1 /* $NetBSD: nfs_vfsops.c,v 1.217 2010/08/12 09:41:10 pooka 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 1.217 2010/08/12 09:41:10 pooka Exp $"); 39 40 #if defined(_KERNEL_OPT) 41 #include "opt_nfs.h" 42 #endif 43 44 #include <sys/param.h> 45 #include <sys/ioctl.h> 46 #include <sys/signal.h> 47 #include <sys/proc.h> 48 #include <sys/namei.h> 49 #include <sys/device.h> 50 #include <sys/vnode.h> 51 #include <sys/kernel.h> 52 #include <sys/mount.h> 53 #include <sys/buf.h> 54 #include <sys/mbuf.h> 55 #include <sys/dirent.h> 56 #include <sys/socket.h> 57 #include <sys/socketvar.h> 58 #include <sys/sysctl.h> 59 #include <sys/systm.h> 60 #include <sys/timetc.h> 61 #include <sys/kauth.h> 62 #include <sys/module.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 MODULE(MODULE_CLASS_VFS, nfs, NULL); 79 80 extern struct nfsstats nfsstats; 81 extern int nfs_ticks; 82 83 /* 84 * keep a count of the nfs mounts to generate ficticious drive names 85 * for the per drive stats. 86 */ 87 unsigned int nfs_mount_count = 0; 88 89 int nfs_commitsize; 90 91 /* 92 * nfs vfs operations. 93 */ 94 95 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc; 96 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc; 97 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc; 98 99 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = { 100 &nfsv2_vnodeop_opv_desc, 101 &spec_nfsv2nodeop_opv_desc, 102 &fifo_nfsv2nodeop_opv_desc, 103 NULL, 104 }; 105 106 struct vfsops nfs_vfsops = { 107 MOUNT_NFS, 108 sizeof (struct nfs_args), 109 nfs_mount, 110 nfs_start, 111 nfs_unmount, 112 nfs_root, 113 (void *)eopnotsupp, /* vfs_quotactl */ 114 nfs_statvfs, 115 nfs_sync, 116 nfs_vget, 117 nfs_fhtovp, 118 nfs_vptofh, 119 nfs_vfs_init, 120 NULL, 121 nfs_vfs_done, 122 nfs_mountroot, 123 (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp, 124 vfs_stdextattrctl, 125 (void *)eopnotsupp, /* vfs_suspendctl */ 126 genfs_renamelock_enter, 127 genfs_renamelock_exit, 128 (void *)eopnotsupp, 129 nfs_vnodeopv_descs, 130 0, 131 { NULL, NULL }, 132 }; 133 134 extern u_int32_t nfs_procids[NFS_NPROCS]; 135 extern u_int32_t nfs_prog, nfs_vers; 136 static struct sysctllog *nfs_clog; 137 138 static int nfs_mount_diskless(struct nfs_dlmount *, const char *, 139 struct mount **, struct vnode **, struct lwp *); 140 static void nfs_sysctl_init(void); 141 static void nfs_sysctl_fini(void); 142 143 static int 144 nfs_modcmd(modcmd_t cmd, void *arg) 145 { 146 int error; 147 148 switch (cmd) { 149 case MODULE_CMD_INIT: 150 error = vfs_attach(&nfs_vfsops); 151 if (error == 0) { 152 nfs_sysctl_init(); 153 } 154 return error; 155 case MODULE_CMD_FINI: 156 error = vfs_detach(&nfs_vfsops); 157 if (error == 0) { 158 nfs_sysctl_fini(); 159 } 160 return error; 161 default: 162 return ENOTTY; 163 } 164 } 165 166 /* 167 * nfs statvfs call 168 */ 169 int 170 nfs_statvfs(struct mount *mp, struct statvfs *sbp) 171 { 172 struct lwp *l = curlwp; 173 struct vnode *vp; 174 struct nfs_statfs *sfp; 175 char *cp; 176 u_int32_t *tl; 177 int32_t t1, t2; 178 char *bpos, *dpos, *cp2; 179 struct nfsmount *nmp = VFSTONFS(mp); 180 int error = 0, retattr; 181 #ifdef NFS_V2_ONLY 182 const int v3 = 0; 183 #else 184 int v3 = (nmp->nm_flag & NFSMNT_NFSV3); 185 #endif 186 struct mbuf *mreq, *mrep = NULL, *md, *mb; 187 kauth_cred_t cred; 188 u_quad_t tquad; 189 struct nfsnode *np; 190 191 #ifndef nolint 192 sfp = (struct nfs_statfs *)0; 193 #endif 194 vp = nmp->nm_vnode; 195 np = VTONFS(vp); 196 cred = kauth_cred_alloc(); 197 #ifndef NFS_V2_ONLY 198 if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 199 (void)nfs_fsinfo(nmp, vp, cred, l); 200 #endif 201 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 202 nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3)); 203 nfsm_fhtom(np, v3); 204 nfsm_request(np, NFSPROC_FSSTAT, l, cred); 205 if (v3) 206 nfsm_postop_attr(vp, retattr, 0); 207 if (error) { 208 if (mrep != NULL) { 209 if (mrep->m_next != NULL) 210 printf("nfs_vfsops: nfs_statvfs would lose buffers\n"); 211 m_freem(mrep); 212 } 213 goto nfsmout; 214 } 215 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 216 sbp->f_flag = nmp->nm_flag; 217 sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize); 218 if (v3) { 219 sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE; 220 tquad = fxdr_hyper(&sfp->sf_tbytes); 221 sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 222 tquad = fxdr_hyper(&sfp->sf_fbytes); 223 sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 224 tquad = fxdr_hyper(&sfp->sf_abytes); 225 tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 226 sbp->f_bresvd = sbp->f_bfree - tquad; 227 sbp->f_bavail = tquad; 228 /* Handle older NFS servers returning negative values */ 229 if ((quad_t)sbp->f_bavail < 0) 230 sbp->f_bavail = 0; 231 tquad = fxdr_hyper(&sfp->sf_tfiles); 232 sbp->f_files = tquad; 233 tquad = fxdr_hyper(&sfp->sf_ffiles); 234 sbp->f_ffree = tquad; 235 sbp->f_favail = tquad; 236 sbp->f_fresvd = 0; 237 sbp->f_namemax = NFS_MAXNAMLEN; 238 } else { 239 sbp->f_bsize = NFS_FABLKSIZE; 240 sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 241 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 242 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 243 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 244 sbp->f_fresvd = 0; 245 sbp->f_files = 0; 246 sbp->f_ffree = 0; 247 sbp->f_favail = 0; 248 sbp->f_fresvd = 0; 249 sbp->f_namemax = NFS_MAXNAMLEN; 250 } 251 copy_statvfs_info(sbp, mp); 252 nfsm_reqdone; 253 kauth_cred_free(cred); 254 return (error); 255 } 256 257 #ifndef NFS_V2_ONLY 258 /* 259 * nfs version 3 fsinfo rpc call 260 */ 261 int 262 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, kauth_cred_t cred, struct lwp *l) 263 { 264 struct nfsv3_fsinfo *fsp; 265 char *cp; 266 int32_t t1, t2; 267 u_int32_t *tl, pref, xmax; 268 char *bpos, *dpos, *cp2; 269 int error = 0, retattr; 270 struct mbuf *mreq, *mrep, *md, *mb; 271 u_int64_t maxfsize; 272 struct nfsnode *np = VTONFS(vp); 273 274 nfsstats.rpccnt[NFSPROC_FSINFO]++; 275 nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1)); 276 nfsm_fhtom(np, 1); 277 nfsm_request(np, NFSPROC_FSINFO, l, cred); 278 nfsm_postop_attr(vp, retattr, 0); 279 if (!error) { 280 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 281 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 282 if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 && 283 pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 284 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 285 ~(NFS_FABLKSIZE - 1); 286 xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 287 if (xmax < nmp->nm_wsize && xmax > 0) { 288 nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1); 289 if (nmp->nm_wsize == 0) 290 nmp->nm_wsize = xmax; 291 } 292 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 293 if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 && 294 pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 295 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 296 ~(NFS_FABLKSIZE - 1); 297 xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 298 if (xmax < nmp->nm_rsize && xmax > 0) { 299 nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1); 300 if (nmp->nm_rsize == 0) 301 nmp->nm_rsize = xmax; 302 } 303 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 304 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ) 305 nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) & 306 ~(NFS_DIRFRAGSIZ - 1); 307 if (xmax < nmp->nm_readdirsize && xmax > 0) { 308 nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1); 309 if (nmp->nm_readdirsize == 0) 310 nmp->nm_readdirsize = xmax; 311 } 312 /* XXX */ 313 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 314 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 315 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 316 nmp->nm_maxfilesize = maxfsize; 317 nmp->nm_mountp->mnt_fs_bshift = 318 ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 319 nmp->nm_iflag |= NFSMNT_GOTFSINFO; 320 } 321 nfsm_reqdone; 322 return (error); 323 } 324 #endif 325 326 /* 327 * Mount a remote root fs via. NFS. It goes like this: 328 * - Call nfs_boot_init() to fill in the nfs_diskless struct 329 * - build the rootfs mount point and call mountnfs() to do the rest. 330 */ 331 int 332 nfs_mountroot(void) 333 { 334 struct timespec ts; 335 struct nfs_diskless *nd; 336 struct vattr attr; 337 struct mount *mp; 338 struct vnode *vp; 339 struct lwp *l; 340 long n; 341 int error; 342 343 l = curlwp; /* XXX */ 344 345 if (device_class(root_device) != DV_IFNET) 346 return (ENODEV); 347 348 /* 349 * XXX time must be non-zero when we init the interface or else 350 * the arp code will wedge. [Fixed now in if_ether.c] 351 * However, the NFS attribute cache gives false "hits" when the 352 * current time < nfs_attrtimeo(nmp, np) so keep this in for now. 353 */ 354 if (time_second < NFS_MAXATTRTIMO) { 355 ts.tv_sec = NFS_MAXATTRTIMO; 356 ts.tv_nsec = 0; 357 tc_setclock(&ts); 358 } 359 360 /* 361 * Call nfs_boot_init() to fill in the nfs_diskless struct. 362 * Side effect: Finds and configures a network interface. 363 */ 364 nd = kmem_zalloc(sizeof(*nd), KM_SLEEP); 365 error = nfs_boot_init(nd, l); 366 if (error) { 367 kmem_free(nd, sizeof(*nd)); 368 return (error); 369 } 370 371 /* 372 * Create the root mount point. 373 */ 374 error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l); 375 if (error) 376 goto out; 377 printf("root on %s\n", nd->nd_root.ndm_host); 378 379 /* 380 * Link it into the mount list. 381 */ 382 mutex_enter(&mountlist_lock); 383 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 384 mutex_exit(&mountlist_lock); 385 rootvp = vp; 386 mp->mnt_vnodecovered = NULLVP; 387 vfs_unbusy(mp, false, NULL); 388 389 /* Get root attributes (for the time). */ 390 error = VOP_GETATTR(vp, &attr, l->l_cred); 391 if (error) 392 panic("nfs_mountroot: getattr for root"); 393 n = attr.va_atime.tv_sec; 394 #ifdef DEBUG 395 printf("root time: 0x%lx\n", n); 396 #endif 397 setrootfstime(n); 398 399 out: 400 if (error) 401 nfs_boot_cleanup(nd, l); 402 kmem_free(nd, sizeof(*nd)); 403 return (error); 404 } 405 406 /* 407 * Internal version of mount system call for diskless setup. 408 * Separate function because we used to call it twice. 409 * (once for root and once for swap) 410 */ 411 static int 412 nfs_mount_diskless(struct nfs_dlmount *ndmntp, const char *mntname, struct mount **mpp, struct vnode **vpp, struct lwp *l) 413 /* mntname: mount point name */ 414 { 415 struct mount *mp; 416 struct mbuf *m; 417 int error; 418 419 vfs_rootmountalloc(MOUNT_NFS, mntname, &mp); 420 421 mp->mnt_op = &nfs_vfsops; 422 423 /* 424 * Historical practice expects NFS root file systems to 425 * be initially mounted r/w. 426 */ 427 mp->mnt_flag &= ~MNT_RDONLY; 428 429 /* Get mbuf for server sockaddr. */ 430 m = m_get(M_WAIT, MT_SONAME); 431 if (m == NULL) 432 panic("nfs_mountroot: mget soname for %s", mntname); 433 MCLAIM(m, &nfs_mowner); 434 memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr, 435 (m->m_len = ndmntp->ndm_args.addr->sa_len)); 436 437 error = mountnfs(&ndmntp->ndm_args, mp, m, mntname, 438 ndmntp->ndm_args.hostname, vpp, l); 439 if (error) { 440 vfs_unbusy(mp, false, NULL); 441 vfs_destroy(mp); 442 printf("nfs_mountroot: mount %s failed: %d\n", 443 mntname, error); 444 } else 445 *mpp = mp; 446 447 return (error); 448 } 449 450 void 451 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp, struct lwp *l) 452 { 453 int s; 454 int adjsock; 455 int maxio; 456 457 s = splsoftnet(); 458 459 /* 460 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 461 * no sense in that context. 462 */ 463 if (argp->sotype == SOCK_STREAM) 464 argp->flags &= ~NFSMNT_NOCONN; 465 466 /* 467 * Cookie translation is not needed for v2, silently ignore it. 468 */ 469 if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) == 470 NFSMNT_XLATECOOKIE) 471 argp->flags &= ~NFSMNT_XLATECOOKIE; 472 473 /* Re-bind if rsrvd port requested and wasn't on one */ 474 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 475 && (argp->flags & NFSMNT_RESVPORT); 476 /* Also re-bind if we're switching to/from a connected UDP socket */ 477 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 478 (argp->flags & NFSMNT_NOCONN)); 479 480 /* Update flags. */ 481 nmp->nm_flag = argp->flags; 482 splx(s); 483 484 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 485 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 486 if (nmp->nm_timeo < NFS_MINTIMEO) 487 nmp->nm_timeo = NFS_MINTIMEO; 488 else if (nmp->nm_timeo > NFS_MAXTIMEO) 489 nmp->nm_timeo = NFS_MAXTIMEO; 490 } 491 492 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 493 nmp->nm_retry = argp->retrans; 494 if (nmp->nm_retry > NFS_MAXREXMIT) 495 nmp->nm_retry = NFS_MAXREXMIT; 496 } 497 498 #ifndef NFS_V2_ONLY 499 if (argp->flags & NFSMNT_NFSV3) { 500 if (argp->sotype == SOCK_DGRAM) 501 maxio = NFS_MAXDGRAMDATA; 502 else 503 maxio = NFS_MAXDATA; 504 } else 505 #endif 506 maxio = NFS_V2MAXDATA; 507 508 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 509 int osize = nmp->nm_wsize; 510 nmp->nm_wsize = argp->wsize; 511 /* Round down to multiple of blocksize */ 512 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 513 if (nmp->nm_wsize <= 0) 514 nmp->nm_wsize = NFS_FABLKSIZE; 515 adjsock |= (nmp->nm_wsize != osize); 516 } 517 if (nmp->nm_wsize > maxio) 518 nmp->nm_wsize = maxio; 519 if (nmp->nm_wsize > MAXBSIZE) 520 nmp->nm_wsize = MAXBSIZE; 521 522 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 523 int osize = nmp->nm_rsize; 524 nmp->nm_rsize = argp->rsize; 525 /* Round down to multiple of blocksize */ 526 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 527 if (nmp->nm_rsize <= 0) 528 nmp->nm_rsize = NFS_FABLKSIZE; 529 adjsock |= (nmp->nm_rsize != osize); 530 } 531 if (nmp->nm_rsize > maxio) 532 nmp->nm_rsize = maxio; 533 if (nmp->nm_rsize > MAXBSIZE) 534 nmp->nm_rsize = MAXBSIZE; 535 536 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 537 nmp->nm_readdirsize = argp->readdirsize; 538 /* Round down to multiple of minimum blocksize */ 539 nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1); 540 if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ) 541 nmp->nm_readdirsize = NFS_DIRFRAGSIZ; 542 /* Bigger than buffer size makes no sense */ 543 if (nmp->nm_readdirsize > NFS_DIRBLKSIZ) 544 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 545 } else if (argp->flags & NFSMNT_RSIZE) 546 nmp->nm_readdirsize = nmp->nm_rsize; 547 548 if (nmp->nm_readdirsize > maxio) 549 nmp->nm_readdirsize = maxio; 550 551 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && 552 argp->maxgrouplist <= NFS_MAXGRPS) 553 nmp->nm_numgrps = argp->maxgrouplist; 554 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && 555 argp->readahead <= NFS_MAXRAHEAD) 556 nmp->nm_readahead = argp->readahead; 557 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && 558 argp->deadthresh <= NFS_NEVERDEAD) 559 nmp->nm_deadthresh = argp->deadthresh; 560 561 adjsock |= ((nmp->nm_sotype != argp->sotype) || 562 (nmp->nm_soproto != argp->proto)); 563 nmp->nm_sotype = argp->sotype; 564 nmp->nm_soproto = argp->proto; 565 566 if (nmp->nm_so && adjsock) { 567 nfs_safedisconnect(nmp); 568 if (nmp->nm_sotype == SOCK_DGRAM) 569 while (nfs_connect(nmp, (struct nfsreq *)0, l)) { 570 printf("nfs_args: retrying connect\n"); 571 kpause("nfscn3", false, hz, NULL); 572 } 573 } 574 } 575 576 /* 577 * VFS Operations. 578 * 579 * mount system call 580 * It seems a bit dumb to copyinstr() the host and path here and then 581 * memcpy() them in mountnfs(), but I wanted to detect errors before 582 * doing the sockargs() call because sockargs() allocates an mbuf and 583 * an error after that means that I have to release the mbuf. 584 */ 585 /* ARGSUSED */ 586 int 587 nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 588 { 589 struct lwp *l = curlwp; 590 int error; 591 struct nfs_args *args = data; 592 struct mbuf *nam; 593 struct nfsmount *nmp = VFSTONFS(mp); 594 struct sockaddr *sa; 595 struct vnode *vp; 596 char *pth, *hst; 597 struct proc *p; 598 size_t len; 599 u_char *nfh; 600 601 if (*data_len < sizeof *args) 602 return EINVAL; 603 604 p = l->l_proc; 605 if (mp->mnt_flag & MNT_GETARGS) { 606 607 if (nmp == NULL) 608 return (EIO); 609 if (args->addr != NULL) { 610 sa = mtod(nmp->nm_nam, struct sockaddr *); 611 error = copyout(sa, args->addr, sa->sa_len); 612 if (error) 613 return (error); 614 args->addrlen = sa->sa_len; 615 } else 616 args->addrlen = 0; 617 618 args->version = NFS_ARGSVERSION; 619 args->sotype = nmp->nm_sotype; 620 args->proto = nmp->nm_soproto; 621 args->fh = NULL; 622 args->fhsize = 0; 623 args->flags = nmp->nm_flag; 624 args->wsize = nmp->nm_wsize; 625 args->rsize = nmp->nm_rsize; 626 args->readdirsize = nmp->nm_readdirsize; 627 args->timeo = nmp->nm_timeo; 628 args->retrans = nmp->nm_retry; 629 args->maxgrouplist = nmp->nm_numgrps; 630 args->readahead = nmp->nm_readahead; 631 args->leaseterm = 0; /* dummy */ 632 args->deadthresh = nmp->nm_deadthresh; 633 args->hostname = NULL; 634 *data_len = sizeof *args; 635 return 0; 636 } 637 638 if (args->version != NFS_ARGSVERSION) 639 return (EPROGMISMATCH); 640 if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB)) 641 return (EPROGUNAVAIL); 642 #ifdef NFS_V2_ONLY 643 if (args->flags & NFSMNT_NFSV3) 644 return (EPROGMISMATCH); 645 #endif 646 if (mp->mnt_flag & MNT_UPDATE) { 647 if (nmp == NULL) 648 return (EIO); 649 /* 650 * When doing an update, we can't change from or to 651 * v3, or change cookie translation 652 */ 653 args->flags = (args->flags & ~(NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)) | 654 (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)); 655 nfs_decode_args(nmp, args, l); 656 return (0); 657 } 658 if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX) 659 return (EINVAL); 660 nfh = malloc(NFSX_V3FHMAX, M_TEMP, M_WAITOK); 661 error = copyin(args->fh, nfh, args->fhsize); 662 if (error) 663 goto free_nfh; 664 pth = malloc(MNAMELEN, M_TEMP, M_WAITOK); 665 error = copyinstr(path, pth, MNAMELEN - 1, &len); 666 if (error) 667 goto free_pth; 668 memset(&pth[len], 0, MNAMELEN - len); 669 hst = malloc(MNAMELEN, M_TEMP, M_WAITOK); 670 error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len); 671 if (error) 672 goto free_hst; 673 memset(&hst[len], 0, MNAMELEN - len); 674 /* sockargs() call must be after above copyin() calls */ 675 error = sockargs(&nam, args->addr, args->addrlen, MT_SONAME); 676 if (error) 677 goto free_hst; 678 MCLAIM(nam, &nfs_mowner); 679 args->fh = nfh; 680 error = mountnfs(args, mp, nam, pth, hst, &vp, l); 681 682 free_hst: 683 free(hst, M_TEMP); 684 free_pth: 685 free(pth, M_TEMP); 686 free_nfh: 687 free(nfh, M_TEMP); 688 689 return (error); 690 } 691 692 /* 693 * Common code for mount and mountroot 694 */ 695 int 696 mountnfs(struct nfs_args *argp, struct mount *mp, struct mbuf *nam, const char *pth, const char *hst, struct vnode **vpp, struct lwp *l) 697 { 698 struct nfsmount *nmp; 699 struct nfsnode *np; 700 struct vnode *vp; 701 int error; 702 struct vattr *attrs; 703 kauth_cred_t cr; 704 char iosname[IOSTATNAMELEN]; 705 706 /* 707 * If the number of nfs iothreads to use has never 708 * been set, create a reasonable number of them. 709 */ 710 711 if (nfs_niothreads < 0) { 712 nfs_set_niothreads(NFS_DEFAULT_NIOTHREADS); 713 } 714 715 if (mp->mnt_flag & MNT_UPDATE) { 716 nmp = VFSTONFS(mp); 717 /* update paths, file handles, etc, here XXX */ 718 m_freem(nam); 719 return (0); 720 } else { 721 nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP); 722 mp->mnt_data = nmp; 723 TAILQ_INIT(&nmp->nm_uidlruhead); 724 TAILQ_INIT(&nmp->nm_bufq); 725 rw_init(&nmp->nm_writeverflock); 726 mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE); 727 rw_init(&nmp->nm_rbtlock); 728 cv_init(&nmp->nm_rcvcv, "nfsrcv"); 729 cv_init(&nmp->nm_sndcv, "nfssnd"); 730 cv_init(&nmp->nm_aiocv, "nfsaio"); 731 cv_init(&nmp->nm_disconcv, "nfsdis"); 732 nfs_rbtinit(nmp); 733 } 734 vfs_getnewfsid(mp); 735 nmp->nm_mountp = mp; 736 737 #ifndef NFS_V2_ONLY 738 if ((argp->flags & NFSMNT_NFSV3) == 0) 739 #endif 740 { 741 if (argp->fhsize != NFSX_V2FH) { 742 return EINVAL; 743 } 744 } 745 746 /* 747 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo 748 * will overwrite this. 749 */ 750 nmp->nm_maxfilesize = 0xffffffffLL; 751 752 nmp->nm_timeo = NFS_TIMEO; 753 nmp->nm_retry = NFS_RETRANS; 754 nmp->nm_wsize = NFS_WSIZE; 755 nmp->nm_rsize = NFS_RSIZE; 756 nmp->nm_readdirsize = NFS_READDIRSIZE; 757 nmp->nm_numgrps = NFS_MAXGRPS; 758 nmp->nm_readahead = NFS_DEFRAHEAD; 759 nmp->nm_deadthresh = NFS_DEFDEADTHRESH; 760 error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, 761 mp->mnt_op->vfs_name, mp, l); 762 if (error) 763 goto bad; 764 nmp->nm_nam = nam; 765 766 /* Set up the sockets and per-host congestion */ 767 nmp->nm_sotype = argp->sotype; 768 nmp->nm_soproto = argp->proto; 769 770 nfs_decode_args(nmp, argp, l); 771 772 mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 773 mp->mnt_dev_bshift = DEV_BSHIFT; 774 775 /* 776 * For Connection based sockets (TCP,...) defer the connect until 777 * the first request, in case the server is not responding. 778 */ 779 if (nmp->nm_sotype == SOCK_DGRAM && 780 (error = nfs_connect(nmp, (struct nfsreq *)0, l))) 781 goto bad; 782 783 /* 784 * This is silly, but it has to be set so that vinifod() works. 785 * We do not want to do an nfs_statvfs() here since we can get 786 * stuck on a dead server and we are holding a lock on the mount 787 * point. 788 */ 789 mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; 790 error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np); 791 if (error) 792 goto bad; 793 vp = NFSTOV(np); 794 attrs = malloc(sizeof(struct vattr), M_TEMP, M_WAITOK); 795 VOP_GETATTR(vp, attrs, l->l_cred); 796 if ((nmp->nm_flag & NFSMNT_NFSV3) && (vp->v_type == VDIR)) { 797 cr = kauth_cred_alloc(); 798 kauth_cred_setuid(cr, attrs->va_uid); 799 kauth_cred_seteuid(cr, attrs->va_uid); 800 kauth_cred_setsvuid(cr, attrs->va_uid); 801 kauth_cred_setgid(cr, attrs->va_gid); 802 kauth_cred_setegid(cr, attrs->va_gid); 803 kauth_cred_setsvgid(cr, attrs->va_gid); 804 nfs_cookieheuristic(vp, &nmp->nm_iflag, l, cr); 805 kauth_cred_free(cr); 806 } 807 free(attrs, M_TEMP); 808 809 /* 810 * A reference count is needed on the nfsnode representing the 811 * remote root. If this object is not persistent, then backward 812 * traversals of the mount point (i.e. "..") will not work if 813 * the nfsnode gets flushed out of the cache. Ufs does not have 814 * this problem, because one can identify root inodes by their 815 * number == ROOTINO (2). So, just unlock, but no rele. 816 */ 817 818 nmp->nm_vnode = vp; 819 if (vp->v_type == VNON) 820 vp->v_type = VDIR; 821 vp->v_vflag |= VV_ROOT; 822 VOP_UNLOCK(vp); 823 *vpp = vp; 824 825 snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++); 826 nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname); 827 828 return (0); 829 bad: 830 nfs_disconnect(nmp); 831 rw_destroy(&nmp->nm_writeverflock); 832 rw_destroy(&nmp->nm_rbtlock); 833 mutex_destroy(&nmp->nm_lock); 834 cv_destroy(&nmp->nm_rcvcv); 835 cv_destroy(&nmp->nm_sndcv); 836 cv_destroy(&nmp->nm_aiocv); 837 cv_destroy(&nmp->nm_disconcv); 838 kmem_free(nmp, sizeof(*nmp)); 839 m_freem(nam); 840 return (error); 841 } 842 843 /* 844 * unmount system call 845 */ 846 int 847 nfs_unmount(struct mount *mp, int mntflags) 848 { 849 struct nfsmount *nmp; 850 struct vnode *vp; 851 int error, flags = 0; 852 853 if (mntflags & MNT_FORCE) 854 flags |= FORCECLOSE; 855 nmp = VFSTONFS(mp); 856 /* 857 * Goes something like this.. 858 * - Check for activity on the root vnode (other than ourselves). 859 * - Call vflush() to clear out vnodes for this file system, 860 * except for the root vnode. 861 * - Decrement reference on the vnode representing remote root. 862 * - Close the socket 863 * - Free up the data structures 864 */ 865 /* 866 * We need to decrement the ref. count on the nfsnode representing 867 * the remote root. See comment in mountnfs(). 868 */ 869 vp = nmp->nm_vnode; 870 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 871 if (error != 0) 872 return error; 873 874 if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 1) { 875 VOP_UNLOCK(vp); 876 return (EBUSY); 877 } 878 879 error = vflush(mp, vp, flags); 880 if (error) { 881 VOP_UNLOCK(vp); 882 return (error); 883 } 884 885 /* 886 * We are now committed to the unmount; mark the mount structure 887 * as doomed so that any sleepers kicked awake by nfs_disconnect 888 * will go away cleanly. 889 */ 890 nmp->nm_iflag |= NFSMNT_DISMNT; 891 892 /* 893 * Clean up the stats... note that we carefully avoid decrementing 894 * nfs_mount_count here for good reason - we may not be unmounting 895 * the last thing mounted. 896 */ 897 iostat_free(nmp->nm_stats); 898 899 /* 900 * There is one reference count to get rid of here 901 * (see comment in mountnfs()). 902 */ 903 VOP_UNLOCK(vp); 904 vgone(vp); 905 nfs_disconnect(nmp); 906 m_freem(nmp->nm_nam); 907 908 rw_destroy(&nmp->nm_writeverflock); 909 rw_destroy(&nmp->nm_rbtlock); 910 mutex_destroy(&nmp->nm_lock); 911 cv_destroy(&nmp->nm_rcvcv); 912 cv_destroy(&nmp->nm_sndcv); 913 cv_destroy(&nmp->nm_aiocv); 914 cv_destroy(&nmp->nm_disconcv); 915 kmem_free(nmp, sizeof(*nmp)); 916 return (0); 917 } 918 919 /* 920 * Return root of a filesystem 921 */ 922 int 923 nfs_root(struct mount *mp, struct vnode **vpp) 924 { 925 struct vnode *vp; 926 struct nfsmount *nmp; 927 int error; 928 929 nmp = VFSTONFS(mp); 930 vp = nmp->nm_vnode; 931 vref(vp); 932 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 933 if (error != 0) { 934 vrele(vp); 935 return error; 936 } 937 *vpp = vp; 938 return (0); 939 } 940 941 extern int syncprt; 942 943 /* 944 * Flush out the buffer cache 945 */ 946 /* ARGSUSED */ 947 int 948 nfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred) 949 { 950 struct vnode *vp, *mvp; 951 int error, allerror = 0; 952 953 /* 954 * Force stale buffer cache information to be flushed. 955 */ 956 if ((mvp = vnalloc(mp)) == NULL) 957 return (ENOMEM); 958 loop: 959 /* 960 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone() 961 * and vclean() can be called indirectly 962 */ 963 mutex_enter(&mntvnode_lock); 964 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) { 965 vmark(mvp, vp); 966 if (vp->v_mount != mp || vismarker(vp)) 967 continue; 968 mutex_enter(&vp->v_interlock); 969 /* XXX MNT_LAZY cannot be right? */ 970 if (waitfor == MNT_LAZY || 971 (LIST_EMPTY(&vp->v_dirtyblkhd) && 972 UVM_OBJ_IS_CLEAN(&vp->v_uobj))) { 973 mutex_exit(&vp->v_interlock); 974 continue; 975 } 976 mutex_exit(&mntvnode_lock); 977 error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT); 978 if (error != 0) { 979 if (error != ENOENT) { 980 mutex_enter(&mntvnode_lock); 981 continue; 982 } 983 (void)vunmark(mvp); 984 goto loop; 985 } 986 error = VOP_FSYNC(vp, cred, 987 waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0); 988 if (error) 989 allerror = error; 990 vput(vp); 991 mutex_enter(&mntvnode_lock); 992 } 993 mutex_exit(&mntvnode_lock); 994 vnfree(mvp); 995 return (allerror); 996 } 997 998 /* 999 * NFS flat namespace lookup. 1000 * Currently unsupported. 1001 */ 1002 /* ARGSUSED */ 1003 int 1004 nfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1005 { 1006 1007 return (EOPNOTSUPP); 1008 } 1009 1010 /* 1011 * Do that sysctl thang... 1012 */ 1013 static int 1014 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS) 1015 { 1016 struct sysctlnode node; 1017 int val; 1018 int error; 1019 1020 val = nfs_niothreads; 1021 node = *rnode; 1022 node.sysctl_data = &val; 1023 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1024 if (error || newp == NULL) 1025 return error; 1026 1027 return nfs_set_niothreads(val); 1028 } 1029 1030 static void 1031 nfs_sysctl_init(void) 1032 { 1033 1034 sysctl_createv(&nfs_clog, 0, NULL, NULL, 1035 CTLFLAG_PERMANENT, 1036 CTLTYPE_NODE, "vfs", NULL, 1037 NULL, 0, NULL, 0, 1038 CTL_VFS, CTL_EOL); 1039 sysctl_createv(&nfs_clog, 0, NULL, NULL, 1040 CTLFLAG_PERMANENT, 1041 CTLTYPE_NODE, "nfs", 1042 SYSCTL_DESCR("NFS vfs options"), 1043 NULL, 0, NULL, 0, 1044 CTL_VFS, 2, CTL_EOL); 1045 /* 1046 * XXX the "2" above could be dynamic, thereby eliminating one 1047 * more instance of the "number to vfs" mapping problem, but 1048 * "2" is the order as taken from sys/mount.h 1049 */ 1050 1051 sysctl_createv(&nfs_clog, 0, NULL, NULL, 1052 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1053 CTLTYPE_STRUCT, "nfsstats", 1054 SYSCTL_DESCR("NFS operation statistics"), 1055 NULL, 0, &nfsstats, sizeof(nfsstats), 1056 CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL); 1057 sysctl_createv(&nfs_clog, 0, NULL, NULL, 1058 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1059 CTLTYPE_INT, "iothreads", 1060 SYSCTL_DESCR("Number of NFS client processes desired"), 1061 sysctl_vfs_nfs_iothreads, 0, NULL, 0, 1062 CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL); 1063 } 1064 1065 static void 1066 nfs_sysctl_fini(void) 1067 { 1068 1069 sysctl_teardown(&nfs_clog); 1070 } 1071 1072 /* ARGSUSED */ 1073 int 1074 nfs_fhtovp(struct mount *mp, struct fid *fid, struct vnode **vpp) 1075 { 1076 size_t fidsize; 1077 size_t fhsize; 1078 struct nfsnode *np; 1079 int error; 1080 struct vattr va; 1081 1082 fidsize = fid->fid_len; 1083 if (fidsize < sizeof(*fid)) { 1084 return EINVAL; 1085 } 1086 fhsize = fidsize - sizeof(*fid); 1087 if ((fhsize % NFSX_UNSIGNED) != 0) { 1088 return EINVAL; 1089 } 1090 if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) { 1091 if (fhsize > NFSX_V3FHMAX || fhsize == 0) { 1092 return EINVAL; 1093 } 1094 } else { 1095 if (fhsize != NFSX_V2FH) { 1096 return EINVAL; 1097 } 1098 } 1099 error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np); 1100 if (error) { 1101 return error; 1102 } 1103 *vpp = NFSTOV(np); 1104 error = VOP_GETATTR(*vpp, &va, kauth_cred_get()); 1105 if (error != 0) { 1106 vput(*vpp); 1107 *vpp = NULLVP; 1108 } 1109 return error; 1110 } 1111 1112 /* ARGSUSED */ 1113 int 1114 nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize) 1115 { 1116 struct nfsnode *np; 1117 struct fid *fid; 1118 size_t fidsize; 1119 int error = 0; 1120 1121 np = VTONFS(vp); 1122 fidsize = sizeof(*fid) + np->n_fhsize; 1123 if (*bufsize < fidsize) { 1124 error = E2BIG; 1125 } 1126 *bufsize = fidsize; 1127 if (error == 0) { 1128 struct fid fid_store; 1129 1130 fid = &fid_store; 1131 memset(fid, 0, sizeof(*fid)); 1132 fid->fid_len = fidsize; 1133 memcpy(buf, fid, sizeof(*fid)); 1134 memcpy(buf->fid_data, np->n_fhp, np->n_fhsize); 1135 } 1136 return error; 1137 } 1138 1139 /* 1140 * Vfs start routine, a no-op. 1141 */ 1142 /* ARGSUSED */ 1143 int 1144 nfs_start(struct mount *mp, int flags) 1145 { 1146 1147 return (0); 1148 } 1149 1150 /* 1151 * Called once at VFS init to initialize client-specific data structures. 1152 */ 1153 void 1154 nfs_vfs_init(void) 1155 { 1156 1157 /* Initialize NFS server / client shared data. */ 1158 nfs_init(); 1159 nfs_node_init(); 1160 1161 /* Initialize the kqueue structures */ 1162 nfs_kqinit(); 1163 /* Initialize the iod structures */ 1164 nfs_iodinit(); 1165 1166 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1167 } 1168 1169 void 1170 nfs_vfs_done(void) 1171 { 1172 1173 nfs_node_done(); 1174 nfs_kqfini(); 1175 nfs_iodfini(); 1176 } 1177