* ipconfig cmd writes a file in /tmp that is ready for direct evaluation.
We can use this instead of parsing the output, to do this some variable
renames are needed.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Remove IDE stuff.
* Remove modules that can be added via relevant hooks.
* Replace sata with virtio in baseline profile.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Support img_label=
* Allow use img_dev= as cow_device=
* Use new function _mnt_dev() from archiso.
* Check for error while setting loopback device.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
This allow to take control again of these mountpoints outside initramfs.
(i.e: on deinitramfs stage at shutdown for unmount it)
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Add some options to control where all COW files will be located.
Until this moment all files are located in a tmpfs filesystem.
Now is posible to set a device via a filesystem label or device node
plus a directory, where all these files will be stored.
All dm-snapshot devices will be persistent by default,
but this can be changed if wanted.
Take care, a filesystem that does not support sparse files maybe
is not the best choice for COW files, because they are created with
the same size (is apparent) like the read-only device (the image.fs inside .sfs).
Of course sooner or later, depending on use, these files actually end up
being as big as the read-only device.
KNOW-ISSUE: On shutdown in step "Unmounting Filesystems" will [FAIL],
all filesystem except $cow_device, will be unmounted/remounted-ro.
For this reason this change needs an archiso_shutdown hook, thats
unmount all block devices used by archiso. (Work in progress)
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Separate this code from main mount hook, and make it more generic.
_mnt_dev(device, mountpoint, flags) ->
wait for device and mount, launch a shell if something goes wrong.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* These files was added during developement stage of dm-snapshot branch
of archiso. Never was used, sinse archiso works directly with dmsetup
and with device nodes in /dev/mapper/
* Added 11-dm-initramfs.rules to keep dm info about nodes persistent
in udev db across initramfs/real-root.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
This makes archiso_pxe_nbd functionallity full-independient from
archiso profiles (configs/*).
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Preparing terrain for UEFI support.
* Also make isohybrid hack in one step.
* Removed UDF layer, since xorriso does not support it.
* Removed unsupported options by xorriso (-uid/-gid/-allow-limited-size)
* Removed option already default in xorriso (-input-charset utf-8)
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Default (if not specified) -w work -o out.
-o <out_dir> is only used by 'iso' command.
Adjust build.sh of releng and baseline profile to reflect this change.
<project>
- work <- $work_dir
- out <- $out_dir
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Also put _is_directory_changed() inside 'if' avoiding premature execution if it returns non-zero.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Default value is "arch". This implies that all device mapper node
for snapshot devices will be for example: arch_root-image.
This helps AIF to hidden these devices from "Device Configuration".
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Use device mapper + snapshot module, instead union layer filesystem.
* A block-level approach vs vfs-level.
* No more unofficial (Linux) things.
* More memory is needed.
* Refactor mkarchiso.
* Refactor hooks/archiso.
* Fix install/archiso_pxe_nbd
(due recent change in mkinitcpio-0.6.15 on checked_modules()/all_modules())
[Thanks Dave for the improved workaround]
* New configs/releng to build official images.
* Works with a Bash script instead of Makefile.
(better control and easy to maintain)
* Remove configs/syslinux-iso.
* Remove archiso2dual script. Integrate functionality in configs/releng.
* New configs/baseline to build the most basic live medium or use as template.
* New README (draft). [Thanks Dieter for fixing english grammar]
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Change from none -> union (looks like util-linux is confused BUG?)
Avoid the message: "umount aufs failed: not found" at shutdown.
#1 mount -t tmpfs none /something
#2 mount -t tmpfs xxxx /something
The output of findmnt -rnu -o SOURCE will be blank in #1.
* Change from aufs -> auto, make its more generic.
* Update devpts and shm entries.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Add two parameters:
img_dev=/dev/sdNM
where /dev/sdNM is the device where .iso is located.
img_loop=/path/to/arch.iso
where /path/to/arch.iso is the full path of the .iso in the device img_dev.
Original idea from Baurzhan Muftakhidinov.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
-D <install_dir> allow to select install directory on target media
defaults to "arch".
There is a new %INSTALL_DIR% macro for syslinux.cfg.
This is how ISO will look now:
/syslinux only syslinux related files (syslinux.cfg, *.c32, etc)
/arch/ isomounts + *.sqfs images
/arch/boot/ Linux + initramfs + Memtest86+ + other files for early boot stage
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Add a note about supported status in Linux versions.
Needs squashfs-tools-4.1 (now on extra).
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Newer nbd-server uses config file instead of parameters via cmd line (deprecated).
Also does not use a fixed port (oldstyle).
Sample nbd-server.conf that works with this change:
-- CUT HERE --
[generic]
[archiso]
readonly = true
exportname = /tmp/archlinux.iso
-- CUT HERE --
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
The name of the bootloader is syslinux, while isolinux is just one
of many components. isolinux.bin now also accepts syslinux.cfg
as a configuration file name (as do all other loaders).
Thus, rename the isolinux/ folder to syslinux/, and rename
isolinux.cfg to syslinux.cfg. The only occurrence of 'isolinux'
is now the actual loader file 'isolinux.bin'. This makes
the transition from isolinux to the other syslinux loaders
easier when remastering the ISO onto another medium.
* squashfs images that will be part of read-only branchs for union mount in /
are mounted on /ro_branch/{image_name} instead of /tmpfs/mnt/{image_name}
(and avoid empty dirs on live env under /mnt)
* tmpfs that will be part of read-write branch for union mount in /
is mounted on /rw_branch instead of /tmpfs
* tmpfs that is for store *.sqfs images when copytoram=y, is mounted on
/copytoram instead of /tmpfs.sqfs
* tmpfs_size= parameter renamed to rw_branch_size=
* tmpfs_sqfs_size= parameter renamed to copytoram_size=
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
More generic way than current method, and fix if logic (inverted).
Takes the same actions independent of medium (disk/cdrom).
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Currently works partially since in bash "*" is not expanded.
Old dirs removed, always empty dirs removed from list.
Also cleanup sync databases from pacman 3.4.0, avoiding problems with
pacman -Sy since "sync" directory is deleted.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
A quick & dirty changes to test iso withs differents combinations of
I={virtio,scsi,ide} x M={disk,cdrom}
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Latest init script mounts a /dev FS, so /dev/loop0 will disappear.
Create loop0 on /lib/udev/devices so udev copy to /dev on startup.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
When remastering the image onto a USB volume, you might want to reuse
an existing partition without filling the root with lots of archiso
files. The archisobasedir parameter allows to specify a new root
directory, isomounts and all images will be searched in this directory.
The archiso_pxe_nbd hook adds all network modules, including ones that
are not supported: The hook only supports ethernet devices, so there
is no need to add wireless, wimax, ppp, irda and plip modules.
Adding the wireless modules had an undesired side effect: The
wireless driver was loaded in the initramfs stage, where the
required crypto modules were unavailable. This caused the
initialization of the wireless devices to fail.
This patch removes all network modules except ethernet.
Otherwise archiso_pxe_nbd hook can configure the incorrect network card,
because ipconfig will configure all interfaces available.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Register a mount_hook if booted via PXE. This allows to do all needed
steps before and after calling archiso_mount_handler.
* umount and disconnect network device if copytoram=y
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
nbd (network block device utilities) package is required. This package is in AUR
at this moment: http://aur.archlinux.org/packages.php?ID=9691
Also mkinitcpio-nfs-utils is required for ipconfig.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Takes the advantage of already symlinks created on /dev/disk/by-label/ by
60-persistent-storage.rules add by udev hook.
Tested on:
* Real machine as CD-ROM and USB key drive.
* KVM as CD-ROM and DISK (for both modes, in combination with IDE and SCSI)
* VirtualBox as IDE CD-ROM.
IMPORTANT note: Since isohybrid is a hack, if more partitions are added to
USB key drive, filesystem on these partitions MUST have LABEL, otherwise
USB key will fail to boot. Anyway there is an option archisodevice=,
to force and specific device where live-media reside.
(Maybe isohybrid will change the hack method used in a future)
More info at:
http://mailman.archlinux.org/pipermail/arch-releng/2010-March/000890.htmlhttp://mailman.archlinux.org/pipermail/arch-releng/2010-March/000913.html
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
New function _mnt_aufs(). Don't display incorrect message (Adding new aufs...)
when no aufs branch is mounted, because mount type is bind.
Also, makes the code more clean :)
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Fixes FS#17182: copytoram=y does not work with aufs
The solution is simple, just use a directory outside aufs tree.
* If copytoram=y, then another tmpfs is mounted but on /tmpfs.sqfs
* Add a new option: tmpfs_sqfs_size= (default to 75%)
* Lock cdrom drive is not locked if copytoram=y
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
To make compatible with disablehooks= param. Since "hook_archiso-early"
is not a valid variable name.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Make first /dev/loop0 device at archiso install hook.
* Remove unneeded losetup, all is done directly via mount.
* Images are now mounted on /tmpfs/mnt/image-name instead of
/tmpfs/mnt/loopN, removing unneeded counter.
* Add some variables to make code more readable.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Next version of mkinitcpio > 0.6.2 implements mount hook functionality.
So can avoid skipping normal flow of main init script, and removing common
end code with it.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Split from commit 00dda7d4 from Svenstaro repo: changed -offset for mkarchiso
from default 0 to 1 so that the other partition on the usb device can
still be edited after dd'ing to it
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Add archisodevide= boot option (suggested by brain0)
(if declared use it, if not use /dev/archiso created via udev rule)
* Removed redundant modprobe commands,
(modules are loaded automatically on mount -t)
* Other minor changes
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Some basic changes to make archiso hooks work with the next mkinitcpio 0.6
Tested under KVM, a ""2010.02"" and works fine with it.
NOTE: current mkinitcpio-0.5.99.2-2 does not include "losetup", maybe
will be included in next version, see #[1]. For now must be added manually.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Also use ARCH_YYYYMM format instead of random chars, removing pwgen dep.
Closing FS#14919.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Signed-off-by: Aaron Griffin <aaronmgriffin@gmail.com>
Only support syslinux now. Grub and other things can still
be managed manually by running mkisofs manually.
Signed-off-by: Aaron Griffin <aaronmgriffin@gmail.com>
As in title, this patch cleans up opt flags that exist for no reason and
adds additional flags to modify the disk meta info which would otherwise
be hardcoded.
I wasn't quite sure about the user directory part but it doesn't seem
like a good idea to put that part into the script and therefore I
commented it out. It would probably interfere with rc.local scripts that
take care of skel copying. I replaced the part after the TODO with a
safe guard to make sure that people who are not aware of the changes
will not be caught by surprise.
Signed-off-by: Aaron Griffin <aaronmgriffin@gmail.com>
Rename ramdisk_size parameter to tmpfs_size to avoid conflict with
the kernel parameter that is for old ramdisk based blockdevs (/dev/ramN)
and confusion to users.
The kernel parameter ramdisk_size=N is in "N * k size", while tmpfs
size=N is in "N * [ k | m | g ] byte size" or "in N% of RAM",
like is used by default on archiso.
This parameter is used only for tmpfs in size option.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Check if there are any modification of files/directories inside
the target directory for squashfs image.
Implements in some way this item in TODO:
* Add 'needsupdate' function to check if a squashfs image in the
iso dir is up to date, if so, skip it; else rebuild it. No more
need for the -f flag
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Implement this feature request:
FS#16835 - isomounts file whithin archiso hook is inflexible
Add an optional "isomounts" kernel command line parameter.
In this way, can have differents combinations of
kernels/ramdisk/filesystems.sqfs on the same media.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
The current implementation in how partition is created for ext2 img
it looks a bit bad.
This patch makes the partition in more standarized way, respecting
cylinder alignement:
* The size of resulting image will be in cylinder multiple ~8MB.
* Use fdisk instead of sfdisk (sfdisk write some bad information)
* Make the result image in one pass, instead of concatenating.
Also the advantage is that with this can add another partitions
without any issues in the usb-flash-drive with this .img.
For example of current situation:
qemu-system-x86_64 -hda archlinux-avr.toolchain.img -serial stdio
---------------------------------------------------------------------
[root@avr ~]# fdisk /dev/sda
Command (m for help): p
Disk /dev/sda: 223 MB, 223974400 bytes
59 heads, 41 sectors/track, 180 cylinders
Units = cylinders of 2419 * 512 = 1238528 bytes
Disk identifier: 0x00000000
Device Boot Start End Blocks Id System
/dev/sda1 * 1 181 218693+ 83 Linux
Partition 1 has different physical/logical beginnings (non-Linux?):
phys=(0, 1, 1) logical=(0, 1, 23)
Partition 1 has different physical/logical endings:
phys=(27, 58, 41) logical=(180, 49, 21)
Command (m for help): v
Partition 1 has different physical/logical beginnings (non-Linux?):
phys=(0, 1, 1) logical=(0, 1, 23)
Partition 1 has different physical/logical endings:
phys=(27, 58, 41) logical=(180, 49, 21)
Partition 1: previous sectors 437449 disagrees with total 67731
62 unallocated 512-byte sectors
Command (m for help):
---------------------------------------------------------------------
qemu-system-x86_64 -hda archlinux-avr.toolchain-fix.img -serial stdio
---------------------------------------------------------------------
[root@avr ~]# fdisk /dev/sda
Command (m for help): p
Disk /dev/sda: 230 MB, 230307840 bytes
255 heads, 63 sectors/track, 28 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes
Disk identifier: 0x5c94ca4f
Device Boot Start End Blocks Id System
/dev/sda1 * 1 28 224878+ 83 Linux
Command (m for help): v
62 unallocated 512-byte sectors
---------------------------------------------------------------------
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
fstype from klibc don't detect a vfat formatted blockdevice when try to
mount /dev/archiso to /bootmnt. Some users don't use the recommened
method with dd to bring the image to their USB stick. If they for ex.
use Unetbootin to put the iso image on their (v)fat formatted stick they
got lost with a kernel panic cause fstype won't detect vfat, so the
mount and the later squashfs bindings traps.
Signed-off-by: Gerhard Brauer <gerbra@archlinux.de>
Missing slash caused bootmnt to be bound to /real_rootbootmnt
instead of /real_root/bootmnt
See FS#15356
Signed-off-by: Aaron Griffin <aaronmgriffin@gmail.com>
This will generate a label ARCHISO_$randomstring using the pwgen
utility and add this label to the kernel command line.
A new label will be generated for each image, so it can be uniquely
identified.
Patch from brain0 <thomas@archlinux.org>
Signed-off-by: Gerhard Brauer <gerbra@archlinux.de>
Signed-off-by: Aaron Griffin <aaronmgriffin@gmail.com>
