1/* $NetBSD: nfs_vfsops.c,v 1.246 2024/05/13 00:11:22 msaitoh 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 1.246 2024/05/13 00:11:22 msaitoh 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 78MODULE(MODULE_CLASS_VFS, nfs, NULL); 79 80extern struct nfsstats nfsstats; 81extern int nfs_ticks; 82 83/* 84 * keep a count of the nfs mounts to generate fictitious drive names 85 * for the per drive stats. 86 */ 87unsigned int nfs_mount_count = 0; 88 89int nfs_commitsize; 90 91/* 92 * nfs vfs operations. 93 */ 94 95extern const struct vnodeopv_desc nfsv2_vnodeop_opv_desc; 96extern const struct vnodeopv_desc spec_nfsv2nodeop_opv_desc; 97extern const struct vnodeopv_desc fifo_nfsv2nodeop_opv_desc; 98 99const 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 106struct 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 132extern u_int32_t nfs_procids[NFS_NPROCS]; 133extern u_int32_t nfs_prog, nfs_vers; 134 135static int nfs_mount_diskless(struct nfs_dlmount *, const char *, 136 struct mount **, struct vnode **, struct lwp *); 137 138static int 139nfs_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 */ 158int 159nfs_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 } else { 227 sbp->f_bsize = NFS_FABLKSIZE; 228 sbp->f_frsize = fxdr_unsigned(int32_t, sfp->sf_bsize); 229 sbp->f_blocks = fxdr_unsigned(int32_t, sfp->sf_blocks); 230 sbp->f_bfree = fxdr_unsigned(int32_t, sfp->sf_bfree); 231 sbp->f_bavail = fxdr_unsigned(int32_t, sfp->sf_bavail); 232 sbp->f_fresvd = 0; 233 sbp->f_files = 0; 234 sbp->f_ffree = 0; 235 sbp->f_favail = 0; 236 sbp->f_fresvd = 0; 237 } 238 copy_statvfs_info(sbp, mp); 239 nfsm_reqdone; 240 kauth_cred_free(cred); 241 return (error); 242} 243 244#ifndef NFS_V2_ONLY 245/* 246 * nfs version 3 fsinfo rpc call 247 */ 248int 249nfs_fsinfo(struct nfsmount *nmp, struct vnode *vp, kauth_cred_t cred, struct lwp *l) 250{ 251 struct nfsv3_fsinfo *fsp; 252 char *cp; 253 int32_t t1, t2; 254 u_int32_t *tl, pref, xmax; 255 char *bpos, *dpos, *cp2; 256 int error = 0, retattr; 257 struct mbuf *mreq, *mrep, *md, *mb; 258 u_int64_t maxfsize; 259 struct nfsnode *np = VTONFS(vp); 260 261 nfsstats.rpccnt[NFSPROC_FSINFO]++; 262 nfsm_reqhead(np, NFSPROC_FSINFO, NFSX_FH(1)); 263 nfsm_fhtom(np, 1); 264 nfsm_request(np, NFSPROC_FSINFO, l, cred); 265 nfsm_postop_attr(vp, retattr, 0); 266 if (!error) { 267 nfsm_dissect(fsp, struct nfsv3_fsinfo *, NFSX_V3FSINFO); 268 pref = fxdr_unsigned(u_int32_t, fsp->fs_wtpref); 269 if ((nmp->nm_flag & NFSMNT_WSIZE) == 0 && 270 pref < nmp->nm_wsize && pref >= NFS_FABLKSIZE) 271 nmp->nm_wsize = (pref + NFS_FABLKSIZE - 1) & 272 ~(NFS_FABLKSIZE - 1); 273 xmax = fxdr_unsigned(u_int32_t, fsp->fs_wtmax); 274 if (xmax < nmp->nm_wsize && xmax > 0) { 275 nmp->nm_wsize = xmax & ~(NFS_FABLKSIZE - 1); 276 if (nmp->nm_wsize == 0) 277 nmp->nm_wsize = xmax; 278 } 279 pref = fxdr_unsigned(u_int32_t, fsp->fs_rtpref); 280 if ((nmp->nm_flag & NFSMNT_RSIZE) == 0 && 281 pref < nmp->nm_rsize && pref >= NFS_FABLKSIZE) 282 nmp->nm_rsize = (pref + NFS_FABLKSIZE - 1) & 283 ~(NFS_FABLKSIZE - 1); 284 xmax = fxdr_unsigned(u_int32_t, fsp->fs_rtmax); 285 if (xmax < nmp->nm_rsize && xmax > 0) { 286 nmp->nm_rsize = xmax & ~(NFS_FABLKSIZE - 1); 287 if (nmp->nm_rsize == 0) 288 nmp->nm_rsize = xmax; 289 } 290 pref = fxdr_unsigned(u_int32_t, fsp->fs_dtpref); 291 if (pref < nmp->nm_readdirsize && pref >= NFS_DIRFRAGSIZ) 292 nmp->nm_readdirsize = (pref + NFS_DIRFRAGSIZ - 1) & 293 ~(NFS_DIRFRAGSIZ - 1); 294 if (xmax < nmp->nm_readdirsize && xmax > 0) { 295 nmp->nm_readdirsize = xmax & ~(NFS_DIRFRAGSIZ - 1); 296 if (nmp->nm_readdirsize == 0) 297 nmp->nm_readdirsize = xmax; 298 } 299 nmp->nm_maxfilesize = 0xffffffffffffffffull; 300 maxfsize = fxdr_hyper(&fsp->fs_maxfilesize); 301 if (maxfsize > 0 && maxfsize < nmp->nm_maxfilesize) 302 nmp->nm_maxfilesize = maxfsize; 303 nmp->nm_mountp->mnt_fs_bshift = 304 ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 305 nmp->nm_iflag |= NFSMNT_GOTFSINFO; 306 } 307 nfsm_reqdone; 308 return (error); 309} 310#endif 311 312/* 313 * Mount a remote root fs via. NFS. It goes like this: 314 * - Call nfs_boot_init() to fill in the nfs_diskless struct 315 * - build the rootfs mount point and call mountnfs() to do the rest. 316 */ 317int 318nfs_mountroot(void) 319{ 320 struct timespec ts; 321 struct nfs_diskless *nd; 322 struct vattr attr; 323 struct mount *mp; 324 struct vnode *vp; 325 struct lwp *l; 326 time_t n; 327 int error; 328 329 l = curlwp; /* XXX */ 330 331 if (device_class(root_device) != DV_IFNET) 332 return (ENODEV); 333 334 /* 335 * XXX time must be non-zero when we init the interface or else 336 * the arp code will wedge. [Fixed now in if_ether.c] 337 * However, the NFS attribute cache gives false "hits" when the 338 * current time < nfs_attrtimeo(nmp, np) so keep this in for now. 339 */ 340 if (time_second < NFS_MAXATTRTIMO) { 341 ts.tv_sec = NFS_MAXATTRTIMO; 342 ts.tv_nsec = 0; 343 tc_setclock(&ts); 344 } 345 346 /* 347 * Call nfs_boot_init() to fill in the nfs_diskless struct. 348 * Side effect: Finds and configures a network interface. 349 */ 350 nd = kmem_zalloc(sizeof(*nd), KM_SLEEP); 351 error = nfs_boot_init(nd, l); 352 if (error) { 353 kmem_free(nd, sizeof(*nd)); 354 return (error); 355 } 356 357 /* 358 * Create the root mount point. 359 */ 360 error = nfs_mount_diskless(&nd->nd_root, "/", &mp, &vp, l); 361 if (error) 362 goto out; 363 printf("root on %s\n", nd->nd_root.ndm_host); 364 365 /* 366 * Link it into the mount list. 367 */ 368 mountlist_append(mp); 369 rootvp = vp; 370 mp->mnt_vnodecovered = NULLVP; 371 vfs_unbusy(mp); 372 373 /* Get root attributes (for the time). */ 374 vn_lock(vp, LK_SHARED | LK_RETRY); 375 error = VOP_GETATTR(vp, &attr, l->l_cred); 376 VOP_UNLOCK(vp); 377 if (error) 378 panic("nfs_mountroot: getattr for root"); 379 n = attr.va_atime.tv_sec; 380#ifdef DEBUG 381 printf("root time: 0x%jx\n", (intmax_t)n); 382#endif 383 setrootfstime(n); 384 385out: 386 if (error) 387 nfs_boot_cleanup(nd, l); 388 kmem_free(nd, sizeof(*nd)); 389 return (error); 390} 391 392/* 393 * Internal version of mount system call for diskless setup. 394 * Separate function because we used to call it twice. 395 * (once for root and once for swap) 396 */ 397static int 398nfs_mount_diskless(struct nfs_dlmount *ndmntp, const char *mntname, struct mount **mpp, struct vnode **vpp, struct lwp *l) 399 /* mntname: mount point name */ 400{ 401 struct mount *mp; 402 struct mbuf *m; 403 int error; 404 405 vfs_rootmountalloc(MOUNT_NFS, mntname, &mp); 406 407 mp->mnt_op = &nfs_vfsops; 408 409 /* 410 * Historical practice expects NFS root file systems to 411 * be initially mounted r/w. 412 */ 413 mp->mnt_flag &= ~MNT_RDONLY; 414 415 /* Get mbuf for server sockaddr. */ 416 m = m_get(M_WAIT, MT_SONAME); 417 if (m == NULL) 418 panic("nfs_mountroot: mget soname for %s", mntname); 419 MCLAIM(m, &nfs_mowner); 420 memcpy(mtod(m, void *), (void *)ndmntp->ndm_args.addr, 421 (m->m_len = ndmntp->ndm_args.addr->sa_len)); 422 423 error = mountnfs(&ndmntp->ndm_args, mp, m, mntname, 424 ndmntp->ndm_args.hostname, vpp, l); 425 if (error) { 426 vfs_unbusy(mp); 427 vfs_rele(mp); 428 printf("nfs_mountroot: mount %s failed: %d\n", 429 mntname, error); 430 } else 431 *mpp = mp; 432 433 return (error); 434} 435 436void 437nfs_decode_args(struct nfsmount *nmp, struct nfs_args *argp, struct lwp *l) 438{ 439 int s; 440 int adjsock; 441 int maxio; 442 443 s = splsoftnet(); 444 445 /* 446 * Silently clear NFSMNT_NOCONN if it's a TCP mount, it makes 447 * no sense in that context. 448 */ 449 if (argp->sotype == SOCK_STREAM) 450 argp->flags &= ~NFSMNT_NOCONN; 451 452 /* 453 * Cookie translation is not needed for v2, silently ignore it. 454 */ 455 if ((argp->flags & (NFSMNT_XLATECOOKIE|NFSMNT_NFSV3)) == 456 NFSMNT_XLATECOOKIE) 457 argp->flags &= ~NFSMNT_XLATECOOKIE; 458 459 /* Re-bind if rsrvd port requested and wasn't on one */ 460 adjsock = !(nmp->nm_flag & NFSMNT_RESVPORT) 461 && (argp->flags & NFSMNT_RESVPORT); 462 /* Also re-bind if we're switching to/from a connected UDP socket */ 463 adjsock |= ((nmp->nm_flag & NFSMNT_NOCONN) != 464 (argp->flags & NFSMNT_NOCONN)); 465 466 /* Update flags. */ 467 nmp->nm_flag = argp->flags; 468 splx(s); 469 470 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 471 nmp->nm_timeo = (argp->timeo * NFS_HZ + 5) / 10; 472 if (nmp->nm_timeo < NFS_MINTIMEO) 473 nmp->nm_timeo = NFS_MINTIMEO; 474 else if (nmp->nm_timeo > NFS_MAXTIMEO) 475 nmp->nm_timeo = NFS_MAXTIMEO; 476 } 477 478 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 479 nmp->nm_retry = argp->retrans; 480 if (nmp->nm_retry > NFS_MAXREXMIT) 481 nmp->nm_retry = NFS_MAXREXMIT; 482 } 483 484#ifndef NFS_V2_ONLY 485 if (argp->flags & NFSMNT_NFSV3) { 486 if (argp->sotype == SOCK_DGRAM) 487 maxio = NFS_MAXDGRAMDATA; 488 else 489 maxio = NFS_MAXDATA; 490 } else 491#endif 492 maxio = NFS_V2MAXDATA; 493 494 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 495 int osize = nmp->nm_wsize; 496 nmp->nm_wsize = argp->wsize; 497 /* Round down to multiple of blocksize */ 498 nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1); 499 if (nmp->nm_wsize <= 0) 500 nmp->nm_wsize = NFS_FABLKSIZE; 501 adjsock |= (nmp->nm_wsize != osize); 502 } 503 if (nmp->nm_wsize > maxio) 504 nmp->nm_wsize = maxio; 505 if (nmp->nm_wsize > MAXBSIZE) 506 nmp->nm_wsize = MAXBSIZE; 507 508 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 509 int osize = nmp->nm_rsize; 510 nmp->nm_rsize = argp->rsize; 511 /* Round down to multiple of blocksize */ 512 nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1); 513 if (nmp->nm_rsize <= 0) 514 nmp->nm_rsize = NFS_FABLKSIZE; 515 adjsock |= (nmp->nm_rsize != osize); 516 } 517 if (nmp->nm_rsize > maxio) 518 nmp->nm_rsize = maxio; 519 if (nmp->nm_rsize > MAXBSIZE) 520 nmp->nm_rsize = MAXBSIZE; 521 522 if ((argp->flags & NFSMNT_READDIRSIZE) && argp->readdirsize > 0) { 523 nmp->nm_readdirsize = argp->readdirsize; 524 /* Round down to multiple of minimum blocksize */ 525 nmp->nm_readdirsize &= ~(NFS_DIRFRAGSIZ - 1); 526 if (nmp->nm_readdirsize < NFS_DIRFRAGSIZ) 527 nmp->nm_readdirsize = NFS_DIRFRAGSIZ; 528 /* Bigger than buffer size makes no sense */ 529 if (nmp->nm_readdirsize > NFS_DIRBLKSIZ) 530 nmp->nm_readdirsize = NFS_DIRBLKSIZ; 531 } else if (argp->flags & NFSMNT_RSIZE) 532 nmp->nm_readdirsize = nmp->nm_rsize; 533 534 if (nmp->nm_readdirsize > maxio) 535 nmp->nm_readdirsize = maxio; 536 537 if ((argp->flags & NFSMNT_MAXGRPS) && argp->maxgrouplist >= 0 && 538 argp->maxgrouplist <= NFS_MAXGRPS) 539 nmp->nm_numgrps = argp->maxgrouplist; 540 if ((argp->flags & NFSMNT_READAHEAD) && argp->readahead >= 0 && 541 argp->readahead <= NFS_MAXRAHEAD) 542 nmp->nm_readahead = argp->readahead; 543 if ((argp->flags & NFSMNT_DEADTHRESH) && argp->deadthresh >= 1 && 544 argp->deadthresh <= NFS_NEVERDEAD) 545 nmp->nm_deadthresh = argp->deadthresh; 546 547 adjsock |= ((nmp->nm_sotype != argp->sotype) || 548 (nmp->nm_soproto != argp->proto)); 549 nmp->nm_sotype = argp->sotype; 550 nmp->nm_soproto = argp->proto; 551 552 if (nmp->nm_so && adjsock) { 553 nfs_safedisconnect(nmp); 554 if (nmp->nm_sotype == SOCK_DGRAM) 555 while (nfs_connect(nmp, (struct nfsreq *)0, l)) { 556 printf("nfs_args: retrying connect\n"); 557 kpause("nfscn3", false, hz, NULL); 558 } 559 } 560} 561 562/* 563 * VFS Operations. 564 * 565 * mount system call 566 * It seems a bit dumb to copyinstr() the host and path here and then 567 * memcpy() them in mountnfs(), but I wanted to detect errors before 568 * doing the sockargs() call because sockargs() allocates an mbuf and 569 * an error after that means that I have to release the mbuf. 570 */ 571/* ARGSUSED */ 572int 573nfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len) 574{ 575 struct lwp *l = curlwp; 576 int error; 577 struct nfs_args *args = data; 578 struct mbuf *nam; 579 struct nfsmount *nmp = VFSTONFS(mp); 580 struct sockaddr *sa; 581 struct vnode *vp; 582 char *pth, *hst; 583 size_t len; 584 u_char *nfh; 585 586 if (args == NULL) 587 return EINVAL; 588 if (*data_len < sizeof *args) 589 return EINVAL; 590 591 if (mp->mnt_flag & MNT_GETARGS) { 592 593 if (nmp == NULL) 594 return (EIO); 595 if (args->addr != NULL) { 596 sa = mtod(nmp->nm_nam, struct sockaddr *); 597 error = copyout(sa, args->addr, sa->sa_len); 598 if (error) 599 return (error); 600 args->addrlen = sa->sa_len; 601 } else 602 args->addrlen = 0; 603 604 args->version = NFS_ARGSVERSION; 605 args->sotype = nmp->nm_sotype; 606 args->proto = nmp->nm_soproto; 607 args->fh = NULL; 608 args->fhsize = 0; 609 args->flags = nmp->nm_flag; 610 args->wsize = nmp->nm_wsize; 611 args->rsize = nmp->nm_rsize; 612 args->readdirsize = nmp->nm_readdirsize; 613 args->timeo = nmp->nm_timeo; 614 args->retrans = nmp->nm_retry; 615 args->maxgrouplist = nmp->nm_numgrps; 616 args->readahead = nmp->nm_readahead; 617 args->leaseterm = 0; /* dummy */ 618 args->deadthresh = nmp->nm_deadthresh; 619 args->hostname = NULL; 620 *data_len = sizeof *args; 621 return 0; 622 } 623 624 if (args->version != NFS_ARGSVERSION) 625 return (EPROGMISMATCH); 626 if (args->flags & (NFSMNT_NQNFS|NFSMNT_KERB)) 627 return (EPROGUNAVAIL); 628#ifdef NFS_V2_ONLY 629 if (args->flags & NFSMNT_NFSV3) 630 return (EPROGMISMATCH); 631#endif 632 if (mp->mnt_flag & MNT_UPDATE) { 633 if (nmp == NULL) 634 return (EIO); 635 /* 636 * When doing an update, we can't change from or to 637 * v3, or change cookie translation 638 */ 639 args->flags = (args->flags & ~(NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)) | 640 (nmp->nm_flag & (NFSMNT_NFSV3|NFSMNT_XLATECOOKIE)); 641 nfs_decode_args(nmp, args, l); 642 return (0); 643 } 644 if (args->fhsize < 0 || args->fhsize > NFSX_V3FHMAX) 645 return (EINVAL); 646 nfh = malloc(NFSX_V3FHMAX, M_TEMP, M_WAITOK); 647 error = copyin(args->fh, nfh, args->fhsize); 648 if (error) 649 goto free_nfh; 650 pth = malloc(MNAMELEN, M_TEMP, M_WAITOK); 651 error = copyinstr(path, pth, MNAMELEN - 1, &len); 652 if (error) 653 goto free_pth; 654 memset(&pth[len], 0, MNAMELEN - len); 655 hst = malloc(MNAMELEN, M_TEMP, M_WAITOK); 656 error = copyinstr(args->hostname, hst, MNAMELEN - 1, &len); 657 if (error) 658 goto free_hst; 659 memset(&hst[len], 0, MNAMELEN - len); 660 /* sockargs() call must be after above copyin() calls */ 661 error = sockargs(&nam, args->addr, args->addrlen, UIO_USERSPACE, 662 MT_SONAME); 663 if (error) 664 goto free_hst; 665 MCLAIM(nam, &nfs_mowner); 666 args->fh = nfh; 667 error = mountnfs(args, mp, nam, pth, hst, &vp, l); 668 669free_hst: 670 free(hst, M_TEMP); 671free_pth: 672 free(pth, M_TEMP); 673free_nfh: 674 free(nfh, M_TEMP); 675 676 return (error); 677} 678 679/* 680 * Common code for mount and mountroot 681 */ 682int 683mountnfs(struct nfs_args *argp, struct mount *mp, struct mbuf *nam, const char *pth, const char *hst, struct vnode **vpp, struct lwp *l) 684{ 685 struct nfsmount *nmp; 686 struct nfsnode *np; 687 struct vnode *vp; 688 int error; 689 struct vattr *attrs; 690 kauth_cred_t cr; 691 char iosname[IOSTATNAMELEN]; 692 693 /* 694 * If the number of nfs iothreads to use has never 695 * been set, create a reasonable number of them. 696 */ 697 698 if (nfs_niothreads < 0) { 699 nfs_set_niothreads(NFS_DEFAULT_NIOTHREADS); 700 } 701 702 if (mp->mnt_flag & MNT_UPDATE) { 703 nmp = VFSTONFS(mp); 704 /* update paths, file handles, etc, here XXX */ 705 m_freem(nam); 706 return 0; 707 } 708 nmp = kmem_zalloc(sizeof(*nmp), KM_SLEEP); 709 TAILQ_INIT(&nmp->nm_uidlruhead); 710 TAILQ_INIT(&nmp->nm_bufq); 711 rw_init(&nmp->nm_writeverflock); 712 mutex_init(&nmp->nm_lock, MUTEX_DEFAULT, IPL_NONE); 713 cv_init(&nmp->nm_rcvcv, "nfsrcv"); 714 cv_init(&nmp->nm_sndcv, "nfssnd"); 715 cv_init(&nmp->nm_aiocv, "nfsaio"); 716 cv_init(&nmp->nm_disconcv, "nfsdis"); 717 718 mp->mnt_data = nmp; 719 mp->mnt_stat.f_namemax = NFS_MAXNAMLEN; 720 vfs_getnewfsid(mp); 721 nmp->nm_mountp = mp; 722 723#ifndef NFS_V2_ONLY 724 if ((argp->flags & NFSMNT_NFSV3) == 0) 725#endif 726 { 727 if (argp->fhsize != NFSX_V2FH) { 728 return EINVAL; 729 } 730 } 731 732 /* 733 * V2 can only handle 32 bit filesizes. For v3, nfs_fsinfo 734 * will overwrite this. 735 */ 736 nmp->nm_maxfilesize = 0xffffffffLL; 737 738 nmp->nm_timeo = NFS_TIMEO; 739 nmp->nm_retry = NFS_RETRANS; 740 nmp->nm_wsize = NFS_WSIZE; 741 nmp->nm_rsize = NFS_RSIZE; 742 nmp->nm_readdirsize = NFS_READDIRSIZE; 743 nmp->nm_numgrps = NFS_MAXGRPS; 744 nmp->nm_readahead = NFS_DEFRAHEAD; 745 nmp->nm_deadthresh = NFS_DEFDEADTHRESH; 746 error = set_statvfs_info(pth, UIO_SYSSPACE, hst, UIO_SYSSPACE, 747 mp->mnt_op->vfs_name, mp, l); 748 if (error) 749 goto bad; 750 nmp->nm_nam = nam; 751 752 /* Set up the sockets and per-host congestion */ 753 nmp->nm_sotype = argp->sotype; 754 nmp->nm_soproto = argp->proto; 755 756 nfs_decode_args(nmp, argp, l); 757 758 mp->mnt_fs_bshift = ffs(MIN(nmp->nm_rsize, nmp->nm_wsize)) - 1; 759 mp->mnt_dev_bshift = DEV_BSHIFT; 760 761 /* 762 * For Connection based sockets (TCP,...) defer the connect until 763 * the first request, in case the server is not responding. 764 */ 765 if (nmp->nm_sotype == SOCK_DGRAM && 766 (error = nfs_connect(nmp, (struct nfsreq *)0, l))) 767 goto bad; 768 769 /* 770 * This is silly, but it has to be set so that vinifod() works. 771 * We do not want to do an nfs_statvfs() here since we can get 772 * stuck on a dead server and we are holding a lock on the mount 773 * point. 774 */ 775 mp->mnt_stat.f_iosize = NFS_MAXDGRAMDATA; 776 error = nfs_nget(mp, (nfsfh_t *)argp->fh, argp->fhsize, &np); 777 if (error) 778 goto bad; 779 vp = NFSTOV(np); 780 attrs = malloc(sizeof(struct vattr), M_TEMP, M_WAITOK); 781 VOP_GETATTR(vp, attrs, l->l_cred); 782 if ((nmp->nm_flag & NFSMNT_NFSV3) && (vp->v_type == VDIR)) { 783 cr = kauth_cred_alloc(); 784 kauth_cred_setuid(cr, attrs->va_uid); 785 kauth_cred_seteuid(cr, attrs->va_uid); 786 kauth_cred_setsvuid(cr, attrs->va_uid); 787 kauth_cred_setgid(cr, attrs->va_gid); 788 kauth_cred_setegid(cr, attrs->va_gid); 789 kauth_cred_setsvgid(cr, attrs->va_gid); 790 nfs_cookieheuristic(vp, &nmp->nm_iflag, l, cr); 791 kauth_cred_free(cr); 792 } 793 free(attrs, M_TEMP); 794 795 /* 796 * A reference count is needed on the nfsnode representing the 797 * remote root. If this object is not persistent, then backward 798 * traversals of the mount point (i.e. "..") will not work if 799 * the nfsnode gets flushed out of the cache. Ufs does not have 800 * this problem, because one can identify root inodes by their 801 * number == UFS_ROOTINO (2). So, just unlock, but no rele. 802 */ 803 804 nmp->nm_vnode = vp; 805 if (vp->v_type == VNON) 806 vp->v_type = VDIR; 807 vp->v_vflag |= VV_ROOT; 808 VOP_UNLOCK(vp); 809 *vpp = vp; 810 811 snprintf(iosname, sizeof(iosname), "nfs%u", nfs_mount_count++); 812 nmp->nm_stats = iostat_alloc(IOSTAT_NFS, nmp, iosname); 813 814 return (0); 815bad: 816 nfs_disconnect(nmp); 817 rw_destroy(&nmp->nm_writeverflock); 818 mutex_destroy(&nmp->nm_lock); 819 cv_destroy(&nmp->nm_rcvcv); 820 cv_destroy(&nmp->nm_sndcv); 821 cv_destroy(&nmp->nm_aiocv); 822 cv_destroy(&nmp->nm_disconcv); 823 kmem_free(nmp, sizeof(*nmp)); 824 m_freem(nam); 825 return (error); 826} 827 828/* 829 * unmount system call 830 */ 831int 832nfs_unmount(struct mount *mp, int mntflags) 833{ 834 struct nfsmount *nmp = VFSTONFS(mp); 835 struct vnode *vp; 836 int error, flags = 0; 837 838 if (mntflags & MNT_FORCE) { 839 mutex_enter(&nmp->nm_lock); 840 flags |= FORCECLOSE; 841 nmp->nm_iflag |= NFSMNT_DISMNTFORCE; 842 mutex_exit(&nmp->nm_lock); 843 844 } 845 846 /* 847 * Goes something like this.. 848 * - Check for activity on the root vnode (other than ourselves). 849 * - Call vflush() to clear out vnodes for this file system, 850 * except for the root vnode. 851 * - Decrement reference on the vnode representing remote root. 852 * - Close the socket 853 * - Free up the data structures 854 */ 855 /* 856 * We need to decrement the ref. count on the nfsnode representing 857 * the remote root. See comment in mountnfs(). 858 */ 859 vp = nmp->nm_vnode; 860 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 861 if (error != 0) 862 goto err; 863 864 if ((mntflags & MNT_FORCE) == 0 && vrefcnt(vp) > 1) { 865 VOP_UNLOCK(vp); 866 error = EBUSY; 867 goto err; 868 } 869 870 error = vflush(mp, vp, flags); 871 if (error) { 872 VOP_UNLOCK(vp); 873 goto err; 874 } 875 876 /* 877 * We are now committed to the unmount; mark the mount structure 878 * as doomed so that any sleepers kicked awake by nfs_disconnect 879 * will go away cleanly. 880 */ 881 nmp->nm_iflag |= NFSMNT_DISMNT; 882 883 /* 884 * No new async I/O will be added, but await for pending 885 * ones to drain. 886 */ 887 while (nfs_iodbusy(nmp)) 888 kpause("nfsumnt", false, hz, NULL); 889 890 /* 891 * Clean up the stats... note that we carefully avoid decrementing 892 * nfs_mount_count here for good reason - we may not be unmounting 893 * the last thing mounted. 894 */ 895 iostat_free(nmp->nm_stats); 896 897 /* 898 * There is one reference count to get rid of here 899 * (see comment in mountnfs()). 900 */ 901 VOP_UNLOCK(vp); 902 vgone(vp); 903 nfs_disconnect(nmp); 904 m_freem(nmp->nm_nam); 905 906 rw_destroy(&nmp->nm_writeverflock); 907 mutex_destroy(&nmp->nm_lock); 908 cv_destroy(&nmp->nm_rcvcv); 909 cv_destroy(&nmp->nm_sndcv); 910 cv_destroy(&nmp->nm_aiocv); 911 cv_destroy(&nmp->nm_disconcv); 912 kmem_free(nmp, sizeof(*nmp)); 913 return (0); 914 915err: 916 if (mntflags & MNT_FORCE) { 917 mutex_enter(&nmp->nm_lock); 918 nmp->nm_iflag &= ~NFSMNT_DISMNTFORCE; 919 mutex_exit(&nmp->nm_lock); 920 } 921 922 return error; 923} 924 925/* 926 * Return root of a filesystem 927 */ 928int 929nfs_root(struct mount *mp, int lktype, struct vnode **vpp) 930{ 931 struct vnode *vp; 932 struct nfsmount *nmp; 933 int error; 934 935 nmp = VFSTONFS(mp); 936 vp = nmp->nm_vnode; 937 vref(vp); 938 error = vn_lock(vp, lktype | LK_RETRY); 939 if (error != 0) { 940 vrele(vp); 941 return error; 942 } 943 *vpp = vp; 944 return (0); 945} 946 947extern int syncprt; 948 949static bool 950nfs_sync_selector(void *cl, struct vnode *vp) 951{ 952 953 KASSERT(mutex_owned(vp->v_interlock)); 954 955 return !LIST_EMPTY(&vp->v_dirtyblkhd) || 956 (vp->v_iflag & VI_ONWORKLST) != 0; 957} 958 959/* 960 * Flush out the buffer cache 961 */ 962/* ARGSUSED */ 963int 964nfs_sync(struct mount *mp, int waitfor, kauth_cred_t cred) 965{ 966 struct vnode *vp; 967 struct vnode_iterator *marker; 968 int error, allerror = 0; 969 970 /* 971 * Force stale buffer cache information to be flushed. 972 */ 973 vfs_vnode_iterator_init(mp, &marker); 974 while ((vp = vfs_vnode_iterator_next(marker, nfs_sync_selector, 975 NULL))) 976 { 977 error = vn_lock(vp, LK_EXCLUSIVE); 978 if (error) { 979 vrele(vp); 980 continue; 981 } 982 error = VOP_FSYNC(vp, cred, 983 waitfor == MNT_WAIT ? FSYNC_WAIT : 0, 0, 0); 984 if (error) 985 allerror = error; 986 vput(vp); 987 } 988 vfs_vnode_iterator_destroy(marker); 989 return allerror; 990} 991 992/* 993 * NFS flat namespace lookup. 994 * Currently unsupported. 995 */ 996/* ARGSUSED */ 997int 998nfs_vget(struct mount *mp, ino_t ino, int lktype, struct vnode **vpp) 999{ 1000 1001 return (EOPNOTSUPP); 1002} 1003 1004/* 1005 * Do that sysctl thang... 1006 */ 1007static int 1008sysctl_vfs_nfs_iothreads(SYSCTLFN_ARGS) 1009{ 1010 struct sysctlnode node; 1011 int val; 1012 int error; 1013 1014 val = nfs_niothreads; 1015 node = *rnode; 1016 node.sysctl_data = &val; 1017 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 1018 if (error || newp == NULL) 1019 return error; 1020 1021 return nfs_set_niothreads(val); 1022} 1023 1024SYSCTL_SETUP(nfs_sysctl_init, "nfs sysctl") 1025{ 1026 1027 sysctl_createv(clog, 0, NULL, NULL, 1028 CTLFLAG_PERMANENT, 1029 CTLTYPE_NODE, "nfs", 1030 SYSCTL_DESCR("NFS vfs options"), 1031 NULL, 0, NULL, 0, 1032 CTL_VFS, 2, CTL_EOL); 1033 /* 1034 * XXX the "2" above could be dynamic, thereby eliminating one 1035 * more instance of the "number to vfs" mapping problem, but 1036 * "2" is the order as taken from sys/mount.h 1037 */ 1038 1039 sysctl_createv(clog, 0, NULL, NULL, 1040 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1041 CTLTYPE_STRUCT, "nfsstats", 1042 SYSCTL_DESCR("NFS operation statistics"), 1043 NULL, 0, &nfsstats, sizeof(nfsstats), 1044 CTL_VFS, 2, NFS_NFSSTATS, CTL_EOL); 1045 sysctl_createv(clog, 0, NULL, NULL, 1046 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 1047 CTLTYPE_INT, "iothreads", 1048 SYSCTL_DESCR("Number of NFS client processes desired"), 1049 sysctl_vfs_nfs_iothreads, 0, NULL, 0, 1050 CTL_VFS, 2, NFS_IOTHREADS, CTL_EOL); 1051} 1052 1053/* ARGSUSED */ 1054int 1055nfs_fhtovp(struct mount *mp, struct fid *fid, int lktype, struct vnode **vpp) 1056{ 1057 size_t fidsize; 1058 size_t fhsize; 1059 struct nfsnode *np; 1060 int error; 1061 struct vattr va; 1062 1063 fidsize = fid->fid_len; 1064 if (fidsize < sizeof(*fid)) { 1065 return EINVAL; 1066 } 1067 fhsize = fidsize - sizeof(*fid); 1068 if ((fhsize % NFSX_UNSIGNED) != 0) { 1069 return EINVAL; 1070 } 1071 if ((VFSTONFS(mp)->nm_flag & NFSMNT_NFSV3) != 0) { 1072 if (fhsize > NFSX_V3FHMAX || fhsize == 0) { 1073 return EINVAL; 1074 } 1075 } else { 1076 if (fhsize != NFSX_V2FH) { 1077 return EINVAL; 1078 } 1079 } 1080 /* XXX lktype ignored */ 1081 error = nfs_nget(mp, (void *)fid->fid_data, fhsize, &np); 1082 if (error) { 1083 return error; 1084 } 1085 *vpp = NFSTOV(np); 1086 error = VOP_GETATTR(*vpp, &va, kauth_cred_get()); 1087 if (error != 0) { 1088 vput(*vpp); 1089 *vpp = NULLVP; 1090 } 1091 return error; 1092} 1093 1094/* ARGSUSED */ 1095int 1096nfs_vptofh(struct vnode *vp, struct fid *buf, size_t *bufsize) 1097{ 1098 struct nfsnode *np; 1099 struct fid *fid; 1100 size_t fidsize; 1101 int error = 0; 1102 1103 np = VTONFS(vp); 1104 fidsize = sizeof(*fid) + np->n_fhsize; 1105 if (*bufsize < fidsize) { 1106 error = E2BIG; 1107 } 1108 *bufsize = fidsize; 1109 if (error == 0) { 1110 struct fid fid_store; 1111 1112 fid = &fid_store; 1113 memset(fid, 0, sizeof(*fid)); 1114 fid->fid_len = fidsize; 1115 memcpy(buf, fid, sizeof(*fid)); 1116 memcpy(buf->fid_data, np->n_fhp, np->n_fhsize); 1117 } 1118 return error; 1119} 1120 1121/* 1122 * Vfs start routine, a no-op. 1123 */ 1124/* ARGSUSED */ 1125int 1126nfs_start(struct mount *mp, int flags) 1127{ 1128 1129 return (0); 1130} 1131 1132/* 1133 * Called once at VFS init to initialize client-specific data structures. 1134 */ 1135void 1136nfs_vfs_init(void) 1137{ 1138 unsigned scale; 1139 1140 /* Initialize NFS server / client shared data. */ 1141 nfs_init(); 1142 nfs_node_init(); 1143 1144 /* Initialize the kqueue structures */ 1145 nfs_kqinit(); 1146 /* Initialize the iod structures */ 1147 nfs_iodinit(); 1148 1149 scale = PAGE_SHIFT - 4; 1150 nfs_commitsize = uimin(uvmexp.npages, INT_MAX >> scale) << scale; 1151} 1152 1153void 1154nfs_vfs_done(void) 1155{ 1156 1157 nfs_node_done(); 1158 nfs_kqfini(); 1159 nfs_iodfini(); 1160 nfs_fini(); 1161} 1162