1 /* 2 * Copyright (c) 1989 The Regents of the University of California. 3 * All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Rick Macklem at The University of Guelph. 7 * 8 * %sccs.include.redist.c% 9 * 10 * @(#)nfs_vfsops.c 7.31 (Berkeley) 05/06/91 11 */ 12 13 #include "param.h" 14 #include "conf.h" 15 #include "ioctl.h" 16 #include "signal.h" 17 #include "proc.h" 18 #include "namei.h" 19 #include "vnode.h" 20 #include "mount.h" 21 #include "buf.h" 22 #include "mbuf.h" 23 #include "socket.h" 24 #include "systm.h" 25 26 #include "../net/if.h" 27 #include "../net/route.h" 28 #include "../netinet/in.h" 29 30 #include "nfsv2.h" 31 #include "nfsnode.h" 32 #include "nfsmount.h" 33 #include "nfs.h" 34 #include "xdr_subs.h" 35 #include "nfsm_subs.h" 36 #include "nfsdiskless.h" 37 38 /* 39 * nfs vfs operations. 40 */ 41 struct vfsops nfs_vfsops = { 42 nfs_mount, 43 nfs_start, 44 nfs_unmount, 45 nfs_root, 46 nfs_quotactl, 47 nfs_statfs, 48 nfs_sync, 49 nfs_fhtovp, 50 nfs_vptofh, 51 nfs_init, 52 }; 53 54 static u_char nfs_mntid; 55 extern u_long nfs_procids[NFS_NPROCS]; 56 extern u_long nfs_prog, nfs_vers; 57 struct nfs_diskless nfs_diskless; 58 void nfs_disconnect(); 59 60 #define TRUE 1 61 #define FALSE 0 62 63 /* 64 * nfs statfs call 65 */ 66 nfs_statfs(mp, sbp, p) 67 struct mount *mp; 68 register struct statfs *sbp; 69 struct proc *p; 70 { 71 register struct vnode *vp; 72 register struct nfsv2_statfs *sfp; 73 register caddr_t cp; 74 register long t1; 75 caddr_t bpos, dpos, cp2; 76 u_long xid; 77 int error = 0; 78 struct mbuf *mreq, *mrep, *md, *mb, *mb2; 79 struct nfsmount *nmp; 80 struct ucred *cred; 81 struct nfsnode *np; 82 83 nmp = VFSTONFS(mp); 84 if (error = nfs_nget(mp, &nmp->nm_fh, &np)) 85 return (error); 86 vp = NFSTOV(np); 87 nfsstats.rpccnt[NFSPROC_STATFS]++; 88 cred = crget(); 89 cred->cr_ngroups = 1; 90 nfsm_reqhead(nfs_procids[NFSPROC_STATFS], cred, NFSX_FH); 91 nfsm_fhtom(vp); 92 nfsm_request(vp, NFSPROC_STATFS, p, 0); 93 nfsm_disect(sfp, struct nfsv2_statfs *, NFSX_STATFS); 94 sbp->f_type = MOUNT_NFS; 95 sbp->f_flags = nmp->nm_flag; 96 sbp->f_bsize = fxdr_unsigned(long, sfp->sf_tsize); 97 sbp->f_fsize = fxdr_unsigned(long, sfp->sf_bsize); 98 sbp->f_blocks = fxdr_unsigned(long, sfp->sf_blocks); 99 sbp->f_bfree = fxdr_unsigned(long, sfp->sf_bfree); 100 sbp->f_bavail = fxdr_unsigned(long, sfp->sf_bavail); 101 sbp->f_files = 0; 102 sbp->f_ffree = 0; 103 if (sbp != &mp->mnt_stat) { 104 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN); 105 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN); 106 } 107 nfsm_reqdone; 108 nfs_nput(vp); 109 crfree(cred); 110 return (error); 111 } 112 113 /* 114 * Mount a remote root fs via. nfs. This depends on the info in the 115 * nfs_diskless structure that has been filled in properly by some primary 116 * bootstrap. 117 * It goes something like this: 118 * - do enough of "ifconfig" by calling ifioctl() so that the system 119 * can talk to the server 120 * - If nfs_diskless.mygateway is filled in, use that address as 121 * a default gateway. 122 * (This is done the 4.3 way with rtioctl() and should be changed) 123 * - hand craft the swap nfs vnode hanging off a fake mount point 124 * - build the rootfs mount point and call mountnfs() to do the rest. 125 */ 126 nfs_mountroot() 127 { 128 register struct mount *mp; 129 register struct mbuf *m; 130 struct socket *so; 131 struct vnode *vp; 132 int error; 133 134 /* 135 * Do enough of ifconfig(8) so that critical net interface can 136 * talk to the server. 137 */ 138 if (socreate(nfs_diskless.myif.ifra_addr.sa_family, &so, SOCK_DGRAM, 0)) 139 panic("nfs ifconf"); 140 if (ifioctl(so, SIOCAIFADDR, &nfs_diskless.myif)) 141 panic("nfs ifconf2"); 142 soclose(so); 143 144 /* 145 * If the gateway field is filled in, set it as the default route. 146 */ 147 #ifdef COMPAT_43 148 if (nfs_diskless.mygateway.sa_family == AF_INET) { 149 struct ortentry rt; 150 struct sockaddr_in *sin; 151 152 sin = (struct sockaddr_in *) &rt.rt_dst; 153 sin->sin_len = sizeof (struct sockaddr_in); 154 sin->sin_family = AF_INET; 155 sin->sin_addr.s_addr = 0; /* default */ 156 bcopy((caddr_t)&nfs_diskless.mygateway, (caddr_t)&rt.rt_gateway, 157 sizeof (struct sockaddr_in)); 158 rt.rt_flags = (RTF_UP | RTF_GATEWAY); 159 if (rtioctl(SIOCADDRT, (caddr_t)&rt)) 160 panic("nfs root route"); 161 } 162 #endif /* COMPAT_43 */ 163 164 /* 165 * If swapping to an nfs node (indicated by swdevt[0].sw_dev == NODEV): 166 * Create a fake mount point just for the swap vnode so that the 167 * swap file can be on a different server from the rootfs. 168 */ 169 if (swdevt[0].sw_dev == NODEV) { 170 mp = (struct mount *)malloc((u_long)sizeof(struct mount), 171 M_MOUNT, M_NOWAIT); 172 if (mp == NULL) 173 panic("nfs root mount"); 174 mp->mnt_op = &nfs_vfsops; 175 mp->mnt_flag = 0; 176 mp->mnt_exroot = 0; 177 mp->mnt_mounth = NULLVP; 178 179 /* 180 * Set up the diskless nfs_args for the swap mount point 181 * and then call mountnfs() to mount it. 182 * Since the swap file is not the root dir of a file system, 183 * hack it to a regular file. 184 */ 185 nfs_diskless.swap_args.fh = (nfsv2fh_t *)nfs_diskless.swap_fh; 186 MGET(m, MT_SONAME, M_DONTWAIT); 187 if (m == NULL) 188 panic("nfs root mbuf"); 189 bcopy((caddr_t)&nfs_diskless.swap_saddr, mtod(m, caddr_t), 190 nfs_diskless.swap_saddr.sa_len); 191 m->m_len = nfs_diskless.swap_saddr.sa_len; 192 if (mountnfs(&nfs_diskless.swap_args, mp, m, "/swap", 193 nfs_diskless.swap_hostnam, &vp)) 194 panic("nfs swap"); 195 vp->v_type = VREG; 196 vp->v_flag = 0; 197 swapdev_vp = vp; 198 VREF(vp); 199 swdevt[0].sw_vp = vp; 200 } 201 202 /* 203 * Create the rootfs mount point. 204 */ 205 mp = (struct mount *)malloc((u_long)sizeof(struct mount), 206 M_MOUNT, M_NOWAIT); 207 if (mp == NULL) 208 panic("nfs root mount2"); 209 mp->mnt_op = &nfs_vfsops; 210 mp->mnt_flag = MNT_RDONLY; 211 mp->mnt_exroot = 0; 212 mp->mnt_mounth = NULLVP; 213 214 /* 215 * Set up the root fs args and call mountnfs() to do the rest. 216 */ 217 nfs_diskless.root_args.fh = (nfsv2fh_t *)nfs_diskless.root_fh; 218 MGET(m, MT_SONAME, M_DONTWAIT); 219 if (m == NULL) 220 panic("nfs root mbuf2"); 221 bcopy((caddr_t)&nfs_diskless.root_saddr, mtod(m, caddr_t), 222 nfs_diskless.root_saddr.sa_len); 223 m->m_len = nfs_diskless.root_saddr.sa_len; 224 if (mountnfs(&nfs_diskless.root_args, mp, m, "/", 225 nfs_diskless.root_hostnam, &vp)) 226 panic("nfs root"); 227 if (vfs_lock(mp)) 228 panic("nfs root2"); 229 rootfs = mp; 230 mp->mnt_next = mp; 231 mp->mnt_prev = mp; 232 mp->mnt_vnodecovered = NULLVP; 233 vfs_unlock(mp); 234 rootvp = vp; 235 inittodr((time_t)0); /* There is no time in the nfs fsstat so ?? */ 236 return (0); 237 } 238 239 /* 240 * VFS Operations. 241 * 242 * mount system call 243 * It seems a bit dumb to copyinstr() the host and path here and then 244 * bcopy() them in mountnfs(), but I wanted to detect errors before 245 * doing the sockargs() call because sockargs() allocates an mbuf and 246 * an error after that means that I have to release the mbuf. 247 */ 248 /* ARGSUSED */ 249 nfs_mount(mp, path, data, ndp, p) 250 struct mount *mp; 251 char *path; 252 caddr_t data; 253 struct nameidata *ndp; 254 struct proc *p; 255 { 256 int error; 257 struct nfs_args args; 258 struct mbuf *nam; 259 struct vnode *vp; 260 char pth[MNAMELEN], hst[MNAMELEN]; 261 u_int len; 262 nfsv2fh_t nfh; 263 264 if (mp->mnt_flag & MNT_UPDATE) 265 return (0); 266 if (error = copyin(data, (caddr_t)&args, sizeof (struct nfs_args))) 267 return (error); 268 if (error = copyin((caddr_t)args.fh, (caddr_t)&nfh, sizeof (nfsv2fh_t))) 269 return (error); 270 if (error = copyinstr(path, pth, MNAMELEN-1, &len)) 271 return (error); 272 bzero(&pth[len], MNAMELEN - len); 273 if (error = copyinstr(args.hostname, hst, MNAMELEN-1, &len)) 274 return (error); 275 bzero(&hst[len], MNAMELEN - len); 276 /* sockargs() call must be after above copyin() calls */ 277 if (error = sockargs(&nam, (caddr_t)args.addr, 278 sizeof (struct sockaddr), MT_SONAME)) 279 return (error); 280 args.fh = &nfh; 281 error = mountnfs(&args, mp, nam, pth, hst, &vp); 282 return (error); 283 } 284 285 /* 286 * Common code for mount and mountroot 287 */ 288 mountnfs(argp, mp, nam, pth, hst, vpp) 289 register struct nfs_args *argp; 290 register struct mount *mp; 291 struct mbuf *nam; 292 char *pth, *hst; 293 struct vnode **vpp; 294 { 295 register struct nfsmount *nmp; 296 struct proc *p = curproc; /* XXX */ 297 struct nfsnode *np; 298 int error; 299 fsid_t tfsid; 300 301 MALLOC(nmp, struct nfsmount *, sizeof *nmp, M_NFSMNT, M_WAITOK); 302 bzero((caddr_t)nmp, sizeof *nmp); 303 mp->mnt_data = (qaddr_t)nmp; 304 /* 305 * Generate a unique nfs mount id. The problem is that a dev number 306 * is not unique across multiple systems. The techique is as follows: 307 * 1) Set to nblkdev,0 which will never be used otherwise 308 * 2) Generate a first guess as nblkdev,nfs_mntid where nfs_mntid is 309 * NOT 0 310 * 3) Loop searching the mount list for another one with same id 311 * If a match, increment val[0] and try again 312 * NB: I increment val[0] { a long } instead of nfs_mntid { a u_char } 313 * so that nfs is not limited to 255 mount points 314 * Incrementing the high order bits does no real harm, since it 315 * simply makes the major dev number tick up. The upper bound is 316 * set to major dev 127 to avoid any sign extention problems 317 */ 318 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev, 0); 319 mp->mnt_stat.f_fsid.val[1] = MOUNT_NFS; 320 if (++nfs_mntid == 0) 321 ++nfs_mntid; 322 tfsid.val[0] = makedev(nblkdev, nfs_mntid); 323 tfsid.val[1] = MOUNT_NFS; 324 while (rootfs && getvfs(&tfsid)) { 325 tfsid.val[0]++; 326 nfs_mntid++; 327 } 328 if (major(tfsid.val[0]) > 127) { 329 error = ENOENT; 330 goto bad; 331 } 332 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 333 nmp->nm_mountp = mp; 334 nmp->nm_flag = argp->flags; 335 nmp->nm_rto = NFS_TIMEO; 336 nmp->nm_rtt = -1; 337 nmp->nm_rttvar = nmp->nm_rto << 1; 338 nmp->nm_retry = NFS_RETRANS; 339 nmp->nm_wsize = NFS_WSIZE; 340 nmp->nm_rsize = NFS_RSIZE; 341 bcopy((caddr_t)argp->fh, (caddr_t)&nmp->nm_fh, sizeof(nfsv2fh_t)); 342 mp->mnt_stat.f_type = MOUNT_NFS; 343 bcopy(hst, mp->mnt_stat.f_mntfromname, MNAMELEN); 344 bcopy(pth, mp->mnt_stat.f_mntonname, MNAMELEN); 345 nmp->nm_nam = nam; 346 347 if ((argp->flags & NFSMNT_TIMEO) && argp->timeo > 0) { 348 nmp->nm_rto = argp->timeo; 349 /* NFS timeouts are specified in 1/10 sec. */ 350 nmp->nm_rto = (nmp->nm_rto * 10) / NFS_HZ; 351 if (nmp->nm_rto < NFS_MINTIMEO) 352 nmp->nm_rto = NFS_MINTIMEO; 353 else if (nmp->nm_rto > NFS_MAXTIMEO) 354 nmp->nm_rto = NFS_MAXTIMEO; 355 nmp->nm_rttvar = nmp->nm_rto << 1; 356 } 357 358 if ((argp->flags & NFSMNT_RETRANS) && argp->retrans > 1) { 359 nmp->nm_retry = argp->retrans; 360 if (nmp->nm_retry > NFS_MAXREXMIT) 361 nmp->nm_retry = NFS_MAXREXMIT; 362 } 363 364 if ((argp->flags & NFSMNT_WSIZE) && argp->wsize > 0) { 365 nmp->nm_wsize = argp->wsize; 366 /* Round down to multiple of blocksize */ 367 nmp->nm_wsize &= ~0x1ff; 368 if (nmp->nm_wsize <= 0) 369 nmp->nm_wsize = 512; 370 else if (nmp->nm_wsize > NFS_MAXDATA) 371 nmp->nm_wsize = NFS_MAXDATA; 372 } 373 if (nmp->nm_wsize > MAXBSIZE) 374 nmp->nm_wsize = MAXBSIZE; 375 376 if ((argp->flags & NFSMNT_RSIZE) && argp->rsize > 0) { 377 nmp->nm_rsize = argp->rsize; 378 /* Round down to multiple of blocksize */ 379 nmp->nm_rsize &= ~0x1ff; 380 if (nmp->nm_rsize <= 0) 381 nmp->nm_rsize = 512; 382 else if (nmp->nm_rsize > NFS_MAXDATA) 383 nmp->nm_rsize = NFS_MAXDATA; 384 } 385 if (nmp->nm_rsize > MAXBSIZE) 386 nmp->nm_rsize = MAXBSIZE; 387 /* Set up the sockets and per-host congestion */ 388 nmp->nm_sotype = argp->sotype; 389 nmp->nm_soproto = argp->proto; 390 if (error = nfs_connect(nmp)) 391 goto bad; 392 393 if (error = nfs_statfs(mp, &mp->mnt_stat, p)) 394 goto bad; 395 /* 396 * A reference count is needed on the nfsnode representing the 397 * remote root. If this object is not persistent, then backward 398 * traversals of the mount point (i.e. "..") will not work if 399 * the nfsnode gets flushed out of the cache. Ufs does not have 400 * this problem, because one can identify root inodes by their 401 * number == ROOTINO (2). 402 */ 403 if (error = nfs_nget(mp, &nmp->nm_fh, &np)) 404 goto bad; 405 /* 406 * Unlock it, but keep the reference count. 407 */ 408 nfs_unlock(NFSTOV(np)); 409 *vpp = NFSTOV(np); 410 411 return (0); 412 bad: 413 nfs_disconnect(nmp); 414 FREE(nmp, M_NFSMNT); 415 m_freem(nam); 416 return (error); 417 } 418 419 /* 420 * unmount system call 421 */ 422 nfs_unmount(mp, mntflags, p) 423 struct mount *mp; 424 int mntflags; 425 struct proc *p; 426 { 427 register struct nfsmount *nmp; 428 struct nfsnode *np; 429 struct vnode *vp; 430 int error, flags = 0; 431 extern int doforce; 432 433 if (mntflags & MNT_FORCE) { 434 if (!doforce || mp == rootfs) 435 return (EINVAL); 436 flags |= FORCECLOSE; 437 } 438 nmp = VFSTONFS(mp); 439 /* 440 * Clear out the buffer cache 441 */ 442 mntflushbuf(mp, 0); 443 if (mntinvalbuf(mp)) 444 return (EBUSY); 445 /* 446 * Goes something like this.. 447 * - Check for activity on the root vnode (other than ourselves). 448 * - Call vflush() to clear out vnodes for this file system, 449 * except for the root vnode. 450 * - Decrement reference on the vnode representing remote root. 451 * - Close the socket 452 * - Free up the data structures 453 */ 454 /* 455 * We need to decrement the ref. count on the nfsnode representing 456 * the remote root. See comment in mountnfs(). The VFS unmount() 457 * has done vput on this vnode, otherwise we would get deadlock! 458 */ 459 if (error = nfs_nget(mp, &nmp->nm_fh, &np)) 460 return(error); 461 vp = NFSTOV(np); 462 if (vp->v_usecount > 2) { 463 vput(vp); 464 return (EBUSY); 465 } 466 if (error = vflush(mp, vp, flags)) { 467 vput(vp); 468 return (error); 469 } 470 /* 471 * Get rid of two reference counts, and unlock it on the second. 472 */ 473 vrele(vp); 474 vput(vp); 475 nfs_disconnect(nmp); 476 m_freem(nmp->nm_nam); 477 free((caddr_t)nmp, M_NFSMNT); 478 return (0); 479 } 480 481 /* 482 * Return root of a filesystem 483 */ 484 nfs_root(mp, vpp) 485 struct mount *mp; 486 struct vnode **vpp; 487 { 488 register struct vnode *vp; 489 struct nfsmount *nmp; 490 struct nfsnode *np; 491 int error; 492 493 nmp = VFSTONFS(mp); 494 if (error = nfs_nget(mp, &nmp->nm_fh, &np)) 495 return (error); 496 vp = NFSTOV(np); 497 vp->v_type = VDIR; 498 vp->v_flag = VROOT; 499 *vpp = vp; 500 return (0); 501 } 502 503 extern int syncprt; 504 505 /* 506 * Flush out the buffer cache 507 */ 508 /* ARGSUSED */ 509 nfs_sync(mp, waitfor) 510 struct mount *mp; 511 int waitfor; 512 { 513 if (syncprt) 514 bufstats(); 515 /* 516 * Force stale buffer cache information to be flushed. 517 */ 518 mntflushbuf(mp, waitfor == MNT_WAIT ? B_SYNC : 0); 519 return (0); 520 } 521 522 /* 523 * At this point, this should never happen 524 */ 525 /* ARGSUSED */ 526 nfs_fhtovp(mp, fhp, vpp) 527 struct mount *mp; 528 struct fid *fhp; 529 struct vnode **vpp; 530 { 531 532 return (EINVAL); 533 } 534 535 /* 536 * Vnode pointer to File handle, should never happen either 537 */ 538 /* ARGSUSED */ 539 nfs_vptofh(vp, fhp) 540 struct vnode *vp; 541 struct fid *fhp; 542 { 543 544 return (EINVAL); 545 } 546 547 /* 548 * Vfs start routine, a no-op. 549 */ 550 /* ARGSUSED */ 551 nfs_start(mp, flags, p) 552 struct mount *mp; 553 int flags; 554 struct proc *p; 555 { 556 557 return (0); 558 } 559 560 /* 561 * Do operations associated with quotas, not supported 562 */ 563 nfs_quotactl(mp, cmd, uid, arg, p) 564 struct mount *mp; 565 int cmd; 566 uid_t uid; 567 caddr_t arg; 568 struct proc *p; 569 { 570 #ifdef lint 571 mp = mp; cmd = cmd; uid = uid; arg = arg; 572 #endif /* lint */ 573 return (EOPNOTSUPP); 574 } 575