2 * Copyright (c) 1998 Robert Nordier
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
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
16 #include <sys/cdefs.h>
17 __FBSDID("$FreeBSD$");
19 #include <sys/param.h>
20 #include <sys/errno.h>
21 #include <sys/diskmbr.h>
25 #include <sys/reboot.h>
26 #include <sys/queue.h>
28 #include <machine/bootinfo.h>
29 #include <machine/elf.h>
30 #include <machine/pc/bios.h>
48 #define PATH_DOTCONFIG "/boot.config"
49 #define PATH_CONFIG "/boot/config"
50 #define PATH_BOOT3 "/boot/zfsloader"
51 #define PATH_KERNEL "/boot/kernel/kernel"
57 #define BIOS_NUMDRIVES 0x475
63 #define TYPE_MAXHARD TYPE_DA
69 static const uuid_t freebsd_zfs_uuid = GPT_ENT_TYPE_FREEBSD_ZFS;
71 static const char optstr[NOPT] = "DhaCcdgmnpqrsv"; /* Also 'P', 'S' */
72 static const unsigned char flags[NOPT] = {
90 static const char *const dev_nm[NDEV] = {"ad", "da", "fd"};
91 static const unsigned char dev_maj[NDEV] = {30, 4, 2};
94 static char cmddup[512];
95 static char kname[1024];
96 static char rootname[256];
97 static int comspeed = SIOSPD;
98 static struct bootinfo bootinfo;
99 static uint32_t bootdev;
100 static struct zfs_boot_args zfsargs;
101 static struct zfsmount zfsmount;
103 vm_offset_t high_heap_base;
104 uint32_t bios_basemem, bios_extmem, high_heap_size;
106 static struct bios_smap smap;
109 * The minimum amount of memory to reserve in bios_extmem for the heap.
111 #define HEAP_MIN (3 * 1024 * 1024)
113 static char *heap_next;
114 static char *heap_end;
116 /* Buffers that must not span a 64k boundary. */
117 #define READ_BUF_SIZE 8192
119 char rdbuf[READ_BUF_SIZE]; /* for reading large things */
120 char secbuf[READ_BUF_SIZE]; /* for MBR/disklabel */
122 static struct dmadat *dmadat;
125 static void load(void);
126 static int parse(void);
127 static void bios_getmem(void);
133 if (p + n > heap_end) {
134 printf("malloc failure\n");
144 strdup(const char *s)
146 char *p = malloc(strlen(s) + 1);
154 * Read from a dnode (which must be from a ZPL filesystem).
157 zfs_read(spa_t *spa, const dnode_phys_t *dnode, off_t *offp, void *start, size_t size)
159 const znode_phys_t *zp = (const znode_phys_t *) dnode->dn_bonus;
164 if (*offp + n > zp->zp_size)
165 n = zp->zp_size - *offp;
167 rc = dnode_read(spa, dnode, *offp, start, n);
179 static spa_t *primary_spa;
180 static vdev_t *primary_vdev;
183 * A wrapper for dskread that doesn't have to worry about whether the
184 * buffer pointer crosses a 64k boundary.
187 vdev_read(vdev_t *vdev, void *priv, off_t off, void *buf, size_t bytes)
192 struct dsk *dsk = (struct dsk *) priv;
194 if ((off & (DEV_BSIZE - 1)) || (bytes & (DEV_BSIZE - 1)))
198 lba = off / DEV_BSIZE;
201 nb = bytes / DEV_BSIZE;
202 if (nb > READ_BUF_SIZE / DEV_BSIZE)
203 nb = READ_BUF_SIZE / DEV_BSIZE;
204 if (drvread(dsk, dmadat->rdbuf, lba, nb))
206 memcpy(p, dmadat->rdbuf, nb * DEV_BSIZE);
209 bytes -= nb * DEV_BSIZE;
216 xfsread(const dnode_phys_t *dnode, off_t *offp, void *buf, size_t nbyte)
218 if ((size_t)zfs_read(spa, dnode, offp, buf, nbyte) != nbyte) {
219 printf("Invalid format\n");
230 /* Parse system memory map */
234 v86.addr = 0x15; /* int 0x15 function 0xe820*/
236 v86.ecx = sizeof(struct bios_smap);
238 v86.es = VTOPSEG(&smap);
239 v86.edi = VTOPOFF(&smap);
241 if ((v86.efl & 1) || (v86.eax != SMAP_SIG))
243 /* look for a low-memory segment that's large enough */
244 if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0) &&
245 (smap.length >= (512 * 1024)))
246 bios_basemem = smap.length;
247 /* look for the first segment in 'extended' memory */
248 if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base == 0x100000)) {
249 bios_extmem = smap.length;
253 * Look for the largest segment in 'extended' memory beyond
256 if ((smap.type == SMAP_TYPE_MEMORY) && (smap.base > 0x100000) &&
257 (smap.base < 0x100000000ull)) {
261 * If this segment crosses the 4GB boundary, truncate it.
263 if (smap.base + size > 0x100000000ull)
264 size = 0x100000000ull - smap.base;
266 if (size > high_heap_size) {
267 high_heap_size = size;
268 high_heap_base = smap.base;
271 } while (v86.ebx != 0);
273 /* Fall back to the old compatibility function for base memory */
274 if (bios_basemem == 0) {
276 v86.addr = 0x12; /* int 0x12 */
279 bios_basemem = (v86.eax & 0xffff) * 1024;
282 /* Fall back through several compatibility functions for extended memory */
283 if (bios_extmem == 0) {
285 v86.addr = 0x15; /* int 0x15 function 0xe801*/
288 if (!(v86.efl & 1)) {
289 bios_extmem = ((v86.ecx & 0xffff) + ((v86.edx & 0xffff) * 64)) * 1024;
292 if (bios_extmem == 0) {
294 v86.addr = 0x15; /* int 0x15 function 0x88*/
297 bios_extmem = (v86.eax & 0xffff) * 1024;
301 * If we have extended memory and did not find a suitable heap
302 * region in the SMAP, use the last 3MB of 'extended' memory as a
303 * high heap candidate.
305 if (bios_extmem >= HEAP_MIN && high_heap_size < HEAP_MIN) {
306 high_heap_size = HEAP_MIN;
307 high_heap_base = bios_extmem + 0x100000 - HEAP_MIN;
312 * Try to detect a device supported by the legacy int13 BIOS
315 int13probe(int drive)
323 if (!(v86.efl & 0x1) && /* carry clear */
324 ((v86.edx & 0xff) != (drive & DRV_MASK))) { /* unit # OK */
325 if ((v86.ecx & 0x3f) == 0) { /* absurd sector size */
326 return(0); /* skip device */
334 * We call this when we find a ZFS vdev - ZFS consumes the dsk
335 * structure so we must make a new one.
338 copy_dsk(struct dsk *dsk)
342 newdsk = malloc(sizeof(struct dsk));
348 probe_drive(struct dsk *dsk)
354 unsigned part, entries_per_sec;
356 struct dos_partition *dp;
361 * If we find a vdev on the whole disk, stop here. Otherwise dig
362 * out the partition table and probe each slice/partition
363 * in turn for a vdev.
365 if (vdev_probe(vdev_read, dsk, NULL) == 0)
368 sec = dmadat->secbuf;
373 * First check for GPT.
375 if (drvread(dsk, sec, 1, 1)) {
378 memcpy(&hdr, sec, sizeof(hdr));
379 if (memcmp(hdr.hdr_sig, GPT_HDR_SIG, sizeof(hdr.hdr_sig)) != 0 ||
380 hdr.hdr_lba_self != 1 || hdr.hdr_revision < 0x00010000 ||
381 hdr.hdr_entsz < sizeof(*ent) || DEV_BSIZE % hdr.hdr_entsz != 0) {
386 * Probe all GPT partitions for the presense of ZFS pools. We
387 * return the spa_t for the first we find (if requested). This
388 * will have the effect of booting from the first pool on the
391 entries_per_sec = DEV_BSIZE / hdr.hdr_entsz;
392 slba = hdr.hdr_lba_table;
393 elba = slba + hdr.hdr_entries / entries_per_sec;
394 while (slba < elba) {
396 if (drvread(dsk, sec, slba, 1))
398 for (part = 0; part < entries_per_sec; part++) {
399 ent = (struct gpt_ent *)(sec + part * hdr.hdr_entsz);
400 if (memcmp(&ent->ent_type, &freebsd_zfs_uuid,
401 sizeof(uuid_t)) == 0) {
402 dsk->start = ent->ent_lba_start;
403 if (vdev_probe(vdev_read, dsk, NULL) == 0) {
405 * This slice had a vdev. We need a new dsk
406 * structure now since the vdev now owns this one.
418 if (drvread(dsk, sec, DOSBBSECTOR, 1))
420 dp = (void *)(sec + DOSPARTOFF);
422 for (i = 0; i < NDOSPART; i++) {
425 dsk->start = dp[i].dp_start;
426 if (vdev_probe(vdev_read, dsk, NULL) == 0) {
428 * This slice had a vdev. We need a new dsk structure now
429 * since the vdev now owns this one.
444 dmadat = (void *)(roundup2(__base + (int32_t)&_end, 0x10000) - __base);
448 if (high_heap_size > 0) {
449 heap_end = PTOV(high_heap_base + high_heap_size);
450 heap_next = PTOV(high_heap_base);
452 heap_next = (char *) dmadat + sizeof(*dmadat);
453 heap_end = (char *) PTOV(bios_basemem);
456 dsk = malloc(sizeof(struct dsk));
457 dsk->drive = *(uint8_t *)PTOV(ARGS);
458 dsk->type = dsk->drive & DRV_HARD ? TYPE_AD : TYPE_FD;
459 dsk->unit = dsk->drive & DRV_MASK;
460 dsk->slice = *(uint8_t *)PTOV(ARGS + 1) + 1;
465 bootinfo.bi_version = BOOTINFO_VERSION;
466 bootinfo.bi_size = sizeof(bootinfo);
467 bootinfo.bi_basemem = bios_basemem / 1024;
468 bootinfo.bi_extmem = bios_extmem / 1024;
469 bootinfo.bi_memsizes_valid++;
470 bootinfo.bi_bios_dev = dsk->drive;
472 bootdev = MAKEBOOTDEV(dev_maj[dsk->type],
473 dsk->slice, dsk->unit, dsk->part),
475 /* Process configuration file */
482 * Probe the boot drive first - we will try to boot from whatever
483 * pool we find on that drive.
488 * Probe the rest of the drives that the bios knows about. This
489 * will find any other available pools and it may fill in missing
490 * vdevs for the boot pool.
493 for (i = 0; i < *(unsigned char *)PTOV(BIOS_NUMDRIVES); i++)
495 for (i = 0; i < MAXBDDEV; i++)
498 if ((i | DRV_HARD) == *(uint8_t *)PTOV(ARGS))
501 if (!int13probe(i | DRV_HARD))
504 dsk = malloc(sizeof(struct dsk));
505 dsk->drive = i | DRV_HARD;
506 dsk->type = dsk->drive & TYPE_AD;
516 * The first discovered pool, if any, is the pool.
518 spa = spa_get_primary();
520 printf("%s: No ZFS pools located, can't boot\n", BOOTPROG);
526 primary_vdev = spa_get_primary_vdev(spa);
528 if (zfs_spa_init(spa) != 0 || zfs_mount(spa, 0, &zfsmount) != 0) {
529 printf("%s: failed to mount default pool %s\n",
530 BOOTPROG, spa->spa_name);
532 } else if (zfs_lookup(&zfsmount, PATH_CONFIG, &dn) == 0 ||
533 zfs_lookup(&zfsmount, PATH_DOTCONFIG, &dn) == 0) {
535 zfs_read(spa, &dn, &off, cmd, sizeof(cmd));
540 * Note that parse() is destructive to cmd[] and we also want
541 * to honor RBX_QUIET option that could be present in cmd[].
543 memcpy(cmddup, cmd, sizeof(cmd));
546 if (!OPT_CHECK(RBX_QUIET))
547 printf("%s: %s\n", PATH_CONFIG, cmddup);
548 /* Do not process this command twice */
553 * Try to exec stage 3 boot loader. If interrupted by a keypress,
554 * or in case of failure, try to load a kernel directly instead.
557 if (autoboot && !*kname) {
558 memcpy(kname, PATH_BOOT3, sizeof(PATH_BOOT3));
561 memcpy(kname, PATH_KERNEL, sizeof(PATH_KERNEL));
565 /* Present the user with the boot2 prompt. */
568 if (!autoboot || !OPT_CHECK(RBX_QUIET)) {
569 printf("\nFreeBSD/x86 boot\n");
570 if (zfs_rlookup(spa, zfsmount.rootobj, rootname) != 0)
571 printf("Default: %s/<0x%llx>:%s\n"
573 spa->spa_name, zfsmount.rootobj, kname);
574 else if (rootname[0] != '\0')
575 printf("Default: %s/%s:%s\n"
577 spa->spa_name, rootname, kname);
579 printf("Default: %s:%s\n"
581 spa->spa_name, kname);
583 if (ioctrl & IO_SERIAL)
585 if (!autoboot || keyhit(5))
586 getstr(cmd, sizeof(cmd));
587 else if (!autoboot || !OPT_CHECK(RBX_QUIET))
597 /* XXX - Needed for btxld to link the boot2 binary; do not remove. */
610 static Elf32_Phdr ep[2];
611 static Elf32_Shdr es[2];
618 if (zfs_lookup(&zfsmount, kname, &dn)) {
619 printf("\nCan't find %s\n", kname);
623 if (xfsread(&dn, &off, &hdr, sizeof(hdr)))
625 if (N_GETMAGIC(hdr.ex) == ZMAGIC)
627 else if (IS_ELF(hdr.eh))
630 printf("Invalid %s\n", "format");
634 addr = hdr.ex.a_entry & 0xffffff;
637 if (xfsread(&dn, &off, p, hdr.ex.a_text))
639 p += roundup2(hdr.ex.a_text, PAGE_SIZE);
640 if (xfsread(&dn, &off, p, hdr.ex.a_data))
642 p += hdr.ex.a_data + roundup2(hdr.ex.a_bss, PAGE_SIZE);
643 bootinfo.bi_symtab = VTOP(p);
644 memcpy(p, &hdr.ex.a_syms, sizeof(hdr.ex.a_syms));
645 p += sizeof(hdr.ex.a_syms);
647 if (xfsread(&dn, &off, p, hdr.ex.a_syms))
650 if (xfsread(&dn, &off, p, sizeof(int)))
655 if (xfsread(&dn, &off, p, x))
660 off = hdr.eh.e_phoff;
661 for (j = i = 0; i < hdr.eh.e_phnum && j < 2; i++) {
662 if (xfsread(&dn, &off, ep + j, sizeof(ep[0])))
664 if (ep[j].p_type == PT_LOAD)
667 for (i = 0; i < 2; i++) {
668 p = PTOV(ep[i].p_paddr & 0xffffff);
669 off = ep[i].p_offset;
670 if (xfsread(&dn, &off, p, ep[i].p_filesz))
673 p += roundup2(ep[1].p_memsz, PAGE_SIZE);
674 bootinfo.bi_symtab = VTOP(p);
675 if (hdr.eh.e_shnum == hdr.eh.e_shstrndx + 3) {
676 off = hdr.eh.e_shoff + sizeof(es[0]) *
677 (hdr.eh.e_shstrndx + 1);
678 if (xfsread(&dn, &off, &es, sizeof(es)))
680 for (i = 0; i < 2; i++) {
681 memcpy(p, &es[i].sh_size, sizeof(es[i].sh_size));
682 p += sizeof(es[i].sh_size);
683 off = es[i].sh_offset;
684 if (xfsread(&dn, &off, p, es[i].sh_size))
689 addr = hdr.eh.e_entry & 0xffffff;
691 bootinfo.bi_esymtab = VTOP(p);
692 bootinfo.bi_kernelname = VTOP(kname);
693 zfsargs.size = sizeof(zfsargs);
694 zfsargs.pool = zfsmount.spa->spa_guid;
695 zfsargs.root = zfsmount.rootobj;
696 zfsargs.primary_pool = primary_spa->spa_guid;
697 if (primary_vdev != NULL)
698 zfsargs.primary_vdev = primary_vdev->v_guid;
700 printf("failed to detect primary vdev\n");
701 __exec((caddr_t)addr, RB_BOOTINFO | (opts & RBX_MASK),
703 KARGS_FLAGS_ZFS | KARGS_FLAGS_EXTARG,
704 (uint32_t) spa->spa_guid,
705 (uint32_t) (spa->spa_guid >> 32),
711 zfs_mount_ds(char *dsname)
717 q = strchr(dsname, '/');
720 newspa = spa_find_by_name(dsname);
721 if (newspa == NULL) {
722 printf("\nCan't find ZFS pool %s\n", dsname);
726 if (zfs_spa_init(newspa))
731 if (zfs_lookup_dataset(newspa, q, &newroot)) {
732 printf("\nCan't find dataset %s in ZFS pool %s\n",
733 q, newspa->spa_name);
737 if (zfs_mount(newspa, newroot, &zfsmount)) {
738 printf("\nCan't mount ZFS dataset\n");
753 while ((c = *arg++)) {
754 if (c == ' ' || c == '\t' || c == '\n')
756 for (p = arg; *p && *p != '\n' && *p != ' ' && *p != '\t'; p++);
761 while ((c = *arg++)) {
763 if (*(uint8_t *)PTOV(0x496) & 0x10) {
766 opts |= OPT_SET(RBX_DUAL) | OPT_SET(RBX_SERIAL);
769 printf("Keyboard: %s\n", cp);
771 } else if (c == 'S') {
773 while ((unsigned int)(i = *arg++ - '0') <= 9)
775 if (j > 0 && i == -'0') {
779 /* Fall through to error below ('S' not in optstr[]). */
781 for (i = 0; c != optstr[i]; i++)
784 opts ^= OPT_SET(flags[i]);
786 ioctrl = OPT_CHECK(RBX_DUAL) ? (IO_SERIAL|IO_KEYBOARD) :
787 OPT_CHECK(RBX_SERIAL) ? IO_SERIAL : IO_KEYBOARD;
788 if (ioctrl & IO_SERIAL) {
789 if (sio_init(115200 / comspeed) != 0)
790 ioctrl &= ~IO_SERIAL;
795 if (zfs_lookup(&zfsmount, arg, &dn) == 0) {
803 * Report pool status if the comment is 'status'. Lets
804 * hope no-one wants to load /status as a kernel.
806 if (!strcmp(arg, "status")) {
812 * If there is "zfs:" prefix simply ignore it.
814 if (strncmp(arg, "zfs:", 4) == 0)
818 * If there is a colon, switch pools.
820 q = strchr(arg, ':');
823 if (zfs_mount_ds(arg) != 0)
827 if ((i = ep - arg)) {
828 if ((size_t)i >= sizeof(kname))
830 memcpy(kname, arg, i + 1);