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 1.6 2003/07/26 21:48:49 rob 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 <nfs/rpcv2.h> 65 #include <nfs/nfsproto.h> 66 #include <nfs/nfs.h> 67 #include <nfs/nfsnode.h> 68 #include <nfs/nfsmount.h> 69 #include <nfs/xdr_subs.h> 70 #include <nfs/nfsm_subs.h> 71 #include <nfs/nfsdiskless.h> 72 #include <nfs/nqnfs.h> 73 74 extern int nfs_mountroot __P((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 __P((struct nfsmount *nmp)); 102 static void nfs_decode_args __P((struct nfsmount *nmp, 103 struct nfs_args *argp)); 104 static int mountnfs __P((struct nfs_args *,struct mount *, 105 struct sockaddr *,char *,char *,struct vnode **)); 106 static int nfs_mount __P(( struct mount *mp, char *path, caddr_t data, 107 struct nameidata *ndp, struct thread *td)); 108 static int nfs_unmount __P(( struct mount *mp, int mntflags, 109 struct thread *td)); 110 static int nfs_root __P(( struct mount *mp, struct vnode **vpp)); 111 static int nfs_statfs __P(( struct mount *mp, struct statfs *sbp, 112 struct thread *td)); 113 static int nfs_sync __P(( 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 __P((struct nfs_args *)); 165 static int nfs_mountdiskless __P((char *, char *, int, 166 struct sockaddr_in *, struct nfs_args *, 167 struct thread *, struct vnode **, 168 struct mount **)); 169 static void nfs_convert_diskless __P((void)); 170 static void nfs_convert_oargs __P((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 nfsm_reqdone; 313 vput(vp); 314 crfree(cred); 315 return (error); 316 } 317 318 /* 319 * nfs version 3 fsinfo rpc call 320 */ 321 int 322 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td) 323 { 324 struct nfsv3_fsinfo *fsp; 325 caddr_t cp; 326 int32_t t1, t2; 327 u_int32_t *tl, pref, max; 328 caddr_t bpos, dpos, cp2; 329 int error = 0, retattr; 330 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 331 u_int64_t maxfsize; 332 333 nfsstats.rpccnt[NFSPROC_FSINFO]++; 334 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1)); 335 nfsm_fhtom(vp, 1); 336 nfsm_request(vp, NFSPROC_FSINFO, td, NFSVPCRED(vp)); 337 nfsm_postop_attr(vp, retattr); 338 if (!error) { 339 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 340 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 341 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 342 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 343 ~(NFS_FABLKSIZE - 1); 344 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 345 if (max < nmp->nm_wsize && max > 0) { 346 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1); 347 if (nmp->nm_wsize == 0) 348 nmp->nm_wsize = max; 349 } 350 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 351 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 352 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 353 ~(NFS_FABLKSIZE - 1); 354 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 355 if (max < nmp->nm_rsize && max > 0) { 356 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1); 357 if (nmp->nm_rsize == 0) 358 nmp->nm_rsize = max; 359 } 360 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 361 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ) 362 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) & 363 ~(NFS_DIRBLKSIZ - 1); 364 if (max < nmp->nm_readdirsize && max > 0) { 365 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1); 366 if (nmp->nm_readdirsize == 0) 367 nmp->nm_readdirsize = max; 368 } 369 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 370 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 371 nmp->nm_maxfilesize = maxfsize; 372 nmp->nm_state |= NFSSTA_GOTFSINFO; 373 } 374 nfsm_reqdone; 375 return (error); 376 } 377 378 /* 379 * Mount a remote root fs via. nfs. This depends on the info in the 380 * nfs_diskless structure that has been filled in properly by some primary 381 * bootstrap. 382 * It goes something like this: 383 * - do enough of "ifconfig" by calling ifioctl() so that the system 384 * can talk to the server 385 * - If nfs_diskless.mygateway is filled in, use that address as 386 * a default gateway. 387 * - build the rootfs mount point and call mountnfs() to do the rest. 388 */ 389 int 390 nfs_mountroot(mp) 391 struct mount *mp; 392 { 393 struct mount *swap_mp; 394 struct nfsv3_diskless *nd = &nfsv3_diskless; 395 struct socket *so; 396 struct vnode *vp; 397 struct thread *td = curthread; /* XXX */ 398 int error, i; 399 u_long l; 400 char buf[128]; 401 402 #if defined(BOOTP_NFSROOT) && defined(BOOTP) 403 bootpc_init(); /* use bootp to get nfs_diskless filled in */ 404 #endif 405 406 /* 407 * XXX time must be non-zero when we init the interface or else 408 * the arp code will wedge... 409 */ 410 while (time_second == 0) 411 tsleep(&time_second, 0, "arpkludge", 10); 412 413 if (nfs_diskless_valid==1) 414 nfs_convert_diskless(); 415 416 /* 417 * XXX splnet, so networks will receive... 418 */ 419 splnet(); 420 421 #ifdef notyet 422 /* Set up swap credentials. */ 423 proc0.p_ucred->cr_uid = ntohl(nd->swap_ucred.cr_uid); 424 proc0.p_ucred->cr_gid = ntohl(nd->swap_ucred.cr_gid); 425 if ((proc0.p_ucred->cr_ngroups = ntohs(nd->swap_ucred.cr_ngroups)) > 426 NGROUPS) 427 proc0.p_ucred->cr_ngroups = NGROUPS; 428 for (i = 0; i < proc0.p_ucred->cr_ngroups; i++) 429 proc0.p_ucred->cr_groups[i] = ntohl(nd->swap_ucred.cr_groups[i]); 430 #endif 431 432 /* 433 * Do enough of ifconfig(8) so that the critical net interface can 434 * talk to the server. 435 */ 436 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td); 437 if (error) 438 panic("nfs_mountroot: socreate(%04x): %d", 439 nd->myif.ifra_addr.sa_family, error); 440 441 #if 0 /* XXX Bad idea */ 442 /* 443 * We might not have been told the right interface, so we pass 444 * over the first ten interfaces of the same kind, until we get 445 * one of them configured. 446 */ 447 448 for (i = strlen(nd->myif.ifra_name) - 1; 449 nd->myif.ifra_name[i] >= '0' && 450 nd->myif.ifra_name[i] <= '9'; 451 nd->myif.ifra_name[i] ++) { 452 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); 453 if(!error) 454 break; 455 } 456 #endif 457 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); 458 if (error) 459 panic("nfs_mountroot: SIOCAIFADDR: %d", error); 460 soclose(so); 461 462 /* 463 * If the gateway field is filled in, set it as the default route. 464 */ 465 if (nd->mygateway.sin_len != 0) { 466 struct sockaddr_in mask, sin; 467 468 bzero((caddr_t)&mask, sizeof(mask)); 469 sin = mask; 470 sin.sin_family = AF_INET; 471 sin.sin_len = sizeof(sin); 472 error = rtrequest(RTM_ADD, (struct sockaddr *)&sin, 473 (struct sockaddr *)&nd->mygateway, 474 (struct sockaddr *)&mask, 475 RTF_UP | RTF_GATEWAY, (struct rtentry **)0); 476 if (error) 477 panic("nfs_mountroot: RTM_ADD: %d", error); 478 } 479 480 /* 481 * Create the rootfs mount point. 482 */ 483 nd->root_args.fh = nd->root_fh; 484 nd->root_args.fhsize = nd->root_fhsize; 485 l = ntohl(nd->root_saddr.sin_addr.s_addr); 486 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 487 (l >> 24) & 0xff, (l >> 16) & 0xff, 488 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam); 489 printf("NFS ROOT: %s\n",buf); 490 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY, 491 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) { 492 if (swap_mp) { 493 mp->mnt_vfc->vfc_refcount--; 494 free(swap_mp, M_MOUNT); 495 } 496 return (error); 497 } 498 499 swap_mp = NULL; 500 if (nd->swap_nblks) { 501 502 /* Convert to DEV_BSIZE instead of Kilobyte */ 503 nd->swap_nblks *= 2; 504 505 /* 506 * Create a fake mount point just for the swap vnode so that the 507 * swap file can be on a different server from the rootfs. 508 */ 509 nd->swap_args.fh = nd->swap_fh; 510 nd->swap_args.fhsize = nd->swap_fhsize; 511 l = ntohl(nd->swap_saddr.sin_addr.s_addr); 512 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 513 (l >> 24) & 0xff, (l >> 16) & 0xff, 514 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam); 515 printf("NFS SWAP: %s\n",buf); 516 if ((error = nfs_mountdiskless(buf, "/swap", 0, 517 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) 518 return (error); 519 vfs_unbusy(swap_mp, td); 520 521 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size = 522 nd->swap_nblks * DEV_BSIZE ; 523 524 /* 525 * Since the swap file is not the root dir of a file system, 526 * hack it to a regular file. 527 */ 528 vp->v_type = VREG; 529 vp->v_flag = 0; 530 VREF(vp); 531 swaponvp(td, vp, NODEV, nd->swap_nblks); 532 } 533 534 mp->mnt_flag |= MNT_ROOTFS; 535 mp->mnt_vnodecovered = NULLVP; 536 rootvp = vp; 537 vfs_unbusy(mp, td); 538 539 /* 540 * This is not really an nfs issue, but it is much easier to 541 * set hostname here and then let the "/etc/rc.xxx" files 542 * mount the right /var based upon its preset value. 543 */ 544 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN); 545 hostname[MAXHOSTNAMELEN - 1] = '\0'; 546 for (i = 0; i < MAXHOSTNAMELEN; i++) 547 if (hostname[i] == '\0') 548 break; 549 inittodr(ntohl(nd->root_time)); 550 return (0); 551 } 552 553 /* 554 * Internal version of mount system call for diskless setup. 555 */ 556 static int 557 nfs_mountdiskless(char *path, char *which, int mountflag, 558 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td, 559 struct vnode **vpp, struct mount **mpp) 560 { 561 struct mount *mp; 562 struct sockaddr *nam; 563 int error; 564 int didalloc = 0; 565 566 mp = *mpp; 567 568 if (mp == NULL) { 569 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) { 570 printf("nfs_mountroot: NFS not configured"); 571 return (error); 572 } 573 didalloc = 1; 574 } 575 576 mp->mnt_kern_flag = 0; 577 mp->mnt_flag = mountflag; 578 nam = dup_sockaddr((struct sockaddr *)sin, 1); 579 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) { 580 printf("nfs_mountroot: mount %s on %s: %d", path, which, error); 581 mp->mnt_vfc->vfc_refcount--; 582 vfs_unbusy(mp, td); 583 if (didalloc) 584 free(mp, M_MOUNT); 585 FREE(nam, M_SONAME); 586 return (error); 587 } 588 (void) copystr(which, mp->mnt_stat.f_mntonname, MNAMELEN - 1, 0); 589 *mpp = mp; 590 return (0); 591 } 592 593 static void 594 nfs_decode_args(nmp, argp) 595 struct nfsmount *nmp; 596 struct nfs_args *argp; 597 { 598 int s; 599 int adjsock; 600 int maxio; 601 602 s = splnet(); 603 /* 604 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 605 * no sense in that context. 606 */ 607 if (argp->sotype == SOCK_STREAM) 608 nmp->nm_flag &= ~NFSMNT_NOCONN; 609 610 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */ 611 if ((argp->flags & NFSMNT_NFSV3) == 0) 612 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 613 614 /* Re-bind if rsrvd port requested and wasn't on one */ 615 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 616 && (argp->flags & NFSMNT_RESVPORT); 617 /* Also re-bind if we're switching to/from a connected UDP socket */ 618 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 619 (argp->flags & NFSMNT_NOCONN)); 620 621 /* Update flags atomically. Don't change the lock bits. */ 622 nmp->nm_flag = argp->flags | nmp->nm_flag; 623 splx(s); 624 625 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 626 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 627 if (nmp->nm_timeo < NFS_MINTIMEO) 628 nmp->nm_timeo = NFS_MINTIMEO; 629 else if (nmp->nm_timeo > NFS_MAXTIMEO) 630 nmp->nm_timeo = NFS_MAXTIMEO; 631 } 632 633 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 634 nmp->nm_retry = argp->retrans; 635 if (nmp->nm_retry > NFS_MAXREXMIT) 636 nmp->nm_retry = NFS_MAXREXMIT; 637 } 638 639 if (argp->flags & NFSMNT_NFSV3) { 640 if (argp->sotype == SOCK_DGRAM) 641 maxio = NFS_MAXDGRAMDATA; 642 else 643 maxio = NFS_MAXDATA; 644 } else 645 maxio = NFS_V2MAXDATA; 646 647 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 648 nmp->nm_wsize = argp->wsize; 649 /* Round down to multiple of blocksize */ 650 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 651 if (nmp->nm_wsize <= 0) 652 nmp->nm_wsize = NFS_FABLKSIZE; 653 } 654 if (nmp->nm_wsize > maxio) 655 nmp->nm_wsize = maxio; 656 if (nmp->nm_wsize > MAXBSIZE) 657 nmp->nm_wsize = MAXBSIZE; 658 659 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 660 nmp->nm_rsize = argp->rsize; 661 /* Round down to multiple of blocksize */ 662 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 663 if (nmp->nm_rsize <= 0) 664 nmp->nm_rsize = NFS_FABLKSIZE; 665 } 666 if (nmp->nm_rsize > maxio) 667 nmp->nm_rsize = maxio; 668 if (nmp->nm_rsize > MAXBSIZE) 669 nmp->nm_rsize = MAXBSIZE; 670 671 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 672 nmp->nm_readdirsize = argp->readdirsize; 673 } 674 if (nmp->nm_readdirsize > maxio) 675 nmp->nm_readdirsize = maxio; 676 if (nmp->nm_readdirsize > nmp->nm_rsize) 677 nmp->nm_readdirsize = nmp->nm_rsize; 678 679 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0) 680 nmp->nm_acregmin = argp->acregmin; 681 else 682 nmp->nm_acregmin = NFS_MINATTRTIMO; 683 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0) 684 nmp->nm_acregmax = argp->acregmax; 685 else 686 nmp->nm_acregmax = NFS_MAXATTRTIMO; 687 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0) 688 nmp->nm_acdirmin = argp->acdirmin; 689 else 690 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO; 691 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0) 692 nmp->nm_acdirmax = argp->acdirmax; 693 else 694 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO; 695 if (nmp->nm_acdirmin > nmp->nm_acdirmax) 696 nmp->nm_acdirmin = nmp->nm_acdirmax; 697 if (nmp->nm_acregmin > nmp->nm_acregmax) 698 nmp->nm_acregmin = nmp->nm_acregmax; 699 700 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) { 701 if (argp->maxgrouplist <= NFS_MAXGRPS) 702 nmp->nm_numgrps = argp->maxgrouplist; 703 else 704 nmp->nm_numgrps = NFS_MAXGRPS; 705 } 706 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) { 707 if (argp->readahead <= NFS_MAXRAHEAD) 708 nmp->nm_readahead = argp->readahead; 709 else 710 nmp->nm_readahead = NFS_MAXRAHEAD; 711 } 712 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2) { 713 if (argp->leaseterm <= NQ_MAXLEASE) 714 nmp->nm_leaseterm = argp->leaseterm; 715 else 716 nmp->nm_leaseterm = NQ_MAXLEASE; 717 } 718 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) { 719 if (argp->deadthresh <= NQ_NEVERDEAD) 720 nmp->nm_deadthresh = argp->deadthresh; 721 else 722 nmp->nm_deadthresh = NQ_NEVERDEAD; 723 } 724 725 adjsock |= ((nmp->nm_sotype != argp->sotype) || 726 (nmp->nm_soproto != argp->proto)); 727 nmp->nm_sotype = argp->sotype; 728 nmp->nm_soproto = argp->proto; 729 730 if (nmp->nm_so && adjsock) { 731 nfs_safedisconnect(nmp); 732 if (nmp->nm_sotype == SOCK_DGRAM) 733 while (nfs_connect(nmp, (struct nfsreq *)0)) { 734 printf("nfs_args: retrying connect\n"); 735 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0); 736 } 737 } 738 } 739 740 /* 741 * VFS Operations. 742 * 743 * mount system call 744 * It seems a bit dumb to copyinstr() the host and path here and then 745 * bcopy() them in mountnfs(), but I wanted to detect errors before 746 * doing the sockargs() call because sockargs() allocates an mbuf and 747 * an error after that means that I have to release the mbuf. 748 */ 749 /* ARGSUSED */ 750 static int 751 nfs_mount(struct mount *mp, char *path, caddr_t data, 752 struct nameidata *ndp, struct thread *td) 753 { 754 int error; 755 struct nfs_args args; 756 struct sockaddr *nam; 757 struct vnode *vp; 758 char pth[MNAMELEN], hst[MNAMELEN]; 759 size_t len; 760 u_char nfh[NFSX_V3FHMAX]; 761 762 if (path == NULL) { 763 nfs_mountroot(mp); 764 return (0); 765 } 766 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)); 767 if (error) 768 return (error); 769 if (args.version != NFS_ARGSVERSION) { 770 #ifdef COMPAT_PRELITE2 771 /* 772 * If the argument version is unknown, then assume the 773 * caller is a pre-lite2 4.4BSD client and convert its 774 * arguments. 775 */ 776 struct onfs_args oargs; 777 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args)); 778 if (error) 779 return (error); 780 nfs_convert_oargs(&args,&oargs); 781 #else /* !COMPAT_PRELITE2 */ 782 return (EPROGMISMATCH); 783 #endif /* COMPAT_PRELITE2 */ 784 } 785 if (mp->mnt_flag & MNT_UPDATE) { 786 struct nfsmount *nmp = VFSTONFS(mp); 787 788 if (nmp == NULL) 789 return (EIO); 790 /* 791 * When doing an update, we can't change from or to 792 * v3 and/or nqnfs, or change cookie translation 793 */ 794 args.flags = (args.flags & 795 ~(NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)) | 796 (nmp->nm_flag & 797 (NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)); 798 nfs_decode_args(nmp, &args); 799 return (0); 800 } 801 802 /* 803 * Make the nfs_ip_paranoia sysctl serve as the default connection 804 * or no-connection mode for those protocols that support 805 * no-connection mode (the flag will be cleared later for protocols 806 * that do not support no-connection mode). This will allow a client 807 * to receive replies from a different IP then the request was 808 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid), 809 * not 0. 810 */ 811 if (nfs_ip_paranoia == 0) 812 args.flags |= NFSMNT_NOCONN; 813 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) 814 return (EINVAL); 815 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); 816 if (error) 817 return (error); 818 error = copyinstr(path, pth, MNAMELEN-1, &len); 819 if (error) 820 return (error); 821 bzero(&pth[len], MNAMELEN - len); 822 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len); 823 if (error) 824 return (error); 825 bzero(&hst[len], MNAMELEN - len); 826 /* sockargs() call must be after above copyin() calls */ 827 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen); 828 if (error) 829 return (error); 830 args.fh = nfh; 831 error = mountnfs(&args, mp, nam, pth, hst, &vp); 832 return (error); 833 } 834 835 /* 836 * Common code for mount and mountroot 837 */ 838 static int 839 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam, 840 char *pth, char *hst, struct vnode **vpp) 841 { 842 struct nfsmount *nmp; 843 struct nfsnode *np; 844 int error; 845 struct vattr attrs; 846 847 if (mp->mnt_flag & MNT_UPDATE) { 848 nmp = VFSTONFS(mp); 849 /* update paths, file handles, etc, here XXX */ 850 FREE(nam, M_SONAME); 851 return (0); 852 } else { 853 nmp = zalloc(nfsmount_zone); 854 bzero((caddr_t)nmp, sizeof (struct nfsmount)); 855 TAILQ_INIT(&nmp->nm_uidlruhead); 856 TAILQ_INIT(&nmp->nm_bufq); 857 mp->mnt_data = (qaddr_t)nmp; 858 } 859 vfs_getnewfsid(mp); 860 nmp->nm_mountp = mp; 861 if (argp->flags & NFSMNT_NQNFS) 862 /* 863 * We have to set mnt_maxsymlink to a non-zero value so 864 * that COMPAT_43 routines will know that we are setting 865 * the d_type field in directories (and can zero it for 866 * unsuspecting binaries). 867 */ 868 mp->mnt_maxsymlinklen = 1; 869 870 /* 871 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too 872 * high, depending on whether we end up with negative offsets in 873 * the client or server somewhere. 2GB-1 may be safer. 874 * 875 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum 876 * that we can handle until we find out otherwise. 877 * XXX Our "safe" limit on the client is what we can store in our 878 * buffer cache using signed(!) block numbers. 879 */ 880 if ((argp->flags & NFSMNT_NFSV3) == 0) 881 nmp->nm_maxfilesize = 0xffffffffLL; 882 else 883 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 884 885 nmp->nm_timeo = NFS_TIMEO; 886 nmp->nm_retry = NFS_RETRANS; 887 nmp->nm_wsize = NFS_WSIZE; 888 nmp->nm_rsize = NFS_RSIZE; 889 nmp->nm_readdirsize = NFS_READDIRSIZE; 890 nmp->nm_numgrps = NFS_MAXGRPS; 891 nmp->nm_readahead = NFS_DEFRAHEAD; 892 nmp->nm_leaseterm = NQ_DEFLEASE; 893 nmp->nm_deadthresh = NQ_DEADTHRESH; 894 CIRCLEQ_INIT(&nmp->nm_timerhead); 895 nmp->nm_inprog = NULLVP; 896 nmp->nm_fhsize = argp->fhsize; 897 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize); 898 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); 899 bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN); 900 nmp->nm_nam = nam; 901 /* Set up the sockets and per-host congestion */ 902 nmp->nm_sotype = argp->sotype; 903 nmp->nm_soproto = argp->proto; 904 nmp->nm_cred = crhold(proc0.p_ucred); 905 906 nfs_decode_args(nmp, argp); 907 908 /* 909 * For Connection based sockets (TCP,...) defer the connect until 910 * the first request, in case the server is not responding. 911 */ 912 if (nmp->nm_sotype == SOCK_DGRAM && 913 (error = nfs_connect(nmp, (struct nfsreq *)0))) 914 goto bad; 915 916 /* 917 * This is silly, but it has to be set so that vinifod() works. 918 * We do not want to do an nfs_statfs() here since we can get 919 * stuck on a dead server and we are holding a lock on the mount 920 * point. 921 */ 922 mp->mnt_stat.f_iosize = nfs_iosize(nmp); 923 /* 924 * A reference count is needed on the nfsnode representing the 925 * remote root. If this object is not persistent, then backward 926 * traversals of the mount point (i.e. "..") will not work if 927 * the nfsnode gets flushed out of the cache. Ufs does not have 928 * this problem, because one can identify root inodes by their 929 * number == ROOTINO (2). 930 */ 931 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 932 if (error) 933 goto bad; 934 *vpp = NFSTOV(np); 935 936 /* 937 * Get file attributes for the mountpoint. This has the side 938 * effect of filling in (*vpp)->v_type with the correct value. 939 */ 940 VOP_GETATTR(*vpp, &attrs, curthread); 941 942 /* 943 * Lose the lock but keep the ref. 944 */ 945 VOP_UNLOCK(*vpp, 0, curthread); 946 947 return (0); 948 bad: 949 nfs_disconnect(nmp); 950 nfs_free_mount(nmp); 951 FREE(nam, M_SONAME); 952 return (error); 953 } 954 955 /* 956 * unmount system call 957 */ 958 static int 959 nfs_unmount(struct mount *mp, int mntflags, struct thread *td) 960 { 961 struct nfsmount *nmp; 962 int error, flags = 0; 963 964 if (mntflags & MNT_FORCE) 965 flags |= FORCECLOSE; 966 nmp = VFSTONFS(mp); 967 /* 968 * Goes something like this.. 969 * - Call vflush() to clear out vnodes for this file system 970 * - Close the socket 971 * - Free up the data structures 972 */ 973 /* In the forced case, cancel any outstanding requests. */ 974 if (flags & FORCECLOSE) { 975 error = nfs_nmcancelreqs(nmp); 976 if (error) 977 return (error); 978 } 979 /* 980 * Must handshake with nqnfs_clientd() if it is active. 981 */ 982 nmp->nm_state |= NFSSTA_DISMINPROG; 983 while (nmp->nm_inprog != NULLVP) 984 (void) tsleep((caddr_t)&lbolt, 0, "nfsdism", 0); 985 986 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */ 987 error = vflush(mp, 1, flags); 988 if (error) { 989 nmp->nm_state &= ~NFSSTA_DISMINPROG; 990 return (error); 991 } 992 993 /* 994 * We are now committed to the unmount. 995 * For NQNFS, let the server daemon free the nfsmount structure. 996 */ 997 if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) 998 nmp->nm_state |= NFSSTA_DISMNT; 999 1000 nfs_disconnect(nmp); 1001 FREE(nmp->nm_nam, M_SONAME); 1002 1003 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0) 1004 nfs_free_mount(nmp); 1005 return (0); 1006 } 1007 1008 void 1009 nfs_free_mount(struct nfsmount *nmp) 1010 { 1011 if (nmp->nm_cred) { 1012 crfree(nmp->nm_cred); 1013 nmp->nm_cred = NULL; 1014 } 1015 zfree(nfsmount_zone, nmp); 1016 } 1017 1018 /* 1019 * Return root of a filesystem 1020 */ 1021 static int 1022 nfs_root(mp, vpp) 1023 struct mount *mp; 1024 struct vnode **vpp; 1025 { 1026 struct vnode *vp; 1027 struct nfsmount *nmp; 1028 struct nfsnode *np; 1029 int error; 1030 1031 nmp = VFSTONFS(mp); 1032 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 1033 if (error) 1034 return (error); 1035 vp = NFSTOV(np); 1036 if (vp->v_type == VNON) 1037 vp->v_type = VDIR; 1038 vp->v_flag = VROOT; 1039 *vpp = vp; 1040 return (0); 1041 } 1042 1043 extern int syncprt; 1044 1045 /* 1046 * Flush out the buffer cache 1047 */ 1048 /* ARGSUSED */ 1049 static int 1050 nfs_sync(struct mount *mp, int waitfor, struct thread *td) 1051 { 1052 struct vnode *vp; 1053 int error, allerror = 0; 1054 1055 /* 1056 * Force stale buffer cache information to be flushed. 1057 */ 1058 loop: 1059 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1060 vp != NULL; 1061 vp = TAILQ_NEXT(vp, v_nmntvnodes)) { 1062 /* 1063 * If the vnode that we are about to sync is no longer 1064 * associated with this mount point, start over. 1065 */ 1066 if (vp->v_mount != mp) 1067 goto loop; 1068 if (VOP_ISLOCKED(vp, NULL) || TAILQ_EMPTY(&vp->v_dirtyblkhd) || 1069 waitfor == MNT_LAZY) 1070 continue; 1071 if (vget(vp, LK_EXCLUSIVE, td)) 1072 goto loop; 1073 error = VOP_FSYNC(vp, waitfor, td); 1074 if (error) 1075 allerror = error; 1076 vput(vp); 1077 } 1078 return (allerror); 1079 } 1080 1081