xref: /freebsd/stand/i386/zfsboot/zfsboot.c (revision daceb336172a6b0572de864b97e70b28451ca636)
1 /*-
2  * Copyright (c) 1998 Robert Nordier
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms are freely
6  * permitted provided that the above copyright notice and this
7  * paragraph and the following disclaimer are duplicated in all
8  * such forms.
9  *
10  * This software is provided "AS IS" and without any express or
11  * implied warranties, including, without limitation, the implied
12  * warranties of merchantability and fitness for a particular
13  * purpose.
14  */
15 
16 #include <sys/cdefs.h>
17 __FBSDID("$FreeBSD$");
18 
19 #include "stand.h"
20 
21 #include <sys/param.h>
22 #include <sys/errno.h>
23 #include <sys/diskmbr.h>
24 #ifdef GPT
25 #include <sys/gpt.h>
26 #endif
27 #include <sys/reboot.h>
28 #include <sys/queue.h>
29 
30 #include <machine/bootinfo.h>
31 #include <machine/elf.h>
32 #include <machine/pc/bios.h>
33 
34 #include <stdarg.h>
35 #include <stddef.h>
36 
37 #include <a.out.h>
38 
39 #include <btxv86.h>
40 
41 #include "lib.h"
42 #include "rbx.h"
43 #include "drv.h"
44 #include "edd.h"
45 #include "cons.h"
46 #include "bootargs.h"
47 #include "paths.h"
48 
49 #include "libzfs.h"
50 
51 #define ARGS			0x900
52 #define NOPT			14
53 #define NDEV			3
54 
55 #define BIOS_NUMDRIVES		0x475
56 #define DRV_HARD		0x80
57 #define DRV_MASK		0x7f
58 
59 #define TYPE_AD			0
60 #define TYPE_DA			1
61 #define TYPE_MAXHARD		TYPE_DA
62 #define TYPE_FD			2
63 
64 #define DEV_GELIBOOT_BSIZE	4096
65 
66 extern uint32_t _end;
67 
68 #ifdef GPT
69 static const uuid_t freebsd_zfs_uuid = GPT_ENT_TYPE_FREEBSD_ZFS;
70 #endif
71 static const char optstr[NOPT] = "DhaCcdgmnpqrsv"; /* Also 'P', 'S' */
72 static const unsigned char flags[NOPT] = {
73     RBX_DUAL,
74     RBX_SERIAL,
75     RBX_ASKNAME,
76     RBX_CDROM,
77     RBX_CONFIG,
78     RBX_KDB,
79     RBX_GDB,
80     RBX_MUTE,
81     RBX_NOINTR,
82     RBX_PAUSE,
83     RBX_QUIET,
84     RBX_DFLTROOT,
85     RBX_SINGLE,
86     RBX_VERBOSE
87 };
88 uint32_t opts;
89 
90 static const unsigned char dev_maj[NDEV] = {30, 4, 2};
91 
92 static char cmd[512];
93 static char cmddup[512];
94 static char kname[1024];
95 static char rootname[256];
96 static int comspeed = SIOSPD;
97 static struct bootinfo bootinfo;
98 static uint32_t bootdev;
99 static struct zfs_boot_args zfsargs;
100 
101 vm_offset_t	high_heap_base;
102 uint32_t	bios_basemem, bios_extmem, high_heap_size;
103 
104 static struct bios_smap smap;
105 
106 /*
107  * The minimum amount of memory to reserve in bios_extmem for the heap.
108  */
109 #define	HEAP_MIN		(64 * 1024 * 1024)
110 
111 static char *heap_next;
112 static char *heap_end;
113 
114 /* Buffers that must not span a 64k boundary. */
115 #define READ_BUF_SIZE		8192
116 struct dmadat {
117 	char rdbuf[READ_BUF_SIZE];	/* for reading large things */
118 	char secbuf[READ_BUF_SIZE];	/* for MBR/disklabel */
119 };
120 static struct dmadat *dmadat;
121 
122 void exit(int);
123 void reboot(void);
124 static void load(void);
125 static int parse_cmd(void);
126 static void bios_getmem(void);
127 int main(void);
128 
129 #ifdef LOADER_GELI_SUPPORT
130 #include "geliboot.h"
131 static char gelipw[GELI_PW_MAXLEN];
132 #endif
133 
134 struct zfsdsk {
135 	struct dsk       dsk;
136 #ifdef LOADER_GELI_SUPPORT
137 	struct geli_dev *gdev;
138 #endif
139 };
140 
141 #include "zfsimpl.c"
142 
143 /*
144  * Read from a dnode (which must be from a ZPL filesystem).
145  */
146 static int
147 zfs_read(spa_t *spa, const dnode_phys_t *dnode, off_t *offp, void *start, size_t size)
148 {
149 	const znode_phys_t *zp = (const znode_phys_t *) dnode->dn_bonus;
150 	size_t n;
151 	int rc;
152 
153 	n = size;
154 	if (*offp + n > zp->zp_size)
155 		n = zp->zp_size - *offp;
156 
157 	rc = dnode_read(spa, dnode, *offp, start, n);
158 	if (rc)
159 		return (-1);
160 	*offp += n;
161 
162 	return (n);
163 }
164 
165 /*
166  * Current ZFS pool
167  */
168 static spa_t *spa;
169 static spa_t *primary_spa;
170 static vdev_t *primary_vdev;
171 
172 /*
173  * A wrapper for dskread that doesn't have to worry about whether the
174  * buffer pointer crosses a 64k boundary.
175  */
176 static int
177 vdev_read(void *xvdev, void *priv, off_t off, void *buf, size_t bytes)
178 {
179 	char *p;
180 	daddr_t lba, alignlba;
181 	off_t diff;
182 	unsigned int nb, alignnb;
183 	struct zfsdsk *zdsk = (struct zfsdsk *) priv;
184 
185 	if ((off & (DEV_BSIZE - 1)) || (bytes & (DEV_BSIZE - 1)))
186 		return -1;
187 
188 	p = buf;
189 	lba = off / DEV_BSIZE;
190 	lba += zdsk->dsk.start;
191 	/*
192 	 * Align reads to 4k else 4k sector GELIs will not decrypt.
193 	 * Round LBA down to nearest multiple of DEV_GELIBOOT_BSIZE bytes.
194 	 */
195 	alignlba = rounddown2(off, DEV_GELIBOOT_BSIZE) / DEV_BSIZE;
196 	/*
197 	 * The read must be aligned to DEV_GELIBOOT_BSIZE bytes relative to the
198 	 * start of the GELI partition, not the start of the actual disk.
199 	 */
200 	alignlba += zdsk->dsk.start;
201 	diff = (lba - alignlba) * DEV_BSIZE;
202 
203 	while (bytes > 0) {
204 		nb = bytes / DEV_BSIZE;
205 		/*
206 		 * Ensure that the read size plus the leading offset does not
207 		 * exceed the size of the read buffer.
208 		 */
209 		if (nb > (READ_BUF_SIZE - diff) / DEV_BSIZE)
210 			nb = (READ_BUF_SIZE - diff) / DEV_BSIZE;
211 		/*
212 		 * Round the number of blocks to read up to the nearest multiple
213 		 * of DEV_GELIBOOT_BSIZE.
214 		 */
215 		alignnb = roundup2(nb * DEV_BSIZE + diff, DEV_GELIBOOT_BSIZE)
216 		    / DEV_BSIZE;
217 
218 		if (zdsk->dsk.size > 0 && alignlba + alignnb >
219 		    zdsk->dsk.size + zdsk->dsk.start) {
220 			printf("Shortening read at %lld from %d to %lld\n",
221 			    alignlba, alignnb,
222 			    (zdsk->dsk.size + zdsk->dsk.start) - alignlba);
223 			alignnb = (zdsk->dsk.size + zdsk->dsk.start) - alignlba;
224 		}
225 
226 		if (drvread(&zdsk->dsk, dmadat->rdbuf, alignlba, alignnb))
227 			return -1;
228 #ifdef LOADER_GELI_SUPPORT
229 		/* decrypt */
230 		if (zdsk->gdev != NULL) {
231 			if (geli_read(zdsk->gdev, ((alignlba - zdsk->dsk.start) *
232 			    DEV_BSIZE), dmadat->rdbuf, alignnb * DEV_BSIZE))
233 				return (-1);
234 		}
235 #endif
236 		memcpy(p, dmadat->rdbuf + diff, nb * DEV_BSIZE);
237 		p += nb * DEV_BSIZE;
238 		lba += nb;
239 		alignlba += alignnb;
240 		bytes -= nb * DEV_BSIZE;
241 		/* Don't need the leading offset after the first block. */
242 		diff = 0;
243 	}
244 
245 	return 0;
246 }
247 /* Match the signature exactly due to signature madness */
248 static int
249 vdev_read2(vdev_t *vdev, void *priv, off_t off, void *buf, size_t bytes)
250 {
251 	return vdev_read(vdev, priv, off, buf, bytes);
252 }
253 
254 
255 static int
256 vdev_write(vdev_t *vdev, void *priv, off_t off, void *buf, size_t bytes)
257 {
258 	char *p;
259 	daddr_t lba;
260 	unsigned int nb;
261 	struct zfsdsk *zdsk = (struct zfsdsk *) priv;
262 
263 	if ((off & (DEV_BSIZE - 1)) || (bytes & (DEV_BSIZE - 1)))
264 		return -1;
265 
266 	p = buf;
267 	lba = off / DEV_BSIZE;
268 	lba += zdsk->dsk.start;
269 	while (bytes > 0) {
270 		nb = bytes / DEV_BSIZE;
271 		if (nb > READ_BUF_SIZE / DEV_BSIZE)
272 			nb = READ_BUF_SIZE / DEV_BSIZE;
273 		memcpy(dmadat->rdbuf, p, nb * DEV_BSIZE);
274 		if (drvwrite(&zdsk->dsk, dmadat->rdbuf, lba, nb))
275 			return -1;
276 		p += nb * DEV_BSIZE;
277 		lba += nb;
278 		bytes -= nb * DEV_BSIZE;
279 	}
280 
281 	return 0;
282 }
283 
284 static int
285 xfsread(const dnode_phys_t *dnode, off_t *offp, void *buf, size_t nbyte)
286 {
287     if ((size_t)zfs_read(spa, dnode, offp, buf, nbyte) != nbyte) {
288 	printf("Invalid format\n");
289 	return -1;
290     }
291     return 0;
292 }
293 
294 /*
295  * Read Pad2 (formerly "Boot Block Header") area of the first
296  * vdev label of the given vdev.
297  */
298 static int
299 vdev_read_pad2(vdev_t *vdev, char *buf, size_t size)
300 {
301 	blkptr_t bp;
302 	char *tmp = zap_scratch;
303 	off_t off = offsetof(vdev_label_t, vl_pad2);
304 
305 	if (size > VDEV_PAD_SIZE)
306 		size = VDEV_PAD_SIZE;
307 
308 	BP_ZERO(&bp);
309 	BP_SET_LSIZE(&bp, VDEV_PAD_SIZE);
310 	BP_SET_PSIZE(&bp, VDEV_PAD_SIZE);
311 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
312 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
313 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
314 	if (vdev_read_phys(vdev, &bp, tmp, off, 0))
315 		return (EIO);
316 	memcpy(buf, tmp, size);
317 	return (0);
318 }
319 
320 static int
321 vdev_clear_pad2(vdev_t *vdev)
322 {
323 	char *zeroes = zap_scratch;
324 	uint64_t *end;
325 	off_t off = offsetof(vdev_label_t, vl_pad2);
326 
327 	memset(zeroes, 0, VDEV_PAD_SIZE);
328 	end = (uint64_t *)(zeroes + VDEV_PAD_SIZE);
329 	/* ZIO_CHECKSUM_LABEL magic and pre-calcualted checksum for all zeros */
330 	end[-5] = 0x0210da7ab10c7a11;
331 	end[-4] = 0x97f48f807f6e2a3f;
332 	end[-3] = 0xaf909f1658aacefc;
333 	end[-2] = 0xcbd1ea57ff6db48b;
334 	end[-1] = 0x6ec692db0d465fab;
335 	if (vdev_write(vdev, vdev->v_read_priv, off, zeroes, VDEV_PAD_SIZE))
336 		return (EIO);
337 	return (0);
338 }
339 
340 static void
341 bios_getmem(void)
342 {
343     uint64_t size;
344 
345     /* Parse system memory map */
346     v86.ebx = 0;
347     do {
348 	v86.ctl = V86_FLAGS;
349 	v86.addr = 0x15;		/* int 0x15 function 0xe820*/
350 	v86.eax = 0xe820;
351 	v86.ecx = sizeof(struct bios_smap);
352 	v86.edx = SMAP_SIG;
353 	v86.es = VTOPSEG(&smap);
354 	v86.edi = VTOPOFF(&smap);
355 	v86int();
356 	if (V86_CY(v86.efl) || (v86.eax != SMAP_SIG))
357 	    break;
358 	/* look for a low-memory segment that's large enough */
359 	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0) &&
360 	    (smap.length >= (512 * 1024)))
361 	    bios_basemem = smap.length;
362 	/* look for the first segment in 'extended' memory */
363 	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0x100000)) {
364 	    bios_extmem = smap.length;
365 	}
366 
367 	/*
368 	 * Look for the largest segment in 'extended' memory beyond
369 	 * 1MB but below 4GB.
370 	 */
371 	if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base > 0x100000) &&
372 	    (smap.base < 0x100000000ull)) {
373 	    size = smap.length;
374 
375 	    /*
376 	     * If this segment crosses the 4GB boundary, truncate it.
377 	     */
378 	    if (smap.base + size > 0x100000000ull)
379 		size = 0x100000000ull - smap.base;
380 
381 	    if (size > high_heap_size) {
382 		high_heap_size = size;
383 		high_heap_base = smap.base;
384 	    }
385 	}
386     } while (v86.ebx != 0);
387 
388     /* Fall back to the old compatibility function for base memory */
389     if (bios_basemem == 0) {
390 	v86.ctl = 0;
391 	v86.addr = 0x12;		/* int 0x12 */
392 	v86int();
393 
394 	bios_basemem = (v86.eax & 0xffff) * 1024;
395     }
396 
397     /* Fall back through several compatibility functions for extended memory */
398     if (bios_extmem == 0) {
399 	v86.ctl = V86_FLAGS;
400 	v86.addr = 0x15;		/* int 0x15 function 0xe801*/
401 	v86.eax = 0xe801;
402 	v86int();
403 	if (!V86_CY(v86.efl)) {
404 	    bios_extmem = ((v86.ecx & 0xffff) + ((v86.edx & 0xffff) * 64)) * 1024;
405 	}
406     }
407     if (bios_extmem == 0) {
408 	v86.ctl = 0;
409 	v86.addr = 0x15;		/* int 0x15 function 0x88*/
410 	v86.eax = 0x8800;
411 	v86int();
412 	bios_extmem = (v86.eax & 0xffff) * 1024;
413     }
414 
415     /*
416      * If we have extended memory and did not find a suitable heap
417      * region in the SMAP, use the last 3MB of 'extended' memory as a
418      * high heap candidate.
419      */
420     if (bios_extmem >= HEAP_MIN && high_heap_size < HEAP_MIN) {
421 	high_heap_size = HEAP_MIN;
422 	high_heap_base = bios_extmem + 0x100000 - HEAP_MIN;
423     }
424 }
425 
426 /*
427  * Try to detect a device supported by the legacy int13 BIOS
428  */
429 static int
430 int13probe(int drive)
431 {
432     v86.ctl = V86_FLAGS;
433     v86.addr = 0x13;
434     v86.eax = 0x800;
435     v86.edx = drive;
436     v86int();
437 
438     if (!V86_CY(v86.efl) &&				/* carry clear */
439 	((v86.edx & 0xff) != (drive & DRV_MASK))) {	/* unit # OK */
440 	if ((v86.ecx & 0x3f) == 0) {			/* absurd sector size */
441 		return(0);				/* skip device */
442 	}
443 	return (1);
444     }
445     return(0);
446 }
447 
448 /*
449  * We call this when we find a ZFS vdev - ZFS consumes the dsk
450  * structure so we must make a new one.
451  */
452 static struct zfsdsk *
453 copy_dsk(struct zfsdsk *zdsk)
454 {
455     struct zfsdsk *newdsk;
456 
457     newdsk = malloc(sizeof(struct zfsdsk));
458     *newdsk = *zdsk;
459     return (newdsk);
460 }
461 
462 /*
463  * Get disk size from GPT.
464  */
465 static uint64_t
466 drvsize_gpt(struct dsk *dskp)
467 {
468 #ifdef GPT
469 	struct gpt_hdr hdr;
470 	char *sec;
471 
472 	sec = dmadat->secbuf;
473 	if (drvread(dskp, sec, 1, 1))
474 		return (0);
475 
476 	memcpy(&hdr, sec, sizeof(hdr));
477 	if (memcmp(hdr.hdr_sig, GPT_HDR_SIG, sizeof(hdr.hdr_sig)) != 0 ||
478 	    hdr.hdr_lba_self != 1 || hdr.hdr_revision < 0x00010000 ||
479 	    hdr.hdr_entsz < sizeof(struct gpt_ent) ||
480 	    DEV_BSIZE % hdr.hdr_entsz != 0) {
481 		return (0);
482 	}
483 	return (hdr.hdr_lba_alt + 1);
484 #else
485 	return (0);
486 #endif
487 }
488 
489 /*
490  * Get disk size from eax=0x800 and 0x4800. We need to probe both
491  * because 0x4800 may not be available and we would like to get more
492  * or less correct disk size - if it is possible at all.
493  * Note we do not really want to touch drv.c because that code is shared
494  * with boot2 and we can not afford to grow that code.
495  */
496 static uint64_t
497 drvsize_ext(struct zfsdsk *zdsk)
498 {
499 	struct dsk *dskp;
500 	uint64_t size, tmp;
501 	int cyl, hds, sec;
502 
503 	dskp = &zdsk->dsk;
504 
505 	/* Try to read disk size from GPT */
506 	size = drvsize_gpt(dskp);
507 	if (size != 0)
508 		return (size);
509 
510 	v86.ctl = V86_FLAGS;
511 	v86.addr = 0x13;
512 	v86.eax = 0x800;
513 	v86.edx = dskp->drive;
514 	v86int();
515 
516 	/* Don't error out if we get bad sector number, try EDD as well */
517 	if (V86_CY(v86.efl) ||	/* carry set */
518 	    (v86.edx & 0xff) <= (unsigned)(dskp->drive & 0x7f)) /* unit # bad */
519 		return (0);
520 	cyl = ((v86.ecx & 0xc0) << 2) + ((v86.ecx & 0xff00) >> 8) + 1;
521 	/* Convert max head # -> # of heads */
522 	hds = ((v86.edx & 0xff00) >> 8) + 1;
523 	sec = v86.ecx & 0x3f;
524 
525 	size = (uint64_t)cyl * hds * sec;
526 
527 	/* Determine if we can use EDD with this device. */
528 	v86.ctl = V86_FLAGS;
529 	v86.addr = 0x13;
530 	v86.eax = 0x4100;
531 	v86.edx = dskp->drive;
532 	v86.ebx = 0x55aa;
533 	v86int();
534 	if (V86_CY(v86.efl) ||  /* carry set */
535 	    (v86.ebx & 0xffff) != 0xaa55 || /* signature */
536 	    (v86.ecx & EDD_INTERFACE_FIXED_DISK) == 0)
537 		return (size);
538 
539 	tmp = drvsize(dskp);
540 	if (tmp > size)
541 		size = tmp;
542 
543 	return (size);
544 }
545 
546 /*
547  * The "layered" ioctl to read disk/partition size. Unfortunately
548  * the zfsboot case is hardest, because we do not have full software
549  * stack available, so we need to do some manual work here.
550  */
551 uint64_t
552 ldi_get_size(void *priv)
553 {
554 	struct zfsdsk *zdsk = priv;
555 	uint64_t size = zdsk->dsk.size;
556 
557 	if (zdsk->dsk.start == 0)
558 		size = drvsize_ext(zdsk);
559 
560 	return (size * DEV_BSIZE);
561 }
562 
563 static void
564 probe_drive(struct zfsdsk *zdsk)
565 {
566 #ifdef GPT
567     struct gpt_hdr hdr;
568     struct gpt_ent *ent;
569     unsigned part, entries_per_sec;
570     daddr_t slba;
571 #endif
572 #if defined(GPT) || defined(LOADER_GELI_SUPPORT)
573     daddr_t elba;
574 #endif
575 
576     struct dos_partition *dp;
577     char *sec;
578     unsigned i;
579 
580 #ifdef LOADER_GELI_SUPPORT
581     /*
582      * Taste the disk, if it is GELI encrypted, decrypt it then dig out the
583      * partition table and probe each slice/partition in turn for a vdev or
584      * GELI encrypted vdev.
585      */
586     elba = drvsize_ext(zdsk);
587     if (elba > 0) {
588 	elba--;
589     }
590     zdsk->gdev = geli_taste(vdev_read, zdsk, elba, "disk%u:0:");
591     if ((zdsk->gdev != NULL) && (geli_havekey(zdsk->gdev) == 0))
592 	    geli_passphrase(zdsk->gdev, gelipw);
593 #endif /* LOADER_GELI_SUPPORT */
594 
595     sec = dmadat->secbuf;
596     zdsk->dsk.start = 0;
597 
598 #ifdef GPT
599     /*
600      * First check for GPT.
601      */
602     if (drvread(&zdsk->dsk, sec, 1, 1)) {
603 	return;
604     }
605     memcpy(&hdr, sec, sizeof(hdr));
606     if (memcmp(hdr.hdr_sig, GPT_HDR_SIG, sizeof(hdr.hdr_sig)) != 0 ||
607 	hdr.hdr_lba_self != 1 || hdr.hdr_revision < 0x00010000 ||
608 	hdr.hdr_entsz < sizeof(*ent) || DEV_BSIZE % hdr.hdr_entsz != 0) {
609 	goto trymbr;
610     }
611 
612     /*
613      * Probe all GPT partitions for the presence of ZFS pools. We
614      * return the spa_t for the first we find (if requested). This
615      * will have the effect of booting from the first pool on the
616      * disk.
617      *
618      * If no vdev is found, GELI decrypting the device and try again
619      */
620     entries_per_sec = DEV_BSIZE / hdr.hdr_entsz;
621     slba = hdr.hdr_lba_table;
622     elba = slba + hdr.hdr_entries / entries_per_sec;
623     while (slba < elba) {
624 	zdsk->dsk.start = 0;
625 	if (drvread(&zdsk->dsk, sec, slba, 1))
626 	    return;
627 	for (part = 0; part < entries_per_sec; part++) {
628 	    ent = (struct gpt_ent *)(sec + part * hdr.hdr_entsz);
629 	    if (memcmp(&ent->ent_type, &freebsd_zfs_uuid,
630 		     sizeof(uuid_t)) == 0) {
631 		zdsk->dsk.start = ent->ent_lba_start;
632 		zdsk->dsk.size = ent->ent_lba_end - ent->ent_lba_start + 1;
633 		zdsk->dsk.slice = part + 1;
634 		zdsk->dsk.part = 255;
635 		if (vdev_probe(vdev_read2, zdsk, NULL) == 0) {
636 		    /*
637 		     * This slice had a vdev. We need a new dsk
638 		     * structure now since the vdev now owns this one.
639 		     */
640 		    zdsk = copy_dsk(zdsk);
641 		}
642 #ifdef LOADER_GELI_SUPPORT
643 		else if ((zdsk->gdev = geli_taste(vdev_read, zdsk,
644 		    ent->ent_lba_end - ent->ent_lba_start, "disk%up%u:",
645 		    zdsk->dsk.unit, zdsk->dsk.slice)) != NULL) {
646 		    if (geli_havekey(zdsk->gdev) == 0 ||
647 			geli_passphrase(zdsk->gdev, gelipw) == 0) {
648 			/*
649 			 * This slice has GELI, check it for ZFS.
650 			 */
651 			if (vdev_probe(vdev_read2, zdsk, NULL) == 0) {
652 			    /*
653 			     * This slice had a vdev. We need a new dsk
654 			     * structure now since the vdev now owns this one.
655 			     */
656 			    zdsk = copy_dsk(zdsk);
657 			}
658 			break;
659 		    }
660 		}
661 #endif /* LOADER_GELI_SUPPORT */
662 	    }
663 	}
664 	slba++;
665     }
666     return;
667 trymbr:
668 #endif /* GPT */
669 
670     if (drvread(&zdsk->dsk, sec, DOSBBSECTOR, 1))
671 	return;
672     dp = (void *)(sec + DOSPARTOFF);
673 
674     for (i = 0; i < NDOSPART; i++) {
675 	if (!dp[i].dp_typ)
676 	    continue;
677 	zdsk->dsk.start = dp[i].dp_start;
678 	zdsk->dsk.size = dp[i].dp_size;
679 	zdsk->dsk.slice = i + 1;
680 	if (vdev_probe(vdev_read2, zdsk, NULL) == 0) {
681 	    zdsk = copy_dsk(zdsk);
682 	}
683 #ifdef LOADER_GELI_SUPPORT
684 	else if ((zdsk->gdev = geli_taste(vdev_read, zdsk, dp[i].dp_size -
685 		 dp[i].dp_start, "disk%us%u:")) != NULL) {
686 	    if (geli_havekey(zdsk->gdev) == 0 ||
687 		geli_passphrase(zdsk->gdev, gelipw) == 0) {
688 		/*
689 		 * This slice has GELI, check it for ZFS.
690 		 */
691 		if (vdev_probe(vdev_read2, zdsk, NULL) == 0) {
692 		    /*
693 		     * This slice had a vdev. We need a new dsk
694 		     * structure now since the vdev now owns this one.
695 		     */
696 		    zdsk = copy_dsk(zdsk);
697 		}
698 		break;
699 	    }
700 	}
701 #endif /* LOADER_GELI_SUPPORT */
702     }
703 }
704 
705 int
706 main(void)
707 {
708     dnode_phys_t dn;
709     off_t off;
710     struct zfsdsk *zdsk;
711     int autoboot, i;
712     int nextboot;
713     int rc;
714 
715     dmadat = (void *)(roundup2(__base + (int32_t)&_end, 0x10000) - __base);
716 
717     bios_getmem();
718 
719     if (high_heap_size > 0) {
720 	heap_end = PTOV(high_heap_base + high_heap_size);
721 	heap_next = PTOV(high_heap_base);
722     } else {
723 	heap_next = (char *)dmadat + sizeof(*dmadat);
724 	heap_end = (char *)PTOV(bios_basemem);
725     }
726     setheap(heap_next, heap_end);
727 
728     zdsk = calloc(1, sizeof(struct zfsdsk));
729     zdsk->dsk.drive = *(uint8_t *)PTOV(ARGS);
730     zdsk->dsk.type = zdsk->dsk.drive & DRV_HARD ? TYPE_AD : TYPE_FD;
731     zdsk->dsk.unit = zdsk->dsk.drive & DRV_MASK;
732     zdsk->dsk.slice = *(uint8_t *)PTOV(ARGS + 1) + 1;
733     zdsk->dsk.part = 0;
734     zdsk->dsk.start = 0;
735     zdsk->dsk.size = drvsize_ext(zdsk);
736 
737     bootinfo.bi_version = BOOTINFO_VERSION;
738     bootinfo.bi_size = sizeof(bootinfo);
739     bootinfo.bi_basemem = bios_basemem / 1024;
740     bootinfo.bi_extmem = bios_extmem / 1024;
741     bootinfo.bi_memsizes_valid++;
742     bootinfo.bi_bios_dev = zdsk->dsk.drive;
743 
744     bootdev = MAKEBOOTDEV(dev_maj[zdsk->dsk.type],
745 			  zdsk->dsk.slice, zdsk->dsk.unit, zdsk->dsk.part);
746 
747     /* Process configuration file */
748 
749     autoboot = 1;
750 
751     zfs_init();
752 
753     /*
754      * Probe the boot drive first - we will try to boot from whatever
755      * pool we find on that drive.
756      */
757     probe_drive(zdsk);
758 
759     /*
760      * Probe the rest of the drives that the bios knows about. This
761      * will find any other available pools and it may fill in missing
762      * vdevs for the boot pool.
763      */
764 #ifndef VIRTUALBOX
765     for (i = 0; i < *(unsigned char *)PTOV(BIOS_NUMDRIVES); i++)
766 #else
767     for (i = 0; i < MAXBDDEV; i++)
768 #endif
769     {
770 	if ((i | DRV_HARD) == *(uint8_t *)PTOV(ARGS))
771 	    continue;
772 
773 	if (!int13probe(i | DRV_HARD))
774 	    break;
775 
776 	zdsk = calloc(1, sizeof(struct zfsdsk));
777 	zdsk->dsk.drive = i | DRV_HARD;
778 	zdsk->dsk.type = zdsk->dsk.drive & TYPE_AD;
779 	zdsk->dsk.unit = i;
780 	zdsk->dsk.slice = 0;
781 	zdsk->dsk.part = 0;
782 	zdsk->dsk.start = 0;
783 	zdsk->dsk.size = drvsize_ext(zdsk);
784 	probe_drive(zdsk);
785     }
786 
787     /*
788      * The first discovered pool, if any, is the pool.
789      */
790     spa = spa_get_primary();
791     if (!spa) {
792 	printf("%s: No ZFS pools located, can't boot\n", BOOTPROG);
793 	for (;;)
794 	    ;
795     }
796 
797     primary_spa = spa;
798     primary_vdev = spa_get_primary_vdev(spa);
799 
800     nextboot = 0;
801     rc  = vdev_read_pad2(primary_vdev, cmd, sizeof(cmd));
802     if (vdev_clear_pad2(primary_vdev))
803 	printf("failed to clear pad2 area of primary vdev\n");
804     if (rc == 0) {
805 	if (*cmd) {
806 	    /*
807 	     * We could find an old-style ZFS Boot Block header here.
808 	     * Simply ignore it.
809 	     */
810 	    if (*(uint64_t *)cmd != 0x2f5b007b10c) {
811 		/*
812 		 * Note that parse() is destructive to cmd[] and we also want
813 		 * to honor RBX_QUIET option that could be present in cmd[].
814 		 */
815 		nextboot = 1;
816 		memcpy(cmddup, cmd, sizeof(cmd));
817 		if (parse_cmd()) {
818 		    printf("failed to parse pad2 area of primary vdev\n");
819 		    reboot();
820 		}
821 		if (!OPT_CHECK(RBX_QUIET))
822 		    printf("zfs nextboot: %s\n", cmddup);
823 	    }
824 	    /* Do not process this command twice */
825 	    *cmd = 0;
826 	}
827     } else
828 	printf("failed to read pad2 area of primary vdev\n");
829 
830     /* Mount ZFS only if it's not already mounted via nextboot parsing. */
831     if (zfsmount.spa == NULL &&
832 	(zfs_spa_init(spa) != 0 || zfs_mount(spa, 0, &zfsmount) != 0)) {
833 	printf("%s: failed to mount default pool %s\n",
834 	    BOOTPROG, spa->spa_name);
835 	autoboot = 0;
836     } else if (zfs_lookup(&zfsmount, PATH_CONFIG, &dn) == 0 ||
837         zfs_lookup(&zfsmount, PATH_DOTCONFIG, &dn) == 0) {
838 	off = 0;
839 	zfs_read(spa, &dn, &off, cmd, sizeof(cmd));
840     }
841 
842     if (*cmd) {
843 	/*
844 	 * Note that parse_cmd() is destructive to cmd[] and we also want
845 	 * to honor RBX_QUIET option that could be present in cmd[].
846 	 */
847 	memcpy(cmddup, cmd, sizeof(cmd));
848 	if (parse_cmd())
849 	    autoboot = 0;
850 	if (!OPT_CHECK(RBX_QUIET))
851 	    printf("%s: %s\n", PATH_CONFIG, cmddup);
852 	/* Do not process this command twice */
853 	*cmd = 0;
854     }
855 
856     /* Do not risk waiting at the prompt forever. */
857     if (nextboot && !autoboot)
858 	reboot();
859 
860     /*
861      * Try to exec /boot/loader. If interrupted by a keypress,
862      * or in case of failure, try to load a kernel directly instead.
863      */
864 
865     if (autoboot && !*kname) {
866 	memcpy(kname, PATH_LOADER, sizeof(PATH_LOADER));
867 	if (!keyhit(3)) {
868 	    load();
869 	    memcpy(kname, PATH_KERNEL, sizeof(PATH_KERNEL));
870 	}
871     }
872 
873     /* Present the user with the boot2 prompt. */
874 
875     for (;;) {
876 	if (!autoboot || !OPT_CHECK(RBX_QUIET)) {
877 	    printf("\nFreeBSD/x86 boot\n");
878 	    if (zfs_rlookup(spa, zfsmount.rootobj, rootname) != 0)
879 		printf("Default: %s/<0x%llx>:%s\n"
880 		       "boot: ",
881 		       spa->spa_name, zfsmount.rootobj, kname);
882 	    else if (rootname[0] != '\0')
883 		printf("Default: %s/%s:%s\n"
884 		       "boot: ",
885 		       spa->spa_name, rootname, kname);
886 	    else
887 		printf("Default: %s:%s\n"
888 		       "boot: ",
889 		       spa->spa_name, kname);
890 	}
891 	if (ioctrl & IO_SERIAL)
892 	    sio_flush();
893 	if (!autoboot || keyhit(5))
894 	    getstr(cmd, sizeof(cmd));
895 	else if (!autoboot || !OPT_CHECK(RBX_QUIET))
896 	    putchar('\n');
897 	autoboot = 0;
898 	if (parse_cmd())
899 	    putchar('\a');
900 	else
901 	    load();
902     }
903 }
904 
905 /* XXX - Needed for btxld to link the boot2 binary; do not remove. */
906 void
907 exit(int x)
908 {
909     __exit(x);
910 }
911 
912 void
913 reboot(void)
914 {
915     __exit(0);
916 }
917 
918 static void
919 load(void)
920 {
921     union {
922 	struct exec ex;
923 	Elf32_Ehdr eh;
924     } hdr;
925     static Elf32_Phdr ep[2];
926     static Elf32_Shdr es[2];
927     caddr_t p;
928     dnode_phys_t dn;
929     off_t off;
930     uint32_t addr, x;
931     int fmt, i, j;
932 
933     if (zfs_lookup(&zfsmount, kname, &dn)) {
934 	printf("\nCan't find %s\n", kname);
935 	return;
936     }
937     off = 0;
938     if (xfsread(&dn, &off, &hdr, sizeof(hdr)))
939 	return;
940     if (N_GETMAGIC(hdr.ex) == ZMAGIC)
941 	fmt = 0;
942     else if (IS_ELF(hdr.eh))
943 	fmt = 1;
944     else {
945 	printf("Invalid %s\n", "format");
946 	return;
947     }
948     if (fmt == 0) {
949 	addr = hdr.ex.a_entry & 0xffffff;
950 	p = PTOV(addr);
951 	off = PAGE_SIZE;
952 	if (xfsread(&dn, &off, p, hdr.ex.a_text))
953 	    return;
954 	p += roundup2(hdr.ex.a_text, PAGE_SIZE);
955 	if (xfsread(&dn, &off, p, hdr.ex.a_data))
956 	    return;
957 	p += hdr.ex.a_data + roundup2(hdr.ex.a_bss, PAGE_SIZE);
958 	bootinfo.bi_symtab = VTOP(p);
959 	memcpy(p, &hdr.ex.a_syms, sizeof(hdr.ex.a_syms));
960 	p += sizeof(hdr.ex.a_syms);
961 	if (hdr.ex.a_syms) {
962 	    if (xfsread(&dn, &off, p, hdr.ex.a_syms))
963 		return;
964 	    p += hdr.ex.a_syms;
965 	    if (xfsread(&dn, &off, p, sizeof(int)))
966 		return;
967 	    x = *(uint32_t *)p;
968 	    p += sizeof(int);
969 	    x -= sizeof(int);
970 	    if (xfsread(&dn, &off, p, x))
971 		return;
972 	    p += x;
973 	}
974     } else {
975 	off = hdr.eh.e_phoff;
976 	for (j = i = 0; i < hdr.eh.e_phnum && j < 2; i++) {
977 	    if (xfsread(&dn, &off, ep + j, sizeof(ep[0])))
978 		return;
979 	    if (ep[j].p_type == PT_LOAD)
980 		j++;
981 	}
982 	for (i = 0; i < 2; i++) {
983 	    p = PTOV(ep[i].p_paddr & 0xffffff);
984 	    off = ep[i].p_offset;
985 	    if (xfsread(&dn, &off, p, ep[i].p_filesz))
986 		return;
987 	}
988 	p += roundup2(ep[1].p_memsz, PAGE_SIZE);
989 	bootinfo.bi_symtab = VTOP(p);
990 	if (hdr.eh.e_shnum == hdr.eh.e_shstrndx + 3) {
991 	    off = hdr.eh.e_shoff + sizeof(es[0]) *
992 		(hdr.eh.e_shstrndx + 1);
993 	    if (xfsread(&dn, &off, &es, sizeof(es)))
994 		return;
995 	    for (i = 0; i < 2; i++) {
996 		memcpy(p, &es[i].sh_size, sizeof(es[i].sh_size));
997 		p += sizeof(es[i].sh_size);
998 		off = es[i].sh_offset;
999 		if (xfsread(&dn, &off, p, es[i].sh_size))
1000 		    return;
1001 		p += es[i].sh_size;
1002 	    }
1003 	}
1004 	addr = hdr.eh.e_entry & 0xffffff;
1005     }
1006     bootinfo.bi_esymtab = VTOP(p);
1007     bootinfo.bi_kernelname = VTOP(kname);
1008     zfsargs.size = sizeof(zfsargs);
1009     zfsargs.pool = zfsmount.spa->spa_guid;
1010     zfsargs.root = zfsmount.rootobj;
1011     zfsargs.primary_pool = primary_spa->spa_guid;
1012 #ifdef LOADER_GELI_SUPPORT
1013     explicit_bzero(gelipw, sizeof(gelipw));
1014     export_geli_boot_data(&zfsargs.gelidata);
1015 #endif
1016     if (primary_vdev != NULL)
1017 	zfsargs.primary_vdev = primary_vdev->v_guid;
1018     else
1019 	printf("failed to detect primary vdev\n");
1020     /*
1021      * Note that the zfsargs struct is passed by value, not by pointer.  Code in
1022      * btxldr.S copies the values from the entry stack to a fixed location
1023      * within loader(8) at startup due to the presence of KARGS_FLAGS_EXTARG.
1024      */
1025     __exec((caddr_t)addr, RB_BOOTINFO | (opts & RBX_MASK),
1026 	   bootdev,
1027 	   KARGS_FLAGS_ZFS | KARGS_FLAGS_EXTARG,
1028 	   (uint32_t) spa->spa_guid,
1029 	   (uint32_t) (spa->spa_guid >> 32),
1030 	   VTOP(&bootinfo),
1031 	   zfsargs);
1032 }
1033 
1034 static int
1035 zfs_mount_ds(char *dsname)
1036 {
1037     uint64_t newroot;
1038     spa_t *newspa;
1039     char *q;
1040 
1041     q = strchr(dsname, '/');
1042     if (q)
1043 	*q++ = '\0';
1044     newspa = spa_find_by_name(dsname);
1045     if (newspa == NULL) {
1046 	printf("\nCan't find ZFS pool %s\n", dsname);
1047 	return -1;
1048     }
1049 
1050     if (zfs_spa_init(newspa))
1051 	return -1;
1052 
1053     newroot = 0;
1054     if (q) {
1055 	if (zfs_lookup_dataset(newspa, q, &newroot)) {
1056 	    printf("\nCan't find dataset %s in ZFS pool %s\n",
1057 		    q, newspa->spa_name);
1058 	    return -1;
1059 	}
1060     }
1061     if (zfs_mount(newspa, newroot, &zfsmount)) {
1062 	printf("\nCan't mount ZFS dataset\n");
1063 	return -1;
1064     }
1065     spa = newspa;
1066     return (0);
1067 }
1068 
1069 static int
1070 parse_cmd(void)
1071 {
1072     char *arg = cmd;
1073     char *ep, *p, *q;
1074     const char *cp;
1075     int c, i, j;
1076 
1077     while ((c = *arg++)) {
1078 	if (c == ' ' || c == '\t' || c == '\n')
1079 	    continue;
1080 	for (p = arg; *p && *p != '\n' && *p != ' ' && *p != '\t'; p++);
1081 	ep = p;
1082 	if (*p)
1083 	    *p++ = 0;
1084 	if (c == '-') {
1085 	    while ((c = *arg++)) {
1086 		if (c == 'P') {
1087 		    if (*(uint8_t *)PTOV(0x496) & 0x10) {
1088 			cp = "yes";
1089 		    } else {
1090 			opts |= OPT_SET(RBX_DUAL) | OPT_SET(RBX_SERIAL);
1091 			cp = "no";
1092 		    }
1093 		    printf("Keyboard: %s\n", cp);
1094 		    continue;
1095 		} else if (c == 'S') {
1096 		    j = 0;
1097 		    while ((unsigned int)(i = *arg++ - '0') <= 9)
1098 			j = j * 10 + i;
1099 		    if (j > 0 && i == -'0') {
1100 			comspeed = j;
1101 			break;
1102 		    }
1103 		    /* Fall through to error below ('S' not in optstr[]). */
1104 		}
1105 		for (i = 0; c != optstr[i]; i++)
1106 		    if (i == NOPT - 1)
1107 			return -1;
1108 		opts ^= OPT_SET(flags[i]);
1109 	    }
1110 	    ioctrl = OPT_CHECK(RBX_DUAL) ? (IO_SERIAL|IO_KEYBOARD) :
1111 		     OPT_CHECK(RBX_SERIAL) ? IO_SERIAL : IO_KEYBOARD;
1112 	    if (ioctrl & IO_SERIAL) {
1113 	        if (sio_init(115200 / comspeed) != 0)
1114 		    ioctrl &= ~IO_SERIAL;
1115 	    }
1116 	} if (c == '?') {
1117 	    dnode_phys_t dn;
1118 
1119 	    if (zfs_lookup(&zfsmount, arg, &dn) == 0) {
1120 		zap_list(spa, &dn);
1121 	    }
1122 	    return -1;
1123 	} else {
1124 	    arg--;
1125 
1126 	    /*
1127 	     * Report pool status if the comment is 'status'. Lets
1128 	     * hope no-one wants to load /status as a kernel.
1129 	     */
1130 	    if (!strcmp(arg, "status")) {
1131 		spa_all_status();
1132 		return -1;
1133 	    }
1134 
1135 	    /*
1136 	     * If there is "zfs:" prefix simply ignore it.
1137 	     */
1138 	    if (strncmp(arg, "zfs:", 4) == 0)
1139 		arg += 4;
1140 
1141 	    /*
1142 	     * If there is a colon, switch pools.
1143 	     */
1144 	    q = strchr(arg, ':');
1145 	    if (q) {
1146 		*q++ = '\0';
1147 		if (zfs_mount_ds(arg) != 0)
1148 		    return -1;
1149 		arg = q;
1150 	    }
1151 	    if ((i = ep - arg)) {
1152 		if ((size_t)i >= sizeof(kname))
1153 		    return -1;
1154 		memcpy(kname, arg, i + 1);
1155 	    }
1156 	}
1157 	arg = p;
1158     }
1159     return 0;
1160 }
1161