xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision c27f7d6b9cf6d4ab01cb3d0972726c14e0aca146)
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 #include <vm/vm_param.h>
55 #include <vm/vnode_pager.h>
56 
57 FEATURE(nfsd, "NFSv4 server");
58 
59 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
60 extern int nfsrv_useacl;
61 extern int newnfs_numnfsd;
62 extern int nfsrv_sessionhashsize;
63 extern struct nfslayouthash *nfslayouthash;
64 extern int nfsrv_layouthashsize;
65 extern struct mtx nfsrv_dslock_mtx;
66 extern int nfs_pnfsiothreads;
67 extern volatile int nfsrv_devidcnt;
68 extern int nfsrv_maxpnfsmirror;
69 extern uint32_t nfs_srvmaxio;
70 extern int nfs_bufpackets;
71 extern u_long sb_max_adj;
72 extern struct nfsv4lock nfsv4rootfs_lock;
73 
74 NFSD_VNET_DECLARE(int, nfsrv_numnfsd);
75 NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst);
76 NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool);
77 NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash);
78 NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash);
79 NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash);
80 NFSD_VNET_DECLARE(struct nfsv4lock, nfsd_suspend_lock);
81 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
82 
83 NFSDLOCKMUTEX;
84 NFSSTATESPINLOCK;
85 struct mtx nfsrc_udpmtx;
86 struct mtx nfs_v4root_mutex;
87 struct mtx nfsrv_dontlistlock_mtx;
88 struct mtx nfsrv_recalllock_mtx;
89 struct nfsrvfh nfs_pubfh;
90 int nfs_pubfhset = 0;
91 int nfsd_debuglevel = 0;
92 static pid_t nfsd_master_pid = (pid_t)-1;
93 static char nfsd_master_comm[MAXCOMLEN + 1];
94 static struct timeval nfsd_master_start;
95 static uint32_t nfsv4_sysid = 0;
96 static fhandle_t zerofh;
97 
98 NFSD_VNET_DEFINE(struct proc *, nfsd_master_proc) = NULL;
99 NFSD_VNET_DEFINE(struct nfsrvhashhead *, nfsrvudphashtbl);
100 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrchash_table);
101 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrcahash_table);
102 NFSD_VNET_DEFINE(struct nfsrvfh, nfs_rootfh);
103 NFSD_VNET_DEFINE(int, nfs_rootfhset) = 0;
104 NFSD_VNET_DEFINE(struct callout, nfsd_callout);
105 NFSD_VNET_DEFINE_STATIC(struct mount *, nfsv4root_mnt);
106 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_opt);
107 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_newopt);
108 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_suspend_nfsd) = false;
109 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_mntinited) = false;
110 
111 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
112     struct ucred *);
113 static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *);
114 
115 int nfsrv_enable_crossmntpt = 1;
116 static int nfs_commit_blks;
117 static int nfs_commit_miss;
118 extern int nfsrv_issuedelegs;
119 extern int nfsrv_dolocallocks;
120 extern struct nfsdevicehead nfsrv_devidhead;
121 
122 /* Map d_type to vnode type. */
123 static uint8_t dtype_to_vnode[DT_WHT + 1] = { VNON, VFIFO, VCHR, VNON, VDIR,
124     VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON };
125 
126 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
127     struct iovec **);
128 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
129     struct mbuf **, struct iovec **);
130 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
131     int *);
132 static void nfs_dtypetovtype(struct nfsvattr *, struct vnode *, uint8_t);
133 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
134     NFSPROC_T *);
135 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
136     int *, char *, fhandle_t *);
137 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
138     NFSPROC_T *);
139 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
140     struct thread *, int, struct mbuf **, char *, struct mbuf **,
141     struct nfsvattr *, struct acl *, off_t *, int, bool *);
142 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
143 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
144     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
145 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
146     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
147     char *, int *);
148 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
149     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
150 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
151     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
152 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
153     struct vnode *, struct nfsmount **, int, struct acl *, int *);
154 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
155     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
156 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
157     struct vnode *, struct nfsmount *, struct nfsvattr *);
158 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
159     NFSPROC_T *, struct nfsmount *);
160 static int nfsrv_putfhname(fhandle_t *, char *);
161 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
162     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
163 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
164     struct vnode *, NFSPROC_T *);
165 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
166 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
167     NFSPROC_T *);
168 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
169 
170 int nfs_pnfsio(task_fn_t *, void *);
171 
172 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
173     "NFS server");
174 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
175     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
176 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
177     0, "");
178 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
179     0, "");
180 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
181     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
182 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
183     0, "Debug level for NFS server");
184 NFSD_VNET_DECLARE(int, nfsd_enable_stringtouid);
185 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid,
186     CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_enable_stringtouid),
187     0, "Enable nfsd to accept numeric owner_names");
188 static int nfsrv_pnfsgetdsattr = 1;
189 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
190     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
191 
192 /*
193  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
194  * not running.
195  * The dsN subdirectories for the increased values must have been created
196  * on all DS servers before this increase is done.
197  */
198 u_int	nfsrv_dsdirsize = 20;
199 static int
200 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
201 {
202 	int error, newdsdirsize;
203 
204 	newdsdirsize = nfsrv_dsdirsize;
205 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
206 	if (error != 0 || req->newptr == NULL)
207 		return (error);
208 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
209 	    newnfs_numnfsd != 0)
210 		return (EINVAL);
211 	nfsrv_dsdirsize = newdsdirsize;
212 	return (0);
213 }
214 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
215     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
216     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
217 
218 /*
219  * nfs_srvmaxio can only be increased and only when the nfsd threads are
220  * not running.  The setting must be a power of 2, with the current limit of
221  * 1Mbyte.
222  */
223 static int
224 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)
225 {
226 	int error;
227 	u_int newsrvmaxio;
228 	uint64_t tval;
229 
230 	newsrvmaxio = nfs_srvmaxio;
231 	error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req);
232 	if (error != 0 || req->newptr == NULL)
233 		return (error);
234 	if (newsrvmaxio == nfs_srvmaxio)
235 		return (0);
236 	if (newsrvmaxio < nfs_srvmaxio) {
237 		printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n");
238 		return (EINVAL);
239 	}
240 	if (newsrvmaxio > 1048576) {
241 		printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n");
242 		return (EINVAL);
243 	}
244 	if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) {
245 		printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n");
246 		return (EINVAL);
247 	}
248 
249 	/*
250 	 * Check that kern.ipc.maxsockbuf is large enough for
251 	 * newsrviomax, given the setting of vfs.nfs.bufpackets.
252 	 */
253 	if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets >
254 	    sb_max_adj) {
255 		/*
256 		 * Suggest vfs.nfs.bufpackets * maximum RPC message for
257 		 * sb_max_adj.
258 		 */
259 		tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets;
260 
261 		/*
262 		 * Convert suggested sb_max_adj value to a suggested
263 		 * sb_max value, which is what is set via kern.ipc.maxsockbuf.
264 		 * Perform the inverse calculation of (from uipc_sockbuf.c):
265 		 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES /
266 		 *     (MSIZE + MCLBYTES);
267 		 * XXX If the calculation of sb_max_adj from sb_max changes,
268 		 *     this calculation must be changed as well.
269 		 */
270 		tval *= (MSIZE + MCLBYTES);  /* Brackets for readability. */
271 		tval += MCLBYTES - 1;        /* Round up divide. */
272 		tval /= MCLBYTES;
273 		printf("nfsd: set kern.ipc.maxsockbuf to a minimum of "
274 		    "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval,
275 		    newsrvmaxio);
276 		return (EINVAL);
277 	}
278 
279 	NFSD_LOCK();
280 	if (newnfs_numnfsd != 0) {
281 		NFSD_UNLOCK();
282 		printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd "
283 		    "threads are running\n");
284 		return (EINVAL);
285 	}
286 
287 
288 	nfs_srvmaxio = newsrvmaxio;
289 	NFSD_UNLOCK();
290 	return (0);
291 }
292 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio,
293     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
294     sysctl_srvmaxio, "IU", "Maximum I/O size in bytes");
295 
296 static int
297 sysctl_dolocallocks(SYSCTL_HANDLER_ARGS)
298 {
299 	int error, igotlock, newdolocallocks;
300 
301 	newdolocallocks = nfsrv_dolocallocks;
302 	error = sysctl_handle_int(oidp, &newdolocallocks, 0, req);
303 	if (error != 0 || req->newptr == NULL)
304 		return (error);
305 	if (newdolocallocks == nfsrv_dolocallocks)
306 		return (0);
307 	if (jailed(curthread->td_ucred))
308 		return (EINVAL);
309 
310 	NFSLOCKV4ROOTMUTEX();
311 	do {
312 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
313 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
314 	} while (!igotlock);
315 	NFSUNLOCKV4ROOTMUTEX();
316 
317 	nfsrv_dolocallocks = newdolocallocks;
318 
319 	NFSLOCKV4ROOTMUTEX();
320 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
321 	NFSUNLOCKV4ROOTMUTEX();
322 	return (0);
323 }
324 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, enable_locallocks,
325     CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
326     sysctl_dolocallocks, "IU", "Enable nfsd to acquire local locks on files");
327 
328 #define	MAX_REORDERED_RPC	16
329 #define	NUM_HEURISTIC		1031
330 #define	NHUSE_INIT		64
331 #define	NHUSE_INC		16
332 #define	NHUSE_MAX		2048
333 
334 static struct nfsheur {
335 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
336 	off_t nh_nextoff;	/* next offset for sequential detection */
337 	int nh_use;		/* use count for selection */
338 	int nh_seqcount;	/* heuristic */
339 } nfsheur[NUM_HEURISTIC];
340 
341 /*
342  * Heuristic to detect sequential operation.
343  */
344 static struct nfsheur *
345 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
346 {
347 	struct nfsheur *nh;
348 	int hi, try;
349 
350 	/* Locate best candidate. */
351 	try = 32;
352 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
353 	nh = &nfsheur[hi];
354 	while (try--) {
355 		if (nfsheur[hi].nh_vp == vp) {
356 			nh = &nfsheur[hi];
357 			break;
358 		}
359 		if (nfsheur[hi].nh_use > 0)
360 			--nfsheur[hi].nh_use;
361 		hi = (hi + 1) % NUM_HEURISTIC;
362 		if (nfsheur[hi].nh_use < nh->nh_use)
363 			nh = &nfsheur[hi];
364 	}
365 
366 	/* Initialize hint if this is a new file. */
367 	if (nh->nh_vp != vp) {
368 		nh->nh_vp = vp;
369 		nh->nh_nextoff = uio->uio_offset;
370 		nh->nh_use = NHUSE_INIT;
371 		if (uio->uio_offset == 0)
372 			nh->nh_seqcount = 4;
373 		else
374 			nh->nh_seqcount = 1;
375 	}
376 
377 	/* Calculate heuristic. */
378 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
379 	    uio->uio_offset == nh->nh_nextoff) {
380 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
381 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
382 		if (nh->nh_seqcount > IO_SEQMAX)
383 			nh->nh_seqcount = IO_SEQMAX;
384 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
385 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
386 		/* Probably a reordered RPC, leave seqcount alone. */
387 	} else if (nh->nh_seqcount > 1) {
388 		nh->nh_seqcount /= 2;
389 	} else {
390 		nh->nh_seqcount = 0;
391 	}
392 	nh->nh_use += NHUSE_INC;
393 	if (nh->nh_use > NHUSE_MAX)
394 		nh->nh_use = NHUSE_MAX;
395 	return (nh);
396 }
397 
398 /*
399  * Get attributes into nfsvattr structure.
400  */
401 int
402 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
403     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
404     nfsattrbit_t *attrbitp)
405 {
406 	int error, gotattr, lockedit = 0;
407 	struct nfsvattr na;
408 
409 	if (vpislocked == 0) {
410 		/*
411 		 * When vpislocked == 0, the vnode is either exclusively
412 		 * locked by this thread or not locked by this thread.
413 		 * As such, shared lock it, if not exclusively locked.
414 		 */
415 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
416 			lockedit = 1;
417 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
418 		}
419 	}
420 
421 	/*
422 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
423 	 * attributes, as required.
424 	 * This needs to be done for regular files if:
425 	 * - non-NFSv4 RPCs or
426 	 * - when attrbitp == NULL or
427 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
428 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
429 	 * these attributes.  nfsrv_proxyds() will return an error if the
430 	 * server is not a pNFS one.
431 	 */
432 	gotattr = 0;
433 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
434 	    (nd->nd_flag & ND_NFSV4) == 0 ||
435 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
436 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
437 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
438 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
439 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
440 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
441 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
442 		    NULL);
443 		if (error == 0)
444 			gotattr = 1;
445 	}
446 
447 	nvap->na_bsdflags = 0;
448 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
449 	if (lockedit != 0)
450 		NFSVOPUNLOCK(vp);
451 
452 	/*
453 	 * If we got the Change, Size and Modify Time from the DS,
454 	 * replace them.
455 	 */
456 	if (gotattr != 0) {
457 		nvap->na_atime = na.na_atime;
458 		nvap->na_mtime = na.na_mtime;
459 		nvap->na_filerev = na.na_filerev;
460 		nvap->na_size = na.na_size;
461 		nvap->na_bytes = na.na_bytes;
462 	}
463 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
464 	    error, (uintmax_t)na.na_filerev);
465 
466 	NFSEXITCODE(error);
467 	return (error);
468 }
469 
470 /*
471  * Get a file handle for a vnode.
472  */
473 int
474 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
475 {
476 	int error;
477 
478 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
479 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
480 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
481 
482 	NFSEXITCODE(error);
483 	return (error);
484 }
485 
486 /*
487  * Perform access checking for vnodes obtained from file handles that would
488  * refer to files already opened by a Unix client. You cannot just use
489  * vn_writechk() and VOP_ACCESSX() for two reasons.
490  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
491  *     case.
492  * 2 - The owner is to be given access irrespective of mode bits for some
493  *     operations, so that processes that chmod after opening a file don't
494  *     break.
495  */
496 int
497 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
498     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
499     u_int32_t *supportedtypep)
500 {
501 	struct vattr vattr;
502 	int error = 0, getret = 0;
503 
504 	if (vpislocked == 0) {
505 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
506 			error = EPERM;
507 			goto out;
508 		}
509 	}
510 	if (accmode & VWRITE) {
511 		/* Just vn_writechk() changed to check rdonly */
512 		/*
513 		 * Disallow write attempts on read-only file systems;
514 		 * unless the file is a socket or a block or character
515 		 * device resident on the file system.
516 		 */
517 		if (NFSVNO_EXRDONLY(exp) ||
518 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
519 			switch (vp->v_type) {
520 			case VREG:
521 			case VDIR:
522 			case VLNK:
523 				error = EROFS;
524 			default:
525 				break;
526 			}
527 		}
528 		/*
529 		 * If there's shared text associated with
530 		 * the inode, try to free it up once.  If
531 		 * we fail, we can't allow writing.
532 		 */
533 		if (VOP_IS_TEXT(vp) && error == 0)
534 			error = ETXTBSY;
535 	}
536 	if (error != 0) {
537 		if (vpislocked == 0)
538 			NFSVOPUNLOCK(vp);
539 		goto out;
540 	}
541 
542 	/*
543 	 * Should the override still be applied when ACLs are enabled?
544 	 */
545 	error = VOP_ACCESSX(vp, accmode, cred, p);
546 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
547 		/*
548 		 * Try again with VEXPLICIT_DENY, to see if the test for
549 		 * deletion is supported.
550 		 */
551 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
552 		if (error == 0) {
553 			if (vp->v_type == VDIR) {
554 				accmode &= ~(VDELETE | VDELETE_CHILD);
555 				accmode |= VWRITE;
556 				error = VOP_ACCESSX(vp, accmode, cred, p);
557 			} else if (supportedtypep != NULL) {
558 				*supportedtypep &= ~NFSACCESS_DELETE;
559 			}
560 		}
561 	}
562 
563 	/*
564 	 * Allow certain operations for the owner (reads and writes
565 	 * on files that are already open).
566 	 */
567 	if (override != NFSACCCHK_NOOVERRIDE &&
568 	    (error == EPERM || error == EACCES)) {
569 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
570 			error = 0;
571 		else if (override & NFSACCCHK_ALLOWOWNER) {
572 			getret = VOP_GETATTR(vp, &vattr, cred);
573 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
574 				error = 0;
575 		}
576 	}
577 	if (vpislocked == 0)
578 		NFSVOPUNLOCK(vp);
579 
580 out:
581 	NFSEXITCODE(error);
582 	return (error);
583 }
584 
585 /*
586  * Set attribute(s) vnop.
587  */
588 int
589 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
590     struct thread *p, struct nfsexstuff *exp)
591 {
592 	u_quad_t savsize = 0;
593 	int error, savedit;
594 	time_t savbtime;
595 
596 	/*
597 	 * If this is an exported file system and a pNFS service is running,
598 	 * don't VOP_SETATTR() of size for the MDS file system.
599 	 */
600 	savedit = 0;
601 	error = 0;
602 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
603 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
604 	    nvap->na_vattr.va_size > 0) {
605 		savsize = nvap->na_vattr.va_size;
606 		nvap->na_vattr.va_size = VNOVAL;
607 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
608 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
609 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
610 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
611 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
612 			savedit = 1;
613 		else
614 			savedit = 2;
615 	}
616 	if (savedit != 2)
617 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
618 	if (savedit != 0)
619 		nvap->na_vattr.va_size = savsize;
620 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
621 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
622 	    nvap->na_vattr.va_size != VNOVAL ||
623 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
624 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
625 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
626 		/* Never modify birthtime on a DS file. */
627 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
628 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
629 		/* For a pNFS server, set the attributes on the DS file. */
630 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
631 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
632 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
633 		if (error == ENOENT)
634 			error = 0;
635 	}
636 	NFSEXITCODE(error);
637 	return (error);
638 }
639 
640 /*
641  * Set up nameidata for a lookup() call and do it.
642  */
643 int
644 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
645     struct vnode *dp, int islocked, struct nfsexstuff *exp,
646     struct vnode **retdirp)
647 {
648 	struct componentname *cnp = &ndp->ni_cnd;
649 	int i;
650 	struct iovec aiov;
651 	struct uio auio;
652 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
653 	int error = 0;
654 	char *cp;
655 
656 	*retdirp = NULL;
657 	cnp->cn_nameptr = cnp->cn_pnbuf;
658 	ndp->ni_lcf = 0;
659 	/*
660 	 * Extract and set starting directory.
661 	 */
662 	if (dp->v_type != VDIR) {
663 		if (islocked)
664 			vput(dp);
665 		else
666 			vrele(dp);
667 		nfsvno_relpathbuf(ndp);
668 		error = ENOTDIR;
669 		goto out1;
670 	}
671 	if (islocked)
672 		NFSVOPUNLOCK(dp);
673 	VREF(dp);
674 	*retdirp = dp;
675 	if (NFSVNO_EXRDONLY(exp))
676 		cnp->cn_flags |= RDONLY;
677 	ndp->ni_segflg = UIO_SYSSPACE;
678 
679 	if (nd->nd_flag & ND_PUBLOOKUP) {
680 		ndp->ni_loopcnt = 0;
681 		if (cnp->cn_pnbuf[0] == '/') {
682 			vrele(dp);
683 			/*
684 			 * Check for degenerate pathnames here, since lookup()
685 			 * panics on them.
686 			 */
687 			for (i = 1; i < ndp->ni_pathlen; i++)
688 				if (cnp->cn_pnbuf[i] != '/')
689 					break;
690 			if (i == ndp->ni_pathlen) {
691 				error = NFSERR_ACCES;
692 				goto out;
693 			}
694 			dp = rootvnode;
695 			VREF(dp);
696 		}
697 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
698 	    (nd->nd_flag & ND_NFSV4) == 0) {
699 		/*
700 		 * Only cross mount points for NFSv4 when doing a
701 		 * mount while traversing the file system above
702 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
703 		 */
704 		cnp->cn_flags |= NOCROSSMOUNT;
705 	}
706 
707 	/*
708 	 * Initialize for scan, set ni_startdir and bump ref on dp again
709 	 * because lookup() will dereference ni_startdir.
710 	 */
711 
712 	ndp->ni_startdir = dp;
713 	ndp->ni_rootdir = rootvnode;
714 	ndp->ni_topdir = NULL;
715 
716 	if (!lockleaf)
717 		cnp->cn_flags |= LOCKLEAF;
718 	for (;;) {
719 		cnp->cn_nameptr = cnp->cn_pnbuf;
720 		/*
721 		 * Call lookup() to do the real work.  If an error occurs,
722 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
723 		 * we do not have to dereference anything before returning.
724 		 * In either case ni_startdir will be dereferenced and NULLed
725 		 * out.
726 		 */
727 		error = vfs_lookup(ndp);
728 		if (error)
729 			break;
730 
731 		/*
732 		 * Check for encountering a symbolic link.  Trivial
733 		 * termination occurs if no symlink encountered.
734 		 */
735 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
736 			if (ndp->ni_vp && !lockleaf)
737 				NFSVOPUNLOCK(ndp->ni_vp);
738 			break;
739 		}
740 
741 		/*
742 		 * Validate symlink
743 		 */
744 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
745 			NFSVOPUNLOCK(ndp->ni_dvp);
746 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
747 			error = EINVAL;
748 			goto badlink2;
749 		}
750 
751 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
752 			error = ELOOP;
753 			goto badlink2;
754 		}
755 		if (ndp->ni_pathlen > 1)
756 			cp = uma_zalloc(namei_zone, M_WAITOK);
757 		else
758 			cp = cnp->cn_pnbuf;
759 		aiov.iov_base = cp;
760 		aiov.iov_len = MAXPATHLEN;
761 		auio.uio_iov = &aiov;
762 		auio.uio_iovcnt = 1;
763 		auio.uio_offset = 0;
764 		auio.uio_rw = UIO_READ;
765 		auio.uio_segflg = UIO_SYSSPACE;
766 		auio.uio_td = NULL;
767 		auio.uio_resid = MAXPATHLEN;
768 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
769 		if (error) {
770 		badlink1:
771 			if (ndp->ni_pathlen > 1)
772 				uma_zfree(namei_zone, cp);
773 		badlink2:
774 			vrele(ndp->ni_dvp);
775 			vput(ndp->ni_vp);
776 			break;
777 		}
778 		linklen = MAXPATHLEN - auio.uio_resid;
779 		if (linklen == 0) {
780 			error = ENOENT;
781 			goto badlink1;
782 		}
783 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
784 			error = ENAMETOOLONG;
785 			goto badlink1;
786 		}
787 
788 		/*
789 		 * Adjust or replace path
790 		 */
791 		if (ndp->ni_pathlen > 1) {
792 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
793 			uma_zfree(namei_zone, cnp->cn_pnbuf);
794 			cnp->cn_pnbuf = cp;
795 		} else
796 			cnp->cn_pnbuf[linklen] = '\0';
797 		ndp->ni_pathlen += linklen;
798 
799 		/*
800 		 * Cleanup refs for next loop and check if root directory
801 		 * should replace current directory.  Normally ni_dvp
802 		 * becomes the new base directory and is cleaned up when
803 		 * we loop.  Explicitly null pointers after invalidation
804 		 * to clarify operation.
805 		 */
806 		vput(ndp->ni_vp);
807 		ndp->ni_vp = NULL;
808 
809 		if (cnp->cn_pnbuf[0] == '/') {
810 			vrele(ndp->ni_dvp);
811 			ndp->ni_dvp = ndp->ni_rootdir;
812 			VREF(ndp->ni_dvp);
813 		}
814 		ndp->ni_startdir = ndp->ni_dvp;
815 		ndp->ni_dvp = NULL;
816 	}
817 	if (!lockleaf)
818 		cnp->cn_flags &= ~LOCKLEAF;
819 
820 out:
821 	if (error) {
822 		nfsvno_relpathbuf(ndp);
823 		ndp->ni_vp = NULL;
824 		ndp->ni_dvp = NULL;
825 		ndp->ni_startdir = NULL;
826 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
827 		ndp->ni_dvp = NULL;
828 	}
829 
830 out1:
831 	NFSEXITCODE2(error, nd);
832 	return (error);
833 }
834 
835 /*
836  * Set up a pathname buffer and return a pointer to it and, optionally
837  * set a hash pointer.
838  */
839 void
840 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
841 {
842 	struct componentname *cnp = &ndp->ni_cnd;
843 
844 	cnp->cn_flags |= (NOMACCHECK);
845 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
846 	if (hashpp != NULL)
847 		*hashpp = NULL;
848 	*bufpp = cnp->cn_pnbuf;
849 }
850 
851 /*
852  * Release the above path buffer, if not released by nfsvno_namei().
853  */
854 void
855 nfsvno_relpathbuf(struct nameidata *ndp)
856 {
857 
858 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
859 	ndp->ni_cnd.cn_pnbuf = NULL;
860 }
861 
862 /*
863  * Readlink vnode op into an mbuf list.
864  */
865 int
866 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz,
867     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
868 {
869 	struct iovec *iv;
870 	struct uio io, *uiop = &io;
871 	struct mbuf *mp, *mp3;
872 	int len, tlen, error = 0;
873 
874 	len = NFS_MAXPATHLEN;
875 	if (maxextsiz > 0)
876 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
877 		    &mp3, &mp, &iv);
878 	else
879 		uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
880 	uiop->uio_iov = iv;
881 	uiop->uio_offset = 0;
882 	uiop->uio_resid = len;
883 	uiop->uio_rw = UIO_READ;
884 	uiop->uio_segflg = UIO_SYSSPACE;
885 	uiop->uio_td = NULL;
886 	error = VOP_READLINK(vp, uiop, cred);
887 	free(iv, M_TEMP);
888 	if (error) {
889 		m_freem(mp3);
890 		*lenp = 0;
891 		goto out;
892 	}
893 	if (uiop->uio_resid > 0) {
894 		len -= uiop->uio_resid;
895 		tlen = NFSM_RNDUP(len);
896 		if (tlen == 0) {
897 			m_freem(mp3);
898 			mp3 = mp = NULL;
899 		} else if (tlen != NFS_MAXPATHLEN || tlen != len)
900 			mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
901 			    tlen - len);
902 	}
903 	*lenp = len;
904 	*mpp = mp3;
905 	*mpendp = mp;
906 
907 out:
908 	NFSEXITCODE(error);
909 	return (error);
910 }
911 
912 /*
913  * Create an mbuf chain and an associated iovec that can be used to Read
914  * or Getextattr of data.
915  * Upon success, return pointers to the first and last mbufs in the chain
916  * plus the malloc'd iovec and its iovlen.
917  */
918 static int
919 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
920     struct iovec **ivp)
921 {
922 	struct mbuf *m, *m2 = NULL, *m3;
923 	struct iovec *iv;
924 	int i, left, siz;
925 
926 	left = len;
927 	m3 = NULL;
928 	/*
929 	 * Generate the mbuf list with the uio_iov ref. to it.
930 	 */
931 	i = 0;
932 	while (left > 0) {
933 		NFSMGET(m);
934 		MCLGET(m, M_WAITOK);
935 		m->m_len = 0;
936 		siz = min(M_TRAILINGSPACE(m), left);
937 		left -= siz;
938 		i++;
939 		if (m3)
940 			m2->m_next = m;
941 		else
942 			m3 = m;
943 		m2 = m;
944 	}
945 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
946 	m = m3;
947 	left = len;
948 	i = 0;
949 	while (left > 0) {
950 		if (m == NULL)
951 			panic("nfsrv_createiovec iov");
952 		siz = min(M_TRAILINGSPACE(m), left);
953 		if (siz > 0) {
954 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
955 			iv->iov_len = siz;
956 			m->m_len += siz;
957 			left -= siz;
958 			iv++;
959 			i++;
960 		}
961 		m = m->m_next;
962 	}
963 	*mpp = m3;
964 	*mpendp = m2;
965 	return (i);
966 }
967 
968 /*
969  * Create an mbuf chain and an associated iovec that can be used to Read
970  * or Getextattr of data.
971  * Upon success, return pointers to the first and last mbufs in the chain
972  * plus the malloc'd iovec and its iovlen.
973  * Same as above, but creates ext_pgs mbuf(s).
974  */
975 static int
976 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp,
977     struct mbuf **mpendp, struct iovec **ivp)
978 {
979 	struct mbuf *m, *m2 = NULL, *m3;
980 	struct iovec *iv;
981 	int i, left, pgno, siz;
982 
983 	left = len;
984 	m3 = NULL;
985 	/*
986 	 * Generate the mbuf list with the uio_iov ref. to it.
987 	 */
988 	i = 0;
989 	while (left > 0) {
990 		siz = min(left, maxextsiz);
991 		m = mb_alloc_ext_plus_pages(siz, M_WAITOK);
992 		left -= siz;
993 		i += m->m_epg_npgs;
994 		if (m3 != NULL)
995 			m2->m_next = m;
996 		else
997 			m3 = m;
998 		m2 = m;
999 	}
1000 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1001 	m = m3;
1002 	left = len;
1003 	i = 0;
1004 	pgno = 0;
1005 	while (left > 0) {
1006 		if (m == NULL)
1007 			panic("nfsvno_createiovec_extpgs iov");
1008 		siz = min(PAGE_SIZE, left);
1009 		if (siz > 0) {
1010 			iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]);
1011 			iv->iov_len = siz;
1012 			m->m_len += siz;
1013 			if (pgno == m->m_epg_npgs - 1)
1014 				m->m_epg_last_len = siz;
1015 			left -= siz;
1016 			iv++;
1017 			i++;
1018 			pgno++;
1019 		}
1020 		if (pgno == m->m_epg_npgs && left > 0) {
1021 			m = m->m_next;
1022 			if (m == NULL)
1023 				panic("nfsvno_createiovec_extpgs iov");
1024 			pgno = 0;
1025 		}
1026 	}
1027 	*mpp = m3;
1028 	*mpendp = m2;
1029 	return (i);
1030 }
1031 
1032 /*
1033  * Read vnode op call into mbuf list.
1034  */
1035 int
1036 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
1037     int maxextsiz, struct thread *p, struct mbuf **mpp,
1038     struct mbuf **mpendp)
1039 {
1040 	struct mbuf *m;
1041 	struct iovec *iv;
1042 	int error = 0, len, tlen, ioflag = 0;
1043 	struct mbuf *m3;
1044 	struct uio io, *uiop = &io;
1045 	struct nfsheur *nh;
1046 
1047 	/*
1048 	 * Attempt to read from a DS file. A return of ENOENT implies
1049 	 * there is no DS file to read.
1050 	 */
1051 	error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
1052 	    NULL, mpendp, NULL, NULL, NULL, 0, NULL);
1053 	if (error != ENOENT)
1054 		return (error);
1055 
1056 	len = NFSM_RNDUP(cnt);
1057 	if (maxextsiz > 0)
1058 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
1059 		    &m3, &m, &iv);
1060 	else
1061 		uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
1062 	uiop->uio_iov = iv;
1063 	uiop->uio_offset = off;
1064 	uiop->uio_resid = len;
1065 	uiop->uio_rw = UIO_READ;
1066 	uiop->uio_segflg = UIO_SYSSPACE;
1067 	uiop->uio_td = NULL;
1068 	nh = nfsrv_sequential_heuristic(uiop, vp);
1069 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
1070 	/* XXX KDM make this more systematic? */
1071 	NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_READ] += uiop->uio_resid;
1072 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
1073 	free(iv, M_TEMP);
1074 	if (error) {
1075 		m_freem(m3);
1076 		*mpp = NULL;
1077 		goto out;
1078 	}
1079 	nh->nh_nextoff = uiop->uio_offset;
1080 	tlen = len - uiop->uio_resid;
1081 	cnt = cnt < tlen ? cnt : tlen;
1082 	tlen = NFSM_RNDUP(cnt);
1083 	if (tlen == 0) {
1084 		m_freem(m3);
1085 		m3 = m = NULL;
1086 	} else if (len != tlen || tlen != cnt)
1087 		m = nfsrv_adj(m3, len - tlen, tlen - cnt);
1088 	*mpp = m3;
1089 	*mpendp = m;
1090 
1091 out:
1092 	NFSEXITCODE(error);
1093 	return (error);
1094 }
1095 
1096 /*
1097  * Create the iovec for the mbuf chain passed in as an argument.
1098  * The "cp" argument is where the data starts within the first mbuf in
1099  * the chain. It returns the iovec and the iovcnt.
1100  */
1101 static int
1102 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
1103     int *iovcntp)
1104 {
1105 	struct mbuf *mp;
1106 	struct iovec *ivp;
1107 	int cnt, i, len;
1108 
1109 	/*
1110 	 * Loop through the mbuf chain, counting how many mbufs are a
1111 	 * part of this write operation, so the iovec size is known.
1112 	 */
1113 	cnt = 0;
1114 	len = retlen;
1115 	mp = m;
1116 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1117 	while (len > 0) {
1118 		if (i > 0) {
1119 			len -= i;
1120 			cnt++;
1121 		}
1122 		mp = mp->m_next;
1123 		if (!mp) {
1124 			if (len > 0)
1125 				return (EBADRPC);
1126 		} else
1127 			i = mp->m_len;
1128 	}
1129 
1130 	/* Now, create the iovec. */
1131 	mp = m;
1132 	*ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1133 	    M_WAITOK);
1134 	*iovcntp = cnt;
1135 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1136 	len = retlen;
1137 	while (len > 0) {
1138 		if (mp == NULL)
1139 			panic("nfsrv_createiovecw");
1140 		if (i > 0) {
1141 			i = min(i, len);
1142 			ivp->iov_base = cp;
1143 			ivp->iov_len = i;
1144 			ivp++;
1145 			len -= i;
1146 		}
1147 		mp = mp->m_next;
1148 		if (mp) {
1149 			i = mp->m_len;
1150 			cp = mtod(mp, caddr_t);
1151 		}
1152 	}
1153 	return (0);
1154 }
1155 
1156 /*
1157  * Write vnode op from an mbuf list.
1158  */
1159 int
1160 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
1161     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
1162 {
1163 	struct iovec *iv;
1164 	int cnt, ioflags, error;
1165 	struct uio io, *uiop = &io;
1166 	struct nfsheur *nh;
1167 
1168 	/*
1169 	 * Attempt to write to a DS file. A return of ENOENT implies
1170 	 * there is no DS file to write.
1171 	 */
1172 	error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
1173 	    &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
1174 	if (error != ENOENT) {
1175 		*stable = NFSWRITE_FILESYNC;
1176 		return (error);
1177 	}
1178 
1179 	if (*stable == NFSWRITE_UNSTABLE)
1180 		ioflags = IO_NODELOCKED;
1181 	else
1182 		ioflags = (IO_SYNC | IO_NODELOCKED);
1183 	error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
1184 	if (error != 0)
1185 		return (error);
1186 	uiop->uio_iov = iv;
1187 	uiop->uio_iovcnt = cnt;
1188 	uiop->uio_resid = retlen;
1189 	uiop->uio_rw = UIO_WRITE;
1190 	uiop->uio_segflg = UIO_SYSSPACE;
1191 	NFSUIOPROC(uiop, p);
1192 	uiop->uio_offset = off;
1193 	nh = nfsrv_sequential_heuristic(uiop, vp);
1194 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1195 	/* XXX KDM make this more systematic? */
1196 	NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
1197 	error = VOP_WRITE(vp, uiop, ioflags, cred);
1198 	if (error == 0)
1199 		nh->nh_nextoff = uiop->uio_offset;
1200 	free(iv, M_TEMP);
1201 
1202 	NFSEXITCODE(error);
1203 	return (error);
1204 }
1205 
1206 /*
1207  * Common code for creating a regular file (plus special files for V2).
1208  */
1209 int
1210 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1211     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1212     int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1213 {
1214 	u_quad_t tempsize;
1215 	int error;
1216 	struct thread *p = curthread;
1217 
1218 	error = nd->nd_repstat;
1219 	if (!error && ndp->ni_vp == NULL) {
1220 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1221 			error = VOP_CREATE(ndp->ni_dvp,
1222 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1223 			/* For a pNFS server, create the data file on a DS. */
1224 			if (error == 0 && nvap->na_type == VREG) {
1225 				/*
1226 				 * Create a data file on a DS for a pNFS server.
1227 				 * This function just returns if not
1228 				 * running a pNFS DS or the creation fails.
1229 				 */
1230 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1231 				    nd->nd_cred, p);
1232 			}
1233 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1234 			    NULL, false);
1235 			nfsvno_relpathbuf(ndp);
1236 			if (!error) {
1237 				if (*exclusive_flagp) {
1238 					*exclusive_flagp = 0;
1239 					NFSVNO_ATTRINIT(nvap);
1240 					nvap->na_atime.tv_sec = cverf[0];
1241 					nvap->na_atime.tv_nsec = cverf[1];
1242 					error = VOP_SETATTR(ndp->ni_vp,
1243 					    &nvap->na_vattr, nd->nd_cred);
1244 					if (error != 0) {
1245 						vput(ndp->ni_vp);
1246 						ndp->ni_vp = NULL;
1247 						error = NFSERR_NOTSUPP;
1248 					}
1249 				}
1250 			}
1251 		/*
1252 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1253 		 * (This implies, just get out on an error.)
1254 		 */
1255 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1256 			nvap->na_type == VFIFO) {
1257 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
1258 				nvap->na_type = VFIFO;
1259                         if (nvap->na_type != VFIFO &&
1260 			    (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1261 				nfsvno_relpathbuf(ndp);
1262 				vput(ndp->ni_dvp);
1263 				goto out;
1264 			}
1265 			nvap->na_rdev = rdev;
1266 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1267 			    &ndp->ni_cnd, &nvap->na_vattr);
1268 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1269 			    NULL, false);
1270 			nfsvno_relpathbuf(ndp);
1271 			if (error)
1272 				goto out;
1273 		} else {
1274 			nfsvno_relpathbuf(ndp);
1275 			vput(ndp->ni_dvp);
1276 			error = ENXIO;
1277 			goto out;
1278 		}
1279 		*vpp = ndp->ni_vp;
1280 	} else {
1281 		/*
1282 		 * Handle cases where error is already set and/or
1283 		 * the file exists.
1284 		 * 1 - clean up the lookup
1285 		 * 2 - iff !error and na_size set, truncate it
1286 		 */
1287 		nfsvno_relpathbuf(ndp);
1288 		*vpp = ndp->ni_vp;
1289 		if (ndp->ni_dvp == *vpp)
1290 			vrele(ndp->ni_dvp);
1291 		else
1292 			vput(ndp->ni_dvp);
1293 		if (!error && nvap->na_size != VNOVAL) {
1294 			error = nfsvno_accchk(*vpp, VWRITE,
1295 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1296 			    NFSACCCHK_VPISLOCKED, NULL);
1297 			if (!error) {
1298 				tempsize = nvap->na_size;
1299 				NFSVNO_ATTRINIT(nvap);
1300 				nvap->na_size = tempsize;
1301 				error = nfsvno_setattr(*vpp, nvap,
1302 				    nd->nd_cred, p, exp);
1303 			}
1304 		}
1305 		if (error)
1306 			vput(*vpp);
1307 	}
1308 
1309 out:
1310 	NFSEXITCODE(error);
1311 	return (error);
1312 }
1313 
1314 /*
1315  * Do a mknod vnode op.
1316  */
1317 int
1318 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1319     struct thread *p)
1320 {
1321 	int error = 0;
1322 	__enum_uint8(vtype) vtyp;
1323 
1324 	vtyp = nvap->na_type;
1325 	/*
1326 	 * Iff doesn't exist, create it.
1327 	 */
1328 	if (ndp->ni_vp) {
1329 		nfsvno_relpathbuf(ndp);
1330 		vput(ndp->ni_dvp);
1331 		vrele(ndp->ni_vp);
1332 		error = EEXIST;
1333 		goto out;
1334 	}
1335 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1336 		nfsvno_relpathbuf(ndp);
1337 		vput(ndp->ni_dvp);
1338 		error = NFSERR_BADTYPE;
1339 		goto out;
1340 	}
1341 	if (vtyp == VSOCK) {
1342 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1343 		    &ndp->ni_cnd, &nvap->na_vattr);
1344 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1345 		    false);
1346 		nfsvno_relpathbuf(ndp);
1347 	} else {
1348 		if (nvap->na_type != VFIFO &&
1349 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1350 			nfsvno_relpathbuf(ndp);
1351 			vput(ndp->ni_dvp);
1352 			goto out;
1353 		}
1354 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1355 		    &ndp->ni_cnd, &nvap->na_vattr);
1356 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1357 		    false);
1358 		nfsvno_relpathbuf(ndp);
1359 		/*
1360 		 * Since VOP_MKNOD returns the ni_vp, I can't
1361 		 * see any reason to do the lookup.
1362 		 */
1363 	}
1364 
1365 out:
1366 	NFSEXITCODE(error);
1367 	return (error);
1368 }
1369 
1370 /*
1371  * Mkdir vnode op.
1372  */
1373 int
1374 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1375     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1376 {
1377 	int error = 0;
1378 
1379 	if (ndp->ni_vp != NULL) {
1380 		if (ndp->ni_dvp == ndp->ni_vp)
1381 			vrele(ndp->ni_dvp);
1382 		else
1383 			vput(ndp->ni_dvp);
1384 		vrele(ndp->ni_vp);
1385 		nfsvno_relpathbuf(ndp);
1386 		error = EEXIST;
1387 		goto out;
1388 	}
1389 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1390 	    &nvap->na_vattr);
1391 	VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false);
1392 	nfsvno_relpathbuf(ndp);
1393 
1394 out:
1395 	NFSEXITCODE(error);
1396 	return (error);
1397 }
1398 
1399 /*
1400  * symlink vnode op.
1401  */
1402 int
1403 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1404     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1405     struct nfsexstuff *exp)
1406 {
1407 	int error = 0;
1408 
1409 	if (ndp->ni_vp) {
1410 		nfsvno_relpathbuf(ndp);
1411 		if (ndp->ni_dvp == ndp->ni_vp)
1412 			vrele(ndp->ni_dvp);
1413 		else
1414 			vput(ndp->ni_dvp);
1415 		vrele(ndp->ni_vp);
1416 		error = EEXIST;
1417 		goto out;
1418 	}
1419 
1420 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1421 	    &nvap->na_vattr, pathcp);
1422 	/*
1423 	 * Although FreeBSD still had the lookup code in
1424 	 * it for 7/current, there doesn't seem to be any
1425 	 * point, since VOP_SYMLINK() returns the ni_vp.
1426 	 * Just vput it for v2.
1427 	 */
1428 	VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0);
1429 	nfsvno_relpathbuf(ndp);
1430 
1431 out:
1432 	NFSEXITCODE(error);
1433 	return (error);
1434 }
1435 
1436 /*
1437  * Parse symbolic link arguments.
1438  * This function has an ugly side effect. It will malloc() an area for
1439  * the symlink and set iov_base to point to it, only if it succeeds.
1440  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1441  * be FREE'd later.
1442  */
1443 int
1444 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1445     struct thread *p, char **pathcpp, int *lenp)
1446 {
1447 	u_int32_t *tl;
1448 	char *pathcp = NULL;
1449 	int error = 0, len;
1450 	struct nfsv2_sattr *sp;
1451 
1452 	*pathcpp = NULL;
1453 	*lenp = 0;
1454 	if ((nd->nd_flag & ND_NFSV3) &&
1455 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1456 		goto nfsmout;
1457 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1458 	len = fxdr_unsigned(int, *tl);
1459 	if (len > NFS_MAXPATHLEN || len <= 0) {
1460 		error = EBADRPC;
1461 		goto nfsmout;
1462 	}
1463 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1464 	error = nfsrv_mtostr(nd, pathcp, len);
1465 	if (error)
1466 		goto nfsmout;
1467 	if (nd->nd_flag & ND_NFSV2) {
1468 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1469 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1470 	}
1471 	*pathcpp = pathcp;
1472 	*lenp = len;
1473 	NFSEXITCODE2(0, nd);
1474 	return (0);
1475 nfsmout:
1476 	if (pathcp)
1477 		free(pathcp, M_TEMP);
1478 	NFSEXITCODE2(error, nd);
1479 	return (error);
1480 }
1481 
1482 /*
1483  * Remove a non-directory object.
1484  */
1485 int
1486 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1487     struct thread *p, struct nfsexstuff *exp)
1488 {
1489 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1490 	int error = 0, mirrorcnt;
1491 	char fname[PNFS_FILENAME_LEN + 1];
1492 	fhandle_t fh;
1493 
1494 	vp = ndp->ni_vp;
1495 	dsdvp[0] = NULL;
1496 	if (vp->v_type == VDIR)
1497 		error = NFSERR_ISDIR;
1498 	else if (is_v4)
1499 		error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1500 		    p);
1501 	if (error == 0)
1502 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1503 	if (!error)
1504 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1505 	if (error == 0 && dsdvp[0] != NULL)
1506 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1507 	if (ndp->ni_dvp == vp)
1508 		vrele(ndp->ni_dvp);
1509 	else
1510 		vput(ndp->ni_dvp);
1511 	vput(vp);
1512 	nfsvno_relpathbuf(ndp);
1513 	NFSEXITCODE(error);
1514 	return (error);
1515 }
1516 
1517 /*
1518  * Remove a directory.
1519  */
1520 int
1521 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1522     struct thread *p, struct nfsexstuff *exp)
1523 {
1524 	struct vnode *vp;
1525 	int error = 0;
1526 
1527 	vp = ndp->ni_vp;
1528 	if (vp->v_type != VDIR) {
1529 		error = ENOTDIR;
1530 		goto out;
1531 	}
1532 	/*
1533 	 * No rmdir "." please.
1534 	 */
1535 	if (ndp->ni_dvp == vp) {
1536 		error = EINVAL;
1537 		goto out;
1538 	}
1539 	/*
1540 	 * The root of a mounted filesystem cannot be deleted.
1541 	 */
1542 	if (vp->v_vflag & VV_ROOT)
1543 		error = EBUSY;
1544 out:
1545 	if (!error)
1546 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1547 	if (ndp->ni_dvp == vp)
1548 		vrele(ndp->ni_dvp);
1549 	else
1550 		vput(ndp->ni_dvp);
1551 	vput(vp);
1552 	nfsvno_relpathbuf(ndp);
1553 	NFSEXITCODE(error);
1554 	return (error);
1555 }
1556 
1557 /*
1558  * Rename vnode op.
1559  */
1560 int
1561 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1562     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1563 {
1564 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1565 	int error = 0, mirrorcnt;
1566 	char fname[PNFS_FILENAME_LEN + 1];
1567 	fhandle_t fh;
1568 
1569 	dsdvp[0] = NULL;
1570 	fvp = fromndp->ni_vp;
1571 	if (ndstat) {
1572 		vrele(fromndp->ni_dvp);
1573 		vrele(fvp);
1574 		error = ndstat;
1575 		goto out1;
1576 	}
1577 	tdvp = tondp->ni_dvp;
1578 	tvp = tondp->ni_vp;
1579 	if (tvp != NULL) {
1580 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1581 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1582 			goto out;
1583 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1584 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1585 			goto out;
1586 		}
1587 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1588 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1589 			goto out;
1590 		}
1591 
1592 		/*
1593 		 * A rename to '.' or '..' results in a prematurely
1594 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1595 		 * here.
1596 		 */
1597 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1598 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1599 		    (tondp->ni_cnd.cn_namelen == 2 &&
1600 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1601 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1602 			error = EINVAL;
1603 			goto out;
1604 		}
1605 	}
1606 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1607 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1608 		goto out;
1609 	}
1610 	if (fvp->v_mount != tdvp->v_mount) {
1611 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1612 		goto out;
1613 	}
1614 	if (fvp == tdvp) {
1615 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1616 		goto out;
1617 	}
1618 	if (fvp == tvp) {
1619 		/*
1620 		 * If source and destination are the same, there is nothing to
1621 		 * do. Set error to -1 to indicate this.
1622 		 */
1623 		error = -1;
1624 		goto out;
1625 	}
1626 	if (ndflag & ND_NFSV4) {
1627 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1628 			error = nfsrv_checkremove(fvp, 0, NULL,
1629 			    (nfsquad_t)((u_quad_t)0), p);
1630 			NFSVOPUNLOCK(fvp);
1631 		} else
1632 			error = EPERM;
1633 		if (tvp && !error)
1634 			error = nfsrv_checkremove(tvp, 1, NULL,
1635 			    (nfsquad_t)((u_quad_t)0), p);
1636 	} else {
1637 		/*
1638 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1639 		 * that the NFSv4 client won't be confused by the rename.
1640 		 * Since nfsd_recalldelegation() can only be called on an
1641 		 * unlocked vnode at this point and fvp is the file that will
1642 		 * still exist after the rename, just do fvp.
1643 		 */
1644 		nfsd_recalldelegation(fvp, p);
1645 	}
1646 	if (error == 0 && tvp != NULL) {
1647 		nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1648 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1649 		    " dsdvp=%p\n", dsdvp[0]);
1650 	}
1651 out:
1652 	if (!error) {
1653 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1654 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1655 		    &tondp->ni_cnd);
1656 	} else {
1657 		if (tdvp == tvp)
1658 			vrele(tdvp);
1659 		else
1660 			vput(tdvp);
1661 		if (tvp)
1662 			vput(tvp);
1663 		vrele(fromndp->ni_dvp);
1664 		vrele(fvp);
1665 		if (error == -1)
1666 			error = 0;
1667 	}
1668 
1669 	/*
1670 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1671 	 * if the rename succeeded, the DS file for the tvp needs to be
1672 	 * removed.
1673 	 */
1674 	if (error == 0 && dsdvp[0] != NULL) {
1675 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1676 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1677 	}
1678 
1679 	nfsvno_relpathbuf(tondp);
1680 out1:
1681 	nfsvno_relpathbuf(fromndp);
1682 	NFSEXITCODE(error);
1683 	return (error);
1684 }
1685 
1686 /*
1687  * Link vnode op.
1688  */
1689 int
1690 nfsvno_link(struct nameidata *ndp, struct vnode *vp, nfsquad_t clientid,
1691     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1692 {
1693 	struct vnode *xp;
1694 	int error = 0;
1695 
1696 	xp = ndp->ni_vp;
1697 	if (xp != NULL) {
1698 		error = EEXIST;
1699 	} else {
1700 		xp = ndp->ni_dvp;
1701 		if (vp->v_mount != xp->v_mount)
1702 			error = EXDEV;
1703 	}
1704 	if (!error) {
1705 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1706 		if (!VN_IS_DOOMED(vp)) {
1707 			error = nfsrv_checkremove(vp, 0, NULL, clientid, p);
1708 			if (error == 0)
1709 				error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1710 		} else
1711 			error = EPERM;
1712 		if (ndp->ni_dvp == vp) {
1713 			vrele(ndp->ni_dvp);
1714 			NFSVOPUNLOCK(vp);
1715 		} else {
1716 			vref(vp);
1717 			VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true);
1718 		}
1719 	} else {
1720 		if (ndp->ni_dvp == ndp->ni_vp)
1721 			vrele(ndp->ni_dvp);
1722 		else
1723 			vput(ndp->ni_dvp);
1724 		if (ndp->ni_vp)
1725 			vrele(ndp->ni_vp);
1726 	}
1727 	nfsvno_relpathbuf(ndp);
1728 	NFSEXITCODE(error);
1729 	return (error);
1730 }
1731 
1732 /*
1733  * Do the fsync() appropriate for the commit.
1734  */
1735 int
1736 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1737     struct thread *td)
1738 {
1739 	int error = 0;
1740 
1741 	/*
1742 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1743 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1744 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1745 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1746 	 * File systems that do not use the buffer cache (as indicated
1747 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1748 	 */
1749 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1750 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1751 		/*
1752 		 * Give up and do the whole thing
1753 		 */
1754 		vnode_pager_clean_sync(vp);
1755 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1756 	} else {
1757 		/*
1758 		 * Locate and synchronously write any buffers that fall
1759 		 * into the requested range.  Note:  we are assuming that
1760 		 * f_iosize is a power of 2.
1761 		 */
1762 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1763 		int iomask = iosize - 1;
1764 		struct bufobj *bo;
1765 		daddr_t lblkno;
1766 
1767 		/*
1768 		 * Align to iosize boundary, super-align to page boundary.
1769 		 */
1770 		if (off & iomask) {
1771 			cnt += off & iomask;
1772 			off &= ~(u_quad_t)iomask;
1773 		}
1774 		if (off & PAGE_MASK) {
1775 			cnt += off & PAGE_MASK;
1776 			off &= ~(u_quad_t)PAGE_MASK;
1777 		}
1778 		lblkno = off / iosize;
1779 
1780 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1781 			VM_OBJECT_WLOCK(vp->v_object);
1782 			vm_object_page_clean(vp->v_object, off, off + cnt,
1783 			    OBJPC_SYNC);
1784 			VM_OBJECT_WUNLOCK(vp->v_object);
1785 		}
1786 
1787 		bo = &vp->v_bufobj;
1788 		BO_LOCK(bo);
1789 		while (cnt > 0) {
1790 			struct buf *bp;
1791 
1792 			/*
1793 			 * If we have a buffer and it is marked B_DELWRI we
1794 			 * have to lock and write it.  Otherwise the prior
1795 			 * write is assumed to have already been committed.
1796 			 *
1797 			 * gbincore() can return invalid buffers now so we
1798 			 * have to check that bit as well (though B_DELWRI
1799 			 * should not be set if B_INVAL is set there could be
1800 			 * a race here since we haven't locked the buffer).
1801 			 */
1802 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1803 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1804 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1805 					BO_LOCK(bo);
1806 					continue; /* retry */
1807 				}
1808 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1809 				    B_DELWRI) {
1810 					bremfree(bp);
1811 					bp->b_flags &= ~B_ASYNC;
1812 					bwrite(bp);
1813 					++nfs_commit_miss;
1814 				} else
1815 					BUF_UNLOCK(bp);
1816 				BO_LOCK(bo);
1817 			}
1818 			++nfs_commit_blks;
1819 			if (cnt < iosize)
1820 				break;
1821 			cnt -= iosize;
1822 			++lblkno;
1823 		}
1824 		BO_UNLOCK(bo);
1825 	}
1826 	NFSEXITCODE(error);
1827 	return (error);
1828 }
1829 
1830 /*
1831  * Statfs vnode op.
1832  */
1833 int
1834 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1835 {
1836 	struct statfs *tsf;
1837 	int error;
1838 
1839 	tsf = NULL;
1840 	if (nfsrv_devidcnt > 0) {
1841 		/* For a pNFS service, get the DS numbers. */
1842 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1843 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1844 		if (error != 0) {
1845 			free(tsf, M_TEMP);
1846 			tsf = NULL;
1847 		}
1848 	}
1849 	error = VFS_STATFS(vp->v_mount, sf);
1850 	if (error == 0) {
1851 		if (tsf != NULL) {
1852 			sf->f_blocks = tsf->f_blocks;
1853 			sf->f_bavail = tsf->f_bavail;
1854 			sf->f_bfree = tsf->f_bfree;
1855 			sf->f_bsize = tsf->f_bsize;
1856 		}
1857 		/*
1858 		 * Since NFS handles these values as unsigned on the
1859 		 * wire, there is no way to represent negative values,
1860 		 * so set them to 0. Without this, they will appear
1861 		 * to be very large positive values for clients like
1862 		 * Solaris10.
1863 		 */
1864 		if (sf->f_bavail < 0)
1865 			sf->f_bavail = 0;
1866 		if (sf->f_ffree < 0)
1867 			sf->f_ffree = 0;
1868 	}
1869 	free(tsf, M_TEMP);
1870 	NFSEXITCODE(error);
1871 	return (error);
1872 }
1873 
1874 /*
1875  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1876  * must handle nfsrv_opencheck() calls after any other access checks.
1877  */
1878 void
1879 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1880     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1881     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1882     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, bool done_namei,
1883     struct nfsexstuff *exp, struct vnode **vpp)
1884 {
1885 	struct vnode *vp = NULL;
1886 	u_quad_t tempsize;
1887 	struct nfsexstuff nes;
1888 	struct thread *p = curthread;
1889 	uint32_t oldrepstat;
1890 
1891 	if (ndp->ni_vp == NULL) {
1892 		/*
1893 		 * If nfsrv_opencheck() sets nd_repstat, done_namei needs to be
1894 		 * set true, since cleanup after nfsvno_namei() is needed.
1895 		 */
1896 		oldrepstat = nd->nd_repstat;
1897 		nd->nd_repstat = nfsrv_opencheck(clientid,
1898 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1899 		if (nd->nd_repstat != 0 && oldrepstat == 0)
1900 			done_namei = true;
1901 	}
1902 	if (!nd->nd_repstat) {
1903 		if (ndp->ni_vp == NULL) {
1904 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1905 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1906 			/* For a pNFS server, create the data file on a DS. */
1907 			if (nd->nd_repstat == 0) {
1908 				/*
1909 				 * Create a data file on a DS for a pNFS server.
1910 				 * This function just returns if not
1911 				 * running a pNFS DS or the creation fails.
1912 				 */
1913 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1914 				    cred, p);
1915 			}
1916 			VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ?
1917 			    &ndp->ni_vp : NULL, false);
1918 			nfsvno_relpathbuf(ndp);
1919 			if (!nd->nd_repstat) {
1920 				if (*exclusive_flagp) {
1921 					*exclusive_flagp = 0;
1922 					NFSVNO_ATTRINIT(nvap);
1923 					nvap->na_atime.tv_sec = cverf[0];
1924 					nvap->na_atime.tv_nsec = cverf[1];
1925 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1926 					    &nvap->na_vattr, cred);
1927 					if (nd->nd_repstat != 0) {
1928 						vput(ndp->ni_vp);
1929 						ndp->ni_vp = NULL;
1930 						nd->nd_repstat = NFSERR_NOTSUPP;
1931 					} else
1932 						NFSSETBIT_ATTRBIT(attrbitp,
1933 						    NFSATTRBIT_TIMEACCESS);
1934 				} else {
1935 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1936 					    aclp, p, attrbitp, exp);
1937 				}
1938 			}
1939 			vp = ndp->ni_vp;
1940 		} else {
1941 			nfsvno_relpathbuf(ndp);
1942 			vp = ndp->ni_vp;
1943 			if (create == NFSV4OPEN_CREATE) {
1944 				if (ndp->ni_dvp == vp)
1945 					vrele(ndp->ni_dvp);
1946 				else
1947 					vput(ndp->ni_dvp);
1948 			}
1949 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1950 				if (ndp->ni_cnd.cn_flags & RDONLY)
1951 					NFSVNO_SETEXRDONLY(&nes);
1952 				else
1953 					NFSVNO_EXINIT(&nes);
1954 				nd->nd_repstat = nfsvno_accchk(vp,
1955 				    VWRITE, cred, &nes, p,
1956 				    NFSACCCHK_NOOVERRIDE,
1957 				    NFSACCCHK_VPISLOCKED, NULL);
1958 				nd->nd_repstat = nfsrv_opencheck(clientid,
1959 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1960 				if (!nd->nd_repstat) {
1961 					tempsize = nvap->na_size;
1962 					NFSVNO_ATTRINIT(nvap);
1963 					nvap->na_size = tempsize;
1964 					nd->nd_repstat = nfsvno_setattr(vp,
1965 					    nvap, cred, p, exp);
1966 				}
1967 			} else if (vp->v_type == VREG) {
1968 				nd->nd_repstat = nfsrv_opencheck(clientid,
1969 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1970 			}
1971 		}
1972 	} else if (done_namei) {
1973 		KASSERT(create == NFSV4OPEN_CREATE,
1974 		    ("nfsvno_open: not create"));
1975 		/*
1976 		 * done_namei is set when nfsvno_namei() has completed
1977 		 * successfully, but a subsequent error was set in
1978 		 * nd_repstat.  As such, cleanup of the nfsvno_namei()
1979 		 * results is required.
1980 		 */
1981 		nfsvno_relpathbuf(ndp);
1982 		if (ndp->ni_dvp == ndp->ni_vp)
1983 			vrele(ndp->ni_dvp);
1984 		else
1985 			vput(ndp->ni_dvp);
1986 		if (ndp->ni_vp)
1987 			vput(ndp->ni_vp);
1988 	}
1989 	*vpp = vp;
1990 
1991 	NFSEXITCODE2(0, nd);
1992 }
1993 
1994 /*
1995  * Updates the file rev and sets the mtime and ctime
1996  * to the current clock time, returning the va_filerev and va_Xtime
1997  * values.
1998  * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1999  */
2000 int
2001 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
2002     struct nfsrv_descript *nd, struct thread *p)
2003 {
2004 	struct vattr va;
2005 
2006 	VATTR_NULL(&va);
2007 	vfs_timestamp(&va.va_mtime);
2008 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
2009 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
2010 		if (VN_IS_DOOMED(vp))
2011 			return (ESTALE);
2012 	}
2013 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
2014 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
2015 	return (0);
2016 }
2017 
2018 /*
2019  * Glue routine to nfsv4_fillattr().
2020  */
2021 int
2022 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
2023     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
2024     struct ucred *cred, struct thread *p, int isdgram, int reterr,
2025     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
2026 {
2027 	struct statfs *sf;
2028 	int error;
2029 
2030 	sf = NULL;
2031 	if (nfsrv_devidcnt > 0 &&
2032 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
2033 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
2034 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
2035 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
2036 		error = nfsrv_pnfsstatfs(sf, mp);
2037 		if (error != 0) {
2038 			free(sf, M_TEMP);
2039 			sf = NULL;
2040 		}
2041 	}
2042 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
2043 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
2044 	    mounted_on_fileno, sf);
2045 	free(sf, M_TEMP);
2046 	NFSEXITCODE2(0, nd);
2047 	return (error);
2048 }
2049 
2050 /*
2051  * Convert a dirent d_type to a vnode type.
2052  */
2053 static void nfs_dtypetovtype(struct nfsvattr *nvap, struct vnode *vp,
2054     uint8_t dtype)
2055 {
2056 
2057 	if ((vn_irflag_read(vp) & VIRF_NAMEDDIR) != 0) {
2058 		nvap->na_type = VREG;
2059 		nvap->na_bsdflags |= SFBSD_NAMEDATTR;
2060 	} else if (dtype <= DT_WHT) {
2061 		nvap->na_type = dtype_to_vnode[dtype];
2062 	} else {
2063 		nvap->na_type = VNON;
2064 	}
2065 }
2066 
2067 /* Since the Readdir vnode ops vary, put the entire functions in here. */
2068 /*
2069  * nfs readdir service
2070  * - mallocs what it thinks is enough to read
2071  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
2072  * - calls VOP_READDIR()
2073  * - loops around building the reply
2074  *	if the output generated exceeds count break out of loop
2075  *	The NFSM_CLGET macro is used here so that the reply will be packed
2076  *	tightly in mbuf clusters.
2077  * - it trims out records with d_fileno == 0
2078  *	this doesn't matter for Unix clients, but they might confuse clients
2079  *	for other os'.
2080  * - it trims out records with d_type == DT_WHT
2081  *	these cannot be seen through NFS (unless we extend the protocol)
2082  *     The alternate call nfsrvd_readdirplus() does lookups as well.
2083  * PS: The NFS protocol spec. does not clarify what the "count" byte
2084  *	argument is a count of.. just name strings and file id's or the
2085  *	entire reply rpc or ...
2086  *	I tried just file name and id sizes and it confused the Sun client,
2087  *	so I am using the full rpc size now. The "paranoia.." comment refers
2088  *	to including the status longwords that are not a part of the dir.
2089  *	"entry" structures, but are in the rpc.
2090  */
2091 int
2092 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
2093     struct vnode *vp, struct nfsexstuff *exp)
2094 {
2095 	struct dirent *dp;
2096 	u_int32_t *tl;
2097 	int dirlen;
2098 	char *cpos, *cend, *rbuf;
2099 	struct nfsvattr at;
2100 	int nlen, error = 0, getret = 1;
2101 	int siz, cnt, fullsiz, eofflag, ncookies;
2102 	u_int64_t off, toff, verf __unused;
2103 	uint64_t *cookies = NULL, *cookiep;
2104 	struct uio io;
2105 	struct iovec iv;
2106 	int is_ufs;
2107 	struct thread *p = curthread;
2108 
2109 	if (nd->nd_repstat) {
2110 		nfsrv_postopattr(nd, getret, &at);
2111 		goto out;
2112 	}
2113 	if (nd->nd_flag & ND_NFSV2) {
2114 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2115 		off = fxdr_unsigned(u_quad_t, *tl++);
2116 	} else {
2117 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2118 		off = fxdr_hyper(tl);
2119 		tl += 2;
2120 		verf = fxdr_hyper(tl);
2121 		tl += 2;
2122 	}
2123 	toff = off;
2124 	cnt = fxdr_unsigned(int, *tl);
2125 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2126 		cnt = NFS_SRVMAXDATA(nd);
2127 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2128 	fullsiz = siz;
2129 	if (nd->nd_flag & ND_NFSV3) {
2130 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
2131 		    NULL);
2132 #if 0
2133 		/*
2134 		 * va_filerev is not sufficient as a cookie verifier,
2135 		 * since it is not supposed to change when entries are
2136 		 * removed/added unless that offset cookies returned to
2137 		 * the client are no longer valid.
2138 		 */
2139 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
2140 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2141 #endif
2142 	}
2143 	if (!nd->nd_repstat && vp->v_type != VDIR)
2144 		nd->nd_repstat = NFSERR_NOTDIR;
2145 	if (nd->nd_repstat == 0 && cnt == 0) {
2146 		if (nd->nd_flag & ND_NFSV2)
2147 			/* NFSv2 does not have NFSERR_TOOSMALL */
2148 			nd->nd_repstat = EPERM;
2149 		else
2150 			nd->nd_repstat = NFSERR_TOOSMALL;
2151 	}
2152 	if (!nd->nd_repstat)
2153 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2154 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2155 		    NFSACCCHK_VPISLOCKED, NULL);
2156 	if (nd->nd_repstat) {
2157 		vput(vp);
2158 		if (nd->nd_flag & ND_NFSV3)
2159 			nfsrv_postopattr(nd, getret, &at);
2160 		goto out;
2161 	}
2162 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2163 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2164 again:
2165 	eofflag = 0;
2166 	if (cookies) {
2167 		free(cookies, M_TEMP);
2168 		cookies = NULL;
2169 	}
2170 
2171 	iv.iov_base = rbuf;
2172 	iv.iov_len = siz;
2173 	io.uio_iov = &iv;
2174 	io.uio_iovcnt = 1;
2175 	io.uio_offset = (off_t)off;
2176 	io.uio_resid = siz;
2177 	io.uio_segflg = UIO_SYSSPACE;
2178 	io.uio_rw = UIO_READ;
2179 	io.uio_td = NULL;
2180 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2181 	    &cookies);
2182 	off = (u_int64_t)io.uio_offset;
2183 	if (io.uio_resid)
2184 		siz -= io.uio_resid;
2185 
2186 	if (!cookies && !nd->nd_repstat)
2187 		nd->nd_repstat = NFSERR_PERM;
2188 	if (nd->nd_flag & ND_NFSV3) {
2189 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2190 		if (!nd->nd_repstat)
2191 			nd->nd_repstat = getret;
2192 	}
2193 
2194 	/*
2195 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
2196 	 */
2197 	if (nd->nd_repstat) {
2198 		vput(vp);
2199 		free(rbuf, M_TEMP);
2200 		if (cookies)
2201 			free(cookies, M_TEMP);
2202 		if (nd->nd_flag & ND_NFSV3)
2203 			nfsrv_postopattr(nd, getret, &at);
2204 		goto out;
2205 	}
2206 	/*
2207 	 * If nothing read, return eof
2208 	 * rpc reply
2209 	 */
2210 	if (siz == 0) {
2211 		vput(vp);
2212 		if (nd->nd_flag & ND_NFSV2) {
2213 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2214 		} else {
2215 			nfsrv_postopattr(nd, getret, &at);
2216 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2217 			txdr_hyper(at.na_filerev, tl);
2218 			tl += 2;
2219 		}
2220 		*tl++ = newnfs_false;
2221 		*tl = newnfs_true;
2222 		free(rbuf, M_TEMP);
2223 		free(cookies, M_TEMP);
2224 		goto out;
2225 	}
2226 
2227 	/*
2228 	 * Check for degenerate cases of nothing useful read.
2229 	 * If so go try again
2230 	 */
2231 	cpos = rbuf;
2232 	cend = rbuf + siz;
2233 	dp = (struct dirent *)cpos;
2234 	cookiep = cookies;
2235 
2236 	/*
2237 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2238 	 * directory offset up to a block boundary, so it is necessary to
2239 	 * skip over the records that precede the requested offset. This
2240 	 * requires the assumption that file offset cookies monotonically
2241 	 * increase.
2242 	 */
2243 	while (cpos < cend && ncookies > 0 &&
2244 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2245 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2246 		cpos += dp->d_reclen;
2247 		dp = (struct dirent *)cpos;
2248 		cookiep++;
2249 		ncookies--;
2250 	}
2251 	if (cpos >= cend || ncookies == 0) {
2252 		siz = fullsiz;
2253 		toff = off;
2254 		goto again;
2255 	}
2256 	vput(vp);
2257 
2258 	/*
2259 	 * If cnt > MCLBYTES and the reply will not be saved, use
2260 	 * ext_pgs mbufs for TLS.
2261 	 * For NFSv4.0, we do not know for sure if the reply will
2262 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2263 	 */
2264 	if (cnt > MCLBYTES && siz > MCLBYTES &&
2265 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2266 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2267 		nd->nd_flag |= ND_EXTPG;
2268 
2269 	/*
2270 	 * dirlen is the size of the reply, including all XDR and must
2271 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2272 	 * if the XDR should be included in "count", but to be safe, we do.
2273 	 * (Include the two booleans at the end of the reply in dirlen now.)
2274 	 */
2275 	if (nd->nd_flag & ND_NFSV3) {
2276 		nfsrv_postopattr(nd, getret, &at);
2277 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2278 		txdr_hyper(at.na_filerev, tl);
2279 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2280 	} else {
2281 		dirlen = 2 * NFSX_UNSIGNED;
2282 	}
2283 
2284 	/* Loop through the records and build reply */
2285 	while (cpos < cend && ncookies > 0) {
2286 		nlen = dp->d_namlen;
2287 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2288 			nlen <= NFS_MAXNAMLEN) {
2289 			if (nd->nd_flag & ND_NFSV3)
2290 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2291 			else
2292 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2293 			if (dirlen > cnt) {
2294 				eofflag = 0;
2295 				break;
2296 			}
2297 
2298 			/*
2299 			 * Build the directory record xdr from
2300 			 * the dirent entry.
2301 			 */
2302 			if (nd->nd_flag & ND_NFSV3) {
2303 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2304 				*tl++ = newnfs_true;
2305 				txdr_hyper(dp->d_fileno, tl);
2306 			} else {
2307 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2308 				*tl++ = newnfs_true;
2309 				*tl = txdr_unsigned(dp->d_fileno);
2310 			}
2311 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2312 			if (nd->nd_flag & ND_NFSV3) {
2313 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2314 				txdr_hyper(*cookiep, tl);
2315 			} else {
2316 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2317 				*tl = txdr_unsigned(*cookiep);
2318 			}
2319 		}
2320 		cpos += dp->d_reclen;
2321 		dp = (struct dirent *)cpos;
2322 		cookiep++;
2323 		ncookies--;
2324 	}
2325 	if (cpos < cend)
2326 		eofflag = 0;
2327 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2328 	*tl++ = newnfs_false;
2329 	if (eofflag)
2330 		*tl = newnfs_true;
2331 	else
2332 		*tl = newnfs_false;
2333 	free(rbuf, M_TEMP);
2334 	free(cookies, M_TEMP);
2335 
2336 out:
2337 	NFSEXITCODE2(0, nd);
2338 	return (0);
2339 nfsmout:
2340 	vput(vp);
2341 	NFSEXITCODE2(error, nd);
2342 	return (error);
2343 }
2344 
2345 /*
2346  * Readdirplus for V3 and Readdir for V4.
2347  */
2348 int
2349 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2350     struct vnode *vp, struct nfsexstuff *exp)
2351 {
2352 	struct dirent *dp;
2353 	u_int32_t *tl;
2354 	int dirlen;
2355 	char *cpos, *cend, *rbuf;
2356 	struct vnode *nvp;
2357 	fhandle_t nfh;
2358 	struct nfsvattr nva, at, *nvap = &nva;
2359 	struct mbuf *mb0, *mb1;
2360 	struct nfsreferral *refp;
2361 	int nlen, r, error = 0, getret = 1, usevget = 1;
2362 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2363 	caddr_t bpos0, bpos1;
2364 	u_int64_t off, toff, verf __unused;
2365 	uint64_t *cookies = NULL, *cookiep;
2366 	nfsattrbit_t attrbits, rderrbits, savbits, refbits;
2367 	struct uio io;
2368 	struct iovec iv;
2369 	struct componentname cn;
2370 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2371 	struct mount *mp, *new_mp;
2372 	uint64_t mounted_on_fileno;
2373 	struct thread *p = curthread;
2374 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2375 
2376 	if (nd->nd_repstat) {
2377 		nfsrv_postopattr(nd, getret, &at);
2378 		goto out;
2379 	}
2380 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2381 	off = fxdr_hyper(tl);
2382 	toff = off;
2383 	tl += 2;
2384 	verf = fxdr_hyper(tl);
2385 	tl += 2;
2386 	siz = fxdr_unsigned(int, *tl++);
2387 	cnt = fxdr_unsigned(int, *tl);
2388 
2389 	/*
2390 	 * Use the server's maximum data transfer size as the upper bound
2391 	 * on reply datalen.
2392 	 */
2393 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2394 		cnt = NFS_SRVMAXDATA(nd);
2395 
2396 	/*
2397 	 * siz is a "hint" of how much directory information (name, fileid,
2398 	 * cookie) should be in the reply. At least one client "hints" 0,
2399 	 * so I set it to cnt for that case. I also round it up to the
2400 	 * next multiple of DIRBLKSIZ.
2401 	 * Since the size of a Readdirplus directory entry reply will always
2402 	 * be greater than a directory entry returned by VOP_READDIR(), it
2403 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2404 	 * VOP_READDIR().
2405 	 */
2406 	if (siz <= 0)
2407 		siz = cnt;
2408 	else if (siz > NFS_SRVMAXDATA(nd))
2409 		siz = NFS_SRVMAXDATA(nd);
2410 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2411 
2412 	if (nd->nd_flag & ND_NFSV4) {
2413 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2414 		if (error)
2415 			goto nfsmout;
2416 		NFSSET_ATTRBIT(&savbits, &attrbits);
2417 		NFSSET_ATTRBIT(&refbits, &attrbits);
2418 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2419 		NFSZERO_ATTRBIT(&rderrbits);
2420 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2421 		/*
2422 		 * If these 4 bits are the only attributes requested by the
2423 		 * client, they can be satisfied without acquiring the vnode
2424 		 * for the file object unless it is a directory.
2425 		 * This will be indicated by savbits being all 0s.
2426 		 */
2427 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_TYPE);
2428 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_FILEID);
2429 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_MOUNTEDONFILEID);
2430 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_RDATTRERROR);
2431 	} else {
2432 		NFSZERO_ATTRBIT(&attrbits);
2433 	}
2434 	fullsiz = siz;
2435 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2436 #if 0
2437 	if (!nd->nd_repstat) {
2438 	    if (off && verf != at.na_filerev) {
2439 		/*
2440 		 * va_filerev is not sufficient as a cookie verifier,
2441 		 * since it is not supposed to change when entries are
2442 		 * removed/added unless that offset cookies returned to
2443 		 * the client are no longer valid.
2444 		 */
2445 		if (nd->nd_flag & ND_NFSV4) {
2446 			nd->nd_repstat = NFSERR_NOTSAME;
2447 		} else {
2448 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2449 		}
2450 	    }
2451 	}
2452 #endif
2453 	if (!nd->nd_repstat && vp->v_type != VDIR)
2454 		nd->nd_repstat = NFSERR_NOTDIR;
2455 	if (!nd->nd_repstat && cnt == 0)
2456 		nd->nd_repstat = NFSERR_TOOSMALL;
2457 	if (!nd->nd_repstat)
2458 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2459 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2460 		    NFSACCCHK_VPISLOCKED, NULL);
2461 	if (nd->nd_repstat) {
2462 		vput(vp);
2463 		if (nd->nd_flag & ND_NFSV3)
2464 			nfsrv_postopattr(nd, getret, &at);
2465 		goto out;
2466 	}
2467 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2468 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2469 
2470 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2471 again:
2472 	eofflag = 0;
2473 	if (cookies) {
2474 		free(cookies, M_TEMP);
2475 		cookies = NULL;
2476 	}
2477 
2478 	iv.iov_base = rbuf;
2479 	iv.iov_len = siz;
2480 	io.uio_iov = &iv;
2481 	io.uio_iovcnt = 1;
2482 	io.uio_offset = (off_t)off;
2483 	io.uio_resid = siz;
2484 	io.uio_segflg = UIO_SYSSPACE;
2485 	io.uio_rw = UIO_READ;
2486 	io.uio_td = NULL;
2487 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2488 	    &cookies);
2489 	off = (u_int64_t)io.uio_offset;
2490 	if (io.uio_resid)
2491 		siz -= io.uio_resid;
2492 
2493 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2494 
2495 	if (!cookies && !nd->nd_repstat)
2496 		nd->nd_repstat = NFSERR_PERM;
2497 	if (!nd->nd_repstat)
2498 		nd->nd_repstat = getret;
2499 	if (nd->nd_repstat) {
2500 		vput(vp);
2501 		if (cookies)
2502 			free(cookies, M_TEMP);
2503 		free(rbuf, M_TEMP);
2504 		if (nd->nd_flag & ND_NFSV3)
2505 			nfsrv_postopattr(nd, getret, &at);
2506 		goto out;
2507 	}
2508 	/*
2509 	 * If nothing read, return eof
2510 	 * rpc reply
2511 	 */
2512 	if (siz == 0) {
2513 		vput(vp);
2514 		if (nd->nd_flag & ND_NFSV3)
2515 			nfsrv_postopattr(nd, getret, &at);
2516 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2517 		txdr_hyper(at.na_filerev, tl);
2518 		tl += 2;
2519 		*tl++ = newnfs_false;
2520 		*tl = newnfs_true;
2521 		free(cookies, M_TEMP);
2522 		free(rbuf, M_TEMP);
2523 		goto out;
2524 	}
2525 
2526 	/*
2527 	 * Check for degenerate cases of nothing useful read.
2528 	 * If so go try again
2529 	 */
2530 	cpos = rbuf;
2531 	cend = rbuf + siz;
2532 	dp = (struct dirent *)cpos;
2533 	cookiep = cookies;
2534 
2535 	/*
2536 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2537 	 * directory offset up to a block boundary, so it is necessary to
2538 	 * skip over the records that precede the requested offset. This
2539 	 * requires the assumption that file offset cookies monotonically
2540 	 * increase.
2541 	 */
2542 	while (cpos < cend && ncookies > 0 &&
2543 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2544 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2545 	   ((nd->nd_flag & ND_NFSV4) &&
2546 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2547 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2548 		cpos += dp->d_reclen;
2549 		dp = (struct dirent *)cpos;
2550 		cookiep++;
2551 		ncookies--;
2552 	}
2553 	if (cpos >= cend || ncookies == 0) {
2554 		siz = fullsiz;
2555 		toff = off;
2556 		goto again;
2557 	}
2558 
2559 	/*
2560 	 * Busy the file system so that the mount point won't go away
2561 	 * and, as such, VFS_VGET() can be used safely.
2562 	 */
2563 	mp = vp->v_mount;
2564 	vfs_ref(mp);
2565 	NFSVOPUNLOCK(vp);
2566 	nd->nd_repstat = vfs_busy(mp, 0);
2567 	vfs_rel(mp);
2568 	if (nd->nd_repstat != 0) {
2569 		vrele(vp);
2570 		free(cookies, M_TEMP);
2571 		free(rbuf, M_TEMP);
2572 		if (nd->nd_flag & ND_NFSV3)
2573 			nfsrv_postopattr(nd, getret, &at);
2574 		goto out;
2575 	}
2576 
2577 	/*
2578 	 * Check to see if entries in this directory can be safely acquired
2579 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2580 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2581 	 * automount of the snapshot directory that is required will
2582 	 * be done.
2583 	 * This needs to be done here for NFSv4, since NFSv4 never does
2584 	 * a VFS_VGET() for "." or "..".
2585 	 */
2586 	if (is_zfs == 1) {
2587 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2588 		if (r == EOPNOTSUPP) {
2589 			usevget = 0;
2590 			cn.cn_nameiop = LOOKUP;
2591 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2592 			cn.cn_cred = nd->nd_cred;
2593 		} else if (r == 0)
2594 			vput(nvp);
2595 	}
2596 
2597 	/*
2598 	 * If the reply is likely to exceed MCLBYTES and the reply will
2599 	 * not be saved, use ext_pgs mbufs for TLS.
2600 	 * It is difficult to predict how large each entry will be and
2601 	 * how many entries have been read, so just assume the directory
2602 	 * entries grow by a factor of 4 when attributes are included.
2603 	 * For NFSv4.0, we do not know for sure if the reply will
2604 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2605 	 */
2606 	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
2607 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2608 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2609 		nd->nd_flag |= ND_EXTPG;
2610 
2611 	/*
2612 	 * Save this position, in case there is an error before one entry
2613 	 * is created.
2614 	 */
2615 	mb0 = nd->nd_mb;
2616 	bpos0 = nd->nd_bpos;
2617 	bextpg0 = nd->nd_bextpg;
2618 	bextpgsiz0 = nd->nd_bextpgsiz;
2619 
2620 	/*
2621 	 * Fill in the first part of the reply.
2622 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2623 	 * (Include the two booleans at the end of the reply in dirlen now,
2624 	 *  so we recognize when we have exceeded cnt.)
2625 	 */
2626 	if (nd->nd_flag & ND_NFSV3) {
2627 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2628 		nfsrv_postopattr(nd, getret, &at);
2629 	} else {
2630 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2631 	}
2632 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2633 	txdr_hyper(at.na_filerev, tl);
2634 
2635 	/*
2636 	 * Save this position, in case there is an empty reply needed.
2637 	 */
2638 	mb1 = nd->nd_mb;
2639 	bpos1 = nd->nd_bpos;
2640 	bextpg1 = nd->nd_bextpg;
2641 	bextpgsiz1 = nd->nd_bextpgsiz;
2642 
2643 	/* Loop through the records and build reply */
2644 	entrycnt = 0;
2645 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2646 		nlen = dp->d_namlen;
2647 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2648 		    nlen <= NFS_MAXNAMLEN &&
2649 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2650 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2651 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2652 			/*
2653 			 * Save the current position in the reply, in case
2654 			 * this entry exceeds cnt.
2655 			 */
2656 			mb1 = nd->nd_mb;
2657 			bpos1 = nd->nd_bpos;
2658 			bextpg1 = nd->nd_bextpg;
2659 			bextpgsiz1 = nd->nd_bextpgsiz;
2660 
2661 			/*
2662 			 * For readdir_and_lookup get the vnode using
2663 			 * the file number.
2664 			 */
2665 			nvp = NULL;
2666 			refp = NULL;
2667 			r = 0;
2668 			at_root = 0;
2669 			needs_unbusy = 0;
2670 			new_mp = mp;
2671 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2672 			if ((nd->nd_flag & ND_NFSV3) ||
2673 			    NFSNONZERO_ATTRBIT(&savbits) ||
2674 			    dp->d_type == DT_UNKNOWN ||
2675 			    (dp->d_type == DT_DIR &&
2676 			     nfsrv_enable_crossmntpt != 0)) {
2677 				if (nd->nd_flag & ND_NFSV4)
2678 					refp = nfsv4root_getreferral(NULL,
2679 					    vp, dp->d_fileno);
2680 				if (refp == NULL) {
2681 					if (usevget)
2682 						r = VFS_VGET(mp, dp->d_fileno,
2683 						    LK_SHARED, &nvp);
2684 					else
2685 						r = EOPNOTSUPP;
2686 					if (r == 0 && (vn_irflag_read(vp) &
2687 					    VIRF_NAMEDDIR) != 0)
2688 						vn_irflag_set_cond(nvp,
2689 						    VIRF_NAMEDATTR);
2690 					if (r == EOPNOTSUPP) {
2691 						if (usevget) {
2692 							usevget = 0;
2693 							cn.cn_nameiop = LOOKUP;
2694 							cn.cn_lkflags =
2695 							    LK_SHARED |
2696 							    LK_RETRY;
2697 							cn.cn_cred =
2698 							    nd->nd_cred;
2699 						}
2700 						cn.cn_nameptr = dp->d_name;
2701 						cn.cn_namelen = nlen;
2702 						cn.cn_flags = ISLASTCN |
2703 						    NOFOLLOW | LOCKLEAF;
2704 						if ((vn_irflag_read(vp) &
2705 						    VIRF_NAMEDDIR) != 0)
2706 							cn.cn_flags |=
2707 							    OPENNAMED;
2708 						if (nlen == 2 &&
2709 						    dp->d_name[0] == '.' &&
2710 						    dp->d_name[1] == '.')
2711 							cn.cn_flags |=
2712 							    ISDOTDOT;
2713 						if (NFSVOPLOCK(vp, LK_SHARED)
2714 						    != 0) {
2715 							nd->nd_repstat = EPERM;
2716 							break;
2717 						}
2718 						if ((vp->v_vflag & VV_ROOT) != 0
2719 						    && (cn.cn_flags & ISDOTDOT)
2720 						    != 0) {
2721 							vref(vp);
2722 							nvp = vp;
2723 							r = 0;
2724 						} else {
2725 							r = VOP_LOOKUP(vp, &nvp,
2726 							    &cn);
2727 							if (vp != nvp)
2728 								NFSVOPUNLOCK(vp);
2729 						}
2730 					}
2731 
2732 					/*
2733 					 * For NFSv4, check to see if nvp is
2734 					 * a mount point and get the mount
2735 					 * point vnode, as required.
2736 					 */
2737 					if (r == 0 &&
2738 					    nfsrv_enable_crossmntpt != 0 &&
2739 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2740 					    nvp->v_type == VDIR &&
2741 					    nvp->v_mountedhere != NULL) {
2742 						new_mp = nvp->v_mountedhere;
2743 						r = vfs_busy(new_mp, 0);
2744 						vput(nvp);
2745 						nvp = NULL;
2746 						if (r == 0) {
2747 							r = VFS_ROOT(new_mp,
2748 							    LK_SHARED, &nvp);
2749 							needs_unbusy = 1;
2750 							if (r == 0)
2751 								at_root = 1;
2752 						}
2753 					}
2754 				}
2755 
2756 				/*
2757 				 * If we failed to look up the entry, then it
2758 				 * has become invalid, most likely removed.
2759 				 */
2760 				if (r != 0) {
2761 					if (needs_unbusy)
2762 						vfs_unbusy(new_mp);
2763 					goto invalid;
2764 				}
2765 				KASSERT(refp != NULL || nvp != NULL,
2766 				    ("%s: undetected lookup error", __func__));
2767 
2768 				if (refp == NULL &&
2769 				    ((nd->nd_flag & ND_NFSV3) ||
2770 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2771 					r = nfsvno_getfh(nvp, &nfh, p);
2772 					if (!r)
2773 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2774 						1, &attrbits);
2775 					if (r == 0 && is_zfs == 1 &&
2776 					    nfsrv_enable_crossmntpt != 0 &&
2777 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2778 					    nvp->v_type == VDIR &&
2779 					    vp->v_mount != nvp->v_mount) {
2780 					    /*
2781 					     * For a ZFS snapshot, there is a
2782 					     * pseudo mount that does not set
2783 					     * v_mountedhere, so it needs to
2784 					     * be detected via a different
2785 					     * mount structure.
2786 					     */
2787 					    at_root = 1;
2788 					    if (new_mp == mp)
2789 						new_mp = nvp->v_mount;
2790 					}
2791 				}
2792 
2793 				/*
2794 				 * If we failed to get attributes of the entry,
2795 				 * then just skip it for NFSv3 (the traditional
2796 				 * behavior in the old NFS server).
2797 				 * For NFSv4 the behavior is controlled by
2798 				 * RDATTRERROR: we either ignore the error or
2799 				 * fail the request.
2800 				 * The exception is EOPNOTSUPP, which can be
2801 				 * returned by nfsvno_getfh() for certain
2802 				 * file systems, such as devfs.  This indicates
2803 				 * that the file system cannot be exported,
2804 				 * so just skip over the entry.
2805 				 * Note that RDATTRERROR is never set for NFSv3.
2806 				 */
2807 				if (r != 0) {
2808 					if (!NFSISSET_ATTRBIT(&attrbits,
2809 					    NFSATTRBIT_RDATTRERROR) ||
2810 					    r == EOPNOTSUPP) {
2811 						vput(nvp);
2812 						if (needs_unbusy != 0)
2813 							vfs_unbusy(new_mp);
2814 						if ((nd->nd_flag & ND_NFSV3) ||
2815 						    r == EOPNOTSUPP)
2816 							goto invalid;
2817 						nd->nd_repstat = r;
2818 						break;
2819 					}
2820 				}
2821 			} else if (NFSNONZERO_ATTRBIT(&attrbits)) {
2822 				/* Only need Type and/or Fileid. */
2823 				VATTR_NULL(&nvap->na_vattr);
2824 				nvap->na_fileid = dp->d_fileno;
2825 				nfs_dtypetovtype(nvap, vp, dp->d_type);
2826 			}
2827 
2828 			/*
2829 			 * Build the directory record xdr
2830 			 */
2831 			if (nd->nd_flag & ND_NFSV3) {
2832 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2833 				*tl++ = newnfs_true;
2834 				txdr_hyper(dp->d_fileno, tl);
2835 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2836 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2837 				txdr_hyper(*cookiep, tl);
2838 				nfsrv_postopattr(nd, 0, nvap);
2839 				dirlen += nfsm_fhtom(NULL, nd, (u_int8_t *)&nfh,
2840 				    0, 1);
2841 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2842 				if (nvp != NULL)
2843 					vput(nvp);
2844 			} else {
2845 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2846 				*tl++ = newnfs_true;
2847 				txdr_hyper(*cookiep, tl);
2848 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2849 				if (nvp != NULL) {
2850 					supports_nfsv4acls =
2851 					    nfs_supportsnfsv4acls(nvp);
2852 					NFSVOPUNLOCK(nvp);
2853 				} else
2854 					supports_nfsv4acls = 0;
2855 				if (refp != NULL) {
2856 					dirlen += nfsrv_putreferralattr(nd,
2857 					    &refbits, refp, 0,
2858 					    &nd->nd_repstat);
2859 					if (nd->nd_repstat) {
2860 						if (nvp != NULL)
2861 							vrele(nvp);
2862 						if (needs_unbusy != 0)
2863 							vfs_unbusy(new_mp);
2864 						break;
2865 					}
2866 				} else if (r) {
2867 					dirlen += nfsvno_fillattr(nd, new_mp,
2868 					    nvp, nvap, &nfh, r, &rderrbits,
2869 					    nd->nd_cred, p, isdgram, 0,
2870 					    supports_nfsv4acls, at_root,
2871 					    mounted_on_fileno);
2872 				} else {
2873 					dirlen += nfsvno_fillattr(nd, new_mp,
2874 					    nvp, nvap, &nfh, r, &attrbits,
2875 					    nd->nd_cred, p, isdgram, 0,
2876 					    supports_nfsv4acls, at_root,
2877 					    mounted_on_fileno);
2878 				}
2879 				if (nvp != NULL)
2880 					vrele(nvp);
2881 				dirlen += (3 * NFSX_UNSIGNED);
2882 			}
2883 			if (needs_unbusy != 0)
2884 				vfs_unbusy(new_mp);
2885 			if (dirlen <= cnt)
2886 				entrycnt++;
2887 		}
2888 invalid:
2889 		cpos += dp->d_reclen;
2890 		dp = (struct dirent *)cpos;
2891 		cookiep++;
2892 		ncookies--;
2893 	}
2894 	vrele(vp);
2895 	vfs_unbusy(mp);
2896 
2897 	/*
2898 	 * If dirlen > cnt, we must strip off the last entry. If that
2899 	 * results in an empty reply, report NFSERR_TOOSMALL.
2900 	 */
2901 	if (dirlen > cnt || nd->nd_repstat) {
2902 		if (!nd->nd_repstat && entrycnt == 0)
2903 			nd->nd_repstat = NFSERR_TOOSMALL;
2904 		if (nd->nd_repstat) {
2905 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2906 			if (nd->nd_flag & ND_NFSV3)
2907 				nfsrv_postopattr(nd, getret, &at);
2908 		} else
2909 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2910 		eofflag = 0;
2911 	} else if (cpos < cend)
2912 		eofflag = 0;
2913 	if (!nd->nd_repstat) {
2914 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2915 		*tl++ = newnfs_false;
2916 		if (eofflag)
2917 			*tl = newnfs_true;
2918 		else
2919 			*tl = newnfs_false;
2920 	}
2921 	free(cookies, M_TEMP);
2922 	free(rbuf, M_TEMP);
2923 
2924 out:
2925 	NFSEXITCODE2(0, nd);
2926 	return (0);
2927 nfsmout:
2928 	vput(vp);
2929 	NFSEXITCODE2(error, nd);
2930 	return (error);
2931 }
2932 
2933 /*
2934  * Get the settable attributes out of the mbuf list.
2935  * (Return 0 or EBADRPC)
2936  */
2937 int
2938 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2939     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2940 {
2941 	u_int32_t *tl;
2942 	struct nfsv2_sattr *sp;
2943 	int error = 0, toclient = 0;
2944 
2945 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2946 	case ND_NFSV2:
2947 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2948 		/*
2949 		 * Some old clients didn't fill in the high order 16bits.
2950 		 * --> check the low order 2 bytes for 0xffff
2951 		 */
2952 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2953 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2954 		if (sp->sa_uid != newnfs_xdrneg1)
2955 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2956 		if (sp->sa_gid != newnfs_xdrneg1)
2957 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2958 		if (sp->sa_size != newnfs_xdrneg1)
2959 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2960 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2961 #ifdef notyet
2962 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2963 #else
2964 			nvap->na_atime.tv_sec =
2965 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2966 			nvap->na_atime.tv_nsec = 0;
2967 #endif
2968 		}
2969 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2970 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2971 		break;
2972 	case ND_NFSV3:
2973 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2974 		if (*tl == newnfs_true) {
2975 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2976 			nvap->na_mode = nfstov_mode(*tl);
2977 		}
2978 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2979 		if (*tl == newnfs_true) {
2980 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2981 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2982 		}
2983 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2984 		if (*tl == newnfs_true) {
2985 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2986 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2987 		}
2988 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2989 		if (*tl == newnfs_true) {
2990 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2991 			nvap->na_size = fxdr_hyper(tl);
2992 		}
2993 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2994 		switch (fxdr_unsigned(int, *tl)) {
2995 		case NFSV3SATTRTIME_TOCLIENT:
2996 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2997 			fxdr_nfsv3time(tl, &nvap->na_atime);
2998 			toclient = 1;
2999 			break;
3000 		case NFSV3SATTRTIME_TOSERVER:
3001 			vfs_timestamp(&nvap->na_atime);
3002 			nvap->na_vaflags |= VA_UTIMES_NULL;
3003 			break;
3004 		}
3005 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3006 		switch (fxdr_unsigned(int, *tl)) {
3007 		case NFSV3SATTRTIME_TOCLIENT:
3008 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3009 			fxdr_nfsv3time(tl, &nvap->na_mtime);
3010 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
3011 			break;
3012 		case NFSV3SATTRTIME_TOSERVER:
3013 			vfs_timestamp(&nvap->na_mtime);
3014 			if (!toclient)
3015 				nvap->na_vaflags |= VA_UTIMES_NULL;
3016 			break;
3017 		}
3018 		break;
3019 	case ND_NFSV4:
3020 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
3021 	}
3022 nfsmout:
3023 	NFSEXITCODE2(error, nd);
3024 	return (error);
3025 }
3026 
3027 /*
3028  * Handle the setable attributes for V4.
3029  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
3030  */
3031 int
3032 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
3033     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
3034 {
3035 	u_int32_t *tl;
3036 	int attrsum = 0;
3037 	int i, j;
3038 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
3039 	int moderet, toclient = 0;
3040 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
3041 	uid_t uid;
3042 	gid_t gid;
3043 	u_short mode, mask;		/* Same type as va_mode. */
3044 	struct vattr va;
3045 
3046 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
3047 	if (error)
3048 		goto nfsmout;
3049 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3050 	attrsize = fxdr_unsigned(int, *tl);
3051 
3052 	/*
3053 	 * Loop around getting the setable attributes. If an unsupported
3054 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
3055 	 * Once nd_repstat != 0, do not set the attribute value, but keep
3056 	 * parsing the attribute(s).
3057 	 */
3058 	if (retnotsup) {
3059 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3060 		bitpos = NFSATTRBIT_MAX;
3061 	} else {
3062 		bitpos = 0;
3063 	}
3064 	moderet = 0;
3065 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
3066 	    if (attrsum > attrsize) {
3067 		error = NFSERR_BADXDR;
3068 		goto nfsmout;
3069 	    }
3070 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
3071 		switch (bitpos) {
3072 		case NFSATTRBIT_SIZE:
3073 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
3074 			if (!nd->nd_repstat) {
3075 				if (vp != NULL && vp->v_type != VREG)
3076 					nd->nd_repstat = (vp->v_type == VDIR) ?
3077 					    NFSERR_ISDIR : NFSERR_INVAL;
3078 				else
3079 					nvap->na_size = fxdr_hyper(tl);
3080 			}
3081 			attrsum += NFSX_HYPER;
3082 			break;
3083 		case NFSATTRBIT_ACL:
3084 			error = nfsrv_dissectacl(nd, aclp, true, &aceerr,
3085 			    &aclsize, p);
3086 			if (error)
3087 				goto nfsmout;
3088 			if (aceerr && !nd->nd_repstat)
3089 				nd->nd_repstat = aceerr;
3090 			attrsum += aclsize;
3091 			break;
3092 		case NFSATTRBIT_ARCHIVE:
3093 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3094 			if (!nd->nd_repstat)
3095 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3096 			attrsum += NFSX_UNSIGNED;
3097 			break;
3098 		case NFSATTRBIT_HIDDEN:
3099 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3100 			if (!nd->nd_repstat)
3101 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3102 			attrsum += NFSX_UNSIGNED;
3103 			break;
3104 		case NFSATTRBIT_MIMETYPE:
3105 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3106 			i = fxdr_unsigned(int, *tl);
3107 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3108 			if (error)
3109 				goto nfsmout;
3110 			if (!nd->nd_repstat)
3111 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3112 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
3113 			break;
3114 		case NFSATTRBIT_MODE:
3115 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
3116 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3117 			if (!nd->nd_repstat)
3118 				nvap->na_mode = nfstov_mode(*tl);
3119 			attrsum += NFSX_UNSIGNED;
3120 			break;
3121 		case NFSATTRBIT_OWNER:
3122 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3123 			j = fxdr_unsigned(int, *tl);
3124 			if (j < 0) {
3125 				error = NFSERR_BADXDR;
3126 				goto nfsmout;
3127 			}
3128 			if (j > NFSV4_SMALLSTR)
3129 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3130 			else
3131 				cp = namestr;
3132 			error = nfsrv_mtostr(nd, cp, j);
3133 			if (error) {
3134 				if (j > NFSV4_SMALLSTR)
3135 					free(cp, M_NFSSTRING);
3136 				goto nfsmout;
3137 			}
3138 			if (!nd->nd_repstat) {
3139 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
3140 				    &uid);
3141 				if (!nd->nd_repstat)
3142 					nvap->na_uid = uid;
3143 			}
3144 			if (j > NFSV4_SMALLSTR)
3145 				free(cp, M_NFSSTRING);
3146 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3147 			break;
3148 		case NFSATTRBIT_OWNERGROUP:
3149 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3150 			j = fxdr_unsigned(int, *tl);
3151 			if (j < 0) {
3152 				error = NFSERR_BADXDR;
3153 				goto nfsmout;
3154 			}
3155 			if (j > NFSV4_SMALLSTR)
3156 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3157 			else
3158 				cp = namestr;
3159 			error = nfsrv_mtostr(nd, cp, j);
3160 			if (error) {
3161 				if (j > NFSV4_SMALLSTR)
3162 					free(cp, M_NFSSTRING);
3163 				goto nfsmout;
3164 			}
3165 			if (!nd->nd_repstat) {
3166 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
3167 				    &gid);
3168 				if (!nd->nd_repstat)
3169 					nvap->na_gid = gid;
3170 			}
3171 			if (j > NFSV4_SMALLSTR)
3172 				free(cp, M_NFSSTRING);
3173 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3174 			break;
3175 		case NFSATTRBIT_SYSTEM:
3176 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3177 			if (!nd->nd_repstat)
3178 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3179 			attrsum += NFSX_UNSIGNED;
3180 			break;
3181 		case NFSATTRBIT_TIMEACCESSSET:
3182 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3183 			attrsum += NFSX_UNSIGNED;
3184 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3185 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3186 			    if (!nd->nd_repstat)
3187 				fxdr_nfsv4time(tl, &nvap->na_atime);
3188 			    toclient = 1;
3189 			    attrsum += NFSX_V4TIME;
3190 			} else if (!nd->nd_repstat) {
3191 			    vfs_timestamp(&nvap->na_atime);
3192 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3193 			}
3194 			break;
3195 		case NFSATTRBIT_TIMEBACKUP:
3196 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3197 			if (!nd->nd_repstat)
3198 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3199 			attrsum += NFSX_V4TIME;
3200 			break;
3201 		case NFSATTRBIT_TIMECREATE:
3202 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3203 			if (!nd->nd_repstat)
3204 				fxdr_nfsv4time(tl, &nvap->na_btime);
3205 			attrsum += NFSX_V4TIME;
3206 			break;
3207 		case NFSATTRBIT_TIMEMODIFYSET:
3208 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3209 			attrsum += NFSX_UNSIGNED;
3210 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3211 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3212 			    if (!nd->nd_repstat)
3213 				fxdr_nfsv4time(tl, &nvap->na_mtime);
3214 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3215 			    attrsum += NFSX_V4TIME;
3216 			} else if (!nd->nd_repstat) {
3217 			    vfs_timestamp(&nvap->na_mtime);
3218 			    if (!toclient)
3219 				nvap->na_vaflags |= VA_UTIMES_NULL;
3220 			}
3221 			break;
3222 		case NFSATTRBIT_MODESETMASKED:
3223 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3224 			mode = fxdr_unsigned(u_short, *tl++);
3225 			mask = fxdr_unsigned(u_short, *tl);
3226 			/*
3227 			 * vp == NULL implies an Open/Create operation.
3228 			 * This attribute can only be used for Setattr and
3229 			 * only for NFSv4.1 or higher.
3230 			 * If moderet != 0, a mode attribute has also been
3231 			 * specified and this attribute cannot be done in the
3232 			 * same Setattr operation.
3233 			 */
3234 			if (!nd->nd_repstat) {
3235 				if ((nd->nd_flag & ND_NFSV41) == 0)
3236 					nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3237 				else if ((mode & ~07777) != 0 ||
3238 				    (mask & ~07777) != 0 || vp == NULL)
3239 					nd->nd_repstat = NFSERR_INVAL;
3240 				else if (moderet == 0)
3241 					moderet = VOP_GETATTR(vp, &va,
3242 					    nd->nd_cred);
3243 				if (moderet == 0)
3244 					nvap->na_mode = (mode & mask) |
3245 					    (va.va_mode & ~mask);
3246 				else
3247 					nd->nd_repstat = moderet;
3248 			}
3249 			attrsum += 2 * NFSX_UNSIGNED;
3250 			break;
3251 		case NFSATTRBIT_MODEUMASK:
3252 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3253 			mode = fxdr_unsigned(u_short, *tl++);
3254 			mask = fxdr_unsigned(u_short, *tl);
3255 			/*
3256 			 * If moderet != 0, mode has already been done.
3257 			 * If vp != NULL, this is not a file object creation.
3258 			 */
3259 			if (!nd->nd_repstat) {
3260 				if ((nd->nd_flag & ND_NFSV42) == 0)
3261 					nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3262 				else if ((mask & ~0777) != 0 || vp != NULL ||
3263 				    moderet != 0)
3264 					nd->nd_repstat = NFSERR_INVAL;
3265 				else
3266 					nvap->na_mode = (mode & ~mask);
3267 			}
3268 			attrsum += 2 * NFSX_UNSIGNED;
3269 			break;
3270 		default:
3271 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3272 			/*
3273 			 * set bitpos so we drop out of the loop.
3274 			 */
3275 			bitpos = NFSATTRBIT_MAX;
3276 			break;
3277 		}
3278 	}
3279 
3280 	/*
3281 	 * some clients pad the attrlist, so we need to skip over the
3282 	 * padding.  This also skips over unparsed non-supported attributes.
3283 	 */
3284 	if (attrsum > attrsize) {
3285 		error = NFSERR_BADXDR;
3286 	} else {
3287 		attrsize = NFSM_RNDUP(attrsize);
3288 		if (attrsum < attrsize)
3289 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3290 	}
3291 nfsmout:
3292 	NFSEXITCODE2(error, nd);
3293 	return (error);
3294 }
3295 
3296 /*
3297  * Check/setup export credentials.
3298  */
3299 int
3300 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3301     struct ucred *credanon, bool testsec)
3302 {
3303 	int error;
3304 
3305 	/*
3306 	 * Check/setup credentials.
3307 	 */
3308 	if (nd->nd_flag & ND_GSS)
3309 		exp->nes_exflag &= ~MNT_EXPORTANON;
3310 
3311 	/*
3312 	 * Check to see if the operation is allowed for this security flavor.
3313 	 */
3314 	error = 0;
3315 	if (testsec) {
3316 		error = nfsvno_testexp(nd, exp);
3317 		if (error != 0)
3318 			goto out;
3319 	}
3320 
3321 	/*
3322 	 * Check to see if the file system is exported V4 only.
3323 	 */
3324 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3325 		error = NFSERR_PROGNOTV4;
3326 		goto out;
3327 	}
3328 
3329 	/*
3330 	 * Now, map the user credentials.
3331 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3332 	 *  Fsinfo RPC. If set for anything else, this code might need
3333 	 *  to change.)
3334 	 */
3335 	if (NFSVNO_EXPORTED(exp)) {
3336 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3337 		     NFSVNO_EXPORTANON(exp) ||
3338 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3339 			nd->nd_cred->cr_uid = credanon->cr_uid;
3340 			/*
3341 			 * 'credanon' is already a 'struct ucred' that was built
3342 			 * internally with calls to crsetgroups_fallback(), so
3343 			 * we don't need a fallback here.
3344 			 */
3345 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3346 			    credanon->cr_groups);
3347 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3348 			/*
3349 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3350 			 * if there is a replacement credential with a group
3351 			 * list set up by "nfsuserd -manage-gids".
3352 			 * If there is no replacement, nfsrv_getgrpscred()
3353 			 * simply returns its argument.
3354 			 */
3355 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3356 		}
3357 	}
3358 
3359 out:
3360 	NFSEXITCODE2(error, nd);
3361 	return (error);
3362 }
3363 
3364 /*
3365  * Check exports.
3366  */
3367 int
3368 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3369     struct ucred **credp)
3370 {
3371 	int error;
3372 
3373 	error = 0;
3374 	*credp = NULL;
3375 	MNT_ILOCK(mp);
3376 	if (mp->mnt_exjail == NULL ||
3377 	    mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison)
3378 		error = EACCES;
3379 	MNT_IUNLOCK(mp);
3380 	if (error == 0)
3381 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3382 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3383 	if (error) {
3384 		if (NFSD_VNET(nfs_rootfhset)) {
3385 			exp->nes_exflag = 0;
3386 			exp->nes_numsecflavor = 0;
3387 			error = 0;
3388 		}
3389 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3390 	    MAXSECFLAVORS) {
3391 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3392 		exp->nes_numsecflavor = 0;
3393 		error = EACCES;
3394 	}
3395 	NFSEXITCODE(error);
3396 	return (error);
3397 }
3398 
3399 /*
3400  * Get a vnode for a file handle and export stuff.
3401  */
3402 int
3403 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3404     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3405     struct ucred **credp)
3406 {
3407 	int error;
3408 
3409 	*credp = NULL;
3410 	exp->nes_numsecflavor = 0;
3411 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3412 	if (error != 0)
3413 		/* Make sure the server replies ESTALE to the client. */
3414 		error = ESTALE;
3415 	if (nam && !error) {
3416 		MNT_ILOCK(mp);
3417 		if (mp->mnt_exjail == NULL ||
3418 		    mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison)
3419 			error = EACCES;
3420 		MNT_IUNLOCK(mp);
3421 		if (error == 0)
3422 			error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3423 			    &exp->nes_numsecflavor, exp->nes_secflavors);
3424 		if (error) {
3425 			if (NFSD_VNET(nfs_rootfhset)) {
3426 				exp->nes_exflag = 0;
3427 				exp->nes_numsecflavor = 0;
3428 				error = 0;
3429 			} else {
3430 				vput(*vpp);
3431 			}
3432 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3433 		    MAXSECFLAVORS) {
3434 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3435 			exp->nes_numsecflavor = 0;
3436 			error = EACCES;
3437 			vput(*vpp);
3438 		}
3439 	}
3440 	NFSEXITCODE(error);
3441 	return (error);
3442 }
3443 
3444 /*
3445  * nfsd_fhtovp() - convert a fh to a vnode ptr
3446  * 	- look up fsid in mount list (if not found ret error)
3447  *	- get vp and export rights by calling nfsvno_fhtovp()
3448  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3449  *	  for AUTH_SYS
3450  *	- if mpp != NULL, return the mount point so that it can
3451  *	  be used for vn_finished_write() by the caller
3452  */
3453 void
3454 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3455     struct vnode **vpp, struct nfsexstuff *exp,
3456     struct mount **mpp, int startwrite, int nextop)
3457 {
3458 	struct mount *mp, *mpw;
3459 	struct ucred *credanon;
3460 	fhandle_t *fhp;
3461 	int error;
3462 
3463 	if (mpp != NULL)
3464 		*mpp = NULL;
3465 	*vpp = NULL;
3466 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3467 	mp = vfs_busyfs(&fhp->fh_fsid);
3468 	if (mp == NULL) {
3469 		nd->nd_repstat = ESTALE;
3470 		goto out;
3471 	}
3472 
3473 	if (startwrite) {
3474 		mpw = mp;
3475 		error = vn_start_write(NULL, &mpw, V_WAIT);
3476 		if (error != 0) {
3477 			mpw = NULL;
3478 			vfs_unbusy(mp);
3479 			nd->nd_repstat = ESTALE;
3480 			goto out;
3481 		}
3482 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3483 			lktype = LK_EXCLUSIVE;
3484 	} else
3485 		mpw = NULL;
3486 
3487 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3488 	    &credanon);
3489 	vfs_unbusy(mp);
3490 
3491 	if (nd->nd_repstat == 0 &&
3492 	    nfp->nfsrvfh_len >= NFSX_MYFH + NFSX_V4NAMEDDIRFH &&
3493 	    nfp->nfsrvfh_len <= NFSX_MYFH + NFSX_V4NAMEDATTRFH) {
3494 		if (nfp->nfsrvfh_len == NFSX_MYFH + NFSX_V4NAMEDDIRFH)
3495 			vn_irflag_set_cond(*vpp, VIRF_NAMEDDIR);
3496 		else
3497 			vn_irflag_set_cond(*vpp, VIRF_NAMEDATTR);
3498 	}
3499 
3500 	/*
3501 	 * For NFSv4 without a pseudo root fs, unexported file handles
3502 	 * can be returned, so that Lookup works everywhere.
3503 	 */
3504 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3505 	    !(nd->nd_flag & ND_NFSV4)) {
3506 		vput(*vpp);
3507 		*vpp = NULL;
3508 		nd->nd_repstat = EACCES;
3509 	}
3510 
3511 	/*
3512 	 * Personally, I've never seen any point in requiring a
3513 	 * reserved port#, since only in the rare case where the
3514 	 * clients are all boxes with secure system privileges,
3515 	 * does it provide any enhanced security, but... some people
3516 	 * believe it to be useful and keep putting this code back in.
3517 	 * (There is also some "security checker" out there that
3518 	 *  complains if the nfs server doesn't enforce this.)
3519 	 * However, note the following:
3520 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3521 	 *	required.
3522 	 * RFC2623 recommends that, if a reserved port# is checked for,
3523 	 *	that there be a way to turn that off--> ifdef'd.
3524 	 */
3525 #ifdef NFS_REQRSVPORT
3526 	if (!nd->nd_repstat) {
3527 		struct sockaddr_in *saddr;
3528 		struct sockaddr_in6 *saddr6;
3529 
3530 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3531 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3532 		if (!(nd->nd_flag & ND_NFSV4) &&
3533 		    ((saddr->sin_family == AF_INET &&
3534 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3535 		     (saddr6->sin6_family == AF_INET6 &&
3536 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3537 			vput(*vpp);
3538 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3539 		}
3540 	}
3541 #endif	/* NFS_REQRSVPORT */
3542 
3543 	/*
3544 	 * Check/setup credentials.
3545 	 */
3546 	if (!nd->nd_repstat) {
3547 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3548 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3549 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3550 		if (nd->nd_repstat)
3551 			vput(*vpp);
3552 	}
3553 	if (credanon != NULL)
3554 		crfree(credanon);
3555 	if (nd->nd_repstat) {
3556 		vn_finished_write(mpw);
3557 		*vpp = NULL;
3558 	} else if (mpp != NULL) {
3559 		*mpp = mpw;
3560 	}
3561 
3562 out:
3563 	NFSEXITCODE2(0, nd);
3564 }
3565 
3566 /*
3567  * glue for fp.
3568  */
3569 static int
3570 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3571 {
3572 	struct filedesc *fdp;
3573 	struct file *fp;
3574 	int error = 0;
3575 
3576 	fdp = p->td_proc->p_fd;
3577 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3578 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3579 		error = EBADF;
3580 		goto out;
3581 	}
3582 	*fpp = fp;
3583 
3584 out:
3585 	NFSEXITCODE(error);
3586 	return (error);
3587 }
3588 
3589 /*
3590  * Called from nfssvc() to update the exports list. Just call
3591  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3592  * in the mount list.
3593  */
3594 int
3595 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3596 {
3597 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3598 	int error = 0;
3599 	struct nameidata nd;
3600 	fhandle_t fh;
3601 
3602 	error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export, false);
3603 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3604 		NFSD_VNET(nfs_rootfhset) = 0;
3605 	else if (error == 0) {
3606 		if (nfsexargp->fspec == NULL) {
3607 			error = EPERM;
3608 			goto out;
3609 		}
3610 		/*
3611 		 * If fspec != NULL, this is the v4root path.
3612 		 */
3613 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec);
3614 		if ((error = namei(&nd)) != 0)
3615 			goto out;
3616 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3617 		vrele(nd.ni_vp);
3618 		if (!error) {
3619 			NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH;
3620 			NFSBCOPY((caddr_t)&fh,
3621 			    NFSD_VNET(nfs_rootfh).nfsrvfh_data,
3622 			    sizeof (fhandle_t));
3623 			NFSD_VNET(nfs_rootfhset) = 1;
3624 		}
3625 	}
3626 
3627 out:
3628 	NFSEXITCODE(error);
3629 	return (error);
3630 }
3631 
3632 /*
3633  * This function needs to test to see if the system is near its limit
3634  * for memory allocation via malloc() or mget() and return True iff
3635  * either of these resources are near their limit.
3636  * XXX (For now, this is just a stub.)
3637  */
3638 int nfsrv_testmalloclimit = 0;
3639 int
3640 nfsrv_mallocmget_limit(void)
3641 {
3642 	static int printmesg = 0;
3643 	static int testval = 1;
3644 
3645 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3646 		if ((printmesg++ % 100) == 0)
3647 			printf("nfsd: malloc/mget near limit\n");
3648 		return (1);
3649 	}
3650 	return (0);
3651 }
3652 
3653 /*
3654  * BSD specific initialization of a mount point.
3655  */
3656 void
3657 nfsd_mntinit(void)
3658 {
3659 
3660 	NFSD_LOCK();
3661 	if (NFSD_VNET(nfsrv_mntinited)) {
3662 		NFSD_UNLOCK();
3663 		return;
3664 	}
3665 	NFSD_VNET(nfsrv_mntinited) = true;
3666 	nfsrvd_init(0);
3667 	NFSD_UNLOCK();
3668 
3669 	NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP,
3670 	    M_WAITOK | M_ZERO);
3671 	NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3672 	mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3673 	lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0);
3674 	TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist);
3675 	TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist);
3676 	NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL;
3677 	TAILQ_INIT(&NFSD_VNET(nfsv4root_opt));
3678 	TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt));
3679 	NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt);
3680 	NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt);
3681 	NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0;
3682 	NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0;
3683 	callout_init(&NFSD_VNET(nfsd_callout), 1);
3684 
3685 	nfsrvd_initcache();
3686 	nfsd_init();
3687 }
3688 
3689 static void
3690 nfsd_timer(void *arg)
3691 {
3692 	struct vnet *vnetp;
3693 
3694 	vnetp = (struct vnet *)arg;
3695 	NFSD_CURVNET_SET_QUIET(vnetp);
3696 	nfsrv_servertimer(vnetp);
3697 	callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer,
3698 	    arg, 0);
3699 	NFSD_CURVNET_RESTORE();
3700 }
3701 
3702 /*
3703  * Get a vnode for a file handle, without checking exports, etc.
3704  */
3705 struct vnode *
3706 nfsvno_getvp(fhandle_t *fhp)
3707 {
3708 	struct mount *mp;
3709 	struct vnode *vp;
3710 	int error;
3711 
3712 	mp = vfs_busyfs(&fhp->fh_fsid);
3713 	if (mp == NULL)
3714 		return (NULL);
3715 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3716 	vfs_unbusy(mp);
3717 	if (error)
3718 		return (NULL);
3719 	return (vp);
3720 }
3721 
3722 /*
3723  * Do a local VOP_ADVLOCK().
3724  */
3725 int
3726 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3727     u_int64_t end, struct thread *td)
3728 {
3729 	int error = 0;
3730 	struct flock fl;
3731 	u_int64_t tlen;
3732 
3733 	if (nfsrv_dolocallocks == 0)
3734 		goto out;
3735 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3736 
3737 	fl.l_whence = SEEK_SET;
3738 	fl.l_type = ftype;
3739 	fl.l_start = (off_t)first;
3740 	if (end == NFS64BITSSET) {
3741 		fl.l_len = 0;
3742 	} else {
3743 		tlen = end - first;
3744 		fl.l_len = (off_t)tlen;
3745 	}
3746 	/*
3747 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3748 	 * values for all calls, so that all locks will be held by the
3749 	 * nfsd server. (The nfsd server handles conflicts between the
3750 	 * various clients.)
3751 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3752 	 * bytes, so it can't be put in l_sysid.
3753 	 */
3754 	if (nfsv4_sysid == 0)
3755 		nfsv4_sysid = nlm_acquire_next_sysid();
3756 	fl.l_pid = (pid_t)0;
3757 	fl.l_sysid = (int)nfsv4_sysid;
3758 
3759 	if (ftype == F_UNLCK)
3760 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3761 		    (F_POSIX | F_REMOTE));
3762 	else
3763 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3764 		    (F_POSIX | F_REMOTE));
3765 
3766 out:
3767 	NFSEXITCODE(error);
3768 	return (error);
3769 }
3770 
3771 /*
3772  * Check the nfsv4 root exports.
3773  */
3774 int
3775 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3776 {
3777 	struct ucred *credanon;
3778 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3779 	uint64_t exflags;
3780 
3781 	error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags,
3782 	    &credanon, &numsecflavor, secflavors);
3783 	if (error) {
3784 		error = NFSERR_PROGUNAVAIL;
3785 		goto out;
3786 	}
3787 	if (credanon != NULL)
3788 		crfree(credanon);
3789 	for (i = 0; i < numsecflavor; i++) {
3790 		if (secflavors[i] == AUTH_SYS)
3791 			nd->nd_flag |= ND_EXAUTHSYS;
3792 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3793 			nd->nd_flag |= ND_EXGSS;
3794 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3795 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3796 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3797 			nd->nd_flag |= ND_EXGSSPRIVACY;
3798 	}
3799 
3800 	/* And set ND_EXxx flags for TLS. */
3801 	if ((exflags & MNT_EXTLS) != 0) {
3802 		nd->nd_flag |= ND_EXTLS;
3803 		if ((exflags & MNT_EXTLSCERT) != 0)
3804 			nd->nd_flag |= ND_EXTLSCERT;
3805 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3806 			nd->nd_flag |= ND_EXTLSCERTUSER;
3807 	}
3808 
3809 out:
3810 	NFSEXITCODE(error);
3811 	return (error);
3812 }
3813 
3814 /*
3815  * Nfs server pseudo system call for the nfsd's
3816  */
3817 /*
3818  * MPSAFE
3819  */
3820 static int
3821 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3822 {
3823 	struct file *fp;
3824 	struct nfsd_addsock_args sockarg;
3825 	struct nfsd_nfsd_args nfsdarg;
3826 	struct nfsd_nfsd_oargs onfsdarg;
3827 	struct nfsd_pnfsd_args pnfsdarg;
3828 	struct vnode *vp, *nvp, *curdvp;
3829 	struct pnfsdsfile *pf;
3830 	struct nfsdevice *ds, *fds;
3831 	cap_rights_t rights;
3832 	int buflen, error, ret;
3833 	char *buf, *cp, *cp2, *cp3;
3834 	char fname[PNFS_FILENAME_LEN + 1];
3835 
3836 	NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td));
3837 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3838 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3839 		if (error)
3840 			goto out;
3841 		/*
3842 		 * Since we don't know what rights might be required,
3843 		 * pretend that we need them all. It is better to be too
3844 		 * careful than too reckless.
3845 		 */
3846 		error = fget(td, sockarg.sock,
3847 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3848 		if (error != 0)
3849 			goto out;
3850 		if (fp->f_type != DTYPE_SOCKET) {
3851 			fdrop(fp, td);
3852 			error = EPERM;
3853 			goto out;
3854 		}
3855 		error = nfsrvd_addsock(fp);
3856 		fdrop(fp, td);
3857 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3858 		if (uap->argp == NULL) {
3859 			error = EINVAL;
3860 			goto out;
3861 		}
3862 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3863 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3864 			if (error == 0) {
3865 				nfsdarg.principal = onfsdarg.principal;
3866 				nfsdarg.minthreads = onfsdarg.minthreads;
3867 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3868 				nfsdarg.version = 1;
3869 				nfsdarg.addr = NULL;
3870 				nfsdarg.addrlen = 0;
3871 				nfsdarg.dnshost = NULL;
3872 				nfsdarg.dnshostlen = 0;
3873 				nfsdarg.dspath = NULL;
3874 				nfsdarg.dspathlen = 0;
3875 				nfsdarg.mdspath = NULL;
3876 				nfsdarg.mdspathlen = 0;
3877 				nfsdarg.mirrorcnt = 1;
3878 			}
3879 		} else
3880 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3881 		if (error)
3882 			goto out;
3883 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3884 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3885 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3886 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3887 		    nfsdarg.mirrorcnt >= 1 &&
3888 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3889 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3890 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3891 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3892 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3893 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3894 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3895 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3896 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3897 			if (error != 0) {
3898 				free(cp, M_TEMP);
3899 				goto out;
3900 			}
3901 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3902 			nfsdarg.addr = cp;
3903 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3904 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3905 			if (error != 0) {
3906 				free(nfsdarg.addr, M_TEMP);
3907 				free(cp, M_TEMP);
3908 				goto out;
3909 			}
3910 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3911 			nfsdarg.dnshost = cp;
3912 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3913 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3914 			if (error != 0) {
3915 				free(nfsdarg.addr, M_TEMP);
3916 				free(nfsdarg.dnshost, M_TEMP);
3917 				free(cp, M_TEMP);
3918 				goto out;
3919 			}
3920 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3921 			nfsdarg.dspath = cp;
3922 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3923 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3924 			if (error != 0) {
3925 				free(nfsdarg.addr, M_TEMP);
3926 				free(nfsdarg.dnshost, M_TEMP);
3927 				free(nfsdarg.dspath, M_TEMP);
3928 				free(cp, M_TEMP);
3929 				goto out;
3930 			}
3931 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3932 			nfsdarg.mdspath = cp;
3933 		} else {
3934 			nfsdarg.addr = NULL;
3935 			nfsdarg.addrlen = 0;
3936 			nfsdarg.dnshost = NULL;
3937 			nfsdarg.dnshostlen = 0;
3938 			nfsdarg.dspath = NULL;
3939 			nfsdarg.dspathlen = 0;
3940 			nfsdarg.mdspath = NULL;
3941 			nfsdarg.mdspathlen = 0;
3942 			nfsdarg.mirrorcnt = 1;
3943 		}
3944 		nfsd_timer(NFSD_TD_TO_VNET(td));
3945 		error = nfsrvd_nfsd(td, &nfsdarg);
3946 		callout_drain(&NFSD_VNET(nfsd_callout));
3947 		free(nfsdarg.addr, M_TEMP);
3948 		free(nfsdarg.dnshost, M_TEMP);
3949 		free(nfsdarg.dspath, M_TEMP);
3950 		free(nfsdarg.mdspath, M_TEMP);
3951 	} else if (uap->flag & NFSSVC_PNFSDS) {
3952 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3953 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3954 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3955 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3956 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3957 			    NULL);
3958 			if (error == 0)
3959 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3960 			free(cp, M_TEMP);
3961 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3962 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3963 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3964 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3965 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3966 			    NULL);
3967 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3968 			if (error == 0 && pnfsdarg.dspath != NULL) {
3969 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3970 				error = copyinstr(pnfsdarg.dspath, cp2,
3971 				    PATH_MAX + 1, NULL);
3972 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3973 				    error);
3974 			} else
3975 				cp2 = NULL;
3976 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3977 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3978 				error = copyinstr(pnfsdarg.curdspath, cp3,
3979 				    PATH_MAX + 1, NULL);
3980 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3981 				    error);
3982 			} else
3983 				cp3 = NULL;
3984 			curdvp = NULL;
3985 			fds = NULL;
3986 			if (error == 0)
3987 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3988 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3989 				    &fds);
3990 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3991 			if (error == 0) {
3992 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3993 					printf("copymr: dsdir out of range\n");
3994 					pf->dsf_dir = 0;
3995 				}
3996 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3997 				error = nfsrv_copymr(vp, nvp,
3998 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3999 				    (struct pnfsdsfile *)buf,
4000 				    buflen / sizeof(*pf), td->td_ucred, td);
4001 				vput(vp);
4002 				vput(nvp);
4003 				if (fds != NULL && error == 0) {
4004 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
4005 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
4006 					if (ret == 0) {
4007 						nfsrv_dsremove(curdvp, fname,
4008 						    td->td_ucred, td);
4009 						NFSVOPUNLOCK(curdvp);
4010 					}
4011 				}
4012 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
4013 			}
4014 			free(cp, M_TEMP);
4015 			free(cp2, M_TEMP);
4016 			free(cp3, M_TEMP);
4017 			free(buf, M_TEMP);
4018 		}
4019 	} else {
4020 		error = nfssvc_srvcall(td, uap, td->td_ucred);
4021 	}
4022 
4023 out:
4024 	NFSD_CURVNET_RESTORE();
4025 	NFSEXITCODE(error);
4026 	return (error);
4027 }
4028 
4029 static int
4030 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
4031 {
4032 	struct nfsex_args export;
4033 	struct nfsex_oldargs oexp;
4034 	struct file *fp = NULL;
4035 	int stablefd, i, len;
4036 	struct nfsd_clid adminrevoke;
4037 	struct nfsd_dumplist dumplist;
4038 	struct nfsd_dumpclients *dumpclients;
4039 	struct nfsd_dumplocklist dumplocklist;
4040 	struct nfsd_dumplocks *dumplocks;
4041 	struct nameidata nd;
4042 	vnode_t vp;
4043 	int error = EINVAL, igotlock;
4044 	struct proc *procp;
4045 	gid_t *grps;
4046 
4047 	if (uap->flag & NFSSVC_PUBLICFH) {
4048 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
4049 		    sizeof (fhandle_t));
4050 		error = copyin(uap->argp,
4051 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
4052 		if (!error)
4053 			nfs_pubfhset = 1;
4054 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
4055 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
4056 		error = copyin(uap->argp,(caddr_t)&export,
4057 		    sizeof (struct nfsex_args));
4058 		if (!error) {
4059 			grps = NULL;
4060 			if (export.export.ex_ngroups > NGROUPS_MAX ||
4061 			    export.export.ex_ngroups < 0)
4062 				error = EINVAL;
4063 			else if (export.export.ex_ngroups > 0) {
4064 				grps = malloc(export.export.ex_ngroups *
4065 				    sizeof(gid_t), M_TEMP, M_WAITOK);
4066 				error = copyin(export.export.ex_groups, grps,
4067 				    export.export.ex_ngroups * sizeof(gid_t));
4068 				export.export.ex_groups = grps;
4069 			} else
4070 				export.export.ex_groups = NULL;
4071 			if (!error)
4072 				error = nfsrv_v4rootexport(&export, cred, p);
4073 			free(grps, M_TEMP);
4074 		}
4075 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
4076 	    NFSSVC_V4ROOTEXPORT) {
4077 		error = copyin(uap->argp,(caddr_t)&oexp,
4078 		    sizeof (struct nfsex_oldargs));
4079 		if (!error) {
4080 			memset(&export.export, 0, sizeof(export.export));
4081 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
4082 			export.export.ex_root = oexp.export.ex_root;
4083 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
4084 			export.export.ex_ngroups =
4085 			    oexp.export.ex_anon.cr_ngroups;
4086 			export.export.ex_groups = NULL;
4087 			if (export.export.ex_ngroups > XU_NGROUPS ||
4088 			    export.export.ex_ngroups < 0)
4089 				error = EINVAL;
4090 			else if (export.export.ex_ngroups > 0) {
4091 				export.export.ex_groups = malloc(
4092 				    export.export.ex_ngroups * sizeof(gid_t),
4093 				    M_TEMP, M_WAITOK);
4094 				for (i = 0; i < export.export.ex_ngroups; i++)
4095 					export.export.ex_groups[i] =
4096 					    oexp.export.ex_anon.cr_groups[i];
4097 			}
4098 			export.export.ex_addr = oexp.export.ex_addr;
4099 			export.export.ex_addrlen = oexp.export.ex_addrlen;
4100 			export.export.ex_mask = oexp.export.ex_mask;
4101 			export.export.ex_masklen = oexp.export.ex_masklen;
4102 			export.export.ex_indexfile = oexp.export.ex_indexfile;
4103 			export.export.ex_numsecflavors =
4104 			    oexp.export.ex_numsecflavors;
4105 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
4106 			    export.export.ex_numsecflavors < 0)
4107 				error = EINVAL;
4108 			else {
4109 				for (i = 0; i < export.export.ex_numsecflavors;
4110 				    i++)
4111 					export.export.ex_secflavors[i] =
4112 					    oexp.export.ex_secflavors[i];
4113 			}
4114 			export.fspec = oexp.fspec;
4115 			if (error == 0)
4116 				error = nfsrv_v4rootexport(&export, cred, p);
4117 			free(export.export.ex_groups, M_TEMP);
4118 		}
4119 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
4120 		nfs_pubfhset = 0;
4121 		error = 0;
4122 	} else if (uap->flag & NFSSVC_STABLERESTART) {
4123 		error = copyin(uap->argp, (caddr_t)&stablefd,
4124 		    sizeof (int));
4125 		if (!error)
4126 			error = fp_getfvp(p, stablefd, &fp, &vp);
4127 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
4128 			error = EBADF;
4129 		if (!error && NFSD_VNET(nfsrv_numnfsd) != 0)
4130 			error = ENXIO;
4131 		if (!error) {
4132 			NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp;
4133 			nfsrv_setupstable(p);
4134 		}
4135 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
4136 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
4137 		    sizeof (struct nfsd_clid));
4138 		if (!error)
4139 			error = nfsrv_adminrevoke(&adminrevoke, p);
4140 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
4141 		error = copyin(uap->argp, (caddr_t)&dumplist,
4142 		    sizeof (struct nfsd_dumplist));
4143 		if (!error && (dumplist.ndl_size < 1 ||
4144 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
4145 			error = EPERM;
4146 		if (!error) {
4147 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
4148 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4149 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
4150 		    error = copyout(dumpclients, dumplist.ndl_list, len);
4151 		    free(dumpclients, M_TEMP);
4152 		}
4153 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
4154 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
4155 		    sizeof (struct nfsd_dumplocklist));
4156 		if (!error && (dumplocklist.ndllck_size < 1 ||
4157 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
4158 			error = EPERM;
4159 		if (!error)
4160 			error = nfsrv_lookupfilename(&nd,
4161 				dumplocklist.ndllck_fname, p);
4162 		if (!error) {
4163 			len = sizeof (struct nfsd_dumplocks) *
4164 				dumplocklist.ndllck_size;
4165 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4166 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
4167 			    dumplocklist.ndllck_size, p);
4168 			vput(nd.ni_vp);
4169 			error = copyout(dumplocks, dumplocklist.ndllck_list,
4170 			    len);
4171 			free(dumplocks, M_TEMP);
4172 		}
4173 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
4174 		procp = p->td_proc;
4175 		PROC_LOCK(procp);
4176 		nfsd_master_pid = procp->p_pid;
4177 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
4178 		nfsd_master_start = procp->p_stats->p_start;
4179 		NFSD_VNET(nfsd_master_proc) = procp;
4180 		PROC_UNLOCK(procp);
4181 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
4182 		NFSLOCKV4ROOTMUTEX();
4183 		if (!NFSD_VNET(nfsrv_suspend_nfsd)) {
4184 			/* Lock out all nfsd threads */
4185 			do {
4186 				igotlock = nfsv4_lock(
4187 				    &NFSD_VNET(nfsd_suspend_lock), 1, NULL,
4188 				    NFSV4ROOTLOCKMUTEXPTR, NULL);
4189 			} while (igotlock == 0 &&
4190 			    !NFSD_VNET(nfsrv_suspend_nfsd));
4191 			NFSD_VNET(nfsrv_suspend_nfsd) = true;
4192 		}
4193 		NFSUNLOCKV4ROOTMUTEX();
4194 		error = 0;
4195 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
4196 		NFSLOCKV4ROOTMUTEX();
4197 		if (NFSD_VNET(nfsrv_suspend_nfsd)) {
4198 			nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0);
4199 			NFSD_VNET(nfsrv_suspend_nfsd) = false;
4200 		}
4201 		NFSUNLOCKV4ROOTMUTEX();
4202 		error = 0;
4203 	}
4204 
4205 	NFSEXITCODE(error);
4206 	return (error);
4207 }
4208 
4209 /*
4210  * Check exports.
4211  * Returns 0 if ok, 1 otherwise.
4212  */
4213 int
4214 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
4215 {
4216 	int i;
4217 
4218 	if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) ||
4219 	    (NFSVNO_EXTLSCERT(exp) &&
4220 	     (nd->nd_flag & ND_TLSCERT) == 0) ||
4221 	    (NFSVNO_EXTLSCERTUSER(exp) &&
4222 	     (nd->nd_flag & ND_TLSCERTUSER) == 0)) {
4223 		if ((nd->nd_flag & ND_NFSV4) != 0)
4224 			return (NFSERR_WRONGSEC);
4225 #ifdef notnow
4226 		/* There is currently no auth_stat for this. */
4227 		else if ((nd->nd_flag & ND_TLS) == 0)
4228 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
4229 		else
4230 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
4231 #endif
4232 		else
4233 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4234 	}
4235 
4236 	/*
4237 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use
4238 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
4239 	 */
4240 	if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO)
4241 		return (0);
4242 
4243 	/*
4244 	 * This seems odd, but allow the case where the security flavor
4245 	 * list is empty. This happens when NFSv4 is traversing non-exported
4246 	 * file systems. Exported file systems should always have a non-empty
4247 	 * security flavor list.
4248 	 */
4249 	if (exp->nes_numsecflavor == 0)
4250 		return (0);
4251 
4252 	for (i = 0; i < exp->nes_numsecflavor; i++) {
4253 		/*
4254 		 * The tests for privacy and integrity must be first,
4255 		 * since ND_GSS is set for everything but AUTH_SYS.
4256 		 */
4257 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
4258 		    (nd->nd_flag & ND_GSSPRIVACY))
4259 			return (0);
4260 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
4261 		    (nd->nd_flag & ND_GSSINTEGRITY))
4262 			return (0);
4263 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
4264 		    (nd->nd_flag & ND_GSS))
4265 			return (0);
4266 		if (exp->nes_secflavors[i] == AUTH_SYS &&
4267 		    (nd->nd_flag & ND_GSS) == 0)
4268 			return (0);
4269 	}
4270 	if ((nd->nd_flag & ND_NFSV4) != 0)
4271 		return (NFSERR_WRONGSEC);
4272 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4273 }
4274 
4275 /*
4276  * Calculate a hash value for the fid in a file handle.
4277  */
4278 uint32_t
4279 nfsrv_hashfh(fhandle_t *fhp)
4280 {
4281 	uint32_t hashval;
4282 
4283 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
4284 	return (hashval);
4285 }
4286 
4287 /*
4288  * Calculate a hash value for the sessionid.
4289  */
4290 uint32_t
4291 nfsrv_hashsessionid(uint8_t *sessionid)
4292 {
4293 	uint32_t hashval;
4294 
4295 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
4296 	return (hashval);
4297 }
4298 
4299 /*
4300  * Signal the userland master nfsd to backup the stable restart file.
4301  */
4302 void
4303 nfsrv_backupstable(void)
4304 {
4305 	struct proc *procp;
4306 
4307 	if (NFSD_VNET(nfsd_master_proc) != NULL) {
4308 		procp = pfind(nfsd_master_pid);
4309 		/* Try to make sure it is the correct process. */
4310 		if (procp == NFSD_VNET(nfsd_master_proc) &&
4311 		    procp->p_stats->p_start.tv_sec ==
4312 		    nfsd_master_start.tv_sec &&
4313 		    procp->p_stats->p_start.tv_usec ==
4314 		    nfsd_master_start.tv_usec &&
4315 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
4316 			kern_psignal(procp, SIGUSR2);
4317 		else
4318 			NFSD_VNET(nfsd_master_proc) = NULL;
4319 
4320 		if (procp != NULL)
4321 			PROC_UNLOCK(procp);
4322 	}
4323 }
4324 
4325 /*
4326  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
4327  * The arguments are in a structure, so that they can be passed through
4328  * taskqueue for a kernel process to execute this function.
4329  */
4330 struct nfsrvdscreate {
4331 	int			done;
4332 	int			inprog;
4333 	struct task		tsk;
4334 	struct ucred		*tcred;
4335 	struct vnode		*dvp;
4336 	NFSPROC_T		*p;
4337 	struct pnfsdsfile	*pf;
4338 	int			err;
4339 	fhandle_t		fh;
4340 	struct vattr		va;
4341 	struct vattr		createva;
4342 };
4343 
4344 int
4345 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
4346     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
4347     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
4348 {
4349 	struct vnode *nvp;
4350 	struct nameidata named;
4351 	struct vattr va;
4352 	char *bufp;
4353 	u_long *hashp;
4354 	struct nfsnode *np;
4355 	struct nfsmount *nmp;
4356 	int error;
4357 
4358 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
4359 	    LOCKPARENT | LOCKLEAF | NOCACHE);
4360 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4361 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
4362 	named.ni_cnd.cn_nameptr = bufp;
4363 	if (fnamep != NULL) {
4364 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
4365 		named.ni_cnd.cn_namelen = strlen(bufp);
4366 	} else
4367 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
4368 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
4369 
4370 	/* Create the date file in the DS mount. */
4371 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4372 	if (error == 0) {
4373 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
4374 		vref(dvp);
4375 		VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false);
4376 		if (error == 0) {
4377 			/* Set the ownership of the file. */
4378 			error = VOP_SETATTR(nvp, nvap, tcred);
4379 			NFSD_DEBUG(4, "nfsrv_dscreate:"
4380 			    " setattr-uid=%d\n", error);
4381 			if (error != 0)
4382 				vput(nvp);
4383 		}
4384 		if (error != 0)
4385 			printf("pNFS: pnfscreate failed=%d\n", error);
4386 	} else
4387 		printf("pNFS: pnfscreate vnlock=%d\n", error);
4388 	if (error == 0) {
4389 		np = VTONFS(nvp);
4390 		nmp = VFSTONFS(nvp->v_mount);
4391 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
4392 		    != 0 || nmp->nm_nam->sa_len > sizeof(
4393 		    struct sockaddr_in6) ||
4394 		    np->n_fhp->nfh_len != NFSX_MYFH) {
4395 			printf("Bad DS file: fstype=%s salen=%d"
4396 			    " fhlen=%d\n",
4397 			    nvp->v_mount->mnt_vfc->vfc_name,
4398 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
4399 			error = ENOENT;
4400 		}
4401 
4402 		/* Set extattrs for the DS on the MDS file. */
4403 		if (error == 0) {
4404 			if (dsa != NULL) {
4405 				error = VOP_GETATTR(nvp, &va, tcred);
4406 				if (error == 0) {
4407 					dsa->dsa_filerev = va.va_filerev;
4408 					dsa->dsa_size = va.va_size;
4409 					dsa->dsa_atime = va.va_atime;
4410 					dsa->dsa_mtime = va.va_mtime;
4411 					dsa->dsa_bytes = va.va_bytes;
4412 				}
4413 			}
4414 			if (error == 0) {
4415 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4416 				    NFSX_MYFH);
4417 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4418 				    nmp->nm_nam->sa_len);
4419 				NFSBCOPY(named.ni_cnd.cn_nameptr,
4420 				    pf->dsf_filename,
4421 				    sizeof(pf->dsf_filename));
4422 			}
4423 		} else
4424 			printf("pNFS: pnfscreate can't get DS"
4425 			    " attr=%d\n", error);
4426 		if (nvpp != NULL && error == 0)
4427 			*nvpp = nvp;
4428 		else
4429 			vput(nvp);
4430 	}
4431 	nfsvno_relpathbuf(&named);
4432 	return (error);
4433 }
4434 
4435 /*
4436  * Start up the thread that will execute nfsrv_dscreate().
4437  */
4438 static void
4439 start_dscreate(void *arg, int pending)
4440 {
4441 	struct nfsrvdscreate *dsc;
4442 
4443 	dsc = (struct nfsrvdscreate *)arg;
4444 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4445 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4446 	dsc->done = 1;
4447 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4448 }
4449 
4450 /*
4451  * Create a pNFS data file on the Data Server(s).
4452  */
4453 static void
4454 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4455     NFSPROC_T *p)
4456 {
4457 	struct nfsrvdscreate *dsc, *tdsc = NULL;
4458 	struct nfsdevice *ds, *tds, *fds;
4459 	struct mount *mp;
4460 	struct pnfsdsfile *pf, *tpf;
4461 	struct pnfsdsattr dsattr;
4462 	struct vattr va;
4463 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4464 	struct nfsmount *nmp;
4465 	fhandle_t fh;
4466 	uid_t vauid;
4467 	gid_t vagid;
4468 	u_short vamode;
4469 	struct ucred *tcred;
4470 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4471 	int failpos, timo;
4472 
4473 	/* Get a DS server directory in a round-robin order. */
4474 	mirrorcnt = 1;
4475 	mp = vp->v_mount;
4476 	ds = fds = NULL;
4477 	NFSDDSLOCK();
4478 	/*
4479 	 * Search for the first entry that handles this MDS fs, but use the
4480 	 * first entry for all MDS fs's otherwise.
4481 	 */
4482 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4483 		if (tds->nfsdev_nmp != NULL) {
4484 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4485 				ds = tds;
4486 			else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4487 			    &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4488 				ds = fds = tds;
4489 				break;
4490 			}
4491 		}
4492 	}
4493 	if (ds == NULL) {
4494 		NFSDDSUNLOCK();
4495 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4496 		return;
4497 	}
4498 	i = dsdir[0] = ds->nfsdev_nextdir;
4499 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4500 	dvp[0] = ds->nfsdev_dsdir[i];
4501 	tds = TAILQ_NEXT(ds, nfsdev_list);
4502 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4503 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4504 			if (tds->nfsdev_nmp != NULL &&
4505 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4506 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4507 			      fsidcmp(&mp->mnt_stat.f_fsid,
4508 			      &tds->nfsdev_mdsfsid) == 0))) {
4509 				dsdir[mirrorcnt] = i;
4510 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4511 				mirrorcnt++;
4512 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
4513 					break;
4514 			}
4515 		}
4516 	}
4517 	/* Put at end of list to implement round-robin usage. */
4518 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4519 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4520 	NFSDDSUNLOCK();
4521 	dsc = NULL;
4522 	if (mirrorcnt > 1)
4523 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4524 		    M_WAITOK | M_ZERO);
4525 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4526 	    M_ZERO);
4527 
4528 	error = nfsvno_getfh(vp, &fh, p);
4529 	if (error == 0)
4530 		error = VOP_GETATTR(vp, &va, cred);
4531 	if (error == 0) {
4532 		/* Set the attributes for "vp" to Setattr the DS vp. */
4533 		vauid = va.va_uid;
4534 		vagid = va.va_gid;
4535 		vamode = va.va_mode;
4536 		VATTR_NULL(&va);
4537 		va.va_uid = vauid;
4538 		va.va_gid = vagid;
4539 		va.va_mode = vamode;
4540 		va.va_size = 0;
4541 	} else
4542 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4543 
4544 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4545 	    cred->cr_gid);
4546 	/* Make data file name based on FH. */
4547 	tcred = newnfs_getcred();
4548 
4549 	/*
4550 	 * Create the file on each DS mirror, using kernel process(es) for the
4551 	 * additional mirrors.
4552 	 */
4553 	failpos = -1;
4554 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4555 		tpf->dsf_dir = dsdir[i];
4556 		tdsc->tcred = tcred;
4557 		tdsc->p = p;
4558 		tdsc->pf = tpf;
4559 		tdsc->createva = *vap;
4560 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4561 		tdsc->va = va;
4562 		tdsc->dvp = dvp[i];
4563 		tdsc->done = 0;
4564 		tdsc->inprog = 0;
4565 		tdsc->err = 0;
4566 		ret = EIO;
4567 		if (nfs_pnfsiothreads != 0) {
4568 			ret = nfs_pnfsio(start_dscreate, tdsc);
4569 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4570 		}
4571 		if (ret != 0) {
4572 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4573 			    NULL, tcred, p, NULL);
4574 			if (ret != 0) {
4575 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4576 				    error));
4577 				if (failpos == -1 && nfsds_failerr(ret))
4578 					failpos = i;
4579 				else
4580 					error = ret;
4581 			}
4582 		}
4583 	}
4584 	if (error == 0) {
4585 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4586 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4587 		    &dsattr, NULL, tcred, p, NULL);
4588 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4589 			failpos = mirrorcnt - 1;
4590 			error = 0;
4591 		}
4592 	}
4593 	timo = hz / 50;		/* Wait for 20msec. */
4594 	if (timo < 1)
4595 		timo = 1;
4596 	/* Wait for kernel task(s) to complete. */
4597 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4598 		while (tdsc->inprog != 0 && tdsc->done == 0)
4599 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4600 		if (tdsc->err != 0) {
4601 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4602 				failpos = i;
4603 			else if (error == 0)
4604 				error = tdsc->err;
4605 		}
4606 	}
4607 
4608 	/*
4609 	 * If failpos has been set, that mirror has failed, so it needs
4610 	 * to be disabled.
4611 	 */
4612 	if (failpos >= 0) {
4613 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4614 		NFSLOCKMNT(nmp);
4615 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4616 		     NFSMNTP_CANCELRPCS)) == 0) {
4617 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4618 			NFSUNLOCKMNT(nmp);
4619 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4620 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4621 			    ds);
4622 			if (ds != NULL)
4623 				nfsrv_killrpcs(nmp);
4624 			NFSLOCKMNT(nmp);
4625 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4626 			wakeup(nmp);
4627 		}
4628 		NFSUNLOCKMNT(nmp);
4629 	}
4630 
4631 	NFSFREECRED(tcred);
4632 	if (error == 0) {
4633 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4634 
4635 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4636 		    mirrorcnt, nfsrv_maxpnfsmirror);
4637 		/*
4638 		 * For all mirrors that couldn't be created, fill in the
4639 		 * *pf structure, but with an IP address == 0.0.0.0.
4640 		 */
4641 		tpf = pf + mirrorcnt;
4642 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4643 			*tpf = *pf;
4644 			tpf->dsf_sin.sin_family = AF_INET;
4645 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4646 			tpf->dsf_sin.sin_addr.s_addr = 0;
4647 			tpf->dsf_sin.sin_port = 0;
4648 		}
4649 
4650 		error = vn_extattr_set(vp, IO_NODELOCKED,
4651 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4652 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4653 		if (error == 0)
4654 			error = vn_extattr_set(vp, IO_NODELOCKED,
4655 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4656 			    sizeof(dsattr), (char *)&dsattr, p);
4657 		if (error != 0)
4658 			printf("pNFS: pnfscreate setextattr=%d\n",
4659 			    error);
4660 	} else
4661 		printf("pNFS: pnfscreate=%d\n", error);
4662 	free(pf, M_TEMP);
4663 	free(dsc, M_TEMP);
4664 }
4665 
4666 /*
4667  * Get the information needed to remove the pNFS Data Server file from the
4668  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4669  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4670  */
4671 static void
4672 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4673     int *mirrorcntp, char *fname, fhandle_t *fhp)
4674 {
4675 	struct vattr va;
4676 	struct ucred *tcred;
4677 	char *buf;
4678 	int buflen, error;
4679 
4680 	dvpp[0] = NULL;
4681 	/* If not an exported regular file or not a pNFS server, just return. */
4682 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4683 	    nfsrv_devidcnt == 0)
4684 		return;
4685 
4686 	/* Check to see if this is the last hard link. */
4687 	tcred = newnfs_getcred();
4688 	error = VOP_GETATTR(vp, &va, tcred);
4689 	NFSFREECRED(tcred);
4690 	if (error != 0) {
4691 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4692 		return;
4693 	}
4694 	if (va.va_nlink > 1)
4695 		return;
4696 
4697 	error = nfsvno_getfh(vp, fhp, p);
4698 	if (error != 0) {
4699 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4700 		return;
4701 	}
4702 
4703 	buflen = 1024;
4704 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4705 	/* Get the directory vnode for the DS mount and the file handle. */
4706 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4707 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4708 	free(buf, M_TEMP);
4709 	if (error != 0)
4710 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4711 }
4712 
4713 /*
4714  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4715  * The arguments are in a structure, so that they can be passed through
4716  * taskqueue for a kernel process to execute this function.
4717  */
4718 struct nfsrvdsremove {
4719 	int			done;
4720 	int			inprog;
4721 	struct task		tsk;
4722 	struct ucred		*tcred;
4723 	struct vnode		*dvp;
4724 	NFSPROC_T		*p;
4725 	int			err;
4726 	char			fname[PNFS_FILENAME_LEN + 1];
4727 };
4728 
4729 static int
4730 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4731     NFSPROC_T *p)
4732 {
4733 	struct nameidata named;
4734 	struct vnode *nvp;
4735 	char *bufp;
4736 	u_long *hashp;
4737 	int error;
4738 
4739 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4740 	if (error != 0)
4741 		return (error);
4742 	named.ni_cnd.cn_nameiop = DELETE;
4743 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4744 	named.ni_cnd.cn_cred = tcred;
4745 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF;
4746 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4747 	named.ni_cnd.cn_nameptr = bufp;
4748 	named.ni_cnd.cn_namelen = strlen(fname);
4749 	strlcpy(bufp, fname, NAME_MAX);
4750 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4751 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4752 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4753 	if (error == 0) {
4754 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4755 		vput(nvp);
4756 	}
4757 	NFSVOPUNLOCK(dvp);
4758 	nfsvno_relpathbuf(&named);
4759 	if (error != 0)
4760 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4761 	return (error);
4762 }
4763 
4764 /*
4765  * Start up the thread that will execute nfsrv_dsremove().
4766  */
4767 static void
4768 start_dsremove(void *arg, int pending)
4769 {
4770 	struct nfsrvdsremove *dsrm;
4771 
4772 	dsrm = (struct nfsrvdsremove *)arg;
4773 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4774 	    dsrm->p);
4775 	dsrm->done = 1;
4776 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4777 }
4778 
4779 /*
4780  * Remove a pNFS data file from a Data Server.
4781  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4782  * removed to set up the dvp and fill in the FH.
4783  */
4784 static void
4785 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4786     NFSPROC_T *p)
4787 {
4788 	struct ucred *tcred;
4789 	struct nfsrvdsremove *dsrm, *tdsrm;
4790 	struct nfsdevice *ds;
4791 	struct nfsmount *nmp;
4792 	int failpos, i, ret, timo;
4793 
4794 	tcred = newnfs_getcred();
4795 	dsrm = NULL;
4796 	if (mirrorcnt > 1)
4797 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4798 	/*
4799 	 * Remove the file on each DS mirror, using kernel process(es) for the
4800 	 * additional mirrors.
4801 	 */
4802 	failpos = -1;
4803 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4804 		tdsrm->tcred = tcred;
4805 		tdsrm->p = p;
4806 		tdsrm->dvp = dvp[i];
4807 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4808 		tdsrm->inprog = 0;
4809 		tdsrm->done = 0;
4810 		tdsrm->err = 0;
4811 		ret = EIO;
4812 		if (nfs_pnfsiothreads != 0) {
4813 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4814 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4815 		}
4816 		if (ret != 0) {
4817 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4818 			if (failpos == -1 && nfsds_failerr(ret))
4819 				failpos = i;
4820 		}
4821 	}
4822 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4823 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4824 		failpos = mirrorcnt - 1;
4825 	timo = hz / 50;		/* Wait for 20msec. */
4826 	if (timo < 1)
4827 		timo = 1;
4828 	/* Wait for kernel task(s) to complete. */
4829 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4830 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4831 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4832 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4833 			failpos = i;
4834 	}
4835 
4836 	/*
4837 	 * If failpos has been set, that mirror has failed, so it needs
4838 	 * to be disabled.
4839 	 */
4840 	if (failpos >= 0) {
4841 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4842 		NFSLOCKMNT(nmp);
4843 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4844 		     NFSMNTP_CANCELRPCS)) == 0) {
4845 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4846 			NFSUNLOCKMNT(nmp);
4847 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4848 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4849 			    ds);
4850 			if (ds != NULL)
4851 				nfsrv_killrpcs(nmp);
4852 			NFSLOCKMNT(nmp);
4853 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4854 			wakeup(nmp);
4855 		}
4856 		NFSUNLOCKMNT(nmp);
4857 	}
4858 
4859 	/* Get rid all layouts for the file. */
4860 	nfsrv_freefilelayouts(fhp);
4861 
4862 	NFSFREECRED(tcred);
4863 	free(dsrm, M_TEMP);
4864 }
4865 
4866 /*
4867  * Generate a file name based on the file handle and put it in *bufp.
4868  * Return the number of bytes generated.
4869  */
4870 static int
4871 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4872 {
4873 	int i;
4874 	uint8_t *cp;
4875 	const uint8_t *hexdigits = "0123456789abcdef";
4876 
4877 	cp = (uint8_t *)fhp;
4878 	for (i = 0; i < sizeof(*fhp); i++) {
4879 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4880 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4881 	}
4882 	bufp[2 * i] = '\0';
4883 	return (2 * i);
4884 }
4885 
4886 /*
4887  * Update the Metadata file's attributes from the DS file when a Read/Write
4888  * layout is returned.
4889  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4890  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4891  */
4892 int
4893 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4894 {
4895 	struct ucred *tcred;
4896 	int error;
4897 
4898 	/* Do this as root so that it won't fail with EACCES. */
4899 	tcred = newnfs_getcred();
4900 	error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4901 	    NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4902 	NFSFREECRED(tcred);
4903 	return (error);
4904 }
4905 
4906 /*
4907  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4908  */
4909 static int
4910 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4911     NFSPROC_T *p)
4912 {
4913 	int error;
4914 
4915 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4916 	    NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4917 	return (error);
4918 }
4919 
4920 static int
4921 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4922     struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4923     struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4924     off_t *offp, int content, bool *eofp)
4925 {
4926 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4927 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4928 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4929 	struct nfsdevice *ds;
4930 	struct pnfsdsattr dsattr;
4931 	struct opnfsdsattr odsattr;
4932 	char *buf;
4933 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4934 
4935 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4936 	/*
4937 	 * If not a regular file, not exported or not a pNFS server,
4938 	 * just return ENOENT.
4939 	 */
4940 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4941 	    nfsrv_devidcnt == 0)
4942 		return (ENOENT);
4943 
4944 	buflen = 1024;
4945 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4946 	error = 0;
4947 
4948 	/*
4949 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4950 	 * from the MetaData file's extended attribute.
4951 	 */
4952 	if (ioproc == NFSPROC_GETATTR) {
4953 		error = vn_extattr_get(vp, IO_NODELOCKED,
4954 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4955 		    p);
4956 		if (error == 0) {
4957 			if (buflen == sizeof(odsattr)) {
4958 				NFSBCOPY(buf, &odsattr, buflen);
4959 				nap->na_filerev = odsattr.dsa_filerev;
4960 				nap->na_size = odsattr.dsa_size;
4961 				nap->na_atime = odsattr.dsa_atime;
4962 				nap->na_mtime = odsattr.dsa_mtime;
4963 				/*
4964 				 * Fake na_bytes by rounding up na_size.
4965 				 * Since we don't know the block size, just
4966 				 * use BLKDEV_IOSIZE.
4967 				 */
4968 				nap->na_bytes = (odsattr.dsa_size +
4969 				    BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4970 			} else if (buflen == sizeof(dsattr)) {
4971 				NFSBCOPY(buf, &dsattr, buflen);
4972 				nap->na_filerev = dsattr.dsa_filerev;
4973 				nap->na_size = dsattr.dsa_size;
4974 				nap->na_atime = dsattr.dsa_atime;
4975 				nap->na_mtime = dsattr.dsa_mtime;
4976 				nap->na_bytes = dsattr.dsa_bytes;
4977 			} else
4978 				error = ENXIO;
4979 		}
4980 		if (error == 0) {
4981 			/*
4982 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4983 			 * returns 0, just return now.  nfsrv_checkdsattr()
4984 			 * returns 0 if there is no Read/Write layout
4985 			 * plus either an Open/Write_access or Write
4986 			 * delegation issued to a client for the file.
4987 			 */
4988 			if (nfsrv_pnfsgetdsattr == 0 ||
4989 			    nfsrv_checkdsattr(vp, p) == 0) {
4990 				free(buf, M_TEMP);
4991 				return (error);
4992 			}
4993 		}
4994 
4995 		/*
4996 		 * Clear ENOATTR so the code below will attempt to do a
4997 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4998 		 * the extended attribute.
4999 		 */
5000 		if (error == ENOATTR)
5001 			error = 0;
5002 	}
5003 
5004 	origmircnt = -1;
5005 	trycnt = 0;
5006 tryagain:
5007 	if (error == 0) {
5008 		buflen = 1024;
5009 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
5010 		    LK_EXCLUSIVE)
5011 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
5012 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
5013 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
5014 		    NULL, NULL);
5015 		if (error == 0) {
5016 			for (i = 0; i < mirrorcnt; i++)
5017 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
5018 		} else
5019 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
5020 	} else
5021 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
5022 	if (error == 0) {
5023 		failpos = -1;
5024 		if (origmircnt == -1)
5025 			origmircnt = mirrorcnt;
5026 		/*
5027 		 * If failpos is set to a mirror#, then that mirror has
5028 		 * failed and will be disabled. For Read, Getattr and Seek, the
5029 		 * function only tries one mirror, so if that mirror has
5030 		 * failed, it will need to be retried. As such, increment
5031 		 * tryitagain for these cases.
5032 		 * For Write, Setattr and Setacl, the function tries all
5033 		 * mirrors and will not return an error for the case where
5034 		 * one mirror has failed. For these cases, the functioning
5035 		 * mirror(s) will have been modified, so a retry isn't
5036 		 * necessary. These functions will set failpos for the
5037 		 * failed mirror#.
5038 		 */
5039 		if (ioproc == NFSPROC_READDS) {
5040 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
5041 			    mpp, mpp2);
5042 			if (nfsds_failerr(error) && mirrorcnt > 1) {
5043 				/*
5044 				 * Setting failpos will cause the mirror
5045 				 * to be disabled and then a retry of this
5046 				 * read is required.
5047 				 */
5048 				failpos = 0;
5049 				error = 0;
5050 				trycnt++;
5051 			}
5052 		} else if (ioproc == NFSPROC_WRITEDS)
5053 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
5054 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
5055 		else if (ioproc == NFSPROC_SETATTR)
5056 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
5057 			    mirrorcnt, nap, &failpos);
5058 		else if (ioproc == NFSPROC_SETACL)
5059 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
5060 			    mirrorcnt, aclp, &failpos);
5061 		else if (ioproc == NFSPROC_SEEKDS) {
5062 			error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
5063 			    p, nmp[0]);
5064 			if (nfsds_failerr(error) && mirrorcnt > 1) {
5065 				/*
5066 				 * Setting failpos will cause the mirror
5067 				 * to be disabled and then a retry of this
5068 				 * read is required.
5069 				 */
5070 				failpos = 0;
5071 				error = 0;
5072 				trycnt++;
5073 			}
5074 		} else if (ioproc == NFSPROC_ALLOCATE)
5075 			error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
5076 			    &nmp[0], mirrorcnt, &failpos);
5077 		else if (ioproc == NFSPROC_DEALLOCATE)
5078 			error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p,
5079 			    vp, &nmp[0], mirrorcnt, &failpos);
5080 		else {
5081 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
5082 			    vp, nmp[mirrorcnt - 1], nap);
5083 			if (nfsds_failerr(error) && mirrorcnt > 1) {
5084 				/*
5085 				 * Setting failpos will cause the mirror
5086 				 * to be disabled and then a retry of this
5087 				 * getattr is required.
5088 				 */
5089 				failpos = mirrorcnt - 1;
5090 				error = 0;
5091 				trycnt++;
5092 			}
5093 		}
5094 		ds = NULL;
5095 		if (failpos >= 0) {
5096 			failnmp = nmp[failpos];
5097 			NFSLOCKMNT(failnmp);
5098 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
5099 			     NFSMNTP_CANCELRPCS)) == 0) {
5100 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
5101 				NFSUNLOCKMNT(failnmp);
5102 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
5103 				    failnmp, p);
5104 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
5105 				    failpos, ds);
5106 				if (ds != NULL)
5107 					nfsrv_killrpcs(failnmp);
5108 				NFSLOCKMNT(failnmp);
5109 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
5110 				wakeup(failnmp);
5111 			}
5112 			NFSUNLOCKMNT(failnmp);
5113 		}
5114 		for (i = 0; i < mirrorcnt; i++)
5115 			NFSVOPUNLOCK(dvp[i]);
5116 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
5117 		    trycnt);
5118 		/* Try the Read/Getattr again if a mirror was deleted. */
5119 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
5120 			goto tryagain;
5121 	} else {
5122 		/* Return ENOENT for any Extended Attribute error. */
5123 		error = ENOENT;
5124 	}
5125 	free(buf, M_TEMP);
5126 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
5127 	return (error);
5128 }
5129 
5130 /*
5131  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
5132  * attribute.
5133  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
5134  *           to be checked.  If it points to a NULL nmp, then it returns
5135  *           a suitable destination.
5136  * curnmp - If non-NULL, it is the source mount for the copy.
5137  */
5138 int
5139 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
5140     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
5141     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
5142     struct nfsmount *curnmp, int *ippos, int *dsdirp)
5143 {
5144 	struct vnode *dvp, *nvp = NULL, **tdvpp;
5145 	struct mount *mp;
5146 	struct nfsmount *nmp, *newnmp;
5147 	struct sockaddr *sad;
5148 	struct sockaddr_in *sin;
5149 	struct nfsdevice *ds, *tds, *fndds;
5150 	struct pnfsdsfile *pf;
5151 	uint32_t dsdir;
5152 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
5153 
5154 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
5155 	*mirrorcntp = 1;
5156 	tdvpp = dvpp;
5157 	if (nvpp != NULL)
5158 		*nvpp = NULL;
5159 	if (dvpp != NULL)
5160 		*dvpp = NULL;
5161 	if (ippos != NULL)
5162 		*ippos = -1;
5163 	if (newnmpp != NULL)
5164 		newnmp = *newnmpp;
5165 	else
5166 		newnmp = NULL;
5167 	mp = vp->v_mount;
5168 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5169 	    "pnfsd.dsfile", buflenp, buf, p);
5170 	mirrorcnt = *buflenp / sizeof(*pf);
5171 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
5172 	    *buflenp != sizeof(*pf) * mirrorcnt))
5173 		error = ENOATTR;
5174 
5175 	pf = (struct pnfsdsfile *)buf;
5176 	/* If curnmp != NULL, check for a match in the mirror list. */
5177 	if (curnmp != NULL && error == 0) {
5178 		fnd = 0;
5179 		for (i = 0; i < mirrorcnt; i++, pf++) {
5180 			sad = (struct sockaddr *)&pf->dsf_sin;
5181 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
5182 				if (ippos != NULL)
5183 					*ippos = i;
5184 				fnd = 1;
5185 				break;
5186 			}
5187 		}
5188 		if (fnd == 0)
5189 			error = ENXIO;
5190 	}
5191 
5192 	gotone = 0;
5193 	pf = (struct pnfsdsfile *)buf;
5194 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
5195 	    error);
5196 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
5197 		fhiszero = 0;
5198 		sad = (struct sockaddr *)&pf->dsf_sin;
5199 		sin = &pf->dsf_sin;
5200 		dsdir = pf->dsf_dir;
5201 		if (dsdir >= nfsrv_dsdirsize) {
5202 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
5203 			error = ENOATTR;
5204 		} else if (nvpp != NULL && newnmp != NULL &&
5205 		    nfsaddr2_match(sad, newnmp->nm_nam))
5206 			error = EEXIST;
5207 		if (error == 0) {
5208 			if (ippos != NULL && curnmp == NULL &&
5209 			    sad->sa_family == AF_INET &&
5210 			    sin->sin_addr.s_addr == 0)
5211 				*ippos = i;
5212 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
5213 				fhiszero = 1;
5214 			/* Use the socket address to find the mount point. */
5215 			fndds = NULL;
5216 			NFSDDSLOCK();
5217 			/* Find a match for the IP address. */
5218 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5219 				if (ds->nfsdev_nmp != NULL) {
5220 					dvp = ds->nfsdev_dvp;
5221 					nmp = VFSTONFS(dvp->v_mount);
5222 					if (nmp != ds->nfsdev_nmp)
5223 						printf("different2 nmp %p %p\n",
5224 						    nmp, ds->nfsdev_nmp);
5225 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
5226 						fndds = ds;
5227 						break;
5228 					}
5229 				}
5230 			}
5231 			if (fndds != NULL && newnmpp != NULL &&
5232 			    newnmp == NULL) {
5233 				/* Search for a place to make a mirror copy. */
5234 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
5235 				    nfsdev_list) {
5236 					if (tds->nfsdev_nmp != NULL &&
5237 					    fndds != tds &&
5238 					    ((tds->nfsdev_mdsisset == 0 &&
5239 					      fndds->nfsdev_mdsisset == 0) ||
5240 					     (tds->nfsdev_mdsisset != 0 &&
5241 					      fndds->nfsdev_mdsisset != 0 &&
5242 					      fsidcmp(&tds->nfsdev_mdsfsid,
5243 					      &mp->mnt_stat.f_fsid) == 0))) {
5244 						*newnmpp = tds->nfsdev_nmp;
5245 						break;
5246 					}
5247 				}
5248 				if (tds != NULL) {
5249 					/*
5250 					 * Move this entry to the end of the
5251 					 * list, so it won't be selected as
5252 					 * easily the next time.
5253 					 */
5254 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
5255 					    nfsdev_list);
5256 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
5257 					    nfsdev_list);
5258 				}
5259 			}
5260 			NFSDDSUNLOCK();
5261 			if (fndds != NULL) {
5262 				dvp = fndds->nfsdev_dsdir[dsdir];
5263 				if (lktype != 0 || fhiszero != 0 ||
5264 				    (nvpp != NULL && *nvpp == NULL)) {
5265 					if (fhiszero != 0)
5266 						error = vn_lock(dvp,
5267 						    LK_EXCLUSIVE);
5268 					else if (lktype != 0)
5269 						error = vn_lock(dvp, lktype);
5270 					else
5271 						error = vn_lock(dvp, LK_SHARED);
5272 					/*
5273 					 * If the file handle is all 0's, try to
5274 					 * do a Lookup against the DS to acquire
5275 					 * it.
5276 					 * If dvpp == NULL or the Lookup fails,
5277 					 * unlock dvp after the call.
5278 					 */
5279 					if (error == 0 && (fhiszero != 0 ||
5280 					    (nvpp != NULL && *nvpp == NULL))) {
5281 						error = nfsrv_pnfslookupds(vp,
5282 						    dvp, pf, &nvp, p);
5283 						if (error == 0) {
5284 							if (fhiszero != 0)
5285 								nfsrv_pnfssetfh(
5286 								    vp, pf,
5287 								    devid,
5288 								    fnamep,
5289 								    nvp, p);
5290 							if (nvpp != NULL &&
5291 							    *nvpp == NULL) {
5292 								*nvpp = nvp;
5293 								*dsdirp = dsdir;
5294 							} else
5295 								vput(nvp);
5296 						}
5297 						if (error != 0 || lktype == 0)
5298 							NFSVOPUNLOCK(dvp);
5299 					}
5300 				}
5301 				if (error == 0) {
5302 					gotone++;
5303 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
5304 					if (devid != NULL) {
5305 						NFSBCOPY(fndds->nfsdev_deviceid,
5306 						    devid, NFSX_V4DEVICEID);
5307 						devid += NFSX_V4DEVICEID;
5308 					}
5309 					if (dvpp != NULL)
5310 						*tdvpp++ = dvp;
5311 					if (fhp != NULL)
5312 						NFSBCOPY(&pf->dsf_fh, fhp++,
5313 						    NFSX_MYFH);
5314 					if (fnamep != NULL && gotone == 1)
5315 						strlcpy(fnamep,
5316 						    pf->dsf_filename,
5317 						    sizeof(pf->dsf_filename));
5318 				} else
5319 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
5320 					    "err=%d\n", error);
5321 			}
5322 		}
5323 	}
5324 	if (error == 0 && gotone == 0)
5325 		error = ENOENT;
5326 
5327 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
5328 	    error);
5329 	if (error == 0)
5330 		*mirrorcntp = gotone;
5331 	else {
5332 		if (gotone > 0 && dvpp != NULL) {
5333 			/*
5334 			 * If the error didn't occur on the first one and
5335 			 * dvpp != NULL, the one(s) prior to the failure will
5336 			 * have locked dvp's that need to be unlocked.
5337 			 */
5338 			for (i = 0; i < gotone; i++) {
5339 				NFSVOPUNLOCK(*dvpp);
5340 				*dvpp++ = NULL;
5341 			}
5342 		}
5343 		/*
5344 		 * If it found the vnode to be copied from before a failure,
5345 		 * it needs to be vput()'d.
5346 		 */
5347 		if (nvpp != NULL && *nvpp != NULL) {
5348 			vput(*nvpp);
5349 			*nvpp = NULL;
5350 		}
5351 	}
5352 	return (error);
5353 }
5354 
5355 /*
5356  * Set the extended attribute for the Change attribute.
5357  */
5358 static int
5359 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
5360 {
5361 	struct pnfsdsattr dsattr;
5362 	int error;
5363 
5364 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
5365 	dsattr.dsa_filerev = nap->na_filerev;
5366 	dsattr.dsa_size = nap->na_size;
5367 	dsattr.dsa_atime = nap->na_atime;
5368 	dsattr.dsa_mtime = nap->na_mtime;
5369 	dsattr.dsa_bytes = nap->na_bytes;
5370 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5371 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
5372 	if (error != 0)
5373 		printf("pNFS: setextattr=%d\n", error);
5374 	return (error);
5375 }
5376 
5377 static int
5378 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5379     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
5380 {
5381 	uint32_t *tl;
5382 	struct nfsrv_descript *nd;
5383 	nfsv4stateid_t st;
5384 	struct mbuf *m, *m2;
5385 	int error = 0, retlen, tlen, trimlen;
5386 
5387 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
5388 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5389 	*mpp = NULL;
5390 	/*
5391 	 * Use a stateid where other is an alternating 01010 pattern and
5392 	 * seqid is 0xffffffff.  This value is not defined as special by
5393 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5394 	 * MDS->DS proxy operation.
5395 	 */
5396 	st.other[0] = 0x55555555;
5397 	st.other[1] = 0x55555555;
5398 	st.other[2] = 0x55555555;
5399 	st.seqid = 0xffffffff;
5400 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5401 	    NULL, NULL, 0, 0, cred);
5402 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5403 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
5404 	txdr_hyper(off, tl);
5405 	*(tl + 2) = txdr_unsigned(len);
5406 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5407 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5408 	if (error != 0) {
5409 		free(nd, M_TEMP);
5410 		return (error);
5411 	}
5412 	if (nd->nd_repstat == 0) {
5413 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5414 		NFSM_STRSIZ(retlen, len);
5415 		if (retlen > 0) {
5416 			/* Trim off the pre-data XDR from the mbuf chain. */
5417 			m = nd->nd_mrep;
5418 			while (m != NULL && m != nd->nd_md) {
5419 				if (m->m_next == nd->nd_md) {
5420 					m->m_next = NULL;
5421 					m_freem(nd->nd_mrep);
5422 					nd->nd_mrep = m = nd->nd_md;
5423 				} else
5424 					m = m->m_next;
5425 			}
5426 			if (m == NULL) {
5427 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5428 				error = ENOENT;
5429 				goto nfsmout;
5430 			}
5431 
5432 			/*
5433 			 * Now, adjust first mbuf so that any XDR before the
5434 			 * read data is skipped over.
5435 			 */
5436 			trimlen = nd->nd_dpos - mtod(m, char *);
5437 			if (trimlen > 0) {
5438 				m->m_len -= trimlen;
5439 				NFSM_DATAP(m, trimlen);
5440 			}
5441 
5442 			/*
5443 			 * Truncate the mbuf chain at retlen bytes of data,
5444 			 * plus XDR padding that brings the length up to a
5445 			 * multiple of 4.
5446 			 */
5447 			tlen = NFSM_RNDUP(retlen);
5448 			do {
5449 				if (m->m_len >= tlen) {
5450 					m->m_len = tlen;
5451 					tlen = 0;
5452 					m2 = m->m_next;
5453 					m->m_next = NULL;
5454 					m_freem(m2);
5455 					break;
5456 				}
5457 				tlen -= m->m_len;
5458 				m = m->m_next;
5459 			} while (m != NULL);
5460 			if (tlen > 0) {
5461 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5462 				error = ENOENT;
5463 				goto nfsmout;
5464 			}
5465 			*mpp = nd->nd_mrep;
5466 			*mpendp = m;
5467 			nd->nd_mrep = NULL;
5468 		}
5469 	} else
5470 		error = nd->nd_repstat;
5471 nfsmout:
5472 	/* If nd->nd_mrep is already NULL, this is a no-op. */
5473 	m_freem(nd->nd_mrep);
5474 	free(nd, M_TEMP);
5475 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5476 	return (error);
5477 }
5478 
5479 /*
5480  * Do a write RPC on a DS data file, using this structure for the arguments,
5481  * so that this function can be executed by a separate kernel process.
5482  */
5483 struct nfsrvwritedsdorpc {
5484 	int			done;
5485 	int			inprog;
5486 	struct task		tsk;
5487 	fhandle_t		fh;
5488 	off_t			off;
5489 	int			len;
5490 	struct nfsmount		*nmp;
5491 	struct ucred		*cred;
5492 	NFSPROC_T		*p;
5493 	struct mbuf		*m;
5494 	int			err;
5495 };
5496 
5497 static int
5498 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5499     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5500 {
5501 	uint32_t *tl;
5502 	struct nfsrv_descript *nd;
5503 	nfsattrbit_t attrbits;
5504 	nfsv4stateid_t st;
5505 	int commit, error, retlen;
5506 
5507 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5508 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5509 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5510 
5511 	/*
5512 	 * Use a stateid where other is an alternating 01010 pattern and
5513 	 * seqid is 0xffffffff.  This value is not defined as special by
5514 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5515 	 * MDS->DS proxy operation.
5516 	 */
5517 	st.other[0] = 0x55555555;
5518 	st.other[1] = 0x55555555;
5519 	st.other[2] = 0x55555555;
5520 	st.seqid = 0xffffffff;
5521 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5522 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5523 	txdr_hyper(off, tl);
5524 	tl += 2;
5525 	/*
5526 	 * Do all writes FileSync, since the server doesn't hold onto dirty
5527 	 * buffers.  Since clients should be accessing the DS servers directly
5528 	 * using the pNFS layouts, this just needs to work correctly as a
5529 	 * fallback.
5530 	 */
5531 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5532 	*tl = txdr_unsigned(len);
5533 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5534 
5535 	/* Put data in mbuf chain. */
5536 	nd->nd_mb->m_next = m;
5537 
5538 	/* Set nd_mb and nd_bpos to end of data. */
5539 	while (m->m_next != NULL)
5540 		m = m->m_next;
5541 	nd->nd_mb = m;
5542 	nfsm_set(nd, m->m_len);
5543 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5544 
5545 	/* Do a Getattr for the attributes that change upon writing. */
5546 	NFSZERO_ATTRBIT(&attrbits);
5547 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5548 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5549 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5550 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5551 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5552 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5553 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5554 	(void) nfsrv_putattrbit(nd, &attrbits);
5555 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5556 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5557 	if (error != 0) {
5558 		free(nd, M_TEMP);
5559 		return (error);
5560 	}
5561 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5562 	/* Get rid of weak cache consistency data for now. */
5563 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5564 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5565 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5566 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
5567 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5568 		if (error != 0)
5569 			goto nfsmout;
5570 		/*
5571 		 * Get rid of Op# and status for next op.
5572 		 */
5573 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5574 		if (*++tl != 0)
5575 			nd->nd_flag |= ND_NOMOREDATA;
5576 	}
5577 	if (nd->nd_repstat == 0) {
5578 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5579 		retlen = fxdr_unsigned(int, *tl++);
5580 		commit = fxdr_unsigned(int, *tl);
5581 		if (commit != NFSWRITE_FILESYNC)
5582 			error = NFSERR_IO;
5583 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5584 		    retlen, commit, error);
5585 	} else
5586 		error = nd->nd_repstat;
5587 	/* We have no use for the Write Verifier since we use FileSync. */
5588 
5589 	/*
5590 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5591 	 * on the Metadata file, so its attributes will be what the file's
5592 	 * would be if it had been written.
5593 	 */
5594 	if (error == 0) {
5595 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5596 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5597 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
5598 	}
5599 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5600 nfsmout:
5601 	m_freem(nd->nd_mrep);
5602 	free(nd, M_TEMP);
5603 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5604 	return (error);
5605 }
5606 
5607 /*
5608  * Start up the thread that will execute nfsrv_writedsdorpc().
5609  */
5610 static void
5611 start_writedsdorpc(void *arg, int pending)
5612 {
5613 	struct nfsrvwritedsdorpc *drpc;
5614 
5615 	drpc = (struct nfsrvwritedsdorpc *)arg;
5616 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5617 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5618 	drpc->done = 1;
5619 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5620 }
5621 
5622 static int
5623 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5624     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5625     struct mbuf **mpp, char *cp, int *failposp)
5626 {
5627 	struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5628 	struct nfsvattr na;
5629 	struct mbuf *m;
5630 	int error, i, offs, ret, timo;
5631 
5632 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5633 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5634 	drpc = NULL;
5635 	if (mirrorcnt > 1)
5636 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5637 		    M_WAITOK);
5638 
5639 	/* Calculate offset in mbuf chain that data starts. */
5640 	offs = cp - mtod(*mpp, char *);
5641 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5642 
5643 	/*
5644 	 * Do the write RPC for every DS, using a separate kernel process
5645 	 * for every DS except the last one.
5646 	 */
5647 	error = 0;
5648 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5649 		tdrpc->done = 0;
5650 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5651 		tdrpc->off = off;
5652 		tdrpc->len = len;
5653 		tdrpc->nmp = *nmpp;
5654 		tdrpc->cred = cred;
5655 		tdrpc->p = p;
5656 		tdrpc->inprog = 0;
5657 		tdrpc->err = 0;
5658 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5659 		ret = EIO;
5660 		if (nfs_pnfsiothreads != 0) {
5661 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5662 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5663 			    ret);
5664 		}
5665 		if (ret != 0) {
5666 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5667 			    tdrpc->m, cred, p);
5668 			if (nfsds_failerr(ret) && *failposp == -1)
5669 				*failposp = i;
5670 			else if (error == 0 && ret != 0)
5671 				error = ret;
5672 		}
5673 		nmpp++;
5674 		fhp++;
5675 	}
5676 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5677 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5678 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5679 		*failposp = mirrorcnt - 1;
5680 	else if (error == 0 && ret != 0)
5681 		error = ret;
5682 	if (error == 0)
5683 		error = nfsrv_setextattr(vp, &na, p);
5684 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5685 	tdrpc = drpc;
5686 	timo = hz / 50;		/* Wait for 20msec. */
5687 	if (timo < 1)
5688 		timo = 1;
5689 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5690 		/* Wait for RPCs on separate threads to complete. */
5691 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5692 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5693 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5694 			*failposp = i;
5695 		else if (error == 0 && tdrpc->err != 0)
5696 			error = tdrpc->err;
5697 	}
5698 	free(drpc, M_TEMP);
5699 	return (error);
5700 }
5701 
5702 /*
5703  * Do a allocate RPC on a DS data file, using this structure for the arguments,
5704  * so that this function can be executed by a separate kernel process.
5705  */
5706 struct nfsrvallocatedsdorpc {
5707 	int			done;
5708 	int			inprog;
5709 	struct task		tsk;
5710 	fhandle_t		fh;
5711 	off_t			off;
5712 	off_t			len;
5713 	struct nfsmount		*nmp;
5714 	struct ucred		*cred;
5715 	NFSPROC_T		*p;
5716 	int			err;
5717 };
5718 
5719 static int
5720 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5721     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5722 {
5723 	uint32_t *tl;
5724 	struct nfsrv_descript *nd;
5725 	nfsattrbit_t attrbits;
5726 	nfsv4stateid_t st;
5727 	int error;
5728 
5729 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5730 	nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5731 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5732 
5733 	/*
5734 	 * Use a stateid where other is an alternating 01010 pattern and
5735 	 * seqid is 0xffffffff.  This value is not defined as special by
5736 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5737 	 * MDS->DS proxy operation.
5738 	 */
5739 	st.other[0] = 0x55555555;
5740 	st.other[1] = 0x55555555;
5741 	st.other[2] = 0x55555555;
5742 	st.seqid = 0xffffffff;
5743 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5744 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5745 	txdr_hyper(off, tl); tl += 2;
5746 	txdr_hyper(len, tl); tl += 2;
5747 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5748 
5749 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5750 	NFSGETATTR_ATTRBIT(&attrbits);
5751 	nfsrv_putattrbit(nd, &attrbits);
5752 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5753 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5754 	if (error != 0) {
5755 		free(nd, M_TEMP);
5756 		return (error);
5757 	}
5758 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5759 	    nd->nd_repstat);
5760 	if (nd->nd_repstat == 0) {
5761 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5762 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5763 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
5764 	} else
5765 		error = nd->nd_repstat;
5766 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5767 nfsmout:
5768 	m_freem(nd->nd_mrep);
5769 	free(nd, M_TEMP);
5770 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5771 	return (error);
5772 }
5773 
5774 /*
5775  * Start up the thread that will execute nfsrv_allocatedsdorpc().
5776  */
5777 static void
5778 start_allocatedsdorpc(void *arg, int pending)
5779 {
5780 	struct nfsrvallocatedsdorpc *drpc;
5781 
5782 	drpc = (struct nfsrvallocatedsdorpc *)arg;
5783 	drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5784 	    drpc->len, NULL, drpc->cred, drpc->p);
5785 	drpc->done = 1;
5786 	NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5787 }
5788 
5789 static int
5790 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5791     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5792     int *failposp)
5793 {
5794 	struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5795 	struct nfsvattr na;
5796 	int error, i, ret, timo;
5797 
5798 	NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5799 	drpc = NULL;
5800 	if (mirrorcnt > 1)
5801 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5802 		    M_WAITOK);
5803 
5804 	/*
5805 	 * Do the allocate RPC for every DS, using a separate kernel process
5806 	 * for every DS except the last one.
5807 	 */
5808 	error = 0;
5809 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5810 		tdrpc->done = 0;
5811 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5812 		tdrpc->off = off;
5813 		tdrpc->len = len;
5814 		tdrpc->nmp = *nmpp;
5815 		tdrpc->cred = cred;
5816 		tdrpc->p = p;
5817 		tdrpc->inprog = 0;
5818 		tdrpc->err = 0;
5819 		ret = EIO;
5820 		if (nfs_pnfsiothreads != 0) {
5821 			ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5822 			NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5823 			    ret);
5824 		}
5825 		if (ret != 0) {
5826 			ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5827 			    cred, p);
5828 			if (nfsds_failerr(ret) && *failposp == -1)
5829 				*failposp = i;
5830 			else if (error == 0 && ret != 0)
5831 				error = ret;
5832 		}
5833 		nmpp++;
5834 		fhp++;
5835 	}
5836 	ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5837 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5838 		*failposp = mirrorcnt - 1;
5839 	else if (error == 0 && ret != 0)
5840 		error = ret;
5841 	if (error == 0)
5842 		error = nfsrv_setextattr(vp, &na, p);
5843 	NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5844 	tdrpc = drpc;
5845 	timo = hz / 50;		/* Wait for 20msec. */
5846 	if (timo < 1)
5847 		timo = 1;
5848 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5849 		/* Wait for RPCs on separate threads to complete. */
5850 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5851 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5852 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5853 			*failposp = i;
5854 		else if (error == 0 && tdrpc->err != 0)
5855 			error = tdrpc->err;
5856 	}
5857 	free(drpc, M_TEMP);
5858 	return (error);
5859 }
5860 
5861 /*
5862  * Do a deallocate RPC on a DS data file, using this structure for the
5863  * arguments, so that this function can be executed by a separate kernel
5864  * process.
5865  */
5866 struct nfsrvdeallocatedsdorpc {
5867 	int			done;
5868 	int			inprog;
5869 	struct task		tsk;
5870 	fhandle_t		fh;
5871 	off_t			off;
5872 	off_t			len;
5873 	struct nfsmount		*nmp;
5874 	struct ucred		*cred;
5875 	NFSPROC_T		*p;
5876 	int			err;
5877 };
5878 
5879 static int
5880 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5881     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5882 {
5883 	uint32_t *tl;
5884 	struct nfsrv_descript *nd;
5885 	nfsattrbit_t attrbits;
5886 	nfsv4stateid_t st;
5887 	int error;
5888 
5889 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5890 	nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp,
5891 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5892 
5893 	/*
5894 	 * Use a stateid where other is an alternating 01010 pattern and
5895 	 * seqid is 0xffffffff.  This value is not defined as special by
5896 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5897 	 * MDS->DS proxy operation.
5898 	 */
5899 	st.other[0] = 0x55555555;
5900 	st.other[1] = 0x55555555;
5901 	st.other[2] = 0x55555555;
5902 	st.seqid = 0xffffffff;
5903 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5904 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5905 	txdr_hyper(off, tl); tl += 2;
5906 	txdr_hyper(len, tl); tl += 2;
5907 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len);
5908 
5909 	/* Do a Getattr for the attributes that change upon writing. */
5910 	NFSZERO_ATTRBIT(&attrbits);
5911 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5912 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5913 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5914 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5915 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5916 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5917 	nfsrv_putattrbit(nd, &attrbits);
5918 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5919 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5920 	if (error != 0) {
5921 		free(nd, M_TEMP);
5922 		return (error);
5923 	}
5924 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n",
5925 	    nd->nd_repstat);
5926 	/* Get rid of weak cache consistency data for now. */
5927 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5928 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5929 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5930 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
5931 		NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error);
5932 		if (error != 0)
5933 			goto nfsmout;
5934 		/*
5935 		 * Get rid of Op# and status for next op.
5936 		 */
5937 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5938 		if (*++tl != 0)
5939 			nd->nd_flag |= ND_NOMOREDATA;
5940 	}
5941 	if (nd->nd_repstat == 0) {
5942 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5943 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5944 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
5945 	} else
5946 		error = nd->nd_repstat;
5947 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error);
5948 nfsmout:
5949 	m_freem(nd->nd_mrep);
5950 	free(nd, M_TEMP);
5951 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error);
5952 	return (error);
5953 }
5954 
5955 /*
5956  * Start up the thread that will execute nfsrv_deallocatedsdorpc().
5957  */
5958 static void
5959 start_deallocatedsdorpc(void *arg, int pending)
5960 {
5961 	struct nfsrvdeallocatedsdorpc *drpc;
5962 
5963 	drpc = (struct nfsrvdeallocatedsdorpc *)arg;
5964 	drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5965 	    drpc->len, NULL, drpc->cred, drpc->p);
5966 	drpc->done = 1;
5967 	NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err);
5968 }
5969 
5970 static int
5971 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5972     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5973     int *failposp)
5974 {
5975 	struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL;
5976 	struct nfsvattr na;
5977 	int error, i, ret, timo;
5978 
5979 	NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n");
5980 	drpc = NULL;
5981 	if (mirrorcnt > 1)
5982 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5983 		    M_WAITOK);
5984 
5985 	/*
5986 	 * Do the deallocate RPC for every DS, using a separate kernel process
5987 	 * for every DS except the last one.
5988 	 */
5989 	error = 0;
5990 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5991 		tdrpc->done = 0;
5992 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5993 		tdrpc->off = off;
5994 		tdrpc->len = len;
5995 		tdrpc->nmp = *nmpp;
5996 		tdrpc->cred = cred;
5997 		tdrpc->p = p;
5998 		tdrpc->inprog = 0;
5999 		tdrpc->err = 0;
6000 		ret = EIO;
6001 		if (nfs_pnfsiothreads != 0) {
6002 			ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc);
6003 			NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n",
6004 			    ret);
6005 		}
6006 		if (ret != 0) {
6007 			ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len,
6008 			    NULL, cred, p);
6009 			if (nfsds_failerr(ret) && *failposp == -1)
6010 				*failposp = i;
6011 			else if (error == 0 && ret != 0)
6012 				error = ret;
6013 		}
6014 		nmpp++;
6015 		fhp++;
6016 	}
6017 	ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
6018 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6019 		*failposp = mirrorcnt - 1;
6020 	else if (error == 0 && ret != 0)
6021 		error = ret;
6022 	if (error == 0)
6023 		error = nfsrv_setextattr(vp, &na, p);
6024 	NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error);
6025 	tdrpc = drpc;
6026 	timo = hz / 50;		/* Wait for 20msec. */
6027 	if (timo < 1)
6028 		timo = 1;
6029 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6030 		/* Wait for RPCs on separate threads to complete. */
6031 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6032 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
6033 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6034 			*failposp = i;
6035 		else if (error == 0 && tdrpc->err != 0)
6036 			error = tdrpc->err;
6037 	}
6038 	free(drpc, M_TEMP);
6039 	return (error);
6040 }
6041 
6042 static int
6043 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6044     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
6045     struct nfsvattr *dsnap)
6046 {
6047 	uint32_t *tl;
6048 	struct nfsrv_descript *nd;
6049 	nfsv4stateid_t st;
6050 	nfsattrbit_t attrbits;
6051 	int error;
6052 
6053 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
6054 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6055 	/*
6056 	 * Use a stateid where other is an alternating 01010 pattern and
6057 	 * seqid is 0xffffffff.  This value is not defined as special by
6058 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6059 	 * MDS->DS proxy operation.
6060 	 */
6061 	st.other[0] = 0x55555555;
6062 	st.other[1] = 0x55555555;
6063 	st.other[2] = 0x55555555;
6064 	st.seqid = 0xffffffff;
6065 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
6066 	    NULL, NULL, 0, 0, cred);
6067 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6068 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
6069 
6070 	/* Do a Getattr for the attributes that change due to writing. */
6071 	NFSZERO_ATTRBIT(&attrbits);
6072 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6073 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6074 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6075 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6076 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6077 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
6078 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
6079 	(void) nfsrv_putattrbit(nd, &attrbits);
6080 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6081 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6082 	if (error != 0) {
6083 		free(nd, M_TEMP);
6084 		return (error);
6085 	}
6086 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
6087 	    nd->nd_repstat);
6088 	/* Get rid of weak cache consistency data for now. */
6089 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
6090 	    (ND_NFSV4 | ND_V4WCCATTR)) {
6091 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
6092 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL, NULL);
6093 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
6094 		if (error != 0)
6095 			goto nfsmout;
6096 		/*
6097 		 * Get rid of Op# and status for next op.
6098 		 */
6099 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
6100 		if (*++tl != 0)
6101 			nd->nd_flag |= ND_NOMOREDATA;
6102 	}
6103 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
6104 	if (error != 0)
6105 		goto nfsmout;
6106 	if (nd->nd_repstat != 0)
6107 		error = nd->nd_repstat;
6108 	/*
6109 	 * Get the Change, Size, Access Time and Modify Time attributes and set
6110 	 * on the Metadata file, so its attributes will be what the file's
6111 	 * would be if it had been written.
6112 	 */
6113 	if (error == 0) {
6114 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
6115 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
6116 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL,
6117 		    NULL);
6118 	}
6119 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
6120 nfsmout:
6121 	m_freem(nd->nd_mrep);
6122 	free(nd, M_TEMP);
6123 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
6124 	return (error);
6125 }
6126 
6127 struct nfsrvsetattrdsdorpc {
6128 	int			done;
6129 	int			inprog;
6130 	struct task		tsk;
6131 	fhandle_t		fh;
6132 	struct nfsmount		*nmp;
6133 	struct vnode		*vp;
6134 	struct ucred		*cred;
6135 	NFSPROC_T		*p;
6136 	struct nfsvattr		na;
6137 	struct nfsvattr		dsna;
6138 	int			err;
6139 };
6140 
6141 /*
6142  * Start up the thread that will execute nfsrv_setattrdsdorpc().
6143  */
6144 static void
6145 start_setattrdsdorpc(void *arg, int pending)
6146 {
6147 	struct nfsrvsetattrdsdorpc *drpc;
6148 
6149 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
6150 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
6151 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
6152 	drpc->done = 1;
6153 }
6154 
6155 static int
6156 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6157     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
6158     struct nfsvattr *nap, int *failposp)
6159 {
6160 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
6161 	struct nfsvattr na;
6162 	int error, i, ret, timo;
6163 
6164 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
6165 	drpc = NULL;
6166 	if (mirrorcnt > 1)
6167 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6168 		    M_WAITOK);
6169 
6170 	/*
6171 	 * Do the setattr RPC for every DS, using a separate kernel process
6172 	 * for every DS except the last one.
6173 	 */
6174 	error = 0;
6175 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6176 		tdrpc->done = 0;
6177 		tdrpc->inprog = 0;
6178 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6179 		tdrpc->nmp = *nmpp;
6180 		tdrpc->vp = vp;
6181 		tdrpc->cred = cred;
6182 		tdrpc->p = p;
6183 		tdrpc->na = *nap;
6184 		tdrpc->err = 0;
6185 		ret = EIO;
6186 		if (nfs_pnfsiothreads != 0) {
6187 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
6188 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
6189 			    ret);
6190 		}
6191 		if (ret != 0) {
6192 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
6193 			    &na);
6194 			if (nfsds_failerr(ret) && *failposp == -1)
6195 				*failposp = i;
6196 			else if (error == 0 && ret != 0)
6197 				error = ret;
6198 		}
6199 		nmpp++;
6200 		fhp++;
6201 	}
6202 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
6203 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6204 		*failposp = mirrorcnt - 1;
6205 	else if (error == 0 && ret != 0)
6206 		error = ret;
6207 	if (error == 0)
6208 		error = nfsrv_setextattr(vp, &na, p);
6209 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
6210 	tdrpc = drpc;
6211 	timo = hz / 50;		/* Wait for 20msec. */
6212 	if (timo < 1)
6213 		timo = 1;
6214 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6215 		/* Wait for RPCs on separate threads to complete. */
6216 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6217 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
6218 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6219 			*failposp = i;
6220 		else if (error == 0 && tdrpc->err != 0)
6221 			error = tdrpc->err;
6222 	}
6223 	free(drpc, M_TEMP);
6224 	return (error);
6225 }
6226 
6227 /*
6228  * Do a Setattr of an NFSv4 ACL on the DS file.
6229  */
6230 static int
6231 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6232     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
6233 {
6234 	struct nfsrv_descript *nd;
6235 	nfsv4stateid_t st;
6236 	nfsattrbit_t attrbits;
6237 	int error;
6238 
6239 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
6240 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6241 	/*
6242 	 * Use a stateid where other is an alternating 01010 pattern and
6243 	 * seqid is 0xffffffff.  This value is not defined as special by
6244 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6245 	 * MDS->DS proxy operation.
6246 	 */
6247 	st.other[0] = 0x55555555;
6248 	st.other[1] = 0x55555555;
6249 	st.other[2] = 0x55555555;
6250 	st.seqid = 0xffffffff;
6251 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
6252 	    NULL, NULL, 0, 0, cred);
6253 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6254 	NFSZERO_ATTRBIT(&attrbits);
6255 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
6256 	/*
6257 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
6258 	 * so passing in the metadata "vp" will be ok, since it is of
6259 	 * the same type (VREG).
6260 	 */
6261 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
6262 	    NULL, 0, 0, 0, 0, 0, NULL);
6263 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6264 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6265 	if (error != 0) {
6266 		free(nd, M_TEMP);
6267 		return (error);
6268 	}
6269 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
6270 	    nd->nd_repstat);
6271 	error = nd->nd_repstat;
6272 	m_freem(nd->nd_mrep);
6273 	free(nd, M_TEMP);
6274 	return (error);
6275 }
6276 
6277 struct nfsrvsetacldsdorpc {
6278 	int			done;
6279 	int			inprog;
6280 	struct task		tsk;
6281 	fhandle_t		fh;
6282 	struct nfsmount		*nmp;
6283 	struct vnode		*vp;
6284 	struct ucred		*cred;
6285 	NFSPROC_T		*p;
6286 	struct acl		*aclp;
6287 	int			err;
6288 };
6289 
6290 /*
6291  * Start up the thread that will execute nfsrv_setacldsdorpc().
6292  */
6293 static void
6294 start_setacldsdorpc(void *arg, int pending)
6295 {
6296 	struct nfsrvsetacldsdorpc *drpc;
6297 
6298 	drpc = (struct nfsrvsetacldsdorpc *)arg;
6299 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
6300 	    drpc->vp, drpc->nmp, drpc->aclp);
6301 	drpc->done = 1;
6302 }
6303 
6304 static int
6305 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6306     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
6307     int *failposp)
6308 {
6309 	struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
6310 	int error, i, ret, timo;
6311 
6312 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
6313 	drpc = NULL;
6314 	if (mirrorcnt > 1)
6315 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6316 		    M_WAITOK);
6317 
6318 	/*
6319 	 * Do the setattr RPC for every DS, using a separate kernel process
6320 	 * for every DS except the last one.
6321 	 */
6322 	error = 0;
6323 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6324 		tdrpc->done = 0;
6325 		tdrpc->inprog = 0;
6326 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6327 		tdrpc->nmp = *nmpp;
6328 		tdrpc->vp = vp;
6329 		tdrpc->cred = cred;
6330 		tdrpc->p = p;
6331 		tdrpc->aclp = aclp;
6332 		tdrpc->err = 0;
6333 		ret = EIO;
6334 		if (nfs_pnfsiothreads != 0) {
6335 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
6336 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
6337 			    ret);
6338 		}
6339 		if (ret != 0) {
6340 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
6341 			    aclp);
6342 			if (nfsds_failerr(ret) && *failposp == -1)
6343 				*failposp = i;
6344 			else if (error == 0 && ret != 0)
6345 				error = ret;
6346 		}
6347 		nmpp++;
6348 		fhp++;
6349 	}
6350 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
6351 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6352 		*failposp = mirrorcnt - 1;
6353 	else if (error == 0 && ret != 0)
6354 		error = ret;
6355 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
6356 	tdrpc = drpc;
6357 	timo = hz / 50;		/* Wait for 20msec. */
6358 	if (timo < 1)
6359 		timo = 1;
6360 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6361 		/* Wait for RPCs on separate threads to complete. */
6362 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6363 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
6364 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6365 			*failposp = i;
6366 		else if (error == 0 && tdrpc->err != 0)
6367 			error = tdrpc->err;
6368 	}
6369 	free(drpc, M_TEMP);
6370 	return (error);
6371 }
6372 
6373 /*
6374  * Getattr call to the DS for the attributes that change due to writing.
6375  */
6376 static int
6377 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6378     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
6379 {
6380 	struct nfsrv_descript *nd;
6381 	int error;
6382 	nfsattrbit_t attrbits;
6383 
6384 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
6385 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6386 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
6387 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6388 	NFSZERO_ATTRBIT(&attrbits);
6389 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6390 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6391 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6392 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6393 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6394 	(void) nfsrv_putattrbit(nd, &attrbits);
6395 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6396 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6397 	if (error != 0) {
6398 		free(nd, M_TEMP);
6399 		return (error);
6400 	}
6401 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
6402 	    nd->nd_repstat);
6403 	if (nd->nd_repstat == 0) {
6404 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
6405 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
6406 		    NULL, NULL, NULL);
6407 		/*
6408 		 * We can only save the updated values in the extended
6409 		 * attribute if the vp is exclusively locked.
6410 		 * This should happen when any of the following operations
6411 		 * occur on the vnode:
6412 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
6413 		 * As such, the updated extended attribute should get saved
6414 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
6415 		 * attributes to be returned without calling this function.
6416 		 */
6417 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
6418 			error = nfsrv_setextattr(vp, nap, p);
6419 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
6420 			    error);
6421 		}
6422 	} else
6423 		error = nd->nd_repstat;
6424 	m_freem(nd->nd_mrep);
6425 	free(nd, M_TEMP);
6426 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
6427 	return (error);
6428 }
6429 
6430 /*
6431  * Seek call to a DS.
6432  */
6433 static int
6434 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
6435     struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
6436 {
6437 	uint32_t *tl;
6438 	struct nfsrv_descript *nd;
6439 	nfsv4stateid_t st;
6440 	int error;
6441 
6442 	NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
6443 	/*
6444 	 * Use a stateid where other is an alternating 01010 pattern and
6445 	 * seqid is 0xffffffff.  This value is not defined as special by
6446 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6447 	 * MDS->DS proxy operation.
6448 	 */
6449 	st.other[0] = 0x55555555;
6450 	st.other[1] = 0x55555555;
6451 	st.other[2] = 0x55555555;
6452 	st.seqid = 0xffffffff;
6453 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6454 	nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
6455 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6456 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6457 	NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
6458 	txdr_hyper(*offp, tl); tl += 2;
6459 	*tl = txdr_unsigned(content);
6460 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6461 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6462 	if (error != 0) {
6463 		free(nd, M_TEMP);
6464 		return (error);
6465 	}
6466 	NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
6467 	if (nd->nd_repstat == 0) {
6468 		NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
6469 		if (*tl++ == newnfs_true)
6470 			*eofp = true;
6471 		else
6472 			*eofp = false;
6473 		*offp = fxdr_hyper(tl);
6474 	} else
6475 		error = nd->nd_repstat;
6476 nfsmout:
6477 	m_freem(nd->nd_mrep);
6478 	free(nd, M_TEMP);
6479 	NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
6480 	return (error);
6481 }
6482 
6483 /*
6484  * Get the device id and file handle for a DS file.
6485  */
6486 int
6487 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
6488     fhandle_t *fhp, char *devid)
6489 {
6490 	int buflen, error;
6491 	char *buf;
6492 
6493 	buflen = 1024;
6494 	buf = malloc(buflen, M_TEMP, M_WAITOK);
6495 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
6496 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
6497 	free(buf, M_TEMP);
6498 	return (error);
6499 }
6500 
6501 /*
6502  * Do a Lookup against the DS for the filename.
6503  */
6504 static int
6505 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
6506     struct vnode **nvpp, NFSPROC_T *p)
6507 {
6508 	struct nameidata named;
6509 	struct ucred *tcred;
6510 	char *bufp;
6511 	u_long *hashp;
6512 	struct vnode *nvp;
6513 	int error;
6514 
6515 	tcred = newnfs_getcred();
6516 	named.ni_cnd.cn_nameiop = LOOKUP;
6517 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
6518 	named.ni_cnd.cn_cred = tcred;
6519 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF;
6520 	nfsvno_setpathbuf(&named, &bufp, &hashp);
6521 	named.ni_cnd.cn_nameptr = bufp;
6522 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
6523 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
6524 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
6525 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
6526 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
6527 	NFSFREECRED(tcred);
6528 	nfsvno_relpathbuf(&named);
6529 	if (error == 0)
6530 		*nvpp = nvp;
6531 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
6532 	return (error);
6533 }
6534 
6535 /*
6536  * Set the file handle to the correct one.
6537  */
6538 static void
6539 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
6540     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
6541 {
6542 	struct nfsnode *np;
6543 	int ret = 0;
6544 
6545 	np = VTONFS(nvp);
6546 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
6547 	/*
6548 	 * We can only do a vn_set_extattr() if the vnode is exclusively
6549 	 * locked and vn_start_write() has been done.  If devid != NULL or
6550 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
6551 	 * may not have been done.
6552 	 * If not done now, it will be done on a future call.
6553 	 */
6554 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
6555 	    LK_EXCLUSIVE)
6556 		ret = vn_extattr_set(vp, IO_NODELOCKED,
6557 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
6558 		    (char *)pf, p);
6559 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
6560 }
6561 
6562 /*
6563  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
6564  * when the DS has failed.
6565  */
6566 void
6567 nfsrv_killrpcs(struct nfsmount *nmp)
6568 {
6569 
6570 	/*
6571 	 * Call newnfs_nmcancelreqs() to cause
6572 	 * any RPCs in progress on the mount point to
6573 	 * fail.
6574 	 * This will cause any process waiting for an
6575 	 * RPC to complete while holding a vnode lock
6576 	 * on the mounted-on vnode (such as "df" or
6577 	 * a non-forced "umount") to fail.
6578 	 * This will unlock the mounted-on vnode so
6579 	 * a forced dismount can succeed.
6580 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
6581 	 * called.
6582 	 */
6583 	newnfs_nmcancelreqs(nmp);
6584 }
6585 
6586 /*
6587  * Sum up the statfs info for each of the DSs, so that the client will
6588  * receive the total for all DSs.
6589  */
6590 static int
6591 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6592 {
6593 	struct statfs *tsf;
6594 	struct nfsdevice *ds;
6595 	struct vnode **dvpp, **tdvpp, *dvp;
6596 	uint64_t tot;
6597 	int cnt, error = 0, i;
6598 
6599 	if (nfsrv_devidcnt <= 0)
6600 		return (ENXIO);
6601 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6602 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6603 
6604 	/* Get an array of the dvps for the DSs. */
6605 	tdvpp = dvpp;
6606 	i = 0;
6607 	NFSDDSLOCK();
6608 	/* First, search for matches for same file system. */
6609 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6610 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6611 		    fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6612 			if (++i > nfsrv_devidcnt)
6613 				break;
6614 			*tdvpp++ = ds->nfsdev_dvp;
6615 		}
6616 	}
6617 	/*
6618 	 * If no matches for same file system, total all servers not assigned
6619 	 * to a file system.
6620 	 */
6621 	if (i == 0) {
6622 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6623 			if (ds->nfsdev_nmp != NULL &&
6624 			    ds->nfsdev_mdsisset == 0) {
6625 				if (++i > nfsrv_devidcnt)
6626 					break;
6627 				*tdvpp++ = ds->nfsdev_dvp;
6628 			}
6629 		}
6630 	}
6631 	NFSDDSUNLOCK();
6632 	cnt = i;
6633 
6634 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
6635 	tdvpp = dvpp;
6636 	for (i = 0; i < cnt && error == 0; i++) {
6637 		dvp = *tdvpp++;
6638 		error = VFS_STATFS(dvp->v_mount, tsf);
6639 		if (error == 0) {
6640 			if (sf->f_bsize == 0) {
6641 				if (tsf->f_bsize > 0)
6642 					sf->f_bsize = tsf->f_bsize;
6643 				else
6644 					sf->f_bsize = 8192;
6645 			}
6646 			if (tsf->f_blocks > 0) {
6647 				if (sf->f_bsize != tsf->f_bsize) {
6648 					tot = tsf->f_blocks * tsf->f_bsize;
6649 					sf->f_blocks += (tot / sf->f_bsize);
6650 				} else
6651 					sf->f_blocks += tsf->f_blocks;
6652 			}
6653 			if (tsf->f_bfree > 0) {
6654 				if (sf->f_bsize != tsf->f_bsize) {
6655 					tot = tsf->f_bfree * tsf->f_bsize;
6656 					sf->f_bfree += (tot / sf->f_bsize);
6657 				} else
6658 					sf->f_bfree += tsf->f_bfree;
6659 			}
6660 			if (tsf->f_bavail > 0) {
6661 				if (sf->f_bsize != tsf->f_bsize) {
6662 					tot = tsf->f_bavail * tsf->f_bsize;
6663 					sf->f_bavail += (tot / sf->f_bsize);
6664 				} else
6665 					sf->f_bavail += tsf->f_bavail;
6666 			}
6667 		}
6668 	}
6669 	free(tsf, M_TEMP);
6670 	free(dvpp, M_TEMP);
6671 	return (error);
6672 }
6673 
6674 /*
6675  * Set an NFSv4 acl.
6676  */
6677 int
6678 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6679 {
6680 	int error;
6681 
6682 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6683 		error = NFSERR_ATTRNOTSUPP;
6684 		goto out;
6685 	}
6686 	/*
6687 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6688 	 * Make sure it has enough room for that - splitting every entry
6689 	 * into two and appending "canonical six" entries at the end.
6690 	 * Cribbed out of kern/vfs_acl.c - Rick M.
6691 	 */
6692 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6693 		error = NFSERR_ATTRNOTSUPP;
6694 		goto out;
6695 	}
6696 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6697 	if (error == 0) {
6698 		error = nfsrv_dssetacl(vp, aclp, cred, p);
6699 		if (error == ENOENT)
6700 			error = 0;
6701 	}
6702 
6703 out:
6704 	NFSEXITCODE(error);
6705 	return (error);
6706 }
6707 
6708 /*
6709  * Seek vnode op call (actually it is a VOP_IOCTL()).
6710  * This function is called with the vnode locked, but unlocks and vrele()s
6711  * the vp before returning.
6712  */
6713 int
6714 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6715     off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6716 {
6717 	struct nfsvattr at;
6718 	int error, ret;
6719 
6720 	ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6721 	/*
6722 	 * Attempt to seek on a DS file. A return of ENOENT implies
6723 	 * there is no DS file to seek on.
6724 	 */
6725 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6726 	    NULL, NULL, NULL, NULL, offp, content, eofp);
6727 	if (error != ENOENT) {
6728 		vput(vp);
6729 		return (error);
6730 	}
6731 
6732 	/*
6733 	 * Do the VOP_IOCTL() call.  For the case where *offp == file_size,
6734 	 * VOP_IOCTL() will return ENXIO.  However, the correct reply for
6735 	 * NFSv4.2 is *eofp == true and error == 0 for this case.
6736 	 */
6737 	NFSVOPUNLOCK(vp);
6738 	error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6739 	*eofp = false;
6740 	if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6741 		/* Handle the cases where we might be at EOF. */
6742 		ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6743 		if (ret == 0 && *offp == at.na_size) {
6744 			*eofp = true;
6745 			error = 0;
6746 		}
6747 		if (ret != 0 && error == 0)
6748 			error = ret;
6749 	}
6750 	vrele(vp);
6751 	NFSEXITCODE(error);
6752 	return (error);
6753 }
6754 
6755 /*
6756  * Allocate vnode op call.
6757  */
6758 int
6759 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6760     NFSPROC_T *p)
6761 {
6762 	int error;
6763 	off_t olen;
6764 
6765 	ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6766 	/*
6767 	 * Attempt to allocate on a DS file. A return of ENOENT implies
6768 	 * there is no DS file to allocate on.
6769 	 */
6770 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6771 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6772 	if (error != ENOENT)
6773 		return (error);
6774 
6775 	/*
6776 	 * Do the actual VOP_ALLOCATE(), looping so long as
6777 	 * progress is being made, to achieve completion.
6778 	 */
6779 	do {
6780 		olen = len;
6781 		error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred);
6782 		if (error == 0 && len > 0 && olen > len)
6783 			maybe_yield();
6784 	} while (error == 0 && len > 0 && olen > len);
6785 	if (error == 0 && len > 0)
6786 		error = NFSERR_IO;
6787 	NFSEXITCODE(error);
6788 	return (error);
6789 }
6790 
6791 /*
6792  * Deallocate vnode op call.
6793  */
6794 int
6795 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6796     NFSPROC_T *p)
6797 {
6798 	int error;
6799 	off_t olen;
6800 
6801 	ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp");
6802 	/*
6803 	 * Attempt to deallocate on a DS file. A return of ENOENT implies
6804 	 * there is no DS file to deallocate on.
6805 	 */
6806 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL,
6807 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6808 	if (error != ENOENT)
6809 		return (error);
6810 
6811 	/*
6812 	 * Do the actual VOP_DEALLOCATE(), looping so long as
6813 	 * progress is being made, to achieve completion.
6814 	 */
6815 	do {
6816 		olen = len;
6817 		error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred);
6818 		if (error == 0 && len > 0 && olen > len)
6819 			maybe_yield();
6820 	} while (error == 0 && len > 0 && olen > len);
6821 	if (error == 0 && len > 0)
6822 		error = NFSERR_IO;
6823 	NFSEXITCODE(error);
6824 	return (error);
6825 }
6826 
6827 /*
6828  * Get Extended Atribute vnode op into an mbuf list.
6829  */
6830 int
6831 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6832     struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p,
6833     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
6834 {
6835 	struct iovec *iv;
6836 	struct uio io, *uiop = &io;
6837 	struct mbuf *m, *m2;
6838 	int alen, error, len, tlen;
6839 	size_t siz;
6840 
6841 	/* First, find out the size of the extended attribute. */
6842 	error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6843 	    &siz, cred, p);
6844 	if (error != 0)
6845 		return (NFSERR_NOXATTR);
6846 	if (siz > maxresp - NFS_MAXXDR)
6847 		return (NFSERR_XATTR2BIG);
6848 	len = siz;
6849 	tlen = NFSM_RNDUP(len);
6850 	if (tlen > 0) {
6851 		/*
6852 		 * If cnt > MCLBYTES and the reply will not be saved, use
6853 		 * ext_pgs mbufs for TLS.
6854 		 * For NFSv4.0, we do not know for sure if the reply will
6855 		 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
6856 		 * Always use ext_pgs mbufs if ND_EXTPG is set.
6857 		 */
6858 		if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES &&
6859 		    (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS &&
6860 		    (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4))
6861 			uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen,
6862 			    maxextsiz, &m, &m2, &iv);
6863 		else
6864 			uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2,
6865 			    &iv);
6866 		uiop->uio_iov = iv;
6867 	} else {
6868 		uiop->uio_iovcnt = 0;
6869 		uiop->uio_iov = iv = NULL;
6870 		m = m2 = NULL;
6871 	}
6872 	uiop->uio_offset = 0;
6873 	uiop->uio_resid = tlen;
6874 	uiop->uio_rw = UIO_READ;
6875 	uiop->uio_segflg = UIO_SYSSPACE;
6876 	uiop->uio_td = p;
6877 #ifdef MAC
6878 	error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6879 	    name);
6880 	if (error != 0)
6881 		goto out;
6882 #endif
6883 
6884 	if (tlen > 0)
6885 		error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6886 		    NULL, cred, p);
6887 	if (error != 0)
6888 		goto out;
6889 	if (uiop->uio_resid > 0) {
6890 		alen = tlen;
6891 		len = tlen - uiop->uio_resid;
6892 		tlen = NFSM_RNDUP(len);
6893 		if (alen != tlen)
6894 			printf("nfsvno_getxattr: weird size read\n");
6895 		if (tlen == 0) {
6896 			m_freem(m);
6897 			m = m2 = NULL;
6898 		} else if (alen != tlen || tlen != len)
6899 			m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6900 	}
6901 	*lenp = len;
6902 	*mpp = m;
6903 	*mpendp = m2;
6904 
6905 out:
6906 	if (error != 0) {
6907 		if (m != NULL)
6908 			m_freem(m);
6909 		*lenp = 0;
6910 	}
6911 	free(iv, M_TEMP);
6912 	NFSEXITCODE(error);
6913 	return (error);
6914 }
6915 
6916 /*
6917  * Set Extended attribute vnode op from an mbuf list.
6918  */
6919 int
6920 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6921     char *cp, struct ucred *cred, struct thread *p)
6922 {
6923 	struct iovec *iv;
6924 	struct uio uio, *uiop = &uio;
6925 	int cnt, error;
6926 
6927 	error = 0;
6928 #ifdef MAC
6929 	error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6930 	    name);
6931 #endif
6932 	if (error != 0)
6933 		goto out;
6934 
6935 	uiop->uio_rw = UIO_WRITE;
6936 	uiop->uio_segflg = UIO_SYSSPACE;
6937 	uiop->uio_td = p;
6938 	uiop->uio_offset = 0;
6939 	uiop->uio_resid = len;
6940 	if (len > 0) {
6941 		error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6942 		uiop->uio_iov = iv;
6943 		uiop->uio_iovcnt = cnt;
6944 	} else {
6945 		uiop->uio_iov = iv = NULL;
6946 		uiop->uio_iovcnt = 0;
6947 	}
6948 	if (error == 0) {
6949 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6950 		    cred, p);
6951 		if (error == 0) {
6952 			if (vp->v_type == VREG && nfsrv_devidcnt != 0)
6953 				nfsvno_updateds(vp, cred, p);
6954 			error = VOP_FSYNC(vp, MNT_WAIT, p);
6955 		}
6956 		free(iv, M_TEMP);
6957 	}
6958 
6959 out:
6960 	NFSEXITCODE(error);
6961 	return (error);
6962 }
6963 
6964 /*
6965  * For a pNFS server, the DS file's ctime and
6966  * va_filerev (TimeMetadata and Change) needs to
6967  * be updated.  This is a hack, but works by
6968  * flipping the S_ISGID bit in va_mode and then
6969  * flipping it back.
6970  * It does result in two MDS->DS RPCs, but creating
6971  * a custom RPC just to do this seems overkill, since
6972  * Setxattr/Rmxattr will not be done that frequently.
6973  * If it fails part way through, that is not too
6974  * serious, since the DS file is never executed.
6975  */
6976 static void
6977 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p)
6978 {
6979 	struct nfsvattr nva;
6980 	int ret;
6981 	u_short tmode;
6982 
6983 	ret = VOP_GETATTR(vp, &nva.na_vattr, cred);
6984 	if (ret == 0) {
6985 		tmode = nva.na_mode;
6986 		NFSVNO_ATTRINIT(&nva);
6987 		tmode ^= S_ISGID;
6988 		NFSVNO_SETATTRVAL(&nva, mode, tmode);
6989 		ret = nfsrv_proxyds(vp, 0, 0, cred, p,
6990 		    NFSPROC_SETATTR, NULL, NULL, NULL, &nva,
6991 		    NULL, NULL, 0, NULL);
6992 		if (ret == 0) {
6993 			tmode ^= S_ISGID;
6994 			NFSVNO_SETATTRVAL(&nva, mode, tmode);
6995 			ret = nfsrv_proxyds(vp, 0, 0, cred, p,
6996 			    NFSPROC_SETATTR, NULL, NULL, NULL,
6997 			    &nva, NULL, NULL, 0, NULL);
6998 		}
6999 	}
7000 }
7001 
7002 /*
7003  * Remove Extended attribute vnode op.
7004  */
7005 int
7006 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
7007     struct ucred *cred, struct thread *p)
7008 {
7009 	int error;
7010 
7011 	/*
7012 	 * Get rid of any delegations.  I am not sure why this is required,
7013 	 * but RFC-8276 says so.
7014 	 */
7015 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
7016 	if (error != 0)
7017 		goto out;
7018 #ifdef MAC
7019 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
7020 	    name);
7021 	if (error != 0)
7022 		goto out;
7023 #endif
7024 
7025 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
7026 	if (error == EOPNOTSUPP)
7027 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
7028 		    cred, p);
7029 	if (error == 0) {
7030 		if (vp->v_type == VREG && nfsrv_devidcnt != 0)
7031 			nfsvno_updateds(vp, cred, p);
7032 		error = VOP_FSYNC(vp, MNT_WAIT, p);
7033 	}
7034 out:
7035 	NFSEXITCODE(error);
7036 	return (error);
7037 }
7038 
7039 /*
7040  * List Extended Atribute vnode op into an mbuf list.
7041  */
7042 int
7043 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
7044     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
7045 {
7046 	struct iovec iv;
7047 	struct uio io;
7048 	int error;
7049 	size_t siz;
7050 
7051 	*bufp = NULL;
7052 	/* First, find out the size of the extended attribute. */
7053 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
7054 	    p);
7055 	if (error != 0)
7056 		return (NFSERR_NOXATTR);
7057 	if (siz <= cookie) {
7058 		*lenp = 0;
7059 		*eofp = true;
7060 		goto out;
7061 	}
7062 	if (siz > cookie + *lenp) {
7063 		siz = cookie + *lenp;
7064 		*eofp = false;
7065 	} else
7066 		*eofp = true;
7067 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
7068 	if (siz > 10 * 1024 * 1024) {
7069 		error = NFSERR_XATTR2BIG;
7070 		goto out;
7071 	}
7072 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
7073 	iv.iov_base = *bufp;
7074 	iv.iov_len = siz;
7075 	io.uio_iovcnt = 1;
7076 	io.uio_iov = &iv;
7077 	io.uio_offset = 0;
7078 	io.uio_resid = siz;
7079 	io.uio_rw = UIO_READ;
7080 	io.uio_segflg = UIO_SYSSPACE;
7081 	io.uio_td = p;
7082 #ifdef MAC
7083 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
7084 	if (error != 0)
7085 		goto out;
7086 #endif
7087 
7088 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
7089 	    p);
7090 	if (error != 0)
7091 		goto out;
7092 	if (io.uio_resid > 0)
7093 		siz -= io.uio_resid;
7094 	*lenp = siz;
7095 
7096 out:
7097 	if (error != 0) {
7098 		free(*bufp, M_TEMP);
7099 		*bufp = NULL;
7100 	}
7101 	NFSEXITCODE(error);
7102 	return (error);
7103 }
7104 
7105 /*
7106  * Trim trailing data off the mbuf list being built.
7107  */
7108 void
7109 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
7110     int bextpg, int bextpgsiz)
7111 {
7112 	vm_page_t pg;
7113 	int fullpgsiz, i;
7114 
7115 	if (mb->m_next != NULL) {
7116 		m_freem(mb->m_next);
7117 		mb->m_next = NULL;
7118 	}
7119 	if ((mb->m_flags & M_EXTPG) != 0) {
7120 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
7121 		    ("nfsm_trimtrailing: bextpg out of range"));
7122 		KASSERT(bpos == (char *)(void *)
7123 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
7124 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
7125 
7126 		/* First, get rid of any pages after this position. */
7127 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
7128 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
7129 			vm_page_unwire_noq(pg);
7130 			vm_page_free(pg);
7131 		}
7132 		mb->m_epg_npgs = bextpg + 1;
7133 		if (bextpg == 0)
7134 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
7135 		else
7136 			fullpgsiz = PAGE_SIZE;
7137 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
7138 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
7139 		for (i = 1; i < mb->m_epg_npgs; i++)
7140 			mb->m_len += m_epg_pagelen(mb, i, 0);
7141 		nd->nd_bextpgsiz = bextpgsiz;
7142 		nd->nd_bextpg = bextpg;
7143 	} else
7144 		mb->m_len = bpos - mtod(mb, char *);
7145 	nd->nd_mb = mb;
7146 	nd->nd_bpos = bpos;
7147 }
7148 
7149 
7150 /*
7151  * Check to see if a put file handle operation should test for
7152  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
7153  * When Open is the next operation, NFSERR_WRONGSEC cannot be
7154  * replied for the Open cases that use a component.  This can
7155  * be identified by the fact that the file handle's type is VDIR.
7156  */
7157 bool
7158 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, __enum_uint8(vtype) vtyp)
7159 {
7160 
7161 	if ((nd->nd_flag & ND_NFSV4) == 0)
7162 		return (true);
7163 
7164 	if ((nd->nd_flag & ND_LASTOP) != 0)
7165 		return (false);
7166 
7167 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
7168 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
7169 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
7170 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
7171 		return (false);
7172 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
7173 		return (false);
7174 	return (true);
7175 }
7176 
7177 /*
7178  * Check DSs marked no space.
7179  */
7180 void
7181 nfsrv_checknospc(void)
7182 {
7183 	struct statfs *tsf;
7184 	struct nfsdevice *ds;
7185 	struct vnode **dvpp, **tdvpp, *dvp;
7186 	char *devid, *tdevid;
7187 	int cnt, error = 0, i;
7188 
7189 	if (nfsrv_devidcnt <= 0)
7190 		return;
7191 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
7192 	devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK);
7193 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
7194 
7195 	/* Get an array of the dvps for the DSs. */
7196 	tdvpp = dvpp;
7197 	tdevid = devid;
7198 	i = 0;
7199 	NFSDDSLOCK();
7200 	/* First, search for matches for same file system. */
7201 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7202 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) {
7203 			if (++i > nfsrv_devidcnt)
7204 				break;
7205 			*tdvpp++ = ds->nfsdev_dvp;
7206 			NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID);
7207 			tdevid += NFSX_V4DEVICEID;
7208 		}
7209 	}
7210 	NFSDDSUNLOCK();
7211 
7212 	/* Do a VFS_STATFS() for each of the DSs and clear no space. */
7213 	cnt = i;
7214 	tdvpp = dvpp;
7215 	tdevid = devid;
7216 	for (i = 0; i < cnt && error == 0; i++) {
7217 		dvp = *tdvpp++;
7218 		error = VFS_STATFS(dvp->v_mount, tsf);
7219 		if (error == 0 && tsf->f_bavail > 0) {
7220 			NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n");
7221 			nfsrv_marknospc(tdevid, false);
7222 		}
7223 		tdevid += NFSX_V4DEVICEID;
7224 	}
7225 	free(tsf, M_TEMP);
7226 	free(dvpp, M_TEMP);
7227 	free(devid, M_TEMP);
7228 }
7229 
7230 /*
7231  * Initialize everything that needs to be initialized for a vnet.
7232  */
7233 static void
7234 nfsrv_vnetinit(const void *unused __unused)
7235 {
7236 
7237 	nfsd_mntinit();
7238 }
7239 VNET_SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY,
7240     nfsrv_vnetinit, NULL);
7241 
7242 /*
7243  * Clean up everything that is in a vnet and needs to be
7244  * done when the jail is destroyed or the module unloaded.
7245  */
7246 static void
7247 nfsrv_cleanup(const void *unused __unused)
7248 {
7249 	int i;
7250 
7251 	NFSD_LOCK();
7252 	if (!NFSD_VNET(nfsrv_mntinited)) {
7253 		NFSD_UNLOCK();
7254 		return;
7255 	}
7256 	NFSD_VNET(nfsrv_mntinited) = false;
7257 	NFSD_UNLOCK();
7258 
7259 	/* Clean out all NFSv4 state. */
7260 	nfsrv_throwawayallstate(curthread);
7261 
7262 	/* Clean the NFS server reply cache */
7263 	nfsrvd_cleancache();
7264 
7265 	/* Clean out v4root exports. */
7266 	if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) {
7267 		vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export);
7268 		free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT);
7269 		NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL;
7270 	}
7271 
7272 	/* Free up the krpc server pool. */
7273 	if (NFSD_VNET(nfsrvd_pool) != NULL)
7274 		svcpool_destroy(NFSD_VNET(nfsrvd_pool));
7275 
7276 	/* and get rid of the locks */
7277 	for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7278 		mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx);
7279 		mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx);
7280 	}
7281 	mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx);
7282 	for (i = 0; i < nfsrv_sessionhashsize; i++)
7283 		mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx);
7284 	lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock);
7285 	free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE);
7286 	free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE);
7287 	free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE);
7288 	free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT);
7289 	free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE);
7290 	free(NFSD_VNET(nfssessionhash), M_NFSDSESSION);
7291 	free(NFSD_VNET(nfsv4root_mnt), M_TEMP);
7292 	NFSD_VNET(nfsv4root_mnt) = NULL;
7293 }
7294 VNET_SYSUNINIT(nfsrv_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY,
7295     nfsrv_cleanup, NULL);
7296 
7297 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
7298 
7299 /*
7300  * Called once to initialize data structures...
7301  */
7302 static int
7303 nfsd_modevent(module_t mod, int type, void *data)
7304 {
7305 	int error = 0, i;
7306 	static int loaded = 0;
7307 
7308 	switch (type) {
7309 	case MOD_LOAD:
7310 		if (loaded)
7311 			goto out;
7312 		newnfs_portinit();
7313 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
7314 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
7315 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
7316 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
7317 #ifdef VV_DISABLEDELEG
7318 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
7319 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
7320 #endif
7321 		nfsd_call_nfsd = nfssvc_nfsd;
7322 		loaded = 1;
7323 		break;
7324 
7325 	case MOD_UNLOAD:
7326 		if (newnfs_numnfsd != 0) {
7327 			error = EBUSY;
7328 			break;
7329 		}
7330 
7331 #ifdef VV_DISABLEDELEG
7332 		vn_deleg_ops.vndeleg_recall = NULL;
7333 		vn_deleg_ops.vndeleg_disable = NULL;
7334 #endif
7335 		nfsd_call_nfsd = NULL;
7336 		mtx_destroy(&nfsrc_udpmtx);
7337 		mtx_destroy(&nfs_v4root_mutex);
7338 		mtx_destroy(&nfsrv_dontlistlock_mtx);
7339 		mtx_destroy(&nfsrv_recalllock_mtx);
7340 		if (nfslayouthash != NULL) {
7341 			for (i = 0; i < nfsrv_layouthashsize; i++)
7342 				mtx_destroy(&nfslayouthash[i].mtx);
7343 			free(nfslayouthash, M_NFSDSESSION);
7344 		}
7345 		loaded = 0;
7346 		break;
7347 	default:
7348 		error = EOPNOTSUPP;
7349 		break;
7350 	}
7351 
7352 out:
7353 	NFSEXITCODE(error);
7354 	return (error);
7355 }
7356 static moduledata_t nfsd_mod = {
7357 	"nfsd",
7358 	nfsd_modevent,
7359 	NULL,
7360 };
7361 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
7362 
7363 /* So that loader and kldload(2) can find us, wherever we are.. */
7364 MODULE_VERSION(nfsd, 1);
7365 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
7366 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
7367 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
7368 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
7369