wiki:linux/ubi

Context Navigation

Version 13 (modified by Bobby Jones, 6 years ago) ( diff )
Removed extra "for" loop renaming, which is only necessary when moving the .gz extension on compressed part files

Unsorted Block Images (UBI)
FLASH chip geometry
UBI use on Gateworks Ventana

Unsorted Block Images (UBI)

Unsorted block images (UBI) is an erase block management layer for flash memory devices such as raw NAND devices. UBI serves two purposes:

tracking NAND flash bad blocks
providing wear leveling

UBI was written specifically for UBIFS so that the filesystem does not have to deal with these issues. UBIFS is a successor to JFFS2 and a competitor to LogFS.

References:

FLASH chip geometry

Because UBI takes care of bad block tracking and wear leveling, it needs to know some details about the geometry of the raw FLASH device when creating an image.

The geometry data can be found in the datasheet for the specific flash device used. Gateworks uses two different sized devices which do not share geometries.

UBI use on Gateworks Ventana

UBI is used on the Gateworks Ventana for raw NAND storage. Prior Gateworks product families typically JFFS2.

The Gateworks Ventana default U-Boot environment scripts make a few assumptions about booting from NAND flash:

the mtdparts env var partitions the NAND device into sections for uboot (16MB as this accounts for the SPL, u-boot.img and redundancy for bad blocks), u-boot env (1MB), rootfs (the remainder of the device)
the 'rootfs' partition (defined by the mtdparts env var) contains a UBI device/image
the UBI device/image in the rootfs partition contains a volume also called 'rootfs'
an optional bootscript may exist in boot/6x_bootscript-ventana (the image env var) and if found will be sourced
a flattened device tree blob (dtb) may exist in boot/ (named by the automatic env vars fdt_file, fdt_file1, or fdt_file2)
a kernel may exist in boot/uImage

The above assumptions are not requirements - the U-Boot env scripts can be changed to suit your needs.

Creating a UBI using existing rootfs, kernel, and device-tree blobs

To create a UBI image suitable for booting on a standard Gateworks Ventana bootloader (taking into account the assumptions above) you can do the following on a Linux system provided you have a root filesystem (myrootfs.tar.gz), kernel (uImage), and device-tree dtb files (imx6*-gw*.dtb).

256MB FLASH devices

To create a UBI image suitable for the 256MB FLASH devices:

create a temporary directory for our rootfs and untar your filesystem to it
```
mkdir tmp
fakeroot tar --numeric-owner -xvf myrootfs.tar.gz -C tmp
```
- Note that fakeroot is used because this allows you preserve the ownership of files that you have in your rootfs such as root owned /dev files

copy your kernel files (and u-boot bootscript if you have one):

mkdir tmp/boot
cp uImage imx6*-gw*.dtb 6x_bootscript-ventana tmp/boot/

create a ubifs image for the 256MB device geometry:

mkfs.ubifs -F -m 2048 -e 124KiB -c 1912 -x zlib -o root.ubifs -d tmp

use ubinize to create ubi images for the 256MB device geometry which can be written to a suitable raw NAND devices in U-Boot using nand write:

# for the small layout device
cat <<EOF > ubinize.cfg
[ubifs]
mode=ubi
image=root.ubifs
vol_id=0
vol_type=dynamic
vol_name=rootfs
vol_flags=autoresize
EOF
ubinize -m 2048 -p 128KiB -s 2048 -o image.ubi ubinize.cfg

1GB or 2GB FLASH devices

To create a UBI image suitable for the 1GB or larger FLASH devices (Note that the parameters to mkfs.ubifs and ubinize change):

create a temporary directory for our rootfs and untar your filesystem to it
```
mkdir tmp
fakeroot tar --numeric-owner -xvf myrootfs.tar.gz -C tmp
```
- Note that fakeroot is used because this allows you preserve the ownership of files that you have in your rootfs such as root owned /dev files

copy your kernel files (and u-boot bootscript if you have one):

mkdir tmp/boot
cp uImage imx6*-gw*.dtb 6x_bootscript-ventana tmp/boot/

create ubifs images for the 'large' device geometry:

mkfs.ubifs -F -m 4096 -e 248KiB -c 8124 -x zlib -o root.ubifs -d tmp

use ubinize to create ubi image for the 'large' device geometry which can be written to a suitable raw NAND devices in U-Boot using nand write:

# for the small layout device
cat <<EOF > ubinize.cfg
[ubifs]
mode=ubi
image=root.ubifs #generate this image with mkfs.ubifs
vol_id=0
vol_type=dynamic
vol_name=rootfs
vol_flags=autoresize
EOF
ubinize -m 4096 -p 256KiB -s 4096 -o image.ubi ubinize.cfg

Creating your own .ubifs file

sudo mkfs.ubifs -d /your/rootfs/ -F -m 4096 -e 248KiB -c 8124 -x zlib -o A_large.ubifs

Essentials:

-r Build file system from directory DIR
-F The -F option causes mkfs.ubifs to set a special flag in the superblock, which triggers a "free space fixup" procedure in the kernel the very first time the filesystem is mounted. This fixup procedure involves finding all empty pages in the UBIFS file system and re-erasing them. This ensures that NAND pages which contain all 0xFF data get fully erased, which removes any problematic non-0xFF data from their OOB areas.
-m Minimum I/O unit size
-e Logical erase block size
-c Maximum logical erase block count
-x Compression type - "lzo", "favor_lzo", "zlib" or "none" (default: "lzo")
-o Output to FILENAME
-h Help, provides full list of options

Bad blocks

NAND flashes have a random amount of initial bad eraseblocks. This means that different devices may have slightly different volume sizes (especially if the UBI auto-resize feature is used).

Working with a UBI in Linux OS (extracting or manipulating contents)

Linux has support for ubi:

mounting a ubi
creating/renaming/removing ubi volumes
reading/writing ubi volumes

Working with UBI and UBIFS volumes on Linux require use of the following:

tools (from mtd-utils package):
- ubiattach
- ubidetach
- ubiformat
- ubirename
- ubiupdatevol
- ubiinfo
- ubimkvol
- ubirmvol
kernel support:
- mtd (the MTD layer)
- mtdblock (the block layer over MTD)
- ubi (ubi support)
- ubifs (ubi filesystem support)
- nandsim (if you want to simulate a NAND device on a development host)

When using the above tools the MTD partition must not be mounted by your Linux OS meaning you either have to:

Note that the creation of ubifs and ubi images is covered in the section above. This section is all about extracting and / or manipulating the contents of a ubi volume.

References:

On a (NAND-less) Linux Development Host

Often you want to work with UBI and UBIFS images on a Linux Development Host which has no NAND itself. While it is not currently possible to mount a UBI or UBIFS as a loopback device (commonly used for other types of block devices) you can use the nandsim kernel module to simulate a NAND device on a NAND-less Linux development host.

To create a simulated NAND device you need to know the first four bytes returned by the READ_ID flash command, which identifies the characteristics of the device. These values can be found in the datasheet of the FLASH devices. As Gateworks boards support two flash sizes, we will show both examples.

To create a virtual NAND device on a NAND-less Linux development system:

load nandsim module with parameters for specific flash device simulating:
- 256MB FLASH (what we refer to as the 'normal' flash device):
```
sudo rmmod nandsim
sudo modprobe nandsim
  first_id_byte=0x2c \
  second_id_byte=0xda \
  third_id_byte=0x90 \
  fourth_id_byte=0x95
```
- 1GB or 2GB FLASH (what we refer to as the 'large' flash device):
```
sudo rmmod nandsim
sudo modprobe nandsim \
  first_id_byte=0x2c \
  second_id_byte=0xd5 \
  third_id_byte=0xd1 \
  fourth_id_byte=0xa6
```
- the above create a /dev/mtd0 device on your development host
- by default nandsim will use RAM backed storage which makes for much faster read/write access, however if you are cramped for RAM space and simulating a large device (ie 2GB) you can add a cache_file=/tmp/nandimage param to nandsim to tell it to use a file-backed storage instead. You can remove the file once you rmmod nandsim.
populate NAND with an existing ubi:
```
sudo modprobe mtdblock
sudo dd if=image.ubi of=/dev/mtdblock0 bs=2048
```
- the mtdblock module creates a /dev/mtdblock0 device on your development host
- Note that instead of the dd command above you could also have used ubiformat /dev/mtd0 to wipe and format a new ubi but this section is documenting how to extract and alter contents of an existing ubi.

attach the UBI

sudo modprobe ubi
sudo ubiattach /dev/ubi_ctrl -m0 -O2048
sudo ubiinfo -a # optionally show info about the UBI

mount the ubifs to host

sudo modprobe ubifs
sudo mkdir /mnt/ubi # create a mountpoint
sudo mount -t ubifs ubi0 /mnt/ubi
ls /mnt/ubi # optionally list contents of mounted filesystem

Once you are done working with the files in /mnt/ubi simply unmount the device:

sudo umount /mnt/ubi

If you have altered the filesystem and wish to extract the altered UBI you can detach and use dd to copy it to a file:

sudo ubidetach /dev/ubi_ctrl -m0
sudo dd if=/dev/mtdblock0 of=image-updated.ubi bs=2048

And when you are done working with the simulated NAND device completely, remove the nandsim module to free up your system memory:

sudo rmmod nandsim

References:

http://www.linux-mtd.infradead.org/faq/nand.html#L_nand_nandsim

On a target board with NAND FLASH

To extract or manipulate the contents of a UBI within the NAND Flash of a Linux system, Linux must be booted from a secondary medium without the NAND FLASH partition being mounted.

The following examples assume that UBI has been configured to use the 3rd (zero based) MTD FLASH partition (ubi.mtd=2 kernel command line) as we typically use the mtdparts=nand:16m(uboot),1m(env),-(rootfs) env variable in U-Boot to dynamically modify the Linux Device-Tree to partition the NAND flash device into 3 partitions: bootloader, bootloader-env, and rootfs. Thus the bootloader is /dev/mtd0, the bootloader env is /dev/mtd1, and the rootfs is /dev/mtd2.

Examples:

Mounting UBI NAND FLASH volume (from /dev/mtd2):

ubidetach /dev/ubi_ctrl -m 2 # Make sure /dev/mtd2 is not currently attached
ubiattach /dev/ubi_ctrl -m 2 # attach /dev/mtd2 as /dev/ubi0
mount -t ubifs ubi0:rootfs /mnt/ubi #mount 'rootfs' volume from /dev/ubi0 to /mnt/ubi

Updating NAND FLASH using linux target with NAND(/dev/mtd2) with a UBI image:

ubidetach /dev/ubi_ctrl -m 2 # Make sure /dev/mtd2 is not currently attached
sync
ubiformat /dev/mtd2 -y -f image.ubi

Updating NAND FLASH partition (/dev/mtd2) with a UBIFS image

ubidetach /dev/ubi_ctrl -m 2 # Make sure /dev/mtd2 is not currently attached
ubiformat /dev/mtd2 -y # format (will wipe contents!)
ubiattach /dev/ubi_ctrl -m 2 # attach /dev/mtd2 as /dev/ubi0
ubimkvol /dev/ubi0 -N rootfs -m # create a volume on /dev/ubi0 named 'rootfs'
ubiupdatevol /dev/ubi0_0 image.ubifs

Updating NAND FLASH partition (/dev/mtd2) with a tarball

ubidetach /dev/ubi_ctrl -m 2 # Make sure /dev/mtd2 is not currently attached
ubiformat /dev/mtd2 -y # format (will wipe contents!)
ubiattach /dev/ubi_ctrl -m 2 # attach /dev/mtd2 as /dev/ubi0
ubimkvol /dev/ubi0 -N rootfs -m # create a volume on /dev/ubi0 named 'rootfs'
mkdir /mnt/ubi # create a mount-point
mount -t ubifs ubi0:rootfs /mnt/ubi # mount 'rootfs' volume from /dev/ubi0 to /mnt/ubi
tar xvf rootfs.tar.gz -C /mnt/ubi # extract tarball
umount /mnt/ubi # unmount filesystem

Troubleshooting:

The output beneath indicates that /dev/mtd2 is not attached and you may proceed with manipulating the NAND flash.

root@OpenWrt:/# ubidetach /dev/ubi_ctrl -m 2
ubidetach: error!: cannot detach mtd2
           error 16 (Resource busy)

Working with UBI in U-boot Bootloader

The U-boot bootloader used by Ventana has support for ubi:

mounting a ubi
removing ubi volumes
creating ubi volumes
writing ubi volumes

Updating NAND FLASH with a UBI image

If your ubi image is on a tftp server as image.ubi you can use the following to update the 'rootfs' NAND FLASH partition in U-Boot (update the ip addresses as appropriate for your network environment):

setenv ipaddr 192.168.1.1
setenv serverip 192.168.1.254
tftp ${loadaddr} image.ubi && \
  nand erase.part rootfs && \
  nand write ${loadaddr} rootfs ${filesize}

Note the nand_update script will perform the last step for you using the image name specified in the image_rootfs env variable such that you could do a setenv image_rootfs image.ubi; run nand_update

If your ubi image is on the first partition of a FAT formated MMC called image.ubi you can do the following:

mmc rescan
fatload mmc 0:1 ${loadaddr} image.ubi && \
  nand erase.part rootfs && \
  nand write ${loadaddr} rootfs ${filesize}

If your ubi image is on the first partition of a ext2/3/4 formated MMC called image.ubi you can do the following:

mmc rescan
ext2load mmc 0:1 ${loadaddr} image.ubi && \
  nand erase.part rootfs && \
  nand write ${loadaddr} rootfs ${filesize}

If your ubi image is on the first partition of a FAT formatted USB stick called image.ubi you can do the following:

usb start
usb dev 0
fatload usb 0:1 ${loadaddr} image.ubi && \
  nand erase.part rootfs && \
  nand write ${loadaddr} rootfs ${filesize}

Note that operating on the ubi device as described here destroys the wear-leveling information contained in the UBI therefore it is recommended to work at the ubi volume layer to replace ubifs volumes as described below

Updating UBI with low RAM devices:

For some devices, the size of the NAND flash exceeds the size of RAM. In this scenario the instructions from the prior section would fail as you would run out of RAM space before you could finish retrieving the file to program the NAND flash with.

To work around this limitation Gateworks has created an installable U-Boot script that will apply UBI files containing a single partition image. This can be particularly useful for updating the firmware of a running target.

To update using a split UBI:

1. Split your UBI
In most situations the RAM of your device will be smaller than the image file required to update your system, therefore it is necessary to split the image file and apply the resulting compressed parts in a piece wise update. The below example uses the split command to split the input file into parts that are reasonably smaller than the RAM size of the board which will be consuming the update.
FILE=xenial-large.ubi
# Split file every 200M with suffix ".part" and decimal increment
split -d -b 200M ${FILE} ${FILE}.part

# Result will be multiple files such as xenial-large.ubi.partXX that are <= 200M and
# which adhere to the naming convention required by the script (ends with .partXX in sequential order)
2. Install the U-Boot script
On the target machine, source the nand_split_update.scr U-Boot script attached to this page (or copy its contents with an editor). For example via tftp:
setenv serverip 192.168.1.100
tftp nand_split_update.scr
source $loadaddr
3. Configure your U-Boot environment
From the script's usage:
The following environment variables need to be set in order for this script to run:

    splitfile    - full path file name prefix of the splitted files (eg file.ubi for file.ubi.part00)
An example configuration:
setenv splitfile xenial-large.ubi          # ubi file from example above

4. Run the added script with run nand_split_update

Updating NAND FLASH with a ubifs image

To update the contents of a ubi with a ubifs in U-Boot:

ubi part rootfs # select the 'rootfs' partition (from mtdparts env var) for ubi
ubi info layout # list volumes
ubi remove rootfs # remove rootfs volume
ubi create rootfs a00000 static # create a 10MB volume named rootfs
tftp ${loadaddr} ventana/root.ubifs && ubi write ${loadaddr} rootfs ${filesize}
ubifsmount ubi0:rootfs # mount rootfs volume from ubi
ubifsls /
ubifsumount

Operating on a NAND device at the ubi volume layer like this preserves the wear-leveling information for your flash device and is therefore recommended, especially if you are doing this often.

Use the information in the section above for examples on how to load data in uboot from various sources such as network, mmc, usb.

Attachments (1)

nand_split_update.scr (1.6 KB ) - added by Bobby Jones 6 years ago. sourceable uboot script to add the nand_split_update command

Download all attachments as: .zip

Note: See TracWiki for help on using the wiki.

Download in other formats:

Plain Text