1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2009 Rick Macklem, University of Guelph 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 */ 29 30 #include <sys/cdefs.h> 31 __FBSDID("$FreeBSD$"); 32 33 #include "opt_inet.h" 34 #include "opt_inet6.h" 35 #include <sys/extattr.h> 36 #include <fs/nfs/nfsport.h> 37 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; 45 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; 59 NFSV4ROOTLOCKMUTEX; 60 NFSSTATESPINLOCK; 61 extern struct nfsdontlisthead nfsrv_dontlisthead; 62 extern volatile int nfsrv_devidcnt; 63 extern struct nfslayouthead nfsrv_recalllisthead; 64 extern char *nfsrv_zeropnfsdat; 65 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"); 71 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"); 76 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"); 81 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"); 86 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"); 91 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"); 96 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"); 101 102 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"); 106 107 int nfsrv_pnfsatime = 0; 108 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW, 109 &nfsrv_pnfsatime, 0, 110 "For pNFS service, do Getattr ops to keep atime up-to-date"); 111 112 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"); 116 117 /* 118 * Hash lists for nfs V4. 119 */ 120 struct nfsclienthashhead *nfsclienthash; 121 struct nfslockhashhead *nfslockhash; 122 struct nfssessionhash *nfssessionhash; 123 struct nfslayouthash *nfslayouthash; 124 volatile int nfsrv_dontlistlen = 0; 125 126 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; 133 134 /* 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, 138 NFSPROC_T *p); 139 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, 140 NFSPROC_T *p); 141 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep, 142 NFSPROC_T *p); 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, 164 u_int32_t flags); 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, 194 NFSPROC_T *p); 195 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, 196 NFSPROC_T *p); 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, 211 int iomode); 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, 223 NFSPROC_T *p); 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, 234 vnode_t *tvpp); 235 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp); 236 237 /* 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. 242 */ 243 int 244 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp, 245 nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p) 246 { 247 struct nfsclient *clp = NULL, *new_clp = *new_clpp; 248 int i, error = 0, ret; 249 struct nfsstate *stp, *tstp; 250 #ifdef INET 251 struct sockaddr_in *sin, *rin; 252 #endif 253 #ifdef INET6 254 struct sockaddr_in6 *sin6, *rin6; 255 #endif 256 struct nfsdsession *sep, *nsep; 257 int zapit = 0, gotit, hasstate = 0, igotlock; 258 static u_int64_t confirm_index = 0; 259 260 /* 261 * Check for state resource limit exceeded. 262 */ 263 if (nfsrv_openpluslock > nfsrv_v4statelimit) { 264 error = NFSERR_RESOURCE; 265 goto out; 266 } 267 268 if (nfsrv_issuedelegs == 0 || 269 ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0)) 270 /* 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 281 * the client. 282 */ 283 new_clp->lc_program = 0; 284 285 /* Lock out other nfsd threads */ 286 NFSLOCKV4ROOTMUTEX(); 287 nfsv4_relref(&nfsv4rootfs_lock); 288 do { 289 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 290 NFSV4ROOTLOCKMUTEXPTR, NULL); 291 } while (!igotlock); 292 NFSUNLOCKV4ROOTMUTEX(); 293 294 /* 295 * Search for a match in the client list. 296 */ 297 gotit = i = 0; 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)) { 302 gotit = 1; 303 break; 304 } 305 } 306 if (gotit == 0) 307 i++; 308 } 309 if (!gotit || 310 (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) { 311 if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) { 312 /* 313 * For NFSv4.1, if confirmp->lval[1] is non-zero, the 314 * client is trying to update a confirmed clientid. 315 */ 316 NFSLOCKV4ROOTMUTEX(); 317 nfsv4_unlock(&nfsv4rootfs_lock, 1); 318 NFSUNLOCKV4ROOTMUTEX(); 319 confirmp->lval[1] = 0; 320 error = NFSERR_NOENT; 321 goto out; 322 } 323 /* 324 * Get rid of the old one. 325 */ 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); 331 zapit = 1; 332 } 333 /* 334 * Add it after assigning a client id to it. 335 */ 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] = 339 ++confirm_index; 340 else 341 confirmp->qval = new_clp->lc_confirm.qval = 342 ++confirm_index; 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, 358 lc_hash); 359 nfsstatsv1.srvclients++; 360 nfsrv_openpluslock++; 361 nfsrv_clients++; 362 NFSLOCKV4ROOTMUTEX(); 363 nfsv4_unlock(&nfsv4rootfs_lock, 1); 364 NFSUNLOCKV4ROOTMUTEX(); 365 if (zapit) 366 nfsrv_zapclient(clp, p); 367 *new_clpp = NULL; 368 goto out; 369 } 370 371 /* 372 * Now, handle the cases where the id is already issued. 373 */ 374 if (nfsrv_notsamecredname(nd, clp)) { 375 /* 376 * Check to see if there is expired state that should go away. 377 */ 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); 382 } 383 384 /* 385 * If there is outstanding state, then reply NFSERR_CLIDINUSE per 386 * RFC3530 Sec. 8.1.2 last para. 387 */ 388 if (!LIST_EMPTY(&clp->lc_deleg)) { 389 hasstate = 1; 390 } else if (LIST_EMPTY(&clp->lc_open)) { 391 hasstate = 0; 392 } else { 393 hasstate = 0; 394 /* Look for an Open on the OpenOwner */ 395 LIST_FOREACH(stp, &clp->lc_open, ls_list) { 396 if (!LIST_EMPTY(&stp->ls_open)) { 397 hasstate = 1; 398 break; 399 } 400 } 401 } 402 if (hasstate) { 403 /* 404 * If the uid doesn't match, return NFSERR_CLIDINUSE after 405 * filling out the correct ipaddr and portnum. 406 */ 407 switch (clp->lc_req.nr_nam->sa_family) { 408 #ifdef INET 409 case AF_INET: 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; 414 break; 415 #endif 416 #ifdef INET6 417 case AF_INET6: 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; 422 break; 423 #endif 424 } 425 NFSLOCKV4ROOTMUTEX(); 426 nfsv4_unlock(&nfsv4rootfs_lock, 1); 427 NFSUNLOCKV4ROOTMUTEX(); 428 error = NFSERR_CLIDINUSE; 429 goto out; 430 } 431 } 432 433 if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) { 434 /* 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. 438 */ 439 LIST_REMOVE(clp, lc_hash); 440 441 /* 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); 444 if (ret != 0) 445 printf("nfsrv_setclient: verifier changed free" 446 " session failed=%d\n", ret); 447 } 448 449 new_clp->lc_flags |= LCL_NEEDSCONFIRM; 450 if ((nd->nd_flag & ND_NFSV41) != 0) 451 new_clp->lc_confirm.lval[0] = confirmp->lval[0] = 452 ++confirm_index; 453 else 454 confirmp->qval = new_clp->lc_confirm.qval = 455 ++confirm_index; 456 clientidp->lval[0] = new_clp->lc_clientid.lval[0] = 457 nfsrvboottime; 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(); 464 465 /* 466 * Save the state until confirmed. 467 */ 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, 475 ls_list); 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; 483 } 484 LIST_INIT(&new_clp->lc_session); 485 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, 486 lc_hash); 487 nfsstatsv1.srvclients++; 488 nfsrv_openpluslock++; 489 nfsrv_clients++; 490 NFSLOCKV4ROOTMUTEX(); 491 nfsv4_unlock(&nfsv4rootfs_lock, 1); 492 NFSUNLOCKV4ROOTMUTEX(); 493 494 /* 495 * Must wait until any outstanding callback on the old clp 496 * completes. 497 */ 498 NFSLOCKSTATE(); 499 while (clp->lc_cbref) { 500 clp->lc_flags |= LCL_WAKEUPWANTED; 501 (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, 502 "nfsd clp", 10 * hz); 503 } 504 NFSUNLOCKSTATE(); 505 nfsrv_zapclient(clp, p); 506 *new_clpp = NULL; 507 goto out; 508 } 509 510 /* 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; 516 } else { 517 /* 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. 521 */ 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; 548 } 549 LIST_INIT(&new_clp->lc_session); 550 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, 551 lc_hash); 552 nfsstatsv1.srvclients++; 553 nfsrv_openpluslock++; 554 nfsrv_clients++; 555 } 556 NFSLOCKV4ROOTMUTEX(); 557 nfsv4_unlock(&nfsv4rootfs_lock, 1); 558 NFSUNLOCKV4ROOTMUTEX(); 559 560 if ((nd->nd_flag & ND_NFSV41) == 0) { 561 /* 562 * Must wait until any outstanding callback on the old clp 563 * completes. 564 */ 565 NFSLOCKSTATE(); 566 while (clp->lc_cbref) { 567 clp->lc_flags |= LCL_WAKEUPWANTED; 568 (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, 569 "nfsdclp", 10 * hz); 570 } 571 NFSUNLOCKSTATE(); 572 nfsrv_zapclient(clp, p); 573 *new_clpp = NULL; 574 } 575 576 out: 577 NFSEXITCODE2(error, nd); 578 return (error); 579 } 580 581 /* 582 * Check to see if the client id exists and optionally confirm it. 583 */ 584 int 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) 588 { 589 struct nfsclient *clp; 590 struct nfsstate *stp; 591 int i; 592 struct nfsclienthashhead *hp; 593 int error = 0, igotlock, doneok; 594 struct nfssessionhash *shp; 595 struct nfsdsession *sep; 596 uint64_t sessid[2]; 597 static uint64_t next_sess = 0; 598 599 if (clpp) 600 *clpp = NULL; 601 if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 || 602 opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) { 603 error = NFSERR_STALECLIENTID; 604 goto out; 605 } 606 607 /* 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. 611 */ 612 if (opflags & CLOPS_CONFIRM) { 613 NFSLOCKV4ROOTMUTEX(); 614 nfsv4_relref(&nfsv4rootfs_lock); 615 do { 616 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 617 NFSV4ROOTLOCKMUTEXPTR, NULL); 618 } while (!igotlock); 619 /* 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. 622 */ 623 if ((nd->nd_flag & ND_NFSV41) != 0) { 624 sessid[0] = ++next_sess; 625 sessid[1] = clientid.qval; 626 } 627 NFSUNLOCKV4ROOTMUTEX(); 628 } else if (opflags != CLOPS_RENEW) { 629 NFSLOCKSTATE(); 630 } 631 632 /* 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) { 635 clp = NULL; 636 if ((nd->nd_flag & ND_HASSEQUENCE) != 0) { 637 shp = NFSSESSIONHASH(nd->nd_sessionid); 638 NFSLOCKSESSION(shp); 639 sep = nfsrv_findsession(nd->nd_sessionid); 640 if (sep != NULL) 641 clp = sep->sess_clp; 642 NFSUNLOCKSESSION(shp); 643 } 644 } else { 645 hp = NFSCLIENTHASH(clientid); 646 LIST_FOREACH(clp, hp, lc_hash) { 647 if (clp->lc_clientid.lval[1] == clientid.lval[1]) 648 break; 649 } 650 } 651 if (clp == NULL) { 652 if (opflags & CLOPS_CONFIRM) 653 error = NFSERR_STALECLIENTID; 654 else 655 error = NFSERR_EXPIRED; 656 } else if (clp->lc_flags & LCL_ADMINREVOKED) { 657 /* 658 * If marked admin revoked, just return the error. 659 */ 660 error = NFSERR_ADMINREVOKED; 661 } 662 if (error) { 663 if (opflags & CLOPS_CONFIRM) { 664 NFSLOCKV4ROOTMUTEX(); 665 nfsv4_unlock(&nfsv4rootfs_lock, 1); 666 NFSUNLOCKV4ROOTMUTEX(); 667 } else if (opflags != CLOPS_RENEW) { 668 NFSUNLOCKSTATE(); 669 } 670 goto out; 671 } 672 673 /* 674 * Perform any operations specified by the opflags. 675 */ 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; 684 685 if (!error) { 686 if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) == 687 LCL_NEEDSCONFIRM) { 688 /* 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. 692 */ 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); 698 } else { 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, 704 ls_list); 705 } 706 if ((nd->nd_flag & ND_NFSV41) != 0) 707 clp->lc_program = cbprogram; 708 } 709 clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN); 710 if (clp->lc_program) 711 clp->lc_flags |= LCL_NEEDSCBNULL; 712 /* For NFSv4.1, link the session onto the client. */ 713 if (nsep != NULL) { 714 /* Hold a reference on the xprt for a backchannel. */ 715 if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) 716 != 0) { 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); 723 nd->nd_xprt->xp_p2 = 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; 728 } else 729 nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN; 730 } 731 NFSBCOPY(sessid, nsep->sess_sessionid, 732 NFSX_V4SESSIONID); 733 NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid, 734 NFSX_V4SESSIONID); 735 shp = NFSSESSIONHASH(nsep->sess_sessionid); 736 NFSLOCKSTATE(); 737 NFSLOCKSESSION(shp); 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); 742 NFSUNLOCKSTATE(); 743 } 744 } 745 } else if (clp->lc_flags & LCL_NEEDSCONFIRM) { 746 error = NFSERR_EXPIRED; 747 } 748 749 /* 750 * If called by the Renew Op, we must check the principal. 751 */ 752 if (!error && (opflags & CLOPS_RENEWOP)) { 753 if (nfsrv_notsamecredname(nd, clp)) { 754 doneok = 0; 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) { 759 doneok = 1; 760 break; 761 } 762 } 763 } 764 if (!doneok) 765 error = NFSERR_ACCES; 766 } 767 if (!error && (clp->lc_flags & LCL_CBDOWN)) 768 error = NFSERR_CBPATHDOWN; 769 } 770 if ((!error || error == NFSERR_CBPATHDOWN) && 771 (opflags & CLOPS_RENEW)) { 772 clp->lc_expiry = nfsrv_leaseexpiry(); 773 } 774 if (opflags & CLOPS_CONFIRM) { 775 NFSLOCKV4ROOTMUTEX(); 776 nfsv4_unlock(&nfsv4rootfs_lock, 1); 777 NFSUNLOCKV4ROOTMUTEX(); 778 } else if (opflags != CLOPS_RENEW) { 779 NFSUNLOCKSTATE(); 780 } 781 if (clpp) 782 *clpp = clp; 783 784 out: 785 NFSEXITCODE2(error, nd); 786 return (error); 787 } 788 789 /* 790 * Perform the NFSv4.1 destroy clientid. 791 */ 792 int 793 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p) 794 { 795 struct nfsclient *clp; 796 struct nfsclienthashhead *hp; 797 int error = 0, i, igotlock; 798 799 if (nfsrvboottime != clientid.lval[0]) { 800 error = NFSERR_STALECLIENTID; 801 goto out; 802 } 803 804 /* Lock out other nfsd threads */ 805 NFSLOCKV4ROOTMUTEX(); 806 nfsv4_relref(&nfsv4rootfs_lock); 807 do { 808 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 809 NFSV4ROOTLOCKMUTEXPTR, NULL); 810 } while (igotlock == 0); 811 NFSUNLOCKV4ROOTMUTEX(); 812 813 hp = NFSCLIENTHASH(clientid); 814 LIST_FOREACH(clp, hp, lc_hash) { 815 if (clp->lc_clientid.lval[1] == clientid.lval[1]) 816 break; 817 } 818 if (clp == NULL) { 819 NFSLOCKV4ROOTMUTEX(); 820 nfsv4_unlock(&nfsv4rootfs_lock, 1); 821 NFSUNLOCKV4ROOTMUTEX(); 822 /* Just return ok, since it is gone. */ 823 goto out; 824 } 825 826 /* 827 * Free up all layouts on the clientid. Should the client return the 828 * layouts? 829 */ 830 nfsrv_freelayoutlist(clientid); 831 832 /* 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; 839 goto out; 840 } 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; 846 goto out; 847 } 848 849 /* 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); 858 out: 859 NFSEXITCODE2(error, nd); 860 return (error); 861 } 862 863 /* 864 * Called from the new nfssvc syscall to admin revoke a clientid. 865 * Returns 0 for success, error otherwise. 866 */ 867 int 868 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p) 869 { 870 struct nfsclient *clp = NULL; 871 int i, error = 0; 872 int gotit, igotlock; 873 874 /* 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 877 * file. 878 */ 879 NFSLOCKV4ROOTMUTEX(); 880 do { 881 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 882 NFSV4ROOTLOCKMUTEXPTR, NULL); 883 } while (!igotlock); 884 NFSUNLOCKV4ROOTMUTEX(); 885 886 /* 887 * Search for a match in the client list. 888 */ 889 gotit = i = 0; 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)) { 894 gotit = 1; 895 break; 896 } 897 } 898 i++; 899 } 900 if (!gotit) { 901 NFSLOCKV4ROOTMUTEX(); 902 nfsv4_unlock(&nfsv4rootfs_lock, 0); 903 NFSUNLOCKV4ROOTMUTEX(); 904 error = EPERM; 905 goto out; 906 } 907 908 /* 909 * Now, write out the revocation record 910 */ 911 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); 912 nfsrv_backupstable(); 913 914 /* 915 * and clear out the state, marking the clientid revoked. 916 */ 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(); 925 926 out: 927 NFSEXITCODE(error); 928 return (error); 929 } 930 931 /* 932 * Dump out stats for all clients. Called from nfssvc(2), that is used 933 * nfsstatsv1. 934 */ 935 void 936 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt) 937 { 938 struct nfsclient *clp; 939 int i = 0, cnt = 0; 940 941 /* 942 * First, get a reference on the nfsv4rootfs_lock so that an 943 * exclusive lock cannot be acquired while dumping the clients. 944 */ 945 NFSLOCKV4ROOTMUTEX(); 946 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); 947 NFSUNLOCKV4ROOTMUTEX(); 948 NFSLOCKSTATE(); 949 /* 950 * Rattle through the client lists until done. 951 */ 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]); 956 cnt++; 957 clp = LIST_NEXT(clp, lc_hash); 958 } 959 i++; 960 } 961 if (cnt < maxcnt) 962 dumpp[cnt].ndcl_clid.nclid_idlen = 0; 963 NFSUNLOCKSTATE(); 964 NFSLOCKV4ROOTMUTEX(); 965 nfsv4_relref(&nfsv4rootfs_lock); 966 NFSUNLOCKV4ROOTMUTEX(); 967 } 968 969 /* 970 * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd. 971 */ 972 static void 973 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp) 974 { 975 struct nfsstate *stp, *openstp, *lckownstp; 976 struct nfslock *lop; 977 sa_family_t af; 978 #ifdef INET 979 struct sockaddr_in *rin; 980 #endif 981 #ifdef INET6 982 struct sockaddr_in6 *rin6; 983 #endif 984 985 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; 993 switch (af) { 994 #ifdef INET 995 case AF_INET: 996 rin = (struct sockaddr_in *)clp->lc_req.nr_nam; 997 dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr; 998 break; 999 #endif 1000 #ifdef INET6 1001 case AF_INET6: 1002 rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; 1003 dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr; 1004 break; 1005 #endif 1006 } 1007 1008 /* 1009 * Now, scan the state lists and total up the opens and locks. 1010 */ 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++; 1019 } 1020 } 1021 } 1022 } 1023 1024 /* 1025 * and the delegation lists. 1026 */ 1027 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) { 1028 dumpp->ndcl_ndelegs++; 1029 } 1030 LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) { 1031 dumpp->ndcl_nolddelegs++; 1032 } 1033 } 1034 1035 /* 1036 * Dump out lock stats for a file. 1037 */ 1038 void 1039 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt, 1040 NFSPROC_T *p) 1041 { 1042 struct nfsstate *stp; 1043 struct nfslock *lop; 1044 int cnt = 0; 1045 struct nfslockfile *lfp; 1046 sa_family_t af; 1047 #ifdef INET 1048 struct sockaddr_in *rin; 1049 #endif 1050 #ifdef INET6 1051 struct sockaddr_in6 *rin6; 1052 #endif 1053 int ret; 1054 fhandle_t nfh; 1055 1056 ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p); 1057 /* 1058 * First, get a reference on the nfsv4rootfs_lock so that an 1059 * exclusive lock on it cannot be acquired while dumping the locks. 1060 */ 1061 NFSLOCKV4ROOTMUTEX(); 1062 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); 1063 NFSUNLOCKV4ROOTMUTEX(); 1064 NFSLOCKSTATE(); 1065 if (!ret) 1066 ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0); 1067 if (ret) { 1068 ldumpp[0].ndlck_clid.nclid_idlen = 0; 1069 NFSUNLOCKSTATE(); 1070 NFSLOCKV4ROOTMUTEX(); 1071 nfsv4_relref(&nfsv4rootfs_lock); 1072 NFSUNLOCKV4ROOTMUTEX(); 1073 return; 1074 } 1075 1076 /* 1077 * For each open share on file, dump it out. 1078 */ 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; 1096 switch (af) { 1097 #ifdef INET 1098 case AF_INET: 1099 rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; 1100 ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; 1101 break; 1102 #endif 1103 #ifdef INET6 1104 case AF_INET6: 1105 rin6 = (struct sockaddr_in6 *) 1106 stp->ls_clp->lc_req.nr_nam; 1107 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; 1108 break; 1109 #endif 1110 } 1111 stp = LIST_NEXT(stp, ls_file); 1112 cnt++; 1113 } 1114 1115 /* 1116 * and all locks. 1117 */ 1118 lop = LIST_FIRST(&lfp->lf_lock); 1119 while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) { 1120 stp = lop->lo_stp; 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, 1130 stp->ls_ownerlen); 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; 1136 switch (af) { 1137 #ifdef INET 1138 case AF_INET: 1139 rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; 1140 ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; 1141 break; 1142 #endif 1143 #ifdef INET6 1144 case AF_INET6: 1145 rin6 = (struct sockaddr_in6 *) 1146 stp->ls_clp->lc_req.nr_nam; 1147 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; 1148 break; 1149 #endif 1150 } 1151 lop = LIST_NEXT(lop, lo_lckfile); 1152 cnt++; 1153 } 1154 1155 /* 1156 * and the delegations. 1157 */ 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; 1171 switch (af) { 1172 #ifdef INET 1173 case AF_INET: 1174 rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; 1175 ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; 1176 break; 1177 #endif 1178 #ifdef INET6 1179 case AF_INET6: 1180 rin6 = (struct sockaddr_in6 *) 1181 stp->ls_clp->lc_req.nr_nam; 1182 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; 1183 break; 1184 #endif 1185 } 1186 stp = LIST_NEXT(stp, ls_file); 1187 cnt++; 1188 } 1189 1190 /* 1191 * If list isn't full, mark end of list by setting the client name 1192 * to zero length. 1193 */ 1194 if (cnt < maxcnt) 1195 ldumpp[cnt].ndlck_clid.nclid_idlen = 0; 1196 NFSUNLOCKSTATE(); 1197 NFSLOCKV4ROOTMUTEX(); 1198 nfsv4_relref(&nfsv4rootfs_lock); 1199 NFSUNLOCKV4ROOTMUTEX(); 1200 } 1201 1202 /* 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 1205 * nfsv4rootfs_lock. 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. 1210 */ 1211 void 1212 nfsrv_servertimer(void) 1213 { 1214 struct nfsclient *clp, *nclp; 1215 struct nfsstate *stp, *nstp; 1216 int got_ref, i; 1217 1218 /* 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. 1222 */ 1223 if (nfsboottime.tv_sec == 0) 1224 NFSSETBOOTTIME(nfsboottime); 1225 1226 /* 1227 * If server hasn't started yet, just return. 1228 */ 1229 NFSLOCKSTATE(); 1230 if (nfsrv_stablefirst.nsf_eograce == 0) { 1231 NFSUNLOCKSTATE(); 1232 return; 1233 } 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); 1239 NFSUNLOCKSTATE(); 1240 return; 1241 } 1242 1243 /* 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. 1247 */ 1248 NFSLOCKV4ROOTMUTEX(); 1249 got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock); 1250 NFSUNLOCKV4ROOTMUTEX(); 1251 if (got_ref == 0) { 1252 NFSUNLOCKSTATE(); 1253 return; 1254 } 1255 1256 /* 1257 * For each client... 1258 */ 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)) { 1271 /* 1272 * Lease has expired several nfsrv_lease times ago: 1273 * PLUS 1274 * - no state is associated with it 1275 * OR 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?) 1282 * OR 1283 * Lease has expired a very long time ago 1284 * OR 1285 * Lease has expired PLUS the number of opens + locks 1286 * has exceeded 90% of capacity 1287 * 1288 * --> Mark for expiry. The actual expiry will be done 1289 * by an nfsd sometime soon. 1290 */ 1291 clp->lc_flags |= LCL_EXPIREIT; 1292 nfsrv_stablefirst.nsf_flags |= 1293 (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT); 1294 } else { 1295 /* 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. 1301 */ 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)) { 1306 stp->ls_noopens++; 1307 if (stp->ls_noopens > NFSNOOPEN || 1308 (nfsrv_openpluslock * 2) > 1309 nfsrv_v4statelimit) 1310 nfsrv_stablefirst.nsf_flags |= 1311 NFSNSF_NOOPENS; 1312 } else { 1313 stp->ls_noopens = 0; 1314 } 1315 stp = nstp; 1316 } 1317 } 1318 } 1319 clp = nclp; 1320 } 1321 } 1322 NFSUNLOCKSTATE(); 1323 NFSLOCKV4ROOTMUTEX(); 1324 nfsv4_relref(&nfsv4rootfs_lock); 1325 NFSUNLOCKV4ROOTMUTEX(); 1326 } 1327 1328 /* 1329 * The following set of functions free up the various data structures. 1330 */ 1331 /* 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. 1335 */ 1336 void 1337 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p) 1338 { 1339 struct nfsstate *stp, *nstp; 1340 struct nfsdsession *sep, *nsep; 1341 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); 1347 } 1348 1349 /* 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.) 1354 */ 1355 void 1356 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p) 1357 { 1358 1359 #ifdef notyet 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); 1368 } 1369 #endif 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); 1375 NFSLOCKSTATE(); 1376 nfsstatsv1.srvclients--; 1377 nfsrv_openpluslock--; 1378 nfsrv_clients--; 1379 NFSUNLOCKSTATE(); 1380 } 1381 1382 /* 1383 * Free a list of delegation state structures. 1384 * (This function will also free all nfslockfile structures that no 1385 * longer have associated state.) 1386 */ 1387 void 1388 nfsrv_freedeleglist(struct nfsstatehead *sthp) 1389 { 1390 struct nfsstate *stp, *nstp; 1391 1392 LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) { 1393 nfsrv_freedeleg(stp); 1394 } 1395 LIST_INIT(sthp); 1396 } 1397 1398 /* 1399 * Free up a delegation. 1400 */ 1401 static void 1402 nfsrv_freedeleg(struct nfsstate *stp) 1403 { 1404 struct nfslockfile *lfp; 1405 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--; 1411 lfp = stp->ls_lfp; 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--; 1422 } 1423 1424 /* 1425 * This function frees an open owner and all associated opens. 1426 */ 1427 static void 1428 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p) 1429 { 1430 struct nfsstate *nstp, *tstp; 1431 1432 LIST_REMOVE(stp, ls_list); 1433 /* 1434 * Now, free all associated opens. 1435 */ 1436 nstp = LIST_FIRST(&stp->ls_open); 1437 while (nstp != LIST_END(&stp->ls_open)) { 1438 tstp = nstp; 1439 nstp = LIST_NEXT(nstp, ls_list); 1440 (void) nfsrv_freeopen(tstp, NULL, cansleep, p); 1441 } 1442 if (stp->ls_op) 1443 nfsrvd_derefcache(stp->ls_op); 1444 free(stp, M_NFSDSTATE); 1445 nfsstatsv1.srvopenowners--; 1446 nfsrv_openpluslock--; 1447 } 1448 1449 /* 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. 1454 */ 1455 static int 1456 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p) 1457 { 1458 struct nfsstate *nstp, *tstp; 1459 struct nfslockfile *lfp; 1460 int ret; 1461 1462 LIST_REMOVE(stp, ls_hash); 1463 LIST_REMOVE(stp, ls_list); 1464 LIST_REMOVE(stp, ls_file); 1465 1466 lfp = stp->ls_lfp; 1467 /* 1468 * Now, free all lockowners associated with this open. 1469 */ 1470 LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp) 1471 nfsrv_freelockowner(tstp, vp, cansleep, p); 1472 1473 /* 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(). 1480 */ 1481 if (cansleep != 0) 1482 NFSLOCKSTATE(); 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); 1489 ret = 1; 1490 } else 1491 ret = 0; 1492 if (cansleep != 0) 1493 NFSUNLOCKSTATE(); 1494 free(stp, M_NFSDSTATE); 1495 nfsstatsv1.srvopens--; 1496 nfsrv_openpluslock--; 1497 return (ret); 1498 } 1499 1500 /* 1501 * Frees a lockowner and all associated locks. 1502 */ 1503 static void 1504 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep, 1505 NFSPROC_T *p) 1506 { 1507 1508 LIST_REMOVE(stp, ls_hash); 1509 LIST_REMOVE(stp, ls_list); 1510 nfsrv_freeallnfslocks(stp, vp, cansleep, p); 1511 if (stp->ls_op) 1512 nfsrvd_derefcache(stp->ls_op); 1513 free(stp, M_NFSDSTATE); 1514 nfsstatsv1.srvlockowners--; 1515 nfsrv_openpluslock--; 1516 } 1517 1518 /* 1519 * Free all the nfs locks on a lockowner. 1520 */ 1521 static void 1522 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep, 1523 NFSPROC_T *p) 1524 { 1525 struct nfslock *lop, *nlop; 1526 struct nfsrollback *rlp, *nrlp; 1527 struct nfslockfile *lfp = NULL; 1528 int gottvp = 0; 1529 vnode_t tvp = NULL; 1530 uint64_t first, end; 1531 1532 if (vp != NULL) 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); 1537 /* 1538 * Since all locks should be for the same file, lfp should 1539 * not change. 1540 */ 1541 if (lfp == NULL) 1542 lfp = lop->lo_lfp; 1543 else if (lfp != lop->lo_lfp) 1544 panic("allnfslocks"); 1545 /* 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(). 1549 */ 1550 if (gottvp == 0) { 1551 if (nfsrv_dolocallocks == 0) 1552 tvp = NULL; 1553 else if (vp == NULL && cansleep != 0) { 1554 tvp = nfsvno_getvp(&lfp->lf_fh); 1555 if (tvp != NULL) 1556 NFSVOPUNLOCK(tvp); 1557 } else 1558 tvp = vp; 1559 gottvp = 1; 1560 } 1561 1562 if (tvp != NULL) { 1563 if (cansleep == 0) 1564 panic("allnfs2"); 1565 first = lop->lo_first; 1566 end = lop->lo_end; 1567 nfsrv_freenfslock(lop); 1568 nfsrv_localunlock(tvp, lfp, first, end, p); 1569 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, 1570 nrlp) 1571 free(rlp, M_NFSDROLLBACK); 1572 LIST_INIT(&lfp->lf_rollback); 1573 } else 1574 nfsrv_freenfslock(lop); 1575 lop = nlop; 1576 } 1577 if (vp == NULL && tvp != NULL) 1578 vrele(tvp); 1579 } 1580 1581 /* 1582 * Free an nfslock structure. 1583 */ 1584 static void 1585 nfsrv_freenfslock(struct nfslock *lop) 1586 { 1587 1588 if (lop->lo_lckfile.le_prev != NULL) { 1589 LIST_REMOVE(lop, lo_lckfile); 1590 nfsstatsv1.srvlocks--; 1591 nfsrv_openpluslock--; 1592 } 1593 LIST_REMOVE(lop, lo_lckowner); 1594 free(lop, M_NFSDLOCK); 1595 } 1596 1597 /* 1598 * This function frees an nfslockfile structure. 1599 */ 1600 static void 1601 nfsrv_freenfslockfile(struct nfslockfile *lfp) 1602 { 1603 1604 LIST_REMOVE(lfp, lf_hash); 1605 free(lfp, M_NFSDLOCKFILE); 1606 } 1607 1608 /* 1609 * This function looks up an nfsstate structure via stateid. 1610 */ 1611 static int 1612 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags, 1613 struct nfsstate **stpp) 1614 { 1615 struct nfsstate *stp; 1616 struct nfsstatehead *hp; 1617 int error = 0; 1618 1619 *stpp = NULL; 1620 hp = NFSSTATEHASH(clp, *stateidp); 1621 LIST_FOREACH(stp, hp, ls_hash) { 1622 if (!NFSBCMP(stp->ls_stateid.other, stateidp->other, 1623 NFSX_STATEIDOTHER)) 1624 break; 1625 } 1626 1627 /* 1628 * If no state id in list, return NFSERR_BADSTATEID. 1629 */ 1630 if (stp == LIST_END(hp)) { 1631 error = NFSERR_BADSTATEID; 1632 goto out; 1633 } 1634 *stpp = stp; 1635 1636 out: 1637 NFSEXITCODE(error); 1638 return (error); 1639 } 1640 1641 /* 1642 * This function gets an nfsstate structure via owner string. 1643 */ 1644 static void 1645 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp, 1646 struct nfsstate **stpp) 1647 { 1648 struct nfsstate *stp; 1649 1650 *stpp = NULL; 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)) { 1654 *stpp = stp; 1655 return; 1656 } 1657 } 1658 } 1659 1660 /* 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. 1670 */ 1671 int 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) 1677 { 1678 struct nfslock *lop; 1679 struct nfsstate *new_stp = *new_stpp; 1680 struct nfslock *new_lop = *new_lopp; 1681 struct nfsstate *tstp, *mystp, *nstp; 1682 int specialid = 0; 1683 struct nfslockfile *lfp; 1684 struct nfslock *other_lop = NULL; 1685 struct nfsstate *stp, *lckstp = NULL; 1686 struct nfsclient *clp = NULL; 1687 u_int32_t bits; 1688 int error = 0, haslock = 0, ret, reterr; 1689 int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0; 1690 fhandle_t nfh; 1691 uint64_t first, end; 1692 uint32_t lock_flags; 1693 1694 if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) { 1695 /* 1696 * Note the special cases of "all 1s" or "all 0s" stateids and 1697 * let reads with all 1s go ahead. 1698 */ 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) 1703 specialid = 1; 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) 1708 specialid = 2; 1709 } 1710 1711 /* 1712 * Check for restart conditions (client and server). 1713 */ 1714 error = nfsrv_checkrestart(clientid, new_stp->ls_flags, 1715 &new_stp->ls_stateid, specialid); 1716 if (error) 1717 goto out; 1718 1719 /* 1720 * Check for state resource limit exceeded. 1721 */ 1722 if ((new_stp->ls_flags & NFSLCK_LOCK) && 1723 nfsrv_openpluslock > nfsrv_v4statelimit) { 1724 error = NFSERR_RESOURCE; 1725 goto out; 1726 } 1727 1728 /* 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. 1733 */ 1734 tryagain: 1735 if (new_stp->ls_flags & NFSLCK_LOCK) 1736 other_lop = malloc(sizeof (struct nfslock), 1737 M_NFSDLOCK, M_WAITOK); 1738 filestruct_locked = 0; 1739 reterr = 0; 1740 lfp = NULL; 1741 1742 /* 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. 1751 */ 1752 getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p); 1753 NFSLOCKSTATE(); 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, 1758 &lfp, &nfh, 1); 1759 if (getlckret == 0) 1760 filestruct_locked = 1; 1761 } else 1762 getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL, 1763 &lfp, &nfh, 0); 1764 } 1765 if (getlckret != 0 && getlckret != -1) 1766 reterr = getlckret; 1767 1768 if (filestruct_locked != 0) { 1769 LIST_INIT(&lfp->lf_rollback); 1770 if ((new_stp->ls_flags & NFSLCK_LOCK)) { 1771 /* 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. 1777 */ 1778 NFSUNLOCKSTATE(); 1779 if (vnode_unlocked == 0) { 1780 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1"); 1781 vnode_unlocked = 1; 1782 NFSVOPUNLOCK(vp); 1783 } 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); 1787 NFSLOCKSTATE(); 1788 } 1789 } 1790 1791 if (specialid == 0) { 1792 if (new_stp->ls_flags & NFSLCK_TEST) { 1793 /* 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. 1797 */ 1798 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 1799 (nfsquad_t)((u_quad_t)0), 0, nd, p); 1800 1801 /* 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) 1806 */ 1807 if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED) 1808 error = 0; 1809 lckstp = new_stp; 1810 } else { 1811 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 1812 (nfsquad_t)((u_quad_t)0), 0, nd, p); 1813 if (error == 0) 1814 /* 1815 * Look up the stateid 1816 */ 1817 error = nfsrv_getstate(clp, &new_stp->ls_stateid, 1818 new_stp->ls_flags, &stp); 1819 /* 1820 * do some sanity checks for an unconfirmed open or a 1821 * stateid that refers to the wrong file, for an open stateid 1822 */ 1823 if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) && 1824 ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) || 1825 (getlckret == 0 && stp->ls_lfp != lfp))){ 1826 /* 1827 * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID 1828 * The only exception is using SETATTR with SIZE. 1829 * */ 1830 if ((new_stp->ls_flags & 1831 (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR) 1832 error = NFSERR_BADSTATEID; 1833 } 1834 1835 if (error == 0 && 1836 (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) && 1837 getlckret == 0 && stp->ls_lfp != lfp) 1838 error = NFSERR_BADSTATEID; 1839 1840 /* 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 1844 * on all files. 1845 * For now, log it as a diagnostic, instead of considering it 1846 * a BadStateid. 1847 */ 1848 if (error == 0 && (stp->ls_flags & 1849 (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 && 1850 getlckret == 0 && stp->ls_lfp != lfp) { 1851 #ifdef DIAGNOSTIC 1852 printf("Got a lock statid for different file open\n"); 1853 #endif 1854 /* 1855 error = NFSERR_BADSTATEID; 1856 */ 1857 } 1858 1859 if (error == 0) { 1860 if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) { 1861 /* 1862 * If haslock set, we've already checked the seqid. 1863 */ 1864 if (!haslock) { 1865 if (stp->ls_flags & NFSLCK_OPEN) 1866 error = nfsrv_checkseqid(nd, new_stp->ls_seq, 1867 stp->ls_openowner, new_stp->ls_op); 1868 else 1869 error = NFSERR_BADSTATEID; 1870 } 1871 if (!error) 1872 nfsrv_getowner(&stp->ls_open, new_stp, &lckstp); 1873 if (lckstp) 1874 /* 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. 1878 */ 1879 error = NFSERR_INVAL; 1880 else 1881 lckstp = new_stp; 1882 } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) { 1883 /* 1884 * If haslock set, ditto above. 1885 */ 1886 if (!haslock) { 1887 if (stp->ls_flags & NFSLCK_OPEN) 1888 error = NFSERR_BADSTATEID; 1889 else 1890 error = nfsrv_checkseqid(nd, new_stp->ls_seq, 1891 stp, new_stp->ls_op); 1892 } 1893 lckstp = stp; 1894 } else { 1895 lckstp = stp; 1896 } 1897 } 1898 /* 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.) 1907 */ 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; 1915 } 1916 } 1917 1918 /* 1919 * Now we can check for grace. 1920 */ 1921 if (!error) 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; 1926 /* 1927 * If we successfully Reclaimed state, note that. 1928 */ 1929 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error) 1930 nfsrv_markstable(clp); 1931 1932 /* 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 1941 * function call.) 1942 */ 1943 if (error != 0 || nd->nd_repstat != 0 || reterr != 0) { 1944 if (error == 0) { 1945 if (nd->nd_repstat != 0) 1946 error = nd->nd_repstat; 1947 else 1948 error = reterr; 1949 } 1950 if (filestruct_locked != 0) { 1951 /* Roll back local locks. */ 1952 NFSUNLOCKSTATE(); 1953 if (vnode_unlocked == 0) { 1954 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2"); 1955 vnode_unlocked = 1; 1956 NFSVOPUNLOCK(vp); 1957 } 1958 nfsrv_locallock_rollback(vp, lfp, p); 1959 NFSLOCKSTATE(); 1960 nfsrv_unlocklf(lfp); 1961 } 1962 NFSUNLOCKSTATE(); 1963 goto out; 1964 } 1965 1966 /* 1967 * Check the nfsrv_getlockfile return. 1968 * Returned -1 if no structure found. 1969 */ 1970 if (getlckret == -1) { 1971 error = NFSERR_EXPIRED; 1972 /* 1973 * Called from lockt, so no lock is OK. 1974 */ 1975 if (new_stp->ls_flags & NFSLCK_TEST) { 1976 error = 0; 1977 } else if (new_stp->ls_flags & 1978 (NFSLCK_CHECK | NFSLCK_SETATTR)) { 1979 /* 1980 * Called to check for a lock, OK if the stateid is all 1981 * 1s or all 0s, but there should be an nfsstate 1982 * otherwise. 1983 * (ie. If there is no open, I'll assume no share 1984 * deny bits.) 1985 */ 1986 if (specialid) 1987 error = 0; 1988 else 1989 error = NFSERR_BADSTATEID; 1990 } 1991 NFSUNLOCKSTATE(); 1992 goto out; 1993 } 1994 1995 /* 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. 2000 */ 2001 if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) { 2002 /* 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 2008 */ 2009 if (!specialid) { 2010 if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) { 2011 delegation = 1; 2012 mystp = stp; 2013 nfsrv_delaydelegtimeout(stp); 2014 } else if (stp->ls_flags & NFSLCK_OPEN) { 2015 mystp = stp; 2016 } else { 2017 mystp = stp->ls_openstp; 2018 } 2019 /* 2020 * If locking or checking, require correct access 2021 * bit set. 2022 */ 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. */ 2035 NFSUNLOCKSTATE(); 2036 if (vnode_unlocked == 0) { 2037 ASSERT_VOP_ELOCKED(vp, 2038 "nfsrv_lockctrl3"); 2039 vnode_unlocked = 1; 2040 NFSVOPUNLOCK(vp); 2041 } 2042 nfsrv_locallock_rollback(vp, lfp, p); 2043 NFSLOCKSTATE(); 2044 nfsrv_unlocklf(lfp); 2045 } 2046 NFSUNLOCKSTATE(); 2047 error = NFSERR_OPENMODE; 2048 goto out; 2049 } 2050 } else 2051 mystp = NULL; 2052 if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) { 2053 /* 2054 * Check for a conflicting deny bit. 2055 */ 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, 2064 vp, p); 2065 if (ret == 1) { 2066 /* 2067 * nfsrv_clientconflict unlocks state 2068 * when it returns non-zero. 2069 */ 2070 lckstp = NULL; 2071 goto tryagain; 2072 } 2073 if (ret == 0) 2074 NFSUNLOCKSTATE(); 2075 if (ret == 2) 2076 error = NFSERR_PERM; 2077 else 2078 error = NFSERR_OPENMODE; 2079 goto out; 2080 } 2081 } 2082 } 2083 2084 /* We're outta here */ 2085 NFSUNLOCKSTATE(); 2086 goto out; 2087 } 2088 } 2089 2090 /* 2091 * For setattr, just get rid of all the Delegations for other clients. 2092 */ 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); 2097 if (ret) { 2098 /* 2099 * nfsrv_cleandeleg() unlocks state when it 2100 * returns non-zero. 2101 */ 2102 if (ret == -1) { 2103 lckstp = NULL; 2104 goto tryagain; 2105 } 2106 error = ret; 2107 goto out; 2108 } 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))) { 2112 NFSUNLOCKSTATE(); 2113 goto out; 2114 } 2115 } 2116 2117 /* 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?) 2133 */ 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)))) { 2146 ret = 0; 2147 if (filestruct_locked != 0) { 2148 /* Roll back local locks. */ 2149 NFSUNLOCKSTATE(); 2150 if (vnode_unlocked == 0) { 2151 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4"); 2152 NFSVOPUNLOCK(vp); 2153 } 2154 nfsrv_locallock_rollback(vp, lfp, p); 2155 NFSLOCKSTATE(); 2156 nfsrv_unlocklf(lfp); 2157 NFSUNLOCKSTATE(); 2158 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2159 vnode_unlocked = 0; 2160 if (VN_IS_DOOMED(vp)) 2161 ret = NFSERR_SERVERFAULT; 2162 NFSLOCKSTATE(); 2163 } 2164 if (ret == 0) 2165 ret = nfsrv_delegconflict(tstp, &haslock, p, vp); 2166 if (ret) { 2167 /* 2168 * nfsrv_delegconflict unlocks state when it 2169 * returns non-zero, which it always does. 2170 */ 2171 if (other_lop) { 2172 free(other_lop, M_NFSDLOCK); 2173 other_lop = NULL; 2174 } 2175 if (ret == -1) { 2176 lckstp = NULL; 2177 goto tryagain; 2178 } 2179 error = ret; 2180 goto out; 2181 } 2182 /* Never gets here. */ 2183 } 2184 tstp = nstp; 2185 } 2186 2187 /* 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.) 2191 */ 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) { 2203 NFSUNLOCKSTATE(); 2204 if (vnode_unlocked == 0) { 2205 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5"); 2206 vnode_unlocked = 1; 2207 NFSVOPUNLOCK(vp); 2208 } 2209 /* Update the local locks. */ 2210 nfsrv_localunlock(vp, lfp, first, end, p); 2211 NFSLOCKSTATE(); 2212 nfsrv_unlocklf(lfp); 2213 } 2214 NFSUNLOCKSTATE(); 2215 goto out; 2216 } 2217 2218 /* 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 2223 */ 2224 if (!delegation) { 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))) { 2235 if (other_lop) { 2236 free(other_lop, M_NFSDLOCK); 2237 other_lop = NULL; 2238 } 2239 if (vnode_unlocked != 0) 2240 ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock, 2241 NULL, p); 2242 else 2243 ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock, 2244 vp, p); 2245 if (ret == 1) { 2246 if (filestruct_locked != 0) { 2247 if (vnode_unlocked == 0) { 2248 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6"); 2249 NFSVOPUNLOCK(vp); 2250 } 2251 /* Roll back local locks. */ 2252 nfsrv_locallock_rollback(vp, lfp, p); 2253 NFSLOCKSTATE(); 2254 nfsrv_unlocklf(lfp); 2255 NFSUNLOCKSTATE(); 2256 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2257 vnode_unlocked = 0; 2258 if (VN_IS_DOOMED(vp)) { 2259 error = NFSERR_SERVERFAULT; 2260 goto out; 2261 } 2262 } 2263 /* 2264 * nfsrv_clientconflict() unlocks state when it 2265 * returns non-zero. 2266 */ 2267 lckstp = NULL; 2268 goto tryagain; 2269 } 2270 /* 2271 * Found a conflicting lock, so record the conflict and 2272 * return the error. 2273 */ 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, 2282 cfp->cl_ownerlen); 2283 } 2284 if (ret == 2) 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; 2290 else 2291 error = NFSERR_DENIED; 2292 if (filestruct_locked != 0 && ret == 0) { 2293 /* Roll back local locks. */ 2294 NFSUNLOCKSTATE(); 2295 if (vnode_unlocked == 0) { 2296 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7"); 2297 vnode_unlocked = 1; 2298 NFSVOPUNLOCK(vp); 2299 } 2300 nfsrv_locallock_rollback(vp, lfp, p); 2301 NFSLOCKSTATE(); 2302 nfsrv_unlocklf(lfp); 2303 } 2304 if (ret == 0) 2305 NFSUNLOCKSTATE(); 2306 goto out; 2307 } 2308 } 2309 } 2310 2311 /* 2312 * We only get here if there was no lock that conflicted. 2313 */ 2314 if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) { 2315 NFSUNLOCKSTATE(); 2316 goto out; 2317 } 2318 2319 /* 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 2324 */ 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]; 2336 } else { 2337 /* 2338 * The new open_to_lock_owner case. 2339 * Link the new nfsstate into the lists. 2340 */ 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, 2355 lfp); 2356 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid), 2357 new_stp, ls_hash); 2358 LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list); 2359 *new_lopp = NULL; 2360 *new_stpp = NULL; 2361 nfsstatsv1.srvlockowners++; 2362 nfsrv_openpluslock++; 2363 } 2364 if (filestruct_locked != 0) { 2365 NFSUNLOCKSTATE(); 2366 nfsrv_locallock_commit(lfp, lock_flags, first, end); 2367 NFSLOCKSTATE(); 2368 nfsrv_unlocklf(lfp); 2369 } 2370 NFSUNLOCKSTATE(); 2371 2372 out: 2373 if (haslock) { 2374 NFSLOCKV4ROOTMUTEX(); 2375 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2376 NFSUNLOCKV4ROOTMUTEX(); 2377 } 2378 if (vnode_unlocked != 0) { 2379 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2380 if (error == 0 && VN_IS_DOOMED(vp)) 2381 error = NFSERR_SERVERFAULT; 2382 } 2383 if (other_lop) 2384 free(other_lop, M_NFSDLOCK); 2385 NFSEXITCODE2(error, nd); 2386 return (error); 2387 } 2388 2389 /* 2390 * Check for state errors for Open. 2391 * repstat is passed back out as an error if more critical errors 2392 * are not detected. 2393 */ 2394 int 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) 2398 { 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; 2404 2405 if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS) 2406 readonly = 1; 2407 /* 2408 * Check for restart conditions (client and server). 2409 */ 2410 error = nfsrv_checkrestart(clientid, new_stp->ls_flags, 2411 &new_stp->ls_stateid, 0); 2412 if (error) 2413 goto out; 2414 2415 /* 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. 2421 */ 2422 if (nfsrv_openpluslock > nfsrv_v4statelimit) { 2423 error = NFSERR_RESOURCE; 2424 goto out; 2425 } 2426 2427 tryagain: 2428 new_lfp = malloc(sizeof (struct nfslockfile), 2429 M_NFSDLOCKFILE, M_WAITOK); 2430 if (vp) 2431 getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp, 2432 NULL, p); 2433 NFSLOCKSTATE(); 2434 /* 2435 * Get the nfsclient structure. 2436 */ 2437 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 2438 (nfsquad_t)((u_quad_t)0), 0, nd, p); 2439 2440 /* 2441 * Look up the open owner. See if it needs confirmation and 2442 * check the seq#, as required. 2443 */ 2444 if (!error) 2445 nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp); 2446 2447 if (!error && ownerstp) { 2448 error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp, 2449 new_stp->ls_op); 2450 /* 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. 2454 */ 2455 if (error == NFSERR_BADSEQID && 2456 (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM)) 2457 error = 0; 2458 } 2459 2460 /* 2461 * Check for grace. 2462 */ 2463 if (!error) 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; 2468 2469 /* 2470 * If none of the above errors occurred, let repstat be 2471 * returned. 2472 */ 2473 if (repstat && !error) 2474 error = repstat; 2475 if (error) { 2476 NFSUNLOCKSTATE(); 2477 if (haslock) { 2478 NFSLOCKV4ROOTMUTEX(); 2479 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2480 NFSUNLOCKV4ROOTMUTEX(); 2481 } 2482 free(new_lfp, M_NFSDLOCKFILE); 2483 goto out; 2484 } 2485 2486 /* 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.) 2489 */ 2490 if (vp == NULL) { 2491 NFSUNLOCKSTATE(); 2492 free(new_lfp, M_NFSDLOCKFILE); 2493 goto out; 2494 } 2495 2496 /* 2497 * Get the structure for the underlying file. 2498 */ 2499 if (getfhret) 2500 error = getfhret; 2501 else 2502 error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp, 2503 NULL, 0); 2504 if (new_lfp) 2505 free(new_lfp, M_NFSDLOCKFILE); 2506 if (error) { 2507 NFSUNLOCKSTATE(); 2508 if (haslock) { 2509 NFSLOCKV4ROOTMUTEX(); 2510 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2511 NFSUNLOCKV4ROOTMUTEX(); 2512 } 2513 goto out; 2514 } 2515 2516 /* 2517 * Search for a conflicting open/share. 2518 */ 2519 if (new_stp->ls_flags & NFSLCK_DELEGCUR) { 2520 /* 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 2526 * match? 2527 */ 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, 2534 NFSX_STATEIDOTHER)) 2535 break; 2536 } 2537 if (stp == LIST_END(&lfp->lf_deleg) || 2538 ((new_stp->ls_flags & NFSLCK_WRITEACCESS) && 2539 (stp->ls_flags & NFSLCK_DELEGREAD))) { 2540 NFSUNLOCKSTATE(); 2541 if (haslock) { 2542 NFSLOCKV4ROOTMUTEX(); 2543 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2544 NFSUNLOCKV4ROOTMUTEX(); 2545 } 2546 error = NFSERR_EXPIRED; 2547 goto out; 2548 } 2549 } 2550 2551 /* 2552 * Check for access/deny bit conflicts. I check for the same 2553 * owner as well, in case the client didn't bother. 2554 */ 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); 2562 if (ret == 1) { 2563 /* 2564 * nfsrv_clientconflict() unlocks 2565 * state when it returns non-zero. 2566 */ 2567 goto tryagain; 2568 } 2569 if (ret == 2) 2570 error = NFSERR_PERM; 2571 else if (new_stp->ls_flags & NFSLCK_RECLAIM) 2572 error = NFSERR_RECLAIMCONFLICT; 2573 else 2574 error = NFSERR_SHAREDENIED; 2575 if (ret == 0) 2576 NFSUNLOCKSTATE(); 2577 if (haslock) { 2578 NFSLOCKV4ROOTMUTEX(); 2579 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2580 NFSUNLOCKV4ROOTMUTEX(); 2581 } 2582 goto out; 2583 } 2584 } 2585 2586 /* 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(). 2605 */ 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); 2616 if (ret) { 2617 /* 2618 * nfsrv_delegconflict() unlocks state 2619 * when it returns non-zero. 2620 */ 2621 if (ret == -1) 2622 goto tryagain; 2623 error = ret; 2624 goto out; 2625 } 2626 } 2627 stp = nstp; 2628 } 2629 } 2630 NFSUNLOCKSTATE(); 2631 if (haslock) { 2632 NFSLOCKV4ROOTMUTEX(); 2633 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2634 NFSUNLOCKV4ROOTMUTEX(); 2635 } 2636 2637 out: 2638 NFSEXITCODE2(error, nd); 2639 return (error); 2640 } 2641 2642 /* 2643 * Open control function to create/update open state for an open. 2644 */ 2645 int 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) 2650 { 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; 2660 2661 if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS) 2662 readonly = 1; 2663 /* 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. 2668 */ 2669 error = nfsrv_checkrestart(clientid, new_stp->ls_flags, 2670 &new_stp->ls_stateid, 0); 2671 if (error) { 2672 printf("Nfsd: openctrl unexpected restart err=%d\n", 2673 error); 2674 error = NFSERR_EXPIRED; 2675 goto out; 2676 } 2677 2678 clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK); 2679 tryagain: 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, 2687 NULL, p); 2688 NFSLOCKSTATE(); 2689 /* 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() 2694 * or 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 2701 * expiration. 2702 */ 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) && 2706 clp->lc_program) { 2707 /* 2708 * This happens on the first open for a client 2709 * that supports callbacks. 2710 */ 2711 NFSUNLOCKSTATE(); 2712 /* 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 2718 * harmless. 2719 */ 2720 cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL, 2721 NULL, 0, NULL, NULL, NULL, 0, p); 2722 NFSLOCKSTATE(); 2723 clp->lc_flags &= ~LCL_NEEDSCBNULL; 2724 if (!cbret) 2725 clp->lc_flags |= LCL_CALLBACKSON; 2726 } 2727 2728 /* 2729 * Look up the open owner. See if it needs confirmation and 2730 * check the seq#, as required. 2731 */ 2732 if (!error) 2733 nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp); 2734 2735 if (error) { 2736 NFSUNLOCKSTATE(); 2737 printf("Nfsd: openctrl unexpected state err=%d\n", 2738 error); 2739 free(new_lfp, M_NFSDLOCKFILE); 2740 free(new_open, M_NFSDSTATE); 2741 free(new_deleg, M_NFSDSTATE); 2742 if (haslock) { 2743 NFSLOCKV4ROOTMUTEX(); 2744 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2745 NFSUNLOCKV4ROOTMUTEX(); 2746 } 2747 error = NFSERR_EXPIRED; 2748 goto out; 2749 } 2750 2751 if (new_stp->ls_flags & NFSLCK_RECLAIM) 2752 nfsrv_markstable(clp); 2753 2754 /* 2755 * Get the structure for the underlying file. 2756 */ 2757 if (getfhret) 2758 error = getfhret; 2759 else 2760 error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp, 2761 NULL, 0); 2762 if (new_lfp) 2763 free(new_lfp, M_NFSDLOCKFILE); 2764 if (error) { 2765 NFSUNLOCKSTATE(); 2766 printf("Nfsd openctrl unexpected getlockfile err=%d\n", 2767 error); 2768 free(new_open, M_NFSDSTATE); 2769 free(new_deleg, M_NFSDSTATE); 2770 if (haslock) { 2771 NFSLOCKV4ROOTMUTEX(); 2772 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2773 NFSUNLOCKV4ROOTMUTEX(); 2774 } 2775 goto out; 2776 } 2777 2778 /* 2779 * Search for a conflicting open/share. 2780 */ 2781 if (new_stp->ls_flags & NFSLCK_DELEGCUR) { 2782 /* 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 2788 * match? 2789 */ 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, 2796 NFSX_STATEIDOTHER)) 2797 break; 2798 } 2799 if (stp == LIST_END(&lfp->lf_deleg) || 2800 ((new_stp->ls_flags & NFSLCK_WRITEACCESS) && 2801 (stp->ls_flags & NFSLCK_DELEGREAD))) { 2802 NFSUNLOCKSTATE(); 2803 printf("Nfsd openctrl unexpected expiry\n"); 2804 free(new_open, M_NFSDSTATE); 2805 free(new_deleg, M_NFSDSTATE); 2806 if (haslock) { 2807 NFSLOCKV4ROOTMUTEX(); 2808 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2809 NFSUNLOCKV4ROOTMUTEX(); 2810 } 2811 error = NFSERR_EXPIRED; 2812 goto out; 2813 } 2814 2815 /* 2816 * Don't issue a Delegation, since one already exists and 2817 * delay delegation timeout, as required. 2818 */ 2819 delegate = 0; 2820 nfsrv_delaydelegtimeout(stp); 2821 } 2822 2823 /* 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. 2829 */ 2830 LIST_FOREACH(stp, &lfp->lf_open, ls_file) { 2831 if (ownerstp && stp->ls_openowner == ownerstp) 2832 openstp = stp; 2833 if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) { 2834 /* 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. 2838 */ 2839 if (clp != stp->ls_clp) { 2840 if ((stp->ls_flags & NFSLCK_SHAREBITS) == 2841 NFSLCK_READACCESS) 2842 writedeleg = 0; 2843 else 2844 delegate = 0; 2845 } 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); 2851 if (ret == 1) { 2852 /* 2853 * nfsrv_clientconflict() unlocks state 2854 * when it returns non-zero. 2855 */ 2856 free(new_open, M_NFSDSTATE); 2857 free(new_deleg, M_NFSDSTATE); 2858 openstp = NULL; 2859 goto tryagain; 2860 } 2861 if (ret == 2) 2862 error = NFSERR_PERM; 2863 else if (new_stp->ls_flags & NFSLCK_RECLAIM) 2864 error = NFSERR_RECLAIMCONFLICT; 2865 else 2866 error = NFSERR_SHAREDENIED; 2867 if (ret == 0) 2868 NFSUNLOCKSTATE(); 2869 if (haslock) { 2870 NFSLOCKV4ROOTMUTEX(); 2871 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2872 NFSUNLOCKV4ROOTMUTEX(); 2873 } 2874 free(new_open, M_NFSDSTATE); 2875 free(new_deleg, M_NFSDSTATE); 2876 printf("nfsd openctrl unexpected client cnfl\n"); 2877 goto out; 2878 } 2879 } 2880 } 2881 2882 /* 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.) 2899 */ 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)) 2905 writedeleg = 0; 2906 else 2907 delegate = 0; 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) { 2913 delegate = 2; 2914 } else { 2915 ret = nfsrv_delegconflict(stp, &haslock, p, vp); 2916 if (ret) { 2917 /* 2918 * nfsrv_delegconflict() unlocks state 2919 * when it returns non-zero. 2920 */ 2921 printf("Nfsd openctrl unexpected deleg cnfl\n"); 2922 free(new_open, M_NFSDSTATE); 2923 free(new_deleg, M_NFSDSTATE); 2924 if (ret == -1) { 2925 openstp = NULL; 2926 goto tryagain; 2927 } 2928 error = ret; 2929 goto out; 2930 } 2931 } 2932 } 2933 stp = nstp; 2934 } 2935 } 2936 2937 /* 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. 2943 */ 2944 if (new_stp->ls_flags & NFSLCK_DELEGPREV) { 2945 /* 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 2949 * NFSERR_EXPIRED. 2950 */ 2951 LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) { 2952 if (stp->ls_lfp == lfp) { 2953 /* Found it */ 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; 2972 else 2973 *rflagsp |= NFSV4OPEN_READDELEGATE; 2974 clp->lc_delegtime = NFSD_MONOSEC + 2975 nfsrv_lease + NFSRV_LEASEDELTA; 2976 2977 /* 2978 * Now, do the associated open. 2979 */ 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)| 2985 NFSLCK_OPEN; 2986 if (stp->ls_flags & NFSLCK_DELEGWRITE) 2987 new_open->ls_flags |= (NFSLCK_READACCESS | 2988 NFSLCK_WRITEACCESS); 2989 else 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), 2997 new_open, ls_hash); 2998 /* 2999 * and handle the open owner 3000 */ 3001 if (ownerstp) { 3002 new_open->ls_openowner = ownerstp; 3003 LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list); 3004 } else { 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); 3012 *new_stpp = NULL; 3013 nfsstatsv1.srvopenowners++; 3014 nfsrv_openpluslock++; 3015 } 3016 openstp = new_open; 3017 new_open = NULL; 3018 nfsstatsv1.srvopens++; 3019 nfsrv_openpluslock++; 3020 break; 3021 } 3022 } 3023 if (stp == LIST_END(&clp->lc_olddeleg)) 3024 error = NFSERR_EXPIRED; 3025 } else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) { 3026 /* 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 3030 * openowner. 3031 */ 3032 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { 3033 if (stp->ls_clp == clp) 3034 break; 3035 } 3036 if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) { 3037 /* 3038 * This is the Claim_Previous case with a delegation 3039 * type != Delegate_None. 3040 */ 3041 /* 3042 * First, add the delegation. (Although we must issue the 3043 * delegation, we can also ask for an immediate return.) 3044 */ 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++; 3057 } else { 3058 new_deleg->ls_flags = (NFSLCK_DELEGREAD | 3059 NFSLCK_READACCESS); 3060 *rflagsp |= NFSV4OPEN_READDELEGATE; 3061 } 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); 3071 new_deleg = NULL; 3072 if (delegate == 2 || nfsrv_issuedelegs == 0 || 3073 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) != 3074 LCL_CALLBACKSON || 3075 NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) || 3076 !NFSVNO_DELEGOK(vp)) 3077 *rflagsp |= NFSV4OPEN_RECALL; 3078 nfsstatsv1.srvdelegates++; 3079 nfsrv_openpluslock++; 3080 nfsrv_delegatecnt++; 3081 3082 /* 3083 * Now, do the associated open. 3084 */ 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) | 3090 NFSLCK_OPEN; 3091 if (new_stp->ls_flags & NFSLCK_DELEGWRITE) 3092 new_open->ls_flags |= (NFSLCK_READACCESS | 3093 NFSLCK_WRITEACCESS); 3094 else 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), 3102 new_open, ls_hash); 3103 /* 3104 * and handle the open owner 3105 */ 3106 if (ownerstp) { 3107 new_open->ls_openowner = ownerstp; 3108 LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list); 3109 } else { 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); 3117 *new_stpp = NULL; 3118 nfsstatsv1.srvopenowners++; 3119 nfsrv_openpluslock++; 3120 } 3121 openstp = new_open; 3122 new_open = NULL; 3123 nfsstatsv1.srvopens++; 3124 nfsrv_openpluslock++; 3125 } else { 3126 error = NFSERR_RECLAIMCONFLICT; 3127 } 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) | 3139 NFSLCK_OPEN; 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); 3145 stp->ls_lfp = lfp; 3146 } 3147 openstp = stp; 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; 3154 3155 /* 3156 * This is where we can choose to issue a delegation. 3157 */ 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)) != 3170 LCL_CALLBACKSON) 3171 *rflagsp |= NFSV4OPEN_WDCONTENTION; 3172 else { 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); 3193 new_deleg = NULL; 3194 nfsstatsv1.srvdelegates++; 3195 nfsrv_openpluslock++; 3196 nfsrv_delegatecnt++; 3197 } 3198 } else { 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)| 3204 NFSLCK_OPEN; 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), 3213 new_open, ls_hash); 3214 openstp = new_open; 3215 new_open = NULL; 3216 nfsstatsv1.srvopens++; 3217 nfsrv_openpluslock++; 3218 3219 /* 3220 * This is where we can choose to issue a delegation. 3221 */ 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)) != 3231 LCL_CALLBACKSON) 3232 *rflagsp |= NFSV4OPEN_WDCONTENTION; 3233 else { 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++; 3248 } else { 3249 new_deleg->ls_flags = (NFSLCK_DELEGREAD | 3250 NFSLCK_READACCESS); 3251 *rflagsp |= NFSV4OPEN_READDELEGATE; 3252 } 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); 3262 new_deleg = NULL; 3263 nfsstatsv1.srvdelegates++; 3264 nfsrv_openpluslock++; 3265 nfsrv_delegatecnt++; 3266 } 3267 } 3268 } else { 3269 /* 3270 * New owner case. Start the open_owner sequence with a 3271 * Needs confirmation (unless a reclaim) and hang the 3272 * new open off it. 3273 */ 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) | 3279 NFSLCK_OPEN; 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; 3291 3292 /* 3293 * This is where we can choose to issue a delegation. 3294 */ 3295 if (delegate && nfsrv_issuedelegs && 3296 (writedeleg || readonly) && 3297 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) == 3298 LCL_CALLBACKSON && 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) == 3318 0)) { 3319 new_deleg->ls_flags = 3320 (NFSLCK_DELEGWRITE | 3321 NFSLCK_READACCESS | 3322 NFSLCK_WRITEACCESS); 3323 *rflagsp |= NFSV4OPEN_WRITEDELEGATE; 3324 nfsrv_writedelegcnt++; 3325 } else { 3326 new_deleg->ls_flags = 3327 (NFSLCK_DELEGREAD | 3328 NFSLCK_READACCESS); 3329 *rflagsp |= NFSV4OPEN_READDELEGATE; 3330 } 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, 3337 ls_file); 3338 LIST_INSERT_HEAD(NFSSTATEHASH(clp, 3339 new_deleg->ls_stateid), new_deleg, ls_hash); 3340 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, 3341 ls_list); 3342 new_deleg = NULL; 3343 nfsstatsv1.srvdelegates++; 3344 nfsrv_openpluslock++; 3345 nfsrv_delegatecnt++; 3346 } 3347 /* 3348 * Since NFSv4.1 never does an OpenConfirm, the first 3349 * open state will be acquired here. 3350 */ 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); 3355 gotstate = 1; 3356 } 3357 } else { 3358 *rflagsp |= NFSV4OPEN_RESULTCONFIRM; 3359 new_stp->ls_flags = NFSLCK_NEEDSCONFIRM; 3360 } 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), 3367 new_open, ls_hash); 3368 openstp = new_open; 3369 new_open = NULL; 3370 *new_stpp = NULL; 3371 nfsstatsv1.srvopens++; 3372 nfsrv_openpluslock++; 3373 nfsstatsv1.srvopenowners++; 3374 nfsrv_openpluslock++; 3375 } 3376 if (!error) { 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]; 3381 } 3382 NFSUNLOCKSTATE(); 3383 if (haslock) { 3384 NFSLOCKV4ROOTMUTEX(); 3385 nfsv4_unlock(&nfsv4rootfs_lock, 1); 3386 NFSUNLOCKV4ROOTMUTEX(); 3387 } 3388 if (new_open) 3389 free(new_open, M_NFSDSTATE); 3390 if (new_deleg) 3391 free(new_deleg, M_NFSDSTATE); 3392 3393 /* 3394 * If the NFSv4.1 client just acquired its first open, write a timestamp 3395 * to the stable storage file. 3396 */ 3397 if (gotstate != 0) { 3398 nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p); 3399 nfsrv_backupstable(); 3400 } 3401 3402 out: 3403 free(clidp, M_TEMP); 3404 NFSEXITCODE2(error, nd); 3405 return (error); 3406 } 3407 3408 /* 3409 * Open update. Does the confirm, downgrade and close. 3410 */ 3411 int 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) 3415 { 3416 struct nfsstate *stp; 3417 struct nfsclient *clp; 3418 struct nfslockfile *lfp; 3419 u_int32_t bits; 3420 int error = 0, gotstate = 0, len = 0; 3421 u_char *clidp = NULL; 3422 3423 /* 3424 * Check for restart conditions (client and server). 3425 */ 3426 error = nfsrv_checkrestart(clientid, new_stp->ls_flags, 3427 &new_stp->ls_stateid, 0); 3428 if (error) 3429 goto out; 3430 3431 clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK); 3432 NFSLOCKSTATE(); 3433 /* 3434 * Get the open structure via clientid and stateid. 3435 */ 3436 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 3437 (nfsquad_t)((u_quad_t)0), 0, nd, p); 3438 if (!error) 3439 error = nfsrv_getstate(clp, &new_stp->ls_stateid, 3440 new_stp->ls_flags, &stp); 3441 3442 /* 3443 * Sanity check the open. 3444 */ 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; 3451 3452 if (!error) 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; 3464 else 3465 error = NFSERR_INVAL; 3466 } 3467 3468 if (error) { 3469 /* 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#. 3474 */ 3475 if (error == NFSERR_BADSEQID && 3476 (new_stp->ls_flags & NFSLCK_CONFIRM) && 3477 nfsrv_nootherstate(stp)) 3478 nfsrv_freeopenowner(stp->ls_openowner, 0, p); 3479 NFSUNLOCKSTATE(); 3480 goto out; 3481 } 3482 3483 /* 3484 * Set the return stateid. 3485 */ 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]; 3492 /* 3493 * Now, handle the three cases. 3494 */ 3495 if (new_stp->ls_flags & NFSLCK_CONFIRM) { 3496 /* 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? 3499 */ 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); 3511 gotstate = 1; 3512 } 3513 NFSUNLOCKSTATE(); 3514 } else if (new_stp->ls_flags & NFSLCK_CLOSE) { 3515 lfp = stp->ls_lfp; 3516 if (retwriteaccessp != NULL) { 3517 if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0) 3518 *retwriteaccessp = 1; 3519 else 3520 *retwriteaccessp = 0; 3521 } 3522 if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) { 3523 /* Get the lf lock */ 3524 nfsrv_locklf(lfp); 3525 NFSUNLOCKSTATE(); 3526 ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate"); 3527 NFSVOPUNLOCK(vp); 3528 if (nfsrv_freeopen(stp, vp, 1, p) == 0) { 3529 NFSLOCKSTATE(); 3530 nfsrv_unlocklf(lfp); 3531 NFSUNLOCKSTATE(); 3532 } 3533 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3534 } else { 3535 (void) nfsrv_freeopen(stp, NULL, 0, p); 3536 NFSUNLOCKSTATE(); 3537 } 3538 } else { 3539 /* 3540 * Update the share bits, making sure that the new set are a 3541 * subset of the old ones. 3542 */ 3543 bits = (new_stp->ls_flags & NFSLCK_SHAREBITS); 3544 if (~(stp->ls_flags) & bits) { 3545 NFSUNLOCKSTATE(); 3546 error = NFSERR_INVAL; 3547 goto out; 3548 } 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; 3554 NFSUNLOCKSTATE(); 3555 } 3556 3557 /* 3558 * If the client just confirmed its first open, write a timestamp 3559 * to the stable storage file. 3560 */ 3561 if (gotstate != 0) { 3562 nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p); 3563 nfsrv_backupstable(); 3564 } 3565 3566 out: 3567 free(clidp, M_TEMP); 3568 NFSEXITCODE2(error, nd); 3569 return (error); 3570 } 3571 3572 /* 3573 * Delegation update. Does the purge and return. 3574 */ 3575 int 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) 3579 { 3580 struct nfsstate *stp; 3581 struct nfsclient *clp; 3582 int error = 0; 3583 fhandle_t fh; 3584 3585 /* 3586 * Do a sanity check against the file handle for DelegReturn. 3587 */ 3588 if (vp) { 3589 error = nfsvno_getfh(vp, &fh, p); 3590 if (error) 3591 goto out; 3592 } 3593 /* 3594 * Check for restart conditions (client and server). 3595 */ 3596 if (op == NFSV4OP_DELEGRETURN) 3597 error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN, 3598 stateidp, 0); 3599 else 3600 error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE, 3601 stateidp, 0); 3602 3603 NFSLOCKSTATE(); 3604 /* 3605 * Get the open structure via clientid and stateid. 3606 */ 3607 if (!error) 3608 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 3609 (nfsquad_t)((u_quad_t)0), 0, nd, p); 3610 if (error) { 3611 if (error == NFSERR_CBPATHDOWN) 3612 error = 0; 3613 if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN) 3614 error = NFSERR_STALESTATEID; 3615 } 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; 3621 } 3622 /* 3623 * NFSERR_EXPIRED means that the state has gone away, 3624 * so Delegations have been purged. Just return ok. 3625 */ 3626 if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) { 3627 NFSUNLOCKSTATE(); 3628 error = 0; 3629 goto out; 3630 } 3631 if (error) { 3632 NFSUNLOCKSTATE(); 3633 goto out; 3634 } 3635 3636 if (op == NFSV4OP_DELEGRETURN) { 3637 if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh, 3638 sizeof (fhandle_t))) { 3639 NFSUNLOCKSTATE(); 3640 error = NFSERR_BADSTATEID; 3641 goto out; 3642 } 3643 if (retwriteaccessp != NULL) { 3644 if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0) 3645 *retwriteaccessp = 1; 3646 else 3647 *retwriteaccessp = 0; 3648 } 3649 nfsrv_freedeleg(stp); 3650 } else { 3651 nfsrv_freedeleglist(&clp->lc_olddeleg); 3652 } 3653 NFSUNLOCKSTATE(); 3654 error = 0; 3655 3656 out: 3657 NFSEXITCODE(error); 3658 return (error); 3659 } 3660 3661 /* 3662 * Release lock owner. 3663 */ 3664 int 3665 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid, 3666 NFSPROC_T *p) 3667 { 3668 struct nfsstate *stp, *nstp, *openstp, *ownstp; 3669 struct nfsclient *clp; 3670 int error = 0; 3671 3672 /* 3673 * Check for restart conditions (client and server). 3674 */ 3675 error = nfsrv_checkrestart(clientid, new_stp->ls_flags, 3676 &new_stp->ls_stateid, 0); 3677 if (error) 3678 goto out; 3679 3680 NFSLOCKSTATE(); 3681 /* 3682 * Get the lock owner by name. 3683 */ 3684 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 3685 (nfsquad_t)((u_quad_t)0), 0, NULL, p); 3686 if (error) { 3687 NFSUNLOCKSTATE(); 3688 goto out; 3689 } 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); 3695 /* 3696 * If the owner matches, check for locks and 3697 * then free or return an error. 3698 */ 3699 if (stp->ls_ownerlen == new_stp->ls_ownerlen && 3700 !NFSBCMP(stp->ls_owner, new_stp->ls_owner, 3701 stp->ls_ownerlen)){ 3702 if (LIST_EMPTY(&stp->ls_lock)) { 3703 nfsrv_freelockowner(stp, NULL, 0, p); 3704 } else { 3705 NFSUNLOCKSTATE(); 3706 error = NFSERR_LOCKSHELD; 3707 goto out; 3708 } 3709 } 3710 stp = nstp; 3711 } 3712 } 3713 } 3714 NFSUNLOCKSTATE(); 3715 3716 out: 3717 NFSEXITCODE(error); 3718 return (error); 3719 } 3720 3721 /* 3722 * Get the file handle for a lock structure. 3723 */ 3724 static int 3725 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp, 3726 fhandle_t *nfhp, NFSPROC_T *p) 3727 { 3728 fhandle_t *fhp = NULL; 3729 int error; 3730 3731 /* 3732 * For lock, use the new nfslock structure, otherwise just 3733 * a fhandle_t on the stack. 3734 */ 3735 if (flags & NFSLCK_OPEN) { 3736 KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL")); 3737 fhp = &new_lfp->lf_fh; 3738 } else if (nfhp) { 3739 fhp = nfhp; 3740 } else { 3741 panic("nfsrv_getlockfh"); 3742 } 3743 error = nfsvno_getfh(vp, fhp, p); 3744 NFSEXITCODE(error); 3745 return (error); 3746 } 3747 3748 /* 3749 * Get an nfs lock structure. Allocate one, as required, and return a 3750 * pointer to it. 3751 * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock. 3752 */ 3753 static int 3754 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp, 3755 struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit) 3756 { 3757 struct nfslockfile *lfp; 3758 fhandle_t *fhp = NULL, *tfhp; 3759 struct nfslockhashhead *hp; 3760 struct nfslockfile *new_lfp = NULL; 3761 3762 /* 3763 * For lock, use the new nfslock structure, otherwise just 3764 * a fhandle_t on the stack. 3765 */ 3766 if (flags & NFSLCK_OPEN) { 3767 new_lfp = *new_lfpp; 3768 fhp = &new_lfp->lf_fh; 3769 } else if (nfhp) { 3770 fhp = nfhp; 3771 } else { 3772 panic("nfsrv_getlockfile"); 3773 } 3774 3775 hp = NFSLOCKHASH(fhp); 3776 LIST_FOREACH(lfp, hp, lf_hash) { 3777 tfhp = &lfp->lf_fh; 3778 if (NFSVNO_CMPFH(fhp, tfhp)) { 3779 if (lockit) 3780 nfsrv_locklf(lfp); 3781 *lfpp = lfp; 3782 return (0); 3783 } 3784 } 3785 if (!(flags & NFSLCK_OPEN)) 3786 return (-1); 3787 3788 /* 3789 * No match, so chain the new one into the list. 3790 */ 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); 3800 *lfpp = new_lfp; 3801 *new_lfpp = NULL; 3802 return (0); 3803 } 3804 3805 /* 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. 3809 */ 3810 static void 3811 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop, 3812 struct nfsstate *stp, struct nfslockfile *lfp) 3813 { 3814 struct nfslock *lop, *nlop; 3815 3816 new_lop->lo_stp = stp; 3817 new_lop->lo_lfp = lfp; 3818 3819 if (stp != NULL) { 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); 3825 } else { 3826 nlop = LIST_NEXT(lop, lo_lckfile); 3827 while (nlop != LIST_END(&lfp->lf_lock) && 3828 nlop->lo_first < new_lop->lo_first) { 3829 lop = nlop; 3830 nlop = LIST_NEXT(lop, lo_lckfile); 3831 } 3832 LIST_INSERT_AFTER(lop, new_lop, lo_lckfile); 3833 } 3834 } else { 3835 new_lop->lo_lckfile.le_prev = NULL; /* list not used */ 3836 } 3837 3838 /* 3839 * Insert after insert_lop, which is overloaded as stp or lfp for 3840 * an empty list. 3841 */ 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); 3846 else 3847 LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner); 3848 if (stp != NULL) { 3849 nfsstatsv1.srvlocks++; 3850 nfsrv_openpluslock++; 3851 } 3852 } 3853 3854 /* 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 3859 * at by other_lopp. 3860 */ 3861 static void 3862 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp, 3863 struct nfslock **other_lopp, struct nfslockfile *lfp) 3864 { 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; 3869 u_int64_t tmp; 3870 3871 /* 3872 * Work down the list until the lock is merged. 3873 */ 3874 if (new_lop->lo_flags & NFSLCK_UNLOCK) 3875 unlock = 1; 3876 if (stp != NULL) { 3877 ilop = (struct nfslock *)stp; 3878 lop = LIST_FIRST(&stp->ls_lock); 3879 } else { 3880 ilop = (struct nfslock *)lfp; 3881 lop = LIST_FIRST(&lfp->lf_locallock); 3882 } 3883 while (lop != NULL) { 3884 /* 3885 * Only check locks for this file that aren't before the start of 3886 * new lock's range. 3887 */ 3888 if (lop->lo_lfp == lfp) { 3889 myfile = 1; 3890 if (lop->lo_end >= new_lop->lo_first) { 3891 if (new_lop->lo_end < lop->lo_first) { 3892 /* 3893 * If the new lock ends before the start of the 3894 * current lock's range, no merge, just insert 3895 * the new lock. 3896 */ 3897 break; 3898 } 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)) { 3902 /* 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 3907 * unlock. 3908 */ 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; 3913 tlop = lop; 3914 lop = LIST_NEXT(lop, lo_lckowner); 3915 nfsrv_freenfslock(tlop); 3916 continue; 3917 } 3918 3919 /* 3920 * All these cases are for contiguous locks that are not the 3921 * same type, so they can't be merged. 3922 */ 3923 if (new_lop->lo_first <= lop->lo_first) { 3924 /* 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.) 3930 */ 3931 lop->lo_first = new_lop->lo_end; 3932 break; 3933 } 3934 if (new_lop->lo_end >= lop->lo_end) { 3935 /* 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. 3942 */ 3943 lop->lo_end = new_lop->lo_first; 3944 ilop = lop; 3945 lop = LIST_NEXT(lop, lo_lckowner); 3946 continue; 3947 } 3948 /* 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. 3957 */ 3958 tmp = new_lop->lo_first; 3959 if (other_lop == NULL) { 3960 if (!unlock) 3961 panic("nfsd srv update unlock"); 3962 other_lop = new_lop; 3963 *new_lopp = NULL; 3964 } 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; 3970 lop->lo_end = tmp; 3971 nfsrv_insertlock(other_lop, lop, stp, lfp); 3972 *other_lopp = NULL; 3973 ilop = lop; 3974 break; 3975 } 3976 } 3977 ilop = lop; 3978 lop = LIST_NEXT(lop, lo_lckowner); 3979 if (myfile && (lop == NULL || lop->lo_lfp != lfp)) 3980 break; 3981 } 3982 3983 /* 3984 * Insert the new lock in the list at the appropriate place. 3985 */ 3986 if (!unlock) { 3987 nfsrv_insertlock(new_lop, ilop, stp, lfp); 3988 *new_lopp = NULL; 3989 } 3990 } 3991 3992 /* 3993 * This function handles sequencing of locks, etc. 3994 * It returns an error that indicates what the caller should do. 3995 */ 3996 static int 3997 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid, 3998 struct nfsstate *stp, struct nfsrvcache *op) 3999 { 4000 int error = 0; 4001 4002 if ((nd->nd_flag & ND_NFSV41) != 0) 4003 /* NFSv4.1 ignores the open_seqid and lock_seqid. */ 4004 goto out; 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"); 4012 } 4013 if ((stp->ls_seq + 1) == seqid) { 4014 if (stp->ls_op) 4015 nfsrvd_derefcache(stp->ls_op); 4016 stp->ls_op = op; 4017 nfsrvd_refcache(op); 4018 stp->ls_seq = seqid; 4019 goto out; 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; 4027 goto out; 4028 } 4029 nd->nd_rp = stp->ls_op; 4030 nd->nd_rp->rc_flag |= RC_INPROG; 4031 nfsrvd_delcache(op); 4032 error = NFSERR_REPLYFROMCACHE; 4033 goto out; 4034 } 4035 error = NFSERR_BADSEQID; 4036 4037 out: 4038 NFSEXITCODE2(error, nd); 4039 return (error); 4040 } 4041 4042 /* 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. 4049 */ 4050 int 4051 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp) 4052 { 4053 u_int32_t *tl; 4054 u_char *cp, *cp2; 4055 int i, j, maxalen = 0, minalen = 0; 4056 sa_family_t af; 4057 #ifdef INET 4058 struct sockaddr_in *rin = NULL, *sin; 4059 #endif 4060 #ifdef INET6 4061 struct sockaddr_in6 *rin6 = NULL, *sin6; 4062 #endif 4063 u_char *addr; 4064 int error = 0, cantparse = 0; 4065 union { 4066 in_addr_t ival; 4067 u_char cval[4]; 4068 } ip; 4069 union { 4070 in_port_t sval; 4071 u_char cval[2]; 4072 } port; 4073 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; 4078 af = AF_UNSPEC; 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); 4083 if (error) 4084 goto nfsmout; 4085 #ifdef INET 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; 4090 af = AF_INET; 4091 } else if (!strcmp(addr, "udp")) { 4092 clp->lc_req.nr_sotype = SOCK_DGRAM; 4093 clp->lc_req.nr_soproto = IPPROTO_UDP; 4094 af = AF_INET; 4095 } 4096 #endif 4097 #ifdef INET6 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; 4103 af = AF_INET6; 4104 } else if (!strcmp(addr, "udp6")) { 4105 clp->lc_req.nr_sotype = SOCK_DGRAM; 4106 clp->lc_req.nr_soproto = IPPROTO_UDP; 4107 af = AF_INET6; 4108 } 4109 } 4110 #endif 4111 if (af == AF_UNSPEC) { 4112 cantparse = 1; 4113 } 4114 } else { 4115 cantparse = 1; 4116 if (i > 0) { 4117 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 4118 if (error) 4119 goto nfsmout; 4120 } 4121 } 4122 /* 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. 4131 */ 4132 switch (af) { 4133 #ifdef INET 4134 case AF_INET: 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; 4139 minalen = 7 + 4; 4140 break; 4141 #endif 4142 #ifdef INET6 4143 case AF_INET6: 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; 4148 minalen = 3 + 4; 4149 break; 4150 #endif 4151 } 4152 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 4153 i = fxdr_unsigned(int, *tl); 4154 if (i < 0) { 4155 error = NFSERR_BADXDR; 4156 goto nfsmout; 4157 } else if (i == 0) { 4158 cantparse = 1; 4159 } else if (!cantparse && i <= maxalen && i >= minalen) { 4160 error = nfsrv_mtostr(nd, addr, i); 4161 if (error) 4162 goto nfsmout; 4163 4164 /* 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. 4168 */ 4169 cp = NULL; 4170 switch (af) { 4171 #ifdef INET6 4172 /* 4173 * For AF_INET6, first parse the host address. 4174 */ 4175 case AF_INET6: 4176 cp = strchr(addr, '.'); 4177 if (cp != NULL) { 4178 *cp++ = '\0'; 4179 if (inet_pton(af, addr, &rin6->sin6_addr) == 1) 4180 i = 4; 4181 else { 4182 cp = NULL; 4183 cantparse = 1; 4184 } 4185 } 4186 break; 4187 #endif 4188 #ifdef INET 4189 case AF_INET: 4190 cp = addr; 4191 i = 0; 4192 break; 4193 #endif 4194 } 4195 while (cp != NULL && *cp && i < 6) { 4196 cp2 = cp; 4197 while (*cp2 && *cp2 != '.') 4198 cp2++; 4199 if (*cp2) 4200 *cp2++ = '\0'; 4201 else if (i != 5) { 4202 cantparse = 1; 4203 break; 4204 } 4205 j = nfsrv_getipnumber(cp); 4206 if (j >= 0) { 4207 if (i < 4) 4208 ip.cval[3 - i] = j; 4209 else 4210 port.cval[5 - i] = j; 4211 } else { 4212 cantparse = 1; 4213 break; 4214 } 4215 cp = cp2; 4216 i++; 4217 } 4218 if (!cantparse) { 4219 /* 4220 * The host address INADDR_ANY is (mis)used to indicate 4221 * "there is no valid callback address". 4222 */ 4223 switch (af) { 4224 #ifdef INET6 4225 case AF_INET6: 4226 if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr, 4227 &in6addr_any)) 4228 rin6->sin6_port = htons(port.sval); 4229 else 4230 cantparse = 1; 4231 break; 4232 #endif 4233 #ifdef INET 4234 case AF_INET: 4235 if (ip.ival != INADDR_ANY) { 4236 rin->sin_addr.s_addr = htonl(ip.ival); 4237 rin->sin_port = htons(port.sval); 4238 } else { 4239 cantparse = 1; 4240 } 4241 break; 4242 #endif 4243 } 4244 } 4245 } else { 4246 cantparse = 1; 4247 if (i > 0) { 4248 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 4249 if (error) 4250 goto nfsmout; 4251 } 4252 } 4253 if (cantparse) { 4254 switch (nd->nd_nam->sa_family) { 4255 #ifdef INET 4256 case AF_INET: 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; 4263 break; 4264 #endif 4265 #ifdef INET6 4266 case AF_INET6: 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; 4273 break; 4274 #endif 4275 } 4276 clp->lc_program = 0; 4277 } 4278 nfsmout: 4279 free(addr, M_TEMP); 4280 NFSEXITCODE2(error, nd); 4281 return (error); 4282 } 4283 4284 /* 4285 * Turn a string of up to three decimal digits into a number. Return -1 upon 4286 * error. 4287 */ 4288 static int 4289 nfsrv_getipnumber(u_char *cp) 4290 { 4291 int i = 0, j = 0; 4292 4293 while (*cp) { 4294 if (j > 2 || *cp < '0' || *cp > '9') 4295 return (-1); 4296 i *= 10; 4297 i += (*cp - '0'); 4298 cp++; 4299 j++; 4300 } 4301 if (i < 256) 4302 return (i); 4303 return (-1); 4304 } 4305 4306 /* 4307 * This function checks for restart conditions. 4308 */ 4309 static int 4310 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags, 4311 nfsv4stateid_t *stateidp, int specialid) 4312 { 4313 int ret = 0; 4314 4315 /* 4316 * First check for a server restart. Open, LockT, ReleaseLockOwner 4317 * and DelegPurge have a clientid, the rest a stateid. 4318 */ 4319 if (flags & 4320 (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) { 4321 if (clientid.lval[0] != nfsrvboottime) { 4322 ret = NFSERR_STALECLIENTID; 4323 goto out; 4324 } 4325 } else if (stateidp->other[0] != nfsrvboottime && 4326 specialid == 0) { 4327 ret = NFSERR_STALESTATEID; 4328 goto out; 4329 } 4330 4331 /* 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.) 4335 */ 4336 if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST))) 4337 goto out; 4338 4339 NFSLOCKSTATE(); 4340 ret = nfsrv_checkgrace(NULL, NULL, flags); 4341 NFSUNLOCKSTATE(); 4342 4343 out: 4344 NFSEXITCODE(ret); 4345 return (ret); 4346 } 4347 4348 /* 4349 * Check for grace. 4350 */ 4351 static int 4352 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp, 4353 u_int32_t flags) 4354 { 4355 int error = 0, notreclaimed; 4356 struct nfsrv_stable *sp; 4357 4358 if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE | 4359 NFSNSF_GRACEOVER)) == 0) { 4360 /* 4361 * First, check to see if all of the clients have done a 4362 * ReclaimComplete. If so, grace can end now. 4363 */ 4364 notreclaimed = 0; 4365 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { 4366 if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) { 4367 notreclaimed = 1; 4368 break; 4369 } 4370 } 4371 if (notreclaimed == 0) 4372 nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER | 4373 NFSNSF_NEEDLOCK); 4374 } 4375 4376 if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) { 4377 if (flags & NFSLCK_RECLAIM) { 4378 error = NFSERR_NOGRACE; 4379 goto out; 4380 } 4381 } else { 4382 if (!(flags & NFSLCK_RECLAIM)) { 4383 error = NFSERR_GRACE; 4384 goto out; 4385 } 4386 if (nd != NULL && clp != NULL && 4387 (nd->nd_flag & ND_NFSV41) != 0 && 4388 (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) { 4389 error = NFSERR_NOGRACE; 4390 goto out; 4391 } 4392 4393 /* 4394 * If grace is almost over and we are still getting Reclaims, 4395 * extend grace a bit. 4396 */ 4397 if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) > 4398 nfsrv_stablefirst.nsf_eograce) 4399 nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC + 4400 NFSRV_LEASEDELTA; 4401 } 4402 4403 out: 4404 NFSEXITCODE(error); 4405 return (error); 4406 } 4407 4408 /* 4409 * Do a server callback. 4410 * The "trunc" argument is slightly overloaded and refers to different 4411 * boolean arguments for CBRECALL and CBLAYOUTRECALL. 4412 */ 4413 static int 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) 4417 { 4418 struct mbuf *m; 4419 u_int32_t *tl; 4420 struct nfsrv_descript *nd; 4421 struct ucred *cred; 4422 int error = 0; 4423 u_int32_t callback; 4424 struct nfsdsession *sep = NULL; 4425 uint64_t tval; 4426 bool dotls; 4427 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) { 4432 NFSUNLOCKSTATE(); 4433 panic("docallb"); 4434 } 4435 clp->lc_cbref++; 4436 4437 /* 4438 * Fill the callback program# and version into the request 4439 * structure for newnfs_connect() to use. 4440 */ 4441 clp->lc_req.nr_prog = clp->lc_program; 4442 #ifdef notnow 4443 if ((clp->lc_flags & LCL_NFSV41) != 0) 4444 clp->lc_req.nr_vers = NFSV41_CBVERS; 4445 else 4446 #endif 4447 clp->lc_req.nr_vers = NFSV4_CBVERS; 4448 4449 /* 4450 * First, fill in some of the fields of nd and cr. 4451 */ 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; 4459 nd->nd_repstat = 0; 4460 cred->cr_uid = clp->lc_uid; 4461 cred->cr_gid = clp->lc_gid; 4462 callback = clp->lc_callback; 4463 NFSUNLOCKSTATE(); 4464 cred->cr_ngroups = 1; 4465 4466 /* 4467 * Get the first mbuf for the request. 4468 */ 4469 MGET(m, M_WAITOK, MT_DATA); 4470 m->m_len = 0; 4471 nd->nd_mreq = nd->nd_mb = m; 4472 nd->nd_bpos = mtod(m, caddr_t); 4473 4474 /* 4475 * and build the callback request. 4476 */ 4477 if (procnum == NFSV4OP_CBGETATTR) { 4478 nd->nd_procnum = NFSV4PROC_CBCOMPOUND; 4479 error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR, 4480 "CB Getattr", &sep); 4481 if (error != 0) { 4482 m_freem(nd->nd_mreq); 4483 goto errout; 4484 } 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, 4490 "CB Recall", &sep); 4491 if (error != 0) { 4492 m_freem(nd->nd_mreq); 4493 goto errout; 4494 } 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, 4498 NFSX_STATEIDOTHER); 4499 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED); 4500 if (trunc) 4501 *tl = newnfs_true; 4502 else 4503 *tl = newnfs_false; 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); 4511 if (error != 0) { 4512 m_freem(nd->nd_mreq); 4513 goto errout; 4514 } 4515 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 4516 *tl++ = txdr_unsigned(laytype); 4517 *tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY); 4518 if (trunc) 4519 *tl++ = newnfs_true; 4520 else 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); 4525 tval = 0; 4526 txdr_hyper(tval, tl); tl += 2; 4527 tval = UINT64_MAX; 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); 4537 if (error != 0) { 4538 m_freem(nd->nd_mreq); 4539 goto errout; 4540 } 4541 } 4542 } else { 4543 error = NFSERR_SERVERFAULT; 4544 m_freem(nd->nd_mreq); 4545 goto errout; 4546 } 4547 4548 /* 4549 * Call newnfs_connect(), as required, and then newnfs_request(). 4550 */ 4551 dotls = false; 4552 if ((clp->lc_flags & LCL_TLSCB) != 0) 4553 dotls = true; 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, 4561 NULL, 1, dotls); 4562 else 4563 error = newnfs_connect(NULL, &clp->lc_req, cred, 4564 NULL, 3, dotls); 4565 } 4566 newnfs_sndunlock(&clp->lc_req.nr_lock); 4567 NFSD_DEBUG(4, "aft sndunlock=%d\n", error); 4568 if (!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); 4576 else { 4577 /* 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. 4584 */ 4585 printf("nfsrv_docallback: no xprt\n"); 4586 error = ECONNREFUSED; 4587 } 4588 NFSD_DEBUG(4, "aft newnfs_request=%d\n", error); 4589 nfsrv_freesession(sep, NULL); 4590 } else 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); 4594 } 4595 errout: 4596 NFSFREECRED(cred); 4597 4598 /* 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.) 4603 */ 4604 if (error) { 4605 /* 4606 * Mark the callback pathway down, which disabled issuing 4607 * of delegations and gets Renew to return NFSERR_CBPATHDOWN. 4608 */ 4609 NFSLOCKSTATE(); 4610 clp->lc_flags |= LCL_CBDOWN; 4611 NFSUNLOCKSTATE(); 4612 } else { 4613 /* 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 4617 * flakey.) 4618 */ 4619 NFSLOCKSTATE(); 4620 if (clp->lc_flags & LCL_CBDOWN) 4621 clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON); 4622 NFSUNLOCKSTATE(); 4623 if (nd->nd_repstat) { 4624 error = nd->nd_repstat; 4625 NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n", 4626 procnum, error); 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, 4630 p, NULL); 4631 m_freem(nd->nd_mrep); 4632 } 4633 NFSLOCKSTATE(); 4634 clp->lc_cbref--; 4635 if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) { 4636 clp->lc_flags &= ~LCL_WAKEUPWANTED; 4637 wakeup(clp); 4638 } 4639 NFSUNLOCKSTATE(); 4640 4641 free(nd, M_TEMP); 4642 NFSEXITCODE(error); 4643 return (error); 4644 } 4645 4646 /* 4647 * Set up the compound RPC for the callback. 4648 */ 4649 static int 4650 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp, 4651 uint32_t callback, int op, const char *optag, struct nfsdsession **sepp) 4652 { 4653 uint32_t *tl; 4654 int error, len; 4655 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); 4662 else 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); 4668 if (error != 0) 4669 return (error); 4670 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 4671 *tl = txdr_unsigned(op); 4672 } else { 4673 *tl++ = txdr_unsigned(NFSV4_MINORVERSION); 4674 *tl++ = txdr_unsigned(callback); 4675 *tl++ = txdr_unsigned(1); 4676 *tl = txdr_unsigned(op); 4677 } 4678 return (0); 4679 } 4680 4681 /* 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.) 4688 */ 4689 static u_int32_t 4690 nfsrv_nextclientindex(void) 4691 { 4692 static u_int32_t client_index = 0; 4693 4694 client_index++; 4695 if (client_index != 0) 4696 return (client_index); 4697 4698 printf("%s: out of clientids\n", __func__); 4699 return (client_index); 4700 } 4701 4702 /* 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. 4707 */ 4708 static u_int32_t 4709 nfsrv_nextstateindex(struct nfsclient *clp) 4710 { 4711 struct nfsstate *stp; 4712 int i; 4713 u_int32_t canuse, min_index, max_index; 4714 4715 if (!(clp->lc_flags & LCL_INDEXNOTOK)) { 4716 clp->lc_stateindex++; 4717 if (clp->lc_stateindex != clp->lc_statemaxindex) 4718 return (clp->lc_stateindex); 4719 } 4720 4721 /* 4722 * Yuck, we've hit the end. 4723 * Look for a new min and max. 4724 */ 4725 min_index = 0; 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]; 4732 } else { 4733 if (stp->ls_stateid.other[2] > min_index) 4734 min_index = stp->ls_stateid.other[2]; 4735 } 4736 } 4737 } 4738 4739 /* 4740 * Yikes, highly unlikely, but I'll handle it anyhow. 4741 */ 4742 if (min_index == 0x80000000 && max_index == 0x80000001) { 4743 canuse = 0; 4744 /* 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.) 4749 */ 4750 tryagain: 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) { 4754 canuse++; 4755 goto tryagain; 4756 } 4757 } 4758 } 4759 clp->lc_flags |= LCL_INDEXNOTOK; 4760 return (canuse); 4761 } 4762 4763 /* 4764 * Ok to start again from min + 1. 4765 */ 4766 clp->lc_stateindex = min_index + 1; 4767 clp->lc_statemaxindex = max_index; 4768 clp->lc_flags &= ~LCL_INDEXNOTOK; 4769 return (clp->lc_stateindex); 4770 } 4771 4772 /* 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 4782 * be recognized.) 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. 4791 * 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. 4795 * 4796 * Subsequent records are appended when the first state is issued to 4797 * a client and when state is revoked for a client. 4798 * 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. 4803 */ 4804 4805 /* 4806 * Read in the stable storage file. Called by nfssvc() before the nfsd 4807 * processes start servicing requests. 4808 */ 4809 void 4810 nfsrv_setupstable(NFSPROC_T *p) 4811 { 4812 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; 4813 struct nfsrv_stable *sp, *nsp; 4814 struct nfst_rec *tsp; 4815 int error, i, tryagain; 4816 off_t off = 0; 4817 ssize_t aresid, len; 4818 4819 /* 4820 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without 4821 * a reboot, so state has not been lost. 4822 */ 4823 if (sf->nsf_flags & NFSNSF_UPDATEDONE) 4824 return; 4825 /* 4826 * Set Grace over just until the file reads successfully. 4827 */ 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) 4833 return; 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) 4839 return; 4840 4841 /* 4842 * Now, read in the boottimes. 4843 */ 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; 4853 return; 4854 } 4855 4856 /* 4857 * Make sure this nfsrvboottime is different from all recorded 4858 * previous ones. 4859 */ 4860 do { 4861 tryagain = 0; 4862 for (i = 0; i < sf->nsf_numboots; i++) { 4863 if (nfsrvboottime == sf->nsf_bootvals[i]) { 4864 nfsrvboottime++; 4865 tryagain = 1; 4866 break; 4867 } 4868 } 4869 } while (tryagain); 4870 4871 sf->nsf_flags |= NFSNSF_OK; 4872 off += (sf->nsf_numboots * sizeof (time_t)); 4873 4874 /* 4875 * Read through the file, building a list of records for grace 4876 * checking. 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. 4880 */ 4881 tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) + 4882 NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK); 4883 do { 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)))) { 4890 /* 4891 * Yuck, the file has been corrupted, so just return 4892 * after clearing out any restart state, so the grace period 4893 * is over. 4894 */ 4895 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) { 4896 LIST_REMOVE(sp, nst_list); 4897 free(sp, M_TEMP); 4898 } 4899 free(tsp, M_TEMP); 4900 sf->nsf_flags &= ~NFSNSF_OK; 4901 free(sf->nsf_bootvals, M_TEMP); 4902 sf->nsf_bootvals = NULL; 4903 return; 4904 } 4905 if (len > 0) { 4906 off += sizeof (struct nfst_rec) + tsp->len - 1; 4907 /* 4908 * Search the list for a matching client. 4909 */ 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)) 4913 break; 4914 } 4915 if (sp == LIST_END(&sf->nsf_head)) { 4916 sp = (struct nfsrv_stable *)malloc(tsp->len + 4917 sizeof (struct nfsrv_stable) - 1, M_TEMP, 4918 M_WAITOK); 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); 4922 } else { 4923 if (tsp->flag == NFSNST_REVOKE) 4924 sp->nst_flag |= NFSNST_REVOKE; 4925 else 4926 /* 4927 * A subsequent timestamp indicates the client 4928 * did a setclientid/confirm and any previous 4929 * revoke is no longer relevant. 4930 */ 4931 sp->nst_flag &= ~NFSNST_REVOKE; 4932 } 4933 } 4934 } while (len > 0); 4935 free(tsp, M_TEMP); 4936 sf->nsf_flags = NFSNSF_OK; 4937 sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease + 4938 NFSRV_LEASEDELTA; 4939 } 4940 4941 /* 4942 * Update the stable storage file, now that the grace period is over. 4943 */ 4944 void 4945 nfsrv_updatestable(NFSPROC_T *p) 4946 { 4947 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; 4948 struct nfsrv_stable *sp, *nsp; 4949 int i; 4950 struct nfsvattr nva; 4951 vnode_t vp; 4952 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000) 4953 mount_t mp = NULL; 4954 #endif 4955 int error; 4956 4957 if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE)) 4958 return; 4959 sf->nsf_flags |= NFSNSF_UPDATEDONE; 4960 /* 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 4966 */ 4967 if (sf->nsf_bootvals) { 4968 sf->nsf_numboots++; 4969 for (i = sf->nsf_numboots - 2; i >= 0; i--) 4970 sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i]; 4971 } else { 4972 sf->nsf_numboots = 1; 4973 sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t), 4974 M_TEMP, M_WAITOK); 4975 } 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, 4984 NULL); 4985 NFSVOPUNLOCK(vp); 4986 } else 4987 error = EPERM; 4988 vn_finished_write(mp); 4989 if (!error) 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); 4993 if (!error) 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; 5001 if (error) { 5002 sf->nsf_flags &= ~NFSNSF_OK; 5003 printf("EEK! Can't write NfsV4 stable storage file\n"); 5004 return; 5005 } 5006 sf->nsf_flags |= NFSNSF_OK; 5007 5008 /* 5009 * Loop through the list and write out timestamp records for 5010 * any clients that successfully reclaimed state. 5011 */ 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; 5017 } 5018 LIST_REMOVE(sp, nst_list); 5019 free(sp, M_TEMP); 5020 } 5021 nfsrv_backupstable(); 5022 } 5023 5024 /* 5025 * Append a record to the stable storage file. 5026 */ 5027 void 5028 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p) 5029 { 5030 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; 5031 struct nfst_rec *sp; 5032 int error; 5033 5034 if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL) 5035 return; 5036 sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) + 5037 len - 1, M_TEMP, M_WAITOK); 5038 sp->len = len; 5039 NFSBCOPY(client, sp->client, len); 5040 sp->flag = flag; 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); 5044 free(sp, M_TEMP); 5045 if (error) { 5046 sf->nsf_flags &= ~NFSNSF_OK; 5047 printf("EEK! Can't write NfsV4 stable storage file\n"); 5048 } 5049 } 5050 5051 /* 5052 * This function is called during the grace period to mark a client 5053 * that successfully reclaimed state. 5054 */ 5055 static void 5056 nfsrv_markstable(struct nfsclient *clp) 5057 { 5058 struct nfsrv_stable *sp; 5059 5060 /* 5061 * First find the client structure. 5062 */ 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)) 5066 break; 5067 } 5068 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head)) 5069 return; 5070 5071 /* 5072 * Now, just mark it and set the nfsclient back pointer. 5073 */ 5074 sp->nst_flag |= NFSNST_GOTSTATE; 5075 sp->nst_clp = clp; 5076 } 5077 5078 /* 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. 5081 */ 5082 static void 5083 nfsrv_markreclaim(struct nfsclient *clp) 5084 { 5085 struct nfsrv_stable *sp; 5086 5087 /* 5088 * First find the client structure. 5089 */ 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)) 5093 break; 5094 } 5095 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head)) 5096 return; 5097 5098 /* 5099 * Now, just set the flag. 5100 */ 5101 sp->nst_flag |= NFSNST_RECLAIMED; 5102 } 5103 5104 /* 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. 5107 */ 5108 static int 5109 nfsrv_checkstable(struct nfsclient *clp) 5110 { 5111 struct nfsrv_stable *sp; 5112 5113 /* 5114 * First, find the entry for the client. 5115 */ 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)) 5119 break; 5120 } 5121 5122 /* 5123 * If not in the list, state was revoked or no state was issued 5124 * since the previous reboot, a reclaim is denied. 5125 */ 5126 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) || 5127 (sp->nst_flag & NFSNST_REVOKE) || 5128 !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK)) 5129 return (1); 5130 return (0); 5131 } 5132 5133 /* 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 5136 * a found. 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. 5145 */ 5146 static int 5147 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp, 5148 NFSPROC_T *p) 5149 { 5150 int gotlock, lktype = 0; 5151 5152 /* 5153 * If lease hasn't expired, we can't fix it. 5154 */ 5155 if (clp->lc_expiry >= NFSD_MONOSEC || 5156 !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) 5157 return (0); 5158 if (*haslockp == 0) { 5159 NFSUNLOCKSTATE(); 5160 if (vp != NULL) { 5161 lktype = NFSVOPISLOCKED(vp); 5162 NFSVOPUNLOCK(vp); 5163 } 5164 NFSLOCKV4ROOTMUTEX(); 5165 nfsv4_relref(&nfsv4rootfs_lock); 5166 do { 5167 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 5168 NFSV4ROOTLOCKMUTEXPTR, NULL); 5169 } while (!gotlock); 5170 NFSUNLOCKV4ROOTMUTEX(); 5171 *haslockp = 1; 5172 if (vp != NULL) { 5173 NFSVOPLOCK(vp, lktype | LK_RETRY); 5174 if (VN_IS_DOOMED(vp)) 5175 return (2); 5176 } 5177 return (1); 5178 } 5179 NFSUNLOCKSTATE(); 5180 5181 /* 5182 * Ok, we can expire the conflicting client. 5183 */ 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); 5191 return (1); 5192 } 5193 5194 /* 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. 5199 * 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. 5212 */ 5213 static int 5214 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p, 5215 vnode_t vp) 5216 { 5217 struct nfsclient *clp = stp->ls_clp; 5218 int gotlock, error, lktype = 0, retrycnt, zapped_clp; 5219 nfsv4stateid_t tstateid; 5220 fhandle_t tfh; 5221 5222 /* 5223 * If the conflict is with an old delegation... 5224 */ 5225 if (stp->ls_flags & NFSLCK_OLDDELEG) { 5226 /* 5227 * You can delete it, if it has expired. 5228 */ 5229 if (clp->lc_delegtime < NFSD_MONOSEC) { 5230 nfsrv_freedeleg(stp); 5231 NFSUNLOCKSTATE(); 5232 error = -1; 5233 goto out; 5234 } 5235 NFSUNLOCKSTATE(); 5236 /* 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. 5241 */ 5242 if (*haslockp) { 5243 *haslockp = 0; 5244 NFSLOCKV4ROOTMUTEX(); 5245 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5246 NFSUNLOCKV4ROOTMUTEX(); 5247 } 5248 error = NFSERR_DELAY; 5249 goto out; 5250 } 5251 5252 /* 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 5256 */ 5257 if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) { 5258 /* 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.) 5271 */ 5272 stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) + 5273 NFSRV_LEASEDELTA; 5274 stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) + 5275 NFSRV_LEASEDELTA; 5276 stp->ls_flags |= NFSLCK_DELEGRECALL; 5277 5278 /* 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 5285 * the call. 5286 */ 5287 NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid, 5288 sizeof (tstateid)); 5289 NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh, 5290 sizeof (tfh)); 5291 NFSUNLOCKSTATE(); 5292 if (*haslockp) { 5293 *haslockp = 0; 5294 NFSLOCKV4ROOTMUTEX(); 5295 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5296 NFSUNLOCKV4ROOTMUTEX(); 5297 } 5298 retrycnt = 0; 5299 do { 5300 error = nfsrv_docallback(clp, NFSV4OP_CBRECALL, 5301 &tstateid, 0, &tfh, NULL, NULL, 0, p); 5302 retrycnt++; 5303 } while ((error == NFSERR_BADSTATEID || 5304 error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT); 5305 error = NFSERR_DELAY; 5306 goto out; 5307 } 5308 5309 if (clp->lc_expiry >= NFSD_MONOSEC && 5310 stp->ls_delegtime >= NFSD_MONOSEC) { 5311 NFSUNLOCKSTATE(); 5312 /* 5313 * A recall has been done, but it has not yet expired. 5314 * So, RETURN_DELAY. 5315 */ 5316 if (*haslockp) { 5317 *haslockp = 0; 5318 NFSLOCKV4ROOTMUTEX(); 5319 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5320 NFSUNLOCKV4ROOTMUTEX(); 5321 } 5322 error = NFSERR_DELAY; 5323 goto out; 5324 } 5325 5326 /* 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 5329 * this delegation. 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. 5332 */ 5333 if (*haslockp == 0) { 5334 NFSUNLOCKSTATE(); 5335 if (vp != NULL) { 5336 lktype = NFSVOPISLOCKED(vp); 5337 NFSVOPUNLOCK(vp); 5338 } 5339 NFSLOCKV4ROOTMUTEX(); 5340 nfsv4_relref(&nfsv4rootfs_lock); 5341 do { 5342 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 5343 NFSV4ROOTLOCKMUTEXPTR, NULL); 5344 } while (!gotlock); 5345 NFSUNLOCKV4ROOTMUTEX(); 5346 *haslockp = 1; 5347 if (vp != NULL) { 5348 NFSVOPLOCK(vp, lktype | LK_RETRY); 5349 if (VN_IS_DOOMED(vp)) { 5350 *haslockp = 0; 5351 NFSLOCKV4ROOTMUTEX(); 5352 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5353 NFSUNLOCKV4ROOTMUTEX(); 5354 error = NFSERR_PERM; 5355 goto out; 5356 } 5357 } 5358 error = -1; 5359 goto out; 5360 } 5361 5362 NFSUNLOCKSTATE(); 5363 /* 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. 5369 */ 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); 5377 zapped_clp = 1; 5378 } else { 5379 nfsrv_freedeleg(stp); 5380 zapped_clp = 0; 5381 } 5382 if (zapped_clp) 5383 nfsrv_zapclient(clp, p); 5384 error = -1; 5385 5386 out: 5387 NFSEXITCODE(error); 5388 return (error); 5389 } 5390 5391 /* 5392 * Check for a remove allowed, if remove is set to 1 and get rid of 5393 * delegations. 5394 */ 5395 int 5396 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd, 5397 nfsquad_t clientid, NFSPROC_T *p) 5398 { 5399 struct nfsclient *clp; 5400 struct nfsstate *stp; 5401 struct nfslockfile *lfp; 5402 int error, haslock = 0; 5403 fhandle_t nfh; 5404 5405 clp = NULL; 5406 /* 5407 * First, get the lock file structure. 5408 * (A return of -1 means no associated state, so remove ok.) 5409 */ 5410 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p); 5411 tryagain: 5412 NFSLOCKSTATE(); 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); 5416 if (!error) 5417 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0); 5418 if (error) { 5419 NFSUNLOCKSTATE(); 5420 if (haslock) { 5421 NFSLOCKV4ROOTMUTEX(); 5422 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5423 NFSUNLOCKV4ROOTMUTEX(); 5424 } 5425 if (error == -1) 5426 error = 0; 5427 goto out; 5428 } 5429 5430 /* 5431 * Now, we must Recall any delegations. 5432 */ 5433 error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p); 5434 if (error) { 5435 /* 5436 * nfsrv_cleandeleg() unlocks state for non-zero 5437 * return. 5438 */ 5439 if (error == -1) 5440 goto tryagain; 5441 if (haslock) { 5442 NFSLOCKV4ROOTMUTEX(); 5443 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5444 NFSUNLOCKV4ROOTMUTEX(); 5445 } 5446 goto out; 5447 } 5448 5449 /* 5450 * Now, look for a conflicting open share. 5451 */ 5452 if (remove) { 5453 /* 5454 * If the entry in the directory was the last reference to the 5455 * corresponding filesystem object, the object can be destroyed 5456 * */ 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; 5461 break; 5462 } 5463 } 5464 } 5465 5466 NFSUNLOCKSTATE(); 5467 if (haslock) { 5468 NFSLOCKV4ROOTMUTEX(); 5469 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5470 NFSUNLOCKV4ROOTMUTEX(); 5471 } 5472 5473 out: 5474 NFSEXITCODE(error); 5475 return (error); 5476 } 5477 5478 /* 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. 5486 */ 5487 static int 5488 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp, 5489 struct nfsclient *clp, int *haslockp, NFSPROC_T *p) 5490 { 5491 struct nfsstate *stp, *nstp; 5492 int ret = 0; 5493 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); 5499 if (ret) { 5500 /* 5501 * nfsrv_delegconflict() unlocks state 5502 * when it returns non-zero. 5503 */ 5504 goto out; 5505 } 5506 } 5507 stp = nstp; 5508 } 5509 out: 5510 NFSEXITCODE(ret); 5511 return (ret); 5512 } 5513 5514 /* 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. 5539 */ 5540 void 5541 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p) 5542 { 5543 time_t starttime; 5544 int error; 5545 5546 /* 5547 * First, check to see if the server is currently running and it has 5548 * been called for a regular file when issuing delegations. 5549 */ 5550 if (newnfs_numnfsd == 0 || vp->v_type != VREG || 5551 nfsrv_issuedelegs == 0) 5552 return; 5553 5554 KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp)); 5555 /* 5556 * First, get a reference on the nfsv4rootfs_lock so that an 5557 * exclusive lock cannot be acquired by another thread. 5558 */ 5559 NFSLOCKV4ROOTMUTEX(); 5560 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); 5561 NFSUNLOCKV4ROOTMUTEX(); 5562 5563 /* 5564 * Now, call nfsrv_checkremove() in a loop while it returns 5565 * NFSERR_DELAY. Return upon any other error or when timed out. 5566 */ 5567 starttime = NFSD_MONOSEC; 5568 do { 5569 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) { 5570 error = nfsrv_checkremove(vp, 0, NULL, 5571 (nfsquad_t)((u_quad_t)0), p); 5572 NFSVOPUNLOCK(vp); 5573 } else 5574 error = EPERM; 5575 if (error == NFSERR_DELAY) { 5576 if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO) 5577 break; 5578 /* Sleep for a short period of time */ 5579 (void) nfs_catnap(PZERO, 0, "nfsremove"); 5580 } 5581 } while (error == NFSERR_DELAY); 5582 NFSLOCKV4ROOTMUTEX(); 5583 nfsv4_relref(&nfsv4rootfs_lock); 5584 NFSUNLOCKV4ROOTMUTEX(); 5585 } 5586 5587 void 5588 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p) 5589 { 5590 5591 #ifdef VV_DISABLEDELEG 5592 /* 5593 * First, flag issuance of delegations disabled. 5594 */ 5595 atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG); 5596 #endif 5597 5598 /* 5599 * Then call nfsd_recalldelegation() to get rid of all extant 5600 * delegations. 5601 */ 5602 nfsd_recalldelegation(vp, p); 5603 } 5604 5605 /* 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 5615 * different?) 5616 */ 5617 int 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) 5621 { 5622 struct nfsstate st, *stp = &st; 5623 struct nfslock lo, *lop = &lo; 5624 int error = 0; 5625 nfsquad_t clientid; 5626 5627 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) { 5628 stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS); 5629 lop->lo_first = nvap->na_size; 5630 } else { 5631 stp->ls_flags = 0; 5632 lop->lo_first = 0; 5633 } 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) 5640 goto out; 5641 lop->lo_end = NFS64BITSSET; 5642 lop->lo_flags = NFSLCK_WRITE; 5643 stp->ls_ownerlen = 0; 5644 stp->ls_op = NULL; 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); 5652 5653 out: 5654 NFSEXITCODE2(error, nd); 5655 return (error); 5656 } 5657 5658 /* 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 5662 * just return ok. 5663 */ 5664 int 5665 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp, 5666 struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p) 5667 { 5668 struct nfsstate *stp; 5669 struct nfslockfile *lfp; 5670 struct nfsclient *clp; 5671 struct nfsvattr nva; 5672 fhandle_t nfh; 5673 int error = 0; 5674 nfsattrbit_t cbbits; 5675 u_quad_t delegfilerev; 5676 5677 NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits); 5678 if (!NFSNONZERO_ATTRBIT(&cbbits)) 5679 goto out; 5680 if (nfsrv_writedelegcnt == 0) 5681 goto out; 5682 5683 /* 5684 * Get the lock file structure. 5685 * (A return of -1 means no associated state, so return ok.) 5686 */ 5687 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p); 5688 NFSLOCKSTATE(); 5689 if (!error) 5690 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0); 5691 if (error) { 5692 NFSUNLOCKSTATE(); 5693 if (error == -1) 5694 error = 0; 5695 goto out; 5696 } 5697 5698 /* 5699 * Now, look for a write delegation. 5700 */ 5701 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { 5702 if (stp->ls_flags & NFSLCK_DELEGWRITE) 5703 break; 5704 } 5705 if (stp == LIST_END(&lfp->lf_deleg)) { 5706 NFSUNLOCKSTATE(); 5707 goto out; 5708 } 5709 clp = stp->ls_clp; 5710 delegfilerev = stp->ls_filerev; 5711 5712 /* 5713 * If the Write delegation was issued as a part of this Compound RPC 5714 * or if we have an Implied Clientid (used in a previous Op in this 5715 * compound) and it is the client the delegation was issued to, 5716 * just return ok. 5717 * I also assume that it is from the same client iff the network 5718 * host IP address is the same as the callback address. (Not 5719 * exactly correct by the RFC, but avoids a lot of Getattr 5720 * callbacks.) 5721 */ 5722 if (nd->nd_compref == stp->ls_compref || 5723 ((nd->nd_flag & ND_IMPLIEDCLID) && 5724 clp->lc_clientid.qval == nd->nd_clientid.qval) || 5725 nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) { 5726 NFSUNLOCKSTATE(); 5727 goto out; 5728 } 5729 5730 /* 5731 * We are now done with the delegation state structure, 5732 * so the statelock can be released and we can now tsleep(). 5733 */ 5734 5735 /* 5736 * Now, we must do the CB Getattr callback, to see if Change or Size 5737 * has changed. 5738 */ 5739 if (clp->lc_expiry >= NFSD_MONOSEC) { 5740 NFSUNLOCKSTATE(); 5741 NFSVNO_ATTRINIT(&nva); 5742 nva.na_filerev = NFS64BITSSET; 5743 error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL, 5744 0, &nfh, &nva, &cbbits, 0, p); 5745 if (!error) { 5746 if ((nva.na_filerev != NFS64BITSSET && 5747 nva.na_filerev > delegfilerev) || 5748 (NFSVNO_ISSETSIZE(&nva) && 5749 nva.na_size != nvap->na_size)) { 5750 error = nfsvno_updfilerev(vp, nvap, nd, p); 5751 if (NFSVNO_ISSETSIZE(&nva)) 5752 nvap->na_size = nva.na_size; 5753 } 5754 } else 5755 error = 0; /* Ignore callback errors for now. */ 5756 } else { 5757 NFSUNLOCKSTATE(); 5758 } 5759 5760 out: 5761 NFSEXITCODE2(error, nd); 5762 return (error); 5763 } 5764 5765 /* 5766 * This function looks for openowners that haven't had any opens for 5767 * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS 5768 * is set. 5769 */ 5770 void 5771 nfsrv_throwawayopens(NFSPROC_T *p) 5772 { 5773 struct nfsclient *clp, *nclp; 5774 struct nfsstate *stp, *nstp; 5775 int i; 5776 5777 NFSLOCKSTATE(); 5778 nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS; 5779 /* 5780 * For each client... 5781 */ 5782 for (i = 0; i < nfsrv_clienthashsize; i++) { 5783 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) { 5784 LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) { 5785 if (LIST_EMPTY(&stp->ls_open) && 5786 (stp->ls_noopens > NFSNOOPEN || 5787 (nfsrv_openpluslock * 2) > 5788 nfsrv_v4statelimit)) 5789 nfsrv_freeopenowner(stp, 0, p); 5790 } 5791 } 5792 } 5793 NFSUNLOCKSTATE(); 5794 } 5795 5796 /* 5797 * This function checks to see if the credentials are the same. 5798 * Returns 1 for not same, 0 otherwise. 5799 */ 5800 static int 5801 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp) 5802 { 5803 5804 if (nd->nd_flag & ND_GSS) { 5805 if (!(clp->lc_flags & LCL_GSS)) 5806 return (1); 5807 if (clp->lc_flags & LCL_NAME) { 5808 if (nd->nd_princlen != clp->lc_namelen || 5809 NFSBCMP(nd->nd_principal, clp->lc_name, 5810 clp->lc_namelen)) 5811 return (1); 5812 else 5813 return (0); 5814 } 5815 if (nd->nd_cred->cr_uid == clp->lc_uid) 5816 return (0); 5817 else 5818 return (1); 5819 } else if (clp->lc_flags & LCL_GSS) 5820 return (1); 5821 /* 5822 * For AUTH_SYS, allow the same uid or root. (This is underspecified 5823 * in RFC3530, which talks about principals, but doesn't say anything 5824 * about uids for AUTH_SYS.) 5825 */ 5826 if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0) 5827 return (0); 5828 else 5829 return (1); 5830 } 5831 5832 /* 5833 * Calculate the lease expiry time. 5834 */ 5835 static time_t 5836 nfsrv_leaseexpiry(void) 5837 { 5838 5839 if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC) 5840 return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA)); 5841 return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA); 5842 } 5843 5844 /* 5845 * Delay the delegation timeout as far as ls_delegtimelimit, as required. 5846 */ 5847 static void 5848 nfsrv_delaydelegtimeout(struct nfsstate *stp) 5849 { 5850 5851 if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) 5852 return; 5853 5854 if ((stp->ls_delegtime + 15) > NFSD_MONOSEC && 5855 stp->ls_delegtime < stp->ls_delegtimelimit) { 5856 stp->ls_delegtime += nfsrv_lease; 5857 if (stp->ls_delegtime > stp->ls_delegtimelimit) 5858 stp->ls_delegtime = stp->ls_delegtimelimit; 5859 } 5860 } 5861 5862 /* 5863 * This function checks to see if there is any other state associated 5864 * with the openowner for this Open. 5865 * It returns 1 if there is no other state, 0 otherwise. 5866 */ 5867 static int 5868 nfsrv_nootherstate(struct nfsstate *stp) 5869 { 5870 struct nfsstate *tstp; 5871 5872 LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) { 5873 if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock)) 5874 return (0); 5875 } 5876 return (1); 5877 } 5878 5879 /* 5880 * Create a list of lock deltas (changes to local byte range locking 5881 * that can be rolled back using the list) and apply the changes via 5882 * nfsvno_advlock(). Optionally, lock the list. It is expected that either 5883 * the rollback or update function will be called after this. 5884 * It returns an error (and rolls back, as required), if any nfsvno_advlock() 5885 * call fails. If it returns an error, it will unlock the list. 5886 */ 5887 static int 5888 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags, 5889 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p) 5890 { 5891 struct nfslock *lop, *nlop; 5892 int error = 0; 5893 5894 /* Loop through the list of locks. */ 5895 lop = LIST_FIRST(&lfp->lf_locallock); 5896 while (first < end && lop != NULL) { 5897 nlop = LIST_NEXT(lop, lo_lckowner); 5898 if (first >= lop->lo_end) { 5899 /* not there yet */ 5900 lop = nlop; 5901 } else if (first < lop->lo_first) { 5902 /* new one starts before entry in list */ 5903 if (end <= lop->lo_first) { 5904 /* no overlap between old and new */ 5905 error = nfsrv_dolocal(vp, lfp, flags, 5906 NFSLCK_UNLOCK, first, end, cfp, p); 5907 if (error != 0) 5908 break; 5909 first = end; 5910 } else { 5911 /* handle fragment overlapped with new one */ 5912 error = nfsrv_dolocal(vp, lfp, flags, 5913 NFSLCK_UNLOCK, first, lop->lo_first, cfp, 5914 p); 5915 if (error != 0) 5916 break; 5917 first = lop->lo_first; 5918 } 5919 } else { 5920 /* new one overlaps this entry in list */ 5921 if (end <= lop->lo_end) { 5922 /* overlaps all of new one */ 5923 error = nfsrv_dolocal(vp, lfp, flags, 5924 lop->lo_flags, first, end, cfp, p); 5925 if (error != 0) 5926 break; 5927 first = end; 5928 } else { 5929 /* handle fragment overlapped with new one */ 5930 error = nfsrv_dolocal(vp, lfp, flags, 5931 lop->lo_flags, first, lop->lo_end, cfp, p); 5932 if (error != 0) 5933 break; 5934 first = lop->lo_end; 5935 lop = nlop; 5936 } 5937 } 5938 } 5939 if (first < end && error == 0) 5940 /* handle fragment past end of list */ 5941 error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first, 5942 end, cfp, p); 5943 5944 NFSEXITCODE(error); 5945 return (error); 5946 } 5947 5948 /* 5949 * Local lock unlock. Unlock all byte ranges that are no longer locked 5950 * by NFSv4. To do this, unlock any subranges of first-->end that 5951 * do not overlap with the byte ranges of any lock in the lfp->lf_lock 5952 * list. This list has all locks for the file held by other 5953 * <clientid, lockowner> tuples. The list is ordered by increasing 5954 * lo_first value, but may have entries that overlap each other, for 5955 * the case of read locks. 5956 */ 5957 static void 5958 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first, 5959 uint64_t init_end, NFSPROC_T *p) 5960 { 5961 struct nfslock *lop; 5962 uint64_t first, end, prevfirst __unused; 5963 5964 first = init_first; 5965 end = init_end; 5966 while (first < init_end) { 5967 /* Loop through all nfs locks, adjusting first and end */ 5968 prevfirst = 0; 5969 LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) { 5970 KASSERT(prevfirst <= lop->lo_first, 5971 ("nfsv4 locks out of order")); 5972 KASSERT(lop->lo_first < lop->lo_end, 5973 ("nfsv4 bogus lock")); 5974 prevfirst = lop->lo_first; 5975 if (first >= lop->lo_first && 5976 first < lop->lo_end) 5977 /* 5978 * Overlaps with initial part, so trim 5979 * off that initial part by moving first past 5980 * it. 5981 */ 5982 first = lop->lo_end; 5983 else if (end > lop->lo_first && 5984 lop->lo_first > first) { 5985 /* 5986 * This lock defines the end of the 5987 * segment to unlock, so set end to the 5988 * start of it and break out of the loop. 5989 */ 5990 end = lop->lo_first; 5991 break; 5992 } 5993 if (first >= end) 5994 /* 5995 * There is no segment left to do, so 5996 * break out of this loop and then exit 5997 * the outer while() since first will be set 5998 * to end, which must equal init_end here. 5999 */ 6000 break; 6001 } 6002 if (first < end) { 6003 /* Unlock this segment */ 6004 (void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK, 6005 NFSLCK_READ, first, end, NULL, p); 6006 nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK, 6007 first, end); 6008 } 6009 /* 6010 * Now move past this segment and look for any further 6011 * segment in the range, if there is one. 6012 */ 6013 first = end; 6014 end = init_end; 6015 } 6016 } 6017 6018 /* 6019 * Do the local lock operation and update the rollback list, as required. 6020 * Perform the rollback and return the error if nfsvno_advlock() fails. 6021 */ 6022 static int 6023 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags, 6024 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p) 6025 { 6026 struct nfsrollback *rlp; 6027 int error = 0, ltype, oldltype; 6028 6029 if (flags & NFSLCK_WRITE) 6030 ltype = F_WRLCK; 6031 else if (flags & NFSLCK_READ) 6032 ltype = F_RDLCK; 6033 else 6034 ltype = F_UNLCK; 6035 if (oldflags & NFSLCK_WRITE) 6036 oldltype = F_WRLCK; 6037 else if (oldflags & NFSLCK_READ) 6038 oldltype = F_RDLCK; 6039 else 6040 oldltype = F_UNLCK; 6041 if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK)) 6042 /* nothing to do */ 6043 goto out; 6044 error = nfsvno_advlock(vp, ltype, first, end, p); 6045 if (error != 0) { 6046 if (cfp != NULL) { 6047 cfp->cl_clientid.lval[0] = 0; 6048 cfp->cl_clientid.lval[1] = 0; 6049 cfp->cl_first = 0; 6050 cfp->cl_end = NFS64BITSSET; 6051 cfp->cl_flags = NFSLCK_WRITE; 6052 cfp->cl_ownerlen = 5; 6053 NFSBCOPY("LOCAL", cfp->cl_owner, 5); 6054 } 6055 nfsrv_locallock_rollback(vp, lfp, p); 6056 } else if (ltype != F_UNLCK) { 6057 rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK, 6058 M_WAITOK); 6059 rlp->rlck_first = first; 6060 rlp->rlck_end = end; 6061 rlp->rlck_type = oldltype; 6062 LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list); 6063 } 6064 6065 out: 6066 NFSEXITCODE(error); 6067 return (error); 6068 } 6069 6070 /* 6071 * Roll back local lock changes and free up the rollback list. 6072 */ 6073 static void 6074 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p) 6075 { 6076 struct nfsrollback *rlp, *nrlp; 6077 6078 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) { 6079 (void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first, 6080 rlp->rlck_end, p); 6081 free(rlp, M_NFSDROLLBACK); 6082 } 6083 LIST_INIT(&lfp->lf_rollback); 6084 } 6085 6086 /* 6087 * Update local lock list and delete rollback list (ie now committed to the 6088 * local locks). Most of the work is done by the internal function. 6089 */ 6090 static void 6091 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first, 6092 uint64_t end) 6093 { 6094 struct nfsrollback *rlp, *nrlp; 6095 struct nfslock *new_lop, *other_lop; 6096 6097 new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK); 6098 if (flags & (NFSLCK_READ | NFSLCK_WRITE)) 6099 other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, 6100 M_WAITOK); 6101 else 6102 other_lop = NULL; 6103 new_lop->lo_flags = flags; 6104 new_lop->lo_first = first; 6105 new_lop->lo_end = end; 6106 nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp); 6107 if (new_lop != NULL) 6108 free(new_lop, M_NFSDLOCK); 6109 if (other_lop != NULL) 6110 free(other_lop, M_NFSDLOCK); 6111 6112 /* and get rid of the rollback list */ 6113 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) 6114 free(rlp, M_NFSDROLLBACK); 6115 LIST_INIT(&lfp->lf_rollback); 6116 } 6117 6118 /* 6119 * Lock the struct nfslockfile for local lock updating. 6120 */ 6121 static void 6122 nfsrv_locklf(struct nfslockfile *lfp) 6123 { 6124 int gotlock; 6125 6126 /* lf_usecount ensures *lfp won't be free'd */ 6127 lfp->lf_usecount++; 6128 do { 6129 gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL, 6130 NFSSTATEMUTEXPTR, NULL); 6131 } while (gotlock == 0); 6132 lfp->lf_usecount--; 6133 } 6134 6135 /* 6136 * Unlock the struct nfslockfile after local lock updating. 6137 */ 6138 static void 6139 nfsrv_unlocklf(struct nfslockfile *lfp) 6140 { 6141 6142 nfsv4_unlock(&lfp->lf_locallock_lck, 0); 6143 } 6144 6145 /* 6146 * Clear out all state for the NFSv4 server. 6147 * Must be called by a thread that can sleep when no nfsds are running. 6148 */ 6149 void 6150 nfsrv_throwawayallstate(NFSPROC_T *p) 6151 { 6152 struct nfsclient *clp, *nclp; 6153 struct nfslockfile *lfp, *nlfp; 6154 int i; 6155 6156 /* 6157 * For each client, clean out the state and then free the structure. 6158 */ 6159 for (i = 0; i < nfsrv_clienthashsize; i++) { 6160 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) { 6161 nfsrv_cleanclient(clp, p); 6162 nfsrv_freedeleglist(&clp->lc_deleg); 6163 nfsrv_freedeleglist(&clp->lc_olddeleg); 6164 free(clp->lc_stateid, M_NFSDCLIENT); 6165 free(clp, M_NFSDCLIENT); 6166 } 6167 } 6168 6169 /* 6170 * Also, free up any remaining lock file structures. 6171 */ 6172 for (i = 0; i < nfsrv_lockhashsize; i++) { 6173 LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) { 6174 printf("nfsd unload: fnd a lock file struct\n"); 6175 nfsrv_freenfslockfile(lfp); 6176 } 6177 } 6178 6179 /* And get rid of the deviceid structures and layouts. */ 6180 nfsrv_freealllayoutsanddevids(); 6181 } 6182 6183 /* 6184 * Check the sequence# for the session and slot provided as an argument. 6185 * Also, renew the lease if the session will return NFS_OK. 6186 */ 6187 int 6188 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid, 6189 uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this, 6190 uint32_t *sflagsp, NFSPROC_T *p) 6191 { 6192 struct nfsdsession *sep; 6193 struct nfssessionhash *shp; 6194 int error; 6195 SVCXPRT *savxprt; 6196 6197 shp = NFSSESSIONHASH(nd->nd_sessionid); 6198 NFSLOCKSESSION(shp); 6199 sep = nfsrv_findsession(nd->nd_sessionid); 6200 if (sep == NULL) { 6201 NFSUNLOCKSESSION(shp); 6202 return (NFSERR_BADSESSION); 6203 } 6204 error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp, 6205 sep->sess_slots, NULL, NFSV4_SLOTS - 1); 6206 if (error != 0) { 6207 NFSUNLOCKSESSION(shp); 6208 return (error); 6209 } 6210 if (cache_this != 0) 6211 nd->nd_flag |= ND_SAVEREPLY; 6212 /* Renew the lease. */ 6213 sep->sess_clp->lc_expiry = nfsrv_leaseexpiry(); 6214 nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval; 6215 nd->nd_flag |= ND_IMPLIEDCLID; 6216 6217 /* Save maximum request and reply sizes. */ 6218 nd->nd_maxreq = sep->sess_maxreq; 6219 nd->nd_maxresp = sep->sess_maxresp; 6220 6221 /* 6222 * If this session handles the backchannel, save the nd_xprt for this 6223 * RPC, since this is the one being used. 6224 * RFC-5661 specifies that the fore channel will be implicitly 6225 * bound by a Sequence operation. However, since some NFSv4.1 clients 6226 * erroneously assumed that the back channel would be implicitly 6227 * bound as well, do the implicit binding unless a 6228 * BindConnectiontoSession has already been done on the session. 6229 */ 6230 if (sep->sess_clp->lc_req.nr_client != NULL && 6231 sep->sess_cbsess.nfsess_xprt != nd->nd_xprt && 6232 (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0 && 6233 (sep->sess_clp->lc_flags & LCL_DONEBINDCONN) == 0) { 6234 NFSD_DEBUG(2, 6235 "nfsrv_checksequence: implicit back channel bind\n"); 6236 savxprt = sep->sess_cbsess.nfsess_xprt; 6237 SVC_ACQUIRE(nd->nd_xprt); 6238 nd->nd_xprt->xp_p2 = 6239 sep->sess_clp->lc_req.nr_client->cl_private; 6240 nd->nd_xprt->xp_idletimeout = 0; /* Disable timeout. */ 6241 sep->sess_cbsess.nfsess_xprt = nd->nd_xprt; 6242 if (savxprt != NULL) 6243 SVC_RELEASE(savxprt); 6244 } 6245 6246 *sflagsp = 0; 6247 if (sep->sess_clp->lc_req.nr_client == NULL) 6248 *sflagsp |= NFSV4SEQ_CBPATHDOWN; 6249 NFSUNLOCKSESSION(shp); 6250 if (error == NFSERR_EXPIRED) { 6251 *sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED; 6252 error = 0; 6253 } else if (error == NFSERR_ADMINREVOKED) { 6254 *sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED; 6255 error = 0; 6256 } 6257 *highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1; 6258 return (0); 6259 } 6260 6261 /* 6262 * Check/set reclaim complete for this session/clientid. 6263 */ 6264 int 6265 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs) 6266 { 6267 struct nfsdsession *sep; 6268 struct nfssessionhash *shp; 6269 int error = 0; 6270 6271 shp = NFSSESSIONHASH(nd->nd_sessionid); 6272 NFSLOCKSTATE(); 6273 NFSLOCKSESSION(shp); 6274 sep = nfsrv_findsession(nd->nd_sessionid); 6275 if (sep == NULL) { 6276 NFSUNLOCKSESSION(shp); 6277 NFSUNLOCKSTATE(); 6278 return (NFSERR_BADSESSION); 6279 } 6280 6281 if (onefs != 0) 6282 sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS; 6283 /* Check to see if reclaim complete has already happened. */ 6284 else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) 6285 error = NFSERR_COMPLETEALREADY; 6286 else { 6287 sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE; 6288 nfsrv_markreclaim(sep->sess_clp); 6289 } 6290 NFSUNLOCKSESSION(shp); 6291 NFSUNLOCKSTATE(); 6292 return (error); 6293 } 6294 6295 /* 6296 * Cache the reply in a session slot. 6297 */ 6298 void 6299 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m) 6300 { 6301 struct nfsdsession *sep; 6302 struct nfssessionhash *shp; 6303 char *buf, *cp; 6304 #ifdef INET 6305 struct sockaddr_in *sin; 6306 #endif 6307 #ifdef INET6 6308 struct sockaddr_in6 *sin6; 6309 #endif 6310 6311 shp = NFSSESSIONHASH(nd->nd_sessionid); 6312 NFSLOCKSESSION(shp); 6313 sep = nfsrv_findsession(nd->nd_sessionid); 6314 if (sep == NULL) { 6315 NFSUNLOCKSESSION(shp); 6316 if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) { 6317 buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK); 6318 switch (nd->nd_nam->sa_family) { 6319 #ifdef INET 6320 case AF_INET: 6321 sin = (struct sockaddr_in *)nd->nd_nam; 6322 cp = inet_ntop(sin->sin_family, 6323 &sin->sin_addr.s_addr, buf, 6324 INET6_ADDRSTRLEN); 6325 break; 6326 #endif 6327 #ifdef INET6 6328 case AF_INET6: 6329 sin6 = (struct sockaddr_in6 *)nd->nd_nam; 6330 cp = inet_ntop(sin6->sin6_family, 6331 &sin6->sin6_addr, buf, INET6_ADDRSTRLEN); 6332 break; 6333 #endif 6334 default: 6335 cp = NULL; 6336 } 6337 if (cp != NULL) 6338 printf("nfsrv_cache_session: no session " 6339 "IPaddr=%s\n", cp); 6340 else 6341 printf("nfsrv_cache_session: no session\n"); 6342 free(buf, M_TEMP); 6343 } 6344 m_freem(*m); 6345 return; 6346 } 6347 nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat, 6348 m); 6349 NFSUNLOCKSESSION(shp); 6350 } 6351 6352 /* 6353 * Search for a session that matches the sessionid. 6354 */ 6355 static struct nfsdsession * 6356 nfsrv_findsession(uint8_t *sessionid) 6357 { 6358 struct nfsdsession *sep; 6359 struct nfssessionhash *shp; 6360 6361 shp = NFSSESSIONHASH(sessionid); 6362 LIST_FOREACH(sep, &shp->list, sess_hash) { 6363 if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID)) 6364 break; 6365 } 6366 return (sep); 6367 } 6368 6369 /* 6370 * Destroy a session. 6371 */ 6372 int 6373 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid) 6374 { 6375 int error, igotlock, samesess; 6376 6377 samesess = 0; 6378 if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) && 6379 (nd->nd_flag & ND_HASSEQUENCE) != 0) { 6380 samesess = 1; 6381 if ((nd->nd_flag & ND_LASTOP) == 0) 6382 return (NFSERR_BADSESSION); 6383 } 6384 6385 /* Lock out other nfsd threads */ 6386 NFSLOCKV4ROOTMUTEX(); 6387 nfsv4_relref(&nfsv4rootfs_lock); 6388 do { 6389 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 6390 NFSV4ROOTLOCKMUTEXPTR, NULL); 6391 } while (igotlock == 0); 6392 NFSUNLOCKV4ROOTMUTEX(); 6393 6394 error = nfsrv_freesession(NULL, sessionid); 6395 if (error == 0 && samesess != 0) 6396 nd->nd_flag &= ~ND_HASSEQUENCE; 6397 6398 NFSLOCKV4ROOTMUTEX(); 6399 nfsv4_unlock(&nfsv4rootfs_lock, 1); 6400 NFSUNLOCKV4ROOTMUTEX(); 6401 return (error); 6402 } 6403 6404 /* 6405 * Bind a connection to a session. 6406 * For now, only certain variants are supported, since the current session 6407 * structure can only handle a single backchannel entry, which will be 6408 * applied to all connections if it is set. 6409 */ 6410 int 6411 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp) 6412 { 6413 struct nfssessionhash *shp; 6414 struct nfsdsession *sep; 6415 struct nfsclient *clp; 6416 SVCXPRT *savxprt; 6417 int error; 6418 6419 error = 0; 6420 shp = NFSSESSIONHASH(sessionid); 6421 NFSLOCKSTATE(); 6422 NFSLOCKSESSION(shp); 6423 sep = nfsrv_findsession(sessionid); 6424 if (sep != NULL) { 6425 clp = sep->sess_clp; 6426 if (*foreaftp == NFSCDFC4_BACK || 6427 *foreaftp == NFSCDFC4_BACK_OR_BOTH || 6428 *foreaftp == NFSCDFC4_FORE_OR_BOTH) { 6429 /* Try to set up a backchannel. */ 6430 if (clp->lc_req.nr_client == NULL) { 6431 NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire " 6432 "backchannel\n"); 6433 clp->lc_req.nr_client = (struct __rpc_client *) 6434 clnt_bck_create(nd->nd_xprt->xp_socket, 6435 sep->sess_cbprogram, NFSV4_CBVERS); 6436 } 6437 if (clp->lc_req.nr_client != NULL) { 6438 NFSD_DEBUG(2, "nfsrv_bindconnsess: set up " 6439 "backchannel\n"); 6440 savxprt = sep->sess_cbsess.nfsess_xprt; 6441 SVC_ACQUIRE(nd->nd_xprt); 6442 nd->nd_xprt->xp_p2 = 6443 clp->lc_req.nr_client->cl_private; 6444 /* Disable idle timeout. */ 6445 nd->nd_xprt->xp_idletimeout = 0; 6446 sep->sess_cbsess.nfsess_xprt = nd->nd_xprt; 6447 if (savxprt != NULL) 6448 SVC_RELEASE(savxprt); 6449 sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN; 6450 clp->lc_flags |= LCL_DONEBINDCONN; 6451 if (*foreaftp == NFSCDFS4_BACK) 6452 *foreaftp = NFSCDFS4_BACK; 6453 else 6454 *foreaftp = NFSCDFS4_BOTH; 6455 } else if (*foreaftp != NFSCDFC4_BACK) { 6456 NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set " 6457 "up backchannel\n"); 6458 sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN; 6459 clp->lc_flags |= LCL_DONEBINDCONN; 6460 *foreaftp = NFSCDFS4_FORE; 6461 } else { 6462 error = NFSERR_NOTSUPP; 6463 printf("nfsrv_bindconnsess: Can't add " 6464 "backchannel\n"); 6465 } 6466 } else { 6467 NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n"); 6468 clp->lc_flags |= LCL_DONEBINDCONN; 6469 *foreaftp = NFSCDFS4_FORE; 6470 } 6471 } else 6472 error = NFSERR_BADSESSION; 6473 NFSUNLOCKSESSION(shp); 6474 NFSUNLOCKSTATE(); 6475 return (error); 6476 } 6477 6478 /* 6479 * Free up a session structure. 6480 */ 6481 static int 6482 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid) 6483 { 6484 struct nfssessionhash *shp; 6485 int i; 6486 6487 NFSLOCKSTATE(); 6488 if (sep == NULL) { 6489 shp = NFSSESSIONHASH(sessionid); 6490 NFSLOCKSESSION(shp); 6491 sep = nfsrv_findsession(sessionid); 6492 } else { 6493 shp = NFSSESSIONHASH(sep->sess_sessionid); 6494 NFSLOCKSESSION(shp); 6495 } 6496 if (sep != NULL) { 6497 sep->sess_refcnt--; 6498 if (sep->sess_refcnt > 0) { 6499 NFSUNLOCKSESSION(shp); 6500 NFSUNLOCKSTATE(); 6501 return (NFSERR_BACKCHANBUSY); 6502 } 6503 LIST_REMOVE(sep, sess_hash); 6504 LIST_REMOVE(sep, sess_list); 6505 } 6506 NFSUNLOCKSESSION(shp); 6507 NFSUNLOCKSTATE(); 6508 if (sep == NULL) 6509 return (NFSERR_BADSESSION); 6510 for (i = 0; i < NFSV4_SLOTS; i++) 6511 if (sep->sess_slots[i].nfssl_reply != NULL) 6512 m_freem(sep->sess_slots[i].nfssl_reply); 6513 if (sep->sess_cbsess.nfsess_xprt != NULL) 6514 SVC_RELEASE(sep->sess_cbsess.nfsess_xprt); 6515 free(sep, M_NFSDSESSION); 6516 return (0); 6517 } 6518 6519 /* 6520 * Free a stateid. 6521 * RFC5661 says that it should fail when there are associated opens, locks 6522 * or delegations. Since stateids represent opens, I don't see how you can 6523 * free an open stateid (it will be free'd when closed), so this function 6524 * only works for lock stateids (freeing the lock_owner) or delegations. 6525 */ 6526 int 6527 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, 6528 NFSPROC_T *p) 6529 { 6530 struct nfsclient *clp; 6531 struct nfsstate *stp; 6532 int error; 6533 6534 NFSLOCKSTATE(); 6535 /* 6536 * Look up the stateid 6537 */ 6538 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, 6539 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); 6540 if (error == 0) { 6541 /* First, check for a delegation. */ 6542 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) { 6543 if (!NFSBCMP(stp->ls_stateid.other, stateidp->other, 6544 NFSX_STATEIDOTHER)) 6545 break; 6546 } 6547 if (stp != NULL) { 6548 nfsrv_freedeleg(stp); 6549 NFSUNLOCKSTATE(); 6550 return (error); 6551 } 6552 } 6553 /* Not a delegation, try for a lock_owner. */ 6554 if (error == 0) 6555 error = nfsrv_getstate(clp, stateidp, 0, &stp); 6556 if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD | 6557 NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0)) 6558 /* Not a lock_owner stateid. */ 6559 error = NFSERR_LOCKSHELD; 6560 if (error == 0 && !LIST_EMPTY(&stp->ls_lock)) 6561 error = NFSERR_LOCKSHELD; 6562 if (error == 0) 6563 nfsrv_freelockowner(stp, NULL, 0, p); 6564 NFSUNLOCKSTATE(); 6565 return (error); 6566 } 6567 6568 /* 6569 * Test a stateid. 6570 */ 6571 int 6572 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, 6573 NFSPROC_T *p) 6574 { 6575 struct nfsclient *clp; 6576 struct nfsstate *stp; 6577 int error; 6578 6579 NFSLOCKSTATE(); 6580 /* 6581 * Look up the stateid 6582 */ 6583 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, 6584 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); 6585 if (error == 0) 6586 error = nfsrv_getstate(clp, stateidp, 0, &stp); 6587 if (error == 0 && stateidp->seqid != 0 && 6588 SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid)) 6589 error = NFSERR_OLDSTATEID; 6590 NFSUNLOCKSTATE(); 6591 return (error); 6592 } 6593 6594 /* 6595 * Generate the xdr for an NFSv4.1 CBSequence Operation. 6596 */ 6597 static int 6598 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp, 6599 int dont_replycache, struct nfsdsession **sepp) 6600 { 6601 struct nfsdsession *sep; 6602 uint32_t *tl, slotseq = 0; 6603 int maxslot, slotpos; 6604 uint8_t sessionid[NFSX_V4SESSIONID]; 6605 int error; 6606 6607 error = nfsv4_getcbsession(clp, sepp); 6608 if (error != 0) 6609 return (error); 6610 sep = *sepp; 6611 (void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot, 6612 &slotseq, sessionid); 6613 KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot")); 6614 6615 /* Build the Sequence arguments. */ 6616 NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED); 6617 bcopy(sessionid, tl, NFSX_V4SESSIONID); 6618 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; 6619 nd->nd_slotseq = tl; 6620 *tl++ = txdr_unsigned(slotseq); 6621 *tl++ = txdr_unsigned(slotpos); 6622 *tl++ = txdr_unsigned(maxslot); 6623 if (dont_replycache == 0) 6624 *tl++ = newnfs_true; 6625 else 6626 *tl++ = newnfs_false; 6627 *tl = 0; /* No referring call list, for now. */ 6628 nd->nd_flag |= ND_HASSEQUENCE; 6629 return (0); 6630 } 6631 6632 /* 6633 * Get a session for the callback. 6634 */ 6635 static int 6636 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp) 6637 { 6638 struct nfsdsession *sep; 6639 6640 NFSLOCKSTATE(); 6641 LIST_FOREACH(sep, &clp->lc_session, sess_list) { 6642 if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) 6643 break; 6644 } 6645 if (sep == NULL) { 6646 NFSUNLOCKSTATE(); 6647 return (NFSERR_BADSESSION); 6648 } 6649 sep->sess_refcnt++; 6650 *sepp = sep; 6651 NFSUNLOCKSTATE(); 6652 return (0); 6653 } 6654 6655 /* 6656 * Free up all backchannel xprts. This needs to be done when the nfsd threads 6657 * exit, since those transports will all be going away. 6658 * This is only called after all the nfsd threads are done performing RPCs, 6659 * so locking shouldn't be an issue. 6660 */ 6661 void 6662 nfsrv_freeallbackchannel_xprts(void) 6663 { 6664 struct nfsdsession *sep; 6665 struct nfsclient *clp; 6666 SVCXPRT *xprt; 6667 int i; 6668 6669 for (i = 0; i < nfsrv_clienthashsize; i++) { 6670 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) { 6671 LIST_FOREACH(sep, &clp->lc_session, sess_list) { 6672 xprt = sep->sess_cbsess.nfsess_xprt; 6673 sep->sess_cbsess.nfsess_xprt = NULL; 6674 if (xprt != NULL) 6675 SVC_RELEASE(xprt); 6676 } 6677 } 6678 } 6679 } 6680 6681 /* 6682 * Do a layout commit. Actually just call nfsrv_updatemdsattr(). 6683 * I have no idea if the rest of these arguments will ever be useful? 6684 */ 6685 int 6686 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype, 6687 int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len, 6688 int hasnewmtime, struct timespec *newmtimep, int reclaim, 6689 nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep, 6690 uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p) 6691 { 6692 struct nfsvattr na; 6693 int error; 6694 6695 error = nfsrv_updatemdsattr(vp, &na, p); 6696 if (error == 0) { 6697 *hasnewsizep = 1; 6698 *newsizep = na.na_size; 6699 } 6700 return (error); 6701 } 6702 6703 /* 6704 * Try and get a layout. 6705 */ 6706 int 6707 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp, 6708 int layouttype, int *iomode, uint64_t *offset, uint64_t *len, 6709 uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose, 6710 int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p) 6711 { 6712 struct nfslayouthash *lhyp; 6713 struct nfslayout *lyp; 6714 char *devid; 6715 fhandle_t fh, *dsfhp; 6716 int error, mirrorcnt; 6717 6718 if (nfsrv_devidcnt == 0) 6719 return (NFSERR_UNKNLAYOUTTYPE); 6720 6721 if (*offset != 0) 6722 printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset, 6723 (uintmax_t)*len); 6724 error = nfsvno_getfh(vp, &fh, p); 6725 NFSD_DEBUG(4, "layoutget getfh=%d\n", error); 6726 if (error != 0) 6727 return (error); 6728 6729 /* 6730 * For now, all layouts are for entire files. 6731 * Only issue Read/Write layouts if requested for a non-readonly fs. 6732 */ 6733 if (NFSVNO_EXRDONLY(exp)) { 6734 if (*iomode == NFSLAYOUTIOMODE_RW) 6735 return (NFSERR_LAYOUTTRYLATER); 6736 *iomode = NFSLAYOUTIOMODE_READ; 6737 } 6738 if (*iomode != NFSLAYOUTIOMODE_RW) 6739 *iomode = NFSLAYOUTIOMODE_READ; 6740 6741 /* 6742 * Check to see if a write layout can be issued for this file. 6743 * This is used during mirror recovery to avoid RW layouts being 6744 * issued for a file while it is being copied to the recovered 6745 * mirror. 6746 */ 6747 if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0) 6748 return (NFSERR_LAYOUTTRYLATER); 6749 6750 *retonclose = 0; 6751 *offset = 0; 6752 *len = UINT64_MAX; 6753 6754 /* First, see if a layout already exists and return if found. */ 6755 lhyp = NFSLAYOUTHASH(&fh); 6756 NFSLOCKLAYOUT(lhyp); 6757 error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp); 6758 NFSD_DEBUG(4, "layoutget findlay=%d\n", error); 6759 /* 6760 * Not sure if the seqid must be the same, so I won't check it. 6761 */ 6762 if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] || 6763 stateidp->other[1] != lyp->lay_stateid.other[1] || 6764 stateidp->other[2] != lyp->lay_stateid.other[2])) { 6765 if ((lyp->lay_flags & NFSLAY_CALLB) == 0) { 6766 NFSUNLOCKLAYOUT(lhyp); 6767 NFSD_DEBUG(1, "ret bad stateid\n"); 6768 return (NFSERR_BADSTATEID); 6769 } 6770 /* 6771 * I believe we get here because there is a race between 6772 * the client processing the CBLAYOUTRECALL and the layout 6773 * being deleted here on the server. 6774 * The client has now done a LayoutGet with a non-layout 6775 * stateid, as it would when there is no layout. 6776 * As such, free this layout and set error == NFSERR_BADSTATEID 6777 * so the code below will create a new layout structure as 6778 * would happen if no layout was found. 6779 * "lyp" will be set before being used below, but set it NULL 6780 * as a safety belt. 6781 */ 6782 nfsrv_freelayout(&lhyp->list, lyp); 6783 lyp = NULL; 6784 error = NFSERR_BADSTATEID; 6785 } 6786 if (error == 0) { 6787 if (lyp->lay_layoutlen > maxcnt) { 6788 NFSUNLOCKLAYOUT(lhyp); 6789 NFSD_DEBUG(1, "ret layout too small\n"); 6790 return (NFSERR_TOOSMALL); 6791 } 6792 if (*iomode == NFSLAYOUTIOMODE_RW) 6793 lyp->lay_flags |= NFSLAY_RW; 6794 else 6795 lyp->lay_flags |= NFSLAY_READ; 6796 NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen); 6797 *layoutlenp = lyp->lay_layoutlen; 6798 if (++lyp->lay_stateid.seqid == 0) 6799 lyp->lay_stateid.seqid = 1; 6800 stateidp->seqid = lyp->lay_stateid.seqid; 6801 NFSUNLOCKLAYOUT(lhyp); 6802 NFSD_DEBUG(4, "ret fnd layout\n"); 6803 return (0); 6804 } 6805 NFSUNLOCKLAYOUT(lhyp); 6806 6807 /* Find the device id and file handle. */ 6808 dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK); 6809 devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK); 6810 error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid); 6811 NFSD_DEBUG(4, "layoutget devandfh=%d\n", error); 6812 if (error == 0) { 6813 if (layouttype == NFSLAYOUT_NFSV4_1_FILES) { 6814 if (NFSX_V4FILELAYOUT > maxcnt) 6815 error = NFSERR_TOOSMALL; 6816 else 6817 lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp, 6818 devid, vp->v_mount->mnt_stat.f_fsid); 6819 } else { 6820 if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt) 6821 error = NFSERR_TOOSMALL; 6822 else 6823 lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt, 6824 &fh, dsfhp, devid, 6825 vp->v_mount->mnt_stat.f_fsid); 6826 } 6827 } 6828 free(dsfhp, M_TEMP); 6829 free(devid, M_TEMP); 6830 if (error != 0) 6831 return (error); 6832 6833 /* 6834 * Now, add this layout to the list. 6835 */ 6836 error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p); 6837 NFSD_DEBUG(4, "layoutget addl=%d\n", error); 6838 /* 6839 * The lyp will be set to NULL by nfsrv_addlayout() if it 6840 * linked the new structure into the lists. 6841 */ 6842 free(lyp, M_NFSDSTATE); 6843 return (error); 6844 } 6845 6846 /* 6847 * Generate a File Layout. 6848 */ 6849 static struct nfslayout * 6850 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp, 6851 fhandle_t *dsfhp, char *devid, fsid_t fs) 6852 { 6853 uint32_t *tl; 6854 struct nfslayout *lyp; 6855 uint64_t pattern_offset; 6856 6857 lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE, 6858 M_WAITOK | M_ZERO); 6859 lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES; 6860 if (iomode == NFSLAYOUTIOMODE_RW) 6861 lyp->lay_flags = NFSLAY_RW; 6862 else 6863 lyp->lay_flags = NFSLAY_READ; 6864 NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp)); 6865 lyp->lay_clientid.qval = nd->nd_clientid.qval; 6866 lyp->lay_fsid = fs; 6867 6868 /* Fill in the xdr for the files layout. */ 6869 tl = (uint32_t *)lyp->lay_xdr; 6870 NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */ 6871 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); 6872 6873 /* 6874 * Make the stripe size as many 64K blocks as will fit in the stripe 6875 * mask. Since there is only one stripe, the stripe size doesn't really 6876 * matter, except that the Linux client will only handle an exact 6877 * multiple of their PAGE_SIZE (usually 4K). I chose 64K as a value 6878 * that should cover most/all arches w.r.t. PAGE_SIZE. 6879 */ 6880 *tl++ = txdr_unsigned(NFSFLAYUTIL_STRIPE_MASK & ~0xffff); 6881 *tl++ = 0; /* 1st stripe index. */ 6882 pattern_offset = 0; 6883 txdr_hyper(pattern_offset, tl); tl += 2; /* Pattern offset. */ 6884 *tl++ = txdr_unsigned(1); /* 1 file handle. */ 6885 *tl++ = txdr_unsigned(NFSX_V4PNFSFH); 6886 NFSBCOPY(dsfhp, tl, sizeof(*dsfhp)); 6887 lyp->lay_layoutlen = NFSX_V4FILELAYOUT; 6888 return (lyp); 6889 } 6890 6891 #define FLEX_OWNERID "999" 6892 #define FLEX_UID0 "0" 6893 /* 6894 * Generate a Flex File Layout. 6895 * The FLEX_OWNERID can be any string of 3 decimal digits. Although this 6896 * string goes on the wire, it isn't supposed to be used by the client, 6897 * since this server uses tight coupling. 6898 * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use 6899 * a string of "0". This works around the Linux Flex File Layout driver bug 6900 * which uses the synthetic uid/gid strings for the "tightly coupled" case. 6901 */ 6902 static struct nfslayout * 6903 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt, 6904 fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs) 6905 { 6906 uint32_t *tl; 6907 struct nfslayout *lyp; 6908 uint64_t lenval; 6909 int i; 6910 6911 lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt), 6912 M_NFSDSTATE, M_WAITOK | M_ZERO); 6913 lyp->lay_type = NFSLAYOUT_FLEXFILE; 6914 if (iomode == NFSLAYOUTIOMODE_RW) 6915 lyp->lay_flags = NFSLAY_RW; 6916 else 6917 lyp->lay_flags = NFSLAY_READ; 6918 NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp)); 6919 lyp->lay_clientid.qval = nd->nd_clientid.qval; 6920 lyp->lay_fsid = fs; 6921 lyp->lay_mirrorcnt = mirrorcnt; 6922 6923 /* Fill in the xdr for the files layout. */ 6924 tl = (uint32_t *)lyp->lay_xdr; 6925 lenval = 0; 6926 txdr_hyper(lenval, tl); tl += 2; /* Stripe unit. */ 6927 *tl++ = txdr_unsigned(mirrorcnt); /* # of mirrors. */ 6928 for (i = 0; i < mirrorcnt; i++) { 6929 *tl++ = txdr_unsigned(1); /* One stripe. */ 6930 NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */ 6931 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); 6932 devid += NFSX_V4DEVICEID; 6933 *tl++ = txdr_unsigned(1); /* Efficiency. */ 6934 *tl++ = 0; /* Proxy Stateid. */ 6935 *tl++ = 0x55555555; 6936 *tl++ = 0x55555555; 6937 *tl++ = 0x55555555; 6938 *tl++ = txdr_unsigned(1); /* 1 file handle. */ 6939 *tl++ = txdr_unsigned(NFSX_V4PNFSFH); 6940 NFSBCOPY(dsfhp, tl, sizeof(*dsfhp)); 6941 tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED); 6942 dsfhp++; 6943 if (nfsrv_flexlinuxhack != 0) { 6944 *tl++ = txdr_unsigned(strlen(FLEX_UID0)); 6945 *tl = 0; /* 0 pad string. */ 6946 NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0)); 6947 *tl++ = txdr_unsigned(strlen(FLEX_UID0)); 6948 *tl = 0; /* 0 pad string. */ 6949 NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0)); 6950 } else { 6951 *tl++ = txdr_unsigned(strlen(FLEX_OWNERID)); 6952 NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED); 6953 *tl++ = txdr_unsigned(strlen(FLEX_OWNERID)); 6954 NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED); 6955 } 6956 } 6957 *tl++ = txdr_unsigned(0); /* ff_flags. */ 6958 *tl = txdr_unsigned(60); /* Status interval hint. */ 6959 lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt); 6960 return (lyp); 6961 } 6962 6963 /* 6964 * Parse and process Flex File errors returned via LayoutReturn. 6965 */ 6966 static void 6967 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt, 6968 NFSPROC_T *p) 6969 { 6970 uint32_t *tl; 6971 int cnt, errcnt, i, j, opnum, stat; 6972 char devid[NFSX_V4DEVICEID]; 6973 6974 tl = layp; 6975 cnt = fxdr_unsigned(int, *tl++); 6976 NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt); 6977 for (i = 0; i < cnt; i++) { 6978 /* Skip offset, length and stateid for now. */ 6979 tl += (4 + NFSX_STATEID / NFSX_UNSIGNED); 6980 errcnt = fxdr_unsigned(int, *tl++); 6981 NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt); 6982 for (j = 0; j < errcnt; j++) { 6983 NFSBCOPY(tl, devid, NFSX_V4DEVICEID); 6984 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); 6985 stat = fxdr_unsigned(int, *tl++); 6986 opnum = fxdr_unsigned(int, *tl++); 6987 NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum, 6988 stat); 6989 /* 6990 * Except for NFSERR_ACCES and NFSERR_STALE errors, 6991 * disable the mirror. 6992 */ 6993 if (stat != NFSERR_ACCES && stat != NFSERR_STALE) 6994 nfsrv_delds(devid, p); 6995 } 6996 } 6997 } 6998 6999 /* 7000 * This function removes all flex file layouts which has a mirror with 7001 * a device id that matches the argument. 7002 * Called when the DS represented by the device id has failed. 7003 */ 7004 void 7005 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p) 7006 { 7007 uint32_t *tl; 7008 struct nfslayout *lyp, *nlyp; 7009 struct nfslayouthash *lhyp; 7010 struct nfslayouthead loclyp; 7011 int i, j; 7012 7013 NFSD_DEBUG(4, "flexmirrordel\n"); 7014 /* Move all layouts found onto a local list. */ 7015 TAILQ_INIT(&loclyp); 7016 for (i = 0; i < nfsrv_layouthashsize; i++) { 7017 lhyp = &nfslayouthash[i]; 7018 NFSLOCKLAYOUT(lhyp); 7019 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { 7020 if (lyp->lay_type == NFSLAYOUT_FLEXFILE && 7021 lyp->lay_mirrorcnt > 1) { 7022 NFSD_DEBUG(4, "possible match\n"); 7023 tl = lyp->lay_xdr; 7024 tl += 3; 7025 for (j = 0; j < lyp->lay_mirrorcnt; j++) { 7026 tl++; 7027 if (NFSBCMP(devid, tl, NFSX_V4DEVICEID) 7028 == 0) { 7029 /* Found one. */ 7030 NFSD_DEBUG(4, "fnd one\n"); 7031 TAILQ_REMOVE(&lhyp->list, lyp, 7032 lay_list); 7033 TAILQ_INSERT_HEAD(&loclyp, lyp, 7034 lay_list); 7035 break; 7036 } 7037 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED + 7038 NFSM_RNDUP(NFSX_V4PNFSFH) / 7039 NFSX_UNSIGNED + 11 * NFSX_UNSIGNED); 7040 } 7041 } 7042 } 7043 NFSUNLOCKLAYOUT(lhyp); 7044 } 7045 7046 /* Now, try to do a Layout recall for each one found. */ 7047 TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) { 7048 NFSD_DEBUG(4, "do layout recall\n"); 7049 /* 7050 * The layout stateid.seqid needs to be incremented 7051 * before doing a LAYOUT_RECALL callback. 7052 */ 7053 if (++lyp->lay_stateid.seqid == 0) 7054 lyp->lay_stateid.seqid = 1; 7055 nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid, 7056 &lyp->lay_fh, lyp, 1, lyp->lay_type, p); 7057 nfsrv_freelayout(&loclyp, lyp); 7058 } 7059 } 7060 7061 /* 7062 * Do a recall callback to the client for this layout. 7063 */ 7064 static int 7065 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp, 7066 struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p) 7067 { 7068 struct nfsclient *clp; 7069 int error; 7070 7071 NFSD_DEBUG(4, "nfsrv_recalllayout\n"); 7072 error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0), 7073 0, NULL, p); 7074 NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error); 7075 if (error != 0) { 7076 printf("nfsrv_recalllayout: getclient err=%d\n", error); 7077 return (error); 7078 } 7079 if ((clp->lc_flags & LCL_NFSV41) != 0) { 7080 error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL, 7081 stateidp, changed, fhp, NULL, NULL, laytype, p); 7082 /* If lyp != NULL, handle an error return here. */ 7083 if (error != 0 && lyp != NULL) { 7084 NFSDRECALLLOCK(); 7085 /* 7086 * Mark it returned, since no layout recall 7087 * has been done. 7088 * All errors seem to be non-recoverable, although 7089 * NFSERR_NOMATCHLAYOUT is a normal event. 7090 */ 7091 if ((lyp->lay_flags & NFSLAY_RECALL) != 0) { 7092 lyp->lay_flags |= NFSLAY_RETURNED; 7093 wakeup(lyp); 7094 } 7095 NFSDRECALLUNLOCK(); 7096 if (error != NFSERR_NOMATCHLAYOUT) 7097 printf("nfsrv_recalllayout: err=%d\n", error); 7098 } 7099 } else 7100 printf("nfsrv_recalllayout: clp not NFSv4.1\n"); 7101 return (error); 7102 } 7103 7104 /* 7105 * Find a layout to recall when we exceed our high water mark. 7106 */ 7107 void 7108 nfsrv_recalloldlayout(NFSPROC_T *p) 7109 { 7110 struct nfslayouthash *lhyp; 7111 struct nfslayout *lyp; 7112 nfsquad_t clientid; 7113 nfsv4stateid_t stateid; 7114 fhandle_t fh; 7115 int error, laytype = 0, ret; 7116 7117 lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize]; 7118 NFSLOCKLAYOUT(lhyp); 7119 TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) { 7120 if ((lyp->lay_flags & NFSLAY_CALLB) == 0) { 7121 lyp->lay_flags |= NFSLAY_CALLB; 7122 /* 7123 * The layout stateid.seqid needs to be incremented 7124 * before doing a LAYOUT_RECALL callback. 7125 */ 7126 if (++lyp->lay_stateid.seqid == 0) 7127 lyp->lay_stateid.seqid = 1; 7128 clientid = lyp->lay_clientid; 7129 stateid = lyp->lay_stateid; 7130 NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh)); 7131 laytype = lyp->lay_type; 7132 break; 7133 } 7134 } 7135 NFSUNLOCKLAYOUT(lhyp); 7136 if (lyp != NULL) { 7137 error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0, 7138 laytype, p); 7139 if (error != 0 && error != NFSERR_NOMATCHLAYOUT) 7140 NFSD_DEBUG(4, "recallold=%d\n", error); 7141 if (error != 0) { 7142 NFSLOCKLAYOUT(lhyp); 7143 /* 7144 * Since the hash list was unlocked, we need to 7145 * find it again. 7146 */ 7147 ret = nfsrv_findlayout(&clientid, &fh, laytype, p, 7148 &lyp); 7149 if (ret == 0 && 7150 (lyp->lay_flags & NFSLAY_CALLB) != 0 && 7151 lyp->lay_stateid.other[0] == stateid.other[0] && 7152 lyp->lay_stateid.other[1] == stateid.other[1] && 7153 lyp->lay_stateid.other[2] == stateid.other[2]) { 7154 /* 7155 * The client no longer knows this layout, so 7156 * it can be free'd now. 7157 */ 7158 if (error == NFSERR_NOMATCHLAYOUT) 7159 nfsrv_freelayout(&lhyp->list, lyp); 7160 else { 7161 /* 7162 * Leave it to be tried later by 7163 * clearing NFSLAY_CALLB and moving 7164 * it to the head of the list, so it 7165 * won't be tried again for a while. 7166 */ 7167 lyp->lay_flags &= ~NFSLAY_CALLB; 7168 TAILQ_REMOVE(&lhyp->list, lyp, 7169 lay_list); 7170 TAILQ_INSERT_HEAD(&lhyp->list, lyp, 7171 lay_list); 7172 } 7173 } 7174 NFSUNLOCKLAYOUT(lhyp); 7175 } 7176 } 7177 } 7178 7179 /* 7180 * Try and return layout(s). 7181 */ 7182 int 7183 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp, 7184 int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim, 7185 int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp, 7186 struct ucred *cred, NFSPROC_T *p) 7187 { 7188 struct nfsvattr na; 7189 struct nfslayouthash *lhyp; 7190 struct nfslayout *lyp; 7191 fhandle_t fh; 7192 int error = 0; 7193 7194 *fndp = 0; 7195 if (kind == NFSV4LAYOUTRET_FILE) { 7196 error = nfsvno_getfh(vp, &fh, p); 7197 if (error == 0) { 7198 error = nfsrv_updatemdsattr(vp, &na, p); 7199 if (error != 0) 7200 printf("nfsrv_layoutreturn: updatemdsattr" 7201 " failed=%d\n", error); 7202 } 7203 if (error == 0) { 7204 if (reclaim == newnfs_true) { 7205 error = nfsrv_checkgrace(NULL, NULL, 7206 NFSLCK_RECLAIM); 7207 if (error != NFSERR_NOGRACE) 7208 error = 0; 7209 return (error); 7210 } 7211 lhyp = NFSLAYOUTHASH(&fh); 7212 NFSDRECALLLOCK(); 7213 NFSLOCKLAYOUT(lhyp); 7214 error = nfsrv_findlayout(&nd->nd_clientid, &fh, 7215 layouttype, p, &lyp); 7216 NFSD_DEBUG(4, "layoutret findlay=%d\n", error); 7217 if (error == 0 && 7218 stateidp->other[0] == lyp->lay_stateid.other[0] && 7219 stateidp->other[1] == lyp->lay_stateid.other[1] && 7220 stateidp->other[2] == lyp->lay_stateid.other[2]) { 7221 NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d" 7222 " %x %x %x laystateid %d %x %x %x" 7223 " off=%ju len=%ju flgs=0x%x\n", 7224 stateidp->seqid, stateidp->other[0], 7225 stateidp->other[1], stateidp->other[2], 7226 lyp->lay_stateid.seqid, 7227 lyp->lay_stateid.other[0], 7228 lyp->lay_stateid.other[1], 7229 lyp->lay_stateid.other[2], 7230 (uintmax_t)offset, (uintmax_t)len, 7231 lyp->lay_flags); 7232 if (++lyp->lay_stateid.seqid == 0) 7233 lyp->lay_stateid.seqid = 1; 7234 stateidp->seqid = lyp->lay_stateid.seqid; 7235 if (offset == 0 && len == UINT64_MAX) { 7236 if ((iomode & NFSLAYOUTIOMODE_READ) != 7237 0) 7238 lyp->lay_flags &= ~NFSLAY_READ; 7239 if ((iomode & NFSLAYOUTIOMODE_RW) != 0) 7240 lyp->lay_flags &= ~NFSLAY_RW; 7241 if ((lyp->lay_flags & (NFSLAY_READ | 7242 NFSLAY_RW)) == 0) 7243 nfsrv_freelayout(&lhyp->list, 7244 lyp); 7245 else 7246 *fndp = 1; 7247 } else 7248 *fndp = 1; 7249 } 7250 NFSUNLOCKLAYOUT(lhyp); 7251 /* Search the nfsrv_recalllist for a match. */ 7252 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { 7253 if (NFSBCMP(&lyp->lay_fh, &fh, 7254 sizeof(fh)) == 0 && 7255 lyp->lay_clientid.qval == 7256 nd->nd_clientid.qval && 7257 stateidp->other[0] == 7258 lyp->lay_stateid.other[0] && 7259 stateidp->other[1] == 7260 lyp->lay_stateid.other[1] && 7261 stateidp->other[2] == 7262 lyp->lay_stateid.other[2]) { 7263 lyp->lay_flags |= NFSLAY_RETURNED; 7264 wakeup(lyp); 7265 error = 0; 7266 } 7267 } 7268 NFSDRECALLUNLOCK(); 7269 } 7270 if (layouttype == NFSLAYOUT_FLEXFILE) 7271 nfsrv_flexlayouterr(nd, layp, maxcnt, p); 7272 } else if (kind == NFSV4LAYOUTRET_FSID) 7273 nfsrv_freelayouts(&nd->nd_clientid, 7274 &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode); 7275 else if (kind == NFSV4LAYOUTRET_ALL) 7276 nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode); 7277 else 7278 error = NFSERR_INVAL; 7279 if (error == -1) 7280 error = 0; 7281 return (error); 7282 } 7283 7284 /* 7285 * Look for an existing layout. 7286 */ 7287 static int 7288 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype, 7289 NFSPROC_T *p, struct nfslayout **lypp) 7290 { 7291 struct nfslayouthash *lhyp; 7292 struct nfslayout *lyp; 7293 int ret; 7294 7295 *lypp = NULL; 7296 ret = 0; 7297 lhyp = NFSLAYOUTHASH(fhp); 7298 TAILQ_FOREACH(lyp, &lhyp->list, lay_list) { 7299 if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 && 7300 lyp->lay_clientid.qval == clientidp->qval && 7301 lyp->lay_type == laytype) 7302 break; 7303 } 7304 if (lyp != NULL) 7305 *lypp = lyp; 7306 else 7307 ret = -1; 7308 return (ret); 7309 } 7310 7311 /* 7312 * Add the new layout, as required. 7313 */ 7314 static int 7315 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp, 7316 nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p) 7317 { 7318 struct nfsclient *clp; 7319 struct nfslayouthash *lhyp; 7320 struct nfslayout *lyp, *nlyp; 7321 fhandle_t *fhp; 7322 int error; 7323 7324 KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0, 7325 ("nfsrv_layoutget: no nd_clientid\n")); 7326 lyp = *lypp; 7327 fhp = &lyp->lay_fh; 7328 NFSLOCKSTATE(); 7329 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, 7330 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); 7331 if (error != 0) { 7332 NFSUNLOCKSTATE(); 7333 return (error); 7334 } 7335 lyp->lay_stateid.seqid = stateidp->seqid = 1; 7336 lyp->lay_stateid.other[0] = stateidp->other[0] = 7337 clp->lc_clientid.lval[0]; 7338 lyp->lay_stateid.other[1] = stateidp->other[1] = 7339 clp->lc_clientid.lval[1]; 7340 lyp->lay_stateid.other[2] = stateidp->other[2] = 7341 nfsrv_nextstateindex(clp); 7342 NFSUNLOCKSTATE(); 7343 7344 lhyp = NFSLAYOUTHASH(fhp); 7345 NFSLOCKLAYOUT(lhyp); 7346 TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) { 7347 if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 && 7348 nlyp->lay_clientid.qval == nd->nd_clientid.qval) 7349 break; 7350 } 7351 if (nlyp != NULL) { 7352 /* A layout already exists, so use it. */ 7353 nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)); 7354 NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen); 7355 *layoutlenp = nlyp->lay_layoutlen; 7356 if (++nlyp->lay_stateid.seqid == 0) 7357 nlyp->lay_stateid.seqid = 1; 7358 stateidp->seqid = nlyp->lay_stateid.seqid; 7359 stateidp->other[0] = nlyp->lay_stateid.other[0]; 7360 stateidp->other[1] = nlyp->lay_stateid.other[1]; 7361 stateidp->other[2] = nlyp->lay_stateid.other[2]; 7362 NFSUNLOCKLAYOUT(lhyp); 7363 return (0); 7364 } 7365 7366 /* Insert the new layout in the lists. */ 7367 *lypp = NULL; 7368 atomic_add_int(&nfsrv_layoutcnt, 1); 7369 NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen); 7370 *layoutlenp = lyp->lay_layoutlen; 7371 TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list); 7372 NFSUNLOCKLAYOUT(lhyp); 7373 return (0); 7374 } 7375 7376 /* 7377 * Get the devinfo for a deviceid. 7378 */ 7379 int 7380 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt, 7381 uint32_t *notify, int *devaddrlen, char **devaddr) 7382 { 7383 struct nfsdevice *ds; 7384 7385 if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype != 7386 NFSLAYOUT_FLEXFILE) || 7387 (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES)) 7388 return (NFSERR_UNKNLAYOUTTYPE); 7389 7390 /* 7391 * Now, search for the device id. Note that the structures won't go 7392 * away, but the order changes in the list. As such, the lock only 7393 * needs to be held during the search through the list. 7394 */ 7395 NFSDDSLOCK(); 7396 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7397 if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 && 7398 ds->nfsdev_nmp != NULL) 7399 break; 7400 } 7401 NFSDDSUNLOCK(); 7402 if (ds == NULL) 7403 return (NFSERR_NOENT); 7404 7405 /* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */ 7406 *devaddrlen = 0; 7407 if (layouttype == NFSLAYOUT_NFSV4_1_FILES) { 7408 *devaddrlen = ds->nfsdev_fileaddrlen; 7409 *devaddr = ds->nfsdev_fileaddr; 7410 } else if (layouttype == NFSLAYOUT_FLEXFILE) { 7411 *devaddrlen = ds->nfsdev_flexaddrlen; 7412 *devaddr = ds->nfsdev_flexaddr; 7413 } 7414 if (*devaddrlen == 0) 7415 return (NFSERR_UNKNLAYOUTTYPE); 7416 7417 /* 7418 * The XDR overhead is 3 unsigned values: layout_type, 7419 * length_of_address and notify bitmap. 7420 * If the notify array is changed to not all zeros, the 7421 * count of unsigned values must be increased. 7422 */ 7423 if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) + 7424 3 * NFSX_UNSIGNED) { 7425 *maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED; 7426 return (NFSERR_TOOSMALL); 7427 } 7428 return (0); 7429 } 7430 7431 /* 7432 * Free a list of layout state structures. 7433 */ 7434 static void 7435 nfsrv_freelayoutlist(nfsquad_t clientid) 7436 { 7437 struct nfslayouthash *lhyp; 7438 struct nfslayout *lyp, *nlyp; 7439 int i; 7440 7441 for (i = 0; i < nfsrv_layouthashsize; i++) { 7442 lhyp = &nfslayouthash[i]; 7443 NFSLOCKLAYOUT(lhyp); 7444 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { 7445 if (lyp->lay_clientid.qval == clientid.qval) 7446 nfsrv_freelayout(&lhyp->list, lyp); 7447 } 7448 NFSUNLOCKLAYOUT(lhyp); 7449 } 7450 } 7451 7452 /* 7453 * Free up a layout. 7454 */ 7455 static void 7456 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp) 7457 { 7458 7459 NFSD_DEBUG(4, "Freelayout=%p\n", lyp); 7460 atomic_add_int(&nfsrv_layoutcnt, -1); 7461 TAILQ_REMOVE(lhp, lyp, lay_list); 7462 free(lyp, M_NFSDSTATE); 7463 } 7464 7465 /* 7466 * Free up a device id. 7467 */ 7468 void 7469 nfsrv_freeonedevid(struct nfsdevice *ds) 7470 { 7471 int i; 7472 7473 atomic_add_int(&nfsrv_devidcnt, -1); 7474 vrele(ds->nfsdev_dvp); 7475 for (i = 0; i < nfsrv_dsdirsize; i++) 7476 if (ds->nfsdev_dsdir[i] != NULL) 7477 vrele(ds->nfsdev_dsdir[i]); 7478 free(ds->nfsdev_fileaddr, M_NFSDSTATE); 7479 free(ds->nfsdev_flexaddr, M_NFSDSTATE); 7480 free(ds->nfsdev_host, M_NFSDSTATE); 7481 free(ds, M_NFSDSTATE); 7482 } 7483 7484 /* 7485 * Free up a device id and its mirrors. 7486 */ 7487 static void 7488 nfsrv_freedevid(struct nfsdevice *ds) 7489 { 7490 7491 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 7492 nfsrv_freeonedevid(ds); 7493 } 7494 7495 /* 7496 * Free all layouts and device ids. 7497 * Done when the nfsd threads are shut down since there may be a new 7498 * modified device id list created when the nfsd is restarted. 7499 */ 7500 void 7501 nfsrv_freealllayoutsanddevids(void) 7502 { 7503 struct nfsdontlist *mrp, *nmrp; 7504 struct nfslayout *lyp, *nlyp; 7505 7506 /* Get rid of the deviceid structures. */ 7507 nfsrv_freealldevids(); 7508 TAILQ_INIT(&nfsrv_devidhead); 7509 nfsrv_devidcnt = 0; 7510 7511 /* Get rid of all layouts. */ 7512 nfsrv_freealllayouts(); 7513 7514 /* Get rid of any nfsdontlist entries. */ 7515 LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp) 7516 free(mrp, M_NFSDSTATE); 7517 LIST_INIT(&nfsrv_dontlisthead); 7518 nfsrv_dontlistlen = 0; 7519 7520 /* Free layouts in the recall list. */ 7521 TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp) 7522 nfsrv_freelayout(&nfsrv_recalllisthead, lyp); 7523 TAILQ_INIT(&nfsrv_recalllisthead); 7524 } 7525 7526 /* 7527 * Free layouts that match the arguments. 7528 */ 7529 static void 7530 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode) 7531 { 7532 struct nfslayouthash *lhyp; 7533 struct nfslayout *lyp, *nlyp; 7534 int i; 7535 7536 for (i = 0; i < nfsrv_layouthashsize; i++) { 7537 lhyp = &nfslayouthash[i]; 7538 NFSLOCKLAYOUT(lhyp); 7539 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { 7540 if (clid->qval != lyp->lay_clientid.qval) 7541 continue; 7542 if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0) 7543 continue; 7544 if (laytype != lyp->lay_type) 7545 continue; 7546 if ((iomode & NFSLAYOUTIOMODE_READ) != 0) 7547 lyp->lay_flags &= ~NFSLAY_READ; 7548 if ((iomode & NFSLAYOUTIOMODE_RW) != 0) 7549 lyp->lay_flags &= ~NFSLAY_RW; 7550 if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0) 7551 nfsrv_freelayout(&lhyp->list, lyp); 7552 } 7553 NFSUNLOCKLAYOUT(lhyp); 7554 } 7555 } 7556 7557 /* 7558 * Free all layouts for the argument file. 7559 */ 7560 void 7561 nfsrv_freefilelayouts(fhandle_t *fhp) 7562 { 7563 struct nfslayouthash *lhyp; 7564 struct nfslayout *lyp, *nlyp; 7565 7566 lhyp = NFSLAYOUTHASH(fhp); 7567 NFSLOCKLAYOUT(lhyp); 7568 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { 7569 if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0) 7570 nfsrv_freelayout(&lhyp->list, lyp); 7571 } 7572 NFSUNLOCKLAYOUT(lhyp); 7573 } 7574 7575 /* 7576 * Free all layouts. 7577 */ 7578 static void 7579 nfsrv_freealllayouts(void) 7580 { 7581 struct nfslayouthash *lhyp; 7582 struct nfslayout *lyp, *nlyp; 7583 int i; 7584 7585 for (i = 0; i < nfsrv_layouthashsize; i++) { 7586 lhyp = &nfslayouthash[i]; 7587 NFSLOCKLAYOUT(lhyp); 7588 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) 7589 nfsrv_freelayout(&lhyp->list, lyp); 7590 NFSUNLOCKLAYOUT(lhyp); 7591 } 7592 } 7593 7594 /* 7595 * Look up the mount path for the DS server. 7596 */ 7597 static int 7598 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p, 7599 struct nfsdevice **dsp) 7600 { 7601 struct nameidata nd; 7602 struct nfsdevice *ds; 7603 struct mount *mp; 7604 int error, i; 7605 char *dsdirpath; 7606 size_t dsdirsize; 7607 7608 NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp); 7609 *dsp = NULL; 7610 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, 7611 dspathp, p); 7612 error = namei(&nd); 7613 NFSD_DEBUG(4, "lookup=%d\n", error); 7614 if (error != 0) 7615 return (error); 7616 if (nd.ni_vp->v_type != VDIR) { 7617 vput(nd.ni_vp); 7618 NFSD_DEBUG(4, "dspath not dir\n"); 7619 return (ENOTDIR); 7620 } 7621 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { 7622 vput(nd.ni_vp); 7623 NFSD_DEBUG(4, "dspath not an NFS mount\n"); 7624 return (ENXIO); 7625 } 7626 7627 /* 7628 * Allocate a DS server structure with the NFS mounted directory 7629 * vnode reference counted, so that a non-forced dismount will 7630 * fail with EBUSY. 7631 * This structure is always linked into the list, even if an error 7632 * is being returned. The caller will free the entire list upon 7633 * an error return. 7634 */ 7635 *dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t), 7636 M_NFSDSTATE, M_WAITOK | M_ZERO); 7637 ds->nfsdev_dvp = nd.ni_vp; 7638 ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount); 7639 NFSVOPUNLOCK(nd.ni_vp); 7640 7641 dsdirsize = strlen(dspathp) + 16; 7642 dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK); 7643 /* Now, create the DS directory structures. */ 7644 for (i = 0; i < nfsrv_dsdirsize; i++) { 7645 snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i); 7646 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 7647 UIO_SYSSPACE, dsdirpath, p); 7648 error = namei(&nd); 7649 NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error); 7650 if (error != 0) 7651 break; 7652 if (nd.ni_vp->v_type != VDIR) { 7653 vput(nd.ni_vp); 7654 error = ENOTDIR; 7655 NFSD_DEBUG(4, "dsdirpath not a VDIR\n"); 7656 break; 7657 } 7658 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { 7659 vput(nd.ni_vp); 7660 error = ENXIO; 7661 NFSD_DEBUG(4, "dsdirpath not an NFS mount\n"); 7662 break; 7663 } 7664 ds->nfsdev_dsdir[i] = nd.ni_vp; 7665 NFSVOPUNLOCK(nd.ni_vp); 7666 } 7667 free(dsdirpath, M_TEMP); 7668 7669 if (strlen(mdspathp) > 0) { 7670 /* 7671 * This DS stores file for a specific MDS exported file 7672 * system. 7673 */ 7674 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 7675 UIO_SYSSPACE, mdspathp, p); 7676 error = namei(&nd); 7677 NFSD_DEBUG(4, "mds lookup=%d\n", error); 7678 if (error != 0) 7679 goto out; 7680 if (nd.ni_vp->v_type != VDIR) { 7681 vput(nd.ni_vp); 7682 error = ENOTDIR; 7683 NFSD_DEBUG(4, "mdspath not dir\n"); 7684 goto out; 7685 } 7686 mp = nd.ni_vp->v_mount; 7687 if ((mp->mnt_flag & MNT_EXPORTED) == 0) { 7688 vput(nd.ni_vp); 7689 error = ENXIO; 7690 NFSD_DEBUG(4, "mdspath not an exported fs\n"); 7691 goto out; 7692 } 7693 ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid; 7694 ds->nfsdev_mdsisset = 1; 7695 vput(nd.ni_vp); 7696 } 7697 7698 out: 7699 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 7700 atomic_add_int(&nfsrv_devidcnt, 1); 7701 return (error); 7702 } 7703 7704 /* 7705 * Look up the mount path for the DS server and delete it. 7706 */ 7707 int 7708 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p) 7709 { 7710 struct mount *mp; 7711 struct nfsmount *nmp; 7712 struct nfsdevice *ds; 7713 int error; 7714 7715 NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp); 7716 /* 7717 * Search for the path in the mount list. Avoid looking the path 7718 * up, since this mount point may be hung, with associated locked 7719 * vnodes, etc. 7720 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked 7721 * until this completes. 7722 * As noted in the man page, this should be done before any forced 7723 * dismount on the mount point, but at least the handshake on 7724 * NFSMNTP_CANCELRPCS should make it safe. 7725 */ 7726 error = 0; 7727 ds = NULL; 7728 nmp = NULL; 7729 mtx_lock(&mountlist_mtx); 7730 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 7731 if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 && 7732 strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 && 7733 mp->mnt_data != NULL) { 7734 nmp = VFSTONFS(mp); 7735 NFSLOCKMNT(nmp); 7736 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 7737 NFSMNTP_CANCELRPCS)) == 0) { 7738 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 7739 NFSUNLOCKMNT(nmp); 7740 } else { 7741 NFSUNLOCKMNT(nmp); 7742 nmp = NULL; 7743 } 7744 break; 7745 } 7746 } 7747 mtx_unlock(&mountlist_mtx); 7748 7749 if (nmp != NULL) { 7750 ds = nfsrv_deldsnmp(op, nmp, p); 7751 NFSD_DEBUG(4, "deldsnmp=%p\n", ds); 7752 if (ds != NULL) { 7753 nfsrv_killrpcs(nmp); 7754 NFSD_DEBUG(4, "aft killrpcs\n"); 7755 } else 7756 error = ENXIO; 7757 NFSLOCKMNT(nmp); 7758 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 7759 wakeup(nmp); 7760 NFSUNLOCKMNT(nmp); 7761 } else 7762 error = EINVAL; 7763 return (error); 7764 } 7765 7766 /* 7767 * Search for and remove a DS entry which matches the "nmp" argument. 7768 * The nfsdevice structure pointer is returned so that the caller can 7769 * free it via nfsrv_freeonedevid(). 7770 * For the forced case, do not try to do LayoutRecalls, since the server 7771 * must be shut down now anyhow. 7772 */ 7773 struct nfsdevice * 7774 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p) 7775 { 7776 struct nfsdevice *fndds; 7777 7778 NFSD_DEBUG(4, "deldsdvp\n"); 7779 NFSDDSLOCK(); 7780 if (op == PNFSDOP_FORCEDELDS) 7781 fndds = nfsv4_findmirror(nmp); 7782 else 7783 fndds = nfsrv_findmirroredds(nmp); 7784 if (fndds != NULL) 7785 nfsrv_deleteds(fndds); 7786 NFSDDSUNLOCK(); 7787 if (fndds != NULL) { 7788 if (op != PNFSDOP_FORCEDELDS) 7789 nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p); 7790 printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host); 7791 } 7792 return (fndds); 7793 } 7794 7795 /* 7796 * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid. 7797 * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount 7798 * point. 7799 * Also, returns an error instead of the nfsdevice found. 7800 */ 7801 int 7802 nfsrv_delds(char *devid, NFSPROC_T *p) 7803 { 7804 struct nfsdevice *ds, *fndds; 7805 struct nfsmount *nmp; 7806 int fndmirror; 7807 7808 NFSD_DEBUG(4, "delds\n"); 7809 /* 7810 * Search the DS server list for a match with devid. 7811 * Remove the DS entry if found and there is a mirror. 7812 */ 7813 fndds = NULL; 7814 nmp = NULL; 7815 fndmirror = 0; 7816 NFSDDSLOCK(); 7817 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7818 if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 && 7819 ds->nfsdev_nmp != NULL) { 7820 NFSD_DEBUG(4, "fnd main ds\n"); 7821 fndds = ds; 7822 break; 7823 } 7824 } 7825 if (fndds == NULL) { 7826 NFSDDSUNLOCK(); 7827 return (ENXIO); 7828 } 7829 if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0) 7830 fndmirror = 1; 7831 else if (fndds->nfsdev_mdsisset != 0) { 7832 /* For the fsid is set case, search for a mirror. */ 7833 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7834 if (ds != fndds && ds->nfsdev_nmp != NULL && 7835 ds->nfsdev_mdsisset != 0 && 7836 fsidcmp(&ds->nfsdev_mdsfsid, 7837 &fndds->nfsdev_mdsfsid) == 0) { 7838 fndmirror = 1; 7839 break; 7840 } 7841 } 7842 } 7843 if (fndmirror != 0) { 7844 nmp = fndds->nfsdev_nmp; 7845 NFSLOCKMNT(nmp); 7846 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 7847 NFSMNTP_CANCELRPCS)) == 0) { 7848 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 7849 NFSUNLOCKMNT(nmp); 7850 nfsrv_deleteds(fndds); 7851 } else { 7852 NFSUNLOCKMNT(nmp); 7853 nmp = NULL; 7854 } 7855 } 7856 NFSDDSUNLOCK(); 7857 if (nmp != NULL) { 7858 nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p); 7859 printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host); 7860 nfsrv_killrpcs(nmp); 7861 NFSLOCKMNT(nmp); 7862 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 7863 wakeup(nmp); 7864 NFSUNLOCKMNT(nmp); 7865 return (0); 7866 } 7867 return (ENXIO); 7868 } 7869 7870 /* 7871 * Mark a DS as disabled by setting nfsdev_nmp = NULL. 7872 */ 7873 static void 7874 nfsrv_deleteds(struct nfsdevice *fndds) 7875 { 7876 7877 NFSD_DEBUG(4, "deleteds: deleting a mirror\n"); 7878 fndds->nfsdev_nmp = NULL; 7879 if (fndds->nfsdev_mdsisset == 0) 7880 nfsrv_faildscnt--; 7881 } 7882 7883 /* 7884 * Fill in the addr structures for the File and Flex File layouts. 7885 */ 7886 static void 7887 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost) 7888 { 7889 uint32_t *tl; 7890 char *netprot; 7891 int addrlen; 7892 static uint64_t new_devid = 0; 7893 7894 if (strchr(addr, ':') != NULL) 7895 netprot = "tcp6"; 7896 else 7897 netprot = "tcp"; 7898 7899 /* Fill in the device id. */ 7900 NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time)); 7901 new_devid++; 7902 NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)], 7903 sizeof(new_devid)); 7904 7905 /* 7906 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4 7907 * as defined in RFC5661) in XDR. 7908 */ 7909 addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) + 7910 6 * NFSX_UNSIGNED; 7911 NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot); 7912 ds->nfsdev_fileaddrlen = addrlen; 7913 tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO); 7914 ds->nfsdev_fileaddr = (char *)tl; 7915 *tl++ = txdr_unsigned(1); /* One stripe with index 0. */ 7916 *tl++ = 0; 7917 *tl++ = txdr_unsigned(1); /* One multipath list */ 7918 *tl++ = txdr_unsigned(1); /* with one entry in it. */ 7919 /* The netaddr for this one entry. */ 7920 *tl++ = txdr_unsigned(strlen(netprot)); 7921 NFSBCOPY(netprot, tl, strlen(netprot)); 7922 tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED); 7923 *tl++ = txdr_unsigned(strlen(addr)); 7924 NFSBCOPY(addr, tl, strlen(addr)); 7925 7926 /* 7927 * Fill in the flex file addr (actually the ff_device_addr4 7928 * as defined for Flexible File Layout) in XDR. 7929 */ 7930 addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) + 7931 14 * NFSX_UNSIGNED; 7932 ds->nfsdev_flexaddrlen = addrlen; 7933 tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO); 7934 ds->nfsdev_flexaddr = (char *)tl; 7935 *tl++ = txdr_unsigned(1); /* One multipath entry. */ 7936 /* The netaddr for this one entry. */ 7937 *tl++ = txdr_unsigned(strlen(netprot)); 7938 NFSBCOPY(netprot, tl, strlen(netprot)); 7939 tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED); 7940 *tl++ = txdr_unsigned(strlen(addr)); 7941 NFSBCOPY(addr, tl, strlen(addr)); 7942 tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED); 7943 *tl++ = txdr_unsigned(2); /* Two NFS Versions. */ 7944 *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */ 7945 *tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */ 7946 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max rsize. */ 7947 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max wsize. */ 7948 *tl++ = newnfs_true; /* Tightly coupled. */ 7949 *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */ 7950 *tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */ 7951 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max rsize. */ 7952 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max wsize. */ 7953 *tl = newnfs_true; /* Tightly coupled. */ 7954 7955 ds->nfsdev_hostnamelen = strlen(dnshost); 7956 ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE, 7957 M_WAITOK); 7958 NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1); 7959 } 7960 7961 7962 /* 7963 * Create the device id list. 7964 * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument 7965 * is misconfigured. 7966 */ 7967 int 7968 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p) 7969 { 7970 struct nfsdevice *ds; 7971 char *addrp, *dnshostp, *dspathp, *mdspathp; 7972 int error, i; 7973 7974 addrp = args->addr; 7975 dnshostp = args->dnshost; 7976 dspathp = args->dspath; 7977 mdspathp = args->mdspath; 7978 nfsrv_maxpnfsmirror = args->mirrorcnt; 7979 if (addrp == NULL || dnshostp == NULL || dspathp == NULL || 7980 mdspathp == NULL) 7981 return (0); 7982 7983 /* 7984 * Loop around for each nul-terminated string in args->addr, 7985 * args->dnshost, args->dnspath and args->mdspath. 7986 */ 7987 while (addrp < (args->addr + args->addrlen) && 7988 dnshostp < (args->dnshost + args->dnshostlen) && 7989 dspathp < (args->dspath + args->dspathlen) && 7990 mdspathp < (args->mdspath + args->mdspathlen)) { 7991 error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds); 7992 if (error != 0) { 7993 /* Free all DS servers. */ 7994 nfsrv_freealldevids(); 7995 nfsrv_devidcnt = 0; 7996 return (ENXIO); 7997 } 7998 nfsrv_allocdevid(ds, addrp, dnshostp); 7999 addrp += (strlen(addrp) + 1); 8000 dnshostp += (strlen(dnshostp) + 1); 8001 dspathp += (strlen(dspathp) + 1); 8002 mdspathp += (strlen(mdspathp) + 1); 8003 } 8004 if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) { 8005 /* Free all DS servers. */ 8006 nfsrv_freealldevids(); 8007 nfsrv_devidcnt = 0; 8008 nfsrv_maxpnfsmirror = 1; 8009 return (ENXIO); 8010 } 8011 /* We can fail at most one less DS than the mirror level. */ 8012 nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1; 8013 8014 /* 8015 * Allocate the nfslayout hash table now, since this is a pNFS server. 8016 * Make it 1% of the high water mark and at least 100. 8017 */ 8018 if (nfslayouthash == NULL) { 8019 nfsrv_layouthashsize = nfsrv_layouthighwater / 100; 8020 if (nfsrv_layouthashsize < 100) 8021 nfsrv_layouthashsize = 100; 8022 nfslayouthash = mallocarray(nfsrv_layouthashsize, 8023 sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK | 8024 M_ZERO); 8025 for (i = 0; i < nfsrv_layouthashsize; i++) { 8026 mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF); 8027 TAILQ_INIT(&nfslayouthash[i].list); 8028 } 8029 } 8030 return (0); 8031 } 8032 8033 /* 8034 * Free all device ids. 8035 */ 8036 static void 8037 nfsrv_freealldevids(void) 8038 { 8039 struct nfsdevice *ds, *nds; 8040 8041 TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds) 8042 nfsrv_freedevid(ds); 8043 } 8044 8045 /* 8046 * Check to see if there is a Read/Write Layout plus either: 8047 * - A Write Delegation 8048 * or 8049 * - An Open with Write_access. 8050 * Return 1 if this is the case and 0 otherwise. 8051 * This function is used by nfsrv_proxyds() to decide if doing a Proxy 8052 * Getattr RPC to the Data Server (DS) is necessary. 8053 */ 8054 #define NFSCLIDVECSIZE 6 8055 int 8056 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p) 8057 { 8058 fhandle_t fh, *tfhp; 8059 struct nfsstate *stp; 8060 struct nfslayout *lyp; 8061 struct nfslayouthash *lhyp; 8062 struct nfslockhashhead *hp; 8063 struct nfslockfile *lfp; 8064 nfsquad_t clid[NFSCLIDVECSIZE]; 8065 int clidcnt, ret; 8066 8067 ret = nfsvno_getfh(vp, &fh, p); 8068 if (ret != 0) 8069 return (0); 8070 8071 /* First check for a Read/Write Layout. */ 8072 clidcnt = 0; 8073 lhyp = NFSLAYOUTHASH(&fh); 8074 NFSLOCKLAYOUT(lhyp); 8075 TAILQ_FOREACH(lyp, &lhyp->list, lay_list) { 8076 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && 8077 ((lyp->lay_flags & NFSLAY_RW) != 0 || 8078 ((lyp->lay_flags & NFSLAY_READ) != 0 && 8079 nfsrv_pnfsatime != 0))) { 8080 if (clidcnt < NFSCLIDVECSIZE) 8081 clid[clidcnt].qval = lyp->lay_clientid.qval; 8082 clidcnt++; 8083 } 8084 } 8085 NFSUNLOCKLAYOUT(lhyp); 8086 if (clidcnt == 0) { 8087 /* None found, so return 0. */ 8088 return (0); 8089 } 8090 8091 /* Get the nfslockfile for this fh. */ 8092 NFSLOCKSTATE(); 8093 hp = NFSLOCKHASH(&fh); 8094 LIST_FOREACH(lfp, hp, lf_hash) { 8095 tfhp = &lfp->lf_fh; 8096 if (NFSVNO_CMPFH(&fh, tfhp)) 8097 break; 8098 } 8099 if (lfp == NULL) { 8100 /* None found, so return 0. */ 8101 NFSUNLOCKSTATE(); 8102 return (0); 8103 } 8104 8105 /* Now, look for a Write delegation for this clientid. */ 8106 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { 8107 if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 && 8108 nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0) 8109 break; 8110 } 8111 if (stp != NULL) { 8112 /* Found one, so return 1. */ 8113 NFSUNLOCKSTATE(); 8114 return (1); 8115 } 8116 8117 /* No Write delegation, so look for an Open with Write_access. */ 8118 LIST_FOREACH(stp, &lfp->lf_open, ls_file) { 8119 KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0, 8120 ("nfsrv_checkdsattr: Non-open in Open list\n")); 8121 if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 && 8122 nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0) 8123 break; 8124 } 8125 NFSUNLOCKSTATE(); 8126 if (stp != NULL) 8127 return (1); 8128 return (0); 8129 } 8130 8131 /* 8132 * Look for a matching clientid in the vector. Return 1 if one might match. 8133 */ 8134 static int 8135 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt) 8136 { 8137 int i; 8138 8139 /* If too many for the vector, return 1 since there might be a match. */ 8140 if (clidcnt > NFSCLIDVECSIZE) 8141 return (1); 8142 8143 for (i = 0; i < clidcnt; i++) 8144 if (clidvec[i].qval == clid.qval) 8145 return (1); 8146 return (0); 8147 } 8148 8149 /* 8150 * Check the don't list for "vp" and see if issuing an rw layout is allowed. 8151 * Return 1 if issuing an rw layout isn't allowed, 0 otherwise. 8152 */ 8153 static int 8154 nfsrv_dontlayout(fhandle_t *fhp) 8155 { 8156 struct nfsdontlist *mrp; 8157 int ret; 8158 8159 if (nfsrv_dontlistlen == 0) 8160 return (0); 8161 ret = 0; 8162 NFSDDONTLISTLOCK(); 8163 LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) { 8164 if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 && 8165 (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) { 8166 ret = 1; 8167 break; 8168 } 8169 } 8170 NFSDDONTLISTUNLOCK(); 8171 return (ret); 8172 } 8173 8174 #define PNFSDS_COPYSIZ 65536 8175 /* 8176 * Create a new file on a DS and copy the contents of an extant DS file to it. 8177 * This can be used for recovery of a DS file onto a recovered DS. 8178 * The steps are: 8179 * - When called, the MDS file's vnode is locked, blocking LayoutGet operations. 8180 * - Disable issuing of read/write layouts for the file via the nfsdontlist, 8181 * so that they will be disabled after the MDS file's vnode is unlocked. 8182 * - Set up the nfsrv_recalllist so that recall of read/write layouts can 8183 * be done. 8184 * - Unlock the MDS file's vnode, so that the client(s) can perform proxied 8185 * writes, LayoutCommits and LayoutReturns for the file when completing the 8186 * LayoutReturn requested by the LayoutRecall callback. 8187 * - Issue a LayoutRecall callback for all read/write layouts and wait for 8188 * them to be returned. (If the LayoutRecall callback replies 8189 * NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.) 8190 * - Exclusively lock the MDS file's vnode. This ensures that no proxied 8191 * writes are in progress or can occur during the DS file copy. 8192 * It also blocks Setattr operations. 8193 * - Create the file on the recovered mirror. 8194 * - Copy the file from the operational DS. 8195 * - Copy any ACL from the MDS file to the new DS file. 8196 * - Set the modify time of the new DS file to that of the MDS file. 8197 * - Update the extended attribute for the MDS file. 8198 * - Enable issuing of rw layouts by deleting the nfsdontlist entry. 8199 * - The caller will unlock the MDS file's vnode allowing operations 8200 * to continue normally, since it is now on the mirror again. 8201 */ 8202 int 8203 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds, 8204 struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt, 8205 struct ucred *cred, NFSPROC_T *p) 8206 { 8207 struct nfsdontlist *mrp, *nmrp; 8208 struct nfslayouthash *lhyp; 8209 struct nfslayout *lyp, *nlyp; 8210 struct nfslayouthead thl; 8211 struct mount *mp, *tvmp; 8212 struct acl *aclp; 8213 struct vattr va; 8214 struct timespec mtime; 8215 fhandle_t fh; 8216 vnode_t tvp; 8217 off_t rdpos, wrpos; 8218 ssize_t aresid; 8219 char *dat; 8220 int didprintf, ret, retacl, xfer; 8221 8222 ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp"); 8223 ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp"); 8224 /* 8225 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag 8226 * so that no more RW layouts will get issued. 8227 */ 8228 ret = nfsvno_getfh(vp, &fh, p); 8229 if (ret != 0) { 8230 NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret); 8231 return (ret); 8232 } 8233 nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK); 8234 nmrp->nfsmr_flags = NFSMR_DONTLAYOUT; 8235 NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh)); 8236 NFSDDONTLISTLOCK(); 8237 LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) { 8238 if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0) 8239 break; 8240 } 8241 if (mrp == NULL) { 8242 LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list); 8243 mrp = nmrp; 8244 nmrp = NULL; 8245 nfsrv_dontlistlen++; 8246 NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n"); 8247 } else { 8248 NFSDDONTLISTUNLOCK(); 8249 free(nmrp, M_NFSDSTATE); 8250 NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n"); 8251 return (ENXIO); 8252 } 8253 NFSDDONTLISTUNLOCK(); 8254 8255 /* 8256 * Search for all RW layouts for this file. Move them to the 8257 * recall list, so they can be recalled and their return noted. 8258 */ 8259 lhyp = NFSLAYOUTHASH(&fh); 8260 NFSDRECALLLOCK(); 8261 NFSLOCKLAYOUT(lhyp); 8262 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { 8263 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && 8264 (lyp->lay_flags & NFSLAY_RW) != 0) { 8265 TAILQ_REMOVE(&lhyp->list, lyp, lay_list); 8266 TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list); 8267 lyp->lay_trycnt = 0; 8268 } 8269 } 8270 NFSUNLOCKLAYOUT(lhyp); 8271 NFSDRECALLUNLOCK(); 8272 8273 ret = 0; 8274 mp = tvmp = NULL; 8275 didprintf = 0; 8276 TAILQ_INIT(&thl); 8277 /* Unlock the MDS vp, so that a LayoutReturn can be done on it. */ 8278 NFSVOPUNLOCK(vp); 8279 /* Now, do a recall for all layouts not yet recalled. */ 8280 tryagain: 8281 NFSDRECALLLOCK(); 8282 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { 8283 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && 8284 (lyp->lay_flags & NFSLAY_RECALL) == 0) { 8285 lyp->lay_flags |= NFSLAY_RECALL; 8286 /* 8287 * The layout stateid.seqid needs to be incremented 8288 * before doing a LAYOUT_RECALL callback. 8289 */ 8290 if (++lyp->lay_stateid.seqid == 0) 8291 lyp->lay_stateid.seqid = 1; 8292 NFSDRECALLUNLOCK(); 8293 nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid, 8294 &lyp->lay_fh, lyp, 0, lyp->lay_type, p); 8295 NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n"); 8296 goto tryagain; 8297 } 8298 } 8299 8300 /* Now wait for them to be returned. */ 8301 tryagain2: 8302 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { 8303 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) { 8304 if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) { 8305 TAILQ_REMOVE(&nfsrv_recalllisthead, lyp, 8306 lay_list); 8307 TAILQ_INSERT_HEAD(&thl, lyp, lay_list); 8308 NFSD_DEBUG(4, 8309 "nfsrv_copymr: layout returned\n"); 8310 } else { 8311 lyp->lay_trycnt++; 8312 ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR, 8313 PVFS | PCATCH, "nfsmrl", hz); 8314 NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n", 8315 ret); 8316 if (ret == EINTR || ret == ERESTART) 8317 break; 8318 if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) { 8319 /* 8320 * Give up after 60sec and return 8321 * ENXIO, failing the copymr. 8322 * This layout will remain on the 8323 * recalllist. It can only be cleared 8324 * by restarting the nfsd. 8325 * This seems the safe way to handle 8326 * it, since it cannot be safely copied 8327 * with an outstanding RW layout. 8328 */ 8329 if (lyp->lay_trycnt >= 60) { 8330 ret = ENXIO; 8331 break; 8332 } 8333 if (didprintf == 0) { 8334 printf("nfsrv_copymr: layout " 8335 "not returned\n"); 8336 didprintf = 1; 8337 } 8338 } 8339 } 8340 goto tryagain2; 8341 } 8342 } 8343 NFSDRECALLUNLOCK(); 8344 /* We can now get rid of the layouts that have been returned. */ 8345 TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp) 8346 nfsrv_freelayout(&thl, lyp); 8347 8348 /* 8349 * Do the vn_start_write() calls here, before the MDS vnode is 8350 * locked and the tvp is created (locked) in the NFS file system 8351 * that dvp is in. 8352 * For tvmp, this probably isn't necessary, since it will be an 8353 * NFS mount and they are not suspendable at this time. 8354 */ 8355 if (ret == 0) 8356 ret = vn_start_write(vp, &mp, V_WAIT | PCATCH); 8357 if (ret == 0) { 8358 tvmp = dvp->v_mount; 8359 ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH); 8360 } 8361 8362 /* 8363 * LK_EXCLUSIVE lock the MDS vnode, so that any 8364 * proxied writes through the MDS will be blocked until we have 8365 * completed the copy and update of the extended attributes. 8366 * This will also ensure that any attributes and ACL will not be 8367 * changed until the copy is complete. 8368 */ 8369 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 8370 if (ret == 0 && VN_IS_DOOMED(vp)) { 8371 NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n"); 8372 ret = ESTALE; 8373 } 8374 8375 /* Create the data file on the recovered DS. */ 8376 if (ret == 0) 8377 ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp); 8378 8379 /* Copy the DS file, if created successfully. */ 8380 if (ret == 0) { 8381 /* 8382 * Get any NFSv4 ACL on the MDS file, so that it can be set 8383 * on the new DS file. 8384 */ 8385 aclp = acl_alloc(M_WAITOK | M_ZERO); 8386 retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); 8387 if (retacl != 0 && retacl != ENOATTR) 8388 NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl); 8389 dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK); 8390 /* Malloc a block of 0s used to check for holes. */ 8391 if (nfsrv_zeropnfsdat == NULL) 8392 nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP, 8393 M_WAITOK | M_ZERO); 8394 rdpos = wrpos = 0; 8395 ret = VOP_GETATTR(fvp, &va, cred); 8396 aresid = 0; 8397 while (ret == 0 && aresid == 0) { 8398 ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ, 8399 rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL, 8400 &aresid, p); 8401 xfer = PNFSDS_COPYSIZ - aresid; 8402 if (ret == 0 && xfer > 0) { 8403 rdpos += xfer; 8404 /* 8405 * Skip the write for holes, except for the 8406 * last block. 8407 */ 8408 if (xfer < PNFSDS_COPYSIZ || rdpos == 8409 va.va_size || NFSBCMP(dat, 8410 nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0) 8411 ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer, 8412 wrpos, UIO_SYSSPACE, IO_NODELOCKED, 8413 cred, NULL, NULL, p); 8414 if (ret == 0) 8415 wrpos += xfer; 8416 } 8417 } 8418 8419 /* If there is an ACL and the copy succeeded, set the ACL. */ 8420 if (ret == 0 && retacl == 0) { 8421 ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p); 8422 /* 8423 * Don't consider these as errors, since VOP_GETACL() 8424 * can return an ACL when they are not actually 8425 * supported. For example, for UFS, VOP_GETACL() 8426 * will return a trivial ACL based on the uid/gid/mode 8427 * when there is no ACL on the file. 8428 * This case should be recognized as a trivial ACL 8429 * by UFS's VOP_SETACL() and succeed, but... 8430 */ 8431 if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM) 8432 ret = 0; 8433 } 8434 8435 if (ret == 0) 8436 ret = VOP_FSYNC(tvp, MNT_WAIT, p); 8437 8438 /* Set the DS data file's modify time that of the MDS file. */ 8439 if (ret == 0) 8440 ret = VOP_GETATTR(vp, &va, cred); 8441 if (ret == 0) { 8442 mtime = va.va_mtime; 8443 VATTR_NULL(&va); 8444 va.va_mtime = mtime; 8445 ret = VOP_SETATTR(tvp, &va, cred); 8446 } 8447 8448 vput(tvp); 8449 acl_free(aclp); 8450 free(dat, M_TEMP); 8451 } 8452 if (tvmp != NULL) 8453 vn_finished_write(tvmp); 8454 8455 /* Update the extended attributes for the newly created DS file. */ 8456 if (ret == 0) 8457 ret = vn_extattr_set(vp, IO_NODELOCKED, 8458 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 8459 sizeof(*wpf) * mirrorcnt, (char *)wpf, p); 8460 if (mp != NULL) 8461 vn_finished_write(mp); 8462 8463 /* Get rid of the dontlist entry, so that Layouts can be issued. */ 8464 NFSDDONTLISTLOCK(); 8465 LIST_REMOVE(mrp, nfsmr_list); 8466 NFSDDONTLISTUNLOCK(); 8467 free(mrp, M_NFSDSTATE); 8468 return (ret); 8469 } 8470 8471 /* 8472 * Create a data storage file on the recovered DS. 8473 */ 8474 static int 8475 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf, 8476 vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p, 8477 vnode_t *tvpp) 8478 { 8479 struct vattr va, nva; 8480 int error; 8481 8482 /* Make data file name based on FH. */ 8483 error = VOP_GETATTR(vp, &va, cred); 8484 if (error == 0) { 8485 /* Set the attributes for "vp" to Setattr the DS vp. */ 8486 VATTR_NULL(&nva); 8487 nva.va_uid = va.va_uid; 8488 nva.va_gid = va.va_gid; 8489 nva.va_mode = va.va_mode; 8490 nva.va_size = 0; 8491 VATTR_NULL(&va); 8492 va.va_type = VREG; 8493 va.va_mode = nva.va_mode; 8494 NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf); 8495 error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL, 8496 pf->dsf_filename, cred, p, tvpp); 8497 } 8498 return (error); 8499 } 8500 8501 /* 8502 * Look up the MDS file shared locked, and then get the extended attribute 8503 * to find the extant DS file to be copied to the new mirror. 8504 * If successful, *vpp is set to the MDS file's vp and *nvpp is 8505 * set to a DS data file for the MDS file, both exclusively locked. 8506 * The "buf" argument has the pnfsdsfile structure from the MDS file 8507 * in it and buflen is set to its length. 8508 */ 8509 int 8510 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf, 8511 int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp, 8512 struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp, 8513 struct nfsdevice **fdsp) 8514 { 8515 struct nameidata nd; 8516 struct vnode *vp, *curvp; 8517 struct pnfsdsfile *pf; 8518 struct nfsmount *nmp, *curnmp; 8519 int dsdir, error, mirrorcnt, ippos; 8520 8521 vp = NULL; 8522 curvp = NULL; 8523 curnmp = NULL; 8524 *dsp = NULL; 8525 *fdsp = NULL; 8526 if (dspathp == NULL && curdspathp != NULL) 8527 return (EPERM); 8528 8529 /* 8530 * Look up the MDS file shared locked. The lock will be upgraded 8531 * to an exclusive lock after any rw layouts have been returned. 8532 */ 8533 NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp); 8534 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, 8535 mdspathp, p); 8536 error = namei(&nd); 8537 NFSD_DEBUG(4, "lookup=%d\n", error); 8538 if (error != 0) 8539 return (error); 8540 if (nd.ni_vp->v_type != VREG) { 8541 vput(nd.ni_vp); 8542 NFSD_DEBUG(4, "mdspath not reg\n"); 8543 return (EISDIR); 8544 } 8545 vp = nd.ni_vp; 8546 8547 if (curdspathp != NULL) { 8548 /* 8549 * Look up the current DS path and find the nfsdev structure for 8550 * it. 8551 */ 8552 NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp); 8553 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 8554 UIO_SYSSPACE, curdspathp, p); 8555 error = namei(&nd); 8556 NFSD_DEBUG(4, "ds lookup=%d\n", error); 8557 if (error != 0) { 8558 vput(vp); 8559 return (error); 8560 } 8561 if (nd.ni_vp->v_type != VDIR) { 8562 vput(nd.ni_vp); 8563 vput(vp); 8564 NFSD_DEBUG(4, "curdspath not dir\n"); 8565 return (ENOTDIR); 8566 } 8567 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { 8568 vput(nd.ni_vp); 8569 vput(vp); 8570 NFSD_DEBUG(4, "curdspath not an NFS mount\n"); 8571 return (ENXIO); 8572 } 8573 curnmp = VFSTONFS(nd.ni_vp->v_mount); 8574 8575 /* Search the nfsdev list for a match. */ 8576 NFSDDSLOCK(); 8577 *fdsp = nfsv4_findmirror(curnmp); 8578 NFSDDSUNLOCK(); 8579 if (*fdsp == NULL) 8580 curnmp = NULL; 8581 if (curnmp == NULL) { 8582 vput(nd.ni_vp); 8583 vput(vp); 8584 NFSD_DEBUG(4, "mdscopymr: no current ds\n"); 8585 return (ENXIO); 8586 } 8587 curvp = nd.ni_vp; 8588 } 8589 8590 if (dspathp != NULL) { 8591 /* Look up the nfsdev path and find the nfsdev structure. */ 8592 NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp); 8593 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 8594 UIO_SYSSPACE, dspathp, p); 8595 error = namei(&nd); 8596 NFSD_DEBUG(4, "ds lookup=%d\n", error); 8597 if (error != 0) { 8598 vput(vp); 8599 if (curvp != NULL) 8600 vput(curvp); 8601 return (error); 8602 } 8603 if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) { 8604 vput(nd.ni_vp); 8605 vput(vp); 8606 if (curvp != NULL) 8607 vput(curvp); 8608 NFSD_DEBUG(4, "dspath not dir\n"); 8609 if (nd.ni_vp == curvp) 8610 return (EPERM); 8611 return (ENOTDIR); 8612 } 8613 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { 8614 vput(nd.ni_vp); 8615 vput(vp); 8616 if (curvp != NULL) 8617 vput(curvp); 8618 NFSD_DEBUG(4, "dspath not an NFS mount\n"); 8619 return (ENXIO); 8620 } 8621 nmp = VFSTONFS(nd.ni_vp->v_mount); 8622 8623 /* 8624 * Search the nfsdevice list for a match. If curnmp == NULL, 8625 * this is a recovery and there must be a mirror. 8626 */ 8627 NFSDDSLOCK(); 8628 if (curnmp == NULL) 8629 *dsp = nfsrv_findmirroredds(nmp); 8630 else 8631 *dsp = nfsv4_findmirror(nmp); 8632 NFSDDSUNLOCK(); 8633 if (*dsp == NULL) { 8634 vput(nd.ni_vp); 8635 vput(vp); 8636 if (curvp != NULL) 8637 vput(curvp); 8638 NFSD_DEBUG(4, "mdscopymr: no ds\n"); 8639 return (ENXIO); 8640 } 8641 } else { 8642 nd.ni_vp = NULL; 8643 nmp = NULL; 8644 } 8645 8646 /* 8647 * Get a vp for an available DS data file using the extended 8648 * attribute on the MDS file. 8649 * If there is a valid entry for the new DS in the extended attribute 8650 * on the MDS file (as checked via the nmp argument), 8651 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur. 8652 */ 8653 error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p, 8654 NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir); 8655 if (curvp != NULL) 8656 vput(curvp); 8657 if (nd.ni_vp == NULL) { 8658 if (error == 0 && nmp != NULL) { 8659 /* Search the nfsdev list for a match. */ 8660 NFSDDSLOCK(); 8661 *dsp = nfsrv_findmirroredds(nmp); 8662 NFSDDSUNLOCK(); 8663 } 8664 if (error == 0 && (nmp == NULL || *dsp == NULL)) { 8665 if (nvpp != NULL && *nvpp != NULL) { 8666 vput(*nvpp); 8667 *nvpp = NULL; 8668 } 8669 error = ENXIO; 8670 } 8671 } else 8672 vput(nd.ni_vp); 8673 8674 /* 8675 * When dspathp != NULL and curdspathp == NULL, this is a recovery 8676 * and is only allowed if there is a 0.0.0.0 IP address entry. 8677 * When curdspathp != NULL, the ippos will be set to that entry. 8678 */ 8679 if (error == 0 && dspathp != NULL && ippos == -1) { 8680 if (nvpp != NULL && *nvpp != NULL) { 8681 vput(*nvpp); 8682 *nvpp = NULL; 8683 } 8684 error = ENXIO; 8685 } 8686 if (error == 0) { 8687 *vpp = vp; 8688 8689 pf = (struct pnfsdsfile *)buf; 8690 if (ippos == -1) { 8691 /* If no zeroip pnfsdsfile, add one. */ 8692 ippos = *buflenp / sizeof(*pf); 8693 *buflenp += sizeof(*pf); 8694 pf += ippos; 8695 pf->dsf_dir = dsdir; 8696 strlcpy(pf->dsf_filename, fname, 8697 sizeof(pf->dsf_filename)); 8698 } else 8699 pf += ippos; 8700 *pfp = pf; 8701 } else 8702 vput(vp); 8703 return (error); 8704 } 8705 8706 /* 8707 * Search for a matching pnfsd mirror device structure, base on the nmp arg. 8708 * Return one if found, NULL otherwise. 8709 */ 8710 static struct nfsdevice * 8711 nfsrv_findmirroredds(struct nfsmount *nmp) 8712 { 8713 struct nfsdevice *ds, *fndds; 8714 int fndmirror; 8715 8716 mtx_assert(NFSDDSMUTEXPTR, MA_OWNED); 8717 /* 8718 * Search the DS server list for a match with nmp. 8719 * Remove the DS entry if found and there is a mirror. 8720 */ 8721 fndds = NULL; 8722 fndmirror = 0; 8723 if (nfsrv_devidcnt == 0) 8724 return (fndds); 8725 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 8726 if (ds->nfsdev_nmp == nmp) { 8727 NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n"); 8728 fndds = ds; 8729 break; 8730 } 8731 } 8732 if (fndds == NULL) 8733 return (fndds); 8734 if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0) 8735 fndmirror = 1; 8736 else if (fndds->nfsdev_mdsisset != 0) { 8737 /* For the fsid is set case, search for a mirror. */ 8738 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 8739 if (ds != fndds && ds->nfsdev_nmp != NULL && 8740 ds->nfsdev_mdsisset != 0 && 8741 fsidcmp(&ds->nfsdev_mdsfsid, 8742 &fndds->nfsdev_mdsfsid) == 0) { 8743 fndmirror = 1; 8744 break; 8745 } 8746 } 8747 } 8748 if (fndmirror == 0) { 8749 NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n"); 8750 return (NULL); 8751 } 8752 return (fndds); 8753 } 8754 8755