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 /* 258 * Copy the NFS handle passed from the diskless code. 259 * 260 * XXX CURRENTLY DISABLED - bootp passes us a NFSv2 handle which 261 * will fail utterly with HAMMER due to limitations with NFSv2 262 * directory cookies. 263 */ 264 bcopy(nfs_diskless.swap_fh, nfsv3_diskless.swap_fh, NFSX_V2FH); 265 nfsv3_diskless.swap_fhsize = NFSX_V2FH; 266 for (i = NFSX_V2FH - 1; i >= 0; --i) { 267 if (nfs_diskless.swap_fh[i]) 268 break; 269 } 270 if (i < 0) 271 nfsv3_diskless.swap_fhsize = 0; 272 nfsv3_diskless.swap_fhsize = 0; /* FORCE DISABLE */ 273 274 bcopy(&nfs_diskless.swap_saddr,&nfsv3_diskless.swap_saddr, 275 sizeof(struct sockaddr_in)); 276 bcopy(nfs_diskless.swap_hostnam,nfsv3_diskless.swap_hostnam, MNAMELEN); 277 nfsv3_diskless.swap_nblks = nfs_diskless.swap_nblks; 278 bcopy(&nfs_diskless.swap_ucred, &nfsv3_diskless.swap_ucred, 279 sizeof(struct ucred)); 280 nfs_convert_oargs(&nfsv3_diskless.root_args, &nfs_diskless.root_args); 281 282 /* 283 * Copy the NFS handle passed from the diskless code. 284 * 285 * XXX CURRENTLY DISABLED - bootp passes us a NFSv2 handle which 286 * will fail utterly with HAMMER due to limitations with NFSv2 287 * directory cookies. 288 */ 289 bcopy(nfs_diskless.root_fh, nfsv3_diskless.root_fh, NFSX_V2FH); 290 nfsv3_diskless.root_fhsize = NFSX_V2FH; 291 for (i = NFSX_V2FH - 1; i >= 0; --i) { 292 if (nfs_diskless.root_fh[i]) 293 break; 294 } 295 if (i < 0) 296 nfsv3_diskless.root_fhsize = 0; 297 nfsv3_diskless.root_fhsize = 0; /* FORCE DISABLE */ 298 299 bcopy(&nfs_diskless.root_saddr,&nfsv3_diskless.root_saddr, 300 sizeof(struct sockaddr_in)); 301 bcopy(nfs_diskless.root_hostnam,nfsv3_diskless.root_hostnam, MNAMELEN); 302 nfsv3_diskless.root_time = nfs_diskless.root_time; 303 bcopy(nfs_diskless.my_hostnam,nfsv3_diskless.my_hostnam, 304 MAXHOSTNAMELEN); 305 nfs_diskless_valid = 3; 306 } 307 308 /* 309 * nfs statfs call 310 */ 311 int 312 nfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred) 313 { 314 struct vnode *vp; 315 struct nfs_statfs *sfp; 316 struct nfsmount *nmp = VFSTONFS(mp); 317 thread_t td = curthread; 318 int error = 0, retattr; 319 struct nfsnode *np; 320 u_quad_t tquad; 321 struct nfsm_info info; 322 323 info.mrep = NULL; 324 info.v3 = (nmp->nm_flag & NFSMNT_NFSV3); 325 326 lwkt_gettoken(&nmp->nm_token); 327 328 #ifndef nolint 329 sfp = NULL; 330 #endif 331 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 332 if (error) { 333 lwkt_reltoken(&nmp->nm_token); 334 return (error); 335 } 336 vp = NFSTOV(np); 337 /* ignore the passed cred */ 338 cred = crget(); 339 cred->cr_ngroups = 1; 340 if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 341 (void)nfs_fsinfo(nmp, vp, td); 342 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 343 nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3)); 344 ERROROUT(nfsm_fhtom(&info, vp)); 345 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error)); 346 if (info.v3) { 347 ERROROUT(nfsm_postop_attr(&info, vp, &retattr, 348 NFS_LATTR_NOSHRINK)); 349 } 350 if (error) { 351 if (info.mrep != NULL) 352 m_freem(info.mrep); 353 goto nfsmout; 354 } 355 NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3))); 356 sbp->f_flags = nmp->nm_flag; 357 358 if (info.v3) { 359 sbp->f_bsize = NFS_FABLKSIZE; 360 tquad = fxdr_hyper(&sfp->sf_tbytes); 361 sbp->f_blocks = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 362 tquad = fxdr_hyper(&sfp->sf_fbytes); 363 sbp->f_bfree = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 364 tquad = fxdr_hyper(&sfp->sf_abytes); 365 sbp->f_bavail = (long)(tquad / ((u_quad_t)NFS_FABLKSIZE)); 366 sbp->f_files = (fxdr_unsigned(int32_t, 367 sfp->sf_tfiles.nfsuquad[1]) & 0x7fffffff); 368 sbp->f_ffree = (fxdr_unsigned(int32_t, 369 sfp->sf_ffiles.nfsuquad[1]) & 0x7fffffff); 370 } else { 371 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 372 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 373 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 374 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 375 sbp->f_files = 0; 376 sbp->f_ffree = 0; 377 } 378 379 /* 380 * Some values are pre-set in mnt_stat. Note in particular f_iosize 381 * cannot be changed once the filesystem is mounted as it is used 382 * as the basis for BIOs. 383 */ 384 if (sbp != &mp->mnt_stat) { 385 sbp->f_type = mp->mnt_vfc->vfc_typenum; 386 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); 387 sbp->f_iosize = mp->mnt_stat.f_iosize; 388 } 389 m_freem(info.mrep); 390 info.mrep = NULL; 391 nfsmout: 392 vput(vp); 393 crfree(cred); 394 lwkt_reltoken(&nmp->nm_token); 395 return (error); 396 } 397 398 static int 399 nfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred) 400 { 401 struct vnode *vp; 402 struct nfs_statfs *sfp; 403 struct nfsmount *nmp = VFSTONFS(mp); 404 thread_t td = curthread; 405 int error = 0, retattr; 406 struct nfsnode *np; 407 struct nfsm_info info; 408 409 info.mrep = NULL; 410 info.v3 = (nmp->nm_flag & NFSMNT_NFSV3); 411 lwkt_gettoken(&nmp->nm_token); 412 413 #ifndef nolint 414 sfp = NULL; 415 #endif 416 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 417 if (error) { 418 lwkt_reltoken(&nmp->nm_token); 419 return (error); 420 } 421 vp = NFSTOV(np); 422 /* ignore the passed cred */ 423 cred = crget(); 424 cred->cr_ngroups = 1; 425 if (info.v3 && (nmp->nm_state & NFSSTA_GOTFSINFO) == 0) 426 (void)nfs_fsinfo(nmp, vp, td); 427 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 428 nfsm_reqhead(&info, vp, NFSPROC_FSSTAT, NFSX_FH(info.v3)); 429 ERROROUT(nfsm_fhtom(&info, vp)); 430 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSSTAT, td, cred, &error)); 431 if (info.v3) { 432 ERROROUT(nfsm_postop_attr(&info, vp, &retattr, 433 NFS_LATTR_NOSHRINK)); 434 } 435 if (error) { 436 if (info.mrep != NULL) 437 m_freem(info.mrep); 438 goto nfsmout; 439 } 440 NULLOUT(sfp = nfsm_dissect(&info, NFSX_STATFS(info.v3))); 441 sbp->f_flag = nmp->nm_flag; 442 sbp->f_owner = nmp->nm_cred->cr_ruid; 443 444 if (info.v3) { 445 sbp->f_bsize = NFS_FABLKSIZE; 446 sbp->f_frsize = NFS_FABLKSIZE; 447 sbp->f_blocks = (fxdr_hyper(&sfp->sf_tbytes) / 448 ((u_quad_t)NFS_FABLKSIZE)); 449 sbp->f_bfree = (fxdr_hyper(&sfp->sf_fbytes) / 450 ((u_quad_t)NFS_FABLKSIZE)); 451 sbp->f_bavail = (fxdr_hyper(&sfp->sf_abytes) / 452 ((u_quad_t)NFS_FABLKSIZE)); 453 sbp->f_files = fxdr_hyper(&sfp->sf_tfiles); 454 sbp->f_ffree = fxdr_hyper(&sfp->sf_ffiles); 455 sbp->f_favail = fxdr_hyper(&sfp->sf_afiles); 456 } else { 457 sbp->f_bsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 458 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 459 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 460 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 461 sbp->f_files = 0; 462 sbp->f_ffree = 0; 463 sbp->f_favail = 0; 464 } 465 sbp->f_syncreads = 0; 466 sbp->f_syncwrites = 0; 467 sbp->f_asyncreads = 0; 468 sbp->f_asyncwrites = 0; 469 sbp->f_type = mp->mnt_vfc->vfc_typenum; 470 471 m_freem(info.mrep); 472 info.mrep = NULL; 473 nfsmout: 474 vput(vp); 475 crfree(cred); 476 lwkt_reltoken(&nmp->nm_token); 477 return (error); 478 } 479 480 /* 481 * nfs version 3 fsinfo rpc call 482 */ 483 int 484 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, struct thread *td) 485 { 486 struct nfsv3_fsinfo *fsp; 487 u_int32_t pref, max; 488 int error = 0, retattr; 489 u_int64_t maxfsize; 490 struct nfsm_info info; 491 492 info.v3 = 1; 493 nfsstats.rpccnt[NFSPROC_FSINFO]++; 494 nfsm_reqhead(&info, vp, NFSPROC_FSINFO, NFSX_FH(1)); 495 ERROROUT(nfsm_fhtom(&info, vp)); 496 NEGKEEPOUT(nfsm_request(&info, vp, NFSPROC_FSINFO, td, 497 nfs_vpcred(vp, ND_READ), &error)); 498 ERROROUT(nfsm_postop_attr(&info, vp, &retattr, NFS_LATTR_NOSHRINK)); 499 if (error == 0) { 500 NULLOUT(fsp = nfsm_dissect(&info, NFSX_V3FSINFO)); 501 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 502 if (pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 503 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 504 ~(NFS_FABLKSIZE - 1); 505 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 506 if (max < nmp->nm_wsize && max > 0) { 507 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1); 508 if (nmp->nm_wsize == 0) 509 nmp->nm_wsize = max; 510 } 511 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 512 if (pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 513 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 514 ~(NFS_FABLKSIZE - 1); 515 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 516 if (max < nmp->nm_rsize && max > 0) { 517 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1); 518 if (nmp->nm_rsize == 0) 519 nmp->nm_rsize = max; 520 } 521 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 522 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRBLKSIZ) 523 nmp->nm_readdirsize = (pref + NFS_DIRBLKSIZ - 1) & 524 ~(NFS_DIRBLKSIZ - 1); 525 if (max < nmp->nm_readdirsize && max > 0) { 526 nmp->nm_readdirsize = max & ~(NFS_DIRBLKSIZ - 1); 527 if (nmp->nm_readdirsize == 0) 528 nmp->nm_readdirsize = max; 529 } 530 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 531 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 532 nmp->nm_maxfilesize = maxfsize; 533 nmp->nm_state |= NFSSTA_GOTFSINFO; 534 535 /* 536 * Use the smaller of rsize/wsize for the biosize. 537 */ 538 if (nmp->nm_rsize < nmp->nm_wsize) 539 nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_rsize; 540 else 541 nmp->nm_mountp->mnt_stat.f_iosize = nmp->nm_wsize; 542 } 543 m_freem(info.mrep); 544 info.mrep = NULL; 545 nfsmout: 546 return (error); 547 } 548 549 /* 550 * Mount a remote root fs via. nfs. This depends on the info in the 551 * nfs_diskless structure that has been filled in properly by some primary 552 * bootstrap. 553 * It goes something like this: 554 * - do enough of "ifconfig" by calling ifioctl() so that the system 555 * can talk to the server 556 * - If nfs_diskless.mygateway is filled in, use that address as 557 * a default gateway. 558 * - build the rootfs mount point and call mountnfs() to do the rest. 559 */ 560 int 561 nfs_mountroot(struct mount *mp) 562 { 563 struct mount *swap_mp; 564 struct nfsv3_diskless *nd = &nfsv3_diskless; 565 struct socket *so; 566 struct vnode *vp; 567 struct thread *td = curthread; /* XXX */ 568 int error, i; 569 u_long l; 570 char buf[128]; 571 572 #if defined(BOOTP_NFSROOT) && defined(BOOTP) 573 bootpc_init(); /* use bootp to get nfs_diskless filled in */ 574 #endif 575 576 /* 577 * XXX time must be non-zero when we init the interface or else 578 * the arp code will wedge... 579 */ 580 while (mycpu->gd_time_seconds == 0) 581 tsleep(mycpu, 0, "arpkludge", 10); 582 583 /* 584 * The boot code may have passed us a diskless structure. 585 */ 586 kprintf("DISKLESS %d\n", nfs_diskless_valid); 587 if (nfs_diskless_valid == 1) 588 nfs_convert_diskless(); 589 590 /* 591 * NFSv3 is required. 592 */ 593 nd->root_args.flags |= NFSMNT_NFSV3 | NFSMNT_RDIRPLUS; 594 nd->swap_args.flags |= NFSMNT_NFSV3; 595 596 #define SINP(sockaddr) ((struct sockaddr_in *)(sockaddr)) 597 kprintf("nfs_mountroot: interface %s ip %s", 598 nd->myif.ifra_name, 599 inet_ntoa(SINP(&nd->myif.ifra_addr)->sin_addr)); 600 kprintf(" bcast %s", 601 inet_ntoa(SINP(&nd->myif.ifra_broadaddr)->sin_addr)); 602 kprintf(" mask %s\n", 603 inet_ntoa(SINP(&nd->myif.ifra_mask)->sin_addr)); 604 #undef SINP 605 606 /* 607 * XXX splnet, so networks will receive... 608 */ 609 crit_enter(); 610 611 /* 612 * BOOTP does not necessarily have to be compiled into the kernel 613 * for an NFS root to work. If we inherited the network 614 * configuration for PXEBOOT then pxe_setup_nfsdiskless() has figured 615 * out our interface for us and all we need to do is ifconfig the 616 * interface. We only do this if the interface has not already been 617 * ifconfig'd by e.g. BOOTP. 618 */ 619 error = socreate(nd->myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0, td); 620 if (error) { 621 panic("nfs_mountroot: socreate(%04x): %d", 622 nd->myif.ifra_addr.sa_family, error); 623 } 624 625 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&nd->myif, proc0.p_ucred); 626 if (error) 627 panic("nfs_mountroot: SIOCAIFADDR: %d", error); 628 629 soclose(so, FNONBLOCK); 630 631 /* 632 * If the gateway field is filled in, set it as the default route. 633 */ 634 if (nd->mygateway.sin_len != 0) { 635 struct sockaddr_in mask, sin; 636 637 bzero((caddr_t)&mask, sizeof(mask)); 638 sin = mask; 639 sin.sin_family = AF_INET; 640 sin.sin_len = sizeof(sin); 641 kprintf("nfs_mountroot: gateway %s\n", 642 inet_ntoa(nd->mygateway.sin_addr)); 643 error = rtrequest_global(RTM_ADD, (struct sockaddr *)&sin, 644 (struct sockaddr *)&nd->mygateway, 645 (struct sockaddr *)&mask, 646 RTF_UP | RTF_GATEWAY); 647 if (error) 648 kprintf("nfs_mountroot: unable to set gateway, error %d, continuing anyway\n", error); 649 } 650 651 /* 652 * Create the rootfs mount point. 653 */ 654 nd->root_args.fh = nd->root_fh; 655 nd->root_args.fhsize = nd->root_fhsize; 656 l = ntohl(nd->root_saddr.sin_addr.s_addr); 657 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 658 (l >> 24) & 0xff, (l >> 16) & 0xff, 659 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->root_hostnam); 660 kprintf("NFS_ROOT: %s\n",buf); 661 error = nfs_mountdiskless(buf, "/", MNT_RDONLY, &nd->root_saddr, 662 &nd->root_args, td, &vp, &mp); 663 if (error) { 664 mp->mnt_vfc->vfc_refcount--; 665 crit_exit(); 666 return (error); 667 } 668 669 swap_mp = NULL; 670 if (nd->swap_nblks) { 671 672 /* Convert to DEV_BSIZE instead of Kilobyte */ 673 nd->swap_nblks *= 2; 674 675 /* 676 * Create a fake mount point just for the swap vnode so that the 677 * swap file can be on a different server from the rootfs. 678 */ 679 nd->swap_args.fh = nd->swap_fh; 680 nd->swap_args.fhsize = nd->swap_fhsize; 681 l = ntohl(nd->swap_saddr.sin_addr.s_addr); 682 ksnprintf(buf, sizeof(buf), "%ld.%ld.%ld.%ld:%s", 683 (l >> 24) & 0xff, (l >> 16) & 0xff, 684 (l >> 8) & 0xff, (l >> 0) & 0xff,nd->swap_hostnam); 685 kprintf("NFS SWAP: %s\n",buf); 686 error = nfs_mountdiskless(buf, "/swap", 0, &nd->swap_saddr, 687 &nd->swap_args, td, &vp, &swap_mp); 688 if (error) { 689 crit_exit(); 690 return (error); 691 } 692 vfs_unbusy(swap_mp); 693 694 VTONFS(vp)->n_size = VTONFS(vp)->n_vattr.va_size = 695 nd->swap_nblks * DEV_BSIZE ; 696 697 /* 698 * Since the swap file is not the root dir of a file system, 699 * hack it to a regular file. 700 */ 701 vclrflags(vp, VROOT); 702 vref(vp); 703 nfs_setvtype(vp, VREG); 704 swaponvp(td, vp, nd->swap_nblks); 705 } 706 707 mp->mnt_flag |= MNT_ROOTFS; 708 vfs_unbusy(mp); 709 710 /* 711 * This is not really an nfs issue, but it is much easier to 712 * set hostname here and then let the "/etc/rc.xxx" files 713 * mount the right /var based upon its preset value. 714 */ 715 bcopy(nd->my_hostnam, hostname, MAXHOSTNAMELEN); 716 hostname[MAXHOSTNAMELEN - 1] = '\0'; 717 for (i = 0; i < MAXHOSTNAMELEN; i++) 718 if (hostname[i] == '\0') 719 break; 720 inittodr(ntohl(nd->root_time)); 721 crit_exit(); 722 return (0); 723 } 724 725 /* 726 * Internal version of mount system call for diskless setup. 727 */ 728 static int 729 nfs_mountdiskless(char *path, char *which, int mountflag, 730 struct sockaddr_in *sin, struct nfs_args *args, struct thread *td, 731 struct vnode **vpp, struct mount **mpp) 732 { 733 struct mount *mp; 734 struct sockaddr *nam; 735 int didalloc = 0; 736 int error; 737 738 mp = *mpp; 739 740 if (mp == NULL) { 741 if ((error = vfs_rootmountalloc("nfs", path, &mp)) != 0) { 742 kprintf("nfs_mountroot: NFS not configured"); 743 return (error); 744 } 745 didalloc = 1; 746 } 747 mp->mnt_kern_flag = 0; 748 mp->mnt_flag = mountflag; 749 nam = dup_sockaddr((struct sockaddr *)sin); 750 751 #if defined(BOOTP) || defined(NFS_ROOT) 752 if (args->fhsize == 0) { 753 char *xpath = path; 754 755 kprintf("NFS_ROOT: No FH passed from loader, attempting " 756 "mount rpc..."); 757 while (*xpath && *xpath != ':') 758 ++xpath; 759 if (*xpath) 760 ++xpath; 761 args->fhsize = 0; 762 error = md_mount(sin, xpath, args->fh, &args->fhsize, args, td); 763 if (error) { 764 kprintf("failed error %d.\n", error); 765 goto haderror; 766 } 767 kprintf("success!\n"); 768 } 769 #endif 770 771 if ((error = mountnfs(args, mp, nam, which, path, vpp)) != 0) { 772 #if defined(BOOTP) || defined(NFS_ROOT) 773 haderror: 774 #endif 775 kprintf("nfs_mountroot: mount %s on %s: %d", path, which, error); 776 mp->mnt_vfc->vfc_refcount--; 777 vfs_unbusy(mp); 778 if (didalloc) 779 kfree(mp, M_MOUNT); 780 FREE(nam, M_SONAME); 781 return (error); 782 } 783 *mpp = mp; 784 return (0); 785 } 786 787 static void 788 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp) 789 { 790 int adjsock; 791 int maxio; 792 793 crit_enter(); 794 /* 795 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 796 * no sense in that context. 797 */ 798 if (nmp->nm_sotype == SOCK_STREAM) { 799 nmp->nm_flag &= ~NFSMNT_NOCONN; 800 argp->flags &= ~NFSMNT_NOCONN; 801 } 802 803 /* 804 * readdirplus is NFSv3 only. 805 */ 806 if ((argp->flags & NFSMNT_NFSV3) == 0) { 807 nmp->nm_flag &= ~NFSMNT_RDIRPLUS; 808 argp->flags &= ~NFSMNT_RDIRPLUS; 809 } 810 811 /* 812 * Re-bind if rsrvd port flag has changed 813 */ 814 adjsock = (nmp->nm_flag & NFSMNT_RESVPORT) != 815 (argp->flags & NFSMNT_RESVPORT); 816 817 /* Update flags atomically. Don't change the lock bits. */ 818 nmp->nm_flag = argp->flags | nmp->nm_flag; 819 crit_exit(); 820 821 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 822 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 823 if (nmp->nm_timeo < NFS_MINTIMEO) 824 nmp->nm_timeo = NFS_MINTIMEO; 825 else if (nmp->nm_timeo > NFS_MAXTIMEO) 826 nmp->nm_timeo = NFS_MAXTIMEO; 827 } 828 829 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 830 nmp->nm_retry = argp->retrans; 831 if (nmp->nm_retry > NFS_MAXREXMIT) 832 nmp->nm_retry = NFS_MAXREXMIT; 833 } 834 835 /* 836 * These parameters effect the buffer cache and cannot be changed 837 * once we've successfully mounted. 838 */ 839 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { 840 maxio = nfs_iosize(argp->flags & NFSMNT_NFSV3, nmp->nm_sotype); 841 842 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 843 nmp->nm_wsize = argp->wsize; 844 /* Round down to multiple of blocksize */ 845 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 846 if (nmp->nm_wsize <= 0) 847 nmp->nm_wsize = NFS_FABLKSIZE; 848 } 849 if (nmp->nm_wsize > maxio) 850 nmp->nm_wsize = maxio; 851 if (nmp->nm_wsize > MAXBSIZE) 852 nmp->nm_wsize = MAXBSIZE; 853 854 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 855 nmp->nm_rsize = argp->rsize; 856 /* Round down to multiple of blocksize */ 857 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 858 if (nmp->nm_rsize <= 0) 859 nmp->nm_rsize = NFS_FABLKSIZE; 860 } 861 if (nmp->nm_rsize > maxio) 862 nmp->nm_rsize = maxio; 863 if (nmp->nm_rsize > MAXBSIZE) 864 nmp->nm_rsize = MAXBSIZE; 865 866 if ((argp->flags & NFSMNT_READDIRSIZE) && 867 argp->readdirsize > 0) { 868 nmp->nm_readdirsize = argp->readdirsize; 869 } 870 if (nmp->nm_readdirsize > maxio) 871 nmp->nm_readdirsize = maxio; 872 if (nmp->nm_readdirsize > nmp->nm_rsize) 873 nmp->nm_readdirsize = nmp->nm_rsize; 874 } 875 876 if ((argp->flags & NFSMNT_ACREGMIN) && argp->acregmin >= 0) 877 nmp->nm_acregmin = argp->acregmin; 878 else 879 nmp->nm_acregmin = NFS_MINATTRTIMO; 880 if ((argp->flags & NFSMNT_ACREGMAX) && argp->acregmax >= 0) 881 nmp->nm_acregmax = argp->acregmax; 882 else 883 nmp->nm_acregmax = NFS_MAXATTRTIMO; 884 if ((argp->flags & NFSMNT_ACDIRMIN) && argp->acdirmin >= 0) 885 nmp->nm_acdirmin = argp->acdirmin; 886 else 887 nmp->nm_acdirmin = NFS_MINDIRATTRTIMO; 888 if ((argp->flags & NFSMNT_ACDIRMAX) && argp->acdirmax >= 0) 889 nmp->nm_acdirmax = argp->acdirmax; 890 else 891 nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO; 892 if (nmp->nm_acdirmin > nmp->nm_acdirmax) 893 nmp->nm_acdirmin = nmp->nm_acdirmax; 894 if (nmp->nm_acregmin > nmp->nm_acregmax) 895 nmp->nm_acregmin = nmp->nm_acregmax; 896 897 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0) { 898 if (argp->maxgrouplist <= NFS_MAXGRPS) 899 nmp->nm_numgrps = argp->maxgrouplist; 900 else 901 nmp->nm_numgrps = NFS_MAXGRPS; 902 } 903 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0) { 904 if (argp->readahead <= NFS_MAXRAHEAD) 905 nmp->nm_readahead = argp->readahead; 906 else 907 nmp->nm_readahead = NFS_MAXRAHEAD; 908 } 909 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1) { 910 if (argp->deadthresh <= NFS_NEVERDEAD) 911 nmp->nm_deadthresh = argp->deadthresh; 912 else 913 nmp->nm_deadthresh = NFS_NEVERDEAD; 914 } 915 916 if (nmp->nm_so && adjsock) { 917 nfs_safedisconnect(nmp); 918 if (nmp->nm_sotype == SOCK_DGRAM) 919 while (nfs_connect(nmp, NULL)) { 920 kprintf("nfs_args: retrying connect\n"); 921 (void) tsleep((caddr_t)&lbolt, 0, "nfscon", 0); 922 } 923 } 924 } 925 926 /* 927 * VFS Operations. 928 * 929 * mount system call 930 * It seems a bit dumb to copyinstr() the host and path here and then 931 * bcopy() them in mountnfs(), but I wanted to detect errors before 932 * doing the sockargs() call because sockargs() allocates an mbuf and 933 * an error after that means that I have to release the mbuf. 934 */ 935 /* ARGSUSED */ 936 static int 937 nfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred) 938 { 939 int error; 940 struct nfs_args args; 941 struct sockaddr *nam; 942 struct vnode *vp; 943 char pth[MNAMELEN], hst[MNAMELEN]; 944 size_t len; 945 u_char nfh[NFSX_V3FHMAX]; 946 947 if (path == NULL) { 948 nfs_mountroot(mp); 949 return (0); 950 } 951 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)); 952 if (error) 953 return (error); 954 if (args.version != NFS_ARGSVERSION) { 955 #ifdef COMPAT_PRELITE2 956 /* 957 * If the argument version is unknown, then assume the 958 * caller is a pre-lite2 4.4BSD client and convert its 959 * arguments. 960 */ 961 struct onfs_args oargs; 962 error = copyin(data, (caddr_t)&oargs, sizeof (struct onfs_args)); 963 if (error) 964 return (error); 965 nfs_convert_oargs(&args,&oargs); 966 #else /* !COMPAT_PRELITE2 */ 967 return (EPROGMISMATCH); 968 #endif /* COMPAT_PRELITE2 */ 969 } 970 if (mp->mnt_flag & MNT_UPDATE) { 971 struct nfsmount *nmp = VFSTONFS(mp); 972 973 if (nmp == NULL) 974 return (EIO); 975 /* 976 * When doing an update, we can't change from or to 977 * v3, or change cookie translation, or rsize or wsize. 978 */ 979 args.flags &= ~(NFSMNT_NFSV3 | NFSMNT_RSIZE | NFSMNT_WSIZE); 980 args.flags |= nmp->nm_flag & (NFSMNT_NFSV3); 981 nfs_decode_args(nmp, &args); 982 return (0); 983 } 984 985 /* 986 * Make the nfs_ip_paranoia sysctl serve as the default connection 987 * or no-connection mode for those protocols that support 988 * no-connection mode (the flag will be cleared later for protocols 989 * that do not support no-connection mode). This will allow a client 990 * to receive replies from a different IP then the request was 991 * sent to. Note: default value for nfs_ip_paranoia is 1 (paranoid), 992 * not 0. 993 */ 994 if (nfs_ip_paranoia == 0) 995 args.flags |= NFSMNT_NOCONN; 996 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) 997 return (EINVAL); 998 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); 999 if (error) 1000 return (error); 1001 error = copyinstr(path, pth, MNAMELEN-1, &len); 1002 if (error) 1003 return (error); 1004 bzero(&pth[len], MNAMELEN - len); 1005 error = copyinstr(args.hostname, hst, MNAMELEN-1, &len); 1006 if (error) 1007 return (error); 1008 bzero(&hst[len], MNAMELEN - len); 1009 /* sockargs() call must be after above copyin() calls */ 1010 error = getsockaddr(&nam, (caddr_t)args.addr, args.addrlen); 1011 if (error) 1012 return (error); 1013 args.fh = nfh; 1014 error = mountnfs(&args, mp, nam, pth, hst, &vp); 1015 return (error); 1016 } 1017 1018 /* 1019 * Common code for mount and mountroot 1020 */ 1021 static int 1022 mountnfs(struct nfs_args *argp, struct mount *mp, struct sockaddr *nam, 1023 char *pth, char *hst, struct vnode **vpp) 1024 { 1025 struct nfsmount *nmp; 1026 struct nfsnode *np; 1027 int error; 1028 int rxcpu; 1029 int txcpu; 1030 1031 if (mp->mnt_flag & MNT_UPDATE) { 1032 nmp = VFSTONFS(mp); 1033 /* update paths, file handles, etc, here XXX */ 1034 FREE(nam, M_SONAME); 1035 return (0); 1036 } else { 1037 nmp = zalloc(nfsmount_zone); 1038 bzero((caddr_t)nmp, sizeof (struct nfsmount)); 1039 mtx_init(&nmp->nm_rxlock); 1040 mtx_init(&nmp->nm_txlock); 1041 TAILQ_INIT(&nmp->nm_uidlruhead); 1042 TAILQ_INIT(&nmp->nm_bioq); 1043 TAILQ_INIT(&nmp->nm_reqq); 1044 TAILQ_INIT(&nmp->nm_reqtxq); 1045 TAILQ_INIT(&nmp->nm_reqrxq); 1046 mp->mnt_data = (qaddr_t)nmp; 1047 lwkt_token_init(&nmp->nm_token, 1, "nfs_token"); 1048 } 1049 vfs_getnewfsid(mp); 1050 nmp->nm_mountp = mp; 1051 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; 1052 1053 lwkt_gettoken(&nmp->nm_token); 1054 1055 /* 1056 * V2 can only handle 32 bit filesizes. A 4GB-1 limit may be too 1057 * high, depending on whether we end up with negative offsets in 1058 * the client or server somewhere. 2GB-1 may be safer. 1059 * 1060 * For V3, nfs_fsinfo will adjust this as necessary. Assume maximum 1061 * that we can handle until we find out otherwise. 1062 * XXX Our "safe" limit on the client is what we can store in our 1063 * buffer cache using signed(!) block numbers. 1064 */ 1065 if ((argp->flags & NFSMNT_NFSV3) == 0) 1066 nmp->nm_maxfilesize = 0xffffffffLL; 1067 else 1068 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 1069 1070 nmp->nm_timeo = NFS_TIMEO; 1071 nmp->nm_retry = NFS_RETRANS; 1072 nmp->nm_wsize = nfs_iosize(argp->flags & NFSMNT_NFSV3, argp->sotype); 1073 nmp->nm_rsize = nmp->nm_wsize; 1074 nmp->nm_readdirsize = NFS_READDIRSIZE; 1075 nmp->nm_numgrps = NFS_MAXGRPS; 1076 nmp->nm_readahead = NFS_DEFRAHEAD; 1077 nmp->nm_deadthresh = NFS_DEADTHRESH; 1078 nmp->nm_fhsize = argp->fhsize; 1079 bcopy((caddr_t)argp->fh, (caddr_t)nmp->nm_fh, argp->fhsize); 1080 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); 1081 nmp->nm_nam = nam; 1082 /* Set up the sockets and per-host congestion */ 1083 nmp->nm_sotype = argp->sotype; 1084 nmp->nm_soproto = argp->proto; 1085 nmp->nm_cred = crhold(proc0.p_ucred); 1086 1087 nfs_decode_args(nmp, argp); 1088 1089 /* 1090 * For Connection based sockets (TCP,...) defer the connect until 1091 * the first request, in case the server is not responding. 1092 */ 1093 if (nmp->nm_sotype == SOCK_DGRAM && 1094 (error = nfs_connect(nmp, NULL))) 1095 goto bad; 1096 1097 /* 1098 * This is silly, but it has to be set so that vinifod() works. 1099 * We do not want to do an nfs_statfs() here since we can get 1100 * stuck on a dead server and we are holding a lock on the mount 1101 * point. 1102 */ 1103 mp->mnt_stat.f_iosize = 1104 nfs_iosize(nmp->nm_flag & NFSMNT_NFSV3, nmp->nm_sotype); 1105 1106 /* 1107 * Install vop_ops for our vnops 1108 */ 1109 vfs_add_vnodeops(mp, &nfsv2_vnode_vops, &mp->mnt_vn_norm_ops); 1110 vfs_add_vnodeops(mp, &nfsv2_spec_vops, &mp->mnt_vn_spec_ops); 1111 vfs_add_vnodeops(mp, &nfsv2_fifo_vops, &mp->mnt_vn_fifo_ops); 1112 1113 /* 1114 * A reference count is needed on the nfsnode representing the 1115 * remote root. If this object is not persistent, then backward 1116 * traversals of the mount point (i.e. "..") will not work if 1117 * the nfsnode gets flushed out of the cache. Ufs does not have 1118 * this problem, because one can identify root inodes by their 1119 * number == ROOTINO (2). 1120 */ 1121 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 1122 if (error) 1123 goto bad; 1124 *vpp = NFSTOV(np); 1125 1126 /* 1127 * Retrieval of mountpoint attributes is delayed until nfs_rot 1128 * or nfs_statfs are first called. This will happen either when 1129 * we first traverse the mount point or if somebody does a df(1). 1130 * 1131 * NFSSTA_GOTFSINFO is used to flag if we have successfully 1132 * retrieved mountpoint attributes. In the case of NFSv3 we 1133 * also flag static fsinfo. 1134 */ 1135 if (*vpp != NULL) 1136 (*vpp)->v_type = VNON; 1137 1138 /* 1139 * Lose the lock but keep the ref. 1140 */ 1141 vn_unlock(*vpp); 1142 lwkt_gettoken(&nfs_token); 1143 TAILQ_INSERT_TAIL(&nfs_mountq, nmp, nm_entry); 1144 lwkt_reltoken(&nfs_token); 1145 1146 #ifdef SMP 1147 switch(ncpus) { 1148 case 0: 1149 case 1: 1150 rxcpu = 0; 1151 txcpu = 0; 1152 break; 1153 case 2: 1154 rxcpu = 0; 1155 txcpu = 1; 1156 break; 1157 default: 1158 rxcpu = 1; 1159 txcpu = 2; 1160 break; 1161 } 1162 #else 1163 rxcpu = 0; 1164 txcpu = 0; 1165 #endif 1166 1167 /* 1168 * Start the reader and writer threads. 1169 */ 1170 lwkt_create(nfssvc_iod_reader, nmp, &nmp->nm_rxthread, 1171 NULL, 0, rxcpu, "nfsiod_rx"); 1172 lwkt_create(nfssvc_iod_writer, nmp, &nmp->nm_txthread, 1173 NULL, 0, txcpu, "nfsiod_tx"); 1174 lwkt_reltoken(&nmp->nm_token); 1175 return (0); 1176 bad: 1177 nfs_disconnect(nmp); 1178 lwkt_reltoken(&nmp->nm_token); 1179 nfs_free_mount(nmp); 1180 return (error); 1181 } 1182 1183 /* 1184 * unmount system call 1185 */ 1186 static int 1187 nfs_unmount(struct mount *mp, int mntflags) 1188 { 1189 struct nfsmount *nmp; 1190 int error, flags = 0; 1191 1192 nmp = VFSTONFS(mp); 1193 lwkt_gettoken(&nmp->nm_token); 1194 if (mntflags & MNT_FORCE) { 1195 flags |= FORCECLOSE; 1196 nmp->nm_flag |= NFSMNT_FORCE; 1197 } 1198 1199 /* 1200 * Goes something like this.. 1201 * - Call vflush() to clear out vnodes for this file system 1202 * - Close the socket 1203 * - Free up the data structures 1204 */ 1205 /* In the forced case, cancel any outstanding requests. */ 1206 if (flags & FORCECLOSE) { 1207 error = nfs_nmcancelreqs(nmp); 1208 if (error) { 1209 kprintf("NFS: %s: Unable to cancel all requests\n", 1210 mp->mnt_stat.f_mntfromname); 1211 /* continue anyway */ 1212 } 1213 } 1214 1215 /* 1216 * Must handshake with nfs_clientd() if it is active. XXX 1217 */ 1218 nmp->nm_state |= NFSSTA_DISMINPROG; 1219 1220 /* 1221 * We hold 1 extra ref on the root vnode; see comment in mountnfs(). 1222 * 1223 * If this doesn't work and we are doing a forced unmount we continue 1224 * anyway. 1225 */ 1226 error = vflush(mp, 1, flags); 1227 if (error) { 1228 nmp->nm_state &= ~NFSSTA_DISMINPROG; 1229 if ((flags & FORCECLOSE) == 0) { 1230 lwkt_reltoken(&nmp->nm_token); 1231 return (error); 1232 } 1233 } 1234 1235 /* 1236 * We are now committed to the unmount. 1237 * For NQNFS, let the server daemon free the nfsmount structure. 1238 */ 1239 if (nmp->nm_flag & NFSMNT_KERB) 1240 nmp->nm_state |= NFSSTA_DISMNT; 1241 nfssvc_iod_stop1(nmp); 1242 nfs_disconnect(nmp); 1243 nfssvc_iod_stop2(nmp); 1244 1245 lwkt_gettoken(&nfs_token); 1246 TAILQ_REMOVE(&nfs_mountq, nmp, nm_entry); 1247 lwkt_reltoken(&nfs_token); 1248 1249 lwkt_reltoken(&nmp->nm_token); 1250 1251 if ((nmp->nm_flag & NFSMNT_KERB) == 0) { 1252 nfs_free_mount(nmp); 1253 } 1254 return (0); 1255 } 1256 1257 void 1258 nfs_free_mount(struct nfsmount *nmp) 1259 { 1260 if (nmp->nm_cred) { 1261 crfree(nmp->nm_cred); 1262 nmp->nm_cred = NULL; 1263 } 1264 if (nmp->nm_nam) { 1265 FREE(nmp->nm_nam, M_SONAME); 1266 nmp->nm_nam = NULL; 1267 } 1268 zfree(nfsmount_zone, nmp); 1269 } 1270 1271 /* 1272 * Return root of a filesystem 1273 */ 1274 static int 1275 nfs_root(struct mount *mp, struct vnode **vpp) 1276 { 1277 struct vnode *vp; 1278 struct nfsmount *nmp; 1279 struct vattr attrs; 1280 struct nfsnode *np; 1281 int error; 1282 1283 nmp = VFSTONFS(mp); 1284 lwkt_gettoken(&nmp->nm_token); 1285 error = nfs_nget(mp, (nfsfh_t *)nmp->nm_fh, nmp->nm_fhsize, &np); 1286 if (error) { 1287 lwkt_reltoken(&nmp->nm_token); 1288 return (error); 1289 } 1290 vp = NFSTOV(np); 1291 1292 /* 1293 * Get transfer parameters and root vnode attributes 1294 * 1295 * NOTE: nfs_fsinfo() is expected to override the default 1296 * f_iosize we set. 1297 */ 1298 if ((nmp->nm_state & NFSSTA_GOTFSINFO) == 0) { 1299 if (nmp->nm_flag & NFSMNT_NFSV3) { 1300 mp->mnt_stat.f_iosize = nfs_iosize(1, nmp->nm_sotype); 1301 error = nfs_fsinfo(nmp, vp, curthread); 1302 } else { 1303 if ((error = VOP_GETATTR(vp, &attrs)) == 0) 1304 nmp->nm_state |= NFSSTA_GOTFSINFO; 1305 1306 } 1307 } else { 1308 /* 1309 * The root vnode is usually cached by the namecache so do not 1310 * try to avoid going over the wire even if we have previous 1311 * information cached. A stale NFS mount can loop 1312 * forever resolving the root vnode if we return no-error when 1313 * there is in fact an error. 1314 */ 1315 np->n_attrstamp = 0; 1316 error = VOP_GETATTR(vp, &attrs); 1317 } 1318 if (vp->v_type == VNON) 1319 nfs_setvtype(vp, VDIR); 1320 vsetflags(vp, VROOT); 1321 if (error) 1322 vput(vp); 1323 else 1324 *vpp = vp; 1325 lwkt_reltoken(&nmp->nm_token); 1326 return (error); 1327 } 1328 1329 struct scaninfo { 1330 int rescan; 1331 int waitfor; 1332 int allerror; 1333 }; 1334 1335 static int nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data); 1336 static int nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data); 1337 1338 /* 1339 * Flush out the buffer cache 1340 */ 1341 /* ARGSUSED */ 1342 static int 1343 nfs_sync(struct mount *mp, int waitfor) 1344 { 1345 struct nfsmount *nmp = VFSTONFS(mp); 1346 struct scaninfo scaninfo; 1347 int error; 1348 1349 scaninfo.rescan = 1; 1350 scaninfo.waitfor = waitfor; 1351 scaninfo.allerror = 0; 1352 1353 /* 1354 * Force stale buffer cache information to be flushed. 1355 */ 1356 lwkt_gettoken(&nmp->nm_token); 1357 error = 0; 1358 while (error == 0 && scaninfo.rescan) { 1359 scaninfo.rescan = 0; 1360 error = vmntvnodescan(mp, VMSC_GETVP, nfs_sync_scan1, 1361 nfs_sync_scan2, &scaninfo); 1362 } 1363 lwkt_reltoken(&nmp->nm_token); 1364 return(error); 1365 } 1366 1367 static int 1368 nfs_sync_scan1(struct mount *mp, struct vnode *vp, void *data) 1369 { 1370 struct scaninfo *info = data; 1371 1372 if (vn_islocked(vp) || RB_EMPTY(&vp->v_rbdirty_tree)) 1373 return(-1); 1374 if (info->waitfor == MNT_LAZY) 1375 return(-1); 1376 return(0); 1377 } 1378 1379 static int 1380 nfs_sync_scan2(struct mount *mp, struct vnode *vp, void *data) 1381 { 1382 struct scaninfo *info = data; 1383 int error; 1384 1385 error = VOP_FSYNC(vp, info->waitfor, 0); 1386 if (error) 1387 info->allerror = error; 1388 return(0); 1389 } 1390 1391