nfs_vfsops.c revision 1.151.6.1
1 /* $NetBSD: nfs_vfsops.c,v 1.151.6.1 2006/02/04 14:10:02 simonb Exp $ */ 2 3 /* 4 * Copyright (c) 1989, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Rick Macklem at The University of Guelph. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)nfs_vfsops.c 8.12 (Berkeley) 5/20/95 35 */ 36 37 #include <sys/cdefs.h> 38 __KERNEL_RCSID(0, "$NetBSD: nfs_vfsops.c,v 1.151.6.1 2006/02/04 14:10:02 simonb Exp $"); 39 40 #if defined(_KERNEL_OPT) 41 #include "opt_compat_netbsd.h" 42 #include "opt_nfs.h" 43 #endif 44 45 #include <sys/param.h> 46 #include <sys/ioctl.h> 47 #include <sys/signal.h> 48 #include <sys/proc.h> 49 #include <sys/namei.h> 50 #include <sys/device.h> 51 #include <sys/vnode.h> 52 #include <sys/kernel.h> 53 #include <sys/mount.h> 54 #include <sys/buf.h> 55 #include <sys/mbuf.h> 56 #include <sys/dirent.h> 57 #include <sys/socket.h> 58 #include <sys/socketvar.h> 59 #include <sys/sysctl.h> 60 #include <sys/systm.h> 61 #include <sys/timetc.h> 62 63 #include <net/if.h> 64 #include <net/route.h> 65 #include <netinet/in.h> 66 67 #include <nfs/rpcv2.h> 68 #include <nfs/nfsproto.h> 69 #include <nfs/nfsnode.h> 70 #include <nfs/nfs.h> 71 #include <nfs/nfsmount.h> 72 #include <nfs/xdr_subs.h> 73 #include <nfs/nfsm_subs.h> 74 #include <nfs/nfsdiskless.h> 75 #include <nfs/nqnfs.h> 76 #include <nfs/nfs_var.h> 77 78 extern struct nfsstats nfsstats; 79 extern int nfs_ticks; 80 81 MALLOC_DEFINE(M_NFSMNT, "NFS mount", "NFS mount structure"); 82 83 /* 84 * nfs vfs operations. 85 */ 86 87 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc; 88 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc; 89 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc; 90 91 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = { 92 &nfsv2_vnodeop_opv_desc, 93 &spec_nfsv2nodeop_opv_desc, 94 &fifo_nfsv2nodeop_opv_desc, 95 NULL, 96 }; 97 98 struct vfsops nfs_vfsops = { 99 MOUNT_NFS, 100 nfs_mount, 101 nfs_start, 102 nfs_unmount, 103 nfs_root, 104 nfs_quotactl, 105 nfs_statvfs, 106 nfs_sync, 107 nfs_vget, 108 nfs_fhtovp, 109 nfs_vptofh, 110 nfs_vfs_init, 111 nfs_vfs_reinit, 112 nfs_vfs_done, 113 nfs_mountroot, 114 (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp, 115 vfs_stdextattrctl, 116 nfs_vnodeopv_descs, 117 }; 118 VFS_ATTACH(nfs_vfsops); 119 120 extern u_int32_t nfs_procids[NFS_NPROCS]; 121 extern u_int32_t nfs_prog, nfs_vers; 122 123 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *, 124 struct mount **, struct vnode **, struct lwp *)); 125 126 /* 127 * nfs statvfs call 128 */ 129 int 130 nfs_statvfs(mp, sbp, l) 131 struct mount *mp; 132 struct statvfs *sbp; 133 struct lwp *l; 134 { 135 struct vnode *vp; 136 struct nfs_statfs *sfp; 137 caddr_t cp; 138 u_int32_t *tl; 139 int32_t t1, t2; 140 caddr_t bpos, dpos, cp2; 141 struct nfsmount *nmp = VFSTONFS(mp); 142 int error = 0, retattr; 143 #ifdef NFS_V2_ONLY 144 const int v3 = 0; 145 #else 146 int v3 = (nmp->nm_flag & NFSMNT_NFSV3); 147 #endif 148 struct mbuf *mreq, *mrep = NULL, *md, *mb; 149 struct ucred *cred; 150 u_quad_t tquad; 151 struct nfsnode *np; 152 153 #ifndef nolint 154 sfp = (struct nfs_statfs *)0; 155 #endif 156 vp = nmp->nm_vnode; 157 np = VTONFS(vp); 158 cred = crget(); 159 cred->cr_ngroups = 0; 160 #ifndef NFS_V2_ONLY 161 if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 162 (void)nfs_fsinfo(nmp, vp, cred, l); 163 #endif 164 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 165 nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3)); 166 nfsm_fhtom(np, v3); 167 nfsm_request(np, NFSPROC_FSSTAT, l, cred); 168 if (v3) 169 nfsm_postop_attr(vp, retattr, 0); 170 if (error) { 171 if (mrep != NULL) { 172 if (mrep->m_next != NULL) 173 printf("nfs_vfsops: nfs_statvfs would lose buffers\n"); 174 m_freem(mrep); 175 } 176 goto nfsmout; 177 } 178 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 179 sbp->f_flag = nmp->nm_flag; 180 sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize); 181 if (v3) { 182 sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE; 183 tquad = fxdr_hyper(&sfp->sf_tbytes); 184 sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 185 tquad = fxdr_hyper(&sfp->sf_fbytes); 186 sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 187 tquad = fxdr_hyper(&sfp->sf_abytes); 188 tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 189 sbp->f_bresvd = sbp->f_bfree - tquad; 190 sbp->f_bavail = tquad; 191 #ifdef COMPAT_20 192 /* Handle older NFS servers returning negative values */ 193 if ((quad_t)sbp->f_bavail < 0) 194 sbp->f_bavail = 0; 195 #endif 196 tquad = fxdr_hyper(&sfp->sf_tfiles); 197 sbp->f_files = tquad; 198 tquad = fxdr_hyper(&sfp->sf_ffiles); 199 sbp->f_ffree = tquad; 200 sbp->f_favail = tquad; 201 sbp->f_fresvd = 0; 202 sbp->f_namemax = MAXNAMLEN; 203 } else { 204 sbp->f_bsize = NFS_FABLKSIZE; 205 sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 206 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 207 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 208 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 209 sbp->f_fresvd = 0; 210 sbp->f_files = 0; 211 sbp->f_ffree = 0; 212 sbp->f_favail = 0; 213 sbp->f_fresvd = 0; 214 sbp->f_namemax = MAXNAMLEN; 215 } 216 copy_statvfs_info(sbp, mp); 217 nfsm_reqdone; 218 crfree(cred); 219 return (error); 220 } 221 222 #ifndef NFS_V2_ONLY 223 /* 224 * nfs version 3 fsinfo rpc call 225 */ 226 int 227 nfs_fsinfo(nmp, vp, cred, l) 228 struct nfsmount *nmp; 229 struct vnode *vp; 230 struct ucred *cred; 231 struct lwp *l; 232 { 233 struct nfsv3_fsinfo *fsp; 234 caddr_t cp; 235 int32_t t1, t2; 236 u_int32_t *tl, pref, xmax; 237 caddr_t bpos, dpos, cp2; 238 int error = 0, retattr; 239 struct mbuf *mreq, *mrep, *md, *mb; 240 u_int64_t maxfsize; 241 struct nfsnode *np = VTONFS(vp); 242 243 nfsstats.rpccnt[NFSPROC_FSINFO]++; 244 nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1)); 245 nfsm_fhtom(np, 1); 246 nfsm_request(np, NFSPROC_FSINFO, l, cred); 247 nfsm_postop_attr(vp, retattr, 0); 248 if (!error) { 249 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 250 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 251 if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 && 252 pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 253 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 254 ~(NFS_FABLKSIZE - 1); 255 xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 256 if (xmax < nmp->nm_wsize && xmax > 0) { 257 nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1); 258 if (nmp->nm_wsize == 0) 259 nmp->nm_wsize = xmax; 260 } 261 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 262 if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 && 263 pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 264 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 265 ~(NFS_FABLKSIZE - 1); 266 xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 267 if (xmax < nmp->nm_rsize && xmax > 0) { 268 nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1); 269 if (nmp->nm_rsize == 0) 270 nmp->nm_rsize = xmax; 271 } 272 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 273 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ) 274 nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) & 275 ~(NFS_DIRFRAGSIZ - 1); 276 if (xmax < nmp->nm_readdirsize && xmax > 0) { 277 nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1); 278 if (nmp->nm_readdirsize == 0) 279 nmp->nm_readdirsize = xmax; 280 } 281 /* XXX */ 282 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 283 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 284 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 285 nmp->nm_maxfilesize = maxfsize; 286 nmp->nm_mountp->mnt_fs_bshift = 287 ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 288 nmp->nm_iflag |= NFSMNT_GOTFSINFO; 289 } 290 nfsm_reqdone; 291 return (error); 292 } 293 #endif 294 295 /* 296 * Mount a remote root fs via. NFS. It goes like this: 297 * - Call nfs_boot_init() to fill in the nfs_diskless struct 298 * - build the rootfs mount point and call mountnfs() to do the rest. 299 */ 300 int 301 nfs_mountroot() 302 { 303 #ifdef __HAVE_TIMECOUNTER 304 struct timespec ts; 305 #endif 306 struct nfs_diskless *nd; 307 struct vattr attr; 308 struct mount *mp; 309 struct vnode *vp; 310 struct lwp *l; 311 long n; 312 int error; 313 314 l = curlwp; /* XXX */ 315 316 if (root_device->dv_class != DV_IFNET) 317 return (ENODEV); 318 319 /* 320 * XXX time must be non-zero when we init the interface or else 321 * the arp code will wedge. [Fixed now in if_ether.c] 322 * However, the NFS attribute cache gives false "hits" when the 323 * current time < NFS_ATTRTIMEO(nmp, np) so keep this in for now. 324 */ 325 if (time_second < NFS_MAXATTRTIMO) { 326 #ifdef __HAVE_TIMECOUNTER 327 ts.tv_sec = NFS_MAXATTRTIMO; 328 ts.tv_nsec = 0; 329 tc_setclock(&ts); 330 #else /* !__HAVE_TIMECOUNTER */ 331 time.tv_sec = NFS_MAXATTRTIMO; 332 #endif /* !__HAVE_TIMECOUNTER */ 333 } 334 335 /* 336 * Call nfs_boot_init() to fill in the nfs_diskless struct. 337 * Side effect: Finds and configures a network interface. 338 */ 339 nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK); 340 memset((caddr_t)nd, 0, sizeof(*nd)); 341 error = nfs_boot_init(nd, l); 342 if (error) { 343 free(nd, M_NFSMNT); 344 return (error); 345 } 346 347 /* 348 * Create the root mount point. 349 */ 350 error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l); 351 if (error) 352 goto out; 353 printf("root on %s\n", nd->nd_root.ndm_host); 354 355 /* 356 * Link it into the mount list. 357 */ 358 simple_lock(&mountlist_slock); 359 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 360 simple_unlock(&mountlist_slock); 361 rootvp = vp; 362 mp->mnt_vnodecovered = NULLVP; 363 vfs_unbusy(mp); 364 365 /* Get root attributes (for the time). */ 366 error = VOP_GETATTR(vp, &attr, l->l_proc->p_ucred, l); 367 if (error) 368 panic("nfs_mountroot: getattr for root"); 369 n = attr.va_atime.tv_sec; 370 #ifdef DEBUG 371 printf("root time: 0x%lx\n", n); 372 #endif 373 setrootfstime(n); 374 375 out: 376 if (error) 377 nfs_boot_cleanup(nd, l); 378 free(nd, M_NFSMNT); 379 return (error); 380 } 381 382 /* 383 * Internal version of mount system call for diskless setup. 384 * Separate function because we used to call it twice. 385 * (once for root and once for swap) 386 */ 387 static int 388 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, l) 389 struct nfs_dlmount *ndmntp; 390 const char *mntname; /* mount point name */ 391 struct mount **mpp; 392 struct vnode **vpp; 393 struct lwp *l; 394 { 395 struct mount *mp; 396 struct mbuf *m; 397 int error; 398 399 vfs_rootmountalloc(MOUNT_NFS, mntname, &mp); 400 401 mp->mnt_op = &nfs_vfsops; 402 403 /* 404 * Historical practice expects NFS root file systems to 405 * be initially mounted r/w. 406 */ 407 mp->mnt_flag &= ~MNT_RDONLY; 408 409 /* Get mbuf for server sockaddr. */ 410 m = m_get(M_WAIT, MT_SONAME); 411 if (m == NULL) 412 panic("nfs_mountroot: mget soname for %s", mntname); 413 MCLAIM(m, &nfs_mowner); 414 memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr, 415 (m->m_len = ndmntp->ndm_args.addr->sa_len)); 416 417 error = mountnfs(&ndmntp->ndm_args, mp, m, mntname, 418 ndmntp->ndm_args.hostname, vpp, l); 419 if (error) { 420 mp->mnt_op->vfs_refcount--; 421 vfs_unbusy(mp); 422 printf("nfs_mountroot: mount %s failed: %d\n", 423 mntname, error); 424 free(mp, M_MOUNT); 425 } else 426 *mpp = mp; 427 428 return (error); 429 } 430 431 void 432 nfs_decode_args(nmp, argp, l) 433 struct nfsmount *nmp; 434 struct nfs_args *argp; 435 struct lwp *l; 436 { 437 int s; 438 int adjsock; 439 int maxio; 440 441 s = splsoftnet(); 442 443 /* 444 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 445 * no sense in that context. 446 */ 447 if (argp->sotype == SOCK_STREAM) 448 argp->flags &= ~NFSMNT_NOCONN; 449 450 /* 451 * Cookie translation is not needed for v2, silently ignore it. 452 */ 453 if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) == 454 NFSMNT_XLATECOOKIE) 455 argp->flags &= ~NFSMNT_XLATECOOKIE; 456 457 /* Re-bind if rsrvd port requested and wasn't on one */ 458 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 459 && (argp->flags & NFSMNT_RESVPORT); 460 /* Also re-bind if we're switching to/from a connected UDP socket */ 461 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 462 (argp->flags & NFSMNT_NOCONN)); 463 464 /* Update flags. */ 465 nmp->nm_flag = argp->flags; 466 splx(s); 467 468 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 469 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 470 if (nmp->nm_timeo < NFS_MINTIMEO) 471 nmp->nm_timeo = NFS_MINTIMEO; 472 else if (nmp->nm_timeo > NFS_MAXTIMEO) 473 nmp->nm_timeo = NFS_MAXTIMEO; 474 } 475 476 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 477 nmp->nm_retry = argp->retrans; 478 if (nmp->nm_retry > NFS_MAXREXMIT) 479 nmp->nm_retry = NFS_MAXREXMIT; 480 } 481 482 #ifndef NFS_V2_ONLY 483 if (argp->flags & NFSMNT_NFSV3) { 484 if (argp->sotype == SOCK_DGRAM) 485 maxio = NFS_MAXDGRAMDATA; 486 else 487 maxio = NFS_MAXDATA; 488 } else 489 #endif 490 maxio = NFS_V2MAXDATA; 491 492 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 493 int osize = nmp->nm_wsize; 494 nmp->nm_wsize = argp->wsize; 495 /* Round down to multiple of blocksize */ 496 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 497 if (nmp->nm_wsize <= 0) 498 nmp->nm_wsize = NFS_FABLKSIZE; 499 adjsock |= (nmp->nm_wsize != osize); 500 } 501 if (nmp->nm_wsize > maxio) 502 nmp->nm_wsize = maxio; 503 if (nmp->nm_wsize > MAXBSIZE) 504 nmp->nm_wsize = MAXBSIZE; 505 506 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 507 int osize = nmp->nm_rsize; 508 nmp->nm_rsize = argp->rsize; 509 /* Round down to multiple of blocksize */ 510 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 511 if (nmp->nm_rsize <= 0) 512 nmp->nm_rsize = NFS_FABLKSIZE; 513 adjsock |= (nmp->nm_rsize != osize); 514 } 515 if (nmp->nm_rsize > maxio) 516 nmp->nm_rsize = maxio; 517 if (nmp->nm_rsize > MAXBSIZE) 518 nmp->nm_rsize = MAXBSIZE; 519 520 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 521 nmp->nm_readdirsize = argp->readdirsize; 522 /* Round down to multiple of minimum blocksize */ 523 nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1); 524 if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ) 525 nmp->nm_readdirsize = NFS_DIRFRAGSIZ; 526 /* Bigger than buffer size makes no sense */ 527 if (nmp->nm_readdirsize > NFS_DIRBLKSIZ) 528 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 529 } else if (argp->flags & NFSMNT_RSIZE) 530 nmp->nm_readdirsize = nmp->nm_rsize; 531 532 if (nmp->nm_readdirsize > maxio) 533 nmp->nm_readdirsize = maxio; 534 535 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && 536 argp->maxgrouplist <= NFS_MAXGRPS) 537 nmp->nm_numgrps = argp->maxgrouplist; 538 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && 539 argp->readahead <= NFS_MAXRAHEAD) 540 nmp->nm_readahead = argp->readahead; 541 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 && 542 argp->leaseterm <= NQ_MAXLEASE) 543 nmp->nm_leaseterm = argp->leaseterm; 544 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && 545 argp->deadthresh <= NQ_NEVERDEAD) 546 nmp->nm_deadthresh = argp->deadthresh; 547 548 adjsock |= ((nmp->nm_sotype != argp->sotype) || 549 (nmp->nm_soproto != argp->proto)); 550 nmp->nm_sotype = argp->sotype; 551 nmp->nm_soproto = argp->proto; 552 553 if (nmp->nm_so && adjsock) { 554 nfs_safedisconnect(nmp); 555 if (nmp->nm_sotype == SOCK_DGRAM) 556 while (nfs_connect(nmp, (struct nfsreq *)0, l)) { 557 printf("nfs_args: retrying connect\n"); 558 (void) tsleep((caddr_t)&lbolt, 559 PSOCK, "nfscn3", 0); 560 } 561 } 562 } 563 564 /* 565 * VFS Operations. 566 * 567 * mount system call 568 * It seems a bit dumb to copyinstr() the host and path here and then 569 * memcpy() them in mountnfs(), but I wanted to detect errors before 570 * doing the sockargs() call because sockargs() allocates an mbuf and 571 * an error after that means that I have to release the mbuf. 572 */ 573 /* ARGSUSED */ 574 int 575 nfs_mount(mp, path, data, ndp, l) 576 struct mount *mp; 577 const char *path; 578 void *data; 579 struct nameidata *ndp; 580 struct lwp *l; 581 { 582 int error; 583 struct nfs_args args; 584 struct mbuf *nam; 585 struct nfsmount *nmp = VFSTONFS(mp); 586 struct sockaddr *sa; 587 struct vnode *vp; 588 char *pth, *hst; 589 struct proc *p; 590 size_t len; 591 u_char *nfh; 592 593 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)); 594 if (error) 595 return (error); 596 597 p = l->l_proc; 598 if (mp->mnt_flag & MNT_GETARGS) { 599 600 if (nmp == NULL) 601 return (EIO); 602 if (args.addr != NULL) { 603 sa = mtod(nmp->nm_nam, struct sockaddr *); 604 error = copyout(sa, args.addr, sa->sa_len); 605 if (error) 606 return (error); 607 args.addrlen = sa->sa_len; 608 } else 609 args.addrlen = 0; 610 611 args.version = NFS_ARGSVERSION; 612 args.sotype = nmp->nm_sotype; 613 args.proto = nmp->nm_soproto; 614 args.fh = NULL; 615 args.fhsize = 0; 616 args.flags = nmp->nm_flag; 617 args.wsize = nmp->nm_wsize; 618 args.rsize = nmp->nm_rsize; 619 args.readdirsize = nmp->nm_readdirsize; 620 args.timeo = nmp->nm_timeo; 621 args.retrans = nmp->nm_retry; 622 args.maxgrouplist = nmp->nm_numgrps; 623 args.readahead = nmp->nm_readahead; 624 args.leaseterm = nmp->nm_leaseterm; 625 args.deadthresh = nmp->nm_deadthresh; 626 args.hostname = NULL; 627 return (copyout(&args, data, sizeof(args))); 628 } 629 630 if (args.version != NFS_ARGSVERSION) 631 return (EPROGMISMATCH); 632 #ifdef NFS_V2_ONLY 633 if (args.flags & NFSMNT_NQNFS) 634 return (EPROGUNAVAIL); 635 if (args.flags & NFSMNT_NFSV3) 636 return (EPROGMISMATCH); 637 #endif 638 if (mp->mnt_flag & MNT_UPDATE) { 639 if (nmp == NULL) 640 return (EIO); 641 /* 642 * When doing an update, we can't change from or to 643 * v3 and/or nqnfs, or change cookie translation 644 */ 645 args.flags = (args.flags & 646 ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) | 647 (nmp->nm_flag & 648 (NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)); 649 nfs_decode_args(nmp, &args, l); 650 return (0); 651 } 652 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) 653 return (EINVAL); 654 MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK); 655 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); 656 if (error) 657 return (error); 658 MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK); 659 error = copyinstr(path, pth, MNAMELEN - 1, &len); 660 if (error) 661 goto free_nfh; 662 memset(&pth[len], 0, MNAMELEN - len); 663 MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK); 664 error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len); 665 if (error) 666 goto free_pth; 667 memset(&hst[len], 0, MNAMELEN - len); 668 /* sockargs() call must be after above copyin() calls */ 669 error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME); 670 if (error) 671 goto free_hst; 672 MCLAIM(nam, &nfs_mowner); 673 args.fh = nfh; 674 error = mountnfs(&args, mp, nam, pth, hst, &vp, l); 675 676 free_hst: 677 FREE(hst, M_TEMP); 678 free_pth: 679 FREE(pth, M_TEMP); 680 free_nfh: 681 FREE(nfh, M_TEMP); 682 683 return (error); 684 } 685 686 /* 687 * Common code for mount and mountroot 688 */ 689 int 690 mountnfs(argp, mp, nam, pth, hst, vpp, l) 691 struct nfs_args *argp; 692 struct mount *mp; 693 struct mbuf *nam; 694 const char *pth, *hst; 695 struct vnode **vpp; 696 struct lwp *l; 697 { 698 struct nfsmount *nmp; 699 struct nfsnode *np; 700 int error; 701 struct vattr *attrs; 702 struct ucred *cr; 703 704 /* 705 * If the number of nfs iothreads to use has never 706 * been set, create a reasonable number of them. 707 */ 708 709 if (nfs_niothreads < 0) { 710 nfs_niothreads = NFS_DEFAULT_NIOTHREADS; 711 nfs_getset_niothreads(TRUE); 712 } 713 714 if (mp->mnt_flag & MNT_UPDATE) { 715 nmp = VFSTONFS(mp); 716 /* update paths, file handles, etc, here XXX */ 717 m_freem(nam); 718 return (0); 719 } else { 720 MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount), 721 M_NFSMNT, M_WAITOK); 722 memset((caddr_t)nmp, 0, sizeof (struct nfsmount)); 723 mp->mnt_data = nmp; 724 TAILQ_INIT(&nmp->nm_uidlruhead); 725 TAILQ_INIT(&nmp->nm_bufq); 726 lockinit(&nmp->nm_writeverflock, PRIBIO, "nfswverf", 0, 0); 727 simple_lock_init(&nmp->nm_slock); 728 } 729 vfs_getnewfsid(mp); 730 nmp->nm_mountp = mp; 731 732 #ifndef NFS_V2_ONLY 733 if (argp->flags & NFSMNT_NQNFS) 734 mp->mnt_iflag |= IMNT_DTYPE; 735 #endif 736 737 #ifndef NFS_V2_ONLY 738 if ((argp->flags & NFSMNT_NFSV3) == 0) 739 #endif 740 /* 741 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo 742 * will fill this in. 743 */ 744 nmp->nm_maxfilesize = 0xffffffffLL; 745 746 nmp->nm_timeo = NFS_TIMEO; 747 nmp->nm_retry = NFS_RETRANS; 748 nmp->nm_wsize = NFS_WSIZE; 749 nmp->nm_rsize = NFS_RSIZE; 750 nmp->nm_readdirsize = NFS_READDIRSIZE; 751 nmp->nm_numgrps = NFS_MAXGRPS; 752 nmp->nm_readahead = NFS_DEFRAHEAD; 753 nmp->nm_leaseterm = NQ_DEFLEASE; 754 nmp->nm_deadthresh = NQ_DEADTHRESH; 755 CIRCLEQ_INIT(&nmp->nm_timerhead); 756 nmp->nm_inprog = NULLVP; 757 error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, mp, l); 758 if (error) 759 goto bad; 760 nmp->nm_nam = nam; 761 762 /* Set up the sockets and per-host congestion */ 763 nmp->nm_sotype = argp->sotype; 764 nmp->nm_soproto = argp->proto; 765 766 nfs_decode_args(nmp, argp, l); 767 768 mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 769 mp->mnt_dev_bshift = DEV_BSHIFT; 770 771 /* 772 * For Connection based sockets (TCP,...) defer the connect until 773 * the first request, in case the server is not responding. 774 */ 775 if (nmp->nm_sotype == SOCK_DGRAM && 776 (error = nfs_connect(nmp, (struct nfsreq *)0, l))) 777 goto bad; 778 779 /* 780 * This is silly, but it has to be set so that vinifod() works. 781 * We do not want to do an nfs_statvfs() here since we can get 782 * stuck on a dead server and we are holding a lock on the mount 783 * point. 784 */ 785 mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; 786 error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np); 787 if (error) 788 goto bad; 789 *vpp = NFSTOV(np); 790 MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK); 791 VOP_GETATTR(*vpp, attrs, l->l_proc->p_ucred, l); 792 if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) { 793 cr = crget(); 794 cr->cr_uid = attrs->va_uid; 795 cr->cr_gid = attrs->va_gid; 796 cr->cr_ngroups = 0; 797 nfs_cookieheuristic(*vpp, &nmp->nm_iflag, l, cr); 798 crfree(cr); 799 } 800 FREE(attrs, M_TEMP); 801 802 /* 803 * A reference count is needed on the nfsnode representing the 804 * remote root. If this object is not persistent, then backward 805 * traversals of the mount point (i.e. "..") will not work if 806 * the nfsnode gets flushed out of the cache. Ufs does not have 807 * this problem, because one can identify root inodes by their 808 * number == ROOTINO (2). So, just unlock, but no rele. 809 */ 810 811 nmp->nm_vnode = *vpp; 812 VOP_UNLOCK(*vpp, 0); 813 814 return (0); 815 bad: 816 nfs_disconnect(nmp); 817 free((caddr_t)nmp, M_NFSMNT); 818 m_freem(nam); 819 return (error); 820 } 821 822 /* 823 * unmount system call 824 */ 825 int 826 nfs_unmount(mp, mntflags, l) 827 struct mount *mp; 828 int mntflags; 829 struct lwp *l; 830 { 831 struct nfsmount *nmp; 832 struct vnode *vp; 833 int error, flags = 0; 834 835 if (mntflags & MNT_FORCE) 836 flags |= FORCECLOSE; 837 nmp = VFSTONFS(mp); 838 /* 839 * Goes something like this.. 840 * - Check for activity on the root vnode (other than ourselves). 841 * - Call vflush() to clear out vnodes for this file system, 842 * except for the root vnode. 843 * - Decrement reference on the vnode representing remote root. 844 * - Close the socket 845 * - Free up the data structures 846 */ 847 /* 848 * We need to decrement the ref. count on the nfsnode representing 849 * the remote root. See comment in mountnfs(). The VFS unmount() 850 * has done vput on this vnode, otherwise we would get deadlock! 851 */ 852 vp = nmp->nm_vnode; 853 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 854 if (error != 0) 855 return error; 856 857 if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) { 858 vput(vp); 859 return (EBUSY); 860 } 861 862 /* 863 * Must handshake with nqnfs_clientd() if it is active. 864 */ 865 nmp->nm_iflag |= NFSMNT_DISMINPROG; 866 while (nmp->nm_inprog != NULLVP) 867 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0); 868 error = vflush(mp, vp, flags); 869 if (error) { 870 vput(vp); 871 nmp->nm_iflag &= ~NFSMNT_DISMINPROG; 872 return (error); 873 } 874 875 /* 876 * We are now committed to the unmount; mark the mount structure 877 * as doomed so that any sleepers kicked awake by nfs_disconnect 878 * will go away cleanly. 879 */ 880 nmp->nm_iflag |= NFSMNT_DISMNT; 881 882 /* 883 * There are two reference counts to get rid of here 884 * (see comment in mountnfs()). 885 */ 886 vrele(vp); 887 vput(vp); 888 vgone(vp); 889 nfs_disconnect(nmp); 890 m_freem(nmp->nm_nam); 891 892 /* 893 * For NQNFS, let the server daemon free the nfsmount structure. 894 */ 895 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0) 896 free((caddr_t)nmp, M_NFSMNT); 897 return (0); 898 } 899 900 /* 901 * Return root of a filesystem 902 */ 903 int 904 nfs_root(mp, vpp) 905 struct mount *mp; 906 struct vnode **vpp; 907 { 908 struct vnode *vp; 909 struct nfsmount *nmp; 910 int error; 911 912 nmp = VFSTONFS(mp); 913 vp = nmp->nm_vnode; 914 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 915 if (error != 0) 916 return error; 917 if (vp->v_type == VNON) 918 vp->v_type = VDIR; 919 vp->v_flag = VROOT; 920 *vpp = vp; 921 return (0); 922 } 923 924 extern int syncprt; 925 926 /* 927 * Flush out the buffer cache 928 */ 929 /* ARGSUSED */ 930 int 931 nfs_sync(mp, waitfor, cred, l) 932 struct mount *mp; 933 int waitfor; 934 struct ucred *cred; 935 struct lwp *l; 936 { 937 struct vnode *vp; 938 int error, allerror = 0; 939 940 /* 941 * Force stale buffer cache information to be flushed. 942 */ 943 loop: 944 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 945 /* 946 * If the vnode that we are about to sync is no longer 947 * associated with this mount point, start over. 948 */ 949 if (vp->v_mount != mp) 950 goto loop; 951 if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) || 952 (LIST_EMPTY(&vp->v_dirtyblkhd) && 953 vp->v_uobj.uo_npages == 0)) 954 continue; 955 if (vget(vp, LK_EXCLUSIVE)) 956 goto loop; 957 error = VOP_FSYNC(vp, cred, 958 waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, l); 959 if (error) 960 allerror = error; 961 vput(vp); 962 } 963 return (allerror); 964 } 965 966 /* 967 * NFS flat namespace lookup. 968 * Currently unsupported. 969 */ 970 /* ARGSUSED */ 971 int 972 nfs_vget(mp, ino, vpp) 973 struct mount *mp; 974 ino_t ino; 975 struct vnode **vpp; 976 { 977 978 return (EOPNOTSUPP); 979 } 980 981 /* 982 * Do that sysctl thang... 983 */ 984 static int 985 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS) 986 { 987 int error; 988 989 nfs_getset_niothreads(0); 990 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 991 if (error || newp == NULL) 992 return (error); 993 nfs_getset_niothreads(1); 994 995 return (0); 996 } 997 998 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup") 999 { 1000 1001 sysctl_createv(clog, 0, NULL, NULL, 1002 CTLFLAG_PERMANENT, 1003 CTLTYPE_NODE, "vfs", NULL, 1004 NULL, 0, NULL, 0, 1005 CTL_VFS, CTL_EOL); 1006 sysctl_createv(clog, 0, NULL, NULL, 1007 CTLFLAG_PERMANENT, 1008 CTLTYPE_NODE, "nfs", 1009 SYSCTL_DESCR("NFS vfs options"), 1010 NULL, 0, NULL, 0, 1011 CTL_VFS, 2, CTL_EOL); 1012 /* 1013 * XXX the "2" above could be dynamic, thereby eliminating one 1014 * more instance of the "number to vfs" mapping problem, but 1015 * "2" is the order as taken from sys/mount.h 1016 */ 1017 1018 sysctl_createv(clog, 0, NULL, NULL, 1019 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1020 CTLTYPE_STRUCT, "nfsstats", 1021 SYSCTL_DESCR("NFS operation statistics"), 1022 NULL, 0, &nfsstats, sizeof(nfsstats), 1023 CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL); 1024 sysctl_createv(clog, 0, NULL, NULL, 1025 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1026 CTLTYPE_INT, "iothreads", 1027 SYSCTL_DESCR("Number of NFS client processes desired"), 1028 sysctl_vfs_nfs_iothreads, 0, &nfs_niothreads, 0, 1029 CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL); 1030 } 1031 1032 /* 1033 * At this point, this should never happen 1034 */ 1035 /* ARGSUSED */ 1036 int 1037 nfs_fhtovp(mp, fhp, vpp) 1038 struct mount *mp; 1039 struct fid *fhp; 1040 struct vnode **vpp; 1041 { 1042 1043 return (EINVAL); 1044 } 1045 1046 /* 1047 * Vnode pointer to File handle, should never happen either 1048 */ 1049 /* ARGSUSED */ 1050 int 1051 nfs_vptofh(vp, fhp) 1052 struct vnode *vp; 1053 struct fid *fhp; 1054 { 1055 1056 return (EINVAL); 1057 } 1058 1059 /* 1060 * Vfs start routine, a no-op. 1061 */ 1062 /* ARGSUSED */ 1063 int 1064 nfs_start(mp, flags, l) 1065 struct mount *mp; 1066 int flags; 1067 struct lwp *l; 1068 { 1069 1070 return (0); 1071 } 1072 1073 /* 1074 * Do operations associated with quotas, not supported 1075 */ 1076 /* ARGSUSED */ 1077 int 1078 nfs_quotactl(mp, cmd, uid, arg, l) 1079 struct mount *mp; 1080 int cmd; 1081 uid_t uid; 1082 void *arg; 1083 struct lwp *l; 1084 { 1085 1086 return (EOPNOTSUPP); 1087 } 1088
Indexes created Wed Oct 01 07:09:59 GMT 2025