Oct 202011
 

So there I was, installing a Linux distribution on my new laptop. Got to the end of the installation when it refused to install grub in the master boot record. Opted to try another partition, and rebooted. At which point the infamous error “Error: the symbol ‘grub_xputs’ not found” was shown with a “grub rescue” prompt.

At which point I had a laptop that wouldn’t boot of course.

To cut a long story short, because it’s only the fix I’m interested in recording for posterity, I sorted this out by booting off an emergency USB stick (unetbootin is a good tool for writing one … if you have a working system). Once booted, I setup an environment where chroot would function well. This is basically where you start a shell whose root directory is a directory under the normal root directory. This allows commands to be run almost as if the non-bootable system was booted.

mount /dev/sda5 /mnt # Mount the root filesystem of the unbootable system under /mnt
mount /dev/sda1 /mnt/boot # And the /boot filesystem
mount -o bind /proc /mnt/proc
mount -o bind /dev /mnt/dev
mount -o bind /sys /mnt/sys
chroot /mnt

Once that is done, there are quite a few things that can be done to repair a broken system, but I just needed to re-install grub to the MBR of /dev/sda :-

grub-install /dev/sda

Once that was done, everything booted fine.

Of course all that comes with the experience of a lot of time with Linux. Those who have not used it since the 1990s will not be as lucky, but there’s a few key points there :-

  1. Don’t panic. Just because it won’t boot doesn’t mean everything is lost.
  2. Write down the error message exactly as it appears on screen. A small mistake here can make searching for the error almost impossible.
  3. Get a rescue USB stick. Ideally before you break a system, but afterwards is usually possible even if you don’t have another working system – you have friends, or there are ways to write a USB stick at work.
  4. Search the Internet for the problem. You may have to spend quite a while reading other people’s problems that may or may not relate to your problem. You may have to improve your search methodology. Putting the error message in quotes is usually a good method.
  5. And if you find a solution to your problem online, check the date of the solution. Something that worked 5 years ago may not be the best solution today. And that applies to this page just as much as any other.
Oh! And to those who would jump and down screaming about this wouldn’t happen with Windows or OSX, please grow up. Such problems occur with any operating system – and I’ve seen them.
May 052011
 

For my own future reference …

Today I encountered an interesting little issue where I could not send an ABORT signal to a running process to kill it with a core dump because the process had a limit of 0 for the core dump size. Try as I might, I could not find a way to change that process’s core dump limit.

Turns out there is another way of tackling the problem, which is to use gdb to generate a core image :-

gdb
>attach PID
>gcore /var/tmp/core.PID

There is of course the gcore shell script wrapper for this, but that may not work if the working directory of the process no longer exists.

Jan 292011
 

In the dim and distant past when keyboards were enclosed in metal cases and you certainly didn’t tuck one under the arm and walk around with it (actually I don’t do now either), the placement of many keys was continually up for debate. But apart from the main QWERTY section, one of the key placements you could rely on was the Control key next to the “A” key. These days it’s been turned into one of those silly CapsLock keys.

Back when I previously did some keymapping, I neglected to mention how I mapped CapsLock into a Control key. As appropriate punishment, changing window managers has somehow meant that my previous mapping had been lost. So I had to figure out how to do it again.

First thing to do is to switch to a text console – I’ll be mapping this at a very low level.

Next thing to do is to find out the scancode of the key I want to map :-

# showkey -s

One started I have to press the key I am interested in within 10 seconds or the program will edit. I press CapsLock and I get two numbers displayed – 3a and ba (they’re in hexadecimal for the base-16 challenged). The first is the key press, and the second is the key release. We can discard the second as Linux is clever enough to figure out one from the other.

The next thing we want to do is to obtain the keycode of the key that we want to map to – in this case the left control key. It probably doesn’t matter here, but it is worth noting that the left and right control keys are different scancodes and keycodes. So you could map then to different things. Anyway, to obtain the keycode of the key we want, run :-

# showkey -k

And press the key to map to.

Lastly we want to construct the command to actually do the mapping :-

setkeycodes 3a 29

This of course has to be added to a script being run when the system boots – you want this mapped as early as possible.

Aug 112010
 

Ten years ago? Bloomin’ heck! Still as this still seems to get frequent hits, I suppose quick update is in order – I no longer use the Unicomp “naked” but go through a Soarer converter which is an easier way of mapping the keys and allows macros to be added.

Thanks to these guys (the thread eventually gets to the meaty details), I have changed my Unicomp 122-key keyboard (UB40T56) from a funky “be friendly to Windows IBM Terminal emulators” mode into something a little more interesting. Specifically each key should be sending a unique keycode – which if you select the right model from Unicomp, you’ll get.

This can be done by opening up the keyboard and removing the jumper from JP3 (just below a small chip and close to the scroll LED). In addition, it is strongly suggested that you set the kernel parameter “atkbd.softraw=0” which can be done with Ubuntu 10.04 with the following :-

  1. Edit /etc/default/grub and change the variable GRUB_CMDLINE_LINUX_DEFAULT to include “atkbd.softraw=0” at the end of what is already there.
  2. Finish editing and run grub-update and finally reboot.

Without this, showkey -s only shows the scancodes of keys that the kernel knows about – not what scancodes are sent by the keyboard! It is possible to show the unknown scancodes by enabling the i8042 module debug mode, but atkbd.softraw does the same thing more effectively.

This is currently a work in progress, and will need further checking before it can be taken as gospel, but …

Group Key Make Break Keycode
Function Keys
F13 5b db
F14 5c dc 95
F15 5d dd 183
F16 63 e3
F17 64 e4
F18 65 e5
F19 66 e6
F20 67 e7
F21 68 e8
F22 69 e9
F23 6a ea
F24 6b eb
(next row) F1 3b bb 59
F2 3c bc 60
F3 3d bd 61
F4 3e be 62
F5 3f bf 63
F6 40 c0 64
F7 41 c1 65
F8 42 c2 66
F9 43 c3 67
F10 44 c4 68
F11 57 d7 87
F12 58 d8 88
Left Keypad (Top left is “Esc”)
Esc 7e fe 121
Cent 76 f6 85
Print Screen 72 f2
Pause e1 1d 45 e1 9d c5 119
Print 74 f4
Help 6d ed
Record e0 2a e0 37 e0 b7 e0 aa 99
Play 6f ef
GUI (Windows) 75 f5
Menu 6c ec
Editing Pad (between QWERTY and Number Pad)
Backtab 5a da
Insert e0 49 e0 c9 104
PageUp e0 51 e0 d1 109
(next row) Blue Return e0 4f e0 cf 107
Delete e0 52 e0 d2 110
Page Down e0 53 e0 d3 111
(next row) Up Arrow e0 48 e0 c8 103
(next row) Left Arrow e0 4b e0 cb 105
Home e0 47 e0 c7 102
Right Arrow e0 4d e0 cd 106
(next row) Down Arrow e0 50 e0 d0 108
Number Pad
(top row) End 01 81 1
Scroll Lock 46 c6 70
(shifted Scroll Lock) Number Lock 45 c5 69
/ 37 b7 55
* e0 c5 e0 b5 98
(next row) KP-7 47 c7 71
KP-8 48 c8 72
KP-9 49 c9 73
4e ce 78
(next row) KP-4 41 cb 75
KP-5 4c cc 76
KP-6 4d cd 77
+ 4a ca 74
(next row) KP-1 4f cf 79
KP-2 50 d0 80
KP-3 51 d1 81
Enter e0 1c e0 9c 96
(next row) KP-0 52 d2 82
KP-. 53 d3 83
  • Group. To break things up a little, I’ve grouped the keys into the 5 separate parts of the keyboard – the function keys, the keypad to the left, the qwerty pad, the editing pad, and the number pad (“keypad”). The details of the qwerty pad will be the last as the other groups are more interesting (‘qwerty’ keys just work).
  • Key. This is the label on the key on my keyboard. This may be different on different variants so in all cases I have started with the top left, worked left and down (the “qwerty” row before the “asdf” row).
  • Make. This is the scancode produced when the key is pressed.
  • Break. This is the scancode produced when the key is released.
  • Keycode. The configured keycode produced on the Linux console. The red cells are values that are wrong, but in addition many are missing because they are not produced on a keypress. I say wrong because the keycode gives a result that does not match the key legend – in some cases dangerously wrong such as PageDown generating Delete. One thing to be aware of is that you must use “showkey -k” at the console to get the same numbers I have – X seems to add 8 to each keycode.

Three interesting oddities here … F14 and F15 have somehow been graced with keycodes by default; their scan codes must coincide with keys defined on more popular keyboards. And of course Num Lock and Scroll Lock sharing the same key is a little … odd. And lastly the Record key is effectively sending two keystrokes in one.

Fixing The Wrong Keys

The first place to start is to map the keys that return a keycode that represents a key other than that written on the keycap – such as the key marked “End” which thinks it is an “Esc” key. I have left out two of the wrong keys from this group as they fit better into the next section.

The wrong keys can be fixed with the following commands :-

setkeycodes 7e         1    # Esc
setkeycodes e049      82    # Insert
setkeycodes e051     105    # PageUp
setkeycodes e052     111    # Delete
setkeycodes e053     109    # PageDown
setkeycodes 01       107    # End
setkeycodes 37        98    # KP-/
setkeycodes e035      55    # KP-*
setkeycodes 4e        74    # KP--
setkeycodes 4a        78    # KP-+

I have left out a couple of the wrong keys from this section as they do not return dangerously incorrect values, and they fit more logically into the next section (being Record and Blue Return).

Dealing With The Extra Keys

Now onto dealing with the extra keys. The tricky bit here was coming up with new keycodes for these keys that did not conflict with existing keycodes, and were reasonable. This is effectively impossible, as xmodmap -pk appears to show no significant range of unused keycodes although some of the used keycodes are for things like “Shop” buttons!

So I picked a range with a larger number of useless key symbols and some unused ones :-

setkeycodes 5b       222    # F13
setkeycodes 5c       223    # F14
setkeycodes 5d       224    # F15
setkeycodes 63       225    # F16
setkeycodes 64       237    # F17
setkeycodes 65       238    # F18
setkeycodes 66       228    # F19
setkeycodes 67       229    # F20
setkeycodes 68       230    # F21
setkeycodes 69       231    # F22
setkeycodes 6a       232    # F23
setkeycodes 6b       233    # F24
setkeycodes 72        99    # Record (after keyswap)
setkeycodes 74       209    # Print
setkeycodes 6d       138    # Help
setkeycodes 6f       239    # Play
setkeycodes 75       234    # Windows (GUI)
setkeycodes 6c       240    # Menu
setkeycodes 5a       235    # Backtab
setkeycodes e04f     236    # BlueReturn

Once this has run, we can look at fixing the X mappings … which is why F17 and F18 are out of sequence in the above! One key has to be (at least until someone comes up with a better solution!) sorted out with a keycap swap. Take the keycap from the Record key and swap it for the one marked “Print Screen”. This is because the scancode for Record is effectively two scancodes in one and attempting to remap it will result in strange things happening.

Sorting Out X11

Once you have a set of keycodes that don’t do funny things under X (for instance F17 and F18 when in sequence produce not a keystroke under X11 but some other event), you can move onto configuring the X keyboard. The following attempts to map as close to the keycaps as possible without going to extremes :-

xmodmap -e "keycode 230 = F13"
xmodmap -e "keycode 231 = F14"
xmodmap -e "keycode 232 = F15"
xmodmap -e "keycode 233 = F16"
xmodmap -e "keycode 245 = F17"
xmodmap -e "keycode 246 = F18"
xmodmap -e "keycode 236 = F19"
xmodmap -e "keycode 237 = F20"
xmodmap -e "keycode 238 = F21"
xmodmap -e "keycode 239 = F22"
xmodmap -e "keycode 240 = F23"
xmodmap -e "keycode 241 = F24"
xmodmap -e "keycode 217 = Print"
xmodmap -e "keycode 9 = Escape 3270_Attn"
xmodmap -e "keycode  93 = cent bar"
xmodmap -e "keycode 175 = 3270_Record"
xmodmap -e "keycode 175 ="
xmodmap -e "keycode 247 = 3270_Play"
xmodmap -e "keycode 242 = Super_L"
xmodmap -e "keycode 248 = Multi_key"
xmodmap -e "keycode 243 = 3270_BackTab"
xmodmap -e "keycode 118 = Insert 3270_Duplicate"
xmodmap -e "keycode 112 = Prior 3270_Jump"
xmodmap -e "keycode 117 = Next 3270_Rule"

This results in a keyboard that more or less matches the key caps. For some of the blue symbols, you press the key in combination with shift.

The number pad could do with a little more attention in the realm of X-mapping, and there are a few blue symbols on the main qwerty pad that might be usefully mapped, but this is sufficient for my purposes.

Apr 012010
 

… is a good clone of SMIT.

SMIT for those who have not been exposed to AIX, is a system administration tool that allows you to perform tasks through a graphical (or text-based) interface. Just like any other tool really, but the killer feature is that once you have built up a task in SMIT such as extend a logical volume by a certain amount, you can then ask it what the command-line equivalent is.

Now all you point and drool fans out there are probably thinking “So what?”. Well perhaps this feature is not for you, but it does allow those who work at the command line to find out what command is necessary to perform a certain task. Once you know the command to perform a task you can :-

  • Use it to setup a cron job to run the task at a particular time. No need to stay up late to perform a task in a “maintenance window” after midnight.
  • Run that command on a whole rack full of servers using a tool like pssh. Much easier than repeating the same steps on a dozen computers one by one – aren’t these computers supposed to automate tedious jobs for us ?
  • Use that command to put into a script to run on certain events. For instance you could monitor the available space on all your filesystems and grow the filesystem when the available space drops below a certain trigger point. You might also want to automatically order a new hard disk when the “volume group” runs out of space 🙂

Linux is currently undergoing a process of fragmentation where different distributions are operated in different ways much like the old commercial Unix variants went through in the 1990s. A good clone of SMIT would go a long way to allowing different distributions to go their own way in system administration commands, but allow system administrators to use a standard tool to manage any Linux distribution.