1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* 27 * Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T. 28 * All Rights Reserved 29 */ 30 31 #include <sys/param.h> 32 #include <sys/types.h> 33 #include <sys/systm.h> 34 #include <sys/cred.h> 35 #include <sys/vfs.h> 36 #include <sys/vfs_opreg.h> 37 #include <sys/vnode.h> 38 #include <sys/pathname.h> 39 #include <sys/sysmacros.h> 40 #include <sys/kmem.h> 41 #include <sys/mkdev.h> 42 #include <sys/mount.h> 43 #include <sys/statvfs.h> 44 #include <sys/errno.h> 45 #include <sys/debug.h> 46 #include <sys/cmn_err.h> 47 #include <sys/utsname.h> 48 #include <sys/bootconf.h> 49 #include <sys/modctl.h> 50 #include <sys/acl.h> 51 #include <sys/flock.h> 52 #include <sys/time.h> 53 #include <sys/disp.h> 54 #include <sys/policy.h> 55 #include <sys/socket.h> 56 #include <sys/netconfig.h> 57 #include <sys/dnlc.h> 58 #include <sys/list.h> 59 #include <sys/mntent.h> 60 #include <sys/tsol/label.h> 61 62 #include <rpc/types.h> 63 #include <rpc/auth.h> 64 #include <rpc/rpcsec_gss.h> 65 #include <rpc/clnt.h> 66 67 #include <nfs/nfs.h> 68 #include <nfs/nfs_clnt.h> 69 #include <nfs/mount.h> 70 #include <nfs/nfs_acl.h> 71 72 #include <fs/fs_subr.h> 73 74 #include <nfs/nfs4.h> 75 #include <nfs/rnode4.h> 76 #include <nfs/nfs4_clnt.h> 77 #include <sys/fs/autofs.h> 78 79 80 /* 81 * Arguments passed to thread to free data structures from forced unmount. 82 */ 83 84 typedef struct { 85 vfs_t *fm_vfsp; 86 int fm_flag; 87 cred_t *fm_cr; 88 } freemountargs_t; 89 90 static void async_free_mount(vfs_t *, int, cred_t *); 91 static void nfs4_free_mount(vfs_t *, int, cred_t *); 92 static void nfs4_free_mount_thread(freemountargs_t *); 93 static int nfs4_chkdup_servinfo4(servinfo4_t *, servinfo4_t *); 94 95 /* 96 * From rpcsec module (common/rpcsec). 97 */ 98 extern int sec_clnt_loadinfo(struct sec_data *, struct sec_data **, model_t); 99 extern void sec_clnt_freeinfo(struct sec_data *); 100 101 /* 102 * The order and contents of this structure must be kept in sync with that of 103 * rfsreqcnt_v4_tmpl in nfs_stats.c 104 */ 105 static char *rfsnames_v4[] = { 106 "null", "compound", "reserved", "access", "close", "commit", "create", 107 "delegpurge", "delegreturn", "getattr", "getfh", "link", "lock", 108 "lockt", "locku", "lookup", "lookupp", "nverify", "open", "openattr", 109 "open_confirm", "open_downgrade", "putfh", "putpubfh", "putrootfh", 110 "read", "readdir", "readlink", "remove", "rename", "renew", 111 "restorefh", "savefh", "secinfo", "setattr", "setclientid", 112 "setclientid_confirm", "verify", "write" 113 }; 114 115 /* 116 * nfs4_max_mount_retry is the number of times the client will redrive 117 * a mount compound before giving up and returning failure. The intent 118 * is to redrive mount compounds which fail NFS4ERR_STALE so that 119 * if a component of the server path being mounted goes stale, it can 120 * "recover" by redriving the mount compund (LOOKUP ops). This recovery 121 * code is needed outside of the recovery framework because mount is a 122 * special case. The client doesn't create vnodes/rnodes for components 123 * of the server path being mounted. The recovery code recovers real 124 * client objects, not STALE FHs which map to components of the server 125 * path being mounted. 126 * 127 * We could just fail the mount on the first time, but that would 128 * instantly trigger failover (from nfs4_mount), and the client should 129 * try to re-lookup the STALE FH before doing failover. The easiest 130 * way to "re-lookup" is to simply redrive the mount compound. 131 */ 132 static int nfs4_max_mount_retry = 2; 133 134 /* 135 * nfs4 vfs operations. 136 */ 137 int nfs4_mount(vfs_t *, vnode_t *, struct mounta *, cred_t *); 138 static int nfs4_unmount(vfs_t *, int, cred_t *); 139 static int nfs4_root(vfs_t *, vnode_t **); 140 static int nfs4_statvfs(vfs_t *, struct statvfs64 *); 141 static int nfs4_sync(vfs_t *, short, cred_t *); 142 static int nfs4_vget(vfs_t *, vnode_t **, fid_t *); 143 static int nfs4_mountroot(vfs_t *, whymountroot_t); 144 static void nfs4_freevfs(vfs_t *); 145 146 static int nfs4rootvp(vnode_t **, vfs_t *, struct servinfo4 *, 147 int, cred_t *, zone_t *); 148 149 vfsops_t *nfs4_vfsops; 150 151 int nfs4_vfsinit(void); 152 void nfs4_vfsfini(void); 153 static void nfs4setclientid_init(void); 154 static void nfs4setclientid_fini(void); 155 static void nfs4setclientid_otw(mntinfo4_t *, servinfo4_t *, cred_t *, 156 struct nfs4_server *, nfs4_error_t *, int *); 157 static void destroy_nfs4_server(nfs4_server_t *); 158 static void remove_mi(nfs4_server_t *, mntinfo4_t *); 159 160 extern void nfs4_ephemeral_init(void); 161 extern void nfs4_ephemeral_fini(void); 162 163 /* 164 * Initialize the vfs structure 165 */ 166 167 static int nfs4fstyp; 168 169 170 /* 171 * Debug variable to check for rdma based 172 * transport startup and cleanup. Controlled 173 * through /etc/system. Off by default. 174 */ 175 extern int rdma_debug; 176 177 int 178 nfs4init(int fstyp, char *name) 179 { 180 static const fs_operation_def_t nfs4_vfsops_template[] = { 181 VFSNAME_MOUNT, { .vfs_mount = nfs4_mount }, 182 VFSNAME_UNMOUNT, { .vfs_unmount = nfs4_unmount }, 183 VFSNAME_ROOT, { .vfs_root = nfs4_root }, 184 VFSNAME_STATVFS, { .vfs_statvfs = nfs4_statvfs }, 185 VFSNAME_SYNC, { .vfs_sync = nfs4_sync }, 186 VFSNAME_VGET, { .vfs_vget = nfs4_vget }, 187 VFSNAME_MOUNTROOT, { .vfs_mountroot = nfs4_mountroot }, 188 VFSNAME_FREEVFS, { .vfs_freevfs = nfs4_freevfs }, 189 NULL, NULL 190 }; 191 int error; 192 193 nfs4_vfsops = NULL; 194 nfs4_vnodeops = NULL; 195 nfs4_trigger_vnodeops = NULL; 196 197 error = vfs_setfsops(fstyp, nfs4_vfsops_template, &nfs4_vfsops); 198 if (error != 0) { 199 zcmn_err(GLOBAL_ZONEID, CE_WARN, 200 "nfs4init: bad vfs ops template"); 201 goto out; 202 } 203 204 error = vn_make_ops(name, nfs4_vnodeops_template, &nfs4_vnodeops); 205 if (error != 0) { 206 zcmn_err(GLOBAL_ZONEID, CE_WARN, 207 "nfs4init: bad vnode ops template"); 208 goto out; 209 } 210 211 error = vn_make_ops("nfs4_trigger", nfs4_trigger_vnodeops_template, 212 &nfs4_trigger_vnodeops); 213 if (error != 0) { 214 zcmn_err(GLOBAL_ZONEID, CE_WARN, 215 "nfs4init: bad trigger vnode ops template"); 216 goto out; 217 } 218 219 nfs4fstyp = fstyp; 220 (void) nfs4_vfsinit(); 221 (void) nfs4_init_dot_entries(); 222 223 out: 224 if (error) { 225 if (nfs4_trigger_vnodeops != NULL) 226 vn_freevnodeops(nfs4_trigger_vnodeops); 227 228 if (nfs4_vnodeops != NULL) 229 vn_freevnodeops(nfs4_vnodeops); 230 231 (void) vfs_freevfsops_by_type(fstyp); 232 } 233 234 return (error); 235 } 236 237 void 238 nfs4fini(void) 239 { 240 (void) nfs4_destroy_dot_entries(); 241 nfs4_vfsfini(); 242 } 243 244 /* 245 * Create a new sec_data structure to store AUTH_DH related data: 246 * netname, syncaddr, knetconfig. There is no AUTH_F_RPCTIMESYNC 247 * flag set for NFS V4 since we are avoiding to contact the rpcbind 248 * daemon and is using the IP time service (IPPORT_TIMESERVER). 249 * 250 * sec_data can be freed by sec_clnt_freeinfo(). 251 */ 252 static struct sec_data * 253 create_authdh_data(char *netname, int nlen, struct netbuf *syncaddr, 254 struct knetconfig *knconf) { 255 struct sec_data *secdata; 256 dh_k4_clntdata_t *data; 257 char *pf, *p; 258 259 if (syncaddr == NULL || syncaddr->buf == NULL || nlen == 0) 260 return (NULL); 261 262 secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP); 263 secdata->flags = 0; 264 265 data = kmem_alloc(sizeof (*data), KM_SLEEP); 266 267 data->syncaddr.maxlen = syncaddr->maxlen; 268 data->syncaddr.len = syncaddr->len; 269 data->syncaddr.buf = (char *)kmem_alloc(syncaddr->len, KM_SLEEP); 270 bcopy(syncaddr->buf, data->syncaddr.buf, syncaddr->len); 271 272 /* 273 * duplicate the knconf information for the 274 * new opaque data. 275 */ 276 data->knconf = kmem_alloc(sizeof (*knconf), KM_SLEEP); 277 *data->knconf = *knconf; 278 pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP); 279 p = kmem_alloc(KNC_STRSIZE, KM_SLEEP); 280 bcopy(knconf->knc_protofmly, pf, KNC_STRSIZE); 281 bcopy(knconf->knc_proto, p, KNC_STRSIZE); 282 data->knconf->knc_protofmly = pf; 283 data->knconf->knc_proto = p; 284 285 /* move server netname to the sec_data structure */ 286 data->netname = kmem_alloc(nlen, KM_SLEEP); 287 bcopy(netname, data->netname, nlen); 288 data->netnamelen = (int)nlen; 289 290 secdata->secmod = AUTH_DH; 291 secdata->rpcflavor = AUTH_DH; 292 secdata->data = (caddr_t)data; 293 294 return (secdata); 295 } 296 297 /* 298 * Returns (deep) copy of sec_data_t. Allocates all memory required; caller 299 * is responsible for freeing. 300 */ 301 sec_data_t * 302 copy_sec_data(sec_data_t *fsecdata) { 303 sec_data_t *tsecdata; 304 305 if (fsecdata == NULL) 306 return (NULL); 307 308 if (fsecdata->rpcflavor == AUTH_DH) { 309 dh_k4_clntdata_t *fdata = (dh_k4_clntdata_t *)fsecdata->data; 310 311 if (fdata == NULL) 312 return (NULL); 313 314 tsecdata = (sec_data_t *)create_authdh_data(fdata->netname, 315 fdata->netnamelen, &fdata->syncaddr, fdata->knconf); 316 317 return (tsecdata); 318 } 319 320 tsecdata = kmem_zalloc(sizeof (sec_data_t), KM_SLEEP); 321 322 tsecdata->secmod = fsecdata->secmod; 323 tsecdata->rpcflavor = fsecdata->rpcflavor; 324 tsecdata->flags = fsecdata->flags; 325 tsecdata->uid = fsecdata->uid; 326 327 if (fsecdata->rpcflavor == RPCSEC_GSS) { 328 gss_clntdata_t *gcd = (gss_clntdata_t *)fsecdata->data; 329 330 tsecdata->data = (caddr_t)copy_sec_data_gss(gcd); 331 } else { 332 tsecdata->data = NULL; 333 } 334 335 return (tsecdata); 336 } 337 338 gss_clntdata_t * 339 copy_sec_data_gss(gss_clntdata_t *fdata) 340 { 341 gss_clntdata_t *tdata; 342 343 if (fdata == NULL) 344 return (NULL); 345 346 tdata = kmem_zalloc(sizeof (gss_clntdata_t), KM_SLEEP); 347 348 tdata->mechanism.length = fdata->mechanism.length; 349 tdata->mechanism.elements = kmem_zalloc(fdata->mechanism.length, 350 KM_SLEEP); 351 bcopy(fdata->mechanism.elements, tdata->mechanism.elements, 352 fdata->mechanism.length); 353 354 tdata->service = fdata->service; 355 356 (void) strcpy(tdata->uname, fdata->uname); 357 (void) strcpy(tdata->inst, fdata->inst); 358 (void) strcpy(tdata->realm, fdata->realm); 359 360 tdata->qop = fdata->qop; 361 362 return (tdata); 363 } 364 365 static int 366 nfs4_chkdup_servinfo4(servinfo4_t *svp_head, servinfo4_t *svp) 367 { 368 servinfo4_t *si; 369 370 /* 371 * Iterate over the servinfo4 list to make sure 372 * we do not have a duplicate. Skip any servinfo4 373 * that has been marked "NOT IN USE" 374 */ 375 for (si = svp_head; si; si = si->sv_next) { 376 (void) nfs_rw_enter_sig(&si->sv_lock, RW_READER, 0); 377 if (si->sv_flags & SV4_NOTINUSE) { 378 nfs_rw_exit(&si->sv_lock); 379 continue; 380 } 381 nfs_rw_exit(&si->sv_lock); 382 if (si == svp) 383 continue; 384 if (si->sv_addr.len == svp->sv_addr.len && 385 strcmp(si->sv_knconf->knc_protofmly, 386 svp->sv_knconf->knc_protofmly) == 0 && 387 bcmp(si->sv_addr.buf, svp->sv_addr.buf, 388 si->sv_addr.len) == 0) { 389 /* it's a duplicate */ 390 return (1); 391 } 392 } 393 /* it's not a duplicate */ 394 return (0); 395 } 396 397 void 398 nfs4_free_args(struct nfs_args *nargs) 399 { 400 if (nargs->knconf) { 401 if (nargs->knconf->knc_protofmly) 402 kmem_free(nargs->knconf->knc_protofmly, 403 KNC_STRSIZE); 404 if (nargs->knconf->knc_proto) 405 kmem_free(nargs->knconf->knc_proto, KNC_STRSIZE); 406 kmem_free(nargs->knconf, sizeof (*nargs->knconf)); 407 nargs->knconf = NULL; 408 } 409 410 if (nargs->fh) { 411 kmem_free(nargs->fh, strlen(nargs->fh) + 1); 412 nargs->fh = NULL; 413 } 414 415 if (nargs->hostname) { 416 kmem_free(nargs->hostname, strlen(nargs->hostname) + 1); 417 nargs->hostname = NULL; 418 } 419 420 if (nargs->addr) { 421 if (nargs->addr->buf) { 422 ASSERT(nargs->addr->len); 423 kmem_free(nargs->addr->buf, nargs->addr->len); 424 } 425 kmem_free(nargs->addr, sizeof (struct netbuf)); 426 nargs->addr = NULL; 427 } 428 429 if (nargs->syncaddr) { 430 ASSERT(nargs->syncaddr->len); 431 if (nargs->syncaddr->buf) { 432 ASSERT(nargs->syncaddr->len); 433 kmem_free(nargs->syncaddr->buf, nargs->syncaddr->len); 434 } 435 kmem_free(nargs->syncaddr, sizeof (struct netbuf)); 436 nargs->syncaddr = NULL; 437 } 438 439 if (nargs->netname) { 440 kmem_free(nargs->netname, strlen(nargs->netname) + 1); 441 nargs->netname = NULL; 442 } 443 444 if (nargs->nfs_ext_u.nfs_extA.secdata) { 445 sec_clnt_freeinfo( 446 nargs->nfs_ext_u.nfs_extA.secdata); 447 nargs->nfs_ext_u.nfs_extA.secdata = NULL; 448 } 449 } 450 451 452 int 453 nfs4_copyin(char *data, int datalen, struct nfs_args *nargs) 454 { 455 456 int error; 457 size_t hlen; /* length of hostname */ 458 size_t nlen; /* length of netname */ 459 char netname[MAXNETNAMELEN+1]; /* server's netname */ 460 struct netbuf addr; /* server's address */ 461 struct netbuf syncaddr; /* AUTH_DES time sync addr */ 462 struct knetconfig *knconf; /* transport structure */ 463 struct sec_data *secdata = NULL; /* security data */ 464 STRUCT_DECL(nfs_args, args); /* nfs mount arguments */ 465 STRUCT_DECL(knetconfig, knconf_tmp); 466 STRUCT_DECL(netbuf, addr_tmp); 467 int flags; 468 char *p, *pf; 469 struct pathname pn; 470 char *userbufptr; 471 472 473 bzero(nargs, sizeof (*nargs)); 474 475 STRUCT_INIT(args, get_udatamodel()); 476 bzero(STRUCT_BUF(args), SIZEOF_STRUCT(nfs_args, DATAMODEL_NATIVE)); 477 if (copyin(data, STRUCT_BUF(args), MIN(datalen, 478 STRUCT_SIZE(args)))) 479 return (EFAULT); 480 481 nargs->wsize = STRUCT_FGET(args, wsize); 482 nargs->rsize = STRUCT_FGET(args, rsize); 483 nargs->timeo = STRUCT_FGET(args, timeo); 484 nargs->retrans = STRUCT_FGET(args, retrans); 485 nargs->acregmin = STRUCT_FGET(args, acregmin); 486 nargs->acregmax = STRUCT_FGET(args, acregmax); 487 nargs->acdirmin = STRUCT_FGET(args, acdirmin); 488 nargs->acdirmax = STRUCT_FGET(args, acdirmax); 489 490 flags = STRUCT_FGET(args, flags); 491 nargs->flags = flags; 492 493 addr.buf = NULL; 494 syncaddr.buf = NULL; 495 496 497 /* 498 * Allocate space for a knetconfig structure and 499 * its strings and copy in from user-land. 500 */ 501 knconf = kmem_zalloc(sizeof (*knconf), KM_SLEEP); 502 STRUCT_INIT(knconf_tmp, get_udatamodel()); 503 if (copyin(STRUCT_FGETP(args, knconf), STRUCT_BUF(knconf_tmp), 504 STRUCT_SIZE(knconf_tmp))) { 505 kmem_free(knconf, sizeof (*knconf)); 506 return (EFAULT); 507 } 508 509 knconf->knc_semantics = STRUCT_FGET(knconf_tmp, knc_semantics); 510 knconf->knc_protofmly = STRUCT_FGETP(knconf_tmp, knc_protofmly); 511 knconf->knc_proto = STRUCT_FGETP(knconf_tmp, knc_proto); 512 if (get_udatamodel() != DATAMODEL_LP64) { 513 knconf->knc_rdev = expldev(STRUCT_FGET(knconf_tmp, knc_rdev)); 514 } else { 515 knconf->knc_rdev = STRUCT_FGET(knconf_tmp, knc_rdev); 516 } 517 518 pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP); 519 p = kmem_alloc(KNC_STRSIZE, KM_SLEEP); 520 error = copyinstr(knconf->knc_protofmly, pf, KNC_STRSIZE, NULL); 521 if (error) { 522 kmem_free(pf, KNC_STRSIZE); 523 kmem_free(p, KNC_STRSIZE); 524 kmem_free(knconf, sizeof (*knconf)); 525 return (error); 526 } 527 528 error = copyinstr(knconf->knc_proto, p, KNC_STRSIZE, NULL); 529 if (error) { 530 kmem_free(pf, KNC_STRSIZE); 531 kmem_free(p, KNC_STRSIZE); 532 kmem_free(knconf, sizeof (*knconf)); 533 return (error); 534 } 535 536 537 knconf->knc_protofmly = pf; 538 knconf->knc_proto = p; 539 540 nargs->knconf = knconf; 541 542 /* 543 * Get server address 544 */ 545 STRUCT_INIT(addr_tmp, get_udatamodel()); 546 if (copyin(STRUCT_FGETP(args, addr), STRUCT_BUF(addr_tmp), 547 STRUCT_SIZE(addr_tmp))) { 548 error = EFAULT; 549 goto errout; 550 } 551 552 nargs->addr = kmem_alloc(sizeof (struct netbuf), KM_SLEEP); 553 userbufptr = STRUCT_FGETP(addr_tmp, buf); 554 addr.len = STRUCT_FGET(addr_tmp, len); 555 addr.buf = kmem_alloc(addr.len, KM_SLEEP); 556 addr.maxlen = addr.len; 557 if (copyin(userbufptr, addr.buf, addr.len)) { 558 kmem_free(addr.buf, addr.len); 559 error = EFAULT; 560 goto errout; 561 } 562 bcopy(&addr, nargs->addr, sizeof (struct netbuf)); 563 564 /* 565 * Get the root fhandle 566 */ 567 error = pn_get(STRUCT_FGETP(args, fh), UIO_USERSPACE, &pn); 568 if (error) 569 goto errout; 570 571 /* Volatile fh: keep server paths, so use actual-size strings */ 572 nargs->fh = kmem_alloc(pn.pn_pathlen + 1, KM_SLEEP); 573 bcopy(pn.pn_path, nargs->fh, pn.pn_pathlen); 574 nargs->fh[pn.pn_pathlen] = '\0'; 575 pn_free(&pn); 576 577 578 /* 579 * Get server's hostname 580 */ 581 if (flags & NFSMNT_HOSTNAME) { 582 error = copyinstr(STRUCT_FGETP(args, hostname), 583 netname, sizeof (netname), &hlen); 584 if (error) 585 goto errout; 586 nargs->hostname = kmem_zalloc(hlen, KM_SLEEP); 587 (void) strcpy(nargs->hostname, netname); 588 589 } else { 590 nargs->hostname = NULL; 591 } 592 593 594 /* 595 * If there are syncaddr and netname data, load them in. This is 596 * to support data needed for NFSV4 when AUTH_DH is the negotiated 597 * flavor via SECINFO. (instead of using MOUNT protocol in V3). 598 */ 599 netname[0] = '\0'; 600 if (flags & NFSMNT_SECURE) { 601 602 /* get syncaddr */ 603 STRUCT_INIT(addr_tmp, get_udatamodel()); 604 if (copyin(STRUCT_FGETP(args, syncaddr), STRUCT_BUF(addr_tmp), 605 STRUCT_SIZE(addr_tmp))) { 606 error = EINVAL; 607 goto errout; 608 } 609 userbufptr = STRUCT_FGETP(addr_tmp, buf); 610 syncaddr.len = STRUCT_FGET(addr_tmp, len); 611 syncaddr.buf = kmem_alloc(syncaddr.len, KM_SLEEP); 612 syncaddr.maxlen = syncaddr.len; 613 if (copyin(userbufptr, syncaddr.buf, syncaddr.len)) { 614 kmem_free(syncaddr.buf, syncaddr.len); 615 error = EFAULT; 616 goto errout; 617 } 618 619 nargs->syncaddr = kmem_alloc(sizeof (struct netbuf), KM_SLEEP); 620 bcopy(&syncaddr, nargs->syncaddr, sizeof (struct netbuf)); 621 622 /* get server's netname */ 623 if (copyinstr(STRUCT_FGETP(args, netname), netname, 624 sizeof (netname), &nlen)) { 625 error = EFAULT; 626 goto errout; 627 } 628 629 netname[nlen] = '\0'; 630 nargs->netname = kmem_zalloc(nlen, KM_SLEEP); 631 (void) strcpy(nargs->netname, netname); 632 } 633 634 /* 635 * Get the extention data which has the security data structure. 636 * This includes data for AUTH_SYS as well. 637 */ 638 if (flags & NFSMNT_NEWARGS) { 639 nargs->nfs_args_ext = STRUCT_FGET(args, nfs_args_ext); 640 if (nargs->nfs_args_ext == NFS_ARGS_EXTA || 641 nargs->nfs_args_ext == NFS_ARGS_EXTB) { 642 /* 643 * Indicating the application is using the new 644 * sec_data structure to pass in the security 645 * data. 646 */ 647 if (STRUCT_FGETP(args, 648 nfs_ext_u.nfs_extA.secdata) != NULL) { 649 error = sec_clnt_loadinfo( 650 (struct sec_data *)STRUCT_FGETP(args, 651 nfs_ext_u.nfs_extA.secdata), 652 &secdata, get_udatamodel()); 653 } 654 nargs->nfs_ext_u.nfs_extA.secdata = secdata; 655 } 656 } 657 658 if (error) 659 goto errout; 660 661 /* 662 * Failover support: 663 * 664 * We may have a linked list of nfs_args structures, 665 * which means the user is looking for failover. If 666 * the mount is either not "read-only" or "soft", 667 * we want to bail out with EINVAL. 668 */ 669 if (nargs->nfs_args_ext == NFS_ARGS_EXTB) 670 nargs->nfs_ext_u.nfs_extB.next = 671 STRUCT_FGETP(args, nfs_ext_u.nfs_extB.next); 672 673 errout: 674 if (error) 675 nfs4_free_args(nargs); 676 677 return (error); 678 } 679 680 681 /* 682 * nfs mount vfsop 683 * Set up mount info record and attach it to vfs struct. 684 */ 685 int 686 nfs4_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr) 687 { 688 char *data = uap->dataptr; 689 int error; 690 vnode_t *rtvp; /* the server's root */ 691 mntinfo4_t *mi; /* mount info, pointed at by vfs */ 692 struct knetconfig *rdma_knconf; /* rdma transport structure */ 693 rnode4_t *rp; 694 struct servinfo4 *svp; /* nfs server info */ 695 struct servinfo4 *svp_tail = NULL; /* previous nfs server info */ 696 struct servinfo4 *svp_head; /* first nfs server info */ 697 struct servinfo4 *svp_2ndlast; /* 2nd last in server info list */ 698 struct sec_data *secdata; /* security data */ 699 struct nfs_args *args = NULL; 700 int flags, addr_type, removed; 701 zone_t *zone = nfs_zone(); 702 nfs4_error_t n4e; 703 zone_t *mntzone = NULL; 704 705 if (secpolicy_fs_mount(cr, mvp, vfsp) != 0) 706 return (EPERM); 707 if (mvp->v_type != VDIR) 708 return (ENOTDIR); 709 710 /* 711 * get arguments 712 * 713 * nfs_args is now versioned and is extensible, so 714 * uap->datalen might be different from sizeof (args) 715 * in a compatible situation. 716 */ 717 more: 718 if (!(uap->flags & MS_SYSSPACE)) { 719 if (args == NULL) 720 args = kmem_zalloc(sizeof (struct nfs_args), KM_SLEEP); 721 else 722 nfs4_free_args(args); 723 error = nfs4_copyin(data, uap->datalen, args); 724 if (error) { 725 if (args) { 726 kmem_free(args, sizeof (*args)); 727 } 728 return (error); 729 } 730 } else { 731 args = (struct nfs_args *)data; 732 } 733 734 flags = args->flags; 735 736 /* 737 * If the request changes the locking type, disallow the remount, 738 * because it's questionable whether we can transfer the 739 * locking state correctly. 740 */ 741 if (uap->flags & MS_REMOUNT) { 742 if (!(uap->flags & MS_SYSSPACE)) { 743 nfs4_free_args(args); 744 kmem_free(args, sizeof (*args)); 745 } 746 if ((mi = VFTOMI4(vfsp)) != NULL) { 747 uint_t new_mi_llock; 748 uint_t old_mi_llock; 749 new_mi_llock = (flags & NFSMNT_LLOCK) ? 1 : 0; 750 old_mi_llock = (mi->mi_flags & MI4_LLOCK) ? 1 : 0; 751 if (old_mi_llock != new_mi_llock) 752 return (EBUSY); 753 } 754 return (0); 755 } 756 757 /* 758 * For ephemeral mount trigger stub vnodes, we have two problems 759 * to solve: racing threads will likely fail the v_count check, and 760 * we want only one to proceed with the mount. 761 * 762 * For stubs, if the mount has already occurred (via a racing thread), 763 * just return success. If not, skip the v_count check and proceed. 764 * Note that we are already serialised at this point. 765 */ 766 mutex_enter(&mvp->v_lock); 767 if (vn_matchops(mvp, nfs4_trigger_vnodeops)) { 768 /* mntpt is a v4 stub vnode */ 769 ASSERT(RP_ISSTUB(VTOR4(mvp))); 770 ASSERT(!(uap->flags & MS_OVERLAY)); 771 ASSERT(!(mvp->v_flag & VROOT)); 772 if (vn_mountedvfs(mvp) != NULL) { 773 /* ephemeral mount has already occurred */ 774 ASSERT(uap->flags & MS_SYSSPACE); 775 mutex_exit(&mvp->v_lock); 776 return (0); 777 } 778 } else { 779 /* mntpt is a non-v4 or v4 non-stub vnode */ 780 if (!(uap->flags & MS_OVERLAY) && 781 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { 782 mutex_exit(&mvp->v_lock); 783 if (!(uap->flags & MS_SYSSPACE)) { 784 nfs4_free_args(args); 785 kmem_free(args, sizeof (*args)); 786 } 787 return (EBUSY); 788 } 789 } 790 mutex_exit(&mvp->v_lock); 791 792 /* make sure things are zeroed for errout: */ 793 rtvp = NULL; 794 mi = NULL; 795 secdata = NULL; 796 797 /* 798 * A valid knetconfig structure is required. 799 */ 800 if (!(flags & NFSMNT_KNCONF) || 801 args->knconf == NULL || args->knconf->knc_protofmly == NULL || 802 args->knconf->knc_proto == NULL || 803 (strcmp(args->knconf->knc_proto, NC_UDP) == 0)) { 804 if (!(uap->flags & MS_SYSSPACE)) { 805 nfs4_free_args(args); 806 kmem_free(args, sizeof (*args)); 807 } 808 return (EINVAL); 809 } 810 811 if ((strlen(args->knconf->knc_protofmly) >= KNC_STRSIZE) || 812 (strlen(args->knconf->knc_proto) >= KNC_STRSIZE)) { 813 if (!(uap->flags & MS_SYSSPACE)) { 814 nfs4_free_args(args); 815 kmem_free(args, sizeof (*args)); 816 } 817 return (EINVAL); 818 } 819 820 /* 821 * Allocate a servinfo4 struct. 822 */ 823 svp = kmem_zalloc(sizeof (*svp), KM_SLEEP); 824 nfs_rw_init(&svp->sv_lock, NULL, RW_DEFAULT, NULL); 825 if (svp_tail) { 826 svp_2ndlast = svp_tail; 827 svp_tail->sv_next = svp; 828 } else { 829 svp_head = svp; 830 svp_2ndlast = svp; 831 } 832 833 svp_tail = svp; 834 svp->sv_knconf = args->knconf; 835 args->knconf = NULL; 836 837 /* 838 * Get server address 839 */ 840 if (args->addr == NULL || args->addr->buf == NULL) { 841 error = EINVAL; 842 goto errout; 843 } 844 845 svp->sv_addr.maxlen = args->addr->maxlen; 846 svp->sv_addr.len = args->addr->len; 847 svp->sv_addr.buf = args->addr->buf; 848 args->addr->buf = NULL; 849 850 /* 851 * Get the root fhandle 852 */ 853 if (args->fh == NULL || (strlen(args->fh) >= MAXPATHLEN)) { 854 error = EINVAL; 855 goto errout; 856 } 857 858 svp->sv_path = args->fh; 859 svp->sv_pathlen = strlen(args->fh) + 1; 860 args->fh = NULL; 861 862 /* 863 * Get server's hostname 864 */ 865 if (flags & NFSMNT_HOSTNAME) { 866 if (args->hostname == NULL || (strlen(args->hostname) > 867 MAXNETNAMELEN)) { 868 error = EINVAL; 869 goto errout; 870 } 871 svp->sv_hostnamelen = strlen(args->hostname) + 1; 872 svp->sv_hostname = args->hostname; 873 args->hostname = NULL; 874 } else { 875 char *p = "unknown-host"; 876 svp->sv_hostnamelen = strlen(p) + 1; 877 svp->sv_hostname = kmem_zalloc(svp->sv_hostnamelen, KM_SLEEP); 878 (void) strcpy(svp->sv_hostname, p); 879 } 880 881 /* 882 * RDMA MOUNT SUPPORT FOR NFS v4. 883 * Establish, is it possible to use RDMA, if so overload the 884 * knconf with rdma specific knconf and free the orignal knconf. 885 */ 886 if ((flags & NFSMNT_TRYRDMA) || (flags & NFSMNT_DORDMA)) { 887 /* 888 * Determine the addr type for RDMA, IPv4 or v6. 889 */ 890 if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET) == 0) 891 addr_type = AF_INET; 892 else if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET6) == 0) 893 addr_type = AF_INET6; 894 895 if (rdma_reachable(addr_type, &svp->sv_addr, 896 &rdma_knconf) == 0) { 897 /* 898 * If successful, hijack the orignal knconf and 899 * replace with the new one, depending on the flags. 900 */ 901 svp->sv_origknconf = svp->sv_knconf; 902 svp->sv_knconf = rdma_knconf; 903 } else { 904 if (flags & NFSMNT_TRYRDMA) { 905 #ifdef DEBUG 906 if (rdma_debug) 907 zcmn_err(getzoneid(), CE_WARN, 908 "no RDMA onboard, revert\n"); 909 #endif 910 } 911 912 if (flags & NFSMNT_DORDMA) { 913 /* 914 * If proto=rdma is specified and no RDMA 915 * path to this server is avialable then 916 * ditch this server. 917 * This is not included in the mountable 918 * server list or the replica list. 919 * Check if more servers are specified; 920 * Failover case, otherwise bail out of mount. 921 */ 922 if (args->nfs_args_ext == NFS_ARGS_EXTB && 923 args->nfs_ext_u.nfs_extB.next != NULL) { 924 data = (char *) 925 args->nfs_ext_u.nfs_extB.next; 926 if (uap->flags & MS_RDONLY && 927 !(flags & NFSMNT_SOFT)) { 928 if (svp_head->sv_next == NULL) { 929 svp_tail = NULL; 930 svp_2ndlast = NULL; 931 sv4_free(svp_head); 932 goto more; 933 } else { 934 svp_tail = svp_2ndlast; 935 svp_2ndlast->sv_next = 936 NULL; 937 sv4_free(svp); 938 goto more; 939 } 940 } 941 } else { 942 /* 943 * This is the last server specified 944 * in the nfs_args list passed down 945 * and its not rdma capable. 946 */ 947 if (svp_head->sv_next == NULL) { 948 /* 949 * Is this the only one 950 */ 951 error = EINVAL; 952 #ifdef DEBUG 953 if (rdma_debug) 954 zcmn_err(getzoneid(), 955 CE_WARN, 956 "No RDMA srv"); 957 #endif 958 goto errout; 959 } else { 960 /* 961 * There is list, since some 962 * servers specified before 963 * this passed all requirements 964 */ 965 svp_tail = svp_2ndlast; 966 svp_2ndlast->sv_next = NULL; 967 sv4_free(svp); 968 goto proceed; 969 } 970 } 971 } 972 } 973 } 974 975 /* 976 * If there are syncaddr and netname data, load them in. This is 977 * to support data needed for NFSV4 when AUTH_DH is the negotiated 978 * flavor via SECINFO. (instead of using MOUNT protocol in V3). 979 */ 980 if (args->flags & NFSMNT_SECURE) { 981 svp->sv_dhsec = create_authdh_data(args->netname, 982 strlen(args->netname), 983 args->syncaddr, svp->sv_knconf); 984 } 985 986 /* 987 * Get the extention data which has the security data structure. 988 * This includes data for AUTH_SYS as well. 989 */ 990 if (flags & NFSMNT_NEWARGS) { 991 switch (args->nfs_args_ext) { 992 case NFS_ARGS_EXTA: 993 case NFS_ARGS_EXTB: 994 /* 995 * Indicating the application is using the new 996 * sec_data structure to pass in the security 997 * data. 998 */ 999 secdata = args->nfs_ext_u.nfs_extA.secdata; 1000 if (secdata == NULL) { 1001 error = EINVAL; 1002 } else if (uap->flags & MS_SYSSPACE) { 1003 /* 1004 * Need to validate the flavor here if 1005 * sysspace, userspace was already 1006 * validate from the nfs_copyin function. 1007 */ 1008 switch (secdata->rpcflavor) { 1009 case AUTH_NONE: 1010 case AUTH_UNIX: 1011 case AUTH_LOOPBACK: 1012 case AUTH_DES: 1013 case RPCSEC_GSS: 1014 break; 1015 default: 1016 error = EINVAL; 1017 goto errout; 1018 } 1019 } 1020 args->nfs_ext_u.nfs_extA.secdata = NULL; 1021 break; 1022 1023 default: 1024 error = EINVAL; 1025 break; 1026 } 1027 1028 } else if (flags & NFSMNT_SECURE) { 1029 /* 1030 * NFSMNT_SECURE is deprecated but we keep it 1031 * to support the rogue user-generated application 1032 * that may use this undocumented interface to do 1033 * AUTH_DH security, e.g. our own rexd. 1034 * 1035 * Also note that NFSMNT_SECURE is used for passing 1036 * AUTH_DH info to be used in negotiation. 1037 */ 1038 secdata = create_authdh_data(args->netname, 1039 strlen(args->netname), args->syncaddr, svp->sv_knconf); 1040 1041 } else { 1042 secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP); 1043 secdata->secmod = secdata->rpcflavor = AUTH_SYS; 1044 secdata->data = NULL; 1045 } 1046 1047 svp->sv_secdata = secdata; 1048 1049 /* 1050 * User does not explictly specify a flavor, and a user 1051 * defined default flavor is passed down. 1052 */ 1053 if (flags & NFSMNT_SECDEFAULT) { 1054 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0); 1055 svp->sv_flags |= SV4_TRYSECDEFAULT; 1056 nfs_rw_exit(&svp->sv_lock); 1057 } 1058 1059 /* 1060 * Failover support: 1061 * 1062 * We may have a linked list of nfs_args structures, 1063 * which means the user is looking for failover. If 1064 * the mount is either not "read-only" or "soft", 1065 * we want to bail out with EINVAL. 1066 */ 1067 if (args->nfs_args_ext == NFS_ARGS_EXTB && 1068 args->nfs_ext_u.nfs_extB.next != NULL) { 1069 if (uap->flags & MS_RDONLY && !(flags & NFSMNT_SOFT)) { 1070 data = (char *)args->nfs_ext_u.nfs_extB.next; 1071 goto more; 1072 } 1073 error = EINVAL; 1074 goto errout; 1075 } 1076 1077 /* 1078 * Determine the zone we're being mounted into. 1079 */ 1080 zone_hold(mntzone = zone); /* start with this assumption */ 1081 if (getzoneid() == GLOBAL_ZONEID) { 1082 zone_rele(mntzone); 1083 mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt)); 1084 ASSERT(mntzone != NULL); 1085 if (mntzone != zone) { 1086 error = EBUSY; 1087 goto errout; 1088 } 1089 } 1090 1091 if (is_system_labeled()) { 1092 error = nfs_mount_label_policy(vfsp, &svp->sv_addr, 1093 svp->sv_knconf, cr); 1094 1095 if (error > 0) 1096 goto errout; 1097 1098 if (error == -1) { 1099 /* change mount to read-only to prevent write-down */ 1100 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 1101 } 1102 } 1103 1104 /* 1105 * Stop the mount from going any further if the zone is going away. 1106 */ 1107 if (zone_status_get(mntzone) >= ZONE_IS_SHUTTING_DOWN) { 1108 error = EBUSY; 1109 goto errout; 1110 } 1111 1112 /* 1113 * Get root vnode. 1114 */ 1115 proceed: 1116 error = nfs4rootvp(&rtvp, vfsp, svp_head, flags, cr, mntzone); 1117 if (error) { 1118 /* if nfs4rootvp failed, it will free svp_head */ 1119 svp_head = NULL; 1120 goto errout; 1121 } 1122 1123 mi = VTOMI4(rtvp); 1124 1125 /* 1126 * Send client id to the server, if necessary 1127 */ 1128 nfs4_error_zinit(&n4e); 1129 nfs4setclientid(mi, cr, FALSE, &n4e); 1130 1131 error = n4e.error; 1132 1133 if (error) 1134 goto errout; 1135 1136 /* 1137 * Set option fields in the mount info record 1138 */ 1139 1140 if (svp_head->sv_next) { 1141 mutex_enter(&mi->mi_lock); 1142 mi->mi_flags |= MI4_LLOCK; 1143 mutex_exit(&mi->mi_lock); 1144 } 1145 error = nfs4_setopts(rtvp, DATAMODEL_NATIVE, args); 1146 if (error) 1147 goto errout; 1148 1149 /* 1150 * Time to tie in the mirror mount info at last! 1151 */ 1152 if (flags & NFSMNT_EPHEMERAL) 1153 error = nfs4_record_ephemeral_mount(mi, mvp); 1154 1155 errout: 1156 if (error) { 1157 if (rtvp != NULL) { 1158 rp = VTOR4(rtvp); 1159 if (rp->r_flags & R4HASHED) 1160 rp4_rmhash(rp); 1161 } 1162 if (mi != NULL) { 1163 nfs4_async_stop(vfsp); 1164 nfs4_async_manager_stop(vfsp); 1165 nfs4_remove_mi_from_server(mi, NULL); 1166 if (rtvp != NULL) 1167 VN_RELE(rtvp); 1168 if (mntzone != NULL) 1169 zone_rele(mntzone); 1170 /* need to remove it from the zone */ 1171 removed = nfs4_mi_zonelist_remove(mi); 1172 if (removed) 1173 zone_rele(mi->mi_zone); 1174 MI4_RELE(mi); 1175 if (!(uap->flags & MS_SYSSPACE) && args) { 1176 nfs4_free_args(args); 1177 kmem_free(args, sizeof (*args)); 1178 } 1179 return (error); 1180 } 1181 if (svp_head) 1182 sv4_free(svp_head); 1183 } 1184 1185 if (!(uap->flags & MS_SYSSPACE) && args) { 1186 nfs4_free_args(args); 1187 kmem_free(args, sizeof (*args)); 1188 } 1189 if (rtvp != NULL) 1190 VN_RELE(rtvp); 1191 1192 if (mntzone != NULL) 1193 zone_rele(mntzone); 1194 1195 return (error); 1196 } 1197 1198 #ifdef DEBUG 1199 #define VERS_MSG "NFS4 server " 1200 #else 1201 #define VERS_MSG "NFS server " 1202 #endif 1203 1204 #define READ_MSG \ 1205 VERS_MSG "%s returned 0 for read transfer size" 1206 #define WRITE_MSG \ 1207 VERS_MSG "%s returned 0 for write transfer size" 1208 #define SIZE_MSG \ 1209 VERS_MSG "%s returned 0 for maximum file size" 1210 1211 /* 1212 * Get the symbolic link text from the server for a given filehandle 1213 * of that symlink. 1214 * 1215 * (get symlink text) PUTFH READLINK 1216 */ 1217 static int 1218 getlinktext_otw(mntinfo4_t *mi, nfs_fh4 *fh, char **linktextp, cred_t *cr, 1219 int flags) 1220 { 1221 COMPOUND4args_clnt args; 1222 COMPOUND4res_clnt res; 1223 int doqueue; 1224 nfs_argop4 argop[2]; 1225 nfs_resop4 *resop; 1226 READLINK4res *lr_res; 1227 uint_t len; 1228 bool_t needrecov = FALSE; 1229 nfs4_recov_state_t recov_state; 1230 nfs4_sharedfh_t *sfh; 1231 nfs4_error_t e; 1232 int num_retry = nfs4_max_mount_retry; 1233 int recovery = !(flags & NFS4_GETFH_NEEDSOP); 1234 1235 sfh = sfh4_get(fh, mi); 1236 recov_state.rs_flags = 0; 1237 recov_state.rs_num_retry_despite_err = 0; 1238 1239 recov_retry: 1240 nfs4_error_zinit(&e); 1241 1242 args.array_len = 2; 1243 args.array = argop; 1244 args.ctag = TAG_GET_SYMLINK; 1245 1246 if (! recovery) { 1247 e.error = nfs4_start_op(mi, NULL, NULL, &recov_state); 1248 if (e.error) { 1249 sfh4_rele(&sfh); 1250 return (e.error); 1251 } 1252 } 1253 1254 /* 0. putfh symlink fh */ 1255 argop[0].argop = OP_CPUTFH; 1256 argop[0].nfs_argop4_u.opcputfh.sfh = sfh; 1257 1258 /* 1. readlink */ 1259 argop[1].argop = OP_READLINK; 1260 1261 doqueue = 1; 1262 1263 rfs4call(mi, &args, &res, cr, &doqueue, 0, &e); 1264 1265 needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp); 1266 1267 if (needrecov && !recovery && num_retry-- > 0) { 1268 1269 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE, 1270 "getlinktext_otw: initiating recovery\n")); 1271 1272 if (nfs4_start_recovery(&e, mi, NULL, NULL, NULL, NULL, 1273 OP_READLINK, NULL) == FALSE) { 1274 nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov); 1275 if (!e.error) 1276 (void) xdr_free(xdr_COMPOUND4res_clnt, 1277 (caddr_t)&res); 1278 goto recov_retry; 1279 } 1280 } 1281 1282 /* 1283 * If non-NFS4 pcol error and/or we weren't able to recover. 1284 */ 1285 if (e.error != 0) { 1286 if (! recovery) 1287 nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov); 1288 sfh4_rele(&sfh); 1289 return (e.error); 1290 } 1291 1292 if (res.status) { 1293 e.error = geterrno4(res.status); 1294 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 1295 if (! recovery) 1296 nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov); 1297 sfh4_rele(&sfh); 1298 return (e.error); 1299 } 1300 1301 /* res.status == NFS4_OK */ 1302 ASSERT(res.status == NFS4_OK); 1303 1304 resop = &res.array[1]; /* readlink res */ 1305 lr_res = &resop->nfs_resop4_u.opreadlink; 1306 1307 /* treat symlink name as data */ 1308 *linktextp = utf8_to_str(&lr_res->link, &len, NULL); 1309 1310 if (! recovery) 1311 nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov); 1312 sfh4_rele(&sfh); 1313 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 1314 return (0); 1315 } 1316 1317 /* 1318 * Skip over consecutive slashes and "/./" in a pathname. 1319 */ 1320 void 1321 pathname_skipslashdot(struct pathname *pnp) 1322 { 1323 char *c1, *c2; 1324 1325 while (pnp->pn_pathlen > 0 && *pnp->pn_path == '/') { 1326 1327 c1 = pnp->pn_path + 1; 1328 c2 = pnp->pn_path + 2; 1329 1330 if (*c1 == '.' && (*c2 == '/' || *c2 == '\0')) { 1331 pnp->pn_path = pnp->pn_path + 2; /* skip "/." */ 1332 pnp->pn_pathlen = pnp->pn_pathlen - 2; 1333 } else { 1334 pnp->pn_path++; 1335 pnp->pn_pathlen--; 1336 } 1337 } 1338 } 1339 1340 /* 1341 * Resolve a symbolic link path. The symlink is in the nth component of 1342 * svp->sv_path and has an nfs4 file handle "fh". 1343 * Upon return, the sv_path will point to the new path that has the nth 1344 * component resolved to its symlink text. 1345 */ 1346 int 1347 resolve_sympath(mntinfo4_t *mi, servinfo4_t *svp, int nth, nfs_fh4 *fh, 1348 cred_t *cr, int flags) 1349 { 1350 char *oldpath; 1351 char *symlink, *newpath; 1352 struct pathname oldpn, newpn; 1353 char component[MAXNAMELEN]; 1354 int i, addlen, error = 0; 1355 int oldpathlen; 1356 1357 /* Get the symbolic link text over the wire. */ 1358 error = getlinktext_otw(mi, fh, &symlink, cr, flags); 1359 1360 if (error || symlink == NULL || strlen(symlink) == 0) 1361 return (error); 1362 1363 /* 1364 * Compose the new pathname. 1365 * Note: 1366 * - only the nth component is resolved for the pathname. 1367 * - pathname.pn_pathlen does not count the ending null byte. 1368 */ 1369 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 1370 oldpath = svp->sv_path; 1371 oldpathlen = svp->sv_pathlen; 1372 if (error = pn_get(oldpath, UIO_SYSSPACE, &oldpn)) { 1373 nfs_rw_exit(&svp->sv_lock); 1374 kmem_free(symlink, strlen(symlink) + 1); 1375 return (error); 1376 } 1377 nfs_rw_exit(&svp->sv_lock); 1378 pn_alloc(&newpn); 1379 1380 /* 1381 * Skip over previous components from the oldpath so that the 1382 * oldpn.pn_path will point to the symlink component. Skip 1383 * leading slashes and "/./" (no OP_LOOKUP on ".") so that 1384 * pn_getcompnent can get the component. 1385 */ 1386 for (i = 1; i < nth; i++) { 1387 pathname_skipslashdot(&oldpn); 1388 error = pn_getcomponent(&oldpn, component); 1389 if (error) 1390 goto out; 1391 } 1392 1393 /* 1394 * Copy the old path upto the component right before the symlink 1395 * if the symlink is not an absolute path. 1396 */ 1397 if (symlink[0] != '/') { 1398 addlen = oldpn.pn_path - oldpn.pn_buf; 1399 bcopy(oldpn.pn_buf, newpn.pn_path, addlen); 1400 newpn.pn_pathlen += addlen; 1401 newpn.pn_path += addlen; 1402 newpn.pn_buf[newpn.pn_pathlen] = '/'; 1403 newpn.pn_pathlen++; 1404 newpn.pn_path++; 1405 } 1406 1407 /* copy the resolved symbolic link text */ 1408 addlen = strlen(symlink); 1409 if (newpn.pn_pathlen + addlen >= newpn.pn_bufsize) { 1410 error = ENAMETOOLONG; 1411 goto out; 1412 } 1413 bcopy(symlink, newpn.pn_path, addlen); 1414 newpn.pn_pathlen += addlen; 1415 newpn.pn_path += addlen; 1416 1417 /* 1418 * Check if there is any remaining path after the symlink component. 1419 * First, skip the symlink component. 1420 */ 1421 pathname_skipslashdot(&oldpn); 1422 if (error = pn_getcomponent(&oldpn, component)) 1423 goto out; 1424 1425 addlen = pn_pathleft(&oldpn); /* includes counting the slash */ 1426 1427 /* 1428 * Copy the remaining path to the new pathname if there is any. 1429 */ 1430 if (addlen > 0) { 1431 if (newpn.pn_pathlen + addlen >= newpn.pn_bufsize) { 1432 error = ENAMETOOLONG; 1433 goto out; 1434 } 1435 bcopy(oldpn.pn_path, newpn.pn_path, addlen); 1436 newpn.pn_pathlen += addlen; 1437 } 1438 newpn.pn_buf[newpn.pn_pathlen] = '\0'; 1439 1440 /* get the newpath and store it in the servinfo4_t */ 1441 newpath = kmem_alloc(newpn.pn_pathlen + 1, KM_SLEEP); 1442 bcopy(newpn.pn_buf, newpath, newpn.pn_pathlen); 1443 newpath[newpn.pn_pathlen] = '\0'; 1444 1445 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0); 1446 svp->sv_path = newpath; 1447 svp->sv_pathlen = strlen(newpath) + 1; 1448 nfs_rw_exit(&svp->sv_lock); 1449 1450 kmem_free(oldpath, oldpathlen); 1451 out: 1452 kmem_free(symlink, strlen(symlink) + 1); 1453 pn_free(&newpn); 1454 pn_free(&oldpn); 1455 1456 return (error); 1457 } 1458 1459 /* 1460 * Get the root filehandle for the given filesystem and server, and update 1461 * svp. 1462 * 1463 * If NFS4_GETFH_NEEDSOP is set, then use nfs4_start_fop and nfs4_end_fop 1464 * to coordinate with recovery. Otherwise, the caller is assumed to be 1465 * the recovery thread or have already done a start_fop. 1466 * 1467 * Errors are returned by the nfs4_error_t parameter. 1468 */ 1469 1470 static void 1471 nfs4getfh_otw(struct mntinfo4 *mi, servinfo4_t *svp, vtype_t *vtp, 1472 int flags, cred_t *cr, nfs4_error_t *ep) 1473 { 1474 COMPOUND4args_clnt args; 1475 COMPOUND4res_clnt res; 1476 int doqueue = 1; 1477 nfs_argop4 *argop; 1478 nfs_resop4 *resop; 1479 nfs4_ga_res_t *garp; 1480 int num_argops; 1481 lookup4_param_t lookuparg; 1482 nfs_fh4 *tmpfhp; 1483 nfs_fh4 *resfhp; 1484 bool_t needrecov = FALSE; 1485 nfs4_recov_state_t recov_state; 1486 int llndx; 1487 int nthcomp; 1488 int recovery = !(flags & NFS4_GETFH_NEEDSOP); 1489 1490 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 1491 ASSERT(svp->sv_path != NULL); 1492 if (svp->sv_path[0] == '\0') { 1493 nfs_rw_exit(&svp->sv_lock); 1494 nfs4_error_init(ep, EINVAL); 1495 return; 1496 } 1497 nfs_rw_exit(&svp->sv_lock); 1498 1499 recov_state.rs_flags = 0; 1500 recov_state.rs_num_retry_despite_err = 0; 1501 recov_retry: 1502 nfs4_error_zinit(ep); 1503 1504 if (!recovery) { 1505 ep->error = nfs4_start_fop(mi, NULL, NULL, OH_MOUNT, 1506 &recov_state, NULL); 1507 1508 /* 1509 * If recovery has been started and this request as 1510 * initiated by a mount, then we must wait for recovery 1511 * to finish before proceeding, otherwise, the error 1512 * cleanup would remove data structures needed by the 1513 * recovery thread. 1514 */ 1515 if (ep->error) { 1516 mutex_enter(&mi->mi_lock); 1517 if (mi->mi_flags & MI4_MOUNTING) { 1518 mi->mi_flags |= MI4_RECOV_FAIL; 1519 mi->mi_error = EIO; 1520 1521 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE, 1522 "nfs4getfh_otw: waiting 4 recovery\n")); 1523 1524 while (mi->mi_flags & MI4_RECOV_ACTIV) 1525 cv_wait(&mi->mi_failover_cv, 1526 &mi->mi_lock); 1527 } 1528 mutex_exit(&mi->mi_lock); 1529 return; 1530 } 1531 1532 /* 1533 * If the client does not specify a specific flavor to use 1534 * and has not gotten a secinfo list from the server yet, 1535 * retrieve the secinfo list from the server and use a 1536 * flavor from the list to mount. 1537 * 1538 * If fail to get the secinfo list from the server, then 1539 * try the default flavor. 1540 */ 1541 if ((svp->sv_flags & SV4_TRYSECDEFAULT) && 1542 svp->sv_secinfo == NULL) { 1543 (void) nfs4_secinfo_path(mi, cr, FALSE); 1544 } 1545 } 1546 1547 if (recovery) 1548 args.ctag = TAG_REMAP_MOUNT; 1549 else 1550 args.ctag = TAG_MOUNT; 1551 1552 lookuparg.l4_getattrs = LKP4_ALL_ATTRIBUTES; 1553 lookuparg.argsp = &args; 1554 lookuparg.resp = &res; 1555 lookuparg.header_len = 2; /* Putrootfh, getfh */ 1556 lookuparg.trailer_len = 0; 1557 lookuparg.ga_bits = FATTR4_FSINFO_MASK; 1558 lookuparg.mi = mi; 1559 1560 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 1561 ASSERT(svp->sv_path != NULL); 1562 llndx = nfs4lookup_setup(svp->sv_path, &lookuparg, 0); 1563 nfs_rw_exit(&svp->sv_lock); 1564 1565 argop = args.array; 1566 num_argops = args.array_len; 1567 1568 /* choose public or root filehandle */ 1569 if (flags & NFS4_GETFH_PUBLIC) 1570 argop[0].argop = OP_PUTPUBFH; 1571 else 1572 argop[0].argop = OP_PUTROOTFH; 1573 1574 /* get fh */ 1575 argop[1].argop = OP_GETFH; 1576 1577 NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE, 1578 "nfs4getfh_otw: %s call, mi 0x%p", 1579 needrecov ? "recov" : "first", (void *)mi)); 1580 1581 rfs4call(mi, &args, &res, cr, &doqueue, RFSCALL_SOFT, ep); 1582 1583 needrecov = nfs4_needs_recovery(ep, FALSE, mi->mi_vfsp); 1584 1585 if (needrecov) { 1586 bool_t abort; 1587 1588 if (recovery) { 1589 nfs4args_lookup_free(argop, num_argops); 1590 kmem_free(argop, 1591 lookuparg.arglen * sizeof (nfs_argop4)); 1592 if (!ep->error) 1593 (void) xdr_free(xdr_COMPOUND4res_clnt, 1594 (caddr_t)&res); 1595 return; 1596 } 1597 1598 NFS4_DEBUG(nfs4_client_recov_debug, 1599 (CE_NOTE, "nfs4getfh_otw: initiating recovery\n")); 1600 1601 abort = nfs4_start_recovery(ep, mi, NULL, 1602 NULL, NULL, NULL, OP_GETFH, NULL); 1603 if (!ep->error) { 1604 ep->error = geterrno4(res.status); 1605 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 1606 } 1607 nfs4args_lookup_free(argop, num_argops); 1608 kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4)); 1609 nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state, needrecov); 1610 /* have another go? */ 1611 if (abort == FALSE) 1612 goto recov_retry; 1613 return; 1614 } 1615 1616 /* 1617 * No recovery, but check if error is set. 1618 */ 1619 if (ep->error) { 1620 nfs4args_lookup_free(argop, num_argops); 1621 kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4)); 1622 if (!recovery) 1623 nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state, 1624 needrecov); 1625 return; 1626 } 1627 1628 is_link_err: 1629 1630 /* for non-recovery errors */ 1631 if (res.status && res.status != NFS4ERR_SYMLINK) { 1632 if (!recovery) { 1633 nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state, 1634 needrecov); 1635 } 1636 nfs4args_lookup_free(argop, num_argops); 1637 kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4)); 1638 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 1639 return; 1640 } 1641 1642 /* 1643 * If any intermediate component in the path is a symbolic link, 1644 * resolve the symlink, then try mount again using the new path. 1645 */ 1646 if (res.status == NFS4ERR_SYMLINK) { 1647 int where; 1648 1649 /* 1650 * This must be from OP_LOOKUP failure. The (cfh) for this 1651 * OP_LOOKUP is a symlink node. Found out where the 1652 * OP_GETFH is for the (cfh) that is a symlink node. 1653 * 1654 * Example: 1655 * (mount) PUTROOTFH, GETFH, LOOKUP comp1, GETFH, GETATTR, 1656 * LOOKUP comp2, GETFH, GETATTR, LOOKUP comp3, GETFH, GETATTR 1657 * 1658 * LOOKUP comp3 fails with SYMLINK because comp2 is a symlink. 1659 * In this case, where = 7, nthcomp = 2. 1660 */ 1661 where = res.array_len - 2; 1662 ASSERT(where > 0); 1663 1664 resop = &res.array[where - 1]; 1665 ASSERT(resop->resop == OP_GETFH); 1666 tmpfhp = &resop->nfs_resop4_u.opgetfh.object; 1667 nthcomp = res.array_len/3 - 1; 1668 1669 /* 1670 * Need to call nfs4_end_op before resolve_sympath to avoid 1671 * potential nfs4_start_op deadlock. 1672 */ 1673 if (!recovery) 1674 nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state, 1675 needrecov); 1676 1677 ep->error = resolve_sympath(mi, svp, nthcomp, tmpfhp, cr, 1678 flags); 1679 1680 nfs4args_lookup_free(argop, num_argops); 1681 kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4)); 1682 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 1683 1684 if (ep->error) 1685 return; 1686 1687 goto recov_retry; 1688 } 1689 1690 /* getfh */ 1691 resop = &res.array[res.array_len - 2]; 1692 ASSERT(resop->resop == OP_GETFH); 1693 resfhp = &resop->nfs_resop4_u.opgetfh.object; 1694 1695 /* getattr fsinfo res */ 1696 resop++; 1697 garp = &resop->nfs_resop4_u.opgetattr.ga_res; 1698 1699 *vtp = garp->n4g_va.va_type; 1700 1701 mi->mi_fh_expire_type = garp->n4g_ext_res->n4g_fet; 1702 1703 mutex_enter(&mi->mi_lock); 1704 if (garp->n4g_ext_res->n4g_pc4.pc4_link_support) 1705 mi->mi_flags |= MI4_LINK; 1706 if (garp->n4g_ext_res->n4g_pc4.pc4_symlink_support) 1707 mi->mi_flags |= MI4_SYMLINK; 1708 if (garp->n4g_ext_res->n4g_suppattrs & FATTR4_ACL_MASK) 1709 mi->mi_flags |= MI4_ACL; 1710 mutex_exit(&mi->mi_lock); 1711 1712 if (garp->n4g_ext_res->n4g_maxread == 0) 1713 mi->mi_tsize = 1714 MIN(MAXBSIZE, mi->mi_tsize); 1715 else 1716 mi->mi_tsize = 1717 MIN(garp->n4g_ext_res->n4g_maxread, 1718 mi->mi_tsize); 1719 1720 if (garp->n4g_ext_res->n4g_maxwrite == 0) 1721 mi->mi_stsize = 1722 MIN(MAXBSIZE, mi->mi_stsize); 1723 else 1724 mi->mi_stsize = 1725 MIN(garp->n4g_ext_res->n4g_maxwrite, 1726 mi->mi_stsize); 1727 1728 if (garp->n4g_ext_res->n4g_maxfilesize != 0) 1729 mi->mi_maxfilesize = 1730 MIN(garp->n4g_ext_res->n4g_maxfilesize, 1731 mi->mi_maxfilesize); 1732 1733 /* 1734 * If the final component is a a symbolic link, resolve the symlink, 1735 * then try mount again using the new path. 1736 * 1737 * Assume no symbolic link for root filesysm "/". 1738 */ 1739 if (*vtp == VLNK) { 1740 /* 1741 * nthcomp is the total result length minus 1742 * the 1st 2 OPs (PUTROOTFH, GETFH), 1743 * then divided by 3 (LOOKUP,GETFH,GETATTR) 1744 * 1745 * e.g. PUTROOTFH GETFH LOOKUP 1st-comp GETFH GETATTR 1746 * LOOKUP 2nd-comp GETFH GETATTR 1747 * 1748 * (8 - 2)/3 = 2 1749 */ 1750 nthcomp = (res.array_len - 2)/3; 1751 1752 /* 1753 * Need to call nfs4_end_op before resolve_sympath to avoid 1754 * potential nfs4_start_op deadlock. See RFE 4777612. 1755 */ 1756 if (!recovery) 1757 nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state, 1758 needrecov); 1759 1760 ep->error = resolve_sympath(mi, svp, nthcomp, resfhp, cr, 1761 flags); 1762 1763 nfs4args_lookup_free(argop, num_argops); 1764 kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4)); 1765 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 1766 1767 if (ep->error) 1768 return; 1769 1770 goto recov_retry; 1771 } 1772 1773 /* 1774 * We need to figure out where in the compound the getfh 1775 * for the parent directory is. If the object to be mounted is 1776 * the root, then there is no lookup at all: 1777 * PUTROOTFH, GETFH. 1778 * If the object to be mounted is in the root, then the compound is: 1779 * PUTROOTFH, GETFH, LOOKUP, GETFH, GETATTR. 1780 * In either of these cases, the index of the GETFH is 1. 1781 * If it is not at the root, then it's something like: 1782 * PUTROOTFH, GETFH, LOOKUP, GETFH, GETATTR, 1783 * LOOKUP, GETFH, GETATTR 1784 * In this case, the index is llndx (last lookup index) - 2. 1785 */ 1786 if (llndx == -1 || llndx == 2) 1787 resop = &res.array[1]; 1788 else { 1789 ASSERT(llndx > 2); 1790 resop = &res.array[llndx-2]; 1791 } 1792 1793 ASSERT(resop->resop == OP_GETFH); 1794 tmpfhp = &resop->nfs_resop4_u.opgetfh.object; 1795 1796 /* save the filehandles for the replica */ 1797 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0); 1798 ASSERT(tmpfhp->nfs_fh4_len <= NFS4_FHSIZE); 1799 svp->sv_pfhandle.fh_len = tmpfhp->nfs_fh4_len; 1800 bcopy(tmpfhp->nfs_fh4_val, svp->sv_pfhandle.fh_buf, 1801 tmpfhp->nfs_fh4_len); 1802 ASSERT(resfhp->nfs_fh4_len <= NFS4_FHSIZE); 1803 svp->sv_fhandle.fh_len = resfhp->nfs_fh4_len; 1804 bcopy(resfhp->nfs_fh4_val, svp->sv_fhandle.fh_buf, resfhp->nfs_fh4_len); 1805 1806 /* initialize fsid and supp_attrs for server fs */ 1807 svp->sv_fsid = garp->n4g_fsid; 1808 svp->sv_supp_attrs = 1809 garp->n4g_ext_res->n4g_suppattrs | FATTR4_MANDATTR_MASK; 1810 1811 nfs_rw_exit(&svp->sv_lock); 1812 1813 nfs4args_lookup_free(argop, num_argops); 1814 kmem_free(argop, lookuparg.arglen * sizeof (nfs_argop4)); 1815 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 1816 if (!recovery) 1817 nfs4_end_fop(mi, NULL, NULL, OH_MOUNT, &recov_state, needrecov); 1818 } 1819 1820 static ushort_t nfs4_max_threads = 8; /* max number of active async threads */ 1821 static uint_t nfs4_bsize = 32 * 1024; /* client `block' size */ 1822 static uint_t nfs4_async_clusters = 1; /* # of reqs from each async queue */ 1823 static uint_t nfs4_cots_timeo = NFS_COTS_TIMEO; 1824 1825 /* 1826 * Remap the root filehandle for the given filesystem. 1827 * 1828 * results returned via the nfs4_error_t parameter. 1829 */ 1830 void 1831 nfs4_remap_root(mntinfo4_t *mi, nfs4_error_t *ep, int flags) 1832 { 1833 struct servinfo4 *svp; 1834 vtype_t vtype; 1835 nfs_fh4 rootfh; 1836 int getfh_flags; 1837 char *orig_sv_path; 1838 int orig_sv_pathlen, num_retry; 1839 1840 mutex_enter(&mi->mi_lock); 1841 1842 remap_retry: 1843 svp = mi->mi_curr_serv; 1844 getfh_flags = 1845 (flags & NFS4_REMAP_NEEDSOP) ? NFS4_GETFH_NEEDSOP : 0; 1846 getfh_flags |= 1847 (mi->mi_flags & MI4_PUBLIC) ? NFS4_GETFH_PUBLIC : 0; 1848 mutex_exit(&mi->mi_lock); 1849 1850 /* 1851 * Just in case server path being mounted contains 1852 * symlinks and fails w/STALE, save the initial sv_path 1853 * so we can redrive the initial mount compound with the 1854 * initial sv_path -- not a symlink-expanded version. 1855 * 1856 * This could only happen if a symlink was expanded 1857 * and the expanded mount compound failed stale. Because 1858 * it could be the case that the symlink was removed at 1859 * the server (and replaced with another symlink/dir, 1860 * we need to use the initial sv_path when attempting 1861 * to re-lookup everything and recover. 1862 */ 1863 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 1864 orig_sv_pathlen = svp->sv_pathlen; 1865 orig_sv_path = kmem_alloc(orig_sv_pathlen, KM_SLEEP); 1866 bcopy(svp->sv_path, orig_sv_path, orig_sv_pathlen); 1867 nfs_rw_exit(&svp->sv_lock); 1868 1869 num_retry = nfs4_max_mount_retry; 1870 1871 do { 1872 /* 1873 * Get the root fh from the server. Retry nfs4_max_mount_retry 1874 * (2) times if it fails with STALE since the recovery 1875 * infrastructure doesn't do STALE recovery for components 1876 * of the server path to the object being mounted. 1877 */ 1878 nfs4getfh_otw(mi, svp, &vtype, getfh_flags, CRED(), ep); 1879 1880 if (ep->error == 0 && ep->stat == NFS4_OK) 1881 break; 1882 1883 /* 1884 * For some reason, the mount compound failed. Before 1885 * retrying, we need to restore the original sv_path 1886 * because it might have contained symlinks that were 1887 * expanded by nfsgetfh_otw before the failure occurred. 1888 * replace current sv_path with orig sv_path -- just in case 1889 * it changed due to embedded symlinks. 1890 */ 1891 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 1892 if (orig_sv_pathlen != svp->sv_pathlen) { 1893 kmem_free(svp->sv_path, svp->sv_pathlen); 1894 svp->sv_path = kmem_alloc(orig_sv_pathlen, KM_SLEEP); 1895 svp->sv_pathlen = orig_sv_pathlen; 1896 } 1897 bcopy(orig_sv_path, svp->sv_path, orig_sv_pathlen); 1898 nfs_rw_exit(&svp->sv_lock); 1899 1900 } while (num_retry-- > 0); 1901 1902 kmem_free(orig_sv_path, orig_sv_pathlen); 1903 1904 if (ep->error != 0 || ep->stat != 0) { 1905 return; 1906 } 1907 1908 if (vtype != VNON && vtype != mi->mi_type) { 1909 /* shouldn't happen */ 1910 zcmn_err(mi->mi_zone->zone_id, CE_WARN, 1911 "nfs4_remap_root: server root vnode type (%d) doesn't " 1912 "match mount info (%d)", vtype, mi->mi_type); 1913 } 1914 1915 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 1916 rootfh.nfs_fh4_val = svp->sv_fhandle.fh_buf; 1917 rootfh.nfs_fh4_len = svp->sv_fhandle.fh_len; 1918 nfs_rw_exit(&svp->sv_lock); 1919 sfh4_update(mi->mi_rootfh, &rootfh); 1920 1921 /* 1922 * It's possible that recovery took place on the filesystem 1923 * and the server has been updated between the time we did 1924 * the nfs4getfh_otw and now. Re-drive the otw operation 1925 * to make sure we have a good fh. 1926 */ 1927 mutex_enter(&mi->mi_lock); 1928 if (mi->mi_curr_serv != svp) 1929 goto remap_retry; 1930 1931 mutex_exit(&mi->mi_lock); 1932 } 1933 1934 static int 1935 nfs4rootvp(vnode_t **rtvpp, vfs_t *vfsp, struct servinfo4 *svp_head, 1936 int flags, cred_t *cr, zone_t *zone) 1937 { 1938 vnode_t *rtvp = NULL; 1939 mntinfo4_t *mi; 1940 dev_t nfs_dev; 1941 int error = 0; 1942 rnode4_t *rp; 1943 int i; 1944 struct vattr va; 1945 vtype_t vtype = VNON; 1946 vtype_t tmp_vtype = VNON; 1947 struct servinfo4 *firstsvp = NULL, *svp = svp_head; 1948 nfs4_oo_hash_bucket_t *bucketp; 1949 nfs_fh4 fh; 1950 char *droptext = ""; 1951 struct nfs_stats *nfsstatsp; 1952 nfs4_fname_t *mfname; 1953 nfs4_error_t e; 1954 char *orig_sv_path; 1955 int orig_sv_pathlen, num_retry, removed; 1956 cred_t *lcr = NULL, *tcr = cr; 1957 1958 nfsstatsp = zone_getspecific(nfsstat_zone_key, nfs_zone()); 1959 ASSERT(nfsstatsp != NULL); 1960 1961 ASSERT(nfs_zone() == zone); 1962 ASSERT(crgetref(cr)); 1963 1964 /* 1965 * Create a mount record and link it to the vfs struct. 1966 */ 1967 mi = kmem_zalloc(sizeof (*mi), KM_SLEEP); 1968 mutex_init(&mi->mi_lock, NULL, MUTEX_DEFAULT, NULL); 1969 nfs_rw_init(&mi->mi_recovlock, NULL, RW_DEFAULT, NULL); 1970 nfs_rw_init(&mi->mi_rename_lock, NULL, RW_DEFAULT, NULL); 1971 nfs_rw_init(&mi->mi_fh_lock, NULL, RW_DEFAULT, NULL); 1972 1973 if (!(flags & NFSMNT_SOFT)) 1974 mi->mi_flags |= MI4_HARD; 1975 if ((flags & NFSMNT_NOPRINT)) 1976 mi->mi_flags |= MI4_NOPRINT; 1977 if (flags & NFSMNT_INT) 1978 mi->mi_flags |= MI4_INT; 1979 if (flags & NFSMNT_PUBLIC) 1980 mi->mi_flags |= MI4_PUBLIC; 1981 if (flags & NFSMNT_MIRRORMOUNT) 1982 mi->mi_flags |= MI4_MIRRORMOUNT; 1983 mi->mi_retrans = NFS_RETRIES; 1984 if (svp->sv_knconf->knc_semantics == NC_TPI_COTS_ORD || 1985 svp->sv_knconf->knc_semantics == NC_TPI_COTS) 1986 mi->mi_timeo = nfs4_cots_timeo; 1987 else 1988 mi->mi_timeo = NFS_TIMEO; 1989 mi->mi_prog = NFS_PROGRAM; 1990 mi->mi_vers = NFS_V4; 1991 mi->mi_rfsnames = rfsnames_v4; 1992 mi->mi_reqs = nfsstatsp->nfs_stats_v4.rfsreqcnt_ptr; 1993 cv_init(&mi->mi_failover_cv, NULL, CV_DEFAULT, NULL); 1994 mi->mi_servers = svp; 1995 mi->mi_curr_serv = svp; 1996 mi->mi_acregmin = SEC2HR(ACREGMIN); 1997 mi->mi_acregmax = SEC2HR(ACREGMAX); 1998 mi->mi_acdirmin = SEC2HR(ACDIRMIN); 1999 mi->mi_acdirmax = SEC2HR(ACDIRMAX); 2000 mi->mi_fh_expire_type = FH4_PERSISTENT; 2001 mi->mi_clientid_next = NULL; 2002 mi->mi_clientid_prev = NULL; 2003 mi->mi_grace_wait = 0; 2004 mi->mi_error = 0; 2005 mi->mi_srvsettime = 0; 2006 2007 mi->mi_count = 1; 2008 2009 mi->mi_tsize = nfs4_tsize(svp->sv_knconf); 2010 mi->mi_stsize = mi->mi_tsize; 2011 2012 if (flags & NFSMNT_DIRECTIO) 2013 mi->mi_flags |= MI4_DIRECTIO; 2014 2015 mi->mi_flags |= MI4_MOUNTING; 2016 2017 /* 2018 * Make a vfs struct for nfs. We do this here instead of below 2019 * because rtvp needs a vfs before we can do a getattr on it. 2020 * 2021 * Assign a unique device id to the mount 2022 */ 2023 mutex_enter(&nfs_minor_lock); 2024 do { 2025 nfs_minor = (nfs_minor + 1) & MAXMIN32; 2026 nfs_dev = makedevice(nfs_major, nfs_minor); 2027 } while (vfs_devismounted(nfs_dev)); 2028 mutex_exit(&nfs_minor_lock); 2029 2030 vfsp->vfs_dev = nfs_dev; 2031 vfs_make_fsid(&vfsp->vfs_fsid, nfs_dev, nfs4fstyp); 2032 vfsp->vfs_data = (caddr_t)mi; 2033 vfsp->vfs_fstype = nfsfstyp; 2034 vfsp->vfs_bsize = nfs4_bsize; 2035 2036 /* 2037 * Initialize fields used to support async putpage operations. 2038 */ 2039 for (i = 0; i < NFS4_ASYNC_TYPES; i++) 2040 mi->mi_async_clusters[i] = nfs4_async_clusters; 2041 mi->mi_async_init_clusters = nfs4_async_clusters; 2042 mi->mi_async_curr = &mi->mi_async_reqs[0]; 2043 mi->mi_max_threads = nfs4_max_threads; 2044 mutex_init(&mi->mi_async_lock, NULL, MUTEX_DEFAULT, NULL); 2045 cv_init(&mi->mi_async_reqs_cv, NULL, CV_DEFAULT, NULL); 2046 cv_init(&mi->mi_async_work_cv, NULL, CV_DEFAULT, NULL); 2047 cv_init(&mi->mi_async_cv, NULL, CV_DEFAULT, NULL); 2048 cv_init(&mi->mi_inact_req_cv, NULL, CV_DEFAULT, NULL); 2049 2050 mi->mi_vfsp = vfsp; 2051 zone_hold(mi->mi_zone = zone); 2052 nfs4_mi_zonelist_add(mi); 2053 2054 /* 2055 * Initialize the <open owner/cred> hash table. 2056 */ 2057 for (i = 0; i < NFS4_NUM_OO_BUCKETS; i++) { 2058 bucketp = &(mi->mi_oo_list[i]); 2059 mutex_init(&bucketp->b_lock, NULL, MUTEX_DEFAULT, NULL); 2060 list_create(&bucketp->b_oo_hash_list, 2061 sizeof (nfs4_open_owner_t), 2062 offsetof(nfs4_open_owner_t, oo_hash_node)); 2063 } 2064 2065 /* 2066 * Initialize the freed open owner list. 2067 */ 2068 mi->mi_foo_num = 0; 2069 mi->mi_foo_max = NFS4_NUM_FREED_OPEN_OWNERS; 2070 list_create(&mi->mi_foo_list, sizeof (nfs4_open_owner_t), 2071 offsetof(nfs4_open_owner_t, oo_foo_node)); 2072 2073 list_create(&mi->mi_lost_state, sizeof (nfs4_lost_rqst_t), 2074 offsetof(nfs4_lost_rqst_t, lr_node)); 2075 2076 list_create(&mi->mi_bseqid_list, sizeof (nfs4_bseqid_entry_t), 2077 offsetof(nfs4_bseqid_entry_t, bs_node)); 2078 2079 /* 2080 * Initialize the msg buffer. 2081 */ 2082 list_create(&mi->mi_msg_list, sizeof (nfs4_debug_msg_t), 2083 offsetof(nfs4_debug_msg_t, msg_node)); 2084 mi->mi_msg_count = 0; 2085 mutex_init(&mi->mi_msg_list_lock, NULL, MUTEX_DEFAULT, NULL); 2086 2087 /* 2088 * Initialize kstats 2089 */ 2090 nfs4_mnt_kstat_init(vfsp); 2091 2092 /* 2093 * Initialize the shared filehandle pool, and get the fname for 2094 * the filesystem root. 2095 */ 2096 sfh4_createtab(&mi->mi_filehandles); 2097 mi->mi_fname = fn_get(NULL, "."); 2098 2099 /* 2100 * Save server path we're attempting to mount. 2101 */ 2102 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0); 2103 orig_sv_pathlen = svp_head->sv_pathlen; 2104 orig_sv_path = kmem_alloc(svp_head->sv_pathlen, KM_SLEEP); 2105 bcopy(svp_head->sv_path, orig_sv_path, svp_head->sv_pathlen); 2106 nfs_rw_exit(&svp->sv_lock); 2107 2108 /* 2109 * Make the GETFH call to get root fh for each replica. 2110 */ 2111 if (svp_head->sv_next) 2112 droptext = ", dropping replica"; 2113 2114 /* 2115 * If the uid is set then set the creds for secure mounts 2116 * by proxy processes such as automountd. 2117 */ 2118 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 2119 if (svp->sv_secdata->uid != 0) { 2120 lcr = crdup(cr); 2121 (void) crsetugid(lcr, svp->sv_secdata->uid, crgetgid(cr)); 2122 tcr = lcr; 2123 } 2124 nfs_rw_exit(&svp->sv_lock); 2125 for (svp = svp_head; svp; svp = svp->sv_next) { 2126 if (nfs4_chkdup_servinfo4(svp_head, svp)) { 2127 nfs_cmn_err(error, CE_WARN, 2128 VERS_MSG "Host %s is a duplicate%s", 2129 svp->sv_hostname, droptext); 2130 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0); 2131 svp->sv_flags |= SV4_NOTINUSE; 2132 nfs_rw_exit(&svp->sv_lock); 2133 continue; 2134 } 2135 mi->mi_curr_serv = svp; 2136 2137 /* 2138 * Just in case server path being mounted contains 2139 * symlinks and fails w/STALE, save the initial sv_path 2140 * so we can redrive the initial mount compound with the 2141 * initial sv_path -- not a symlink-expanded version. 2142 * 2143 * This could only happen if a symlink was expanded 2144 * and the expanded mount compound failed stale. Because 2145 * it could be the case that the symlink was removed at 2146 * the server (and replaced with another symlink/dir, 2147 * we need to use the initial sv_path when attempting 2148 * to re-lookup everything and recover. 2149 * 2150 * Other mount errors should evenutally be handled here also 2151 * (NFS4ERR_DELAY, NFS4ERR_RESOURCE). For now, all mount 2152 * failures will result in mount being redriven a few times. 2153 */ 2154 num_retry = nfs4_max_mount_retry; 2155 do { 2156 nfs4getfh_otw(mi, svp, &tmp_vtype, 2157 ((flags & NFSMNT_PUBLIC) ? NFS4_GETFH_PUBLIC : 0) | 2158 NFS4_GETFH_NEEDSOP, tcr, &e); 2159 2160 if (e.error == 0 && e.stat == NFS4_OK) 2161 break; 2162 2163 /* 2164 * replace current sv_path with orig sv_path -- just in 2165 * case it changed due to embedded symlinks. 2166 */ 2167 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 2168 if (orig_sv_pathlen != svp->sv_pathlen) { 2169 kmem_free(svp->sv_path, svp->sv_pathlen); 2170 svp->sv_path = kmem_alloc(orig_sv_pathlen, 2171 KM_SLEEP); 2172 svp->sv_pathlen = orig_sv_pathlen; 2173 } 2174 bcopy(orig_sv_path, svp->sv_path, orig_sv_pathlen); 2175 nfs_rw_exit(&svp->sv_lock); 2176 2177 } while (num_retry-- > 0); 2178 2179 error = e.error ? e.error : geterrno4(e.stat); 2180 if (error) { 2181 nfs_cmn_err(error, CE_WARN, 2182 VERS_MSG "initial call to %s failed%s: %m", 2183 svp->sv_hostname, droptext); 2184 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0); 2185 svp->sv_flags |= SV4_NOTINUSE; 2186 nfs_rw_exit(&svp->sv_lock); 2187 mi->mi_flags &= ~MI4_RECOV_FAIL; 2188 mi->mi_error = 0; 2189 continue; 2190 } 2191 2192 if (tmp_vtype == VBAD) { 2193 zcmn_err(mi->mi_zone->zone_id, CE_WARN, 2194 VERS_MSG "%s returned a bad file type for " 2195 "root%s", svp->sv_hostname, droptext); 2196 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0); 2197 svp->sv_flags |= SV4_NOTINUSE; 2198 nfs_rw_exit(&svp->sv_lock); 2199 continue; 2200 } 2201 2202 if (vtype == VNON) { 2203 vtype = tmp_vtype; 2204 } else if (vtype != tmp_vtype) { 2205 zcmn_err(mi->mi_zone->zone_id, CE_WARN, 2206 VERS_MSG "%s returned a different file type " 2207 "for root%s", svp->sv_hostname, droptext); 2208 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0); 2209 svp->sv_flags |= SV4_NOTINUSE; 2210 nfs_rw_exit(&svp->sv_lock); 2211 continue; 2212 } 2213 if (firstsvp == NULL) 2214 firstsvp = svp; 2215 } 2216 2217 kmem_free(orig_sv_path, orig_sv_pathlen); 2218 2219 if (firstsvp == NULL) { 2220 if (error == 0) 2221 error = ENOENT; 2222 goto bad; 2223 } 2224 2225 mi->mi_curr_serv = svp = firstsvp; 2226 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 2227 ASSERT((mi->mi_curr_serv->sv_flags & SV4_NOTINUSE) == 0); 2228 fh.nfs_fh4_len = svp->sv_fhandle.fh_len; 2229 fh.nfs_fh4_val = svp->sv_fhandle.fh_buf; 2230 mi->mi_rootfh = sfh4_get(&fh, mi); 2231 fh.nfs_fh4_len = svp->sv_pfhandle.fh_len; 2232 fh.nfs_fh4_val = svp->sv_pfhandle.fh_buf; 2233 mi->mi_srvparentfh = sfh4_get(&fh, mi); 2234 nfs_rw_exit(&svp->sv_lock); 2235 2236 /* 2237 * Make the root vnode without attributes. 2238 */ 2239 mfname = mi->mi_fname; 2240 fn_hold(mfname); 2241 rtvp = makenfs4node_by_fh(mi->mi_rootfh, NULL, 2242 &mfname, NULL, mi, cr, gethrtime()); 2243 rtvp->v_type = vtype; 2244 2245 mi->mi_curread = mi->mi_tsize; 2246 mi->mi_curwrite = mi->mi_stsize; 2247 2248 /* 2249 * Start the manager thread responsible for handling async worker 2250 * threads. 2251 */ 2252 MI4_HOLD(mi); 2253 VFS_HOLD(vfsp); /* add reference for thread */ 2254 mi->mi_manager_thread = zthread_create(NULL, 0, nfs4_async_manager, 2255 vfsp, 0, minclsyspri); 2256 ASSERT(mi->mi_manager_thread != NULL); 2257 2258 /* 2259 * Create the thread that handles over-the-wire calls for 2260 * VOP_INACTIVE. 2261 * This needs to happen after the manager thread is created. 2262 */ 2263 MI4_HOLD(mi); 2264 mi->mi_inactive_thread = zthread_create(NULL, 0, nfs4_inactive_thread, 2265 mi, 0, minclsyspri); 2266 ASSERT(mi->mi_inactive_thread != NULL); 2267 2268 /* If we didn't get a type, get one now */ 2269 if (rtvp->v_type == VNON) { 2270 va.va_mask = AT_TYPE; 2271 error = nfs4getattr(rtvp, &va, tcr); 2272 if (error) 2273 goto bad; 2274 rtvp->v_type = va.va_type; 2275 } 2276 2277 mi->mi_type = rtvp->v_type; 2278 2279 mutex_enter(&mi->mi_lock); 2280 mi->mi_flags &= ~MI4_MOUNTING; 2281 mutex_exit(&mi->mi_lock); 2282 2283 *rtvpp = rtvp; 2284 if (lcr != NULL) 2285 crfree(lcr); 2286 2287 return (0); 2288 bad: 2289 /* 2290 * An error occurred somewhere, need to clean up... 2291 */ 2292 if (lcr != NULL) 2293 crfree(lcr); 2294 2295 if (rtvp != NULL) { 2296 /* 2297 * We need to release our reference to the root vnode and 2298 * destroy the mntinfo4 struct that we just created. 2299 */ 2300 rp = VTOR4(rtvp); 2301 if (rp->r_flags & R4HASHED) 2302 rp4_rmhash(rp); 2303 VN_RELE(rtvp); 2304 } 2305 nfs4_async_stop(vfsp); 2306 nfs4_async_manager_stop(vfsp); 2307 removed = nfs4_mi_zonelist_remove(mi); 2308 if (removed) 2309 zone_rele(mi->mi_zone); 2310 2311 /* 2312 * This releases the initial "hold" of the mi since it will never 2313 * be referenced by the vfsp. Also, when mount returns to vfs.c 2314 * with an error, the vfsp will be destroyed, not rele'd. 2315 */ 2316 MI4_RELE(mi); 2317 2318 *rtvpp = NULL; 2319 return (error); 2320 } 2321 2322 /* 2323 * vfs operations 2324 */ 2325 static int 2326 nfs4_unmount(vfs_t *vfsp, int flag, cred_t *cr) 2327 { 2328 mntinfo4_t *mi; 2329 ushort_t omax; 2330 int removed; 2331 2332 bool_t must_unlock; 2333 2334 nfs4_ephemeral_tree_t *eph_tree; 2335 2336 if (secpolicy_fs_unmount(cr, vfsp) != 0) 2337 return (EPERM); 2338 2339 mi = VFTOMI4(vfsp); 2340 2341 if (flag & MS_FORCE) { 2342 vfsp->vfs_flag |= VFS_UNMOUNTED; 2343 if (nfs_zone() != mi->mi_zone) { 2344 /* 2345 * If the request is coming from the wrong zone, 2346 * we don't want to create any new threads, and 2347 * performance is not a concern. Do everything 2348 * inline. 2349 */ 2350 NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE, 2351 "nfs4_unmount x-zone forced unmount of vfs %p\n", 2352 (void *)vfsp)); 2353 nfs4_free_mount(vfsp, flag, cr); 2354 } else { 2355 /* 2356 * Free data structures asynchronously, to avoid 2357 * blocking the current thread (for performance 2358 * reasons only). 2359 */ 2360 async_free_mount(vfsp, flag, cr); 2361 } 2362 2363 return (0); 2364 } 2365 2366 /* 2367 * Wait until all asynchronous putpage operations on 2368 * this file system are complete before flushing rnodes 2369 * from the cache. 2370 */ 2371 omax = mi->mi_max_threads; 2372 if (nfs4_async_stop_sig(vfsp)) 2373 return (EINTR); 2374 2375 r4flush(vfsp, cr); 2376 2377 /* 2378 * About the only reason that this would fail would be 2379 * that the harvester is already busy tearing down this 2380 * node. So we fail back to the caller and let them try 2381 * again when needed. 2382 */ 2383 if (nfs4_ephemeral_umount(mi, flag, cr, 2384 &must_unlock, &eph_tree)) { 2385 ASSERT(must_unlock == FALSE); 2386 mutex_enter(&mi->mi_async_lock); 2387 mi->mi_max_threads = omax; 2388 mutex_exit(&mi->mi_async_lock); 2389 2390 return (EBUSY); 2391 } 2392 2393 /* 2394 * If there are any active vnodes on this file system, 2395 * then the file system is busy and can't be unmounted. 2396 */ 2397 if (check_rtable4(vfsp)) { 2398 nfs4_ephemeral_umount_unlock(&must_unlock, &eph_tree); 2399 2400 mutex_enter(&mi->mi_async_lock); 2401 mi->mi_max_threads = omax; 2402 mutex_exit(&mi->mi_async_lock); 2403 2404 return (EBUSY); 2405 } 2406 2407 /* 2408 * The unmount can't fail from now on, so record any 2409 * ephemeral changes. 2410 */ 2411 nfs4_ephemeral_umount_activate(mi, &must_unlock, &eph_tree); 2412 2413 /* 2414 * There are no active files that could require over-the-wire 2415 * calls to the server, so stop the async manager and the 2416 * inactive thread. 2417 */ 2418 nfs4_async_manager_stop(vfsp); 2419 2420 /* 2421 * Destroy all rnodes belonging to this file system from the 2422 * rnode hash queues and purge any resources allocated to 2423 * them. 2424 */ 2425 destroy_rtable4(vfsp, cr); 2426 vfsp->vfs_flag |= VFS_UNMOUNTED; 2427 2428 nfs4_remove_mi_from_server(mi, NULL); 2429 removed = nfs4_mi_zonelist_remove(mi); 2430 if (removed) 2431 zone_rele(mi->mi_zone); 2432 2433 return (0); 2434 } 2435 2436 /* 2437 * find root of nfs 2438 */ 2439 static int 2440 nfs4_root(vfs_t *vfsp, vnode_t **vpp) 2441 { 2442 mntinfo4_t *mi; 2443 vnode_t *vp; 2444 nfs4_fname_t *mfname; 2445 servinfo4_t *svp; 2446 2447 mi = VFTOMI4(vfsp); 2448 2449 if (nfs_zone() != mi->mi_zone) 2450 return (EPERM); 2451 2452 svp = mi->mi_curr_serv; 2453 if (svp) { 2454 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 2455 if (svp->sv_flags & SV4_ROOT_STALE) { 2456 nfs_rw_exit(&svp->sv_lock); 2457 2458 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_WRITER, 0); 2459 if (svp->sv_flags & SV4_ROOT_STALE) { 2460 svp->sv_flags &= ~SV4_ROOT_STALE; 2461 nfs_rw_exit(&svp->sv_lock); 2462 return (ENOENT); 2463 } 2464 nfs_rw_exit(&svp->sv_lock); 2465 } else 2466 nfs_rw_exit(&svp->sv_lock); 2467 } 2468 2469 mfname = mi->mi_fname; 2470 fn_hold(mfname); 2471 vp = makenfs4node_by_fh(mi->mi_rootfh, NULL, &mfname, NULL, 2472 VFTOMI4(vfsp), CRED(), gethrtime()); 2473 2474 if (VTOR4(vp)->r_flags & R4STALE) { 2475 VN_RELE(vp); 2476 return (ENOENT); 2477 } 2478 2479 ASSERT(vp->v_type == VNON || vp->v_type == mi->mi_type); 2480 2481 vp->v_type = mi->mi_type; 2482 2483 *vpp = vp; 2484 2485 return (0); 2486 } 2487 2488 static int 2489 nfs4_statfs_otw(vnode_t *vp, struct statvfs64 *sbp, cred_t *cr) 2490 { 2491 int error; 2492 nfs4_ga_res_t gar; 2493 nfs4_ga_ext_res_t ger; 2494 2495 gar.n4g_ext_res = &ger; 2496 2497 if (error = nfs4_attr_otw(vp, TAG_FSINFO, &gar, 2498 NFS4_STATFS_ATTR_MASK, cr)) 2499 return (error); 2500 2501 *sbp = gar.n4g_ext_res->n4g_sb; 2502 2503 return (0); 2504 } 2505 2506 /* 2507 * Get file system statistics. 2508 */ 2509 static int 2510 nfs4_statvfs(vfs_t *vfsp, struct statvfs64 *sbp) 2511 { 2512 int error; 2513 vnode_t *vp; 2514 cred_t *cr; 2515 2516 error = nfs4_root(vfsp, &vp); 2517 if (error) 2518 return (error); 2519 2520 cr = CRED(); 2521 2522 error = nfs4_statfs_otw(vp, sbp, cr); 2523 if (!error) { 2524 (void) strncpy(sbp->f_basetype, 2525 vfssw[vfsp->vfs_fstype].vsw_name, FSTYPSZ); 2526 sbp->f_flag = vf_to_stf(vfsp->vfs_flag); 2527 } else { 2528 nfs4_purge_stale_fh(error, vp, cr); 2529 } 2530 2531 VN_RELE(vp); 2532 2533 return (error); 2534 } 2535 2536 static kmutex_t nfs4_syncbusy; 2537 2538 /* 2539 * Flush dirty nfs files for file system vfsp. 2540 * If vfsp == NULL, all nfs files are flushed. 2541 * 2542 * SYNC_CLOSE in flag is passed to us to 2543 * indicate that we are shutting down and or 2544 * rebooting. 2545 */ 2546 static int 2547 nfs4_sync(vfs_t *vfsp, short flag, cred_t *cr) 2548 { 2549 /* 2550 * Cross-zone calls are OK here, since this translates to a 2551 * VOP_PUTPAGE(B_ASYNC), which gets picked up by the right zone. 2552 */ 2553 if (!(flag & SYNC_ATTR) && mutex_tryenter(&nfs4_syncbusy) != 0) { 2554 r4flush(vfsp, cr); 2555 mutex_exit(&nfs4_syncbusy); 2556 } 2557 2558 /* 2559 * if SYNC_CLOSE is set then we know that 2560 * the system is rebooting, mark the mntinfo 2561 * for later examination. 2562 */ 2563 if (vfsp && (flag & SYNC_CLOSE)) { 2564 mntinfo4_t *mi; 2565 2566 mi = VFTOMI4(vfsp); 2567 if (!(mi->mi_flags & MI4_SHUTDOWN)) { 2568 mutex_enter(&mi->mi_lock); 2569 mi->mi_flags |= MI4_SHUTDOWN; 2570 mutex_exit(&mi->mi_lock); 2571 } 2572 } 2573 return (0); 2574 } 2575 2576 /* 2577 * vget is difficult, if not impossible, to support in v4 because we don't 2578 * know the parent directory or name, which makes it impossible to create a 2579 * useful shadow vnode. And we need the shadow vnode for things like 2580 * OPEN. 2581 */ 2582 2583 /* ARGSUSED */ 2584 /* 2585 * XXX Check nfs4_vget_pseudo() for dependency. 2586 */ 2587 static int 2588 nfs4_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp) 2589 { 2590 return (EREMOTE); 2591 } 2592 2593 /* 2594 * nfs4_mountroot get called in the case where we are diskless booting. All 2595 * we need from here is the ability to get the server info and from there we 2596 * can simply call nfs4_rootvp. 2597 */ 2598 /* ARGSUSED */ 2599 static int 2600 nfs4_mountroot(vfs_t *vfsp, whymountroot_t why) 2601 { 2602 vnode_t *rtvp; 2603 char root_hostname[SYS_NMLN+1]; 2604 struct servinfo4 *svp; 2605 int error; 2606 int vfsflags; 2607 size_t size; 2608 char *root_path; 2609 struct pathname pn; 2610 char *name; 2611 cred_t *cr; 2612 mntinfo4_t *mi; 2613 struct nfs_args args; /* nfs mount arguments */ 2614 static char token[10]; 2615 nfs4_error_t n4e; 2616 2617 bzero(&args, sizeof (args)); 2618 2619 /* do this BEFORE getfile which causes xid stamps to be initialized */ 2620 clkset(-1L); /* hack for now - until we get time svc? */ 2621 2622 if (why == ROOT_REMOUNT) { 2623 /* 2624 * Shouldn't happen. 2625 */ 2626 panic("nfs4_mountroot: why == ROOT_REMOUNT"); 2627 } 2628 2629 if (why == ROOT_UNMOUNT) { 2630 /* 2631 * Nothing to do for NFS. 2632 */ 2633 return (0); 2634 } 2635 2636 /* 2637 * why == ROOT_INIT 2638 */ 2639 2640 name = token; 2641 *name = 0; 2642 (void) getfsname("root", name, sizeof (token)); 2643 2644 pn_alloc(&pn); 2645 root_path = pn.pn_path; 2646 2647 svp = kmem_zalloc(sizeof (*svp), KM_SLEEP); 2648 nfs_rw_init(&svp->sv_lock, NULL, RW_DEFAULT, NULL); 2649 svp->sv_knconf = kmem_zalloc(sizeof (*svp->sv_knconf), KM_SLEEP); 2650 svp->sv_knconf->knc_protofmly = kmem_alloc(KNC_STRSIZE, KM_SLEEP); 2651 svp->sv_knconf->knc_proto = kmem_alloc(KNC_STRSIZE, KM_SLEEP); 2652 2653 /* 2654 * Get server address 2655 * Get the root path 2656 * Get server's transport 2657 * Get server's hostname 2658 * Get options 2659 */ 2660 args.addr = &svp->sv_addr; 2661 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 2662 args.fh = (char *)&svp->sv_fhandle; 2663 args.knconf = svp->sv_knconf; 2664 args.hostname = root_hostname; 2665 vfsflags = 0; 2666 if (error = mount_root(*name ? name : "root", root_path, NFS_V4, 2667 &args, &vfsflags)) { 2668 if (error == EPROTONOSUPPORT) 2669 nfs_cmn_err(error, CE_WARN, "nfs4_mountroot: " 2670 "mount_root failed: server doesn't support NFS V4"); 2671 else 2672 nfs_cmn_err(error, CE_WARN, 2673 "nfs4_mountroot: mount_root failed: %m"); 2674 nfs_rw_exit(&svp->sv_lock); 2675 sv4_free(svp); 2676 pn_free(&pn); 2677 return (error); 2678 } 2679 nfs_rw_exit(&svp->sv_lock); 2680 svp->sv_hostnamelen = (int)(strlen(root_hostname) + 1); 2681 svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP); 2682 (void) strcpy(svp->sv_hostname, root_hostname); 2683 2684 svp->sv_pathlen = (int)(strlen(root_path) + 1); 2685 svp->sv_path = kmem_alloc(svp->sv_pathlen, KM_SLEEP); 2686 (void) strcpy(svp->sv_path, root_path); 2687 2688 /* 2689 * Force root partition to always be mounted with AUTH_UNIX for now 2690 */ 2691 svp->sv_secdata = kmem_alloc(sizeof (*svp->sv_secdata), KM_SLEEP); 2692 svp->sv_secdata->secmod = AUTH_UNIX; 2693 svp->sv_secdata->rpcflavor = AUTH_UNIX; 2694 svp->sv_secdata->data = NULL; 2695 2696 cr = crgetcred(); 2697 rtvp = NULL; 2698 2699 error = nfs4rootvp(&rtvp, vfsp, svp, args.flags, cr, global_zone); 2700 2701 if (error) { 2702 crfree(cr); 2703 pn_free(&pn); 2704 sv4_free(svp); 2705 return (error); 2706 } 2707 2708 mi = VTOMI4(rtvp); 2709 2710 /* 2711 * Send client id to the server, if necessary 2712 */ 2713 nfs4_error_zinit(&n4e); 2714 nfs4setclientid(mi, cr, FALSE, &n4e); 2715 error = n4e.error; 2716 2717 crfree(cr); 2718 2719 if (error) { 2720 pn_free(&pn); 2721 goto errout; 2722 } 2723 2724 error = nfs4_setopts(rtvp, DATAMODEL_NATIVE, &args); 2725 if (error) { 2726 nfs_cmn_err(error, CE_WARN, 2727 "nfs4_mountroot: invalid root mount options"); 2728 pn_free(&pn); 2729 goto errout; 2730 } 2731 2732 (void) vfs_lock_wait(vfsp); 2733 vfs_add(NULL, vfsp, vfsflags); 2734 vfs_unlock(vfsp); 2735 2736 size = strlen(svp->sv_hostname); 2737 (void) strcpy(rootfs.bo_name, svp->sv_hostname); 2738 rootfs.bo_name[size] = ':'; 2739 (void) strcpy(&rootfs.bo_name[size + 1], root_path); 2740 2741 pn_free(&pn); 2742 2743 errout: 2744 if (error) { 2745 sv4_free(svp); 2746 nfs4_async_stop(vfsp); 2747 nfs4_async_manager_stop(vfsp); 2748 } 2749 2750 if (rtvp != NULL) 2751 VN_RELE(rtvp); 2752 2753 return (error); 2754 } 2755 2756 /* 2757 * Initialization routine for VFS routines. Should only be called once 2758 */ 2759 int 2760 nfs4_vfsinit(void) 2761 { 2762 mutex_init(&nfs4_syncbusy, NULL, MUTEX_DEFAULT, NULL); 2763 nfs4setclientid_init(); 2764 nfs4_ephemeral_init(); 2765 return (0); 2766 } 2767 2768 void 2769 nfs4_vfsfini(void) 2770 { 2771 nfs4_ephemeral_fini(); 2772 nfs4setclientid_fini(); 2773 mutex_destroy(&nfs4_syncbusy); 2774 } 2775 2776 void 2777 nfs4_freevfs(vfs_t *vfsp) 2778 { 2779 mntinfo4_t *mi; 2780 2781 /* need to release the initial hold */ 2782 mi = VFTOMI4(vfsp); 2783 MI4_RELE(mi); 2784 } 2785 2786 /* 2787 * Client side SETCLIENTID and SETCLIENTID_CONFIRM 2788 */ 2789 struct nfs4_server nfs4_server_lst = 2790 { &nfs4_server_lst, &nfs4_server_lst }; 2791 2792 kmutex_t nfs4_server_lst_lock; 2793 2794 static void 2795 nfs4setclientid_init(void) 2796 { 2797 mutex_init(&nfs4_server_lst_lock, NULL, MUTEX_DEFAULT, NULL); 2798 } 2799 2800 static void 2801 nfs4setclientid_fini(void) 2802 { 2803 mutex_destroy(&nfs4_server_lst_lock); 2804 } 2805 2806 int nfs4_retry_sclid_delay = NFS4_RETRY_SCLID_DELAY; 2807 int nfs4_num_sclid_retries = NFS4_NUM_SCLID_RETRIES; 2808 2809 /* 2810 * Set the clientid for the server for "mi". No-op if the clientid is 2811 * already set. 2812 * 2813 * The recovery boolean should be set to TRUE if this function was called 2814 * by the recovery code, and FALSE otherwise. This is used to determine 2815 * if we need to call nfs4_start/end_op as well as grab the mi_recovlock 2816 * for adding a mntinfo4_t to a nfs4_server_t. 2817 * 2818 * Error is returned via 'n4ep'. If there was a 'n4ep->stat' error, then 2819 * 'n4ep->error' is set to geterrno4(n4ep->stat). 2820 */ 2821 void 2822 nfs4setclientid(mntinfo4_t *mi, cred_t *cr, bool_t recovery, nfs4_error_t *n4ep) 2823 { 2824 struct nfs4_server *np; 2825 struct servinfo4 *svp = mi->mi_curr_serv; 2826 nfs4_recov_state_t recov_state; 2827 int num_retries = 0; 2828 bool_t retry; 2829 cred_t *lcr = NULL; 2830 int retry_inuse = 1; /* only retry once on NFS4ERR_CLID_INUSE */ 2831 time_t lease_time = 0; 2832 2833 recov_state.rs_flags = 0; 2834 recov_state.rs_num_retry_despite_err = 0; 2835 ASSERT(n4ep != NULL); 2836 2837 recov_retry: 2838 retry = FALSE; 2839 nfs4_error_zinit(n4ep); 2840 if (!recovery) 2841 (void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, 0); 2842 2843 mutex_enter(&nfs4_server_lst_lock); 2844 np = servinfo4_to_nfs4_server(svp); /* This locks np if it is found */ 2845 mutex_exit(&nfs4_server_lst_lock); 2846 if (!np) { 2847 struct nfs4_server *tnp; 2848 np = new_nfs4_server(svp, cr); 2849 mutex_enter(&np->s_lock); 2850 2851 mutex_enter(&nfs4_server_lst_lock); 2852 tnp = servinfo4_to_nfs4_server(svp); 2853 if (tnp) { 2854 /* 2855 * another thread snuck in and put server on list. 2856 * since we aren't adding it to the nfs4_server_list 2857 * we need to set the ref count to 0 and destroy it. 2858 */ 2859 np->s_refcnt = 0; 2860 destroy_nfs4_server(np); 2861 np = tnp; 2862 } else { 2863 /* 2864 * do not give list a reference until everything 2865 * succeeds 2866 */ 2867 insque(np, &nfs4_server_lst); 2868 } 2869 mutex_exit(&nfs4_server_lst_lock); 2870 } 2871 ASSERT(MUTEX_HELD(&np->s_lock)); 2872 /* 2873 * If we find the server already has N4S_CLIENTID_SET, then 2874 * just return, we've already done SETCLIENTID to that server 2875 */ 2876 if (np->s_flags & N4S_CLIENTID_SET) { 2877 /* add mi to np's mntinfo4_list */ 2878 nfs4_add_mi_to_server(np, mi); 2879 if (!recovery) 2880 nfs_rw_exit(&mi->mi_recovlock); 2881 mutex_exit(&np->s_lock); 2882 nfs4_server_rele(np); 2883 return; 2884 } 2885 mutex_exit(&np->s_lock); 2886 2887 2888 /* 2889 * Drop the mi_recovlock since nfs4_start_op will 2890 * acquire it again for us. 2891 */ 2892 if (!recovery) { 2893 nfs_rw_exit(&mi->mi_recovlock); 2894 2895 n4ep->error = nfs4_start_op(mi, NULL, NULL, &recov_state); 2896 if (n4ep->error) { 2897 nfs4_server_rele(np); 2898 return; 2899 } 2900 } 2901 2902 mutex_enter(&np->s_lock); 2903 while (np->s_flags & N4S_CLIENTID_PEND) { 2904 if (!cv_wait_sig(&np->s_clientid_pend, &np->s_lock)) { 2905 mutex_exit(&np->s_lock); 2906 nfs4_server_rele(np); 2907 if (!recovery) 2908 nfs4_end_op(mi, NULL, NULL, &recov_state, 2909 recovery); 2910 n4ep->error = EINTR; 2911 return; 2912 } 2913 } 2914 2915 if (np->s_flags & N4S_CLIENTID_SET) { 2916 /* XXX copied/pasted from above */ 2917 /* add mi to np's mntinfo4_list */ 2918 nfs4_add_mi_to_server(np, mi); 2919 mutex_exit(&np->s_lock); 2920 nfs4_server_rele(np); 2921 if (!recovery) 2922 nfs4_end_op(mi, NULL, NULL, &recov_state, recovery); 2923 return; 2924 } 2925 2926 /* 2927 * Reset the N4S_CB_PINGED flag. This is used to 2928 * indicate if we have received a CB_NULL from the 2929 * server. Also we reset the waiter flag. 2930 */ 2931 np->s_flags &= ~(N4S_CB_PINGED | N4S_CB_WAITER); 2932 /* any failure must now clear this flag */ 2933 np->s_flags |= N4S_CLIENTID_PEND; 2934 mutex_exit(&np->s_lock); 2935 nfs4setclientid_otw(mi, svp, cr, np, n4ep, &retry_inuse); 2936 2937 if (n4ep->error == EACCES) { 2938 /* 2939 * If the uid is set then set the creds for secure mounts 2940 * by proxy processes such as automountd. 2941 */ 2942 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0); 2943 if (svp->sv_secdata->uid != 0) { 2944 lcr = crdup(cr); 2945 (void) crsetugid(lcr, svp->sv_secdata->uid, 2946 crgetgid(cr)); 2947 } 2948 nfs_rw_exit(&svp->sv_lock); 2949 2950 if (lcr != NULL) { 2951 mutex_enter(&np->s_lock); 2952 crfree(np->s_cred); 2953 np->s_cred = lcr; 2954 mutex_exit(&np->s_lock); 2955 nfs4setclientid_otw(mi, svp, lcr, np, n4ep, 2956 &retry_inuse); 2957 } 2958 } 2959 mutex_enter(&np->s_lock); 2960 lease_time = np->s_lease_time; 2961 np->s_flags &= ~N4S_CLIENTID_PEND; 2962 mutex_exit(&np->s_lock); 2963 2964 if (n4ep->error != 0 || n4ep->stat != NFS4_OK) { 2965 /* 2966 * Start recovery if failover is a possibility. If 2967 * invoked by the recovery thread itself, then just 2968 * return and let it handle the failover first. NB: 2969 * recovery is not allowed if the mount is in progress 2970 * since the infrastructure is not sufficiently setup 2971 * to allow it. Just return the error (after suitable 2972 * retries). 2973 */ 2974 if (FAILOVER_MOUNT4(mi) && nfs4_try_failover(n4ep)) { 2975 (void) nfs4_start_recovery(n4ep, mi, NULL, 2976 NULL, NULL, NULL, OP_SETCLIENTID, NULL); 2977 /* 2978 * Don't retry here, just return and let 2979 * recovery take over. 2980 */ 2981 if (recovery) 2982 retry = FALSE; 2983 } else if (nfs4_rpc_retry_error(n4ep->error) || 2984 n4ep->stat == NFS4ERR_RESOURCE || 2985 n4ep->stat == NFS4ERR_STALE_CLIENTID) { 2986 2987 retry = TRUE; 2988 /* 2989 * Always retry if in recovery or once had 2990 * contact with the server (but now it's 2991 * overloaded). 2992 */ 2993 if (recovery == TRUE || 2994 n4ep->error == ETIMEDOUT || 2995 n4ep->error == ECONNRESET) 2996 num_retries = 0; 2997 } else if (retry_inuse && n4ep->error == 0 && 2998 n4ep->stat == NFS4ERR_CLID_INUSE) { 2999 retry = TRUE; 3000 num_retries = 0; 3001 } 3002 } else { 3003 /* 3004 * Since everything succeeded give the list a reference count if 3005 * it hasn't been given one by add_new_nfs4_server() or if this 3006 * is not a recovery situation in which case it is already on 3007 * the list. 3008 */ 3009 mutex_enter(&np->s_lock); 3010 if ((np->s_flags & N4S_INSERTED) == 0) { 3011 np->s_refcnt++; 3012 np->s_flags |= N4S_INSERTED; 3013 } 3014 mutex_exit(&np->s_lock); 3015 } 3016 3017 if (!recovery) 3018 nfs4_end_op(mi, NULL, NULL, &recov_state, recovery); 3019 3020 3021 if (retry && num_retries++ < nfs4_num_sclid_retries) { 3022 if (retry_inuse) { 3023 delay(SEC_TO_TICK(lease_time + nfs4_retry_sclid_delay)); 3024 retry_inuse = 0; 3025 } else 3026 delay(SEC_TO_TICK(nfs4_retry_sclid_delay)); 3027 3028 nfs4_server_rele(np); 3029 goto recov_retry; 3030 } 3031 3032 3033 if (n4ep->error == 0) 3034 n4ep->error = geterrno4(n4ep->stat); 3035 3036 /* broadcast before release in case no other threads are waiting */ 3037 cv_broadcast(&np->s_clientid_pend); 3038 nfs4_server_rele(np); 3039 } 3040 3041 int nfs4setclientid_otw_debug = 0; 3042 3043 /* 3044 * This function handles the recovery of STALE_CLIENTID for SETCLIENTID_CONFRIM, 3045 * but nothing else; the calling function must be designed to handle those 3046 * other errors. 3047 */ 3048 static void 3049 nfs4setclientid_otw(mntinfo4_t *mi, struct servinfo4 *svp, cred_t *cr, 3050 struct nfs4_server *np, nfs4_error_t *ep, int *retry_inusep) 3051 { 3052 COMPOUND4args_clnt args; 3053 COMPOUND4res_clnt res; 3054 nfs_argop4 argop[3]; 3055 SETCLIENTID4args *s_args; 3056 SETCLIENTID4resok *s_resok; 3057 int doqueue = 1; 3058 nfs4_ga_res_t *garp = NULL; 3059 timespec_t prop_time, after_time; 3060 verifier4 verf; 3061 clientid4 tmp_clientid; 3062 3063 ASSERT(!MUTEX_HELD(&np->s_lock)); 3064 3065 args.ctag = TAG_SETCLIENTID; 3066 3067 args.array = argop; 3068 args.array_len = 3; 3069 3070 /* PUTROOTFH */ 3071 argop[0].argop = OP_PUTROOTFH; 3072 3073 /* GETATTR */ 3074 argop[1].argop = OP_GETATTR; 3075 argop[1].nfs_argop4_u.opgetattr.attr_request = FATTR4_LEASE_TIME_MASK; 3076 argop[1].nfs_argop4_u.opgetattr.mi = mi; 3077 3078 /* SETCLIENTID */ 3079 argop[2].argop = OP_SETCLIENTID; 3080 3081 s_args = &argop[2].nfs_argop4_u.opsetclientid; 3082 3083 mutex_enter(&np->s_lock); 3084 3085 s_args->client.verifier = np->clidtosend.verifier; 3086 s_args->client.id_len = np->clidtosend.id_len; 3087 ASSERT(s_args->client.id_len <= NFS4_OPAQUE_LIMIT); 3088 s_args->client.id_val = np->clidtosend.id_val; 3089 3090 /* 3091 * Callback needs to happen on non-RDMA transport 3092 * Check if we have saved the original knetconfig 3093 * if so, use that instead. 3094 */ 3095 if (svp->sv_origknconf != NULL) 3096 nfs4_cb_args(np, svp->sv_origknconf, s_args); 3097 else 3098 nfs4_cb_args(np, svp->sv_knconf, s_args); 3099 3100 mutex_exit(&np->s_lock); 3101 3102 rfs4call(mi, &args, &res, cr, &doqueue, 0, ep); 3103 3104 if (ep->error) 3105 return; 3106 3107 /* getattr lease_time res */ 3108 if (res.array_len >= 2) { 3109 garp = &res.array[1].nfs_resop4_u.opgetattr.ga_res; 3110 3111 #ifndef _LP64 3112 /* 3113 * The 32 bit client cannot handle a lease time greater than 3114 * (INT32_MAX/1000000). This is due to the use of the 3115 * lease_time in calls to drv_usectohz() in 3116 * nfs4_renew_lease_thread(). The problem is that 3117 * drv_usectohz() takes a time_t (which is just a long = 4 3118 * bytes) as its parameter. The lease_time is multiplied by 3119 * 1000000 to convert seconds to usecs for the parameter. If 3120 * a number bigger than (INT32_MAX/1000000) is used then we 3121 * overflow on the 32bit client. 3122 */ 3123 if (garp->n4g_ext_res->n4g_leasetime > (INT32_MAX/1000000)) { 3124 garp->n4g_ext_res->n4g_leasetime = INT32_MAX/1000000; 3125 } 3126 #endif 3127 3128 mutex_enter(&np->s_lock); 3129 np->s_lease_time = garp->n4g_ext_res->n4g_leasetime; 3130 3131 /* 3132 * Keep track of the lease period for the mi's 3133 * mi_msg_list. We need an appropiate time 3134 * bound to associate past facts with a current 3135 * event. The lease period is perfect for this. 3136 */ 3137 mutex_enter(&mi->mi_msg_list_lock); 3138 mi->mi_lease_period = np->s_lease_time; 3139 mutex_exit(&mi->mi_msg_list_lock); 3140 mutex_exit(&np->s_lock); 3141 } 3142 3143 3144 if (res.status == NFS4ERR_CLID_INUSE) { 3145 clientaddr4 *clid_inuse; 3146 3147 if (!(*retry_inusep)) { 3148 clid_inuse = &res.array->nfs_resop4_u. 3149 opsetclientid.SETCLIENTID4res_u.client_using; 3150 3151 zcmn_err(mi->mi_zone->zone_id, CE_NOTE, 3152 "NFS4 mount (SETCLIENTID failed)." 3153 " nfs4_client_id.id is in" 3154 "use already by: r_netid<%s> r_addr<%s>", 3155 clid_inuse->r_netid, clid_inuse->r_addr); 3156 } 3157 3158 /* 3159 * XXX - The client should be more robust in its 3160 * handling of clientid in use errors (regen another 3161 * clientid and try again?) 3162 */ 3163 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 3164 return; 3165 } 3166 3167 if (res.status) { 3168 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 3169 return; 3170 } 3171 3172 s_resok = &res.array[2].nfs_resop4_u. 3173 opsetclientid.SETCLIENTID4res_u.resok4; 3174 3175 tmp_clientid = s_resok->clientid; 3176 3177 verf = s_resok->setclientid_confirm; 3178 3179 #ifdef DEBUG 3180 if (nfs4setclientid_otw_debug) { 3181 union { 3182 clientid4 clientid; 3183 int foo[2]; 3184 } cid; 3185 3186 cid.clientid = s_resok->clientid; 3187 3188 zcmn_err(mi->mi_zone->zone_id, CE_NOTE, 3189 "nfs4setclientid_otw: OK, clientid = %x,%x, " 3190 "verifier = %" PRIx64 "\n", cid.foo[0], cid.foo[1], verf); 3191 } 3192 #endif 3193 3194 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 3195 3196 /* Confirm the client id and get the lease_time attribute */ 3197 3198 args.ctag = TAG_SETCLIENTID_CF; 3199 3200 args.array = argop; 3201 args.array_len = 1; 3202 3203 argop[0].argop = OP_SETCLIENTID_CONFIRM; 3204 3205 argop[0].nfs_argop4_u.opsetclientid_confirm.clientid = tmp_clientid; 3206 argop[0].nfs_argop4_u.opsetclientid_confirm.setclientid_confirm = verf; 3207 3208 /* used to figure out RTT for np */ 3209 gethrestime(&prop_time); 3210 3211 NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4setlientid_otw: " 3212 "start time: %ld sec %ld nsec", prop_time.tv_sec, 3213 prop_time.tv_nsec)); 3214 3215 rfs4call(mi, &args, &res, cr, &doqueue, 0, ep); 3216 3217 gethrestime(&after_time); 3218 mutex_enter(&np->s_lock); 3219 np->propagation_delay.tv_sec = 3220 MAX(1, after_time.tv_sec - prop_time.tv_sec); 3221 mutex_exit(&np->s_lock); 3222 3223 NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4setlcientid_otw: " 3224 "finish time: %ld sec ", after_time.tv_sec)); 3225 3226 NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4setclientid_otw: " 3227 "propagation delay set to %ld sec", 3228 np->propagation_delay.tv_sec)); 3229 3230 if (ep->error) 3231 return; 3232 3233 if (res.status == NFS4ERR_CLID_INUSE) { 3234 clientaddr4 *clid_inuse; 3235 3236 if (!(*retry_inusep)) { 3237 clid_inuse = &res.array->nfs_resop4_u. 3238 opsetclientid.SETCLIENTID4res_u.client_using; 3239 3240 zcmn_err(mi->mi_zone->zone_id, CE_NOTE, 3241 "SETCLIENTID_CONFIRM failed. " 3242 "nfs4_client_id.id is in use already by: " 3243 "r_netid<%s> r_addr<%s>", 3244 clid_inuse->r_netid, clid_inuse->r_addr); 3245 } 3246 3247 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 3248 return; 3249 } 3250 3251 if (res.status) { 3252 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 3253 return; 3254 } 3255 3256 mutex_enter(&np->s_lock); 3257 np->clientid = tmp_clientid; 3258 np->s_flags |= N4S_CLIENTID_SET; 3259 3260 /* Add mi to np's mntinfo4 list */ 3261 nfs4_add_mi_to_server(np, mi); 3262 3263 if (np->lease_valid == NFS4_LEASE_NOT_STARTED) { 3264 /* 3265 * Start lease management thread. 3266 * Keep trying until we succeed. 3267 */ 3268 3269 np->s_refcnt++; /* pass reference to thread */ 3270 (void) zthread_create(NULL, 0, nfs4_renew_lease_thread, np, 0, 3271 minclsyspri); 3272 } 3273 mutex_exit(&np->s_lock); 3274 3275 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res); 3276 } 3277 3278 /* 3279 * Add mi to sp's mntinfo4_list if it isn't already in the list. Makes 3280 * mi's clientid the same as sp's. 3281 * Assumes sp is locked down. 3282 */ 3283 void 3284 nfs4_add_mi_to_server(nfs4_server_t *sp, mntinfo4_t *mi) 3285 { 3286 mntinfo4_t *tmi; 3287 int in_list = 0; 3288 3289 ASSERT(nfs_rw_lock_held(&mi->mi_recovlock, RW_READER) || 3290 nfs_rw_lock_held(&mi->mi_recovlock, RW_WRITER)); 3291 ASSERT(sp != &nfs4_server_lst); 3292 ASSERT(MUTEX_HELD(&sp->s_lock)); 3293 3294 NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3295 "nfs4_add_mi_to_server: add mi %p to sp %p", 3296 (void*)mi, (void*)sp)); 3297 3298 for (tmi = sp->mntinfo4_list; 3299 tmi != NULL; 3300 tmi = tmi->mi_clientid_next) { 3301 if (tmi == mi) { 3302 NFS4_DEBUG(nfs4_client_lease_debug, 3303 (CE_NOTE, 3304 "nfs4_add_mi_to_server: mi in list")); 3305 in_list = 1; 3306 } 3307 } 3308 3309 /* 3310 * First put a hold on the mntinfo4's vfsp so that references via 3311 * mntinfo4_list will be valid. 3312 */ 3313 if (!in_list) 3314 VFS_HOLD(mi->mi_vfsp); 3315 3316 NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4_add_mi_to_server: " 3317 "hold vfs %p for mi: %p", (void*)mi->mi_vfsp, (void*)mi)); 3318 3319 if (!in_list) { 3320 if (sp->mntinfo4_list) 3321 sp->mntinfo4_list->mi_clientid_prev = mi; 3322 mi->mi_clientid_next = sp->mntinfo4_list; 3323 sp->mntinfo4_list = mi; 3324 mi->mi_srvsettime = gethrestime_sec(); 3325 } 3326 3327 /* set mi's clientid to that of sp's for later matching */ 3328 mi->mi_clientid = sp->clientid; 3329 3330 /* 3331 * Update the clientid for any other mi's belonging to sp. This 3332 * must be done here while we hold sp->s_lock, so that 3333 * find_nfs4_server() continues to work. 3334 */ 3335 3336 for (tmi = sp->mntinfo4_list; 3337 tmi != NULL; 3338 tmi = tmi->mi_clientid_next) { 3339 if (tmi != mi) { 3340 tmi->mi_clientid = sp->clientid; 3341 } 3342 } 3343 } 3344 3345 /* 3346 * Remove the mi from sp's mntinfo4_list and release its reference. 3347 * Exception: if mi still has open files, flag it for later removal (when 3348 * all the files are closed). 3349 * 3350 * If this is the last mntinfo4 in sp's list then tell the lease renewal 3351 * thread to exit. 3352 */ 3353 static void 3354 nfs4_remove_mi_from_server_nolock(mntinfo4_t *mi, nfs4_server_t *sp) 3355 { 3356 NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3357 "nfs4_remove_mi_from_server_nolock: remove mi %p from sp %p", 3358 (void*)mi, (void*)sp)); 3359 3360 ASSERT(sp != NULL); 3361 ASSERT(MUTEX_HELD(&sp->s_lock)); 3362 ASSERT(mi->mi_open_files >= 0); 3363 3364 /* 3365 * First make sure this mntinfo4 can be taken off of the list, 3366 * ie: it doesn't have any open files remaining. 3367 */ 3368 if (mi->mi_open_files > 0) { 3369 NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3370 "nfs4_remove_mi_from_server_nolock: don't " 3371 "remove mi since it still has files open")); 3372 3373 mutex_enter(&mi->mi_lock); 3374 mi->mi_flags |= MI4_REMOVE_ON_LAST_CLOSE; 3375 mutex_exit(&mi->mi_lock); 3376 return; 3377 } 3378 3379 VFS_HOLD(mi->mi_vfsp); 3380 remove_mi(sp, mi); 3381 VFS_RELE(mi->mi_vfsp); 3382 3383 if (sp->mntinfo4_list == NULL) { 3384 /* last fs unmounted, kill the thread */ 3385 NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, 3386 "remove_mi_from_nfs4_server_nolock: kill the thread")); 3387 nfs4_mark_srv_dead(sp); 3388 } 3389 } 3390 3391 /* 3392 * Remove mi from sp's mntinfo4_list and release the vfs reference. 3393 */ 3394 static void 3395 remove_mi(nfs4_server_t *sp, mntinfo4_t *mi) 3396 { 3397 ASSERT(MUTEX_HELD(&sp->s_lock)); 3398 3399 /* 3400 * We release a reference, and the caller must still have a 3401 * reference. 3402 */ 3403 ASSERT(mi->mi_vfsp->vfs_count >= 2); 3404 3405 if (mi->mi_clientid_prev) { 3406 mi->mi_clientid_prev->mi_clientid_next = mi->mi_clientid_next; 3407 } else { 3408 /* This is the first mi in sp's mntinfo4_list */ 3409 /* 3410 * Make sure the first mntinfo4 in the list is the actual 3411 * mntinfo4 passed in. 3412 */ 3413 ASSERT(sp->mntinfo4_list == mi); 3414 3415 sp->mntinfo4_list = mi->mi_clientid_next; 3416 } 3417 if (mi->mi_clientid_next) 3418 mi->mi_clientid_next->mi_clientid_prev = mi->mi_clientid_prev; 3419 3420 /* Now mark the mntinfo4's links as being removed */ 3421 mi->mi_clientid_prev = mi->mi_clientid_next = NULL; 3422 3423 VFS_RELE(mi->mi_vfsp); 3424 } 3425 3426 /* 3427 * Free all the entries in sp's mntinfo4_list. 3428 */ 3429 static void 3430 remove_all_mi(nfs4_server_t *sp) 3431 { 3432 mntinfo4_t *mi; 3433 3434 ASSERT(MUTEX_HELD(&sp->s_lock)); 3435 3436 while (sp->mntinfo4_list != NULL) { 3437 mi = sp->mntinfo4_list; 3438 /* 3439 * Grab a reference in case there is only one left (which 3440 * remove_mi() frees). 3441 */ 3442 VFS_HOLD(mi->mi_vfsp); 3443 remove_mi(sp, mi); 3444 VFS_RELE(mi->mi_vfsp); 3445 } 3446 } 3447 3448 /* 3449 * Remove the mi from sp's mntinfo4_list as above, and rele the vfs. 3450 * 3451 * This version can be called with a null nfs4_server_t arg, 3452 * and will either find the right one and handle locking, or 3453 * do nothing because the mi wasn't added to an sp's mntinfo4_list. 3454 */ 3455 void 3456 nfs4_remove_mi_from_server(mntinfo4_t *mi, nfs4_server_t *esp) 3457 { 3458 nfs4_server_t *sp; 3459 3460 if (esp == NULL) { 3461 (void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, 0); 3462 sp = find_nfs4_server_all(mi, 1); 3463 } else 3464 sp = esp; 3465 3466 if (sp != NULL) 3467 nfs4_remove_mi_from_server_nolock(mi, sp); 3468 3469 /* 3470 * If we had a valid esp as input, the calling function will be 3471 * responsible for unlocking the esp nfs4_server. 3472 */ 3473 if (esp == NULL) { 3474 if (sp != NULL) 3475 mutex_exit(&sp->s_lock); 3476 nfs_rw_exit(&mi->mi_recovlock); 3477 if (sp != NULL) 3478 nfs4_server_rele(sp); 3479 } 3480 } 3481 3482 /* 3483 * Return TRUE if the given server has any non-unmounted filesystems. 3484 */ 3485 3486 bool_t 3487 nfs4_fs_active(nfs4_server_t *sp) 3488 { 3489 mntinfo4_t *mi; 3490 3491 ASSERT(MUTEX_HELD(&sp->s_lock)); 3492 3493 for (mi = sp->mntinfo4_list; mi != NULL; mi = mi->mi_clientid_next) { 3494 if (!(mi->mi_vfsp->vfs_flag & VFS_UNMOUNTED)) 3495 return (TRUE); 3496 } 3497 3498 return (FALSE); 3499 } 3500 3501 /* 3502 * Mark sp as finished and notify any waiters. 3503 */ 3504 3505 void 3506 nfs4_mark_srv_dead(nfs4_server_t *sp) 3507 { 3508 ASSERT(MUTEX_HELD(&sp->s_lock)); 3509 3510 sp->s_thread_exit = NFS4_THREAD_EXIT; 3511 cv_broadcast(&sp->cv_thread_exit); 3512 } 3513 3514 /* 3515 * Create a new nfs4_server_t structure. 3516 * Returns new node unlocked and not in list, but with a reference count of 3517 * 1. 3518 */ 3519 struct nfs4_server * 3520 new_nfs4_server(struct servinfo4 *svp, cred_t *cr) 3521 { 3522 struct nfs4_server *np; 3523 timespec_t tt; 3524 union { 3525 struct { 3526 uint32_t sec; 3527 uint32_t subsec; 3528 } un_curtime; 3529 verifier4 un_verifier; 3530 } nfs4clientid_verifier; 3531 char id_val[] = "Solaris: %s, NFSv4 kernel client"; 3532 int len; 3533 3534 np = kmem_zalloc(sizeof (struct nfs4_server), KM_SLEEP); 3535 np->saddr.len = svp->sv_addr.len; 3536 np->saddr.maxlen = svp->sv_addr.maxlen; 3537 np->saddr.buf = kmem_alloc(svp->sv_addr.maxlen, KM_SLEEP); 3538 bcopy(svp->sv_addr.buf, np->saddr.buf, svp->sv_addr.len); 3539 np->s_refcnt = 1; 3540 3541 /* 3542 * Build the nfs_client_id4 for this server mount. Ensure 3543 * the verifier is useful and that the identification is 3544 * somehow based on the server's address for the case of 3545 * multi-homed servers. 3546 */ 3547 nfs4clientid_verifier.un_verifier = 0; 3548 gethrestime(&tt); 3549 nfs4clientid_verifier.un_curtime.sec = (uint32_t)tt.tv_sec; 3550 nfs4clientid_verifier.un_curtime.subsec = (uint32_t)tt.tv_nsec; 3551 np->clidtosend.verifier = nfs4clientid_verifier.un_verifier; 3552 3553 /* 3554 * calculate the length of the opaque identifier. Subtract 2 3555 * for the "%s" and add the traditional +1 for null 3556 * termination. 3557 */ 3558 len = strlen(id_val) - 2 + strlen(uts_nodename()) + 1; 3559 np->clidtosend.id_len = len + np->saddr.maxlen; 3560 3561 np->clidtosend.id_val = kmem_alloc(np->clidtosend.id_len, KM_SLEEP); 3562 (void) sprintf(np->clidtosend.id_val, id_val, uts_nodename()); 3563 bcopy(np->saddr.buf, &np->clidtosend.id_val[len], np->saddr.len); 3564 3565 np->s_flags = 0; 3566 np->mntinfo4_list = NULL; 3567 /* save cred for issuing rfs4calls inside the renew thread */ 3568 crhold(cr); 3569 np->s_cred = cr; 3570 cv_init(&np->cv_thread_exit, NULL, CV_DEFAULT, NULL); 3571 mutex_init(&np->s_lock, NULL, MUTEX_DEFAULT, NULL); 3572 nfs_rw_init(&np->s_recovlock, NULL, RW_DEFAULT, NULL); 3573 list_create(&np->s_deleg_list, sizeof (rnode4_t), 3574 offsetof(rnode4_t, r_deleg_link)); 3575 np->s_thread_exit = 0; 3576 np->state_ref_count = 0; 3577 np->lease_valid = NFS4_LEASE_NOT_STARTED; 3578 cv_init(&np->s_cv_otw_count, NULL, CV_DEFAULT, NULL); 3579 cv_init(&np->s_clientid_pend, NULL, CV_DEFAULT, NULL); 3580 np->s_otw_call_count = 0; 3581 cv_init(&np->wait_cb_null, NULL, CV_DEFAULT, NULL); 3582 np->zoneid = getzoneid(); 3583 np->zone_globals = nfs4_get_callback_globals(); 3584 ASSERT(np->zone_globals != NULL); 3585 return (np); 3586 } 3587 3588 /* 3589 * Create a new nfs4_server_t structure and add it to the list. 3590 * Returns new node locked; reference must eventually be freed. 3591 */ 3592 static struct nfs4_server * 3593 add_new_nfs4_server(struct servinfo4 *svp, cred_t *cr) 3594 { 3595 nfs4_server_t *sp; 3596 3597 ASSERT(MUTEX_HELD(&nfs4_server_lst_lock)); 3598 sp = new_nfs4_server(svp, cr); 3599 mutex_enter(&sp->s_lock); 3600 insque(sp, &nfs4_server_lst); 3601 sp->s_refcnt++; /* list gets a reference */ 3602 sp->s_flags |= N4S_INSERTED; 3603 sp->clientid = 0; 3604 return (sp); 3605 } 3606 3607 int nfs4_server_t_debug = 0; 3608 3609 #ifdef lint 3610 extern void 3611 dumpnfs4slist(char *, mntinfo4_t *, clientid4, servinfo4_t *); 3612 #endif 3613 3614 #ifndef lint 3615 #ifdef DEBUG 3616 void 3617 dumpnfs4slist(char *txt, mntinfo4_t *mi, clientid4 clientid, servinfo4_t *srv_p) 3618 { 3619 int hash16(void *p, int len); 3620 nfs4_server_t *np; 3621 3622 NFS4_DEBUG(nfs4_server_t_debug, (CE_NOTE, 3623 "dumping nfs4_server_t list in %s", txt)); 3624 NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT, 3625 "mi 0x%p, want clientid %llx, addr %d/%04X", 3626 mi, (longlong_t)clientid, srv_p->sv_addr.len, 3627 hash16((void *)srv_p->sv_addr.buf, srv_p->sv_addr.len))); 3628 for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; 3629 np = np->forw) { 3630 NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT, 3631 "node 0x%p, clientid %llx, addr %d/%04X, cnt %d", 3632 np, (longlong_t)np->clientid, np->saddr.len, 3633 hash16((void *)np->saddr.buf, np->saddr.len), 3634 np->state_ref_count)); 3635 if (np->saddr.len == srv_p->sv_addr.len && 3636 bcmp(np->saddr.buf, srv_p->sv_addr.buf, 3637 np->saddr.len) == 0) 3638 NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT, 3639 " - address matches")); 3640 if (np->clientid == clientid || np->clientid == 0) 3641 NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT, 3642 " - clientid matches")); 3643 if (np->s_thread_exit != NFS4_THREAD_EXIT) 3644 NFS4_DEBUG(nfs4_server_t_debug, (CE_CONT, 3645 " - thread not exiting")); 3646 } 3647 delay(hz); 3648 } 3649 #endif 3650 #endif 3651 3652 3653 /* 3654 * Move a mntinfo4_t from one server list to another. 3655 * Locking of the two nfs4_server_t nodes will be done in list order. 3656 * 3657 * Returns NULL if the current nfs4_server_t for the filesystem could not 3658 * be found (e.g., due to forced unmount). Otherwise returns a reference 3659 * to the new nfs4_server_t, which must eventually be freed. 3660 */ 3661 nfs4_server_t * 3662 nfs4_move_mi(mntinfo4_t *mi, servinfo4_t *old, servinfo4_t *new) 3663 { 3664 nfs4_server_t *p, *op = NULL, *np = NULL; 3665 int num_open; 3666 zoneid_t zoneid = nfs_zoneid(); 3667 3668 ASSERT(nfs_zone() == mi->mi_zone); 3669 3670 mutex_enter(&nfs4_server_lst_lock); 3671 #ifdef DEBUG 3672 if (nfs4_server_t_debug) 3673 dumpnfs4slist("nfs4_move_mi", mi, (clientid4)0, new); 3674 #endif 3675 for (p = nfs4_server_lst.forw; p != &nfs4_server_lst; p = p->forw) { 3676 if (p->zoneid != zoneid) 3677 continue; 3678 if (p->saddr.len == old->sv_addr.len && 3679 bcmp(p->saddr.buf, old->sv_addr.buf, p->saddr.len) == 0 && 3680 p->s_thread_exit != NFS4_THREAD_EXIT) { 3681 op = p; 3682 mutex_enter(&op->s_lock); 3683 op->s_refcnt++; 3684 } 3685 if (p->saddr.len == new->sv_addr.len && 3686 bcmp(p->saddr.buf, new->sv_addr.buf, p->saddr.len) == 0 && 3687 p->s_thread_exit != NFS4_THREAD_EXIT) { 3688 np = p; 3689 mutex_enter(&np->s_lock); 3690 } 3691 if (op != NULL && np != NULL) 3692 break; 3693 } 3694 if (op == NULL) { 3695 /* 3696 * Filesystem has been forcibly unmounted. Bail out. 3697 */ 3698 if (np != NULL) 3699 mutex_exit(&np->s_lock); 3700 mutex_exit(&nfs4_server_lst_lock); 3701 return (NULL); 3702 } 3703 if (np != NULL) { 3704 np->s_refcnt++; 3705 } else { 3706 #ifdef DEBUG 3707 NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE, 3708 "nfs4_move_mi: no target nfs4_server, will create.")); 3709 #endif 3710 np = add_new_nfs4_server(new, kcred); 3711 } 3712 mutex_exit(&nfs4_server_lst_lock); 3713 3714 NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE, 3715 "nfs4_move_mi: for mi 0x%p, " 3716 "old servinfo4 0x%p, new servinfo4 0x%p, " 3717 "old nfs4_server 0x%p, new nfs4_server 0x%p, ", 3718 (void*)mi, (void*)old, (void*)new, 3719 (void*)op, (void*)np)); 3720 ASSERT(op != NULL && np != NULL); 3721 3722 /* discard any delegations */ 3723 nfs4_deleg_discard(mi, op); 3724 3725 num_open = mi->mi_open_files; 3726 mi->mi_open_files = 0; 3727 op->state_ref_count -= num_open; 3728 ASSERT(op->state_ref_count >= 0); 3729 np->state_ref_count += num_open; 3730 nfs4_remove_mi_from_server_nolock(mi, op); 3731 mi->mi_open_files = num_open; 3732 NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE, 3733 "nfs4_move_mi: mi_open_files %d, op->cnt %d, np->cnt %d", 3734 mi->mi_open_files, op->state_ref_count, np->state_ref_count)); 3735 3736 nfs4_add_mi_to_server(np, mi); 3737 3738 mutex_exit(&op->s_lock); 3739 nfs4_server_rele(op); 3740 mutex_exit(&np->s_lock); 3741 3742 return (np); 3743 } 3744 3745 /* 3746 * Need to have the nfs4_server_lst_lock. 3747 * Search the nfs4_server list to find a match on this servinfo4 3748 * based on its address. 3749 * 3750 * Returns NULL if no match is found. Otherwise returns a reference (which 3751 * must eventually be freed) to a locked nfs4_server. 3752 */ 3753 nfs4_server_t * 3754 servinfo4_to_nfs4_server(servinfo4_t *srv_p) 3755 { 3756 nfs4_server_t *np; 3757 zoneid_t zoneid = nfs_zoneid(); 3758 3759 ASSERT(MUTEX_HELD(&nfs4_server_lst_lock)); 3760 for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; np = np->forw) { 3761 if (np->zoneid == zoneid && 3762 np->saddr.len == srv_p->sv_addr.len && 3763 bcmp(np->saddr.buf, srv_p->sv_addr.buf, 3764 np->saddr.len) == 0 && 3765 np->s_thread_exit != NFS4_THREAD_EXIT) { 3766 mutex_enter(&np->s_lock); 3767 np->s_refcnt++; 3768 return (np); 3769 } 3770 } 3771 return (NULL); 3772 } 3773 3774 /* 3775 * Search the nfs4_server_lst to find a match based on clientid and 3776 * addr. 3777 * Locks the nfs4_server down if it is found and returns a reference that 3778 * must eventually be freed. 3779 * 3780 * Returns NULL it no match is found. This means one of two things: either 3781 * mi is in the process of being mounted, or mi has been unmounted. 3782 * 3783 * The caller should be holding mi->mi_recovlock, and it should continue to 3784 * hold the lock until done with the returned nfs4_server_t. Once 3785 * mi->mi_recovlock is released, there is no guarantee that the returned 3786 * mi->nfs4_server_t will continue to correspond to mi. 3787 */ 3788 nfs4_server_t * 3789 find_nfs4_server(mntinfo4_t *mi) 3790 { 3791 return (find_nfs4_server_all(mi, 0)); 3792 } 3793 3794 /* 3795 * Same as above, but takes an "all" parameter which can be 3796 * set to 1 if the caller wishes to find nfs4_server_t's which 3797 * have been marked for termination by the exit of the renew 3798 * thread. This should only be used by operations which are 3799 * cleaning up and will not cause an OTW op. 3800 */ 3801 nfs4_server_t * 3802 find_nfs4_server_all(mntinfo4_t *mi, int all) 3803 { 3804 nfs4_server_t *np; 3805 servinfo4_t *svp; 3806 zoneid_t zoneid = mi->mi_zone->zone_id; 3807 3808 ASSERT(nfs_rw_lock_held(&mi->mi_recovlock, RW_READER) || 3809 nfs_rw_lock_held(&mi->mi_recovlock, RW_WRITER)); 3810 /* 3811 * This can be called from nfs4_unmount() which can be called from the 3812 * global zone, hence it's legal for the global zone to muck with 3813 * another zone's server list, as long as it doesn't try to contact 3814 * them. 3815 */ 3816 ASSERT(zoneid == getzoneid() || getzoneid() == GLOBAL_ZONEID || 3817 nfs_global_client_only != 0); 3818 3819 /* 3820 * The nfs4_server_lst_lock global lock is held when we get a new 3821 * clientid (via SETCLIENTID OTW). Holding this global lock and 3822 * mi_recovlock (READER is fine) ensures that the nfs4_server 3823 * and this mntinfo4 can't get out of sync, so the following search is 3824 * always valid. 3825 */ 3826 mutex_enter(&nfs4_server_lst_lock); 3827 #ifdef DEBUG 3828 if (nfs4_server_t_debug) { 3829 /* mi->mi_clientid is unprotected, ok for debug output */ 3830 dumpnfs4slist("find_nfs4_server", mi, mi->mi_clientid, 3831 mi->mi_curr_serv); 3832 } 3833 #endif 3834 for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; np = np->forw) { 3835 mutex_enter(&np->s_lock); 3836 svp = mi->mi_curr_serv; 3837 3838 if (np->zoneid == zoneid && 3839 np->clientid == mi->mi_clientid && 3840 np->saddr.len == svp->sv_addr.len && 3841 bcmp(np->saddr.buf, svp->sv_addr.buf, np->saddr.len) == 0 && 3842 (np->s_thread_exit != NFS4_THREAD_EXIT || all != 0)) { 3843 mutex_exit(&nfs4_server_lst_lock); 3844 np->s_refcnt++; 3845 return (np); 3846 } 3847 mutex_exit(&np->s_lock); 3848 } 3849 mutex_exit(&nfs4_server_lst_lock); 3850 3851 return (NULL); 3852 } 3853 3854 /* 3855 * Release the reference to sp and destroy it if that's the last one. 3856 */ 3857 3858 void 3859 nfs4_server_rele(nfs4_server_t *sp) 3860 { 3861 mutex_enter(&sp->s_lock); 3862 ASSERT(sp->s_refcnt > 0); 3863 sp->s_refcnt--; 3864 if (sp->s_refcnt > 0) { 3865 mutex_exit(&sp->s_lock); 3866 return; 3867 } 3868 mutex_exit(&sp->s_lock); 3869 3870 mutex_enter(&nfs4_server_lst_lock); 3871 mutex_enter(&sp->s_lock); 3872 if (sp->s_refcnt > 0) { 3873 mutex_exit(&sp->s_lock); 3874 mutex_exit(&nfs4_server_lst_lock); 3875 return; 3876 } 3877 remque(sp); 3878 sp->forw = sp->back = NULL; 3879 mutex_exit(&nfs4_server_lst_lock); 3880 destroy_nfs4_server(sp); 3881 } 3882 3883 static void 3884 destroy_nfs4_server(nfs4_server_t *sp) 3885 { 3886 ASSERT(MUTEX_HELD(&sp->s_lock)); 3887 ASSERT(sp->s_refcnt == 0); 3888 ASSERT(sp->s_otw_call_count == 0); 3889 3890 remove_all_mi(sp); 3891 3892 crfree(sp->s_cred); 3893 kmem_free(sp->saddr.buf, sp->saddr.maxlen); 3894 kmem_free(sp->clidtosend.id_val, sp->clidtosend.id_len); 3895 mutex_exit(&sp->s_lock); 3896 3897 /* destroy the nfs4_server */ 3898 nfs4callback_destroy(sp); 3899 list_destroy(&sp->s_deleg_list); 3900 mutex_destroy(&sp->s_lock); 3901 cv_destroy(&sp->cv_thread_exit); 3902 cv_destroy(&sp->s_cv_otw_count); 3903 cv_destroy(&sp->s_clientid_pend); 3904 cv_destroy(&sp->wait_cb_null); 3905 nfs_rw_destroy(&sp->s_recovlock); 3906 kmem_free(sp, sizeof (*sp)); 3907 } 3908 3909 /* 3910 * Lock sp, but only if it's still active (in the list and hasn't been 3911 * flagged as exiting) or 'all' is non-zero. 3912 * Returns TRUE if sp got locked and adds a reference to sp. 3913 */ 3914 bool_t 3915 nfs4_server_vlock(nfs4_server_t *sp, int all) 3916 { 3917 nfs4_server_t *np; 3918 3919 mutex_enter(&nfs4_server_lst_lock); 3920 for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; np = np->forw) { 3921 if (sp == np && (np->s_thread_exit != NFS4_THREAD_EXIT || 3922 all != 0)) { 3923 mutex_enter(&np->s_lock); 3924 np->s_refcnt++; 3925 mutex_exit(&nfs4_server_lst_lock); 3926 return (TRUE); 3927 } 3928 } 3929 mutex_exit(&nfs4_server_lst_lock); 3930 return (FALSE); 3931 } 3932 3933 /* 3934 * Fork off a thread to free the data structures for a mount. 3935 */ 3936 3937 static void 3938 async_free_mount(vfs_t *vfsp, int flag, cred_t *cr) 3939 { 3940 freemountargs_t *args; 3941 args = kmem_alloc(sizeof (freemountargs_t), KM_SLEEP); 3942 args->fm_vfsp = vfsp; 3943 VFS_HOLD(vfsp); 3944 MI4_HOLD(VFTOMI4(vfsp)); 3945 args->fm_flag = flag; 3946 args->fm_cr = cr; 3947 crhold(cr); 3948 (void) zthread_create(NULL, 0, nfs4_free_mount_thread, args, 0, 3949 minclsyspri); 3950 } 3951 3952 static void 3953 nfs4_free_mount_thread(freemountargs_t *args) 3954 { 3955 mntinfo4_t *mi; 3956 nfs4_free_mount(args->fm_vfsp, args->fm_flag, args->fm_cr); 3957 mi = VFTOMI4(args->fm_vfsp); 3958 crfree(args->fm_cr); 3959 VFS_RELE(args->fm_vfsp); 3960 MI4_RELE(mi); 3961 kmem_free(args, sizeof (freemountargs_t)); 3962 zthread_exit(); 3963 /* NOTREACHED */ 3964 } 3965 3966 /* 3967 * Thread to free the data structures for a given filesystem. 3968 */ 3969 static void 3970 nfs4_free_mount(vfs_t *vfsp, int flag, cred_t *cr) 3971 { 3972 mntinfo4_t *mi = VFTOMI4(vfsp); 3973 nfs4_server_t *sp; 3974 callb_cpr_t cpr_info; 3975 kmutex_t cpr_lock; 3976 boolean_t async_thread; 3977 int removed; 3978 3979 bool_t must_unlock; 3980 nfs4_ephemeral_tree_t *eph_tree; 3981 3982 /* 3983 * We need to participate in the CPR framework if this is a kernel 3984 * thread. 3985 */ 3986 async_thread = (curproc == nfs_zone()->zone_zsched); 3987 if (async_thread) { 3988 mutex_init(&cpr_lock, NULL, MUTEX_DEFAULT, NULL); 3989 CALLB_CPR_INIT(&cpr_info, &cpr_lock, callb_generic_cpr, 3990 "nfsv4AsyncUnmount"); 3991 } 3992 3993 /* 3994 * We need to wait for all outstanding OTW calls 3995 * and recovery to finish before we remove the mi 3996 * from the nfs4_server_t, as current pending 3997 * calls might still need this linkage (in order 3998 * to find a nfs4_server_t from a mntinfo4_t). 3999 */ 4000 (void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER, FALSE); 4001 sp = find_nfs4_server(mi); 4002 nfs_rw_exit(&mi->mi_recovlock); 4003 4004 if (sp) { 4005 while (sp->s_otw_call_count != 0) { 4006 if (async_thread) { 4007 mutex_enter(&cpr_lock); 4008 CALLB_CPR_SAFE_BEGIN(&cpr_info); 4009 mutex_exit(&cpr_lock); 4010 } 4011 cv_wait(&sp->s_cv_otw_count, &sp->s_lock); 4012 if (async_thread) { 4013 mutex_enter(&cpr_lock); 4014 CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock); 4015 mutex_exit(&cpr_lock); 4016 } 4017 } 4018 mutex_exit(&sp->s_lock); 4019 nfs4_server_rele(sp); 4020 sp = NULL; 4021 } 4022 4023 mutex_enter(&mi->mi_lock); 4024 while (mi->mi_in_recovery != 0) { 4025 if (async_thread) { 4026 mutex_enter(&cpr_lock); 4027 CALLB_CPR_SAFE_BEGIN(&cpr_info); 4028 mutex_exit(&cpr_lock); 4029 } 4030 cv_wait(&mi->mi_cv_in_recov, &mi->mi_lock); 4031 if (async_thread) { 4032 mutex_enter(&cpr_lock); 4033 CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock); 4034 mutex_exit(&cpr_lock); 4035 } 4036 } 4037 mutex_exit(&mi->mi_lock); 4038 4039 /* 4040 * If we got an error, then do not nuke the 4041 * tree. Either the harvester is busy reclaiming 4042 * this node or we ran into some busy condition. 4043 * 4044 * The harvester will eventually come along and cleanup. 4045 * The only problem would be the root mount point. 4046 * 4047 * Since the busy node can occur for a variety 4048 * of reasons and can result in an entry staying 4049 * in df output but no longer accessible from the 4050 * directory tree, we are okay. 4051 */ 4052 if (!nfs4_ephemeral_umount(mi, flag, cr, 4053 &must_unlock, &eph_tree)) 4054 nfs4_ephemeral_umount_activate(mi, &must_unlock, 4055 &eph_tree); 4056 4057 /* 4058 * The original purge of the dnlc via 'dounmount' 4059 * doesn't guarantee that another dnlc entry was not 4060 * added while we waitied for all outstanding OTW 4061 * and recovery calls to finish. So re-purge the 4062 * dnlc now. 4063 */ 4064 (void) dnlc_purge_vfsp(vfsp, 0); 4065 4066 /* 4067 * We need to explicitly stop the manager thread; the asyc worker 4068 * threads can timeout and exit on their own. 4069 */ 4070 mutex_enter(&mi->mi_async_lock); 4071 mi->mi_max_threads = 0; 4072 cv_broadcast(&mi->mi_async_work_cv); 4073 mutex_exit(&mi->mi_async_lock); 4074 if (mi->mi_manager_thread) 4075 nfs4_async_manager_stop(vfsp); 4076 4077 destroy_rtable4(vfsp, cr); 4078 4079 nfs4_remove_mi_from_server(mi, NULL); 4080 4081 if (async_thread) { 4082 mutex_enter(&cpr_lock); 4083 CALLB_CPR_EXIT(&cpr_info); /* drops cpr_lock */ 4084 mutex_destroy(&cpr_lock); 4085 } 4086 4087 removed = nfs4_mi_zonelist_remove(mi); 4088 if (removed) 4089 zone_rele(mi->mi_zone); 4090 } 4091