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 1.26 2005/04/15 19:08:21 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 extern void bootpc_init(void); 76 77 extern int nfs_ticks; 78 extern struct vnodeopv_entry_desc nfsv2_vnodeop_entries[]; 79 extern struct vnodeopv_entry_desc nfsv2_fifoop_entries[]; 80 extern struct vnodeopv_entry_desc nfsv2_specop_entries[]; 81 82 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header"); 83 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle"); 84 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure"); 85 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data"); 86 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor"); 87 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure"); 88 MALLOC_DEFINE(M_NQLEASE, "NQNFS Lease", "Nqnfs lease"); 89 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables"); 90 91 vm_zone_t nfsmount_zone; 92 93 struct nfsstats nfsstats; 94 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem"); 95 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD, 96 &nfsstats, nfsstats, ""); 97 static int nfs_ip_paranoia = 1; 98 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW, 99 &nfs_ip_paranoia, 0, ""); 100 #ifdef NFS_DEBUG 101 int nfs_debug; 102 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, ""); 103 #endif 104 105 /* 106 * Tunable to determine the Read/Write unit size. Maximum value 107 * is NFS_MAXDATA. We also default to NFS_MAXDATA. 108 */ 109 static int nfs_io_size = NFS_MAXDATA; 110 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW, 111 &nfs_io_size, 0, "NFS optimal I/O unit size"); 112 113 static void nfs_decode_args (struct nfsmount *nmp, 114 struct nfs_args *argp); 115 static int mountnfs (struct nfs_args *,struct mount *, 116 struct sockaddr *,char *,char *,struct vnode **); 117 static int nfs_mount ( struct mount *mp, char *path, caddr_t data, 118 struct thread *td); 119 static int nfs_unmount ( struct mount *mp, int mntflags, 120 struct thread *td); 121 static int nfs_root ( struct mount *mp, struct vnode **vpp); 122 static int nfs_statfs ( struct mount *mp, struct statfs *sbp, 123 struct thread *td); 124 static int nfs_sync ( struct mount *mp, int waitfor, 125 struct thread *td); 126 127 /* 128 * nfs vfs operations. 129 */ 130 static struct vfsops nfs_vfsops = { 131 nfs_mount, 132 vfs_stdstart, 133 nfs_unmount, 134 nfs_root, 135 vfs_stdquotactl, 136 nfs_statfs, 137 nfs_sync, 138 vfs_stdvget, 139 vfs_stdfhtovp, /* shouldn't happen */ 140 vfs_stdcheckexp, 141 vfs_stdvptofh, /* shouldn't happen */ 142 nfs_init, 143 nfs_uninit, 144 vfs_stdextattrctl, 145 }; 146 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK); 147 148 /* 149 * This structure must be filled in by a primary bootstrap or bootstrap 150 * server for a diskless/dataless machine. It is initialized below just 151 * to ensure that it is allocated to initialized data (.data not .bss). 152 */ 153 struct nfs_diskless nfs_diskless = { { { 0 } } }; 154 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } }; 155 int nfs_diskless_valid = 0; 156 157 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD, 158 &nfs_diskless_valid, 0, ""); 159 160 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD, 161 nfsv3_diskless.root_hostnam, 0, ""); 162 163 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD, 164 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr, 165 "%Ssockaddr_in", ""); 166 167 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD, 168 nfsv3_diskless.swap_hostnam, 0, ""); 169 170 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD, 171 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr, 172 "%Ssockaddr_in",""); 173 174 175 void nfsargs_ntoh (struct nfs_args *); 176 static int nfs_mountdiskless (char *, char *, int, 177 struct sockaddr_in *, struct nfs_args *, 178 struct thread *, struct vnode **, 179 struct mount **); 180 static void nfs_convert_diskless (void); 181 static void nfs_convert_oargs (struct nfs_args *args, 182 struct onfs_args *oargs); 183 184 /* 185 * Calculate the buffer I/O block size to use. The maximum V2 block size 186 * is typically 8K, the maximum datagram size is typically 16K, and the 187 * maximum V3 block size is typically 32K. The buffer cache tends to work 188 * best with 16K blocks but we allow 32K for TCP connections. 189 * 190 * We force the block size to be at least a page for buffer cache efficiency. 191 */ 192 static 193 int 194 nfs_iosize(int v3, int sotype) 195 { 196 int iosize; 197 int iomax; 198 199 if (v3) { 200 if (sotype == SOCK_STREAM) 201 iomax = NFS_MAXDATA; 202 else 203 iomax = NFS_MAXDGRAMDATA; 204 } else { 205 iomax = NFS_V2MAXDATA; 206 } 207 if ((iosize = nfs_io_size) > iomax) 208 iosize = iomax; 209 if (iosize < PAGE_SIZE) 210 iosize = PAGE_SIZE; 211 212 /* 213 * This is an aweful hack but until the buffer cache is rewritten 214 * we need it. The problem is that when you combine write() with 215 * mmap() the vm_page->valid bits can become weird looking 216 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers 217 * at the file EOF. To solve the problem the BIO system needs to 218 * be guarenteed that the NFS iosize for regular files will be a 219 * multiple of PAGE_SIZE so it can invalidate the whole page 220 * rather then just the piece of it owned by the buffer when 221 * NFS does vinvalbuf() calls. 222 */ 223 if (iosize & PAGE_MASK) 224 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE; 225 return iosize; 226 } 227 228 static void 229 nfs_convert_oargs(args, oargs) 230 struct nfs_args *args; 231 struct onfs_args *oargs; 232 { 233 args->version = NFS_ARGSVERSION; 234 args->addr = oargs->addr; 235 args->addrlen = oargs->addrlen; 236 args->sotype = oargs->sotype; 237 args->proto = oargs->proto; 238 args->fh = oargs->fh; 239 args->fhsize = oargs->fhsize; 240 args->flags = oargs->flags; 241 args->wsize = oargs->wsize; 242 args->rsize = oargs->rsize; 243 args->readdirsize = oargs->readdirsize; 244 args->timeo = oargs->timeo; 245 args->retrans = oargs->retrans; 246 args->maxgrouplist = oargs->maxgrouplist; 247 args->readahead = oargs->readahead; 248 args->leaseterm = oargs->leaseterm; 249 args->deadthresh = oargs->deadthresh; 250 args->hostname = oargs->hostname; 251 } 252 253 static void 254 nfs_convert_diskless() 255 { 256 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif, 257 sizeof(struct ifaliasreq)); 258 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway, 259 sizeof(struct sockaddr_in)); 260 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args); 261 nfsv3_diskless.swap_fhsize = NFSX_V2FH; 262 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH); 263 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr, 264 sizeof(struct sockaddr_in)); 265 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN); 266 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks; 267 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred, 268 sizeof(struct ucred)); 269 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args); 270 nfsv3_diskless.root_fhsize = NFSX_V2FH; 271 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH); 272 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr, 273 sizeof(struct sockaddr_in)); 274 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN); 275 nfsv3_diskless.root_time = nfs_diskless.root_time; 276 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam, 277 MAXHOSTNAMELEN); 278 nfs_diskless_valid = 3; 279 } 280 281 /* 282 * nfs statfs call 283 */ 284 int 285 nfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 286 { 287 struct vnode *vp; 288 struct nfs_statfs *sfp; 289 caddr_t cp; 290 u_int32_t *tl; 291 int32_t t1, t2; 292 caddr_t bpos, dpos, cp2; 293 struct nfsmount *nmp = VFSTONFS(mp); 294 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr; 295 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 296 struct ucred *cred; 297 struct nfsnode *np; 298 u_quad_t tquad; 299 300 #ifndef nolint 301 sfp = (struct nfs_statfs *)0; 302 #endif 303 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 304 if (error) 305 return (error); 306 vp = NFSTOV(np); 307 cred = crget(); 308 cred->cr_ngroups = 1; 309 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 310 (void)nfs_fsinfo(nmp, vp, td); 311 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 312 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3)); 313 nfsm_fhtom(vp, v3); 314 nfsm_request(vp, NFSPROC_FSSTAT, td, cred); 315 if (v3) 316 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK); 317 if (error) { 318 if (mrep != NULL) 319 m_freem(mrep); 320 goto nfsmout; 321 } 322 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 323 sbp->f_flags = nmp->nm_flag; 324 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype); 325 326 if (v3) { 327 sbp->f_bsize = NFS_FABLKSIZE; 328 tquad = fxdr_hyper(&sfp->sf_tbytes); 329 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 330 tquad = fxdr_hyper(&sfp->sf_fbytes); 331 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 332 tquad = fxdr_hyper(&sfp->sf_abytes); 333 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 334 sbp->f_files = (fxdr_unsigned(int32_t, 335 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff); 336 sbp->f_ffree = (fxdr_unsigned(int32_t, 337 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff); 338 } else { 339 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 340 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 341 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 342 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 343 sbp->f_files = 0; 344 sbp->f_ffree = 0; 345 } 346 if (sbp != &mp->mnt_stat) { 347 sbp->f_type = mp->mnt_vfc->vfc_typenum; 348 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); 349 } 350 m_freem(mrep); 351 nfsmout: 352 vput(vp); 353 crfree(cred); 354 return (error); 355 } 356 357 /* 358 * nfs version 3 fsinfo rpc call 359 */ 360 int 361 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td) 362 { 363 struct nfsv3_fsinfo *fsp; 364 caddr_t cp; 365 int32_t t1, t2; 366 u_int32_t *tl, pref, max; 367 caddr_t bpos, dpos, cp2; 368 int error = 0, retattr; 369 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 370 u_int64_t maxfsize; 371 372 nfsstats.rpccnt[NFSPROC_FSINFO]++; 373 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1)); 374 nfsm_fhtom(vp, 1); 375 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ)); 376 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK); 377 if (!error) { 378 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 379 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 380 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 381 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 382 ~(NFS_FABLKSIZE - 1); 383 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 384 if (max < nmp->nm_wsize && max > 0) { 385 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1); 386 if (nmp->nm_wsize == 0) 387 nmp->nm_wsize = max; 388 } 389 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 390 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 391 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 392 ~(NFS_FABLKSIZE - 1); 393 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 394 if (max < nmp->nm_rsize && max > 0) { 395 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1); 396 if (nmp->nm_rsize == 0) 397 nmp->nm_rsize = max; 398 } 399 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 400 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ) 401 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) & 402 ~(NFS_DIRBLKSIZ - 1); 403 if (max < nmp->nm_readdirsize && max > 0) { 404 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1); 405 if (nmp->nm_readdirsize == 0) 406 nmp->nm_readdirsize = max; 407 } 408 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 409 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 410 nmp->nm_maxfilesize = maxfsize; 411 nmp->nm_state |= NFSSTA_GOTFSINFO; 412 } 413 m_freem(mrep); 414 nfsmout: 415 return (error); 416 } 417 418 /* 419 * Mount a remote root fs via. nfs. This depends on the info in the 420 * nfs_diskless structure that has been filled in properly by some primary 421 * bootstrap. 422 * It goes something like this: 423 * - do enough of "ifconfig" by calling ifioctl() so that the system 424 * can talk to the server 425 * - If nfs_diskless.mygateway is filled in, use that address as 426 * a default gateway. 427 * - build the rootfs mount point and call mountnfs() to do the rest. 428 */ 429 int 430 nfs_mountroot(mp) 431 struct mount *mp; 432 { 433 struct mount *swap_mp; 434 struct nfsv3_diskless *nd = &nfsv3_diskless; 435 struct socket *so; 436 struct vnode *vp; 437 struct thread *td = curthread; /* XXX */ 438 int error, i; 439 u_long l; 440 char buf[128]; 441 442 #if defined(BOOTP_NFSROOT) && defined(BOOTP) 443 bootpc_init(); /* use bootp to get nfs_diskless filled in */ 444 #endif 445 446 /* 447 * XXX time must be non-zero when we init the interface or else 448 * the arp code will wedge... 449 */ 450 while (mycpu->gd_time_seconds == 0) 451 tsleep(mycpu, 0, "arpkludge", 10); 452 453 if (nfs_diskless_valid==1) 454 nfs_convert_diskless(); 455 456 /* 457 * XXX splnet, so networks will receive... 458 */ 459 splnet(); 460 461 #ifdef notyet 462 /* Set up swap credentials. */ 463 proc0.p_ucred->cr_uid = ntohl(nd->swap_ucred.cr_uid); 464 proc0.p_ucred->cr_gid = ntohl(nd->swap_ucred.cr_gid); 465 if ((proc0.p_ucred->cr_ngroups = ntohs(nd->swap_ucred.cr_ngroups)) > 466 NGROUPS) 467 proc0.p_ucred->cr_ngroups = NGROUPS; 468 for (i = 0; i < proc0.p_ucred->cr_ngroups; i++) 469 proc0.p_ucred->cr_groups[i] = ntohl(nd->swap_ucred.cr_groups[i]); 470 #endif 471 472 /* 473 * Do enough of ifconfig(8) so that the critical net interface can 474 * talk to the server. 475 */ 476 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td); 477 if (error) 478 panic("nfs_mountroot: socreate(%04x): %d", 479 nd->myif.ifra_addr.sa_family, error); 480 481 #if 0 /* XXX Bad idea */ 482 /* 483 * We might not have been told the right interface, so we pass 484 * over the first ten interfaces of the same kind, until we get 485 * one of them configured. 486 */ 487 488 for (i = strlen(nd->myif.ifra_name) - 1; 489 nd->myif.ifra_name[i] >= '0' && 490 nd->myif.ifra_name[i] <= '9'; 491 nd->myif.ifra_name[i] ++) { 492 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); 493 if(!error) 494 break; 495 } 496 #endif 497 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, td); 498 if (error) 499 panic("nfs_mountroot: SIOCAIFADDR: %d", error); 500 soclose(so); 501 502 /* 503 * If the gateway field is filled in, set it as the default route. 504 */ 505 if (nd->mygateway.sin_len != 0) { 506 struct sockaddr_in mask, sin; 507 508 bzero((caddr_t)&mask, sizeof(mask)); 509 sin = mask; 510 sin.sin_family = AF_INET; 511 sin.sin_len = sizeof(sin); 512 error = rtrequest(RTM_ADD, (struct sockaddr *)&sin, 513 (struct sockaddr *)&nd->mygateway, 514 (struct sockaddr *)&mask, 515 RTF_UP | RTF_GATEWAY, (struct rtentry **)0); 516 if (error) 517 panic("nfs_mountroot: RTM_ADD: %d", error); 518 } 519 520 /* 521 * Create the rootfs mount point. 522 */ 523 nd->root_args.fh = nd->root_fh; 524 nd->root_args.fhsize = nd->root_fhsize; 525 l = ntohl(nd->root_saddr.sin_addr.s_addr); 526 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 527 (l >> 24) & 0xff, (l >> 16) & 0xff, 528 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam); 529 printf("NFS ROOT: %s\n",buf); 530 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY, 531 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) { 532 if (swap_mp) { 533 mp->mnt_vfc->vfc_refcount--; 534 free(swap_mp, M_MOUNT); 535 } 536 return (error); 537 } 538 539 swap_mp = NULL; 540 if (nd->swap_nblks) { 541 542 /* Convert to DEV_BSIZE instead of Kilobyte */ 543 nd->swap_nblks *= 2; 544 545 /* 546 * Create a fake mount point just for the swap vnode so that the 547 * swap file can be on a different server from the rootfs. 548 */ 549 nd->swap_args.fh = nd->swap_fh; 550 nd->swap_args.fhsize = nd->swap_fhsize; 551 l = ntohl(nd->swap_saddr.sin_addr.s_addr); 552 snprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 553 (l >> 24) & 0xff, (l >> 16) & 0xff, 554 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam); 555 printf("NFS SWAP: %s\n",buf); 556 if ((error = nfs_mountdiskless(buf, "/swap", 0, 557 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) 558 return (error); 559 vfs_unbusy(swap_mp, td); 560 561 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size = 562 nd->swap_nblks * DEV_BSIZE ; 563 564 /* 565 * Since the swap file is not the root dir of a file system, 566 * hack it to a regular file. 567 */ 568 vp->v_type = VREG; 569 vp->v_flag = 0; 570 vref(vp); 571 swaponvp(td, vp, nd->swap_nblks); 572 } 573 574 mp->mnt_flag |= MNT_ROOTFS; 575 mp->mnt_vnodecovered = NULLVP; 576 vfs_unbusy(mp, td); 577 578 /* 579 * This is not really an nfs issue, but it is much easier to 580 * set hostname here and then let the "/etc/rc.xxx" files 581 * mount the right /var based upon its preset value. 582 */ 583 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN); 584 hostname[MAXHOSTNAMELEN - 1] = '\0'; 585 for (i = 0; i < MAXHOSTNAMELEN; i++) 586 if (hostname[i] == '\0') 587 break; 588 inittodr(ntohl(nd->root_time)); 589 return (0); 590 } 591 592 /* 593 * Internal version of mount system call for diskless setup. 594 */ 595 static int 596 nfs_mountdiskless(char *path, char *which, int mountflag, 597 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td, 598 struct vnode **vpp, struct mount **mpp) 599 { 600 struct mount *mp; 601 struct sockaddr *nam; 602 int error; 603 int didalloc = 0; 604 605 mp = *mpp; 606 607 if (mp == NULL) { 608 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) { 609 printf("nfs_mountroot: NFS not configured"); 610 return (error); 611 } 612 didalloc = 1; 613 } 614 615 mp->mnt_kern_flag = 0; 616 mp->mnt_flag = mountflag; 617 nam = dup_sockaddr((struct sockaddr *)sin); 618 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) { 619 printf("nfs_mountroot: mount %s on %s: %d", path, which, error); 620 mp->mnt_vfc->vfc_refcount--; 621 vfs_unbusy(mp, td); 622 if (didalloc) 623 free(mp, M_MOUNT); 624 FREE(nam, M_SONAME); 625 return (error); 626 } 627 *mpp = mp; 628 return (0); 629 } 630 631 static void 632 nfs_decode_args(nmp, argp) 633 struct nfsmount *nmp; 634 struct nfs_args *argp; 635 { 636 int s; 637 int adjsock; 638 int maxio; 639 640 s = splnet(); 641 /* 642 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 643 * no sense in that context. 644 */ 645 if (argp->sotype == SOCK_STREAM) 646 nmp->nm_flag &= ~NFSMNT_NOCONN; 647 648 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */ 649 if ((argp->flags & NFSMNT_NFSV3) == 0) 650 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 651 652 /* Re-bind if rsrvd port requested and wasn't on one */ 653 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 654 && (argp->flags & NFSMNT_RESVPORT); 655 /* Also re-bind if we're switching to/from a connected UDP socket */ 656 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 657 (argp->flags & NFSMNT_NOCONN)); 658 659 /* Update flags atomically. Don't change the lock bits. */ 660 nmp->nm_flag = argp->flags | nmp->nm_flag; 661 splx(s); 662 663 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 664 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 665 if (nmp->nm_timeo < NFS_MINTIMEO) 666 nmp->nm_timeo = NFS_MINTIMEO; 667 else if (nmp->nm_timeo > NFS_MAXTIMEO) 668 nmp->nm_timeo = NFS_MAXTIMEO; 669 } 670 671 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 672 nmp->nm_retry = argp->retrans; 673 if (nmp->nm_retry > NFS_MAXREXMIT) 674 nmp->nm_retry = NFS_MAXREXMIT; 675 } 676 677 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype); 678 679 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 680 nmp->nm_wsize = argp->wsize; 681 /* Round down to multiple of blocksize */ 682 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 683 if (nmp->nm_wsize <= 0) 684 nmp->nm_wsize = NFS_FABLKSIZE; 685 } 686 if (nmp->nm_wsize > maxio) 687 nmp->nm_wsize = maxio; 688 if (nmp->nm_wsize > MAXBSIZE) 689 nmp->nm_wsize = MAXBSIZE; 690 691 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 692 nmp->nm_rsize = argp->rsize; 693 /* Round down to multiple of blocksize */ 694 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 695 if (nmp->nm_rsize <= 0) 696 nmp->nm_rsize = NFS_FABLKSIZE; 697 } 698 if (nmp->nm_rsize > maxio) 699 nmp->nm_rsize = maxio; 700 if (nmp->nm_rsize > MAXBSIZE) 701 nmp->nm_rsize = MAXBSIZE; 702 703 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 704 nmp->nm_readdirsize = argp->readdirsize; 705 } 706 if (nmp->nm_readdirsize > maxio) 707 nmp->nm_readdirsize = maxio; 708 if (nmp->nm_readdirsize > nmp->nm_rsize) 709 nmp->nm_readdirsize = nmp->nm_rsize; 710 711 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0) 712 nmp->nm_acregmin = argp->acregmin; 713 else 714 nmp->nm_acregmin = NFS_MINATTRTIMO; 715 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0) 716 nmp->nm_acregmax = argp->acregmax; 717 else 718 nmp->nm_acregmax = NFS_MAXATTRTIMO; 719 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0) 720 nmp->nm_acdirmin = argp->acdirmin; 721 else 722 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO; 723 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0) 724 nmp->nm_acdirmax = argp->acdirmax; 725 else 726 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO; 727 if (nmp->nm_acdirmin > nmp->nm_acdirmax) 728 nmp->nm_acdirmin = nmp->nm_acdirmax; 729 if (nmp->nm_acregmin > nmp->nm_acregmax) 730 nmp->nm_acregmin = nmp->nm_acregmax; 731 732 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) { 733 if (argp->maxgrouplist <= NFS_MAXGRPS) 734 nmp->nm_numgrps = argp->maxgrouplist; 735 else 736 nmp->nm_numgrps = NFS_MAXGRPS; 737 } 738 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) { 739 if (argp->readahead <= NFS_MAXRAHEAD) 740 nmp->nm_readahead = argp->readahead; 741 else 742 nmp->nm_readahead = NFS_MAXRAHEAD; 743 } 744 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2) { 745 if (argp->leaseterm <= NQ_MAXLEASE) 746 nmp->nm_leaseterm = argp->leaseterm; 747 else 748 nmp->nm_leaseterm = NQ_MAXLEASE; 749 } 750 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) { 751 if (argp->deadthresh <= NQ_NEVERDEAD) 752 nmp->nm_deadthresh = argp->deadthresh; 753 else 754 nmp->nm_deadthresh = NQ_NEVERDEAD; 755 } 756 757 adjsock |= ((nmp->nm_sotype != argp->sotype) || 758 (nmp->nm_soproto != argp->proto)); 759 nmp->nm_sotype = argp->sotype; 760 nmp->nm_soproto = argp->proto; 761 762 if (nmp->nm_so && adjsock) { 763 nfs_safedisconnect(nmp); 764 if (nmp->nm_sotype == SOCK_DGRAM) 765 while (nfs_connect(nmp, (struct nfsreq *)0)) { 766 printf("nfs_args: retrying connect\n"); 767 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0); 768 } 769 } 770 } 771 772 /* 773 * VFS Operations. 774 * 775 * mount system call 776 * It seems a bit dumb to copyinstr() the host and path here and then 777 * bcopy() them in mountnfs(), but I wanted to detect errors before 778 * doing the sockargs() call because sockargs() allocates an mbuf and 779 * an error after that means that I have to release the mbuf. 780 */ 781 /* ARGSUSED */ 782 static int 783 nfs_mount(struct mount *mp, char *path, caddr_t data, struct thread *td) 784 { 785 int error; 786 struct nfs_args args; 787 struct sockaddr *nam; 788 struct vnode *vp; 789 char pth[MNAMELEN], hst[MNAMELEN]; 790 size_t len; 791 u_char nfh[NFSX_V3FHMAX]; 792 793 if (path == NULL) { 794 nfs_mountroot(mp); 795 return (0); 796 } 797 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)); 798 if (error) 799 return (error); 800 if (args.version != NFS_ARGSVERSION) { 801 #ifdef COMPAT_PRELITE2 802 /* 803 * If the argument version is unknown, then assume the 804 * caller is a pre-lite2 4.4BSD client and convert its 805 * arguments. 806 */ 807 struct onfs_args oargs; 808 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args)); 809 if (error) 810 return (error); 811 nfs_convert_oargs(&args,&oargs); 812 #else /* !COMPAT_PRELITE2 */ 813 return (EPROGMISMATCH); 814 #endif /* COMPAT_PRELITE2 */ 815 } 816 if (mp->mnt_flag & MNT_UPDATE) { 817 struct nfsmount *nmp = VFSTONFS(mp); 818 819 if (nmp == NULL) 820 return (EIO); 821 /* 822 * When doing an update, we can't change from or to 823 * v3 and/or nqnfs, or change cookie translation 824 */ 825 args.flags = (args.flags & 826 ~(NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)) | 827 (nmp->nm_flag & 828 (NFSMNT_NFSV3|NFSMNT_NQNFS /*|NFSMNT_XLATECOOKIE*/)); 829 nfs_decode_args(nmp, &args); 830 return (0); 831 } 832 833 /* 834 * Make the nfs_ip_paranoia sysctl serve as the default connection 835 * or no-connection mode for those protocols that support 836 * no-connection mode (the flag will be cleared later for protocols 837 * that do not support no-connection mode). This will allow a client 838 * to receive replies from a different IP then the request was 839 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid), 840 * not 0. 841 */ 842 if (nfs_ip_paranoia == 0) 843 args.flags |= NFSMNT_NOCONN; 844 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) 845 return (EINVAL); 846 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); 847 if (error) 848 return (error); 849 error = copyinstr(path, pth, MNAMELEN-1, &len); 850 if (error) 851 return (error); 852 bzero(&pth[len], MNAMELEN - len); 853 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len); 854 if (error) 855 return (error); 856 bzero(&hst[len], MNAMELEN - len); 857 /* sockargs() call must be after above copyin() calls */ 858 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen); 859 if (error) 860 return (error); 861 args.fh = nfh; 862 error = mountnfs(&args, mp, nam, pth, hst, &vp); 863 return (error); 864 } 865 866 /* 867 * Common code for mount and mountroot 868 */ 869 static int 870 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam, 871 char *pth, char *hst, struct vnode **vpp) 872 { 873 struct nfsmount *nmp; 874 struct nfsnode *np; 875 int error; 876 877 if (mp->mnt_flag & MNT_UPDATE) { 878 nmp = VFSTONFS(mp); 879 /* update paths, file handles, etc, here XXX */ 880 FREE(nam, M_SONAME); 881 return (0); 882 } else { 883 nmp = zalloc(nfsmount_zone); 884 bzero((caddr_t)nmp, sizeof (struct nfsmount)); 885 TAILQ_INIT(&nmp->nm_uidlruhead); 886 TAILQ_INIT(&nmp->nm_bufq); 887 mp->mnt_data = (qaddr_t)nmp; 888 } 889 vfs_getnewfsid(mp); 890 nmp->nm_mountp = mp; 891 if (argp->flags & NFSMNT_NQNFS) 892 /* 893 * We have to set mnt_maxsymlink to a non-zero value so 894 * that COMPAT_43 routines will know that we are setting 895 * the d_type field in directories (and can zero it for 896 * unsuspecting binaries). 897 */ 898 mp->mnt_maxsymlinklen = 1; 899 900 /* 901 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too 902 * high, depending on whether we end up with negative offsets in 903 * the client or server somewhere. 2GB-1 may be safer. 904 * 905 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum 906 * that we can handle until we find out otherwise. 907 * XXX Our "safe" limit on the client is what we can store in our 908 * buffer cache using signed(!) block numbers. 909 */ 910 if ((argp->flags & NFSMNT_NFSV3) == 0) 911 nmp->nm_maxfilesize = 0xffffffffLL; 912 else 913 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 914 915 nmp->nm_timeo = NFS_TIMEO; 916 nmp->nm_retry = NFS_RETRANS; 917 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype); 918 nmp->nm_rsize = nmp->nm_wsize; 919 nmp->nm_readdirsize = NFS_READDIRSIZE; 920 nmp->nm_numgrps = NFS_MAXGRPS; 921 nmp->nm_readahead = NFS_DEFRAHEAD; 922 nmp->nm_leaseterm = NQ_DEFLEASE; 923 nmp->nm_deadthresh = NQ_DEADTHRESH; 924 CIRCLEQ_INIT(&nmp->nm_timerhead); 925 nmp->nm_inprog = NULLVP; 926 nmp->nm_fhsize = argp->fhsize; 927 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize); 928 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); 929 nmp->nm_nam = nam; 930 /* Set up the sockets and per-host congestion */ 931 nmp->nm_sotype = argp->sotype; 932 nmp->nm_soproto = argp->proto; 933 nmp->nm_cred = crhold(proc0.p_ucred); 934 935 nfs_decode_args(nmp, argp); 936 937 /* 938 * For Connection based sockets (TCP,...) defer the connect until 939 * the first request, in case the server is not responding. 940 */ 941 if (nmp->nm_sotype == SOCK_DGRAM && 942 (error = nfs_connect(nmp, (struct nfsreq *)0))) 943 goto bad; 944 945 /* 946 * This is silly, but it has to be set so that vinifod() works. 947 * We do not want to do an nfs_statfs() here since we can get 948 * stuck on a dead server and we are holding a lock on the mount 949 * point. 950 */ 951 mp->mnt_stat.f_iosize = 952 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype); 953 954 /* 955 * Install vop_ops for our vnops 956 */ 957 vfs_add_vnodeops(mp, &mp->mnt_vn_norm_ops, nfsv2_vnodeop_entries); 958 vfs_add_vnodeops(mp, &mp->mnt_vn_spec_ops, nfsv2_specop_entries); 959 vfs_add_vnodeops(mp, &mp->mnt_vn_fifo_ops, nfsv2_fifoop_entries); 960 961 /* 962 * A reference count is needed on the nfsnode representing the 963 * remote root. If this object is not persistent, then backward 964 * traversals of the mount point (i.e. "..") will not work if 965 * the nfsnode gets flushed out of the cache. Ufs does not have 966 * this problem, because one can identify root inodes by their 967 * number == ROOTINO (2). 968 */ 969 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 970 if (error) 971 goto bad; 972 *vpp = NFSTOV(np); 973 974 /* 975 * Retrieval of mountpoint attributes is delayed until nfs_rot 976 * or nfs_statfs are first called. This will happen either when 977 * we first traverse the mount point or if somebody does a df(1). 978 * 979 * NFSSTA_GOTFSINFO is used to flag if we have successfully 980 * retrieved mountpoint attributes. In the case of NFSv3 we 981 * also flag static fsinfo. 982 */ 983 if (*vpp != NULL) 984 (*vpp)->v_type = VNON; 985 986 /* 987 * Lose the lock but keep the ref. 988 */ 989 VOP_UNLOCK(*vpp, 0, curthread); 990 991 return (0); 992 bad: 993 nfs_disconnect(nmp); 994 nfs_free_mount(nmp); 995 FREE(nam, M_SONAME); 996 return (error); 997 } 998 999 /* 1000 * unmount system call 1001 */ 1002 static int 1003 nfs_unmount(struct mount *mp, int mntflags, struct thread *td) 1004 { 1005 struct nfsmount *nmp; 1006 int error, flags = 0; 1007 1008 if (mntflags & MNT_FORCE) 1009 flags |= FORCECLOSE; 1010 nmp = VFSTONFS(mp); 1011 /* 1012 * Goes something like this.. 1013 * - Call vflush() to clear out vnodes for this file system 1014 * - Close the socket 1015 * - Free up the data structures 1016 */ 1017 /* In the forced case, cancel any outstanding requests. */ 1018 if (flags & FORCECLOSE) { 1019 error = nfs_nmcancelreqs(nmp); 1020 if (error) 1021 return (error); 1022 } 1023 /* 1024 * Must handshake with nqnfs_clientd() if it is active. 1025 */ 1026 nmp->nm_state |= NFSSTA_DISMINPROG; 1027 while (nmp->nm_inprog != NULLVP) 1028 (void) tsleep((caddr_t)&lbolt, 0, "nfsdism", 0); 1029 1030 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */ 1031 error = vflush(mp, 1, flags); 1032 if (error) { 1033 nmp->nm_state &= ~NFSSTA_DISMINPROG; 1034 return (error); 1035 } 1036 1037 /* 1038 * We are now committed to the unmount. 1039 * For NQNFS, let the server daemon free the nfsmount structure. 1040 */ 1041 if (nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) 1042 nmp->nm_state |= NFSSTA_DISMNT; 1043 1044 nfs_disconnect(nmp); 1045 FREE(nmp->nm_nam, M_SONAME); 1046 1047 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0) 1048 nfs_free_mount(nmp); 1049 return (0); 1050 } 1051 1052 void 1053 nfs_free_mount(struct nfsmount *nmp) 1054 { 1055 if (nmp->nm_cred) { 1056 crfree(nmp->nm_cred); 1057 nmp->nm_cred = NULL; 1058 } 1059 zfree(nfsmount_zone, nmp); 1060 } 1061 1062 /* 1063 * Return root of a filesystem 1064 */ 1065 static int 1066 nfs_root(mp, vpp) 1067 struct mount *mp; 1068 struct vnode **vpp; 1069 { 1070 struct vnode *vp; 1071 struct nfsmount *nmp; 1072 struct vattr attrs; 1073 struct nfsnode *np; 1074 int error; 1075 1076 nmp = VFSTONFS(mp); 1077 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 1078 if (error) 1079 return (error); 1080 vp = NFSTOV(np); 1081 1082 /* 1083 * Get transfer parameters and root vnode attributes 1084 */ 1085 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { 1086 if (nmp->nm_flag & NFSMNT_NFSV3) { 1087 nfs_fsinfo(nmp, vp, curthread); 1088 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype); 1089 } else { 1090 if ((error = VOP_GETATTR(vp, &attrs, curthread)) == 0) 1091 nmp->nm_state |= NFSSTA_GOTFSINFO; 1092 1093 } 1094 } 1095 if (vp->v_type == VNON) 1096 vp->v_type = VDIR; 1097 vp->v_flag = VROOT; 1098 *vpp = vp; 1099 return (0); 1100 } 1101 1102 extern int syncprt; 1103 1104 struct scaninfo { 1105 int rescan; 1106 thread_t td; 1107 int waitfor; 1108 int allerror; 1109 }; 1110 1111 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data); 1112 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data); 1113 1114 /* 1115 * Flush out the buffer cache 1116 */ 1117 /* ARGSUSED */ 1118 static int 1119 nfs_sync(struct mount *mp, int waitfor, struct thread *td) 1120 { 1121 struct scaninfo scaninfo; 1122 int error; 1123 1124 scaninfo.rescan = 0; 1125 scaninfo.td = td; 1126 scaninfo.waitfor = waitfor; 1127 scaninfo.allerror = 0; 1128 1129 /* 1130 * Force stale buffer cache information to be flushed. 1131 */ 1132 error = 0; 1133 while (error == 0 && scaninfo.rescan) { 1134 scaninfo.rescan = 0; 1135 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1, 1136 nfs_sync_scan2, &scaninfo); 1137 } 1138 return(error); 1139 } 1140 1141 static 1142 int 1143 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data) 1144 { 1145 struct scaninfo *info = data; 1146 1147 if (VOP_ISLOCKED(vp, NULL) || RB_EMPTY(&vp->v_rbdirty_tree)) 1148 return(-1); 1149 if (info->waitfor == MNT_LAZY) 1150 return(-1); 1151 return(0); 1152 } 1153 1154 static 1155 int 1156 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data) 1157 { 1158 struct scaninfo *info = data; 1159 int error; 1160 1161 error = VOP_FSYNC(vp, info->waitfor, info->td); 1162 if (error) 1163 info->allerror = error; 1164 return(0); 1165 } 1166 1167