Category: Linux / Gentoo Linux

While playing around with various mainline kernels I found out that the vanilla NVidia driver (310.xx and 313.xx) do not compile on the linux 3.8.0-rc3 kernel. Testing the xorg-edgers and ubuntu-x-swat versions of the NVidia driver package did not help either.

Until the NVidia driver officially supports the 3.8.0 kernel, this workaround will allow you to fix the NVidia driver using the DKMS source installation. This might also still apply for kernel 3.7.x. Begin by adding the ubuntu-x-swat repository to get at least version 304.60 of the NVidia driver:

sudo apt-add-repository ppa:ubuntu-x-swat/x-updates
sudo apt-get update
sudo apt-get install nvidia-current

Patch (by hand if you have to) the following 2 files: /usr/src/nvidia-current-304.64/conftest.sh and /usr/src/nvidia-current-304.64/nv-mmap.c. These patches are based on 310.xx but can be used for manual modification:

--- conftest.sh.dist    2012-10-11 19:18:22.704848496 -0400
+++ conftest.sh 2012-10-12 20:35:55.707213868 -0400
@@ -20,6 +20,7 @@
 ISYSTEM=`$CC -print-file-name=include 2> /dev/null`
 SOURCES=$4
 HEADERS=$SOURCES/include
+HEADERSA=$SOURCES/include/uapi
 OUTPUT=$5
 XEN_PRESENT=1
 
@@ -118,7 +119,7 @@
         fi
     fi
 
-    CFLAGS="$CFLAGS $OUTPUT_CFLAGS -I$HEADERS $AUTOCONF_CFLAGS"
+    CFLAGS="$CFLAGS $OUTPUT_CFLAGS -I$HEADERS -I$HEADERSA $AUTOCONF_CFLAGS"
 
     test_xen
 
@@ -146,10 +147,10 @@
         fi
     fi
 
-    CFLAGS="$BASE_CFLAGS $MACH_CFLAGS $OUTPUT_CFLAGS -I$HEADERS $AUTOCONF_CFLAGS"
+    CFLAGS="$BASE_CFLAGS $MACH_CFLAGS $OUTPUT_CFLAGS -I$HEADERS -I$HEADERSA $AUTOCONF_CFLAGS"
 
     if [ "$ARCH" = "i386" -o "$ARCH" = "x86_64" ]; then
-        CFLAGS="$CFLAGS -I$SOURCES/arch/x86/include -I$OUTPUT/arch/x86/include/generated"
+        CFLAGS="$CFLAGS -I$SOURCES/arch/x86/include -I$SOURCES/arch/x86/include/uapi -I$OUTPUT/arch/x86/include/generated -I$OUTPUT/arch/x86/include/generated/uapi"
     elif [ "$ARCH" = "arm" ]; then
         CFLAGS="$CFLAGS -I$SOURCES/arch/arm/include -I$OUTPUT/arch/arm/include/generated"
     fi
--- nv-mmap.c.dist 2012-08-08 22:52:53.000000000 -0400
+++ nv-mmap.c 2012-08-14 23:52:41.257235863 -0400
@@ -450,7 +450,7 @@
NV_PRINT_AT(NV_DBG_MEMINFO, at);
nv_vm_list_page_count(&at->page_table[i], pages);

- vma->vm_flags |= (VM_IO | VM_LOCKED | VM_RESERVED);
+ vma->vm_flags |= (VM_IO | VM_LOCKED | (VM_DONTEXPAND | VM_DONTDUMP));

Patches are taken from here After patching, the NVidia kernel module will not be built unless you install a new kernel or force an update using DKMS. To rebuild the kernel module for your currently running kernel, run this:

sudo dkms autoinstall -k $(uname -r)

I had to perform this emergency procedure every now and then and I seem to reinvent the wheel every time so I reckoned it was time to document it.

This time, its a failing master drive in a simple home server – without a RAID. The SMART monitoring has not yet caught on but the self-tests fail and some files are read-only as writes to the drive fail. Also the logs are full with SATA errors like: Apr 29 10:23:57 rivenscryr smartd[3621]: Device: /dev/sda, 18 Currently unreadable (pending) sectors

To save the installation and all the files, I want to clone or backup the entire root filesystem (which is ext3, but that is not relevant) to a new drive (formatted ext4) in another machine. The reason to do this between machines can be numerous: perhaps you are migrating a live machine or (in this case) perhaps you worry that the current drive may fail when you power-cycle the system when the replacement drive is installed.

Backup/Clone Over Network Recipe ∞

1. Mount the root filesystem of the original machine on another location in order to have a clean copy:

     mkdir /mnt/rootfs && mount / /mnt/rootfs -o bind

   Check if it worked by listing the /mnt/rootfs/ folder. Note that /mnt/rootfs/proc will be empty and dev will be empty or very sparse.

2. Use netcat and tar on the receiving machine (my case: random system with Ubuntu on USB stick and new HDD plugged in) to receive and write the data.

     nc -l 7777 | tar -xz -C /mnt/targetdir

3. Use tar and netcat on the original machine to stream-copy the root filesystem over the network.
   Big fat warning: My cloning is done on an internal, trusted network. You will be cloning using an unsecured stream over this network.

     tar -czp --atime-preserve --numeric-owner \
       --xattrs --recursion /mnt/rootfs | nc remotehost 7777

Here is what all these options do:

• -c: Create a new archive
• -z: Use GZip compression to speed up the transfer
• -p: Preserve file permissions
• –atime-preserve: Do not update the accessed timestamp
• –numeric-owner: Copy using numeric ids only – useful when cloning into a foreign system
• –xattrs: Include ACL attributes and SElinux attributes
• –recursion: Recurse down into sub-folders
• –ignore-failed-read: Useful when a drive is heavily damaged and you only care to extract as much data as possible. Note: do not enable this by default as you might end up with an unusable clone if some files are missing. (Aka, this is a last resort option for non-functional clones)

For some reason, my attempts to use GPT with Windows 7 ran aground and I ended up with a working Master Boot Record (MBR) and an empty GPT. While Windows 7 was quite happy with this, the Ubuntu installer detected the GPT and attempts to use it – which effectively prevents me from installing it unless I wipe the entire hard drive for Ubuntu.

In this little article I will clear the GPT from the disk so only the MBR remains. This is an advanced technique which WILL destroy your partition layout and thus your data when it fails. Make very very sure you understand each step and have backups.

Start by making a backup of your MBR, which is 512 bytes long at the start of the disk:
sudo dd if=/dev/sda of=disk_mbr.raw bs=512 count=1
I made a backup of the MBR and the GPT as well, just in case (GPT is 512 bytes for the header followed by 16kB of content, total with MBR is 17kB):
sudo dd if=/dev/sda of=disk_mbr_gpt.raw bs=1024 count=17
The GPT is present at the start and end of the disk. Lets start by erasing the GPT from the start:
sudo dd if=/dev/zero of=/dev/sda bs=512 seek=1 count=33
Now find the end of your disk, use fdisk /dev/sda and note the number of bytes on the disk. In my case it is: 240,057,409,536 bytes, since I want to use kilobytes with dd, lets convert it to kB: 234,431,064 kB. We need to erase the last 16.5 kB, but 16kB will suffice (as the header is in the last 512 bytes).

So subtracts 16 from the drive size to generate the GPT offset for the end of the disk and zero-fill it to the end: sudo dd if=/dev/zero of=/dev/sda seek=234431048 bs=1024 count=17

All done! Now fdisk will no longer complain the drive has a GPT table and the Ubuntu installer will only find the MBR and happily install using that.

From Wikipedia:

A Universally Unique Identifier (UUID) is an identifier standard used in software construction, standardized by the Open Software Foundation (OSF) as part of the Distributed Computing Environment (DCE).

The intent of UUIDs is to enable distributed systems to uniquely identify information without significant central coordination. Thus, anyone can create a UUID and use it to identify something with reasonable confidence that the identifier will never be unintentionally used by anyone for anything else. Information labeled with UUIDs can therefore be later combined into a single database without needing to resolve name conflicts.

Since CentOS (and Ubuntu and others) use the UUID of partitions to mount or perform other operations on them, I needed to find the UUID of my storage partition. The downside to using UUIDs is obvious: I know the device name (/dev/sdb1) but I have no clue what the UUID of the partition is.

The upside is that if I move the drives around, so my /dev/sdb1 becomes /dev/sda1 for example, everything will still work because the identifiers remained unchanged.

Enough talking, I found 3 ways of finding the UUID of a drive on Linux. Note that this is not all of them though, just enough for me.

Option 1: Use /dev ∞

Since the UUID to device mapping needs to be done somewhere for the kernel and system tools, why not use that ourselves?
ls -l /dev/disk/by-uuid

Option 2: Use udev ∞

The vol_id program pulls the information from udev and presents it all in a fairly readable form. For some reason, my Ubuntu 10.10 does not have it anymore but possibly others still do.
vol_id /dev/sdb1

Option 3: Use blkid ∞

The option I tend to use it to use the program called blkid. Which as sole purpose (as its name suggests) is to show the block device ID of a block device *gasp*.
blkid /dev/sdb1