xref: /freebsd/stand/i386/libi386/pxe.c (revision bf2fa8d9d11c9f2ceff09bacc406876fa37096be) 
1 /*-
2  * Copyright (c) 2000 Alfred Perlstein <alfred@freebsd.org>
3  * Copyright (c) 2000 Paul Saab <ps@freebsd.org>
4  * All rights reserved.
5  * Copyright (c) 2000 John Baldwin <jhb@freebsd.org>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <stand.h>
33 #include <errno.h>
34 #include <stdbool.h>
35 #include <stddef.h>
36 #include <string.h>
37 #include <stdarg.h>
38 #include <sys/param.h>
39 
40 #include <net/ethernet.h>
41 #include <netinet/in_systm.h>
42 #include <netinet/in.h>
43 #include <netinet/ip.h>
44 #include <netinet/udp.h>
45 
46 #include <net.h>
47 #include <netif.h>
48 #include <nfsv2.h>
49 #include <iodesc.h>
50 
51 #include <bootp.h>
52 #include <bootstrap.h>
53 #include "libi386.h"
54 #include "btxv86.h"
55 #include "pxe.h"
56 
57 static pxenv_t *pxenv_p = NULL;	/* PXENV+ */
58 static pxe_t *pxe_p = NULL;		/* !PXE */
59 
60 #ifdef PXE_DEBUG
61 static int	pxe_debug = 0;
62 #endif
63 
64 void		pxe_enable(void *pxeinfo);
65 static void	(*pxe_call)(int func, void *ptr);
66 static void	pxenv_call(int func, void *ptr);
67 static void	bangpxe_call(int func, void *ptr);
68 
69 static int	pxe_init(void);
70 static int	pxe_print(int verbose);
71 static void	pxe_cleanup(void);
72 
73 static void	pxe_perror(int error);
74 static int	pxe_netif_match(struct netif *nif, void *machdep_hint);
75 static int	pxe_netif_probe(struct netif *nif, void *machdep_hint);
76 static void	pxe_netif_init(struct iodesc *desc, void *machdep_hint);
77 static ssize_t	pxe_netif_get(struct iodesc *, void **, time_t);
78 static ssize_t	pxe_netif_put(struct iodesc *desc, void *pkt, size_t len);
79 static void	pxe_netif_end(struct netif *nif);
80 
81 extern struct netif_stats	pxe_st[];
82 extern uint16_t			__bangpxeseg;
83 extern uint16_t			__bangpxeoff;
84 extern void			__bangpxeentry(void);
85 extern uint16_t			__pxenvseg;
86 extern uint16_t			__pxenvoff;
87 extern void			__pxenventry(void);
88 
89 struct netif_dif pxe_ifs[] = {
90 /*	dif_unit        dif_nsel        dif_stats       dif_private     */
91 	{0,             1,              &pxe_st[0],     0}
92 };
93 
94 struct netif_stats pxe_st[nitems(pxe_ifs)];
95 
96 struct netif_driver pxenetif = {
97 	.netif_bname = "pxenet",
98 	.netif_match = pxe_netif_match,
99 	.netif_probe = pxe_netif_probe,
100 	.netif_init = pxe_netif_init,
101 	.netif_get = pxe_netif_get,
102 	.netif_put = pxe_netif_put,
103 	.netif_end = pxe_netif_end,
104 	.netif_ifs = pxe_ifs,
105 	.netif_nifs = nitems(pxe_ifs)
106 };
107 
108 struct netif_driver *netif_drivers[] = {
109 	&pxenetif,
110 	NULL
111 };
112 
113 struct devsw pxedisk = {
114 	.dv_name = "net",
115 	.dv_type = DEVT_NET,
116 	.dv_init = pxe_init,
117 	.dv_strategy = NULL,	/* Will be set in pxe_init */
118 	.dv_open = NULL,	/* Will be set in pxe_init */
119 	.dv_close = NULL,	/* Will be set in pxe_init */
120 	.dv_ioctl = noioctl,
121 	.dv_print = pxe_print,
122 	.dv_cleanup = pxe_cleanup
123 };
124 
125 /*
126  * This function is called by the loader to enable PXE support if we
127  * are booted by PXE. The passed in pointer is a pointer to the PXENV+
128  * structure.
129  */
130 void
131 pxe_enable(void *pxeinfo)
132 {
133 	pxenv_p  = (pxenv_t *)pxeinfo;
134 	pxe_p    = (pxe_t *)PTOV(pxenv_p->PXEPtr.segment * 16 +
135 				 pxenv_p->PXEPtr.offset);
136 	pxe_call = NULL;
137 }
138 
139 /*
140  * return true if pxe structures are found/initialized,
141  * also figures out our IP information via the pxe cached info struct
142  */
143 static int
144 pxe_init(void)
145 {
146 	t_PXENV_GET_CACHED_INFO *gci_p;
147 	int counter;
148 	uint8_t checksum;
149 	uint8_t *checkptr;
150 	extern struct devsw netdev;
151 
152 	if (pxenv_p == NULL)
153 		return (0);
154 
155 	/* look for "PXENV+" */
156 	if (bcmp((void *)pxenv_p->Signature, S_SIZE("PXENV+"))) {
157 		pxenv_p = NULL;
158 		return (0);
159 	}
160 
161 	/* make sure the size is something we can handle */
162 	if (pxenv_p->Length > sizeof(*pxenv_p)) {
163 		printf("PXENV+ structure too large, ignoring\n");
164 		pxenv_p = NULL;
165 		return (0);
166 	}
167 
168 	/*
169 	 * do byte checksum:
170 	 * add up each byte in the structure, the total should be 0
171 	 */
172 	checksum = 0;
173 	checkptr = (uint8_t *) pxenv_p;
174 	for (counter = 0; counter < pxenv_p->Length; counter++)
175 		checksum += *checkptr++;
176 	if (checksum != 0) {
177 		printf("PXENV+ structure failed checksum, ignoring\n");
178 		pxenv_p = NULL;
179 		return (0);
180 	}
181 
182 	/*
183 	 * PXENV+ passed, so use that if !PXE is not available or
184 	 * the checksum fails.
185 	 */
186 	pxe_call = pxenv_call;
187 	if (pxenv_p->Version >= 0x0200) {
188 		for (;;) {
189 			if (bcmp((void *)pxe_p->Signature, S_SIZE("!PXE"))) {
190 				pxe_p = NULL;
191 				break;
192 			}
193 			checksum = 0;
194 			checkptr = (uint8_t *)pxe_p;
195 			for (counter = 0; counter < pxe_p->StructLength;
196 			    counter++)
197 				checksum += *checkptr++;
198 			if (checksum != 0) {
199 				pxe_p = NULL;
200 				break;
201 			}
202 			pxe_call = bangpxe_call;
203 			break;
204 		}
205 	}
206 
207 	pxedisk.dv_open = netdev.dv_open;
208 	pxedisk.dv_close = netdev.dv_close;
209 	pxedisk.dv_strategy = netdev.dv_strategy;
210 
211 	printf("\nPXE version %d.%d, real mode entry point ",
212 	    (uint8_t) (pxenv_p->Version >> 8),
213 	    (uint8_t) (pxenv_p->Version & 0xFF));
214 	if (pxe_call == bangpxe_call)
215 		printf("@%04x:%04x\n",
216 		    pxe_p->EntryPointSP.segment,
217 		    pxe_p->EntryPointSP.offset);
218 	else
219 		printf("@%04x:%04x\n",
220 		    pxenv_p->RMEntry.segment, pxenv_p->RMEntry.offset);
221 
222 	gci_p = bio_alloc(sizeof(*gci_p));
223 	if (gci_p == NULL) {
224 		pxe_p = NULL;
225 		return (0);
226 	}
227 	bzero(gci_p, sizeof(*gci_p));
228 	gci_p->PacketType = PXENV_PACKET_TYPE_BINL_REPLY;
229 	pxe_call(PXENV_GET_CACHED_INFO, gci_p);
230 	if (gci_p->Status != 0) {
231 		pxe_perror(gci_p->Status);
232 		bio_free(gci_p, sizeof(*gci_p));
233 		pxe_p = NULL;
234 		return (0);
235 	}
236 	free(bootp_response);
237 	if ((bootp_response = malloc(gci_p->BufferSize)) != NULL) {
238 		bootp_response_size = gci_p->BufferSize;
239 		bcopy(PTOV((gci_p->Buffer.segment << 4) + gci_p->Buffer.offset),
240 		    bootp_response, bootp_response_size);
241 	}
242 	bio_free(gci_p, sizeof(*gci_p));
243 	return (1);
244 }
245 
246 static int
247 pxe_print(int verbose)
248 {
249 	if (pxe_call == NULL)
250 		return (0);
251 
252 	printf("%s devices:", pxedisk.dv_name);
253 	if (pager_output("\n") != 0)
254 		return (1);
255 	printf("    %s0:", pxedisk.dv_name);
256 	if (verbose) {
257 		printf("    %s:%s", inet_ntoa(rootip), rootpath);
258 	}
259 	return (pager_output("\n"));
260 }
261 
262 static void
263 pxe_cleanup(void)
264 {
265 	t_PXENV_UNLOAD_STACK *unload_stack_p;
266 	t_PXENV_UNDI_SHUTDOWN *undi_shutdown_p;
267 
268 	if (pxe_call == NULL)
269 		return;
270 
271 	undi_shutdown_p = bio_alloc(sizeof(*undi_shutdown_p));
272 	if (undi_shutdown_p != NULL) {
273 		bzero(undi_shutdown_p, sizeof(*undi_shutdown_p));
274 		pxe_call(PXENV_UNDI_SHUTDOWN, undi_shutdown_p);
275 
276 #ifdef PXE_DEBUG
277 		if (pxe_debug && undi_shutdown_p->Status != 0)
278 			printf("pxe_cleanup: UNDI_SHUTDOWN failed %x\n",
279 			    undi_shutdown_p->Status);
280 #endif
281 		bio_free(undi_shutdown_p, sizeof(*undi_shutdown_p));
282 	}
283 
284 	unload_stack_p = bio_alloc(sizeof(*unload_stack_p));
285 	if (unload_stack_p != NULL) {
286 		bzero(unload_stack_p, sizeof(*unload_stack_p));
287 		pxe_call(PXENV_UNLOAD_STACK, unload_stack_p);
288 
289 #ifdef PXE_DEBUG
290 		if (pxe_debug && unload_stack_p->Status != 0)
291 			printf("pxe_cleanup: UNLOAD_STACK failed %x\n",
292 			    unload_stack_p->Status);
293 #endif
294 		bio_free(unload_stack_p, sizeof(*unload_stack_p));
295 	}
296 }
297 
298 void
299 pxe_perror(int err)
300 {
301 	return;
302 }
303 
304 void
305 pxenv_call(int func, void *ptr)
306 {
307 #ifdef PXE_DEBUG
308 	if (pxe_debug)
309 		printf("pxenv_call %x\n", func);
310 #endif
311 
312 	bzero(&v86, sizeof(v86));
313 
314 	__pxenvseg = pxenv_p->RMEntry.segment;
315 	__pxenvoff = pxenv_p->RMEntry.offset;
316 
317 	v86.ctl  = V86_ADDR | V86_CALLF | V86_FLAGS;
318 	v86.es   = VTOPSEG(ptr);
319 	v86.edi  = VTOPOFF(ptr);
320 	v86.addr = (VTOPSEG(__pxenventry) << 16) | VTOPOFF(__pxenventry);
321 	v86.ebx  = func;
322 	v86int();
323 	v86.ctl  = V86_FLAGS;
324 }
325 
326 void
327 bangpxe_call(int func, void *ptr)
328 {
329 #ifdef PXE_DEBUG
330 	if (pxe_debug)
331 		printf("bangpxe_call %x\n", func);
332 #endif
333 
334 	bzero(&v86, sizeof(v86));
335 
336 	__bangpxeseg = pxe_p->EntryPointSP.segment;
337 	__bangpxeoff = pxe_p->EntryPointSP.offset;
338 
339 	v86.ctl  = V86_ADDR | V86_CALLF | V86_FLAGS;
340 	v86.edx  = VTOPSEG(ptr);
341 	v86.eax  = VTOPOFF(ptr);
342 	v86.addr = (VTOPSEG(__bangpxeentry) << 16) | VTOPOFF(__bangpxeentry);
343 	v86.ebx  = func;
344 	v86int();
345 	v86.ctl  = V86_FLAGS;
346 }
347 
348 
349 static int
350 pxe_netif_match(struct netif *nif, void *machdep_hint)
351 {
352 	return (1);
353 }
354 
355 static int
356 pxe_netif_probe(struct netif *nif, void *machdep_hint)
357 {
358 	if (pxe_call == NULL)
359 		return (-1);
360 
361 	return (0);
362 }
363 
364 static void
365 pxe_netif_end(struct netif *nif)
366 {
367 	t_PXENV_UNDI_CLOSE *undi_close_p;
368 
369 	undi_close_p = bio_alloc(sizeof(*undi_close_p));
370 	if (undi_close_p != NULL) {
371 		bzero(undi_close_p, sizeof(*undi_close_p));
372 		pxe_call(PXENV_UNDI_CLOSE, undi_close_p);
373 		if (undi_close_p->Status != 0)
374 			printf("undi close failed: %x\n", undi_close_p->Status);
375 		bio_free(undi_close_p, sizeof(*undi_close_p));
376 	}
377 }
378 
379 static void
380 pxe_netif_init(struct iodesc *desc, void *machdep_hint)
381 {
382 	t_PXENV_UNDI_GET_INFORMATION *undi_info_p;
383 	t_PXENV_UNDI_OPEN *undi_open_p;
384 	uint8_t *mac;
385 	int i, len;
386 
387 	undi_info_p = bio_alloc(sizeof(*undi_info_p));
388 	if (undi_info_p == NULL)
389 		return;
390 
391 	bzero(undi_info_p, sizeof(*undi_info_p));
392 	pxe_call(PXENV_UNDI_GET_INFORMATION, undi_info_p);
393 	if (undi_info_p->Status != 0) {
394 		printf("undi get info failed: %x\n", undi_info_p->Status);
395 		bio_free(undi_info_p, sizeof(*undi_info_p));
396 		return;
397 	}
398 
399 	/* Make sure the CurrentNodeAddress is valid. */
400 	for (i = 0; i < undi_info_p->HwAddrLen; ++i) {
401 		if (undi_info_p->CurrentNodeAddress[i] != 0)
402 			break;
403 	}
404 	if (i < undi_info_p->HwAddrLen) {
405 		for (i = 0; i < undi_info_p->HwAddrLen; ++i) {
406 			if (undi_info_p->CurrentNodeAddress[i] != 0xff)
407 				break;
408 		}
409 	}
410 	if (i < undi_info_p->HwAddrLen)
411 		mac = undi_info_p->CurrentNodeAddress;
412 	else
413 		mac = undi_info_p->PermNodeAddress;
414 
415 	len = min(sizeof (desc->myea), undi_info_p->HwAddrLen);
416 	for (i = 0; i < len; ++i)
417 		desc->myea[i] = mac[i];
418 
419 	if (bootp_response != NULL)
420 		desc->xid = bootp_response->bp_xid;
421 	else
422 		desc->xid = 0;
423 
424 	bio_free(undi_info_p, sizeof(*undi_info_p));
425 	undi_open_p = bio_alloc(sizeof(*undi_open_p));
426 	if (undi_open_p == NULL)
427 		return;
428 	bzero(undi_open_p, sizeof(*undi_open_p));
429 	undi_open_p->PktFilter = FLTR_DIRECTED | FLTR_BRDCST;
430 	pxe_call(PXENV_UNDI_OPEN, undi_open_p);
431 	if (undi_open_p->Status != 0)
432 		printf("undi open failed: %x\n", undi_open_p->Status);
433 	bio_free(undi_open_p, sizeof(*undi_open_p));
434 }
435 
436 static int
437 pxe_netif_receive_isr(t_PXENV_UNDI_ISR *isr, void **pkt, ssize_t *retsize)
438 {
439 	static bool data_pending;
440 	char *buf, *ptr, *frame;
441 	size_t size, rsize;
442 
443 	buf = NULL;
444 	size = rsize = 0;
445 
446 	/*
447 	 * We can save ourselves the next two pxe calls because we already know
448 	 * we weren't done grabbing everything.
449 	 */
450 	if (data_pending) {
451 		data_pending = false;
452 		goto nextbuf;
453 	}
454 
455 	/*
456 	 * We explicitly don't check for OURS/NOT_OURS as a result of START;
457 	 * it's been reported that some cards are known to mishandle these.
458 	 */
459 	bzero(isr, sizeof(*isr));
460 	isr->FuncFlag = PXENV_UNDI_ISR_IN_START;
461 	pxe_call(PXENV_UNDI_ISR, isr);
462 	/* We could translate Status... */
463 	if (isr->Status != 0) {
464 		return (ENXIO);
465 	}
466 
467 	bzero(isr, sizeof(*isr));
468 	isr->FuncFlag = PXENV_UNDI_ISR_IN_PROCESS;
469 	pxe_call(PXENV_UNDI_ISR, isr);
470 	if (isr->Status != 0) {
471 		return (ENXIO);
472 	}
473 	if (isr->FuncFlag == PXENV_UNDI_ISR_OUT_BUSY) {
474 		/*
475 		 * Let the caller decide if we need to be restarted.  It will
476 		 * currently blindly restart us, but it could check timeout in
477 		 * the future.
478 		 */
479 		return (ERESTART);
480 	}
481 
482 	/*
483 	 * By design, we'll hardly ever hit this terminal condition unless we
484 	 * pick up nothing but tx interrupts here.  More frequently, we will
485 	 * process rx buffers until we hit the terminal condition in the middle.
486 	 */
487 	while (isr->FuncFlag != PXENV_UNDI_ISR_OUT_DONE) {
488 		/*
489 		 * This might have given us PXENV_UNDI_ISR_OUT_TRANSMIT, in
490 		 * which case we can just disregard and move on to the next
491 		 * buffer/frame.
492 		 */
493 		if (isr->FuncFlag != PXENV_UNDI_ISR_OUT_RECEIVE)
494 			goto nextbuf;
495 
496 		if (buf == NULL) {
497 			/*
498 			 * Grab size from the first Frame that we picked up,
499 			 * allocate an rx buf to hold.  Careful here, as we may
500 			 * see a fragmented frame that's spread out across
501 			 * multiple GET_NEXT calls.
502 			 */
503 			size = isr->FrameLength;
504 			buf = malloc(size + ETHER_ALIGN);
505 			if (buf == NULL)
506 				return (ENOMEM);
507 
508 			ptr = buf + ETHER_ALIGN;
509 		}
510 
511 		frame = (char *)((uintptr_t)isr->Frame.segment << 4);
512 		frame += isr->Frame.offset;
513 		bcopy(PTOV(frame), ptr, isr->BufferLength);
514 		ptr += isr->BufferLength;
515 		rsize += isr->BufferLength;
516 
517 		/*
518 		 * Stop here before we risk catching the start of another frame.
519 		 * It would be nice to continue reading until we actually get a
520 		 * PXENV_UNDI_ISR_OUT_DONE, but our network stack in libsa isn't
521 		 * suitable for reading more than one packet at a time.
522 		 */
523 		if (rsize >= size) {
524 			data_pending = true;
525 			break;
526 		}
527 
528 nextbuf:
529 		bzero(isr, sizeof(*isr));
530 		isr->FuncFlag = PXENV_UNDI_ISR_IN_GET_NEXT;
531 		pxe_call(PXENV_UNDI_ISR, isr);
532 		if (isr->Status != 0) {
533 			free(buf);
534 			return (ENXIO);
535 		}
536 	}
537 
538 	/*
539 	 * We may have never picked up a frame at all (all tx), in which case
540 	 * the caller should restart us.
541 	 */
542 	if (rsize == 0) {
543 		return (ERESTART);
544 	}
545 
546 	*pkt = buf;
547 	*retsize = rsize;
548 	return (0);
549 }
550 
551 static int
552 pxe_netif_receive(void **pkt, ssize_t *size)
553 {
554 	t_PXENV_UNDI_ISR *isr;
555 	int ret;
556 
557 	isr = bio_alloc(sizeof(*isr));
558 	if (isr == NULL)
559 		return (ENOMEM);
560 
561 	/*
562 	 * This completely ignores the timeout specified in pxe_netif_get(), but
563 	 * we shouldn't be running long enough here for that to make a
564 	 * difference.
565 	 */
566 	for (;;) {
567 		/* We'll only really re-enter for PXENV_UNDI_ISR_OUT_BUSY. */
568 		ret = pxe_netif_receive_isr(isr, pkt, size);
569 		if (ret != ERESTART)
570 			break;
571 	}
572 
573 	bio_free(isr, sizeof(*isr));
574 	return (ret);
575 }
576 
577 static ssize_t
578 pxe_netif_get(struct iodesc *desc, void **pkt, time_t timeout)
579 {
580 	time_t t;
581 	void *ptr;
582 	int ret = -1;
583 	ssize_t size;
584 
585 	t = getsecs();
586 	size = 0;
587 	while ((getsecs() - t) < timeout) {
588 		ret = pxe_netif_receive(&ptr, &size);
589 		if (ret != -1) {
590 			*pkt = ptr;
591 			break;
592 		}
593 	}
594 
595 	return (ret == 0 ? size : -1);
596 }
597 
598 static ssize_t
599 pxe_netif_put(struct iodesc *desc, void *pkt, size_t len)
600 {
601 	t_PXENV_UNDI_TRANSMIT *trans_p;
602 	t_PXENV_UNDI_TBD *tbd_p;
603 	char *data;
604 	ssize_t rv = -1;
605 
606 	trans_p = bio_alloc(sizeof(*trans_p));
607 	tbd_p = bio_alloc(sizeof(*tbd_p));
608 	data = bio_alloc(len);
609 
610 	if (trans_p != NULL && tbd_p != NULL && data != NULL) {
611 		bzero(trans_p, sizeof(*trans_p));
612 		bzero(tbd_p, sizeof(*tbd_p));
613 
614 		trans_p->TBD.segment = VTOPSEG(tbd_p);
615 		trans_p->TBD.offset  = VTOPOFF(tbd_p);
616 
617 		tbd_p->ImmedLength = len;
618 		tbd_p->Xmit.segment = VTOPSEG(data);
619 		tbd_p->Xmit.offset  = VTOPOFF(data);
620 		bcopy(pkt, data, len);
621 
622 		pxe_call(PXENV_UNDI_TRANSMIT, trans_p);
623 		if (trans_p->Status == 0)
624 			rv = len;
625 	}
626 
627 	bio_free(data, len);
628 	bio_free(tbd_p, sizeof(*tbd_p));
629 	bio_free(trans_p, sizeof(*trans_p));
630 	return (rv);
631 }
632