nfs_vfsops.c revision 1.196.10.3
1 /* $NetBSD: nfs_vfsops.c,v 1.196.10.3 2009/05/04 08:14:22 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 1.196.10.3 2009/05/04 08:14:22 yamt 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/kmem.h> 63 #include <sys/module.h> 64 65 #include <net/if.h> 66 #include <net/route.h> 67 #include <netinet/in.h> 68 69 #include <nfs/rpcv2.h> 70 #include <nfs/nfsproto.h> 71 #include <nfs/nfsnode.h> 72 #include <nfs/nfs.h> 73 #include <nfs/nfsmount.h> 74 #include <nfs/xdr_subs.h> 75 #include <nfs/nfsm_subs.h> 76 #include <nfs/nfsdiskless.h> 77 #include <nfs/nfs_var.h> 78 79 MODULE(MODULE_CLASS_VFS, nfs, NULL); 80 81 extern struct nfsstats nfsstats; 82 extern int nfs_ticks; 83 84 /* 85 * keep a count of the nfs mounts to generate ficticious drive names 86 * for the per drive stats. 87 */ 88 unsigned int nfs_mount_count = 0; 89 90 /* 91 * nfs vfs operations. 92 */ 93 94 extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc; 95 extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc; 96 extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc; 97 98 const struct vnodeopv_desc * const nfs_vnodeopv_descs[] = { 99 &nfsv2_vnodeop_opv_desc, 100 &spec_nfsv2nodeop_opv_desc, 101 &fifo_nfsv2nodeop_opv_desc, 102 NULL, 103 }; 104 105 struct vfsops nfs_vfsops = { 106 MOUNT_NFS, 107 sizeof (struct nfs_args), 108 nfs_mount, 109 nfs_start, 110 nfs_unmount, 111 nfs_root, 112 (void *)eopnotsupp, /* vfs_quotactl */ 113 nfs_statvfs, 114 nfs_sync, 115 nfs_vget, 116 nfs_fhtovp, 117 nfs_vptofh, 118 nfs_vfs_init, 119 NULL, 120 nfs_vfs_done, 121 nfs_mountroot, 122 (int (*)(struct mount *, struct vnode *, struct timespec *)) eopnotsupp, 123 vfs_stdextattrctl, 124 (void *)eopnotsupp, /* vfs_suspendctl */ 125 genfs_renamelock_enter, 126 genfs_renamelock_exit, 127 (void *)eopnotsupp, 128 nfs_vnodeopv_descs, 129 0, 130 { NULL, NULL }, 131 }; 132 133 extern u_int32_t nfs_procids[NFS_NPROCS]; 134 extern u_int32_t nfs_prog, nfs_vers; 135 static struct sysctllog *nfs_clog; 136 137 static int nfs_mount_diskless(struct nfs_dlmount *, const char *, 138 struct mount **, struct vnode **, struct lwp *); 139 static void nfs_sysctl_init(void); 140 static void nfs_sysctl_fini(void); 141 142 static int 143 nfs_modcmd(modcmd_t cmd, void *arg) 144 { 145 int error; 146 147 switch (cmd) { 148 case MODULE_CMD_INIT: 149 error = vfs_attach(&nfs_vfsops); 150 if (error == 0) { 151 nfs_sysctl_init(); 152 } 153 return error; 154 case MODULE_CMD_FINI: 155 error = vfs_detach(&nfs_vfsops); 156 if (error == 0) { 157 nfs_sysctl_fini(); 158 } 159 return error; 160 default: 161 return ENOTTY; 162 } 163 } 164 165 /* 166 * nfs statvfs call 167 */ 168 int 169 nfs_statvfs(struct mount *mp, struct statvfs *sbp) 170 { 171 struct lwp *l = curlwp; 172 struct vnode *vp; 173 struct nfs_statfs *sfp; 174 char *cp; 175 u_int32_t *tl; 176 int32_t t1, t2; 177 char *bpos, *dpos, *cp2; 178 struct nfsmount *nmp = VFSTONFS(mp); 179 int error = 0, retattr; 180 #ifdef NFS_V2_ONLY 181 const int v3 = 0; 182 #else 183 int v3 = (nmp->nm_flag & NFSMNT_NFSV3); 184 #endif 185 struct mbuf *mreq, *mrep = NULL, *md, *mb; 186 kauth_cred_t cred; 187 u_quad_t tquad; 188 struct nfsnode *np; 189 190 #ifndef nolint 191 sfp = (struct nfs_statfs *)0; 192 #endif 193 vp = nmp->nm_vnode; 194 np = VTONFS(vp); 195 cred = kauth_cred_alloc(); 196 #ifndef NFS_V2_ONLY 197 if (v3 && (nmp->nm_iflag & NFSMNT_GOTFSINFO) == 0) 198 (void)nfs_fsinfo(nmp, vp, cred, l); 199 #endif 200 nfsstats.rpccnt[NFSPROC_FSSTAT]++; 201 nfsm_reqhead(np, NFSPROC_FSSTAT, NFSX_FH(v3)); 202 nfsm_fhtom(np, v3); 203 nfsm_request(np, NFSPROC_FSSTAT, l, cred); 204 if (v3) 205 nfsm_postop_attr(vp, retattr, 0); 206 if (error) { 207 if (mrep != NULL) { 208 if (mrep->m_next != NULL) 209 printf("nfs_vfsops: nfs_statvfs would lose buffers\n"); 210 m_freem(mrep); 211 } 212 goto nfsmout; 213 } 214 nfsm_dissect(sfp, struct nfs_statfs *, NFSX_STATFS(v3)); 215 sbp->f_flag = nmp->nm_flag; 216 sbp->f_iosize = min(nmp->nm_rsize, nmp->nm_wsize); 217 if (v3) { 218 sbp->f_frsize = sbp->f_bsize = NFS_FABLKSIZE; 219 tquad = fxdr_hyper(&sfp->sf_tbytes); 220 sbp->f_blocks = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 221 tquad = fxdr_hyper(&sfp->sf_fbytes); 222 sbp->f_bfree = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 223 tquad = fxdr_hyper(&sfp->sf_abytes); 224 tquad = ((quad_t)tquad / (quad_t)NFS_FABLKSIZE); 225 sbp->f_bresvd = sbp->f_bfree - tquad; 226 sbp->f_bavail = tquad; 227 /* Handle older NFS servers returning negative values */ 228 if ((quad_t)sbp->f_bavail < 0) 229 sbp->f_bavail = 0; 230 tquad = fxdr_hyper(&sfp->sf_tfiles); 231 sbp->f_files = tquad; 232 tquad = fxdr_hyper(&sfp->sf_ffiles); 233 sbp->f_ffree = tquad; 234 sbp->f_favail = tquad; 235 sbp->f_fresvd = 0; 236 sbp->f_namemax = MAXNAMLEN; 237 } else { 238 sbp->f_bsize = NFS_FABLKSIZE; 239 sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 240 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 241 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 242 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 243 sbp->f_fresvd = 0; 244 sbp->f_files = 0; 245 sbp->f_ffree = 0; 246 sbp->f_favail = 0; 247 sbp->f_fresvd = 0; 248 sbp->f_namemax = MAXNAMLEN; 249 } 250 copy_statvfs_info(sbp, mp); 251 nfsm_reqdone; 252 kauth_cred_free(cred); 253 return (error); 254 } 255 256 #ifndef NFS_V2_ONLY 257 /* 258 * nfs version 3 fsinfo rpc call 259 */ 260 int 261 nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, kauth_cred_t cred, struct lwp *l) 262 { 263 struct nfsv3_fsinfo *fsp; 264 char *cp; 265 int32_t t1, t2; 266 u_int32_t *tl, pref, xmax; 267 char *bpos, *dpos, *cp2; 268 int error = 0, retattr; 269 struct mbuf *mreq, *mrep, *md, *mb; 270 u_int64_t maxfsize; 271 struct nfsnode *np = VTONFS(vp); 272 273 nfsstats.rpccnt[NFSPROC_FSINFO]++; 274 nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1)); 275 nfsm_fhtom(np, 1); 276 nfsm_request(np, NFSPROC_FSINFO, l, cred); 277 nfsm_postop_attr(vp, retattr, 0); 278 if (!error) { 279 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 280 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 281 if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 && 282 pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 283 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 284 ~(NFS_FABLKSIZE - 1); 285 xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 286 if (xmax < nmp->nm_wsize && xmax > 0) { 287 nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1); 288 if (nmp->nm_wsize == 0) 289 nmp->nm_wsize = xmax; 290 } 291 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 292 if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 && 293 pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 294 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 295 ~(NFS_FABLKSIZE - 1); 296 xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 297 if (xmax < nmp->nm_rsize && xmax > 0) { 298 nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1); 299 if (nmp->nm_rsize == 0) 300 nmp->nm_rsize = xmax; 301 } 302 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 303 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ) 304 nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) & 305 ~(NFS_DIRFRAGSIZ - 1); 306 if (xmax < nmp->nm_readdirsize && xmax > 0) { 307 nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1); 308 if (nmp->nm_readdirsize == 0) 309 nmp->nm_readdirsize = xmax; 310 } 311 /* XXX */ 312 nmp->nm_maxfilesize = (u_int64_t)0x80000000 * DEV_BSIZE - 1; 313 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 314 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 315 nmp->nm_maxfilesize = maxfsize; 316 nmp->nm_mountp->mnt_fs_bshift = 317 ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 318 nmp->nm_iflag |= NFSMNT_GOTFSINFO; 319 } 320 nfsm_reqdone; 321 return (error); 322 } 323 #endif 324 325 /* 326 * Mount a remote root fs via. NFS. It goes like this: 327 * - Call nfs_boot_init() to fill in the nfs_diskless struct 328 * - build the rootfs mount point and call mountnfs() to do the rest. 329 */ 330 int 331 nfs_mountroot(void) 332 { 333 struct timespec ts; 334 struct nfs_diskless *nd; 335 struct vattr attr; 336 struct mount *mp; 337 struct vnode *vp; 338 struct lwp *l; 339 long n; 340 int error; 341 342 l = curlwp; /* XXX */ 343 344 if (device_class(root_device) != DV_IFNET) 345 return (ENODEV); 346 347 /* 348 * XXX time must be non-zero when we init the interface or else 349 * the arp code will wedge. [Fixed now in if_ether.c] 350 * However, the NFS attribute cache gives false "hits" when the 351 * current time < nfs_attrtimeo(nmp, np) so keep this in for now. 352 */ 353 if (time_second < NFS_MAXATTRTIMO) { 354 ts.tv_sec = NFS_MAXATTRTIMO; 355 ts.tv_nsec = 0; 356 tc_setclock(&ts); 357 } 358 359 /* 360 * Call nfs_boot_init() to fill in the nfs_diskless struct. 361 * Side effect: Finds and configures a network interface. 362 */ 363 nd = kmem_zalloc(sizeof(*nd), KM_SLEEP); 364 error = nfs_boot_init(nd, l); 365 if (error) { 366 kmem_free(nd, sizeof(*nd)); 367 return (error); 368 } 369 370 /* 371 * Create the root mount point. 372 */ 373 error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l); 374 if (error) 375 goto out; 376 printf("root on %s\n", nd->nd_root.ndm_host); 377 378 /* 379 * Link it into the mount list. 380 */ 381 mutex_enter(&mountlist_lock); 382 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list); 383 mutex_exit(&mountlist_lock); 384 rootvp = vp; 385 mp->mnt_vnodecovered = NULLVP; 386 vfs_unbusy(mp, false, NULL); 387 388 /* Get root attributes (for the time). */ 389 error = VOP_GETATTR(vp, &attr, l->l_cred); 390 if (error) 391 panic("nfs_mountroot: getattr for root"); 392 n = attr.va_atime.tv_sec; 393 #ifdef DEBUG 394 printf("root time: 0x%lx\n", n); 395 #endif 396 setrootfstime(n); 397 398 out: 399 if (error) 400 nfs_boot_cleanup(nd, l); 401 kmem_free(nd, sizeof(*nd)); 402 return (error); 403 } 404 405 /* 406 * Internal version of mount system call for diskless setup. 407 * Separate function because we used to call it twice. 408 * (once for root and once for swap) 409 */ 410 static int 411 nfs_mount_diskless(struct nfs_dlmount *ndmntp, const char *mntname, struct mount **mpp, struct vnode **vpp, struct lwp *l) 412 /* mntname: mount point name */ 413 { 414 struct mount *mp; 415 struct mbuf *m; 416 int error; 417 418 vfs_rootmountalloc(MOUNT_NFS, mntname, &mp); 419 420 mp->mnt_op = &nfs_vfsops; 421 422 /* 423 * Historical practice expects NFS root file systems to 424 * be initially mounted r/w. 425 */ 426 mp->mnt_flag &= ~MNT_RDONLY; 427 428 /* Get mbuf for server sockaddr. */ 429 m = m_get(M_WAIT, MT_SONAME); 430 if (m == NULL) 431 panic("nfs_mountroot: mget soname for %s", mntname); 432 MCLAIM(m, &nfs_mowner); 433 memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr, 434 (m->m_len = ndmntp->ndm_args.addr->sa_len)); 435 436 error = mountnfs(&ndmntp->ndm_args, mp, m, mntname, 437 ndmntp->ndm_args.hostname, vpp, l); 438 if (error) { 439 vfs_unbusy(mp, false, NULL); 440 vfs_destroy(mp); 441 printf("nfs_mountroot: mount %s failed: %d\n", 442 mntname, error); 443 } else 444 *mpp = mp; 445 446 return (error); 447 } 448 449 void 450 nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp, struct lwp *l) 451 { 452 int adjsock; 453 int maxio; 454 455 rw_enter(&nmp->nm_solock, RW_WRITER); 456 457 /* 458 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 459 * no sense in that context. 460 */ 461 if (argp->sotype == SOCK_STREAM) 462 argp->flags &= ~NFSMNT_NOCONN; 463 464 /* 465 * Cookie translation is not needed for v2, silently ignore it. 466 */ 467 if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) == 468 NFSMNT_XLATECOOKIE) 469 argp->flags &= ~NFSMNT_XLATECOOKIE; 470 471 /* Re-bind if rsrvd port requested and wasn't on one */ 472 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 473 && (argp->flags & NFSMNT_RESVPORT); 474 /* Also re-bind if we're switching to/from a connected UDP socket */ 475 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 476 (argp->flags & NFSMNT_NOCONN)); 477 478 /* Update flags. */ 479 nmp->nm_flag = argp->flags; 480 481 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 482 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 483 if (nmp->nm_timeo < NFS_MINTIMEO) 484 nmp->nm_timeo = NFS_MINTIMEO; 485 else if (nmp->nm_timeo > NFS_MAXTIMEO) 486 nmp->nm_timeo = NFS_MAXTIMEO; 487 } 488 489 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 490 nmp->nm_retry = argp->retrans; 491 if (nmp->nm_retry > NFS_MAXREXMIT) 492 nmp->nm_retry = NFS_MAXREXMIT; 493 } 494 495 #ifndef NFS_V2_ONLY 496 if (argp->flags & NFSMNT_NFSV3) { 497 if (argp->sotype == SOCK_DGRAM) 498 maxio = NFS_MAXDGRAMDATA; 499 else 500 maxio = NFS_MAXDATA; 501 } else 502 #endif 503 maxio = NFS_V2MAXDATA; 504 505 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 506 int osize = nmp->nm_wsize; 507 nmp->nm_wsize = argp->wsize; 508 /* Round down to multiple of blocksize */ 509 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 510 if (nmp->nm_wsize <= 0) 511 nmp->nm_wsize = NFS_FABLKSIZE; 512 adjsock |= (nmp->nm_wsize != osize); 513 } 514 if (nmp->nm_wsize > maxio) 515 nmp->nm_wsize = maxio; 516 if (nmp->nm_wsize > MAXBSIZE) 517 nmp->nm_wsize = MAXBSIZE; 518 519 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 520 int osize = nmp->nm_rsize; 521 nmp->nm_rsize = argp->rsize; 522 /* Round down to multiple of blocksize */ 523 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 524 if (nmp->nm_rsize <= 0) 525 nmp->nm_rsize = NFS_FABLKSIZE; 526 adjsock |= (nmp->nm_rsize != osize); 527 } 528 if (nmp->nm_rsize > maxio) 529 nmp->nm_rsize = maxio; 530 if (nmp->nm_rsize > MAXBSIZE) 531 nmp->nm_rsize = MAXBSIZE; 532 533 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 534 nmp->nm_readdirsize = argp->readdirsize; 535 /* Round down to multiple of minimum blocksize */ 536 nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1); 537 if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ) 538 nmp->nm_readdirsize = NFS_DIRFRAGSIZ; 539 /* Bigger than buffer size makes no sense */ 540 if (nmp->nm_readdirsize > NFS_DIRBLKSIZ) 541 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 542 } else if (argp->flags & NFSMNT_RSIZE) 543 nmp->nm_readdirsize = nmp->nm_rsize; 544 545 if (nmp->nm_readdirsize > maxio) 546 nmp->nm_readdirsize = maxio; 547 548 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && 549 argp->maxgrouplist <= NFS_MAXGRPS) 550 nmp->nm_numgrps = argp->maxgrouplist; 551 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && 552 argp->readahead <= NFS_MAXRAHEAD) 553 nmp->nm_readahead = argp->readahead; 554 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && 555 argp->deadthresh <= NFS_NEVERDEAD) 556 nmp->nm_deadthresh = argp->deadthresh; 557 558 adjsock |= ((nmp->nm_sotype != argp->sotype) || 559 (nmp->nm_soproto != argp->proto)); 560 nmp->nm_sotype = argp->sotype; 561 nmp->nm_soproto = argp->proto; 562 563 if (nmp->nm_so && adjsock) { 564 nfs_safedisconnect(nmp); 565 if (nmp->nm_sotype == SOCK_DGRAM) 566 while (nfs_connect(nmp, (struct nfsreq *)0, l)) { 567 printf("nfs_args: retrying connect\n"); 568 kpause("nfscn3", false, hz, NULL); 569 } 570 } 571 572 rw_exit(&nmp->nm_solock); 573 } 574 575 /* 576 * VFS Operations. 577 * 578 * mount system call 579 * It seems a bit dumb to copyinstr() the host and path here and then 580 * memcpy() them in mountnfs(), but I wanted to detect errors before 581 * doing the sockargs() call because sockargs() allocates an mbuf and 582 * an error after that means that I have to release the mbuf. 583 */ 584 /* ARGSUSED */ 585 int 586 nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 587 { 588 struct lwp *l = curlwp; 589 int error; 590 struct nfs_args *args = data; 591 struct mbuf *nam; 592 struct nfsmount *nmp = VFSTONFS(mp); 593 struct sockaddr *sa; 594 struct vnode *vp; 595 char *pth, *hst; 596 struct proc *p; 597 size_t len; 598 u_char *nfh; 599 600 if (*data_len < sizeof *args) 601 return EINVAL; 602 603 p = l->l_proc; 604 if (mp->mnt_flag & MNT_GETARGS) { 605 606 if (nmp == NULL) 607 return (EIO); 608 if (args->addr != NULL) { 609 sa = mtod(nmp->nm_nam, struct sockaddr *); 610 error = copyout(sa, args->addr, sa->sa_len); 611 if (error) 612 return (error); 613 args->addrlen = sa->sa_len; 614 } else 615 args->addrlen = 0; 616 617 args->version = NFS_ARGSVERSION; 618 args->sotype = nmp->nm_sotype; 619 args->proto = nmp->nm_soproto; 620 args->fh = NULL; 621 args->fhsize = 0; 622 args->flags = nmp->nm_flag; 623 args->wsize = nmp->nm_wsize; 624 args->rsize = nmp->nm_rsize; 625 args->readdirsize = nmp->nm_readdirsize; 626 args->timeo = nmp->nm_timeo; 627 args->retrans = nmp->nm_retry; 628 args->maxgrouplist = nmp->nm_numgrps; 629 args->readahead = nmp->nm_readahead; 630 args->leaseterm = 0; /* dummy */ 631 args->deadthresh = nmp->nm_deadthresh; 632 args->hostname = NULL; 633 *data_len = sizeof *args; 634 return 0; 635 } 636 637 if (args->version != NFS_ARGSVERSION) 638 return (EPROGMISMATCH); 639 if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB)) 640 return (EPROGUNAVAIL); 641 #ifdef NFS_V2_ONLY 642 if (args->flags & NFSMNT_NFSV3) 643 return (EPROGMISMATCH); 644 #endif 645 if (mp->mnt_flag & MNT_UPDATE) { 646 #if 0 647 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 #endif 658 return EOPNOTSUPP; 659 } 660 if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX) 661 return (EINVAL); 662 nfh = malloc(NFSX_V3FHMAX, M_TEMP, M_WAITOK); 663 error = copyin(args->fh, nfh, args->fhsize); 664 if (error) 665 goto free_nfh; 666 pth = malloc(MNAMELEN, M_TEMP, M_WAITOK); 667 error = copyinstr(path, pth, MNAMELEN - 1, &len); 668 if (error) 669 goto free_pth; 670 memset(&pth[len], 0, MNAMELEN - len); 671 hst = malloc(MNAMELEN, M_TEMP, M_WAITOK); 672 error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len); 673 if (error) 674 goto free_hst; 675 memset(&hst[len], 0, MNAMELEN - len); 676 /* sockargs() call must be after above copyin() calls */ 677 error = sockargs(&nam, args->addr, args->addrlen, MT_SONAME); 678 if (error) 679 goto free_hst; 680 MCLAIM(nam, &nfs_mowner); 681 args->fh = nfh; 682 error = mountnfs(args, mp, nam, pth, hst, &vp, l); 683 684 free_hst: 685 free(hst, M_TEMP); 686 free_pth: 687 free(pth, M_TEMP); 688 free_nfh: 689 free(nfh, M_TEMP); 690 691 return (error); 692 } 693 694 /* 695 * Common code for mount and mountroot 696 */ 697 int 698 mountnfs(struct nfs_args *argp, struct mount *mp, struct mbuf *nam, const char *pth, const char *hst, struct vnode **vpp, struct lwp *l) 699 { 700 struct nfsmount *nmp; 701 struct nfsnode *np; 702 struct vnode *vp; 703 int error; 704 struct vattr *attrs; 705 kauth_cred_t cr; 706 char iosname[IOSTATNAMELEN]; 707 708 /* 709 * If the number of nfs iothreads to use has never 710 * been set, create a reasonable number of them. 711 */ 712 713 if (nfs_niothreads < 0) { 714 nfs_set_niothreads(NFS_DEFAULT_NIOTHREADS); 715 } 716 717 if (mp->mnt_flag & MNT_UPDATE) { 718 nmp = VFSTONFS(mp); 719 /* update paths, file handles, etc, here XXX */ 720 m_freem(nam); 721 return (0); 722 } else { 723 nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP); 724 mp->mnt_data = nmp; 725 TAILQ_INIT(&nmp->nm_uidlruhead); 726 TAILQ_INIT(&nmp->nm_bufq); 727 rw_init(&nmp->nm_writeverflock); 728 mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE); 729 rw_init(&nmp->nm_solock); 730 rw_init(&nmp->nm_rbtlock); 731 cv_init(&nmp->nm_rcvcv, "nfsrcv"); 732 cv_init(&nmp->nm_sndcv, "nfssnd"); 733 cv_init(&nmp->nm_aiocv, "nfsaio"); 734 cv_init(&nmp->nm_disconcv, "nfsdis"); 735 nfs_rbtinit(nmp); 736 } 737 vfs_getnewfsid(mp); 738 nmp->nm_mountp = mp; 739 740 #ifndef NFS_V2_ONLY 741 if ((argp->flags & NFSMNT_NFSV3) == 0) 742 #endif 743 { 744 if (argp->fhsize != NFSX_V2FH) { 745 return EINVAL; 746 } 747 } 748 749 /* 750 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo 751 * will overwrite this. 752 */ 753 nmp->nm_maxfilesize = 0xffffffffLL; 754 755 nmp->nm_timeo = NFS_TIMEO; 756 nmp->nm_retry = NFS_RETRANS; 757 nmp->nm_wsize = NFS_WSIZE; 758 nmp->nm_rsize = NFS_RSIZE; 759 nmp->nm_readdirsize = NFS_READDIRSIZE; 760 nmp->nm_numgrps = NFS_MAXGRPS; 761 nmp->nm_readahead = NFS_DEFRAHEAD; 762 nmp->nm_deadthresh = NFS_DEFDEADTHRESH; 763 error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, 764 mp->mnt_op->vfs_name, mp, l); 765 if (error) 766 goto bad; 767 nmp->nm_nam = nam; 768 769 /* Set up the sockets and per-host congestion */ 770 nmp->nm_sotype = argp->sotype; 771 nmp->nm_soproto = argp->proto; 772 773 nfs_decode_args(nmp, argp, l); 774 775 mp->mnt_iflag |= IMNT_MPSAFE; 776 777 mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 778 mp->mnt_dev_bshift = DEV_BSHIFT; 779 780 /* 781 * For Connection based sockets (TCP,...) defer the connect until 782 * the first request, in case the server is not responding. 783 */ 784 rw_enter(&nmp->nm_solock, RW_WRITER); 785 if (nmp->nm_sotype == SOCK_DGRAM && 786 (error = nfs_connect(nmp, NULL, l))) { 787 rw_exit(&nmp->nm_solock); 788 goto bad; 789 } 790 rw_exit(&nmp->nm_solock); 791 792 /* 793 * This is silly, but it has to be set so that vinifod() works. 794 * We do not want to do an nfs_statvfs() here since we can get 795 * stuck on a dead server and we are holding a lock on the mount 796 * point. 797 */ 798 mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; 799 error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np); 800 if (error) 801 goto bad; 802 vp = NFSTOV(np); 803 attrs = kmem_alloc(sizeof(*attrs), KM_SLEEP); 804 VOP_GETATTR(vp, attrs, l->l_cred); 805 if ((nmp->nm_flag & NFSMNT_NFSV3) && (vp->v_type == VDIR)) { 806 cr = kauth_cred_alloc(); 807 kauth_cred_setuid(cr, attrs->va_uid); 808 kauth_cred_seteuid(cr, attrs->va_uid); 809 kauth_cred_setsvuid(cr, attrs->va_uid); 810 kauth_cred_setgid(cr, attrs->va_gid); 811 kauth_cred_setegid(cr, attrs->va_gid); 812 kauth_cred_setsvgid(cr, attrs->va_gid); 813 nfs_cookieheuristic(vp, &nmp->nm_iflag, l, cr); 814 kauth_cred_free(cr); 815 } 816 kmem_free(attrs, sizeof(*attrs)); 817 818 /* 819 * A reference count is needed on the nfsnode representing the 820 * remote root. If this object is not persistent, then backward 821 * traversals of the mount point (i.e. "..") will not work if 822 * the nfsnode gets flushed out of the cache. Ufs does not have 823 * this problem, because one can identify root inodes by their 824 * number == ROOTINO (2). So, just unlock, but no rele. 825 */ 826 827 nmp->nm_vnode = vp; 828 if (vp->v_type == VNON) 829 vp->v_type = VDIR; 830 vp->v_vflag |= VV_ROOT; 831 VOP_UNLOCK(vp, 0); 832 *vpp = vp; 833 834 snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++); 835 nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname); 836 837 return (0); 838 bad: 839 rw_enter(&nmp->nm_solock, RW_WRITER); 840 nfs_disconnect(nmp); 841 rw_exit(&nmp->nm_solock); 842 rw_destroy(&nmp->nm_writeverflock); 843 rw_destroy(&nmp->nm_rbtlock); 844 mutex_destroy(&nmp->nm_lock); 845 rw_destroy(&nmp->nm_solock); 846 cv_destroy(&nmp->nm_rcvcv); 847 cv_destroy(&nmp->nm_sndcv); 848 cv_destroy(&nmp->nm_aiocv); 849 cv_destroy(&nmp->nm_disconcv); 850 kmem_free(nmp, sizeof(*nmp)); 851 m_freem(nam); 852 return (error); 853 } 854 855 /* 856 * unmount system call 857 */ 858 int 859 nfs_unmount(struct mount *mp, int mntflags) 860 { 861 struct nfsmount *nmp; 862 struct vnode *vp; 863 int error, flags = 0; 864 865 if (mntflags & MNT_FORCE) 866 flags |= FORCECLOSE; 867 nmp = VFSTONFS(mp); 868 /* 869 * Goes something like this.. 870 * - Check for activity on the root vnode (other than ourselves). 871 * - Call vflush() to clear out vnodes for this file system, 872 * except for the root vnode. 873 * - Decrement reference on the vnode representing remote root. 874 * - Close the socket 875 * - Free up the data structures 876 */ 877 /* 878 * We need to decrement the ref. count on the nfsnode representing 879 * the remote root. See comment in mountnfs(). The VFS unmount() 880 * has done vput on this vnode, otherwise we would get deadlock! 881 */ 882 vp = nmp->nm_vnode; 883 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 884 if (error != 0) 885 return error; 886 887 if ((mntflags & MNT_FORCE) == 0 && vp->v_usecount > 2) { 888 vput(vp); 889 return (EBUSY); 890 } 891 892 error = vflush(mp, vp, flags); 893 if (error) { 894 vput(vp); 895 return (error); 896 } 897 898 /* 899 * We are now committed to the unmount; mark the mount structure 900 * as doomed so that any sleepers kicked awake by nfs_disconnect 901 * will go away cleanly. 902 */ 903 nmp->nm_iflag |= NFSMNT_DISMNT; 904 905 /* 906 * Clean up the stats... note that we carefully avoid decrementing 907 * nfs_mount_count here for good reason - we may not be unmounting 908 * the last thing mounted. 909 */ 910 iostat_free(nmp->nm_stats); 911 912 /* 913 * There are two reference counts to get rid of here 914 * (see comment in mountnfs()). 915 */ 916 vput(vp); 917 vgone(vp); 918 rw_enter(&nmp->nm_solock, RW_WRITER); 919 nfs_disconnect(nmp); 920 rw_exit(&nmp->nm_solock); 921 m_freem(nmp->nm_nam); 922 923 rw_destroy(&nmp->nm_writeverflock); 924 rw_destroy(&nmp->nm_rbtlock); 925 mutex_destroy(&nmp->nm_lock); 926 rw_destroy(&nmp->nm_solock); 927 cv_destroy(&nmp->nm_rcvcv); 928 cv_destroy(&nmp->nm_sndcv); 929 cv_destroy(&nmp->nm_aiocv); 930 cv_destroy(&nmp->nm_disconcv); 931 kmem_free(nmp, sizeof(*nmp)); 932 return (0); 933 } 934 935 /* 936 * Return root of a filesystem 937 */ 938 int 939 nfs_root(struct mount *mp, struct vnode **vpp) 940 { 941 struct vnode *vp; 942 struct nfsmount *nmp; 943 int error; 944 945 nmp = VFSTONFS(mp); 946 vp = nmp->nm_vnode; 947 error = vget(vp, LK_EXCLUSIVE | LK_RETRY); 948 if (error != 0) 949 return error; 950 *vpp = vp; 951 return (0); 952 } 953 954 extern int syncprt; 955 956 /* 957 * Flush out the buffer cache 958 */ 959 /* ARGSUSED */ 960 int 961 nfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred) 962 { 963 struct vnode *vp, *mvp; 964 int error, allerror = 0; 965 966 /* 967 * Force stale buffer cache information to be flushed. 968 */ 969 if ((mvp = vnalloc(mp)) == NULL) 970 return (ENOMEM); 971 loop: 972 /* 973 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone() 974 * and vclean() can be called indirectly 975 */ 976 mutex_enter(&mntvnode_lock); 977 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) { 978 vmark(mvp, vp); 979 if (vp->v_mount != mp || vismarker(vp)) 980 continue; 981 mutex_enter(&vp->v_interlock); 982 /* XXX MNT_LAZY cannot be right? */ 983 if (waitfor == MNT_LAZY || VOP_ISLOCKED(vp) || 984 (LIST_EMPTY(&vp->v_dirtyblkhd) && 985 UVM_OBJ_IS_CLEAN(&vp->v_uobj))) { 986 mutex_exit(&vp->v_interlock); 987 continue; 988 } 989 mutex_exit(&mntvnode_lock); 990 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) { 991 (void)vunmark(mvp); 992 goto loop; 993 } 994 error = VOP_FSYNC(vp, cred, 995 waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0); 996 if (error) 997 allerror = error; 998 vput(vp); 999 mutex_enter(&mntvnode_lock); 1000 } 1001 mutex_exit(&mntvnode_lock); 1002 vnfree(mvp); 1003 return (allerror); 1004 } 1005 1006 /* 1007 * NFS flat namespace lookup. 1008 * Currently unsupported. 1009 */ 1010 /* ARGSUSED */ 1011 int 1012 nfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp) 1013 { 1014 1015 return (EOPNOTSUPP); 1016 } 1017 1018 /* 1019 * Do that sysctl thang... 1020 */ 1021 static int 1022 sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS) 1023 { 1024 struct sysctlnode node; 1025 int val; 1026 int error; 1027 1028 val = nfs_niothreads; 1029 node = *rnode; 1030 node.sysctl_data = &val; 1031 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1032 if (error || newp == NULL) 1033 return error; 1034 1035 return nfs_set_niothreads(val); 1036 } 1037 1038 static void 1039 nfs_sysctl_init(void) 1040 { 1041 1042 sysctl_createv(&nfs_clog, 0, NULL, NULL, 1043 CTLFLAG_PERMANENT, 1044 CTLTYPE_NODE, "vfs", NULL, 1045 NULL, 0, NULL, 0, 1046 CTL_VFS, CTL_EOL); 1047 sysctl_createv(&nfs_clog, 0, NULL, NULL, 1048 CTLFLAG_PERMANENT, 1049 CTLTYPE_NODE, "nfs", 1050 SYSCTL_DESCR("NFS vfs options"), 1051 NULL, 0, NULL, 0, 1052 CTL_VFS, 2, CTL_EOL); 1053 /* 1054 * XXX the "2" above could be dynamic, thereby eliminating one 1055 * more instance of the "number to vfs" mapping problem, but 1056 * "2" is the order as taken from sys/mount.h 1057 */ 1058 1059 sysctl_createv(&nfs_clog, 0, NULL, NULL, 1060 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1061 CTLTYPE_STRUCT, "nfsstats", 1062 SYSCTL_DESCR("NFS operation statistics"), 1063 NULL, 0, &nfsstats, sizeof(nfsstats), 1064 CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL); 1065 sysctl_createv(&nfs_clog, 0, NULL, NULL, 1066 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1067 CTLTYPE_INT, "iothreads", 1068 SYSCTL_DESCR("Number of NFS client processes desired"), 1069 sysctl_vfs_nfs_iothreads, 0, NULL, 0, 1070 CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL); 1071 } 1072 1073 static void 1074 nfs_sysctl_fini(void) 1075 { 1076 1077 sysctl_teardown(&nfs_clog); 1078 } 1079 1080 /* ARGSUSED */ 1081 int 1082 nfs_fhtovp(struct mount *mp, struct fid *fid, struct vnode **vpp) 1083 { 1084 size_t fidsize; 1085 size_t fhsize; 1086 struct nfsnode *np; 1087 int error; 1088 struct vattr va; 1089 1090 fidsize = fid->fid_len; 1091 if (fidsize < sizeof(*fid)) { 1092 return EINVAL; 1093 } 1094 fhsize = fidsize - sizeof(*fid); 1095 if ((fhsize % NFSX_UNSIGNED) != 0) { 1096 return EINVAL; 1097 } 1098 if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) { 1099 if (fhsize > NFSX_V3FHMAX || fhsize == 0) { 1100 return EINVAL; 1101 } 1102 } else { 1103 if (fhsize != NFSX_V2FH) { 1104 return EINVAL; 1105 } 1106 } 1107 error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np); 1108 if (error) { 1109 return error; 1110 } 1111 *vpp = NFSTOV(np); 1112 error = VOP_GETATTR(*vpp, &va, kauth_cred_get()); 1113 if (error != 0) { 1114 vput(*vpp); 1115 } 1116 return error; 1117 } 1118 1119 /* ARGSUSED */ 1120 int 1121 nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize) 1122 { 1123 struct nfsnode *np; 1124 struct fid *fid; 1125 size_t fidsize; 1126 int error = 0; 1127 1128 np = VTONFS(vp); 1129 fidsize = sizeof(*fid) + np->n_fhsize; 1130 if (*bufsize < fidsize) { 1131 error = E2BIG; 1132 } 1133 *bufsize = fidsize; 1134 if (error == 0) { 1135 struct fid fid_store; 1136 1137 fid = &fid_store; 1138 memset(fid, 0, sizeof(*fid)); 1139 fid->fid_len = fidsize; 1140 memcpy(buf, fid, sizeof(*fid)); 1141 memcpy(buf->fid_data, np->n_fhp, np->n_fhsize); 1142 } 1143 return error; 1144 } 1145 1146 /* 1147 * Vfs start routine, a no-op. 1148 */ 1149 /* ARGSUSED */ 1150 int 1151 nfs_start(struct mount *mp, int flags) 1152 { 1153 1154 return (0); 1155 } 1156
Indexes created Wed Oct 01 07:09:59 GMT 2025