xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 06c3fb2749bda94cb5201f81ffdb8fa6c3161b2e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/capsicum.h>
37 #include <sys/extattr.h>
38 
39 /*
40  * Functions that perform the vfs operations required by the routines in
41  * nfsd_serv.c. It is hoped that this change will make the server more
42  * portable.
43  */
44 
45 #include <fs/nfs/nfsport.h>
46 #include <security/mac/mac_framework.h>
47 #include <sys/callout.h>
48 #include <sys/filio.h>
49 #include <sys/hash.h>
50 #include <sys/osd.h>
51 #include <sys/sysctl.h>
52 #include <nlm/nlm_prot.h>
53 #include <nlm/nlm.h>
54 
55 FEATURE(nfsd, "NFSv4 server");
56 
57 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
58 extern int nfsrv_useacl;
59 extern int newnfs_numnfsd;
60 extern int nfsrv_sessionhashsize;
61 extern struct nfslayouthash *nfslayouthash;
62 extern int nfsrv_layouthashsize;
63 extern struct mtx nfsrv_dslock_mtx;
64 extern int nfs_pnfsiothreads;
65 extern volatile int nfsrv_devidcnt;
66 extern int nfsrv_maxpnfsmirror;
67 extern uint32_t nfs_srvmaxio;
68 extern int nfs_bufpackets;
69 extern u_long sb_max_adj;
70 
71 NFSD_VNET_DECLARE(int, nfsrv_numnfsd);
72 NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst);
73 NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool);
74 NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash);
75 NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash);
76 NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash);
77 NFSD_VNET_DECLARE(struct nfsv4lock, nfsd_suspend_lock);
78 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
79 
80 NFSDLOCKMUTEX;
81 NFSSTATESPINLOCK;
82 struct mtx nfsrc_udpmtx;
83 struct mtx nfs_v4root_mutex;
84 struct mtx nfsrv_dontlistlock_mtx;
85 struct mtx nfsrv_recalllock_mtx;
86 struct nfsrvfh nfs_pubfh;
87 int nfs_pubfhset = 0;
88 int nfsd_debuglevel = 0;
89 static pid_t nfsd_master_pid = (pid_t)-1;
90 static char nfsd_master_comm[MAXCOMLEN + 1];
91 static struct timeval nfsd_master_start;
92 static uint32_t nfsv4_sysid = 0;
93 static fhandle_t zerofh;
94 
95 NFSD_VNET_DEFINE(struct proc *, nfsd_master_proc) = NULL;
96 NFSD_VNET_DEFINE(struct nfsrvhashhead *, nfsrvudphashtbl);
97 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrchash_table);
98 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrcahash_table);
99 NFSD_VNET_DEFINE(struct nfsrvfh, nfs_rootfh);
100 NFSD_VNET_DEFINE(int, nfs_rootfhset) = 0;
101 NFSD_VNET_DEFINE(struct callout, nfsd_callout);
102 NFSD_VNET_DEFINE_STATIC(struct mount *, nfsv4root_mnt);
103 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_opt);
104 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_newopt);
105 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_suspend_nfsd) = false;
106 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_mntinited) = false;
107 
108 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
109     struct ucred *);
110 static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *);
111 
112 int nfsrv_enable_crossmntpt = 1;
113 static int nfs_commit_blks;
114 static int nfs_commit_miss;
115 extern int nfsrv_issuedelegs;
116 extern int nfsrv_dolocallocks;
117 extern struct nfsdevicehead nfsrv_devidhead;
118 
119 /* Map d_type to vnode type. */
120 static uint8_t dtype_to_vnode[DT_WHT + 1] = { VNON, VFIFO, VCHR, VNON, VDIR,
121     VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON };
122 #define	NFS_DTYPETOVTYPE(t)	((t) <= DT_WHT ? dtype_to_vnode[(t)] : VNON)
123 
124 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
125     struct iovec **);
126 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
127     struct mbuf **, struct iovec **);
128 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
129     int *);
130 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
131     NFSPROC_T *);
132 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
133     int *, char *, fhandle_t *);
134 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
135     NFSPROC_T *);
136 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
137     struct thread *, int, struct mbuf **, char *, struct mbuf **,
138     struct nfsvattr *, struct acl *, off_t *, int, bool *);
139 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
140 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
141     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
142 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
143     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
144     char *, int *);
145 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
146     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
147 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
148     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
149 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
150     struct vnode *, struct nfsmount **, int, struct acl *, int *);
151 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
152     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
153 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
154     struct vnode *, struct nfsmount *, struct nfsvattr *);
155 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
156     NFSPROC_T *, struct nfsmount *);
157 static int nfsrv_putfhname(fhandle_t *, char *);
158 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
159     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
160 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
161     struct vnode *, NFSPROC_T *);
162 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
163 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
164     NFSPROC_T *);
165 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
166 
167 int nfs_pnfsio(task_fn_t *, void *);
168 
169 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
170     "NFS server");
171 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
172     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
173 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
174     0, "");
175 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
176     0, "");
177 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
178     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
179 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
180     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
181 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
182     0, "Debug level for NFS server");
183 NFSD_VNET_DECLARE(int, nfsd_enable_stringtouid);
184 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid,
185     CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_enable_stringtouid),
186     0, "Enable nfsd to accept numeric owner_names");
187 static int nfsrv_pnfsgetdsattr = 1;
188 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
189     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
190 
191 /*
192  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
193  * not running.
194  * The dsN subdirectories for the increased values must have been created
195  * on all DS servers before this increase is done.
196  */
197 u_int	nfsrv_dsdirsize = 20;
198 static int
199 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
200 {
201 	int error, newdsdirsize;
202 
203 	newdsdirsize = nfsrv_dsdirsize;
204 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
205 	if (error != 0 || req->newptr == NULL)
206 		return (error);
207 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
208 	    newnfs_numnfsd != 0)
209 		return (EINVAL);
210 	nfsrv_dsdirsize = newdsdirsize;
211 	return (0);
212 }
213 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
214     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
215     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
216 
217 /*
218  * nfs_srvmaxio can only be increased and only when the nfsd threads are
219  * not running.  The setting must be a power of 2, with the current limit of
220  * 1Mbyte.
221  */
222 static int
223 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)
224 {
225 	int error;
226 	u_int newsrvmaxio;
227 	uint64_t tval;
228 
229 	newsrvmaxio = nfs_srvmaxio;
230 	error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req);
231 	if (error != 0 || req->newptr == NULL)
232 		return (error);
233 	if (newsrvmaxio == nfs_srvmaxio)
234 		return (0);
235 	if (newsrvmaxio < nfs_srvmaxio) {
236 		printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n");
237 		return (EINVAL);
238 	}
239 	if (newsrvmaxio > 1048576) {
240 		printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n");
241 		return (EINVAL);
242 	}
243 	if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) {
244 		printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n");
245 		return (EINVAL);
246 	}
247 
248 	/*
249 	 * Check that kern.ipc.maxsockbuf is large enough for
250 	 * newsrviomax, given the setting of vfs.nfs.bufpackets.
251 	 */
252 	if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets >
253 	    sb_max_adj) {
254 		/*
255 		 * Suggest vfs.nfs.bufpackets * maximum RPC message for
256 		 * sb_max_adj.
257 		 */
258 		tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets;
259 
260 		/*
261 		 * Convert suggested sb_max_adj value to a suggested
262 		 * sb_max value, which is what is set via kern.ipc.maxsockbuf.
263 		 * Perform the inverse calculation of (from uipc_sockbuf.c):
264 		 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES /
265 		 *     (MSIZE + MCLBYTES);
266 		 * XXX If the calculation of sb_max_adj from sb_max changes,
267 		 *     this calculation must be changed as well.
268 		 */
269 		tval *= (MSIZE + MCLBYTES);  /* Brackets for readability. */
270 		tval += MCLBYTES - 1;        /* Round up divide. */
271 		tval /= MCLBYTES;
272 		printf("nfsd: set kern.ipc.maxsockbuf to a minimum of "
273 		    "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval,
274 		    newsrvmaxio);
275 		return (EINVAL);
276 	}
277 
278 	NFSD_LOCK();
279 	if (newnfs_numnfsd != 0) {
280 		NFSD_UNLOCK();
281 		printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd "
282 		    "threads are running\n");
283 		return (EINVAL);
284 	}
285 
286 
287 	nfs_srvmaxio = newsrvmaxio;
288 	NFSD_UNLOCK();
289 	return (0);
290 }
291 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio,
292     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
293     sysctl_srvmaxio, "IU", "Maximum I/O size in bytes");
294 
295 #define	MAX_REORDERED_RPC	16
296 #define	NUM_HEURISTIC		1031
297 #define	NHUSE_INIT		64
298 #define	NHUSE_INC		16
299 #define	NHUSE_MAX		2048
300 
301 static struct nfsheur {
302 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
303 	off_t nh_nextoff;	/* next offset for sequential detection */
304 	int nh_use;		/* use count for selection */
305 	int nh_seqcount;	/* heuristic */
306 } nfsheur[NUM_HEURISTIC];
307 
308 /*
309  * Heuristic to detect sequential operation.
310  */
311 static struct nfsheur *
312 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
313 {
314 	struct nfsheur *nh;
315 	int hi, try;
316 
317 	/* Locate best candidate. */
318 	try = 32;
319 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
320 	nh = &nfsheur[hi];
321 	while (try--) {
322 		if (nfsheur[hi].nh_vp == vp) {
323 			nh = &nfsheur[hi];
324 			break;
325 		}
326 		if (nfsheur[hi].nh_use > 0)
327 			--nfsheur[hi].nh_use;
328 		hi = (hi + 1) % NUM_HEURISTIC;
329 		if (nfsheur[hi].nh_use < nh->nh_use)
330 			nh = &nfsheur[hi];
331 	}
332 
333 	/* Initialize hint if this is a new file. */
334 	if (nh->nh_vp != vp) {
335 		nh->nh_vp = vp;
336 		nh->nh_nextoff = uio->uio_offset;
337 		nh->nh_use = NHUSE_INIT;
338 		if (uio->uio_offset == 0)
339 			nh->nh_seqcount = 4;
340 		else
341 			nh->nh_seqcount = 1;
342 	}
343 
344 	/* Calculate heuristic. */
345 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
346 	    uio->uio_offset == nh->nh_nextoff) {
347 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
348 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
349 		if (nh->nh_seqcount > IO_SEQMAX)
350 			nh->nh_seqcount = IO_SEQMAX;
351 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
352 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
353 		/* Probably a reordered RPC, leave seqcount alone. */
354 	} else if (nh->nh_seqcount > 1) {
355 		nh->nh_seqcount /= 2;
356 	} else {
357 		nh->nh_seqcount = 0;
358 	}
359 	nh->nh_use += NHUSE_INC;
360 	if (nh->nh_use > NHUSE_MAX)
361 		nh->nh_use = NHUSE_MAX;
362 	return (nh);
363 }
364 
365 /*
366  * Get attributes into nfsvattr structure.
367  */
368 int
369 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
370     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
371     nfsattrbit_t *attrbitp)
372 {
373 	int error, gotattr, lockedit = 0;
374 	struct nfsvattr na;
375 
376 	if (vpislocked == 0) {
377 		/*
378 		 * When vpislocked == 0, the vnode is either exclusively
379 		 * locked by this thread or not locked by this thread.
380 		 * As such, shared lock it, if not exclusively locked.
381 		 */
382 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
383 			lockedit = 1;
384 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
385 		}
386 	}
387 
388 	/*
389 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
390 	 * attributes, as required.
391 	 * This needs to be done for regular files if:
392 	 * - non-NFSv4 RPCs or
393 	 * - when attrbitp == NULL or
394 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
395 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
396 	 * these attributes.  nfsrv_proxyds() will return an error if the
397 	 * server is not a pNFS one.
398 	 */
399 	gotattr = 0;
400 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
401 	    (nd->nd_flag & ND_NFSV4) == 0 ||
402 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
403 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
404 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
405 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
406 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
407 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
408 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
409 		    NULL);
410 		if (error == 0)
411 			gotattr = 1;
412 	}
413 
414 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
415 	if (lockedit != 0)
416 		NFSVOPUNLOCK(vp);
417 
418 	/*
419 	 * If we got the Change, Size and Modify Time from the DS,
420 	 * replace them.
421 	 */
422 	if (gotattr != 0) {
423 		nvap->na_atime = na.na_atime;
424 		nvap->na_mtime = na.na_mtime;
425 		nvap->na_filerev = na.na_filerev;
426 		nvap->na_size = na.na_size;
427 		nvap->na_bytes = na.na_bytes;
428 	}
429 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
430 	    error, (uintmax_t)na.na_filerev);
431 
432 	NFSEXITCODE(error);
433 	return (error);
434 }
435 
436 /*
437  * Get a file handle for a vnode.
438  */
439 int
440 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
441 {
442 	int error;
443 
444 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
445 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
446 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
447 
448 	NFSEXITCODE(error);
449 	return (error);
450 }
451 
452 /*
453  * Perform access checking for vnodes obtained from file handles that would
454  * refer to files already opened by a Unix client. You cannot just use
455  * vn_writechk() and VOP_ACCESSX() for two reasons.
456  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
457  *     case.
458  * 2 - The owner is to be given access irrespective of mode bits for some
459  *     operations, so that processes that chmod after opening a file don't
460  *     break.
461  */
462 int
463 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
464     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
465     u_int32_t *supportedtypep)
466 {
467 	struct vattr vattr;
468 	int error = 0, getret = 0;
469 
470 	if (vpislocked == 0) {
471 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
472 			error = EPERM;
473 			goto out;
474 		}
475 	}
476 	if (accmode & VWRITE) {
477 		/* Just vn_writechk() changed to check rdonly */
478 		/*
479 		 * Disallow write attempts on read-only file systems;
480 		 * unless the file is a socket or a block or character
481 		 * device resident on the file system.
482 		 */
483 		if (NFSVNO_EXRDONLY(exp) ||
484 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
485 			switch (vp->v_type) {
486 			case VREG:
487 			case VDIR:
488 			case VLNK:
489 				error = EROFS;
490 			default:
491 				break;
492 			}
493 		}
494 		/*
495 		 * If there's shared text associated with
496 		 * the inode, try to free it up once.  If
497 		 * we fail, we can't allow writing.
498 		 */
499 		if (VOP_IS_TEXT(vp) && error == 0)
500 			error = ETXTBSY;
501 	}
502 	if (error != 0) {
503 		if (vpislocked == 0)
504 			NFSVOPUNLOCK(vp);
505 		goto out;
506 	}
507 
508 	/*
509 	 * Should the override still be applied when ACLs are enabled?
510 	 */
511 	error = VOP_ACCESSX(vp, accmode, cred, p);
512 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
513 		/*
514 		 * Try again with VEXPLICIT_DENY, to see if the test for
515 		 * deletion is supported.
516 		 */
517 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
518 		if (error == 0) {
519 			if (vp->v_type == VDIR) {
520 				accmode &= ~(VDELETE | VDELETE_CHILD);
521 				accmode |= VWRITE;
522 				error = VOP_ACCESSX(vp, accmode, cred, p);
523 			} else if (supportedtypep != NULL) {
524 				*supportedtypep &= ~NFSACCESS_DELETE;
525 			}
526 		}
527 	}
528 
529 	/*
530 	 * Allow certain operations for the owner (reads and writes
531 	 * on files that are already open).
532 	 */
533 	if (override != NFSACCCHK_NOOVERRIDE &&
534 	    (error == EPERM || error == EACCES)) {
535 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
536 			error = 0;
537 		else if (override & NFSACCCHK_ALLOWOWNER) {
538 			getret = VOP_GETATTR(vp, &vattr, cred);
539 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
540 				error = 0;
541 		}
542 	}
543 	if (vpislocked == 0)
544 		NFSVOPUNLOCK(vp);
545 
546 out:
547 	NFSEXITCODE(error);
548 	return (error);
549 }
550 
551 /*
552  * Set attribute(s) vnop.
553  */
554 int
555 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
556     struct thread *p, struct nfsexstuff *exp)
557 {
558 	u_quad_t savsize = 0;
559 	int error, savedit;
560 	time_t savbtime;
561 
562 	/*
563 	 * If this is an exported file system and a pNFS service is running,
564 	 * don't VOP_SETATTR() of size for the MDS file system.
565 	 */
566 	savedit = 0;
567 	error = 0;
568 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
569 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
570 	    nvap->na_vattr.va_size > 0) {
571 		savsize = nvap->na_vattr.va_size;
572 		nvap->na_vattr.va_size = VNOVAL;
573 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
574 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
575 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
576 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
577 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
578 			savedit = 1;
579 		else
580 			savedit = 2;
581 	}
582 	if (savedit != 2)
583 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
584 	if (savedit != 0)
585 		nvap->na_vattr.va_size = savsize;
586 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
587 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
588 	    nvap->na_vattr.va_size != VNOVAL ||
589 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
590 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
591 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
592 		/* Never modify birthtime on a DS file. */
593 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
594 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
595 		/* For a pNFS server, set the attributes on the DS file. */
596 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
597 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
598 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
599 		if (error == ENOENT)
600 			error = 0;
601 	}
602 	NFSEXITCODE(error);
603 	return (error);
604 }
605 
606 /*
607  * Set up nameidata for a lookup() call and do it.
608  */
609 int
610 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
611     struct vnode *dp, int islocked, struct nfsexstuff *exp,
612     struct vnode **retdirp)
613 {
614 	struct componentname *cnp = &ndp->ni_cnd;
615 	int i;
616 	struct iovec aiov;
617 	struct uio auio;
618 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
619 	int error = 0;
620 	char *cp;
621 
622 	*retdirp = NULL;
623 	cnp->cn_nameptr = cnp->cn_pnbuf;
624 	ndp->ni_lcf = 0;
625 	/*
626 	 * Extract and set starting directory.
627 	 */
628 	if (dp->v_type != VDIR) {
629 		if (islocked)
630 			vput(dp);
631 		else
632 			vrele(dp);
633 		nfsvno_relpathbuf(ndp);
634 		error = ENOTDIR;
635 		goto out1;
636 	}
637 	if (islocked)
638 		NFSVOPUNLOCK(dp);
639 	VREF(dp);
640 	*retdirp = dp;
641 	if (NFSVNO_EXRDONLY(exp))
642 		cnp->cn_flags |= RDONLY;
643 	ndp->ni_segflg = UIO_SYSSPACE;
644 
645 	if (nd->nd_flag & ND_PUBLOOKUP) {
646 		ndp->ni_loopcnt = 0;
647 		if (cnp->cn_pnbuf[0] == '/') {
648 			vrele(dp);
649 			/*
650 			 * Check for degenerate pathnames here, since lookup()
651 			 * panics on them.
652 			 */
653 			for (i = 1; i < ndp->ni_pathlen; i++)
654 				if (cnp->cn_pnbuf[i] != '/')
655 					break;
656 			if (i == ndp->ni_pathlen) {
657 				error = NFSERR_ACCES;
658 				goto out;
659 			}
660 			dp = rootvnode;
661 			VREF(dp);
662 		}
663 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
664 	    (nd->nd_flag & ND_NFSV4) == 0) {
665 		/*
666 		 * Only cross mount points for NFSv4 when doing a
667 		 * mount while traversing the file system above
668 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
669 		 */
670 		cnp->cn_flags |= NOCROSSMOUNT;
671 	}
672 
673 	/*
674 	 * Initialize for scan, set ni_startdir and bump ref on dp again
675 	 * because lookup() will dereference ni_startdir.
676 	 */
677 
678 	ndp->ni_startdir = dp;
679 	ndp->ni_rootdir = rootvnode;
680 	ndp->ni_topdir = NULL;
681 
682 	if (!lockleaf)
683 		cnp->cn_flags |= LOCKLEAF;
684 	for (;;) {
685 		cnp->cn_nameptr = cnp->cn_pnbuf;
686 		/*
687 		 * Call lookup() to do the real work.  If an error occurs,
688 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
689 		 * we do not have to dereference anything before returning.
690 		 * In either case ni_startdir will be dereferenced and NULLed
691 		 * out.
692 		 */
693 		error = vfs_lookup(ndp);
694 		if (error)
695 			break;
696 
697 		/*
698 		 * Check for encountering a symbolic link.  Trivial
699 		 * termination occurs if no symlink encountered.
700 		 */
701 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
702 			if (ndp->ni_vp && !lockleaf)
703 				NFSVOPUNLOCK(ndp->ni_vp);
704 			break;
705 		}
706 
707 		/*
708 		 * Validate symlink
709 		 */
710 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
711 			NFSVOPUNLOCK(ndp->ni_dvp);
712 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
713 			error = EINVAL;
714 			goto badlink2;
715 		}
716 
717 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
718 			error = ELOOP;
719 			goto badlink2;
720 		}
721 		if (ndp->ni_pathlen > 1)
722 			cp = uma_zalloc(namei_zone, M_WAITOK);
723 		else
724 			cp = cnp->cn_pnbuf;
725 		aiov.iov_base = cp;
726 		aiov.iov_len = MAXPATHLEN;
727 		auio.uio_iov = &aiov;
728 		auio.uio_iovcnt = 1;
729 		auio.uio_offset = 0;
730 		auio.uio_rw = UIO_READ;
731 		auio.uio_segflg = UIO_SYSSPACE;
732 		auio.uio_td = NULL;
733 		auio.uio_resid = MAXPATHLEN;
734 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
735 		if (error) {
736 		badlink1:
737 			if (ndp->ni_pathlen > 1)
738 				uma_zfree(namei_zone, cp);
739 		badlink2:
740 			vrele(ndp->ni_dvp);
741 			vput(ndp->ni_vp);
742 			break;
743 		}
744 		linklen = MAXPATHLEN - auio.uio_resid;
745 		if (linklen == 0) {
746 			error = ENOENT;
747 			goto badlink1;
748 		}
749 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
750 			error = ENAMETOOLONG;
751 			goto badlink1;
752 		}
753 
754 		/*
755 		 * Adjust or replace path
756 		 */
757 		if (ndp->ni_pathlen > 1) {
758 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
759 			uma_zfree(namei_zone, cnp->cn_pnbuf);
760 			cnp->cn_pnbuf = cp;
761 		} else
762 			cnp->cn_pnbuf[linklen] = '\0';
763 		ndp->ni_pathlen += linklen;
764 
765 		/*
766 		 * Cleanup refs for next loop and check if root directory
767 		 * should replace current directory.  Normally ni_dvp
768 		 * becomes the new base directory and is cleaned up when
769 		 * we loop.  Explicitly null pointers after invalidation
770 		 * to clarify operation.
771 		 */
772 		vput(ndp->ni_vp);
773 		ndp->ni_vp = NULL;
774 
775 		if (cnp->cn_pnbuf[0] == '/') {
776 			vrele(ndp->ni_dvp);
777 			ndp->ni_dvp = ndp->ni_rootdir;
778 			VREF(ndp->ni_dvp);
779 		}
780 		ndp->ni_startdir = ndp->ni_dvp;
781 		ndp->ni_dvp = NULL;
782 	}
783 	if (!lockleaf)
784 		cnp->cn_flags &= ~LOCKLEAF;
785 
786 out:
787 	if (error) {
788 		nfsvno_relpathbuf(ndp);
789 		ndp->ni_vp = NULL;
790 		ndp->ni_dvp = NULL;
791 		ndp->ni_startdir = NULL;
792 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
793 		ndp->ni_dvp = NULL;
794 	}
795 
796 out1:
797 	NFSEXITCODE2(error, nd);
798 	return (error);
799 }
800 
801 /*
802  * Set up a pathname buffer and return a pointer to it and, optionally
803  * set a hash pointer.
804  */
805 void
806 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
807 {
808 	struct componentname *cnp = &ndp->ni_cnd;
809 
810 	cnp->cn_flags |= (NOMACCHECK);
811 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
812 	if (hashpp != NULL)
813 		*hashpp = NULL;
814 	*bufpp = cnp->cn_pnbuf;
815 }
816 
817 /*
818  * Release the above path buffer, if not released by nfsvno_namei().
819  */
820 void
821 nfsvno_relpathbuf(struct nameidata *ndp)
822 {
823 
824 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
825 	ndp->ni_cnd.cn_pnbuf = NULL;
826 }
827 
828 /*
829  * Readlink vnode op into an mbuf list.
830  */
831 int
832 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz,
833     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
834 {
835 	struct iovec *iv;
836 	struct uio io, *uiop = &io;
837 	struct mbuf *mp, *mp3;
838 	int len, tlen, error = 0;
839 
840 	len = NFS_MAXPATHLEN;
841 	if (maxextsiz > 0)
842 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
843 		    &mp3, &mp, &iv);
844 	else
845 		uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
846 	uiop->uio_iov = iv;
847 	uiop->uio_offset = 0;
848 	uiop->uio_resid = len;
849 	uiop->uio_rw = UIO_READ;
850 	uiop->uio_segflg = UIO_SYSSPACE;
851 	uiop->uio_td = NULL;
852 	error = VOP_READLINK(vp, uiop, cred);
853 	free(iv, M_TEMP);
854 	if (error) {
855 		m_freem(mp3);
856 		*lenp = 0;
857 		goto out;
858 	}
859 	if (uiop->uio_resid > 0) {
860 		len -= uiop->uio_resid;
861 		tlen = NFSM_RNDUP(len);
862 		if (tlen == 0) {
863 			m_freem(mp3);
864 			mp3 = mp = NULL;
865 		} else if (tlen != NFS_MAXPATHLEN || tlen != len)
866 			mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
867 			    tlen - len);
868 	}
869 	*lenp = len;
870 	*mpp = mp3;
871 	*mpendp = mp;
872 
873 out:
874 	NFSEXITCODE(error);
875 	return (error);
876 }
877 
878 /*
879  * Create an mbuf chain and an associated iovec that can be used to Read
880  * or Getextattr of data.
881  * Upon success, return pointers to the first and last mbufs in the chain
882  * plus the malloc'd iovec and its iovlen.
883  */
884 static int
885 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
886     struct iovec **ivp)
887 {
888 	struct mbuf *m, *m2 = NULL, *m3;
889 	struct iovec *iv;
890 	int i, left, siz;
891 
892 	left = len;
893 	m3 = NULL;
894 	/*
895 	 * Generate the mbuf list with the uio_iov ref. to it.
896 	 */
897 	i = 0;
898 	while (left > 0) {
899 		NFSMGET(m);
900 		MCLGET(m, M_WAITOK);
901 		m->m_len = 0;
902 		siz = min(M_TRAILINGSPACE(m), left);
903 		left -= siz;
904 		i++;
905 		if (m3)
906 			m2->m_next = m;
907 		else
908 			m3 = m;
909 		m2 = m;
910 	}
911 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
912 	m = m3;
913 	left = len;
914 	i = 0;
915 	while (left > 0) {
916 		if (m == NULL)
917 			panic("nfsrv_createiovec iov");
918 		siz = min(M_TRAILINGSPACE(m), left);
919 		if (siz > 0) {
920 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
921 			iv->iov_len = siz;
922 			m->m_len += siz;
923 			left -= siz;
924 			iv++;
925 			i++;
926 		}
927 		m = m->m_next;
928 	}
929 	*mpp = m3;
930 	*mpendp = m2;
931 	return (i);
932 }
933 
934 /*
935  * Create an mbuf chain and an associated iovec that can be used to Read
936  * or Getextattr of data.
937  * Upon success, return pointers to the first and last mbufs in the chain
938  * plus the malloc'd iovec and its iovlen.
939  * Same as above, but creates ext_pgs mbuf(s).
940  */
941 static int
942 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp,
943     struct mbuf **mpendp, struct iovec **ivp)
944 {
945 	struct mbuf *m, *m2 = NULL, *m3;
946 	struct iovec *iv;
947 	int i, left, pgno, siz;
948 
949 	left = len;
950 	m3 = NULL;
951 	/*
952 	 * Generate the mbuf list with the uio_iov ref. to it.
953 	 */
954 	i = 0;
955 	while (left > 0) {
956 		siz = min(left, maxextsiz);
957 		m = mb_alloc_ext_plus_pages(siz, M_WAITOK);
958 		left -= siz;
959 		i += m->m_epg_npgs;
960 		if (m3 != NULL)
961 			m2->m_next = m;
962 		else
963 			m3 = m;
964 		m2 = m;
965 	}
966 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
967 	m = m3;
968 	left = len;
969 	i = 0;
970 	pgno = 0;
971 	while (left > 0) {
972 		if (m == NULL)
973 			panic("nfsvno_createiovec_extpgs iov");
974 		siz = min(PAGE_SIZE, left);
975 		if (siz > 0) {
976 			iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]);
977 			iv->iov_len = siz;
978 			m->m_len += siz;
979 			if (pgno == m->m_epg_npgs - 1)
980 				m->m_epg_last_len = siz;
981 			left -= siz;
982 			iv++;
983 			i++;
984 			pgno++;
985 		}
986 		if (pgno == m->m_epg_npgs && left > 0) {
987 			m = m->m_next;
988 			if (m == NULL)
989 				panic("nfsvno_createiovec_extpgs iov");
990 			pgno = 0;
991 		}
992 	}
993 	*mpp = m3;
994 	*mpendp = m2;
995 	return (i);
996 }
997 
998 /*
999  * Read vnode op call into mbuf list.
1000  */
1001 int
1002 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
1003     int maxextsiz, struct thread *p, struct mbuf **mpp,
1004     struct mbuf **mpendp)
1005 {
1006 	struct mbuf *m;
1007 	struct iovec *iv;
1008 	int error = 0, len, tlen, ioflag = 0;
1009 	struct mbuf *m3;
1010 	struct uio io, *uiop = &io;
1011 	struct nfsheur *nh;
1012 
1013 	/*
1014 	 * Attempt to read from a DS file. A return of ENOENT implies
1015 	 * there is no DS file to read.
1016 	 */
1017 	error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
1018 	    NULL, mpendp, NULL, NULL, NULL, 0, NULL);
1019 	if (error != ENOENT)
1020 		return (error);
1021 
1022 	len = NFSM_RNDUP(cnt);
1023 	if (maxextsiz > 0)
1024 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
1025 		    &m3, &m, &iv);
1026 	else
1027 		uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
1028 	uiop->uio_iov = iv;
1029 	uiop->uio_offset = off;
1030 	uiop->uio_resid = len;
1031 	uiop->uio_rw = UIO_READ;
1032 	uiop->uio_segflg = UIO_SYSSPACE;
1033 	uiop->uio_td = NULL;
1034 	nh = nfsrv_sequential_heuristic(uiop, vp);
1035 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
1036 	/* XXX KDM make this more systematic? */
1037 	NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_READ] += uiop->uio_resid;
1038 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
1039 	free(iv, M_TEMP);
1040 	if (error) {
1041 		m_freem(m3);
1042 		*mpp = NULL;
1043 		goto out;
1044 	}
1045 	nh->nh_nextoff = uiop->uio_offset;
1046 	tlen = len - uiop->uio_resid;
1047 	cnt = cnt < tlen ? cnt : tlen;
1048 	tlen = NFSM_RNDUP(cnt);
1049 	if (tlen == 0) {
1050 		m_freem(m3);
1051 		m3 = m = NULL;
1052 	} else if (len != tlen || tlen != cnt)
1053 		m = nfsrv_adj(m3, len - tlen, tlen - cnt);
1054 	*mpp = m3;
1055 	*mpendp = m;
1056 
1057 out:
1058 	NFSEXITCODE(error);
1059 	return (error);
1060 }
1061 
1062 /*
1063  * Create the iovec for the mbuf chain passed in as an argument.
1064  * The "cp" argument is where the data starts within the first mbuf in
1065  * the chain. It returns the iovec and the iovcnt.
1066  */
1067 static int
1068 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
1069     int *iovcntp)
1070 {
1071 	struct mbuf *mp;
1072 	struct iovec *ivp;
1073 	int cnt, i, len;
1074 
1075 	/*
1076 	 * Loop through the mbuf chain, counting how many mbufs are a
1077 	 * part of this write operation, so the iovec size is known.
1078 	 */
1079 	cnt = 0;
1080 	len = retlen;
1081 	mp = m;
1082 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1083 	while (len > 0) {
1084 		if (i > 0) {
1085 			len -= i;
1086 			cnt++;
1087 		}
1088 		mp = mp->m_next;
1089 		if (!mp) {
1090 			if (len > 0)
1091 				return (EBADRPC);
1092 		} else
1093 			i = mp->m_len;
1094 	}
1095 
1096 	/* Now, create the iovec. */
1097 	mp = m;
1098 	*ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1099 	    M_WAITOK);
1100 	*iovcntp = cnt;
1101 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1102 	len = retlen;
1103 	while (len > 0) {
1104 		if (mp == NULL)
1105 			panic("nfsrv_createiovecw");
1106 		if (i > 0) {
1107 			i = min(i, len);
1108 			ivp->iov_base = cp;
1109 			ivp->iov_len = i;
1110 			ivp++;
1111 			len -= i;
1112 		}
1113 		mp = mp->m_next;
1114 		if (mp) {
1115 			i = mp->m_len;
1116 			cp = mtod(mp, caddr_t);
1117 		}
1118 	}
1119 	return (0);
1120 }
1121 
1122 /*
1123  * Write vnode op from an mbuf list.
1124  */
1125 int
1126 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
1127     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
1128 {
1129 	struct iovec *iv;
1130 	int cnt, ioflags, error;
1131 	struct uio io, *uiop = &io;
1132 	struct nfsheur *nh;
1133 
1134 	/*
1135 	 * Attempt to write to a DS file. A return of ENOENT implies
1136 	 * there is no DS file to write.
1137 	 */
1138 	error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
1139 	    &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
1140 	if (error != ENOENT) {
1141 		*stable = NFSWRITE_FILESYNC;
1142 		return (error);
1143 	}
1144 
1145 	if (*stable == NFSWRITE_UNSTABLE)
1146 		ioflags = IO_NODELOCKED;
1147 	else
1148 		ioflags = (IO_SYNC | IO_NODELOCKED);
1149 	error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
1150 	if (error != 0)
1151 		return (error);
1152 	uiop->uio_iov = iv;
1153 	uiop->uio_iovcnt = cnt;
1154 	uiop->uio_resid = retlen;
1155 	uiop->uio_rw = UIO_WRITE;
1156 	uiop->uio_segflg = UIO_SYSSPACE;
1157 	NFSUIOPROC(uiop, p);
1158 	uiop->uio_offset = off;
1159 	nh = nfsrv_sequential_heuristic(uiop, vp);
1160 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1161 	/* XXX KDM make this more systematic? */
1162 	NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
1163 	error = VOP_WRITE(vp, uiop, ioflags, cred);
1164 	if (error == 0)
1165 		nh->nh_nextoff = uiop->uio_offset;
1166 	free(iv, M_TEMP);
1167 
1168 	NFSEXITCODE(error);
1169 	return (error);
1170 }
1171 
1172 /*
1173  * Common code for creating a regular file (plus special files for V2).
1174  */
1175 int
1176 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1177     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1178     int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1179 {
1180 	u_quad_t tempsize;
1181 	int error;
1182 	struct thread *p = curthread;
1183 
1184 	error = nd->nd_repstat;
1185 	if (!error && ndp->ni_vp == NULL) {
1186 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1187 			error = VOP_CREATE(ndp->ni_dvp,
1188 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1189 			/* For a pNFS server, create the data file on a DS. */
1190 			if (error == 0 && nvap->na_type == VREG) {
1191 				/*
1192 				 * Create a data file on a DS for a pNFS server.
1193 				 * This function just returns if not
1194 				 * running a pNFS DS or the creation fails.
1195 				 */
1196 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1197 				    nd->nd_cred, p);
1198 			}
1199 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1200 			    NULL, false);
1201 			nfsvno_relpathbuf(ndp);
1202 			if (!error) {
1203 				if (*exclusive_flagp) {
1204 					*exclusive_flagp = 0;
1205 					NFSVNO_ATTRINIT(nvap);
1206 					nvap->na_atime.tv_sec = cverf[0];
1207 					nvap->na_atime.tv_nsec = cverf[1];
1208 					error = VOP_SETATTR(ndp->ni_vp,
1209 					    &nvap->na_vattr, nd->nd_cred);
1210 					if (error != 0) {
1211 						vput(ndp->ni_vp);
1212 						ndp->ni_vp = NULL;
1213 						error = NFSERR_NOTSUPP;
1214 					}
1215 				}
1216 			}
1217 		/*
1218 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1219 		 * (This implies, just get out on an error.)
1220 		 */
1221 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1222 			nvap->na_type == VFIFO) {
1223 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
1224 				nvap->na_type = VFIFO;
1225                         if (nvap->na_type != VFIFO &&
1226 			    (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1227 				nfsvno_relpathbuf(ndp);
1228 				vput(ndp->ni_dvp);
1229 				goto out;
1230 			}
1231 			nvap->na_rdev = rdev;
1232 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1233 			    &ndp->ni_cnd, &nvap->na_vattr);
1234 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1235 			    NULL, false);
1236 			nfsvno_relpathbuf(ndp);
1237 			if (error)
1238 				goto out;
1239 		} else {
1240 			nfsvno_relpathbuf(ndp);
1241 			vput(ndp->ni_dvp);
1242 			error = ENXIO;
1243 			goto out;
1244 		}
1245 		*vpp = ndp->ni_vp;
1246 	} else {
1247 		/*
1248 		 * Handle cases where error is already set and/or
1249 		 * the file exists.
1250 		 * 1 - clean up the lookup
1251 		 * 2 - iff !error and na_size set, truncate it
1252 		 */
1253 		nfsvno_relpathbuf(ndp);
1254 		*vpp = ndp->ni_vp;
1255 		if (ndp->ni_dvp == *vpp)
1256 			vrele(ndp->ni_dvp);
1257 		else
1258 			vput(ndp->ni_dvp);
1259 		if (!error && nvap->na_size != VNOVAL) {
1260 			error = nfsvno_accchk(*vpp, VWRITE,
1261 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1262 			    NFSACCCHK_VPISLOCKED, NULL);
1263 			if (!error) {
1264 				tempsize = nvap->na_size;
1265 				NFSVNO_ATTRINIT(nvap);
1266 				nvap->na_size = tempsize;
1267 				error = nfsvno_setattr(*vpp, nvap,
1268 				    nd->nd_cred, p, exp);
1269 			}
1270 		}
1271 		if (error)
1272 			vput(*vpp);
1273 	}
1274 
1275 out:
1276 	NFSEXITCODE(error);
1277 	return (error);
1278 }
1279 
1280 /*
1281  * Do a mknod vnode op.
1282  */
1283 int
1284 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1285     struct thread *p)
1286 {
1287 	int error = 0;
1288 	__enum_uint8(vtype) vtyp;
1289 
1290 	vtyp = nvap->na_type;
1291 	/*
1292 	 * Iff doesn't exist, create it.
1293 	 */
1294 	if (ndp->ni_vp) {
1295 		nfsvno_relpathbuf(ndp);
1296 		vput(ndp->ni_dvp);
1297 		vrele(ndp->ni_vp);
1298 		error = EEXIST;
1299 		goto out;
1300 	}
1301 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1302 		nfsvno_relpathbuf(ndp);
1303 		vput(ndp->ni_dvp);
1304 		error = NFSERR_BADTYPE;
1305 		goto out;
1306 	}
1307 	if (vtyp == VSOCK) {
1308 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1309 		    &ndp->ni_cnd, &nvap->na_vattr);
1310 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1311 		    false);
1312 		nfsvno_relpathbuf(ndp);
1313 	} else {
1314 		if (nvap->na_type != VFIFO &&
1315 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1316 			nfsvno_relpathbuf(ndp);
1317 			vput(ndp->ni_dvp);
1318 			goto out;
1319 		}
1320 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1321 		    &ndp->ni_cnd, &nvap->na_vattr);
1322 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1323 		    false);
1324 		nfsvno_relpathbuf(ndp);
1325 		/*
1326 		 * Since VOP_MKNOD returns the ni_vp, I can't
1327 		 * see any reason to do the lookup.
1328 		 */
1329 	}
1330 
1331 out:
1332 	NFSEXITCODE(error);
1333 	return (error);
1334 }
1335 
1336 /*
1337  * Mkdir vnode op.
1338  */
1339 int
1340 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1341     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1342 {
1343 	int error = 0;
1344 
1345 	if (ndp->ni_vp != NULL) {
1346 		if (ndp->ni_dvp == ndp->ni_vp)
1347 			vrele(ndp->ni_dvp);
1348 		else
1349 			vput(ndp->ni_dvp);
1350 		vrele(ndp->ni_vp);
1351 		nfsvno_relpathbuf(ndp);
1352 		error = EEXIST;
1353 		goto out;
1354 	}
1355 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1356 	    &nvap->na_vattr);
1357 	VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false);
1358 	nfsvno_relpathbuf(ndp);
1359 
1360 out:
1361 	NFSEXITCODE(error);
1362 	return (error);
1363 }
1364 
1365 /*
1366  * symlink vnode op.
1367  */
1368 int
1369 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1370     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1371     struct nfsexstuff *exp)
1372 {
1373 	int error = 0;
1374 
1375 	if (ndp->ni_vp) {
1376 		nfsvno_relpathbuf(ndp);
1377 		if (ndp->ni_dvp == ndp->ni_vp)
1378 			vrele(ndp->ni_dvp);
1379 		else
1380 			vput(ndp->ni_dvp);
1381 		vrele(ndp->ni_vp);
1382 		error = EEXIST;
1383 		goto out;
1384 	}
1385 
1386 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1387 	    &nvap->na_vattr, pathcp);
1388 	/*
1389 	 * Although FreeBSD still had the lookup code in
1390 	 * it for 7/current, there doesn't seem to be any
1391 	 * point, since VOP_SYMLINK() returns the ni_vp.
1392 	 * Just vput it for v2.
1393 	 */
1394 	VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0);
1395 	nfsvno_relpathbuf(ndp);
1396 
1397 out:
1398 	NFSEXITCODE(error);
1399 	return (error);
1400 }
1401 
1402 /*
1403  * Parse symbolic link arguments.
1404  * This function has an ugly side effect. It will malloc() an area for
1405  * the symlink and set iov_base to point to it, only if it succeeds.
1406  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1407  * be FREE'd later.
1408  */
1409 int
1410 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1411     struct thread *p, char **pathcpp, int *lenp)
1412 {
1413 	u_int32_t *tl;
1414 	char *pathcp = NULL;
1415 	int error = 0, len;
1416 	struct nfsv2_sattr *sp;
1417 
1418 	*pathcpp = NULL;
1419 	*lenp = 0;
1420 	if ((nd->nd_flag & ND_NFSV3) &&
1421 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1422 		goto nfsmout;
1423 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1424 	len = fxdr_unsigned(int, *tl);
1425 	if (len > NFS_MAXPATHLEN || len <= 0) {
1426 		error = EBADRPC;
1427 		goto nfsmout;
1428 	}
1429 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1430 	error = nfsrv_mtostr(nd, pathcp, len);
1431 	if (error)
1432 		goto nfsmout;
1433 	if (nd->nd_flag & ND_NFSV2) {
1434 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1435 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1436 	}
1437 	*pathcpp = pathcp;
1438 	*lenp = len;
1439 	NFSEXITCODE2(0, nd);
1440 	return (0);
1441 nfsmout:
1442 	if (pathcp)
1443 		free(pathcp, M_TEMP);
1444 	NFSEXITCODE2(error, nd);
1445 	return (error);
1446 }
1447 
1448 /*
1449  * Remove a non-directory object.
1450  */
1451 int
1452 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1453     struct thread *p, struct nfsexstuff *exp)
1454 {
1455 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1456 	int error = 0, mirrorcnt;
1457 	char fname[PNFS_FILENAME_LEN + 1];
1458 	fhandle_t fh;
1459 
1460 	vp = ndp->ni_vp;
1461 	dsdvp[0] = NULL;
1462 	if (vp->v_type == VDIR)
1463 		error = NFSERR_ISDIR;
1464 	else if (is_v4)
1465 		error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1466 		    p);
1467 	if (error == 0)
1468 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1469 	if (!error)
1470 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1471 	if (error == 0 && dsdvp[0] != NULL)
1472 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1473 	if (ndp->ni_dvp == vp)
1474 		vrele(ndp->ni_dvp);
1475 	else
1476 		vput(ndp->ni_dvp);
1477 	vput(vp);
1478 	nfsvno_relpathbuf(ndp);
1479 	NFSEXITCODE(error);
1480 	return (error);
1481 }
1482 
1483 /*
1484  * Remove a directory.
1485  */
1486 int
1487 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1488     struct thread *p, struct nfsexstuff *exp)
1489 {
1490 	struct vnode *vp;
1491 	int error = 0;
1492 
1493 	vp = ndp->ni_vp;
1494 	if (vp->v_type != VDIR) {
1495 		error = ENOTDIR;
1496 		goto out;
1497 	}
1498 	/*
1499 	 * No rmdir "." please.
1500 	 */
1501 	if (ndp->ni_dvp == vp) {
1502 		error = EINVAL;
1503 		goto out;
1504 	}
1505 	/*
1506 	 * The root of a mounted filesystem cannot be deleted.
1507 	 */
1508 	if (vp->v_vflag & VV_ROOT)
1509 		error = EBUSY;
1510 out:
1511 	if (!error)
1512 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1513 	if (ndp->ni_dvp == vp)
1514 		vrele(ndp->ni_dvp);
1515 	else
1516 		vput(ndp->ni_dvp);
1517 	vput(vp);
1518 	nfsvno_relpathbuf(ndp);
1519 	NFSEXITCODE(error);
1520 	return (error);
1521 }
1522 
1523 /*
1524  * Rename vnode op.
1525  */
1526 int
1527 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1528     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1529 {
1530 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1531 	int error = 0, mirrorcnt;
1532 	char fname[PNFS_FILENAME_LEN + 1];
1533 	fhandle_t fh;
1534 
1535 	dsdvp[0] = NULL;
1536 	fvp = fromndp->ni_vp;
1537 	if (ndstat) {
1538 		vrele(fromndp->ni_dvp);
1539 		vrele(fvp);
1540 		error = ndstat;
1541 		goto out1;
1542 	}
1543 	tdvp = tondp->ni_dvp;
1544 	tvp = tondp->ni_vp;
1545 	if (tvp != NULL) {
1546 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1547 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1548 			goto out;
1549 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1550 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1551 			goto out;
1552 		}
1553 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1554 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1555 			goto out;
1556 		}
1557 
1558 		/*
1559 		 * A rename to '.' or '..' results in a prematurely
1560 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1561 		 * here.
1562 		 */
1563 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1564 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1565 		    (tondp->ni_cnd.cn_namelen == 2 &&
1566 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1567 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1568 			error = EINVAL;
1569 			goto out;
1570 		}
1571 	}
1572 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1573 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1574 		goto out;
1575 	}
1576 	if (fvp->v_mount != tdvp->v_mount) {
1577 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1578 		goto out;
1579 	}
1580 	if (fvp == tdvp) {
1581 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1582 		goto out;
1583 	}
1584 	if (fvp == tvp) {
1585 		/*
1586 		 * If source and destination are the same, there is nothing to
1587 		 * do. Set error to -1 to indicate this.
1588 		 */
1589 		error = -1;
1590 		goto out;
1591 	}
1592 	if (ndflag & ND_NFSV4) {
1593 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1594 			error = nfsrv_checkremove(fvp, 0, NULL,
1595 			    (nfsquad_t)((u_quad_t)0), p);
1596 			NFSVOPUNLOCK(fvp);
1597 		} else
1598 			error = EPERM;
1599 		if (tvp && !error)
1600 			error = nfsrv_checkremove(tvp, 1, NULL,
1601 			    (nfsquad_t)((u_quad_t)0), p);
1602 	} else {
1603 		/*
1604 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1605 		 * that the NFSv4 client won't be confused by the rename.
1606 		 * Since nfsd_recalldelegation() can only be called on an
1607 		 * unlocked vnode at this point and fvp is the file that will
1608 		 * still exist after the rename, just do fvp.
1609 		 */
1610 		nfsd_recalldelegation(fvp, p);
1611 	}
1612 	if (error == 0 && tvp != NULL) {
1613 		nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1614 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1615 		    " dsdvp=%p\n", dsdvp[0]);
1616 	}
1617 out:
1618 	if (!error) {
1619 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1620 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1621 		    &tondp->ni_cnd);
1622 	} else {
1623 		if (tdvp == tvp)
1624 			vrele(tdvp);
1625 		else
1626 			vput(tdvp);
1627 		if (tvp)
1628 			vput(tvp);
1629 		vrele(fromndp->ni_dvp);
1630 		vrele(fvp);
1631 		if (error == -1)
1632 			error = 0;
1633 	}
1634 
1635 	/*
1636 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1637 	 * if the rename succeeded, the DS file for the tvp needs to be
1638 	 * removed.
1639 	 */
1640 	if (error == 0 && dsdvp[0] != NULL) {
1641 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1642 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1643 	}
1644 
1645 	nfsvno_relpathbuf(tondp);
1646 out1:
1647 	nfsvno_relpathbuf(fromndp);
1648 	NFSEXITCODE(error);
1649 	return (error);
1650 }
1651 
1652 /*
1653  * Link vnode op.
1654  */
1655 int
1656 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1657     struct thread *p, struct nfsexstuff *exp)
1658 {
1659 	struct vnode *xp;
1660 	int error = 0;
1661 
1662 	xp = ndp->ni_vp;
1663 	if (xp != NULL) {
1664 		error = EEXIST;
1665 	} else {
1666 		xp = ndp->ni_dvp;
1667 		if (vp->v_mount != xp->v_mount)
1668 			error = EXDEV;
1669 	}
1670 	if (!error) {
1671 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1672 		if (!VN_IS_DOOMED(vp))
1673 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1674 		else
1675 			error = EPERM;
1676 		if (ndp->ni_dvp == vp) {
1677 			vrele(ndp->ni_dvp);
1678 			NFSVOPUNLOCK(vp);
1679 		} else {
1680 			vref(vp);
1681 			VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true);
1682 		}
1683 	} else {
1684 		if (ndp->ni_dvp == ndp->ni_vp)
1685 			vrele(ndp->ni_dvp);
1686 		else
1687 			vput(ndp->ni_dvp);
1688 		if (ndp->ni_vp)
1689 			vrele(ndp->ni_vp);
1690 	}
1691 	nfsvno_relpathbuf(ndp);
1692 	NFSEXITCODE(error);
1693 	return (error);
1694 }
1695 
1696 /*
1697  * Do the fsync() appropriate for the commit.
1698  */
1699 int
1700 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1701     struct thread *td)
1702 {
1703 	int error = 0;
1704 
1705 	/*
1706 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1707 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1708 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1709 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1710 	 * File systems that do not use the buffer cache (as indicated
1711 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1712 	 */
1713 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1714 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1715 		/*
1716 		 * Give up and do the whole thing
1717 		 */
1718 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1719 			VM_OBJECT_WLOCK(vp->v_object);
1720 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1721 			VM_OBJECT_WUNLOCK(vp->v_object);
1722 		}
1723 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1724 	} else {
1725 		/*
1726 		 * Locate and synchronously write any buffers that fall
1727 		 * into the requested range.  Note:  we are assuming that
1728 		 * f_iosize is a power of 2.
1729 		 */
1730 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1731 		int iomask = iosize - 1;
1732 		struct bufobj *bo;
1733 		daddr_t lblkno;
1734 
1735 		/*
1736 		 * Align to iosize boundary, super-align to page boundary.
1737 		 */
1738 		if (off & iomask) {
1739 			cnt += off & iomask;
1740 			off &= ~(u_quad_t)iomask;
1741 		}
1742 		if (off & PAGE_MASK) {
1743 			cnt += off & PAGE_MASK;
1744 			off &= ~(u_quad_t)PAGE_MASK;
1745 		}
1746 		lblkno = off / iosize;
1747 
1748 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1749 			VM_OBJECT_WLOCK(vp->v_object);
1750 			vm_object_page_clean(vp->v_object, off, off + cnt,
1751 			    OBJPC_SYNC);
1752 			VM_OBJECT_WUNLOCK(vp->v_object);
1753 		}
1754 
1755 		bo = &vp->v_bufobj;
1756 		BO_LOCK(bo);
1757 		while (cnt > 0) {
1758 			struct buf *bp;
1759 
1760 			/*
1761 			 * If we have a buffer and it is marked B_DELWRI we
1762 			 * have to lock and write it.  Otherwise the prior
1763 			 * write is assumed to have already been committed.
1764 			 *
1765 			 * gbincore() can return invalid buffers now so we
1766 			 * have to check that bit as well (though B_DELWRI
1767 			 * should not be set if B_INVAL is set there could be
1768 			 * a race here since we haven't locked the buffer).
1769 			 */
1770 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1771 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1772 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1773 					BO_LOCK(bo);
1774 					continue; /* retry */
1775 				}
1776 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1777 				    B_DELWRI) {
1778 					bremfree(bp);
1779 					bp->b_flags &= ~B_ASYNC;
1780 					bwrite(bp);
1781 					++nfs_commit_miss;
1782 				} else
1783 					BUF_UNLOCK(bp);
1784 				BO_LOCK(bo);
1785 			}
1786 			++nfs_commit_blks;
1787 			if (cnt < iosize)
1788 				break;
1789 			cnt -= iosize;
1790 			++lblkno;
1791 		}
1792 		BO_UNLOCK(bo);
1793 	}
1794 	NFSEXITCODE(error);
1795 	return (error);
1796 }
1797 
1798 /*
1799  * Statfs vnode op.
1800  */
1801 int
1802 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1803 {
1804 	struct statfs *tsf;
1805 	int error;
1806 
1807 	tsf = NULL;
1808 	if (nfsrv_devidcnt > 0) {
1809 		/* For a pNFS service, get the DS numbers. */
1810 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1811 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1812 		if (error != 0) {
1813 			free(tsf, M_TEMP);
1814 			tsf = NULL;
1815 		}
1816 	}
1817 	error = VFS_STATFS(vp->v_mount, sf);
1818 	if (error == 0) {
1819 		if (tsf != NULL) {
1820 			sf->f_blocks = tsf->f_blocks;
1821 			sf->f_bavail = tsf->f_bavail;
1822 			sf->f_bfree = tsf->f_bfree;
1823 			sf->f_bsize = tsf->f_bsize;
1824 		}
1825 		/*
1826 		 * Since NFS handles these values as unsigned on the
1827 		 * wire, there is no way to represent negative values,
1828 		 * so set them to 0. Without this, they will appear
1829 		 * to be very large positive values for clients like
1830 		 * Solaris10.
1831 		 */
1832 		if (sf->f_bavail < 0)
1833 			sf->f_bavail = 0;
1834 		if (sf->f_ffree < 0)
1835 			sf->f_ffree = 0;
1836 	}
1837 	free(tsf, M_TEMP);
1838 	NFSEXITCODE(error);
1839 	return (error);
1840 }
1841 
1842 /*
1843  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1844  * must handle nfsrv_opencheck() calls after any other access checks.
1845  */
1846 void
1847 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1848     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1849     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1850     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, bool done_namei,
1851     struct nfsexstuff *exp, struct vnode **vpp)
1852 {
1853 	struct vnode *vp = NULL;
1854 	u_quad_t tempsize;
1855 	struct nfsexstuff nes;
1856 	struct thread *p = curthread;
1857 	uint32_t oldrepstat;
1858 
1859 	if (ndp->ni_vp == NULL) {
1860 		/*
1861 		 * If nfsrv_opencheck() sets nd_repstat, done_namei needs to be
1862 		 * set true, since cleanup after nfsvno_namei() is needed.
1863 		 */
1864 		oldrepstat = nd->nd_repstat;
1865 		nd->nd_repstat = nfsrv_opencheck(clientid,
1866 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1867 		if (nd->nd_repstat != 0 && oldrepstat == 0)
1868 			done_namei = true;
1869 	}
1870 	if (!nd->nd_repstat) {
1871 		if (ndp->ni_vp == NULL) {
1872 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1873 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1874 			/* For a pNFS server, create the data file on a DS. */
1875 			if (nd->nd_repstat == 0) {
1876 				/*
1877 				 * Create a data file on a DS for a pNFS server.
1878 				 * This function just returns if not
1879 				 * running a pNFS DS or the creation fails.
1880 				 */
1881 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1882 				    cred, p);
1883 			}
1884 			VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ?
1885 			    &ndp->ni_vp : NULL, false);
1886 			nfsvno_relpathbuf(ndp);
1887 			if (!nd->nd_repstat) {
1888 				if (*exclusive_flagp) {
1889 					*exclusive_flagp = 0;
1890 					NFSVNO_ATTRINIT(nvap);
1891 					nvap->na_atime.tv_sec = cverf[0];
1892 					nvap->na_atime.tv_nsec = cverf[1];
1893 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1894 					    &nvap->na_vattr, cred);
1895 					if (nd->nd_repstat != 0) {
1896 						vput(ndp->ni_vp);
1897 						ndp->ni_vp = NULL;
1898 						nd->nd_repstat = NFSERR_NOTSUPP;
1899 					} else
1900 						NFSSETBIT_ATTRBIT(attrbitp,
1901 						    NFSATTRBIT_TIMEACCESS);
1902 				} else {
1903 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1904 					    aclp, p, attrbitp, exp);
1905 				}
1906 			}
1907 			vp = ndp->ni_vp;
1908 		} else {
1909 			nfsvno_relpathbuf(ndp);
1910 			vp = ndp->ni_vp;
1911 			if (create == NFSV4OPEN_CREATE) {
1912 				if (ndp->ni_dvp == vp)
1913 					vrele(ndp->ni_dvp);
1914 				else
1915 					vput(ndp->ni_dvp);
1916 			}
1917 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1918 				if (ndp->ni_cnd.cn_flags & RDONLY)
1919 					NFSVNO_SETEXRDONLY(&nes);
1920 				else
1921 					NFSVNO_EXINIT(&nes);
1922 				nd->nd_repstat = nfsvno_accchk(vp,
1923 				    VWRITE, cred, &nes, p,
1924 				    NFSACCCHK_NOOVERRIDE,
1925 				    NFSACCCHK_VPISLOCKED, NULL);
1926 				nd->nd_repstat = nfsrv_opencheck(clientid,
1927 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1928 				if (!nd->nd_repstat) {
1929 					tempsize = nvap->na_size;
1930 					NFSVNO_ATTRINIT(nvap);
1931 					nvap->na_size = tempsize;
1932 					nd->nd_repstat = nfsvno_setattr(vp,
1933 					    nvap, cred, p, exp);
1934 				}
1935 			} else if (vp->v_type == VREG) {
1936 				nd->nd_repstat = nfsrv_opencheck(clientid,
1937 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1938 			}
1939 		}
1940 	} else if (done_namei) {
1941 		KASSERT(create == NFSV4OPEN_CREATE,
1942 		    ("nfsvno_open: not create"));
1943 		/*
1944 		 * done_namei is set when nfsvno_namei() has completed
1945 		 * successfully, but a subsequent error was set in
1946 		 * nd_repstat.  As such, cleanup of the nfsvno_namei()
1947 		 * results is required.
1948 		 */
1949 		nfsvno_relpathbuf(ndp);
1950 		if (ndp->ni_dvp == ndp->ni_vp)
1951 			vrele(ndp->ni_dvp);
1952 		else
1953 			vput(ndp->ni_dvp);
1954 		if (ndp->ni_vp)
1955 			vput(ndp->ni_vp);
1956 	}
1957 	*vpp = vp;
1958 
1959 	NFSEXITCODE2(0, nd);
1960 }
1961 
1962 /*
1963  * Updates the file rev and sets the mtime and ctime
1964  * to the current clock time, returning the va_filerev and va_Xtime
1965  * values.
1966  * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1967  */
1968 int
1969 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1970     struct nfsrv_descript *nd, struct thread *p)
1971 {
1972 	struct vattr va;
1973 
1974 	VATTR_NULL(&va);
1975 	vfs_timestamp(&va.va_mtime);
1976 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1977 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1978 		if (VN_IS_DOOMED(vp))
1979 			return (ESTALE);
1980 	}
1981 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
1982 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1983 	return (0);
1984 }
1985 
1986 /*
1987  * Glue routine to nfsv4_fillattr().
1988  */
1989 int
1990 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1991     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1992     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1993     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1994 {
1995 	struct statfs *sf;
1996 	int error;
1997 
1998 	sf = NULL;
1999 	if (nfsrv_devidcnt > 0 &&
2000 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
2001 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
2002 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
2003 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
2004 		error = nfsrv_pnfsstatfs(sf, mp);
2005 		if (error != 0) {
2006 			free(sf, M_TEMP);
2007 			sf = NULL;
2008 		}
2009 	}
2010 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
2011 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
2012 	    mounted_on_fileno, sf);
2013 	free(sf, M_TEMP);
2014 	NFSEXITCODE2(0, nd);
2015 	return (error);
2016 }
2017 
2018 /* Since the Readdir vnode ops vary, put the entire functions in here. */
2019 /*
2020  * nfs readdir service
2021  * - mallocs what it thinks is enough to read
2022  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
2023  * - calls VOP_READDIR()
2024  * - loops around building the reply
2025  *	if the output generated exceeds count break out of loop
2026  *	The NFSM_CLGET macro is used here so that the reply will be packed
2027  *	tightly in mbuf clusters.
2028  * - it trims out records with d_fileno == 0
2029  *	this doesn't matter for Unix clients, but they might confuse clients
2030  *	for other os'.
2031  * - it trims out records with d_type == DT_WHT
2032  *	these cannot be seen through NFS (unless we extend the protocol)
2033  *     The alternate call nfsrvd_readdirplus() does lookups as well.
2034  * PS: The NFS protocol spec. does not clarify what the "count" byte
2035  *	argument is a count of.. just name strings and file id's or the
2036  *	entire reply rpc or ...
2037  *	I tried just file name and id sizes and it confused the Sun client,
2038  *	so I am using the full rpc size now. The "paranoia.." comment refers
2039  *	to including the status longwords that are not a part of the dir.
2040  *	"entry" structures, but are in the rpc.
2041  */
2042 int
2043 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
2044     struct vnode *vp, struct nfsexstuff *exp)
2045 {
2046 	struct dirent *dp;
2047 	u_int32_t *tl;
2048 	int dirlen;
2049 	char *cpos, *cend, *rbuf;
2050 	struct nfsvattr at;
2051 	int nlen, error = 0, getret = 1;
2052 	int siz, cnt, fullsiz, eofflag, ncookies;
2053 	u_int64_t off, toff, verf __unused;
2054 	uint64_t *cookies = NULL, *cookiep;
2055 	struct uio io;
2056 	struct iovec iv;
2057 	int is_ufs;
2058 	struct thread *p = curthread;
2059 
2060 	if (nd->nd_repstat) {
2061 		nfsrv_postopattr(nd, getret, &at);
2062 		goto out;
2063 	}
2064 	if (nd->nd_flag & ND_NFSV2) {
2065 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2066 		off = fxdr_unsigned(u_quad_t, *tl++);
2067 	} else {
2068 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2069 		off = fxdr_hyper(tl);
2070 		tl += 2;
2071 		verf = fxdr_hyper(tl);
2072 		tl += 2;
2073 	}
2074 	toff = off;
2075 	cnt = fxdr_unsigned(int, *tl);
2076 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2077 		cnt = NFS_SRVMAXDATA(nd);
2078 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2079 	fullsiz = siz;
2080 	if (nd->nd_flag & ND_NFSV3) {
2081 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
2082 		    NULL);
2083 #if 0
2084 		/*
2085 		 * va_filerev is not sufficient as a cookie verifier,
2086 		 * since it is not supposed to change when entries are
2087 		 * removed/added unless that offset cookies returned to
2088 		 * the client are no longer valid.
2089 		 */
2090 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
2091 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2092 #endif
2093 	}
2094 	if (!nd->nd_repstat && vp->v_type != VDIR)
2095 		nd->nd_repstat = NFSERR_NOTDIR;
2096 	if (nd->nd_repstat == 0 && cnt == 0) {
2097 		if (nd->nd_flag & ND_NFSV2)
2098 			/* NFSv2 does not have NFSERR_TOOSMALL */
2099 			nd->nd_repstat = EPERM;
2100 		else
2101 			nd->nd_repstat = NFSERR_TOOSMALL;
2102 	}
2103 	if (!nd->nd_repstat)
2104 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2105 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2106 		    NFSACCCHK_VPISLOCKED, NULL);
2107 	if (nd->nd_repstat) {
2108 		vput(vp);
2109 		if (nd->nd_flag & ND_NFSV3)
2110 			nfsrv_postopattr(nd, getret, &at);
2111 		goto out;
2112 	}
2113 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2114 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2115 again:
2116 	eofflag = 0;
2117 	if (cookies) {
2118 		free(cookies, M_TEMP);
2119 		cookies = NULL;
2120 	}
2121 
2122 	iv.iov_base = rbuf;
2123 	iv.iov_len = siz;
2124 	io.uio_iov = &iv;
2125 	io.uio_iovcnt = 1;
2126 	io.uio_offset = (off_t)off;
2127 	io.uio_resid = siz;
2128 	io.uio_segflg = UIO_SYSSPACE;
2129 	io.uio_rw = UIO_READ;
2130 	io.uio_td = NULL;
2131 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2132 	    &cookies);
2133 	off = (u_int64_t)io.uio_offset;
2134 	if (io.uio_resid)
2135 		siz -= io.uio_resid;
2136 
2137 	if (!cookies && !nd->nd_repstat)
2138 		nd->nd_repstat = NFSERR_PERM;
2139 	if (nd->nd_flag & ND_NFSV3) {
2140 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2141 		if (!nd->nd_repstat)
2142 			nd->nd_repstat = getret;
2143 	}
2144 
2145 	/*
2146 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
2147 	 */
2148 	if (nd->nd_repstat) {
2149 		vput(vp);
2150 		free(rbuf, M_TEMP);
2151 		if (cookies)
2152 			free(cookies, M_TEMP);
2153 		if (nd->nd_flag & ND_NFSV3)
2154 			nfsrv_postopattr(nd, getret, &at);
2155 		goto out;
2156 	}
2157 	/*
2158 	 * If nothing read, return eof
2159 	 * rpc reply
2160 	 */
2161 	if (siz == 0) {
2162 		vput(vp);
2163 		if (nd->nd_flag & ND_NFSV2) {
2164 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2165 		} else {
2166 			nfsrv_postopattr(nd, getret, &at);
2167 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2168 			txdr_hyper(at.na_filerev, tl);
2169 			tl += 2;
2170 		}
2171 		*tl++ = newnfs_false;
2172 		*tl = newnfs_true;
2173 		free(rbuf, M_TEMP);
2174 		free(cookies, M_TEMP);
2175 		goto out;
2176 	}
2177 
2178 	/*
2179 	 * Check for degenerate cases of nothing useful read.
2180 	 * If so go try again
2181 	 */
2182 	cpos = rbuf;
2183 	cend = rbuf + siz;
2184 	dp = (struct dirent *)cpos;
2185 	cookiep = cookies;
2186 
2187 	/*
2188 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2189 	 * directory offset up to a block boundary, so it is necessary to
2190 	 * skip over the records that precede the requested offset. This
2191 	 * requires the assumption that file offset cookies monotonically
2192 	 * increase.
2193 	 */
2194 	while (cpos < cend && ncookies > 0 &&
2195 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2196 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2197 		cpos += dp->d_reclen;
2198 		dp = (struct dirent *)cpos;
2199 		cookiep++;
2200 		ncookies--;
2201 	}
2202 	if (cpos >= cend || ncookies == 0) {
2203 		siz = fullsiz;
2204 		toff = off;
2205 		goto again;
2206 	}
2207 	vput(vp);
2208 
2209 	/*
2210 	 * If cnt > MCLBYTES and the reply will not be saved, use
2211 	 * ext_pgs mbufs for TLS.
2212 	 * For NFSv4.0, we do not know for sure if the reply will
2213 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2214 	 */
2215 	if (cnt > MCLBYTES && siz > MCLBYTES &&
2216 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2217 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2218 		nd->nd_flag |= ND_EXTPG;
2219 
2220 	/*
2221 	 * dirlen is the size of the reply, including all XDR and must
2222 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2223 	 * if the XDR should be included in "count", but to be safe, we do.
2224 	 * (Include the two booleans at the end of the reply in dirlen now.)
2225 	 */
2226 	if (nd->nd_flag & ND_NFSV3) {
2227 		nfsrv_postopattr(nd, getret, &at);
2228 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2229 		txdr_hyper(at.na_filerev, tl);
2230 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2231 	} else {
2232 		dirlen = 2 * NFSX_UNSIGNED;
2233 	}
2234 
2235 	/* Loop through the records and build reply */
2236 	while (cpos < cend && ncookies > 0) {
2237 		nlen = dp->d_namlen;
2238 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2239 			nlen <= NFS_MAXNAMLEN) {
2240 			if (nd->nd_flag & ND_NFSV3)
2241 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2242 			else
2243 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2244 			if (dirlen > cnt) {
2245 				eofflag = 0;
2246 				break;
2247 			}
2248 
2249 			/*
2250 			 * Build the directory record xdr from
2251 			 * the dirent entry.
2252 			 */
2253 			if (nd->nd_flag & ND_NFSV3) {
2254 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2255 				*tl++ = newnfs_true;
2256 				txdr_hyper(dp->d_fileno, tl);
2257 			} else {
2258 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2259 				*tl++ = newnfs_true;
2260 				*tl = txdr_unsigned(dp->d_fileno);
2261 			}
2262 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2263 			if (nd->nd_flag & ND_NFSV3) {
2264 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2265 				txdr_hyper(*cookiep, tl);
2266 			} else {
2267 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2268 				*tl = txdr_unsigned(*cookiep);
2269 			}
2270 		}
2271 		cpos += dp->d_reclen;
2272 		dp = (struct dirent *)cpos;
2273 		cookiep++;
2274 		ncookies--;
2275 	}
2276 	if (cpos < cend)
2277 		eofflag = 0;
2278 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2279 	*tl++ = newnfs_false;
2280 	if (eofflag)
2281 		*tl = newnfs_true;
2282 	else
2283 		*tl = newnfs_false;
2284 	free(rbuf, M_TEMP);
2285 	free(cookies, M_TEMP);
2286 
2287 out:
2288 	NFSEXITCODE2(0, nd);
2289 	return (0);
2290 nfsmout:
2291 	vput(vp);
2292 	NFSEXITCODE2(error, nd);
2293 	return (error);
2294 }
2295 
2296 /*
2297  * Readdirplus for V3 and Readdir for V4.
2298  */
2299 int
2300 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2301     struct vnode *vp, struct nfsexstuff *exp)
2302 {
2303 	struct dirent *dp;
2304 	u_int32_t *tl;
2305 	int dirlen;
2306 	char *cpos, *cend, *rbuf;
2307 	struct vnode *nvp;
2308 	fhandle_t nfh;
2309 	struct nfsvattr nva, at, *nvap = &nva;
2310 	struct mbuf *mb0, *mb1;
2311 	struct nfsreferral *refp;
2312 	int nlen, r, error = 0, getret = 1, usevget = 1;
2313 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2314 	caddr_t bpos0, bpos1;
2315 	u_int64_t off, toff, verf __unused;
2316 	uint64_t *cookies = NULL, *cookiep;
2317 	nfsattrbit_t attrbits, rderrbits, savbits, refbits;
2318 	struct uio io;
2319 	struct iovec iv;
2320 	struct componentname cn;
2321 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2322 	struct mount *mp, *new_mp;
2323 	uint64_t mounted_on_fileno;
2324 	struct thread *p = curthread;
2325 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2326 
2327 	if (nd->nd_repstat) {
2328 		nfsrv_postopattr(nd, getret, &at);
2329 		goto out;
2330 	}
2331 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2332 	off = fxdr_hyper(tl);
2333 	toff = off;
2334 	tl += 2;
2335 	verf = fxdr_hyper(tl);
2336 	tl += 2;
2337 	siz = fxdr_unsigned(int, *tl++);
2338 	cnt = fxdr_unsigned(int, *tl);
2339 
2340 	/*
2341 	 * Use the server's maximum data transfer size as the upper bound
2342 	 * on reply datalen.
2343 	 */
2344 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2345 		cnt = NFS_SRVMAXDATA(nd);
2346 
2347 	/*
2348 	 * siz is a "hint" of how much directory information (name, fileid,
2349 	 * cookie) should be in the reply. At least one client "hints" 0,
2350 	 * so I set it to cnt for that case. I also round it up to the
2351 	 * next multiple of DIRBLKSIZ.
2352 	 * Since the size of a Readdirplus directory entry reply will always
2353 	 * be greater than a directory entry returned by VOP_READDIR(), it
2354 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2355 	 * VOP_READDIR().
2356 	 */
2357 	if (siz <= 0)
2358 		siz = cnt;
2359 	else if (siz > NFS_SRVMAXDATA(nd))
2360 		siz = NFS_SRVMAXDATA(nd);
2361 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2362 
2363 	if (nd->nd_flag & ND_NFSV4) {
2364 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2365 		if (error)
2366 			goto nfsmout;
2367 		NFSSET_ATTRBIT(&savbits, &attrbits);
2368 		NFSSET_ATTRBIT(&refbits, &attrbits);
2369 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2370 		NFSZERO_ATTRBIT(&rderrbits);
2371 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2372 		/*
2373 		 * If these 4 bits are the only attributes requested by the
2374 		 * client, they can be satisfied without acquiring the vnode
2375 		 * for the file object unless it is a directory.
2376 		 * This will be indicated by savbits being all 0s.
2377 		 */
2378 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_TYPE);
2379 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_FILEID);
2380 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_MOUNTEDONFILEID);
2381 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_RDATTRERROR);
2382 	} else {
2383 		NFSZERO_ATTRBIT(&attrbits);
2384 	}
2385 	fullsiz = siz;
2386 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2387 #if 0
2388 	if (!nd->nd_repstat) {
2389 	    if (off && verf != at.na_filerev) {
2390 		/*
2391 		 * va_filerev is not sufficient as a cookie verifier,
2392 		 * since it is not supposed to change when entries are
2393 		 * removed/added unless that offset cookies returned to
2394 		 * the client are no longer valid.
2395 		 */
2396 		if (nd->nd_flag & ND_NFSV4) {
2397 			nd->nd_repstat = NFSERR_NOTSAME;
2398 		} else {
2399 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2400 		}
2401 	    }
2402 	}
2403 #endif
2404 	if (!nd->nd_repstat && vp->v_type != VDIR)
2405 		nd->nd_repstat = NFSERR_NOTDIR;
2406 	if (!nd->nd_repstat && cnt == 0)
2407 		nd->nd_repstat = NFSERR_TOOSMALL;
2408 	if (!nd->nd_repstat)
2409 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2410 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2411 		    NFSACCCHK_VPISLOCKED, NULL);
2412 	if (nd->nd_repstat) {
2413 		vput(vp);
2414 		if (nd->nd_flag & ND_NFSV3)
2415 			nfsrv_postopattr(nd, getret, &at);
2416 		goto out;
2417 	}
2418 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2419 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2420 
2421 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2422 again:
2423 	eofflag = 0;
2424 	if (cookies) {
2425 		free(cookies, M_TEMP);
2426 		cookies = NULL;
2427 	}
2428 
2429 	iv.iov_base = rbuf;
2430 	iv.iov_len = siz;
2431 	io.uio_iov = &iv;
2432 	io.uio_iovcnt = 1;
2433 	io.uio_offset = (off_t)off;
2434 	io.uio_resid = siz;
2435 	io.uio_segflg = UIO_SYSSPACE;
2436 	io.uio_rw = UIO_READ;
2437 	io.uio_td = NULL;
2438 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2439 	    &cookies);
2440 	off = (u_int64_t)io.uio_offset;
2441 	if (io.uio_resid)
2442 		siz -= io.uio_resid;
2443 
2444 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2445 
2446 	if (!cookies && !nd->nd_repstat)
2447 		nd->nd_repstat = NFSERR_PERM;
2448 	if (!nd->nd_repstat)
2449 		nd->nd_repstat = getret;
2450 	if (nd->nd_repstat) {
2451 		vput(vp);
2452 		if (cookies)
2453 			free(cookies, M_TEMP);
2454 		free(rbuf, M_TEMP);
2455 		if (nd->nd_flag & ND_NFSV3)
2456 			nfsrv_postopattr(nd, getret, &at);
2457 		goto out;
2458 	}
2459 	/*
2460 	 * If nothing read, return eof
2461 	 * rpc reply
2462 	 */
2463 	if (siz == 0) {
2464 		vput(vp);
2465 		if (nd->nd_flag & ND_NFSV3)
2466 			nfsrv_postopattr(nd, getret, &at);
2467 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2468 		txdr_hyper(at.na_filerev, tl);
2469 		tl += 2;
2470 		*tl++ = newnfs_false;
2471 		*tl = newnfs_true;
2472 		free(cookies, M_TEMP);
2473 		free(rbuf, M_TEMP);
2474 		goto out;
2475 	}
2476 
2477 	/*
2478 	 * Check for degenerate cases of nothing useful read.
2479 	 * If so go try again
2480 	 */
2481 	cpos = rbuf;
2482 	cend = rbuf + siz;
2483 	dp = (struct dirent *)cpos;
2484 	cookiep = cookies;
2485 
2486 	/*
2487 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2488 	 * directory offset up to a block boundary, so it is necessary to
2489 	 * skip over the records that precede the requested offset. This
2490 	 * requires the assumption that file offset cookies monotonically
2491 	 * increase.
2492 	 */
2493 	while (cpos < cend && ncookies > 0 &&
2494 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2495 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2496 	   ((nd->nd_flag & ND_NFSV4) &&
2497 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2498 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2499 		cpos += dp->d_reclen;
2500 		dp = (struct dirent *)cpos;
2501 		cookiep++;
2502 		ncookies--;
2503 	}
2504 	if (cpos >= cend || ncookies == 0) {
2505 		siz = fullsiz;
2506 		toff = off;
2507 		goto again;
2508 	}
2509 
2510 	/*
2511 	 * Busy the file system so that the mount point won't go away
2512 	 * and, as such, VFS_VGET() can be used safely.
2513 	 */
2514 	mp = vp->v_mount;
2515 	vfs_ref(mp);
2516 	NFSVOPUNLOCK(vp);
2517 	nd->nd_repstat = vfs_busy(mp, 0);
2518 	vfs_rel(mp);
2519 	if (nd->nd_repstat != 0) {
2520 		vrele(vp);
2521 		free(cookies, M_TEMP);
2522 		free(rbuf, M_TEMP);
2523 		if (nd->nd_flag & ND_NFSV3)
2524 			nfsrv_postopattr(nd, getret, &at);
2525 		goto out;
2526 	}
2527 
2528 	/*
2529 	 * Check to see if entries in this directory can be safely acquired
2530 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2531 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2532 	 * automount of the snapshot directory that is required will
2533 	 * be done.
2534 	 * This needs to be done here for NFSv4, since NFSv4 never does
2535 	 * a VFS_VGET() for "." or "..".
2536 	 */
2537 	if (is_zfs == 1) {
2538 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2539 		if (r == EOPNOTSUPP) {
2540 			usevget = 0;
2541 			cn.cn_nameiop = LOOKUP;
2542 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2543 			cn.cn_cred = nd->nd_cred;
2544 		} else if (r == 0)
2545 			vput(nvp);
2546 	}
2547 
2548 	/*
2549 	 * If the reply is likely to exceed MCLBYTES and the reply will
2550 	 * not be saved, use ext_pgs mbufs for TLS.
2551 	 * It is difficult to predict how large each entry will be and
2552 	 * how many entries have been read, so just assume the directory
2553 	 * entries grow by a factor of 4 when attributes are included.
2554 	 * For NFSv4.0, we do not know for sure if the reply will
2555 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2556 	 */
2557 	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
2558 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2559 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2560 		nd->nd_flag |= ND_EXTPG;
2561 
2562 	/*
2563 	 * Save this position, in case there is an error before one entry
2564 	 * is created.
2565 	 */
2566 	mb0 = nd->nd_mb;
2567 	bpos0 = nd->nd_bpos;
2568 	bextpg0 = nd->nd_bextpg;
2569 	bextpgsiz0 = nd->nd_bextpgsiz;
2570 
2571 	/*
2572 	 * Fill in the first part of the reply.
2573 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2574 	 * (Include the two booleans at the end of the reply in dirlen now,
2575 	 *  so we recognize when we have exceeded cnt.)
2576 	 */
2577 	if (nd->nd_flag & ND_NFSV3) {
2578 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2579 		nfsrv_postopattr(nd, getret, &at);
2580 	} else {
2581 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2582 	}
2583 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2584 	txdr_hyper(at.na_filerev, tl);
2585 
2586 	/*
2587 	 * Save this position, in case there is an empty reply needed.
2588 	 */
2589 	mb1 = nd->nd_mb;
2590 	bpos1 = nd->nd_bpos;
2591 	bextpg1 = nd->nd_bextpg;
2592 	bextpgsiz1 = nd->nd_bextpgsiz;
2593 
2594 	/* Loop through the records and build reply */
2595 	entrycnt = 0;
2596 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2597 		nlen = dp->d_namlen;
2598 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2599 		    nlen <= NFS_MAXNAMLEN &&
2600 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2601 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2602 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2603 			/*
2604 			 * Save the current position in the reply, in case
2605 			 * this entry exceeds cnt.
2606 			 */
2607 			mb1 = nd->nd_mb;
2608 			bpos1 = nd->nd_bpos;
2609 			bextpg1 = nd->nd_bextpg;
2610 			bextpgsiz1 = nd->nd_bextpgsiz;
2611 
2612 			/*
2613 			 * For readdir_and_lookup get the vnode using
2614 			 * the file number.
2615 			 */
2616 			nvp = NULL;
2617 			refp = NULL;
2618 			r = 0;
2619 			at_root = 0;
2620 			needs_unbusy = 0;
2621 			new_mp = mp;
2622 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2623 			if ((nd->nd_flag & ND_NFSV3) ||
2624 			    NFSNONZERO_ATTRBIT(&savbits) ||
2625 			    dp->d_type == DT_UNKNOWN ||
2626 			    (dp->d_type == DT_DIR &&
2627 			     nfsrv_enable_crossmntpt != 0)) {
2628 				if (nd->nd_flag & ND_NFSV4)
2629 					refp = nfsv4root_getreferral(NULL,
2630 					    vp, dp->d_fileno);
2631 				if (refp == NULL) {
2632 					if (usevget)
2633 						r = VFS_VGET(mp, dp->d_fileno,
2634 						    LK_SHARED, &nvp);
2635 					else
2636 						r = EOPNOTSUPP;
2637 					if (r == EOPNOTSUPP) {
2638 						if (usevget) {
2639 							usevget = 0;
2640 							cn.cn_nameiop = LOOKUP;
2641 							cn.cn_lkflags =
2642 							    LK_SHARED |
2643 							    LK_RETRY;
2644 							cn.cn_cred =
2645 							    nd->nd_cred;
2646 						}
2647 						cn.cn_nameptr = dp->d_name;
2648 						cn.cn_namelen = nlen;
2649 						cn.cn_flags = ISLASTCN |
2650 						    NOFOLLOW | LOCKLEAF;
2651 						if (nlen == 2 &&
2652 						    dp->d_name[0] == '.' &&
2653 						    dp->d_name[1] == '.')
2654 							cn.cn_flags |=
2655 							    ISDOTDOT;
2656 						if (NFSVOPLOCK(vp, LK_SHARED)
2657 						    != 0) {
2658 							nd->nd_repstat = EPERM;
2659 							break;
2660 						}
2661 						if ((vp->v_vflag & VV_ROOT) != 0
2662 						    && (cn.cn_flags & ISDOTDOT)
2663 						    != 0) {
2664 							vref(vp);
2665 							nvp = vp;
2666 							r = 0;
2667 						} else {
2668 							r = VOP_LOOKUP(vp, &nvp,
2669 							    &cn);
2670 							if (vp != nvp)
2671 								NFSVOPUNLOCK(vp);
2672 						}
2673 					}
2674 
2675 					/*
2676 					 * For NFSv4, check to see if nvp is
2677 					 * a mount point and get the mount
2678 					 * point vnode, as required.
2679 					 */
2680 					if (r == 0 &&
2681 					    nfsrv_enable_crossmntpt != 0 &&
2682 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2683 					    nvp->v_type == VDIR &&
2684 					    nvp->v_mountedhere != NULL) {
2685 						new_mp = nvp->v_mountedhere;
2686 						r = vfs_busy(new_mp, 0);
2687 						vput(nvp);
2688 						nvp = NULL;
2689 						if (r == 0) {
2690 							r = VFS_ROOT(new_mp,
2691 							    LK_SHARED, &nvp);
2692 							needs_unbusy = 1;
2693 							if (r == 0)
2694 								at_root = 1;
2695 						}
2696 					}
2697 				}
2698 
2699 				/*
2700 				 * If we failed to look up the entry, then it
2701 				 * has become invalid, most likely removed.
2702 				 */
2703 				if (r != 0) {
2704 					if (needs_unbusy)
2705 						vfs_unbusy(new_mp);
2706 					goto invalid;
2707 				}
2708 				KASSERT(refp != NULL || nvp != NULL,
2709 				    ("%s: undetected lookup error", __func__));
2710 
2711 				if (refp == NULL &&
2712 				    ((nd->nd_flag & ND_NFSV3) ||
2713 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2714 					r = nfsvno_getfh(nvp, &nfh, p);
2715 					if (!r)
2716 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2717 						1, &attrbits);
2718 					if (r == 0 && is_zfs == 1 &&
2719 					    nfsrv_enable_crossmntpt != 0 &&
2720 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2721 					    nvp->v_type == VDIR &&
2722 					    vp->v_mount != nvp->v_mount) {
2723 					    /*
2724 					     * For a ZFS snapshot, there is a
2725 					     * pseudo mount that does not set
2726 					     * v_mountedhere, so it needs to
2727 					     * be detected via a different
2728 					     * mount structure.
2729 					     */
2730 					    at_root = 1;
2731 					    if (new_mp == mp)
2732 						new_mp = nvp->v_mount;
2733 					}
2734 				}
2735 
2736 				/*
2737 				 * If we failed to get attributes of the entry,
2738 				 * then just skip it for NFSv3 (the traditional
2739 				 * behavior in the old NFS server).
2740 				 * For NFSv4 the behavior is controlled by
2741 				 * RDATTRERROR: we either ignore the error or
2742 				 * fail the request.
2743 				 * The exception is EOPNOTSUPP, which can be
2744 				 * returned by nfsvno_getfh() for certain
2745 				 * file systems, such as devfs.  This indicates
2746 				 * that the file system cannot be exported,
2747 				 * so just skip over the entry.
2748 				 * Note that RDATTRERROR is never set for NFSv3.
2749 				 */
2750 				if (r != 0) {
2751 					if (!NFSISSET_ATTRBIT(&attrbits,
2752 					    NFSATTRBIT_RDATTRERROR) ||
2753 					    r == EOPNOTSUPP) {
2754 						vput(nvp);
2755 						if (needs_unbusy != 0)
2756 							vfs_unbusy(new_mp);
2757 						if ((nd->nd_flag & ND_NFSV3) ||
2758 						    r == EOPNOTSUPP)
2759 							goto invalid;
2760 						nd->nd_repstat = r;
2761 						break;
2762 					}
2763 				}
2764 			} else if (NFSNONZERO_ATTRBIT(&attrbits)) {
2765 				/* Only need Type and/or Fileid. */
2766 				VATTR_NULL(&nvap->na_vattr);
2767 				nvap->na_fileid = dp->d_fileno;
2768 				nvap->na_type = NFS_DTYPETOVTYPE(dp->d_type);
2769 			}
2770 
2771 			/*
2772 			 * Build the directory record xdr
2773 			 */
2774 			if (nd->nd_flag & ND_NFSV3) {
2775 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2776 				*tl++ = newnfs_true;
2777 				txdr_hyper(dp->d_fileno, tl);
2778 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2779 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2780 				txdr_hyper(*cookiep, tl);
2781 				nfsrv_postopattr(nd, 0, nvap);
2782 				dirlen += nfsm_fhtom(NULL, nd, (u_int8_t *)&nfh,
2783 				    0, 1);
2784 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2785 				if (nvp != NULL)
2786 					vput(nvp);
2787 			} else {
2788 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2789 				*tl++ = newnfs_true;
2790 				txdr_hyper(*cookiep, tl);
2791 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2792 				if (nvp != NULL) {
2793 					supports_nfsv4acls =
2794 					    nfs_supportsnfsv4acls(nvp);
2795 					NFSVOPUNLOCK(nvp);
2796 				} else
2797 					supports_nfsv4acls = 0;
2798 				if (refp != NULL) {
2799 					dirlen += nfsrv_putreferralattr(nd,
2800 					    &refbits, refp, 0,
2801 					    &nd->nd_repstat);
2802 					if (nd->nd_repstat) {
2803 						if (nvp != NULL)
2804 							vrele(nvp);
2805 						if (needs_unbusy != 0)
2806 							vfs_unbusy(new_mp);
2807 						break;
2808 					}
2809 				} else if (r) {
2810 					dirlen += nfsvno_fillattr(nd, new_mp,
2811 					    nvp, nvap, &nfh, r, &rderrbits,
2812 					    nd->nd_cred, p, isdgram, 0,
2813 					    supports_nfsv4acls, at_root,
2814 					    mounted_on_fileno);
2815 				} else {
2816 					dirlen += nfsvno_fillattr(nd, new_mp,
2817 					    nvp, nvap, &nfh, r, &attrbits,
2818 					    nd->nd_cred, p, isdgram, 0,
2819 					    supports_nfsv4acls, at_root,
2820 					    mounted_on_fileno);
2821 				}
2822 				if (nvp != NULL)
2823 					vrele(nvp);
2824 				dirlen += (3 * NFSX_UNSIGNED);
2825 			}
2826 			if (needs_unbusy != 0)
2827 				vfs_unbusy(new_mp);
2828 			if (dirlen <= cnt)
2829 				entrycnt++;
2830 		}
2831 invalid:
2832 		cpos += dp->d_reclen;
2833 		dp = (struct dirent *)cpos;
2834 		cookiep++;
2835 		ncookies--;
2836 	}
2837 	vrele(vp);
2838 	vfs_unbusy(mp);
2839 
2840 	/*
2841 	 * If dirlen > cnt, we must strip off the last entry. If that
2842 	 * results in an empty reply, report NFSERR_TOOSMALL.
2843 	 */
2844 	if (dirlen > cnt || nd->nd_repstat) {
2845 		if (!nd->nd_repstat && entrycnt == 0)
2846 			nd->nd_repstat = NFSERR_TOOSMALL;
2847 		if (nd->nd_repstat) {
2848 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2849 			if (nd->nd_flag & ND_NFSV3)
2850 				nfsrv_postopattr(nd, getret, &at);
2851 		} else
2852 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2853 		eofflag = 0;
2854 	} else if (cpos < cend)
2855 		eofflag = 0;
2856 	if (!nd->nd_repstat) {
2857 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2858 		*tl++ = newnfs_false;
2859 		if (eofflag)
2860 			*tl = newnfs_true;
2861 		else
2862 			*tl = newnfs_false;
2863 	}
2864 	free(cookies, M_TEMP);
2865 	free(rbuf, M_TEMP);
2866 
2867 out:
2868 	NFSEXITCODE2(0, nd);
2869 	return (0);
2870 nfsmout:
2871 	vput(vp);
2872 	NFSEXITCODE2(error, nd);
2873 	return (error);
2874 }
2875 
2876 /*
2877  * Get the settable attributes out of the mbuf list.
2878  * (Return 0 or EBADRPC)
2879  */
2880 int
2881 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2882     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2883 {
2884 	u_int32_t *tl;
2885 	struct nfsv2_sattr *sp;
2886 	int error = 0, toclient = 0;
2887 
2888 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2889 	case ND_NFSV2:
2890 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2891 		/*
2892 		 * Some old clients didn't fill in the high order 16bits.
2893 		 * --> check the low order 2 bytes for 0xffff
2894 		 */
2895 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2896 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2897 		if (sp->sa_uid != newnfs_xdrneg1)
2898 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2899 		if (sp->sa_gid != newnfs_xdrneg1)
2900 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2901 		if (sp->sa_size != newnfs_xdrneg1)
2902 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2903 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2904 #ifdef notyet
2905 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2906 #else
2907 			nvap->na_atime.tv_sec =
2908 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2909 			nvap->na_atime.tv_nsec = 0;
2910 #endif
2911 		}
2912 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2913 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2914 		break;
2915 	case ND_NFSV3:
2916 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2917 		if (*tl == newnfs_true) {
2918 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2919 			nvap->na_mode = nfstov_mode(*tl);
2920 		}
2921 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2922 		if (*tl == newnfs_true) {
2923 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2924 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2925 		}
2926 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2927 		if (*tl == newnfs_true) {
2928 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2929 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2930 		}
2931 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2932 		if (*tl == newnfs_true) {
2933 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2934 			nvap->na_size = fxdr_hyper(tl);
2935 		}
2936 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2937 		switch (fxdr_unsigned(int, *tl)) {
2938 		case NFSV3SATTRTIME_TOCLIENT:
2939 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2940 			fxdr_nfsv3time(tl, &nvap->na_atime);
2941 			toclient = 1;
2942 			break;
2943 		case NFSV3SATTRTIME_TOSERVER:
2944 			vfs_timestamp(&nvap->na_atime);
2945 			nvap->na_vaflags |= VA_UTIMES_NULL;
2946 			break;
2947 		}
2948 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2949 		switch (fxdr_unsigned(int, *tl)) {
2950 		case NFSV3SATTRTIME_TOCLIENT:
2951 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2952 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2953 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2954 			break;
2955 		case NFSV3SATTRTIME_TOSERVER:
2956 			vfs_timestamp(&nvap->na_mtime);
2957 			if (!toclient)
2958 				nvap->na_vaflags |= VA_UTIMES_NULL;
2959 			break;
2960 		}
2961 		break;
2962 	case ND_NFSV4:
2963 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2964 	}
2965 nfsmout:
2966 	NFSEXITCODE2(error, nd);
2967 	return (error);
2968 }
2969 
2970 /*
2971  * Handle the setable attributes for V4.
2972  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2973  */
2974 int
2975 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2976     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2977 {
2978 	u_int32_t *tl;
2979 	int attrsum = 0;
2980 	int i, j;
2981 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2982 	int moderet, toclient = 0;
2983 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2984 	uid_t uid;
2985 	gid_t gid;
2986 	u_short mode, mask;		/* Same type as va_mode. */
2987 	struct vattr va;
2988 
2989 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2990 	if (error)
2991 		goto nfsmout;
2992 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2993 	attrsize = fxdr_unsigned(int, *tl);
2994 
2995 	/*
2996 	 * Loop around getting the setable attributes. If an unsupported
2997 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2998 	 */
2999 	if (retnotsup) {
3000 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3001 		bitpos = NFSATTRBIT_MAX;
3002 	} else {
3003 		bitpos = 0;
3004 	}
3005 	moderet = 0;
3006 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
3007 	    if (attrsum > attrsize) {
3008 		error = NFSERR_BADXDR;
3009 		goto nfsmout;
3010 	    }
3011 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
3012 		switch (bitpos) {
3013 		case NFSATTRBIT_SIZE:
3014 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
3015                      if (vp != NULL && vp->v_type != VREG) {
3016                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
3017                                 NFSERR_INVAL;
3018                             goto nfsmout;
3019 			}
3020 			nvap->na_size = fxdr_hyper(tl);
3021 			attrsum += NFSX_HYPER;
3022 			break;
3023 		case NFSATTRBIT_ACL:
3024 			error = nfsrv_dissectacl(nd, aclp, true, &aceerr,
3025 			    &aclsize, p);
3026 			if (error)
3027 				goto nfsmout;
3028 			if (aceerr && !nd->nd_repstat)
3029 				nd->nd_repstat = aceerr;
3030 			attrsum += aclsize;
3031 			break;
3032 		case NFSATTRBIT_ARCHIVE:
3033 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3034 			if (!nd->nd_repstat)
3035 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3036 			attrsum += NFSX_UNSIGNED;
3037 			break;
3038 		case NFSATTRBIT_HIDDEN:
3039 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3040 			if (!nd->nd_repstat)
3041 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3042 			attrsum += NFSX_UNSIGNED;
3043 			break;
3044 		case NFSATTRBIT_MIMETYPE:
3045 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3046 			i = fxdr_unsigned(int, *tl);
3047 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3048 			if (error)
3049 				goto nfsmout;
3050 			if (!nd->nd_repstat)
3051 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3052 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
3053 			break;
3054 		case NFSATTRBIT_MODE:
3055 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
3056 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3057 			nvap->na_mode = nfstov_mode(*tl);
3058 			attrsum += NFSX_UNSIGNED;
3059 			break;
3060 		case NFSATTRBIT_OWNER:
3061 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3062 			j = fxdr_unsigned(int, *tl);
3063 			if (j < 0) {
3064 				error = NFSERR_BADXDR;
3065 				goto nfsmout;
3066 			}
3067 			if (j > NFSV4_SMALLSTR)
3068 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3069 			else
3070 				cp = namestr;
3071 			error = nfsrv_mtostr(nd, cp, j);
3072 			if (error) {
3073 				if (j > NFSV4_SMALLSTR)
3074 					free(cp, M_NFSSTRING);
3075 				goto nfsmout;
3076 			}
3077 			if (!nd->nd_repstat) {
3078 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
3079 				    &uid);
3080 				if (!nd->nd_repstat)
3081 					nvap->na_uid = uid;
3082 			}
3083 			if (j > NFSV4_SMALLSTR)
3084 				free(cp, M_NFSSTRING);
3085 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3086 			break;
3087 		case NFSATTRBIT_OWNERGROUP:
3088 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3089 			j = fxdr_unsigned(int, *tl);
3090 			if (j < 0) {
3091 				error = NFSERR_BADXDR;
3092 				goto nfsmout;
3093 			}
3094 			if (j > NFSV4_SMALLSTR)
3095 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3096 			else
3097 				cp = namestr;
3098 			error = nfsrv_mtostr(nd, cp, j);
3099 			if (error) {
3100 				if (j > NFSV4_SMALLSTR)
3101 					free(cp, M_NFSSTRING);
3102 				goto nfsmout;
3103 			}
3104 			if (!nd->nd_repstat) {
3105 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
3106 				    &gid);
3107 				if (!nd->nd_repstat)
3108 					nvap->na_gid = gid;
3109 			}
3110 			if (j > NFSV4_SMALLSTR)
3111 				free(cp, M_NFSSTRING);
3112 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3113 			break;
3114 		case NFSATTRBIT_SYSTEM:
3115 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3116 			if (!nd->nd_repstat)
3117 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3118 			attrsum += NFSX_UNSIGNED;
3119 			break;
3120 		case NFSATTRBIT_TIMEACCESSSET:
3121 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3122 			attrsum += NFSX_UNSIGNED;
3123 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3124 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3125 			    fxdr_nfsv4time(tl, &nvap->na_atime);
3126 			    toclient = 1;
3127 			    attrsum += NFSX_V4TIME;
3128 			} else {
3129 			    vfs_timestamp(&nvap->na_atime);
3130 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3131 			}
3132 			break;
3133 		case NFSATTRBIT_TIMEBACKUP:
3134 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3135 			if (!nd->nd_repstat)
3136 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3137 			attrsum += NFSX_V4TIME;
3138 			break;
3139 		case NFSATTRBIT_TIMECREATE:
3140 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3141 			fxdr_nfsv4time(tl, &nvap->na_btime);
3142 			attrsum += NFSX_V4TIME;
3143 			break;
3144 		case NFSATTRBIT_TIMEMODIFYSET:
3145 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3146 			attrsum += NFSX_UNSIGNED;
3147 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3148 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3149 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
3150 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3151 			    attrsum += NFSX_V4TIME;
3152 			} else {
3153 			    vfs_timestamp(&nvap->na_mtime);
3154 			    if (!toclient)
3155 				nvap->na_vaflags |= VA_UTIMES_NULL;
3156 			}
3157 			break;
3158 		case NFSATTRBIT_MODESETMASKED:
3159 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3160 			mode = fxdr_unsigned(u_short, *tl++);
3161 			mask = fxdr_unsigned(u_short, *tl);
3162 			/*
3163 			 * vp == NULL implies an Open/Create operation.
3164 			 * This attribute can only be used for Setattr and
3165 			 * only for NFSv4.1 or higher.
3166 			 * If moderet != 0, a mode attribute has also been
3167 			 * specified and this attribute cannot be done in the
3168 			 * same Setattr operation.
3169 			 */
3170 			if ((nd->nd_flag & ND_NFSV41) == 0)
3171 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3172 			else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
3173 			    vp == NULL)
3174 				nd->nd_repstat = NFSERR_INVAL;
3175 			else if (moderet == 0)
3176 				moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
3177 			if (moderet == 0)
3178 				nvap->na_mode = (mode & mask) |
3179 				    (va.va_mode & ~mask);
3180 			else
3181 				nd->nd_repstat = moderet;
3182 			attrsum += 2 * NFSX_UNSIGNED;
3183 			break;
3184 		default:
3185 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3186 			/*
3187 			 * set bitpos so we drop out of the loop.
3188 			 */
3189 			bitpos = NFSATTRBIT_MAX;
3190 			break;
3191 		}
3192 	}
3193 
3194 	/*
3195 	 * some clients pad the attrlist, so we need to skip over the
3196 	 * padding.
3197 	 */
3198 	if (attrsum > attrsize) {
3199 		error = NFSERR_BADXDR;
3200 	} else {
3201 		attrsize = NFSM_RNDUP(attrsize);
3202 		if (attrsum < attrsize)
3203 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3204 	}
3205 nfsmout:
3206 	NFSEXITCODE2(error, nd);
3207 	return (error);
3208 }
3209 
3210 /*
3211  * Check/setup export credentials.
3212  */
3213 int
3214 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3215     struct ucred *credanon, bool testsec)
3216 {
3217 	int error;
3218 
3219 	/*
3220 	 * Check/setup credentials.
3221 	 */
3222 	if (nd->nd_flag & ND_GSS)
3223 		exp->nes_exflag &= ~MNT_EXPORTANON;
3224 
3225 	/*
3226 	 * Check to see if the operation is allowed for this security flavor.
3227 	 */
3228 	error = 0;
3229 	if (testsec) {
3230 		error = nfsvno_testexp(nd, exp);
3231 		if (error != 0)
3232 			goto out;
3233 	}
3234 
3235 	/*
3236 	 * Check to see if the file system is exported V4 only.
3237 	 */
3238 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3239 		error = NFSERR_PROGNOTV4;
3240 		goto out;
3241 	}
3242 
3243 	/*
3244 	 * Now, map the user credentials.
3245 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3246 	 *  Fsinfo RPC. If set for anything else, this code might need
3247 	 *  to change.)
3248 	 */
3249 	if (NFSVNO_EXPORTED(exp)) {
3250 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3251 		     NFSVNO_EXPORTANON(exp) ||
3252 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3253 			nd->nd_cred->cr_uid = credanon->cr_uid;
3254 			nd->nd_cred->cr_gid = credanon->cr_gid;
3255 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3256 			    credanon->cr_groups);
3257 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3258 			/*
3259 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3260 			 * if there is a replacement credential with a group
3261 			 * list set up by "nfsuserd -manage-gids".
3262 			 * If there is no replacement, nfsrv_getgrpscred()
3263 			 * simply returns its argument.
3264 			 */
3265 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3266 		}
3267 	}
3268 
3269 out:
3270 	NFSEXITCODE2(error, nd);
3271 	return (error);
3272 }
3273 
3274 /*
3275  * Check exports.
3276  */
3277 int
3278 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3279     struct ucred **credp)
3280 {
3281 	int error;
3282 
3283 	error = 0;
3284 	*credp = NULL;
3285 	MNT_ILOCK(mp);
3286 	if (mp->mnt_exjail == NULL ||
3287 	    mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison)
3288 		error = EACCES;
3289 	MNT_IUNLOCK(mp);
3290 	if (error == 0)
3291 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3292 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3293 	if (error) {
3294 		if (NFSD_VNET(nfs_rootfhset)) {
3295 			exp->nes_exflag = 0;
3296 			exp->nes_numsecflavor = 0;
3297 			error = 0;
3298 		}
3299 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3300 	    MAXSECFLAVORS) {
3301 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3302 		exp->nes_numsecflavor = 0;
3303 		error = EACCES;
3304 	}
3305 	NFSEXITCODE(error);
3306 	return (error);
3307 }
3308 
3309 /*
3310  * Get a vnode for a file handle and export stuff.
3311  */
3312 int
3313 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3314     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3315     struct ucred **credp)
3316 {
3317 	int error;
3318 
3319 	*credp = NULL;
3320 	exp->nes_numsecflavor = 0;
3321 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3322 	if (error != 0)
3323 		/* Make sure the server replies ESTALE to the client. */
3324 		error = ESTALE;
3325 	if (nam && !error) {
3326 		MNT_ILOCK(mp);
3327 		if (mp->mnt_exjail == NULL ||
3328 		    mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison)
3329 			error = EACCES;
3330 		MNT_IUNLOCK(mp);
3331 		if (error == 0)
3332 			error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3333 			    &exp->nes_numsecflavor, exp->nes_secflavors);
3334 		if (error) {
3335 			if (NFSD_VNET(nfs_rootfhset)) {
3336 				exp->nes_exflag = 0;
3337 				exp->nes_numsecflavor = 0;
3338 				error = 0;
3339 			} else {
3340 				vput(*vpp);
3341 			}
3342 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3343 		    MAXSECFLAVORS) {
3344 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3345 			exp->nes_numsecflavor = 0;
3346 			error = EACCES;
3347 			vput(*vpp);
3348 		}
3349 	}
3350 	NFSEXITCODE(error);
3351 	return (error);
3352 }
3353 
3354 /*
3355  * nfsd_fhtovp() - convert a fh to a vnode ptr
3356  * 	- look up fsid in mount list (if not found ret error)
3357  *	- get vp and export rights by calling nfsvno_fhtovp()
3358  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3359  *	  for AUTH_SYS
3360  *	- if mpp != NULL, return the mount point so that it can
3361  *	  be used for vn_finished_write() by the caller
3362  */
3363 void
3364 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3365     struct vnode **vpp, struct nfsexstuff *exp,
3366     struct mount **mpp, int startwrite, int nextop)
3367 {
3368 	struct mount *mp, *mpw;
3369 	struct ucred *credanon;
3370 	fhandle_t *fhp;
3371 	int error;
3372 
3373 	if (mpp != NULL)
3374 		*mpp = NULL;
3375 	*vpp = NULL;
3376 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3377 	mp = vfs_busyfs(&fhp->fh_fsid);
3378 	if (mp == NULL) {
3379 		nd->nd_repstat = ESTALE;
3380 		goto out;
3381 	}
3382 
3383 	if (startwrite) {
3384 		mpw = mp;
3385 		error = vn_start_write(NULL, &mpw, V_WAIT);
3386 		if (error != 0) {
3387 			mpw = NULL;
3388 			vfs_unbusy(mp);
3389 			nd->nd_repstat = ESTALE;
3390 			goto out;
3391 		}
3392 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3393 			lktype = LK_EXCLUSIVE;
3394 	} else
3395 		mpw = NULL;
3396 
3397 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3398 	    &credanon);
3399 	vfs_unbusy(mp);
3400 
3401 	/*
3402 	 * For NFSv4 without a pseudo root fs, unexported file handles
3403 	 * can be returned, so that Lookup works everywhere.
3404 	 */
3405 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3406 	    !(nd->nd_flag & ND_NFSV4)) {
3407 		vput(*vpp);
3408 		*vpp = NULL;
3409 		nd->nd_repstat = EACCES;
3410 	}
3411 
3412 	/*
3413 	 * Personally, I've never seen any point in requiring a
3414 	 * reserved port#, since only in the rare case where the
3415 	 * clients are all boxes with secure system privileges,
3416 	 * does it provide any enhanced security, but... some people
3417 	 * believe it to be useful and keep putting this code back in.
3418 	 * (There is also some "security checker" out there that
3419 	 *  complains if the nfs server doesn't enforce this.)
3420 	 * However, note the following:
3421 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3422 	 *	required.
3423 	 * RFC2623 recommends that, if a reserved port# is checked for,
3424 	 *	that there be a way to turn that off--> ifdef'd.
3425 	 */
3426 #ifdef NFS_REQRSVPORT
3427 	if (!nd->nd_repstat) {
3428 		struct sockaddr_in *saddr;
3429 		struct sockaddr_in6 *saddr6;
3430 
3431 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3432 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3433 		if (!(nd->nd_flag & ND_NFSV4) &&
3434 		    ((saddr->sin_family == AF_INET &&
3435 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3436 		     (saddr6->sin6_family == AF_INET6 &&
3437 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3438 			vput(*vpp);
3439 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3440 		}
3441 	}
3442 #endif	/* NFS_REQRSVPORT */
3443 
3444 	/*
3445 	 * Check/setup credentials.
3446 	 */
3447 	if (!nd->nd_repstat) {
3448 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3449 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3450 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3451 		if (nd->nd_repstat)
3452 			vput(*vpp);
3453 	}
3454 	if (credanon != NULL)
3455 		crfree(credanon);
3456 	if (nd->nd_repstat) {
3457 		vn_finished_write(mpw);
3458 		*vpp = NULL;
3459 	} else if (mpp != NULL) {
3460 		*mpp = mpw;
3461 	}
3462 
3463 out:
3464 	NFSEXITCODE2(0, nd);
3465 }
3466 
3467 /*
3468  * glue for fp.
3469  */
3470 static int
3471 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3472 {
3473 	struct filedesc *fdp;
3474 	struct file *fp;
3475 	int error = 0;
3476 
3477 	fdp = p->td_proc->p_fd;
3478 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3479 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3480 		error = EBADF;
3481 		goto out;
3482 	}
3483 	*fpp = fp;
3484 
3485 out:
3486 	NFSEXITCODE(error);
3487 	return (error);
3488 }
3489 
3490 /*
3491  * Called from nfssvc() to update the exports list. Just call
3492  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3493  * in the mount list.
3494  */
3495 int
3496 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3497 {
3498 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3499 	int error = 0;
3500 	struct nameidata nd;
3501 	fhandle_t fh;
3502 
3503 	error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export, false);
3504 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3505 		NFSD_VNET(nfs_rootfhset) = 0;
3506 	else if (error == 0) {
3507 		if (nfsexargp->fspec == NULL) {
3508 			error = EPERM;
3509 			goto out;
3510 		}
3511 		/*
3512 		 * If fspec != NULL, this is the v4root path.
3513 		 */
3514 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec);
3515 		if ((error = namei(&nd)) != 0)
3516 			goto out;
3517 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3518 		vrele(nd.ni_vp);
3519 		if (!error) {
3520 			NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH;
3521 			NFSBCOPY((caddr_t)&fh,
3522 			    NFSD_VNET(nfs_rootfh).nfsrvfh_data,
3523 			    sizeof (fhandle_t));
3524 			NFSD_VNET(nfs_rootfhset) = 1;
3525 		}
3526 	}
3527 
3528 out:
3529 	NFSEXITCODE(error);
3530 	return (error);
3531 }
3532 
3533 /*
3534  * This function needs to test to see if the system is near its limit
3535  * for memory allocation via malloc() or mget() and return True iff
3536  * either of these resources are near their limit.
3537  * XXX (For now, this is just a stub.)
3538  */
3539 int nfsrv_testmalloclimit = 0;
3540 int
3541 nfsrv_mallocmget_limit(void)
3542 {
3543 	static int printmesg = 0;
3544 	static int testval = 1;
3545 
3546 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3547 		if ((printmesg++ % 100) == 0)
3548 			printf("nfsd: malloc/mget near limit\n");
3549 		return (1);
3550 	}
3551 	return (0);
3552 }
3553 
3554 /*
3555  * BSD specific initialization of a mount point.
3556  */
3557 void
3558 nfsd_mntinit(void)
3559 {
3560 
3561 	NFSD_LOCK();
3562 	if (NFSD_VNET(nfsrv_mntinited)) {
3563 		NFSD_UNLOCK();
3564 		return;
3565 	}
3566 	NFSD_VNET(nfsrv_mntinited) = true;
3567 	nfsrvd_init(0);
3568 	NFSD_UNLOCK();
3569 
3570 	NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP,
3571 	    M_WAITOK | M_ZERO);
3572 	NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3573 	mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3574 	lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0);
3575 	TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist);
3576 	TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist);
3577 	NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL;
3578 	TAILQ_INIT(&NFSD_VNET(nfsv4root_opt));
3579 	TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt));
3580 	NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt);
3581 	NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt);
3582 	NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0;
3583 	NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0;
3584 	callout_init(&NFSD_VNET(nfsd_callout), 1);
3585 
3586 	nfsrvd_initcache();
3587 	nfsd_init();
3588 }
3589 
3590 static void
3591 nfsd_timer(void *arg)
3592 {
3593 	struct vnet *vnetp;
3594 
3595 	vnetp = (struct vnet *)arg;
3596 	NFSD_CURVNET_SET_QUIET(vnetp);
3597 	nfsrv_servertimer(vnetp);
3598 	callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer,
3599 	    arg, 0);
3600 	NFSD_CURVNET_RESTORE();
3601 }
3602 
3603 /*
3604  * Get a vnode for a file handle, without checking exports, etc.
3605  */
3606 struct vnode *
3607 nfsvno_getvp(fhandle_t *fhp)
3608 {
3609 	struct mount *mp;
3610 	struct vnode *vp;
3611 	int error;
3612 
3613 	mp = vfs_busyfs(&fhp->fh_fsid);
3614 	if (mp == NULL)
3615 		return (NULL);
3616 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3617 	vfs_unbusy(mp);
3618 	if (error)
3619 		return (NULL);
3620 	return (vp);
3621 }
3622 
3623 /*
3624  * Do a local VOP_ADVLOCK().
3625  */
3626 int
3627 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3628     u_int64_t end, struct thread *td)
3629 {
3630 	int error = 0;
3631 	struct flock fl;
3632 	u_int64_t tlen;
3633 
3634 	if (nfsrv_dolocallocks == 0)
3635 		goto out;
3636 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3637 
3638 	fl.l_whence = SEEK_SET;
3639 	fl.l_type = ftype;
3640 	fl.l_start = (off_t)first;
3641 	if (end == NFS64BITSSET) {
3642 		fl.l_len = 0;
3643 	} else {
3644 		tlen = end - first;
3645 		fl.l_len = (off_t)tlen;
3646 	}
3647 	/*
3648 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3649 	 * values for all calls, so that all locks will be held by the
3650 	 * nfsd server. (The nfsd server handles conflicts between the
3651 	 * various clients.)
3652 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3653 	 * bytes, so it can't be put in l_sysid.
3654 	 */
3655 	if (nfsv4_sysid == 0)
3656 		nfsv4_sysid = nlm_acquire_next_sysid();
3657 	fl.l_pid = (pid_t)0;
3658 	fl.l_sysid = (int)nfsv4_sysid;
3659 
3660 	if (ftype == F_UNLCK)
3661 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3662 		    (F_POSIX | F_REMOTE));
3663 	else
3664 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3665 		    (F_POSIX | F_REMOTE));
3666 
3667 out:
3668 	NFSEXITCODE(error);
3669 	return (error);
3670 }
3671 
3672 /*
3673  * Check the nfsv4 root exports.
3674  */
3675 int
3676 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3677 {
3678 	struct ucred *credanon;
3679 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3680 	uint64_t exflags;
3681 
3682 	error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags,
3683 	    &credanon, &numsecflavor, secflavors);
3684 	if (error) {
3685 		error = NFSERR_PROGUNAVAIL;
3686 		goto out;
3687 	}
3688 	if (credanon != NULL)
3689 		crfree(credanon);
3690 	for (i = 0; i < numsecflavor; i++) {
3691 		if (secflavors[i] == AUTH_SYS)
3692 			nd->nd_flag |= ND_EXAUTHSYS;
3693 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3694 			nd->nd_flag |= ND_EXGSS;
3695 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3696 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3697 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3698 			nd->nd_flag |= ND_EXGSSPRIVACY;
3699 	}
3700 
3701 	/* And set ND_EXxx flags for TLS. */
3702 	if ((exflags & MNT_EXTLS) != 0) {
3703 		nd->nd_flag |= ND_EXTLS;
3704 		if ((exflags & MNT_EXTLSCERT) != 0)
3705 			nd->nd_flag |= ND_EXTLSCERT;
3706 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3707 			nd->nd_flag |= ND_EXTLSCERTUSER;
3708 	}
3709 
3710 out:
3711 	NFSEXITCODE(error);
3712 	return (error);
3713 }
3714 
3715 /*
3716  * Nfs server pseudo system call for the nfsd's
3717  */
3718 /*
3719  * MPSAFE
3720  */
3721 static int
3722 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3723 {
3724 	struct file *fp;
3725 	struct nfsd_addsock_args sockarg;
3726 	struct nfsd_nfsd_args nfsdarg;
3727 	struct nfsd_nfsd_oargs onfsdarg;
3728 	struct nfsd_pnfsd_args pnfsdarg;
3729 	struct vnode *vp, *nvp, *curdvp;
3730 	struct pnfsdsfile *pf;
3731 	struct nfsdevice *ds, *fds;
3732 	cap_rights_t rights;
3733 	int buflen, error, ret;
3734 	char *buf, *cp, *cp2, *cp3;
3735 	char fname[PNFS_FILENAME_LEN + 1];
3736 
3737 	NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td));
3738 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3739 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3740 		if (error)
3741 			goto out;
3742 		/*
3743 		 * Since we don't know what rights might be required,
3744 		 * pretend that we need them all. It is better to be too
3745 		 * careful than too reckless.
3746 		 */
3747 		error = fget(td, sockarg.sock,
3748 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3749 		if (error != 0)
3750 			goto out;
3751 		if (fp->f_type != DTYPE_SOCKET) {
3752 			fdrop(fp, td);
3753 			error = EPERM;
3754 			goto out;
3755 		}
3756 		error = nfsrvd_addsock(fp);
3757 		fdrop(fp, td);
3758 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3759 		if (uap->argp == NULL) {
3760 			error = EINVAL;
3761 			goto out;
3762 		}
3763 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3764 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3765 			if (error == 0) {
3766 				nfsdarg.principal = onfsdarg.principal;
3767 				nfsdarg.minthreads = onfsdarg.minthreads;
3768 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3769 				nfsdarg.version = 1;
3770 				nfsdarg.addr = NULL;
3771 				nfsdarg.addrlen = 0;
3772 				nfsdarg.dnshost = NULL;
3773 				nfsdarg.dnshostlen = 0;
3774 				nfsdarg.dspath = NULL;
3775 				nfsdarg.dspathlen = 0;
3776 				nfsdarg.mdspath = NULL;
3777 				nfsdarg.mdspathlen = 0;
3778 				nfsdarg.mirrorcnt = 1;
3779 			}
3780 		} else
3781 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3782 		if (error)
3783 			goto out;
3784 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3785 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3786 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3787 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3788 		    nfsdarg.mirrorcnt >= 1 &&
3789 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3790 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3791 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3792 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3793 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3794 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3795 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3796 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3797 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3798 			if (error != 0) {
3799 				free(cp, M_TEMP);
3800 				goto out;
3801 			}
3802 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3803 			nfsdarg.addr = cp;
3804 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3805 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3806 			if (error != 0) {
3807 				free(nfsdarg.addr, M_TEMP);
3808 				free(cp, M_TEMP);
3809 				goto out;
3810 			}
3811 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3812 			nfsdarg.dnshost = cp;
3813 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3814 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3815 			if (error != 0) {
3816 				free(nfsdarg.addr, M_TEMP);
3817 				free(nfsdarg.dnshost, M_TEMP);
3818 				free(cp, M_TEMP);
3819 				goto out;
3820 			}
3821 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3822 			nfsdarg.dspath = cp;
3823 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3824 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3825 			if (error != 0) {
3826 				free(nfsdarg.addr, M_TEMP);
3827 				free(nfsdarg.dnshost, M_TEMP);
3828 				free(nfsdarg.dspath, M_TEMP);
3829 				free(cp, M_TEMP);
3830 				goto out;
3831 			}
3832 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3833 			nfsdarg.mdspath = cp;
3834 		} else {
3835 			nfsdarg.addr = NULL;
3836 			nfsdarg.addrlen = 0;
3837 			nfsdarg.dnshost = NULL;
3838 			nfsdarg.dnshostlen = 0;
3839 			nfsdarg.dspath = NULL;
3840 			nfsdarg.dspathlen = 0;
3841 			nfsdarg.mdspath = NULL;
3842 			nfsdarg.mdspathlen = 0;
3843 			nfsdarg.mirrorcnt = 1;
3844 		}
3845 		nfsd_timer(NFSD_TD_TO_VNET(td));
3846 		error = nfsrvd_nfsd(td, &nfsdarg);
3847 		callout_drain(&NFSD_VNET(nfsd_callout));
3848 		free(nfsdarg.addr, M_TEMP);
3849 		free(nfsdarg.dnshost, M_TEMP);
3850 		free(nfsdarg.dspath, M_TEMP);
3851 		free(nfsdarg.mdspath, M_TEMP);
3852 	} else if (uap->flag & NFSSVC_PNFSDS) {
3853 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3854 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3855 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3856 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3857 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3858 			    NULL);
3859 			if (error == 0)
3860 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3861 			free(cp, M_TEMP);
3862 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3863 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3864 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3865 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3866 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3867 			    NULL);
3868 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3869 			if (error == 0 && pnfsdarg.dspath != NULL) {
3870 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3871 				error = copyinstr(pnfsdarg.dspath, cp2,
3872 				    PATH_MAX + 1, NULL);
3873 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3874 				    error);
3875 			} else
3876 				cp2 = NULL;
3877 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3878 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3879 				error = copyinstr(pnfsdarg.curdspath, cp3,
3880 				    PATH_MAX + 1, NULL);
3881 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3882 				    error);
3883 			} else
3884 				cp3 = NULL;
3885 			curdvp = NULL;
3886 			fds = NULL;
3887 			if (error == 0)
3888 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3889 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3890 				    &fds);
3891 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3892 			if (error == 0) {
3893 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3894 					printf("copymr: dsdir out of range\n");
3895 					pf->dsf_dir = 0;
3896 				}
3897 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3898 				error = nfsrv_copymr(vp, nvp,
3899 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3900 				    (struct pnfsdsfile *)buf,
3901 				    buflen / sizeof(*pf), td->td_ucred, td);
3902 				vput(vp);
3903 				vput(nvp);
3904 				if (fds != NULL && error == 0) {
3905 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3906 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
3907 					if (ret == 0) {
3908 						nfsrv_dsremove(curdvp, fname,
3909 						    td->td_ucred, td);
3910 						NFSVOPUNLOCK(curdvp);
3911 					}
3912 				}
3913 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3914 			}
3915 			free(cp, M_TEMP);
3916 			free(cp2, M_TEMP);
3917 			free(cp3, M_TEMP);
3918 			free(buf, M_TEMP);
3919 		}
3920 	} else {
3921 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3922 	}
3923 
3924 out:
3925 	NFSD_CURVNET_RESTORE();
3926 	NFSEXITCODE(error);
3927 	return (error);
3928 }
3929 
3930 static int
3931 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3932 {
3933 	struct nfsex_args export;
3934 	struct nfsex_oldargs oexp;
3935 	struct file *fp = NULL;
3936 	int stablefd, i, len;
3937 	struct nfsd_clid adminrevoke;
3938 	struct nfsd_dumplist dumplist;
3939 	struct nfsd_dumpclients *dumpclients;
3940 	struct nfsd_dumplocklist dumplocklist;
3941 	struct nfsd_dumplocks *dumplocks;
3942 	struct nameidata nd;
3943 	vnode_t vp;
3944 	int error = EINVAL, igotlock;
3945 	struct proc *procp;
3946 	gid_t *grps;
3947 
3948 	if (uap->flag & NFSSVC_PUBLICFH) {
3949 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3950 		    sizeof (fhandle_t));
3951 		error = copyin(uap->argp,
3952 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3953 		if (!error)
3954 			nfs_pubfhset = 1;
3955 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3956 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3957 		error = copyin(uap->argp,(caddr_t)&export,
3958 		    sizeof (struct nfsex_args));
3959 		if (!error) {
3960 			grps = NULL;
3961 			if (export.export.ex_ngroups > NGROUPS_MAX ||
3962 			    export.export.ex_ngroups < 0)
3963 				error = EINVAL;
3964 			else if (export.export.ex_ngroups > 0) {
3965 				grps = malloc(export.export.ex_ngroups *
3966 				    sizeof(gid_t), M_TEMP, M_WAITOK);
3967 				error = copyin(export.export.ex_groups, grps,
3968 				    export.export.ex_ngroups * sizeof(gid_t));
3969 				export.export.ex_groups = grps;
3970 			} else
3971 				export.export.ex_groups = NULL;
3972 			if (!error)
3973 				error = nfsrv_v4rootexport(&export, cred, p);
3974 			free(grps, M_TEMP);
3975 		}
3976 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3977 	    NFSSVC_V4ROOTEXPORT) {
3978 		error = copyin(uap->argp,(caddr_t)&oexp,
3979 		    sizeof (struct nfsex_oldargs));
3980 		if (!error) {
3981 			memset(&export.export, 0, sizeof(export.export));
3982 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3983 			export.export.ex_root = oexp.export.ex_root;
3984 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3985 			export.export.ex_ngroups =
3986 			    oexp.export.ex_anon.cr_ngroups;
3987 			export.export.ex_groups = NULL;
3988 			if (export.export.ex_ngroups > XU_NGROUPS ||
3989 			    export.export.ex_ngroups < 0)
3990 				error = EINVAL;
3991 			else if (export.export.ex_ngroups > 0) {
3992 				export.export.ex_groups = malloc(
3993 				    export.export.ex_ngroups * sizeof(gid_t),
3994 				    M_TEMP, M_WAITOK);
3995 				for (i = 0; i < export.export.ex_ngroups; i++)
3996 					export.export.ex_groups[i] =
3997 					    oexp.export.ex_anon.cr_groups[i];
3998 			}
3999 			export.export.ex_addr = oexp.export.ex_addr;
4000 			export.export.ex_addrlen = oexp.export.ex_addrlen;
4001 			export.export.ex_mask = oexp.export.ex_mask;
4002 			export.export.ex_masklen = oexp.export.ex_masklen;
4003 			export.export.ex_indexfile = oexp.export.ex_indexfile;
4004 			export.export.ex_numsecflavors =
4005 			    oexp.export.ex_numsecflavors;
4006 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
4007 			    export.export.ex_numsecflavors < 0)
4008 				error = EINVAL;
4009 			else {
4010 				for (i = 0; i < export.export.ex_numsecflavors;
4011 				    i++)
4012 					export.export.ex_secflavors[i] =
4013 					    oexp.export.ex_secflavors[i];
4014 			}
4015 			export.fspec = oexp.fspec;
4016 			if (error == 0)
4017 				error = nfsrv_v4rootexport(&export, cred, p);
4018 			free(export.export.ex_groups, M_TEMP);
4019 		}
4020 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
4021 		nfs_pubfhset = 0;
4022 		error = 0;
4023 	} else if (uap->flag & NFSSVC_STABLERESTART) {
4024 		error = copyin(uap->argp, (caddr_t)&stablefd,
4025 		    sizeof (int));
4026 		if (!error)
4027 			error = fp_getfvp(p, stablefd, &fp, &vp);
4028 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
4029 			error = EBADF;
4030 		if (!error && NFSD_VNET(nfsrv_numnfsd) != 0)
4031 			error = ENXIO;
4032 		if (!error) {
4033 			NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp;
4034 			nfsrv_setupstable(p);
4035 		}
4036 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
4037 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
4038 		    sizeof (struct nfsd_clid));
4039 		if (!error)
4040 			error = nfsrv_adminrevoke(&adminrevoke, p);
4041 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
4042 		error = copyin(uap->argp, (caddr_t)&dumplist,
4043 		    sizeof (struct nfsd_dumplist));
4044 		if (!error && (dumplist.ndl_size < 1 ||
4045 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
4046 			error = EPERM;
4047 		if (!error) {
4048 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
4049 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4050 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
4051 		    error = copyout(dumpclients, dumplist.ndl_list, len);
4052 		    free(dumpclients, M_TEMP);
4053 		}
4054 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
4055 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
4056 		    sizeof (struct nfsd_dumplocklist));
4057 		if (!error && (dumplocklist.ndllck_size < 1 ||
4058 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
4059 			error = EPERM;
4060 		if (!error)
4061 			error = nfsrv_lookupfilename(&nd,
4062 				dumplocklist.ndllck_fname, p);
4063 		if (!error) {
4064 			len = sizeof (struct nfsd_dumplocks) *
4065 				dumplocklist.ndllck_size;
4066 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4067 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
4068 			    dumplocklist.ndllck_size, p);
4069 			vput(nd.ni_vp);
4070 			error = copyout(dumplocks, dumplocklist.ndllck_list,
4071 			    len);
4072 			free(dumplocks, M_TEMP);
4073 		}
4074 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
4075 		procp = p->td_proc;
4076 		PROC_LOCK(procp);
4077 		nfsd_master_pid = procp->p_pid;
4078 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
4079 		nfsd_master_start = procp->p_stats->p_start;
4080 		NFSD_VNET(nfsd_master_proc) = procp;
4081 		PROC_UNLOCK(procp);
4082 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
4083 		NFSLOCKV4ROOTMUTEX();
4084 		if (!NFSD_VNET(nfsrv_suspend_nfsd)) {
4085 			/* Lock out all nfsd threads */
4086 			do {
4087 				igotlock = nfsv4_lock(
4088 				    &NFSD_VNET(nfsd_suspend_lock), 1, NULL,
4089 				    NFSV4ROOTLOCKMUTEXPTR, NULL);
4090 			} while (igotlock == 0 &&
4091 			    !NFSD_VNET(nfsrv_suspend_nfsd));
4092 			NFSD_VNET(nfsrv_suspend_nfsd) = true;
4093 		}
4094 		NFSUNLOCKV4ROOTMUTEX();
4095 		error = 0;
4096 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
4097 		NFSLOCKV4ROOTMUTEX();
4098 		if (NFSD_VNET(nfsrv_suspend_nfsd)) {
4099 			nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0);
4100 			NFSD_VNET(nfsrv_suspend_nfsd) = false;
4101 		}
4102 		NFSUNLOCKV4ROOTMUTEX();
4103 		error = 0;
4104 	}
4105 
4106 	NFSEXITCODE(error);
4107 	return (error);
4108 }
4109 
4110 /*
4111  * Check exports.
4112  * Returns 0 if ok, 1 otherwise.
4113  */
4114 int
4115 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
4116 {
4117 	int i;
4118 
4119 	if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) ||
4120 	    (NFSVNO_EXTLSCERT(exp) &&
4121 	     (nd->nd_flag & ND_TLSCERT) == 0) ||
4122 	    (NFSVNO_EXTLSCERTUSER(exp) &&
4123 	     (nd->nd_flag & ND_TLSCERTUSER) == 0)) {
4124 		if ((nd->nd_flag & ND_NFSV4) != 0)
4125 			return (NFSERR_WRONGSEC);
4126 #ifdef notnow
4127 		/* There is currently no auth_stat for this. */
4128 		else if ((nd->nd_flag & ND_TLS) == 0)
4129 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
4130 		else
4131 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
4132 #endif
4133 		else
4134 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4135 	}
4136 
4137 	/*
4138 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use
4139 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
4140 	 */
4141 	if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO)
4142 		return (0);
4143 
4144 	/*
4145 	 * This seems odd, but allow the case where the security flavor
4146 	 * list is empty. This happens when NFSv4 is traversing non-exported
4147 	 * file systems. Exported file systems should always have a non-empty
4148 	 * security flavor list.
4149 	 */
4150 	if (exp->nes_numsecflavor == 0)
4151 		return (0);
4152 
4153 	for (i = 0; i < exp->nes_numsecflavor; i++) {
4154 		/*
4155 		 * The tests for privacy and integrity must be first,
4156 		 * since ND_GSS is set for everything but AUTH_SYS.
4157 		 */
4158 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
4159 		    (nd->nd_flag & ND_GSSPRIVACY))
4160 			return (0);
4161 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
4162 		    (nd->nd_flag & ND_GSSINTEGRITY))
4163 			return (0);
4164 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
4165 		    (nd->nd_flag & ND_GSS))
4166 			return (0);
4167 		if (exp->nes_secflavors[i] == AUTH_SYS &&
4168 		    (nd->nd_flag & ND_GSS) == 0)
4169 			return (0);
4170 	}
4171 	if ((nd->nd_flag & ND_NFSV4) != 0)
4172 		return (NFSERR_WRONGSEC);
4173 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4174 }
4175 
4176 /*
4177  * Calculate a hash value for the fid in a file handle.
4178  */
4179 uint32_t
4180 nfsrv_hashfh(fhandle_t *fhp)
4181 {
4182 	uint32_t hashval;
4183 
4184 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
4185 	return (hashval);
4186 }
4187 
4188 /*
4189  * Calculate a hash value for the sessionid.
4190  */
4191 uint32_t
4192 nfsrv_hashsessionid(uint8_t *sessionid)
4193 {
4194 	uint32_t hashval;
4195 
4196 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
4197 	return (hashval);
4198 }
4199 
4200 /*
4201  * Signal the userland master nfsd to backup the stable restart file.
4202  */
4203 void
4204 nfsrv_backupstable(void)
4205 {
4206 	struct proc *procp;
4207 
4208 	if (NFSD_VNET(nfsd_master_proc) != NULL) {
4209 		procp = pfind(nfsd_master_pid);
4210 		/* Try to make sure it is the correct process. */
4211 		if (procp == NFSD_VNET(nfsd_master_proc) &&
4212 		    procp->p_stats->p_start.tv_sec ==
4213 		    nfsd_master_start.tv_sec &&
4214 		    procp->p_stats->p_start.tv_usec ==
4215 		    nfsd_master_start.tv_usec &&
4216 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
4217 			kern_psignal(procp, SIGUSR2);
4218 		else
4219 			NFSD_VNET(nfsd_master_proc) = NULL;
4220 
4221 		if (procp != NULL)
4222 			PROC_UNLOCK(procp);
4223 	}
4224 }
4225 
4226 /*
4227  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
4228  * The arguments are in a structure, so that they can be passed through
4229  * taskqueue for a kernel process to execute this function.
4230  */
4231 struct nfsrvdscreate {
4232 	int			done;
4233 	int			inprog;
4234 	struct task		tsk;
4235 	struct ucred		*tcred;
4236 	struct vnode		*dvp;
4237 	NFSPROC_T		*p;
4238 	struct pnfsdsfile	*pf;
4239 	int			err;
4240 	fhandle_t		fh;
4241 	struct vattr		va;
4242 	struct vattr		createva;
4243 };
4244 
4245 int
4246 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
4247     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
4248     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
4249 {
4250 	struct vnode *nvp;
4251 	struct nameidata named;
4252 	struct vattr va;
4253 	char *bufp;
4254 	u_long *hashp;
4255 	struct nfsnode *np;
4256 	struct nfsmount *nmp;
4257 	int error;
4258 
4259 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
4260 	    LOCKPARENT | LOCKLEAF | NOCACHE);
4261 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4262 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
4263 	named.ni_cnd.cn_nameptr = bufp;
4264 	if (fnamep != NULL) {
4265 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
4266 		named.ni_cnd.cn_namelen = strlen(bufp);
4267 	} else
4268 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
4269 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
4270 
4271 	/* Create the date file in the DS mount. */
4272 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4273 	if (error == 0) {
4274 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
4275 		vref(dvp);
4276 		VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false);
4277 		if (error == 0) {
4278 			/* Set the ownership of the file. */
4279 			error = VOP_SETATTR(nvp, nvap, tcred);
4280 			NFSD_DEBUG(4, "nfsrv_dscreate:"
4281 			    " setattr-uid=%d\n", error);
4282 			if (error != 0)
4283 				vput(nvp);
4284 		}
4285 		if (error != 0)
4286 			printf("pNFS: pnfscreate failed=%d\n", error);
4287 	} else
4288 		printf("pNFS: pnfscreate vnlock=%d\n", error);
4289 	if (error == 0) {
4290 		np = VTONFS(nvp);
4291 		nmp = VFSTONFS(nvp->v_mount);
4292 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
4293 		    != 0 || nmp->nm_nam->sa_len > sizeof(
4294 		    struct sockaddr_in6) ||
4295 		    np->n_fhp->nfh_len != NFSX_MYFH) {
4296 			printf("Bad DS file: fstype=%s salen=%d"
4297 			    " fhlen=%d\n",
4298 			    nvp->v_mount->mnt_vfc->vfc_name,
4299 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
4300 			error = ENOENT;
4301 		}
4302 
4303 		/* Set extattrs for the DS on the MDS file. */
4304 		if (error == 0) {
4305 			if (dsa != NULL) {
4306 				error = VOP_GETATTR(nvp, &va, tcred);
4307 				if (error == 0) {
4308 					dsa->dsa_filerev = va.va_filerev;
4309 					dsa->dsa_size = va.va_size;
4310 					dsa->dsa_atime = va.va_atime;
4311 					dsa->dsa_mtime = va.va_mtime;
4312 					dsa->dsa_bytes = va.va_bytes;
4313 				}
4314 			}
4315 			if (error == 0) {
4316 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4317 				    NFSX_MYFH);
4318 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4319 				    nmp->nm_nam->sa_len);
4320 				NFSBCOPY(named.ni_cnd.cn_nameptr,
4321 				    pf->dsf_filename,
4322 				    sizeof(pf->dsf_filename));
4323 			}
4324 		} else
4325 			printf("pNFS: pnfscreate can't get DS"
4326 			    " attr=%d\n", error);
4327 		if (nvpp != NULL && error == 0)
4328 			*nvpp = nvp;
4329 		else
4330 			vput(nvp);
4331 	}
4332 	nfsvno_relpathbuf(&named);
4333 	return (error);
4334 }
4335 
4336 /*
4337  * Start up the thread that will execute nfsrv_dscreate().
4338  */
4339 static void
4340 start_dscreate(void *arg, int pending)
4341 {
4342 	struct nfsrvdscreate *dsc;
4343 
4344 	dsc = (struct nfsrvdscreate *)arg;
4345 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4346 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4347 	dsc->done = 1;
4348 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4349 }
4350 
4351 /*
4352  * Create a pNFS data file on the Data Server(s).
4353  */
4354 static void
4355 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4356     NFSPROC_T *p)
4357 {
4358 	struct nfsrvdscreate *dsc, *tdsc = NULL;
4359 	struct nfsdevice *ds, *tds, *fds;
4360 	struct mount *mp;
4361 	struct pnfsdsfile *pf, *tpf;
4362 	struct pnfsdsattr dsattr;
4363 	struct vattr va;
4364 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4365 	struct nfsmount *nmp;
4366 	fhandle_t fh;
4367 	uid_t vauid;
4368 	gid_t vagid;
4369 	u_short vamode;
4370 	struct ucred *tcred;
4371 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4372 	int failpos, timo;
4373 
4374 	/* Get a DS server directory in a round-robin order. */
4375 	mirrorcnt = 1;
4376 	mp = vp->v_mount;
4377 	ds = fds = NULL;
4378 	NFSDDSLOCK();
4379 	/*
4380 	 * Search for the first entry that handles this MDS fs, but use the
4381 	 * first entry for all MDS fs's otherwise.
4382 	 */
4383 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4384 		if (tds->nfsdev_nmp != NULL) {
4385 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4386 				ds = tds;
4387 			else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4388 			    &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4389 				ds = fds = tds;
4390 				break;
4391 			}
4392 		}
4393 	}
4394 	if (ds == NULL) {
4395 		NFSDDSUNLOCK();
4396 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4397 		return;
4398 	}
4399 	i = dsdir[0] = ds->nfsdev_nextdir;
4400 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4401 	dvp[0] = ds->nfsdev_dsdir[i];
4402 	tds = TAILQ_NEXT(ds, nfsdev_list);
4403 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4404 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4405 			if (tds->nfsdev_nmp != NULL &&
4406 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4407 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4408 			      fsidcmp(&mp->mnt_stat.f_fsid,
4409 			      &tds->nfsdev_mdsfsid) == 0))) {
4410 				dsdir[mirrorcnt] = i;
4411 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4412 				mirrorcnt++;
4413 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
4414 					break;
4415 			}
4416 		}
4417 	}
4418 	/* Put at end of list to implement round-robin usage. */
4419 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4420 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4421 	NFSDDSUNLOCK();
4422 	dsc = NULL;
4423 	if (mirrorcnt > 1)
4424 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4425 		    M_WAITOK | M_ZERO);
4426 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4427 	    M_ZERO);
4428 
4429 	error = nfsvno_getfh(vp, &fh, p);
4430 	if (error == 0)
4431 		error = VOP_GETATTR(vp, &va, cred);
4432 	if (error == 0) {
4433 		/* Set the attributes for "vp" to Setattr the DS vp. */
4434 		vauid = va.va_uid;
4435 		vagid = va.va_gid;
4436 		vamode = va.va_mode;
4437 		VATTR_NULL(&va);
4438 		va.va_uid = vauid;
4439 		va.va_gid = vagid;
4440 		va.va_mode = vamode;
4441 		va.va_size = 0;
4442 	} else
4443 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4444 
4445 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4446 	    cred->cr_gid);
4447 	/* Make data file name based on FH. */
4448 	tcred = newnfs_getcred();
4449 
4450 	/*
4451 	 * Create the file on each DS mirror, using kernel process(es) for the
4452 	 * additional mirrors.
4453 	 */
4454 	failpos = -1;
4455 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4456 		tpf->dsf_dir = dsdir[i];
4457 		tdsc->tcred = tcred;
4458 		tdsc->p = p;
4459 		tdsc->pf = tpf;
4460 		tdsc->createva = *vap;
4461 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4462 		tdsc->va = va;
4463 		tdsc->dvp = dvp[i];
4464 		tdsc->done = 0;
4465 		tdsc->inprog = 0;
4466 		tdsc->err = 0;
4467 		ret = EIO;
4468 		if (nfs_pnfsiothreads != 0) {
4469 			ret = nfs_pnfsio(start_dscreate, tdsc);
4470 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4471 		}
4472 		if (ret != 0) {
4473 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4474 			    NULL, tcred, p, NULL);
4475 			if (ret != 0) {
4476 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4477 				    error));
4478 				if (failpos == -1 && nfsds_failerr(ret))
4479 					failpos = i;
4480 				else
4481 					error = ret;
4482 			}
4483 		}
4484 	}
4485 	if (error == 0) {
4486 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4487 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4488 		    &dsattr, NULL, tcred, p, NULL);
4489 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4490 			failpos = mirrorcnt - 1;
4491 			error = 0;
4492 		}
4493 	}
4494 	timo = hz / 50;		/* Wait for 20msec. */
4495 	if (timo < 1)
4496 		timo = 1;
4497 	/* Wait for kernel task(s) to complete. */
4498 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4499 		while (tdsc->inprog != 0 && tdsc->done == 0)
4500 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4501 		if (tdsc->err != 0) {
4502 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4503 				failpos = i;
4504 			else if (error == 0)
4505 				error = tdsc->err;
4506 		}
4507 	}
4508 
4509 	/*
4510 	 * If failpos has been set, that mirror has failed, so it needs
4511 	 * to be disabled.
4512 	 */
4513 	if (failpos >= 0) {
4514 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4515 		NFSLOCKMNT(nmp);
4516 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4517 		     NFSMNTP_CANCELRPCS)) == 0) {
4518 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4519 			NFSUNLOCKMNT(nmp);
4520 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4521 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4522 			    ds);
4523 			if (ds != NULL)
4524 				nfsrv_killrpcs(nmp);
4525 			NFSLOCKMNT(nmp);
4526 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4527 			wakeup(nmp);
4528 		}
4529 		NFSUNLOCKMNT(nmp);
4530 	}
4531 
4532 	NFSFREECRED(tcred);
4533 	if (error == 0) {
4534 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4535 
4536 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4537 		    mirrorcnt, nfsrv_maxpnfsmirror);
4538 		/*
4539 		 * For all mirrors that couldn't be created, fill in the
4540 		 * *pf structure, but with an IP address == 0.0.0.0.
4541 		 */
4542 		tpf = pf + mirrorcnt;
4543 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4544 			*tpf = *pf;
4545 			tpf->dsf_sin.sin_family = AF_INET;
4546 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4547 			tpf->dsf_sin.sin_addr.s_addr = 0;
4548 			tpf->dsf_sin.sin_port = 0;
4549 		}
4550 
4551 		error = vn_extattr_set(vp, IO_NODELOCKED,
4552 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4553 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4554 		if (error == 0)
4555 			error = vn_extattr_set(vp, IO_NODELOCKED,
4556 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4557 			    sizeof(dsattr), (char *)&dsattr, p);
4558 		if (error != 0)
4559 			printf("pNFS: pnfscreate setextattr=%d\n",
4560 			    error);
4561 	} else
4562 		printf("pNFS: pnfscreate=%d\n", error);
4563 	free(pf, M_TEMP);
4564 	free(dsc, M_TEMP);
4565 }
4566 
4567 /*
4568  * Get the information needed to remove the pNFS Data Server file from the
4569  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4570  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4571  */
4572 static void
4573 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4574     int *mirrorcntp, char *fname, fhandle_t *fhp)
4575 {
4576 	struct vattr va;
4577 	struct ucred *tcred;
4578 	char *buf;
4579 	int buflen, error;
4580 
4581 	dvpp[0] = NULL;
4582 	/* If not an exported regular file or not a pNFS server, just return. */
4583 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4584 	    nfsrv_devidcnt == 0)
4585 		return;
4586 
4587 	/* Check to see if this is the last hard link. */
4588 	tcred = newnfs_getcred();
4589 	error = VOP_GETATTR(vp, &va, tcred);
4590 	NFSFREECRED(tcred);
4591 	if (error != 0) {
4592 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4593 		return;
4594 	}
4595 	if (va.va_nlink > 1)
4596 		return;
4597 
4598 	error = nfsvno_getfh(vp, fhp, p);
4599 	if (error != 0) {
4600 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4601 		return;
4602 	}
4603 
4604 	buflen = 1024;
4605 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4606 	/* Get the directory vnode for the DS mount and the file handle. */
4607 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4608 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4609 	free(buf, M_TEMP);
4610 	if (error != 0)
4611 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4612 }
4613 
4614 /*
4615  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4616  * The arguments are in a structure, so that they can be passed through
4617  * taskqueue for a kernel process to execute this function.
4618  */
4619 struct nfsrvdsremove {
4620 	int			done;
4621 	int			inprog;
4622 	struct task		tsk;
4623 	struct ucred		*tcred;
4624 	struct vnode		*dvp;
4625 	NFSPROC_T		*p;
4626 	int			err;
4627 	char			fname[PNFS_FILENAME_LEN + 1];
4628 };
4629 
4630 static int
4631 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4632     NFSPROC_T *p)
4633 {
4634 	struct nameidata named;
4635 	struct vnode *nvp;
4636 	char *bufp;
4637 	u_long *hashp;
4638 	int error;
4639 
4640 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4641 	if (error != 0)
4642 		return (error);
4643 	named.ni_cnd.cn_nameiop = DELETE;
4644 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4645 	named.ni_cnd.cn_cred = tcred;
4646 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF;
4647 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4648 	named.ni_cnd.cn_nameptr = bufp;
4649 	named.ni_cnd.cn_namelen = strlen(fname);
4650 	strlcpy(bufp, fname, NAME_MAX);
4651 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4652 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4653 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4654 	if (error == 0) {
4655 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4656 		vput(nvp);
4657 	}
4658 	NFSVOPUNLOCK(dvp);
4659 	nfsvno_relpathbuf(&named);
4660 	if (error != 0)
4661 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4662 	return (error);
4663 }
4664 
4665 /*
4666  * Start up the thread that will execute nfsrv_dsremove().
4667  */
4668 static void
4669 start_dsremove(void *arg, int pending)
4670 {
4671 	struct nfsrvdsremove *dsrm;
4672 
4673 	dsrm = (struct nfsrvdsremove *)arg;
4674 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4675 	    dsrm->p);
4676 	dsrm->done = 1;
4677 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4678 }
4679 
4680 /*
4681  * Remove a pNFS data file from a Data Server.
4682  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4683  * removed to set up the dvp and fill in the FH.
4684  */
4685 static void
4686 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4687     NFSPROC_T *p)
4688 {
4689 	struct ucred *tcred;
4690 	struct nfsrvdsremove *dsrm, *tdsrm;
4691 	struct nfsdevice *ds;
4692 	struct nfsmount *nmp;
4693 	int failpos, i, ret, timo;
4694 
4695 	tcred = newnfs_getcred();
4696 	dsrm = NULL;
4697 	if (mirrorcnt > 1)
4698 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4699 	/*
4700 	 * Remove the file on each DS mirror, using kernel process(es) for the
4701 	 * additional mirrors.
4702 	 */
4703 	failpos = -1;
4704 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4705 		tdsrm->tcred = tcred;
4706 		tdsrm->p = p;
4707 		tdsrm->dvp = dvp[i];
4708 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4709 		tdsrm->inprog = 0;
4710 		tdsrm->done = 0;
4711 		tdsrm->err = 0;
4712 		ret = EIO;
4713 		if (nfs_pnfsiothreads != 0) {
4714 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4715 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4716 		}
4717 		if (ret != 0) {
4718 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4719 			if (failpos == -1 && nfsds_failerr(ret))
4720 				failpos = i;
4721 		}
4722 	}
4723 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4724 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4725 		failpos = mirrorcnt - 1;
4726 	timo = hz / 50;		/* Wait for 20msec. */
4727 	if (timo < 1)
4728 		timo = 1;
4729 	/* Wait for kernel task(s) to complete. */
4730 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4731 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4732 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4733 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4734 			failpos = i;
4735 	}
4736 
4737 	/*
4738 	 * If failpos has been set, that mirror has failed, so it needs
4739 	 * to be disabled.
4740 	 */
4741 	if (failpos >= 0) {
4742 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4743 		NFSLOCKMNT(nmp);
4744 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4745 		     NFSMNTP_CANCELRPCS)) == 0) {
4746 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4747 			NFSUNLOCKMNT(nmp);
4748 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4749 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4750 			    ds);
4751 			if (ds != NULL)
4752 				nfsrv_killrpcs(nmp);
4753 			NFSLOCKMNT(nmp);
4754 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4755 			wakeup(nmp);
4756 		}
4757 		NFSUNLOCKMNT(nmp);
4758 	}
4759 
4760 	/* Get rid all layouts for the file. */
4761 	nfsrv_freefilelayouts(fhp);
4762 
4763 	NFSFREECRED(tcred);
4764 	free(dsrm, M_TEMP);
4765 }
4766 
4767 /*
4768  * Generate a file name based on the file handle and put it in *bufp.
4769  * Return the number of bytes generated.
4770  */
4771 static int
4772 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4773 {
4774 	int i;
4775 	uint8_t *cp;
4776 	const uint8_t *hexdigits = "0123456789abcdef";
4777 
4778 	cp = (uint8_t *)fhp;
4779 	for (i = 0; i < sizeof(*fhp); i++) {
4780 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4781 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4782 	}
4783 	bufp[2 * i] = '\0';
4784 	return (2 * i);
4785 }
4786 
4787 /*
4788  * Update the Metadata file's attributes from the DS file when a Read/Write
4789  * layout is returned.
4790  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4791  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4792  */
4793 int
4794 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4795 {
4796 	struct ucred *tcred;
4797 	int error;
4798 
4799 	/* Do this as root so that it won't fail with EACCES. */
4800 	tcred = newnfs_getcred();
4801 	error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4802 	    NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4803 	NFSFREECRED(tcred);
4804 	return (error);
4805 }
4806 
4807 /*
4808  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4809  */
4810 static int
4811 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4812     NFSPROC_T *p)
4813 {
4814 	int error;
4815 
4816 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4817 	    NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4818 	return (error);
4819 }
4820 
4821 static int
4822 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4823     struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4824     struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4825     off_t *offp, int content, bool *eofp)
4826 {
4827 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4828 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4829 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4830 	struct nfsdevice *ds;
4831 	struct pnfsdsattr dsattr;
4832 	struct opnfsdsattr odsattr;
4833 	char *buf;
4834 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4835 
4836 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4837 	/*
4838 	 * If not a regular file, not exported or not a pNFS server,
4839 	 * just return ENOENT.
4840 	 */
4841 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4842 	    nfsrv_devidcnt == 0)
4843 		return (ENOENT);
4844 
4845 	buflen = 1024;
4846 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4847 	error = 0;
4848 
4849 	/*
4850 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4851 	 * from the MetaData file's extended attribute.
4852 	 */
4853 	if (ioproc == NFSPROC_GETATTR) {
4854 		error = vn_extattr_get(vp, IO_NODELOCKED,
4855 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4856 		    p);
4857 		if (error == 0) {
4858 			if (buflen == sizeof(odsattr)) {
4859 				NFSBCOPY(buf, &odsattr, buflen);
4860 				nap->na_filerev = odsattr.dsa_filerev;
4861 				nap->na_size = odsattr.dsa_size;
4862 				nap->na_atime = odsattr.dsa_atime;
4863 				nap->na_mtime = odsattr.dsa_mtime;
4864 				/*
4865 				 * Fake na_bytes by rounding up na_size.
4866 				 * Since we don't know the block size, just
4867 				 * use BLKDEV_IOSIZE.
4868 				 */
4869 				nap->na_bytes = (odsattr.dsa_size +
4870 				    BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4871 			} else if (buflen == sizeof(dsattr)) {
4872 				NFSBCOPY(buf, &dsattr, buflen);
4873 				nap->na_filerev = dsattr.dsa_filerev;
4874 				nap->na_size = dsattr.dsa_size;
4875 				nap->na_atime = dsattr.dsa_atime;
4876 				nap->na_mtime = dsattr.dsa_mtime;
4877 				nap->na_bytes = dsattr.dsa_bytes;
4878 			} else
4879 				error = ENXIO;
4880 		}
4881 		if (error == 0) {
4882 			/*
4883 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4884 			 * returns 0, just return now.  nfsrv_checkdsattr()
4885 			 * returns 0 if there is no Read/Write layout
4886 			 * plus either an Open/Write_access or Write
4887 			 * delegation issued to a client for the file.
4888 			 */
4889 			if (nfsrv_pnfsgetdsattr == 0 ||
4890 			    nfsrv_checkdsattr(vp, p) == 0) {
4891 				free(buf, M_TEMP);
4892 				return (error);
4893 			}
4894 		}
4895 
4896 		/*
4897 		 * Clear ENOATTR so the code below will attempt to do a
4898 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4899 		 * the extended attribute.
4900 		 */
4901 		if (error == ENOATTR)
4902 			error = 0;
4903 	}
4904 
4905 	origmircnt = -1;
4906 	trycnt = 0;
4907 tryagain:
4908 	if (error == 0) {
4909 		buflen = 1024;
4910 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4911 		    LK_EXCLUSIVE)
4912 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4913 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4914 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4915 		    NULL, NULL);
4916 		if (error == 0) {
4917 			for (i = 0; i < mirrorcnt; i++)
4918 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
4919 		} else
4920 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4921 	} else
4922 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4923 	if (error == 0) {
4924 		failpos = -1;
4925 		if (origmircnt == -1)
4926 			origmircnt = mirrorcnt;
4927 		/*
4928 		 * If failpos is set to a mirror#, then that mirror has
4929 		 * failed and will be disabled. For Read, Getattr and Seek, the
4930 		 * function only tries one mirror, so if that mirror has
4931 		 * failed, it will need to be retried. As such, increment
4932 		 * tryitagain for these cases.
4933 		 * For Write, Setattr and Setacl, the function tries all
4934 		 * mirrors and will not return an error for the case where
4935 		 * one mirror has failed. For these cases, the functioning
4936 		 * mirror(s) will have been modified, so a retry isn't
4937 		 * necessary. These functions will set failpos for the
4938 		 * failed mirror#.
4939 		 */
4940 		if (ioproc == NFSPROC_READDS) {
4941 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4942 			    mpp, mpp2);
4943 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4944 				/*
4945 				 * Setting failpos will cause the mirror
4946 				 * to be disabled and then a retry of this
4947 				 * read is required.
4948 				 */
4949 				failpos = 0;
4950 				error = 0;
4951 				trycnt++;
4952 			}
4953 		} else if (ioproc == NFSPROC_WRITEDS)
4954 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4955 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
4956 		else if (ioproc == NFSPROC_SETATTR)
4957 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4958 			    mirrorcnt, nap, &failpos);
4959 		else if (ioproc == NFSPROC_SETACL)
4960 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4961 			    mirrorcnt, aclp, &failpos);
4962 		else if (ioproc == NFSPROC_SEEKDS) {
4963 			error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
4964 			    p, nmp[0]);
4965 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4966 				/*
4967 				 * Setting failpos will cause the mirror
4968 				 * to be disabled and then a retry of this
4969 				 * read is required.
4970 				 */
4971 				failpos = 0;
4972 				error = 0;
4973 				trycnt++;
4974 			}
4975 		} else if (ioproc == NFSPROC_ALLOCATE)
4976 			error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
4977 			    &nmp[0], mirrorcnt, &failpos);
4978 		else if (ioproc == NFSPROC_DEALLOCATE)
4979 			error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p,
4980 			    vp, &nmp[0], mirrorcnt, &failpos);
4981 		else {
4982 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4983 			    vp, nmp[mirrorcnt - 1], nap);
4984 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4985 				/*
4986 				 * Setting failpos will cause the mirror
4987 				 * to be disabled and then a retry of this
4988 				 * getattr is required.
4989 				 */
4990 				failpos = mirrorcnt - 1;
4991 				error = 0;
4992 				trycnt++;
4993 			}
4994 		}
4995 		ds = NULL;
4996 		if (failpos >= 0) {
4997 			failnmp = nmp[failpos];
4998 			NFSLOCKMNT(failnmp);
4999 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
5000 			     NFSMNTP_CANCELRPCS)) == 0) {
5001 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
5002 				NFSUNLOCKMNT(failnmp);
5003 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
5004 				    failnmp, p);
5005 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
5006 				    failpos, ds);
5007 				if (ds != NULL)
5008 					nfsrv_killrpcs(failnmp);
5009 				NFSLOCKMNT(failnmp);
5010 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
5011 				wakeup(failnmp);
5012 			}
5013 			NFSUNLOCKMNT(failnmp);
5014 		}
5015 		for (i = 0; i < mirrorcnt; i++)
5016 			NFSVOPUNLOCK(dvp[i]);
5017 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
5018 		    trycnt);
5019 		/* Try the Read/Getattr again if a mirror was deleted. */
5020 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
5021 			goto tryagain;
5022 	} else {
5023 		/* Return ENOENT for any Extended Attribute error. */
5024 		error = ENOENT;
5025 	}
5026 	free(buf, M_TEMP);
5027 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
5028 	return (error);
5029 }
5030 
5031 /*
5032  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
5033  * attribute.
5034  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
5035  *           to be checked.  If it points to a NULL nmp, then it returns
5036  *           a suitable destination.
5037  * curnmp - If non-NULL, it is the source mount for the copy.
5038  */
5039 int
5040 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
5041     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
5042     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
5043     struct nfsmount *curnmp, int *ippos, int *dsdirp)
5044 {
5045 	struct vnode *dvp, *nvp = NULL, **tdvpp;
5046 	struct mount *mp;
5047 	struct nfsmount *nmp, *newnmp;
5048 	struct sockaddr *sad;
5049 	struct sockaddr_in *sin;
5050 	struct nfsdevice *ds, *tds, *fndds;
5051 	struct pnfsdsfile *pf;
5052 	uint32_t dsdir;
5053 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
5054 
5055 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
5056 	*mirrorcntp = 1;
5057 	tdvpp = dvpp;
5058 	if (nvpp != NULL)
5059 		*nvpp = NULL;
5060 	if (dvpp != NULL)
5061 		*dvpp = NULL;
5062 	if (ippos != NULL)
5063 		*ippos = -1;
5064 	if (newnmpp != NULL)
5065 		newnmp = *newnmpp;
5066 	else
5067 		newnmp = NULL;
5068 	mp = vp->v_mount;
5069 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5070 	    "pnfsd.dsfile", buflenp, buf, p);
5071 	mirrorcnt = *buflenp / sizeof(*pf);
5072 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
5073 	    *buflenp != sizeof(*pf) * mirrorcnt))
5074 		error = ENOATTR;
5075 
5076 	pf = (struct pnfsdsfile *)buf;
5077 	/* If curnmp != NULL, check for a match in the mirror list. */
5078 	if (curnmp != NULL && error == 0) {
5079 		fnd = 0;
5080 		for (i = 0; i < mirrorcnt; i++, pf++) {
5081 			sad = (struct sockaddr *)&pf->dsf_sin;
5082 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
5083 				if (ippos != NULL)
5084 					*ippos = i;
5085 				fnd = 1;
5086 				break;
5087 			}
5088 		}
5089 		if (fnd == 0)
5090 			error = ENXIO;
5091 	}
5092 
5093 	gotone = 0;
5094 	pf = (struct pnfsdsfile *)buf;
5095 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
5096 	    error);
5097 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
5098 		fhiszero = 0;
5099 		sad = (struct sockaddr *)&pf->dsf_sin;
5100 		sin = &pf->dsf_sin;
5101 		dsdir = pf->dsf_dir;
5102 		if (dsdir >= nfsrv_dsdirsize) {
5103 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
5104 			error = ENOATTR;
5105 		} else if (nvpp != NULL && newnmp != NULL &&
5106 		    nfsaddr2_match(sad, newnmp->nm_nam))
5107 			error = EEXIST;
5108 		if (error == 0) {
5109 			if (ippos != NULL && curnmp == NULL &&
5110 			    sad->sa_family == AF_INET &&
5111 			    sin->sin_addr.s_addr == 0)
5112 				*ippos = i;
5113 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
5114 				fhiszero = 1;
5115 			/* Use the socket address to find the mount point. */
5116 			fndds = NULL;
5117 			NFSDDSLOCK();
5118 			/* Find a match for the IP address. */
5119 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5120 				if (ds->nfsdev_nmp != NULL) {
5121 					dvp = ds->nfsdev_dvp;
5122 					nmp = VFSTONFS(dvp->v_mount);
5123 					if (nmp != ds->nfsdev_nmp)
5124 						printf("different2 nmp %p %p\n",
5125 						    nmp, ds->nfsdev_nmp);
5126 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
5127 						fndds = ds;
5128 						break;
5129 					}
5130 				}
5131 			}
5132 			if (fndds != NULL && newnmpp != NULL &&
5133 			    newnmp == NULL) {
5134 				/* Search for a place to make a mirror copy. */
5135 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
5136 				    nfsdev_list) {
5137 					if (tds->nfsdev_nmp != NULL &&
5138 					    fndds != tds &&
5139 					    ((tds->nfsdev_mdsisset == 0 &&
5140 					      fndds->nfsdev_mdsisset == 0) ||
5141 					     (tds->nfsdev_mdsisset != 0 &&
5142 					      fndds->nfsdev_mdsisset != 0 &&
5143 					      fsidcmp(&tds->nfsdev_mdsfsid,
5144 					      &mp->mnt_stat.f_fsid) == 0))) {
5145 						*newnmpp = tds->nfsdev_nmp;
5146 						break;
5147 					}
5148 				}
5149 				if (tds != NULL) {
5150 					/*
5151 					 * Move this entry to the end of the
5152 					 * list, so it won't be selected as
5153 					 * easily the next time.
5154 					 */
5155 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
5156 					    nfsdev_list);
5157 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
5158 					    nfsdev_list);
5159 				}
5160 			}
5161 			NFSDDSUNLOCK();
5162 			if (fndds != NULL) {
5163 				dvp = fndds->nfsdev_dsdir[dsdir];
5164 				if (lktype != 0 || fhiszero != 0 ||
5165 				    (nvpp != NULL && *nvpp == NULL)) {
5166 					if (fhiszero != 0)
5167 						error = vn_lock(dvp,
5168 						    LK_EXCLUSIVE);
5169 					else if (lktype != 0)
5170 						error = vn_lock(dvp, lktype);
5171 					else
5172 						error = vn_lock(dvp, LK_SHARED);
5173 					/*
5174 					 * If the file handle is all 0's, try to
5175 					 * do a Lookup against the DS to acquire
5176 					 * it.
5177 					 * If dvpp == NULL or the Lookup fails,
5178 					 * unlock dvp after the call.
5179 					 */
5180 					if (error == 0 && (fhiszero != 0 ||
5181 					    (nvpp != NULL && *nvpp == NULL))) {
5182 						error = nfsrv_pnfslookupds(vp,
5183 						    dvp, pf, &nvp, p);
5184 						if (error == 0) {
5185 							if (fhiszero != 0)
5186 								nfsrv_pnfssetfh(
5187 								    vp, pf,
5188 								    devid,
5189 								    fnamep,
5190 								    nvp, p);
5191 							if (nvpp != NULL &&
5192 							    *nvpp == NULL) {
5193 								*nvpp = nvp;
5194 								*dsdirp = dsdir;
5195 							} else
5196 								vput(nvp);
5197 						}
5198 						if (error != 0 || lktype == 0)
5199 							NFSVOPUNLOCK(dvp);
5200 					}
5201 				}
5202 				if (error == 0) {
5203 					gotone++;
5204 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
5205 					if (devid != NULL) {
5206 						NFSBCOPY(fndds->nfsdev_deviceid,
5207 						    devid, NFSX_V4DEVICEID);
5208 						devid += NFSX_V4DEVICEID;
5209 					}
5210 					if (dvpp != NULL)
5211 						*tdvpp++ = dvp;
5212 					if (fhp != NULL)
5213 						NFSBCOPY(&pf->dsf_fh, fhp++,
5214 						    NFSX_MYFH);
5215 					if (fnamep != NULL && gotone == 1)
5216 						strlcpy(fnamep,
5217 						    pf->dsf_filename,
5218 						    sizeof(pf->dsf_filename));
5219 				} else
5220 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
5221 					    "err=%d\n", error);
5222 			}
5223 		}
5224 	}
5225 	if (error == 0 && gotone == 0)
5226 		error = ENOENT;
5227 
5228 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
5229 	    error);
5230 	if (error == 0)
5231 		*mirrorcntp = gotone;
5232 	else {
5233 		if (gotone > 0 && dvpp != NULL) {
5234 			/*
5235 			 * If the error didn't occur on the first one and
5236 			 * dvpp != NULL, the one(s) prior to the failure will
5237 			 * have locked dvp's that need to be unlocked.
5238 			 */
5239 			for (i = 0; i < gotone; i++) {
5240 				NFSVOPUNLOCK(*dvpp);
5241 				*dvpp++ = NULL;
5242 			}
5243 		}
5244 		/*
5245 		 * If it found the vnode to be copied from before a failure,
5246 		 * it needs to be vput()'d.
5247 		 */
5248 		if (nvpp != NULL && *nvpp != NULL) {
5249 			vput(*nvpp);
5250 			*nvpp = NULL;
5251 		}
5252 	}
5253 	return (error);
5254 }
5255 
5256 /*
5257  * Set the extended attribute for the Change attribute.
5258  */
5259 static int
5260 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
5261 {
5262 	struct pnfsdsattr dsattr;
5263 	int error;
5264 
5265 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
5266 	dsattr.dsa_filerev = nap->na_filerev;
5267 	dsattr.dsa_size = nap->na_size;
5268 	dsattr.dsa_atime = nap->na_atime;
5269 	dsattr.dsa_mtime = nap->na_mtime;
5270 	dsattr.dsa_bytes = nap->na_bytes;
5271 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5272 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
5273 	if (error != 0)
5274 		printf("pNFS: setextattr=%d\n", error);
5275 	return (error);
5276 }
5277 
5278 static int
5279 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5280     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
5281 {
5282 	uint32_t *tl;
5283 	struct nfsrv_descript *nd;
5284 	nfsv4stateid_t st;
5285 	struct mbuf *m, *m2;
5286 	int error = 0, retlen, tlen, trimlen;
5287 
5288 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
5289 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5290 	*mpp = NULL;
5291 	/*
5292 	 * Use a stateid where other is an alternating 01010 pattern and
5293 	 * seqid is 0xffffffff.  This value is not defined as special by
5294 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5295 	 * MDS->DS proxy operation.
5296 	 */
5297 	st.other[0] = 0x55555555;
5298 	st.other[1] = 0x55555555;
5299 	st.other[2] = 0x55555555;
5300 	st.seqid = 0xffffffff;
5301 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5302 	    NULL, NULL, 0, 0, cred);
5303 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5304 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
5305 	txdr_hyper(off, tl);
5306 	*(tl + 2) = txdr_unsigned(len);
5307 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5308 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5309 	if (error != 0) {
5310 		free(nd, M_TEMP);
5311 		return (error);
5312 	}
5313 	if (nd->nd_repstat == 0) {
5314 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5315 		NFSM_STRSIZ(retlen, len);
5316 		if (retlen > 0) {
5317 			/* Trim off the pre-data XDR from the mbuf chain. */
5318 			m = nd->nd_mrep;
5319 			while (m != NULL && m != nd->nd_md) {
5320 				if (m->m_next == nd->nd_md) {
5321 					m->m_next = NULL;
5322 					m_freem(nd->nd_mrep);
5323 					nd->nd_mrep = m = nd->nd_md;
5324 				} else
5325 					m = m->m_next;
5326 			}
5327 			if (m == NULL) {
5328 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5329 				error = ENOENT;
5330 				goto nfsmout;
5331 			}
5332 
5333 			/*
5334 			 * Now, adjust first mbuf so that any XDR before the
5335 			 * read data is skipped over.
5336 			 */
5337 			trimlen = nd->nd_dpos - mtod(m, char *);
5338 			if (trimlen > 0) {
5339 				m->m_len -= trimlen;
5340 				NFSM_DATAP(m, trimlen);
5341 			}
5342 
5343 			/*
5344 			 * Truncate the mbuf chain at retlen bytes of data,
5345 			 * plus XDR padding that brings the length up to a
5346 			 * multiple of 4.
5347 			 */
5348 			tlen = NFSM_RNDUP(retlen);
5349 			do {
5350 				if (m->m_len >= tlen) {
5351 					m->m_len = tlen;
5352 					tlen = 0;
5353 					m2 = m->m_next;
5354 					m->m_next = NULL;
5355 					m_freem(m2);
5356 					break;
5357 				}
5358 				tlen -= m->m_len;
5359 				m = m->m_next;
5360 			} while (m != NULL);
5361 			if (tlen > 0) {
5362 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5363 				error = ENOENT;
5364 				goto nfsmout;
5365 			}
5366 			*mpp = nd->nd_mrep;
5367 			*mpendp = m;
5368 			nd->nd_mrep = NULL;
5369 		}
5370 	} else
5371 		error = nd->nd_repstat;
5372 nfsmout:
5373 	/* If nd->nd_mrep is already NULL, this is a no-op. */
5374 	m_freem(nd->nd_mrep);
5375 	free(nd, M_TEMP);
5376 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5377 	return (error);
5378 }
5379 
5380 /*
5381  * Do a write RPC on a DS data file, using this structure for the arguments,
5382  * so that this function can be executed by a separate kernel process.
5383  */
5384 struct nfsrvwritedsdorpc {
5385 	int			done;
5386 	int			inprog;
5387 	struct task		tsk;
5388 	fhandle_t		fh;
5389 	off_t			off;
5390 	int			len;
5391 	struct nfsmount		*nmp;
5392 	struct ucred		*cred;
5393 	NFSPROC_T		*p;
5394 	struct mbuf		*m;
5395 	int			err;
5396 };
5397 
5398 static int
5399 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5400     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5401 {
5402 	uint32_t *tl;
5403 	struct nfsrv_descript *nd;
5404 	nfsattrbit_t attrbits;
5405 	nfsv4stateid_t st;
5406 	int commit, error, retlen;
5407 
5408 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5409 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5410 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5411 
5412 	/*
5413 	 * Use a stateid where other is an alternating 01010 pattern and
5414 	 * seqid is 0xffffffff.  This value is not defined as special by
5415 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5416 	 * MDS->DS proxy operation.
5417 	 */
5418 	st.other[0] = 0x55555555;
5419 	st.other[1] = 0x55555555;
5420 	st.other[2] = 0x55555555;
5421 	st.seqid = 0xffffffff;
5422 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5423 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5424 	txdr_hyper(off, tl);
5425 	tl += 2;
5426 	/*
5427 	 * Do all writes FileSync, since the server doesn't hold onto dirty
5428 	 * buffers.  Since clients should be accessing the DS servers directly
5429 	 * using the pNFS layouts, this just needs to work correctly as a
5430 	 * fallback.
5431 	 */
5432 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5433 	*tl = txdr_unsigned(len);
5434 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5435 
5436 	/* Put data in mbuf chain. */
5437 	nd->nd_mb->m_next = m;
5438 
5439 	/* Set nd_mb and nd_bpos to end of data. */
5440 	while (m->m_next != NULL)
5441 		m = m->m_next;
5442 	nd->nd_mb = m;
5443 	nfsm_set(nd, m->m_len);
5444 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5445 
5446 	/* Do a Getattr for the attributes that change upon writing. */
5447 	NFSZERO_ATTRBIT(&attrbits);
5448 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5449 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5450 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5451 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5452 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5453 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5454 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5455 	(void) nfsrv_putattrbit(nd, &attrbits);
5456 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5457 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5458 	if (error != 0) {
5459 		free(nd, M_TEMP);
5460 		return (error);
5461 	}
5462 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5463 	/* Get rid of weak cache consistency data for now. */
5464 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5465 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5466 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5467 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5468 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5469 		if (error != 0)
5470 			goto nfsmout;
5471 		/*
5472 		 * Get rid of Op# and status for next op.
5473 		 */
5474 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5475 		if (*++tl != 0)
5476 			nd->nd_flag |= ND_NOMOREDATA;
5477 	}
5478 	if (nd->nd_repstat == 0) {
5479 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5480 		retlen = fxdr_unsigned(int, *tl++);
5481 		commit = fxdr_unsigned(int, *tl);
5482 		if (commit != NFSWRITE_FILESYNC)
5483 			error = NFSERR_IO;
5484 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5485 		    retlen, commit, error);
5486 	} else
5487 		error = nd->nd_repstat;
5488 	/* We have no use for the Write Verifier since we use FileSync. */
5489 
5490 	/*
5491 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5492 	 * on the Metadata file, so its attributes will be what the file's
5493 	 * would be if it had been written.
5494 	 */
5495 	if (error == 0) {
5496 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5497 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5498 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5499 	}
5500 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5501 nfsmout:
5502 	m_freem(nd->nd_mrep);
5503 	free(nd, M_TEMP);
5504 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5505 	return (error);
5506 }
5507 
5508 /*
5509  * Start up the thread that will execute nfsrv_writedsdorpc().
5510  */
5511 static void
5512 start_writedsdorpc(void *arg, int pending)
5513 {
5514 	struct nfsrvwritedsdorpc *drpc;
5515 
5516 	drpc = (struct nfsrvwritedsdorpc *)arg;
5517 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5518 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5519 	drpc->done = 1;
5520 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5521 }
5522 
5523 static int
5524 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5525     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5526     struct mbuf **mpp, char *cp, int *failposp)
5527 {
5528 	struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5529 	struct nfsvattr na;
5530 	struct mbuf *m;
5531 	int error, i, offs, ret, timo;
5532 
5533 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5534 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5535 	drpc = NULL;
5536 	if (mirrorcnt > 1)
5537 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5538 		    M_WAITOK);
5539 
5540 	/* Calculate offset in mbuf chain that data starts. */
5541 	offs = cp - mtod(*mpp, char *);
5542 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5543 
5544 	/*
5545 	 * Do the write RPC for every DS, using a separate kernel process
5546 	 * for every DS except the last one.
5547 	 */
5548 	error = 0;
5549 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5550 		tdrpc->done = 0;
5551 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5552 		tdrpc->off = off;
5553 		tdrpc->len = len;
5554 		tdrpc->nmp = *nmpp;
5555 		tdrpc->cred = cred;
5556 		tdrpc->p = p;
5557 		tdrpc->inprog = 0;
5558 		tdrpc->err = 0;
5559 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5560 		ret = EIO;
5561 		if (nfs_pnfsiothreads != 0) {
5562 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5563 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5564 			    ret);
5565 		}
5566 		if (ret != 0) {
5567 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5568 			    tdrpc->m, cred, p);
5569 			if (nfsds_failerr(ret) && *failposp == -1)
5570 				*failposp = i;
5571 			else if (error == 0 && ret != 0)
5572 				error = ret;
5573 		}
5574 		nmpp++;
5575 		fhp++;
5576 	}
5577 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5578 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5579 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5580 		*failposp = mirrorcnt - 1;
5581 	else if (error == 0 && ret != 0)
5582 		error = ret;
5583 	if (error == 0)
5584 		error = nfsrv_setextattr(vp, &na, p);
5585 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5586 	tdrpc = drpc;
5587 	timo = hz / 50;		/* Wait for 20msec. */
5588 	if (timo < 1)
5589 		timo = 1;
5590 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5591 		/* Wait for RPCs on separate threads to complete. */
5592 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5593 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5594 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5595 			*failposp = i;
5596 		else if (error == 0 && tdrpc->err != 0)
5597 			error = tdrpc->err;
5598 	}
5599 	free(drpc, M_TEMP);
5600 	return (error);
5601 }
5602 
5603 /*
5604  * Do a allocate RPC on a DS data file, using this structure for the arguments,
5605  * so that this function can be executed by a separate kernel process.
5606  */
5607 struct nfsrvallocatedsdorpc {
5608 	int			done;
5609 	int			inprog;
5610 	struct task		tsk;
5611 	fhandle_t		fh;
5612 	off_t			off;
5613 	off_t			len;
5614 	struct nfsmount		*nmp;
5615 	struct ucred		*cred;
5616 	NFSPROC_T		*p;
5617 	int			err;
5618 };
5619 
5620 static int
5621 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5622     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5623 {
5624 	uint32_t *tl;
5625 	struct nfsrv_descript *nd;
5626 	nfsattrbit_t attrbits;
5627 	nfsv4stateid_t st;
5628 	int error;
5629 
5630 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5631 	nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5632 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5633 
5634 	/*
5635 	 * Use a stateid where other is an alternating 01010 pattern and
5636 	 * seqid is 0xffffffff.  This value is not defined as special by
5637 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5638 	 * MDS->DS proxy operation.
5639 	 */
5640 	st.other[0] = 0x55555555;
5641 	st.other[1] = 0x55555555;
5642 	st.other[2] = 0x55555555;
5643 	st.seqid = 0xffffffff;
5644 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5645 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5646 	txdr_hyper(off, tl); tl += 2;
5647 	txdr_hyper(len, tl); tl += 2;
5648 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5649 
5650 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5651 	NFSGETATTR_ATTRBIT(&attrbits);
5652 	nfsrv_putattrbit(nd, &attrbits);
5653 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5654 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5655 	if (error != 0) {
5656 		free(nd, M_TEMP);
5657 		return (error);
5658 	}
5659 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5660 	    nd->nd_repstat);
5661 	if (nd->nd_repstat == 0) {
5662 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5663 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5664 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5665 	} else
5666 		error = nd->nd_repstat;
5667 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5668 nfsmout:
5669 	m_freem(nd->nd_mrep);
5670 	free(nd, M_TEMP);
5671 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5672 	return (error);
5673 }
5674 
5675 /*
5676  * Start up the thread that will execute nfsrv_allocatedsdorpc().
5677  */
5678 static void
5679 start_allocatedsdorpc(void *arg, int pending)
5680 {
5681 	struct nfsrvallocatedsdorpc *drpc;
5682 
5683 	drpc = (struct nfsrvallocatedsdorpc *)arg;
5684 	drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5685 	    drpc->len, NULL, drpc->cred, drpc->p);
5686 	drpc->done = 1;
5687 	NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5688 }
5689 
5690 static int
5691 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5692     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5693     int *failposp)
5694 {
5695 	struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5696 	struct nfsvattr na;
5697 	int error, i, ret, timo;
5698 
5699 	NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5700 	drpc = NULL;
5701 	if (mirrorcnt > 1)
5702 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5703 		    M_WAITOK);
5704 
5705 	/*
5706 	 * Do the allocate RPC for every DS, using a separate kernel process
5707 	 * for every DS except the last one.
5708 	 */
5709 	error = 0;
5710 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5711 		tdrpc->done = 0;
5712 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5713 		tdrpc->off = off;
5714 		tdrpc->len = len;
5715 		tdrpc->nmp = *nmpp;
5716 		tdrpc->cred = cred;
5717 		tdrpc->p = p;
5718 		tdrpc->inprog = 0;
5719 		tdrpc->err = 0;
5720 		ret = EIO;
5721 		if (nfs_pnfsiothreads != 0) {
5722 			ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5723 			NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5724 			    ret);
5725 		}
5726 		if (ret != 0) {
5727 			ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5728 			    cred, p);
5729 			if (nfsds_failerr(ret) && *failposp == -1)
5730 				*failposp = i;
5731 			else if (error == 0 && ret != 0)
5732 				error = ret;
5733 		}
5734 		nmpp++;
5735 		fhp++;
5736 	}
5737 	ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5738 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5739 		*failposp = mirrorcnt - 1;
5740 	else if (error == 0 && ret != 0)
5741 		error = ret;
5742 	if (error == 0)
5743 		error = nfsrv_setextattr(vp, &na, p);
5744 	NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5745 	tdrpc = drpc;
5746 	timo = hz / 50;		/* Wait for 20msec. */
5747 	if (timo < 1)
5748 		timo = 1;
5749 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5750 		/* Wait for RPCs on separate threads to complete. */
5751 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5752 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5753 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5754 			*failposp = i;
5755 		else if (error == 0 && tdrpc->err != 0)
5756 			error = tdrpc->err;
5757 	}
5758 	free(drpc, M_TEMP);
5759 	return (error);
5760 }
5761 
5762 /*
5763  * Do a deallocate RPC on a DS data file, using this structure for the
5764  * arguments, so that this function can be executed by a separate kernel
5765  * process.
5766  */
5767 struct nfsrvdeallocatedsdorpc {
5768 	int			done;
5769 	int			inprog;
5770 	struct task		tsk;
5771 	fhandle_t		fh;
5772 	off_t			off;
5773 	off_t			len;
5774 	struct nfsmount		*nmp;
5775 	struct ucred		*cred;
5776 	NFSPROC_T		*p;
5777 	int			err;
5778 };
5779 
5780 static int
5781 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5782     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5783 {
5784 	uint32_t *tl;
5785 	struct nfsrv_descript *nd;
5786 	nfsattrbit_t attrbits;
5787 	nfsv4stateid_t st;
5788 	int error;
5789 
5790 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5791 	nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp,
5792 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5793 
5794 	/*
5795 	 * Use a stateid where other is an alternating 01010 pattern and
5796 	 * seqid is 0xffffffff.  This value is not defined as special by
5797 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5798 	 * MDS->DS proxy operation.
5799 	 */
5800 	st.other[0] = 0x55555555;
5801 	st.other[1] = 0x55555555;
5802 	st.other[2] = 0x55555555;
5803 	st.seqid = 0xffffffff;
5804 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5805 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5806 	txdr_hyper(off, tl); tl += 2;
5807 	txdr_hyper(len, tl); tl += 2;
5808 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len);
5809 
5810 	/* Do a Getattr for the attributes that change upon writing. */
5811 	NFSZERO_ATTRBIT(&attrbits);
5812 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5813 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5814 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5815 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5816 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5817 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5818 	nfsrv_putattrbit(nd, &attrbits);
5819 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5820 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5821 	if (error != 0) {
5822 		free(nd, M_TEMP);
5823 		return (error);
5824 	}
5825 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n",
5826 	    nd->nd_repstat);
5827 	/* Get rid of weak cache consistency data for now. */
5828 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5829 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5830 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5831 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5832 		NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error);
5833 		if (error != 0)
5834 			goto nfsmout;
5835 		/*
5836 		 * Get rid of Op# and status for next op.
5837 		 */
5838 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5839 		if (*++tl != 0)
5840 			nd->nd_flag |= ND_NOMOREDATA;
5841 	}
5842 	if (nd->nd_repstat == 0) {
5843 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5844 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5845 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5846 	} else
5847 		error = nd->nd_repstat;
5848 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error);
5849 nfsmout:
5850 	m_freem(nd->nd_mrep);
5851 	free(nd, M_TEMP);
5852 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error);
5853 	return (error);
5854 }
5855 
5856 /*
5857  * Start up the thread that will execute nfsrv_deallocatedsdorpc().
5858  */
5859 static void
5860 start_deallocatedsdorpc(void *arg, int pending)
5861 {
5862 	struct nfsrvdeallocatedsdorpc *drpc;
5863 
5864 	drpc = (struct nfsrvdeallocatedsdorpc *)arg;
5865 	drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5866 	    drpc->len, NULL, drpc->cred, drpc->p);
5867 	drpc->done = 1;
5868 	NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err);
5869 }
5870 
5871 static int
5872 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5873     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5874     int *failposp)
5875 {
5876 	struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL;
5877 	struct nfsvattr na;
5878 	int error, i, ret, timo;
5879 
5880 	NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n");
5881 	drpc = NULL;
5882 	if (mirrorcnt > 1)
5883 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5884 		    M_WAITOK);
5885 
5886 	/*
5887 	 * Do the deallocate RPC for every DS, using a separate kernel process
5888 	 * for every DS except the last one.
5889 	 */
5890 	error = 0;
5891 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5892 		tdrpc->done = 0;
5893 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5894 		tdrpc->off = off;
5895 		tdrpc->len = len;
5896 		tdrpc->nmp = *nmpp;
5897 		tdrpc->cred = cred;
5898 		tdrpc->p = p;
5899 		tdrpc->inprog = 0;
5900 		tdrpc->err = 0;
5901 		ret = EIO;
5902 		if (nfs_pnfsiothreads != 0) {
5903 			ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc);
5904 			NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n",
5905 			    ret);
5906 		}
5907 		if (ret != 0) {
5908 			ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len,
5909 			    NULL, cred, p);
5910 			if (nfsds_failerr(ret) && *failposp == -1)
5911 				*failposp = i;
5912 			else if (error == 0 && ret != 0)
5913 				error = ret;
5914 		}
5915 		nmpp++;
5916 		fhp++;
5917 	}
5918 	ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5919 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5920 		*failposp = mirrorcnt - 1;
5921 	else if (error == 0 && ret != 0)
5922 		error = ret;
5923 	if (error == 0)
5924 		error = nfsrv_setextattr(vp, &na, p);
5925 	NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error);
5926 	tdrpc = drpc;
5927 	timo = hz / 50;		/* Wait for 20msec. */
5928 	if (timo < 1)
5929 		timo = 1;
5930 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5931 		/* Wait for RPCs on separate threads to complete. */
5932 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5933 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5934 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5935 			*failposp = i;
5936 		else if (error == 0 && tdrpc->err != 0)
5937 			error = tdrpc->err;
5938 	}
5939 	free(drpc, M_TEMP);
5940 	return (error);
5941 }
5942 
5943 static int
5944 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5945     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5946     struct nfsvattr *dsnap)
5947 {
5948 	uint32_t *tl;
5949 	struct nfsrv_descript *nd;
5950 	nfsv4stateid_t st;
5951 	nfsattrbit_t attrbits;
5952 	int error;
5953 
5954 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5955 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5956 	/*
5957 	 * Use a stateid where other is an alternating 01010 pattern and
5958 	 * seqid is 0xffffffff.  This value is not defined as special by
5959 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5960 	 * MDS->DS proxy operation.
5961 	 */
5962 	st.other[0] = 0x55555555;
5963 	st.other[1] = 0x55555555;
5964 	st.other[2] = 0x55555555;
5965 	st.seqid = 0xffffffff;
5966 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5967 	    NULL, NULL, 0, 0, cred);
5968 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5969 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5970 
5971 	/* Do a Getattr for the attributes that change due to writing. */
5972 	NFSZERO_ATTRBIT(&attrbits);
5973 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5974 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5975 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5976 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5977 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5978 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5979 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5980 	(void) nfsrv_putattrbit(nd, &attrbits);
5981 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5982 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5983 	if (error != 0) {
5984 		free(nd, M_TEMP);
5985 		return (error);
5986 	}
5987 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5988 	    nd->nd_repstat);
5989 	/* Get rid of weak cache consistency data for now. */
5990 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5991 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5992 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5993 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5994 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5995 		if (error != 0)
5996 			goto nfsmout;
5997 		/*
5998 		 * Get rid of Op# and status for next op.
5999 		 */
6000 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
6001 		if (*++tl != 0)
6002 			nd->nd_flag |= ND_NOMOREDATA;
6003 	}
6004 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
6005 	if (error != 0)
6006 		goto nfsmout;
6007 	if (nd->nd_repstat != 0)
6008 		error = nd->nd_repstat;
6009 	/*
6010 	 * Get the Change, Size, Access Time and Modify Time attributes and set
6011 	 * on the Metadata file, so its attributes will be what the file's
6012 	 * would be if it had been written.
6013 	 */
6014 	if (error == 0) {
6015 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
6016 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
6017 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
6018 	}
6019 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
6020 nfsmout:
6021 	m_freem(nd->nd_mrep);
6022 	free(nd, M_TEMP);
6023 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
6024 	return (error);
6025 }
6026 
6027 struct nfsrvsetattrdsdorpc {
6028 	int			done;
6029 	int			inprog;
6030 	struct task		tsk;
6031 	fhandle_t		fh;
6032 	struct nfsmount		*nmp;
6033 	struct vnode		*vp;
6034 	struct ucred		*cred;
6035 	NFSPROC_T		*p;
6036 	struct nfsvattr		na;
6037 	struct nfsvattr		dsna;
6038 	int			err;
6039 };
6040 
6041 /*
6042  * Start up the thread that will execute nfsrv_setattrdsdorpc().
6043  */
6044 static void
6045 start_setattrdsdorpc(void *arg, int pending)
6046 {
6047 	struct nfsrvsetattrdsdorpc *drpc;
6048 
6049 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
6050 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
6051 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
6052 	drpc->done = 1;
6053 }
6054 
6055 static int
6056 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6057     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
6058     struct nfsvattr *nap, int *failposp)
6059 {
6060 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
6061 	struct nfsvattr na;
6062 	int error, i, ret, timo;
6063 
6064 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
6065 	drpc = NULL;
6066 	if (mirrorcnt > 1)
6067 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6068 		    M_WAITOK);
6069 
6070 	/*
6071 	 * Do the setattr RPC for every DS, using a separate kernel process
6072 	 * for every DS except the last one.
6073 	 */
6074 	error = 0;
6075 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6076 		tdrpc->done = 0;
6077 		tdrpc->inprog = 0;
6078 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6079 		tdrpc->nmp = *nmpp;
6080 		tdrpc->vp = vp;
6081 		tdrpc->cred = cred;
6082 		tdrpc->p = p;
6083 		tdrpc->na = *nap;
6084 		tdrpc->err = 0;
6085 		ret = EIO;
6086 		if (nfs_pnfsiothreads != 0) {
6087 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
6088 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
6089 			    ret);
6090 		}
6091 		if (ret != 0) {
6092 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
6093 			    &na);
6094 			if (nfsds_failerr(ret) && *failposp == -1)
6095 				*failposp = i;
6096 			else if (error == 0 && ret != 0)
6097 				error = ret;
6098 		}
6099 		nmpp++;
6100 		fhp++;
6101 	}
6102 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
6103 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6104 		*failposp = mirrorcnt - 1;
6105 	else if (error == 0 && ret != 0)
6106 		error = ret;
6107 	if (error == 0)
6108 		error = nfsrv_setextattr(vp, &na, p);
6109 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
6110 	tdrpc = drpc;
6111 	timo = hz / 50;		/* Wait for 20msec. */
6112 	if (timo < 1)
6113 		timo = 1;
6114 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6115 		/* Wait for RPCs on separate threads to complete. */
6116 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6117 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
6118 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6119 			*failposp = i;
6120 		else if (error == 0 && tdrpc->err != 0)
6121 			error = tdrpc->err;
6122 	}
6123 	free(drpc, M_TEMP);
6124 	return (error);
6125 }
6126 
6127 /*
6128  * Do a Setattr of an NFSv4 ACL on the DS file.
6129  */
6130 static int
6131 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6132     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
6133 {
6134 	struct nfsrv_descript *nd;
6135 	nfsv4stateid_t st;
6136 	nfsattrbit_t attrbits;
6137 	int error;
6138 
6139 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
6140 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6141 	/*
6142 	 * Use a stateid where other is an alternating 01010 pattern and
6143 	 * seqid is 0xffffffff.  This value is not defined as special by
6144 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6145 	 * MDS->DS proxy operation.
6146 	 */
6147 	st.other[0] = 0x55555555;
6148 	st.other[1] = 0x55555555;
6149 	st.other[2] = 0x55555555;
6150 	st.seqid = 0xffffffff;
6151 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
6152 	    NULL, NULL, 0, 0, cred);
6153 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6154 	NFSZERO_ATTRBIT(&attrbits);
6155 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
6156 	/*
6157 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
6158 	 * so passing in the metadata "vp" will be ok, since it is of
6159 	 * the same type (VREG).
6160 	 */
6161 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
6162 	    NULL, 0, 0, 0, 0, 0, NULL);
6163 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6164 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6165 	if (error != 0) {
6166 		free(nd, M_TEMP);
6167 		return (error);
6168 	}
6169 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
6170 	    nd->nd_repstat);
6171 	error = nd->nd_repstat;
6172 	m_freem(nd->nd_mrep);
6173 	free(nd, M_TEMP);
6174 	return (error);
6175 }
6176 
6177 struct nfsrvsetacldsdorpc {
6178 	int			done;
6179 	int			inprog;
6180 	struct task		tsk;
6181 	fhandle_t		fh;
6182 	struct nfsmount		*nmp;
6183 	struct vnode		*vp;
6184 	struct ucred		*cred;
6185 	NFSPROC_T		*p;
6186 	struct acl		*aclp;
6187 	int			err;
6188 };
6189 
6190 /*
6191  * Start up the thread that will execute nfsrv_setacldsdorpc().
6192  */
6193 static void
6194 start_setacldsdorpc(void *arg, int pending)
6195 {
6196 	struct nfsrvsetacldsdorpc *drpc;
6197 
6198 	drpc = (struct nfsrvsetacldsdorpc *)arg;
6199 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
6200 	    drpc->vp, drpc->nmp, drpc->aclp);
6201 	drpc->done = 1;
6202 }
6203 
6204 static int
6205 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6206     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
6207     int *failposp)
6208 {
6209 	struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
6210 	int error, i, ret, timo;
6211 
6212 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
6213 	drpc = NULL;
6214 	if (mirrorcnt > 1)
6215 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6216 		    M_WAITOK);
6217 
6218 	/*
6219 	 * Do the setattr RPC for every DS, using a separate kernel process
6220 	 * for every DS except the last one.
6221 	 */
6222 	error = 0;
6223 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6224 		tdrpc->done = 0;
6225 		tdrpc->inprog = 0;
6226 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6227 		tdrpc->nmp = *nmpp;
6228 		tdrpc->vp = vp;
6229 		tdrpc->cred = cred;
6230 		tdrpc->p = p;
6231 		tdrpc->aclp = aclp;
6232 		tdrpc->err = 0;
6233 		ret = EIO;
6234 		if (nfs_pnfsiothreads != 0) {
6235 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
6236 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
6237 			    ret);
6238 		}
6239 		if (ret != 0) {
6240 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
6241 			    aclp);
6242 			if (nfsds_failerr(ret) && *failposp == -1)
6243 				*failposp = i;
6244 			else if (error == 0 && ret != 0)
6245 				error = ret;
6246 		}
6247 		nmpp++;
6248 		fhp++;
6249 	}
6250 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
6251 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6252 		*failposp = mirrorcnt - 1;
6253 	else if (error == 0 && ret != 0)
6254 		error = ret;
6255 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
6256 	tdrpc = drpc;
6257 	timo = hz / 50;		/* Wait for 20msec. */
6258 	if (timo < 1)
6259 		timo = 1;
6260 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6261 		/* Wait for RPCs on separate threads to complete. */
6262 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6263 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
6264 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6265 			*failposp = i;
6266 		else if (error == 0 && tdrpc->err != 0)
6267 			error = tdrpc->err;
6268 	}
6269 	free(drpc, M_TEMP);
6270 	return (error);
6271 }
6272 
6273 /*
6274  * Getattr call to the DS for the attributes that change due to writing.
6275  */
6276 static int
6277 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6278     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
6279 {
6280 	struct nfsrv_descript *nd;
6281 	int error;
6282 	nfsattrbit_t attrbits;
6283 
6284 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
6285 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6286 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
6287 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6288 	NFSZERO_ATTRBIT(&attrbits);
6289 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6290 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6291 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6292 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6293 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6294 	(void) nfsrv_putattrbit(nd, &attrbits);
6295 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6296 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6297 	if (error != 0) {
6298 		free(nd, M_TEMP);
6299 		return (error);
6300 	}
6301 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
6302 	    nd->nd_repstat);
6303 	if (nd->nd_repstat == 0) {
6304 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
6305 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
6306 		    NULL, NULL);
6307 		/*
6308 		 * We can only save the updated values in the extended
6309 		 * attribute if the vp is exclusively locked.
6310 		 * This should happen when any of the following operations
6311 		 * occur on the vnode:
6312 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
6313 		 * As such, the updated extended attribute should get saved
6314 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
6315 		 * attributes to be returned without calling this function.
6316 		 */
6317 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
6318 			error = nfsrv_setextattr(vp, nap, p);
6319 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
6320 			    error);
6321 		}
6322 	} else
6323 		error = nd->nd_repstat;
6324 	m_freem(nd->nd_mrep);
6325 	free(nd, M_TEMP);
6326 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
6327 	return (error);
6328 }
6329 
6330 /*
6331  * Seek call to a DS.
6332  */
6333 static int
6334 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
6335     struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
6336 {
6337 	uint32_t *tl;
6338 	struct nfsrv_descript *nd;
6339 	nfsv4stateid_t st;
6340 	int error;
6341 
6342 	NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
6343 	/*
6344 	 * Use a stateid where other is an alternating 01010 pattern and
6345 	 * seqid is 0xffffffff.  This value is not defined as special by
6346 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6347 	 * MDS->DS proxy operation.
6348 	 */
6349 	st.other[0] = 0x55555555;
6350 	st.other[1] = 0x55555555;
6351 	st.other[2] = 0x55555555;
6352 	st.seqid = 0xffffffff;
6353 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6354 	nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
6355 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6356 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6357 	NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
6358 	txdr_hyper(*offp, tl); tl += 2;
6359 	*tl = txdr_unsigned(content);
6360 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6361 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6362 	if (error != 0) {
6363 		free(nd, M_TEMP);
6364 		return (error);
6365 	}
6366 	NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
6367 	if (nd->nd_repstat == 0) {
6368 		NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
6369 		if (*tl++ == newnfs_true)
6370 			*eofp = true;
6371 		else
6372 			*eofp = false;
6373 		*offp = fxdr_hyper(tl);
6374 	} else
6375 		error = nd->nd_repstat;
6376 nfsmout:
6377 	m_freem(nd->nd_mrep);
6378 	free(nd, M_TEMP);
6379 	NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
6380 	return (error);
6381 }
6382 
6383 /*
6384  * Get the device id and file handle for a DS file.
6385  */
6386 int
6387 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
6388     fhandle_t *fhp, char *devid)
6389 {
6390 	int buflen, error;
6391 	char *buf;
6392 
6393 	buflen = 1024;
6394 	buf = malloc(buflen, M_TEMP, M_WAITOK);
6395 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
6396 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
6397 	free(buf, M_TEMP);
6398 	return (error);
6399 }
6400 
6401 /*
6402  * Do a Lookup against the DS for the filename.
6403  */
6404 static int
6405 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
6406     struct vnode **nvpp, NFSPROC_T *p)
6407 {
6408 	struct nameidata named;
6409 	struct ucred *tcred;
6410 	char *bufp;
6411 	u_long *hashp;
6412 	struct vnode *nvp;
6413 	int error;
6414 
6415 	tcred = newnfs_getcred();
6416 	named.ni_cnd.cn_nameiop = LOOKUP;
6417 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
6418 	named.ni_cnd.cn_cred = tcred;
6419 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF;
6420 	nfsvno_setpathbuf(&named, &bufp, &hashp);
6421 	named.ni_cnd.cn_nameptr = bufp;
6422 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
6423 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
6424 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
6425 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
6426 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
6427 	NFSFREECRED(tcred);
6428 	nfsvno_relpathbuf(&named);
6429 	if (error == 0)
6430 		*nvpp = nvp;
6431 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
6432 	return (error);
6433 }
6434 
6435 /*
6436  * Set the file handle to the correct one.
6437  */
6438 static void
6439 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
6440     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
6441 {
6442 	struct nfsnode *np;
6443 	int ret = 0;
6444 
6445 	np = VTONFS(nvp);
6446 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
6447 	/*
6448 	 * We can only do a vn_set_extattr() if the vnode is exclusively
6449 	 * locked and vn_start_write() has been done.  If devid != NULL or
6450 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
6451 	 * may not have been done.
6452 	 * If not done now, it will be done on a future call.
6453 	 */
6454 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
6455 	    LK_EXCLUSIVE)
6456 		ret = vn_extattr_set(vp, IO_NODELOCKED,
6457 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
6458 		    (char *)pf, p);
6459 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
6460 }
6461 
6462 /*
6463  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
6464  * when the DS has failed.
6465  */
6466 void
6467 nfsrv_killrpcs(struct nfsmount *nmp)
6468 {
6469 
6470 	/*
6471 	 * Call newnfs_nmcancelreqs() to cause
6472 	 * any RPCs in progress on the mount point to
6473 	 * fail.
6474 	 * This will cause any process waiting for an
6475 	 * RPC to complete while holding a vnode lock
6476 	 * on the mounted-on vnode (such as "df" or
6477 	 * a non-forced "umount") to fail.
6478 	 * This will unlock the mounted-on vnode so
6479 	 * a forced dismount can succeed.
6480 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
6481 	 * called.
6482 	 */
6483 	newnfs_nmcancelreqs(nmp);
6484 }
6485 
6486 /*
6487  * Sum up the statfs info for each of the DSs, so that the client will
6488  * receive the total for all DSs.
6489  */
6490 static int
6491 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6492 {
6493 	struct statfs *tsf;
6494 	struct nfsdevice *ds;
6495 	struct vnode **dvpp, **tdvpp, *dvp;
6496 	uint64_t tot;
6497 	int cnt, error = 0, i;
6498 
6499 	if (nfsrv_devidcnt <= 0)
6500 		return (ENXIO);
6501 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6502 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6503 
6504 	/* Get an array of the dvps for the DSs. */
6505 	tdvpp = dvpp;
6506 	i = 0;
6507 	NFSDDSLOCK();
6508 	/* First, search for matches for same file system. */
6509 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6510 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6511 		    fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6512 			if (++i > nfsrv_devidcnt)
6513 				break;
6514 			*tdvpp++ = ds->nfsdev_dvp;
6515 		}
6516 	}
6517 	/*
6518 	 * If no matches for same file system, total all servers not assigned
6519 	 * to a file system.
6520 	 */
6521 	if (i == 0) {
6522 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6523 			if (ds->nfsdev_nmp != NULL &&
6524 			    ds->nfsdev_mdsisset == 0) {
6525 				if (++i > nfsrv_devidcnt)
6526 					break;
6527 				*tdvpp++ = ds->nfsdev_dvp;
6528 			}
6529 		}
6530 	}
6531 	NFSDDSUNLOCK();
6532 	cnt = i;
6533 
6534 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
6535 	tdvpp = dvpp;
6536 	for (i = 0; i < cnt && error == 0; i++) {
6537 		dvp = *tdvpp++;
6538 		error = VFS_STATFS(dvp->v_mount, tsf);
6539 		if (error == 0) {
6540 			if (sf->f_bsize == 0) {
6541 				if (tsf->f_bsize > 0)
6542 					sf->f_bsize = tsf->f_bsize;
6543 				else
6544 					sf->f_bsize = 8192;
6545 			}
6546 			if (tsf->f_blocks > 0) {
6547 				if (sf->f_bsize != tsf->f_bsize) {
6548 					tot = tsf->f_blocks * tsf->f_bsize;
6549 					sf->f_blocks += (tot / sf->f_bsize);
6550 				} else
6551 					sf->f_blocks += tsf->f_blocks;
6552 			}
6553 			if (tsf->f_bfree > 0) {
6554 				if (sf->f_bsize != tsf->f_bsize) {
6555 					tot = tsf->f_bfree * tsf->f_bsize;
6556 					sf->f_bfree += (tot / sf->f_bsize);
6557 				} else
6558 					sf->f_bfree += tsf->f_bfree;
6559 			}
6560 			if (tsf->f_bavail > 0) {
6561 				if (sf->f_bsize != tsf->f_bsize) {
6562 					tot = tsf->f_bavail * tsf->f_bsize;
6563 					sf->f_bavail += (tot / sf->f_bsize);
6564 				} else
6565 					sf->f_bavail += tsf->f_bavail;
6566 			}
6567 		}
6568 	}
6569 	free(tsf, M_TEMP);
6570 	free(dvpp, M_TEMP);
6571 	return (error);
6572 }
6573 
6574 /*
6575  * Set an NFSv4 acl.
6576  */
6577 int
6578 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6579 {
6580 	int error;
6581 
6582 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6583 		error = NFSERR_ATTRNOTSUPP;
6584 		goto out;
6585 	}
6586 	/*
6587 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6588 	 * Make sure it has enough room for that - splitting every entry
6589 	 * into two and appending "canonical six" entries at the end.
6590 	 * Cribbed out of kern/vfs_acl.c - Rick M.
6591 	 */
6592 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6593 		error = NFSERR_ATTRNOTSUPP;
6594 		goto out;
6595 	}
6596 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6597 	if (error == 0) {
6598 		error = nfsrv_dssetacl(vp, aclp, cred, p);
6599 		if (error == ENOENT)
6600 			error = 0;
6601 	}
6602 
6603 out:
6604 	NFSEXITCODE(error);
6605 	return (error);
6606 }
6607 
6608 /*
6609  * Seek vnode op call (actually it is a VOP_IOCTL()).
6610  * This function is called with the vnode locked, but unlocks and vrele()s
6611  * the vp before returning.
6612  */
6613 int
6614 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6615     off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6616 {
6617 	struct nfsvattr at;
6618 	int error, ret;
6619 
6620 	ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6621 	/*
6622 	 * Attempt to seek on a DS file. A return of ENOENT implies
6623 	 * there is no DS file to seek on.
6624 	 */
6625 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6626 	    NULL, NULL, NULL, NULL, offp, content, eofp);
6627 	if (error != ENOENT) {
6628 		vput(vp);
6629 		return (error);
6630 	}
6631 
6632 	/*
6633 	 * Do the VOP_IOCTL() call.  For the case where *offp == file_size,
6634 	 * VOP_IOCTL() will return ENXIO.  However, the correct reply for
6635 	 * NFSv4.2 is *eofp == true and error == 0 for this case.
6636 	 */
6637 	NFSVOPUNLOCK(vp);
6638 	error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6639 	*eofp = false;
6640 	if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6641 		/* Handle the cases where we might be at EOF. */
6642 		ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6643 		if (ret == 0 && *offp == at.na_size) {
6644 			*eofp = true;
6645 			error = 0;
6646 		}
6647 		if (ret != 0 && error == 0)
6648 			error = ret;
6649 	}
6650 	vrele(vp);
6651 	NFSEXITCODE(error);
6652 	return (error);
6653 }
6654 
6655 /*
6656  * Allocate vnode op call.
6657  */
6658 int
6659 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6660     NFSPROC_T *p)
6661 {
6662 	int error;
6663 	off_t olen;
6664 
6665 	ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6666 	/*
6667 	 * Attempt to allocate on a DS file. A return of ENOENT implies
6668 	 * there is no DS file to allocate on.
6669 	 */
6670 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6671 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6672 	if (error != ENOENT)
6673 		return (error);
6674 
6675 	/*
6676 	 * Do the actual VOP_ALLOCATE(), looping so long as
6677 	 * progress is being made, to achieve completion.
6678 	 */
6679 	do {
6680 		olen = len;
6681 		error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred);
6682 		if (error == 0 && len > 0 && olen > len)
6683 			maybe_yield();
6684 	} while (error == 0 && len > 0 && olen > len);
6685 	if (error == 0 && len > 0)
6686 		error = NFSERR_IO;
6687 	NFSEXITCODE(error);
6688 	return (error);
6689 }
6690 
6691 /*
6692  * Deallocate vnode op call.
6693  */
6694 int
6695 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6696     NFSPROC_T *p)
6697 {
6698 	int error;
6699 	off_t olen;
6700 
6701 	ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp");
6702 	/*
6703 	 * Attempt to deallocate on a DS file. A return of ENOENT implies
6704 	 * there is no DS file to deallocate on.
6705 	 */
6706 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL,
6707 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6708 	if (error != ENOENT)
6709 		return (error);
6710 
6711 	/*
6712 	 * Do the actual VOP_DEALLOCATE(), looping so long as
6713 	 * progress is being made, to achieve completion.
6714 	 */
6715 	do {
6716 		olen = len;
6717 		error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred);
6718 		if (error == 0 && len > 0 && olen > len)
6719 			maybe_yield();
6720 	} while (error == 0 && len > 0 && olen > len);
6721 	if (error == 0 && len > 0)
6722 		error = NFSERR_IO;
6723 	NFSEXITCODE(error);
6724 	return (error);
6725 }
6726 
6727 /*
6728  * Get Extended Atribute vnode op into an mbuf list.
6729  */
6730 int
6731 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6732     struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p,
6733     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
6734 {
6735 	struct iovec *iv;
6736 	struct uio io, *uiop = &io;
6737 	struct mbuf *m, *m2;
6738 	int alen, error, len, tlen;
6739 	size_t siz;
6740 
6741 	/* First, find out the size of the extended attribute. */
6742 	error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6743 	    &siz, cred, p);
6744 	if (error != 0)
6745 		return (NFSERR_NOXATTR);
6746 	if (siz > maxresp - NFS_MAXXDR)
6747 		return (NFSERR_XATTR2BIG);
6748 	len = siz;
6749 	tlen = NFSM_RNDUP(len);
6750 	if (tlen > 0) {
6751 		/*
6752 		 * If cnt > MCLBYTES and the reply will not be saved, use
6753 		 * ext_pgs mbufs for TLS.
6754 		 * For NFSv4.0, we do not know for sure if the reply will
6755 		 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
6756 		 * Always use ext_pgs mbufs if ND_EXTPG is set.
6757 		 */
6758 		if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES &&
6759 		    (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS &&
6760 		    (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4))
6761 			uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen,
6762 			    maxextsiz, &m, &m2, &iv);
6763 		else
6764 			uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2,
6765 			    &iv);
6766 		uiop->uio_iov = iv;
6767 	} else {
6768 		uiop->uio_iovcnt = 0;
6769 		uiop->uio_iov = iv = NULL;
6770 		m = m2 = NULL;
6771 	}
6772 	uiop->uio_offset = 0;
6773 	uiop->uio_resid = tlen;
6774 	uiop->uio_rw = UIO_READ;
6775 	uiop->uio_segflg = UIO_SYSSPACE;
6776 	uiop->uio_td = p;
6777 #ifdef MAC
6778 	error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6779 	    name);
6780 	if (error != 0)
6781 		goto out;
6782 #endif
6783 
6784 	if (tlen > 0)
6785 		error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6786 		    NULL, cred, p);
6787 	if (error != 0)
6788 		goto out;
6789 	if (uiop->uio_resid > 0) {
6790 		alen = tlen;
6791 		len = tlen - uiop->uio_resid;
6792 		tlen = NFSM_RNDUP(len);
6793 		if (alen != tlen)
6794 			printf("nfsvno_getxattr: weird size read\n");
6795 		if (tlen == 0) {
6796 			m_freem(m);
6797 			m = m2 = NULL;
6798 		} else if (alen != tlen || tlen != len)
6799 			m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6800 	}
6801 	*lenp = len;
6802 	*mpp = m;
6803 	*mpendp = m2;
6804 
6805 out:
6806 	if (error != 0) {
6807 		if (m != NULL)
6808 			m_freem(m);
6809 		*lenp = 0;
6810 	}
6811 	free(iv, M_TEMP);
6812 	NFSEXITCODE(error);
6813 	return (error);
6814 }
6815 
6816 /*
6817  * Set Extended attribute vnode op from an mbuf list.
6818  */
6819 int
6820 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6821     char *cp, struct ucred *cred, struct thread *p)
6822 {
6823 	struct iovec *iv;
6824 	struct uio uio, *uiop = &uio;
6825 	int cnt, error;
6826 
6827 	error = 0;
6828 #ifdef MAC
6829 	error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6830 	    name);
6831 #endif
6832 	if (error != 0)
6833 		goto out;
6834 
6835 	uiop->uio_rw = UIO_WRITE;
6836 	uiop->uio_segflg = UIO_SYSSPACE;
6837 	uiop->uio_td = p;
6838 	uiop->uio_offset = 0;
6839 	uiop->uio_resid = len;
6840 	if (len > 0) {
6841 		error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6842 		uiop->uio_iov = iv;
6843 		uiop->uio_iovcnt = cnt;
6844 	} else {
6845 		uiop->uio_iov = iv = NULL;
6846 		uiop->uio_iovcnt = 0;
6847 	}
6848 	if (error == 0) {
6849 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6850 		    cred, p);
6851 		if (error == 0) {
6852 			if (vp->v_type == VREG && nfsrv_devidcnt != 0)
6853 				nfsvno_updateds(vp, cred, p);
6854 			error = VOP_FSYNC(vp, MNT_WAIT, p);
6855 		}
6856 		free(iv, M_TEMP);
6857 	}
6858 
6859 out:
6860 	NFSEXITCODE(error);
6861 	return (error);
6862 }
6863 
6864 /*
6865  * For a pNFS server, the DS file's ctime and
6866  * va_filerev (TimeMetadata and Change) needs to
6867  * be updated.  This is a hack, but works by
6868  * flipping the S_ISGID bit in va_mode and then
6869  * flipping it back.
6870  * It does result in two MDS->DS RPCs, but creating
6871  * a custom RPC just to do this seems overkill, since
6872  * Setxattr/Rmxattr will not be done that frequently.
6873  * If it fails part way through, that is not too
6874  * serious, since the DS file is never executed.
6875  */
6876 static void
6877 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p)
6878 {
6879 	struct nfsvattr nva;
6880 	int ret;
6881 	u_short tmode;
6882 
6883 	ret = VOP_GETATTR(vp, &nva.na_vattr, cred);
6884 	if (ret == 0) {
6885 		tmode = nva.na_mode;
6886 		NFSVNO_ATTRINIT(&nva);
6887 		tmode ^= S_ISGID;
6888 		NFSVNO_SETATTRVAL(&nva, mode, tmode);
6889 		ret = nfsrv_proxyds(vp, 0, 0, cred, p,
6890 		    NFSPROC_SETATTR, NULL, NULL, NULL, &nva,
6891 		    NULL, NULL, 0, NULL);
6892 		if (ret == 0) {
6893 			tmode ^= S_ISGID;
6894 			NFSVNO_SETATTRVAL(&nva, mode, tmode);
6895 			ret = nfsrv_proxyds(vp, 0, 0, cred, p,
6896 			    NFSPROC_SETATTR, NULL, NULL, NULL,
6897 			    &nva, NULL, NULL, 0, NULL);
6898 		}
6899 	}
6900 }
6901 
6902 /*
6903  * Remove Extended attribute vnode op.
6904  */
6905 int
6906 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6907     struct ucred *cred, struct thread *p)
6908 {
6909 	int error;
6910 
6911 	/*
6912 	 * Get rid of any delegations.  I am not sure why this is required,
6913 	 * but RFC-8276 says so.
6914 	 */
6915 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6916 	if (error != 0)
6917 		goto out;
6918 #ifdef MAC
6919 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6920 	    name);
6921 	if (error != 0)
6922 		goto out;
6923 #endif
6924 
6925 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6926 	if (error == EOPNOTSUPP)
6927 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6928 		    cred, p);
6929 	if (error == 0) {
6930 		if (vp->v_type == VREG && nfsrv_devidcnt != 0)
6931 			nfsvno_updateds(vp, cred, p);
6932 		error = VOP_FSYNC(vp, MNT_WAIT, p);
6933 	}
6934 out:
6935 	NFSEXITCODE(error);
6936 	return (error);
6937 }
6938 
6939 /*
6940  * List Extended Atribute vnode op into an mbuf list.
6941  */
6942 int
6943 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6944     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6945 {
6946 	struct iovec iv;
6947 	struct uio io;
6948 	int error;
6949 	size_t siz;
6950 
6951 	*bufp = NULL;
6952 	/* First, find out the size of the extended attribute. */
6953 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6954 	    p);
6955 	if (error != 0)
6956 		return (NFSERR_NOXATTR);
6957 	if (siz <= cookie) {
6958 		*lenp = 0;
6959 		*eofp = true;
6960 		goto out;
6961 	}
6962 	if (siz > cookie + *lenp) {
6963 		siz = cookie + *lenp;
6964 		*eofp = false;
6965 	} else
6966 		*eofp = true;
6967 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6968 	if (siz > 10 * 1024 * 1024) {
6969 		error = NFSERR_XATTR2BIG;
6970 		goto out;
6971 	}
6972 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
6973 	iv.iov_base = *bufp;
6974 	iv.iov_len = siz;
6975 	io.uio_iovcnt = 1;
6976 	io.uio_iov = &iv;
6977 	io.uio_offset = 0;
6978 	io.uio_resid = siz;
6979 	io.uio_rw = UIO_READ;
6980 	io.uio_segflg = UIO_SYSSPACE;
6981 	io.uio_td = p;
6982 #ifdef MAC
6983 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6984 	if (error != 0)
6985 		goto out;
6986 #endif
6987 
6988 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6989 	    p);
6990 	if (error != 0)
6991 		goto out;
6992 	if (io.uio_resid > 0)
6993 		siz -= io.uio_resid;
6994 	*lenp = siz;
6995 
6996 out:
6997 	if (error != 0) {
6998 		free(*bufp, M_TEMP);
6999 		*bufp = NULL;
7000 	}
7001 	NFSEXITCODE(error);
7002 	return (error);
7003 }
7004 
7005 /*
7006  * Trim trailing data off the mbuf list being built.
7007  */
7008 void
7009 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
7010     int bextpg, int bextpgsiz)
7011 {
7012 	vm_page_t pg;
7013 	int fullpgsiz, i;
7014 
7015 	if (mb->m_next != NULL) {
7016 		m_freem(mb->m_next);
7017 		mb->m_next = NULL;
7018 	}
7019 	if ((mb->m_flags & M_EXTPG) != 0) {
7020 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
7021 		    ("nfsm_trimtrailing: bextpg out of range"));
7022 		KASSERT(bpos == (char *)(void *)
7023 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
7024 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
7025 
7026 		/* First, get rid of any pages after this position. */
7027 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
7028 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
7029 			vm_page_unwire_noq(pg);
7030 			vm_page_free(pg);
7031 		}
7032 		mb->m_epg_npgs = bextpg + 1;
7033 		if (bextpg == 0)
7034 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
7035 		else
7036 			fullpgsiz = PAGE_SIZE;
7037 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
7038 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
7039 		for (i = 1; i < mb->m_epg_npgs; i++)
7040 			mb->m_len += m_epg_pagelen(mb, i, 0);
7041 		nd->nd_bextpgsiz = bextpgsiz;
7042 		nd->nd_bextpg = bextpg;
7043 	} else
7044 		mb->m_len = bpos - mtod(mb, char *);
7045 	nd->nd_mb = mb;
7046 	nd->nd_bpos = bpos;
7047 }
7048 
7049 
7050 /*
7051  * Check to see if a put file handle operation should test for
7052  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
7053  * When Open is the next operation, NFSERR_WRONGSEC cannot be
7054  * replied for the Open cases that use a component.  This can
7055  * be identified by the fact that the file handle's type is VDIR.
7056  */
7057 bool
7058 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, __enum_uint8(vtype) vtyp)
7059 {
7060 
7061 	if ((nd->nd_flag & ND_NFSV4) == 0)
7062 		return (true);
7063 
7064 	if ((nd->nd_flag & ND_LASTOP) != 0)
7065 		return (false);
7066 
7067 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
7068 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
7069 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
7070 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
7071 		return (false);
7072 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
7073 		return (false);
7074 	return (true);
7075 }
7076 
7077 /*
7078  * Check DSs marked no space.
7079  */
7080 void
7081 nfsrv_checknospc(void)
7082 {
7083 	struct statfs *tsf;
7084 	struct nfsdevice *ds;
7085 	struct vnode **dvpp, **tdvpp, *dvp;
7086 	char *devid, *tdevid;
7087 	int cnt, error = 0, i;
7088 
7089 	if (nfsrv_devidcnt <= 0)
7090 		return;
7091 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
7092 	devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK);
7093 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
7094 
7095 	/* Get an array of the dvps for the DSs. */
7096 	tdvpp = dvpp;
7097 	tdevid = devid;
7098 	i = 0;
7099 	NFSDDSLOCK();
7100 	/* First, search for matches for same file system. */
7101 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7102 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) {
7103 			if (++i > nfsrv_devidcnt)
7104 				break;
7105 			*tdvpp++ = ds->nfsdev_dvp;
7106 			NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID);
7107 			tdevid += NFSX_V4DEVICEID;
7108 		}
7109 	}
7110 	NFSDDSUNLOCK();
7111 
7112 	/* Do a VFS_STATFS() for each of the DSs and clear no space. */
7113 	cnt = i;
7114 	tdvpp = dvpp;
7115 	tdevid = devid;
7116 	for (i = 0; i < cnt && error == 0; i++) {
7117 		dvp = *tdvpp++;
7118 		error = VFS_STATFS(dvp->v_mount, tsf);
7119 		if (error == 0 && tsf->f_bavail > 0) {
7120 			NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n");
7121 			nfsrv_marknospc(tdevid, false);
7122 		}
7123 		tdevid += NFSX_V4DEVICEID;
7124 	}
7125 	free(tsf, M_TEMP);
7126 	free(dvpp, M_TEMP);
7127 	free(devid, M_TEMP);
7128 }
7129 
7130 /*
7131  * Initialize everything that needs to be initialized for a vnet.
7132  */
7133 static void
7134 nfsrv_vnetinit(const void *unused __unused)
7135 {
7136 
7137 	nfsd_mntinit();
7138 }
7139 VNET_SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY,
7140     nfsrv_vnetinit, NULL);
7141 
7142 /*
7143  * Clean up everything that is in a vnet and needs to be
7144  * done when the jail is destroyed or the module unloaded.
7145  */
7146 static void
7147 nfsrv_cleanup(const void *unused __unused)
7148 {
7149 	int i;
7150 
7151 	NFSD_LOCK();
7152 	if (!NFSD_VNET(nfsrv_mntinited)) {
7153 		NFSD_UNLOCK();
7154 		return;
7155 	}
7156 	NFSD_VNET(nfsrv_mntinited) = false;
7157 	NFSD_UNLOCK();
7158 
7159 	/* Clean out all NFSv4 state. */
7160 	nfsrv_throwawayallstate(curthread);
7161 
7162 	/* Clean the NFS server reply cache */
7163 	nfsrvd_cleancache();
7164 
7165 	/* Clean out v4root exports. */
7166 	if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) {
7167 		vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export);
7168 		free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT);
7169 		NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL;
7170 	}
7171 
7172 	/* Free up the krpc server pool. */
7173 	if (NFSD_VNET(nfsrvd_pool) != NULL)
7174 		svcpool_destroy(NFSD_VNET(nfsrvd_pool));
7175 
7176 	/* and get rid of the locks */
7177 	for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7178 		mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx);
7179 		mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx);
7180 	}
7181 	mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx);
7182 	for (i = 0; i < nfsrv_sessionhashsize; i++)
7183 		mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx);
7184 	lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock);
7185 	free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE);
7186 	free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE);
7187 	free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE);
7188 	free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT);
7189 	free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE);
7190 	free(NFSD_VNET(nfssessionhash), M_NFSDSESSION);
7191 	free(NFSD_VNET(nfsv4root_mnt), M_TEMP);
7192 	NFSD_VNET(nfsv4root_mnt) = NULL;
7193 }
7194 VNET_SYSUNINIT(nfsrv_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY,
7195     nfsrv_cleanup, NULL);
7196 
7197 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
7198 
7199 /*
7200  * Called once to initialize data structures...
7201  */
7202 static int
7203 nfsd_modevent(module_t mod, int type, void *data)
7204 {
7205 	int error = 0, i;
7206 	static int loaded = 0;
7207 
7208 	switch (type) {
7209 	case MOD_LOAD:
7210 		if (loaded)
7211 			goto out;
7212 		newnfs_portinit();
7213 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
7214 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
7215 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
7216 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
7217 #ifdef VV_DISABLEDELEG
7218 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
7219 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
7220 #endif
7221 		nfsd_call_nfsd = nfssvc_nfsd;
7222 		loaded = 1;
7223 		break;
7224 
7225 	case MOD_UNLOAD:
7226 		if (newnfs_numnfsd != 0) {
7227 			error = EBUSY;
7228 			break;
7229 		}
7230 
7231 #ifdef VV_DISABLEDELEG
7232 		vn_deleg_ops.vndeleg_recall = NULL;
7233 		vn_deleg_ops.vndeleg_disable = NULL;
7234 #endif
7235 		nfsd_call_nfsd = NULL;
7236 		mtx_destroy(&nfsrc_udpmtx);
7237 		mtx_destroy(&nfs_v4root_mutex);
7238 		mtx_destroy(&nfsrv_dontlistlock_mtx);
7239 		mtx_destroy(&nfsrv_recalllock_mtx);
7240 		if (nfslayouthash != NULL) {
7241 			for (i = 0; i < nfsrv_layouthashsize; i++)
7242 				mtx_destroy(&nfslayouthash[i].mtx);
7243 			free(nfslayouthash, M_NFSDSESSION);
7244 		}
7245 		loaded = 0;
7246 		break;
7247 	default:
7248 		error = EOPNOTSUPP;
7249 		break;
7250 	}
7251 
7252 out:
7253 	NFSEXITCODE(error);
7254 	return (error);
7255 }
7256 static moduledata_t nfsd_mod = {
7257 	"nfsd",
7258 	nfsd_modevent,
7259 	NULL,
7260 };
7261 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
7262 
7263 /* So that loader and kldload(2) can find us, wherever we are.. */
7264 MODULE_VERSION(nfsd, 1);
7265 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
7266 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
7267 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
7268 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
7269