nfs_vfsops.c revision 1.152.4.2
1 /* $NetBSD: nfs_vfsops.c,v 1.152.4.2 2006/04/14 10:01:35 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 1.152.4.2 2006/04/14 10:01:35 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 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 nfs_mountroot, 113 (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp, 114 vfs_stdextattrctl, 115 nfs_vnodeopv_descs, 116 }; 117 VFS_ATTACH(nfs_vfsops); 118 119 extern u_int32_t nfs_procids[NFS_NPROCS]; 120 extern u_int32_t nfs_prog, nfs_vers; 121 122 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *, 123 struct mount **, struct vnode **, struct lwp *)); 124 125 /* 126 * nfs statvfs call 127 */ 128 int 129 nfs_statvfs(mp, sbp, l) 130 struct mount *mp; 131 struct statvfs *sbp; 132 struct lwp *l; 133 { 134 struct vnode *vp; 135 struct nfs_statfs *sfp; 136 caddr_t cp; 137 u_int32_t *tl; 138 int32_t t1, t2; 139 caddr_t bpos, dpos, cp2; 140 struct nfsmount *nmp = VFSTONFS(mp); 141 int error = 0, retattr; 142 #ifdef NFS_V2_ONLY 143 const int v3 = 0; 144 #else 145 int v3 = (nmp->nm_flag & NFSMNT_NFSV3); 146 #endif 147 struct mbuf *mreq, *mrep = NULL, *md, *mb; 148 kauth_cred_t cred; 149 u_quad_t tquad; 150 struct nfsnode *np; 151 152 #ifndef nolint 153 sfp = (struct nfs_statfs *)0; 154 #endif 155 vp = nmp->nm_vnode; 156 np = VTONFS(vp); 157 cred = kauth_cred_alloc(); 158 #ifndef NFS_V2_ONLY 159 if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 160 (void)nfs_fsinfo(nmp, vp, cred, l); 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, l, 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 kauth_cred_free(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, l) 226 struct nfsmount *nmp; 227 struct vnode *vp; 228 kauth_cred_t cred; 229 struct lwp *l; 230 { 231 struct nfsv3_fsinfo *fsp; 232 caddr_t cp; 233 int32_t t1, t2; 234 u_int32_t *tl, pref, xmax; 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, l, 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 xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 254 if (xmax < nmp->nm_wsize && xmax > 0) { 255 nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1); 256 if (nmp->nm_wsize == 0) 257 nmp->nm_wsize = xmax; 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 xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 265 if (xmax < nmp->nm_rsize && xmax > 0) { 266 nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1); 267 if (nmp->nm_rsize == 0) 268 nmp->nm_rsize = xmax; 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 (xmax < nmp->nm_readdirsize && xmax > 0) { 275 nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1); 276 if (nmp->nm_readdirsize == 0) 277 nmp->nm_readdirsize = xmax; 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_mountp->mnt_fs_bshift = 285 ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 286 nmp->nm_iflag |= NFSMNT_GOTFSINFO; 287 } 288 nfsm_reqdone; 289 return (error); 290 } 291 #endif 292 293 /* 294 * Mount a remote root fs via. NFS. It goes like this: 295 * - Call nfs_boot_init() to fill in the nfs_diskless struct 296 * - build the rootfs mount point and call mountnfs() to do the rest. 297 */ 298 int 299 nfs_mountroot() 300 { 301 struct nfs_diskless *nd; 302 struct vattr attr; 303 struct mount *mp; 304 struct vnode *vp; 305 struct lwp *l; 306 long n; 307 int error; 308 309 l = curlwp; /* XXX */ 310 311 if (device_class(root_device) != DV_IFNET) 312 return (ENODEV); 313 314 /* 315 * XXX time must be non-zero when we init the interface or else 316 * the arp code will wedge. [Fixed now in if_ether.c] 317 * However, the NFS attribute cache gives false "hits" when 318 * time.tv_sec < NFS_ATTRTIMEO(nmp, np) so keep this in for now. 319 */ 320 if (time.tv_sec < NFS_MAXATTRTIMO) 321 time.tv_sec = NFS_MAXATTRTIMO; 322 323 /* 324 * Call nfs_boot_init() to fill in the nfs_diskless struct. 325 * Side effect: Finds and configures a network interface. 326 */ 327 nd = malloc(sizeof(*nd), M_NFSMNT, M_WAITOK); 328 memset((caddr_t)nd, 0, sizeof(*nd)); 329 error = nfs_boot_init(nd, l); 330 if (error) { 331 free(nd, M_NFSMNT); 332 return (error); 333 } 334 335 /* 336 * Create the root mount point. 337 */ 338 error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l); 339 if (error) 340 goto out; 341 printf("root on %s\n", nd->nd_root.ndm_host); 342 343 /* 344 * Link it into the mount list. 345 */ 346 simple_lock(&mountlist_slock); 347 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 348 simple_unlock(&mountlist_slock); 349 rootvp = vp; 350 mp->mnt_vnodecovered = NULLVP; 351 vfs_unbusy(mp); 352 353 /* Get root attributes (for the time). */ 354 error = VOP_GETATTR(vp, &attr, l->l_proc->p_cred, l); 355 if (error) 356 panic("nfs_mountroot: getattr for root"); 357 n = attr.va_atime.tv_sec; 358 #ifdef DEBUG 359 printf("root time: 0x%lx\n", n); 360 #endif 361 setrootfstime(n); 362 363 out: 364 if (error) 365 nfs_boot_cleanup(nd, l); 366 free(nd, M_NFSMNT); 367 return (error); 368 } 369 370 /* 371 * Internal version of mount system call for diskless setup. 372 * Separate function because we used to call it twice. 373 * (once for root and once for swap) 374 */ 375 static int 376 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, l) 377 struct nfs_dlmount *ndmntp; 378 const char *mntname; /* mount point name */ 379 struct mount **mpp; 380 struct vnode **vpp; 381 struct lwp *l; 382 { 383 struct mount *mp; 384 struct mbuf *m; 385 int error; 386 387 vfs_rootmountalloc(MOUNT_NFS, mntname, &mp); 388 389 mp->mnt_op = &nfs_vfsops; 390 391 /* 392 * Historical practice expects NFS root file systems to 393 * be initially mounted r/w. 394 */ 395 mp->mnt_flag &= ~MNT_RDONLY; 396 397 /* Get mbuf for server sockaddr. */ 398 m = m_get(M_WAIT, MT_SONAME); 399 if (m == NULL) 400 panic("nfs_mountroot: mget soname for %s", mntname); 401 MCLAIM(m, &nfs_mowner); 402 memcpy(mtod(m, caddr_t), (caddr_t)ndmntp->ndm_args.addr, 403 (m->m_len = ndmntp->ndm_args.addr->sa_len)); 404 405 error = mountnfs(&ndmntp->ndm_args, mp, m, mntname, 406 ndmntp->ndm_args.hostname, vpp, l); 407 if (error) { 408 mp->mnt_op->vfs_refcount--; 409 vfs_unbusy(mp); 410 printf("nfs_mountroot: mount %s failed: %d\n", 411 mntname, error); 412 free(mp, M_MOUNT); 413 } else 414 *mpp = mp; 415 416 return (error); 417 } 418 419 void 420 nfs_decode_args(nmp, argp, l) 421 struct nfsmount *nmp; 422 struct nfs_args *argp; 423 struct lwp *l; 424 { 425 int s; 426 int adjsock; 427 int maxio; 428 429 s = splsoftnet(); 430 431 /* 432 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 433 * no sense in that context. 434 */ 435 if (argp->sotype == SOCK_STREAM) 436 argp->flags &= ~NFSMNT_NOCONN; 437 438 /* 439 * Cookie translation is not needed for v2, silently ignore it. 440 */ 441 if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) == 442 NFSMNT_XLATECOOKIE) 443 argp->flags &= ~NFSMNT_XLATECOOKIE; 444 445 /* Re-bind if rsrvd port requested and wasn't on one */ 446 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 447 && (argp->flags & NFSMNT_RESVPORT); 448 /* Also re-bind if we're switching to/from a connected UDP socket */ 449 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 450 (argp->flags & NFSMNT_NOCONN)); 451 452 /* Update flags. */ 453 nmp->nm_flag = argp->flags; 454 splx(s); 455 456 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 457 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 458 if (nmp->nm_timeo < NFS_MINTIMEO) 459 nmp->nm_timeo = NFS_MINTIMEO; 460 else if (nmp->nm_timeo > NFS_MAXTIMEO) 461 nmp->nm_timeo = NFS_MAXTIMEO; 462 } 463 464 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 465 nmp->nm_retry = argp->retrans; 466 if (nmp->nm_retry > NFS_MAXREXMIT) 467 nmp->nm_retry = NFS_MAXREXMIT; 468 } 469 470 #ifndef NFS_V2_ONLY 471 if (argp->flags & NFSMNT_NFSV3) { 472 if (argp->sotype == SOCK_DGRAM) 473 maxio = NFS_MAXDGRAMDATA; 474 else 475 maxio = NFS_MAXDATA; 476 } else 477 #endif 478 maxio = NFS_V2MAXDATA; 479 480 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 481 int osize = nmp->nm_wsize; 482 nmp->nm_wsize = argp->wsize; 483 /* Round down to multiple of blocksize */ 484 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 485 if (nmp->nm_wsize <= 0) 486 nmp->nm_wsize = NFS_FABLKSIZE; 487 adjsock |= (nmp->nm_wsize != osize); 488 } 489 if (nmp->nm_wsize > maxio) 490 nmp->nm_wsize = maxio; 491 if (nmp->nm_wsize > MAXBSIZE) 492 nmp->nm_wsize = MAXBSIZE; 493 494 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 495 int osize = nmp->nm_rsize; 496 nmp->nm_rsize = argp->rsize; 497 /* Round down to multiple of blocksize */ 498 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 499 if (nmp->nm_rsize <= 0) 500 nmp->nm_rsize = NFS_FABLKSIZE; 501 adjsock |= (nmp->nm_rsize != osize); 502 } 503 if (nmp->nm_rsize > maxio) 504 nmp->nm_rsize = maxio; 505 if (nmp->nm_rsize > MAXBSIZE) 506 nmp->nm_rsize = MAXBSIZE; 507 508 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 509 nmp->nm_readdirsize = argp->readdirsize; 510 /* Round down to multiple of minimum blocksize */ 511 nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1); 512 if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ) 513 nmp->nm_readdirsize = NFS_DIRFRAGSIZ; 514 /* Bigger than buffer size makes no sense */ 515 if (nmp->nm_readdirsize > NFS_DIRBLKSIZ) 516 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 517 } else if (argp->flags & NFSMNT_RSIZE) 518 nmp->nm_readdirsize = nmp->nm_rsize; 519 520 if (nmp->nm_readdirsize > maxio) 521 nmp->nm_readdirsize = maxio; 522 523 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && 524 argp->maxgrouplist <= NFS_MAXGRPS) 525 nmp->nm_numgrps = argp->maxgrouplist; 526 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && 527 argp->readahead <= NFS_MAXRAHEAD) 528 nmp->nm_readahead = argp->readahead; 529 if ((argp->flags & NFSMNT_LEASETERM) && argp->leaseterm >= 2 && 530 argp->leaseterm <= NQ_MAXLEASE) 531 nmp->nm_leaseterm = argp->leaseterm; 532 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && 533 argp->deadthresh <= NQ_NEVERDEAD) 534 nmp->nm_deadthresh = argp->deadthresh; 535 536 adjsock |= ((nmp->nm_sotype != argp->sotype) || 537 (nmp->nm_soproto != argp->proto)); 538 nmp->nm_sotype = argp->sotype; 539 nmp->nm_soproto = argp->proto; 540 541 if (nmp->nm_so && adjsock) { 542 nfs_safedisconnect(nmp); 543 if (nmp->nm_sotype == SOCK_DGRAM) 544 while (nfs_connect(nmp, (struct nfsreq *)0, l)) { 545 printf("nfs_args: retrying connect\n"); 546 (void) tsleep((caddr_t)&lbolt, 547 PSOCK, "nfscn3", 0); 548 } 549 } 550 } 551 552 /* 553 * VFS Operations. 554 * 555 * mount system call 556 * It seems a bit dumb to copyinstr() the host and path here and then 557 * memcpy() them in mountnfs(), but I wanted to detect errors before 558 * doing the sockargs() call because sockargs() allocates an mbuf and 559 * an error after that means that I have to release the mbuf. 560 */ 561 /* ARGSUSED */ 562 int 563 nfs_mount(mp, path, data, ndp, l) 564 struct mount *mp; 565 const char *path; 566 void *data; 567 struct nameidata *ndp; 568 struct lwp *l; 569 { 570 int error; 571 struct nfs_args args; 572 struct mbuf *nam; 573 struct nfsmount *nmp = VFSTONFS(mp); 574 struct sockaddr *sa; 575 struct vnode *vp; 576 char *pth, *hst; 577 struct proc *p; 578 size_t len; 579 u_char *nfh; 580 581 error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args)); 582 if (error) 583 return (error); 584 585 p = l->l_proc; 586 if (mp->mnt_flag & MNT_GETARGS) { 587 588 if (nmp == NULL) 589 return (EIO); 590 if (args.addr != NULL) { 591 sa = mtod(nmp->nm_nam, struct sockaddr *); 592 error = copyout(sa, args.addr, sa->sa_len); 593 if (error) 594 return (error); 595 args.addrlen = sa->sa_len; 596 } else 597 args.addrlen = 0; 598 599 args.version = NFS_ARGSVERSION; 600 args.sotype = nmp->nm_sotype; 601 args.proto = nmp->nm_soproto; 602 args.fh = NULL; 603 args.fhsize = 0; 604 args.flags = nmp->nm_flag; 605 args.wsize = nmp->nm_wsize; 606 args.rsize = nmp->nm_rsize; 607 args.readdirsize = nmp->nm_readdirsize; 608 args.timeo = nmp->nm_timeo; 609 args.retrans = nmp->nm_retry; 610 args.maxgrouplist = nmp->nm_numgrps; 611 args.readahead = nmp->nm_readahead; 612 args.leaseterm = nmp->nm_leaseterm; 613 args.deadthresh = nmp->nm_deadthresh; 614 args.hostname = NULL; 615 return (copyout(&args, data, sizeof(args))); 616 } 617 618 if (args.version != NFS_ARGSVERSION) 619 return (EPROGMISMATCH); 620 #ifdef NFS_V2_ONLY 621 if (args.flags & NFSMNT_NQNFS) 622 return (EPROGUNAVAIL); 623 if (args.flags & NFSMNT_NFSV3) 624 return (EPROGMISMATCH); 625 #endif 626 if (mp->mnt_flag & MNT_UPDATE) { 627 if (nmp == NULL) 628 return (EIO); 629 /* 630 * When doing an update, we can't change from or to 631 * v3 and/or nqnfs, or change cookie translation 632 */ 633 args.flags = (args.flags & 634 ~(NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)) | 635 (nmp->nm_flag & 636 (NFSMNT_NFSV3|NFSMNT_NQNFS|NFSMNT_XLATECOOKIE)); 637 nfs_decode_args(nmp, &args, l); 638 return (0); 639 } 640 if (args.fhsize < 0 || args.fhsize > NFSX_V3FHMAX) 641 return (EINVAL); 642 MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK); 643 error = copyin((caddr_t)args.fh, (caddr_t)nfh, args.fhsize); 644 if (error) 645 return (error); 646 MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK); 647 error = copyinstr(path, pth, MNAMELEN - 1, &len); 648 if (error) 649 goto free_nfh; 650 memset(&pth[len], 0, MNAMELEN - len); 651 MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK); 652 error = copyinstr(args.hostname, hst, MNAMELEN - 1, &len); 653 if (error) 654 goto free_pth; 655 memset(&hst[len], 0, MNAMELEN - len); 656 /* sockargs() call must be after above copyin() calls */ 657 error = sockargs(&nam, (caddr_t)args.addr, args.addrlen, MT_SONAME); 658 if (error) 659 goto free_hst; 660 MCLAIM(nam, &nfs_mowner); 661 args.fh = nfh; 662 error = mountnfs(&args, mp, nam, pth, hst, &vp, l); 663 664 free_hst: 665 FREE(hst, M_TEMP); 666 free_pth: 667 FREE(pth, M_TEMP); 668 free_nfh: 669 FREE(nfh, M_TEMP); 670 671 return (error); 672 } 673 674 /* 675 * Common code for mount and mountroot 676 */ 677 int 678 mountnfs(argp, mp, nam, pth, hst, vpp, l) 679 struct nfs_args *argp; 680 struct mount *mp; 681 struct mbuf *nam; 682 const char *pth, *hst; 683 struct vnode **vpp; 684 struct lwp *l; 685 { 686 struct nfsmount *nmp; 687 struct nfsnode *np; 688 int error; 689 struct vattr *attrs; 690 kauth_cred_t cr; 691 692 /* 693 * If the number of nfs iothreads to use has never 694 * been set, create a reasonable number of them. 695 */ 696 697 if (nfs_niothreads < 0) { 698 nfs_niothreads = NFS_DEFAULT_NIOTHREADS; 699 nfs_getset_niothreads(TRUE); 700 } 701 702 if (mp->mnt_flag & MNT_UPDATE) { 703 nmp = VFSTONFS(mp); 704 /* update paths, file handles, etc, here XXX */ 705 m_freem(nam); 706 return (0); 707 } else { 708 MALLOC(nmp, struct nfsmount *, sizeof (struct nfsmount), 709 M_NFSMNT, M_WAITOK); 710 memset((caddr_t)nmp, 0, sizeof (struct nfsmount)); 711 mp->mnt_data = nmp; 712 TAILQ_INIT(&nmp->nm_uidlruhead); 713 TAILQ_INIT(&nmp->nm_bufq); 714 lockinit(&nmp->nm_writeverflock, PRIBIO, "nfswverf", 0, 0); 715 simple_lock_init(&nmp->nm_slock); 716 } 717 vfs_getnewfsid(mp); 718 nmp->nm_mountp = mp; 719 720 #ifndef NFS_V2_ONLY 721 if (argp->flags & NFSMNT_NQNFS) 722 mp->mnt_iflag |= IMNT_DTYPE; 723 #endif 724 725 #ifndef NFS_V2_ONLY 726 if ((argp->flags & NFSMNT_NFSV3) == 0) 727 #endif 728 /* 729 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo 730 * will fill this in. 731 */ 732 nmp->nm_maxfilesize = 0xffffffffLL; 733 734 nmp->nm_timeo = NFS_TIMEO; 735 nmp->nm_retry = NFS_RETRANS; 736 nmp->nm_wsize = NFS_WSIZE; 737 nmp->nm_rsize = NFS_RSIZE; 738 nmp->nm_readdirsize = NFS_READDIRSIZE; 739 nmp->nm_numgrps = NFS_MAXGRPS; 740 nmp->nm_readahead = NFS_DEFRAHEAD; 741 nmp->nm_leaseterm = NQ_DEFLEASE; 742 nmp->nm_deadthresh = NQ_DEADTHRESH; 743 CIRCLEQ_INIT(&nmp->nm_timerhead); 744 nmp->nm_inprog = NULLVP; 745 error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, mp, l); 746 if (error) 747 goto bad; 748 nmp->nm_nam = nam; 749 750 /* Set up the sockets and per-host congestion */ 751 nmp->nm_sotype = argp->sotype; 752 nmp->nm_soproto = argp->proto; 753 754 nfs_decode_args(nmp, argp, l); 755 756 mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 757 mp->mnt_dev_bshift = DEV_BSHIFT; 758 759 /* 760 * For Connection based sockets (TCP,...) defer the connect until 761 * the first request, in case the server is not responding. 762 */ 763 if (nmp->nm_sotype == SOCK_DGRAM && 764 (error = nfs_connect(nmp, (struct nfsreq *)0, l))) 765 goto bad; 766 767 /* 768 * This is silly, but it has to be set so that vinifod() works. 769 * We do not want to do an nfs_statvfs() here since we can get 770 * stuck on a dead server and we are holding a lock on the mount 771 * point. 772 */ 773 mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; 774 error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np); 775 if (error) 776 goto bad; 777 *vpp = NFSTOV(np); 778 MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK); 779 VOP_GETATTR(*vpp, attrs, l->l_proc->p_cred, l); 780 if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) { 781 cr = kauth_cred_alloc(); 782 kauth_cred_setuid(cr, attrs->va_uid); 783 kauth_cred_seteuid(cr, attrs->va_uid); 784 kauth_cred_setsvuid(cr, attrs->va_uid); 785 kauth_cred_setgid(cr, attrs->va_gid); 786 kauth_cred_setegid(cr, attrs->va_gid); 787 kauth_cred_setsvgid(cr, attrs->va_gid); 788 nfs_cookieheuristic(*vpp, &nmp->nm_iflag, l, cr); 789 kauth_cred_free(cr); 790 } 791 FREE(attrs, M_TEMP); 792 793 /* 794 * A reference count is needed on the nfsnode representing the 795 * remote root. If this object is not persistent, then backward 796 * traversals of the mount point (i.e. "..") will not work if 797 * the nfsnode gets flushed out of the cache. Ufs does not have 798 * this problem, because one can identify root inodes by their 799 * number == ROOTINO (2). So, just unlock, but no rele. 800 */ 801 802 nmp->nm_vnode = *vpp; 803 VOP_UNLOCK(*vpp, 0); 804 805 return (0); 806 bad: 807 nfs_disconnect(nmp); 808 free((caddr_t)nmp, M_NFSMNT); 809 m_freem(nam); 810 return (error); 811 } 812 813 /* 814 * unmount system call 815 */ 816 int 817 nfs_unmount(mp, mntflags, l) 818 struct mount *mp; 819 int mntflags; 820 struct lwp *l; 821 { 822 struct nfsmount *nmp; 823 struct vnode *vp; 824 int error, flags = 0; 825 826 if (mntflags & MNT_FORCE) 827 flags |= FORCECLOSE; 828 nmp = VFSTONFS(mp); 829 /* 830 * Goes something like this.. 831 * - Check for activity on the root vnode (other than ourselves). 832 * - Call vflush() to clear out vnodes for this file system, 833 * except for the root vnode. 834 * - Decrement reference on the vnode representing remote root. 835 * - Close the socket 836 * - Free up the data structures 837 */ 838 /* 839 * We need to decrement the ref. count on the nfsnode representing 840 * the remote root. See comment in mountnfs(). The VFS unmount() 841 * has done vput on this vnode, otherwise we would get deadlock! 842 */ 843 vp = nmp->nm_vnode; 844 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 845 if (error != 0) 846 return error; 847 848 if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) { 849 vput(vp); 850 return (EBUSY); 851 } 852 853 /* 854 * Must handshake with nqnfs_clientd() if it is active. 855 */ 856 nmp->nm_iflag |= NFSMNT_DISMINPROG; 857 while (nmp->nm_inprog != NULLVP) 858 (void) tsleep((caddr_t)&lbolt, PSOCK, "nfsdism", 0); 859 error = vflush(mp, vp, flags); 860 if (error) { 861 vput(vp); 862 nmp->nm_iflag &= ~NFSMNT_DISMINPROG; 863 return (error); 864 } 865 866 /* 867 * We are now committed to the unmount; mark the mount structure 868 * as doomed so that any sleepers kicked awake by nfs_disconnect 869 * will go away cleanly. 870 */ 871 nmp->nm_iflag |= NFSMNT_DISMNT; 872 873 /* 874 * There are two reference counts to get rid of here 875 * (see comment in mountnfs()). 876 */ 877 vrele(vp); 878 vput(vp); 879 vgone(vp); 880 nfs_disconnect(nmp); 881 m_freem(nmp->nm_nam); 882 883 /* 884 * For NQNFS, let the server daemon free the nfsmount structure. 885 */ 886 if ((nmp->nm_flag & (NFSMNT_NQNFS | NFSMNT_KERB)) == 0) 887 free((caddr_t)nmp, M_NFSMNT); 888 return (0); 889 } 890 891 /* 892 * Return root of a filesystem 893 */ 894 int 895 nfs_root(mp, vpp) 896 struct mount *mp; 897 struct vnode **vpp; 898 { 899 struct vnode *vp; 900 struct nfsmount *nmp; 901 int error; 902 903 nmp = VFSTONFS(mp); 904 vp = nmp->nm_vnode; 905 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 906 if (error != 0) 907 return error; 908 if (vp->v_type == VNON) 909 vp->v_type = VDIR; 910 vp->v_flag = VROOT; 911 *vpp = vp; 912 return (0); 913 } 914 915 extern int syncprt; 916 917 /* 918 * Flush out the buffer cache 919 */ 920 /* ARGSUSED */ 921 int 922 nfs_sync(mp, waitfor, cred, l) 923 struct mount *mp; 924 int waitfor; 925 kauth_cred_t cred; 926 struct lwp *l; 927 { 928 struct vnode *vp; 929 int error, allerror = 0; 930 931 /* 932 * Force stale buffer cache information to be flushed. 933 */ 934 loop: 935 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 936 /* 937 * If the vnode that we are about to sync is no longer 938 * associated with this mount point, start over. 939 */ 940 if (vp->v_mount != mp) 941 goto loop; 942 if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) || 943 (LIST_EMPTY(&vp->v_dirtyblkhd) && 944 vp->v_uobj.uo_npages == 0)) 945 continue; 946 if (vget(vp, LK_EXCLUSIVE)) 947 goto loop; 948 error = VOP_FSYNC(vp, cred, 949 waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0, l); 950 if (error) 951 allerror = error; 952 vput(vp); 953 } 954 return (allerror); 955 } 956 957 /* 958 * NFS flat namespace lookup. 959 * Currently unsupported. 960 */ 961 /* ARGSUSED */ 962 int 963 nfs_vget(mp, ino, vpp) 964 struct mount *mp; 965 ino_t ino; 966 struct vnode **vpp; 967 { 968 969 return (EOPNOTSUPP); 970 } 971 972 /* 973 * Do that sysctl thang... 974 */ 975 static int 976 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS) 977 { 978 int error; 979 980 nfs_getset_niothreads(0); 981 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 982 if (error || newp == NULL) 983 return (error); 984 nfs_getset_niothreads(1); 985 986 return (0); 987 } 988 989 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup") 990 { 991 992 sysctl_createv(clog, 0, NULL, NULL, 993 CTLFLAG_PERMANENT, 994 CTLTYPE_NODE, "vfs", NULL, 995 NULL, 0, NULL, 0, 996 CTL_VFS, CTL_EOL); 997 sysctl_createv(clog, 0, NULL, NULL, 998 CTLFLAG_PERMANENT, 999 CTLTYPE_NODE, "nfs", 1000 SYSCTL_DESCR("NFS vfs options"), 1001 NULL, 0, NULL, 0, 1002 CTL_VFS, 2, CTL_EOL); 1003 /* 1004 * XXX the "2" above could be dynamic, thereby eliminating one 1005 * more instance of the "number to vfs" mapping problem, but 1006 * "2" is the order as taken from sys/mount.h 1007 */ 1008 1009 sysctl_createv(clog, 0, NULL, NULL, 1010 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1011 CTLTYPE_STRUCT, "nfsstats", 1012 SYSCTL_DESCR("NFS operation statistics"), 1013 NULL, 0, &nfsstats, sizeof(nfsstats), 1014 CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL); 1015 sysctl_createv(clog, 0, NULL, NULL, 1016 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1017 CTLTYPE_INT, "iothreads", 1018 SYSCTL_DESCR("Number of NFS client processes desired"), 1019 sysctl_vfs_nfs_iothreads, 0, &nfs_niothreads, 0, 1020 CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL); 1021 } 1022 1023 /* 1024 * At this point, this should never happen 1025 */ 1026 /* ARGSUSED */ 1027 int 1028 nfs_fhtovp(mp, fhp, vpp) 1029 struct mount *mp; 1030 struct fid *fhp; 1031 struct vnode **vpp; 1032 { 1033 1034 return (EINVAL); 1035 } 1036 1037 /* 1038 * Vnode pointer to File handle, should never happen either 1039 */ 1040 /* ARGSUSED */ 1041 int 1042 nfs_vptofh(vp, fhp) 1043 struct vnode *vp; 1044 struct fid *fhp; 1045 { 1046 1047 return (EINVAL); 1048 } 1049 1050 /* 1051 * Vfs start routine, a no-op. 1052 */ 1053 /* ARGSUSED */ 1054 int 1055 nfs_start(mp, flags, l) 1056 struct mount *mp; 1057 int flags; 1058 struct lwp *l; 1059 { 1060 1061 return (0); 1062 } 1063 1064 /* 1065 * Do operations associated with quotas, not supported 1066 */ 1067 /* ARGSUSED */ 1068 int 1069 nfs_quotactl(mp, cmd, uid, arg, l) 1070 struct mount *mp; 1071 int cmd; 1072 uid_t uid; 1073 void *arg; 1074 struct lwp *l; 1075 { 1076 1077 return (EOPNOTSUPP); 1078 } 1079
Indexes created Wed Oct 01 07:09:59 GMT 2025