2 * Copyright (c) 2009 Rick Macklem, University of Guelph
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
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.
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
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
32 #include <fs/nfs/nfsport.h>
34 struct nfsrv_stablefirst nfsrv_stablefirst;
35 int nfsrv_issuedelegs = 0;
36 int nfsrv_dolocallocks = 0;
37 struct nfsv4lock nfsv4rootfs_lock;
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;
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");
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");
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");
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");
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");
78 static int nfsrv_writedelegifpos = 0;
79 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
80 &nfsrv_writedelegifpos, 0,
81 "Issue a write delegation for read opens if possible");
84 * Hash lists for nfs V4.
86 struct nfsclienthashhead *nfsclienthash;
87 struct nfslockhashhead *nfslockhash;
88 struct nfssessionhash *nfssessionhash;
89 #endif /* !APPLEKEXT */
91 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
92 static time_t nfsrvboottime;
93 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
94 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
95 static int nfsrv_nogsscallback = 0;
98 static void nfsrv_dumpaclient(struct nfsclient *clp,
99 struct nfsd_dumpclients *dumpp);
100 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
102 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
104 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
106 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
107 int cansleep, NFSPROC_T *p);
108 static void nfsrv_freenfslock(struct nfslock *lop);
109 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
110 static void nfsrv_freedeleg(struct nfsstate *);
111 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
112 u_int32_t flags, struct nfsstate **stpp);
113 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
114 struct nfsstate **stpp);
115 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
116 struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
117 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
118 struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
119 static void nfsrv_insertlock(struct nfslock *new_lop,
120 struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
121 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
122 struct nfslock **other_lopp, struct nfslockfile *lfp);
123 static int nfsrv_getipnumber(u_char *cp);
124 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
125 nfsv4stateid_t *stateidp, int specialid);
126 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
128 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
129 nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
130 struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
131 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
132 uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
133 static u_int32_t nfsrv_nextclientindex(void);
134 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
135 static void nfsrv_markstable(struct nfsclient *clp);
136 static int nfsrv_checkstable(struct nfsclient *clp);
137 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
138 vnode *vp, NFSPROC_T *p);
139 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
140 NFSPROC_T *p, vnode_t vp);
141 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
142 struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
143 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
144 struct nfsclient *clp);
145 static time_t nfsrv_leaseexpiry(void);
146 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
147 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
148 struct nfsstate *stp, struct nfsrvcache *op);
149 static int nfsrv_nootherstate(struct nfsstate *stp);
150 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
151 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
152 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
153 uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
154 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
155 int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
157 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
159 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
160 uint64_t first, uint64_t end);
161 static void nfsrv_locklf(struct nfslockfile *lfp);
162 static void nfsrv_unlocklf(struct nfslockfile *lfp);
163 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
164 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
165 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
166 int dont_replycache, struct nfsdsession **sepp);
167 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
170 * Scan the client list for a match and either return the current one,
171 * create a new entry or return an error.
172 * If returning a non-error, the clp structure must either be linked into
173 * the client list or free'd.
176 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
177 nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
179 struct nfsclient *clp = NULL, *new_clp = *new_clpp;
181 struct nfsstate *stp, *tstp;
182 struct sockaddr_in *sad, *rad;
183 int zapit = 0, gotit, hasstate = 0, igotlock;
184 static u_int64_t confirm_index = 0;
187 * Check for state resource limit exceeded.
189 if (nfsrv_openpluslock > nfsrv_v4statelimit) {
190 error = NFSERR_RESOURCE;
194 if (nfsrv_issuedelegs == 0 ||
195 ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
197 * Don't do callbacks when delegations are disabled or
198 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
199 * If establishing a callback connection is attempted
200 * when a firewall is blocking the callback path, the
201 * server may wait too long for the connect attempt to
202 * succeed during the Open. Some clients, such as Linux,
203 * may timeout and give up on the Open before the server
204 * replies. Also, since AUTH_GSS callbacks are not
205 * yet interoperability tested, they might cause the
206 * server to crap out, if they get past the Init call to
209 new_clp->lc_program = 0;
211 /* Lock out other nfsd threads */
212 NFSLOCKV4ROOTMUTEX();
213 nfsv4_relref(&nfsv4rootfs_lock);
215 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
216 NFSV4ROOTLOCKMUTEXPTR, NULL);
218 NFSUNLOCKV4ROOTMUTEX();
221 * Search for a match in the client list.
224 while (i < nfsrv_clienthashsize && !gotit) {
225 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
226 if (new_clp->lc_idlen == clp->lc_idlen &&
227 !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
236 (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
237 if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
239 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
240 * client is trying to update a confirmed clientid.
242 NFSLOCKV4ROOTMUTEX();
243 nfsv4_unlock(&nfsv4rootfs_lock, 1);
244 NFSUNLOCKV4ROOTMUTEX();
245 confirmp->lval[1] = 0;
246 error = NFSERR_NOENT;
250 * Get rid of the old one.
252 if (i != nfsrv_clienthashsize) {
253 LIST_REMOVE(clp, lc_hash);
254 nfsrv_cleanclient(clp, p);
255 nfsrv_freedeleglist(&clp->lc_deleg);
256 nfsrv_freedeleglist(&clp->lc_olddeleg);
260 * Add it after assigning a client id to it.
262 new_clp->lc_flags |= LCL_NEEDSCONFIRM;
263 if ((nd->nd_flag & ND_NFSV41) != 0)
264 new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
267 confirmp->qval = new_clp->lc_confirm.qval =
269 clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
270 (u_int32_t)nfsrvboottime;
271 clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
272 nfsrv_nextclientindex();
273 new_clp->lc_stateindex = 0;
274 new_clp->lc_statemaxindex = 0;
275 new_clp->lc_cbref = 0;
276 new_clp->lc_expiry = nfsrv_leaseexpiry();
277 LIST_INIT(&new_clp->lc_open);
278 LIST_INIT(&new_clp->lc_deleg);
279 LIST_INIT(&new_clp->lc_olddeleg);
280 LIST_INIT(&new_clp->lc_session);
281 for (i = 0; i < nfsrv_statehashsize; i++)
282 LIST_INIT(&new_clp->lc_stateid[i]);
283 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
285 newnfsstats.srvclients++;
286 nfsrv_openpluslock++;
288 NFSLOCKV4ROOTMUTEX();
289 nfsv4_unlock(&nfsv4rootfs_lock, 1);
290 NFSUNLOCKV4ROOTMUTEX();
292 nfsrv_zapclient(clp, p);
298 * Now, handle the cases where the id is already issued.
300 if (nfsrv_notsamecredname(nd, clp)) {
302 * Check to see if there is expired state that should go away.
304 if (clp->lc_expiry < NFSD_MONOSEC &&
305 (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
306 nfsrv_cleanclient(clp, p);
307 nfsrv_freedeleglist(&clp->lc_deleg);
311 * If there is outstanding state, then reply NFSERR_CLIDINUSE per
312 * RFC3530 Sec. 8.1.2 last para.
314 if (!LIST_EMPTY(&clp->lc_deleg)) {
316 } else if (LIST_EMPTY(&clp->lc_open)) {
320 /* Look for an Open on the OpenOwner */
321 LIST_FOREACH(stp, &clp->lc_open, ls_list) {
322 if (!LIST_EMPTY(&stp->ls_open)) {
330 * If the uid doesn't match, return NFSERR_CLIDINUSE after
331 * filling out the correct ipaddr and portnum.
333 sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
334 rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
335 sad->sin_addr.s_addr = rad->sin_addr.s_addr;
336 sad->sin_port = rad->sin_port;
337 NFSLOCKV4ROOTMUTEX();
338 nfsv4_unlock(&nfsv4rootfs_lock, 1);
339 NFSUNLOCKV4ROOTMUTEX();
340 error = NFSERR_CLIDINUSE;
345 if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
347 * If the verifier has changed, the client has rebooted
348 * and a new client id is issued. The old state info
349 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
351 LIST_REMOVE(clp, lc_hash);
352 new_clp->lc_flags |= LCL_NEEDSCONFIRM;
353 if ((nd->nd_flag & ND_NFSV41) != 0)
354 new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
357 confirmp->qval = new_clp->lc_confirm.qval =
359 clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
361 clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
362 nfsrv_nextclientindex();
363 new_clp->lc_stateindex = 0;
364 new_clp->lc_statemaxindex = 0;
365 new_clp->lc_cbref = 0;
366 new_clp->lc_expiry = nfsrv_leaseexpiry();
369 * Save the state until confirmed.
371 LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
372 LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
373 tstp->ls_clp = new_clp;
374 LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
375 LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
376 tstp->ls_clp = new_clp;
377 LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
379 LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
380 tstp->ls_clp = new_clp;
381 for (i = 0; i < nfsrv_statehashsize; i++) {
382 LIST_NEWHEAD(&new_clp->lc_stateid[i],
383 &clp->lc_stateid[i], ls_hash);
384 LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
385 tstp->ls_clp = new_clp;
387 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
389 newnfsstats.srvclients++;
390 nfsrv_openpluslock++;
392 NFSLOCKV4ROOTMUTEX();
393 nfsv4_unlock(&nfsv4rootfs_lock, 1);
394 NFSUNLOCKV4ROOTMUTEX();
397 * Must wait until any outstanding callback on the old clp
401 while (clp->lc_cbref) {
402 clp->lc_flags |= LCL_WAKEUPWANTED;
403 (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
404 "nfsd clp", 10 * hz);
407 nfsrv_zapclient(clp, p);
412 /* For NFSv4.1, mark that we found a confirmed clientid. */
413 if ((nd->nd_flag & ND_NFSV41) != 0) {
414 clientidp->lval[0] = clp->lc_clientid.lval[0];
415 clientidp->lval[1] = clp->lc_clientid.lval[1];
416 confirmp->lval[0] = 0; /* Ignored by client */
417 confirmp->lval[1] = 1;
420 * id and verifier match, so update the net address info
421 * and get rid of any existing callback authentication
422 * handle, so a new one will be acquired.
424 LIST_REMOVE(clp, lc_hash);
425 new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
426 new_clp->lc_expiry = nfsrv_leaseexpiry();
427 confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
428 clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
429 clp->lc_clientid.lval[0];
430 clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
431 clp->lc_clientid.lval[1];
432 new_clp->lc_delegtime = clp->lc_delegtime;
433 new_clp->lc_stateindex = clp->lc_stateindex;
434 new_clp->lc_statemaxindex = clp->lc_statemaxindex;
435 new_clp->lc_cbref = 0;
436 LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
437 LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
438 tstp->ls_clp = new_clp;
439 LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
440 LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
441 tstp->ls_clp = new_clp;
442 LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
443 LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
444 tstp->ls_clp = new_clp;
445 for (i = 0; i < nfsrv_statehashsize; i++) {
446 LIST_NEWHEAD(&new_clp->lc_stateid[i],
447 &clp->lc_stateid[i], ls_hash);
448 LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
449 tstp->ls_clp = new_clp;
451 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
453 newnfsstats.srvclients++;
454 nfsrv_openpluslock++;
457 NFSLOCKV4ROOTMUTEX();
458 nfsv4_unlock(&nfsv4rootfs_lock, 1);
459 NFSUNLOCKV4ROOTMUTEX();
461 if ((nd->nd_flag & ND_NFSV41) == 0) {
463 * Must wait until any outstanding callback on the old clp
467 while (clp->lc_cbref) {
468 clp->lc_flags |= LCL_WAKEUPWANTED;
469 (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
473 nfsrv_zapclient(clp, p);
478 NFSEXITCODE2(error, nd);
483 * Check to see if the client id exists and optionally confirm it.
486 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
487 struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
488 struct nfsrv_descript *nd, NFSPROC_T *p)
490 struct nfsclient *clp;
491 struct nfsstate *stp;
493 struct nfsclienthashhead *hp;
494 int error = 0, igotlock, doneok;
495 struct nfssessionhash *shp;
496 struct nfsdsession *sep;
498 static uint64_t next_sess = 0;
502 if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
503 opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
504 error = NFSERR_STALECLIENTID;
509 * If called with opflags == CLOPS_RENEW, the State Lock is
510 * already held. Otherwise, we need to get either that or,
511 * for the case of Confirm, lock out the nfsd threads.
513 if (opflags & CLOPS_CONFIRM) {
514 NFSLOCKV4ROOTMUTEX();
515 nfsv4_relref(&nfsv4rootfs_lock);
517 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
518 NFSV4ROOTLOCKMUTEXPTR, NULL);
521 * Create a new sessionid here, since we need to do it where
522 * there is a mutex held to serialize update of next_sess.
524 if ((nd->nd_flag & ND_NFSV41) != 0) {
525 sessid[0] = ++next_sess;
526 sessid[1] = clientid.qval;
528 NFSUNLOCKV4ROOTMUTEX();
529 } else if (opflags != CLOPS_RENEW) {
533 /* For NFSv4.1, the clp is acquired from the associated session. */
534 if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
535 opflags == CLOPS_RENEW) {
537 if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
538 shp = NFSSESSIONHASH(nd->nd_sessionid);
540 sep = nfsrv_findsession(nd->nd_sessionid);
543 NFSUNLOCKSESSION(shp);
546 hp = NFSCLIENTHASH(clientid);
547 LIST_FOREACH(clp, hp, lc_hash) {
548 if (clp->lc_clientid.lval[1] == clientid.lval[1])
553 if (opflags & CLOPS_CONFIRM)
554 error = NFSERR_STALECLIENTID;
556 error = NFSERR_EXPIRED;
557 } else if (clp->lc_flags & LCL_ADMINREVOKED) {
559 * If marked admin revoked, just return the error.
561 error = NFSERR_ADMINREVOKED;
564 if (opflags & CLOPS_CONFIRM) {
565 NFSLOCKV4ROOTMUTEX();
566 nfsv4_unlock(&nfsv4rootfs_lock, 1);
567 NFSUNLOCKV4ROOTMUTEX();
568 } else if (opflags != CLOPS_RENEW) {
575 * Perform any operations specified by the opflags.
577 if (opflags & CLOPS_CONFIRM) {
578 if (((nd->nd_flag & ND_NFSV41) != 0 &&
579 clp->lc_confirm.lval[0] != confirm.lval[0]) ||
580 ((nd->nd_flag & ND_NFSV41) == 0 &&
581 clp->lc_confirm.qval != confirm.qval))
582 error = NFSERR_STALECLIENTID;
583 else if (nfsrv_notsamecredname(nd, clp))
584 error = NFSERR_CLIDINUSE;
587 if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
590 * Hang onto the delegations (as old delegations)
591 * for an Open with CLAIM_DELEGATE_PREV unless in
592 * grace, but get rid of the rest of the state.
594 nfsrv_cleanclient(clp, p);
595 nfsrv_freedeleglist(&clp->lc_olddeleg);
596 if (nfsrv_checkgrace(nd, clp, 0)) {
597 /* In grace, so just delete delegations */
598 nfsrv_freedeleglist(&clp->lc_deleg);
600 LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
601 stp->ls_flags |= NFSLCK_OLDDELEG;
602 clp->lc_delegtime = NFSD_MONOSEC +
603 nfsrv_lease + NFSRV_LEASEDELTA;
604 LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
607 if ((nd->nd_flag & ND_NFSV41) != 0)
608 clp->lc_program = cbprogram;
610 clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
612 clp->lc_flags |= LCL_NEEDSCBNULL;
613 /* For NFSv4.1, link the session onto the client. */
615 /* Hold a reference on the xprt for a backchannel. */
616 if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
617 != 0 && clp->lc_req.nr_client == NULL) {
618 clp->lc_req.nr_client = (struct __rpc_client *)
619 clnt_bck_create(nd->nd_xprt->xp_socket,
620 cbprogram, NFSV4_CBVERS);
621 if (clp->lc_req.nr_client != NULL) {
622 SVC_ACQUIRE(nd->nd_xprt);
624 clp->lc_req.nr_client->cl_private;
625 /* Disable idle timeout. */
626 nd->nd_xprt->xp_idletimeout = 0;
627 nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
629 nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
631 NFSBCOPY(sessid, nsep->sess_sessionid,
633 NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
635 shp = NFSSESSIONHASH(nsep->sess_sessionid);
638 LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
639 LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
640 nsep->sess_clp = clp;
641 NFSUNLOCKSESSION(shp);
645 } else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
646 error = NFSERR_EXPIRED;
650 * If called by the Renew Op, we must check the principal.
652 if (!error && (opflags & CLOPS_RENEWOP)) {
653 if (nfsrv_notsamecredname(nd, clp)) {
655 for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
656 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
657 if ((stp->ls_flags & NFSLCK_OPEN) &&
658 stp->ls_uid == nd->nd_cred->cr_uid) {
665 error = NFSERR_ACCES;
667 if (!error && (clp->lc_flags & LCL_CBDOWN))
668 error = NFSERR_CBPATHDOWN;
670 if ((!error || error == NFSERR_CBPATHDOWN) &&
671 (opflags & CLOPS_RENEW)) {
672 clp->lc_expiry = nfsrv_leaseexpiry();
674 if (opflags & CLOPS_CONFIRM) {
675 NFSLOCKV4ROOTMUTEX();
676 nfsv4_unlock(&nfsv4rootfs_lock, 1);
677 NFSUNLOCKV4ROOTMUTEX();
678 } else if (opflags != CLOPS_RENEW) {
685 NFSEXITCODE2(error, nd);
690 * Perform the NFSv4.1 destroy clientid.
693 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
695 struct nfsclient *clp;
696 struct nfsclienthashhead *hp;
697 int error = 0, i, igotlock;
699 if (nfsrvboottime != clientid.lval[0]) {
700 error = NFSERR_STALECLIENTID;
704 /* Lock out other nfsd threads */
705 NFSLOCKV4ROOTMUTEX();
706 nfsv4_relref(&nfsv4rootfs_lock);
708 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
709 NFSV4ROOTLOCKMUTEXPTR, NULL);
710 } while (igotlock == 0);
711 NFSUNLOCKV4ROOTMUTEX();
713 hp = NFSCLIENTHASH(clientid);
714 LIST_FOREACH(clp, hp, lc_hash) {
715 if (clp->lc_clientid.lval[1] == clientid.lval[1])
719 NFSLOCKV4ROOTMUTEX();
720 nfsv4_unlock(&nfsv4rootfs_lock, 1);
721 NFSUNLOCKV4ROOTMUTEX();
722 /* Just return ok, since it is gone. */
726 /* Scan for state on the clientid. */
727 for (i = 0; i < nfsrv_statehashsize; i++)
728 if (!LIST_EMPTY(&clp->lc_stateid[i])) {
729 NFSLOCKV4ROOTMUTEX();
730 nfsv4_unlock(&nfsv4rootfs_lock, 1);
731 NFSUNLOCKV4ROOTMUTEX();
732 error = NFSERR_CLIENTIDBUSY;
735 if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
736 NFSLOCKV4ROOTMUTEX();
737 nfsv4_unlock(&nfsv4rootfs_lock, 1);
738 NFSUNLOCKV4ROOTMUTEX();
739 error = NFSERR_CLIENTIDBUSY;
743 /* Destroy the clientid and return ok. */
744 nfsrv_cleanclient(clp, p);
745 nfsrv_freedeleglist(&clp->lc_deleg);
746 nfsrv_freedeleglist(&clp->lc_olddeleg);
747 LIST_REMOVE(clp, lc_hash);
748 NFSLOCKV4ROOTMUTEX();
749 nfsv4_unlock(&nfsv4rootfs_lock, 1);
750 NFSUNLOCKV4ROOTMUTEX();
751 nfsrv_zapclient(clp, p);
753 NFSEXITCODE2(error, nd);
758 * Called from the new nfssvc syscall to admin revoke a clientid.
759 * Returns 0 for success, error otherwise.
762 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
764 struct nfsclient *clp = NULL;
769 * First, lock out the nfsd so that state won't change while the
770 * revocation record is being written to the stable storage restart
773 NFSLOCKV4ROOTMUTEX();
775 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
776 NFSV4ROOTLOCKMUTEXPTR, NULL);
778 NFSUNLOCKV4ROOTMUTEX();
781 * Search for a match in the client list.
784 while (i < nfsrv_clienthashsize && !gotit) {
785 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
786 if (revokep->nclid_idlen == clp->lc_idlen &&
787 !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
795 NFSLOCKV4ROOTMUTEX();
796 nfsv4_unlock(&nfsv4rootfs_lock, 0);
797 NFSUNLOCKV4ROOTMUTEX();
803 * Now, write out the revocation record
805 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
806 nfsrv_backupstable();
809 * and clear out the state, marking the clientid revoked.
811 clp->lc_flags &= ~LCL_CALLBACKSON;
812 clp->lc_flags |= LCL_ADMINREVOKED;
813 nfsrv_cleanclient(clp, p);
814 nfsrv_freedeleglist(&clp->lc_deleg);
815 nfsrv_freedeleglist(&clp->lc_olddeleg);
816 NFSLOCKV4ROOTMUTEX();
817 nfsv4_unlock(&nfsv4rootfs_lock, 0);
818 NFSUNLOCKV4ROOTMUTEX();
826 * Dump out stats for all clients. Called from nfssvc(2), that is used
830 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
832 struct nfsclient *clp;
836 * First, get a reference on the nfsv4rootfs_lock so that an
837 * exclusive lock cannot be acquired while dumping the clients.
839 NFSLOCKV4ROOTMUTEX();
840 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
841 NFSUNLOCKV4ROOTMUTEX();
844 * Rattle through the client lists until done.
846 while (i < nfsrv_clienthashsize && cnt < maxcnt) {
847 clp = LIST_FIRST(&nfsclienthash[i]);
848 while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
849 nfsrv_dumpaclient(clp, &dumpp[cnt]);
851 clp = LIST_NEXT(clp, lc_hash);
856 dumpp[cnt].ndcl_clid.nclid_idlen = 0;
858 NFSLOCKV4ROOTMUTEX();
859 nfsv4_relref(&nfsv4rootfs_lock);
860 NFSUNLOCKV4ROOTMUTEX();
864 * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
867 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
869 struct nfsstate *stp, *openstp, *lckownstp;
871 struct sockaddr *sad;
872 struct sockaddr_in *rad;
873 struct sockaddr_in6 *rad6;
875 dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
876 dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
877 dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
878 dumpp->ndcl_flags = clp->lc_flags;
879 dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
880 NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
881 sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
882 dumpp->ndcl_addrfam = sad->sa_family;
883 if (sad->sa_family == AF_INET) {
884 rad = (struct sockaddr_in *)sad;
885 dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
887 rad6 = (struct sockaddr_in6 *)sad;
888 dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
892 * Now, scan the state lists and total up the opens and locks.
894 LIST_FOREACH(stp, &clp->lc_open, ls_list) {
895 dumpp->ndcl_nopenowners++;
896 LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
897 dumpp->ndcl_nopens++;
898 LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
899 dumpp->ndcl_nlockowners++;
900 LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
901 dumpp->ndcl_nlocks++;
908 * and the delegation lists.
910 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
911 dumpp->ndcl_ndelegs++;
913 LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
914 dumpp->ndcl_nolddelegs++;
919 * Dump out lock stats for a file.
922 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
925 struct nfsstate *stp;
928 struct nfslockfile *lfp;
929 struct sockaddr *sad;
930 struct sockaddr_in *rad;
931 struct sockaddr_in6 *rad6;
935 ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
937 * First, get a reference on the nfsv4rootfs_lock so that an
938 * exclusive lock on it cannot be acquired while dumping the locks.
940 NFSLOCKV4ROOTMUTEX();
941 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
942 NFSUNLOCKV4ROOTMUTEX();
945 ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
947 ldumpp[0].ndlck_clid.nclid_idlen = 0;
949 NFSLOCKV4ROOTMUTEX();
950 nfsv4_relref(&nfsv4rootfs_lock);
951 NFSUNLOCKV4ROOTMUTEX();
956 * For each open share on file, dump it out.
958 stp = LIST_FIRST(&lfp->lf_open);
959 while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
960 ldumpp[cnt].ndlck_flags = stp->ls_flags;
961 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
962 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
963 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
964 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
965 ldumpp[cnt].ndlck_owner.nclid_idlen =
966 stp->ls_openowner->ls_ownerlen;
967 NFSBCOPY(stp->ls_openowner->ls_owner,
968 ldumpp[cnt].ndlck_owner.nclid_id,
969 stp->ls_openowner->ls_ownerlen);
970 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
971 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
972 stp->ls_clp->lc_idlen);
973 sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
974 ldumpp[cnt].ndlck_addrfam = sad->sa_family;
975 if (sad->sa_family == AF_INET) {
976 rad = (struct sockaddr_in *)sad;
977 ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
979 rad6 = (struct sockaddr_in6 *)sad;
980 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
982 stp = LIST_NEXT(stp, ls_file);
989 lop = LIST_FIRST(&lfp->lf_lock);
990 while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
992 ldumpp[cnt].ndlck_flags = lop->lo_flags;
993 ldumpp[cnt].ndlck_first = lop->lo_first;
994 ldumpp[cnt].ndlck_end = lop->lo_end;
995 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
996 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
997 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
998 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
999 ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1000 NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1002 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1003 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1004 stp->ls_clp->lc_idlen);
1005 sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1006 ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1007 if (sad->sa_family == AF_INET) {
1008 rad = (struct sockaddr_in *)sad;
1009 ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1011 rad6 = (struct sockaddr_in6 *)sad;
1012 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1014 lop = LIST_NEXT(lop, lo_lckfile);
1019 * and the delegations.
1021 stp = LIST_FIRST(&lfp->lf_deleg);
1022 while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1023 ldumpp[cnt].ndlck_flags = stp->ls_flags;
1024 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1025 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1026 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1027 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1028 ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1029 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1030 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1031 stp->ls_clp->lc_idlen);
1032 sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1033 ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1034 if (sad->sa_family == AF_INET) {
1035 rad = (struct sockaddr_in *)sad;
1036 ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1038 rad6 = (struct sockaddr_in6 *)sad;
1039 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1041 stp = LIST_NEXT(stp, ls_file);
1046 * If list isn't full, mark end of list by setting the client name
1050 ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1052 NFSLOCKV4ROOTMUTEX();
1053 nfsv4_relref(&nfsv4rootfs_lock);
1054 NFSUNLOCKV4ROOTMUTEX();
1058 * Server timer routine. It can scan any linked list, so long
1059 * as it holds the spin/mutex lock and there is no exclusive lock on
1061 * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1062 * to do this from a callout, since the spin locks work. For
1063 * Darwin, I'm not sure what will work correctly yet.)
1064 * Should be called once per second.
1067 nfsrv_servertimer(void)
1069 struct nfsclient *clp, *nclp;
1070 struct nfsstate *stp, *nstp;
1074 * Make sure nfsboottime is set. This is used by V3 as well
1075 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1076 * only used by the V4 server for leases.
1078 if (nfsboottime.tv_sec == 0)
1079 NFSSETBOOTTIME(nfsboottime);
1082 * If server hasn't started yet, just return.
1085 if (nfsrv_stablefirst.nsf_eograce == 0) {
1089 if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1090 if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1091 NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1092 nfsrv_stablefirst.nsf_flags |=
1093 (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1099 * Try and get a reference count on the nfsv4rootfs_lock so that
1100 * no nfsd thread can acquire an exclusive lock on it before this
1101 * call is done. If it is already exclusively locked, just return.
1103 NFSLOCKV4ROOTMUTEX();
1104 got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1105 NFSUNLOCKV4ROOTMUTEX();
1112 * For each client...
1114 for (i = 0; i < nfsrv_clienthashsize; i++) {
1115 clp = LIST_FIRST(&nfsclienthash[i]);
1116 while (clp != LIST_END(&nfsclienthash[i])) {
1117 nclp = LIST_NEXT(clp, lc_hash);
1118 if (!(clp->lc_flags & LCL_EXPIREIT)) {
1119 if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1120 && ((LIST_EMPTY(&clp->lc_deleg)
1121 && LIST_EMPTY(&clp->lc_open)) ||
1122 nfsrv_clients > nfsrv_clienthighwater)) ||
1123 (clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1124 (clp->lc_expiry < NFSD_MONOSEC &&
1125 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1127 * Lease has expired several nfsrv_lease times ago:
1129 * - no state is associated with it
1131 * - above high water mark for number of clients
1132 * (nfsrv_clienthighwater should be large enough
1133 * that this only occurs when clients fail to
1134 * use the same nfs_client_id4.id. Maybe somewhat
1135 * higher that the maximum number of clients that
1136 * will mount this server?)
1138 * Lease has expired a very long time ago
1140 * Lease has expired PLUS the number of opens + locks
1141 * has exceeded 90% of capacity
1143 * --> Mark for expiry. The actual expiry will be done
1144 * by an nfsd sometime soon.
1146 clp->lc_flags |= LCL_EXPIREIT;
1147 nfsrv_stablefirst.nsf_flags |=
1148 (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1151 * If there are no opens, increment no open tick cnt
1152 * If time exceeds NFSNOOPEN, mark it to be thrown away
1153 * otherwise, if there is an open, reset no open time
1154 * Hopefully, this will avoid excessive re-creation
1155 * of open owners and subsequent open confirms.
1157 stp = LIST_FIRST(&clp->lc_open);
1158 while (stp != LIST_END(&clp->lc_open)) {
1159 nstp = LIST_NEXT(stp, ls_list);
1160 if (LIST_EMPTY(&stp->ls_open)) {
1162 if (stp->ls_noopens > NFSNOOPEN ||
1163 (nfsrv_openpluslock * 2) >
1165 nfsrv_stablefirst.nsf_flags |=
1168 stp->ls_noopens = 0;
1178 NFSLOCKV4ROOTMUTEX();
1179 nfsv4_relref(&nfsv4rootfs_lock);
1180 NFSUNLOCKV4ROOTMUTEX();
1184 * The following set of functions free up the various data structures.
1187 * Clear out all open/lock state related to this nfsclient.
1188 * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1189 * there are no other active nfsd threads.
1192 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1194 struct nfsstate *stp, *nstp;
1195 struct nfsdsession *sep, *nsep;
1197 LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1198 nfsrv_freeopenowner(stp, 1, p);
1199 if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1200 LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1201 (void)nfsrv_freesession(sep, NULL);
1205 * Free a client that has been cleaned. It should also already have been
1206 * removed from the lists.
1207 * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1208 * softclock interrupts are enabled.)
1211 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1215 if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1216 (LCL_GSS | LCL_CALLBACKSON) &&
1217 (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1218 clp->lc_handlelen > 0) {
1219 clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1220 clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1221 (void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1222 NULL, 0, NULL, NULL, NULL, p);
1225 newnfs_disconnect(&clp->lc_req);
1226 NFSSOCKADDRFREE(clp->lc_req.nr_nam);
1227 NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1228 free(clp->lc_stateid, M_NFSDCLIENT);
1229 free(clp, M_NFSDCLIENT);
1231 newnfsstats.srvclients--;
1232 nfsrv_openpluslock--;
1238 * Free a list of delegation state structures.
1239 * (This function will also free all nfslockfile structures that no
1240 * longer have associated state.)
1243 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1245 struct nfsstate *stp, *nstp;
1247 LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1248 nfsrv_freedeleg(stp);
1254 * Free up a delegation.
1257 nfsrv_freedeleg(struct nfsstate *stp)
1259 struct nfslockfile *lfp;
1261 LIST_REMOVE(stp, ls_hash);
1262 LIST_REMOVE(stp, ls_list);
1263 LIST_REMOVE(stp, ls_file);
1265 if (LIST_EMPTY(&lfp->lf_open) &&
1266 LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1267 LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1268 lfp->lf_usecount == 0 &&
1269 nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1270 nfsrv_freenfslockfile(lfp);
1271 FREE((caddr_t)stp, M_NFSDSTATE);
1272 newnfsstats.srvdelegates--;
1273 nfsrv_openpluslock--;
1274 nfsrv_delegatecnt--;
1278 * This function frees an open owner and all associated opens.
1281 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1283 struct nfsstate *nstp, *tstp;
1285 LIST_REMOVE(stp, ls_list);
1287 * Now, free all associated opens.
1289 nstp = LIST_FIRST(&stp->ls_open);
1290 while (nstp != LIST_END(&stp->ls_open)) {
1292 nstp = LIST_NEXT(nstp, ls_list);
1293 (void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1296 nfsrvd_derefcache(stp->ls_op);
1297 FREE((caddr_t)stp, M_NFSDSTATE);
1298 newnfsstats.srvopenowners--;
1299 nfsrv_openpluslock--;
1303 * This function frees an open (nfsstate open structure) with all associated
1304 * lock_owners and locks. It also frees the nfslockfile structure iff there
1305 * are no other opens on the file.
1306 * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1309 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1311 struct nfsstate *nstp, *tstp;
1312 struct nfslockfile *lfp;
1315 LIST_REMOVE(stp, ls_hash);
1316 LIST_REMOVE(stp, ls_list);
1317 LIST_REMOVE(stp, ls_file);
1321 * Now, free all lockowners associated with this open.
1323 LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1324 nfsrv_freelockowner(tstp, vp, cansleep, p);
1327 * The nfslockfile is freed here if there are no locks
1328 * associated with the open.
1329 * If there are locks associated with the open, the
1330 * nfslockfile structure can be freed via nfsrv_freelockowner().
1331 * Acquire the state mutex to avoid races with calls to
1332 * nfsrv_getlockfile().
1336 if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1337 LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1338 LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1339 lfp->lf_usecount == 0 &&
1340 (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1341 nfsrv_freenfslockfile(lfp);
1347 FREE((caddr_t)stp, M_NFSDSTATE);
1348 newnfsstats.srvopens--;
1349 nfsrv_openpluslock--;
1354 * Frees a lockowner and all associated locks.
1357 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1361 LIST_REMOVE(stp, ls_hash);
1362 LIST_REMOVE(stp, ls_list);
1363 nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1365 nfsrvd_derefcache(stp->ls_op);
1366 FREE((caddr_t)stp, M_NFSDSTATE);
1367 newnfsstats.srvlockowners--;
1368 nfsrv_openpluslock--;
1372 * Free all the nfs locks on a lockowner.
1375 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1378 struct nfslock *lop, *nlop;
1379 struct nfsrollback *rlp, *nrlp;
1380 struct nfslockfile *lfp = NULL;
1383 uint64_t first, end;
1386 ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1387 lop = LIST_FIRST(&stp->ls_lock);
1388 while (lop != LIST_END(&stp->ls_lock)) {
1389 nlop = LIST_NEXT(lop, lo_lckowner);
1391 * Since all locks should be for the same file, lfp should
1396 else if (lfp != lop->lo_lfp)
1397 panic("allnfslocks");
1399 * If vp is NULL and cansleep != 0, a vnode must be acquired
1400 * from the file handle. This only occurs when called from
1401 * nfsrv_cleanclient().
1404 if (nfsrv_dolocallocks == 0)
1406 else if (vp == NULL && cansleep != 0) {
1407 tvp = nfsvno_getvp(&lfp->lf_fh);
1408 NFSVOPUNLOCK(tvp, 0);
1417 first = lop->lo_first;
1419 nfsrv_freenfslock(lop);
1420 nfsrv_localunlock(tvp, lfp, first, end, p);
1421 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1423 free(rlp, M_NFSDROLLBACK);
1424 LIST_INIT(&lfp->lf_rollback);
1426 nfsrv_freenfslock(lop);
1429 if (vp == NULL && tvp != NULL)
1434 * Free an nfslock structure.
1437 nfsrv_freenfslock(struct nfslock *lop)
1440 if (lop->lo_lckfile.le_prev != NULL) {
1441 LIST_REMOVE(lop, lo_lckfile);
1442 newnfsstats.srvlocks--;
1443 nfsrv_openpluslock--;
1445 LIST_REMOVE(lop, lo_lckowner);
1446 FREE((caddr_t)lop, M_NFSDLOCK);
1450 * This function frees an nfslockfile structure.
1453 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1456 LIST_REMOVE(lfp, lf_hash);
1457 FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1461 * This function looks up an nfsstate structure via stateid.
1464 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1465 struct nfsstate **stpp)
1467 struct nfsstate *stp;
1468 struct nfsstatehead *hp;
1472 hp = NFSSTATEHASH(clp, *stateidp);
1473 LIST_FOREACH(stp, hp, ls_hash) {
1474 if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1480 * If no state id in list, return NFSERR_BADSTATEID.
1482 if (stp == LIST_END(hp)) {
1483 error = NFSERR_BADSTATEID;
1494 * This function gets an nfsstate structure via owner string.
1497 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1498 struct nfsstate **stpp)
1500 struct nfsstate *stp;
1503 LIST_FOREACH(stp, hp, ls_list) {
1504 if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1505 !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1513 * Lock control function called to update lock status.
1514 * Returns 0 upon success, -1 if there is no lock and the flags indicate
1515 * that one isn't to be created and an NFSERR_xxx for other errors.
1516 * The structures new_stp and new_lop are passed in as pointers that should
1517 * be set to NULL if the structure is used and shouldn't be free'd.
1518 * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1519 * never used and can safely be allocated on the stack. For all other
1520 * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1521 * in case they are used.
1524 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1525 struct nfslock **new_lopp, struct nfslockconflict *cfp,
1526 nfsquad_t clientid, nfsv4stateid_t *stateidp,
1527 __unused struct nfsexstuff *exp,
1528 struct nfsrv_descript *nd, NFSPROC_T *p)
1530 struct nfslock *lop;
1531 struct nfsstate *new_stp = *new_stpp;
1532 struct nfslock *new_lop = *new_lopp;
1533 struct nfsstate *tstp, *mystp, *nstp;
1535 struct nfslockfile *lfp;
1536 struct nfslock *other_lop = NULL;
1537 struct nfsstate *stp, *lckstp = NULL;
1538 struct nfsclient *clp = NULL;
1540 int error = 0, haslock = 0, ret, reterr;
1541 int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1543 uint64_t first, end;
1544 uint32_t lock_flags;
1546 if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1548 * Note the special cases of "all 1s" or "all 0s" stateids and
1549 * let reads with all 1s go ahead.
1551 if (new_stp->ls_stateid.seqid == 0x0 &&
1552 new_stp->ls_stateid.other[0] == 0x0 &&
1553 new_stp->ls_stateid.other[1] == 0x0 &&
1554 new_stp->ls_stateid.other[2] == 0x0)
1556 else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1557 new_stp->ls_stateid.other[0] == 0xffffffff &&
1558 new_stp->ls_stateid.other[1] == 0xffffffff &&
1559 new_stp->ls_stateid.other[2] == 0xffffffff)
1564 * Check for restart conditions (client and server).
1566 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1567 &new_stp->ls_stateid, specialid);
1572 * Check for state resource limit exceeded.
1574 if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1575 nfsrv_openpluslock > nfsrv_v4statelimit) {
1576 error = NFSERR_RESOURCE;
1581 * For the lock case, get another nfslock structure,
1582 * just in case we need it.
1583 * Malloc now, before we start sifting through the linked lists,
1584 * in case we have to wait for memory.
1587 if (new_stp->ls_flags & NFSLCK_LOCK)
1588 MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1589 M_NFSDLOCK, M_WAITOK);
1590 filestruct_locked = 0;
1595 * Get the lockfile structure for CFH now, so we can do a sanity
1596 * check against the stateid, before incrementing the seqid#, since
1597 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1598 * shouldn't be incremented for this case.
1599 * If nfsrv_getlockfile() returns -1, it means "not found", which
1600 * will be handled later.
1601 * If we are doing Lock/LockU and local locking is enabled, sleep
1602 * lock the nfslockfile structure.
1604 getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1606 if (getlckret == 0) {
1607 if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1608 nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1609 getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1612 filestruct_locked = 1;
1614 getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1617 if (getlckret != 0 && getlckret != -1)
1620 if (filestruct_locked != 0) {
1621 LIST_INIT(&lfp->lf_rollback);
1622 if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1624 * For local locking, do the advisory locking now, so
1625 * that any conflict can be detected. A failure later
1626 * can be rolled back locally. If an error is returned,
1627 * struct nfslockfile has been unlocked and any local
1628 * locking rolled back.
1631 if (vnode_unlocked == 0) {
1632 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1634 NFSVOPUNLOCK(vp, 0);
1636 reterr = nfsrv_locallock(vp, lfp,
1637 (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1638 new_lop->lo_first, new_lop->lo_end, cfp, p);
1643 if (specialid == 0) {
1644 if (new_stp->ls_flags & NFSLCK_TEST) {
1646 * RFC 3530 does not list LockT as an op that renews a
1647 * lease, but the concensus seems to be that it is ok
1648 * for a server to do so.
1650 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1651 (nfsquad_t)((u_quad_t)0), 0, nd, p);
1654 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1655 * error returns for LockT, just go ahead and test for a lock,
1656 * since there are no locks for this client, but other locks
1657 * can conflict. (ie. same client will always be false)
1659 if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1663 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1664 (nfsquad_t)((u_quad_t)0), 0, nd, p);
1667 * Look up the stateid
1669 error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1670 new_stp->ls_flags, &stp);
1672 * do some sanity checks for an unconfirmed open or a
1673 * stateid that refers to the wrong file, for an open stateid
1675 if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1676 ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1677 (getlckret == 0 && stp->ls_lfp != lfp)))
1678 error = NFSERR_BADSTATEID;
1680 (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1681 getlckret == 0 && stp->ls_lfp != lfp)
1682 error = NFSERR_BADSTATEID;
1685 * If the lockowner stateid doesn't refer to the same file,
1686 * I believe that is considered ok, since some clients will
1687 * only create a single lockowner and use that for all locks
1689 * For now, log it as a diagnostic, instead of considering it
1692 if (error == 0 && (stp->ls_flags &
1693 (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1694 getlckret == 0 && stp->ls_lfp != lfp) {
1696 printf("Got a lock statid for different file open\n");
1699 error = NFSERR_BADSTATEID;
1704 if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1706 * If haslock set, we've already checked the seqid.
1709 if (stp->ls_flags & NFSLCK_OPEN)
1710 error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1711 stp->ls_openowner, new_stp->ls_op);
1713 error = NFSERR_BADSTATEID;
1716 nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1719 * I believe this should be an error, but it
1720 * isn't obvious what NFSERR_xxx would be
1721 * appropriate, so I'll use NFSERR_INVAL for now.
1723 error = NFSERR_INVAL;
1726 } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1728 * If haslock set, ditto above.
1731 if (stp->ls_flags & NFSLCK_OPEN)
1732 error = NFSERR_BADSTATEID;
1734 error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1735 stp, new_stp->ls_op);
1743 * If the seqid part of the stateid isn't the same, return
1744 * NFSERR_OLDSTATEID for cases other than I/O Ops.
1745 * For I/O Ops, only return NFSERR_OLDSTATEID if
1746 * nfsrv_returnoldstateid is set. (The concensus on the email
1747 * list was that most clients would prefer to not receive
1748 * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1749 * is what will happen, so I use the nfsrv_returnoldstateid to
1750 * allow for either server configuration.)
1752 if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1753 (((nd->nd_flag & ND_NFSV41) == 0 &&
1754 (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1755 nfsrv_returnoldstateid)) ||
1756 ((nd->nd_flag & ND_NFSV41) != 0 &&
1757 new_stp->ls_stateid.seqid != 0)))
1758 error = NFSERR_OLDSTATEID;
1763 * Now we can check for grace.
1766 error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1767 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1768 nfsrv_checkstable(clp))
1769 error = NFSERR_NOGRACE;
1771 * If we successfully Reclaimed state, note that.
1773 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1774 nfsrv_markstable(clp);
1777 * At this point, either error == NFSERR_BADSTATEID or the
1778 * seqid# has been updated, so we can return any error.
1779 * If error == 0, there may be an error in:
1780 * nd_repstat - Set by the calling function.
1781 * reterr - Set above, if getting the nfslockfile structure
1782 * or acquiring the local lock failed.
1783 * (If both of these are set, nd_repstat should probably be
1784 * returned, since that error was detected before this
1787 if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1789 if (nd->nd_repstat != 0)
1790 error = nd->nd_repstat;
1794 if (filestruct_locked != 0) {
1795 /* Roll back local locks. */
1797 if (vnode_unlocked == 0) {
1798 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1800 NFSVOPUNLOCK(vp, 0);
1802 nfsrv_locallock_rollback(vp, lfp, p);
1804 nfsrv_unlocklf(lfp);
1811 * Check the nfsrv_getlockfile return.
1812 * Returned -1 if no structure found.
1814 if (getlckret == -1) {
1815 error = NFSERR_EXPIRED;
1817 * Called from lockt, so no lock is OK.
1819 if (new_stp->ls_flags & NFSLCK_TEST) {
1821 } else if (new_stp->ls_flags &
1822 (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1824 * Called to check for a lock, OK if the stateid is all
1825 * 1s or all 0s, but there should be an nfsstate
1827 * (ie. If there is no open, I'll assume no share
1833 error = NFSERR_BADSTATEID;
1840 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1841 * For NFSLCK_CHECK, allow a read if write access is granted,
1842 * but check for a deny. For NFSLCK_LOCK, require correct access,
1843 * which implies a conflicting deny can't exist.
1845 if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1847 * Four kinds of state id:
1848 * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1849 * - stateid for an open
1850 * - stateid for a delegation
1851 * - stateid for a lock owner
1854 if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1857 nfsrv_delaydelegtimeout(stp);
1858 } else if (stp->ls_flags & NFSLCK_OPEN) {
1861 mystp = stp->ls_openstp;
1864 * If locking or checking, require correct access
1867 if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1868 !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1869 mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1870 ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1871 (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1872 !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1873 ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1874 (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1875 !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1876 if (filestruct_locked != 0) {
1877 /* Roll back local locks. */
1879 if (vnode_unlocked == 0) {
1880 ASSERT_VOP_ELOCKED(vp,
1883 NFSVOPUNLOCK(vp, 0);
1885 nfsrv_locallock_rollback(vp, lfp, p);
1887 nfsrv_unlocklf(lfp);
1890 error = NFSERR_OPENMODE;
1895 if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1897 * Check for a conflicting deny bit.
1899 LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1900 if (tstp != mystp) {
1901 bits = tstp->ls_flags;
1902 bits >>= NFSLCK_SHIFT;
1903 if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1904 KASSERT(vnode_unlocked == 0,
1905 ("nfsrv_lockctrl: vnode unlocked1"));
1906 ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1910 * nfsrv_clientconflict unlocks state
1911 * when it returns non-zero.
1919 error = NFSERR_PERM;
1921 error = NFSERR_OPENMODE;
1927 /* We're outta here */
1934 * For setattr, just get rid of all the Delegations for other clients.
1936 if (new_stp->ls_flags & NFSLCK_SETATTR) {
1937 KASSERT(vnode_unlocked == 0,
1938 ("nfsrv_lockctrl: vnode unlocked2"));
1939 ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1942 * nfsrv_cleandeleg() unlocks state when it
1952 if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1953 (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1954 LIST_EMPTY(&lfp->lf_deleg))) {
1961 * Check for a conflicting delegation. If one is found, call
1962 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1963 * been set yet, it will get the lock. Otherwise, it will recall
1964 * the delegation. Then, we try try again...
1965 * I currently believe the conflict algorithm to be:
1966 * For Lock Ops (Lock/LockT/LockU)
1967 * - there is a conflict iff a different client has a write delegation
1968 * For Reading (Read Op)
1969 * - there is a conflict iff a different client has a write delegation
1970 * (the specialids are always a different client)
1971 * For Writing (Write/Setattr of size)
1972 * - there is a conflict if a different client has any delegation
1973 * - there is a conflict if the same client has a read delegation
1974 * (I don't understand why this isn't allowed, but that seems to be
1975 * the current concensus?)
1977 tstp = LIST_FIRST(&lfp->lf_deleg);
1978 while (tstp != LIST_END(&lfp->lf_deleg)) {
1979 nstp = LIST_NEXT(tstp, ls_file);
1980 if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1981 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1982 (new_lop->lo_flags & NFSLCK_READ))) &&
1983 clp != tstp->ls_clp &&
1984 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1985 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1986 (new_lop->lo_flags & NFSLCK_WRITE) &&
1987 (clp != tstp->ls_clp ||
1988 (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1990 if (filestruct_locked != 0) {
1991 /* Roll back local locks. */
1993 if (vnode_unlocked == 0) {
1994 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
1995 NFSVOPUNLOCK(vp, 0);
1997 nfsrv_locallock_rollback(vp, lfp, p);
1999 nfsrv_unlocklf(lfp);
2001 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2003 if ((vp->v_iflag & VI_DOOMED) != 0)
2004 ret = NFSERR_SERVERFAULT;
2008 ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2011 * nfsrv_delegconflict unlocks state when it
2012 * returns non-zero, which it always does.
2015 FREE((caddr_t)other_lop, M_NFSDLOCK);
2025 /* Never gets here. */
2031 * Handle the unlock case by calling nfsrv_updatelock().
2032 * (Should I have done some access checking above for unlock? For now,
2033 * just let it happen.)
2035 if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2036 first = new_lop->lo_first;
2037 end = new_lop->lo_end;
2038 nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2039 stateidp->seqid = ++(stp->ls_stateid.seqid);
2040 if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2041 stateidp->seqid = stp->ls_stateid.seqid = 1;
2042 stateidp->other[0] = stp->ls_stateid.other[0];
2043 stateidp->other[1] = stp->ls_stateid.other[1];
2044 stateidp->other[2] = stp->ls_stateid.other[2];
2045 if (filestruct_locked != 0) {
2047 if (vnode_unlocked == 0) {
2048 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2050 NFSVOPUNLOCK(vp, 0);
2052 /* Update the local locks. */
2053 nfsrv_localunlock(vp, lfp, first, end, p);
2055 nfsrv_unlocklf(lfp);
2062 * Search for a conflicting lock. A lock conflicts if:
2063 * - the lock range overlaps and
2064 * - at least one lock is a write lock and
2065 * - it is not owned by the same lock owner
2068 LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2069 if (new_lop->lo_end > lop->lo_first &&
2070 new_lop->lo_first < lop->lo_end &&
2071 (new_lop->lo_flags == NFSLCK_WRITE ||
2072 lop->lo_flags == NFSLCK_WRITE) &&
2073 lckstp != lop->lo_stp &&
2074 (clp != lop->lo_stp->ls_clp ||
2075 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2076 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2077 lckstp->ls_ownerlen))) {
2079 FREE((caddr_t)other_lop, M_NFSDLOCK);
2082 if (vnode_unlocked != 0)
2083 ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2086 ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2089 if (filestruct_locked != 0) {
2090 if (vnode_unlocked == 0) {
2091 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2092 NFSVOPUNLOCK(vp, 0);
2094 /* Roll back local locks. */
2095 nfsrv_locallock_rollback(vp, lfp, p);
2097 nfsrv_unlocklf(lfp);
2099 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2101 if ((vp->v_iflag & VI_DOOMED) != 0) {
2102 error = NFSERR_SERVERFAULT;
2107 * nfsrv_clientconflict() unlocks state when it
2114 * Found a conflicting lock, so record the conflict and
2117 if (cfp != NULL && ret == 0) {
2118 cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2119 cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2120 cfp->cl_first = lop->lo_first;
2121 cfp->cl_end = lop->lo_end;
2122 cfp->cl_flags = lop->lo_flags;
2123 cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2124 NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2128 error = NFSERR_PERM;
2129 else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2130 error = NFSERR_RECLAIMCONFLICT;
2131 else if (new_stp->ls_flags & NFSLCK_CHECK)
2132 error = NFSERR_LOCKED;
2134 error = NFSERR_DENIED;
2135 if (filestruct_locked != 0 && ret == 0) {
2136 /* Roll back local locks. */
2138 if (vnode_unlocked == 0) {
2139 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2141 NFSVOPUNLOCK(vp, 0);
2143 nfsrv_locallock_rollback(vp, lfp, p);
2145 nfsrv_unlocklf(lfp);
2155 * We only get here if there was no lock that conflicted.
2157 if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2163 * We only get here when we are creating or modifying a lock.
2164 * There are two variants:
2165 * - exist_lock_owner where lock_owner exists
2166 * - open_to_lock_owner with new lock_owner
2168 first = new_lop->lo_first;
2169 end = new_lop->lo_end;
2170 lock_flags = new_lop->lo_flags;
2171 if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2172 nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2173 stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2174 if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2175 stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2176 stateidp->other[0] = lckstp->ls_stateid.other[0];
2177 stateidp->other[1] = lckstp->ls_stateid.other[1];
2178 stateidp->other[2] = lckstp->ls_stateid.other[2];
2181 * The new open_to_lock_owner case.
2182 * Link the new nfsstate into the lists.
2184 new_stp->ls_seq = new_stp->ls_opentolockseq;
2185 nfsrvd_refcache(new_stp->ls_op);
2186 stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2187 stateidp->other[0] = new_stp->ls_stateid.other[0] =
2188 clp->lc_clientid.lval[0];
2189 stateidp->other[1] = new_stp->ls_stateid.other[1] =
2190 clp->lc_clientid.lval[1];
2191 stateidp->other[2] = new_stp->ls_stateid.other[2] =
2192 nfsrv_nextstateindex(clp);
2193 new_stp->ls_clp = clp;
2194 LIST_INIT(&new_stp->ls_lock);
2195 new_stp->ls_openstp = stp;
2196 new_stp->ls_lfp = lfp;
2197 nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2199 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2201 LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2204 newnfsstats.srvlockowners++;
2205 nfsrv_openpluslock++;
2207 if (filestruct_locked != 0) {
2209 nfsrv_locallock_commit(lfp, lock_flags, first, end);
2211 nfsrv_unlocklf(lfp);
2217 NFSLOCKV4ROOTMUTEX();
2218 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2219 NFSUNLOCKV4ROOTMUTEX();
2221 if (vnode_unlocked != 0) {
2222 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2223 if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2224 error = NFSERR_SERVERFAULT;
2227 FREE((caddr_t)other_lop, M_NFSDLOCK);
2228 NFSEXITCODE2(error, nd);
2233 * Check for state errors for Open.
2234 * repstat is passed back out as an error if more critical errors
2238 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2239 struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2240 NFSPROC_T *p, int repstat)
2242 struct nfsstate *stp, *nstp;
2243 struct nfsclient *clp;
2244 struct nfsstate *ownerstp;
2245 struct nfslockfile *lfp, *new_lfp;
2246 int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2248 if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2251 * Check for restart conditions (client and server).
2253 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2254 &new_stp->ls_stateid, 0);
2259 * Check for state resource limit exceeded.
2260 * Technically this should be SMP protected, but the worst
2261 * case error is "out by one or two" on the count when it
2262 * returns NFSERR_RESOURCE and the limit is just a rather
2263 * arbitrary high water mark, so no harm is done.
2265 if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2266 error = NFSERR_RESOURCE;
2271 MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2272 M_NFSDLOCKFILE, M_WAITOK);
2274 getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2278 * Get the nfsclient structure.
2280 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2281 (nfsquad_t)((u_quad_t)0), 0, nd, p);
2284 * Look up the open owner. See if it needs confirmation and
2285 * check the seq#, as required.
2288 nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2290 if (!error && ownerstp) {
2291 error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2294 * If the OpenOwner hasn't been confirmed, assume the
2295 * old one was a replay and this one is ok.
2296 * See: RFC3530 Sec. 14.2.18.
2298 if (error == NFSERR_BADSEQID &&
2299 (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2307 error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2308 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2309 nfsrv_checkstable(clp))
2310 error = NFSERR_NOGRACE;
2313 * If none of the above errors occurred, let repstat be
2316 if (repstat && !error)
2321 NFSLOCKV4ROOTMUTEX();
2322 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2323 NFSUNLOCKV4ROOTMUTEX();
2325 free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2330 * If vp == NULL, the file doesn't exist yet, so return ok.
2331 * (This always happens on the first pass, so haslock must be 0.)
2335 FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2340 * Get the structure for the underlying file.
2345 error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2348 FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2352 NFSLOCKV4ROOTMUTEX();
2353 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2354 NFSUNLOCKV4ROOTMUTEX();
2360 * Search for a conflicting open/share.
2362 if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2364 * For Delegate_Cur, search for the matching Delegation,
2365 * which indicates no conflict.
2366 * An old delegation should have been recovered by the
2367 * client doing a Claim_DELEGATE_Prev, so I won't let
2368 * it match and return NFSERR_EXPIRED. Should I let it
2371 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2372 if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2373 (((nd->nd_flag & ND_NFSV41) != 0 &&
2374 stateidp->seqid == 0) ||
2375 stateidp->seqid == stp->ls_stateid.seqid) &&
2376 !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2380 if (stp == LIST_END(&lfp->lf_deleg) ||
2381 ((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2382 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2385 NFSLOCKV4ROOTMUTEX();
2386 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2387 NFSUNLOCKV4ROOTMUTEX();
2389 error = NFSERR_EXPIRED;
2395 * Check for access/deny bit conflicts. I check for the same
2396 * owner as well, in case the client didn't bother.
2398 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2399 if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2400 (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2401 ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2402 ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2403 ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2404 ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2407 * nfsrv_clientconflict() unlocks
2408 * state when it returns non-zero.
2413 error = NFSERR_PERM;
2414 else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2415 error = NFSERR_RECLAIMCONFLICT;
2417 error = NFSERR_SHAREDENIED;
2421 NFSLOCKV4ROOTMUTEX();
2422 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2423 NFSUNLOCKV4ROOTMUTEX();
2430 * Check for a conflicting delegation. If one is found, call
2431 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2432 * been set yet, it will get the lock. Otherwise, it will recall
2433 * the delegation. Then, we try try again...
2434 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2435 * isn't a conflict.)
2436 * I currently believe the conflict algorithm to be:
2437 * For Open with Read Access and Deny None
2438 * - there is a conflict iff a different client has a write delegation
2439 * For Open with other Write Access or any Deny except None
2440 * - there is a conflict if a different client has any delegation
2441 * - there is a conflict if the same client has a read delegation
2442 * (The current concensus is that this last case should be
2443 * considered a conflict since the client with a read delegation
2444 * could have done an Open with ReadAccess and WriteDeny
2445 * locally and then not have checked for the WriteDeny.)
2446 * Don't check for a Reclaim, since that will be dealt with
2447 * by nfsrv_openctrl().
2449 if (!(new_stp->ls_flags &
2450 (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2451 stp = LIST_FIRST(&lfp->lf_deleg);
2452 while (stp != LIST_END(&lfp->lf_deleg)) {
2453 nstp = LIST_NEXT(stp, ls_file);
2454 if ((readonly && stp->ls_clp != clp &&
2455 (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2456 (!readonly && (stp->ls_clp != clp ||
2457 (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2458 ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2461 * nfsrv_delegconflict() unlocks state
2462 * when it returns non-zero.
2475 NFSLOCKV4ROOTMUTEX();
2476 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2477 NFSUNLOCKV4ROOTMUTEX();
2481 NFSEXITCODE2(error, nd);
2486 * Open control function to create/update open state for an open.
2489 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2490 struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2491 nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2492 NFSPROC_T *p, u_quad_t filerev)
2494 struct nfsstate *new_stp = *new_stpp;
2495 struct nfsstate *stp, *nstp;
2496 struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2497 struct nfslockfile *lfp, *new_lfp;
2498 struct nfsclient *clp;
2499 int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2500 int readonly = 0, cbret = 1, getfhret = 0;
2501 int gotstate = 0, len = 0;
2502 u_char *clidp = NULL;
2504 if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2507 * Check for restart conditions (client and server).
2508 * (Paranoia, should have been detected by nfsrv_opencheck().)
2509 * If an error does show up, return NFSERR_EXPIRED, since the
2510 * the seqid# has already been incremented.
2512 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2513 &new_stp->ls_stateid, 0);
2515 printf("Nfsd: openctrl unexpected restart err=%d\n",
2517 error = NFSERR_EXPIRED;
2521 clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2523 MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2524 M_NFSDLOCKFILE, M_WAITOK);
2525 MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2526 M_NFSDSTATE, M_WAITOK);
2527 MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2528 M_NFSDSTATE, M_WAITOK);
2529 getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2533 * Get the client structure. Since the linked lists could be changed
2534 * by other nfsd processes if this process does a tsleep(), one of
2535 * two things must be done.
2536 * 1 - don't tsleep()
2538 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2539 * before using the lists, since this lock stops the other
2540 * nfsd. This should only be used for rare cases, since it
2541 * essentially single threads the nfsd.
2542 * At this time, it is only done for cases where the stable
2543 * storage file must be written prior to completion of state
2546 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2547 (nfsquad_t)((u_quad_t)0), 0, nd, p);
2548 if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2551 * This happens on the first open for a client
2552 * that supports callbacks.
2556 * Although nfsrv_docallback() will sleep, clp won't
2557 * go away, since they are only removed when the
2558 * nfsv4_lock() has blocked the nfsd threads. The
2559 * fields in clp can change, but having multiple
2560 * threads do this Null callback RPC should be
2563 cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2564 NULL, 0, NULL, NULL, NULL, p);
2566 clp->lc_flags &= ~LCL_NEEDSCBNULL;
2568 clp->lc_flags |= LCL_CALLBACKSON;
2572 * Look up the open owner. See if it needs confirmation and
2573 * check the seq#, as required.
2576 nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2580 printf("Nfsd: openctrl unexpected state err=%d\n",
2582 free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2583 free((caddr_t)new_open, M_NFSDSTATE);
2584 free((caddr_t)new_deleg, M_NFSDSTATE);
2586 NFSLOCKV4ROOTMUTEX();
2587 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2588 NFSUNLOCKV4ROOTMUTEX();
2590 error = NFSERR_EXPIRED;
2594 if (new_stp->ls_flags & NFSLCK_RECLAIM)
2595 nfsrv_markstable(clp);
2598 * Get the structure for the underlying file.
2603 error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2606 FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2609 printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2611 free((caddr_t)new_open, M_NFSDSTATE);
2612 free((caddr_t)new_deleg, M_NFSDSTATE);
2614 NFSLOCKV4ROOTMUTEX();
2615 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2616 NFSUNLOCKV4ROOTMUTEX();
2622 * Search for a conflicting open/share.
2624 if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2626 * For Delegate_Cur, search for the matching Delegation,
2627 * which indicates no conflict.
2628 * An old delegation should have been recovered by the
2629 * client doing a Claim_DELEGATE_Prev, so I won't let
2630 * it match and return NFSERR_EXPIRED. Should I let it
2633 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2634 if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2635 (((nd->nd_flag & ND_NFSV41) != 0 &&
2636 stateidp->seqid == 0) ||
2637 stateidp->seqid == stp->ls_stateid.seqid) &&
2638 !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2642 if (stp == LIST_END(&lfp->lf_deleg) ||
2643 ((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2644 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2646 printf("Nfsd openctrl unexpected expiry\n");
2647 free((caddr_t)new_open, M_NFSDSTATE);
2648 free((caddr_t)new_deleg, M_NFSDSTATE);
2650 NFSLOCKV4ROOTMUTEX();
2651 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2652 NFSUNLOCKV4ROOTMUTEX();
2654 error = NFSERR_EXPIRED;
2659 * Don't issue a Delegation, since one already exists and
2660 * delay delegation timeout, as required.
2663 nfsrv_delaydelegtimeout(stp);
2667 * Check for access/deny bit conflicts. I also check for the
2668 * same owner, since the client might not have bothered to check.
2669 * Also, note an open for the same file and owner, if found,
2670 * which is all we do here for Delegate_Cur, since conflict
2671 * checking is already done.
2673 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2674 if (ownerstp && stp->ls_openowner == ownerstp)
2676 if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2678 * If another client has the file open, the only
2679 * delegation that can be issued is a Read delegation
2680 * and only if it is a Read open with Deny none.
2682 if (clp != stp->ls_clp) {
2683 if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2689 if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2690 ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2691 ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2692 ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2693 ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2696 * nfsrv_clientconflict() unlocks state
2697 * when it returns non-zero.
2699 free((caddr_t)new_open, M_NFSDSTATE);
2700 free((caddr_t)new_deleg, M_NFSDSTATE);
2705 error = NFSERR_PERM;
2706 else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2707 error = NFSERR_RECLAIMCONFLICT;
2709 error = NFSERR_SHAREDENIED;
2713 NFSLOCKV4ROOTMUTEX();
2714 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2715 NFSUNLOCKV4ROOTMUTEX();
2717 free((caddr_t)new_open, M_NFSDSTATE);
2718 free((caddr_t)new_deleg, M_NFSDSTATE);
2719 printf("nfsd openctrl unexpected client cnfl\n");
2726 * Check for a conflicting delegation. If one is found, call
2727 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2728 * been set yet, it will get the lock. Otherwise, it will recall
2729 * the delegation. Then, we try try again...
2730 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2731 * isn't a conflict.)
2732 * I currently believe the conflict algorithm to be:
2733 * For Open with Read Access and Deny None
2734 * - there is a conflict iff a different client has a write delegation
2735 * For Open with other Write Access or any Deny except None
2736 * - there is a conflict if a different client has any delegation
2737 * - there is a conflict if the same client has a read delegation
2738 * (The current concensus is that this last case should be
2739 * considered a conflict since the client with a read delegation
2740 * could have done an Open with ReadAccess and WriteDeny
2741 * locally and then not have checked for the WriteDeny.)
2743 if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2744 stp = LIST_FIRST(&lfp->lf_deleg);
2745 while (stp != LIST_END(&lfp->lf_deleg)) {
2746 nstp = LIST_NEXT(stp, ls_file);
2747 if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2751 if ((readonly && stp->ls_clp != clp &&
2752 (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2753 (!readonly && (stp->ls_clp != clp ||
2754 (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2755 if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2758 ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2761 * nfsrv_delegconflict() unlocks state
2762 * when it returns non-zero.
2764 printf("Nfsd openctrl unexpected deleg cnfl\n");
2765 free((caddr_t)new_open, M_NFSDSTATE);
2766 free((caddr_t)new_deleg, M_NFSDSTATE);
2781 * We only get here if there was no open that conflicted.
2782 * If an open for the owner exists, or in the access/deny bits.
2783 * Otherwise it is a new open. If the open_owner hasn't been
2784 * confirmed, replace the open with the new one needing confirmation,
2785 * otherwise add the open.
2787 if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2789 * Handle NFSLCK_DELEGPREV by searching the old delegations for
2790 * a match. If found, just move the old delegation to the current
2791 * delegation list and issue open. If not found, return
2794 LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2795 if (stp->ls_lfp == lfp) {
2797 if (stp->ls_clp != clp)
2798 panic("olddeleg clp");
2799 LIST_REMOVE(stp, ls_list);
2800 LIST_REMOVE(stp, ls_hash);
2801 stp->ls_flags &= ~NFSLCK_OLDDELEG;
2802 stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2803 stp->ls_stateid.other[0] = delegstateidp->other[0] =
2804 clp->lc_clientid.lval[0];
2805 stp->ls_stateid.other[1] = delegstateidp->other[1] =
2806 clp->lc_clientid.lval[1];
2807 stp->ls_stateid.other[2] = delegstateidp->other[2] =
2808 nfsrv_nextstateindex(clp);
2809 stp->ls_compref = nd->nd_compref;
2810 LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2811 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2812 stp->ls_stateid), stp, ls_hash);
2813 if (stp->ls_flags & NFSLCK_DELEGWRITE)
2814 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2816 *rflagsp |= NFSV4OPEN_READDELEGATE;
2817 clp->lc_delegtime = NFSD_MONOSEC +
2818 nfsrv_lease + NFSRV_LEASEDELTA;
2821 * Now, do the associated open.
2823 new_open->ls_stateid.seqid = 1;
2824 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2825 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2826 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2827 new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2829 if (stp->ls_flags & NFSLCK_DELEGWRITE)
2830 new_open->ls_flags |= (NFSLCK_READACCESS |
2831 NFSLCK_WRITEACCESS);
2833 new_open->ls_flags |= NFSLCK_READACCESS;
2834 new_open->ls_uid = new_stp->ls_uid;
2835 new_open->ls_lfp = lfp;
2836 new_open->ls_clp = clp;
2837 LIST_INIT(&new_open->ls_open);
2838 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2839 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2842 * and handle the open owner
2845 new_open->ls_openowner = ownerstp;
2846 LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2848 new_open->ls_openowner = new_stp;
2849 new_stp->ls_flags = 0;
2850 nfsrvd_refcache(new_stp->ls_op);
2851 new_stp->ls_noopens = 0;
2852 LIST_INIT(&new_stp->ls_open);
2853 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2854 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2856 newnfsstats.srvopenowners++;
2857 nfsrv_openpluslock++;
2861 newnfsstats.srvopens++;
2862 nfsrv_openpluslock++;
2866 if (stp == LIST_END(&clp->lc_olddeleg))
2867 error = NFSERR_EXPIRED;
2868 } else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2870 * Scan to see that no delegation for this client and file
2871 * doesn't already exist.
2872 * There also shouldn't yet be an Open for this file and
2875 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2876 if (stp->ls_clp == clp)
2879 if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2881 * This is the Claim_Previous case with a delegation
2882 * type != Delegate_None.
2885 * First, add the delegation. (Although we must issue the
2886 * delegation, we can also ask for an immediate return.)
2888 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2889 new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2890 clp->lc_clientid.lval[0];
2891 new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2892 clp->lc_clientid.lval[1];
2893 new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2894 nfsrv_nextstateindex(clp);
2895 if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2896 new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2897 NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2898 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2900 new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2902 *rflagsp |= NFSV4OPEN_READDELEGATE;
2904 new_deleg->ls_uid = new_stp->ls_uid;
2905 new_deleg->ls_lfp = lfp;
2906 new_deleg->ls_clp = clp;
2907 new_deleg->ls_filerev = filerev;
2908 new_deleg->ls_compref = nd->nd_compref;
2909 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2910 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2911 new_deleg->ls_stateid), new_deleg, ls_hash);
2912 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2914 if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2915 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2917 NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2918 !NFSVNO_DELEGOK(vp))
2919 *rflagsp |= NFSV4OPEN_RECALL;
2920 newnfsstats.srvdelegates++;
2921 nfsrv_openpluslock++;
2922 nfsrv_delegatecnt++;
2925 * Now, do the associated open.
2927 new_open->ls_stateid.seqid = 1;
2928 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2929 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2930 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2931 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2933 if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2934 new_open->ls_flags |= (NFSLCK_READACCESS |
2935 NFSLCK_WRITEACCESS);
2937 new_open->ls_flags |= NFSLCK_READACCESS;
2938 new_open->ls_uid = new_stp->ls_uid;
2939 new_open->ls_lfp = lfp;
2940 new_open->ls_clp = clp;
2941 LIST_INIT(&new_open->ls_open);
2942 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2943 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2946 * and handle the open owner
2949 new_open->ls_openowner = ownerstp;
2950 LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2952 new_open->ls_openowner = new_stp;
2953 new_stp->ls_flags = 0;
2954 nfsrvd_refcache(new_stp->ls_op);
2955 new_stp->ls_noopens = 0;
2956 LIST_INIT(&new_stp->ls_open);
2957 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2958 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2960 newnfsstats.srvopenowners++;
2961 nfsrv_openpluslock++;
2965 newnfsstats.srvopens++;
2966 nfsrv_openpluslock++;
2968 error = NFSERR_RECLAIMCONFLICT;
2970 } else if (ownerstp) {
2971 if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2972 /* Replace the open */
2973 if (ownerstp->ls_op)
2974 nfsrvd_derefcache(ownerstp->ls_op);
2975 ownerstp->ls_op = new_stp->ls_op;
2976 nfsrvd_refcache(ownerstp->ls_op);
2977 ownerstp->ls_seq = new_stp->ls_seq;
2978 *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2979 stp = LIST_FIRST(&ownerstp->ls_open);
2980 stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2982 stp->ls_stateid.seqid = 1;
2983 stp->ls_uid = new_stp->ls_uid;
2984 if (lfp != stp->ls_lfp) {
2985 LIST_REMOVE(stp, ls_file);
2986 LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2990 } else if (openstp) {
2991 openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2992 openstp->ls_stateid.seqid++;
2993 if ((nd->nd_flag & ND_NFSV41) != 0 &&
2994 openstp->ls_stateid.seqid == 0)
2995 openstp->ls_stateid.seqid = 1;
2998 * This is where we can choose to issue a delegation.
3000 if (delegate == 0 || writedeleg == 0 ||
3001 NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3002 nfsrv_writedelegifpos == 0) ||
3003 !NFSVNO_DELEGOK(vp) ||
3004 (new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3005 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3007 *rflagsp |= NFSV4OPEN_WDCONTENTION;
3008 else if (nfsrv_issuedelegs == 0 ||
3009 NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3010 *rflagsp |= NFSV4OPEN_WDRESOURCE;
3011 else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3012 *rflagsp |= NFSV4OPEN_WDNOTWANTED;
3014 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3015 new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3016 = clp->lc_clientid.lval[0];
3017 new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3018 = clp->lc_clientid.lval[1];
3019 new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3020 = nfsrv_nextstateindex(clp);
3021 new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3022 NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3023 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3024 new_deleg->ls_uid = new_stp->ls_uid;
3025 new_deleg->ls_lfp = lfp;
3026 new_deleg->ls_clp = clp;
3027 new_deleg->ls_filerev = filerev;
3028 new_deleg->ls_compref = nd->nd_compref;
3029 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3030 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3031 new_deleg->ls_stateid), new_deleg, ls_hash);
3032 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3034 newnfsstats.srvdelegates++;
3035 nfsrv_openpluslock++;
3036 nfsrv_delegatecnt++;
3039 new_open->ls_stateid.seqid = 1;
3040 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3041 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3042 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3043 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3045 new_open->ls_uid = new_stp->ls_uid;
3046 new_open->ls_openowner = ownerstp;
3047 new_open->ls_lfp = lfp;
3048 new_open->ls_clp = clp;
3049 LIST_INIT(&new_open->ls_open);
3050 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3051 LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3052 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3056 newnfsstats.srvopens++;
3057 nfsrv_openpluslock++;
3060 * This is where we can choose to issue a delegation.
3062 if (delegate == 0 || (writedeleg == 0 && readonly == 0) ||
3063 !NFSVNO_DELEGOK(vp) ||
3064 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3066 *rflagsp |= NFSV4OPEN_WDCONTENTION;
3067 else if (nfsrv_issuedelegs == 0 ||
3068 NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3069 *rflagsp |= NFSV4OPEN_WDRESOURCE;
3070 else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3071 *rflagsp |= NFSV4OPEN_WDNOTWANTED;
3073 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3074 new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3075 = clp->lc_clientid.lval[0];
3076 new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3077 = clp->lc_clientid.lval[1];
3078 new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3079 = nfsrv_nextstateindex(clp);
3080 if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3081 (nfsrv_writedelegifpos || !readonly) &&
3082 (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3083 new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3084 NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3085 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3087 new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3089 *rflagsp |= NFSV4OPEN_READDELEGATE;
3091 new_deleg->ls_uid = new_stp->ls_uid;
3092 new_deleg->ls_lfp = lfp;
3093 new_deleg->ls_clp = clp;
3094 new_deleg->ls_filerev = filerev;
3095 new_deleg->ls_compref = nd->nd_compref;
3096 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3097 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3098 new_deleg->ls_stateid), new_deleg, ls_hash);
3099 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3101 newnfsstats.srvdelegates++;
3102 nfsrv_openpluslock++;
3103 nfsrv_delegatecnt++;
3108 * New owner case. Start the open_owner sequence with a
3109 * Needs confirmation (unless a reclaim) and hang the
3112 new_open->ls_stateid.seqid = 1;
3113 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3114 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3115 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3116 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3118 new_open->ls_uid = new_stp->ls_uid;
3119 LIST_INIT(&new_open->ls_open);
3120 new_open->ls_openowner = new_stp;
3121 new_open->ls_lfp = lfp;
3122 new_open->ls_clp = clp;
3123 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3124 if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3125 new_stp->ls_flags = 0;
3126 } else if ((nd->nd_flag & ND_NFSV41) != 0) {
3127 /* NFSv4.1 never needs confirmation. */
3128 new_stp->ls_flags = 0;
3131 * This is where we can choose to issue a delegation.
3133 if (delegate && nfsrv_issuedelegs &&
3134 (writedeleg || readonly) &&
3135 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3137 !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3138 NFSVNO_DELEGOK(vp) &&
3139 ((nd->nd_flag & ND_NFSV41) == 0 ||
3140 (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3141 new_deleg->ls_stateid.seqid =
3142 delegstateidp->seqid = 1;
3143 new_deleg->ls_stateid.other[0] =
3144 delegstateidp->other[0]
3145 = clp->lc_clientid.lval[0];
3146 new_deleg->ls_stateid.other[1] =
3147 delegstateidp->other[1]
3148 = clp->lc_clientid.lval[1];
3149 new_deleg->ls_stateid.other[2] =
3150 delegstateidp->other[2]
3151 = nfsrv_nextstateindex(clp);
3152 if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3153 (nfsrv_writedelegifpos || !readonly) &&
3154 ((nd->nd_flag & ND_NFSV41) == 0 ||
3155 (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3157 new_deleg->ls_flags =
3158 (NFSLCK_DELEGWRITE |
3160 NFSLCK_WRITEACCESS);
3161 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3163 new_deleg->ls_flags =
3166 *rflagsp |= NFSV4OPEN_READDELEGATE;
3168 new_deleg->ls_uid = new_stp->ls_uid;
3169 new_deleg->ls_lfp = lfp;
3170 new_deleg->ls_clp = clp;
3171 new_deleg->ls_filerev = filerev;
3172 new_deleg->ls_compref = nd->nd_compref;
3173 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3175 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3176 new_deleg->ls_stateid), new_deleg, ls_hash);
3177 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3180 newnfsstats.srvdelegates++;
3181 nfsrv_openpluslock++;
3182 nfsrv_delegatecnt++;
3185 * Since NFSv4.1 never does an OpenConfirm, the first
3186 * open state will be acquired here.
3188 if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3189 clp->lc_flags |= LCL_STAMPEDSTABLE;
3190 len = clp->lc_idlen;
3191 NFSBCOPY(clp->lc_id, clidp, len);
3195 *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3196 new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3198 nfsrvd_refcache(new_stp->ls_op);
3199 new_stp->ls_noopens = 0;
3200 LIST_INIT(&new_stp->ls_open);
3201 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3202 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3203 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3208 newnfsstats.srvopens++;
3209 nfsrv_openpluslock++;
3210 newnfsstats.srvopenowners++;
3211 nfsrv_openpluslock++;
3214 stateidp->seqid = openstp->ls_stateid.seqid;
3215 stateidp->other[0] = openstp->ls_stateid.other[0];
3216 stateidp->other[1] = openstp->ls_stateid.other[1];
3217 stateidp->other[2] = openstp->ls_stateid.other[2];
3221 NFSLOCKV4ROOTMUTEX();
3222 nfsv4_unlock(&nfsv4rootfs_lock, 1);
3223 NFSUNLOCKV4ROOTMUTEX();
3226 FREE((caddr_t)new_open, M_NFSDSTATE);
3228 FREE((caddr_t)new_deleg, M_NFSDSTATE);
3231 * If the NFSv4.1 client just acquired its first open, write a timestamp
3232 * to the stable storage file.
3234 if (gotstate != 0) {
3235 nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3236 nfsrv_backupstable();
3240 free(clidp, M_TEMP);
3241 NFSEXITCODE2(error, nd);
3246 * Open update. Does the confirm, downgrade and close.
3249 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3250 nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3252 struct nfsstate *stp, *ownerstp;
3253 struct nfsclient *clp;
3254 struct nfslockfile *lfp;
3256 int error = 0, gotstate = 0, len = 0;
3257 u_char *clidp = NULL;
3260 * Check for restart conditions (client and server).
3262 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3263 &new_stp->ls_stateid, 0);
3267 clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3270 * Get the open structure via clientid and stateid.
3272 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3273 (nfsquad_t)((u_quad_t)0), 0, nd, p);
3275 error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3276 new_stp->ls_flags, &stp);
3279 * Sanity check the open.
3281 if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3282 (!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3283 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3284 ((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3285 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3286 error = NFSERR_BADSTATEID;
3289 error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3290 stp->ls_openowner, new_stp->ls_op);
3291 if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3292 (((nd->nd_flag & ND_NFSV41) == 0 &&
3293 !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3294 ((nd->nd_flag & ND_NFSV41) != 0 &&
3295 new_stp->ls_stateid.seqid != 0)))
3296 error = NFSERR_OLDSTATEID;
3297 if (!error && vnode_vtype(vp) != VREG) {
3298 if (vnode_vtype(vp) == VDIR)
3299 error = NFSERR_ISDIR;
3301 error = NFSERR_INVAL;
3306 * If a client tries to confirm an Open with a bad
3307 * seqid# and there are no byte range locks or other Opens
3308 * on the openowner, just throw it away, so the next use of the
3309 * openowner will start a fresh seq#.
3311 if (error == NFSERR_BADSEQID &&
3312 (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3313 nfsrv_nootherstate(stp))
3314 nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3320 * Set the return stateid.
3322 stateidp->seqid = stp->ls_stateid.seqid + 1;
3323 if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3324 stateidp->seqid = 1;
3325 stateidp->other[0] = stp->ls_stateid.other[0];
3326 stateidp->other[1] = stp->ls_stateid.other[1];
3327 stateidp->other[2] = stp->ls_stateid.other[2];
3329 * Now, handle the three cases.
3331 if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3333 * If the open doesn't need confirmation, it seems to me that
3334 * there is a client error, but I'll just log it and keep going?
3336 if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3337 printf("Nfsv4d: stray open confirm\n");
3338 stp->ls_openowner->ls_flags = 0;
3339 stp->ls_stateid.seqid++;
3340 if ((nd->nd_flag & ND_NFSV41) != 0 &&
3341 stp->ls_stateid.seqid == 0)
3342 stp->ls_stateid.seqid = 1;
3343 if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3344 clp->lc_flags |= LCL_STAMPEDSTABLE;
3345 len = clp->lc_idlen;
3346 NFSBCOPY(clp->lc_id, clidp, len);
3350 } else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3351 ownerstp = stp->ls_openowner;
3353 if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3354 /* Get the lf lock */
3357 ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3358 NFSVOPUNLOCK(vp, 0);
3359 if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3361 nfsrv_unlocklf(lfp);
3364 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3366 (void) nfsrv_freeopen(stp, NULL, 0, p);
3371 * Update the share bits, making sure that the new set are a
3372 * subset of the old ones.
3374 bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3375 if (~(stp->ls_flags) & bits) {
3377 error = NFSERR_INVAL;
3380 stp->ls_flags = (bits | NFSLCK_OPEN);
3381 stp->ls_stateid.seqid++;
3382 if ((nd->nd_flag & ND_NFSV41) != 0 &&
3383 stp->ls_stateid.seqid == 0)
3384 stp->ls_stateid.seqid = 1;
3389 * If the client just confirmed its first open, write a timestamp
3390 * to the stable storage file.
3392 if (gotstate != 0) {
3393 nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3394 nfsrv_backupstable();
3398 free(clidp, M_TEMP);
3399 NFSEXITCODE2(error, nd);
3404 * Delegation update. Does the purge and return.
3407 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3408 nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3411 struct nfsstate *stp;
3412 struct nfsclient *clp;
3417 * Do a sanity check against the file handle for DelegReturn.
3420 error = nfsvno_getfh(vp, &fh, p);
3425 * Check for restart conditions (client and server).
3427 if (op == NFSV4OP_DELEGRETURN)
3428 error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3431 error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3436 * Get the open structure via clientid and stateid.
3439 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3440 (nfsquad_t)((u_quad_t)0), 0, nd, p);
3442 if (error == NFSERR_CBPATHDOWN)
3444 if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3445 error = NFSERR_STALESTATEID;
3447 if (!error && op == NFSV4OP_DELEGRETURN) {
3448 error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3449 if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3450 ((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3451 error = NFSERR_OLDSTATEID;
3454 * NFSERR_EXPIRED means that the state has gone away,
3455 * so Delegations have been purged. Just return ok.
3457 if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3467 if (op == NFSV4OP_DELEGRETURN) {
3468 if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3469 sizeof (fhandle_t))) {
3471 error = NFSERR_BADSTATEID;
3474 nfsrv_freedeleg(stp);
3476 nfsrv_freedeleglist(&clp->lc_olddeleg);
3487 * Release lock owner.
3490 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3493 struct nfsstate *stp, *nstp, *openstp, *ownstp;
3494 struct nfsclient *clp;
3498 * Check for restart conditions (client and server).
3500 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3501 &new_stp->ls_stateid, 0);
3507 * Get the lock owner by name.
3509 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3510 (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3515 LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3516 LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3517 stp = LIST_FIRST(&openstp->ls_open);
3518 while (stp != LIST_END(&openstp->ls_open)) {
3519 nstp = LIST_NEXT(stp, ls_list);
3521 * If the owner matches, check for locks and
3522 * then free or return an error.
3524 if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3525 !NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3527 if (LIST_EMPTY(&stp->ls_lock)) {
3528 nfsrv_freelockowner(stp, NULL, 0, p);
3531 error = NFSERR_LOCKSHELD;
3547 * Get the file handle for a lock structure.
3550 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3551 fhandle_t *nfhp, NFSPROC_T *p)
3553 fhandle_t *fhp = NULL;
3557 * For lock, use the new nfslock structure, otherwise just
3558 * a fhandle_t on the stack.
3560 if (flags & NFSLCK_OPEN) {
3561 KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3562 fhp = &new_lfp->lf_fh;
3566 panic("nfsrv_getlockfh");
3568 error = nfsvno_getfh(vp, fhp, p);
3574 * Get an nfs lock structure. Allocate one, as required, and return a
3576 * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3579 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3580 struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3582 struct nfslockfile *lfp;
3583 fhandle_t *fhp = NULL, *tfhp;
3584 struct nfslockhashhead *hp;
3585 struct nfslockfile *new_lfp = NULL;
3588 * For lock, use the new nfslock structure, otherwise just
3589 * a fhandle_t on the stack.
3591 if (flags & NFSLCK_OPEN) {
3592 new_lfp = *new_lfpp;
3593 fhp = &new_lfp->lf_fh;
3597 panic("nfsrv_getlockfile");
3600 hp = NFSLOCKHASH(fhp);
3601 LIST_FOREACH(lfp, hp, lf_hash) {
3603 if (NFSVNO_CMPFH(fhp, tfhp)) {
3610 if (!(flags & NFSLCK_OPEN))
3614 * No match, so chain the new one into the list.
3616 LIST_INIT(&new_lfp->lf_open);
3617 LIST_INIT(&new_lfp->lf_lock);
3618 LIST_INIT(&new_lfp->lf_deleg);
3619 LIST_INIT(&new_lfp->lf_locallock);
3620 LIST_INIT(&new_lfp->lf_rollback);
3621 new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3622 new_lfp->lf_locallock_lck.nfslock_lock = 0;
3623 new_lfp->lf_usecount = 0;
3624 LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3631 * This function adds a nfslock lock structure to the list for the associated
3632 * nfsstate and nfslockfile structures. It will be inserted after the
3633 * entry pointed at by insert_lop.
3636 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3637 struct nfsstate *stp, struct nfslockfile *lfp)
3639 struct nfslock *lop, *nlop;
3641 new_lop->lo_stp = stp;
3642 new_lop->lo_lfp = lfp;
3645 /* Insert in increasing lo_first order */
3646 lop = LIST_FIRST(&lfp->lf_lock);
3647 if (lop == LIST_END(&lfp->lf_lock) ||
3648 new_lop->lo_first <= lop->lo_first) {
3649 LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3651 nlop = LIST_NEXT(lop, lo_lckfile);
3652 while (nlop != LIST_END(&lfp->lf_lock) &&
3653 nlop->lo_first < new_lop->lo_first) {
3655 nlop = LIST_NEXT(lop, lo_lckfile);
3657 LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3660 new_lop->lo_lckfile.le_prev = NULL; /* list not used */
3664 * Insert after insert_lop, which is overloaded as stp or lfp for
3667 if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3668 LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3669 else if ((struct nfsstate *)insert_lop == stp)
3670 LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3672 LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3674 newnfsstats.srvlocks++;
3675 nfsrv_openpluslock++;
3680 * This function updates the locking for a lock owner and given file. It
3681 * maintains a list of lock ranges ordered on increasing file offset that
3682 * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3683 * It always adds new_lop to the list and sometimes uses the one pointed
3687 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3688 struct nfslock **other_lopp, struct nfslockfile *lfp)
3690 struct nfslock *new_lop = *new_lopp;
3691 struct nfslock *lop, *tlop, *ilop;
3692 struct nfslock *other_lop = *other_lopp;
3693 int unlock = 0, myfile = 0;
3697 * Work down the list until the lock is merged.
3699 if (new_lop->lo_flags & NFSLCK_UNLOCK)
3702 ilop = (struct nfslock *)stp;
3703 lop = LIST_FIRST(&stp->ls_lock);
3705 ilop = (struct nfslock *)lfp;
3706 lop = LIST_FIRST(&lfp->lf_locallock);
3708 while (lop != NULL) {
3710 * Only check locks for this file that aren't before the start of
3713 if (lop->lo_lfp == lfp) {
3715 if (lop->lo_end >= new_lop->lo_first) {
3716 if (new_lop->lo_end < lop->lo_first) {
3718 * If the new lock ends before the start of the
3719 * current lock's range, no merge, just insert
3724 if (new_lop->lo_flags == lop->lo_flags ||
3725 (new_lop->lo_first <= lop->lo_first &&
3726 new_lop->lo_end >= lop->lo_end)) {
3728 * This lock can be absorbed by the new lock/unlock.
3729 * This happens when it covers the entire range
3730 * of the old lock or is contiguous
3731 * with the old lock and is of the same type or an
3734 if (lop->lo_first < new_lop->lo_first)
3735 new_lop->lo_first = lop->lo_first;
3736 if (lop->lo_end > new_lop->lo_end)
3737 new_lop->lo_end = lop->lo_end;
3739 lop = LIST_NEXT(lop, lo_lckowner);
3740 nfsrv_freenfslock(tlop);
3745 * All these cases are for contiguous locks that are not the
3746 * same type, so they can't be merged.
3748 if (new_lop->lo_first <= lop->lo_first) {
3750 * This case is where the new lock overlaps with the
3751 * first part of the old lock. Move the start of the
3752 * old lock to just past the end of the new lock. The
3753 * new lock will be inserted in front of the old, since
3754 * ilop hasn't been updated. (We are done now.)
3756 lop->lo_first = new_lop->lo_end;
3759 if (new_lop->lo_end >= lop->lo_end) {
3761 * This case is where the new lock overlaps with the
3762 * end of the old lock's range. Move the old lock's
3763 * end to just before the new lock's first and insert
3764 * the new lock after the old lock.
3765 * Might not be done yet, since the new lock could
3766 * overlap further locks with higher ranges.
3768 lop->lo_end = new_lop->lo_first;
3770 lop = LIST_NEXT(lop, lo_lckowner);
3774 * The final case is where the new lock's range is in the
3775 * middle of the current lock's and splits the current lock
3776 * up. Use *other_lopp to handle the second part of the
3777 * split old lock range. (We are done now.)
3778 * For unlock, we use new_lop as other_lop and tmp, since
3779 * other_lop and new_lop are the same for this case.
3780 * We noted the unlock case above, so we don't need
3781 * new_lop->lo_flags any longer.
3783 tmp = new_lop->lo_first;
3784 if (other_lop == NULL) {
3786 panic("nfsd srv update unlock");
3787 other_lop = new_lop;
3790 other_lop->lo_first = new_lop->lo_end;
3791 other_lop->lo_end = lop->lo_end;
3792 other_lop->lo_flags = lop->lo_flags;
3793 other_lop->lo_stp = stp;
3794 other_lop->lo_lfp = lfp;
3796 nfsrv_insertlock(other_lop, lop, stp, lfp);
3803 lop = LIST_NEXT(lop, lo_lckowner);
3804 if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3809 * Insert the new lock in the list at the appropriate place.
3812 nfsrv_insertlock(new_lop, ilop, stp, lfp);
3818 * This function handles sequencing of locks, etc.
3819 * It returns an error that indicates what the caller should do.
3822 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3823 struct nfsstate *stp, struct nfsrvcache *op)
3827 if ((nd->nd_flag & ND_NFSV41) != 0)
3828 /* NFSv4.1 ignores the open_seqid and lock_seqid. */
3830 if (op != nd->nd_rp)
3831 panic("nfsrvstate checkseqid");
3832 if (!(op->rc_flag & RC_INPROG))
3833 panic("nfsrvstate not inprog");
3834 if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3835 printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3836 panic("nfsrvstate op refcnt");
3838 if ((stp->ls_seq + 1) == seqid) {
3840 nfsrvd_derefcache(stp->ls_op);
3842 nfsrvd_refcache(op);
3843 stp->ls_seq = seqid;
3845 } else if (stp->ls_seq == seqid && stp->ls_op &&
3846 op->rc_xid == stp->ls_op->rc_xid &&
3847 op->rc_refcnt == 0 &&
3848 op->rc_reqlen == stp->ls_op->rc_reqlen &&
3849 op->rc_cksum == stp->ls_op->rc_cksum) {
3850 if (stp->ls_op->rc_flag & RC_INPROG) {
3851 error = NFSERR_DONTREPLY;
3854 nd->nd_rp = stp->ls_op;
3855 nd->nd_rp->rc_flag |= RC_INPROG;
3856 nfsrvd_delcache(op);
3857 error = NFSERR_REPLYFROMCACHE;
3860 error = NFSERR_BADSEQID;
3863 NFSEXITCODE2(error, nd);
3868 * Get the client ip address for callbacks. If the strings can't be parsed,
3869 * just set lc_program to 0 to indicate no callbacks are possible.
3870 * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3871 * the address to the client's transport address. This won't be used
3872 * for callbacks, but can be printed out by newnfsstats for info.)
3873 * Return error if the xdr can't be parsed, 0 otherwise.
3876 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3881 struct sockaddr_in *rad, *sad;
3882 u_char protocol[5], addr[24];
3883 int error = 0, cantparse = 0;
3893 rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3894 rad->sin_family = AF_INET;
3895 rad->sin_len = sizeof (struct sockaddr_in);
3896 rad->sin_addr.s_addr = 0;
3898 clp->lc_req.nr_client = NULL;
3899 clp->lc_req.nr_lock = 0;
3900 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3901 i = fxdr_unsigned(int, *tl);
3902 if (i >= 3 && i <= 4) {
3903 error = nfsrv_mtostr(nd, protocol, i);
3906 if (!strcmp(protocol, "tcp")) {
3907 clp->lc_flags |= LCL_TCPCALLBACK;
3908 clp->lc_req.nr_sotype = SOCK_STREAM;
3909 clp->lc_req.nr_soproto = IPPROTO_TCP;
3910 } else if (!strcmp(protocol, "udp")) {
3911 clp->lc_req.nr_sotype = SOCK_DGRAM;
3912 clp->lc_req.nr_soproto = IPPROTO_UDP;
3919 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3924 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3925 i = fxdr_unsigned(int, *tl);
3927 error = NFSERR_BADXDR;
3929 } else if (i == 0) {
3931 } else if (!cantparse && i <= 23 && i >= 11) {
3932 error = nfsrv_mtostr(nd, addr, i);
3937 * Parse out the address fields. We expect 6 decimal numbers
3938 * separated by '.'s.
3942 while (*cp && i < 6) {
3944 while (*cp2 && *cp2 != '.')
3952 j = nfsrv_getipnumber(cp);
3957 port.cval[5 - i] = j;
3966 if (ip.ival != 0x0) {
3967 rad->sin_addr.s_addr = htonl(ip.ival);
3968 rad->sin_port = htons(port.sval);
3976 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3982 sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3983 rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3984 rad->sin_port = 0x0;
3985 clp->lc_program = 0;
3988 NFSEXITCODE2(error, nd);
3993 * Turn a string of up to three decimal digits into a number. Return -1 upon
3997 nfsrv_getipnumber(u_char *cp)
4002 if (j > 2 || *cp < '0' || *cp > '9')
4015 * This function checks for restart conditions.
4018 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4019 nfsv4stateid_t *stateidp, int specialid)
4024 * First check for a server restart. Open, LockT, ReleaseLockOwner
4025 * and DelegPurge have a clientid, the rest a stateid.
4028 (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4029 if (clientid.lval[0] != nfsrvboottime) {
4030 ret = NFSERR_STALECLIENTID;
4033 } else if (stateidp->other[0] != nfsrvboottime &&
4035 ret = NFSERR_STALESTATEID;
4040 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4041 * not use a lock/open owner seqid#, so the check can be done now.
4042 * (The others will be checked, as required, later.)
4044 if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4048 ret = nfsrv_checkgrace(NULL, NULL, flags);
4060 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4065 if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4066 if (flags & NFSLCK_RECLAIM) {
4067 error = NFSERR_NOGRACE;
4071 if (!(flags & NFSLCK_RECLAIM)) {
4072 error = NFSERR_GRACE;
4075 if (nd != NULL && clp != NULL &&
4076 (nd->nd_flag & ND_NFSV41) != 0 &&
4077 (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4078 error = NFSERR_NOGRACE;
4083 * If grace is almost over and we are still getting Reclaims,
4084 * extend grace a bit.
4086 if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4087 nfsrv_stablefirst.nsf_eograce)
4088 nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4098 * Do a server callback.
4101 nfsrv_docallback(struct nfsclient *clp, int procnum,
4102 nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
4103 struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
4107 struct nfsrv_descript nfsd, *nd = &nfsd;
4111 struct nfsdsession *sep = NULL;
4113 cred = newnfs_getcred();
4114 NFSLOCKSTATE(); /* mostly for lc_cbref++ */
4115 if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4122 * Fill the callback program# and version into the request
4123 * structure for newnfs_connect() to use.
4125 clp->lc_req.nr_prog = clp->lc_program;
4127 if ((clp->lc_flags & LCL_NFSV41) != 0)
4128 clp->lc_req.nr_vers = NFSV41_CBVERS;
4131 clp->lc_req.nr_vers = NFSV4_CBVERS;
4134 * First, fill in some of the fields of nd and cr.
4136 nd->nd_flag = ND_NFSV4;
4137 if (clp->lc_flags & LCL_GSS)
4138 nd->nd_flag |= ND_KERBV;
4139 if ((clp->lc_flags & LCL_NFSV41) != 0)
4140 nd->nd_flag |= ND_NFSV41;
4142 cred->cr_uid = clp->lc_uid;
4143 cred->cr_gid = clp->lc_gid;
4144 callback = clp->lc_callback;
4146 cred->cr_ngroups = 1;
4149 * Get the first mbuf for the request.
4151 MGET(m, M_WAITOK, MT_DATA);
4153 nd->nd_mreq = nd->nd_mb = m;
4154 nd->nd_bpos = NFSMTOD(m, caddr_t);
4157 * and build the callback request.
4159 if (procnum == NFSV4OP_CBGETATTR) {
4160 nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4161 error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4162 "CB Getattr", &sep);
4164 mbuf_freem(nd->nd_mreq);
4167 (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4168 (void)nfsrv_putattrbit(nd, attrbitp);
4169 } else if (procnum == NFSV4OP_CBRECALL) {
4170 nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4171 error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4174 mbuf_freem(nd->nd_mreq);
4177 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4178 *tl++ = txdr_unsigned(stateidp->seqid);
4179 NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4181 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4186 (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4187 } else if (procnum == NFSV4PROC_CBNULL) {
4188 nd->nd_procnum = NFSV4PROC_CBNULL;
4189 if ((clp->lc_flags & LCL_NFSV41) != 0) {
4190 error = nfsv4_getcbsession(clp, &sep);
4192 mbuf_freem(nd->nd_mreq);
4197 error = NFSERR_SERVERFAULT;
4198 mbuf_freem(nd->nd_mreq);
4203 * Call newnfs_connect(), as required, and then newnfs_request().
4205 (void) newnfs_sndlock(&clp->lc_req.nr_lock);
4206 if (clp->lc_req.nr_client == NULL) {
4207 if ((clp->lc_flags & LCL_NFSV41) != 0)
4208 error = ECONNREFUSED;
4209 else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4210 error = newnfs_connect(NULL, &clp->lc_req, cred,
4213 error = newnfs_connect(NULL, &clp->lc_req, cred,
4216 newnfs_sndunlock(&clp->lc_req.nr_lock);
4218 if ((nd->nd_flag & ND_NFSV41) != 0) {
4219 KASSERT(sep != NULL, ("sep NULL"));
4220 if (sep->sess_cbsess.nfsess_xprt != NULL)
4221 error = newnfs_request(nd, NULL, clp,
4222 &clp->lc_req, NULL, NULL, cred,
4223 clp->lc_program, clp->lc_req.nr_vers, NULL,
4224 1, NULL, &sep->sess_cbsess);
4227 * This should probably never occur, but if a
4228 * client somehow does an RPC without a
4229 * SequenceID Op that causes a callback just
4230 * after the nfsd threads have been terminated
4231 * and restared we could conceivably get here
4232 * without a backchannel xprt.
4234 printf("nfsrv_docallback: no xprt\n");
4235 error = ECONNREFUSED;
4237 nfsrv_freesession(sep, NULL);
4239 error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4240 NULL, NULL, cred, clp->lc_program,
4241 clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4247 * If error is set here, the Callback path isn't working
4248 * properly, so twiddle the appropriate LCL_ flags.
4249 * (nd_repstat != 0 indicates the Callback path is working,
4250 * but the callback failed on the client.)
4254 * Mark the callback pathway down, which disabled issuing
4255 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4258 clp->lc_flags |= LCL_CBDOWN;
4262 * Callback worked. If the callback path was down, disable
4263 * callbacks, so no more delegations will be issued. (This
4264 * is done on the assumption that the callback pathway is
4268 if (clp->lc_flags & LCL_CBDOWN)
4269 clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4272 error = nd->nd_repstat;
4273 else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4274 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4275 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4277 mbuf_freem(nd->nd_mrep);
4281 if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4282 clp->lc_flags &= ~LCL_WAKEUPWANTED;
4292 * Set up the compound RPC for the callback.
4295 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4296 uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4301 len = strlen(optag);
4302 (void)nfsm_strtom(nd, optag, len);
4303 NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4304 if ((nd->nd_flag & ND_NFSV41) != 0) {
4305 *tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4306 *tl++ = txdr_unsigned(callback);
4307 *tl++ = txdr_unsigned(2);
4308 *tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4309 error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4312 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4313 *tl = txdr_unsigned(op);
4315 *tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4316 *tl++ = txdr_unsigned(callback);
4317 *tl++ = txdr_unsigned(1);
4318 *tl = txdr_unsigned(op);
4324 * Return the next index# for a clientid. Mostly just increment and return
4325 * the next one, but... if the 32bit unsigned does actually wrap around,
4326 * it should be rebooted.
4327 * At an average rate of one new client per second, it will wrap around in
4328 * approximately 136 years. (I think the server will have been shut
4329 * down or rebooted before then.)
4332 nfsrv_nextclientindex(void)
4334 static u_int32_t client_index = 0;
4337 if (client_index != 0)
4338 return (client_index);
4340 printf("%s: out of clientids\n", __func__);
4341 return (client_index);
4345 * Return the next index# for a stateid. Mostly just increment and return
4346 * the next one, but... if the 32bit unsigned does actually wrap around
4347 * (will a BSD server stay up that long?), find
4348 * new start and end values.
4351 nfsrv_nextstateindex(struct nfsclient *clp)
4353 struct nfsstate *stp;
4355 u_int32_t canuse, min_index, max_index;
4357 if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4358 clp->lc_stateindex++;
4359 if (clp->lc_stateindex != clp->lc_statemaxindex)
4360 return (clp->lc_stateindex);
4364 * Yuck, we've hit the end.
4365 * Look for a new min and max.
4368 max_index = 0xffffffff;
4369 for (i = 0; i < nfsrv_statehashsize; i++) {
4370 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4371 if (stp->ls_stateid.other[2] > 0x80000000) {
4372 if (stp->ls_stateid.other[2] < max_index)
4373 max_index = stp->ls_stateid.other[2];
4375 if (stp->ls_stateid.other[2] > min_index)
4376 min_index = stp->ls_stateid.other[2];
4382 * Yikes, highly unlikely, but I'll handle it anyhow.
4384 if (min_index == 0x80000000 && max_index == 0x80000001) {
4387 * Loop around until we find an unused entry. Return that
4388 * and set LCL_INDEXNOTOK, so the search will continue next time.
4389 * (This is one of those rare cases where a goto is the
4390 * cleanest way to code the loop.)
4393 for (i = 0; i < nfsrv_statehashsize; i++) {
4394 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4395 if (stp->ls_stateid.other[2] == canuse) {
4401 clp->lc_flags |= LCL_INDEXNOTOK;
4406 * Ok to start again from min + 1.
4408 clp->lc_stateindex = min_index + 1;
4409 clp->lc_statemaxindex = max_index;
4410 clp->lc_flags &= ~LCL_INDEXNOTOK;
4411 return (clp->lc_stateindex);
4415 * The following functions handle the stable storage file that deals with
4416 * the edge conditions described in RFC3530 Sec. 8.6.3.
4417 * The file is as follows:
4418 * - a single record at the beginning that has the lease time of the
4419 * previous server instance (before the last reboot) and the nfsrvboottime
4420 * values for the previous server boots.
4421 * These previous boot times are used to ensure that the current
4422 * nfsrvboottime does not, somehow, get set to a previous one.
4423 * (This is important so that Stale ClientIDs and StateIDs can
4425 * The number of previous nfsvrboottime values preceeds the list.
4426 * - followed by some number of appended records with:
4427 * - client id string
4428 * - flag that indicates it is a record revoking state via lease
4429 * expiration or similar
4430 * OR has successfully acquired state.
4431 * These structures vary in length, with the client string at the end, up
4432 * to NFSV4_OPAQUELIMIT in size.
4434 * At the end of the grace period, the file is truncated, the first
4435 * record is rewritten with updated information and any acquired state
4436 * records for successful reclaims of state are written.
4438 * Subsequent records are appended when the first state is issued to
4439 * a client and when state is revoked for a client.
4441 * When reading the file in, state issued records that come later in
4442 * the file override older ones, since the append log is in cronological order.
4443 * If, for some reason, the file can't be read, the grace period is
4444 * immediately terminated and all reclaims get NFSERR_NOGRACE.
4448 * Read in the stable storage file. Called by nfssvc() before the nfsd
4449 * processes start servicing requests.
4452 nfsrv_setupstable(NFSPROC_T *p)
4454 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4455 struct nfsrv_stable *sp, *nsp;
4456 struct nfst_rec *tsp;
4457 int error, i, tryagain;
4459 ssize_t aresid, len;
4462 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4463 * a reboot, so state has not been lost.
4465 if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4468 * Set Grace over just until the file reads successfully.
4470 nfsrvboottime = time_second;
4471 LIST_INIT(&sf->nsf_head);
4472 sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4473 sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4474 if (sf->nsf_fp == NULL)
4476 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4477 (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4478 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4479 if (error || aresid || sf->nsf_numboots == 0 ||
4480 sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4484 * Now, read in the boottimes.
4486 sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4487 sizeof (time_t), M_TEMP, M_WAITOK);
4488 off = sizeof (struct nfsf_rec);
4489 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4490 (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4491 UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4492 if (error || aresid) {
4493 free((caddr_t)sf->nsf_bootvals, M_TEMP);
4494 sf->nsf_bootvals = NULL;
4499 * Make sure this nfsrvboottime is different from all recorded
4504 for (i = 0; i < sf->nsf_numboots; i++) {
4505 if (nfsrvboottime == sf->nsf_bootvals[i]) {
4513 sf->nsf_flags |= NFSNSF_OK;
4514 off += (sf->nsf_numboots * sizeof (time_t));
4517 * Read through the file, building a list of records for grace
4519 * Each record is between sizeof (struct nfst_rec) and
4520 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4521 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4523 tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4524 NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4526 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4527 (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4528 off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4529 len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4530 if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4531 len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4533 * Yuck, the file has been corrupted, so just return
4534 * after clearing out any restart state, so the grace period
4537 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4538 LIST_REMOVE(sp, nst_list);
4539 free((caddr_t)sp, M_TEMP);
4541 free((caddr_t)tsp, M_TEMP);
4542 sf->nsf_flags &= ~NFSNSF_OK;
4543 free((caddr_t)sf->nsf_bootvals, M_TEMP);
4544 sf->nsf_bootvals = NULL;
4548 off += sizeof (struct nfst_rec) + tsp->len - 1;
4550 * Search the list for a matching client.
4552 LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4553 if (tsp->len == sp->nst_len &&
4554 !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4557 if (sp == LIST_END(&sf->nsf_head)) {
4558 sp = (struct nfsrv_stable *)malloc(tsp->len +
4559 sizeof (struct nfsrv_stable) - 1, M_TEMP,
4561 NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4562 sizeof (struct nfst_rec) + tsp->len - 1);
4563 LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4565 if (tsp->flag == NFSNST_REVOKE)
4566 sp->nst_flag |= NFSNST_REVOKE;
4569 * A subsequent timestamp indicates the client
4570 * did a setclientid/confirm and any previous
4571 * revoke is no longer relevant.
4573 sp->nst_flag &= ~NFSNST_REVOKE;
4577 free((caddr_t)tsp, M_TEMP);
4578 sf->nsf_flags = NFSNSF_OK;
4579 sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4584 * Update the stable storage file, now that the grace period is over.
4587 nfsrv_updatestable(NFSPROC_T *p)
4589 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4590 struct nfsrv_stable *sp, *nsp;
4592 struct nfsvattr nva;
4594 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4599 if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4601 sf->nsf_flags |= NFSNSF_UPDATEDONE;
4603 * Ok, we need to rewrite the stable storage file.
4604 * - truncate to 0 length
4605 * - write the new first structure
4606 * - loop through the data structures, writing out any that
4607 * have timestamps older than the old boot
4609 if (sf->nsf_bootvals) {
4611 for (i = sf->nsf_numboots - 2; i >= 0; i--)
4612 sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4614 sf->nsf_numboots = 1;
4615 sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4618 sf->nsf_bootvals[0] = nfsrvboottime;
4619 sf->nsf_lease = nfsrv_lease;
4620 NFSVNO_ATTRINIT(&nva);
4621 NFSVNO_SETATTRVAL(&nva, size, 0);
4622 vp = NFSFPVNODE(sf->nsf_fp);
4623 vn_start_write(vp, &mp, V_WAIT);
4624 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4625 error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4627 NFSVOPUNLOCK(vp, 0);
4630 vn_finished_write(mp);
4632 error = NFSD_RDWR(UIO_WRITE, vp,
4633 (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4634 UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4636 error = NFSD_RDWR(UIO_WRITE, vp,
4637 (caddr_t)sf->nsf_bootvals,
4638 sf->nsf_numboots * sizeof (time_t),
4639 (off_t)(sizeof (struct nfsf_rec)),
4640 UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4641 free((caddr_t)sf->nsf_bootvals, M_TEMP);
4642 sf->nsf_bootvals = NULL;
4644 sf->nsf_flags &= ~NFSNSF_OK;
4645 printf("EEK! Can't write NfsV4 stable storage file\n");
4648 sf->nsf_flags |= NFSNSF_OK;
4651 * Loop through the list and write out timestamp records for
4652 * any clients that successfully reclaimed state.
4654 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4655 if (sp->nst_flag & NFSNST_GOTSTATE) {
4656 nfsrv_writestable(sp->nst_client, sp->nst_len,
4657 NFSNST_NEWSTATE, p);
4658 sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4660 LIST_REMOVE(sp, nst_list);
4661 free((caddr_t)sp, M_TEMP);
4663 nfsrv_backupstable();
4667 * Append a record to the stable storage file.
4670 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4672 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4673 struct nfst_rec *sp;
4676 if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4678 sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4679 len - 1, M_TEMP, M_WAITOK);
4681 NFSBCOPY(client, sp->client, len);
4683 error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4684 (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4685 UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4686 free((caddr_t)sp, M_TEMP);
4688 sf->nsf_flags &= ~NFSNSF_OK;
4689 printf("EEK! Can't write NfsV4 stable storage file\n");
4694 * This function is called during the grace period to mark a client
4695 * that successfully reclaimed state.
4698 nfsrv_markstable(struct nfsclient *clp)
4700 struct nfsrv_stable *sp;
4703 * First find the client structure.
4705 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4706 if (sp->nst_len == clp->lc_idlen &&
4707 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4710 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4714 * Now, just mark it and set the nfsclient back pointer.
4716 sp->nst_flag |= NFSNST_GOTSTATE;
4721 * This function is called for a reclaim, to see if it gets grace.
4722 * It returns 0 if a reclaim is allowed, 1 otherwise.
4725 nfsrv_checkstable(struct nfsclient *clp)
4727 struct nfsrv_stable *sp;
4730 * First, find the entry for the client.
4732 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4733 if (sp->nst_len == clp->lc_idlen &&
4734 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4739 * If not in the list, state was revoked or no state was issued
4740 * since the previous reboot, a reclaim is denied.
4742 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4743 (sp->nst_flag & NFSNST_REVOKE) ||
4744 !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4750 * Test for and try to clear out a conflicting client. This is called by
4751 * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4753 * The trick here is that it can't revoke a conflicting client with an
4754 * expired lease unless it holds the v4root lock, so...
4755 * If no v4root lock, get the lock and return 1 to indicate "try again".
4756 * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4757 * the revocation worked and the conflicting client is "bye, bye", so it
4758 * can be tried again.
4759 * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4760 * Unlocks State before a non-zero value is returned.
4763 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4766 int gotlock, lktype = 0;
4769 * If lease hasn't expired, we can't fix it.
4771 if (clp->lc_expiry >= NFSD_MONOSEC ||
4772 !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4774 if (*haslockp == 0) {
4777 lktype = NFSVOPISLOCKED(vp);
4778 NFSVOPUNLOCK(vp, 0);
4780 NFSLOCKV4ROOTMUTEX();
4781 nfsv4_relref(&nfsv4rootfs_lock);
4783 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4784 NFSV4ROOTLOCKMUTEXPTR, NULL);
4786 NFSUNLOCKV4ROOTMUTEX();
4789 NFSVOPLOCK(vp, lktype | LK_RETRY);
4790 if ((vp->v_iflag & VI_DOOMED) != 0)
4798 * Ok, we can expire the conflicting client.
4800 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4801 nfsrv_backupstable();
4802 nfsrv_cleanclient(clp, p);
4803 nfsrv_freedeleglist(&clp->lc_deleg);
4804 nfsrv_freedeleglist(&clp->lc_olddeleg);
4805 LIST_REMOVE(clp, lc_hash);
4806 nfsrv_zapclient(clp, p);
4811 * Resolve a delegation conflict.
4812 * Returns 0 to indicate the conflict was resolved without sleeping.
4813 * Return -1 to indicate that the caller should check for conflicts again.
4814 * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4816 * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4817 * for a return of 0, since there was no sleep and it could be required
4818 * later. It is released for a return of NFSERR_DELAY, since the caller
4819 * will return that error. It is released when a sleep was done waiting
4820 * for the delegation to be returned or expire (so that other nfsds can
4821 * handle ops). Then, it must be acquired for the write to stable storage.
4822 * (This function is somewhat similar to nfsrv_clientconflict(), but
4823 * the semantics differ in a couple of subtle ways. The return of 0
4824 * indicates the conflict was resolved without sleeping here, not
4825 * that the conflict can't be resolved and the handling of nfsv4root_lock
4826 * differs, as noted above.)
4827 * Unlocks State before returning a non-zero value.
4830 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4833 struct nfsclient *clp = stp->ls_clp;
4834 int gotlock, error, lktype = 0, retrycnt, zapped_clp;
4835 nfsv4stateid_t tstateid;
4839 * If the conflict is with an old delegation...
4841 if (stp->ls_flags & NFSLCK_OLDDELEG) {
4843 * You can delete it, if it has expired.
4845 if (clp->lc_delegtime < NFSD_MONOSEC) {
4846 nfsrv_freedeleg(stp);
4853 * During this delay, the old delegation could expire or it
4854 * could be recovered by the client via an Open with
4855 * CLAIM_DELEGATE_PREV.
4856 * Release the nfsv4root_lock, if held.
4860 NFSLOCKV4ROOTMUTEX();
4861 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4862 NFSUNLOCKV4ROOTMUTEX();
4864 error = NFSERR_DELAY;
4869 * It's a current delegation, so:
4870 * - check to see if the delegation has expired
4871 * - if so, get the v4root lock and then expire it
4873 if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4875 * - do a recall callback, since not yet done
4876 * For now, never allow truncate to be set. To use
4877 * truncate safely, it must be guaranteed that the
4878 * Remove, Rename or Setattr with size of 0 will
4879 * succeed and that would require major changes to
4880 * the VFS/Vnode OPs.
4881 * Set the expiry time large enough so that it won't expire
4882 * until after the callback, then set it correctly, once
4883 * the callback is done. (The delegation will now time
4884 * out whether or not the Recall worked ok. The timeout
4885 * will be extended when ops are done on the delegation
4886 * stateid, up to the timelimit.)
4888 stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4890 stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4892 stp->ls_flags |= NFSLCK_DELEGRECALL;
4895 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4896 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4897 * in order to try and avoid a race that could happen
4898 * when a CBRecall request passed the Open reply with
4899 * the delegation in it when transitting the network.
4900 * Since nfsrv_docallback will sleep, don't use stp after
4903 NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4905 NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4910 NFSLOCKV4ROOTMUTEX();
4911 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4912 NFSUNLOCKV4ROOTMUTEX();
4916 error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4917 &tstateid, 0, &tfh, NULL, NULL, p);
4919 } while ((error == NFSERR_BADSTATEID ||
4920 error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4921 error = NFSERR_DELAY;
4925 if (clp->lc_expiry >= NFSD_MONOSEC &&
4926 stp->ls_delegtime >= NFSD_MONOSEC) {
4929 * A recall has been done, but it has not yet expired.
4934 NFSLOCKV4ROOTMUTEX();
4935 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4936 NFSUNLOCKV4ROOTMUTEX();
4938 error = NFSERR_DELAY;
4943 * If we don't yet have the lock, just get it and then return,
4944 * since we need that before deleting expired state, such as
4946 * When getting the lock, unlock the vnode, so other nfsds that
4947 * are in progress, won't get stuck waiting for the vnode lock.
4949 if (*haslockp == 0) {
4952 lktype = NFSVOPISLOCKED(vp);
4953 NFSVOPUNLOCK(vp, 0);
4955 NFSLOCKV4ROOTMUTEX();
4956 nfsv4_relref(&nfsv4rootfs_lock);
4958 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4959 NFSV4ROOTLOCKMUTEXPTR, NULL);
4961 NFSUNLOCKV4ROOTMUTEX();
4964 NFSVOPLOCK(vp, lktype | LK_RETRY);
4965 if ((vp->v_iflag & VI_DOOMED) != 0) {
4967 NFSLOCKV4ROOTMUTEX();
4968 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4969 NFSUNLOCKV4ROOTMUTEX();
4970 error = NFSERR_PERM;
4980 * Ok, we can delete the expired delegation.
4981 * First, write the Revoke record to stable storage and then
4982 * clear out the conflict.
4983 * Since all other nfsd threads are now blocked, we can safely
4984 * sleep without the state changing.
4986 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4987 nfsrv_backupstable();
4988 if (clp->lc_expiry < NFSD_MONOSEC) {
4989 nfsrv_cleanclient(clp, p);
4990 nfsrv_freedeleglist(&clp->lc_deleg);
4991 nfsrv_freedeleglist(&clp->lc_olddeleg);
4992 LIST_REMOVE(clp, lc_hash);
4995 nfsrv_freedeleg(stp);
4999 nfsrv_zapclient(clp, p);
5008 * Check for a remove allowed, if remove is set to 1 and get rid of
5012 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
5014 struct nfsstate *stp;
5015 struct nfslockfile *lfp;
5016 int error, haslock = 0;
5020 * First, get the lock file structure.
5021 * (A return of -1 means no associated state, so remove ok.)
5023 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5027 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5031 NFSLOCKV4ROOTMUTEX();
5032 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5033 NFSUNLOCKV4ROOTMUTEX();
5041 * Now, we must Recall any delegations.
5043 error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
5046 * nfsrv_cleandeleg() unlocks state for non-zero
5052 NFSLOCKV4ROOTMUTEX();
5053 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5054 NFSUNLOCKV4ROOTMUTEX();
5060 * Now, look for a conflicting open share.
5063 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5064 if (stp->ls_flags & NFSLCK_WRITEDENY) {
5065 error = NFSERR_FILEOPEN;
5073 NFSLOCKV4ROOTMUTEX();
5074 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5075 NFSUNLOCKV4ROOTMUTEX();
5084 * Clear out all delegations for the file referred to by lfp.
5085 * May return NFSERR_DELAY, if there will be a delay waiting for
5086 * delegations to expire.
5087 * Returns -1 to indicate it slept while recalling a delegation.
5088 * This function has the side effect of deleting the nfslockfile structure,
5089 * if it no longer has associated state and didn't have to sleep.
5090 * Unlocks State before a non-zero value is returned.
5093 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5094 struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5096 struct nfsstate *stp, *nstp;
5099 stp = LIST_FIRST(&lfp->lf_deleg);
5100 while (stp != LIST_END(&lfp->lf_deleg)) {
5101 nstp = LIST_NEXT(stp, ls_file);
5102 if (stp->ls_clp != clp) {
5103 ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5106 * nfsrv_delegconflict() unlocks state
5107 * when it returns non-zero.
5120 * There are certain operations that, when being done outside of NFSv4,
5121 * require that any NFSv4 delegation for the file be recalled.
5122 * This function is to be called for those cases:
5123 * VOP_RENAME() - When a delegation is being recalled for any reason,
5124 * the client may have to do Opens against the server, using the file's
5125 * final component name. If the file has been renamed on the server,
5126 * that component name will be incorrect and the Open will fail.
5127 * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5128 * been removed on the server, if there is a delegation issued to
5129 * that client for the file. I say "theoretically" since clients
5130 * normally do an Access Op before the Open and that Access Op will
5131 * fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5132 * they will detect the file's removal in the same manner. (There is
5133 * one case where RFC3530 allows a client to do an Open without first
5134 * doing an Access Op, which is passage of a check against the ACE
5135 * returned with a Write delegation, but current practice is to ignore
5136 * the ACE and always do an Access Op.)
5137 * Since the functions can only be called with an unlocked vnode, this
5138 * can't be done at this time.
5139 * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5140 * locks locally in the client, which are not visible to the server. To
5141 * deal with this, issuing of delegations for a vnode must be disabled
5142 * and all delegations for the vnode recalled. This is done via the
5143 * second function, using the VV_DISABLEDELEG vflag on the vnode.
5146 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5152 * First, check to see if the server is currently running and it has
5153 * been called for a regular file when issuing delegations.
5155 if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5156 nfsrv_issuedelegs == 0)
5159 KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5161 * First, get a reference on the nfsv4rootfs_lock so that an
5162 * exclusive lock cannot be acquired by another thread.
5164 NFSLOCKV4ROOTMUTEX();
5165 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5166 NFSUNLOCKV4ROOTMUTEX();
5169 * Now, call nfsrv_checkremove() in a loop while it returns
5170 * NFSERR_DELAY. Return upon any other error or when timed out.
5172 starttime = NFSD_MONOSEC;
5174 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5175 error = nfsrv_checkremove(vp, 0, p);
5176 NFSVOPUNLOCK(vp, 0);
5179 if (error == NFSERR_DELAY) {
5180 if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5182 /* Sleep for a short period of time */
5183 (void) nfs_catnap(PZERO, 0, "nfsremove");
5185 } while (error == NFSERR_DELAY);
5186 NFSLOCKV4ROOTMUTEX();
5187 nfsv4_relref(&nfsv4rootfs_lock);
5188 NFSUNLOCKV4ROOTMUTEX();
5192 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5195 #ifdef VV_DISABLEDELEG
5197 * First, flag issuance of delegations disabled.
5199 atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5203 * Then call nfsd_recalldelegation() to get rid of all extant
5206 nfsd_recalldelegation(vp, p);
5210 * Check for conflicting locks, etc. and then get rid of delegations.
5211 * (At one point I thought that I should get rid of delegations for any
5212 * Setattr, since it could potentially disallow the I/O op (read or write)
5213 * allowed by the delegation. However, Setattr Ops that aren't changing
5214 * the size get a stateid of all 0s, so you can't tell if it is a delegation
5215 * for the same client or a different one, so I decided to only get rid
5216 * of delegations for other clients when the size is being changed.)
5217 * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5218 * as Write backs, even if there is no delegation, so it really isn't any
5222 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5223 nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5224 struct nfsexstuff *exp, NFSPROC_T *p)
5226 struct nfsstate st, *stp = &st;
5227 struct nfslock lo, *lop = &lo;
5231 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5232 stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5233 lop->lo_first = nvap->na_size;
5238 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5239 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5240 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5241 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5242 stp->ls_flags |= NFSLCK_SETATTR;
5243 if (stp->ls_flags == 0)
5245 lop->lo_end = NFS64BITSSET;
5246 lop->lo_flags = NFSLCK_WRITE;
5247 stp->ls_ownerlen = 0;
5249 stp->ls_uid = nd->nd_cred->cr_uid;
5250 stp->ls_stateid.seqid = stateidp->seqid;
5251 clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5252 clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5253 stp->ls_stateid.other[2] = stateidp->other[2];
5254 error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5255 stateidp, exp, nd, p);
5258 NFSEXITCODE2(error, nd);
5263 * Check for a write delegation and do a CBGETATTR if there is one, updating
5264 * the attributes, as required.
5265 * Should I return an error if I can't get the attributes? (For now, I'll
5269 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5270 struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5273 struct nfsstate *stp;
5274 struct nfslockfile *lfp;
5275 struct nfsclient *clp;
5276 struct nfsvattr nva;
5279 nfsattrbit_t cbbits;
5280 u_quad_t delegfilerev;
5282 NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5283 if (!NFSNONZERO_ATTRBIT(&cbbits))
5287 * Get the lock file structure.
5288 * (A return of -1 means no associated state, so return ok.)
5290 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5293 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5302 * Now, look for a write delegation.
5304 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5305 if (stp->ls_flags & NFSLCK_DELEGWRITE)
5308 if (stp == LIST_END(&lfp->lf_deleg)) {
5313 delegfilerev = stp->ls_filerev;
5316 * If the Write delegation was issued as a part of this Compound RPC
5317 * or if we have an Implied Clientid (used in a previous Op in this
5318 * compound) and it is the client the delegation was issued to,
5320 * I also assume that it is from the same client iff the network
5321 * host IP address is the same as the callback address. (Not
5322 * exactly correct by the RFC, but avoids a lot of Getattr
5325 if (nd->nd_compref == stp->ls_compref ||
5326 ((nd->nd_flag & ND_IMPLIEDCLID) &&
5327 clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5328 nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5334 * We are now done with the delegation state structure,
5335 * so the statelock can be released and we can now tsleep().
5339 * Now, we must do the CB Getattr callback, to see if Change or Size
5342 if (clp->lc_expiry >= NFSD_MONOSEC) {
5344 NFSVNO_ATTRINIT(&nva);
5345 nva.na_filerev = NFS64BITSSET;
5346 error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5347 0, &nfh, &nva, &cbbits, p);
5349 if ((nva.na_filerev != NFS64BITSSET &&
5350 nva.na_filerev > delegfilerev) ||
5351 (NFSVNO_ISSETSIZE(&nva) &&
5352 nva.na_size != nvap->na_size)) {
5353 error = nfsvno_updfilerev(vp, nvap, cred, p);
5354 if (NFSVNO_ISSETSIZE(&nva))
5355 nvap->na_size = nva.na_size;
5358 error = 0; /* Ignore callback errors for now. */
5364 NFSEXITCODE2(error, nd);
5369 * This function looks for openowners that haven't had any opens for
5370 * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5374 nfsrv_throwawayopens(NFSPROC_T *p)
5376 struct nfsclient *clp, *nclp;
5377 struct nfsstate *stp, *nstp;
5381 nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5383 * For each client...
5385 for (i = 0; i < nfsrv_clienthashsize; i++) {
5386 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5387 LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5388 if (LIST_EMPTY(&stp->ls_open) &&
5389 (stp->ls_noopens > NFSNOOPEN ||
5390 (nfsrv_openpluslock * 2) >
5391 nfsrv_v4statelimit))
5392 nfsrv_freeopenowner(stp, 0, p);
5400 * This function checks to see if the credentials are the same.
5401 * Returns 1 for not same, 0 otherwise.
5404 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5407 if (nd->nd_flag & ND_GSS) {
5408 if (!(clp->lc_flags & LCL_GSS))
5410 if (clp->lc_flags & LCL_NAME) {
5411 if (nd->nd_princlen != clp->lc_namelen ||
5412 NFSBCMP(nd->nd_principal, clp->lc_name,
5418 if (nd->nd_cred->cr_uid == clp->lc_uid)
5422 } else if (clp->lc_flags & LCL_GSS)
5425 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5426 * in RFC3530, which talks about principals, but doesn't say anything
5427 * about uids for AUTH_SYS.)
5429 if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5436 * Calculate the lease expiry time.
5439 nfsrv_leaseexpiry(void)
5442 if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5443 return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5444 return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5448 * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5451 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5454 if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5457 if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5458 stp->ls_delegtime < stp->ls_delegtimelimit) {
5459 stp->ls_delegtime += nfsrv_lease;
5460 if (stp->ls_delegtime > stp->ls_delegtimelimit)
5461 stp->ls_delegtime = stp->ls_delegtimelimit;
5466 * This function checks to see if there is any other state associated
5467 * with the openowner for this Open.
5468 * It returns 1 if there is no other state, 0 otherwise.
5471 nfsrv_nootherstate(struct nfsstate *stp)
5473 struct nfsstate *tstp;
5475 LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5476 if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5483 * Create a list of lock deltas (changes to local byte range locking
5484 * that can be rolled back using the list) and apply the changes via
5485 * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5486 * the rollback or update function will be called after this.
5487 * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5488 * call fails. If it returns an error, it will unlock the list.
5491 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5492 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5494 struct nfslock *lop, *nlop;
5497 /* Loop through the list of locks. */
5498 lop = LIST_FIRST(&lfp->lf_locallock);
5499 while (first < end && lop != NULL) {
5500 nlop = LIST_NEXT(lop, lo_lckowner);
5501 if (first >= lop->lo_end) {
5504 } else if (first < lop->lo_first) {
5505 /* new one starts before entry in list */
5506 if (end <= lop->lo_first) {
5507 /* no overlap between old and new */
5508 error = nfsrv_dolocal(vp, lfp, flags,
5509 NFSLCK_UNLOCK, first, end, cfp, p);
5514 /* handle fragment overlapped with new one */
5515 error = nfsrv_dolocal(vp, lfp, flags,
5516 NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5520 first = lop->lo_first;
5523 /* new one overlaps this entry in list */
5524 if (end <= lop->lo_end) {
5525 /* overlaps all of new one */
5526 error = nfsrv_dolocal(vp, lfp, flags,
5527 lop->lo_flags, first, end, cfp, p);
5532 /* handle fragment overlapped with new one */
5533 error = nfsrv_dolocal(vp, lfp, flags,
5534 lop->lo_flags, first, lop->lo_end, cfp, p);
5537 first = lop->lo_end;
5542 if (first < end && error == 0)
5543 /* handle fragment past end of list */
5544 error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5552 * Local lock unlock. Unlock all byte ranges that are no longer locked
5553 * by NFSv4. To do this, unlock any subranges of first-->end that
5554 * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5555 * list. This list has all locks for the file held by other
5556 * <clientid, lockowner> tuples. The list is ordered by increasing
5557 * lo_first value, but may have entries that overlap each other, for
5558 * the case of read locks.
5561 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5562 uint64_t init_end, NFSPROC_T *p)
5564 struct nfslock *lop;
5565 uint64_t first, end, prevfirst;
5569 while (first < init_end) {
5570 /* Loop through all nfs locks, adjusting first and end */
5572 LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5573 KASSERT(prevfirst <= lop->lo_first,
5574 ("nfsv4 locks out of order"));
5575 KASSERT(lop->lo_first < lop->lo_end,
5576 ("nfsv4 bogus lock"));
5577 prevfirst = lop->lo_first;
5578 if (first >= lop->lo_first &&
5579 first < lop->lo_end)
5581 * Overlaps with initial part, so trim
5582 * off that initial part by moving first past
5585 first = lop->lo_end;
5586 else if (end > lop->lo_first &&
5587 lop->lo_first > first) {
5589 * This lock defines the end of the
5590 * segment to unlock, so set end to the
5591 * start of it and break out of the loop.
5593 end = lop->lo_first;
5598 * There is no segment left to do, so
5599 * break out of this loop and then exit
5600 * the outer while() since first will be set
5601 * to end, which must equal init_end here.
5606 /* Unlock this segment */
5607 (void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5608 NFSLCK_READ, first, end, NULL, p);
5609 nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5613 * Now move past this segment and look for any further
5614 * segment in the range, if there is one.
5622 * Do the local lock operation and update the rollback list, as required.
5623 * Perform the rollback and return the error if nfsvno_advlock() fails.
5626 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5627 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5629 struct nfsrollback *rlp;
5630 int error = 0, ltype, oldltype;
5632 if (flags & NFSLCK_WRITE)
5634 else if (flags & NFSLCK_READ)
5638 if (oldflags & NFSLCK_WRITE)
5640 else if (oldflags & NFSLCK_READ)
5644 if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5647 error = nfsvno_advlock(vp, ltype, first, end, p);
5650 cfp->cl_clientid.lval[0] = 0;
5651 cfp->cl_clientid.lval[1] = 0;
5653 cfp->cl_end = NFS64BITSSET;
5654 cfp->cl_flags = NFSLCK_WRITE;
5655 cfp->cl_ownerlen = 5;
5656 NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5658 nfsrv_locallock_rollback(vp, lfp, p);
5659 } else if (ltype != F_UNLCK) {
5660 rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5662 rlp->rlck_first = first;
5663 rlp->rlck_end = end;
5664 rlp->rlck_type = oldltype;
5665 LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5674 * Roll back local lock changes and free up the rollback list.
5677 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5679 struct nfsrollback *rlp, *nrlp;
5681 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5682 (void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5684 free(rlp, M_NFSDROLLBACK);
5686 LIST_INIT(&lfp->lf_rollback);
5690 * Update local lock list and delete rollback list (ie now committed to the
5691 * local locks). Most of the work is done by the internal function.
5694 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5697 struct nfsrollback *rlp, *nrlp;
5698 struct nfslock *new_lop, *other_lop;
5700 new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5701 if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5702 other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5706 new_lop->lo_flags = flags;
5707 new_lop->lo_first = first;
5708 new_lop->lo_end = end;
5709 nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5710 if (new_lop != NULL)
5711 free(new_lop, M_NFSDLOCK);
5712 if (other_lop != NULL)
5713 free(other_lop, M_NFSDLOCK);
5715 /* and get rid of the rollback list */
5716 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5717 free(rlp, M_NFSDROLLBACK);
5718 LIST_INIT(&lfp->lf_rollback);
5722 * Lock the struct nfslockfile for local lock updating.
5725 nfsrv_locklf(struct nfslockfile *lfp)
5729 /* lf_usecount ensures *lfp won't be free'd */
5732 gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5733 NFSSTATEMUTEXPTR, NULL);
5734 } while (gotlock == 0);
5739 * Unlock the struct nfslockfile after local lock updating.
5742 nfsrv_unlocklf(struct nfslockfile *lfp)
5745 nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5749 * Clear out all state for the NFSv4 server.
5750 * Must be called by a thread that can sleep when no nfsds are running.
5753 nfsrv_throwawayallstate(NFSPROC_T *p)
5755 struct nfsclient *clp, *nclp;
5756 struct nfslockfile *lfp, *nlfp;
5760 * For each client, clean out the state and then free the structure.
5762 for (i = 0; i < nfsrv_clienthashsize; i++) {
5763 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5764 nfsrv_cleanclient(clp, p);
5765 nfsrv_freedeleglist(&clp->lc_deleg);
5766 nfsrv_freedeleglist(&clp->lc_olddeleg);
5767 free(clp->lc_stateid, M_NFSDCLIENT);
5768 free(clp, M_NFSDCLIENT);
5773 * Also, free up any remaining lock file structures.
5775 for (i = 0; i < nfsrv_lockhashsize; i++) {
5776 LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
5777 printf("nfsd unload: fnd a lock file struct\n");
5778 nfsrv_freenfslockfile(lfp);
5784 * Check the sequence# for the session and slot provided as an argument.
5785 * Also, renew the lease if the session will return NFS_OK.
5788 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
5789 uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
5790 uint32_t *sflagsp, NFSPROC_T *p)
5792 struct nfsdsession *sep;
5793 struct nfssessionhash *shp;
5797 shp = NFSSESSIONHASH(nd->nd_sessionid);
5798 NFSLOCKSESSION(shp);
5799 sep = nfsrv_findsession(nd->nd_sessionid);
5801 NFSUNLOCKSESSION(shp);
5802 return (NFSERR_BADSESSION);
5804 error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
5805 sep->sess_slots, NULL, NFSV4_SLOTS - 1);
5807 NFSUNLOCKSESSION(shp);
5810 if (cache_this != 0)
5811 nd->nd_flag |= ND_SAVEREPLY;
5812 /* Renew the lease. */
5813 sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
5814 nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
5815 nd->nd_flag |= ND_IMPLIEDCLID;
5818 * If this session handles the backchannel, save the nd_xprt for this
5819 * RPC, since this is the one being used.
5821 if (sep->sess_clp->lc_req.nr_client != NULL &&
5822 (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) {
5823 savxprt = sep->sess_cbsess.nfsess_xprt;
5824 SVC_ACQUIRE(nd->nd_xprt);
5825 nd->nd_xprt->xp_p2 =
5826 sep->sess_clp->lc_req.nr_client->cl_private;
5827 nd->nd_xprt->xp_idletimeout = 0; /* Disable timeout. */
5828 sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
5829 if (savxprt != NULL)
5830 SVC_RELEASE(savxprt);
5834 if (sep->sess_clp->lc_req.nr_client == NULL)
5835 *sflagsp |= NFSV4SEQ_CBPATHDOWN;
5836 NFSUNLOCKSESSION(shp);
5837 if (error == NFSERR_EXPIRED) {
5838 *sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
5840 } else if (error == NFSERR_ADMINREVOKED) {
5841 *sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
5844 *highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
5849 * Check/set reclaim complete for this session/clientid.
5852 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd)
5854 struct nfsdsession *sep;
5855 struct nfssessionhash *shp;
5858 shp = NFSSESSIONHASH(nd->nd_sessionid);
5860 NFSLOCKSESSION(shp);
5861 sep = nfsrv_findsession(nd->nd_sessionid);
5863 NFSUNLOCKSESSION(shp);
5865 return (NFSERR_BADSESSION);
5868 /* Check to see if reclaim complete has already happened. */
5869 if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
5870 error = NFSERR_COMPLETEALREADY;
5872 sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
5873 NFSUNLOCKSESSION(shp);
5879 * Cache the reply in a session slot.
5882 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
5885 struct nfsdsession *sep;
5886 struct nfssessionhash *shp;
5888 shp = NFSSESSIONHASH(sessionid);
5889 NFSLOCKSESSION(shp);
5890 sep = nfsrv_findsession(sessionid);
5892 NFSUNLOCKSESSION(shp);
5893 printf("nfsrv_cache_session: no session\n");
5897 nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
5898 NFSUNLOCKSESSION(shp);
5902 * Search for a session that matches the sessionid.
5904 static struct nfsdsession *
5905 nfsrv_findsession(uint8_t *sessionid)
5907 struct nfsdsession *sep;
5908 struct nfssessionhash *shp;
5910 shp = NFSSESSIONHASH(sessionid);
5911 LIST_FOREACH(sep, &shp->list, sess_hash) {
5912 if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
5919 * Destroy a session.
5922 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
5924 int error, samesess;
5927 if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID)) {
5929 if ((nd->nd_flag & ND_LASTOP) == 0)
5930 return (NFSERR_BADSESSION);
5932 error = nfsrv_freesession(NULL, sessionid);
5933 if (error == 0 && samesess != 0)
5934 nd->nd_flag &= ~ND_HASSEQUENCE;
5939 * Free up a session structure.
5942 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
5944 struct nfssessionhash *shp;
5949 shp = NFSSESSIONHASH(sessionid);
5950 NFSLOCKSESSION(shp);
5951 sep = nfsrv_findsession(sessionid);
5953 shp = NFSSESSIONHASH(sep->sess_sessionid);
5954 NFSLOCKSESSION(shp);
5958 if (sep->sess_refcnt > 0) {
5959 NFSUNLOCKSESSION(shp);
5963 LIST_REMOVE(sep, sess_hash);
5964 LIST_REMOVE(sep, sess_list);
5966 NFSUNLOCKSESSION(shp);
5969 return (NFSERR_BADSESSION);
5970 for (i = 0; i < NFSV4_SLOTS; i++)
5971 if (sep->sess_slots[i].nfssl_reply != NULL)
5972 m_freem(sep->sess_slots[i].nfssl_reply);
5973 if (sep->sess_cbsess.nfsess_xprt != NULL)
5974 SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
5975 free(sep, M_NFSDSESSION);
5981 * RFC5661 says that it should fail when there are associated opens, locks
5982 * or delegations. Since stateids represent opens, I don't see how you can
5983 * free an open stateid (it will be free'd when closed), so this function
5984 * only works for lock stateids (freeing the lock_owner) or delegations.
5987 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
5990 struct nfsclient *clp;
5991 struct nfsstate *stp;
5996 * Look up the stateid
5998 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
5999 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6001 /* First, check for a delegation. */
6002 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6003 if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6008 nfsrv_freedeleg(stp);
6013 /* Not a delegation, try for a lock_owner. */
6015 error = nfsrv_getstate(clp, stateidp, 0, &stp);
6016 if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6017 NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6018 /* Not a lock_owner stateid. */
6019 error = NFSERR_LOCKSHELD;
6020 if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6021 error = NFSERR_LOCKSHELD;
6023 nfsrv_freelockowner(stp, NULL, 0, p);
6029 * Generate the xdr for an NFSv4.1 CBSequence Operation.
6032 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6033 int dont_replycache, struct nfsdsession **sepp)
6035 struct nfsdsession *sep;
6036 uint32_t *tl, slotseq = 0;
6037 int maxslot, slotpos;
6038 uint8_t sessionid[NFSX_V4SESSIONID];
6041 error = nfsv4_getcbsession(clp, sepp);
6045 (void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6046 &slotseq, sessionid);
6047 KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6049 /* Build the Sequence arguments. */
6050 NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6051 bcopy(sessionid, tl, NFSX_V4SESSIONID);
6052 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6053 nd->nd_slotseq = tl;
6054 *tl++ = txdr_unsigned(slotseq);
6055 *tl++ = txdr_unsigned(slotpos);
6056 *tl++ = txdr_unsigned(maxslot);
6057 if (dont_replycache == 0)
6058 *tl++ = newnfs_true;
6060 *tl++ = newnfs_false;
6061 *tl = 0; /* No referring call list, for now. */
6062 nd->nd_flag |= ND_HASSEQUENCE;
6067 * Get a session for the callback.
6070 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6072 struct nfsdsession *sep;
6075 LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6076 if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6081 return (NFSERR_BADSESSION);
6090 * Free up all backchannel xprts. This needs to be done when the nfsd threads
6091 * exit, since those transports will all be going away.
6092 * This is only called after all the nfsd threads are done performing RPCs,
6093 * so locking shouldn't be an issue.
6096 nfsrv_freeallbackchannel_xprts(void)
6098 struct nfsdsession *sep;
6099 struct nfsclient *clp;
6103 for (i = 0; i < nfsrv_clienthashsize; i++) {
6104 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6105 LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6106 xprt = sep->sess_cbsess.nfsess_xprt;
6107 sep->sess_cbsess.nfsess_xprt = NULL;