[Libguestfs] [PATCH nbdkit v3 4/4] Add linuxdisk plugin.

[Eric Blake]

On 2/22/19 3:07 AM, Richard W.M. Jones wrote:
> From: "Richard W.M. Jones" <rjones at redhat.com>
> 
> This plugin allows you to create a complete ext2, ext3 or ext4
> filesystem in a GPT partitioned disk image.  This can be attached as a
> disk to a Linux virtual machine.  It is implemented using e2fsprogs
> mke2fs ‘-d’ option thus allowing the implementation to be very small
> and simple, with all the hard work done by mke2fs.
> 
> Although there is some overlap with nbdkit-iso-plugin and
> nbdkit-floppy-plugin, the implementations and use cases of all three

> +static int64_t
> +estimate_size (void)
> +{

> 
> +  r = pclose (fp);
> +  if (r == -1) {
> +    nbdkit_error ("pclose: %m");
> +    free (line);
> +    return -1;
> +  }
> +  if (WIFEXITED (r) && WEXITSTATUS (r) != 0) {
> +    nbdkit_error ("du command failed with exit code %d", WEXITSTATUS (r));
> +    return -1;
> +  }
> +

Doesn't cover if the process died due to signal.

> +static int
> +mke2fs (const char *filename)
> +{

> +  /* Run the command. */
> +  nbdkit_debug ("%s", command);
> +  r = system (command);
> +  free (command);
> +
> +  if (WIFEXITED (r) && WEXITSTATUS (r) != 0) {
> +    nbdkit_error ("mke2fs command failed with exit code %d", WEXITSTATUS (r));
> +    return -1;
> +  }
> +  else if (WIFSIGNALED (r)) {
> +    nbdkit_error ("mke2fs command was killed by signal %d", WTERMSIG (r));
> +    return -1;
> +  }
> +  else if (WIFSTOPPED (r)) {
> +    nbdkit_error ("mke2fs command was stopped by signal %d", WSTOPSIG (r));
> +    return -1;
> +  }
> +

WIFSTOPPED() branch is unreachable.

> +=item nbdkit --filter=cow linuxdisk /path/to/directory
> +
> +Add a writable overlay (see L<nbdkit-cow-filter(1)>, allowing the disk
> +to be written by the client.  B<Multiple clients must not be allowed
> +to connect at the same time> (even if they all mount it read-only) as
> +this will cause disk corruption.

Not a cause for changing any of the wording, but just to make sure I
understand: even when a filesystem is mounted read-only, the kernel may
still write to the underlying block device (such as fsck or journal
updates), and THESE writes must only be done by a single writer, hence
allowing multiple writers to a cow wrapper around the read-only file
system indeed risks races where one writer corrupts another.


> +=item Create a minimal virtual appliance
> +
> +This creates and boots a minimal L<busybox(1)>-based virtual
> +appliance.  This assumes that your kernel (F</boot/vmlinuz>) contains
> +the ext2 or ext4 driver compiled in, but most Linux distro kernels
> +have that.
> +
> + mkdir root root/bin root/dev root/proc root/sbin root/sys
> + mkdir root/usr root/usr/bin root/usr/sbin

Worth 'mkdir -p' in this example?

> + sudo mknod root/dev/console c 5 1
> + cp /sbin/busybox root/sbin/
> + ln root/sbin/busybox root/sbin/init
> + ln root/sbin/busybox root/bin/ls
> + ln root/sbin/busybox root/bin/sh
> + nbdkit -U - linuxdisk root --run '
> +   qemu-kvm -display none -kernel /boot/vmlinuz -drive file=nbd:unix:$unixsocket,snapshot=on -append "console=ttyS0 root=/dev/sda1 rw" -serial stdio
> + '
> +
> +You can drop any extra files you need into the F<root/> directory and
> +they will be copied into the appliance before boot.  After booting
> +type these commands to complete the environmental setup:
> +
> + /sbin/busybox --install
> + mount -t proc proc /proc
> + mount -t sysfs sys /sys

Pretty cool!

I think you're ready to go, with minor tweaks to child process status
handling.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org