xref: /trueos/sys/fs/nfsserver/nfs_nfsdstate.c (revision 4fc31a2c6b4c624186fb609e12be08ea94c20c1a)
1 /*-
2  * Copyright (c) 2009 Rick Macklem, University of Guelph
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #ifndef APPLEKEXT
32 #include <fs/nfs/nfsport.h>
33 
34 struct nfsrv_stablefirst nfsrv_stablefirst;
35 int nfsrv_issuedelegs = 0;
36 int nfsrv_dolocallocks = 0;
37 struct nfsv4lock nfsv4rootfs_lock;
38 
39 extern int newnfs_numnfsd;
40 extern struct nfsstats newnfsstats;
41 extern int nfsrv_lease;
42 extern struct timeval nfsboottime;
43 extern u_int32_t newnfs_true, newnfs_false;
44 NFSV4ROOTLOCKMUTEX;
45 NFSSTATESPINLOCK;
46 
47 SYSCTL_DECL(_vfs_nfsd);
48 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
49 TUNABLE_INT("vfs.nfsd.statehashsize", &nfsrv_statehashsize);
50 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
51     &nfsrv_statehashsize, 0,
52     "Size of state hash table set via loader.conf");
53 
54 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
55 TUNABLE_INT("vfs.nfsd.clienthashsize", &nfsrv_clienthashsize);
56 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
57     &nfsrv_clienthashsize, 0,
58     "Size of client hash table set via loader.conf");
59 
60 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
61 TUNABLE_INT("vfs.nfsd.fhhashsize", &nfsrv_lockhashsize);
62 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
63     &nfsrv_lockhashsize, 0,
64     "Size of file handle hash table set via loader.conf");
65 
66 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
67 TUNABLE_INT("vfs.nfsd.sessionhashsize", &nfsrv_sessionhashsize);
68 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
69     &nfsrv_sessionhashsize, 0,
70     "Size of session hash table set via loader.conf");
71 
72 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
73 TUNABLE_INT("vfs.nfsd.v4statelimit", &nfsrv_v4statelimit);
74 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
75     &nfsrv_v4statelimit, 0,
76     "High water limit for NFSv4 opens+locks+delegations");
77 
78 /*
79  * Hash lists for nfs V4.
80  */
81 struct nfsclienthashhead	*nfsclienthash;
82 struct nfslockhashhead		*nfslockhash;
83 struct nfssessionhash		*nfssessionhash;
84 #endif	/* !APPLEKEXT */
85 
86 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
87 static time_t nfsrvboottime;
88 static int nfsrv_writedelegifpos = 1;
89 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
90 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
91 static int nfsrv_nogsscallback = 0;
92 
93 /* local functions */
94 static void nfsrv_dumpaclient(struct nfsclient *clp,
95     struct nfsd_dumpclients *dumpp);
96 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
97     NFSPROC_T *p);
98 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
99     NFSPROC_T *p);
100 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
101     NFSPROC_T *p);
102 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
103     int cansleep, NFSPROC_T *p);
104 static void nfsrv_freenfslock(struct nfslock *lop);
105 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
106 static void nfsrv_freedeleg(struct nfsstate *);
107 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
108     u_int32_t flags, struct nfsstate **stpp);
109 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
110     struct nfsstate **stpp);
111 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
112     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
113 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
114     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
115 static void nfsrv_insertlock(struct nfslock *new_lop,
116     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
117 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
118     struct nfslock **other_lopp, struct nfslockfile *lfp);
119 static int nfsrv_getipnumber(u_char *cp);
120 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
121     nfsv4stateid_t *stateidp, int specialid);
122 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
123     u_int32_t flags);
124 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
125     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
126     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
127 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
128     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
129 static u_int32_t nfsrv_nextclientindex(void);
130 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
131 static void nfsrv_markstable(struct nfsclient *clp);
132 static int nfsrv_checkstable(struct nfsclient *clp);
133 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
134     vnode *vp, NFSPROC_T *p);
135 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
136     NFSPROC_T *p, vnode_t vp);
137 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
138     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
139 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
140     struct nfsclient *clp);
141 static time_t nfsrv_leaseexpiry(void);
142 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
143 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
144     struct nfsstate *stp, struct nfsrvcache *op);
145 static int nfsrv_nootherstate(struct nfsstate *stp);
146 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
147     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
148 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
149     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
150 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
151     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
152     NFSPROC_T *p);
153 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
154     NFSPROC_T *p);
155 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
156     uint64_t first, uint64_t end);
157 static void nfsrv_locklf(struct nfslockfile *lfp);
158 static void nfsrv_unlocklf(struct nfslockfile *lfp);
159 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
160 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
161 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
162     int dont_replycache, struct nfsdsession **sepp);
163 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
164 
165 /*
166  * Scan the client list for a match and either return the current one,
167  * create a new entry or return an error.
168  * If returning a non-error, the clp structure must either be linked into
169  * the client list or free'd.
170  */
171 APPLESTATIC int
nfsrv_setclient(struct nfsrv_descript * nd,struct nfsclient ** new_clpp,nfsquad_t * clientidp,nfsquad_t * confirmp,NFSPROC_T * p)172 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
173     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
174 {
175 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
176 	int i, error = 0;
177 	struct nfsstate *stp, *tstp;
178 	struct sockaddr_in *sad, *rad;
179 	int zapit = 0, gotit, hasstate = 0, igotlock;
180 	static u_int64_t confirm_index = 0;
181 
182 	/*
183 	 * Check for state resource limit exceeded.
184 	 */
185 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
186 		error = NFSERR_RESOURCE;
187 		goto out;
188 	}
189 
190 	if (nfsrv_issuedelegs == 0 ||
191 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
192 		/*
193 		 * Don't do callbacks when delegations are disabled or
194 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
195 		 * If establishing a callback connection is attempted
196 		 * when a firewall is blocking the callback path, the
197 		 * server may wait too long for the connect attempt to
198 		 * succeed during the Open. Some clients, such as Linux,
199 		 * may timeout and give up on the Open before the server
200 		 * replies. Also, since AUTH_GSS callbacks are not
201 		 * yet interoperability tested, they might cause the
202 		 * server to crap out, if they get past the Init call to
203 		 * the client.
204 		 */
205 		new_clp->lc_program = 0;
206 
207 	/* Lock out other nfsd threads */
208 	NFSLOCKV4ROOTMUTEX();
209 	nfsv4_relref(&nfsv4rootfs_lock);
210 	do {
211 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
212 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
213 	} while (!igotlock);
214 	NFSUNLOCKV4ROOTMUTEX();
215 
216 	/*
217 	 * Search for a match in the client list.
218 	 */
219 	gotit = i = 0;
220 	while (i < nfsrv_clienthashsize && !gotit) {
221 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
222 		if (new_clp->lc_idlen == clp->lc_idlen &&
223 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
224 			gotit = 1;
225 			break;
226 		}
227 	    }
228 	    i++;
229 	}
230 	if (!gotit ||
231 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
232 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
233 			/*
234 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
235 			 * client is trying to update a confirmed clientid.
236 			 */
237 			NFSLOCKV4ROOTMUTEX();
238 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
239 			NFSUNLOCKV4ROOTMUTEX();
240 			confirmp->lval[1] = 0;
241 			error = NFSERR_NOENT;
242 			goto out;
243 		}
244 		/*
245 		 * Get rid of the old one.
246 		 */
247 		if (i != nfsrv_clienthashsize) {
248 			LIST_REMOVE(clp, lc_hash);
249 			nfsrv_cleanclient(clp, p);
250 			nfsrv_freedeleglist(&clp->lc_deleg);
251 			nfsrv_freedeleglist(&clp->lc_olddeleg);
252 			zapit = 1;
253 		}
254 		/*
255 		 * Add it after assigning a client id to it.
256 		 */
257 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
258 		if ((nd->nd_flag & ND_NFSV41) != 0)
259 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
260 			    ++confirm_index;
261 		else
262 			confirmp->qval = new_clp->lc_confirm.qval =
263 			    ++confirm_index;
264 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
265 		    (u_int32_t)nfsrvboottime;
266 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
267 		    nfsrv_nextclientindex();
268 		new_clp->lc_stateindex = 0;
269 		new_clp->lc_statemaxindex = 0;
270 		new_clp->lc_cbref = 0;
271 		new_clp->lc_expiry = nfsrv_leaseexpiry();
272 		LIST_INIT(&new_clp->lc_open);
273 		LIST_INIT(&new_clp->lc_deleg);
274 		LIST_INIT(&new_clp->lc_olddeleg);
275 		LIST_INIT(&new_clp->lc_session);
276 		for (i = 0; i < nfsrv_statehashsize; i++)
277 			LIST_INIT(&new_clp->lc_stateid[i]);
278 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
279 		    lc_hash);
280 		newnfsstats.srvclients++;
281 		nfsrv_openpluslock++;
282 		nfsrv_clients++;
283 		NFSLOCKV4ROOTMUTEX();
284 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
285 		NFSUNLOCKV4ROOTMUTEX();
286 		if (zapit)
287 			nfsrv_zapclient(clp, p);
288 		*new_clpp = NULL;
289 		goto out;
290 	}
291 
292 	/*
293 	 * Now, handle the cases where the id is already issued.
294 	 */
295 	if (nfsrv_notsamecredname(nd, clp)) {
296 	    /*
297 	     * Check to see if there is expired state that should go away.
298 	     */
299 	    if (clp->lc_expiry < NFSD_MONOSEC &&
300 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
301 		nfsrv_cleanclient(clp, p);
302 		nfsrv_freedeleglist(&clp->lc_deleg);
303 	    }
304 
305 	    /*
306 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
307 	     * RFC3530 Sec. 8.1.2 last para.
308 	     */
309 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
310 		hasstate = 1;
311 	    } else if (LIST_EMPTY(&clp->lc_open)) {
312 		hasstate = 0;
313 	    } else {
314 		hasstate = 0;
315 		/* Look for an Open on the OpenOwner */
316 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
317 		    if (!LIST_EMPTY(&stp->ls_open)) {
318 			hasstate = 1;
319 			break;
320 		    }
321 		}
322 	    }
323 	    if (hasstate) {
324 		/*
325 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
326 		 * filling out the correct ipaddr and portnum.
327 		 */
328 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
329 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
330 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
331 		sad->sin_port = rad->sin_port;
332 		NFSLOCKV4ROOTMUTEX();
333 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
334 		NFSUNLOCKV4ROOTMUTEX();
335 		error = NFSERR_CLIDINUSE;
336 		goto out;
337 	    }
338 	}
339 
340 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
341 		/*
342 		 * If the verifier has changed, the client has rebooted
343 		 * and a new client id is issued. The old state info
344 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
345 		 */
346 		LIST_REMOVE(clp, lc_hash);
347 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
348 		if ((nd->nd_flag & ND_NFSV41) != 0)
349 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
350 			    ++confirm_index;
351 		else
352 			confirmp->qval = new_clp->lc_confirm.qval =
353 			    ++confirm_index;
354 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
355 		    nfsrvboottime;
356 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
357 		    nfsrv_nextclientindex();
358 		new_clp->lc_stateindex = 0;
359 		new_clp->lc_statemaxindex = 0;
360 		new_clp->lc_cbref = 0;
361 		new_clp->lc_expiry = nfsrv_leaseexpiry();
362 
363 		/*
364 		 * Save the state until confirmed.
365 		 */
366 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
367 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
368 			tstp->ls_clp = new_clp;
369 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
370 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
371 			tstp->ls_clp = new_clp;
372 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
373 		    ls_list);
374 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
375 			tstp->ls_clp = new_clp;
376 		for (i = 0; i < nfsrv_statehashsize; i++) {
377 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
378 			    &clp->lc_stateid[i], ls_hash);
379 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
380 				tstp->ls_clp = new_clp;
381 		}
382 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
383 		    lc_hash);
384 		newnfsstats.srvclients++;
385 		nfsrv_openpluslock++;
386 		nfsrv_clients++;
387 		NFSLOCKV4ROOTMUTEX();
388 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
389 		NFSUNLOCKV4ROOTMUTEX();
390 
391 		/*
392 		 * Must wait until any outstanding callback on the old clp
393 		 * completes.
394 		 */
395 		NFSLOCKSTATE();
396 		while (clp->lc_cbref) {
397 			clp->lc_flags |= LCL_WAKEUPWANTED;
398 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
399 			    "nfsd clp", 10 * hz);
400 		}
401 		NFSUNLOCKSTATE();
402 		nfsrv_zapclient(clp, p);
403 		*new_clpp = NULL;
404 		goto out;
405 	}
406 
407 	/* For NFSv4.1, mark that we found a confirmed clientid. */
408 	if ((nd->nd_flag & ND_NFSV41) != 0)
409 		confirmp->lval[1] = 1;
410 	else {
411 		/*
412 		 * id and verifier match, so update the net address info
413 		 * and get rid of any existing callback authentication
414 		 * handle, so a new one will be acquired.
415 		 */
416 		LIST_REMOVE(clp, lc_hash);
417 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
418 		new_clp->lc_expiry = nfsrv_leaseexpiry();
419 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
420 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
421 		    clp->lc_clientid.lval[0];
422 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
423 		    clp->lc_clientid.lval[1];
424 		new_clp->lc_delegtime = clp->lc_delegtime;
425 		new_clp->lc_stateindex = clp->lc_stateindex;
426 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
427 		new_clp->lc_cbref = 0;
428 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
429 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
430 			tstp->ls_clp = new_clp;
431 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
432 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
433 			tstp->ls_clp = new_clp;
434 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
435 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
436 			tstp->ls_clp = new_clp;
437 		for (i = 0; i < nfsrv_statehashsize; i++) {
438 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
439 			    &clp->lc_stateid[i], ls_hash);
440 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
441 				tstp->ls_clp = new_clp;
442 		}
443 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
444 		    lc_hash);
445 		newnfsstats.srvclients++;
446 		nfsrv_openpluslock++;
447 		nfsrv_clients++;
448 	}
449 	NFSLOCKV4ROOTMUTEX();
450 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
451 	NFSUNLOCKV4ROOTMUTEX();
452 
453 	if ((nd->nd_flag & ND_NFSV41) == 0) {
454 		/*
455 		 * Must wait until any outstanding callback on the old clp
456 		 * completes.
457 		 */
458 		NFSLOCKSTATE();
459 		while (clp->lc_cbref) {
460 			clp->lc_flags |= LCL_WAKEUPWANTED;
461 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
462 			    "nfsdclp", 10 * hz);
463 		}
464 		NFSUNLOCKSTATE();
465 		nfsrv_zapclient(clp, p);
466 		*new_clpp = NULL;
467 	}
468 
469 out:
470 	NFSEXITCODE2(error, nd);
471 	return (error);
472 }
473 
474 /*
475  * Check to see if the client id exists and optionally confirm it.
476  */
477 APPLESTATIC 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)478 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
479     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
480     struct nfsrv_descript *nd, NFSPROC_T *p)
481 {
482 	struct nfsclient *clp;
483 	struct nfsstate *stp;
484 	int i;
485 	struct nfsclienthashhead *hp;
486 	int error = 0, igotlock, doneok;
487 	struct nfssessionhash *shp;
488 	struct nfsdsession *sep;
489 	uint64_t sessid[2];
490 	static uint64_t next_sess = 0;
491 
492 	if (clpp)
493 		*clpp = NULL;
494 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
495 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
496 		error = NFSERR_STALECLIENTID;
497 		goto out;
498 	}
499 
500 	/*
501 	 * If called with opflags == CLOPS_RENEW, the State Lock is
502 	 * already held. Otherwise, we need to get either that or,
503 	 * for the case of Confirm, lock out the nfsd threads.
504 	 */
505 	if (opflags & CLOPS_CONFIRM) {
506 		NFSLOCKV4ROOTMUTEX();
507 		nfsv4_relref(&nfsv4rootfs_lock);
508 		do {
509 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
510 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
511 		} while (!igotlock);
512 		/*
513 		 * Create a new sessionid here, since we need to do it where
514 		 * there is a mutex held to serialize update of next_sess.
515 		 */
516 		if ((nd->nd_flag & ND_NFSV41) != 0) {
517 			sessid[0] = ++next_sess;
518 			sessid[1] = clientid.qval;
519 		}
520 		NFSUNLOCKV4ROOTMUTEX();
521 	} else if (opflags != CLOPS_RENEW) {
522 		NFSLOCKSTATE();
523 	}
524 
525 	/* For NFSv4.1, the clp is acquired from the associated session. */
526 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
527 	    opflags == CLOPS_RENEW) {
528 		clp = NULL;
529 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
530 			shp = NFSSESSIONHASH(nd->nd_sessionid);
531 			NFSLOCKSESSION(shp);
532 			sep = nfsrv_findsession(nd->nd_sessionid);
533 			if (sep != NULL)
534 				clp = sep->sess_clp;
535 			NFSUNLOCKSESSION(shp);
536 		}
537 	} else {
538 		hp = NFSCLIENTHASH(clientid);
539 		LIST_FOREACH(clp, hp, lc_hash) {
540 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
541 				break;
542 		}
543 	}
544 	if (clp == NULL) {
545 		if (opflags & CLOPS_CONFIRM)
546 			error = NFSERR_STALECLIENTID;
547 		else
548 			error = NFSERR_EXPIRED;
549 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
550 		/*
551 		 * If marked admin revoked, just return the error.
552 		 */
553 		error = NFSERR_ADMINREVOKED;
554 	}
555 	if (error) {
556 		if (opflags & CLOPS_CONFIRM) {
557 			NFSLOCKV4ROOTMUTEX();
558 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
559 			NFSUNLOCKV4ROOTMUTEX();
560 		} else if (opflags != CLOPS_RENEW) {
561 			NFSUNLOCKSTATE();
562 		}
563 		goto out;
564 	}
565 
566 	/*
567 	 * Perform any operations specified by the opflags.
568 	 */
569 	if (opflags & CLOPS_CONFIRM) {
570 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
571 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
572 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
573 		     clp->lc_confirm.qval != confirm.qval))
574 			error = NFSERR_STALECLIENTID;
575 		else if (nfsrv_notsamecredname(nd, clp))
576 			error = NFSERR_CLIDINUSE;
577 
578 		if (!error) {
579 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
580 			LCL_NEEDSCONFIRM) {
581 			/*
582 			 * Hang onto the delegations (as old delegations)
583 			 * for an Open with CLAIM_DELEGATE_PREV unless in
584 			 * grace, but get rid of the rest of the state.
585 			 */
586 			nfsrv_cleanclient(clp, p);
587 			nfsrv_freedeleglist(&clp->lc_olddeleg);
588 			if (nfsrv_checkgrace(nd, clp, 0)) {
589 			    /* In grace, so just delete delegations */
590 			    nfsrv_freedeleglist(&clp->lc_deleg);
591 			} else {
592 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
593 				stp->ls_flags |= NFSLCK_OLDDELEG;
594 			    clp->lc_delegtime = NFSD_MONOSEC +
595 				nfsrv_lease + NFSRV_LEASEDELTA;
596 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
597 				ls_list);
598 			}
599 			if ((nd->nd_flag & ND_NFSV41) != 0)
600 			    clp->lc_program = cbprogram;
601 		    }
602 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
603 		    if (clp->lc_program)
604 			clp->lc_flags |= LCL_NEEDSCBNULL;
605 		    /* For NFSv4.1, link the session onto the client. */
606 		    if (nsep != NULL) {
607 			/* Hold a reference on the xprt for a backchannel. */
608 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
609 			    != 0 && clp->lc_req.nr_client == NULL) {
610 			    clp->lc_req.nr_client = (struct __rpc_client *)
611 				clnt_bck_create(nd->nd_xprt->xp_socket,
612 				cbprogram, NFSV4_CBVERS);
613 			    if (clp->lc_req.nr_client != NULL) {
614 				SVC_ACQUIRE(nd->nd_xprt);
615 				nd->nd_xprt->xp_p2 =
616 				    clp->lc_req.nr_client->cl_private;
617 				/* Disable idle timeout. */
618 				nd->nd_xprt->xp_idletimeout = 0;
619 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
620 			    } else
621 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
622 			}
623 			NFSBCOPY(sessid, nsep->sess_sessionid,
624 			    NFSX_V4SESSIONID);
625 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
626 			    NFSX_V4SESSIONID);
627 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
628 			NFSLOCKSESSION(shp);
629 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
630 			NFSLOCKSTATE();
631 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
632 			nsep->sess_clp = clp;
633 			NFSUNLOCKSTATE();
634 			NFSUNLOCKSESSION(shp);
635 		    }
636 		}
637 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
638 		error = NFSERR_EXPIRED;
639 	}
640 
641 	/*
642 	 * If called by the Renew Op, we must check the principal.
643 	 */
644 	if (!error && (opflags & CLOPS_RENEWOP)) {
645 	    if (nfsrv_notsamecredname(nd, clp)) {
646 		doneok = 0;
647 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
648 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
649 			if ((stp->ls_flags & NFSLCK_OPEN) &&
650 			    stp->ls_uid == nd->nd_cred->cr_uid) {
651 				doneok = 1;
652 				break;
653 			}
654 		    }
655 		}
656 		if (!doneok)
657 			error = NFSERR_ACCES;
658 	    }
659 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
660 		error = NFSERR_CBPATHDOWN;
661 	}
662 	if ((!error || error == NFSERR_CBPATHDOWN) &&
663 	     (opflags & CLOPS_RENEW)) {
664 		clp->lc_expiry = nfsrv_leaseexpiry();
665 	}
666 	if (opflags & CLOPS_CONFIRM) {
667 		NFSLOCKV4ROOTMUTEX();
668 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
669 		NFSUNLOCKV4ROOTMUTEX();
670 	} else if (opflags != CLOPS_RENEW) {
671 		NFSUNLOCKSTATE();
672 	}
673 	if (clpp)
674 		*clpp = clp;
675 
676 out:
677 	NFSEXITCODE2(error, nd);
678 	return (error);
679 }
680 
681 /*
682  * Perform the NFSv4.1 destroy clientid.
683  */
684 int
nfsrv_destroyclient(nfsquad_t clientid,NFSPROC_T * p)685 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
686 {
687 	struct nfsclient *clp;
688 	struct nfsclienthashhead *hp;
689 	int error = 0, i, igotlock;
690 
691 	if (nfsrvboottime != clientid.lval[0]) {
692 		error = NFSERR_STALECLIENTID;
693 		goto out;
694 	}
695 
696 	/* Lock out other nfsd threads */
697 	NFSLOCKV4ROOTMUTEX();
698 	nfsv4_relref(&nfsv4rootfs_lock);
699 	do {
700 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
701 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
702 	} while (igotlock == 0);
703 	NFSUNLOCKV4ROOTMUTEX();
704 
705 	hp = NFSCLIENTHASH(clientid);
706 	LIST_FOREACH(clp, hp, lc_hash) {
707 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
708 			break;
709 	}
710 	if (clp == NULL) {
711 		NFSLOCKV4ROOTMUTEX();
712 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
713 		NFSUNLOCKV4ROOTMUTEX();
714 		/* Just return ok, since it is gone. */
715 		goto out;
716 	}
717 
718 	/* Scan for state on the clientid. */
719 	for (i = 0; i < nfsrv_statehashsize; i++)
720 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
721 			NFSLOCKV4ROOTMUTEX();
722 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
723 			NFSUNLOCKV4ROOTMUTEX();
724 			error = NFSERR_CLIENTIDBUSY;
725 			goto out;
726 		}
727 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
728 		NFSLOCKV4ROOTMUTEX();
729 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
730 		NFSUNLOCKV4ROOTMUTEX();
731 		error = NFSERR_CLIENTIDBUSY;
732 		goto out;
733 	}
734 
735 	/* Destroy the clientid and return ok. */
736 	nfsrv_cleanclient(clp, p);
737 	nfsrv_freedeleglist(&clp->lc_deleg);
738 	nfsrv_freedeleglist(&clp->lc_olddeleg);
739 	LIST_REMOVE(clp, lc_hash);
740 	NFSLOCKV4ROOTMUTEX();
741 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
742 	NFSUNLOCKV4ROOTMUTEX();
743 	nfsrv_zapclient(clp, p);
744 out:
745 	NFSEXITCODE2(error, nd);
746 	return (error);
747 }
748 
749 /*
750  * Called from the new nfssvc syscall to admin revoke a clientid.
751  * Returns 0 for success, error otherwise.
752  */
753 APPLESTATIC int
nfsrv_adminrevoke(struct nfsd_clid * revokep,NFSPROC_T * p)754 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
755 {
756 	struct nfsclient *clp = NULL;
757 	int i, error = 0;
758 	int gotit, igotlock;
759 
760 	/*
761 	 * First, lock out the nfsd so that state won't change while the
762 	 * revocation record is being written to the stable storage restart
763 	 * file.
764 	 */
765 	NFSLOCKV4ROOTMUTEX();
766 	do {
767 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
768 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
769 	} while (!igotlock);
770 	NFSUNLOCKV4ROOTMUTEX();
771 
772 	/*
773 	 * Search for a match in the client list.
774 	 */
775 	gotit = i = 0;
776 	while (i < nfsrv_clienthashsize && !gotit) {
777 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
778 		if (revokep->nclid_idlen == clp->lc_idlen &&
779 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
780 			gotit = 1;
781 			break;
782 		}
783 	    }
784 	    i++;
785 	}
786 	if (!gotit) {
787 		NFSLOCKV4ROOTMUTEX();
788 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
789 		NFSUNLOCKV4ROOTMUTEX();
790 		error = EPERM;
791 		goto out;
792 	}
793 
794 	/*
795 	 * Now, write out the revocation record
796 	 */
797 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
798 	nfsrv_backupstable();
799 
800 	/*
801 	 * and clear out the state, marking the clientid revoked.
802 	 */
803 	clp->lc_flags &= ~LCL_CALLBACKSON;
804 	clp->lc_flags |= LCL_ADMINREVOKED;
805 	nfsrv_cleanclient(clp, p);
806 	nfsrv_freedeleglist(&clp->lc_deleg);
807 	nfsrv_freedeleglist(&clp->lc_olddeleg);
808 	NFSLOCKV4ROOTMUTEX();
809 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
810 	NFSUNLOCKV4ROOTMUTEX();
811 
812 out:
813 	NFSEXITCODE(error);
814 	return (error);
815 }
816 
817 /*
818  * Dump out stats for all clients. Called from nfssvc(2), that is used
819  * newnfsstats.
820  */
821 APPLESTATIC void
nfsrv_dumpclients(struct nfsd_dumpclients * dumpp,int maxcnt)822 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
823 {
824 	struct nfsclient *clp;
825 	int i = 0, cnt = 0;
826 
827 	/*
828 	 * First, get a reference on the nfsv4rootfs_lock so that an
829 	 * exclusive lock cannot be acquired while dumping the clients.
830 	 */
831 	NFSLOCKV4ROOTMUTEX();
832 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
833 	NFSUNLOCKV4ROOTMUTEX();
834 	NFSLOCKSTATE();
835 	/*
836 	 * Rattle through the client lists until done.
837 	 */
838 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
839 	    clp = LIST_FIRST(&nfsclienthash[i]);
840 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
841 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
842 		cnt++;
843 		clp = LIST_NEXT(clp, lc_hash);
844 	    }
845 	    i++;
846 	}
847 	if (cnt < maxcnt)
848 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
849 	NFSUNLOCKSTATE();
850 	NFSLOCKV4ROOTMUTEX();
851 	nfsv4_relref(&nfsv4rootfs_lock);
852 	NFSUNLOCKV4ROOTMUTEX();
853 }
854 
855 /*
856  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
857  */
858 static void
nfsrv_dumpaclient(struct nfsclient * clp,struct nfsd_dumpclients * dumpp)859 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
860 {
861 	struct nfsstate *stp, *openstp, *lckownstp;
862 	struct nfslock *lop;
863 	struct sockaddr *sad;
864 	struct sockaddr_in *rad;
865 	struct sockaddr_in6 *rad6;
866 
867 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
868 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
869 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
870 	dumpp->ndcl_flags = clp->lc_flags;
871 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
872 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
873 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
874 	dumpp->ndcl_addrfam = sad->sa_family;
875 	if (sad->sa_family == AF_INET) {
876 		rad = (struct sockaddr_in *)sad;
877 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
878 	} else {
879 		rad6 = (struct sockaddr_in6 *)sad;
880 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
881 	}
882 
883 	/*
884 	 * Now, scan the state lists and total up the opens and locks.
885 	 */
886 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
887 	    dumpp->ndcl_nopenowners++;
888 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
889 		dumpp->ndcl_nopens++;
890 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
891 		    dumpp->ndcl_nlockowners++;
892 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
893 			dumpp->ndcl_nlocks++;
894 		    }
895 		}
896 	    }
897 	}
898 
899 	/*
900 	 * and the delegation lists.
901 	 */
902 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
903 	    dumpp->ndcl_ndelegs++;
904 	}
905 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
906 	    dumpp->ndcl_nolddelegs++;
907 	}
908 }
909 
910 /*
911  * Dump out lock stats for a file.
912  */
913 APPLESTATIC void
nfsrv_dumplocks(vnode_t vp,struct nfsd_dumplocks * ldumpp,int maxcnt,NFSPROC_T * p)914 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
915     NFSPROC_T *p)
916 {
917 	struct nfsstate *stp;
918 	struct nfslock *lop;
919 	int cnt = 0;
920 	struct nfslockfile *lfp;
921 	struct sockaddr *sad;
922 	struct sockaddr_in *rad;
923 	struct sockaddr_in6 *rad6;
924 	int ret;
925 	fhandle_t nfh;
926 
927 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
928 	/*
929 	 * First, get a reference on the nfsv4rootfs_lock so that an
930 	 * exclusive lock on it cannot be acquired while dumping the locks.
931 	 */
932 	NFSLOCKV4ROOTMUTEX();
933 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
934 	NFSUNLOCKV4ROOTMUTEX();
935 	NFSLOCKSTATE();
936 	if (!ret)
937 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
938 	if (ret) {
939 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
940 		NFSUNLOCKSTATE();
941 		NFSLOCKV4ROOTMUTEX();
942 		nfsv4_relref(&nfsv4rootfs_lock);
943 		NFSUNLOCKV4ROOTMUTEX();
944 		return;
945 	}
946 
947 	/*
948 	 * For each open share on file, dump it out.
949 	 */
950 	stp = LIST_FIRST(&lfp->lf_open);
951 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
952 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
953 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
954 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
955 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
956 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
957 		ldumpp[cnt].ndlck_owner.nclid_idlen =
958 		    stp->ls_openowner->ls_ownerlen;
959 		NFSBCOPY(stp->ls_openowner->ls_owner,
960 		    ldumpp[cnt].ndlck_owner.nclid_id,
961 		    stp->ls_openowner->ls_ownerlen);
962 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
963 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
964 		    stp->ls_clp->lc_idlen);
965 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
966 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
967 		if (sad->sa_family == AF_INET) {
968 			rad = (struct sockaddr_in *)sad;
969 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
970 		} else {
971 			rad6 = (struct sockaddr_in6 *)sad;
972 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
973 		}
974 		stp = LIST_NEXT(stp, ls_file);
975 		cnt++;
976 	}
977 
978 	/*
979 	 * and all locks.
980 	 */
981 	lop = LIST_FIRST(&lfp->lf_lock);
982 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
983 		stp = lop->lo_stp;
984 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
985 		ldumpp[cnt].ndlck_first = lop->lo_first;
986 		ldumpp[cnt].ndlck_end = lop->lo_end;
987 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
988 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
989 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
990 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
991 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
992 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
993 		    stp->ls_ownerlen);
994 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
995 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
996 		    stp->ls_clp->lc_idlen);
997 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
998 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
999 		if (sad->sa_family == AF_INET) {
1000 			rad = (struct sockaddr_in *)sad;
1001 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1002 		} else {
1003 			rad6 = (struct sockaddr_in6 *)sad;
1004 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1005 		}
1006 		lop = LIST_NEXT(lop, lo_lckfile);
1007 		cnt++;
1008 	}
1009 
1010 	/*
1011 	 * and the delegations.
1012 	 */
1013 	stp = LIST_FIRST(&lfp->lf_deleg);
1014 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1015 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1016 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1017 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1018 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1019 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1020 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1021 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1022 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1023 		    stp->ls_clp->lc_idlen);
1024 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1025 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1026 		if (sad->sa_family == AF_INET) {
1027 			rad = (struct sockaddr_in *)sad;
1028 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1029 		} else {
1030 			rad6 = (struct sockaddr_in6 *)sad;
1031 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1032 		}
1033 		stp = LIST_NEXT(stp, ls_file);
1034 		cnt++;
1035 	}
1036 
1037 	/*
1038 	 * If list isn't full, mark end of list by setting the client name
1039 	 * to zero length.
1040 	 */
1041 	if (cnt < maxcnt)
1042 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1043 	NFSUNLOCKSTATE();
1044 	NFSLOCKV4ROOTMUTEX();
1045 	nfsv4_relref(&nfsv4rootfs_lock);
1046 	NFSUNLOCKV4ROOTMUTEX();
1047 }
1048 
1049 /*
1050  * Server timer routine. It can scan any linked list, so long
1051  * as it holds the spin/mutex lock and there is no exclusive lock on
1052  * nfsv4rootfs_lock.
1053  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1054  *  to do this from a callout, since the spin locks work. For
1055  *  Darwin, I'm not sure what will work correctly yet.)
1056  * Should be called once per second.
1057  */
1058 APPLESTATIC void
nfsrv_servertimer(void)1059 nfsrv_servertimer(void)
1060 {
1061 	struct nfsclient *clp, *nclp;
1062 	struct nfsstate *stp, *nstp;
1063 	int got_ref, i;
1064 
1065 	/*
1066 	 * Make sure nfsboottime is set. This is used by V3 as well
1067 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1068 	 * only used by the V4 server for leases.
1069 	 */
1070 	if (nfsboottime.tv_sec == 0)
1071 		NFSSETBOOTTIME(nfsboottime);
1072 
1073 	/*
1074 	 * If server hasn't started yet, just return.
1075 	 */
1076 	NFSLOCKSTATE();
1077 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1078 		NFSUNLOCKSTATE();
1079 		return;
1080 	}
1081 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1082 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1083 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1084 			nfsrv_stablefirst.nsf_flags |=
1085 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1086 		NFSUNLOCKSTATE();
1087 		return;
1088 	}
1089 
1090 	/*
1091 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1092 	 * no nfsd thread can acquire an exclusive lock on it before this
1093 	 * call is done. If it is already exclusively locked, just return.
1094 	 */
1095 	NFSLOCKV4ROOTMUTEX();
1096 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1097 	NFSUNLOCKV4ROOTMUTEX();
1098 	if (got_ref == 0) {
1099 		NFSUNLOCKSTATE();
1100 		return;
1101 	}
1102 
1103 	/*
1104 	 * For each client...
1105 	 */
1106 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1107 	    clp = LIST_FIRST(&nfsclienthash[i]);
1108 	    while (clp != LIST_END(&nfsclienthash[i])) {
1109 		nclp = LIST_NEXT(clp, lc_hash);
1110 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1111 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1112 			 && ((LIST_EMPTY(&clp->lc_deleg)
1113 			      && LIST_EMPTY(&clp->lc_open)) ||
1114 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1115 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1116 			(clp->lc_expiry < NFSD_MONOSEC &&
1117 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1118 			/*
1119 			 * Lease has expired several nfsrv_lease times ago:
1120 			 * PLUS
1121 			 *    - no state is associated with it
1122 			 *    OR
1123 			 *    - above high water mark for number of clients
1124 			 *      (nfsrv_clienthighwater should be large enough
1125 			 *       that this only occurs when clients fail to
1126 			 *       use the same nfs_client_id4.id. Maybe somewhat
1127 			 *       higher that the maximum number of clients that
1128 			 *       will mount this server?)
1129 			 * OR
1130 			 * Lease has expired a very long time ago
1131 			 * OR
1132 			 * Lease has expired PLUS the number of opens + locks
1133 			 * has exceeded 90% of capacity
1134 			 *
1135 			 * --> Mark for expiry. The actual expiry will be done
1136 			 *     by an nfsd sometime soon.
1137 			 */
1138 			clp->lc_flags |= LCL_EXPIREIT;
1139 			nfsrv_stablefirst.nsf_flags |=
1140 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1141 		    } else {
1142 			/*
1143 			 * If there are no opens, increment no open tick cnt
1144 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1145 			 * otherwise, if there is an open, reset no open time
1146 			 * Hopefully, this will avoid excessive re-creation
1147 			 * of open owners and subsequent open confirms.
1148 			 */
1149 			stp = LIST_FIRST(&clp->lc_open);
1150 			while (stp != LIST_END(&clp->lc_open)) {
1151 				nstp = LIST_NEXT(stp, ls_list);
1152 				if (LIST_EMPTY(&stp->ls_open)) {
1153 					stp->ls_noopens++;
1154 					if (stp->ls_noopens > NFSNOOPEN ||
1155 					    (nfsrv_openpluslock * 2) >
1156 					    nfsrv_v4statelimit)
1157 						nfsrv_stablefirst.nsf_flags |=
1158 							NFSNSF_NOOPENS;
1159 				} else {
1160 					stp->ls_noopens = 0;
1161 				}
1162 				stp = nstp;
1163 			}
1164 		    }
1165 		}
1166 		clp = nclp;
1167 	    }
1168 	}
1169 	NFSUNLOCKSTATE();
1170 	NFSLOCKV4ROOTMUTEX();
1171 	nfsv4_relref(&nfsv4rootfs_lock);
1172 	NFSUNLOCKV4ROOTMUTEX();
1173 }
1174 
1175 /*
1176  * The following set of functions free up the various data structures.
1177  */
1178 /*
1179  * Clear out all open/lock state related to this nfsclient.
1180  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1181  * there are no other active nfsd threads.
1182  */
1183 APPLESTATIC void
nfsrv_cleanclient(struct nfsclient * clp,NFSPROC_T * p)1184 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1185 {
1186 	struct nfsstate *stp, *nstp;
1187 	struct nfsdsession *sep, *nsep;
1188 
1189 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1190 		nfsrv_freeopenowner(stp, 1, p);
1191 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1192 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1193 			(void)nfsrv_freesession(sep, NULL);
1194 }
1195 
1196 /*
1197  * Free a client that has been cleaned. It should also already have been
1198  * removed from the lists.
1199  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1200  *  softclock interrupts are enabled.)
1201  */
1202 APPLESTATIC void
nfsrv_zapclient(struct nfsclient * clp,NFSPROC_T * p)1203 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1204 {
1205 
1206 #ifdef notyet
1207 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1208 	     (LCL_GSS | LCL_CALLBACKSON) &&
1209 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1210 	    clp->lc_handlelen > 0) {
1211 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1212 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1213 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1214 			NULL, 0, NULL, NULL, NULL, p);
1215 	}
1216 #endif
1217 	newnfs_disconnect(&clp->lc_req);
1218 	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
1219 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1220 	free(clp->lc_stateid, M_NFSDCLIENT);
1221 	free(clp, M_NFSDCLIENT);
1222 	NFSLOCKSTATE();
1223 	newnfsstats.srvclients--;
1224 	nfsrv_openpluslock--;
1225 	nfsrv_clients--;
1226 	NFSUNLOCKSTATE();
1227 }
1228 
1229 /*
1230  * Free a list of delegation state structures.
1231  * (This function will also free all nfslockfile structures that no
1232  *  longer have associated state.)
1233  */
1234 APPLESTATIC void
nfsrv_freedeleglist(struct nfsstatehead * sthp)1235 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1236 {
1237 	struct nfsstate *stp, *nstp;
1238 
1239 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1240 		nfsrv_freedeleg(stp);
1241 	}
1242 	LIST_INIT(sthp);
1243 }
1244 
1245 /*
1246  * Free up a delegation.
1247  */
1248 static void
nfsrv_freedeleg(struct nfsstate * stp)1249 nfsrv_freedeleg(struct nfsstate *stp)
1250 {
1251 	struct nfslockfile *lfp;
1252 
1253 	LIST_REMOVE(stp, ls_hash);
1254 	LIST_REMOVE(stp, ls_list);
1255 	LIST_REMOVE(stp, ls_file);
1256 	lfp = stp->ls_lfp;
1257 	if (LIST_EMPTY(&lfp->lf_open) &&
1258 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1259 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1260 	    lfp->lf_usecount == 0 &&
1261 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1262 		nfsrv_freenfslockfile(lfp);
1263 	FREE((caddr_t)stp, M_NFSDSTATE);
1264 	newnfsstats.srvdelegates--;
1265 	nfsrv_openpluslock--;
1266 	nfsrv_delegatecnt--;
1267 }
1268 
1269 /*
1270  * This function frees an open owner and all associated opens.
1271  */
1272 static void
nfsrv_freeopenowner(struct nfsstate * stp,int cansleep,NFSPROC_T * p)1273 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1274 {
1275 	struct nfsstate *nstp, *tstp;
1276 
1277 	LIST_REMOVE(stp, ls_list);
1278 	/*
1279 	 * Now, free all associated opens.
1280 	 */
1281 	nstp = LIST_FIRST(&stp->ls_open);
1282 	while (nstp != LIST_END(&stp->ls_open)) {
1283 		tstp = nstp;
1284 		nstp = LIST_NEXT(nstp, ls_list);
1285 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1286 	}
1287 	if (stp->ls_op)
1288 		nfsrvd_derefcache(stp->ls_op);
1289 	FREE((caddr_t)stp, M_NFSDSTATE);
1290 	newnfsstats.srvopenowners--;
1291 	nfsrv_openpluslock--;
1292 }
1293 
1294 /*
1295  * This function frees an open (nfsstate open structure) with all associated
1296  * lock_owners and locks. It also frees the nfslockfile structure iff there
1297  * are no other opens on the file.
1298  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1299  */
1300 static int
nfsrv_freeopen(struct nfsstate * stp,vnode_t vp,int cansleep,NFSPROC_T * p)1301 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1302 {
1303 	struct nfsstate *nstp, *tstp;
1304 	struct nfslockfile *lfp;
1305 	int ret;
1306 
1307 	LIST_REMOVE(stp, ls_hash);
1308 	LIST_REMOVE(stp, ls_list);
1309 	LIST_REMOVE(stp, ls_file);
1310 
1311 	lfp = stp->ls_lfp;
1312 	/*
1313 	 * Now, free all lockowners associated with this open.
1314 	 */
1315 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1316 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1317 
1318 	/*
1319 	 * The nfslockfile is freed here if there are no locks
1320 	 * associated with the open.
1321 	 * If there are locks associated with the open, the
1322 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1323 	 * Acquire the state mutex to avoid races with calls to
1324 	 * nfsrv_getlockfile().
1325 	 */
1326 	if (cansleep != 0)
1327 		NFSLOCKSTATE();
1328 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1329 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1330 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1331 	    lfp->lf_usecount == 0 &&
1332 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1333 		nfsrv_freenfslockfile(lfp);
1334 		ret = 1;
1335 	} else
1336 		ret = 0;
1337 	if (cansleep != 0)
1338 		NFSUNLOCKSTATE();
1339 	FREE((caddr_t)stp, M_NFSDSTATE);
1340 	newnfsstats.srvopens--;
1341 	nfsrv_openpluslock--;
1342 	return (ret);
1343 }
1344 
1345 /*
1346  * Frees a lockowner and all associated locks.
1347  */
1348 static void
nfsrv_freelockowner(struct nfsstate * stp,vnode_t vp,int cansleep,NFSPROC_T * p)1349 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1350     NFSPROC_T *p)
1351 {
1352 
1353 	LIST_REMOVE(stp, ls_hash);
1354 	LIST_REMOVE(stp, ls_list);
1355 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1356 	if (stp->ls_op)
1357 		nfsrvd_derefcache(stp->ls_op);
1358 	FREE((caddr_t)stp, M_NFSDSTATE);
1359 	newnfsstats.srvlockowners--;
1360 	nfsrv_openpluslock--;
1361 }
1362 
1363 /*
1364  * Free all the nfs locks on a lockowner.
1365  */
1366 static void
nfsrv_freeallnfslocks(struct nfsstate * stp,vnode_t vp,int cansleep,NFSPROC_T * p)1367 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1368     NFSPROC_T *p)
1369 {
1370 	struct nfslock *lop, *nlop;
1371 	struct nfsrollback *rlp, *nrlp;
1372 	struct nfslockfile *lfp = NULL;
1373 	int gottvp = 0;
1374 	vnode_t tvp = NULL;
1375 	uint64_t first, end;
1376 
1377 	if (vp != NULL)
1378 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1379 	lop = LIST_FIRST(&stp->ls_lock);
1380 	while (lop != LIST_END(&stp->ls_lock)) {
1381 		nlop = LIST_NEXT(lop, lo_lckowner);
1382 		/*
1383 		 * Since all locks should be for the same file, lfp should
1384 		 * not change.
1385 		 */
1386 		if (lfp == NULL)
1387 			lfp = lop->lo_lfp;
1388 		else if (lfp != lop->lo_lfp)
1389 			panic("allnfslocks");
1390 		/*
1391 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1392 		 * from the file handle. This only occurs when called from
1393 		 * nfsrv_cleanclient().
1394 		 */
1395 		if (gottvp == 0) {
1396 			if (nfsrv_dolocallocks == 0)
1397 				tvp = NULL;
1398 			else if (vp == NULL && cansleep != 0) {
1399 				tvp = nfsvno_getvp(&lfp->lf_fh);
1400 				NFSVOPUNLOCK(tvp, 0);
1401 			} else
1402 				tvp = vp;
1403 			gottvp = 1;
1404 		}
1405 
1406 		if (tvp != NULL) {
1407 			if (cansleep == 0)
1408 				panic("allnfs2");
1409 			first = lop->lo_first;
1410 			end = lop->lo_end;
1411 			nfsrv_freenfslock(lop);
1412 			nfsrv_localunlock(tvp, lfp, first, end, p);
1413 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1414 			    nrlp)
1415 				free(rlp, M_NFSDROLLBACK);
1416 			LIST_INIT(&lfp->lf_rollback);
1417 		} else
1418 			nfsrv_freenfslock(lop);
1419 		lop = nlop;
1420 	}
1421 	if (vp == NULL && tvp != NULL)
1422 		vrele(tvp);
1423 }
1424 
1425 /*
1426  * Free an nfslock structure.
1427  */
1428 static void
nfsrv_freenfslock(struct nfslock * lop)1429 nfsrv_freenfslock(struct nfslock *lop)
1430 {
1431 
1432 	if (lop->lo_lckfile.le_prev != NULL) {
1433 		LIST_REMOVE(lop, lo_lckfile);
1434 		newnfsstats.srvlocks--;
1435 		nfsrv_openpluslock--;
1436 	}
1437 	LIST_REMOVE(lop, lo_lckowner);
1438 	FREE((caddr_t)lop, M_NFSDLOCK);
1439 }
1440 
1441 /*
1442  * This function frees an nfslockfile structure.
1443  */
1444 static void
nfsrv_freenfslockfile(struct nfslockfile * lfp)1445 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1446 {
1447 
1448 	LIST_REMOVE(lfp, lf_hash);
1449 	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1450 }
1451 
1452 /*
1453  * This function looks up an nfsstate structure via stateid.
1454  */
1455 static int
nfsrv_getstate(struct nfsclient * clp,nfsv4stateid_t * stateidp,__unused u_int32_t flags,struct nfsstate ** stpp)1456 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1457     struct nfsstate **stpp)
1458 {
1459 	struct nfsstate *stp;
1460 	struct nfsstatehead *hp;
1461 	int error = 0;
1462 
1463 	*stpp = NULL;
1464 	hp = NFSSTATEHASH(clp, *stateidp);
1465 	LIST_FOREACH(stp, hp, ls_hash) {
1466 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1467 			NFSX_STATEIDOTHER))
1468 			break;
1469 	}
1470 
1471 	/*
1472 	 * If no state id in list, return NFSERR_BADSTATEID.
1473 	 */
1474 	if (stp == LIST_END(hp)) {
1475 		error = NFSERR_BADSTATEID;
1476 		goto out;
1477 	}
1478 	*stpp = stp;
1479 
1480 out:
1481 	NFSEXITCODE(error);
1482 	return (error);
1483 }
1484 
1485 /*
1486  * This function gets an nfsstate structure via owner string.
1487  */
1488 static void
nfsrv_getowner(struct nfsstatehead * hp,struct nfsstate * new_stp,struct nfsstate ** stpp)1489 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1490     struct nfsstate **stpp)
1491 {
1492 	struct nfsstate *stp;
1493 
1494 	*stpp = NULL;
1495 	LIST_FOREACH(stp, hp, ls_list) {
1496 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1497 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1498 			*stpp = stp;
1499 			return;
1500 		}
1501 	}
1502 }
1503 
1504 /*
1505  * Lock control function called to update lock status.
1506  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1507  * that one isn't to be created and an NFSERR_xxx for other errors.
1508  * The structures new_stp and new_lop are passed in as pointers that should
1509  * be set to NULL if the structure is used and shouldn't be free'd.
1510  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1511  * never used and can safely be allocated on the stack. For all other
1512  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1513  * in case they are used.
1514  */
1515 APPLESTATIC 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)1516 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1517     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1518     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1519     __unused struct nfsexstuff *exp,
1520     struct nfsrv_descript *nd, NFSPROC_T *p)
1521 {
1522 	struct nfslock *lop;
1523 	struct nfsstate *new_stp = *new_stpp;
1524 	struct nfslock *new_lop = *new_lopp;
1525 	struct nfsstate *tstp, *mystp, *nstp;
1526 	int specialid = 0;
1527 	struct nfslockfile *lfp;
1528 	struct nfslock *other_lop = NULL;
1529 	struct nfsstate *stp, *lckstp = NULL;
1530 	struct nfsclient *clp = NULL;
1531 	u_int32_t bits;
1532 	int error = 0, haslock = 0, ret, reterr;
1533 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1534 	fhandle_t nfh;
1535 	uint64_t first, end;
1536 	uint32_t lock_flags;
1537 
1538 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1539 		/*
1540 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1541 		 * let reads with all 1s go ahead.
1542 		 */
1543 		if (new_stp->ls_stateid.seqid == 0x0 &&
1544 		    new_stp->ls_stateid.other[0] == 0x0 &&
1545 		    new_stp->ls_stateid.other[1] == 0x0 &&
1546 		    new_stp->ls_stateid.other[2] == 0x0)
1547 			specialid = 1;
1548 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1549 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1550 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1551 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1552 			specialid = 2;
1553 	}
1554 
1555 	/*
1556 	 * Check for restart conditions (client and server).
1557 	 */
1558 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1559 	    &new_stp->ls_stateid, specialid);
1560 	if (error)
1561 		goto out;
1562 
1563 	/*
1564 	 * Check for state resource limit exceeded.
1565 	 */
1566 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1567 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1568 		error = NFSERR_RESOURCE;
1569 		goto out;
1570 	}
1571 
1572 	/*
1573 	 * For the lock case, get another nfslock structure,
1574 	 * just in case we need it.
1575 	 * Malloc now, before we start sifting through the linked lists,
1576 	 * in case we have to wait for memory.
1577 	 */
1578 tryagain:
1579 	if (new_stp->ls_flags & NFSLCK_LOCK)
1580 		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1581 		    M_NFSDLOCK, M_WAITOK);
1582 	filestruct_locked = 0;
1583 	reterr = 0;
1584 	lfp = NULL;
1585 
1586 	/*
1587 	 * Get the lockfile structure for CFH now, so we can do a sanity
1588 	 * check against the stateid, before incrementing the seqid#, since
1589 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1590 	 * shouldn't be incremented for this case.
1591 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1592 	 * will be handled later.
1593 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1594 	 * lock the nfslockfile structure.
1595 	 */
1596 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1597 	NFSLOCKSTATE();
1598 	if (getlckret == 0) {
1599 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1600 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1601 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1602 			    &lfp, &nfh, 1);
1603 			if (getlckret == 0)
1604 				filestruct_locked = 1;
1605 		} else
1606 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1607 			    &lfp, &nfh, 0);
1608 	}
1609 	if (getlckret != 0 && getlckret != -1)
1610 		reterr = getlckret;
1611 
1612 	if (filestruct_locked != 0) {
1613 		LIST_INIT(&lfp->lf_rollback);
1614 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1615 			/*
1616 			 * For local locking, do the advisory locking now, so
1617 			 * that any conflict can be detected. A failure later
1618 			 * can be rolled back locally. If an error is returned,
1619 			 * struct nfslockfile has been unlocked and any local
1620 			 * locking rolled back.
1621 			 */
1622 			NFSUNLOCKSTATE();
1623 			if (vnode_unlocked == 0) {
1624 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1625 				vnode_unlocked = 1;
1626 				NFSVOPUNLOCK(vp, 0);
1627 			}
1628 			reterr = nfsrv_locallock(vp, lfp,
1629 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1630 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1631 			NFSLOCKSTATE();
1632 		}
1633 	}
1634 
1635 	if (specialid == 0) {
1636 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1637 		/*
1638 		 * RFC 3530 does not list LockT as an op that renews a
1639 		 * lease, but the concensus seems to be that it is ok
1640 		 * for a server to do so.
1641 		 */
1642 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1643 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1644 
1645 		/*
1646 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1647 		 * error returns for LockT, just go ahead and test for a lock,
1648 		 * since there are no locks for this client, but other locks
1649 		 * can conflict. (ie. same client will always be false)
1650 		 */
1651 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1652 		    error = 0;
1653 		lckstp = new_stp;
1654 	    } else {
1655 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1656 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1657 	      if (error == 0)
1658 		/*
1659 		 * Look up the stateid
1660 		 */
1661 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1662 		  new_stp->ls_flags, &stp);
1663 	      /*
1664 	       * do some sanity checks for an unconfirmed open or a
1665 	       * stateid that refers to the wrong file, for an open stateid
1666 	       */
1667 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1668 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1669 		   (getlckret == 0 && stp->ls_lfp != lfp)))
1670 			error = NFSERR_BADSTATEID;
1671 	      if (error == 0 &&
1672 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1673 		  getlckret == 0 && stp->ls_lfp != lfp)
1674 			error = NFSERR_BADSTATEID;
1675 
1676 	      /*
1677 	       * If the lockowner stateid doesn't refer to the same file,
1678 	       * I believe that is considered ok, since some clients will
1679 	       * only create a single lockowner and use that for all locks
1680 	       * on all files.
1681 	       * For now, log it as a diagnostic, instead of considering it
1682 	       * a BadStateid.
1683 	       */
1684 	      if (error == 0 && (stp->ls_flags &
1685 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1686 		  getlckret == 0 && stp->ls_lfp != lfp) {
1687 #ifdef DIAGNOSTIC
1688 		  printf("Got a lock statid for different file open\n");
1689 #endif
1690 		  /*
1691 		  error = NFSERR_BADSTATEID;
1692 		  */
1693 	      }
1694 
1695 	      if (error == 0) {
1696 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1697 			/*
1698 			 * If haslock set, we've already checked the seqid.
1699 			 */
1700 			if (!haslock) {
1701 			    if (stp->ls_flags & NFSLCK_OPEN)
1702 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1703 				    stp->ls_openowner, new_stp->ls_op);
1704 			    else
1705 				error = NFSERR_BADSTATEID;
1706 			}
1707 			if (!error)
1708 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1709 			if (lckstp)
1710 			    /*
1711 			     * I believe this should be an error, but it
1712 			     * isn't obvious what NFSERR_xxx would be
1713 			     * appropriate, so I'll use NFSERR_INVAL for now.
1714 			     */
1715 			    error = NFSERR_INVAL;
1716 			else
1717 			    lckstp = new_stp;
1718 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1719 			/*
1720 			 * If haslock set, ditto above.
1721 			 */
1722 			if (!haslock) {
1723 			    if (stp->ls_flags & NFSLCK_OPEN)
1724 				error = NFSERR_BADSTATEID;
1725 			    else
1726 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1727 				    stp, new_stp->ls_op);
1728 			}
1729 			lckstp = stp;
1730 		    } else {
1731 			lckstp = stp;
1732 		    }
1733 	      }
1734 	      /*
1735 	       * If the seqid part of the stateid isn't the same, return
1736 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1737 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1738 	       * nfsrv_returnoldstateid is set. (The concensus on the email
1739 	       * list was that most clients would prefer to not receive
1740 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1741 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1742 	       * allow for either server configuration.)
1743 	       */
1744 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1745 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1746 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1747 		    nfsrv_returnoldstateid)) ||
1748 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1749 		    new_stp->ls_stateid.seqid != 0)))
1750 		    error = NFSERR_OLDSTATEID;
1751 	    }
1752 	}
1753 
1754 	/*
1755 	 * Now we can check for grace.
1756 	 */
1757 	if (!error)
1758 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1759 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1760 		nfsrv_checkstable(clp))
1761 		error = NFSERR_NOGRACE;
1762 	/*
1763 	 * If we successfully Reclaimed state, note that.
1764 	 */
1765 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1766 		nfsrv_markstable(clp);
1767 
1768 	/*
1769 	 * At this point, either error == NFSERR_BADSTATEID or the
1770 	 * seqid# has been updated, so we can return any error.
1771 	 * If error == 0, there may be an error in:
1772 	 *    nd_repstat - Set by the calling function.
1773 	 *    reterr - Set above, if getting the nfslockfile structure
1774 	 *       or acquiring the local lock failed.
1775 	 *    (If both of these are set, nd_repstat should probably be
1776 	 *     returned, since that error was detected before this
1777 	 *     function call.)
1778 	 */
1779 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1780 		if (error == 0) {
1781 			if (nd->nd_repstat != 0)
1782 				error = nd->nd_repstat;
1783 			else
1784 				error = reterr;
1785 		}
1786 		if (filestruct_locked != 0) {
1787 			/* Roll back local locks. */
1788 			NFSUNLOCKSTATE();
1789 			if (vnode_unlocked == 0) {
1790 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1791 				vnode_unlocked = 1;
1792 				NFSVOPUNLOCK(vp, 0);
1793 			}
1794 			nfsrv_locallock_rollback(vp, lfp, p);
1795 			NFSLOCKSTATE();
1796 			nfsrv_unlocklf(lfp);
1797 		}
1798 		NFSUNLOCKSTATE();
1799 		goto out;
1800 	}
1801 
1802 	/*
1803 	 * Check the nfsrv_getlockfile return.
1804 	 * Returned -1 if no structure found.
1805 	 */
1806 	if (getlckret == -1) {
1807 		error = NFSERR_EXPIRED;
1808 		/*
1809 		 * Called from lockt, so no lock is OK.
1810 		 */
1811 		if (new_stp->ls_flags & NFSLCK_TEST) {
1812 			error = 0;
1813 		} else if (new_stp->ls_flags &
1814 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1815 			/*
1816 			 * Called to check for a lock, OK if the stateid is all
1817 			 * 1s or all 0s, but there should be an nfsstate
1818 			 * otherwise.
1819 			 * (ie. If there is no open, I'll assume no share
1820 			 *  deny bits.)
1821 			 */
1822 			if (specialid)
1823 				error = 0;
1824 			else
1825 				error = NFSERR_BADSTATEID;
1826 		}
1827 		NFSUNLOCKSTATE();
1828 		goto out;
1829 	}
1830 
1831 	/*
1832 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1833 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1834 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1835 	 * which implies a conflicting deny can't exist.
1836 	 */
1837 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1838 	    /*
1839 	     * Four kinds of state id:
1840 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1841 	     * - stateid for an open
1842 	     * - stateid for a delegation
1843 	     * - stateid for a lock owner
1844 	     */
1845 	    if (!specialid) {
1846 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1847 		    delegation = 1;
1848 		    mystp = stp;
1849 		    nfsrv_delaydelegtimeout(stp);
1850 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1851 		    mystp = stp;
1852 		} else {
1853 		    mystp = stp->ls_openstp;
1854 		}
1855 		/*
1856 		 * If locking or checking, require correct access
1857 		 * bit set.
1858 		 */
1859 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1860 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1861 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1862 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1863 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1864 		     !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1865 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1866 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1867 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1868 			if (filestruct_locked != 0) {
1869 				/* Roll back local locks. */
1870 				NFSUNLOCKSTATE();
1871 				if (vnode_unlocked == 0) {
1872 					ASSERT_VOP_ELOCKED(vp,
1873 					    "nfsrv_lockctrl3");
1874 					vnode_unlocked = 1;
1875 					NFSVOPUNLOCK(vp, 0);
1876 				}
1877 				nfsrv_locallock_rollback(vp, lfp, p);
1878 				NFSLOCKSTATE();
1879 				nfsrv_unlocklf(lfp);
1880 			}
1881 			NFSUNLOCKSTATE();
1882 			error = NFSERR_OPENMODE;
1883 			goto out;
1884 		}
1885 	    } else
1886 		mystp = NULL;
1887 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1888 		/*
1889 		 * Check for a conflicting deny bit.
1890 		 */
1891 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1892 		    if (tstp != mystp) {
1893 			bits = tstp->ls_flags;
1894 			bits >>= NFSLCK_SHIFT;
1895 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1896 			    KASSERT(vnode_unlocked == 0,
1897 				("nfsrv_lockctrl: vnode unlocked1"));
1898 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1899 				vp, p);
1900 			    if (ret == 1) {
1901 				/*
1902 				* nfsrv_clientconflict unlocks state
1903 				 * when it returns non-zero.
1904 				 */
1905 				lckstp = NULL;
1906 				goto tryagain;
1907 			    }
1908 			    if (ret == 0)
1909 				NFSUNLOCKSTATE();
1910 			    if (ret == 2)
1911 				error = NFSERR_PERM;
1912 			    else
1913 				error = NFSERR_OPENMODE;
1914 			    goto out;
1915 			}
1916 		    }
1917 		}
1918 
1919 		/* We're outta here */
1920 		NFSUNLOCKSTATE();
1921 		goto out;
1922 	    }
1923 	}
1924 
1925 	/*
1926 	 * For setattr, just get rid of all the Delegations for other clients.
1927 	 */
1928 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1929 		KASSERT(vnode_unlocked == 0,
1930 		    ("nfsrv_lockctrl: vnode unlocked2"));
1931 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1932 		if (ret) {
1933 			/*
1934 			 * nfsrv_cleandeleg() unlocks state when it
1935 			 * returns non-zero.
1936 			 */
1937 			if (ret == -1) {
1938 				lckstp = NULL;
1939 				goto tryagain;
1940 			}
1941 			error = ret;
1942 			goto out;
1943 		}
1944 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1945 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1946 		     LIST_EMPTY(&lfp->lf_deleg))) {
1947 			NFSUNLOCKSTATE();
1948 			goto out;
1949 		}
1950 	}
1951 
1952 	/*
1953 	 * Check for a conflicting delegation. If one is found, call
1954 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1955 	 * been set yet, it will get the lock. Otherwise, it will recall
1956 	 * the delegation. Then, we try try again...
1957 	 * I currently believe the conflict algorithm to be:
1958 	 * For Lock Ops (Lock/LockT/LockU)
1959 	 * - there is a conflict iff a different client has a write delegation
1960 	 * For Reading (Read Op)
1961 	 * - there is a conflict iff a different client has a write delegation
1962 	 *   (the specialids are always a different client)
1963 	 * For Writing (Write/Setattr of size)
1964 	 * - there is a conflict if a different client has any delegation
1965 	 * - there is a conflict if the same client has a read delegation
1966 	 *   (I don't understand why this isn't allowed, but that seems to be
1967 	 *    the current concensus?)
1968 	 */
1969 	tstp = LIST_FIRST(&lfp->lf_deleg);
1970 	while (tstp != LIST_END(&lfp->lf_deleg)) {
1971 	    nstp = LIST_NEXT(tstp, ls_file);
1972 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1973 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1974 		  (new_lop->lo_flags & NFSLCK_READ))) &&
1975 		  clp != tstp->ls_clp &&
1976 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1977 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1978 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
1979 		  (clp != tstp->ls_clp ||
1980 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1981 		ret = 0;
1982 		if (filestruct_locked != 0) {
1983 			/* Roll back local locks. */
1984 			NFSUNLOCKSTATE();
1985 			if (vnode_unlocked == 0) {
1986 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
1987 				NFSVOPUNLOCK(vp, 0);
1988 			}
1989 			nfsrv_locallock_rollback(vp, lfp, p);
1990 			NFSLOCKSTATE();
1991 			nfsrv_unlocklf(lfp);
1992 			NFSUNLOCKSTATE();
1993 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1994 			vnode_unlocked = 0;
1995 			if ((vp->v_iflag & VI_DOOMED) != 0)
1996 				ret = NFSERR_SERVERFAULT;
1997 			NFSLOCKSTATE();
1998 		}
1999 		if (ret == 0)
2000 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2001 		if (ret) {
2002 		    /*
2003 		     * nfsrv_delegconflict unlocks state when it
2004 		     * returns non-zero, which it always does.
2005 		     */
2006 		    if (other_lop) {
2007 			FREE((caddr_t)other_lop, M_NFSDLOCK);
2008 			other_lop = NULL;
2009 		    }
2010 		    if (ret == -1) {
2011 			lckstp = NULL;
2012 			goto tryagain;
2013 		    }
2014 		    error = ret;
2015 		    goto out;
2016 		}
2017 		/* Never gets here. */
2018 	    }
2019 	    tstp = nstp;
2020 	}
2021 
2022 	/*
2023 	 * Handle the unlock case by calling nfsrv_updatelock().
2024 	 * (Should I have done some access checking above for unlock? For now,
2025 	 *  just let it happen.)
2026 	 */
2027 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2028 		first = new_lop->lo_first;
2029 		end = new_lop->lo_end;
2030 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2031 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2032 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2033 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2034 		stateidp->other[0] = stp->ls_stateid.other[0];
2035 		stateidp->other[1] = stp->ls_stateid.other[1];
2036 		stateidp->other[2] = stp->ls_stateid.other[2];
2037 		if (filestruct_locked != 0) {
2038 			NFSUNLOCKSTATE();
2039 			if (vnode_unlocked == 0) {
2040 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2041 				vnode_unlocked = 1;
2042 				NFSVOPUNLOCK(vp, 0);
2043 			}
2044 			/* Update the local locks. */
2045 			nfsrv_localunlock(vp, lfp, first, end, p);
2046 			NFSLOCKSTATE();
2047 			nfsrv_unlocklf(lfp);
2048 		}
2049 		NFSUNLOCKSTATE();
2050 		goto out;
2051 	}
2052 
2053 	/*
2054 	 * Search for a conflicting lock. A lock conflicts if:
2055 	 * - the lock range overlaps and
2056 	 * - at least one lock is a write lock and
2057 	 * - it is not owned by the same lock owner
2058 	 */
2059 	if (!delegation) {
2060 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2061 	    if (new_lop->lo_end > lop->lo_first &&
2062 		new_lop->lo_first < lop->lo_end &&
2063 		(new_lop->lo_flags == NFSLCK_WRITE ||
2064 		 lop->lo_flags == NFSLCK_WRITE) &&
2065 		lckstp != lop->lo_stp &&
2066 		(clp != lop->lo_stp->ls_clp ||
2067 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2068 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2069 		    lckstp->ls_ownerlen))) {
2070 		if (other_lop) {
2071 		    FREE((caddr_t)other_lop, M_NFSDLOCK);
2072 		    other_lop = NULL;
2073 		}
2074 		if (vnode_unlocked != 0)
2075 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2076 			NULL, p);
2077 		else
2078 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2079 			vp, p);
2080 		if (ret == 1) {
2081 		    if (filestruct_locked != 0) {
2082 			if (vnode_unlocked == 0) {
2083 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2084 				NFSVOPUNLOCK(vp, 0);
2085 			}
2086 			/* Roll back local locks. */
2087 			nfsrv_locallock_rollback(vp, lfp, p);
2088 			NFSLOCKSTATE();
2089 			nfsrv_unlocklf(lfp);
2090 			NFSUNLOCKSTATE();
2091 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2092 			vnode_unlocked = 0;
2093 			if ((vp->v_iflag & VI_DOOMED) != 0) {
2094 				error = NFSERR_SERVERFAULT;
2095 				goto out;
2096 			}
2097 		    }
2098 		    /*
2099 		     * nfsrv_clientconflict() unlocks state when it
2100 		     * returns non-zero.
2101 		     */
2102 		    lckstp = NULL;
2103 		    goto tryagain;
2104 		}
2105 		/*
2106 		 * Found a conflicting lock, so record the conflict and
2107 		 * return the error.
2108 		 */
2109 		if (cfp != NULL && ret == 0) {
2110 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2111 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2112 		    cfp->cl_first = lop->lo_first;
2113 		    cfp->cl_end = lop->lo_end;
2114 		    cfp->cl_flags = lop->lo_flags;
2115 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2116 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2117 			cfp->cl_ownerlen);
2118 		}
2119 		if (ret == 2)
2120 		    error = NFSERR_PERM;
2121 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2122 		    error = NFSERR_RECLAIMCONFLICT;
2123 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2124 		    error = NFSERR_LOCKED;
2125 		else
2126 		    error = NFSERR_DENIED;
2127 		if (filestruct_locked != 0 && ret == 0) {
2128 			/* Roll back local locks. */
2129 			NFSUNLOCKSTATE();
2130 			if (vnode_unlocked == 0) {
2131 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2132 				vnode_unlocked = 1;
2133 				NFSVOPUNLOCK(vp, 0);
2134 			}
2135 			nfsrv_locallock_rollback(vp, lfp, p);
2136 			NFSLOCKSTATE();
2137 			nfsrv_unlocklf(lfp);
2138 		}
2139 		if (ret == 0)
2140 			NFSUNLOCKSTATE();
2141 		goto out;
2142 	    }
2143 	  }
2144 	}
2145 
2146 	/*
2147 	 * We only get here if there was no lock that conflicted.
2148 	 */
2149 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2150 		NFSUNLOCKSTATE();
2151 		goto out;
2152 	}
2153 
2154 	/*
2155 	 * We only get here when we are creating or modifying a lock.
2156 	 * There are two variants:
2157 	 * - exist_lock_owner where lock_owner exists
2158 	 * - open_to_lock_owner with new lock_owner
2159 	 */
2160 	first = new_lop->lo_first;
2161 	end = new_lop->lo_end;
2162 	lock_flags = new_lop->lo_flags;
2163 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2164 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2165 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2166 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2167 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2168 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2169 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2170 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2171 	} else {
2172 		/*
2173 		 * The new open_to_lock_owner case.
2174 		 * Link the new nfsstate into the lists.
2175 		 */
2176 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2177 		nfsrvd_refcache(new_stp->ls_op);
2178 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2179 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2180 		    clp->lc_clientid.lval[0];
2181 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2182 		    clp->lc_clientid.lval[1];
2183 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2184 		    nfsrv_nextstateindex(clp);
2185 		new_stp->ls_clp = clp;
2186 		LIST_INIT(&new_stp->ls_lock);
2187 		new_stp->ls_openstp = stp;
2188 		new_stp->ls_lfp = lfp;
2189 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2190 		    lfp);
2191 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2192 		    new_stp, ls_hash);
2193 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2194 		*new_lopp = NULL;
2195 		*new_stpp = NULL;
2196 		newnfsstats.srvlockowners++;
2197 		nfsrv_openpluslock++;
2198 	}
2199 	if (filestruct_locked != 0) {
2200 		NFSUNLOCKSTATE();
2201 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2202 		NFSLOCKSTATE();
2203 		nfsrv_unlocklf(lfp);
2204 	}
2205 	NFSUNLOCKSTATE();
2206 
2207 out:
2208 	if (haslock) {
2209 		NFSLOCKV4ROOTMUTEX();
2210 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2211 		NFSUNLOCKV4ROOTMUTEX();
2212 	}
2213 	if (vnode_unlocked != 0) {
2214 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2215 		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2216 			error = NFSERR_SERVERFAULT;
2217 	}
2218 	if (other_lop)
2219 		FREE((caddr_t)other_lop, M_NFSDLOCK);
2220 	NFSEXITCODE2(error, nd);
2221 	return (error);
2222 }
2223 
2224 /*
2225  * Check for state errors for Open.
2226  * repstat is passed back out as an error if more critical errors
2227  * are not detected.
2228  */
2229 APPLESTATIC 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)2230 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2231     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2232     NFSPROC_T *p, int repstat)
2233 {
2234 	struct nfsstate *stp, *nstp;
2235 	struct nfsclient *clp;
2236 	struct nfsstate *ownerstp;
2237 	struct nfslockfile *lfp, *new_lfp;
2238 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2239 
2240 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2241 		readonly = 1;
2242 	/*
2243 	 * Check for restart conditions (client and server).
2244 	 */
2245 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2246 		&new_stp->ls_stateid, 0);
2247 	if (error)
2248 		goto out;
2249 
2250 	/*
2251 	 * Check for state resource limit exceeded.
2252 	 * Technically this should be SMP protected, but the worst
2253 	 * case error is "out by one or two" on the count when it
2254 	 * returns NFSERR_RESOURCE and the limit is just a rather
2255 	 * arbitrary high water mark, so no harm is done.
2256 	 */
2257 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2258 		error = NFSERR_RESOURCE;
2259 		goto out;
2260 	}
2261 
2262 tryagain:
2263 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2264 	    M_NFSDLOCKFILE, M_WAITOK);
2265 	if (vp)
2266 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2267 		    NULL, p);
2268 	NFSLOCKSTATE();
2269 	/*
2270 	 * Get the nfsclient structure.
2271 	 */
2272 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2273 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2274 
2275 	/*
2276 	 * Look up the open owner. See if it needs confirmation and
2277 	 * check the seq#, as required.
2278 	 */
2279 	if (!error)
2280 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2281 
2282 	if (!error && ownerstp) {
2283 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2284 		    new_stp->ls_op);
2285 		/*
2286 		 * If the OpenOwner hasn't been confirmed, assume the
2287 		 * old one was a replay and this one is ok.
2288 		 * See: RFC3530 Sec. 14.2.18.
2289 		 */
2290 		if (error == NFSERR_BADSEQID &&
2291 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2292 			error = 0;
2293 	}
2294 
2295 	/*
2296 	 * Check for grace.
2297 	 */
2298 	if (!error)
2299 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2300 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2301 		nfsrv_checkstable(clp))
2302 		error = NFSERR_NOGRACE;
2303 
2304 	/*
2305 	 * If none of the above errors occurred, let repstat be
2306 	 * returned.
2307 	 */
2308 	if (repstat && !error)
2309 		error = repstat;
2310 	if (error) {
2311 		NFSUNLOCKSTATE();
2312 		if (haslock) {
2313 			NFSLOCKV4ROOTMUTEX();
2314 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2315 			NFSUNLOCKV4ROOTMUTEX();
2316 		}
2317 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2318 		goto out;
2319 	}
2320 
2321 	/*
2322 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2323 	 * (This always happens on the first pass, so haslock must be 0.)
2324 	 */
2325 	if (vp == NULL) {
2326 		NFSUNLOCKSTATE();
2327 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2328 		goto out;
2329 	}
2330 
2331 	/*
2332 	 * Get the structure for the underlying file.
2333 	 */
2334 	if (getfhret)
2335 		error = getfhret;
2336 	else
2337 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2338 		    NULL, 0);
2339 	if (new_lfp)
2340 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2341 	if (error) {
2342 		NFSUNLOCKSTATE();
2343 		if (haslock) {
2344 			NFSLOCKV4ROOTMUTEX();
2345 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2346 			NFSUNLOCKV4ROOTMUTEX();
2347 		}
2348 		goto out;
2349 	}
2350 
2351 	/*
2352 	 * Search for a conflicting open/share.
2353 	 */
2354 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2355 	    /*
2356 	     * For Delegate_Cur, search for the matching Delegation,
2357 	     * which indicates no conflict.
2358 	     * An old delegation should have been recovered by the
2359 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2360 	     * it match and return NFSERR_EXPIRED. Should I let it
2361 	     * match?
2362 	     */
2363 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2364 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2365 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2366 		    stateidp->seqid == 0) ||
2367 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2368 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2369 			  NFSX_STATEIDOTHER))
2370 			break;
2371 	    }
2372 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2373 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2374 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2375 		NFSUNLOCKSTATE();
2376 		if (haslock) {
2377 			NFSLOCKV4ROOTMUTEX();
2378 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2379 			NFSUNLOCKV4ROOTMUTEX();
2380 		}
2381 		error = NFSERR_EXPIRED;
2382 		goto out;
2383 	    }
2384 	}
2385 
2386 	/*
2387 	 * Check for access/deny bit conflicts. I check for the same
2388 	 * owner as well, in case the client didn't bother.
2389 	 */
2390 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2391 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2392 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2393 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2394 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2395 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2396 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2397 			if (ret == 1) {
2398 				/*
2399 				 * nfsrv_clientconflict() unlocks
2400 				 * state when it returns non-zero.
2401 				 */
2402 				goto tryagain;
2403 			}
2404 			if (ret == 2)
2405 				error = NFSERR_PERM;
2406 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2407 				error = NFSERR_RECLAIMCONFLICT;
2408 			else
2409 				error = NFSERR_SHAREDENIED;
2410 			if (ret == 0)
2411 				NFSUNLOCKSTATE();
2412 			if (haslock) {
2413 				NFSLOCKV4ROOTMUTEX();
2414 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2415 				NFSUNLOCKV4ROOTMUTEX();
2416 			}
2417 			goto out;
2418 		}
2419 	}
2420 
2421 	/*
2422 	 * Check for a conflicting delegation. If one is found, call
2423 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2424 	 * been set yet, it will get the lock. Otherwise, it will recall
2425 	 * the delegation. Then, we try try again...
2426 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2427 	 *  isn't a conflict.)
2428 	 * I currently believe the conflict algorithm to be:
2429 	 * For Open with Read Access and Deny None
2430 	 * - there is a conflict iff a different client has a write delegation
2431 	 * For Open with other Write Access or any Deny except None
2432 	 * - there is a conflict if a different client has any delegation
2433 	 * - there is a conflict if the same client has a read delegation
2434 	 *   (The current concensus is that this last case should be
2435 	 *    considered a conflict since the client with a read delegation
2436 	 *    could have done an Open with ReadAccess and WriteDeny
2437 	 *    locally and then not have checked for the WriteDeny.)
2438 	 * Don't check for a Reclaim, since that will be dealt with
2439 	 * by nfsrv_openctrl().
2440 	 */
2441 	if (!(new_stp->ls_flags &
2442 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2443 	    stp = LIST_FIRST(&lfp->lf_deleg);
2444 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2445 		nstp = LIST_NEXT(stp, ls_file);
2446 		if ((readonly && stp->ls_clp != clp &&
2447 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2448 		    (!readonly && (stp->ls_clp != clp ||
2449 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2450 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2451 			if (ret) {
2452 			    /*
2453 			     * nfsrv_delegconflict() unlocks state
2454 			     * when it returns non-zero.
2455 			     */
2456 			    if (ret == -1)
2457 				goto tryagain;
2458 			    error = ret;
2459 			    goto out;
2460 			}
2461 		}
2462 		stp = nstp;
2463 	    }
2464 	}
2465 	NFSUNLOCKSTATE();
2466 	if (haslock) {
2467 		NFSLOCKV4ROOTMUTEX();
2468 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2469 		NFSUNLOCKV4ROOTMUTEX();
2470 	}
2471 
2472 out:
2473 	NFSEXITCODE2(error, nd);
2474 	return (error);
2475 }
2476 
2477 /*
2478  * Open control function to create/update open state for an open.
2479  */
2480 APPLESTATIC 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)2481 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2482     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2483     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2484     NFSPROC_T *p, u_quad_t filerev)
2485 {
2486 	struct nfsstate *new_stp = *new_stpp;
2487 	struct nfsstate *stp, *nstp;
2488 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2489 	struct nfslockfile *lfp, *new_lfp;
2490 	struct nfsclient *clp;
2491 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2492 	int readonly = 0, cbret = 1, getfhret = 0;
2493 
2494 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2495 		readonly = 1;
2496 	/*
2497 	 * Check for restart conditions (client and server).
2498 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2499 	 * If an error does show up, return NFSERR_EXPIRED, since the
2500 	 * the seqid# has already been incremented.
2501 	 */
2502 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2503 	    &new_stp->ls_stateid, 0);
2504 	if (error) {
2505 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2506 		    error);
2507 		error = NFSERR_EXPIRED;
2508 		goto out;
2509 	}
2510 
2511 tryagain:
2512 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2513 	    M_NFSDLOCKFILE, M_WAITOK);
2514 	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2515 	    M_NFSDSTATE, M_WAITOK);
2516 	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2517 	    M_NFSDSTATE, M_WAITOK);
2518 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2519 	    NULL, p);
2520 	NFSLOCKSTATE();
2521 	/*
2522 	 * Get the client structure. Since the linked lists could be changed
2523 	 * by other nfsd processes if this process does a tsleep(), one of
2524 	 * two things must be done.
2525 	 * 1 - don't tsleep()
2526 	 * or
2527 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2528 	 *     before using the lists, since this lock stops the other
2529 	 *     nfsd. This should only be used for rare cases, since it
2530 	 *     essentially single threads the nfsd.
2531 	 *     At this time, it is only done for cases where the stable
2532 	 *     storage file must be written prior to completion of state
2533 	 *     expiration.
2534 	 */
2535 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2536 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2537 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2538 	    clp->lc_program) {
2539 		/*
2540 		 * This happens on the first open for a client
2541 		 * that supports callbacks.
2542 		 */
2543 		NFSUNLOCKSTATE();
2544 		/*
2545 		 * Although nfsrv_docallback() will sleep, clp won't
2546 		 * go away, since they are only removed when the
2547 		 * nfsv4_lock() has blocked the nfsd threads. The
2548 		 * fields in clp can change, but having multiple
2549 		 * threads do this Null callback RPC should be
2550 		 * harmless.
2551 		 */
2552 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2553 		    NULL, 0, NULL, NULL, NULL, p);
2554 		NFSLOCKSTATE();
2555 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2556 		if (!cbret)
2557 			clp->lc_flags |= LCL_CALLBACKSON;
2558 	}
2559 
2560 	/*
2561 	 * Look up the open owner. See if it needs confirmation and
2562 	 * check the seq#, as required.
2563 	 */
2564 	if (!error)
2565 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2566 
2567 	if (error) {
2568 		NFSUNLOCKSTATE();
2569 		printf("Nfsd: openctrl unexpected state err=%d\n",
2570 			error);
2571 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2572 		free((caddr_t)new_open, M_NFSDSTATE);
2573 		free((caddr_t)new_deleg, M_NFSDSTATE);
2574 		if (haslock) {
2575 			NFSLOCKV4ROOTMUTEX();
2576 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2577 			NFSUNLOCKV4ROOTMUTEX();
2578 		}
2579 		error = NFSERR_EXPIRED;
2580 		goto out;
2581 	}
2582 
2583 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2584 		nfsrv_markstable(clp);
2585 
2586 	/*
2587 	 * Get the structure for the underlying file.
2588 	 */
2589 	if (getfhret)
2590 		error = getfhret;
2591 	else
2592 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2593 		    NULL, 0);
2594 	if (new_lfp)
2595 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2596 	if (error) {
2597 		NFSUNLOCKSTATE();
2598 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2599 		    error);
2600 		free((caddr_t)new_open, M_NFSDSTATE);
2601 		free((caddr_t)new_deleg, M_NFSDSTATE);
2602 		if (haslock) {
2603 			NFSLOCKV4ROOTMUTEX();
2604 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2605 			NFSUNLOCKV4ROOTMUTEX();
2606 		}
2607 		goto out;
2608 	}
2609 
2610 	/*
2611 	 * Search for a conflicting open/share.
2612 	 */
2613 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2614 	    /*
2615 	     * For Delegate_Cur, search for the matching Delegation,
2616 	     * which indicates no conflict.
2617 	     * An old delegation should have been recovered by the
2618 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2619 	     * it match and return NFSERR_EXPIRED. Should I let it
2620 	     * match?
2621 	     */
2622 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2623 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2624 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2625 		    stateidp->seqid == 0) ||
2626 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2627 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2628 			NFSX_STATEIDOTHER))
2629 			break;
2630 	    }
2631 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2632 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2633 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2634 		NFSUNLOCKSTATE();
2635 		printf("Nfsd openctrl unexpected expiry\n");
2636 		free((caddr_t)new_open, M_NFSDSTATE);
2637 		free((caddr_t)new_deleg, M_NFSDSTATE);
2638 		if (haslock) {
2639 			NFSLOCKV4ROOTMUTEX();
2640 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2641 			NFSUNLOCKV4ROOTMUTEX();
2642 		}
2643 		error = NFSERR_EXPIRED;
2644 		goto out;
2645 	    }
2646 
2647 	    /*
2648 	     * Don't issue a Delegation, since one already exists and
2649 	     * delay delegation timeout, as required.
2650 	     */
2651 	    delegate = 0;
2652 	    nfsrv_delaydelegtimeout(stp);
2653 	}
2654 
2655 	/*
2656 	 * Check for access/deny bit conflicts. I also check for the
2657 	 * same owner, since the client might not have bothered to check.
2658 	 * Also, note an open for the same file and owner, if found,
2659 	 * which is all we do here for Delegate_Cur, since conflict
2660 	 * checking is already done.
2661 	 */
2662 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2663 		if (ownerstp && stp->ls_openowner == ownerstp)
2664 			openstp = stp;
2665 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2666 		    /*
2667 		     * If another client has the file open, the only
2668 		     * delegation that can be issued is a Read delegation
2669 		     * and only if it is a Read open with Deny none.
2670 		     */
2671 		    if (clp != stp->ls_clp) {
2672 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2673 			    NFSLCK_READACCESS)
2674 			    writedeleg = 0;
2675 			else
2676 			    delegate = 0;
2677 		    }
2678 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2679 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2680 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2681 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2682 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2683 			if (ret == 1) {
2684 				/*
2685 				 * nfsrv_clientconflict() unlocks state
2686 				 * when it returns non-zero.
2687 				 */
2688 				free((caddr_t)new_open, M_NFSDSTATE);
2689 				free((caddr_t)new_deleg, M_NFSDSTATE);
2690 				openstp = NULL;
2691 				goto tryagain;
2692 			}
2693 			if (ret == 2)
2694 				error = NFSERR_PERM;
2695 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2696 				error = NFSERR_RECLAIMCONFLICT;
2697 			else
2698 				error = NFSERR_SHAREDENIED;
2699 			if (ret == 0)
2700 				NFSUNLOCKSTATE();
2701 			if (haslock) {
2702 				NFSLOCKV4ROOTMUTEX();
2703 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2704 				NFSUNLOCKV4ROOTMUTEX();
2705 			}
2706 			free((caddr_t)new_open, M_NFSDSTATE);
2707 			free((caddr_t)new_deleg, M_NFSDSTATE);
2708 			printf("nfsd openctrl unexpected client cnfl\n");
2709 			goto out;
2710 		    }
2711 		}
2712 	}
2713 
2714 	/*
2715 	 * Check for a conflicting delegation. If one is found, call
2716 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2717 	 * been set yet, it will get the lock. Otherwise, it will recall
2718 	 * the delegation. Then, we try try again...
2719 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2720 	 *  isn't a conflict.)
2721 	 * I currently believe the conflict algorithm to be:
2722 	 * For Open with Read Access and Deny None
2723 	 * - there is a conflict iff a different client has a write delegation
2724 	 * For Open with other Write Access or any Deny except None
2725 	 * - there is a conflict if a different client has any delegation
2726 	 * - there is a conflict if the same client has a read delegation
2727 	 *   (The current concensus is that this last case should be
2728 	 *    considered a conflict since the client with a read delegation
2729 	 *    could have done an Open with ReadAccess and WriteDeny
2730 	 *    locally and then not have checked for the WriteDeny.)
2731 	 */
2732 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2733 	    stp = LIST_FIRST(&lfp->lf_deleg);
2734 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2735 		nstp = LIST_NEXT(stp, ls_file);
2736 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2737 			writedeleg = 0;
2738 		else
2739 			delegate = 0;
2740 		if ((readonly && stp->ls_clp != clp &&
2741 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2742 		    (!readonly && (stp->ls_clp != clp ||
2743 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2744 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2745 			delegate = 2;
2746 		    } else {
2747 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2748 			if (ret) {
2749 			    /*
2750 			     * nfsrv_delegconflict() unlocks state
2751 			     * when it returns non-zero.
2752 			     */
2753 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2754 			    free((caddr_t)new_open, M_NFSDSTATE);
2755 			    free((caddr_t)new_deleg, M_NFSDSTATE);
2756 			    if (ret == -1) {
2757 				openstp = NULL;
2758 				goto tryagain;
2759 			    }
2760 			    error = ret;
2761 			    goto out;
2762 			}
2763 		    }
2764 		}
2765 		stp = nstp;
2766 	    }
2767 	}
2768 
2769 	/*
2770 	 * We only get here if there was no open that conflicted.
2771 	 * If an open for the owner exists, or in the access/deny bits.
2772 	 * Otherwise it is a new open. If the open_owner hasn't been
2773 	 * confirmed, replace the open with the new one needing confirmation,
2774 	 * otherwise add the open.
2775 	 */
2776 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2777 	    /*
2778 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2779 	     * a match. If found, just move the old delegation to the current
2780 	     * delegation list and issue open. If not found, return
2781 	     * NFSERR_EXPIRED.
2782 	     */
2783 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2784 		if (stp->ls_lfp == lfp) {
2785 		    /* Found it */
2786 		    if (stp->ls_clp != clp)
2787 			panic("olddeleg clp");
2788 		    LIST_REMOVE(stp, ls_list);
2789 		    LIST_REMOVE(stp, ls_hash);
2790 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2791 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2792 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2793 			clp->lc_clientid.lval[0];
2794 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2795 			clp->lc_clientid.lval[1];
2796 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2797 			nfsrv_nextstateindex(clp);
2798 		    stp->ls_compref = nd->nd_compref;
2799 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2800 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2801 			stp->ls_stateid), stp, ls_hash);
2802 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2803 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2804 		    else
2805 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2806 		    clp->lc_delegtime = NFSD_MONOSEC +
2807 			nfsrv_lease + NFSRV_LEASEDELTA;
2808 
2809 		    /*
2810 		     * Now, do the associated open.
2811 		     */
2812 		    new_open->ls_stateid.seqid = 1;
2813 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2814 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2815 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2816 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2817 			NFSLCK_OPEN;
2818 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2819 			new_open->ls_flags |= (NFSLCK_READACCESS |
2820 			    NFSLCK_WRITEACCESS);
2821 		    else
2822 			new_open->ls_flags |= NFSLCK_READACCESS;
2823 		    new_open->ls_uid = new_stp->ls_uid;
2824 		    new_open->ls_lfp = lfp;
2825 		    new_open->ls_clp = clp;
2826 		    LIST_INIT(&new_open->ls_open);
2827 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2828 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2829 			new_open, ls_hash);
2830 		    /*
2831 		     * and handle the open owner
2832 		     */
2833 		    if (ownerstp) {
2834 			new_open->ls_openowner = ownerstp;
2835 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2836 		    } else {
2837 			new_open->ls_openowner = new_stp;
2838 			new_stp->ls_flags = 0;
2839 			nfsrvd_refcache(new_stp->ls_op);
2840 			new_stp->ls_noopens = 0;
2841 			LIST_INIT(&new_stp->ls_open);
2842 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2843 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2844 			*new_stpp = NULL;
2845 			newnfsstats.srvopenowners++;
2846 			nfsrv_openpluslock++;
2847 		    }
2848 		    openstp = new_open;
2849 		    new_open = NULL;
2850 		    newnfsstats.srvopens++;
2851 		    nfsrv_openpluslock++;
2852 		    break;
2853 		}
2854 	    }
2855 	    if (stp == LIST_END(&clp->lc_olddeleg))
2856 		error = NFSERR_EXPIRED;
2857 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2858 	    /*
2859 	     * Scan to see that no delegation for this client and file
2860 	     * doesn't already exist.
2861 	     * There also shouldn't yet be an Open for this file and
2862 	     * openowner.
2863 	     */
2864 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2865 		if (stp->ls_clp == clp)
2866 		    break;
2867 	    }
2868 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2869 		/*
2870 		 * This is the Claim_Previous case with a delegation
2871 		 * type != Delegate_None.
2872 		 */
2873 		/*
2874 		 * First, add the delegation. (Although we must issue the
2875 		 * delegation, we can also ask for an immediate return.)
2876 		 */
2877 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2878 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2879 		    clp->lc_clientid.lval[0];
2880 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2881 		    clp->lc_clientid.lval[1];
2882 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2883 		    nfsrv_nextstateindex(clp);
2884 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2885 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2886 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2887 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2888 		} else {
2889 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2890 			NFSLCK_READACCESS);
2891 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2892 		}
2893 		new_deleg->ls_uid = new_stp->ls_uid;
2894 		new_deleg->ls_lfp = lfp;
2895 		new_deleg->ls_clp = clp;
2896 		new_deleg->ls_filerev = filerev;
2897 		new_deleg->ls_compref = nd->nd_compref;
2898 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2899 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2900 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2901 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2902 		new_deleg = NULL;
2903 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2904 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2905 		     LCL_CALLBACKSON ||
2906 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2907 		    !NFSVNO_DELEGOK(vp))
2908 		    *rflagsp |= NFSV4OPEN_RECALL;
2909 		newnfsstats.srvdelegates++;
2910 		nfsrv_openpluslock++;
2911 		nfsrv_delegatecnt++;
2912 
2913 		/*
2914 		 * Now, do the associated open.
2915 		 */
2916 		new_open->ls_stateid.seqid = 1;
2917 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2918 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2919 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2920 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2921 		    NFSLCK_OPEN;
2922 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2923 			new_open->ls_flags |= (NFSLCK_READACCESS |
2924 			    NFSLCK_WRITEACCESS);
2925 		else
2926 			new_open->ls_flags |= NFSLCK_READACCESS;
2927 		new_open->ls_uid = new_stp->ls_uid;
2928 		new_open->ls_lfp = lfp;
2929 		new_open->ls_clp = clp;
2930 		LIST_INIT(&new_open->ls_open);
2931 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2932 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2933 		   new_open, ls_hash);
2934 		/*
2935 		 * and handle the open owner
2936 		 */
2937 		if (ownerstp) {
2938 		    new_open->ls_openowner = ownerstp;
2939 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2940 		} else {
2941 		    new_open->ls_openowner = new_stp;
2942 		    new_stp->ls_flags = 0;
2943 		    nfsrvd_refcache(new_stp->ls_op);
2944 		    new_stp->ls_noopens = 0;
2945 		    LIST_INIT(&new_stp->ls_open);
2946 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2947 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2948 		    *new_stpp = NULL;
2949 		    newnfsstats.srvopenowners++;
2950 		    nfsrv_openpluslock++;
2951 		}
2952 		openstp = new_open;
2953 		new_open = NULL;
2954 		newnfsstats.srvopens++;
2955 		nfsrv_openpluslock++;
2956 	    } else {
2957 		error = NFSERR_RECLAIMCONFLICT;
2958 	    }
2959 	} else if (ownerstp) {
2960 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2961 		    /* Replace the open */
2962 		    if (ownerstp->ls_op)
2963 			nfsrvd_derefcache(ownerstp->ls_op);
2964 		    ownerstp->ls_op = new_stp->ls_op;
2965 		    nfsrvd_refcache(ownerstp->ls_op);
2966 		    ownerstp->ls_seq = new_stp->ls_seq;
2967 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2968 		    stp = LIST_FIRST(&ownerstp->ls_open);
2969 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2970 			NFSLCK_OPEN;
2971 		    stp->ls_stateid.seqid = 1;
2972 		    stp->ls_uid = new_stp->ls_uid;
2973 		    if (lfp != stp->ls_lfp) {
2974 			LIST_REMOVE(stp, ls_file);
2975 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2976 			stp->ls_lfp = lfp;
2977 		    }
2978 		    openstp = stp;
2979 		} else if (openstp) {
2980 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2981 		    openstp->ls_stateid.seqid++;
2982 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
2983 			openstp->ls_stateid.seqid == 0)
2984 			openstp->ls_stateid.seqid = 1;
2985 
2986 		    /*
2987 		     * This is where we can choose to issue a delegation.
2988 		     */
2989 		    if (delegate == 0 || writedeleg == 0 ||
2990 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
2991 			nfsrv_writedelegifpos == 0) ||
2992 			!NFSVNO_DELEGOK(vp) ||
2993 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
2994 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2995 			 LCL_CALLBACKSON)
2996 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
2997 		    else if (nfsrv_issuedelegs == 0 ||
2998 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
2999 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3000 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3001 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3002 		    else {
3003 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3004 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3005 			    = clp->lc_clientid.lval[0];
3006 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3007 			    = clp->lc_clientid.lval[1];
3008 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3009 			    = nfsrv_nextstateindex(clp);
3010 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3011 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3012 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3013 			new_deleg->ls_uid = new_stp->ls_uid;
3014 			new_deleg->ls_lfp = lfp;
3015 			new_deleg->ls_clp = clp;
3016 			new_deleg->ls_filerev = filerev;
3017 			new_deleg->ls_compref = nd->nd_compref;
3018 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3019 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3020 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3021 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3022 			new_deleg = NULL;
3023 			newnfsstats.srvdelegates++;
3024 			nfsrv_openpluslock++;
3025 			nfsrv_delegatecnt++;
3026 		    }
3027 		} else {
3028 		    new_open->ls_stateid.seqid = 1;
3029 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3030 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3031 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3032 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3033 			NFSLCK_OPEN;
3034 		    new_open->ls_uid = new_stp->ls_uid;
3035 		    new_open->ls_openowner = ownerstp;
3036 		    new_open->ls_lfp = lfp;
3037 		    new_open->ls_clp = clp;
3038 		    LIST_INIT(&new_open->ls_open);
3039 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3040 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3041 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3042 			new_open, ls_hash);
3043 		    openstp = new_open;
3044 		    new_open = NULL;
3045 		    newnfsstats.srvopens++;
3046 		    nfsrv_openpluslock++;
3047 
3048 		    /*
3049 		     * This is where we can choose to issue a delegation.
3050 		     */
3051 		    if (delegate == 0 || (writedeleg == 0 && readonly == 0) ||
3052 			!NFSVNO_DELEGOK(vp) ||
3053 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3054 			 LCL_CALLBACKSON)
3055 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3056 		    else if (nfsrv_issuedelegs == 0 ||
3057 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3058 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3059 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3060 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3061 		    else {
3062 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3063 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3064 			    = clp->lc_clientid.lval[0];
3065 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3066 			    = clp->lc_clientid.lval[1];
3067 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3068 			    = nfsrv_nextstateindex(clp);
3069 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3070 			    (nfsrv_writedelegifpos || !readonly) &&
3071 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3072 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3073 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3074 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3075 			} else {
3076 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3077 				NFSLCK_READACCESS);
3078 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3079 			}
3080 			new_deleg->ls_uid = new_stp->ls_uid;
3081 			new_deleg->ls_lfp = lfp;
3082 			new_deleg->ls_clp = clp;
3083 			new_deleg->ls_filerev = filerev;
3084 			new_deleg->ls_compref = nd->nd_compref;
3085 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3086 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3087 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3088 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3089 			new_deleg = NULL;
3090 			newnfsstats.srvdelegates++;
3091 			nfsrv_openpluslock++;
3092 			nfsrv_delegatecnt++;
3093 		    }
3094 		}
3095 	} else {
3096 		/*
3097 		 * New owner case. Start the open_owner sequence with a
3098 		 * Needs confirmation (unless a reclaim) and hang the
3099 		 * new open off it.
3100 		 */
3101 		new_open->ls_stateid.seqid = 1;
3102 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3103 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3104 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3105 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3106 		    NFSLCK_OPEN;
3107 		new_open->ls_uid = new_stp->ls_uid;
3108 		LIST_INIT(&new_open->ls_open);
3109 		new_open->ls_openowner = new_stp;
3110 		new_open->ls_lfp = lfp;
3111 		new_open->ls_clp = clp;
3112 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3113 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3114 			new_stp->ls_flags = 0;
3115 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3116 			/* NFSv4.1 never needs confirmation. */
3117 			new_stp->ls_flags = 0;
3118 
3119 			/*
3120 			 * This is where we can choose to issue a delegation.
3121 			 */
3122 			if (delegate && nfsrv_issuedelegs &&
3123 			    (writedeleg || readonly) &&
3124 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3125 			     LCL_CALLBACKSON &&
3126 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3127 			    NFSVNO_DELEGOK(vp) &&
3128 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3129 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3130 				new_deleg->ls_stateid.seqid =
3131 				    delegstateidp->seqid = 1;
3132 				new_deleg->ls_stateid.other[0] =
3133 				    delegstateidp->other[0]
3134 				    = clp->lc_clientid.lval[0];
3135 				new_deleg->ls_stateid.other[1] =
3136 				    delegstateidp->other[1]
3137 				    = clp->lc_clientid.lval[1];
3138 				new_deleg->ls_stateid.other[2] =
3139 				    delegstateidp->other[2]
3140 				    = nfsrv_nextstateindex(clp);
3141 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3142 				    (nfsrv_writedelegifpos || !readonly) &&
3143 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3144 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3145 				     0)) {
3146 					new_deleg->ls_flags =
3147 					    (NFSLCK_DELEGWRITE |
3148 					     NFSLCK_READACCESS |
3149 					     NFSLCK_WRITEACCESS);
3150 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3151 				} else {
3152 					new_deleg->ls_flags =
3153 					    (NFSLCK_DELEGREAD |
3154 					     NFSLCK_READACCESS);
3155 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3156 				}
3157 				new_deleg->ls_uid = new_stp->ls_uid;
3158 				new_deleg->ls_lfp = lfp;
3159 				new_deleg->ls_clp = clp;
3160 				new_deleg->ls_filerev = filerev;
3161 				new_deleg->ls_compref = nd->nd_compref;
3162 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3163 				    ls_file);
3164 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3165 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3166 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3167 				    ls_list);
3168 				new_deleg = NULL;
3169 				newnfsstats.srvdelegates++;
3170 				nfsrv_openpluslock++;
3171 				nfsrv_delegatecnt++;
3172 			}
3173 		} else {
3174 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3175 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3176 		}
3177 		nfsrvd_refcache(new_stp->ls_op);
3178 		new_stp->ls_noopens = 0;
3179 		LIST_INIT(&new_stp->ls_open);
3180 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3181 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3182 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3183 		    new_open, ls_hash);
3184 		openstp = new_open;
3185 		new_open = NULL;
3186 		*new_stpp = NULL;
3187 		newnfsstats.srvopens++;
3188 		nfsrv_openpluslock++;
3189 		newnfsstats.srvopenowners++;
3190 		nfsrv_openpluslock++;
3191 	}
3192 	if (!error) {
3193 		stateidp->seqid = openstp->ls_stateid.seqid;
3194 		stateidp->other[0] = openstp->ls_stateid.other[0];
3195 		stateidp->other[1] = openstp->ls_stateid.other[1];
3196 		stateidp->other[2] = openstp->ls_stateid.other[2];
3197 	}
3198 	NFSUNLOCKSTATE();
3199 	if (haslock) {
3200 		NFSLOCKV4ROOTMUTEX();
3201 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3202 		NFSUNLOCKV4ROOTMUTEX();
3203 	}
3204 	if (new_open)
3205 		FREE((caddr_t)new_open, M_NFSDSTATE);
3206 	if (new_deleg)
3207 		FREE((caddr_t)new_deleg, M_NFSDSTATE);
3208 
3209 out:
3210 	NFSEXITCODE2(error, nd);
3211 	return (error);
3212 }
3213 
3214 /*
3215  * Open update. Does the confirm, downgrade and close.
3216  */
3217 APPLESTATIC int
nfsrv_openupdate(vnode_t vp,struct nfsstate * new_stp,nfsquad_t clientid,nfsv4stateid_t * stateidp,struct nfsrv_descript * nd,NFSPROC_T * p)3218 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3219     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3220 {
3221 	struct nfsstate *stp, *ownerstp;
3222 	struct nfsclient *clp;
3223 	struct nfslockfile *lfp;
3224 	u_int32_t bits;
3225 	int error = 0, gotstate = 0, len = 0;
3226 	u_char client[NFSV4_OPAQUELIMIT];
3227 
3228 	/*
3229 	 * Check for restart conditions (client and server).
3230 	 */
3231 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3232 	    &new_stp->ls_stateid, 0);
3233 	if (error)
3234 		goto out;
3235 
3236 	NFSLOCKSTATE();
3237 	/*
3238 	 * Get the open structure via clientid and stateid.
3239 	 */
3240 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3241 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3242 	if (!error)
3243 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3244 		    new_stp->ls_flags, &stp);
3245 
3246 	/*
3247 	 * Sanity check the open.
3248 	 */
3249 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3250 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3251 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3252 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3253 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3254 		error = NFSERR_BADSTATEID;
3255 
3256 	if (!error)
3257 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3258 		    stp->ls_openowner, new_stp->ls_op);
3259 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3260 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3261 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3262 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3263 	      new_stp->ls_stateid.seqid != 0)))
3264 		error = NFSERR_OLDSTATEID;
3265 	if (!error && vnode_vtype(vp) != VREG) {
3266 		if (vnode_vtype(vp) == VDIR)
3267 			error = NFSERR_ISDIR;
3268 		else
3269 			error = NFSERR_INVAL;
3270 	}
3271 
3272 	if (error) {
3273 		/*
3274 		 * If a client tries to confirm an Open with a bad
3275 		 * seqid# and there are no byte range locks or other Opens
3276 		 * on the openowner, just throw it away, so the next use of the
3277 		 * openowner will start a fresh seq#.
3278 		 */
3279 		if (error == NFSERR_BADSEQID &&
3280 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3281 		    nfsrv_nootherstate(stp))
3282 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3283 		NFSUNLOCKSTATE();
3284 		goto out;
3285 	}
3286 
3287 	/*
3288 	 * Set the return stateid.
3289 	 */
3290 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3291 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3292 		stateidp->seqid = 1;
3293 	stateidp->other[0] = stp->ls_stateid.other[0];
3294 	stateidp->other[1] = stp->ls_stateid.other[1];
3295 	stateidp->other[2] = stp->ls_stateid.other[2];
3296 	/*
3297 	 * Now, handle the three cases.
3298 	 */
3299 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3300 		/*
3301 		 * If the open doesn't need confirmation, it seems to me that
3302 		 * there is a client error, but I'll just log it and keep going?
3303 		 */
3304 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3305 			printf("Nfsv4d: stray open confirm\n");
3306 		stp->ls_openowner->ls_flags = 0;
3307 		stp->ls_stateid.seqid++;
3308 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3309 		    stp->ls_stateid.seqid == 0)
3310 			stp->ls_stateid.seqid = 1;
3311 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3312 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3313 			len = clp->lc_idlen;
3314 			NFSBCOPY(clp->lc_id, client, len);
3315 			gotstate = 1;
3316 		}
3317 		NFSUNLOCKSTATE();
3318 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3319 		ownerstp = stp->ls_openowner;
3320 		lfp = stp->ls_lfp;
3321 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3322 			/* Get the lf lock */
3323 			nfsrv_locklf(lfp);
3324 			NFSUNLOCKSTATE();
3325 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3326 			NFSVOPUNLOCK(vp, 0);
3327 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3328 				NFSLOCKSTATE();
3329 				nfsrv_unlocklf(lfp);
3330 				NFSUNLOCKSTATE();
3331 			}
3332 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3333 		} else {
3334 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3335 			NFSUNLOCKSTATE();
3336 		}
3337 	} else {
3338 		/*
3339 		 * Update the share bits, making sure that the new set are a
3340 		 * subset of the old ones.
3341 		 */
3342 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3343 		if (~(stp->ls_flags) & bits) {
3344 			NFSUNLOCKSTATE();
3345 			error = NFSERR_INVAL;
3346 			goto out;
3347 		}
3348 		stp->ls_flags = (bits | NFSLCK_OPEN);
3349 		stp->ls_stateid.seqid++;
3350 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3351 		    stp->ls_stateid.seqid == 0)
3352 			stp->ls_stateid.seqid = 1;
3353 		NFSUNLOCKSTATE();
3354 	}
3355 
3356 	/*
3357 	 * If the client just confirmed its first open, write a timestamp
3358 	 * to the stable storage file.
3359 	 */
3360 	if (gotstate != 0) {
3361 		nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
3362 		nfsrv_backupstable();
3363 	}
3364 
3365 out:
3366 	NFSEXITCODE2(error, nd);
3367 	return (error);
3368 }
3369 
3370 /*
3371  * Delegation update. Does the purge and return.
3372  */
3373 APPLESTATIC int
nfsrv_delegupdate(struct nfsrv_descript * nd,nfsquad_t clientid,nfsv4stateid_t * stateidp,vnode_t vp,int op,struct ucred * cred,NFSPROC_T * p)3374 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3375     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3376     NFSPROC_T *p)
3377 {
3378 	struct nfsstate *stp;
3379 	struct nfsclient *clp;
3380 	int error = 0;
3381 	fhandle_t fh;
3382 
3383 	/*
3384 	 * Do a sanity check against the file handle for DelegReturn.
3385 	 */
3386 	if (vp) {
3387 		error = nfsvno_getfh(vp, &fh, p);
3388 		if (error)
3389 			goto out;
3390 	}
3391 	/*
3392 	 * Check for restart conditions (client and server).
3393 	 */
3394 	if (op == NFSV4OP_DELEGRETURN)
3395 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3396 			stateidp, 0);
3397 	else
3398 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3399 			stateidp, 0);
3400 
3401 	NFSLOCKSTATE();
3402 	/*
3403 	 * Get the open structure via clientid and stateid.
3404 	 */
3405 	if (!error)
3406 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3407 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3408 	if (error) {
3409 		if (error == NFSERR_CBPATHDOWN)
3410 			error = 0;
3411 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3412 			error = NFSERR_STALESTATEID;
3413 	}
3414 	if (!error && op == NFSV4OP_DELEGRETURN) {
3415 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3416 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3417 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3418 		error = NFSERR_OLDSTATEID;
3419 	}
3420 	/*
3421 	 * NFSERR_EXPIRED means that the state has gone away,
3422 	 * so Delegations have been purged. Just return ok.
3423 	 */
3424 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3425 		NFSUNLOCKSTATE();
3426 		error = 0;
3427 		goto out;
3428 	}
3429 	if (error) {
3430 		NFSUNLOCKSTATE();
3431 		goto out;
3432 	}
3433 
3434 	if (op == NFSV4OP_DELEGRETURN) {
3435 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3436 		    sizeof (fhandle_t))) {
3437 			NFSUNLOCKSTATE();
3438 			error = NFSERR_BADSTATEID;
3439 			goto out;
3440 		}
3441 		nfsrv_freedeleg(stp);
3442 	} else {
3443 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3444 	}
3445 	NFSUNLOCKSTATE();
3446 	error = 0;
3447 
3448 out:
3449 	NFSEXITCODE(error);
3450 	return (error);
3451 }
3452 
3453 /*
3454  * Release lock owner.
3455  */
3456 APPLESTATIC int
nfsrv_releaselckown(struct nfsstate * new_stp,nfsquad_t clientid,NFSPROC_T * p)3457 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3458     NFSPROC_T *p)
3459 {
3460 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3461 	struct nfsclient *clp;
3462 	int error = 0;
3463 
3464 	/*
3465 	 * Check for restart conditions (client and server).
3466 	 */
3467 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3468 	    &new_stp->ls_stateid, 0);
3469 	if (error)
3470 		goto out;
3471 
3472 	NFSLOCKSTATE();
3473 	/*
3474 	 * Get the lock owner by name.
3475 	 */
3476 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3477 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3478 	if (error) {
3479 		NFSUNLOCKSTATE();
3480 		goto out;
3481 	}
3482 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3483 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3484 		stp = LIST_FIRST(&openstp->ls_open);
3485 		while (stp != LIST_END(&openstp->ls_open)) {
3486 		    nstp = LIST_NEXT(stp, ls_list);
3487 		    /*
3488 		     * If the owner matches, check for locks and
3489 		     * then free or return an error.
3490 		     */
3491 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3492 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3493 			 stp->ls_ownerlen)){
3494 			if (LIST_EMPTY(&stp->ls_lock)) {
3495 			    nfsrv_freelockowner(stp, NULL, 0, p);
3496 			} else {
3497 			    NFSUNLOCKSTATE();
3498 			    error = NFSERR_LOCKSHELD;
3499 			    goto out;
3500 			}
3501 		    }
3502 		    stp = nstp;
3503 		}
3504 	    }
3505 	}
3506 	NFSUNLOCKSTATE();
3507 
3508 out:
3509 	NFSEXITCODE(error);
3510 	return (error);
3511 }
3512 
3513 /*
3514  * Get the file handle for a lock structure.
3515  */
3516 static int
nfsrv_getlockfh(vnode_t vp,u_short flags,struct nfslockfile * new_lfp,fhandle_t * nfhp,NFSPROC_T * p)3517 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3518     fhandle_t *nfhp, NFSPROC_T *p)
3519 {
3520 	fhandle_t *fhp = NULL;
3521 	int error;
3522 
3523 	/*
3524 	 * For lock, use the new nfslock structure, otherwise just
3525 	 * a fhandle_t on the stack.
3526 	 */
3527 	if (flags & NFSLCK_OPEN) {
3528 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3529 		fhp = &new_lfp->lf_fh;
3530 	} else if (nfhp) {
3531 		fhp = nfhp;
3532 	} else {
3533 		panic("nfsrv_getlockfh");
3534 	}
3535 	error = nfsvno_getfh(vp, fhp, p);
3536 	NFSEXITCODE(error);
3537 	return (error);
3538 }
3539 
3540 /*
3541  * Get an nfs lock structure. Allocate one, as required, and return a
3542  * pointer to it.
3543  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3544  */
3545 static int
nfsrv_getlockfile(u_short flags,struct nfslockfile ** new_lfpp,struct nfslockfile ** lfpp,fhandle_t * nfhp,int lockit)3546 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3547     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3548 {
3549 	struct nfslockfile *lfp;
3550 	fhandle_t *fhp = NULL, *tfhp;
3551 	struct nfslockhashhead *hp;
3552 	struct nfslockfile *new_lfp = NULL;
3553 
3554 	/*
3555 	 * For lock, use the new nfslock structure, otherwise just
3556 	 * a fhandle_t on the stack.
3557 	 */
3558 	if (flags & NFSLCK_OPEN) {
3559 		new_lfp = *new_lfpp;
3560 		fhp = &new_lfp->lf_fh;
3561 	} else if (nfhp) {
3562 		fhp = nfhp;
3563 	} else {
3564 		panic("nfsrv_getlockfile");
3565 	}
3566 
3567 	hp = NFSLOCKHASH(fhp);
3568 	LIST_FOREACH(lfp, hp, lf_hash) {
3569 		tfhp = &lfp->lf_fh;
3570 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3571 			if (lockit)
3572 				nfsrv_locklf(lfp);
3573 			*lfpp = lfp;
3574 			return (0);
3575 		}
3576 	}
3577 	if (!(flags & NFSLCK_OPEN))
3578 		return (-1);
3579 
3580 	/*
3581 	 * No match, so chain the new one into the list.
3582 	 */
3583 	LIST_INIT(&new_lfp->lf_open);
3584 	LIST_INIT(&new_lfp->lf_lock);
3585 	LIST_INIT(&new_lfp->lf_deleg);
3586 	LIST_INIT(&new_lfp->lf_locallock);
3587 	LIST_INIT(&new_lfp->lf_rollback);
3588 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3589 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3590 	new_lfp->lf_usecount = 0;
3591 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3592 	*lfpp = new_lfp;
3593 	*new_lfpp = NULL;
3594 	return (0);
3595 }
3596 
3597 /*
3598  * This function adds a nfslock lock structure to the list for the associated
3599  * nfsstate and nfslockfile structures. It will be inserted after the
3600  * entry pointed at by insert_lop.
3601  */
3602 static void
nfsrv_insertlock(struct nfslock * new_lop,struct nfslock * insert_lop,struct nfsstate * stp,struct nfslockfile * lfp)3603 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3604     struct nfsstate *stp, struct nfslockfile *lfp)
3605 {
3606 	struct nfslock *lop, *nlop;
3607 
3608 	new_lop->lo_stp = stp;
3609 	new_lop->lo_lfp = lfp;
3610 
3611 	if (stp != NULL) {
3612 		/* Insert in increasing lo_first order */
3613 		lop = LIST_FIRST(&lfp->lf_lock);
3614 		if (lop == LIST_END(&lfp->lf_lock) ||
3615 		    new_lop->lo_first <= lop->lo_first) {
3616 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3617 		} else {
3618 			nlop = LIST_NEXT(lop, lo_lckfile);
3619 			while (nlop != LIST_END(&lfp->lf_lock) &&
3620 			       nlop->lo_first < new_lop->lo_first) {
3621 				lop = nlop;
3622 				nlop = LIST_NEXT(lop, lo_lckfile);
3623 			}
3624 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3625 		}
3626 	} else {
3627 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3628 	}
3629 
3630 	/*
3631 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3632 	 * an empty list.
3633 	 */
3634 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3635 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3636 	else if ((struct nfsstate *)insert_lop == stp)
3637 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3638 	else
3639 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3640 	if (stp != NULL) {
3641 		newnfsstats.srvlocks++;
3642 		nfsrv_openpluslock++;
3643 	}
3644 }
3645 
3646 /*
3647  * This function updates the locking for a lock owner and given file. It
3648  * maintains a list of lock ranges ordered on increasing file offset that
3649  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3650  * It always adds new_lop to the list and sometimes uses the one pointed
3651  * at by other_lopp.
3652  */
3653 static void
nfsrv_updatelock(struct nfsstate * stp,struct nfslock ** new_lopp,struct nfslock ** other_lopp,struct nfslockfile * lfp)3654 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3655     struct nfslock **other_lopp, struct nfslockfile *lfp)
3656 {
3657 	struct nfslock *new_lop = *new_lopp;
3658 	struct nfslock *lop, *tlop, *ilop;
3659 	struct nfslock *other_lop = *other_lopp;
3660 	int unlock = 0, myfile = 0;
3661 	u_int64_t tmp;
3662 
3663 	/*
3664 	 * Work down the list until the lock is merged.
3665 	 */
3666 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3667 		unlock = 1;
3668 	if (stp != NULL) {
3669 		ilop = (struct nfslock *)stp;
3670 		lop = LIST_FIRST(&stp->ls_lock);
3671 	} else {
3672 		ilop = (struct nfslock *)lfp;
3673 		lop = LIST_FIRST(&lfp->lf_locallock);
3674 	}
3675 	while (lop != NULL) {
3676 	    /*
3677 	     * Only check locks for this file that aren't before the start of
3678 	     * new lock's range.
3679 	     */
3680 	    if (lop->lo_lfp == lfp) {
3681 	      myfile = 1;
3682 	      if (lop->lo_end >= new_lop->lo_first) {
3683 		if (new_lop->lo_end < lop->lo_first) {
3684 			/*
3685 			 * If the new lock ends before the start of the
3686 			 * current lock's range, no merge, just insert
3687 			 * the new lock.
3688 			 */
3689 			break;
3690 		}
3691 		if (new_lop->lo_flags == lop->lo_flags ||
3692 		    (new_lop->lo_first <= lop->lo_first &&
3693 		     new_lop->lo_end >= lop->lo_end)) {
3694 			/*
3695 			 * This lock can be absorbed by the new lock/unlock.
3696 			 * This happens when it covers the entire range
3697 			 * of the old lock or is contiguous
3698 			 * with the old lock and is of the same type or an
3699 			 * unlock.
3700 			 */
3701 			if (lop->lo_first < new_lop->lo_first)
3702 				new_lop->lo_first = lop->lo_first;
3703 			if (lop->lo_end > new_lop->lo_end)
3704 				new_lop->lo_end = lop->lo_end;
3705 			tlop = lop;
3706 			lop = LIST_NEXT(lop, lo_lckowner);
3707 			nfsrv_freenfslock(tlop);
3708 			continue;
3709 		}
3710 
3711 		/*
3712 		 * All these cases are for contiguous locks that are not the
3713 		 * same type, so they can't be merged.
3714 		 */
3715 		if (new_lop->lo_first <= lop->lo_first) {
3716 			/*
3717 			 * This case is where the new lock overlaps with the
3718 			 * first part of the old lock. Move the start of the
3719 			 * old lock to just past the end of the new lock. The
3720 			 * new lock will be inserted in front of the old, since
3721 			 * ilop hasn't been updated. (We are done now.)
3722 			 */
3723 			lop->lo_first = new_lop->lo_end;
3724 			break;
3725 		}
3726 		if (new_lop->lo_end >= lop->lo_end) {
3727 			/*
3728 			 * This case is where the new lock overlaps with the
3729 			 * end of the old lock's range. Move the old lock's
3730 			 * end to just before the new lock's first and insert
3731 			 * the new lock after the old lock.
3732 			 * Might not be done yet, since the new lock could
3733 			 * overlap further locks with higher ranges.
3734 			 */
3735 			lop->lo_end = new_lop->lo_first;
3736 			ilop = lop;
3737 			lop = LIST_NEXT(lop, lo_lckowner);
3738 			continue;
3739 		}
3740 		/*
3741 		 * The final case is where the new lock's range is in the
3742 		 * middle of the current lock's and splits the current lock
3743 		 * up. Use *other_lopp to handle the second part of the
3744 		 * split old lock range. (We are done now.)
3745 		 * For unlock, we use new_lop as other_lop and tmp, since
3746 		 * other_lop and new_lop are the same for this case.
3747 		 * We noted the unlock case above, so we don't need
3748 		 * new_lop->lo_flags any longer.
3749 		 */
3750 		tmp = new_lop->lo_first;
3751 		if (other_lop == NULL) {
3752 			if (!unlock)
3753 				panic("nfsd srv update unlock");
3754 			other_lop = new_lop;
3755 			*new_lopp = NULL;
3756 		}
3757 		other_lop->lo_first = new_lop->lo_end;
3758 		other_lop->lo_end = lop->lo_end;
3759 		other_lop->lo_flags = lop->lo_flags;
3760 		other_lop->lo_stp = stp;
3761 		other_lop->lo_lfp = lfp;
3762 		lop->lo_end = tmp;
3763 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3764 		*other_lopp = NULL;
3765 		ilop = lop;
3766 		break;
3767 	      }
3768 	    }
3769 	    ilop = lop;
3770 	    lop = LIST_NEXT(lop, lo_lckowner);
3771 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3772 		break;
3773 	}
3774 
3775 	/*
3776 	 * Insert the new lock in the list at the appropriate place.
3777 	 */
3778 	if (!unlock) {
3779 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3780 		*new_lopp = NULL;
3781 	}
3782 }
3783 
3784 /*
3785  * This function handles sequencing of locks, etc.
3786  * It returns an error that indicates what the caller should do.
3787  */
3788 static int
nfsrv_checkseqid(struct nfsrv_descript * nd,u_int32_t seqid,struct nfsstate * stp,struct nfsrvcache * op)3789 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3790     struct nfsstate *stp, struct nfsrvcache *op)
3791 {
3792 	int error = 0;
3793 
3794 	if ((nd->nd_flag & ND_NFSV41) != 0)
3795 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3796 		goto out;
3797 	if (op != nd->nd_rp)
3798 		panic("nfsrvstate checkseqid");
3799 	if (!(op->rc_flag & RC_INPROG))
3800 		panic("nfsrvstate not inprog");
3801 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3802 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3803 		panic("nfsrvstate op refcnt");
3804 	}
3805 	if ((stp->ls_seq + 1) == seqid) {
3806 		if (stp->ls_op)
3807 			nfsrvd_derefcache(stp->ls_op);
3808 		stp->ls_op = op;
3809 		nfsrvd_refcache(op);
3810 		stp->ls_seq = seqid;
3811 		goto out;
3812 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3813 		op->rc_xid == stp->ls_op->rc_xid &&
3814 		op->rc_refcnt == 0 &&
3815 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3816 		op->rc_cksum == stp->ls_op->rc_cksum) {
3817 		if (stp->ls_op->rc_flag & RC_INPROG) {
3818 			error = NFSERR_DONTREPLY;
3819 			goto out;
3820 		}
3821 		nd->nd_rp = stp->ls_op;
3822 		nd->nd_rp->rc_flag |= RC_INPROG;
3823 		nfsrvd_delcache(op);
3824 		error = NFSERR_REPLYFROMCACHE;
3825 		goto out;
3826 	}
3827 	error = NFSERR_BADSEQID;
3828 
3829 out:
3830 	NFSEXITCODE2(error, nd);
3831 	return (error);
3832 }
3833 
3834 /*
3835  * Get the client ip address for callbacks. If the strings can't be parsed,
3836  * just set lc_program to 0 to indicate no callbacks are possible.
3837  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3838  *  the address to the client's transport address. This won't be used
3839  *  for callbacks, but can be printed out by newnfsstats for info.)
3840  * Return error if the xdr can't be parsed, 0 otherwise.
3841  */
3842 APPLESTATIC int
nfsrv_getclientipaddr(struct nfsrv_descript * nd,struct nfsclient * clp)3843 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3844 {
3845 	u_int32_t *tl;
3846 	u_char *cp, *cp2;
3847 	int i, j;
3848 	struct sockaddr_in *rad, *sad;
3849 	u_char protocol[5], addr[24];
3850 	int error = 0, cantparse = 0;
3851 	union {
3852 		u_long ival;
3853 		u_char cval[4];
3854 	} ip;
3855 	union {
3856 		u_short sval;
3857 		u_char cval[2];
3858 	} port;
3859 
3860 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3861 	rad->sin_family = AF_INET;
3862 	rad->sin_len = sizeof (struct sockaddr_in);
3863 	rad->sin_addr.s_addr = 0;
3864 	rad->sin_port = 0;
3865 	clp->lc_req.nr_client = NULL;
3866 	clp->lc_req.nr_lock = 0;
3867 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3868 	i = fxdr_unsigned(int, *tl);
3869 	if (i >= 3 && i <= 4) {
3870 		error = nfsrv_mtostr(nd, protocol, i);
3871 		if (error)
3872 			goto nfsmout;
3873 		if (!strcmp(protocol, "tcp")) {
3874 			clp->lc_flags |= LCL_TCPCALLBACK;
3875 			clp->lc_req.nr_sotype = SOCK_STREAM;
3876 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3877 		} else if (!strcmp(protocol, "udp")) {
3878 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3879 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3880 		} else {
3881 			cantparse = 1;
3882 		}
3883 	} else {
3884 		cantparse = 1;
3885 		if (i > 0) {
3886 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3887 			if (error)
3888 				goto nfsmout;
3889 		}
3890 	}
3891 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3892 	i = fxdr_unsigned(int, *tl);
3893 	if (i < 0) {
3894 		error = NFSERR_BADXDR;
3895 		goto nfsmout;
3896 	} else if (i == 0) {
3897 		cantparse = 1;
3898 	} else if (!cantparse && i <= 23 && i >= 11) {
3899 		error = nfsrv_mtostr(nd, addr, i);
3900 		if (error)
3901 			goto nfsmout;
3902 
3903 		/*
3904 		 * Parse out the address fields. We expect 6 decimal numbers
3905 		 * separated by '.'s.
3906 		 */
3907 		cp = addr;
3908 		i = 0;
3909 		while (*cp && i < 6) {
3910 			cp2 = cp;
3911 			while (*cp2 && *cp2 != '.')
3912 				cp2++;
3913 			if (*cp2)
3914 				*cp2++ = '\0';
3915 			else if (i != 5) {
3916 				cantparse = 1;
3917 				break;
3918 			}
3919 			j = nfsrv_getipnumber(cp);
3920 			if (j >= 0) {
3921 				if (i < 4)
3922 					ip.cval[3 - i] = j;
3923 				else
3924 					port.cval[5 - i] = j;
3925 			} else {
3926 				cantparse = 1;
3927 				break;
3928 			}
3929 			cp = cp2;
3930 			i++;
3931 		}
3932 		if (!cantparse) {
3933 			if (ip.ival != 0x0) {
3934 				rad->sin_addr.s_addr = htonl(ip.ival);
3935 				rad->sin_port = htons(port.sval);
3936 			} else {
3937 				cantparse = 1;
3938 			}
3939 		}
3940 	} else {
3941 		cantparse = 1;
3942 		if (i > 0) {
3943 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3944 			if (error)
3945 				goto nfsmout;
3946 		}
3947 	}
3948 	if (cantparse) {
3949 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3950 		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3951 		rad->sin_port = 0x0;
3952 		clp->lc_program = 0;
3953 	}
3954 nfsmout:
3955 	NFSEXITCODE2(error, nd);
3956 	return (error);
3957 }
3958 
3959 /*
3960  * Turn a string of up to three decimal digits into a number. Return -1 upon
3961  * error.
3962  */
3963 static int
nfsrv_getipnumber(u_char * cp)3964 nfsrv_getipnumber(u_char *cp)
3965 {
3966 	int i = 0, j = 0;
3967 
3968 	while (*cp) {
3969 		if (j > 2 || *cp < '0' || *cp > '9')
3970 			return (-1);
3971 		i *= 10;
3972 		i += (*cp - '0');
3973 		cp++;
3974 		j++;
3975 	}
3976 	if (i < 256)
3977 		return (i);
3978 	return (-1);
3979 }
3980 
3981 /*
3982  * This function checks for restart conditions.
3983  */
3984 static int
nfsrv_checkrestart(nfsquad_t clientid,u_int32_t flags,nfsv4stateid_t * stateidp,int specialid)3985 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
3986     nfsv4stateid_t *stateidp, int specialid)
3987 {
3988 	int ret = 0;
3989 
3990 	/*
3991 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
3992 	 * and DelegPurge have a clientid, the rest a stateid.
3993 	 */
3994 	if (flags &
3995 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
3996 		if (clientid.lval[0] != nfsrvboottime) {
3997 			ret = NFSERR_STALECLIENTID;
3998 			goto out;
3999 		}
4000 	} else if (stateidp->other[0] != nfsrvboottime &&
4001 		specialid == 0) {
4002 		ret = NFSERR_STALESTATEID;
4003 		goto out;
4004 	}
4005 
4006 	/*
4007 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4008 	 * not use a lock/open owner seqid#, so the check can be done now.
4009 	 * (The others will be checked, as required, later.)
4010 	 */
4011 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4012 		goto out;
4013 
4014 	NFSLOCKSTATE();
4015 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4016 	NFSUNLOCKSTATE();
4017 
4018 out:
4019 	NFSEXITCODE(ret);
4020 	return (ret);
4021 }
4022 
4023 /*
4024  * Check for grace.
4025  */
4026 static int
nfsrv_checkgrace(struct nfsrv_descript * nd,struct nfsclient * clp,u_int32_t flags)4027 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4028     u_int32_t flags)
4029 {
4030 	int error = 0;
4031 
4032 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4033 		if (flags & NFSLCK_RECLAIM) {
4034 			error = NFSERR_NOGRACE;
4035 			goto out;
4036 		}
4037 	} else {
4038 		if (!(flags & NFSLCK_RECLAIM)) {
4039 			error = NFSERR_GRACE;
4040 			goto out;
4041 		}
4042 		if (nd != NULL && clp != NULL &&
4043 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4044 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4045 			error = NFSERR_NOGRACE;
4046 			goto out;
4047 		}
4048 
4049 		/*
4050 		 * If grace is almost over and we are still getting Reclaims,
4051 		 * extend grace a bit.
4052 		 */
4053 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4054 		    nfsrv_stablefirst.nsf_eograce)
4055 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4056 				NFSRV_LEASEDELTA;
4057 	}
4058 
4059 out:
4060 	NFSEXITCODE(error);
4061 	return (error);
4062 }
4063 
4064 /*
4065  * Do a server callback.
4066  */
4067 static int
nfsrv_docallback(struct nfsclient * clp,int procnum,nfsv4stateid_t * stateidp,int trunc,fhandle_t * fhp,struct nfsvattr * nap,nfsattrbit_t * attrbitp,NFSPROC_T * p)4068 nfsrv_docallback(struct nfsclient *clp, int procnum,
4069     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
4070     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
4071 {
4072 	mbuf_t m;
4073 	u_int32_t *tl;
4074 	struct nfsrv_descript nfsd, *nd = &nfsd;
4075 	struct ucred *cred;
4076 	int error = 0;
4077 	u_int32_t callback;
4078 	struct nfsdsession *sep = NULL;
4079 
4080 	cred = newnfs_getcred();
4081 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4082 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4083 		NFSUNLOCKSTATE();
4084 		panic("docallb");
4085 	}
4086 	clp->lc_cbref++;
4087 
4088 	/*
4089 	 * Fill the callback program# and version into the request
4090 	 * structure for newnfs_connect() to use.
4091 	 */
4092 	clp->lc_req.nr_prog = clp->lc_program;
4093 #ifdef notnow
4094 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4095 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4096 	else
4097 #endif
4098 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4099 
4100 	/*
4101 	 * First, fill in some of the fields of nd and cr.
4102 	 */
4103 	nd->nd_flag = ND_NFSV4;
4104 	if (clp->lc_flags & LCL_GSS)
4105 		nd->nd_flag |= ND_KERBV;
4106 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4107 		nd->nd_flag |= ND_NFSV41;
4108 	nd->nd_repstat = 0;
4109 	cred->cr_uid = clp->lc_uid;
4110 	cred->cr_gid = clp->lc_gid;
4111 	callback = clp->lc_callback;
4112 	NFSUNLOCKSTATE();
4113 	cred->cr_ngroups = 1;
4114 
4115 	/*
4116 	 * Get the first mbuf for the request.
4117 	 */
4118 	MGET(m, M_WAITOK, MT_DATA);
4119 	mbuf_setlen(m, 0);
4120 	nd->nd_mreq = nd->nd_mb = m;
4121 	nd->nd_bpos = NFSMTOD(m, caddr_t);
4122 
4123 	/*
4124 	 * and build the callback request.
4125 	 */
4126 	if (procnum == NFSV4OP_CBGETATTR) {
4127 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4128 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4129 		    "CB Getattr", &sep);
4130 		if (error != 0) {
4131 			mbuf_freem(nd->nd_mreq);
4132 			goto errout;
4133 		}
4134 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4135 		(void)nfsrv_putattrbit(nd, attrbitp);
4136 	} else if (procnum == NFSV4OP_CBRECALL) {
4137 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4138 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4139 		    "CB Recall", &sep);
4140 		if (error != 0) {
4141 			mbuf_freem(nd->nd_mreq);
4142 			goto errout;
4143 		}
4144 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4145 		*tl++ = txdr_unsigned(stateidp->seqid);
4146 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4147 		    NFSX_STATEIDOTHER);
4148 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4149 		if (trunc)
4150 			*tl = newnfs_true;
4151 		else
4152 			*tl = newnfs_false;
4153 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4154 	} else if (procnum == NFSV4PROC_CBNULL) {
4155 		nd->nd_procnum = NFSV4PROC_CBNULL;
4156 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4157 			error = nfsv4_getcbsession(clp, &sep);
4158 			if (error != 0) {
4159 				mbuf_freem(nd->nd_mreq);
4160 				goto errout;
4161 			}
4162 		}
4163 	} else {
4164 		error = NFSERR_SERVERFAULT;
4165 		mbuf_freem(nd->nd_mreq);
4166 		goto errout;
4167 	}
4168 
4169 	/*
4170 	 * Call newnfs_connect(), as required, and then newnfs_request().
4171 	 */
4172 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4173 	if (clp->lc_req.nr_client == NULL) {
4174 		if ((clp->lc_flags & LCL_NFSV41) != 0)
4175 			error = ECONNREFUSED;
4176 		else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4177 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4178 			    NULL, 1);
4179 		else
4180 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4181 			    NULL, 3);
4182 	}
4183 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4184 	if (!error) {
4185 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4186 			KASSERT(sep != NULL, ("sep NULL"));
4187 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4188 			    NULL, NULL, cred, clp->lc_program,
4189 			    clp->lc_req.nr_vers, NULL, 1, NULL,
4190 			    &sep->sess_cbsess);
4191 			nfsrv_freesession(sep, NULL);
4192 		} else
4193 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4194 			    NULL, NULL, cred, clp->lc_program,
4195 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4196 	}
4197 errout:
4198 	NFSFREECRED(cred);
4199 
4200 	/*
4201 	 * If error is set here, the Callback path isn't working
4202 	 * properly, so twiddle the appropriate LCL_ flags.
4203 	 * (nd_repstat != 0 indicates the Callback path is working,
4204 	 *  but the callback failed on the client.)
4205 	 */
4206 	if (error) {
4207 		/*
4208 		 * Mark the callback pathway down, which disabled issuing
4209 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4210 		 */
4211 		NFSLOCKSTATE();
4212 		clp->lc_flags |= LCL_CBDOWN;
4213 		NFSUNLOCKSTATE();
4214 	} else {
4215 		/*
4216 		 * Callback worked. If the callback path was down, disable
4217 		 * callbacks, so no more delegations will be issued. (This
4218 		 * is done on the assumption that the callback pathway is
4219 		 * flakey.)
4220 		 */
4221 		NFSLOCKSTATE();
4222 		if (clp->lc_flags & LCL_CBDOWN)
4223 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4224 		NFSUNLOCKSTATE();
4225 		if (nd->nd_repstat)
4226 			error = nd->nd_repstat;
4227 		else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4228 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4229 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4230 			    p, NULL);
4231 		mbuf_freem(nd->nd_mrep);
4232 	}
4233 	NFSLOCKSTATE();
4234 	clp->lc_cbref--;
4235 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4236 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4237 		wakeup(clp);
4238 	}
4239 	NFSUNLOCKSTATE();
4240 
4241 	NFSEXITCODE(error);
4242 	return (error);
4243 }
4244 
4245 /*
4246  * Set up the compound RPC for the callback.
4247  */
4248 static int
nfsrv_cbcallargs(struct nfsrv_descript * nd,struct nfsclient * clp,uint32_t callback,int op,const char * optag,struct nfsdsession ** sepp)4249 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4250     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4251 {
4252 	uint32_t *tl;
4253 	int error, len;
4254 
4255 	len = strlen(optag);
4256 	(void)nfsm_strtom(nd, optag, len);
4257 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4258 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4259 		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4260 		*tl++ = txdr_unsigned(callback);
4261 		*tl++ = txdr_unsigned(2);
4262 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4263 		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4264 		if (error != 0)
4265 			return (error);
4266 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4267 		*tl = txdr_unsigned(op);
4268 	} else {
4269 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4270 		*tl++ = txdr_unsigned(callback);
4271 		*tl++ = txdr_unsigned(1);
4272 		*tl = txdr_unsigned(op);
4273 	}
4274 	return (0);
4275 }
4276 
4277 /*
4278  * Return the next index# for a clientid. Mostly just increment and return
4279  * the next one, but... if the 32bit unsigned does actually wrap around,
4280  * it should be rebooted.
4281  * At an average rate of one new client per second, it will wrap around in
4282  * approximately 136 years. (I think the server will have been shut
4283  * down or rebooted before then.)
4284  */
4285 static u_int32_t
nfsrv_nextclientindex(void)4286 nfsrv_nextclientindex(void)
4287 {
4288 	static u_int32_t client_index = 0;
4289 
4290 	client_index++;
4291 	if (client_index != 0)
4292 		return (client_index);
4293 
4294 	printf("%s: out of clientids\n", __func__);
4295 	return (client_index);
4296 }
4297 
4298 /*
4299  * Return the next index# for a stateid. Mostly just increment and return
4300  * the next one, but... if the 32bit unsigned does actually wrap around
4301  * (will a BSD server stay up that long?), find
4302  * new start and end values.
4303  */
4304 static u_int32_t
nfsrv_nextstateindex(struct nfsclient * clp)4305 nfsrv_nextstateindex(struct nfsclient *clp)
4306 {
4307 	struct nfsstate *stp;
4308 	int i;
4309 	u_int32_t canuse, min_index, max_index;
4310 
4311 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4312 		clp->lc_stateindex++;
4313 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4314 			return (clp->lc_stateindex);
4315 	}
4316 
4317 	/*
4318 	 * Yuck, we've hit the end.
4319 	 * Look for a new min and max.
4320 	 */
4321 	min_index = 0;
4322 	max_index = 0xffffffff;
4323 	for (i = 0; i < nfsrv_statehashsize; i++) {
4324 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4325 		if (stp->ls_stateid.other[2] > 0x80000000) {
4326 		    if (stp->ls_stateid.other[2] < max_index)
4327 			max_index = stp->ls_stateid.other[2];
4328 		} else {
4329 		    if (stp->ls_stateid.other[2] > min_index)
4330 			min_index = stp->ls_stateid.other[2];
4331 		}
4332 	    }
4333 	}
4334 
4335 	/*
4336 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4337 	 */
4338 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4339 	    canuse = 0;
4340 	    /*
4341 	     * Loop around until we find an unused entry. Return that
4342 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4343 	     * (This is one of those rare cases where a goto is the
4344 	     *  cleanest way to code the loop.)
4345 	     */
4346 tryagain:
4347 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4348 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4349 		    if (stp->ls_stateid.other[2] == canuse) {
4350 			canuse++;
4351 			goto tryagain;
4352 		    }
4353 		}
4354 	    }
4355 	    clp->lc_flags |= LCL_INDEXNOTOK;
4356 	    return (canuse);
4357 	}
4358 
4359 	/*
4360 	 * Ok to start again from min + 1.
4361 	 */
4362 	clp->lc_stateindex = min_index + 1;
4363 	clp->lc_statemaxindex = max_index;
4364 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4365 	return (clp->lc_stateindex);
4366 }
4367 
4368 /*
4369  * The following functions handle the stable storage file that deals with
4370  * the edge conditions described in RFC3530 Sec. 8.6.3.
4371  * The file is as follows:
4372  * - a single record at the beginning that has the lease time of the
4373  *   previous server instance (before the last reboot) and the nfsrvboottime
4374  *   values for the previous server boots.
4375  *   These previous boot times are used to ensure that the current
4376  *   nfsrvboottime does not, somehow, get set to a previous one.
4377  *   (This is important so that Stale ClientIDs and StateIDs can
4378  *    be recognized.)
4379  *   The number of previous nfsvrboottime values preceeds the list.
4380  * - followed by some number of appended records with:
4381  *   - client id string
4382  *   - flag that indicates it is a record revoking state via lease
4383  *     expiration or similar
4384  *     OR has successfully acquired state.
4385  * These structures vary in length, with the client string at the end, up
4386  * to NFSV4_OPAQUELIMIT in size.
4387  *
4388  * At the end of the grace period, the file is truncated, the first
4389  * record is rewritten with updated information and any acquired state
4390  * records for successful reclaims of state are written.
4391  *
4392  * Subsequent records are appended when the first state is issued to
4393  * a client and when state is revoked for a client.
4394  *
4395  * When reading the file in, state issued records that come later in
4396  * the file override older ones, since the append log is in cronological order.
4397  * If, for some reason, the file can't be read, the grace period is
4398  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4399  */
4400 
4401 /*
4402  * Read in the stable storage file. Called by nfssvc() before the nfsd
4403  * processes start servicing requests.
4404  */
4405 APPLESTATIC void
nfsrv_setupstable(NFSPROC_T * p)4406 nfsrv_setupstable(NFSPROC_T *p)
4407 {
4408 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4409 	struct nfsrv_stable *sp, *nsp;
4410 	struct nfst_rec *tsp;
4411 	int error, i, tryagain;
4412 	off_t off = 0;
4413 	ssize_t aresid, len;
4414 
4415 	/*
4416 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4417 	 * a reboot, so state has not been lost.
4418 	 */
4419 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4420 		return;
4421 	/*
4422 	 * Set Grace over just until the file reads successfully.
4423 	 */
4424 	nfsrvboottime = time_second;
4425 	LIST_INIT(&sf->nsf_head);
4426 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4427 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4428 	if (sf->nsf_fp == NULL)
4429 		return;
4430 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4431 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4432 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4433 	if (error || aresid || sf->nsf_numboots == 0 ||
4434 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4435 		return;
4436 
4437 	/*
4438 	 * Now, read in the boottimes.
4439 	 */
4440 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4441 		sizeof (time_t), M_TEMP, M_WAITOK);
4442 	off = sizeof (struct nfsf_rec);
4443 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4444 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4445 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4446 	if (error || aresid) {
4447 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4448 		sf->nsf_bootvals = NULL;
4449 		return;
4450 	}
4451 
4452 	/*
4453 	 * Make sure this nfsrvboottime is different from all recorded
4454 	 * previous ones.
4455 	 */
4456 	do {
4457 		tryagain = 0;
4458 		for (i = 0; i < sf->nsf_numboots; i++) {
4459 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4460 				nfsrvboottime++;
4461 				tryagain = 1;
4462 				break;
4463 			}
4464 		}
4465 	} while (tryagain);
4466 
4467 	sf->nsf_flags |= NFSNSF_OK;
4468 	off += (sf->nsf_numboots * sizeof (time_t));
4469 
4470 	/*
4471 	 * Read through the file, building a list of records for grace
4472 	 * checking.
4473 	 * Each record is between sizeof (struct nfst_rec) and
4474 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4475 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4476 	 */
4477 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4478 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4479 	do {
4480 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4481 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4482 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4483 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4484 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4485 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4486 		/*
4487 		 * Yuck, the file has been corrupted, so just return
4488 		 * after clearing out any restart state, so the grace period
4489 		 * is over.
4490 		 */
4491 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4492 			LIST_REMOVE(sp, nst_list);
4493 			free((caddr_t)sp, M_TEMP);
4494 		}
4495 		free((caddr_t)tsp, M_TEMP);
4496 		sf->nsf_flags &= ~NFSNSF_OK;
4497 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4498 		sf->nsf_bootvals = NULL;
4499 		return;
4500 	    }
4501 	    if (len > 0) {
4502 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4503 		/*
4504 		 * Search the list for a matching client.
4505 		 */
4506 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4507 			if (tsp->len == sp->nst_len &&
4508 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4509 				break;
4510 		}
4511 		if (sp == LIST_END(&sf->nsf_head)) {
4512 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4513 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4514 				M_WAITOK);
4515 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4516 				sizeof (struct nfst_rec) + tsp->len - 1);
4517 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4518 		} else {
4519 			if (tsp->flag == NFSNST_REVOKE)
4520 				sp->nst_flag |= NFSNST_REVOKE;
4521 			else
4522 				/*
4523 				 * A subsequent timestamp indicates the client
4524 				 * did a setclientid/confirm and any previous
4525 				 * revoke is no longer relevant.
4526 				 */
4527 				sp->nst_flag &= ~NFSNST_REVOKE;
4528 		}
4529 	    }
4530 	} while (len > 0);
4531 	free((caddr_t)tsp, M_TEMP);
4532 	sf->nsf_flags = NFSNSF_OK;
4533 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4534 		NFSRV_LEASEDELTA;
4535 }
4536 
4537 /*
4538  * Update the stable storage file, now that the grace period is over.
4539  */
4540 APPLESTATIC void
nfsrv_updatestable(NFSPROC_T * p)4541 nfsrv_updatestable(NFSPROC_T *p)
4542 {
4543 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4544 	struct nfsrv_stable *sp, *nsp;
4545 	int i;
4546 	struct nfsvattr nva;
4547 	vnode_t vp;
4548 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4549 	mount_t mp = NULL;
4550 #endif
4551 	int error;
4552 
4553 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4554 		return;
4555 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4556 	/*
4557 	 * Ok, we need to rewrite the stable storage file.
4558 	 * - truncate to 0 length
4559 	 * - write the new first structure
4560 	 * - loop through the data structures, writing out any that
4561 	 *   have timestamps older than the old boot
4562 	 */
4563 	if (sf->nsf_bootvals) {
4564 		sf->nsf_numboots++;
4565 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4566 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4567 	} else {
4568 		sf->nsf_numboots = 1;
4569 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4570 			M_TEMP, M_WAITOK);
4571 	}
4572 	sf->nsf_bootvals[0] = nfsrvboottime;
4573 	sf->nsf_lease = nfsrv_lease;
4574 	NFSVNO_ATTRINIT(&nva);
4575 	NFSVNO_SETATTRVAL(&nva, size, 0);
4576 	vp = NFSFPVNODE(sf->nsf_fp);
4577 	vn_start_write(vp, &mp, V_WAIT);
4578 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4579 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4580 		    NULL);
4581 		NFSVOPUNLOCK(vp, 0);
4582 	} else
4583 		error = EPERM;
4584 	vn_finished_write(mp);
4585 	if (!error)
4586 	    error = NFSD_RDWR(UIO_WRITE, vp,
4587 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4588 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4589 	if (!error)
4590 	    error = NFSD_RDWR(UIO_WRITE, vp,
4591 		(caddr_t)sf->nsf_bootvals,
4592 		sf->nsf_numboots * sizeof (time_t),
4593 		(off_t)(sizeof (struct nfsf_rec)),
4594 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4595 	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4596 	sf->nsf_bootvals = NULL;
4597 	if (error) {
4598 		sf->nsf_flags &= ~NFSNSF_OK;
4599 		printf("EEK! Can't write NfsV4 stable storage file\n");
4600 		return;
4601 	}
4602 	sf->nsf_flags |= NFSNSF_OK;
4603 
4604 	/*
4605 	 * Loop through the list and write out timestamp records for
4606 	 * any clients that successfully reclaimed state.
4607 	 */
4608 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4609 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4610 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4611 				NFSNST_NEWSTATE, p);
4612 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4613 		}
4614 		LIST_REMOVE(sp, nst_list);
4615 		free((caddr_t)sp, M_TEMP);
4616 	}
4617 	nfsrv_backupstable();
4618 }
4619 
4620 /*
4621  * Append a record to the stable storage file.
4622  */
4623 APPLESTATIC void
nfsrv_writestable(u_char * client,int len,int flag,NFSPROC_T * p)4624 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4625 {
4626 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4627 	struct nfst_rec *sp;
4628 	int error;
4629 
4630 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4631 		return;
4632 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4633 		len - 1, M_TEMP, M_WAITOK);
4634 	sp->len = len;
4635 	NFSBCOPY(client, sp->client, len);
4636 	sp->flag = flag;
4637 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4638 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4639 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4640 	free((caddr_t)sp, M_TEMP);
4641 	if (error) {
4642 		sf->nsf_flags &= ~NFSNSF_OK;
4643 		printf("EEK! Can't write NfsV4 stable storage file\n");
4644 	}
4645 }
4646 
4647 /*
4648  * This function is called during the grace period to mark a client
4649  * that successfully reclaimed state.
4650  */
4651 static void
nfsrv_markstable(struct nfsclient * clp)4652 nfsrv_markstable(struct nfsclient *clp)
4653 {
4654 	struct nfsrv_stable *sp;
4655 
4656 	/*
4657 	 * First find the client structure.
4658 	 */
4659 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4660 		if (sp->nst_len == clp->lc_idlen &&
4661 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4662 			break;
4663 	}
4664 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4665 		return;
4666 
4667 	/*
4668 	 * Now, just mark it and set the nfsclient back pointer.
4669 	 */
4670 	sp->nst_flag |= NFSNST_GOTSTATE;
4671 	sp->nst_clp = clp;
4672 }
4673 
4674 /*
4675  * This function is called for a reclaim, to see if it gets grace.
4676  * It returns 0 if a reclaim is allowed, 1 otherwise.
4677  */
4678 static int
nfsrv_checkstable(struct nfsclient * clp)4679 nfsrv_checkstable(struct nfsclient *clp)
4680 {
4681 	struct nfsrv_stable *sp;
4682 
4683 	/*
4684 	 * First, find the entry for the client.
4685 	 */
4686 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4687 		if (sp->nst_len == clp->lc_idlen &&
4688 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4689 			break;
4690 	}
4691 
4692 	/*
4693 	 * If not in the list, state was revoked or no state was issued
4694 	 * since the previous reboot, a reclaim is denied.
4695 	 */
4696 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4697 	    (sp->nst_flag & NFSNST_REVOKE) ||
4698 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4699 		return (1);
4700 	return (0);
4701 }
4702 
4703 /*
4704  * Test for and try to clear out a conflicting client. This is called by
4705  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4706  * a found.
4707  * The trick here is that it can't revoke a conflicting client with an
4708  * expired lease unless it holds the v4root lock, so...
4709  * If no v4root lock, get the lock and return 1 to indicate "try again".
4710  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4711  * the revocation worked and the conflicting client is "bye, bye", so it
4712  * can be tried again.
4713  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4714  * Unlocks State before a non-zero value is returned.
4715  */
4716 static int
nfsrv_clientconflict(struct nfsclient * clp,int * haslockp,vnode_t vp,NFSPROC_T * p)4717 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4718     NFSPROC_T *p)
4719 {
4720 	int gotlock, lktype = 0;
4721 
4722 	/*
4723 	 * If lease hasn't expired, we can't fix it.
4724 	 */
4725 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4726 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4727 		return (0);
4728 	if (*haslockp == 0) {
4729 		NFSUNLOCKSTATE();
4730 		if (vp != NULL) {
4731 			lktype = NFSVOPISLOCKED(vp);
4732 			NFSVOPUNLOCK(vp, 0);
4733 		}
4734 		NFSLOCKV4ROOTMUTEX();
4735 		nfsv4_relref(&nfsv4rootfs_lock);
4736 		do {
4737 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4738 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4739 		} while (!gotlock);
4740 		NFSUNLOCKV4ROOTMUTEX();
4741 		*haslockp = 1;
4742 		if (vp != NULL) {
4743 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4744 			if ((vp->v_iflag & VI_DOOMED) != 0)
4745 				return (2);
4746 		}
4747 		return (1);
4748 	}
4749 	NFSUNLOCKSTATE();
4750 
4751 	/*
4752 	 * Ok, we can expire the conflicting client.
4753 	 */
4754 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4755 	nfsrv_backupstable();
4756 	nfsrv_cleanclient(clp, p);
4757 	nfsrv_freedeleglist(&clp->lc_deleg);
4758 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4759 	LIST_REMOVE(clp, lc_hash);
4760 	nfsrv_zapclient(clp, p);
4761 	return (1);
4762 }
4763 
4764 /*
4765  * Resolve a delegation conflict.
4766  * Returns 0 to indicate the conflict was resolved without sleeping.
4767  * Return -1 to indicate that the caller should check for conflicts again.
4768  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4769  *
4770  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4771  * for a return of 0, since there was no sleep and it could be required
4772  * later. It is released for a return of NFSERR_DELAY, since the caller
4773  * will return that error. It is released when a sleep was done waiting
4774  * for the delegation to be returned or expire (so that other nfsds can
4775  * handle ops). Then, it must be acquired for the write to stable storage.
4776  * (This function is somewhat similar to nfsrv_clientconflict(), but
4777  *  the semantics differ in a couple of subtle ways. The return of 0
4778  *  indicates the conflict was resolved without sleeping here, not
4779  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4780  *  differs, as noted above.)
4781  * Unlocks State before returning a non-zero value.
4782  */
4783 static int
nfsrv_delegconflict(struct nfsstate * stp,int * haslockp,NFSPROC_T * p,vnode_t vp)4784 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4785     vnode_t vp)
4786 {
4787 	struct nfsclient *clp = stp->ls_clp;
4788 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
4789 	nfsv4stateid_t tstateid;
4790 	fhandle_t tfh;
4791 
4792 	/*
4793 	 * If the conflict is with an old delegation...
4794 	 */
4795 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4796 		/*
4797 		 * You can delete it, if it has expired.
4798 		 */
4799 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4800 			nfsrv_freedeleg(stp);
4801 			NFSUNLOCKSTATE();
4802 			error = -1;
4803 			goto out;
4804 		}
4805 		NFSUNLOCKSTATE();
4806 		/*
4807 		 * During this delay, the old delegation could expire or it
4808 		 * could be recovered by the client via an Open with
4809 		 * CLAIM_DELEGATE_PREV.
4810 		 * Release the nfsv4root_lock, if held.
4811 		 */
4812 		if (*haslockp) {
4813 			*haslockp = 0;
4814 			NFSLOCKV4ROOTMUTEX();
4815 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4816 			NFSUNLOCKV4ROOTMUTEX();
4817 		}
4818 		error = NFSERR_DELAY;
4819 		goto out;
4820 	}
4821 
4822 	/*
4823 	 * It's a current delegation, so:
4824 	 * - check to see if the delegation has expired
4825 	 *   - if so, get the v4root lock and then expire it
4826 	 */
4827 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4828 		/*
4829 		 * - do a recall callback, since not yet done
4830 		 * For now, never allow truncate to be set. To use
4831 		 * truncate safely, it must be guaranteed that the
4832 		 * Remove, Rename or Setattr with size of 0 will
4833 		 * succeed and that would require major changes to
4834 		 * the VFS/Vnode OPs.
4835 		 * Set the expiry time large enough so that it won't expire
4836 		 * until after the callback, then set it correctly, once
4837 		 * the callback is done. (The delegation will now time
4838 		 * out whether or not the Recall worked ok. The timeout
4839 		 * will be extended when ops are done on the delegation
4840 		 * stateid, up to the timelimit.)
4841 		 */
4842 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4843 		    NFSRV_LEASEDELTA;
4844 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4845 		    NFSRV_LEASEDELTA;
4846 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4847 
4848 		/*
4849 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4850 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4851 		 * in order to try and avoid a race that could happen
4852 		 * when a CBRecall request passed the Open reply with
4853 		 * the delegation in it when transitting the network.
4854 		 * Since nfsrv_docallback will sleep, don't use stp after
4855 		 * the call.
4856 		 */
4857 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4858 		    sizeof (tstateid));
4859 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4860 		    sizeof (tfh));
4861 		NFSUNLOCKSTATE();
4862 		if (*haslockp) {
4863 			*haslockp = 0;
4864 			NFSLOCKV4ROOTMUTEX();
4865 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4866 			NFSUNLOCKV4ROOTMUTEX();
4867 		}
4868 		retrycnt = 0;
4869 		do {
4870 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4871 			&tstateid, 0, &tfh, NULL, NULL, p);
4872 		    retrycnt++;
4873 		} while ((error == NFSERR_BADSTATEID ||
4874 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4875 		error = NFSERR_DELAY;
4876 		goto out;
4877 	}
4878 
4879 	if (clp->lc_expiry >= NFSD_MONOSEC &&
4880 	    stp->ls_delegtime >= NFSD_MONOSEC) {
4881 		NFSUNLOCKSTATE();
4882 		/*
4883 		 * A recall has been done, but it has not yet expired.
4884 		 * So, RETURN_DELAY.
4885 		 */
4886 		if (*haslockp) {
4887 			*haslockp = 0;
4888 			NFSLOCKV4ROOTMUTEX();
4889 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4890 			NFSUNLOCKV4ROOTMUTEX();
4891 		}
4892 		error = NFSERR_DELAY;
4893 		goto out;
4894 	}
4895 
4896 	/*
4897 	 * If we don't yet have the lock, just get it and then return,
4898 	 * since we need that before deleting expired state, such as
4899 	 * this delegation.
4900 	 * When getting the lock, unlock the vnode, so other nfsds that
4901 	 * are in progress, won't get stuck waiting for the vnode lock.
4902 	 */
4903 	if (*haslockp == 0) {
4904 		NFSUNLOCKSTATE();
4905 		if (vp != NULL) {
4906 			lktype = NFSVOPISLOCKED(vp);
4907 			NFSVOPUNLOCK(vp, 0);
4908 		}
4909 		NFSLOCKV4ROOTMUTEX();
4910 		nfsv4_relref(&nfsv4rootfs_lock);
4911 		do {
4912 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4913 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4914 		} while (!gotlock);
4915 		NFSUNLOCKV4ROOTMUTEX();
4916 		*haslockp = 1;
4917 		if (vp != NULL) {
4918 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4919 			if ((vp->v_iflag & VI_DOOMED) != 0) {
4920 				*haslockp = 0;
4921 				NFSLOCKV4ROOTMUTEX();
4922 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
4923 				NFSUNLOCKV4ROOTMUTEX();
4924 				error = NFSERR_PERM;
4925 				goto out;
4926 			}
4927 		}
4928 		error = -1;
4929 		goto out;
4930 	}
4931 
4932 	NFSUNLOCKSTATE();
4933 	/*
4934 	 * Ok, we can delete the expired delegation.
4935 	 * First, write the Revoke record to stable storage and then
4936 	 * clear out the conflict.
4937 	 * Since all other nfsd threads are now blocked, we can safely
4938 	 * sleep without the state changing.
4939 	 */
4940 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4941 	nfsrv_backupstable();
4942 	if (clp->lc_expiry < NFSD_MONOSEC) {
4943 		nfsrv_cleanclient(clp, p);
4944 		nfsrv_freedeleglist(&clp->lc_deleg);
4945 		nfsrv_freedeleglist(&clp->lc_olddeleg);
4946 		LIST_REMOVE(clp, lc_hash);
4947 		zapped_clp = 1;
4948 	} else {
4949 		nfsrv_freedeleg(stp);
4950 		zapped_clp = 0;
4951 	}
4952 	if (zapped_clp)
4953 		nfsrv_zapclient(clp, p);
4954 	error = -1;
4955 
4956 out:
4957 	NFSEXITCODE(error);
4958 	return (error);
4959 }
4960 
4961 /*
4962  * Check for a remove allowed, if remove is set to 1 and get rid of
4963  * delegations.
4964  */
4965 APPLESTATIC int
nfsrv_checkremove(vnode_t vp,int remove,NFSPROC_T * p)4966 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
4967 {
4968 	struct nfsstate *stp;
4969 	struct nfslockfile *lfp;
4970 	int error, haslock = 0;
4971 	fhandle_t nfh;
4972 
4973 	/*
4974 	 * First, get the lock file structure.
4975 	 * (A return of -1 means no associated state, so remove ok.)
4976 	 */
4977 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4978 tryagain:
4979 	NFSLOCKSTATE();
4980 	if (!error)
4981 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4982 	if (error) {
4983 		NFSUNLOCKSTATE();
4984 		if (haslock) {
4985 			NFSLOCKV4ROOTMUTEX();
4986 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4987 			NFSUNLOCKV4ROOTMUTEX();
4988 		}
4989 		if (error == -1)
4990 			error = 0;
4991 		goto out;
4992 	}
4993 
4994 	/*
4995 	 * Now, we must Recall any delegations.
4996 	 */
4997 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
4998 	if (error) {
4999 		/*
5000 		 * nfsrv_cleandeleg() unlocks state for non-zero
5001 		 * return.
5002 		 */
5003 		if (error == -1)
5004 			goto tryagain;
5005 		if (haslock) {
5006 			NFSLOCKV4ROOTMUTEX();
5007 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5008 			NFSUNLOCKV4ROOTMUTEX();
5009 		}
5010 		goto out;
5011 	}
5012 
5013 	/*
5014 	 * Now, look for a conflicting open share.
5015 	 */
5016 	if (remove) {
5017 		LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5018 			if (stp->ls_flags & NFSLCK_WRITEDENY) {
5019 				error = NFSERR_FILEOPEN;
5020 				break;
5021 			}
5022 		}
5023 	}
5024 
5025 	NFSUNLOCKSTATE();
5026 	if (haslock) {
5027 		NFSLOCKV4ROOTMUTEX();
5028 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5029 		NFSUNLOCKV4ROOTMUTEX();
5030 	}
5031 
5032 out:
5033 	NFSEXITCODE(error);
5034 	return (error);
5035 }
5036 
5037 /*
5038  * Clear out all delegations for the file referred to by lfp.
5039  * May return NFSERR_DELAY, if there will be a delay waiting for
5040  * delegations to expire.
5041  * Returns -1 to indicate it slept while recalling a delegation.
5042  * This function has the side effect of deleting the nfslockfile structure,
5043  * if it no longer has associated state and didn't have to sleep.
5044  * Unlocks State before a non-zero value is returned.
5045  */
5046 static int
nfsrv_cleandeleg(vnode_t vp,struct nfslockfile * lfp,struct nfsclient * clp,int * haslockp,NFSPROC_T * p)5047 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5048     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5049 {
5050 	struct nfsstate *stp, *nstp;
5051 	int ret = 0;
5052 
5053 	stp = LIST_FIRST(&lfp->lf_deleg);
5054 	while (stp != LIST_END(&lfp->lf_deleg)) {
5055 		nstp = LIST_NEXT(stp, ls_file);
5056 		if (stp->ls_clp != clp) {
5057 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5058 			if (ret) {
5059 				/*
5060 				 * nfsrv_delegconflict() unlocks state
5061 				 * when it returns non-zero.
5062 				 */
5063 				goto out;
5064 			}
5065 		}
5066 		stp = nstp;
5067 	}
5068 out:
5069 	NFSEXITCODE(ret);
5070 	return (ret);
5071 }
5072 
5073 /*
5074  * There are certain operations that, when being done outside of NFSv4,
5075  * require that any NFSv4 delegation for the file be recalled.
5076  * This function is to be called for those cases:
5077  * VOP_RENAME() - When a delegation is being recalled for any reason,
5078  *	the client may have to do Opens against the server, using the file's
5079  *	final component name. If the file has been renamed on the server,
5080  *	that component name will be incorrect and the Open will fail.
5081  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5082  *	been removed on the server, if there is a delegation issued to
5083  *	that client for the file. I say "theoretically" since clients
5084  *	normally do an Access Op before the Open and that Access Op will
5085  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5086  *	they will detect the file's removal in the same manner. (There is
5087  *	one case where RFC3530 allows a client to do an Open without first
5088  *	doing an Access Op, which is passage of a check against the ACE
5089  *	returned with a Write delegation, but current practice is to ignore
5090  *	the ACE and always do an Access Op.)
5091  *	Since the functions can only be called with an unlocked vnode, this
5092  *	can't be done at this time.
5093  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5094  *	locks locally in the client, which are not visible to the server. To
5095  *	deal with this, issuing of delegations for a vnode must be disabled
5096  *	and all delegations for the vnode recalled. This is done via the
5097  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5098  */
5099 APPLESTATIC void
nfsd_recalldelegation(vnode_t vp,NFSPROC_T * p)5100 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5101 {
5102 	time_t starttime;
5103 	int error;
5104 
5105 	/*
5106 	 * First, check to see if the server is currently running and it has
5107 	 * been called for a regular file when issuing delegations.
5108 	 */
5109 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5110 	    nfsrv_issuedelegs == 0)
5111 		return;
5112 
5113 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5114 	/*
5115 	 * First, get a reference on the nfsv4rootfs_lock so that an
5116 	 * exclusive lock cannot be acquired by another thread.
5117 	 */
5118 	NFSLOCKV4ROOTMUTEX();
5119 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5120 	NFSUNLOCKV4ROOTMUTEX();
5121 
5122 	/*
5123 	 * Now, call nfsrv_checkremove() in a loop while it returns
5124 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5125 	 */
5126 	starttime = NFSD_MONOSEC;
5127 	do {
5128 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5129 			error = nfsrv_checkremove(vp, 0, p);
5130 			NFSVOPUNLOCK(vp, 0);
5131 		} else
5132 			error = EPERM;
5133 		if (error == NFSERR_DELAY) {
5134 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5135 				break;
5136 			/* Sleep for a short period of time */
5137 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5138 		}
5139 	} while (error == NFSERR_DELAY);
5140 	NFSLOCKV4ROOTMUTEX();
5141 	nfsv4_relref(&nfsv4rootfs_lock);
5142 	NFSUNLOCKV4ROOTMUTEX();
5143 }
5144 
5145 APPLESTATIC void
nfsd_disabledelegation(vnode_t vp,NFSPROC_T * p)5146 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5147 {
5148 
5149 #ifdef VV_DISABLEDELEG
5150 	/*
5151 	 * First, flag issuance of delegations disabled.
5152 	 */
5153 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5154 #endif
5155 
5156 	/*
5157 	 * Then call nfsd_recalldelegation() to get rid of all extant
5158 	 * delegations.
5159 	 */
5160 	nfsd_recalldelegation(vp, p);
5161 }
5162 
5163 /*
5164  * Check for conflicting locks, etc. and then get rid of delegations.
5165  * (At one point I thought that I should get rid of delegations for any
5166  *  Setattr, since it could potentially disallow the I/O op (read or write)
5167  *  allowed by the delegation. However, Setattr Ops that aren't changing
5168  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5169  *  for the same client or a different one, so I decided to only get rid
5170  *  of delegations for other clients when the size is being changed.)
5171  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5172  * as Write backs, even if there is no delegation, so it really isn't any
5173  * different?)
5174  */
5175 APPLESTATIC 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)5176 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5177     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5178     struct nfsexstuff *exp, NFSPROC_T *p)
5179 {
5180 	struct nfsstate st, *stp = &st;
5181 	struct nfslock lo, *lop = &lo;
5182 	int error = 0;
5183 	nfsquad_t clientid;
5184 
5185 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5186 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5187 		lop->lo_first = nvap->na_size;
5188 	} else {
5189 		stp->ls_flags = 0;
5190 		lop->lo_first = 0;
5191 	}
5192 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5193 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5194 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5195 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5196 		stp->ls_flags |= NFSLCK_SETATTR;
5197 	if (stp->ls_flags == 0)
5198 		goto out;
5199 	lop->lo_end = NFS64BITSSET;
5200 	lop->lo_flags = NFSLCK_WRITE;
5201 	stp->ls_ownerlen = 0;
5202 	stp->ls_op = NULL;
5203 	stp->ls_uid = nd->nd_cred->cr_uid;
5204 	stp->ls_stateid.seqid = stateidp->seqid;
5205 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5206 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5207 	stp->ls_stateid.other[2] = stateidp->other[2];
5208 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5209 	    stateidp, exp, nd, p);
5210 
5211 out:
5212 	NFSEXITCODE2(error, nd);
5213 	return (error);
5214 }
5215 
5216 /*
5217  * Check for a write delegation and do a CBGETATTR if there is one, updating
5218  * the attributes, as required.
5219  * Should I return an error if I can't get the attributes? (For now, I'll
5220  * just return ok.
5221  */
5222 APPLESTATIC int
nfsrv_checkgetattr(struct nfsrv_descript * nd,vnode_t vp,struct nfsvattr * nvap,nfsattrbit_t * attrbitp,struct ucred * cred,NFSPROC_T * p)5223 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5224     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5225     NFSPROC_T *p)
5226 {
5227 	struct nfsstate *stp;
5228 	struct nfslockfile *lfp;
5229 	struct nfsclient *clp;
5230 	struct nfsvattr nva;
5231 	fhandle_t nfh;
5232 	int error = 0;
5233 	nfsattrbit_t cbbits;
5234 	u_quad_t delegfilerev;
5235 
5236 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5237 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5238 		goto out;
5239 
5240 	/*
5241 	 * Get the lock file structure.
5242 	 * (A return of -1 means no associated state, so return ok.)
5243 	 */
5244 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5245 	NFSLOCKSTATE();
5246 	if (!error)
5247 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5248 	if (error) {
5249 		NFSUNLOCKSTATE();
5250 		if (error == -1)
5251 			error = 0;
5252 		goto out;
5253 	}
5254 
5255 	/*
5256 	 * Now, look for a write delegation.
5257 	 */
5258 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5259 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5260 			break;
5261 	}
5262 	if (stp == LIST_END(&lfp->lf_deleg)) {
5263 		NFSUNLOCKSTATE();
5264 		goto out;
5265 	}
5266 	clp = stp->ls_clp;
5267 	delegfilerev = stp->ls_filerev;
5268 
5269 	/*
5270 	 * If the Write delegation was issued as a part of this Compound RPC
5271 	 * or if we have an Implied Clientid (used in a previous Op in this
5272 	 * compound) and it is the client the delegation was issued to,
5273 	 * just return ok.
5274 	 * I also assume that it is from the same client iff the network
5275 	 * host IP address is the same as the callback address. (Not
5276 	 * exactly correct by the RFC, but avoids a lot of Getattr
5277 	 * callbacks.)
5278 	 */
5279 	if (nd->nd_compref == stp->ls_compref ||
5280 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5281 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5282 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5283 		NFSUNLOCKSTATE();
5284 		goto out;
5285 	}
5286 
5287 	/*
5288 	 * We are now done with the delegation state structure,
5289 	 * so the statelock can be released and we can now tsleep().
5290 	 */
5291 
5292 	/*
5293 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5294 	 * has changed.
5295 	 */
5296 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5297 		NFSUNLOCKSTATE();
5298 		NFSVNO_ATTRINIT(&nva);
5299 		nva.na_filerev = NFS64BITSSET;
5300 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5301 		    0, &nfh, &nva, &cbbits, p);
5302 		if (!error) {
5303 			if ((nva.na_filerev != NFS64BITSSET &&
5304 			    nva.na_filerev > delegfilerev) ||
5305 			    (NFSVNO_ISSETSIZE(&nva) &&
5306 			     nva.na_size != nvap->na_size)) {
5307 				error = nfsvno_updfilerev(vp, nvap, cred, p);
5308 				if (NFSVNO_ISSETSIZE(&nva))
5309 					nvap->na_size = nva.na_size;
5310 			}
5311 		} else
5312 			error = 0;	/* Ignore callback errors for now. */
5313 	} else {
5314 		NFSUNLOCKSTATE();
5315 	}
5316 
5317 out:
5318 	NFSEXITCODE2(error, nd);
5319 	return (error);
5320 }
5321 
5322 /*
5323  * This function looks for openowners that haven't had any opens for
5324  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5325  * is set.
5326  */
5327 APPLESTATIC void
nfsrv_throwawayopens(NFSPROC_T * p)5328 nfsrv_throwawayopens(NFSPROC_T *p)
5329 {
5330 	struct nfsclient *clp, *nclp;
5331 	struct nfsstate *stp, *nstp;
5332 	int i;
5333 
5334 	NFSLOCKSTATE();
5335 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5336 	/*
5337 	 * For each client...
5338 	 */
5339 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5340 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5341 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5342 			if (LIST_EMPTY(&stp->ls_open) &&
5343 			    (stp->ls_noopens > NFSNOOPEN ||
5344 			     (nfsrv_openpluslock * 2) >
5345 			     nfsrv_v4statelimit))
5346 				nfsrv_freeopenowner(stp, 0, p);
5347 		}
5348 	    }
5349 	}
5350 	NFSUNLOCKSTATE();
5351 }
5352 
5353 /*
5354  * This function checks to see if the credentials are the same.
5355  * Returns 1 for not same, 0 otherwise.
5356  */
5357 static int
nfsrv_notsamecredname(struct nfsrv_descript * nd,struct nfsclient * clp)5358 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5359 {
5360 
5361 	if (nd->nd_flag & ND_GSS) {
5362 		if (!(clp->lc_flags & LCL_GSS))
5363 			return (1);
5364 		if (clp->lc_flags & LCL_NAME) {
5365 			if (nd->nd_princlen != clp->lc_namelen ||
5366 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5367 				clp->lc_namelen))
5368 				return (1);
5369 			else
5370 				return (0);
5371 		}
5372 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5373 			return (0);
5374 		else
5375 			return (1);
5376 	} else if (clp->lc_flags & LCL_GSS)
5377 		return (1);
5378 	/*
5379 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5380 	 * in RFC3530, which talks about principals, but doesn't say anything
5381 	 * about uids for AUTH_SYS.)
5382 	 */
5383 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5384 		return (0);
5385 	else
5386 		return (1);
5387 }
5388 
5389 /*
5390  * Calculate the lease expiry time.
5391  */
5392 static time_t
nfsrv_leaseexpiry(void)5393 nfsrv_leaseexpiry(void)
5394 {
5395 
5396 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5397 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5398 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5399 }
5400 
5401 /*
5402  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5403  */
5404 static void
nfsrv_delaydelegtimeout(struct nfsstate * stp)5405 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5406 {
5407 
5408 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5409 		return;
5410 
5411 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5412 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5413 		stp->ls_delegtime += nfsrv_lease;
5414 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5415 			stp->ls_delegtime = stp->ls_delegtimelimit;
5416 	}
5417 }
5418 
5419 /*
5420  * This function checks to see if there is any other state associated
5421  * with the openowner for this Open.
5422  * It returns 1 if there is no other state, 0 otherwise.
5423  */
5424 static int
nfsrv_nootherstate(struct nfsstate * stp)5425 nfsrv_nootherstate(struct nfsstate *stp)
5426 {
5427 	struct nfsstate *tstp;
5428 
5429 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5430 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5431 			return (0);
5432 	}
5433 	return (1);
5434 }
5435 
5436 /*
5437  * Create a list of lock deltas (changes to local byte range locking
5438  * that can be rolled back using the list) and apply the changes via
5439  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5440  * the rollback or update function will be called after this.
5441  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5442  * call fails. If it returns an error, it will unlock the list.
5443  */
5444 static int
nfsrv_locallock(vnode_t vp,struct nfslockfile * lfp,int flags,uint64_t first,uint64_t end,struct nfslockconflict * cfp,NFSPROC_T * p)5445 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5446     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5447 {
5448 	struct nfslock *lop, *nlop;
5449 	int error = 0;
5450 
5451 	/* Loop through the list of locks. */
5452 	lop = LIST_FIRST(&lfp->lf_locallock);
5453 	while (first < end && lop != NULL) {
5454 		nlop = LIST_NEXT(lop, lo_lckowner);
5455 		if (first >= lop->lo_end) {
5456 			/* not there yet */
5457 			lop = nlop;
5458 		} else if (first < lop->lo_first) {
5459 			/* new one starts before entry in list */
5460 			if (end <= lop->lo_first) {
5461 				/* no overlap between old and new */
5462 				error = nfsrv_dolocal(vp, lfp, flags,
5463 				    NFSLCK_UNLOCK, first, end, cfp, p);
5464 				if (error != 0)
5465 					break;
5466 				first = end;
5467 			} else {
5468 				/* handle fragment overlapped with new one */
5469 				error = nfsrv_dolocal(vp, lfp, flags,
5470 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5471 				    p);
5472 				if (error != 0)
5473 					break;
5474 				first = lop->lo_first;
5475 			}
5476 		} else {
5477 			/* new one overlaps this entry in list */
5478 			if (end <= lop->lo_end) {
5479 				/* overlaps all of new one */
5480 				error = nfsrv_dolocal(vp, lfp, flags,
5481 				    lop->lo_flags, first, end, cfp, p);
5482 				if (error != 0)
5483 					break;
5484 				first = end;
5485 			} else {
5486 				/* handle fragment overlapped with new one */
5487 				error = nfsrv_dolocal(vp, lfp, flags,
5488 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5489 				if (error != 0)
5490 					break;
5491 				first = lop->lo_end;
5492 				lop = nlop;
5493 			}
5494 		}
5495 	}
5496 	if (first < end && error == 0)
5497 		/* handle fragment past end of list */
5498 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5499 		    end, cfp, p);
5500 
5501 	NFSEXITCODE(error);
5502 	return (error);
5503 }
5504 
5505 /*
5506  * Local lock unlock. Unlock all byte ranges that are no longer locked
5507  * by NFSv4. To do this, unlock any subranges of first-->end that
5508  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5509  * list. This list has all locks for the file held by other
5510  * <clientid, lockowner> tuples. The list is ordered by increasing
5511  * lo_first value, but may have entries that overlap each other, for
5512  * the case of read locks.
5513  */
5514 static void
nfsrv_localunlock(vnode_t vp,struct nfslockfile * lfp,uint64_t init_first,uint64_t init_end,NFSPROC_T * p)5515 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5516     uint64_t init_end, NFSPROC_T *p)
5517 {
5518 	struct nfslock *lop;
5519 	uint64_t first, end, prevfirst;
5520 
5521 	first = init_first;
5522 	end = init_end;
5523 	while (first < init_end) {
5524 		/* Loop through all nfs locks, adjusting first and end */
5525 		prevfirst = 0;
5526 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5527 			KASSERT(prevfirst <= lop->lo_first,
5528 			    ("nfsv4 locks out of order"));
5529 			KASSERT(lop->lo_first < lop->lo_end,
5530 			    ("nfsv4 bogus lock"));
5531 			prevfirst = lop->lo_first;
5532 			if (first >= lop->lo_first &&
5533 			    first < lop->lo_end)
5534 				/*
5535 				 * Overlaps with initial part, so trim
5536 				 * off that initial part by moving first past
5537 				 * it.
5538 				 */
5539 				first = lop->lo_end;
5540 			else if (end > lop->lo_first &&
5541 			    lop->lo_first > first) {
5542 				/*
5543 				 * This lock defines the end of the
5544 				 * segment to unlock, so set end to the
5545 				 * start of it and break out of the loop.
5546 				 */
5547 				end = lop->lo_first;
5548 				break;
5549 			}
5550 			if (first >= end)
5551 				/*
5552 				 * There is no segment left to do, so
5553 				 * break out of this loop and then exit
5554 				 * the outer while() since first will be set
5555 				 * to end, which must equal init_end here.
5556 				 */
5557 				break;
5558 		}
5559 		if (first < end) {
5560 			/* Unlock this segment */
5561 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5562 			    NFSLCK_READ, first, end, NULL, p);
5563 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5564 			    first, end);
5565 		}
5566 		/*
5567 		 * Now move past this segment and look for any further
5568 		 * segment in the range, if there is one.
5569 		 */
5570 		first = end;
5571 		end = init_end;
5572 	}
5573 }
5574 
5575 /*
5576  * Do the local lock operation and update the rollback list, as required.
5577  * Perform the rollback and return the error if nfsvno_advlock() fails.
5578  */
5579 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)5580 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5581     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5582 {
5583 	struct nfsrollback *rlp;
5584 	int error = 0, ltype, oldltype;
5585 
5586 	if (flags & NFSLCK_WRITE)
5587 		ltype = F_WRLCK;
5588 	else if (flags & NFSLCK_READ)
5589 		ltype = F_RDLCK;
5590 	else
5591 		ltype = F_UNLCK;
5592 	if (oldflags & NFSLCK_WRITE)
5593 		oldltype = F_WRLCK;
5594 	else if (oldflags & NFSLCK_READ)
5595 		oldltype = F_RDLCK;
5596 	else
5597 		oldltype = F_UNLCK;
5598 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5599 		/* nothing to do */
5600 		goto out;
5601 	error = nfsvno_advlock(vp, ltype, first, end, p);
5602 	if (error != 0) {
5603 		if (cfp != NULL) {
5604 			cfp->cl_clientid.lval[0] = 0;
5605 			cfp->cl_clientid.lval[1] = 0;
5606 			cfp->cl_first = 0;
5607 			cfp->cl_end = NFS64BITSSET;
5608 			cfp->cl_flags = NFSLCK_WRITE;
5609 			cfp->cl_ownerlen = 5;
5610 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5611 		}
5612 		nfsrv_locallock_rollback(vp, lfp, p);
5613 	} else if (ltype != F_UNLCK) {
5614 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5615 		    M_WAITOK);
5616 		rlp->rlck_first = first;
5617 		rlp->rlck_end = end;
5618 		rlp->rlck_type = oldltype;
5619 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5620 	}
5621 
5622 out:
5623 	NFSEXITCODE(error);
5624 	return (error);
5625 }
5626 
5627 /*
5628  * Roll back local lock changes and free up the rollback list.
5629  */
5630 static void
nfsrv_locallock_rollback(vnode_t vp,struct nfslockfile * lfp,NFSPROC_T * p)5631 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5632 {
5633 	struct nfsrollback *rlp, *nrlp;
5634 
5635 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5636 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5637 		    rlp->rlck_end, p);
5638 		free(rlp, M_NFSDROLLBACK);
5639 	}
5640 	LIST_INIT(&lfp->lf_rollback);
5641 }
5642 
5643 /*
5644  * Update local lock list and delete rollback list (ie now committed to the
5645  * local locks). Most of the work is done by the internal function.
5646  */
5647 static void
nfsrv_locallock_commit(struct nfslockfile * lfp,int flags,uint64_t first,uint64_t end)5648 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5649     uint64_t end)
5650 {
5651 	struct nfsrollback *rlp, *nrlp;
5652 	struct nfslock *new_lop, *other_lop;
5653 
5654 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5655 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5656 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5657 		    M_WAITOK);
5658 	else
5659 		other_lop = NULL;
5660 	new_lop->lo_flags = flags;
5661 	new_lop->lo_first = first;
5662 	new_lop->lo_end = end;
5663 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5664 	if (new_lop != NULL)
5665 		free(new_lop, M_NFSDLOCK);
5666 	if (other_lop != NULL)
5667 		free(other_lop, M_NFSDLOCK);
5668 
5669 	/* and get rid of the rollback list */
5670 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5671 		free(rlp, M_NFSDROLLBACK);
5672 	LIST_INIT(&lfp->lf_rollback);
5673 }
5674 
5675 /*
5676  * Lock the struct nfslockfile for local lock updating.
5677  */
5678 static void
nfsrv_locklf(struct nfslockfile * lfp)5679 nfsrv_locklf(struct nfslockfile *lfp)
5680 {
5681 	int gotlock;
5682 
5683 	/* lf_usecount ensures *lfp won't be free'd */
5684 	lfp->lf_usecount++;
5685 	do {
5686 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5687 		    NFSSTATEMUTEXPTR, NULL);
5688 	} while (gotlock == 0);
5689 	lfp->lf_usecount--;
5690 }
5691 
5692 /*
5693  * Unlock the struct nfslockfile after local lock updating.
5694  */
5695 static void
nfsrv_unlocklf(struct nfslockfile * lfp)5696 nfsrv_unlocklf(struct nfslockfile *lfp)
5697 {
5698 
5699 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5700 }
5701 
5702 /*
5703  * Clear out all state for the NFSv4 server.
5704  * Must be called by a thread that can sleep when no nfsds are running.
5705  */
5706 void
nfsrv_throwawayallstate(NFSPROC_T * p)5707 nfsrv_throwawayallstate(NFSPROC_T *p)
5708 {
5709 	struct nfsclient *clp, *nclp;
5710 	struct nfslockfile *lfp, *nlfp;
5711 	int i;
5712 
5713 	/*
5714 	 * For each client, clean out the state and then free the structure.
5715 	 */
5716 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5717 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5718 			nfsrv_cleanclient(clp, p);
5719 			nfsrv_freedeleglist(&clp->lc_deleg);
5720 			nfsrv_freedeleglist(&clp->lc_olddeleg);
5721 			free(clp->lc_stateid, M_NFSDCLIENT);
5722 			free(clp, M_NFSDCLIENT);
5723 		}
5724 	}
5725 
5726 	/*
5727 	 * Also, free up any remaining lock file structures.
5728 	 */
5729 	for (i = 0; i < nfsrv_lockhashsize; i++) {
5730 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
5731 			printf("nfsd unload: fnd a lock file struct\n");
5732 			nfsrv_freenfslockfile(lfp);
5733 		}
5734 	}
5735 }
5736 
5737 /*
5738  * Check the sequence# for the session and slot provided as an argument.
5739  * Also, renew the lease if the session will return NFS_OK.
5740  */
5741 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)5742 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
5743     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
5744     uint32_t *sflagsp, NFSPROC_T *p)
5745 {
5746 	struct nfsdsession *sep;
5747 	struct nfssessionhash *shp;
5748 	int error;
5749 	SVCXPRT *savxprt;
5750 
5751 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5752 	NFSLOCKSESSION(shp);
5753 	sep = nfsrv_findsession(nd->nd_sessionid);
5754 	if (sep == NULL) {
5755 		NFSUNLOCKSESSION(shp);
5756 		return (NFSERR_BADSESSION);
5757 	}
5758 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
5759 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
5760 	if (error != 0) {
5761 		NFSUNLOCKSESSION(shp);
5762 		return (error);
5763 	}
5764 	if (cache_this != 0)
5765 		nd->nd_flag |= ND_SAVEREPLY;
5766 	/* Renew the lease. */
5767 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
5768 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
5769 	nd->nd_flag |= ND_IMPLIEDCLID;
5770 
5771 	/*
5772 	 * If this session handles the backchannel, save the nd_xprt for this
5773 	 * RPC, since this is the one being used.
5774 	 */
5775 	if (sep->sess_cbsess.nfsess_xprt != NULL &&
5776 	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) {
5777 		savxprt = sep->sess_cbsess.nfsess_xprt;
5778 		SVC_ACQUIRE(nd->nd_xprt);
5779 		nd->nd_xprt->xp_p2 = savxprt->xp_p2;
5780 		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
5781 		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
5782 		SVC_RELEASE(savxprt);
5783 	}
5784 
5785 	*sflagsp = 0;
5786 	if (sep->sess_clp->lc_req.nr_client == NULL)
5787 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
5788 	NFSUNLOCKSESSION(shp);
5789 	if (error == NFSERR_EXPIRED) {
5790 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
5791 		error = 0;
5792 	} else if (error == NFSERR_ADMINREVOKED) {
5793 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
5794 		error = 0;
5795 	}
5796 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
5797 	return (0);
5798 }
5799 
5800 /*
5801  * Check/set reclaim complete for this session/clientid.
5802  */
5803 int
nfsrv_checkreclaimcomplete(struct nfsrv_descript * nd)5804 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd)
5805 {
5806 	struct nfsdsession *sep;
5807 	struct nfssessionhash *shp;
5808 	int error = 0;
5809 
5810 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5811 	NFSLOCKSTATE();
5812 	NFSLOCKSESSION(shp);
5813 	sep = nfsrv_findsession(nd->nd_sessionid);
5814 	if (sep == NULL) {
5815 		NFSUNLOCKSESSION(shp);
5816 		NFSUNLOCKSTATE();
5817 		return (NFSERR_BADSESSION);
5818 	}
5819 
5820 	/* Check to see if reclaim complete has already happened. */
5821 	if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
5822 		error = NFSERR_COMPLETEALREADY;
5823 	else
5824 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
5825 	NFSUNLOCKSESSION(shp);
5826 	NFSUNLOCKSTATE();
5827 	return (error);
5828 }
5829 
5830 /*
5831  * Cache the reply in a session slot.
5832  */
5833 void
nfsrv_cache_session(uint8_t * sessionid,uint32_t slotid,int repstat,struct mbuf ** m)5834 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
5835    struct mbuf **m)
5836 {
5837 	struct nfsdsession *sep;
5838 	struct nfssessionhash *shp;
5839 
5840 	shp = NFSSESSIONHASH(sessionid);
5841 	NFSLOCKSESSION(shp);
5842 	sep = nfsrv_findsession(sessionid);
5843 	if (sep == NULL) {
5844 		NFSUNLOCKSESSION(shp);
5845 		printf("nfsrv_cache_session: no session\n");
5846 		m_freem(*m);
5847 		return;
5848 	}
5849 	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
5850 	NFSUNLOCKSESSION(shp);
5851 }
5852 
5853 /*
5854  * Search for a session that matches the sessionid.
5855  */
5856 static struct nfsdsession *
nfsrv_findsession(uint8_t * sessionid)5857 nfsrv_findsession(uint8_t *sessionid)
5858 {
5859 	struct nfsdsession *sep;
5860 	struct nfssessionhash *shp;
5861 
5862 	shp = NFSSESSIONHASH(sessionid);
5863 	LIST_FOREACH(sep, &shp->list, sess_hash) {
5864 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
5865 			break;
5866 	}
5867 	return (sep);
5868 }
5869 
5870 /*
5871  * Destroy a session.
5872  */
5873 int
nfsrv_destroysession(struct nfsrv_descript * nd,uint8_t * sessionid)5874 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
5875 {
5876 	int error, samesess;
5877 
5878 	samesess = 0;
5879 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID)) {
5880 		samesess = 1;
5881 		if ((nd->nd_flag & ND_LASTOP) == 0)
5882 			return (NFSERR_BADSESSION);
5883 	}
5884 	error = nfsrv_freesession(NULL, sessionid);
5885 	if (error == 0 && samesess != 0)
5886 		nd->nd_flag &= ~ND_HASSEQUENCE;
5887 	return (error);
5888 }
5889 
5890 /*
5891  * Free up a session structure.
5892  */
5893 static int
nfsrv_freesession(struct nfsdsession * sep,uint8_t * sessionid)5894 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
5895 {
5896 	struct nfssessionhash *shp;
5897 	int i;
5898 
5899 	if (sep == NULL) {
5900 		shp = NFSSESSIONHASH(sessionid);
5901 		NFSLOCKSESSION(shp);
5902 		sep = nfsrv_findsession(sessionid);
5903 	} else {
5904 		shp = NFSSESSIONHASH(sep->sess_sessionid);
5905 		NFSLOCKSESSION(shp);
5906 	}
5907 	if (sep != NULL) {
5908 		NFSLOCKSTATE();
5909 		sep->sess_refcnt--;
5910 		if (sep->sess_refcnt > 0) {
5911 			NFSUNLOCKSTATE();
5912 			NFSUNLOCKSESSION(shp);
5913 			return (0);
5914 		}
5915 		LIST_REMOVE(sep, sess_hash);
5916 		LIST_REMOVE(sep, sess_list);
5917 		NFSUNLOCKSTATE();
5918 	}
5919 	NFSUNLOCKSESSION(shp);
5920 	if (sep == NULL)
5921 		return (NFSERR_BADSESSION);
5922 	for (i = 0; i < NFSV4_SLOTS; i++)
5923 		if (sep->sess_slots[i].nfssl_reply != NULL)
5924 			m_freem(sep->sess_slots[i].nfssl_reply);
5925 	if (sep->sess_cbsess.nfsess_xprt != NULL)
5926 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
5927 	free(sep, M_NFSDSESSION);
5928 	return (0);
5929 }
5930 
5931 /*
5932  * Free a stateid.
5933  * RFC5661 says that it should fail when there are associated opens, locks
5934  * or delegations. Since stateids represent opens, I don't see how you can
5935  * free an open stateid (it will be free'd when closed), so this function
5936  * only works for lock stateids (freeing the lock_owner) or delegations.
5937  */
5938 int
nfsrv_freestateid(struct nfsrv_descript * nd,nfsv4stateid_t * stateidp,NFSPROC_T * p)5939 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
5940     NFSPROC_T *p)
5941 {
5942 	struct nfsclient *clp;
5943 	struct nfsstate *stp;
5944 	int error;
5945 
5946 	NFSLOCKSTATE();
5947 	/*
5948 	 * Look up the stateid
5949 	 */
5950 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
5951 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
5952 	if (error == 0) {
5953 		/* First, check for a delegation. */
5954 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
5955 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
5956 			    NFSX_STATEIDOTHER))
5957 				break;
5958 		}
5959 		if (stp != NULL) {
5960 			nfsrv_freedeleg(stp);
5961 			NFSUNLOCKSTATE();
5962 			return (error);
5963 		}
5964 	}
5965 	/* Not a delegation, try for a lock_owner. */
5966 	if (error == 0)
5967 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
5968 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
5969 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
5970 		/* Not a lock_owner stateid. */
5971 		error = NFSERR_LOCKSHELD;
5972 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
5973 		error = NFSERR_LOCKSHELD;
5974 	if (error == 0)
5975 		nfsrv_freelockowner(stp, NULL, 0, p);
5976 	NFSUNLOCKSTATE();
5977 	return (error);
5978 }
5979 
5980 /*
5981  * Generate the xdr for an NFSv4.1 CBSequence Operation.
5982  */
5983 static int
nfsv4_setcbsequence(struct nfsrv_descript * nd,struct nfsclient * clp,int dont_replycache,struct nfsdsession ** sepp)5984 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
5985     int dont_replycache, struct nfsdsession **sepp)
5986 {
5987 	struct nfsdsession *sep;
5988 	uint32_t *tl, slotseq = 0;
5989 	int maxslot, slotpos;
5990 	uint8_t sessionid[NFSX_V4SESSIONID];
5991 	int error;
5992 
5993 	error = nfsv4_getcbsession(clp, sepp);
5994 	if (error != 0)
5995 		return (error);
5996 	sep = *sepp;
5997 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
5998 	    &slotseq, sessionid);
5999 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6000 
6001 	/* Build the Sequence arguments. */
6002 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6003 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6004 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6005 	nd->nd_slotseq = tl;
6006 	*tl++ = txdr_unsigned(slotseq);
6007 	*tl++ = txdr_unsigned(slotpos);
6008 	*tl++ = txdr_unsigned(maxslot);
6009 	if (dont_replycache == 0)
6010 		*tl++ = newnfs_true;
6011 	else
6012 		*tl++ = newnfs_false;
6013 	*tl = 0;			/* No referring call list, for now. */
6014 	nd->nd_flag |= ND_HASSEQUENCE;
6015 	return (0);
6016 }
6017 
6018 /*
6019  * Get a session for the callback.
6020  */
6021 static int
nfsv4_getcbsession(struct nfsclient * clp,struct nfsdsession ** sepp)6022 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6023 {
6024 	struct nfsdsession *sep;
6025 
6026 	NFSLOCKSTATE();
6027 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6028 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6029 			break;
6030 	}
6031 	if (sep == NULL) {
6032 		NFSUNLOCKSTATE();
6033 		return (NFSERR_BADSESSION);
6034 	}
6035 	sep->sess_refcnt++;
6036 	*sepp = sep;
6037 	NFSUNLOCKSTATE();
6038 	return (0);
6039 }
6040 
6041