nfs_boot.c revision 1.50.4.1 1 /* $NetBSD: nfs_boot.c,v 1.50.4.1 1999/08/02 22:38:26 thorpej Exp $ */
2
3 /*-
4 * Copyright (c) 1995, 1997 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Adam Glass and Gordon W. Ross.
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. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the NetBSD
21 * Foundation, Inc. and its contributors.
22 * 4. Neither the name of The NetBSD Foundation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 * POSSIBILITY OF SUCH DAMAGE.
37 */
38
39 /*
40 * Support for NFS diskless booting, specifically getting information
41 * about where to mount root from, what pathnames, etc.
42 */
43
44 #include "opt_nfs_boot.h"
45
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/conf.h>
50 #include <sys/device.h>
51 #include <sys/ioctl.h>
52 #include <sys/proc.h>
53 #include <sys/mount.h>
54 #include <sys/mbuf.h>
55 #include <sys/reboot.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
58
59 #include <net/if.h>
60 #include <net/route.h>
61 #include <net/if_ether.h>
62 #include <net/if_types.h>
63
64 #include <netinet/in.h>
65 #include <netinet/if_inarp.h>
66
67 #include <nfs/rpcv2.h>
68 #include <nfs/krpc.h>
69 #include <nfs/xdr_subs.h>
70
71 #include <nfs/nfsproto.h>
72 #include <nfs/nfs.h>
73 #include <nfs/nfsmount.h>
74 #include <nfs/nfsdiskless.h>
75
76 /*
77 * There are two implementations of NFS diskless boot.
78 * One implementation uses BOOTP (RFC951, RFC1048),
79 * the other uses Sun RPC/bootparams. See the files:
80 * nfs_bootp.c: BOOTP (RFC951, RFC1048)
81 * nfs_bootsun.c: Sun RPC/bootparams
82 */
83 #if defined(NFS_BOOT_BOOTP) || defined(NFS_BOOT_DHCP)
84 int nfs_boot_rfc951 = 1; /* BOOTP enabled (default) */
85 #endif
86 #ifdef NFS_BOOT_BOOTPARAM
87 int nfs_boot_bootparam = 1; /* BOOTPARAM enabled (default) */
88 #endif
89
90 /* mountd RPC */
91 static int md_mount __P((struct sockaddr_in *mdsin, char *path,
92 struct nfs_args *argp));
93
94 static void nfs_boot_defrt __P((struct in_addr *));
95 static int nfs_boot_getfh __P((struct nfs_dlmount *ndm));
96
97
98 /*
99 * Called with an empty nfs_diskless struct to be filled in.
100 * Find an interface, determine its ip address (etc.) and
101 * save all the boot parameters in the nfs_diskless struct.
102 */
103 int
104 nfs_boot_init(nd, procp)
105 struct nfs_diskless *nd;
106 struct proc *procp;
107 {
108 struct ifnet *ifp;
109 int error;
110
111 /*
112 * Find the network interface.
113 */
114 ifp = ifunit(root_device->dv_xname);
115 if (ifp == NULL) {
116 printf("nfs_boot: '%s' not found\n",
117 root_device->dv_xname);
118 return (ENXIO);
119 }
120 nd->nd_ifp = ifp;
121
122 error = EADDRNOTAVAIL; /* ??? */
123 #if defined(NFS_BOOT_BOOTP) || defined(NFS_BOOT_DHCP)
124 if (error && nfs_boot_rfc951) {
125 #if defined(NFS_BOOT_DHCP)
126 printf("nfs_boot: trying DHCP/BOOTP\n");
127 #else
128 printf("nfs_boot: trying BOOTP\n");
129 #endif
130 error = nfs_bootdhcp(nd, procp);
131 }
132 #endif
133 #ifdef NFS_BOOT_BOOTPARAM
134 if (error && nfs_boot_bootparam) {
135 printf("nfs_boot: trying RARP (and RPC/bootparam)\n");
136 error = nfs_bootparam(nd, procp);
137 }
138 #endif
139 if (error)
140 return (error);
141
142 /*
143 * If the gateway address is set, add a default route.
144 * (The mountd RPCs may go across a gateway.)
145 */
146 if (nd->nd_gwip.s_addr)
147 nfs_boot_defrt(&nd->nd_gwip);
148
149 /*
150 * Now fetch the NFS file handles as appropriate.
151 */
152 error = nfs_boot_getfh(&nd->nd_root);
153
154 if (error)
155 nfs_boot_cleanup(nd, procp);
156
157 return (error);
158 }
159
160 void
161 nfs_boot_cleanup(nd, procp)
162 struct nfs_diskless *nd;
163 struct proc *procp;
164 {
165
166 nfs_boot_deladdress(nd->nd_ifp, procp, nd->nd_myip.s_addr);
167 nfs_boot_ifupdown(nd->nd_ifp, procp, 0);
168 nfs_boot_flushrt(nd->nd_ifp);
169 }
170
171 int
172 nfs_boot_ifupdown(ifp, procp, up)
173 struct ifnet *ifp;
174 struct proc *procp;
175 int up;
176 {
177 struct socket *so;
178 struct ifreq ireq;
179 int error;
180
181 memset(&ireq, 0, sizeof(ireq));
182 memcpy(ireq.ifr_name, ifp->if_xname, IFNAMSIZ);
183
184 /*
185 * Get a socket to use for various things in here.
186 * After this, use "goto out" to cleanup and return.
187 */
188 error = socreate(AF_INET, &so, SOCK_DGRAM, 0);
189 if (error) {
190 printf("ifupdown: socreate, error=%d\n", error);
191 return (error);
192 }
193
194 /*
195 * Bring up the interface. (just set the "up" flag)
196 * Get the old interface flags and or IFF_UP into them so
197 * things like media selection flags are not clobbered.
198 */
199 error = ifioctl(so, SIOCGIFFLAGS, (caddr_t)&ireq, procp);
200 if (error) {
201 printf("ifupdown: GIFFLAGS, error=%d\n", error);
202 goto out;
203 }
204 if (up)
205 ireq.ifr_flags |= IFF_UP;
206 else
207 ireq.ifr_flags &= ~IFF_UP;
208 error = ifioctl(so, SIOCSIFFLAGS, (caddr_t)&ireq, procp);
209 if (error) {
210 printf("ifupdown: SIFFLAGS, error=%d\n", error);
211 goto out;
212 }
213
214 out:
215 soclose(so);
216 return (error);
217 }
218
219 int
220 nfs_boot_setaddress(ifp, procp, addr, netmask, braddr)
221 struct ifnet *ifp;
222 struct proc *procp;
223 u_int32_t addr, netmask, braddr;
224 {
225 struct socket *so;
226 struct ifaliasreq iareq;
227 struct sockaddr_in *sin;
228 int error;
229
230 /*
231 * Get a socket to use for various things in here.
232 * After this, use "goto out" to cleanup and return.
233 */
234 error = socreate(AF_INET, &so, SOCK_DGRAM, 0);
235 if (error) {
236 printf("setaddress: socreate, error=%d\n", error);
237 return (error);
238 }
239
240 memset(&iareq, 0, sizeof(iareq));
241 memcpy(iareq.ifra_name, ifp->if_xname, IFNAMSIZ);
242
243 /* Set the I/F address */
244 sin = (struct sockaddr_in *)&iareq.ifra_addr;
245 sin->sin_len = sizeof(*sin);
246 sin->sin_family = AF_INET;
247 sin->sin_addr.s_addr = addr;
248
249 /* Set the netmask */
250 if (netmask != INADDR_ANY) {
251 sin = (struct sockaddr_in *)&iareq.ifra_mask;
252 sin->sin_len = sizeof(*sin);
253 sin->sin_family = AF_INET;
254 sin->sin_addr.s_addr = netmask;
255 } /* else leave subnetmask unspecified (len=0) */
256
257 /* Set the broadcast addr. */
258 if (braddr != INADDR_ANY) {
259 sin = (struct sockaddr_in *)&iareq.ifra_broadaddr;
260 sin->sin_len = sizeof(*sin);
261 sin->sin_family = AF_INET;
262 sin->sin_addr.s_addr = braddr;
263 } /* else leave broadcast addr unspecified (len=0) */
264
265 error = ifioctl(so, SIOCAIFADDR, (caddr_t)&iareq, procp);
266 if (error) {
267 printf("setaddress, error=%d\n", error);
268 goto out;
269 }
270
271 out:
272 soclose(so);
273 return (error);
274 }
275
276 int
277 nfs_boot_deladdress(ifp, procp, addr)
278 struct ifnet *ifp;
279 struct proc *procp;
280 u_int32_t addr;
281 {
282 struct socket *so;
283 struct ifreq ireq;
284 struct sockaddr_in *sin;
285 int error;
286
287 /*
288 * Get a socket to use for various things in here.
289 * After this, use "goto out" to cleanup and return.
290 */
291 error = socreate(AF_INET, &so, SOCK_DGRAM, 0);
292 if (error) {
293 printf("deladdress: socreate, error=%d\n", error);
294 return (error);
295 }
296
297 memset(&ireq, 0, sizeof(ireq));
298 memcpy(ireq.ifr_name, ifp->if_xname, IFNAMSIZ);
299
300 sin = (struct sockaddr_in *)&ireq.ifr_addr;
301 sin->sin_len = sizeof(*sin);
302 sin->sin_family = AF_INET;
303 sin->sin_addr.s_addr = addr;
304
305 error = ifioctl(so, SIOCDIFADDR, (caddr_t)&ireq, procp);
306 if (error) {
307 printf("deladdress, error=%d\n", error);
308 goto out;
309 }
310
311 out:
312 soclose(so);
313 return (error);
314 }
315
316 int
317 nfs_boot_setrecvtimo(so)
318 struct socket *so;
319 {
320 struct mbuf *m;
321 struct timeval *tv;
322
323 m = m_get(M_WAIT, MT_SOOPTS);
324 tv = mtod(m, struct timeval *);
325 m->m_len = sizeof(*tv);
326 tv->tv_sec = 1;
327 tv->tv_usec = 0;
328 return (sosetopt(so, SOL_SOCKET, SO_RCVTIMEO, m));
329 }
330
331 int
332 nfs_boot_enbroadcast(so)
333 struct socket *so;
334 {
335 struct mbuf *m;
336 int32_t *on;
337
338 m = m_get(M_WAIT, MT_SOOPTS);
339 on = mtod(m, int32_t *);
340 m->m_len = sizeof(*on);
341 *on = 1;
342 return (sosetopt(so, SOL_SOCKET, SO_BROADCAST, m));
343 }
344
345 int
346 nfs_boot_sobind_ipport(so, port)
347 struct socket *so;
348 u_int16_t port;
349 {
350 struct mbuf *m;
351 struct sockaddr_in *sin;
352 int error;
353
354 m = m_getclr(M_WAIT, MT_SONAME);
355 sin = mtod(m, struct sockaddr_in *);
356 sin->sin_len = m->m_len = sizeof(*sin);
357 sin->sin_family = AF_INET;
358 sin->sin_addr.s_addr = INADDR_ANY;
359 sin->sin_port = htons(port);
360 error = sobind(so, m);
361 m_freem(m);
362 return (error);
363 }
364
365 /*
366 * What is the longest we will wait before re-sending a request?
367 * Note this is also the frequency of "timeout" messages.
368 * The re-send loop counts up linearly to this maximum, so the
369 * first complaint will happen after (1+2+3+4+5)=15 seconds.
370 */
371 #define MAX_RESEND_DELAY 5 /* seconds */
372 #define TOTAL_TIMEOUT 30 /* seconds */
373
374 int
375 nfs_boot_sendrecv(so, nam, sndproc, snd, rcvproc, rcv, from_p, context)
376 struct socket *so;
377 struct mbuf *nam;
378 int (*sndproc) __P((struct mbuf*, void*, int));
379 struct mbuf *snd;
380 int (*rcvproc) __P((struct mbuf*, void*));
381 struct mbuf **rcv, **from_p;
382 void *context;
383 {
384 int error, rcvflg, timo, secs, waited;
385 struct mbuf *m, *from;
386 struct uio uio;
387
388 /* Free at end if not null. */
389 from = NULL;
390
391 /*
392 * Send it, repeatedly, until a reply is received,
393 * but delay each re-send by an increasing amount.
394 * If the delay hits the maximum, start complaining.
395 */
396 waited = timo = 0;
397 send_again:
398 waited += timo;
399 if (waited >= TOTAL_TIMEOUT)
400 return (ETIMEDOUT);
401
402 /* Determine new timeout. */
403 if (timo < MAX_RESEND_DELAY)
404 timo++;
405 else
406 printf("nfs_boot: timeout...\n");
407
408 if (sndproc) {
409 error = (*sndproc)(snd, context, waited);
410 if (error)
411 goto out;
412 }
413
414 /* Send request (or re-send). */
415 m = m_copypacket(snd, M_WAIT);
416 if (m == NULL) {
417 error = ENOBUFS;
418 goto out;
419 }
420 error = (*so->so_send)(so, nam, NULL, m, NULL, 0);
421 if (error) {
422 printf("nfs_boot: sosend: %d\n", error);
423 goto out;
424 }
425 m = NULL;
426
427 /*
428 * Wait for up to timo seconds for a reply.
429 * The socket receive timeout was set to 1 second.
430 */
431
432 secs = timo;
433 for (;;) {
434 if (from) {
435 m_freem(from);
436 from = NULL;
437 }
438 if (m) {
439 m_freem(m);
440 m = NULL;
441 }
442 uio.uio_resid = 1 << 16; /* ??? */
443 rcvflg = 0;
444 error = (*so->so_receive)(so, &from, &uio, &m, NULL, &rcvflg);
445 if (error == EWOULDBLOCK) {
446 if (--secs <= 0)
447 goto send_again;
448 continue;
449 }
450 if (error)
451 goto out;
452 #ifdef DIAGNOSTIC
453 if (!m || !(m->m_flags & M_PKTHDR)
454 || (1 << 16) - uio.uio_resid != m->m_pkthdr.len)
455 panic("nfs_boot_sendrecv: return size");
456 #endif
457
458 if ((*rcvproc)(m, context))
459 continue;
460
461 if (rcv)
462 *rcv = m;
463 else
464 m_freem(m);
465 if (from_p) {
466 *from_p = from;
467 from = NULL;
468 }
469 break;
470 }
471 out:
472 if (from) m_freem(from);
473 return (error);
474 }
475
476 /*
477 * Install a default route to the passed IP address.
478 */
479 static void
480 nfs_boot_defrt(gw_ip)
481 struct in_addr *gw_ip;
482 {
483 struct sockaddr dst, gw, mask;
484 struct sockaddr_in *sin;
485 int error;
486
487 /* Destination: (default) */
488 memset((caddr_t)&dst, 0, sizeof(dst));
489 dst.sa_len = sizeof(dst);
490 dst.sa_family = AF_INET;
491 /* Gateway: */
492 memset((caddr_t)&gw, 0, sizeof(gw));
493 sin = (struct sockaddr_in *)&gw;
494 sin->sin_len = sizeof(*sin);
495 sin->sin_family = AF_INET;
496 sin->sin_addr.s_addr = gw_ip->s_addr;
497 /* Mask: (zero length) */
498 /* XXX - Just pass a null pointer? */
499 memset(&mask, 0, sizeof(mask));
500
501 /* add, dest, gw, mask, flags, 0 */
502 error = rtrequest(RTM_ADD, &dst, &gw, &mask,
503 (RTF_UP | RTF_GATEWAY | RTF_STATIC), NULL);
504 if (error) {
505 printf("nfs_boot: add route, error=%d\n", error);
506 error = 0;
507 }
508 }
509
510 static int nfs_boot_delroute __P((struct radix_node *, void *));
511 static int
512 nfs_boot_delroute(rn, w)
513 struct radix_node *rn;
514 void *w;
515 {
516 struct rtentry *rt = (struct rtentry *)rn;
517 int error;
518
519 if (rt->rt_ifp != (struct ifnet *)w)
520 return (0);
521
522 error = rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 0, NULL);
523 if (error)
524 printf("nfs_boot: del route, error=%d\n", error);
525
526 return (0);
527 }
528
529 void
530 nfs_boot_flushrt(ifp)
531 struct ifnet *ifp;
532 {
533
534 rn_walktree(rt_tables[AF_INET], nfs_boot_delroute, ifp);
535 }
536
537 /*
538 * Get an initial NFS file handle using Sun RPC/mountd.
539 * Separate function because we used to call it twice.
540 * (once for root and once for swap)
541 */
542 static int
543 nfs_boot_getfh(ndm)
544 struct nfs_dlmount *ndm; /* output */
545 {
546 struct nfs_args *args;
547 struct sockaddr_in *sin;
548 char *pathname;
549 int error;
550 u_int16_t port;
551
552 args = &ndm->ndm_args;
553
554 /* Initialize mount args. */
555 memset((caddr_t) args, 0, sizeof(*args));
556 args->addr = &ndm->ndm_saddr;
557 args->addrlen = args->addr->sa_len;
558 #ifdef NFS_BOOT_TCP
559 args->sotype = SOCK_STREAM;
560 #else
561 args->sotype = SOCK_DGRAM;
562 #endif
563 args->fh = ndm->ndm_fh;
564 args->hostname = ndm->ndm_host;
565 args->flags = NFSMNT_NFSV3 | NFSMNT_NOCONN | NFSMNT_RESVPORT;
566
567 #ifdef NFS_BOOT_OPTIONS
568 args->flags |= NFS_BOOT_OPTIONS;
569 #endif
570 #ifdef NFS_BOOT_RWSIZE
571 /*
572 * Reduce rsize,wsize for interfaces that consistently
573 * drop fragments of long UDP messages. (i.e. wd8003).
574 * You can always change these later via remount.
575 */
576 args->flags |= NFSMNT_WSIZE | NFSMNT_RSIZE;
577 args->wsize = NFS_BOOT_RWSIZE;
578 args->rsize = NFS_BOOT_RWSIZE;
579 #endif
580
581 /*
582 * Find the pathname part of the "server:pathname"
583 * string left in ndm->ndm_host by nfs_boot_init.
584 */
585 pathname = strchr(ndm->ndm_host, ':');
586 if (pathname == 0) {
587 printf("nfs_boot: getfh - no pathname\n");
588 return (EIO);
589 }
590 pathname++;
591
592 /*
593 * Get file handle using RPC to mountd/mount
594 */
595 sin = (struct sockaddr_in *)&ndm->ndm_saddr;
596 error = md_mount(sin, pathname, args);
597 if (error) {
598 printf("nfs_boot: mountd `%s', error=%d\n",
599 ndm->ndm_host, error);
600 return (error);
601 }
602
603 /* Set port number for NFS use. */
604 /* XXX: NFS port is always 2049, right? */
605 #ifdef NFS_BOOT_TCP
606 retry:
607 #endif
608 error = krpc_portmap(sin, NFS_PROG,
609 (args->flags & NFSMNT_NFSV3) ? NFS_VER3 : NFS_VER2,
610 (args->sotype == SOCK_STREAM) ? IPPROTO_TCP : IPPROTO_UDP,
611 &port);
612 if (port == htons(0))
613 error = EIO;
614 if (error) {
615 #ifdef NFS_BOOT_TCP
616 if (args->sotype == SOCK_STREAM) {
617 args->sotype = SOCK_DGRAM;
618 goto retry;
619 }
620 #endif
621 printf("nfs_boot: portmap NFS, error=%d\n", error);
622 return (error);
623 }
624 sin->sin_port = port;
625 return (0);
626 }
627
628
629 /*
630 * RPC: mountd/mount
631 * Given a server pathname, get an NFS file handle.
632 * Also, sets sin->sin_port to the NFS service port.
633 */
634 static int
635 md_mount(mdsin, path, argp)
636 struct sockaddr_in *mdsin; /* mountd server address */
637 char *path;
638 struct nfs_args *argp;
639 {
640 /* The RPC structures */
641 struct rdata {
642 u_int32_t errno;
643 union {
644 u_int8_t v2fh[NFSX_V2FH];
645 struct {
646 u_int32_t fhlen;
647 u_int8_t fh[1];
648 } v3fh;
649 } fh;
650 } *rdata;
651 struct mbuf *m;
652 u_int8_t *fh;
653 int minlen, error;
654 int mntver;
655
656 mntver = (argp->flags & NFSMNT_NFSV3) ? 3 : 2;
657 do {
658 /*
659 * Get port number for MOUNTD.
660 */
661 error = krpc_portmap(mdsin, RPCPROG_MNT, mntver,
662 IPPROTO_UDP, &mdsin->sin_port);
663 if (error)
664 continue;
665
666 /* This mbuf is consumed by krpc_call. */
667 m = xdr_string_encode(path, strlen(path));
668 if (m == NULL)
669 return ENOMEM;
670
671 /* Do RPC to mountd. */
672 error = krpc_call(mdsin, RPCPROG_MNT, mntver,
673 RPCMNT_MOUNT, &m, NULL);
674 if (error != EPROGMISMATCH)
675 break;
676 /* Try lower version of mountd. */
677 } while (--mntver >= 1);
678 if (error) {
679 printf("nfs_boot: mountd error=%d\n", error);
680 return error;
681 }
682 if (mntver != 3)
683 argp->flags &= ~NFSMNT_NFSV3;
684
685 /* The reply might have only the errno. */
686 if (m->m_len < 4)
687 goto bad;
688 /* Have at least errno, so check that. */
689 rdata = mtod(m, struct rdata *);
690 error = fxdr_unsigned(u_int32_t, rdata->errno);
691 if (error)
692 goto out;
693
694 /* Have errno==0, so the fh must be there. */
695 if (mntver == 3) {
696 argp->fhsize = fxdr_unsigned(u_int32_t, rdata->fh.v3fh.fhlen);
697 if (argp->fhsize > NFSX_V3FHMAX)
698 goto bad;
699 minlen = 2 * sizeof(u_int32_t) + argp->fhsize;
700 } else {
701 argp->fhsize = NFSX_V2FH;
702 minlen = sizeof(u_int32_t) + argp->fhsize;
703 }
704
705 if (m->m_len < minlen) {
706 m = m_pullup(m, minlen);
707 if (m == NULL)
708 return(EBADRPC);
709 rdata = mtod(m, struct rdata *);
710 }
711
712 fh = (mntver == 3) ?
713 rdata->fh.v3fh.fh : rdata->fh.v2fh;
714 memcpy(argp->fh, fh, argp->fhsize);
715
716 goto out;
717
718 bad:
719 error = EBADRPC;
720
721 out:
722 m_freem(m);
723 return error;
724 }
725