2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2009 Rick Macklem, University of Guelph
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
34 #include "opt_inet6.h"
35 #include <sys/extattr.h>
36 #include <fs/nfs/nfsport.h>
38 struct nfsrv_stablefirst nfsrv_stablefirst;
39 int nfsrv_issuedelegs = 0;
40 int nfsrv_dolocallocks = 0;
41 struct nfsv4lock nfsv4rootfs_lock;
42 time_t nfsdev_time = 0;
43 int nfsrv_layouthashsize;
44 volatile int nfsrv_layoutcnt = 0;
46 extern int newnfs_numnfsd;
47 extern struct nfsstatsv1 nfsstatsv1;
48 extern int nfsrv_lease;
49 extern struct timeval nfsboottime;
50 extern u_int32_t newnfs_true, newnfs_false;
51 extern struct mtx nfsrv_dslock_mtx;
52 extern struct mtx nfsrv_recalllock_mtx;
53 extern struct mtx nfsrv_dontlistlock_mtx;
54 extern int nfsd_debuglevel;
55 extern u_int nfsrv_dsdirsize;
56 extern struct nfsdevicehead nfsrv_devidhead;
57 extern int nfsrv_doflexfile;
58 extern int nfsrv_maxpnfsmirror;
61 extern struct nfsdontlisthead nfsrv_dontlisthead;
62 extern volatile int nfsrv_devidcnt;
63 extern struct nfslayouthead nfsrv_recalllisthead;
64 extern char *nfsrv_zeropnfsdat;
66 SYSCTL_DECL(_vfs_nfsd);
67 int nfsrv_statehashsize = NFSSTATEHASHSIZE;
68 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
69 &nfsrv_statehashsize, 0,
70 "Size of state hash table set via loader.conf");
72 int nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
73 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
74 &nfsrv_clienthashsize, 0,
75 "Size of client hash table set via loader.conf");
77 int nfsrv_lockhashsize = NFSLOCKHASHSIZE;
78 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
79 &nfsrv_lockhashsize, 0,
80 "Size of file handle hash table set via loader.conf");
82 int nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
83 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
84 &nfsrv_sessionhashsize, 0,
85 "Size of session hash table set via loader.conf");
87 int nfsrv_layouthighwater = NFSLAYOUTHIGHWATER;
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN,
89 &nfsrv_layouthighwater, 0,
90 "High water mark for number of layouts set via loader.conf");
92 static int nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
93 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
94 &nfsrv_v4statelimit, 0,
95 "High water limit for NFSv4 opens+locks+delegations");
97 static int nfsrv_writedelegifpos = 0;
98 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
99 &nfsrv_writedelegifpos, 0,
100 "Issue a write delegation for read opens if possible");
102 static int nfsrv_allowreadforwriteopen = 1;
103 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
104 &nfsrv_allowreadforwriteopen, 0,
105 "Allow Reads to be done with Write Access StateIDs");
107 int nfsrv_pnfsatime = 0;
108 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW,
110 "For pNFS service, do Getattr ops to keep atime up-to-date");
112 int nfsrv_flexlinuxhack = 0;
113 SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW,
114 &nfsrv_flexlinuxhack, 0,
115 "For Linux clients, hack around Flex File Layout bug");
118 * Hash lists for nfs V4.
120 struct nfsclienthashhead *nfsclienthash;
121 struct nfslockhashhead *nfslockhash;
122 struct nfssessionhash *nfssessionhash;
123 struct nfslayouthash *nfslayouthash;
124 volatile int nfsrv_dontlistlen = 0;
126 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
127 static time_t nfsrvboottime;
128 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
129 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
130 static int nfsrv_nogsscallback = 0;
131 static volatile int nfsrv_writedelegcnt = 0;
132 static int nfsrv_faildscnt;
134 /* local functions */
135 static void nfsrv_dumpaclient(struct nfsclient *clp,
136 struct nfsd_dumpclients *dumpp);
137 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
139 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
141 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
143 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
144 int cansleep, NFSPROC_T *p);
145 static void nfsrv_freenfslock(struct nfslock *lop);
146 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
147 static void nfsrv_freedeleg(struct nfsstate *);
148 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
149 u_int32_t flags, struct nfsstate **stpp);
150 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
151 struct nfsstate **stpp);
152 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
153 struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
154 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
155 struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
156 static void nfsrv_insertlock(struct nfslock *new_lop,
157 struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
158 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
159 struct nfslock **other_lopp, struct nfslockfile *lfp);
160 static int nfsrv_getipnumber(u_char *cp);
161 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
162 nfsv4stateid_t *stateidp, int specialid);
163 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
165 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
166 nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
167 struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
168 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
169 uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
170 static u_int32_t nfsrv_nextclientindex(void);
171 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
172 static void nfsrv_markstable(struct nfsclient *clp);
173 static void nfsrv_markreclaim(struct nfsclient *clp);
174 static int nfsrv_checkstable(struct nfsclient *clp);
175 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
176 vnode *vp, NFSPROC_T *p);
177 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
178 NFSPROC_T *p, vnode_t vp);
179 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
180 struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
181 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
182 struct nfsclient *clp);
183 static time_t nfsrv_leaseexpiry(void);
184 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
185 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
186 struct nfsstate *stp, struct nfsrvcache *op);
187 static int nfsrv_nootherstate(struct nfsstate *stp);
188 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
189 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
190 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
191 uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
192 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
193 int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
195 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
197 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
198 uint64_t first, uint64_t end);
199 static void nfsrv_locklf(struct nfslockfile *lfp);
200 static void nfsrv_unlocklf(struct nfslockfile *lfp);
201 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
202 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
203 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
204 int dont_replycache, struct nfsdsession **sepp);
205 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
206 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
207 nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
208 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
209 static void nfsrv_freelayoutlist(nfsquad_t clientid);
210 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
212 static void nfsrv_freealllayouts(void);
213 static void nfsrv_freedevid(struct nfsdevice *ds);
214 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
215 struct nfsdevice **dsp);
216 static void nfsrv_deleteds(struct nfsdevice *fndds);
217 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
218 static void nfsrv_freealldevids(void);
219 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
220 int maxcnt, NFSPROC_T *p);
221 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
222 fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
224 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
225 NFSPROC_T *, struct nfslayout **lypp);
226 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
227 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
228 fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
229 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
230 int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
231 static int nfsrv_dontlayout(fhandle_t *fhp);
232 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
233 vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
235 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
238 * Scan the client list for a match and either return the current one,
239 * create a new entry or return an error.
240 * If returning a non-error, the clp structure must either be linked into
241 * the client list or free'd.
244 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
245 nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
247 struct nfsclient *clp = NULL, *new_clp = *new_clpp;
248 int i, error = 0, ret;
249 struct nfsstate *stp, *tstp;
251 struct sockaddr_in *sin, *rin;
254 struct sockaddr_in6 *sin6, *rin6;
256 struct nfsdsession *sep, *nsep;
257 int zapit = 0, gotit, hasstate = 0, igotlock;
258 static u_int64_t confirm_index = 0;
261 * Check for state resource limit exceeded.
263 if (nfsrv_openpluslock > nfsrv_v4statelimit) {
264 error = NFSERR_RESOURCE;
268 if (nfsrv_issuedelegs == 0 ||
269 ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
271 * Don't do callbacks when delegations are disabled or
272 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
273 * If establishing a callback connection is attempted
274 * when a firewall is blocking the callback path, the
275 * server may wait too long for the connect attempt to
276 * succeed during the Open. Some clients, such as Linux,
277 * may timeout and give up on the Open before the server
278 * replies. Also, since AUTH_GSS callbacks are not
279 * yet interoperability tested, they might cause the
280 * server to crap out, if they get past the Init call to
283 new_clp->lc_program = 0;
285 /* Lock out other nfsd threads */
286 NFSLOCKV4ROOTMUTEX();
287 nfsv4_relref(&nfsv4rootfs_lock);
289 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
290 NFSV4ROOTLOCKMUTEXPTR, NULL);
292 NFSUNLOCKV4ROOTMUTEX();
295 * Search for a match in the client list.
298 while (i < nfsrv_clienthashsize && !gotit) {
299 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
300 if (new_clp->lc_idlen == clp->lc_idlen &&
301 !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
310 (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
311 if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
313 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
314 * client is trying to update a confirmed clientid.
316 NFSLOCKV4ROOTMUTEX();
317 nfsv4_unlock(&nfsv4rootfs_lock, 1);
318 NFSUNLOCKV4ROOTMUTEX();
319 confirmp->lval[1] = 0;
320 error = NFSERR_NOENT;
324 * Get rid of the old one.
326 if (i != nfsrv_clienthashsize) {
327 LIST_REMOVE(clp, lc_hash);
328 nfsrv_cleanclient(clp, p);
329 nfsrv_freedeleglist(&clp->lc_deleg);
330 nfsrv_freedeleglist(&clp->lc_olddeleg);
334 * Add it after assigning a client id to it.
336 new_clp->lc_flags |= LCL_NEEDSCONFIRM;
337 if ((nd->nd_flag & ND_NFSV41) != 0)
338 new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
341 confirmp->qval = new_clp->lc_confirm.qval =
343 clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
344 (u_int32_t)nfsrvboottime;
345 clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
346 nfsrv_nextclientindex();
347 new_clp->lc_stateindex = 0;
348 new_clp->lc_statemaxindex = 0;
349 new_clp->lc_cbref = 0;
350 new_clp->lc_expiry = nfsrv_leaseexpiry();
351 LIST_INIT(&new_clp->lc_open);
352 LIST_INIT(&new_clp->lc_deleg);
353 LIST_INIT(&new_clp->lc_olddeleg);
354 LIST_INIT(&new_clp->lc_session);
355 for (i = 0; i < nfsrv_statehashsize; i++)
356 LIST_INIT(&new_clp->lc_stateid[i]);
357 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
359 nfsstatsv1.srvclients++;
360 nfsrv_openpluslock++;
362 NFSLOCKV4ROOTMUTEX();
363 nfsv4_unlock(&nfsv4rootfs_lock, 1);
364 NFSUNLOCKV4ROOTMUTEX();
366 nfsrv_zapclient(clp, p);
372 * Now, handle the cases where the id is already issued.
374 if (nfsrv_notsamecredname(nd, clp)) {
376 * Check to see if there is expired state that should go away.
378 if (clp->lc_expiry < NFSD_MONOSEC &&
379 (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
380 nfsrv_cleanclient(clp, p);
381 nfsrv_freedeleglist(&clp->lc_deleg);
385 * If there is outstanding state, then reply NFSERR_CLIDINUSE per
386 * RFC3530 Sec. 8.1.2 last para.
388 if (!LIST_EMPTY(&clp->lc_deleg)) {
390 } else if (LIST_EMPTY(&clp->lc_open)) {
394 /* Look for an Open on the OpenOwner */
395 LIST_FOREACH(stp, &clp->lc_open, ls_list) {
396 if (!LIST_EMPTY(&stp->ls_open)) {
404 * If the uid doesn't match, return NFSERR_CLIDINUSE after
405 * filling out the correct ipaddr and portnum.
407 switch (clp->lc_req.nr_nam->sa_family) {
410 sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
411 rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
412 sin->sin_addr.s_addr = rin->sin_addr.s_addr;
413 sin->sin_port = rin->sin_port;
418 sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
419 rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
420 sin6->sin6_addr = rin6->sin6_addr;
421 sin6->sin6_port = rin6->sin6_port;
425 NFSLOCKV4ROOTMUTEX();
426 nfsv4_unlock(&nfsv4rootfs_lock, 1);
427 NFSUNLOCKV4ROOTMUTEX();
428 error = NFSERR_CLIDINUSE;
433 if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
435 * If the verifier has changed, the client has rebooted
436 * and a new client id is issued. The old state info
437 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
439 LIST_REMOVE(clp, lc_hash);
441 /* Get rid of all sessions on this clientid. */
442 LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
443 ret = nfsrv_freesession(sep, NULL);
445 printf("nfsrv_setclient: verifier changed free"
446 " session failed=%d\n", ret);
449 new_clp->lc_flags |= LCL_NEEDSCONFIRM;
450 if ((nd->nd_flag & ND_NFSV41) != 0)
451 new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
454 confirmp->qval = new_clp->lc_confirm.qval =
456 clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
458 clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
459 nfsrv_nextclientindex();
460 new_clp->lc_stateindex = 0;
461 new_clp->lc_statemaxindex = 0;
462 new_clp->lc_cbref = 0;
463 new_clp->lc_expiry = nfsrv_leaseexpiry();
466 * Save the state until confirmed.
468 LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
469 LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
470 tstp->ls_clp = new_clp;
471 LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
472 LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
473 tstp->ls_clp = new_clp;
474 LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
476 LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
477 tstp->ls_clp = new_clp;
478 for (i = 0; i < nfsrv_statehashsize; i++) {
479 LIST_NEWHEAD(&new_clp->lc_stateid[i],
480 &clp->lc_stateid[i], ls_hash);
481 LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
482 tstp->ls_clp = new_clp;
484 LIST_INIT(&new_clp->lc_session);
485 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
487 nfsstatsv1.srvclients++;
488 nfsrv_openpluslock++;
490 NFSLOCKV4ROOTMUTEX();
491 nfsv4_unlock(&nfsv4rootfs_lock, 1);
492 NFSUNLOCKV4ROOTMUTEX();
495 * Must wait until any outstanding callback on the old clp
499 while (clp->lc_cbref) {
500 clp->lc_flags |= LCL_WAKEUPWANTED;
501 (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
502 "nfsd clp", 10 * hz);
505 nfsrv_zapclient(clp, p);
510 /* For NFSv4.1, mark that we found a confirmed clientid. */
511 if ((nd->nd_flag & ND_NFSV41) != 0) {
512 clientidp->lval[0] = clp->lc_clientid.lval[0];
513 clientidp->lval[1] = clp->lc_clientid.lval[1];
514 confirmp->lval[0] = 0; /* Ignored by client */
515 confirmp->lval[1] = 1;
518 * id and verifier match, so update the net address info
519 * and get rid of any existing callback authentication
520 * handle, so a new one will be acquired.
522 LIST_REMOVE(clp, lc_hash);
523 new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
524 new_clp->lc_expiry = nfsrv_leaseexpiry();
525 confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
526 clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
527 clp->lc_clientid.lval[0];
528 clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
529 clp->lc_clientid.lval[1];
530 new_clp->lc_delegtime = clp->lc_delegtime;
531 new_clp->lc_stateindex = clp->lc_stateindex;
532 new_clp->lc_statemaxindex = clp->lc_statemaxindex;
533 new_clp->lc_cbref = 0;
534 LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
535 LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
536 tstp->ls_clp = new_clp;
537 LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
538 LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
539 tstp->ls_clp = new_clp;
540 LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
541 LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
542 tstp->ls_clp = new_clp;
543 for (i = 0; i < nfsrv_statehashsize; i++) {
544 LIST_NEWHEAD(&new_clp->lc_stateid[i],
545 &clp->lc_stateid[i], ls_hash);
546 LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
547 tstp->ls_clp = new_clp;
549 LIST_INIT(&new_clp->lc_session);
550 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
552 nfsstatsv1.srvclients++;
553 nfsrv_openpluslock++;
556 NFSLOCKV4ROOTMUTEX();
557 nfsv4_unlock(&nfsv4rootfs_lock, 1);
558 NFSUNLOCKV4ROOTMUTEX();
560 if ((nd->nd_flag & ND_NFSV41) == 0) {
562 * Must wait until any outstanding callback on the old clp
566 while (clp->lc_cbref) {
567 clp->lc_flags |= LCL_WAKEUPWANTED;
568 (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
572 nfsrv_zapclient(clp, p);
577 NFSEXITCODE2(error, nd);
582 * Check to see if the client id exists and optionally confirm it.
585 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
586 struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
587 struct nfsrv_descript *nd, NFSPROC_T *p)
589 struct nfsclient *clp;
590 struct nfsstate *stp;
592 struct nfsclienthashhead *hp;
593 int error = 0, igotlock, doneok;
594 struct nfssessionhash *shp;
595 struct nfsdsession *sep;
597 static uint64_t next_sess = 0;
601 if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
602 opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
603 error = NFSERR_STALECLIENTID;
608 * If called with opflags == CLOPS_RENEW, the State Lock is
609 * already held. Otherwise, we need to get either that or,
610 * for the case of Confirm, lock out the nfsd threads.
612 if (opflags & CLOPS_CONFIRM) {
613 NFSLOCKV4ROOTMUTEX();
614 nfsv4_relref(&nfsv4rootfs_lock);
616 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
617 NFSV4ROOTLOCKMUTEXPTR, NULL);
620 * Create a new sessionid here, since we need to do it where
621 * there is a mutex held to serialize update of next_sess.
623 if ((nd->nd_flag & ND_NFSV41) != 0) {
624 sessid[0] = ++next_sess;
625 sessid[1] = clientid.qval;
627 NFSUNLOCKV4ROOTMUTEX();
628 } else if (opflags != CLOPS_RENEW) {
632 /* For NFSv4.1, the clp is acquired from the associated session. */
633 if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
634 opflags == CLOPS_RENEW) {
636 if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
637 shp = NFSSESSIONHASH(nd->nd_sessionid);
639 sep = nfsrv_findsession(nd->nd_sessionid);
642 NFSUNLOCKSESSION(shp);
645 hp = NFSCLIENTHASH(clientid);
646 LIST_FOREACH(clp, hp, lc_hash) {
647 if (clp->lc_clientid.lval[1] == clientid.lval[1])
652 if (opflags & CLOPS_CONFIRM)
653 error = NFSERR_STALECLIENTID;
655 error = NFSERR_EXPIRED;
656 } else if (clp->lc_flags & LCL_ADMINREVOKED) {
658 * If marked admin revoked, just return the error.
660 error = NFSERR_ADMINREVOKED;
663 if (opflags & CLOPS_CONFIRM) {
664 NFSLOCKV4ROOTMUTEX();
665 nfsv4_unlock(&nfsv4rootfs_lock, 1);
666 NFSUNLOCKV4ROOTMUTEX();
667 } else if (opflags != CLOPS_RENEW) {
674 * Perform any operations specified by the opflags.
676 if (opflags & CLOPS_CONFIRM) {
677 if (((nd->nd_flag & ND_NFSV41) != 0 &&
678 clp->lc_confirm.lval[0] != confirm.lval[0]) ||
679 ((nd->nd_flag & ND_NFSV41) == 0 &&
680 clp->lc_confirm.qval != confirm.qval))
681 error = NFSERR_STALECLIENTID;
682 else if (nfsrv_notsamecredname(nd, clp))
683 error = NFSERR_CLIDINUSE;
686 if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
689 * Hang onto the delegations (as old delegations)
690 * for an Open with CLAIM_DELEGATE_PREV unless in
691 * grace, but get rid of the rest of the state.
693 nfsrv_cleanclient(clp, p);
694 nfsrv_freedeleglist(&clp->lc_olddeleg);
695 if (nfsrv_checkgrace(nd, clp, 0)) {
696 /* In grace, so just delete delegations */
697 nfsrv_freedeleglist(&clp->lc_deleg);
699 LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
700 stp->ls_flags |= NFSLCK_OLDDELEG;
701 clp->lc_delegtime = NFSD_MONOSEC +
702 nfsrv_lease + NFSRV_LEASEDELTA;
703 LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
706 if ((nd->nd_flag & ND_NFSV41) != 0)
707 clp->lc_program = cbprogram;
709 clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
711 clp->lc_flags |= LCL_NEEDSCBNULL;
712 /* For NFSv4.1, link the session onto the client. */
714 /* Hold a reference on the xprt for a backchannel. */
715 if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
717 if (clp->lc_req.nr_client == NULL)
718 clp->lc_req.nr_client = (struct __rpc_client *)
719 clnt_bck_create(nd->nd_xprt->xp_socket,
720 cbprogram, NFSV4_CBVERS);
721 if (clp->lc_req.nr_client != NULL) {
722 SVC_ACQUIRE(nd->nd_xprt);
724 clp->lc_req.nr_client->cl_private;
725 /* Disable idle timeout. */
726 nd->nd_xprt->xp_idletimeout = 0;
727 nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
729 nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
731 NFSBCOPY(sessid, nsep->sess_sessionid,
733 NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
735 shp = NFSSESSIONHASH(nsep->sess_sessionid);
738 LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
739 LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
740 nsep->sess_clp = clp;
741 NFSUNLOCKSESSION(shp);
745 } else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
746 error = NFSERR_EXPIRED;
750 * If called by the Renew Op, we must check the principal.
752 if (!error && (opflags & CLOPS_RENEWOP)) {
753 if (nfsrv_notsamecredname(nd, clp)) {
755 for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
756 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
757 if ((stp->ls_flags & NFSLCK_OPEN) &&
758 stp->ls_uid == nd->nd_cred->cr_uid) {
765 error = NFSERR_ACCES;
767 if (!error && (clp->lc_flags & LCL_CBDOWN))
768 error = NFSERR_CBPATHDOWN;
770 if ((!error || error == NFSERR_CBPATHDOWN) &&
771 (opflags & CLOPS_RENEW)) {
772 clp->lc_expiry = nfsrv_leaseexpiry();
774 if (opflags & CLOPS_CONFIRM) {
775 NFSLOCKV4ROOTMUTEX();
776 nfsv4_unlock(&nfsv4rootfs_lock, 1);
777 NFSUNLOCKV4ROOTMUTEX();
778 } else if (opflags != CLOPS_RENEW) {
785 NFSEXITCODE2(error, nd);
790 * Perform the NFSv4.1 destroy clientid.
793 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
795 struct nfsclient *clp;
796 struct nfsclienthashhead *hp;
797 int error = 0, i, igotlock;
799 if (nfsrvboottime != clientid.lval[0]) {
800 error = NFSERR_STALECLIENTID;
804 /* Lock out other nfsd threads */
805 NFSLOCKV4ROOTMUTEX();
806 nfsv4_relref(&nfsv4rootfs_lock);
808 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
809 NFSV4ROOTLOCKMUTEXPTR, NULL);
810 } while (igotlock == 0);
811 NFSUNLOCKV4ROOTMUTEX();
813 hp = NFSCLIENTHASH(clientid);
814 LIST_FOREACH(clp, hp, lc_hash) {
815 if (clp->lc_clientid.lval[1] == clientid.lval[1])
819 NFSLOCKV4ROOTMUTEX();
820 nfsv4_unlock(&nfsv4rootfs_lock, 1);
821 NFSUNLOCKV4ROOTMUTEX();
822 /* Just return ok, since it is gone. */
827 * Free up all layouts on the clientid. Should the client return the
830 nfsrv_freelayoutlist(clientid);
832 /* Scan for state on the clientid. */
833 for (i = 0; i < nfsrv_statehashsize; i++)
834 if (!LIST_EMPTY(&clp->lc_stateid[i])) {
835 NFSLOCKV4ROOTMUTEX();
836 nfsv4_unlock(&nfsv4rootfs_lock, 1);
837 NFSUNLOCKV4ROOTMUTEX();
838 error = NFSERR_CLIENTIDBUSY;
841 if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
842 NFSLOCKV4ROOTMUTEX();
843 nfsv4_unlock(&nfsv4rootfs_lock, 1);
844 NFSUNLOCKV4ROOTMUTEX();
845 error = NFSERR_CLIENTIDBUSY;
849 /* Destroy the clientid and return ok. */
850 nfsrv_cleanclient(clp, p);
851 nfsrv_freedeleglist(&clp->lc_deleg);
852 nfsrv_freedeleglist(&clp->lc_olddeleg);
853 LIST_REMOVE(clp, lc_hash);
854 NFSLOCKV4ROOTMUTEX();
855 nfsv4_unlock(&nfsv4rootfs_lock, 1);
856 NFSUNLOCKV4ROOTMUTEX();
857 nfsrv_zapclient(clp, p);
859 NFSEXITCODE2(error, nd);
864 * Called from the new nfssvc syscall to admin revoke a clientid.
865 * Returns 0 for success, error otherwise.
868 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
870 struct nfsclient *clp = NULL;
875 * First, lock out the nfsd so that state won't change while the
876 * revocation record is being written to the stable storage restart
879 NFSLOCKV4ROOTMUTEX();
881 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
882 NFSV4ROOTLOCKMUTEXPTR, NULL);
884 NFSUNLOCKV4ROOTMUTEX();
887 * Search for a match in the client list.
890 while (i < nfsrv_clienthashsize && !gotit) {
891 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
892 if (revokep->nclid_idlen == clp->lc_idlen &&
893 !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
901 NFSLOCKV4ROOTMUTEX();
902 nfsv4_unlock(&nfsv4rootfs_lock, 0);
903 NFSUNLOCKV4ROOTMUTEX();
909 * Now, write out the revocation record
911 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
912 nfsrv_backupstable();
915 * and clear out the state, marking the clientid revoked.
917 clp->lc_flags &= ~LCL_CALLBACKSON;
918 clp->lc_flags |= LCL_ADMINREVOKED;
919 nfsrv_cleanclient(clp, p);
920 nfsrv_freedeleglist(&clp->lc_deleg);
921 nfsrv_freedeleglist(&clp->lc_olddeleg);
922 NFSLOCKV4ROOTMUTEX();
923 nfsv4_unlock(&nfsv4rootfs_lock, 0);
924 NFSUNLOCKV4ROOTMUTEX();
932 * Dump out stats for all clients. Called from nfssvc(2), that is used
936 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
938 struct nfsclient *clp;
942 * First, get a reference on the nfsv4rootfs_lock so that an
943 * exclusive lock cannot be acquired while dumping the clients.
945 NFSLOCKV4ROOTMUTEX();
946 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
947 NFSUNLOCKV4ROOTMUTEX();
950 * Rattle through the client lists until done.
952 while (i < nfsrv_clienthashsize && cnt < maxcnt) {
953 clp = LIST_FIRST(&nfsclienthash[i]);
954 while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
955 nfsrv_dumpaclient(clp, &dumpp[cnt]);
957 clp = LIST_NEXT(clp, lc_hash);
962 dumpp[cnt].ndcl_clid.nclid_idlen = 0;
964 NFSLOCKV4ROOTMUTEX();
965 nfsv4_relref(&nfsv4rootfs_lock);
966 NFSUNLOCKV4ROOTMUTEX();
970 * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
973 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
975 struct nfsstate *stp, *openstp, *lckownstp;
979 struct sockaddr_in *rin;
982 struct sockaddr_in6 *rin6;
985 dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
986 dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
987 dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
988 dumpp->ndcl_flags = clp->lc_flags;
989 dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
990 NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
991 af = clp->lc_req.nr_nam->sa_family;
992 dumpp->ndcl_addrfam = af;
996 rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
997 dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
1002 rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1003 dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1009 * Now, scan the state lists and total up the opens and locks.
1011 LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1012 dumpp->ndcl_nopenowners++;
1013 LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1014 dumpp->ndcl_nopens++;
1015 LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1016 dumpp->ndcl_nlockowners++;
1017 LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1018 dumpp->ndcl_nlocks++;
1025 * and the delegation lists.
1027 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1028 dumpp->ndcl_ndelegs++;
1030 LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1031 dumpp->ndcl_nolddelegs++;
1036 * Dump out lock stats for a file.
1039 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1042 struct nfsstate *stp;
1043 struct nfslock *lop;
1045 struct nfslockfile *lfp;
1048 struct sockaddr_in *rin;
1051 struct sockaddr_in6 *rin6;
1056 ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1058 * First, get a reference on the nfsv4rootfs_lock so that an
1059 * exclusive lock on it cannot be acquired while dumping the locks.
1061 NFSLOCKV4ROOTMUTEX();
1062 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1063 NFSUNLOCKV4ROOTMUTEX();
1066 ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1068 ldumpp[0].ndlck_clid.nclid_idlen = 0;
1070 NFSLOCKV4ROOTMUTEX();
1071 nfsv4_relref(&nfsv4rootfs_lock);
1072 NFSUNLOCKV4ROOTMUTEX();
1077 * For each open share on file, dump it out.
1079 stp = LIST_FIRST(&lfp->lf_open);
1080 while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1081 ldumpp[cnt].ndlck_flags = stp->ls_flags;
1082 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1083 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1084 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1085 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1086 ldumpp[cnt].ndlck_owner.nclid_idlen =
1087 stp->ls_openowner->ls_ownerlen;
1088 NFSBCOPY(stp->ls_openowner->ls_owner,
1089 ldumpp[cnt].ndlck_owner.nclid_id,
1090 stp->ls_openowner->ls_ownerlen);
1091 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1092 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1093 stp->ls_clp->lc_idlen);
1094 af = stp->ls_clp->lc_req.nr_nam->sa_family;
1095 ldumpp[cnt].ndlck_addrfam = af;
1099 rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1100 ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1105 rin6 = (struct sockaddr_in6 *)
1106 stp->ls_clp->lc_req.nr_nam;
1107 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1111 stp = LIST_NEXT(stp, ls_file);
1118 lop = LIST_FIRST(&lfp->lf_lock);
1119 while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1121 ldumpp[cnt].ndlck_flags = lop->lo_flags;
1122 ldumpp[cnt].ndlck_first = lop->lo_first;
1123 ldumpp[cnt].ndlck_end = lop->lo_end;
1124 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1125 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1126 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1127 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1128 ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1129 NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1131 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1132 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1133 stp->ls_clp->lc_idlen);
1134 af = stp->ls_clp->lc_req.nr_nam->sa_family;
1135 ldumpp[cnt].ndlck_addrfam = af;
1139 rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1140 ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1145 rin6 = (struct sockaddr_in6 *)
1146 stp->ls_clp->lc_req.nr_nam;
1147 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1151 lop = LIST_NEXT(lop, lo_lckfile);
1156 * and the delegations.
1158 stp = LIST_FIRST(&lfp->lf_deleg);
1159 while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1160 ldumpp[cnt].ndlck_flags = stp->ls_flags;
1161 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1162 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1163 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1164 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1165 ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1166 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1167 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1168 stp->ls_clp->lc_idlen);
1169 af = stp->ls_clp->lc_req.nr_nam->sa_family;
1170 ldumpp[cnt].ndlck_addrfam = af;
1174 rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1175 ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1180 rin6 = (struct sockaddr_in6 *)
1181 stp->ls_clp->lc_req.nr_nam;
1182 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1186 stp = LIST_NEXT(stp, ls_file);
1191 * If list isn't full, mark end of list by setting the client name
1195 ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1197 NFSLOCKV4ROOTMUTEX();
1198 nfsv4_relref(&nfsv4rootfs_lock);
1199 NFSUNLOCKV4ROOTMUTEX();
1203 * Server timer routine. It can scan any linked list, so long
1204 * as it holds the spin/mutex lock and there is no exclusive lock on
1206 * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1207 * to do this from a callout, since the spin locks work. For
1208 * Darwin, I'm not sure what will work correctly yet.)
1209 * Should be called once per second.
1212 nfsrv_servertimer(void)
1214 struct nfsclient *clp, *nclp;
1215 struct nfsstate *stp, *nstp;
1219 * Make sure nfsboottime is set. This is used by V3 as well
1220 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1221 * only used by the V4 server for leases.
1223 if (nfsboottime.tv_sec == 0)
1224 NFSSETBOOTTIME(nfsboottime);
1227 * If server hasn't started yet, just return.
1230 if (nfsrv_stablefirst.nsf_eograce == 0) {
1234 if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1235 if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1236 NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1237 nfsrv_stablefirst.nsf_flags |=
1238 (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1244 * Try and get a reference count on the nfsv4rootfs_lock so that
1245 * no nfsd thread can acquire an exclusive lock on it before this
1246 * call is done. If it is already exclusively locked, just return.
1248 NFSLOCKV4ROOTMUTEX();
1249 got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1250 NFSUNLOCKV4ROOTMUTEX();
1257 * For each client...
1259 for (i = 0; i < nfsrv_clienthashsize; i++) {
1260 clp = LIST_FIRST(&nfsclienthash[i]);
1261 while (clp != LIST_END(&nfsclienthash[i])) {
1262 nclp = LIST_NEXT(clp, lc_hash);
1263 if (!(clp->lc_flags & LCL_EXPIREIT)) {
1264 if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1265 && ((LIST_EMPTY(&clp->lc_deleg)
1266 && LIST_EMPTY(&clp->lc_open)) ||
1267 nfsrv_clients > nfsrv_clienthighwater)) ||
1268 (clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1269 (clp->lc_expiry < NFSD_MONOSEC &&
1270 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1272 * Lease has expired several nfsrv_lease times ago:
1274 * - no state is associated with it
1276 * - above high water mark for number of clients
1277 * (nfsrv_clienthighwater should be large enough
1278 * that this only occurs when clients fail to
1279 * use the same nfs_client_id4.id. Maybe somewhat
1280 * higher that the maximum number of clients that
1281 * will mount this server?)
1283 * Lease has expired a very long time ago
1285 * Lease has expired PLUS the number of opens + locks
1286 * has exceeded 90% of capacity
1288 * --> Mark for expiry. The actual expiry will be done
1289 * by an nfsd sometime soon.
1291 clp->lc_flags |= LCL_EXPIREIT;
1292 nfsrv_stablefirst.nsf_flags |=
1293 (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1296 * If there are no opens, increment no open tick cnt
1297 * If time exceeds NFSNOOPEN, mark it to be thrown away
1298 * otherwise, if there is an open, reset no open time
1299 * Hopefully, this will avoid excessive re-creation
1300 * of open owners and subsequent open confirms.
1302 stp = LIST_FIRST(&clp->lc_open);
1303 while (stp != LIST_END(&clp->lc_open)) {
1304 nstp = LIST_NEXT(stp, ls_list);
1305 if (LIST_EMPTY(&stp->ls_open)) {
1307 if (stp->ls_noopens > NFSNOOPEN ||
1308 (nfsrv_openpluslock * 2) >
1310 nfsrv_stablefirst.nsf_flags |=
1313 stp->ls_noopens = 0;
1323 NFSLOCKV4ROOTMUTEX();
1324 nfsv4_relref(&nfsv4rootfs_lock);
1325 NFSUNLOCKV4ROOTMUTEX();
1329 * The following set of functions free up the various data structures.
1332 * Clear out all open/lock state related to this nfsclient.
1333 * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1334 * there are no other active nfsd threads.
1337 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1339 struct nfsstate *stp, *nstp;
1340 struct nfsdsession *sep, *nsep;
1342 LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1343 nfsrv_freeopenowner(stp, 1, p);
1344 if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1345 LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1346 (void)nfsrv_freesession(sep, NULL);
1350 * Free a client that has been cleaned. It should also already have been
1351 * removed from the lists.
1352 * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1353 * softclock interrupts are enabled.)
1356 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1360 if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1361 (LCL_GSS | LCL_CALLBACKSON) &&
1362 (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1363 clp->lc_handlelen > 0) {
1364 clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1365 clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1366 (void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1367 NULL, 0, NULL, NULL, NULL, 0, p);
1370 newnfs_disconnect(&clp->lc_req);
1371 free(clp->lc_req.nr_nam, M_SONAME);
1372 NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1373 free(clp->lc_stateid, M_NFSDCLIENT);
1374 free(clp, M_NFSDCLIENT);
1376 nfsstatsv1.srvclients--;
1377 nfsrv_openpluslock--;
1383 * Free a list of delegation state structures.
1384 * (This function will also free all nfslockfile structures that no
1385 * longer have associated state.)
1388 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1390 struct nfsstate *stp, *nstp;
1392 LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1393 nfsrv_freedeleg(stp);
1399 * Free up a delegation.
1402 nfsrv_freedeleg(struct nfsstate *stp)
1404 struct nfslockfile *lfp;
1406 LIST_REMOVE(stp, ls_hash);
1407 LIST_REMOVE(stp, ls_list);
1408 LIST_REMOVE(stp, ls_file);
1409 if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1410 nfsrv_writedelegcnt--;
1412 if (LIST_EMPTY(&lfp->lf_open) &&
1413 LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1414 LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1415 lfp->lf_usecount == 0 &&
1416 nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1417 nfsrv_freenfslockfile(lfp);
1418 free(stp, M_NFSDSTATE);
1419 nfsstatsv1.srvdelegates--;
1420 nfsrv_openpluslock--;
1421 nfsrv_delegatecnt--;
1425 * This function frees an open owner and all associated opens.
1428 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1430 struct nfsstate *nstp, *tstp;
1432 LIST_REMOVE(stp, ls_list);
1434 * Now, free all associated opens.
1436 nstp = LIST_FIRST(&stp->ls_open);
1437 while (nstp != LIST_END(&stp->ls_open)) {
1439 nstp = LIST_NEXT(nstp, ls_list);
1440 (void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1443 nfsrvd_derefcache(stp->ls_op);
1444 free(stp, M_NFSDSTATE);
1445 nfsstatsv1.srvopenowners--;
1446 nfsrv_openpluslock--;
1450 * This function frees an open (nfsstate open structure) with all associated
1451 * lock_owners and locks. It also frees the nfslockfile structure iff there
1452 * are no other opens on the file.
1453 * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1456 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1458 struct nfsstate *nstp, *tstp;
1459 struct nfslockfile *lfp;
1462 LIST_REMOVE(stp, ls_hash);
1463 LIST_REMOVE(stp, ls_list);
1464 LIST_REMOVE(stp, ls_file);
1468 * Now, free all lockowners associated with this open.
1470 LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1471 nfsrv_freelockowner(tstp, vp, cansleep, p);
1474 * The nfslockfile is freed here if there are no locks
1475 * associated with the open.
1476 * If there are locks associated with the open, the
1477 * nfslockfile structure can be freed via nfsrv_freelockowner().
1478 * Acquire the state mutex to avoid races with calls to
1479 * nfsrv_getlockfile().
1483 if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1484 LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1485 LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1486 lfp->lf_usecount == 0 &&
1487 (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1488 nfsrv_freenfslockfile(lfp);
1494 free(stp, M_NFSDSTATE);
1495 nfsstatsv1.srvopens--;
1496 nfsrv_openpluslock--;
1501 * Frees a lockowner and all associated locks.
1504 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1508 LIST_REMOVE(stp, ls_hash);
1509 LIST_REMOVE(stp, ls_list);
1510 nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1512 nfsrvd_derefcache(stp->ls_op);
1513 free(stp, M_NFSDSTATE);
1514 nfsstatsv1.srvlockowners--;
1515 nfsrv_openpluslock--;
1519 * Free all the nfs locks on a lockowner.
1522 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1525 struct nfslock *lop, *nlop;
1526 struct nfsrollback *rlp, *nrlp;
1527 struct nfslockfile *lfp = NULL;
1530 uint64_t first, end;
1533 ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1534 lop = LIST_FIRST(&stp->ls_lock);
1535 while (lop != LIST_END(&stp->ls_lock)) {
1536 nlop = LIST_NEXT(lop, lo_lckowner);
1538 * Since all locks should be for the same file, lfp should
1543 else if (lfp != lop->lo_lfp)
1544 panic("allnfslocks");
1546 * If vp is NULL and cansleep != 0, a vnode must be acquired
1547 * from the file handle. This only occurs when called from
1548 * nfsrv_cleanclient().
1551 if (nfsrv_dolocallocks == 0)
1553 else if (vp == NULL && cansleep != 0) {
1554 tvp = nfsvno_getvp(&lfp->lf_fh);
1565 first = lop->lo_first;
1567 nfsrv_freenfslock(lop);
1568 nfsrv_localunlock(tvp, lfp, first, end, p);
1569 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1571 free(rlp, M_NFSDROLLBACK);
1572 LIST_INIT(&lfp->lf_rollback);
1574 nfsrv_freenfslock(lop);
1577 if (vp == NULL && tvp != NULL)
1582 * Free an nfslock structure.
1585 nfsrv_freenfslock(struct nfslock *lop)
1588 if (lop->lo_lckfile.le_prev != NULL) {
1589 LIST_REMOVE(lop, lo_lckfile);
1590 nfsstatsv1.srvlocks--;
1591 nfsrv_openpluslock--;
1593 LIST_REMOVE(lop, lo_lckowner);
1594 free(lop, M_NFSDLOCK);
1598 * This function frees an nfslockfile structure.
1601 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1604 LIST_REMOVE(lfp, lf_hash);
1605 free(lfp, M_NFSDLOCKFILE);
1609 * This function looks up an nfsstate structure via stateid.
1612 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1613 struct nfsstate **stpp)
1615 struct nfsstate *stp;
1616 struct nfsstatehead *hp;
1620 hp = NFSSTATEHASH(clp, *stateidp);
1621 LIST_FOREACH(stp, hp, ls_hash) {
1622 if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1628 * If no state id in list, return NFSERR_BADSTATEID.
1630 if (stp == LIST_END(hp)) {
1631 error = NFSERR_BADSTATEID;
1642 * This function gets an nfsstate structure via owner string.
1645 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1646 struct nfsstate **stpp)
1648 struct nfsstate *stp;
1651 LIST_FOREACH(stp, hp, ls_list) {
1652 if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1653 !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1661 * Lock control function called to update lock status.
1662 * Returns 0 upon success, -1 if there is no lock and the flags indicate
1663 * that one isn't to be created and an NFSERR_xxx for other errors.
1664 * The structures new_stp and new_lop are passed in as pointers that should
1665 * be set to NULL if the structure is used and shouldn't be free'd.
1666 * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1667 * never used and can safely be allocated on the stack. For all other
1668 * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1669 * in case they are used.
1672 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1673 struct nfslock **new_lopp, struct nfslockconflict *cfp,
1674 nfsquad_t clientid, nfsv4stateid_t *stateidp,
1675 __unused struct nfsexstuff *exp,
1676 struct nfsrv_descript *nd, NFSPROC_T *p)
1678 struct nfslock *lop;
1679 struct nfsstate *new_stp = *new_stpp;
1680 struct nfslock *new_lop = *new_lopp;
1681 struct nfsstate *tstp, *mystp, *nstp;
1683 struct nfslockfile *lfp;
1684 struct nfslock *other_lop = NULL;
1685 struct nfsstate *stp, *lckstp = NULL;
1686 struct nfsclient *clp = NULL;
1688 int error = 0, haslock = 0, ret, reterr;
1689 int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1691 uint64_t first, end;
1692 uint32_t lock_flags;
1694 if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1696 * Note the special cases of "all 1s" or "all 0s" stateids and
1697 * let reads with all 1s go ahead.
1699 if (new_stp->ls_stateid.seqid == 0x0 &&
1700 new_stp->ls_stateid.other[0] == 0x0 &&
1701 new_stp->ls_stateid.other[1] == 0x0 &&
1702 new_stp->ls_stateid.other[2] == 0x0)
1704 else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1705 new_stp->ls_stateid.other[0] == 0xffffffff &&
1706 new_stp->ls_stateid.other[1] == 0xffffffff &&
1707 new_stp->ls_stateid.other[2] == 0xffffffff)
1712 * Check for restart conditions (client and server).
1714 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1715 &new_stp->ls_stateid, specialid);
1720 * Check for state resource limit exceeded.
1722 if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1723 nfsrv_openpluslock > nfsrv_v4statelimit) {
1724 error = NFSERR_RESOURCE;
1729 * For the lock case, get another nfslock structure,
1730 * just in case we need it.
1731 * Malloc now, before we start sifting through the linked lists,
1732 * in case we have to wait for memory.
1735 if (new_stp->ls_flags & NFSLCK_LOCK)
1736 other_lop = malloc(sizeof (struct nfslock),
1737 M_NFSDLOCK, M_WAITOK);
1738 filestruct_locked = 0;
1743 * Get the lockfile structure for CFH now, so we can do a sanity
1744 * check against the stateid, before incrementing the seqid#, since
1745 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1746 * shouldn't be incremented for this case.
1747 * If nfsrv_getlockfile() returns -1, it means "not found", which
1748 * will be handled later.
1749 * If we are doing Lock/LockU and local locking is enabled, sleep
1750 * lock the nfslockfile structure.
1752 getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1754 if (getlckret == 0) {
1755 if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1756 nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1757 getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1760 filestruct_locked = 1;
1762 getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1765 if (getlckret != 0 && getlckret != -1)
1768 if (filestruct_locked != 0) {
1769 LIST_INIT(&lfp->lf_rollback);
1770 if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1772 * For local locking, do the advisory locking now, so
1773 * that any conflict can be detected. A failure later
1774 * can be rolled back locally. If an error is returned,
1775 * struct nfslockfile has been unlocked and any local
1776 * locking rolled back.
1779 if (vnode_unlocked == 0) {
1780 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1784 reterr = nfsrv_locallock(vp, lfp,
1785 (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1786 new_lop->lo_first, new_lop->lo_end, cfp, p);
1791 if (specialid == 0) {
1792 if (new_stp->ls_flags & NFSLCK_TEST) {
1794 * RFC 3530 does not list LockT as an op that renews a
1795 * lease, but the consensus seems to be that it is ok
1796 * for a server to do so.
1798 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1799 (nfsquad_t)((u_quad_t)0), 0, nd, p);
1802 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1803 * error returns for LockT, just go ahead and test for a lock,
1804 * since there are no locks for this client, but other locks
1805 * can conflict. (ie. same client will always be false)
1807 if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1811 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1812 (nfsquad_t)((u_quad_t)0), 0, nd, p);
1815 * Look up the stateid
1817 error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1818 new_stp->ls_flags, &stp);
1820 * do some sanity checks for an unconfirmed open or a
1821 * stateid that refers to the wrong file, for an open stateid
1823 if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1824 ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1825 (getlckret == 0 && stp->ls_lfp != lfp))){
1827 * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1828 * The only exception is using SETATTR with SIZE.
1830 if ((new_stp->ls_flags &
1831 (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1832 error = NFSERR_BADSTATEID;
1836 (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1837 getlckret == 0 && stp->ls_lfp != lfp)
1838 error = NFSERR_BADSTATEID;
1841 * If the lockowner stateid doesn't refer to the same file,
1842 * I believe that is considered ok, since some clients will
1843 * only create a single lockowner and use that for all locks
1845 * For now, log it as a diagnostic, instead of considering it
1848 if (error == 0 && (stp->ls_flags &
1849 (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1850 getlckret == 0 && stp->ls_lfp != lfp) {
1852 printf("Got a lock statid for different file open\n");
1855 error = NFSERR_BADSTATEID;
1860 if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1862 * If haslock set, we've already checked the seqid.
1865 if (stp->ls_flags & NFSLCK_OPEN)
1866 error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1867 stp->ls_openowner, new_stp->ls_op);
1869 error = NFSERR_BADSTATEID;
1872 nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1875 * I believe this should be an error, but it
1876 * isn't obvious what NFSERR_xxx would be
1877 * appropriate, so I'll use NFSERR_INVAL for now.
1879 error = NFSERR_INVAL;
1882 } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1884 * If haslock set, ditto above.
1887 if (stp->ls_flags & NFSLCK_OPEN)
1888 error = NFSERR_BADSTATEID;
1890 error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1891 stp, new_stp->ls_op);
1899 * If the seqid part of the stateid isn't the same, return
1900 * NFSERR_OLDSTATEID for cases other than I/O Ops.
1901 * For I/O Ops, only return NFSERR_OLDSTATEID if
1902 * nfsrv_returnoldstateid is set. (The consensus on the email
1903 * list was that most clients would prefer to not receive
1904 * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1905 * is what will happen, so I use the nfsrv_returnoldstateid to
1906 * allow for either server configuration.)
1908 if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1909 (((nd->nd_flag & ND_NFSV41) == 0 &&
1910 (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1911 nfsrv_returnoldstateid)) ||
1912 ((nd->nd_flag & ND_NFSV41) != 0 &&
1913 new_stp->ls_stateid.seqid != 0)))
1914 error = NFSERR_OLDSTATEID;
1919 * Now we can check for grace.
1922 error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1923 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1924 nfsrv_checkstable(clp))
1925 error = NFSERR_NOGRACE;
1927 * If we successfully Reclaimed state, note that.
1929 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1930 nfsrv_markstable(clp);
1933 * At this point, either error == NFSERR_BADSTATEID or the
1934 * seqid# has been updated, so we can return any error.
1935 * If error == 0, there may be an error in:
1936 * nd_repstat - Set by the calling function.
1937 * reterr - Set above, if getting the nfslockfile structure
1938 * or acquiring the local lock failed.
1939 * (If both of these are set, nd_repstat should probably be
1940 * returned, since that error was detected before this
1943 if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1945 if (nd->nd_repstat != 0)
1946 error = nd->nd_repstat;
1950 if (filestruct_locked != 0) {
1951 /* Roll back local locks. */
1953 if (vnode_unlocked == 0) {
1954 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1958 nfsrv_locallock_rollback(vp, lfp, p);
1960 nfsrv_unlocklf(lfp);
1967 * Check the nfsrv_getlockfile return.
1968 * Returned -1 if no structure found.
1970 if (getlckret == -1) {
1971 error = NFSERR_EXPIRED;
1973 * Called from lockt, so no lock is OK.
1975 if (new_stp->ls_flags & NFSLCK_TEST) {
1977 } else if (new_stp->ls_flags &
1978 (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1980 * Called to check for a lock, OK if the stateid is all
1981 * 1s or all 0s, but there should be an nfsstate
1983 * (ie. If there is no open, I'll assume no share
1989 error = NFSERR_BADSTATEID;
1996 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1997 * For NFSLCK_CHECK, allow a read if write access is granted,
1998 * but check for a deny. For NFSLCK_LOCK, require correct access,
1999 * which implies a conflicting deny can't exist.
2001 if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2003 * Four kinds of state id:
2004 * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2005 * - stateid for an open
2006 * - stateid for a delegation
2007 * - stateid for a lock owner
2010 if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2013 nfsrv_delaydelegtimeout(stp);
2014 } else if (stp->ls_flags & NFSLCK_OPEN) {
2017 mystp = stp->ls_openstp;
2020 * If locking or checking, require correct access
2023 if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2024 !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2025 mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2026 ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2027 (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2028 !(mystp->ls_flags & NFSLCK_READACCESS) &&
2029 nfsrv_allowreadforwriteopen == 0) ||
2030 ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2031 (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2032 !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2033 if (filestruct_locked != 0) {
2034 /* Roll back local locks. */
2036 if (vnode_unlocked == 0) {
2037 ASSERT_VOP_ELOCKED(vp,
2042 nfsrv_locallock_rollback(vp, lfp, p);
2044 nfsrv_unlocklf(lfp);
2047 error = NFSERR_OPENMODE;
2052 if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2054 * Check for a conflicting deny bit.
2056 LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2057 if (tstp != mystp) {
2058 bits = tstp->ls_flags;
2059 bits >>= NFSLCK_SHIFT;
2060 if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2061 KASSERT(vnode_unlocked == 0,
2062 ("nfsrv_lockctrl: vnode unlocked1"));
2063 ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2067 * nfsrv_clientconflict unlocks state
2068 * when it returns non-zero.
2076 error = NFSERR_PERM;
2078 error = NFSERR_OPENMODE;
2084 /* We're outta here */
2091 * For setattr, just get rid of all the Delegations for other clients.
2093 if (new_stp->ls_flags & NFSLCK_SETATTR) {
2094 KASSERT(vnode_unlocked == 0,
2095 ("nfsrv_lockctrl: vnode unlocked2"));
2096 ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2099 * nfsrv_cleandeleg() unlocks state when it
2109 if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2110 (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2111 LIST_EMPTY(&lfp->lf_deleg))) {
2118 * Check for a conflicting delegation. If one is found, call
2119 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2120 * been set yet, it will get the lock. Otherwise, it will recall
2121 * the delegation. Then, we try try again...
2122 * I currently believe the conflict algorithm to be:
2123 * For Lock Ops (Lock/LockT/LockU)
2124 * - there is a conflict iff a different client has a write delegation
2125 * For Reading (Read Op)
2126 * - there is a conflict iff a different client has a write delegation
2127 * (the specialids are always a different client)
2128 * For Writing (Write/Setattr of size)
2129 * - there is a conflict if a different client has any delegation
2130 * - there is a conflict if the same client has a read delegation
2131 * (I don't understand why this isn't allowed, but that seems to be
2132 * the current consensus?)
2134 tstp = LIST_FIRST(&lfp->lf_deleg);
2135 while (tstp != LIST_END(&lfp->lf_deleg)) {
2136 nstp = LIST_NEXT(tstp, ls_file);
2137 if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2138 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2139 (new_lop->lo_flags & NFSLCK_READ))) &&
2140 clp != tstp->ls_clp &&
2141 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2142 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2143 (new_lop->lo_flags & NFSLCK_WRITE) &&
2144 (clp != tstp->ls_clp ||
2145 (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2147 if (filestruct_locked != 0) {
2148 /* Roll back local locks. */
2150 if (vnode_unlocked == 0) {
2151 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2154 nfsrv_locallock_rollback(vp, lfp, p);
2156 nfsrv_unlocklf(lfp);
2158 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2160 if (VN_IS_DOOMED(vp))
2161 ret = NFSERR_SERVERFAULT;
2165 ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2168 * nfsrv_delegconflict unlocks state when it
2169 * returns non-zero, which it always does.
2172 free(other_lop, M_NFSDLOCK);
2182 /* Never gets here. */
2188 * Handle the unlock case by calling nfsrv_updatelock().
2189 * (Should I have done some access checking above for unlock? For now,
2190 * just let it happen.)
2192 if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2193 first = new_lop->lo_first;
2194 end = new_lop->lo_end;
2195 nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2196 stateidp->seqid = ++(stp->ls_stateid.seqid);
2197 if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2198 stateidp->seqid = stp->ls_stateid.seqid = 1;
2199 stateidp->other[0] = stp->ls_stateid.other[0];
2200 stateidp->other[1] = stp->ls_stateid.other[1];
2201 stateidp->other[2] = stp->ls_stateid.other[2];
2202 if (filestruct_locked != 0) {
2204 if (vnode_unlocked == 0) {
2205 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2209 /* Update the local locks. */
2210 nfsrv_localunlock(vp, lfp, first, end, p);
2212 nfsrv_unlocklf(lfp);
2219 * Search for a conflicting lock. A lock conflicts if:
2220 * - the lock range overlaps and
2221 * - at least one lock is a write lock and
2222 * - it is not owned by the same lock owner
2225 LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2226 if (new_lop->lo_end > lop->lo_first &&
2227 new_lop->lo_first < lop->lo_end &&
2228 (new_lop->lo_flags == NFSLCK_WRITE ||
2229 lop->lo_flags == NFSLCK_WRITE) &&
2230 lckstp != lop->lo_stp &&
2231 (clp != lop->lo_stp->ls_clp ||
2232 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2233 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2234 lckstp->ls_ownerlen))) {
2236 free(other_lop, M_NFSDLOCK);
2239 if (vnode_unlocked != 0)
2240 ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2243 ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2246 if (filestruct_locked != 0) {
2247 if (vnode_unlocked == 0) {
2248 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2251 /* Roll back local locks. */
2252 nfsrv_locallock_rollback(vp, lfp, p);
2254 nfsrv_unlocklf(lfp);
2256 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2258 if (VN_IS_DOOMED(vp)) {
2259 error = NFSERR_SERVERFAULT;
2264 * nfsrv_clientconflict() unlocks state when it
2271 * Found a conflicting lock, so record the conflict and
2274 if (cfp != NULL && ret == 0) {
2275 cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2276 cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2277 cfp->cl_first = lop->lo_first;
2278 cfp->cl_end = lop->lo_end;
2279 cfp->cl_flags = lop->lo_flags;
2280 cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2281 NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2285 error = NFSERR_PERM;
2286 else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2287 error = NFSERR_RECLAIMCONFLICT;
2288 else if (new_stp->ls_flags & NFSLCK_CHECK)
2289 error = NFSERR_LOCKED;
2291 error = NFSERR_DENIED;
2292 if (filestruct_locked != 0 && ret == 0) {
2293 /* Roll back local locks. */
2295 if (vnode_unlocked == 0) {
2296 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2300 nfsrv_locallock_rollback(vp, lfp, p);
2302 nfsrv_unlocklf(lfp);
2312 * We only get here if there was no lock that conflicted.
2314 if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2320 * We only get here when we are creating or modifying a lock.
2321 * There are two variants:
2322 * - exist_lock_owner where lock_owner exists
2323 * - open_to_lock_owner with new lock_owner
2325 first = new_lop->lo_first;
2326 end = new_lop->lo_end;
2327 lock_flags = new_lop->lo_flags;
2328 if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2329 nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2330 stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2331 if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2332 stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2333 stateidp->other[0] = lckstp->ls_stateid.other[0];
2334 stateidp->other[1] = lckstp->ls_stateid.other[1];
2335 stateidp->other[2] = lckstp->ls_stateid.other[2];
2338 * The new open_to_lock_owner case.
2339 * Link the new nfsstate into the lists.
2341 new_stp->ls_seq = new_stp->ls_opentolockseq;
2342 nfsrvd_refcache(new_stp->ls_op);
2343 stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2344 stateidp->other[0] = new_stp->ls_stateid.other[0] =
2345 clp->lc_clientid.lval[0];
2346 stateidp->other[1] = new_stp->ls_stateid.other[1] =
2347 clp->lc_clientid.lval[1];
2348 stateidp->other[2] = new_stp->ls_stateid.other[2] =
2349 nfsrv_nextstateindex(clp);
2350 new_stp->ls_clp = clp;
2351 LIST_INIT(&new_stp->ls_lock);
2352 new_stp->ls_openstp = stp;
2353 new_stp->ls_lfp = lfp;
2354 nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2356 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2358 LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2361 nfsstatsv1.srvlockowners++;
2362 nfsrv_openpluslock++;
2364 if (filestruct_locked != 0) {
2366 nfsrv_locallock_commit(lfp, lock_flags, first, end);
2368 nfsrv_unlocklf(lfp);
2374 NFSLOCKV4ROOTMUTEX();
2375 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2376 NFSUNLOCKV4ROOTMUTEX();
2378 if (vnode_unlocked != 0) {
2379 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2380 if (error == 0 && VN_IS_DOOMED(vp))
2381 error = NFSERR_SERVERFAULT;
2384 free(other_lop, M_NFSDLOCK);
2385 NFSEXITCODE2(error, nd);
2390 * Check for state errors for Open.
2391 * repstat is passed back out as an error if more critical errors
2395 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2396 struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2397 NFSPROC_T *p, int repstat)
2399 struct nfsstate *stp, *nstp;
2400 struct nfsclient *clp;
2401 struct nfsstate *ownerstp;
2402 struct nfslockfile *lfp, *new_lfp;
2403 int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2405 if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2408 * Check for restart conditions (client and server).
2410 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2411 &new_stp->ls_stateid, 0);
2416 * Check for state resource limit exceeded.
2417 * Technically this should be SMP protected, but the worst
2418 * case error is "out by one or two" on the count when it
2419 * returns NFSERR_RESOURCE and the limit is just a rather
2420 * arbitrary high water mark, so no harm is done.
2422 if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2423 error = NFSERR_RESOURCE;
2428 new_lfp = malloc(sizeof (struct nfslockfile),
2429 M_NFSDLOCKFILE, M_WAITOK);
2431 getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2435 * Get the nfsclient structure.
2437 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2438 (nfsquad_t)((u_quad_t)0), 0, nd, p);
2441 * Look up the open owner. See if it needs confirmation and
2442 * check the seq#, as required.
2445 nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2447 if (!error && ownerstp) {
2448 error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2451 * If the OpenOwner hasn't been confirmed, assume the
2452 * old one was a replay and this one is ok.
2453 * See: RFC3530 Sec. 14.2.18.
2455 if (error == NFSERR_BADSEQID &&
2456 (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2464 error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2465 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2466 nfsrv_checkstable(clp))
2467 error = NFSERR_NOGRACE;
2470 * If none of the above errors occurred, let repstat be
2473 if (repstat && !error)
2478 NFSLOCKV4ROOTMUTEX();
2479 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2480 NFSUNLOCKV4ROOTMUTEX();
2482 free(new_lfp, M_NFSDLOCKFILE);
2487 * If vp == NULL, the file doesn't exist yet, so return ok.
2488 * (This always happens on the first pass, so haslock must be 0.)
2492 free(new_lfp, M_NFSDLOCKFILE);
2497 * Get the structure for the underlying file.
2502 error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2505 free(new_lfp, M_NFSDLOCKFILE);
2509 NFSLOCKV4ROOTMUTEX();
2510 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2511 NFSUNLOCKV4ROOTMUTEX();
2517 * Search for a conflicting open/share.
2519 if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2521 * For Delegate_Cur, search for the matching Delegation,
2522 * which indicates no conflict.
2523 * An old delegation should have been recovered by the
2524 * client doing a Claim_DELEGATE_Prev, so I won't let
2525 * it match and return NFSERR_EXPIRED. Should I let it
2528 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2529 if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2530 (((nd->nd_flag & ND_NFSV41) != 0 &&
2531 stateidp->seqid == 0) ||
2532 stateidp->seqid == stp->ls_stateid.seqid) &&
2533 !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2537 if (stp == LIST_END(&lfp->lf_deleg) ||
2538 ((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2539 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2542 NFSLOCKV4ROOTMUTEX();
2543 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2544 NFSUNLOCKV4ROOTMUTEX();
2546 error = NFSERR_EXPIRED;
2552 * Check for access/deny bit conflicts. I check for the same
2553 * owner as well, in case the client didn't bother.
2555 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2556 if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2557 (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2558 ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2559 ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2560 ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2561 ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2564 * nfsrv_clientconflict() unlocks
2565 * state when it returns non-zero.
2570 error = NFSERR_PERM;
2571 else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2572 error = NFSERR_RECLAIMCONFLICT;
2574 error = NFSERR_SHAREDENIED;
2578 NFSLOCKV4ROOTMUTEX();
2579 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2580 NFSUNLOCKV4ROOTMUTEX();
2587 * Check for a conflicting delegation. If one is found, call
2588 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2589 * been set yet, it will get the lock. Otherwise, it will recall
2590 * the delegation. Then, we try try again...
2591 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2592 * isn't a conflict.)
2593 * I currently believe the conflict algorithm to be:
2594 * For Open with Read Access and Deny None
2595 * - there is a conflict iff a different client has a write delegation
2596 * For Open with other Write Access or any Deny except None
2597 * - there is a conflict if a different client has any delegation
2598 * - there is a conflict if the same client has a read delegation
2599 * (The current consensus is that this last case should be
2600 * considered a conflict since the client with a read delegation
2601 * could have done an Open with ReadAccess and WriteDeny
2602 * locally and then not have checked for the WriteDeny.)
2603 * Don't check for a Reclaim, since that will be dealt with
2604 * by nfsrv_openctrl().
2606 if (!(new_stp->ls_flags &
2607 (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2608 stp = LIST_FIRST(&lfp->lf_deleg);
2609 while (stp != LIST_END(&lfp->lf_deleg)) {
2610 nstp = LIST_NEXT(stp, ls_file);
2611 if ((readonly && stp->ls_clp != clp &&
2612 (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2613 (!readonly && (stp->ls_clp != clp ||
2614 (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2615 ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2618 * nfsrv_delegconflict() unlocks state
2619 * when it returns non-zero.
2632 NFSLOCKV4ROOTMUTEX();
2633 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2634 NFSUNLOCKV4ROOTMUTEX();
2638 NFSEXITCODE2(error, nd);
2643 * Open control function to create/update open state for an open.
2646 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2647 struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2648 nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2649 NFSPROC_T *p, u_quad_t filerev)
2651 struct nfsstate *new_stp = *new_stpp;
2652 struct nfsstate *stp, *nstp;
2653 struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2654 struct nfslockfile *lfp, *new_lfp;
2655 struct nfsclient *clp;
2656 int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2657 int readonly = 0, cbret = 1, getfhret = 0;
2658 int gotstate = 0, len = 0;
2659 u_char *clidp = NULL;
2661 if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2664 * Check for restart conditions (client and server).
2665 * (Paranoia, should have been detected by nfsrv_opencheck().)
2666 * If an error does show up, return NFSERR_EXPIRED, since the
2667 * the seqid# has already been incremented.
2669 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2670 &new_stp->ls_stateid, 0);
2672 printf("Nfsd: openctrl unexpected restart err=%d\n",
2674 error = NFSERR_EXPIRED;
2678 clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2680 new_lfp = malloc(sizeof (struct nfslockfile),
2681 M_NFSDLOCKFILE, M_WAITOK);
2682 new_open = malloc(sizeof (struct nfsstate),
2683 M_NFSDSTATE, M_WAITOK);
2684 new_deleg = malloc(sizeof (struct nfsstate),
2685 M_NFSDSTATE, M_WAITOK);
2686 getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2690 * Get the client structure. Since the linked lists could be changed
2691 * by other nfsd processes if this process does a tsleep(), one of
2692 * two things must be done.
2693 * 1 - don't tsleep()
2695 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2696 * before using the lists, since this lock stops the other
2697 * nfsd. This should only be used for rare cases, since it
2698 * essentially single threads the nfsd.
2699 * At this time, it is only done for cases where the stable
2700 * storage file must be written prior to completion of state
2703 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2704 (nfsquad_t)((u_quad_t)0), 0, nd, p);
2705 if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2708 * This happens on the first open for a client
2709 * that supports callbacks.
2713 * Although nfsrv_docallback() will sleep, clp won't
2714 * go away, since they are only removed when the
2715 * nfsv4_lock() has blocked the nfsd threads. The
2716 * fields in clp can change, but having multiple
2717 * threads do this Null callback RPC should be
2720 cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2721 NULL, 0, NULL, NULL, NULL, 0, p);
2723 clp->lc_flags &= ~LCL_NEEDSCBNULL;
2725 clp->lc_flags |= LCL_CALLBACKSON;
2729 * Look up the open owner. See if it needs confirmation and
2730 * check the seq#, as required.
2733 nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2737 printf("Nfsd: openctrl unexpected state err=%d\n",
2739 free(new_lfp, M_NFSDLOCKFILE);
2740 free(new_open, M_NFSDSTATE);
2741 free(new_deleg, M_NFSDSTATE);
2743 NFSLOCKV4ROOTMUTEX();
2744 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2745 NFSUNLOCKV4ROOTMUTEX();
2747 error = NFSERR_EXPIRED;
2751 if (new_stp->ls_flags & NFSLCK_RECLAIM)
2752 nfsrv_markstable(clp);
2755 * Get the structure for the underlying file.
2760 error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2763 free(new_lfp, M_NFSDLOCKFILE);
2766 printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2768 free(new_open, M_NFSDSTATE);
2769 free(new_deleg, M_NFSDSTATE);
2771 NFSLOCKV4ROOTMUTEX();
2772 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2773 NFSUNLOCKV4ROOTMUTEX();
2779 * Search for a conflicting open/share.
2781 if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2783 * For Delegate_Cur, search for the matching Delegation,
2784 * which indicates no conflict.
2785 * An old delegation should have been recovered by the
2786 * client doing a Claim_DELEGATE_Prev, so I won't let
2787 * it match and return NFSERR_EXPIRED. Should I let it
2790 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2791 if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2792 (((nd->nd_flag & ND_NFSV41) != 0 &&
2793 stateidp->seqid == 0) ||
2794 stateidp->seqid == stp->ls_stateid.seqid) &&
2795 !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2799 if (stp == LIST_END(&lfp->lf_deleg) ||
2800 ((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2801 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2803 printf("Nfsd openctrl unexpected expiry\n");
2804 free(new_open, M_NFSDSTATE);
2805 free(new_deleg, M_NFSDSTATE);
2807 NFSLOCKV4ROOTMUTEX();
2808 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2809 NFSUNLOCKV4ROOTMUTEX();
2811 error = NFSERR_EXPIRED;
2816 * Don't issue a Delegation, since one already exists and
2817 * delay delegation timeout, as required.
2820 nfsrv_delaydelegtimeout(stp);
2824 * Check for access/deny bit conflicts. I also check for the
2825 * same owner, since the client might not have bothered to check.
2826 * Also, note an open for the same file and owner, if found,
2827 * which is all we do here for Delegate_Cur, since conflict
2828 * checking is already done.
2830 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2831 if (ownerstp && stp->ls_openowner == ownerstp)
2833 if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2835 * If another client has the file open, the only
2836 * delegation that can be issued is a Read delegation
2837 * and only if it is a Read open with Deny none.
2839 if (clp != stp->ls_clp) {
2840 if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2846 if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2847 ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2848 ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2849 ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2850 ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2853 * nfsrv_clientconflict() unlocks state
2854 * when it returns non-zero.
2856 free(new_open, M_NFSDSTATE);
2857 free(new_deleg, M_NFSDSTATE);
2862 error = NFSERR_PERM;
2863 else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2864 error = NFSERR_RECLAIMCONFLICT;
2866 error = NFSERR_SHAREDENIED;
2870 NFSLOCKV4ROOTMUTEX();
2871 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2872 NFSUNLOCKV4ROOTMUTEX();
2874 free(new_open, M_NFSDSTATE);
2875 free(new_deleg, M_NFSDSTATE);
2876 printf("nfsd openctrl unexpected client cnfl\n");
2883 * Check for a conflicting delegation. If one is found, call
2884 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2885 * been set yet, it will get the lock. Otherwise, it will recall
2886 * the delegation. Then, we try try again...
2887 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2888 * isn't a conflict.)
2889 * I currently believe the conflict algorithm to be:
2890 * For Open with Read Access and Deny None
2891 * - there is a conflict iff a different client has a write delegation
2892 * For Open with other Write Access or any Deny except None
2893 * - there is a conflict if a different client has any delegation
2894 * - there is a conflict if the same client has a read delegation
2895 * (The current consensus is that this last case should be
2896 * considered a conflict since the client with a read delegation
2897 * could have done an Open with ReadAccess and WriteDeny
2898 * locally and then not have checked for the WriteDeny.)
2900 if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2901 stp = LIST_FIRST(&lfp->lf_deleg);
2902 while (stp != LIST_END(&lfp->lf_deleg)) {
2903 nstp = LIST_NEXT(stp, ls_file);
2904 if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2908 if ((readonly && stp->ls_clp != clp &&
2909 (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2910 (!readonly && (stp->ls_clp != clp ||
2911 (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2912 if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2915 ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2918 * nfsrv_delegconflict() unlocks state
2919 * when it returns non-zero.
2921 printf("Nfsd openctrl unexpected deleg cnfl\n");
2922 free(new_open, M_NFSDSTATE);
2923 free(new_deleg, M_NFSDSTATE);
2938 * We only get here if there was no open that conflicted.
2939 * If an open for the owner exists, or in the access/deny bits.
2940 * Otherwise it is a new open. If the open_owner hasn't been
2941 * confirmed, replace the open with the new one needing confirmation,
2942 * otherwise add the open.
2944 if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2946 * Handle NFSLCK_DELEGPREV by searching the old delegations for
2947 * a match. If found, just move the old delegation to the current
2948 * delegation list and issue open. If not found, return
2951 LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2952 if (stp->ls_lfp == lfp) {
2954 if (stp->ls_clp != clp)
2955 panic("olddeleg clp");
2956 LIST_REMOVE(stp, ls_list);
2957 LIST_REMOVE(stp, ls_hash);
2958 stp->ls_flags &= ~NFSLCK_OLDDELEG;
2959 stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2960 stp->ls_stateid.other[0] = delegstateidp->other[0] =
2961 clp->lc_clientid.lval[0];
2962 stp->ls_stateid.other[1] = delegstateidp->other[1] =
2963 clp->lc_clientid.lval[1];
2964 stp->ls_stateid.other[2] = delegstateidp->other[2] =
2965 nfsrv_nextstateindex(clp);
2966 stp->ls_compref = nd->nd_compref;
2967 LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2968 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2969 stp->ls_stateid), stp, ls_hash);
2970 if (stp->ls_flags & NFSLCK_DELEGWRITE)
2971 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2973 *rflagsp |= NFSV4OPEN_READDELEGATE;
2974 clp->lc_delegtime = NFSD_MONOSEC +
2975 nfsrv_lease + NFSRV_LEASEDELTA;
2978 * Now, do the associated open.
2980 new_open->ls_stateid.seqid = 1;
2981 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2982 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2983 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2984 new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2986 if (stp->ls_flags & NFSLCK_DELEGWRITE)
2987 new_open->ls_flags |= (NFSLCK_READACCESS |
2988 NFSLCK_WRITEACCESS);
2990 new_open->ls_flags |= NFSLCK_READACCESS;
2991 new_open->ls_uid = new_stp->ls_uid;
2992 new_open->ls_lfp = lfp;
2993 new_open->ls_clp = clp;
2994 LIST_INIT(&new_open->ls_open);
2995 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2996 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2999 * and handle the open owner
3002 new_open->ls_openowner = ownerstp;
3003 LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3005 new_open->ls_openowner = new_stp;
3006 new_stp->ls_flags = 0;
3007 nfsrvd_refcache(new_stp->ls_op);
3008 new_stp->ls_noopens = 0;
3009 LIST_INIT(&new_stp->ls_open);
3010 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3011 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3013 nfsstatsv1.srvopenowners++;
3014 nfsrv_openpluslock++;
3018 nfsstatsv1.srvopens++;
3019 nfsrv_openpluslock++;
3023 if (stp == LIST_END(&clp->lc_olddeleg))
3024 error = NFSERR_EXPIRED;
3025 } else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3027 * Scan to see that no delegation for this client and file
3028 * doesn't already exist.
3029 * There also shouldn't yet be an Open for this file and
3032 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3033 if (stp->ls_clp == clp)
3036 if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3038 * This is the Claim_Previous case with a delegation
3039 * type != Delegate_None.
3042 * First, add the delegation. (Although we must issue the
3043 * delegation, we can also ask for an immediate return.)
3045 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3046 new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3047 clp->lc_clientid.lval[0];
3048 new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3049 clp->lc_clientid.lval[1];
3050 new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3051 nfsrv_nextstateindex(clp);
3052 if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3053 new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3054 NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3055 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3056 nfsrv_writedelegcnt++;
3058 new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3060 *rflagsp |= NFSV4OPEN_READDELEGATE;
3062 new_deleg->ls_uid = new_stp->ls_uid;
3063 new_deleg->ls_lfp = lfp;
3064 new_deleg->ls_clp = clp;
3065 new_deleg->ls_filerev = filerev;
3066 new_deleg->ls_compref = nd->nd_compref;
3067 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3068 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3069 new_deleg->ls_stateid), new_deleg, ls_hash);
3070 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3072 if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3073 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3075 NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3076 !NFSVNO_DELEGOK(vp))
3077 *rflagsp |= NFSV4OPEN_RECALL;
3078 nfsstatsv1.srvdelegates++;
3079 nfsrv_openpluslock++;
3080 nfsrv_delegatecnt++;
3083 * Now, do the associated open.
3085 new_open->ls_stateid.seqid = 1;
3086 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3087 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3088 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3089 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3091 if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3092 new_open->ls_flags |= (NFSLCK_READACCESS |
3093 NFSLCK_WRITEACCESS);
3095 new_open->ls_flags |= NFSLCK_READACCESS;
3096 new_open->ls_uid = new_stp->ls_uid;
3097 new_open->ls_lfp = lfp;
3098 new_open->ls_clp = clp;
3099 LIST_INIT(&new_open->ls_open);
3100 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3101 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3104 * and handle the open owner
3107 new_open->ls_openowner = ownerstp;
3108 LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3110 new_open->ls_openowner = new_stp;
3111 new_stp->ls_flags = 0;
3112 nfsrvd_refcache(new_stp->ls_op);
3113 new_stp->ls_noopens = 0;
3114 LIST_INIT(&new_stp->ls_open);
3115 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3116 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3118 nfsstatsv1.srvopenowners++;
3119 nfsrv_openpluslock++;
3123 nfsstatsv1.srvopens++;
3124 nfsrv_openpluslock++;
3126 error = NFSERR_RECLAIMCONFLICT;
3128 } else if (ownerstp) {
3129 if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3130 /* Replace the open */
3131 if (ownerstp->ls_op)
3132 nfsrvd_derefcache(ownerstp->ls_op);
3133 ownerstp->ls_op = new_stp->ls_op;
3134 nfsrvd_refcache(ownerstp->ls_op);
3135 ownerstp->ls_seq = new_stp->ls_seq;
3136 *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3137 stp = LIST_FIRST(&ownerstp->ls_open);
3138 stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3140 stp->ls_stateid.seqid = 1;
3141 stp->ls_uid = new_stp->ls_uid;
3142 if (lfp != stp->ls_lfp) {
3143 LIST_REMOVE(stp, ls_file);
3144 LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3148 } else if (openstp) {
3149 openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3150 openstp->ls_stateid.seqid++;
3151 if ((nd->nd_flag & ND_NFSV41) != 0 &&
3152 openstp->ls_stateid.seqid == 0)
3153 openstp->ls_stateid.seqid = 1;
3156 * This is where we can choose to issue a delegation.
3158 if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3159 *rflagsp |= NFSV4OPEN_WDNOTWANTED;
3160 else if (nfsrv_issuedelegs == 0)
3161 *rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3162 else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3163 *rflagsp |= NFSV4OPEN_WDRESOURCE;
3164 else if (delegate == 0 || writedeleg == 0 ||
3165 NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3166 nfsrv_writedelegifpos == 0) ||
3167 !NFSVNO_DELEGOK(vp) ||
3168 (new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3169 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3171 *rflagsp |= NFSV4OPEN_WDCONTENTION;
3173 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3174 new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3175 = clp->lc_clientid.lval[0];
3176 new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3177 = clp->lc_clientid.lval[1];
3178 new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3179 = nfsrv_nextstateindex(clp);
3180 new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3181 NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3182 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3183 new_deleg->ls_uid = new_stp->ls_uid;
3184 new_deleg->ls_lfp = lfp;
3185 new_deleg->ls_clp = clp;
3186 new_deleg->ls_filerev = filerev;
3187 new_deleg->ls_compref = nd->nd_compref;
3188 nfsrv_writedelegcnt++;
3189 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3190 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3191 new_deleg->ls_stateid), new_deleg, ls_hash);
3192 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3194 nfsstatsv1.srvdelegates++;
3195 nfsrv_openpluslock++;
3196 nfsrv_delegatecnt++;
3199 new_open->ls_stateid.seqid = 1;
3200 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3201 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3202 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3203 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3205 new_open->ls_uid = new_stp->ls_uid;
3206 new_open->ls_openowner = ownerstp;
3207 new_open->ls_lfp = lfp;
3208 new_open->ls_clp = clp;
3209 LIST_INIT(&new_open->ls_open);
3210 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3211 LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3212 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3216 nfsstatsv1.srvopens++;
3217 nfsrv_openpluslock++;
3220 * This is where we can choose to issue a delegation.
3222 if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3223 *rflagsp |= NFSV4OPEN_WDNOTWANTED;
3224 else if (nfsrv_issuedelegs == 0)
3225 *rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3226 else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3227 *rflagsp |= NFSV4OPEN_WDRESOURCE;
3228 else if (delegate == 0 || (writedeleg == 0 &&
3229 readonly == 0) || !NFSVNO_DELEGOK(vp) ||
3230 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3232 *rflagsp |= NFSV4OPEN_WDCONTENTION;
3234 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3235 new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3236 = clp->lc_clientid.lval[0];
3237 new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3238 = clp->lc_clientid.lval[1];
3239 new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3240 = nfsrv_nextstateindex(clp);
3241 if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3242 (nfsrv_writedelegifpos || !readonly) &&
3243 (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3244 new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3245 NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3246 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3247 nfsrv_writedelegcnt++;
3249 new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3251 *rflagsp |= NFSV4OPEN_READDELEGATE;
3253 new_deleg->ls_uid = new_stp->ls_uid;
3254 new_deleg->ls_lfp = lfp;
3255 new_deleg->ls_clp = clp;
3256 new_deleg->ls_filerev = filerev;
3257 new_deleg->ls_compref = nd->nd_compref;
3258 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3259 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3260 new_deleg->ls_stateid), new_deleg, ls_hash);
3261 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3263 nfsstatsv1.srvdelegates++;
3264 nfsrv_openpluslock++;
3265 nfsrv_delegatecnt++;
3270 * New owner case. Start the open_owner sequence with a
3271 * Needs confirmation (unless a reclaim) and hang the
3274 new_open->ls_stateid.seqid = 1;
3275 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3276 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3277 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3278 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3280 new_open->ls_uid = new_stp->ls_uid;
3281 LIST_INIT(&new_open->ls_open);
3282 new_open->ls_openowner = new_stp;
3283 new_open->ls_lfp = lfp;
3284 new_open->ls_clp = clp;
3285 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3286 if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3287 new_stp->ls_flags = 0;
3288 } else if ((nd->nd_flag & ND_NFSV41) != 0) {
3289 /* NFSv4.1 never needs confirmation. */
3290 new_stp->ls_flags = 0;
3293 * This is where we can choose to issue a delegation.
3295 if (delegate && nfsrv_issuedelegs &&
3296 (writedeleg || readonly) &&
3297 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3299 !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3300 NFSVNO_DELEGOK(vp) &&
3301 ((nd->nd_flag & ND_NFSV41) == 0 ||
3302 (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3303 new_deleg->ls_stateid.seqid =
3304 delegstateidp->seqid = 1;
3305 new_deleg->ls_stateid.other[0] =
3306 delegstateidp->other[0]
3307 = clp->lc_clientid.lval[0];
3308 new_deleg->ls_stateid.other[1] =
3309 delegstateidp->other[1]
3310 = clp->lc_clientid.lval[1];
3311 new_deleg->ls_stateid.other[2] =
3312 delegstateidp->other[2]
3313 = nfsrv_nextstateindex(clp);
3314 if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3315 (nfsrv_writedelegifpos || !readonly) &&
3316 ((nd->nd_flag & ND_NFSV41) == 0 ||
3317 (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3319 new_deleg->ls_flags =
3320 (NFSLCK_DELEGWRITE |
3322 NFSLCK_WRITEACCESS);
3323 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3324 nfsrv_writedelegcnt++;
3326 new_deleg->ls_flags =
3329 *rflagsp |= NFSV4OPEN_READDELEGATE;
3331 new_deleg->ls_uid = new_stp->ls_uid;
3332 new_deleg->ls_lfp = lfp;
3333 new_deleg->ls_clp = clp;
3334 new_deleg->ls_filerev = filerev;
3335 new_deleg->ls_compref = nd->nd_compref;
3336 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3338 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3339 new_deleg->ls_stateid), new_deleg, ls_hash);
3340 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3343 nfsstatsv1.srvdelegates++;
3344 nfsrv_openpluslock++;
3345 nfsrv_delegatecnt++;
3348 * Since NFSv4.1 never does an OpenConfirm, the first
3349 * open state will be acquired here.
3351 if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3352 clp->lc_flags |= LCL_STAMPEDSTABLE;
3353 len = clp->lc_idlen;
3354 NFSBCOPY(clp->lc_id, clidp, len);
3358 *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3359 new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3361 nfsrvd_refcache(new_stp->ls_op);
3362 new_stp->ls_noopens = 0;
3363 LIST_INIT(&new_stp->ls_open);
3364 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3365 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3366 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3371 nfsstatsv1.srvopens++;
3372 nfsrv_openpluslock++;
3373 nfsstatsv1.srvopenowners++;
3374 nfsrv_openpluslock++;
3377 stateidp->seqid = openstp->ls_stateid.seqid;
3378 stateidp->other[0] = openstp->ls_stateid.other[0];
3379 stateidp->other[1] = openstp->ls_stateid.other[1];
3380 stateidp->other[2] = openstp->ls_stateid.other[2];
3384 NFSLOCKV4ROOTMUTEX();
3385 nfsv4_unlock(&nfsv4rootfs_lock, 1);
3386 NFSUNLOCKV4ROOTMUTEX();
3389 free(new_open, M_NFSDSTATE);
3391 free(new_deleg, M_NFSDSTATE);
3394 * If the NFSv4.1 client just acquired its first open, write a timestamp
3395 * to the stable storage file.
3397 if (gotstate != 0) {
3398 nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3399 nfsrv_backupstable();
3403 free(clidp, M_TEMP);
3404 NFSEXITCODE2(error, nd);
3409 * Open update. Does the confirm, downgrade and close.
3412 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3413 nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3414 int *retwriteaccessp)
3416 struct nfsstate *stp;
3417 struct nfsclient *clp;
3418 struct nfslockfile *lfp;
3420 int error = 0, gotstate = 0, len = 0;
3421 u_char *clidp = NULL;
3424 * Check for restart conditions (client and server).
3426 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3427 &new_stp->ls_stateid, 0);
3431 clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3434 * Get the open structure via clientid and stateid.
3436 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3437 (nfsquad_t)((u_quad_t)0), 0, nd, p);
3439 error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3440 new_stp->ls_flags, &stp);
3443 * Sanity check the open.
3445 if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3446 (!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3447 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3448 ((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3449 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3450 error = NFSERR_BADSTATEID;
3453 error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3454 stp->ls_openowner, new_stp->ls_op);
3455 if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3456 (((nd->nd_flag & ND_NFSV41) == 0 &&
3457 !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3458 ((nd->nd_flag & ND_NFSV41) != 0 &&
3459 new_stp->ls_stateid.seqid != 0)))
3460 error = NFSERR_OLDSTATEID;
3461 if (!error && vnode_vtype(vp) != VREG) {
3462 if (vnode_vtype(vp) == VDIR)
3463 error = NFSERR_ISDIR;
3465 error = NFSERR_INVAL;
3470 * If a client tries to confirm an Open with a bad
3471 * seqid# and there are no byte range locks or other Opens
3472 * on the openowner, just throw it away, so the next use of the
3473 * openowner will start a fresh seq#.
3475 if (error == NFSERR_BADSEQID &&
3476 (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3477 nfsrv_nootherstate(stp))
3478 nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3484 * Set the return stateid.
3486 stateidp->seqid = stp->ls_stateid.seqid + 1;
3487 if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3488 stateidp->seqid = 1;
3489 stateidp->other[0] = stp->ls_stateid.other[0];
3490 stateidp->other[1] = stp->ls_stateid.other[1];
3491 stateidp->other[2] = stp->ls_stateid.other[2];
3493 * Now, handle the three cases.
3495 if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3497 * If the open doesn't need confirmation, it seems to me that
3498 * there is a client error, but I'll just log it and keep going?
3500 if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3501 printf("Nfsv4d: stray open confirm\n");
3502 stp->ls_openowner->ls_flags = 0;
3503 stp->ls_stateid.seqid++;
3504 if ((nd->nd_flag & ND_NFSV41) != 0 &&
3505 stp->ls_stateid.seqid == 0)
3506 stp->ls_stateid.seqid = 1;
3507 if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3508 clp->lc_flags |= LCL_STAMPEDSTABLE;
3509 len = clp->lc_idlen;
3510 NFSBCOPY(clp->lc_id, clidp, len);
3514 } else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3516 if (retwriteaccessp != NULL) {
3517 if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3518 *retwriteaccessp = 1;
3520 *retwriteaccessp = 0;
3522 if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3523 /* Get the lf lock */
3526 ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3528 if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3530 nfsrv_unlocklf(lfp);
3533 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3535 (void) nfsrv_freeopen(stp, NULL, 0, p);
3540 * Update the share bits, making sure that the new set are a
3541 * subset of the old ones.
3543 bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3544 if (~(stp->ls_flags) & bits) {
3546 error = NFSERR_INVAL;
3549 stp->ls_flags = (bits | NFSLCK_OPEN);
3550 stp->ls_stateid.seqid++;
3551 if ((nd->nd_flag & ND_NFSV41) != 0 &&
3552 stp->ls_stateid.seqid == 0)
3553 stp->ls_stateid.seqid = 1;
3558 * If the client just confirmed its first open, write a timestamp
3559 * to the stable storage file.
3561 if (gotstate != 0) {
3562 nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3563 nfsrv_backupstable();
3567 free(clidp, M_TEMP);
3568 NFSEXITCODE2(error, nd);
3573 * Delegation update. Does the purge and return.
3576 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3577 nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3578 NFSPROC_T *p, int *retwriteaccessp)
3580 struct nfsstate *stp;
3581 struct nfsclient *clp;
3586 * Do a sanity check against the file handle for DelegReturn.
3589 error = nfsvno_getfh(vp, &fh, p);
3594 * Check for restart conditions (client and server).
3596 if (op == NFSV4OP_DELEGRETURN)
3597 error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3600 error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3605 * Get the open structure via clientid and stateid.
3608 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3609 (nfsquad_t)((u_quad_t)0), 0, nd, p);
3611 if (error == NFSERR_CBPATHDOWN)
3613 if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3614 error = NFSERR_STALESTATEID;
3616 if (!error && op == NFSV4OP_DELEGRETURN) {
3617 error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3618 if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3619 ((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3620 error = NFSERR_OLDSTATEID;
3623 * NFSERR_EXPIRED means that the state has gone away,
3624 * so Delegations have been purged. Just return ok.
3626 if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3636 if (op == NFSV4OP_DELEGRETURN) {
3637 if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3638 sizeof (fhandle_t))) {
3640 error = NFSERR_BADSTATEID;
3643 if (retwriteaccessp != NULL) {
3644 if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3645 *retwriteaccessp = 1;
3647 *retwriteaccessp = 0;
3649 nfsrv_freedeleg(stp);
3651 nfsrv_freedeleglist(&clp->lc_olddeleg);
3662 * Release lock owner.
3665 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3668 struct nfsstate *stp, *nstp, *openstp, *ownstp;
3669 struct nfsclient *clp;
3673 * Check for restart conditions (client and server).
3675 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3676 &new_stp->ls_stateid, 0);
3682 * Get the lock owner by name.
3684 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3685 (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3690 LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3691 LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3692 stp = LIST_FIRST(&openstp->ls_open);
3693 while (stp != LIST_END(&openstp->ls_open)) {
3694 nstp = LIST_NEXT(stp, ls_list);
3696 * If the owner matches, check for locks and
3697 * then free or return an error.
3699 if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3700 !NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3702 if (LIST_EMPTY(&stp->ls_lock)) {
3703 nfsrv_freelockowner(stp, NULL, 0, p);
3706 error = NFSERR_LOCKSHELD;
3722 * Get the file handle for a lock structure.
3725 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3726 fhandle_t *nfhp, NFSPROC_T *p)
3728 fhandle_t *fhp = NULL;
3732 * For lock, use the new nfslock structure, otherwise just
3733 * a fhandle_t on the stack.
3735 if (flags & NFSLCK_OPEN) {
3736 KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3737 fhp = &new_lfp->lf_fh;
3741 panic("nfsrv_getlockfh");
3743 error = nfsvno_getfh(vp, fhp, p);
3749 * Get an nfs lock structure. Allocate one, as required, and return a
3751 * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3754 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3755 struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3757 struct nfslockfile *lfp;
3758 fhandle_t *fhp = NULL, *tfhp;
3759 struct nfslockhashhead *hp;
3760 struct nfslockfile *new_lfp = NULL;
3763 * For lock, use the new nfslock structure, otherwise just
3764 * a fhandle_t on the stack.
3766 if (flags & NFSLCK_OPEN) {
3767 new_lfp = *new_lfpp;
3768 fhp = &new_lfp->lf_fh;
3772 panic("nfsrv_getlockfile");
3775 hp = NFSLOCKHASH(fhp);
3776 LIST_FOREACH(lfp, hp, lf_hash) {
3778 if (NFSVNO_CMPFH(fhp, tfhp)) {
3785 if (!(flags & NFSLCK_OPEN))
3789 * No match, so chain the new one into the list.
3791 LIST_INIT(&new_lfp->lf_open);
3792 LIST_INIT(&new_lfp->lf_lock);
3793 LIST_INIT(&new_lfp->lf_deleg);
3794 LIST_INIT(&new_lfp->lf_locallock);
3795 LIST_INIT(&new_lfp->lf_rollback);
3796 new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3797 new_lfp->lf_locallock_lck.nfslock_lock = 0;
3798 new_lfp->lf_usecount = 0;
3799 LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3806 * This function adds a nfslock lock structure to the list for the associated
3807 * nfsstate and nfslockfile structures. It will be inserted after the
3808 * entry pointed at by insert_lop.
3811 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3812 struct nfsstate *stp, struct nfslockfile *lfp)
3814 struct nfslock *lop, *nlop;
3816 new_lop->lo_stp = stp;
3817 new_lop->lo_lfp = lfp;
3820 /* Insert in increasing lo_first order */
3821 lop = LIST_FIRST(&lfp->lf_lock);
3822 if (lop == LIST_END(&lfp->lf_lock) ||
3823 new_lop->lo_first <= lop->lo_first) {
3824 LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3826 nlop = LIST_NEXT(lop, lo_lckfile);
3827 while (nlop != LIST_END(&lfp->lf_lock) &&
3828 nlop->lo_first < new_lop->lo_first) {
3830 nlop = LIST_NEXT(lop, lo_lckfile);
3832 LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3835 new_lop->lo_lckfile.le_prev = NULL; /* list not used */
3839 * Insert after insert_lop, which is overloaded as stp or lfp for
3842 if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3843 LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3844 else if ((struct nfsstate *)insert_lop == stp)
3845 LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3847 LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3849 nfsstatsv1.srvlocks++;
3850 nfsrv_openpluslock++;
3855 * This function updates the locking for a lock owner and given file. It
3856 * maintains a list of lock ranges ordered on increasing file offset that
3857 * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3858 * It always adds new_lop to the list and sometimes uses the one pointed
3862 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3863 struct nfslock **other_lopp, struct nfslockfile *lfp)
3865 struct nfslock *new_lop = *new_lopp;
3866 struct nfslock *lop, *tlop, *ilop;
3867 struct nfslock *other_lop = *other_lopp;
3868 int unlock = 0, myfile = 0;
3872 * Work down the list until the lock is merged.
3874 if (new_lop->lo_flags & NFSLCK_UNLOCK)
3877 ilop = (struct nfslock *)stp;
3878 lop = LIST_FIRST(&stp->ls_lock);
3880 ilop = (struct nfslock *)lfp;
3881 lop = LIST_FIRST(&lfp->lf_locallock);
3883 while (lop != NULL) {
3885 * Only check locks for this file that aren't before the start of
3888 if (lop->lo_lfp == lfp) {
3890 if (lop->lo_end >= new_lop->lo_first) {
3891 if (new_lop->lo_end < lop->lo_first) {
3893 * If the new lock ends before the start of the
3894 * current lock's range, no merge, just insert
3899 if (new_lop->lo_flags == lop->lo_flags ||
3900 (new_lop->lo_first <= lop->lo_first &&
3901 new_lop->lo_end >= lop->lo_end)) {
3903 * This lock can be absorbed by the new lock/unlock.
3904 * This happens when it covers the entire range
3905 * of the old lock or is contiguous
3906 * with the old lock and is of the same type or an
3909 if (lop->lo_first < new_lop->lo_first)
3910 new_lop->lo_first = lop->lo_first;
3911 if (lop->lo_end > new_lop->lo_end)
3912 new_lop->lo_end = lop->lo_end;
3914 lop = LIST_NEXT(lop, lo_lckowner);
3915 nfsrv_freenfslock(tlop);
3920 * All these cases are for contiguous locks that are not the
3921 * same type, so they can't be merged.
3923 if (new_lop->lo_first <= lop->lo_first) {
3925 * This case is where the new lock overlaps with the
3926 * first part of the old lock. Move the start of the
3927 * old lock to just past the end of the new lock. The
3928 * new lock will be inserted in front of the old, since
3929 * ilop hasn't been updated. (We are done now.)
3931 lop->lo_first = new_lop->lo_end;
3934 if (new_lop->lo_end >= lop->lo_end) {
3936 * This case is where the new lock overlaps with the
3937 * end of the old lock's range. Move the old lock's
3938 * end to just before the new lock's first and insert
3939 * the new lock after the old lock.
3940 * Might not be done yet, since the new lock could
3941 * overlap further locks with higher ranges.
3943 lop->lo_end = new_lop->lo_first;
3945 lop = LIST_NEXT(lop, lo_lckowner);
3949 * The final case is where the new lock's range is in the
3950 * middle of the current lock's and splits the current lock
3951 * up. Use *other_lopp to handle the second part of the
3952 * split old lock range. (We are done now.)
3953 * For unlock, we use new_lop as other_lop and tmp, since
3954 * other_lop and new_lop are the same for this case.
3955 * We noted the unlock case above, so we don't need
3956 * new_lop->lo_flags any longer.
3958 tmp = new_lop->lo_first;
3959 if (other_lop == NULL) {
3961 panic("nfsd srv update unlock");
3962 other_lop = new_lop;
3965 other_lop->lo_first = new_lop->lo_end;
3966 other_lop->lo_end = lop->lo_end;
3967 other_lop->lo_flags = lop->lo_flags;
3968 other_lop->lo_stp = stp;
3969 other_lop->lo_lfp = lfp;
3971 nfsrv_insertlock(other_lop, lop, stp, lfp);
3978 lop = LIST_NEXT(lop, lo_lckowner);
3979 if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3984 * Insert the new lock in the list at the appropriate place.
3987 nfsrv_insertlock(new_lop, ilop, stp, lfp);
3993 * This function handles sequencing of locks, etc.
3994 * It returns an error that indicates what the caller should do.
3997 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3998 struct nfsstate *stp, struct nfsrvcache *op)
4002 if ((nd->nd_flag & ND_NFSV41) != 0)
4003 /* NFSv4.1 ignores the open_seqid and lock_seqid. */
4005 if (op != nd->nd_rp)
4006 panic("nfsrvstate checkseqid");
4007 if (!(op->rc_flag & RC_INPROG))
4008 panic("nfsrvstate not inprog");
4009 if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4010 printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4011 panic("nfsrvstate op refcnt");
4013 if ((stp->ls_seq + 1) == seqid) {
4015 nfsrvd_derefcache(stp->ls_op);
4017 nfsrvd_refcache(op);
4018 stp->ls_seq = seqid;
4020 } else if (stp->ls_seq == seqid && stp->ls_op &&
4021 op->rc_xid == stp->ls_op->rc_xid &&
4022 op->rc_refcnt == 0 &&
4023 op->rc_reqlen == stp->ls_op->rc_reqlen &&
4024 op->rc_cksum == stp->ls_op->rc_cksum) {
4025 if (stp->ls_op->rc_flag & RC_INPROG) {
4026 error = NFSERR_DONTREPLY;
4029 nd->nd_rp = stp->ls_op;
4030 nd->nd_rp->rc_flag |= RC_INPROG;
4031 nfsrvd_delcache(op);
4032 error = NFSERR_REPLYFROMCACHE;
4035 error = NFSERR_BADSEQID;
4038 NFSEXITCODE2(error, nd);
4043 * Get the client ip address for callbacks. If the strings can't be parsed,
4044 * just set lc_program to 0 to indicate no callbacks are possible.
4045 * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
4046 * the address to the client's transport address. This won't be used
4047 * for callbacks, but can be printed out by nfsstats for info.)
4048 * Return error if the xdr can't be parsed, 0 otherwise.
4051 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4055 int i, j, maxalen = 0, minalen = 0;
4058 struct sockaddr_in *rin = NULL, *sin;
4061 struct sockaddr_in6 *rin6 = NULL, *sin6;
4064 int error = 0, cantparse = 0;
4074 /* 8 is the maximum length of the port# string. */
4075 addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
4076 clp->lc_req.nr_client = NULL;
4077 clp->lc_req.nr_lock = 0;
4079 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4080 i = fxdr_unsigned(int, *tl);
4081 if (i >= 3 && i <= 4) {
4082 error = nfsrv_mtostr(nd, addr, i);
4086 if (!strcmp(addr, "tcp")) {
4087 clp->lc_flags |= LCL_TCPCALLBACK;
4088 clp->lc_req.nr_sotype = SOCK_STREAM;
4089 clp->lc_req.nr_soproto = IPPROTO_TCP;
4091 } else if (!strcmp(addr, "udp")) {
4092 clp->lc_req.nr_sotype = SOCK_DGRAM;
4093 clp->lc_req.nr_soproto = IPPROTO_UDP;
4098 if (af == AF_UNSPEC) {
4099 if (!strcmp(addr, "tcp6")) {
4100 clp->lc_flags |= LCL_TCPCALLBACK;
4101 clp->lc_req.nr_sotype = SOCK_STREAM;
4102 clp->lc_req.nr_soproto = IPPROTO_TCP;
4104 } else if (!strcmp(addr, "udp6")) {
4105 clp->lc_req.nr_sotype = SOCK_DGRAM;
4106 clp->lc_req.nr_soproto = IPPROTO_UDP;
4111 if (af == AF_UNSPEC) {
4117 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4123 * The caller has allocated clp->lc_req.nr_nam to be large enough
4124 * for either AF_INET or AF_INET6 and zeroed out the contents.
4125 * maxalen is set to the maximum length of the host IP address string
4126 * plus 8 for the maximum length of the port#.
4127 * minalen is set to the minimum length of the host IP address string
4128 * plus 4 for the minimum length of the port#.
4129 * These lengths do not include NULL termination,
4130 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4135 rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4136 rin->sin_family = AF_INET;
4137 rin->sin_len = sizeof(struct sockaddr_in);
4138 maxalen = INET_ADDRSTRLEN - 1 + 8;
4144 rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4145 rin6->sin6_family = AF_INET6;
4146 rin6->sin6_len = sizeof(struct sockaddr_in6);
4147 maxalen = INET6_ADDRSTRLEN - 1 + 8;
4152 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4153 i = fxdr_unsigned(int, *tl);
4155 error = NFSERR_BADXDR;
4157 } else if (i == 0) {
4159 } else if (!cantparse && i <= maxalen && i >= minalen) {
4160 error = nfsrv_mtostr(nd, addr, i);
4165 * Parse out the address fields. We expect 6 decimal numbers
4166 * separated by '.'s for AF_INET and two decimal numbers
4167 * preceeded by '.'s for AF_INET6.
4173 * For AF_INET6, first parse the host address.
4176 cp = strchr(addr, '.');
4179 if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4195 while (cp != NULL && *cp && i < 6) {
4197 while (*cp2 && *cp2 != '.')
4205 j = nfsrv_getipnumber(cp);
4210 port.cval[5 - i] = j;
4220 * The host address INADDR_ANY is (mis)used to indicate
4221 * "there is no valid callback address".
4226 if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4228 rin6->sin6_port = htons(port.sval);
4235 if (ip.ival != INADDR_ANY) {
4236 rin->sin_addr.s_addr = htonl(ip.ival);
4237 rin->sin_port = htons(port.sval);
4248 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4254 switch (nd->nd_nam->sa_family) {
4257 sin = (struct sockaddr_in *)nd->nd_nam;
4258 rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4259 rin->sin_family = AF_INET;
4260 rin->sin_len = sizeof(struct sockaddr_in);
4261 rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4262 rin->sin_port = 0x0;
4267 sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4268 rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4269 rin6->sin6_family = AF_INET6;
4270 rin6->sin6_len = sizeof(struct sockaddr_in6);
4271 rin6->sin6_addr = sin6->sin6_addr;
4272 rin6->sin6_port = 0x0;
4276 clp->lc_program = 0;
4280 NFSEXITCODE2(error, nd);
4285 * Turn a string of up to three decimal digits into a number. Return -1 upon
4289 nfsrv_getipnumber(u_char *cp)
4294 if (j > 2 || *cp < '0' || *cp > '9')
4307 * This function checks for restart conditions.
4310 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4311 nfsv4stateid_t *stateidp, int specialid)
4316 * First check for a server restart. Open, LockT, ReleaseLockOwner
4317 * and DelegPurge have a clientid, the rest a stateid.
4320 (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4321 if (clientid.lval[0] != nfsrvboottime) {
4322 ret = NFSERR_STALECLIENTID;
4325 } else if (stateidp->other[0] != nfsrvboottime &&
4327 ret = NFSERR_STALESTATEID;
4332 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4333 * not use a lock/open owner seqid#, so the check can be done now.
4334 * (The others will be checked, as required, later.)
4336 if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4340 ret = nfsrv_checkgrace(NULL, NULL, flags);
4352 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4355 int error = 0, notreclaimed;
4356 struct nfsrv_stable *sp;
4358 if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE |
4359 NFSNSF_GRACEOVER)) == 0) {
4361 * First, check to see if all of the clients have done a
4362 * ReclaimComplete. If so, grace can end now.
4365 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4366 if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4371 if (notreclaimed == 0)
4372 nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER |
4376 if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4377 if (flags & NFSLCK_RECLAIM) {
4378 error = NFSERR_NOGRACE;
4382 if (!(flags & NFSLCK_RECLAIM)) {
4383 error = NFSERR_GRACE;
4386 if (nd != NULL && clp != NULL &&
4387 (nd->nd_flag & ND_NFSV41) != 0 &&
4388 (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4389 error = NFSERR_NOGRACE;
4394 * If grace is almost over and we are still getting Reclaims,
4395 * extend grace a bit.
4397 if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4398 nfsrv_stablefirst.nsf_eograce)
4399 nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4409 * Do a server callback.
4410 * The "trunc" argument is slightly overloaded and refers to different
4411 * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4414 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4415 int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4416 int laytype, NFSPROC_T *p)
4420 struct nfsrv_descript *nd;
4424 struct nfsdsession *sep = NULL;
4427 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4428 cred = newnfs_getcred();
4429 NFSLOCKSTATE(); /* mostly for lc_cbref++ */
4430 if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4437 * Fill the callback program# and version into the request
4438 * structure for newnfs_connect() to use.
4440 clp->lc_req.nr_prog = clp->lc_program;
4442 if ((clp->lc_flags & LCL_NFSV41) != 0)
4443 clp->lc_req.nr_vers = NFSV41_CBVERS;
4446 clp->lc_req.nr_vers = NFSV4_CBVERS;
4449 * First, fill in some of the fields of nd and cr.
4451 nd->nd_flag = ND_NFSV4;
4452 if (clp->lc_flags & LCL_GSS)
4453 nd->nd_flag |= ND_KERBV;
4454 if ((clp->lc_flags & LCL_NFSV41) != 0)
4455 nd->nd_flag |= ND_NFSV41;
4456 if ((clp->lc_flags & LCL_NFSV42) != 0)
4457 nd->nd_flag |= ND_NFSV42;
4459 cred->cr_uid = clp->lc_uid;
4460 cred->cr_gid = clp->lc_gid;
4461 callback = clp->lc_callback;
4463 cred->cr_ngroups = 1;
4466 * Get the first mbuf for the request.
4468 MGET(m, M_WAITOK, MT_DATA);
4470 nd->nd_mreq = nd->nd_mb = m;
4471 nd->nd_bpos = mtod(m, caddr_t);
4474 * and build the callback request.
4476 if (procnum == NFSV4OP_CBGETATTR) {
4477 nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4478 error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4479 "CB Getattr", &sep);
4481 m_freem(nd->nd_mreq);
4484 (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4485 (void)nfsrv_putattrbit(nd, attrbitp);
4486 } else if (procnum == NFSV4OP_CBRECALL) {
4487 nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4488 error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4491 m_freem(nd->nd_mreq);
4494 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4495 *tl++ = txdr_unsigned(stateidp->seqid);
4496 NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4498 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4503 (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4504 } else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4505 NFSD_DEBUG(4, "docallback layout recall\n");
4506 nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4507 error = nfsrv_cbcallargs(nd, clp, callback,
4508 NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep);
4509 NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4511 m_freem(nd->nd_mreq);
4514 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4515 *tl++ = txdr_unsigned(laytype);
4516 *tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4518 *tl++ = newnfs_true;
4520 *tl++ = newnfs_false;
4521 *tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4522 nfsm_fhtom(nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4523 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4525 txdr_hyper(tval, tl); tl += 2;
4527 txdr_hyper(tval, tl); tl += 2;
4528 *tl++ = txdr_unsigned(stateidp->seqid);
4529 NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4530 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4531 NFSD_DEBUG(4, "aft args\n");
4532 } else if (procnum == NFSV4PROC_CBNULL) {
4533 nd->nd_procnum = NFSV4PROC_CBNULL;
4534 if ((clp->lc_flags & LCL_NFSV41) != 0) {
4535 error = nfsv4_getcbsession(clp, &sep);
4537 m_freem(nd->nd_mreq);
4542 error = NFSERR_SERVERFAULT;
4543 m_freem(nd->nd_mreq);
4548 * Call newnfs_connect(), as required, and then newnfs_request().
4550 (void) newnfs_sndlock(&clp->lc_req.nr_lock);
4551 if (clp->lc_req.nr_client == NULL) {
4552 if ((clp->lc_flags & LCL_NFSV41) != 0) {
4553 error = ECONNREFUSED;
4554 nfsrv_freesession(sep, NULL);
4555 } else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4556 error = newnfs_connect(NULL, &clp->lc_req, cred,
4559 error = newnfs_connect(NULL, &clp->lc_req, cred,
4562 newnfs_sndunlock(&clp->lc_req.nr_lock);
4563 NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4565 if ((nd->nd_flag & ND_NFSV41) != 0) {
4566 KASSERT(sep != NULL, ("sep NULL"));
4567 if (sep->sess_cbsess.nfsess_xprt != NULL)
4568 error = newnfs_request(nd, NULL, clp,
4569 &clp->lc_req, NULL, NULL, cred,
4570 clp->lc_program, clp->lc_req.nr_vers, NULL,
4571 1, NULL, &sep->sess_cbsess);
4574 * This should probably never occur, but if a
4575 * client somehow does an RPC without a
4576 * SequenceID Op that causes a callback just
4577 * after the nfsd threads have been terminated
4578 * and restared we could conceivably get here
4579 * without a backchannel xprt.
4581 printf("nfsrv_docallback: no xprt\n");
4582 error = ECONNREFUSED;
4584 NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4585 nfsrv_freesession(sep, NULL);
4587 error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4588 NULL, NULL, cred, clp->lc_program,
4589 clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4595 * If error is set here, the Callback path isn't working
4596 * properly, so twiddle the appropriate LCL_ flags.
4597 * (nd_repstat != 0 indicates the Callback path is working,
4598 * but the callback failed on the client.)
4602 * Mark the callback pathway down, which disabled issuing
4603 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4606 clp->lc_flags |= LCL_CBDOWN;
4610 * Callback worked. If the callback path was down, disable
4611 * callbacks, so no more delegations will be issued. (This
4612 * is done on the assumption that the callback pathway is
4616 if (clp->lc_flags & LCL_CBDOWN)
4617 clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4619 if (nd->nd_repstat) {
4620 error = nd->nd_repstat;
4621 NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4623 } else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4624 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4625 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4627 m_freem(nd->nd_mrep);
4631 if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4632 clp->lc_flags &= ~LCL_WAKEUPWANTED;
4643 * Set up the compound RPC for the callback.
4646 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4647 uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4652 len = strlen(optag);
4653 (void)nfsm_strtom(nd, optag, len);
4654 NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4655 if ((nd->nd_flag & ND_NFSV41) != 0) {
4656 if ((nd->nd_flag & ND_NFSV42) != 0)
4657 *tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4659 *tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4660 *tl++ = txdr_unsigned(callback);
4661 *tl++ = txdr_unsigned(2);
4662 *tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4663 error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4666 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4667 *tl = txdr_unsigned(op);
4669 *tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4670 *tl++ = txdr_unsigned(callback);
4671 *tl++ = txdr_unsigned(1);
4672 *tl = txdr_unsigned(op);
4678 * Return the next index# for a clientid. Mostly just increment and return
4679 * the next one, but... if the 32bit unsigned does actually wrap around,
4680 * it should be rebooted.
4681 * At an average rate of one new client per second, it will wrap around in
4682 * approximately 136 years. (I think the server will have been shut
4683 * down or rebooted before then.)
4686 nfsrv_nextclientindex(void)
4688 static u_int32_t client_index = 0;
4691 if (client_index != 0)
4692 return (client_index);
4694 printf("%s: out of clientids\n", __func__);
4695 return (client_index);
4699 * Return the next index# for a stateid. Mostly just increment and return
4700 * the next one, but... if the 32bit unsigned does actually wrap around
4701 * (will a BSD server stay up that long?), find
4702 * new start and end values.
4705 nfsrv_nextstateindex(struct nfsclient *clp)
4707 struct nfsstate *stp;
4709 u_int32_t canuse, min_index, max_index;
4711 if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4712 clp->lc_stateindex++;
4713 if (clp->lc_stateindex != clp->lc_statemaxindex)
4714 return (clp->lc_stateindex);
4718 * Yuck, we've hit the end.
4719 * Look for a new min and max.
4722 max_index = 0xffffffff;
4723 for (i = 0; i < nfsrv_statehashsize; i++) {
4724 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4725 if (stp->ls_stateid.other[2] > 0x80000000) {
4726 if (stp->ls_stateid.other[2] < max_index)
4727 max_index = stp->ls_stateid.other[2];
4729 if (stp->ls_stateid.other[2] > min_index)
4730 min_index = stp->ls_stateid.other[2];
4736 * Yikes, highly unlikely, but I'll handle it anyhow.
4738 if (min_index == 0x80000000 && max_index == 0x80000001) {
4741 * Loop around until we find an unused entry. Return that
4742 * and set LCL_INDEXNOTOK, so the search will continue next time.
4743 * (This is one of those rare cases where a goto is the
4744 * cleanest way to code the loop.)
4747 for (i = 0; i < nfsrv_statehashsize; i++) {
4748 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4749 if (stp->ls_stateid.other[2] == canuse) {
4755 clp->lc_flags |= LCL_INDEXNOTOK;
4760 * Ok to start again from min + 1.
4762 clp->lc_stateindex = min_index + 1;
4763 clp->lc_statemaxindex = max_index;
4764 clp->lc_flags &= ~LCL_INDEXNOTOK;
4765 return (clp->lc_stateindex);
4769 * The following functions handle the stable storage file that deals with
4770 * the edge conditions described in RFC3530 Sec. 8.6.3.
4771 * The file is as follows:
4772 * - a single record at the beginning that has the lease time of the
4773 * previous server instance (before the last reboot) and the nfsrvboottime
4774 * values for the previous server boots.
4775 * These previous boot times are used to ensure that the current
4776 * nfsrvboottime does not, somehow, get set to a previous one.
4777 * (This is important so that Stale ClientIDs and StateIDs can
4779 * The number of previous nfsvrboottime values precedes the list.
4780 * - followed by some number of appended records with:
4781 * - client id string
4782 * - flag that indicates it is a record revoking state via lease
4783 * expiration or similar
4784 * OR has successfully acquired state.
4785 * These structures vary in length, with the client string at the end, up
4786 * to NFSV4_OPAQUELIMIT in size.
4788 * At the end of the grace period, the file is truncated, the first
4789 * record is rewritten with updated information and any acquired state
4790 * records for successful reclaims of state are written.
4792 * Subsequent records are appended when the first state is issued to
4793 * a client and when state is revoked for a client.
4795 * When reading the file in, state issued records that come later in
4796 * the file override older ones, since the append log is in cronological order.
4797 * If, for some reason, the file can't be read, the grace period is
4798 * immediately terminated and all reclaims get NFSERR_NOGRACE.
4802 * Read in the stable storage file. Called by nfssvc() before the nfsd
4803 * processes start servicing requests.
4806 nfsrv_setupstable(NFSPROC_T *p)
4808 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4809 struct nfsrv_stable *sp, *nsp;
4810 struct nfst_rec *tsp;
4811 int error, i, tryagain;
4813 ssize_t aresid, len;
4816 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4817 * a reboot, so state has not been lost.
4819 if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4822 * Set Grace over just until the file reads successfully.
4824 nfsrvboottime = time_second;
4825 LIST_INIT(&sf->nsf_head);
4826 sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4827 sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4828 if (sf->nsf_fp == NULL)
4830 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4831 (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4832 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4833 if (error || aresid || sf->nsf_numboots == 0 ||
4834 sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4838 * Now, read in the boottimes.
4840 sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4841 sizeof (time_t), M_TEMP, M_WAITOK);
4842 off = sizeof (struct nfsf_rec);
4843 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4844 (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4845 UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4846 if (error || aresid) {
4847 free(sf->nsf_bootvals, M_TEMP);
4848 sf->nsf_bootvals = NULL;
4853 * Make sure this nfsrvboottime is different from all recorded
4858 for (i = 0; i < sf->nsf_numboots; i++) {
4859 if (nfsrvboottime == sf->nsf_bootvals[i]) {
4867 sf->nsf_flags |= NFSNSF_OK;
4868 off += (sf->nsf_numboots * sizeof (time_t));
4871 * Read through the file, building a list of records for grace
4873 * Each record is between sizeof (struct nfst_rec) and
4874 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4875 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4877 tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4878 NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4880 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4881 (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4882 off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4883 len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4884 if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4885 len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4887 * Yuck, the file has been corrupted, so just return
4888 * after clearing out any restart state, so the grace period
4891 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4892 LIST_REMOVE(sp, nst_list);
4896 sf->nsf_flags &= ~NFSNSF_OK;
4897 free(sf->nsf_bootvals, M_TEMP);
4898 sf->nsf_bootvals = NULL;
4902 off += sizeof (struct nfst_rec) + tsp->len - 1;
4904 * Search the list for a matching client.
4906 LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4907 if (tsp->len == sp->nst_len &&
4908 !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4911 if (sp == LIST_END(&sf->nsf_head)) {
4912 sp = (struct nfsrv_stable *)malloc(tsp->len +
4913 sizeof (struct nfsrv_stable) - 1, M_TEMP,
4915 NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4916 sizeof (struct nfst_rec) + tsp->len - 1);
4917 LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4919 if (tsp->flag == NFSNST_REVOKE)
4920 sp->nst_flag |= NFSNST_REVOKE;
4923 * A subsequent timestamp indicates the client
4924 * did a setclientid/confirm and any previous
4925 * revoke is no longer relevant.
4927 sp->nst_flag &= ~NFSNST_REVOKE;
4932 sf->nsf_flags = NFSNSF_OK;
4933 sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4938 * Update the stable storage file, now that the grace period is over.
4941 nfsrv_updatestable(NFSPROC_T *p)
4943 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4944 struct nfsrv_stable *sp, *nsp;
4946 struct nfsvattr nva;
4948 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4953 if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4955 sf->nsf_flags |= NFSNSF_UPDATEDONE;
4957 * Ok, we need to rewrite the stable storage file.
4958 * - truncate to 0 length
4959 * - write the new first structure
4960 * - loop through the data structures, writing out any that
4961 * have timestamps older than the old boot
4963 if (sf->nsf_bootvals) {
4965 for (i = sf->nsf_numboots - 2; i >= 0; i--)
4966 sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4968 sf->nsf_numboots = 1;
4969 sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4972 sf->nsf_bootvals[0] = nfsrvboottime;
4973 sf->nsf_lease = nfsrv_lease;
4974 NFSVNO_ATTRINIT(&nva);
4975 NFSVNO_SETATTRVAL(&nva, size, 0);
4976 vp = NFSFPVNODE(sf->nsf_fp);
4977 vn_start_write(vp, &mp, V_WAIT);
4978 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4979 error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4984 vn_finished_write(mp);
4986 error = NFSD_RDWR(UIO_WRITE, vp,
4987 (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4988 UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4990 error = NFSD_RDWR(UIO_WRITE, vp,
4991 (caddr_t)sf->nsf_bootvals,
4992 sf->nsf_numboots * sizeof (time_t),
4993 (off_t)(sizeof (struct nfsf_rec)),
4994 UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4995 free(sf->nsf_bootvals, M_TEMP);
4996 sf->nsf_bootvals = NULL;
4998 sf->nsf_flags &= ~NFSNSF_OK;
4999 printf("EEK! Can't write NfsV4 stable storage file\n");
5002 sf->nsf_flags |= NFSNSF_OK;
5005 * Loop through the list and write out timestamp records for
5006 * any clients that successfully reclaimed state.
5008 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5009 if (sp->nst_flag & NFSNST_GOTSTATE) {
5010 nfsrv_writestable(sp->nst_client, sp->nst_len,
5011 NFSNST_NEWSTATE, p);
5012 sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5014 LIST_REMOVE(sp, nst_list);
5017 nfsrv_backupstable();
5021 * Append a record to the stable storage file.
5024 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5026 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
5027 struct nfst_rec *sp;
5030 if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5032 sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5033 len - 1, M_TEMP, M_WAITOK);
5035 NFSBCOPY(client, sp->client, len);
5037 error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5038 (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5039 UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5042 sf->nsf_flags &= ~NFSNSF_OK;
5043 printf("EEK! Can't write NfsV4 stable storage file\n");
5048 * This function is called during the grace period to mark a client
5049 * that successfully reclaimed state.
5052 nfsrv_markstable(struct nfsclient *clp)
5054 struct nfsrv_stable *sp;
5057 * First find the client structure.
5059 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5060 if (sp->nst_len == clp->lc_idlen &&
5061 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5064 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5068 * Now, just mark it and set the nfsclient back pointer.
5070 sp->nst_flag |= NFSNST_GOTSTATE;
5075 * This function is called when a NFSv4.1 client does a ReclaimComplete.
5076 * Very similar to nfsrv_markstable(), except for the flag being set.
5079 nfsrv_markreclaim(struct nfsclient *clp)
5081 struct nfsrv_stable *sp;
5084 * First find the client structure.
5086 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5087 if (sp->nst_len == clp->lc_idlen &&
5088 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5091 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5095 * Now, just set the flag.
5097 sp->nst_flag |= NFSNST_RECLAIMED;
5101 * This function is called for a reclaim, to see if it gets grace.
5102 * It returns 0 if a reclaim is allowed, 1 otherwise.
5105 nfsrv_checkstable(struct nfsclient *clp)
5107 struct nfsrv_stable *sp;
5110 * First, find the entry for the client.
5112 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5113 if (sp->nst_len == clp->lc_idlen &&
5114 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5119 * If not in the list, state was revoked or no state was issued
5120 * since the previous reboot, a reclaim is denied.
5122 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
5123 (sp->nst_flag & NFSNST_REVOKE) ||
5124 !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
5130 * Test for and try to clear out a conflicting client. This is called by
5131 * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5133 * The trick here is that it can't revoke a conflicting client with an
5134 * expired lease unless it holds the v4root lock, so...
5135 * If no v4root lock, get the lock and return 1 to indicate "try again".
5136 * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5137 * the revocation worked and the conflicting client is "bye, bye", so it
5138 * can be tried again.
5139 * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5140 * Unlocks State before a non-zero value is returned.
5143 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5146 int gotlock, lktype = 0;
5149 * If lease hasn't expired, we can't fix it.
5151 if (clp->lc_expiry >= NFSD_MONOSEC ||
5152 !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
5154 if (*haslockp == 0) {
5157 lktype = NFSVOPISLOCKED(vp);
5160 NFSLOCKV4ROOTMUTEX();
5161 nfsv4_relref(&nfsv4rootfs_lock);
5163 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5164 NFSV4ROOTLOCKMUTEXPTR, NULL);
5166 NFSUNLOCKV4ROOTMUTEX();
5169 NFSVOPLOCK(vp, lktype | LK_RETRY);
5170 if (VN_IS_DOOMED(vp))
5178 * Ok, we can expire the conflicting client.
5180 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5181 nfsrv_backupstable();
5182 nfsrv_cleanclient(clp, p);
5183 nfsrv_freedeleglist(&clp->lc_deleg);
5184 nfsrv_freedeleglist(&clp->lc_olddeleg);
5185 LIST_REMOVE(clp, lc_hash);
5186 nfsrv_zapclient(clp, p);
5191 * Resolve a delegation conflict.
5192 * Returns 0 to indicate the conflict was resolved without sleeping.
5193 * Return -1 to indicate that the caller should check for conflicts again.
5194 * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5196 * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5197 * for a return of 0, since there was no sleep and it could be required
5198 * later. It is released for a return of NFSERR_DELAY, since the caller
5199 * will return that error. It is released when a sleep was done waiting
5200 * for the delegation to be returned or expire (so that other nfsds can
5201 * handle ops). Then, it must be acquired for the write to stable storage.
5202 * (This function is somewhat similar to nfsrv_clientconflict(), but
5203 * the semantics differ in a couple of subtle ways. The return of 0
5204 * indicates the conflict was resolved without sleeping here, not
5205 * that the conflict can't be resolved and the handling of nfsv4root_lock
5206 * differs, as noted above.)
5207 * Unlocks State before returning a non-zero value.
5210 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5213 struct nfsclient *clp = stp->ls_clp;
5214 int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5215 nfsv4stateid_t tstateid;
5219 * If the conflict is with an old delegation...
5221 if (stp->ls_flags & NFSLCK_OLDDELEG) {
5223 * You can delete it, if it has expired.
5225 if (clp->lc_delegtime < NFSD_MONOSEC) {
5226 nfsrv_freedeleg(stp);
5233 * During this delay, the old delegation could expire or it
5234 * could be recovered by the client via an Open with
5235 * CLAIM_DELEGATE_PREV.
5236 * Release the nfsv4root_lock, if held.
5240 NFSLOCKV4ROOTMUTEX();
5241 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5242 NFSUNLOCKV4ROOTMUTEX();
5244 error = NFSERR_DELAY;
5249 * It's a current delegation, so:
5250 * - check to see if the delegation has expired
5251 * - if so, get the v4root lock and then expire it
5253 if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
5255 * - do a recall callback, since not yet done
5256 * For now, never allow truncate to be set. To use
5257 * truncate safely, it must be guaranteed that the
5258 * Remove, Rename or Setattr with size of 0 will
5259 * succeed and that would require major changes to
5260 * the VFS/Vnode OPs.
5261 * Set the expiry time large enough so that it won't expire
5262 * until after the callback, then set it correctly, once
5263 * the callback is done. (The delegation will now time
5264 * out whether or not the Recall worked ok. The timeout
5265 * will be extended when ops are done on the delegation
5266 * stateid, up to the timelimit.)
5268 stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5270 stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
5272 stp->ls_flags |= NFSLCK_DELEGRECALL;
5275 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5276 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5277 * in order to try and avoid a race that could happen
5278 * when a CBRecall request passed the Open reply with
5279 * the delegation in it when transitting the network.
5280 * Since nfsrv_docallback will sleep, don't use stp after
5283 NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5285 NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5290 NFSLOCKV4ROOTMUTEX();
5291 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5292 NFSUNLOCKV4ROOTMUTEX();
5296 error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5297 &tstateid, 0, &tfh, NULL, NULL, 0, p);
5299 } while ((error == NFSERR_BADSTATEID ||
5300 error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5301 error = NFSERR_DELAY;
5305 if (clp->lc_expiry >= NFSD_MONOSEC &&
5306 stp->ls_delegtime >= NFSD_MONOSEC) {
5309 * A recall has been done, but it has not yet expired.
5314 NFSLOCKV4ROOTMUTEX();
5315 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5316 NFSUNLOCKV4ROOTMUTEX();
5318 error = NFSERR_DELAY;
5323 * If we don't yet have the lock, just get it and then return,
5324 * since we need that before deleting expired state, such as
5326 * When getting the lock, unlock the vnode, so other nfsds that
5327 * are in progress, won't get stuck waiting for the vnode lock.
5329 if (*haslockp == 0) {
5332 lktype = NFSVOPISLOCKED(vp);
5335 NFSLOCKV4ROOTMUTEX();
5336 nfsv4_relref(&nfsv4rootfs_lock);
5338 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5339 NFSV4ROOTLOCKMUTEXPTR, NULL);
5341 NFSUNLOCKV4ROOTMUTEX();
5344 NFSVOPLOCK(vp, lktype | LK_RETRY);
5345 if (VN_IS_DOOMED(vp)) {
5347 NFSLOCKV4ROOTMUTEX();
5348 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5349 NFSUNLOCKV4ROOTMUTEX();
5350 error = NFSERR_PERM;
5360 * Ok, we can delete the expired delegation.
5361 * First, write the Revoke record to stable storage and then
5362 * clear out the conflict.
5363 * Since all other nfsd threads are now blocked, we can safely
5364 * sleep without the state changing.
5366 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5367 nfsrv_backupstable();
5368 if (clp->lc_expiry < NFSD_MONOSEC) {
5369 nfsrv_cleanclient(clp, p);
5370 nfsrv_freedeleglist(&clp->lc_deleg);
5371 nfsrv_freedeleglist(&clp->lc_olddeleg);
5372 LIST_REMOVE(clp, lc_hash);
5375 nfsrv_freedeleg(stp);
5379 nfsrv_zapclient(clp, p);
5388 * Check for a remove allowed, if remove is set to 1 and get rid of
5392 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5393 nfsquad_t clientid, NFSPROC_T *p)
5395 struct nfsclient *clp;
5396 struct nfsstate *stp;
5397 struct nfslockfile *lfp;
5398 int error, haslock = 0;
5403 * First, get the lock file structure.
5404 * (A return of -1 means no associated state, so remove ok.)
5406 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5409 if (error == 0 && clientid.qval != 0)
5410 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5411 (nfsquad_t)((u_quad_t)0), 0, nd, p);
5413 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5417 NFSLOCKV4ROOTMUTEX();
5418 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5419 NFSUNLOCKV4ROOTMUTEX();
5427 * Now, we must Recall any delegations.
5429 error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5432 * nfsrv_cleandeleg() unlocks state for non-zero
5438 NFSLOCKV4ROOTMUTEX();
5439 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5440 NFSUNLOCKV4ROOTMUTEX();
5446 * Now, look for a conflicting open share.
5450 * If the entry in the directory was the last reference to the
5451 * corresponding filesystem object, the object can be destroyed
5453 if(lfp->lf_usecount>1)
5454 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5455 if (stp->ls_flags & NFSLCK_WRITEDENY) {
5456 error = NFSERR_FILEOPEN;
5464 NFSLOCKV4ROOTMUTEX();
5465 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5466 NFSUNLOCKV4ROOTMUTEX();
5475 * Clear out all delegations for the file referred to by lfp.
5476 * May return NFSERR_DELAY, if there will be a delay waiting for
5477 * delegations to expire.
5478 * Returns -1 to indicate it slept while recalling a delegation.
5479 * This function has the side effect of deleting the nfslockfile structure,
5480 * if it no longer has associated state and didn't have to sleep.
5481 * Unlocks State before a non-zero value is returned.
5484 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5485 struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5487 struct nfsstate *stp, *nstp;
5490 stp = LIST_FIRST(&lfp->lf_deleg);
5491 while (stp != LIST_END(&lfp->lf_deleg)) {
5492 nstp = LIST_NEXT(stp, ls_file);
5493 if (stp->ls_clp != clp) {
5494 ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5497 * nfsrv_delegconflict() unlocks state
5498 * when it returns non-zero.
5511 * There are certain operations that, when being done outside of NFSv4,
5512 * require that any NFSv4 delegation for the file be recalled.
5513 * This function is to be called for those cases:
5514 * VOP_RENAME() - When a delegation is being recalled for any reason,
5515 * the client may have to do Opens against the server, using the file's
5516 * final component name. If the file has been renamed on the server,
5517 * that component name will be incorrect and the Open will fail.
5518 * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5519 * been removed on the server, if there is a delegation issued to
5520 * that client for the file. I say "theoretically" since clients
5521 * normally do an Access Op before the Open and that Access Op will
5522 * fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5523 * they will detect the file's removal in the same manner. (There is
5524 * one case where RFC3530 allows a client to do an Open without first
5525 * doing an Access Op, which is passage of a check against the ACE
5526 * returned with a Write delegation, but current practice is to ignore
5527 * the ACE and always do an Access Op.)
5528 * Since the functions can only be called with an unlocked vnode, this
5529 * can't be done at this time.
5530 * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5531 * locks locally in the client, which are not visible to the server. To
5532 * deal with this, issuing of delegations for a vnode must be disabled
5533 * and all delegations for the vnode recalled. This is done via the
5534 * second function, using the VV_DISABLEDELEG vflag on the vnode.
5537 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5543 * First, check to see if the server is currently running and it has
5544 * been called for a regular file when issuing delegations.
5546 if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5547 nfsrv_issuedelegs == 0)
5550 KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5552 * First, get a reference on the nfsv4rootfs_lock so that an
5553 * exclusive lock cannot be acquired by another thread.
5555 NFSLOCKV4ROOTMUTEX();
5556 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5557 NFSUNLOCKV4ROOTMUTEX();
5560 * Now, call nfsrv_checkremove() in a loop while it returns
5561 * NFSERR_DELAY. Return upon any other error or when timed out.
5563 starttime = NFSD_MONOSEC;
5565 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5566 error = nfsrv_checkremove(vp, 0, NULL,
5567 (nfsquad_t)((u_quad_t)0), p);
5571 if (error == NFSERR_DELAY) {
5572 if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5574 /* Sleep for a short period of time */
5575 (void) nfs_catnap(PZERO, 0, "nfsremove");
5577 } while (error == NFSERR_DELAY);
5578 NFSLOCKV4ROOTMUTEX();
5579 nfsv4_relref(&nfsv4rootfs_lock);
5580 NFSUNLOCKV4ROOTMUTEX();
5584 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5587 #ifdef VV_DISABLEDELEG
5589 * First, flag issuance of delegations disabled.
5591 atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5595 * Then call nfsd_recalldelegation() to get rid of all extant
5598 nfsd_recalldelegation(vp, p);
5602 * Check for conflicting locks, etc. and then get rid of delegations.
5603 * (At one point I thought that I should get rid of delegations for any
5604 * Setattr, since it could potentially disallow the I/O op (read or write)
5605 * allowed by the delegation. However, Setattr Ops that aren't changing
5606 * the size get a stateid of all 0s, so you can't tell if it is a delegation
5607 * for the same client or a different one, so I decided to only get rid
5608 * of delegations for other clients when the size is being changed.)
5609 * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5610 * as Write backs, even if there is no delegation, so it really isn't any
5614 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5615 nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5616 struct nfsexstuff *exp, NFSPROC_T *p)
5618 struct nfsstate st, *stp = &st;
5619 struct nfslock lo, *lop = &lo;
5623 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5624 stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5625 lop->lo_first = nvap->na_size;
5630 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5631 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5632 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5633 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5634 stp->ls_flags |= NFSLCK_SETATTR;
5635 if (stp->ls_flags == 0)
5637 lop->lo_end = NFS64BITSSET;
5638 lop->lo_flags = NFSLCK_WRITE;
5639 stp->ls_ownerlen = 0;
5641 stp->ls_uid = nd->nd_cred->cr_uid;
5642 stp->ls_stateid.seqid = stateidp->seqid;
5643 clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5644 clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5645 stp->ls_stateid.other[2] = stateidp->other[2];
5646 error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5647 stateidp, exp, nd, p);
5650 NFSEXITCODE2(error, nd);
5655 * Check for a write delegation and do a CBGETATTR if there is one, updating
5656 * the attributes, as required.
5657 * Should I return an error if I can't get the attributes? (For now, I'll
5661 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5662 struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5664 struct nfsstate *stp;
5665 struct nfslockfile *lfp;
5666 struct nfsclient *clp;
5667 struct nfsvattr nva;
5670 nfsattrbit_t cbbits;
5671 u_quad_t delegfilerev;
5673 NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5674 if (!NFSNONZERO_ATTRBIT(&cbbits))
5676 if (nfsrv_writedelegcnt == 0)
5680 * Get the lock file structure.
5681 * (A return of -1 means no associated state, so return ok.)
5683 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5686 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5695 * Now, look for a write delegation.
5697 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5698 if (stp->ls_flags & NFSLCK_DELEGWRITE)
5701 if (stp == LIST_END(&lfp->lf_deleg)) {
5706 delegfilerev = stp->ls_filerev;
5709 * If the Write delegation was issued as a part of this Compound RPC
5710 * or if we have an Implied Clientid (used in a previous Op in this
5711 * compound) and it is the client the delegation was issued to,
5713 * I also assume that it is from the same client iff the network
5714 * host IP address is the same as the callback address. (Not
5715 * exactly correct by the RFC, but avoids a lot of Getattr
5718 if (nd->nd_compref == stp->ls_compref ||
5719 ((nd->nd_flag & ND_IMPLIEDCLID) &&
5720 clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5721 nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5727 * We are now done with the delegation state structure,
5728 * so the statelock can be released and we can now tsleep().
5732 * Now, we must do the CB Getattr callback, to see if Change or Size
5735 if (clp->lc_expiry >= NFSD_MONOSEC) {
5737 NFSVNO_ATTRINIT(&nva);
5738 nva.na_filerev = NFS64BITSSET;
5739 error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5740 0, &nfh, &nva, &cbbits, 0, p);
5742 if ((nva.na_filerev != NFS64BITSSET &&
5743 nva.na_filerev > delegfilerev) ||
5744 (NFSVNO_ISSETSIZE(&nva) &&
5745 nva.na_size != nvap->na_size)) {
5746 error = nfsvno_updfilerev(vp, nvap, nd, p);
5747 if (NFSVNO_ISSETSIZE(&nva))
5748 nvap->na_size = nva.na_size;
5751 error = 0; /* Ignore callback errors for now. */
5757 NFSEXITCODE2(error, nd);
5762 * This function looks for openowners that haven't had any opens for
5763 * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5767 nfsrv_throwawayopens(NFSPROC_T *p)
5769 struct nfsclient *clp, *nclp;
5770 struct nfsstate *stp, *nstp;
5774 nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5776 * For each client...
5778 for (i = 0; i < nfsrv_clienthashsize; i++) {
5779 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5780 LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5781 if (LIST_EMPTY(&stp->ls_open) &&
5782 (stp->ls_noopens > NFSNOOPEN ||
5783 (nfsrv_openpluslock * 2) >
5784 nfsrv_v4statelimit))
5785 nfsrv_freeopenowner(stp, 0, p);
5793 * This function checks to see if the credentials are the same.
5794 * Returns 1 for not same, 0 otherwise.
5797 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5800 if (nd->nd_flag & ND_GSS) {
5801 if (!(clp->lc_flags & LCL_GSS))
5803 if (clp->lc_flags & LCL_NAME) {
5804 if (nd->nd_princlen != clp->lc_namelen ||
5805 NFSBCMP(nd->nd_principal, clp->lc_name,
5811 if (nd->nd_cred->cr_uid == clp->lc_uid)
5815 } else if (clp->lc_flags & LCL_GSS)
5818 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5819 * in RFC3530, which talks about principals, but doesn't say anything
5820 * about uids for AUTH_SYS.)
5822 if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5829 * Calculate the lease expiry time.
5832 nfsrv_leaseexpiry(void)
5835 if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5836 return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5837 return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5841 * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5844 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5847 if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5850 if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5851 stp->ls_delegtime < stp->ls_delegtimelimit) {
5852 stp->ls_delegtime += nfsrv_lease;
5853 if (stp->ls_delegtime > stp->ls_delegtimelimit)
5854 stp->ls_delegtime = stp->ls_delegtimelimit;
5859 * This function checks to see if there is any other state associated
5860 * with the openowner for this Open.
5861 * It returns 1 if there is no other state, 0 otherwise.
5864 nfsrv_nootherstate(struct nfsstate *stp)
5866 struct nfsstate *tstp;
5868 LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5869 if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5876 * Create a list of lock deltas (changes to local byte range locking
5877 * that can be rolled back using the list) and apply the changes via
5878 * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5879 * the rollback or update function will be called after this.
5880 * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5881 * call fails. If it returns an error, it will unlock the list.
5884 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5885 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5887 struct nfslock *lop, *nlop;
5890 /* Loop through the list of locks. */
5891 lop = LIST_FIRST(&lfp->lf_locallock);
5892 while (first < end && lop != NULL) {
5893 nlop = LIST_NEXT(lop, lo_lckowner);
5894 if (first >= lop->lo_end) {
5897 } else if (first < lop->lo_first) {
5898 /* new one starts before entry in list */
5899 if (end <= lop->lo_first) {
5900 /* no overlap between old and new */
5901 error = nfsrv_dolocal(vp, lfp, flags,
5902 NFSLCK_UNLOCK, first, end, cfp, p);
5907 /* handle fragment overlapped with new one */
5908 error = nfsrv_dolocal(vp, lfp, flags,
5909 NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5913 first = lop->lo_first;
5916 /* new one overlaps this entry in list */
5917 if (end <= lop->lo_end) {
5918 /* overlaps all of new one */
5919 error = nfsrv_dolocal(vp, lfp, flags,
5920 lop->lo_flags, first, end, cfp, p);
5925 /* handle fragment overlapped with new one */
5926 error = nfsrv_dolocal(vp, lfp, flags,
5927 lop->lo_flags, first, lop->lo_end, cfp, p);
5930 first = lop->lo_end;
5935 if (first < end && error == 0)
5936 /* handle fragment past end of list */
5937 error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5945 * Local lock unlock. Unlock all byte ranges that are no longer locked
5946 * by NFSv4. To do this, unlock any subranges of first-->end that
5947 * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5948 * list. This list has all locks for the file held by other
5949 * <clientid, lockowner> tuples. The list is ordered by increasing
5950 * lo_first value, but may have entries that overlap each other, for
5951 * the case of read locks.
5954 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5955 uint64_t init_end, NFSPROC_T *p)
5957 struct nfslock *lop;
5958 uint64_t first, end, prevfirst __unused;
5962 while (first < init_end) {
5963 /* Loop through all nfs locks, adjusting first and end */
5965 LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5966 KASSERT(prevfirst <= lop->lo_first,
5967 ("nfsv4 locks out of order"));
5968 KASSERT(lop->lo_first < lop->lo_end,
5969 ("nfsv4 bogus lock"));
5970 prevfirst = lop->lo_first;
5971 if (first >= lop->lo_first &&
5972 first < lop->lo_end)
5974 * Overlaps with initial part, so trim
5975 * off that initial part by moving first past
5978 first = lop->lo_end;
5979 else if (end > lop->lo_first &&
5980 lop->lo_first > first) {
5982 * This lock defines the end of the
5983 * segment to unlock, so set end to the
5984 * start of it and break out of the loop.
5986 end = lop->lo_first;
5991 * There is no segment left to do, so
5992 * break out of this loop and then exit
5993 * the outer while() since first will be set
5994 * to end, which must equal init_end here.
5999 /* Unlock this segment */
6000 (void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6001 NFSLCK_READ, first, end, NULL, p);
6002 nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6006 * Now move past this segment and look for any further
6007 * segment in the range, if there is one.
6015 * Do the local lock operation and update the rollback list, as required.
6016 * Perform the rollback and return the error if nfsvno_advlock() fails.
6019 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6020 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6022 struct nfsrollback *rlp;
6023 int error = 0, ltype, oldltype;
6025 if (flags & NFSLCK_WRITE)
6027 else if (flags & NFSLCK_READ)
6031 if (oldflags & NFSLCK_WRITE)
6033 else if (oldflags & NFSLCK_READ)
6037 if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6040 error = nfsvno_advlock(vp, ltype, first, end, p);
6043 cfp->cl_clientid.lval[0] = 0;
6044 cfp->cl_clientid.lval[1] = 0;
6046 cfp->cl_end = NFS64BITSSET;
6047 cfp->cl_flags = NFSLCK_WRITE;
6048 cfp->cl_ownerlen = 5;
6049 NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6051 nfsrv_locallock_rollback(vp, lfp, p);
6052 } else if (ltype != F_UNLCK) {
6053 rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6055 rlp->rlck_first = first;
6056 rlp->rlck_end = end;
6057 rlp->rlck_type = oldltype;
6058 LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6067 * Roll back local lock changes and free up the rollback list.
6070 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6072 struct nfsrollback *rlp, *nrlp;
6074 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6075 (void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6077 free(rlp, M_NFSDROLLBACK);
6079 LIST_INIT(&lfp->lf_rollback);
6083 * Update local lock list and delete rollback list (ie now committed to the
6084 * local locks). Most of the work is done by the internal function.
6087 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6090 struct nfsrollback *rlp, *nrlp;
6091 struct nfslock *new_lop, *other_lop;
6093 new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6094 if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6095 other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6099 new_lop->lo_flags = flags;
6100 new_lop->lo_first = first;
6101 new_lop->lo_end = end;
6102 nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6103 if (new_lop != NULL)
6104 free(new_lop, M_NFSDLOCK);
6105 if (other_lop != NULL)
6106 free(other_lop, M_NFSDLOCK);
6108 /* and get rid of the rollback list */
6109 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6110 free(rlp, M_NFSDROLLBACK);
6111 LIST_INIT(&lfp->lf_rollback);
6115 * Lock the struct nfslockfile for local lock updating.
6118 nfsrv_locklf(struct nfslockfile *lfp)
6122 /* lf_usecount ensures *lfp won't be free'd */
6125 gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6126 NFSSTATEMUTEXPTR, NULL);
6127 } while (gotlock == 0);
6132 * Unlock the struct nfslockfile after local lock updating.
6135 nfsrv_unlocklf(struct nfslockfile *lfp)
6138 nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6142 * Clear out all state for the NFSv4 server.
6143 * Must be called by a thread that can sleep when no nfsds are running.
6146 nfsrv_throwawayallstate(NFSPROC_T *p)
6148 struct nfsclient *clp, *nclp;
6149 struct nfslockfile *lfp, *nlfp;
6153 * For each client, clean out the state and then free the structure.
6155 for (i = 0; i < nfsrv_clienthashsize; i++) {
6156 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
6157 nfsrv_cleanclient(clp, p);
6158 nfsrv_freedeleglist(&clp->lc_deleg);
6159 nfsrv_freedeleglist(&clp->lc_olddeleg);
6160 free(clp->lc_stateid, M_NFSDCLIENT);
6161 free(clp, M_NFSDCLIENT);
6166 * Also, free up any remaining lock file structures.
6168 for (i = 0; i < nfsrv_lockhashsize; i++) {
6169 LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
6170 printf("nfsd unload: fnd a lock file struct\n");
6171 nfsrv_freenfslockfile(lfp);
6175 /* And get rid of the deviceid structures and layouts. */
6176 nfsrv_freealllayoutsanddevids();
6180 * Check the sequence# for the session and slot provided as an argument.
6181 * Also, renew the lease if the session will return NFS_OK.
6184 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6185 uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6186 uint32_t *sflagsp, NFSPROC_T *p)
6188 struct nfsdsession *sep;
6189 struct nfssessionhash *shp;
6193 shp = NFSSESSIONHASH(nd->nd_sessionid);
6194 NFSLOCKSESSION(shp);
6195 sep = nfsrv_findsession(nd->nd_sessionid);
6197 NFSUNLOCKSESSION(shp);
6198 return (NFSERR_BADSESSION);
6200 error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6201 sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6203 NFSUNLOCKSESSION(shp);
6206 if (cache_this != 0)
6207 nd->nd_flag |= ND_SAVEREPLY;
6208 /* Renew the lease. */
6209 sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6210 nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6211 nd->nd_flag |= ND_IMPLIEDCLID;
6213 /* Save maximum request and reply sizes. */
6214 nd->nd_maxreq = sep->sess_maxreq;
6215 nd->nd_maxresp = sep->sess_maxresp;
6218 * If this session handles the backchannel, save the nd_xprt for this
6219 * RPC, since this is the one being used.
6220 * RFC-5661 specifies that the fore channel will be implicitly
6221 * bound by a Sequence operation. However, since some NFSv4.1 clients
6222 * erroneously assumed that the back channel would be implicitly
6223 * bound as well, do the implicit binding unless a
6224 * BindConnectiontoSession has already been done on the session.
6226 if (sep->sess_clp->lc_req.nr_client != NULL &&
6227 sep->sess_cbsess.nfsess_xprt != nd->nd_xprt &&
6228 (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0 &&
6229 (sep->sess_clp->lc_flags & LCL_DONEBINDCONN) == 0) {
6231 "nfsrv_checksequence: implicit back channel bind\n");
6232 savxprt = sep->sess_cbsess.nfsess_xprt;
6233 SVC_ACQUIRE(nd->nd_xprt);
6234 nd->nd_xprt->xp_p2 =
6235 sep->sess_clp->lc_req.nr_client->cl_private;
6236 nd->nd_xprt->xp_idletimeout = 0; /* Disable timeout. */
6237 sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6238 if (savxprt != NULL)
6239 SVC_RELEASE(savxprt);
6243 if (sep->sess_clp->lc_req.nr_client == NULL)
6244 *sflagsp |= NFSV4SEQ_CBPATHDOWN;
6245 NFSUNLOCKSESSION(shp);
6246 if (error == NFSERR_EXPIRED) {
6247 *sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6249 } else if (error == NFSERR_ADMINREVOKED) {
6250 *sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6253 *highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6258 * Check/set reclaim complete for this session/clientid.
6261 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6263 struct nfsdsession *sep;
6264 struct nfssessionhash *shp;
6267 shp = NFSSESSIONHASH(nd->nd_sessionid);
6269 NFSLOCKSESSION(shp);
6270 sep = nfsrv_findsession(nd->nd_sessionid);
6272 NFSUNLOCKSESSION(shp);
6274 return (NFSERR_BADSESSION);
6278 sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6279 /* Check to see if reclaim complete has already happened. */
6280 else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6281 error = NFSERR_COMPLETEALREADY;
6283 sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6284 nfsrv_markreclaim(sep->sess_clp);
6286 NFSUNLOCKSESSION(shp);
6292 * Cache the reply in a session slot.
6295 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6297 struct nfsdsession *sep;
6298 struct nfssessionhash *shp;
6301 struct sockaddr_in *sin;
6304 struct sockaddr_in6 *sin6;
6307 shp = NFSSESSIONHASH(nd->nd_sessionid);
6308 NFSLOCKSESSION(shp);
6309 sep = nfsrv_findsession(nd->nd_sessionid);
6311 NFSUNLOCKSESSION(shp);
6312 if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
6313 buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6314 switch (nd->nd_nam->sa_family) {
6317 sin = (struct sockaddr_in *)nd->nd_nam;
6318 cp = inet_ntop(sin->sin_family,
6319 &sin->sin_addr.s_addr, buf,
6325 sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6326 cp = inet_ntop(sin6->sin6_family,
6327 &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6334 printf("nfsrv_cache_session: no session "
6337 printf("nfsrv_cache_session: no session\n");
6343 nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6345 NFSUNLOCKSESSION(shp);
6349 * Search for a session that matches the sessionid.
6351 static struct nfsdsession *
6352 nfsrv_findsession(uint8_t *sessionid)
6354 struct nfsdsession *sep;
6355 struct nfssessionhash *shp;
6357 shp = NFSSESSIONHASH(sessionid);
6358 LIST_FOREACH(sep, &shp->list, sess_hash) {
6359 if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6366 * Destroy a session.
6369 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6371 int error, igotlock, samesess;
6374 if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6375 (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6377 if ((nd->nd_flag & ND_LASTOP) == 0)
6378 return (NFSERR_BADSESSION);
6381 /* Lock out other nfsd threads */
6382 NFSLOCKV4ROOTMUTEX();
6383 nfsv4_relref(&nfsv4rootfs_lock);
6385 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6386 NFSV4ROOTLOCKMUTEXPTR, NULL);
6387 } while (igotlock == 0);
6388 NFSUNLOCKV4ROOTMUTEX();
6390 error = nfsrv_freesession(NULL, sessionid);
6391 if (error == 0 && samesess != 0)
6392 nd->nd_flag &= ~ND_HASSEQUENCE;
6394 NFSLOCKV4ROOTMUTEX();
6395 nfsv4_unlock(&nfsv4rootfs_lock, 1);
6396 NFSUNLOCKV4ROOTMUTEX();
6401 * Bind a connection to a session.
6402 * For now, only certain variants are supported, since the current session
6403 * structure can only handle a single backchannel entry, which will be
6404 * applied to all connections if it is set.
6407 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6409 struct nfssessionhash *shp;
6410 struct nfsdsession *sep;
6411 struct nfsclient *clp;
6416 shp = NFSSESSIONHASH(sessionid);
6418 NFSLOCKSESSION(shp);
6419 sep = nfsrv_findsession(sessionid);
6421 clp = sep->sess_clp;
6422 if (*foreaftp == NFSCDFC4_BACK ||
6423 *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6424 *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6425 /* Try to set up a backchannel. */
6426 if (clp->lc_req.nr_client == NULL) {
6427 NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6429 clp->lc_req.nr_client = (struct __rpc_client *)
6430 clnt_bck_create(nd->nd_xprt->xp_socket,
6431 sep->sess_cbprogram, NFSV4_CBVERS);
6433 if (clp->lc_req.nr_client != NULL) {
6434 NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6436 savxprt = sep->sess_cbsess.nfsess_xprt;
6437 SVC_ACQUIRE(nd->nd_xprt);
6438 nd->nd_xprt->xp_p2 =
6439 clp->lc_req.nr_client->cl_private;
6440 /* Disable idle timeout. */
6441 nd->nd_xprt->xp_idletimeout = 0;
6442 sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6443 if (savxprt != NULL)
6444 SVC_RELEASE(savxprt);
6445 sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6446 clp->lc_flags |= LCL_DONEBINDCONN;
6447 if (*foreaftp == NFSCDFS4_BACK)
6448 *foreaftp = NFSCDFS4_BACK;
6450 *foreaftp = NFSCDFS4_BOTH;
6451 } else if (*foreaftp != NFSCDFC4_BACK) {
6452 NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6453 "up backchannel\n");
6454 sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6455 clp->lc_flags |= LCL_DONEBINDCONN;
6456 *foreaftp = NFSCDFS4_FORE;
6458 error = NFSERR_NOTSUPP;
6459 printf("nfsrv_bindconnsess: Can't add "
6463 NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6464 clp->lc_flags |= LCL_DONEBINDCONN;
6465 *foreaftp = NFSCDFS4_FORE;
6468 error = NFSERR_BADSESSION;
6469 NFSUNLOCKSESSION(shp);
6475 * Free up a session structure.
6478 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6480 struct nfssessionhash *shp;
6485 shp = NFSSESSIONHASH(sessionid);
6486 NFSLOCKSESSION(shp);
6487 sep = nfsrv_findsession(sessionid);
6489 shp = NFSSESSIONHASH(sep->sess_sessionid);
6490 NFSLOCKSESSION(shp);
6494 if (sep->sess_refcnt > 0) {
6495 NFSUNLOCKSESSION(shp);
6497 return (NFSERR_BACKCHANBUSY);
6499 LIST_REMOVE(sep, sess_hash);
6500 LIST_REMOVE(sep, sess_list);
6502 NFSUNLOCKSESSION(shp);
6505 return (NFSERR_BADSESSION);
6506 for (i = 0; i < NFSV4_SLOTS; i++)
6507 if (sep->sess_slots[i].nfssl_reply != NULL)
6508 m_freem(sep->sess_slots[i].nfssl_reply);
6509 if (sep->sess_cbsess.nfsess_xprt != NULL)
6510 SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6511 free(sep, M_NFSDSESSION);
6517 * RFC5661 says that it should fail when there are associated opens, locks
6518 * or delegations. Since stateids represent opens, I don't see how you can
6519 * free an open stateid (it will be free'd when closed), so this function
6520 * only works for lock stateids (freeing the lock_owner) or delegations.
6523 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6526 struct nfsclient *clp;
6527 struct nfsstate *stp;
6532 * Look up the stateid
6534 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6535 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6537 /* First, check for a delegation. */
6538 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6539 if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6544 nfsrv_freedeleg(stp);
6549 /* Not a delegation, try for a lock_owner. */
6551 error = nfsrv_getstate(clp, stateidp, 0, &stp);
6552 if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6553 NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6554 /* Not a lock_owner stateid. */
6555 error = NFSERR_LOCKSHELD;
6556 if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6557 error = NFSERR_LOCKSHELD;
6559 nfsrv_freelockowner(stp, NULL, 0, p);
6568 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6571 struct nfsclient *clp;
6572 struct nfsstate *stp;
6577 * Look up the stateid
6579 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6580 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6582 error = nfsrv_getstate(clp, stateidp, 0, &stp);
6583 if (error == 0 && stateidp->seqid != 0 &&
6584 SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6585 error = NFSERR_OLDSTATEID;
6591 * Generate the xdr for an NFSv4.1 CBSequence Operation.
6594 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6595 int dont_replycache, struct nfsdsession **sepp)
6597 struct nfsdsession *sep;
6598 uint32_t *tl, slotseq = 0;
6599 int maxslot, slotpos;
6600 uint8_t sessionid[NFSX_V4SESSIONID];
6603 error = nfsv4_getcbsession(clp, sepp);
6607 (void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6608 &slotseq, sessionid);
6609 KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6611 /* Build the Sequence arguments. */
6612 NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6613 bcopy(sessionid, tl, NFSX_V4SESSIONID);
6614 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6615 nd->nd_slotseq = tl;
6616 *tl++ = txdr_unsigned(slotseq);
6617 *tl++ = txdr_unsigned(slotpos);
6618 *tl++ = txdr_unsigned(maxslot);
6619 if (dont_replycache == 0)
6620 *tl++ = newnfs_true;
6622 *tl++ = newnfs_false;
6623 *tl = 0; /* No referring call list, for now. */
6624 nd->nd_flag |= ND_HASSEQUENCE;
6629 * Get a session for the callback.
6632 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6634 struct nfsdsession *sep;
6637 LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6638 if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6643 return (NFSERR_BADSESSION);
6652 * Free up all backchannel xprts. This needs to be done when the nfsd threads
6653 * exit, since those transports will all be going away.
6654 * This is only called after all the nfsd threads are done performing RPCs,
6655 * so locking shouldn't be an issue.
6658 nfsrv_freeallbackchannel_xprts(void)
6660 struct nfsdsession *sep;
6661 struct nfsclient *clp;
6665 for (i = 0; i < nfsrv_clienthashsize; i++) {
6666 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6667 LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6668 xprt = sep->sess_cbsess.nfsess_xprt;
6669 sep->sess_cbsess.nfsess_xprt = NULL;
6678 * Do a layout commit. Actually just call nfsrv_updatemdsattr().
6679 * I have no idea if the rest of these arguments will ever be useful?
6682 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6683 int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6684 int hasnewmtime, struct timespec *newmtimep, int reclaim,
6685 nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6686 uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6691 error = nfsrv_updatemdsattr(vp, &na, p);
6694 *newsizep = na.na_size;
6700 * Try and get a layout.
6703 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6704 int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6705 uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6706 int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6708 struct nfslayouthash *lhyp;
6709 struct nfslayout *lyp;
6711 fhandle_t fh, *dsfhp;
6712 int error, mirrorcnt;
6714 if (nfsrv_devidcnt == 0)
6715 return (NFSERR_UNKNLAYOUTTYPE);
6718 printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6720 error = nfsvno_getfh(vp, &fh, p);
6721 NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6726 * For now, all layouts are for entire files.
6727 * Only issue Read/Write layouts if requested for a non-readonly fs.
6729 if (NFSVNO_EXRDONLY(exp)) {
6730 if (*iomode == NFSLAYOUTIOMODE_RW)
6731 return (NFSERR_LAYOUTTRYLATER);
6732 *iomode = NFSLAYOUTIOMODE_READ;
6734 if (*iomode != NFSLAYOUTIOMODE_RW)
6735 *iomode = NFSLAYOUTIOMODE_READ;
6738 * Check to see if a write layout can be issued for this file.
6739 * This is used during mirror recovery to avoid RW layouts being
6740 * issued for a file while it is being copied to the recovered
6743 if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6744 return (NFSERR_LAYOUTTRYLATER);
6750 /* First, see if a layout already exists and return if found. */
6751 lhyp = NFSLAYOUTHASH(&fh);
6752 NFSLOCKLAYOUT(lhyp);
6753 error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6754 NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6756 * Not sure if the seqid must be the same, so I won't check it.
6758 if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6759 stateidp->other[1] != lyp->lay_stateid.other[1] ||
6760 stateidp->other[2] != lyp->lay_stateid.other[2])) {
6761 if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6762 NFSUNLOCKLAYOUT(lhyp);
6763 NFSD_DEBUG(1, "ret bad stateid\n");
6764 return (NFSERR_BADSTATEID);
6767 * I believe we get here because there is a race between
6768 * the client processing the CBLAYOUTRECALL and the layout
6769 * being deleted here on the server.
6770 * The client has now done a LayoutGet with a non-layout
6771 * stateid, as it would when there is no layout.
6772 * As such, free this layout and set error == NFSERR_BADSTATEID
6773 * so the code below will create a new layout structure as
6774 * would happen if no layout was found.
6775 * "lyp" will be set before being used below, but set it NULL
6778 nfsrv_freelayout(&lhyp->list, lyp);
6780 error = NFSERR_BADSTATEID;
6783 if (lyp->lay_layoutlen > maxcnt) {
6784 NFSUNLOCKLAYOUT(lhyp);
6785 NFSD_DEBUG(1, "ret layout too small\n");
6786 return (NFSERR_TOOSMALL);
6788 if (*iomode == NFSLAYOUTIOMODE_RW)
6789 lyp->lay_flags |= NFSLAY_RW;
6791 lyp->lay_flags |= NFSLAY_READ;
6792 NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6793 *layoutlenp = lyp->lay_layoutlen;
6794 if (++lyp->lay_stateid.seqid == 0)
6795 lyp->lay_stateid.seqid = 1;
6796 stateidp->seqid = lyp->lay_stateid.seqid;
6797 NFSUNLOCKLAYOUT(lhyp);
6798 NFSD_DEBUG(4, "ret fnd layout\n");
6801 NFSUNLOCKLAYOUT(lhyp);
6803 /* Find the device id and file handle. */
6804 dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6805 devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6806 error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6807 NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6809 if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6810 if (NFSX_V4FILELAYOUT > maxcnt)
6811 error = NFSERR_TOOSMALL;
6813 lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6814 devid, vp->v_mount->mnt_stat.f_fsid);
6816 if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6817 error = NFSERR_TOOSMALL;
6819 lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6821 vp->v_mount->mnt_stat.f_fsid);
6824 free(dsfhp, M_TEMP);
6825 free(devid, M_TEMP);
6830 * Now, add this layout to the list.
6832 error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6833 NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6835 * The lyp will be set to NULL by nfsrv_addlayout() if it
6836 * linked the new structure into the lists.
6838 free(lyp, M_NFSDSTATE);
6843 * Generate a File Layout.
6845 static struct nfslayout *
6846 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6847 fhandle_t *dsfhp, char *devid, fsid_t fs)
6850 struct nfslayout *lyp;
6851 uint64_t pattern_offset;
6853 lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6855 lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6856 if (iomode == NFSLAYOUTIOMODE_RW)
6857 lyp->lay_flags = NFSLAY_RW;
6859 lyp->lay_flags = NFSLAY_READ;
6860 NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6861 lyp->lay_clientid.qval = nd->nd_clientid.qval;
6864 /* Fill in the xdr for the files layout. */
6865 tl = (uint32_t *)lyp->lay_xdr;
6866 NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */
6867 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6870 * Make the stripe size as many 64K blocks as will fit in the stripe
6871 * mask. Since there is only one stripe, the stripe size doesn't really
6872 * matter, except that the Linux client will only handle an exact
6873 * multiple of their PAGE_SIZE (usually 4K). I chose 64K as a value
6874 * that should cover most/all arches w.r.t. PAGE_SIZE.
6876 *tl++ = txdr_unsigned(NFSFLAYUTIL_STRIPE_MASK & ~0xffff);
6877 *tl++ = 0; /* 1st stripe index. */
6879 txdr_hyper(pattern_offset, tl); tl += 2; /* Pattern offset. */
6880 *tl++ = txdr_unsigned(1); /* 1 file handle. */
6881 *tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6882 NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6883 lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6887 #define FLEX_OWNERID "999"
6888 #define FLEX_UID0 "0"
6890 * Generate a Flex File Layout.
6891 * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6892 * string goes on the wire, it isn't supposed to be used by the client,
6893 * since this server uses tight coupling.
6894 * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6895 * a string of "0". This works around the Linux Flex File Layout driver bug
6896 * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6898 static struct nfslayout *
6899 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6900 fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
6903 struct nfslayout *lyp;
6907 lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
6908 M_NFSDSTATE, M_WAITOK | M_ZERO);
6909 lyp->lay_type = NFSLAYOUT_FLEXFILE;
6910 if (iomode == NFSLAYOUTIOMODE_RW)
6911 lyp->lay_flags = NFSLAY_RW;
6913 lyp->lay_flags = NFSLAY_READ;
6914 NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6915 lyp->lay_clientid.qval = nd->nd_clientid.qval;
6917 lyp->lay_mirrorcnt = mirrorcnt;
6919 /* Fill in the xdr for the files layout. */
6920 tl = (uint32_t *)lyp->lay_xdr;
6922 txdr_hyper(lenval, tl); tl += 2; /* Stripe unit. */
6923 *tl++ = txdr_unsigned(mirrorcnt); /* # of mirrors. */
6924 for (i = 0; i < mirrorcnt; i++) {
6925 *tl++ = txdr_unsigned(1); /* One stripe. */
6926 NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */
6927 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6928 devid += NFSX_V4DEVICEID;
6929 *tl++ = txdr_unsigned(1); /* Efficiency. */
6930 *tl++ = 0; /* Proxy Stateid. */
6934 *tl++ = txdr_unsigned(1); /* 1 file handle. */
6935 *tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6936 NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6937 tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
6939 if (nfsrv_flexlinuxhack != 0) {
6940 *tl++ = txdr_unsigned(strlen(FLEX_UID0));
6941 *tl = 0; /* 0 pad string. */
6942 NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6943 *tl++ = txdr_unsigned(strlen(FLEX_UID0));
6944 *tl = 0; /* 0 pad string. */
6945 NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6947 *tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6948 NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6949 *tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6950 NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6953 *tl++ = txdr_unsigned(0); /* ff_flags. */
6954 *tl = txdr_unsigned(60); /* Status interval hint. */
6955 lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
6960 * Parse and process Flex File errors returned via LayoutReturn.
6963 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
6967 int cnt, errcnt, i, j, opnum, stat;
6968 char devid[NFSX_V4DEVICEID];
6971 cnt = fxdr_unsigned(int, *tl++);
6972 NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
6973 for (i = 0; i < cnt; i++) {
6974 /* Skip offset, length and stateid for now. */
6975 tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
6976 errcnt = fxdr_unsigned(int, *tl++);
6977 NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
6978 for (j = 0; j < errcnt; j++) {
6979 NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
6980 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6981 stat = fxdr_unsigned(int, *tl++);
6982 opnum = fxdr_unsigned(int, *tl++);
6983 NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
6986 * Except for NFSERR_ACCES and NFSERR_STALE errors,
6987 * disable the mirror.
6989 if (stat != NFSERR_ACCES && stat != NFSERR_STALE)
6990 nfsrv_delds(devid, p);
6996 * This function removes all flex file layouts which has a mirror with
6997 * a device id that matches the argument.
6998 * Called when the DS represented by the device id has failed.
7001 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
7004 struct nfslayout *lyp, *nlyp;
7005 struct nfslayouthash *lhyp;
7006 struct nfslayouthead loclyp;
7009 NFSD_DEBUG(4, "flexmirrordel\n");
7010 /* Move all layouts found onto a local list. */
7011 TAILQ_INIT(&loclyp);
7012 for (i = 0; i < nfsrv_layouthashsize; i++) {
7013 lhyp = &nfslayouthash[i];
7014 NFSLOCKLAYOUT(lhyp);
7015 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7016 if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
7017 lyp->lay_mirrorcnt > 1) {
7018 NFSD_DEBUG(4, "possible match\n");
7021 for (j = 0; j < lyp->lay_mirrorcnt; j++) {
7023 if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
7026 NFSD_DEBUG(4, "fnd one\n");
7027 TAILQ_REMOVE(&lhyp->list, lyp,
7029 TAILQ_INSERT_HEAD(&loclyp, lyp,
7033 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7034 NFSM_RNDUP(NFSX_V4PNFSFH) /
7035 NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7039 NFSUNLOCKLAYOUT(lhyp);
7042 /* Now, try to do a Layout recall for each one found. */
7043 TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7044 NFSD_DEBUG(4, "do layout recall\n");
7046 * The layout stateid.seqid needs to be incremented
7047 * before doing a LAYOUT_RECALL callback.
7049 if (++lyp->lay_stateid.seqid == 0)
7050 lyp->lay_stateid.seqid = 1;
7051 nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7052 &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7053 nfsrv_freelayout(&loclyp, lyp);
7058 * Do a recall callback to the client for this layout.
7061 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7062 struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7064 struct nfsclient *clp;
7067 NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7068 error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7070 NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7072 printf("nfsrv_recalllayout: getclient err=%d\n", error);
7075 if ((clp->lc_flags & LCL_NFSV41) != 0) {
7076 error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7077 stateidp, changed, fhp, NULL, NULL, laytype, p);
7078 /* If lyp != NULL, handle an error return here. */
7079 if (error != 0 && lyp != NULL) {
7082 * Mark it returned, since no layout recall
7084 * All errors seem to be non-recoverable, although
7085 * NFSERR_NOMATCHLAYOUT is a normal event.
7087 if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7088 lyp->lay_flags |= NFSLAY_RETURNED;
7092 if (error != NFSERR_NOMATCHLAYOUT)
7093 printf("nfsrv_recalllayout: err=%d\n", error);
7096 printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7101 * Find a layout to recall when we exceed our high water mark.
7104 nfsrv_recalloldlayout(NFSPROC_T *p)
7106 struct nfslayouthash *lhyp;
7107 struct nfslayout *lyp;
7109 nfsv4stateid_t stateid;
7111 int error, laytype = 0, ret;
7113 lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7114 NFSLOCKLAYOUT(lhyp);
7115 TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7116 if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7117 lyp->lay_flags |= NFSLAY_CALLB;
7119 * The layout stateid.seqid needs to be incremented
7120 * before doing a LAYOUT_RECALL callback.
7122 if (++lyp->lay_stateid.seqid == 0)
7123 lyp->lay_stateid.seqid = 1;
7124 clientid = lyp->lay_clientid;
7125 stateid = lyp->lay_stateid;
7126 NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7127 laytype = lyp->lay_type;
7131 NFSUNLOCKLAYOUT(lhyp);
7133 error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7135 if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7136 NFSD_DEBUG(4, "recallold=%d\n", error);
7138 NFSLOCKLAYOUT(lhyp);
7140 * Since the hash list was unlocked, we need to
7143 ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7146 (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7147 lyp->lay_stateid.other[0] == stateid.other[0] &&
7148 lyp->lay_stateid.other[1] == stateid.other[1] &&
7149 lyp->lay_stateid.other[2] == stateid.other[2]) {
7151 * The client no longer knows this layout, so
7152 * it can be free'd now.
7154 if (error == NFSERR_NOMATCHLAYOUT)
7155 nfsrv_freelayout(&lhyp->list, lyp);
7158 * Leave it to be tried later by
7159 * clearing NFSLAY_CALLB and moving
7160 * it to the head of the list, so it
7161 * won't be tried again for a while.
7163 lyp->lay_flags &= ~NFSLAY_CALLB;
7164 TAILQ_REMOVE(&lhyp->list, lyp,
7166 TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7170 NFSUNLOCKLAYOUT(lhyp);
7176 * Try and return layout(s).
7179 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7180 int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7181 int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7182 struct ucred *cred, NFSPROC_T *p)
7185 struct nfslayouthash *lhyp;
7186 struct nfslayout *lyp;
7191 if (kind == NFSV4LAYOUTRET_FILE) {
7192 error = nfsvno_getfh(vp, &fh, p);
7194 error = nfsrv_updatemdsattr(vp, &na, p);
7196 printf("nfsrv_layoutreturn: updatemdsattr"
7197 " failed=%d\n", error);
7200 if (reclaim == newnfs_true) {
7201 error = nfsrv_checkgrace(NULL, NULL,
7203 if (error != NFSERR_NOGRACE)
7207 lhyp = NFSLAYOUTHASH(&fh);
7209 NFSLOCKLAYOUT(lhyp);
7210 error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7211 layouttype, p, &lyp);
7212 NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7214 stateidp->other[0] == lyp->lay_stateid.other[0] &&
7215 stateidp->other[1] == lyp->lay_stateid.other[1] &&
7216 stateidp->other[2] == lyp->lay_stateid.other[2]) {
7217 NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7218 " %x %x %x laystateid %d %x %x %x"
7219 " off=%ju len=%ju flgs=0x%x\n",
7220 stateidp->seqid, stateidp->other[0],
7221 stateidp->other[1], stateidp->other[2],
7222 lyp->lay_stateid.seqid,
7223 lyp->lay_stateid.other[0],
7224 lyp->lay_stateid.other[1],
7225 lyp->lay_stateid.other[2],
7226 (uintmax_t)offset, (uintmax_t)len,
7228 if (++lyp->lay_stateid.seqid == 0)
7229 lyp->lay_stateid.seqid = 1;
7230 stateidp->seqid = lyp->lay_stateid.seqid;
7231 if (offset == 0 && len == UINT64_MAX) {
7232 if ((iomode & NFSLAYOUTIOMODE_READ) !=
7234 lyp->lay_flags &= ~NFSLAY_READ;
7235 if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7236 lyp->lay_flags &= ~NFSLAY_RW;
7237 if ((lyp->lay_flags & (NFSLAY_READ |
7239 nfsrv_freelayout(&lhyp->list,
7246 NFSUNLOCKLAYOUT(lhyp);
7247 /* Search the nfsrv_recalllist for a match. */
7248 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7249 if (NFSBCMP(&lyp->lay_fh, &fh,
7251 lyp->lay_clientid.qval ==
7252 nd->nd_clientid.qval &&
7253 stateidp->other[0] ==
7254 lyp->lay_stateid.other[0] &&
7255 stateidp->other[1] ==
7256 lyp->lay_stateid.other[1] &&
7257 stateidp->other[2] ==
7258 lyp->lay_stateid.other[2]) {
7259 lyp->lay_flags |= NFSLAY_RETURNED;
7266 if (layouttype == NFSLAYOUT_FLEXFILE)
7267 nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7268 } else if (kind == NFSV4LAYOUTRET_FSID)
7269 nfsrv_freelayouts(&nd->nd_clientid,
7270 &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7271 else if (kind == NFSV4LAYOUTRET_ALL)
7272 nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7274 error = NFSERR_INVAL;
7281 * Look for an existing layout.
7284 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7285 NFSPROC_T *p, struct nfslayout **lypp)
7287 struct nfslayouthash *lhyp;
7288 struct nfslayout *lyp;
7293 lhyp = NFSLAYOUTHASH(fhp);
7294 TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7295 if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7296 lyp->lay_clientid.qval == clientidp->qval &&
7297 lyp->lay_type == laytype)
7308 * Add the new layout, as required.
7311 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7312 nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7314 struct nfsclient *clp;
7315 struct nfslayouthash *lhyp;
7316 struct nfslayout *lyp, *nlyp;
7320 KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7321 ("nfsrv_layoutget: no nd_clientid\n"));
7325 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7326 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7331 lyp->lay_stateid.seqid = stateidp->seqid = 1;
7332 lyp->lay_stateid.other[0] = stateidp->other[0] =
7333 clp->lc_clientid.lval[0];
7334 lyp->lay_stateid.other[1] = stateidp->other[1] =
7335 clp->lc_clientid.lval[1];
7336 lyp->lay_stateid.other[2] = stateidp->other[2] =
7337 nfsrv_nextstateindex(clp);
7340 lhyp = NFSLAYOUTHASH(fhp);
7341 NFSLOCKLAYOUT(lhyp);
7342 TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7343 if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7344 nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7348 /* A layout already exists, so use it. */
7349 nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7350 NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7351 *layoutlenp = nlyp->lay_layoutlen;
7352 if (++nlyp->lay_stateid.seqid == 0)
7353 nlyp->lay_stateid.seqid = 1;
7354 stateidp->seqid = nlyp->lay_stateid.seqid;
7355 stateidp->other[0] = nlyp->lay_stateid.other[0];
7356 stateidp->other[1] = nlyp->lay_stateid.other[1];
7357 stateidp->other[2] = nlyp->lay_stateid.other[2];
7358 NFSUNLOCKLAYOUT(lhyp);
7362 /* Insert the new layout in the lists. */
7364 atomic_add_int(&nfsrv_layoutcnt, 1);
7365 NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7366 *layoutlenp = lyp->lay_layoutlen;
7367 TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7368 NFSUNLOCKLAYOUT(lhyp);
7373 * Get the devinfo for a deviceid.
7376 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7377 uint32_t *notify, int *devaddrlen, char **devaddr)
7379 struct nfsdevice *ds;
7381 if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7382 NFSLAYOUT_FLEXFILE) ||
7383 (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7384 return (NFSERR_UNKNLAYOUTTYPE);
7387 * Now, search for the device id. Note that the structures won't go
7388 * away, but the order changes in the list. As such, the lock only
7389 * needs to be held during the search through the list.
7392 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7393 if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7394 ds->nfsdev_nmp != NULL)
7399 return (NFSERR_NOENT);
7401 /* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7403 if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7404 *devaddrlen = ds->nfsdev_fileaddrlen;
7405 *devaddr = ds->nfsdev_fileaddr;
7406 } else if (layouttype == NFSLAYOUT_FLEXFILE) {
7407 *devaddrlen = ds->nfsdev_flexaddrlen;
7408 *devaddr = ds->nfsdev_flexaddr;
7410 if (*devaddrlen == 0)
7411 return (NFSERR_UNKNLAYOUTTYPE);
7414 * The XDR overhead is 3 unsigned values: layout_type,
7415 * length_of_address and notify bitmap.
7416 * If the notify array is changed to not all zeros, the
7417 * count of unsigned values must be increased.
7419 if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7420 3 * NFSX_UNSIGNED) {
7421 *maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7422 return (NFSERR_TOOSMALL);
7428 * Free a list of layout state structures.
7431 nfsrv_freelayoutlist(nfsquad_t clientid)
7433 struct nfslayouthash *lhyp;
7434 struct nfslayout *lyp, *nlyp;
7437 for (i = 0; i < nfsrv_layouthashsize; i++) {
7438 lhyp = &nfslayouthash[i];
7439 NFSLOCKLAYOUT(lhyp);
7440 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7441 if (lyp->lay_clientid.qval == clientid.qval)
7442 nfsrv_freelayout(&lhyp->list, lyp);
7444 NFSUNLOCKLAYOUT(lhyp);
7452 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7455 NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7456 atomic_add_int(&nfsrv_layoutcnt, -1);
7457 TAILQ_REMOVE(lhp, lyp, lay_list);
7458 free(lyp, M_NFSDSTATE);
7462 * Free up a device id.
7465 nfsrv_freeonedevid(struct nfsdevice *ds)
7469 atomic_add_int(&nfsrv_devidcnt, -1);
7470 vrele(ds->nfsdev_dvp);
7471 for (i = 0; i < nfsrv_dsdirsize; i++)
7472 if (ds->nfsdev_dsdir[i] != NULL)
7473 vrele(ds->nfsdev_dsdir[i]);
7474 free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7475 free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7476 free(ds->nfsdev_host, M_NFSDSTATE);
7477 free(ds, M_NFSDSTATE);
7481 * Free up a device id and its mirrors.
7484 nfsrv_freedevid(struct nfsdevice *ds)
7487 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7488 nfsrv_freeonedevid(ds);
7492 * Free all layouts and device ids.
7493 * Done when the nfsd threads are shut down since there may be a new
7494 * modified device id list created when the nfsd is restarted.
7497 nfsrv_freealllayoutsanddevids(void)
7499 struct nfsdontlist *mrp, *nmrp;
7500 struct nfslayout *lyp, *nlyp;
7502 /* Get rid of the deviceid structures. */
7503 nfsrv_freealldevids();
7504 TAILQ_INIT(&nfsrv_devidhead);
7507 /* Get rid of all layouts. */
7508 nfsrv_freealllayouts();
7510 /* Get rid of any nfsdontlist entries. */
7511 LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7512 free(mrp, M_NFSDSTATE);
7513 LIST_INIT(&nfsrv_dontlisthead);
7514 nfsrv_dontlistlen = 0;
7516 /* Free layouts in the recall list. */
7517 TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7518 nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7519 TAILQ_INIT(&nfsrv_recalllisthead);
7523 * Free layouts that match the arguments.
7526 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7528 struct nfslayouthash *lhyp;
7529 struct nfslayout *lyp, *nlyp;
7532 for (i = 0; i < nfsrv_layouthashsize; i++) {
7533 lhyp = &nfslayouthash[i];
7534 NFSLOCKLAYOUT(lhyp);
7535 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7536 if (clid->qval != lyp->lay_clientid.qval)
7538 if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7540 if (laytype != lyp->lay_type)
7542 if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7543 lyp->lay_flags &= ~NFSLAY_READ;
7544 if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7545 lyp->lay_flags &= ~NFSLAY_RW;
7546 if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7547 nfsrv_freelayout(&lhyp->list, lyp);
7549 NFSUNLOCKLAYOUT(lhyp);
7554 * Free all layouts for the argument file.
7557 nfsrv_freefilelayouts(fhandle_t *fhp)
7559 struct nfslayouthash *lhyp;
7560 struct nfslayout *lyp, *nlyp;
7562 lhyp = NFSLAYOUTHASH(fhp);
7563 NFSLOCKLAYOUT(lhyp);
7564 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7565 if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7566 nfsrv_freelayout(&lhyp->list, lyp);
7568 NFSUNLOCKLAYOUT(lhyp);
7575 nfsrv_freealllayouts(void)
7577 struct nfslayouthash *lhyp;
7578 struct nfslayout *lyp, *nlyp;
7581 for (i = 0; i < nfsrv_layouthashsize; i++) {
7582 lhyp = &nfslayouthash[i];
7583 NFSLOCKLAYOUT(lhyp);
7584 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7585 nfsrv_freelayout(&lhyp->list, lyp);
7586 NFSUNLOCKLAYOUT(lhyp);
7591 * Look up the mount path for the DS server.
7594 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7595 struct nfsdevice **dsp)
7597 struct nameidata nd;
7598 struct nfsdevice *ds;
7604 NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7606 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7609 NFSD_DEBUG(4, "lookup=%d\n", error);
7612 if (nd.ni_vp->v_type != VDIR) {
7614 NFSD_DEBUG(4, "dspath not dir\n");
7617 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7619 NFSD_DEBUG(4, "dspath not an NFS mount\n");
7624 * Allocate a DS server structure with the NFS mounted directory
7625 * vnode reference counted, so that a non-forced dismount will
7627 * This structure is always linked into the list, even if an error
7628 * is being returned. The caller will free the entire list upon
7631 *dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7632 M_NFSDSTATE, M_WAITOK | M_ZERO);
7633 ds->nfsdev_dvp = nd.ni_vp;
7634 ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7635 NFSVOPUNLOCK(nd.ni_vp);
7637 dsdirsize = strlen(dspathp) + 16;
7638 dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7639 /* Now, create the DS directory structures. */
7640 for (i = 0; i < nfsrv_dsdirsize; i++) {
7641 snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7642 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7643 UIO_SYSSPACE, dsdirpath, p);
7645 NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7648 if (nd.ni_vp->v_type != VDIR) {
7651 NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7654 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7657 NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7660 ds->nfsdev_dsdir[i] = nd.ni_vp;
7661 NFSVOPUNLOCK(nd.ni_vp);
7663 free(dsdirpath, M_TEMP);
7665 if (strlen(mdspathp) > 0) {
7667 * This DS stores file for a specific MDS exported file
7670 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7671 UIO_SYSSPACE, mdspathp, p);
7673 NFSD_DEBUG(4, "mds lookup=%d\n", error);
7676 if (nd.ni_vp->v_type != VDIR) {
7679 NFSD_DEBUG(4, "mdspath not dir\n");
7682 mp = nd.ni_vp->v_mount;
7683 if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7686 NFSD_DEBUG(4, "mdspath not an exported fs\n");
7689 ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7690 ds->nfsdev_mdsisset = 1;
7695 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7696 atomic_add_int(&nfsrv_devidcnt, 1);
7701 * Look up the mount path for the DS server and delete it.
7704 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7707 struct nfsmount *nmp;
7708 struct nfsdevice *ds;
7711 NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7713 * Search for the path in the mount list. Avoid looking the path
7714 * up, since this mount point may be hung, with associated locked
7716 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7717 * until this completes.
7718 * As noted in the man page, this should be done before any forced
7719 * dismount on the mount point, but at least the handshake on
7720 * NFSMNTP_CANCELRPCS should make it safe.
7725 mtx_lock(&mountlist_mtx);
7726 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7727 if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7728 strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7729 mp->mnt_data != NULL) {
7732 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7733 NFSMNTP_CANCELRPCS)) == 0) {
7734 nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7743 mtx_unlock(&mountlist_mtx);
7746 ds = nfsrv_deldsnmp(op, nmp, p);
7747 NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7749 nfsrv_killrpcs(nmp);
7750 NFSD_DEBUG(4, "aft killrpcs\n");
7754 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7763 * Search for and remove a DS entry which matches the "nmp" argument.
7764 * The nfsdevice structure pointer is returned so that the caller can
7765 * free it via nfsrv_freeonedevid().
7766 * For the forced case, do not try to do LayoutRecalls, since the server
7767 * must be shut down now anyhow.
7770 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7772 struct nfsdevice *fndds;
7774 NFSD_DEBUG(4, "deldsdvp\n");
7776 if (op == PNFSDOP_FORCEDELDS)
7777 fndds = nfsv4_findmirror(nmp);
7779 fndds = nfsrv_findmirroredds(nmp);
7781 nfsrv_deleteds(fndds);
7783 if (fndds != NULL) {
7784 if (op != PNFSDOP_FORCEDELDS)
7785 nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7786 printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7792 * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7793 * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7795 * Also, returns an error instead of the nfsdevice found.
7798 nfsrv_delds(char *devid, NFSPROC_T *p)
7800 struct nfsdevice *ds, *fndds;
7801 struct nfsmount *nmp;
7804 NFSD_DEBUG(4, "delds\n");
7806 * Search the DS server list for a match with devid.
7807 * Remove the DS entry if found and there is a mirror.
7813 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7814 if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7815 ds->nfsdev_nmp != NULL) {
7816 NFSD_DEBUG(4, "fnd main ds\n");
7821 if (fndds == NULL) {
7825 if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7827 else if (fndds->nfsdev_mdsisset != 0) {
7828 /* For the fsid is set case, search for a mirror. */
7829 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7830 if (ds != fndds && ds->nfsdev_nmp != NULL &&
7831 ds->nfsdev_mdsisset != 0 &&
7832 fsidcmp(&ds->nfsdev_mdsfsid,
7833 &fndds->nfsdev_mdsfsid) == 0) {
7839 if (fndmirror != 0) {
7840 nmp = fndds->nfsdev_nmp;
7842 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7843 NFSMNTP_CANCELRPCS)) == 0) {
7844 nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7846 nfsrv_deleteds(fndds);
7854 nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7855 printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7856 nfsrv_killrpcs(nmp);
7858 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7867 * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7870 nfsrv_deleteds(struct nfsdevice *fndds)
7873 NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7874 fndds->nfsdev_nmp = NULL;
7875 if (fndds->nfsdev_mdsisset == 0)
7880 * Fill in the addr structures for the File and Flex File layouts.
7883 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
7888 static uint64_t new_devid = 0;
7890 if (strchr(addr, ':') != NULL)
7895 /* Fill in the device id. */
7896 NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
7898 NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
7902 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
7903 * as defined in RFC5661) in XDR.
7905 addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7907 NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
7908 ds->nfsdev_fileaddrlen = addrlen;
7909 tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7910 ds->nfsdev_fileaddr = (char *)tl;
7911 *tl++ = txdr_unsigned(1); /* One stripe with index 0. */
7913 *tl++ = txdr_unsigned(1); /* One multipath list */
7914 *tl++ = txdr_unsigned(1); /* with one entry in it. */
7915 /* The netaddr for this one entry. */
7916 *tl++ = txdr_unsigned(strlen(netprot));
7917 NFSBCOPY(netprot, tl, strlen(netprot));
7918 tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7919 *tl++ = txdr_unsigned(strlen(addr));
7920 NFSBCOPY(addr, tl, strlen(addr));
7923 * Fill in the flex file addr (actually the ff_device_addr4
7924 * as defined for Flexible File Layout) in XDR.
7926 addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7928 ds->nfsdev_flexaddrlen = addrlen;
7929 tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7930 ds->nfsdev_flexaddr = (char *)tl;
7931 *tl++ = txdr_unsigned(1); /* One multipath entry. */
7932 /* The netaddr for this one entry. */
7933 *tl++ = txdr_unsigned(strlen(netprot));
7934 NFSBCOPY(netprot, tl, strlen(netprot));
7935 tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7936 *tl++ = txdr_unsigned(strlen(addr));
7937 NFSBCOPY(addr, tl, strlen(addr));
7938 tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
7939 *tl++ = txdr_unsigned(2); /* Two NFS Versions. */
7940 *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */
7941 *tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
7942 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max rsize. */
7943 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max wsize. */
7944 *tl++ = newnfs_true; /* Tightly coupled. */
7945 *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */
7946 *tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
7947 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max rsize. */
7948 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max wsize. */
7949 *tl = newnfs_true; /* Tightly coupled. */
7951 ds->nfsdev_hostnamelen = strlen(dnshost);
7952 ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
7954 NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
7959 * Create the device id list.
7960 * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
7964 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
7966 struct nfsdevice *ds;
7967 char *addrp, *dnshostp, *dspathp, *mdspathp;
7971 dnshostp = args->dnshost;
7972 dspathp = args->dspath;
7973 mdspathp = args->mdspath;
7974 nfsrv_maxpnfsmirror = args->mirrorcnt;
7975 if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
7980 * Loop around for each nul-terminated string in args->addr,
7981 * args->dnshost, args->dnspath and args->mdspath.
7983 while (addrp < (args->addr + args->addrlen) &&
7984 dnshostp < (args->dnshost + args->dnshostlen) &&
7985 dspathp < (args->dspath + args->dspathlen) &&
7986 mdspathp < (args->mdspath + args->mdspathlen)) {
7987 error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
7989 /* Free all DS servers. */
7990 nfsrv_freealldevids();
7994 nfsrv_allocdevid(ds, addrp, dnshostp);
7995 addrp += (strlen(addrp) + 1);
7996 dnshostp += (strlen(dnshostp) + 1);
7997 dspathp += (strlen(dspathp) + 1);
7998 mdspathp += (strlen(mdspathp) + 1);
8000 if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
8001 /* Free all DS servers. */
8002 nfsrv_freealldevids();
8004 nfsrv_maxpnfsmirror = 1;
8007 /* We can fail at most one less DS than the mirror level. */
8008 nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
8011 * Allocate the nfslayout hash table now, since this is a pNFS server.
8012 * Make it 1% of the high water mark and at least 100.
8014 if (nfslayouthash == NULL) {
8015 nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
8016 if (nfsrv_layouthashsize < 100)
8017 nfsrv_layouthashsize = 100;
8018 nfslayouthash = mallocarray(nfsrv_layouthashsize,
8019 sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
8021 for (i = 0; i < nfsrv_layouthashsize; i++) {
8022 mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
8023 TAILQ_INIT(&nfslayouthash[i].list);
8030 * Free all device ids.
8033 nfsrv_freealldevids(void)
8035 struct nfsdevice *ds, *nds;
8037 TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8038 nfsrv_freedevid(ds);
8042 * Check to see if there is a Read/Write Layout plus either:
8043 * - A Write Delegation
8045 * - An Open with Write_access.
8046 * Return 1 if this is the case and 0 otherwise.
8047 * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8048 * Getattr RPC to the Data Server (DS) is necessary.
8050 #define NFSCLIDVECSIZE 6
8052 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8054 fhandle_t fh, *tfhp;
8055 struct nfsstate *stp;
8056 struct nfslayout *lyp;
8057 struct nfslayouthash *lhyp;
8058 struct nfslockhashhead *hp;
8059 struct nfslockfile *lfp;
8060 nfsquad_t clid[NFSCLIDVECSIZE];
8063 ret = nfsvno_getfh(vp, &fh, p);
8067 /* First check for a Read/Write Layout. */
8069 lhyp = NFSLAYOUTHASH(&fh);
8070 NFSLOCKLAYOUT(lhyp);
8071 TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8072 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8073 ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8074 ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8075 nfsrv_pnfsatime != 0))) {
8076 if (clidcnt < NFSCLIDVECSIZE)
8077 clid[clidcnt].qval = lyp->lay_clientid.qval;
8081 NFSUNLOCKLAYOUT(lhyp);
8083 /* None found, so return 0. */
8087 /* Get the nfslockfile for this fh. */
8089 hp = NFSLOCKHASH(&fh);
8090 LIST_FOREACH(lfp, hp, lf_hash) {
8092 if (NFSVNO_CMPFH(&fh, tfhp))
8096 /* None found, so return 0. */
8101 /* Now, look for a Write delegation for this clientid. */
8102 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8103 if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8104 nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8108 /* Found one, so return 1. */
8113 /* No Write delegation, so look for an Open with Write_access. */
8114 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8115 KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8116 ("nfsrv_checkdsattr: Non-open in Open list\n"));
8117 if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8118 nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8128 * Look for a matching clientid in the vector. Return 1 if one might match.
8131 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8135 /* If too many for the vector, return 1 since there might be a match. */
8136 if (clidcnt > NFSCLIDVECSIZE)
8139 for (i = 0; i < clidcnt; i++)
8140 if (clidvec[i].qval == clid.qval)
8146 * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8147 * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8150 nfsrv_dontlayout(fhandle_t *fhp)
8152 struct nfsdontlist *mrp;
8155 if (nfsrv_dontlistlen == 0)
8159 LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8160 if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8161 (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8166 NFSDDONTLISTUNLOCK();
8170 #define PNFSDS_COPYSIZ 65536
8172 * Create a new file on a DS and copy the contents of an extant DS file to it.
8173 * This can be used for recovery of a DS file onto a recovered DS.
8175 * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8176 * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8177 * so that they will be disabled after the MDS file's vnode is unlocked.
8178 * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8180 * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8181 * writes, LayoutCommits and LayoutReturns for the file when completing the
8182 * LayoutReturn requested by the LayoutRecall callback.
8183 * - Issue a LayoutRecall callback for all read/write layouts and wait for
8184 * them to be returned. (If the LayoutRecall callback replies
8185 * NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8186 * - Exclusively lock the MDS file's vnode. This ensures that no proxied
8187 * writes are in progress or can occur during the DS file copy.
8188 * It also blocks Setattr operations.
8189 * - Create the file on the recovered mirror.
8190 * - Copy the file from the operational DS.
8191 * - Copy any ACL from the MDS file to the new DS file.
8192 * - Set the modify time of the new DS file to that of the MDS file.
8193 * - Update the extended attribute for the MDS file.
8194 * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8195 * - The caller will unlock the MDS file's vnode allowing operations
8196 * to continue normally, since it is now on the mirror again.
8199 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8200 struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8201 struct ucred *cred, NFSPROC_T *p)
8203 struct nfsdontlist *mrp, *nmrp;
8204 struct nfslayouthash *lhyp;
8205 struct nfslayout *lyp, *nlyp;
8206 struct nfslayouthead thl;
8207 struct mount *mp, *tvmp;
8210 struct timespec mtime;
8216 int didprintf, ret, retacl, xfer;
8218 ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8219 ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8221 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8222 * so that no more RW layouts will get issued.
8224 ret = nfsvno_getfh(vp, &fh, p);
8226 NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8229 nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8230 nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8231 NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8233 LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8234 if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8238 LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8241 nfsrv_dontlistlen++;
8242 NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8244 NFSDDONTLISTUNLOCK();
8245 free(nmrp, M_NFSDSTATE);
8246 NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8249 NFSDDONTLISTUNLOCK();
8252 * Search for all RW layouts for this file. Move them to the
8253 * recall list, so they can be recalled and their return noted.
8255 lhyp = NFSLAYOUTHASH(&fh);
8257 NFSLOCKLAYOUT(lhyp);
8258 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8259 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8260 (lyp->lay_flags & NFSLAY_RW) != 0) {
8261 TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8262 TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8263 lyp->lay_trycnt = 0;
8266 NFSUNLOCKLAYOUT(lhyp);
8273 /* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8275 /* Now, do a recall for all layouts not yet recalled. */
8278 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8279 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8280 (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8281 lyp->lay_flags |= NFSLAY_RECALL;
8283 * The layout stateid.seqid needs to be incremented
8284 * before doing a LAYOUT_RECALL callback.
8286 if (++lyp->lay_stateid.seqid == 0)
8287 lyp->lay_stateid.seqid = 1;
8289 nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8290 &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8291 NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8296 /* Now wait for them to be returned. */
8298 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8299 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8300 if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8301 TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8303 TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8305 "nfsrv_copymr: layout returned\n");
8308 ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8309 PVFS | PCATCH, "nfsmrl", hz);
8310 NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8312 if (ret == EINTR || ret == ERESTART)
8314 if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8316 * Give up after 60sec and return
8317 * ENXIO, failing the copymr.
8318 * This layout will remain on the
8319 * recalllist. It can only be cleared
8320 * by restarting the nfsd.
8321 * This seems the safe way to handle
8322 * it, since it cannot be safely copied
8323 * with an outstanding RW layout.
8325 if (lyp->lay_trycnt >= 60) {
8329 if (didprintf == 0) {
8330 printf("nfsrv_copymr: layout "
8340 /* We can now get rid of the layouts that have been returned. */
8341 TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8342 nfsrv_freelayout(&thl, lyp);
8345 * Do the vn_start_write() calls here, before the MDS vnode is
8346 * locked and the tvp is created (locked) in the NFS file system
8348 * For tvmp, this probably isn't necessary, since it will be an
8349 * NFS mount and they are not suspendable at this time.
8352 ret = vn_start_write(vp, &mp, V_WAIT | PCATCH);
8354 tvmp = dvp->v_mount;
8355 ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH);
8359 * LK_EXCLUSIVE lock the MDS vnode, so that any
8360 * proxied writes through the MDS will be blocked until we have
8361 * completed the copy and update of the extended attributes.
8362 * This will also ensure that any attributes and ACL will not be
8363 * changed until the copy is complete.
8365 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8366 if (ret == 0 && VN_IS_DOOMED(vp)) {
8367 NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8371 /* Create the data file on the recovered DS. */
8373 ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8375 /* Copy the DS file, if created successfully. */
8378 * Get any NFSv4 ACL on the MDS file, so that it can be set
8379 * on the new DS file.
8381 aclp = acl_alloc(M_WAITOK | M_ZERO);
8382 retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8383 if (retacl != 0 && retacl != ENOATTR)
8384 NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8385 dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8386 /* Malloc a block of 0s used to check for holes. */
8387 if (nfsrv_zeropnfsdat == NULL)
8388 nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8391 ret = VOP_GETATTR(fvp, &va, cred);
8393 while (ret == 0 && aresid == 0) {
8394 ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8395 rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8397 xfer = PNFSDS_COPYSIZ - aresid;
8398 if (ret == 0 && xfer > 0) {
8401 * Skip the write for holes, except for the
8404 if (xfer < PNFSDS_COPYSIZ || rdpos ==
8405 va.va_size || NFSBCMP(dat,
8406 nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8407 ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8408 wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8409 cred, NULL, NULL, p);
8415 /* If there is an ACL and the copy succeeded, set the ACL. */
8416 if (ret == 0 && retacl == 0) {
8417 ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8419 * Don't consider these as errors, since VOP_GETACL()
8420 * can return an ACL when they are not actually
8421 * supported. For example, for UFS, VOP_GETACL()
8422 * will return a trivial ACL based on the uid/gid/mode
8423 * when there is no ACL on the file.
8424 * This case should be recognized as a trivial ACL
8425 * by UFS's VOP_SETACL() and succeed, but...
8427 if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8432 ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8434 /* Set the DS data file's modify time that of the MDS file. */
8436 ret = VOP_GETATTR(vp, &va, cred);
8438 mtime = va.va_mtime;
8440 va.va_mtime = mtime;
8441 ret = VOP_SETATTR(tvp, &va, cred);
8449 vn_finished_write(tvmp);
8451 /* Update the extended attributes for the newly created DS file. */
8453 ret = vn_extattr_set(vp, IO_NODELOCKED,
8454 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8455 sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8457 vn_finished_write(mp);
8459 /* Get rid of the dontlist entry, so that Layouts can be issued. */
8461 LIST_REMOVE(mrp, nfsmr_list);
8462 NFSDDONTLISTUNLOCK();
8463 free(mrp, M_NFSDSTATE);
8468 * Create a data storage file on the recovered DS.
8471 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8472 vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8475 struct vattr va, nva;
8478 /* Make data file name based on FH. */
8479 error = VOP_GETATTR(vp, &va, cred);
8481 /* Set the attributes for "vp" to Setattr the DS vp. */
8483 nva.va_uid = va.va_uid;
8484 nva.va_gid = va.va_gid;
8485 nva.va_mode = va.va_mode;
8489 va.va_mode = nva.va_mode;
8490 NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8491 error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8492 pf->dsf_filename, cred, p, tvpp);
8498 * Look up the MDS file shared locked, and then get the extended attribute
8499 * to find the extant DS file to be copied to the new mirror.
8500 * If successful, *vpp is set to the MDS file's vp and *nvpp is
8501 * set to a DS data file for the MDS file, both exclusively locked.
8502 * The "buf" argument has the pnfsdsfile structure from the MDS file
8503 * in it and buflen is set to its length.
8506 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8507 int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8508 struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8509 struct nfsdevice **fdsp)
8511 struct nameidata nd;
8512 struct vnode *vp, *curvp;
8513 struct pnfsdsfile *pf;
8514 struct nfsmount *nmp, *curnmp;
8515 int dsdir, error, mirrorcnt, ippos;
8522 if (dspathp == NULL && curdspathp != NULL)
8526 * Look up the MDS file shared locked. The lock will be upgraded
8527 * to an exclusive lock after any rw layouts have been returned.
8529 NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8530 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8533 NFSD_DEBUG(4, "lookup=%d\n", error);
8536 if (nd.ni_vp->v_type != VREG) {
8538 NFSD_DEBUG(4, "mdspath not reg\n");
8543 if (curdspathp != NULL) {
8545 * Look up the current DS path and find the nfsdev structure for
8548 NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8549 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8550 UIO_SYSSPACE, curdspathp, p);
8552 NFSD_DEBUG(4, "ds lookup=%d\n", error);
8557 if (nd.ni_vp->v_type != VDIR) {
8560 NFSD_DEBUG(4, "curdspath not dir\n");
8563 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8566 NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8569 curnmp = VFSTONFS(nd.ni_vp->v_mount);
8571 /* Search the nfsdev list for a match. */
8573 *fdsp = nfsv4_findmirror(curnmp);
8577 if (curnmp == NULL) {
8580 NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8586 if (dspathp != NULL) {
8587 /* Look up the nfsdev path and find the nfsdev structure. */
8588 NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8589 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8590 UIO_SYSSPACE, dspathp, p);
8592 NFSD_DEBUG(4, "ds lookup=%d\n", error);
8599 if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8604 NFSD_DEBUG(4, "dspath not dir\n");
8605 if (nd.ni_vp == curvp)
8609 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8614 NFSD_DEBUG(4, "dspath not an NFS mount\n");
8617 nmp = VFSTONFS(nd.ni_vp->v_mount);
8620 * Search the nfsdevice list for a match. If curnmp == NULL,
8621 * this is a recovery and there must be a mirror.
8625 *dsp = nfsrv_findmirroredds(nmp);
8627 *dsp = nfsv4_findmirror(nmp);
8634 NFSD_DEBUG(4, "mdscopymr: no ds\n");
8643 * Get a vp for an available DS data file using the extended
8644 * attribute on the MDS file.
8645 * If there is a valid entry for the new DS in the extended attribute
8646 * on the MDS file (as checked via the nmp argument),
8647 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8649 error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8650 NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8653 if (nd.ni_vp == NULL) {
8654 if (error == 0 && nmp != NULL) {
8655 /* Search the nfsdev list for a match. */
8657 *dsp = nfsrv_findmirroredds(nmp);
8660 if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8661 if (nvpp != NULL && *nvpp != NULL) {
8671 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8672 * and is only allowed if there is a 0.0.0.0 IP address entry.
8673 * When curdspathp != NULL, the ippos will be set to that entry.
8675 if (error == 0 && dspathp != NULL && ippos == -1) {
8676 if (nvpp != NULL && *nvpp != NULL) {
8685 pf = (struct pnfsdsfile *)buf;
8687 /* If no zeroip pnfsdsfile, add one. */
8688 ippos = *buflenp / sizeof(*pf);
8689 *buflenp += sizeof(*pf);
8691 pf->dsf_dir = dsdir;
8692 strlcpy(pf->dsf_filename, fname,
8693 sizeof(pf->dsf_filename));
8703 * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8704 * Return one if found, NULL otherwise.
8706 static struct nfsdevice *
8707 nfsrv_findmirroredds(struct nfsmount *nmp)
8709 struct nfsdevice *ds, *fndds;
8712 mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8714 * Search the DS server list for a match with nmp.
8715 * Remove the DS entry if found and there is a mirror.
8719 if (nfsrv_devidcnt == 0)
8721 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8722 if (ds->nfsdev_nmp == nmp) {
8723 NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8730 if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8732 else if (fndds->nfsdev_mdsisset != 0) {
8733 /* For the fsid is set case, search for a mirror. */
8734 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8735 if (ds != fndds && ds->nfsdev_nmp != NULL &&
8736 ds->nfsdev_mdsisset != 0 &&
8737 fsidcmp(&ds->nfsdev_mdsfsid,
8738 &fndds->nfsdev_mdsfsid) == 0) {
8744 if (fndmirror == 0) {
8745 NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");