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 1.51 2007/09/04 00:48:18 dillon Exp $ 39 */ 40 41 #include "opt_bootp.h" 42 #include "opt_nfsroot.h" 43 44 #include <sys/param.h> 45 #include <sys/sockio.h> 46 #include <sys/proc.h> 47 #include <sys/vnode.h> 48 #include <sys/fcntl.h> 49 #include <sys/kernel.h> 50 #include <sys/sysctl.h> 51 #include <sys/malloc.h> 52 #include <sys/mount.h> 53 #include <sys/mbuf.h> 54 #include <sys/socket.h> 55 #include <sys/socketvar.h> 56 #include <sys/systm.h> 57 58 #include <vm/vm.h> 59 #include <vm/vm_extern.h> 60 #include <vm/vm_zone.h> 61 62 #include <net/if.h> 63 #include <net/route.h> 64 #include <netinet/in.h> 65 66 #include <sys/thread2.h> 67 68 #include "rpcv2.h" 69 #include "nfsproto.h" 70 #include "nfs.h" 71 #include "nfsmount.h" 72 #include "nfsnode.h" 73 #include "xdr_subs.h" 74 #include "nfsm_subs.h" 75 #include "nfsdiskless.h" 76 #include "nfsmountrpc.h" 77 78 extern int nfs_mountroot(struct mount *mp); 79 extern void bootpc_init(void); 80 81 extern int nfs_ticks; 82 extern struct vop_ops nfsv2_vnode_vops; 83 extern struct vop_ops nfsv2_fifo_vops; 84 extern struct vop_ops nfsv2_spec_vops; 85 86 MALLOC_DEFINE(M_NFSREQ, "NFS req", "NFS request header"); 87 MALLOC_DEFINE(M_NFSBIGFH, "NFSV3 bigfh", "NFS version 3 file handle"); 88 MALLOC_DEFINE(M_NFSD, "NFS daemon", "Nfs server daemon structure"); 89 MALLOC_DEFINE(M_NFSDIROFF, "NFSV3 diroff", "NFS directory offset data"); 90 MALLOC_DEFINE(M_NFSRVDESC, "NFSV3 srvdesc", "NFS server socket descriptor"); 91 MALLOC_DEFINE(M_NFSUID, "NFS uid", "Nfs uid mapping structure"); 92 MALLOC_DEFINE(M_NFSHASH, "NFS hash", "NFS hash tables"); 93 94 vm_zone_t nfsmount_zone; 95 96 struct nfsstats nfsstats; 97 SYSCTL_NODE(_vfs, OID_AUTO, nfs, CTLFLAG_RW, 0, "NFS filesystem"); 98 SYSCTL_STRUCT(_vfs_nfs, NFS_NFSSTATS, nfsstats, CTLFLAG_RD, 99 &nfsstats, nfsstats, ""); 100 static int nfs_ip_paranoia = 1; 101 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_ip_paranoia, CTLFLAG_RW, 102 &nfs_ip_paranoia, 0, ""); 103 #ifdef NFS_DEBUG 104 int nfs_debug; 105 SYSCTL_INT(_vfs_nfs, OID_AUTO, debug, CTLFLAG_RW, &nfs_debug, 0, ""); 106 #endif 107 108 /* 109 * Tunable to determine the Read/Write unit size. Maximum value 110 * is NFS_MAXDATA. We also default to NFS_MAXDATA. 111 */ 112 static int nfs_io_size = NFS_MAXDATA; 113 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_io_size, CTLFLAG_RW, 114 &nfs_io_size, 0, "NFS optimal I/O unit size"); 115 116 static void nfs_decode_args (struct nfsmount *nmp, 117 struct nfs_args *argp); 118 static int mountnfs (struct nfs_args *,struct mount *, 119 struct sockaddr *,char *,char *,struct vnode **); 120 static int nfs_mount ( struct mount *mp, char *path, caddr_t data, 121 struct ucred *cred); 122 static int nfs_unmount ( struct mount *mp, int mntflags); 123 static int nfs_root ( struct mount *mp, struct vnode **vpp); 124 static int nfs_statfs ( struct mount *mp, struct statfs *sbp, 125 struct ucred *cred); 126 static int nfs_sync ( struct mount *mp, int waitfor); 127 128 /* 129 * nfs vfs operations. 130 */ 131 static struct vfsops nfs_vfsops = { 132 .vfs_mount = nfs_mount, 133 .vfs_unmount = nfs_unmount, 134 .vfs_root = nfs_root, 135 .vfs_statfs = nfs_statfs, 136 .vfs_sync = nfs_sync, 137 .vfs_init = nfs_init, 138 .vfs_uninit = nfs_uninit 139 }; 140 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK); 141 142 /* 143 * This structure must be filled in by a primary bootstrap or bootstrap 144 * server for a diskless/dataless machine. It is initialized below just 145 * to ensure that it is allocated to initialized data (.data not .bss). 146 */ 147 struct nfs_diskless nfs_diskless = { { { 0 } } }; 148 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } }; 149 int nfs_diskless_valid = 0; 150 151 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD, 152 &nfs_diskless_valid, 0, ""); 153 154 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD, 155 nfsv3_diskless.root_hostnam, 0, ""); 156 157 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD, 158 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr, 159 "%Ssockaddr_in", ""); 160 161 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD, 162 nfsv3_diskless.swap_hostnam, 0, ""); 163 164 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD, 165 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr, 166 "%Ssockaddr_in",""); 167 168 169 void nfsargs_ntoh (struct nfs_args *); 170 static int nfs_mountdiskless (char *, char *, int, 171 struct sockaddr_in *, struct nfs_args *, 172 struct thread *, struct vnode **, 173 struct mount **); 174 static void nfs_convert_diskless (void); 175 static void nfs_convert_oargs (struct nfs_args *args, 176 struct onfs_args *oargs); 177 178 /* 179 * Calculate the buffer I/O block size to use. The maximum V2 block size 180 * is typically 8K, the maximum datagram size is typically 16K, and the 181 * maximum V3 block size is typically 32K. The buffer cache tends to work 182 * best with 16K blocks but we allow 32K for TCP connections. 183 * 184 * We force the block size to be at least a page for buffer cache efficiency. 185 */ 186 static int 187 nfs_iosize(int v3, int sotype) 188 { 189 int iosize; 190 int iomax; 191 192 if (v3) { 193 if (sotype == SOCK_STREAM) 194 iomax = NFS_MAXDATA; 195 else 196 iomax = NFS_MAXDGRAMDATA; 197 } else { 198 iomax = NFS_V2MAXDATA; 199 } 200 if ((iosize = nfs_io_size) > iomax) 201 iosize = iomax; 202 if (iosize < PAGE_SIZE) 203 iosize = PAGE_SIZE; 204 205 /* 206 * This is an aweful hack but until the buffer cache is rewritten 207 * we need it. The problem is that when you combine write() with 208 * mmap() the vm_page->valid bits can become weird looking 209 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers 210 * at the file EOF. To solve the problem the BIO system needs to 211 * be guarenteed that the NFS iosize for regular files will be a 212 * multiple of PAGE_SIZE so it can invalidate the whole page 213 * rather then just the piece of it owned by the buffer when 214 * NFS does vinvalbuf() calls. 215 */ 216 if (iosize & PAGE_MASK) 217 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE; 218 return iosize; 219 } 220 221 static void 222 nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs) 223 { 224 args->version = NFS_ARGSVERSION; 225 args->addr = oargs->addr; 226 args->addrlen = oargs->addrlen; 227 args->sotype = oargs->sotype; 228 args->proto = oargs->proto; 229 args->fh = oargs->fh; 230 args->fhsize = oargs->fhsize; 231 args->flags = oargs->flags; 232 args->wsize = oargs->wsize; 233 args->rsize = oargs->rsize; 234 args->readdirsize = oargs->readdirsize; 235 args->timeo = oargs->timeo; 236 args->retrans = oargs->retrans; 237 args->maxgrouplist = oargs->maxgrouplist; 238 args->readahead = oargs->readahead; 239 args->deadthresh = oargs->deadthresh; 240 args->hostname = oargs->hostname; 241 } 242 243 static void 244 nfs_convert_diskless(void) 245 { 246 int i; 247 248 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif, 249 sizeof(struct ifaliasreq)); 250 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway, 251 sizeof(struct sockaddr_in)); 252 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args); 253 254 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH); 255 nfsv3_diskless.swap_fhsize = NFSX_V2FH; 256 for (i = NFSX_V2FH - 1; i >= 0; --i) { 257 if (nfs_diskless.swap_fh[i]) 258 break; 259 } 260 if (i < 0) 261 nfsv3_diskless.swap_fhsize = 0; 262 263 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); 270 271 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH); 272 nfsv3_diskless.root_fhsize = NFSX_V2FH; 273 for (i = NFSX_V2FH - 1; i >= 0; --i) { 274 if (nfs_diskless.root_fh[i]) 275 break; 276 } 277 if (i < 0) 278 nfsv3_diskless.root_fhsize = 0; 279 280 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr, 281 sizeof(struct sockaddr_in)); 282 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN); 283 nfsv3_diskless.root_time = nfs_diskless.root_time; 284 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam, 285 MAXHOSTNAMELEN); 286 nfs_diskless_valid = 3; 287 } 288 289 /* 290 * nfs statfs call 291 */ 292 int 293 nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred) 294 { 295 struct vnode *vp; 296 struct nfs_statfs *sfp; 297 caddr_t cp; 298 u_int32_t *tl; 299 int32_t t1, t2; 300 caddr_t bpos, dpos, cp2; 301 struct nfsmount *nmp = VFSTONFS(mp); 302 thread_t td = curthread; 303 int error = 0, v3 = (nmp->nm_flag & NFSMNT_NFSV3), retattr; 304 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 305 struct nfsnode *np; 306 u_quad_t tquad; 307 308 #ifndef nolint 309 sfp = (struct nfs_statfs *)0; 310 #endif 311 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 312 if (error) 313 return (error); 314 vp = NFSTOV(np); 315 /* ignore the passed cred */ 316 cred = crget(); 317 cred->cr_ngroups = 1; 318 if (v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 319 (void)nfs_fsinfo(nmp, vp, td); 320 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 321 nfsm_reqhead(vp, NFSPROC_FSSTAT, NFSX_FH(v3)); 322 nfsm_fhtom(vp, v3); 323 nfsm_request(vp, NFSPROC_FSSTAT, td, cred); 324 if (v3) 325 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK); 326 if (error) { 327 if (mrep != NULL) 328 m_freem(mrep); 329 goto nfsmout; 330 } 331 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 332 sbp->f_flags = nmp->nm_flag; 333 sbp->f_iosize = nfs_iosize(v3, nmp->nm_sotype); 334 335 if (v3) { 336 sbp->f_bsize = NFS_FABLKSIZE; 337 tquad = fxdr_hyper(&sfp->sf_tbytes); 338 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 339 tquad = fxdr_hyper(&sfp->sf_fbytes); 340 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 341 tquad = fxdr_hyper(&sfp->sf_abytes); 342 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 343 sbp->f_files = (fxdr_unsigned(int32_t, 344 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff); 345 sbp->f_ffree = (fxdr_unsigned(int32_t, 346 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff); 347 } else { 348 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 349 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 350 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 351 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 352 sbp->f_files = 0; 353 sbp->f_ffree = 0; 354 } 355 if (sbp != &mp->mnt_stat) { 356 sbp->f_type = mp->mnt_vfc->vfc_typenum; 357 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); 358 } 359 m_freem(mrep); 360 nfsmout: 361 vput(vp); 362 crfree(cred); 363 return (error); 364 } 365 366 /* 367 * nfs version 3 fsinfo rpc call 368 */ 369 int 370 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td) 371 { 372 struct nfsv3_fsinfo *fsp; 373 caddr_t cp; 374 int32_t t1, t2; 375 u_int32_t *tl, pref, max; 376 caddr_t bpos, dpos, cp2; 377 int error = 0, retattr; 378 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 379 u_int64_t maxfsize; 380 381 nfsstats.rpccnt[NFSPROC_FSINFO]++; 382 nfsm_reqhead(vp, NFSPROC_FSINFO, NFSX_FH(1)); 383 nfsm_fhtom(vp, 1); 384 nfsm_request(vp, NFSPROC_FSINFO, td, nfs_vpcred(vp, ND_READ)); 385 nfsm_postop_attr(vp, retattr, NFS_LATTR_NOSHRINK); 386 if (!error) { 387 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 388 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 389 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 390 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 391 ~(NFS_FABLKSIZE - 1); 392 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 393 if (max < nmp->nm_wsize && max > 0) { 394 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1); 395 if (nmp->nm_wsize == 0) 396 nmp->nm_wsize = max; 397 } 398 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 399 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 400 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 401 ~(NFS_FABLKSIZE - 1); 402 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 403 if (max < nmp->nm_rsize && max > 0) { 404 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1); 405 if (nmp->nm_rsize == 0) 406 nmp->nm_rsize = max; 407 } 408 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 409 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ) 410 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) & 411 ~(NFS_DIRBLKSIZ - 1); 412 if (max < nmp->nm_readdirsize && max > 0) { 413 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1); 414 if (nmp->nm_readdirsize == 0) 415 nmp->nm_readdirsize = max; 416 } 417 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 418 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 419 nmp->nm_maxfilesize = maxfsize; 420 nmp->nm_state |= NFSSTA_GOTFSINFO; 421 } 422 m_freem(mrep); 423 nfsmout: 424 return (error); 425 } 426 427 /* 428 * Mount a remote root fs via. nfs. This depends on the info in the 429 * nfs_diskless structure that has been filled in properly by some primary 430 * bootstrap. 431 * It goes something like this: 432 * - do enough of "ifconfig" by calling ifioctl() so that the system 433 * can talk to the server 434 * - If nfs_diskless.mygateway is filled in, use that address as 435 * a default gateway. 436 * - build the rootfs mount point and call mountnfs() to do the rest. 437 */ 438 int 439 nfs_mountroot(struct mount *mp) 440 { 441 struct mount *swap_mp; 442 struct nfsv3_diskless *nd = &nfsv3_diskless; 443 struct socket *so; 444 struct vnode *vp; 445 struct thread *td = curthread; /* XXX */ 446 int error, i; 447 u_long l; 448 char buf[128]; 449 450 #if defined(BOOTP_NFSROOT) && defined(BOOTP) 451 bootpc_init(); /* use bootp to get nfs_diskless filled in */ 452 #endif 453 454 /* 455 * XXX time must be non-zero when we init the interface or else 456 * the arp code will wedge... 457 */ 458 while (mycpu->gd_time_seconds == 0) 459 tsleep(mycpu, 0, "arpkludge", 10); 460 461 /* 462 * The boot code may have passed us a diskless structure. 463 */ 464 if (nfs_diskless_valid == 1) 465 nfs_convert_diskless(); 466 467 #define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr)) 468 kprintf("nfs_mountroot: interface %s ip %s", 469 nd->myif.ifra_name, 470 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr)); 471 kprintf(" bcast %s", 472 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr)); 473 kprintf(" mask %s\n", 474 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr)); 475 #undef SINP 476 477 /* 478 * XXX splnet, so networks will receive... 479 */ 480 crit_enter(); 481 482 /* 483 * BOOTP does not necessarily have to be compiled into the kernel 484 * for an NFS root to work. If we inherited the network 485 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured 486 * out our interface for us and all we need to do is ifconfig the 487 * interface. We only do this if the interface has not already been 488 * ifconfig'd by e.g. BOOTP. 489 */ 490 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td); 491 if (error) { 492 panic("nfs_mountroot: socreate(%04x): %d", 493 nd->myif.ifra_addr.sa_family, error); 494 } 495 496 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred); 497 if (error) 498 panic("nfs_mountroot: SIOCAIFADDR: %d", error); 499 500 soclose(so, FNONBLOCK); 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 kprintf("nfs_mountroot: gateway %s\n", 513 inet_ntoa(nd->mygateway.sin_addr)); 514 error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin, 515 (struct sockaddr *)&nd->mygateway, 516 (struct sockaddr *)&mask, 517 RTF_UP | RTF_GATEWAY); 518 if (error) 519 kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error); 520 } 521 522 /* 523 * Create the rootfs mount point. 524 */ 525 nd->root_args.fh = nd->root_fh; 526 nd->root_args.fhsize = nd->root_fhsize; 527 l = ntohl(nd->root_saddr.sin_addr.s_addr); 528 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 529 (l >> 24) & 0xff, (l >> 16) & 0xff, 530 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam); 531 kprintf("NFS ROOT: %s\n",buf); 532 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY, 533 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) { 534 if (swap_mp) { 535 mp->mnt_vfc->vfc_refcount--; 536 kfree(swap_mp, M_MOUNT); 537 } 538 crit_exit(); 539 return (error); 540 } 541 542 swap_mp = NULL; 543 if (nd->swap_nblks) { 544 545 /* Convert to DEV_BSIZE instead of Kilobyte */ 546 nd->swap_nblks *= 2; 547 548 /* 549 * Create a fake mount point just for the swap vnode so that the 550 * swap file can be on a different server from the rootfs. 551 */ 552 nd->swap_args.fh = nd->swap_fh; 553 nd->swap_args.fhsize = nd->swap_fhsize; 554 l = ntohl(nd->swap_saddr.sin_addr.s_addr); 555 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 556 (l >> 24) & 0xff, (l >> 16) & 0xff, 557 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam); 558 kprintf("NFS SWAP: %s\n",buf); 559 if ((error = nfs_mountdiskless(buf, "/swap", 0, 560 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) { 561 crit_exit(); 562 return (error); 563 } 564 vfs_unbusy(swap_mp); 565 566 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size = 567 nd->swap_nblks * DEV_BSIZE ; 568 569 /* 570 * Since the swap file is not the root dir of a file system, 571 * hack it to a regular file. 572 */ 573 vp->v_flag = 0; 574 vref(vp); 575 nfs_setvtype(vp, VREG); 576 swaponvp(td, vp, nd->swap_nblks); 577 } 578 579 mp->mnt_flag |= MNT_ROOTFS; 580 vfs_unbusy(mp); 581 582 /* 583 * This is not really an nfs issue, but it is much easier to 584 * set hostname here and then let the "/etc/rc.xxx" files 585 * mount the right /var based upon its preset value. 586 */ 587 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN); 588 hostname[MAXHOSTNAMELEN - 1] = '\0'; 589 for (i = 0; i < MAXHOSTNAMELEN; i++) 590 if (hostname[i] == '\0') 591 break; 592 inittodr(ntohl(nd->root_time)); 593 crit_exit(); 594 return (0); 595 } 596 597 /* 598 * Internal version of mount system call for diskless setup. 599 */ 600 static int 601 nfs_mountdiskless(char *path, char *which, int mountflag, 602 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td, 603 struct vnode **vpp, struct mount **mpp) 604 { 605 struct mount *mp; 606 struct sockaddr *nam; 607 int didalloc = 0; 608 int error; 609 610 mp = *mpp; 611 612 if (mp == NULL) { 613 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) { 614 kprintf("nfs_mountroot: NFS not configured"); 615 return (error); 616 } 617 didalloc = 1; 618 } 619 mp->mnt_kern_flag = 0; 620 mp->mnt_flag = mountflag; 621 nam = dup_sockaddr((struct sockaddr *)sin); 622 623 #if defined(BOOTP) || defined(NFS_ROOT) 624 if (args->fhsize == 0) { 625 kprintf("NFS_ROOT: No FH passed from loader, attempting mount rpc..."); 626 args->fhsize = 0; 627 error = md_mount(sin, which, args->fh, &args->fhsize, args, td); 628 if (error) { 629 kprintf("failed.\n"); 630 goto haderror; 631 } 632 kprintf("success!\n"); 633 } 634 #endif 635 636 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) { 637 #if defined(BOOTP) || defined(NFS_ROOT) 638 haderror: 639 #endif 640 kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error); 641 mp->mnt_vfc->vfc_refcount--; 642 vfs_unbusy(mp); 643 if (didalloc) 644 kfree(mp, M_MOUNT); 645 FREE(nam, M_SONAME); 646 return (error); 647 } 648 *mpp = mp; 649 return (0); 650 } 651 652 static void 653 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp) 654 { 655 int adjsock; 656 int maxio; 657 658 crit_enter(); 659 /* 660 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 661 * no sense in that context. 662 */ 663 if (argp->sotype == SOCK_STREAM) 664 nmp->nm_flag &= ~NFSMNT_NOCONN; 665 666 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */ 667 if ((argp->flags & NFSMNT_NFSV3) == 0) 668 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 669 670 /* Re-bind if rsrvd port requested and wasn't on one */ 671 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 672 && (argp->flags & NFSMNT_RESVPORT); 673 /* Also re-bind if we're switching to/from a connected UDP socket */ 674 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 675 (argp->flags & NFSMNT_NOCONN)); 676 677 /* Update flags atomically. Don't change the lock bits. */ 678 nmp->nm_flag = argp->flags | nmp->nm_flag; 679 crit_exit(); 680 681 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 682 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 683 if (nmp->nm_timeo < NFS_MINTIMEO) 684 nmp->nm_timeo = NFS_MINTIMEO; 685 else if (nmp->nm_timeo > NFS_MAXTIMEO) 686 nmp->nm_timeo = NFS_MAXTIMEO; 687 } 688 689 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 690 nmp->nm_retry = argp->retrans; 691 if (nmp->nm_retry > NFS_MAXREXMIT) 692 nmp->nm_retry = NFS_MAXREXMIT; 693 } 694 695 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype); 696 697 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 698 nmp->nm_wsize = argp->wsize; 699 /* Round down to multiple of blocksize */ 700 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 701 if (nmp->nm_wsize <= 0) 702 nmp->nm_wsize = NFS_FABLKSIZE; 703 } 704 if (nmp->nm_wsize > maxio) 705 nmp->nm_wsize = maxio; 706 if (nmp->nm_wsize > MAXBSIZE) 707 nmp->nm_wsize = MAXBSIZE; 708 709 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 710 nmp->nm_rsize = argp->rsize; 711 /* Round down to multiple of blocksize */ 712 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 713 if (nmp->nm_rsize <= 0) 714 nmp->nm_rsize = NFS_FABLKSIZE; 715 } 716 if (nmp->nm_rsize > maxio) 717 nmp->nm_rsize = maxio; 718 if (nmp->nm_rsize > MAXBSIZE) 719 nmp->nm_rsize = MAXBSIZE; 720 721 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 722 nmp->nm_readdirsize = argp->readdirsize; 723 } 724 if (nmp->nm_readdirsize > maxio) 725 nmp->nm_readdirsize = maxio; 726 if (nmp->nm_readdirsize > nmp->nm_rsize) 727 nmp->nm_readdirsize = nmp->nm_rsize; 728 729 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0) 730 nmp->nm_acregmin = argp->acregmin; 731 else 732 nmp->nm_acregmin = NFS_MINATTRTIMO; 733 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0) 734 nmp->nm_acregmax = argp->acregmax; 735 else 736 nmp->nm_acregmax = NFS_MAXATTRTIMO; 737 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0) 738 nmp->nm_acdirmin = argp->acdirmin; 739 else 740 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO; 741 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0) 742 nmp->nm_acdirmax = argp->acdirmax; 743 else 744 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO; 745 if (nmp->nm_acdirmin > nmp->nm_acdirmax) 746 nmp->nm_acdirmin = nmp->nm_acdirmax; 747 if (nmp->nm_acregmin > nmp->nm_acregmax) 748 nmp->nm_acregmin = nmp->nm_acregmax; 749 750 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) { 751 if (argp->maxgrouplist <= NFS_MAXGRPS) 752 nmp->nm_numgrps = argp->maxgrouplist; 753 else 754 nmp->nm_numgrps = NFS_MAXGRPS; 755 } 756 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) { 757 if (argp->readahead <= NFS_MAXRAHEAD) 758 nmp->nm_readahead = argp->readahead; 759 else 760 nmp->nm_readahead = NFS_MAXRAHEAD; 761 } 762 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) { 763 if (argp->deadthresh <= NFS_NEVERDEAD) 764 nmp->nm_deadthresh = argp->deadthresh; 765 else 766 nmp->nm_deadthresh = NFS_NEVERDEAD; 767 } 768 769 adjsock |= ((nmp->nm_sotype != argp->sotype) || 770 (nmp->nm_soproto != argp->proto)); 771 nmp->nm_sotype = argp->sotype; 772 nmp->nm_soproto = argp->proto; 773 774 if (nmp->nm_so && adjsock) { 775 nfs_safedisconnect(nmp); 776 if (nmp->nm_sotype == SOCK_DGRAM) 777 while (nfs_connect(nmp, (struct nfsreq *)0)) { 778 kprintf("nfs_args: retrying connect\n"); 779 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0); 780 } 781 } 782 } 783 784 /* 785 * VFS Operations. 786 * 787 * mount system call 788 * It seems a bit dumb to copyinstr() the host and path here and then 789 * bcopy() them in mountnfs(), but I wanted to detect errors before 790 * doing the sockargs() call because sockargs() allocates an mbuf and 791 * an error after that means that I have to release the mbuf. 792 */ 793 /* ARGSUSED */ 794 static int 795 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred) 796 { 797 int error; 798 struct nfs_args args; 799 struct sockaddr *nam; 800 struct vnode *vp; 801 char pth[MNAMELEN], hst[MNAMELEN]; 802 size_t len; 803 u_char nfh[NFSX_V3FHMAX]; 804 805 if (path == NULL) { 806 nfs_mountroot(mp); 807 return (0); 808 } 809 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)); 810 if (error) 811 return (error); 812 if (args.version != NFS_ARGSVERSION) { 813 #ifdef COMPAT_PRELITE2 814 /* 815 * If the argument version is unknown, then assume the 816 * caller is a pre-lite2 4.4BSD client and convert its 817 * arguments. 818 */ 819 struct onfs_args oargs; 820 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args)); 821 if (error) 822 return (error); 823 nfs_convert_oargs(&args,&oargs); 824 #else /* !COMPAT_PRELITE2 */ 825 return (EPROGMISMATCH); 826 #endif /* COMPAT_PRELITE2 */ 827 } 828 if (mp->mnt_flag & MNT_UPDATE) { 829 struct nfsmount *nmp = VFSTONFS(mp); 830 831 if (nmp == NULL) 832 return (EIO); 833 /* 834 * When doing an update, we can't change from or to 835 * v3, or change cookie translation 836 */ 837 args.flags = (args.flags & 838 ~(NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/)) | 839 (nmp->nm_flag & 840 (NFSMNT_NFSV3/*|NFSMNT_XLATECOOKIE*/)); 841 nfs_decode_args(nmp, &args); 842 return (0); 843 } 844 845 /* 846 * Make the nfs_ip_paranoia sysctl serve as the default connection 847 * or no-connection mode for those protocols that support 848 * no-connection mode (the flag will be cleared later for protocols 849 * that do not support no-connection mode). This will allow a client 850 * to receive replies from a different IP then the request was 851 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid), 852 * not 0. 853 */ 854 if (nfs_ip_paranoia == 0) 855 args.flags |= NFSMNT_NOCONN; 856 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) 857 return (EINVAL); 858 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); 859 if (error) 860 return (error); 861 error = copyinstr(path, pth, MNAMELEN-1, &len); 862 if (error) 863 return (error); 864 bzero(&pth[len], MNAMELEN - len); 865 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len); 866 if (error) 867 return (error); 868 bzero(&hst[len], MNAMELEN - len); 869 /* sockargs() call must be after above copyin() calls */ 870 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen); 871 if (error) 872 return (error); 873 args.fh = nfh; 874 error = mountnfs(&args, mp, nam, pth, hst, &vp); 875 return (error); 876 } 877 878 /* 879 * Common code for mount and mountroot 880 */ 881 static int 882 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam, 883 char *pth, char *hst, struct vnode **vpp) 884 { 885 struct nfsmount *nmp; 886 struct nfsnode *np; 887 int error; 888 889 if (mp->mnt_flag & MNT_UPDATE) { 890 nmp = VFSTONFS(mp); 891 /* update paths, file handles, etc, here XXX */ 892 FREE(nam, M_SONAME); 893 return (0); 894 } else { 895 nmp = zalloc(nfsmount_zone); 896 bzero((caddr_t)nmp, sizeof (struct nfsmount)); 897 TAILQ_INIT(&nmp->nm_uidlruhead); 898 TAILQ_INIT(&nmp->nm_bioq); 899 mp->mnt_data = (qaddr_t)nmp; 900 } 901 vfs_getnewfsid(mp); 902 nmp->nm_mountp = mp; 903 904 /* 905 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too 906 * high, depending on whether we end up with negative offsets in 907 * the client or server somewhere. 2GB-1 may be safer. 908 * 909 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum 910 * that we can handle until we find out otherwise. 911 * XXX Our "safe" limit on the client is what we can store in our 912 * buffer cache using signed(!) block numbers. 913 */ 914 if ((argp->flags & NFSMNT_NFSV3) == 0) 915 nmp->nm_maxfilesize = 0xffffffffLL; 916 else 917 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 918 919 nmp->nm_timeo = NFS_TIMEO; 920 nmp->nm_retry = NFS_RETRANS; 921 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype); 922 nmp->nm_rsize = nmp->nm_wsize; 923 nmp->nm_readdirsize = NFS_READDIRSIZE; 924 nmp->nm_numgrps = NFS_MAXGRPS; 925 nmp->nm_readahead = NFS_DEFRAHEAD; 926 nmp->nm_deadthresh = NFS_DEADTHRESH; 927 nmp->nm_fhsize = argp->fhsize; 928 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize); 929 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); 930 nmp->nm_nam = nam; 931 /* Set up the sockets and per-host congestion */ 932 nmp->nm_sotype = argp->sotype; 933 nmp->nm_soproto = argp->proto; 934 nmp->nm_cred = crhold(proc0.p_ucred); 935 936 nfs_decode_args(nmp, argp); 937 938 /* 939 * For Connection based sockets (TCP,...) defer the connect until 940 * the first request, in case the server is not responding. 941 */ 942 if (nmp->nm_sotype == SOCK_DGRAM && 943 (error = nfs_connect(nmp, (struct nfsreq *)0))) 944 goto bad; 945 946 /* 947 * This is silly, but it has to be set so that vinifod() works. 948 * We do not want to do an nfs_statfs() here since we can get 949 * stuck on a dead server and we are holding a lock on the mount 950 * point. 951 */ 952 mp->mnt_stat.f_iosize = 953 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype); 954 955 /* 956 * Install vop_ops for our vnops 957 */ 958 vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops); 959 vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops); 960 vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops); 961 962 /* 963 * A reference count is needed on the nfsnode representing the 964 * remote root. If this object is not persistent, then backward 965 * traversals of the mount point (i.e. "..") will not work if 966 * the nfsnode gets flushed out of the cache. Ufs does not have 967 * this problem, because one can identify root inodes by their 968 * number == ROOTINO (2). 969 */ 970 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 971 if (error) 972 goto bad; 973 *vpp = NFSTOV(np); 974 975 /* 976 * Retrieval of mountpoint attributes is delayed until nfs_rot 977 * or nfs_statfs are first called. This will happen either when 978 * we first traverse the mount point or if somebody does a df(1). 979 * 980 * NFSSTA_GOTFSINFO is used to flag if we have successfully 981 * retrieved mountpoint attributes. In the case of NFSv3 we 982 * also flag static fsinfo. 983 */ 984 if (*vpp != NULL) 985 (*vpp)->v_type = VNON; 986 987 /* 988 * Lose the lock but keep the ref. 989 */ 990 vn_unlock(*vpp); 991 992 return (0); 993 bad: 994 nfs_disconnect(nmp); 995 nfs_free_mount(nmp); 996 FREE(nam, M_SONAME); 997 return (error); 998 } 999 1000 /* 1001 * unmount system call 1002 */ 1003 static int 1004 nfs_unmount(struct mount *mp, int mntflags) 1005 { 1006 struct nfsmount *nmp; 1007 int error, flags = 0; 1008 1009 if (mntflags & MNT_FORCE) 1010 flags |= FORCECLOSE; 1011 nmp = VFSTONFS(mp); 1012 /* 1013 * Goes something like this.. 1014 * - Call vflush() to clear out vnodes for this file system 1015 * - Close the socket 1016 * - Free up the data structures 1017 */ 1018 /* In the forced case, cancel any outstanding requests. */ 1019 if (flags & FORCECLOSE) { 1020 error = nfs_nmcancelreqs(nmp); 1021 if (error) 1022 return (error); 1023 } 1024 /* 1025 * Must handshake with nfs_clientd() if it is active. XXX 1026 */ 1027 nmp->nm_state |= NFSSTA_DISMINPROG; 1028 1029 /* We hold 1 extra ref on the root vnode; see comment in mountnfs(). */ 1030 error = vflush(mp, 1, flags); 1031 if (error) { 1032 nmp->nm_state &= ~NFSSTA_DISMINPROG; 1033 return (error); 1034 } 1035 1036 /* 1037 * We are now committed to the unmount. 1038 * For NQNFS, let the server daemon free the nfsmount structure. 1039 */ 1040 if (nmp->nm_flag & NFSMNT_KERB) 1041 nmp->nm_state |= NFSSTA_DISMNT; 1042 1043 nfs_disconnect(nmp); 1044 FREE(nmp->nm_nam, M_SONAME); 1045 1046 if ((nmp->nm_flag & NFSMNT_KERB) == 0) 1047 nfs_free_mount(nmp); 1048 return (0); 1049 } 1050 1051 void 1052 nfs_free_mount(struct nfsmount *nmp) 1053 { 1054 if (nmp->nm_cred) { 1055 crfree(nmp->nm_cred); 1056 nmp->nm_cred = NULL; 1057 } 1058 zfree(nfsmount_zone, nmp); 1059 } 1060 1061 /* 1062 * Return root of a filesystem 1063 */ 1064 static int 1065 nfs_root(struct mount *mp, struct vnode **vpp) 1066 { 1067 struct vnode *vp; 1068 struct nfsmount *nmp; 1069 struct vattr attrs; 1070 struct nfsnode *np; 1071 int error; 1072 1073 nmp = VFSTONFS(mp); 1074 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 1075 if (error) 1076 return (error); 1077 vp = NFSTOV(np); 1078 1079 /* 1080 * Get transfer parameters and root vnode attributes 1081 */ 1082 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { 1083 if (nmp->nm_flag & NFSMNT_NFSV3) { 1084 error = nfs_fsinfo(nmp, vp, curthread); 1085 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype); 1086 } else { 1087 if ((error = VOP_GETATTR(vp, &attrs)) == 0) 1088 nmp->nm_state |= NFSSTA_GOTFSINFO; 1089 1090 } 1091 } else { 1092 /* 1093 * The root vnode is usually cached by the namecache so do not 1094 * try to avoid going over the wire even if we have previous 1095 * information cached. A stale NFS mount can loop 1096 * forever resolving the root vnode if we return no-error when 1097 * there is in fact an error. 1098 */ 1099 np->n_attrstamp = 0; 1100 error = VOP_GETATTR(vp, &attrs); 1101 } 1102 if (vp->v_type == VNON) 1103 nfs_setvtype(vp, VDIR); 1104 vp->v_flag = VROOT; 1105 if (error) 1106 vput(vp); 1107 else 1108 *vpp = vp; 1109 return (error); 1110 } 1111 1112 extern int syncprt; 1113 1114 struct scaninfo { 1115 int rescan; 1116 int waitfor; 1117 int allerror; 1118 }; 1119 1120 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data); 1121 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data); 1122 1123 /* 1124 * Flush out the buffer cache 1125 */ 1126 /* ARGSUSED */ 1127 static int 1128 nfs_sync(struct mount *mp, int waitfor) 1129 { 1130 struct scaninfo scaninfo; 1131 int error; 1132 1133 scaninfo.rescan = 0; 1134 scaninfo.waitfor = waitfor; 1135 scaninfo.allerror = 0; 1136 1137 /* 1138 * Force stale buffer cache information to be flushed. 1139 */ 1140 error = 0; 1141 while (error == 0 && scaninfo.rescan) { 1142 scaninfo.rescan = 0; 1143 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1, 1144 nfs_sync_scan2, &scaninfo); 1145 } 1146 return(error); 1147 } 1148 1149 static int 1150 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data) 1151 { 1152 struct scaninfo *info = data; 1153 1154 if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree)) 1155 return(-1); 1156 if (info->waitfor == MNT_LAZY) 1157 return(-1); 1158 return(0); 1159 } 1160 1161 static int 1162 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data) 1163 { 1164 struct scaninfo *info = data; 1165 int error; 1166 1167 error = VOP_FSYNC(vp, info->waitfor); 1168 if (error) 1169 info->allerror = error; 1170 return(0); 1171 } 1172 1173