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, 154 insertions, 129 deletions

diff --git a/usr/src/lib/libbe/Makefile.com b/usr/src/lib/libbe/Makefile.com
index b71a0f85b5..bc04dd8a47 100644
--- a/usr/src/lib/libbe/Makefile.com
+++ b/usr/src/lib/libbe/Makefile.com
@@ -22,8 +22,8 @@
 #
 # Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
 #
-# Copyright 2011 Nexenta Systems, Inc. All rights reserved.
 # Copyright 2012 OmniTI Computer Consulting, Inc.  All rights reserved.
+# Copyright 2015 Nexenta Systems, Inc. All rights reserved.
 #
 
 
@@ -50,7 +50,7 @@ INCS += -I$(SRCDIR)
 
 C99MODE= 	$(C99_ENABLE)
 
-LDLIBS +=	-lzfs -linstzones -luuid -lnvpair -lc -lgen -ldevinfo
+LDLIBS +=	-lzfs -linstzones -luuid -lnvpair -lc -lgen -ldevinfo -lefi
 CPPFLAGS +=	$(INCS)
 CERRWARN +=	-_gcc=-Wno-unused-label
 CERRWARN +=	-_gcc=-Wno-uninitialized
diff --git a/usr/src/lib/libbe/common/be_activate.c b/usr/src/lib/libbe/common/be_activate.c
index 5b4555d924..da6ed3fb18 100644
--- a/usr/src/lib/libbe/common/be_activate.c
+++ b/usr/src/lib/libbe/common/be_activate.c
@@ -24,7 +24,7 @@
  */
 
 /*
- * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
  */
 
 #include <assert.h>
@@ -34,11 +34,14 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <strings.h>
 #include <errno.h>
 #include <sys/mnttab.h>
 #include <sys/types.h>
 #include <sys/stat.h>
+#include <fcntl.h>
 #include <unistd.h>
+#include <sys/efi_partition.h>
 
 #include <libbe.h>
 #include <libbe_priv.h>
@@ -50,6 +53,9 @@ char	*mnttab = MNTTAB;
  */
 static int set_bootfs(char *boot_rpool, char *be_root_ds);
 static int set_canmount(be_node_list_t *, char *);
+static boolean_t be_do_installgrub_mbr(char *, nvlist_t *);
+static int be_do_installgrub_helper(zpool_handle_t *, nvlist_t *, char *,
+    char *);
 static int be_do_installgrub(be_transaction_data_t *);
 static int be_get_grub_vers(be_transaction_data_t *, char **, char **);
 static int get_ver_from_capfile(char *, char **);
@@ -750,6 +756,136 @@ get_ver_from_capfile(char *file, char **vers)
 }
 
 /*
+ * To be able to boot EFI labeled disks, GRUB stage1 needs to be written
+ * into the MBR. We do not do this if we're on disks with a traditional
+ * fdisk partition table only, or if any foreign EFI partitions exist.
+ * In the trivial case of a whole-disk vdev we always write stage1 into
+ * the MBR.
+ */
+static boolean_t
+be_do_installgrub_mbr(char *diskname, nvlist_t *child)
+{
+	struct uuid allowed_uuids[] = {
+		EFI_UNUSED,
+		EFI_RESV1,
+		EFI_BOOT,
+		EFI_ROOT,
+		EFI_SWAP,
+		EFI_USR,
+		EFI_BACKUP,
+		EFI_RESV2,
+		EFI_VAR,
+		EFI_HOME,
+		EFI_ALTSCTR,
+		EFI_RESERVED,
+		EFI_SYSTEM,
+		EFI_BIOS_BOOT,
+		EFI_SYMC_PUB,
+		EFI_SYMC_CDS
+	};
+
+	uint64_t whole;
+	struct dk_gpt *gpt;
+	struct uuid *u;
+	int fd, npart, i, j;
+
+	(void) nvlist_lookup_uint64(child, ZPOOL_CONFIG_WHOLE_DISK,
+	    &whole);
+
+	if (whole)
+		return (B_TRUE);
+
+	if ((fd = open(diskname, O_RDONLY|O_NDELAY)) < 0)
+		return (B_FALSE);
+
+	if ((npart = efi_alloc_and_read(fd, &gpt)) <= 0)
+		return (B_FALSE);
+
+	for (i = 0; i != npart; i++) {
+		int match = 0;
+
+		u = &gpt->efi_parts[i].p_guid;
+
+		for (j = 0;
+		    j != sizeof (allowed_uuids) / sizeof (struct uuid);
+		    j++)
+			if (bcmp(u, &allowed_uuids[j],
+			    sizeof (struct uuid)) == 0)
+				match++;
+
+		if (match == 0)
+			return (B_FALSE);
+	}
+
+	return (B_TRUE);
+}
+
+static int
+be_do_installgrub_helper(zpool_handle_t *zphp, nvlist_t *child, char *stage1,
+    char *stage2)
+{
+	char installgrub_cmd[MAXPATHLEN];
+	char be_run_cmd_errbuf[BUFSIZ];
+	char diskname[MAXPATHLEN];
+	char *vname;
+	char *path, *dsk_ptr;
+	char *m_flag = "";
+
+	if (nvlist_lookup_string(child, ZPOOL_CONFIG_PATH, &path) != 0) {
+		be_print_err(gettext("be_do_installgrub: "
+		    "failed to get device path\n"));
+		return (BE_ERR_NODEV);
+	}
+
+	/*
+	 * Modify the vdev path to point to the raw disk.
+	 */
+	path = strdup(path);
+	if (path == NULL)
+		return (BE_ERR_NOMEM);
+
+	dsk_ptr = strstr(path, "/dsk/");
+	if (dsk_ptr != NULL) {
+		*dsk_ptr = '\0';
+		dsk_ptr++;
+	} else {
+		dsk_ptr = "";
+	}
+
+	(void) snprintf(diskname, sizeof (diskname), "%s/r%s", path, dsk_ptr);
+	free(path);
+
+	if (be_do_installgrub_mbr(diskname, child))
+		m_flag = "-m -f";
+
+	vname = zpool_vdev_name(g_zfs, zphp, child, B_FALSE);
+	if (vname == NULL) {
+		be_print_err(gettext("be_do_installgrub: "
+		    "failed to get device name: %s\n"),
+		    libzfs_error_description(g_zfs));
+		return (zfs_err_to_be_err(g_zfs));
+	}
+
+	(void) snprintf(installgrub_cmd, sizeof (installgrub_cmd),
+	    "%s %s %s %s %s", BE_INSTALL_GRUB, m_flag, stage1, stage2,
+	    diskname);
+	if (be_run_cmd(installgrub_cmd, be_run_cmd_errbuf, BUFSIZ, NULL, 0)
+	    != BE_SUCCESS) {
+		be_print_err(gettext("be_do_installgrub: installgrub "
+		    "failed for device %s.\n"), vname);
+		/* Assume localized cmd err output. */
+		be_print_err(gettext("  Command: \"%s\"\n"),
+		    installgrub_cmd);
+		be_print_err("%s", be_run_cmd_errbuf);
+		free(vname);
+		return (BE_ERR_BOOTFILE_INST);
+	}
+	free(vname);
+
+	return (BE_SUCCESS);
+}
+
+/*
  * Function:	be_do_installgrub
  * Description:	This function runs installgrub using the grub loader files
  *              from the BE we're activating and installing them on the
@@ -782,9 +918,7 @@ be_do_installgrub(be_transaction_data_t *bt)
 	char zpool_cap_file[MAXPATHLEN];
 	char stage1[MAXPATHLEN];
 	char stage2[MAXPATHLEN];
-	char installgrub_cmd[MAXPATHLEN];
 	char *vname;
-	char be_run_cmd_errbuf[BUFSIZ];
 	int ret = BE_SUCCESS;
 	int err = 0;
 	boolean_t be_mounted = B_FALSE;
@@ -868,6 +1002,7 @@ be_do_installgrub(be_transaction_data_t *bt)
 			goto done;
 		}
 		if (strcmp(vname, "mirror") == 0 || vname[0] != 'c') {
+			free(vname);
 
 			if (nvlist_lookup_nvlist_array(child[c],
 			    ZPOOL_CONFIG_CHILDREN, &nvchild, &nchildren) != 0) {
@@ -879,56 +1014,18 @@ be_do_installgrub(be_transaction_data_t *bt)
 			}
 
 			for (i = 0; i < nchildren; i++) {
-				vname = zpool_vdev_name(g_zfs, zphp,
-				    nvchild[i], B_FALSE);
-				if (vname == NULL) {
-					be_print_err(gettext(
-					    "be_do_installgrub: "
-					    "failed to get device name: %s\n"),
-					    libzfs_error_description(g_zfs));
-					ret = zfs_err_to_be_err(g_zfs);
+				ret = be_do_installgrub_helper(zphp, nvchild[i],
+				    stage1, stage2);
+				if (ret != BE_SUCCESS)
 					goto done;
-				}
-
-				(void) snprintf(installgrub_cmd,
-				    sizeof (installgrub_cmd),
-				    "%s %s %s /dev/rdsk/%s",
-				    BE_INSTALL_GRUB, stage1, stage2, vname);
-				if (be_run_cmd(installgrub_cmd,
-				    be_run_cmd_errbuf, BUFSIZ, NULL, 0) !=
-				    BE_SUCCESS) {
-					be_print_err(gettext(
-					    "be_do_installgrub: installgrub "
-					    "failed for device %s.\n"), vname);
-					/* Assume localized cmd err output. */
-					be_print_err(gettext(
-					    "  Command: \"%s\"\n"),
-					    installgrub_cmd);
-					be_print_err("%s", be_run_cmd_errbuf);
-					free(vname);
-					ret = BE_ERR_BOOTFILE_INST;
-					goto done;
-				}
-				free(vname);
 			}
 		} else {
-			(void) snprintf(installgrub_cmd,
-			    sizeof (installgrub_cmd), "%s %s %s /dev/rdsk/%s",
-			    BE_INSTALL_GRUB, stage1, stage2, vname);
-			if (be_run_cmd(installgrub_cmd, be_run_cmd_errbuf,
-			    BUFSIZ, NULL, 0) != BE_SUCCESS) {
-				be_print_err(gettext(
-				    "be_do_installgrub: installgrub "
-				    "failed for device %s.\n"), vname);
-				/* Assume localized cmd err output. */
-				be_print_err(gettext("  Command: \"%s\"\n"),
-				    installgrub_cmd);
-				be_print_err("%s", be_run_cmd_errbuf);
-				free(vname);
-				ret = BE_ERR_BOOTFILE_INST;
-				goto done;
-			}
 			free(vname);
+
+			ret = be_do_installgrub_helper(zphp, child[c], stage1,
+			    stage2);
+			if (ret != BE_SUCCESS)
+				goto done;
 		}
 	}
 
diff --git a/usr/src/lib/libgrubmgmt/common/libgrub_cmd.c b/usr/src/lib/libgrubmgmt/common/libgrub_cmd.c
index 85304d45e7..5dc12247b1 100644
--- a/usr/src/lib/libgrubmgmt/common/libgrub_cmd.c
+++ b/usr/src/lib/libgrubmgmt/common/libgrub_cmd.c
@@ -22,6 +22,9 @@
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
+/*
+ * Copyright 2015 Nexenta Systems, Inc.
+ */
 
 /*
  * This file contains all the functions that implement the following
@@ -399,7 +402,7 @@ findroot(const grub_line_t *lp, grub_barg_t *barg)
 			return (EG_FINDROOTFMT);
 
 		++pos;
-		if (pos[0] != ',' ||
+		if ((pos[0] != ',' && pos[0] != ')') ||
 		    !IS_SLCNUM_VALID(barg->gb_slcnum = pos[1]) ||
 		    pos[2] != ')')
 			return (EG_FINDROOTFMT);
diff --git a/usr/src/lib/libzfs/common/libzfs_pool.c b/usr/src/lib/libzfs/common/libzfs_pool.c
index 4f8d0b6148..0418091025 100644
--- a/usr/src/lib/libzfs/common/libzfs_pool.c
+++ b/usr/src/lib/libzfs/common/libzfs_pool.c
@@ -373,27 +373,6 @@ bootfs_name_valid(const char *pool, char *bootfs)
 	return (B_FALSE);
 }
 
-/*
- * Inspect the configuration to determine if any of the devices contain
- * an EFI label.
- */
-static boolean_t
-pool_uses_efi(nvlist_t *config)
-{
-	nvlist_t **child;
-	uint_t c, children;
-
-	if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
-	    &child, &children) != 0)
-		return (read_efi_label(config, NULL) >= 0);
-
-	for (c = 0; c < children; c++) {
-		if (pool_uses_efi(child[c]))
-			return (B_TRUE);
-	}
-	return (B_FALSE);
-}
-
 boolean_t
 zpool_is_bootable(zpool_handle_t *zhp)
 {
@@ -422,7 +401,6 @@ zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 	char *slash, *check;
 	struct stat64 statbuf;
 	zpool_handle_t *zhp;
-	nvlist_t *nvroot;
 
 	if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 		(void) no_memory(hdl);
@@ -541,21 +519,6 @@ zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 				(void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 				goto error;
 			}
-			verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
-			    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
-
-			/*
-			 * bootfs property cannot be set on a disk which has
-			 * been EFI labeled.
-			 */
-			if (pool_uses_efi(nvroot)) {
-				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
-				    "property '%s' not supported on "
-				    "EFI labeled devices"), propname);
-				(void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
-				zpool_close(zhp);
-				goto error;
-			}
 			zpool_close(zhp);
 			break;
 
@@ -1278,25 +1241,6 @@ zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
 		return (zfs_error(hdl, EZFS_BADVERSION, msg));
 	}
 
-	if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
-	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
-		uint64_t s;
-
-		for (s = 0; s < nspares; s++) {
-			char *path;
-
-			if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
-			    &path) == 0 && pool_uses_efi(spares[s])) {
-				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
-				    "device '%s' contains an EFI label and "
-				    "cannot be used on root pools."),
-				    zpool_vdev_name(hdl, NULL, spares[s],
-				    B_FALSE));
-				return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
-			}
-		}
-	}
-
 	if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
 	    SPA_VERSION_L2CACHE &&
 	    nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
@@ -2311,11 +2255,9 @@ zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
 		return (EZFS_INVALCONFIG);
 
 	/*
-	 * root pool can not have EFI labeled disks and can only have
-	 * a single top-level vdev.
+	 * root pool can only have a single top-level vdev.
 	 */
-	if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
-	    pool_uses_efi(vdev_root))
+	if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1)
 		return (EZFS_POOL_INVALARG);
 
 	(void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
@@ -2617,16 +2559,6 @@ zpool_vdev_attach(zpool_handle_t *zhp,
 		(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
 		    "cannot attach %s to %s"), new_disk, old_disk);
 
-	/*
-	 * If this is a root pool, make sure that we're not attaching an
-	 * EFI labeled device.
-	 */
-	if (rootpool && pool_uses_efi(nvroot)) {
-		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
-		    "EFI labeled devices are not supported on root pools."));
-		return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
-	}
-
 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 	if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
 	    &islog)) == 0)
@@ -3902,13 +3834,6 @@ zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
 	if (zhp) {
 		nvlist_t *nvroot;
 
-		if (zpool_is_bootable(zhp)) {
-			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
-			    "EFI labeled devices are not supported on root "
-			    "pools."));
-			return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
-		}
-
 		verify(nvlist_lookup_nvlist(zhp->zpool_config,
 		    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);