nfs_vfsops.c revision 1.148.2.11
1 /* $NetBSD: nfs_vfsops.c,v 1.148.2.11 2008/02/27 08:37:05 yamt 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 1.148.2.11 2008/02/27 08:37:05 yamt 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_zalloc(sizeof(*nd), KM_SLEEP); 343 error = nfs_boot_init(nd, l); 344 if (error) { 345 kmem_free(nd, sizeof(*nd)); 346 return (error); 347 } 348 349 /* 350 * Create the root mount point. 351 */ 352 error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l); 353 if (error) 354 goto out; 355 printf("root on %s\n", nd->nd_root.ndm_host); 356 357 /* 358 * Link it into the mount list. 359 */ 360 mutex_enter(&mountlist_lock); 361 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 362 mutex_exit(&mountlist_lock); 363 rootvp = vp; 364 mp->mnt_vnodecovered = NULLVP; 365 vfs_unbusy(mp, false); 366 367 /* Get root attributes (for the time). */ 368 error = VOP_GETATTR(vp, &attr, l->l_cred); 369 if (error) 370 panic("nfs_mountroot: getattr for root"); 371 n = attr.va_atime.tv_sec; 372 #ifdef DEBUG 373 printf("root time: 0x%lx\n", n); 374 #endif 375 setrootfstime(n); 376 377 out: 378 if (error) 379 nfs_boot_cleanup(nd, l); 380 kmem_free(nd, sizeof(*nd)); 381 return (error); 382 } 383 384 /* 385 * Internal version of mount system call for diskless setup. 386 * Separate function because we used to call it twice. 387 * (once for root and once for swap) 388 */ 389 static int 390 nfs_mount_diskless(ndmntp, mntname, mpp, vpp, l) 391 struct nfs_dlmount *ndmntp; 392 const char *mntname; /* mount point name */ 393 struct mount **mpp; 394 struct vnode **vpp; 395 struct lwp *l; 396 { 397 struct mount *mp; 398 struct mbuf *m; 399 int error; 400 401 vfs_rootmountalloc(MOUNT_NFS, mntname, &mp); 402 403 mp->mnt_op = &nfs_vfsops; 404 405 /* 406 * Historical practice expects NFS root file systems to 407 * be initially mounted r/w. 408 */ 409 mp->mnt_flag &= ~MNT_RDONLY; 410 411 /* Get mbuf for server sockaddr. */ 412 m = m_get(M_WAIT, MT_SONAME); 413 if (m == NULL) 414 panic("nfs_mountroot: mget soname for %s", mntname); 415 MCLAIM(m, &nfs_mowner); 416 memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr, 417 (m->m_len = ndmntp->ndm_args.addr->sa_len)); 418 419 error = mountnfs(&ndmntp->ndm_args, mp, m, mntname, 420 ndmntp->ndm_args.hostname, vpp, l); 421 if (error) { 422 vfs_unbusy(mp, false); 423 vfs_destroy(mp); 424 printf("nfs_mountroot: mount %s failed: %d\n", 425 mntname, error); 426 } else 427 *mpp = mp; 428 429 return (error); 430 } 431 432 void 433 nfs_decode_args(nmp, argp, l) 434 struct nfsmount *nmp; 435 struct nfs_args *argp; 436 struct lwp *l; 437 { 438 int adjsock; 439 int maxio; 440 441 /* 442 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 443 * no sense in that context. 444 */ 445 if (argp->sotype == SOCK_STREAM) 446 argp->flags &= ~NFSMNT_NOCONN; 447 448 /* 449 * Cookie translation is not needed for v2, silently ignore it. 450 */ 451 if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) == 452 NFSMNT_XLATECOOKIE) 453 argp->flags &= ~NFSMNT_XLATECOOKIE; 454 455 /* Re-bind if rsrvd port requested and wasn't on one */ 456 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 457 && (argp->flags & NFSMNT_RESVPORT); 458 /* Also re-bind if we're switching to/from a connected UDP socket */ 459 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 460 (argp->flags & NFSMNT_NOCONN)); 461 462 /* Update flags. */ 463 nmp->nm_flag = argp->flags; 464 465 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 466 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 467 if (nmp->nm_timeo < NFS_MINTIMEO) 468 nmp->nm_timeo = NFS_MINTIMEO; 469 else if (nmp->nm_timeo > NFS_MAXTIMEO) 470 nmp->nm_timeo = NFS_MAXTIMEO; 471 } 472 473 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 474 nmp->nm_retry = argp->retrans; 475 if (nmp->nm_retry > NFS_MAXREXMIT) 476 nmp->nm_retry = NFS_MAXREXMIT; 477 } 478 479 #ifndef NFS_V2_ONLY 480 if (argp->flags & NFSMNT_NFSV3) { 481 if (argp->sotype == SOCK_DGRAM) 482 maxio = NFS_MAXDGRAMDATA; 483 else 484 maxio = NFS_MAXDATA; 485 } else 486 #endif 487 maxio = NFS_V2MAXDATA; 488 489 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 490 int osize = nmp->nm_wsize; 491 nmp->nm_wsize = argp->wsize; 492 /* Round down to multiple of blocksize */ 493 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 494 if (nmp->nm_wsize <= 0) 495 nmp->nm_wsize = NFS_FABLKSIZE; 496 adjsock |= (nmp->nm_wsize != osize); 497 } 498 if (nmp->nm_wsize > maxio) 499 nmp->nm_wsize = maxio; 500 if (nmp->nm_wsize > MAXBSIZE) 501 nmp->nm_wsize = MAXBSIZE; 502 503 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 504 int osize = nmp->nm_rsize; 505 nmp->nm_rsize = argp->rsize; 506 /* Round down to multiple of blocksize */ 507 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 508 if (nmp->nm_rsize <= 0) 509 nmp->nm_rsize = NFS_FABLKSIZE; 510 adjsock |= (nmp->nm_rsize != osize); 511 } 512 if (nmp->nm_rsize > maxio) 513 nmp->nm_rsize = maxio; 514 if (nmp->nm_rsize > MAXBSIZE) 515 nmp->nm_rsize = MAXBSIZE; 516 517 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 518 nmp->nm_readdirsize = argp->readdirsize; 519 /* Round down to multiple of minimum blocksize */ 520 nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1); 521 if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ) 522 nmp->nm_readdirsize = NFS_DIRFRAGSIZ; 523 /* Bigger than buffer size makes no sense */ 524 if (nmp->nm_readdirsize > NFS_DIRBLKSIZ) 525 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 526 } else if (argp->flags & NFSMNT_RSIZE) 527 nmp->nm_readdirsize = nmp->nm_rsize; 528 529 if (nmp->nm_readdirsize > maxio) 530 nmp->nm_readdirsize = maxio; 531 532 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && 533 argp->maxgrouplist <= NFS_MAXGRPS) 534 nmp->nm_numgrps = argp->maxgrouplist; 535 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && 536 argp->readahead <= NFS_MAXRAHEAD) 537 nmp->nm_readahead = argp->readahead; 538 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && 539 argp->deadthresh <= NFS_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 kpause("nfscn3", false, hz, NULL); 553 } 554 } 555 } 556 557 /* 558 * VFS Operations. 559 * 560 * mount system call 561 * It seems a bit dumb to copyinstr() the host and path here and then 562 * memcpy() them in mountnfs(), but I wanted to detect errors before 563 * doing the sockargs() call because sockargs() allocates an mbuf and 564 * an error after that means that I have to release the mbuf. 565 */ 566 /* ARGSUSED */ 567 int 568 nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 569 { 570 struct lwp *l = curlwp; 571 int error; 572 struct nfs_args *args = data; 573 struct mbuf *nam; 574 struct nfsmount *nmp = VFSTONFS(mp); 575 struct sockaddr *sa; 576 struct vnode *vp; 577 char *pth, *hst; 578 struct proc *p; 579 size_t len; 580 u_char *nfh; 581 582 if (*data_len < sizeof *args) 583 return EINVAL; 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 = 0; /* dummy */ 613 args->deadthresh = nmp->nm_deadthresh; 614 args->hostname = NULL; 615 *data_len = sizeof *args; 616 return 0; 617 } 618 619 if (args->version != NFS_ARGSVERSION) 620 return (EPROGMISMATCH); 621 if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB)) 622 return (EPROGUNAVAIL); 623 #ifdef NFS_V2_ONLY 624 if (args->flags & NFSMNT_NFSV3) 625 return (EPROGMISMATCH); 626 #endif 627 if (mp->mnt_flag & MNT_UPDATE) { 628 #if 0 629 if (nmp == NULL) 630 return (EIO); 631 /* 632 * When doing an update, we can't change from or to 633 * v3, or change cookie translation 634 */ 635 args->flags = (args->flags & ~(NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)) | 636 (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)); 637 nfs_decode_args(nmp, args, l); 638 return (0); 639 #endif 640 return EOPNOTSUPP; 641 } 642 if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX) 643 return (EINVAL); 644 MALLOC(nfh, u_char *, NFSX_V3FHMAX, M_TEMP, M_WAITOK); 645 error = copyin(args->fh, nfh, args->fhsize); 646 if (error) 647 goto free_nfh; 648 MALLOC(pth, char *, MNAMELEN, M_TEMP, M_WAITOK); 649 error = copyinstr(path, pth, MNAMELEN - 1, &len); 650 if (error) 651 goto free_pth; 652 memset(&pth[len], 0, MNAMELEN - len); 653 MALLOC(hst, char *, MNAMELEN, M_TEMP, M_WAITOK); 654 error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len); 655 if (error) 656 goto free_hst; 657 memset(&hst[len], 0, MNAMELEN - len); 658 /* sockargs() call must be after above copyin() calls */ 659 error = sockargs(&nam, args->addr, args->addrlen, MT_SONAME); 660 if (error) 661 goto free_hst; 662 MCLAIM(nam, &nfs_mowner); 663 args->fh = nfh; 664 error = mountnfs(args, mp, nam, pth, hst, &vp, l); 665 666 free_hst: 667 FREE(hst, M_TEMP); 668 free_pth: 669 FREE(pth, M_TEMP); 670 free_nfh: 671 FREE(nfh, M_TEMP); 672 673 return (error); 674 } 675 676 /* 677 * Common code for mount and mountroot 678 */ 679 int 680 mountnfs(argp, mp, nam, pth, hst, vpp, l) 681 struct nfs_args *argp; 682 struct mount *mp; 683 struct mbuf *nam; 684 const char *pth, *hst; 685 struct vnode **vpp; 686 struct lwp *l; 687 { 688 struct nfsmount *nmp; 689 struct nfsnode *np; 690 int error; 691 struct vattr *attrs; 692 kauth_cred_t cr; 693 char iosname[IOSTATNAMELEN]; 694 695 /* 696 * If the number of nfs iothreads to use has never 697 * been set, create a reasonable number of them. 698 */ 699 700 if (nfs_niothreads < 0) { 701 nfs_set_niothreads(NFS_DEFAULT_NIOTHREADS); 702 } 703 704 if (mp->mnt_flag & MNT_UPDATE) { 705 nmp = VFSTONFS(mp); 706 /* update paths, file handles, etc, here XXX */ 707 m_freem(nam); 708 return (0); 709 } else { 710 nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP); 711 mp->mnt_data = nmp; 712 TAILQ_INIT(&nmp->nm_uidlruhead); 713 TAILQ_INIT(&nmp->nm_bufq); 714 rw_init(&nmp->nm_writeverflock); 715 mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE); 716 cv_init(&nmp->nm_rcvcv, "nfsrcv"); 717 cv_init(&nmp->nm_sndcv, "nfssnd"); 718 cv_init(&nmp->nm_aiocv, "nfsaio"); 719 cv_init(&nmp->nm_disconcv, "nfsdis"); 720 } 721 vfs_getnewfsid(mp); 722 nmp->nm_mountp = mp; 723 724 #ifndef NFS_V2_ONLY 725 if ((argp->flags & NFSMNT_NFSV3) == 0) 726 #endif 727 { 728 if (argp->fhsize != NFSX_V2FH) { 729 return EINVAL; 730 } 731 } 732 733 /* 734 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo 735 * will overwrite this. 736 */ 737 nmp->nm_maxfilesize = 0xffffffffLL; 738 739 nmp->nm_timeo = NFS_TIMEO; 740 nmp->nm_retry = NFS_RETRANS; 741 nmp->nm_wsize = NFS_WSIZE; 742 nmp->nm_rsize = NFS_RSIZE; 743 nmp->nm_readdirsize = NFS_READDIRSIZE; 744 nmp->nm_numgrps = NFS_MAXGRPS; 745 nmp->nm_readahead = NFS_DEFRAHEAD; 746 nmp->nm_deadthresh = NFS_DEFDEADTHRESH; 747 error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, 748 mp->mnt_op->vfs_name, mp, l); 749 if (error) 750 goto bad; 751 nmp->nm_nam = nam; 752 753 /* Set up the sockets and per-host congestion */ 754 nmp->nm_sotype = argp->sotype; 755 nmp->nm_soproto = argp->proto; 756 757 nfs_decode_args(nmp, argp, l); 758 759 mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 760 mp->mnt_dev_bshift = DEV_BSHIFT; 761 762 /* 763 * For Connection based sockets (TCP,...) defer the connect until 764 * the first request, in case the server is not responding. 765 */ 766 if (nmp->nm_sotype == SOCK_DGRAM && 767 (error = nfs_connect(nmp, (struct nfsreq *)0, l))) 768 goto bad; 769 770 /* 771 * This is silly, but it has to be set so that vinifod() works. 772 * We do not want to do an nfs_statvfs() here since we can get 773 * stuck on a dead server and we are holding a lock on the mount 774 * point. 775 */ 776 mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; 777 error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np); 778 if (error) 779 goto bad; 780 *vpp = NFSTOV(np); 781 MALLOC(attrs, struct vattr *, sizeof(struct vattr), M_TEMP, M_WAITOK); 782 VOP_GETATTR(*vpp, attrs, l->l_cred); 783 if ((nmp->nm_flag & NFSMNT_NFSV3) && ((*vpp)->v_type == VDIR)) { 784 cr = kauth_cred_alloc(); 785 kauth_cred_setuid(cr, attrs->va_uid); 786 kauth_cred_seteuid(cr, attrs->va_uid); 787 kauth_cred_setsvuid(cr, attrs->va_uid); 788 kauth_cred_setgid(cr, attrs->va_gid); 789 kauth_cred_setegid(cr, attrs->va_gid); 790 kauth_cred_setsvgid(cr, attrs->va_gid); 791 nfs_cookieheuristic(*vpp, &nmp->nm_iflag, l, cr); 792 kauth_cred_free(cr); 793 } 794 FREE(attrs, M_TEMP); 795 796 /* 797 * A reference count is needed on the nfsnode representing the 798 * remote root. If this object is not persistent, then backward 799 * traversals of the mount point (i.e. "..") will not work if 800 * the nfsnode gets flushed out of the cache. Ufs does not have 801 * this problem, because one can identify root inodes by their 802 * number == ROOTINO (2). So, just unlock, but no rele. 803 */ 804 805 (*vpp)->v_vflag |= VV_ROOT; 806 nmp->nm_vnode = *vpp; 807 VOP_UNLOCK(*vpp, 0); 808 809 snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++); 810 nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname); 811 812 return (0); 813 bad: 814 nfs_disconnect(nmp); 815 rw_destroy(&nmp->nm_writeverflock); 816 mutex_destroy(&nmp->nm_lock); 817 cv_destroy(&nmp->nm_rcvcv); 818 cv_destroy(&nmp->nm_sndcv); 819 cv_destroy(&nmp->nm_aiocv); 820 cv_destroy(&nmp->nm_disconcv); 821 kmem_free(nmp, sizeof(*nmp)); 822 m_freem(nam); 823 return (error); 824 } 825 826 /* 827 * unmount system call 828 */ 829 int 830 nfs_unmount(struct mount *mp, int mntflags) 831 { 832 struct nfsmount *nmp; 833 struct vnode *vp; 834 int error, flags = 0; 835 836 if (mntflags & MNT_FORCE) 837 flags |= FORCECLOSE; 838 nmp = VFSTONFS(mp); 839 /* 840 * Goes something like this.. 841 * - Check for activity on the root vnode (other than ourselves). 842 * - Call vflush() to clear out vnodes for this file system, 843 * except for the root vnode. 844 * - Decrement reference on the vnode representing remote root. 845 * - Close the socket 846 * - Free up the data structures 847 */ 848 /* 849 * We need to decrement the ref. count on the nfsnode representing 850 * the remote root. See comment in mountnfs(). The VFS unmount() 851 * has done vput on this vnode, otherwise we would get deadlock! 852 */ 853 vp = nmp->nm_vnode; 854 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 855 if (error != 0) 856 return error; 857 858 if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) { 859 vput(vp); 860 return (EBUSY); 861 } 862 863 error = vflush(mp, vp, flags); 864 if (error) { 865 vput(vp); 866 return (error); 867 } 868 869 /* 870 * We are now committed to the unmount; mark the mount structure 871 * as doomed so that any sleepers kicked awake by nfs_disconnect 872 * will go away cleanly. 873 */ 874 nmp->nm_iflag |= NFSMNT_DISMNT; 875 876 /* 877 * Clean up the stats... note that we carefully avoid decrementing 878 * nfs_mount_count here for good reason - we may not be unmounting 879 * the last thing mounted. 880 */ 881 iostat_free(nmp->nm_stats); 882 883 /* 884 * There are two reference counts to get rid of here 885 * (see comment in mountnfs()). 886 */ 887 vput(vp); 888 vgone(vp); 889 nfs_disconnect(nmp); 890 m_freem(nmp->nm_nam); 891 892 rw_destroy(&nmp->nm_writeverflock); 893 mutex_destroy(&nmp->nm_lock); 894 cv_destroy(&nmp->nm_rcvcv); 895 cv_destroy(&nmp->nm_sndcv); 896 cv_destroy(&nmp->nm_aiocv); 897 cv_destroy(&nmp->nm_disconcv); 898 kmem_free(nmp, sizeof(*nmp)); 899 return (0); 900 } 901 902 /* 903 * Return root of a filesystem 904 */ 905 int 906 nfs_root(mp, vpp) 907 struct mount *mp; 908 struct vnode **vpp; 909 { 910 struct vnode *vp; 911 struct nfsmount *nmp; 912 int error; 913 914 nmp = VFSTONFS(mp); 915 vp = nmp->nm_vnode; 916 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 917 if (error != 0) 918 return error; 919 if (vp->v_type == VNON) 920 vp->v_type = VDIR; 921 *vpp = vp; 922 return (0); 923 } 924 925 extern int syncprt; 926 927 /* 928 * Flush out the buffer cache 929 */ 930 /* ARGSUSED */ 931 int 932 nfs_sync(mp, waitfor, cred) 933 struct mount *mp; 934 int waitfor; 935 kauth_cred_t cred; 936 { 937 struct vnode *vp, *mvp; 938 int error, allerror = 0; 939 940 /* 941 * Force stale buffer cache information to be flushed. 942 */ 943 if ((mvp = vnalloc(mp)) == NULL) 944 return (ENOMEM); 945 loop: 946 /* 947 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone() 948 * and vclean() can be called indirectly 949 */ 950 mutex_enter(&mntvnode_lock); 951 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) { 952 vmark(mvp, vp); 953 if (vp->v_mount != mp || vismarker(vp)) 954 continue; 955 mutex_enter(&vp->v_interlock); 956 /* XXX MNT_LAZY cannot be right? */ 957 if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) || 958 (LIST_EMPTY(&vp->v_dirtyblkhd) && 959 UVM_OBJ_IS_CLEAN(&vp->v_uobj))) { 960 mutex_exit(&vp->v_interlock); 961 continue; 962 } 963 mutex_exit(&mntvnode_lock); 964 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) { 965 (void)vunmark(mvp); 966 goto loop; 967 } 968 error = VOP_FSYNC(vp, cred, 969 waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0); 970 if (error) 971 allerror = error; 972 vput(vp); 973 mutex_enter(&mntvnode_lock); 974 } 975 mutex_exit(&mntvnode_lock); 976 vnfree(mvp); 977 return (allerror); 978 } 979 980 /* 981 * NFS flat namespace lookup. 982 * Currently unsupported. 983 */ 984 /* ARGSUSED */ 985 int 986 nfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 987 { 988 989 return (EOPNOTSUPP); 990 } 991 992 /* 993 * Do that sysctl thang... 994 */ 995 static int 996 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS) 997 { 998 struct sysctlnode node; 999 int val; 1000 int error; 1001 1002 val = nfs_niothreads; 1003 node = *rnode; 1004 node.sysctl_data = &val; 1005 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1006 if (error || newp == NULL) 1007 return error; 1008 1009 return nfs_set_niothreads(val); 1010 } 1011 1012 SYSCTL_SETUP(sysctl_vfs_nfs_setup, "sysctl vfs.nfs subtree setup") 1013 { 1014 1015 sysctl_createv(clog, 0, NULL, NULL, 1016 CTLFLAG_PERMANENT, 1017 CTLTYPE_NODE, "vfs", NULL, 1018 NULL, 0, NULL, 0, 1019 CTL_VFS, CTL_EOL); 1020 sysctl_createv(clog, 0, NULL, NULL, 1021 CTLFLAG_PERMANENT, 1022 CTLTYPE_NODE, "nfs", 1023 SYSCTL_DESCR("NFS vfs options"), 1024 NULL, 0, NULL, 0, 1025 CTL_VFS, 2, CTL_EOL); 1026 /* 1027 * XXX the "2" above could be dynamic, thereby eliminating one 1028 * more instance of the "number to vfs" mapping problem, but 1029 * "2" is the order as taken from sys/mount.h 1030 */ 1031 1032 sysctl_createv(clog, 0, NULL, NULL, 1033 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1034 CTLTYPE_STRUCT, "nfsstats", 1035 SYSCTL_DESCR("NFS operation statistics"), 1036 NULL, 0, &nfsstats, sizeof(nfsstats), 1037 CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL); 1038 sysctl_createv(clog, 0, NULL, NULL, 1039 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1040 CTLTYPE_INT, "iothreads", 1041 SYSCTL_DESCR("Number of NFS client processes desired"), 1042 sysctl_vfs_nfs_iothreads, 0, NULL, 0, 1043 CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL); 1044 } 1045 1046 /* ARGSUSED */ 1047 int 1048 nfs_fhtovp(struct mount *mp, struct fid *fid, struct vnode **vpp) 1049 { 1050 size_t fidsize; 1051 size_t fhsize; 1052 struct nfsnode *np; 1053 int error; 1054 struct vattr va; 1055 1056 fidsize = fid->fid_len; 1057 if (fidsize < sizeof(*fid)) { 1058 return EINVAL; 1059 } 1060 fhsize = fidsize - sizeof(*fid); 1061 if ((fhsize % NFSX_UNSIGNED) != 0) { 1062 return EINVAL; 1063 } 1064 if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) { 1065 if (fhsize > NFSX_V3FHMAX || fhsize == 0) { 1066 return EINVAL; 1067 } 1068 } else { 1069 if (fhsize != NFSX_V2FH) { 1070 return EINVAL; 1071 } 1072 } 1073 error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np); 1074 if (error) { 1075 return error; 1076 } 1077 *vpp = NFSTOV(np); 1078 error = VOP_GETATTR(*vpp, &va, kauth_cred_get()); 1079 if (error != 0) { 1080 vput(*vpp); 1081 } 1082 return error; 1083 } 1084 1085 /* ARGSUSED */ 1086 int 1087 nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize) 1088 { 1089 struct nfsnode *np; 1090 struct fid *fid; 1091 size_t fidsize; 1092 int error = 0; 1093 1094 np = VTONFS(vp); 1095 fidsize = sizeof(*fid) + np->n_fhsize; 1096 if (*bufsize < fidsize) { 1097 error = E2BIG; 1098 } 1099 *bufsize = fidsize; 1100 if (error == 0) { 1101 struct fid fid_store; 1102 1103 fid = &fid_store; 1104 memset(fid, 0, sizeof(*fid)); 1105 fid->fid_len = fidsize; 1106 memcpy(buf, fid, sizeof(*fid)); 1107 memcpy(buf->fid_data, np->n_fhp, np->n_fhsize); 1108 } 1109 return error; 1110 } 1111 1112 /* 1113 * Vfs start routine, a no-op. 1114 */ 1115 /* ARGSUSED */ 1116 int 1117 nfs_start(struct mount *mp, int flags) 1118 { 1119 1120 return (0); 1121 } 1122
Indexes created Wed Oct 01 07:09:59 GMT 2025