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 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)nfs_vfsops.c 8.12 (Berkeley) 5/20/95 37 * $FreeBSD: src/sys/nfs/nfs_vfsops.c,v 1.91.2.7 2003/01/27 20:04:08 dillon Exp $ 38 * $DragonFly: src/sys/vfs/nfs/nfs_vfsops.c,v 1.10 2003/11/15 21:05:44 dillon Exp $ 39 */ 40 41 #include "opt_bootp.h" 42 43 #include <sys/param.h> 44 #include <sys/sockio.h> 45 #include <sys/proc.h> 46 #include <sys/vnode.h> 47 #include <sys/kernel.h> 48 #include <sys/sysctl.h> 49 #include <sys/malloc.h> 50 #include <sys/mount.h> 51 #include <sys/mbuf.h> 52 #include <sys/socket.h> 53 #include <sys/socketvar.h> 54 #include <sys/systm.h> 55 56 #include <vm/vm.h> 57 #include <vm/vm_extern.h> 58 #include <vm/vm_zone.h> 59 60 #include <net/if.h> 61 #include <net/route.h> 62 #include <netinet/in.h> 63 64 #include "rpcv2.h" 65 #include "nfsproto.h" 66 #include "nfs.h" 67 #include "nfsmount.h" 68 #include "nfsnode.h" 69 #include "xdr_subs.h" 70 #include "nfsm_subs.h" 71 #include "nfsdiskless.h" 72 #include "nqnfs.h" 73 74 extern int nfs_mountroot (struct mount *mp); 75 76 extern int nfs_ticks; 77 78 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header"); 79 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle"); 80 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure"); 81 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data"); 82 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor"); 83 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure"); 84 MALLOC_DEFINE(M_NQLEASE, "NQNFS Lease", "Nqnfs lease"); 85 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables"); 86 87 vm_zone_t nfsmount_zone; 88 89 struct nfsstats nfsstats; 90 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem"); 91 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD, 92 &nfsstats, nfsstats, ""); 93 static int nfs_ip_paranoia = 1; 94 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW, 95 &nfs_ip_paranoia, 0, ""); 96 #ifdef NFS_DEBUG 97 int nfs_debug; 98 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, ""); 99 #endif 100 101 static int nfs_iosize (struct nfsmount *nmp); 102 static void nfs_decode_args (struct nfsmount *nmp, 103 struct nfs_args *argp); 104 static int mountnfs (struct nfs_args *,struct mount *, 105 struct sockaddr *,char *,char *,struct vnode **); 106 static int nfs_mount ( struct mount *mp, char *path, caddr_t data, 107 struct nameidata *ndp, struct thread *td); 108 static int nfs_unmount ( struct mount *mp, int mntflags, 109 struct thread *td); 110 static int nfs_root ( struct mount *mp, struct vnode **vpp); 111 static int nfs_statfs ( struct mount *mp, struct statfs *sbp, 112 struct thread *td); 113 static int nfs_sync ( struct mount *mp, int waitfor, 114 struct thread *td); 115 116 /* 117 * nfs vfs operations. 118 */ 119 static struct vfsops nfs_vfsops = { 120 nfs_mount, 121 vfs_stdstart, 122 nfs_unmount, 123 nfs_root, 124 vfs_stdquotactl, 125 nfs_statfs, 126 nfs_sync, 127 vfs_stdvget, 128 vfs_stdfhtovp, /* shouldn't happen */ 129 vfs_stdcheckexp, 130 vfs_stdvptofh, /* shouldn't happen */ 131 nfs_init, 132 nfs_uninit, 133 vfs_stdextattrctl, 134 }; 135 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK); 136 137 /* 138 * This structure must be filled in by a primary bootstrap or bootstrap 139 * server for a diskless/dataless machine. It is initialized below just 140 * to ensure that it is allocated to initialized data (.data not .bss). 141 */ 142 struct nfs_diskless nfs_diskless = { { { 0 } } }; 143 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } }; 144 int nfs_diskless_valid = 0; 145 146 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD, 147 &nfs_diskless_valid, 0, ""); 148 149 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD, 150 nfsv3_diskless.root_hostnam, 0, ""); 151 152 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD, 153 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr, 154 "%Ssockaddr_in", ""); 155 156 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD, 157 nfsv3_diskless.swap_hostnam, 0, ""); 158 159 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD, 160 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr, 161 "%Ssockaddr_in",""); 162 163 164 void nfsargs_ntoh (struct nfs_args *); 165 static int nfs_mountdiskless (char *, char *, int, 166 struct sockaddr_in *, struct nfs_args *, 167 struct thread *, struct vnode **, 168 struct mount **); 169 static void nfs_convert_diskless (void); 170 static void nfs_convert_oargs (struct nfs_args *args, 171 struct onfs_args *oargs); 172 173 static int 174 nfs_iosize(nmp) 175 struct nfsmount* nmp; 176 { 177 int iosize; 178 179 /* 180 * Calculate the size used for io buffers. Use the larger 181 * of the two sizes to minimise nfs requests but make sure 182 * that it is at least one VM page to avoid wasting buffer 183 * space. 184 */ 185 iosize = max(nmp->nm_rsize, nmp->nm_wsize); 186 if (iosize < PAGE_SIZE) iosize = PAGE_SIZE; 187 return iosize; 188 } 189 190 static void 191 nfs_convert_oargs(args, oargs) 192 struct nfs_args *args; 193 struct onfs_args *oargs; 194 { 195 args->version = NFS_ARGSVERSION; 196 args->addr = oargs->addr; 197 args->addrlen = oargs->addrlen; 198 args->sotype = oargs->sotype; 199 args->proto = oargs->proto; 200 args->fh = oargs->fh; 201 args->fhsize = oargs->fhsize; 202 args->flags = oargs->flags; 203 args->wsize = oargs->wsize; 204 args->rsize = oargs->rsize; 205 args->readdirsize = oargs->readdirsize; 206 args->timeo = oargs->timeo; 207 args->retrans = oargs->retrans; 208 args->maxgrouplist = oargs->maxgrouplist; 209 args->readahead = oargs->readahead; 210 args->leaseterm = oargs->leaseterm; 211 args->deadthresh = oargs->deadthresh; 212 args->hostname = oargs->hostname; 213 } 214 215 static void 216 nfs_convert_diskless() 217 { 218 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif, 219 sizeof(struct ifaliasreq)); 220 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway, 221 sizeof(struct sockaddr_in)); 222 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args); 223 nfsv3_diskless.swap_fhsize = NFSX_V2FH; 224 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH); 225 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr, 226 sizeof(struct sockaddr_in)); 227 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN); 228 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks; 229 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred, 230 sizeof(struct ucred)); 231 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args); 232 nfsv3_diskless.root_fhsize = NFSX_V2FH; 233 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH); 234 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr, 235 sizeof(struct sockaddr_in)); 236 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN); 237 nfsv3_diskless.root_time = nfs_diskless.root_time; 238 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam, 239 MAXHOSTNAMELEN); 240 nfs_diskless_valid = 3; 241 } 242 243 /* 244 * nfs statfs call 245 */ 246 int 247 nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 248 { 249 struct vnode *vp; 250 struct nfs_statfs *sfp; 251 caddr_t cp; 252 u_int32_t *tl; 253 int32_t t1, t2; 254 caddr_t bpos, dpos, cp2; 255 struct nfsmount *nmp = VFSTONFS(mp); 256 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr; 257 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 258 struct ucred *cred; 259 struct nfsnode *np; 260 u_quad_t tquad; 261 262 #ifndef nolint 263 sfp = (struct nfs_statfs *)0; 264 #endif 265 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 266 if (error) 267 return (error); 268 vp = NFSTOV(np); 269 cred = crget(); 270 cred->cr_ngroups = 1; 271 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 272 (void)nfs_fsinfo(nmp, vp, td); 273 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 274 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3)); 275 nfsm_fhtom(vp, v3); 276 nfsm_request(vp, NFSPROC_FSSTAT, td, cred); 277 if (v3) 278 nfsm_postop_attr(vp, retattr); 279 if (error) { 280 if (mrep != NULL) 281 m_freem(mrep); 282 goto nfsmout; 283 } 284 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 285 sbp->f_flags = nmp->nm_flag; 286 sbp->f_iosize = nfs_iosize(nmp); 287 if (v3) { 288 sbp->f_bsize = NFS_FABLKSIZE; 289 tquad = fxdr_hyper(&sfp->sf_tbytes); 290 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 291 tquad = fxdr_hyper(&sfp->sf_fbytes); 292 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 293 tquad = fxdr_hyper(&sfp->sf_abytes); 294 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 295 sbp->f_files = (fxdr_unsigned(int32_t, 296 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff); 297 sbp->f_ffree = (fxdr_unsigned(int32_t, 298 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff); 299 } else { 300 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 301 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 302 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 303 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 304 sbp->f_files = 0; 305 sbp->f_ffree = 0; 306 } 307 if (sbp != &mp->mnt_stat) { 308 sbp->f_type = mp->mnt_vfc->vfc_typenum; 309 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN); 310 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); 311 } 312 m_freem(mrep); 313 nfsmout: 314 vput(vp); 315 crfree(cred); 316 return (error); 317 } 318 319 /* 320 * nfs version 3 fsinfo rpc call 321 */ 322 int 323 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td) 324 { 325 struct nfsv3_fsinfo *fsp; 326 caddr_t cp; 327 int32_t t1, t2; 328 u_int32_t *tl, pref, max; 329 caddr_t bpos, dpos, cp2; 330 int error = 0, retattr; 331 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 332 u_int64_t maxfsize; 333 334 nfsstats.rpccnt[NFSPROC_FSINFO]++; 335 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1)); 336 nfsm_fhtom(vp, 1); 337 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ)); 338 nfsm_postop_attr(vp, retattr); 339 if (!error) { 340 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 341 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 342 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 343 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 344 ~(NFS_FABLKSIZE - 1); 345 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 346 if (max < nmp->nm_wsize && max > 0) { 347 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1); 348 if (nmp->nm_wsize == 0) 349 nmp->nm_wsize = max; 350 } 351 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 352 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 353 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 354 ~(NFS_FABLKSIZE - 1); 355 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 356 if (max < nmp->nm_rsize && max > 0) { 357 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1); 358 if (nmp->nm_rsize == 0) 359 nmp->nm_rsize = max; 360 } 361 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 362 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ) 363 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) & 364 ~(NFS_DIRBLKSIZ - 1); 365 if (max < nmp->nm_readdirsize && max > 0) { 366 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1); 367 if (nmp->nm_readdirsize == 0) 368 nmp->nm_readdirsize = max; 369 } 370 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 371 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 372 nmp->nm_maxfilesize = maxfsize; 373 nmp->nm_state |= NFSSTA_GOTFSINFO; 374 } 375 m_freem(mrep); 376 nfsmout: 377 return (error); 378 } 379 380 /* 381 * Mount a remote root fs via. nfs. This depends on the info in the 382 * nfs_diskless structure that has been filled in properly by some primary 383 * bootstrap. 384 * It goes something like this: 385 * - do enough of "ifconfig" by calling ifioctl() so that the system 386 * can talk to the server 387 * - If nfs_diskless.mygateway is filled in, use that address as 388 * a default gateway. 389 * - build the rootfs mount point and call mountnfs() to do the rest. 390 */ 391 int 392 nfs_mountroot(mp) 393 struct mount *mp; 394 { 395 struct mount *swap_mp; 396 struct nfsv3_diskless *nd = &nfsv3_diskless; 397 struct socket *so; 398 struct vnode *vp; 399 struct thread *td = curthread; /* XXX */ 400 int error, i; 401 u_long l; 402 char buf[128]; 403 404 #if defined(BOOTP_NFSROOT) && defined(BOOTP) 405 bootpc_init(); /* use bootp to get nfs_diskless filled in */ 406 #endif 407 408 /* 409 * XXX time must be non-zero when we init the interface or else 410 * the arp code will wedge... 411 */ 412 while (time_second == 0) 413 tsleep(&time_second, 0, "arpkludge", 10); 414 415 if (nfs_diskless_valid==1) 416 nfs_convert_diskless(); 417 418 /* 419 * XXX splnet, so networks will receive... 420 */ 421 splnet(); 422 423 #ifdef notyet 424 /* Set up swap credentials. */ 425 proc0.p_ucred->cr_uid = ntohl(nd->swap_ucred.cr_uid); 426 proc0.p_ucred->cr_gid = ntohl(nd->swap_ucred.cr_gid); 427 if ((proc0.p_ucred->cr_ngroups = ntohs(nd->swap_ucred.cr_ngroups)) > 428 NGROUPS) 429 proc0.p_ucred->cr_ngroups = NGROUPS; 430 for (i = 0; i < proc0.p_ucred->cr_ngroups; i++) 431 proc0.p_ucred->cr_groups[i] = ntohl(nd->swap_ucred.cr_groups[i]); 432 #endif 433 434 /* 435 * Do enough of ifconfig(8) so that the critical net interface can 436 * talk to the server. 437 */ 438 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td); 439 if (error) 440 panic("nfs_mountroot: socreate(%04x): %d", 441 nd->myif.ifra_addr.sa_family, error); 442 443 #if 0 /* XXX Bad idea */ 444 /* 445 * We might not have been told the right interface, so we pass 446 * over the first ten interfaces of the same kind, until we get 447 * one of them configured. 448 */ 449 450 for (i = strlen(nd->myif.ifra_name) - 1; 451 nd->myif.ifra_name[i] >= '0' && 452 nd->myif.ifra_name[i] <= '9'; 453 nd->myif.ifra_name[i] ++) { 454 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); 455 if(!error) 456 break; 457 } 458 #endif 459 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); 460 if (error) 461 panic("nfs_mountroot: SIOCAIFADDR: %d", error); 462 soclose(so); 463 464 /* 465 * If the gateway field is filled in, set it as the default route. 466 */ 467 if (nd->mygateway.sin_len != 0) { 468 struct sockaddr_in mask, sin; 469 470 bzero((caddr_t)&mask, sizeof(mask)); 471 sin = mask; 472 sin.sin_family = AF_INET; 473 sin.sin_len = sizeof(sin); 474 error = rtrequest(RTM_ADD, (struct sockaddr *)&sin, 475 (struct sockaddr *)&nd->mygateway, 476 (struct sockaddr *)&mask, 477 RTF_UP | RTF_GATEWAY, (struct rtentry **)0); 478 if (error) 479 panic("nfs_mountroot: RTM_ADD: %d", error); 480 } 481 482 /* 483 * Create the rootfs mount point. 484 */ 485 nd->root_args.fh = nd->root_fh; 486 nd->root_args.fhsize = nd->root_fhsize; 487 l = ntohl(nd->root_saddr.sin_addr.s_addr); 488 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 489 (l >> 24) & 0xff, (l >> 16) & 0xff, 490 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam); 491 printf("NFS ROOT: %s\n",buf); 492 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY, 493 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) { 494 if (swap_mp) { 495 mp->mnt_vfc->vfc_refcount--; 496 free(swap_mp, M_MOUNT); 497 } 498 return (error); 499 } 500 501 swap_mp = NULL; 502 if (nd->swap_nblks) { 503 504 /* Convert to DEV_BSIZE instead of Kilobyte */ 505 nd->swap_nblks *= 2; 506 507 /* 508 * Create a fake mount point just for the swap vnode so that the 509 * swap file can be on a different server from the rootfs. 510 */ 511 nd->swap_args.fh = nd->swap_fh; 512 nd->swap_args.fhsize = nd->swap_fhsize; 513 l = ntohl(nd->swap_saddr.sin_addr.s_addr); 514 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 515 (l >> 24) & 0xff, (l >> 16) & 0xff, 516 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam); 517 printf("NFS SWAP: %s\n",buf); 518 if ((error = nfs_mountdiskless(buf, "/swap", 0, 519 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) 520 return (error); 521 vfs_unbusy(swap_mp, td); 522 523 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size = 524 nd->swap_nblks * DEV_BSIZE ; 525 526 /* 527 * Since the swap file is not the root dir of a file system, 528 * hack it to a regular file. 529 */ 530 vp->v_type = VREG; 531 vp->v_flag = 0; 532 VREF(vp); 533 swaponvp(td, vp, NODEV, nd->swap_nblks); 534 } 535 536 mp->mnt_flag |= MNT_ROOTFS; 537 mp->mnt_vnodecovered = NULLVP; 538 rootvp = vp; 539 vfs_unbusy(mp, td); 540 541 /* 542 * This is not really an nfs issue, but it is much easier to 543 * set hostname here and then let the "/etc/rc.xxx" files 544 * mount the right /var based upon its preset value. 545 */ 546 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN); 547 hostname[MAXHOSTNAMELEN - 1] = '\0'; 548 for (i = 0; i < MAXHOSTNAMELEN; i++) 549 if (hostname[i] == '\0') 550 break; 551 inittodr(ntohl(nd->root_time)); 552 return (0); 553 } 554 555 /* 556 * Internal version of mount system call for diskless setup. 557 */ 558 static int 559 nfs_mountdiskless(char *path, char *which, int mountflag, 560 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td, 561 struct vnode **vpp, struct mount **mpp) 562 { 563 struct mount *mp; 564 struct sockaddr *nam; 565 int error; 566 int didalloc = 0; 567 568 mp = *mpp; 569 570 if (mp == NULL) { 571 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) { 572 printf("nfs_mountroot: NFS not configured"); 573 return (error); 574 } 575 didalloc = 1; 576 } 577 578 mp->mnt_kern_flag = 0; 579 mp->mnt_flag = mountflag; 580 nam = dup_sockaddr((struct sockaddr *)sin, 1); 581 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) { 582 printf("nfs_mountroot: mount %s on %s: %d", path, which, error); 583 mp->mnt_vfc->vfc_refcount--; 584 vfs_unbusy(mp, td); 585 if (didalloc) 586 free(mp, M_MOUNT); 587 FREE(nam, M_SONAME); 588 return (error); 589 } 590 (void) copystr(which, mp->mnt_stat.f_mntonname, MNAMELEN - 1, 0); 591 *mpp = mp; 592 return (0); 593 } 594 595 static void 596 nfs_decode_args(nmp, argp) 597 struct nfsmount *nmp; 598 struct nfs_args *argp; 599 { 600 int s; 601 int adjsock; 602 int maxio; 603 604 s = splnet(); 605 /* 606 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 607 * no sense in that context. 608 */ 609 if (argp->sotype == SOCK_STREAM) 610 nmp->nm_flag &= ~NFSMNT_NOCONN; 611 612 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */ 613 if ((argp->flags & NFSMNT_NFSV3) == 0) 614 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 615 616 /* Re-bind if rsrvd port requested and wasn't on one */ 617 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 618 && (argp->flags & NFSMNT_RESVPORT); 619 /* Also re-bind if we're switching to/from a connected UDP socket */ 620 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 621 (argp->flags & NFSMNT_NOCONN)); 622 623 /* Update flags atomically. Don't change the lock bits. */ 624 nmp->nm_flag = argp->flags | nmp->nm_flag; 625 splx(s); 626 627 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 628 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 629 if (nmp->nm_timeo < NFS_MINTIMEO) 630 nmp->nm_timeo = NFS_MINTIMEO; 631 else if (nmp->nm_timeo > NFS_MAXTIMEO) 632 nmp->nm_timeo = NFS_MAXTIMEO; 633 } 634 635 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 636 nmp->nm_retry = argp->retrans; 637 if (nmp->nm_retry > NFS_MAXREXMIT) 638 nmp->nm_retry = NFS_MAXREXMIT; 639 } 640 641 if (argp->flags & NFSMNT_NFSV3) { 642 if (argp->sotype == SOCK_DGRAM) 643 maxio = NFS_MAXDGRAMDATA; 644 else 645 maxio = NFS_MAXDATA; 646 } else 647 maxio = NFS_V2MAXDATA; 648 649 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 650 nmp->nm_wsize = argp->wsize; 651 /* Round down to multiple of blocksize */ 652 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 653 if (nmp->nm_wsize <= 0) 654 nmp->nm_wsize = NFS_FABLKSIZE; 655 } 656 if (nmp->nm_wsize > maxio) 657 nmp->nm_wsize = maxio; 658 if (nmp->nm_wsize > MAXBSIZE) 659 nmp->nm_wsize = MAXBSIZE; 660 661 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 662 nmp->nm_rsize = argp->rsize; 663 /* Round down to multiple of blocksize */ 664 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 665 if (nmp->nm_rsize <= 0) 666 nmp->nm_rsize = NFS_FABLKSIZE; 667 } 668 if (nmp->nm_rsize > maxio) 669 nmp->nm_rsize = maxio; 670 if (nmp->nm_rsize > MAXBSIZE) 671 nmp->nm_rsize = MAXBSIZE; 672 673 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 674 nmp->nm_readdirsize = argp->readdirsize; 675 } 676 if (nmp->nm_readdirsize > maxio) 677 nmp->nm_readdirsize = maxio; 678 if (nmp->nm_readdirsize > nmp->nm_rsize) 679 nmp->nm_readdirsize = nmp->nm_rsize; 680 681 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0) 682 nmp->nm_acregmin = argp->acregmin; 683 else 684 nmp->nm_acregmin = NFS_MINATTRTIMO; 685 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0) 686 nmp->nm_acregmax = argp->acregmax; 687 else 688 nmp->nm_acregmax = NFS_MAXATTRTIMO; 689 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0) 690 nmp->nm_acdirmin = argp->acdirmin; 691 else 692 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO; 693 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0) 694 nmp->nm_acdirmax = argp->acdirmax; 695 else 696 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO; 697 if (nmp->nm_acdirmin > nmp->nm_acdirmax) 698 nmp->nm_acdirmin = nmp->nm_acdirmax; 699 if (nmp->nm_acregmin > nmp->nm_acregmax) 700 nmp->nm_acregmin = nmp->nm_acregmax; 701 702 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) { 703 if (argp->maxgrouplist <= NFS_MAXGRPS) 704 nmp->nm_numgrps = argp->maxgrouplist; 705 else 706 nmp->nm_numgrps = NFS_MAXGRPS; 707 } 708 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) { 709 if (argp->readahead <= NFS_MAXRAHEAD) 710 nmp->nm_readahead = argp->readahead; 711 else 712 nmp->nm_readahead = NFS_MAXRAHEAD; 713 } 714 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2) { 715 if (argp->leaseterm <= NQ_MAXLEASE) 716 nmp->nm_leaseterm = argp->leaseterm; 717 else 718 nmp->nm_leaseterm = NQ_MAXLEASE; 719 } 720 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) { 721 if (argp->deadthresh <= NQ_NEVERDEAD) 722 nmp->nm_deadthresh = argp->deadthresh; 723 else 724 nmp->nm_deadthresh = NQ_NEVERDEAD; 725 } 726 727 adjsock |= ((nmp->nm_sotype != argp->sotype) || 728 (nmp->nm_soproto != argp->proto)); 729 nmp->nm_sotype = argp->sotype; 730 nmp->nm_soproto = argp->proto; 731 732 if (nmp->nm_so && adjsock) { 733 nfs_safedisconnect(nmp); 734 if (nmp->nm_sotype == SOCK_DGRAM) 735 while (nfs_connect(nmp, (struct nfsreq *)0)) { 736 printf("nfs_args: retrying connect\n"); 737 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0); 738 } 739 } 740 } 741 742 /* 743 * VFS Operations. 744 * 745 * mount system call 746 * It seems a bit dumb to copyinstr() the host and path here and then 747 * bcopy() them in mountnfs(), but I wanted to detect errors before 748 * doing the sockargs() call because sockargs() allocates an mbuf and 749 * an error after that means that I have to release the mbuf. 750 */ 751 /* ARGSUSED */ 752 static int 753 nfs_mount(struct mount *mp, char *path, caddr_t data, 754 struct nameidata *ndp, struct thread *td) 755 { 756 int error; 757 struct nfs_args args; 758 struct sockaddr *nam; 759 struct vnode *vp; 760 char pth[MNAMELEN], hst[MNAMELEN]; 761 size_t len; 762 u_char nfh[NFSX_V3FHMAX]; 763 764 if (path == NULL) { 765 nfs_mountroot(mp); 766 return (0); 767 } 768 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)); 769 if (error) 770 return (error); 771 if (args.version != NFS_ARGSVERSION) { 772 #ifdef COMPAT_PRELITE2 773 /* 774 * If the argument version is unknown, then assume the 775 * caller is a pre-lite2 4.4BSD client and convert its 776 * arguments. 777 */ 778 struct onfs_args oargs; 779 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args)); 780 if (error) 781 return (error); 782 nfs_convert_oargs(&args,&oargs); 783 #else /* !COMPAT_PRELITE2 */ 784 return (EPROGMISMATCH); 785 #endif /* COMPAT_PRELITE2 */ 786 } 787 if (mp->mnt_flag & MNT_UPDATE) { 788 struct nfsmount *nmp = VFSTONFS(mp); 789 790 if (nmp == NULL) 791 return (EIO); 792 /* 793 * When doing an update, we can't change from or to 794 * v3 and/or nqnfs, or change cookie translation 795 */ 796 args.flags = (args.flags & 797 ~(NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)) | 798 (nmp->nm_flag & 799 (NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)); 800 nfs_decode_args(nmp, &args); 801 return (0); 802 } 803 804 /* 805 * Make the nfs_ip_paranoia sysctl serve as the default connection 806 * or no-connection mode for those protocols that support 807 * no-connection mode (the flag will be cleared later for protocols 808 * that do not support no-connection mode). This will allow a client 809 * to receive replies from a different IP then the request was 810 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid), 811 * not 0. 812 */ 813 if (nfs_ip_paranoia == 0) 814 args.flags |= NFSMNT_NOCONN; 815 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) 816 return (EINVAL); 817 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); 818 if (error) 819 return (error); 820 error = copyinstr(path, pth, MNAMELEN-1, &len); 821 if (error) 822 return (error); 823 bzero(&pth[len], MNAMELEN - len); 824 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len); 825 if (error) 826 return (error); 827 bzero(&hst[len], MNAMELEN - len); 828 /* sockargs() call must be after above copyin() calls */ 829 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen); 830 if (error) 831 return (error); 832 args.fh = nfh; 833 error = mountnfs(&args, mp, nam, pth, hst, &vp); 834 return (error); 835 } 836 837 /* 838 * Common code for mount and mountroot 839 */ 840 static int 841 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam, 842 char *pth, char *hst, struct vnode **vpp) 843 { 844 struct nfsmount *nmp; 845 struct nfsnode *np; 846 int error; 847 struct vattr attrs; 848 849 if (mp->mnt_flag & MNT_UPDATE) { 850 nmp = VFSTONFS(mp); 851 /* update paths, file handles, etc, here XXX */ 852 FREE(nam, M_SONAME); 853 return (0); 854 } else { 855 nmp = zalloc(nfsmount_zone); 856 bzero((caddr_t)nmp, sizeof (struct nfsmount)); 857 TAILQ_INIT(&nmp->nm_uidlruhead); 858 TAILQ_INIT(&nmp->nm_bufq); 859 mp->mnt_data = (qaddr_t)nmp; 860 } 861 vfs_getnewfsid(mp); 862 nmp->nm_mountp = mp; 863 if (argp->flags & NFSMNT_NQNFS) 864 /* 865 * We have to set mnt_maxsymlink to a non-zero value so 866 * that COMPAT_43 routines will know that we are setting 867 * the d_type field in directories (and can zero it for 868 * unsuspecting binaries). 869 */ 870 mp->mnt_maxsymlinklen = 1; 871 872 /* 873 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too 874 * high, depending on whether we end up with negative offsets in 875 * the client or server somewhere. 2GB-1 may be safer. 876 * 877 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum 878 * that we can handle until we find out otherwise. 879 * XXX Our "safe" limit on the client is what we can store in our 880 * buffer cache using signed(!) block numbers. 881 */ 882 if ((argp->flags & NFSMNT_NFSV3) == 0) 883 nmp->nm_maxfilesize = 0xffffffffLL; 884 else 885 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 886 887 nmp->nm_timeo = NFS_TIMEO; 888 nmp->nm_retry = NFS_RETRANS; 889 nmp->nm_wsize = NFS_WSIZE; 890 nmp->nm_rsize = NFS_RSIZE; 891 nmp->nm_readdirsize = NFS_READDIRSIZE; 892 nmp->nm_numgrps = NFS_MAXGRPS; 893 nmp->nm_readahead = NFS_DEFRAHEAD; 894 nmp->nm_leaseterm = NQ_DEFLEASE; 895 nmp->nm_deadthresh = NQ_DEADTHRESH; 896 CIRCLEQ_INIT(&nmp->nm_timerhead); 897 nmp->nm_inprog = NULLVP; 898 nmp->nm_fhsize = argp->fhsize; 899 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize); 900 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); 901 bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN); 902 nmp->nm_nam = nam; 903 /* Set up the sockets and per-host congestion */ 904 nmp->nm_sotype = argp->sotype; 905 nmp->nm_soproto = argp->proto; 906 nmp->nm_cred = crhold(proc0.p_ucred); 907 908 nfs_decode_args(nmp, argp); 909 910 /* 911 * For Connection based sockets (TCP,...) defer the connect until 912 * the first request, in case the server is not responding. 913 */ 914 if (nmp->nm_sotype == SOCK_DGRAM && 915 (error = nfs_connect(nmp, (struct nfsreq *)0))) 916 goto bad; 917 918 /* 919 * This is silly, but it has to be set so that vinifod() works. 920 * We do not want to do an nfs_statfs() here since we can get 921 * stuck on a dead server and we are holding a lock on the mount 922 * point. 923 */ 924 mp->mnt_stat.f_iosize = nfs_iosize(nmp); 925 /* 926 * A reference count is needed on the nfsnode representing the 927 * remote root. If this object is not persistent, then backward 928 * traversals of the mount point (i.e. "..") will not work if 929 * the nfsnode gets flushed out of the cache. Ufs does not have 930 * this problem, because one can identify root inodes by their 931 * number == ROOTINO (2). 932 */ 933 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 934 if (error) 935 goto bad; 936 *vpp = NFSTOV(np); 937 938 /* 939 * Get file attributes for the mountpoint. This has the side 940 * effect of filling in (*vpp)->v_type with the correct value. 941 */ 942 VOP_GETATTR(*vpp, &attrs, curthread); 943 944 /* 945 * Lose the lock but keep the ref. 946 */ 947 VOP_UNLOCK(*vpp, 0, curthread); 948 949 return (0); 950 bad: 951 nfs_disconnect(nmp); 952 nfs_free_mount(nmp); 953 FREE(nam, M_SONAME); 954 return (error); 955 } 956 957 /* 958 * unmount system call 959 */ 960 static int 961 nfs_unmount(struct mount *mp, int mntflags, struct thread *td) 962 { 963 struct nfsmount *nmp; 964 int error, flags = 0; 965 966 if (mntflags & MNT_FORCE) 967 flags |= FORCECLOSE; 968 nmp = VFSTONFS(mp); 969 /* 970 * Goes something like this.. 971 * - Call vflush() to clear out vnodes for this file system 972 * - Close the socket 973 * - Free up the data structures 974 */ 975 /* In the forced case, cancel any outstanding requests. */ 976 if (flags & FORCECLOSE) { 977 error = nfs_nmcancelreqs(nmp); 978 if (error) 979 return (error); 980 } 981 /* 982 * Must handshake with nqnfs_clientd() if it is active. 983 */ 984 nmp->nm_state |= NFSSTA_DISMINPROG; 985 while (nmp->nm_inprog != NULLVP) 986 (void) tsleep((caddr_t)&lbolt, 0, "nfsdism", 0); 987 988 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */ 989 error = vflush(mp, 1, flags); 990 if (error) { 991 nmp->nm_state &= ~NFSSTA_DISMINPROG; 992 return (error); 993 } 994 995 /* 996 * We are now committed to the unmount. 997 * For NQNFS, let the server daemon free the nfsmount structure. 998 */ 999 if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) 1000 nmp->nm_state |= NFSSTA_DISMNT; 1001 1002 nfs_disconnect(nmp); 1003 FREE(nmp->nm_nam, M_SONAME); 1004 1005 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0) 1006 nfs_free_mount(nmp); 1007 return (0); 1008 } 1009 1010 void 1011 nfs_free_mount(struct nfsmount *nmp) 1012 { 1013 if (nmp->nm_cred) { 1014 crfree(nmp->nm_cred); 1015 nmp->nm_cred = NULL; 1016 } 1017 zfree(nfsmount_zone, nmp); 1018 } 1019 1020 /* 1021 * Return root of a filesystem 1022 */ 1023 static int 1024 nfs_root(mp, vpp) 1025 struct mount *mp; 1026 struct vnode **vpp; 1027 { 1028 struct vnode *vp; 1029 struct nfsmount *nmp; 1030 struct nfsnode *np; 1031 int error; 1032 1033 nmp = VFSTONFS(mp); 1034 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 1035 if (error) 1036 return (error); 1037 vp = NFSTOV(np); 1038 if (vp->v_type == VNON) 1039 vp->v_type = VDIR; 1040 vp->v_flag = VROOT; 1041 *vpp = vp; 1042 return (0); 1043 } 1044 1045 extern int syncprt; 1046 1047 /* 1048 * Flush out the buffer cache 1049 */ 1050 /* ARGSUSED */ 1051 static int 1052 nfs_sync(struct mount *mp, int waitfor, struct thread *td) 1053 { 1054 struct vnode *vp; 1055 int error, allerror = 0; 1056 1057 /* 1058 * Force stale buffer cache information to be flushed. 1059 */ 1060 loop: 1061 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1062 vp != NULL; 1063 vp = TAILQ_NEXT(vp, v_nmntvnodes)) { 1064 /* 1065 * If the vnode that we are about to sync is no longer 1066 * associated with this mount point, start over. 1067 */ 1068 if (vp->v_mount != mp) 1069 goto loop; 1070 if (VOP_ISLOCKED(vp, NULL) || TAILQ_EMPTY(&vp->v_dirtyblkhd) || 1071 waitfor == MNT_LAZY) 1072 continue; 1073 if (vget(vp, LK_EXCLUSIVE, td)) 1074 goto loop; 1075 error = VOP_FSYNC(vp, waitfor, td); 1076 if (error) 1077 allerror = error; 1078 vput(vp); 1079 } 1080 return (allerror); 1081 } 1082 1083