5561 support root pools on EFI/GPT partitioned disks

5125 update zpool/libzfs to manage bootable whole disk pools (EFI/GPT labeled disks)
Reviewed by: Jean McCormack <jean.mccormack@nexenta.com>
Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
Approved by: Dan McDonald <danmcd@omniti.com>

4 files changed, 94 insertions, 26 deletions

diff --git a/usr/src/cmd/boot/bootadm/bootadm.c b/usr/src/cmd/boot/bootadm/bootadm.c
index c9466725f5..454b266d45 100644
--- a/usr/src/cmd/boot/bootadm/bootadm.c
+++ b/usr/src/cmd/boot/bootadm/bootadm.c
@@ -24,7 +24,7 @@
  */
 
 /*
- * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
  */
 
 /*
@@ -116,6 +116,9 @@ typedef struct {
 #define	ENTRY_INIT	-1	/* entryNum initial value */
 #define	ALL_ENTRIES	-2	/* selects all boot entries */
 
+#define	PARTNO_NOTFOUND -1	/* Solaris partition not found */
+#define	PARTNO_EFI	-2	/* EFI partition table found */
+
 #define	GRUB_DIR		"/boot/grub"
 #define	GRUB_STAGE2		GRUB_DIR "/stage2"
 #define	GRUB_MENU		"/boot/grub/menu.lst"
@@ -4916,14 +4919,14 @@ create_diskmap(char *osroot)
 static int
 get_partition(char *device)
 {
-	int i, fd, is_pcfs, partno = -1;
+	int i, fd, is_pcfs, partno = PARTNO_NOTFOUND;
 	struct mboot *mboot;
 	char boot_sect[SECTOR_SIZE];
 	char *wholedisk, *slice;
 #ifdef i386
 	ext_part_t *epp;
 	uint32_t secnum, numsec;
-	int rval, pno, ext_partno = -1;
+	int rval, pno, ext_partno = PARTNO_NOTFOUND;
 #endif
 
 	/* form whole disk (p0) */
@@ -4979,6 +4982,11 @@ get_partition(char *device)
 				break;
 			}
 		} else {	/* look for solaris partition, old and new */
+			if (part->systid == EFI_PMBR) {
+				partno = PARTNO_EFI;
+				break;
+			}
+
 #ifdef i386
 			if ((part->systid == SUNIXOS &&
 			    (fdisk_is_linux_swap(epp, part->relsect,
@@ -4999,7 +5007,7 @@ get_partition(char *device)
 	}
 #ifdef i386
 	/* If no primary solaris partition, check extended partition */
-	if ((partno == -1) && (ext_partno != -1)) {
+	if ((partno == PARTNO_NOTFOUND) && (ext_partno != PARTNO_NOTFOUND)) {
 		rval = fdisk_get_solaris_part(epp, &pno, &secnum, &numsec);
 		if (rval == FDISK_SUCCESS) {
 			partno = pno - 1;
@@ -5072,13 +5080,18 @@ get_grubroot(char *osroot, char *osdev, char *menu_root)
 	}
 
 	fdiskpart = get_partition(osdev);
-	INJECT_ERROR1("GRUBROOT_FDISK_FAIL", fdiskpart = -1);
-	if (fdiskpart == -1) {
+	INJECT_ERROR1("GRUBROOT_FDISK_FAIL", fdiskpart = PARTNO_NOTFOUND);
+	if (fdiskpart == PARTNO_NOTFOUND) {
 		bam_error(FDISKPART_FAIL, osdev);
 		return (NULL);
 	}
 
 	grubroot = s_calloc(1, 10);
+	if (fdiskpart == PARTNO_EFI) {
+		fdiskpart = atoi(&slice[1]);
+		slice = NULL;
+	}
+
 	if (slice) {
 		(void) snprintf(grubroot, 10, "(hd%s,%d,%c)",
 		    grubhd, fdiskpart, slice[1] + 'a' - '0');
@@ -7103,8 +7116,8 @@ get_grubsign(char *osroot, char *osdev)
 		bam_print(GRUBSIGN_FOUND_OR_CREATED, sign, osdev);
 
 	fdiskpart = get_partition(osdev);
-	INJECT_ERROR1("GET_GRUBSIGN_FDISK", fdiskpart = -1);
-	if (fdiskpart == -1) {
+	INJECT_ERROR1("GET_GRUBSIGN_FDISK", fdiskpart = PARTNO_NOTFOUND);
+	if (fdiskpart == PARTNO_NOTFOUND) {
 		bam_error(FDISKPART_FAIL, osdev);
 		free(sign);
 		return (NULL);
@@ -7112,6 +7125,11 @@ get_grubsign(char *osroot, char *osdev)
 
 	slice = strrchr(osdev, 's');
 
+	if (fdiskpart == PARTNO_EFI) {
+		fdiskpart = atoi(&slice[1]);
+		slice = NULL;
+	}
+
 	grubsign = s_calloc(1, MAXNAMELEN + 10);
 	if (slice) {
 		(void) snprintf(grubsign, MAXNAMELEN + 10, "(%s,%d,%c)",
diff --git a/usr/src/cmd/boot/installgrub/Makefile b/usr/src/cmd/boot/installgrub/Makefile
index 4cc69eccd5..aecf0f7096 100644
--- a/usr/src/cmd/boot/installgrub/Makefile
+++ b/usr/src/cmd/boot/installgrub/Makefile
@@ -19,6 +19,7 @@
 # CDDL HEADER END
 #
 # Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+# Copyright 2015, Nexenta Systems, Inc.
 # 
 
 PROG= installgrub
@@ -45,7 +46,7 @@ LDLIBS += -lmd5
 i386_CFLAGS += -D_LARGEFILE64_SOURCE
 i386_CFLAGS += -D_FILE_OFFSET_BITS=64
 
-LDLIBS += -lfdisk
+LDLIBS += -lfdisk -lefi -lfstyp
 
 LINTFLAGS += \
 	-erroff=E_BAD_PTR_CAST_ALIGN \
diff --git a/usr/src/cmd/boot/installgrub/installgrub.c b/usr/src/cmd/boot/installgrub/installgrub.c
index c27dca802b..4d29d0236e 100644
--- a/usr/src/cmd/boot/installgrub/installgrub.c
+++ b/usr/src/cmd/boot/installgrub/installgrub.c
@@ -21,7 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2012 Milan Jurik. All rights reserved.
- * Copyright 2012 Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
  */
 
 #include <stdio.h>
@@ -48,6 +48,10 @@
 #include <sys/stat.h>
 #include <sys/multiboot.h>
 #include <sys/sysmacros.h>
+#include <sys/efi_partition.h>
+
+#include <libnvpair.h>
+#include <libfstyp.h>
 
 #include "message.h"
 #include "installgrub.h"
@@ -107,7 +111,7 @@ static void usage(char *);
 static int read_stage1_from_file(char *, ig_data_t *);
 static int read_stage2_from_file(char *, ig_data_t *);
 static int read_stage1_from_disk(int, char *);
-static int read_stage2_from_disk(int, ig_stage2_t *);
+static int read_stage2_from_disk(int, ig_stage2_t *, int);
 static int prepare_stage1(ig_data_t *);
 static int prepare_stage2(ig_data_t *, char *);
 static void prepare_fake_multiboot(ig_stage2_t *);
@@ -389,7 +393,7 @@ handle_getinfo(char *progname, char **argv)
 		goto out_dev;
 	}
 
-	ret = read_stage2_from_disk(device->part_fd, stage2);
+	ret = read_stage2_from_disk(device->part_fd, stage2, device->type);
 	if (ret == BC_ERROR) {
 		(void) fprintf(stderr, gettext("Error reading stage2 from "
 		    "%s\n"), device_path);
@@ -487,7 +491,8 @@ handle_mirror(char *progname, char **argv)
 		goto out_devs;
 	}
 
-	ret = read_stage2_from_disk(curr_device->part_fd, stage2_curr);
+	ret = read_stage2_from_disk(curr_device->part_fd, stage2_curr,
+	    curr_device->type);
 	if (ret == BC_ERROR) {
 		BOOT_DEBUG("Error reading first stage2 blocks from %s\n",
 		    curr_device->path);
@@ -623,6 +628,10 @@ propagate_bootblock(ig_data_t *source, ig_data_t *target, char *updt_str)
 static int
 init_device(ig_device_t *device, char *path)
 {
+	struct dk_gpt *vtoc;
+	fstyp_handle_t fhdl;
+	const char *fident;
+
 	bzero(device, sizeof (*device));
 	device->part_fd = -1;
 	device->disk_fd = -1;
@@ -655,9 +664,25 @@ init_device(ig_device_t *device, char *path)
 		return (BC_ERROR);
 	}
 
+	if (efi_alloc_and_read(device->disk_fd, &vtoc) > 0) {
+		device->type = IG_DEV_EFI;
+		efi_free(vtoc);
+	}
+
 	if (get_raw_partition_fd(device) != BC_SUCCESS)
 		return (BC_ERROR);
 
+	if (fstyp_init(device->part_fd, 0, NULL, &fhdl) != 0)
+		return (BC_ERROR);
+
+	if (fstyp_ident(fhdl, "zfs", &fident) != 0) {
+		fstyp_fini(fhdl);
+		(void) fprintf(stderr, gettext("Booting of EFI labeled disks "
+		    "is only supported with ZFS\n"));
+		return (BC_ERROR);
+	}
+	fstyp_fini(fhdl);
+
 	if (get_start_sector(device) != BC_SUCCESS)
 		return (BC_ERROR);
 
@@ -701,6 +726,21 @@ get_start_sector(ig_device_t *device)
 	struct part_info	dkpi;
 	struct extpart_info	edkpi;
 
+	if (is_efi(device->type)) {
+		struct dk_gpt *vtoc;
+
+		if (efi_alloc_and_read(device->disk_fd, &vtoc) <= 0)
+			return (BC_ERROR);
+
+		device->start_sector = vtoc->efi_parts[device->slice].p_start;
+		/* GPT doesn't use traditional slice letters */
+		device->slice = 0xff;
+		device->partition = 0;
+
+		efi_free(vtoc);
+		goto found_part;
+	}
+
 	mboot = (struct mboot *)device->boot_sector;
 
 	if (is_bootpar(device->type)) {
@@ -959,7 +999,8 @@ write_stage2(ig_data_t *install)
 	 * Note that we use stage2->buf rather than stage2->file, because we
 	 * may have extended information after the latter.
 	 */
-	offset = STAGE2_BLKOFF * SECTOR_SIZE;
+	offset = STAGE2_BLKOFF(device->type) * SECTOR_SIZE;
+
 	if (write_out(device->part_fd, stage2->buf, stage2->buf_size,
 	    offset) != BC_SUCCESS) {
 		perror("write");
@@ -968,7 +1009,7 @@ write_stage2(ig_data_t *install)
 
 	/* Simulate the "old" installgrub output. */
 	(void) fprintf(stdout, WRITE_STAGE2_DISK, device->partition,
-	    (stage2->buf_size / SECTOR_SIZE) + 1, STAGE2_BLKOFF,
+	    (stage2->buf_size / SECTOR_SIZE) + 1, STAGE2_BLKOFF(device->type),
 	    stage2->first_sector);
 
 	return (BC_SUCCESS);
@@ -1162,7 +1203,7 @@ read_stage1_from_disk(int dev_fd, char *stage1_buf)
 }
 
 static int
-read_stage2_from_disk(int dev_fd, ig_stage2_t *stage2)
+read_stage2_from_disk(int dev_fd, ig_stage2_t *stage2, int type)
 {
 	uint32_t		size;
 	uint32_t		buf_size;
@@ -1173,7 +1214,7 @@ read_stage2_from_disk(int dev_fd, ig_stage2_t *stage2)
 	assert(dev_fd != -1);
 
 	if (read_in(dev_fd, mboot_scan, sizeof (mboot_scan),
-	    STAGE2_BLKOFF * SECTOR_SIZE) != BC_SUCCESS) {
+	    STAGE2_BLKOFF(type) * SECTOR_SIZE) != BC_SUCCESS) {
 		perror(gettext("Error reading stage2 sectors"));
 		return (BC_ERROR);
 	}
@@ -1209,7 +1250,7 @@ read_stage2_from_disk(int dev_fd, ig_stage2_t *stage2)
 	}
 	stage2->buf_size = buf_size;
 
-	if (read_in(dev_fd, stage2->buf, buf_size, STAGE2_BLKOFF *
+	if (read_in(dev_fd, stage2->buf, buf_size, STAGE2_BLKOFF(type) *
 	    SECTOR_SIZE) != BC_SUCCESS) {
 		perror("read");
 		free(stage2->buf);
@@ -1243,7 +1284,8 @@ is_update_necessary(ig_data_t *data, char *updt_str)
 	bzero(&stage2_disk, sizeof (ig_stage2_t));
 
 	/* Gather stage2 (if present) from the target device. */
-	if (read_stage2_from_disk(dev_fd, &stage2_disk) != BC_SUCCESS) {
+	if (read_stage2_from_disk(dev_fd, &stage2_disk, device->type)
+	    != BC_SUCCESS) {
 		BOOT_DEBUG("Unable to read stage2 from %s\n", device->path);
 		BOOT_DEBUG("No multiboot wrapped stage2 on %s\n", device->path);
 		return (B_TRUE);
@@ -1367,7 +1409,8 @@ prepare_stage2(ig_data_t *install, char *updt_str)
 		}
 	} else {
 		/* Solaris VTOC */
-		stage2->first_sector = device->start_sector + STAGE2_BLKOFF;
+		stage2->first_sector = device->start_sector +
+		    STAGE2_BLKOFF(device->type);
 		BOOT_DEBUG("stage2 first sector: %d\n", stage2->first_sector);
 		/*
 		 * In a solaris partition, stage2 is written to contiguous
@@ -1441,15 +1484,18 @@ get_raw_partition_path(ig_device_t *device)
 	}
 
 	len = strlen(raw);
-	if (raw[len - 2] != 's' || raw[len - 1] == '2') {
+	if (!is_efi(device->type) &&
+	    (raw[len - 2] != 's' || raw[len - 1] == '2')) {
 		(void) fprintf(stderr, NOT_ROOT_SLICE);
 		free(raw);
 		return (NULL);
 	}
 	device->slice = atoi(&raw[len - 1]);
 
-	raw[len - 2] = 's';
-	raw[len - 1] = '2';
+	if (!is_efi(device->type)) {
+		raw[len - 2] = 's';
+		raw[len - 1] = '2';
+	}
 
 	return (raw);
 }
diff --git a/usr/src/cmd/boot/installgrub/installgrub.h b/usr/src/cmd/boot/installgrub/installgrub.h
index af6e60b973..037ffad4ce 100644
--- a/usr/src/cmd/boot/installgrub/installgrub.h
+++ b/usr/src/cmd/boot/installgrub/installgrub.h
@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright 2012 Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
  */
 
 #ifndef	_INSTALLGRUB_H
@@ -68,10 +68,12 @@ typedef struct _ig_data {
 
 enum ig_devtype_t {
 	IG_DEV_X86BOOTPAR = 1,
-	IG_DEV_SOLVTOC
+	IG_DEV_SOLVTOC,
+	IG_DEV_EFI
 };
 
 #define	is_bootpar(type)	(type == IG_DEV_X86BOOTPAR)
+#define	is_efi(type)		(type == IG_DEV_EFI)
 
 #define	STAGE2_MEMADDR		(0x8000)	/* loading addr of stage2 */
 
@@ -86,7 +88,8 @@ enum ig_devtype_t {
 #define	STAGE2_BLOCKLIST	(SECTOR_SIZE - 0x8)
 #define	STAGE2_INSTALLPART	(SECTOR_SIZE + 0x8)
 #define	STAGE2_FORCE_LBA	(SECTOR_SIZE + 0x11)
-#define	STAGE2_BLKOFF		(50)	/* offset from start of fdisk part */
+#define	STAGE2_BLKOFF(type)	\
+	(is_efi(type) ? 1024 : 50)	/* offset from start of part */
 
 #ifdef	__cplusplus
 }