nfs_vfsops.c revision 1.152.4.3
1 /* $NetBSD: nfs_vfsops.c,v 1.152.4.3 2006/04/19 05:06:37 elad 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 1.152.4.3 2006/04/19 05:06:37 elad 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 /* 81 * keep a count of the nfs mounts to generate ficticious drive names 82 * for the per drive stats. 83 */ 84 unsigned int nfs_mount_count = 0; 85 86 MALLOC_DEFINE(M_NFSMNT, "NFS mount", "NFS mount structure"); 87 88 /* 89 * nfs vfs operations. 90 */ 91 92 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc; 93 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc; 94 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc; 95 96 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = { 97 &nfsv2_vnodeop_opv_desc, 98 &spec_nfsv2nodeop_opv_desc, 99 &fifo_nfsv2nodeop_opv_desc, 100 NULL, 101 }; 102 103 struct vfsops nfs_vfsops = { 104 MOUNT_NFS, 105 nfs_mount, 106 nfs_start, 107 nfs_unmount, 108 nfs_root, 109 nfs_quotactl, 110 nfs_statvfs, 111 nfs_sync, 112 nfs_vget, 113 nfs_fhtovp, 114 nfs_vptofh, 115 nfs_vfs_init, 116 nfs_vfs_reinit, 117 nfs_vfs_done, 118 nfs_mountroot, 119 (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp, 120 vfs_stdextattrctl, 121 nfs_vnodeopv_descs, 122 }; 123 VFS_ATTACH(nfs_vfsops); 124 125 extern u_int32_t nfs_procids[NFS_NPROCS]; 126 extern u_int32_t nfs_prog, nfs_vers; 127 128 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *, 129 struct mount **, struct vnode **, struct lwp *)); 130 131 /* 132 * nfs statvfs call 133 */ 134 int 135 nfs_statvfs(mp, sbp, l) 136 struct mount *mp; 137 struct statvfs *sbp; 138 struct lwp *l; 139 { 140 struct vnode *vp; 141 struct nfs_statfs *sfp; 142 caddr_t cp; 143 u_int32_t *tl; 144 int32_t t1, t2; 145 caddr_t bpos, dpos, cp2; 146 struct nfsmount *nmp = VFSTONFS(mp); 147 int error = 0, retattr; 148 #ifdef NFS_V2_ONLY 149 const int v3 = 0; 150 #else 151 int v3 = (nmp->nm_flag & NFSMNT_NFSV3); 152 #endif 153 struct mbuf *mreq, *mrep = NULL, *md, *mb; 154 kauth_cred_t cred; 155 u_quad_t tquad; 156 struct nfsnode *np; 157 158 #ifndef nolint 159 sfp = (struct nfs_statfs *)0; 160 #endif 161 vp = nmp->nm_vnode; 162 np = VTONFS(vp); 163 cred = kauth_cred_alloc(); 164 #ifndef NFS_V2_ONLY 165 if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 166 (void)nfs_fsinfo(nmp, vp, cred, l); 167 #endif 168 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 169 nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3)); 170 nfsm_fhtom(np, v3); 171 nfsm_request(np, NFSPROC_FSSTAT, l, cred); 172 if (v3) 173 nfsm_postop_attr(vp, retattr, 0); 174 if (error) { 175 if (mrep != NULL) { 176 if (mrep->m_next != NULL) 177 printf("nfs_vfsops: nfs_statvfs would lose buffers\n"); 178 m_freem(mrep); 179 } 180 goto nfsmout; 181 } 182 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 183 sbp->f_flag = nmp->nm_flag; 184 sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize); 185 if (v3) { 186 sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE; 187 tquad = fxdr_hyper(&sfp->sf_tbytes); 188 sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 189 tquad = fxdr_hyper(&sfp->sf_fbytes); 190 sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 191 tquad = fxdr_hyper(&sfp->sf_abytes); 192 tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 193 sbp->f_bresvd = sbp->f_bfree - tquad; 194 sbp->f_bavail = tquad; 195 #ifdef COMPAT_20 196 /* Handle older NFS servers returning negative values */ 197 if ((quad_t)sbp->f_bavail < 0) 198 sbp->f_bavail = 0; 199 #endif 200 tquad = fxdr_hyper(&sfp->sf_tfiles); 201 sbp->f_files = tquad; 202 tquad = fxdr_hyper(&sfp->sf_ffiles); 203 sbp->f_ffree = tquad; 204 sbp->f_favail = tquad; 205 sbp->f_fresvd = 0; 206 sbp->f_namemax = MAXNAMLEN; 207 } else { 208 sbp->f_bsize = NFS_FABLKSIZE; 209 sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 210 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 211 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 212 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 213 sbp->f_fresvd = 0; 214 sbp->f_files = 0; 215 sbp->f_ffree = 0; 216 sbp->f_favail = 0; 217 sbp->f_fresvd = 0; 218 sbp->f_namemax = MAXNAMLEN; 219 } 220 copy_statvfs_info(sbp, mp); 221 nfsm_reqdone; 222 kauth_cred_free(cred); 223 return (error); 224 } 225 226 #ifndef NFS_V2_ONLY 227 /* 228 * nfs version 3 fsinfo rpc call 229 */ 230 int 231 nfs_fsinfo(nmp, vp, cred, l) 232 struct nfsmount *nmp; 233 struct vnode *vp; 234 kauth_cred_t cred; 235 struct lwp *l; 236 { 237 struct nfsv3_fsinfo *fsp; 238 caddr_t cp; 239 int32_t t1, t2; 240 u_int32_t *tl, pref, xmax; 241 caddr_t bpos, dpos, cp2; 242 int error = 0, retattr; 243 struct mbuf *mreq, *mrep, *md, *mb; 244 u_int64_t maxfsize; 245 struct nfsnode *np = VTONFS(vp); 246 247 nfsstats.rpccnt[NFSPROC_FSINFO]++; 248 nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1)); 249 nfsm_fhtom(np, 1); 250 nfsm_request(np, NFSPROC_FSINFO, l, cred); 251 nfsm_postop_attr(vp, retattr, 0); 252 if (!error) { 253 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 254 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 255 if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 && 256 pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 257 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 258 ~(NFS_FABLKSIZE - 1); 259 xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 260 if (xmax < nmp->nm_wsize && xmax > 0) { 261 nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1); 262 if (nmp->nm_wsize == 0) 263 nmp->nm_wsize = xmax; 264 } 265 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 266 if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 && 267 pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 268 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 269 ~(NFS_FABLKSIZE - 1); 270 xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 271 if (xmax < nmp->nm_rsize && xmax > 0) { 272 nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1); 273 if (nmp->nm_rsize == 0) 274 nmp->nm_rsize = xmax; 275 } 276 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 277 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ) 278 nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) & 279 ~(NFS_DIRFRAGSIZ - 1); 280 if (xmax < nmp->nm_readdirsize && xmax > 0) { 281 nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1); 282 if (nmp->nm_readdirsize == 0) 283 nmp->nm_readdirsize = xmax; 284 } 285 /* XXX */ 286 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 287 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 288 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 289 nmp->nm_maxfilesize = maxfsize; 290 nmp->nm_mountp->mnt_fs_bshift = 291 ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 292 nmp->nm_iflag |= NFSMNT_GOTFSINFO; 293 } 294 nfsm_reqdone; 295 return (error); 296 } 297 #endif 298 299 /* 300 * Mount a remote root fs via. NFS. It goes like this: 301 * - Call nfs_boot_init() to fill in the nfs_diskless struct 302 * - build the rootfs mount point and call mountnfs() to do the rest. 303 */ 304 int 305 nfs_mountroot() 306 { 307 struct nfs_diskless *nd; 308 struct vattr attr; 309 struct mount *mp; 310 struct vnode *vp; 311 struct lwp *l; 312 long n; 313 int error; 314 315 l = curlwp; /* XXX */ 316 317 if (device_class(root_device) != DV_IFNET) 318 return (ENODEV); 319 320 /* 321 * XXX time must be non-zero when we init the interface or else 322 * the arp code will wedge. [Fixed now in if_ether.c] 323 * However, the NFS attribute cache gives false "hits" when 324 * time.tv_sec < NFS_ATTRTIMEO(nmp, np) so keep this in for now. 325 */ 326 if (time.tv_sec < NFS_MAXATTRTIMO) 327 time.tv_sec = NFS_MAXATTRTIMO; 328 329 /* 330 * Call nfs_boot_init() to fill in the nfs_diskless struct. 331 * Side effect: Finds and configures a network interface. 332 */ 333 nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK); 334 memset((caddr_t)nd, 0, sizeof(*nd)); 335 error = nfs_boot_init(nd, l); 336 if (error) { 337 free(nd, M_NFSMNT); 338 return (error); 339 } 340 341 /* 342 * Create the root mount point. 343 */ 344 error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l); 345 if (error) 346 goto out; 347 printf("root on %s\n", nd->nd_root.ndm_host); 348 349 /* 350 * Link it into the mount list. 351 */ 352 simple_lock(&mountlist_slock); 353 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 354 simple_unlock(&mountlist_slock); 355 rootvp = vp; 356 mp->mnt_vnodecovered = NULLVP; 357 vfs_unbusy(mp); 358 359 /* Get root attributes (for the time). */ 360 error = VOP_GETATTR(vp, &attr, l->l_proc->p_cred, l); 361 if (error) 362 panic("nfs_mountroot: getattr for root"); 363 n = attr.va_atime.tv_sec; 364 #ifdef DEBUG 365 printf("root time: 0x%lx\n", n); 366 #endif 367 setrootfstime(n); 368 369 out: 370 if (error) 371 nfs_boot_cleanup(nd, l); 372 free(nd, M_NFSMNT); 373 return (error); 374 } 375 376 /* 377 * Internal version of mount system call for diskless setup. 378 * Separate function because we used to call it twice. 379 * (once for root and once for swap) 380 */ 381 static int 382 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, l) 383 struct nfs_dlmount *ndmntp; 384 const char *mntname; /* mount point name */ 385 struct mount **mpp; 386 struct vnode **vpp; 387 struct lwp *l; 388 { 389 struct mount *mp; 390 struct mbuf *m; 391 int error; 392 393 vfs_rootmountalloc(MOUNT_NFS, mntname, &mp); 394 395 mp->mnt_op = &nfs_vfsops; 396 397 /* 398 * Historical practice expects NFS root file systems to 399 * be initially mounted r/w. 400 */ 401 mp->mnt_flag &= ~MNT_RDONLY; 402 403 /* Get mbuf for server sockaddr. */ 404 m = m_get(M_WAIT, MT_SONAME); 405 if (m == NULL) 406 panic("nfs_mountroot: mget soname for %s", mntname); 407 MCLAIM(m, &nfs_mowner); 408 memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr, 409 (m->m_len = ndmntp->ndm_args.addr->sa_len)); 410 411 error = mountnfs(&ndmntp->ndm_args, mp, m, mntname, 412 ndmntp->ndm_args.hostname, vpp, l); 413 if (error) { 414 mp->mnt_op->vfs_refcount--; 415 vfs_unbusy(mp); 416 printf("nfs_mountroot: mount %s failed: %d\n", 417 mntname, error); 418 free(mp, M_MOUNT); 419 } else 420 *mpp = mp; 421 422 return (error); 423 } 424 425 void 426 nfs_decode_args(nmp, argp, l) 427 struct nfsmount *nmp; 428 struct nfs_args *argp; 429 struct lwp *l; 430 { 431 int s; 432 int adjsock; 433 int maxio; 434 435 s = splsoftnet(); 436 437 /* 438 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 439 * no sense in that context. 440 */ 441 if (argp->sotype == SOCK_STREAM) 442 argp->flags &= ~NFSMNT_NOCONN; 443 444 /* 445 * Cookie translation is not needed for v2, silently ignore it. 446 */ 447 if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) == 448 NFSMNT_XLATECOOKIE) 449 argp->flags &= ~NFSMNT_XLATECOOKIE; 450 451 /* Re-bind if rsrvd port requested and wasn't on one */ 452 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 453 && (argp->flags & NFSMNT_RESVPORT); 454 /* Also re-bind if we're switching to/from a connected UDP socket */ 455 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 456 (argp->flags & NFSMNT_NOCONN)); 457 458 /* Update flags. */ 459 nmp->nm_flag = argp->flags; 460 splx(s); 461 462 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 463 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 464 if (nmp->nm_timeo < NFS_MINTIMEO) 465 nmp->nm_timeo = NFS_MINTIMEO; 466 else if (nmp->nm_timeo > NFS_MAXTIMEO) 467 nmp->nm_timeo = NFS_MAXTIMEO; 468 } 469 470 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 471 nmp->nm_retry = argp->retrans; 472 if (nmp->nm_retry > NFS_MAXREXMIT) 473 nmp->nm_retry = NFS_MAXREXMIT; 474 } 475 476 #ifndef NFS_V2_ONLY 477 if (argp->flags & NFSMNT_NFSV3) { 478 if (argp->sotype == SOCK_DGRAM) 479 maxio = NFS_MAXDGRAMDATA; 480 else 481 maxio = NFS_MAXDATA; 482 } else 483 #endif 484 maxio = NFS_V2MAXDATA; 485 486 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 487 int osize = nmp->nm_wsize; 488 nmp->nm_wsize = argp->wsize; 489 /* Round down to multiple of blocksize */ 490 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 491 if (nmp->nm_wsize <= 0) 492 nmp->nm_wsize = NFS_FABLKSIZE; 493 adjsock |= (nmp->nm_wsize != osize); 494 } 495 if (nmp->nm_wsize > maxio) 496 nmp->nm_wsize = maxio; 497 if (nmp->nm_wsize > MAXBSIZE) 498 nmp->nm_wsize = MAXBSIZE; 499 500 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 501 int osize = nmp->nm_rsize; 502 nmp->nm_rsize = argp->rsize; 503 /* Round down to multiple of blocksize */ 504 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 505 if (nmp->nm_rsize <= 0) 506 nmp->nm_rsize = NFS_FABLKSIZE; 507 adjsock |= (nmp->nm_rsize != osize); 508 } 509 if (nmp->nm_rsize > maxio) 510 nmp->nm_rsize = maxio; 511 if (nmp->nm_rsize > MAXBSIZE) 512 nmp->nm_rsize = MAXBSIZE; 513 514 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 515 nmp->nm_readdirsize = argp->readdirsize; 516 /* Round down to multiple of minimum blocksize */ 517 nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1); 518 if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ) 519 nmp->nm_readdirsize = NFS_DIRFRAGSIZ; 520 /* Bigger than buffer size makes no sense */ 521 if (nmp->nm_readdirsize > NFS_DIRBLKSIZ) 522 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 523 } else if (argp->flags & NFSMNT_RSIZE) 524 nmp->nm_readdirsize = nmp->nm_rsize; 525 526 if (nmp->nm_readdirsize > maxio) 527 nmp->nm_readdirsize = maxio; 528 529 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && 530 argp->maxgrouplist <= NFS_MAXGRPS) 531 nmp->nm_numgrps = argp->maxgrouplist; 532 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && 533 argp->readahead <= NFS_MAXRAHEAD) 534 nmp->nm_readahead = argp->readahead; 535 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 && 536 argp->leaseterm <= NQ_MAXLEASE) 537 nmp->nm_leaseterm = argp->leaseterm; 538 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && 539 argp->deadthresh <= NQ_NEVERDEAD) 540 nmp->nm_deadthresh = argp->deadthresh; 541 542 adjsock |= ((nmp->nm_sotype != argp->sotype) || 543 (nmp->nm_soproto != argp->proto)); 544 nmp->nm_sotype = argp->sotype; 545 nmp->nm_soproto = argp->proto; 546 547 if (nmp->nm_so && adjsock) { 548 nfs_safedisconnect(nmp); 549 if (nmp->nm_sotype == SOCK_DGRAM) 550 while (nfs_connect(nmp, (struct nfsreq *)0, l)) { 551 printf("nfs_args: retrying connect\n"); 552 (void) tsleep((caddr_t)&lbolt, 553 PSOCK, "nfscn3", 0); 554 } 555 } 556 } 557 558 /* 559 * VFS Operations. 560 * 561 * mount system call 562 * It seems a bit dumb to copyinstr() the host and path here and then 563 * memcpy() them in mountnfs(), but I wanted to detect errors before 564 * doing the sockargs() call because sockargs() allocates an mbuf and 565 * an error after that means that I have to release the mbuf. 566 */ 567 /* ARGSUSED */ 568 int 569 nfs_mount(mp, path, data, ndp, l) 570 struct mount *mp; 571 const char *path; 572 void *data; 573 struct nameidata *ndp; 574 struct lwp *l; 575 { 576 int error; 577 struct nfs_args args; 578 struct mbuf *nam; 579 struct nfsmount *nmp = VFSTONFS(mp); 580 struct sockaddr *sa; 581 struct vnode *vp; 582 char *pth, *hst; 583 struct proc *p; 584 size_t len; 585 u_char *nfh; 586 587 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)); 588 if (error) 589 return (error); 590 591 p = l->l_proc; 592 if (mp->mnt_flag & MNT_GETARGS) { 593 594 if (nmp == NULL) 595 return (EIO); 596 if (args.addr != NULL) { 597 sa = mtod(nmp->nm_nam, struct sockaddr *); 598 error = copyout(sa, args.addr, sa->sa_len); 599 if (error) 600 return (error); 601 args.addrlen = sa->sa_len; 602 } else 603 args.addrlen = 0; 604 605 args.version = NFS_ARGSVERSION; 606 args.sotype = nmp->nm_sotype; 607 args.proto = nmp->nm_soproto; 608 args.fh = NULL; 609 args.fhsize = 0; 610 args.flags = nmp->nm_flag; 611 args.wsize = nmp->nm_wsize; 612 args.rsize = nmp->nm_rsize; 613 args.readdirsize = nmp->nm_readdirsize; 614 args.timeo = nmp->nm_timeo; 615 args.retrans = nmp->nm_retry; 616 args.maxgrouplist = nmp->nm_numgrps; 617 args.readahead = nmp->nm_readahead; 618 args.leaseterm = nmp->nm_leaseterm; 619 args.deadthresh = nmp->nm_deadthresh; 620 args.hostname = NULL; 621 return (copyout(&args, data, sizeof(args))); 622 } 623 624 if (args.version != NFS_ARGSVERSION) 625 return (EPROGMISMATCH); 626 #ifdef NFS_V2_ONLY 627 if (args.flags & NFSMNT_NQNFS) 628 return (EPROGUNAVAIL); 629 if (args.flags & NFSMNT_NFSV3) 630 return (EPROGMISMATCH); 631 #endif 632 if (mp->mnt_flag & MNT_UPDATE) { 633 if (nmp == NULL) 634 return (EIO); 635 /* 636 * When doing an update, we can't change from or to 637 * v3 and/or nqnfs, or change cookie translation 638 */ 639 args.flags = (args.flags & 640 ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) | 641 (nmp->nm_flag & 642 (NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)); 643 nfs_decode_args(nmp, &args, l); 644 return (0); 645 } 646 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) 647 return (EINVAL); 648 MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK); 649 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); 650 if (error) 651 return (error); 652 MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK); 653 error = copyinstr(path, pth, MNAMELEN - 1, &len); 654 if (error) 655 goto free_nfh; 656 memset(&pth[len], 0, MNAMELEN - len); 657 MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK); 658 error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len); 659 if (error) 660 goto free_pth; 661 memset(&hst[len], 0, MNAMELEN - len); 662 /* sockargs() call must be after above copyin() calls */ 663 error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME); 664 if (error) 665 goto free_hst; 666 MCLAIM(nam, &nfs_mowner); 667 args.fh = nfh; 668 error = mountnfs(&args, mp, nam, pth, hst, &vp, l); 669 670 free_hst: 671 FREE(hst, M_TEMP); 672 free_pth: 673 FREE(pth, M_TEMP); 674 free_nfh: 675 FREE(nfh, M_TEMP); 676 677 return (error); 678 } 679 680 /* 681 * Common code for mount and mountroot 682 */ 683 int 684 mountnfs(argp, mp, nam, pth, hst, vpp, l) 685 struct nfs_args *argp; 686 struct mount *mp; 687 struct mbuf *nam; 688 const char *pth, *hst; 689 struct vnode **vpp; 690 struct lwp *l; 691 { 692 struct nfsmount *nmp; 693 struct nfsnode *np; 694 int error; 695 struct vattr *attrs; 696 kauth_cred_t cr; 697 698 /* 699 * If the number of nfs iothreads to use has never 700 * been set, create a reasonable number of them. 701 */ 702 703 if (nfs_niothreads < 0) { 704 nfs_niothreads = NFS_DEFAULT_NIOTHREADS; 705 nfs_getset_niothreads(TRUE); 706 } 707 708 if (mp->mnt_flag & MNT_UPDATE) { 709 nmp = VFSTONFS(mp); 710 /* update paths, file handles, etc, here XXX */ 711 m_freem(nam); 712 return (0); 713 } else { 714 MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount), 715 M_NFSMNT, M_WAITOK); 716 memset((caddr_t)nmp, 0, sizeof (struct nfsmount)); 717 mp->mnt_data = nmp; 718 TAILQ_INIT(&nmp->nm_uidlruhead); 719 TAILQ_INIT(&nmp->nm_bufq); 720 lockinit(&nmp->nm_writeverflock, PRIBIO, "nfswverf", 0, 0); 721 simple_lock_init(&nmp->nm_slock); 722 } 723 vfs_getnewfsid(mp); 724 nmp->nm_mountp = mp; 725 726 #ifndef NFS_V2_ONLY 727 if (argp->flags & NFSMNT_NQNFS) 728 mp->mnt_iflag |= IMNT_DTYPE; 729 #endif 730 731 #ifndef NFS_V2_ONLY 732 if ((argp->flags & NFSMNT_NFSV3) == 0) 733 #endif 734 /* 735 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo 736 * will fill this in. 737 */ 738 nmp->nm_maxfilesize = 0xffffffffLL; 739 740 nmp->nm_timeo = NFS_TIMEO; 741 nmp->nm_retry = NFS_RETRANS; 742 nmp->nm_wsize = NFS_WSIZE; 743 nmp->nm_rsize = NFS_RSIZE; 744 nmp->nm_readdirsize = NFS_READDIRSIZE; 745 nmp->nm_numgrps = NFS_MAXGRPS; 746 nmp->nm_readahead = NFS_DEFRAHEAD; 747 nmp->nm_leaseterm = NQ_DEFLEASE; 748 nmp->nm_deadthresh = NQ_DEADTHRESH; 749 CIRCLEQ_INIT(&nmp->nm_timerhead); 750 nmp->nm_inprog = NULLVP; 751 error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, mp, l); 752 if (error) 753 goto bad; 754 nmp->nm_nam = nam; 755 756 /* Set up the sockets and per-host congestion */ 757 nmp->nm_sotype = argp->sotype; 758 nmp->nm_soproto = argp->proto; 759 760 nfs_decode_args(nmp, argp, l); 761 762 mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 763 mp->mnt_dev_bshift = DEV_BSHIFT; 764 765 /* 766 * For Connection based sockets (TCP,...) defer the connect until 767 * the first request, in case the server is not responding. 768 */ 769 if (nmp->nm_sotype == SOCK_DGRAM && 770 (error = nfs_connect(nmp, (struct nfsreq *)0, l))) 771 goto bad; 772 773 /* 774 * This is silly, but it has to be set so that vinifod() works. 775 * We do not want to do an nfs_statvfs() here since we can get 776 * stuck on a dead server and we are holding a lock on the mount 777 * point. 778 */ 779 mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; 780 error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np); 781 if (error) 782 goto bad; 783 *vpp = NFSTOV(np); 784 MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK); 785 VOP_GETATTR(*vpp, attrs, l->l_proc->p_cred, l); 786 if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) { 787 cr = kauth_cred_alloc(); 788 kauth_cred_setuid(cr, attrs->va_uid); 789 kauth_cred_seteuid(cr, attrs->va_uid); 790 kauth_cred_setsvuid(cr, attrs->va_uid); 791 kauth_cred_setgid(cr, attrs->va_gid); 792 kauth_cred_setegid(cr, attrs->va_gid); 793 kauth_cred_setsvgid(cr, attrs->va_gid); 794 nfs_cookieheuristic(*vpp, &nmp->nm_iflag, l, cr); 795 kauth_cred_free(cr); 796 } 797 FREE(attrs, M_TEMP); 798 799 /* 800 * A reference count is needed on the nfsnode representing the 801 * remote root. If this object is not persistent, then backward 802 * traversals of the mount point (i.e. "..") will not work if 803 * the nfsnode gets flushed out of the cache. Ufs does not have 804 * this problem, because one can identify root inodes by their 805 * number == ROOTINO (2). So, just unlock, but no rele. 806 */ 807 808 nmp->nm_vnode = *vpp; 809 VOP_UNLOCK(*vpp, 0); 810 811 nmp->nm_stats = iostat_alloc(IOSTAT_NFS); 812 nmp->nm_stats->parent = nmp; 813 /* generate a ficticious drive name for the nfs mount */ 814 MALLOC(nmp->nm_stats->name, char *, IOSTATNAMELEN, M_NFSMNT, M_WAITOK); 815 snprintf(nmp->nm_stats->name, IOSTATNAMELEN, "nfs%u", nfs_mount_count); 816 nfs_mount_count++; 817 818 return (0); 819 bad: 820 nfs_disconnect(nmp); 821 free((caddr_t)nmp, M_NFSMNT); 822 m_freem(nam); 823 return (error); 824 } 825 826 /* 827 * unmount system call 828 */ 829 int 830 nfs_unmount(mp, mntflags, l) 831 struct mount *mp; 832 int mntflags; 833 struct lwp *l; 834 { 835 struct nfsmount *nmp; 836 struct vnode *vp; 837 int error, flags = 0; 838 839 if (mntflags & MNT_FORCE) 840 flags |= FORCECLOSE; 841 nmp = VFSTONFS(mp); 842 /* 843 * Goes something like this.. 844 * - Check for activity on the root vnode (other than ourselves). 845 * - Call vflush() to clear out vnodes for this file system, 846 * except for the root vnode. 847 * - Decrement reference on the vnode representing remote root. 848 * - Close the socket 849 * - Free up the data structures 850 */ 851 /* 852 * We need to decrement the ref. count on the nfsnode representing 853 * the remote root. See comment in mountnfs(). The VFS unmount() 854 * has done vput on this vnode, otherwise we would get deadlock! 855 */ 856 vp = nmp->nm_vnode; 857 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 858 if (error != 0) 859 return error; 860 861 if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) { 862 vput(vp); 863 return (EBUSY); 864 } 865 866 /* 867 * Must handshake with nqnfs_clientd() if it is active. 868 */ 869 nmp->nm_iflag |= NFSMNT_DISMINPROG; 870 while (nmp->nm_inprog != NULLVP) 871 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0); 872 error = vflush(mp, vp, flags); 873 if (error) { 874 vput(vp); 875 nmp->nm_iflag &= ~NFSMNT_DISMINPROG; 876 return (error); 877 } 878 879 /* 880 * We are now committed to the unmount; mark the mount structure 881 * as doomed so that any sleepers kicked awake by nfs_disconnect 882 * will go away cleanly. 883 */ 884 nmp->nm_iflag |= NFSMNT_DISMNT; 885 886 /* 887 * Clean up the stats... note that we carefully avoid decrementing 888 * nfs_mount_count here for good reason - we may not be unmounting 889 * the last thing mounted. 890 */ 891 FREE(nmp->nm_stats->name, M_NFSMNT); 892 iostat_free(nmp->nm_stats); 893 894 /* 895 * There are two reference counts to get rid of here 896 * (see comment in mountnfs()). 897 */ 898 vrele(vp); 899 vput(vp); 900 vgone(vp); 901 nfs_disconnect(nmp); 902 m_freem(nmp->nm_nam); 903 904 /* 905 * For NQNFS, let the server daemon free the nfsmount structure. 906 */ 907 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0) 908 free((caddr_t)nmp, M_NFSMNT); 909 return (0); 910 } 911 912 /* 913 * Return root of a filesystem 914 */ 915 int 916 nfs_root(mp, vpp) 917 struct mount *mp; 918 struct vnode **vpp; 919 { 920 struct vnode *vp; 921 struct nfsmount *nmp; 922 int error; 923 924 nmp = VFSTONFS(mp); 925 vp = nmp->nm_vnode; 926 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 927 if (error != 0) 928 return error; 929 if (vp->v_type == VNON) 930 vp->v_type = VDIR; 931 vp->v_flag = VROOT; 932 *vpp = vp; 933 return (0); 934 } 935 936 extern int syncprt; 937 938 /* 939 * Flush out the buffer cache 940 */ 941 /* ARGSUSED */ 942 int 943 nfs_sync(mp, waitfor, cred, l) 944 struct mount *mp; 945 int waitfor; 946 kauth_cred_t cred; 947 struct lwp *l; 948 { 949 struct vnode *vp; 950 int error, allerror = 0; 951 952 /* 953 * Force stale buffer cache information to be flushed. 954 */ 955 loop: 956 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 957 /* 958 * If the vnode that we are about to sync is no longer 959 * associated with this mount point, start over. 960 */ 961 if (vp->v_mount != mp) 962 goto loop; 963 if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) || 964 (LIST_EMPTY(&vp->v_dirtyblkhd) && 965 vp->v_uobj.uo_npages == 0)) 966 continue; 967 if (vget(vp, LK_EXCLUSIVE)) 968 goto loop; 969 error = VOP_FSYNC(vp, cred, 970 waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, l); 971 if (error) 972 allerror = error; 973 vput(vp); 974 } 975 return (allerror); 976 } 977 978 /* 979 * NFS flat namespace lookup. 980 * Currently unsupported. 981 */ 982 /* ARGSUSED */ 983 int 984 nfs_vget(mp, ino, vpp) 985 struct mount *mp; 986 ino_t ino; 987 struct vnode **vpp; 988 { 989 990 return (EOPNOTSUPP); 991 } 992 993 /* 994 * Do that sysctl thang... 995 */ 996 static int 997 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS) 998 { 999 int error; 1000 1001 nfs_getset_niothreads(0); 1002 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 1003 if (error || newp == NULL) 1004 return (error); 1005 nfs_getset_niothreads(1); 1006 1007 return (0); 1008 } 1009 1010 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup") 1011 { 1012 1013 sysctl_createv(clog, 0, NULL, NULL, 1014 CTLFLAG_PERMANENT, 1015 CTLTYPE_NODE, "vfs", NULL, 1016 NULL, 0, NULL, 0, 1017 CTL_VFS, CTL_EOL); 1018 sysctl_createv(clog, 0, NULL, NULL, 1019 CTLFLAG_PERMANENT, 1020 CTLTYPE_NODE, "nfs", 1021 SYSCTL_DESCR("NFS vfs options"), 1022 NULL, 0, NULL, 0, 1023 CTL_VFS, 2, CTL_EOL); 1024 /* 1025 * XXX the "2" above could be dynamic, thereby eliminating one 1026 * more instance of the "number to vfs" mapping problem, but 1027 * "2" is the order as taken from sys/mount.h 1028 */ 1029 1030 sysctl_createv(clog, 0, NULL, NULL, 1031 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1032 CTLTYPE_STRUCT, "nfsstats", 1033 SYSCTL_DESCR("NFS operation statistics"), 1034 NULL, 0, &nfsstats, sizeof(nfsstats), 1035 CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL); 1036 sysctl_createv(clog, 0, NULL, NULL, 1037 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1038 CTLTYPE_INT, "iothreads", 1039 SYSCTL_DESCR("Number of NFS client processes desired"), 1040 sysctl_vfs_nfs_iothreads, 0, &nfs_niothreads, 0, 1041 CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL); 1042 } 1043 1044 /* 1045 * At this point, this should never happen 1046 */ 1047 /* ARGSUSED */ 1048 int 1049 nfs_fhtovp(mp, fhp, vpp) 1050 struct mount *mp; 1051 struct fid *fhp; 1052 struct vnode **vpp; 1053 { 1054 1055 return (EINVAL); 1056 } 1057 1058 /* 1059 * Vnode pointer to File handle, should never happen either 1060 */ 1061 /* ARGSUSED */ 1062 int 1063 nfs_vptofh(vp, fhp) 1064 struct vnode *vp; 1065 struct fid *fhp; 1066 { 1067 1068 return (EINVAL); 1069 } 1070 1071 /* 1072 * Vfs start routine, a no-op. 1073 */ 1074 /* ARGSUSED */ 1075 int 1076 nfs_start(mp, flags, l) 1077 struct mount *mp; 1078 int flags; 1079 struct lwp *l; 1080 { 1081 1082 return (0); 1083 } 1084 1085 /* 1086 * Do operations associated with quotas, not supported 1087 */ 1088 /* ARGSUSED */ 1089 int 1090 nfs_quotactl(mp, cmd, uid, arg, l) 1091 struct mount *mp; 1092 int cmd; 1093 uid_t uid; 1094 void *arg; 1095 struct lwp *l; 1096 { 1097 1098 return (EOPNOTSUPP); 1099 } 1100
Indexes created Wed Oct 01 07:09:59 GMT 2025