xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 3dfbbe8586bbd489ffc67941f09f94c2d8152946)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2009 Rick Macklem, University of Guelph
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 #include <sys/extattr.h>
34 #include <fs/nfs/nfsport.h>
35 
36 int nfsrv_issuedelegs = 0;
37 int nfsrv_dolocallocks = 0;
38 struct nfsv4lock nfsv4rootfs_lock;
39 time_t nfsdev_time = 0;
40 int nfsrv_layouthashsize;
41 volatile int nfsrv_layoutcnt = 0;
42 
43 VNET_DEFINE(struct nfsrv_stablefirst, nfsrv_stablefirst);
44 
45 VNET_DECLARE(int, nfsrv_numnfsd);
46 VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
47 
48 extern uint32_t nfs_srvmaxio;
49 extern int nfsrv_lease;
50 extern struct timeval nfsboottime;
51 extern u_int32_t newnfs_true, newnfs_false;
52 extern struct mtx nfsrv_dslock_mtx;
53 extern struct mtx nfsrv_recalllock_mtx;
54 extern struct mtx nfsrv_dontlistlock_mtx;
55 extern int nfsd_debuglevel;
56 extern u_int nfsrv_dsdirsize;
57 extern struct nfsdevicehead nfsrv_devidhead;
58 extern int nfsrv_doflexfile;
59 extern int nfsrv_maxpnfsmirror;
60 NFSV4ROOTLOCKMUTEX;
61 NFSSTATESPINLOCK;
62 extern struct nfsdontlisthead nfsrv_dontlisthead;
63 extern volatile int nfsrv_devidcnt;
64 extern struct nfslayouthead nfsrv_recalllisthead;
65 extern char *nfsrv_zeropnfsdat;
66 extern uint64_t nfsrv_stripesiz;
67 
68 SYSCTL_DECL(_vfs_nfsd);
69 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
70 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
71     &nfsrv_statehashsize, 0,
72     "Size of state hash table set via loader.conf");
73 
74 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
75 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
76     &nfsrv_clienthashsize, 0,
77     "Size of client hash table set via loader.conf");
78 
79 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
80 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
81     &nfsrv_lockhashsize, 0,
82     "Size of file handle hash table set via loader.conf");
83 
84 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
85 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
86     &nfsrv_sessionhashsize, 0,
87     "Size of session hash table set via loader.conf");
88 
89 int	nfsrv_layouthighwater = NFSLAYOUTHIGHWATER;
90 SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN,
91     &nfsrv_layouthighwater, 0,
92     "High water mark for number of layouts set via loader.conf");
93 
94 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
95 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
96     &nfsrv_v4statelimit, 0,
97     "High water limit for NFSv4 opens+locks+delegations");
98 
99 static int	nfsrv_writedelegifpos = 0;
100 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
101     &nfsrv_writedelegifpos, 0,
102     "Issue a write delegation for read opens if possible");
103 
104 static int	nfsrv_allowreadforwriteopen = 1;
105 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
106     &nfsrv_allowreadforwriteopen, 0,
107     "Allow Reads to be done with Write Access StateIDs");
108 
109 int	nfsrv_pnfsatime = 0;
110 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW,
111     &nfsrv_pnfsatime, 0,
112     "For pNFS service, do Getattr ops to keep atime up-to-date");
113 
114 int	nfsrv_flexlinuxhack = 0;
115 SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW,
116     &nfsrv_flexlinuxhack, 0,
117     "For Linux clients, hack around Flex File Layout bug");
118 
119 VNET_DEFINE_STATIC(bool, nfsd_disable_grace) = false;
120 SYSCTL_BOOL(_vfs_nfsd, OID_AUTO, testing_disable_grace,
121     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nfsd_disable_grace),
122     0, "Disable grace for testing");
123 
124 /*
125  * Hash lists for nfs V4.
126  */
127 VNET_DEFINE(struct nfsclienthashhead *, nfsclienthash);
128 VNET_DEFINE(struct nfslockhashhead *, nfslockhash);
129 VNET_DEFINE(struct nfssessionhash *, nfssessionhash);
130 
131 struct nfslayouthash		*nfslayouthash;
132 volatile int nfsrv_dontlistlen = 0;
133 
134 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
135 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
136 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
137 static int nfsrv_nogsscallback = 0;
138 static volatile int nfsrv_writedelegcnt = 0;
139 static int nfsrv_faildscnt;
140 
141 VNET_DEFINE_STATIC(time_t, nfsrvboottime);
142 
143 /* local functions */
144 static void nfsrv_dumpaclient(struct nfsclient *clp,
145     struct nfsd_dumpclients *dumpp);
146 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
147     NFSPROC_T *p);
148 static void nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
149     NFSPROC_T *p);
150 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
151     NFSPROC_T *p);
152 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
153     int cansleep, NFSPROC_T *p);
154 static void nfsrv_freenfslock(struct nfslock *lop);
155 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
156 static void nfsrv_freedeleg(struct nfsstate *);
157 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
158     u_int32_t flags, struct nfsstate **stpp);
159 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
160     struct nfsstate **stpp);
161 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
162     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
163 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
164     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
165 static void nfsrv_insertlock(struct nfslock *new_lop,
166     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
167 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
168     struct nfslock **other_lopp, struct nfslockfile *lfp);
169 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
170     nfsv4stateid_t *stateidp, int specialid);
171 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
172     u_int32_t flags);
173 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
174     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
175     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
176 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
177     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
178     int *slotposp);
179 static u_int32_t nfsrv_nextclientindex(void);
180 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
181 static void nfsrv_markstable(struct nfsclient *clp);
182 static void nfsrv_markreclaim(struct nfsclient *clp);
183 static int nfsrv_checkstable(struct nfsclient *clp);
184 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
185     vnode *vp, NFSPROC_T *p);
186 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
187     NFSPROC_T *p, vnode_t vp);
188 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
189     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
190 static int nfsrv_notsamecredname(int op, struct nfsrv_descript *nd,
191     struct nfsclient *clp);
192 static time_t nfsrv_leaseexpiry(void);
193 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
194 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
195     struct nfsstate *stp, struct nfsrvcache *op);
196 static int nfsrv_nootherstate(struct nfsstate *stp);
197 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
198     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
199 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
200     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
201 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
202     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
203     NFSPROC_T *p);
204 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
205     NFSPROC_T *p);
206 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
207     uint64_t first, uint64_t end);
208 static void nfsrv_locklf(struct nfslockfile *lfp);
209 static void nfsrv_unlocklf(struct nfslockfile *lfp);
210 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
211 static int nfsrv_freesession(struct nfsrv_descript *nd, struct nfsdsession *sep,
212     uint8_t *sessionid, bool locked, SVCXPRT **old_xprtp);
213 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
214     int dont_replycache, struct nfsdsession **sepp, int *slotposp);
215 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
216 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
217     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
218 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
219 static void nfsrv_freelayoutlist(nfsquad_t clientid);
220 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
221     int iomode);
222 static void nfsrv_freealllayouts(void);
223 static void nfsrv_freedevid(struct nfsdevice *ds);
224 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
225     struct nfsdevice **dsp);
226 static void nfsrv_deleteds(struct nfsdevice *fndds);
227 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
228 static void nfsrv_freealldevids(void);
229 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
230     int maxcnt, NFSPROC_T *p);
231 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
232     fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
233     NFSPROC_T *p);
234 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
235     NFSPROC_T *, struct nfslayout **lypp);
236 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
237 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
238     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
239 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
240     int mirrorcnt, uint64_t stripesiz, int stripecnt, fhandle_t *fhp,
241     fhandle_t *dsfhp, char *devid, fsid_t fs);
242 static int nfsrv_dontlayout(fhandle_t *fhp);
243 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
244     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
245     vnode_t *tvpp);
246 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
247 static int nfsrv_checkmachcred(int op, struct nfsrv_descript *nd,
248     struct nfsclient *clp);
249 static void nfsrv_issuedelegation(struct vnode *vp, struct nfsclient *clp,
250     struct nfsrv_descript *nd, int delegate, int writedeleg, int readonly,
251     u_quad_t filerev, uint64_t rdonly, struct nfsstate **new_delegp,
252     struct nfsstate *new_stp, struct nfslockfile *lfp, uint32_t *rflagsp,
253     nfsv4stateid_t *delegstateidp);
254 static void nfsrv_clientlock(bool mlocked);
255 static void nfsrv_clientunlock(bool mlocked);
256 
257 /*
258  * Lock the client structure, either with the mutex or the exclusive nfsd lock.
259  */
260 static void
nfsrv_clientlock(bool mlocked)261 nfsrv_clientlock(bool mlocked)
262 {
263 	int igotlock;
264 
265 	if (mlocked) {
266 		NFSLOCKSTATE();
267 	} else {
268 		NFSLOCKV4ROOTMUTEX();
269 		nfsv4_relref(&nfsv4rootfs_lock);
270 		do {
271 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
272 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
273 		} while (!igotlock);
274 		NFSUNLOCKV4ROOTMUTEX();
275 	}
276 }
277 
278 /*
279  * Unlock the client structure.
280  */
281 static void
nfsrv_clientunlock(bool mlocked)282 nfsrv_clientunlock(bool mlocked)
283 {
284 
285 	if (mlocked) {
286 		NFSUNLOCKSTATE();
287 	} else {
288 		NFSLOCKV4ROOTMUTEX();
289 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
290 		NFSUNLOCKV4ROOTMUTEX();
291 	}
292 }
293 
294 /*
295  * Scan the client list for a match and either return the current one,
296  * create a new entry or return an error.
297  * If returning a non-error, the clp structure must either be linked into
298  * the client list or free'd.
299  */
300 int
nfsrv_setclient(struct nfsrv_descript * nd,struct nfsclient ** new_clpp,nfsquad_t * clientidp,nfsquad_t * confirmp,NFSPROC_T * p)301 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
302     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
303 {
304 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
305 	int i, error = 0, ret;
306 	struct nfsstate *stp, *tstp;
307 #ifdef INET
308 	struct sockaddr_in *sin, *rin;
309 #endif
310 #ifdef INET6
311 	struct sockaddr_in6 *sin6, *rin6;
312 #endif
313 	struct nfsdsession *sep, *nsep;
314 	SVCXPRT *old_xprt;
315 	struct nfssessionhead old_sess;
316 	int zapit = 0, gotit, hasstate = 0;
317 	bool mlocked;
318 	static u_int64_t confirm_index = 0;
319 
320 	/*
321 	 * Check for state resource limit exceeded.
322 	 */
323 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
324 		error = NFSERR_RESOURCE;
325 		goto out;
326 	}
327 
328 	if (nfsrv_issuedelegs == 0 ||
329 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
330 		/*
331 		 * Don't do callbacks when delegations are disabled or
332 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
333 		 * If establishing a callback connection is attempted
334 		 * when a firewall is blocking the callback path, the
335 		 * server may wait too long for the connect attempt to
336 		 * succeed during the Open. Some clients, such as Linux,
337 		 * may timeout and give up on the Open before the server
338 		 * replies. Also, since AUTH_GSS callbacks are not
339 		 * yet interoperability tested, they might cause the
340 		 * server to crap out, if they get past the Init call to
341 		 * the client.
342 		 */
343 		new_clp->lc_program = 0;
344 
345 	mlocked = true;
346 	if (nfsrv_dolocallocks != 0)
347 		mlocked = false;
348 	/* Lock out other nfsd threads */
349 	nfsrv_clientlock(mlocked);
350 
351 	/*
352 	 * Search for a match in the client list.
353 	 */
354 	gotit = i = 0;
355 	while (i < nfsrv_clienthashsize && !gotit) {
356 	    LIST_FOREACH(clp, &VNET(nfsclienthash)[i], lc_hash) {
357 		if (new_clp->lc_idlen == clp->lc_idlen &&
358 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
359 			gotit = 1;
360 			break;
361 		}
362 	    }
363 	    if (gotit == 0)
364 		i++;
365 	}
366 	old_xprt = NULL;
367 	if (!gotit ||
368 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
369 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
370 			/*
371 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
372 			 * client is trying to update a confirmed clientid.
373 			 */
374 			nfsrv_clientunlock(mlocked);
375 			confirmp->lval[1] = 0;
376 			error = NFSERR_NOENT;
377 			goto out;
378 		}
379 		/*
380 		 * Get rid of the old one.
381 		 */
382 		if (i != nfsrv_clienthashsize) {
383 			LIST_REMOVE(clp, lc_hash);
384 			if (mlocked)
385 				nfsrv_cleanclient(clp, p, true, &old_xprt);
386 			else
387 				nfsrv_cleanclient(clp, p, false, NULL);
388 			nfsrv_freedeleglist(&clp->lc_deleg);
389 			nfsrv_freedeleglist(&clp->lc_olddeleg);
390 			zapit = 1;
391 		}
392 		/*
393 		 * Add it after assigning a client id to it.
394 		 */
395 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
396 		if ((nd->nd_flag & ND_NFSV41) != 0) {
397 			confirmp->lval[0] = ++confirm_index;
398 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
399 		} else
400 			confirmp->qval = new_clp->lc_confirm.qval =
401 			    ++confirm_index;
402 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
403 		    VNET(nfsrvboottime);
404 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
405 		    nfsrv_nextclientindex();
406 		new_clp->lc_stateindex = 0;
407 		new_clp->lc_statemaxindex = 0;
408 		new_clp->lc_prevsess = 0;
409 		new_clp->lc_cbref = 0;
410 		new_clp->lc_expiry = nfsrv_leaseexpiry();
411 		LIST_INIT(&new_clp->lc_open);
412 		LIST_INIT(&new_clp->lc_deleg);
413 		LIST_INIT(&new_clp->lc_olddeleg);
414 		LIST_INIT(&new_clp->lc_session);
415 		for (i = 0; i < nfsrv_statehashsize; i++)
416 			LIST_INIT(&new_clp->lc_stateid[i]);
417 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
418 		    lc_hash);
419 		VNET(nfsstatsv1_p)->srvclients++;
420 		nfsrv_openpluslock++;
421 		nfsrv_clients++;
422 		nfsrv_clientunlock(mlocked);
423 		if (zapit != 0) {
424 			if (old_xprt != NULL)
425 				SVC_RELEASE(old_xprt);
426 			nfsrv_zapclient(clp, p);
427 		}
428 		*new_clpp = NULL;
429 		goto out;
430 	}
431 
432 	/*
433 	 * Now, handle the cases where the id is already issued.
434 	 */
435 	if (nfsrv_notsamecredname(NFSV4OP_EXCHANGEID, nd, clp)) {
436 	    /*
437 	     * Check to see if there is expired state that should go away.
438 	     */
439 	    if (clp->lc_expiry < NFSD_MONOSEC &&
440 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
441 		if (mlocked)
442 		    nfsrv_cleanclient(clp, p, true, &old_xprt);
443 		else
444 		    nfsrv_cleanclient(clp, p, false, NULL);
445 		nfsrv_freedeleglist(&clp->lc_deleg);
446 	    }
447 
448 	    /*
449 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
450 	     * RFC3530 Sec. 8.1.2 last para.
451 	     */
452 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
453 		hasstate = 1;
454 	    } else if (LIST_EMPTY(&clp->lc_open)) {
455 		hasstate = 0;
456 	    } else {
457 		hasstate = 0;
458 		/* Look for an Open on the OpenOwner */
459 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
460 		    if (!LIST_EMPTY(&stp->ls_open)) {
461 			hasstate = 1;
462 			break;
463 		    }
464 		}
465 	    }
466 	    if (hasstate) {
467 		/*
468 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
469 		 * filling out the correct ipaddr and portnum.
470 		 */
471 		switch (clp->lc_req.nr_nam->sa_family) {
472 #ifdef INET
473 		case AF_INET:
474 			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
475 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
476 			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
477 			sin->sin_port = rin->sin_port;
478 			break;
479 #endif
480 #ifdef INET6
481 		case AF_INET6:
482 			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
483 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
484 			sin6->sin6_addr = rin6->sin6_addr;
485 			sin6->sin6_port = rin6->sin6_port;
486 			break;
487 #endif
488 		}
489 		nfsrv_clientunlock(mlocked);
490 		if (old_xprt != NULL)
491 			SVC_RELEASE(old_xprt);
492 		error = NFSERR_CLIDINUSE;
493 		goto out;
494 	    }
495 	}
496 
497 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
498 		/*
499 		 * If the verifier has changed, the client has rebooted
500 		 * and a new client id is issued. The old state info
501 		 * can be thrown away once the SetClientID_Confirm or
502 		 * Create_Session that confirms the clientid occurs.
503 		 */
504 		LIST_REMOVE(clp, lc_hash);
505 
506 		LIST_NEWHEAD(&old_sess, &clp->lc_session, sess_list);
507 
508 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
509 		if ((nd->nd_flag & ND_NFSV41) != 0) {
510 			confirmp->lval[0] = ++confirm_index;
511 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
512 		} else
513 			confirmp->qval = new_clp->lc_confirm.qval =
514 			    ++confirm_index;
515 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
516 		    VNET(nfsrvboottime);
517 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
518 		    nfsrv_nextclientindex();
519 		new_clp->lc_stateindex = 0;
520 		new_clp->lc_statemaxindex = 0;
521 		new_clp->lc_prevsess = 0;
522 		new_clp->lc_cbref = 0;
523 		new_clp->lc_expiry = nfsrv_leaseexpiry();
524 
525 		/*
526 		 * Save the state until confirmed.
527 		 */
528 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
529 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
530 			tstp->ls_clp = new_clp;
531 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
532 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
533 			tstp->ls_clp = new_clp;
534 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
535 		    ls_list);
536 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
537 			tstp->ls_clp = new_clp;
538 		for (i = 0; i < nfsrv_statehashsize; i++) {
539 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
540 			    &clp->lc_stateid[i], ls_hash);
541 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
542 				tstp->ls_clp = new_clp;
543 		}
544 		LIST_INIT(&new_clp->lc_session);
545 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
546 		    lc_hash);
547 		VNET(nfsstatsv1_p)->srvclients++;
548 		nfsrv_openpluslock++;
549 		nfsrv_clients++;
550 		if (!mlocked) {
551 			nfsrv_clientunlock(mlocked);
552 			NFSLOCKSTATE();
553 		}
554 
555 		/*
556 		 * Must wait until any outstanding callback on the old clp
557 		 * completes.
558 		 */
559 		while (clp->lc_cbref) {
560 			clp->lc_flags |= LCL_WAKEUPWANTED;
561 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PVFS,
562 			    "nfsd clp", 10 * hz);
563 		}
564 		NFSUNLOCKSTATE();
565 		if (old_xprt != NULL)
566 			SVC_RELEASE(old_xprt);
567 		/* Get rid of all sessions on this clientid. */
568 		LIST_FOREACH_SAFE(sep, &old_sess, sess_list, nsep) {
569 			ret = nfsrv_freesession(NULL, sep, NULL, false, NULL);
570 			if (ret != 0)
571 				printf("nfsrv_setclient: verifier changed free"
572 				    " session failed=%d\n", ret);
573 		}
574 
575 		nfsrv_zapclient(clp, p);
576 		*new_clpp = NULL;
577 		goto out;
578 	}
579 
580 	/* For NFSv4.1, mark that we found a confirmed clientid. */
581 	if ((nd->nd_flag & ND_NFSV41) != 0) {
582 		clientidp->lval[0] = clp->lc_clientid.lval[0];
583 		clientidp->lval[1] = clp->lc_clientid.lval[1];
584 		confirmp->lval[0] = 0;	/* Ignored by client */
585 		confirmp->lval[1] = 1;
586 	} else {
587 		/*
588 		 * id and verifier match, so update the net address info
589 		 * and get rid of any existing callback authentication
590 		 * handle, so a new one will be acquired.
591 		 */
592 		LIST_REMOVE(clp, lc_hash);
593 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
594 		new_clp->lc_expiry = nfsrv_leaseexpiry();
595 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
596 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
597 		    clp->lc_clientid.lval[0];
598 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
599 		    clp->lc_clientid.lval[1];
600 		new_clp->lc_delegtime = clp->lc_delegtime;
601 		new_clp->lc_stateindex = clp->lc_stateindex;
602 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
603 		new_clp->lc_cbref = 0;
604 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
605 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
606 			tstp->ls_clp = new_clp;
607 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
608 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
609 			tstp->ls_clp = new_clp;
610 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
611 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
612 			tstp->ls_clp = new_clp;
613 		for (i = 0; i < nfsrv_statehashsize; i++) {
614 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
615 			    &clp->lc_stateid[i], ls_hash);
616 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
617 				tstp->ls_clp = new_clp;
618 		}
619 		LIST_INIT(&new_clp->lc_session);
620 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
621 		    lc_hash);
622 		VNET(nfsstatsv1_p)->srvclients++;
623 		nfsrv_openpluslock++;
624 		nfsrv_clients++;
625 	}
626 	if (!mlocked)
627 		nfsrv_clientunlock(mlocked);
628 
629 	if ((nd->nd_flag & ND_NFSV41) == 0) {
630 		/*
631 		 * Must wait until any outstanding callback on the old clp
632 		 * completes.
633 		 */
634 		if (!mlocked)
635 			NFSLOCKSTATE();
636 		while (clp->lc_cbref) {
637 			clp->lc_flags |= LCL_WAKEUPWANTED;
638 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PVFS,
639 			    "nfsdclp", 10 * hz);
640 		}
641 		NFSUNLOCKSTATE();
642 		if (old_xprt != NULL)
643 			SVC_RELEASE(old_xprt);
644 		nfsrv_zapclient(clp, p);
645 		*new_clpp = NULL;
646 	} else {
647 		if (mlocked)
648 			NFSUNLOCKSTATE();
649 		if (old_xprt != NULL)
650 			SVC_RELEASE(old_xprt);
651 	}
652 
653 out:
654 	NFSEXITCODE2(error, nd);
655 	return (error);
656 }
657 
658 /*
659  * Check to see if the client id exists and optionally confirm it.
660  */
661 int
nfsrv_getclient(nfsquad_t clientid,int opflags,struct nfsclient ** clpp,struct nfsdsession * nsep,nfsquad_t confirm,uint32_t cbprogram,struct nfsrv_descript * nd,NFSPROC_T * p)662 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
663     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
664     struct nfsrv_descript *nd, NFSPROC_T *p)
665 {
666 	struct nfsclient *clp;
667 	struct nfsstate *stp;
668 	int i;
669 	struct nfsclienthashhead *hp;
670 	int error = 0, doneok, igotlock;
671 	struct nfssessionhash *shp;
672 	struct nfsdsession *sep;
673 	uint64_t sessid[2];
674 	CLIENT *client;
675 	SVCXPRT *old_xprt;
676 	bool mlocked, sess_replay;
677 	static uint64_t next_sess = 0;
678 
679 	if (clpp)
680 		*clpp = NULL;
681 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
682 	    opflags != CLOPS_RENEW) && VNET(nfsrvboottime) !=
683 	    clientid.lval[0]) {
684 		error = NFSERR_STALECLIENTID;
685 		goto out;
686 	}
687 
688 	/*
689 	 * If called with opflags == CLOPS_RENEW, the State Lock is
690 	 * already held. Otherwise, we need to get either that or,
691 	 * for the case of Confirm, lock out the nfsd threads.
692 	 */
693 	client = NULL;
694 	old_xprt = NULL;
695 	mlocked = true;
696 	if (nfsrv_dolocallocks != 0)
697 		mlocked = false;
698 	if (opflags & CLOPS_CONFIRM) {
699 		if (nsep != NULL &&
700 		    (nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
701 			client = (struct __rpc_client *)
702 			    clnt_bck_create(nd->nd_xprt->xp_socket,
703 			    cbprogram, NFSV4_CBVERS);
704 		if (mlocked) {
705 			nfsrv_clientlock(mlocked);
706 		} else {
707 			NFSLOCKV4ROOTMUTEX();
708 			nfsv4_relref(&nfsv4rootfs_lock);
709 			do {
710 				igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1,
711 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
712 			} while (!igotlock);
713 		}
714 		/*
715 		 * Create a new sessionid here, since we need to do it where
716 		 * there is a mutex held to serialize update of next_sess.
717 		 */
718 		if ((nd->nd_flag & ND_NFSV41) != 0) {
719 			sessid[0] = ++next_sess;
720 			sessid[1] = clientid.qval;
721 		}
722 		if (!mlocked)
723 			NFSUNLOCKV4ROOTMUTEX();
724 	} else if (opflags != CLOPS_RENEW) {
725 		NFSLOCKSTATE();
726 	}
727 
728 	/* For NFSv4.1, the clp is acquired from the associated session. */
729 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
730 	    opflags == CLOPS_RENEW) {
731 		clp = NULL;
732 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
733 			shp = NFSSESSIONHASH(nd->nd_sessionid);
734 			NFSLOCKSESSION(shp);
735 			sep = nfsrv_findsession(nd->nd_sessionid);
736 			if (sep != NULL)
737 				clp = sep->sess_clp;
738 			NFSUNLOCKSESSION(shp);
739 		}
740 	} else {
741 		hp = NFSCLIENTHASH(clientid);
742 		LIST_FOREACH(clp, hp, lc_hash) {
743 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
744 				break;
745 		}
746 	}
747 	if (clp == NULL) {
748 		if (opflags & CLOPS_CONFIRM)
749 			error = NFSERR_STALECLIENTID;
750 		else
751 			error = NFSERR_EXPIRED;
752 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
753 		/*
754 		 * If marked admin revoked, just return the error.
755 		 */
756 		error = NFSERR_ADMINREVOKED;
757 	}
758 	if (error) {
759 		if (opflags & CLOPS_CONFIRM) {
760 			nfsrv_clientunlock(mlocked);
761 			if (client != NULL)
762 				CLNT_RELEASE(client);
763 		} else if (opflags != CLOPS_RENEW) {
764 			NFSUNLOCKSTATE();
765 		}
766 		goto out;
767 	}
768 
769 	/*
770 	 * Perform any operations specified by the opflags.
771 	 */
772 	if (opflags & CLOPS_CONFIRM) {
773 		sess_replay = false;
774 		if ((nd->nd_flag & ND_NFSV41) != 0) {
775 		    /*
776 		     * For the case where lc_confirm.lval[0] == confirm.lval[0],
777 		     * use the new session, but with the previous sessionid.
778 		     * This is not exactly what the RFC describes, but should
779 		     * result in the same reply as the previous CreateSession.
780 		     */
781 		    if (clp->lc_confirm.lval[0] + 1 == confirm.lval[0]) {
782 			clp->lc_confirm.lval[0] = confirm.lval[0];
783 			clp->lc_prevsess = sessid[0];
784 		    } else if (clp->lc_confirm.lval[0] == confirm.lval[0]) {
785 			if (clp->lc_prevsess == 0)
786 			    error = NFSERR_SEQMISORDERED;
787 			else
788 			    sessid[0] = clp->lc_prevsess;
789 			sess_replay = true;
790 		    } else
791 			error = NFSERR_SEQMISORDERED;
792 		} else if ((nd->nd_flag & ND_NFSV41) == 0 &&
793 		     clp->lc_confirm.qval != confirm.qval)
794 			error = NFSERR_STALECLIENTID;
795 		if (error == 0 && nfsrv_notsamecredname(NFSV4OP_CREATESESSION,
796 		    nd, clp))
797 			error = NFSERR_CLIDINUSE;
798 
799 		if (!error) {
800 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
801 			LCL_NEEDSCONFIRM) {
802 			/*
803 			 * Hang onto the delegations (as old delegations)
804 			 * for an Open with CLAIM_DELEGATE_PREV unless in
805 			 * grace, but get rid of the rest of the state.
806 			 */
807 			if (mlocked)
808 				nfsrv_cleanclient(clp, p, true, &old_xprt);
809 			else
810 				nfsrv_cleanclient(clp, p, false, NULL);
811 			nfsrv_freedeleglist(&clp->lc_olddeleg);
812 			if (nfsrv_checkgrace(nd, clp, 0)) {
813 			    /* In grace, so just delete delegations */
814 			    nfsrv_freedeleglist(&clp->lc_deleg);
815 			} else {
816 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
817 				stp->ls_flags |= NFSLCK_OLDDELEG;
818 			    clp->lc_delegtime = NFSD_MONOSEC +
819 				nfsrv_lease + NFSRV_LEASEDELTA;
820 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
821 				ls_list);
822 			}
823 			if ((nd->nd_flag & ND_NFSV41) != 0)
824 			    clp->lc_program = cbprogram;
825 		    }
826 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
827 		    if (clp->lc_program)
828 			clp->lc_flags |= LCL_NEEDSCBNULL;
829 		    /* For NFSv4.1, link the session onto the client. */
830 		    if (nsep != NULL) {
831 			/* Hold a reference on the xprt for a backchannel. */
832 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
833 			    != 0 && !sess_replay) {
834 			    if (clp->lc_req.nr_client == NULL) {
835 				clp->lc_req.nr_client = client;
836 				client = NULL;
837 			    }
838 			    if (clp->lc_req.nr_client != NULL) {
839 				SVC_ACQUIRE(nd->nd_xprt);
840 				CLNT_ACQUIRE(clp->lc_req.nr_client);
841 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
842 				/* Disable idle timeout. */
843 				nd->nd_xprt->xp_idletimeout = 0;
844 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
845 			    } else
846 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
847 			}
848 			NFSBCOPY(sessid, nsep->sess_sessionid,
849 			    NFSX_V4SESSIONID);
850 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
851 			    NFSX_V4SESSIONID);
852 			if (!sess_replay) {
853 			    shp = NFSSESSIONHASH(nsep->sess_sessionid);
854 			    if (!mlocked)
855 				NFSLOCKSTATE();
856 			    NFSLOCKSESSION(shp);
857 			    LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
858 			    LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
859 			    nsep->sess_clp = clp;
860 			    NFSUNLOCKSESSION(shp);
861 			    if (!mlocked)
862 				NFSUNLOCKSTATE();
863 			}
864 		    }
865 		}
866 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
867 		error = NFSERR_EXPIRED;
868 	}
869 
870 	/*
871 	 * If called by the Renew Op, we must check the principal.
872 	 */
873 	if (!error && (opflags & CLOPS_RENEWOP)) {
874 	    if (nfsrv_notsamecredname(0, nd, clp)) {
875 		doneok = 0;
876 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
877 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
878 			if ((stp->ls_flags & NFSLCK_OPEN) &&
879 			    stp->ls_uid == nd->nd_cred->cr_uid) {
880 				doneok = 1;
881 				break;
882 			}
883 		    }
884 		}
885 		if (!doneok)
886 			error = NFSERR_ACCES;
887 	    }
888 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
889 		error = NFSERR_CBPATHDOWN;
890 	}
891 	if ((!error || error == NFSERR_CBPATHDOWN) &&
892 	     (opflags & CLOPS_RENEW)) {
893 		clp->lc_expiry = nfsrv_leaseexpiry();
894 	}
895 	if (opflags & CLOPS_CONFIRM) {
896 		nfsrv_clientunlock(mlocked);
897 		if (client != NULL)
898 			CLNT_RELEASE(client);
899 		if (old_xprt != NULL)
900 			SVC_RELEASE(old_xprt);
901 	} else if (opflags != CLOPS_RENEW) {
902 		NFSUNLOCKSTATE();
903 	}
904 	if (clpp)
905 		*clpp = clp;
906 
907 out:
908 	NFSEXITCODE2(error, nd);
909 	return (error);
910 }
911 
912 /*
913  * Perform the NFSv4.1 destroy clientid.
914  */
915 int
nfsrv_destroyclient(struct nfsrv_descript * nd,nfsquad_t clientid,NFSPROC_T * p)916 nfsrv_destroyclient(struct nfsrv_descript *nd, nfsquad_t clientid, NFSPROC_T *p)
917 {
918 	struct nfsclient *clp;
919 	struct nfsclienthashhead *hp;
920 	SVCXPRT *old_xprt;
921 	int error = 0, i;
922 	bool mlocked;
923 
924 	if (VNET(nfsrvboottime) != clientid.lval[0]) {
925 		error = NFSERR_STALECLIENTID;
926 		goto out;
927 	}
928 
929 	mlocked = true;
930 	if (nfsrv_dolocallocks != 0)
931 		mlocked = false;
932 	/* Lock out other nfsd threads */
933 	nfsrv_clientlock(mlocked);
934 
935 	hp = NFSCLIENTHASH(clientid);
936 	LIST_FOREACH(clp, hp, lc_hash) {
937 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
938 			break;
939 	}
940 	if (clp == NULL) {
941 		nfsrv_clientunlock(mlocked);
942 		/* Just return ok, since it is gone. */
943 		goto out;
944 	}
945 
946 	/* Check for the SP4_MACH_CRED case. */
947 	error = nfsrv_checkmachcred(NFSV4OP_DESTROYCLIENTID, nd, clp);
948 	if (error != 0) {
949 		nfsrv_clientunlock(mlocked);
950 		goto out;
951 	}
952 
953 	/*
954 	 * Free up all layouts on the clientid.  Should the client return the
955 	 * layouts?
956 	 */
957 	nfsrv_freelayoutlist(clientid);
958 
959 	/* Scan for state on the clientid. */
960 	for (i = 0; i < nfsrv_statehashsize; i++)
961 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
962 			nfsrv_clientunlock(mlocked);
963 			error = NFSERR_CLIENTIDBUSY;
964 			goto out;
965 		}
966 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
967 		nfsrv_clientunlock(mlocked);
968 		error = NFSERR_CLIENTIDBUSY;
969 		goto out;
970 	}
971 
972 	/* Destroy the clientid and return ok. */
973 	old_xprt = NULL;
974 	if (mlocked)
975 		nfsrv_cleanclient(clp, p, true, &old_xprt);
976 	else
977 		nfsrv_cleanclient(clp, p, false, NULL);
978 	nfsrv_freedeleglist(&clp->lc_deleg);
979 	nfsrv_freedeleglist(&clp->lc_olddeleg);
980 	LIST_REMOVE(clp, lc_hash);
981 	nfsrv_clientunlock(mlocked);
982 	if (old_xprt != NULL)
983 		SVC_RELEASE(old_xprt);
984 	nfsrv_zapclient(clp, p);
985 out:
986 	NFSEXITCODE2(error, nd);
987 	return (error);
988 }
989 
990 /*
991  * Called from the new nfssvc syscall to admin revoke a clientid.
992  * Returns 0 for success, error otherwise.
993  */
994 int
nfsrv_adminrevoke(struct nfsd_clid * revokep,NFSPROC_T * p)995 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
996 {
997 	struct nfsclient *clp = NULL;
998 	int i, error = 0;
999 	int gotit, igotlock;
1000 
1001 	/*
1002 	 * First, lock out the nfsd so that state won't change while the
1003 	 * revocation record is being written to the stable storage restart
1004 	 * file.
1005 	 */
1006 	NFSLOCKV4ROOTMUTEX();
1007 	do {
1008 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
1009 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
1010 	} while (!igotlock);
1011 	NFSUNLOCKV4ROOTMUTEX();
1012 
1013 	/*
1014 	 * Search for a match in the client list.
1015 	 */
1016 	gotit = i = 0;
1017 	while (i < nfsrv_clienthashsize && !gotit) {
1018 	    LIST_FOREACH(clp, &VNET(nfsclienthash)[i], lc_hash) {
1019 		if (revokep->nclid_idlen == clp->lc_idlen &&
1020 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
1021 			gotit = 1;
1022 			break;
1023 		}
1024 	    }
1025 	    i++;
1026 	}
1027 	if (!gotit) {
1028 		NFSLOCKV4ROOTMUTEX();
1029 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
1030 		NFSUNLOCKV4ROOTMUTEX();
1031 		error = EPERM;
1032 		goto out;
1033 	}
1034 
1035 	/*
1036 	 * Now, write out the revocation record
1037 	 */
1038 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
1039 	nfsrv_backupstable();
1040 
1041 	/*
1042 	 * and clear out the state, marking the clientid revoked.
1043 	 */
1044 	clp->lc_flags &= ~LCL_CALLBACKSON;
1045 	clp->lc_flags |= LCL_ADMINREVOKED;
1046 	nfsrv_cleanclient(clp, p, false, NULL);
1047 	nfsrv_freedeleglist(&clp->lc_deleg);
1048 	nfsrv_freedeleglist(&clp->lc_olddeleg);
1049 	NFSLOCKV4ROOTMUTEX();
1050 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
1051 	NFSUNLOCKV4ROOTMUTEX();
1052 
1053 out:
1054 	NFSEXITCODE(error);
1055 	return (error);
1056 }
1057 
1058 /*
1059  * Dump out stats for all clients. Called from nfssvc(2), that is used
1060  * nfsstatsv1.
1061  */
1062 void
nfsrv_dumpclients(struct nfsd_dumpclients * dumpp,int maxcnt)1063 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
1064 {
1065 	struct nfsclient *clp;
1066 	int i = 0, cnt = 0;
1067 
1068 	/*
1069 	 * First, get a reference on the nfsv4rootfs_lock so that an
1070 	 * exclusive lock cannot be acquired while dumping the clients.
1071 	 */
1072 	NFSLOCKV4ROOTMUTEX();
1073 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1074 	NFSUNLOCKV4ROOTMUTEX();
1075 	NFSLOCKSTATE();
1076 	/*
1077 	 * Rattle through the client lists until done.
1078 	 */
1079 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
1080 	    clp = LIST_FIRST(&VNET(nfsclienthash)[i]);
1081 	    while (clp != LIST_END(&VNET(nfsclienthash)[i]) && cnt <
1082 		maxcnt) {
1083 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
1084 		cnt++;
1085 		clp = LIST_NEXT(clp, lc_hash);
1086 	    }
1087 	    i++;
1088 	}
1089 	if (cnt < maxcnt)
1090 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
1091 	NFSUNLOCKSTATE();
1092 	NFSLOCKV4ROOTMUTEX();
1093 	nfsv4_relref(&nfsv4rootfs_lock);
1094 	NFSUNLOCKV4ROOTMUTEX();
1095 }
1096 
1097 /*
1098  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
1099  */
1100 static void
nfsrv_dumpaclient(struct nfsclient * clp,struct nfsd_dumpclients * dumpp)1101 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
1102 {
1103 	struct nfsstate *stp, *openstp, *lckownstp;
1104 	struct nfslock *lop;
1105 	sa_family_t af;
1106 #ifdef INET
1107 	struct sockaddr_in *rin;
1108 #endif
1109 #ifdef INET6
1110 	struct sockaddr_in6 *rin6;
1111 #endif
1112 
1113 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
1114 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
1115 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
1116 	dumpp->ndcl_flags = clp->lc_flags;
1117 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
1118 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
1119 	af = clp->lc_req.nr_nam->sa_family;
1120 	dumpp->ndcl_addrfam = af;
1121 	switch (af) {
1122 #ifdef INET
1123 	case AF_INET:
1124 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
1125 		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
1126 		break;
1127 #endif
1128 #ifdef INET6
1129 	case AF_INET6:
1130 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1131 		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1132 		break;
1133 #endif
1134 	}
1135 
1136 	/*
1137 	 * Now, scan the state lists and total up the opens and locks.
1138 	 */
1139 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1140 	    dumpp->ndcl_nopenowners++;
1141 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1142 		dumpp->ndcl_nopens++;
1143 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1144 		    dumpp->ndcl_nlockowners++;
1145 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1146 			dumpp->ndcl_nlocks++;
1147 		    }
1148 		}
1149 	    }
1150 	}
1151 
1152 	/*
1153 	 * and the delegation lists.
1154 	 */
1155 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1156 	    dumpp->ndcl_ndelegs++;
1157 	}
1158 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1159 	    dumpp->ndcl_nolddelegs++;
1160 	}
1161 }
1162 
1163 /*
1164  * Dump out lock stats for a file.
1165  */
1166 void
nfsrv_dumplocks(vnode_t vp,struct nfsd_dumplocks * ldumpp,int maxcnt,NFSPROC_T * p)1167 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1168     NFSPROC_T *p)
1169 {
1170 	struct nfsstate *stp;
1171 	struct nfslock *lop;
1172 	int cnt = 0;
1173 	struct nfslockfile *lfp;
1174 	sa_family_t af;
1175 #ifdef INET
1176 	struct sockaddr_in *rin;
1177 #endif
1178 #ifdef INET6
1179 	struct sockaddr_in6 *rin6;
1180 #endif
1181 	int ret;
1182 	fhandle_t nfh;
1183 
1184 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1185 	/*
1186 	 * First, get a reference on the nfsv4rootfs_lock so that an
1187 	 * exclusive lock on it cannot be acquired while dumping the locks.
1188 	 */
1189 	NFSLOCKV4ROOTMUTEX();
1190 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1191 	NFSUNLOCKV4ROOTMUTEX();
1192 	NFSLOCKSTATE();
1193 	if (!ret)
1194 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1195 	if (ret) {
1196 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1197 		NFSUNLOCKSTATE();
1198 		NFSLOCKV4ROOTMUTEX();
1199 		nfsv4_relref(&nfsv4rootfs_lock);
1200 		NFSUNLOCKV4ROOTMUTEX();
1201 		return;
1202 	}
1203 
1204 	/*
1205 	 * For each open share on file, dump it out.
1206 	 */
1207 	stp = LIST_FIRST(&lfp->lf_open);
1208 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1209 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1210 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1211 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1212 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1213 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1214 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1215 		    stp->ls_openowner->ls_ownerlen;
1216 		NFSBCOPY(stp->ls_openowner->ls_owner,
1217 		    ldumpp[cnt].ndlck_owner.nclid_id,
1218 		    stp->ls_openowner->ls_ownerlen);
1219 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1220 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1221 		    stp->ls_clp->lc_idlen);
1222 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1223 		ldumpp[cnt].ndlck_addrfam = af;
1224 		switch (af) {
1225 #ifdef INET
1226 		case AF_INET:
1227 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1228 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1229 			break;
1230 #endif
1231 #ifdef INET6
1232 		case AF_INET6:
1233 			rin6 = (struct sockaddr_in6 *)
1234 			    stp->ls_clp->lc_req.nr_nam;
1235 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1236 			break;
1237 #endif
1238 		}
1239 		stp = LIST_NEXT(stp, ls_file);
1240 		cnt++;
1241 	}
1242 
1243 	/*
1244 	 * and all locks.
1245 	 */
1246 	lop = LIST_FIRST(&lfp->lf_lock);
1247 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1248 		stp = lop->lo_stp;
1249 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1250 		ldumpp[cnt].ndlck_first = lop->lo_first;
1251 		ldumpp[cnt].ndlck_end = lop->lo_end;
1252 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1253 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1254 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1255 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1256 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1257 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1258 		    stp->ls_ownerlen);
1259 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1260 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1261 		    stp->ls_clp->lc_idlen);
1262 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1263 		ldumpp[cnt].ndlck_addrfam = af;
1264 		switch (af) {
1265 #ifdef INET
1266 		case AF_INET:
1267 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1268 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1269 			break;
1270 #endif
1271 #ifdef INET6
1272 		case AF_INET6:
1273 			rin6 = (struct sockaddr_in6 *)
1274 			    stp->ls_clp->lc_req.nr_nam;
1275 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1276 			break;
1277 #endif
1278 		}
1279 		lop = LIST_NEXT(lop, lo_lckfile);
1280 		cnt++;
1281 	}
1282 
1283 	/*
1284 	 * and the delegations.
1285 	 */
1286 	stp = LIST_FIRST(&lfp->lf_deleg);
1287 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1288 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1289 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1290 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1291 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1292 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1293 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1294 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1295 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1296 		    stp->ls_clp->lc_idlen);
1297 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1298 		ldumpp[cnt].ndlck_addrfam = af;
1299 		switch (af) {
1300 #ifdef INET
1301 		case AF_INET:
1302 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1303 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1304 			break;
1305 #endif
1306 #ifdef INET6
1307 		case AF_INET6:
1308 			rin6 = (struct sockaddr_in6 *)
1309 			    stp->ls_clp->lc_req.nr_nam;
1310 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1311 			break;
1312 #endif
1313 		}
1314 		stp = LIST_NEXT(stp, ls_file);
1315 		cnt++;
1316 	}
1317 
1318 	/*
1319 	 * If list isn't full, mark end of list by setting the client name
1320 	 * to zero length.
1321 	 */
1322 	if (cnt < maxcnt)
1323 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1324 	NFSUNLOCKSTATE();
1325 	NFSLOCKV4ROOTMUTEX();
1326 	nfsv4_relref(&nfsv4rootfs_lock);
1327 	NFSUNLOCKV4ROOTMUTEX();
1328 }
1329 
1330 /*
1331  * Server timer routine. It can scan any linked list, so long
1332  * as it holds the spin/mutex lock and there is no exclusive lock on
1333  * nfsv4rootfs_lock.
1334  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1335  *  to do this from a callout, since the spin locks work. For
1336  *  Darwin, I'm not sure what will work correctly yet.)
1337  * Should be called once per second.
1338  */
1339 void
nfsrv_servertimer(void * arg __unused)1340 nfsrv_servertimer(void *arg __unused)
1341 {
1342 	struct nfsclient *clp, *nclp;
1343 	struct nfsstate *stp, *nstp;
1344 	int got_ref, i;
1345 
1346 	/*
1347 	 * Make sure nfsboottime is set. This is used by V3 as well
1348 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1349 	 * only used by the V4 server for leases.
1350 	 */
1351 	if (nfsboottime.tv_sec == 0)
1352 		NFSSETBOOTTIME(nfsboottime);
1353 
1354 	/*
1355 	 * If server hasn't started yet, just return.
1356 	 */
1357 	NFSLOCKSTATE();
1358 	if (VNET(nfsrv_stablefirst).nsf_eograce == 0) {
1359 		NFSUNLOCKSTATE();
1360 		return;
1361 	}
1362 	if (!(VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE)) {
1363 		if (!(VNET(nfsrv_stablefirst).nsf_flags &
1364 		      NFSNSF_GRACEOVER) &&
1365 		    NFSD_MONOSEC > VNET(nfsrv_stablefirst).nsf_eograce)
1366 			VNET(nfsrv_stablefirst).nsf_flags |=
1367 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1368 		NFSUNLOCKSTATE();
1369 		return;
1370 	}
1371 
1372 	/*
1373 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1374 	 * no nfsd thread can acquire an exclusive lock on it before this
1375 	 * call is done. If it is already exclusively locked, just return.
1376 	 */
1377 	NFSLOCKV4ROOTMUTEX();
1378 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1379 	NFSUNLOCKV4ROOTMUTEX();
1380 	if (got_ref == 0) {
1381 		NFSUNLOCKSTATE();
1382 		return;
1383 	}
1384 
1385 	/*
1386 	 * For each client...
1387 	 */
1388 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1389 	    clp = LIST_FIRST(&VNET(nfsclienthash)[i]);
1390 	    while (clp != LIST_END(&VNET(nfsclienthash)[i])) {
1391 		nclp = LIST_NEXT(clp, lc_hash);
1392 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1393 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1394 			 && ((LIST_EMPTY(&clp->lc_deleg)
1395 			      && LIST_EMPTY(&clp->lc_open)) ||
1396 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1397 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1398 			(clp->lc_expiry < NFSD_MONOSEC &&
1399 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1400 			/*
1401 			 * Lease has expired several nfsrv_lease times ago:
1402 			 * PLUS
1403 			 *    - no state is associated with it
1404 			 *    OR
1405 			 *    - above high water mark for number of clients
1406 			 *      (nfsrv_clienthighwater should be large enough
1407 			 *       that this only occurs when clients fail to
1408 			 *       use the same nfs_client_id4.id. Maybe somewhat
1409 			 *       higher that the maximum number of clients that
1410 			 *       will mount this server?)
1411 			 * OR
1412 			 * Lease has expired a very long time ago
1413 			 * OR
1414 			 * Lease has expired PLUS the number of opens + locks
1415 			 * has exceeded 90% of capacity
1416 			 *
1417 			 * --> Mark for expiry. The actual expiry will be done
1418 			 *     by an nfsd sometime soon.
1419 			 */
1420 			clp->lc_flags |= LCL_EXPIREIT;
1421 			VNET(nfsrv_stablefirst).nsf_flags |=
1422 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1423 		    } else {
1424 			/*
1425 			 * If there are no opens, increment no open tick cnt
1426 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1427 			 * otherwise, if there is an open, reset no open time
1428 			 * Hopefully, this will avoid excessive re-creation
1429 			 * of open owners and subsequent open confirms.
1430 			 */
1431 			stp = LIST_FIRST(&clp->lc_open);
1432 			while (stp != LIST_END(&clp->lc_open)) {
1433 				nstp = LIST_NEXT(stp, ls_list);
1434 				if (LIST_EMPTY(&stp->ls_open)) {
1435 					stp->ls_noopens++;
1436 					if (stp->ls_noopens > NFSNOOPEN ||
1437 					    (nfsrv_openpluslock * 2) >
1438 					    nfsrv_v4statelimit)
1439 						VNET(nfsrv_stablefirst).nsf_flags |=
1440 							NFSNSF_NOOPENS;
1441 				} else {
1442 					stp->ls_noopens = 0;
1443 				}
1444 				stp = nstp;
1445 			}
1446 		    }
1447 		}
1448 		clp = nclp;
1449 	    }
1450 	}
1451 	NFSUNLOCKSTATE();
1452 	NFSLOCKV4ROOTMUTEX();
1453 	nfsv4_relref(&nfsv4rootfs_lock);
1454 	NFSUNLOCKV4ROOTMUTEX();
1455 }
1456 
1457 /*
1458  * The following set of functions free up the various data structures.
1459  */
1460 /*
1461  * Clear out all open/lock state related to this nfsclient.
1462  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1463  * there are no other active nfsd threads.
1464  */
1465 void
nfsrv_cleanclient(struct nfsclient * clp,NFSPROC_T * p,bool locked,SVCXPRT ** old_xprtp)1466 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p, bool locked,
1467     SVCXPRT **old_xprtp)
1468 {
1469 	struct nfsstate *stp, *nstp;
1470 	struct nfsdsession *sep, *nsep;
1471 
1472 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
1473 		if (locked)
1474 			nfsrv_freeopenowner(stp, 0, p);
1475 		else
1476 			nfsrv_freeopenowner(stp, 1, p);
1477 	}
1478 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1479 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1480 			(void)nfsrv_freesession(NULL, sep, NULL, locked,
1481 			    old_xprtp);
1482 }
1483 
1484 /*
1485  * Free a client that has been cleaned. It should also already have been
1486  * removed from the lists.
1487  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1488  *  softclock interrupts are enabled.)
1489  */
1490 void
nfsrv_zapclient(struct nfsclient * clp,NFSPROC_T * p)1491 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1492 {
1493 
1494 #ifdef notyet
1495 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1496 	     (LCL_GSS | LCL_CALLBACKSON) &&
1497 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1498 	    clp->lc_handlelen > 0) {
1499 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1500 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1501 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1502 			NULL, 0, NULL, NULL, NULL, 0, p);
1503 	}
1504 #endif
1505 	newnfs_disconnect(NULL, &clp->lc_req);
1506 	free(clp->lc_req.nr_nam, M_SONAME);
1507 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1508 	crfree(clp->lc_req.nr_cred);
1509 	free(clp->lc_stateid, M_NFSDCLIENT);
1510 	free(clp, M_NFSDCLIENT);
1511 	NFSLOCKSTATE();
1512 	VNET(nfsstatsv1_p)->srvclients--;
1513 	nfsrv_openpluslock--;
1514 	nfsrv_clients--;
1515 	NFSUNLOCKSTATE();
1516 }
1517 
1518 /*
1519  * Free a list of delegation state structures.
1520  * (This function will also free all nfslockfile structures that no
1521  *  longer have associated state.)
1522  */
1523 void
nfsrv_freedeleglist(struct nfsstatehead * sthp)1524 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1525 {
1526 	struct nfsstate *stp, *nstp;
1527 
1528 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1529 		nfsrv_freedeleg(stp);
1530 	}
1531 	LIST_INIT(sthp);
1532 }
1533 
1534 /*
1535  * Free up a delegation.
1536  */
1537 static void
nfsrv_freedeleg(struct nfsstate * stp)1538 nfsrv_freedeleg(struct nfsstate *stp)
1539 {
1540 	struct nfslockfile *lfp;
1541 
1542 	LIST_REMOVE(stp, ls_hash);
1543 	LIST_REMOVE(stp, ls_list);
1544 	LIST_REMOVE(stp, ls_file);
1545 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1546 		nfsrv_writedelegcnt--;
1547 	lfp = stp->ls_lfp;
1548 	if (LIST_EMPTY(&lfp->lf_open) &&
1549 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1550 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1551 	    lfp->lf_usecount == 0 &&
1552 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1553 		nfsrv_freenfslockfile(lfp);
1554 	free(stp, M_NFSDSTATE);
1555 	VNET(nfsstatsv1_p)->srvdelegates--;
1556 	nfsrv_openpluslock--;
1557 	nfsrv_delegatecnt--;
1558 }
1559 
1560 /*
1561  * This function frees an open owner and all associated opens.
1562  */
1563 static void
nfsrv_freeopenowner(struct nfsstate * stp,int cansleep,NFSPROC_T * p)1564 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1565 {
1566 	struct nfsstate *nstp, *tstp;
1567 
1568 	LIST_REMOVE(stp, ls_list);
1569 	/*
1570 	 * Now, free all associated opens.
1571 	 */
1572 	nstp = LIST_FIRST(&stp->ls_open);
1573 	while (nstp != LIST_END(&stp->ls_open)) {
1574 		tstp = nstp;
1575 		nstp = LIST_NEXT(nstp, ls_list);
1576 		nfsrv_freeopen(tstp, NULL, cansleep, p);
1577 	}
1578 	if (stp->ls_op)
1579 		nfsrvd_derefcache(stp->ls_op);
1580 	free(stp, M_NFSDSTATE);
1581 	VNET(nfsstatsv1_p)->srvopenowners--;
1582 	nfsrv_openpluslock--;
1583 }
1584 
1585 /*
1586  * This function frees an open (nfsstate open structure) with all associated
1587  * lock_owners and locks. It also frees the nfslockfile structure iff there
1588  * are no other opens on the file.
1589  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1590  */
1591 static void
nfsrv_freeopen(struct nfsstate * stp,vnode_t vp,int cansleep,NFSPROC_T * p)1592 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1593 {
1594 	struct nfsstate *nstp, *tstp;
1595 	struct nfslockfile *lfp;
1596 
1597 	LIST_REMOVE(stp, ls_hash);
1598 	LIST_REMOVE(stp, ls_list);
1599 	LIST_REMOVE(stp, ls_file);
1600 
1601 	lfp = stp->ls_lfp;
1602 	/*
1603 	 * Now, free all lockowners associated with this open.
1604 	 * Note that, if vp != NULL, nfsrv_freelockowner() will
1605 	 * not call nfsrv_freeallnfslocks(), so it needs to be called, below.
1606 	 */
1607 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1608 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1609 
1610 	if (vp != NULL) {
1611 		KASSERT(cansleep != 0, ("nfsrv_freeopen: cansleep == 0"));
1612 		mtx_assert(NFSSTATEMUTEXPTR, MA_OWNED);
1613 		/*
1614 		 * Only called with vp != NULL for Close when
1615 		 * vfs.nfsd.enable_locallocks != 0.
1616 		 * Lock the lfp so that it will not go away and do the
1617 		 * nfsrv_freeallnfslocks() call that was not done by
1618 		 * nfsrv_freelockowner().
1619 		 */
1620 		nfsrv_locklf(lfp);
1621 		NFSUNLOCKSTATE();
1622 		NFSVOPUNLOCK(vp);
1623 		nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1624 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1625 		NFSLOCKSTATE();
1626 		nfsrv_unlocklf(lfp);
1627 	}
1628 
1629 	/*
1630 	 * The nfslockfile is freed here if there are no locks
1631 	 * associated with the open.
1632 	 * If there are locks associated with the open, the
1633 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1634 	 */
1635 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1636 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1637 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1638 	    lfp->lf_usecount == 0 &&
1639 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1640 		nfsrv_freenfslockfile(lfp);
1641 	free(stp, M_NFSDSTATE);
1642 	VNET(nfsstatsv1_p)->srvopens--;
1643 	nfsrv_openpluslock--;
1644 }
1645 
1646 /*
1647  * Frees a lockowner and all associated locks.
1648  */
1649 static void
nfsrv_freelockowner(struct nfsstate * stp,vnode_t vp,int cansleep,NFSPROC_T * p)1650 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1651     NFSPROC_T *p)
1652 {
1653 
1654 	LIST_REMOVE(stp, ls_hash);
1655 	LIST_REMOVE(stp, ls_list);
1656 	if (vp == NULL)
1657 		nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1658 	if (stp->ls_op)
1659 		nfsrvd_derefcache(stp->ls_op);
1660 	free(stp, M_NFSDSTATE);
1661 	VNET(nfsstatsv1_p)->srvlockowners--;
1662 	nfsrv_openpluslock--;
1663 }
1664 
1665 /*
1666  * Free all the nfs locks on a lockowner.
1667  */
1668 static void
nfsrv_freeallnfslocks(struct nfsstate * stp,vnode_t vp,int cansleep,NFSPROC_T * p)1669 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1670     NFSPROC_T *p)
1671 {
1672 	struct nfslock *lop, *nlop;
1673 	struct nfsrollback *rlp, *nrlp;
1674 	struct nfslockfile *lfp = NULL;
1675 	int gottvp = 0;
1676 	vnode_t tvp = NULL;
1677 	uint64_t first, end;
1678 
1679 	if (vp != NULL)
1680 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1681 	lop = LIST_FIRST(&stp->ls_lock);
1682 	while (lop != LIST_END(&stp->ls_lock)) {
1683 		nlop = LIST_NEXT(lop, lo_lckowner);
1684 		/*
1685 		 * Since all locks should be for the same file, lfp should
1686 		 * not change.
1687 		 */
1688 		if (lfp == NULL)
1689 			lfp = lop->lo_lfp;
1690 		else if (lfp != lop->lo_lfp)
1691 			panic("allnfslocks");
1692 		/*
1693 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1694 		 * from the file handle. This only occurs when called from
1695 		 * nfsrv_cleanclient().
1696 		 */
1697 		if (gottvp == 0) {
1698 			if (nfsrv_dolocallocks == 0)
1699 				tvp = NULL;
1700 			else if (vp == NULL && cansleep != 0) {
1701 				tvp = nfsvno_getvp(&lfp->lf_fh);
1702 				if (tvp != NULL)
1703 					NFSVOPUNLOCK(tvp);
1704 			} else
1705 				tvp = vp;
1706 			gottvp = 1;
1707 		}
1708 
1709 		if (tvp != NULL) {
1710 			if (cansleep == 0)
1711 				panic("allnfs2");
1712 			first = lop->lo_first;
1713 			end = lop->lo_end;
1714 			nfsrv_freenfslock(lop);
1715 			nfsrv_localunlock(tvp, lfp, first, end, p);
1716 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1717 			    nrlp)
1718 				free(rlp, M_NFSDROLLBACK);
1719 			LIST_INIT(&lfp->lf_rollback);
1720 		} else
1721 			nfsrv_freenfslock(lop);
1722 		lop = nlop;
1723 	}
1724 	if (vp == NULL && tvp != NULL)
1725 		vrele(tvp);
1726 }
1727 
1728 /*
1729  * Free an nfslock structure.
1730  */
1731 static void
nfsrv_freenfslock(struct nfslock * lop)1732 nfsrv_freenfslock(struct nfslock *lop)
1733 {
1734 
1735 	if (lop->lo_lckfile.le_prev != NULL) {
1736 		LIST_REMOVE(lop, lo_lckfile);
1737 		VNET(nfsstatsv1_p)->srvlocks--;
1738 		nfsrv_openpluslock--;
1739 	}
1740 	LIST_REMOVE(lop, lo_lckowner);
1741 	free(lop, M_NFSDLOCK);
1742 }
1743 
1744 /*
1745  * This function frees an nfslockfile structure.
1746  */
1747 static void
nfsrv_freenfslockfile(struct nfslockfile * lfp)1748 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1749 {
1750 
1751 	LIST_REMOVE(lfp, lf_hash);
1752 	free(lfp, M_NFSDLOCKFILE);
1753 }
1754 
1755 /*
1756  * This function looks up an nfsstate structure via stateid.
1757  */
1758 static int
nfsrv_getstate(struct nfsclient * clp,nfsv4stateid_t * stateidp,__unused u_int32_t flags,struct nfsstate ** stpp)1759 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1760     struct nfsstate **stpp)
1761 {
1762 	struct nfsstate *stp;
1763 	struct nfsstatehead *hp;
1764 	int error = 0;
1765 
1766 	*stpp = NULL;
1767 	hp = NFSSTATEHASH(clp, *stateidp);
1768 	LIST_FOREACH(stp, hp, ls_hash) {
1769 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1770 			NFSX_STATEIDOTHER))
1771 			break;
1772 	}
1773 
1774 	/*
1775 	 * If no state id in list, return NFSERR_BADSTATEID.
1776 	 */
1777 	if (stp == LIST_END(hp)) {
1778 		error = NFSERR_BADSTATEID;
1779 		goto out;
1780 	}
1781 	*stpp = stp;
1782 
1783 out:
1784 	NFSEXITCODE(error);
1785 	return (error);
1786 }
1787 
1788 /*
1789  * This function gets an nfsstate structure via owner string.
1790  */
1791 static void
nfsrv_getowner(struct nfsstatehead * hp,struct nfsstate * new_stp,struct nfsstate ** stpp)1792 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1793     struct nfsstate **stpp)
1794 {
1795 	struct nfsstate *stp;
1796 
1797 	*stpp = NULL;
1798 	LIST_FOREACH(stp, hp, ls_list) {
1799 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1800 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1801 			*stpp = stp;
1802 			return;
1803 		}
1804 	}
1805 }
1806 
1807 /*
1808  * Lock control function called to update lock status.
1809  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1810  * that one isn't to be created and an NFSERR_xxx for other errors.
1811  * The structures new_stp and new_lop are passed in as pointers that should
1812  * be set to NULL if the structure is used and shouldn't be free'd.
1813  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1814  * never used and can safely be allocated on the stack. For all other
1815  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1816  * in case they are used.
1817  */
1818 int
nfsrv_lockctrl(vnode_t vp,struct nfsstate ** new_stpp,struct nfslock ** new_lopp,struct nfslockconflict * cfp,nfsquad_t clientid,nfsv4stateid_t * stateidp,__unused struct nfsexstuff * exp,struct nfsrv_descript * nd,NFSPROC_T * p)1819 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1820     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1821     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1822     __unused struct nfsexstuff *exp,
1823     struct nfsrv_descript *nd, NFSPROC_T *p)
1824 {
1825 	struct nfslock *lop;
1826 	struct nfsstate *new_stp = *new_stpp;
1827 	struct nfslock *new_lop = *new_lopp;
1828 	struct nfsstate *tstp, *mystp, *nstp;
1829 	int specialid = 0;
1830 	struct nfslockfile *lfp;
1831 	struct nfslock *other_lop = NULL;
1832 	struct nfsstate *stp, *lckstp = NULL;
1833 	struct nfsclient *clp = NULL;
1834 	u_int32_t bits;
1835 	int error = 0, haslock = 0, ret, reterr;
1836 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1837 	fhandle_t nfh;
1838 	uint64_t first, end;
1839 	uint32_t lock_flags;
1840 
1841 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1842 		/*
1843 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1844 		 * let reads with all 1s go ahead.
1845 		 */
1846 		if (new_stp->ls_stateid.seqid == 0x0 &&
1847 		    new_stp->ls_stateid.other[0] == 0x0 &&
1848 		    new_stp->ls_stateid.other[1] == 0x0 &&
1849 		    new_stp->ls_stateid.other[2] == 0x0)
1850 			specialid = 1;
1851 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1852 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1853 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1854 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1855 			specialid = 2;
1856 	}
1857 
1858 	/*
1859 	 * Check for restart conditions (client and server).
1860 	 */
1861 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1862 	    &new_stp->ls_stateid, specialid);
1863 	if (error)
1864 		goto out;
1865 
1866 	/*
1867 	 * Check for state resource limit exceeded.
1868 	 */
1869 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1870 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1871 		error = NFSERR_RESOURCE;
1872 		goto out;
1873 	}
1874 
1875 	/*
1876 	 * For the lock case, get another nfslock structure,
1877 	 * just in case we need it.
1878 	 * Malloc now, before we start sifting through the linked lists,
1879 	 * in case we have to wait for memory.
1880 	 */
1881 tryagain:
1882 	if (new_stp->ls_flags & NFSLCK_LOCK)
1883 		other_lop = malloc(sizeof (struct nfslock),
1884 		    M_NFSDLOCK, M_WAITOK);
1885 	filestruct_locked = 0;
1886 	reterr = 0;
1887 	lfp = NULL;
1888 
1889 	/*
1890 	 * Get the lockfile structure for CFH now, so we can do a sanity
1891 	 * check against the stateid, before incrementing the seqid#, since
1892 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1893 	 * shouldn't be incremented for this case.
1894 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1895 	 * will be handled later.
1896 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1897 	 * lock the nfslockfile structure.
1898 	 */
1899 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1900 	NFSLOCKSTATE();
1901 	if (getlckret == 0) {
1902 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1903 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1904 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1905 			    &lfp, &nfh, 1);
1906 			if (getlckret == 0)
1907 				filestruct_locked = 1;
1908 		} else
1909 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1910 			    &lfp, &nfh, 0);
1911 	}
1912 	if (getlckret != 0 && getlckret != -1)
1913 		reterr = getlckret;
1914 
1915 	if (filestruct_locked != 0) {
1916 		LIST_INIT(&lfp->lf_rollback);
1917 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1918 			/*
1919 			 * For local locking, do the advisory locking now, so
1920 			 * that any conflict can be detected. A failure later
1921 			 * can be rolled back locally. If an error is returned,
1922 			 * struct nfslockfile has been unlocked and any local
1923 			 * locking rolled back.
1924 			 */
1925 			NFSUNLOCKSTATE();
1926 			if (vnode_unlocked == 0) {
1927 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1928 				vnode_unlocked = 1;
1929 				NFSVOPUNLOCK(vp);
1930 			}
1931 			reterr = nfsrv_locallock(vp, lfp,
1932 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1933 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1934 			NFSLOCKSTATE();
1935 		}
1936 	}
1937 
1938 	if (specialid == 0) {
1939 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1940 		/*
1941 		 * RFC 3530 does not list LockT as an op that renews a
1942 		 * lease, but the consensus seems to be that it is ok
1943 		 * for a server to do so.
1944 		 */
1945 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1946 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1947 
1948 		/*
1949 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1950 		 * error returns for LockT, just go ahead and test for a lock,
1951 		 * since there are no locks for this client, but other locks
1952 		 * can conflict. (ie. same client will always be false)
1953 		 */
1954 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1955 		    error = 0;
1956 		lckstp = new_stp;
1957 	    } else {
1958 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1959 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1960 	      if (error == 0)
1961 		/*
1962 		 * Look up the stateid
1963 		 */
1964 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1965 		  new_stp->ls_flags, &stp);
1966 	      /*
1967 	       * do some sanity checks for an unconfirmed open or a
1968 	       * stateid that refers to the wrong file, for an open stateid
1969 	       */
1970 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1971 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1972 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1973 		      /*
1974 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1975 		       * The only exception is using SETATTR with SIZE.
1976 		       * */
1977                     if ((new_stp->ls_flags &
1978                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1979 			     error = NFSERR_BADSTATEID;
1980 	      }
1981 
1982 	      /*
1983 	       * Sanity check the stateid for the Lock/LockU cases.
1984 	       */
1985 	      if (error == 0 && (new_stp->ls_flags & NFSLCK_LOCK) != 0 &&
1986 		  (((new_stp->ls_flags & NFSLCK_OPENTOLOCK) != 0 &&
1987 		    (stp->ls_flags & NFSLCK_OPEN) == 0) ||
1988 		   ((new_stp->ls_flags & NFSLCK_OPENTOLOCK) == 0 &&
1989 		    (stp->ls_flags & NFSLCK_LOCK) == 0)))
1990 			error = NFSERR_BADSTATEID;
1991 	      if (error == 0 && (new_stp->ls_flags & NFSLCK_UNLOCK) != 0 &&
1992 		  (stp->ls_flags & NFSLCK_LOCK) == 0)
1993 			error = NFSERR_BADSTATEID;
1994 
1995 		/* Sanity check the delegation stateid. */
1996 		if (error == 0 &&
1997 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1998 		  getlckret == 0 && stp->ls_lfp != lfp)
1999 			error = NFSERR_BADSTATEID;
2000 
2001 	      /*
2002 	       * If the lockowner stateid doesn't refer to the same file,
2003 	       * I believe that is considered ok, since some clients will
2004 	       * only create a single lockowner and use that for all locks
2005 	       * on all files.
2006 	       * For now, log it as a diagnostic, instead of considering it
2007 	       * a BadStateid.
2008 	       */
2009 	      if (error == 0 && (stp->ls_flags &
2010 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
2011 		  getlckret == 0 && stp->ls_lfp != lfp) {
2012 #ifdef DIAGNOSTIC
2013 		  printf("Got a lock statid for different file open\n");
2014 #endif
2015 		  /*
2016 		  error = NFSERR_BADSTATEID;
2017 		  */
2018 	      }
2019 
2020 	      if (error == 0) {
2021 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
2022 			/*
2023 			 * If haslock set, we've already checked the seqid.
2024 			 */
2025 			if (!haslock) {
2026 			    if (stp->ls_flags & NFSLCK_OPEN)
2027 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2028 				    stp->ls_openowner, new_stp->ls_op);
2029 			    else
2030 				error = NFSERR_BADSTATEID;
2031 			}
2032 			if (!error)
2033 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
2034 			if (lckstp) {
2035 			    /*
2036 			     * For NFSv4.1 and NFSv4.2 allow an
2037 			     * open_to_lock_owner when the lock_owner already
2038 			     * exists.  Just clear NFSLCK_OPENTOLOCK so that
2039 			     * a new lock_owner will not be created.
2040 			     * RFC7530 states that the error for NFSv4.0
2041 			     * is NFS4ERR_BAD_SEQID.
2042 			     */
2043 			    if ((nd->nd_flag & ND_NFSV41) != 0)
2044 				new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK;
2045 			    else
2046 				error = NFSERR_BADSEQID;
2047 			} else
2048 			    lckstp = new_stp;
2049 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
2050 			/*
2051 			 * If haslock set, ditto above.
2052 			 */
2053 			if (!haslock) {
2054 			    if (stp->ls_flags & NFSLCK_OPEN)
2055 				error = NFSERR_BADSTATEID;
2056 			    else
2057 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2058 				    stp, new_stp->ls_op);
2059 			}
2060 			lckstp = stp;
2061 		    } else {
2062 			lckstp = stp;
2063 		    }
2064 	      }
2065 	      /*
2066 	       * If the seqid part of the stateid isn't the same, return
2067 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
2068 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
2069 	       * nfsrv_returnoldstateid is set. (The consensus on the email
2070 	       * list was that most clients would prefer to not receive
2071 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
2072 	       * is what will happen, so I use the nfsrv_returnoldstateid to
2073 	       * allow for either server configuration.)
2074 	       */
2075 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
2076 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
2077 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2078 		    nfsrv_returnoldstateid)) ||
2079 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
2080 		    new_stp->ls_stateid.seqid != 0)))
2081 		    error = NFSERR_OLDSTATEID;
2082 	    }
2083 	}
2084 
2085 	/*
2086 	 * Now we can check for grace.
2087 	 */
2088 	if (!error)
2089 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2090 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2091 		nfsrv_checkstable(clp))
2092 		error = NFSERR_NOGRACE;
2093 	/*
2094 	 * If we successfully Reclaimed state, note that.
2095 	 */
2096 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
2097 		nfsrv_markstable(clp);
2098 
2099 	/*
2100 	 * At this point, either error == NFSERR_BADSTATEID or the
2101 	 * seqid# has been updated, so we can return any error.
2102 	 * If error == 0, there may be an error in:
2103 	 *    nd_repstat - Set by the calling function.
2104 	 *    reterr - Set above, if getting the nfslockfile structure
2105 	 *       or acquiring the local lock failed.
2106 	 *    (If both of these are set, nd_repstat should probably be
2107 	 *     returned, since that error was detected before this
2108 	 *     function call.)
2109 	 */
2110 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
2111 		if (error == 0) {
2112 			if (nd->nd_repstat != 0)
2113 				error = nd->nd_repstat;
2114 			else
2115 				error = reterr;
2116 		}
2117 		if (filestruct_locked != 0) {
2118 			/* Roll back local locks. */
2119 			NFSUNLOCKSTATE();
2120 			if (vnode_unlocked == 0) {
2121 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
2122 				vnode_unlocked = 1;
2123 				NFSVOPUNLOCK(vp);
2124 			}
2125 			nfsrv_locallock_rollback(vp, lfp, p);
2126 			NFSLOCKSTATE();
2127 			nfsrv_unlocklf(lfp);
2128 		}
2129 		NFSUNLOCKSTATE();
2130 		goto out;
2131 	}
2132 
2133 	/*
2134 	 * Check the nfsrv_getlockfile return.
2135 	 * Returned -1 if no structure found.
2136 	 */
2137 	if (getlckret == -1) {
2138 		error = NFSERR_EXPIRED;
2139 		/*
2140 		 * Called from lockt, so no lock is OK.
2141 		 */
2142 		if (new_stp->ls_flags & NFSLCK_TEST) {
2143 			error = 0;
2144 		} else if (new_stp->ls_flags &
2145 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
2146 			/*
2147 			 * Called to check for a lock, OK if the stateid is all
2148 			 * 1s or all 0s, but there should be an nfsstate
2149 			 * otherwise.
2150 			 * (ie. If there is no open, I'll assume no share
2151 			 *  deny bits.)
2152 			 */
2153 			if (specialid)
2154 				error = 0;
2155 			else
2156 				error = NFSERR_BADSTATEID;
2157 		}
2158 		NFSUNLOCKSTATE();
2159 		goto out;
2160 	}
2161 
2162 	/*
2163 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
2164 	 * For NFSLCK_CHECK, allow a read if write access is granted,
2165 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
2166 	 * which implies a conflicting deny can't exist.
2167 	 */
2168 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2169 	    /*
2170 	     * Four kinds of state id:
2171 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2172 	     * - stateid for an open
2173 	     * - stateid for a delegation
2174 	     * - stateid for a lock owner
2175 	     */
2176 	    if (!specialid) {
2177 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2178 		    delegation = 1;
2179 		    mystp = stp;
2180 		    nfsrv_delaydelegtimeout(stp);
2181 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
2182 		    mystp = stp;
2183 		} else {
2184 		    mystp = stp->ls_openstp;
2185 		}
2186 		/*
2187 		 * If locking or checking, require correct access
2188 		 * bit set.
2189 		 */
2190 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2191 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2192 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2193 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2194 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2195 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
2196 		     nfsrv_allowreadforwriteopen == 0) ||
2197 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2198 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2199 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2200 			if (filestruct_locked != 0) {
2201 				/* Roll back local locks. */
2202 				NFSUNLOCKSTATE();
2203 				if (vnode_unlocked == 0) {
2204 					ASSERT_VOP_ELOCKED(vp,
2205 					    "nfsrv_lockctrl3");
2206 					vnode_unlocked = 1;
2207 					NFSVOPUNLOCK(vp);
2208 				}
2209 				nfsrv_locallock_rollback(vp, lfp, p);
2210 				NFSLOCKSTATE();
2211 				nfsrv_unlocklf(lfp);
2212 			}
2213 			NFSUNLOCKSTATE();
2214 			error = NFSERR_OPENMODE;
2215 			goto out;
2216 		}
2217 	    } else
2218 		mystp = NULL;
2219 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2220 		/*
2221 		 * Check for a conflicting deny bit.
2222 		 */
2223 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2224 		    if (tstp != mystp) {
2225 			bits = tstp->ls_flags;
2226 			bits >>= NFSLCK_SHIFT;
2227 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2228 			    KASSERT(vnode_unlocked == 0,
2229 				("nfsrv_lockctrl: vnode unlocked1"));
2230 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2231 				vp, p);
2232 			    if (ret == 1) {
2233 				/*
2234 				* nfsrv_clientconflict unlocks state
2235 				 * when it returns non-zero.
2236 				 */
2237 				lckstp = NULL;
2238 				goto tryagain;
2239 			    }
2240 			    if (ret == 0)
2241 				NFSUNLOCKSTATE();
2242 			    if (ret == 2)
2243 				error = NFSERR_PERM;
2244 			    else
2245 				error = NFSERR_OPENMODE;
2246 			    goto out;
2247 			}
2248 		    }
2249 		}
2250 
2251 		/* We're outta here */
2252 		NFSUNLOCKSTATE();
2253 		goto out;
2254 	    }
2255 	}
2256 
2257 	/*
2258 	 * For setattr, just get rid of all the Delegations for other clients.
2259 	 */
2260 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2261 		KASSERT(vnode_unlocked == 0,
2262 		    ("nfsrv_lockctrl: vnode unlocked2"));
2263 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2264 		if (ret) {
2265 			/*
2266 			 * nfsrv_cleandeleg() unlocks state when it
2267 			 * returns non-zero.
2268 			 */
2269 			if (ret == -1) {
2270 				lckstp = NULL;
2271 				goto tryagain;
2272 			}
2273 			error = ret;
2274 			goto out;
2275 		}
2276 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2277 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2278 		     LIST_EMPTY(&lfp->lf_deleg))) {
2279 			NFSUNLOCKSTATE();
2280 			goto out;
2281 		}
2282 	}
2283 
2284 	/*
2285 	 * Check for a conflicting delegation. If one is found, call
2286 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2287 	 * been set yet, it will get the lock. Otherwise, it will recall
2288 	 * the delegation. Then, we try try again...
2289 	 * I currently believe the conflict algorithm to be:
2290 	 * For Lock Ops (Lock/LockT/LockU)
2291 	 * - there is a conflict iff a different client has a write delegation
2292 	 * For Reading (Read Op)
2293 	 * - there is a conflict iff a different client has a write delegation
2294 	 *   (the specialids are always a different client)
2295 	 * For Writing (Write/Setattr of size)
2296 	 * - there is a conflict if a different client has any delegation
2297 	 * - there is a conflict if the same client has a read delegation
2298 	 *   (I don't understand why this isn't allowed, but that seems to be
2299 	 *    the current consensus?)
2300 	 */
2301 	tstp = LIST_FIRST(&lfp->lf_deleg);
2302 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2303 	    nstp = LIST_NEXT(tstp, ls_file);
2304 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2305 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2306 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2307 		  clp != tstp->ls_clp &&
2308 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2309 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2310 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2311 		  (clp != tstp->ls_clp ||
2312 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2313 		ret = 0;
2314 		if (filestruct_locked != 0) {
2315 			/* Roll back local locks. */
2316 			NFSUNLOCKSTATE();
2317 			if (vnode_unlocked == 0) {
2318 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2319 				NFSVOPUNLOCK(vp);
2320 			}
2321 			nfsrv_locallock_rollback(vp, lfp, p);
2322 			NFSLOCKSTATE();
2323 			nfsrv_unlocklf(lfp);
2324 			NFSUNLOCKSTATE();
2325 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2326 			vnode_unlocked = 0;
2327 			if (VN_IS_DOOMED(vp))
2328 				ret = NFSERR_SERVERFAULT;
2329 			NFSLOCKSTATE();
2330 		}
2331 		if (ret == 0)
2332 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2333 		if (ret) {
2334 		    /*
2335 		     * nfsrv_delegconflict unlocks state when it
2336 		     * returns non-zero, which it always does.
2337 		     */
2338 		    if (other_lop) {
2339 			free(other_lop, M_NFSDLOCK);
2340 			other_lop = NULL;
2341 		    }
2342 		    if (ret == -1) {
2343 			lckstp = NULL;
2344 			goto tryagain;
2345 		    }
2346 		    error = ret;
2347 		    goto out;
2348 		}
2349 		/* Never gets here. */
2350 	    }
2351 	    tstp = nstp;
2352 	}
2353 
2354 	/*
2355 	 * Handle the unlock case by calling nfsrv_updatelock().
2356 	 * (Should I have done some access checking above for unlock? For now,
2357 	 *  just let it happen.)
2358 	 */
2359 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2360 		first = new_lop->lo_first;
2361 		end = new_lop->lo_end;
2362 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2363 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2364 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2365 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2366 		stateidp->other[0] = stp->ls_stateid.other[0];
2367 		stateidp->other[1] = stp->ls_stateid.other[1];
2368 		stateidp->other[2] = stp->ls_stateid.other[2];
2369 		if (filestruct_locked != 0) {
2370 			NFSUNLOCKSTATE();
2371 			if (vnode_unlocked == 0) {
2372 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2373 				vnode_unlocked = 1;
2374 				NFSVOPUNLOCK(vp);
2375 			}
2376 			/* Update the local locks. */
2377 			nfsrv_localunlock(vp, lfp, first, end, p);
2378 			NFSLOCKSTATE();
2379 			nfsrv_unlocklf(lfp);
2380 		}
2381 		NFSUNLOCKSTATE();
2382 		goto out;
2383 	}
2384 
2385 	/*
2386 	 * Search for a conflicting lock. A lock conflicts if:
2387 	 * - the lock range overlaps and
2388 	 * - at least one lock is a write lock and
2389 	 * - it is not owned by the same lock owner
2390 	 */
2391 	if (!delegation) {
2392 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2393 	    if (new_lop->lo_end > lop->lo_first &&
2394 		new_lop->lo_first < lop->lo_end &&
2395 		(new_lop->lo_flags == NFSLCK_WRITE ||
2396 		 lop->lo_flags == NFSLCK_WRITE) &&
2397 		lckstp != lop->lo_stp &&
2398 		(clp != lop->lo_stp->ls_clp ||
2399 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2400 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2401 		    lckstp->ls_ownerlen))) {
2402 		if (other_lop) {
2403 		    free(other_lop, M_NFSDLOCK);
2404 		    other_lop = NULL;
2405 		}
2406 		if (vnode_unlocked != 0)
2407 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2408 			NULL, p);
2409 		else
2410 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2411 			vp, p);
2412 		if (ret == 1) {
2413 		    if (filestruct_locked != 0) {
2414 			if (vnode_unlocked == 0) {
2415 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2416 				NFSVOPUNLOCK(vp);
2417 			}
2418 			/* Roll back local locks. */
2419 			nfsrv_locallock_rollback(vp, lfp, p);
2420 			NFSLOCKSTATE();
2421 			nfsrv_unlocklf(lfp);
2422 			NFSUNLOCKSTATE();
2423 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2424 			vnode_unlocked = 0;
2425 			if (VN_IS_DOOMED(vp)) {
2426 				error = NFSERR_SERVERFAULT;
2427 				goto out;
2428 			}
2429 		    }
2430 		    /*
2431 		     * nfsrv_clientconflict() unlocks state when it
2432 		     * returns non-zero.
2433 		     */
2434 		    lckstp = NULL;
2435 		    goto tryagain;
2436 		}
2437 		/*
2438 		 * Found a conflicting lock, so record the conflict and
2439 		 * return the error.
2440 		 */
2441 		if (cfp != NULL && ret == 0) {
2442 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2443 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2444 		    cfp->cl_first = lop->lo_first;
2445 		    cfp->cl_end = lop->lo_end;
2446 		    cfp->cl_flags = lop->lo_flags;
2447 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2448 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2449 			cfp->cl_ownerlen);
2450 		}
2451 		if (ret == 2)
2452 		    error = NFSERR_PERM;
2453 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2454 		    error = NFSERR_RECLAIMCONFLICT;
2455 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2456 		    error = NFSERR_LOCKED;
2457 		else
2458 		    error = NFSERR_DENIED;
2459 		if (filestruct_locked != 0 && ret == 0) {
2460 			/* Roll back local locks. */
2461 			NFSUNLOCKSTATE();
2462 			if (vnode_unlocked == 0) {
2463 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2464 				vnode_unlocked = 1;
2465 				NFSVOPUNLOCK(vp);
2466 			}
2467 			nfsrv_locallock_rollback(vp, lfp, p);
2468 			NFSLOCKSTATE();
2469 			nfsrv_unlocklf(lfp);
2470 		}
2471 		if (ret == 0)
2472 			NFSUNLOCKSTATE();
2473 		goto out;
2474 	    }
2475 	  }
2476 	}
2477 
2478 	/*
2479 	 * We only get here if there was no lock that conflicted.
2480 	 */
2481 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2482 		NFSUNLOCKSTATE();
2483 		goto out;
2484 	}
2485 
2486 	/*
2487 	 * We only get here when we are creating or modifying a lock.
2488 	 * There are two variants:
2489 	 * - exist_lock_owner where lock_owner exists
2490 	 * - open_to_lock_owner with new lock_owner
2491 	 */
2492 	first = new_lop->lo_first;
2493 	end = new_lop->lo_end;
2494 	lock_flags = new_lop->lo_flags;
2495 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2496 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2497 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2498 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2499 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2500 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2501 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2502 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2503 	} else {
2504 		/*
2505 		 * The new open_to_lock_owner case.
2506 		 * Link the new nfsstate into the lists.
2507 		 */
2508 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2509 		nfsrvd_refcache(new_stp->ls_op);
2510 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2511 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2512 		    clp->lc_clientid.lval[0];
2513 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2514 		    clp->lc_clientid.lval[1];
2515 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2516 		    nfsrv_nextstateindex(clp);
2517 		new_stp->ls_clp = clp;
2518 		LIST_INIT(&new_stp->ls_lock);
2519 		new_stp->ls_openstp = stp;
2520 		new_stp->ls_lfp = lfp;
2521 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2522 		    lfp);
2523 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2524 		    new_stp, ls_hash);
2525 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2526 		*new_lopp = NULL;
2527 		*new_stpp = NULL;
2528 		VNET(nfsstatsv1_p)->srvlockowners++;
2529 		nfsrv_openpluslock++;
2530 	}
2531 	if (filestruct_locked != 0) {
2532 		NFSUNLOCKSTATE();
2533 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2534 		NFSLOCKSTATE();
2535 		nfsrv_unlocklf(lfp);
2536 	}
2537 	NFSUNLOCKSTATE();
2538 
2539 out:
2540 	if (haslock) {
2541 		NFSLOCKV4ROOTMUTEX();
2542 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2543 		NFSUNLOCKV4ROOTMUTEX();
2544 	}
2545 	if (vnode_unlocked != 0) {
2546 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2547 		if (error == 0 && VN_IS_DOOMED(vp))
2548 			error = NFSERR_SERVERFAULT;
2549 	}
2550 	if (other_lop)
2551 		free(other_lop, M_NFSDLOCK);
2552 	NFSEXITCODE2(error, nd);
2553 	return (error);
2554 }
2555 
2556 /*
2557  * Check for state errors for Open.
2558  * repstat is passed back out as an error if more critical errors
2559  * are not detected.
2560  */
2561 int
nfsrv_opencheck(nfsquad_t clientid,nfsv4stateid_t * stateidp,struct nfsstate * new_stp,vnode_t vp,struct nfsrv_descript * nd,NFSPROC_T * p,int repstat)2562 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2563     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2564     NFSPROC_T *p, int repstat)
2565 {
2566 	struct nfsstate *stp, *nstp;
2567 	struct nfsclient *clp;
2568 	struct nfsstate *ownerstp;
2569 	struct nfslockfile *lfp, *new_lfp;
2570 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2571 
2572 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2573 		readonly = 1;
2574 	/*
2575 	 * Check for restart conditions (client and server).
2576 	 */
2577 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2578 		&new_stp->ls_stateid, 0);
2579 	if (error)
2580 		goto out;
2581 
2582 	/*
2583 	 * Check for state resource limit exceeded.
2584 	 * Technically this should be SMP protected, but the worst
2585 	 * case error is "out by one or two" on the count when it
2586 	 * returns NFSERR_RESOURCE and the limit is just a rather
2587 	 * arbitrary high water mark, so no harm is done.
2588 	 */
2589 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2590 		error = NFSERR_RESOURCE;
2591 		goto out;
2592 	}
2593 
2594 tryagain:
2595 	new_lfp = malloc(sizeof (struct nfslockfile),
2596 	    M_NFSDLOCKFILE, M_WAITOK);
2597 	if (vp)
2598 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2599 		    NULL, p);
2600 	NFSLOCKSTATE();
2601 	/*
2602 	 * Get the nfsclient structure.
2603 	 */
2604 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2605 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2606 
2607 	/*
2608 	 * Look up the open owner. See if it needs confirmation and
2609 	 * check the seq#, as required.
2610 	 */
2611 	if (!error)
2612 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2613 
2614 	if (!error && ownerstp) {
2615 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2616 		    new_stp->ls_op);
2617 		/*
2618 		 * If the OpenOwner hasn't been confirmed, assume the
2619 		 * old one was a replay and this one is ok.
2620 		 * See: RFC3530 Sec. 14.2.18.
2621 		 */
2622 		if (error == NFSERR_BADSEQID &&
2623 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2624 			error = 0;
2625 	}
2626 
2627 	/*
2628 	 * Check for grace.
2629 	 */
2630 	if (!error)
2631 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2632 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2633 		nfsrv_checkstable(clp))
2634 		error = NFSERR_NOGRACE;
2635 
2636 	/*
2637 	 * If none of the above errors occurred, let repstat be
2638 	 * returned.
2639 	 */
2640 	if (repstat && !error)
2641 		error = repstat;
2642 	if (error) {
2643 		NFSUNLOCKSTATE();
2644 		if (haslock) {
2645 			NFSLOCKV4ROOTMUTEX();
2646 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2647 			NFSUNLOCKV4ROOTMUTEX();
2648 		}
2649 		free(new_lfp, M_NFSDLOCKFILE);
2650 		goto out;
2651 	}
2652 
2653 	/*
2654 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2655 	 * (This always happens on the first pass, so haslock must be 0.)
2656 	 */
2657 	if (vp == NULL) {
2658 		NFSUNLOCKSTATE();
2659 		free(new_lfp, M_NFSDLOCKFILE);
2660 		goto out;
2661 	}
2662 
2663 	/*
2664 	 * Get the structure for the underlying file.
2665 	 */
2666 	if (getfhret)
2667 		error = getfhret;
2668 	else
2669 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2670 		    NULL, 0);
2671 	if (new_lfp)
2672 		free(new_lfp, M_NFSDLOCKFILE);
2673 	if (error) {
2674 		NFSUNLOCKSTATE();
2675 		if (haslock) {
2676 			NFSLOCKV4ROOTMUTEX();
2677 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2678 			NFSUNLOCKV4ROOTMUTEX();
2679 		}
2680 		goto out;
2681 	}
2682 
2683 	/*
2684 	 * Search for a conflicting open/share.
2685 	 */
2686 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2687 	    /*
2688 	     * For Delegate_Cur, search for the matching Delegation,
2689 	     * which indicates no conflict.
2690 	     * An old delegation should have been recovered by the
2691 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2692 	     * it match and return NFSERR_EXPIRED. Should I let it
2693 	     * match?
2694 	     */
2695 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2696 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2697 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2698 		    stateidp->seqid == 0) ||
2699 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2700 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2701 			  NFSX_STATEIDOTHER))
2702 			break;
2703 	    }
2704 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2705 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2706 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2707 		NFSUNLOCKSTATE();
2708 		if (haslock) {
2709 			NFSLOCKV4ROOTMUTEX();
2710 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2711 			NFSUNLOCKV4ROOTMUTEX();
2712 		}
2713 		error = NFSERR_EXPIRED;
2714 		goto out;
2715 	    }
2716 	}
2717 
2718 	/*
2719 	 * Check for access/deny bit conflicts. I check for the same
2720 	 * owner as well, in case the client didn't bother.
2721 	 */
2722 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2723 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2724 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2725 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2726 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2727 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2728 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2729 			if (ret == 1) {
2730 				/*
2731 				 * nfsrv_clientconflict() unlocks
2732 				 * state when it returns non-zero.
2733 				 */
2734 				goto tryagain;
2735 			}
2736 			if (ret == 2)
2737 				error = NFSERR_PERM;
2738 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2739 				error = NFSERR_RECLAIMCONFLICT;
2740 			else
2741 				error = NFSERR_SHAREDENIED;
2742 			if (ret == 0)
2743 				NFSUNLOCKSTATE();
2744 			if (haslock) {
2745 				NFSLOCKV4ROOTMUTEX();
2746 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2747 				NFSUNLOCKV4ROOTMUTEX();
2748 			}
2749 			goto out;
2750 		}
2751 	}
2752 
2753 	/*
2754 	 * Check for a conflicting delegation. If one is found, call
2755 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2756 	 * been set yet, it will get the lock. Otherwise, it will recall
2757 	 * the delegation. Then, we try try again...
2758 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2759 	 *  isn't a conflict.)
2760 	 * I currently believe the conflict algorithm to be:
2761 	 * For Open with Read Access and Deny None
2762 	 * - there is a conflict iff a different client has a write delegation
2763 	 * For Open with other Write Access or any Deny except None
2764 	 * - there is a conflict if a different client has any delegation
2765 	 * - there is a conflict if the same client has a read delegation
2766 	 *   (The current consensus is that this last case should be
2767 	 *    considered a conflict since the client with a read delegation
2768 	 *    could have done an Open with ReadAccess and WriteDeny
2769 	 *    locally and then not have checked for the WriteDeny.)
2770 	 *    The exception is a NFSv4.1/4.2 client that has requested
2771 	 *    an atomic upgrade to a write delegation.
2772 	 * Don't check for a Reclaim, since that will be dealt with
2773 	 * by nfsrv_openctrl().
2774 	 */
2775 	if (!(new_stp->ls_flags &
2776 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2777 	    stp = LIST_FIRST(&lfp->lf_deleg);
2778 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2779 		nstp = LIST_NEXT(stp, ls_file);
2780 		if ((readonly && stp->ls_clp != clp &&
2781 		     (stp->ls_flags & NFSLCK_DELEGWRITE) != 0) ||
2782 		    (!readonly && (stp->ls_clp != clp ||
2783 		     ((stp->ls_flags & NFSLCK_DELEGREAD) != 0 &&
2784 		      (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)))) {
2785 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2786 			if (ret) {
2787 			    /*
2788 			     * nfsrv_delegconflict() unlocks state
2789 			     * when it returns non-zero.
2790 			     */
2791 			    if (ret == -1)
2792 				goto tryagain;
2793 			    error = ret;
2794 			    goto out;
2795 			}
2796 		}
2797 		stp = nstp;
2798 	    }
2799 	}
2800 	NFSUNLOCKSTATE();
2801 	if (haslock) {
2802 		NFSLOCKV4ROOTMUTEX();
2803 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2804 		NFSUNLOCKV4ROOTMUTEX();
2805 	}
2806 
2807 out:
2808 	NFSEXITCODE2(error, nd);
2809 	return (error);
2810 }
2811 
2812 /*
2813  * Open control function to create/update open state for an open.
2814  */
2815 int
nfsrv_openctrl(struct nfsrv_descript * nd,vnode_t vp,struct nfsstate ** new_stpp,nfsquad_t clientid,nfsv4stateid_t * stateidp,nfsv4stateid_t * delegstateidp,u_int32_t * rflagsp,struct nfsexstuff * exp,NFSPROC_T * p,u_quad_t filerev)2816 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2817     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2818     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2819     NFSPROC_T *p, u_quad_t filerev)
2820 {
2821 	struct nfsstate *new_stp = *new_stpp;
2822 	struct nfsstate *stp, *nstp;
2823 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2824 	struct nfslockfile *lfp, *new_lfp;
2825 	struct nfsclient *clp;
2826 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2827 	int readonly = 0, cbret = 1, getfhret = 0;
2828 	int gotstate = 0, len = 0;
2829 	u_char *clidp = NULL;
2830 
2831 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2832 		readonly = 1;
2833 	/*
2834 	 * Check for restart conditions (client and server).
2835 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2836 	 * If an error does show up, return NFSERR_EXPIRED, since the
2837 	 * the seqid# has already been incremented.
2838 	 */
2839 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2840 	    &new_stp->ls_stateid, 0);
2841 	if (error) {
2842 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2843 		    error);
2844 		error = NFSERR_EXPIRED;
2845 		goto out;
2846 	}
2847 
2848 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2849 tryagain:
2850 	new_lfp = malloc(sizeof (struct nfslockfile),
2851 	    M_NFSDLOCKFILE, M_WAITOK);
2852 	new_open = malloc(sizeof (struct nfsstate),
2853 	    M_NFSDSTATE, M_WAITOK);
2854 	new_deleg = malloc(sizeof (struct nfsstate),
2855 	    M_NFSDSTATE, M_WAITOK);
2856 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2857 	    NULL, p);
2858 	NFSLOCKSTATE();
2859 	/*
2860 	 * Get the client structure. Since the linked lists could be changed
2861 	 * by other nfsd processes if this process does a tsleep(), one of
2862 	 * two things must be done.
2863 	 * 1 - don't tsleep()
2864 	 * or
2865 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2866 	 *     before using the lists, since this lock stops the other
2867 	 *     nfsd. This should only be used for rare cases, since it
2868 	 *     essentially single threads the nfsd.
2869 	 *     At this time, it is only done for cases where the stable
2870 	 *     storage file must be written prior to completion of state
2871 	 *     expiration.
2872 	 */
2873 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2874 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2875 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2876 	    clp->lc_program) {
2877 		/*
2878 		 * This happens on the first open for a client
2879 		 * that supports callbacks.
2880 		 */
2881 		NFSUNLOCKSTATE();
2882 		/*
2883 		 * Although nfsrv_docallback() will sleep, clp won't
2884 		 * go away, since they are only removed when the
2885 		 * nfsv4_lock() has blocked the nfsd threads. The
2886 		 * fields in clp can change, but having multiple
2887 		 * threads do this Null callback RPC should be
2888 		 * harmless.
2889 		 */
2890 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2891 		    NULL, 0, NULL, NULL, NULL, 0, p);
2892 		NFSLOCKSTATE();
2893 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2894 		if (!cbret)
2895 			clp->lc_flags |= LCL_CALLBACKSON;
2896 	}
2897 
2898 	/*
2899 	 * Look up the open owner. See if it needs confirmation and
2900 	 * check the seq#, as required.
2901 	 */
2902 	if (!error)
2903 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2904 
2905 	if (error) {
2906 		NFSUNLOCKSTATE();
2907 		printf("Nfsd: openctrl unexpected state err=%d\n",
2908 			error);
2909 		free(new_lfp, M_NFSDLOCKFILE);
2910 		free(new_open, M_NFSDSTATE);
2911 		free(new_deleg, M_NFSDSTATE);
2912 		if (haslock) {
2913 			NFSLOCKV4ROOTMUTEX();
2914 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2915 			NFSUNLOCKV4ROOTMUTEX();
2916 		}
2917 		error = NFSERR_EXPIRED;
2918 		goto out;
2919 	}
2920 
2921 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2922 		nfsrv_markstable(clp);
2923 
2924 	/*
2925 	 * Get the structure for the underlying file.
2926 	 */
2927 	if (getfhret)
2928 		error = getfhret;
2929 	else
2930 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2931 		    NULL, 0);
2932 	if (new_lfp)
2933 		free(new_lfp, M_NFSDLOCKFILE);
2934 	if (error) {
2935 		NFSUNLOCKSTATE();
2936 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2937 		    error);
2938 		free(new_open, M_NFSDSTATE);
2939 		free(new_deleg, M_NFSDSTATE);
2940 		if (haslock) {
2941 			NFSLOCKV4ROOTMUTEX();
2942 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2943 			NFSUNLOCKV4ROOTMUTEX();
2944 		}
2945 		goto out;
2946 	}
2947 
2948 	/*
2949 	 * Search for a conflicting open/share.
2950 	 */
2951 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2952 	    /*
2953 	     * For Delegate_Cur, search for the matching Delegation,
2954 	     * which indicates no conflict.
2955 	     * An old delegation should have been recovered by the
2956 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2957 	     * it match and return NFSERR_EXPIRED. Should I let it
2958 	     * match?
2959 	     */
2960 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2961 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2962 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2963 		    stateidp->seqid == 0) ||
2964 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2965 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2966 			NFSX_STATEIDOTHER))
2967 			break;
2968 	    }
2969 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2970 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2971 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2972 		NFSUNLOCKSTATE();
2973 		printf("Nfsd openctrl unexpected expiry\n");
2974 		free(new_open, M_NFSDSTATE);
2975 		free(new_deleg, M_NFSDSTATE);
2976 		if (haslock) {
2977 			NFSLOCKV4ROOTMUTEX();
2978 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2979 			NFSUNLOCKV4ROOTMUTEX();
2980 		}
2981 		error = NFSERR_EXPIRED;
2982 		goto out;
2983 	    }
2984 
2985 	    /*
2986 	     * Don't issue a Delegation, since one already exists and
2987 	     * delay delegation timeout, as required.
2988 	     */
2989 	    delegate = 0;
2990 	    nfsrv_delaydelegtimeout(stp);
2991 	}
2992 
2993 	/*
2994 	 * Check for access/deny bit conflicts. I also check for the
2995 	 * same owner, since the client might not have bothered to check.
2996 	 * Also, note an open for the same file and owner, if found,
2997 	 * which is all we do here for Delegate_Cur, since conflict
2998 	 * checking is already done.
2999 	 */
3000 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
3001 		if (ownerstp && stp->ls_openowner == ownerstp)
3002 			openstp = stp;
3003 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
3004 		    /*
3005 		     * If another client has the file open, the only
3006 		     * delegation that can be issued is a Read delegation
3007 		     * and only if it is a Read open with Deny none.
3008 		     */
3009 		    if (clp != stp->ls_clp) {
3010 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
3011 			    NFSLCK_READACCESS)
3012 			    writedeleg = 0;
3013 			else
3014 			    delegate = 0;
3015 		    }
3016 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
3017 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
3018 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
3019 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
3020 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
3021 			if (ret == 1) {
3022 				/*
3023 				 * nfsrv_clientconflict() unlocks state
3024 				 * when it returns non-zero.
3025 				 */
3026 				free(new_open, M_NFSDSTATE);
3027 				free(new_deleg, M_NFSDSTATE);
3028 				openstp = NULL;
3029 				goto tryagain;
3030 			}
3031 			if (ret == 2)
3032 				error = NFSERR_PERM;
3033 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
3034 				error = NFSERR_RECLAIMCONFLICT;
3035 			else
3036 				error = NFSERR_SHAREDENIED;
3037 			if (ret == 0)
3038 				NFSUNLOCKSTATE();
3039 			if (haslock) {
3040 				NFSLOCKV4ROOTMUTEX();
3041 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
3042 				NFSUNLOCKV4ROOTMUTEX();
3043 			}
3044 			free(new_open, M_NFSDSTATE);
3045 			free(new_deleg, M_NFSDSTATE);
3046 			printf("nfsd openctrl unexpected client cnfl\n");
3047 			goto out;
3048 		    }
3049 		}
3050 	}
3051 
3052 	/*
3053 	 * Check for a conflicting delegation. If one is found, call
3054 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
3055 	 * been set yet, it will get the lock. Otherwise, it will recall
3056 	 * the delegation. Then, we try try again...
3057 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
3058 	 *  isn't a conflict.)
3059 	 * I currently believe the conflict algorithm to be:
3060 	 * For Open with Read Access and Deny None
3061 	 * - there is a conflict iff a different client has a write delegation
3062 	 * For Open with other Write Access or any Deny except None
3063 	 * - there is a conflict if a different client has any delegation
3064 	 * - there is a conflict if the same client has a read delegation
3065 	 *   (The current consensus is that this last case should be
3066 	 *    considered a conflict since the client with a read delegation
3067 	 *    could have done an Open with ReadAccess and WriteDeny
3068 	 *    locally and then not have checked for the WriteDeny.)
3069 	 *    The exception is a NFSv4.1/4.2 client that has requested
3070 	 *    an atomic upgrade to a write delegation.
3071 	 */
3072 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
3073 	    stp = LIST_FIRST(&lfp->lf_deleg);
3074 	    while (stp != LIST_END(&lfp->lf_deleg)) {
3075 		nstp = LIST_NEXT(stp, ls_file);
3076 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
3077 			writedeleg = 0;
3078 		else if (stp->ls_clp != clp ||
3079 		    (stp->ls_flags & NFSLCK_DELEGWRITE) != 0 ||
3080 		    (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)
3081 			delegate = 0;
3082 		if ((readonly && stp->ls_clp != clp &&
3083 		     (stp->ls_flags & NFSLCK_DELEGWRITE) != 0) ||
3084 		    (!readonly && (stp->ls_clp != clp ||
3085 		     ((stp->ls_flags & NFSLCK_DELEGREAD) != 0 &&
3086 		      (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)))) {
3087 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3088 			delegate = 2;
3089 		    } else {
3090 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
3091 			if (ret) {
3092 			    /*
3093 			     * nfsrv_delegconflict() unlocks state
3094 			     * when it returns non-zero.
3095 			     */
3096 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
3097 			    free(new_open, M_NFSDSTATE);
3098 			    free(new_deleg, M_NFSDSTATE);
3099 			    if (ret == -1) {
3100 				openstp = NULL;
3101 				goto tryagain;
3102 			    }
3103 			    error = ret;
3104 			    goto out;
3105 			}
3106 		    }
3107 		}
3108 		stp = nstp;
3109 	    }
3110 	}
3111 
3112 	/*
3113 	 * We only get here if there was no open that conflicted.
3114 	 * If an open for the owner exists, or in the access/deny bits.
3115 	 * Otherwise it is a new open. If the open_owner hasn't been
3116 	 * confirmed, replace the open with the new one needing confirmation,
3117 	 * otherwise add the open.
3118 	 */
3119 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
3120 	    /*
3121 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
3122 	     * a match. If found, just move the old delegation to the current
3123 	     * delegation list and issue open. If not found, return
3124 	     * NFSERR_EXPIRED.
3125 	     */
3126 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
3127 		if (stp->ls_lfp == lfp) {
3128 		    /* Found it */
3129 		    if (stp->ls_clp != clp)
3130 			panic("olddeleg clp");
3131 		    LIST_REMOVE(stp, ls_list);
3132 		    LIST_REMOVE(stp, ls_hash);
3133 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
3134 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
3135 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
3136 			clp->lc_clientid.lval[0];
3137 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
3138 			clp->lc_clientid.lval[1];
3139 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
3140 			nfsrv_nextstateindex(clp);
3141 		    stp->ls_compref = nd->nd_compref;
3142 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
3143 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3144 			stp->ls_stateid), stp, ls_hash);
3145 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3146 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3147 		    else
3148 			*rflagsp |= NFSV4OPEN_READDELEGATE;
3149 		    clp->lc_delegtime = NFSD_MONOSEC +
3150 			nfsrv_lease + NFSRV_LEASEDELTA;
3151 
3152 		    /*
3153 		     * Now, do the associated open.
3154 		     */
3155 		    new_open->ls_stateid.seqid = 1;
3156 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3157 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3158 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3159 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
3160 			NFSLCK_OPEN;
3161 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3162 			new_open->ls_flags |= (NFSLCK_READACCESS |
3163 			    NFSLCK_WRITEACCESS);
3164 		    else
3165 			new_open->ls_flags |= NFSLCK_READACCESS;
3166 		    new_open->ls_uid = new_stp->ls_uid;
3167 		    new_open->ls_lfp = lfp;
3168 		    new_open->ls_clp = clp;
3169 		    LIST_INIT(&new_open->ls_open);
3170 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3171 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3172 			new_open, ls_hash);
3173 		    /*
3174 		     * and handle the open owner
3175 		     */
3176 		    if (ownerstp) {
3177 			new_open->ls_openowner = ownerstp;
3178 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3179 		    } else {
3180 			new_open->ls_openowner = new_stp;
3181 			new_stp->ls_flags = 0;
3182 			nfsrvd_refcache(new_stp->ls_op);
3183 			new_stp->ls_noopens = 0;
3184 			LIST_INIT(&new_stp->ls_open);
3185 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3186 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3187 			*new_stpp = NULL;
3188 			VNET(nfsstatsv1_p)->srvopenowners++;
3189 			nfsrv_openpluslock++;
3190 		    }
3191 		    openstp = new_open;
3192 		    new_open = NULL;
3193 		    VNET(nfsstatsv1_p)->srvopens++;
3194 		    nfsrv_openpluslock++;
3195 		    break;
3196 		}
3197 	    }
3198 	    if (stp == LIST_END(&clp->lc_olddeleg))
3199 		error = NFSERR_EXPIRED;
3200 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3201 	    /*
3202 	     * Scan to see that no delegation for this client and file
3203 	     * doesn't already exist.
3204 	     * There also shouldn't yet be an Open for this file and
3205 	     * openowner.
3206 	     */
3207 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3208 		if (stp->ls_clp == clp)
3209 		    break;
3210 	    }
3211 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3212 		/*
3213 		 * This is the Claim_Previous case with a delegation
3214 		 * type != Delegate_None.
3215 		 */
3216 		/*
3217 		 * First, add the delegation. (Although we must issue the
3218 		 * delegation, we can also ask for an immediate return.)
3219 		 */
3220 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3221 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3222 		    clp->lc_clientid.lval[0];
3223 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3224 		    clp->lc_clientid.lval[1];
3225 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3226 		    nfsrv_nextstateindex(clp);
3227 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3228 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3229 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3230 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3231 		    nfsrv_writedelegcnt++;
3232 		} else {
3233 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3234 			NFSLCK_READACCESS);
3235 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3236 		}
3237 		new_deleg->ls_uid = new_stp->ls_uid;
3238 		new_deleg->ls_lfp = lfp;
3239 		new_deleg->ls_clp = clp;
3240 		new_deleg->ls_filerev = filerev;
3241 		new_deleg->ls_compref = nd->nd_compref;
3242 		new_deleg->ls_lastrecall = 0;
3243 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3244 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3245 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3246 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3247 		new_deleg = NULL;
3248 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3249 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3250 		     LCL_CALLBACKSON ||
3251 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3252 		    !NFSVNO_DELEGOK(vp))
3253 		    *rflagsp |= NFSV4OPEN_RECALL;
3254 		VNET(nfsstatsv1_p)->srvdelegates++;
3255 		nfsrv_openpluslock++;
3256 		nfsrv_delegatecnt++;
3257 
3258 		/*
3259 		 * Now, do the associated open.
3260 		 */
3261 		new_open->ls_stateid.seqid = 1;
3262 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3263 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3264 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3265 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3266 		    NFSLCK_OPEN;
3267 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3268 			new_open->ls_flags |= (NFSLCK_READACCESS |
3269 			    NFSLCK_WRITEACCESS);
3270 		else
3271 			new_open->ls_flags |= NFSLCK_READACCESS;
3272 		new_open->ls_uid = new_stp->ls_uid;
3273 		new_open->ls_lfp = lfp;
3274 		new_open->ls_clp = clp;
3275 		LIST_INIT(&new_open->ls_open);
3276 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3277 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3278 		   new_open, ls_hash);
3279 		/*
3280 		 * and handle the open owner
3281 		 */
3282 		if (ownerstp) {
3283 		    new_open->ls_openowner = ownerstp;
3284 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3285 		} else {
3286 		    new_open->ls_openowner = new_stp;
3287 		    new_stp->ls_flags = 0;
3288 		    nfsrvd_refcache(new_stp->ls_op);
3289 		    new_stp->ls_noopens = 0;
3290 		    LIST_INIT(&new_stp->ls_open);
3291 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3292 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3293 		    *new_stpp = NULL;
3294 		    VNET(nfsstatsv1_p)->srvopenowners++;
3295 		    nfsrv_openpluslock++;
3296 		}
3297 		openstp = new_open;
3298 		new_open = NULL;
3299 		VNET(nfsstatsv1_p)->srvopens++;
3300 		nfsrv_openpluslock++;
3301 	    } else {
3302 		error = NFSERR_RECLAIMCONFLICT;
3303 	    }
3304 	} else if (ownerstp) {
3305 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3306 		    /* Replace the open */
3307 		    if (ownerstp->ls_op)
3308 			nfsrvd_derefcache(ownerstp->ls_op);
3309 		    ownerstp->ls_op = new_stp->ls_op;
3310 		    nfsrvd_refcache(ownerstp->ls_op);
3311 		    ownerstp->ls_seq = new_stp->ls_seq;
3312 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3313 		    stp = LIST_FIRST(&ownerstp->ls_open);
3314 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3315 			NFSLCK_OPEN;
3316 		    stp->ls_stateid.seqid = 1;
3317 		    stp->ls_uid = new_stp->ls_uid;
3318 		    if (lfp != stp->ls_lfp) {
3319 			LIST_REMOVE(stp, ls_file);
3320 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3321 			stp->ls_lfp = lfp;
3322 		    }
3323 		    openstp = stp;
3324 		} else if (openstp) {
3325 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3326 		    openstp->ls_stateid.seqid++;
3327 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3328 			openstp->ls_stateid.seqid == 0)
3329 			openstp->ls_stateid.seqid = 1;
3330 
3331 		    /*
3332 		     * This is where we can choose to issue a delegation.
3333 		     */
3334 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3335 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3336 			new_stp, lfp, rflagsp, delegstateidp);
3337 		} else {
3338 		    new_open->ls_stateid.seqid = 1;
3339 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3340 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3341 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3342 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3343 			NFSLCK_OPEN;
3344 		    new_open->ls_uid = new_stp->ls_uid;
3345 		    new_open->ls_openowner = ownerstp;
3346 		    new_open->ls_lfp = lfp;
3347 		    new_open->ls_clp = clp;
3348 		    LIST_INIT(&new_open->ls_open);
3349 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3350 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3351 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3352 			new_open, ls_hash);
3353 		    openstp = new_open;
3354 		    new_open = NULL;
3355 		    VNET(nfsstatsv1_p)->srvopens++;
3356 		    nfsrv_openpluslock++;
3357 
3358 		    /*
3359 		     * This is where we can choose to issue a delegation.
3360 		     */
3361 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3362 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3363 			new_stp, lfp, rflagsp, delegstateidp);
3364 		}
3365 	} else {
3366 		/*
3367 		 * New owner case. Start the open_owner sequence with a
3368 		 * Needs confirmation (unless a reclaim) and hang the
3369 		 * new open off it.
3370 		 */
3371 		new_open->ls_stateid.seqid = 1;
3372 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3373 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3374 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3375 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3376 		    NFSLCK_OPEN;
3377 		new_open->ls_uid = new_stp->ls_uid;
3378 		LIST_INIT(&new_open->ls_open);
3379 		new_open->ls_openowner = new_stp;
3380 		new_open->ls_lfp = lfp;
3381 		new_open->ls_clp = clp;
3382 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3383 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3384 			new_stp->ls_flags = 0;
3385 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3386 		    /*
3387 		     * This is where we can choose to issue a delegation.
3388 		     */
3389 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3390 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3391 			new_stp, lfp, rflagsp, delegstateidp);
3392 		    /* NFSv4.1 never needs confirmation. */
3393 		    new_stp->ls_flags = 0;
3394 
3395 		    /*
3396 		     * Since NFSv4.1 never does an OpenConfirm, the first
3397 		     * open state will be acquired here.
3398 		     */
3399 		    if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3400 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3401 			len = clp->lc_idlen;
3402 			NFSBCOPY(clp->lc_id, clidp, len);
3403 			gotstate = 1;
3404 		    }
3405 		} else {
3406 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3407 		    new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3408 		}
3409 		nfsrvd_refcache(new_stp->ls_op);
3410 		new_stp->ls_noopens = 0;
3411 		LIST_INIT(&new_stp->ls_open);
3412 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3413 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3414 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3415 		    new_open, ls_hash);
3416 		openstp = new_open;
3417 		new_open = NULL;
3418 		*new_stpp = NULL;
3419 		VNET(nfsstatsv1_p)->srvopens++;
3420 		nfsrv_openpluslock++;
3421 		VNET(nfsstatsv1_p)->srvopenowners++;
3422 		nfsrv_openpluslock++;
3423 	}
3424 	if (!error) {
3425 		stateidp->seqid = openstp->ls_stateid.seqid;
3426 		stateidp->other[0] = openstp->ls_stateid.other[0];
3427 		stateidp->other[1] = openstp->ls_stateid.other[1];
3428 		stateidp->other[2] = openstp->ls_stateid.other[2];
3429 	}
3430 	NFSUNLOCKSTATE();
3431 	if (haslock) {
3432 		NFSLOCKV4ROOTMUTEX();
3433 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3434 		NFSUNLOCKV4ROOTMUTEX();
3435 	}
3436 	if (new_open)
3437 		free(new_open, M_NFSDSTATE);
3438 	if (new_deleg)
3439 		free(new_deleg, M_NFSDSTATE);
3440 
3441 	/*
3442 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3443 	 * to the stable storage file.
3444 	 */
3445 	if (gotstate != 0) {
3446 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3447 		nfsrv_backupstable();
3448 	}
3449 
3450 out:
3451 	free(clidp, M_TEMP);
3452 	NFSEXITCODE2(error, nd);
3453 	return (error);
3454 }
3455 
3456 /*
3457  * Open update. Does the confirm, downgrade and close.
3458  */
3459 int
nfsrv_openupdate(vnode_t vp,struct nfsstate * new_stp,nfsquad_t clientid,nfsv4stateid_t * stateidp,struct nfsrv_descript * nd,NFSPROC_T * p,int * retwriteaccessp)3460 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3461     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3462     int *retwriteaccessp)
3463 {
3464 	struct nfsstate *stp;
3465 	struct nfsclient *clp;
3466 	u_int32_t bits;
3467 	int error = 0, gotstate = 0, len = 0;
3468 	u_char *clidp = NULL;
3469 
3470 	/*
3471 	 * Check for restart conditions (client and server).
3472 	 */
3473 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3474 	    &new_stp->ls_stateid, 0);
3475 	if (error)
3476 		goto out;
3477 
3478 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3479 	NFSLOCKSTATE();
3480 	/*
3481 	 * Get the open structure via clientid and stateid.
3482 	 */
3483 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3484 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3485 	if (!error)
3486 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3487 		    new_stp->ls_flags, &stp);
3488 
3489 	/*
3490 	 * Sanity check the open.
3491 	 */
3492 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3493 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3494 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3495 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3496 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3497 		error = NFSERR_BADSTATEID;
3498 
3499 	if (!error)
3500 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3501 		    stp->ls_openowner, new_stp->ls_op);
3502 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3503 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3504 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3505 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3506 	      new_stp->ls_stateid.seqid != 0)))
3507 		error = NFSERR_OLDSTATEID;
3508 	if (!error && vp->v_type != VREG) {
3509 		if (vp->v_type == VDIR)
3510 			error = NFSERR_ISDIR;
3511 		else
3512 			error = NFSERR_INVAL;
3513 	}
3514 
3515 	if (error) {
3516 		/*
3517 		 * If a client tries to confirm an Open with a bad
3518 		 * seqid# and there are no byte range locks or other Opens
3519 		 * on the openowner, just throw it away, so the next use of the
3520 		 * openowner will start a fresh seq#.
3521 		 */
3522 		if (error == NFSERR_BADSEQID &&
3523 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3524 		    nfsrv_nootherstate(stp))
3525 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3526 		NFSUNLOCKSTATE();
3527 		goto out;
3528 	}
3529 
3530 	/*
3531 	 * Set the return stateid.
3532 	 */
3533 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3534 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3535 		stateidp->seqid = 1;
3536 	stateidp->other[0] = stp->ls_stateid.other[0];
3537 	stateidp->other[1] = stp->ls_stateid.other[1];
3538 	stateidp->other[2] = stp->ls_stateid.other[2];
3539 	/*
3540 	 * Now, handle the three cases.
3541 	 */
3542 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3543 		/*
3544 		 * If the open doesn't need confirmation, it seems to me that
3545 		 * there is a client error, but I'll just log it and keep going?
3546 		 */
3547 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3548 			printf("Nfsv4d: stray open confirm\n");
3549 		stp->ls_openowner->ls_flags = 0;
3550 		stp->ls_stateid.seqid++;
3551 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3552 		    stp->ls_stateid.seqid == 0)
3553 			stp->ls_stateid.seqid = 1;
3554 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3555 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3556 			len = clp->lc_idlen;
3557 			NFSBCOPY(clp->lc_id, clidp, len);
3558 			gotstate = 1;
3559 		}
3560 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3561 		if (retwriteaccessp != NULL) {
3562 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3563 				*retwriteaccessp = 1;
3564 			else
3565 				*retwriteaccessp = 0;
3566 		}
3567 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3568 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3569 			nfsrv_freeopen(stp, vp, 1, p);
3570 		} else {
3571 			nfsrv_freeopen(stp, NULL, 0, p);
3572 		}
3573 	} else {
3574 		/*
3575 		 * Update the share bits, making sure that the new set are a
3576 		 * subset of the old ones.
3577 		 */
3578 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3579 		if (~(stp->ls_flags) & bits) {
3580 			NFSUNLOCKSTATE();
3581 			error = NFSERR_INVAL;
3582 			goto out;
3583 		}
3584 		stp->ls_flags = (bits | NFSLCK_OPEN);
3585 		stp->ls_stateid.seqid++;
3586 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3587 		    stp->ls_stateid.seqid == 0)
3588 			stp->ls_stateid.seqid = 1;
3589 	}
3590 	NFSUNLOCKSTATE();
3591 
3592 	/*
3593 	 * If the client just confirmed its first open, write a timestamp
3594 	 * to the stable storage file.
3595 	 */
3596 	if (gotstate != 0) {
3597 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3598 		nfsrv_backupstable();
3599 	}
3600 
3601 out:
3602 	free(clidp, M_TEMP);
3603 	NFSEXITCODE2(error, nd);
3604 	return (error);
3605 }
3606 
3607 /*
3608  * Delegation update. Does the purge and return.
3609  */
3610 int
nfsrv_delegupdate(struct nfsrv_descript * nd,nfsquad_t clientid,nfsv4stateid_t * stateidp,vnode_t vp,int op,struct ucred * cred,NFSPROC_T * p,int * retwriteaccessp)3611 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3612     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3613     NFSPROC_T *p, int *retwriteaccessp)
3614 {
3615 	struct nfsstate *stp;
3616 	struct nfsclient *clp;
3617 	int error = 0;
3618 	fhandle_t fh;
3619 
3620 	/*
3621 	 * Do a sanity check against the file handle for DelegReturn.
3622 	 */
3623 	if (vp) {
3624 		error = nfsvno_getfh(vp, &fh, p);
3625 		if (error)
3626 			goto out;
3627 	}
3628 	/*
3629 	 * Check for restart conditions (client and server).
3630 	 */
3631 	if (op == NFSV4OP_DELEGRETURN)
3632 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3633 			stateidp, 0);
3634 	else
3635 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3636 			stateidp, 0);
3637 
3638 	NFSLOCKSTATE();
3639 	/*
3640 	 * Get the open structure via clientid and stateid.
3641 	 */
3642 	if (!error)
3643 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3644 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3645 	if (error) {
3646 		if (error == NFSERR_CBPATHDOWN)
3647 			error = 0;
3648 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3649 			error = NFSERR_STALESTATEID;
3650 	}
3651 	if (!error && op == NFSV4OP_DELEGRETURN) {
3652 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3653 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3654 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3655 		error = NFSERR_OLDSTATEID;
3656 	}
3657 	/*
3658 	 * NFSERR_EXPIRED means that the state has gone away,
3659 	 * so Delegations have been purged. Just return ok.
3660 	 */
3661 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3662 		NFSUNLOCKSTATE();
3663 		error = 0;
3664 		goto out;
3665 	}
3666 	if (error) {
3667 		NFSUNLOCKSTATE();
3668 		goto out;
3669 	}
3670 
3671 	if (op == NFSV4OP_DELEGRETURN) {
3672 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3673 		    sizeof (fhandle_t))) {
3674 			NFSUNLOCKSTATE();
3675 			error = NFSERR_BADSTATEID;
3676 			goto out;
3677 		}
3678 		if (retwriteaccessp != NULL) {
3679 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3680 				*retwriteaccessp = 1;
3681 			else
3682 				*retwriteaccessp = 0;
3683 		}
3684 		nfsrv_freedeleg(stp);
3685 	} else {
3686 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3687 	}
3688 	NFSUNLOCKSTATE();
3689 	error = 0;
3690 
3691 out:
3692 	NFSEXITCODE(error);
3693 	return (error);
3694 }
3695 
3696 /*
3697  * Release lock owner.
3698  */
3699 int
nfsrv_releaselckown(struct nfsstate * new_stp,nfsquad_t clientid,NFSPROC_T * p)3700 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3701     NFSPROC_T *p)
3702 {
3703 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3704 	struct nfsclient *clp;
3705 	int error = 0;
3706 
3707 	/*
3708 	 * Check for restart conditions (client and server).
3709 	 */
3710 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3711 	    &new_stp->ls_stateid, 0);
3712 	if (error)
3713 		goto out;
3714 
3715 	NFSLOCKSTATE();
3716 	/*
3717 	 * Get the lock owner by name.
3718 	 */
3719 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3720 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3721 	if (error) {
3722 		NFSUNLOCKSTATE();
3723 		goto out;
3724 	}
3725 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3726 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3727 		stp = LIST_FIRST(&openstp->ls_open);
3728 		while (stp != LIST_END(&openstp->ls_open)) {
3729 		    nstp = LIST_NEXT(stp, ls_list);
3730 		    /*
3731 		     * If the owner matches, check for locks and
3732 		     * then free or return an error.
3733 		     */
3734 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3735 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3736 			 stp->ls_ownerlen)){
3737 			if (LIST_EMPTY(&stp->ls_lock)) {
3738 			    nfsrv_freelockowner(stp, NULL, 0, p);
3739 			} else {
3740 			    NFSUNLOCKSTATE();
3741 			    error = NFSERR_LOCKSHELD;
3742 			    goto out;
3743 			}
3744 		    }
3745 		    stp = nstp;
3746 		}
3747 	    }
3748 	}
3749 	NFSUNLOCKSTATE();
3750 
3751 out:
3752 	NFSEXITCODE(error);
3753 	return (error);
3754 }
3755 
3756 /*
3757  * Get the file handle for a lock structure.
3758  */
3759 static int
nfsrv_getlockfh(vnode_t vp,u_short flags,struct nfslockfile * new_lfp,fhandle_t * nfhp,NFSPROC_T * p)3760 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3761     fhandle_t *nfhp, NFSPROC_T *p)
3762 {
3763 	fhandle_t *fhp = NULL;
3764 	int error;
3765 
3766 	/*
3767 	 * For lock, use the new nfslock structure, otherwise just
3768 	 * a fhandle_t on the stack.
3769 	 */
3770 	if (flags & NFSLCK_OPEN) {
3771 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3772 		fhp = &new_lfp->lf_fh;
3773 	} else if (nfhp) {
3774 		fhp = nfhp;
3775 	} else {
3776 		panic("nfsrv_getlockfh");
3777 	}
3778 	error = nfsvno_getfh(vp, fhp, p);
3779 	NFSEXITCODE(error);
3780 	return (error);
3781 }
3782 
3783 /*
3784  * Get an nfs lock structure. Allocate one, as required, and return a
3785  * pointer to it.
3786  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3787  */
3788 static int
nfsrv_getlockfile(u_short flags,struct nfslockfile ** new_lfpp,struct nfslockfile ** lfpp,fhandle_t * nfhp,int lockit)3789 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3790     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3791 {
3792 	struct nfslockfile *lfp;
3793 	fhandle_t *fhp = NULL, *tfhp;
3794 	struct nfslockhashhead *hp;
3795 	struct nfslockfile *new_lfp = NULL;
3796 
3797 	/*
3798 	 * For lock, use the new nfslock structure, otherwise just
3799 	 * a fhandle_t on the stack.
3800 	 */
3801 	if (flags & NFSLCK_OPEN) {
3802 		new_lfp = *new_lfpp;
3803 		fhp = &new_lfp->lf_fh;
3804 	} else if (nfhp) {
3805 		fhp = nfhp;
3806 	} else {
3807 		panic("nfsrv_getlockfile");
3808 	}
3809 
3810 	hp = NFSLOCKHASH(fhp);
3811 	LIST_FOREACH(lfp, hp, lf_hash) {
3812 		tfhp = &lfp->lf_fh;
3813 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3814 			if (lockit)
3815 				nfsrv_locklf(lfp);
3816 			*lfpp = lfp;
3817 			return (0);
3818 		}
3819 	}
3820 	if (!(flags & NFSLCK_OPEN))
3821 		return (-1);
3822 
3823 	/*
3824 	 * No match, so chain the new one into the list.
3825 	 */
3826 	LIST_INIT(&new_lfp->lf_open);
3827 	LIST_INIT(&new_lfp->lf_lock);
3828 	LIST_INIT(&new_lfp->lf_deleg);
3829 	LIST_INIT(&new_lfp->lf_locallock);
3830 	LIST_INIT(&new_lfp->lf_rollback);
3831 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3832 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3833 	new_lfp->lf_usecount = 0;
3834 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3835 	*lfpp = new_lfp;
3836 	*new_lfpp = NULL;
3837 	return (0);
3838 }
3839 
3840 /*
3841  * This function adds a nfslock lock structure to the list for the associated
3842  * nfsstate and nfslockfile structures. It will be inserted after the
3843  * entry pointed at by insert_lop.
3844  */
3845 static void
nfsrv_insertlock(struct nfslock * new_lop,struct nfslock * insert_lop,struct nfsstate * stp,struct nfslockfile * lfp)3846 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3847     struct nfsstate *stp, struct nfslockfile *lfp)
3848 {
3849 	struct nfslock *lop, *nlop;
3850 
3851 	new_lop->lo_stp = stp;
3852 	new_lop->lo_lfp = lfp;
3853 
3854 	if (stp != NULL) {
3855 		/* Insert in increasing lo_first order */
3856 		lop = LIST_FIRST(&lfp->lf_lock);
3857 		if (lop == LIST_END(&lfp->lf_lock) ||
3858 		    new_lop->lo_first <= lop->lo_first) {
3859 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3860 		} else {
3861 			nlop = LIST_NEXT(lop, lo_lckfile);
3862 			while (nlop != LIST_END(&lfp->lf_lock) &&
3863 			       nlop->lo_first < new_lop->lo_first) {
3864 				lop = nlop;
3865 				nlop = LIST_NEXT(lop, lo_lckfile);
3866 			}
3867 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3868 		}
3869 	} else {
3870 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3871 	}
3872 
3873 	/*
3874 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3875 	 * an empty list.
3876 	 */
3877 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3878 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3879 	else if ((struct nfsstate *)insert_lop == stp)
3880 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3881 	else
3882 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3883 	if (stp != NULL) {
3884 		VNET(nfsstatsv1_p)->srvlocks++;
3885 		nfsrv_openpluslock++;
3886 	}
3887 }
3888 
3889 /*
3890  * This function updates the locking for a lock owner and given file. It
3891  * maintains a list of lock ranges ordered on increasing file offset that
3892  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3893  * It always adds new_lop to the list and sometimes uses the one pointed
3894  * at by other_lopp.
3895  */
3896 static void
nfsrv_updatelock(struct nfsstate * stp,struct nfslock ** new_lopp,struct nfslock ** other_lopp,struct nfslockfile * lfp)3897 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3898     struct nfslock **other_lopp, struct nfslockfile *lfp)
3899 {
3900 	struct nfslock *new_lop = *new_lopp;
3901 	struct nfslock *lop, *tlop, *ilop;
3902 	struct nfslock *other_lop = *other_lopp;
3903 	int unlock = 0, myfile = 0;
3904 	u_int64_t tmp;
3905 
3906 	/*
3907 	 * Work down the list until the lock is merged.
3908 	 */
3909 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3910 		unlock = 1;
3911 	if (stp != NULL) {
3912 		ilop = (struct nfslock *)stp;
3913 		lop = LIST_FIRST(&stp->ls_lock);
3914 	} else {
3915 		ilop = (struct nfslock *)lfp;
3916 		lop = LIST_FIRST(&lfp->lf_locallock);
3917 	}
3918 	while (lop != NULL) {
3919 	    /*
3920 	     * Only check locks for this file that aren't before the start of
3921 	     * new lock's range.
3922 	     */
3923 	    if (lop->lo_lfp == lfp) {
3924 	      myfile = 1;
3925 	      if (lop->lo_end >= new_lop->lo_first) {
3926 		if (new_lop->lo_end < lop->lo_first) {
3927 			/*
3928 			 * If the new lock ends before the start of the
3929 			 * current lock's range, no merge, just insert
3930 			 * the new lock.
3931 			 */
3932 			break;
3933 		}
3934 		if (new_lop->lo_flags == lop->lo_flags ||
3935 		    (new_lop->lo_first <= lop->lo_first &&
3936 		     new_lop->lo_end >= lop->lo_end)) {
3937 			/*
3938 			 * This lock can be absorbed by the new lock/unlock.
3939 			 * This happens when it covers the entire range
3940 			 * of the old lock or is contiguous
3941 			 * with the old lock and is of the same type or an
3942 			 * unlock.
3943 			 */
3944 			if (lop->lo_first < new_lop->lo_first)
3945 				new_lop->lo_first = lop->lo_first;
3946 			if (lop->lo_end > new_lop->lo_end)
3947 				new_lop->lo_end = lop->lo_end;
3948 			tlop = lop;
3949 			lop = LIST_NEXT(lop, lo_lckowner);
3950 			nfsrv_freenfslock(tlop);
3951 			continue;
3952 		}
3953 
3954 		/*
3955 		 * All these cases are for contiguous locks that are not the
3956 		 * same type, so they can't be merged.
3957 		 */
3958 		if (new_lop->lo_first <= lop->lo_first) {
3959 			/*
3960 			 * This case is where the new lock overlaps with the
3961 			 * first part of the old lock. Move the start of the
3962 			 * old lock to just past the end of the new lock. The
3963 			 * new lock will be inserted in front of the old, since
3964 			 * ilop hasn't been updated. (We are done now.)
3965 			 */
3966 			lop->lo_first = new_lop->lo_end;
3967 			break;
3968 		}
3969 		if (new_lop->lo_end >= lop->lo_end) {
3970 			/*
3971 			 * This case is where the new lock overlaps with the
3972 			 * end of the old lock's range. Move the old lock's
3973 			 * end to just before the new lock's first and insert
3974 			 * the new lock after the old lock.
3975 			 * Might not be done yet, since the new lock could
3976 			 * overlap further locks with higher ranges.
3977 			 */
3978 			lop->lo_end = new_lop->lo_first;
3979 			ilop = lop;
3980 			lop = LIST_NEXT(lop, lo_lckowner);
3981 			continue;
3982 		}
3983 		/*
3984 		 * The final case is where the new lock's range is in the
3985 		 * middle of the current lock's and splits the current lock
3986 		 * up. Use *other_lopp to handle the second part of the
3987 		 * split old lock range. (We are done now.)
3988 		 * For unlock, we use new_lop as other_lop and tmp, since
3989 		 * other_lop and new_lop are the same for this case.
3990 		 * We noted the unlock case above, so we don't need
3991 		 * new_lop->lo_flags any longer.
3992 		 */
3993 		tmp = new_lop->lo_first;
3994 		if (other_lop == NULL) {
3995 			if (!unlock)
3996 				panic("nfsd srv update unlock");
3997 			other_lop = new_lop;
3998 			*new_lopp = NULL;
3999 		}
4000 		other_lop->lo_first = new_lop->lo_end;
4001 		other_lop->lo_end = lop->lo_end;
4002 		other_lop->lo_flags = lop->lo_flags;
4003 		other_lop->lo_stp = stp;
4004 		other_lop->lo_lfp = lfp;
4005 		lop->lo_end = tmp;
4006 		nfsrv_insertlock(other_lop, lop, stp, lfp);
4007 		*other_lopp = NULL;
4008 		ilop = lop;
4009 		break;
4010 	      }
4011 	    }
4012 	    ilop = lop;
4013 	    lop = LIST_NEXT(lop, lo_lckowner);
4014 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
4015 		break;
4016 	}
4017 
4018 	/*
4019 	 * Insert the new lock in the list at the appropriate place.
4020 	 */
4021 	if (!unlock) {
4022 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
4023 		*new_lopp = NULL;
4024 	}
4025 }
4026 
4027 /*
4028  * This function handles sequencing of locks, etc.
4029  * It returns an error that indicates what the caller should do.
4030  */
4031 static int
nfsrv_checkseqid(struct nfsrv_descript * nd,u_int32_t seqid,struct nfsstate * stp,struct nfsrvcache * op)4032 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
4033     struct nfsstate *stp, struct nfsrvcache *op)
4034 {
4035 	int error = 0;
4036 
4037 	if ((nd->nd_flag & ND_NFSV41) != 0)
4038 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
4039 		goto out;
4040 	if (op != nd->nd_rp)
4041 		panic("nfsrvstate checkseqid");
4042 	if (!(op->rc_flag & RC_INPROG))
4043 		panic("nfsrvstate not inprog");
4044 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4045 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4046 		panic("nfsrvstate op refcnt");
4047 	}
4048 
4049 	/* If ND_ERELOOKUP is set, the seqid has already been handled. */
4050 	if ((nd->nd_flag & ND_ERELOOKUP) != 0)
4051 		goto out;
4052 
4053 	if ((stp->ls_seq + 1) == seqid) {
4054 		if (stp->ls_op)
4055 			nfsrvd_derefcache(stp->ls_op);
4056 		stp->ls_op = op;
4057 		nfsrvd_refcache(op);
4058 		stp->ls_seq = seqid;
4059 		goto out;
4060 	} else if (stp->ls_seq == seqid && stp->ls_op &&
4061 		op->rc_xid == stp->ls_op->rc_xid &&
4062 		op->rc_refcnt == 0 &&
4063 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
4064 		op->rc_cksum == stp->ls_op->rc_cksum) {
4065 		if (stp->ls_op->rc_flag & RC_INPROG) {
4066 			error = NFSERR_DONTREPLY;
4067 			goto out;
4068 		}
4069 		nd->nd_rp = stp->ls_op;
4070 		nd->nd_rp->rc_flag |= RC_INPROG;
4071 		nfsrvd_delcache(op);
4072 		error = NFSERR_REPLYFROMCACHE;
4073 		goto out;
4074 	}
4075 	error = NFSERR_BADSEQID;
4076 
4077 out:
4078 	NFSEXITCODE2(error, nd);
4079 	return (error);
4080 }
4081 
4082 /*
4083  * Just set lc_program to 0 to indicate no callbacks are possible.
4084  * Set the address to the client's transport address. This won't be used
4085  * for callbacks, but can be printed out by nfsstats for info.
4086  * Return error if the xdr can't be parsed, 0 otherwise.
4087  */
4088 int
nfsrv_getclientipaddr(struct nfsrv_descript * nd,struct nfsclient * clp)4089 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4090 {
4091 	uint32_t *tl;
4092 	int error = 0, i;
4093 #ifdef INET
4094 	struct sockaddr_in *rin, *sin;
4095 #endif
4096 #ifdef INET6
4097 	struct sockaddr_in6 *rin6, *sin6;
4098 #endif
4099 
4100 	clp->lc_req.nr_client = NULL;
4101 	clp->lc_req.nr_lock = 0;
4102 	NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED);
4103 	i = fxdr_unsigned(int, *tl);
4104 	if (i < 0) {
4105 		error = NFSERR_BADXDR;
4106 		goto nfsmout;
4107 	} else if (i > 0) {
4108 		error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4109 		if (error)
4110 			goto nfsmout;
4111 	}
4112 	NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED);
4113 	i = fxdr_unsigned(int, *tl);
4114 	if (i < 0) {
4115 		error = NFSERR_BADXDR;
4116 		goto nfsmout;
4117 	} else if (i > 0) {
4118 		error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4119 		if (error)
4120 			goto nfsmout;
4121 	}
4122 	switch (nd->nd_nam->sa_family) {
4123 #ifdef INET
4124 	case AF_INET:
4125 		sin = (struct sockaddr_in *)nd->nd_nam;
4126 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4127 		rin->sin_family = AF_INET;
4128 		rin->sin_len = sizeof(struct sockaddr_in);
4129 		rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4130 		rin->sin_port = 0x0;
4131 		break;
4132 #endif
4133 #ifdef INET6
4134 	case AF_INET6:
4135 		sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4136 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4137 		rin6->sin6_family = AF_INET6;
4138 		rin6->sin6_len = sizeof(struct sockaddr_in6);
4139 		rin6->sin6_addr = sin6->sin6_addr;
4140 		rin6->sin6_port = 0x0;
4141 		break;
4142 #endif
4143 	}
4144 	clp->lc_program = 0;
4145 nfsmout:
4146 	NFSEXITCODE2(error, nd);
4147 	return (error);
4148 }
4149 
4150 /*
4151  * This function checks for restart conditions.
4152  */
4153 static int
nfsrv_checkrestart(nfsquad_t clientid,u_int32_t flags,nfsv4stateid_t * stateidp,int specialid)4154 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4155     nfsv4stateid_t *stateidp, int specialid)
4156 {
4157 	int ret = 0;
4158 
4159 	/*
4160 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4161 	 * and DelegPurge have a clientid, the rest a stateid.
4162 	 */
4163 	if (flags &
4164 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4165 		if (clientid.lval[0] != VNET(nfsrvboottime)) {
4166 			ret = NFSERR_STALECLIENTID;
4167 			goto out;
4168 		}
4169 	} else if (stateidp->other[0] != VNET(nfsrvboottime) &&
4170 		specialid == 0) {
4171 		ret = NFSERR_STALESTATEID;
4172 		goto out;
4173 	}
4174 
4175 	/*
4176 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4177 	 * not use a lock/open owner seqid#, so the check can be done now.
4178 	 * (The others will be checked, as required, later.)
4179 	 */
4180 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4181 		goto out;
4182 
4183 	NFSLOCKSTATE();
4184 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4185 	NFSUNLOCKSTATE();
4186 
4187 out:
4188 	NFSEXITCODE(ret);
4189 	return (ret);
4190 }
4191 
4192 /*
4193  * Check for grace.
4194  */
4195 static int
nfsrv_checkgrace(struct nfsrv_descript * nd,struct nfsclient * clp,u_int32_t flags)4196 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4197     u_int32_t flags)
4198 {
4199 	int error = 0, notreclaimed;
4200 	struct nfsrv_stable *sp;
4201 
4202 	if ((VNET(nfsrv_stablefirst).nsf_flags & (NFSNSF_UPDATEDONE |
4203 	     NFSNSF_GRACEOVER)) == 0) {
4204 		/*
4205 		 * First, check to see if all of the clients have done a
4206 		 * ReclaimComplete.  If so, grace can end now.
4207 		 */
4208 		notreclaimed = 0;
4209 		if (!VNET(nfsd_disable_grace)) {
4210 			LIST_FOREACH(sp, &VNET(nfsrv_stablefirst).nsf_head,
4211 			    nst_list) {
4212 				if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4213 					notreclaimed = 1;
4214 					break;
4215 				}
4216 			}
4217 		}
4218 		if (notreclaimed == 0)
4219 			VNET(nfsrv_stablefirst).nsf_flags |=
4220 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4221 	}
4222 
4223 	if ((VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_GRACEOVER) != 0) {
4224 		if (flags & NFSLCK_RECLAIM) {
4225 			error = NFSERR_NOGRACE;
4226 			goto out;
4227 		}
4228 	} else {
4229 		if (!(flags & NFSLCK_RECLAIM)) {
4230 			error = NFSERR_GRACE;
4231 			goto out;
4232 		}
4233 		if (nd != NULL && clp != NULL &&
4234 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4235 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4236 			error = NFSERR_NOGRACE;
4237 			goto out;
4238 		}
4239 
4240 		/*
4241 		 * If grace is almost over and we are still getting Reclaims,
4242 		 * extend grace a bit.
4243 		 */
4244 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4245 		    VNET(nfsrv_stablefirst).nsf_eograce)
4246 			VNET(nfsrv_stablefirst).nsf_eograce =
4247 				NFSD_MONOSEC + NFSRV_LEASEDELTA;
4248 	}
4249 
4250 out:
4251 	NFSEXITCODE(error);
4252 	return (error);
4253 }
4254 
4255 /*
4256  * Do a server callback.
4257  * The "trunc" argument is slightly overloaded and refers to different
4258  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4259  */
4260 static int
nfsrv_docallback(struct nfsclient * clp,int procnum,nfsv4stateid_t * stateidp,int trunc,fhandle_t * fhp,struct nfsvattr * nap,nfsattrbit_t * attrbitp,int laytype,NFSPROC_T * p)4261 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4262     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4263     int laytype, NFSPROC_T *p)
4264 {
4265 	struct mbuf *m;
4266 	u_int32_t *tl;
4267 	struct nfsrv_descript *nd;
4268 	struct ucred *cred;
4269 	int error = 0, slotpos;
4270 	u_int32_t callback;
4271 	struct nfsdsession *sep = NULL;
4272 	uint64_t tval;
4273 	bool dotls;
4274 
4275 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4276 	cred = newnfs_getcred();
4277 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4278 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4279 		NFSUNLOCKSTATE();
4280 		panic("docallb");
4281 	}
4282 	clp->lc_cbref++;
4283 
4284 	/*
4285 	 * Fill the callback program# and version into the request
4286 	 * structure for newnfs_connect() to use.
4287 	 */
4288 	clp->lc_req.nr_prog = clp->lc_program;
4289 #ifdef notnow
4290 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4291 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4292 	else
4293 #endif
4294 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4295 
4296 	/*
4297 	 * First, fill in some of the fields of nd and cr.
4298 	 */
4299 	nd->nd_flag = ND_NFSV4;
4300 	if (clp->lc_flags & LCL_GSS)
4301 		nd->nd_flag |= ND_KERBV;
4302 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4303 		nd->nd_flag |= ND_NFSV41;
4304 	if ((clp->lc_flags & LCL_NFSV42) != 0)
4305 		nd->nd_flag |= ND_NFSV42;
4306 	nd->nd_repstat = 0;
4307 	cred->cr_uid = clp->lc_uid;
4308 	cred->cr_gid = clp->lc_gid;
4309 	callback = clp->lc_callback;
4310 	NFSUNLOCKSTATE();
4311 	cred->cr_ngroups = 1;
4312 
4313 	/*
4314 	 * Get the first mbuf for the request.
4315 	 */
4316 	MGET(m, M_WAITOK, MT_DATA);
4317 	m->m_len = 0;
4318 	nd->nd_mreq = nd->nd_mb = m;
4319 	nd->nd_bpos = mtod(m, caddr_t);
4320 
4321 	/*
4322 	 * and build the callback request.
4323 	 */
4324 	if (procnum == NFSV4OP_CBGETATTR) {
4325 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4326 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4327 		    "CB Getattr", &sep, &slotpos);
4328 		if (error != 0) {
4329 			m_freem(nd->nd_mreq);
4330 			goto errout;
4331 		}
4332 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4333 		(void)nfsrv_putattrbit(nd, attrbitp);
4334 	} else if (procnum == NFSV4OP_CBRECALL) {
4335 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4336 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4337 		    "CB Recall", &sep, &slotpos);
4338 		if (error != 0) {
4339 			m_freem(nd->nd_mreq);
4340 			goto errout;
4341 		}
4342 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4343 		*tl++ = txdr_unsigned(stateidp->seqid);
4344 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4345 		    NFSX_STATEIDOTHER);
4346 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4347 		if (trunc)
4348 			*tl = newnfs_true;
4349 		else
4350 			*tl = newnfs_false;
4351 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4352 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4353 		NFSD_DEBUG(4, "docallback layout recall\n");
4354 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4355 		error = nfsrv_cbcallargs(nd, clp, callback,
4356 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos);
4357 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4358 		if (error != 0) {
4359 			m_freem(nd->nd_mreq);
4360 			goto errout;
4361 		}
4362 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4363 		*tl++ = txdr_unsigned(laytype);
4364 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4365 		if (trunc)
4366 			*tl++ = newnfs_true;
4367 		else
4368 			*tl++ = newnfs_false;
4369 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4370 		(void)nfsm_fhtom(NULL, nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4371 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4372 		tval = 0;
4373 		txdr_hyper(tval, tl); tl += 2;
4374 		tval = UINT64_MAX;
4375 		txdr_hyper(tval, tl); tl += 2;
4376 		*tl++ = txdr_unsigned(stateidp->seqid);
4377 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4378 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4379 		NFSD_DEBUG(4, "aft args\n");
4380 	} else if (procnum == NFSV4PROC_CBNULL) {
4381 		nd->nd_procnum = NFSV4PROC_CBNULL;
4382 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4383 			error = nfsv4_getcbsession(clp, &sep);
4384 			if (error != 0) {
4385 				m_freem(nd->nd_mreq);
4386 				goto errout;
4387 			}
4388 		}
4389 	} else {
4390 		error = NFSERR_SERVERFAULT;
4391 		m_freem(nd->nd_mreq);
4392 		goto errout;
4393 	}
4394 
4395 	/*
4396 	 * Call newnfs_connect(), as required, and then newnfs_request().
4397 	 */
4398 	dotls = false;
4399 	if ((clp->lc_flags & LCL_TLSCB) != 0)
4400 		dotls = true;
4401 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4402 	if (clp->lc_req.nr_client == NULL) {
4403 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4404 			error = ECONNREFUSED;
4405 			if (procnum != NFSV4PROC_CBNULL)
4406 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4407 				    true);
4408 			nfsrv_freesession(NULL, sep, NULL, false, NULL);
4409 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4410 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4411 			    NULL, 1, dotls, &clp->lc_req.nr_client);
4412 		else
4413 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4414 			    NULL, 3, dotls, &clp->lc_req.nr_client);
4415 	}
4416 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4417 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4418 	if (!error) {
4419 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4420 			KASSERT(sep != NULL, ("sep NULL"));
4421 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4422 				error = newnfs_request(nd, NULL, clp,
4423 				    &clp->lc_req, NULL, NULL, cred,
4424 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4425 				    1, NULL, &sep->sess_cbsess);
4426 			else {
4427 				/*
4428 				 * This should probably never occur, but if a
4429 				 * client somehow does an RPC without a
4430 				 * SequenceID Op that causes a callback just
4431 				 * after the nfsd threads have been terminated
4432 				 * and restarted we could conceivably get here
4433 				 * without a backchannel xprt.
4434 				 */
4435 				printf("nfsrv_docallback: no xprt\n");
4436 				error = ECONNREFUSED;
4437 			}
4438 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4439 			if (error != 0 && procnum != NFSV4PROC_CBNULL) {
4440 				/*
4441 				 * It is likely that the callback was never
4442 				 * processed by the client and, as such,
4443 				 * the sequence# for the session slot needs
4444 				 * to be backed up by one to avoid a
4445 				 * NFSERR_SEQMISORDERED error reply.
4446 				 * For the unlikely case where the callback
4447 				 * was processed by the client, this will
4448 				 * make the next callback on the slot
4449 				 * appear to be a retry.
4450 				 * Since callbacks never specify that the
4451 				 * reply be cached, this "apparent retry"
4452 				 * should not be a problem.
4453 				 */
4454 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4455 				    true);
4456 			}
4457 			nfsrv_freesession(NULL, sep, NULL, false, NULL);
4458 		} else
4459 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4460 			    NULL, NULL, cred, clp->lc_program,
4461 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4462 	}
4463 errout:
4464 	NFSFREECRED(cred);
4465 
4466 	/*
4467 	 * If error is set here, the Callback path isn't working
4468 	 * properly, so twiddle the appropriate LCL_ flags.
4469 	 * (nd_repstat != 0 indicates the Callback path is working,
4470 	 *  but the callback failed on the client.)
4471 	 */
4472 	if (error) {
4473 		/*
4474 		 * Mark the callback pathway down, which disabled issuing
4475 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4476 		 */
4477 		NFSLOCKSTATE();
4478 		clp->lc_flags |= LCL_CBDOWN;
4479 		NFSUNLOCKSTATE();
4480 	} else {
4481 		/*
4482 		 * Callback worked. If the callback path was down, disable
4483 		 * callbacks, so no more delegations will be issued. (This
4484 		 * is done on the assumption that the callback pathway is
4485 		 * flakey.)
4486 		 */
4487 		NFSLOCKSTATE();
4488 		if (clp->lc_flags & LCL_CBDOWN)
4489 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4490 		NFSUNLOCKSTATE();
4491 		if (nd->nd_repstat) {
4492 			error = nd->nd_repstat;
4493 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4494 			    procnum, error);
4495 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4496 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4497 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4498 			    NULL, NULL, NULL, p, NULL);
4499 		m_freem(nd->nd_mrep);
4500 	}
4501 	NFSLOCKSTATE();
4502 	clp->lc_cbref--;
4503 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4504 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4505 		wakeup(clp);
4506 	}
4507 	NFSUNLOCKSTATE();
4508 
4509 	free(nd, M_TEMP);
4510 	NFSEXITCODE(error);
4511 	return (error);
4512 }
4513 
4514 /*
4515  * Set up the compound RPC for the callback.
4516  */
4517 static int
nfsrv_cbcallargs(struct nfsrv_descript * nd,struct nfsclient * clp,uint32_t callback,int op,const char * optag,struct nfsdsession ** sepp,int * slotposp)4518 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4519     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
4520     int *slotposp)
4521 {
4522 	uint32_t *tl;
4523 	int error, len;
4524 
4525 	len = strlen(optag);
4526 	(void)nfsm_strtom(nd, optag, len);
4527 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4528 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4529 		if ((nd->nd_flag & ND_NFSV42) != 0)
4530 			*tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4531 		else
4532 			*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4533 		*tl++ = txdr_unsigned(callback);
4534 		*tl++ = txdr_unsigned(2);
4535 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4536 		error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp);
4537 		if (error != 0)
4538 			return (error);
4539 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4540 		*tl = txdr_unsigned(op);
4541 	} else {
4542 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4543 		*tl++ = txdr_unsigned(callback);
4544 		*tl++ = txdr_unsigned(1);
4545 		*tl = txdr_unsigned(op);
4546 	}
4547 	return (0);
4548 }
4549 
4550 /*
4551  * Return the next index# for a clientid. Mostly just increment and return
4552  * the next one, but... if the 32bit unsigned does actually wrap around,
4553  * it should be rebooted.
4554  * At an average rate of one new client per second, it will wrap around in
4555  * approximately 136 years. (I think the server will have been shut
4556  * down or rebooted before then.)
4557  */
4558 static u_int32_t
nfsrv_nextclientindex(void)4559 nfsrv_nextclientindex(void)
4560 {
4561 	static u_int32_t client_index = 0;
4562 
4563 	client_index++;
4564 	if (client_index != 0)
4565 		return (client_index);
4566 
4567 	printf("%s: out of clientids\n", __func__);
4568 	return (client_index);
4569 }
4570 
4571 /*
4572  * Return the next index# for a stateid. Mostly just increment and return
4573  * the next one, but... if the 32bit unsigned does actually wrap around
4574  * (will a BSD server stay up that long?), find
4575  * new start and end values.
4576  */
4577 static u_int32_t
nfsrv_nextstateindex(struct nfsclient * clp)4578 nfsrv_nextstateindex(struct nfsclient *clp)
4579 {
4580 	struct nfsstate *stp;
4581 	int i;
4582 	u_int32_t canuse, min_index, max_index;
4583 
4584 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4585 		clp->lc_stateindex++;
4586 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4587 			return (clp->lc_stateindex);
4588 	}
4589 
4590 	/*
4591 	 * Yuck, we've hit the end.
4592 	 * Look for a new min and max.
4593 	 */
4594 	min_index = 0;
4595 	max_index = 0xffffffff;
4596 	for (i = 0; i < nfsrv_statehashsize; i++) {
4597 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4598 		if (stp->ls_stateid.other[2] > 0x80000000) {
4599 		    if (stp->ls_stateid.other[2] < max_index)
4600 			max_index = stp->ls_stateid.other[2];
4601 		} else {
4602 		    if (stp->ls_stateid.other[2] > min_index)
4603 			min_index = stp->ls_stateid.other[2];
4604 		}
4605 	    }
4606 	}
4607 
4608 	/*
4609 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4610 	 */
4611 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4612 	    canuse = 0;
4613 	    /*
4614 	     * Loop around until we find an unused entry. Return that
4615 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4616 	     * (This is one of those rare cases where a goto is the
4617 	     *  cleanest way to code the loop.)
4618 	     */
4619 tryagain:
4620 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4621 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4622 		    if (stp->ls_stateid.other[2] == canuse) {
4623 			canuse++;
4624 			goto tryagain;
4625 		    }
4626 		}
4627 	    }
4628 	    clp->lc_flags |= LCL_INDEXNOTOK;
4629 	    return (canuse);
4630 	}
4631 
4632 	/*
4633 	 * Ok to start again from min + 1.
4634 	 */
4635 	clp->lc_stateindex = min_index + 1;
4636 	clp->lc_statemaxindex = max_index;
4637 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4638 	return (clp->lc_stateindex);
4639 }
4640 
4641 /*
4642  * The following functions handle the stable storage file that deals with
4643  * the edge conditions described in RFC3530 Sec. 8.6.3.
4644  * The file is as follows:
4645  * - a single record at the beginning that has the lease time of the
4646  *   previous server instance (before the last reboot) and the nfsrvboottime
4647  *   values for the previous server boots.
4648  *   These previous boot times are used to ensure that the current
4649  *   nfsrvboottime does not, somehow, get set to a previous one.
4650  *   (This is important so that Stale ClientIDs and StateIDs can
4651  *    be recognized.)
4652  *   The number of previous nfsvrboottime values precedes the list.
4653  * - followed by some number of appended records with:
4654  *   - client id string
4655  *   - flag that indicates it is a record revoking state via lease
4656  *     expiration or similar
4657  *     OR has successfully acquired state.
4658  * These structures vary in length, with the client string at the end, up
4659  * to NFSV4_OPAQUELIMIT in size.
4660  *
4661  * At the end of the grace period, the file is truncated, the first
4662  * record is rewritten with updated information and any acquired state
4663  * records for successful reclaims of state are written.
4664  *
4665  * Subsequent records are appended when the first state is issued to
4666  * a client and when state is revoked for a client.
4667  *
4668  * When reading the file in, state issued records that come later in
4669  * the file override older ones, since the append log is in cronological order.
4670  * If, for some reason, the file can't be read, the grace period is
4671  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4672  */
4673 
4674 /*
4675  * Read in the stable storage file. Called by nfssvc() before the nfsd
4676  * processes start servicing requests.
4677  */
4678 void
nfsrv_setupstable(NFSPROC_T * p)4679 nfsrv_setupstable(NFSPROC_T *p)
4680 {
4681 	struct nfsrv_stablefirst *sf = &VNET(nfsrv_stablefirst);
4682 	struct nfsrv_stable *sp, *nsp;
4683 	struct nfst_rec *tsp;
4684 	int error, i, tryagain;
4685 	off_t off = 0;
4686 	ssize_t aresid, len;
4687 
4688 	/*
4689 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4690 	 * a reboot, so state has not been lost.
4691 	 */
4692 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4693 		return;
4694 	/*
4695 	 * Set Grace over just until the file reads successfully.
4696 	 */
4697 	VNET(nfsrvboottime) = time_second;
4698 	LIST_INIT(&sf->nsf_head);
4699 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4700 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4701 	if (sf->nsf_fp == NULL)
4702 		return;
4703 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4704 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4705 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4706 	if (error || aresid || sf->nsf_numboots == 0 ||
4707 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4708 		return;
4709 
4710 	/*
4711 	 * Now, read in the boottimes.
4712 	 */
4713 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4714 		sizeof(time_t), M_TEMP, M_WAITOK);
4715 	off = sizeof (struct nfsf_rec);
4716 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4717 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4718 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4719 	if (error || aresid) {
4720 		free(sf->nsf_bootvals, M_TEMP);
4721 		sf->nsf_bootvals = NULL;
4722 		return;
4723 	}
4724 
4725 	/*
4726 	 * Make sure this nfsrvboottime is different from all recorded
4727 	 * previous ones.
4728 	 */
4729 	do {
4730 		tryagain = 0;
4731 		for (i = 0; i < sf->nsf_numboots; i++) {
4732 			if (VNET(nfsrvboottime) == sf->nsf_bootvals[i]) {
4733 				VNET(nfsrvboottime)++;
4734 				tryagain = 1;
4735 				break;
4736 			}
4737 		}
4738 	} while (tryagain);
4739 
4740 	sf->nsf_flags |= NFSNSF_OK;
4741 	off += (sf->nsf_numboots * sizeof (time_t));
4742 
4743 	/*
4744 	 * Read through the file, building a list of records for grace
4745 	 * checking.
4746 	 * Each record is between sizeof (struct nfst_rec) and
4747 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4748 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4749 	 */
4750 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4751 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4752 	do {
4753 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4754 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4755 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4756 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4757 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4758 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4759 		/*
4760 		 * Yuck, the file has been corrupted, so just return
4761 		 * after clearing out any restart state, so the grace period
4762 		 * is over.
4763 		 */
4764 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4765 			LIST_REMOVE(sp, nst_list);
4766 			free(sp, M_TEMP);
4767 		}
4768 		free(tsp, M_TEMP);
4769 		sf->nsf_flags &= ~NFSNSF_OK;
4770 		free(sf->nsf_bootvals, M_TEMP);
4771 		sf->nsf_bootvals = NULL;
4772 		return;
4773 	    }
4774 	    if (len > 0) {
4775 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4776 		/*
4777 		 * Search the list for a matching client.
4778 		 */
4779 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4780 			if (tsp->len == sp->nst_len &&
4781 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4782 				break;
4783 		}
4784 		if (sp == LIST_END(&sf->nsf_head)) {
4785 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4786 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4787 				M_WAITOK);
4788 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4789 				sizeof (struct nfst_rec) + tsp->len - 1);
4790 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4791 		} else {
4792 			if (tsp->flag == NFSNST_REVOKE)
4793 				sp->nst_flag |= NFSNST_REVOKE;
4794 			else
4795 				/*
4796 				 * A subsequent timestamp indicates the client
4797 				 * did a setclientid/confirm and any previous
4798 				 * revoke is no longer relevant.
4799 				 */
4800 				sp->nst_flag &= ~NFSNST_REVOKE;
4801 		}
4802 	    }
4803 	} while (len > 0);
4804 	free(tsp, M_TEMP);
4805 	sf->nsf_flags = NFSNSF_OK;
4806 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4807 		NFSRV_LEASEDELTA;
4808 }
4809 
4810 /*
4811  * Update the stable storage file, now that the grace period is over.
4812  */
4813 void
nfsrv_updatestable(NFSPROC_T * p)4814 nfsrv_updatestable(NFSPROC_T *p)
4815 {
4816 	struct nfsrv_stablefirst *sf = &VNET(nfsrv_stablefirst);
4817 	struct nfsrv_stable *sp, *nsp;
4818 	int i;
4819 	struct nfsvattr nva;
4820 	vnode_t vp;
4821 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4822 	mount_t mp = NULL;
4823 #endif
4824 	int error;
4825 
4826 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4827 		return;
4828 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4829 	/*
4830 	 * Ok, we need to rewrite the stable storage file.
4831 	 * - truncate to 0 length
4832 	 * - write the new first structure
4833 	 * - loop through the data structures, writing out any that
4834 	 *   have timestamps older than the old boot
4835 	 */
4836 	if (sf->nsf_bootvals) {
4837 		sf->nsf_numboots++;
4838 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4839 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4840 	} else {
4841 		sf->nsf_numboots = 1;
4842 		sf->nsf_bootvals = (time_t *)malloc(sizeof(time_t),
4843 			M_TEMP, M_WAITOK);
4844 	}
4845 	sf->nsf_bootvals[0] = VNET(nfsrvboottime);
4846 	sf->nsf_lease = nfsrv_lease;
4847 	NFSVNO_ATTRINIT(&nva);
4848 	NFSVNO_SETATTRVAL(&nva, size, 0);
4849 	vp = NFSFPVNODE(sf->nsf_fp);
4850 	vn_start_write(vp, &mp, V_WAIT);
4851 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4852 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4853 		    NULL);
4854 		NFSVOPUNLOCK(vp);
4855 	} else
4856 		error = EPERM;
4857 	vn_finished_write(mp);
4858 	if (!error)
4859 	    error = NFSD_RDWR(UIO_WRITE, vp,
4860 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4861 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4862 	if (!error)
4863 	    error = NFSD_RDWR(UIO_WRITE, vp,
4864 		(caddr_t)sf->nsf_bootvals,
4865 		sf->nsf_numboots * sizeof (time_t),
4866 		(off_t)(sizeof (struct nfsf_rec)),
4867 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4868 	free(sf->nsf_bootvals, M_TEMP);
4869 	sf->nsf_bootvals = NULL;
4870 	if (error) {
4871 		sf->nsf_flags &= ~NFSNSF_OK;
4872 		printf("EEK! Can't write NfsV4 stable storage file\n");
4873 		return;
4874 	}
4875 	sf->nsf_flags |= NFSNSF_OK;
4876 
4877 	/*
4878 	 * Loop through the list and write out timestamp records for
4879 	 * any clients that successfully reclaimed state.
4880 	 */
4881 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4882 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4883 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4884 				NFSNST_NEWSTATE, p);
4885 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4886 		}
4887 		LIST_REMOVE(sp, nst_list);
4888 		free(sp, M_TEMP);
4889 	}
4890 	nfsrv_backupstable();
4891 }
4892 
4893 /*
4894  * Append a record to the stable storage file.
4895  */
4896 void
nfsrv_writestable(u_char * client,int len,int flag,NFSPROC_T * p)4897 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4898 {
4899 	struct nfsrv_stablefirst *sf = &VNET(nfsrv_stablefirst);
4900 	struct nfst_rec *sp;
4901 	int error;
4902 
4903 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4904 		return;
4905 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4906 		len - 1, M_TEMP, M_WAITOK);
4907 	sp->len = len;
4908 	NFSBCOPY(client, sp->client, len);
4909 	sp->flag = flag;
4910 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4911 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4912 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4913 	free(sp, M_TEMP);
4914 	if (error) {
4915 		sf->nsf_flags &= ~NFSNSF_OK;
4916 		printf("EEK! Can't write NfsV4 stable storage file\n");
4917 	}
4918 }
4919 
4920 /*
4921  * This function is called during the grace period to mark a client
4922  * that successfully reclaimed state.
4923  */
4924 static void
nfsrv_markstable(struct nfsclient * clp)4925 nfsrv_markstable(struct nfsclient *clp)
4926 {
4927 	struct nfsrv_stable *sp;
4928 
4929 	/*
4930 	 * First find the client structure.
4931 	 */
4932 	LIST_FOREACH(sp, &VNET(nfsrv_stablefirst).nsf_head, nst_list) {
4933 		if (sp->nst_len == clp->lc_idlen &&
4934 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4935 			break;
4936 	}
4937 	if (sp == LIST_END(&VNET(nfsrv_stablefirst).nsf_head))
4938 		return;
4939 
4940 	/*
4941 	 * Now, just mark it and set the nfsclient back pointer.
4942 	 */
4943 	sp->nst_flag |= NFSNST_GOTSTATE;
4944 	sp->nst_clp = clp;
4945 }
4946 
4947 /*
4948  * This function is called when a NFSv4.1 client does a ReclaimComplete.
4949  * Very similar to nfsrv_markstable(), except for the flag being set.
4950  */
4951 static void
nfsrv_markreclaim(struct nfsclient * clp)4952 nfsrv_markreclaim(struct nfsclient *clp)
4953 {
4954 	struct nfsrv_stable *sp;
4955 
4956 	/*
4957 	 * First find the client structure.
4958 	 */
4959 	LIST_FOREACH(sp, &VNET(nfsrv_stablefirst).nsf_head, nst_list) {
4960 		if (sp->nst_len == clp->lc_idlen &&
4961 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4962 			break;
4963 	}
4964 	if (sp == LIST_END(&VNET(nfsrv_stablefirst).nsf_head))
4965 		return;
4966 
4967 	/*
4968 	 * Now, just set the flag.
4969 	 */
4970 	sp->nst_flag |= NFSNST_RECLAIMED;
4971 
4972 	/*
4973 	 * Free up any old delegations.
4974 	 */
4975 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4976 }
4977 
4978 /*
4979  * This function is called for a reclaim, to see if it gets grace.
4980  * It returns 0 if a reclaim is allowed, 1 otherwise.
4981  */
4982 static int
nfsrv_checkstable(struct nfsclient * clp)4983 nfsrv_checkstable(struct nfsclient *clp)
4984 {
4985 	struct nfsrv_stable *sp;
4986 
4987 	/*
4988 	 * First, find the entry for the client.
4989 	 */
4990 	LIST_FOREACH(sp, &VNET(nfsrv_stablefirst).nsf_head, nst_list) {
4991 		if (sp->nst_len == clp->lc_idlen &&
4992 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4993 			break;
4994 	}
4995 
4996 	/*
4997 	 * If not in the list, state was revoked or no state was issued
4998 	 * since the previous reboot, a reclaim is denied.
4999 	 */
5000 	if (sp == LIST_END(&VNET(nfsrv_stablefirst).nsf_head) ||
5001 	    (sp->nst_flag & NFSNST_REVOKE) ||
5002 	    !(VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_OK))
5003 		return (1);
5004 	return (0);
5005 }
5006 
5007 /*
5008  * Test for and try to clear out a conflicting client. This is called by
5009  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5010  * a found.
5011  * The trick here is that it can't revoke a conflicting client with an
5012  * expired lease unless it holds the v4root lock, so...
5013  * If no v4root lock, get the lock and return 1 to indicate "try again".
5014  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5015  * the revocation worked and the conflicting client is "bye, bye", so it
5016  * can be tried again.
5017  * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5018  * Unlocks State before a non-zero value is returned.
5019  */
5020 static int
nfsrv_clientconflict(struct nfsclient * clp,int * haslockp,vnode_t vp,NFSPROC_T * p)5021 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5022     NFSPROC_T *p)
5023 {
5024 	int gotlock, lktype = 0;
5025 
5026 	/*
5027 	 * If lease hasn't expired, we can't fix it.
5028 	 */
5029 	if (clp->lc_expiry >= NFSD_MONOSEC ||
5030 	    !(VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE))
5031 		return (0);
5032 	if (*haslockp == 0) {
5033 		NFSUNLOCKSTATE();
5034 		if (vp != NULL) {
5035 			lktype = NFSVOPISLOCKED(vp);
5036 			NFSVOPUNLOCK(vp);
5037 		}
5038 		NFSLOCKV4ROOTMUTEX();
5039 		nfsv4_relref(&nfsv4rootfs_lock);
5040 		do {
5041 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5042 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5043 		} while (!gotlock);
5044 		NFSUNLOCKV4ROOTMUTEX();
5045 		*haslockp = 1;
5046 		if (vp != NULL) {
5047 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5048 			if (VN_IS_DOOMED(vp))
5049 				return (2);
5050 		}
5051 		return (1);
5052 	}
5053 	NFSUNLOCKSTATE();
5054 
5055 	/*
5056 	 * Ok, we can expire the conflicting client.
5057 	 */
5058 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5059 	nfsrv_backupstable();
5060 	nfsrv_cleanclient(clp, p, false, NULL);
5061 	nfsrv_freedeleglist(&clp->lc_deleg);
5062 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5063 	LIST_REMOVE(clp, lc_hash);
5064 	nfsrv_zapclient(clp, p);
5065 	return (1);
5066 }
5067 
5068 /*
5069  * Resolve a delegation conflict.
5070  * Returns 0 to indicate the conflict was resolved without sleeping.
5071  * Return -1 to indicate that the caller should check for conflicts again.
5072  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5073  *
5074  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5075  * for a return of 0, since there was no sleep and it could be required
5076  * later. It is released for a return of NFSERR_DELAY, since the caller
5077  * will return that error. It is released when a sleep was done waiting
5078  * for the delegation to be returned or expire (so that other nfsds can
5079  * handle ops). Then, it must be acquired for the write to stable storage.
5080  * (This function is somewhat similar to nfsrv_clientconflict(), but
5081  *  the semantics differ in a couple of subtle ways. The return of 0
5082  *  indicates the conflict was resolved without sleeping here, not
5083  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5084  *  differs, as noted above.)
5085  * Unlocks State before returning a non-zero value.
5086  */
5087 static int
nfsrv_delegconflict(struct nfsstate * stp,int * haslockp,NFSPROC_T * p,vnode_t vp)5088 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5089     vnode_t vp)
5090 {
5091 	struct nfsclient *clp = stp->ls_clp;
5092 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5093 	nfsv4stateid_t tstateid;
5094 	fhandle_t tfh;
5095 
5096 	/*
5097 	 * If the conflict is with an old delegation...
5098 	 */
5099 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5100 		/*
5101 		 * You can delete it, if it has expired.
5102 		 */
5103 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5104 			nfsrv_freedeleg(stp);
5105 			NFSUNLOCKSTATE();
5106 			error = -1;
5107 			goto out;
5108 		}
5109 		NFSUNLOCKSTATE();
5110 		/*
5111 		 * During this delay, the old delegation could expire or it
5112 		 * could be recovered by the client via an Open with
5113 		 * CLAIM_DELEGATE_PREV.
5114 		 * Release the nfsv4root_lock, if held.
5115 		 */
5116 		if (*haslockp) {
5117 			*haslockp = 0;
5118 			NFSLOCKV4ROOTMUTEX();
5119 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5120 			NFSUNLOCKV4ROOTMUTEX();
5121 		}
5122 		error = NFSERR_DELAY;
5123 		goto out;
5124 	}
5125 
5126 	/*
5127 	 * It's a current delegation, so:
5128 	 * - check to see if the delegation has expired
5129 	 *   - if so, get the v4root lock and then expire it
5130 	 */
5131 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall <
5132 	    NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC &&
5133 	    stp->ls_delegtime >= NFSD_MONOSEC)) {
5134 		/*
5135 		 * - do a recall callback, since not yet done
5136 		 * For now, never allow truncate to be set. To use
5137 		 * truncate safely, it must be guaranteed that the
5138 		 * Remove, Rename or Setattr with size of 0 will
5139 		 * succeed and that would require major changes to
5140 		 * the VFS/Vnode OPs.
5141 		 * Set the expiry time large enough so that it won't expire
5142 		 * until after the callback, then set it correctly, once
5143 		 * the callback is done. (The delegation will now time
5144 		 * out whether or not the Recall worked ok. The timeout
5145 		 * will be extended when ops are done on the delegation
5146 		 * stateid, up to the timelimit.)
5147 		 */
5148 		if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) {
5149 			stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5150 			    NFSRV_LEASEDELTA;
5151 			stp->ls_delegtimelimit = NFSD_MONOSEC + (6 *
5152 			    nfsrv_lease) + NFSRV_LEASEDELTA;
5153 			stp->ls_flags |= NFSLCK_DELEGRECALL;
5154 		}
5155 		stp->ls_lastrecall = time_uptime + 1;
5156 
5157 		/*
5158 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5159 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5160 		 * in order to try and avoid a race that could happen
5161 		 * when a CBRecall request passed the Open reply with
5162 		 * the delegation in it when transitting the network.
5163 		 * Since nfsrv_docallback will sleep, don't use stp after
5164 		 * the call.
5165 		 */
5166 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5167 		    sizeof (tstateid));
5168 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5169 		    sizeof (tfh));
5170 		NFSUNLOCKSTATE();
5171 		if (*haslockp) {
5172 			*haslockp = 0;
5173 			NFSLOCKV4ROOTMUTEX();
5174 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5175 			NFSUNLOCKV4ROOTMUTEX();
5176 		}
5177 		retrycnt = 0;
5178 		do {
5179 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5180 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5181 		    retrycnt++;
5182 		} while ((error == NFSERR_BADSTATEID ||
5183 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5184 		error = NFSERR_DELAY;
5185 		goto out;
5186 	}
5187 
5188 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5189 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5190 		NFSUNLOCKSTATE();
5191 		/*
5192 		 * A recall has been done, but it has not yet expired.
5193 		 * So, RETURN_DELAY.
5194 		 */
5195 		if (*haslockp) {
5196 			*haslockp = 0;
5197 			NFSLOCKV4ROOTMUTEX();
5198 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5199 			NFSUNLOCKV4ROOTMUTEX();
5200 		}
5201 		error = NFSERR_DELAY;
5202 		goto out;
5203 	}
5204 
5205 	/*
5206 	 * If we don't yet have the lock, just get it and then return,
5207 	 * since we need that before deleting expired state, such as
5208 	 * this delegation.
5209 	 * When getting the lock, unlock the vnode, so other nfsds that
5210 	 * are in progress, won't get stuck waiting for the vnode lock.
5211 	 */
5212 	if (*haslockp == 0) {
5213 		NFSUNLOCKSTATE();
5214 		if (vp != NULL) {
5215 			lktype = NFSVOPISLOCKED(vp);
5216 			NFSVOPUNLOCK(vp);
5217 		}
5218 		NFSLOCKV4ROOTMUTEX();
5219 		nfsv4_relref(&nfsv4rootfs_lock);
5220 		do {
5221 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5222 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5223 		} while (!gotlock);
5224 		NFSUNLOCKV4ROOTMUTEX();
5225 		*haslockp = 1;
5226 		if (vp != NULL) {
5227 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5228 			if (VN_IS_DOOMED(vp)) {
5229 				*haslockp = 0;
5230 				NFSLOCKV4ROOTMUTEX();
5231 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5232 				NFSUNLOCKV4ROOTMUTEX();
5233 				error = NFSERR_PERM;
5234 				goto out;
5235 			}
5236 		}
5237 		error = -1;
5238 		goto out;
5239 	}
5240 
5241 	NFSUNLOCKSTATE();
5242 	/*
5243 	 * Ok, we can delete the expired delegation.
5244 	 * First, write the Revoke record to stable storage and then
5245 	 * clear out the conflict.
5246 	 * Since all other nfsd threads are now blocked, we can safely
5247 	 * sleep without the state changing.
5248 	 */
5249 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5250 	nfsrv_backupstable();
5251 	if (clp->lc_expiry < NFSD_MONOSEC) {
5252 		nfsrv_cleanclient(clp, p, false, NULL);
5253 		nfsrv_freedeleglist(&clp->lc_deleg);
5254 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5255 		LIST_REMOVE(clp, lc_hash);
5256 		zapped_clp = 1;
5257 	} else {
5258 		nfsrv_freedeleg(stp);
5259 		zapped_clp = 0;
5260 	}
5261 	if (zapped_clp)
5262 		nfsrv_zapclient(clp, p);
5263 	error = -1;
5264 
5265 out:
5266 	NFSEXITCODE(error);
5267 	return (error);
5268 }
5269 
5270 /*
5271  * Check for a remove allowed, if remove is set to 1 and get rid of
5272  * delegations.
5273  */
5274 int
nfsrv_checkremove(vnode_t vp,int remove,struct nfsrv_descript * nd,nfsquad_t clientid,NFSPROC_T * p)5275 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5276     nfsquad_t clientid, NFSPROC_T *p)
5277 {
5278 	struct nfsclient *clp;
5279 	struct nfsstate *stp;
5280 	struct nfslockfile *lfp;
5281 	int error, haslock = 0;
5282 	fhandle_t nfh;
5283 
5284 	clp = NULL;
5285 	/*
5286 	 * First, get the lock file structure.
5287 	 * (A return of -1 means no associated state, so remove ok.)
5288 	 */
5289 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5290 tryagain:
5291 	NFSLOCKSTATE();
5292 	if (error == 0 && clientid.qval != 0)
5293 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5294 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
5295 	if (!error)
5296 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5297 	if (error) {
5298 		NFSUNLOCKSTATE();
5299 		if (haslock) {
5300 			NFSLOCKV4ROOTMUTEX();
5301 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5302 			NFSUNLOCKV4ROOTMUTEX();
5303 		}
5304 		if (error == -1)
5305 			error = 0;
5306 		goto out;
5307 	}
5308 
5309 	/*
5310 	 * Now, we must Recall any delegations.
5311 	 */
5312 	error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5313 	if (error) {
5314 		/*
5315 		 * nfsrv_cleandeleg() unlocks state for non-zero
5316 		 * return.
5317 		 */
5318 		if (error == -1)
5319 			goto tryagain;
5320 		if (haslock) {
5321 			NFSLOCKV4ROOTMUTEX();
5322 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5323 			NFSUNLOCKV4ROOTMUTEX();
5324 		}
5325 		goto out;
5326 	}
5327 
5328 	/*
5329 	 * Now, look for a conflicting open share.
5330 	 */
5331 	if (remove) {
5332 		/*
5333 		 * If the entry in the directory was the last reference to the
5334 		 * corresponding filesystem object, the object can be destroyed
5335 		 * */
5336 		if(lfp->lf_usecount>1)
5337 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5338 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5339 					error = NFSERR_FILEOPEN;
5340 					break;
5341 				}
5342 			}
5343 	}
5344 
5345 	NFSUNLOCKSTATE();
5346 	if (haslock) {
5347 		NFSLOCKV4ROOTMUTEX();
5348 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5349 		NFSUNLOCKV4ROOTMUTEX();
5350 	}
5351 
5352 out:
5353 	NFSEXITCODE(error);
5354 	return (error);
5355 }
5356 
5357 /*
5358  * Clear out all delegations for the file referred to by lfp.
5359  * May return NFSERR_DELAY, if there will be a delay waiting for
5360  * delegations to expire.
5361  * Returns -1 to indicate it slept while recalling a delegation.
5362  * This function has the side effect of deleting the nfslockfile structure,
5363  * if it no longer has associated state and didn't have to sleep.
5364  * Unlocks State before a non-zero value is returned.
5365  */
5366 static int
nfsrv_cleandeleg(vnode_t vp,struct nfslockfile * lfp,struct nfsclient * clp,int * haslockp,NFSPROC_T * p)5367 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5368     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5369 {
5370 	struct nfsstate *stp, *nstp;
5371 	int ret = 0;
5372 
5373 	stp = LIST_FIRST(&lfp->lf_deleg);
5374 	while (stp != LIST_END(&lfp->lf_deleg)) {
5375 		nstp = LIST_NEXT(stp, ls_file);
5376 		if (stp->ls_clp != clp) {
5377 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5378 			if (ret) {
5379 				/*
5380 				 * nfsrv_delegconflict() unlocks state
5381 				 * when it returns non-zero.
5382 				 */
5383 				goto out;
5384 			}
5385 		}
5386 		stp = nstp;
5387 	}
5388 out:
5389 	NFSEXITCODE(ret);
5390 	return (ret);
5391 }
5392 
5393 /*
5394  * There are certain operations that, when being done outside of NFSv4,
5395  * require that any NFSv4 delegation for the file be recalled.
5396  * This function is to be called for those cases:
5397  * VOP_RENAME() - When a delegation is being recalled for any reason,
5398  *	the client may have to do Opens against the server, using the file's
5399  *	final component name. If the file has been renamed on the server,
5400  *	that component name will be incorrect and the Open will fail.
5401  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5402  *	been removed on the server, if there is a delegation issued to
5403  *	that client for the file. I say "theoretically" since clients
5404  *	normally do an Access Op before the Open and that Access Op will
5405  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5406  *	they will detect the file's removal in the same manner. (There is
5407  *	one case where RFC3530 allows a client to do an Open without first
5408  *	doing an Access Op, which is passage of a check against the ACE
5409  *	returned with a Write delegation, but current practice is to ignore
5410  *	the ACE and always do an Access Op.)
5411  *	Since the functions can only be called with an unlocked vnode, this
5412  *	can't be done at this time.
5413  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5414  *	locks locally in the client, which are not visible to the server. To
5415  *	deal with this, issuing of delegations for a vnode must be disabled
5416  *	and all delegations for the vnode recalled. This is done via the
5417  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5418  */
5419 void
nfsd_recalldelegation(vnode_t vp,NFSPROC_T * p)5420 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5421 {
5422 	time_t starttime;
5423 	int error;
5424 
5425 	/*
5426 	 * First, check to see if the server is currently running and it has
5427 	 * been called for a regular file when issuing delegations.
5428 	 */
5429 	if (VNET(nfsrv_numnfsd) == 0 || vp->v_type != VREG ||
5430 	    nfsrv_issuedelegs == 0)
5431 		return;
5432 
5433 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5434 	/*
5435 	 * First, get a reference on the nfsv4rootfs_lock so that an
5436 	 * exclusive lock cannot be acquired by another thread.
5437 	 */
5438 	NFSLOCKV4ROOTMUTEX();
5439 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5440 	NFSUNLOCKV4ROOTMUTEX();
5441 
5442 	/*
5443 	 * Now, call nfsrv_checkremove() in a loop while it returns
5444 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5445 	 */
5446 	starttime = NFSD_MONOSEC;
5447 	do {
5448 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5449 			error = nfsrv_checkremove(vp, 0, NULL,
5450 			    (nfsquad_t)((u_quad_t)0), p);
5451 			NFSVOPUNLOCK(vp);
5452 		} else
5453 			error = EPERM;
5454 		if (error == NFSERR_DELAY) {
5455 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5456 				break;
5457 			/* Sleep for a short period of time */
5458 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5459 		}
5460 	} while (error == NFSERR_DELAY);
5461 	NFSLOCKV4ROOTMUTEX();
5462 	nfsv4_relref(&nfsv4rootfs_lock);
5463 	NFSUNLOCKV4ROOTMUTEX();
5464 }
5465 
5466 void
nfsd_disabledelegation(vnode_t vp,NFSPROC_T * p)5467 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5468 {
5469 
5470 #ifdef VV_DISABLEDELEG
5471 	/*
5472 	 * First, flag issuance of delegations disabled.
5473 	 */
5474 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5475 #endif
5476 
5477 	/*
5478 	 * Then call nfsd_recalldelegation() to get rid of all extant
5479 	 * delegations.
5480 	 */
5481 	nfsd_recalldelegation(vp, p);
5482 }
5483 
5484 /*
5485  * Check for conflicting locks, etc. and then get rid of delegations.
5486  * (At one point I thought that I should get rid of delegations for any
5487  *  Setattr, since it could potentially disallow the I/O op (read or write)
5488  *  allowed by the delegation. However, Setattr Ops that aren't changing
5489  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5490  *  for the same client or a different one, so I decided to only get rid
5491  *  of delegations for other clients when the size is being changed.)
5492  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5493  * as Write backs, even if there is no delegation, so it really isn't any
5494  * different?)
5495  */
5496 int
nfsrv_checksetattr(vnode_t vp,struct nfsrv_descript * nd,nfsv4stateid_t * stateidp,struct nfsvattr * nvap,nfsattrbit_t * attrbitp,struct nfsexstuff * exp,NFSPROC_T * p)5497 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5498     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5499     struct nfsexstuff *exp, NFSPROC_T *p)
5500 {
5501 	struct nfsstate st, *stp = &st;
5502 	struct nfslock lo, *lop = &lo;
5503 	int error = 0;
5504 	nfsquad_t clientid;
5505 
5506 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5507 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5508 		lop->lo_first = nvap->na_size;
5509 	} else {
5510 		stp->ls_flags = 0;
5511 		lop->lo_first = 0;
5512 	}
5513 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5514 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5515 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5516 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5517 		stp->ls_flags |= NFSLCK_SETATTR;
5518 	if (stp->ls_flags == 0)
5519 		goto out;
5520 	lop->lo_end = NFS64BITSSET;
5521 	lop->lo_flags = NFSLCK_WRITE;
5522 	stp->ls_ownerlen = 0;
5523 	stp->ls_op = NULL;
5524 	stp->ls_uid = nd->nd_cred->cr_uid;
5525 	stp->ls_stateid.seqid = stateidp->seqid;
5526 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5527 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5528 	stp->ls_stateid.other[2] = stateidp->other[2];
5529 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5530 	    stateidp, exp, nd, p);
5531 
5532 out:
5533 	NFSEXITCODE2(error, nd);
5534 	return (error);
5535 }
5536 
5537 /*
5538  * Check for a write delegation and do a CBGETATTR if there is one, updating
5539  * the attributes, as required.
5540  * Should I return an error if I can't get the attributes? (For now, I'll
5541  * just return ok.
5542  */
5543 int
nfsrv_checkgetattr(struct nfsrv_descript * nd,vnode_t vp,struct nfsvattr * nvap,nfsattrbit_t * attrbitp,NFSPROC_T * p)5544 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5545     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5546 {
5547 	struct nfsstate *stp;
5548 	struct nfslockfile *lfp;
5549 	struct nfsclient *clp;
5550 	struct nfsvattr nva;
5551 	fhandle_t nfh;
5552 	int error = 0;
5553 	nfsattrbit_t cbbits;
5554 	u_quad_t delegfilerev;
5555 
5556 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5557 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5558 		goto out;
5559 	if (nfsrv_writedelegcnt == 0)
5560 		goto out;
5561 
5562 	/*
5563 	 * Get the lock file structure.
5564 	 * (A return of -1 means no associated state, so return ok.)
5565 	 */
5566 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5567 	NFSLOCKSTATE();
5568 	if (!error)
5569 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5570 	if (error) {
5571 		NFSUNLOCKSTATE();
5572 		if (error == -1)
5573 			error = 0;
5574 		goto out;
5575 	}
5576 
5577 	/*
5578 	 * Now, look for a write delegation.
5579 	 */
5580 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5581 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5582 			break;
5583 	}
5584 	if (stp == LIST_END(&lfp->lf_deleg)) {
5585 		NFSUNLOCKSTATE();
5586 		goto out;
5587 	}
5588 	clp = stp->ls_clp;
5589 
5590 	/* If the clientid is not confirmed, ignore the delegation. */
5591 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
5592 		NFSUNLOCKSTATE();
5593 		goto out;
5594 	}
5595 
5596 	delegfilerev = stp->ls_filerev;
5597 	/*
5598 	 * If the Write delegation was issued as a part of this Compound RPC
5599 	 * or if we have an Implied Clientid (used in a previous Op in this
5600 	 * compound) and it is the client the delegation was issued to,
5601 	 * just return ok.
5602 	 * I also assume that it is from the same client iff the network
5603 	 * host IP address is the same as the callback address. (Not
5604 	 * exactly correct by the RFC, but avoids a lot of Getattr
5605 	 * callbacks.)
5606 	 */
5607 	if (nd->nd_compref == stp->ls_compref ||
5608 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5609 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5610 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5611 		NFSUNLOCKSTATE();
5612 		goto out;
5613 	}
5614 
5615 	/*
5616 	 * We are now done with the delegation state structure,
5617 	 * so the statelock can be released and we can now tsleep().
5618 	 */
5619 
5620 	/*
5621 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5622 	 * has changed.
5623 	 */
5624 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5625 		NFSUNLOCKSTATE();
5626 		NFSVNO_ATTRINIT(&nva);
5627 		nva.na_filerev = NFS64BITSSET;
5628 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5629 		    0, &nfh, &nva, &cbbits, 0, p);
5630 		if (!error) {
5631 			if ((nva.na_filerev != NFS64BITSSET &&
5632 			    nva.na_filerev > delegfilerev) ||
5633 			    (NFSVNO_ISSETSIZE(&nva) &&
5634 			     nva.na_size != nvap->na_size)) {
5635 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5636 				if (NFSVNO_ISSETSIZE(&nva))
5637 					nvap->na_size = nva.na_size;
5638 			}
5639 		} else
5640 			error = 0;	/* Ignore callback errors for now. */
5641 	} else {
5642 		NFSUNLOCKSTATE();
5643 	}
5644 
5645 out:
5646 	NFSEXITCODE2(error, nd);
5647 	return (error);
5648 }
5649 
5650 /*
5651  * This function looks for openowners that haven't had any opens for
5652  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5653  * is set.
5654  */
5655 void
nfsrv_throwawayopens(NFSPROC_T * p)5656 nfsrv_throwawayopens(NFSPROC_T *p)
5657 {
5658 	struct nfsclient *clp, *nclp;
5659 	struct nfsstate *stp, *nstp;
5660 	int i;
5661 
5662 	NFSLOCKSTATE();
5663 	VNET(nfsrv_stablefirst).nsf_flags &= ~NFSNSF_NOOPENS;
5664 	/*
5665 	 * For each client...
5666 	 */
5667 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5668 	    LIST_FOREACH_SAFE(clp, &VNET(nfsclienthash)[i], lc_hash,
5669 		nclp) {
5670 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5671 			if (LIST_EMPTY(&stp->ls_open) &&
5672 			    (stp->ls_noopens > NFSNOOPEN ||
5673 			     (nfsrv_openpluslock * 2) >
5674 			     nfsrv_v4statelimit))
5675 				nfsrv_freeopenowner(stp, 0, p);
5676 		}
5677 	    }
5678 	}
5679 	NFSUNLOCKSTATE();
5680 }
5681 
5682 /*
5683  * This function checks to see if the credentials are the same.
5684  * The check for same credentials is needed for state management operations
5685  * for NFSv4.0 or NFSv4.1/4.2 when SP4_MACH_CRED is configured via
5686  * ExchangeID.
5687  * Returns 1 for not same, 0 otherwise.
5688  */
5689 static int
nfsrv_notsamecredname(int op,struct nfsrv_descript * nd,struct nfsclient * clp)5690 nfsrv_notsamecredname(int op, struct nfsrv_descript *nd, struct nfsclient *clp)
5691 {
5692 
5693 	/* Check for the SP4_MACH_CRED case. */
5694 	if (op != 0 && nfsrv_checkmachcred(op, nd, clp) != 0)
5695 		return (1);
5696 
5697 	/* For NFSv4.1/4.2, SP4_NONE always allows this. */
5698 	if ((nd->nd_flag & ND_NFSV41) != 0)
5699 		return (0);
5700 
5701 	if (nd->nd_flag & ND_GSS) {
5702 		if (!(clp->lc_flags & LCL_GSS))
5703 			return (1);
5704 		if (clp->lc_flags & LCL_NAME) {
5705 			if (nd->nd_princlen != clp->lc_namelen ||
5706 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5707 				clp->lc_namelen))
5708 				return (1);
5709 			else
5710 				return (0);
5711 		}
5712 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5713 			return (0);
5714 		else
5715 			return (1);
5716 	} else if (clp->lc_flags & LCL_GSS)
5717 		return (1);
5718 	/*
5719 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5720 	 * in RFC3530, which talks about principals, but doesn't say anything
5721 	 * about uids for AUTH_SYS.)
5722 	 */
5723 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5724 		return (0);
5725 	else
5726 		return (1);
5727 }
5728 
5729 /*
5730  * Calculate the lease expiry time.
5731  */
5732 static time_t
nfsrv_leaseexpiry(void)5733 nfsrv_leaseexpiry(void)
5734 {
5735 
5736 	if (VNET(nfsrv_stablefirst).nsf_eograce > NFSD_MONOSEC)
5737 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5738 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5739 }
5740 
5741 /*
5742  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5743  */
5744 static void
nfsrv_delaydelegtimeout(struct nfsstate * stp)5745 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5746 {
5747 
5748 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5749 		return;
5750 
5751 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5752 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5753 		stp->ls_delegtime += nfsrv_lease;
5754 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5755 			stp->ls_delegtime = stp->ls_delegtimelimit;
5756 	}
5757 }
5758 
5759 /*
5760  * This function checks to see if there is any other state associated
5761  * with the openowner for this Open.
5762  * It returns 1 if there is no other state, 0 otherwise.
5763  */
5764 static int
nfsrv_nootherstate(struct nfsstate * stp)5765 nfsrv_nootherstate(struct nfsstate *stp)
5766 {
5767 	struct nfsstate *tstp;
5768 
5769 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5770 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5771 			return (0);
5772 	}
5773 	return (1);
5774 }
5775 
5776 /*
5777  * Create a list of lock deltas (changes to local byte range locking
5778  * that can be rolled back using the list) and apply the changes via
5779  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5780  * the rollback or update function will be called after this.
5781  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5782  * call fails. If it returns an error, it will unlock the list.
5783  */
5784 static int
nfsrv_locallock(vnode_t vp,struct nfslockfile * lfp,int flags,uint64_t first,uint64_t end,struct nfslockconflict * cfp,NFSPROC_T * p)5785 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5786     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5787 {
5788 	struct nfslock *lop, *nlop;
5789 	int error = 0;
5790 
5791 	/* Loop through the list of locks. */
5792 	lop = LIST_FIRST(&lfp->lf_locallock);
5793 	while (first < end && lop != NULL) {
5794 		nlop = LIST_NEXT(lop, lo_lckowner);
5795 		if (first >= lop->lo_end) {
5796 			/* not there yet */
5797 			lop = nlop;
5798 		} else if (first < lop->lo_first) {
5799 			/* new one starts before entry in list */
5800 			if (end <= lop->lo_first) {
5801 				/* no overlap between old and new */
5802 				error = nfsrv_dolocal(vp, lfp, flags,
5803 				    NFSLCK_UNLOCK, first, end, cfp, p);
5804 				if (error != 0)
5805 					break;
5806 				first = end;
5807 			} else {
5808 				/* handle fragment overlapped with new one */
5809 				error = nfsrv_dolocal(vp, lfp, flags,
5810 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5811 				    p);
5812 				if (error != 0)
5813 					break;
5814 				first = lop->lo_first;
5815 			}
5816 		} else {
5817 			/* new one overlaps this entry in list */
5818 			if (end <= lop->lo_end) {
5819 				/* overlaps all of new one */
5820 				error = nfsrv_dolocal(vp, lfp, flags,
5821 				    lop->lo_flags, first, end, cfp, p);
5822 				if (error != 0)
5823 					break;
5824 				first = end;
5825 			} else {
5826 				/* handle fragment overlapped with new one */
5827 				error = nfsrv_dolocal(vp, lfp, flags,
5828 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5829 				if (error != 0)
5830 					break;
5831 				first = lop->lo_end;
5832 				lop = nlop;
5833 			}
5834 		}
5835 	}
5836 	if (first < end && error == 0)
5837 		/* handle fragment past end of list */
5838 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5839 		    end, cfp, p);
5840 
5841 	NFSEXITCODE(error);
5842 	return (error);
5843 }
5844 
5845 /*
5846  * Local lock unlock. Unlock all byte ranges that are no longer locked
5847  * by NFSv4. To do this, unlock any subranges of first-->end that
5848  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5849  * list. This list has all locks for the file held by other
5850  * <clientid, lockowner> tuples. The list is ordered by increasing
5851  * lo_first value, but may have entries that overlap each other, for
5852  * the case of read locks.
5853  */
5854 static void
nfsrv_localunlock(vnode_t vp,struct nfslockfile * lfp,uint64_t init_first,uint64_t init_end,NFSPROC_T * p)5855 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5856     uint64_t init_end, NFSPROC_T *p)
5857 {
5858 	struct nfslock *lop;
5859 	uint64_t first, end, prevfirst __unused;
5860 
5861 	first = init_first;
5862 	end = init_end;
5863 	while (first < init_end) {
5864 		/* Loop through all nfs locks, adjusting first and end */
5865 		prevfirst = 0;
5866 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5867 			KASSERT(prevfirst <= lop->lo_first,
5868 			    ("nfsv4 locks out of order"));
5869 			KASSERT(lop->lo_first < lop->lo_end,
5870 			    ("nfsv4 bogus lock"));
5871 			prevfirst = lop->lo_first;
5872 			if (first >= lop->lo_first &&
5873 			    first < lop->lo_end)
5874 				/*
5875 				 * Overlaps with initial part, so trim
5876 				 * off that initial part by moving first past
5877 				 * it.
5878 				 */
5879 				first = lop->lo_end;
5880 			else if (end > lop->lo_first &&
5881 			    lop->lo_first > first) {
5882 				/*
5883 				 * This lock defines the end of the
5884 				 * segment to unlock, so set end to the
5885 				 * start of it and break out of the loop.
5886 				 */
5887 				end = lop->lo_first;
5888 				break;
5889 			}
5890 			if (first >= end)
5891 				/*
5892 				 * There is no segment left to do, so
5893 				 * break out of this loop and then exit
5894 				 * the outer while() since first will be set
5895 				 * to end, which must equal init_end here.
5896 				 */
5897 				break;
5898 		}
5899 		if (first < end) {
5900 			/* Unlock this segment */
5901 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5902 			    NFSLCK_READ, first, end, NULL, p);
5903 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5904 			    first, end);
5905 		}
5906 		/*
5907 		 * Now move past this segment and look for any further
5908 		 * segment in the range, if there is one.
5909 		 */
5910 		first = end;
5911 		end = init_end;
5912 	}
5913 }
5914 
5915 /*
5916  * Do the local lock operation and update the rollback list, as required.
5917  * Perform the rollback and return the error if nfsvno_advlock() fails.
5918  */
5919 static int
nfsrv_dolocal(vnode_t vp,struct nfslockfile * lfp,int flags,int oldflags,uint64_t first,uint64_t end,struct nfslockconflict * cfp,NFSPROC_T * p)5920 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5921     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5922 {
5923 	struct nfsrollback *rlp;
5924 	int error = 0, ltype, oldltype;
5925 
5926 	if (flags & NFSLCK_WRITE)
5927 		ltype = F_WRLCK;
5928 	else if (flags & NFSLCK_READ)
5929 		ltype = F_RDLCK;
5930 	else
5931 		ltype = F_UNLCK;
5932 	if (oldflags & NFSLCK_WRITE)
5933 		oldltype = F_WRLCK;
5934 	else if (oldflags & NFSLCK_READ)
5935 		oldltype = F_RDLCK;
5936 	else
5937 		oldltype = F_UNLCK;
5938 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5939 		/* nothing to do */
5940 		goto out;
5941 	error = nfsvno_advlock(vp, ltype, first, end, p);
5942 	if (error != 0) {
5943 		if (cfp != NULL) {
5944 			cfp->cl_clientid.lval[0] = 0;
5945 			cfp->cl_clientid.lval[1] = 0;
5946 			cfp->cl_first = 0;
5947 			cfp->cl_end = NFS64BITSSET;
5948 			cfp->cl_flags = NFSLCK_WRITE;
5949 			cfp->cl_ownerlen = 5;
5950 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5951 		}
5952 		nfsrv_locallock_rollback(vp, lfp, p);
5953 	} else if (ltype != F_UNLCK) {
5954 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5955 		    M_WAITOK);
5956 		rlp->rlck_first = first;
5957 		rlp->rlck_end = end;
5958 		rlp->rlck_type = oldltype;
5959 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5960 	}
5961 
5962 out:
5963 	NFSEXITCODE(error);
5964 	return (error);
5965 }
5966 
5967 /*
5968  * Roll back local lock changes and free up the rollback list.
5969  */
5970 static void
nfsrv_locallock_rollback(vnode_t vp,struct nfslockfile * lfp,NFSPROC_T * p)5971 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5972 {
5973 	struct nfsrollback *rlp, *nrlp;
5974 
5975 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5976 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5977 		    rlp->rlck_end, p);
5978 		free(rlp, M_NFSDROLLBACK);
5979 	}
5980 	LIST_INIT(&lfp->lf_rollback);
5981 }
5982 
5983 /*
5984  * Update local lock list and delete rollback list (ie now committed to the
5985  * local locks). Most of the work is done by the internal function.
5986  */
5987 static void
nfsrv_locallock_commit(struct nfslockfile * lfp,int flags,uint64_t first,uint64_t end)5988 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5989     uint64_t end)
5990 {
5991 	struct nfsrollback *rlp, *nrlp;
5992 	struct nfslock *new_lop, *other_lop;
5993 
5994 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5995 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5996 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5997 		    M_WAITOK);
5998 	else
5999 		other_lop = NULL;
6000 	new_lop->lo_flags = flags;
6001 	new_lop->lo_first = first;
6002 	new_lop->lo_end = end;
6003 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6004 	if (new_lop != NULL)
6005 		free(new_lop, M_NFSDLOCK);
6006 	if (other_lop != NULL)
6007 		free(other_lop, M_NFSDLOCK);
6008 
6009 	/* and get rid of the rollback list */
6010 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6011 		free(rlp, M_NFSDROLLBACK);
6012 	LIST_INIT(&lfp->lf_rollback);
6013 }
6014 
6015 /*
6016  * Lock the struct nfslockfile for local lock updating.
6017  */
6018 static void
nfsrv_locklf(struct nfslockfile * lfp)6019 nfsrv_locklf(struct nfslockfile *lfp)
6020 {
6021 	int gotlock;
6022 
6023 	/* lf_usecount ensures *lfp won't be free'd */
6024 	lfp->lf_usecount++;
6025 	do {
6026 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6027 		    NFSSTATEMUTEXPTR, NULL);
6028 	} while (gotlock == 0);
6029 	lfp->lf_usecount--;
6030 }
6031 
6032 /*
6033  * Unlock the struct nfslockfile after local lock updating.
6034  */
6035 static void
nfsrv_unlocklf(struct nfslockfile * lfp)6036 nfsrv_unlocklf(struct nfslockfile *lfp)
6037 {
6038 
6039 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6040 }
6041 
6042 /*
6043  * Clear out all state for the NFSv4 server.
6044  * Must be called by a thread that can sleep when no nfsds are running.
6045  */
6046 void
nfsrv_throwawayallstate(NFSPROC_T * p)6047 nfsrv_throwawayallstate(NFSPROC_T *p)
6048 {
6049 	struct nfsclient *clp, *nclp;
6050 	struct nfslockfile *lfp, *nlfp;
6051 	int i;
6052 
6053 	/*
6054 	 * For each client, clean out the state and then free the structure.
6055 	 */
6056 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6057 		LIST_FOREACH_SAFE(clp, &VNET(nfsclienthash)[i], lc_hash,
6058 		    nclp) {
6059 			nfsrv_cleanclient(clp, p, false, NULL);
6060 			nfsrv_freedeleglist(&clp->lc_deleg);
6061 			nfsrv_freedeleglist(&clp->lc_olddeleg);
6062 			free(clp->lc_stateid, M_NFSDCLIENT);
6063 			free(clp, M_NFSDCLIENT);
6064 		}
6065 	}
6066 
6067 	/*
6068 	 * Also, free up any remaining lock file structures.
6069 	 */
6070 	for (i = 0; i < nfsrv_lockhashsize; i++) {
6071 		LIST_FOREACH_SAFE(lfp, &VNET(nfslockhash)[i], lf_hash,
6072 		    nlfp) {
6073 			printf("nfsd unload: fnd a lock file struct\n");
6074 			nfsrv_freenfslockfile(lfp);
6075 		}
6076 	}
6077 
6078 	/* And get rid of the deviceid structures and layouts. */
6079 	nfsrv_freealllayoutsanddevids();
6080 }
6081 
6082 /*
6083  * Check the sequence# for the session and slot provided as an argument.
6084  * Also, renew the lease if the session will return NFS_OK.
6085  */
6086 int
nfsrv_checksequence(struct nfsrv_descript * nd,uint32_t sequenceid,uint32_t * highest_slotidp,uint32_t * target_highest_slotidp,int cache_this,uint32_t * sflagsp,NFSPROC_T * p)6087 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6088     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6089     uint32_t *sflagsp, NFSPROC_T *p)
6090 {
6091 	struct nfsdsession *sep;
6092 	struct nfssessionhash *shp;
6093 	int error;
6094 
6095 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6096 	NFSLOCKSESSION(shp);
6097 	sep = nfsrv_findsession(nd->nd_sessionid);
6098 	if (sep == NULL) {
6099 		NFSUNLOCKSESSION(shp);
6100 		return (NFSERR_BADSESSION);
6101 	}
6102 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6103 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6104 	if (error != 0) {
6105 		NFSUNLOCKSESSION(shp);
6106 		return (error);
6107 	}
6108 	if (cache_this != 0)
6109 		nd->nd_flag |= ND_SAVEREPLY;
6110 	/* Renew the lease. */
6111 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6112 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6113 	nd->nd_flag |= ND_IMPLIEDCLID;
6114 
6115 	/* Handle the SP4_MECH_CRED case for NFSv4.1/4.2. */
6116 	if ((sep->sess_clp->lc_flags & LCL_MACHCRED) != 0 &&
6117 	    (nd->nd_flag & (ND_GSSINTEGRITY | ND_GSSPRIVACY)) != 0 &&
6118 	    nd->nd_princlen == sep->sess_clp->lc_namelen &&
6119 	    !NFSBCMP(sep->sess_clp->lc_name, nd->nd_principal,
6120 	    nd->nd_princlen)) {
6121 		nd->nd_flag |= ND_MACHCRED;
6122 		NFSSET_OPBIT(&nd->nd_allowops, &sep->sess_clp->lc_allowops);
6123 	}
6124 
6125 	/* Save maximum request and reply sizes. */
6126 	nd->nd_maxreq = sep->sess_maxreq;
6127 	nd->nd_maxresp = sep->sess_maxresp;
6128 
6129 	*sflagsp = 0;
6130 	if (sep->sess_clp->lc_req.nr_client == NULL ||
6131 	    (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0)
6132 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6133 	NFSUNLOCKSESSION(shp);
6134 	if (error == NFSERR_EXPIRED) {
6135 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6136 		error = 0;
6137 	} else if (error == NFSERR_ADMINREVOKED) {
6138 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6139 		error = 0;
6140 	}
6141 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6142 	return (0);
6143 }
6144 
6145 /*
6146  * Check/set reclaim complete for this session/clientid.
6147  */
6148 int
nfsrv_checkreclaimcomplete(struct nfsrv_descript * nd,int onefs)6149 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6150 {
6151 	struct nfsdsession *sep;
6152 	struct nfssessionhash *shp;
6153 	int error = 0;
6154 
6155 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6156 	NFSLOCKSTATE();
6157 	NFSLOCKSESSION(shp);
6158 	sep = nfsrv_findsession(nd->nd_sessionid);
6159 	if (sep == NULL) {
6160 		NFSUNLOCKSESSION(shp);
6161 		NFSUNLOCKSTATE();
6162 		return (NFSERR_BADSESSION);
6163 	}
6164 
6165 	if (onefs != 0)
6166 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6167 		/* Check to see if reclaim complete has already happened. */
6168 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6169 		error = NFSERR_COMPLETEALREADY;
6170 	else {
6171 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6172 		nfsrv_markreclaim(sep->sess_clp);
6173 	}
6174 	NFSUNLOCKSESSION(shp);
6175 	NFSUNLOCKSTATE();
6176 	return (error);
6177 }
6178 
6179 /*
6180  * Cache the reply in a session slot.
6181  */
6182 void
nfsrv_cache_session(struct nfsrv_descript * nd,struct mbuf ** m)6183 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6184 {
6185 	struct nfsdsession *sep;
6186 	struct nfssessionhash *shp;
6187 	char *buf, *cp;
6188 #ifdef INET
6189 	struct sockaddr_in *sin;
6190 #endif
6191 #ifdef INET6
6192 	struct sockaddr_in6 *sin6;
6193 #endif
6194 
6195 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6196 	NFSLOCKSESSION(shp);
6197 	sep = nfsrv_findsession(nd->nd_sessionid);
6198 	if (sep == NULL) {
6199 		NFSUNLOCKSESSION(shp);
6200 		if ((VNET(nfsrv_stablefirst).nsf_flags &
6201 		     NFSNSF_GRACEOVER) != 0) {
6202 			buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6203 			switch (nd->nd_nam->sa_family) {
6204 #ifdef INET
6205 			case AF_INET:
6206 				sin = (struct sockaddr_in *)nd->nd_nam;
6207 				cp = inet_ntop(sin->sin_family,
6208 				    &sin->sin_addr.s_addr, buf,
6209 				    INET6_ADDRSTRLEN);
6210 				break;
6211 #endif
6212 #ifdef INET6
6213 			case AF_INET6:
6214 				sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6215 				cp = inet_ntop(sin6->sin6_family,
6216 				    &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6217 				break;
6218 #endif
6219 			default:
6220 				cp = NULL;
6221 			}
6222 			if (cp != NULL)
6223 				printf("nfsrv_cache_session: no session "
6224 				    "IPaddr=%s, check NFS clients for unique "
6225 				    "/etc/hostid's\n", cp);
6226 			else
6227 				printf("nfsrv_cache_session: no session, "
6228 				    "check NFS clients for unique "
6229 				    "/etc/hostid's\n");
6230 			free(buf, M_TEMP);
6231 		}
6232 		m_freem(*m);
6233 		return;
6234 	}
6235 	nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6236 	    m);
6237 	NFSUNLOCKSESSION(shp);
6238 }
6239 
6240 /*
6241  * Search for a session that matches the sessionid.
6242  */
6243 static struct nfsdsession *
nfsrv_findsession(uint8_t * sessionid)6244 nfsrv_findsession(uint8_t *sessionid)
6245 {
6246 	struct nfsdsession *sep;
6247 	struct nfssessionhash *shp;
6248 
6249 	shp = NFSSESSIONHASH(sessionid);
6250 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6251 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6252 			break;
6253 	}
6254 	return (sep);
6255 }
6256 
6257 /*
6258  * Destroy a session.
6259  */
6260 int
nfsrv_destroysession(struct nfsrv_descript * nd,uint8_t * sessionid)6261 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6262 {
6263 	int error, igotlock, samesess;
6264 
6265 	samesess = 0;
6266 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6267 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6268 		samesess = 1;
6269 		if ((nd->nd_flag & ND_LASTOP) == 0)
6270 			return (NFSERR_BADSESSION);
6271 	}
6272 
6273 	/* Lock out other nfsd threads */
6274 	NFSLOCKV4ROOTMUTEX();
6275 	nfsv4_relref(&nfsv4rootfs_lock);
6276 	do {
6277 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6278 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6279 	} while (igotlock == 0);
6280 	NFSUNLOCKV4ROOTMUTEX();
6281 
6282 	error = nfsrv_freesession(nd, NULL, sessionid, false, NULL);
6283 	if (error == 0 && samesess != 0)
6284 		nd->nd_flag &= ~ND_HASSEQUENCE;
6285 
6286 	NFSLOCKV4ROOTMUTEX();
6287 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6288 	NFSUNLOCKV4ROOTMUTEX();
6289 	return (error);
6290 }
6291 
6292 /*
6293  * Bind a connection to a session.
6294  * For now, only certain variants are supported, since the current session
6295  * structure can only handle a single backchannel entry, which will be
6296  * applied to all connections if it is set.
6297  */
6298 int
nfsrv_bindconnsess(struct nfsrv_descript * nd,uint8_t * sessionid,int * foreaftp)6299 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6300 {
6301 	struct nfssessionhash *shp;
6302 	struct nfsdsession *sep;
6303 	struct nfsclient *clp;
6304 	SVCXPRT *savxprt;
6305 	int error;
6306 
6307 	error = 0;
6308 	savxprt = NULL;
6309 	shp = NFSSESSIONHASH(sessionid);
6310 	NFSLOCKSTATE();
6311 	NFSLOCKSESSION(shp);
6312 	sep = nfsrv_findsession(sessionid);
6313 	if (sep != NULL) {
6314 		clp = sep->sess_clp;
6315 		error = nfsrv_checkmachcred(NFSV4OP_BINDCONNTOSESS, nd, clp);
6316 		if (error != 0)
6317 			goto out;
6318 		if (*foreaftp == NFSCDFC4_BACK ||
6319 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6320 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6321 			/* Try to set up a backchannel. */
6322 			if (clp->lc_req.nr_client == NULL) {
6323 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6324 				    "backchannel\n");
6325 				clp->lc_req.nr_client = (struct __rpc_client *)
6326 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6327 				    sep->sess_cbprogram, NFSV4_CBVERS);
6328 			}
6329 			if (clp->lc_req.nr_client != NULL) {
6330 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6331 				    "backchannel\n");
6332 				savxprt = sep->sess_cbsess.nfsess_xprt;
6333 				SVC_ACQUIRE(nd->nd_xprt);
6334 				CLNT_ACQUIRE(clp->lc_req.nr_client);
6335 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
6336 				/* Disable idle timeout. */
6337 				nd->nd_xprt->xp_idletimeout = 0;
6338 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6339 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6340 				clp->lc_flags |= LCL_DONEBINDCONN |
6341 				    LCL_NEEDSCBNULL;
6342 				clp->lc_flags &= ~LCL_CBDOWN;
6343 				if (*foreaftp == NFSCDFS4_BACK)
6344 					*foreaftp = NFSCDFS4_BACK;
6345 				else
6346 					*foreaftp = NFSCDFS4_BOTH;
6347 			} else if (*foreaftp != NFSCDFC4_BACK) {
6348 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6349 				    "up backchannel\n");
6350 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6351 				clp->lc_flags |= LCL_DONEBINDCONN;
6352 				*foreaftp = NFSCDFS4_FORE;
6353 			} else {
6354 				error = NFSERR_NOTSUPP;
6355 				printf("nfsrv_bindconnsess: Can't add "
6356 				    "backchannel\n");
6357 			}
6358 		} else {
6359 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6360 			clp->lc_flags |= LCL_DONEBINDCONN;
6361 			*foreaftp = NFSCDFS4_FORE;
6362 		}
6363 	} else
6364 		error = NFSERR_BADSESSION;
6365 out:
6366 	NFSUNLOCKSESSION(shp);
6367 	NFSUNLOCKSTATE();
6368 	if (savxprt != NULL)
6369 		SVC_RELEASE(savxprt);
6370 	return (error);
6371 }
6372 
6373 /*
6374  * Free up a session structure.
6375  */
6376 static int
nfsrv_freesession(struct nfsrv_descript * nd,struct nfsdsession * sep,uint8_t * sessionid,bool locked,SVCXPRT ** old_xprtp)6377 nfsrv_freesession(struct nfsrv_descript *nd, struct nfsdsession *sep,
6378     uint8_t *sessionid, bool locked, SVCXPRT **old_xprtp)
6379 {
6380 	struct nfssessionhash *shp;
6381 	int i;
6382 
6383 	if (!locked)
6384 		NFSLOCKSTATE();
6385 	if (sep == NULL) {
6386 		shp = NFSSESSIONHASH(sessionid);
6387 		NFSLOCKSESSION(shp);
6388 		sep = nfsrv_findsession(sessionid);
6389 	} else {
6390 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6391 		NFSLOCKSESSION(shp);
6392 	}
6393 	if (sep != NULL) {
6394 		/* Check for the SP4_MACH_CRED case. */
6395 		if (nd != NULL && nfsrv_checkmachcred(NFSV4OP_DESTROYSESSION,
6396 		    nd, sep->sess_clp) != 0) {
6397 			NFSUNLOCKSESSION(shp);
6398 			if (!locked)
6399 				NFSUNLOCKSTATE();
6400 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
6401 		}
6402 
6403 		sep->sess_refcnt--;
6404 		if (sep->sess_refcnt > 0) {
6405 			NFSUNLOCKSESSION(shp);
6406 			if (!locked)
6407 				NFSUNLOCKSTATE();
6408 			return (NFSERR_BACKCHANBUSY);
6409 		}
6410 		LIST_REMOVE(sep, sess_hash);
6411 		LIST_REMOVE(sep, sess_list);
6412 	}
6413 	NFSUNLOCKSESSION(shp);
6414 	if (!locked)
6415 		NFSUNLOCKSTATE();
6416 	if (sep == NULL)
6417 		return (NFSERR_BADSESSION);
6418 	for (i = 0; i < NFSV4_SLOTS; i++)
6419 		if (sep->sess_slots[i].nfssl_reply != NULL)
6420 			m_freem(sep->sess_slots[i].nfssl_reply);
6421 	if (!locked) {
6422 		if (sep->sess_cbsess.nfsess_xprt != NULL)
6423 			SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6424 		if (old_xprtp != NULL)
6425 			*old_xprtp = NULL;
6426 	} else if (old_xprtp != NULL)
6427 		*old_xprtp = sep->sess_cbsess.nfsess_xprt;
6428 	free(sep, M_NFSDSESSION);
6429 	return (0);
6430 }
6431 
6432 /*
6433  * Free a stateid.
6434  * RFC5661 says that it should fail when there are associated opens, locks
6435  * or delegations. Since stateids represent opens, I don't see how you can
6436  * free an open stateid (it will be free'd when closed), so this function
6437  * only works for lock stateids (freeing the lock_owner) or delegations.
6438  */
6439 int
nfsrv_freestateid(struct nfsrv_descript * nd,nfsv4stateid_t * stateidp,NFSPROC_T * p)6440 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6441     NFSPROC_T *p)
6442 {
6443 	struct nfsclient *clp;
6444 	struct nfsstate *stp;
6445 	int error;
6446 
6447 	NFSLOCKSTATE();
6448 	/*
6449 	 * Look up the stateid
6450 	 */
6451 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6452 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6453 	if (error == 0) {
6454 		/* First, check for a delegation. */
6455 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6456 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6457 			    NFSX_STATEIDOTHER))
6458 				break;
6459 		}
6460 		if (stp != NULL) {
6461 			nfsrv_freedeleg(stp);
6462 			NFSUNLOCKSTATE();
6463 			return (error);
6464 		}
6465 	}
6466 	/* Not a delegation, try for a lock_owner. */
6467 	if (error == 0)
6468 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6469 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6470 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6471 		/* Not a lock_owner stateid. */
6472 		error = NFSERR_LOCKSHELD;
6473 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6474 		error = NFSERR_LOCKSHELD;
6475 	if (error == 0)
6476 		nfsrv_freelockowner(stp, NULL, 0, p);
6477 	NFSUNLOCKSTATE();
6478 	return (error);
6479 }
6480 
6481 /*
6482  * Test a stateid.
6483  */
6484 int
nfsrv_teststateid(struct nfsrv_descript * nd,nfsv4stateid_t * stateidp,NFSPROC_T * p)6485 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6486     NFSPROC_T *p)
6487 {
6488 	struct nfsclient *clp;
6489 	struct nfsstate *stp;
6490 	int error;
6491 
6492 	NFSLOCKSTATE();
6493 	/*
6494 	 * Look up the stateid
6495 	 */
6496 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6497 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6498 	if (error == 0)
6499 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6500 	if (error == 0 && stateidp->seqid != 0 &&
6501 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6502 		error = NFSERR_OLDSTATEID;
6503 	NFSUNLOCKSTATE();
6504 	return (error);
6505 }
6506 
6507 /*
6508  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6509  */
6510 static int
nfsv4_setcbsequence(struct nfsrv_descript * nd,struct nfsclient * clp,int dont_replycache,struct nfsdsession ** sepp,int * slotposp)6511 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6512     int dont_replycache, struct nfsdsession **sepp, int *slotposp)
6513 {
6514 	struct nfsdsession *sep;
6515 	uint32_t *tl, slotseq = 0;
6516 	int maxslot;
6517 	uint8_t sessionid[NFSX_V4SESSIONID];
6518 	int error;
6519 
6520 	error = nfsv4_getcbsession(clp, sepp);
6521 	if (error != 0)
6522 		return (error);
6523 	sep = *sepp;
6524 	nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot,
6525 	    &slotseq, sessionid, true);
6526 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6527 
6528 	/* Build the Sequence arguments. */
6529 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6530 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6531 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6532 	nd->nd_slotseq = tl;
6533 	nd->nd_slotid = *slotposp;
6534 	nd->nd_flag |= ND_HASSLOTID;
6535 	*tl++ = txdr_unsigned(slotseq);
6536 	*tl++ = txdr_unsigned(*slotposp);
6537 	*tl++ = txdr_unsigned(maxslot);
6538 	if (dont_replycache == 0)
6539 		*tl++ = newnfs_true;
6540 	else
6541 		*tl++ = newnfs_false;
6542 	*tl = 0;			/* No referring call list, for now. */
6543 	nd->nd_flag |= ND_HASSEQUENCE;
6544 	return (0);
6545 }
6546 
6547 /*
6548  * Get a session for the callback.
6549  */
6550 static int
nfsv4_getcbsession(struct nfsclient * clp,struct nfsdsession ** sepp)6551 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6552 {
6553 	struct nfsdsession *sep;
6554 
6555 	NFSLOCKSTATE();
6556 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6557 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6558 			break;
6559 	}
6560 	if (sep == NULL) {
6561 		NFSUNLOCKSTATE();
6562 		return (NFSERR_BADSESSION);
6563 	}
6564 	sep->sess_refcnt++;
6565 	*sepp = sep;
6566 	NFSUNLOCKSTATE();
6567 	return (0);
6568 }
6569 
6570 /*
6571  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6572  * exit, since those transports will all be going away.
6573  * This is only called after all the nfsd threads are done performing RPCs,
6574  * so locking shouldn't be an issue.
6575  */
6576 void
nfsrv_freeallbackchannel_xprts(void)6577 nfsrv_freeallbackchannel_xprts(void)
6578 {
6579 	struct nfsdsession *sep;
6580 	struct nfsclient *clp;
6581 	SVCXPRT *xprt;
6582 	int i;
6583 
6584 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6585 		LIST_FOREACH(clp, &VNET(nfsclienthash)[i], lc_hash) {
6586 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6587 				xprt = sep->sess_cbsess.nfsess_xprt;
6588 				sep->sess_cbsess.nfsess_xprt = NULL;
6589 				if (xprt != NULL)
6590 					SVC_RELEASE(xprt);
6591 			}
6592 		}
6593 	}
6594 }
6595 
6596 /*
6597  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6598  * I have no idea if the rest of these arguments will ever be useful?
6599  */
6600 int
nfsrv_layoutcommit(struct nfsrv_descript * nd,vnode_t vp,int layouttype,int hasnewoff,uint64_t newoff,uint64_t offset,uint64_t len,int hasnewmtime,struct timespec * newmtimep,int reclaim,nfsv4stateid_t * stateidp,int maxcnt,char * layp,int * hasnewsizep,uint64_t * newsizep,struct ucred * cred,NFSPROC_T * p)6601 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6602     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6603     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6604     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6605     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6606 {
6607 	struct nfsvattr na;
6608 	int error;
6609 
6610 	error = nfsrv_updatemdsattr(vp, &na, p);
6611 	if (error == 0) {
6612 		*hasnewsizep = 1;
6613 		*newsizep = na.na_size;
6614 	}
6615 	return (error);
6616 }
6617 
6618 /*
6619  * Try and get a layout.
6620  */
6621 int
nfsrv_layoutget(struct nfsrv_descript * nd,vnode_t vp,struct nfsexstuff * exp,int layouttype,int * iomode,uint64_t * offset,uint64_t * len,uint64_t minlen,nfsv4stateid_t * stateidp,int maxcnt,int * retonclose,int * layoutlenp,char * layp,struct ucred * cred,NFSPROC_T * p)6622 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6623     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6624     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6625     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6626 {
6627 	struct nfslayouthash *lhyp;
6628 	struct nfslayout *lyp;
6629 	char *devid;
6630 	fhandle_t fh, *dsfhp;
6631 	int error, mirrorcnt, stripecnt;
6632 	uint64_t stripesiz;
6633 
6634 	if (nfsrv_devidcnt == 0)
6635 		return (NFSERR_UNKNLAYOUTTYPE);
6636 
6637 	if (*offset != 0)
6638 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6639 		    (uintmax_t)*len);
6640 	error = nfsvno_getfh(vp, &fh, p);
6641 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6642 	if (error != 0)
6643 		return (error);
6644 
6645 	/*
6646 	 * For now, all layouts are for entire files.
6647 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6648 	 */
6649 	if (NFSVNO_EXRDONLY(exp)) {
6650 		if (*iomode == NFSLAYOUTIOMODE_RW)
6651 			return (NFSERR_LAYOUTTRYLATER);
6652 		*iomode = NFSLAYOUTIOMODE_READ;
6653 	}
6654 	if (*iomode != NFSLAYOUTIOMODE_RW)
6655 		*iomode = NFSLAYOUTIOMODE_READ;
6656 
6657 	/*
6658 	 * Check to see if a write layout can be issued for this file.
6659 	 * This is used during mirror recovery to avoid RW layouts being
6660 	 * issued for a file while it is being copied to the recovered
6661 	 * mirror.
6662 	 */
6663 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6664 		return (NFSERR_LAYOUTTRYLATER);
6665 
6666 	*retonclose = 0;
6667 	*offset = 0;
6668 	*len = UINT64_MAX;
6669 
6670 	/* First, see if a layout already exists and return if found. */
6671 	lhyp = NFSLAYOUTHASH(&fh);
6672 	NFSLOCKLAYOUT(lhyp);
6673 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6674 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6675 	/*
6676 	 * Not sure if the seqid must be the same, so I won't check it.
6677 	 */
6678 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6679 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6680 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6681 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6682 			NFSUNLOCKLAYOUT(lhyp);
6683 			NFSD_DEBUG(1, "ret bad stateid\n");
6684 			return (NFSERR_BADSTATEID);
6685 		}
6686 		/*
6687 		 * I believe we get here because there is a race between
6688 		 * the client processing the CBLAYOUTRECALL and the layout
6689 		 * being deleted here on the server.
6690 		 * The client has now done a LayoutGet with a non-layout
6691 		 * stateid, as it would when there is no layout.
6692 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6693 		 * so the code below will create a new layout structure as
6694 		 * would happen if no layout was found.
6695 		 * "lyp" will be set before being used below, but set it NULL
6696 		 * as a safety belt.
6697 		 */
6698 		nfsrv_freelayout(&lhyp->list, lyp);
6699 		lyp = NULL;
6700 		error = NFSERR_BADSTATEID;
6701 	}
6702 	if (error == 0) {
6703 		if (lyp->lay_layoutlen > maxcnt) {
6704 			NFSUNLOCKLAYOUT(lhyp);
6705 			NFSD_DEBUG(1, "ret layout too small\n");
6706 			return (NFSERR_TOOSMALL);
6707 		}
6708 		if (*iomode == NFSLAYOUTIOMODE_RW) {
6709 			if ((lyp->lay_flags & NFSLAY_NOSPC) != 0) {
6710 				NFSUNLOCKLAYOUT(lhyp);
6711 				NFSD_DEBUG(1, "ret layout nospace\n");
6712 				return (NFSERR_NOSPC);
6713 			}
6714 			lyp->lay_flags |= NFSLAY_RW;
6715 		} else
6716 			lyp->lay_flags |= NFSLAY_READ;
6717 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6718 		*layoutlenp = lyp->lay_layoutlen;
6719 		if (++lyp->lay_stateid.seqid == 0)
6720 			lyp->lay_stateid.seqid = 1;
6721 		stateidp->seqid = lyp->lay_stateid.seqid;
6722 		NFSUNLOCKLAYOUT(lhyp);
6723 		NFSD_DEBUG(4, "ret fnd layout\n");
6724 		return (0);
6725 	}
6726 	NFSUNLOCKLAYOUT(lhyp);
6727 
6728 	/* Find the device id and file handle. */
6729 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, &stripesiz, &stripecnt,
6730 	    &dsfhp, &devid);
6731 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6732 	if (error == 0) {
6733 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6734 			if (NFSX_V4FILELAYOUT > maxcnt)
6735 				error = NFSERR_TOOSMALL;
6736 			else
6737 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6738 				    devid, vp->v_mount->mnt_stat.f_fsid);
6739 		} else {
6740 			if (NFSX_V4FLEXLAYOUT(mirrorcnt, stripecnt) > maxcnt)
6741 				error = NFSERR_TOOSMALL;
6742 			else
6743 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6744 				    stripesiz, stripecnt, &fh, dsfhp, devid,
6745 				    vp->v_mount->mnt_stat.f_fsid);
6746 		}
6747 	}
6748 	free(dsfhp, M_TEMP);
6749 	free(devid, M_TEMP);
6750 	if (error != 0)
6751 		return (error);
6752 
6753 	/*
6754 	 * Now, add this layout to the list.
6755 	 */
6756 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6757 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6758 	/*
6759 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6760 	 * linked the new structure into the lists.
6761 	 */
6762 	free(lyp, M_NFSDSTATE);
6763 	return (error);
6764 }
6765 
6766 /*
6767  * Generate a File Layout.
6768  */
6769 static struct nfslayout *
nfsrv_filelayout(struct nfsrv_descript * nd,int iomode,fhandle_t * fhp,fhandle_t * dsfhp,char * devid,fsid_t fs)6770 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6771     fhandle_t *dsfhp, char *devid, fsid_t fs)
6772 {
6773 	uint32_t *tl;
6774 	struct nfslayout *lyp;
6775 	uint64_t pattern_offset;
6776 
6777 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6778 	    M_WAITOK | M_ZERO);
6779 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6780 	if (iomode == NFSLAYOUTIOMODE_RW)
6781 		lyp->lay_flags = NFSLAY_RW;
6782 	else
6783 		lyp->lay_flags = NFSLAY_READ;
6784 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6785 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6786 	lyp->lay_fsid = fs;
6787 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6788 
6789 	/* Fill in the xdr for the files layout. */
6790 	tl = (uint32_t *)lyp->lay_xdr;
6791 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6792 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6793 
6794 	/* Set the stripe size to the maximum I/O size. */
6795 	*tl++ = txdr_unsigned(nfs_srvmaxio & NFSFLAYUTIL_STRIPE_MASK);
6796 	*tl++ = 0;					/* 1st stripe index. */
6797 	pattern_offset = 0;
6798 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
6799 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
6800 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6801 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6802 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6803 	return (lyp);
6804 }
6805 
6806 #define	FLEX_OWNERID	"999"
6807 #define	FLEX_UID0	"0"
6808 /*
6809  * Generate a Flex File Layout.
6810  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6811  * string goes on the wire, it isn't supposed to be used by the client,
6812  * since this server uses tight coupling.
6813  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6814  * a string of "0". This works around the Linux Flex File Layout driver bug
6815  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6816  */
6817 static struct nfslayout *
nfsrv_flexlayout(struct nfsrv_descript * nd,int iomode,int mirrorcnt,uint64_t stripesiz,int stripecnt,fhandle_t * fhp,fhandle_t * dsfhp,char * devid,fsid_t fs)6818 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6819     uint64_t stripesiz, int stripecnt, fhandle_t *fhp, fhandle_t *dsfhp,
6820     char *devid, fsid_t fs)
6821 {
6822 	uint32_t *tl;
6823 	struct nfslayout *lyp;
6824 	int i, j;
6825 
6826 	lyp = malloc(sizeof(struct nfslayout) +
6827 	    NFSX_V4FLEXLAYOUT(mirrorcnt, stripecnt), M_NFSDSTATE,
6828 	    M_WAITOK | M_ZERO);
6829 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
6830 	if (iomode == NFSLAYOUTIOMODE_RW)
6831 		lyp->lay_flags = NFSLAY_RW;
6832 	else
6833 		lyp->lay_flags = NFSLAY_READ;
6834 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6835 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6836 	lyp->lay_fsid = fs;
6837 	lyp->lay_mirrorcnt = mirrorcnt;
6838 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6839 
6840 	/* Fill in the xdr for the files layout. */
6841 	tl = (uint32_t *)lyp->lay_xdr;
6842 	txdr_hyper(stripesiz, tl); tl += 2;		/* Stripe unit. */
6843 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
6844 	for (i = 0; i < mirrorcnt; i++) {
6845 		*tl++ = txdr_unsigned(stripecnt);	/* Stripe cnt. */
6846 		for (j = 0; j < stripecnt; j++) {
6847 			NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
6848 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6849 			devid += NFSX_V4DEVICEID;
6850 			*tl++ = txdr_unsigned(1);	/* Efficiency. */
6851 			*tl++ = 0;			/* Proxy Stateid. */
6852 			*tl++ = 0x55555555;
6853 			*tl++ = 0x55555555;
6854 			*tl++ = 0x55555555;
6855 			*tl++ = txdr_unsigned(1);	/* 1 file handle. */
6856 			*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6857 			NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6858 			tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
6859 			dsfhp++;
6860 			if (nfsrv_flexlinuxhack != 0) {
6861 				*tl++ = txdr_unsigned(strlen(FLEX_UID0));
6862 				*tl = 0;		/* 0 pad string. */
6863 				NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6864 				*tl++ = txdr_unsigned(strlen(FLEX_UID0));
6865 				*tl = 0;		/* 0 pad string. */
6866 				NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6867 			} else {
6868 				*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6869 				NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6870 				*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6871 				NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6872 			}
6873 		}
6874 	}
6875 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
6876 	*tl = txdr_unsigned(60);		/* Status interval hint. */
6877 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt, stripecnt);
6878 	return (lyp);
6879 }
6880 
6881 /*
6882  * Parse and process Flex File errors returned via LayoutReturn.
6883  */
6884 static void
nfsrv_flexlayouterr(struct nfsrv_descript * nd,uint32_t * layp,int maxcnt,NFSPROC_T * p)6885 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
6886     NFSPROC_T *p)
6887 {
6888 	uint32_t *tl;
6889 	int cnt, errcnt, i, j, opnum, stat;
6890 	char devid[NFSX_V4DEVICEID];
6891 
6892 	tl = layp;
6893 	maxcnt -= NFSX_UNSIGNED;
6894 	if (maxcnt > 0)
6895 		cnt = fxdr_unsigned(int, *tl++);
6896 	else
6897 		cnt = 0;
6898 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
6899 	for (i = 0; i < cnt; i++) {
6900 		maxcnt -= NFSX_STATEID + 2 * NFSX_HYPER +
6901 		    NFSX_UNSIGNED;
6902 		if (maxcnt <= 0)
6903 			break;
6904 		/* Skip offset, length and stateid for now. */
6905 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
6906 		errcnt = fxdr_unsigned(int, *tl++);
6907 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
6908 		for (j = 0; j < errcnt; j++) {
6909 			maxcnt -= NFSX_V4DEVICEID + 2 * NFSX_UNSIGNED;
6910 			if (maxcnt < 0)
6911 				break;
6912 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
6913 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6914 			stat = fxdr_unsigned(int, *tl++);
6915 			opnum = fxdr_unsigned(int, *tl++);
6916 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
6917 			    stat);
6918 			/*
6919 			 * Except for NFSERR_ACCES, NFSERR_STALE and
6920 			 * NFSERR_NOSPC errors, disable the mirror.
6921 			 */
6922 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE &&
6923 			    stat != NFSERR_NOSPC)
6924 				nfsrv_delds(devid, p);
6925 
6926 			/* For NFSERR_NOSPC, mark all devids and layouts. */
6927 			if (stat == NFSERR_NOSPC)
6928 				nfsrv_marknospc(devid, true);
6929 		}
6930 	}
6931 }
6932 
6933 /*
6934  * This function removes all flex file layouts which has a mirror with
6935  * a device id that matches the argument.
6936  * Called when the DS represented by the device id has failed.
6937  */
6938 void
nfsrv_flexmirrordel(char * devid,NFSPROC_T * p)6939 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
6940 {
6941 	uint32_t *tl;
6942 	struct nfslayout *lyp, *nlyp;
6943 	struct nfslayouthash *lhyp;
6944 	struct nfslayouthead loclyp;
6945 	int i, j;
6946 
6947 	NFSD_DEBUG(4, "flexmirrordel\n");
6948 	/* Move all layouts found onto a local list. */
6949 	TAILQ_INIT(&loclyp);
6950 	for (i = 0; i < nfsrv_layouthashsize; i++) {
6951 		lhyp = &nfslayouthash[i];
6952 		NFSLOCKLAYOUT(lhyp);
6953 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
6954 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
6955 			    lyp->lay_mirrorcnt > 1) {
6956 				NFSD_DEBUG(4, "possible match\n");
6957 				tl = lyp->lay_xdr;
6958 				tl += 3;
6959 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
6960 					tl++;
6961 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
6962 					    == 0) {
6963 						/* Found one. */
6964 						NFSD_DEBUG(4, "fnd one\n");
6965 						TAILQ_REMOVE(&lhyp->list, lyp,
6966 						    lay_list);
6967 						TAILQ_INSERT_HEAD(&loclyp, lyp,
6968 						    lay_list);
6969 						break;
6970 					}
6971 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
6972 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
6973 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
6974 				}
6975 			}
6976 		}
6977 		NFSUNLOCKLAYOUT(lhyp);
6978 	}
6979 
6980 	/* Now, try to do a Layout recall for each one found. */
6981 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
6982 		NFSD_DEBUG(4, "do layout recall\n");
6983 		/*
6984 		 * The layout stateid.seqid needs to be incremented
6985 		 * before doing a LAYOUT_RECALL callback.
6986 		 */
6987 		if (++lyp->lay_stateid.seqid == 0)
6988 			lyp->lay_stateid.seqid = 1;
6989 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
6990 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
6991 		nfsrv_freelayout(&loclyp, lyp);
6992 	}
6993 }
6994 
6995 /*
6996  * Do a recall callback to the client for this layout.
6997  */
6998 static int
nfsrv_recalllayout(nfsquad_t clid,nfsv4stateid_t * stateidp,fhandle_t * fhp,struct nfslayout * lyp,int changed,int laytype,NFSPROC_T * p)6999 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7000     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7001 {
7002 	struct nfsclient *clp;
7003 	int error;
7004 
7005 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7006 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7007 	    0, NULL, p);
7008 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7009 	if (error != 0) {
7010 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
7011 		return (error);
7012 	}
7013 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
7014 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7015 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
7016 		/* If lyp != NULL, handle an error return here. */
7017 		if (error != 0 && lyp != NULL) {
7018 			NFSDRECALLLOCK();
7019 			/*
7020 			 * Mark it returned, since no layout recall
7021 			 * has been done.
7022 			 * All errors seem to be non-recoverable, although
7023 			 * NFSERR_NOMATCHLAYOUT is a normal event.
7024 			 */
7025 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7026 				lyp->lay_flags |= NFSLAY_RETURNED;
7027 				wakeup(lyp);
7028 			}
7029 			NFSDRECALLUNLOCK();
7030 			if (error != NFSERR_NOMATCHLAYOUT)
7031 				printf("nfsrv_recalllayout: err=%d\n", error);
7032 		}
7033 	} else
7034 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7035 	return (error);
7036 }
7037 
7038 /*
7039  * Find a layout to recall when we exceed our high water mark.
7040  */
7041 void
nfsrv_recalloldlayout(NFSPROC_T * p)7042 nfsrv_recalloldlayout(NFSPROC_T *p)
7043 {
7044 	struct nfslayouthash *lhyp;
7045 	struct nfslayout *lyp;
7046 	nfsquad_t clientid;
7047 	nfsv4stateid_t stateid;
7048 	fhandle_t fh;
7049 	int error, laytype = 0, ret;
7050 
7051 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7052 	NFSLOCKLAYOUT(lhyp);
7053 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7054 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7055 			lyp->lay_flags |= NFSLAY_CALLB;
7056 			/*
7057 			 * The layout stateid.seqid needs to be incremented
7058 			 * before doing a LAYOUT_RECALL callback.
7059 			 */
7060 			if (++lyp->lay_stateid.seqid == 0)
7061 				lyp->lay_stateid.seqid = 1;
7062 			clientid = lyp->lay_clientid;
7063 			stateid = lyp->lay_stateid;
7064 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7065 			laytype = lyp->lay_type;
7066 			break;
7067 		}
7068 	}
7069 	NFSUNLOCKLAYOUT(lhyp);
7070 	if (lyp != NULL) {
7071 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7072 		    laytype, p);
7073 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7074 			NFSD_DEBUG(4, "recallold=%d\n", error);
7075 		if (error != 0) {
7076 			NFSLOCKLAYOUT(lhyp);
7077 			/*
7078 			 * Since the hash list was unlocked, we need to
7079 			 * find it again.
7080 			 */
7081 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7082 			    &lyp);
7083 			if (ret == 0 &&
7084 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7085 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
7086 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
7087 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
7088 				/*
7089 				 * The client no longer knows this layout, so
7090 				 * it can be free'd now.
7091 				 */
7092 				if (error == NFSERR_NOMATCHLAYOUT)
7093 					nfsrv_freelayout(&lhyp->list, lyp);
7094 				else {
7095 					/*
7096 					 * Leave it to be tried later by
7097 					 * clearing NFSLAY_CALLB and moving
7098 					 * it to the head of the list, so it
7099 					 * won't be tried again for a while.
7100 					 */
7101 					lyp->lay_flags &= ~NFSLAY_CALLB;
7102 					TAILQ_REMOVE(&lhyp->list, lyp,
7103 					    lay_list);
7104 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7105 					    lay_list);
7106 				}
7107 			}
7108 			NFSUNLOCKLAYOUT(lhyp);
7109 		}
7110 	}
7111 }
7112 
7113 /*
7114  * Try and return layout(s).
7115  */
7116 int
nfsrv_layoutreturn(struct nfsrv_descript * nd,vnode_t vp,int layouttype,int iomode,uint64_t offset,uint64_t len,int reclaim,int kind,nfsv4stateid_t * stateidp,int maxcnt,uint32_t * layp,int * fndp,struct ucred * cred,NFSPROC_T * p)7117 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7118     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7119     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7120     struct ucred *cred, NFSPROC_T *p)
7121 {
7122 	struct nfsvattr na;
7123 	struct nfslayouthash *lhyp;
7124 	struct nfslayout *lyp;
7125 	fhandle_t fh;
7126 	int error = 0;
7127 
7128 	*fndp = 0;
7129 	if (kind == NFSV4LAYOUTRET_FILE) {
7130 		error = nfsvno_getfh(vp, &fh, p);
7131 		if (error == 0) {
7132 			error = nfsrv_updatemdsattr(vp, &na, p);
7133 			if (error != 0 && error != ESTALE)
7134 				printf("nfsrv_layoutreturn: updatemdsattr"
7135 				    " failed=%d\n", error);
7136 		}
7137 		if (error == 0) {
7138 			if (reclaim == newnfs_true) {
7139 				error = nfsrv_checkgrace(NULL, NULL,
7140 				    NFSLCK_RECLAIM);
7141 				if (error != NFSERR_NOGRACE)
7142 					error = 0;
7143 				return (error);
7144 			}
7145 			lhyp = NFSLAYOUTHASH(&fh);
7146 			NFSDRECALLLOCK();
7147 			NFSLOCKLAYOUT(lhyp);
7148 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7149 			    layouttype, p, &lyp);
7150 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7151 			if (error == 0 &&
7152 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
7153 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
7154 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
7155 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7156 				    " %x %x %x laystateid %d %x %x %x"
7157 				    " off=%ju len=%ju flgs=0x%x\n",
7158 				    stateidp->seqid, stateidp->other[0],
7159 				    stateidp->other[1], stateidp->other[2],
7160 				    lyp->lay_stateid.seqid,
7161 				    lyp->lay_stateid.other[0],
7162 				    lyp->lay_stateid.other[1],
7163 				    lyp->lay_stateid.other[2],
7164 				    (uintmax_t)offset, (uintmax_t)len,
7165 				    lyp->lay_flags);
7166 				if (++lyp->lay_stateid.seqid == 0)
7167 					lyp->lay_stateid.seqid = 1;
7168 				stateidp->seqid = lyp->lay_stateid.seqid;
7169 				if (offset == 0 && len == UINT64_MAX) {
7170 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7171 					    0)
7172 						lyp->lay_flags &= ~NFSLAY_READ;
7173 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7174 						lyp->lay_flags &= ~NFSLAY_RW;
7175 					if ((lyp->lay_flags & (NFSLAY_READ |
7176 					    NFSLAY_RW)) == 0)
7177 						nfsrv_freelayout(&lhyp->list,
7178 						    lyp);
7179 					else
7180 						*fndp = 1;
7181 				} else
7182 					*fndp = 1;
7183 			}
7184 			NFSUNLOCKLAYOUT(lhyp);
7185 			/* Search the nfsrv_recalllist for a match. */
7186 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7187 				if (NFSBCMP(&lyp->lay_fh, &fh,
7188 				    sizeof(fh)) == 0 &&
7189 				    lyp->lay_clientid.qval ==
7190 				    nd->nd_clientid.qval &&
7191 				    stateidp->other[0] ==
7192 				    lyp->lay_stateid.other[0] &&
7193 				    stateidp->other[1] ==
7194 				    lyp->lay_stateid.other[1] &&
7195 				    stateidp->other[2] ==
7196 				    lyp->lay_stateid.other[2]) {
7197 					lyp->lay_flags |= NFSLAY_RETURNED;
7198 					wakeup(lyp);
7199 					error = 0;
7200 				}
7201 			}
7202 			NFSDRECALLUNLOCK();
7203 		}
7204 		if (layouttype == NFSLAYOUT_FLEXFILE && layp != NULL)
7205 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7206 	} else if (kind == NFSV4LAYOUTRET_FSID)
7207 		nfsrv_freelayouts(&nd->nd_clientid,
7208 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7209 	else if (kind == NFSV4LAYOUTRET_ALL)
7210 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7211 	else
7212 		error = NFSERR_INVAL;
7213 	if (error == -1)
7214 		error = 0;
7215 	return (error);
7216 }
7217 
7218 /*
7219  * Look for an existing layout.
7220  */
7221 static int
nfsrv_findlayout(nfsquad_t * clientidp,fhandle_t * fhp,int laytype,NFSPROC_T * p,struct nfslayout ** lypp)7222 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7223     NFSPROC_T *p, struct nfslayout **lypp)
7224 {
7225 	struct nfslayouthash *lhyp;
7226 	struct nfslayout *lyp;
7227 	int ret;
7228 
7229 	*lypp = NULL;
7230 	ret = 0;
7231 	lhyp = NFSLAYOUTHASH(fhp);
7232 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7233 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7234 		    lyp->lay_clientid.qval == clientidp->qval &&
7235 		    lyp->lay_type == laytype)
7236 			break;
7237 	}
7238 	if (lyp != NULL)
7239 		*lypp = lyp;
7240 	else
7241 		ret = -1;
7242 	return (ret);
7243 }
7244 
7245 /*
7246  * Add the new layout, as required.
7247  */
7248 static int
nfsrv_addlayout(struct nfsrv_descript * nd,struct nfslayout ** lypp,nfsv4stateid_t * stateidp,char * layp,int * layoutlenp,NFSPROC_T * p)7249 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7250     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7251 {
7252 	struct nfsclient *clp;
7253 	struct nfslayouthash *lhyp;
7254 	struct nfslayout *lyp, *nlyp;
7255 	fhandle_t *fhp;
7256 	int error;
7257 
7258 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7259 	    ("nfsrv_layoutget: no nd_clientid\n"));
7260 	lyp = *lypp;
7261 	fhp = &lyp->lay_fh;
7262 	NFSLOCKSTATE();
7263 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7264 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7265 	if (error != 0) {
7266 		NFSUNLOCKSTATE();
7267 		return (error);
7268 	}
7269 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7270 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7271 	    clp->lc_clientid.lval[0];
7272 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7273 	    clp->lc_clientid.lval[1];
7274 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7275 	    nfsrv_nextstateindex(clp);
7276 	NFSUNLOCKSTATE();
7277 
7278 	lhyp = NFSLAYOUTHASH(fhp);
7279 	NFSLOCKLAYOUT(lhyp);
7280 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7281 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7282 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7283 			break;
7284 	}
7285 	if (nlyp != NULL) {
7286 		/* A layout already exists, so use it. */
7287 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7288 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7289 		*layoutlenp = nlyp->lay_layoutlen;
7290 		if (++nlyp->lay_stateid.seqid == 0)
7291 			nlyp->lay_stateid.seqid = 1;
7292 		stateidp->seqid = nlyp->lay_stateid.seqid;
7293 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7294 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7295 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7296 		NFSUNLOCKLAYOUT(lhyp);
7297 		return (0);
7298 	}
7299 
7300 	/* Insert the new layout in the lists. */
7301 	*lypp = NULL;
7302 	atomic_add_int(&nfsrv_layoutcnt, 1);
7303 	VNET(nfsstatsv1_p)->srvlayouts++;
7304 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7305 	*layoutlenp = lyp->lay_layoutlen;
7306 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7307 	NFSUNLOCKLAYOUT(lhyp);
7308 	return (0);
7309 }
7310 
7311 /*
7312  * Get the devinfo for a deviceid.
7313  */
7314 int
nfsrv_getdevinfo(char * devid,int layouttype,uint32_t * maxcnt,uint32_t * notify,int * devaddrlen,char ** devaddr)7315 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7316     uint32_t *notify, int *devaddrlen, char **devaddr)
7317 {
7318 	struct nfsdevice *ds;
7319 
7320 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7321 	     NFSLAYOUT_FLEXFILE) ||
7322 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7323 		return (NFSERR_UNKNLAYOUTTYPE);
7324 
7325 	/*
7326 	 * Now, search for the device id.  Note that the structures won't go
7327 	 * away, but the order changes in the list.  As such, the lock only
7328 	 * needs to be held during the search through the list.
7329 	 */
7330 	NFSDDSLOCK();
7331 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7332 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7333 		    ds->nfsdev_nmp != NULL)
7334 			break;
7335 	}
7336 	NFSDDSUNLOCK();
7337 	if (ds == NULL)
7338 		return (NFSERR_NOENT);
7339 
7340 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7341 	*devaddrlen = 0;
7342 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7343 		*devaddrlen = ds->nfsdev_fileaddrlen;
7344 		*devaddr = ds->nfsdev_fileaddr;
7345 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7346 		*devaddrlen = ds->nfsdev_flexaddrlen;
7347 		*devaddr = ds->nfsdev_flexaddr;
7348 	}
7349 	if (*devaddrlen == 0)
7350 		return (NFSERR_UNKNLAYOUTTYPE);
7351 
7352 	/*
7353 	 * The XDR overhead is 3 unsigned values: layout_type,
7354 	 * length_of_address and notify bitmap.
7355 	 * If the notify array is changed to not all zeros, the
7356 	 * count of unsigned values must be increased.
7357 	 */
7358 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7359 	    3 * NFSX_UNSIGNED) {
7360 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7361 		return (NFSERR_TOOSMALL);
7362 	}
7363 	return (0);
7364 }
7365 
7366 /*
7367  * Free a list of layout state structures.
7368  */
7369 static void
nfsrv_freelayoutlist(nfsquad_t clientid)7370 nfsrv_freelayoutlist(nfsquad_t clientid)
7371 {
7372 	struct nfslayouthash *lhyp;
7373 	struct nfslayout *lyp, *nlyp;
7374 	int i;
7375 
7376 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7377 		lhyp = &nfslayouthash[i];
7378 		NFSLOCKLAYOUT(lhyp);
7379 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7380 			if (lyp->lay_clientid.qval == clientid.qval)
7381 				nfsrv_freelayout(&lhyp->list, lyp);
7382 		}
7383 		NFSUNLOCKLAYOUT(lhyp);
7384 	}
7385 }
7386 
7387 /*
7388  * Free up a layout.
7389  */
7390 static void
nfsrv_freelayout(struct nfslayouthead * lhp,struct nfslayout * lyp)7391 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7392 {
7393 
7394 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7395 	atomic_add_int(&nfsrv_layoutcnt, -1);
7396 	VNET(nfsstatsv1_p)->srvlayouts--;
7397 	TAILQ_REMOVE(lhp, lyp, lay_list);
7398 	free(lyp, M_NFSDSTATE);
7399 }
7400 
7401 /*
7402  * Free up a device id.
7403  */
7404 void
nfsrv_freeonedevid(struct nfsdevice * ds)7405 nfsrv_freeonedevid(struct nfsdevice *ds)
7406 {
7407 	int i;
7408 
7409 	atomic_add_int(&nfsrv_devidcnt, -1);
7410 	vrele(ds->nfsdev_dvp);
7411 	for (i = 0; i < nfsrv_dsdirsize; i++)
7412 		if (ds->nfsdev_dsdir[i] != NULL)
7413 			vrele(ds->nfsdev_dsdir[i]);
7414 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7415 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7416 	free(ds->nfsdev_host, M_NFSDSTATE);
7417 	free(ds, M_NFSDSTATE);
7418 }
7419 
7420 /*
7421  * Free up a device id and its mirrors.
7422  */
7423 static void
nfsrv_freedevid(struct nfsdevice * ds)7424 nfsrv_freedevid(struct nfsdevice *ds)
7425 {
7426 
7427 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7428 	nfsrv_freeonedevid(ds);
7429 }
7430 
7431 /*
7432  * Free all layouts and device ids.
7433  * Done when the nfsd threads are shut down since there may be a new
7434  * modified device id list created when the nfsd is restarted.
7435  */
7436 void
nfsrv_freealllayoutsanddevids(void)7437 nfsrv_freealllayoutsanddevids(void)
7438 {
7439 	struct nfsdontlist *mrp, *nmrp;
7440 	struct nfslayout *lyp, *nlyp;
7441 
7442 	/* Get rid of the deviceid structures. */
7443 	nfsrv_freealldevids();
7444 	TAILQ_INIT(&nfsrv_devidhead);
7445 	nfsrv_devidcnt = 0;
7446 
7447 	/* Get rid of all layouts. */
7448 	nfsrv_freealllayouts();
7449 
7450 	/* Get rid of any nfsdontlist entries. */
7451 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7452 		free(mrp, M_NFSDSTATE);
7453 	LIST_INIT(&nfsrv_dontlisthead);
7454 	nfsrv_dontlistlen = 0;
7455 
7456 	/* Free layouts in the recall list. */
7457 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7458 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7459 	TAILQ_INIT(&nfsrv_recalllisthead);
7460 }
7461 
7462 /*
7463  * Free layouts that match the arguments.
7464  */
7465 static void
nfsrv_freelayouts(nfsquad_t * clid,fsid_t * fs,int laytype,int iomode)7466 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7467 {
7468 	struct nfslayouthash *lhyp;
7469 	struct nfslayout *lyp, *nlyp;
7470 	int i;
7471 
7472 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7473 		lhyp = &nfslayouthash[i];
7474 		NFSLOCKLAYOUT(lhyp);
7475 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7476 			if (clid->qval != lyp->lay_clientid.qval)
7477 				continue;
7478 			if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7479 				continue;
7480 			if (laytype != lyp->lay_type)
7481 				continue;
7482 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7483 				lyp->lay_flags &= ~NFSLAY_READ;
7484 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7485 				lyp->lay_flags &= ~NFSLAY_RW;
7486 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7487 				nfsrv_freelayout(&lhyp->list, lyp);
7488 		}
7489 		NFSUNLOCKLAYOUT(lhyp);
7490 	}
7491 }
7492 
7493 /*
7494  * Free all layouts for the argument file.
7495  */
7496 void
nfsrv_freefilelayouts(fhandle_t * fhp)7497 nfsrv_freefilelayouts(fhandle_t *fhp)
7498 {
7499 	struct nfslayouthash *lhyp;
7500 	struct nfslayout *lyp, *nlyp;
7501 
7502 	lhyp = NFSLAYOUTHASH(fhp);
7503 	NFSLOCKLAYOUT(lhyp);
7504 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7505 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7506 			nfsrv_freelayout(&lhyp->list, lyp);
7507 	}
7508 	NFSUNLOCKLAYOUT(lhyp);
7509 }
7510 
7511 /*
7512  * Free all layouts.
7513  */
7514 static void
nfsrv_freealllayouts(void)7515 nfsrv_freealllayouts(void)
7516 {
7517 	struct nfslayouthash *lhyp;
7518 	struct nfslayout *lyp, *nlyp;
7519 	int i;
7520 
7521 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7522 		lhyp = &nfslayouthash[i];
7523 		NFSLOCKLAYOUT(lhyp);
7524 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7525 			nfsrv_freelayout(&lhyp->list, lyp);
7526 		NFSUNLOCKLAYOUT(lhyp);
7527 	}
7528 }
7529 
7530 /*
7531  * Look up the mount path for the DS server.
7532  */
7533 static int
nfsrv_setdsserver(char * dspathp,char * mdspathp,NFSPROC_T * p,struct nfsdevice ** dsp)7534 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7535     struct nfsdevice **dsp)
7536 {
7537 	struct nameidata nd;
7538 	struct nfsdevice *ds;
7539 	struct mount *mp;
7540 	int error, i;
7541 	char *cp, *dsdirpath, *endcp;
7542 	size_t dsdirsize;
7543 	u_quad_t stripesiz;
7544 
7545 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7546 	*dsp = NULL;
7547 	if (jailed(p->td_ucred)) {
7548 		printf("A pNFS nfsd cannot run in a jail\n");
7549 		return (EPERM);
7550 	}
7551 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7552 	    dspathp);
7553 	error = namei(&nd);
7554 	NFSD_DEBUG(4, "lookup=%d\n", error);
7555 	if (error != 0)
7556 		return (error);
7557 	NDFREE_PNBUF(&nd);
7558 	if (nd.ni_vp->v_type != VDIR) {
7559 		vput(nd.ni_vp);
7560 		NFSD_DEBUG(4, "dspath not dir\n");
7561 		return (ENOTDIR);
7562 	}
7563 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7564 		vput(nd.ni_vp);
7565 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7566 		return (ENXIO);
7567 	}
7568 
7569 	/*
7570 	 * Allocate a DS server structure with the NFS mounted directory
7571 	 * vnode reference counted, so that a non-forced dismount will
7572 	 * fail with EBUSY.
7573 	 * This structure is always linked into the list, even if an error
7574 	 * is being returned.  The caller will free the entire list upon
7575 	 * an error return.
7576 	 */
7577 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7578 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7579 	ds->nfsdev_dvp = nd.ni_vp;
7580 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7581 	ds->nfsdev_mdsstripesiz = nfsrv_stripesiz;
7582 	NFSVOPUNLOCK(nd.ni_vp);
7583 
7584 	dsdirsize = strlen(dspathp) + 16;
7585 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7586 	/* Now, create the DS directory structures. */
7587 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7588 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7589 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7590 		    UIO_SYSSPACE, dsdirpath);
7591 		error = namei(&nd);
7592 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7593 		if (error != 0)
7594 			break;
7595 		NDFREE_PNBUF(&nd);
7596 		if (nd.ni_vp->v_type != VDIR) {
7597 			vput(nd.ni_vp);
7598 			error = ENOTDIR;
7599 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7600 			break;
7601 		}
7602 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7603 			vput(nd.ni_vp);
7604 			error = ENXIO;
7605 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7606 			break;
7607 		}
7608 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7609 		NFSVOPUNLOCK(nd.ni_vp);
7610 	}
7611 	free(dsdirpath, M_TEMP);
7612 
7613 	if (strlen(mdspathp) > 0) {
7614 		cp = strchr(mdspathp, '@');
7615 		if (cp != NULL)
7616 			*cp = '\0';
7617 		/*
7618 		 * This DS stores file for a specific MDS exported file
7619 		 * system.
7620 		 */
7621 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7622 		    UIO_SYSSPACE, mdspathp);
7623 		error = namei(&nd);
7624 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7625 		if (error != 0)
7626 			goto out;
7627 		NDFREE_PNBUF(&nd);
7628 		if (nd.ni_vp->v_type != VDIR) {
7629 			vput(nd.ni_vp);
7630 			error = ENOTDIR;
7631 			NFSD_DEBUG(4, "mdspath not dir\n");
7632 			goto out;
7633 		}
7634 		mp = nd.ni_vp->v_mount;
7635 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7636 			vput(nd.ni_vp);
7637 			error = ENXIO;
7638 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7639 			goto out;
7640 		}
7641 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7642 		ds->nfsdev_mdsisset = 1;
7643 		vput(nd.ni_vp);
7644 		if (cp != NULL) {
7645 			/* There is a stripesiz specified. */
7646 			endcp = NULL;
7647 			if (*(cp + 1) != '\0')
7648 				stripesiz = strtouq(cp + 1, &endcp, 10);
7649 			if (endcp == NULL || *endcp != '\0') {
7650 				error = ENXIO;
7651 				NFSD_DEBUG(4, "mds stripesiz invalid\n");
7652 				goto out;
7653 			}
7654 			ds->nfsdev_mdsstripesiz = stripesiz;
7655 			*cp = '@';
7656 		}
7657 	}
7658 
7659 out:
7660 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7661 	atomic_add_int(&nfsrv_devidcnt, 1);
7662 	return (error);
7663 }
7664 
7665 /*
7666  * Look up the mount path for the DS server and delete it.
7667  */
7668 int
nfsrv_deldsserver(int op,char * dspathp,NFSPROC_T * p)7669 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7670 {
7671 	struct mount *mp;
7672 	struct nfsmount *nmp;
7673 	struct nfsdevice *ds;
7674 	int error;
7675 
7676 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7677 	/*
7678 	 * Search for the path in the mount list.  Avoid looking the path
7679 	 * up, since this mount point may be hung, with associated locked
7680 	 * vnodes, etc.
7681 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7682 	 * until this completes.
7683 	 * As noted in the man page, this should be done before any forced
7684 	 * dismount on the mount point, but at least the handshake on
7685 	 * NFSMNTP_CANCELRPCS should make it safe.
7686 	 */
7687 	error = 0;
7688 	ds = NULL;
7689 	nmp = NULL;
7690 	mtx_lock(&mountlist_mtx);
7691 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7692 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7693 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7694 		    mp->mnt_data != NULL) {
7695 			nmp = VFSTONFS(mp);
7696 			NFSLOCKMNT(nmp);
7697 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7698 			     NFSMNTP_CANCELRPCS)) == 0) {
7699 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7700 				NFSUNLOCKMNT(nmp);
7701 			} else {
7702 				NFSUNLOCKMNT(nmp);
7703 				nmp = NULL;
7704 			}
7705 			break;
7706 		}
7707 	}
7708 	mtx_unlock(&mountlist_mtx);
7709 
7710 	if (nmp != NULL) {
7711 		ds = nfsrv_deldsnmp(op, nmp, p);
7712 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7713 		if (ds != NULL) {
7714 			nfsrv_killrpcs(nmp);
7715 			NFSD_DEBUG(4, "aft killrpcs\n");
7716 		} else
7717 			error = ENXIO;
7718 		NFSLOCKMNT(nmp);
7719 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7720 		wakeup(nmp);
7721 		NFSUNLOCKMNT(nmp);
7722 	} else
7723 		error = EINVAL;
7724 	return (error);
7725 }
7726 
7727 /*
7728  * Search for and remove a DS entry which matches the "nmp" argument.
7729  * The nfsdevice structure pointer is returned so that the caller can
7730  * free it via nfsrv_freeonedevid().
7731  * For the forced case, do not try to do LayoutRecalls, since the server
7732  * must be shut down now anyhow.
7733  */
7734 struct nfsdevice *
nfsrv_deldsnmp(int op,struct nfsmount * nmp,NFSPROC_T * p)7735 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7736 {
7737 	struct nfsdevice *fndds;
7738 
7739 	NFSD_DEBUG(4, "deldsdvp\n");
7740 	NFSDDSLOCK();
7741 	if (op == PNFSDOP_FORCEDELDS)
7742 		fndds = nfsv4_findmirror(nmp);
7743 	else
7744 		fndds = nfsrv_findmirroredds(nmp);
7745 	if (fndds != NULL)
7746 		nfsrv_deleteds(fndds);
7747 	NFSDDSUNLOCK();
7748 	if (fndds != NULL) {
7749 		if (op != PNFSDOP_FORCEDELDS)
7750 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7751 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7752 	}
7753 	return (fndds);
7754 }
7755 
7756 /*
7757  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7758  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7759  * point.
7760  * Also, returns an error instead of the nfsdevice found.
7761  */
7762 int
nfsrv_delds(char * devid,NFSPROC_T * p)7763 nfsrv_delds(char *devid, NFSPROC_T *p)
7764 {
7765 	struct nfsdevice *ds, *fndds;
7766 	struct nfsmount *nmp;
7767 	int fndmirror;
7768 
7769 	NFSD_DEBUG(4, "delds\n");
7770 	/*
7771 	 * Search the DS server list for a match with devid.
7772 	 * Remove the DS entry if found and there is a mirror.
7773 	 */
7774 	fndds = NULL;
7775 	nmp = NULL;
7776 	fndmirror = 0;
7777 	NFSDDSLOCK();
7778 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7779 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7780 		    ds->nfsdev_nmp != NULL) {
7781 			NFSD_DEBUG(4, "fnd main ds\n");
7782 			fndds = ds;
7783 			break;
7784 		}
7785 	}
7786 	if (fndds == NULL) {
7787 		NFSDDSUNLOCK();
7788 		return (ENXIO);
7789 	}
7790 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7791 		fndmirror = 1;
7792 	else if (fndds->nfsdev_mdsisset != 0) {
7793 		/* For the fsid is set case, search for a mirror. */
7794 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7795 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7796 			    ds->nfsdev_mdsisset != 0 &&
7797 			    fsidcmp(&ds->nfsdev_mdsfsid,
7798 			    &fndds->nfsdev_mdsfsid) == 0) {
7799 				fndmirror = 1;
7800 				break;
7801 			}
7802 		}
7803 	}
7804 	if (fndmirror != 0) {
7805 		nmp = fndds->nfsdev_nmp;
7806 		NFSLOCKMNT(nmp);
7807 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7808 		     NFSMNTP_CANCELRPCS)) == 0) {
7809 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7810 			NFSUNLOCKMNT(nmp);
7811 			nfsrv_deleteds(fndds);
7812 		} else {
7813 			NFSUNLOCKMNT(nmp);
7814 			nmp = NULL;
7815 		}
7816 	}
7817 	NFSDDSUNLOCK();
7818 	if (nmp != NULL) {
7819 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7820 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7821 		nfsrv_killrpcs(nmp);
7822 		NFSLOCKMNT(nmp);
7823 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7824 		wakeup(nmp);
7825 		NFSUNLOCKMNT(nmp);
7826 		return (0);
7827 	}
7828 	return (ENXIO);
7829 }
7830 
7831 /*
7832  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7833  */
7834 static void
nfsrv_deleteds(struct nfsdevice * fndds)7835 nfsrv_deleteds(struct nfsdevice *fndds)
7836 {
7837 
7838 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7839 	fndds->nfsdev_nmp = NULL;
7840 	if (fndds->nfsdev_mdsisset == 0)
7841 		nfsrv_faildscnt--;
7842 }
7843 
7844 /*
7845  * Fill in the addr structures for the File and Flex File layouts.
7846  */
7847 static void
nfsrv_allocdevid(struct nfsdevice * ds,char * addr,char * dnshost)7848 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
7849 {
7850 	uint32_t *tl;
7851 	char *netprot;
7852 	int addrlen;
7853 	static uint64_t new_devid = 0;
7854 
7855 	if (strchr(addr, ':') != NULL)
7856 		netprot = "tcp6";
7857 	else
7858 		netprot = "tcp";
7859 
7860 	/* Fill in the device id. */
7861 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
7862 	new_devid++;
7863 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
7864 	    sizeof(new_devid));
7865 
7866 	/*
7867 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
7868 	 * as defined in RFC5661) in XDR.
7869 	 */
7870 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7871 	    6 * NFSX_UNSIGNED;
7872 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
7873 	ds->nfsdev_fileaddrlen = addrlen;
7874 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7875 	ds->nfsdev_fileaddr = (char *)tl;
7876 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
7877 	*tl++ = 0;
7878 	*tl++ = txdr_unsigned(1);		/* One multipath list */
7879 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
7880 	/* The netaddr for this one entry. */
7881 	*tl++ = txdr_unsigned(strlen(netprot));
7882 	NFSBCOPY(netprot, tl, strlen(netprot));
7883 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7884 	*tl++ = txdr_unsigned(strlen(addr));
7885 	NFSBCOPY(addr, tl, strlen(addr));
7886 
7887 	/*
7888 	 * Fill in the flex file addr (actually the ff_device_addr4
7889 	 * as defined for Flexible File Layout) in XDR.
7890 	 */
7891 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7892 	    14 * NFSX_UNSIGNED;
7893 	ds->nfsdev_flexaddrlen = addrlen;
7894 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7895 	ds->nfsdev_flexaddr = (char *)tl;
7896 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
7897 	/* The netaddr for this one entry. */
7898 	*tl++ = txdr_unsigned(strlen(netprot));
7899 	NFSBCOPY(netprot, tl, strlen(netprot));
7900 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7901 	*tl++ = txdr_unsigned(strlen(addr));
7902 	NFSBCOPY(addr, tl, strlen(addr));
7903 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
7904 	*tl++ = txdr_unsigned(2);		/* Two NFS Versions. */
7905 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
7906 	*tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
7907 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
7908 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
7909 	*tl++ = newnfs_true;			/* Tightly coupled. */
7910 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
7911 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
7912 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
7913 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
7914 	*tl = newnfs_true;			/* Tightly coupled. */
7915 
7916 	ds->nfsdev_hostnamelen = strlen(dnshost);
7917 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
7918 	    M_WAITOK);
7919 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
7920 }
7921 
7922 /*
7923  * Create the device id list.
7924  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
7925  * is misconfigured.
7926  */
7927 int
nfsrv_createdevids(struct nfsd_nfsd_args * args,NFSPROC_T * p)7928 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
7929 {
7930 	struct nfsdevice *ds;
7931 	char *addrp, *dnshostp, *dspathp, *mdspathp;
7932 	int error, i;
7933 
7934 	addrp = args->addr;
7935 	dnshostp = args->dnshost;
7936 	dspathp = args->dspath;
7937 	mdspathp = args->mdspath;
7938 	nfsrv_maxpnfsmirror = args->mirrorcnt;
7939 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
7940 	    mdspathp == NULL)
7941 		return (0);
7942 
7943 	/*
7944 	 * Loop around for each nul-terminated string in args->addr,
7945 	 * args->dnshost, args->dnspath and args->mdspath.
7946 	 */
7947 	while (addrp < (args->addr + args->addrlen) &&
7948 	    dnshostp < (args->dnshost + args->dnshostlen) &&
7949 	    dspathp < (args->dspath + args->dspathlen) &&
7950 	    mdspathp < (args->mdspath + args->mdspathlen)) {
7951 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
7952 		if (error != 0) {
7953 			/* Free all DS servers. */
7954 			nfsrv_freealldevids();
7955 			nfsrv_devidcnt = 0;
7956 			return (ENXIO);
7957 		}
7958 		nfsrv_allocdevid(ds, addrp, dnshostp);
7959 		addrp += (strlen(addrp) + 1);
7960 		dnshostp += (strlen(dnshostp) + 1);
7961 		dspathp += (strlen(dspathp) + 1);
7962 		mdspathp += (strlen(mdspathp) + 1);
7963 	}
7964 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
7965 		/* Free all DS servers. */
7966 		nfsrv_freealldevids();
7967 		nfsrv_devidcnt = 0;
7968 		nfsrv_maxpnfsmirror = 1;
7969 		return (ENXIO);
7970 	}
7971 	/* We can fail at most one less DS than the mirror level. */
7972 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
7973 
7974 	/*
7975 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
7976 	 * Make it 1% of the high water mark and at least 100.
7977 	 */
7978 	if (nfslayouthash == NULL) {
7979 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
7980 		if (nfsrv_layouthashsize < 100)
7981 			nfsrv_layouthashsize = 100;
7982 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
7983 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
7984 		    M_ZERO);
7985 		for (i = 0; i < nfsrv_layouthashsize; i++) {
7986 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
7987 			TAILQ_INIT(&nfslayouthash[i].list);
7988 		}
7989 	}
7990 	return (0);
7991 }
7992 
7993 /*
7994  * Free all device ids.
7995  */
7996 static void
nfsrv_freealldevids(void)7997 nfsrv_freealldevids(void)
7998 {
7999 	struct nfsdevice *ds, *nds;
8000 
8001 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8002 		nfsrv_freedevid(ds);
8003 }
8004 
8005 /*
8006  * Check to see if there is a Read/Write Layout plus either:
8007  * - A Write Delegation
8008  * or
8009  * - An Open with Write_access.
8010  * Return 1 if this is the case and 0 otherwise.
8011  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8012  * Getattr RPC to the Data Server (DS) is necessary.
8013  */
8014 #define	NFSCLIDVECSIZE	6
8015 int
nfsrv_checkdsattr(vnode_t vp,NFSPROC_T * p)8016 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8017 {
8018 	fhandle_t fh, *tfhp;
8019 	struct nfsstate *stp;
8020 	struct nfslayout *lyp;
8021 	struct nfslayouthash *lhyp;
8022 	struct nfslockhashhead *hp;
8023 	struct nfslockfile *lfp;
8024 	nfsquad_t clid[NFSCLIDVECSIZE];
8025 	int clidcnt, ret;
8026 
8027 	ret = nfsvno_getfh(vp, &fh, p);
8028 	if (ret != 0)
8029 		return (0);
8030 
8031 	/* First check for a Read/Write Layout. */
8032 	clidcnt = 0;
8033 	lhyp = NFSLAYOUTHASH(&fh);
8034 	NFSLOCKLAYOUT(lhyp);
8035 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8036 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8037 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8038 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8039 		      nfsrv_pnfsatime != 0))) {
8040 			if (clidcnt < NFSCLIDVECSIZE)
8041 				clid[clidcnt].qval = lyp->lay_clientid.qval;
8042 			clidcnt++;
8043 		}
8044 	}
8045 	NFSUNLOCKLAYOUT(lhyp);
8046 	if (clidcnt == 0) {
8047 		/* None found, so return 0. */
8048 		return (0);
8049 	}
8050 
8051 	/* Get the nfslockfile for this fh. */
8052 	NFSLOCKSTATE();
8053 	hp = NFSLOCKHASH(&fh);
8054 	LIST_FOREACH(lfp, hp, lf_hash) {
8055 		tfhp = &lfp->lf_fh;
8056 		if (NFSVNO_CMPFH(&fh, tfhp))
8057 			break;
8058 	}
8059 	if (lfp == NULL) {
8060 		/* None found, so return 0. */
8061 		NFSUNLOCKSTATE();
8062 		return (0);
8063 	}
8064 
8065 	/* Now, look for a Write delegation for this clientid. */
8066 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8067 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8068 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8069 			break;
8070 	}
8071 	if (stp != NULL) {
8072 		/* Found one, so return 1. */
8073 		NFSUNLOCKSTATE();
8074 		return (1);
8075 	}
8076 
8077 	/* No Write delegation, so look for an Open with Write_access. */
8078 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8079 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8080 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
8081 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8082 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8083 			break;
8084 	}
8085 	NFSUNLOCKSTATE();
8086 	if (stp != NULL)
8087 		return (1);
8088 	return (0);
8089 }
8090 
8091 /*
8092  * Look for a matching clientid in the vector. Return 1 if one might match.
8093  */
8094 static int
nfsrv_fndclid(nfsquad_t * clidvec,nfsquad_t clid,int clidcnt)8095 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8096 {
8097 	int i;
8098 
8099 	/* If too many for the vector, return 1 since there might be a match. */
8100 	if (clidcnt > NFSCLIDVECSIZE)
8101 		return (1);
8102 
8103 	for (i = 0; i < clidcnt; i++)
8104 		if (clidvec[i].qval == clid.qval)
8105 			return (1);
8106 	return (0);
8107 }
8108 
8109 /*
8110  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8111  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8112  */
8113 static int
nfsrv_dontlayout(fhandle_t * fhp)8114 nfsrv_dontlayout(fhandle_t *fhp)
8115 {
8116 	struct nfsdontlist *mrp;
8117 	int ret;
8118 
8119 	if (nfsrv_dontlistlen == 0)
8120 		return (0);
8121 	ret = 0;
8122 	NFSDDONTLISTLOCK();
8123 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8124 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8125 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8126 			ret = 1;
8127 			break;
8128 		}
8129 	}
8130 	NFSDDONTLISTUNLOCK();
8131 	return (ret);
8132 }
8133 
8134 #define	PNFSDS_COPYSIZ	65536
8135 /*
8136  * Create a new file on a DS and copy the contents of an extant DS file to it.
8137  * This can be used for recovery of a DS file onto a recovered DS.
8138  * The steps are:
8139  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8140  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8141  *   so that they will be disabled after the MDS file's vnode is unlocked.
8142  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8143  *   be done.
8144  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8145  *   writes, LayoutCommits and LayoutReturns for the file when completing the
8146  *   LayoutReturn requested by the LayoutRecall callback.
8147  * - Issue a LayoutRecall callback for all read/write layouts and wait for
8148  *   them to be returned. (If the LayoutRecall callback replies
8149  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8150  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
8151  *   writes are in progress or can occur during the DS file copy.
8152  *   It also blocks Setattr operations.
8153  * - Create the file on the recovered mirror.
8154  * - Copy the file from the operational DS.
8155  * - Copy any ACL from the MDS file to the new DS file.
8156  * - Set the modify time of the new DS file to that of the MDS file.
8157  * - Update the extended attribute for the MDS file.
8158  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8159  * - The caller will unlock the MDS file's vnode allowing operations
8160  *   to continue normally, since it is now on the mirror again.
8161  */
8162 int
nfsrv_copymr(vnode_t vp,vnode_t fvp,vnode_t dvp,struct nfsdevice * ds,struct pnfsdsfile * pf,struct pnfsdsfile * wpf,int mirrorcnt,struct ucred * cred,NFSPROC_T * p)8163 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8164     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8165     struct ucred *cred, NFSPROC_T *p)
8166 {
8167 	struct nfsdontlist *mrp, *nmrp;
8168 	struct nfslayouthash *lhyp;
8169 	struct nfslayout *lyp, *nlyp;
8170 	struct nfslayouthead thl;
8171 	struct mount *mp, *tvmp;
8172 	struct acl *aclp;
8173 	struct vattr va;
8174 	struct timespec mtime;
8175 	fhandle_t fh;
8176 	vnode_t tvp;
8177 	off_t rdpos, wrpos;
8178 	ssize_t aresid;
8179 	char *dat;
8180 	int didprintf, ret, retacl, xfer;
8181 
8182 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8183 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8184 	/*
8185 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8186 	 * so that no more RW layouts will get issued.
8187 	 */
8188 	ret = nfsvno_getfh(vp, &fh, p);
8189 	if (ret != 0) {
8190 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8191 		return (ret);
8192 	}
8193 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8194 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8195 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8196 	NFSDDONTLISTLOCK();
8197 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8198 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8199 			break;
8200 	}
8201 	if (mrp == NULL) {
8202 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8203 		mrp = nmrp;
8204 		nmrp = NULL;
8205 		nfsrv_dontlistlen++;
8206 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8207 	} else {
8208 		NFSDDONTLISTUNLOCK();
8209 		free(nmrp, M_NFSDSTATE);
8210 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8211 		return (ENXIO);
8212 	}
8213 	NFSDDONTLISTUNLOCK();
8214 
8215 	/*
8216 	 * Search for all RW layouts for this file.  Move them to the
8217 	 * recall list, so they can be recalled and their return noted.
8218 	 */
8219 	lhyp = NFSLAYOUTHASH(&fh);
8220 	NFSDRECALLLOCK();
8221 	NFSLOCKLAYOUT(lhyp);
8222 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8223 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8224 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8225 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8226 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8227 			lyp->lay_trycnt = 0;
8228 		}
8229 	}
8230 	NFSUNLOCKLAYOUT(lhyp);
8231 	NFSDRECALLUNLOCK();
8232 
8233 	ret = 0;
8234 	mp = tvmp = NULL;
8235 	didprintf = 0;
8236 	TAILQ_INIT(&thl);
8237 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8238 	NFSVOPUNLOCK(vp);
8239 	/* Now, do a recall for all layouts not yet recalled. */
8240 tryagain:
8241 	NFSDRECALLLOCK();
8242 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8243 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8244 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8245 			lyp->lay_flags |= NFSLAY_RECALL;
8246 			/*
8247 			 * The layout stateid.seqid needs to be incremented
8248 			 * before doing a LAYOUT_RECALL callback.
8249 			 */
8250 			if (++lyp->lay_stateid.seqid == 0)
8251 				lyp->lay_stateid.seqid = 1;
8252 			NFSDRECALLUNLOCK();
8253 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8254 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8255 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8256 			goto tryagain;
8257 		}
8258 	}
8259 
8260 	/* Now wait for them to be returned. */
8261 tryagain2:
8262 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8263 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8264 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8265 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8266 				    lay_list);
8267 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8268 				NFSD_DEBUG(4,
8269 				    "nfsrv_copymr: layout returned\n");
8270 			} else {
8271 				lyp->lay_trycnt++;
8272 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8273 				    PVFS | PCATCH, "nfsmrl", hz);
8274 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8275 				    ret);
8276 				if (ret == EINTR || ret == ERESTART)
8277 					break;
8278 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8279 					/*
8280 					 * Give up after 60sec and return
8281 					 * ENXIO, failing the copymr.
8282 					 * This layout will remain on the
8283 					 * recalllist.  It can only be cleared
8284 					 * by restarting the nfsd.
8285 					 * This seems the safe way to handle
8286 					 * it, since it cannot be safely copied
8287 					 * with an outstanding RW layout.
8288 					 */
8289 					if (lyp->lay_trycnt >= 60) {
8290 						ret = ENXIO;
8291 						break;
8292 					}
8293 					if (didprintf == 0) {
8294 						printf("nfsrv_copymr: layout "
8295 						    "not returned\n");
8296 						didprintf = 1;
8297 					}
8298 				}
8299 			}
8300 			goto tryagain2;
8301 		}
8302 	}
8303 	NFSDRECALLUNLOCK();
8304 	/* We can now get rid of the layouts that have been returned. */
8305 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8306 		nfsrv_freelayout(&thl, lyp);
8307 
8308 	/*
8309 	 * Do the vn_start_write() calls here, before the MDS vnode is
8310 	 * locked and the tvp is created (locked) in the NFS file system
8311 	 * that dvp is in.
8312 	 * For tvmp, this probably isn't necessary, since it will be an
8313 	 * NFS mount and they are not suspendable at this time.
8314 	 */
8315 	if (ret == 0)
8316 		ret = vn_start_write(vp, &mp, V_WAIT | V_PCATCH);
8317 	if (ret == 0) {
8318 		tvmp = dvp->v_mount;
8319 		ret = vn_start_write(NULL, &tvmp, V_WAIT | V_PCATCH);
8320 	}
8321 
8322 	/*
8323 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8324 	 * proxied writes through the MDS will be blocked until we have
8325 	 * completed the copy and update of the extended attributes.
8326 	 * This will also ensure that any attributes and ACL will not be
8327 	 * changed until the copy is complete.
8328 	 */
8329 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8330 	if (ret == 0 && VN_IS_DOOMED(vp)) {
8331 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8332 		ret = ESTALE;
8333 	}
8334 
8335 	/* Create the data file on the recovered DS. */
8336 	if (ret == 0)
8337 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8338 
8339 	/* Copy the DS file, if created successfully. */
8340 	if (ret == 0) {
8341 		/*
8342 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8343 		 * on the new DS file.
8344 		 */
8345 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8346 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8347 		if (retacl != 0 && retacl != ENOATTR)
8348 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8349 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8350 		/* Malloc a block of 0s used to check for holes. */
8351 		if (nfsrv_zeropnfsdat == NULL)
8352 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8353 			    M_WAITOK | M_ZERO);
8354 		rdpos = wrpos = 0;
8355 		ret = VOP_GETATTR(fvp, &va, cred);
8356 		aresid = 0;
8357 		while (ret == 0 && aresid == 0) {
8358 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8359 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8360 			    &aresid, p);
8361 			xfer = PNFSDS_COPYSIZ - aresid;
8362 			if (ret == 0 && xfer > 0) {
8363 				rdpos += xfer;
8364 				/*
8365 				 * Skip the write for holes, except for the
8366 				 * last block.
8367 				 */
8368 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8369 				    va.va_size || NFSBCMP(dat,
8370 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8371 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8372 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8373 					    cred, NULL, NULL, p);
8374 				if (ret == 0)
8375 					wrpos += xfer;
8376 			}
8377 		}
8378 
8379 		/* If there is an ACL and the copy succeeded, set the ACL. */
8380 		if (ret == 0 && retacl == 0) {
8381 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8382 			/*
8383 			 * Don't consider these as errors, since VOP_GETACL()
8384 			 * can return an ACL when they are not actually
8385 			 * supported.  For example, for UFS, VOP_GETACL()
8386 			 * will return a trivial ACL based on the uid/gid/mode
8387 			 * when there is no ACL on the file.
8388 			 * This case should be recognized as a trivial ACL
8389 			 * by UFS's VOP_SETACL() and succeed, but...
8390 			 */
8391 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8392 				ret = 0;
8393 		}
8394 
8395 		if (ret == 0)
8396 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8397 
8398 		/* Set the DS data file's modify time that of the MDS file. */
8399 		if (ret == 0)
8400 			ret = VOP_GETATTR(vp, &va, cred);
8401 		if (ret == 0) {
8402 			mtime = va.va_mtime;
8403 			VATTR_NULL(&va);
8404 			va.va_mtime = mtime;
8405 			ret = VOP_SETATTR(tvp, &va, cred);
8406 		}
8407 
8408 		vput(tvp);
8409 		acl_free(aclp);
8410 		free(dat, M_TEMP);
8411 	}
8412 	if (tvmp != NULL)
8413 		vn_finished_write(tvmp);
8414 
8415 	/* Update the extended attributes for the newly created DS file. */
8416 	if (ret == 0)
8417 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8418 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8419 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8420 	if (mp != NULL)
8421 		vn_finished_write(mp);
8422 
8423 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8424 	NFSDDONTLISTLOCK();
8425 	LIST_REMOVE(mrp, nfsmr_list);
8426 	NFSDDONTLISTUNLOCK();
8427 	free(mrp, M_NFSDSTATE);
8428 	return (ret);
8429 }
8430 
8431 /*
8432  * Create a data storage file on the recovered DS.
8433  */
8434 static int
nfsrv_createdsfile(vnode_t vp,fhandle_t * fhp,struct pnfsdsfile * pf,vnode_t dvp,struct nfsdevice * ds,struct ucred * cred,NFSPROC_T * p,vnode_t * tvpp)8435 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8436     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8437     vnode_t *tvpp)
8438 {
8439 	struct vattr va, nva;
8440 	int error;
8441 
8442 	/* Make data file name based on FH. */
8443 	error = VOP_GETATTR(vp, &va, cred);
8444 	if (error == 0) {
8445 		/* Set the attributes for "vp" to Setattr the DS vp. */
8446 		VATTR_NULL(&nva);
8447 		nva.va_uid = va.va_uid;
8448 		nva.va_gid = va.va_gid;
8449 		nva.va_mode = va.va_mode;
8450 		nva.va_size = 0;
8451 		VATTR_NULL(&va);
8452 		va.va_type = VREG;
8453 		va.va_mode = nva.va_mode;
8454 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8455 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8456 		    pf->dsf_filename, cred, p, tvpp);
8457 	}
8458 	return (error);
8459 }
8460 
8461 /*
8462  * Look up the MDS file shared locked, and then get the extended attribute
8463  * to find the extant DS file to be copied to the new mirror.
8464  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8465  * set to a DS data file for the MDS file, both exclusively locked.
8466  * The "buf" argument has the pnfsdsfile structure from the MDS file
8467  * in it and buflen is set to its length.
8468  */
8469 int
nfsrv_mdscopymr(char * mdspathp,char * dspathp,char * curdspathp,char * buf,int * buflenp,char * fname,NFSPROC_T * p,struct vnode ** vpp,struct vnode ** nvpp,struct pnfsdsfile ** pfp,struct nfsdevice ** dsp,struct nfsdevice ** fdsp)8470 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8471     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8472     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8473     struct nfsdevice **fdsp)
8474 {
8475 	struct nameidata nd;
8476 	struct vnode *vp, *curvp;
8477 	struct pnfsdsfile *pf;
8478 	struct nfsmount *nmp, *curnmp;
8479 	int dsdir, error, ippos;
8480 
8481 	vp = NULL;
8482 	curvp = NULL;
8483 	curnmp = NULL;
8484 	*dsp = NULL;
8485 	*fdsp = NULL;
8486 	if (dspathp == NULL && curdspathp != NULL)
8487 		return (EPERM);
8488 
8489 	/*
8490 	 * Look up the MDS file shared locked.  The lock will be upgraded
8491 	 * to an exclusive lock after any rw layouts have been returned.
8492 	 */
8493 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8494 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8495 	    mdspathp);
8496 	error = namei(&nd);
8497 	NFSD_DEBUG(4, "lookup=%d\n", error);
8498 	if (error != 0)
8499 		return (error);
8500 	NDFREE_PNBUF(&nd);
8501 	if (nd.ni_vp->v_type != VREG) {
8502 		vput(nd.ni_vp);
8503 		NFSD_DEBUG(4, "mdspath not reg\n");
8504 		return (EISDIR);
8505 	}
8506 	vp = nd.ni_vp;
8507 
8508 	if (curdspathp != NULL) {
8509 		/*
8510 		 * Look up the current DS path and find the nfsdev structure for
8511 		 * it.
8512 		 */
8513 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8514 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8515 		    UIO_SYSSPACE, curdspathp);
8516 		error = namei(&nd);
8517 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8518 		if (error != 0) {
8519 			vput(vp);
8520 			return (error);
8521 		}
8522 		NDFREE_PNBUF(&nd);
8523 		if (nd.ni_vp->v_type != VDIR) {
8524 			vput(nd.ni_vp);
8525 			vput(vp);
8526 			NFSD_DEBUG(4, "curdspath not dir\n");
8527 			return (ENOTDIR);
8528 		}
8529 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8530 			vput(nd.ni_vp);
8531 			vput(vp);
8532 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8533 			return (ENXIO);
8534 		}
8535 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8536 
8537 		/* Search the nfsdev list for a match. */
8538 		NFSDDSLOCK();
8539 		*fdsp = nfsv4_findmirror(curnmp);
8540 		NFSDDSUNLOCK();
8541 		if (*fdsp == NULL)
8542 			curnmp = NULL;
8543 		if (curnmp == NULL) {
8544 			vput(nd.ni_vp);
8545 			vput(vp);
8546 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8547 			return (ENXIO);
8548 		}
8549 		curvp = nd.ni_vp;
8550 	}
8551 
8552 	if (dspathp != NULL) {
8553 		/* Look up the nfsdev path and find the nfsdev structure. */
8554 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8555 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8556 		    UIO_SYSSPACE, dspathp);
8557 		error = namei(&nd);
8558 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8559 		if (error != 0) {
8560 			vput(vp);
8561 			if (curvp != NULL)
8562 				vput(curvp);
8563 			return (error);
8564 		}
8565 		NDFREE_PNBUF(&nd);
8566 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8567 			vput(nd.ni_vp);
8568 			vput(vp);
8569 			if (curvp != NULL)
8570 				vput(curvp);
8571 			NFSD_DEBUG(4, "dspath not dir\n");
8572 			if (nd.ni_vp == curvp)
8573 				return (EPERM);
8574 			return (ENOTDIR);
8575 		}
8576 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8577 			vput(nd.ni_vp);
8578 			vput(vp);
8579 			if (curvp != NULL)
8580 				vput(curvp);
8581 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8582 			return (ENXIO);
8583 		}
8584 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8585 
8586 		/*
8587 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8588 		 * this is a recovery and there must be a mirror.
8589 		 */
8590 		NFSDDSLOCK();
8591 		if (curnmp == NULL)
8592 			*dsp = nfsrv_findmirroredds(nmp);
8593 		else
8594 			*dsp = nfsv4_findmirror(nmp);
8595 		NFSDDSUNLOCK();
8596 		if (*dsp == NULL) {
8597 			vput(nd.ni_vp);
8598 			vput(vp);
8599 			if (curvp != NULL)
8600 				vput(curvp);
8601 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8602 			return (ENXIO);
8603 		}
8604 	} else {
8605 		nd.ni_vp = NULL;
8606 		nmp = NULL;
8607 	}
8608 
8609 	/*
8610 	 * Get a vp for an available DS data file using the extended
8611 	 * attribute on the MDS file.
8612 	 * If there is a valid entry for the new DS in the extended attribute
8613 	 * on the MDS file (as checked via the nmp argument),
8614 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8615 	 */
8616 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, NULL, NULL, NULL, p,
8617 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8618 	if (curvp != NULL)
8619 		vput(curvp);
8620 	if (nd.ni_vp == NULL) {
8621 		if (error == 0 && nmp != NULL) {
8622 			/* Search the nfsdev list for a match. */
8623 			NFSDDSLOCK();
8624 			*dsp = nfsrv_findmirroredds(nmp);
8625 			NFSDDSUNLOCK();
8626 		}
8627 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8628 			if (nvpp != NULL && *nvpp != NULL) {
8629 				vput(*nvpp);
8630 				*nvpp = NULL;
8631 			}
8632 			error = ENXIO;
8633 		}
8634 	} else
8635 		vput(nd.ni_vp);
8636 
8637 	/*
8638 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8639 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8640 	 * When curdspathp != NULL, the ippos will be set to that entry.
8641 	 */
8642 	if (error == 0 && dspathp != NULL && ippos == -1) {
8643 		if (nvpp != NULL && *nvpp != NULL) {
8644 			vput(*nvpp);
8645 			*nvpp = NULL;
8646 		}
8647 		error = ENXIO;
8648 	}
8649 	if (error == 0) {
8650 		*vpp = vp;
8651 
8652 		pf = (struct pnfsdsfile *)buf;
8653 		if (ippos == -1) {
8654 			/* If no zeroip pnfsdsfile, add one. */
8655 			ippos = *buflenp / sizeof(*pf);
8656 			*buflenp += sizeof(*pf);
8657 			pf += ippos;
8658 			pf->dsf_dir = dsdir;
8659 			strlcpy(pf->dsf_filename, fname,
8660 			    sizeof(pf->dsf_filename));
8661 		} else
8662 			pf += ippos;
8663 		*pfp = pf;
8664 	} else
8665 		vput(vp);
8666 	return (error);
8667 }
8668 
8669 /*
8670  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8671  * Return one if found, NULL otherwise.
8672  */
8673 static struct nfsdevice *
nfsrv_findmirroredds(struct nfsmount * nmp)8674 nfsrv_findmirroredds(struct nfsmount *nmp)
8675 {
8676 	struct nfsdevice *ds, *fndds;
8677 	int fndmirror;
8678 
8679 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8680 	/*
8681 	 * Search the DS server list for a match with nmp.
8682 	 * Remove the DS entry if found and there is a mirror.
8683 	 */
8684 	fndds = NULL;
8685 	fndmirror = 0;
8686 	if (nfsrv_devidcnt == 0)
8687 		return (fndds);
8688 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8689 		if (ds->nfsdev_nmp == nmp) {
8690 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8691 			fndds = ds;
8692 			break;
8693 		}
8694 	}
8695 	if (fndds == NULL)
8696 		return (fndds);
8697 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8698 		fndmirror = 1;
8699 	else if (fndds->nfsdev_mdsisset != 0) {
8700 		/* For the fsid is set case, search for a mirror. */
8701 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8702 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8703 			    ds->nfsdev_mdsisset != 0 &&
8704 			    fsidcmp(&ds->nfsdev_mdsfsid,
8705 			    &fndds->nfsdev_mdsfsid) == 0) {
8706 				fndmirror = 1;
8707 				break;
8708 			}
8709 		}
8710 	}
8711 	if (fndmirror == 0) {
8712 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8713 		return (NULL);
8714 	}
8715 	return (fndds);
8716 }
8717 
8718 /*
8719  * Mark the appropriate devid and all associated layout as "out of space".
8720  */
8721 void
nfsrv_marknospc(char * devid,bool setit)8722 nfsrv_marknospc(char *devid, bool setit)
8723 {
8724 	struct nfsdevice *ds;
8725 	struct nfslayout *lyp;
8726 	struct nfslayouthash *lhyp;
8727 	int i;
8728 
8729 	NFSDDSLOCK();
8730 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8731 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0) {
8732 			NFSD_DEBUG(1, "nfsrv_marknospc: devid %d\n", setit);
8733 			ds->nfsdev_nospc = setit;
8734 		}
8735 	}
8736 	NFSDDSUNLOCK();
8737 
8738 	for (i = 0; i < nfsrv_layouthashsize; i++) {
8739 		lhyp = &nfslayouthash[i];
8740 		NFSLOCKLAYOUT(lhyp);
8741 		TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8742 			if (NFSBCMP(lyp->lay_deviceid, devid,
8743 			    NFSX_V4DEVICEID) == 0) {
8744 				NFSD_DEBUG(1, "nfsrv_marknospc: layout %d\n",
8745 				    setit);
8746 				if (setit)
8747 					lyp->lay_flags |= NFSLAY_NOSPC;
8748 				else
8749 					lyp->lay_flags &= ~NFSLAY_NOSPC;
8750 			}
8751 		}
8752 		NFSUNLOCKLAYOUT(lhyp);
8753 	}
8754 }
8755 
8756 /*
8757  * Check to see if SP4_MACH_CRED is in use and, if it is, check that the
8758  * correct machine credential is being used.
8759  */
8760 static int
nfsrv_checkmachcred(int op,struct nfsrv_descript * nd,struct nfsclient * clp)8761 nfsrv_checkmachcred(int op, struct nfsrv_descript *nd, struct nfsclient *clp)
8762 {
8763 
8764 	if ((clp->lc_flags & LCL_MACHCRED) == 0 ||
8765 	    !NFSISSET_OPBIT(&clp->lc_mustops, op))
8766 		return (0);
8767 	KASSERT((nd->nd_flag & ND_NFSV41) != 0,
8768 	    ("nfsrv_checkmachcred: MachCred for NFSv4.0"));
8769 	if ((nd->nd_flag & (ND_GSSINTEGRITY | ND_GSSPRIVACY)) != 0 &&
8770 	    nd->nd_princlen == clp->lc_namelen &&
8771 	    !NFSBCMP(nd->nd_principal, clp->lc_name, nd->nd_princlen))
8772 		return (0);
8773 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
8774 }
8775 
8776 /*
8777  * Issue a delegation and, optionally set rflagsp for why not.
8778  */
8779 static void
nfsrv_issuedelegation(struct vnode * vp,struct nfsclient * clp,struct nfsrv_descript * nd,int delegate,int writedeleg,int readonly,u_quad_t filerev,uint64_t rdonly,struct nfsstate ** new_delegp,struct nfsstate * new_stp,struct nfslockfile * lfp,uint32_t * rflagsp,nfsv4stateid_t * delegstateidp)8780 nfsrv_issuedelegation(struct vnode *vp, struct nfsclient *clp,
8781     struct nfsrv_descript *nd, int delegate, int writedeleg, int readonly,
8782     u_quad_t filerev, uint64_t rdonly, struct nfsstate **new_delegp,
8783     struct nfsstate *new_stp, struct nfslockfile *lfp, uint32_t *rflagsp,
8784     nfsv4stateid_t *delegstateidp)
8785 {
8786 	struct nfsstate *up_deleg, *new_deleg;
8787 
8788 	new_deleg = *new_delegp;
8789 	up_deleg = LIST_FIRST(&lfp->lf_deleg);
8790 	if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
8791 		*rflagsp |= NFSV4OPEN_WDNOTWANTED;
8792 	else if (nfsrv_issuedelegs == 0)
8793 		*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
8794 	else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
8795 		*rflagsp |= NFSV4OPEN_WDRESOURCE;
8796 	else if (delegate == 0 || !NFSVNO_DELEGOK(vp) ||
8797 	    (writedeleg == 0 && (readonly == 0 ||
8798 	    (new_stp->ls_flags & NFSLCK_WANTWDELEG) != 0)) ||
8799 	    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
8800 	     LCL_CALLBACKSON) {
8801 		/* Is this a downgrade attempt? */
8802 		if (up_deleg != NULL && up_deleg->ls_clp == clp &&
8803 		    (up_deleg->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8804 		    (new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0)
8805 			*rflagsp |= NFSV4OPEN_WDNOTSUPPDOWNGRADE;
8806 		else
8807 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
8808 	} else if (up_deleg != NULL &&
8809 	    (up_deleg->ls_flags & NFSLCK_DELEGREAD) != 0 &&
8810 	    (new_stp->ls_flags & NFSLCK_WANTWDELEG) != 0) {
8811 		/* This is an atomic upgrade. */
8812 		up_deleg->ls_stateid.seqid++;
8813 		delegstateidp->seqid = up_deleg->ls_stateid.seqid;
8814 		delegstateidp->other[0] = up_deleg->ls_stateid.other[0];
8815 		delegstateidp->other[1] = up_deleg->ls_stateid.other[1];
8816 		delegstateidp->other[2] = up_deleg->ls_stateid.other[2];
8817 		up_deleg->ls_flags = (NFSLCK_DELEGWRITE |
8818 		    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
8819 		*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
8820 		nfsrv_writedelegcnt++;
8821 	} else {
8822 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
8823 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
8824 		    = clp->lc_clientid.lval[0];
8825 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
8826 		    = clp->lc_clientid.lval[1];
8827 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
8828 		    = nfsrv_nextstateindex(clp);
8829 		if (writedeleg && !rdonly &&
8830 		    (nfsrv_writedelegifpos || !readonly) &&
8831 		    (new_stp->ls_flags & (NFSLCK_WANTRDELEG |
8832 		     NFSLCK_WANTWDELEG)) != NFSLCK_WANTRDELEG) {
8833 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
8834 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
8835 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
8836 			nfsrv_writedelegcnt++;
8837 		} else {
8838 			new_deleg->ls_flags = (NFSLCK_DELEGREAD |
8839 			    NFSLCK_READACCESS);
8840 			*rflagsp |= NFSV4OPEN_READDELEGATE;
8841 		}
8842 		new_deleg->ls_uid = new_stp->ls_uid;
8843 		new_deleg->ls_lfp = lfp;
8844 		new_deleg->ls_clp = clp;
8845 		new_deleg->ls_filerev = filerev;
8846 		new_deleg->ls_compref = nd->nd_compref;
8847 		new_deleg->ls_lastrecall = 0;
8848 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
8849 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid),
8850 		    new_deleg, ls_hash);
8851 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
8852 		*new_delegp = NULL;
8853 		VNET(nfsstatsv1_p)->srvdelegates++;
8854 		nfsrv_openpluslock++;
8855 		nfsrv_delegatecnt++;
8856 	}
8857 }
8858 
8859 /*
8860  * Find and remove any delegations for the fh.
8861  */
8862 void
nfsrv_removedeleg(fhandle_t * fhp,struct nfsrv_descript * nd,NFSPROC_T * p)8863 nfsrv_removedeleg(fhandle_t *fhp, struct nfsrv_descript *nd, NFSPROC_T *p)
8864 {
8865 	struct nfsclient *clp;
8866 	struct nfsstate *stp, *nstp;
8867 	struct nfslockfile *lfp;
8868 	int error;
8869 
8870 	NFSLOCKSTATE();
8871 	error = nfsrv_getclient(nd->nd_clientid, CLOPS_RENEW, &clp, NULL,
8872 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
8873 	if (error == 0)
8874 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, fhp, 0);
8875 	/*
8876 	 * Now we must free any delegations.
8877 	 */
8878 	if (error == 0) {
8879 		LIST_FOREACH_SAFE(stp, &lfp->lf_deleg, ls_file, nstp)
8880 			nfsrv_freedeleg(stp);
8881 	}
8882 	NFSUNLOCKSTATE();
8883 }
8884