Category Archives: Linux

Build and package your own software for OpenWRT

8 Replies

Today I am going to discuss how to build and package your own software for OpenWRT.

When I say “your own software” in this case I am referring to a C program which you want to cross-compile for the target SoC and install using the opkg package manager included in OpenWRT.

The program I wrote is a little more complicated than your standard “Hello World” application. Here’s what I wanted to do:
1) use libconfig to read a configuration file in /etc/config/ and then perform actions based on the configuration described in this file
2) use sqlite3 to create a database
3) write some meaningful data to the database

Here’s the program flow:
1) Open /etc/config/example-sqlite and read the values into variables
2) Open (or create) a new SQLite3 database file at the location defined in the above configuration file
3) Determine if the SQLite file is initialized with the target table we want to write to, and if not, create the table
4) Write the system load average to the database
5) Quit

To recap, this program is different from “Hello World” in the following ways:
1) It must read and understand a configuration file in libconfig syntax; this requires linking against the libconfig library, which we must tell opkg is a dependency
2) It must create or open an SQLite 3 database; this requires linking against the sqlite3 library, which we must tell opkg is a depenedency
3) It must perform some useful operations on this SQLite file

Let’s start with compiling the C file on your native architecture. Sure, you can just use cc/gcc from bash, but this isn’t any good to OpenWRT SDK, which expects that each package will have a makefile which can be used to compile the software.

load2sqlite.c

#include <sys/types.h>
#include <string.h>
#include <stdio.h>
#include <stdlib.h>
#include <sqlite3.h>
#include <libconfig.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>

int main(int argc, char *argv[]) {
// ...

Most importantly above, we are including sqlite3.h for SQLite support, libconfig.h, and sys/stat.h, fcntl.h,errno.h to check if the SQLite3 database file exists or not.

You can compile this by hand quite easily, just by doing:
cc load2sqlite.c -lsqlite3 -lconfig -o load2sqlite

Okay, but how do we make this ready for OpenWRT SDK? By writing a makefile!

makefile

PROFILE = -O2 -s
CFLAGS = $(PROFILE)
LDFLAGS = -lsqlite3 -lconfig

all: main

# build it
main:
	$(CC) $(CFLAGS) load2sqlite.c $(LDFLAGS) -o load2sqlite

# clean it
clean:
	rm load2sqlite

Okay, so now if you type make in the directory, magically you will end up with an executable called load2sqlite!

But, this is a native binary, and it’s somewhat unlikely that your OpenWRT device is on the same architecture.

load2sqlite: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=9661b88e92b553d0556cbeeafccf04d2526c770f, stripped

If you run it, you’ll see that it looks for the sqlite database file, can’t find it, and so initalizes a new one with the “readings” table.

[hmartin@localhost src]$ ./load2sqlite 
Database file /tmp/sqlite3.db does not exist
Initialized database with readings table
[hmartin@localhost src]$ echo "select * from readings;" | sqlite3 /tmp/sqlite3.db 
2015-10-28 22:48:42|0.57|0.56|0.57

And if you run it again, without removing the SQLite3 file that was created, you’ll see this output:

[hmartin@localhost src]$ ./load2sqlite
SQLite database opened
Found readings table
[hmartin@localhost src]$ echo "select * from readings;" | sqlite3 /tmp/sqlite3.db 
2015-10-28 22:48:42|0.57|0.56|0.57
2015-10-28 22:49:00|0.47|0.54|0.57

Before we proceed further, I want to show you the directory structure so you have an idea of where we just were when we did this compilation. We are currently in the the src directory.

load2sqlite/
|-- Makefile
|-- README
`-- src
    |-- load2sqlite.c
    |-- load2sqlite.conf
    `-- makefile

Now let’s move up to the load2sqlite directory and work on the OpenWRT Makefile (seen above).

Here is the complete file, and then we will discuss it section by section:
Makefile

#
# Copyright (C) 2006-2015 OpenWrt.org
#
# This is free software, licensed under the GNU General Public License v2.
# See /LICENSE for more information.
#

include $(TOPDIR)/rules.mk

PKG_NAME:=load2sqlite
PKG_VERSION:=1.0.1
PKG_RELEASE:=5
PKG_MAINTAINER:=Hal Martin 
PKG_LICENSE:=GPL-2
PKG_CONFIG_DEPENDS:=libsqlite3 libconfig

include $(INCLUDE_DIR)/package.mk

PKG_BUILD_DIR := $(BUILD_DIR)/$(PKG_NAME)-$(PKG_VERSION)

TARGET_LDFLAGS+= \
  -Wl,-rpath-link=$(STAGING_DIR)/usr/lib \
  -Wl,-rpath-link=$(STAGING_DIR)/usr/lib/libconfig/lib \
  -Wl,-rpath-link=$(STAGING_DIR)/usr/lib/sqlite/lib

define Package/load2sqlite
  SECTION:=utils
  CATEGORY:=Utilities
  DEPENDS:=+libsqlite3 +libconfig
  TITLE:=SQLite example program, creates or opens a user defined SQLite database
  URL:=https://github.com/halmartin/load2sqlite
  MENU:=1
endef

define Package/load2sqlite/description
 Example SQLite is a sample program built using libsqlite3 and libconfig
 which creates or opens a user-defined SQLite3 database and performs some
 simple verification checks on the file to ensure that the target table (readings)
 exists, and if not creates the table, then inserts a row with the current system
 time, and the load (1 minute, 5 minute, 15 minute).
endef

define Build/Prepare
	mkdir -p $(PKG_BUILD_DIR)
	$(CP) ./src/* $(PKG_BUILD_DIR)/
endef

define Build/Configure
endef

define Build/Compile
	$(MAKE) -C $(PKG_BUILD_DIR) $(TARGET_CONFIGURE_OPTS)
endef

define Package/load2sqlite/install
	$(INSTALL_DIR) $(1)/bin
	$(INSTALL_BIN) $(PKG_BUILD_DIR)/load2sqlite $(1)/bin/
	$(INSTALL_DIR) $(1)/etc/config
	$(INSTALL_CONF) $(PKG_BUILD_DIR)/load2sqlite.conf $(1)/etc/config/load2sqlite
endef

$(eval $(call BuildPackage,load2sqlite))

If you clone the OpenWRT source and take a look at basically any package, you’ll see a Makefile that looks similar to the one above.

Let’s look at the package information:

PKG_NAME:=load2sqlite
PKG_VERSION:=1.0.1
PKG_RELEASE:=5
PKG_MAINTAINER:=Hal Martin 
PKG_LICENSE:=GPL-2

Here is where we define core details of our package, such as the name (e.g. what opkg will know it as), the version (useful for upgrading later), maintainer, and license.

TARGET_LDFLAGS+= \
  -Wl,-rpath-link=$(STAGING_DIR)/usr/lib \
  -Wl,-rpath-link=$(STAGING_DIR)/usr/lib/libconfig/lib \
  -Wl,-rpath-link=$(STAGING_DIR)/usr/lib/sqlite/lib

Since we want to build a program which links against external libraries, we must also tell the compiler where to find the header files for these libraries, so that the linking process does not fail during compilation. Above you can see that we are linking to libconfig and sqlite libraries.

define Package/load2sqlite
  SECTION:=utils
  CATEGORY:=Utilities
  DEPENDS:=+libsqlite3 +libconfig
  TITLE:=SQLite example program, creates or opens a user defined SQLite database
  URL:=https://github.com/halmartin/load2sqlite
  MENU:=1
endef

This is where you define the package for the OpenWRT build system and declare things like dependencies, and the description that will be present when you run menuconfig (which is how you will select your package to be built as part of an image).

Without declaring dependencies, you may find that you can build, package, and install your software, but it won’t run! So, by declaring the dependencies (packages which provide the libraries we link against) we ensure that when we type opkg install load2sqlite and libconfig and libsqlite3 are not installed, opkg knows to go and install them before installing our program. Now we can safely run the program because all the required libraries are installed on the device! 

define Build/Prepare
	mkdir -p $(PKG_BUILD_DIR)
	$(CP) ./src/* $(PKG_BUILD_DIR)/
endef

define Build/Configure
endef

Since our utility is quite simple, as *NIX software goes, the preparation steps are to create the build directory and copy the source from the source directory to the build directory. Since there is nothing to configure in our sample program, the configure step is empty (otherwise the OpenWRT build system will attempt to configure the package and fail because we haven’t bothered to implement this).

define Build/Compile
	$(MAKE) -C $(PKG_BUILD_DIR) $(TARGET_CONFIGURE_OPTS)
endef

define Package/load2sqlite/install
	$(INSTALL_DIR) $(1)/bin
	$(INSTALL_BIN) $(PKG_BUILD_DIR)/load2sqlite $(1)/bin/
	$(INSTALL_DIR) $(1)/etc/config
	$(INSTALL_CONF) $(PKG_BUILD_DIR)/load2sqlite.conf $(1)/etc/config/load2sqlite
endef

Finally, compile and install the software. As you can see above, I didn’t include an install directive in the makefile of the application, it is instead done manually within the OpenWRT Makefile. This is your choice, since I was designing this program specifically to run on OpenWRT, I saw no need to incorporate the installation steps in the makefile of the program.

And, finally:

$(eval $(call BuildPackage,load2sqlite))

This line is required for OpenWRT to build the package. Forget this line, and you will sit there wondering why your package is not being built!

Okay, now we have prepared our software to be built for OpenWRT. It would be stupid of me to get this far and not tell you how to compile it using the OpenWRT toolchain!

Following the excellent OpenWRT documentation, we need to set up a buildroot.

Install the dependencies (instructions for Debian/Ubuntu):

sudo apt-get install git-core build-essential libssl-dev libncurses5-dev unzip subversion mercurial

Clone the OpenWRT Chaos Calmer release:

git clone git://git.openwrt.org/15.05/openwrt.git

I find that the stock OpenWRT repository is a bit light on some of the software I like to have on my routers, so I take step 3 and install the additional feeds:

cd openwrt
./scripts/feeds update -a
./scripts/feeds install -a

Follow step 4 to ensure you have all the required dependencies installed on your host system!

make defconfig
make prereq
# don't forget to copy load2sqlite to package/utils/ before running this step, or the package won't appear in the menu!
make menuconfig

If everything has gone well thus far (e.g. no errors in the OpenWRT Makefile, and you put load2sqlite in package/utils/ then you should see the following in your menuconfig:

menuconfig_load2sqlite

menuconfig_load2sqlite_desc

Now I already have an official OpenWRT build installed on my router, so I don’t need to build an entire image, just the package I want to install. To do this, we must first build the cross compilation toolchain required to compile for a different CPU architecture.

Warning: the OpenWRT buildroot is around 6GB on disk, so ensure you have the necessary space before starting!

make tools/install
# this will take a while the first time
make toolchain/install
# this will also take a while the first time

When we have the tools and toolchain compiled, we can compile our package:

make package/load2sqlite/compile

This will create an ipkg file in bin/ramips/packages/base/load2sqlite_1.0.1-5_ramips_24kec.ipk which we need to copy to our router to install:

scp bin/ramips/packages/base/load2sqlite_1.0.1-5_ramips_24kec.ipk [email protected]:/tmp/
# scp completes
ssh [email protected]
[email protected]'s password:

BusyBox v1.23.2 (2015-07-25 03:03:02 CEST) built-in shell (ash)

  _______                     ________        __
 |       |.-----.-----.-----.|  |  |  |.----.|  |_
 |   -   ||  _  |  -__|     ||  |  |  ||   _||   _|
 |_______||   __|_____|__|__||________||__|  |____|
          |__| W I R E L E S S   F R E E D O M
 -----------------------------------------------------
 CHAOS CALMER (15.05, r46767)
 -----------------------------------------------------
  * 1 1/2 oz Gin            Shake with a glassful
  * 1/4 oz Triple Sec       of broken ice and pour
  * 3/4 oz Lime Juice       unstrained into a goblet.
  * 1 1/2 oz Orange Juice
  * 1 tsp. Grenadine Syrup
 -----------------------------------------------------
root@OpenWrt:~# opkg install /tmp/load2sqlite_1.0.1-5_ramips_24kec.ipk 
Installing load2sqlite (1.0.1-4) to root...
Installing libsqlite3 (3081101-1) to root...
Downloading http://downloads.openwrt.org/chaos_calmer/15.05/ramips/mt7620/packages/packages/libsqlite3_3081101-1_ramips_24kec.ipk.
Installing libpthread (0.9.33.2-1) to root...
Downloading http://downloads.openwrt.org/chaos_calmer/15.05/ramips/mt7620/packages/base/libpthread_0.9.33.2-1_ramips_24kec.ipk.
Installing libconfig (1.4.9-1) to root...
Downloading http://downloads.openwrt.org/chaos_calmer/15.05/ramips/mt7620/packages/base/libconfig_1.4.9-1_ramips_24kec.ipk.
Configuring libpthread.
Configuring libconfig.
Configuring libsqlite3.
Configuring load2sqlite.

Now that our package is installed, we can test it!

root@OpenWrt:~# /bin/load2sqlite 
Database file /tmp/sqlite3.db does not exist
Initialized database with readings table

If you install sqlite3-cli we can inspect the row added to the file:

root@OpenWrt:~# opkg install sqlite3-cli
root@OpenWrt:~# echo "select * from readings;" | sqlite3 /tmp/sqlite3.db 
2015-10-31 20:47:33|0.76|0.4|0.25

Since this is just an example program, it is one-shot (e.g. not a daemon). If you really do want to track the load of our OpenWRT router, just add /bin/load2sqlite to crontab (e.g. every hour) and you’ll have this tracking info in the SQLite database.

If you run it multiple times, you get another row added to the file each time the program is run:

root@OpenWrt:~# /bin/load2sqlite 
SQLite database opened
Found readings table
root@OpenWrt:~# echo "select * from readings;" | sqlite3 /tmp/sqlite3.db 
2015-10-31 20:47:33|0.76|0.4|0.25
2015-10-31 20:53:43|0.02|0.2|0.22
2015-10-31 21:23:19|0.08|0.04|0.05

Note that by default the file is saved to /tmp/, which on OpenWRT is a ramdisk. This means that the file will be lost when you reboot, or if you leave it running unattended for too long, the file size will grow to the point where the ramdisk will consume all available memory and the router will crash. For this reason, I suggest you modify the configuration file /etc/config/load2sqlite to point to non-volatile storage (such as a USB stick).

Source code: https://github.com/halmartin/load2sqlite

Why write another OpenWRT software guide?

Well, while I was looking for resources on how to build and package software for OpenWRT, I ran into a lot of posts about people compiling simple “Hello World” programs for OpenWRT, but for my particular use case, I wanted to utilize multiple libraries in my program, and I couldn’t find any good instructions on how to compile a program with linked libraries for OpenWRT.

Disclaimer: I’m not a C expert, so maybe there are some headers there which are not strictly necessary, but it works for me and the executable size is quite small.

If you wish to further reduce the size of your executable, you can tell the compiler to strip it of the symbol table and relocation information. Do this by appending -s to the PROFILE line in the makefile. When I did this on my laptop, the output went from 9.9KB to 7.0KB, or a savings of 30%

I have tested this on Chaos Calmer (15.05), and I expect the instructions would also work on Barrier Breaker (14.07) however I didn’t try this, so I cannot say certainly that it will work.

Building Linux 4.1 for the Banana Pi

2 Replies

This post is a follow up to my original post Building BananaPi LeMaker Kernel.

If you’re just looking for a vanilla Debian or Ubuntu image for your Banana Pi that utilizes a Linux kernel newer than 3.4.xxx, then stop reading and go to this page maintained by Igor Pečovnik. He provides pre-built Debian and Ubuntu images for a variety of Banana Pi boards.

If you want to manually build an image, he has put the build scripts he uses up on his GitHub repository. While I tried to do everything manually starting from my last post, I ended up building a kernel that would not boot. So I shamelessly stole the kernel configuration from Igor, and the resulting kernel boots.

The GMAC driver which required so much patching for the 3.4 kernel was mainlined in 3.17. As such, these instructions should work for any kernel newer than 3.17. I am building 4.1.3 in my script.

Here is the Jenkins/bash script to build the kernel, modules, and boot goodness you need (a direct link to the .sh file is at the end of the post):

if [ ! -d "linux-4.1.3" ]; then
wget https://www.kernel.org/pub/linux/kernel/v4.x/linux-4.1.3.tar.xz
tar -Jxvf linux-4.1.3.tar.xz
fi
cd linux-4.1.3
wget https://watchmysys.com/blog/wp-content/uploads/2015/07/banana-pi-linux-4.1.3-config.txt -O .config
make ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- clean
make -j4 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- LOADADDR=0x40008000 zImage dtbs
make ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- INSTALL_MOD_PATH=output modules
make ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- INSTALL_MOD_PATH=output modules_install
mkdir -p output/boot/
cp arch/arm/boot/zImage output/boot/
cp arch/arm/boot/dts/sun7i-a20-bananapi.dtb output/boot/
cat > output/boot/boot.cmd < output/boot/uEnv.txt << EOF
fatload mmc 0 0x46000000 zImage
fatload mmc 0 0x49000000 sun7i-a20-bananapi.dtb
setenv bootargs console=ttyS0,115200 [earlyprintk] root=/dev/mmcblk0p2 rootwait panic=10 rootfstype=ext4 rw ${extra}
bootz 0x46000000 - 0x49000000
EOF
mkimage -C none -A arm -T script -d output/boot/boot.cmd output/boot/boot.scr
cd ..
if [ ! -d "u-boot-2015.04" ]; then
wget ftp://ftp.denx.de/pub/u-boot/u-boot-2015.04.tar.bz2
tar -jxvf u-boot-2015.04.tar.bz2
fi
cd u-boot-2015.04
make -s CROSS_COMPILE=arm-linux-gnueabihf- clean
make ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf- Bananapi_defconfig
make -j4 ARCH=arm CROSS_COMPILE=arm-linux-gnueabihf-
cp u-boot-sunxi-with-spl.bin ../linux-4.1.3/output/boot/
cd ../linux-4.1.3/
cat > output/boot/uEnv.txt << EOF
fatload mmc 0 0x46000000 zImage
fatload mmc 0 0x49000000 sun7i-a20-bananapi.dtb
setenv bootargs console=ttyS0,115200 [earlyprintk] root=/dev/mmcblk0p2 rootwait panic=10 rootfstype=ext4 rw ${extra}
bootz 0x46000000 - 0x49000000
EOF
tar -C output -cjvf ../linux-bananapi-4.1.3.tar.bz2 boot/ lib/

Here is what the partition layout of my SDHC card:

root@bpi:~# fdisk -l /dev/mmcblk0

Disk /dev/mmcblk0: 7948 MB, 7948206080 bytes
4 heads, 16 sectors/track, 242560 cylinders, total 15523840 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disk identifier: 0x00000000

        Device Boot      Start         End      Blocks   Id  System
/dev/mmcblk0p1   *        2048      133119       65536    c  W95 FAT32 (LBA)
/dev/mmcblk0p2          133120    15523839     7695360   83  Linux

Here is the contents of the boot partition (mmcblk0p1, vfat):

root@bpi:~# ls /boot
boot.cmd  boot.scr  sun7i-a20-bananapi.dtb  uEnv.txt  zImage

boot.cmd

fatload mmc 0 0x46000000 zImage
fatload mmc 0 0x49000000 sun7i-a20-bananapi.dtb
setenv bootargs console=ttyS0,115200 earlyprintk root=/dev/mmcblk0p2 rw rootwait panic=10 
bootm 0x46000000 - 0x49000000

uEnv.txt

fatload mmc 0 0x46000000 zImage
fatload mmc 0 0x49000000 sun7i-a20-bananapi.dtb
setenv bootargs console=ttyS0,115200 [earlyprintk] root=/dev/mmcblk0p2 rootwait panic=10 rootfstype=ext4 rw ${extra}
bootz 0x46000000 - 0x49000000

You will need to update u-boot on the SD card to v2015.04. If you use the script I provide above, this file is in boot/u-boot-sunxi-with-spl.bin. You need to write it to the SD card using dd:

dd if=boot/u-boot-sunxi-with-spl.bin of=/dev/mmcblk0 bs=1024 seek=8

More notes:
I was unable to build this kernel running Debian 7 (Wheezy) because binutils is too old. Unfortunately the official repositories do not have a newer version available for Wheezy (slash) I was too lazy to look for a repository that might have a newer version. As such, I upgraded by Jenkins build box to Debian 8 (Jessie) to build this kernel.

Additionally, I had to upgrade from a 1GB SD Card in my Banana Pi to an 8GB SDHC Card because the new u-boot does not seem to like small (non-SDHC) cards.

Banana Pi info:

root@bpi:~# free -m
             total       used       free     shared    buffers     cached
Mem:           996         72        923          0          4         19
-/+ buffers/cache:         48        948
Swap:            0          0          0
root@bpi:~# uname -a
Linux bpi 4.1.3-bananapi #2 SMP Sun Jul 26 15:41:54 CEST 2015 armv7l GNU/Linux
root@bpi:~# lsmod
Module                  Size  Used by
root@bpi:~#

Speed test of the network interface:

# Laptop with Gigabit wired connection, saving to /tmp/ ramdisk
[derp@laptop ~]$ nc -lp 5000 | dd of=/tmp/zerofile
# Banana Pi
root@bpi:~# dd if=/dev/zero bs=1M count=2000 | nc laptop 5000
2000+0 records in
2000+0 records out
2097152000 bytes (2.1 GB) copied, 73.5536 s, 28.5 MB/s

Yes, it is a Gigabit link, but not the fastest. It does seem to be quite stable, and since I am not using my Banana Pi for a bandwidth intensive purpose, this speed is fine for me.

The build script does everything with relative paths, and can be run as a normal user. The output is a tar.bz2 archive containing u-boot binary, boot folder, and kernel modules. You will need sudo/root to install the u-boot bin file with dd as described above.

Download kernel .config: here
Download build script: here
Download u-boot 2015.04: here (SHA1SUM: 8bf4f738ba8aa18ab5d45fca324587f0749f7c10)
Download tar archive with u-boot, kernel, and modules: here (SHA1SUM: d94a8da66ce6d77a1ceb4569740cadf2c8c67e72)

Installing CentOS 7 with a chroot

2 Replies

I needed to install CentOS 7 on an embedded PC with UEFI and 2 SSD disks in mdadm RAID1.

While I’m sure the guys at Red Hat work very hard on CentOS, the installer is a piece of cr*p, especially when it comes to disk partitioning. I have never hated any installer more than the CentOS disk partitioner. I don’t know what happened. The disk partitioning tool in CentOS 6 installer was fine, I had no problems using it, but in the 7 installer it’s just a nightmare to do anything. In my opinion Windows installer does a better job of disk partitioning than the CentOS 7 installer.

While many people who like CentOS will proclaim that the error is between the keyboard and chair, I welcome them to provide a write-up and screenshots of how to accomplish my desired partitioning scheme using the graphical installer. If the instructions are shorter than this blog post, next time I won’t use a chroot to install.

I loosely based my method off of this post, but immediately found I had to deviate because I didn’t have USB install media with all the required commands on it.

Required tools:
1) USB stick with some Linux distro on it (I prefer the Gentoo minimal installer, it’s small and it includes lots of useful utilities)
2) USB stick with the CentOS minimal installer on it
3) About 2GB of free space, you can use a ramdisk, or create an extra partition using the free space on the USB sticks (CentOS occupies about 800MB, Gentoo about 300MB)

Steps:
1) Boot the Gentoo installer off the USB stick
2) Partition your disks however you like using gdisk, fdisk, or parted
3) Create your mdadm array(s)
4) Plug in the CentOS 7 USB stick and mount it to a temporary mount point (e.g. /tmp/centos)

Inside the CentOS USB stick you will find LiveOS/squashfs.img, you need to loopback mount this:

livecd ~ # mount -o loop /tmp/cinstall/LiveOS/squashfs.img /tmp/csquashfs/

Now we have yet another image to mount, this one within the squashfs image:

livecd ~ # mount -o loop /tmp/csquashfs/LiveOS/rootfs.img /tmp/croot

Finally we have a Linux filesystem. But unfortunately we cannot use it for anything as it is mounted read-only and there is no resolv.conf present, so no domain names can be resolved. This is why you need ~2GB of free space somewhere (or 4GB of RAM).

livecd ~ # mkfs -q /dev/ram1 1572864
livecd ~ # mkdir -p /tmp/ramdisk
livecd ~ # mount /dev/ram1 /tmp/ramdisk
livecd ~ # rsync -avHp /tmp/croot /tmp/ramdisk

Now that we have the installer rootfs somewhere writable, copy /etc/resolv.conf to the filesystem:

livecd ~ # cp /etc/resolv.conf /tmp/ramdisk/etc/resolv.conf

Mount your destination partition for CentOS somewhere you can access from within the chroot:

livecd ~ # mount /dev/vg0/centos /tmp/ramdisk/mnt

Chroot to the installer environment:

livecd ~ # chroot /tmp/ramdisk

Download the CentOS release RPM and install it to the destination partition:

bash-4.2# wget http://mirror.centos.org/centos/7/os/x86_64/Packages/centos-release-7-0.1406.el7.centos.2.3.x86_64.rpm
bash-4.2# rpm --root=/mnt --nodeps -i centos-release-7-0.1406.el7.centos.2.3.x86_64.rpm

Because yum is missing the yummain module in the installation environment, we need to download and install the yum RPM on the installer partition:

bash-4.2# wget http://mirror.centos.org/centos/7/os/x86_64/Packages/yum-3.4.3-118.el7.centos.noarch.rpm
bash-4.2# rpm -i --nodeps yum-3.4.3-118.el7.centos.noarch.rpm

Now finally we can run yum on the destination partition to install CentOS:

bash-4.2# yum --installroot=/mnt update
bash-4.2# yum --installroot=/mnt install -y yum
bash-4.2# yum --installroot=/mnt install -y @core kernel
bash-4.2# yum --installroot=/mnt install -y grub2-efi efibootmgr lvm2 mdadm \
dosfstools kernel

Now unfortunately I hit a small snag: the Gentoo installer isn’t EFI aware. Exit the chroot, but remember to copy /etc/resolv.conf to the destination partition:

livecd ~ # cp /etc/resolv.conf /tmp/ramdisk/mnt/etc/

Poweroff and unplug the Gentoo installer USB stick. Plug in the CentOS installer stick and boot to the rescue environment. Skip rootfs detection.

Mount the partition/LV slice containing your CentOS installation:

sh-4.2# mkdir /mnt/centos
sh-4.2# mount /dev/vg0/centos /mnt/centos
sh-4.2# mount -t proc proc /mnt/centos/proc
sh-4.2# mount --rbind /dev /mnt/centos/dev
sh-4.2# mount --rbind /sys /mnt/centos/sys
sh-4.2# chroot /mnt/centos /bin/bash

Save the mdadm array information to the mdadm.conf configuration file:

bash-4.2# mdadm --detail --scan > /etc/mdadm.conf

Format your EFI boot partition:

bash-4.2# mkfs.vfat /dev/sda1
bash-4.2# mkdir /boot/efi
bash-4.2# mount /dev/sda1 /boot/efi

Install grub:

bash-4.2# grub2-install
Installing for x86_64-efi platform
...
Installation finished. No error reported.

Now, dear reader, this is the part where you do not see the hours I spent debugging why dracut would not find my root partition (hint: see Fedora wiki for dracut debugging steps). The tl;dr is that the mdadm array wasn’t being assembled, for reasons still unknown. To solve this we need to add our array UUID as an additional kernel parameter for grub:

bash-4.2# MD_UUID=$(mdadm -D /dev/md0 | grep UUID | awk '{print $3}')
bash-4.2# grubby --update-kernel=/boot/vmlinuz-3.10.0-123.20.1.el7.x86_64 \
--args="rd_MD_UUID=$MD_UUID"
bash-4.2# grub2-mkconfig -o /boot/grub2/grub.cfg

Absolutely verify in /boot/grub2/grub.cfg that the correct rd_MD_UUID was appended to linuxefi, or like me, you may be left wondering why your system won’t boot.

Check with efibootmgr that a menu entry was created:

bash-4.2# efibootmgr -v
...
Boot0014* grub  HD(1,800,32000,SUPER-LONG-UUID)File(\EFI\grub\grubx64.efi)

Don’t forget to set a root password:

bash-4.2# passwd
Changing password for user root.
New password:
Retype new password:
passwd: all authentication tokens updated successfully

Configuring the hostname (using hostnamectl, or editing /etc/sysconfig/network and /etc/hostname), udev rules for eth* interface names, and static network configuration in /etc/sysconfig/network-scripts/ifcfg-eth* is left as an exercise for the reader.

After rebooting

If when you reboot you find that you cannot login as root using the password you specified, it’s probably SELinux. Normally I hate disabling SELinux, but in this case I was so tired of spending a day and a half debugging booting issues, I just disabled it and went on with the setup.

Despite what the wonderful CentOS installer tells you, you do not require a separate /boot partition. Right now this is what the partition layout looks like:

Disk /dev/sda: 128.0 GB, 128035676160 bytes, 250069680 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 4096 bytes
I/O size (minimum/optimal): 4096 bytes / 4096 bytes
Disk label type: gpt


#         Start          End    Size  Type            Name
 1         2048       206847    100M  EFI System      EFI System
 2       206848    250069646  119.1G  Linux RAID      Linux RAID

Because the contents of /boot/efi is static, and cannot be mdraid, remember to copy the contents of /dev/sda1 to /dev/sdb1 so that if your first drive ever dies, you will still have the required EFI components to boot off the second drive.

Overall, I can’t say this was a lot of fun. But I did learn a lot more about dracut and the CentOS booting process, and I can still say with confidence that I hate the CentOS 7 installer.