nfs_vfsops.c revision 1.193
1 /* $NetBSD: nfs_vfsops.c,v 1.193 2008/01/28 14:31:20 dholland 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 1.193 2008/01/28 14:31:20 dholland 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 #include <sys/kauth.h> 63 64 #include <net/if.h> 65 #include <net/route.h> 66 #include <netinet/in.h> 67 68 #include <nfs/rpcv2.h> 69 #include <nfs/nfsproto.h> 70 #include <nfs/nfsnode.h> 71 #include <nfs/nfs.h> 72 #include <nfs/nfsmount.h> 73 #include <nfs/xdr_subs.h> 74 #include <nfs/nfsm_subs.h> 75 #include <nfs/nfsdiskless.h> 76 #include <nfs/nfs_var.h> 77 78 extern struct nfsstats nfsstats; 79 extern int nfs_ticks; 80 81 /* 82 * keep a count of the nfs mounts to generate ficticious drive names 83 * for the per drive stats. 84 */ 85 unsigned int nfs_mount_count = 0; 86 87 /* 88 * nfs vfs operations. 89 */ 90 91 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc; 92 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc; 93 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc; 94 95 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = { 96 &nfsv2_vnodeop_opv_desc, 97 &spec_nfsv2nodeop_opv_desc, 98 &fifo_nfsv2nodeop_opv_desc, 99 NULL, 100 }; 101 102 struct vfsops nfs_vfsops = { 103 MOUNT_NFS, 104 sizeof (struct nfs_args), 105 nfs_mount, 106 nfs_start, 107 nfs_unmount, 108 nfs_root, 109 (void *)eopnotsupp, /* vfs_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 (void *)eopnotsupp, /* vfs_suspendctl */ 122 genfs_renamelock_enter, 123 genfs_renamelock_exit, 124 nfs_vnodeopv_descs, 125 0, 126 { NULL, NULL }, 127 }; 128 VFS_ATTACH(nfs_vfsops); 129 130 extern u_int32_t nfs_procids[NFS_NPROCS]; 131 extern u_int32_t nfs_prog, nfs_vers; 132 133 static int nfs_mount_diskless __P((struct nfs_dlmount *, const char *, 134 struct mount **, struct vnode **, struct lwp *)); 135 136 /* 137 * nfs statvfs call 138 */ 139 int 140 nfs_statvfs(mp, sbp) 141 struct mount *mp; 142 struct statvfs *sbp; 143 { 144 struct lwp *l = curlwp; 145 struct vnode *vp; 146 struct nfs_statfs *sfp; 147 char *cp; 148 u_int32_t *tl; 149 int32_t t1, t2; 150 char *bpos, *dpos, *cp2; 151 struct nfsmount *nmp = VFSTONFS(mp); 152 int error = 0, retattr; 153 #ifdef NFS_V2_ONLY 154 const int v3 = 0; 155 #else 156 int v3 = (nmp->nm_flag & NFSMNT_NFSV3); 157 #endif 158 struct mbuf *mreq, *mrep = NULL, *md, *mb; 159 kauth_cred_t cred; 160 u_quad_t tquad; 161 struct nfsnode *np; 162 163 #ifndef nolint 164 sfp = (struct nfs_statfs *)0; 165 #endif 166 vp = nmp->nm_vnode; 167 np = VTONFS(vp); 168 cred = kauth_cred_alloc(); 169 #ifndef NFS_V2_ONLY 170 if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 171 (void)nfs_fsinfo(nmp, vp, cred, l); 172 #endif 173 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 174 nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3)); 175 nfsm_fhtom(np, v3); 176 nfsm_request(np, NFSPROC_FSSTAT, l, cred); 177 if (v3) 178 nfsm_postop_attr(vp, retattr, 0); 179 if (error) { 180 if (mrep != NULL) { 181 if (mrep->m_next != NULL) 182 printf("nfs_vfsops: nfs_statvfs would lose buffers\n"); 183 m_freem(mrep); 184 } 185 goto nfsmout; 186 } 187 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 188 sbp->f_flag = nmp->nm_flag; 189 sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize); 190 if (v3) { 191 sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE; 192 tquad = fxdr_hyper(&sfp->sf_tbytes); 193 sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 194 tquad = fxdr_hyper(&sfp->sf_fbytes); 195 sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 196 tquad = fxdr_hyper(&sfp->sf_abytes); 197 tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 198 sbp->f_bresvd = sbp->f_bfree - tquad; 199 sbp->f_bavail = tquad; 200 #ifdef COMPAT_20 201 /* Handle older NFS servers returning negative values */ 202 if ((quad_t)sbp->f_bavail < 0) 203 sbp->f_bavail = 0; 204 #endif 205 tquad = fxdr_hyper(&sfp->sf_tfiles); 206 sbp->f_files = tquad; 207 tquad = fxdr_hyper(&sfp->sf_ffiles); 208 sbp->f_ffree = tquad; 209 sbp->f_favail = tquad; 210 sbp->f_fresvd = 0; 211 sbp->f_namemax = MAXNAMLEN; 212 } else { 213 sbp->f_bsize = NFS_FABLKSIZE; 214 sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 215 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 216 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 217 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 218 sbp->f_fresvd = 0; 219 sbp->f_files = 0; 220 sbp->f_ffree = 0; 221 sbp->f_favail = 0; 222 sbp->f_fresvd = 0; 223 sbp->f_namemax = MAXNAMLEN; 224 } 225 copy_statvfs_info(sbp, mp); 226 nfsm_reqdone; 227 kauth_cred_free(cred); 228 return (error); 229 } 230 231 #ifndef NFS_V2_ONLY 232 /* 233 * nfs version 3 fsinfo rpc call 234 */ 235 int 236 nfs_fsinfo(nmp, vp, cred, l) 237 struct nfsmount *nmp; 238 struct vnode *vp; 239 kauth_cred_t cred; 240 struct lwp *l; 241 { 242 struct nfsv3_fsinfo *fsp; 243 char *cp; 244 int32_t t1, t2; 245 u_int32_t *tl, pref, xmax; 246 char *bpos, *dpos, *cp2; 247 int error = 0, retattr; 248 struct mbuf *mreq, *mrep, *md, *mb; 249 u_int64_t maxfsize; 250 struct nfsnode *np = VTONFS(vp); 251 252 nfsstats.rpccnt[NFSPROC_FSINFO]++; 253 nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1)); 254 nfsm_fhtom(np, 1); 255 nfsm_request(np, NFSPROC_FSINFO, l, cred); 256 nfsm_postop_attr(vp, retattr, 0); 257 if (!error) { 258 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 259 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 260 if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 && 261 pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 262 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 263 ~(NFS_FABLKSIZE - 1); 264 xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 265 if (xmax < nmp->nm_wsize && xmax > 0) { 266 nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1); 267 if (nmp->nm_wsize == 0) 268 nmp->nm_wsize = xmax; 269 } 270 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 271 if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 && 272 pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 273 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 274 ~(NFS_FABLKSIZE - 1); 275 xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 276 if (xmax < nmp->nm_rsize && xmax > 0) { 277 nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1); 278 if (nmp->nm_rsize == 0) 279 nmp->nm_rsize = xmax; 280 } 281 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 282 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ) 283 nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) & 284 ~(NFS_DIRFRAGSIZ - 1); 285 if (xmax < nmp->nm_readdirsize && xmax > 0) { 286 nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1); 287 if (nmp->nm_readdirsize == 0) 288 nmp->nm_readdirsize = xmax; 289 } 290 /* XXX */ 291 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 292 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 293 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 294 nmp->nm_maxfilesize = maxfsize; 295 nmp->nm_mountp->mnt_fs_bshift = 296 ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 297 nmp->nm_iflag |= NFSMNT_GOTFSINFO; 298 } 299 nfsm_reqdone; 300 return (error); 301 } 302 #endif 303 304 /* 305 * Mount a remote root fs via. NFS. It goes like this: 306 * - Call nfs_boot_init() to fill in the nfs_diskless struct 307 * - build the rootfs mount point and call mountnfs() to do the rest. 308 */ 309 int 310 nfs_mountroot() 311 { 312 struct timespec ts; 313 struct nfs_diskless *nd; 314 struct vattr attr; 315 struct mount *mp; 316 struct vnode *vp; 317 struct lwp *l; 318 long n; 319 int error; 320 321 l = curlwp; /* XXX */ 322 323 if (device_class(root_device) != DV_IFNET) 324 return (ENODEV); 325 326 /* 327 * XXX time must be non-zero when we init the interface or else 328 * the arp code will wedge. [Fixed now in if_ether.c] 329 * However, the NFS attribute cache gives false "hits" when the 330 * current time < nfs_attrtimeo(nmp, np) so keep this in for now. 331 */ 332 if (time_second < NFS_MAXATTRTIMO) { 333 ts.tv_sec = NFS_MAXATTRTIMO; 334 ts.tv_nsec = 0; 335 tc_setclock(&ts); 336 } 337 338 /* 339 * Call nfs_boot_init() to fill in the nfs_diskless struct. 340 * Side effect: Finds and configures a network interface. 341 */ 342 nd = kmem_alloc(sizeof(*nd), KM_SLEEP); 343 memset(nd, 0, sizeof(*nd)); 344 error = nfs_boot_init(nd, l); 345 if (error) { 346 kmem_free(nd, sizeof(*nd)); 347 return (error); 348 } 349 350 /* 351 * Create the root mount point. 352 */ 353 error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l); 354 if (error) 355 goto out; 356 printf("root on %s\n", nd->nd_root.ndm_host); 357 358 /* 359 * Link it into the mount list. 360 */ 361 mutex_enter(&mountlist_lock); 362 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 363 mutex_exit(&mountlist_lock); 364 rootvp = vp; 365 mp->mnt_vnodecovered = NULLVP; 366 vfs_unbusy(mp); 367 368 /* Get root attributes (for the time). */ 369 error = VOP_GETATTR(vp, &attr, l->l_cred); 370 if (error) 371 panic("nfs_mountroot: getattr for root"); 372 n = attr.va_atime.tv_sec; 373 #ifdef DEBUG 374 printf("root time: 0x%lx\n", n); 375 #endif 376 setrootfstime(n); 377 378 out: 379 if (error) 380 nfs_boot_cleanup(nd, l); 381 kmem_free(nd, sizeof(*nd)); 382 return (error); 383 } 384 385 /* 386 * Internal version of mount system call for diskless setup. 387 * Separate function because we used to call it twice. 388 * (once for root and once for swap) 389 */ 390 static int 391 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, l) 392 struct nfs_dlmount *ndmntp; 393 const char *mntname; /* mount point name */ 394 struct mount **mpp; 395 struct vnode **vpp; 396 struct lwp *l; 397 { 398 struct mount *mp; 399 struct mbuf *m; 400 int error; 401 402 vfs_rootmountalloc(MOUNT_NFS, mntname, &mp); 403 404 mp->mnt_op = &nfs_vfsops; 405 406 /* 407 * Historical practice expects NFS root file systems to 408 * be initially mounted r/w. 409 */ 410 mp->mnt_flag &= ~MNT_RDONLY; 411 412 /* Get mbuf for server sockaddr. */ 413 m = m_get(M_WAIT, MT_SONAME); 414 if (m == NULL) 415 panic("nfs_mountroot: mget soname for %s", mntname); 416 MCLAIM(m, &nfs_mowner); 417 memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr, 418 (m->m_len = ndmntp->ndm_args.addr->sa_len)); 419 420 error = mountnfs(&ndmntp->ndm_args, mp, m, mntname, 421 ndmntp->ndm_args.hostname, vpp, l); 422 if (error) { 423 mp->mnt_op->vfs_refcount--; 424 vfs_unbusy(mp); 425 printf("nfs_mountroot: mount %s failed: %d\n", 426 mntname, error); 427 free(mp, M_MOUNT); 428 } else 429 *mpp = mp; 430 431 return (error); 432 } 433 434 void 435 nfs_decode_args(nmp, argp, l) 436 struct nfsmount *nmp; 437 struct nfs_args *argp; 438 struct lwp *l; 439 { 440 int s; 441 int adjsock; 442 int maxio; 443 444 s = splsoftnet(); 445 446 /* 447 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 448 * no sense in that context. 449 */ 450 if (argp->sotype == SOCK_STREAM) 451 argp->flags &= ~NFSMNT_NOCONN; 452 453 /* 454 * Cookie translation is not needed for v2, silently ignore it. 455 */ 456 if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) == 457 NFSMNT_XLATECOOKIE) 458 argp->flags &= ~NFSMNT_XLATECOOKIE; 459 460 /* Re-bind if rsrvd port requested and wasn't on one */ 461 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 462 && (argp->flags & NFSMNT_RESVPORT); 463 /* Also re-bind if we're switching to/from a connected UDP socket */ 464 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 465 (argp->flags & NFSMNT_NOCONN)); 466 467 /* Update flags. */ 468 nmp->nm_flag = argp->flags; 469 splx(s); 470 471 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 472 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 473 if (nmp->nm_timeo < NFS_MINTIMEO) 474 nmp->nm_timeo = NFS_MINTIMEO; 475 else if (nmp->nm_timeo > NFS_MAXTIMEO) 476 nmp->nm_timeo = NFS_MAXTIMEO; 477 } 478 479 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 480 nmp->nm_retry = argp->retrans; 481 if (nmp->nm_retry > NFS_MAXREXMIT) 482 nmp->nm_retry = NFS_MAXREXMIT; 483 } 484 485 #ifndef NFS_V2_ONLY 486 if (argp->flags & NFSMNT_NFSV3) { 487 if (argp->sotype == SOCK_DGRAM) 488 maxio = NFS_MAXDGRAMDATA; 489 else 490 maxio = NFS_MAXDATA; 491 } else 492 #endif 493 maxio = NFS_V2MAXDATA; 494 495 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 496 int osize = nmp->nm_wsize; 497 nmp->nm_wsize = argp->wsize; 498 /* Round down to multiple of blocksize */ 499 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 500 if (nmp->nm_wsize <= 0) 501 nmp->nm_wsize = NFS_FABLKSIZE; 502 adjsock |= (nmp->nm_wsize != osize); 503 } 504 if (nmp->nm_wsize > maxio) 505 nmp->nm_wsize = maxio; 506 if (nmp->nm_wsize > MAXBSIZE) 507 nmp->nm_wsize = MAXBSIZE; 508 509 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 510 int osize = nmp->nm_rsize; 511 nmp->nm_rsize = argp->rsize; 512 /* Round down to multiple of blocksize */ 513 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 514 if (nmp->nm_rsize <= 0) 515 nmp->nm_rsize = NFS_FABLKSIZE; 516 adjsock |= (nmp->nm_rsize != osize); 517 } 518 if (nmp->nm_rsize > maxio) 519 nmp->nm_rsize = maxio; 520 if (nmp->nm_rsize > MAXBSIZE) 521 nmp->nm_rsize = MAXBSIZE; 522 523 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 524 nmp->nm_readdirsize = argp->readdirsize; 525 /* Round down to multiple of minimum blocksize */ 526 nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1); 527 if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ) 528 nmp->nm_readdirsize = NFS_DIRFRAGSIZ; 529 /* Bigger than buffer size makes no sense */ 530 if (nmp->nm_readdirsize > NFS_DIRBLKSIZ) 531 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 532 } else if (argp->flags & NFSMNT_RSIZE) 533 nmp->nm_readdirsize = nmp->nm_rsize; 534 535 if (nmp->nm_readdirsize > maxio) 536 nmp->nm_readdirsize = maxio; 537 538 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && 539 argp->maxgrouplist <= NFS_MAXGRPS) 540 nmp->nm_numgrps = argp->maxgrouplist; 541 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && 542 argp->readahead <= NFS_MAXRAHEAD) 543 nmp->nm_readahead = argp->readahead; 544 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && 545 argp->deadthresh <= NFS_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 kpause("nfscn3", false, hz, NULL); 559 } 560 } 561 } 562 563 /* 564 * VFS Operations. 565 * 566 * mount system call 567 * It seems a bit dumb to copyinstr() the host and path here and then 568 * memcpy() them in mountnfs(), but I wanted to detect errors before 569 * doing the sockargs() call because sockargs() allocates an mbuf and 570 * an error after that means that I have to release the mbuf. 571 */ 572 /* ARGSUSED */ 573 int 574 nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 575 { 576 struct lwp *l = curlwp; 577 int error; 578 struct nfs_args *args = data; 579 struct mbuf *nam; 580 struct nfsmount *nmp = VFSTONFS(mp); 581 struct sockaddr *sa; 582 struct vnode *vp; 583 char *pth, *hst; 584 struct proc *p; 585 size_t len; 586 u_char *nfh; 587 588 if (*data_len < sizeof *args) 589 return EINVAL; 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 = 0; /* dummy */ 619 args->deadthresh = nmp->nm_deadthresh; 620 args->hostname = NULL; 621 *data_len = sizeof *args; 622 return 0; 623 } 624 625 if (args->version != NFS_ARGSVERSION) 626 return (EPROGMISMATCH); 627 if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB)) 628 return (EPROGUNAVAIL); 629 #ifdef NFS_V2_ONLY 630 if (args->flags & NFSMNT_NFSV3) 631 return (EPROGMISMATCH); 632 #endif 633 if (mp->mnt_flag & MNT_UPDATE) { 634 if (nmp == NULL) 635 return (EIO); 636 /* 637 * When doing an update, we can't change from or to 638 * v3, or change cookie translation 639 */ 640 args->flags = (args->flags & ~(NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)) | 641 (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)); 642 nfs_decode_args(nmp, args, l); 643 return (0); 644 } 645 if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX) 646 return (EINVAL); 647 MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK); 648 error = copyin(args->fh, nfh, args->fhsize); 649 if (error) 650 goto free_nfh; 651 MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK); 652 error = copyinstr(path, pth, MNAMELEN - 1, &len); 653 if (error) 654 goto free_pth; 655 memset(&pth[len], 0, MNAMELEN - len); 656 MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK); 657 error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len); 658 if (error) 659 goto free_hst; 660 memset(&hst[len], 0, MNAMELEN - len); 661 /* sockargs() call must be after above copyin() calls */ 662 error = sockargs(&nam, args->addr, args->addrlen, MT_SONAME); 663 if (error) 664 goto free_hst; 665 MCLAIM(nam, &nfs_mowner); 666 args->fh = nfh; 667 error = mountnfs(args, mp, nam, pth, hst, &vp, l); 668 669 free_hst: 670 FREE(hst, M_TEMP); 671 free_pth: 672 FREE(pth, M_TEMP); 673 free_nfh: 674 FREE(nfh, M_TEMP); 675 676 return (error); 677 } 678 679 /* 680 * Common code for mount and mountroot 681 */ 682 int 683 mountnfs(argp, mp, nam, pth, hst, vpp, l) 684 struct nfs_args *argp; 685 struct mount *mp; 686 struct mbuf *nam; 687 const char *pth, *hst; 688 struct vnode **vpp; 689 struct lwp *l; 690 { 691 struct nfsmount *nmp; 692 struct nfsnode *np; 693 int error; 694 struct vattr *attrs; 695 kauth_cred_t cr; 696 char iosname[IOSTATNAMELEN]; 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_set_niothreads(NFS_DEFAULT_NIOTHREADS); 705 } 706 707 if (mp->mnt_flag & MNT_UPDATE) { 708 nmp = VFSTONFS(mp); 709 /* update paths, file handles, etc, here XXX */ 710 m_freem(nam); 711 return (0); 712 } else { 713 nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP); 714 mp->mnt_data = nmp; 715 TAILQ_INIT(&nmp->nm_uidlruhead); 716 TAILQ_INIT(&nmp->nm_bufq); 717 rw_init(&nmp->nm_writeverflock); 718 mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE); 719 cv_init(&nmp->nm_rcvcv, "nfsrcv"); 720 cv_init(&nmp->nm_sndcv, "nfssnd"); 721 cv_init(&nmp->nm_aiocv, "nfsaio"); 722 cv_init(&nmp->nm_disconcv, "nfsdis"); 723 } 724 vfs_getnewfsid(mp); 725 nmp->nm_mountp = mp; 726 727 #ifndef NFS_V2_ONLY 728 if ((argp->flags & NFSMNT_NFSV3) == 0) 729 #endif 730 { 731 /* 732 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo 733 * will fill this in. 734 */ 735 nmp->nm_maxfilesize = 0xffffffffLL; 736 if (argp->fhsize != NFSX_V2FH) { 737 return EINVAL; 738 } 739 } 740 741 nmp->nm_timeo = NFS_TIMEO; 742 nmp->nm_retry = NFS_RETRANS; 743 nmp->nm_wsize = NFS_WSIZE; 744 nmp->nm_rsize = NFS_RSIZE; 745 nmp->nm_readdirsize = NFS_READDIRSIZE; 746 nmp->nm_numgrps = NFS_MAXGRPS; 747 nmp->nm_readahead = NFS_DEFRAHEAD; 748 nmp->nm_deadthresh = NFS_DEFDEADTHRESH; 749 error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, 750 mp->mnt_op->vfs_name, mp, l); 751 if (error) 752 goto bad; 753 nmp->nm_nam = nam; 754 755 /* Set up the sockets and per-host congestion */ 756 nmp->nm_sotype = argp->sotype; 757 nmp->nm_soproto = argp->proto; 758 759 nfs_decode_args(nmp, argp, l); 760 761 mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 762 mp->mnt_dev_bshift = DEV_BSHIFT; 763 764 /* 765 * For Connection based sockets (TCP,...) defer the connect until 766 * the first request, in case the server is not responding. 767 */ 768 if (nmp->nm_sotype == SOCK_DGRAM && 769 (error = nfs_connect(nmp, (struct nfsreq *)0, l))) 770 goto bad; 771 772 /* 773 * This is silly, but it has to be set so that vinifod() works. 774 * We do not want to do an nfs_statvfs() here since we can get 775 * stuck on a dead server and we are holding a lock on the mount 776 * point. 777 */ 778 mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; 779 error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np); 780 if (error) 781 goto bad; 782 *vpp = NFSTOV(np); 783 MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK); 784 VOP_GETATTR(*vpp, attrs, l->l_cred); 785 if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) { 786 cr = kauth_cred_alloc(); 787 kauth_cred_setuid(cr, attrs->va_uid); 788 kauth_cred_seteuid(cr, attrs->va_uid); 789 kauth_cred_setsvuid(cr, attrs->va_uid); 790 kauth_cred_setgid(cr, attrs->va_gid); 791 kauth_cred_setegid(cr, attrs->va_gid); 792 kauth_cred_setsvgid(cr, attrs->va_gid); 793 nfs_cookieheuristic(*vpp, &nmp->nm_iflag, l, cr); 794 kauth_cred_free(cr); 795 } 796 FREE(attrs, M_TEMP); 797 798 /* 799 * A reference count is needed on the nfsnode representing the 800 * remote root. If this object is not persistent, then backward 801 * traversals of the mount point (i.e. "..") will not work if 802 * the nfsnode gets flushed out of the cache. Ufs does not have 803 * this problem, because one can identify root inodes by their 804 * number == ROOTINO (2). So, just unlock, but no rele. 805 */ 806 807 nmp->nm_vnode = *vpp; 808 VOP_UNLOCK(*vpp, 0); 809 810 snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++); 811 nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname); 812 813 return (0); 814 bad: 815 nfs_disconnect(nmp); 816 rw_destroy(&nmp->nm_writeverflock); 817 mutex_destroy(&nmp->nm_lock); 818 cv_destroy(&nmp->nm_rcvcv); 819 cv_destroy(&nmp->nm_sndcv); 820 cv_destroy(&nmp->nm_aiocv); 821 cv_destroy(&nmp->nm_disconcv); 822 kmem_free(nmp, sizeof(*nmp)); 823 m_freem(nam); 824 return (error); 825 } 826 827 /* 828 * unmount system call 829 */ 830 int 831 nfs_unmount(struct mount *mp, int mntflags) 832 { 833 struct nfsmount *nmp; 834 struct vnode *vp; 835 int error, flags = 0; 836 837 if (mntflags & MNT_FORCE) 838 flags |= FORCECLOSE; 839 nmp = VFSTONFS(mp); 840 /* 841 * Goes something like this.. 842 * - Check for activity on the root vnode (other than ourselves). 843 * - Call vflush() to clear out vnodes for this file system, 844 * except for the root vnode. 845 * - Decrement reference on the vnode representing remote root. 846 * - Close the socket 847 * - Free up the data structures 848 */ 849 /* 850 * We need to decrement the ref. count on the nfsnode representing 851 * the remote root. See comment in mountnfs(). The VFS unmount() 852 * has done vput on this vnode, otherwise we would get deadlock! 853 */ 854 vp = nmp->nm_vnode; 855 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 856 if (error != 0) 857 return error; 858 859 if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) { 860 vput(vp); 861 return (EBUSY); 862 } 863 864 error = vflush(mp, vp, flags); 865 if (error) { 866 vput(vp); 867 return (error); 868 } 869 870 /* 871 * We are now committed to the unmount; mark the mount structure 872 * as doomed so that any sleepers kicked awake by nfs_disconnect 873 * will go away cleanly. 874 */ 875 nmp->nm_iflag |= NFSMNT_DISMNT; 876 877 /* 878 * Clean up the stats... note that we carefully avoid decrementing 879 * nfs_mount_count here for good reason - we may not be unmounting 880 * the last thing mounted. 881 */ 882 iostat_free(nmp->nm_stats); 883 884 /* 885 * There are two reference counts to get rid of here 886 * (see comment in mountnfs()). 887 */ 888 vput(vp); 889 vgone(vp); 890 nfs_disconnect(nmp); 891 m_freem(nmp->nm_nam); 892 893 rw_destroy(&nmp->nm_writeverflock); 894 mutex_destroy(&nmp->nm_lock); 895 cv_destroy(&nmp->nm_rcvcv); 896 cv_destroy(&nmp->nm_sndcv); 897 cv_destroy(&nmp->nm_aiocv); 898 cv_destroy(&nmp->nm_disconcv); 899 kmem_free(nmp, sizeof(*nmp)); 900 return (0); 901 } 902 903 /* 904 * Return root of a filesystem 905 */ 906 int 907 nfs_root(mp, vpp) 908 struct mount *mp; 909 struct vnode **vpp; 910 { 911 struct vnode *vp; 912 struct nfsmount *nmp; 913 int error; 914 915 nmp = VFSTONFS(mp); 916 vp = nmp->nm_vnode; 917 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 918 if (error != 0) 919 return error; 920 if (vp->v_type == VNON) 921 vp->v_type = VDIR; 922 vp->v_vflag = VV_ROOT; 923 *vpp = vp; 924 return (0); 925 } 926 927 extern int syncprt; 928 929 /* 930 * Flush out the buffer cache 931 */ 932 /* ARGSUSED */ 933 int 934 nfs_sync(mp, waitfor, cred) 935 struct mount *mp; 936 int waitfor; 937 kauth_cred_t cred; 938 { 939 struct vnode *vp, *mvp; 940 int error, allerror = 0; 941 942 /* 943 * Force stale buffer cache information to be flushed. 944 */ 945 if ((mvp = vnalloc(mp)) == NULL) 946 return (ENOMEM); 947 loop: 948 /* 949 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone() 950 * and vclean() can be called indirectly 951 */ 952 mutex_enter(&mntvnode_lock); 953 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) { 954 vmark(mvp, vp); 955 if (vp->v_mount != mp || vismarker(vp)) 956 continue; 957 mutex_enter(&vp->v_interlock); 958 /* XXX MNT_LAZY cannot be right? */ 959 if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) || 960 (LIST_EMPTY(&vp->v_dirtyblkhd) && 961 UVM_OBJ_IS_CLEAN(&vp->v_uobj))) { 962 mutex_exit(&vp->v_interlock); 963 continue; 964 } 965 mutex_exit(&mntvnode_lock); 966 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) { 967 (void)vunmark(mvp); 968 goto loop; 969 } 970 error = VOP_FSYNC(vp, cred, 971 waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0); 972 if (error) 973 allerror = error; 974 vput(vp); 975 mutex_enter(&mntvnode_lock); 976 } 977 mutex_exit(&mntvnode_lock); 978 vnfree(mvp); 979 return (allerror); 980 } 981 982 /* 983 * NFS flat namespace lookup. 984 * Currently unsupported. 985 */ 986 /* ARGSUSED */ 987 int 988 nfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 989 { 990 991 return (EOPNOTSUPP); 992 } 993 994 /* 995 * Do that sysctl thang... 996 */ 997 static int 998 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS) 999 { 1000 struct sysctlnode node; 1001 int val; 1002 int error; 1003 1004 val = nfs_niothreads; 1005 node = *rnode; 1006 node.sysctl_data = &val; 1007 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1008 if (error || newp == NULL) 1009 return error; 1010 1011 return nfs_set_niothreads(val); 1012 } 1013 1014 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup") 1015 { 1016 1017 sysctl_createv(clog, 0, NULL, NULL, 1018 CTLFLAG_PERMANENT, 1019 CTLTYPE_NODE, "vfs", NULL, 1020 NULL, 0, NULL, 0, 1021 CTL_VFS, CTL_EOL); 1022 sysctl_createv(clog, 0, NULL, NULL, 1023 CTLFLAG_PERMANENT, 1024 CTLTYPE_NODE, "nfs", 1025 SYSCTL_DESCR("NFS vfs options"), 1026 NULL, 0, NULL, 0, 1027 CTL_VFS, 2, CTL_EOL); 1028 /* 1029 * XXX the "2" above could be dynamic, thereby eliminating one 1030 * more instance of the "number to vfs" mapping problem, but 1031 * "2" is the order as taken from sys/mount.h 1032 */ 1033 1034 sysctl_createv(clog, 0, NULL, NULL, 1035 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1036 CTLTYPE_STRUCT, "nfsstats", 1037 SYSCTL_DESCR("NFS operation statistics"), 1038 NULL, 0, &nfsstats, sizeof(nfsstats), 1039 CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL); 1040 sysctl_createv(clog, 0, NULL, NULL, 1041 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1042 CTLTYPE_INT, "iothreads", 1043 SYSCTL_DESCR("Number of NFS client processes desired"), 1044 sysctl_vfs_nfs_iothreads, 0, NULL, 0, 1045 CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL); 1046 } 1047 1048 /* ARGSUSED */ 1049 int 1050 nfs_fhtovp(struct mount *mp, struct fid *fid, struct vnode **vpp) 1051 { 1052 size_t fidsize; 1053 size_t fhsize; 1054 struct nfsnode *np; 1055 int error; 1056 struct vattr va; 1057 1058 fidsize = fid->fid_len; 1059 if (fidsize < sizeof(*fid)) { 1060 return EINVAL; 1061 } 1062 fhsize = fidsize - sizeof(*fid); 1063 if ((fhsize % NFSX_UNSIGNED) != 0) { 1064 return EINVAL; 1065 } 1066 if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) { 1067 if (fhsize > NFSX_V3FHMAX || fhsize == 0) { 1068 return EINVAL; 1069 } 1070 } else { 1071 if (fhsize != NFSX_V2FH) { 1072 return EINVAL; 1073 } 1074 } 1075 error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np); 1076 if (error) { 1077 return error; 1078 } 1079 *vpp = NFSTOV(np); 1080 error = VOP_GETATTR(*vpp, &va, kauth_cred_get()); 1081 if (error != 0) { 1082 vput(*vpp); 1083 } 1084 return error; 1085 } 1086 1087 /* ARGSUSED */ 1088 int 1089 nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize) 1090 { 1091 struct nfsnode *np; 1092 struct fid *fid; 1093 size_t fidsize; 1094 int error = 0; 1095 1096 np = VTONFS(vp); 1097 fidsize = sizeof(*fid) + np->n_fhsize; 1098 if (*bufsize < fidsize) { 1099 error = E2BIG; 1100 } 1101 *bufsize = fidsize; 1102 if (error == 0) { 1103 struct fid fid_store; 1104 1105 fid = &fid_store; 1106 memset(fid, 0, sizeof(*fid)); 1107 fid->fid_len = fidsize; 1108 memcpy(buf, fid, sizeof(*fid)); 1109 memcpy(buf->fid_data, np->n_fhp, np->n_fhsize); 1110 } 1111 return error; 1112 } 1113 1114 /* 1115 * Vfs start routine, a no-op. 1116 */ 1117 /* ARGSUSED */ 1118 int 1119 nfs_start(struct mount *mp, int flags) 1120 { 1121 1122 return (0); 1123 } 1124
Indexes created Thu Sep 25 16:09:42 GMT 2025