xref: /freebsd/stand/i386/libi386/pxe.c (revision 06c3fb27) 
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 #include <stand.h>
31 #include <errno.h>
32 #include <stdbool.h>
33 #include <stddef.h>
34 #include <string.h>
35 #include <stdarg.h>
36 #include <sys/param.h>
37 
38 #include <net/ethernet.h>
39 #include <netinet/in_systm.h>
40 #include <netinet/in.h>
41 #include <netinet/ip.h>
42 #include <netinet/udp.h>
43 
44 #include <net.h>
45 #include <netif.h>
46 #include <nfsv2.h>
47 #include <iodesc.h>
48 
49 #include <bootp.h>
50 #include <bootstrap.h>
51 #include "libi386.h"
52 #include "btxv86.h"
53 #include "pxe.h"
54 
55 static pxenv_t *pxenv_p = NULL;	/* PXENV+ */
56 static pxe_t *pxe_p = NULL;		/* !PXE */
57 
58 #ifdef PXE_DEBUG
59 static int	pxe_debug = 0;
60 #endif
61 
62 void		pxe_enable(void *pxeinfo);
63 static void	(*pxe_call)(int func, void *ptr);
64 static void	pxenv_call(int func, void *ptr);
65 static void	bangpxe_call(int func, void *ptr);
66 
67 static int	pxe_init(void);
68 static int	pxe_print(int verbose);
69 static void	pxe_cleanup(void);
70 
71 static void	pxe_perror(int error);
72 static int	pxe_netif_match(struct netif *nif, void *machdep_hint);
73 static int	pxe_netif_probe(struct netif *nif, void *machdep_hint);
74 static void	pxe_netif_init(struct iodesc *desc, void *machdep_hint);
75 static ssize_t	pxe_netif_get(struct iodesc *, void **, time_t);
76 static ssize_t	pxe_netif_put(struct iodesc *desc, void *pkt, size_t len);
77 static void	pxe_netif_end(struct netif *nif);
78 
79 extern struct netif_stats	pxe_st[];
80 extern uint16_t			__bangpxeseg;
81 extern uint16_t			__bangpxeoff;
82 extern void			__bangpxeentry(void);
83 extern uint16_t			__pxenvseg;
84 extern uint16_t			__pxenvoff;
85 extern void			__pxenventry(void);
86 
87 struct netif_dif pxe_ifs[] = {
88 /*	dif_unit        dif_nsel        dif_stats       dif_private     */
89 	{0,             1,              &pxe_st[0],     0}
90 };
91 
92 struct netif_stats pxe_st[nitems(pxe_ifs)];
93 
94 struct netif_driver pxenetif = {
95 	.netif_bname = "pxenet",
96 	.netif_match = pxe_netif_match,
97 	.netif_probe = pxe_netif_probe,
98 	.netif_init = pxe_netif_init,
99 	.netif_get = pxe_netif_get,
100 	.netif_put = pxe_netif_put,
101 	.netif_end = pxe_netif_end,
102 	.netif_ifs = pxe_ifs,
103 	.netif_nifs = nitems(pxe_ifs)
104 };
105 
106 struct netif_driver *netif_drivers[] = {
107 	&pxenetif,
108 	NULL
109 };
110 
111 struct devsw pxedisk = {
112 	.dv_name = "net",
113 	.dv_type = DEVT_NET,
114 	.dv_init = pxe_init,
115 	.dv_strategy = NULL,	/* Will be set in pxe_init */
116 	.dv_open = NULL,	/* Will be set in pxe_init */
117 	.dv_close = NULL,	/* Will be set in pxe_init */
118 	.dv_ioctl = noioctl,
119 	.dv_print = pxe_print,
120 	.dv_cleanup = pxe_cleanup,
121 };
122 
123 /*
124  * This function is called by the loader to enable PXE support if we
125  * are booted by PXE. The passed in pointer is a pointer to the PXENV+
126  * structure.
127  */
128 void
129 pxe_enable(void *pxeinfo)
130 {
131 	pxenv_p  = (pxenv_t *)pxeinfo;
132 	pxe_p    = (pxe_t *)PTOV(pxenv_p->PXEPtr.segment * 16 +
133 				 pxenv_p->PXEPtr.offset);
134 	pxe_call = NULL;
135 }
136 
137 /*
138  * return true if pxe structures are found/initialized,
139  * also figures out our IP information via the pxe cached info struct
140  */
141 static int
142 pxe_init(void)
143 {
144 	t_PXENV_GET_CACHED_INFO *gci_p;
145 	int counter;
146 	uint8_t checksum;
147 	uint8_t *checkptr;
148 	extern struct devsw netdev;
149 
150 	if (pxenv_p == NULL)
151 		return (0);
152 
153 	/* look for "PXENV+" */
154 	if (bcmp((void *)pxenv_p->Signature, S_SIZE("PXENV+"))) {
155 		pxenv_p = NULL;
156 		return (0);
157 	}
158 
159 	/* make sure the size is something we can handle */
160 	if (pxenv_p->Length > sizeof(*pxenv_p)) {
161 		printf("PXENV+ structure too large, ignoring\n");
162 		pxenv_p = NULL;
163 		return (0);
164 	}
165 
166 	/*
167 	 * do byte checksum:
168 	 * add up each byte in the structure, the total should be 0
169 	 */
170 	checksum = 0;
171 	checkptr = (uint8_t *) pxenv_p;
172 	for (counter = 0; counter < pxenv_p->Length; counter++)
173 		checksum += *checkptr++;
174 	if (checksum != 0) {
175 		printf("PXENV+ structure failed checksum, ignoring\n");
176 		pxenv_p = NULL;
177 		return (0);
178 	}
179 
180 	/*
181 	 * PXENV+ passed, so use that if !PXE is not available or
182 	 * the checksum fails.
183 	 */
184 	pxe_call = pxenv_call;
185 	if (pxenv_p->Version >= 0x0200) {
186 		for (;;) {
187 			if (bcmp((void *)pxe_p->Signature, S_SIZE("!PXE"))) {
188 				pxe_p = NULL;
189 				break;
190 			}
191 			checksum = 0;
192 			checkptr = (uint8_t *)pxe_p;
193 			for (counter = 0; counter < pxe_p->StructLength;
194 			    counter++)
195 				checksum += *checkptr++;
196 			if (checksum != 0) {
197 				pxe_p = NULL;
198 				break;
199 			}
200 			pxe_call = bangpxe_call;
201 			break;
202 		}
203 	}
204 
205 	pxedisk.dv_open = netdev.dv_open;
206 	pxedisk.dv_close = netdev.dv_close;
207 	pxedisk.dv_strategy = netdev.dv_strategy;
208 
209 	printf("\nPXE version %d.%d, real mode entry point ",
210 	    (uint8_t) (pxenv_p->Version >> 8),
211 	    (uint8_t) (pxenv_p->Version & 0xFF));
212 	if (pxe_call == bangpxe_call)
213 		printf("@%04x:%04x\n",
214 		    pxe_p->EntryPointSP.segment,
215 		    pxe_p->EntryPointSP.offset);
216 	else
217 		printf("@%04x:%04x\n",
218 		    pxenv_p->RMEntry.segment, pxenv_p->RMEntry.offset);
219 
220 	gci_p = bio_alloc(sizeof(*gci_p));
221 	if (gci_p == NULL) {
222 		pxe_p = NULL;
223 		return (0);
224 	}
225 	bzero(gci_p, sizeof(*gci_p));
226 	gci_p->PacketType = PXENV_PACKET_TYPE_BINL_REPLY;
227 	pxe_call(PXENV_GET_CACHED_INFO, gci_p);
228 	if (gci_p->Status != 0) {
229 		pxe_perror(gci_p->Status);
230 		bio_free(gci_p, sizeof(*gci_p));
231 		pxe_p = NULL;
232 		return (0);
233 	}
234 	free(bootp_response);
235 	if ((bootp_response = malloc(gci_p->BufferSize)) != NULL) {
236 		bootp_response_size = gci_p->BufferSize;
237 		bcopy(PTOV((gci_p->Buffer.segment << 4) + gci_p->Buffer.offset),
238 		    bootp_response, bootp_response_size);
239 	}
240 	bio_free(gci_p, sizeof(*gci_p));
241 	return (1);
242 }
243 
244 static int
245 pxe_print(int verbose)
246 {
247 	if (pxe_call == NULL)
248 		return (0);
249 
250 	printf("%s devices:", pxedisk.dv_name);
251 	if (pager_output("\n") != 0)
252 		return (1);
253 	printf("    %s0:", pxedisk.dv_name);
254 	if (verbose) {
255 		printf("    %s:%s", inet_ntoa(rootip), rootpath);
256 	}
257 	return (pager_output("\n"));
258 }
259 
260 static void
261 pxe_cleanup(void)
262 {
263 	t_PXENV_UNLOAD_STACK *unload_stack_p;
264 	t_PXENV_UNDI_SHUTDOWN *undi_shutdown_p;
265 
266 	if (pxe_call == NULL)
267 		return;
268 
269 	undi_shutdown_p = bio_alloc(sizeof(*undi_shutdown_p));
270 	if (undi_shutdown_p != NULL) {
271 		bzero(undi_shutdown_p, sizeof(*undi_shutdown_p));
272 		pxe_call(PXENV_UNDI_SHUTDOWN, undi_shutdown_p);
273 
274 #ifdef PXE_DEBUG
275 		if (pxe_debug && undi_shutdown_p->Status != 0)
276 			printf("pxe_cleanup: UNDI_SHUTDOWN failed %x\n",
277 			    undi_shutdown_p->Status);
278 #endif
279 		bio_free(undi_shutdown_p, sizeof(*undi_shutdown_p));
280 	}
281 
282 	unload_stack_p = bio_alloc(sizeof(*unload_stack_p));
283 	if (unload_stack_p != NULL) {
284 		bzero(unload_stack_p, sizeof(*unload_stack_p));
285 		pxe_call(PXENV_UNLOAD_STACK, unload_stack_p);
286 
287 #ifdef PXE_DEBUG
288 		if (pxe_debug && unload_stack_p->Status != 0)
289 			printf("pxe_cleanup: UNLOAD_STACK failed %x\n",
290 			    unload_stack_p->Status);
291 #endif
292 		bio_free(unload_stack_p, sizeof(*unload_stack_p));
293 	}
294 }
295 
296 void
297 pxe_perror(int err)
298 {
299 	return;
300 }
301 
302 void
303 pxenv_call(int func, void *ptr)
304 {
305 #ifdef PXE_DEBUG
306 	if (pxe_debug)
307 		printf("pxenv_call %x\n", func);
308 #endif
309 
310 	bzero(&v86, sizeof(v86));
311 
312 	__pxenvseg = pxenv_p->RMEntry.segment;
313 	__pxenvoff = pxenv_p->RMEntry.offset;
314 
315 	v86.ctl  = V86_ADDR | V86_CALLF | V86_FLAGS;
316 	v86.es   = VTOPSEG(ptr);
317 	v86.edi  = VTOPOFF(ptr);
318 	v86.addr = (VTOPSEG(__pxenventry) << 16) | VTOPOFF(__pxenventry);
319 	v86.ebx  = func;
320 	v86int();
321 	v86.ctl  = V86_FLAGS;
322 }
323 
324 void
325 bangpxe_call(int func, void *ptr)
326 {
327 #ifdef PXE_DEBUG
328 	if (pxe_debug)
329 		printf("bangpxe_call %x\n", func);
330 #endif
331 
332 	bzero(&v86, sizeof(v86));
333 
334 	__bangpxeseg = pxe_p->EntryPointSP.segment;
335 	__bangpxeoff = pxe_p->EntryPointSP.offset;
336 
337 	v86.ctl  = V86_ADDR | V86_CALLF | V86_FLAGS;
338 	v86.edx  = VTOPSEG(ptr);
339 	v86.eax  = VTOPOFF(ptr);
340 	v86.addr = (VTOPSEG(__bangpxeentry) << 16) | VTOPOFF(__bangpxeentry);
341 	v86.ebx  = func;
342 	v86int();
343 	v86.ctl  = V86_FLAGS;
344 }
345 
346 
347 static int
348 pxe_netif_match(struct netif *nif, void *machdep_hint)
349 {
350 	return (1);
351 }
352 
353 static int
354 pxe_netif_probe(struct netif *nif, void *machdep_hint)
355 {
356 	if (pxe_call == NULL)
357 		return (-1);
358 
359 	return (0);
360 }
361 
362 static void
363 pxe_netif_end(struct netif *nif)
364 {
365 	t_PXENV_UNDI_CLOSE *undi_close_p;
366 
367 	undi_close_p = bio_alloc(sizeof(*undi_close_p));
368 	if (undi_close_p != NULL) {
369 		bzero(undi_close_p, sizeof(*undi_close_p));
370 		pxe_call(PXENV_UNDI_CLOSE, undi_close_p);
371 		if (undi_close_p->Status != 0)
372 			printf("undi close failed: %x\n", undi_close_p->Status);
373 		bio_free(undi_close_p, sizeof(*undi_close_p));
374 	}
375 }
376 
377 static void
378 pxe_netif_init(struct iodesc *desc, void *machdep_hint)
379 {
380 	t_PXENV_UNDI_GET_INFORMATION *undi_info_p;
381 	t_PXENV_UNDI_OPEN *undi_open_p;
382 	uint8_t *mac;
383 	int i, len;
384 
385 	undi_info_p = bio_alloc(sizeof(*undi_info_p));
386 	if (undi_info_p == NULL)
387 		return;
388 
389 	bzero(undi_info_p, sizeof(*undi_info_p));
390 	pxe_call(PXENV_UNDI_GET_INFORMATION, undi_info_p);
391 	if (undi_info_p->Status != 0) {
392 		printf("undi get info failed: %x\n", undi_info_p->Status);
393 		bio_free(undi_info_p, sizeof(*undi_info_p));
394 		return;
395 	}
396 
397 	/* Make sure the CurrentNodeAddress is valid. */
398 	for (i = 0; i < undi_info_p->HwAddrLen; ++i) {
399 		if (undi_info_p->CurrentNodeAddress[i] != 0)
400 			break;
401 	}
402 	if (i < undi_info_p->HwAddrLen) {
403 		for (i = 0; i < undi_info_p->HwAddrLen; ++i) {
404 			if (undi_info_p->CurrentNodeAddress[i] != 0xff)
405 				break;
406 		}
407 	}
408 	if (i < undi_info_p->HwAddrLen)
409 		mac = undi_info_p->CurrentNodeAddress;
410 	else
411 		mac = undi_info_p->PermNodeAddress;
412 
413 	len = min(sizeof (desc->myea), undi_info_p->HwAddrLen);
414 	for (i = 0; i < len; ++i)
415 		desc->myea[i] = mac[i];
416 
417 	if (bootp_response != NULL)
418 		desc->xid = bootp_response->bp_xid;
419 	else
420 		desc->xid = 0;
421 
422 	bio_free(undi_info_p, sizeof(*undi_info_p));
423 	undi_open_p = bio_alloc(sizeof(*undi_open_p));
424 	if (undi_open_p == NULL)
425 		return;
426 	bzero(undi_open_p, sizeof(*undi_open_p));
427 	undi_open_p->PktFilter = FLTR_DIRECTED | FLTR_BRDCST;
428 	pxe_call(PXENV_UNDI_OPEN, undi_open_p);
429 	if (undi_open_p->Status != 0)
430 		printf("undi open failed: %x\n", undi_open_p->Status);
431 	bio_free(undi_open_p, sizeof(*undi_open_p));
432 }
433 
434 static int
435 pxe_netif_receive_isr(t_PXENV_UNDI_ISR *isr, void **pkt, ssize_t *retsize)
436 {
437 	static bool data_pending;
438 	char *buf, *ptr, *frame;
439 	size_t size, rsize;
440 
441 	buf = NULL;
442 	size = rsize = 0;
443 
444 	/*
445 	 * We can save ourselves the next two pxe calls because we already know
446 	 * we weren't done grabbing everything.
447 	 */
448 	if (data_pending) {
449 		data_pending = false;
450 		goto nextbuf;
451 	}
452 
453 	/*
454 	 * We explicitly don't check for OURS/NOT_OURS as a result of START;
455 	 * it's been reported that some cards are known to mishandle these.
456 	 */
457 	bzero(isr, sizeof(*isr));
458 	isr->FuncFlag = PXENV_UNDI_ISR_IN_START;
459 	pxe_call(PXENV_UNDI_ISR, isr);
460 	/* We could translate Status... */
461 	if (isr->Status != 0) {
462 		return (ENXIO);
463 	}
464 
465 	bzero(isr, sizeof(*isr));
466 	isr->FuncFlag = PXENV_UNDI_ISR_IN_PROCESS;
467 	pxe_call(PXENV_UNDI_ISR, isr);
468 	if (isr->Status != 0) {
469 		return (ENXIO);
470 	}
471 	if (isr->FuncFlag == PXENV_UNDI_ISR_OUT_BUSY) {
472 		/*
473 		 * Let the caller decide if we need to be restarted.  It will
474 		 * currently blindly restart us, but it could check timeout in
475 		 * the future.
476 		 */
477 		return (ERESTART);
478 	}
479 
480 	/*
481 	 * By design, we'll hardly ever hit this terminal condition unless we
482 	 * pick up nothing but tx interrupts here.  More frequently, we will
483 	 * process rx buffers until we hit the terminal condition in the middle.
484 	 */
485 	while (isr->FuncFlag != PXENV_UNDI_ISR_OUT_DONE) {
486 		/*
487 		 * This might have given us PXENV_UNDI_ISR_OUT_TRANSMIT, in
488 		 * which case we can just disregard and move on to the next
489 		 * buffer/frame.
490 		 */
491 		if (isr->FuncFlag != PXENV_UNDI_ISR_OUT_RECEIVE)
492 			goto nextbuf;
493 
494 		if (buf == NULL) {
495 			/*
496 			 * Grab size from the first Frame that we picked up,
497 			 * allocate an rx buf to hold.  Careful here, as we may
498 			 * see a fragmented frame that's spread out across
499 			 * multiple GET_NEXT calls.
500 			 */
501 			size = isr->FrameLength;
502 			buf = malloc(size + ETHER_ALIGN);
503 			if (buf == NULL)
504 				return (ENOMEM);
505 
506 			ptr = buf + ETHER_ALIGN;
507 		}
508 
509 		frame = (char *)((uintptr_t)isr->Frame.segment << 4);
510 		frame += isr->Frame.offset;
511 		bcopy(PTOV(frame), ptr, isr->BufferLength);
512 		ptr += isr->BufferLength;
513 		rsize += isr->BufferLength;
514 
515 		/*
516 		 * Stop here before we risk catching the start of another frame.
517 		 * It would be nice to continue reading until we actually get a
518 		 * PXENV_UNDI_ISR_OUT_DONE, but our network stack in libsa isn't
519 		 * suitable for reading more than one packet at a time.
520 		 */
521 		if (rsize >= size) {
522 			data_pending = true;
523 			break;
524 		}
525 
526 nextbuf:
527 		bzero(isr, sizeof(*isr));
528 		isr->FuncFlag = PXENV_UNDI_ISR_IN_GET_NEXT;
529 		pxe_call(PXENV_UNDI_ISR, isr);
530 		if (isr->Status != 0) {
531 			free(buf);
532 			return (ENXIO);
533 		}
534 	}
535 
536 	/*
537 	 * We may have never picked up a frame at all (all tx), in which case
538 	 * the caller should restart us.
539 	 */
540 	if (rsize == 0) {
541 		return (ERESTART);
542 	}
543 
544 	*pkt = buf;
545 	*retsize = rsize;
546 	return (0);
547 }
548 
549 static int
550 pxe_netif_receive(void **pkt, ssize_t *size)
551 {
552 	t_PXENV_UNDI_ISR *isr;
553 	int ret;
554 
555 	isr = bio_alloc(sizeof(*isr));
556 	if (isr == NULL)
557 		return (ENOMEM);
558 
559 	/*
560 	 * This completely ignores the timeout specified in pxe_netif_get(), but
561 	 * we shouldn't be running long enough here for that to make a
562 	 * difference.
563 	 */
564 	for (;;) {
565 		/* We'll only really re-enter for PXENV_UNDI_ISR_OUT_BUSY. */
566 		ret = pxe_netif_receive_isr(isr, pkt, size);
567 		if (ret != ERESTART)
568 			break;
569 	}
570 
571 	bio_free(isr, sizeof(*isr));
572 	return (ret);
573 }
574 
575 static ssize_t
576 pxe_netif_get(struct iodesc *desc, void **pkt, time_t timeout)
577 {
578 	time_t t;
579 	void *ptr;
580 	int ret = -1;
581 	ssize_t size;
582 
583 	t = getsecs();
584 	size = 0;
585 	while ((getsecs() - t) < timeout) {
586 		ret = pxe_netif_receive(&ptr, &size);
587 		if (ret != -1) {
588 			*pkt = ptr;
589 			break;
590 		}
591 	}
592 
593 	return (ret == 0 ? size : -1);
594 }
595 
596 static ssize_t
597 pxe_netif_put(struct iodesc *desc, void *pkt, size_t len)
598 {
599 	t_PXENV_UNDI_TRANSMIT *trans_p;
600 	t_PXENV_UNDI_TBD *tbd_p;
601 	char *data;
602 	ssize_t rv = -1;
603 
604 	trans_p = bio_alloc(sizeof(*trans_p));
605 	tbd_p = bio_alloc(sizeof(*tbd_p));
606 	data = bio_alloc(len);
607 
608 	if (trans_p != NULL && tbd_p != NULL && data != NULL) {
609 		bzero(trans_p, sizeof(*trans_p));
610 		bzero(tbd_p, sizeof(*tbd_p));
611 
612 		trans_p->TBD.segment = VTOPSEG(tbd_p);
613 		trans_p->TBD.offset  = VTOPOFF(tbd_p);
614 
615 		tbd_p->ImmedLength = len;
616 		tbd_p->Xmit.segment = VTOPSEG(data);
617 		tbd_p->Xmit.offset  = VTOPOFF(data);
618 		bcopy(pkt, data, len);
619 
620 		pxe_call(PXENV_UNDI_TRANSMIT, trans_p);
621 		if (trans_p->Status == 0)
622 			rv = len;
623 	}
624 
625 	bio_free(data, len);
626 	bio_free(tbd_p, sizeof(*tbd_p));
627 	bio_free(trans_p, sizeof(*trans_p));
628 	return (rv);
629 }
630