1 /* $NetBSD: nfs_vfsops.c,v 1.241 2020/04/13 19:23:20 ad 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 1.241 2020/04/13 19:23:20 ad 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 .vfs_name = MOUNT_NFS, 108 .vfs_min_mount_data = sizeof (struct nfs_args), 109 .vfs_mount = nfs_mount, 110 .vfs_start = nfs_start, 111 .vfs_unmount = nfs_unmount, 112 .vfs_root = nfs_root, 113 .vfs_quotactl = (void *)eopnotsupp, 114 .vfs_statvfs = nfs_statvfs, 115 .vfs_sync = nfs_sync, 116 .vfs_loadvnode = nfs_loadvnode, 117 .vfs_vget = nfs_vget, 118 .vfs_fhtovp = nfs_fhtovp, 119 .vfs_vptofh = nfs_vptofh, 120 .vfs_init = nfs_vfs_init, 121 .vfs_done = nfs_vfs_done, 122 .vfs_mountroot = nfs_mountroot, 123 .vfs_snapshot = (void *)eopnotsupp, 124 .vfs_extattrctl = vfs_stdextattrctl, 125 .vfs_suspendctl = genfs_suspendctl, 126 .vfs_renamelock_enter = genfs_renamelock_enter, 127 .vfs_renamelock_exit = genfs_renamelock_exit, 128 .vfs_fsync = (void *)eopnotsupp, 129 .vfs_opv_descs = nfs_vnodeopv_descs 130 }; 131 132 extern u_int32_t nfs_procids[NFS_NPROCS]; 133 extern u_int32_t nfs_prog, nfs_vers; 134 135 static int nfs_mount_diskless(struct nfs_dlmount *, const char *, 136 struct mount **, struct vnode **, struct lwp *); 137 138 static int 139 nfs_modcmd(modcmd_t cmd, void *arg) 140 { 141 int error; 142 143 switch (cmd) { 144 case MODULE_CMD_INIT: 145 error = vfs_attach(&nfs_vfsops); 146 return error; 147 case MODULE_CMD_FINI: 148 error = vfs_detach(&nfs_vfsops); 149 return error; 150 default: 151 return ENOTTY; 152 } 153 } 154 155 /* 156 * nfs statvfs call 157 */ 158 int 159 nfs_statvfs(struct mount *mp, struct statvfs *sbp) 160 { 161 struct lwp *l = curlwp; 162 struct vnode *vp; 163 struct nfs_statfs *sfp; 164 char *cp; 165 u_int32_t *tl; 166 int32_t t1, t2; 167 char *bpos, *dpos, *cp2; 168 struct nfsmount *nmp = VFSTONFS(mp); 169 int error = 0, retattr; 170 #ifdef NFS_V2_ONLY 171 const int v3 = 0; 172 #else 173 int v3 = (nmp->nm_flag & NFSMNT_NFSV3); 174 #endif 175 struct mbuf *mreq, *mrep = NULL, *md, *mb; 176 kauth_cred_t cred; 177 u_quad_t tquad; 178 struct nfsnode *np; 179 180 #ifndef nolint 181 sfp = (struct nfs_statfs *)0; 182 #endif 183 vp = nmp->nm_vnode; 184 np = VTONFS(vp); 185 cred = kauth_cred_alloc(); 186 #ifndef NFS_V2_ONLY 187 if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 188 (void)nfs_fsinfo(nmp, vp, cred, l); 189 #endif 190 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 191 nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3)); 192 nfsm_fhtom(np, v3); 193 nfsm_request(np, NFSPROC_FSSTAT, l, cred); 194 if (v3) 195 nfsm_postop_attr(vp, retattr, 0); 196 if (error) { 197 if (mrep != NULL) { 198 if (mrep->m_next != NULL) 199 printf("nfs_vfsops: nfs_statvfs would lose buffers\n"); 200 m_freem(mrep); 201 } 202 goto nfsmout; 203 } 204 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 205 sbp->f_flag = nmp->nm_flag; 206 sbp->f_iosize = uimin(nmp->nm_rsize, nmp->nm_wsize); 207 if (v3) { 208 sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE; 209 tquad = fxdr_hyper(&sfp->sf_tbytes); 210 sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 211 tquad = fxdr_hyper(&sfp->sf_fbytes); 212 sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 213 tquad = fxdr_hyper(&sfp->sf_abytes); 214 tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 215 sbp->f_bresvd = sbp->f_bfree - tquad; 216 sbp->f_bavail = tquad; 217 /* Handle older NFS servers returning negative values */ 218 if ((quad_t)sbp->f_bavail < 0) 219 sbp->f_bavail = 0; 220 tquad = fxdr_hyper(&sfp->sf_tfiles); 221 sbp->f_files = tquad; 222 tquad = fxdr_hyper(&sfp->sf_ffiles); 223 sbp->f_ffree = tquad; 224 sbp->f_favail = tquad; 225 sbp->f_fresvd = 0; 226 sbp->f_namemax = NFS_MAXNAMLEN; 227 } else { 228 sbp->f_bsize = NFS_FABLKSIZE; 229 sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 230 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 231 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 232 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 233 sbp->f_fresvd = 0; 234 sbp->f_files = 0; 235 sbp->f_ffree = 0; 236 sbp->f_favail = 0; 237 sbp->f_fresvd = 0; 238 sbp->f_namemax = NFS_MAXNAMLEN; 239 } 240 copy_statvfs_info(sbp, mp); 241 nfsm_reqdone; 242 kauth_cred_free(cred); 243 return (error); 244 } 245 246 #ifndef NFS_V2_ONLY 247 /* 248 * nfs version 3 fsinfo rpc call 249 */ 250 int 251 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, kauth_cred_t cred, struct lwp *l) 252 { 253 struct nfsv3_fsinfo *fsp; 254 char *cp; 255 int32_t t1, t2; 256 u_int32_t *tl, pref, xmax; 257 char *bpos, *dpos, *cp2; 258 int error = 0, retattr; 259 struct mbuf *mreq, *mrep, *md, *mb; 260 u_int64_t maxfsize; 261 struct nfsnode *np = VTONFS(vp); 262 263 nfsstats.rpccnt[NFSPROC_FSINFO]++; 264 nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1)); 265 nfsm_fhtom(np, 1); 266 nfsm_request(np, NFSPROC_FSINFO, l, cred); 267 nfsm_postop_attr(vp, retattr, 0); 268 if (!error) { 269 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 270 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 271 if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 && 272 pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 273 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 274 ~(NFS_FABLKSIZE - 1); 275 xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 276 if (xmax < nmp->nm_wsize && xmax > 0) { 277 nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1); 278 if (nmp->nm_wsize == 0) 279 nmp->nm_wsize = xmax; 280 } 281 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 282 if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 && 283 pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 284 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 285 ~(NFS_FABLKSIZE - 1); 286 xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 287 if (xmax < nmp->nm_rsize && xmax > 0) { 288 nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1); 289 if (nmp->nm_rsize == 0) 290 nmp->nm_rsize = xmax; 291 } 292 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 293 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ) 294 nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) & 295 ~(NFS_DIRFRAGSIZ - 1); 296 if (xmax < nmp->nm_readdirsize && xmax > 0) { 297 nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1); 298 if (nmp->nm_readdirsize == 0) 299 nmp->nm_readdirsize = xmax; 300 } 301 /* XXX */ 302 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 303 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 304 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 305 nmp->nm_maxfilesize = maxfsize; 306 nmp->nm_mountp->mnt_fs_bshift = 307 ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 308 nmp->nm_iflag |= NFSMNT_GOTFSINFO; 309 } 310 nfsm_reqdone; 311 return (error); 312 } 313 #endif 314 315 /* 316 * Mount a remote root fs via. NFS. It goes like this: 317 * - Call nfs_boot_init() to fill in the nfs_diskless struct 318 * - build the rootfs mount point and call mountnfs() to do the rest. 319 */ 320 int 321 nfs_mountroot(void) 322 { 323 struct timespec ts; 324 struct nfs_diskless *nd; 325 struct vattr attr; 326 struct mount *mp; 327 struct vnode *vp; 328 struct lwp *l; 329 long n; 330 int error; 331 332 l = curlwp; /* XXX */ 333 334 if (device_class(root_device) != DV_IFNET) 335 return (ENODEV); 336 337 /* 338 * XXX time must be non-zero when we init the interface or else 339 * the arp code will wedge. [Fixed now in if_ether.c] 340 * However, the NFS attribute cache gives false "hits" when the 341 * current time < nfs_attrtimeo(nmp, np) so keep this in for now. 342 */ 343 if (time_second < NFS_MAXATTRTIMO) { 344 ts.tv_sec = NFS_MAXATTRTIMO; 345 ts.tv_nsec = 0; 346 tc_setclock(&ts); 347 } 348 349 /* 350 * Call nfs_boot_init() to fill in the nfs_diskless struct. 351 * Side effect: Finds and configures a network interface. 352 */ 353 nd = kmem_zalloc(sizeof(*nd), KM_SLEEP); 354 error = nfs_boot_init(nd, l); 355 if (error) { 356 kmem_free(nd, sizeof(*nd)); 357 return (error); 358 } 359 360 /* 361 * Create the root mount point. 362 */ 363 error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l); 364 if (error) 365 goto out; 366 printf("root on %s\n", nd->nd_root.ndm_host); 367 368 /* 369 * Link it into the mount list. 370 */ 371 mountlist_append(mp); 372 rootvp = vp; 373 mp->mnt_vnodecovered = NULLVP; 374 vfs_unbusy(mp); 375 376 /* Get root attributes (for the time). */ 377 vn_lock(vp, LK_SHARED | LK_RETRY); 378 error = VOP_GETATTR(vp, &attr, l->l_cred); 379 VOP_UNLOCK(vp); 380 if (error) 381 panic("nfs_mountroot: getattr for root"); 382 n = attr.va_atime.tv_sec; 383 #ifdef DEBUG 384 printf("root time: 0x%lx\n", n); 385 #endif 386 setrootfstime(n); 387 388 out: 389 if (error) 390 nfs_boot_cleanup(nd, l); 391 kmem_free(nd, sizeof(*nd)); 392 return (error); 393 } 394 395 /* 396 * Internal version of mount system call for diskless setup. 397 * Separate function because we used to call it twice. 398 * (once for root and once for swap) 399 */ 400 static int 401 nfs_mount_diskless(struct nfs_dlmount *ndmntp, const char *mntname, struct mount **mpp, struct vnode **vpp, struct lwp *l) 402 /* mntname: mount point name */ 403 { 404 struct mount *mp; 405 struct mbuf *m; 406 int error; 407 408 vfs_rootmountalloc(MOUNT_NFS, mntname, &mp); 409 410 mp->mnt_op = &nfs_vfsops; 411 412 /* 413 * Historical practice expects NFS root file systems to 414 * be initially mounted r/w. 415 */ 416 mp->mnt_flag &= ~MNT_RDONLY; 417 418 /* Get mbuf for server sockaddr. */ 419 m = m_get(M_WAIT, MT_SONAME); 420 if (m == NULL) 421 panic("nfs_mountroot: mget soname for %s", mntname); 422 MCLAIM(m, &nfs_mowner); 423 memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr, 424 (m->m_len = ndmntp->ndm_args.addr->sa_len)); 425 426 error = mountnfs(&ndmntp->ndm_args, mp, m, mntname, 427 ndmntp->ndm_args.hostname, vpp, l); 428 if (error) { 429 vfs_unbusy(mp); 430 vfs_rele(mp); 431 printf("nfs_mountroot: mount %s failed: %d\n", 432 mntname, error); 433 } else 434 *mpp = mp; 435 436 return (error); 437 } 438 439 void 440 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp, struct lwp *l) 441 { 442 int s; 443 int adjsock; 444 int maxio; 445 446 s = splsoftnet(); 447 448 /* 449 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 450 * no sense in that context. 451 */ 452 if (argp->sotype == SOCK_STREAM) 453 argp->flags &= ~NFSMNT_NOCONN; 454 455 /* 456 * Cookie translation is not needed for v2, silently ignore it. 457 */ 458 if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) == 459 NFSMNT_XLATECOOKIE) 460 argp->flags &= ~NFSMNT_XLATECOOKIE; 461 462 /* Re-bind if rsrvd port requested and wasn't on one */ 463 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 464 && (argp->flags & NFSMNT_RESVPORT); 465 /* Also re-bind if we're switching to/from a connected UDP socket */ 466 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 467 (argp->flags & NFSMNT_NOCONN)); 468 469 /* Update flags. */ 470 nmp->nm_flag = argp->flags; 471 splx(s); 472 473 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 474 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 475 if (nmp->nm_timeo < NFS_MINTIMEO) 476 nmp->nm_timeo = NFS_MINTIMEO; 477 else if (nmp->nm_timeo > NFS_MAXTIMEO) 478 nmp->nm_timeo = NFS_MAXTIMEO; 479 } 480 481 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 482 nmp->nm_retry = argp->retrans; 483 if (nmp->nm_retry > NFS_MAXREXMIT) 484 nmp->nm_retry = NFS_MAXREXMIT; 485 } 486 487 #ifndef NFS_V2_ONLY 488 if (argp->flags & NFSMNT_NFSV3) { 489 if (argp->sotype == SOCK_DGRAM) 490 maxio = NFS_MAXDGRAMDATA; 491 else 492 maxio = NFS_MAXDATA; 493 } else 494 #endif 495 maxio = NFS_V2MAXDATA; 496 497 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 498 int osize = nmp->nm_wsize; 499 nmp->nm_wsize = argp->wsize; 500 /* Round down to multiple of blocksize */ 501 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 502 if (nmp->nm_wsize <= 0) 503 nmp->nm_wsize = NFS_FABLKSIZE; 504 adjsock |= (nmp->nm_wsize != osize); 505 } 506 if (nmp->nm_wsize > maxio) 507 nmp->nm_wsize = maxio; 508 if (nmp->nm_wsize > MAXBSIZE) 509 nmp->nm_wsize = MAXBSIZE; 510 511 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 512 int osize = nmp->nm_rsize; 513 nmp->nm_rsize = argp->rsize; 514 /* Round down to multiple of blocksize */ 515 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 516 if (nmp->nm_rsize <= 0) 517 nmp->nm_rsize = NFS_FABLKSIZE; 518 adjsock |= (nmp->nm_rsize != osize); 519 } 520 if (nmp->nm_rsize > maxio) 521 nmp->nm_rsize = maxio; 522 if (nmp->nm_rsize > MAXBSIZE) 523 nmp->nm_rsize = MAXBSIZE; 524 525 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 526 nmp->nm_readdirsize = argp->readdirsize; 527 /* Round down to multiple of minimum blocksize */ 528 nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1); 529 if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ) 530 nmp->nm_readdirsize = NFS_DIRFRAGSIZ; 531 /* Bigger than buffer size makes no sense */ 532 if (nmp->nm_readdirsize > NFS_DIRBLKSIZ) 533 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 534 } else if (argp->flags & NFSMNT_RSIZE) 535 nmp->nm_readdirsize = nmp->nm_rsize; 536 537 if (nmp->nm_readdirsize > maxio) 538 nmp->nm_readdirsize = maxio; 539 540 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && 541 argp->maxgrouplist <= NFS_MAXGRPS) 542 nmp->nm_numgrps = argp->maxgrouplist; 543 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && 544 argp->readahead <= NFS_MAXRAHEAD) 545 nmp->nm_readahead = argp->readahead; 546 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && 547 argp->deadthresh <= NFS_NEVERDEAD) 548 nmp->nm_deadthresh = argp->deadthresh; 549 550 adjsock |= ((nmp->nm_sotype != argp->sotype) || 551 (nmp->nm_soproto != argp->proto)); 552 nmp->nm_sotype = argp->sotype; 553 nmp->nm_soproto = argp->proto; 554 555 if (nmp->nm_so && adjsock) { 556 nfs_safedisconnect(nmp); 557 if (nmp->nm_sotype == SOCK_DGRAM) 558 while (nfs_connect(nmp, (struct nfsreq *)0, l)) { 559 printf("nfs_args: retrying connect\n"); 560 kpause("nfscn3", false, hz, NULL); 561 } 562 } 563 } 564 565 /* 566 * VFS Operations. 567 * 568 * mount system call 569 * It seems a bit dumb to copyinstr() the host and path here and then 570 * memcpy() them in mountnfs(), but I wanted to detect errors before 571 * doing the sockargs() call because sockargs() allocates an mbuf and 572 * an error after that means that I have to release the mbuf. 573 */ 574 /* ARGSUSED */ 575 int 576 nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 577 { 578 struct lwp *l = curlwp; 579 int error; 580 struct nfs_args *args = data; 581 struct mbuf *nam; 582 struct nfsmount *nmp = VFSTONFS(mp); 583 struct sockaddr *sa; 584 struct vnode *vp; 585 char *pth, *hst; 586 size_t len; 587 u_char *nfh; 588 589 if (args == NULL) 590 return EINVAL; 591 if (*data_len < sizeof *args) 592 return EINVAL; 593 594 if (mp->mnt_flag & MNT_GETARGS) { 595 596 if (nmp == NULL) 597 return (EIO); 598 if (args->addr != NULL) { 599 sa = mtod(nmp->nm_nam, struct sockaddr *); 600 error = copyout(sa, args->addr, sa->sa_len); 601 if (error) 602 return (error); 603 args->addrlen = sa->sa_len; 604 } else 605 args->addrlen = 0; 606 607 args->version = NFS_ARGSVERSION; 608 args->sotype = nmp->nm_sotype; 609 args->proto = nmp->nm_soproto; 610 args->fh = NULL; 611 args->fhsize = 0; 612 args->flags = nmp->nm_flag; 613 args->wsize = nmp->nm_wsize; 614 args->rsize = nmp->nm_rsize; 615 args->readdirsize = nmp->nm_readdirsize; 616 args->timeo = nmp->nm_timeo; 617 args->retrans = nmp->nm_retry; 618 args->maxgrouplist = nmp->nm_numgrps; 619 args->readahead = nmp->nm_readahead; 620 args->leaseterm = 0; /* dummy */ 621 args->deadthresh = nmp->nm_deadthresh; 622 args->hostname = NULL; 623 *data_len = sizeof *args; 624 return 0; 625 } 626 627 if (args->version != NFS_ARGSVERSION) 628 return (EPROGMISMATCH); 629 if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB)) 630 return (EPROGUNAVAIL); 631 #ifdef NFS_V2_ONLY 632 if (args->flags & NFSMNT_NFSV3) 633 return (EPROGMISMATCH); 634 #endif 635 if (mp->mnt_flag & MNT_UPDATE) { 636 if (nmp == NULL) 637 return (EIO); 638 /* 639 * When doing an update, we can't change from or to 640 * v3, or change cookie translation 641 */ 642 args->flags = (args->flags & ~(NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)) | 643 (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)); 644 nfs_decode_args(nmp, args, l); 645 return (0); 646 } 647 if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX) 648 return (EINVAL); 649 nfh = malloc(NFSX_V3FHMAX, M_TEMP, M_WAITOK); 650 error = copyin(args->fh, nfh, args->fhsize); 651 if (error) 652 goto free_nfh; 653 pth = malloc(MNAMELEN, M_TEMP, M_WAITOK); 654 error = copyinstr(path, pth, MNAMELEN - 1, &len); 655 if (error) 656 goto free_pth; 657 memset(&pth[len], 0, MNAMELEN - len); 658 hst = malloc(MNAMELEN, M_TEMP, M_WAITOK); 659 error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len); 660 if (error) 661 goto free_hst; 662 memset(&hst[len], 0, MNAMELEN - len); 663 /* sockargs() call must be after above copyin() calls */ 664 error = sockargs(&nam, args->addr, args->addrlen, UIO_USERSPACE, 665 MT_SONAME); 666 if (error) 667 goto free_hst; 668 MCLAIM(nam, &nfs_mowner); 669 args->fh = nfh; 670 error = mountnfs(args, mp, nam, pth, hst, &vp, l); 671 672 free_hst: 673 free(hst, M_TEMP); 674 free_pth: 675 free(pth, M_TEMP); 676 free_nfh: 677 free(nfh, M_TEMP); 678 679 return (error); 680 } 681 682 /* 683 * Common code for mount and mountroot 684 */ 685 int 686 mountnfs(struct nfs_args *argp, struct mount *mp, struct mbuf *nam, const char *pth, const char *hst, struct vnode **vpp, struct lwp *l) 687 { 688 struct nfsmount *nmp; 689 struct nfsnode *np; 690 struct vnode *vp; 691 int error; 692 struct vattr *attrs; 693 kauth_cred_t cr; 694 char iosname[IOSTATNAMELEN]; 695 696 /* 697 * If the number of nfs iothreads to use has never 698 * been set, create a reasonable number of them. 699 */ 700 701 if (nfs_niothreads < 0) { 702 nfs_set_niothreads(NFS_DEFAULT_NIOTHREADS); 703 } 704 705 if (mp->mnt_flag & MNT_UPDATE) { 706 nmp = VFSTONFS(mp); 707 /* update paths, file handles, etc, here XXX */ 708 m_freem(nam); 709 return (0); 710 } else { 711 nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP); 712 mp->mnt_data = nmp; 713 TAILQ_INIT(&nmp->nm_uidlruhead); 714 TAILQ_INIT(&nmp->nm_bufq); 715 rw_init(&nmp->nm_writeverflock); 716 mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE); 717 cv_init(&nmp->nm_rcvcv, "nfsrcv"); 718 cv_init(&nmp->nm_sndcv, "nfssnd"); 719 cv_init(&nmp->nm_aiocv, "nfsaio"); 720 cv_init(&nmp->nm_disconcv, "nfsdis"); 721 } 722 vfs_getnewfsid(mp); 723 nmp->nm_mountp = mp; 724 725 #ifndef NFS_V2_ONLY 726 if ((argp->flags & NFSMNT_NFSV3) == 0) 727 #endif 728 { 729 if (argp->fhsize != NFSX_V2FH) { 730 return EINVAL; 731 } 732 } 733 734 /* 735 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo 736 * will overwrite this. 737 */ 738 nmp->nm_maxfilesize = 0xffffffffLL; 739 740 nmp->nm_timeo = NFS_TIMEO; 741 nmp->nm_retry = NFS_RETRANS; 742 nmp->nm_wsize = NFS_WSIZE; 743 nmp->nm_rsize = NFS_RSIZE; 744 nmp->nm_readdirsize = NFS_READDIRSIZE; 745 nmp->nm_numgrps = NFS_MAXGRPS; 746 nmp->nm_readahead = NFS_DEFRAHEAD; 747 nmp->nm_deadthresh = NFS_DEFDEADTHRESH; 748 error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, 749 mp->mnt_op->vfs_name, mp, l); 750 if (error) 751 goto bad; 752 nmp->nm_nam = nam; 753 754 /* Set up the sockets and per-host congestion */ 755 nmp->nm_sotype = argp->sotype; 756 nmp->nm_soproto = argp->proto; 757 758 nfs_decode_args(nmp, argp, l); 759 760 mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 761 mp->mnt_dev_bshift = DEV_BSHIFT; 762 763 /* 764 * For Connection based sockets (TCP,...) defer the connect until 765 * the first request, in case the server is not responding. 766 */ 767 if (nmp->nm_sotype == SOCK_DGRAM && 768 (error = nfs_connect(nmp, (struct nfsreq *)0, l))) 769 goto bad; 770 771 /* 772 * This is silly, but it has to be set so that vinifod() works. 773 * We do not want to do an nfs_statvfs() here since we can get 774 * stuck on a dead server and we are holding a lock on the mount 775 * point. 776 */ 777 mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; 778 error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np); 779 if (error) 780 goto bad; 781 vp = NFSTOV(np); 782 attrs = malloc(sizeof(struct vattr), M_TEMP, M_WAITOK); 783 VOP_GETATTR(vp, attrs, l->l_cred); 784 if ((nmp->nm_flag & NFSMNT_NFSV3) && (vp->v_type == VDIR)) { 785 cr = kauth_cred_alloc(); 786 kauth_cred_setuid(cr, attrs->va_uid); 787 kauth_cred_seteuid(cr, attrs->va_uid); 788 kauth_cred_setsvuid(cr, attrs->va_uid); 789 kauth_cred_setgid(cr, attrs->va_gid); 790 kauth_cred_setegid(cr, attrs->va_gid); 791 kauth_cred_setsvgid(cr, attrs->va_gid); 792 nfs_cookieheuristic(vp, &nmp->nm_iflag, l, cr); 793 kauth_cred_free(cr); 794 } 795 free(attrs, M_TEMP); 796 797 /* 798 * A reference count is needed on the nfsnode representing the 799 * remote root. If this object is not persistent, then backward 800 * traversals of the mount point (i.e. "..") will not work if 801 * the nfsnode gets flushed out of the cache. Ufs does not have 802 * this problem, because one can identify root inodes by their 803 * number == UFS_ROOTINO (2). So, just unlock, but no rele. 804 */ 805 806 nmp->nm_vnode = vp; 807 if (vp->v_type == VNON) 808 vp->v_type = VDIR; 809 vp->v_vflag |= VV_ROOT; 810 VOP_UNLOCK(vp); 811 *vpp = vp; 812 813 snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++); 814 nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname); 815 816 return (0); 817 bad: 818 nfs_disconnect(nmp); 819 rw_destroy(&nmp->nm_writeverflock); 820 mutex_destroy(&nmp->nm_lock); 821 cv_destroy(&nmp->nm_rcvcv); 822 cv_destroy(&nmp->nm_sndcv); 823 cv_destroy(&nmp->nm_aiocv); 824 cv_destroy(&nmp->nm_disconcv); 825 kmem_free(nmp, sizeof(*nmp)); 826 m_freem(nam); 827 return (error); 828 } 829 830 /* 831 * unmount system call 832 */ 833 int 834 nfs_unmount(struct mount *mp, int mntflags) 835 { 836 struct nfsmount *nmp = VFSTONFS(mp); 837 struct vnode *vp; 838 int error, flags = 0; 839 840 if (mntflags & MNT_FORCE) { 841 mutex_enter(&nmp->nm_lock); 842 flags |= FORCECLOSE; 843 nmp->nm_iflag |= NFSMNT_DISMNTFORCE; 844 mutex_exit(&nmp->nm_lock); 845 846 } 847 848 /* 849 * Goes something like this.. 850 * - Check for activity on the root vnode (other than ourselves). 851 * - Call vflush() to clear out vnodes for this file system, 852 * except for the root vnode. 853 * - Decrement reference on the vnode representing remote root. 854 * - Close the socket 855 * - Free up the data structures 856 */ 857 /* 858 * We need to decrement the ref. count on the nfsnode representing 859 * the remote root. See comment in mountnfs(). 860 */ 861 vp = nmp->nm_vnode; 862 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 863 if (error != 0) 864 goto err; 865 866 if ((mntflags & MNT_FORCE) == 0 && vrefcnt(vp) > 1) { 867 VOP_UNLOCK(vp); 868 error = EBUSY; 869 goto err; 870 } 871 872 error = vflush(mp, vp, flags); 873 if (error) { 874 VOP_UNLOCK(vp); 875 goto err; 876 } 877 878 /* 879 * We are now committed to the unmount; mark the mount structure 880 * as doomed so that any sleepers kicked awake by nfs_disconnect 881 * will go away cleanly. 882 */ 883 nmp->nm_iflag |= NFSMNT_DISMNT; 884 885 /* 886 * No new async I/O will be added, but await for pending 887 * ones to drain. 888 */ 889 while (nfs_iodbusy(nmp)) 890 kpause("nfsumnt", false, hz, NULL); 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 mutex_destroy(&nmp->nm_lock); 910 cv_destroy(&nmp->nm_rcvcv); 911 cv_destroy(&nmp->nm_sndcv); 912 cv_destroy(&nmp->nm_aiocv); 913 cv_destroy(&nmp->nm_disconcv); 914 kmem_free(nmp, sizeof(*nmp)); 915 return (0); 916 917 err: 918 if (mntflags & MNT_FORCE) { 919 mutex_enter(&nmp->nm_lock); 920 nmp->nm_iflag &= ~NFSMNT_DISMNTFORCE; 921 mutex_exit(&nmp->nm_lock); 922 } 923 924 return error; 925 } 926 927 /* 928 * Return root of a filesystem 929 */ 930 int 931 nfs_root(struct mount *mp, int lktype, struct vnode **vpp) 932 { 933 struct vnode *vp; 934 struct nfsmount *nmp; 935 int error; 936 937 nmp = VFSTONFS(mp); 938 vp = nmp->nm_vnode; 939 vref(vp); 940 error = vn_lock(vp, lktype | LK_RETRY); 941 if (error != 0) { 942 vrele(vp); 943 return error; 944 } 945 *vpp = vp; 946 return (0); 947 } 948 949 extern int syncprt; 950 951 static bool 952 nfs_sync_selector(void *cl, struct vnode *vp) 953 { 954 955 KASSERT(mutex_owned(vp->v_interlock)); 956 957 return !LIST_EMPTY(&vp->v_dirtyblkhd) || 958 (vp->v_iflag & VI_ONWORKLST) != 0; 959 } 960 961 /* 962 * Flush out the buffer cache 963 */ 964 /* ARGSUSED */ 965 int 966 nfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred) 967 { 968 struct vnode *vp; 969 struct vnode_iterator *marker; 970 int error, allerror = 0; 971 972 /* 973 * Force stale buffer cache information to be flushed. 974 */ 975 vfs_vnode_iterator_init(mp, &marker); 976 while ((vp = vfs_vnode_iterator_next(marker, nfs_sync_selector, 977 NULL))) 978 { 979 error = vn_lock(vp, LK_EXCLUSIVE); 980 if (error) { 981 vrele(vp); 982 continue; 983 } 984 error = VOP_FSYNC(vp, cred, 985 waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0); 986 if (error) 987 allerror = error; 988 vput(vp); 989 } 990 vfs_vnode_iterator_destroy(marker); 991 return allerror; 992 } 993 994 /* 995 * NFS flat namespace lookup. 996 * Currently unsupported. 997 */ 998 /* ARGSUSED */ 999 int 1000 nfs_vget(struct mount *mp, ino_t ino, int lktype, struct vnode **vpp) 1001 { 1002 1003 return (EOPNOTSUPP); 1004 } 1005 1006 /* 1007 * Do that sysctl thang... 1008 */ 1009 static int 1010 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS) 1011 { 1012 struct sysctlnode node; 1013 int val; 1014 int error; 1015 1016 val = nfs_niothreads; 1017 node = *rnode; 1018 node.sysctl_data = &val; 1019 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1020 if (error || newp == NULL) 1021 return error; 1022 1023 return nfs_set_niothreads(val); 1024 } 1025 1026 SYSCTL_SETUP(nfs_sysctl_init, "nfs sysctl") 1027 { 1028 1029 sysctl_createv(clog, 0, NULL, NULL, 1030 CTLFLAG_PERMANENT, 1031 CTLTYPE_NODE, "nfs", 1032 SYSCTL_DESCR("NFS vfs options"), 1033 NULL, 0, NULL, 0, 1034 CTL_VFS, 2, CTL_EOL); 1035 /* 1036 * XXX the "2" above could be dynamic, thereby eliminating one 1037 * more instance of the "number to vfs" mapping problem, but 1038 * "2" is the order as taken from sys/mount.h 1039 */ 1040 1041 sysctl_createv(clog, 0, NULL, NULL, 1042 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1043 CTLTYPE_STRUCT, "nfsstats", 1044 SYSCTL_DESCR("NFS operation statistics"), 1045 NULL, 0, &nfsstats, sizeof(nfsstats), 1046 CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL); 1047 sysctl_createv(clog, 0, NULL, NULL, 1048 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1049 CTLTYPE_INT, "iothreads", 1050 SYSCTL_DESCR("Number of NFS client processes desired"), 1051 sysctl_vfs_nfs_iothreads, 0, NULL, 0, 1052 CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL); 1053 } 1054 1055 /* ARGSUSED */ 1056 int 1057 nfs_fhtovp(struct mount *mp, struct fid *fid, int lktype, struct vnode **vpp) 1058 { 1059 size_t fidsize; 1060 size_t fhsize; 1061 struct nfsnode *np; 1062 int error; 1063 struct vattr va; 1064 1065 fidsize = fid->fid_len; 1066 if (fidsize < sizeof(*fid)) { 1067 return EINVAL; 1068 } 1069 fhsize = fidsize - sizeof(*fid); 1070 if ((fhsize % NFSX_UNSIGNED) != 0) { 1071 return EINVAL; 1072 } 1073 if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) { 1074 if (fhsize > NFSX_V3FHMAX || fhsize == 0) { 1075 return EINVAL; 1076 } 1077 } else { 1078 if (fhsize != NFSX_V2FH) { 1079 return EINVAL; 1080 } 1081 } 1082 /* XXX lktype ignored */ 1083 error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np); 1084 if (error) { 1085 return error; 1086 } 1087 *vpp = NFSTOV(np); 1088 error = VOP_GETATTR(*vpp, &va, kauth_cred_get()); 1089 if (error != 0) { 1090 vput(*vpp); 1091 *vpp = NULLVP; 1092 } 1093 return error; 1094 } 1095 1096 /* ARGSUSED */ 1097 int 1098 nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize) 1099 { 1100 struct nfsnode *np; 1101 struct fid *fid; 1102 size_t fidsize; 1103 int error = 0; 1104 1105 np = VTONFS(vp); 1106 fidsize = sizeof(*fid) + np->n_fhsize; 1107 if (*bufsize < fidsize) { 1108 error = E2BIG; 1109 } 1110 *bufsize = fidsize; 1111 if (error == 0) { 1112 struct fid fid_store; 1113 1114 fid = &fid_store; 1115 memset(fid, 0, sizeof(*fid)); 1116 fid->fid_len = fidsize; 1117 memcpy(buf, fid, sizeof(*fid)); 1118 memcpy(buf->fid_data, np->n_fhp, np->n_fhsize); 1119 } 1120 return error; 1121 } 1122 1123 /* 1124 * Vfs start routine, a no-op. 1125 */ 1126 /* ARGSUSED */ 1127 int 1128 nfs_start(struct mount *mp, int flags) 1129 { 1130 1131 return (0); 1132 } 1133 1134 /* 1135 * Called once at VFS init to initialize client-specific data structures. 1136 */ 1137 void 1138 nfs_vfs_init(void) 1139 { 1140 1141 /* Initialize NFS server / client shared data. */ 1142 nfs_init(); 1143 nfs_node_init(); 1144 1145 /* Initialize the kqueue structures */ 1146 nfs_kqinit(); 1147 /* Initialize the iod structures */ 1148 nfs_iodinit(); 1149 1150 nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4); 1151 } 1152 1153 void 1154 nfs_vfs_done(void) 1155 { 1156 1157 nfs_node_done(); 1158 nfs_kqfini(); 1159 nfs_iodfini(); 1160 nfs_fini(); 1161 } 1162