Difference between revisions of "Talk:USB flash installation media"

From ArchWiki
Jump to: navigation, search
(Universal USB Installer: Remove closed discussion.)
(33 intermediate revisions by 10 users not shown)
Line 1: Line 1:
== unetbootin ==
 
I was unable to get my usb drive bootable with the steps provided. I ended up using UNetbootin, a small application which made my USB drive bootable without a charm. The application doesn't require any installation procedure.
 
 
--[[User:Serpent|Serpent]] 04:22, 25 March 2009 (EDT)
 
 
== General ==
 
I was not able to boot my Thinkpad X31 from USB stick without the lilo part first.
 
I understood it like that: syslinux puts a bootloader at the beginning of the first partition, but nothing in the MBR, so when you try booting from the stick, the bootloader cannot be found.
 
-anonymous
 
 
Shouldn't the info that you need to select boot from usbdisk in BIOS be selected for this to work be mentioned? Or isn't it needed to be able to boot from an usb disk? If it's needed perhaps one could get grub  and/or lilo to bootup an usb disk if the BIOS didn't support it and so how one would do that would also be needed as information.
 
-nut543
 
 
What about merging this article with [[Usb Drive Arch Install]]?
 
-Thujone
 
 
== Does not work anymore... ==
 
 
I was unable to run the new "live" arch-core-install-2008.04-rc-i686.iso off the USB stick.  I've copied the kernel, initrd, and .squashfs files to the stick, and added a minimal entry in syslinux.cfg.  Kernel + initrd boot fine, I even saw that the USB disk is detected, partitions parsed, and /dev/ entries created, but then init halts with something like 'unable to find /dev/cd/*'.  I've spent few minutes reading the initialisation code, but didn't find a kernel commandline option to override it.  I'm sure there is some easy way to make it work, but the process of mounting the CD-ROM should probably be implemented in a more robust way, so the initrd code finds and mounts the compressed filesystem even when booting from USB HDD.
 
  
 
== Verifying the USB ==
 
== Verifying the USB ==
Line 40: Line 21:
 
--[[User:Liquen|Liquen]] 14:55, 4 April 2009 (EDT)
 
--[[User:Liquen|Liquen]] 14:55, 4 April 2009 (EDT)
  
== But I don't want to overwrite the entire USB stick... ==
+
== About making the installation media without overwriting ==
 +
 
 +
I'm not totally sure if I misunderstood something, but I had to change the path of the entries of the *.cfg files. For instance:
 +
 
 +
INCLUDE boot/syslinux/archiso_sys.cfg
 +
 
 +
became:
 +
 +
INCLUDE syslinux/archiso_sys.cfg
 +
 
 +
It was the only way it worked with the unofficial ISO x86_64 image of march 13th, 2012. Looks like the syslinux command described in the page doesn't get the path as it should.
 +
 
 +
I edited all of the .cfg files, but probably only editing this ones should have been enough:
 +
 
 +
archiso.cfg
 +
archiso_head.cfg
 +
archiso_sys_inc.cfg
 +
 
 +
I hope it could be useful to somebody, because I spend some time with this (I even thought that was a problem with the hardware). I think it could be possible to make a simple script (or give some command lines) to patch the files once they are copied into the USB and run syslinux.
 +
 
 +
Thanks !!
 +
 
 +
== Recovering the USB drive afterwards ==
 +
 
 +
This didn't work for me:
 +
 
 +
# dd count=1 bs=512 if=/dev/zero of=/dev/sdx
 +
 
 +
I tried this multiple times. No matter how I formatted the disk, 'devmon' always mounted '/dev/sdd' as /media/ARCH_whatever.
 +
 
 +
I finally just zeroed as much of the disk as I thought the ISO might have been written to.
  
''I'll add this into the article soonish; recording here for reference.''
+
# dd count=100 bs=4M if=/dev/zero of=/dev/sdx; sync
  
So you don't have to. The easier way is to download the ISO image, mount, then copy and install GRUB manually (a la "old method").
+
That worked. I believe we need to zero MORE than just the initial 512 bytes, but I have no idea how much. Maybe 2048?
  
However, I already downloaded the IMG file and don't want to waste time. Then:
+
At any rate, put '; sync' in there somewhere.
  
sfdisk -l -uS /path/to/img
+
This is what I did eventually, taking a tip from:
 +
[http://www.patriotmemory.com/forums/showthread.php?3696-HOWTO-Increase-write-speed-by-aligning-FAT32]
  
outputs:
+
# dd count=100 bs=4M if=/dev/zero of=/dev/sdx; sync
 +
# fdisk -H 224 -S 56 /dev/sdx
  
?
+
(new partition, primary, 1, 2048, whatever, type of partition, c (fat32 LBA), x, beginning data sector, 256, write)
  
Note the starting sector of the first partition (63, in this example). Then:
+
# mkfs.vfat -F 32 -n volume_label -s 32 -v /dev/sdx1; sync
  
dd if=/path/to/img of=arch.img skip=63
+
See the link for more details on the beginning data sector.
  
Now, I can mount the .img file without issue:
+
* Um, zeroing out the first 512 bytes is fine for MBR-formatted drives. Were you using GPT? Because then you would need to zero out the first 512+512+16k, and the last 16k+512. See [[GPT]]. But assuming you used {{ic|dd}}, we're talking about MBR-formatted because the ISO contains a MBR (hybrid) partition table.--[[User:DSpider|DSpider]] ([[User talk:DSpider|talk]]) 06:25, 8 September 2012 (UTC)
  
mount -o loop arch.img /mnt/usb/
+
== flush file system buffers after dd ==
  
== Remove Unetbootin ==
+
I found that after using dd to write the data to my USB I had to wait for the file system to actually write it to my drive. (as dd completed and returned its stats)
 +
This could cause confusion, should we add 'sync' after the dd command on the artical?
  
Can we remove the Unetbootin paragraph? AFAIK, unetbootin has not been working with arch isos for a long time. [[User:Serpent|Serpent]] reported above to have made an install with it in May 2009, but I suspect he used an archboot image (which still works with unetbootin).
+
== Rationale for block size ==
In my experience, the kernel boots fine but it somehow fails to mount the usb drive and therefore can't load the fs overlay stuff. Often people also report on bbs that they tried Unetbootin without luck. Just tried with the new 2010.05 isos, same problem.
+
If someone else can confirm this, I suggest removing Unetbootin altogether since it only adds confusion.
+
  
Similary, is the third option (Gujin) still working? [[User:Hokasch|Hokasch]] 07:04, 18 May 2010 (EDT)
+
Why specifically ''bs=4M'' in the [[USB_Installation_Media#Overwrite_the_USB_drive|example]]:
 +
[[USB_Installation_Media#Overwrite_the_USB_drive|# dd '''bs=4M''' if=/path/to/archlinux.iso of=/dev/sdx]]
 +
[[User:NoobCp|NoobCp]] ([[User talk:NoobCp|talk]]) 11:08, 20 December 2012 (UTC)
  
: Ahh, understood why it is failing and updated the instructions. Still unsure if it wouldn't be better to remove it. [https://bugs.launchpad.net/unetbootin/+bug/582213 Bug report] [[User:Hokasch|Hokasch]] 07:33, 18 May 2010 (EDT)
+
:Because it speeds up the process, that's why. I guess somebody decided that one warning, two notes and a tip would be too rainbow-like. See [https://wiki.archlinux.org/index.php?title=USB_Installation_Media&oldid=218405 this] older edit (which references [http://sprunge.us/SGIY this] script). I reduces the time it needs almost by half. --[[User:DSpider|DSpider]] ([[User talk:DSpider|talk]]) 12:17, 4 February 2013 (UTC)

Revision as of 09:06, 18 February 2013

Verifying the USB

Before and after having performed the dd onto the USB disk, check that the md5sums are correct. For example:

- $ md5sum archlinux-2008.06-core-x86_64.img && echo && cat md5sums.x86_64

The next command will give similar results, but will also let you confirm that the data was written correctly and can be read correctly:

- dd if=/dev/sdb count=661159 status=noxfer | md5sum && echo && cat md5sums.x86_64

--Zatricky 06:45, 22 January 2009 (EST)

dd for Windows

There is also dd for Windows. I tried it and it works perfectly: [1]

 dd if=file.img of=\\.\e:

where e: is your USB drive letter.

--Liquen 14:55, 4 April 2009 (EDT)

About making the installation media without overwriting

I'm not totally sure if I misunderstood something, but I had to change the path of the entries of the *.cfg files. For instance:

INCLUDE boot/syslinux/archiso_sys.cfg

became:

INCLUDE syslinux/archiso_sys.cfg

It was the only way it worked with the unofficial ISO x86_64 image of march 13th, 2012. Looks like the syslinux command described in the page doesn't get the path as it should.

I edited all of the .cfg files, but probably only editing this ones should have been enough:

archiso.cfg archiso_head.cfg archiso_sys_inc.cfg

I hope it could be useful to somebody, because I spend some time with this (I even thought that was a problem with the hardware). I think it could be possible to make a simple script (or give some command lines) to patch the files once they are copied into the USB and run syslinux.

Thanks !!

Recovering the USB drive afterwards

This didn't work for me:

  1. dd count=1 bs=512 if=/dev/zero of=/dev/sdx

I tried this multiple times. No matter how I formatted the disk, 'devmon' always mounted '/dev/sdd' as /media/ARCH_whatever.

I finally just zeroed as much of the disk as I thought the ISO might have been written to.

  1. dd count=100 bs=4M if=/dev/zero of=/dev/sdx; sync

That worked. I believe we need to zero MORE than just the initial 512 bytes, but I have no idea how much. Maybe 2048?

At any rate, put '; sync' in there somewhere.

This is what I did eventually, taking a tip from: [2]

  1. dd count=100 bs=4M if=/dev/zero of=/dev/sdx; sync
  2. fdisk -H 224 -S 56 /dev/sdx

(new partition, primary, 1, 2048, whatever, type of partition, c (fat32 LBA), x, beginning data sector, 256, write)

  1. mkfs.vfat -F 32 -n volume_label -s 32 -v /dev/sdx1; sync

See the link for more details on the beginning data sector.

  • Um, zeroing out the first 512 bytes is fine for MBR-formatted drives. Were you using GPT? Because then you would need to zero out the first 512+512+16k, and the last 16k+512. See GPT. But assuming you used dd, we're talking about MBR-formatted because the ISO contains a MBR (hybrid) partition table.--DSpider (talk) 06:25, 8 September 2012 (UTC)

flush file system buffers after dd

I found that after using dd to write the data to my USB I had to wait for the file system to actually write it to my drive. (as dd completed and returned its stats) This could cause confusion, should we add 'sync' after the dd command on the artical?

Rationale for block size

Why specifically bs=4M in the example:

# dd bs=4M if=/path/to/archlinux.iso of=/dev/sdx

NoobCp (talk) 11:08, 20 December 2012 (UTC)

Because it speeds up the process, that's why. I guess somebody decided that one warning, two notes and a tip would be too rainbow-like. See this older edit (which references this script). I reduces the time it needs almost by half. --DSpider (talk) 12:17, 4 February 2013 (UTC)