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 #ifndef APPLEKEXT 36 #include <sys/extattr.h> 37 #include <fs/nfs/nfsport.h> 38 39 struct nfsrv_stablefirst nfsrv_stablefirst; 40 int nfsrv_issuedelegs = 0; 41 int nfsrv_dolocallocks = 0; 42 struct nfsv4lock nfsv4rootfs_lock; 43 time_t nfsdev_time = 0; 44 int nfsrv_layouthashsize; 45 volatile int nfsrv_layoutcnt = 0; 46 47 extern int newnfs_numnfsd; 48 extern struct nfsstatsv1 nfsstatsv1; 49 extern int nfsrv_lease; 50 extern struct timeval nfsboottime; 51 extern u_int32_t newnfs_true, newnfs_false; 52 extern struct mtx nfsrv_dslock_mtx; 53 extern struct mtx nfsrv_recalllock_mtx; 54 extern struct mtx nfsrv_dontlistlock_mtx; 55 extern int nfsd_debuglevel; 56 extern u_int nfsrv_dsdirsize; 57 extern struct nfsdevicehead nfsrv_devidhead; 58 extern int nfsrv_doflexfile; 59 extern int nfsrv_maxpnfsmirror; 60 NFSV4ROOTLOCKMUTEX; 61 NFSSTATESPINLOCK; 62 extern struct nfsdontlisthead nfsrv_dontlisthead; 63 extern volatile int nfsrv_devidcnt; 64 extern struct nfslayouthead nfsrv_recalllisthead; 65 extern char *nfsrv_zeropnfsdat; 66 67 SYSCTL_DECL(_vfs_nfsd); 68 int nfsrv_statehashsize = NFSSTATEHASHSIZE; 69 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN, 70 &nfsrv_statehashsize, 0, 71 "Size of state hash table set via loader.conf"); 72 73 int nfsrv_clienthashsize = NFSCLIENTHASHSIZE; 74 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN, 75 &nfsrv_clienthashsize, 0, 76 "Size of client hash table set via loader.conf"); 77 78 int nfsrv_lockhashsize = NFSLOCKHASHSIZE; 79 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN, 80 &nfsrv_lockhashsize, 0, 81 "Size of file handle hash table set via loader.conf"); 82 83 int nfsrv_sessionhashsize = NFSSESSIONHASHSIZE; 84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN, 85 &nfsrv_sessionhashsize, 0, 86 "Size of session hash table set via loader.conf"); 87 88 int nfsrv_layouthighwater = NFSLAYOUTHIGHWATER; 89 SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN, 90 &nfsrv_layouthighwater, 0, 91 "High water mark for number of layouts set via loader.conf"); 92 93 static int nfsrv_v4statelimit = NFSRV_V4STATELIMIT; 94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN, 95 &nfsrv_v4statelimit, 0, 96 "High water limit for NFSv4 opens+locks+delegations"); 97 98 static int nfsrv_writedelegifpos = 0; 99 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW, 100 &nfsrv_writedelegifpos, 0, 101 "Issue a write delegation for read opens if possible"); 102 103 static int nfsrv_allowreadforwriteopen = 1; 104 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW, 105 &nfsrv_allowreadforwriteopen, 0, 106 "Allow Reads to be done with Write Access StateIDs"); 107 108 int nfsrv_pnfsatime = 0; 109 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW, 110 &nfsrv_pnfsatime, 0, 111 "For pNFS service, do Getattr ops to keep atime up-to-date"); 112 113 int nfsrv_flexlinuxhack = 0; 114 SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW, 115 &nfsrv_flexlinuxhack, 0, 116 "For Linux clients, hack around Flex File Layout bug"); 117 118 /* 119 * Hash lists for nfs V4. 120 */ 121 struct nfsclienthashhead *nfsclienthash; 122 struct nfslockhashhead *nfslockhash; 123 struct nfssessionhash *nfssessionhash; 124 struct nfslayouthash *nfslayouthash; 125 volatile int nfsrv_dontlistlen = 0; 126 #endif /* !APPLEKEXT */ 127 128 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0; 129 static time_t nfsrvboottime; 130 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0; 131 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER; 132 static int nfsrv_nogsscallback = 0; 133 static volatile int nfsrv_writedelegcnt = 0; 134 static int nfsrv_faildscnt; 135 136 /* local functions */ 137 static void nfsrv_dumpaclient(struct nfsclient *clp, 138 struct nfsd_dumpclients *dumpp); 139 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, 140 NFSPROC_T *p); 141 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, 142 NFSPROC_T *p); 143 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep, 144 NFSPROC_T *p); 145 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, 146 int cansleep, NFSPROC_T *p); 147 static void nfsrv_freenfslock(struct nfslock *lop); 148 static void nfsrv_freenfslockfile(struct nfslockfile *lfp); 149 static void nfsrv_freedeleg(struct nfsstate *); 150 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, 151 u_int32_t flags, struct nfsstate **stpp); 152 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp, 153 struct nfsstate **stpp); 154 static int nfsrv_getlockfh(vnode_t vp, u_short flags, 155 struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p); 156 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp, 157 struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit); 158 static void nfsrv_insertlock(struct nfslock *new_lop, 159 struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp); 160 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp, 161 struct nfslock **other_lopp, struct nfslockfile *lfp); 162 static int nfsrv_getipnumber(u_char *cp); 163 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags, 164 nfsv4stateid_t *stateidp, int specialid); 165 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp, 166 u_int32_t flags); 167 static int nfsrv_docallback(struct nfsclient *clp, int procnum, 168 nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp, 169 struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p); 170 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp, 171 uint32_t callback, int op, const char *optag, struct nfsdsession **sepp); 172 static u_int32_t nfsrv_nextclientindex(void); 173 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp); 174 static void nfsrv_markstable(struct nfsclient *clp); 175 static void nfsrv_markreclaim(struct nfsclient *clp); 176 static int nfsrv_checkstable(struct nfsclient *clp); 177 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct 178 vnode *vp, NFSPROC_T *p); 179 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, 180 NFSPROC_T *p, vnode_t vp); 181 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp, 182 struct nfsclient *clp, int *haslockp, NFSPROC_T *p); 183 static int nfsrv_notsamecredname(struct nfsrv_descript *nd, 184 struct nfsclient *clp); 185 static time_t nfsrv_leaseexpiry(void); 186 static void nfsrv_delaydelegtimeout(struct nfsstate *stp); 187 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid, 188 struct nfsstate *stp, struct nfsrvcache *op); 189 static int nfsrv_nootherstate(struct nfsstate *stp); 190 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags, 191 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p); 192 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, 193 uint64_t init_first, uint64_t init_end, NFSPROC_T *p); 194 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, 195 int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp, 196 NFSPROC_T *p); 197 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, 198 NFSPROC_T *p); 199 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, 200 uint64_t first, uint64_t end); 201 static void nfsrv_locklf(struct nfslockfile *lfp); 202 static void nfsrv_unlocklf(struct nfslockfile *lfp); 203 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid); 204 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid); 205 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp, 206 int dont_replycache, struct nfsdsession **sepp); 207 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp); 208 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp, 209 nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p); 210 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp); 211 static void nfsrv_freelayoutlist(nfsquad_t clientid); 212 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, 213 int iomode); 214 static void nfsrv_freealllayouts(void); 215 static void nfsrv_freedevid(struct nfsdevice *ds); 216 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p, 217 struct nfsdevice **dsp); 218 static int nfsrv_delds(char *devid, NFSPROC_T *p); 219 static void nfsrv_deleteds(struct nfsdevice *fndds); 220 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost); 221 static void nfsrv_freealldevids(void); 222 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, 223 int maxcnt, NFSPROC_T *p); 224 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, 225 fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype, 226 NFSPROC_T *p); 227 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype, 228 NFSPROC_T *, struct nfslayout **lypp); 229 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt); 230 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, 231 fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs); 232 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, 233 int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs); 234 static int nfsrv_dontlayout(fhandle_t *fhp); 235 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf, 236 vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p, 237 vnode_t *tvpp); 238 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp); 239 240 /* 241 * Scan the client list for a match and either return the current one, 242 * create a new entry or return an error. 243 * If returning a non-error, the clp structure must either be linked into 244 * the client list or free'd. 245 */ 246 APPLESTATIC int 247 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp, 248 nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p) 249 { 250 struct nfsclient *clp = NULL, *new_clp = *new_clpp; 251 int i, error = 0, ret; 252 struct nfsstate *stp, *tstp; 253 #ifdef INET 254 struct sockaddr_in *sin, *rin; 255 #endif 256 #ifdef INET6 257 struct sockaddr_in6 *sin6, *rin6; 258 #endif 259 struct nfsdsession *sep, *nsep; 260 int zapit = 0, gotit, hasstate = 0, igotlock; 261 static u_int64_t confirm_index = 0; 262 263 /* 264 * Check for state resource limit exceeded. 265 */ 266 if (nfsrv_openpluslock > nfsrv_v4statelimit) { 267 error = NFSERR_RESOURCE; 268 goto out; 269 } 270 271 if (nfsrv_issuedelegs == 0 || 272 ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0)) 273 /* 274 * Don't do callbacks when delegations are disabled or 275 * for AUTH_GSS unless enabled via nfsrv_nogsscallback. 276 * If establishing a callback connection is attempted 277 * when a firewall is blocking the callback path, the 278 * server may wait too long for the connect attempt to 279 * succeed during the Open. Some clients, such as Linux, 280 * may timeout and give up on the Open before the server 281 * replies. Also, since AUTH_GSS callbacks are not 282 * yet interoperability tested, they might cause the 283 * server to crap out, if they get past the Init call to 284 * the client. 285 */ 286 new_clp->lc_program = 0; 287 288 /* Lock out other nfsd threads */ 289 NFSLOCKV4ROOTMUTEX(); 290 nfsv4_relref(&nfsv4rootfs_lock); 291 do { 292 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 293 NFSV4ROOTLOCKMUTEXPTR, NULL); 294 } while (!igotlock); 295 NFSUNLOCKV4ROOTMUTEX(); 296 297 /* 298 * Search for a match in the client list. 299 */ 300 gotit = i = 0; 301 while (i < nfsrv_clienthashsize && !gotit) { 302 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) { 303 if (new_clp->lc_idlen == clp->lc_idlen && 304 !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) { 305 gotit = 1; 306 break; 307 } 308 } 309 if (gotit == 0) 310 i++; 311 } 312 if (!gotit || 313 (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) { 314 if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) { 315 /* 316 * For NFSv4.1, if confirmp->lval[1] is non-zero, the 317 * client is trying to update a confirmed clientid. 318 */ 319 NFSLOCKV4ROOTMUTEX(); 320 nfsv4_unlock(&nfsv4rootfs_lock, 1); 321 NFSUNLOCKV4ROOTMUTEX(); 322 confirmp->lval[1] = 0; 323 error = NFSERR_NOENT; 324 goto out; 325 } 326 /* 327 * Get rid of the old one. 328 */ 329 if (i != nfsrv_clienthashsize) { 330 LIST_REMOVE(clp, lc_hash); 331 nfsrv_cleanclient(clp, p); 332 nfsrv_freedeleglist(&clp->lc_deleg); 333 nfsrv_freedeleglist(&clp->lc_olddeleg); 334 zapit = 1; 335 } 336 /* 337 * Add it after assigning a client id to it. 338 */ 339 new_clp->lc_flags |= LCL_NEEDSCONFIRM; 340 if ((nd->nd_flag & ND_NFSV41) != 0) 341 new_clp->lc_confirm.lval[0] = confirmp->lval[0] = 342 ++confirm_index; 343 else 344 confirmp->qval = new_clp->lc_confirm.qval = 345 ++confirm_index; 346 clientidp->lval[0] = new_clp->lc_clientid.lval[0] = 347 (u_int32_t)nfsrvboottime; 348 clientidp->lval[1] = new_clp->lc_clientid.lval[1] = 349 nfsrv_nextclientindex(); 350 new_clp->lc_stateindex = 0; 351 new_clp->lc_statemaxindex = 0; 352 new_clp->lc_cbref = 0; 353 new_clp->lc_expiry = nfsrv_leaseexpiry(); 354 LIST_INIT(&new_clp->lc_open); 355 LIST_INIT(&new_clp->lc_deleg); 356 LIST_INIT(&new_clp->lc_olddeleg); 357 LIST_INIT(&new_clp->lc_session); 358 for (i = 0; i < nfsrv_statehashsize; i++) 359 LIST_INIT(&new_clp->lc_stateid[i]); 360 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, 361 lc_hash); 362 nfsstatsv1.srvclients++; 363 nfsrv_openpluslock++; 364 nfsrv_clients++; 365 NFSLOCKV4ROOTMUTEX(); 366 nfsv4_unlock(&nfsv4rootfs_lock, 1); 367 NFSUNLOCKV4ROOTMUTEX(); 368 if (zapit) 369 nfsrv_zapclient(clp, p); 370 *new_clpp = NULL; 371 goto out; 372 } 373 374 /* 375 * Now, handle the cases where the id is already issued. 376 */ 377 if (nfsrv_notsamecredname(nd, clp)) { 378 /* 379 * Check to see if there is expired state that should go away. 380 */ 381 if (clp->lc_expiry < NFSD_MONOSEC && 382 (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) { 383 nfsrv_cleanclient(clp, p); 384 nfsrv_freedeleglist(&clp->lc_deleg); 385 } 386 387 /* 388 * If there is outstanding state, then reply NFSERR_CLIDINUSE per 389 * RFC3530 Sec. 8.1.2 last para. 390 */ 391 if (!LIST_EMPTY(&clp->lc_deleg)) { 392 hasstate = 1; 393 } else if (LIST_EMPTY(&clp->lc_open)) { 394 hasstate = 0; 395 } else { 396 hasstate = 0; 397 /* Look for an Open on the OpenOwner */ 398 LIST_FOREACH(stp, &clp->lc_open, ls_list) { 399 if (!LIST_EMPTY(&stp->ls_open)) { 400 hasstate = 1; 401 break; 402 } 403 } 404 } 405 if (hasstate) { 406 /* 407 * If the uid doesn't match, return NFSERR_CLIDINUSE after 408 * filling out the correct ipaddr and portnum. 409 */ 410 switch (clp->lc_req.nr_nam->sa_family) { 411 #ifdef INET 412 case AF_INET: 413 sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam; 414 rin = (struct sockaddr_in *)clp->lc_req.nr_nam; 415 sin->sin_addr.s_addr = rin->sin_addr.s_addr; 416 sin->sin_port = rin->sin_port; 417 break; 418 #endif 419 #ifdef INET6 420 case AF_INET6: 421 sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam; 422 rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; 423 sin6->sin6_addr = rin6->sin6_addr; 424 sin6->sin6_port = rin6->sin6_port; 425 break; 426 #endif 427 } 428 NFSLOCKV4ROOTMUTEX(); 429 nfsv4_unlock(&nfsv4rootfs_lock, 1); 430 NFSUNLOCKV4ROOTMUTEX(); 431 error = NFSERR_CLIDINUSE; 432 goto out; 433 } 434 } 435 436 if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) { 437 /* 438 * If the verifier has changed, the client has rebooted 439 * and a new client id is issued. The old state info 440 * can be thrown away once the SETCLIENTID_CONFIRM occurs. 441 */ 442 LIST_REMOVE(clp, lc_hash); 443 444 /* Get rid of all sessions on this clientid. */ 445 LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) { 446 ret = nfsrv_freesession(sep, NULL); 447 if (ret != 0) 448 printf("nfsrv_setclient: verifier changed free" 449 " session failed=%d\n", ret); 450 } 451 452 new_clp->lc_flags |= LCL_NEEDSCONFIRM; 453 if ((nd->nd_flag & ND_NFSV41) != 0) 454 new_clp->lc_confirm.lval[0] = confirmp->lval[0] = 455 ++confirm_index; 456 else 457 confirmp->qval = new_clp->lc_confirm.qval = 458 ++confirm_index; 459 clientidp->lval[0] = new_clp->lc_clientid.lval[0] = 460 nfsrvboottime; 461 clientidp->lval[1] = new_clp->lc_clientid.lval[1] = 462 nfsrv_nextclientindex(); 463 new_clp->lc_stateindex = 0; 464 new_clp->lc_statemaxindex = 0; 465 new_clp->lc_cbref = 0; 466 new_clp->lc_expiry = nfsrv_leaseexpiry(); 467 468 /* 469 * Save the state until confirmed. 470 */ 471 LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list); 472 LIST_FOREACH(tstp, &new_clp->lc_open, ls_list) 473 tstp->ls_clp = new_clp; 474 LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list); 475 LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list) 476 tstp->ls_clp = new_clp; 477 LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, 478 ls_list); 479 LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list) 480 tstp->ls_clp = new_clp; 481 for (i = 0; i < nfsrv_statehashsize; i++) { 482 LIST_NEWHEAD(&new_clp->lc_stateid[i], 483 &clp->lc_stateid[i], ls_hash); 484 LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash) 485 tstp->ls_clp = new_clp; 486 } 487 LIST_INIT(&new_clp->lc_session); 488 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, 489 lc_hash); 490 nfsstatsv1.srvclients++; 491 nfsrv_openpluslock++; 492 nfsrv_clients++; 493 NFSLOCKV4ROOTMUTEX(); 494 nfsv4_unlock(&nfsv4rootfs_lock, 1); 495 NFSUNLOCKV4ROOTMUTEX(); 496 497 /* 498 * Must wait until any outstanding callback on the old clp 499 * completes. 500 */ 501 NFSLOCKSTATE(); 502 while (clp->lc_cbref) { 503 clp->lc_flags |= LCL_WAKEUPWANTED; 504 (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, 505 "nfsd clp", 10 * hz); 506 } 507 NFSUNLOCKSTATE(); 508 nfsrv_zapclient(clp, p); 509 *new_clpp = NULL; 510 goto out; 511 } 512 513 /* For NFSv4.1, mark that we found a confirmed clientid. */ 514 if ((nd->nd_flag & ND_NFSV41) != 0) { 515 clientidp->lval[0] = clp->lc_clientid.lval[0]; 516 clientidp->lval[1] = clp->lc_clientid.lval[1]; 517 confirmp->lval[0] = 0; /* Ignored by client */ 518 confirmp->lval[1] = 1; 519 } else { 520 /* 521 * id and verifier match, so update the net address info 522 * and get rid of any existing callback authentication 523 * handle, so a new one will be acquired. 524 */ 525 LIST_REMOVE(clp, lc_hash); 526 new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN); 527 new_clp->lc_expiry = nfsrv_leaseexpiry(); 528 confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index; 529 clientidp->lval[0] = new_clp->lc_clientid.lval[0] = 530 clp->lc_clientid.lval[0]; 531 clientidp->lval[1] = new_clp->lc_clientid.lval[1] = 532 clp->lc_clientid.lval[1]; 533 new_clp->lc_delegtime = clp->lc_delegtime; 534 new_clp->lc_stateindex = clp->lc_stateindex; 535 new_clp->lc_statemaxindex = clp->lc_statemaxindex; 536 new_clp->lc_cbref = 0; 537 LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list); 538 LIST_FOREACH(tstp, &new_clp->lc_open, ls_list) 539 tstp->ls_clp = new_clp; 540 LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list); 541 LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list) 542 tstp->ls_clp = new_clp; 543 LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list); 544 LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list) 545 tstp->ls_clp = new_clp; 546 for (i = 0; i < nfsrv_statehashsize; i++) { 547 LIST_NEWHEAD(&new_clp->lc_stateid[i], 548 &clp->lc_stateid[i], ls_hash); 549 LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash) 550 tstp->ls_clp = new_clp; 551 } 552 LIST_INIT(&new_clp->lc_session); 553 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp, 554 lc_hash); 555 nfsstatsv1.srvclients++; 556 nfsrv_openpluslock++; 557 nfsrv_clients++; 558 } 559 NFSLOCKV4ROOTMUTEX(); 560 nfsv4_unlock(&nfsv4rootfs_lock, 1); 561 NFSUNLOCKV4ROOTMUTEX(); 562 563 if ((nd->nd_flag & ND_NFSV41) == 0) { 564 /* 565 * Must wait until any outstanding callback on the old clp 566 * completes. 567 */ 568 NFSLOCKSTATE(); 569 while (clp->lc_cbref) { 570 clp->lc_flags |= LCL_WAKEUPWANTED; 571 (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, 572 "nfsdclp", 10 * hz); 573 } 574 NFSUNLOCKSTATE(); 575 nfsrv_zapclient(clp, p); 576 *new_clpp = NULL; 577 } 578 579 out: 580 NFSEXITCODE2(error, nd); 581 return (error); 582 } 583 584 /* 585 * Check to see if the client id exists and optionally confirm it. 586 */ 587 APPLESTATIC int 588 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp, 589 struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram, 590 struct nfsrv_descript *nd, NFSPROC_T *p) 591 { 592 struct nfsclient *clp; 593 struct nfsstate *stp; 594 int i; 595 struct nfsclienthashhead *hp; 596 int error = 0, igotlock, doneok; 597 struct nfssessionhash *shp; 598 struct nfsdsession *sep; 599 uint64_t sessid[2]; 600 static uint64_t next_sess = 0; 601 602 if (clpp) 603 *clpp = NULL; 604 if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 || 605 opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) { 606 error = NFSERR_STALECLIENTID; 607 goto out; 608 } 609 610 /* 611 * If called with opflags == CLOPS_RENEW, the State Lock is 612 * already held. Otherwise, we need to get either that or, 613 * for the case of Confirm, lock out the nfsd threads. 614 */ 615 if (opflags & CLOPS_CONFIRM) { 616 NFSLOCKV4ROOTMUTEX(); 617 nfsv4_relref(&nfsv4rootfs_lock); 618 do { 619 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 620 NFSV4ROOTLOCKMUTEXPTR, NULL); 621 } while (!igotlock); 622 /* 623 * Create a new sessionid here, since we need to do it where 624 * there is a mutex held to serialize update of next_sess. 625 */ 626 if ((nd->nd_flag & ND_NFSV41) != 0) { 627 sessid[0] = ++next_sess; 628 sessid[1] = clientid.qval; 629 } 630 NFSUNLOCKV4ROOTMUTEX(); 631 } else if (opflags != CLOPS_RENEW) { 632 NFSLOCKSTATE(); 633 } 634 635 /* For NFSv4.1, the clp is acquired from the associated session. */ 636 if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 && 637 opflags == CLOPS_RENEW) { 638 clp = NULL; 639 if ((nd->nd_flag & ND_HASSEQUENCE) != 0) { 640 shp = NFSSESSIONHASH(nd->nd_sessionid); 641 NFSLOCKSESSION(shp); 642 sep = nfsrv_findsession(nd->nd_sessionid); 643 if (sep != NULL) 644 clp = sep->sess_clp; 645 NFSUNLOCKSESSION(shp); 646 } 647 } else { 648 hp = NFSCLIENTHASH(clientid); 649 LIST_FOREACH(clp, hp, lc_hash) { 650 if (clp->lc_clientid.lval[1] == clientid.lval[1]) 651 break; 652 } 653 } 654 if (clp == NULL) { 655 if (opflags & CLOPS_CONFIRM) 656 error = NFSERR_STALECLIENTID; 657 else 658 error = NFSERR_EXPIRED; 659 } else if (clp->lc_flags & LCL_ADMINREVOKED) { 660 /* 661 * If marked admin revoked, just return the error. 662 */ 663 error = NFSERR_ADMINREVOKED; 664 } 665 if (error) { 666 if (opflags & CLOPS_CONFIRM) { 667 NFSLOCKV4ROOTMUTEX(); 668 nfsv4_unlock(&nfsv4rootfs_lock, 1); 669 NFSUNLOCKV4ROOTMUTEX(); 670 } else if (opflags != CLOPS_RENEW) { 671 NFSUNLOCKSTATE(); 672 } 673 goto out; 674 } 675 676 /* 677 * Perform any operations specified by the opflags. 678 */ 679 if (opflags & CLOPS_CONFIRM) { 680 if (((nd->nd_flag & ND_NFSV41) != 0 && 681 clp->lc_confirm.lval[0] != confirm.lval[0]) || 682 ((nd->nd_flag & ND_NFSV41) == 0 && 683 clp->lc_confirm.qval != confirm.qval)) 684 error = NFSERR_STALECLIENTID; 685 else if (nfsrv_notsamecredname(nd, clp)) 686 error = NFSERR_CLIDINUSE; 687 688 if (!error) { 689 if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) == 690 LCL_NEEDSCONFIRM) { 691 /* 692 * Hang onto the delegations (as old delegations) 693 * for an Open with CLAIM_DELEGATE_PREV unless in 694 * grace, but get rid of the rest of the state. 695 */ 696 nfsrv_cleanclient(clp, p); 697 nfsrv_freedeleglist(&clp->lc_olddeleg); 698 if (nfsrv_checkgrace(nd, clp, 0)) { 699 /* In grace, so just delete delegations */ 700 nfsrv_freedeleglist(&clp->lc_deleg); 701 } else { 702 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) 703 stp->ls_flags |= NFSLCK_OLDDELEG; 704 clp->lc_delegtime = NFSD_MONOSEC + 705 nfsrv_lease + NFSRV_LEASEDELTA; 706 LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg, 707 ls_list); 708 } 709 if ((nd->nd_flag & ND_NFSV41) != 0) 710 clp->lc_program = cbprogram; 711 } 712 clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN); 713 if (clp->lc_program) 714 clp->lc_flags |= LCL_NEEDSCBNULL; 715 /* For NFSv4.1, link the session onto the client. */ 716 if (nsep != NULL) { 717 /* Hold a reference on the xprt for a backchannel. */ 718 if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) 719 != 0) { 720 if (clp->lc_req.nr_client == NULL) 721 clp->lc_req.nr_client = (struct __rpc_client *) 722 clnt_bck_create(nd->nd_xprt->xp_socket, 723 cbprogram, NFSV4_CBVERS); 724 if (clp->lc_req.nr_client != NULL) { 725 SVC_ACQUIRE(nd->nd_xprt); 726 nd->nd_xprt->xp_p2 = 727 clp->lc_req.nr_client->cl_private; 728 /* Disable idle timeout. */ 729 nd->nd_xprt->xp_idletimeout = 0; 730 nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt; 731 } else 732 nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN; 733 } 734 NFSBCOPY(sessid, nsep->sess_sessionid, 735 NFSX_V4SESSIONID); 736 NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid, 737 NFSX_V4SESSIONID); 738 shp = NFSSESSIONHASH(nsep->sess_sessionid); 739 NFSLOCKSTATE(); 740 NFSLOCKSESSION(shp); 741 LIST_INSERT_HEAD(&shp->list, nsep, sess_hash); 742 LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list); 743 nsep->sess_clp = clp; 744 NFSUNLOCKSESSION(shp); 745 NFSUNLOCKSTATE(); 746 } 747 } 748 } else if (clp->lc_flags & LCL_NEEDSCONFIRM) { 749 error = NFSERR_EXPIRED; 750 } 751 752 /* 753 * If called by the Renew Op, we must check the principal. 754 */ 755 if (!error && (opflags & CLOPS_RENEWOP)) { 756 if (nfsrv_notsamecredname(nd, clp)) { 757 doneok = 0; 758 for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) { 759 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) { 760 if ((stp->ls_flags & NFSLCK_OPEN) && 761 stp->ls_uid == nd->nd_cred->cr_uid) { 762 doneok = 1; 763 break; 764 } 765 } 766 } 767 if (!doneok) 768 error = NFSERR_ACCES; 769 } 770 if (!error && (clp->lc_flags & LCL_CBDOWN)) 771 error = NFSERR_CBPATHDOWN; 772 } 773 if ((!error || error == NFSERR_CBPATHDOWN) && 774 (opflags & CLOPS_RENEW)) { 775 clp->lc_expiry = nfsrv_leaseexpiry(); 776 } 777 if (opflags & CLOPS_CONFIRM) { 778 NFSLOCKV4ROOTMUTEX(); 779 nfsv4_unlock(&nfsv4rootfs_lock, 1); 780 NFSUNLOCKV4ROOTMUTEX(); 781 } else if (opflags != CLOPS_RENEW) { 782 NFSUNLOCKSTATE(); 783 } 784 if (clpp) 785 *clpp = clp; 786 787 out: 788 NFSEXITCODE2(error, nd); 789 return (error); 790 } 791 792 /* 793 * Perform the NFSv4.1 destroy clientid. 794 */ 795 int 796 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p) 797 { 798 struct nfsclient *clp; 799 struct nfsclienthashhead *hp; 800 int error = 0, i, igotlock; 801 802 if (nfsrvboottime != clientid.lval[0]) { 803 error = NFSERR_STALECLIENTID; 804 goto out; 805 } 806 807 /* Lock out other nfsd threads */ 808 NFSLOCKV4ROOTMUTEX(); 809 nfsv4_relref(&nfsv4rootfs_lock); 810 do { 811 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 812 NFSV4ROOTLOCKMUTEXPTR, NULL); 813 } while (igotlock == 0); 814 NFSUNLOCKV4ROOTMUTEX(); 815 816 hp = NFSCLIENTHASH(clientid); 817 LIST_FOREACH(clp, hp, lc_hash) { 818 if (clp->lc_clientid.lval[1] == clientid.lval[1]) 819 break; 820 } 821 if (clp == NULL) { 822 NFSLOCKV4ROOTMUTEX(); 823 nfsv4_unlock(&nfsv4rootfs_lock, 1); 824 NFSUNLOCKV4ROOTMUTEX(); 825 /* Just return ok, since it is gone. */ 826 goto out; 827 } 828 829 /* 830 * Free up all layouts on the clientid. Should the client return the 831 * layouts? 832 */ 833 nfsrv_freelayoutlist(clientid); 834 835 /* Scan for state on the clientid. */ 836 for (i = 0; i < nfsrv_statehashsize; i++) 837 if (!LIST_EMPTY(&clp->lc_stateid[i])) { 838 NFSLOCKV4ROOTMUTEX(); 839 nfsv4_unlock(&nfsv4rootfs_lock, 1); 840 NFSUNLOCKV4ROOTMUTEX(); 841 error = NFSERR_CLIENTIDBUSY; 842 goto out; 843 } 844 if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) { 845 NFSLOCKV4ROOTMUTEX(); 846 nfsv4_unlock(&nfsv4rootfs_lock, 1); 847 NFSUNLOCKV4ROOTMUTEX(); 848 error = NFSERR_CLIENTIDBUSY; 849 goto out; 850 } 851 852 /* Destroy the clientid and return ok. */ 853 nfsrv_cleanclient(clp, p); 854 nfsrv_freedeleglist(&clp->lc_deleg); 855 nfsrv_freedeleglist(&clp->lc_olddeleg); 856 LIST_REMOVE(clp, lc_hash); 857 NFSLOCKV4ROOTMUTEX(); 858 nfsv4_unlock(&nfsv4rootfs_lock, 1); 859 NFSUNLOCKV4ROOTMUTEX(); 860 nfsrv_zapclient(clp, p); 861 out: 862 NFSEXITCODE2(error, nd); 863 return (error); 864 } 865 866 /* 867 * Called from the new nfssvc syscall to admin revoke a clientid. 868 * Returns 0 for success, error otherwise. 869 */ 870 APPLESTATIC int 871 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p) 872 { 873 struct nfsclient *clp = NULL; 874 int i, error = 0; 875 int gotit, igotlock; 876 877 /* 878 * First, lock out the nfsd so that state won't change while the 879 * revocation record is being written to the stable storage restart 880 * file. 881 */ 882 NFSLOCKV4ROOTMUTEX(); 883 do { 884 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 885 NFSV4ROOTLOCKMUTEXPTR, NULL); 886 } while (!igotlock); 887 NFSUNLOCKV4ROOTMUTEX(); 888 889 /* 890 * Search for a match in the client list. 891 */ 892 gotit = i = 0; 893 while (i < nfsrv_clienthashsize && !gotit) { 894 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) { 895 if (revokep->nclid_idlen == clp->lc_idlen && 896 !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) { 897 gotit = 1; 898 break; 899 } 900 } 901 i++; 902 } 903 if (!gotit) { 904 NFSLOCKV4ROOTMUTEX(); 905 nfsv4_unlock(&nfsv4rootfs_lock, 0); 906 NFSUNLOCKV4ROOTMUTEX(); 907 error = EPERM; 908 goto out; 909 } 910 911 /* 912 * Now, write out the revocation record 913 */ 914 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); 915 nfsrv_backupstable(); 916 917 /* 918 * and clear out the state, marking the clientid revoked. 919 */ 920 clp->lc_flags &= ~LCL_CALLBACKSON; 921 clp->lc_flags |= LCL_ADMINREVOKED; 922 nfsrv_cleanclient(clp, p); 923 nfsrv_freedeleglist(&clp->lc_deleg); 924 nfsrv_freedeleglist(&clp->lc_olddeleg); 925 NFSLOCKV4ROOTMUTEX(); 926 nfsv4_unlock(&nfsv4rootfs_lock, 0); 927 NFSUNLOCKV4ROOTMUTEX(); 928 929 out: 930 NFSEXITCODE(error); 931 return (error); 932 } 933 934 /* 935 * Dump out stats for all clients. Called from nfssvc(2), that is used 936 * nfsstatsv1. 937 */ 938 APPLESTATIC void 939 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt) 940 { 941 struct nfsclient *clp; 942 int i = 0, cnt = 0; 943 944 /* 945 * First, get a reference on the nfsv4rootfs_lock so that an 946 * exclusive lock cannot be acquired while dumping the clients. 947 */ 948 NFSLOCKV4ROOTMUTEX(); 949 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); 950 NFSUNLOCKV4ROOTMUTEX(); 951 NFSLOCKSTATE(); 952 /* 953 * Rattle through the client lists until done. 954 */ 955 while (i < nfsrv_clienthashsize && cnt < maxcnt) { 956 clp = LIST_FIRST(&nfsclienthash[i]); 957 while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) { 958 nfsrv_dumpaclient(clp, &dumpp[cnt]); 959 cnt++; 960 clp = LIST_NEXT(clp, lc_hash); 961 } 962 i++; 963 } 964 if (cnt < maxcnt) 965 dumpp[cnt].ndcl_clid.nclid_idlen = 0; 966 NFSUNLOCKSTATE(); 967 NFSLOCKV4ROOTMUTEX(); 968 nfsv4_relref(&nfsv4rootfs_lock); 969 NFSUNLOCKV4ROOTMUTEX(); 970 } 971 972 /* 973 * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd. 974 */ 975 static void 976 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp) 977 { 978 struct nfsstate *stp, *openstp, *lckownstp; 979 struct nfslock *lop; 980 sa_family_t af; 981 #ifdef INET 982 struct sockaddr_in *rin; 983 #endif 984 #ifdef INET6 985 struct sockaddr_in6 *rin6; 986 #endif 987 988 dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0; 989 dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0; 990 dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0; 991 dumpp->ndcl_flags = clp->lc_flags; 992 dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen; 993 NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen); 994 af = clp->lc_req.nr_nam->sa_family; 995 dumpp->ndcl_addrfam = af; 996 switch (af) { 997 #ifdef INET 998 case AF_INET: 999 rin = (struct sockaddr_in *)clp->lc_req.nr_nam; 1000 dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr; 1001 break; 1002 #endif 1003 #ifdef INET6 1004 case AF_INET6: 1005 rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; 1006 dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr; 1007 break; 1008 #endif 1009 } 1010 1011 /* 1012 * Now, scan the state lists and total up the opens and locks. 1013 */ 1014 LIST_FOREACH(stp, &clp->lc_open, ls_list) { 1015 dumpp->ndcl_nopenowners++; 1016 LIST_FOREACH(openstp, &stp->ls_open, ls_list) { 1017 dumpp->ndcl_nopens++; 1018 LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) { 1019 dumpp->ndcl_nlockowners++; 1020 LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) { 1021 dumpp->ndcl_nlocks++; 1022 } 1023 } 1024 } 1025 } 1026 1027 /* 1028 * and the delegation lists. 1029 */ 1030 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) { 1031 dumpp->ndcl_ndelegs++; 1032 } 1033 LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) { 1034 dumpp->ndcl_nolddelegs++; 1035 } 1036 } 1037 1038 /* 1039 * Dump out lock stats for a file. 1040 */ 1041 APPLESTATIC void 1042 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt, 1043 NFSPROC_T *p) 1044 { 1045 struct nfsstate *stp; 1046 struct nfslock *lop; 1047 int cnt = 0; 1048 struct nfslockfile *lfp; 1049 sa_family_t af; 1050 #ifdef INET 1051 struct sockaddr_in *rin; 1052 #endif 1053 #ifdef INET6 1054 struct sockaddr_in6 *rin6; 1055 #endif 1056 int ret; 1057 fhandle_t nfh; 1058 1059 ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p); 1060 /* 1061 * First, get a reference on the nfsv4rootfs_lock so that an 1062 * exclusive lock on it cannot be acquired while dumping the locks. 1063 */ 1064 NFSLOCKV4ROOTMUTEX(); 1065 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); 1066 NFSUNLOCKV4ROOTMUTEX(); 1067 NFSLOCKSTATE(); 1068 if (!ret) 1069 ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0); 1070 if (ret) { 1071 ldumpp[0].ndlck_clid.nclid_idlen = 0; 1072 NFSUNLOCKSTATE(); 1073 NFSLOCKV4ROOTMUTEX(); 1074 nfsv4_relref(&nfsv4rootfs_lock); 1075 NFSUNLOCKV4ROOTMUTEX(); 1076 return; 1077 } 1078 1079 /* 1080 * For each open share on file, dump it out. 1081 */ 1082 stp = LIST_FIRST(&lfp->lf_open); 1083 while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) { 1084 ldumpp[cnt].ndlck_flags = stp->ls_flags; 1085 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid; 1086 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0]; 1087 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1]; 1088 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2]; 1089 ldumpp[cnt].ndlck_owner.nclid_idlen = 1090 stp->ls_openowner->ls_ownerlen; 1091 NFSBCOPY(stp->ls_openowner->ls_owner, 1092 ldumpp[cnt].ndlck_owner.nclid_id, 1093 stp->ls_openowner->ls_ownerlen); 1094 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen; 1095 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id, 1096 stp->ls_clp->lc_idlen); 1097 af = stp->ls_clp->lc_req.nr_nam->sa_family; 1098 ldumpp[cnt].ndlck_addrfam = af; 1099 switch (af) { 1100 #ifdef INET 1101 case AF_INET: 1102 rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; 1103 ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; 1104 break; 1105 #endif 1106 #ifdef INET6 1107 case AF_INET6: 1108 rin6 = (struct sockaddr_in6 *) 1109 stp->ls_clp->lc_req.nr_nam; 1110 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; 1111 break; 1112 #endif 1113 } 1114 stp = LIST_NEXT(stp, ls_file); 1115 cnt++; 1116 } 1117 1118 /* 1119 * and all locks. 1120 */ 1121 lop = LIST_FIRST(&lfp->lf_lock); 1122 while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) { 1123 stp = lop->lo_stp; 1124 ldumpp[cnt].ndlck_flags = lop->lo_flags; 1125 ldumpp[cnt].ndlck_first = lop->lo_first; 1126 ldumpp[cnt].ndlck_end = lop->lo_end; 1127 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid; 1128 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0]; 1129 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1]; 1130 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2]; 1131 ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen; 1132 NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id, 1133 stp->ls_ownerlen); 1134 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen; 1135 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id, 1136 stp->ls_clp->lc_idlen); 1137 af = stp->ls_clp->lc_req.nr_nam->sa_family; 1138 ldumpp[cnt].ndlck_addrfam = af; 1139 switch (af) { 1140 #ifdef INET 1141 case AF_INET: 1142 rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; 1143 ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; 1144 break; 1145 #endif 1146 #ifdef INET6 1147 case AF_INET6: 1148 rin6 = (struct sockaddr_in6 *) 1149 stp->ls_clp->lc_req.nr_nam; 1150 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; 1151 break; 1152 #endif 1153 } 1154 lop = LIST_NEXT(lop, lo_lckfile); 1155 cnt++; 1156 } 1157 1158 /* 1159 * and the delegations. 1160 */ 1161 stp = LIST_FIRST(&lfp->lf_deleg); 1162 while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) { 1163 ldumpp[cnt].ndlck_flags = stp->ls_flags; 1164 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid; 1165 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0]; 1166 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1]; 1167 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2]; 1168 ldumpp[cnt].ndlck_owner.nclid_idlen = 0; 1169 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen; 1170 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id, 1171 stp->ls_clp->lc_idlen); 1172 af = stp->ls_clp->lc_req.nr_nam->sa_family; 1173 ldumpp[cnt].ndlck_addrfam = af; 1174 switch (af) { 1175 #ifdef INET 1176 case AF_INET: 1177 rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam; 1178 ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr; 1179 break; 1180 #endif 1181 #ifdef INET6 1182 case AF_INET6: 1183 rin6 = (struct sockaddr_in6 *) 1184 stp->ls_clp->lc_req.nr_nam; 1185 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr; 1186 break; 1187 #endif 1188 } 1189 stp = LIST_NEXT(stp, ls_file); 1190 cnt++; 1191 } 1192 1193 /* 1194 * If list isn't full, mark end of list by setting the client name 1195 * to zero length. 1196 */ 1197 if (cnt < maxcnt) 1198 ldumpp[cnt].ndlck_clid.nclid_idlen = 0; 1199 NFSUNLOCKSTATE(); 1200 NFSLOCKV4ROOTMUTEX(); 1201 nfsv4_relref(&nfsv4rootfs_lock); 1202 NFSUNLOCKV4ROOTMUTEX(); 1203 } 1204 1205 /* 1206 * Server timer routine. It can scan any linked list, so long 1207 * as it holds the spin/mutex lock and there is no exclusive lock on 1208 * nfsv4rootfs_lock. 1209 * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok 1210 * to do this from a callout, since the spin locks work. For 1211 * Darwin, I'm not sure what will work correctly yet.) 1212 * Should be called once per second. 1213 */ 1214 APPLESTATIC void 1215 nfsrv_servertimer(void) 1216 { 1217 struct nfsclient *clp, *nclp; 1218 struct nfsstate *stp, *nstp; 1219 int got_ref, i; 1220 1221 /* 1222 * Make sure nfsboottime is set. This is used by V3 as well 1223 * as V4. Note that nfsboottime is not nfsrvboottime, which is 1224 * only used by the V4 server for leases. 1225 */ 1226 if (nfsboottime.tv_sec == 0) 1227 NFSSETBOOTTIME(nfsboottime); 1228 1229 /* 1230 * If server hasn't started yet, just return. 1231 */ 1232 NFSLOCKSTATE(); 1233 if (nfsrv_stablefirst.nsf_eograce == 0) { 1234 NFSUNLOCKSTATE(); 1235 return; 1236 } 1237 if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) { 1238 if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) && 1239 NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce) 1240 nfsrv_stablefirst.nsf_flags |= 1241 (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK); 1242 NFSUNLOCKSTATE(); 1243 return; 1244 } 1245 1246 /* 1247 * Try and get a reference count on the nfsv4rootfs_lock so that 1248 * no nfsd thread can acquire an exclusive lock on it before this 1249 * call is done. If it is already exclusively locked, just return. 1250 */ 1251 NFSLOCKV4ROOTMUTEX(); 1252 got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock); 1253 NFSUNLOCKV4ROOTMUTEX(); 1254 if (got_ref == 0) { 1255 NFSUNLOCKSTATE(); 1256 return; 1257 } 1258 1259 /* 1260 * For each client... 1261 */ 1262 for (i = 0; i < nfsrv_clienthashsize; i++) { 1263 clp = LIST_FIRST(&nfsclienthash[i]); 1264 while (clp != LIST_END(&nfsclienthash[i])) { 1265 nclp = LIST_NEXT(clp, lc_hash); 1266 if (!(clp->lc_flags & LCL_EXPIREIT)) { 1267 if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC 1268 && ((LIST_EMPTY(&clp->lc_deleg) 1269 && LIST_EMPTY(&clp->lc_open)) || 1270 nfsrv_clients > nfsrv_clienthighwater)) || 1271 (clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC || 1272 (clp->lc_expiry < NFSD_MONOSEC && 1273 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) { 1274 /* 1275 * Lease has expired several nfsrv_lease times ago: 1276 * PLUS 1277 * - no state is associated with it 1278 * OR 1279 * - above high water mark for number of clients 1280 * (nfsrv_clienthighwater should be large enough 1281 * that this only occurs when clients fail to 1282 * use the same nfs_client_id4.id. Maybe somewhat 1283 * higher that the maximum number of clients that 1284 * will mount this server?) 1285 * OR 1286 * Lease has expired a very long time ago 1287 * OR 1288 * Lease has expired PLUS the number of opens + locks 1289 * has exceeded 90% of capacity 1290 * 1291 * --> Mark for expiry. The actual expiry will be done 1292 * by an nfsd sometime soon. 1293 */ 1294 clp->lc_flags |= LCL_EXPIREIT; 1295 nfsrv_stablefirst.nsf_flags |= 1296 (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT); 1297 } else { 1298 /* 1299 * If there are no opens, increment no open tick cnt 1300 * If time exceeds NFSNOOPEN, mark it to be thrown away 1301 * otherwise, if there is an open, reset no open time 1302 * Hopefully, this will avoid excessive re-creation 1303 * of open owners and subsequent open confirms. 1304 */ 1305 stp = LIST_FIRST(&clp->lc_open); 1306 while (stp != LIST_END(&clp->lc_open)) { 1307 nstp = LIST_NEXT(stp, ls_list); 1308 if (LIST_EMPTY(&stp->ls_open)) { 1309 stp->ls_noopens++; 1310 if (stp->ls_noopens > NFSNOOPEN || 1311 (nfsrv_openpluslock * 2) > 1312 nfsrv_v4statelimit) 1313 nfsrv_stablefirst.nsf_flags |= 1314 NFSNSF_NOOPENS; 1315 } else { 1316 stp->ls_noopens = 0; 1317 } 1318 stp = nstp; 1319 } 1320 } 1321 } 1322 clp = nclp; 1323 } 1324 } 1325 NFSUNLOCKSTATE(); 1326 NFSLOCKV4ROOTMUTEX(); 1327 nfsv4_relref(&nfsv4rootfs_lock); 1328 NFSUNLOCKV4ROOTMUTEX(); 1329 } 1330 1331 /* 1332 * The following set of functions free up the various data structures. 1333 */ 1334 /* 1335 * Clear out all open/lock state related to this nfsclient. 1336 * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that 1337 * there are no other active nfsd threads. 1338 */ 1339 APPLESTATIC void 1340 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p) 1341 { 1342 struct nfsstate *stp, *nstp; 1343 struct nfsdsession *sep, *nsep; 1344 1345 LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) 1346 nfsrv_freeopenowner(stp, 1, p); 1347 if ((clp->lc_flags & LCL_ADMINREVOKED) == 0) 1348 LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) 1349 (void)nfsrv_freesession(sep, NULL); 1350 } 1351 1352 /* 1353 * Free a client that has been cleaned. It should also already have been 1354 * removed from the lists. 1355 * (Just to be safe w.r.t. newnfs_disconnect(), call this function when 1356 * softclock interrupts are enabled.) 1357 */ 1358 APPLESTATIC void 1359 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p) 1360 { 1361 1362 #ifdef notyet 1363 if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) == 1364 (LCL_GSS | LCL_CALLBACKSON) && 1365 (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) && 1366 clp->lc_handlelen > 0) { 1367 clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE; 1368 clp->lc_hand.nfsh_flag |= NFSG_DESTROYED; 1369 (void) nfsrv_docallback(clp, NFSV4PROC_CBNULL, 1370 NULL, 0, NULL, NULL, NULL, 0, p); 1371 } 1372 #endif 1373 newnfs_disconnect(&clp->lc_req); 1374 free(clp->lc_req.nr_nam, M_SONAME); 1375 NFSFREEMUTEX(&clp->lc_req.nr_mtx); 1376 free(clp->lc_stateid, M_NFSDCLIENT); 1377 free(clp, M_NFSDCLIENT); 1378 NFSLOCKSTATE(); 1379 nfsstatsv1.srvclients--; 1380 nfsrv_openpluslock--; 1381 nfsrv_clients--; 1382 NFSUNLOCKSTATE(); 1383 } 1384 1385 /* 1386 * Free a list of delegation state structures. 1387 * (This function will also free all nfslockfile structures that no 1388 * longer have associated state.) 1389 */ 1390 APPLESTATIC void 1391 nfsrv_freedeleglist(struct nfsstatehead *sthp) 1392 { 1393 struct nfsstate *stp, *nstp; 1394 1395 LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) { 1396 nfsrv_freedeleg(stp); 1397 } 1398 LIST_INIT(sthp); 1399 } 1400 1401 /* 1402 * Free up a delegation. 1403 */ 1404 static void 1405 nfsrv_freedeleg(struct nfsstate *stp) 1406 { 1407 struct nfslockfile *lfp; 1408 1409 LIST_REMOVE(stp, ls_hash); 1410 LIST_REMOVE(stp, ls_list); 1411 LIST_REMOVE(stp, ls_file); 1412 if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0) 1413 nfsrv_writedelegcnt--; 1414 lfp = stp->ls_lfp; 1415 if (LIST_EMPTY(&lfp->lf_open) && 1416 LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) && 1417 LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) && 1418 lfp->lf_usecount == 0 && 1419 nfsv4_testlock(&lfp->lf_locallock_lck) == 0) 1420 nfsrv_freenfslockfile(lfp); 1421 free(stp, M_NFSDSTATE); 1422 nfsstatsv1.srvdelegates--; 1423 nfsrv_openpluslock--; 1424 nfsrv_delegatecnt--; 1425 } 1426 1427 /* 1428 * This function frees an open owner and all associated opens. 1429 */ 1430 static void 1431 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p) 1432 { 1433 struct nfsstate *nstp, *tstp; 1434 1435 LIST_REMOVE(stp, ls_list); 1436 /* 1437 * Now, free all associated opens. 1438 */ 1439 nstp = LIST_FIRST(&stp->ls_open); 1440 while (nstp != LIST_END(&stp->ls_open)) { 1441 tstp = nstp; 1442 nstp = LIST_NEXT(nstp, ls_list); 1443 (void) nfsrv_freeopen(tstp, NULL, cansleep, p); 1444 } 1445 if (stp->ls_op) 1446 nfsrvd_derefcache(stp->ls_op); 1447 free(stp, M_NFSDSTATE); 1448 nfsstatsv1.srvopenowners--; 1449 nfsrv_openpluslock--; 1450 } 1451 1452 /* 1453 * This function frees an open (nfsstate open structure) with all associated 1454 * lock_owners and locks. It also frees the nfslockfile structure iff there 1455 * are no other opens on the file. 1456 * Returns 1 if it free'd the nfslockfile, 0 otherwise. 1457 */ 1458 static int 1459 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p) 1460 { 1461 struct nfsstate *nstp, *tstp; 1462 struct nfslockfile *lfp; 1463 int ret; 1464 1465 LIST_REMOVE(stp, ls_hash); 1466 LIST_REMOVE(stp, ls_list); 1467 LIST_REMOVE(stp, ls_file); 1468 1469 lfp = stp->ls_lfp; 1470 /* 1471 * Now, free all lockowners associated with this open. 1472 */ 1473 LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp) 1474 nfsrv_freelockowner(tstp, vp, cansleep, p); 1475 1476 /* 1477 * The nfslockfile is freed here if there are no locks 1478 * associated with the open. 1479 * If there are locks associated with the open, the 1480 * nfslockfile structure can be freed via nfsrv_freelockowner(). 1481 * Acquire the state mutex to avoid races with calls to 1482 * nfsrv_getlockfile(). 1483 */ 1484 if (cansleep != 0) 1485 NFSLOCKSTATE(); 1486 if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) && 1487 LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) && 1488 LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) && 1489 lfp->lf_usecount == 0 && 1490 (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) { 1491 nfsrv_freenfslockfile(lfp); 1492 ret = 1; 1493 } else 1494 ret = 0; 1495 if (cansleep != 0) 1496 NFSUNLOCKSTATE(); 1497 free(stp, M_NFSDSTATE); 1498 nfsstatsv1.srvopens--; 1499 nfsrv_openpluslock--; 1500 return (ret); 1501 } 1502 1503 /* 1504 * Frees a lockowner and all associated locks. 1505 */ 1506 static void 1507 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep, 1508 NFSPROC_T *p) 1509 { 1510 1511 LIST_REMOVE(stp, ls_hash); 1512 LIST_REMOVE(stp, ls_list); 1513 nfsrv_freeallnfslocks(stp, vp, cansleep, p); 1514 if (stp->ls_op) 1515 nfsrvd_derefcache(stp->ls_op); 1516 free(stp, M_NFSDSTATE); 1517 nfsstatsv1.srvlockowners--; 1518 nfsrv_openpluslock--; 1519 } 1520 1521 /* 1522 * Free all the nfs locks on a lockowner. 1523 */ 1524 static void 1525 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep, 1526 NFSPROC_T *p) 1527 { 1528 struct nfslock *lop, *nlop; 1529 struct nfsrollback *rlp, *nrlp; 1530 struct nfslockfile *lfp = NULL; 1531 int gottvp = 0; 1532 vnode_t tvp = NULL; 1533 uint64_t first, end; 1534 1535 if (vp != NULL) 1536 ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked"); 1537 lop = LIST_FIRST(&stp->ls_lock); 1538 while (lop != LIST_END(&stp->ls_lock)) { 1539 nlop = LIST_NEXT(lop, lo_lckowner); 1540 /* 1541 * Since all locks should be for the same file, lfp should 1542 * not change. 1543 */ 1544 if (lfp == NULL) 1545 lfp = lop->lo_lfp; 1546 else if (lfp != lop->lo_lfp) 1547 panic("allnfslocks"); 1548 /* 1549 * If vp is NULL and cansleep != 0, a vnode must be acquired 1550 * from the file handle. This only occurs when called from 1551 * nfsrv_cleanclient(). 1552 */ 1553 if (gottvp == 0) { 1554 if (nfsrv_dolocallocks == 0) 1555 tvp = NULL; 1556 else if (vp == NULL && cansleep != 0) { 1557 tvp = nfsvno_getvp(&lfp->lf_fh); 1558 NFSVOPUNLOCK(tvp, 0); 1559 } else 1560 tvp = vp; 1561 gottvp = 1; 1562 } 1563 1564 if (tvp != NULL) { 1565 if (cansleep == 0) 1566 panic("allnfs2"); 1567 first = lop->lo_first; 1568 end = lop->lo_end; 1569 nfsrv_freenfslock(lop); 1570 nfsrv_localunlock(tvp, lfp, first, end, p); 1571 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, 1572 nrlp) 1573 free(rlp, M_NFSDROLLBACK); 1574 LIST_INIT(&lfp->lf_rollback); 1575 } else 1576 nfsrv_freenfslock(lop); 1577 lop = nlop; 1578 } 1579 if (vp == NULL && tvp != NULL) 1580 vrele(tvp); 1581 } 1582 1583 /* 1584 * Free an nfslock structure. 1585 */ 1586 static void 1587 nfsrv_freenfslock(struct nfslock *lop) 1588 { 1589 1590 if (lop->lo_lckfile.le_prev != NULL) { 1591 LIST_REMOVE(lop, lo_lckfile); 1592 nfsstatsv1.srvlocks--; 1593 nfsrv_openpluslock--; 1594 } 1595 LIST_REMOVE(lop, lo_lckowner); 1596 free(lop, M_NFSDLOCK); 1597 } 1598 1599 /* 1600 * This function frees an nfslockfile structure. 1601 */ 1602 static void 1603 nfsrv_freenfslockfile(struct nfslockfile *lfp) 1604 { 1605 1606 LIST_REMOVE(lfp, lf_hash); 1607 free(lfp, M_NFSDLOCKFILE); 1608 } 1609 1610 /* 1611 * This function looks up an nfsstate structure via stateid. 1612 */ 1613 static int 1614 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags, 1615 struct nfsstate **stpp) 1616 { 1617 struct nfsstate *stp; 1618 struct nfsstatehead *hp; 1619 int error = 0; 1620 1621 *stpp = NULL; 1622 hp = NFSSTATEHASH(clp, *stateidp); 1623 LIST_FOREACH(stp, hp, ls_hash) { 1624 if (!NFSBCMP(stp->ls_stateid.other, stateidp->other, 1625 NFSX_STATEIDOTHER)) 1626 break; 1627 } 1628 1629 /* 1630 * If no state id in list, return NFSERR_BADSTATEID. 1631 */ 1632 if (stp == LIST_END(hp)) { 1633 error = NFSERR_BADSTATEID; 1634 goto out; 1635 } 1636 *stpp = stp; 1637 1638 out: 1639 NFSEXITCODE(error); 1640 return (error); 1641 } 1642 1643 /* 1644 * This function gets an nfsstate structure via owner string. 1645 */ 1646 static void 1647 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp, 1648 struct nfsstate **stpp) 1649 { 1650 struct nfsstate *stp; 1651 1652 *stpp = NULL; 1653 LIST_FOREACH(stp, hp, ls_list) { 1654 if (new_stp->ls_ownerlen == stp->ls_ownerlen && 1655 !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) { 1656 *stpp = stp; 1657 return; 1658 } 1659 } 1660 } 1661 1662 /* 1663 * Lock control function called to update lock status. 1664 * Returns 0 upon success, -1 if there is no lock and the flags indicate 1665 * that one isn't to be created and an NFSERR_xxx for other errors. 1666 * The structures new_stp and new_lop are passed in as pointers that should 1667 * be set to NULL if the structure is used and shouldn't be free'd. 1668 * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are 1669 * never used and can safely be allocated on the stack. For all other 1670 * cases, *new_stpp and *new_lopp should be malloc'd before the call, 1671 * in case they are used. 1672 */ 1673 APPLESTATIC int 1674 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp, 1675 struct nfslock **new_lopp, struct nfslockconflict *cfp, 1676 nfsquad_t clientid, nfsv4stateid_t *stateidp, 1677 __unused struct nfsexstuff *exp, 1678 struct nfsrv_descript *nd, NFSPROC_T *p) 1679 { 1680 struct nfslock *lop; 1681 struct nfsstate *new_stp = *new_stpp; 1682 struct nfslock *new_lop = *new_lopp; 1683 struct nfsstate *tstp, *mystp, *nstp; 1684 int specialid = 0; 1685 struct nfslockfile *lfp; 1686 struct nfslock *other_lop = NULL; 1687 struct nfsstate *stp, *lckstp = NULL; 1688 struct nfsclient *clp = NULL; 1689 u_int32_t bits; 1690 int error = 0, haslock = 0, ret, reterr; 1691 int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0; 1692 fhandle_t nfh; 1693 uint64_t first, end; 1694 uint32_t lock_flags; 1695 1696 if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) { 1697 /* 1698 * Note the special cases of "all 1s" or "all 0s" stateids and 1699 * let reads with all 1s go ahead. 1700 */ 1701 if (new_stp->ls_stateid.seqid == 0x0 && 1702 new_stp->ls_stateid.other[0] == 0x0 && 1703 new_stp->ls_stateid.other[1] == 0x0 && 1704 new_stp->ls_stateid.other[2] == 0x0) 1705 specialid = 1; 1706 else if (new_stp->ls_stateid.seqid == 0xffffffff && 1707 new_stp->ls_stateid.other[0] == 0xffffffff && 1708 new_stp->ls_stateid.other[1] == 0xffffffff && 1709 new_stp->ls_stateid.other[2] == 0xffffffff) 1710 specialid = 2; 1711 } 1712 1713 /* 1714 * Check for restart conditions (client and server). 1715 */ 1716 error = nfsrv_checkrestart(clientid, new_stp->ls_flags, 1717 &new_stp->ls_stateid, specialid); 1718 if (error) 1719 goto out; 1720 1721 /* 1722 * Check for state resource limit exceeded. 1723 */ 1724 if ((new_stp->ls_flags & NFSLCK_LOCK) && 1725 nfsrv_openpluslock > nfsrv_v4statelimit) { 1726 error = NFSERR_RESOURCE; 1727 goto out; 1728 } 1729 1730 /* 1731 * For the lock case, get another nfslock structure, 1732 * just in case we need it. 1733 * Malloc now, before we start sifting through the linked lists, 1734 * in case we have to wait for memory. 1735 */ 1736 tryagain: 1737 if (new_stp->ls_flags & NFSLCK_LOCK) 1738 other_lop = malloc(sizeof (struct nfslock), 1739 M_NFSDLOCK, M_WAITOK); 1740 filestruct_locked = 0; 1741 reterr = 0; 1742 lfp = NULL; 1743 1744 /* 1745 * Get the lockfile structure for CFH now, so we can do a sanity 1746 * check against the stateid, before incrementing the seqid#, since 1747 * we want to return NFSERR_BADSTATEID on failure and the seqid# 1748 * shouldn't be incremented for this case. 1749 * If nfsrv_getlockfile() returns -1, it means "not found", which 1750 * will be handled later. 1751 * If we are doing Lock/LockU and local locking is enabled, sleep 1752 * lock the nfslockfile structure. 1753 */ 1754 getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p); 1755 NFSLOCKSTATE(); 1756 if (getlckret == 0) { 1757 if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 && 1758 nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) { 1759 getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL, 1760 &lfp, &nfh, 1); 1761 if (getlckret == 0) 1762 filestruct_locked = 1; 1763 } else 1764 getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL, 1765 &lfp, &nfh, 0); 1766 } 1767 if (getlckret != 0 && getlckret != -1) 1768 reterr = getlckret; 1769 1770 if (filestruct_locked != 0) { 1771 LIST_INIT(&lfp->lf_rollback); 1772 if ((new_stp->ls_flags & NFSLCK_LOCK)) { 1773 /* 1774 * For local locking, do the advisory locking now, so 1775 * that any conflict can be detected. A failure later 1776 * can be rolled back locally. If an error is returned, 1777 * struct nfslockfile has been unlocked and any local 1778 * locking rolled back. 1779 */ 1780 NFSUNLOCKSTATE(); 1781 if (vnode_unlocked == 0) { 1782 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1"); 1783 vnode_unlocked = 1; 1784 NFSVOPUNLOCK(vp, 0); 1785 } 1786 reterr = nfsrv_locallock(vp, lfp, 1787 (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)), 1788 new_lop->lo_first, new_lop->lo_end, cfp, p); 1789 NFSLOCKSTATE(); 1790 } 1791 } 1792 1793 if (specialid == 0) { 1794 if (new_stp->ls_flags & NFSLCK_TEST) { 1795 /* 1796 * RFC 3530 does not list LockT as an op that renews a 1797 * lease, but the consensus seems to be that it is ok 1798 * for a server to do so. 1799 */ 1800 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 1801 (nfsquad_t)((u_quad_t)0), 0, nd, p); 1802 1803 /* 1804 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid 1805 * error returns for LockT, just go ahead and test for a lock, 1806 * since there are no locks for this client, but other locks 1807 * can conflict. (ie. same client will always be false) 1808 */ 1809 if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED) 1810 error = 0; 1811 lckstp = new_stp; 1812 } else { 1813 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 1814 (nfsquad_t)((u_quad_t)0), 0, nd, p); 1815 if (error == 0) 1816 /* 1817 * Look up the stateid 1818 */ 1819 error = nfsrv_getstate(clp, &new_stp->ls_stateid, 1820 new_stp->ls_flags, &stp); 1821 /* 1822 * do some sanity checks for an unconfirmed open or a 1823 * stateid that refers to the wrong file, for an open stateid 1824 */ 1825 if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) && 1826 ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) || 1827 (getlckret == 0 && stp->ls_lfp != lfp))){ 1828 /* 1829 * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID 1830 * The only exception is using SETATTR with SIZE. 1831 * */ 1832 if ((new_stp->ls_flags & 1833 (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR) 1834 error = NFSERR_BADSTATEID; 1835 } 1836 1837 if (error == 0 && 1838 (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) && 1839 getlckret == 0 && stp->ls_lfp != lfp) 1840 error = NFSERR_BADSTATEID; 1841 1842 /* 1843 * If the lockowner stateid doesn't refer to the same file, 1844 * I believe that is considered ok, since some clients will 1845 * only create a single lockowner and use that for all locks 1846 * on all files. 1847 * For now, log it as a diagnostic, instead of considering it 1848 * a BadStateid. 1849 */ 1850 if (error == 0 && (stp->ls_flags & 1851 (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 && 1852 getlckret == 0 && stp->ls_lfp != lfp) { 1853 #ifdef DIAGNOSTIC 1854 printf("Got a lock statid for different file open\n"); 1855 #endif 1856 /* 1857 error = NFSERR_BADSTATEID; 1858 */ 1859 } 1860 1861 if (error == 0) { 1862 if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) { 1863 /* 1864 * If haslock set, we've already checked the seqid. 1865 */ 1866 if (!haslock) { 1867 if (stp->ls_flags & NFSLCK_OPEN) 1868 error = nfsrv_checkseqid(nd, new_stp->ls_seq, 1869 stp->ls_openowner, new_stp->ls_op); 1870 else 1871 error = NFSERR_BADSTATEID; 1872 } 1873 if (!error) 1874 nfsrv_getowner(&stp->ls_open, new_stp, &lckstp); 1875 if (lckstp) 1876 /* 1877 * I believe this should be an error, but it 1878 * isn't obvious what NFSERR_xxx would be 1879 * appropriate, so I'll use NFSERR_INVAL for now. 1880 */ 1881 error = NFSERR_INVAL; 1882 else 1883 lckstp = new_stp; 1884 } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) { 1885 /* 1886 * If haslock set, ditto above. 1887 */ 1888 if (!haslock) { 1889 if (stp->ls_flags & NFSLCK_OPEN) 1890 error = NFSERR_BADSTATEID; 1891 else 1892 error = nfsrv_checkseqid(nd, new_stp->ls_seq, 1893 stp, new_stp->ls_op); 1894 } 1895 lckstp = stp; 1896 } else { 1897 lckstp = stp; 1898 } 1899 } 1900 /* 1901 * If the seqid part of the stateid isn't the same, return 1902 * NFSERR_OLDSTATEID for cases other than I/O Ops. 1903 * For I/O Ops, only return NFSERR_OLDSTATEID if 1904 * nfsrv_returnoldstateid is set. (The consensus on the email 1905 * list was that most clients would prefer to not receive 1906 * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that 1907 * is what will happen, so I use the nfsrv_returnoldstateid to 1908 * allow for either server configuration.) 1909 */ 1910 if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid && 1911 (((nd->nd_flag & ND_NFSV41) == 0 && 1912 (!(new_stp->ls_flags & NFSLCK_CHECK) || 1913 nfsrv_returnoldstateid)) || 1914 ((nd->nd_flag & ND_NFSV41) != 0 && 1915 new_stp->ls_stateid.seqid != 0))) 1916 error = NFSERR_OLDSTATEID; 1917 } 1918 } 1919 1920 /* 1921 * Now we can check for grace. 1922 */ 1923 if (!error) 1924 error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags); 1925 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error && 1926 nfsrv_checkstable(clp)) 1927 error = NFSERR_NOGRACE; 1928 /* 1929 * If we successfully Reclaimed state, note that. 1930 */ 1931 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error) 1932 nfsrv_markstable(clp); 1933 1934 /* 1935 * At this point, either error == NFSERR_BADSTATEID or the 1936 * seqid# has been updated, so we can return any error. 1937 * If error == 0, there may be an error in: 1938 * nd_repstat - Set by the calling function. 1939 * reterr - Set above, if getting the nfslockfile structure 1940 * or acquiring the local lock failed. 1941 * (If both of these are set, nd_repstat should probably be 1942 * returned, since that error was detected before this 1943 * function call.) 1944 */ 1945 if (error != 0 || nd->nd_repstat != 0 || reterr != 0) { 1946 if (error == 0) { 1947 if (nd->nd_repstat != 0) 1948 error = nd->nd_repstat; 1949 else 1950 error = reterr; 1951 } 1952 if (filestruct_locked != 0) { 1953 /* Roll back local locks. */ 1954 NFSUNLOCKSTATE(); 1955 if (vnode_unlocked == 0) { 1956 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2"); 1957 vnode_unlocked = 1; 1958 NFSVOPUNLOCK(vp, 0); 1959 } 1960 nfsrv_locallock_rollback(vp, lfp, p); 1961 NFSLOCKSTATE(); 1962 nfsrv_unlocklf(lfp); 1963 } 1964 NFSUNLOCKSTATE(); 1965 goto out; 1966 } 1967 1968 /* 1969 * Check the nfsrv_getlockfile return. 1970 * Returned -1 if no structure found. 1971 */ 1972 if (getlckret == -1) { 1973 error = NFSERR_EXPIRED; 1974 /* 1975 * Called from lockt, so no lock is OK. 1976 */ 1977 if (new_stp->ls_flags & NFSLCK_TEST) { 1978 error = 0; 1979 } else if (new_stp->ls_flags & 1980 (NFSLCK_CHECK | NFSLCK_SETATTR)) { 1981 /* 1982 * Called to check for a lock, OK if the stateid is all 1983 * 1s or all 0s, but there should be an nfsstate 1984 * otherwise. 1985 * (ie. If there is no open, I'll assume no share 1986 * deny bits.) 1987 */ 1988 if (specialid) 1989 error = 0; 1990 else 1991 error = NFSERR_BADSTATEID; 1992 } 1993 NFSUNLOCKSTATE(); 1994 goto out; 1995 } 1996 1997 /* 1998 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict. 1999 * For NFSLCK_CHECK, allow a read if write access is granted, 2000 * but check for a deny. For NFSLCK_LOCK, require correct access, 2001 * which implies a conflicting deny can't exist. 2002 */ 2003 if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) { 2004 /* 2005 * Four kinds of state id: 2006 * - specialid (all 0s or all 1s), only for NFSLCK_CHECK 2007 * - stateid for an open 2008 * - stateid for a delegation 2009 * - stateid for a lock owner 2010 */ 2011 if (!specialid) { 2012 if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) { 2013 delegation = 1; 2014 mystp = stp; 2015 nfsrv_delaydelegtimeout(stp); 2016 } else if (stp->ls_flags & NFSLCK_OPEN) { 2017 mystp = stp; 2018 } else { 2019 mystp = stp->ls_openstp; 2020 } 2021 /* 2022 * If locking or checking, require correct access 2023 * bit set. 2024 */ 2025 if (((new_stp->ls_flags & NFSLCK_LOCK) && 2026 !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) & 2027 mystp->ls_flags & NFSLCK_ACCESSBITS)) || 2028 ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) == 2029 (NFSLCK_CHECK | NFSLCK_READACCESS) && 2030 !(mystp->ls_flags & NFSLCK_READACCESS) && 2031 nfsrv_allowreadforwriteopen == 0) || 2032 ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) == 2033 (NFSLCK_CHECK | NFSLCK_WRITEACCESS) && 2034 !(mystp->ls_flags & NFSLCK_WRITEACCESS))) { 2035 if (filestruct_locked != 0) { 2036 /* Roll back local locks. */ 2037 NFSUNLOCKSTATE(); 2038 if (vnode_unlocked == 0) { 2039 ASSERT_VOP_ELOCKED(vp, 2040 "nfsrv_lockctrl3"); 2041 vnode_unlocked = 1; 2042 NFSVOPUNLOCK(vp, 0); 2043 } 2044 nfsrv_locallock_rollback(vp, lfp, p); 2045 NFSLOCKSTATE(); 2046 nfsrv_unlocklf(lfp); 2047 } 2048 NFSUNLOCKSTATE(); 2049 error = NFSERR_OPENMODE; 2050 goto out; 2051 } 2052 } else 2053 mystp = NULL; 2054 if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) { 2055 /* 2056 * Check for a conflicting deny bit. 2057 */ 2058 LIST_FOREACH(tstp, &lfp->lf_open, ls_file) { 2059 if (tstp != mystp) { 2060 bits = tstp->ls_flags; 2061 bits >>= NFSLCK_SHIFT; 2062 if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) { 2063 KASSERT(vnode_unlocked == 0, 2064 ("nfsrv_lockctrl: vnode unlocked1")); 2065 ret = nfsrv_clientconflict(tstp->ls_clp, &haslock, 2066 vp, p); 2067 if (ret == 1) { 2068 /* 2069 * nfsrv_clientconflict unlocks state 2070 * when it returns non-zero. 2071 */ 2072 lckstp = NULL; 2073 goto tryagain; 2074 } 2075 if (ret == 0) 2076 NFSUNLOCKSTATE(); 2077 if (ret == 2) 2078 error = NFSERR_PERM; 2079 else 2080 error = NFSERR_OPENMODE; 2081 goto out; 2082 } 2083 } 2084 } 2085 2086 /* We're outta here */ 2087 NFSUNLOCKSTATE(); 2088 goto out; 2089 } 2090 } 2091 2092 /* 2093 * For setattr, just get rid of all the Delegations for other clients. 2094 */ 2095 if (new_stp->ls_flags & NFSLCK_SETATTR) { 2096 KASSERT(vnode_unlocked == 0, 2097 ("nfsrv_lockctrl: vnode unlocked2")); 2098 ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p); 2099 if (ret) { 2100 /* 2101 * nfsrv_cleandeleg() unlocks state when it 2102 * returns non-zero. 2103 */ 2104 if (ret == -1) { 2105 lckstp = NULL; 2106 goto tryagain; 2107 } 2108 error = ret; 2109 goto out; 2110 } 2111 if (!(new_stp->ls_flags & NFSLCK_CHECK) || 2112 (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) && 2113 LIST_EMPTY(&lfp->lf_deleg))) { 2114 NFSUNLOCKSTATE(); 2115 goto out; 2116 } 2117 } 2118 2119 /* 2120 * Check for a conflicting delegation. If one is found, call 2121 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't 2122 * been set yet, it will get the lock. Otherwise, it will recall 2123 * the delegation. Then, we try try again... 2124 * I currently believe the conflict algorithm to be: 2125 * For Lock Ops (Lock/LockT/LockU) 2126 * - there is a conflict iff a different client has a write delegation 2127 * For Reading (Read Op) 2128 * - there is a conflict iff a different client has a write delegation 2129 * (the specialids are always a different client) 2130 * For Writing (Write/Setattr of size) 2131 * - there is a conflict if a different client has any delegation 2132 * - there is a conflict if the same client has a read delegation 2133 * (I don't understand why this isn't allowed, but that seems to be 2134 * the current consensus?) 2135 */ 2136 tstp = LIST_FIRST(&lfp->lf_deleg); 2137 while (tstp != LIST_END(&lfp->lf_deleg)) { 2138 nstp = LIST_NEXT(tstp, ls_file); 2139 if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))|| 2140 ((new_stp->ls_flags & NFSLCK_CHECK) && 2141 (new_lop->lo_flags & NFSLCK_READ))) && 2142 clp != tstp->ls_clp && 2143 (tstp->ls_flags & NFSLCK_DELEGWRITE)) || 2144 ((new_stp->ls_flags & NFSLCK_CHECK) && 2145 (new_lop->lo_flags & NFSLCK_WRITE) && 2146 (clp != tstp->ls_clp || 2147 (tstp->ls_flags & NFSLCK_DELEGREAD)))) { 2148 ret = 0; 2149 if (filestruct_locked != 0) { 2150 /* Roll back local locks. */ 2151 NFSUNLOCKSTATE(); 2152 if (vnode_unlocked == 0) { 2153 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4"); 2154 NFSVOPUNLOCK(vp, 0); 2155 } 2156 nfsrv_locallock_rollback(vp, lfp, p); 2157 NFSLOCKSTATE(); 2158 nfsrv_unlocklf(lfp); 2159 NFSUNLOCKSTATE(); 2160 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2161 vnode_unlocked = 0; 2162 if ((vp->v_iflag & VI_DOOMED) != 0) 2163 ret = NFSERR_SERVERFAULT; 2164 NFSLOCKSTATE(); 2165 } 2166 if (ret == 0) 2167 ret = nfsrv_delegconflict(tstp, &haslock, p, vp); 2168 if (ret) { 2169 /* 2170 * nfsrv_delegconflict unlocks state when it 2171 * returns non-zero, which it always does. 2172 */ 2173 if (other_lop) { 2174 free(other_lop, M_NFSDLOCK); 2175 other_lop = NULL; 2176 } 2177 if (ret == -1) { 2178 lckstp = NULL; 2179 goto tryagain; 2180 } 2181 error = ret; 2182 goto out; 2183 } 2184 /* Never gets here. */ 2185 } 2186 tstp = nstp; 2187 } 2188 2189 /* 2190 * Handle the unlock case by calling nfsrv_updatelock(). 2191 * (Should I have done some access checking above for unlock? For now, 2192 * just let it happen.) 2193 */ 2194 if (new_stp->ls_flags & NFSLCK_UNLOCK) { 2195 first = new_lop->lo_first; 2196 end = new_lop->lo_end; 2197 nfsrv_updatelock(stp, new_lopp, &other_lop, lfp); 2198 stateidp->seqid = ++(stp->ls_stateid.seqid); 2199 if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) 2200 stateidp->seqid = stp->ls_stateid.seqid = 1; 2201 stateidp->other[0] = stp->ls_stateid.other[0]; 2202 stateidp->other[1] = stp->ls_stateid.other[1]; 2203 stateidp->other[2] = stp->ls_stateid.other[2]; 2204 if (filestruct_locked != 0) { 2205 NFSUNLOCKSTATE(); 2206 if (vnode_unlocked == 0) { 2207 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5"); 2208 vnode_unlocked = 1; 2209 NFSVOPUNLOCK(vp, 0); 2210 } 2211 /* Update the local locks. */ 2212 nfsrv_localunlock(vp, lfp, first, end, p); 2213 NFSLOCKSTATE(); 2214 nfsrv_unlocklf(lfp); 2215 } 2216 NFSUNLOCKSTATE(); 2217 goto out; 2218 } 2219 2220 /* 2221 * Search for a conflicting lock. A lock conflicts if: 2222 * - the lock range overlaps and 2223 * - at least one lock is a write lock and 2224 * - it is not owned by the same lock owner 2225 */ 2226 if (!delegation) { 2227 LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) { 2228 if (new_lop->lo_end > lop->lo_first && 2229 new_lop->lo_first < lop->lo_end && 2230 (new_lop->lo_flags == NFSLCK_WRITE || 2231 lop->lo_flags == NFSLCK_WRITE) && 2232 lckstp != lop->lo_stp && 2233 (clp != lop->lo_stp->ls_clp || 2234 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen || 2235 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner, 2236 lckstp->ls_ownerlen))) { 2237 if (other_lop) { 2238 free(other_lop, M_NFSDLOCK); 2239 other_lop = NULL; 2240 } 2241 if (vnode_unlocked != 0) 2242 ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock, 2243 NULL, p); 2244 else 2245 ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock, 2246 vp, p); 2247 if (ret == 1) { 2248 if (filestruct_locked != 0) { 2249 if (vnode_unlocked == 0) { 2250 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6"); 2251 NFSVOPUNLOCK(vp, 0); 2252 } 2253 /* Roll back local locks. */ 2254 nfsrv_locallock_rollback(vp, lfp, p); 2255 NFSLOCKSTATE(); 2256 nfsrv_unlocklf(lfp); 2257 NFSUNLOCKSTATE(); 2258 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2259 vnode_unlocked = 0; 2260 if ((vp->v_iflag & VI_DOOMED) != 0) { 2261 error = NFSERR_SERVERFAULT; 2262 goto out; 2263 } 2264 } 2265 /* 2266 * nfsrv_clientconflict() unlocks state when it 2267 * returns non-zero. 2268 */ 2269 lckstp = NULL; 2270 goto tryagain; 2271 } 2272 /* 2273 * Found a conflicting lock, so record the conflict and 2274 * return the error. 2275 */ 2276 if (cfp != NULL && ret == 0) { 2277 cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0]; 2278 cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1]; 2279 cfp->cl_first = lop->lo_first; 2280 cfp->cl_end = lop->lo_end; 2281 cfp->cl_flags = lop->lo_flags; 2282 cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen; 2283 NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner, 2284 cfp->cl_ownerlen); 2285 } 2286 if (ret == 2) 2287 error = NFSERR_PERM; 2288 else if (new_stp->ls_flags & NFSLCK_RECLAIM) 2289 error = NFSERR_RECLAIMCONFLICT; 2290 else if (new_stp->ls_flags & NFSLCK_CHECK) 2291 error = NFSERR_LOCKED; 2292 else 2293 error = NFSERR_DENIED; 2294 if (filestruct_locked != 0 && ret == 0) { 2295 /* Roll back local locks. */ 2296 NFSUNLOCKSTATE(); 2297 if (vnode_unlocked == 0) { 2298 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7"); 2299 vnode_unlocked = 1; 2300 NFSVOPUNLOCK(vp, 0); 2301 } 2302 nfsrv_locallock_rollback(vp, lfp, p); 2303 NFSLOCKSTATE(); 2304 nfsrv_unlocklf(lfp); 2305 } 2306 if (ret == 0) 2307 NFSUNLOCKSTATE(); 2308 goto out; 2309 } 2310 } 2311 } 2312 2313 /* 2314 * We only get here if there was no lock that conflicted. 2315 */ 2316 if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) { 2317 NFSUNLOCKSTATE(); 2318 goto out; 2319 } 2320 2321 /* 2322 * We only get here when we are creating or modifying a lock. 2323 * There are two variants: 2324 * - exist_lock_owner where lock_owner exists 2325 * - open_to_lock_owner with new lock_owner 2326 */ 2327 first = new_lop->lo_first; 2328 end = new_lop->lo_end; 2329 lock_flags = new_lop->lo_flags; 2330 if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) { 2331 nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp); 2332 stateidp->seqid = ++(lckstp->ls_stateid.seqid); 2333 if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) 2334 stateidp->seqid = lckstp->ls_stateid.seqid = 1; 2335 stateidp->other[0] = lckstp->ls_stateid.other[0]; 2336 stateidp->other[1] = lckstp->ls_stateid.other[1]; 2337 stateidp->other[2] = lckstp->ls_stateid.other[2]; 2338 } else { 2339 /* 2340 * The new open_to_lock_owner case. 2341 * Link the new nfsstate into the lists. 2342 */ 2343 new_stp->ls_seq = new_stp->ls_opentolockseq; 2344 nfsrvd_refcache(new_stp->ls_op); 2345 stateidp->seqid = new_stp->ls_stateid.seqid = 1; 2346 stateidp->other[0] = new_stp->ls_stateid.other[0] = 2347 clp->lc_clientid.lval[0]; 2348 stateidp->other[1] = new_stp->ls_stateid.other[1] = 2349 clp->lc_clientid.lval[1]; 2350 stateidp->other[2] = new_stp->ls_stateid.other[2] = 2351 nfsrv_nextstateindex(clp); 2352 new_stp->ls_clp = clp; 2353 LIST_INIT(&new_stp->ls_lock); 2354 new_stp->ls_openstp = stp; 2355 new_stp->ls_lfp = lfp; 2356 nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp, 2357 lfp); 2358 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid), 2359 new_stp, ls_hash); 2360 LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list); 2361 *new_lopp = NULL; 2362 *new_stpp = NULL; 2363 nfsstatsv1.srvlockowners++; 2364 nfsrv_openpluslock++; 2365 } 2366 if (filestruct_locked != 0) { 2367 NFSUNLOCKSTATE(); 2368 nfsrv_locallock_commit(lfp, lock_flags, first, end); 2369 NFSLOCKSTATE(); 2370 nfsrv_unlocklf(lfp); 2371 } 2372 NFSUNLOCKSTATE(); 2373 2374 out: 2375 if (haslock) { 2376 NFSLOCKV4ROOTMUTEX(); 2377 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2378 NFSUNLOCKV4ROOTMUTEX(); 2379 } 2380 if (vnode_unlocked != 0) { 2381 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 2382 if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) 2383 error = NFSERR_SERVERFAULT; 2384 } 2385 if (other_lop) 2386 free(other_lop, M_NFSDLOCK); 2387 NFSEXITCODE2(error, nd); 2388 return (error); 2389 } 2390 2391 /* 2392 * Check for state errors for Open. 2393 * repstat is passed back out as an error if more critical errors 2394 * are not detected. 2395 */ 2396 APPLESTATIC int 2397 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp, 2398 struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd, 2399 NFSPROC_T *p, int repstat) 2400 { 2401 struct nfsstate *stp, *nstp; 2402 struct nfsclient *clp; 2403 struct nfsstate *ownerstp; 2404 struct nfslockfile *lfp, *new_lfp; 2405 int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0; 2406 2407 if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS) 2408 readonly = 1; 2409 /* 2410 * Check for restart conditions (client and server). 2411 */ 2412 error = nfsrv_checkrestart(clientid, new_stp->ls_flags, 2413 &new_stp->ls_stateid, 0); 2414 if (error) 2415 goto out; 2416 2417 /* 2418 * Check for state resource limit exceeded. 2419 * Technically this should be SMP protected, but the worst 2420 * case error is "out by one or two" on the count when it 2421 * returns NFSERR_RESOURCE and the limit is just a rather 2422 * arbitrary high water mark, so no harm is done. 2423 */ 2424 if (nfsrv_openpluslock > nfsrv_v4statelimit) { 2425 error = NFSERR_RESOURCE; 2426 goto out; 2427 } 2428 2429 tryagain: 2430 new_lfp = malloc(sizeof (struct nfslockfile), 2431 M_NFSDLOCKFILE, M_WAITOK); 2432 if (vp) 2433 getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp, 2434 NULL, p); 2435 NFSLOCKSTATE(); 2436 /* 2437 * Get the nfsclient structure. 2438 */ 2439 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 2440 (nfsquad_t)((u_quad_t)0), 0, nd, p); 2441 2442 /* 2443 * Look up the open owner. See if it needs confirmation and 2444 * check the seq#, as required. 2445 */ 2446 if (!error) 2447 nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp); 2448 2449 if (!error && ownerstp) { 2450 error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp, 2451 new_stp->ls_op); 2452 /* 2453 * If the OpenOwner hasn't been confirmed, assume the 2454 * old one was a replay and this one is ok. 2455 * See: RFC3530 Sec. 14.2.18. 2456 */ 2457 if (error == NFSERR_BADSEQID && 2458 (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM)) 2459 error = 0; 2460 } 2461 2462 /* 2463 * Check for grace. 2464 */ 2465 if (!error) 2466 error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags); 2467 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error && 2468 nfsrv_checkstable(clp)) 2469 error = NFSERR_NOGRACE; 2470 2471 /* 2472 * If none of the above errors occurred, let repstat be 2473 * returned. 2474 */ 2475 if (repstat && !error) 2476 error = repstat; 2477 if (error) { 2478 NFSUNLOCKSTATE(); 2479 if (haslock) { 2480 NFSLOCKV4ROOTMUTEX(); 2481 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2482 NFSUNLOCKV4ROOTMUTEX(); 2483 } 2484 free(new_lfp, M_NFSDLOCKFILE); 2485 goto out; 2486 } 2487 2488 /* 2489 * If vp == NULL, the file doesn't exist yet, so return ok. 2490 * (This always happens on the first pass, so haslock must be 0.) 2491 */ 2492 if (vp == NULL) { 2493 NFSUNLOCKSTATE(); 2494 free(new_lfp, M_NFSDLOCKFILE); 2495 goto out; 2496 } 2497 2498 /* 2499 * Get the structure for the underlying file. 2500 */ 2501 if (getfhret) 2502 error = getfhret; 2503 else 2504 error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp, 2505 NULL, 0); 2506 if (new_lfp) 2507 free(new_lfp, M_NFSDLOCKFILE); 2508 if (error) { 2509 NFSUNLOCKSTATE(); 2510 if (haslock) { 2511 NFSLOCKV4ROOTMUTEX(); 2512 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2513 NFSUNLOCKV4ROOTMUTEX(); 2514 } 2515 goto out; 2516 } 2517 2518 /* 2519 * Search for a conflicting open/share. 2520 */ 2521 if (new_stp->ls_flags & NFSLCK_DELEGCUR) { 2522 /* 2523 * For Delegate_Cur, search for the matching Delegation, 2524 * which indicates no conflict. 2525 * An old delegation should have been recovered by the 2526 * client doing a Claim_DELEGATE_Prev, so I won't let 2527 * it match and return NFSERR_EXPIRED. Should I let it 2528 * match? 2529 */ 2530 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { 2531 if (!(stp->ls_flags & NFSLCK_OLDDELEG) && 2532 (((nd->nd_flag & ND_NFSV41) != 0 && 2533 stateidp->seqid == 0) || 2534 stateidp->seqid == stp->ls_stateid.seqid) && 2535 !NFSBCMP(stateidp->other, stp->ls_stateid.other, 2536 NFSX_STATEIDOTHER)) 2537 break; 2538 } 2539 if (stp == LIST_END(&lfp->lf_deleg) || 2540 ((new_stp->ls_flags & NFSLCK_WRITEACCESS) && 2541 (stp->ls_flags & NFSLCK_DELEGREAD))) { 2542 NFSUNLOCKSTATE(); 2543 if (haslock) { 2544 NFSLOCKV4ROOTMUTEX(); 2545 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2546 NFSUNLOCKV4ROOTMUTEX(); 2547 } 2548 error = NFSERR_EXPIRED; 2549 goto out; 2550 } 2551 } 2552 2553 /* 2554 * Check for access/deny bit conflicts. I check for the same 2555 * owner as well, in case the client didn't bother. 2556 */ 2557 LIST_FOREACH(stp, &lfp->lf_open, ls_file) { 2558 if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) && 2559 (((new_stp->ls_flags & NFSLCK_ACCESSBITS) & 2560 ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))|| 2561 ((stp->ls_flags & NFSLCK_ACCESSBITS) & 2562 ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){ 2563 ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p); 2564 if (ret == 1) { 2565 /* 2566 * nfsrv_clientconflict() unlocks 2567 * state when it returns non-zero. 2568 */ 2569 goto tryagain; 2570 } 2571 if (ret == 2) 2572 error = NFSERR_PERM; 2573 else if (new_stp->ls_flags & NFSLCK_RECLAIM) 2574 error = NFSERR_RECLAIMCONFLICT; 2575 else 2576 error = NFSERR_SHAREDENIED; 2577 if (ret == 0) 2578 NFSUNLOCKSTATE(); 2579 if (haslock) { 2580 NFSLOCKV4ROOTMUTEX(); 2581 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2582 NFSUNLOCKV4ROOTMUTEX(); 2583 } 2584 goto out; 2585 } 2586 } 2587 2588 /* 2589 * Check for a conflicting delegation. If one is found, call 2590 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't 2591 * been set yet, it will get the lock. Otherwise, it will recall 2592 * the delegation. Then, we try try again... 2593 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there 2594 * isn't a conflict.) 2595 * I currently believe the conflict algorithm to be: 2596 * For Open with Read Access and Deny None 2597 * - there is a conflict iff a different client has a write delegation 2598 * For Open with other Write Access or any Deny except None 2599 * - there is a conflict if a different client has any delegation 2600 * - there is a conflict if the same client has a read delegation 2601 * (The current consensus is that this last case should be 2602 * considered a conflict since the client with a read delegation 2603 * could have done an Open with ReadAccess and WriteDeny 2604 * locally and then not have checked for the WriteDeny.) 2605 * Don't check for a Reclaim, since that will be dealt with 2606 * by nfsrv_openctrl(). 2607 */ 2608 if (!(new_stp->ls_flags & 2609 (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) { 2610 stp = LIST_FIRST(&lfp->lf_deleg); 2611 while (stp != LIST_END(&lfp->lf_deleg)) { 2612 nstp = LIST_NEXT(stp, ls_file); 2613 if ((readonly && stp->ls_clp != clp && 2614 (stp->ls_flags & NFSLCK_DELEGWRITE)) || 2615 (!readonly && (stp->ls_clp != clp || 2616 (stp->ls_flags & NFSLCK_DELEGREAD)))) { 2617 ret = nfsrv_delegconflict(stp, &haslock, p, vp); 2618 if (ret) { 2619 /* 2620 * nfsrv_delegconflict() unlocks state 2621 * when it returns non-zero. 2622 */ 2623 if (ret == -1) 2624 goto tryagain; 2625 error = ret; 2626 goto out; 2627 } 2628 } 2629 stp = nstp; 2630 } 2631 } 2632 NFSUNLOCKSTATE(); 2633 if (haslock) { 2634 NFSLOCKV4ROOTMUTEX(); 2635 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2636 NFSUNLOCKV4ROOTMUTEX(); 2637 } 2638 2639 out: 2640 NFSEXITCODE2(error, nd); 2641 return (error); 2642 } 2643 2644 /* 2645 * Open control function to create/update open state for an open. 2646 */ 2647 APPLESTATIC int 2648 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp, 2649 struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp, 2650 nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp, 2651 NFSPROC_T *p, u_quad_t filerev) 2652 { 2653 struct nfsstate *new_stp = *new_stpp; 2654 struct nfsstate *stp, *nstp; 2655 struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg; 2656 struct nfslockfile *lfp, *new_lfp; 2657 struct nfsclient *clp; 2658 int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1; 2659 int readonly = 0, cbret = 1, getfhret = 0; 2660 int gotstate = 0, len = 0; 2661 u_char *clidp = NULL; 2662 2663 if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS) 2664 readonly = 1; 2665 /* 2666 * Check for restart conditions (client and server). 2667 * (Paranoia, should have been detected by nfsrv_opencheck().) 2668 * If an error does show up, return NFSERR_EXPIRED, since the 2669 * the seqid# has already been incremented. 2670 */ 2671 error = nfsrv_checkrestart(clientid, new_stp->ls_flags, 2672 &new_stp->ls_stateid, 0); 2673 if (error) { 2674 printf("Nfsd: openctrl unexpected restart err=%d\n", 2675 error); 2676 error = NFSERR_EXPIRED; 2677 goto out; 2678 } 2679 2680 clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK); 2681 tryagain: 2682 new_lfp = malloc(sizeof (struct nfslockfile), 2683 M_NFSDLOCKFILE, M_WAITOK); 2684 new_open = malloc(sizeof (struct nfsstate), 2685 M_NFSDSTATE, M_WAITOK); 2686 new_deleg = malloc(sizeof (struct nfsstate), 2687 M_NFSDSTATE, M_WAITOK); 2688 getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp, 2689 NULL, p); 2690 NFSLOCKSTATE(); 2691 /* 2692 * Get the client structure. Since the linked lists could be changed 2693 * by other nfsd processes if this process does a tsleep(), one of 2694 * two things must be done. 2695 * 1 - don't tsleep() 2696 * or 2697 * 2 - get the nfsv4_lock() { indicated by haslock == 1 } 2698 * before using the lists, since this lock stops the other 2699 * nfsd. This should only be used for rare cases, since it 2700 * essentially single threads the nfsd. 2701 * At this time, it is only done for cases where the stable 2702 * storage file must be written prior to completion of state 2703 * expiration. 2704 */ 2705 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 2706 (nfsquad_t)((u_quad_t)0), 0, nd, p); 2707 if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) && 2708 clp->lc_program) { 2709 /* 2710 * This happens on the first open for a client 2711 * that supports callbacks. 2712 */ 2713 NFSUNLOCKSTATE(); 2714 /* 2715 * Although nfsrv_docallback() will sleep, clp won't 2716 * go away, since they are only removed when the 2717 * nfsv4_lock() has blocked the nfsd threads. The 2718 * fields in clp can change, but having multiple 2719 * threads do this Null callback RPC should be 2720 * harmless. 2721 */ 2722 cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL, 2723 NULL, 0, NULL, NULL, NULL, 0, p); 2724 NFSLOCKSTATE(); 2725 clp->lc_flags &= ~LCL_NEEDSCBNULL; 2726 if (!cbret) 2727 clp->lc_flags |= LCL_CALLBACKSON; 2728 } 2729 2730 /* 2731 * Look up the open owner. See if it needs confirmation and 2732 * check the seq#, as required. 2733 */ 2734 if (!error) 2735 nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp); 2736 2737 if (error) { 2738 NFSUNLOCKSTATE(); 2739 printf("Nfsd: openctrl unexpected state err=%d\n", 2740 error); 2741 free(new_lfp, M_NFSDLOCKFILE); 2742 free(new_open, M_NFSDSTATE); 2743 free(new_deleg, M_NFSDSTATE); 2744 if (haslock) { 2745 NFSLOCKV4ROOTMUTEX(); 2746 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2747 NFSUNLOCKV4ROOTMUTEX(); 2748 } 2749 error = NFSERR_EXPIRED; 2750 goto out; 2751 } 2752 2753 if (new_stp->ls_flags & NFSLCK_RECLAIM) 2754 nfsrv_markstable(clp); 2755 2756 /* 2757 * Get the structure for the underlying file. 2758 */ 2759 if (getfhret) 2760 error = getfhret; 2761 else 2762 error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp, 2763 NULL, 0); 2764 if (new_lfp) 2765 free(new_lfp, M_NFSDLOCKFILE); 2766 if (error) { 2767 NFSUNLOCKSTATE(); 2768 printf("Nfsd openctrl unexpected getlockfile err=%d\n", 2769 error); 2770 free(new_open, M_NFSDSTATE); 2771 free(new_deleg, M_NFSDSTATE); 2772 if (haslock) { 2773 NFSLOCKV4ROOTMUTEX(); 2774 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2775 NFSUNLOCKV4ROOTMUTEX(); 2776 } 2777 goto out; 2778 } 2779 2780 /* 2781 * Search for a conflicting open/share. 2782 */ 2783 if (new_stp->ls_flags & NFSLCK_DELEGCUR) { 2784 /* 2785 * For Delegate_Cur, search for the matching Delegation, 2786 * which indicates no conflict. 2787 * An old delegation should have been recovered by the 2788 * client doing a Claim_DELEGATE_Prev, so I won't let 2789 * it match and return NFSERR_EXPIRED. Should I let it 2790 * match? 2791 */ 2792 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { 2793 if (!(stp->ls_flags & NFSLCK_OLDDELEG) && 2794 (((nd->nd_flag & ND_NFSV41) != 0 && 2795 stateidp->seqid == 0) || 2796 stateidp->seqid == stp->ls_stateid.seqid) && 2797 !NFSBCMP(stateidp->other, stp->ls_stateid.other, 2798 NFSX_STATEIDOTHER)) 2799 break; 2800 } 2801 if (stp == LIST_END(&lfp->lf_deleg) || 2802 ((new_stp->ls_flags & NFSLCK_WRITEACCESS) && 2803 (stp->ls_flags & NFSLCK_DELEGREAD))) { 2804 NFSUNLOCKSTATE(); 2805 printf("Nfsd openctrl unexpected expiry\n"); 2806 free(new_open, M_NFSDSTATE); 2807 free(new_deleg, M_NFSDSTATE); 2808 if (haslock) { 2809 NFSLOCKV4ROOTMUTEX(); 2810 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2811 NFSUNLOCKV4ROOTMUTEX(); 2812 } 2813 error = NFSERR_EXPIRED; 2814 goto out; 2815 } 2816 2817 /* 2818 * Don't issue a Delegation, since one already exists and 2819 * delay delegation timeout, as required. 2820 */ 2821 delegate = 0; 2822 nfsrv_delaydelegtimeout(stp); 2823 } 2824 2825 /* 2826 * Check for access/deny bit conflicts. I also check for the 2827 * same owner, since the client might not have bothered to check. 2828 * Also, note an open for the same file and owner, if found, 2829 * which is all we do here for Delegate_Cur, since conflict 2830 * checking is already done. 2831 */ 2832 LIST_FOREACH(stp, &lfp->lf_open, ls_file) { 2833 if (ownerstp && stp->ls_openowner == ownerstp) 2834 openstp = stp; 2835 if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) { 2836 /* 2837 * If another client has the file open, the only 2838 * delegation that can be issued is a Read delegation 2839 * and only if it is a Read open with Deny none. 2840 */ 2841 if (clp != stp->ls_clp) { 2842 if ((stp->ls_flags & NFSLCK_SHAREBITS) == 2843 NFSLCK_READACCESS) 2844 writedeleg = 0; 2845 else 2846 delegate = 0; 2847 } 2848 if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) & 2849 ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))|| 2850 ((stp->ls_flags & NFSLCK_ACCESSBITS) & 2851 ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){ 2852 ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p); 2853 if (ret == 1) { 2854 /* 2855 * nfsrv_clientconflict() unlocks state 2856 * when it returns non-zero. 2857 */ 2858 free(new_open, M_NFSDSTATE); 2859 free(new_deleg, M_NFSDSTATE); 2860 openstp = NULL; 2861 goto tryagain; 2862 } 2863 if (ret == 2) 2864 error = NFSERR_PERM; 2865 else if (new_stp->ls_flags & NFSLCK_RECLAIM) 2866 error = NFSERR_RECLAIMCONFLICT; 2867 else 2868 error = NFSERR_SHAREDENIED; 2869 if (ret == 0) 2870 NFSUNLOCKSTATE(); 2871 if (haslock) { 2872 NFSLOCKV4ROOTMUTEX(); 2873 nfsv4_unlock(&nfsv4rootfs_lock, 1); 2874 NFSUNLOCKV4ROOTMUTEX(); 2875 } 2876 free(new_open, M_NFSDSTATE); 2877 free(new_deleg, M_NFSDSTATE); 2878 printf("nfsd openctrl unexpected client cnfl\n"); 2879 goto out; 2880 } 2881 } 2882 } 2883 2884 /* 2885 * Check for a conflicting delegation. If one is found, call 2886 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't 2887 * been set yet, it will get the lock. Otherwise, it will recall 2888 * the delegation. Then, we try try again... 2889 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there 2890 * isn't a conflict.) 2891 * I currently believe the conflict algorithm to be: 2892 * For Open with Read Access and Deny None 2893 * - there is a conflict iff a different client has a write delegation 2894 * For Open with other Write Access or any Deny except None 2895 * - there is a conflict if a different client has any delegation 2896 * - there is a conflict if the same client has a read delegation 2897 * (The current consensus is that this last case should be 2898 * considered a conflict since the client with a read delegation 2899 * could have done an Open with ReadAccess and WriteDeny 2900 * locally and then not have checked for the WriteDeny.) 2901 */ 2902 if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) { 2903 stp = LIST_FIRST(&lfp->lf_deleg); 2904 while (stp != LIST_END(&lfp->lf_deleg)) { 2905 nstp = LIST_NEXT(stp, ls_file); 2906 if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD)) 2907 writedeleg = 0; 2908 else 2909 delegate = 0; 2910 if ((readonly && stp->ls_clp != clp && 2911 (stp->ls_flags & NFSLCK_DELEGWRITE)) || 2912 (!readonly && (stp->ls_clp != clp || 2913 (stp->ls_flags & NFSLCK_DELEGREAD)))) { 2914 if (new_stp->ls_flags & NFSLCK_RECLAIM) { 2915 delegate = 2; 2916 } else { 2917 ret = nfsrv_delegconflict(stp, &haslock, p, vp); 2918 if (ret) { 2919 /* 2920 * nfsrv_delegconflict() unlocks state 2921 * when it returns non-zero. 2922 */ 2923 printf("Nfsd openctrl unexpected deleg cnfl\n"); 2924 free(new_open, M_NFSDSTATE); 2925 free(new_deleg, M_NFSDSTATE); 2926 if (ret == -1) { 2927 openstp = NULL; 2928 goto tryagain; 2929 } 2930 error = ret; 2931 goto out; 2932 } 2933 } 2934 } 2935 stp = nstp; 2936 } 2937 } 2938 2939 /* 2940 * We only get here if there was no open that conflicted. 2941 * If an open for the owner exists, or in the access/deny bits. 2942 * Otherwise it is a new open. If the open_owner hasn't been 2943 * confirmed, replace the open with the new one needing confirmation, 2944 * otherwise add the open. 2945 */ 2946 if (new_stp->ls_flags & NFSLCK_DELEGPREV) { 2947 /* 2948 * Handle NFSLCK_DELEGPREV by searching the old delegations for 2949 * a match. If found, just move the old delegation to the current 2950 * delegation list and issue open. If not found, return 2951 * NFSERR_EXPIRED. 2952 */ 2953 LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) { 2954 if (stp->ls_lfp == lfp) { 2955 /* Found it */ 2956 if (stp->ls_clp != clp) 2957 panic("olddeleg clp"); 2958 LIST_REMOVE(stp, ls_list); 2959 LIST_REMOVE(stp, ls_hash); 2960 stp->ls_flags &= ~NFSLCK_OLDDELEG; 2961 stp->ls_stateid.seqid = delegstateidp->seqid = 1; 2962 stp->ls_stateid.other[0] = delegstateidp->other[0] = 2963 clp->lc_clientid.lval[0]; 2964 stp->ls_stateid.other[1] = delegstateidp->other[1] = 2965 clp->lc_clientid.lval[1]; 2966 stp->ls_stateid.other[2] = delegstateidp->other[2] = 2967 nfsrv_nextstateindex(clp); 2968 stp->ls_compref = nd->nd_compref; 2969 LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list); 2970 LIST_INSERT_HEAD(NFSSTATEHASH(clp, 2971 stp->ls_stateid), stp, ls_hash); 2972 if (stp->ls_flags & NFSLCK_DELEGWRITE) 2973 *rflagsp |= NFSV4OPEN_WRITEDELEGATE; 2974 else 2975 *rflagsp |= NFSV4OPEN_READDELEGATE; 2976 clp->lc_delegtime = NFSD_MONOSEC + 2977 nfsrv_lease + NFSRV_LEASEDELTA; 2978 2979 /* 2980 * Now, do the associated open. 2981 */ 2982 new_open->ls_stateid.seqid = 1; 2983 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; 2984 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; 2985 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); 2986 new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)| 2987 NFSLCK_OPEN; 2988 if (stp->ls_flags & NFSLCK_DELEGWRITE) 2989 new_open->ls_flags |= (NFSLCK_READACCESS | 2990 NFSLCK_WRITEACCESS); 2991 else 2992 new_open->ls_flags |= NFSLCK_READACCESS; 2993 new_open->ls_uid = new_stp->ls_uid; 2994 new_open->ls_lfp = lfp; 2995 new_open->ls_clp = clp; 2996 LIST_INIT(&new_open->ls_open); 2997 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); 2998 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), 2999 new_open, ls_hash); 3000 /* 3001 * and handle the open owner 3002 */ 3003 if (ownerstp) { 3004 new_open->ls_openowner = ownerstp; 3005 LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list); 3006 } else { 3007 new_open->ls_openowner = new_stp; 3008 new_stp->ls_flags = 0; 3009 nfsrvd_refcache(new_stp->ls_op); 3010 new_stp->ls_noopens = 0; 3011 LIST_INIT(&new_stp->ls_open); 3012 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list); 3013 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list); 3014 *new_stpp = NULL; 3015 nfsstatsv1.srvopenowners++; 3016 nfsrv_openpluslock++; 3017 } 3018 openstp = new_open; 3019 new_open = NULL; 3020 nfsstatsv1.srvopens++; 3021 nfsrv_openpluslock++; 3022 break; 3023 } 3024 } 3025 if (stp == LIST_END(&clp->lc_olddeleg)) 3026 error = NFSERR_EXPIRED; 3027 } else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) { 3028 /* 3029 * Scan to see that no delegation for this client and file 3030 * doesn't already exist. 3031 * There also shouldn't yet be an Open for this file and 3032 * openowner. 3033 */ 3034 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { 3035 if (stp->ls_clp == clp) 3036 break; 3037 } 3038 if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) { 3039 /* 3040 * This is the Claim_Previous case with a delegation 3041 * type != Delegate_None. 3042 */ 3043 /* 3044 * First, add the delegation. (Although we must issue the 3045 * delegation, we can also ask for an immediate return.) 3046 */ 3047 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; 3048 new_deleg->ls_stateid.other[0] = delegstateidp->other[0] = 3049 clp->lc_clientid.lval[0]; 3050 new_deleg->ls_stateid.other[1] = delegstateidp->other[1] = 3051 clp->lc_clientid.lval[1]; 3052 new_deleg->ls_stateid.other[2] = delegstateidp->other[2] = 3053 nfsrv_nextstateindex(clp); 3054 if (new_stp->ls_flags & NFSLCK_DELEGWRITE) { 3055 new_deleg->ls_flags = (NFSLCK_DELEGWRITE | 3056 NFSLCK_READACCESS | NFSLCK_WRITEACCESS); 3057 *rflagsp |= NFSV4OPEN_WRITEDELEGATE; 3058 nfsrv_writedelegcnt++; 3059 } else { 3060 new_deleg->ls_flags = (NFSLCK_DELEGREAD | 3061 NFSLCK_READACCESS); 3062 *rflagsp |= NFSV4OPEN_READDELEGATE; 3063 } 3064 new_deleg->ls_uid = new_stp->ls_uid; 3065 new_deleg->ls_lfp = lfp; 3066 new_deleg->ls_clp = clp; 3067 new_deleg->ls_filerev = filerev; 3068 new_deleg->ls_compref = nd->nd_compref; 3069 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); 3070 LIST_INSERT_HEAD(NFSSTATEHASH(clp, 3071 new_deleg->ls_stateid), new_deleg, ls_hash); 3072 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); 3073 new_deleg = NULL; 3074 if (delegate == 2 || nfsrv_issuedelegs == 0 || 3075 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) != 3076 LCL_CALLBACKSON || 3077 NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) || 3078 !NFSVNO_DELEGOK(vp)) 3079 *rflagsp |= NFSV4OPEN_RECALL; 3080 nfsstatsv1.srvdelegates++; 3081 nfsrv_openpluslock++; 3082 nfsrv_delegatecnt++; 3083 3084 /* 3085 * Now, do the associated open. 3086 */ 3087 new_open->ls_stateid.seqid = 1; 3088 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; 3089 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; 3090 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); 3091 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) | 3092 NFSLCK_OPEN; 3093 if (new_stp->ls_flags & NFSLCK_DELEGWRITE) 3094 new_open->ls_flags |= (NFSLCK_READACCESS | 3095 NFSLCK_WRITEACCESS); 3096 else 3097 new_open->ls_flags |= NFSLCK_READACCESS; 3098 new_open->ls_uid = new_stp->ls_uid; 3099 new_open->ls_lfp = lfp; 3100 new_open->ls_clp = clp; 3101 LIST_INIT(&new_open->ls_open); 3102 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); 3103 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), 3104 new_open, ls_hash); 3105 /* 3106 * and handle the open owner 3107 */ 3108 if (ownerstp) { 3109 new_open->ls_openowner = ownerstp; 3110 LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list); 3111 } else { 3112 new_open->ls_openowner = new_stp; 3113 new_stp->ls_flags = 0; 3114 nfsrvd_refcache(new_stp->ls_op); 3115 new_stp->ls_noopens = 0; 3116 LIST_INIT(&new_stp->ls_open); 3117 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list); 3118 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list); 3119 *new_stpp = NULL; 3120 nfsstatsv1.srvopenowners++; 3121 nfsrv_openpluslock++; 3122 } 3123 openstp = new_open; 3124 new_open = NULL; 3125 nfsstatsv1.srvopens++; 3126 nfsrv_openpluslock++; 3127 } else { 3128 error = NFSERR_RECLAIMCONFLICT; 3129 } 3130 } else if (ownerstp) { 3131 if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) { 3132 /* Replace the open */ 3133 if (ownerstp->ls_op) 3134 nfsrvd_derefcache(ownerstp->ls_op); 3135 ownerstp->ls_op = new_stp->ls_op; 3136 nfsrvd_refcache(ownerstp->ls_op); 3137 ownerstp->ls_seq = new_stp->ls_seq; 3138 *rflagsp |= NFSV4OPEN_RESULTCONFIRM; 3139 stp = LIST_FIRST(&ownerstp->ls_open); 3140 stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) | 3141 NFSLCK_OPEN; 3142 stp->ls_stateid.seqid = 1; 3143 stp->ls_uid = new_stp->ls_uid; 3144 if (lfp != stp->ls_lfp) { 3145 LIST_REMOVE(stp, ls_file); 3146 LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file); 3147 stp->ls_lfp = lfp; 3148 } 3149 openstp = stp; 3150 } else if (openstp) { 3151 openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS); 3152 openstp->ls_stateid.seqid++; 3153 if ((nd->nd_flag & ND_NFSV41) != 0 && 3154 openstp->ls_stateid.seqid == 0) 3155 openstp->ls_stateid.seqid = 1; 3156 3157 /* 3158 * This is where we can choose to issue a delegation. 3159 */ 3160 if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0) 3161 *rflagsp |= NFSV4OPEN_WDNOTWANTED; 3162 else if (nfsrv_issuedelegs == 0) 3163 *rflagsp |= NFSV4OPEN_WDSUPPFTYPE; 3164 else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt)) 3165 *rflagsp |= NFSV4OPEN_WDRESOURCE; 3166 else if (delegate == 0 || writedeleg == 0 || 3167 NFSVNO_EXRDONLY(exp) || (readonly != 0 && 3168 nfsrv_writedelegifpos == 0) || 3169 !NFSVNO_DELEGOK(vp) || 3170 (new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 || 3171 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) != 3172 LCL_CALLBACKSON) 3173 *rflagsp |= NFSV4OPEN_WDCONTENTION; 3174 else { 3175 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; 3176 new_deleg->ls_stateid.other[0] = delegstateidp->other[0] 3177 = clp->lc_clientid.lval[0]; 3178 new_deleg->ls_stateid.other[1] = delegstateidp->other[1] 3179 = clp->lc_clientid.lval[1]; 3180 new_deleg->ls_stateid.other[2] = delegstateidp->other[2] 3181 = nfsrv_nextstateindex(clp); 3182 new_deleg->ls_flags = (NFSLCK_DELEGWRITE | 3183 NFSLCK_READACCESS | NFSLCK_WRITEACCESS); 3184 *rflagsp |= NFSV4OPEN_WRITEDELEGATE; 3185 new_deleg->ls_uid = new_stp->ls_uid; 3186 new_deleg->ls_lfp = lfp; 3187 new_deleg->ls_clp = clp; 3188 new_deleg->ls_filerev = filerev; 3189 new_deleg->ls_compref = nd->nd_compref; 3190 nfsrv_writedelegcnt++; 3191 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); 3192 LIST_INSERT_HEAD(NFSSTATEHASH(clp, 3193 new_deleg->ls_stateid), new_deleg, ls_hash); 3194 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); 3195 new_deleg = NULL; 3196 nfsstatsv1.srvdelegates++; 3197 nfsrv_openpluslock++; 3198 nfsrv_delegatecnt++; 3199 } 3200 } else { 3201 new_open->ls_stateid.seqid = 1; 3202 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; 3203 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; 3204 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); 3205 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)| 3206 NFSLCK_OPEN; 3207 new_open->ls_uid = new_stp->ls_uid; 3208 new_open->ls_openowner = ownerstp; 3209 new_open->ls_lfp = lfp; 3210 new_open->ls_clp = clp; 3211 LIST_INIT(&new_open->ls_open); 3212 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); 3213 LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list); 3214 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), 3215 new_open, ls_hash); 3216 openstp = new_open; 3217 new_open = NULL; 3218 nfsstatsv1.srvopens++; 3219 nfsrv_openpluslock++; 3220 3221 /* 3222 * This is where we can choose to issue a delegation. 3223 */ 3224 if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0) 3225 *rflagsp |= NFSV4OPEN_WDNOTWANTED; 3226 else if (nfsrv_issuedelegs == 0) 3227 *rflagsp |= NFSV4OPEN_WDSUPPFTYPE; 3228 else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt)) 3229 *rflagsp |= NFSV4OPEN_WDRESOURCE; 3230 else if (delegate == 0 || (writedeleg == 0 && 3231 readonly == 0) || !NFSVNO_DELEGOK(vp) || 3232 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) != 3233 LCL_CALLBACKSON) 3234 *rflagsp |= NFSV4OPEN_WDCONTENTION; 3235 else { 3236 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1; 3237 new_deleg->ls_stateid.other[0] = delegstateidp->other[0] 3238 = clp->lc_clientid.lval[0]; 3239 new_deleg->ls_stateid.other[1] = delegstateidp->other[1] 3240 = clp->lc_clientid.lval[1]; 3241 new_deleg->ls_stateid.other[2] = delegstateidp->other[2] 3242 = nfsrv_nextstateindex(clp); 3243 if (writedeleg && !NFSVNO_EXRDONLY(exp) && 3244 (nfsrv_writedelegifpos || !readonly) && 3245 (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) { 3246 new_deleg->ls_flags = (NFSLCK_DELEGWRITE | 3247 NFSLCK_READACCESS | NFSLCK_WRITEACCESS); 3248 *rflagsp |= NFSV4OPEN_WRITEDELEGATE; 3249 nfsrv_writedelegcnt++; 3250 } else { 3251 new_deleg->ls_flags = (NFSLCK_DELEGREAD | 3252 NFSLCK_READACCESS); 3253 *rflagsp |= NFSV4OPEN_READDELEGATE; 3254 } 3255 new_deleg->ls_uid = new_stp->ls_uid; 3256 new_deleg->ls_lfp = lfp; 3257 new_deleg->ls_clp = clp; 3258 new_deleg->ls_filerev = filerev; 3259 new_deleg->ls_compref = nd->nd_compref; 3260 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file); 3261 LIST_INSERT_HEAD(NFSSTATEHASH(clp, 3262 new_deleg->ls_stateid), new_deleg, ls_hash); 3263 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list); 3264 new_deleg = NULL; 3265 nfsstatsv1.srvdelegates++; 3266 nfsrv_openpluslock++; 3267 nfsrv_delegatecnt++; 3268 } 3269 } 3270 } else { 3271 /* 3272 * New owner case. Start the open_owner sequence with a 3273 * Needs confirmation (unless a reclaim) and hang the 3274 * new open off it. 3275 */ 3276 new_open->ls_stateid.seqid = 1; 3277 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0]; 3278 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1]; 3279 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp); 3280 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) | 3281 NFSLCK_OPEN; 3282 new_open->ls_uid = new_stp->ls_uid; 3283 LIST_INIT(&new_open->ls_open); 3284 new_open->ls_openowner = new_stp; 3285 new_open->ls_lfp = lfp; 3286 new_open->ls_clp = clp; 3287 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file); 3288 if (new_stp->ls_flags & NFSLCK_RECLAIM) { 3289 new_stp->ls_flags = 0; 3290 } else if ((nd->nd_flag & ND_NFSV41) != 0) { 3291 /* NFSv4.1 never needs confirmation. */ 3292 new_stp->ls_flags = 0; 3293 3294 /* 3295 * This is where we can choose to issue a delegation. 3296 */ 3297 if (delegate && nfsrv_issuedelegs && 3298 (writedeleg || readonly) && 3299 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) == 3300 LCL_CALLBACKSON && 3301 !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) && 3302 NFSVNO_DELEGOK(vp) && 3303 ((nd->nd_flag & ND_NFSV41) == 0 || 3304 (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) { 3305 new_deleg->ls_stateid.seqid = 3306 delegstateidp->seqid = 1; 3307 new_deleg->ls_stateid.other[0] = 3308 delegstateidp->other[0] 3309 = clp->lc_clientid.lval[0]; 3310 new_deleg->ls_stateid.other[1] = 3311 delegstateidp->other[1] 3312 = clp->lc_clientid.lval[1]; 3313 new_deleg->ls_stateid.other[2] = 3314 delegstateidp->other[2] 3315 = nfsrv_nextstateindex(clp); 3316 if (writedeleg && !NFSVNO_EXRDONLY(exp) && 3317 (nfsrv_writedelegifpos || !readonly) && 3318 ((nd->nd_flag & ND_NFSV41) == 0 || 3319 (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 3320 0)) { 3321 new_deleg->ls_flags = 3322 (NFSLCK_DELEGWRITE | 3323 NFSLCK_READACCESS | 3324 NFSLCK_WRITEACCESS); 3325 *rflagsp |= NFSV4OPEN_WRITEDELEGATE; 3326 nfsrv_writedelegcnt++; 3327 } else { 3328 new_deleg->ls_flags = 3329 (NFSLCK_DELEGREAD | 3330 NFSLCK_READACCESS); 3331 *rflagsp |= NFSV4OPEN_READDELEGATE; 3332 } 3333 new_deleg->ls_uid = new_stp->ls_uid; 3334 new_deleg->ls_lfp = lfp; 3335 new_deleg->ls_clp = clp; 3336 new_deleg->ls_filerev = filerev; 3337 new_deleg->ls_compref = nd->nd_compref; 3338 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, 3339 ls_file); 3340 LIST_INSERT_HEAD(NFSSTATEHASH(clp, 3341 new_deleg->ls_stateid), new_deleg, ls_hash); 3342 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, 3343 ls_list); 3344 new_deleg = NULL; 3345 nfsstatsv1.srvdelegates++; 3346 nfsrv_openpluslock++; 3347 nfsrv_delegatecnt++; 3348 } 3349 /* 3350 * Since NFSv4.1 never does an OpenConfirm, the first 3351 * open state will be acquired here. 3352 */ 3353 if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) { 3354 clp->lc_flags |= LCL_STAMPEDSTABLE; 3355 len = clp->lc_idlen; 3356 NFSBCOPY(clp->lc_id, clidp, len); 3357 gotstate = 1; 3358 } 3359 } else { 3360 *rflagsp |= NFSV4OPEN_RESULTCONFIRM; 3361 new_stp->ls_flags = NFSLCK_NEEDSCONFIRM; 3362 } 3363 nfsrvd_refcache(new_stp->ls_op); 3364 new_stp->ls_noopens = 0; 3365 LIST_INIT(&new_stp->ls_open); 3366 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list); 3367 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list); 3368 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid), 3369 new_open, ls_hash); 3370 openstp = new_open; 3371 new_open = NULL; 3372 *new_stpp = NULL; 3373 nfsstatsv1.srvopens++; 3374 nfsrv_openpluslock++; 3375 nfsstatsv1.srvopenowners++; 3376 nfsrv_openpluslock++; 3377 } 3378 if (!error) { 3379 stateidp->seqid = openstp->ls_stateid.seqid; 3380 stateidp->other[0] = openstp->ls_stateid.other[0]; 3381 stateidp->other[1] = openstp->ls_stateid.other[1]; 3382 stateidp->other[2] = openstp->ls_stateid.other[2]; 3383 } 3384 NFSUNLOCKSTATE(); 3385 if (haslock) { 3386 NFSLOCKV4ROOTMUTEX(); 3387 nfsv4_unlock(&nfsv4rootfs_lock, 1); 3388 NFSUNLOCKV4ROOTMUTEX(); 3389 } 3390 if (new_open) 3391 free(new_open, M_NFSDSTATE); 3392 if (new_deleg) 3393 free(new_deleg, M_NFSDSTATE); 3394 3395 /* 3396 * If the NFSv4.1 client just acquired its first open, write a timestamp 3397 * to the stable storage file. 3398 */ 3399 if (gotstate != 0) { 3400 nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p); 3401 nfsrv_backupstable(); 3402 } 3403 3404 out: 3405 free(clidp, M_TEMP); 3406 NFSEXITCODE2(error, nd); 3407 return (error); 3408 } 3409 3410 /* 3411 * Open update. Does the confirm, downgrade and close. 3412 */ 3413 APPLESTATIC int 3414 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid, 3415 nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p, 3416 int *retwriteaccessp) 3417 { 3418 struct nfsstate *stp; 3419 struct nfsclient *clp; 3420 struct nfslockfile *lfp; 3421 u_int32_t bits; 3422 int error = 0, gotstate = 0, len = 0; 3423 u_char *clidp = NULL; 3424 3425 /* 3426 * Check for restart conditions (client and server). 3427 */ 3428 error = nfsrv_checkrestart(clientid, new_stp->ls_flags, 3429 &new_stp->ls_stateid, 0); 3430 if (error) 3431 goto out; 3432 3433 clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK); 3434 NFSLOCKSTATE(); 3435 /* 3436 * Get the open structure via clientid and stateid. 3437 */ 3438 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 3439 (nfsquad_t)((u_quad_t)0), 0, nd, p); 3440 if (!error) 3441 error = nfsrv_getstate(clp, &new_stp->ls_stateid, 3442 new_stp->ls_flags, &stp); 3443 3444 /* 3445 * Sanity check the open. 3446 */ 3447 if (!error && (!(stp->ls_flags & NFSLCK_OPEN) || 3448 (!(new_stp->ls_flags & NFSLCK_CONFIRM) && 3449 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) || 3450 ((new_stp->ls_flags & NFSLCK_CONFIRM) && 3451 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))))) 3452 error = NFSERR_BADSTATEID; 3453 3454 if (!error) 3455 error = nfsrv_checkseqid(nd, new_stp->ls_seq, 3456 stp->ls_openowner, new_stp->ls_op); 3457 if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid && 3458 (((nd->nd_flag & ND_NFSV41) == 0 && 3459 !(new_stp->ls_flags & NFSLCK_CONFIRM)) || 3460 ((nd->nd_flag & ND_NFSV41) != 0 && 3461 new_stp->ls_stateid.seqid != 0))) 3462 error = NFSERR_OLDSTATEID; 3463 if (!error && vnode_vtype(vp) != VREG) { 3464 if (vnode_vtype(vp) == VDIR) 3465 error = NFSERR_ISDIR; 3466 else 3467 error = NFSERR_INVAL; 3468 } 3469 3470 if (error) { 3471 /* 3472 * If a client tries to confirm an Open with a bad 3473 * seqid# and there are no byte range locks or other Opens 3474 * on the openowner, just throw it away, so the next use of the 3475 * openowner will start a fresh seq#. 3476 */ 3477 if (error == NFSERR_BADSEQID && 3478 (new_stp->ls_flags & NFSLCK_CONFIRM) && 3479 nfsrv_nootherstate(stp)) 3480 nfsrv_freeopenowner(stp->ls_openowner, 0, p); 3481 NFSUNLOCKSTATE(); 3482 goto out; 3483 } 3484 3485 /* 3486 * Set the return stateid. 3487 */ 3488 stateidp->seqid = stp->ls_stateid.seqid + 1; 3489 if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0) 3490 stateidp->seqid = 1; 3491 stateidp->other[0] = stp->ls_stateid.other[0]; 3492 stateidp->other[1] = stp->ls_stateid.other[1]; 3493 stateidp->other[2] = stp->ls_stateid.other[2]; 3494 /* 3495 * Now, handle the three cases. 3496 */ 3497 if (new_stp->ls_flags & NFSLCK_CONFIRM) { 3498 /* 3499 * If the open doesn't need confirmation, it seems to me that 3500 * there is a client error, but I'll just log it and keep going? 3501 */ 3502 if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) 3503 printf("Nfsv4d: stray open confirm\n"); 3504 stp->ls_openowner->ls_flags = 0; 3505 stp->ls_stateid.seqid++; 3506 if ((nd->nd_flag & ND_NFSV41) != 0 && 3507 stp->ls_stateid.seqid == 0) 3508 stp->ls_stateid.seqid = 1; 3509 if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) { 3510 clp->lc_flags |= LCL_STAMPEDSTABLE; 3511 len = clp->lc_idlen; 3512 NFSBCOPY(clp->lc_id, clidp, len); 3513 gotstate = 1; 3514 } 3515 NFSUNLOCKSTATE(); 3516 } else if (new_stp->ls_flags & NFSLCK_CLOSE) { 3517 lfp = stp->ls_lfp; 3518 if (retwriteaccessp != NULL) { 3519 if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0) 3520 *retwriteaccessp = 1; 3521 else 3522 *retwriteaccessp = 0; 3523 } 3524 if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) { 3525 /* Get the lf lock */ 3526 nfsrv_locklf(lfp); 3527 NFSUNLOCKSTATE(); 3528 ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate"); 3529 NFSVOPUNLOCK(vp, 0); 3530 if (nfsrv_freeopen(stp, vp, 1, p) == 0) { 3531 NFSLOCKSTATE(); 3532 nfsrv_unlocklf(lfp); 3533 NFSUNLOCKSTATE(); 3534 } 3535 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 3536 } else { 3537 (void) nfsrv_freeopen(stp, NULL, 0, p); 3538 NFSUNLOCKSTATE(); 3539 } 3540 } else { 3541 /* 3542 * Update the share bits, making sure that the new set are a 3543 * subset of the old ones. 3544 */ 3545 bits = (new_stp->ls_flags & NFSLCK_SHAREBITS); 3546 if (~(stp->ls_flags) & bits) { 3547 NFSUNLOCKSTATE(); 3548 error = NFSERR_INVAL; 3549 goto out; 3550 } 3551 stp->ls_flags = (bits | NFSLCK_OPEN); 3552 stp->ls_stateid.seqid++; 3553 if ((nd->nd_flag & ND_NFSV41) != 0 && 3554 stp->ls_stateid.seqid == 0) 3555 stp->ls_stateid.seqid = 1; 3556 NFSUNLOCKSTATE(); 3557 } 3558 3559 /* 3560 * If the client just confirmed its first open, write a timestamp 3561 * to the stable storage file. 3562 */ 3563 if (gotstate != 0) { 3564 nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p); 3565 nfsrv_backupstable(); 3566 } 3567 3568 out: 3569 free(clidp, M_TEMP); 3570 NFSEXITCODE2(error, nd); 3571 return (error); 3572 } 3573 3574 /* 3575 * Delegation update. Does the purge and return. 3576 */ 3577 APPLESTATIC int 3578 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid, 3579 nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred, 3580 NFSPROC_T *p, int *retwriteaccessp) 3581 { 3582 struct nfsstate *stp; 3583 struct nfsclient *clp; 3584 int error = 0; 3585 fhandle_t fh; 3586 3587 /* 3588 * Do a sanity check against the file handle for DelegReturn. 3589 */ 3590 if (vp) { 3591 error = nfsvno_getfh(vp, &fh, p); 3592 if (error) 3593 goto out; 3594 } 3595 /* 3596 * Check for restart conditions (client and server). 3597 */ 3598 if (op == NFSV4OP_DELEGRETURN) 3599 error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN, 3600 stateidp, 0); 3601 else 3602 error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE, 3603 stateidp, 0); 3604 3605 NFSLOCKSTATE(); 3606 /* 3607 * Get the open structure via clientid and stateid. 3608 */ 3609 if (!error) 3610 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 3611 (nfsquad_t)((u_quad_t)0), 0, nd, p); 3612 if (error) { 3613 if (error == NFSERR_CBPATHDOWN) 3614 error = 0; 3615 if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN) 3616 error = NFSERR_STALESTATEID; 3617 } 3618 if (!error && op == NFSV4OP_DELEGRETURN) { 3619 error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp); 3620 if (!error && stp->ls_stateid.seqid != stateidp->seqid && 3621 ((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0)) 3622 error = NFSERR_OLDSTATEID; 3623 } 3624 /* 3625 * NFSERR_EXPIRED means that the state has gone away, 3626 * so Delegations have been purged. Just return ok. 3627 */ 3628 if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) { 3629 NFSUNLOCKSTATE(); 3630 error = 0; 3631 goto out; 3632 } 3633 if (error) { 3634 NFSUNLOCKSTATE(); 3635 goto out; 3636 } 3637 3638 if (op == NFSV4OP_DELEGRETURN) { 3639 if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh, 3640 sizeof (fhandle_t))) { 3641 NFSUNLOCKSTATE(); 3642 error = NFSERR_BADSTATEID; 3643 goto out; 3644 } 3645 if (retwriteaccessp != NULL) { 3646 if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0) 3647 *retwriteaccessp = 1; 3648 else 3649 *retwriteaccessp = 0; 3650 } 3651 nfsrv_freedeleg(stp); 3652 } else { 3653 nfsrv_freedeleglist(&clp->lc_olddeleg); 3654 } 3655 NFSUNLOCKSTATE(); 3656 error = 0; 3657 3658 out: 3659 NFSEXITCODE(error); 3660 return (error); 3661 } 3662 3663 /* 3664 * Release lock owner. 3665 */ 3666 APPLESTATIC int 3667 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid, 3668 NFSPROC_T *p) 3669 { 3670 struct nfsstate *stp, *nstp, *openstp, *ownstp; 3671 struct nfsclient *clp; 3672 int error = 0; 3673 3674 /* 3675 * Check for restart conditions (client and server). 3676 */ 3677 error = nfsrv_checkrestart(clientid, new_stp->ls_flags, 3678 &new_stp->ls_stateid, 0); 3679 if (error) 3680 goto out; 3681 3682 NFSLOCKSTATE(); 3683 /* 3684 * Get the lock owner by name. 3685 */ 3686 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL, 3687 (nfsquad_t)((u_quad_t)0), 0, NULL, p); 3688 if (error) { 3689 NFSUNLOCKSTATE(); 3690 goto out; 3691 } 3692 LIST_FOREACH(ownstp, &clp->lc_open, ls_list) { 3693 LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) { 3694 stp = LIST_FIRST(&openstp->ls_open); 3695 while (stp != LIST_END(&openstp->ls_open)) { 3696 nstp = LIST_NEXT(stp, ls_list); 3697 /* 3698 * If the owner matches, check for locks and 3699 * then free or return an error. 3700 */ 3701 if (stp->ls_ownerlen == new_stp->ls_ownerlen && 3702 !NFSBCMP(stp->ls_owner, new_stp->ls_owner, 3703 stp->ls_ownerlen)){ 3704 if (LIST_EMPTY(&stp->ls_lock)) { 3705 nfsrv_freelockowner(stp, NULL, 0, p); 3706 } else { 3707 NFSUNLOCKSTATE(); 3708 error = NFSERR_LOCKSHELD; 3709 goto out; 3710 } 3711 } 3712 stp = nstp; 3713 } 3714 } 3715 } 3716 NFSUNLOCKSTATE(); 3717 3718 out: 3719 NFSEXITCODE(error); 3720 return (error); 3721 } 3722 3723 /* 3724 * Get the file handle for a lock structure. 3725 */ 3726 static int 3727 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp, 3728 fhandle_t *nfhp, NFSPROC_T *p) 3729 { 3730 fhandle_t *fhp = NULL; 3731 int error; 3732 3733 /* 3734 * For lock, use the new nfslock structure, otherwise just 3735 * a fhandle_t on the stack. 3736 */ 3737 if (flags & NFSLCK_OPEN) { 3738 KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL")); 3739 fhp = &new_lfp->lf_fh; 3740 } else if (nfhp) { 3741 fhp = nfhp; 3742 } else { 3743 panic("nfsrv_getlockfh"); 3744 } 3745 error = nfsvno_getfh(vp, fhp, p); 3746 NFSEXITCODE(error); 3747 return (error); 3748 } 3749 3750 /* 3751 * Get an nfs lock structure. Allocate one, as required, and return a 3752 * pointer to it. 3753 * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock. 3754 */ 3755 static int 3756 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp, 3757 struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit) 3758 { 3759 struct nfslockfile *lfp; 3760 fhandle_t *fhp = NULL, *tfhp; 3761 struct nfslockhashhead *hp; 3762 struct nfslockfile *new_lfp = NULL; 3763 3764 /* 3765 * For lock, use the new nfslock structure, otherwise just 3766 * a fhandle_t on the stack. 3767 */ 3768 if (flags & NFSLCK_OPEN) { 3769 new_lfp = *new_lfpp; 3770 fhp = &new_lfp->lf_fh; 3771 } else if (nfhp) { 3772 fhp = nfhp; 3773 } else { 3774 panic("nfsrv_getlockfile"); 3775 } 3776 3777 hp = NFSLOCKHASH(fhp); 3778 LIST_FOREACH(lfp, hp, lf_hash) { 3779 tfhp = &lfp->lf_fh; 3780 if (NFSVNO_CMPFH(fhp, tfhp)) { 3781 if (lockit) 3782 nfsrv_locklf(lfp); 3783 *lfpp = lfp; 3784 return (0); 3785 } 3786 } 3787 if (!(flags & NFSLCK_OPEN)) 3788 return (-1); 3789 3790 /* 3791 * No match, so chain the new one into the list. 3792 */ 3793 LIST_INIT(&new_lfp->lf_open); 3794 LIST_INIT(&new_lfp->lf_lock); 3795 LIST_INIT(&new_lfp->lf_deleg); 3796 LIST_INIT(&new_lfp->lf_locallock); 3797 LIST_INIT(&new_lfp->lf_rollback); 3798 new_lfp->lf_locallock_lck.nfslock_usecnt = 0; 3799 new_lfp->lf_locallock_lck.nfslock_lock = 0; 3800 new_lfp->lf_usecount = 0; 3801 LIST_INSERT_HEAD(hp, new_lfp, lf_hash); 3802 *lfpp = new_lfp; 3803 *new_lfpp = NULL; 3804 return (0); 3805 } 3806 3807 /* 3808 * This function adds a nfslock lock structure to the list for the associated 3809 * nfsstate and nfslockfile structures. It will be inserted after the 3810 * entry pointed at by insert_lop. 3811 */ 3812 static void 3813 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop, 3814 struct nfsstate *stp, struct nfslockfile *lfp) 3815 { 3816 struct nfslock *lop, *nlop; 3817 3818 new_lop->lo_stp = stp; 3819 new_lop->lo_lfp = lfp; 3820 3821 if (stp != NULL) { 3822 /* Insert in increasing lo_first order */ 3823 lop = LIST_FIRST(&lfp->lf_lock); 3824 if (lop == LIST_END(&lfp->lf_lock) || 3825 new_lop->lo_first <= lop->lo_first) { 3826 LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile); 3827 } else { 3828 nlop = LIST_NEXT(lop, lo_lckfile); 3829 while (nlop != LIST_END(&lfp->lf_lock) && 3830 nlop->lo_first < new_lop->lo_first) { 3831 lop = nlop; 3832 nlop = LIST_NEXT(lop, lo_lckfile); 3833 } 3834 LIST_INSERT_AFTER(lop, new_lop, lo_lckfile); 3835 } 3836 } else { 3837 new_lop->lo_lckfile.le_prev = NULL; /* list not used */ 3838 } 3839 3840 /* 3841 * Insert after insert_lop, which is overloaded as stp or lfp for 3842 * an empty list. 3843 */ 3844 if (stp == NULL && (struct nfslockfile *)insert_lop == lfp) 3845 LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner); 3846 else if ((struct nfsstate *)insert_lop == stp) 3847 LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner); 3848 else 3849 LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner); 3850 if (stp != NULL) { 3851 nfsstatsv1.srvlocks++; 3852 nfsrv_openpluslock++; 3853 } 3854 } 3855 3856 /* 3857 * This function updates the locking for a lock owner and given file. It 3858 * maintains a list of lock ranges ordered on increasing file offset that 3859 * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style). 3860 * It always adds new_lop to the list and sometimes uses the one pointed 3861 * at by other_lopp. 3862 */ 3863 static void 3864 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp, 3865 struct nfslock **other_lopp, struct nfslockfile *lfp) 3866 { 3867 struct nfslock *new_lop = *new_lopp; 3868 struct nfslock *lop, *tlop, *ilop; 3869 struct nfslock *other_lop = *other_lopp; 3870 int unlock = 0, myfile = 0; 3871 u_int64_t tmp; 3872 3873 /* 3874 * Work down the list until the lock is merged. 3875 */ 3876 if (new_lop->lo_flags & NFSLCK_UNLOCK) 3877 unlock = 1; 3878 if (stp != NULL) { 3879 ilop = (struct nfslock *)stp; 3880 lop = LIST_FIRST(&stp->ls_lock); 3881 } else { 3882 ilop = (struct nfslock *)lfp; 3883 lop = LIST_FIRST(&lfp->lf_locallock); 3884 } 3885 while (lop != NULL) { 3886 /* 3887 * Only check locks for this file that aren't before the start of 3888 * new lock's range. 3889 */ 3890 if (lop->lo_lfp == lfp) { 3891 myfile = 1; 3892 if (lop->lo_end >= new_lop->lo_first) { 3893 if (new_lop->lo_end < lop->lo_first) { 3894 /* 3895 * If the new lock ends before the start of the 3896 * current lock's range, no merge, just insert 3897 * the new lock. 3898 */ 3899 break; 3900 } 3901 if (new_lop->lo_flags == lop->lo_flags || 3902 (new_lop->lo_first <= lop->lo_first && 3903 new_lop->lo_end >= lop->lo_end)) { 3904 /* 3905 * This lock can be absorbed by the new lock/unlock. 3906 * This happens when it covers the entire range 3907 * of the old lock or is contiguous 3908 * with the old lock and is of the same type or an 3909 * unlock. 3910 */ 3911 if (lop->lo_first < new_lop->lo_first) 3912 new_lop->lo_first = lop->lo_first; 3913 if (lop->lo_end > new_lop->lo_end) 3914 new_lop->lo_end = lop->lo_end; 3915 tlop = lop; 3916 lop = LIST_NEXT(lop, lo_lckowner); 3917 nfsrv_freenfslock(tlop); 3918 continue; 3919 } 3920 3921 /* 3922 * All these cases are for contiguous locks that are not the 3923 * same type, so they can't be merged. 3924 */ 3925 if (new_lop->lo_first <= lop->lo_first) { 3926 /* 3927 * This case is where the new lock overlaps with the 3928 * first part of the old lock. Move the start of the 3929 * old lock to just past the end of the new lock. The 3930 * new lock will be inserted in front of the old, since 3931 * ilop hasn't been updated. (We are done now.) 3932 */ 3933 lop->lo_first = new_lop->lo_end; 3934 break; 3935 } 3936 if (new_lop->lo_end >= lop->lo_end) { 3937 /* 3938 * This case is where the new lock overlaps with the 3939 * end of the old lock's range. Move the old lock's 3940 * end to just before the new lock's first and insert 3941 * the new lock after the old lock. 3942 * Might not be done yet, since the new lock could 3943 * overlap further locks with higher ranges. 3944 */ 3945 lop->lo_end = new_lop->lo_first; 3946 ilop = lop; 3947 lop = LIST_NEXT(lop, lo_lckowner); 3948 continue; 3949 } 3950 /* 3951 * The final case is where the new lock's range is in the 3952 * middle of the current lock's and splits the current lock 3953 * up. Use *other_lopp to handle the second part of the 3954 * split old lock range. (We are done now.) 3955 * For unlock, we use new_lop as other_lop and tmp, since 3956 * other_lop and new_lop are the same for this case. 3957 * We noted the unlock case above, so we don't need 3958 * new_lop->lo_flags any longer. 3959 */ 3960 tmp = new_lop->lo_first; 3961 if (other_lop == NULL) { 3962 if (!unlock) 3963 panic("nfsd srv update unlock"); 3964 other_lop = new_lop; 3965 *new_lopp = NULL; 3966 } 3967 other_lop->lo_first = new_lop->lo_end; 3968 other_lop->lo_end = lop->lo_end; 3969 other_lop->lo_flags = lop->lo_flags; 3970 other_lop->lo_stp = stp; 3971 other_lop->lo_lfp = lfp; 3972 lop->lo_end = tmp; 3973 nfsrv_insertlock(other_lop, lop, stp, lfp); 3974 *other_lopp = NULL; 3975 ilop = lop; 3976 break; 3977 } 3978 } 3979 ilop = lop; 3980 lop = LIST_NEXT(lop, lo_lckowner); 3981 if (myfile && (lop == NULL || lop->lo_lfp != lfp)) 3982 break; 3983 } 3984 3985 /* 3986 * Insert the new lock in the list at the appropriate place. 3987 */ 3988 if (!unlock) { 3989 nfsrv_insertlock(new_lop, ilop, stp, lfp); 3990 *new_lopp = NULL; 3991 } 3992 } 3993 3994 /* 3995 * This function handles sequencing of locks, etc. 3996 * It returns an error that indicates what the caller should do. 3997 */ 3998 static int 3999 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid, 4000 struct nfsstate *stp, struct nfsrvcache *op) 4001 { 4002 int error = 0; 4003 4004 if ((nd->nd_flag & ND_NFSV41) != 0) 4005 /* NFSv4.1 ignores the open_seqid and lock_seqid. */ 4006 goto out; 4007 if (op != nd->nd_rp) 4008 panic("nfsrvstate checkseqid"); 4009 if (!(op->rc_flag & RC_INPROG)) 4010 panic("nfsrvstate not inprog"); 4011 if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) { 4012 printf("refcnt=%d\n", stp->ls_op->rc_refcnt); 4013 panic("nfsrvstate op refcnt"); 4014 } 4015 if ((stp->ls_seq + 1) == seqid) { 4016 if (stp->ls_op) 4017 nfsrvd_derefcache(stp->ls_op); 4018 stp->ls_op = op; 4019 nfsrvd_refcache(op); 4020 stp->ls_seq = seqid; 4021 goto out; 4022 } else if (stp->ls_seq == seqid && stp->ls_op && 4023 op->rc_xid == stp->ls_op->rc_xid && 4024 op->rc_refcnt == 0 && 4025 op->rc_reqlen == stp->ls_op->rc_reqlen && 4026 op->rc_cksum == stp->ls_op->rc_cksum) { 4027 if (stp->ls_op->rc_flag & RC_INPROG) { 4028 error = NFSERR_DONTREPLY; 4029 goto out; 4030 } 4031 nd->nd_rp = stp->ls_op; 4032 nd->nd_rp->rc_flag |= RC_INPROG; 4033 nfsrvd_delcache(op); 4034 error = NFSERR_REPLYFROMCACHE; 4035 goto out; 4036 } 4037 error = NFSERR_BADSEQID; 4038 4039 out: 4040 NFSEXITCODE2(error, nd); 4041 return (error); 4042 } 4043 4044 /* 4045 * Get the client ip address for callbacks. If the strings can't be parsed, 4046 * just set lc_program to 0 to indicate no callbacks are possible. 4047 * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set 4048 * the address to the client's transport address. This won't be used 4049 * for callbacks, but can be printed out by nfsstats for info.) 4050 * Return error if the xdr can't be parsed, 0 otherwise. 4051 */ 4052 APPLESTATIC int 4053 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp) 4054 { 4055 u_int32_t *tl; 4056 u_char *cp, *cp2; 4057 int i, j, maxalen = 0, minalen = 0; 4058 sa_family_t af; 4059 #ifdef INET 4060 struct sockaddr_in *rin, *sin; 4061 #endif 4062 #ifdef INET6 4063 struct sockaddr_in6 *rin6, *sin6; 4064 #endif 4065 u_char *addr; 4066 int error = 0, cantparse = 0; 4067 union { 4068 in_addr_t ival; 4069 u_char cval[4]; 4070 } ip; 4071 union { 4072 in_port_t sval; 4073 u_char cval[2]; 4074 } port; 4075 4076 /* 8 is the maximum length of the port# string. */ 4077 addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK); 4078 clp->lc_req.nr_client = NULL; 4079 clp->lc_req.nr_lock = 0; 4080 af = AF_UNSPEC; 4081 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 4082 i = fxdr_unsigned(int, *tl); 4083 if (i >= 3 && i <= 4) { 4084 error = nfsrv_mtostr(nd, addr, i); 4085 if (error) 4086 goto nfsmout; 4087 #ifdef INET 4088 if (!strcmp(addr, "tcp")) { 4089 clp->lc_flags |= LCL_TCPCALLBACK; 4090 clp->lc_req.nr_sotype = SOCK_STREAM; 4091 clp->lc_req.nr_soproto = IPPROTO_TCP; 4092 af = AF_INET; 4093 } else if (!strcmp(addr, "udp")) { 4094 clp->lc_req.nr_sotype = SOCK_DGRAM; 4095 clp->lc_req.nr_soproto = IPPROTO_UDP; 4096 af = AF_INET; 4097 } 4098 #endif 4099 #ifdef INET6 4100 if (af == AF_UNSPEC) { 4101 if (!strcmp(addr, "tcp6")) { 4102 clp->lc_flags |= LCL_TCPCALLBACK; 4103 clp->lc_req.nr_sotype = SOCK_STREAM; 4104 clp->lc_req.nr_soproto = IPPROTO_TCP; 4105 af = AF_INET6; 4106 } else if (!strcmp(addr, "udp6")) { 4107 clp->lc_req.nr_sotype = SOCK_DGRAM; 4108 clp->lc_req.nr_soproto = IPPROTO_UDP; 4109 af = AF_INET6; 4110 } 4111 } 4112 #endif 4113 if (af == AF_UNSPEC) { 4114 cantparse = 1; 4115 } 4116 } else { 4117 cantparse = 1; 4118 if (i > 0) { 4119 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 4120 if (error) 4121 goto nfsmout; 4122 } 4123 } 4124 /* 4125 * The caller has allocated clp->lc_req.nr_nam to be large enough 4126 * for either AF_INET or AF_INET6 and zeroed out the contents. 4127 * maxalen is set to the maximum length of the host IP address string 4128 * plus 8 for the maximum length of the port#. 4129 * minalen is set to the minimum length of the host IP address string 4130 * plus 4 for the minimum length of the port#. 4131 * These lengths do not include NULL termination, 4132 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations. 4133 */ 4134 switch (af) { 4135 #ifdef INET 4136 case AF_INET: 4137 rin = (struct sockaddr_in *)clp->lc_req.nr_nam; 4138 rin->sin_family = AF_INET; 4139 rin->sin_len = sizeof(struct sockaddr_in); 4140 maxalen = INET_ADDRSTRLEN - 1 + 8; 4141 minalen = 7 + 4; 4142 break; 4143 #endif 4144 #ifdef INET6 4145 case AF_INET6: 4146 rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; 4147 rin6->sin6_family = AF_INET6; 4148 rin6->sin6_len = sizeof(struct sockaddr_in6); 4149 maxalen = INET6_ADDRSTRLEN - 1 + 8; 4150 minalen = 3 + 4; 4151 break; 4152 #endif 4153 } 4154 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED); 4155 i = fxdr_unsigned(int, *tl); 4156 if (i < 0) { 4157 error = NFSERR_BADXDR; 4158 goto nfsmout; 4159 } else if (i == 0) { 4160 cantparse = 1; 4161 } else if (!cantparse && i <= maxalen && i >= minalen) { 4162 error = nfsrv_mtostr(nd, addr, i); 4163 if (error) 4164 goto nfsmout; 4165 4166 /* 4167 * Parse out the address fields. We expect 6 decimal numbers 4168 * separated by '.'s for AF_INET and two decimal numbers 4169 * preceeded by '.'s for AF_INET6. 4170 */ 4171 cp = NULL; 4172 switch (af) { 4173 #ifdef INET6 4174 /* 4175 * For AF_INET6, first parse the host address. 4176 */ 4177 case AF_INET6: 4178 cp = strchr(addr, '.'); 4179 if (cp != NULL) { 4180 *cp++ = '\0'; 4181 if (inet_pton(af, addr, &rin6->sin6_addr) == 1) 4182 i = 4; 4183 else { 4184 cp = NULL; 4185 cantparse = 1; 4186 } 4187 } 4188 break; 4189 #endif 4190 #ifdef INET 4191 case AF_INET: 4192 cp = addr; 4193 i = 0; 4194 break; 4195 #endif 4196 } 4197 while (cp != NULL && *cp && i < 6) { 4198 cp2 = cp; 4199 while (*cp2 && *cp2 != '.') 4200 cp2++; 4201 if (*cp2) 4202 *cp2++ = '\0'; 4203 else if (i != 5) { 4204 cantparse = 1; 4205 break; 4206 } 4207 j = nfsrv_getipnumber(cp); 4208 if (j >= 0) { 4209 if (i < 4) 4210 ip.cval[3 - i] = j; 4211 else 4212 port.cval[5 - i] = j; 4213 } else { 4214 cantparse = 1; 4215 break; 4216 } 4217 cp = cp2; 4218 i++; 4219 } 4220 if (!cantparse) { 4221 /* 4222 * The host address INADDR_ANY is (mis)used to indicate 4223 * "there is no valid callback address". 4224 */ 4225 switch (af) { 4226 #ifdef INET6 4227 case AF_INET6: 4228 if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr, 4229 &in6addr_any)) 4230 rin6->sin6_port = htons(port.sval); 4231 else 4232 cantparse = 1; 4233 break; 4234 #endif 4235 #ifdef INET 4236 case AF_INET: 4237 if (ip.ival != INADDR_ANY) { 4238 rin->sin_addr.s_addr = htonl(ip.ival); 4239 rin->sin_port = htons(port.sval); 4240 } else { 4241 cantparse = 1; 4242 } 4243 break; 4244 #endif 4245 } 4246 } 4247 } else { 4248 cantparse = 1; 4249 if (i > 0) { 4250 error = nfsm_advance(nd, NFSM_RNDUP(i), -1); 4251 if (error) 4252 goto nfsmout; 4253 } 4254 } 4255 if (cantparse) { 4256 switch (nd->nd_nam->sa_family) { 4257 #ifdef INET 4258 case AF_INET: 4259 sin = (struct sockaddr_in *)nd->nd_nam; 4260 rin = (struct sockaddr_in *)clp->lc_req.nr_nam; 4261 rin->sin_family = AF_INET; 4262 rin->sin_len = sizeof(struct sockaddr_in); 4263 rin->sin_addr.s_addr = sin->sin_addr.s_addr; 4264 rin->sin_port = 0x0; 4265 break; 4266 #endif 4267 #ifdef INET6 4268 case AF_INET6: 4269 sin6 = (struct sockaddr_in6 *)nd->nd_nam; 4270 rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam; 4271 rin6->sin6_family = AF_INET6; 4272 rin6->sin6_len = sizeof(struct sockaddr_in6); 4273 rin6->sin6_addr = sin6->sin6_addr; 4274 rin6->sin6_port = 0x0; 4275 break; 4276 #endif 4277 } 4278 clp->lc_program = 0; 4279 } 4280 nfsmout: 4281 free(addr, M_TEMP); 4282 NFSEXITCODE2(error, nd); 4283 return (error); 4284 } 4285 4286 /* 4287 * Turn a string of up to three decimal digits into a number. Return -1 upon 4288 * error. 4289 */ 4290 static int 4291 nfsrv_getipnumber(u_char *cp) 4292 { 4293 int i = 0, j = 0; 4294 4295 while (*cp) { 4296 if (j > 2 || *cp < '0' || *cp > '9') 4297 return (-1); 4298 i *= 10; 4299 i += (*cp - '0'); 4300 cp++; 4301 j++; 4302 } 4303 if (i < 256) 4304 return (i); 4305 return (-1); 4306 } 4307 4308 /* 4309 * This function checks for restart conditions. 4310 */ 4311 static int 4312 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags, 4313 nfsv4stateid_t *stateidp, int specialid) 4314 { 4315 int ret = 0; 4316 4317 /* 4318 * First check for a server restart. Open, LockT, ReleaseLockOwner 4319 * and DelegPurge have a clientid, the rest a stateid. 4320 */ 4321 if (flags & 4322 (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) { 4323 if (clientid.lval[0] != nfsrvboottime) { 4324 ret = NFSERR_STALECLIENTID; 4325 goto out; 4326 } 4327 } else if (stateidp->other[0] != nfsrvboottime && 4328 specialid == 0) { 4329 ret = NFSERR_STALESTATEID; 4330 goto out; 4331 } 4332 4333 /* 4334 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do 4335 * not use a lock/open owner seqid#, so the check can be done now. 4336 * (The others will be checked, as required, later.) 4337 */ 4338 if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST))) 4339 goto out; 4340 4341 NFSLOCKSTATE(); 4342 ret = nfsrv_checkgrace(NULL, NULL, flags); 4343 NFSUNLOCKSTATE(); 4344 4345 out: 4346 NFSEXITCODE(ret); 4347 return (ret); 4348 } 4349 4350 /* 4351 * Check for grace. 4352 */ 4353 static int 4354 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp, 4355 u_int32_t flags) 4356 { 4357 int error = 0, notreclaimed; 4358 struct nfsrv_stable *sp; 4359 4360 if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE | 4361 NFSNSF_GRACEOVER)) == 0) { 4362 /* 4363 * First, check to see if all of the clients have done a 4364 * ReclaimComplete. If so, grace can end now. 4365 */ 4366 notreclaimed = 0; 4367 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { 4368 if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) { 4369 notreclaimed = 1; 4370 break; 4371 } 4372 } 4373 if (notreclaimed == 0) 4374 nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER | 4375 NFSNSF_NEEDLOCK); 4376 } 4377 4378 if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) { 4379 if (flags & NFSLCK_RECLAIM) { 4380 error = NFSERR_NOGRACE; 4381 goto out; 4382 } 4383 } else { 4384 if (!(flags & NFSLCK_RECLAIM)) { 4385 error = NFSERR_GRACE; 4386 goto out; 4387 } 4388 if (nd != NULL && clp != NULL && 4389 (nd->nd_flag & ND_NFSV41) != 0 && 4390 (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) { 4391 error = NFSERR_NOGRACE; 4392 goto out; 4393 } 4394 4395 /* 4396 * If grace is almost over and we are still getting Reclaims, 4397 * extend grace a bit. 4398 */ 4399 if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) > 4400 nfsrv_stablefirst.nsf_eograce) 4401 nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC + 4402 NFSRV_LEASEDELTA; 4403 } 4404 4405 out: 4406 NFSEXITCODE(error); 4407 return (error); 4408 } 4409 4410 /* 4411 * Do a server callback. 4412 * The "trunc" argument is slightly overloaded and refers to different 4413 * boolean arguments for CBRECALL and CBLAYOUTRECALL. 4414 */ 4415 static int 4416 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp, 4417 int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp, 4418 int laytype, NFSPROC_T *p) 4419 { 4420 mbuf_t m; 4421 u_int32_t *tl; 4422 struct nfsrv_descript *nd; 4423 struct ucred *cred; 4424 int error = 0; 4425 u_int32_t callback; 4426 struct nfsdsession *sep = NULL; 4427 uint64_t tval; 4428 4429 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO); 4430 cred = newnfs_getcred(); 4431 NFSLOCKSTATE(); /* mostly for lc_cbref++ */ 4432 if (clp->lc_flags & LCL_NEEDSCONFIRM) { 4433 NFSUNLOCKSTATE(); 4434 panic("docallb"); 4435 } 4436 clp->lc_cbref++; 4437 4438 /* 4439 * Fill the callback program# and version into the request 4440 * structure for newnfs_connect() to use. 4441 */ 4442 clp->lc_req.nr_prog = clp->lc_program; 4443 #ifdef notnow 4444 if ((clp->lc_flags & LCL_NFSV41) != 0) 4445 clp->lc_req.nr_vers = NFSV41_CBVERS; 4446 else 4447 #endif 4448 clp->lc_req.nr_vers = NFSV4_CBVERS; 4449 4450 /* 4451 * First, fill in some of the fields of nd and cr. 4452 */ 4453 nd->nd_flag = ND_NFSV4; 4454 if (clp->lc_flags & LCL_GSS) 4455 nd->nd_flag |= ND_KERBV; 4456 if ((clp->lc_flags & LCL_NFSV41) != 0) 4457 nd->nd_flag |= ND_NFSV41; 4458 nd->nd_repstat = 0; 4459 cred->cr_uid = clp->lc_uid; 4460 cred->cr_gid = clp->lc_gid; 4461 callback = clp->lc_callback; 4462 NFSUNLOCKSTATE(); 4463 cred->cr_ngroups = 1; 4464 4465 /* 4466 * Get the first mbuf for the request. 4467 */ 4468 MGET(m, M_WAITOK, MT_DATA); 4469 mbuf_setlen(m, 0); 4470 nd->nd_mreq = nd->nd_mb = m; 4471 nd->nd_bpos = NFSMTOD(m, caddr_t); 4472 4473 /* 4474 * and build the callback request. 4475 */ 4476 if (procnum == NFSV4OP_CBGETATTR) { 4477 nd->nd_procnum = NFSV4PROC_CBCOMPOUND; 4478 error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR, 4479 "CB Getattr", &sep); 4480 if (error != 0) { 4481 mbuf_freem(nd->nd_mreq); 4482 goto errout; 4483 } 4484 (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0); 4485 (void)nfsrv_putattrbit(nd, attrbitp); 4486 } else if (procnum == NFSV4OP_CBRECALL) { 4487 nd->nd_procnum = NFSV4PROC_CBCOMPOUND; 4488 error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL, 4489 "CB Recall", &sep); 4490 if (error != 0) { 4491 mbuf_freem(nd->nd_mreq); 4492 goto errout; 4493 } 4494 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID); 4495 *tl++ = txdr_unsigned(stateidp->seqid); 4496 NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl, 4497 NFSX_STATEIDOTHER); 4498 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED); 4499 if (trunc) 4500 *tl = newnfs_true; 4501 else 4502 *tl = newnfs_false; 4503 (void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0); 4504 } else if (procnum == NFSV4OP_CBLAYOUTRECALL) { 4505 NFSD_DEBUG(4, "docallback layout recall\n"); 4506 nd->nd_procnum = NFSV4PROC_CBCOMPOUND; 4507 error = nfsrv_cbcallargs(nd, clp, callback, 4508 NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep); 4509 NFSD_DEBUG(4, "aft cbcallargs=%d\n", error); 4510 if (error != 0) { 4511 mbuf_freem(nd->nd_mreq); 4512 goto errout; 4513 } 4514 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED); 4515 *tl++ = txdr_unsigned(laytype); 4516 *tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY); 4517 if (trunc) 4518 *tl++ = newnfs_true; 4519 else 4520 *tl++ = newnfs_false; 4521 *tl = txdr_unsigned(NFSV4LAYOUTRET_FILE); 4522 nfsm_fhtom(nd, (uint8_t *)fhp, NFSX_MYFH, 0); 4523 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID); 4524 tval = 0; 4525 txdr_hyper(tval, tl); tl += 2; 4526 tval = UINT64_MAX; 4527 txdr_hyper(tval, tl); tl += 2; 4528 *tl++ = txdr_unsigned(stateidp->seqid); 4529 NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER); 4530 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED); 4531 NFSD_DEBUG(4, "aft args\n"); 4532 } else if (procnum == NFSV4PROC_CBNULL) { 4533 nd->nd_procnum = NFSV4PROC_CBNULL; 4534 if ((clp->lc_flags & LCL_NFSV41) != 0) { 4535 error = nfsv4_getcbsession(clp, &sep); 4536 if (error != 0) { 4537 mbuf_freem(nd->nd_mreq); 4538 goto errout; 4539 } 4540 } 4541 } else { 4542 error = NFSERR_SERVERFAULT; 4543 mbuf_freem(nd->nd_mreq); 4544 goto errout; 4545 } 4546 4547 /* 4548 * Call newnfs_connect(), as required, and then newnfs_request(). 4549 */ 4550 (void) newnfs_sndlock(&clp->lc_req.nr_lock); 4551 if (clp->lc_req.nr_client == NULL) { 4552 if ((clp->lc_flags & LCL_NFSV41) != 0) { 4553 error = ECONNREFUSED; 4554 nfsrv_freesession(sep, NULL); 4555 } else if (nd->nd_procnum == NFSV4PROC_CBNULL) 4556 error = newnfs_connect(NULL, &clp->lc_req, cred, 4557 NULL, 1); 4558 else 4559 error = newnfs_connect(NULL, &clp->lc_req, cred, 4560 NULL, 3); 4561 } 4562 newnfs_sndunlock(&clp->lc_req.nr_lock); 4563 NFSD_DEBUG(4, "aft sndunlock=%d\n", error); 4564 if (!error) { 4565 if ((nd->nd_flag & ND_NFSV41) != 0) { 4566 KASSERT(sep != NULL, ("sep NULL")); 4567 if (sep->sess_cbsess.nfsess_xprt != NULL) 4568 error = newnfs_request(nd, NULL, clp, 4569 &clp->lc_req, NULL, NULL, cred, 4570 clp->lc_program, clp->lc_req.nr_vers, NULL, 4571 1, NULL, &sep->sess_cbsess); 4572 else { 4573 /* 4574 * This should probably never occur, but if a 4575 * client somehow does an RPC without a 4576 * SequenceID Op that causes a callback just 4577 * after the nfsd threads have been terminated 4578 * and restared we could conceivably get here 4579 * without a backchannel xprt. 4580 */ 4581 printf("nfsrv_docallback: no xprt\n"); 4582 error = ECONNREFUSED; 4583 } 4584 NFSD_DEBUG(4, "aft newnfs_request=%d\n", error); 4585 nfsrv_freesession(sep, NULL); 4586 } else 4587 error = newnfs_request(nd, NULL, clp, &clp->lc_req, 4588 NULL, NULL, cred, clp->lc_program, 4589 clp->lc_req.nr_vers, NULL, 1, NULL, NULL); 4590 } 4591 errout: 4592 NFSFREECRED(cred); 4593 4594 /* 4595 * If error is set here, the Callback path isn't working 4596 * properly, so twiddle the appropriate LCL_ flags. 4597 * (nd_repstat != 0 indicates the Callback path is working, 4598 * but the callback failed on the client.) 4599 */ 4600 if (error) { 4601 /* 4602 * Mark the callback pathway down, which disabled issuing 4603 * of delegations and gets Renew to return NFSERR_CBPATHDOWN. 4604 */ 4605 NFSLOCKSTATE(); 4606 clp->lc_flags |= LCL_CBDOWN; 4607 NFSUNLOCKSTATE(); 4608 } else { 4609 /* 4610 * Callback worked. If the callback path was down, disable 4611 * callbacks, so no more delegations will be issued. (This 4612 * is done on the assumption that the callback pathway is 4613 * flakey.) 4614 */ 4615 NFSLOCKSTATE(); 4616 if (clp->lc_flags & LCL_CBDOWN) 4617 clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON); 4618 NFSUNLOCKSTATE(); 4619 if (nd->nd_repstat) { 4620 error = nd->nd_repstat; 4621 NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n", 4622 procnum, error); 4623 } else if (error == 0 && procnum == NFSV4OP_CBGETATTR) 4624 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, 4625 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, 4626 p, NULL); 4627 mbuf_freem(nd->nd_mrep); 4628 } 4629 NFSLOCKSTATE(); 4630 clp->lc_cbref--; 4631 if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) { 4632 clp->lc_flags &= ~LCL_WAKEUPWANTED; 4633 wakeup(clp); 4634 } 4635 NFSUNLOCKSTATE(); 4636 4637 free(nd, M_TEMP); 4638 NFSEXITCODE(error); 4639 return (error); 4640 } 4641 4642 /* 4643 * Set up the compound RPC for the callback. 4644 */ 4645 static int 4646 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp, 4647 uint32_t callback, int op, const char *optag, struct nfsdsession **sepp) 4648 { 4649 uint32_t *tl; 4650 int error, len; 4651 4652 len = strlen(optag); 4653 (void)nfsm_strtom(nd, optag, len); 4654 NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED); 4655 if ((nd->nd_flag & ND_NFSV41) != 0) { 4656 *tl++ = txdr_unsigned(NFSV41_MINORVERSION); 4657 *tl++ = txdr_unsigned(callback); 4658 *tl++ = txdr_unsigned(2); 4659 *tl = txdr_unsigned(NFSV4OP_CBSEQUENCE); 4660 error = nfsv4_setcbsequence(nd, clp, 1, sepp); 4661 if (error != 0) 4662 return (error); 4663 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED); 4664 *tl = txdr_unsigned(op); 4665 } else { 4666 *tl++ = txdr_unsigned(NFSV4_MINORVERSION); 4667 *tl++ = txdr_unsigned(callback); 4668 *tl++ = txdr_unsigned(1); 4669 *tl = txdr_unsigned(op); 4670 } 4671 return (0); 4672 } 4673 4674 /* 4675 * Return the next index# for a clientid. Mostly just increment and return 4676 * the next one, but... if the 32bit unsigned does actually wrap around, 4677 * it should be rebooted. 4678 * At an average rate of one new client per second, it will wrap around in 4679 * approximately 136 years. (I think the server will have been shut 4680 * down or rebooted before then.) 4681 */ 4682 static u_int32_t 4683 nfsrv_nextclientindex(void) 4684 { 4685 static u_int32_t client_index = 0; 4686 4687 client_index++; 4688 if (client_index != 0) 4689 return (client_index); 4690 4691 printf("%s: out of clientids\n", __func__); 4692 return (client_index); 4693 } 4694 4695 /* 4696 * Return the next index# for a stateid. Mostly just increment and return 4697 * the next one, but... if the 32bit unsigned does actually wrap around 4698 * (will a BSD server stay up that long?), find 4699 * new start and end values. 4700 */ 4701 static u_int32_t 4702 nfsrv_nextstateindex(struct nfsclient *clp) 4703 { 4704 struct nfsstate *stp; 4705 int i; 4706 u_int32_t canuse, min_index, max_index; 4707 4708 if (!(clp->lc_flags & LCL_INDEXNOTOK)) { 4709 clp->lc_stateindex++; 4710 if (clp->lc_stateindex != clp->lc_statemaxindex) 4711 return (clp->lc_stateindex); 4712 } 4713 4714 /* 4715 * Yuck, we've hit the end. 4716 * Look for a new min and max. 4717 */ 4718 min_index = 0; 4719 max_index = 0xffffffff; 4720 for (i = 0; i < nfsrv_statehashsize; i++) { 4721 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) { 4722 if (stp->ls_stateid.other[2] > 0x80000000) { 4723 if (stp->ls_stateid.other[2] < max_index) 4724 max_index = stp->ls_stateid.other[2]; 4725 } else { 4726 if (stp->ls_stateid.other[2] > min_index) 4727 min_index = stp->ls_stateid.other[2]; 4728 } 4729 } 4730 } 4731 4732 /* 4733 * Yikes, highly unlikely, but I'll handle it anyhow. 4734 */ 4735 if (min_index == 0x80000000 && max_index == 0x80000001) { 4736 canuse = 0; 4737 /* 4738 * Loop around until we find an unused entry. Return that 4739 * and set LCL_INDEXNOTOK, so the search will continue next time. 4740 * (This is one of those rare cases where a goto is the 4741 * cleanest way to code the loop.) 4742 */ 4743 tryagain: 4744 for (i = 0; i < nfsrv_statehashsize; i++) { 4745 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) { 4746 if (stp->ls_stateid.other[2] == canuse) { 4747 canuse++; 4748 goto tryagain; 4749 } 4750 } 4751 } 4752 clp->lc_flags |= LCL_INDEXNOTOK; 4753 return (canuse); 4754 } 4755 4756 /* 4757 * Ok to start again from min + 1. 4758 */ 4759 clp->lc_stateindex = min_index + 1; 4760 clp->lc_statemaxindex = max_index; 4761 clp->lc_flags &= ~LCL_INDEXNOTOK; 4762 return (clp->lc_stateindex); 4763 } 4764 4765 /* 4766 * The following functions handle the stable storage file that deals with 4767 * the edge conditions described in RFC3530 Sec. 8.6.3. 4768 * The file is as follows: 4769 * - a single record at the beginning that has the lease time of the 4770 * previous server instance (before the last reboot) and the nfsrvboottime 4771 * values for the previous server boots. 4772 * These previous boot times are used to ensure that the current 4773 * nfsrvboottime does not, somehow, get set to a previous one. 4774 * (This is important so that Stale ClientIDs and StateIDs can 4775 * be recognized.) 4776 * The number of previous nfsvrboottime values precedes the list. 4777 * - followed by some number of appended records with: 4778 * - client id string 4779 * - flag that indicates it is a record revoking state via lease 4780 * expiration or similar 4781 * OR has successfully acquired state. 4782 * These structures vary in length, with the client string at the end, up 4783 * to NFSV4_OPAQUELIMIT in size. 4784 * 4785 * At the end of the grace period, the file is truncated, the first 4786 * record is rewritten with updated information and any acquired state 4787 * records for successful reclaims of state are written. 4788 * 4789 * Subsequent records are appended when the first state is issued to 4790 * a client and when state is revoked for a client. 4791 * 4792 * When reading the file in, state issued records that come later in 4793 * the file override older ones, since the append log is in cronological order. 4794 * If, for some reason, the file can't be read, the grace period is 4795 * immediately terminated and all reclaims get NFSERR_NOGRACE. 4796 */ 4797 4798 /* 4799 * Read in the stable storage file. Called by nfssvc() before the nfsd 4800 * processes start servicing requests. 4801 */ 4802 APPLESTATIC void 4803 nfsrv_setupstable(NFSPROC_T *p) 4804 { 4805 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; 4806 struct nfsrv_stable *sp, *nsp; 4807 struct nfst_rec *tsp; 4808 int error, i, tryagain; 4809 off_t off = 0; 4810 ssize_t aresid, len; 4811 4812 /* 4813 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without 4814 * a reboot, so state has not been lost. 4815 */ 4816 if (sf->nsf_flags & NFSNSF_UPDATEDONE) 4817 return; 4818 /* 4819 * Set Grace over just until the file reads successfully. 4820 */ 4821 nfsrvboottime = time_second; 4822 LIST_INIT(&sf->nsf_head); 4823 sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK); 4824 sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA; 4825 if (sf->nsf_fp == NULL) 4826 return; 4827 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp), 4828 (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE, 4829 0, NFSFPCRED(sf->nsf_fp), &aresid, p); 4830 if (error || aresid || sf->nsf_numboots == 0 || 4831 sf->nsf_numboots > NFSNSF_MAXNUMBOOTS) 4832 return; 4833 4834 /* 4835 * Now, read in the boottimes. 4836 */ 4837 sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) * 4838 sizeof (time_t), M_TEMP, M_WAITOK); 4839 off = sizeof (struct nfsf_rec); 4840 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp), 4841 (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off, 4842 UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p); 4843 if (error || aresid) { 4844 free(sf->nsf_bootvals, M_TEMP); 4845 sf->nsf_bootvals = NULL; 4846 return; 4847 } 4848 4849 /* 4850 * Make sure this nfsrvboottime is different from all recorded 4851 * previous ones. 4852 */ 4853 do { 4854 tryagain = 0; 4855 for (i = 0; i < sf->nsf_numboots; i++) { 4856 if (nfsrvboottime == sf->nsf_bootvals[i]) { 4857 nfsrvboottime++; 4858 tryagain = 1; 4859 break; 4860 } 4861 } 4862 } while (tryagain); 4863 4864 sf->nsf_flags |= NFSNSF_OK; 4865 off += (sf->nsf_numboots * sizeof (time_t)); 4866 4867 /* 4868 * Read through the file, building a list of records for grace 4869 * checking. 4870 * Each record is between sizeof (struct nfst_rec) and 4871 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1 4872 * and is actually sizeof (struct nfst_rec) + nst_len - 1. 4873 */ 4874 tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) + 4875 NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK); 4876 do { 4877 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp), 4878 (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1, 4879 off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p); 4880 len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid; 4881 if (error || (len > 0 && (len < sizeof (struct nfst_rec) || 4882 len < (sizeof (struct nfst_rec) + tsp->len - 1)))) { 4883 /* 4884 * Yuck, the file has been corrupted, so just return 4885 * after clearing out any restart state, so the grace period 4886 * is over. 4887 */ 4888 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) { 4889 LIST_REMOVE(sp, nst_list); 4890 free(sp, M_TEMP); 4891 } 4892 free(tsp, M_TEMP); 4893 sf->nsf_flags &= ~NFSNSF_OK; 4894 free(sf->nsf_bootvals, M_TEMP); 4895 sf->nsf_bootvals = NULL; 4896 return; 4897 } 4898 if (len > 0) { 4899 off += sizeof (struct nfst_rec) + tsp->len - 1; 4900 /* 4901 * Search the list for a matching client. 4902 */ 4903 LIST_FOREACH(sp, &sf->nsf_head, nst_list) { 4904 if (tsp->len == sp->nst_len && 4905 !NFSBCMP(tsp->client, sp->nst_client, tsp->len)) 4906 break; 4907 } 4908 if (sp == LIST_END(&sf->nsf_head)) { 4909 sp = (struct nfsrv_stable *)malloc(tsp->len + 4910 sizeof (struct nfsrv_stable) - 1, M_TEMP, 4911 M_WAITOK); 4912 NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec, 4913 sizeof (struct nfst_rec) + tsp->len - 1); 4914 LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list); 4915 } else { 4916 if (tsp->flag == NFSNST_REVOKE) 4917 sp->nst_flag |= NFSNST_REVOKE; 4918 else 4919 /* 4920 * A subsequent timestamp indicates the client 4921 * did a setclientid/confirm and any previous 4922 * revoke is no longer relevant. 4923 */ 4924 sp->nst_flag &= ~NFSNST_REVOKE; 4925 } 4926 } 4927 } while (len > 0); 4928 free(tsp, M_TEMP); 4929 sf->nsf_flags = NFSNSF_OK; 4930 sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease + 4931 NFSRV_LEASEDELTA; 4932 } 4933 4934 /* 4935 * Update the stable storage file, now that the grace period is over. 4936 */ 4937 APPLESTATIC void 4938 nfsrv_updatestable(NFSPROC_T *p) 4939 { 4940 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; 4941 struct nfsrv_stable *sp, *nsp; 4942 int i; 4943 struct nfsvattr nva; 4944 vnode_t vp; 4945 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000) 4946 mount_t mp = NULL; 4947 #endif 4948 int error; 4949 4950 if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE)) 4951 return; 4952 sf->nsf_flags |= NFSNSF_UPDATEDONE; 4953 /* 4954 * Ok, we need to rewrite the stable storage file. 4955 * - truncate to 0 length 4956 * - write the new first structure 4957 * - loop through the data structures, writing out any that 4958 * have timestamps older than the old boot 4959 */ 4960 if (sf->nsf_bootvals) { 4961 sf->nsf_numboots++; 4962 for (i = sf->nsf_numboots - 2; i >= 0; i--) 4963 sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i]; 4964 } else { 4965 sf->nsf_numboots = 1; 4966 sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t), 4967 M_TEMP, M_WAITOK); 4968 } 4969 sf->nsf_bootvals[0] = nfsrvboottime; 4970 sf->nsf_lease = nfsrv_lease; 4971 NFSVNO_ATTRINIT(&nva); 4972 NFSVNO_SETATTRVAL(&nva, size, 0); 4973 vp = NFSFPVNODE(sf->nsf_fp); 4974 vn_start_write(vp, &mp, V_WAIT); 4975 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) { 4976 error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p, 4977 NULL); 4978 NFSVOPUNLOCK(vp, 0); 4979 } else 4980 error = EPERM; 4981 vn_finished_write(mp); 4982 if (!error) 4983 error = NFSD_RDWR(UIO_WRITE, vp, 4984 (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0, 4985 UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p); 4986 if (!error) 4987 error = NFSD_RDWR(UIO_WRITE, vp, 4988 (caddr_t)sf->nsf_bootvals, 4989 sf->nsf_numboots * sizeof (time_t), 4990 (off_t)(sizeof (struct nfsf_rec)), 4991 UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p); 4992 free(sf->nsf_bootvals, M_TEMP); 4993 sf->nsf_bootvals = NULL; 4994 if (error) { 4995 sf->nsf_flags &= ~NFSNSF_OK; 4996 printf("EEK! Can't write NfsV4 stable storage file\n"); 4997 return; 4998 } 4999 sf->nsf_flags |= NFSNSF_OK; 5000 5001 /* 5002 * Loop through the list and write out timestamp records for 5003 * any clients that successfully reclaimed state. 5004 */ 5005 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) { 5006 if (sp->nst_flag & NFSNST_GOTSTATE) { 5007 nfsrv_writestable(sp->nst_client, sp->nst_len, 5008 NFSNST_NEWSTATE, p); 5009 sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE; 5010 } 5011 LIST_REMOVE(sp, nst_list); 5012 free(sp, M_TEMP); 5013 } 5014 nfsrv_backupstable(); 5015 } 5016 5017 /* 5018 * Append a record to the stable storage file. 5019 */ 5020 APPLESTATIC void 5021 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p) 5022 { 5023 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst; 5024 struct nfst_rec *sp; 5025 int error; 5026 5027 if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL) 5028 return; 5029 sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) + 5030 len - 1, M_TEMP, M_WAITOK); 5031 sp->len = len; 5032 NFSBCOPY(client, sp->client, len); 5033 sp->flag = flag; 5034 error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp), 5035 (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0, 5036 UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p); 5037 free(sp, M_TEMP); 5038 if (error) { 5039 sf->nsf_flags &= ~NFSNSF_OK; 5040 printf("EEK! Can't write NfsV4 stable storage file\n"); 5041 } 5042 } 5043 5044 /* 5045 * This function is called during the grace period to mark a client 5046 * that successfully reclaimed state. 5047 */ 5048 static void 5049 nfsrv_markstable(struct nfsclient *clp) 5050 { 5051 struct nfsrv_stable *sp; 5052 5053 /* 5054 * First find the client structure. 5055 */ 5056 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { 5057 if (sp->nst_len == clp->lc_idlen && 5058 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len)) 5059 break; 5060 } 5061 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head)) 5062 return; 5063 5064 /* 5065 * Now, just mark it and set the nfsclient back pointer. 5066 */ 5067 sp->nst_flag |= NFSNST_GOTSTATE; 5068 sp->nst_clp = clp; 5069 } 5070 5071 /* 5072 * This function is called when a NFSv4.1 client does a ReclaimComplete. 5073 * Very similar to nfsrv_markstable(), except for the flag being set. 5074 */ 5075 static void 5076 nfsrv_markreclaim(struct nfsclient *clp) 5077 { 5078 struct nfsrv_stable *sp; 5079 5080 /* 5081 * First find the client structure. 5082 */ 5083 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { 5084 if (sp->nst_len == clp->lc_idlen && 5085 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len)) 5086 break; 5087 } 5088 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head)) 5089 return; 5090 5091 /* 5092 * Now, just set the flag. 5093 */ 5094 sp->nst_flag |= NFSNST_RECLAIMED; 5095 } 5096 5097 /* 5098 * This function is called for a reclaim, to see if it gets grace. 5099 * It returns 0 if a reclaim is allowed, 1 otherwise. 5100 */ 5101 static int 5102 nfsrv_checkstable(struct nfsclient *clp) 5103 { 5104 struct nfsrv_stable *sp; 5105 5106 /* 5107 * First, find the entry for the client. 5108 */ 5109 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) { 5110 if (sp->nst_len == clp->lc_idlen && 5111 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len)) 5112 break; 5113 } 5114 5115 /* 5116 * If not in the list, state was revoked or no state was issued 5117 * since the previous reboot, a reclaim is denied. 5118 */ 5119 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) || 5120 (sp->nst_flag & NFSNST_REVOKE) || 5121 !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK)) 5122 return (1); 5123 return (0); 5124 } 5125 5126 /* 5127 * Test for and try to clear out a conflicting client. This is called by 5128 * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients 5129 * a found. 5130 * The trick here is that it can't revoke a conflicting client with an 5131 * expired lease unless it holds the v4root lock, so... 5132 * If no v4root lock, get the lock and return 1 to indicate "try again". 5133 * Return 0 to indicate the conflict can't be revoked and 1 to indicate 5134 * the revocation worked and the conflicting client is "bye, bye", so it 5135 * can be tried again. 5136 * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK(). 5137 * Unlocks State before a non-zero value is returned. 5138 */ 5139 static int 5140 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp, 5141 NFSPROC_T *p) 5142 { 5143 int gotlock, lktype = 0; 5144 5145 /* 5146 * If lease hasn't expired, we can't fix it. 5147 */ 5148 if (clp->lc_expiry >= NFSD_MONOSEC || 5149 !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) 5150 return (0); 5151 if (*haslockp == 0) { 5152 NFSUNLOCKSTATE(); 5153 if (vp != NULL) { 5154 lktype = NFSVOPISLOCKED(vp); 5155 NFSVOPUNLOCK(vp, 0); 5156 } 5157 NFSLOCKV4ROOTMUTEX(); 5158 nfsv4_relref(&nfsv4rootfs_lock); 5159 do { 5160 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 5161 NFSV4ROOTLOCKMUTEXPTR, NULL); 5162 } while (!gotlock); 5163 NFSUNLOCKV4ROOTMUTEX(); 5164 *haslockp = 1; 5165 if (vp != NULL) { 5166 NFSVOPLOCK(vp, lktype | LK_RETRY); 5167 if ((vp->v_iflag & VI_DOOMED) != 0) 5168 return (2); 5169 } 5170 return (1); 5171 } 5172 NFSUNLOCKSTATE(); 5173 5174 /* 5175 * Ok, we can expire the conflicting client. 5176 */ 5177 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); 5178 nfsrv_backupstable(); 5179 nfsrv_cleanclient(clp, p); 5180 nfsrv_freedeleglist(&clp->lc_deleg); 5181 nfsrv_freedeleglist(&clp->lc_olddeleg); 5182 LIST_REMOVE(clp, lc_hash); 5183 nfsrv_zapclient(clp, p); 5184 return (1); 5185 } 5186 5187 /* 5188 * Resolve a delegation conflict. 5189 * Returns 0 to indicate the conflict was resolved without sleeping. 5190 * Return -1 to indicate that the caller should check for conflicts again. 5191 * Return > 0 for an error that should be returned, normally NFSERR_DELAY. 5192 * 5193 * Also, manipulate the nfsv4root_lock, as required. It isn't changed 5194 * for a return of 0, since there was no sleep and it could be required 5195 * later. It is released for a return of NFSERR_DELAY, since the caller 5196 * will return that error. It is released when a sleep was done waiting 5197 * for the delegation to be returned or expire (so that other nfsds can 5198 * handle ops). Then, it must be acquired for the write to stable storage. 5199 * (This function is somewhat similar to nfsrv_clientconflict(), but 5200 * the semantics differ in a couple of subtle ways. The return of 0 5201 * indicates the conflict was resolved without sleeping here, not 5202 * that the conflict can't be resolved and the handling of nfsv4root_lock 5203 * differs, as noted above.) 5204 * Unlocks State before returning a non-zero value. 5205 */ 5206 static int 5207 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p, 5208 vnode_t vp) 5209 { 5210 struct nfsclient *clp = stp->ls_clp; 5211 int gotlock, error, lktype = 0, retrycnt, zapped_clp; 5212 nfsv4stateid_t tstateid; 5213 fhandle_t tfh; 5214 5215 /* 5216 * If the conflict is with an old delegation... 5217 */ 5218 if (stp->ls_flags & NFSLCK_OLDDELEG) { 5219 /* 5220 * You can delete it, if it has expired. 5221 */ 5222 if (clp->lc_delegtime < NFSD_MONOSEC) { 5223 nfsrv_freedeleg(stp); 5224 NFSUNLOCKSTATE(); 5225 error = -1; 5226 goto out; 5227 } 5228 NFSUNLOCKSTATE(); 5229 /* 5230 * During this delay, the old delegation could expire or it 5231 * could be recovered by the client via an Open with 5232 * CLAIM_DELEGATE_PREV. 5233 * Release the nfsv4root_lock, if held. 5234 */ 5235 if (*haslockp) { 5236 *haslockp = 0; 5237 NFSLOCKV4ROOTMUTEX(); 5238 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5239 NFSUNLOCKV4ROOTMUTEX(); 5240 } 5241 error = NFSERR_DELAY; 5242 goto out; 5243 } 5244 5245 /* 5246 * It's a current delegation, so: 5247 * - check to see if the delegation has expired 5248 * - if so, get the v4root lock and then expire it 5249 */ 5250 if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) { 5251 /* 5252 * - do a recall callback, since not yet done 5253 * For now, never allow truncate to be set. To use 5254 * truncate safely, it must be guaranteed that the 5255 * Remove, Rename or Setattr with size of 0 will 5256 * succeed and that would require major changes to 5257 * the VFS/Vnode OPs. 5258 * Set the expiry time large enough so that it won't expire 5259 * until after the callback, then set it correctly, once 5260 * the callback is done. (The delegation will now time 5261 * out whether or not the Recall worked ok. The timeout 5262 * will be extended when ops are done on the delegation 5263 * stateid, up to the timelimit.) 5264 */ 5265 stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) + 5266 NFSRV_LEASEDELTA; 5267 stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) + 5268 NFSRV_LEASEDELTA; 5269 stp->ls_flags |= NFSLCK_DELEGRECALL; 5270 5271 /* 5272 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies 5273 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done 5274 * in order to try and avoid a race that could happen 5275 * when a CBRecall request passed the Open reply with 5276 * the delegation in it when transitting the network. 5277 * Since nfsrv_docallback will sleep, don't use stp after 5278 * the call. 5279 */ 5280 NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid, 5281 sizeof (tstateid)); 5282 NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh, 5283 sizeof (tfh)); 5284 NFSUNLOCKSTATE(); 5285 if (*haslockp) { 5286 *haslockp = 0; 5287 NFSLOCKV4ROOTMUTEX(); 5288 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5289 NFSUNLOCKV4ROOTMUTEX(); 5290 } 5291 retrycnt = 0; 5292 do { 5293 error = nfsrv_docallback(clp, NFSV4OP_CBRECALL, 5294 &tstateid, 0, &tfh, NULL, NULL, 0, p); 5295 retrycnt++; 5296 } while ((error == NFSERR_BADSTATEID || 5297 error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT); 5298 error = NFSERR_DELAY; 5299 goto out; 5300 } 5301 5302 if (clp->lc_expiry >= NFSD_MONOSEC && 5303 stp->ls_delegtime >= NFSD_MONOSEC) { 5304 NFSUNLOCKSTATE(); 5305 /* 5306 * A recall has been done, but it has not yet expired. 5307 * So, RETURN_DELAY. 5308 */ 5309 if (*haslockp) { 5310 *haslockp = 0; 5311 NFSLOCKV4ROOTMUTEX(); 5312 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5313 NFSUNLOCKV4ROOTMUTEX(); 5314 } 5315 error = NFSERR_DELAY; 5316 goto out; 5317 } 5318 5319 /* 5320 * If we don't yet have the lock, just get it and then return, 5321 * since we need that before deleting expired state, such as 5322 * this delegation. 5323 * When getting the lock, unlock the vnode, so other nfsds that 5324 * are in progress, won't get stuck waiting for the vnode lock. 5325 */ 5326 if (*haslockp == 0) { 5327 NFSUNLOCKSTATE(); 5328 if (vp != NULL) { 5329 lktype = NFSVOPISLOCKED(vp); 5330 NFSVOPUNLOCK(vp, 0); 5331 } 5332 NFSLOCKV4ROOTMUTEX(); 5333 nfsv4_relref(&nfsv4rootfs_lock); 5334 do { 5335 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 5336 NFSV4ROOTLOCKMUTEXPTR, NULL); 5337 } while (!gotlock); 5338 NFSUNLOCKV4ROOTMUTEX(); 5339 *haslockp = 1; 5340 if (vp != NULL) { 5341 NFSVOPLOCK(vp, lktype | LK_RETRY); 5342 if ((vp->v_iflag & VI_DOOMED) != 0) { 5343 *haslockp = 0; 5344 NFSLOCKV4ROOTMUTEX(); 5345 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5346 NFSUNLOCKV4ROOTMUTEX(); 5347 error = NFSERR_PERM; 5348 goto out; 5349 } 5350 } 5351 error = -1; 5352 goto out; 5353 } 5354 5355 NFSUNLOCKSTATE(); 5356 /* 5357 * Ok, we can delete the expired delegation. 5358 * First, write the Revoke record to stable storage and then 5359 * clear out the conflict. 5360 * Since all other nfsd threads are now blocked, we can safely 5361 * sleep without the state changing. 5362 */ 5363 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p); 5364 nfsrv_backupstable(); 5365 if (clp->lc_expiry < NFSD_MONOSEC) { 5366 nfsrv_cleanclient(clp, p); 5367 nfsrv_freedeleglist(&clp->lc_deleg); 5368 nfsrv_freedeleglist(&clp->lc_olddeleg); 5369 LIST_REMOVE(clp, lc_hash); 5370 zapped_clp = 1; 5371 } else { 5372 nfsrv_freedeleg(stp); 5373 zapped_clp = 0; 5374 } 5375 if (zapped_clp) 5376 nfsrv_zapclient(clp, p); 5377 error = -1; 5378 5379 out: 5380 NFSEXITCODE(error); 5381 return (error); 5382 } 5383 5384 /* 5385 * Check for a remove allowed, if remove is set to 1 and get rid of 5386 * delegations. 5387 */ 5388 APPLESTATIC int 5389 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p) 5390 { 5391 struct nfsstate *stp; 5392 struct nfslockfile *lfp; 5393 int error, haslock = 0; 5394 fhandle_t nfh; 5395 5396 /* 5397 * First, get the lock file structure. 5398 * (A return of -1 means no associated state, so remove ok.) 5399 */ 5400 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p); 5401 tryagain: 5402 NFSLOCKSTATE(); 5403 if (!error) 5404 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0); 5405 if (error) { 5406 NFSUNLOCKSTATE(); 5407 if (haslock) { 5408 NFSLOCKV4ROOTMUTEX(); 5409 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5410 NFSUNLOCKV4ROOTMUTEX(); 5411 } 5412 if (error == -1) 5413 error = 0; 5414 goto out; 5415 } 5416 5417 /* 5418 * Now, we must Recall any delegations. 5419 */ 5420 error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p); 5421 if (error) { 5422 /* 5423 * nfsrv_cleandeleg() unlocks state for non-zero 5424 * return. 5425 */ 5426 if (error == -1) 5427 goto tryagain; 5428 if (haslock) { 5429 NFSLOCKV4ROOTMUTEX(); 5430 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5431 NFSUNLOCKV4ROOTMUTEX(); 5432 } 5433 goto out; 5434 } 5435 5436 /* 5437 * Now, look for a conflicting open share. 5438 */ 5439 if (remove) { 5440 /* 5441 * If the entry in the directory was the last reference to the 5442 * corresponding filesystem object, the object can be destroyed 5443 * */ 5444 if(lfp->lf_usecount>1) 5445 LIST_FOREACH(stp, &lfp->lf_open, ls_file) { 5446 if (stp->ls_flags & NFSLCK_WRITEDENY) { 5447 error = NFSERR_FILEOPEN; 5448 break; 5449 } 5450 } 5451 } 5452 5453 NFSUNLOCKSTATE(); 5454 if (haslock) { 5455 NFSLOCKV4ROOTMUTEX(); 5456 nfsv4_unlock(&nfsv4rootfs_lock, 1); 5457 NFSUNLOCKV4ROOTMUTEX(); 5458 } 5459 5460 out: 5461 NFSEXITCODE(error); 5462 return (error); 5463 } 5464 5465 /* 5466 * Clear out all delegations for the file referred to by lfp. 5467 * May return NFSERR_DELAY, if there will be a delay waiting for 5468 * delegations to expire. 5469 * Returns -1 to indicate it slept while recalling a delegation. 5470 * This function has the side effect of deleting the nfslockfile structure, 5471 * if it no longer has associated state and didn't have to sleep. 5472 * Unlocks State before a non-zero value is returned. 5473 */ 5474 static int 5475 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp, 5476 struct nfsclient *clp, int *haslockp, NFSPROC_T *p) 5477 { 5478 struct nfsstate *stp, *nstp; 5479 int ret = 0; 5480 5481 stp = LIST_FIRST(&lfp->lf_deleg); 5482 while (stp != LIST_END(&lfp->lf_deleg)) { 5483 nstp = LIST_NEXT(stp, ls_file); 5484 if (stp->ls_clp != clp) { 5485 ret = nfsrv_delegconflict(stp, haslockp, p, vp); 5486 if (ret) { 5487 /* 5488 * nfsrv_delegconflict() unlocks state 5489 * when it returns non-zero. 5490 */ 5491 goto out; 5492 } 5493 } 5494 stp = nstp; 5495 } 5496 out: 5497 NFSEXITCODE(ret); 5498 return (ret); 5499 } 5500 5501 /* 5502 * There are certain operations that, when being done outside of NFSv4, 5503 * require that any NFSv4 delegation for the file be recalled. 5504 * This function is to be called for those cases: 5505 * VOP_RENAME() - When a delegation is being recalled for any reason, 5506 * the client may have to do Opens against the server, using the file's 5507 * final component name. If the file has been renamed on the server, 5508 * that component name will be incorrect and the Open will fail. 5509 * VOP_REMOVE() - Theoretically, a client could Open a file after it has 5510 * been removed on the server, if there is a delegation issued to 5511 * that client for the file. I say "theoretically" since clients 5512 * normally do an Access Op before the Open and that Access Op will 5513 * fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so 5514 * they will detect the file's removal in the same manner. (There is 5515 * one case where RFC3530 allows a client to do an Open without first 5516 * doing an Access Op, which is passage of a check against the ACE 5517 * returned with a Write delegation, but current practice is to ignore 5518 * the ACE and always do an Access Op.) 5519 * Since the functions can only be called with an unlocked vnode, this 5520 * can't be done at this time. 5521 * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range 5522 * locks locally in the client, which are not visible to the server. To 5523 * deal with this, issuing of delegations for a vnode must be disabled 5524 * and all delegations for the vnode recalled. This is done via the 5525 * second function, using the VV_DISABLEDELEG vflag on the vnode. 5526 */ 5527 APPLESTATIC void 5528 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p) 5529 { 5530 time_t starttime; 5531 int error; 5532 5533 /* 5534 * First, check to see if the server is currently running and it has 5535 * been called for a regular file when issuing delegations. 5536 */ 5537 if (newnfs_numnfsd == 0 || vp->v_type != VREG || 5538 nfsrv_issuedelegs == 0) 5539 return; 5540 5541 KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp)); 5542 /* 5543 * First, get a reference on the nfsv4rootfs_lock so that an 5544 * exclusive lock cannot be acquired by another thread. 5545 */ 5546 NFSLOCKV4ROOTMUTEX(); 5547 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL); 5548 NFSUNLOCKV4ROOTMUTEX(); 5549 5550 /* 5551 * Now, call nfsrv_checkremove() in a loop while it returns 5552 * NFSERR_DELAY. Return upon any other error or when timed out. 5553 */ 5554 starttime = NFSD_MONOSEC; 5555 do { 5556 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) { 5557 error = nfsrv_checkremove(vp, 0, p); 5558 NFSVOPUNLOCK(vp, 0); 5559 } else 5560 error = EPERM; 5561 if (error == NFSERR_DELAY) { 5562 if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO) 5563 break; 5564 /* Sleep for a short period of time */ 5565 (void) nfs_catnap(PZERO, 0, "nfsremove"); 5566 } 5567 } while (error == NFSERR_DELAY); 5568 NFSLOCKV4ROOTMUTEX(); 5569 nfsv4_relref(&nfsv4rootfs_lock); 5570 NFSUNLOCKV4ROOTMUTEX(); 5571 } 5572 5573 APPLESTATIC void 5574 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p) 5575 { 5576 5577 #ifdef VV_DISABLEDELEG 5578 /* 5579 * First, flag issuance of delegations disabled. 5580 */ 5581 atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG); 5582 #endif 5583 5584 /* 5585 * Then call nfsd_recalldelegation() to get rid of all extant 5586 * delegations. 5587 */ 5588 nfsd_recalldelegation(vp, p); 5589 } 5590 5591 /* 5592 * Check for conflicting locks, etc. and then get rid of delegations. 5593 * (At one point I thought that I should get rid of delegations for any 5594 * Setattr, since it could potentially disallow the I/O op (read or write) 5595 * allowed by the delegation. However, Setattr Ops that aren't changing 5596 * the size get a stateid of all 0s, so you can't tell if it is a delegation 5597 * for the same client or a different one, so I decided to only get rid 5598 * of delegations for other clients when the size is being changed.) 5599 * In general, a Setattr can disable NFS I/O Ops that are outstanding, such 5600 * as Write backs, even if there is no delegation, so it really isn't any 5601 * different?) 5602 */ 5603 APPLESTATIC int 5604 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd, 5605 nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp, 5606 struct nfsexstuff *exp, NFSPROC_T *p) 5607 { 5608 struct nfsstate st, *stp = &st; 5609 struct nfslock lo, *lop = &lo; 5610 int error = 0; 5611 nfsquad_t clientid; 5612 5613 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) { 5614 stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS); 5615 lop->lo_first = nvap->na_size; 5616 } else { 5617 stp->ls_flags = 0; 5618 lop->lo_first = 0; 5619 } 5620 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) || 5621 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) || 5622 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) || 5623 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL)) 5624 stp->ls_flags |= NFSLCK_SETATTR; 5625 if (stp->ls_flags == 0) 5626 goto out; 5627 lop->lo_end = NFS64BITSSET; 5628 lop->lo_flags = NFSLCK_WRITE; 5629 stp->ls_ownerlen = 0; 5630 stp->ls_op = NULL; 5631 stp->ls_uid = nd->nd_cred->cr_uid; 5632 stp->ls_stateid.seqid = stateidp->seqid; 5633 clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0]; 5634 clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1]; 5635 stp->ls_stateid.other[2] = stateidp->other[2]; 5636 error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid, 5637 stateidp, exp, nd, p); 5638 5639 out: 5640 NFSEXITCODE2(error, nd); 5641 return (error); 5642 } 5643 5644 /* 5645 * Check for a write delegation and do a CBGETATTR if there is one, updating 5646 * the attributes, as required. 5647 * Should I return an error if I can't get the attributes? (For now, I'll 5648 * just return ok. 5649 */ 5650 APPLESTATIC int 5651 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp, 5652 struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p) 5653 { 5654 struct nfsstate *stp; 5655 struct nfslockfile *lfp; 5656 struct nfsclient *clp; 5657 struct nfsvattr nva; 5658 fhandle_t nfh; 5659 int error = 0; 5660 nfsattrbit_t cbbits; 5661 u_quad_t delegfilerev; 5662 5663 NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits); 5664 if (!NFSNONZERO_ATTRBIT(&cbbits)) 5665 goto out; 5666 if (nfsrv_writedelegcnt == 0) 5667 goto out; 5668 5669 /* 5670 * Get the lock file structure. 5671 * (A return of -1 means no associated state, so return ok.) 5672 */ 5673 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p); 5674 NFSLOCKSTATE(); 5675 if (!error) 5676 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0); 5677 if (error) { 5678 NFSUNLOCKSTATE(); 5679 if (error == -1) 5680 error = 0; 5681 goto out; 5682 } 5683 5684 /* 5685 * Now, look for a write delegation. 5686 */ 5687 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { 5688 if (stp->ls_flags & NFSLCK_DELEGWRITE) 5689 break; 5690 } 5691 if (stp == LIST_END(&lfp->lf_deleg)) { 5692 NFSUNLOCKSTATE(); 5693 goto out; 5694 } 5695 clp = stp->ls_clp; 5696 delegfilerev = stp->ls_filerev; 5697 5698 /* 5699 * If the Write delegation was issued as a part of this Compound RPC 5700 * or if we have an Implied Clientid (used in a previous Op in this 5701 * compound) and it is the client the delegation was issued to, 5702 * just return ok. 5703 * I also assume that it is from the same client iff the network 5704 * host IP address is the same as the callback address. (Not 5705 * exactly correct by the RFC, but avoids a lot of Getattr 5706 * callbacks.) 5707 */ 5708 if (nd->nd_compref == stp->ls_compref || 5709 ((nd->nd_flag & ND_IMPLIEDCLID) && 5710 clp->lc_clientid.qval == nd->nd_clientid.qval) || 5711 nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) { 5712 NFSUNLOCKSTATE(); 5713 goto out; 5714 } 5715 5716 /* 5717 * We are now done with the delegation state structure, 5718 * so the statelock can be released and we can now tsleep(). 5719 */ 5720 5721 /* 5722 * Now, we must do the CB Getattr callback, to see if Change or Size 5723 * has changed. 5724 */ 5725 if (clp->lc_expiry >= NFSD_MONOSEC) { 5726 NFSUNLOCKSTATE(); 5727 NFSVNO_ATTRINIT(&nva); 5728 nva.na_filerev = NFS64BITSSET; 5729 error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL, 5730 0, &nfh, &nva, &cbbits, 0, p); 5731 if (!error) { 5732 if ((nva.na_filerev != NFS64BITSSET && 5733 nva.na_filerev > delegfilerev) || 5734 (NFSVNO_ISSETSIZE(&nva) && 5735 nva.na_size != nvap->na_size)) { 5736 error = nfsvno_updfilerev(vp, nvap, nd, p); 5737 if (NFSVNO_ISSETSIZE(&nva)) 5738 nvap->na_size = nva.na_size; 5739 } 5740 } else 5741 error = 0; /* Ignore callback errors for now. */ 5742 } else { 5743 NFSUNLOCKSTATE(); 5744 } 5745 5746 out: 5747 NFSEXITCODE2(error, nd); 5748 return (error); 5749 } 5750 5751 /* 5752 * This function looks for openowners that haven't had any opens for 5753 * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS 5754 * is set. 5755 */ 5756 APPLESTATIC void 5757 nfsrv_throwawayopens(NFSPROC_T *p) 5758 { 5759 struct nfsclient *clp, *nclp; 5760 struct nfsstate *stp, *nstp; 5761 int i; 5762 5763 NFSLOCKSTATE(); 5764 nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS; 5765 /* 5766 * For each client... 5767 */ 5768 for (i = 0; i < nfsrv_clienthashsize; i++) { 5769 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) { 5770 LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) { 5771 if (LIST_EMPTY(&stp->ls_open) && 5772 (stp->ls_noopens > NFSNOOPEN || 5773 (nfsrv_openpluslock * 2) > 5774 nfsrv_v4statelimit)) 5775 nfsrv_freeopenowner(stp, 0, p); 5776 } 5777 } 5778 } 5779 NFSUNLOCKSTATE(); 5780 } 5781 5782 /* 5783 * This function checks to see if the credentials are the same. 5784 * Returns 1 for not same, 0 otherwise. 5785 */ 5786 static int 5787 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp) 5788 { 5789 5790 if (nd->nd_flag & ND_GSS) { 5791 if (!(clp->lc_flags & LCL_GSS)) 5792 return (1); 5793 if (clp->lc_flags & LCL_NAME) { 5794 if (nd->nd_princlen != clp->lc_namelen || 5795 NFSBCMP(nd->nd_principal, clp->lc_name, 5796 clp->lc_namelen)) 5797 return (1); 5798 else 5799 return (0); 5800 } 5801 if (nd->nd_cred->cr_uid == clp->lc_uid) 5802 return (0); 5803 else 5804 return (1); 5805 } else if (clp->lc_flags & LCL_GSS) 5806 return (1); 5807 /* 5808 * For AUTH_SYS, allow the same uid or root. (This is underspecified 5809 * in RFC3530, which talks about principals, but doesn't say anything 5810 * about uids for AUTH_SYS.) 5811 */ 5812 if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0) 5813 return (0); 5814 else 5815 return (1); 5816 } 5817 5818 /* 5819 * Calculate the lease expiry time. 5820 */ 5821 static time_t 5822 nfsrv_leaseexpiry(void) 5823 { 5824 5825 if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC) 5826 return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA)); 5827 return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA); 5828 } 5829 5830 /* 5831 * Delay the delegation timeout as far as ls_delegtimelimit, as required. 5832 */ 5833 static void 5834 nfsrv_delaydelegtimeout(struct nfsstate *stp) 5835 { 5836 5837 if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) 5838 return; 5839 5840 if ((stp->ls_delegtime + 15) > NFSD_MONOSEC && 5841 stp->ls_delegtime < stp->ls_delegtimelimit) { 5842 stp->ls_delegtime += nfsrv_lease; 5843 if (stp->ls_delegtime > stp->ls_delegtimelimit) 5844 stp->ls_delegtime = stp->ls_delegtimelimit; 5845 } 5846 } 5847 5848 /* 5849 * This function checks to see if there is any other state associated 5850 * with the openowner for this Open. 5851 * It returns 1 if there is no other state, 0 otherwise. 5852 */ 5853 static int 5854 nfsrv_nootherstate(struct nfsstate *stp) 5855 { 5856 struct nfsstate *tstp; 5857 5858 LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) { 5859 if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock)) 5860 return (0); 5861 } 5862 return (1); 5863 } 5864 5865 /* 5866 * Create a list of lock deltas (changes to local byte range locking 5867 * that can be rolled back using the list) and apply the changes via 5868 * nfsvno_advlock(). Optionally, lock the list. It is expected that either 5869 * the rollback or update function will be called after this. 5870 * It returns an error (and rolls back, as required), if any nfsvno_advlock() 5871 * call fails. If it returns an error, it will unlock the list. 5872 */ 5873 static int 5874 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags, 5875 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p) 5876 { 5877 struct nfslock *lop, *nlop; 5878 int error = 0; 5879 5880 /* Loop through the list of locks. */ 5881 lop = LIST_FIRST(&lfp->lf_locallock); 5882 while (first < end && lop != NULL) { 5883 nlop = LIST_NEXT(lop, lo_lckowner); 5884 if (first >= lop->lo_end) { 5885 /* not there yet */ 5886 lop = nlop; 5887 } else if (first < lop->lo_first) { 5888 /* new one starts before entry in list */ 5889 if (end <= lop->lo_first) { 5890 /* no overlap between old and new */ 5891 error = nfsrv_dolocal(vp, lfp, flags, 5892 NFSLCK_UNLOCK, first, end, cfp, p); 5893 if (error != 0) 5894 break; 5895 first = end; 5896 } else { 5897 /* handle fragment overlapped with new one */ 5898 error = nfsrv_dolocal(vp, lfp, flags, 5899 NFSLCK_UNLOCK, first, lop->lo_first, cfp, 5900 p); 5901 if (error != 0) 5902 break; 5903 first = lop->lo_first; 5904 } 5905 } else { 5906 /* new one overlaps this entry in list */ 5907 if (end <= lop->lo_end) { 5908 /* overlaps all of new one */ 5909 error = nfsrv_dolocal(vp, lfp, flags, 5910 lop->lo_flags, first, end, cfp, p); 5911 if (error != 0) 5912 break; 5913 first = end; 5914 } else { 5915 /* handle fragment overlapped with new one */ 5916 error = nfsrv_dolocal(vp, lfp, flags, 5917 lop->lo_flags, first, lop->lo_end, cfp, p); 5918 if (error != 0) 5919 break; 5920 first = lop->lo_end; 5921 lop = nlop; 5922 } 5923 } 5924 } 5925 if (first < end && error == 0) 5926 /* handle fragment past end of list */ 5927 error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first, 5928 end, cfp, p); 5929 5930 NFSEXITCODE(error); 5931 return (error); 5932 } 5933 5934 /* 5935 * Local lock unlock. Unlock all byte ranges that are no longer locked 5936 * by NFSv4. To do this, unlock any subranges of first-->end that 5937 * do not overlap with the byte ranges of any lock in the lfp->lf_lock 5938 * list. This list has all locks for the file held by other 5939 * <clientid, lockowner> tuples. The list is ordered by increasing 5940 * lo_first value, but may have entries that overlap each other, for 5941 * the case of read locks. 5942 */ 5943 static void 5944 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first, 5945 uint64_t init_end, NFSPROC_T *p) 5946 { 5947 struct nfslock *lop; 5948 uint64_t first, end, prevfirst __unused; 5949 5950 first = init_first; 5951 end = init_end; 5952 while (first < init_end) { 5953 /* Loop through all nfs locks, adjusting first and end */ 5954 prevfirst = 0; 5955 LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) { 5956 KASSERT(prevfirst <= lop->lo_first, 5957 ("nfsv4 locks out of order")); 5958 KASSERT(lop->lo_first < lop->lo_end, 5959 ("nfsv4 bogus lock")); 5960 prevfirst = lop->lo_first; 5961 if (first >= lop->lo_first && 5962 first < lop->lo_end) 5963 /* 5964 * Overlaps with initial part, so trim 5965 * off that initial part by moving first past 5966 * it. 5967 */ 5968 first = lop->lo_end; 5969 else if (end > lop->lo_first && 5970 lop->lo_first > first) { 5971 /* 5972 * This lock defines the end of the 5973 * segment to unlock, so set end to the 5974 * start of it and break out of the loop. 5975 */ 5976 end = lop->lo_first; 5977 break; 5978 } 5979 if (first >= end) 5980 /* 5981 * There is no segment left to do, so 5982 * break out of this loop and then exit 5983 * the outer while() since first will be set 5984 * to end, which must equal init_end here. 5985 */ 5986 break; 5987 } 5988 if (first < end) { 5989 /* Unlock this segment */ 5990 (void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK, 5991 NFSLCK_READ, first, end, NULL, p); 5992 nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK, 5993 first, end); 5994 } 5995 /* 5996 * Now move past this segment and look for any further 5997 * segment in the range, if there is one. 5998 */ 5999 first = end; 6000 end = init_end; 6001 } 6002 } 6003 6004 /* 6005 * Do the local lock operation and update the rollback list, as required. 6006 * Perform the rollback and return the error if nfsvno_advlock() fails. 6007 */ 6008 static int 6009 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags, 6010 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p) 6011 { 6012 struct nfsrollback *rlp; 6013 int error = 0, ltype, oldltype; 6014 6015 if (flags & NFSLCK_WRITE) 6016 ltype = F_WRLCK; 6017 else if (flags & NFSLCK_READ) 6018 ltype = F_RDLCK; 6019 else 6020 ltype = F_UNLCK; 6021 if (oldflags & NFSLCK_WRITE) 6022 oldltype = F_WRLCK; 6023 else if (oldflags & NFSLCK_READ) 6024 oldltype = F_RDLCK; 6025 else 6026 oldltype = F_UNLCK; 6027 if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK)) 6028 /* nothing to do */ 6029 goto out; 6030 error = nfsvno_advlock(vp, ltype, first, end, p); 6031 if (error != 0) { 6032 if (cfp != NULL) { 6033 cfp->cl_clientid.lval[0] = 0; 6034 cfp->cl_clientid.lval[1] = 0; 6035 cfp->cl_first = 0; 6036 cfp->cl_end = NFS64BITSSET; 6037 cfp->cl_flags = NFSLCK_WRITE; 6038 cfp->cl_ownerlen = 5; 6039 NFSBCOPY("LOCAL", cfp->cl_owner, 5); 6040 } 6041 nfsrv_locallock_rollback(vp, lfp, p); 6042 } else if (ltype != F_UNLCK) { 6043 rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK, 6044 M_WAITOK); 6045 rlp->rlck_first = first; 6046 rlp->rlck_end = end; 6047 rlp->rlck_type = oldltype; 6048 LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list); 6049 } 6050 6051 out: 6052 NFSEXITCODE(error); 6053 return (error); 6054 } 6055 6056 /* 6057 * Roll back local lock changes and free up the rollback list. 6058 */ 6059 static void 6060 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p) 6061 { 6062 struct nfsrollback *rlp, *nrlp; 6063 6064 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) { 6065 (void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first, 6066 rlp->rlck_end, p); 6067 free(rlp, M_NFSDROLLBACK); 6068 } 6069 LIST_INIT(&lfp->lf_rollback); 6070 } 6071 6072 /* 6073 * Update local lock list and delete rollback list (ie now committed to the 6074 * local locks). Most of the work is done by the internal function. 6075 */ 6076 static void 6077 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first, 6078 uint64_t end) 6079 { 6080 struct nfsrollback *rlp, *nrlp; 6081 struct nfslock *new_lop, *other_lop; 6082 6083 new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK); 6084 if (flags & (NFSLCK_READ | NFSLCK_WRITE)) 6085 other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, 6086 M_WAITOK); 6087 else 6088 other_lop = NULL; 6089 new_lop->lo_flags = flags; 6090 new_lop->lo_first = first; 6091 new_lop->lo_end = end; 6092 nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp); 6093 if (new_lop != NULL) 6094 free(new_lop, M_NFSDLOCK); 6095 if (other_lop != NULL) 6096 free(other_lop, M_NFSDLOCK); 6097 6098 /* and get rid of the rollback list */ 6099 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) 6100 free(rlp, M_NFSDROLLBACK); 6101 LIST_INIT(&lfp->lf_rollback); 6102 } 6103 6104 /* 6105 * Lock the struct nfslockfile for local lock updating. 6106 */ 6107 static void 6108 nfsrv_locklf(struct nfslockfile *lfp) 6109 { 6110 int gotlock; 6111 6112 /* lf_usecount ensures *lfp won't be free'd */ 6113 lfp->lf_usecount++; 6114 do { 6115 gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL, 6116 NFSSTATEMUTEXPTR, NULL); 6117 } while (gotlock == 0); 6118 lfp->lf_usecount--; 6119 } 6120 6121 /* 6122 * Unlock the struct nfslockfile after local lock updating. 6123 */ 6124 static void 6125 nfsrv_unlocklf(struct nfslockfile *lfp) 6126 { 6127 6128 nfsv4_unlock(&lfp->lf_locallock_lck, 0); 6129 } 6130 6131 /* 6132 * Clear out all state for the NFSv4 server. 6133 * Must be called by a thread that can sleep when no nfsds are running. 6134 */ 6135 void 6136 nfsrv_throwawayallstate(NFSPROC_T *p) 6137 { 6138 struct nfsclient *clp, *nclp; 6139 struct nfslockfile *lfp, *nlfp; 6140 int i; 6141 6142 /* 6143 * For each client, clean out the state and then free the structure. 6144 */ 6145 for (i = 0; i < nfsrv_clienthashsize; i++) { 6146 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) { 6147 nfsrv_cleanclient(clp, p); 6148 nfsrv_freedeleglist(&clp->lc_deleg); 6149 nfsrv_freedeleglist(&clp->lc_olddeleg); 6150 free(clp->lc_stateid, M_NFSDCLIENT); 6151 free(clp, M_NFSDCLIENT); 6152 } 6153 } 6154 6155 /* 6156 * Also, free up any remaining lock file structures. 6157 */ 6158 for (i = 0; i < nfsrv_lockhashsize; i++) { 6159 LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) { 6160 printf("nfsd unload: fnd a lock file struct\n"); 6161 nfsrv_freenfslockfile(lfp); 6162 } 6163 } 6164 6165 /* And get rid of the deviceid structures and layouts. */ 6166 nfsrv_freealllayoutsanddevids(); 6167 } 6168 6169 /* 6170 * Check the sequence# for the session and slot provided as an argument. 6171 * Also, renew the lease if the session will return NFS_OK. 6172 */ 6173 int 6174 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid, 6175 uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this, 6176 uint32_t *sflagsp, NFSPROC_T *p) 6177 { 6178 struct nfsdsession *sep; 6179 struct nfssessionhash *shp; 6180 int error; 6181 SVCXPRT *savxprt; 6182 6183 shp = NFSSESSIONHASH(nd->nd_sessionid); 6184 NFSLOCKSESSION(shp); 6185 sep = nfsrv_findsession(nd->nd_sessionid); 6186 if (sep == NULL) { 6187 NFSUNLOCKSESSION(shp); 6188 return (NFSERR_BADSESSION); 6189 } 6190 error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp, 6191 sep->sess_slots, NULL, NFSV4_SLOTS - 1); 6192 if (error != 0) { 6193 NFSUNLOCKSESSION(shp); 6194 return (error); 6195 } 6196 if (cache_this != 0) 6197 nd->nd_flag |= ND_SAVEREPLY; 6198 /* Renew the lease. */ 6199 sep->sess_clp->lc_expiry = nfsrv_leaseexpiry(); 6200 nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval; 6201 nd->nd_flag |= ND_IMPLIEDCLID; 6202 6203 /* 6204 * If this session handles the backchannel, save the nd_xprt for this 6205 * RPC, since this is the one being used. 6206 * RFC-5661 specifies that the fore channel will be implicitly 6207 * bound by a Sequence operation. However, since some NFSv4.1 clients 6208 * erroneously assumed that the back channel would be implicitly 6209 * bound as well, do the implicit binding unless a 6210 * BindConnectiontoSession has already been done on the session. 6211 */ 6212 if (sep->sess_clp->lc_req.nr_client != NULL && 6213 sep->sess_cbsess.nfsess_xprt != nd->nd_xprt && 6214 (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0 && 6215 (sep->sess_clp->lc_flags & LCL_DONEBINDCONN) == 0) { 6216 NFSD_DEBUG(2, 6217 "nfsrv_checksequence: implicit back channel bind\n"); 6218 savxprt = sep->sess_cbsess.nfsess_xprt; 6219 SVC_ACQUIRE(nd->nd_xprt); 6220 nd->nd_xprt->xp_p2 = 6221 sep->sess_clp->lc_req.nr_client->cl_private; 6222 nd->nd_xprt->xp_idletimeout = 0; /* Disable timeout. */ 6223 sep->sess_cbsess.nfsess_xprt = nd->nd_xprt; 6224 if (savxprt != NULL) 6225 SVC_RELEASE(savxprt); 6226 } 6227 6228 *sflagsp = 0; 6229 if (sep->sess_clp->lc_req.nr_client == NULL) 6230 *sflagsp |= NFSV4SEQ_CBPATHDOWN; 6231 NFSUNLOCKSESSION(shp); 6232 if (error == NFSERR_EXPIRED) { 6233 *sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED; 6234 error = 0; 6235 } else if (error == NFSERR_ADMINREVOKED) { 6236 *sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED; 6237 error = 0; 6238 } 6239 *highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1; 6240 return (0); 6241 } 6242 6243 /* 6244 * Check/set reclaim complete for this session/clientid. 6245 */ 6246 int 6247 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs) 6248 { 6249 struct nfsdsession *sep; 6250 struct nfssessionhash *shp; 6251 int error = 0; 6252 6253 shp = NFSSESSIONHASH(nd->nd_sessionid); 6254 NFSLOCKSTATE(); 6255 NFSLOCKSESSION(shp); 6256 sep = nfsrv_findsession(nd->nd_sessionid); 6257 if (sep == NULL) { 6258 NFSUNLOCKSESSION(shp); 6259 NFSUNLOCKSTATE(); 6260 return (NFSERR_BADSESSION); 6261 } 6262 6263 if (onefs != 0) 6264 sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS; 6265 /* Check to see if reclaim complete has already happened. */ 6266 else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) 6267 error = NFSERR_COMPLETEALREADY; 6268 else { 6269 sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE; 6270 nfsrv_markreclaim(sep->sess_clp); 6271 } 6272 NFSUNLOCKSESSION(shp); 6273 NFSUNLOCKSTATE(); 6274 return (error); 6275 } 6276 6277 /* 6278 * Cache the reply in a session slot. 6279 */ 6280 void 6281 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat, 6282 struct mbuf **m) 6283 { 6284 struct nfsdsession *sep; 6285 struct nfssessionhash *shp; 6286 6287 shp = NFSSESSIONHASH(sessionid); 6288 NFSLOCKSESSION(shp); 6289 sep = nfsrv_findsession(sessionid); 6290 if (sep == NULL) { 6291 NFSUNLOCKSESSION(shp); 6292 printf("nfsrv_cache_session: no session\n"); 6293 m_freem(*m); 6294 return; 6295 } 6296 nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m); 6297 NFSUNLOCKSESSION(shp); 6298 } 6299 6300 /* 6301 * Search for a session that matches the sessionid. 6302 */ 6303 static struct nfsdsession * 6304 nfsrv_findsession(uint8_t *sessionid) 6305 { 6306 struct nfsdsession *sep; 6307 struct nfssessionhash *shp; 6308 6309 shp = NFSSESSIONHASH(sessionid); 6310 LIST_FOREACH(sep, &shp->list, sess_hash) { 6311 if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID)) 6312 break; 6313 } 6314 return (sep); 6315 } 6316 6317 /* 6318 * Destroy a session. 6319 */ 6320 int 6321 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid) 6322 { 6323 int error, igotlock, samesess; 6324 6325 samesess = 0; 6326 if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) && 6327 (nd->nd_flag & ND_HASSEQUENCE) != 0) { 6328 samesess = 1; 6329 if ((nd->nd_flag & ND_LASTOP) == 0) 6330 return (NFSERR_BADSESSION); 6331 } 6332 6333 /* Lock out other nfsd threads */ 6334 NFSLOCKV4ROOTMUTEX(); 6335 nfsv4_relref(&nfsv4rootfs_lock); 6336 do { 6337 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL, 6338 NFSV4ROOTLOCKMUTEXPTR, NULL); 6339 } while (igotlock == 0); 6340 NFSUNLOCKV4ROOTMUTEX(); 6341 6342 error = nfsrv_freesession(NULL, sessionid); 6343 if (error == 0 && samesess != 0) 6344 nd->nd_flag &= ~ND_HASSEQUENCE; 6345 6346 NFSLOCKV4ROOTMUTEX(); 6347 nfsv4_unlock(&nfsv4rootfs_lock, 1); 6348 NFSUNLOCKV4ROOTMUTEX(); 6349 return (error); 6350 } 6351 6352 /* 6353 * Bind a connection to a session. 6354 * For now, only certain variants are supported, since the current session 6355 * structure can only handle a single backchannel entry, which will be 6356 * applied to all connections if it is set. 6357 */ 6358 int 6359 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp) 6360 { 6361 struct nfssessionhash *shp; 6362 struct nfsdsession *sep; 6363 struct nfsclient *clp; 6364 SVCXPRT *savxprt; 6365 int error; 6366 6367 error = 0; 6368 shp = NFSSESSIONHASH(sessionid); 6369 NFSLOCKSTATE(); 6370 NFSLOCKSESSION(shp); 6371 sep = nfsrv_findsession(sessionid); 6372 if (sep != NULL) { 6373 clp = sep->sess_clp; 6374 if (*foreaftp == NFSCDFC4_BACK || 6375 *foreaftp == NFSCDFC4_BACK_OR_BOTH || 6376 *foreaftp == NFSCDFC4_FORE_OR_BOTH) { 6377 /* Try to set up a backchannel. */ 6378 if (clp->lc_req.nr_client == NULL) { 6379 NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire " 6380 "backchannel\n"); 6381 clp->lc_req.nr_client = (struct __rpc_client *) 6382 clnt_bck_create(nd->nd_xprt->xp_socket, 6383 sep->sess_cbprogram, NFSV4_CBVERS); 6384 } 6385 if (clp->lc_req.nr_client != NULL) { 6386 NFSD_DEBUG(2, "nfsrv_bindconnsess: set up " 6387 "backchannel\n"); 6388 savxprt = sep->sess_cbsess.nfsess_xprt; 6389 SVC_ACQUIRE(nd->nd_xprt); 6390 nd->nd_xprt->xp_p2 = 6391 clp->lc_req.nr_client->cl_private; 6392 /* Disable idle timeout. */ 6393 nd->nd_xprt->xp_idletimeout = 0; 6394 sep->sess_cbsess.nfsess_xprt = nd->nd_xprt; 6395 if (savxprt != NULL) 6396 SVC_RELEASE(savxprt); 6397 sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN; 6398 clp->lc_flags |= LCL_DONEBINDCONN; 6399 if (*foreaftp == NFSCDFS4_BACK) 6400 *foreaftp = NFSCDFS4_BACK; 6401 else 6402 *foreaftp = NFSCDFS4_BOTH; 6403 } else if (*foreaftp != NFSCDFC4_BACK) { 6404 NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set " 6405 "up backchannel\n"); 6406 sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN; 6407 clp->lc_flags |= LCL_DONEBINDCONN; 6408 *foreaftp = NFSCDFS4_FORE; 6409 } else { 6410 error = NFSERR_NOTSUPP; 6411 printf("nfsrv_bindconnsess: Can't add " 6412 "backchannel\n"); 6413 } 6414 } else { 6415 NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n"); 6416 clp->lc_flags |= LCL_DONEBINDCONN; 6417 *foreaftp = NFSCDFS4_FORE; 6418 } 6419 } else 6420 error = NFSERR_BADSESSION; 6421 NFSUNLOCKSESSION(shp); 6422 NFSUNLOCKSTATE(); 6423 return (error); 6424 } 6425 6426 /* 6427 * Free up a session structure. 6428 */ 6429 static int 6430 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid) 6431 { 6432 struct nfssessionhash *shp; 6433 int i; 6434 6435 NFSLOCKSTATE(); 6436 if (sep == NULL) { 6437 shp = NFSSESSIONHASH(sessionid); 6438 NFSLOCKSESSION(shp); 6439 sep = nfsrv_findsession(sessionid); 6440 } else { 6441 shp = NFSSESSIONHASH(sep->sess_sessionid); 6442 NFSLOCKSESSION(shp); 6443 } 6444 if (sep != NULL) { 6445 sep->sess_refcnt--; 6446 if (sep->sess_refcnt > 0) { 6447 NFSUNLOCKSESSION(shp); 6448 NFSUNLOCKSTATE(); 6449 return (NFSERR_BACKCHANBUSY); 6450 } 6451 LIST_REMOVE(sep, sess_hash); 6452 LIST_REMOVE(sep, sess_list); 6453 } 6454 NFSUNLOCKSESSION(shp); 6455 NFSUNLOCKSTATE(); 6456 if (sep == NULL) 6457 return (NFSERR_BADSESSION); 6458 for (i = 0; i < NFSV4_SLOTS; i++) 6459 if (sep->sess_slots[i].nfssl_reply != NULL) 6460 m_freem(sep->sess_slots[i].nfssl_reply); 6461 if (sep->sess_cbsess.nfsess_xprt != NULL) 6462 SVC_RELEASE(sep->sess_cbsess.nfsess_xprt); 6463 free(sep, M_NFSDSESSION); 6464 return (0); 6465 } 6466 6467 /* 6468 * Free a stateid. 6469 * RFC5661 says that it should fail when there are associated opens, locks 6470 * or delegations. Since stateids represent opens, I don't see how you can 6471 * free an open stateid (it will be free'd when closed), so this function 6472 * only works for lock stateids (freeing the lock_owner) or delegations. 6473 */ 6474 int 6475 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, 6476 NFSPROC_T *p) 6477 { 6478 struct nfsclient *clp; 6479 struct nfsstate *stp; 6480 int error; 6481 6482 NFSLOCKSTATE(); 6483 /* 6484 * Look up the stateid 6485 */ 6486 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, 6487 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); 6488 if (error == 0) { 6489 /* First, check for a delegation. */ 6490 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) { 6491 if (!NFSBCMP(stp->ls_stateid.other, stateidp->other, 6492 NFSX_STATEIDOTHER)) 6493 break; 6494 } 6495 if (stp != NULL) { 6496 nfsrv_freedeleg(stp); 6497 NFSUNLOCKSTATE(); 6498 return (error); 6499 } 6500 } 6501 /* Not a delegation, try for a lock_owner. */ 6502 if (error == 0) 6503 error = nfsrv_getstate(clp, stateidp, 0, &stp); 6504 if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD | 6505 NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0)) 6506 /* Not a lock_owner stateid. */ 6507 error = NFSERR_LOCKSHELD; 6508 if (error == 0 && !LIST_EMPTY(&stp->ls_lock)) 6509 error = NFSERR_LOCKSHELD; 6510 if (error == 0) 6511 nfsrv_freelockowner(stp, NULL, 0, p); 6512 NFSUNLOCKSTATE(); 6513 return (error); 6514 } 6515 6516 /* 6517 * Test a stateid. 6518 */ 6519 int 6520 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp, 6521 NFSPROC_T *p) 6522 { 6523 struct nfsclient *clp; 6524 struct nfsstate *stp; 6525 int error; 6526 6527 NFSLOCKSTATE(); 6528 /* 6529 * Look up the stateid 6530 */ 6531 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, 6532 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); 6533 if (error == 0) 6534 error = nfsrv_getstate(clp, stateidp, 0, &stp); 6535 if (error == 0 && stateidp->seqid != 0 && 6536 SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid)) 6537 error = NFSERR_OLDSTATEID; 6538 NFSUNLOCKSTATE(); 6539 return (error); 6540 } 6541 6542 /* 6543 * Generate the xdr for an NFSv4.1 CBSequence Operation. 6544 */ 6545 static int 6546 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp, 6547 int dont_replycache, struct nfsdsession **sepp) 6548 { 6549 struct nfsdsession *sep; 6550 uint32_t *tl, slotseq = 0; 6551 int maxslot, slotpos; 6552 uint8_t sessionid[NFSX_V4SESSIONID]; 6553 int error; 6554 6555 error = nfsv4_getcbsession(clp, sepp); 6556 if (error != 0) 6557 return (error); 6558 sep = *sepp; 6559 (void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot, 6560 &slotseq, sessionid); 6561 KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot")); 6562 6563 /* Build the Sequence arguments. */ 6564 NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED); 6565 bcopy(sessionid, tl, NFSX_V4SESSIONID); 6566 tl += NFSX_V4SESSIONID / NFSX_UNSIGNED; 6567 nd->nd_slotseq = tl; 6568 *tl++ = txdr_unsigned(slotseq); 6569 *tl++ = txdr_unsigned(slotpos); 6570 *tl++ = txdr_unsigned(maxslot); 6571 if (dont_replycache == 0) 6572 *tl++ = newnfs_true; 6573 else 6574 *tl++ = newnfs_false; 6575 *tl = 0; /* No referring call list, for now. */ 6576 nd->nd_flag |= ND_HASSEQUENCE; 6577 return (0); 6578 } 6579 6580 /* 6581 * Get a session for the callback. 6582 */ 6583 static int 6584 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp) 6585 { 6586 struct nfsdsession *sep; 6587 6588 NFSLOCKSTATE(); 6589 LIST_FOREACH(sep, &clp->lc_session, sess_list) { 6590 if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) 6591 break; 6592 } 6593 if (sep == NULL) { 6594 NFSUNLOCKSTATE(); 6595 return (NFSERR_BADSESSION); 6596 } 6597 sep->sess_refcnt++; 6598 *sepp = sep; 6599 NFSUNLOCKSTATE(); 6600 return (0); 6601 } 6602 6603 /* 6604 * Free up all backchannel xprts. This needs to be done when the nfsd threads 6605 * exit, since those transports will all be going away. 6606 * This is only called after all the nfsd threads are done performing RPCs, 6607 * so locking shouldn't be an issue. 6608 */ 6609 APPLESTATIC void 6610 nfsrv_freeallbackchannel_xprts(void) 6611 { 6612 struct nfsdsession *sep; 6613 struct nfsclient *clp; 6614 SVCXPRT *xprt; 6615 int i; 6616 6617 for (i = 0; i < nfsrv_clienthashsize; i++) { 6618 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) { 6619 LIST_FOREACH(sep, &clp->lc_session, sess_list) { 6620 xprt = sep->sess_cbsess.nfsess_xprt; 6621 sep->sess_cbsess.nfsess_xprt = NULL; 6622 if (xprt != NULL) 6623 SVC_RELEASE(xprt); 6624 } 6625 } 6626 } 6627 } 6628 6629 /* 6630 * Do a layout commit. Actually just call nfsrv_updatemdsattr(). 6631 * I have no idea if the rest of these arguments will ever be useful? 6632 */ 6633 int 6634 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype, 6635 int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len, 6636 int hasnewmtime, struct timespec *newmtimep, int reclaim, 6637 nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep, 6638 uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p) 6639 { 6640 struct nfsvattr na; 6641 int error; 6642 6643 error = nfsrv_updatemdsattr(vp, &na, p); 6644 if (error == 0) { 6645 *hasnewsizep = 1; 6646 *newsizep = na.na_size; 6647 } 6648 return (error); 6649 } 6650 6651 /* 6652 * Try and get a layout. 6653 */ 6654 int 6655 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp, 6656 int layouttype, int *iomode, uint64_t *offset, uint64_t *len, 6657 uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose, 6658 int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p) 6659 { 6660 struct nfslayouthash *lhyp; 6661 struct nfslayout *lyp; 6662 char *devid; 6663 fhandle_t fh, *dsfhp; 6664 int error, mirrorcnt; 6665 6666 if (nfsrv_devidcnt == 0) 6667 return (NFSERR_UNKNLAYOUTTYPE); 6668 6669 if (*offset != 0) 6670 printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset, 6671 (uintmax_t)*len); 6672 error = nfsvno_getfh(vp, &fh, p); 6673 NFSD_DEBUG(4, "layoutget getfh=%d\n", error); 6674 if (error != 0) 6675 return (error); 6676 6677 /* 6678 * For now, all layouts are for entire files. 6679 * Only issue Read/Write layouts if requested for a non-readonly fs. 6680 */ 6681 if (NFSVNO_EXRDONLY(exp)) { 6682 if (*iomode == NFSLAYOUTIOMODE_RW) 6683 return (NFSERR_LAYOUTTRYLATER); 6684 *iomode = NFSLAYOUTIOMODE_READ; 6685 } 6686 if (*iomode != NFSLAYOUTIOMODE_RW) 6687 *iomode = NFSLAYOUTIOMODE_READ; 6688 6689 /* 6690 * Check to see if a write layout can be issued for this file. 6691 * This is used during mirror recovery to avoid RW layouts being 6692 * issued for a file while it is being copied to the recovered 6693 * mirror. 6694 */ 6695 if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0) 6696 return (NFSERR_LAYOUTTRYLATER); 6697 6698 *retonclose = 0; 6699 *offset = 0; 6700 *len = UINT64_MAX; 6701 6702 /* First, see if a layout already exists and return if found. */ 6703 lhyp = NFSLAYOUTHASH(&fh); 6704 NFSLOCKLAYOUT(lhyp); 6705 error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp); 6706 NFSD_DEBUG(4, "layoutget findlay=%d\n", error); 6707 /* 6708 * Not sure if the seqid must be the same, so I won't check it. 6709 */ 6710 if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] || 6711 stateidp->other[1] != lyp->lay_stateid.other[1] || 6712 stateidp->other[2] != lyp->lay_stateid.other[2])) { 6713 if ((lyp->lay_flags & NFSLAY_CALLB) == 0) { 6714 NFSUNLOCKLAYOUT(lhyp); 6715 NFSD_DEBUG(1, "ret bad stateid\n"); 6716 return (NFSERR_BADSTATEID); 6717 } 6718 /* 6719 * I believe we get here because there is a race between 6720 * the client processing the CBLAYOUTRECALL and the layout 6721 * being deleted here on the server. 6722 * The client has now done a LayoutGet with a non-layout 6723 * stateid, as it would when there is no layout. 6724 * As such, free this layout and set error == NFSERR_BADSTATEID 6725 * so the code below will create a new layout structure as 6726 * would happen if no layout was found. 6727 * "lyp" will be set before being used below, but set it NULL 6728 * as a safety belt. 6729 */ 6730 nfsrv_freelayout(&lhyp->list, lyp); 6731 lyp = NULL; 6732 error = NFSERR_BADSTATEID; 6733 } 6734 if (error == 0) { 6735 if (lyp->lay_layoutlen > maxcnt) { 6736 NFSUNLOCKLAYOUT(lhyp); 6737 NFSD_DEBUG(1, "ret layout too small\n"); 6738 return (NFSERR_TOOSMALL); 6739 } 6740 if (*iomode == NFSLAYOUTIOMODE_RW) 6741 lyp->lay_flags |= NFSLAY_RW; 6742 else 6743 lyp->lay_flags |= NFSLAY_READ; 6744 NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen); 6745 *layoutlenp = lyp->lay_layoutlen; 6746 if (++lyp->lay_stateid.seqid == 0) 6747 lyp->lay_stateid.seqid = 1; 6748 stateidp->seqid = lyp->lay_stateid.seqid; 6749 NFSUNLOCKLAYOUT(lhyp); 6750 NFSD_DEBUG(4, "ret fnd layout\n"); 6751 return (0); 6752 } 6753 NFSUNLOCKLAYOUT(lhyp); 6754 6755 /* Find the device id and file handle. */ 6756 dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK); 6757 devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK); 6758 error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid); 6759 NFSD_DEBUG(4, "layoutget devandfh=%d\n", error); 6760 if (error == 0) { 6761 if (layouttype == NFSLAYOUT_NFSV4_1_FILES) { 6762 if (NFSX_V4FILELAYOUT > maxcnt) 6763 error = NFSERR_TOOSMALL; 6764 else 6765 lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp, 6766 devid, vp->v_mount->mnt_stat.f_fsid); 6767 } else { 6768 if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt) 6769 error = NFSERR_TOOSMALL; 6770 else 6771 lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt, 6772 &fh, dsfhp, devid, 6773 vp->v_mount->mnt_stat.f_fsid); 6774 } 6775 } 6776 free(dsfhp, M_TEMP); 6777 free(devid, M_TEMP); 6778 if (error != 0) 6779 return (error); 6780 6781 /* 6782 * Now, add this layout to the list. 6783 */ 6784 error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p); 6785 NFSD_DEBUG(4, "layoutget addl=%d\n", error); 6786 /* 6787 * The lyp will be set to NULL by nfsrv_addlayout() if it 6788 * linked the new structure into the lists. 6789 */ 6790 free(lyp, M_NFSDSTATE); 6791 return (error); 6792 } 6793 6794 /* 6795 * Generate a File Layout. 6796 */ 6797 static struct nfslayout * 6798 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp, 6799 fhandle_t *dsfhp, char *devid, fsid_t fs) 6800 { 6801 uint32_t *tl; 6802 struct nfslayout *lyp; 6803 uint64_t pattern_offset; 6804 6805 lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE, 6806 M_WAITOK | M_ZERO); 6807 lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES; 6808 if (iomode == NFSLAYOUTIOMODE_RW) 6809 lyp->lay_flags = NFSLAY_RW; 6810 else 6811 lyp->lay_flags = NFSLAY_READ; 6812 NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp)); 6813 lyp->lay_clientid.qval = nd->nd_clientid.qval; 6814 lyp->lay_fsid = fs; 6815 6816 /* Fill in the xdr for the files layout. */ 6817 tl = (uint32_t *)lyp->lay_xdr; 6818 NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */ 6819 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); 6820 6821 /* 6822 * Make the stripe size as many 64K blocks as will fit in the stripe 6823 * mask. Since there is only one stripe, the stripe size doesn't really 6824 * matter, except that the Linux client will only handle an exact 6825 * multiple of their PAGE_SIZE (usually 4K). I chose 64K as a value 6826 * that should cover most/all arches w.r.t. PAGE_SIZE. 6827 */ 6828 *tl++ = txdr_unsigned(NFSFLAYUTIL_STRIPE_MASK & ~0xffff); 6829 *tl++ = 0; /* 1st stripe index. */ 6830 pattern_offset = 0; 6831 txdr_hyper(pattern_offset, tl); tl += 2; /* Pattern offset. */ 6832 *tl++ = txdr_unsigned(1); /* 1 file handle. */ 6833 *tl++ = txdr_unsigned(NFSX_V4PNFSFH); 6834 NFSBCOPY(dsfhp, tl, sizeof(*dsfhp)); 6835 lyp->lay_layoutlen = NFSX_V4FILELAYOUT; 6836 return (lyp); 6837 } 6838 6839 #define FLEX_OWNERID "999" 6840 #define FLEX_UID0 "0" 6841 /* 6842 * Generate a Flex File Layout. 6843 * The FLEX_OWNERID can be any string of 3 decimal digits. Although this 6844 * string goes on the wire, it isn't supposed to be used by the client, 6845 * since this server uses tight coupling. 6846 * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use 6847 * a string of "0". This works around the Linux Flex File Layout driver bug 6848 * which uses the synthetic uid/gid strings for the "tightly coupled" case. 6849 */ 6850 static struct nfslayout * 6851 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt, 6852 fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs) 6853 { 6854 uint32_t *tl; 6855 struct nfslayout *lyp; 6856 uint64_t lenval; 6857 int i; 6858 6859 lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt), 6860 M_NFSDSTATE, M_WAITOK | M_ZERO); 6861 lyp->lay_type = NFSLAYOUT_FLEXFILE; 6862 if (iomode == NFSLAYOUTIOMODE_RW) 6863 lyp->lay_flags = NFSLAY_RW; 6864 else 6865 lyp->lay_flags = NFSLAY_READ; 6866 NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp)); 6867 lyp->lay_clientid.qval = nd->nd_clientid.qval; 6868 lyp->lay_fsid = fs; 6869 lyp->lay_mirrorcnt = mirrorcnt; 6870 6871 /* Fill in the xdr for the files layout. */ 6872 tl = (uint32_t *)lyp->lay_xdr; 6873 lenval = 0; 6874 txdr_hyper(lenval, tl); tl += 2; /* Stripe unit. */ 6875 *tl++ = txdr_unsigned(mirrorcnt); /* # of mirrors. */ 6876 for (i = 0; i < mirrorcnt; i++) { 6877 *tl++ = txdr_unsigned(1); /* One stripe. */ 6878 NFSBCOPY(devid, tl, NFSX_V4DEVICEID); /* Device ID. */ 6879 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); 6880 devid += NFSX_V4DEVICEID; 6881 *tl++ = txdr_unsigned(1); /* Efficiency. */ 6882 *tl++ = 0; /* Proxy Stateid. */ 6883 *tl++ = 0x55555555; 6884 *tl++ = 0x55555555; 6885 *tl++ = 0x55555555; 6886 *tl++ = txdr_unsigned(1); /* 1 file handle. */ 6887 *tl++ = txdr_unsigned(NFSX_V4PNFSFH); 6888 NFSBCOPY(dsfhp, tl, sizeof(*dsfhp)); 6889 tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED); 6890 dsfhp++; 6891 if (nfsrv_flexlinuxhack != 0) { 6892 *tl++ = txdr_unsigned(strlen(FLEX_UID0)); 6893 *tl = 0; /* 0 pad string. */ 6894 NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0)); 6895 *tl++ = txdr_unsigned(strlen(FLEX_UID0)); 6896 *tl = 0; /* 0 pad string. */ 6897 NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0)); 6898 } else { 6899 *tl++ = txdr_unsigned(strlen(FLEX_OWNERID)); 6900 NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED); 6901 *tl++ = txdr_unsigned(strlen(FLEX_OWNERID)); 6902 NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED); 6903 } 6904 } 6905 *tl++ = txdr_unsigned(0); /* ff_flags. */ 6906 *tl = txdr_unsigned(60); /* Status interval hint. */ 6907 lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt); 6908 return (lyp); 6909 } 6910 6911 /* 6912 * Parse and process Flex File errors returned via LayoutReturn. 6913 */ 6914 static void 6915 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt, 6916 NFSPROC_T *p) 6917 { 6918 uint32_t *tl; 6919 int cnt, errcnt, i, j, opnum, stat; 6920 char devid[NFSX_V4DEVICEID]; 6921 6922 tl = layp; 6923 cnt = fxdr_unsigned(int, *tl++); 6924 NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt); 6925 for (i = 0; i < cnt; i++) { 6926 /* Skip offset, length and stateid for now. */ 6927 tl += (4 + NFSX_STATEID / NFSX_UNSIGNED); 6928 errcnt = fxdr_unsigned(int, *tl++); 6929 NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt); 6930 for (j = 0; j < errcnt; j++) { 6931 NFSBCOPY(tl, devid, NFSX_V4DEVICEID); 6932 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED); 6933 stat = fxdr_unsigned(int, *tl++); 6934 opnum = fxdr_unsigned(int, *tl++); 6935 NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum, 6936 stat); 6937 /* 6938 * Except for NFSERR_ACCES and NFSERR_STALE errors, 6939 * disable the mirror. 6940 */ 6941 if (stat != NFSERR_ACCES && stat != NFSERR_STALE) 6942 nfsrv_delds(devid, p); 6943 } 6944 } 6945 } 6946 6947 /* 6948 * This function removes all flex file layouts which has a mirror with 6949 * a device id that matches the argument. 6950 * Called when the DS represented by the device id has failed. 6951 */ 6952 void 6953 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p) 6954 { 6955 uint32_t *tl; 6956 struct nfslayout *lyp, *nlyp; 6957 struct nfslayouthash *lhyp; 6958 struct nfslayouthead loclyp; 6959 int i, j; 6960 6961 NFSD_DEBUG(4, "flexmirrordel\n"); 6962 /* Move all layouts found onto a local list. */ 6963 TAILQ_INIT(&loclyp); 6964 for (i = 0; i < nfsrv_layouthashsize; i++) { 6965 lhyp = &nfslayouthash[i]; 6966 NFSLOCKLAYOUT(lhyp); 6967 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { 6968 if (lyp->lay_type == NFSLAYOUT_FLEXFILE && 6969 lyp->lay_mirrorcnt > 1) { 6970 NFSD_DEBUG(4, "possible match\n"); 6971 tl = lyp->lay_xdr; 6972 tl += 3; 6973 for (j = 0; j < lyp->lay_mirrorcnt; j++) { 6974 tl++; 6975 if (NFSBCMP(devid, tl, NFSX_V4DEVICEID) 6976 == 0) { 6977 /* Found one. */ 6978 NFSD_DEBUG(4, "fnd one\n"); 6979 TAILQ_REMOVE(&lhyp->list, lyp, 6980 lay_list); 6981 TAILQ_INSERT_HEAD(&loclyp, lyp, 6982 lay_list); 6983 break; 6984 } 6985 tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED + 6986 NFSM_RNDUP(NFSX_V4PNFSFH) / 6987 NFSX_UNSIGNED + 11 * NFSX_UNSIGNED); 6988 } 6989 } 6990 } 6991 NFSUNLOCKLAYOUT(lhyp); 6992 } 6993 6994 /* Now, try to do a Layout recall for each one found. */ 6995 TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) { 6996 NFSD_DEBUG(4, "do layout recall\n"); 6997 /* 6998 * The layout stateid.seqid needs to be incremented 6999 * before doing a LAYOUT_RECALL callback. 7000 */ 7001 if (++lyp->lay_stateid.seqid == 0) 7002 lyp->lay_stateid.seqid = 1; 7003 nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid, 7004 &lyp->lay_fh, lyp, 1, lyp->lay_type, p); 7005 nfsrv_freelayout(&loclyp, lyp); 7006 } 7007 } 7008 7009 /* 7010 * Do a recall callback to the client for this layout. 7011 */ 7012 static int 7013 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp, 7014 struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p) 7015 { 7016 struct nfsclient *clp; 7017 int error; 7018 7019 NFSD_DEBUG(4, "nfsrv_recalllayout\n"); 7020 error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0), 7021 0, NULL, p); 7022 NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error); 7023 if (error != 0) { 7024 printf("nfsrv_recalllayout: getclient err=%d\n", error); 7025 return (error); 7026 } 7027 if ((clp->lc_flags & LCL_NFSV41) != 0) { 7028 error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL, 7029 stateidp, changed, fhp, NULL, NULL, laytype, p); 7030 /* If lyp != NULL, handle an error return here. */ 7031 if (error != 0 && lyp != NULL) { 7032 NFSDRECALLLOCK(); 7033 /* 7034 * Mark it returned, since no layout recall 7035 * has been done. 7036 * All errors seem to be non-recoverable, although 7037 * NFSERR_NOMATCHLAYOUT is a normal event. 7038 */ 7039 if ((lyp->lay_flags & NFSLAY_RECALL) != 0) { 7040 lyp->lay_flags |= NFSLAY_RETURNED; 7041 wakeup(lyp); 7042 } 7043 NFSDRECALLUNLOCK(); 7044 if (error != NFSERR_NOMATCHLAYOUT) 7045 printf("nfsrv_recalllayout: err=%d\n", error); 7046 } 7047 } else 7048 printf("nfsrv_recalllayout: clp not NFSv4.1\n"); 7049 return (error); 7050 } 7051 7052 /* 7053 * Find a layout to recall when we exceed our high water mark. 7054 */ 7055 void 7056 nfsrv_recalloldlayout(NFSPROC_T *p) 7057 { 7058 struct nfslayouthash *lhyp; 7059 struct nfslayout *lyp; 7060 nfsquad_t clientid; 7061 nfsv4stateid_t stateid; 7062 fhandle_t fh; 7063 int error, laytype, ret; 7064 7065 lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize]; 7066 NFSLOCKLAYOUT(lhyp); 7067 TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) { 7068 if ((lyp->lay_flags & NFSLAY_CALLB) == 0) { 7069 lyp->lay_flags |= NFSLAY_CALLB; 7070 /* 7071 * The layout stateid.seqid needs to be incremented 7072 * before doing a LAYOUT_RECALL callback. 7073 */ 7074 if (++lyp->lay_stateid.seqid == 0) 7075 lyp->lay_stateid.seqid = 1; 7076 clientid = lyp->lay_clientid; 7077 stateid = lyp->lay_stateid; 7078 NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh)); 7079 laytype = lyp->lay_type; 7080 break; 7081 } 7082 } 7083 NFSUNLOCKLAYOUT(lhyp); 7084 if (lyp != NULL) { 7085 error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0, 7086 laytype, p); 7087 if (error != 0 && error != NFSERR_NOMATCHLAYOUT) 7088 NFSD_DEBUG(4, "recallold=%d\n", error); 7089 if (error != 0) { 7090 NFSLOCKLAYOUT(lhyp); 7091 /* 7092 * Since the hash list was unlocked, we need to 7093 * find it again. 7094 */ 7095 ret = nfsrv_findlayout(&clientid, &fh, laytype, p, 7096 &lyp); 7097 if (ret == 0 && 7098 (lyp->lay_flags & NFSLAY_CALLB) != 0 && 7099 lyp->lay_stateid.other[0] == stateid.other[0] && 7100 lyp->lay_stateid.other[1] == stateid.other[1] && 7101 lyp->lay_stateid.other[2] == stateid.other[2]) { 7102 /* 7103 * The client no longer knows this layout, so 7104 * it can be free'd now. 7105 */ 7106 if (error == NFSERR_NOMATCHLAYOUT) 7107 nfsrv_freelayout(&lhyp->list, lyp); 7108 else { 7109 /* 7110 * Leave it to be tried later by 7111 * clearing NFSLAY_CALLB and moving 7112 * it to the head of the list, so it 7113 * won't be tried again for a while. 7114 */ 7115 lyp->lay_flags &= ~NFSLAY_CALLB; 7116 TAILQ_REMOVE(&lhyp->list, lyp, 7117 lay_list); 7118 TAILQ_INSERT_HEAD(&lhyp->list, lyp, 7119 lay_list); 7120 } 7121 } 7122 NFSUNLOCKLAYOUT(lhyp); 7123 } 7124 } 7125 } 7126 7127 /* 7128 * Try and return layout(s). 7129 */ 7130 int 7131 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp, 7132 int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim, 7133 int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp, 7134 struct ucred *cred, NFSPROC_T *p) 7135 { 7136 struct nfsvattr na; 7137 struct nfslayouthash *lhyp; 7138 struct nfslayout *lyp; 7139 fhandle_t fh; 7140 int error = 0; 7141 7142 *fndp = 0; 7143 if (kind == NFSV4LAYOUTRET_FILE) { 7144 error = nfsvno_getfh(vp, &fh, p); 7145 if (error == 0) { 7146 error = nfsrv_updatemdsattr(vp, &na, p); 7147 if (error != 0) 7148 printf("nfsrv_layoutreturn: updatemdsattr" 7149 " failed=%d\n", error); 7150 } 7151 if (error == 0) { 7152 if (reclaim == newnfs_true) { 7153 error = nfsrv_checkgrace(NULL, NULL, 7154 NFSLCK_RECLAIM); 7155 if (error != NFSERR_NOGRACE) 7156 error = 0; 7157 return (error); 7158 } 7159 lhyp = NFSLAYOUTHASH(&fh); 7160 NFSDRECALLLOCK(); 7161 NFSLOCKLAYOUT(lhyp); 7162 error = nfsrv_findlayout(&nd->nd_clientid, &fh, 7163 layouttype, p, &lyp); 7164 NFSD_DEBUG(4, "layoutret findlay=%d\n", error); 7165 if (error == 0 && 7166 stateidp->other[0] == lyp->lay_stateid.other[0] && 7167 stateidp->other[1] == lyp->lay_stateid.other[1] && 7168 stateidp->other[2] == lyp->lay_stateid.other[2]) { 7169 NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d" 7170 " %x %x %x laystateid %d %x %x %x" 7171 " off=%ju len=%ju flgs=0x%x\n", 7172 stateidp->seqid, stateidp->other[0], 7173 stateidp->other[1], stateidp->other[2], 7174 lyp->lay_stateid.seqid, 7175 lyp->lay_stateid.other[0], 7176 lyp->lay_stateid.other[1], 7177 lyp->lay_stateid.other[2], 7178 (uintmax_t)offset, (uintmax_t)len, 7179 lyp->lay_flags); 7180 if (++lyp->lay_stateid.seqid == 0) 7181 lyp->lay_stateid.seqid = 1; 7182 stateidp->seqid = lyp->lay_stateid.seqid; 7183 if (offset == 0 && len == UINT64_MAX) { 7184 if ((iomode & NFSLAYOUTIOMODE_READ) != 7185 0) 7186 lyp->lay_flags &= ~NFSLAY_READ; 7187 if ((iomode & NFSLAYOUTIOMODE_RW) != 0) 7188 lyp->lay_flags &= ~NFSLAY_RW; 7189 if ((lyp->lay_flags & (NFSLAY_READ | 7190 NFSLAY_RW)) == 0) 7191 nfsrv_freelayout(&lhyp->list, 7192 lyp); 7193 else 7194 *fndp = 1; 7195 } else 7196 *fndp = 1; 7197 } 7198 NFSUNLOCKLAYOUT(lhyp); 7199 /* Search the nfsrv_recalllist for a match. */ 7200 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { 7201 if (NFSBCMP(&lyp->lay_fh, &fh, 7202 sizeof(fh)) == 0 && 7203 lyp->lay_clientid.qval == 7204 nd->nd_clientid.qval && 7205 stateidp->other[0] == 7206 lyp->lay_stateid.other[0] && 7207 stateidp->other[1] == 7208 lyp->lay_stateid.other[1] && 7209 stateidp->other[2] == 7210 lyp->lay_stateid.other[2]) { 7211 lyp->lay_flags |= NFSLAY_RETURNED; 7212 wakeup(lyp); 7213 error = 0; 7214 } 7215 } 7216 NFSDRECALLUNLOCK(); 7217 } 7218 if (layouttype == NFSLAYOUT_FLEXFILE) 7219 nfsrv_flexlayouterr(nd, layp, maxcnt, p); 7220 } else if (kind == NFSV4LAYOUTRET_FSID) 7221 nfsrv_freelayouts(&nd->nd_clientid, 7222 &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode); 7223 else if (kind == NFSV4LAYOUTRET_ALL) 7224 nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode); 7225 else 7226 error = NFSERR_INVAL; 7227 if (error == -1) 7228 error = 0; 7229 return (error); 7230 } 7231 7232 /* 7233 * Look for an existing layout. 7234 */ 7235 static int 7236 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype, 7237 NFSPROC_T *p, struct nfslayout **lypp) 7238 { 7239 struct nfslayouthash *lhyp; 7240 struct nfslayout *lyp; 7241 int ret; 7242 7243 *lypp = NULL; 7244 ret = 0; 7245 lhyp = NFSLAYOUTHASH(fhp); 7246 TAILQ_FOREACH(lyp, &lhyp->list, lay_list) { 7247 if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 && 7248 lyp->lay_clientid.qval == clientidp->qval && 7249 lyp->lay_type == laytype) 7250 break; 7251 } 7252 if (lyp != NULL) 7253 *lypp = lyp; 7254 else 7255 ret = -1; 7256 return (ret); 7257 } 7258 7259 /* 7260 * Add the new layout, as required. 7261 */ 7262 static int 7263 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp, 7264 nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p) 7265 { 7266 struct nfsclient *clp; 7267 struct nfslayouthash *lhyp; 7268 struct nfslayout *lyp, *nlyp; 7269 fhandle_t *fhp; 7270 int error; 7271 7272 KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0, 7273 ("nfsrv_layoutget: no nd_clientid\n")); 7274 lyp = *lypp; 7275 fhp = &lyp->lay_fh; 7276 NFSLOCKSTATE(); 7277 error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp, 7278 NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p); 7279 if (error != 0) { 7280 NFSUNLOCKSTATE(); 7281 return (error); 7282 } 7283 lyp->lay_stateid.seqid = stateidp->seqid = 1; 7284 lyp->lay_stateid.other[0] = stateidp->other[0] = 7285 clp->lc_clientid.lval[0]; 7286 lyp->lay_stateid.other[1] = stateidp->other[1] = 7287 clp->lc_clientid.lval[1]; 7288 lyp->lay_stateid.other[2] = stateidp->other[2] = 7289 nfsrv_nextstateindex(clp); 7290 NFSUNLOCKSTATE(); 7291 7292 lhyp = NFSLAYOUTHASH(fhp); 7293 NFSLOCKLAYOUT(lhyp); 7294 TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) { 7295 if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 && 7296 nlyp->lay_clientid.qval == nd->nd_clientid.qval) 7297 break; 7298 } 7299 if (nlyp != NULL) { 7300 /* A layout already exists, so use it. */ 7301 nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)); 7302 NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen); 7303 *layoutlenp = nlyp->lay_layoutlen; 7304 if (++nlyp->lay_stateid.seqid == 0) 7305 nlyp->lay_stateid.seqid = 1; 7306 stateidp->seqid = nlyp->lay_stateid.seqid; 7307 stateidp->other[0] = nlyp->lay_stateid.other[0]; 7308 stateidp->other[1] = nlyp->lay_stateid.other[1]; 7309 stateidp->other[2] = nlyp->lay_stateid.other[2]; 7310 NFSUNLOCKLAYOUT(lhyp); 7311 return (0); 7312 } 7313 7314 /* Insert the new layout in the lists. */ 7315 *lypp = NULL; 7316 atomic_add_int(&nfsrv_layoutcnt, 1); 7317 NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen); 7318 *layoutlenp = lyp->lay_layoutlen; 7319 TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list); 7320 NFSUNLOCKLAYOUT(lhyp); 7321 return (0); 7322 } 7323 7324 /* 7325 * Get the devinfo for a deviceid. 7326 */ 7327 int 7328 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt, 7329 uint32_t *notify, int *devaddrlen, char **devaddr) 7330 { 7331 struct nfsdevice *ds; 7332 7333 if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype != 7334 NFSLAYOUT_FLEXFILE) || 7335 (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES)) 7336 return (NFSERR_UNKNLAYOUTTYPE); 7337 7338 /* 7339 * Now, search for the device id. Note that the structures won't go 7340 * away, but the order changes in the list. As such, the lock only 7341 * needs to be held during the search through the list. 7342 */ 7343 NFSDDSLOCK(); 7344 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7345 if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 && 7346 ds->nfsdev_nmp != NULL) 7347 break; 7348 } 7349 NFSDDSUNLOCK(); 7350 if (ds == NULL) 7351 return (NFSERR_NOENT); 7352 7353 /* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */ 7354 *devaddrlen = 0; 7355 if (layouttype == NFSLAYOUT_NFSV4_1_FILES) { 7356 *devaddrlen = ds->nfsdev_fileaddrlen; 7357 *devaddr = ds->nfsdev_fileaddr; 7358 } else if (layouttype == NFSLAYOUT_FLEXFILE) { 7359 *devaddrlen = ds->nfsdev_flexaddrlen; 7360 *devaddr = ds->nfsdev_flexaddr; 7361 } 7362 if (*devaddrlen == 0) 7363 return (NFSERR_UNKNLAYOUTTYPE); 7364 7365 /* 7366 * The XDR overhead is 3 unsigned values: layout_type, 7367 * length_of_address and notify bitmap. 7368 * If the notify array is changed to not all zeros, the 7369 * count of unsigned values must be increased. 7370 */ 7371 if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) + 7372 3 * NFSX_UNSIGNED) { 7373 *maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED; 7374 return (NFSERR_TOOSMALL); 7375 } 7376 return (0); 7377 } 7378 7379 /* 7380 * Free a list of layout state structures. 7381 */ 7382 static void 7383 nfsrv_freelayoutlist(nfsquad_t clientid) 7384 { 7385 struct nfslayouthash *lhyp; 7386 struct nfslayout *lyp, *nlyp; 7387 int i; 7388 7389 for (i = 0; i < nfsrv_layouthashsize; i++) { 7390 lhyp = &nfslayouthash[i]; 7391 NFSLOCKLAYOUT(lhyp); 7392 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { 7393 if (lyp->lay_clientid.qval == clientid.qval) 7394 nfsrv_freelayout(&lhyp->list, lyp); 7395 } 7396 NFSUNLOCKLAYOUT(lhyp); 7397 } 7398 } 7399 7400 /* 7401 * Free up a layout. 7402 */ 7403 static void 7404 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp) 7405 { 7406 7407 NFSD_DEBUG(4, "Freelayout=%p\n", lyp); 7408 atomic_add_int(&nfsrv_layoutcnt, -1); 7409 TAILQ_REMOVE(lhp, lyp, lay_list); 7410 free(lyp, M_NFSDSTATE); 7411 } 7412 7413 /* 7414 * Free up a device id. 7415 */ 7416 void 7417 nfsrv_freeonedevid(struct nfsdevice *ds) 7418 { 7419 int i; 7420 7421 atomic_add_int(&nfsrv_devidcnt, -1); 7422 vrele(ds->nfsdev_dvp); 7423 for (i = 0; i < nfsrv_dsdirsize; i++) 7424 if (ds->nfsdev_dsdir[i] != NULL) 7425 vrele(ds->nfsdev_dsdir[i]); 7426 free(ds->nfsdev_fileaddr, M_NFSDSTATE); 7427 free(ds->nfsdev_flexaddr, M_NFSDSTATE); 7428 free(ds->nfsdev_host, M_NFSDSTATE); 7429 free(ds, M_NFSDSTATE); 7430 } 7431 7432 /* 7433 * Free up a device id and its mirrors. 7434 */ 7435 static void 7436 nfsrv_freedevid(struct nfsdevice *ds) 7437 { 7438 7439 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list); 7440 nfsrv_freeonedevid(ds); 7441 } 7442 7443 /* 7444 * Free all layouts and device ids. 7445 * Done when the nfsd threads are shut down since there may be a new 7446 * modified device id list created when the nfsd is restarted. 7447 */ 7448 void 7449 nfsrv_freealllayoutsanddevids(void) 7450 { 7451 struct nfsdontlist *mrp, *nmrp; 7452 struct nfslayout *lyp, *nlyp; 7453 7454 /* Get rid of the deviceid structures. */ 7455 nfsrv_freealldevids(); 7456 TAILQ_INIT(&nfsrv_devidhead); 7457 nfsrv_devidcnt = 0; 7458 7459 /* Get rid of all layouts. */ 7460 nfsrv_freealllayouts(); 7461 7462 /* Get rid of any nfsdontlist entries. */ 7463 LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp) 7464 free(mrp, M_NFSDSTATE); 7465 LIST_INIT(&nfsrv_dontlisthead); 7466 nfsrv_dontlistlen = 0; 7467 7468 /* Free layouts in the recall list. */ 7469 TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp) 7470 nfsrv_freelayout(&nfsrv_recalllisthead, lyp); 7471 TAILQ_INIT(&nfsrv_recalllisthead); 7472 } 7473 7474 /* 7475 * Free layouts that match the arguments. 7476 */ 7477 static void 7478 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode) 7479 { 7480 struct nfslayouthash *lhyp; 7481 struct nfslayout *lyp, *nlyp; 7482 int i; 7483 7484 for (i = 0; i < nfsrv_layouthashsize; i++) { 7485 lhyp = &nfslayouthash[i]; 7486 NFSLOCKLAYOUT(lhyp); 7487 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { 7488 if (clid->qval != lyp->lay_clientid.qval) 7489 continue; 7490 if (fs != NULL && (fs->val[0] != lyp->lay_fsid.val[0] || 7491 fs->val[1] != lyp->lay_fsid.val[1])) 7492 continue; 7493 if (laytype != lyp->lay_type) 7494 continue; 7495 if ((iomode & NFSLAYOUTIOMODE_READ) != 0) 7496 lyp->lay_flags &= ~NFSLAY_READ; 7497 if ((iomode & NFSLAYOUTIOMODE_RW) != 0) 7498 lyp->lay_flags &= ~NFSLAY_RW; 7499 if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0) 7500 nfsrv_freelayout(&lhyp->list, lyp); 7501 } 7502 NFSUNLOCKLAYOUT(lhyp); 7503 } 7504 } 7505 7506 /* 7507 * Free all layouts for the argument file. 7508 */ 7509 void 7510 nfsrv_freefilelayouts(fhandle_t *fhp) 7511 { 7512 struct nfslayouthash *lhyp; 7513 struct nfslayout *lyp, *nlyp; 7514 7515 lhyp = NFSLAYOUTHASH(fhp); 7516 NFSLOCKLAYOUT(lhyp); 7517 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { 7518 if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0) 7519 nfsrv_freelayout(&lhyp->list, lyp); 7520 } 7521 NFSUNLOCKLAYOUT(lhyp); 7522 } 7523 7524 /* 7525 * Free all layouts. 7526 */ 7527 static void 7528 nfsrv_freealllayouts(void) 7529 { 7530 struct nfslayouthash *lhyp; 7531 struct nfslayout *lyp, *nlyp; 7532 int i; 7533 7534 for (i = 0; i < nfsrv_layouthashsize; i++) { 7535 lhyp = &nfslayouthash[i]; 7536 NFSLOCKLAYOUT(lhyp); 7537 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) 7538 nfsrv_freelayout(&lhyp->list, lyp); 7539 NFSUNLOCKLAYOUT(lhyp); 7540 } 7541 } 7542 7543 /* 7544 * Look up the mount path for the DS server. 7545 */ 7546 static int 7547 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p, 7548 struct nfsdevice **dsp) 7549 { 7550 struct nameidata nd; 7551 struct nfsdevice *ds; 7552 struct mount *mp; 7553 int error, i; 7554 char *dsdirpath; 7555 size_t dsdirsize; 7556 7557 NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp); 7558 *dsp = NULL; 7559 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, 7560 dspathp, p); 7561 error = namei(&nd); 7562 NFSD_DEBUG(4, "lookup=%d\n", error); 7563 if (error != 0) 7564 return (error); 7565 if (nd.ni_vp->v_type != VDIR) { 7566 vput(nd.ni_vp); 7567 NFSD_DEBUG(4, "dspath not dir\n"); 7568 return (ENOTDIR); 7569 } 7570 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { 7571 vput(nd.ni_vp); 7572 NFSD_DEBUG(4, "dspath not an NFS mount\n"); 7573 return (ENXIO); 7574 } 7575 7576 /* 7577 * Allocate a DS server structure with the NFS mounted directory 7578 * vnode reference counted, so that a non-forced dismount will 7579 * fail with EBUSY. 7580 * This structure is always linked into the list, even if an error 7581 * is being returned. The caller will free the entire list upon 7582 * an error return. 7583 */ 7584 *dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t), 7585 M_NFSDSTATE, M_WAITOK | M_ZERO); 7586 ds->nfsdev_dvp = nd.ni_vp; 7587 ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount); 7588 NFSVOPUNLOCK(nd.ni_vp, 0); 7589 7590 dsdirsize = strlen(dspathp) + 16; 7591 dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK); 7592 /* Now, create the DS directory structures. */ 7593 for (i = 0; i < nfsrv_dsdirsize; i++) { 7594 snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i); 7595 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 7596 UIO_SYSSPACE, dsdirpath, p); 7597 error = namei(&nd); 7598 NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error); 7599 if (error != 0) 7600 break; 7601 if (nd.ni_vp->v_type != VDIR) { 7602 vput(nd.ni_vp); 7603 error = ENOTDIR; 7604 NFSD_DEBUG(4, "dsdirpath not a VDIR\n"); 7605 break; 7606 } 7607 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { 7608 vput(nd.ni_vp); 7609 error = ENXIO; 7610 NFSD_DEBUG(4, "dsdirpath not an NFS mount\n"); 7611 break; 7612 } 7613 ds->nfsdev_dsdir[i] = nd.ni_vp; 7614 NFSVOPUNLOCK(nd.ni_vp, 0); 7615 } 7616 free(dsdirpath, M_TEMP); 7617 7618 if (strlen(mdspathp) > 0) { 7619 /* 7620 * This DS stores file for a specific MDS exported file 7621 * system. 7622 */ 7623 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 7624 UIO_SYSSPACE, mdspathp, p); 7625 error = namei(&nd); 7626 NFSD_DEBUG(4, "mds lookup=%d\n", error); 7627 if (error != 0) 7628 goto out; 7629 if (nd.ni_vp->v_type != VDIR) { 7630 vput(nd.ni_vp); 7631 error = ENOTDIR; 7632 NFSD_DEBUG(4, "mdspath not dir\n"); 7633 goto out; 7634 } 7635 mp = nd.ni_vp->v_mount; 7636 if ((mp->mnt_flag & MNT_EXPORTED) == 0) { 7637 vput(nd.ni_vp); 7638 error = ENXIO; 7639 NFSD_DEBUG(4, "mdspath not an exported fs\n"); 7640 goto out; 7641 } 7642 ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid; 7643 ds->nfsdev_mdsisset = 1; 7644 vput(nd.ni_vp); 7645 } 7646 7647 out: 7648 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list); 7649 atomic_add_int(&nfsrv_devidcnt, 1); 7650 return (error); 7651 } 7652 7653 /* 7654 * Look up the mount path for the DS server and delete it. 7655 */ 7656 int 7657 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p) 7658 { 7659 struct mount *mp; 7660 struct nfsmount *nmp; 7661 struct nfsdevice *ds; 7662 int error; 7663 7664 NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp); 7665 /* 7666 * Search for the path in the mount list. Avoid looking the path 7667 * up, since this mount point may be hung, with associated locked 7668 * vnodes, etc. 7669 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked 7670 * until this completes. 7671 * As noted in the man page, this should be done before any forced 7672 * dismount on the mount point, but at least the handshake on 7673 * NFSMNTP_CANCELRPCS should make it safe. 7674 */ 7675 error = 0; 7676 ds = NULL; 7677 nmp = NULL; 7678 mtx_lock(&mountlist_mtx); 7679 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 7680 if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 && 7681 strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 && 7682 mp->mnt_data != NULL) { 7683 nmp = VFSTONFS(mp); 7684 NFSLOCKMNT(nmp); 7685 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 7686 NFSMNTP_CANCELRPCS)) == 0) { 7687 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 7688 NFSUNLOCKMNT(nmp); 7689 } else { 7690 NFSUNLOCKMNT(nmp); 7691 nmp = NULL; 7692 } 7693 break; 7694 } 7695 } 7696 mtx_unlock(&mountlist_mtx); 7697 7698 if (nmp != NULL) { 7699 ds = nfsrv_deldsnmp(op, nmp, p); 7700 NFSD_DEBUG(4, "deldsnmp=%p\n", ds); 7701 if (ds != NULL) { 7702 nfsrv_killrpcs(nmp); 7703 NFSD_DEBUG(4, "aft killrpcs\n"); 7704 } else 7705 error = ENXIO; 7706 NFSLOCKMNT(nmp); 7707 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 7708 wakeup(nmp); 7709 NFSUNLOCKMNT(nmp); 7710 } else 7711 error = EINVAL; 7712 return (error); 7713 } 7714 7715 /* 7716 * Search for and remove a DS entry which matches the "nmp" argument. 7717 * The nfsdevice structure pointer is returned so that the caller can 7718 * free it via nfsrv_freeonedevid(). 7719 * For the forced case, do not try to do LayoutRecalls, since the server 7720 * must be shut down now anyhow. 7721 */ 7722 struct nfsdevice * 7723 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p) 7724 { 7725 struct nfsdevice *fndds; 7726 7727 NFSD_DEBUG(4, "deldsdvp\n"); 7728 NFSDDSLOCK(); 7729 if (op == PNFSDOP_FORCEDELDS) 7730 fndds = nfsv4_findmirror(nmp); 7731 else 7732 fndds = nfsrv_findmirroredds(nmp); 7733 if (fndds != NULL) 7734 nfsrv_deleteds(fndds); 7735 NFSDDSUNLOCK(); 7736 if (fndds != NULL) { 7737 if (op != PNFSDOP_FORCEDELDS) 7738 nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p); 7739 printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host); 7740 } 7741 return (fndds); 7742 } 7743 7744 /* 7745 * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid. 7746 * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount 7747 * point. 7748 * Also, returns an error instead of the nfsdevice found. 7749 */ 7750 static int 7751 nfsrv_delds(char *devid, NFSPROC_T *p) 7752 { 7753 struct nfsdevice *ds, *fndds; 7754 struct nfsmount *nmp; 7755 int fndmirror; 7756 7757 NFSD_DEBUG(4, "delds\n"); 7758 /* 7759 * Search the DS server list for a match with devid. 7760 * Remove the DS entry if found and there is a mirror. 7761 */ 7762 fndds = NULL; 7763 nmp = NULL; 7764 fndmirror = 0; 7765 NFSDDSLOCK(); 7766 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7767 if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 && 7768 ds->nfsdev_nmp != NULL) { 7769 NFSD_DEBUG(4, "fnd main ds\n"); 7770 fndds = ds; 7771 break; 7772 } 7773 } 7774 if (fndds == NULL) { 7775 NFSDDSUNLOCK(); 7776 return (ENXIO); 7777 } 7778 if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0) 7779 fndmirror = 1; 7780 else if (fndds->nfsdev_mdsisset != 0) { 7781 /* For the fsid is set case, search for a mirror. */ 7782 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 7783 if (ds != fndds && ds->nfsdev_nmp != NULL && 7784 ds->nfsdev_mdsisset != 0 && 7785 ds->nfsdev_mdsfsid.val[0] == 7786 fndds->nfsdev_mdsfsid.val[0] && 7787 ds->nfsdev_mdsfsid.val[1] == 7788 fndds->nfsdev_mdsfsid.val[1]) { 7789 fndmirror = 1; 7790 break; 7791 } 7792 } 7793 } 7794 if (fndmirror != 0) { 7795 nmp = fndds->nfsdev_nmp; 7796 NFSLOCKMNT(nmp); 7797 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM | 7798 NFSMNTP_CANCELRPCS)) == 0) { 7799 nmp->nm_privflag |= NFSMNTP_CANCELRPCS; 7800 NFSUNLOCKMNT(nmp); 7801 nfsrv_deleteds(fndds); 7802 } else { 7803 NFSUNLOCKMNT(nmp); 7804 nmp = NULL; 7805 } 7806 } 7807 NFSDDSUNLOCK(); 7808 if (nmp != NULL) { 7809 nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p); 7810 printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host); 7811 nfsrv_killrpcs(nmp); 7812 NFSLOCKMNT(nmp); 7813 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS; 7814 wakeup(nmp); 7815 NFSUNLOCKMNT(nmp); 7816 return (0); 7817 } 7818 return (ENXIO); 7819 } 7820 7821 /* 7822 * Mark a DS as disabled by setting nfsdev_nmp = NULL. 7823 */ 7824 static void 7825 nfsrv_deleteds(struct nfsdevice *fndds) 7826 { 7827 7828 NFSD_DEBUG(4, "deleteds: deleting a mirror\n"); 7829 fndds->nfsdev_nmp = NULL; 7830 if (fndds->nfsdev_mdsisset == 0) 7831 nfsrv_faildscnt--; 7832 } 7833 7834 /* 7835 * Fill in the addr structures for the File and Flex File layouts. 7836 */ 7837 static void 7838 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost) 7839 { 7840 uint32_t *tl; 7841 char *netprot; 7842 int addrlen; 7843 static uint64_t new_devid = 0; 7844 7845 if (strchr(addr, ':') != NULL) 7846 netprot = "tcp6"; 7847 else 7848 netprot = "tcp"; 7849 7850 /* Fill in the device id. */ 7851 NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time)); 7852 new_devid++; 7853 NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)], 7854 sizeof(new_devid)); 7855 7856 /* 7857 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4 7858 * as defined in RFC5661) in XDR. 7859 */ 7860 addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) + 7861 6 * NFSX_UNSIGNED; 7862 NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot); 7863 ds->nfsdev_fileaddrlen = addrlen; 7864 tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO); 7865 ds->nfsdev_fileaddr = (char *)tl; 7866 *tl++ = txdr_unsigned(1); /* One stripe with index 0. */ 7867 *tl++ = 0; 7868 *tl++ = txdr_unsigned(1); /* One multipath list */ 7869 *tl++ = txdr_unsigned(1); /* with one entry in it. */ 7870 /* The netaddr for this one entry. */ 7871 *tl++ = txdr_unsigned(strlen(netprot)); 7872 NFSBCOPY(netprot, tl, strlen(netprot)); 7873 tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED); 7874 *tl++ = txdr_unsigned(strlen(addr)); 7875 NFSBCOPY(addr, tl, strlen(addr)); 7876 7877 /* 7878 * Fill in the flex file addr (actually the ff_device_addr4 7879 * as defined for Flexible File Layout) in XDR. 7880 */ 7881 addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) + 7882 9 * NFSX_UNSIGNED; 7883 ds->nfsdev_flexaddrlen = addrlen; 7884 tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO); 7885 ds->nfsdev_flexaddr = (char *)tl; 7886 *tl++ = txdr_unsigned(1); /* One multipath entry. */ 7887 /* The netaddr for this one entry. */ 7888 *tl++ = txdr_unsigned(strlen(netprot)); 7889 NFSBCOPY(netprot, tl, strlen(netprot)); 7890 tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED); 7891 *tl++ = txdr_unsigned(strlen(addr)); 7892 NFSBCOPY(addr, tl, strlen(addr)); 7893 tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED); 7894 *tl++ = txdr_unsigned(1); /* One NFS Version. */ 7895 *tl++ = txdr_unsigned(NFS_VER4); /* NFSv4. */ 7896 *tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */ 7897 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max rsize. */ 7898 *tl++ = txdr_unsigned(NFS_SRVMAXIO); /* DS max wsize. */ 7899 *tl = newnfs_true; /* Tightly coupled. */ 7900 7901 ds->nfsdev_hostnamelen = strlen(dnshost); 7902 ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE, 7903 M_WAITOK); 7904 NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1); 7905 } 7906 7907 7908 /* 7909 * Create the device id list. 7910 * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument 7911 * is misconfigured. 7912 */ 7913 int 7914 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p) 7915 { 7916 struct nfsdevice *ds; 7917 char *addrp, *dnshostp, *dspathp, *mdspathp; 7918 int error, i; 7919 7920 addrp = args->addr; 7921 dnshostp = args->dnshost; 7922 dspathp = args->dspath; 7923 mdspathp = args->mdspath; 7924 nfsrv_maxpnfsmirror = args->mirrorcnt; 7925 if (addrp == NULL || dnshostp == NULL || dspathp == NULL || 7926 mdspathp == NULL) 7927 return (0); 7928 7929 /* 7930 * Loop around for each nul-terminated string in args->addr, 7931 * args->dnshost, args->dnspath and args->mdspath. 7932 */ 7933 while (addrp < (args->addr + args->addrlen) && 7934 dnshostp < (args->dnshost + args->dnshostlen) && 7935 dspathp < (args->dspath + args->dspathlen) && 7936 mdspathp < (args->mdspath + args->mdspathlen)) { 7937 error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds); 7938 if (error != 0) { 7939 /* Free all DS servers. */ 7940 nfsrv_freealldevids(); 7941 nfsrv_devidcnt = 0; 7942 return (ENXIO); 7943 } 7944 nfsrv_allocdevid(ds, addrp, dnshostp); 7945 addrp += (strlen(addrp) + 1); 7946 dnshostp += (strlen(dnshostp) + 1); 7947 dspathp += (strlen(dspathp) + 1); 7948 mdspathp += (strlen(mdspathp) + 1); 7949 } 7950 if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) { 7951 /* Free all DS servers. */ 7952 nfsrv_freealldevids(); 7953 nfsrv_devidcnt = 0; 7954 nfsrv_maxpnfsmirror = 1; 7955 return (ENXIO); 7956 } 7957 /* We can fail at most one less DS than the mirror level. */ 7958 nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1; 7959 7960 /* 7961 * Allocate the nfslayout hash table now, since this is a pNFS server. 7962 * Make it 1% of the high water mark and at least 100. 7963 */ 7964 if (nfslayouthash == NULL) { 7965 nfsrv_layouthashsize = nfsrv_layouthighwater / 100; 7966 if (nfsrv_layouthashsize < 100) 7967 nfsrv_layouthashsize = 100; 7968 nfslayouthash = mallocarray(nfsrv_layouthashsize, 7969 sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK | 7970 M_ZERO); 7971 for (i = 0; i < nfsrv_layouthashsize; i++) { 7972 mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF); 7973 TAILQ_INIT(&nfslayouthash[i].list); 7974 } 7975 } 7976 return (0); 7977 } 7978 7979 /* 7980 * Free all device ids. 7981 */ 7982 static void 7983 nfsrv_freealldevids(void) 7984 { 7985 struct nfsdevice *ds, *nds; 7986 7987 TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds) 7988 nfsrv_freedevid(ds); 7989 } 7990 7991 /* 7992 * Check to see if there is a Read/Write Layout plus either: 7993 * - A Write Delegation 7994 * or 7995 * - An Open with Write_access. 7996 * Return 1 if this is the case and 0 otherwise. 7997 * This function is used by nfsrv_proxyds() to decide if doing a Proxy 7998 * Getattr RPC to the Data Server (DS) is necessary. 7999 */ 8000 #define NFSCLIDVECSIZE 6 8001 APPLESTATIC int 8002 nfsrv_checkdsattr(struct nfsrv_descript *nd, vnode_t vp, NFSPROC_T *p) 8003 { 8004 fhandle_t fh, *tfhp; 8005 struct nfsstate *stp; 8006 struct nfslayout *lyp; 8007 struct nfslayouthash *lhyp; 8008 struct nfslockhashhead *hp; 8009 struct nfslockfile *lfp; 8010 nfsquad_t clid[NFSCLIDVECSIZE]; 8011 int clidcnt, ret; 8012 8013 ret = nfsvno_getfh(vp, &fh, p); 8014 if (ret != 0) 8015 return (0); 8016 8017 /* First check for a Read/Write Layout. */ 8018 clidcnt = 0; 8019 lhyp = NFSLAYOUTHASH(&fh); 8020 NFSLOCKLAYOUT(lhyp); 8021 TAILQ_FOREACH(lyp, &lhyp->list, lay_list) { 8022 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && 8023 ((lyp->lay_flags & NFSLAY_RW) != 0 || 8024 ((lyp->lay_flags & NFSLAY_READ) != 0 && 8025 nfsrv_pnfsatime != 0))) { 8026 if (clidcnt < NFSCLIDVECSIZE) 8027 clid[clidcnt].qval = lyp->lay_clientid.qval; 8028 clidcnt++; 8029 } 8030 } 8031 NFSUNLOCKLAYOUT(lhyp); 8032 if (clidcnt == 0) { 8033 /* None found, so return 0. */ 8034 return (0); 8035 } 8036 8037 /* Get the nfslockfile for this fh. */ 8038 NFSLOCKSTATE(); 8039 hp = NFSLOCKHASH(&fh); 8040 LIST_FOREACH(lfp, hp, lf_hash) { 8041 tfhp = &lfp->lf_fh; 8042 if (NFSVNO_CMPFH(&fh, tfhp)) 8043 break; 8044 } 8045 if (lfp == NULL) { 8046 /* None found, so return 0. */ 8047 NFSUNLOCKSTATE(); 8048 return (0); 8049 } 8050 8051 /* Now, look for a Write delegation for this clientid. */ 8052 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) { 8053 if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 && 8054 nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0) 8055 break; 8056 } 8057 if (stp != NULL) { 8058 /* Found one, so return 1. */ 8059 NFSUNLOCKSTATE(); 8060 return (1); 8061 } 8062 8063 /* No Write delegation, so look for an Open with Write_access. */ 8064 LIST_FOREACH(stp, &lfp->lf_open, ls_file) { 8065 KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0, 8066 ("nfsrv_checkdsattr: Non-open in Open list\n")); 8067 if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 && 8068 nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0) 8069 break; 8070 } 8071 NFSUNLOCKSTATE(); 8072 if (stp != NULL) 8073 return (1); 8074 return (0); 8075 } 8076 8077 /* 8078 * Look for a matching clientid in the vector. Return 1 if one might match. 8079 */ 8080 static int 8081 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt) 8082 { 8083 int i; 8084 8085 /* If too many for the vector, return 1 since there might be a match. */ 8086 if (clidcnt > NFSCLIDVECSIZE) 8087 return (1); 8088 8089 for (i = 0; i < clidcnt; i++) 8090 if (clidvec[i].qval == clid.qval) 8091 return (1); 8092 return (0); 8093 } 8094 8095 /* 8096 * Check the don't list for "vp" and see if issuing an rw layout is allowed. 8097 * Return 1 if issuing an rw layout isn't allowed, 0 otherwise. 8098 */ 8099 static int 8100 nfsrv_dontlayout(fhandle_t *fhp) 8101 { 8102 struct nfsdontlist *mrp; 8103 int ret; 8104 8105 if (nfsrv_dontlistlen == 0) 8106 return (0); 8107 ret = 0; 8108 NFSDDONTLISTLOCK(); 8109 LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) { 8110 if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 && 8111 (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) { 8112 ret = 1; 8113 break; 8114 } 8115 } 8116 NFSDDONTLISTUNLOCK(); 8117 return (ret); 8118 } 8119 8120 #define PNFSDS_COPYSIZ 65536 8121 /* 8122 * Create a new file on a DS and copy the contents of an extant DS file to it. 8123 * This can be used for recovery of a DS file onto a recovered DS. 8124 * The steps are: 8125 * - When called, the MDS file's vnode is locked, blocking LayoutGet operations. 8126 * - Disable issuing of read/write layouts for the file via the nfsdontlist, 8127 * so that they will be disabled after the MDS file's vnode is unlocked. 8128 * - Set up the nfsrv_recalllist so that recall of read/write layouts can 8129 * be done. 8130 * - Unlock the MDS file's vnode, so that the client(s) can perform proxied 8131 * writes, LayoutCommits and LayoutReturns for the file when completing the 8132 * LayoutReturn requested by the LayoutRecall callback. 8133 * - Issue a LayoutRecall callback for all read/write layouts and wait for 8134 * them to be returned. (If the LayoutRecall callback replies 8135 * NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.) 8136 * - Exclusively lock the MDS file's vnode. This ensures that no proxied 8137 * writes are in progress or can occur during the DS file copy. 8138 * It also blocks Setattr operations. 8139 * - Create the file on the recovered mirror. 8140 * - Copy the file from the operational DS. 8141 * - Copy any ACL from the MDS file to the new DS file. 8142 * - Set the modify time of the new DS file to that of the MDS file. 8143 * - Update the extended attribute for the MDS file. 8144 * - Enable issuing of rw layouts by deleting the nfsdontlist entry. 8145 * - The caller will unlock the MDS file's vnode allowing operations 8146 * to continue normally, since it is now on the mirror again. 8147 */ 8148 int 8149 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds, 8150 struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt, 8151 struct ucred *cred, NFSPROC_T *p) 8152 { 8153 struct nfsdontlist *mrp, *nmrp; 8154 struct nfslayouthash *lhyp; 8155 struct nfslayout *lyp, *nlyp; 8156 struct nfslayouthead thl; 8157 struct mount *mp, *tvmp; 8158 struct acl *aclp; 8159 struct vattr va; 8160 struct timespec mtime; 8161 fhandle_t fh; 8162 vnode_t tvp; 8163 off_t rdpos, wrpos; 8164 ssize_t aresid; 8165 char *dat; 8166 int didprintf, ret, retacl, xfer; 8167 8168 ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp"); 8169 ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp"); 8170 /* 8171 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag 8172 * so that no more RW layouts will get issued. 8173 */ 8174 ret = nfsvno_getfh(vp, &fh, p); 8175 if (ret != 0) { 8176 NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret); 8177 return (ret); 8178 } 8179 nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK); 8180 nmrp->nfsmr_flags = NFSMR_DONTLAYOUT; 8181 NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh)); 8182 NFSDDONTLISTLOCK(); 8183 LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) { 8184 if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0) 8185 break; 8186 } 8187 if (mrp == NULL) { 8188 LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list); 8189 mrp = nmrp; 8190 nmrp = NULL; 8191 nfsrv_dontlistlen++; 8192 NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n"); 8193 } else { 8194 NFSDDONTLISTUNLOCK(); 8195 free(nmrp, M_NFSDSTATE); 8196 NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n"); 8197 return (ENXIO); 8198 } 8199 NFSDDONTLISTUNLOCK(); 8200 8201 /* 8202 * Search for all RW layouts for this file. Move them to the 8203 * recall list, so they can be recalled and their return noted. 8204 */ 8205 lhyp = NFSLAYOUTHASH(&fh); 8206 NFSDRECALLLOCK(); 8207 NFSLOCKLAYOUT(lhyp); 8208 TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) { 8209 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && 8210 (lyp->lay_flags & NFSLAY_RW) != 0) { 8211 TAILQ_REMOVE(&lhyp->list, lyp, lay_list); 8212 TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list); 8213 lyp->lay_trycnt = 0; 8214 } 8215 } 8216 NFSUNLOCKLAYOUT(lhyp); 8217 NFSDRECALLUNLOCK(); 8218 8219 ret = 0; 8220 mp = tvmp = NULL; 8221 didprintf = 0; 8222 TAILQ_INIT(&thl); 8223 /* Unlock the MDS vp, so that a LayoutReturn can be done on it. */ 8224 NFSVOPUNLOCK(vp, 0); 8225 /* Now, do a recall for all layouts not yet recalled. */ 8226 tryagain: 8227 NFSDRECALLLOCK(); 8228 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { 8229 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 && 8230 (lyp->lay_flags & NFSLAY_RECALL) == 0) { 8231 lyp->lay_flags |= NFSLAY_RECALL; 8232 /* 8233 * The layout stateid.seqid needs to be incremented 8234 * before doing a LAYOUT_RECALL callback. 8235 */ 8236 if (++lyp->lay_stateid.seqid == 0) 8237 lyp->lay_stateid.seqid = 1; 8238 NFSDRECALLUNLOCK(); 8239 nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid, 8240 &lyp->lay_fh, lyp, 0, lyp->lay_type, p); 8241 NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n"); 8242 goto tryagain; 8243 } 8244 } 8245 8246 /* Now wait for them to be returned. */ 8247 tryagain2: 8248 TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) { 8249 if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) { 8250 if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) { 8251 TAILQ_REMOVE(&nfsrv_recalllisthead, lyp, 8252 lay_list); 8253 TAILQ_INSERT_HEAD(&thl, lyp, lay_list); 8254 NFSD_DEBUG(4, 8255 "nfsrv_copymr: layout returned\n"); 8256 } else { 8257 lyp->lay_trycnt++; 8258 ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR, 8259 PVFS | PCATCH, "nfsmrl", hz); 8260 NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n", 8261 ret); 8262 if (ret == EINTR || ret == ERESTART) 8263 break; 8264 if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) { 8265 /* 8266 * Give up after 60sec and return 8267 * ENXIO, failing the copymr. 8268 * This layout will remain on the 8269 * recalllist. It can only be cleared 8270 * by restarting the nfsd. 8271 * This seems the safe way to handle 8272 * it, since it cannot be safely copied 8273 * with an outstanding RW layout. 8274 */ 8275 if (lyp->lay_trycnt >= 60) { 8276 ret = ENXIO; 8277 break; 8278 } 8279 if (didprintf == 0) { 8280 printf("nfsrv_copymr: layout " 8281 "not returned\n"); 8282 didprintf = 1; 8283 } 8284 } 8285 } 8286 goto tryagain2; 8287 } 8288 } 8289 NFSDRECALLUNLOCK(); 8290 /* We can now get rid of the layouts that have been returned. */ 8291 TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp) 8292 nfsrv_freelayout(&thl, lyp); 8293 8294 /* 8295 * Do the vn_start_write() calls here, before the MDS vnode is 8296 * locked and the tvp is created (locked) in the NFS file system 8297 * that dvp is in. 8298 * For tvmp, this probably isn't necessary, since it will be an 8299 * NFS mount and they are not suspendable at this time. 8300 */ 8301 if (ret == 0) 8302 ret = vn_start_write(vp, &mp, V_WAIT | PCATCH); 8303 if (ret == 0) { 8304 tvmp = dvp->v_mount; 8305 ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH); 8306 } 8307 8308 /* 8309 * LK_EXCLUSIVE lock the MDS vnode, so that any 8310 * proxied writes through the MDS will be blocked until we have 8311 * completed the copy and update of the extended attributes. 8312 * This will also ensure that any attributes and ACL will not be 8313 * changed until the copy is complete. 8314 */ 8315 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY); 8316 if (ret == 0 && (vp->v_iflag & VI_DOOMED) != 0) { 8317 NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n"); 8318 ret = ESTALE; 8319 } 8320 8321 /* Create the data file on the recovered DS. */ 8322 if (ret == 0) 8323 ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp); 8324 8325 /* Copy the DS file, if created successfully. */ 8326 if (ret == 0) { 8327 /* 8328 * Get any NFSv4 ACL on the MDS file, so that it can be set 8329 * on the new DS file. 8330 */ 8331 aclp = acl_alloc(M_WAITOK | M_ZERO); 8332 retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p); 8333 if (retacl != 0 && retacl != ENOATTR) 8334 NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl); 8335 dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK); 8336 /* Malloc a block of 0s used to check for holes. */ 8337 if (nfsrv_zeropnfsdat == NULL) 8338 nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP, 8339 M_WAITOK | M_ZERO); 8340 rdpos = wrpos = 0; 8341 ret = VOP_GETATTR(fvp, &va, cred); 8342 aresid = 0; 8343 while (ret == 0 && aresid == 0) { 8344 ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ, 8345 rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL, 8346 &aresid, p); 8347 xfer = PNFSDS_COPYSIZ - aresid; 8348 if (ret == 0 && xfer > 0) { 8349 rdpos += xfer; 8350 /* 8351 * Skip the write for holes, except for the 8352 * last block. 8353 */ 8354 if (xfer < PNFSDS_COPYSIZ || rdpos == 8355 va.va_size || NFSBCMP(dat, 8356 nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0) 8357 ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer, 8358 wrpos, UIO_SYSSPACE, IO_NODELOCKED, 8359 cred, NULL, NULL, p); 8360 if (ret == 0) 8361 wrpos += xfer; 8362 } 8363 } 8364 8365 /* If there is an ACL and the copy succeeded, set the ACL. */ 8366 if (ret == 0 && retacl == 0) { 8367 ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p); 8368 /* 8369 * Don't consider these as errors, since VOP_GETACL() 8370 * can return an ACL when they are not actually 8371 * supported. For example, for UFS, VOP_GETACL() 8372 * will return a trivial ACL based on the uid/gid/mode 8373 * when there is no ACL on the file. 8374 * This case should be recognized as a trivial ACL 8375 * by UFS's VOP_SETACL() and succeed, but... 8376 */ 8377 if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM) 8378 ret = 0; 8379 } 8380 8381 if (ret == 0) 8382 ret = VOP_FSYNC(tvp, MNT_WAIT, p); 8383 8384 /* Set the DS data file's modify time that of the MDS file. */ 8385 if (ret == 0) 8386 ret = VOP_GETATTR(vp, &va, cred); 8387 if (ret == 0) { 8388 mtime = va.va_mtime; 8389 VATTR_NULL(&va); 8390 va.va_mtime = mtime; 8391 ret = VOP_SETATTR(tvp, &va, cred); 8392 } 8393 8394 vput(tvp); 8395 acl_free(aclp); 8396 free(dat, M_TEMP); 8397 } 8398 if (tvmp != NULL) 8399 vn_finished_write(tvmp); 8400 8401 /* Update the extended attributes for the newly created DS file. */ 8402 if (ret == 0) 8403 ret = vn_extattr_set(vp, IO_NODELOCKED, 8404 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", 8405 sizeof(*wpf) * mirrorcnt, (char *)wpf, p); 8406 if (mp != NULL) 8407 vn_finished_write(mp); 8408 8409 /* Get rid of the dontlist entry, so that Layouts can be issued. */ 8410 NFSDDONTLISTLOCK(); 8411 LIST_REMOVE(mrp, nfsmr_list); 8412 NFSDDONTLISTUNLOCK(); 8413 free(mrp, M_NFSDSTATE); 8414 return (ret); 8415 } 8416 8417 /* 8418 * Create a data storage file on the recovered DS. 8419 */ 8420 static int 8421 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf, 8422 vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p, 8423 vnode_t *tvpp) 8424 { 8425 struct vattr va, nva; 8426 int error; 8427 8428 /* Make data file name based on FH. */ 8429 error = VOP_GETATTR(vp, &va, cred); 8430 if (error == 0) { 8431 /* Set the attributes for "vp" to Setattr the DS vp. */ 8432 VATTR_NULL(&nva); 8433 nva.va_uid = va.va_uid; 8434 nva.va_gid = va.va_gid; 8435 nva.va_mode = va.va_mode; 8436 nva.va_size = 0; 8437 VATTR_NULL(&va); 8438 va.va_type = VREG; 8439 va.va_mode = nva.va_mode; 8440 NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf); 8441 error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL, 8442 pf->dsf_filename, cred, p, tvpp); 8443 } 8444 return (error); 8445 } 8446 8447 /* 8448 * Look up the MDS file shared locked, and then get the extended attribute 8449 * to find the extant DS file to be copied to the new mirror. 8450 * If successful, *vpp is set to the MDS file's vp and *nvpp is 8451 * set to a DS data file for the MDS file, both exclusively locked. 8452 * The "buf" argument has the pnfsdsfile structure from the MDS file 8453 * in it and buflen is set to its length. 8454 */ 8455 int 8456 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf, 8457 int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp, 8458 struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp, 8459 struct nfsdevice **fdsp) 8460 { 8461 struct nameidata nd; 8462 struct vnode *vp, *curvp; 8463 struct pnfsdsfile *pf; 8464 struct nfsmount *nmp, *curnmp; 8465 int dsdir, error, mirrorcnt, ippos; 8466 8467 vp = NULL; 8468 curvp = NULL; 8469 curnmp = NULL; 8470 *dsp = NULL; 8471 *fdsp = NULL; 8472 if (dspathp == NULL && curdspathp != NULL) 8473 return (EPERM); 8474 8475 /* 8476 * Look up the MDS file shared locked. The lock will be upgraded 8477 * to an exclusive lock after any rw layouts have been returned. 8478 */ 8479 NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp); 8480 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, 8481 mdspathp, p); 8482 error = namei(&nd); 8483 NFSD_DEBUG(4, "lookup=%d\n", error); 8484 if (error != 0) 8485 return (error); 8486 if (nd.ni_vp->v_type != VREG) { 8487 vput(nd.ni_vp); 8488 NFSD_DEBUG(4, "mdspath not reg\n"); 8489 return (EISDIR); 8490 } 8491 vp = nd.ni_vp; 8492 8493 if (curdspathp != NULL) { 8494 /* 8495 * Look up the current DS path and find the nfsdev structure for 8496 * it. 8497 */ 8498 NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp); 8499 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 8500 UIO_SYSSPACE, curdspathp, p); 8501 error = namei(&nd); 8502 NFSD_DEBUG(4, "ds lookup=%d\n", error); 8503 if (error != 0) { 8504 vput(vp); 8505 return (error); 8506 } 8507 if (nd.ni_vp->v_type != VDIR) { 8508 vput(nd.ni_vp); 8509 vput(vp); 8510 NFSD_DEBUG(4, "curdspath not dir\n"); 8511 return (ENOTDIR); 8512 } 8513 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { 8514 vput(nd.ni_vp); 8515 vput(vp); 8516 NFSD_DEBUG(4, "curdspath not an NFS mount\n"); 8517 return (ENXIO); 8518 } 8519 curnmp = VFSTONFS(nd.ni_vp->v_mount); 8520 8521 /* Search the nfsdev list for a match. */ 8522 NFSDDSLOCK(); 8523 *fdsp = nfsv4_findmirror(curnmp); 8524 NFSDDSUNLOCK(); 8525 if (*fdsp == NULL) 8526 curnmp = NULL; 8527 if (curnmp == NULL) { 8528 vput(nd.ni_vp); 8529 vput(vp); 8530 NFSD_DEBUG(4, "mdscopymr: no current ds\n"); 8531 return (ENXIO); 8532 } 8533 curvp = nd.ni_vp; 8534 } 8535 8536 if (dspathp != NULL) { 8537 /* Look up the nfsdev path and find the nfsdev structure. */ 8538 NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp); 8539 NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, 8540 UIO_SYSSPACE, dspathp, p); 8541 error = namei(&nd); 8542 NFSD_DEBUG(4, "ds lookup=%d\n", error); 8543 if (error != 0) { 8544 vput(vp); 8545 if (curvp != NULL) 8546 vput(curvp); 8547 return (error); 8548 } 8549 if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) { 8550 vput(nd.ni_vp); 8551 vput(vp); 8552 if (curvp != NULL) 8553 vput(curvp); 8554 NFSD_DEBUG(4, "dspath not dir\n"); 8555 if (nd.ni_vp == curvp) 8556 return (EPERM); 8557 return (ENOTDIR); 8558 } 8559 if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) { 8560 vput(nd.ni_vp); 8561 vput(vp); 8562 if (curvp != NULL) 8563 vput(curvp); 8564 NFSD_DEBUG(4, "dspath not an NFS mount\n"); 8565 return (ENXIO); 8566 } 8567 nmp = VFSTONFS(nd.ni_vp->v_mount); 8568 8569 /* 8570 * Search the nfsdevice list for a match. If curnmp == NULL, 8571 * this is a recovery and there must be a mirror. 8572 */ 8573 NFSDDSLOCK(); 8574 if (curnmp == NULL) 8575 *dsp = nfsrv_findmirroredds(nmp); 8576 else 8577 *dsp = nfsv4_findmirror(nmp); 8578 NFSDDSUNLOCK(); 8579 if (*dsp == NULL) { 8580 vput(nd.ni_vp); 8581 vput(vp); 8582 if (curvp != NULL) 8583 vput(curvp); 8584 NFSD_DEBUG(4, "mdscopymr: no ds\n"); 8585 return (ENXIO); 8586 } 8587 } else { 8588 nd.ni_vp = NULL; 8589 nmp = NULL; 8590 } 8591 8592 /* 8593 * Get a vp for an available DS data file using the extended 8594 * attribute on the MDS file. 8595 * If there is a valid entry for the new DS in the extended attribute 8596 * on the MDS file (as checked via the nmp argument), 8597 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur. 8598 */ 8599 error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p, 8600 NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir); 8601 if (curvp != NULL) 8602 vput(curvp); 8603 if (nd.ni_vp == NULL) { 8604 if (error == 0 && nmp != NULL) { 8605 /* Search the nfsdev list for a match. */ 8606 NFSDDSLOCK(); 8607 *dsp = nfsrv_findmirroredds(nmp); 8608 NFSDDSUNLOCK(); 8609 } 8610 if (error == 0 && (nmp == NULL || *dsp == NULL)) { 8611 if (nvpp != NULL && *nvpp != NULL) { 8612 vput(*nvpp); 8613 *nvpp = NULL; 8614 } 8615 error = ENXIO; 8616 } 8617 } else 8618 vput(nd.ni_vp); 8619 8620 /* 8621 * When dspathp != NULL and curdspathp == NULL, this is a recovery 8622 * and is only allowed if there is a 0.0.0.0 IP address entry. 8623 * When curdspathp != NULL, the ippos will be set to that entry. 8624 */ 8625 if (error == 0 && dspathp != NULL && ippos == -1) { 8626 if (nvpp != NULL && *nvpp != NULL) { 8627 vput(*nvpp); 8628 *nvpp = NULL; 8629 } 8630 error = ENXIO; 8631 } 8632 if (error == 0) { 8633 *vpp = vp; 8634 8635 pf = (struct pnfsdsfile *)buf; 8636 if (ippos == -1) { 8637 /* If no zeroip pnfsdsfile, add one. */ 8638 ippos = *buflenp / sizeof(*pf); 8639 *buflenp += sizeof(*pf); 8640 pf += ippos; 8641 pf->dsf_dir = dsdir; 8642 strlcpy(pf->dsf_filename, fname, 8643 sizeof(pf->dsf_filename)); 8644 } else 8645 pf += ippos; 8646 *pfp = pf; 8647 } else 8648 vput(vp); 8649 return (error); 8650 } 8651 8652 /* 8653 * Search for a matching pnfsd mirror device structure, base on the nmp arg. 8654 * Return one if found, NULL otherwise. 8655 */ 8656 static struct nfsdevice * 8657 nfsrv_findmirroredds(struct nfsmount *nmp) 8658 { 8659 struct nfsdevice *ds, *fndds; 8660 int fndmirror; 8661 8662 mtx_assert(NFSDDSMUTEXPTR, MA_OWNED); 8663 /* 8664 * Search the DS server list for a match with nmp. 8665 * Remove the DS entry if found and there is a mirror. 8666 */ 8667 fndds = NULL; 8668 fndmirror = 0; 8669 if (nfsrv_devidcnt == 0) 8670 return (fndds); 8671 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 8672 if (ds->nfsdev_nmp == nmp) { 8673 NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n"); 8674 fndds = ds; 8675 break; 8676 } 8677 } 8678 if (fndds == NULL) 8679 return (fndds); 8680 if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0) 8681 fndmirror = 1; 8682 else if (fndds->nfsdev_mdsisset != 0) { 8683 /* For the fsid is set case, search for a mirror. */ 8684 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) { 8685 if (ds != fndds && ds->nfsdev_nmp != NULL && 8686 ds->nfsdev_mdsisset != 0 && 8687 ds->nfsdev_mdsfsid.val[0] == 8688 fndds->nfsdev_mdsfsid.val[0] && 8689 ds->nfsdev_mdsfsid.val[1] == 8690 fndds->nfsdev_mdsfsid.val[1]) { 8691 fndmirror = 1; 8692 break; 8693 } 8694 } 8695 } 8696 if (fndmirror == 0) { 8697 NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n"); 8698 return (NULL); 8699 } 8700 return (fndds); 8701 } 8702 8703