xref: /freebsd/stand/i386/libi386/pxe.c (revision a91a2465) 
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 <stand.h>
30 #include <errno.h>
31 #include <stdbool.h>
32 #include <stddef.h>
33 #include <string.h>
34 #include <stdarg.h>
35 #include <sys/param.h>
36 
37 #include <net/ethernet.h>
38 #include <netinet/in_systm.h>
39 #include <netinet/in.h>
40 #include <netinet/ip.h>
41 #include <netinet/udp.h>
42 
43 #include <net.h>
44 #include <netif.h>
45 #include <nfsv2.h>
46 #include <iodesc.h>
47 
48 #include <bootp.h>
49 #include <bootstrap.h>
50 #include "libi386.h"
51 #include "btxv86.h"
52 #include "pxe.h"
53 
54 static pxenv_t *pxenv_p = NULL;	/* PXENV+ */
55 static pxe_t *pxe_p = NULL;		/* !PXE */
56 
57 #ifdef PXE_DEBUG
58 static int	pxe_debug = 0;
59 #endif
60 
61 void		pxe_enable(void *pxeinfo);
62 static void	(*pxe_call)(int func, void *ptr);
63 static void	pxenv_call(int func, void *ptr);
64 static void	bangpxe_call(int func, void *ptr);
65 
66 static int	pxe_init(void);
67 static int	pxe_print(int verbose);
68 static void	pxe_cleanup(void);
69 
70 static void	pxe_perror(int error);
71 static int	pxe_netif_match(struct netif *nif, void *machdep_hint);
72 static int	pxe_netif_probe(struct netif *nif, void *machdep_hint);
73 static void	pxe_netif_init(struct iodesc *desc, void *machdep_hint);
74 static ssize_t	pxe_netif_get(struct iodesc *, void **, time_t);
75 static ssize_t	pxe_netif_put(struct iodesc *desc, void *pkt, size_t len);
76 static void	pxe_netif_end(struct netif *nif);
77 
78 extern struct netif_stats	pxe_st[];
79 extern uint16_t			__bangpxeseg;
80 extern uint16_t			__bangpxeoff;
81 extern void			__bangpxeentry(void);
82 extern uint16_t			__pxenvseg;
83 extern uint16_t			__pxenvoff;
84 extern void			__pxenventry(void);
85 
86 struct netif_dif pxe_ifs[] = {
87 /*	dif_unit        dif_nsel        dif_stats       dif_private     */
88 	{0,             1,              &pxe_st[0],     0}
89 };
90 
91 struct netif_stats pxe_st[nitems(pxe_ifs)];
92 
93 struct netif_driver pxenetif = {
94 	.netif_bname = "pxenet",
95 	.netif_match = pxe_netif_match,
96 	.netif_probe = pxe_netif_probe,
97 	.netif_init = pxe_netif_init,
98 	.netif_get = pxe_netif_get,
99 	.netif_put = pxe_netif_put,
100 	.netif_end = pxe_netif_end,
101 	.netif_ifs = pxe_ifs,
102 	.netif_nifs = nitems(pxe_ifs)
103 };
104 
105 struct netif_driver *netif_drivers[] = {
106 	&pxenetif,
107 	NULL
108 };
109 
110 struct devsw pxedisk = {
111 	.dv_name = "net",
112 	.dv_type = DEVT_NET,
113 	.dv_init = pxe_init,
114 	.dv_strategy = NULL,	/* Will be set in pxe_init */
115 	.dv_open = NULL,	/* Will be set in pxe_init */
116 	.dv_close = NULL,	/* Will be set in pxe_init */
117 	.dv_ioctl = noioctl,
118 	.dv_print = pxe_print,
119 	.dv_cleanup = pxe_cleanup,
120 };
121 
122 /*
123  * This function is called by the loader to enable PXE support if we
124  * are booted by PXE. The passed in pointer is a pointer to the PXENV+
125  * structure.
126  */
127 void
128 pxe_enable(void *pxeinfo)
129 {
130 	pxenv_p  = (pxenv_t *)pxeinfo;
131 	pxe_p    = (pxe_t *)PTOV(pxenv_p->PXEPtr.segment * 16 +
132 				 pxenv_p->PXEPtr.offset);
133 	pxe_call = NULL;
134 }
135 
136 /*
137  * return true if pxe structures are found/initialized,
138  * also figures out our IP information via the pxe cached info struct
139  */
140 static int
141 pxe_init(void)
142 {
143 	t_PXENV_GET_CACHED_INFO *gci_p;
144 	int counter;
145 	uint8_t checksum;
146 	uint8_t *checkptr;
147 	extern struct devsw netdev;
148 
149 	if (pxenv_p == NULL)
150 		return (0);
151 
152 	/* look for "PXENV+" */
153 	if (bcmp((void *)pxenv_p->Signature, S_SIZE("PXENV+"))) {
154 		pxenv_p = NULL;
155 		return (0);
156 	}
157 
158 	/* make sure the size is something we can handle */
159 	if (pxenv_p->Length > sizeof(*pxenv_p)) {
160 		printf("PXENV+ structure too large, ignoring\n");
161 		pxenv_p = NULL;
162 		return (0);
163 	}
164 
165 	/*
166 	 * do byte checksum:
167 	 * add up each byte in the structure, the total should be 0
168 	 */
169 	checksum = 0;
170 	checkptr = (uint8_t *) pxenv_p;
171 	for (counter = 0; counter < pxenv_p->Length; counter++)
172 		checksum += *checkptr++;
173 	if (checksum != 0) {
174 		printf("PXENV+ structure failed checksum, ignoring\n");
175 		pxenv_p = NULL;
176 		return (0);
177 	}
178 
179 	/*
180 	 * PXENV+ passed, so use that if !PXE is not available or
181 	 * the checksum fails.
182 	 */
183 	pxe_call = pxenv_call;
184 	if (pxenv_p->Version >= 0x0200) {
185 		for (;;) {
186 			if (bcmp((void *)pxe_p->Signature, S_SIZE("!PXE"))) {
187 				pxe_p = NULL;
188 				break;
189 			}
190 			checksum = 0;
191 			checkptr = (uint8_t *)pxe_p;
192 			for (counter = 0; counter < pxe_p->StructLength;
193 			    counter++)
194 				checksum += *checkptr++;
195 			if (checksum != 0) {
196 				pxe_p = NULL;
197 				break;
198 			}
199 			pxe_call = bangpxe_call;
200 			break;
201 		}
202 	}
203 
204 	pxedisk.dv_open = netdev.dv_open;
205 	pxedisk.dv_close = netdev.dv_close;
206 	pxedisk.dv_strategy = netdev.dv_strategy;
207 
208 	printf("\nPXE version %d.%d, real mode entry point ",
209 	    (uint8_t) (pxenv_p->Version >> 8),
210 	    (uint8_t) (pxenv_p->Version & 0xFF));
211 	if (pxe_call == bangpxe_call)
212 		printf("@%04x:%04x\n",
213 		    pxe_p->EntryPointSP.segment,
214 		    pxe_p->EntryPointSP.offset);
215 	else
216 		printf("@%04x:%04x\n",
217 		    pxenv_p->RMEntry.segment, pxenv_p->RMEntry.offset);
218 
219 	gci_p = bio_alloc(sizeof(*gci_p));
220 	if (gci_p == NULL) {
221 		pxe_p = NULL;
222 		return (0);
223 	}
224 	bzero(gci_p, sizeof(*gci_p));
225 	gci_p->PacketType = PXENV_PACKET_TYPE_BINL_REPLY;
226 	pxe_call(PXENV_GET_CACHED_INFO, gci_p);
227 	if (gci_p->Status != 0) {
228 		pxe_perror(gci_p->Status);
229 		bio_free(gci_p, sizeof(*gci_p));
230 		pxe_p = NULL;
231 		return (0);
232 	}
233 	free(bootp_response);
234 	if ((bootp_response = malloc(gci_p->BufferSize)) != NULL) {
235 		bootp_response_size = gci_p->BufferSize;
236 		bcopy(PTOV((gci_p->Buffer.segment << 4) + gci_p->Buffer.offset),
237 		    bootp_response, bootp_response_size);
238 	}
239 	bio_free(gci_p, sizeof(*gci_p));
240 	return (1);
241 }
242 
243 static int
244 pxe_print(int verbose)
245 {
246 	if (pxe_call == NULL)
247 		return (0);
248 
249 	printf("%s devices:", pxedisk.dv_name);
250 	if (pager_output("\n") != 0)
251 		return (1);
252 	printf("    %s0:", pxedisk.dv_name);
253 	if (verbose) {
254 		printf("    %s:%s", inet_ntoa(rootip), rootpath);
255 	}
256 	return (pager_output("\n"));
257 }
258 
259 static void
260 pxe_cleanup(void)
261 {
262 	t_PXENV_UNLOAD_STACK *unload_stack_p;
263 	t_PXENV_UNDI_SHUTDOWN *undi_shutdown_p;
264 
265 	if (pxe_call == NULL)
266 		return;
267 
268 	undi_shutdown_p = bio_alloc(sizeof(*undi_shutdown_p));
269 	if (undi_shutdown_p != NULL) {
270 		bzero(undi_shutdown_p, sizeof(*undi_shutdown_p));
271 		pxe_call(PXENV_UNDI_SHUTDOWN, undi_shutdown_p);
272 
273 #ifdef PXE_DEBUG
274 		if (pxe_debug && undi_shutdown_p->Status != 0)
275 			printf("pxe_cleanup: UNDI_SHUTDOWN failed %x\n",
276 			    undi_shutdown_p->Status);
277 #endif
278 		bio_free(undi_shutdown_p, sizeof(*undi_shutdown_p));
279 	}
280 
281 	unload_stack_p = bio_alloc(sizeof(*unload_stack_p));
282 	if (unload_stack_p != NULL) {
283 		bzero(unload_stack_p, sizeof(*unload_stack_p));
284 		pxe_call(PXENV_UNLOAD_STACK, unload_stack_p);
285 
286 #ifdef PXE_DEBUG
287 		if (pxe_debug && unload_stack_p->Status != 0)
288 			printf("pxe_cleanup: UNLOAD_STACK failed %x\n",
289 			    unload_stack_p->Status);
290 #endif
291 		bio_free(unload_stack_p, sizeof(*unload_stack_p));
292 	}
293 }
294 
295 void
296 pxe_perror(int err)
297 {
298 	return;
299 }
300 
301 void
302 pxenv_call(int func, void *ptr)
303 {
304 #ifdef PXE_DEBUG
305 	if (pxe_debug)
306 		printf("pxenv_call %x\n", func);
307 #endif
308 
309 	bzero(&v86, sizeof(v86));
310 
311 	__pxenvseg = pxenv_p->RMEntry.segment;
312 	__pxenvoff = pxenv_p->RMEntry.offset;
313 
314 	v86.ctl  = V86_ADDR | V86_CALLF | V86_FLAGS;
315 	v86.es   = VTOPSEG(ptr);
316 	v86.edi  = VTOPOFF(ptr);
317 	v86.addr = (VTOPSEG(__pxenventry) << 16) | VTOPOFF(__pxenventry);
318 	v86.ebx  = func;
319 	v86int();
320 	v86.ctl  = V86_FLAGS;
321 }
322 
323 void
324 bangpxe_call(int func, void *ptr)
325 {
326 #ifdef PXE_DEBUG
327 	if (pxe_debug)
328 		printf("bangpxe_call %x\n", func);
329 #endif
330 
331 	bzero(&v86, sizeof(v86));
332 
333 	__bangpxeseg = pxe_p->EntryPointSP.segment;
334 	__bangpxeoff = pxe_p->EntryPointSP.offset;
335 
336 	v86.ctl  = V86_ADDR | V86_CALLF | V86_FLAGS;
337 	v86.edx  = VTOPSEG(ptr);
338 	v86.eax  = VTOPOFF(ptr);
339 	v86.addr = (VTOPSEG(__bangpxeentry) << 16) | VTOPOFF(__bangpxeentry);
340 	v86.ebx  = func;
341 	v86int();
342 	v86.ctl  = V86_FLAGS;
343 }
344 
345 
346 static int
347 pxe_netif_match(struct netif *nif, void *machdep_hint)
348 {
349 	return (1);
350 }
351 
352 static int
353 pxe_netif_probe(struct netif *nif, void *machdep_hint)
354 {
355 	if (pxe_call == NULL)
356 		return (-1);
357 
358 	return (0);
359 }
360 
361 static void
362 pxe_netif_end(struct netif *nif)
363 {
364 	t_PXENV_UNDI_CLOSE *undi_close_p;
365 
366 	undi_close_p = bio_alloc(sizeof(*undi_close_p));
367 	if (undi_close_p != NULL) {
368 		bzero(undi_close_p, sizeof(*undi_close_p));
369 		pxe_call(PXENV_UNDI_CLOSE, undi_close_p);
370 		if (undi_close_p->Status != 0)
371 			printf("undi close failed: %x\n", undi_close_p->Status);
372 		bio_free(undi_close_p, sizeof(*undi_close_p));
373 	}
374 }
375 
376 static void
377 pxe_netif_init(struct iodesc *desc, void *machdep_hint)
378 {
379 	t_PXENV_UNDI_GET_INFORMATION *undi_info_p;
380 	t_PXENV_UNDI_OPEN *undi_open_p;
381 	uint8_t *mac;
382 	int i, len;
383 
384 	undi_info_p = bio_alloc(sizeof(*undi_info_p));
385 	if (undi_info_p == NULL)
386 		return;
387 
388 	bzero(undi_info_p, sizeof(*undi_info_p));
389 	pxe_call(PXENV_UNDI_GET_INFORMATION, undi_info_p);
390 	if (undi_info_p->Status != 0) {
391 		printf("undi get info failed: %x\n", undi_info_p->Status);
392 		bio_free(undi_info_p, sizeof(*undi_info_p));
393 		return;
394 	}
395 
396 	/* Make sure the CurrentNodeAddress is valid. */
397 	for (i = 0; i < undi_info_p->HwAddrLen; ++i) {
398 		if (undi_info_p->CurrentNodeAddress[i] != 0)
399 			break;
400 	}
401 	if (i < undi_info_p->HwAddrLen) {
402 		for (i = 0; i < undi_info_p->HwAddrLen; ++i) {
403 			if (undi_info_p->CurrentNodeAddress[i] != 0xff)
404 				break;
405 		}
406 	}
407 	if (i < undi_info_p->HwAddrLen)
408 		mac = undi_info_p->CurrentNodeAddress;
409 	else
410 		mac = undi_info_p->PermNodeAddress;
411 
412 	len = min(sizeof (desc->myea), undi_info_p->HwAddrLen);
413 	for (i = 0; i < len; ++i)
414 		desc->myea[i] = mac[i];
415 
416 	if (bootp_response != NULL)
417 		desc->xid = bootp_response->bp_xid;
418 	else
419 		desc->xid = 0;
420 
421 	bio_free(undi_info_p, sizeof(*undi_info_p));
422 	undi_open_p = bio_alloc(sizeof(*undi_open_p));
423 	if (undi_open_p == NULL)
424 		return;
425 	bzero(undi_open_p, sizeof(*undi_open_p));
426 	undi_open_p->PktFilter = FLTR_DIRECTED | FLTR_BRDCST;
427 	pxe_call(PXENV_UNDI_OPEN, undi_open_p);
428 	if (undi_open_p->Status != 0)
429 		printf("undi open failed: %x\n", undi_open_p->Status);
430 	bio_free(undi_open_p, sizeof(*undi_open_p));
431 }
432 
433 static int
434 pxe_netif_receive_isr(t_PXENV_UNDI_ISR *isr, void **pkt, ssize_t *retsize)
435 {
436 	static bool data_pending;
437 	char *buf, *ptr, *frame;
438 	size_t size, rsize;
439 
440 	buf = NULL;
441 	size = rsize = 0;
442 
443 	/*
444 	 * We can save ourselves the next two pxe calls because we already know
445 	 * we weren't done grabbing everything.
446 	 */
447 	if (data_pending) {
448 		data_pending = false;
449 		goto nextbuf;
450 	}
451 
452 	/*
453 	 * We explicitly don't check for OURS/NOT_OURS as a result of START;
454 	 * it's been reported that some cards are known to mishandle these.
455 	 */
456 	bzero(isr, sizeof(*isr));
457 	isr->FuncFlag = PXENV_UNDI_ISR_IN_START;
458 	pxe_call(PXENV_UNDI_ISR, isr);
459 	/* We could translate Status... */
460 	if (isr->Status != 0) {
461 		return (ENXIO);
462 	}
463 
464 	bzero(isr, sizeof(*isr));
465 	isr->FuncFlag = PXENV_UNDI_ISR_IN_PROCESS;
466 	pxe_call(PXENV_UNDI_ISR, isr);
467 	if (isr->Status != 0) {
468 		return (ENXIO);
469 	}
470 	if (isr->FuncFlag == PXENV_UNDI_ISR_OUT_BUSY) {
471 		/*
472 		 * Let the caller decide if we need to be restarted.  It will
473 		 * currently blindly restart us, but it could check timeout in
474 		 * the future.
475 		 */
476 		return (ERESTART);
477 	}
478 
479 	/*
480 	 * By design, we'll hardly ever hit this terminal condition unless we
481 	 * pick up nothing but tx interrupts here.  More frequently, we will
482 	 * process rx buffers until we hit the terminal condition in the middle.
483 	 */
484 	while (isr->FuncFlag != PXENV_UNDI_ISR_OUT_DONE) {
485 		/*
486 		 * This might have given us PXENV_UNDI_ISR_OUT_TRANSMIT, in
487 		 * which case we can just disregard and move on to the next
488 		 * buffer/frame.
489 		 */
490 		if (isr->FuncFlag != PXENV_UNDI_ISR_OUT_RECEIVE)
491 			goto nextbuf;
492 
493 		if (buf == NULL) {
494 			/*
495 			 * Grab size from the first Frame that we picked up,
496 			 * allocate an rx buf to hold.  Careful here, as we may
497 			 * see a fragmented frame that's spread out across
498 			 * multiple GET_NEXT calls.
499 			 */
500 			size = isr->FrameLength;
501 			buf = malloc(size + ETHER_ALIGN);
502 			if (buf == NULL)
503 				return (ENOMEM);
504 
505 			ptr = buf + ETHER_ALIGN;
506 		}
507 
508 		frame = (char *)((uintptr_t)isr->Frame.segment << 4);
509 		frame += isr->Frame.offset;
510 		bcopy(PTOV(frame), ptr, isr->BufferLength);
511 		ptr += isr->BufferLength;
512 		rsize += isr->BufferLength;
513 
514 		/*
515 		 * Stop here before we risk catching the start of another frame.
516 		 * It would be nice to continue reading until we actually get a
517 		 * PXENV_UNDI_ISR_OUT_DONE, but our network stack in libsa isn't
518 		 * suitable for reading more than one packet at a time.
519 		 */
520 		if (rsize >= size) {
521 			data_pending = true;
522 			break;
523 		}
524 
525 nextbuf:
526 		bzero(isr, sizeof(*isr));
527 		isr->FuncFlag = PXENV_UNDI_ISR_IN_GET_NEXT;
528 		pxe_call(PXENV_UNDI_ISR, isr);
529 		if (isr->Status != 0) {
530 			free(buf);
531 			return (ENXIO);
532 		}
533 	}
534 
535 	/*
536 	 * We may have never picked up a frame at all (all tx), in which case
537 	 * the caller should restart us.
538 	 */
539 	if (rsize == 0) {
540 		return (ERESTART);
541 	}
542 
543 	*pkt = buf;
544 	*retsize = rsize;
545 	return (0);
546 }
547 
548 static int
549 pxe_netif_receive(void **pkt, ssize_t *size)
550 {
551 	t_PXENV_UNDI_ISR *isr;
552 	int ret;
553 
554 	isr = bio_alloc(sizeof(*isr));
555 	if (isr == NULL)
556 		return (ENOMEM);
557 
558 	/*
559 	 * This completely ignores the timeout specified in pxe_netif_get(), but
560 	 * we shouldn't be running long enough here for that to make a
561 	 * difference.
562 	 */
563 	for (;;) {
564 		/* We'll only really re-enter for PXENV_UNDI_ISR_OUT_BUSY. */
565 		ret = pxe_netif_receive_isr(isr, pkt, size);
566 		if (ret != ERESTART)
567 			break;
568 	}
569 
570 	bio_free(isr, sizeof(*isr));
571 	return (ret);
572 }
573 
574 static ssize_t
575 pxe_netif_get(struct iodesc *desc, void **pkt, time_t timeout)
576 {
577 	time_t t;
578 	void *ptr;
579 	int ret = -1;
580 	ssize_t size;
581 
582 	t = getsecs();
583 	size = 0;
584 	while ((getsecs() - t) < timeout) {
585 		ret = pxe_netif_receive(&ptr, &size);
586 		if (ret != -1) {
587 			*pkt = ptr;
588 			break;
589 		}
590 	}
591 
592 	return (ret == 0 ? size : -1);
593 }
594 
595 static ssize_t
596 pxe_netif_put(struct iodesc *desc, void *pkt, size_t len)
597 {
598 	t_PXENV_UNDI_TRANSMIT *trans_p;
599 	t_PXENV_UNDI_TBD *tbd_p;
600 	char *data;
601 	ssize_t rv = -1;
602 
603 	trans_p = bio_alloc(sizeof(*trans_p));
604 	tbd_p = bio_alloc(sizeof(*tbd_p));
605 	data = bio_alloc(len);
606 
607 	if (trans_p != NULL && tbd_p != NULL && data != NULL) {
608 		bzero(trans_p, sizeof(*trans_p));
609 		bzero(tbd_p, sizeof(*tbd_p));
610 
611 		trans_p->TBD.segment = VTOPSEG(tbd_p);
612 		trans_p->TBD.offset  = VTOPOFF(tbd_p);
613 
614 		tbd_p->ImmedLength = len;
615 		tbd_p->Xmit.segment = VTOPSEG(data);
616 		tbd_p->Xmit.offset  = VTOPOFF(data);
617 		bcopy(pkt, data, len);
618 
619 		pxe_call(PXENV_UNDI_TRANSMIT, trans_p);
620 		if (trans_p->Status == 0)
621 			rv = len;
622 	}
623 
624 	bio_free(data, len);
625 	bio_free(tbd_p, sizeof(*tbd_p));
626 	bio_free(trans_p, sizeof(*trans_p));
627 	return (rv);
628 }
629