args->callback_ident);
8063 
8064 			/* everything okay -- move ahead */
8065 			*cs->statusp = res->status = NFS4_OK;
8066 			res->SETCLIENTID4res_u.resok4.clientid =
8067 			    cp_confirmed->rc_clientid;
8068 
8069 			/* update the confirm_verifier and return it */
8070 			rfs4_client_scv_next(cp_confirmed);
8071 			res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8072 			    cp_confirmed->rc_confirm_verf;
8073 
8074 			rfs4_client_rele(cp_confirmed);
8075 			goto out;
8076 		}
8077 
8078 		/*
8079 		 * Creds match but the verifier doesn't.  Must search
8080 		 * for an unconfirmed client that would be replaced by
8081 		 * this request.
8082 		 */
8083 		create = FALSE;
8084 		cp_unconfirmed = rfs4_findclient(&args->client, &create,
8085 		    cp_confirmed);
8086 	}
8087 
8088 	/*
8089 	 * At this point, we have taken care of the brand new client
8090 	 * struct, INUSE case, update of an existing, and confirmed
8091 	 * client struct.
8092 	 */
8093 
8094 	/*
8095 	 * check to see if things have changed while we originally
8096 	 * picked up the client struct.  If they have, then return and
8097 	 * retry the processing of this SETCLIENTID request.
8098 	 */
8099 	if (cp_unconfirmed) {
8100 		rfs4_dbe_lock(cp_unconfirmed->rc_dbe);
8101 		if (!cp_unconfirmed->rc_need_confirm) {
8102 			rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8103 			rfs4_client_rele(cp_unconfirmed);
8104 			if (cp_confirmed)
8105 				rfs4_client_rele(cp_confirmed);
8106 			goto retry;
8107 		}
8108 		/* do away with the old unconfirmed one */
8109 		rfs4_dbe_invalidate(cp_unconfirmed->rc_dbe);
8110 		rfs4_dbe_unlock(cp_unconfirmed->rc_dbe);
8111 		rfs4_client_rele(cp_unconfirmed);
8112 		cp_unconfirmed = NULL;
8113 	}
8114 
8115 	/*
8116 	 * This search will temporarily hide the confirmed client
8117 	 * struct while a new client struct is created as the
8118 	 * unconfirmed one.
8119 	 */
8120 	create = TRUE;
8121 	newcp = rfs4_findclient(&args->client, &create, cp_confirmed);
8122 
8123 	ASSERT(newcp != NULL);
8124 
8125 	if (newcp == NULL) {
8126 		*cs->statusp = res->status = NFS4ERR_SERVERFAULT;
8127 		rfs4_client_rele(cp_confirmed);
8128 		goto out;
8129 	}
8130 
8131 	/*
8132 	 * If one was not created, then a similar request must be in
8133 	 * process so release and start over with this one
8134 	 */
8135 	if (create != TRUE) {
8136 		rfs4_client_rele(newcp);
8137 		if (cp_confirmed)
8138 			rfs4_client_rele(cp_confirmed);
8139 		goto retry;
8140 	}
8141 
8142 	*cs->statusp = res->status = NFS4_OK;
8143 	res->SETCLIENTID4res_u.resok4.clientid = newcp->rc_clientid;
8144 	res->SETCLIENTID4res_u.resok4.setclientid_confirm =
8145 	    newcp->rc_confirm_verf;
8146 	/* Setup callback information; CB_NULL confirmation later */
8147 	rfs4_client_setcb(newcp, &args->callback, args->callback_ident);
8148 
8149 	newcp->rc_cp_confirmed = cp_confirmed;
8150 
8151 	rfs4_client_rele(newcp);
8152 
8153 out:
8154 	DTRACE_NFSV4_2(op__setclientid__done, struct compound_state *, cs,
8155 	    SETCLIENTID4res *, res);
8156 }
8157 
8158 /*ARGSUSED*/
8159 void
8160 rfs4_op_setclientid_confirm(nfs_argop4 *argop, nfs_resop4 *resop,
8161     struct svc_req *req, struct compound_state *cs)
8162 {
8163 	SETCLIENTID_CONFIRM4args *args =
8164 	    &argop->nfs_argop4_u.opsetclientid_confirm;
8165 	SETCLIENTID_CONFIRM4res *res =
8166 	    &resop->nfs_resop4_u.opsetclientid_confirm;
8167 	rfs4_client_t *cp, *cptoclose = NULL;
8168 
8169 	DTRACE_NFSV4_2(op__setclientid__confirm__start,
8170 	    struct compound_state *, cs,
8171 	    SETCLIENTID_CONFIRM4args *, args);
8172 
8173 	*cs->statusp = res->status = NFS4_OK;
8174 
8175 	cp = rfs4_findclient_by_id(args->clientid, TRUE);
8176 
8177 	if (cp == NULL) {
8178 		*cs->statusp = res->status =
8179 		    rfs4_check_clientid(&args->clientid, 1);
8180 		goto out;
8181 	}
8182 
8183 	if (!creds_ok(cp, req, cs)) {
8184 		*cs->statusp = res->status = NFS4ERR_CLID_INUSE;
8185 		rfs4_client_rele(cp);
8186 		goto out;
8187 	}
8188 
8189 	/* If the verifier doesn't match, the record doesn't match */
8190 	if (cp->rc_confirm_verf != args->setclientid_confirm) {
8191 		*cs->statusp = res->status = NFS4ERR_STALE_CLIENTID;
8192 		rfs4_client_rele(cp);
8193 		goto out;
8194 	}
8195 
8196 	rfs4_dbe_lock(cp->rc_dbe);
8197 	cp->rc_need_confirm = FALSE;
8198 	if (cp->rc_cp_confirmed) {
8199 		cptoclose = cp->rc_cp_confirmed;
8200 		cptoclose->rc_ss_remove = 1;
8201 		cp->rc_cp_confirmed = NULL;
8202 	}
8203 
8204 	/*
8205 	 * Update the client's associated server instance, if it's changed
8206 	 * since the client was created.
8207 	 */
8208 	if (rfs4_servinst(cp) != rfs4_cur_servinst)
8209 		rfs4_servinst_assign(cp, rfs4_cur_servinst);
8210 
8211 	/*
8212 	 * Record clientid in stable storage.
8213 	 * Must be done after server instance has been assigned.
8214 	 */
8215 	rfs4_ss_clid(cp);
8216 
8217 	rfs4_dbe_unlock(cp->rc_dbe);
8218 
8219 	if (cptoclose)
8220 		/* don't need to rele, client_close does it */
8221 		rfs4_client_close(cptoclose);
8222 
8223 	/* If needed, initiate CB_NULL call for callback path */
8224 	rfs4_deleg_cb_check(cp);
8225 	rfs4_update_lease(cp);
8226 
8227 	/*
8228 	 * Check to see if client can perform reclaims
8229 	 */
8230 	rfs4_ss_chkclid(cp);
8231 
8232 	rfs4_client_rele(cp);
8233 
8234 out:
8235 	DTRACE_NFSV4_2(op__setclientid__confirm__done,
8236 	    struct compound_state *, cs,
8237 	    SETCLIENTID_CONFIRM4 *, res);
8238 }
8239 
8240 
8241 /*ARGSUSED*/
8242 void
8243 rfs4_op_close(nfs_argop4 *argop, nfs_resop4 *resop,
8244     struct svc_req *req, struct compound_state *cs)
8245 {
8246 	CLOSE4args *args = &argop->nfs_argop4_u.opclose;
8247 	CLOSE4res *resp = &resop->nfs_resop4_u.opclose;
8248 	rfs4_state_t *sp;
8249 	nfsstat4 status;
8250 
8251 	DTRACE_NFSV4_2(op__close__start, struct compound_state *, cs,
8252 	    CLOSE4args *, args);
8253 
8254 	if (cs->vp == NULL) {
8255 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8256 		goto out;
8257 	}
8258 
8259 	status = rfs4_get_state(&args->open_stateid, &sp, RFS4_DBS_INVALID);
8260 	if (status != NFS4_OK) {
8261 		*cs->statusp = resp->status = status;
8262 		goto out;
8263 	}
8264 
8265 	/* Ensure specified filehandle matches */
8266 	if (cs->vp != sp->rs_finfo->rf_vp) {
8267 		rfs4_state_rele(sp);
8268 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8269 		goto out;
8270 	}
8271 
8272 	/* hold off other access to open_owner while we tinker */
8273 	rfs4_sw_enter(&sp->rs_owner->ro_sw);
8274 
8275 	switch (rfs4_check_stateid_seqid(sp, &args->open_stateid)) {
8276 	case NFS4_CHECK_STATEID_OKAY:
8277 		if (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8278 		    resop) != NFS4_CHKSEQ_OKAY) {
8279 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8280 			goto end;
8281 		}
8282 		break;
8283 	case NFS4_CHECK_STATEID_OLD:
8284 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8285 		goto end;
8286 	case NFS4_CHECK_STATEID_BAD:
8287 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8288 		goto end;
8289 	case NFS4_CHECK_STATEID_EXPIRED:
8290 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8291 		goto end;
8292 	case NFS4_CHECK_STATEID_CLOSED:
8293 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8294 		goto end;
8295 	case NFS4_CHECK_STATEID_UNCONFIRMED:
8296 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8297 		goto end;
8298 	case NFS4_CHECK_STATEID_REPLAY:
8299 		/* Check the sequence id for the open owner */
8300 		switch (rfs4_check_open_seqid(args->seqid, sp->rs_owner,
8301 		    resop)) {
8302 		case NFS4_CHKSEQ_OKAY:
8303 			/*
8304 			 * This is replayed stateid; if seqid matches
8305 			 * next expected, then client is using wrong seqid.
8306 			 */
8307 			/* FALL THROUGH */
8308 		case NFS4_CHKSEQ_BAD:
8309 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8310 			goto end;
8311 		case NFS4_CHKSEQ_REPLAY:
8312 			/*
8313 			 * Note this case is the duplicate case so
8314 			 * resp->status is already set.
8315 			 */
8316 			*cs->statusp = resp->status;
8317 			rfs4_update_lease(sp->rs_owner->ro_client);
8318 			goto end;
8319 		}
8320 		break;
8321 	default:
8322 		ASSERT(FALSE);
8323 		break;
8324 	}
8325 
8326 	rfs4_dbe_lock(sp->rs_dbe);
8327 
8328 	/* Update the stateid. */
8329 	next_stateid(&sp->rs_stateid);
8330 	resp->open_stateid = sp->rs_stateid.stateid;
8331 
8332 	rfs4_dbe_unlock(sp->rs_dbe);
8333 
8334 	rfs4_update_lease(sp->rs_owner->ro_client);
8335 	rfs4_update_open_sequence(sp->rs_owner);
8336 	rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8337 
8338 	rfs4_state_close(sp, FALSE, FALSE, cs->cr);
8339 
8340 	*cs->statusp = resp->status = status;
8341 
8342 end:
8343 	rfs4_sw_exit(&sp->rs_owner->ro_sw);
8344 	rfs4_state_rele(sp);
8345 out:
8346 	DTRACE_NFSV4_2(op__close__done, struct compound_state *, cs,
8347 	    CLOSE4res *, resp);
8348 }
8349 
8350 /*
8351  * Manage the counts on the file struct and close all file locks
8352  */
8353 /*ARGSUSED*/
8354 void
8355 rfs4_release_share_lock_state(rfs4_state_t *sp, cred_t *cr,
8356     bool_t close_of_client)
8357 {
8358 	rfs4_file_t *fp = sp->rs_finfo;
8359 	rfs4_lo_state_t *lsp;
8360 	int fflags = 0;
8361 
8362 	/*
8363 	 * If this call is part of the larger closing down of client
8364 	 * state then it is just easier to release all locks
8365 	 * associated with this client instead of going through each
8366 	 * individual file and cleaning locks there.
8367 	 */
8368 	if (close_of_client) {
8369 		if (sp->rs_owner->ro_client->rc_unlksys_completed == FALSE &&
8370 		    !list_is_empty(&sp->rs_lostatelist) &&
8371 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID) {
8372 			/* Is the PxFS kernel module loaded? */
8373 			if (lm_remove_file_locks != NULL) {
8374 				int new_sysid;
8375 
8376 				/* Encode the cluster nodeid in new sysid */
8377 				new_sysid = sp->rs_owner->ro_client->rc_sysidt;
8378 				lm_set_nlmid_flk(&new_sysid);
8379 
8380 				/*
8381 				 * This PxFS routine removes file locks for a
8382 				 * client over all nodes of a cluster.
8383 				 */
8384 				NFS4_DEBUG(rfs4_debug, (CE_NOTE,
8385 				    "lm_remove_file_locks(sysid=0x%x)\n",
8386 				    new_sysid));
8387 				(*lm_remove_file_locks)(new_sysid);
8388 			} else {
8389 				struct flock64 flk;
8390 
8391 				/* Release all locks for this client */
8392 				flk.l_type = F_UNLKSYS;
8393 				flk.l_whence = 0;
8394 				flk.l_start = 0;
8395 				flk.l_len = 0;
8396 				flk.l_sysid =
8397 				    sp->rs_owner->ro_client->rc_sysidt;
8398 				flk.l_pid = 0;
8399 				(void) VOP_FRLOCK(sp->rs_finfo->rf_vp, F_SETLK,
8400 				    &flk, F_REMOTELOCK | FREAD | FWRITE,
8401 				    (u_offset_t)0, NULL, CRED(), NULL);
8402 			}
8403 
8404 			sp->rs_owner->ro_client->rc_unlksys_completed = TRUE;
8405 		}
8406 	}
8407 
8408 	/*
8409 	 * Release all locks on this file by this lock owner or at
8410 	 * least mark the locks as having been released
8411 	 */
8412 	for (lsp = list_head(&sp->rs_lostatelist); lsp != NULL;
8413 	    lsp = list_next(&sp->rs_lostatelist, lsp)) {
8414 		lsp->rls_locks_cleaned = TRUE;
8415 
8416 		/* Was this already taken care of above? */
8417 		if (!close_of_client &&
8418 		    sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8419 			(void) cleanlocks(sp->rs_finfo->rf_vp,
8420 			    lsp->rls_locker->rl_pid,
8421 			    lsp->rls_locker->rl_client->rc_sysidt);
8422 	}
8423 
8424 	/*
8425 	 * Release any shrlocks associated with this open state ID.
8426 	 * This must be done before the rfs4_state gets marked closed.
8427 	 */
8428 	if (sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID)
8429 		(void) rfs4_unshare(sp);
8430 
8431 	if (sp->rs_open_access) {
8432 		rfs4_dbe_lock(fp->rf_dbe);
8433 
8434 		/*
8435 		 * Decrement the count for each access and deny bit that this
8436 		 * state has contributed to the file.
8437 		 * If the file counts go to zero
8438 		 * clear the appropriate bit in the appropriate mask.
8439 		 */
8440 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_READ) {
8441 			fp->rf_access_read--;
8442 			fflags |= FREAD;
8443 			if (fp->rf_access_read == 0)
8444 				fp->rf_share_access &= ~OPEN4_SHARE_ACCESS_READ;
8445 		}
8446 		if (sp->rs_open_access & OPEN4_SHARE_ACCESS_WRITE) {
8447 			fp->rf_access_write--;
8448 			fflags |= FWRITE;
8449 			if (fp->rf_access_write == 0)
8450 				fp->rf_share_access &=
8451 				    ~OPEN4_SHARE_ACCESS_WRITE;
8452 		}
8453 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_READ) {
8454 			fp->rf_deny_read--;
8455 			if (fp->rf_deny_read == 0)
8456 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_READ;
8457 		}
8458 		if (sp->rs_open_deny & OPEN4_SHARE_DENY_WRITE) {
8459 			fp->rf_deny_write--;
8460 			if (fp->rf_deny_write == 0)
8461 				fp->rf_share_deny &= ~OPEN4_SHARE_DENY_WRITE;
8462 		}
8463 
8464 		(void) VOP_CLOSE(fp->rf_vp, fflags, 1, (offset_t)0, cr, NULL);
8465 
8466 		rfs4_dbe_unlock(fp->rf_dbe);
8467 
8468 		sp->rs_open_access = 0;
8469 		sp->rs_open_deny = 0;
8470 	}
8471 }
8472 
8473 /*
8474  * lock_denied: Fill in a LOCK4deneid structure given an flock64 structure.
8475  */
8476 static nfsstat4
8477 lock_denied(LOCK4denied *dp, struct flock64 *flk)
8478 {
8479 	rfs4_lockowner_t *lo;
8480 	rfs4_client_t *cp;
8481 	uint32_t len;
8482 
8483 	lo = rfs4_findlockowner_by_pid(flk->l_pid);
8484 	if (lo != NULL) {
8485 		cp = lo->rl_client;
8486 		if (rfs4_lease_expired(cp)) {
8487 			rfs4_lockowner_rele(lo);
8488 			rfs4_dbe_hold(cp->rc_dbe);
8489 			rfs4_client_close(cp);
8490 			return (NFS4ERR_EXPIRED);
8491 		}
8492 		dp->owner.clientid = lo->rl_owner.clientid;
8493 		len = lo->rl_owner.owner_len;
8494 		dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8495 		bcopy(lo->rl_owner.owner_val, dp->owner.owner_val, len);
8496 		dp->owner.owner_len = len;
8497 		rfs4_lockowner_rele(lo);
8498 		goto finish;
8499 	}
8500 
8501 	/*
8502 	 * Its not a NFS4 lock. We take advantage that the upper 32 bits
8503 	 * of the client id contain the boot time for a NFS4 lock. So we
8504 	 * fabricate and identity by setting clientid to the sysid, and
8505 	 * the lock owner to the pid.
8506 	 */
8507 	dp->owner.clientid = flk->l_sysid;
8508 	len = sizeof (pid_t);
8509 	dp->owner.owner_len = len;
8510 	dp->owner.owner_val = kmem_alloc(len, KM_SLEEP);
8511 	bcopy(&flk->l_pid, dp->owner.owner_val, len);
8512 finish:
8513 	dp->offset = flk->l_start;
8514 	dp->length = flk->l_len;
8515 
8516 	if (flk->l_type == F_RDLCK)
8517 		dp->locktype = READ_LT;
8518 	else if (flk->l_type == F_WRLCK)
8519 		dp->locktype = WRITE_LT;
8520 	else
8521 		return (NFS4ERR_INVAL);	/* no mapping from POSIX ltype to v4 */
8522 
8523 	return (NFS4_OK);
8524 }
8525 
8526 static int
8527 setlock(vnode_t *vp, struct flock64 *flock, int flag, cred_t *cred)
8528 {
8529 	int error;
8530 	struct flock64 flk;
8531 	int i;
8532 	clock_t delaytime;
8533 	int cmd;
8534 
8535 	cmd = nbl_need_check(vp) ? F_SETLK_NBMAND : F_SETLK;
8536 retry:
8537 	delaytime = MSEC_TO_TICK_ROUNDUP(rfs4_lock_delay);
8538 
8539 	for (i = 0; i < rfs4_maxlock_tries; i++) {
8540 		LOCK_PRINT(rfs4_debug, "setlock", cmd, flock);
8541 		error = VOP_FRLOCK(vp, cmd,
8542 		    flock, flag, (u_offset_t)0, NULL, cred, NULL);
8543 
8544 		if (error != EAGAIN && error != EACCES)
8545 			break;
8546 
8547 		if (i < rfs4_maxlock_tries - 1) {
8548 			delay(delaytime);
8549 			delaytime *= 2;
8550 		}
8551 	}
8552 
8553 	if (error == EAGAIN || error == EACCES) {
8554 		/* Get the owner of the lock */
8555 		flk = *flock;
8556 		LOCK_PRINT(rfs4_debug, "setlock", F_GETLK, &flk);
8557 		if (VOP_FRLOCK(vp, F_GETLK, &flk, flag,
8558 		    (u_offset_t)0, NULL, cred, NULL) == 0) {
8559 			if (flk.l_type == F_UNLCK) {
8560 				/* No longer locked, retry */
8561 				goto retry;
8562 			}
8563 			*flock = flk;
8564 			LOCK_PRINT(rfs4_debug, "setlock(blocking lock)",
8565 			    F_GETLK, &flk);
8566 		}
8567 	}
8568 
8569 	return (error);
8570 }
8571 
8572 /*ARGSUSED*/
8573 static nfsstat4
8574 rfs4_do_lock(rfs4_lo_state_t *lsp, nfs_lock_type4 locktype,
8575     offset4 offset, length4 length, cred_t *cred, nfs_resop4 *resop)
8576 {
8577 	nfsstat4 status;
8578 	rfs4_lockowner_t *lo = lsp->rls_locker;
8579 	rfs4_state_t *sp = lsp->rls_state;
8580 	struct flock64 flock;
8581 	int16_t ltype;
8582 	int flag;
8583 	int error;
8584 	sysid_t sysid;
8585 	LOCK4res *lres;
8586 
8587 	if (rfs4_lease_expired(lo->rl_client)) {
8588 		return (NFS4ERR_EXPIRED);
8589 	}
8590 
8591 	if ((status = rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
8592 		return (status);
8593 
8594 	/* Check for zero length. To lock to end of file use all ones for V4 */
8595 	if (length == 0)
8596 		return (NFS4ERR_INVAL);
8597 	else if (length == (length4)(~0))
8598 		length = 0;		/* Posix to end of file  */
8599 
8600 retry:
8601 	rfs4_dbe_lock(sp->rs_dbe);
8602 	if (sp->rs_closed) {
8603 		rfs4_dbe_unlock(sp->rs_dbe);
8604 		return (NFS4ERR_OLD_STATEID);
8605 	}
8606 
8607 	if (resop->resop != OP_LOCKU) {
8608 		switch (locktype) {
8609 		case READ_LT:
8610 		case READW_LT:
8611 			if ((sp->rs_share_access
8612 			    & OPEN4_SHARE_ACCESS_READ) == 0) {
8613 				rfs4_dbe_unlock(sp->rs_dbe);
8614 
8615 				return (NFS4ERR_OPENMODE);
8616 			}
8617 			ltype = F_RDLCK;
8618 			break;
8619 		case WRITE_LT:
8620 		case WRITEW_LT:
8621 			if ((sp->rs_share_access
8622 			    & OPEN4_SHARE_ACCESS_WRITE) == 0) {
8623 				rfs4_dbe_unlock(sp->rs_dbe);
8624 
8625 				return (NFS4ERR_OPENMODE);
8626 			}
8627 			ltype = F_WRLCK;
8628 			break;
8629 		}
8630 	} else
8631 		ltype = F_UNLCK;
8632 
8633 	flock.l_type = ltype;
8634 	flock.l_whence = 0;		/* SEEK_SET */
8635 	flock.l_start = offset;
8636 	flock.l_len = length;
8637 	flock.l_sysid = sysid;
8638 	flock.l_pid = lsp->rls_locker->rl_pid;
8639 
8640 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
8641 	if (flock.l_len < 0 || flock.l_start < 0) {
8642 		rfs4_dbe_unlock(sp->rs_dbe);
8643 		return (NFS4ERR_INVAL);
8644 	}
8645 
8646 	/*
8647 	 * N.B. FREAD has the same value as OPEN4_SHARE_ACCESS_READ and
8648 	 * FWRITE has the same value as OPEN4_SHARE_ACCESS_WRITE.
8649 	 */
8650 	flag = (int)sp->rs_share_access | F_REMOTELOCK;
8651 
8652 	error = setlock(sp->rs_finfo->rf_vp, &flock, flag, cred);
8653 	if (error == 0) {
8654 		rfs4_dbe_lock(lsp->rls_dbe);
8655 		next_stateid(&lsp->rls_lockid);
8656 		rfs4_dbe_unlock(lsp->rls_dbe);
8657 	}
8658 
8659 	rfs4_dbe_unlock(sp->rs_dbe);
8660 
8661 	/*
8662 	 * N.B. We map error values to nfsv4 errors. This is differrent
8663 	 * than puterrno4 routine.
8664 	 */
8665 	switch (error) {
8666 	case 0:
8667 		status = NFS4_OK;
8668 		break;
8669 	case EAGAIN:
8670 	case EACCES:		/* Old value */
8671 		/* Can only get here if op is OP_LOCK */
8672 		ASSERT(resop->resop == OP_LOCK);
8673 		lres = &resop->nfs_resop4_u.oplock;
8674 		status = NFS4ERR_DENIED;
8675 		if (lock_denied(&lres->LOCK4res_u.denied, &flock)
8676 		    == NFS4ERR_EXPIRED)
8677 			goto retry;
8678 		break;
8679 	case ENOLCK:
8680 		status = NFS4ERR_DELAY;
8681 		break;
8682 	case EOVERFLOW:
8683 		status = NFS4ERR_INVAL;
8684 		break;
8685 	case EINVAL:
8686 		status = NFS4ERR_NOTSUPP;
8687 		break;
8688 	default:
8689 		status = NFS4ERR_SERVERFAULT;
8690 		break;
8691 	}
8692 
8693 	return (status);
8694 }
8695 
8696 /*ARGSUSED*/
8697 void
8698 rfs4_op_lock(nfs_argop4 *argop, nfs_resop4 *resop,
8699     struct svc_req *req, struct compound_state *cs)
8700 {
8701 	LOCK4args *args = &argop->nfs_argop4_u.oplock;
8702 	LOCK4res *resp = &resop->nfs_resop4_u.oplock;
8703 	nfsstat4 status;
8704 	stateid4 *stateid;
8705 	rfs4_lockowner_t *lo;
8706 	rfs4_client_t *cp;
8707 	rfs4_state_t *sp = NULL;
8708 	rfs4_lo_state_t *lsp = NULL;
8709 	bool_t ls_sw_held = FALSE;
8710 	bool_t create = TRUE;
8711 	bool_t lcreate = TRUE;
8712 	bool_t dup_lock = FALSE;
8713 	int rc;
8714 
8715 	DTRACE_NFSV4_2(op__lock__start, struct compound_state *, cs,
8716 	    LOCK4args *, args);
8717 
8718 	if (cs->vp == NULL) {
8719 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
8720 		DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8721 		    cs, LOCK4res *, resp);
8722 		return;
8723 	}
8724 
8725 	if (args->locker.new_lock_owner) {
8726 		/* Create a new lockowner for this instance */
8727 		open_to_lock_owner4 *olo = &args->locker.locker4_u.open_owner;
8728 
8729 		NFS4_DEBUG(rfs4_debug, (CE_NOTE, "Creating new lock owner"));
8730 
8731 		stateid = &olo->open_stateid;
8732 		status = rfs4_get_state(stateid, &sp, RFS4_DBS_VALID);
8733 		if (status != NFS4_OK) {
8734 			NFS4_DEBUG(rfs4_debug,
8735 			    (CE_NOTE, "Get state failed in lock %d", status));
8736 			*cs->statusp = resp->status = status;
8737 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8738 			    cs, LOCK4res *, resp);
8739 			return;
8740 		}
8741 
8742 		/* Ensure specified filehandle matches */
8743 		if (cs->vp != sp->rs_finfo->rf_vp) {
8744 			rfs4_state_rele(sp);
8745 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8746 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8747 			    cs, LOCK4res *, resp);
8748 			return;
8749 		}
8750 
8751 		/* hold off other access to open_owner while we tinker */
8752 		rfs4_sw_enter(&sp->rs_owner->ro_sw);
8753 
8754 		switch (rc = rfs4_check_stateid_seqid(sp, stateid)) {
8755 		case NFS4_CHECK_STATEID_OLD:
8756 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8757 			goto end;
8758 		case NFS4_CHECK_STATEID_BAD:
8759 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8760 			goto end;
8761 		case NFS4_CHECK_STATEID_EXPIRED:
8762 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8763 			goto end;
8764 		case NFS4_CHECK_STATEID_UNCONFIRMED:
8765 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8766 			goto end;
8767 		case NFS4_CHECK_STATEID_CLOSED:
8768 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8769 			goto end;
8770 		case NFS4_CHECK_STATEID_OKAY:
8771 		case NFS4_CHECK_STATEID_REPLAY:
8772 			switch (rfs4_check_olo_seqid(olo->open_seqid,
8773 			    sp->rs_owner, resop)) {
8774 			case NFS4_CHKSEQ_OKAY:
8775 				if (rc == NFS4_CHECK_STATEID_OKAY)
8776 					break;
8777 				/*
8778 				 * This is replayed stateid; if seqid
8779 				 * matches next expected, then client
8780 				 * is using wrong seqid.
8781 				 */
8782 				/* FALLTHROUGH */
8783 			case NFS4_CHKSEQ_BAD:
8784 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8785 				goto end;
8786 			case NFS4_CHKSEQ_REPLAY:
8787 				/* This is a duplicate LOCK request */
8788 				dup_lock = TRUE;
8789 
8790 				/*
8791 				 * For a duplicate we do not want to
8792 				 * create a new lockowner as it should
8793 				 * already exist.
8794 				 * Turn off the lockowner create flag.
8795 				 */
8796 				lcreate = FALSE;
8797 			}
8798 			break;
8799 		}
8800 
8801 		lo = rfs4_findlockowner(&olo->lock_owner, &lcreate);
8802 		if (lo == NULL) {
8803 			NFS4_DEBUG(rfs4_debug,
8804 			    (CE_NOTE, "rfs4_op_lock: no lock owner"));
8805 			*cs->statusp = resp->status = NFS4ERR_RESOURCE;
8806 			goto end;
8807 		}
8808 
8809 		lsp = rfs4_findlo_state_by_owner(lo, sp, &create);
8810 		if (lsp == NULL) {
8811 			rfs4_update_lease(sp->rs_owner->ro_client);
8812 			/*
8813 			 * Only update theh open_seqid if this is not
8814 			 * a duplicate request
8815 			 */
8816 			if (dup_lock == FALSE) {
8817 				rfs4_update_open_sequence(sp->rs_owner);
8818 			}
8819 
8820 			NFS4_DEBUG(rfs4_debug,
8821 			    (CE_NOTE, "rfs4_op_lock: no state"));
8822 			*cs->statusp = resp->status = NFS4ERR_SERVERFAULT;
8823 			rfs4_update_open_resp(sp->rs_owner, resop, NULL);
8824 			rfs4_lockowner_rele(lo);
8825 			goto end;
8826 		}
8827 
8828 		/*
8829 		 * This is the new_lock_owner branch and the client is
8830 		 * supposed to be associating a new lock_owner with
8831 		 * the open file at this point.  If we find that a
8832 		 * lock_owner/state association already exists and a
8833 		 * successful LOCK request was returned to the client,
8834 		 * an error is returned to the client since this is
8835 		 * not appropriate.  The client should be using the
8836 		 * existing lock_owner branch.
8837 		 */
8838 		if (dup_lock == FALSE && create == FALSE) {
8839 			if (lsp->rls_lock_completed == TRUE) {
8840 				*cs->statusp =
8841 				    resp->status = NFS4ERR_BAD_SEQID;
8842 				rfs4_lockowner_rele(lo);
8843 				goto end;
8844 			}
8845 		}
8846 
8847 		rfs4_update_lease(sp->rs_owner->ro_client);
8848 
8849 		/*
8850 		 * Only update theh open_seqid if this is not
8851 		 * a duplicate request
8852 		 */
8853 		if (dup_lock == FALSE) {
8854 			rfs4_update_open_sequence(sp->rs_owner);
8855 		}
8856 
8857 		/*
8858 		 * If this is a duplicate lock request, just copy the
8859 		 * previously saved reply and return.
8860 		 */
8861 		if (dup_lock == TRUE) {
8862 			/* verify that lock_seqid's match */
8863 			if (lsp->rls_seqid != olo->lock_seqid) {
8864 				NFS4_DEBUG(rfs4_debug,
8865 				    (CE_NOTE, "rfs4_op_lock: Dup-Lock seqid bad"
8866 				    "lsp->seqid=%d old->seqid=%d",
8867 				    lsp->rls_seqid, olo->lock_seqid));
8868 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8869 			} else {
8870 				rfs4_copy_reply(resop, &lsp->rls_reply);
8871 				/*
8872 				 * Make sure to copy the just
8873 				 * retrieved reply status into the
8874 				 * overall compound status
8875 				 */
8876 				*cs->statusp = resp->status;
8877 			}
8878 			rfs4_lockowner_rele(lo);
8879 			goto end;
8880 		}
8881 
8882 		rfs4_dbe_lock(lsp->rls_dbe);
8883 
8884 		/* Make sure to update the lock sequence id */
8885 		lsp->rls_seqid = olo->lock_seqid;
8886 
8887 		NFS4_DEBUG(rfs4_debug,
8888 		    (CE_NOTE, "Lock seqid established as %d", lsp->rls_seqid));
8889 
8890 		/*
8891 		 * This is used to signify the newly created lockowner
8892 		 * stateid and its sequence number.  The checks for
8893 		 * sequence number and increment don't occur on the
8894 		 * very first lock request for a lockowner.
8895 		 */
8896 		lsp->rls_skip_seqid_check = TRUE;
8897 
8898 		/* hold off other access to lsp while we tinker */
8899 		rfs4_sw_enter(&lsp->rls_sw);
8900 		ls_sw_held = TRUE;
8901 
8902 		rfs4_dbe_unlock(lsp->rls_dbe);
8903 
8904 		rfs4_lockowner_rele(lo);
8905 	} else {
8906 		stateid = &args->locker.locker4_u.lock_owner.lock_stateid;
8907 		/* get lsp and hold the lock on the underlying file struct */
8908 		if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE))
8909 		    != NFS4_OK) {
8910 			*cs->statusp = resp->status = status;
8911 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8912 			    cs, LOCK4res *, resp);
8913 			return;
8914 		}
8915 		create = FALSE;	/* We didn't create lsp */
8916 
8917 		/* Ensure specified filehandle matches */
8918 		if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
8919 			rfs4_lo_state_rele(lsp, TRUE);
8920 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8921 			DTRACE_NFSV4_2(op__lock__done, struct compound_state *,
8922 			    cs, LOCK4res *, resp);
8923 			return;
8924 		}
8925 
8926 		/* hold off other access to lsp while we tinker */
8927 		rfs4_sw_enter(&lsp->rls_sw);
8928 		ls_sw_held = TRUE;
8929 
8930 		switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
8931 		/*
8932 		 * The stateid looks like it was okay (expected to be
8933 		 * the next one)
8934 		 */
8935 		case NFS4_CHECK_STATEID_OKAY:
8936 			/*
8937 			 * The sequence id is now checked.  Determine
8938 			 * if this is a replay or if it is in the
8939 			 * expected (next) sequence.  In the case of a
8940 			 * replay, there are two replay conditions
8941 			 * that may occur.  The first is the normal
8942 			 * condition where a LOCK is done with a
8943 			 * NFS4_OK response and the stateid is
8944 			 * updated.  That case is handled below when
8945 			 * the stateid is identified as a REPLAY.  The
8946 			 * second is the case where an error is
8947 			 * returned, like NFS4ERR_DENIED, and the
8948 			 * sequence number is updated but the stateid
8949 			 * is not updated.  This second case is dealt
8950 			 * with here.  So it may seem odd that the
8951 			 * stateid is okay but the sequence id is a
8952 			 * replay but it is okay.
8953 			 */
8954 			switch (rfs4_check_lock_seqid(
8955 			    args->locker.locker4_u.lock_owner.lock_seqid,
8956 			    lsp, resop)) {
8957 			case NFS4_CHKSEQ_REPLAY:
8958 				if (resp->status != NFS4_OK) {
8959 					/*
8960 					 * Here is our replay and need
8961 					 * to verify that the last
8962 					 * response was an error.
8963 					 */
8964 					*cs->statusp = resp->status;
8965 					goto end;
8966 				}
8967 				/*
8968 				 * This is done since the sequence id
8969 				 * looked like a replay but it didn't
8970 				 * pass our check so a BAD_SEQID is
8971 				 * returned as a result.
8972 				 */
8973 				/*FALLTHROUGH*/
8974 			case NFS4_CHKSEQ_BAD:
8975 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
8976 				goto end;
8977 			case NFS4_CHKSEQ_OKAY:
8978 				/* Everything looks okay move ahead */
8979 				break;
8980 			}
8981 			break;
8982 		case NFS4_CHECK_STATEID_OLD:
8983 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8984 			goto end;
8985 		case NFS4_CHECK_STATEID_BAD:
8986 			*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
8987 			goto end;
8988 		case NFS4_CHECK_STATEID_EXPIRED:
8989 			*cs->statusp = resp->status = NFS4ERR_EXPIRED;
8990 			goto end;
8991 		case NFS4_CHECK_STATEID_CLOSED:
8992 			*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
8993 			goto end;
8994 		case NFS4_CHECK_STATEID_REPLAY:
8995 			switch (rfs4_check_lock_seqid(
8996 			    args->locker.locker4_u.lock_owner.lock_seqid,
8997 			    lsp, resop)) {
8998 			case NFS4_CHKSEQ_OKAY:
8999 				/*
9000 				 * This is a replayed stateid; if
9001 				 * seqid matches the next expected,
9002 				 * then client is using wrong seqid.
9003 				 */
9004 			case NFS4_CHKSEQ_BAD:
9005 				*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9006 				goto end;
9007 			case NFS4_CHKSEQ_REPLAY:
9008 				rfs4_update_lease(lsp->rls_locker->rl_client);
9009 				*cs->statusp = status = resp->status;
9010 				goto end;
9011 			}
9012 			break;
9013 		default:
9014 			ASSERT(FALSE);
9015 			break;
9016 		}
9017 
9018 		rfs4_update_lock_sequence(lsp);
9019 		rfs4_update_lease(lsp->rls_locker->rl_client);
9020 	}
9021 
9022 	/*
9023 	 * NFS4 only allows locking on regular files, so
9024 	 * verify type of object.
9025 	 */
9026 	if (cs->vp->v_type != VREG) {
9027 		if (cs->vp->v_type == VDIR)
9028 			status = NFS4ERR_ISDIR;
9029 		else
9030 			status = NFS4ERR_INVAL;
9031 		goto out;
9032 	}
9033 
9034 	cp = lsp->rls_state->rs_owner->ro_client;
9035 
9036 	if (rfs4_clnt_in_grace(cp) && !args->reclaim) {
9037 		status = NFS4ERR_GRACE;
9038 		goto out;
9039 	}
9040 
9041 	if (rfs4_clnt_in_grace(cp) && args->reclaim && !cp->rc_can_reclaim) {
9042 		status = NFS4ERR_NO_GRACE;
9043 		goto out;
9044 	}
9045 
9046 	if (!rfs4_clnt_in_grace(cp) && args->reclaim) {
9047 		status = NFS4ERR_NO_GRACE;
9048 		goto out;
9049 	}
9050 
9051 	if (lsp->rls_state->rs_finfo->rf_dinfo.rd_dtype == OPEN_DELEGATE_WRITE)
9052 		cs->deleg = TRUE;
9053 
9054 	status = rfs4_do_lock(lsp, args->locktype,
9055 	    args->offset, args->length, cs->cr, resop);
9056 
9057 out:
9058 	lsp->rls_skip_seqid_check = FALSE;
9059 
9060 	*cs->statusp = resp->status = status;
9061 
9062 	if (status == NFS4_OK) {
9063 		resp->LOCK4res_u.lock_stateid = lsp->rls_lockid.stateid;
9064 		lsp->rls_lock_completed = TRUE;
9065 	}
9066 	/*
9067 	 * Only update the "OPEN" response here if this was a new
9068 	 * lock_owner
9069 	 */
9070 	if (sp)
9071 		rfs4_update_open_resp(sp->rs_owner, resop, NULL);
9072 
9073 	rfs4_update_lock_resp(lsp, resop);
9074 
9075 end:
9076 	if (lsp) {
9077 		if (ls_sw_held)
9078 			rfs4_sw_exit(&lsp->rls_sw);
9079 		/*
9080 		 * If an sp obtained, then the lsp does not represent
9081 		 * a lock on the file struct.
9082 		 */
9083 		if (sp != NULL)
9084 			rfs4_lo_state_rele(lsp, FALSE);
9085 		else
9086 			rfs4_lo_state_rele(lsp, TRUE);
9087 	}
9088 	if (sp) {
9089 		rfs4_sw_exit(&sp->rs_owner->ro_sw);
9090 		rfs4_state_rele(sp);
9091 	}
9092 
9093 	DTRACE_NFSV4_2(op__lock__done, struct compound_state *, cs,
9094 	    LOCK4res *, resp);
9095 }
9096 
9097 /* free function for LOCK/LOCKT */
9098 static void
9099 lock_denied_free(nfs_resop4 *resop)
9100 {
9101 	LOCK4denied *dp = NULL;
9102 
9103 	switch (resop->resop) {
9104 	case OP_LOCK:
9105 		if (resop->nfs_resop4_u.oplock.status == NFS4ERR_DENIED)
9106 			dp = &resop->nfs_resop4_u.oplock.LOCK4res_u.denied;
9107 		break;
9108 	case OP_LOCKT:
9109 		if (resop->nfs_resop4_u.oplockt.status == NFS4ERR_DENIED)
9110 			dp = &resop->nfs_resop4_u.oplockt.denied;
9111 		break;
9112 	default:
9113 		break;
9114 	}
9115 
9116 	if (dp)
9117 		kmem_free(dp->owner.owner_val, dp->owner.owner_len);
9118 }
9119 
9120 /*ARGSUSED*/
9121 void
9122 rfs4_op_locku(nfs_argop4 *argop, nfs_resop4 *resop,
9123     struct svc_req *req, struct compound_state *cs)
9124 {
9125 	LOCKU4args *args = &argop->nfs_argop4_u.oplocku;
9126 	LOCKU4res *resp = &resop->nfs_resop4_u.oplocku;
9127 	nfsstat4 status;
9128 	stateid4 *stateid = &args->lock_stateid;
9129 	rfs4_lo_state_t *lsp;
9130 
9131 	DTRACE_NFSV4_2(op__locku__start, struct compound_state *, cs,
9132 	    LOCKU4args *, args);
9133 
9134 	if (cs->vp == NULL) {
9135 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9136 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9137 		    LOCKU4res *, resp);
9138 		return;
9139 	}
9140 
9141 	if ((status = rfs4_get_lo_state(stateid, &lsp, TRUE)) != NFS4_OK) {
9142 		*cs->statusp = resp->status = status;
9143 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9144 		    LOCKU4res *, resp);
9145 		return;
9146 	}
9147 
9148 	/* Ensure specified filehandle matches */
9149 	if (cs->vp != lsp->rls_state->rs_finfo->rf_vp) {
9150 		rfs4_lo_state_rele(lsp, TRUE);
9151 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9152 		DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9153 		    LOCKU4res *, resp);
9154 		return;
9155 	}
9156 
9157 	/* hold off other access to lsp while we tinker */
9158 	rfs4_sw_enter(&lsp->rls_sw);
9159 
9160 	switch (rfs4_check_lo_stateid_seqid(lsp, stateid)) {
9161 	case NFS4_CHECK_STATEID_OKAY:
9162 		if (rfs4_check_lock_seqid(args->seqid, lsp, resop)
9163 		    != NFS4_CHKSEQ_OKAY) {
9164 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9165 			goto end;
9166 		}
9167 		break;
9168 	case NFS4_CHECK_STATEID_OLD:
9169 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9170 		goto end;
9171 	case NFS4_CHECK_STATEID_BAD:
9172 		*cs->statusp = resp->status = NFS4ERR_BAD_STATEID;
9173 		goto end;
9174 	case NFS4_CHECK_STATEID_EXPIRED:
9175 		*cs->statusp = resp->status = NFS4ERR_EXPIRED;
9176 		goto end;
9177 	case NFS4_CHECK_STATEID_CLOSED:
9178 		*cs->statusp = resp->status = NFS4ERR_OLD_STATEID;
9179 		goto end;
9180 	case NFS4_CHECK_STATEID_REPLAY:
9181 		switch (rfs4_check_lock_seqid(args->seqid, lsp, resop)) {
9182 		case NFS4_CHKSEQ_OKAY:
9183 				/*
9184 				 * This is a replayed stateid; if
9185 				 * seqid matches the next expected,
9186 				 * then client is using wrong seqid.
9187 				 */
9188 		case NFS4_CHKSEQ_BAD:
9189 			*cs->statusp = resp->status = NFS4ERR_BAD_SEQID;
9190 			goto end;
9191 		case NFS4_CHKSEQ_REPLAY:
9192 			rfs4_update_lease(lsp->rls_locker->rl_client);
9193 			*cs->statusp = status = resp->status;
9194 			goto end;
9195 		}
9196 		break;
9197 	default:
9198 		ASSERT(FALSE);
9199 		break;
9200 	}
9201 
9202 	rfs4_update_lock_sequence(lsp);
9203 	rfs4_update_lease(lsp->rls_locker->rl_client);
9204 
9205 	/*
9206 	 * NFS4 only allows locking on regular files, so
9207 	 * verify type of object.
9208 	 */
9209 	if (cs->vp->v_type != VREG) {
9210 		if (cs->vp->v_type == VDIR)
9211 			status = NFS4ERR_ISDIR;
9212 		else
9213 			status = NFS4ERR_INVAL;
9214 		goto out;
9215 	}
9216 
9217 	if (rfs4_clnt_in_grace(lsp->rls_state->rs_owner->ro_client)) {
9218 		status = NFS4ERR_GRACE;
9219 		goto out;
9220 	}
9221 
9222 	status = rfs4_do_lock(lsp, args->locktype,
9223 	    args->offset, args->length, cs->cr, resop);
9224 
9225 out:
9226 	*cs->statusp = resp->status = status;
9227 
9228 	if (status == NFS4_OK)
9229 		resp->lock_stateid = lsp->rls_lockid.stateid;
9230 
9231 	rfs4_update_lock_resp(lsp, resop);
9232 
9233 end:
9234 	rfs4_sw_exit(&lsp->rls_sw);
9235 	rfs4_lo_state_rele(lsp, TRUE);
9236 
9237 	DTRACE_NFSV4_2(op__locku__done, struct compound_state *, cs,
9238 	    LOCKU4res *, resp);
9239 }
9240 
9241 /*
9242  * LOCKT is a best effort routine, the client can not be guaranteed that
9243  * the status return is still in effect by the time the reply is received.
9244  * They are numerous race conditions in this routine, but we are not required
9245  * and can not be accurate.
9246  */
9247 /*ARGSUSED*/
9248 void
9249 rfs4_op_lockt(nfs_argop4 *argop, nfs_resop4 *resop,
9250     struct svc_req *req, struct compound_state *cs)
9251 {
9252 	LOCKT4args *args = &argop->nfs_argop4_u.oplockt;
9253 	LOCKT4res *resp = &resop->nfs_resop4_u.oplockt;
9254 	rfs4_lockowner_t *lo;
9255 	rfs4_client_t *cp;
9256 	bool_t create = FALSE;
9257 	struct flock64 flk;
9258 	int error;
9259 	int flag = FREAD | FWRITE;
9260 	int ltype;
9261 	length4 posix_length;
9262 	sysid_t sysid;
9263 	pid_t pid;
9264 
9265 	DTRACE_NFSV4_2(op__lockt__start, struct compound_state *, cs,
9266 	    LOCKT4args *, args);
9267 
9268 	if (cs->vp == NULL) {
9269 		*cs->statusp = resp->status = NFS4ERR_NOFILEHANDLE;
9270 		goto out;
9271 	}
9272 
9273 	/*
9274 	 * NFS4 only allows locking on regular files, so
9275 	 * verify type of object.
9276 	 */
9277 	if (cs->vp->v_type != VREG) {
9278 		if (cs->vp->v_type == VDIR)
9279 			*cs->statusp = resp->status = NFS4ERR_ISDIR;
9280 		else
9281 			*cs->statusp = resp->status =  NFS4ERR_INVAL;
9282 		goto out;
9283 	}
9284 
9285 	/*
9286 	 * Check out the clientid to ensure the server knows about it
9287 	 * so that we correctly inform the client of a server reboot.
9288 	 */
9289 	if ((cp = rfs4_findclient_by_id(args->owner.clientid, FALSE))
9290 	    == NULL) {
9291 		*cs->statusp = resp->status =
9292 		    rfs4_check_clientid(&args->owner.clientid, 0);
9293 		goto out;
9294 	}
9295 	if (rfs4_lease_expired(cp)) {
9296 		rfs4_client_close(cp);
9297 		/*
9298 		 * Protocol doesn't allow returning NFS4ERR_STALE as
9299 		 * other operations do on this check so STALE_CLIENTID
9300 		 * is returned instead
9301 		 */
9302 		*cs->statusp = resp->status = NFS4ERR_STALE_CLIENTID;
9303 		goto out;
9304 	}
9305 
9306 	if (rfs4_clnt_in_grace(cp) && !(cp->rc_can_reclaim)) {
9307 		*cs->statusp = resp->status = NFS4ERR_GRACE;
9308 		rfs4_client_rele(cp);
9309 		goto out;
9310 	}
9311 	rfs4_client_rele(cp);
9312 
9313 	resp->status = NFS4_OK;
9314 
9315 	switch (args->locktype) {
9316 	case READ_LT:
9317 	case READW_LT:
9318 		ltype = F_RDLCK;
9319 		break;
9320 	case WRITE_LT:
9321 	case WRITEW_LT:
9322 		ltype = F_WRLCK;
9323 		break;
9324 	}
9325 
9326 	posix_length = args->length;
9327 	/* Check for zero length. To lock to end of file use all ones for V4 */
9328 	if (posix_length == 0) {
9329 		*cs->statusp = resp->status = NFS4ERR_INVAL;
9330 		goto out;
9331 	} else if (posix_length == (length4)(~0)) {
9332 		posix_length = 0;	/* Posix to end of file  */
9333 	}
9334 
9335 	/* Find or create a lockowner */
9336 	lo = rfs4_findlockowner(&args->owner, &create);
9337 
9338 	if (lo) {
9339 		pid = lo->rl_pid;
9340 		if ((resp->status =
9341 		    rfs4_client_sysid(lo->rl_client, &sysid)) != NFS4_OK)
9342 			goto err;
9343 	} else {
9344 		pid = 0;
9345 		sysid = lockt_sysid;
9346 	}
9347 retry:
9348 	flk.l_type = ltype;
9349 	flk.l_whence = 0;		/* SEEK_SET */
9350 	flk.l_start = args->offset;
9351 	flk.l_len = posix_length;
9352 	flk.l_sysid = sysid;
9353 	flk.l_pid = pid;
9354 	flag |= F_REMOTELOCK;
9355 
9356 	LOCK_PRINT(rfs4_debug, "rfs4_op_lockt", F_GETLK, &flk);
9357 
9358 	/* Note that length4 is uint64_t but l_len and l_start are off64_t */
9359 	if (flk.l_len < 0 || flk.l_start < 0) {
9360 		resp->status = NFS4ERR_INVAL;
9361 		goto err;
9362 	}
9363 	error = VOP_FRLOCK(cs->vp, F_GETLK, &flk, flag, (u_offset_t)0,
9364 	    NULL, cs->cr, NULL);
9365 
9366 	/*
9367 	 * N.B. We map error values to nfsv4 errors. This is differrent
9368 	 * than puterrno4 routine.
9369 	 */
9370 	switch (error) {
9371 	case 0:
9372 		if (flk.l_type == F_UNLCK)
9373 			resp->status = NFS4_OK;
9374 		else {
9375 			if (lock_denied(&resp->denied, &flk) == NFS4ERR_EXPIRED)
9376 				goto retry;
9377 			resp->status = NFS4ERR_DENIED;
9378 		}
9379 		break;
9380 	case EOVERFLOW:
9381 		resp->status = NFS4ERR_INVAL;
9382 		break;
9383 	case EINVAL:
9384 		resp->status = NFS4ERR_NOTSUPP;
9385 		break;
9386 	default:
9387 		cmn_err(CE_WARN, "rfs4_op_lockt: unexpected errno (%d)",
9388 		    error);
9389 		resp->status = NFS4ERR_SERVERFAULT;
9390 		break;
9391 	}
9392 
9393 err:
9394 	if (lo)
9395 		rfs4_lockowner_rele(lo);
9396 	*cs->statusp = resp->status;
9397 out:
9398 	DTRACE_NFSV4_2(op__lockt__done, struct compound_state *, cs,
9399 	    LOCKT4res *, resp);
9400 }
9401 
9402 int
9403 rfs4_share(rfs4_state_t *sp, uint32_t access, uint32_t deny)
9404 {
9405 	int err;
9406 	int cmd;
9407 	vnode_t *vp;
9408 	struct shrlock shr;
9409 	struct shr_locowner shr_loco;
9410 	int fflags = 0;
9411 
9412 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9413 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9414 
9415 	if (sp->rs_closed)
9416 		return (NFS4ERR_OLD_STATEID);
9417 
9418 	vp = sp->rs_finfo->rf_vp;
9419 	ASSERT(vp);
9420 
9421 	shr.s_access = shr.s_deny = 0;
9422 
9423 	if (access & OPEN4_SHARE_ACCESS_READ) {
9424 		fflags |= FREAD;
9425 		shr.s_access |= F_RDACC;
9426 	}
9427 	if (access & OPEN4_SHARE_ACCESS_WRITE) {
9428 		fflags |= FWRITE;
9429 		shr.s_access |= F_WRACC;
9430 	}
9431 	ASSERT(shr.s_access);
9432 
9433 	if (deny & OPEN4_SHARE_DENY_READ)
9434 		shr.s_deny |= F_RDDNY;
9435 	if (deny & OPEN4_SHARE_DENY_WRITE)
9436 		shr.s_deny |= F_WRDNY;
9437 
9438 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9439 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9440 	shr_loco.sl_pid = shr.s_pid;
9441 	shr_loco.sl_id = shr.s_sysid;
9442 	shr.s_owner = (caddr_t)&shr_loco;
9443 	shr.s_own_len = sizeof (shr_loco);
9444 
9445 	cmd = nbl_need_check(vp) ? F_SHARE_NBMAND : F_SHARE;
9446 
9447 	err = VOP_SHRLOCK(vp, cmd, &shr, fflags, CRED(), NULL);
9448 	if (err != 0) {
9449 		if (err == EAGAIN)
9450 			err = NFS4ERR_SHARE_DENIED;
9451 		else
9452 			err = puterrno4(err);
9453 		return (err);
9454 	}
9455 
9456 	sp->rs_share_access |= access;
9457 	sp->rs_share_deny |= deny;
9458 
9459 	return (0);
9460 }
9461 
9462 int
9463 rfs4_unshare(rfs4_state_t *sp)
9464 {
9465 	int err;
9466 	struct shrlock shr;
9467 	struct shr_locowner shr_loco;
9468 
9469 	ASSERT(rfs4_dbe_islocked(sp->rs_dbe));
9470 
9471 	if (sp->rs_closed || sp->rs_share_access == 0)
9472 		return (0);
9473 
9474 	ASSERT(sp->rs_owner->ro_client->rc_sysidt != LM_NOSYSID);
9475 	ASSERT(sp->rs_finfo->rf_vp);
9476 
9477 	shr.s_access = shr.s_deny = 0;
9478 	shr.s_pid = rfs4_dbe_getid(sp->rs_owner->ro_dbe);
9479 	shr.s_sysid = sp->rs_owner->ro_client->rc_sysidt;
9480 	shr_loco.sl_pid = shr.s_pid;
9481 	shr_loco.sl_id = shr.s_sysid;
9482 	shr.s_owner = (caddr_t)&shr_loco;
9483 	shr.s_own_len = sizeof (shr_loco);
9484 
9485 	err = VOP_SHRLOCK(sp->rs_finfo->rf_vp, F_UNSHARE, &shr, 0, CRED(),
9486 	    NULL);
9487 	if (err != 0) {
9488 		err = puterrno4(err);
9489 		return (err);
9490 	}
9491 
9492 	sp->rs_share_access = 0;
9493 	sp->rs_share_deny = 0;
9494 
9495 	return (0);
9496 
9497 }
9498 
9499 static int
9500 rdma_setup_read_data4(READ4args *args, READ4res *rok)
9501 {
9502 	struct clist	*wcl;
9503 	count4		count = rok->data_len;
9504 	int		wlist_len;
9505 
9506 	wcl = args->wlist;
9507 	if (rdma_setup_read_chunks(wcl, count, &wlist_len) == FALSE) {
9508 		return (FALSE);
9509 	}
9510 	wcl = args->wlist;
9511 	rok->wlist_len = wlist_len;
9512 	rok->wlist = wcl;
9513 	return (TRUE);
9514 }
9515 
9516 /* tunable to disable server referrals */
9517 int rfs4_no_referrals = 0;
9518 
9519 /*
9520  * Find an NFS record in reparse point data.
9521  * Returns 0 for success and <0 or an errno value on failure.
9522  */
9523 int
9524 vn_find_nfs_record(vnode_t *vp, nvlist_t **nvlp, char **svcp, char **datap)
9525 {
9526 	int err;
9527 	char *stype, *val;
9528 	nvlist_t *nvl;
9529 	nvpair_t *curr;
9530 
9531 	if ((nvl = reparse_init()) == NULL)
9532 		return (-1);
9533 
9534 	if ((err = reparse_vnode_parse(vp, nvl)) != 0) {
9535 		reparse_free(nvl);
9536 		return (err);
9537 	}
9538 
9539 	curr = NULL;
9540 	while ((curr = nvlist_next_nvpair(nvl, curr)) != NULL) {
9541 		if ((stype = nvpair_name(curr)) == NULL) {
9542 			reparse_free(nvl);
9543 			return (-2);
9544 		}
9545 		if (strncasecmp(stype, "NFS", 3) == 0)
9546 			break;
9547 	}
9548 
9549 	if ((curr == NULL) ||
9550 	    (nvpair_value_string(curr, &val))) {
9551 		reparse_free(nvl);
9552 		return (-3);
9553 	}
9554 	*nvlp = nvl;
9555 	*svcp = stype;
9556 	*datap = val;
9557 	return (0);
9558 }
9559 
9560 int
9561 vn_is_nfs_reparse(vnode_t *vp, cred_t *cr)
9562 {
9563 	nvlist_t *nvl;
9564 	char *s, *d;
9565 
9566 	if (rfs4_no_referrals != 0)
9567 		return (B_FALSE);
9568 
9569 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9570 		return (B_FALSE);
9571 
9572 	if (vn_find_nfs_record(vp, &nvl, &s, &d) != 0)
9573 		return (B_FALSE);
9574 
9575 	reparse_free(nvl);
9576 
9577 	return (B_TRUE);
9578 }
9579 
9580 /*
9581  * There is a user-level copy of this routine in ref_subr.c.
9582  * Changes should be kept in sync.
9583  */
9584 static int
9585 nfs4_create_components(char *path, component4 *comp4)
9586 {
9587 	int slen, plen, ncomp;
9588 	char *ori_path, *nxtc, buf[MAXNAMELEN];
9589 
9590 	if (path == NULL)
9591 		return (0);
9592 
9593 	plen = strlen(path) + 1;	/* include the terminator */
9594 	ori_path = path;
9595 	ncomp = 0;
9596 
9597 	/* count number of components in the path */
9598 	for (nxtc = path; nxtc < ori_path + plen; nxtc++) {
9599 		if (*nxtc == '/' || *nxtc == '\0' || *nxtc == '\n') {
9600 			if ((slen = nxtc - path) == 0) {
9601 				path = nxtc + 1;
9602 				continue;
9603 			}
9604 
9605 			if (comp4 != NULL) {
9606 				bcopy(path, buf, slen);
9607 				buf[slen] = '\0';
9608 				(void) str_to_utf8(buf, &comp4[ncomp]);
9609 			}
9610 
9611 			ncomp++;	/* 1 valid component */
9612 			path = nxtc + 1;
9613 		}
9614 		if (*nxtc == '\0' || *nxtc == '\n')
9615 			break;
9616 	}
9617 
9618 	return (ncomp);
9619 }
9620 
9621 /*
9622  * There is a user-level copy of this routine in ref_subr.c.
9623  * Changes should be kept in sync.
9624  */
9625 static int
9626 make_pathname4(char *path, pathname4 *pathname)
9627 {
9628 	int ncomp;
9629 	component4 *comp4;
9630 
9631 	if (pathname == NULL)
9632 		return (0);
9633 
9634 	if (path == NULL) {
9635 		pathname->pathname4_val = NULL;
9636 		pathname->pathname4_len = 0;
9637 		return (0);
9638 	}
9639 
9640 	/* count number of components to alloc buffer */
9641 	if ((ncomp = nfs4_create_components(path, NULL)) == 0) {
9642 		pathname->pathname4_val = NULL;
9643 		pathname->pathname4_len = 0;
9644 		return (0);
9645 	}
9646 	comp4 = kmem_zalloc(ncomp * sizeof (component4), KM_SLEEP);
9647 
9648 	/* copy components into allocated buffer */
9649 	ncomp = nfs4_create_components(path, comp4);
9650 
9651 	pathname->pathname4_val = comp4;
9652 	pathname->pathname4_len = ncomp;
9653 
9654 	return (ncomp);
9655 }
9656 
9657 #define	xdr_fs_locations4 xdr_fattr4_fs_locations
9658 
9659 fs_locations4 *
9660 fetch_referral(vnode_t *vp, cred_t *cr)
9661 {
9662 	nvlist_t *nvl;
9663 	char *stype, *sdata;
9664 	fs_locations4 *result;
9665 	char buf[1024];
9666 	size_t bufsize;
9667 	XDR xdr;
9668 	int err;
9669 
9670 	/*
9671 	 * Check attrs to ensure it's a reparse point
9672 	 */
9673 	if (vn_is_reparse(vp, cr, NULL) == B_FALSE)
9674 		return (NULL);
9675 
9676 	/*
9677 	 * Look for an NFS record and get the type and data
9678 	 */
9679 	if (vn_find_nfs_record(vp, &nvl, &stype, &sdata) != 0)
9680 		return (NULL);
9681 
9682 	/*
9683 	 * With the type and data, upcall to get the referral
9684 	 */
9685 	bufsize = sizeof (buf);
9686 	bzero(buf, sizeof (buf));
9687 	err = reparse_kderef((const char *)stype, (const char *)sdata,
9688 	    buf, &bufsize);
9689 	reparse_free(nvl);
9690 
9691 	DTRACE_PROBE4(nfs4serv__func__referral__upcall,
9692 	    char *, stype, char *, sdata, char *, buf, int, err);
9693 	if (err) {
9694 		cmn_err(CE_NOTE,
9695 		    "reparsed daemon not running: unable to get referral (%d)",
9696 		    err);
9697 		return (NULL);
9698 	}
9699 
9700 	/*
9701 	 * We get an XDR'ed record back from the kderef call
9702 	 */
9703 	xdrmem_create(&xdr, buf, bufsize, XDR_DECODE);
9704 	result = kmem_alloc(sizeof (fs_locations4), KM_SLEEP);
9705 	err = xdr_fs_locations4(&xdr, result);
9706 	XDR_DESTROY(&xdr);
9707 	if (err != TRUE) {
9708 		DTRACE_PROBE1(nfs4serv__func__referral__upcall__xdrfail,
9709 		    int, err);
9710 		return (NULL);
9711 	}
9712 
9713 	/*
9714 	 * Look at path to recover fs_root, ignoring the leading '/'
9715 	 */
9716 	(void) make_pathname4(vp->v_path, &result->fs_root);
9717 
9718 	return (result);
9719 }
9720 
9721 char *
9722 build_symlink(vnode_t *vp, cred_t *cr, size_t *strsz)
9723 {
9724 	fs_locations4 *fsl;
9725 	fs_location4 *fs;
9726 	char *server, *path, *symbuf;
9727 	static char *prefix = "/net/";
9728 	int i, size, npaths;
9729 	uint_t len;
9730 
9731 	/* Get the referral */
9732 	if ((fsl = fetch_referral(vp, cr)) == NULL)
9733 		return (NULL);
9734 
9735 	/* Deal with only the first location and first server */
9736 	fs = &fsl->locations_val[0];
9737 	server = utf8_to_str(&fs->server_val[0], &len, NULL);
9738 	if (server == NULL) {
9739 		rfs4_free_fs_locations4(fsl);
9740 		kmem_free(fsl, sizeof (fs_locations4));
9741 		return (NULL);
9742 	}
9743 
9744 	/* Figure out size for "/net/" + host + /path/path/path + NULL */
9745 	size = strlen(prefix) + len;
9746 	for (i = 0; i < fs->rootpath.pathname4_len; i++)
9747 		size += fs->rootpath.pathname4_val[i].utf8string_len + 1;
9748 
9749 	/* Allocate the symlink buffer and fill it */
9750 	symbuf = kmem_zalloc(size, KM_SLEEP);
9751 	(void) strcat(symbuf, prefix);
9752 	(void) strcat(symbuf, server);
9753 	kmem_free(server, len);
9754 
9755 	npaths = 0;
9756 	for (i = 0; i < fs->rootpath.pathname4_len; i++) {
9757 		path = utf8_to_str(&fs->rootpath.pathname4_val[i], &len, NULL);
9758 		if (path == NULL)
9759 			continue;
9760 		(void) strcat(symbuf, "/");
9761 		(void) strcat(symbuf, path);
9762 		npaths++;
9763 		kmem_free(path, len);
9764 	}
9765 
9766 	rfs4_free_fs_locations4(fsl);
9767 	kmem_free(fsl, sizeof (fs_locations4));
9768 
9769 	if (strsz != NULL)
9770 		*strsz = size;
9771 	return (symbuf);
9772 }
9773 
9774 /*
9775  * Check to see if we have a downrev Solaris client, so that we
9776  * can send it a symlink instead of a referral.
9777  */
9778 int
9779 client_is_downrev(struct svc_req *req)
9780 {
9781 	struct sockaddr *ca;
9782 	rfs4_clntip_t *ci;
9783 	bool_t create = FALSE;
9784 	int is_downrev;
9785 
9786 	ca = (struct sockaddr *)svc_getrpccaller(req->rq_xprt)->buf;
9787 	ASSERT(ca);
9788 	ci = rfs4_find_clntip(ca, &create);
9789 	if (ci == NULL)
9790 		return (0);
9791 	is_downrev = ci->ri_no_referrals;
9792 	rfs4_dbe_rele(ci->ri_dbe);
9793 	return (is_downrev);
9794 }
9795