nfs_vfsops.c revision 1.138
1 /* $NetBSD: nfs_vfsops.c,v 1.138 2004/05/22 22:52:16 jonathan 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 1.138 2004/05/22 22:52:16 jonathan 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 62 #include <net/if.h> 63 #include <net/route.h> 64 #include <netinet/in.h> 65 66 #include <nfs/rpcv2.h> 67 #include <nfs/nfsproto.h> 68 #include <nfs/nfsnode.h> 69 #include <nfs/nfs.h> 70 #include <nfs/nfsmount.h> 71 #include <nfs/xdr_subs.h> 72 #include <nfs/nfsm_subs.h> 73 #include <nfs/nfsdiskless.h> 74 #include <nfs/nqnfs.h> 75 #include <nfs/nfs_var.h> 76 77 extern struct nfsstats nfsstats; 78 extern int nfs_ticks; 79 80 MALLOC_DEFINE(M_NFSMNT, "NFS mount", "NFS mount structure"); 81 82 /* 83 * nfs vfs operations. 84 */ 85 86 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc; 87 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc; 88 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc; 89 90 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = { 91 &nfsv2_vnodeop_opv_desc, 92 &spec_nfsv2nodeop_opv_desc, 93 &fifo_nfsv2nodeop_opv_desc, 94 NULL, 95 }; 96 97 struct vfsops nfs_vfsops = { 98 MOUNT_NFS, 99 nfs_mount, 100 nfs_start, 101 nfs_unmount, 102 nfs_root, 103 nfs_quotactl, 104 nfs_statvfs, 105 nfs_sync, 106 nfs_vget, 107 nfs_fhtovp, 108 nfs_vptofh, 109 nfs_vfs_init, 110 nfs_vfs_reinit, 111 nfs_vfs_done, 112 NULL, 113 nfs_mountroot, 114 nfs_checkexp, 115 nfs_vnodeopv_descs, 116 }; 117 118 extern u_int32_t nfs_procids[NFS_NPROCS]; 119 extern u_int32_t nfs_prog, nfs_vers; 120 121 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *, 122 struct mount **, struct vnode **, struct proc *)); 123 124 /* 125 * nfs statvfs call 126 */ 127 int 128 nfs_statvfs(mp, sbp, p) 129 struct mount *mp; 130 struct statvfs *sbp; 131 struct proc *p; 132 { 133 struct vnode *vp; 134 struct nfs_statfs *sfp; 135 caddr_t cp; 136 u_int32_t *tl; 137 int32_t t1, t2; 138 caddr_t bpos, dpos, cp2; 139 struct nfsmount *nmp = VFSTONFS(mp); 140 int error = 0, retattr; 141 #ifdef NFS_V2_ONLY 142 const int v3 = 0; 143 #else 144 int v3 = (nmp->nm_flag & NFSMNT_NFSV3); 145 #endif 146 struct mbuf *mreq, *mrep = NULL, *md, *mb; 147 struct ucred *cred; 148 u_quad_t tquad; 149 struct nfsnode *np; 150 151 #ifndef nolint 152 sfp = (struct nfs_statfs *)0; 153 #endif 154 vp = nmp->nm_vnode; 155 np = VTONFS(vp); 156 cred = crget(); 157 cred->cr_ngroups = 0; 158 #ifndef NFS_V2_ONLY 159 if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 160 (void)nfs_fsinfo(nmp, vp, cred, p); 161 #endif 162 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 163 nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3)); 164 nfsm_fhtom(np, v3); 165 nfsm_request(np, NFSPROC_FSSTAT, p, cred); 166 if (v3) 167 nfsm_postop_attr(vp, retattr, 0); 168 if (error) { 169 if (mrep != NULL) { 170 if (mrep->m_next != NULL) 171 printf("nfs_vfsops: nfs_statvfs would lose buffers\n"); 172 m_freem(mrep); 173 } 174 goto nfsmout; 175 } 176 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 177 sbp->f_flag = nmp->nm_flag; 178 sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize); 179 if (v3) { 180 sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE; 181 tquad = fxdr_hyper(&sfp->sf_tbytes); 182 sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 183 tquad = fxdr_hyper(&sfp->sf_fbytes); 184 sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 185 tquad = fxdr_hyper(&sfp->sf_abytes); 186 tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 187 sbp->f_bresvd = sbp->f_bfree - tquad; 188 sbp->f_bavail = tquad; 189 #ifdef COMPAT_20 190 /* Handle older NFS servers returning negative values */ 191 if ((quad_t)sbp->f_bavail < 0) 192 sbp->f_bavail = 0; 193 #endif 194 tquad = fxdr_hyper(&sfp->sf_tfiles); 195 sbp->f_files = tquad; 196 tquad = fxdr_hyper(&sfp->sf_ffiles); 197 sbp->f_ffree = tquad; 198 sbp->f_favail = tquad; 199 sbp->f_fresvd = 0; 200 sbp->f_namemax = MAXNAMLEN; 201 } else { 202 sbp->f_bsize = NFS_FABLKSIZE; 203 sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 204 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 205 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 206 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 207 sbp->f_fresvd = 0; 208 sbp->f_files = 0; 209 sbp->f_ffree = 0; 210 sbp->f_favail = 0; 211 sbp->f_fresvd = 0; 212 sbp->f_namemax = MAXNAMLEN; 213 } 214 copy_statvfs_info(sbp, mp); 215 nfsm_reqdone; 216 crfree(cred); 217 return (error); 218 } 219 220 #ifndef NFS_V2_ONLY 221 /* 222 * nfs version 3 fsinfo rpc call 223 */ 224 int 225 nfs_fsinfo(nmp, vp, cred, p) 226 struct nfsmount *nmp; 227 struct vnode *vp; 228 struct ucred *cred; 229 struct proc *p; 230 { 231 struct nfsv3_fsinfo *fsp; 232 caddr_t cp; 233 int32_t t1, t2; 234 u_int32_t *tl, pref, max; 235 caddr_t bpos, dpos, cp2; 236 int error = 0, retattr; 237 struct mbuf *mreq, *mrep, *md, *mb; 238 u_int64_t maxfsize; 239 struct nfsnode *np = VTONFS(vp); 240 241 nfsstats.rpccnt[NFSPROC_FSINFO]++; 242 nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1)); 243 nfsm_fhtom(np, 1); 244 nfsm_request(np, NFSPROC_FSINFO, p, cred); 245 nfsm_postop_attr(vp, retattr, 0); 246 if (!error) { 247 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 248 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 249 if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 && 250 pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 251 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 252 ~(NFS_FABLKSIZE - 1); 253 max = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 254 if (max < nmp->nm_wsize && max > 0) { 255 nmp->nm_wsize = max & ~(NFS_FABLKSIZE - 1); 256 if (nmp->nm_wsize == 0) 257 nmp->nm_wsize = max; 258 } 259 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 260 if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 && 261 pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 262 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 263 ~(NFS_FABLKSIZE - 1); 264 max = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 265 if (max < nmp->nm_rsize && max > 0) { 266 nmp->nm_rsize = max & ~(NFS_FABLKSIZE - 1); 267 if (nmp->nm_rsize == 0) 268 nmp->nm_rsize = max; 269 } 270 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 271 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ) 272 nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) & 273 ~(NFS_DIRFRAGSIZ - 1); 274 if (max < nmp->nm_readdirsize && max > 0) { 275 nmp->nm_readdirsize = max & ~(NFS_DIRFRAGSIZ - 1); 276 if (nmp->nm_readdirsize == 0) 277 nmp->nm_readdirsize = max; 278 } 279 /* XXX */ 280 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 281 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 282 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 283 nmp->nm_maxfilesize = maxfsize; 284 nmp->nm_iflag |= NFSMNT_GOTFSINFO; 285 } 286 nfsm_reqdone; 287 return (error); 288 } 289 #endif 290 291 /* 292 * Mount a remote root fs via. NFS. It goes like this: 293 * - Call nfs_boot_init() to fill in the nfs_diskless struct 294 * - build the rootfs mount point and call mountnfs() to do the rest. 295 */ 296 int 297 nfs_mountroot() 298 { 299 struct nfs_diskless *nd; 300 struct vattr attr; 301 struct mount *mp; 302 struct vnode *vp; 303 struct proc *procp; 304 long n; 305 int error; 306 307 procp = curproc; /* XXX */ 308 309 if (root_device->dv_class != DV_IFNET) 310 return (ENODEV); 311 312 /* 313 * XXX time must be non-zero when we init the interface or else 314 * the arp code will wedge. [Fixed now in if_ether.c] 315 * However, the NFS attribute cache gives false "hits" when 316 * time.tv_sec < NFS_ATTRTIMEO(np) so keep this in for now. 317 */ 318 if (time.tv_sec < NFS_MAXATTRTIMO) 319 time.tv_sec = NFS_MAXATTRTIMO; 320 321 /* 322 * Call nfs_boot_init() to fill in the nfs_diskless struct. 323 * Side effect: Finds and configures a network interface. 324 */ 325 nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK); 326 memset((caddr_t)nd, 0, sizeof(*nd)); 327 error = nfs_boot_init(nd, procp); 328 if (error) { 329 free(nd, M_NFSMNT); 330 return (error); 331 } 332 333 /* 334 * Create the root mount point. 335 */ 336 error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, procp); 337 if (error) 338 goto out; 339 printf("root on %s\n", nd->nd_root.ndm_host); 340 341 /* 342 * Link it into the mount list. 343 */ 344 simple_lock(&mountlist_slock); 345 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 346 simple_unlock(&mountlist_slock); 347 rootvp = vp; 348 mp->mnt_vnodecovered = NULLVP; 349 vfs_unbusy(mp); 350 351 /* Get root attributes (for the time). */ 352 error = VOP_GETATTR(vp, &attr, procp->p_ucred, procp); 353 if (error) 354 panic("nfs_mountroot: getattr for root"); 355 n = attr.va_atime.tv_sec; 356 #ifdef DEBUG 357 printf("root time: 0x%lx\n", n); 358 #endif 359 inittodr(n); 360 361 out: 362 if (error) 363 nfs_boot_cleanup(nd, procp); 364 free(nd, M_NFSMNT); 365 return (error); 366 } 367 368 /* 369 * Internal version of mount system call for diskless setup. 370 * Separate function because we used to call it twice. 371 * (once for root and once for swap) 372 */ 373 static int 374 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, p) 375 struct nfs_dlmount *ndmntp; 376 const char *mntname; /* mount point name */ 377 struct mount **mpp; 378 struct vnode **vpp; 379 struct proc *p; 380 { 381 struct mount *mp; 382 struct mbuf *m; 383 int error; 384 385 vfs_rootmountalloc(MOUNT_NFS, (char *)mntname, &mp); 386 387 mp->mnt_op = &nfs_vfsops; 388 389 /* 390 * Historical practice expects NFS root file systems to 391 * be initially mounted r/w. 392 */ 393 mp->mnt_flag &= ~MNT_RDONLY; 394 395 /* Get mbuf for server sockaddr. */ 396 m = m_get(M_WAIT, MT_SONAME); 397 if (m == NULL) 398 panic("nfs_mountroot: mget soname for %s", mntname); 399 MCLAIM(m, &nfs_mowner); 400 memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr, 401 (m->m_len = ndmntp->ndm_args.addr->sa_len)); 402 403 error = mountnfs(&ndmntp->ndm_args, mp, m, mntname, 404 ndmntp->ndm_args.hostname, vpp, p); 405 if (error) { 406 mp->mnt_op->vfs_refcount--; 407 vfs_unbusy(mp); 408 printf("nfs_mountroot: mount %s failed: %d\n", 409 mntname, error); 410 free(mp, M_MOUNT); 411 } else 412 *mpp = mp; 413 414 return (error); 415 } 416 417 void 418 nfs_decode_args(nmp, argp, p) 419 struct nfsmount *nmp; 420 struct nfs_args *argp; 421 struct proc *p; 422 { 423 int s; 424 int adjsock; 425 int maxio; 426 427 s = splsoftnet(); 428 429 /* 430 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 431 * no sense in that context. 432 */ 433 if (argp->sotype == SOCK_STREAM) 434 argp->flags &= ~NFSMNT_NOCONN; 435 436 /* 437 * Cookie translation is not needed for v2, silently ignore it. 438 */ 439 if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) == 440 NFSMNT_XLATECOOKIE) 441 argp->flags &= ~NFSMNT_XLATECOOKIE; 442 443 /* Re-bind if rsrvd port requested and wasn't on one */ 444 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 445 && (argp->flags & NFSMNT_RESVPORT); 446 /* Also re-bind if we're switching to/from a connected UDP socket */ 447 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 448 (argp->flags & NFSMNT_NOCONN)); 449 450 /* Update flags. */ 451 nmp->nm_flag = argp->flags; 452 splx(s); 453 454 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 455 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 456 if (nmp->nm_timeo < NFS_MINTIMEO) 457 nmp->nm_timeo = NFS_MINTIMEO; 458 else if (nmp->nm_timeo > NFS_MAXTIMEO) 459 nmp->nm_timeo = NFS_MAXTIMEO; 460 } 461 462 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 463 nmp->nm_retry = argp->retrans; 464 if (nmp->nm_retry > NFS_MAXREXMIT) 465 nmp->nm_retry = NFS_MAXREXMIT; 466 } 467 468 #ifndef NFS_V2_ONLY 469 if (argp->flags & NFSMNT_NFSV3) { 470 if (argp->sotype == SOCK_DGRAM) 471 maxio = NFS_MAXDGRAMDATA; 472 else 473 maxio = NFS_MAXDATA; 474 } else 475 #endif 476 maxio = NFS_V2MAXDATA; 477 478 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 479 int osize = nmp->nm_wsize; 480 nmp->nm_wsize = argp->wsize; 481 /* Round down to multiple of blocksize */ 482 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 483 if (nmp->nm_wsize <= 0) 484 nmp->nm_wsize = NFS_FABLKSIZE; 485 adjsock |= (nmp->nm_wsize != osize); 486 } 487 if (nmp->nm_wsize > maxio) 488 nmp->nm_wsize = maxio; 489 if (nmp->nm_wsize > MAXBSIZE) 490 nmp->nm_wsize = MAXBSIZE; 491 492 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 493 int osize = nmp->nm_rsize; 494 nmp->nm_rsize = argp->rsize; 495 /* Round down to multiple of blocksize */ 496 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 497 if (nmp->nm_rsize <= 0) 498 nmp->nm_rsize = NFS_FABLKSIZE; 499 adjsock |= (nmp->nm_rsize != osize); 500 } 501 if (nmp->nm_rsize > maxio) 502 nmp->nm_rsize = maxio; 503 if (nmp->nm_rsize > MAXBSIZE) 504 nmp->nm_rsize = MAXBSIZE; 505 506 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 507 nmp->nm_readdirsize = argp->readdirsize; 508 /* Round down to multiple of minimum blocksize */ 509 nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1); 510 if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ) 511 nmp->nm_readdirsize = NFS_DIRFRAGSIZ; 512 /* Bigger than buffer size makes no sense */ 513 if (nmp->nm_readdirsize > NFS_DIRBLKSIZ) 514 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 515 } else if (argp->flags & NFSMNT_RSIZE) 516 nmp->nm_readdirsize = nmp->nm_rsize; 517 518 if (nmp->nm_readdirsize > maxio) 519 nmp->nm_readdirsize = maxio; 520 521 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && 522 argp->maxgrouplist <= NFS_MAXGRPS) 523 nmp->nm_numgrps = argp->maxgrouplist; 524 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && 525 argp->readahead <= NFS_MAXRAHEAD) 526 nmp->nm_readahead = argp->readahead; 527 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 && 528 argp->leaseterm <= NQ_MAXLEASE) 529 nmp->nm_leaseterm = argp->leaseterm; 530 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && 531 argp->deadthresh <= NQ_NEVERDEAD) 532 nmp->nm_deadthresh = argp->deadthresh; 533 534 adjsock |= ((nmp->nm_sotype != argp->sotype) || 535 (nmp->nm_soproto != argp->proto)); 536 nmp->nm_sotype = argp->sotype; 537 nmp->nm_soproto = argp->proto; 538 539 if (nmp->nm_so && adjsock) { 540 nfs_safedisconnect(nmp); 541 if (nmp->nm_sotype == SOCK_DGRAM) 542 while (nfs_connect(nmp, (struct nfsreq *)0, p)) { 543 printf("nfs_args: retrying connect\n"); 544 (void) tsleep((caddr_t)&lbolt, 545 PSOCK, "nfscn3", 0); 546 } 547 } 548 } 549 550 /* 551 * VFS Operations. 552 * 553 * mount system call 554 * It seems a bit dumb to copyinstr() the host and path here and then 555 * memcpy() them in mountnfs(), but I wanted to detect errors before 556 * doing the sockargs() call because sockargs() allocates an mbuf and 557 * an error after that means that I have to release the mbuf. 558 */ 559 /* ARGSUSED */ 560 int 561 nfs_mount(mp, path, data, ndp, p) 562 struct mount *mp; 563 const char *path; 564 void *data; 565 struct nameidata *ndp; 566 struct proc *p; 567 { 568 int error; 569 struct nfs_args args; 570 struct mbuf *nam; 571 struct nfsmount *nmp = VFSTONFS(mp); 572 struct sockaddr *sa; 573 struct vnode *vp; 574 char *pth, *hst; 575 size_t len; 576 u_char *nfh; 577 578 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)); 579 if (error) 580 return (error); 581 582 if (mp->mnt_flag & MNT_GETARGS) { 583 584 if (nmp == NULL) 585 return (EIO); 586 if (args.addr != NULL) { 587 sa = mtod(nmp->nm_nam, struct sockaddr *); 588 error = copyout(sa, args.addr, sa->sa_len); 589 if (error) 590 return (error); 591 args.addrlen = sa->sa_len; 592 } else 593 args.addrlen = 0; 594 595 args.version = NFS_ARGSVERSION; 596 args.sotype = nmp->nm_sotype; 597 args.proto = nmp->nm_soproto; 598 args.fh = NULL; 599 args.fhsize = 0; 600 args.flags = nmp->nm_flag; 601 args.wsize = nmp->nm_wsize; 602 args.rsize = nmp->nm_rsize; 603 args.readdirsize = nmp->nm_readdirsize; 604 args.timeo = nmp->nm_timeo; 605 args.retrans = nmp->nm_retry; 606 args.maxgrouplist = nmp->nm_numgrps; 607 args.readahead = nmp->nm_readahead; 608 args.leaseterm = nmp->nm_leaseterm; 609 args.deadthresh = nmp->nm_deadthresh; 610 args.hostname = NULL; 611 return (copyout(&args, data, sizeof(args))); 612 } 613 614 if (args.version != NFS_ARGSVERSION) 615 return (EPROGMISMATCH); 616 #ifdef NFS_V2_ONLY 617 if (args.flags & NFSMNT_NQNFS) 618 return (EPROGUNAVAIL); 619 if (args.flags & NFSMNT_NFSV3) 620 return (EPROGMISMATCH); 621 #endif 622 if (mp->mnt_flag & MNT_UPDATE) { 623 if (nmp == NULL) 624 return (EIO); 625 /* 626 * When doing an update, we can't change from or to 627 * v3 and/or nqnfs, or change cookie translation 628 */ 629 args.flags = (args.flags & 630 ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) | 631 (nmp->nm_flag & 632 (NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)); 633 nfs_decode_args(nmp, &args, p); 634 return (0); 635 } 636 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) 637 return (EINVAL); 638 MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK); 639 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); 640 if (error) 641 return (error); 642 MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK); 643 error = copyinstr(path, pth, MNAMELEN - 1, &len); 644 if (error) 645 goto free_nfh; 646 memset(&pth[len], 0, MNAMELEN - len); 647 MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK); 648 error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len); 649 if (error) 650 goto free_pth; 651 memset(&hst[len], 0, MNAMELEN - len); 652 /* sockargs() call must be after above copyin() calls */ 653 error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME); 654 if (error) 655 goto free_hst; 656 MCLAIM(nam, &nfs_mowner); 657 args.fh = nfh; 658 error = mountnfs(&args, mp, nam, pth, hst, &vp, p); 659 660 free_hst: 661 FREE(hst, M_TEMP); 662 free_pth: 663 FREE(pth, M_TEMP); 664 free_nfh: 665 FREE(nfh, M_TEMP); 666 667 return (error); 668 } 669 670 /* 671 * Common code for mount and mountroot 672 */ 673 int 674 mountnfs(argp, mp, nam, pth, hst, vpp, p) 675 struct nfs_args *argp; 676 struct mount *mp; 677 struct mbuf *nam; 678 const char *pth, *hst; 679 struct vnode **vpp; 680 struct proc *p; 681 { 682 struct nfsmount *nmp; 683 struct nfsnode *np; 684 int error; 685 struct vattr *attrs; 686 struct ucred *cr; 687 688 /* 689 * If the number of nfs iothreads to use has never 690 * been set, create a reasonable number of them. 691 */ 692 693 if (nfs_niothreads < 0) { 694 nfs_niothreads = NFS_DEFAULT_NIOTHREADS; 695 nfs_getset_niothreads(TRUE); 696 } 697 698 if (mp->mnt_flag & MNT_UPDATE) { 699 nmp = VFSTONFS(mp); 700 /* update paths, file handles, etc, here XXX */ 701 m_freem(nam); 702 return (0); 703 } else { 704 MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount), 705 M_NFSMNT, M_WAITOK); 706 memset((caddr_t)nmp, 0, sizeof (struct nfsmount)); 707 mp->mnt_data = nmp; 708 TAILQ_INIT(&nmp->nm_uidlruhead); 709 TAILQ_INIT(&nmp->nm_bufq); 710 lockinit(&nmp->nm_writeverflock, PRIBIO, "nfswverf", 0, 0); 711 simple_lock_init(&nmp->nm_slock); 712 } 713 vfs_getnewfsid(mp); 714 nmp->nm_mountp = mp; 715 716 #ifndef NFS_V2_ONLY 717 if (argp->flags & NFSMNT_NQNFS) 718 /* 719 * We have to set mnt_maxsymlink to a non-zero value so 720 * that COMPAT_43 routines will know that we are setting 721 * the d_type field in directories (and can zero it for 722 * unsuspecting binaries). 723 */ 724 mp->mnt_maxsymlinklen = 1; 725 #endif 726 727 #ifndef NFS_V2_ONLY 728 if ((argp->flags & NFSMNT_NFSV3) == 0) 729 #endif 730 /* 731 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo 732 * will fill this in. 733 */ 734 nmp->nm_maxfilesize = 0xffffffffLL; 735 736 nmp->nm_timeo = NFS_TIMEO; 737 nmp->nm_retry = NFS_RETRANS; 738 nmp->nm_wsize = NFS_WSIZE; 739 nmp->nm_rsize = NFS_RSIZE; 740 nmp->nm_readdirsize = NFS_READDIRSIZE; 741 nmp->nm_numgrps = NFS_MAXGRPS; 742 nmp->nm_readahead = NFS_DEFRAHEAD; 743 nmp->nm_leaseterm = NQ_DEFLEASE; 744 nmp->nm_deadthresh = NQ_DEADTHRESH; 745 CIRCLEQ_INIT(&nmp->nm_timerhead); 746 nmp->nm_inprog = NULLVP; 747 error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, mp, p); 748 if (error) 749 goto bad; 750 nmp->nm_nam = nam; 751 752 /* Set up the sockets and per-host congestion */ 753 nmp->nm_sotype = argp->sotype; 754 nmp->nm_soproto = argp->proto; 755 756 nfs_decode_args(nmp, argp, p); 757 758 mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 759 mp->mnt_dev_bshift = DEV_BSHIFT; 760 761 /* 762 * For Connection based sockets (TCP,...) defer the connect until 763 * the first request, in case the server is not responding. 764 */ 765 if (nmp->nm_sotype == SOCK_DGRAM && 766 (error = nfs_connect(nmp, (struct nfsreq *)0, p))) 767 goto bad; 768 769 /* 770 * This is silly, but it has to be set so that vinifod() works. 771 * We do not want to do an nfs_statvfs() here since we can get 772 * stuck on a dead server and we are holding a lock on the mount 773 * point. 774 */ 775 mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; 776 error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np); 777 if (error) 778 goto bad; 779 *vpp = NFSTOV(np); 780 MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK); 781 VOP_GETATTR(*vpp, attrs, p->p_ucred, p); 782 if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) { 783 cr = crget(); 784 cr->cr_uid = attrs->va_uid; 785 cr->cr_gid = attrs->va_gid; 786 cr->cr_ngroups = 0; 787 nfs_cookieheuristic(*vpp, &nmp->nm_iflag, p, cr); 788 crfree(cr); 789 } 790 FREE(attrs, M_TEMP); 791 792 /* 793 * A reference count is needed on the nfsnode representing the 794 * remote root. If this object is not persistent, then backward 795 * traversals of the mount point (i.e. "..") will not work if 796 * the nfsnode gets flushed out of the cache. Ufs does not have 797 * this problem, because one can identify root inodes by their 798 * number == ROOTINO (2). So, just unlock, but no rele. 799 */ 800 801 nmp->nm_vnode = *vpp; 802 VOP_UNLOCK(*vpp, 0); 803 804 return (0); 805 bad: 806 nfs_disconnect(nmp); 807 free((caddr_t)nmp, M_NFSMNT); 808 m_freem(nam); 809 return (error); 810 } 811 812 /* 813 * unmount system call 814 */ 815 int 816 nfs_unmount(mp, mntflags, p) 817 struct mount *mp; 818 int mntflags; 819 struct proc *p; 820 { 821 struct nfsmount *nmp; 822 struct vnode *vp; 823 int error, flags = 0; 824 825 if (mntflags & MNT_FORCE) 826 flags |= FORCECLOSE; 827 nmp = VFSTONFS(mp); 828 /* 829 * Goes something like this.. 830 * - Check for activity on the root vnode (other than ourselves). 831 * - Call vflush() to clear out vnodes for this file system, 832 * except for the root vnode. 833 * - Decrement reference on the vnode representing remote root. 834 * - Close the socket 835 * - Free up the data structures 836 */ 837 /* 838 * We need to decrement the ref. count on the nfsnode representing 839 * the remote root. See comment in mountnfs(). The VFS unmount() 840 * has done vput on this vnode, otherwise we would get deadlock! 841 */ 842 vp = nmp->nm_vnode; 843 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 844 if (error != 0) 845 return error; 846 847 if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) { 848 vput(vp); 849 return (EBUSY); 850 } 851 852 /* 853 * Must handshake with nqnfs_clientd() if it is active. 854 */ 855 nmp->nm_iflag |= NFSMNT_DISMINPROG; 856 while (nmp->nm_inprog != NULLVP) 857 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0); 858 error = vflush(mp, vp, flags); 859 if (error) { 860 vput(vp); 861 nmp->nm_iflag &= ~NFSMNT_DISMINPROG; 862 return (error); 863 } 864 865 /* 866 * We are now committed to the unmount; mark the mount structure 867 * as doomed so that any sleepers kicked awake by nfs_disconnect 868 * will go away cleanly. 869 */ 870 nmp->nm_iflag |= NFSMNT_DISMNT; 871 872 /* 873 * There are two reference counts to get rid of here 874 * (see comment in mountnfs()). 875 */ 876 vrele(vp); 877 vput(vp); 878 vgone(vp); 879 nfs_disconnect(nmp); 880 m_freem(nmp->nm_nam); 881 882 /* 883 * For NQNFS, let the server daemon free the nfsmount structure. 884 */ 885 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0) 886 free((caddr_t)nmp, M_NFSMNT); 887 return (0); 888 } 889 890 /* 891 * Return root of a filesystem 892 */ 893 int 894 nfs_root(mp, vpp) 895 struct mount *mp; 896 struct vnode **vpp; 897 { 898 struct vnode *vp; 899 struct nfsmount *nmp; 900 int error; 901 902 nmp = VFSTONFS(mp); 903 vp = nmp->nm_vnode; 904 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 905 if (error != 0) 906 return error; 907 if (vp->v_type == VNON) 908 vp->v_type = VDIR; 909 vp->v_flag = VROOT; 910 *vpp = vp; 911 return (0); 912 } 913 914 extern int syncprt; 915 916 /* 917 * Flush out the buffer cache 918 */ 919 /* ARGSUSED */ 920 int 921 nfs_sync(mp, waitfor, cred, p) 922 struct mount *mp; 923 int waitfor; 924 struct ucred *cred; 925 struct proc *p; 926 { 927 struct vnode *vp; 928 int error, allerror = 0; 929 930 /* 931 * Force stale buffer cache information to be flushed. 932 */ 933 loop: 934 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 935 /* 936 * If the vnode that we are about to sync is no longer 937 * associated with this mount point, start over. 938 */ 939 if (vp->v_mount != mp) 940 goto loop; 941 if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) || 942 (LIST_EMPTY(&vp->v_dirtyblkhd) && 943 vp->v_uobj.uo_npages == 0)) 944 continue; 945 if (vget(vp, LK_EXCLUSIVE)) 946 goto loop; 947 error = VOP_FSYNC(vp, cred, 948 waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, p); 949 if (error) 950 allerror = error; 951 vput(vp); 952 } 953 return (allerror); 954 } 955 956 /* 957 * NFS flat namespace lookup. 958 * Currently unsupported. 959 */ 960 /* ARGSUSED */ 961 int 962 nfs_vget(mp, ino, vpp) 963 struct mount *mp; 964 ino_t ino; 965 struct vnode **vpp; 966 { 967 968 return (EOPNOTSUPP); 969 } 970 971 /* 972 * Do that sysctl thang... 973 */ 974 static int 975 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS) 976 { 977 int error; 978 979 nfs_getset_niothreads(0); 980 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 981 if (error || newp == NULL) 982 return (error); 983 nfs_getset_niothreads(1); 984 985 return (0); 986 } 987 988 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup") 989 { 990 991 sysctl_createv(clog, 0, NULL, NULL, 992 CTLFLAG_PERMANENT, 993 CTLTYPE_NODE, "vfs", NULL, 994 NULL, 0, NULL, 0, 995 CTL_VFS, CTL_EOL); 996 sysctl_createv(clog, 0, NULL, NULL, 997 CTLFLAG_PERMANENT, 998 CTLTYPE_NODE, "nfs", NULL, 999 NULL, 0, NULL, 0, 1000 CTL_VFS, 2, CTL_EOL); 1001 /* 1002 * XXX the "2" above could be dynamic, thereby eliminating one 1003 * more instance of the "number to vfs" mapping problem, but 1004 * "2" is the order as taken from sys/mount.h 1005 */ 1006 1007 sysctl_createv(clog, 0, NULL, NULL, 1008 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1009 CTLTYPE_STRUCT, "nfsstats", NULL, 1010 NULL, 0, &nfsstats, sizeof(nfsstats), 1011 CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL); 1012 sysctl_createv(clog, 0, NULL, NULL, 1013 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1014 CTLTYPE_INT, "iothreads", NULL, 1015 sysctl_vfs_nfs_iothreads, 0, &nfs_niothreads, 0, 1016 CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL); 1017 } 1018 1019 /* 1020 * At this point, this should never happen 1021 */ 1022 /* ARGSUSED */ 1023 int 1024 nfs_fhtovp(mp, fhp, vpp) 1025 struct mount *mp; 1026 struct fid *fhp; 1027 struct vnode **vpp; 1028 { 1029 1030 return (EINVAL); 1031 } 1032 1033 /* ARGSUSED */ 1034 int 1035 nfs_checkexp(mp, nam, exflagsp, credanonp) 1036 struct mount *mp; 1037 struct mbuf *nam; 1038 int *exflagsp; 1039 struct ucred **credanonp; 1040 { 1041 1042 return (EINVAL); 1043 } 1044 1045 /* 1046 * Vnode pointer to File handle, should never happen either 1047 */ 1048 /* ARGSUSED */ 1049 int 1050 nfs_vptofh(vp, fhp) 1051 struct vnode *vp; 1052 struct fid *fhp; 1053 { 1054 1055 return (EINVAL); 1056 } 1057 1058 /* 1059 * Vfs start routine, a no-op. 1060 */ 1061 /* ARGSUSED */ 1062 int 1063 nfs_start(mp, flags, p) 1064 struct mount *mp; 1065 int flags; 1066 struct proc *p; 1067 { 1068 1069 return (0); 1070 } 1071 1072 /* 1073 * Do operations associated with quotas, not supported 1074 */ 1075 /* ARGSUSED */ 1076 int 1077 nfs_quotactl(mp, cmd, uid, arg, p) 1078 struct mount *mp; 1079 int cmd; 1080 uid_t uid; 1081 void *arg; 1082 struct proc *p; 1083 { 1084 1085 return (EOPNOTSUPP); 1086 } 1087
Indexes created Wed Sep 24 15:09:51 GMT 2025