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;
4428 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4429 cred = newnfs_getcred();
4430 NFSLOCKSTATE(); /* mostly for lc_cbref++ */
4431 if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4438 * Fill the callback program# and version into the request
4439 * structure for newnfs_connect() to use.
4441 clp->lc_req.nr_prog = clp->lc_program;
4443 if ((clp->lc_flags & LCL_NFSV41) != 0)
4444 clp->lc_req.nr_vers = NFSV41_CBVERS;
4447 clp->lc_req.nr_vers = NFSV4_CBVERS;
4450 * First, fill in some of the fields of nd and cr.
4452 nd->nd_flag = ND_NFSV4;
4453 if (clp->lc_flags & LCL_GSS)
4454 nd->nd_flag |= ND_KERBV;
4455 if ((clp->lc_flags & LCL_NFSV41) != 0)
4456 nd->nd_flag |= ND_NFSV41;
4457 if ((clp->lc_flags & LCL_NFSV42) != 0)
4458 nd->nd_flag |= ND_NFSV42;
4460 cred->cr_uid = clp->lc_uid;
4461 cred->cr_gid = clp->lc_gid;
4462 callback = clp->lc_callback;
4464 cred->cr_ngroups = 1;
4467 * Get the first mbuf for the request.
4469 MGET(m, M_WAITOK, MT_DATA);
4471 nd->nd_mreq = nd->nd_mb = m;
4472 nd->nd_bpos = mtod(m, caddr_t);
4475 * and build the callback request.
4477 if (procnum == NFSV4OP_CBGETATTR) {
4478 nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4479 error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4480 "CB Getattr", &sep);
4482 m_freem(nd->nd_mreq);
4485 (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4486 (void)nfsrv_putattrbit(nd, attrbitp);
4487 } else if (procnum == NFSV4OP_CBRECALL) {
4488 nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4489 error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4492 m_freem(nd->nd_mreq);
4495 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4496 *tl++ = txdr_unsigned(stateidp->seqid);
4497 NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4499 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4504 (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4505 } else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4506 NFSD_DEBUG(4, "docallback layout recall\n");
4507 nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4508 error = nfsrv_cbcallargs(nd, clp, callback,
4509 NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep);
4510 NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4512 m_freem(nd->nd_mreq);
4515 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4516 *tl++ = txdr_unsigned(laytype);
4517 *tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4519 *tl++ = newnfs_true;
4521 *tl++ = newnfs_false;
4522 *tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4523 nfsm_fhtom(nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4524 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4526 txdr_hyper(tval, tl); tl += 2;
4528 txdr_hyper(tval, tl); tl += 2;
4529 *tl++ = txdr_unsigned(stateidp->seqid);
4530 NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4531 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4532 NFSD_DEBUG(4, "aft args\n");
4533 } else if (procnum == NFSV4PROC_CBNULL) {
4534 nd->nd_procnum = NFSV4PROC_CBNULL;
4535 if ((clp->lc_flags & LCL_NFSV41) != 0) {
4536 error = nfsv4_getcbsession(clp, &sep);
4538 m_freem(nd->nd_mreq);
4543 error = NFSERR_SERVERFAULT;
4544 m_freem(nd->nd_mreq);
4549 * Call newnfs_connect(), as required, and then newnfs_request().
4552 if ((clp->lc_flags & LCL_TLSCB) != 0)
4554 (void) newnfs_sndlock(&clp->lc_req.nr_lock);
4555 if (clp->lc_req.nr_client == NULL) {
4556 if ((clp->lc_flags & LCL_NFSV41) != 0) {
4557 error = ECONNREFUSED;
4558 nfsrv_freesession(sep, NULL);
4559 } else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4560 error = newnfs_connect(NULL, &clp->lc_req, cred,
4563 error = newnfs_connect(NULL, &clp->lc_req, cred,
4566 newnfs_sndunlock(&clp->lc_req.nr_lock);
4567 NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4569 if ((nd->nd_flag & ND_NFSV41) != 0) {
4570 KASSERT(sep != NULL, ("sep NULL"));
4571 if (sep->sess_cbsess.nfsess_xprt != NULL)
4572 error = newnfs_request(nd, NULL, clp,
4573 &clp->lc_req, NULL, NULL, cred,
4574 clp->lc_program, clp->lc_req.nr_vers, NULL,
4575 1, NULL, &sep->sess_cbsess);
4578 * This should probably never occur, but if a
4579 * client somehow does an RPC without a
4580 * SequenceID Op that causes a callback just
4581 * after the nfsd threads have been terminated
4582 * and restared we could conceivably get here
4583 * without a backchannel xprt.
4585 printf("nfsrv_docallback: no xprt\n");
4586 error = ECONNREFUSED;
4588 NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4589 nfsrv_freesession(sep, NULL);
4591 error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4592 NULL, NULL, cred, clp->lc_program,
4593 clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4599 * If error is set here, the Callback path isn't working
4600 * properly, so twiddle the appropriate LCL_ flags.
4601 * (nd_repstat != 0 indicates the Callback path is working,
4602 * but the callback failed on the client.)
4606 * Mark the callback pathway down, which disabled issuing
4607 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4610 clp->lc_flags |= LCL_CBDOWN;
4614 * Callback worked. If the callback path was down, disable
4615 * callbacks, so no more delegations will be issued. (This
4616 * is done on the assumption that the callback pathway is
4620 if (clp->lc_flags & LCL_CBDOWN)
4621 clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4623 if (nd->nd_repstat) {
4624 error = nd->nd_repstat;
4625 NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4627 } else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4628 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4629 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4631 m_freem(nd->nd_mrep);
4635 if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4636 clp->lc_flags &= ~LCL_WAKEUPWANTED;
4647 * Set up the compound RPC for the callback.
4650 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4651 uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4656 len = strlen(optag);
4657 (void)nfsm_strtom(nd, optag, len);
4658 NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4659 if ((nd->nd_flag & ND_NFSV41) != 0) {
4660 if ((nd->nd_flag & ND_NFSV42) != 0)
4661 *tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4663 *tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4664 *tl++ = txdr_unsigned(callback);
4665 *tl++ = txdr_unsigned(2);
4666 *tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4667 error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4670 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4671 *tl = txdr_unsigned(op);
4673 *tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4674 *tl++ = txdr_unsigned(callback);
4675 *tl++ = txdr_unsigned(1);
4676 *tl = txdr_unsigned(op);
4682 * Return the next index# for a clientid. Mostly just increment and return
4683 * the next one, but... if the 32bit unsigned does actually wrap around,
4684 * it should be rebooted.
4685 * At an average rate of one new client per second, it will wrap around in
4686 * approximately 136 years. (I think the server will have been shut
4687 * down or rebooted before then.)
4690 nfsrv_nextclientindex(void)
4692 static u_int32_t client_index = 0;
4695 if (client_index != 0)
4696 return (client_index);
4698 printf("%s: out of clientids\n", __func__);
4699 return (client_index);
4703 * Return the next index# for a stateid. Mostly just increment and return
4704 * the next one, but... if the 32bit unsigned does actually wrap around
4705 * (will a BSD server stay up that long?), find
4706 * new start and end values.
4709 nfsrv_nextstateindex(struct nfsclient *clp)
4711 struct nfsstate *stp;
4713 u_int32_t canuse, min_index, max_index;
4715 if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4716 clp->lc_stateindex++;
4717 if (clp->lc_stateindex != clp->lc_statemaxindex)
4718 return (clp->lc_stateindex);
4722 * Yuck, we've hit the end.
4723 * Look for a new min and max.
4726 max_index = 0xffffffff;
4727 for (i = 0; i < nfsrv_statehashsize; i++) {
4728 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4729 if (stp->ls_stateid.other[2] > 0x80000000) {
4730 if (stp->ls_stateid.other[2] < max_index)
4731 max_index = stp->ls_stateid.other[2];
4733 if (stp->ls_stateid.other[2] > min_index)
4734 min_index = stp->ls_stateid.other[2];
4740 * Yikes, highly unlikely, but I'll handle it anyhow.
4742 if (min_index == 0x80000000 && max_index == 0x80000001) {
4745 * Loop around until we find an unused entry. Return that
4746 * and set LCL_INDEXNOTOK, so the search will continue next time.
4747 * (This is one of those rare cases where a goto is the
4748 * cleanest way to code the loop.)
4751 for (i = 0; i < nfsrv_statehashsize; i++) {
4752 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4753 if (stp->ls_stateid.other[2] == canuse) {
4759 clp->lc_flags |= LCL_INDEXNOTOK;
4764 * Ok to start again from min + 1.
4766 clp->lc_stateindex = min_index + 1;
4767 clp->lc_statemaxindex = max_index;
4768 clp->lc_flags &= ~LCL_INDEXNOTOK;
4769 return (clp->lc_stateindex);
4773 * The following functions handle the stable storage file that deals with
4774 * the edge conditions described in RFC3530 Sec. 8.6.3.
4775 * The file is as follows:
4776 * - a single record at the beginning that has the lease time of the
4777 * previous server instance (before the last reboot) and the nfsrvboottime
4778 * values for the previous server boots.
4779 * These previous boot times are used to ensure that the current
4780 * nfsrvboottime does not, somehow, get set to a previous one.
4781 * (This is important so that Stale ClientIDs and StateIDs can
4783 * The number of previous nfsvrboottime values precedes the list.
4784 * - followed by some number of appended records with:
4785 * - client id string
4786 * - flag that indicates it is a record revoking state via lease
4787 * expiration or similar
4788 * OR has successfully acquired state.
4789 * These structures vary in length, with the client string at the end, up
4790 * to NFSV4_OPAQUELIMIT in size.
4792 * At the end of the grace period, the file is truncated, the first
4793 * record is rewritten with updated information and any acquired state
4794 * records for successful reclaims of state are written.
4796 * Subsequent records are appended when the first state is issued to
4797 * a client and when state is revoked for a client.
4799 * When reading the file in, state issued records that come later in
4800 * the file override older ones, since the append log is in cronological order.
4801 * If, for some reason, the file can't be read, the grace period is
4802 * immediately terminated and all reclaims get NFSERR_NOGRACE.
4806 * Read in the stable storage file. Called by nfssvc() before the nfsd
4807 * processes start servicing requests.
4810 nfsrv_setupstable(NFSPROC_T *p)
4812 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4813 struct nfsrv_stable *sp, *nsp;
4814 struct nfst_rec *tsp;
4815 int error, i, tryagain;
4817 ssize_t aresid, len;
4820 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4821 * a reboot, so state has not been lost.
4823 if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4826 * Set Grace over just until the file reads successfully.
4828 nfsrvboottime = time_second;
4829 LIST_INIT(&sf->nsf_head);
4830 sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4831 sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4832 if (sf->nsf_fp == NULL)
4834 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4835 (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4836 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4837 if (error || aresid || sf->nsf_numboots == 0 ||
4838 sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4842 * Now, read in the boottimes.
4844 sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4845 sizeof (time_t), M_TEMP, M_WAITOK);
4846 off = sizeof (struct nfsf_rec);
4847 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4848 (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4849 UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4850 if (error || aresid) {
4851 free(sf->nsf_bootvals, M_TEMP);
4852 sf->nsf_bootvals = NULL;
4857 * Make sure this nfsrvboottime is different from all recorded
4862 for (i = 0; i < sf->nsf_numboots; i++) {
4863 if (nfsrvboottime == sf->nsf_bootvals[i]) {
4871 sf->nsf_flags |= NFSNSF_OK;
4872 off += (sf->nsf_numboots * sizeof (time_t));
4875 * Read through the file, building a list of records for grace
4877 * Each record is between sizeof (struct nfst_rec) and
4878 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4879 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4881 tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4882 NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4884 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4885 (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4886 off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4887 len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4888 if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4889 len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4891 * Yuck, the file has been corrupted, so just return
4892 * after clearing out any restart state, so the grace period
4895 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4896 LIST_REMOVE(sp, nst_list);
4900 sf->nsf_flags &= ~NFSNSF_OK;
4901 free(sf->nsf_bootvals, M_TEMP);
4902 sf->nsf_bootvals = NULL;
4906 off += sizeof (struct nfst_rec) + tsp->len - 1;
4908 * Search the list for a matching client.
4910 LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4911 if (tsp->len == sp->nst_len &&
4912 !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4915 if (sp == LIST_END(&sf->nsf_head)) {
4916 sp = (struct nfsrv_stable *)malloc(tsp->len +
4917 sizeof (struct nfsrv_stable) - 1, M_TEMP,
4919 NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4920 sizeof (struct nfst_rec) + tsp->len - 1);
4921 LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4923 if (tsp->flag == NFSNST_REVOKE)
4924 sp->nst_flag |= NFSNST_REVOKE;
4927 * A subsequent timestamp indicates the client
4928 * did a setclientid/confirm and any previous
4929 * revoke is no longer relevant.
4931 sp->nst_flag &= ~NFSNST_REVOKE;
4936 sf->nsf_flags = NFSNSF_OK;
4937 sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4942 * Update the stable storage file, now that the grace period is over.
4945 nfsrv_updatestable(NFSPROC_T *p)
4947 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4948 struct nfsrv_stable *sp, *nsp;
4950 struct nfsvattr nva;
4952 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4957 if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4959 sf->nsf_flags |= NFSNSF_UPDATEDONE;
4961 * Ok, we need to rewrite the stable storage file.
4962 * - truncate to 0 length
4963 * - write the new first structure
4964 * - loop through the data structures, writing out any that
4965 * have timestamps older than the old boot
4967 if (sf->nsf_bootvals) {
4969 for (i = sf->nsf_numboots - 2; i >= 0; i--)
4970 sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4972 sf->nsf_numboots = 1;
4973 sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4976 sf->nsf_bootvals[0] = nfsrvboottime;
4977 sf->nsf_lease = nfsrv_lease;
4978 NFSVNO_ATTRINIT(&nva);
4979 NFSVNO_SETATTRVAL(&nva, size, 0);
4980 vp = NFSFPVNODE(sf->nsf_fp);
4981 vn_start_write(vp, &mp, V_WAIT);
4982 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4983 error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4988 vn_finished_write(mp);
4990 error = NFSD_RDWR(UIO_WRITE, vp,
4991 (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4992 UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4994 error = NFSD_RDWR(UIO_WRITE, vp,
4995 (caddr_t)sf->nsf_bootvals,
4996 sf->nsf_numboots * sizeof (time_t),
4997 (off_t)(sizeof (struct nfsf_rec)),
4998 UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4999 free(sf->nsf_bootvals, M_TEMP);
5000 sf->nsf_bootvals = NULL;
5002 sf->nsf_flags &= ~NFSNSF_OK;
5003 printf("EEK! Can't write NfsV4 stable storage file\n");
5006 sf->nsf_flags |= NFSNSF_OK;
5009 * Loop through the list and write out timestamp records for
5010 * any clients that successfully reclaimed state.
5012 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5013 if (sp->nst_flag & NFSNST_GOTSTATE) {
5014 nfsrv_writestable(sp->nst_client, sp->nst_len,
5015 NFSNST_NEWSTATE, p);
5016 sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5018 LIST_REMOVE(sp, nst_list);
5021 nfsrv_backupstable();
5025 * Append a record to the stable storage file.
5028 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5030 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
5031 struct nfst_rec *sp;
5034 if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5036 sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5037 len - 1, M_TEMP, M_WAITOK);
5039 NFSBCOPY(client, sp->client, len);
5041 error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5042 (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5043 UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5046 sf->nsf_flags &= ~NFSNSF_OK;
5047 printf("EEK! Can't write NfsV4 stable storage file\n");
5052 * This function is called during the grace period to mark a client
5053 * that successfully reclaimed state.
5056 nfsrv_markstable(struct nfsclient *clp)
5058 struct nfsrv_stable *sp;
5061 * First find the client structure.
5063 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5064 if (sp->nst_len == clp->lc_idlen &&
5065 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5068 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5072 * Now, just mark it and set the nfsclient back pointer.
5074 sp->nst_flag |= NFSNST_GOTSTATE;
5079 * This function is called when a NFSv4.1 client does a ReclaimComplete.
5080 * Very similar to nfsrv_markstable(), except for the flag being set.
5083 nfsrv_markreclaim(struct nfsclient *clp)
5085 struct nfsrv_stable *sp;
5088 * First find the client structure.
5090 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5091 if (sp->nst_len == clp->lc_idlen &&
5092 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5095 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5099 * Now, just set the flag.
5101 sp->nst_flag |= NFSNST_RECLAIMED;
5105 * This function is called for a reclaim, to see if it gets grace.
5106 * It returns 0 if a reclaim is allowed, 1 otherwise.
5109 nfsrv_checkstable(struct nfsclient *clp)
5111 struct nfsrv_stable *sp;
5114 * First, find the entry for the client.
5116 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5117 if (sp->nst_len == clp->lc_idlen &&
5118 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5123 * If not in the list, state was revoked or no state was issued
5124 * since the previous reboot, a reclaim is denied.
5126 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
5127 (sp->nst_flag & NFSNST_REVOKE) ||
5128 !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
5134 * Test for and try to clear out a conflicting client. This is called by
5135 * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5137 * The trick here is that it can't revoke a conflicting client with an
5138 * expired lease unless it holds the v4root lock, so...
5139 * If no v4root lock, get the lock and return 1 to indicate "try again".
5140 * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5141 * the revocation worked and the conflicting client is "bye, bye", so it
5142 * can be tried again.
5143 * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5144 * Unlocks State before a non-zero value is returned.
5147 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5150 int gotlock, lktype = 0;
5153 * If lease hasn't expired, we can't fix it.
5155 if (clp->lc_expiry >= NFSD_MONOSEC ||
5156 !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
5158 if (*haslockp == 0) {
5161 lktype = NFSVOPISLOCKED(vp);
5164 NFSLOCKV4ROOTMUTEX();
5165 nfsv4_relref(&nfsv4rootfs_lock);
5167 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5168 NFSV4ROOTLOCKMUTEXPTR, NULL);
5170 NFSUNLOCKV4ROOTMUTEX();
5173 NFSVOPLOCK(vp, lktype | LK_RETRY);
5174 if (VN_IS_DOOMED(vp))
5182 * Ok, we can expire the conflicting client.
5184 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5185 nfsrv_backupstable();
5186 nfsrv_cleanclient(clp, p);
5187 nfsrv_freedeleglist(&clp->lc_deleg);
5188 nfsrv_freedeleglist(&clp->lc_olddeleg);
5189 LIST_REMOVE(clp, lc_hash);
5190 nfsrv_zapclient(clp, p);
5195 * Resolve a delegation conflict.
5196 * Returns 0 to indicate the conflict was resolved without sleeping.
5197 * Return -1 to indicate that the caller should check for conflicts again.
5198 * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5200 * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5201 * for a return of 0, since there was no sleep and it could be required
5202 * later. It is released for a return of NFSERR_DELAY, since the caller
5203 * will return that error. It is released when a sleep was done waiting
5204 * for the delegation to be returned or expire (so that other nfsds can
5205 * handle ops). Then, it must be acquired for the write to stable storage.
5206 * (This function is somewhat similar to nfsrv_clientconflict(), but
5207 * the semantics differ in a couple of subtle ways. The return of 0
5208 * indicates the conflict was resolved without sleeping here, not
5209 * that the conflict can't be resolved and the handling of nfsv4root_lock
5210 * differs, as noted above.)
5211 * Unlocks State before returning a non-zero value.
5214 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5217 struct nfsclient *clp = stp->ls_clp;
5218 int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5219 nfsv4stateid_t tstateid;
5223 * If the conflict is with an old delegation...
5225 if (stp->ls_flags & NFSLCK_OLDDELEG) {
5227 * You can delete it, if it has expired.
5229 if (clp->lc_delegtime < NFSD_MONOSEC) {
5230 nfsrv_freedeleg(stp);
5237 * During this delay, the old delegation could expire or it
5238 * could be recovered by the client via an Open with
5239 * CLAIM_DELEGATE_PREV.
5240 * Release the nfsv4root_lock, if held.
5244 NFSLOCKV4ROOTMUTEX();
5245 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5246 NFSUNLOCKV4ROOTMUTEX();
5248 error = NFSERR_DELAY;
5253 * It's a current delegation, so:
5254 * - check to see if the delegation has expired
5255 * - if so, get the v4root lock and then expire it
5257 if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
5259 * - do a recall callback, since not yet done
5260 * For now, never allow truncate to be set. To use
5261 * truncate safely, it must be guaranteed that the
5262 * Remove, Rename or Setattr with size of 0 will
5263 * succeed and that would require major changes to
5264 * the VFS/Vnode OPs.
5265 * Set the expiry time large enough so that it won't expire
5266 * until after the callback, then set it correctly, once
5267 * the callback is done. (The delegation will now time
5268 * out whether or not the Recall worked ok. The timeout
5269 * will be extended when ops are done on the delegation
5270 * stateid, up to the timelimit.)
5272 stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5274 stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
5276 stp->ls_flags |= NFSLCK_DELEGRECALL;
5279 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5280 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5281 * in order to try and avoid a race that could happen
5282 * when a CBRecall request passed the Open reply with
5283 * the delegation in it when transitting the network.
5284 * Since nfsrv_docallback will sleep, don't use stp after
5287 NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5289 NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5294 NFSLOCKV4ROOTMUTEX();
5295 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5296 NFSUNLOCKV4ROOTMUTEX();
5300 error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5301 &tstateid, 0, &tfh, NULL, NULL, 0, p);
5303 } while ((error == NFSERR_BADSTATEID ||
5304 error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5305 error = NFSERR_DELAY;
5309 if (clp->lc_expiry >= NFSD_MONOSEC &&
5310 stp->ls_delegtime >= NFSD_MONOSEC) {
5313 * A recall has been done, but it has not yet expired.
5318 NFSLOCKV4ROOTMUTEX();
5319 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5320 NFSUNLOCKV4ROOTMUTEX();
5322 error = NFSERR_DELAY;
5327 * If we don't yet have the lock, just get it and then return,
5328 * since we need that before deleting expired state, such as
5330 * When getting the lock, unlock the vnode, so other nfsds that
5331 * are in progress, won't get stuck waiting for the vnode lock.
5333 if (*haslockp == 0) {
5336 lktype = NFSVOPISLOCKED(vp);
5339 NFSLOCKV4ROOTMUTEX();
5340 nfsv4_relref(&nfsv4rootfs_lock);
5342 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5343 NFSV4ROOTLOCKMUTEXPTR, NULL);
5345 NFSUNLOCKV4ROOTMUTEX();
5348 NFSVOPLOCK(vp, lktype | LK_RETRY);
5349 if (VN_IS_DOOMED(vp)) {
5351 NFSLOCKV4ROOTMUTEX();
5352 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5353 NFSUNLOCKV4ROOTMUTEX();
5354 error = NFSERR_PERM;
5364 * Ok, we can delete the expired delegation.
5365 * First, write the Revoke record to stable storage and then
5366 * clear out the conflict.
5367 * Since all other nfsd threads are now blocked, we can safely
5368 * sleep without the state changing.
5370 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5371 nfsrv_backupstable();
5372 if (clp->lc_expiry < NFSD_MONOSEC) {
5373 nfsrv_cleanclient(clp, p);
5374 nfsrv_freedeleglist(&clp->lc_deleg);
5375 nfsrv_freedeleglist(&clp->lc_olddeleg);
5376 LIST_REMOVE(clp, lc_hash);
5379 nfsrv_freedeleg(stp);
5383 nfsrv_zapclient(clp, p);
5392 * Check for a remove allowed, if remove is set to 1 and get rid of
5396 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5397 nfsquad_t clientid, NFSPROC_T *p)
5399 struct nfsclient *clp;
5400 struct nfsstate *stp;
5401 struct nfslockfile *lfp;
5402 int error, haslock = 0;
5407 * First, get the lock file structure.
5408 * (A return of -1 means no associated state, so remove ok.)
5410 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5413 if (error == 0 && clientid.qval != 0)
5414 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5415 (nfsquad_t)((u_quad_t)0), 0, nd, p);
5417 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5421 NFSLOCKV4ROOTMUTEX();
5422 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5423 NFSUNLOCKV4ROOTMUTEX();
5431 * Now, we must Recall any delegations.
5433 error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5436 * nfsrv_cleandeleg() unlocks state for non-zero
5442 NFSLOCKV4ROOTMUTEX();
5443 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5444 NFSUNLOCKV4ROOTMUTEX();
5450 * Now, look for a conflicting open share.
5454 * If the entry in the directory was the last reference to the
5455 * corresponding filesystem object, the object can be destroyed
5457 if(lfp->lf_usecount>1)
5458 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5459 if (stp->ls_flags & NFSLCK_WRITEDENY) {
5460 error = NFSERR_FILEOPEN;
5468 NFSLOCKV4ROOTMUTEX();
5469 nfsv4_unlock(&nfsv4rootfs_lock, 1);
5470 NFSUNLOCKV4ROOTMUTEX();
5479 * Clear out all delegations for the file referred to by lfp.
5480 * May return NFSERR_DELAY, if there will be a delay waiting for
5481 * delegations to expire.
5482 * Returns -1 to indicate it slept while recalling a delegation.
5483 * This function has the side effect of deleting the nfslockfile structure,
5484 * if it no longer has associated state and didn't have to sleep.
5485 * Unlocks State before a non-zero value is returned.
5488 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5489 struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5491 struct nfsstate *stp, *nstp;
5494 stp = LIST_FIRST(&lfp->lf_deleg);
5495 while (stp != LIST_END(&lfp->lf_deleg)) {
5496 nstp = LIST_NEXT(stp, ls_file);
5497 if (stp->ls_clp != clp) {
5498 ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5501 * nfsrv_delegconflict() unlocks state
5502 * when it returns non-zero.
5515 * There are certain operations that, when being done outside of NFSv4,
5516 * require that any NFSv4 delegation for the file be recalled.
5517 * This function is to be called for those cases:
5518 * VOP_RENAME() - When a delegation is being recalled for any reason,
5519 * the client may have to do Opens against the server, using the file's
5520 * final component name. If the file has been renamed on the server,
5521 * that component name will be incorrect and the Open will fail.
5522 * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5523 * been removed on the server, if there is a delegation issued to
5524 * that client for the file. I say "theoretically" since clients
5525 * normally do an Access Op before the Open and that Access Op will
5526 * fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5527 * they will detect the file's removal in the same manner. (There is
5528 * one case where RFC3530 allows a client to do an Open without first
5529 * doing an Access Op, which is passage of a check against the ACE
5530 * returned with a Write delegation, but current practice is to ignore
5531 * the ACE and always do an Access Op.)
5532 * Since the functions can only be called with an unlocked vnode, this
5533 * can't be done at this time.
5534 * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5535 * locks locally in the client, which are not visible to the server. To
5536 * deal with this, issuing of delegations for a vnode must be disabled
5537 * and all delegations for the vnode recalled. This is done via the
5538 * second function, using the VV_DISABLEDELEG vflag on the vnode.
5541 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5547 * First, check to see if the server is currently running and it has
5548 * been called for a regular file when issuing delegations.
5550 if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5551 nfsrv_issuedelegs == 0)
5554 KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5556 * First, get a reference on the nfsv4rootfs_lock so that an
5557 * exclusive lock cannot be acquired by another thread.
5559 NFSLOCKV4ROOTMUTEX();
5560 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5561 NFSUNLOCKV4ROOTMUTEX();
5564 * Now, call nfsrv_checkremove() in a loop while it returns
5565 * NFSERR_DELAY. Return upon any other error or when timed out.
5567 starttime = NFSD_MONOSEC;
5569 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5570 error = nfsrv_checkremove(vp, 0, NULL,
5571 (nfsquad_t)((u_quad_t)0), p);
5575 if (error == NFSERR_DELAY) {
5576 if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5578 /* Sleep for a short period of time */
5579 (void) nfs_catnap(PZERO, 0, "nfsremove");
5581 } while (error == NFSERR_DELAY);
5582 NFSLOCKV4ROOTMUTEX();
5583 nfsv4_relref(&nfsv4rootfs_lock);
5584 NFSUNLOCKV4ROOTMUTEX();
5588 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5591 #ifdef VV_DISABLEDELEG
5593 * First, flag issuance of delegations disabled.
5595 atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5599 * Then call nfsd_recalldelegation() to get rid of all extant
5602 nfsd_recalldelegation(vp, p);
5606 * Check for conflicting locks, etc. and then get rid of delegations.
5607 * (At one point I thought that I should get rid of delegations for any
5608 * Setattr, since it could potentially disallow the I/O op (read or write)
5609 * allowed by the delegation. However, Setattr Ops that aren't changing
5610 * the size get a stateid of all 0s, so you can't tell if it is a delegation
5611 * for the same client or a different one, so I decided to only get rid
5612 * of delegations for other clients when the size is being changed.)
5613 * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5614 * as Write backs, even if there is no delegation, so it really isn't any
5618 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5619 nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5620 struct nfsexstuff *exp, NFSPROC_T *p)
5622 struct nfsstate st, *stp = &st;
5623 struct nfslock lo, *lop = &lo;
5627 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5628 stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5629 lop->lo_first = nvap->na_size;
5634 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5635 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5636 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5637 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5638 stp->ls_flags |= NFSLCK_SETATTR;
5639 if (stp->ls_flags == 0)
5641 lop->lo_end = NFS64BITSSET;
5642 lop->lo_flags = NFSLCK_WRITE;
5643 stp->ls_ownerlen = 0;
5645 stp->ls_uid = nd->nd_cred->cr_uid;
5646 stp->ls_stateid.seqid = stateidp->seqid;
5647 clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5648 clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5649 stp->ls_stateid.other[2] = stateidp->other[2];
5650 error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5651 stateidp, exp, nd, p);
5654 NFSEXITCODE2(error, nd);
5659 * Check for a write delegation and do a CBGETATTR if there is one, updating
5660 * the attributes, as required.
5661 * Should I return an error if I can't get the attributes? (For now, I'll
5665 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5666 struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5668 struct nfsstate *stp;
5669 struct nfslockfile *lfp;
5670 struct nfsclient *clp;
5671 struct nfsvattr nva;
5674 nfsattrbit_t cbbits;
5675 u_quad_t delegfilerev;
5677 NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5678 if (!NFSNONZERO_ATTRBIT(&cbbits))
5680 if (nfsrv_writedelegcnt == 0)
5684 * Get the lock file structure.
5685 * (A return of -1 means no associated state, so return ok.)
5687 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5690 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5699 * Now, look for a write delegation.
5701 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5702 if (stp->ls_flags & NFSLCK_DELEGWRITE)
5705 if (stp == LIST_END(&lfp->lf_deleg)) {
5711 /* If the clientid is not confirmed, ignore the delegation. */
5712 if (clp->lc_flags & LCL_NEEDSCONFIRM) {
5717 delegfilerev = stp->ls_filerev;
5719 * If the Write delegation was issued as a part of this Compound RPC
5720 * or if we have an Implied Clientid (used in a previous Op in this
5721 * compound) and it is the client the delegation was issued to,
5723 * I also assume that it is from the same client iff the network
5724 * host IP address is the same as the callback address. (Not
5725 * exactly correct by the RFC, but avoids a lot of Getattr
5728 if (nd->nd_compref == stp->ls_compref ||
5729 ((nd->nd_flag & ND_IMPLIEDCLID) &&
5730 clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5731 nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5737 * We are now done with the delegation state structure,
5738 * so the statelock can be released and we can now tsleep().
5742 * Now, we must do the CB Getattr callback, to see if Change or Size
5745 if (clp->lc_expiry >= NFSD_MONOSEC) {
5747 NFSVNO_ATTRINIT(&nva);
5748 nva.na_filerev = NFS64BITSSET;
5749 error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5750 0, &nfh, &nva, &cbbits, 0, p);
5752 if ((nva.na_filerev != NFS64BITSSET &&
5753 nva.na_filerev > delegfilerev) ||
5754 (NFSVNO_ISSETSIZE(&nva) &&
5755 nva.na_size != nvap->na_size)) {
5756 error = nfsvno_updfilerev(vp, nvap, nd, p);
5757 if (NFSVNO_ISSETSIZE(&nva))
5758 nvap->na_size = nva.na_size;
5761 error = 0; /* Ignore callback errors for now. */
5767 NFSEXITCODE2(error, nd);
5772 * This function looks for openowners that haven't had any opens for
5773 * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5777 nfsrv_throwawayopens(NFSPROC_T *p)
5779 struct nfsclient *clp, *nclp;
5780 struct nfsstate *stp, *nstp;
5784 nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5786 * For each client...
5788 for (i = 0; i < nfsrv_clienthashsize; i++) {
5789 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5790 LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5791 if (LIST_EMPTY(&stp->ls_open) &&
5792 (stp->ls_noopens > NFSNOOPEN ||
5793 (nfsrv_openpluslock * 2) >
5794 nfsrv_v4statelimit))
5795 nfsrv_freeopenowner(stp, 0, p);
5803 * This function checks to see if the credentials are the same.
5804 * Returns 1 for not same, 0 otherwise.
5807 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5810 if (nd->nd_flag & ND_GSS) {
5811 if (!(clp->lc_flags & LCL_GSS))
5813 if (clp->lc_flags & LCL_NAME) {
5814 if (nd->nd_princlen != clp->lc_namelen ||
5815 NFSBCMP(nd->nd_principal, clp->lc_name,
5821 if (nd->nd_cred->cr_uid == clp->lc_uid)
5825 } else if (clp->lc_flags & LCL_GSS)
5828 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5829 * in RFC3530, which talks about principals, but doesn't say anything
5830 * about uids for AUTH_SYS.)
5832 if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5839 * Calculate the lease expiry time.
5842 nfsrv_leaseexpiry(void)
5845 if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5846 return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5847 return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5851 * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5854 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5857 if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5860 if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5861 stp->ls_delegtime < stp->ls_delegtimelimit) {
5862 stp->ls_delegtime += nfsrv_lease;
5863 if (stp->ls_delegtime > stp->ls_delegtimelimit)
5864 stp->ls_delegtime = stp->ls_delegtimelimit;
5869 * This function checks to see if there is any other state associated
5870 * with the openowner for this Open.
5871 * It returns 1 if there is no other state, 0 otherwise.
5874 nfsrv_nootherstate(struct nfsstate *stp)
5876 struct nfsstate *tstp;
5878 LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5879 if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5886 * Create a list of lock deltas (changes to local byte range locking
5887 * that can be rolled back using the list) and apply the changes via
5888 * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5889 * the rollback or update function will be called after this.
5890 * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5891 * call fails. If it returns an error, it will unlock the list.
5894 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5895 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5897 struct nfslock *lop, *nlop;
5900 /* Loop through the list of locks. */
5901 lop = LIST_FIRST(&lfp->lf_locallock);
5902 while (first < end && lop != NULL) {
5903 nlop = LIST_NEXT(lop, lo_lckowner);
5904 if (first >= lop->lo_end) {
5907 } else if (first < lop->lo_first) {
5908 /* new one starts before entry in list */
5909 if (end <= lop->lo_first) {
5910 /* no overlap between old and new */
5911 error = nfsrv_dolocal(vp, lfp, flags,
5912 NFSLCK_UNLOCK, first, end, cfp, p);
5917 /* handle fragment overlapped with new one */
5918 error = nfsrv_dolocal(vp, lfp, flags,
5919 NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5923 first = lop->lo_first;
5926 /* new one overlaps this entry in list */
5927 if (end <= lop->lo_end) {
5928 /* overlaps all of new one */
5929 error = nfsrv_dolocal(vp, lfp, flags,
5930 lop->lo_flags, first, end, cfp, p);
5935 /* handle fragment overlapped with new one */
5936 error = nfsrv_dolocal(vp, lfp, flags,
5937 lop->lo_flags, first, lop->lo_end, cfp, p);
5940 first = lop->lo_end;
5945 if (first < end && error == 0)
5946 /* handle fragment past end of list */
5947 error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5955 * Local lock unlock. Unlock all byte ranges that are no longer locked
5956 * by NFSv4. To do this, unlock any subranges of first-->end that
5957 * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5958 * list. This list has all locks for the file held by other
5959 * <clientid, lockowner> tuples. The list is ordered by increasing
5960 * lo_first value, but may have entries that overlap each other, for
5961 * the case of read locks.
5964 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5965 uint64_t init_end, NFSPROC_T *p)
5967 struct nfslock *lop;
5968 uint64_t first, end, prevfirst __unused;
5972 while (first < init_end) {
5973 /* Loop through all nfs locks, adjusting first and end */
5975 LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5976 KASSERT(prevfirst <= lop->lo_first,
5977 ("nfsv4 locks out of order"));
5978 KASSERT(lop->lo_first < lop->lo_end,
5979 ("nfsv4 bogus lock"));
5980 prevfirst = lop->lo_first;
5981 if (first >= lop->lo_first &&
5982 first < lop->lo_end)
5984 * Overlaps with initial part, so trim
5985 * off that initial part by moving first past
5988 first = lop->lo_end;
5989 else if (end > lop->lo_first &&
5990 lop->lo_first > first) {
5992 * This lock defines the end of the
5993 * segment to unlock, so set end to the
5994 * start of it and break out of the loop.
5996 end = lop->lo_first;
6001 * There is no segment left to do, so
6002 * break out of this loop and then exit
6003 * the outer while() since first will be set
6004 * to end, which must equal init_end here.
6009 /* Unlock this segment */
6010 (void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6011 NFSLCK_READ, first, end, NULL, p);
6012 nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6016 * Now move past this segment and look for any further
6017 * segment in the range, if there is one.
6025 * Do the local lock operation and update the rollback list, as required.
6026 * Perform the rollback and return the error if nfsvno_advlock() fails.
6029 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6030 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6032 struct nfsrollback *rlp;
6033 int error = 0, ltype, oldltype;
6035 if (flags & NFSLCK_WRITE)
6037 else if (flags & NFSLCK_READ)
6041 if (oldflags & NFSLCK_WRITE)
6043 else if (oldflags & NFSLCK_READ)
6047 if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6050 error = nfsvno_advlock(vp, ltype, first, end, p);
6053 cfp->cl_clientid.lval[0] = 0;
6054 cfp->cl_clientid.lval[1] = 0;
6056 cfp->cl_end = NFS64BITSSET;
6057 cfp->cl_flags = NFSLCK_WRITE;
6058 cfp->cl_ownerlen = 5;
6059 NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6061 nfsrv_locallock_rollback(vp, lfp, p);
6062 } else if (ltype != F_UNLCK) {
6063 rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6065 rlp->rlck_first = first;
6066 rlp->rlck_end = end;
6067 rlp->rlck_type = oldltype;
6068 LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6077 * Roll back local lock changes and free up the rollback list.
6080 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6082 struct nfsrollback *rlp, *nrlp;
6084 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6085 (void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6087 free(rlp, M_NFSDROLLBACK);
6089 LIST_INIT(&lfp->lf_rollback);
6093 * Update local lock list and delete rollback list (ie now committed to the
6094 * local locks). Most of the work is done by the internal function.
6097 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6100 struct nfsrollback *rlp, *nrlp;
6101 struct nfslock *new_lop, *other_lop;
6103 new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6104 if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6105 other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6109 new_lop->lo_flags = flags;
6110 new_lop->lo_first = first;
6111 new_lop->lo_end = end;
6112 nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6113 if (new_lop != NULL)
6114 free(new_lop, M_NFSDLOCK);
6115 if (other_lop != NULL)
6116 free(other_lop, M_NFSDLOCK);
6118 /* and get rid of the rollback list */
6119 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6120 free(rlp, M_NFSDROLLBACK);
6121 LIST_INIT(&lfp->lf_rollback);
6125 * Lock the struct nfslockfile for local lock updating.
6128 nfsrv_locklf(struct nfslockfile *lfp)
6132 /* lf_usecount ensures *lfp won't be free'd */
6135 gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6136 NFSSTATEMUTEXPTR, NULL);
6137 } while (gotlock == 0);
6142 * Unlock the struct nfslockfile after local lock updating.
6145 nfsrv_unlocklf(struct nfslockfile *lfp)
6148 nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6152 * Clear out all state for the NFSv4 server.
6153 * Must be called by a thread that can sleep when no nfsds are running.
6156 nfsrv_throwawayallstate(NFSPROC_T *p)
6158 struct nfsclient *clp, *nclp;
6159 struct nfslockfile *lfp, *nlfp;
6163 * For each client, clean out the state and then free the structure.
6165 for (i = 0; i < nfsrv_clienthashsize; i++) {
6166 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
6167 nfsrv_cleanclient(clp, p);
6168 nfsrv_freedeleglist(&clp->lc_deleg);
6169 nfsrv_freedeleglist(&clp->lc_olddeleg);
6170 free(clp->lc_stateid, M_NFSDCLIENT);
6171 free(clp, M_NFSDCLIENT);
6176 * Also, free up any remaining lock file structures.
6178 for (i = 0; i < nfsrv_lockhashsize; i++) {
6179 LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
6180 printf("nfsd unload: fnd a lock file struct\n");
6181 nfsrv_freenfslockfile(lfp);
6185 /* And get rid of the deviceid structures and layouts. */
6186 nfsrv_freealllayoutsanddevids();
6190 * Check the sequence# for the session and slot provided as an argument.
6191 * Also, renew the lease if the session will return NFS_OK.
6194 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6195 uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6196 uint32_t *sflagsp, NFSPROC_T *p)
6198 struct nfsdsession *sep;
6199 struct nfssessionhash *shp;
6203 shp = NFSSESSIONHASH(nd->nd_sessionid);
6204 NFSLOCKSESSION(shp);
6205 sep = nfsrv_findsession(nd->nd_sessionid);
6207 NFSUNLOCKSESSION(shp);
6208 return (NFSERR_BADSESSION);
6210 error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6211 sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6213 NFSUNLOCKSESSION(shp);
6216 if (cache_this != 0)
6217 nd->nd_flag |= ND_SAVEREPLY;
6218 /* Renew the lease. */
6219 sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6220 nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6221 nd->nd_flag |= ND_IMPLIEDCLID;
6223 /* Save maximum request and reply sizes. */
6224 nd->nd_maxreq = sep->sess_maxreq;
6225 nd->nd_maxresp = sep->sess_maxresp;
6228 * If this session handles the backchannel, save the nd_xprt for this
6229 * RPC, since this is the one being used.
6230 * RFC-5661 specifies that the fore channel will be implicitly
6231 * bound by a Sequence operation. However, since some NFSv4.1 clients
6232 * erroneously assumed that the back channel would be implicitly
6233 * bound as well, do the implicit binding unless a
6234 * BindConnectiontoSession has already been done on the session.
6237 if (sep->sess_clp->lc_req.nr_client != NULL &&
6238 sep->sess_cbsess.nfsess_xprt != nd->nd_xprt &&
6239 (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0 &&
6240 (sep->sess_clp->lc_flags & LCL_DONEBINDCONN) == 0) {
6242 "nfsrv_checksequence: implicit back channel bind\n");
6243 savxprt = sep->sess_cbsess.nfsess_xprt;
6244 SVC_ACQUIRE(nd->nd_xprt);
6245 nd->nd_xprt->xp_p2 =
6246 sep->sess_clp->lc_req.nr_client->cl_private;
6247 nd->nd_xprt->xp_idletimeout = 0; /* Disable timeout. */
6248 sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6252 if (sep->sess_clp->lc_req.nr_client == NULL)
6253 *sflagsp |= NFSV4SEQ_CBPATHDOWN;
6254 NFSUNLOCKSESSION(shp);
6255 if (savxprt != NULL)
6256 SVC_RELEASE(savxprt);
6257 if (error == NFSERR_EXPIRED) {
6258 *sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6260 } else if (error == NFSERR_ADMINREVOKED) {
6261 *sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6264 *highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6269 * Check/set reclaim complete for this session/clientid.
6272 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6274 struct nfsdsession *sep;
6275 struct nfssessionhash *shp;
6278 shp = NFSSESSIONHASH(nd->nd_sessionid);
6280 NFSLOCKSESSION(shp);
6281 sep = nfsrv_findsession(nd->nd_sessionid);
6283 NFSUNLOCKSESSION(shp);
6285 return (NFSERR_BADSESSION);
6289 sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6290 /* Check to see if reclaim complete has already happened. */
6291 else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6292 error = NFSERR_COMPLETEALREADY;
6294 sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6295 nfsrv_markreclaim(sep->sess_clp);
6297 NFSUNLOCKSESSION(shp);
6303 * Cache the reply in a session slot.
6306 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6308 struct nfsdsession *sep;
6309 struct nfssessionhash *shp;
6312 struct sockaddr_in *sin;
6315 struct sockaddr_in6 *sin6;
6318 shp = NFSSESSIONHASH(nd->nd_sessionid);
6319 NFSLOCKSESSION(shp);
6320 sep = nfsrv_findsession(nd->nd_sessionid);
6322 NFSUNLOCKSESSION(shp);
6323 if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
6324 buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6325 switch (nd->nd_nam->sa_family) {
6328 sin = (struct sockaddr_in *)nd->nd_nam;
6329 cp = inet_ntop(sin->sin_family,
6330 &sin->sin_addr.s_addr, buf,
6336 sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6337 cp = inet_ntop(sin6->sin6_family,
6338 &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6345 printf("nfsrv_cache_session: no session "
6348 printf("nfsrv_cache_session: no session\n");
6354 nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6356 NFSUNLOCKSESSION(shp);
6360 * Search for a session that matches the sessionid.
6362 static struct nfsdsession *
6363 nfsrv_findsession(uint8_t *sessionid)
6365 struct nfsdsession *sep;
6366 struct nfssessionhash *shp;
6368 shp = NFSSESSIONHASH(sessionid);
6369 LIST_FOREACH(sep, &shp->list, sess_hash) {
6370 if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6377 * Destroy a session.
6380 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6382 int error, igotlock, samesess;
6385 if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6386 (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6388 if ((nd->nd_flag & ND_LASTOP) == 0)
6389 return (NFSERR_BADSESSION);
6392 /* Lock out other nfsd threads */
6393 NFSLOCKV4ROOTMUTEX();
6394 nfsv4_relref(&nfsv4rootfs_lock);
6396 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6397 NFSV4ROOTLOCKMUTEXPTR, NULL);
6398 } while (igotlock == 0);
6399 NFSUNLOCKV4ROOTMUTEX();
6401 error = nfsrv_freesession(NULL, sessionid);
6402 if (error == 0 && samesess != 0)
6403 nd->nd_flag &= ~ND_HASSEQUENCE;
6405 NFSLOCKV4ROOTMUTEX();
6406 nfsv4_unlock(&nfsv4rootfs_lock, 1);
6407 NFSUNLOCKV4ROOTMUTEX();
6412 * Bind a connection to a session.
6413 * For now, only certain variants are supported, since the current session
6414 * structure can only handle a single backchannel entry, which will be
6415 * applied to all connections if it is set.
6418 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6420 struct nfssessionhash *shp;
6421 struct nfsdsession *sep;
6422 struct nfsclient *clp;
6427 shp = NFSSESSIONHASH(sessionid);
6429 NFSLOCKSESSION(shp);
6430 sep = nfsrv_findsession(sessionid);
6432 clp = sep->sess_clp;
6433 if (*foreaftp == NFSCDFC4_BACK ||
6434 *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6435 *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6436 /* Try to set up a backchannel. */
6437 if (clp->lc_req.nr_client == NULL) {
6438 NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6440 clp->lc_req.nr_client = (struct __rpc_client *)
6441 clnt_bck_create(nd->nd_xprt->xp_socket,
6442 sep->sess_cbprogram, NFSV4_CBVERS);
6444 if (clp->lc_req.nr_client != NULL) {
6445 NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6447 savxprt = sep->sess_cbsess.nfsess_xprt;
6448 SVC_ACQUIRE(nd->nd_xprt);
6449 nd->nd_xprt->xp_p2 =
6450 clp->lc_req.nr_client->cl_private;
6451 /* Disable idle timeout. */
6452 nd->nd_xprt->xp_idletimeout = 0;
6453 sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6454 if (savxprt != NULL)
6455 SVC_RELEASE(savxprt);
6456 sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6457 clp->lc_flags |= LCL_DONEBINDCONN;
6458 if (*foreaftp == NFSCDFS4_BACK)
6459 *foreaftp = NFSCDFS4_BACK;
6461 *foreaftp = NFSCDFS4_BOTH;
6462 } else if (*foreaftp != NFSCDFC4_BACK) {
6463 NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6464 "up backchannel\n");
6465 sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6466 clp->lc_flags |= LCL_DONEBINDCONN;
6467 *foreaftp = NFSCDFS4_FORE;
6469 error = NFSERR_NOTSUPP;
6470 printf("nfsrv_bindconnsess: Can't add "
6474 NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6475 clp->lc_flags |= LCL_DONEBINDCONN;
6476 *foreaftp = NFSCDFS4_FORE;
6479 error = NFSERR_BADSESSION;
6480 NFSUNLOCKSESSION(shp);
6486 * Free up a session structure.
6489 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6491 struct nfssessionhash *shp;
6496 shp = NFSSESSIONHASH(sessionid);
6497 NFSLOCKSESSION(shp);
6498 sep = nfsrv_findsession(sessionid);
6500 shp = NFSSESSIONHASH(sep->sess_sessionid);
6501 NFSLOCKSESSION(shp);
6505 if (sep->sess_refcnt > 0) {
6506 NFSUNLOCKSESSION(shp);
6508 return (NFSERR_BACKCHANBUSY);
6510 LIST_REMOVE(sep, sess_hash);
6511 LIST_REMOVE(sep, sess_list);
6513 NFSUNLOCKSESSION(shp);
6516 return (NFSERR_BADSESSION);
6517 for (i = 0; i < NFSV4_SLOTS; i++)
6518 if (sep->sess_slots[i].nfssl_reply != NULL)
6519 m_freem(sep->sess_slots[i].nfssl_reply);
6520 if (sep->sess_cbsess.nfsess_xprt != NULL)
6521 SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6522 free(sep, M_NFSDSESSION);
6528 * RFC5661 says that it should fail when there are associated opens, locks
6529 * or delegations. Since stateids represent opens, I don't see how you can
6530 * free an open stateid (it will be free'd when closed), so this function
6531 * only works for lock stateids (freeing the lock_owner) or delegations.
6534 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6537 struct nfsclient *clp;
6538 struct nfsstate *stp;
6543 * Look up the stateid
6545 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6546 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6548 /* First, check for a delegation. */
6549 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6550 if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6555 nfsrv_freedeleg(stp);
6560 /* Not a delegation, try for a lock_owner. */
6562 error = nfsrv_getstate(clp, stateidp, 0, &stp);
6563 if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6564 NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6565 /* Not a lock_owner stateid. */
6566 error = NFSERR_LOCKSHELD;
6567 if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6568 error = NFSERR_LOCKSHELD;
6570 nfsrv_freelockowner(stp, NULL, 0, p);
6579 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6582 struct nfsclient *clp;
6583 struct nfsstate *stp;
6588 * Look up the stateid
6590 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6591 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6593 error = nfsrv_getstate(clp, stateidp, 0, &stp);
6594 if (error == 0 && stateidp->seqid != 0 &&
6595 SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6596 error = NFSERR_OLDSTATEID;
6602 * Generate the xdr for an NFSv4.1 CBSequence Operation.
6605 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6606 int dont_replycache, struct nfsdsession **sepp)
6608 struct nfsdsession *sep;
6609 uint32_t *tl, slotseq = 0;
6610 int maxslot, slotpos;
6611 uint8_t sessionid[NFSX_V4SESSIONID];
6614 error = nfsv4_getcbsession(clp, sepp);
6618 (void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6619 &slotseq, sessionid);
6620 KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6622 /* Build the Sequence arguments. */
6623 NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6624 bcopy(sessionid, tl, NFSX_V4SESSIONID);
6625 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6626 nd->nd_slotseq = tl;
6627 *tl++ = txdr_unsigned(slotseq);
6628 *tl++ = txdr_unsigned(slotpos);
6629 *tl++ = txdr_unsigned(maxslot);
6630 if (dont_replycache == 0)
6631 *tl++ = newnfs_true;
6633 *tl++ = newnfs_false;
6634 *tl = 0; /* No referring call list, for now. */
6635 nd->nd_flag |= ND_HASSEQUENCE;
6640 * Get a session for the callback.
6643 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6645 struct nfsdsession *sep;
6648 LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6649 if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6654 return (NFSERR_BADSESSION);
6663 * Free up all backchannel xprts. This needs to be done when the nfsd threads
6664 * exit, since those transports will all be going away.
6665 * This is only called after all the nfsd threads are done performing RPCs,
6666 * so locking shouldn't be an issue.
6669 nfsrv_freeallbackchannel_xprts(void)
6671 struct nfsdsession *sep;
6672 struct nfsclient *clp;
6676 for (i = 0; i < nfsrv_clienthashsize; i++) {
6677 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6678 LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6679 xprt = sep->sess_cbsess.nfsess_xprt;
6680 sep->sess_cbsess.nfsess_xprt = NULL;
6689 * Do a layout commit. Actually just call nfsrv_updatemdsattr().
6690 * I have no idea if the rest of these arguments will ever be useful?
6693 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6694 int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6695 int hasnewmtime, struct timespec *newmtimep, int reclaim,
6696 nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6697 uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6702 error = nfsrv_updatemdsattr(vp, &na, p);
6705 *newsizep = na.na_size;
6711 * Try and get a layout.
6714 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6715 int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6716 uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6717 int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6719 struct nfslayouthash *lhyp;
6720 struct nfslayout *lyp;
6722 fhandle_t fh, *dsfhp;
6723 int error, mirrorcnt;
6725 if (nfsrv_devidcnt == 0)
6726 return (NFSERR_UNKNLAYOUTTYPE);
6729 printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6731 error = nfsvno_getfh(vp, &fh, p);
6732 NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6737 * For now, all layouts are for entire files.
6738 * Only issue Read/Write layouts if requested for a non-readonly fs.
6740 if (NFSVNO_EXRDONLY(exp)) {
6741 if (*iomode == NFSLAYOUTIOMODE_RW)
6742 return (NFSERR_LAYOUTTRYLATER);
6743 *iomode = NFSLAYOUTIOMODE_READ;
6745 if (*iomode != NFSLAYOUTIOMODE_RW)
6746 *iomode = NFSLAYOUTIOMODE_READ;
6749 * Check to see if a write layout can be issued for this file.
6750 * This is used during mirror recovery to avoid RW layouts being
6751 * issued for a file while it is being copied to the recovered
6754 if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6755 return (NFSERR_LAYOUTTRYLATER);
6761 /* First, see if a layout already exists and return if found. */
6762 lhyp = NFSLAYOUTHASH(&fh);
6763 NFSLOCKLAYOUT(lhyp);
6764 error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6765 NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6767 * Not sure if the seqid must be the same, so I won't check it.
6769 if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6770 stateidp->other[1] != lyp->lay_stateid.other[1] ||
6771 stateidp->other[2] != lyp->lay_stateid.other[2])) {
6772 if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6773 NFSUNLOCKLAYOUT(lhyp);
6774 NFSD_DEBUG(1, "ret bad stateid\n");
6775 return (NFSERR_BADSTATEID);
6778 * I believe we get here because there is a race between
6779 * the client processing the CBLAYOUTRECALL and the layout
6780 * being deleted here on the server.
6781 * The client has now done a LayoutGet with a non-layout
6782 * stateid, as it would when there is no layout.
6783 * As such, free this layout and set error == NFSERR_BADSTATEID
6784 * so the code below will create a new layout structure as
6785 * would happen if no layout was found.
6786 * "lyp" will be set before being used below, but set it NULL
6789 nfsrv_freelayout(&lhyp->list, lyp);
6791 error = NFSERR_BADSTATEID;
6794 if (lyp->lay_layoutlen > maxcnt) {
6795 NFSUNLOCKLAYOUT(lhyp);
6796 NFSD_DEBUG(1, "ret layout too small\n");
6797 return (NFSERR_TOOSMALL);
6799 if (*iomode == NFSLAYOUTIOMODE_RW)
6800 lyp->lay_flags |= NFSLAY_RW;
6802 lyp->lay_flags |= NFSLAY_READ;
6803 NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6804 *layoutlenp = lyp->lay_layoutlen;
6805 if (++lyp->lay_stateid.seqid == 0)
6806 lyp->lay_stateid.seqid = 1;
6807 stateidp->seqid = lyp->lay_stateid.seqid;
6808 NFSUNLOCKLAYOUT(lhyp);
6809 NFSD_DEBUG(4, "ret fnd layout\n");
6812 NFSUNLOCKLAYOUT(lhyp);
6814 /* Find the device id and file handle. */
6815 dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6816 devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6817 error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6818 NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6820 if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6821 if (NFSX_V4FILELAYOUT > maxcnt)
6822 error = NFSERR_TOOSMALL;
6824 lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6825 devid, vp->v_mount->mnt_stat.f_fsid);
6827 if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6828 error = NFSERR_TOOSMALL;
6830 lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6832 vp->v_mount->mnt_stat.f_fsid);
6835 free(dsfhp, M_TEMP);
6836 free(devid, M_TEMP);
6841 * Now, add this layout to the list.
6843 error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6844 NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6846 * The lyp will be set to NULL by nfsrv_addlayout() if it
6847 * linked the new structure into the lists.
6849 free(lyp, M_NFSDSTATE);
6854 * Generate a File Layout.
6856 static struct nfslayout *
6857 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6858 fhandle_t *dsfhp, char *devid, fsid_t fs)
6861 struct nfslayout *lyp;
6862 uint64_t pattern_offset;
6864 lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6866 lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6867 if (iomode == NFSLAYOUTIOMODE_RW)
6868 lyp->lay_flags = NFSLAY_RW;
6870 lyp->lay_flags = NFSLAY_READ;
6871 NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6872 lyp->lay_clientid.qval = nd->nd_clientid.qval;
6875 /* Fill in the xdr for the files layout. */
6876 tl = (uint32_t *)lyp->lay_xdr;
6877 NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */
6878 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6881 * Make the stripe size as many 64K blocks as will fit in the stripe
6882 * mask. Since there is only one stripe, the stripe size doesn't really
6883 * matter, except that the Linux client will only handle an exact
6884 * multiple of their PAGE_SIZE (usually 4K). I chose 64K as a value
6885 * that should cover most/all arches w.r.t. PAGE_SIZE.
6887 *tl++ = txdr_unsigned(NFSFLAYUTIL_STRIPE_MASK & ~0xffff);
6888 *tl++ = 0; /* 1st stripe index. */
6890 txdr_hyper(pattern_offset, tl); tl += 2; /* Pattern offset. */
6891 *tl++ = txdr_unsigned(1); /* 1 file handle. */
6892 *tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6893 NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6894 lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6898 #define FLEX_OWNERID "999"
6899 #define FLEX_UID0 "0"
6901 * Generate a Flex File Layout.
6902 * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6903 * string goes on the wire, it isn't supposed to be used by the client,
6904 * since this server uses tight coupling.
6905 * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6906 * a string of "0". This works around the Linux Flex File Layout driver bug
6907 * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6909 static struct nfslayout *
6910 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6911 fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
6914 struct nfslayout *lyp;
6918 lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
6919 M_NFSDSTATE, M_WAITOK | M_ZERO);
6920 lyp->lay_type = NFSLAYOUT_FLEXFILE;
6921 if (iomode == NFSLAYOUTIOMODE_RW)
6922 lyp->lay_flags = NFSLAY_RW;
6924 lyp->lay_flags = NFSLAY_READ;
6925 NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6926 lyp->lay_clientid.qval = nd->nd_clientid.qval;
6928 lyp->lay_mirrorcnt = mirrorcnt;
6930 /* Fill in the xdr for the files layout. */
6931 tl = (uint32_t *)lyp->lay_xdr;
6933 txdr_hyper(lenval, tl); tl += 2; /* Stripe unit. */
6934 *tl++ = txdr_unsigned(mirrorcnt); /* # of mirrors. */
6935 for (i = 0; i < mirrorcnt; i++) {
6936 *tl++ = txdr_unsigned(1); /* One stripe. */
6937 NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */
6938 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6939 devid += NFSX_V4DEVICEID;
6940 *tl++ = txdr_unsigned(1); /* Efficiency. */
6941 *tl++ = 0; /* Proxy Stateid. */
6945 *tl++ = txdr_unsigned(1); /* 1 file handle. */
6946 *tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6947 NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6948 tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
6950 if (nfsrv_flexlinuxhack != 0) {
6951 *tl++ = txdr_unsigned(strlen(FLEX_UID0));
6952 *tl = 0; /* 0 pad string. */
6953 NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6954 *tl++ = txdr_unsigned(strlen(FLEX_UID0));
6955 *tl = 0; /* 0 pad string. */
6956 NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6958 *tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6959 NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6960 *tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6961 NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6964 *tl++ = txdr_unsigned(0); /* ff_flags. */
6965 *tl = txdr_unsigned(60); /* Status interval hint. */
6966 lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
6971 * Parse and process Flex File errors returned via LayoutReturn.
6974 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
6978 int cnt, errcnt, i, j, opnum, stat;
6979 char devid[NFSX_V4DEVICEID];
6982 cnt = fxdr_unsigned(int, *tl++);
6983 NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
6984 for (i = 0; i < cnt; i++) {
6985 /* Skip offset, length and stateid for now. */
6986 tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
6987 errcnt = fxdr_unsigned(int, *tl++);
6988 NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
6989 for (j = 0; j < errcnt; j++) {
6990 NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
6991 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6992 stat = fxdr_unsigned(int, *tl++);
6993 opnum = fxdr_unsigned(int, *tl++);
6994 NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
6997 * Except for NFSERR_ACCES and NFSERR_STALE errors,
6998 * disable the mirror.
7000 if (stat != NFSERR_ACCES && stat != NFSERR_STALE)
7001 nfsrv_delds(devid, p);
7007 * This function removes all flex file layouts which has a mirror with
7008 * a device id that matches the argument.
7009 * Called when the DS represented by the device id has failed.
7012 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
7015 struct nfslayout *lyp, *nlyp;
7016 struct nfslayouthash *lhyp;
7017 struct nfslayouthead loclyp;
7020 NFSD_DEBUG(4, "flexmirrordel\n");
7021 /* Move all layouts found onto a local list. */
7022 TAILQ_INIT(&loclyp);
7023 for (i = 0; i < nfsrv_layouthashsize; i++) {
7024 lhyp = &nfslayouthash[i];
7025 NFSLOCKLAYOUT(lhyp);
7026 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7027 if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
7028 lyp->lay_mirrorcnt > 1) {
7029 NFSD_DEBUG(4, "possible match\n");
7032 for (j = 0; j < lyp->lay_mirrorcnt; j++) {
7034 if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
7037 NFSD_DEBUG(4, "fnd one\n");
7038 TAILQ_REMOVE(&lhyp->list, lyp,
7040 TAILQ_INSERT_HEAD(&loclyp, lyp,
7044 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7045 NFSM_RNDUP(NFSX_V4PNFSFH) /
7046 NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7050 NFSUNLOCKLAYOUT(lhyp);
7053 /* Now, try to do a Layout recall for each one found. */
7054 TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7055 NFSD_DEBUG(4, "do layout recall\n");
7057 * The layout stateid.seqid needs to be incremented
7058 * before doing a LAYOUT_RECALL callback.
7060 if (++lyp->lay_stateid.seqid == 0)
7061 lyp->lay_stateid.seqid = 1;
7062 nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7063 &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7064 nfsrv_freelayout(&loclyp, lyp);
7069 * Do a recall callback to the client for this layout.
7072 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7073 struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7075 struct nfsclient *clp;
7078 NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7079 error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7081 NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7083 printf("nfsrv_recalllayout: getclient err=%d\n", error);
7086 if ((clp->lc_flags & LCL_NFSV41) != 0) {
7087 error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7088 stateidp, changed, fhp, NULL, NULL, laytype, p);
7089 /* If lyp != NULL, handle an error return here. */
7090 if (error != 0 && lyp != NULL) {
7093 * Mark it returned, since no layout recall
7095 * All errors seem to be non-recoverable, although
7096 * NFSERR_NOMATCHLAYOUT is a normal event.
7098 if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7099 lyp->lay_flags |= NFSLAY_RETURNED;
7103 if (error != NFSERR_NOMATCHLAYOUT)
7104 printf("nfsrv_recalllayout: err=%d\n", error);
7107 printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7112 * Find a layout to recall when we exceed our high water mark.
7115 nfsrv_recalloldlayout(NFSPROC_T *p)
7117 struct nfslayouthash *lhyp;
7118 struct nfslayout *lyp;
7120 nfsv4stateid_t stateid;
7122 int error, laytype = 0, ret;
7124 lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7125 NFSLOCKLAYOUT(lhyp);
7126 TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7127 if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7128 lyp->lay_flags |= NFSLAY_CALLB;
7130 * The layout stateid.seqid needs to be incremented
7131 * before doing a LAYOUT_RECALL callback.
7133 if (++lyp->lay_stateid.seqid == 0)
7134 lyp->lay_stateid.seqid = 1;
7135 clientid = lyp->lay_clientid;
7136 stateid = lyp->lay_stateid;
7137 NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7138 laytype = lyp->lay_type;
7142 NFSUNLOCKLAYOUT(lhyp);
7144 error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7146 if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7147 NFSD_DEBUG(4, "recallold=%d\n", error);
7149 NFSLOCKLAYOUT(lhyp);
7151 * Since the hash list was unlocked, we need to
7154 ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7157 (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7158 lyp->lay_stateid.other[0] == stateid.other[0] &&
7159 lyp->lay_stateid.other[1] == stateid.other[1] &&
7160 lyp->lay_stateid.other[2] == stateid.other[2]) {
7162 * The client no longer knows this layout, so
7163 * it can be free'd now.
7165 if (error == NFSERR_NOMATCHLAYOUT)
7166 nfsrv_freelayout(&lhyp->list, lyp);
7169 * Leave it to be tried later by
7170 * clearing NFSLAY_CALLB and moving
7171 * it to the head of the list, so it
7172 * won't be tried again for a while.
7174 lyp->lay_flags &= ~NFSLAY_CALLB;
7175 TAILQ_REMOVE(&lhyp->list, lyp,
7177 TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7181 NFSUNLOCKLAYOUT(lhyp);
7187 * Try and return layout(s).
7190 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7191 int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7192 int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7193 struct ucred *cred, NFSPROC_T *p)
7196 struct nfslayouthash *lhyp;
7197 struct nfslayout *lyp;
7202 if (kind == NFSV4LAYOUTRET_FILE) {
7203 error = nfsvno_getfh(vp, &fh, p);
7205 error = nfsrv_updatemdsattr(vp, &na, p);
7207 printf("nfsrv_layoutreturn: updatemdsattr"
7208 " failed=%d\n", error);
7211 if (reclaim == newnfs_true) {
7212 error = nfsrv_checkgrace(NULL, NULL,
7214 if (error != NFSERR_NOGRACE)
7218 lhyp = NFSLAYOUTHASH(&fh);
7220 NFSLOCKLAYOUT(lhyp);
7221 error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7222 layouttype, p, &lyp);
7223 NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7225 stateidp->other[0] == lyp->lay_stateid.other[0] &&
7226 stateidp->other[1] == lyp->lay_stateid.other[1] &&
7227 stateidp->other[2] == lyp->lay_stateid.other[2]) {
7228 NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7229 " %x %x %x laystateid %d %x %x %x"
7230 " off=%ju len=%ju flgs=0x%x\n",
7231 stateidp->seqid, stateidp->other[0],
7232 stateidp->other[1], stateidp->other[2],
7233 lyp->lay_stateid.seqid,
7234 lyp->lay_stateid.other[0],
7235 lyp->lay_stateid.other[1],
7236 lyp->lay_stateid.other[2],
7237 (uintmax_t)offset, (uintmax_t)len,
7239 if (++lyp->lay_stateid.seqid == 0)
7240 lyp->lay_stateid.seqid = 1;
7241 stateidp->seqid = lyp->lay_stateid.seqid;
7242 if (offset == 0 && len == UINT64_MAX) {
7243 if ((iomode & NFSLAYOUTIOMODE_READ) !=
7245 lyp->lay_flags &= ~NFSLAY_READ;
7246 if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7247 lyp->lay_flags &= ~NFSLAY_RW;
7248 if ((lyp->lay_flags & (NFSLAY_READ |
7250 nfsrv_freelayout(&lhyp->list,
7257 NFSUNLOCKLAYOUT(lhyp);
7258 /* Search the nfsrv_recalllist for a match. */
7259 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7260 if (NFSBCMP(&lyp->lay_fh, &fh,
7262 lyp->lay_clientid.qval ==
7263 nd->nd_clientid.qval &&
7264 stateidp->other[0] ==
7265 lyp->lay_stateid.other[0] &&
7266 stateidp->other[1] ==
7267 lyp->lay_stateid.other[1] &&
7268 stateidp->other[2] ==
7269 lyp->lay_stateid.other[2]) {
7270 lyp->lay_flags |= NFSLAY_RETURNED;
7277 if (layouttype == NFSLAYOUT_FLEXFILE)
7278 nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7279 } else if (kind == NFSV4LAYOUTRET_FSID)
7280 nfsrv_freelayouts(&nd->nd_clientid,
7281 &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7282 else if (kind == NFSV4LAYOUTRET_ALL)
7283 nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7285 error = NFSERR_INVAL;
7292 * Look for an existing layout.
7295 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7296 NFSPROC_T *p, struct nfslayout **lypp)
7298 struct nfslayouthash *lhyp;
7299 struct nfslayout *lyp;
7304 lhyp = NFSLAYOUTHASH(fhp);
7305 TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7306 if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7307 lyp->lay_clientid.qval == clientidp->qval &&
7308 lyp->lay_type == laytype)
7319 * Add the new layout, as required.
7322 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7323 nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7325 struct nfsclient *clp;
7326 struct nfslayouthash *lhyp;
7327 struct nfslayout *lyp, *nlyp;
7331 KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7332 ("nfsrv_layoutget: no nd_clientid\n"));
7336 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7337 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7342 lyp->lay_stateid.seqid = stateidp->seqid = 1;
7343 lyp->lay_stateid.other[0] = stateidp->other[0] =
7344 clp->lc_clientid.lval[0];
7345 lyp->lay_stateid.other[1] = stateidp->other[1] =
7346 clp->lc_clientid.lval[1];
7347 lyp->lay_stateid.other[2] = stateidp->other[2] =
7348 nfsrv_nextstateindex(clp);
7351 lhyp = NFSLAYOUTHASH(fhp);
7352 NFSLOCKLAYOUT(lhyp);
7353 TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7354 if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7355 nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7359 /* A layout already exists, so use it. */
7360 nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7361 NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7362 *layoutlenp = nlyp->lay_layoutlen;
7363 if (++nlyp->lay_stateid.seqid == 0)
7364 nlyp->lay_stateid.seqid = 1;
7365 stateidp->seqid = nlyp->lay_stateid.seqid;
7366 stateidp->other[0] = nlyp->lay_stateid.other[0];
7367 stateidp->other[1] = nlyp->lay_stateid.other[1];
7368 stateidp->other[2] = nlyp->lay_stateid.other[2];
7369 NFSUNLOCKLAYOUT(lhyp);
7373 /* Insert the new layout in the lists. */
7375 atomic_add_int(&nfsrv_layoutcnt, 1);
7376 NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7377 *layoutlenp = lyp->lay_layoutlen;
7378 TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7379 NFSUNLOCKLAYOUT(lhyp);
7384 * Get the devinfo for a deviceid.
7387 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7388 uint32_t *notify, int *devaddrlen, char **devaddr)
7390 struct nfsdevice *ds;
7392 if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7393 NFSLAYOUT_FLEXFILE) ||
7394 (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7395 return (NFSERR_UNKNLAYOUTTYPE);
7398 * Now, search for the device id. Note that the structures won't go
7399 * away, but the order changes in the list. As such, the lock only
7400 * needs to be held during the search through the list.
7403 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7404 if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7405 ds->nfsdev_nmp != NULL)
7410 return (NFSERR_NOENT);
7412 /* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7414 if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7415 *devaddrlen = ds->nfsdev_fileaddrlen;
7416 *devaddr = ds->nfsdev_fileaddr;
7417 } else if (layouttype == NFSLAYOUT_FLEXFILE) {
7418 *devaddrlen = ds->nfsdev_flexaddrlen;
7419 *devaddr = ds->nfsdev_flexaddr;
7421 if (*devaddrlen == 0)
7422 return (NFSERR_UNKNLAYOUTTYPE);
7425 * The XDR overhead is 3 unsigned values: layout_type,
7426 * length_of_address and notify bitmap.
7427 * If the notify array is changed to not all zeros, the
7428 * count of unsigned values must be increased.
7430 if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7431 3 * NFSX_UNSIGNED) {
7432 *maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7433 return (NFSERR_TOOSMALL);
7439 * Free a list of layout state structures.
7442 nfsrv_freelayoutlist(nfsquad_t clientid)
7444 struct nfslayouthash *lhyp;
7445 struct nfslayout *lyp, *nlyp;
7448 for (i = 0; i < nfsrv_layouthashsize; i++) {
7449 lhyp = &nfslayouthash[i];
7450 NFSLOCKLAYOUT(lhyp);
7451 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7452 if (lyp->lay_clientid.qval == clientid.qval)
7453 nfsrv_freelayout(&lhyp->list, lyp);
7455 NFSUNLOCKLAYOUT(lhyp);
7463 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7466 NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7467 atomic_add_int(&nfsrv_layoutcnt, -1);
7468 TAILQ_REMOVE(lhp, lyp, lay_list);
7469 free(lyp, M_NFSDSTATE);
7473 * Free up a device id.
7476 nfsrv_freeonedevid(struct nfsdevice *ds)
7480 atomic_add_int(&nfsrv_devidcnt, -1);
7481 vrele(ds->nfsdev_dvp);
7482 for (i = 0; i < nfsrv_dsdirsize; i++)
7483 if (ds->nfsdev_dsdir[i] != NULL)
7484 vrele(ds->nfsdev_dsdir[i]);
7485 free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7486 free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7487 free(ds->nfsdev_host, M_NFSDSTATE);
7488 free(ds, M_NFSDSTATE);
7492 * Free up a device id and its mirrors.
7495 nfsrv_freedevid(struct nfsdevice *ds)
7498 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7499 nfsrv_freeonedevid(ds);
7503 * Free all layouts and device ids.
7504 * Done when the nfsd threads are shut down since there may be a new
7505 * modified device id list created when the nfsd is restarted.
7508 nfsrv_freealllayoutsanddevids(void)
7510 struct nfsdontlist *mrp, *nmrp;
7511 struct nfslayout *lyp, *nlyp;
7513 /* Get rid of the deviceid structures. */
7514 nfsrv_freealldevids();
7515 TAILQ_INIT(&nfsrv_devidhead);
7518 /* Get rid of all layouts. */
7519 nfsrv_freealllayouts();
7521 /* Get rid of any nfsdontlist entries. */
7522 LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7523 free(mrp, M_NFSDSTATE);
7524 LIST_INIT(&nfsrv_dontlisthead);
7525 nfsrv_dontlistlen = 0;
7527 /* Free layouts in the recall list. */
7528 TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7529 nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7530 TAILQ_INIT(&nfsrv_recalllisthead);
7534 * Free layouts that match the arguments.
7537 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7539 struct nfslayouthash *lhyp;
7540 struct nfslayout *lyp, *nlyp;
7543 for (i = 0; i < nfsrv_layouthashsize; i++) {
7544 lhyp = &nfslayouthash[i];
7545 NFSLOCKLAYOUT(lhyp);
7546 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7547 if (clid->qval != lyp->lay_clientid.qval)
7549 if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7551 if (laytype != lyp->lay_type)
7553 if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7554 lyp->lay_flags &= ~NFSLAY_READ;
7555 if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7556 lyp->lay_flags &= ~NFSLAY_RW;
7557 if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7558 nfsrv_freelayout(&lhyp->list, lyp);
7560 NFSUNLOCKLAYOUT(lhyp);
7565 * Free all layouts for the argument file.
7568 nfsrv_freefilelayouts(fhandle_t *fhp)
7570 struct nfslayouthash *lhyp;
7571 struct nfslayout *lyp, *nlyp;
7573 lhyp = NFSLAYOUTHASH(fhp);
7574 NFSLOCKLAYOUT(lhyp);
7575 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7576 if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7577 nfsrv_freelayout(&lhyp->list, lyp);
7579 NFSUNLOCKLAYOUT(lhyp);
7586 nfsrv_freealllayouts(void)
7588 struct nfslayouthash *lhyp;
7589 struct nfslayout *lyp, *nlyp;
7592 for (i = 0; i < nfsrv_layouthashsize; i++) {
7593 lhyp = &nfslayouthash[i];
7594 NFSLOCKLAYOUT(lhyp);
7595 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7596 nfsrv_freelayout(&lhyp->list, lyp);
7597 NFSUNLOCKLAYOUT(lhyp);
7602 * Look up the mount path for the DS server.
7605 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7606 struct nfsdevice **dsp)
7608 struct nameidata nd;
7609 struct nfsdevice *ds;
7615 NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7617 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7620 NFSD_DEBUG(4, "lookup=%d\n", error);
7623 if (nd.ni_vp->v_type != VDIR) {
7625 NFSD_DEBUG(4, "dspath not dir\n");
7628 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7630 NFSD_DEBUG(4, "dspath not an NFS mount\n");
7635 * Allocate a DS server structure with the NFS mounted directory
7636 * vnode reference counted, so that a non-forced dismount will
7638 * This structure is always linked into the list, even if an error
7639 * is being returned. The caller will free the entire list upon
7642 *dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7643 M_NFSDSTATE, M_WAITOK | M_ZERO);
7644 ds->nfsdev_dvp = nd.ni_vp;
7645 ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7646 NFSVOPUNLOCK(nd.ni_vp);
7648 dsdirsize = strlen(dspathp) + 16;
7649 dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7650 /* Now, create the DS directory structures. */
7651 for (i = 0; i < nfsrv_dsdirsize; i++) {
7652 snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7653 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7654 UIO_SYSSPACE, dsdirpath, p);
7656 NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7659 if (nd.ni_vp->v_type != VDIR) {
7662 NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7665 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7668 NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7671 ds->nfsdev_dsdir[i] = nd.ni_vp;
7672 NFSVOPUNLOCK(nd.ni_vp);
7674 free(dsdirpath, M_TEMP);
7676 if (strlen(mdspathp) > 0) {
7678 * This DS stores file for a specific MDS exported file
7681 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7682 UIO_SYSSPACE, mdspathp, p);
7684 NFSD_DEBUG(4, "mds lookup=%d\n", error);
7687 if (nd.ni_vp->v_type != VDIR) {
7690 NFSD_DEBUG(4, "mdspath not dir\n");
7693 mp = nd.ni_vp->v_mount;
7694 if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7697 NFSD_DEBUG(4, "mdspath not an exported fs\n");
7700 ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7701 ds->nfsdev_mdsisset = 1;
7706 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7707 atomic_add_int(&nfsrv_devidcnt, 1);
7712 * Look up the mount path for the DS server and delete it.
7715 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7718 struct nfsmount *nmp;
7719 struct nfsdevice *ds;
7722 NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7724 * Search for the path in the mount list. Avoid looking the path
7725 * up, since this mount point may be hung, with associated locked
7727 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7728 * until this completes.
7729 * As noted in the man page, this should be done before any forced
7730 * dismount on the mount point, but at least the handshake on
7731 * NFSMNTP_CANCELRPCS should make it safe.
7736 mtx_lock(&mountlist_mtx);
7737 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7738 if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7739 strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7740 mp->mnt_data != NULL) {
7743 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7744 NFSMNTP_CANCELRPCS)) == 0) {
7745 nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7754 mtx_unlock(&mountlist_mtx);
7757 ds = nfsrv_deldsnmp(op, nmp, p);
7758 NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7760 nfsrv_killrpcs(nmp);
7761 NFSD_DEBUG(4, "aft killrpcs\n");
7765 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7774 * Search for and remove a DS entry which matches the "nmp" argument.
7775 * The nfsdevice structure pointer is returned so that the caller can
7776 * free it via nfsrv_freeonedevid().
7777 * For the forced case, do not try to do LayoutRecalls, since the server
7778 * must be shut down now anyhow.
7781 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7783 struct nfsdevice *fndds;
7785 NFSD_DEBUG(4, "deldsdvp\n");
7787 if (op == PNFSDOP_FORCEDELDS)
7788 fndds = nfsv4_findmirror(nmp);
7790 fndds = nfsrv_findmirroredds(nmp);
7792 nfsrv_deleteds(fndds);
7794 if (fndds != NULL) {
7795 if (op != PNFSDOP_FORCEDELDS)
7796 nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7797 printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7803 * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7804 * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7806 * Also, returns an error instead of the nfsdevice found.
7809 nfsrv_delds(char *devid, NFSPROC_T *p)
7811 struct nfsdevice *ds, *fndds;
7812 struct nfsmount *nmp;
7815 NFSD_DEBUG(4, "delds\n");
7817 * Search the DS server list for a match with devid.
7818 * Remove the DS entry if found and there is a mirror.
7824 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7825 if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7826 ds->nfsdev_nmp != NULL) {
7827 NFSD_DEBUG(4, "fnd main ds\n");
7832 if (fndds == NULL) {
7836 if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7838 else if (fndds->nfsdev_mdsisset != 0) {
7839 /* For the fsid is set case, search for a mirror. */
7840 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7841 if (ds != fndds && ds->nfsdev_nmp != NULL &&
7842 ds->nfsdev_mdsisset != 0 &&
7843 fsidcmp(&ds->nfsdev_mdsfsid,
7844 &fndds->nfsdev_mdsfsid) == 0) {
7850 if (fndmirror != 0) {
7851 nmp = fndds->nfsdev_nmp;
7853 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7854 NFSMNTP_CANCELRPCS)) == 0) {
7855 nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7857 nfsrv_deleteds(fndds);
7865 nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7866 printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7867 nfsrv_killrpcs(nmp);
7869 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7878 * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7881 nfsrv_deleteds(struct nfsdevice *fndds)
7884 NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7885 fndds->nfsdev_nmp = NULL;
7886 if (fndds->nfsdev_mdsisset == 0)
7891 * Fill in the addr structures for the File and Flex File layouts.
7894 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
7899 static uint64_t new_devid = 0;
7901 if (strchr(addr, ':') != NULL)
7906 /* Fill in the device id. */
7907 NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
7909 NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
7913 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
7914 * as defined in RFC5661) in XDR.
7916 addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7918 NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
7919 ds->nfsdev_fileaddrlen = addrlen;
7920 tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7921 ds->nfsdev_fileaddr = (char *)tl;
7922 *tl++ = txdr_unsigned(1); /* One stripe with index 0. */
7924 *tl++ = txdr_unsigned(1); /* One multipath list */
7925 *tl++ = txdr_unsigned(1); /* with one entry in it. */
7926 /* The netaddr for this one entry. */
7927 *tl++ = txdr_unsigned(strlen(netprot));
7928 NFSBCOPY(netprot, tl, strlen(netprot));
7929 tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7930 *tl++ = txdr_unsigned(strlen(addr));
7931 NFSBCOPY(addr, tl, strlen(addr));
7934 * Fill in the flex file addr (actually the ff_device_addr4
7935 * as defined for Flexible File Layout) in XDR.
7937 addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7939 ds->nfsdev_flexaddrlen = addrlen;
7940 tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7941 ds->nfsdev_flexaddr = (char *)tl;
7942 *tl++ = txdr_unsigned(1); /* One multipath entry. */
7943 /* The netaddr for this one entry. */
7944 *tl++ = txdr_unsigned(strlen(netprot));
7945 NFSBCOPY(netprot, tl, strlen(netprot));
7946 tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7947 *tl++ = txdr_unsigned(strlen(addr));
7948 NFSBCOPY(addr, tl, strlen(addr));
7949 tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
7950 *tl++ = txdr_unsigned(2); /* Two NFS Versions. */
7951 *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */
7952 *tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
7953 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max rsize. */
7954 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max wsize. */
7955 *tl++ = newnfs_true; /* Tightly coupled. */
7956 *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */
7957 *tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
7958 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max rsize. */
7959 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max wsize. */
7960 *tl = newnfs_true; /* Tightly coupled. */
7962 ds->nfsdev_hostnamelen = strlen(dnshost);
7963 ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
7965 NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
7969 * Create the device id list.
7970 * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
7974 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
7976 struct nfsdevice *ds;
7977 char *addrp, *dnshostp, *dspathp, *mdspathp;
7981 dnshostp = args->dnshost;
7982 dspathp = args->dspath;
7983 mdspathp = args->mdspath;
7984 nfsrv_maxpnfsmirror = args->mirrorcnt;
7985 if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
7990 * Loop around for each nul-terminated string in args->addr,
7991 * args->dnshost, args->dnspath and args->mdspath.
7993 while (addrp < (args->addr + args->addrlen) &&
7994 dnshostp < (args->dnshost + args->dnshostlen) &&
7995 dspathp < (args->dspath + args->dspathlen) &&
7996 mdspathp < (args->mdspath + args->mdspathlen)) {
7997 error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
7999 /* Free all DS servers. */
8000 nfsrv_freealldevids();
8004 nfsrv_allocdevid(ds, addrp, dnshostp);
8005 addrp += (strlen(addrp) + 1);
8006 dnshostp += (strlen(dnshostp) + 1);
8007 dspathp += (strlen(dspathp) + 1);
8008 mdspathp += (strlen(mdspathp) + 1);
8010 if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
8011 /* Free all DS servers. */
8012 nfsrv_freealldevids();
8014 nfsrv_maxpnfsmirror = 1;
8017 /* We can fail at most one less DS than the mirror level. */
8018 nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
8021 * Allocate the nfslayout hash table now, since this is a pNFS server.
8022 * Make it 1% of the high water mark and at least 100.
8024 if (nfslayouthash == NULL) {
8025 nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
8026 if (nfsrv_layouthashsize < 100)
8027 nfsrv_layouthashsize = 100;
8028 nfslayouthash = mallocarray(nfsrv_layouthashsize,
8029 sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
8031 for (i = 0; i < nfsrv_layouthashsize; i++) {
8032 mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
8033 TAILQ_INIT(&nfslayouthash[i].list);
8040 * Free all device ids.
8043 nfsrv_freealldevids(void)
8045 struct nfsdevice *ds, *nds;
8047 TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8048 nfsrv_freedevid(ds);
8052 * Check to see if there is a Read/Write Layout plus either:
8053 * - A Write Delegation
8055 * - An Open with Write_access.
8056 * Return 1 if this is the case and 0 otherwise.
8057 * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8058 * Getattr RPC to the Data Server (DS) is necessary.
8060 #define NFSCLIDVECSIZE 6
8062 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8064 fhandle_t fh, *tfhp;
8065 struct nfsstate *stp;
8066 struct nfslayout *lyp;
8067 struct nfslayouthash *lhyp;
8068 struct nfslockhashhead *hp;
8069 struct nfslockfile *lfp;
8070 nfsquad_t clid[NFSCLIDVECSIZE];
8073 ret = nfsvno_getfh(vp, &fh, p);
8077 /* First check for a Read/Write Layout. */
8079 lhyp = NFSLAYOUTHASH(&fh);
8080 NFSLOCKLAYOUT(lhyp);
8081 TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8082 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8083 ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8084 ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8085 nfsrv_pnfsatime != 0))) {
8086 if (clidcnt < NFSCLIDVECSIZE)
8087 clid[clidcnt].qval = lyp->lay_clientid.qval;
8091 NFSUNLOCKLAYOUT(lhyp);
8093 /* None found, so return 0. */
8097 /* Get the nfslockfile for this fh. */
8099 hp = NFSLOCKHASH(&fh);
8100 LIST_FOREACH(lfp, hp, lf_hash) {
8102 if (NFSVNO_CMPFH(&fh, tfhp))
8106 /* None found, so return 0. */
8111 /* Now, look for a Write delegation for this clientid. */
8112 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8113 if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8114 nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8118 /* Found one, so return 1. */
8123 /* No Write delegation, so look for an Open with Write_access. */
8124 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8125 KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8126 ("nfsrv_checkdsattr: Non-open in Open list\n"));
8127 if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8128 nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8138 * Look for a matching clientid in the vector. Return 1 if one might match.
8141 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8145 /* If too many for the vector, return 1 since there might be a match. */
8146 if (clidcnt > NFSCLIDVECSIZE)
8149 for (i = 0; i < clidcnt; i++)
8150 if (clidvec[i].qval == clid.qval)
8156 * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8157 * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8160 nfsrv_dontlayout(fhandle_t *fhp)
8162 struct nfsdontlist *mrp;
8165 if (nfsrv_dontlistlen == 0)
8169 LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8170 if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8171 (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8176 NFSDDONTLISTUNLOCK();
8180 #define PNFSDS_COPYSIZ 65536
8182 * Create a new file on a DS and copy the contents of an extant DS file to it.
8183 * This can be used for recovery of a DS file onto a recovered DS.
8185 * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8186 * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8187 * so that they will be disabled after the MDS file's vnode is unlocked.
8188 * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8190 * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8191 * writes, LayoutCommits and LayoutReturns for the file when completing the
8192 * LayoutReturn requested by the LayoutRecall callback.
8193 * - Issue a LayoutRecall callback for all read/write layouts and wait for
8194 * them to be returned. (If the LayoutRecall callback replies
8195 * NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8196 * - Exclusively lock the MDS file's vnode. This ensures that no proxied
8197 * writes are in progress or can occur during the DS file copy.
8198 * It also blocks Setattr operations.
8199 * - Create the file on the recovered mirror.
8200 * - Copy the file from the operational DS.
8201 * - Copy any ACL from the MDS file to the new DS file.
8202 * - Set the modify time of the new DS file to that of the MDS file.
8203 * - Update the extended attribute for the MDS file.
8204 * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8205 * - The caller will unlock the MDS file's vnode allowing operations
8206 * to continue normally, since it is now on the mirror again.
8209 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8210 struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8211 struct ucred *cred, NFSPROC_T *p)
8213 struct nfsdontlist *mrp, *nmrp;
8214 struct nfslayouthash *lhyp;
8215 struct nfslayout *lyp, *nlyp;
8216 struct nfslayouthead thl;
8217 struct mount *mp, *tvmp;
8220 struct timespec mtime;
8226 int didprintf, ret, retacl, xfer;
8228 ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8229 ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8231 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8232 * so that no more RW layouts will get issued.
8234 ret = nfsvno_getfh(vp, &fh, p);
8236 NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8239 nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8240 nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8241 NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8243 LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8244 if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8248 LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8251 nfsrv_dontlistlen++;
8252 NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8254 NFSDDONTLISTUNLOCK();
8255 free(nmrp, M_NFSDSTATE);
8256 NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8259 NFSDDONTLISTUNLOCK();
8262 * Search for all RW layouts for this file. Move them to the
8263 * recall list, so they can be recalled and their return noted.
8265 lhyp = NFSLAYOUTHASH(&fh);
8267 NFSLOCKLAYOUT(lhyp);
8268 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8269 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8270 (lyp->lay_flags & NFSLAY_RW) != 0) {
8271 TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8272 TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8273 lyp->lay_trycnt = 0;
8276 NFSUNLOCKLAYOUT(lhyp);
8283 /* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8285 /* Now, do a recall for all layouts not yet recalled. */
8288 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8289 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8290 (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8291 lyp->lay_flags |= NFSLAY_RECALL;
8293 * The layout stateid.seqid needs to be incremented
8294 * before doing a LAYOUT_RECALL callback.
8296 if (++lyp->lay_stateid.seqid == 0)
8297 lyp->lay_stateid.seqid = 1;
8299 nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8300 &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8301 NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8306 /* Now wait for them to be returned. */
8308 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8309 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8310 if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8311 TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8313 TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8315 "nfsrv_copymr: layout returned\n");
8318 ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8319 PVFS | PCATCH, "nfsmrl", hz);
8320 NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8322 if (ret == EINTR || ret == ERESTART)
8324 if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8326 * Give up after 60sec and return
8327 * ENXIO, failing the copymr.
8328 * This layout will remain on the
8329 * recalllist. It can only be cleared
8330 * by restarting the nfsd.
8331 * This seems the safe way to handle
8332 * it, since it cannot be safely copied
8333 * with an outstanding RW layout.
8335 if (lyp->lay_trycnt >= 60) {
8339 if (didprintf == 0) {
8340 printf("nfsrv_copymr: layout "
8350 /* We can now get rid of the layouts that have been returned. */
8351 TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8352 nfsrv_freelayout(&thl, lyp);
8355 * Do the vn_start_write() calls here, before the MDS vnode is
8356 * locked and the tvp is created (locked) in the NFS file system
8358 * For tvmp, this probably isn't necessary, since it will be an
8359 * NFS mount and they are not suspendable at this time.
8362 ret = vn_start_write(vp, &mp, V_WAIT | PCATCH);
8364 tvmp = dvp->v_mount;
8365 ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH);
8369 * LK_EXCLUSIVE lock the MDS vnode, so that any
8370 * proxied writes through the MDS will be blocked until we have
8371 * completed the copy and update of the extended attributes.
8372 * This will also ensure that any attributes and ACL will not be
8373 * changed until the copy is complete.
8375 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8376 if (ret == 0 && VN_IS_DOOMED(vp)) {
8377 NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8381 /* Create the data file on the recovered DS. */
8383 ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8385 /* Copy the DS file, if created successfully. */
8388 * Get any NFSv4 ACL on the MDS file, so that it can be set
8389 * on the new DS file.
8391 aclp = acl_alloc(M_WAITOK | M_ZERO);
8392 retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8393 if (retacl != 0 && retacl != ENOATTR)
8394 NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8395 dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8396 /* Malloc a block of 0s used to check for holes. */
8397 if (nfsrv_zeropnfsdat == NULL)
8398 nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8401 ret = VOP_GETATTR(fvp, &va, cred);
8403 while (ret == 0 && aresid == 0) {
8404 ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8405 rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8407 xfer = PNFSDS_COPYSIZ - aresid;
8408 if (ret == 0 && xfer > 0) {
8411 * Skip the write for holes, except for the
8414 if (xfer < PNFSDS_COPYSIZ || rdpos ==
8415 va.va_size || NFSBCMP(dat,
8416 nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8417 ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8418 wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8419 cred, NULL, NULL, p);
8425 /* If there is an ACL and the copy succeeded, set the ACL. */
8426 if (ret == 0 && retacl == 0) {
8427 ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8429 * Don't consider these as errors, since VOP_GETACL()
8430 * can return an ACL when they are not actually
8431 * supported. For example, for UFS, VOP_GETACL()
8432 * will return a trivial ACL based on the uid/gid/mode
8433 * when there is no ACL on the file.
8434 * This case should be recognized as a trivial ACL
8435 * by UFS's VOP_SETACL() and succeed, but...
8437 if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8442 ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8444 /* Set the DS data file's modify time that of the MDS file. */
8446 ret = VOP_GETATTR(vp, &va, cred);
8448 mtime = va.va_mtime;
8450 va.va_mtime = mtime;
8451 ret = VOP_SETATTR(tvp, &va, cred);
8459 vn_finished_write(tvmp);
8461 /* Update the extended attributes for the newly created DS file. */
8463 ret = vn_extattr_set(vp, IO_NODELOCKED,
8464 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8465 sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8467 vn_finished_write(mp);
8469 /* Get rid of the dontlist entry, so that Layouts can be issued. */
8471 LIST_REMOVE(mrp, nfsmr_list);
8472 NFSDDONTLISTUNLOCK();
8473 free(mrp, M_NFSDSTATE);
8478 * Create a data storage file on the recovered DS.
8481 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8482 vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8485 struct vattr va, nva;
8488 /* Make data file name based on FH. */
8489 error = VOP_GETATTR(vp, &va, cred);
8491 /* Set the attributes for "vp" to Setattr the DS vp. */
8493 nva.va_uid = va.va_uid;
8494 nva.va_gid = va.va_gid;
8495 nva.va_mode = va.va_mode;
8499 va.va_mode = nva.va_mode;
8500 NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8501 error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8502 pf->dsf_filename, cred, p, tvpp);
8508 * Look up the MDS file shared locked, and then get the extended attribute
8509 * to find the extant DS file to be copied to the new mirror.
8510 * If successful, *vpp is set to the MDS file's vp and *nvpp is
8511 * set to a DS data file for the MDS file, both exclusively locked.
8512 * The "buf" argument has the pnfsdsfile structure from the MDS file
8513 * in it and buflen is set to its length.
8516 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8517 int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8518 struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8519 struct nfsdevice **fdsp)
8521 struct nameidata nd;
8522 struct vnode *vp, *curvp;
8523 struct pnfsdsfile *pf;
8524 struct nfsmount *nmp, *curnmp;
8525 int dsdir, error, mirrorcnt, ippos;
8532 if (dspathp == NULL && curdspathp != NULL)
8536 * Look up the MDS file shared locked. The lock will be upgraded
8537 * to an exclusive lock after any rw layouts have been returned.
8539 NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8540 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8543 NFSD_DEBUG(4, "lookup=%d\n", error);
8546 if (nd.ni_vp->v_type != VREG) {
8548 NFSD_DEBUG(4, "mdspath not reg\n");
8553 if (curdspathp != NULL) {
8555 * Look up the current DS path and find the nfsdev structure for
8558 NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8559 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8560 UIO_SYSSPACE, curdspathp, p);
8562 NFSD_DEBUG(4, "ds lookup=%d\n", error);
8567 if (nd.ni_vp->v_type != VDIR) {
8570 NFSD_DEBUG(4, "curdspath not dir\n");
8573 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8576 NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8579 curnmp = VFSTONFS(nd.ni_vp->v_mount);
8581 /* Search the nfsdev list for a match. */
8583 *fdsp = nfsv4_findmirror(curnmp);
8587 if (curnmp == NULL) {
8590 NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8596 if (dspathp != NULL) {
8597 /* Look up the nfsdev path and find the nfsdev structure. */
8598 NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8599 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8600 UIO_SYSSPACE, dspathp, p);
8602 NFSD_DEBUG(4, "ds lookup=%d\n", error);
8609 if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8614 NFSD_DEBUG(4, "dspath not dir\n");
8615 if (nd.ni_vp == curvp)
8619 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8624 NFSD_DEBUG(4, "dspath not an NFS mount\n");
8627 nmp = VFSTONFS(nd.ni_vp->v_mount);
8630 * Search the nfsdevice list for a match. If curnmp == NULL,
8631 * this is a recovery and there must be a mirror.
8635 *dsp = nfsrv_findmirroredds(nmp);
8637 *dsp = nfsv4_findmirror(nmp);
8644 NFSD_DEBUG(4, "mdscopymr: no ds\n");
8653 * Get a vp for an available DS data file using the extended
8654 * attribute on the MDS file.
8655 * If there is a valid entry for the new DS in the extended attribute
8656 * on the MDS file (as checked via the nmp argument),
8657 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8659 error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8660 NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8663 if (nd.ni_vp == NULL) {
8664 if (error == 0 && nmp != NULL) {
8665 /* Search the nfsdev list for a match. */
8667 *dsp = nfsrv_findmirroredds(nmp);
8670 if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8671 if (nvpp != NULL && *nvpp != NULL) {
8681 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8682 * and is only allowed if there is a 0.0.0.0 IP address entry.
8683 * When curdspathp != NULL, the ippos will be set to that entry.
8685 if (error == 0 && dspathp != NULL && ippos == -1) {
8686 if (nvpp != NULL && *nvpp != NULL) {
8695 pf = (struct pnfsdsfile *)buf;
8697 /* If no zeroip pnfsdsfile, add one. */
8698 ippos = *buflenp / sizeof(*pf);
8699 *buflenp += sizeof(*pf);
8701 pf->dsf_dir = dsdir;
8702 strlcpy(pf->dsf_filename, fname,
8703 sizeof(pf->dsf_filename));
8713 * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8714 * Return one if found, NULL otherwise.
8716 static struct nfsdevice *
8717 nfsrv_findmirroredds(struct nfsmount *nmp)
8719 struct nfsdevice *ds, *fndds;
8722 mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8724 * Search the DS server list for a match with nmp.
8725 * Remove the DS entry if found and there is a mirror.
8729 if (nfsrv_devidcnt == 0)
8731 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8732 if (ds->nfsdev_nmp == nmp) {
8733 NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8740 if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8742 else if (fndds->nfsdev_mdsisset != 0) {
8743 /* For the fsid is set case, search for a mirror. */
8744 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8745 if (ds != fndds && ds->nfsdev_nmp != NULL &&
8746 ds->nfsdev_mdsisset != 0 &&
8747 fsidcmp(&ds->nfsdev_mdsfsid,
8748 &fndds->nfsdev_mdsfsid) == 0) {
8754 if (fndmirror == 0) {
8755 NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");