nfs_clvfsops.c revision 221436
1/*- 2 * Copyright (c) 1989, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from nfs_vfsops.c 8.12 (Berkeley) 5/20/95 33 */ 34 35#include <sys/cdefs.h> 36__FBSDID("$FreeBSD: head/sys/fs/nfsclient/nfs_clvfsops.c 221436 2011-05-04 13:27:45Z ru $"); 37 38 39#include "opt_bootp.h" 40#include "opt_nfsroot.h" 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/kernel.h> 45#include <sys/bio.h> 46#include <sys/buf.h> 47#include <sys/clock.h> 48#include <sys/jail.h> 49#include <sys/limits.h> 50#include <sys/lock.h> 51#include <sys/malloc.h> 52#include <sys/mbuf.h> 53#include <sys/module.h> 54#include <sys/mount.h> 55#include <sys/proc.h> 56#include <sys/socket.h> 57#include <sys/socketvar.h> 58#include <sys/sockio.h> 59#include <sys/sysctl.h> 60#include <sys/vnode.h> 61#include <sys/signalvar.h> 62 63#include <vm/vm.h> 64#include <vm/vm_extern.h> 65#include <vm/uma.h> 66 67#include <net/if.h> 68#include <net/route.h> 69#include <netinet/in.h> 70 71#include <fs/nfs/nfsport.h> 72#include <fs/nfsclient/nfsnode.h> 73#include <fs/nfsclient/nfsmount.h> 74#include <fs/nfsclient/nfs.h> 75#include <nfs/nfsdiskless.h> 76 77FEATURE(nfscl, "NFSv4 client"); 78 79extern int nfscl_ticks; 80extern struct timeval nfsboottime; 81extern struct nfsstats newnfsstats; 82 83MALLOC_DEFINE(M_NEWNFSREQ, "newnfsclient_req", "New NFS request header"); 84MALLOC_DEFINE(M_NEWNFSMNT, "newnfsmnt", "New NFS mount struct"); 85 86SYSCTL_DECL(_vfs_newnfs); 87SYSCTL_STRUCT(_vfs_newnfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RW, 88 &newnfsstats, nfsstats, "S,nfsstats"); 89static int nfs_ip_paranoia = 1; 90SYSCTL_INT(_vfs_newnfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW, 91 &nfs_ip_paranoia, 0, ""); 92static int nfs_tprintf_initial_delay = NFS_TPRINTF_INITIAL_DELAY; 93SYSCTL_INT(_vfs_newnfs, NFS_TPRINTF_INITIAL_DELAY, 94 downdelayinitial, CTLFLAG_RW, &nfs_tprintf_initial_delay, 0, ""); 95/* how long between console messages "nfs server foo not responding" */ 96static int nfs_tprintf_delay = NFS_TPRINTF_DELAY; 97SYSCTL_INT(_vfs_newnfs, NFS_TPRINTF_DELAY, 98 downdelayinterval, CTLFLAG_RW, &nfs_tprintf_delay, 0, ""); 99 100static int nfs_mountroot(struct mount *); 101static void nfs_sec_name(char *, int *); 102static void nfs_decode_args(struct mount *mp, struct nfsmount *nmp, 103 struct nfs_args *argp, const char *, struct ucred *, 104 struct thread *); 105static int mountnfs(struct nfs_args *, struct mount *, 106 struct sockaddr *, char *, u_char *, int, u_char *, int, 107 u_char *, int, struct vnode **, struct ucred *, 108 struct thread *, int); 109static void nfs_getnlminfo(struct vnode *, uint8_t *, size_t *, 110 struct sockaddr_storage *, int *, off_t *, 111 struct timeval *); 112static vfs_mount_t nfs_mount; 113static vfs_cmount_t nfs_cmount; 114static vfs_unmount_t nfs_unmount; 115static vfs_root_t nfs_root; 116static vfs_statfs_t nfs_statfs; 117static vfs_sync_t nfs_sync; 118static vfs_sysctl_t nfs_sysctl; 119 120/* 121 * nfs vfs operations. 122 */ 123static struct vfsops nfs_vfsops = { 124 .vfs_init = ncl_init, 125 .vfs_mount = nfs_mount, 126 .vfs_cmount = nfs_cmount, 127 .vfs_root = nfs_root, 128 .vfs_statfs = nfs_statfs, 129 .vfs_sync = nfs_sync, 130 .vfs_uninit = ncl_uninit, 131 .vfs_unmount = nfs_unmount, 132 .vfs_sysctl = nfs_sysctl, 133}; 134VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK); 135 136/* So that loader and kldload(2) can find us, wherever we are.. */ 137MODULE_VERSION(nfs, 1); 138MODULE_DEPEND(nfs, nfscommon, 1, 1, 1); 139MODULE_DEPEND(nfs, krpc, 1, 1, 1); 140MODULE_DEPEND(nfs, nfssvc, 1, 1, 1); 141MODULE_DEPEND(nfs, nfslock, 1, 1, 1); 142 143/* 144 * This structure is now defined in sys/nfs/nfs_diskless.c so that it 145 * can be shared by both NFS clients. It is declared here so that it 146 * will be defined for kernels built without NFS_ROOT, although it 147 * isn't used in that case. 148 */ 149#if !defined(NFS_ROOT) && !defined(NFSCLIENT) 150struct nfs_diskless nfs_diskless = { { { 0 } } }; 151struct nfsv3_diskless nfsv3_diskless = { { { 0 } } }; 152int nfs_diskless_valid = 0; 153#endif 154 155SYSCTL_INT(_vfs_newnfs, OID_AUTO, diskless_valid, CTLFLAG_RD, 156 &nfs_diskless_valid, 0, 157 "Has the diskless struct been filled correctly"); 158 159SYSCTL_STRING(_vfs_newnfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD, 160 nfsv3_diskless.root_hostnam, 0, "Path to nfs root"); 161 162SYSCTL_OPAQUE(_vfs_newnfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD, 163 &nfsv3_diskless.root_saddr, sizeof(nfsv3_diskless.root_saddr), 164 "%Ssockaddr_in", "Diskless root nfs address"); 165 166 167void newnfsargs_ntoh(struct nfs_args *); 168static int nfs_mountdiskless(char *, 169 struct sockaddr_in *, struct nfs_args *, 170 struct thread *, struct vnode **, struct mount *); 171static void nfs_convert_diskless(void); 172static void nfs_convert_oargs(struct nfs_args *args, 173 struct onfs_args *oargs); 174 175int 176newnfs_iosize(struct nfsmount *nmp) 177{ 178 int iosize, maxio; 179 180 /* First, set the upper limit for iosize */ 181 if (nmp->nm_flag & NFSMNT_NFSV4) { 182 maxio = NFS_MAXBSIZE; 183 } else if (nmp->nm_flag & NFSMNT_NFSV3) { 184 if (nmp->nm_sotype == SOCK_DGRAM) 185 maxio = NFS_MAXDGRAMDATA; 186 else 187 maxio = NFS_MAXBSIZE; 188 } else { 189 maxio = NFS_V2MAXDATA; 190 } 191 if (nmp->nm_rsize > maxio || nmp->nm_rsize == 0) 192 nmp->nm_rsize = maxio; 193 if (nmp->nm_rsize > MAXBSIZE) 194 nmp->nm_rsize = MAXBSIZE; 195 if (nmp->nm_readdirsize > maxio || nmp->nm_readdirsize == 0) 196 nmp->nm_readdirsize = maxio; 197 if (nmp->nm_readdirsize > nmp->nm_rsize) 198 nmp->nm_readdirsize = nmp->nm_rsize; 199 if (nmp->nm_wsize > maxio || nmp->nm_wsize == 0) 200 nmp->nm_wsize = maxio; 201 if (nmp->nm_wsize > MAXBSIZE) 202 nmp->nm_wsize = MAXBSIZE; 203 204 /* 205 * Calculate the size used for io buffers. Use the larger 206 * of the two sizes to minimise nfs requests but make sure 207 * that it is at least one VM page to avoid wasting buffer 208 * space. 209 */ 210 iosize = imax(nmp->nm_rsize, nmp->nm_wsize); 211 iosize = imax(iosize, PAGE_SIZE); 212 nmp->nm_mountp->mnt_stat.f_iosize = iosize; 213 return (iosize); 214} 215 216static void 217nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs) 218{ 219 220 args->version = NFS_ARGSVERSION; 221 args->addr = oargs->addr; 222 args->addrlen = oargs->addrlen; 223 args->sotype = oargs->sotype; 224 args->proto = oargs->proto; 225 args->fh = oargs->fh; 226 args->fhsize = oargs->fhsize; 227 args->flags = oargs->flags; 228 args->wsize = oargs->wsize; 229 args->rsize = oargs->rsize; 230 args->readdirsize = oargs->readdirsize; 231 args->timeo = oargs->timeo; 232 args->retrans = oargs->retrans; 233 args->readahead = oargs->readahead; 234 args->hostname = oargs->hostname; 235} 236 237static void 238nfs_convert_diskless(void) 239{ 240 241 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif, 242 sizeof(struct ifaliasreq)); 243 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway, 244 sizeof(struct sockaddr_in)); 245 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args); 246 if (nfsv3_diskless.root_args.flags & NFSMNT_NFSV3) { 247 nfsv3_diskless.root_fhsize = NFSX_MYFH; 248 bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_MYFH); 249 } else { 250 nfsv3_diskless.root_fhsize = NFSX_V2FH; 251 bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V2FH); 252 } 253 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr, 254 sizeof(struct sockaddr_in)); 255 bcopy(nfs_diskless.root_hostnam, nfsv3_diskless.root_hostnam, MNAMELEN); 256 nfsv3_diskless.root_time = nfs_diskless.root_time; 257 bcopy(nfs_diskless.my_hostnam, nfsv3_diskless.my_hostnam, 258 MAXHOSTNAMELEN); 259 nfs_diskless_valid = 3; 260} 261 262/* 263 * nfs statfs call 264 */ 265static int 266nfs_statfs(struct mount *mp, struct statfs *sbp) 267{ 268 struct vnode *vp; 269 struct thread *td; 270 struct nfsmount *nmp = VFSTONFS(mp); 271 struct nfsvattr nfsva; 272 struct nfsfsinfo fs; 273 struct nfsstatfs sb; 274 int error = 0, attrflag, gotfsinfo = 0, ret; 275 struct nfsnode *np; 276 277 td = curthread; 278 279 error = vfs_busy(mp, MBF_NOWAIT); 280 if (error) 281 return (error); 282 error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, LK_EXCLUSIVE); 283 if (error) { 284 vfs_unbusy(mp); 285 return (error); 286 } 287 vp = NFSTOV(np); 288 mtx_lock(&nmp->nm_mtx); 289 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) { 290 mtx_unlock(&nmp->nm_mtx); 291 error = nfsrpc_fsinfo(vp, &fs, td->td_ucred, td, &nfsva, 292 &attrflag, NULL); 293 if (!error) 294 gotfsinfo = 1; 295 } else 296 mtx_unlock(&nmp->nm_mtx); 297 if (!error) 298 error = nfsrpc_statfs(vp, &sb, &fs, td->td_ucred, td, &nfsva, 299 &attrflag, NULL); 300 if (attrflag == 0) { 301 ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1, 302 td->td_ucred, td, &nfsva, NULL); 303 if (ret) { 304 /* 305 * Just set default values to get things going. 306 */ 307 NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr)); 308 nfsva.na_vattr.va_type = VDIR; 309 nfsva.na_vattr.va_mode = 0777; 310 nfsva.na_vattr.va_nlink = 100; 311 nfsva.na_vattr.va_uid = (uid_t)0; 312 nfsva.na_vattr.va_gid = (gid_t)0; 313 nfsva.na_vattr.va_fileid = 2; 314 nfsva.na_vattr.va_gen = 1; 315 nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE; 316 nfsva.na_vattr.va_size = 512 * 1024; 317 } 318 } 319 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1); 320 if (!error) { 321 mtx_lock(&nmp->nm_mtx); 322 if (gotfsinfo || (nmp->nm_flag & NFSMNT_NFSV4)) 323 nfscl_loadfsinfo(nmp, &fs); 324 nfscl_loadsbinfo(nmp, &sb, sbp); 325 sbp->f_iosize = newnfs_iosize(nmp); 326 mtx_unlock(&nmp->nm_mtx); 327 if (sbp != &mp->mnt_stat) { 328 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN); 329 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); 330 } 331 strncpy(&sbp->f_fstypename[0], mp->mnt_vfc->vfc_name, MFSNAMELEN); 332 } else if (NFS_ISV4(vp)) { 333 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 334 } 335 vput(vp); 336 vfs_unbusy(mp); 337 return (error); 338} 339 340/* 341 * nfs version 3 fsinfo rpc call 342 */ 343int 344ncl_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct ucred *cred, 345 struct thread *td) 346{ 347 struct nfsfsinfo fs; 348 struct nfsvattr nfsva; 349 int error, attrflag; 350 351 error = nfsrpc_fsinfo(vp, &fs, cred, td, &nfsva, &attrflag, NULL); 352 if (!error) { 353 if (attrflag) 354 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 355 1); 356 mtx_lock(&nmp->nm_mtx); 357 nfscl_loadfsinfo(nmp, &fs); 358 mtx_unlock(&nmp->nm_mtx); 359 } 360 return (error); 361} 362 363/* 364 * Mount a remote root fs via. nfs. This depends on the info in the 365 * nfs_diskless structure that has been filled in properly by some primary 366 * bootstrap. 367 * It goes something like this: 368 * - do enough of "ifconfig" by calling ifioctl() so that the system 369 * can talk to the server 370 * - If nfs_diskless.mygateway is filled in, use that address as 371 * a default gateway. 372 * - build the rootfs mount point and call mountnfs() to do the rest. 373 * 374 * It is assumed to be safe to read, modify, and write the nfsv3_diskless 375 * structure, as well as other global NFS client variables here, as 376 * nfs_mountroot() will be called once in the boot before any other NFS 377 * client activity occurs. 378 */ 379static int 380nfs_mountroot(struct mount *mp) 381{ 382 struct thread *td = curthread; 383 struct nfsv3_diskless *nd = &nfsv3_diskless; 384 struct socket *so; 385 struct vnode *vp; 386 struct ifreq ir; 387 int error; 388 u_long l; 389 char buf[128]; 390 char *cp; 391 392#if defined(BOOTP_NFSROOT) && defined(BOOTP) 393 bootpc_init(); /* use bootp to get nfs_diskless filled in */ 394#elif defined(NFS_ROOT) 395 nfs_setup_diskless(); 396#endif 397 398 if (nfs_diskless_valid == 0) 399 return (-1); 400 if (nfs_diskless_valid == 1) 401 nfs_convert_diskless(); 402 403 /* 404 * XXX splnet, so networks will receive... 405 */ 406 splnet(); 407 408 /* 409 * Do enough of ifconfig(8) so that the critical net interface can 410 * talk to the server. 411 */ 412 error = socreate(nd->myif.ifra_addr.sa_family, &so, nd->root_args.sotype, 0, 413 td->td_ucred, td); 414 if (error) 415 panic("nfs_mountroot: socreate(%04x): %d", 416 nd->myif.ifra_addr.sa_family, error); 417 418#if 0 /* XXX Bad idea */ 419 /* 420 * We might not have been told the right interface, so we pass 421 * over the first ten interfaces of the same kind, until we get 422 * one of them configured. 423 */ 424 425 for (i = strlen(nd->myif.ifra_name) - 1; 426 nd->myif.ifra_name[i] >= '0' && 427 nd->myif.ifra_name[i] <= '9'; 428 nd->myif.ifra_name[i] ++) { 429 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); 430 if(!error) 431 break; 432 } 433#endif 434 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); 435 if (error) 436 panic("nfs_mountroot: SIOCAIFADDR: %d", error); 437 if ((cp = getenv("boot.netif.mtu")) != NULL) { 438 ir.ifr_mtu = strtol(cp, NULL, 10); 439 bcopy(nd->myif.ifra_name, ir.ifr_name, IFNAMSIZ); 440 freeenv(cp); 441 error = ifioctl(so, SIOCSIFMTU, (caddr_t)&ir, td); 442 if (error) 443 printf("nfs_mountroot: SIOCSIFMTU: %d", error); 444 } 445 soclose(so); 446 447 /* 448 * If the gateway field is filled in, set it as the default route. 449 * Note that pxeboot will set a default route of 0 if the route 450 * is not set by the DHCP server. Check also for a value of 0 451 * to avoid panicking inappropriately in that situation. 452 */ 453 if (nd->mygateway.sin_len != 0 && 454 nd->mygateway.sin_addr.s_addr != 0) { 455 struct sockaddr_in mask, sin; 456 457 bzero((caddr_t)&mask, sizeof(mask)); 458 sin = mask; 459 sin.sin_family = AF_INET; 460 sin.sin_len = sizeof(sin); 461 /* XXX MRT use table 0 for this sort of thing */ 462 CURVNET_SET(TD_TO_VNET(td)); 463 error = rtrequest(RTM_ADD, (struct sockaddr *)&sin, 464 (struct sockaddr *)&nd->mygateway, 465 (struct sockaddr *)&mask, 466 RTF_UP | RTF_GATEWAY, NULL); 467 CURVNET_RESTORE(); 468 if (error) 469 panic("nfs_mountroot: RTM_ADD: %d", error); 470 } 471 472 /* 473 * Create the rootfs mount point. 474 */ 475 nd->root_args.fh = nd->root_fh; 476 nd->root_args.fhsize = nd->root_fhsize; 477 l = ntohl(nd->root_saddr.sin_addr.s_addr); 478 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 479 (l >> 24) & 0xff, (l >> 16) & 0xff, 480 (l >> 8) & 0xff, (l >> 0) & 0xff, nd->root_hostnam); 481 printf("NFS ROOT: %s\n", buf); 482 nd->root_args.hostname = buf; 483 if ((error = nfs_mountdiskless(buf, 484 &nd->root_saddr, &nd->root_args, td, &vp, mp)) != 0) { 485 return (error); 486 } 487 488 /* 489 * This is not really an nfs issue, but it is much easier to 490 * set hostname here and then let the "/etc/rc.xxx" files 491 * mount the right /var based upon its preset value. 492 */ 493 mtx_lock(&prison0.pr_mtx); 494 strlcpy(prison0.pr_hostname, nd->my_hostnam, 495 sizeof(prison0.pr_hostname)); 496 mtx_unlock(&prison0.pr_mtx); 497 inittodr(ntohl(nd->root_time)); 498 return (0); 499} 500 501/* 502 * Internal version of mount system call for diskless setup. 503 */ 504static int 505nfs_mountdiskless(char *path, 506 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td, 507 struct vnode **vpp, struct mount *mp) 508{ 509 struct sockaddr *nam; 510 int dirlen, error; 511 char *dirpath; 512 513 /* 514 * Find the directory path in "path", which also has the server's 515 * name/ip address in it. 516 */ 517 dirpath = strchr(path, ':'); 518 if (dirpath != NULL) 519 dirlen = strlen(++dirpath); 520 else 521 dirlen = 0; 522 nam = sodupsockaddr((struct sockaddr *)sin, M_WAITOK); 523 if ((error = mountnfs(args, mp, nam, path, NULL, 0, dirpath, dirlen, 524 NULL, 0, vpp, td->td_ucred, td, NFS_DEFAULT_NEGNAMETIMEO)) != 0) { 525 printf("nfs_mountroot: mount %s on /: %d\n", path, error); 526 return (error); 527 } 528 return (0); 529} 530 531static void 532nfs_sec_name(char *sec, int *flagsp) 533{ 534 if (!strcmp(sec, "krb5")) 535 *flagsp |= NFSMNT_KERB; 536 else if (!strcmp(sec, "krb5i")) 537 *flagsp |= (NFSMNT_KERB | NFSMNT_INTEGRITY); 538 else if (!strcmp(sec, "krb5p")) 539 *flagsp |= (NFSMNT_KERB | NFSMNT_PRIVACY); 540} 541 542static void 543nfs_decode_args(struct mount *mp, struct nfsmount *nmp, struct nfs_args *argp, 544 const char *hostname, struct ucred *cred, struct thread *td) 545{ 546 int s; 547 int adjsock; 548 char *p; 549 550 s = splnet(); 551 552 /* 553 * Set read-only flag if requested; otherwise, clear it if this is 554 * an update. If this is not an update, then either the read-only 555 * flag is already clear, or this is a root mount and it was set 556 * intentionally at some previous point. 557 */ 558 if (vfs_getopt(mp->mnt_optnew, "ro", NULL, NULL) == 0) { 559 MNT_ILOCK(mp); 560 mp->mnt_flag |= MNT_RDONLY; 561 MNT_IUNLOCK(mp); 562 } else if (mp->mnt_flag & MNT_UPDATE) { 563 MNT_ILOCK(mp); 564 mp->mnt_flag &= ~MNT_RDONLY; 565 MNT_IUNLOCK(mp); 566 } 567 568 /* 569 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 570 * no sense in that context. Also, set up appropriate retransmit 571 * and soft timeout behavior. 572 */ 573 if (argp->sotype == SOCK_STREAM) { 574 nmp->nm_flag &= ~NFSMNT_NOCONN; 575 nmp->nm_timeo = NFS_MAXTIMEO; 576 if ((argp->flags & NFSMNT_NFSV4) != 0) 577 nmp->nm_retry = INT_MAX; 578 else 579 nmp->nm_retry = NFS_RETRANS_TCP; 580 } 581 582 /* Also clear RDIRPLUS if NFSv2, it crashes some servers */ 583 if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) { 584 argp->flags &= ~NFSMNT_RDIRPLUS; 585 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 586 } 587 588 /* Clear NFSMNT_RESVPORT for NFSv4, since it is not required. */ 589 if ((argp->flags & NFSMNT_NFSV4) != 0) { 590 argp->flags &= ~NFSMNT_RESVPORT; 591 nmp->nm_flag &= ~NFSMNT_RESVPORT; 592 } 593 594 /* Re-bind if rsrvd port requested and wasn't on one */ 595 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 596 && (argp->flags & NFSMNT_RESVPORT); 597 /* Also re-bind if we're switching to/from a connected UDP socket */ 598 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 599 (argp->flags & NFSMNT_NOCONN)); 600 601 /* Update flags atomically. Don't change the lock bits. */ 602 nmp->nm_flag = argp->flags | nmp->nm_flag; 603 splx(s); 604 605 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 606 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 607 if (nmp->nm_timeo < NFS_MINTIMEO) 608 nmp->nm_timeo = NFS_MINTIMEO; 609 else if (nmp->nm_timeo > NFS_MAXTIMEO) 610 nmp->nm_timeo = NFS_MAXTIMEO; 611 } 612 613 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 614 nmp->nm_retry = argp->retrans; 615 if (nmp->nm_retry > NFS_MAXREXMIT) 616 nmp->nm_retry = NFS_MAXREXMIT; 617 } 618 619 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 620 nmp->nm_wsize = argp->wsize; 621 /* Round down to multiple of blocksize */ 622 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 623 if (nmp->nm_wsize <= 0) 624 nmp->nm_wsize = NFS_FABLKSIZE; 625 } 626 627 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 628 nmp->nm_rsize = argp->rsize; 629 /* Round down to multiple of blocksize */ 630 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 631 if (nmp->nm_rsize <= 0) 632 nmp->nm_rsize = NFS_FABLKSIZE; 633 } 634 635 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 636 nmp->nm_readdirsize = argp->readdirsize; 637 } 638 639 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0) 640 nmp->nm_acregmin = argp->acregmin; 641 else 642 nmp->nm_acregmin = NFS_MINATTRTIMO; 643 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0) 644 nmp->nm_acregmax = argp->acregmax; 645 else 646 nmp->nm_acregmax = NFS_MAXATTRTIMO; 647 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0) 648 nmp->nm_acdirmin = argp->acdirmin; 649 else 650 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO; 651 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0) 652 nmp->nm_acdirmax = argp->acdirmax; 653 else 654 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO; 655 if (nmp->nm_acdirmin > nmp->nm_acdirmax) 656 nmp->nm_acdirmin = nmp->nm_acdirmax; 657 if (nmp->nm_acregmin > nmp->nm_acregmax) 658 nmp->nm_acregmin = nmp->nm_acregmax; 659 660 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) { 661 if (argp->readahead <= NFS_MAXRAHEAD) 662 nmp->nm_readahead = argp->readahead; 663 else 664 nmp->nm_readahead = NFS_MAXRAHEAD; 665 } 666 if ((argp->flags & NFSMNT_WCOMMITSIZE) && argp->wcommitsize >= 0) { 667 if (argp->wcommitsize < nmp->nm_wsize) 668 nmp->nm_wcommitsize = nmp->nm_wsize; 669 else 670 nmp->nm_wcommitsize = argp->wcommitsize; 671 } 672 673 adjsock |= ((nmp->nm_sotype != argp->sotype) || 674 (nmp->nm_soproto != argp->proto)); 675 676 if (nmp->nm_client != NULL && adjsock) { 677 int haslock = 0, error = 0; 678 679 if (nmp->nm_sotype == SOCK_STREAM) { 680 error = newnfs_sndlock(&nmp->nm_sockreq.nr_lock); 681 if (!error) 682 haslock = 1; 683 } 684 if (!error) { 685 newnfs_disconnect(&nmp->nm_sockreq); 686 if (haslock) 687 newnfs_sndunlock(&nmp->nm_sockreq.nr_lock); 688 nmp->nm_sotype = argp->sotype; 689 nmp->nm_soproto = argp->proto; 690 if (nmp->nm_sotype == SOCK_DGRAM) 691 while (newnfs_connect(nmp, &nmp->nm_sockreq, 692 cred, td, 0)) { 693 printf("newnfs_args: retrying connect\n"); 694 (void) nfs_catnap(PSOCK, 0, "newnfscon"); 695 } 696 } 697 } else { 698 nmp->nm_sotype = argp->sotype; 699 nmp->nm_soproto = argp->proto; 700 } 701 702 if (hostname != NULL) { 703 strlcpy(nmp->nm_hostname, hostname, 704 sizeof(nmp->nm_hostname)); 705 p = strchr(nmp->nm_hostname, ':'); 706 if (p != NULL) 707 *p = '\0'; 708 } 709} 710 711static const char *nfs_opts[] = { "from", "nfs_args", 712 "noatime", "noexec", "suiddir", "nosuid", "nosymfollow", "union", 713 "noclusterr", "noclusterw", "multilabel", "acls", "force", "update", 714 "async", "noconn", "nolockd", "conn", "lockd", "intr", "rdirplus", 715 "readdirsize", "soft", "hard", "mntudp", "tcp", "udp", "wsize", "rsize", 716 "retrans", "acregmin", "acregmax", "acdirmin", "acdirmax", "resvport", 717 "readahead", "hostname", "timeout", "addr", "fh", "nfsv3", "sec", 718 "principal", "nfsv4", "gssname", "allgssname", "dirpath", 719 "negnametimeo", "nocto", 720 NULL }; 721 722/* 723 * VFS Operations. 724 * 725 * mount system call 726 * It seems a bit dumb to copyinstr() the host and path here and then 727 * bcopy() them in mountnfs(), but I wanted to detect errors before 728 * doing the sockargs() call because sockargs() allocates an mbuf and 729 * an error after that means that I have to release the mbuf. 730 */ 731/* ARGSUSED */ 732static int 733nfs_mount(struct mount *mp) 734{ 735 struct nfs_args args = { 736 .version = NFS_ARGSVERSION, 737 .addr = NULL, 738 .addrlen = sizeof (struct sockaddr_in), 739 .sotype = SOCK_STREAM, 740 .proto = 0, 741 .fh = NULL, 742 .fhsize = 0, 743 .flags = NFSMNT_RESVPORT, 744 .wsize = NFS_WSIZE, 745 .rsize = NFS_RSIZE, 746 .readdirsize = NFS_READDIRSIZE, 747 .timeo = 10, 748 .retrans = NFS_RETRANS, 749 .readahead = NFS_DEFRAHEAD, 750 .wcommitsize = 0, /* was: NQ_DEFLEASE */ 751 .hostname = NULL, 752 .acregmin = NFS_MINATTRTIMO, 753 .acregmax = NFS_MAXATTRTIMO, 754 .acdirmin = NFS_MINDIRATTRTIMO, 755 .acdirmax = NFS_MAXDIRATTRTIMO, 756 }; 757 int error = 0, ret, len; 758 struct sockaddr *nam = NULL; 759 struct vnode *vp; 760 struct thread *td; 761 char hst[MNAMELEN]; 762 u_char nfh[NFSX_FHMAX], krbname[100], dirpath[100], srvkrbname[100]; 763 char *opt, *name, *secname; 764 int negnametimeo = NFS_DEFAULT_NEGNAMETIMEO; 765 int dirlen, has_nfs_args_opt, krbnamelen, srvkrbnamelen; 766 size_t hstlen; 767 768 has_nfs_args_opt = 0; 769 if (vfs_filteropt(mp->mnt_optnew, nfs_opts)) { 770 error = EINVAL; 771 goto out; 772 } 773 774 td = curthread; 775 if ((mp->mnt_flag & (MNT_ROOTFS | MNT_UPDATE)) == MNT_ROOTFS) { 776 error = nfs_mountroot(mp); 777 goto out; 778 } 779 780 nfscl_init(); 781 782 /* 783 * The old mount_nfs program passed the struct nfs_args 784 * from userspace to kernel. The new mount_nfs program 785 * passes string options via nmount() from userspace to kernel 786 * and we populate the struct nfs_args in the kernel. 787 */ 788 if (vfs_getopt(mp->mnt_optnew, "nfs_args", NULL, NULL) == 0) { 789 error = vfs_copyopt(mp->mnt_optnew, "nfs_args", &args, 790 sizeof(args)); 791 if (error != 0) 792 goto out; 793 794 if (args.version != NFS_ARGSVERSION) { 795 error = EPROGMISMATCH; 796 goto out; 797 } 798 has_nfs_args_opt = 1; 799 } 800 801 /* Handle the new style options. */ 802 if (vfs_getopt(mp->mnt_optnew, "noconn", NULL, NULL) == 0) 803 args.flags |= NFSMNT_NOCONN; 804 if (vfs_getopt(mp->mnt_optnew, "conn", NULL, NULL) == 0) 805 args.flags |= NFSMNT_NOCONN; 806 if (vfs_getopt(mp->mnt_optnew, "nolockd", NULL, NULL) == 0) 807 args.flags |= NFSMNT_NOLOCKD; 808 if (vfs_getopt(mp->mnt_optnew, "lockd", NULL, NULL) == 0) 809 args.flags &= ~NFSMNT_NOLOCKD; 810 if (vfs_getopt(mp->mnt_optnew, "intr", NULL, NULL) == 0) 811 args.flags |= NFSMNT_INT; 812 if (vfs_getopt(mp->mnt_optnew, "rdirplus", NULL, NULL) == 0) 813 args.flags |= NFSMNT_RDIRPLUS; 814 if (vfs_getopt(mp->mnt_optnew, "resvport", NULL, NULL) == 0) 815 args.flags |= NFSMNT_RESVPORT; 816 if (vfs_getopt(mp->mnt_optnew, "noresvport", NULL, NULL) == 0) 817 args.flags &= ~NFSMNT_RESVPORT; 818 if (vfs_getopt(mp->mnt_optnew, "soft", NULL, NULL) == 0) 819 args.flags |= NFSMNT_SOFT; 820 if (vfs_getopt(mp->mnt_optnew, "hard", NULL, NULL) == 0) 821 args.flags &= ~NFSMNT_SOFT; 822 if (vfs_getopt(mp->mnt_optnew, "mntudp", NULL, NULL) == 0) 823 args.sotype = SOCK_DGRAM; 824 if (vfs_getopt(mp->mnt_optnew, "udp", NULL, NULL) == 0) 825 args.sotype = SOCK_DGRAM; 826 if (vfs_getopt(mp->mnt_optnew, "tcp", NULL, NULL) == 0) 827 args.sotype = SOCK_STREAM; 828 if (vfs_getopt(mp->mnt_optnew, "nfsv3", NULL, NULL) == 0) 829 args.flags |= NFSMNT_NFSV3; 830 if (vfs_getopt(mp->mnt_optnew, "nfsv4", NULL, NULL) == 0) { 831 args.flags |= NFSMNT_NFSV4; 832 args.sotype = SOCK_STREAM; 833 } 834 if (vfs_getopt(mp->mnt_optnew, "allgssname", NULL, NULL) == 0) 835 args.flags |= NFSMNT_ALLGSSNAME; 836 if (vfs_getopt(mp->mnt_optnew, "nocto", NULL, NULL) == 0) 837 args.flags |= NFSMNT_NOCTO; 838 if (vfs_getopt(mp->mnt_optnew, "readdirsize", (void **)&opt, NULL) == 0) { 839 if (opt == NULL) { 840 vfs_mount_error(mp, "illegal readdirsize"); 841 error = EINVAL; 842 goto out; 843 } 844 ret = sscanf(opt, "%d", &args.readdirsize); 845 if (ret != 1 || args.readdirsize <= 0) { 846 vfs_mount_error(mp, "illegal readdirsize: %s", 847 opt); 848 error = EINVAL; 849 goto out; 850 } 851 args.flags |= NFSMNT_READDIRSIZE; 852 } 853 if (vfs_getopt(mp->mnt_optnew, "readahead", (void **)&opt, NULL) == 0) { 854 if (opt == NULL) { 855 vfs_mount_error(mp, "illegal readahead"); 856 error = EINVAL; 857 goto out; 858 } 859 ret = sscanf(opt, "%d", &args.readahead); 860 if (ret != 1 || args.readahead <= 0) { 861 vfs_mount_error(mp, "illegal readahead: %s", 862 opt); 863 error = EINVAL; 864 goto out; 865 } 866 args.flags |= NFSMNT_READAHEAD; 867 } 868 if (vfs_getopt(mp->mnt_optnew, "wsize", (void **)&opt, NULL) == 0) { 869 if (opt == NULL) { 870 vfs_mount_error(mp, "illegal wsize"); 871 error = EINVAL; 872 goto out; 873 } 874 ret = sscanf(opt, "%d", &args.wsize); 875 if (ret != 1 || args.wsize <= 0) { 876 vfs_mount_error(mp, "illegal wsize: %s", 877 opt); 878 error = EINVAL; 879 goto out; 880 } 881 args.flags |= NFSMNT_WSIZE; 882 } 883 if (vfs_getopt(mp->mnt_optnew, "rsize", (void **)&opt, NULL) == 0) { 884 if (opt == NULL) { 885 vfs_mount_error(mp, "illegal rsize"); 886 error = EINVAL; 887 goto out; 888 } 889 ret = sscanf(opt, "%d", &args.rsize); 890 if (ret != 1 || args.rsize <= 0) { 891 vfs_mount_error(mp, "illegal wsize: %s", 892 opt); 893 error = EINVAL; 894 goto out; 895 } 896 args.flags |= NFSMNT_RSIZE; 897 } 898 if (vfs_getopt(mp->mnt_optnew, "retrans", (void **)&opt, NULL) == 0) { 899 if (opt == NULL) { 900 vfs_mount_error(mp, "illegal retrans"); 901 error = EINVAL; 902 goto out; 903 } 904 ret = sscanf(opt, "%d", &args.retrans); 905 if (ret != 1 || args.retrans <= 0) { 906 vfs_mount_error(mp, "illegal retrans: %s", 907 opt); 908 error = EINVAL; 909 goto out; 910 } 911 args.flags |= NFSMNT_RETRANS; 912 } 913 if (vfs_getopt(mp->mnt_optnew, "acregmin", (void **)&opt, NULL) == 0) { 914 ret = sscanf(opt, "%d", &args.acregmin); 915 if (ret != 1 || args.acregmin < 0) { 916 vfs_mount_error(mp, "illegal acregmin: %s", 917 opt); 918 error = EINVAL; 919 goto out; 920 } 921 args.flags |= NFSMNT_ACREGMIN; 922 } 923 if (vfs_getopt(mp->mnt_optnew, "acregmax", (void **)&opt, NULL) == 0) { 924 ret = sscanf(opt, "%d", &args.acregmax); 925 if (ret != 1 || args.acregmax < 0) { 926 vfs_mount_error(mp, "illegal acregmax: %s", 927 opt); 928 error = EINVAL; 929 goto out; 930 } 931 args.flags |= NFSMNT_ACREGMAX; 932 } 933 if (vfs_getopt(mp->mnt_optnew, "acdirmin", (void **)&opt, NULL) == 0) { 934 ret = sscanf(opt, "%d", &args.acdirmin); 935 if (ret != 1 || args.acdirmin < 0) { 936 vfs_mount_error(mp, "illegal acdirmin: %s", 937 opt); 938 error = EINVAL; 939 goto out; 940 } 941 args.flags |= NFSMNT_ACDIRMIN; 942 } 943 if (vfs_getopt(mp->mnt_optnew, "acdirmax", (void **)&opt, NULL) == 0) { 944 ret = sscanf(opt, "%d", &args.acdirmax); 945 if (ret != 1 || args.acdirmax < 0) { 946 vfs_mount_error(mp, "illegal acdirmax: %s", 947 opt); 948 error = EINVAL; 949 goto out; 950 } 951 args.flags |= NFSMNT_ACDIRMAX; 952 } 953 if (vfs_getopt(mp->mnt_optnew, "timeout", (void **)&opt, NULL) == 0) { 954 ret = sscanf(opt, "%d", &args.timeo); 955 if (ret != 1 || args.timeo <= 0) { 956 vfs_mount_error(mp, "illegal timeout: %s", 957 opt); 958 error = EINVAL; 959 goto out; 960 } 961 args.flags |= NFSMNT_TIMEO; 962 } 963 if (vfs_getopt(mp->mnt_optnew, "negnametimeo", (void **)&opt, NULL) 964 == 0) { 965 ret = sscanf(opt, "%d", &negnametimeo); 966 if (ret != 1 || negnametimeo < 0) { 967 vfs_mount_error(mp, "illegal negnametimeo: %s", 968 opt); 969 error = EINVAL; 970 goto out; 971 } 972 } 973 if (vfs_getopt(mp->mnt_optnew, "sec", 974 (void **) &secname, NULL) == 0) 975 nfs_sec_name(secname, &args.flags); 976 977 if (mp->mnt_flag & MNT_UPDATE) { 978 struct nfsmount *nmp = VFSTONFS(mp); 979 980 if (nmp == NULL) { 981 error = EIO; 982 goto out; 983 } 984 /* 985 * When doing an update, we can't change version, 986 * security, switch lockd strategies or change cookie 987 * translation 988 */ 989 args.flags = (args.flags & 990 ~(NFSMNT_NFSV3 | 991 NFSMNT_NFSV4 | 992 NFSMNT_KERB | 993 NFSMNT_INTEGRITY | 994 NFSMNT_PRIVACY | 995 NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)) | 996 (nmp->nm_flag & 997 (NFSMNT_NFSV3 | 998 NFSMNT_NFSV4 | 999 NFSMNT_KERB | 1000 NFSMNT_INTEGRITY | 1001 NFSMNT_PRIVACY | 1002 NFSMNT_NOLOCKD /*|NFSMNT_XLATECOOKIE*/)); 1003 nfs_decode_args(mp, nmp, &args, NULL, td->td_ucred, td); 1004 goto out; 1005 } 1006 1007 /* 1008 * Make the nfs_ip_paranoia sysctl serve as the default connection 1009 * or no-connection mode for those protocols that support 1010 * no-connection mode (the flag will be cleared later for protocols 1011 * that do not support no-connection mode). This will allow a client 1012 * to receive replies from a different IP then the request was 1013 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid), 1014 * not 0. 1015 */ 1016 if (nfs_ip_paranoia == 0) 1017 args.flags |= NFSMNT_NOCONN; 1018 1019 if (has_nfs_args_opt != 0) { 1020 /* 1021 * In the 'nfs_args' case, the pointers in the args 1022 * structure are in userland - we copy them in here. 1023 */ 1024 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) { 1025 vfs_mount_error(mp, "Bad file handle"); 1026 error = EINVAL; 1027 goto out; 1028 } 1029 error = copyin((caddr_t)args.fh, (caddr_t)nfh, 1030 args.fhsize); 1031 if (error != 0) 1032 goto out; 1033 error = copyinstr(args.hostname, hst, MNAMELEN - 1, &hstlen); 1034 if (error != 0) 1035 goto out; 1036 bzero(&hst[hstlen], MNAMELEN - hstlen); 1037 args.hostname = hst; 1038 /* sockargs() call must be after above copyin() calls */ 1039 error = getsockaddr(&nam, (caddr_t)args.addr, 1040 args.addrlen); 1041 if (error != 0) 1042 goto out; 1043 } else { 1044 if (vfs_getopt(mp->mnt_optnew, "fh", (void **)&args.fh, 1045 &args.fhsize) == 0) { 1046 if (args.fhsize < 0 || args.fhsize > NFSX_FHMAX) { 1047 vfs_mount_error(mp, "Bad file handle"); 1048 error = EINVAL; 1049 goto out; 1050 } 1051 bcopy(args.fh, nfh, args.fhsize); 1052 } else { 1053 args.fhsize = 0; 1054 } 1055 (void) vfs_getopt(mp->mnt_optnew, "hostname", 1056 (void **)&args.hostname, &len); 1057 if (args.hostname == NULL) { 1058 vfs_mount_error(mp, "Invalid hostname"); 1059 error = EINVAL; 1060 goto out; 1061 } 1062 bcopy(args.hostname, hst, MNAMELEN); 1063 hst[MNAMELEN - 1] = '\0'; 1064 } 1065 1066 if (vfs_getopt(mp->mnt_optnew, "principal", (void **)&name, NULL) == 0) 1067 strlcpy(srvkrbname, name, sizeof (srvkrbname)); 1068 else 1069 snprintf(srvkrbname, sizeof (srvkrbname), "nfs@%s", hst); 1070 srvkrbnamelen = strlen(srvkrbname); 1071 1072 if (vfs_getopt(mp->mnt_optnew, "gssname", (void **)&name, NULL) == 0) 1073 strlcpy(krbname, name, sizeof (krbname)); 1074 else 1075 krbname[0] = '\0'; 1076 krbnamelen = strlen(krbname); 1077 1078 if (vfs_getopt(mp->mnt_optnew, "dirpath", (void **)&name, NULL) == 0) 1079 strlcpy(dirpath, name, sizeof (dirpath)); 1080 else 1081 dirpath[0] = '\0'; 1082 dirlen = strlen(dirpath); 1083 1084 if (has_nfs_args_opt == 0 && vfs_getopt(mp->mnt_optnew, "addr", 1085 (void **)&args.addr, &args.addrlen) == 0) { 1086 if (args.addrlen > SOCK_MAXADDRLEN) { 1087 error = ENAMETOOLONG; 1088 goto out; 1089 } 1090 nam = malloc(args.addrlen, M_SONAME, M_WAITOK); 1091 bcopy(args.addr, nam, args.addrlen); 1092 nam->sa_len = args.addrlen; 1093 } 1094 1095 args.fh = nfh; 1096 error = mountnfs(&args, mp, nam, hst, krbname, krbnamelen, dirpath, 1097 dirlen, srvkrbname, srvkrbnamelen, &vp, td->td_ucred, td, 1098 negnametimeo); 1099out: 1100 if (!error) { 1101 MNT_ILOCK(mp); 1102 mp->mnt_kern_flag |= (MNTK_MPSAFE|MNTK_LOOKUP_SHARED); 1103 MNT_IUNLOCK(mp); 1104 } 1105 return (error); 1106} 1107 1108 1109/* 1110 * VFS Operations. 1111 * 1112 * mount system call 1113 * It seems a bit dumb to copyinstr() the host and path here and then 1114 * bcopy() them in mountnfs(), but I wanted to detect errors before 1115 * doing the sockargs() call because sockargs() allocates an mbuf and 1116 * an error after that means that I have to release the mbuf. 1117 */ 1118/* ARGSUSED */ 1119static int 1120nfs_cmount(struct mntarg *ma, void *data, int flags) 1121{ 1122 int error; 1123 struct nfs_args args; 1124 1125 error = copyin(data, &args, sizeof (struct nfs_args)); 1126 if (error) 1127 return error; 1128 1129 ma = mount_arg(ma, "nfs_args", &args, sizeof args); 1130 1131 error = kernel_mount(ma, flags); 1132 return (error); 1133} 1134 1135/* 1136 * Common code for mount and mountroot 1137 */ 1138static int 1139mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam, 1140 char *hst, u_char *krbname, int krbnamelen, u_char *dirpath, int dirlen, 1141 u_char *srvkrbname, int srvkrbnamelen, struct vnode **vpp, 1142 struct ucred *cred, struct thread *td, int negnametimeo) 1143{ 1144 struct nfsmount *nmp; 1145 struct nfsnode *np; 1146 int error, trycnt, ret; 1147 struct nfsvattr nfsva; 1148 static u_int64_t clval = 0; 1149 1150 if (mp->mnt_flag & MNT_UPDATE) { 1151 nmp = VFSTONFS(mp); 1152 printf("%s: MNT_UPDATE is no longer handled here\n", __func__); 1153 FREE(nam, M_SONAME); 1154 return (0); 1155 } else { 1156 MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount) + 1157 krbnamelen + dirlen + srvkrbnamelen + 2, 1158 M_NEWNFSMNT, M_WAITOK | M_ZERO); 1159 TAILQ_INIT(&nmp->nm_bufq); 1160 if (clval == 0) 1161 clval = (u_int64_t)nfsboottime.tv_sec; 1162 nmp->nm_clval = clval++; 1163 nmp->nm_krbnamelen = krbnamelen; 1164 nmp->nm_dirpathlen = dirlen; 1165 nmp->nm_srvkrbnamelen = srvkrbnamelen; 1166 if (td->td_ucred->cr_uid != (uid_t)0) { 1167 /* 1168 * nm_uid is used to get KerberosV credentials for 1169 * the nfsv4 state handling operations if there is 1170 * no host based principal set. Use the uid of 1171 * this user if not root, since they are doing the 1172 * mount. I don't think setting this for root will 1173 * work, since root normally does not have user 1174 * credentials in a credentials cache. 1175 */ 1176 nmp->nm_uid = td->td_ucred->cr_uid; 1177 } else { 1178 /* 1179 * Just set to -1, so it won't be used. 1180 */ 1181 nmp->nm_uid = (uid_t)-1; 1182 } 1183 1184 /* Copy and null terminate all the names */ 1185 if (nmp->nm_krbnamelen > 0) { 1186 bcopy(krbname, nmp->nm_krbname, nmp->nm_krbnamelen); 1187 nmp->nm_name[nmp->nm_krbnamelen] = '\0'; 1188 } 1189 if (nmp->nm_dirpathlen > 0) { 1190 bcopy(dirpath, NFSMNT_DIRPATH(nmp), 1191 nmp->nm_dirpathlen); 1192 nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen 1193 + 1] = '\0'; 1194 } 1195 if (nmp->nm_srvkrbnamelen > 0) { 1196 bcopy(srvkrbname, NFSMNT_SRVKRBNAME(nmp), 1197 nmp->nm_srvkrbnamelen); 1198 nmp->nm_name[nmp->nm_krbnamelen + nmp->nm_dirpathlen 1199 + nmp->nm_srvkrbnamelen + 2] = '\0'; 1200 } 1201 nmp->nm_sockreq.nr_cred = crhold(cred); 1202 mtx_init(&nmp->nm_sockreq.nr_mtx, "nfssock", NULL, MTX_DEF); 1203 mp->mnt_data = nmp; 1204 nmp->nm_getinfo = nfs_getnlminfo; 1205 nmp->nm_vinvalbuf = ncl_vinvalbuf; 1206 } 1207 vfs_getnewfsid(mp); 1208 nmp->nm_mountp = mp; 1209 mtx_init(&nmp->nm_mtx, "NFSmount lock", NULL, MTX_DEF | MTX_DUPOK); 1210 nmp->nm_negnametimeo = negnametimeo; 1211 1212 nfs_decode_args(mp, nmp, argp, hst, cred, td); 1213 1214 /* 1215 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too 1216 * high, depending on whether we end up with negative offsets in 1217 * the client or server somewhere. 2GB-1 may be safer. 1218 * 1219 * For V3, ncl_fsinfo will adjust this as necessary. Assume maximum 1220 * that we can handle until we find out otherwise. 1221 * XXX Our "safe" limit on the client is what we can store in our 1222 * buffer cache using signed(!) block numbers. 1223 */ 1224 if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) 1225 nmp->nm_maxfilesize = 0xffffffffLL; 1226 else 1227 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 1228 1229 nmp->nm_timeo = NFS_TIMEO; 1230 nmp->nm_retry = NFS_RETRANS; 1231 if ((argp->flags & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) == 0) { 1232 nmp->nm_wsize = NFS_WSIZE; 1233 nmp->nm_rsize = NFS_RSIZE; 1234 nmp->nm_readdirsize = NFS_READDIRSIZE; 1235 } 1236 nmp->nm_wcommitsize = hibufspace / (desiredvnodes / 1000); 1237 nmp->nm_numgrps = NFS_MAXGRPS; 1238 nmp->nm_readahead = NFS_DEFRAHEAD; 1239 nmp->nm_tprintf_delay = nfs_tprintf_delay; 1240 if (nmp->nm_tprintf_delay < 0) 1241 nmp->nm_tprintf_delay = 0; 1242 nmp->nm_tprintf_initial_delay = nfs_tprintf_initial_delay; 1243 if (nmp->nm_tprintf_initial_delay < 0) 1244 nmp->nm_tprintf_initial_delay = 0; 1245 nmp->nm_fhsize = argp->fhsize; 1246 if (nmp->nm_fhsize > 0) 1247 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize); 1248 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); 1249 nmp->nm_nam = nam; 1250 /* Set up the sockets and per-host congestion */ 1251 nmp->nm_sotype = argp->sotype; 1252 nmp->nm_soproto = argp->proto; 1253 nmp->nm_sockreq.nr_prog = NFS_PROG; 1254 if ((argp->flags & NFSMNT_NFSV4)) 1255 nmp->nm_sockreq.nr_vers = NFS_VER4; 1256 else if ((argp->flags & NFSMNT_NFSV3)) 1257 nmp->nm_sockreq.nr_vers = NFS_VER3; 1258 else 1259 nmp->nm_sockreq.nr_vers = NFS_VER2; 1260 1261 1262 if ((error = newnfs_connect(nmp, &nmp->nm_sockreq, cred, td, 0))) 1263 goto bad; 1264 1265 /* 1266 * A reference count is needed on the nfsnode representing the 1267 * remote root. If this object is not persistent, then backward 1268 * traversals of the mount point (i.e. "..") will not work if 1269 * the nfsnode gets flushed out of the cache. Ufs does not have 1270 * this problem, because one can identify root inodes by their 1271 * number == ROOTINO (2). 1272 */ 1273 if (nmp->nm_fhsize == 0 && (nmp->nm_flag & NFSMNT_NFSV4) && 1274 nmp->nm_dirpathlen > 0) { 1275 /* 1276 * If the fhsize on the mount point == 0 for V4, the mount 1277 * path needs to be looked up. 1278 */ 1279 trycnt = 3; 1280 do { 1281 error = nfsrpc_getdirpath(nmp, NFSMNT_DIRPATH(nmp), 1282 cred, td); 1283 if (error) 1284 (void) nfs_catnap(PZERO, error, "nfsgetdirp"); 1285 } while (error && --trycnt > 0); 1286 if (error) { 1287 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0); 1288 goto bad; 1289 } 1290 } 1291 if (nmp->nm_fhsize > 0) { 1292 /* 1293 * Set f_iosize to NFS_DIRBLKSIZ so that bo_bsize gets set 1294 * non-zero for the root vnode. f_iosize will be set correctly 1295 * by nfs_statfs() before any I/O occurs. 1296 */ 1297 mp->mnt_stat.f_iosize = NFS_DIRBLKSIZ; 1298 error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, 1299 LK_EXCLUSIVE); 1300 if (error) 1301 goto bad; 1302 *vpp = NFSTOV(np); 1303 1304 /* 1305 * Get file attributes and transfer parameters for the 1306 * mountpoint. This has the side effect of filling in 1307 * (*vpp)->v_type with the correct value. 1308 */ 1309 ret = nfsrpc_getattrnovp(nmp, nmp->nm_fh, nmp->nm_fhsize, 1, 1310 cred, td, &nfsva, NULL); 1311 if (ret) { 1312 /* 1313 * Just set default values to get things going. 1314 */ 1315 NFSBZERO((caddr_t)&nfsva, sizeof (struct nfsvattr)); 1316 nfsva.na_vattr.va_type = VDIR; 1317 nfsva.na_vattr.va_mode = 0777; 1318 nfsva.na_vattr.va_nlink = 100; 1319 nfsva.na_vattr.va_uid = (uid_t)0; 1320 nfsva.na_vattr.va_gid = (gid_t)0; 1321 nfsva.na_vattr.va_fileid = 2; 1322 nfsva.na_vattr.va_gen = 1; 1323 nfsva.na_vattr.va_blocksize = NFS_FABLKSIZE; 1324 nfsva.na_vattr.va_size = 512 * 1024; 1325 } 1326 (void) nfscl_loadattrcache(vpp, &nfsva, NULL, NULL, 0, 1); 1327 if (argp->flags & NFSMNT_NFSV3) 1328 ncl_fsinfo(nmp, *vpp, cred, td); 1329 1330 /* 1331 * Lose the lock but keep the ref. 1332 */ 1333 VOP_UNLOCK(*vpp, 0); 1334 return (0); 1335 } 1336 error = EIO; 1337 1338bad: 1339 newnfs_disconnect(&nmp->nm_sockreq); 1340 crfree(nmp->nm_sockreq.nr_cred); 1341 mtx_destroy(&nmp->nm_sockreq.nr_mtx); 1342 mtx_destroy(&nmp->nm_mtx); 1343 FREE(nmp, M_NEWNFSMNT); 1344 FREE(nam, M_SONAME); 1345 return (error); 1346} 1347 1348/* 1349 * unmount system call 1350 */ 1351static int 1352nfs_unmount(struct mount *mp, int mntflags) 1353{ 1354 struct thread *td; 1355 struct nfsmount *nmp; 1356 int error, flags = 0, trycnt = 0; 1357 1358 td = curthread; 1359 1360 if (mntflags & MNT_FORCE) 1361 flags |= FORCECLOSE; 1362 nmp = VFSTONFS(mp); 1363 /* 1364 * Goes something like this.. 1365 * - Call vflush() to clear out vnodes for this filesystem 1366 * - Close the socket 1367 * - Free up the data structures 1368 */ 1369 /* In the forced case, cancel any outstanding requests. */ 1370 if (mntflags & MNT_FORCE) { 1371 error = newnfs_nmcancelreqs(nmp); 1372 if (error) 1373 goto out; 1374 /* For a forced close, get rid of the renew thread now */ 1375 nfscl_umount(nmp, td); 1376 } 1377 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */ 1378 do { 1379 error = vflush(mp, 1, flags, td); 1380 if ((mntflags & MNT_FORCE) && error != 0 && ++trycnt < 30) 1381 (void) nfs_catnap(PSOCK, error, "newndm"); 1382 } while ((mntflags & MNT_FORCE) && error != 0 && trycnt < 30); 1383 if (error) 1384 goto out; 1385 1386 /* 1387 * We are now committed to the unmount. 1388 */ 1389 if ((mntflags & MNT_FORCE) == 0) 1390 nfscl_umount(nmp, td); 1391 newnfs_disconnect(&nmp->nm_sockreq); 1392 crfree(nmp->nm_sockreq.nr_cred); 1393 FREE(nmp->nm_nam, M_SONAME); 1394 1395 mtx_destroy(&nmp->nm_sockreq.nr_mtx); 1396 mtx_destroy(&nmp->nm_mtx); 1397 FREE(nmp, M_NEWNFSMNT); 1398out: 1399 return (error); 1400} 1401 1402/* 1403 * Return root of a filesystem 1404 */ 1405static int 1406nfs_root(struct mount *mp, int flags, struct vnode **vpp) 1407{ 1408 struct vnode *vp; 1409 struct nfsmount *nmp; 1410 struct nfsnode *np; 1411 int error; 1412 1413 nmp = VFSTONFS(mp); 1414 error = ncl_nget(mp, nmp->nm_fh, nmp->nm_fhsize, &np, flags); 1415 if (error) 1416 return error; 1417 vp = NFSTOV(np); 1418 /* 1419 * Get transfer parameters and attributes for root vnode once. 1420 */ 1421 mtx_lock(&nmp->nm_mtx); 1422 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp)) { 1423 mtx_unlock(&nmp->nm_mtx); 1424 ncl_fsinfo(nmp, vp, curthread->td_ucred, curthread); 1425 } else 1426 mtx_unlock(&nmp->nm_mtx); 1427 if (vp->v_type == VNON) 1428 vp->v_type = VDIR; 1429 vp->v_vflag |= VV_ROOT; 1430 *vpp = vp; 1431 return (0); 1432} 1433 1434/* 1435 * Flush out the buffer cache 1436 */ 1437/* ARGSUSED */ 1438static int 1439nfs_sync(struct mount *mp, int waitfor) 1440{ 1441 struct vnode *vp, *mvp; 1442 struct thread *td; 1443 int error, allerror = 0; 1444 1445 td = curthread; 1446 1447 /* 1448 * Force stale buffer cache information to be flushed. 1449 */ 1450 MNT_ILOCK(mp); 1451loop: 1452 MNT_VNODE_FOREACH(vp, mp, mvp) { 1453 VI_LOCK(vp); 1454 MNT_IUNLOCK(mp); 1455 /* XXX Racy bv_cnt check. */ 1456 if (VOP_ISLOCKED(vp) || vp->v_bufobj.bo_dirty.bv_cnt == 0 || 1457 waitfor == MNT_LAZY) { 1458 VI_UNLOCK(vp); 1459 MNT_ILOCK(mp); 1460 continue; 1461 } 1462 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { 1463 MNT_ILOCK(mp); 1464 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 1465 goto loop; 1466 } 1467 error = VOP_FSYNC(vp, waitfor, td); 1468 if (error) 1469 allerror = error; 1470 VOP_UNLOCK(vp, 0); 1471 vrele(vp); 1472 1473 MNT_ILOCK(mp); 1474 } 1475 MNT_IUNLOCK(mp); 1476 return (allerror); 1477} 1478 1479static int 1480nfs_sysctl(struct mount *mp, fsctlop_t op, struct sysctl_req *req) 1481{ 1482 struct nfsmount *nmp = VFSTONFS(mp); 1483 struct vfsquery vq; 1484 int error; 1485 1486 bzero(&vq, sizeof(vq)); 1487 switch (op) { 1488#if 0 1489 case VFS_CTL_NOLOCKS: 1490 val = (nmp->nm_flag & NFSMNT_NOLOCKS) ? 1 : 0; 1491 if (req->oldptr != NULL) { 1492 error = SYSCTL_OUT(req, &val, sizeof(val)); 1493 if (error) 1494 return (error); 1495 } 1496 if (req->newptr != NULL) { 1497 error = SYSCTL_IN(req, &val, sizeof(val)); 1498 if (error) 1499 return (error); 1500 if (val) 1501 nmp->nm_flag |= NFSMNT_NOLOCKS; 1502 else 1503 nmp->nm_flag &= ~NFSMNT_NOLOCKS; 1504 } 1505 break; 1506#endif 1507 case VFS_CTL_QUERY: 1508 mtx_lock(&nmp->nm_mtx); 1509 if (nmp->nm_state & NFSSTA_TIMEO) 1510 vq.vq_flags |= VQ_NOTRESP; 1511 mtx_unlock(&nmp->nm_mtx); 1512#if 0 1513 if (!(nmp->nm_flag & NFSMNT_NOLOCKS) && 1514 (nmp->nm_state & NFSSTA_LOCKTIMEO)) 1515 vq.vq_flags |= VQ_NOTRESPLOCK; 1516#endif 1517 error = SYSCTL_OUT(req, &vq, sizeof(vq)); 1518 break; 1519 case VFS_CTL_TIMEO: 1520 if (req->oldptr != NULL) { 1521 error = SYSCTL_OUT(req, &nmp->nm_tprintf_initial_delay, 1522 sizeof(nmp->nm_tprintf_initial_delay)); 1523 if (error) 1524 return (error); 1525 } 1526 if (req->newptr != NULL) { 1527 error = vfs_suser(mp, req->td); 1528 if (error) 1529 return (error); 1530 error = SYSCTL_IN(req, &nmp->nm_tprintf_initial_delay, 1531 sizeof(nmp->nm_tprintf_initial_delay)); 1532 if (error) 1533 return (error); 1534 if (nmp->nm_tprintf_initial_delay < 0) 1535 nmp->nm_tprintf_initial_delay = 0; 1536 } 1537 break; 1538 default: 1539 return (ENOTSUP); 1540 } 1541 return (0); 1542} 1543 1544/* 1545 * Extract the information needed by the nlm from the nfs vnode. 1546 */ 1547static void 1548nfs_getnlminfo(struct vnode *vp, uint8_t *fhp, size_t *fhlenp, 1549 struct sockaddr_storage *sp, int *is_v3p, off_t *sizep, 1550 struct timeval *timeop) 1551{ 1552 struct nfsmount *nmp; 1553 struct nfsnode *np = VTONFS(vp); 1554 1555 nmp = VFSTONFS(vp->v_mount); 1556 if (fhlenp != NULL) 1557 *fhlenp = (size_t)np->n_fhp->nfh_len; 1558 if (fhp != NULL) 1559 bcopy(np->n_fhp->nfh_fh, fhp, np->n_fhp->nfh_len); 1560 if (sp != NULL) 1561 bcopy(nmp->nm_nam, sp, min(nmp->nm_nam->sa_len, sizeof(*sp))); 1562 if (is_v3p != NULL) 1563 *is_v3p = NFS_ISV3(vp); 1564 if (sizep != NULL) 1565 *sizep = np->n_size; 1566 if (timeop != NULL) { 1567 timeop->tv_sec = nmp->nm_timeo / NFS_HZ; 1568 timeop->tv_usec = (nmp->nm_timeo % NFS_HZ) * (1000000 / NFS_HZ); 1569 } 1570} 1571 1572