Using the sync operation with list option fails with --sysroot when
signed database files are around. Instead use the query operation, which
uses the local databases of installed pakages only.
The only downside is that we do no longer record the originating
repository.
Signed-off-by: Christian Hesse <mail@eworm.de>
The kernel has an open file handle after setting up the mapping. We can
remove it early to make sure it is gone on shutdown.
This helps to keep the cow_device clean for non-persistent systems where
cow_directory contains a version specific string.
Signed-off-by: Christian Hesse <mail@eworm.de>
If the ARCHISO_GNUPG_FD environment variable is set, its contents will be interpreted as an open file
descriptor and its contents will be used to create a keyring in the initramfs in /gpg.
Having files on btrfs subvolumes requires to give mount options. Add
boot params archisoflags= and cow_flags= for this purpose. Boot
parameters could look like this:
... archisodevice=/dev/sdaX archisoflags=subvolume=isos
cow_device=/dev/sdaX cow_flags=subvolume=persist ...
Signed-off-by: Christian Hesse <mail@eworm.de>
Currently, when booting loopmounted iso file, it is necessary to
specify not only img_dev and img_loop (which should be sufficient),
but also archisolabel or archisodevice. With this patch, archisodevice
is directly populated with the correct loop device, and it is not
necessary to specify the label when booting from loopmounted iso,
which makes for leaner and cleaner grub.cfg.
We received an IP address from DHCP server and configure it statically.
This is required if we continue to use network connectivity to access
the root device (for example via NBD or NFS).
The lease is not updated, though. This can cause trouble in networks
with low lease times. So let's flush the addresses if root filesystem
has been copied to RAM. A dhcp client in main system can handle the
network connectivity then.
Signed-off-by: Christian Hesse <mail@eworm.de>
Umount detaches the loop device automatically, but let's make it
explicit to be sure. Additionally losetup gives:
losetup: /dev/loop0: detach failed: No such device or address
This is kind of expected, let's silent the error message.
Signed-off-by: Christian Hesse <mail@eworm.de>
I see cases where a stale loop device stays around and fills up my
partition as image file is still in use and does not get unlinked.
Explicitly detach loop device on umount to fix that.
Signed-off-by: Christian Hesse <mail@eworm.de>
This is the first attemp to test overlayfs in archiso.
The current dm-snapshot mode is keep and is enabled by default,
while the new mode is enabled via "-s sfs" to mkarchiso.
No new boot parameters are added, since archiso hooks detects
if the .sfs file is for dm-snapshot (airootfs.img inside)
or for overlayfs.
Persistence is supported in overlayfs mode using the same options
(cowlabel or cowdevice), but warning while in dm-snapshot mode,
only one file is used (airootfs.cow), in overlayfs mode internal
files for workdir/ and upperdir/ are allocated, so you can not use
VFAT or NTFS.
To test this, you need to enable [testing] in pacman.conf from
releng profile and edit build.sh then add "-s sfs" in make_prepare()
Look at:
setarch ${arch} mkarchiso ${verbose} -w "${work_dir}" -D "${install_dir}" prepare
Replace with:
setarch ${arch} mkarchiso ${verbose} -w "${work_dir}" -s sfs -D "${install_dir}" prepare
The build requires just half of space that the build for dm-snapshot,
since there is no ext4 img ;)
Just to remember: there is no space gain in .sfs (just about 2M)
There is at least one thing during boot with machine-id service:
Dec 24 03:31:39 archiso systemd-machine-id-commit[183]: Failed to unmount transient /etc/machine-id file in our private namespace: Invalid argument
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@gmail.com>
Copying big amount of data results in bad performance as data is
written in chunks of 4kiB (8 * 512 bytes).
The default is not changed but can be overwritten.
Signed-off-by: Christian Hesse <mail@eworm.de>
Remove unused feature, just to save space
sizeof ext4 image (empty): 4.3M vs 403K (du airootfs.img)
sizeof used space (empty): 48M vs 20K (df)
Thanks Christian Hesse for initial tip.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@gmail.com>
Currently, when booting via HTTP, the airootfs.md5 file is downloaded to
/run/archiso/bootmnt/arch/. However, the checksum file is later assumed
to be located at "/run/archiso/bootmnt/arch/${arch}". Fix the _curl_get
invocation and directly place the file in the right directory.
Fixes a regression introduced in commit b3e1d31 ([archiso] Rework
checksum function, 2014-06-28).
Signed-off-by: Lukas Fleischer <archlinux@cryptocrack.de>
Among other things, 36459f3 ([archiso] Drop aitab support, 2014-06-28)
removed the possibility to manually set a specific architecture by using
kernel parameters. This, however, is useful, e.g. when installing Arch
Linux on a device that reports itself as i586 but works fine with the
i686 flavor.
Signed-off-by: Lukas Fleischer <archlinux@cryptocrack.de>
Ensure that all inode table is initialized at filesystem creation.
(Anyway filesystem is small, so at first mount is may quickly initialized)
This avoid possible COW usage during runtime ;)
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
On low memory systems installing via PXE is not possible.
Do not force copytoram=y for NFS/NDB.
Broken/hang things:
* On shutdown.
* On run-time if some network settings changed.
Implements FS#32006.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
If not set, nothing is changed in current behaviour.
Since dm-snapshot allows to use as "COW" a file with any size,
without caring about the the size of "origin", we can avoid creating a
"COW" file of the same size as the "origin". This is really useful,
when using as cow_device= a filesystem that is VFAT where sparse files
are not supported, so if root-image.fs is 1000M, passing cowfile_size=25%
will create a root-image.cow of 250M instead of 1000M.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
This reverts commit 6f68bec907.
Using ash was actually correct as only this shell is available in initramfs.
It was the commit message of 27940c420f that was wrong
Simplify build process, using the same method like in shutdown hook by mkinitcpio
we still need a custom shutdown script for archiso,
since "twisted" structure of block devices.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
This command makes a list of installed packages on root-image,
on <ISO>/$archisobasedir/pkglist.$arch.txt with the format
<repo>/<package>-<version>
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
If EFI/archiso/efiboot.img exists then an alternative "El Torito"
boot image is added to the ISO image compatible for EFI.
This image is a FAT filesystem, that is interpreted by EFI as ESP
(EFI System Partition).
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Remove devtools dependency.
* Better control over what files are touched inside chroot (root-image).
Now: NONE :)
* Two new commands:
+ init: To install {base} group and other needed packages (syslinux for now)
+ run: If we want to run some command inside chroot
(mkinitcpio, locale-gen, useradd, etc etc...)
* Renamed command: "created" to "install", says much better what does.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
Needs mkinitpcio >= 0.8.4
Next changes will be for mdadm and mkinitcpio-nfs-utils:
* /lib/initcpio/udev/11-dm-initramfs.rules
* /lib/initcpio/ipconfig , /lib/initcpio/nfsmount
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Add libraries from /usr/lib and /sbin/losetup.
* Use used_block_devices to detach loop devices.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
This list includes block devices of:
* Boot medium [archisodevice] (only if no copytoram= is used)
* Loop medium [img_dev] (only if no copytoram= is used)
* COW space [cowdevice] (only if cowdevice= is used (no tmpfs is used))
* Loop devices used for SquashFS images.
* Loop devices used for device-mapper devices (two per each dm-device, one RO{*.fs} and one RW{*.cow})
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
New losetup from util-linux 2.21 uses new /dev/loop-control from Linux 3.1.
Needs mkinitcpio-0.8.2+ (provides modules.devname, or workaround with earlymodules=loop)
Get advantages from it! Removing custom shell code.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
This is needed for initscripts 2012.01, at least the entry for /.
Otherwise / is remounted ro, but not remounted rw later.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
This was the first implementation that downloads the complete ISO,
but now we have the posibility of download only needed *.sfs files.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
This was useful to me during development of dm-snapshot support
to create very quickly isos without wating for SquashFS compression.
Is time to remove this, I think the is no practical usage,
and make the code a bit more simple.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Now we have more than one PXE boot methods.
* archiso_ndb_srv= accepts special ${pxeserver} parameter like in others PXE hooks.
* This allows to setup network from initramfs without forcing using a special PXE hook.
* Update README
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
On machines with common graphics chipsets (radeon, intel, nvidia), enable
KMS early. Downloading images from a remote server can take time, better
view the status with a nice resolution.
Otherwise does not work.
22100 Nov 29 09:29 ./lib/libnss_dns-2.14.1.so
20 Nov 29 09:29 ./lib/libnss_dns.so.2 -> libnss_dns-2.14.1.so
46732 Nov 29 09:29 ./lib/libnss_files-2.14.1.so
22 Nov 29 09:29 ./lib/libnss_files.so.2 -> libnss_files-2.14.1.so
Looks like my test was made before introducing symlinks
and using directly libnss_dns-$VERSION.so.2 as libnss_dns.so.2.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
We are currently checksuming all files inside $archisobasedir
(aitab, *sfs, kernel/initramfs and boot/syslinux related files)
this is stored in one file only. Its works good for now,
but I think we only need to checksum only aitab and *.sfs if they are needed.
Currently if you boot a dual-iso in i686, also checksum is done
for *.sfs of x86_64. Not a big issue, this just take a bit more time.
The real issue is when booting via PXE with HTTP/FTP methods,
since they download only aitab and needed *.sfs files,
can not use directly the only one file where checksums are stored.
This patch does:
(1) Do not checksum syslinux related files, anyway you are already reach initramfs stage.
(2) Do not checksum kernel/initramfs files, for the same as (1)
(3) Two checksum.${arch}.md5 for each i686 and x86_64 for only aitab and *.sfs.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
[ramfs /]# cat /etc/resolv.conf
nameserver 192.168.0.77
[ramfs /]# curl -O http://releng.archlinux.org/pxeboot/README
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 1129 100 1129 0 0 2191 0 --:--:-- --:--:-- --:--:-- 4589
[ramfs /]#
Do not "add_binary /lib/libnss_dns.so.2", because will add libnss_dns-VERSION.so.2 as libnss_dns.so.2
Is not an issue will work (nobody links directly to VERSION), but is not nice.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
It add support for using *.sfs images via NFS.
It must be located after archiso_pxe_nbd in mkinitcpio.conf, in that way
NBD is the default.
New boot parameters:
archiso_nfs_srv=IP:/path
Set the NFS IP and path. Accepts ${pxeserver} as IP.
archiso_nfs_opt= Sets NFS mount opts (comma separated)
Defaults to:
port = as given by server portmap daemon
rsize = 1024
wsize = 1024
timeo = 7
retrans = 3
acregmin = 3
acregmax = 60
acdirmin = 30
acdirmax = 60
flags = hard, nointr, noposix, cto, ac
:: Running Hook [archiso_pxe_common]
IP-Config: eth0 hardware address 52:54:00:12:34:57 mtu 1500
IP-Config: eth0 guessed broadcast address 192.168.0.255
IP-Config: eth0 guessed nameserver address 192.168.0.218
IP-Config: eth0 complete (from 192.168.0.218):
address: 192.168.0.90 broadcast: 192.168.0.255 netmask: 255.255.255.0
gateway: 192.168.0.218 dns0 : 192.168.0.218 dns1 : 0.0.0.0
rootserver: 192.168.0.218 rootpath:
filename :
:: Running Hook [archiso_pxe_nbd]
:: Running Hook [archiso_pxe_curl]
:: Running Hook [archiso_pxe_nfs]
:: Mounting '192.168.0.77:/mnt/iso'
...
192.168.0.77:/mnt/iso/ on /run/archiso/bootmnt type nfs (rw,relatime,vers=3,rsize=524288,wsize=524288,namlen=255,hard,nolock,proto=tcp,port=65535,timeo=7,retrans=3,sec=sys,local_lock=all,addr=192.168.0.77)
...
My /etc/exports
/mnt/iso 192.168.0.90(ro,no_subtree_check,no_root_squash)
Its add support for downloading the ISO image via HTTP/FTP with curl or download *.sfs files.
The files are downloaded on /run/archiso/curlspace (tmpfs) and setup a loop dev from it
(like in archiso_loop_mnt) in the ISO case, or just do a bind mount to
/run/archiso/bootmnt in the other case.
This hook must be located after archiso_pxe_nbd in mkinitcpio.conf, in that way
NBD is the default.
New bootparm:
* archiso_curl_url=(first form)
archiso_curl_url=http://192.168.0.7/archlinux/iso/archlinux.iso (absolute form)
Use an absolute URL. Fetch the entire ISO
archiso_curl_url=ftp://${pxeserver}/archlinux/iso/archlinux.iso (relative form)
Use the as IP the same server where PXE reside. Fetch the entire ISO
* archiso_curl_url=(second form)
archiso_curl_url=http://192.168.0.7/archlinux/iso/unpacked (absolute form)
Use an absolute URL. Where unpacked (an example name) is a directory where
${archisobasedir} ("arch" by default) is located with all *.sfs.
This fetch only needed files.
archiso_curl_url=ftp://${pxeserver}/archlinux/iso/unpacked (relative form)
Use the as IP the same server where PXE reside. Where unpacked (an example name)
is a directory where ${archisobasedir} ("arch" by default)
is located with all *.sfs. This fetch only needed files.
* curlspace_size= Set a size for a tmpfs filesystem where files are downloaded.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
One hook only setup the network device (archiso_pxe_common) the other
setup the NBD client (archiso_pxe_nbd).
New bootparam:
archiso_pxe_srv=IP
Allow to set an IP different from the PXE server.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
In this way allow to mount /run/archiso/bootmnt from another hook,
but still use the logic from the main mount_hook, skipping only this mount.
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* Now bootmnt/img_dev is always a ro-mount,
and cowspace is first ro-mount then rw-mount.
* Fix a "hidden" bug: at the time of test [[ -ef ]], if devices nodes
are not ready, such test will fail then archisodevice will mounted ro and
when cow_device is mounted to be rw it fails.
(I recently suffered this on a machine with slow USB)
Signed-off-by: Gerardo Exequiel Pozzi <vmlinuz386@yahoo.com.ar>
* 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>