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 1.54 2008/07/31 20:23:40 swildner 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 #include <sys/mutex2.h> 68 69 #include "rpcv2.h" 70 #include "nfsproto.h" 71 #include "nfs.h" 72 #include "nfsmount.h" 73 #include "nfsnode.h" 74 #include "xdr_subs.h" 75 #include "nfsm_subs.h" 76 #include "nfsdiskless.h" 77 #include "nfsmountrpc.h" 78 79 extern int nfs_mountroot(struct mount *mp); 80 extern void bootpc_init(void); 81 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_statvfs(struct mount *mp, struct statvfs *sbp, 127 struct ucred *cred); 128 static int nfs_sync ( struct mount *mp, int waitfor); 129 130 /* 131 * nfs vfs operations. 132 */ 133 static struct vfsops nfs_vfsops = { 134 .vfs_mount = nfs_mount, 135 .vfs_unmount = nfs_unmount, 136 .vfs_root = nfs_root, 137 .vfs_statfs = nfs_statfs, 138 .vfs_statvfs = nfs_statvfs, 139 .vfs_sync = nfs_sync, 140 .vfs_init = nfs_init, 141 .vfs_uninit = nfs_uninit 142 }; 143 VFS_SET(nfs_vfsops, nfs, VFCF_NETWORK); 144 145 /* 146 * This structure must be filled in by a primary bootstrap or bootstrap 147 * server for a diskless/dataless machine. It is initialized below just 148 * to ensure that it is allocated to initialized data (.data not .bss). 149 */ 150 struct nfs_diskless nfs_diskless = { { { 0 } } }; 151 struct nfsv3_diskless nfsv3_diskless = { { { 0 } } }; 152 int nfs_diskless_valid = 0; 153 154 SYSCTL_INT(_vfs_nfs, OID_AUTO, diskless_valid, CTLFLAG_RD, 155 &nfs_diskless_valid, 0, ""); 156 157 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_rootpath, CTLFLAG_RD, 158 nfsv3_diskless.root_hostnam, 0, ""); 159 160 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_rootaddr, CTLFLAG_RD, 161 &nfsv3_diskless.root_saddr, sizeof nfsv3_diskless.root_saddr, 162 "%Ssockaddr_in", ""); 163 164 SYSCTL_STRING(_vfs_nfs, OID_AUTO, diskless_swappath, CTLFLAG_RD, 165 nfsv3_diskless.swap_hostnam, 0, ""); 166 167 SYSCTL_OPAQUE(_vfs_nfs, OID_AUTO, diskless_swapaddr, CTLFLAG_RD, 168 &nfsv3_diskless.swap_saddr, sizeof nfsv3_diskless.swap_saddr, 169 "%Ssockaddr_in",""); 170 171 172 void nfsargs_ntoh (struct nfs_args *); 173 static int nfs_mountdiskless (char *, char *, int, 174 struct sockaddr_in *, struct nfs_args *, 175 struct thread *, struct vnode **, 176 struct mount **); 177 static void nfs_convert_diskless (void); 178 static void nfs_convert_oargs (struct nfs_args *args, 179 struct onfs_args *oargs); 180 181 /* 182 * Calculate the buffer I/O block size to use. The maximum V2 block size 183 * is typically 8K, the maximum datagram size is typically 16K, and the 184 * maximum V3 block size is typically 32K. The buffer cache tends to work 185 * best with 16K blocks but we allow 32K for TCP connections. 186 * 187 * We force the block size to be at least a page for buffer cache efficiency. 188 */ 189 static int 190 nfs_iosize(int v3, int sotype) 191 { 192 int iosize; 193 int iomax; 194 195 if (v3) { 196 if (sotype == SOCK_STREAM) 197 iomax = NFS_MAXDATA; 198 else 199 iomax = NFS_MAXDGRAMDATA; 200 } else { 201 iomax = NFS_V2MAXDATA; 202 } 203 if ((iosize = nfs_io_size) > iomax) 204 iosize = iomax; 205 if (iosize < PAGE_SIZE) 206 iosize = PAGE_SIZE; 207 208 /* 209 * This is an aweful hack but until the buffer cache is rewritten 210 * we need it. The problem is that when you combine write() with 211 * mmap() the vm_page->valid bits can become weird looking 212 * (e.g. 0xfc). This occurs because NFS uses piecemeal buffers 213 * at the file EOF. To solve the problem the BIO system needs to 214 * be guarenteed that the NFS iosize for regular files will be a 215 * multiple of PAGE_SIZE so it can invalidate the whole page 216 * rather then just the piece of it owned by the buffer when 217 * NFS does vinvalbuf() calls. 218 */ 219 if (iosize & PAGE_MASK) 220 iosize = (iosize & ~PAGE_MASK) + PAGE_SIZE; 221 return iosize; 222 } 223 224 static void 225 nfs_convert_oargs(struct nfs_args *args, struct onfs_args *oargs) 226 { 227 args->version = NFS_ARGSVERSION; 228 args->addr = oargs->addr; 229 args->addrlen = oargs->addrlen; 230 args->sotype = oargs->sotype; 231 args->proto = oargs->proto; 232 args->fh = oargs->fh; 233 args->fhsize = oargs->fhsize; 234 args->flags = oargs->flags; 235 args->wsize = oargs->wsize; 236 args->rsize = oargs->rsize; 237 args->readdirsize = oargs->readdirsize; 238 args->timeo = oargs->timeo; 239 args->retrans = oargs->retrans; 240 args->maxgrouplist = oargs->maxgrouplist; 241 args->readahead = oargs->readahead; 242 args->deadthresh = oargs->deadthresh; 243 args->hostname = oargs->hostname; 244 } 245 246 static void 247 nfs_convert_diskless(void) 248 { 249 int i; 250 251 bcopy(&nfs_diskless.myif, &nfsv3_diskless.myif, 252 sizeof(struct ifaliasreq)); 253 bcopy(&nfs_diskless.mygateway, &nfsv3_diskless.mygateway, 254 sizeof(struct sockaddr_in)); 255 nfs_convert_oargs(&nfsv3_diskless.swap_args,&nfs_diskless.swap_args); 256 257 bcopy(nfs_diskless.swap_fh,nfsv3_diskless.swap_fh,NFSX_V2FH); 258 nfsv3_diskless.swap_fhsize = NFSX_V2FH; 259 for (i = NFSX_V2FH - 1; i >= 0; --i) { 260 if (nfs_diskless.swap_fh[i]) 261 break; 262 } 263 if (i < 0) 264 nfsv3_diskless.swap_fhsize = 0; 265 266 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr, 267 sizeof(struct sockaddr_in)); 268 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN); 269 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks; 270 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred, 271 sizeof(struct ucred)); 272 nfs_convert_oargs(&nfsv3_diskless.root_args,&nfs_diskless.root_args); 273 274 bcopy(nfs_diskless.root_fh,nfsv3_diskless.root_fh,NFSX_V2FH); 275 nfsv3_diskless.root_fhsize = NFSX_V2FH; 276 for (i = NFSX_V2FH - 1; i >= 0; --i) { 277 if (nfs_diskless.root_fh[i]) 278 break; 279 } 280 if (i < 0) 281 nfsv3_diskless.root_fhsize = 0; 282 283 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr, 284 sizeof(struct sockaddr_in)); 285 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN); 286 nfsv3_diskless.root_time = nfs_diskless.root_time; 287 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam, 288 MAXHOSTNAMELEN); 289 nfs_diskless_valid = 3; 290 } 291 292 /* 293 * nfs statfs call 294 */ 295 int 296 nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred) 297 { 298 struct vnode *vp; 299 struct nfs_statfs *sfp; 300 struct nfsmount *nmp = VFSTONFS(mp); 301 thread_t td = curthread; 302 int error = 0, retattr; 303 struct nfsnode *np; 304 u_quad_t tquad; 305 struct nfsm_info info; 306 307 info.mrep = NULL; 308 info.v3 = (nmp->nm_flag & NFSMNT_NFSV3); 309 310 #ifndef nolint 311 sfp = NULL; 312 #endif 313 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 314 if (error) 315 return (error); 316 vp = NFSTOV(np); 317 /* ignore the passed cred */ 318 cred = crget(); 319 cred->cr_ngroups = 1; 320 if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 321 (void)nfs_fsinfo(nmp, vp, td); 322 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 323 nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3)); 324 ERROROUT(nfsm_fhtom(&info, vp)); 325 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error)); 326 if (info.v3) { 327 ERROROUT(nfsm_postop_attr(&info, vp, &retattr, 328 NFS_LATTR_NOSHRINK)); 329 } 330 if (error) { 331 if (info.mrep != NULL) 332 m_freem(info.mrep); 333 goto nfsmout; 334 } 335 NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3))); 336 sbp->f_flags = nmp->nm_flag; 337 sbp->f_iosize = nfs_iosize(info.v3, nmp->nm_sotype); 338 339 if (info.v3) { 340 sbp->f_bsize = NFS_FABLKSIZE; 341 tquad = fxdr_hyper(&sfp->sf_tbytes); 342 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 343 tquad = fxdr_hyper(&sfp->sf_fbytes); 344 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 345 tquad = fxdr_hyper(&sfp->sf_abytes); 346 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 347 sbp->f_files = (fxdr_unsigned(int32_t, 348 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff); 349 sbp->f_ffree = (fxdr_unsigned(int32_t, 350 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff); 351 } else { 352 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 353 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 354 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 355 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 356 sbp->f_files = 0; 357 sbp->f_ffree = 0; 358 } 359 if (sbp != &mp->mnt_stat) { 360 sbp->f_type = mp->mnt_vfc->vfc_typenum; 361 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); 362 } 363 m_freem(info.mrep); 364 info.mrep = NULL; 365 nfsmout: 366 vput(vp); 367 crfree(cred); 368 return (error); 369 } 370 371 static int 372 nfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred) 373 { 374 struct vnode *vp; 375 struct nfs_statfs *sfp; 376 struct nfsmount *nmp = VFSTONFS(mp); 377 thread_t td = curthread; 378 int error = 0, retattr; 379 struct nfsnode *np; 380 struct nfsm_info info; 381 382 info.mrep = NULL; 383 info.v3 = (nmp->nm_flag & NFSMNT_NFSV3); 384 385 #ifndef nolint 386 sfp = NULL; 387 #endif 388 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 389 if (error) 390 return (error); 391 vp = NFSTOV(np); 392 /* ignore the passed cred */ 393 cred = crget(); 394 cred->cr_ngroups = 1; 395 if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 396 (void)nfs_fsinfo(nmp, vp, td); 397 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 398 nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3)); 399 ERROROUT(nfsm_fhtom(&info, vp)); 400 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error)); 401 if (info.v3) { 402 ERROROUT(nfsm_postop_attr(&info, vp, &retattr, 403 NFS_LATTR_NOSHRINK)); 404 } 405 if (error) { 406 if (info.mrep != NULL) 407 m_freem(info.mrep); 408 goto nfsmout; 409 } 410 NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3))); 411 sbp->f_flag = nmp->nm_flag; 412 sbp->f_owner = nmp->nm_cred->cr_ruid; 413 414 if (info.v3) { 415 sbp->f_bsize = NFS_FABLKSIZE; 416 sbp->f_frsize = NFS_FABLKSIZE; 417 sbp->f_blocks = (fxdr_hyper(&sfp->sf_tbytes) / 418 ((u_quad_t)NFS_FABLKSIZE)); 419 sbp->f_bfree = (fxdr_hyper(&sfp->sf_fbytes) / 420 ((u_quad_t)NFS_FABLKSIZE)); 421 sbp->f_bavail = (fxdr_hyper(&sfp->sf_abytes) / 422 ((u_quad_t)NFS_FABLKSIZE)); 423 sbp->f_files = fxdr_hyper(&sfp->sf_tfiles); 424 sbp->f_ffree = fxdr_hyper(&sfp->sf_ffiles); 425 sbp->f_favail = fxdr_hyper(&sfp->sf_afiles); 426 } else { 427 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 428 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 429 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 430 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 431 sbp->f_files = 0; 432 sbp->f_ffree = 0; 433 sbp->f_favail = 0; 434 } 435 sbp->f_syncreads = 0; 436 sbp->f_syncwrites = 0; 437 sbp->f_asyncreads = 0; 438 sbp->f_asyncwrites = 0; 439 sbp->f_type = mp->mnt_vfc->vfc_typenum; 440 441 m_freem(info.mrep); 442 info.mrep = NULL; 443 nfsmout: 444 vput(vp); 445 crfree(cred); 446 return (error); 447 } 448 449 /* 450 * nfs version 3 fsinfo rpc call 451 */ 452 int 453 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td) 454 { 455 struct nfsv3_fsinfo *fsp; 456 u_int32_t pref, max; 457 int error = 0, retattr; 458 u_int64_t maxfsize; 459 struct nfsm_info info; 460 461 info.v3 = 1; 462 nfsstats.rpccnt[NFSPROC_FSINFO]++; 463 nfsm_reqhead(&info, vp, NFSPROC_FSINFO, NFSX_FH(1)); 464 ERROROUT(nfsm_fhtom(&info, vp)); 465 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSINFO, td, 466 nfs_vpcred(vp, ND_READ), &error)); 467 ERROROUT(nfsm_postop_attr(&info, vp, &retattr, NFS_LATTR_NOSHRINK)); 468 if (error == 0) { 469 NULLOUT(fsp = nfsm_dissect(&info, NFSX_V3FSINFO)); 470 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 471 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 472 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 473 ~(NFS_FABLKSIZE - 1); 474 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 475 if (max < nmp->nm_wsize && max > 0) { 476 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1); 477 if (nmp->nm_wsize == 0) 478 nmp->nm_wsize = max; 479 } 480 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 481 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 482 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 483 ~(NFS_FABLKSIZE - 1); 484 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 485 if (max < nmp->nm_rsize && max > 0) { 486 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1); 487 if (nmp->nm_rsize == 0) 488 nmp->nm_rsize = max; 489 } 490 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 491 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ) 492 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) & 493 ~(NFS_DIRBLKSIZ - 1); 494 if (max < nmp->nm_readdirsize && max > 0) { 495 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1); 496 if (nmp->nm_readdirsize == 0) 497 nmp->nm_readdirsize = max; 498 } 499 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 500 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 501 nmp->nm_maxfilesize = maxfsize; 502 nmp->nm_state |= NFSSTA_GOTFSINFO; 503 } 504 m_freem(info.mrep); 505 info.mrep = NULL; 506 nfsmout: 507 return (error); 508 } 509 510 /* 511 * Mount a remote root fs via. nfs. This depends on the info in the 512 * nfs_diskless structure that has been filled in properly by some primary 513 * bootstrap. 514 * It goes something like this: 515 * - do enough of "ifconfig" by calling ifioctl() so that the system 516 * can talk to the server 517 * - If nfs_diskless.mygateway is filled in, use that address as 518 * a default gateway. 519 * - build the rootfs mount point and call mountnfs() to do the rest. 520 */ 521 int 522 nfs_mountroot(struct mount *mp) 523 { 524 struct mount *swap_mp; 525 struct nfsv3_diskless *nd = &nfsv3_diskless; 526 struct socket *so; 527 struct vnode *vp; 528 struct thread *td = curthread; /* XXX */ 529 int error, i; 530 u_long l; 531 char buf[128]; 532 533 #if defined(BOOTP_NFSROOT) && defined(BOOTP) 534 bootpc_init(); /* use bootp to get nfs_diskless filled in */ 535 #endif 536 537 /* 538 * XXX time must be non-zero when we init the interface or else 539 * the arp code will wedge... 540 */ 541 while (mycpu->gd_time_seconds == 0) 542 tsleep(mycpu, 0, "arpkludge", 10); 543 544 /* 545 * The boot code may have passed us a diskless structure. 546 */ 547 if (nfs_diskless_valid == 1) 548 nfs_convert_diskless(); 549 550 #define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr)) 551 kprintf("nfs_mountroot: interface %s ip %s", 552 nd->myif.ifra_name, 553 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr)); 554 kprintf(" bcast %s", 555 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr)); 556 kprintf(" mask %s\n", 557 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr)); 558 #undef SINP 559 560 /* 561 * XXX splnet, so networks will receive... 562 */ 563 crit_enter(); 564 565 /* 566 * BOOTP does not necessarily have to be compiled into the kernel 567 * for an NFS root to work. If we inherited the network 568 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured 569 * out our interface for us and all we need to do is ifconfig the 570 * interface. We only do this if the interface has not already been 571 * ifconfig'd by e.g. BOOTP. 572 */ 573 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td); 574 if (error) { 575 panic("nfs_mountroot: socreate(%04x): %d", 576 nd->myif.ifra_addr.sa_family, error); 577 } 578 579 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred); 580 if (error) 581 panic("nfs_mountroot: SIOCAIFADDR: %d", error); 582 583 soclose(so, FNONBLOCK); 584 585 /* 586 * If the gateway field is filled in, set it as the default route. 587 */ 588 if (nd->mygateway.sin_len != 0) { 589 struct sockaddr_in mask, sin; 590 591 bzero((caddr_t)&mask, sizeof(mask)); 592 sin = mask; 593 sin.sin_family = AF_INET; 594 sin.sin_len = sizeof(sin); 595 kprintf("nfs_mountroot: gateway %s\n", 596 inet_ntoa(nd->mygateway.sin_addr)); 597 error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin, 598 (struct sockaddr *)&nd->mygateway, 599 (struct sockaddr *)&mask, 600 RTF_UP | RTF_GATEWAY); 601 if (error) 602 kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error); 603 } 604 605 /* 606 * Create the rootfs mount point. 607 */ 608 nd->root_args.fh = nd->root_fh; 609 nd->root_args.fhsize = nd->root_fhsize; 610 l = ntohl(nd->root_saddr.sin_addr.s_addr); 611 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 612 (l >> 24) & 0xff, (l >> 16) & 0xff, 613 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam); 614 kprintf("NFS_ROOT: %s\n",buf); 615 if ((error = nfs_mountdiskless(buf, "/", MNT_RDONLY, 616 &nd->root_saddr, &nd->root_args, td, &vp, &mp)) != 0) { 617 mp->mnt_vfc->vfc_refcount--; 618 crit_exit(); 619 return (error); 620 } 621 622 swap_mp = NULL; 623 if (nd->swap_nblks) { 624 625 /* Convert to DEV_BSIZE instead of Kilobyte */ 626 nd->swap_nblks *= 2; 627 628 /* 629 * Create a fake mount point just for the swap vnode so that the 630 * swap file can be on a different server from the rootfs. 631 */ 632 nd->swap_args.fh = nd->swap_fh; 633 nd->swap_args.fhsize = nd->swap_fhsize; 634 l = ntohl(nd->swap_saddr.sin_addr.s_addr); 635 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 636 (l >> 24) & 0xff, (l >> 16) & 0xff, 637 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam); 638 kprintf("NFS SWAP: %s\n",buf); 639 if ((error = nfs_mountdiskless(buf, "/swap", 0, 640 &nd->swap_saddr, &nd->swap_args, td, &vp, &swap_mp)) != 0) { 641 crit_exit(); 642 return (error); 643 } 644 vfs_unbusy(swap_mp); 645 646 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size = 647 nd->swap_nblks * DEV_BSIZE ; 648 649 /* 650 * Since the swap file is not the root dir of a file system, 651 * hack it to a regular file. 652 */ 653 vp->v_flag = 0; 654 vref(vp); 655 nfs_setvtype(vp, VREG); 656 swaponvp(td, vp, nd->swap_nblks); 657 } 658 659 mp->mnt_flag |= MNT_ROOTFS; 660 vfs_unbusy(mp); 661 662 /* 663 * This is not really an nfs issue, but it is much easier to 664 * set hostname here and then let the "/etc/rc.xxx" files 665 * mount the right /var based upon its preset value. 666 */ 667 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN); 668 hostname[MAXHOSTNAMELEN - 1] = '\0'; 669 for (i = 0; i < MAXHOSTNAMELEN; i++) 670 if (hostname[i] == '\0') 671 break; 672 inittodr(ntohl(nd->root_time)); 673 crit_exit(); 674 return (0); 675 } 676 677 /* 678 * Internal version of mount system call for diskless setup. 679 */ 680 static int 681 nfs_mountdiskless(char *path, char *which, int mountflag, 682 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td, 683 struct vnode **vpp, struct mount **mpp) 684 { 685 struct mount *mp; 686 struct sockaddr *nam; 687 int didalloc = 0; 688 int error; 689 690 mp = *mpp; 691 692 if (mp == NULL) { 693 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) { 694 kprintf("nfs_mountroot: NFS not configured"); 695 return (error); 696 } 697 didalloc = 1; 698 } 699 mp->mnt_kern_flag = 0; 700 mp->mnt_flag = mountflag; 701 nam = dup_sockaddr((struct sockaddr *)sin); 702 703 #if defined(BOOTP) || defined(NFS_ROOT) 704 if (args->fhsize == 0) { 705 char *xpath = path; 706 707 kprintf("NFS_ROOT: No FH passed from loader, attempting mount rpc..."); 708 while (*xpath && *xpath != ':') 709 ++xpath; 710 if (*xpath) 711 ++xpath; 712 args->fhsize = 0; 713 error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td); 714 if (error) { 715 kprintf("failed error %d.\n", error); 716 goto haderror; 717 } 718 kprintf("success!\n"); 719 } 720 #endif 721 722 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) { 723 #if defined(BOOTP) || defined(NFS_ROOT) 724 haderror: 725 #endif 726 kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error); 727 mp->mnt_vfc->vfc_refcount--; 728 vfs_unbusy(mp); 729 if (didalloc) 730 kfree(mp, M_MOUNT); 731 FREE(nam, M_SONAME); 732 return (error); 733 } 734 *mpp = mp; 735 return (0); 736 } 737 738 static void 739 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp) 740 { 741 int adjsock; 742 int maxio; 743 744 crit_enter(); 745 /* 746 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 747 * no sense in that context. 748 */ 749 if (nmp->nm_sotype == SOCK_STREAM) 750 nmp->nm_flag &= ~NFSMNT_NOCONN; 751 752 /* Also clear RDIRPLUS if not NFSv3, it crashes some servers */ 753 if ((argp->flags & NFSMNT_NFSV3) == 0) 754 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 755 756 /* 757 * Re-bind if rsrvd port flag has changed 758 */ 759 adjsock = (nmp->nm_flag & NFSMNT_RESVPORT) != 760 (argp->flags & NFSMNT_RESVPORT); 761 762 /* Update flags atomically. Don't change the lock bits. */ 763 nmp->nm_flag = argp->flags | nmp->nm_flag; 764 crit_exit(); 765 766 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 767 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 768 if (nmp->nm_timeo < NFS_MINTIMEO) 769 nmp->nm_timeo = NFS_MINTIMEO; 770 else if (nmp->nm_timeo > NFS_MAXTIMEO) 771 nmp->nm_timeo = NFS_MAXTIMEO; 772 } 773 774 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 775 nmp->nm_retry = argp->retrans; 776 if (nmp->nm_retry > NFS_MAXREXMIT) 777 nmp->nm_retry = NFS_MAXREXMIT; 778 } 779 780 /* 781 * These parameters effect the buffer cache and cannot be changed 782 * once we've successfully mounted. 783 */ 784 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { 785 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, nmp->nm_sotype); 786 787 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 788 nmp->nm_wsize = argp->wsize; 789 /* Round down to multiple of blocksize */ 790 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 791 if (nmp->nm_wsize <= 0) 792 nmp->nm_wsize = NFS_FABLKSIZE; 793 } 794 if (nmp->nm_wsize > maxio) 795 nmp->nm_wsize = maxio; 796 if (nmp->nm_wsize > MAXBSIZE) 797 nmp->nm_wsize = MAXBSIZE; 798 799 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 800 nmp->nm_rsize = argp->rsize; 801 /* Round down to multiple of blocksize */ 802 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 803 if (nmp->nm_rsize <= 0) 804 nmp->nm_rsize = NFS_FABLKSIZE; 805 } 806 if (nmp->nm_rsize > maxio) 807 nmp->nm_rsize = maxio; 808 if (nmp->nm_rsize > MAXBSIZE) 809 nmp->nm_rsize = MAXBSIZE; 810 811 if ((argp->flags & NFSMNT_READDIRSIZE) && 812 argp->readdirsize > 0) { 813 nmp->nm_readdirsize = argp->readdirsize; 814 } 815 if (nmp->nm_readdirsize > maxio) 816 nmp->nm_readdirsize = maxio; 817 if (nmp->nm_readdirsize > nmp->nm_rsize) 818 nmp->nm_readdirsize = nmp->nm_rsize; 819 } 820 821 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0) 822 nmp->nm_acregmin = argp->acregmin; 823 else 824 nmp->nm_acregmin = NFS_MINATTRTIMO; 825 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0) 826 nmp->nm_acregmax = argp->acregmax; 827 else 828 nmp->nm_acregmax = NFS_MAXATTRTIMO; 829 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0) 830 nmp->nm_acdirmin = argp->acdirmin; 831 else 832 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO; 833 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0) 834 nmp->nm_acdirmax = argp->acdirmax; 835 else 836 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO; 837 if (nmp->nm_acdirmin > nmp->nm_acdirmax) 838 nmp->nm_acdirmin = nmp->nm_acdirmax; 839 if (nmp->nm_acregmin > nmp->nm_acregmax) 840 nmp->nm_acregmin = nmp->nm_acregmax; 841 842 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) { 843 if (argp->maxgrouplist <= NFS_MAXGRPS) 844 nmp->nm_numgrps = argp->maxgrouplist; 845 else 846 nmp->nm_numgrps = NFS_MAXGRPS; 847 } 848 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) { 849 if (argp->readahead <= NFS_MAXRAHEAD) 850 nmp->nm_readahead = argp->readahead; 851 else 852 nmp->nm_readahead = NFS_MAXRAHEAD; 853 } 854 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) { 855 if (argp->deadthresh <= NFS_NEVERDEAD) 856 nmp->nm_deadthresh = argp->deadthresh; 857 else 858 nmp->nm_deadthresh = NFS_NEVERDEAD; 859 } 860 861 if (nmp->nm_so && adjsock) { 862 nfs_safedisconnect(nmp); 863 if (nmp->nm_sotype == SOCK_DGRAM) 864 while (nfs_connect(nmp, NULL)) { 865 kprintf("nfs_args: retrying connect\n"); 866 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0); 867 } 868 } 869 } 870 871 /* 872 * VFS Operations. 873 * 874 * mount system call 875 * It seems a bit dumb to copyinstr() the host and path here and then 876 * bcopy() them in mountnfs(), but I wanted to detect errors before 877 * doing the sockargs() call because sockargs() allocates an mbuf and 878 * an error after that means that I have to release the mbuf. 879 */ 880 /* ARGSUSED */ 881 static int 882 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred) 883 { 884 int error; 885 struct nfs_args args; 886 struct sockaddr *nam; 887 struct vnode *vp; 888 char pth[MNAMELEN], hst[MNAMELEN]; 889 size_t len; 890 u_char nfh[NFSX_V3FHMAX]; 891 892 if (path == NULL) { 893 nfs_mountroot(mp); 894 return (0); 895 } 896 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)); 897 if (error) 898 return (error); 899 if (args.version != NFS_ARGSVERSION) { 900 #ifdef COMPAT_PRELITE2 901 /* 902 * If the argument version is unknown, then assume the 903 * caller is a pre-lite2 4.4BSD client and convert its 904 * arguments. 905 */ 906 struct onfs_args oargs; 907 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args)); 908 if (error) 909 return (error); 910 nfs_convert_oargs(&args,&oargs); 911 #else /* !COMPAT_PRELITE2 */ 912 return (EPROGMISMATCH); 913 #endif /* COMPAT_PRELITE2 */ 914 } 915 if (mp->mnt_flag & MNT_UPDATE) { 916 struct nfsmount *nmp = VFSTONFS(mp); 917 918 if (nmp == NULL) 919 return (EIO); 920 /* 921 * When doing an update, we can't change from or to 922 * v3, or change cookie translation, or rsize or wsize. 923 */ 924 args.flags &= ~(NFSMNT_NFSV3 | NFSMNT_RSIZE | NFSMNT_WSIZE); 925 args.flags |= nmp->nm_flag & (NFSMNT_NFSV3); 926 nfs_decode_args(nmp, &args); 927 return (0); 928 } 929 930 /* 931 * Make the nfs_ip_paranoia sysctl serve as the default connection 932 * or no-connection mode for those protocols that support 933 * no-connection mode (the flag will be cleared later for protocols 934 * that do not support no-connection mode). This will allow a client 935 * to receive replies from a different IP then the request was 936 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid), 937 * not 0. 938 */ 939 if (nfs_ip_paranoia == 0) 940 args.flags |= NFSMNT_NOCONN; 941 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) 942 return (EINVAL); 943 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); 944 if (error) 945 return (error); 946 error = copyinstr(path, pth, MNAMELEN-1, &len); 947 if (error) 948 return (error); 949 bzero(&pth[len], MNAMELEN - len); 950 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len); 951 if (error) 952 return (error); 953 bzero(&hst[len], MNAMELEN - len); 954 /* sockargs() call must be after above copyin() calls */ 955 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen); 956 if (error) 957 return (error); 958 args.fh = nfh; 959 error = mountnfs(&args, mp, nam, pth, hst, &vp); 960 return (error); 961 } 962 963 /* 964 * Common code for mount and mountroot 965 */ 966 static int 967 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam, 968 char *pth, char *hst, struct vnode **vpp) 969 { 970 struct nfsmount *nmp; 971 struct nfsnode *np; 972 int error; 973 int rxcpu; 974 int txcpu; 975 976 if (mp->mnt_flag & MNT_UPDATE) { 977 nmp = VFSTONFS(mp); 978 /* update paths, file handles, etc, here XXX */ 979 FREE(nam, M_SONAME); 980 return (0); 981 } else { 982 nmp = zalloc(nfsmount_zone); 983 bzero((caddr_t)nmp, sizeof (struct nfsmount)); 984 mtx_init(&nmp->nm_rxlock); 985 mtx_init(&nmp->nm_txlock); 986 TAILQ_INIT(&nmp->nm_uidlruhead); 987 TAILQ_INIT(&nmp->nm_bioq); 988 TAILQ_INIT(&nmp->nm_reqq); 989 TAILQ_INIT(&nmp->nm_reqtxq); 990 TAILQ_INIT(&nmp->nm_reqrxq); 991 mp->mnt_data = (qaddr_t)nmp; 992 } 993 vfs_getnewfsid(mp); 994 nmp->nm_mountp = mp; 995 996 /* 997 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too 998 * high, depending on whether we end up with negative offsets in 999 * the client or server somewhere. 2GB-1 may be safer. 1000 * 1001 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum 1002 * that we can handle until we find out otherwise. 1003 * XXX Our "safe" limit on the client is what we can store in our 1004 * buffer cache using signed(!) block numbers. 1005 */ 1006 if ((argp->flags & NFSMNT_NFSV3) == 0) 1007 nmp->nm_maxfilesize = 0xffffffffLL; 1008 else 1009 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 1010 1011 nmp->nm_timeo = NFS_TIMEO; 1012 nmp->nm_retry = NFS_RETRANS; 1013 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype); 1014 nmp->nm_rsize = nmp->nm_wsize; 1015 nmp->nm_readdirsize = NFS_READDIRSIZE; 1016 nmp->nm_numgrps = NFS_MAXGRPS; 1017 nmp->nm_readahead = NFS_DEFRAHEAD; 1018 nmp->nm_deadthresh = NFS_DEADTHRESH; 1019 nmp->nm_fhsize = argp->fhsize; 1020 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize); 1021 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); 1022 nmp->nm_nam = nam; 1023 /* Set up the sockets and per-host congestion */ 1024 nmp->nm_sotype = argp->sotype; 1025 nmp->nm_soproto = argp->proto; 1026 nmp->nm_cred = crhold(proc0.p_ucred); 1027 1028 nfs_decode_args(nmp, argp); 1029 1030 /* 1031 * For Connection based sockets (TCP,...) defer the connect until 1032 * the first request, in case the server is not responding. 1033 */ 1034 if (nmp->nm_sotype == SOCK_DGRAM && 1035 (error = nfs_connect(nmp, NULL))) 1036 goto bad; 1037 1038 /* 1039 * This is silly, but it has to be set so that vinifod() works. 1040 * We do not want to do an nfs_statfs() here since we can get 1041 * stuck on a dead server and we are holding a lock on the mount 1042 * point. 1043 */ 1044 mp->mnt_stat.f_iosize = 1045 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype); 1046 1047 /* 1048 * Install vop_ops for our vnops 1049 */ 1050 vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops); 1051 vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops); 1052 vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops); 1053 1054 /* 1055 * A reference count is needed on the nfsnode representing the 1056 * remote root. If this object is not persistent, then backward 1057 * traversals of the mount point (i.e. "..") will not work if 1058 * the nfsnode gets flushed out of the cache. Ufs does not have 1059 * this problem, because one can identify root inodes by their 1060 * number == ROOTINO (2). 1061 */ 1062 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 1063 if (error) 1064 goto bad; 1065 *vpp = NFSTOV(np); 1066 1067 /* 1068 * Retrieval of mountpoint attributes is delayed until nfs_rot 1069 * or nfs_statfs are first called. This will happen either when 1070 * we first traverse the mount point or if somebody does a df(1). 1071 * 1072 * NFSSTA_GOTFSINFO is used to flag if we have successfully 1073 * retrieved mountpoint attributes. In the case of NFSv3 we 1074 * also flag static fsinfo. 1075 */ 1076 if (*vpp != NULL) 1077 (*vpp)->v_type = VNON; 1078 1079 /* 1080 * Lose the lock but keep the ref. 1081 */ 1082 vn_unlock(*vpp); 1083 TAILQ_INSERT_TAIL(&nfs_mountq, nmp, nm_entry); 1084 1085 #ifdef SMP 1086 switch(ncpus) { 1087 case 0: 1088 case 1: 1089 rxcpu = 0; 1090 txcpu = 0; 1091 break; 1092 case 2: 1093 rxcpu = 0; 1094 txcpu = 1; 1095 break; 1096 default: 1097 rxcpu = 1; 1098 txcpu = 2; 1099 break; 1100 } 1101 #else 1102 rxcpu = 0; 1103 txcpu = 0; 1104 #endif 1105 1106 /* 1107 * Start the reader and writer threads. 1108 */ 1109 lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread, 1110 NULL, 0, rxcpu, "nfsiod_rx"); 1111 lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread, 1112 NULL, 0, txcpu, "nfsiod_tx"); 1113 1114 return (0); 1115 bad: 1116 nfs_disconnect(nmp); 1117 nfs_free_mount(nmp); 1118 return (error); 1119 } 1120 1121 /* 1122 * unmount system call 1123 */ 1124 static int 1125 nfs_unmount(struct mount *mp, int mntflags) 1126 { 1127 struct nfsmount *nmp; 1128 int error, flags = 0; 1129 1130 nmp = VFSTONFS(mp); 1131 if (mntflags & MNT_FORCE) { 1132 flags |= FORCECLOSE; 1133 nmp->nm_flag |= NFSMNT_FORCE; 1134 } 1135 1136 /* 1137 * Goes something like this.. 1138 * - Call vflush() to clear out vnodes for this file system 1139 * - Close the socket 1140 * - Free up the data structures 1141 */ 1142 /* In the forced case, cancel any outstanding requests. */ 1143 if (flags & FORCECLOSE) { 1144 error = nfs_nmcancelreqs(nmp); 1145 if (error) { 1146 kprintf("NFS: %s: Unable to cancel all requests\n", 1147 mp->mnt_stat.f_mntfromname); 1148 /* continue anyway */ 1149 } 1150 } 1151 1152 /* 1153 * Must handshake with nfs_clientd() if it is active. XXX 1154 */ 1155 nmp->nm_state |= NFSSTA_DISMINPROG; 1156 1157 /* 1158 * We hold 1 extra ref on the root vnode; see comment in mountnfs(). 1159 * 1160 * If this doesn't work and we are doing a forced unmount we continue 1161 * anyway. 1162 */ 1163 error = vflush(mp, 1, flags); 1164 if (error) { 1165 nmp->nm_state &= ~NFSSTA_DISMINPROG; 1166 if ((flags & FORCECLOSE) == 0) 1167 return (error); 1168 } 1169 1170 /* 1171 * We are now committed to the unmount. 1172 * For NQNFS, let the server daemon free the nfsmount structure. 1173 */ 1174 if (nmp->nm_flag & NFSMNT_KERB) 1175 nmp->nm_state |= NFSSTA_DISMNT; 1176 nfssvc_iod_stop1(nmp); 1177 nfs_disconnect(nmp); 1178 nfssvc_iod_stop2(nmp); 1179 TAILQ_REMOVE(&nfs_mountq, nmp, nm_entry); 1180 1181 if ((nmp->nm_flag & NFSMNT_KERB) == 0) { 1182 nfs_free_mount(nmp); 1183 } 1184 return (0); 1185 } 1186 1187 void 1188 nfs_free_mount(struct nfsmount *nmp) 1189 { 1190 if (nmp->nm_cred) { 1191 crfree(nmp->nm_cred); 1192 nmp->nm_cred = NULL; 1193 } 1194 if (nmp->nm_nam) { 1195 FREE(nmp->nm_nam, M_SONAME); 1196 nmp->nm_nam = NULL; 1197 } 1198 zfree(nfsmount_zone, nmp); 1199 } 1200 1201 /* 1202 * Return root of a filesystem 1203 */ 1204 static int 1205 nfs_root(struct mount *mp, struct vnode **vpp) 1206 { 1207 struct vnode *vp; 1208 struct nfsmount *nmp; 1209 struct vattr attrs; 1210 struct nfsnode *np; 1211 int error; 1212 1213 nmp = VFSTONFS(mp); 1214 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 1215 if (error) 1216 return (error); 1217 vp = NFSTOV(np); 1218 1219 /* 1220 * Get transfer parameters and root vnode attributes 1221 */ 1222 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { 1223 if (nmp->nm_flag & NFSMNT_NFSV3) { 1224 error = nfs_fsinfo(nmp, vp, curthread); 1225 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype); 1226 } else { 1227 if ((error = VOP_GETATTR(vp, &attrs)) == 0) 1228 nmp->nm_state |= NFSSTA_GOTFSINFO; 1229 1230 } 1231 } else { 1232 /* 1233 * The root vnode is usually cached by the namecache so do not 1234 * try to avoid going over the wire even if we have previous 1235 * information cached. A stale NFS mount can loop 1236 * forever resolving the root vnode if we return no-error when 1237 * there is in fact an error. 1238 */ 1239 np->n_attrstamp = 0; 1240 error = VOP_GETATTR(vp, &attrs); 1241 } 1242 if (vp->v_type == VNON) 1243 nfs_setvtype(vp, VDIR); 1244 vp->v_flag = VROOT; 1245 if (error) 1246 vput(vp); 1247 else 1248 *vpp = vp; 1249 return (error); 1250 } 1251 1252 struct scaninfo { 1253 int rescan; 1254 int waitfor; 1255 int allerror; 1256 }; 1257 1258 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data); 1259 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data); 1260 1261 /* 1262 * Flush out the buffer cache 1263 */ 1264 /* ARGSUSED */ 1265 static int 1266 nfs_sync(struct mount *mp, int waitfor) 1267 { 1268 struct scaninfo scaninfo; 1269 int error; 1270 1271 scaninfo.rescan = 1; 1272 scaninfo.waitfor = waitfor; 1273 scaninfo.allerror = 0; 1274 1275 /* 1276 * Force stale buffer cache information to be flushed. 1277 */ 1278 error = 0; 1279 while (error == 0 && scaninfo.rescan) { 1280 scaninfo.rescan = 0; 1281 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1, 1282 nfs_sync_scan2, &scaninfo); 1283 } 1284 return(error); 1285 } 1286 1287 static int 1288 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data) 1289 { 1290 struct scaninfo *info = data; 1291 1292 if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree)) 1293 return(-1); 1294 if (info->waitfor == MNT_LAZY) 1295 return(-1); 1296 return(0); 1297 } 1298 1299 static int 1300 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data) 1301 { 1302 struct scaninfo *info = data; 1303 int error; 1304 1305 error = VOP_FSYNC(vp, info->waitfor, 0); 1306 if (error) 1307 info->allerror = error; 1308 return(0); 1309 } 1310 1311