Thomas Lamprecht [Mon, 26 Feb 2024 19:40:14 +0000 (20:40 +0100)]
sys/command: double wait frequency and send SIGKILL once after 0.5s
100 ms is quite plenty, while we would be better of using a event
based wait, i.e., dropping the WNOHANG, that would also mean handling
the time out via alarm, EINTR checking and quite a bit other stuff
making this more convoluted, so for now just go faster..
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Christoph Heiss [Mon, 26 Feb 2024 16:50:43 +0000 (17:50 +0100)]
sys: command: wait for process exit with sub-second granularity
Using full seconds as a granularity for sleeping between waitpid()'s is
way too much and unnecessarily slows down the installation a lot. Most
processes take a few moments after closing their stdin/stdout to
actually exit fully, which means that we would sleep a second in most
cases.
Lower it to 0.1 second, which immensely improves the situation.
Some values for comparison; tui-installer on the same bog-standard
2-core, SeaBIOS, ext4, virtio VM (roughly averaged over multiple runs):
* 8.0 ISO (baseline): ~2:30 min
* w/o patch: ~9:00 min
* w/ patch: ~2:30 min
Values measured are from pressing the 'Install' button until the
autoreboot dialog (aka. install finished) popped up.
Fixes: 152bbef ("sys: command: factor out kill() + waitpid() from run_command()") Reported-by: Stoiko Ivanov <s.ivanov@proxmox.com> Reported-by: Filip Schauer <f.schauer@proxmox.com> Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>
Thomas Lamprecht [Mon, 26 Feb 2024 13:38:11 +0000 (14:38 +0100)]
run command: use explicit return undef in closures on call sites
To avoid a misinterpretation of the auto-return value:
> In the absence of an explicit return, a subroutine, eval, or do FILE
> automatically returns the value of the last expression evaluated.
-- https://perldoc.perl.org/functions/return
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Thomas Lamprecht [Mon, 26 Feb 2024 13:34:09 +0000 (14:34 +0100)]
run command: avoid using 1 as special value
In Perl, the last expression of a block (e.g. of a method, eval) gets
returned if there's no explicit return statement. Quite often that is
truthy, i.e., 1.
As that was chosen as the special value for the CMD_FINISHED flag it
had quite a few false positives, causing weird effects and
installation failure.
Reserve that overly problematic value and chose 2 as new CMD_FINISHED
value, albeit it could be better to signal this even more explicitly,
like with a structured hash reference, but for now this is a good stop
gap.
Fixes: 23c5fbe ("sys: command: allow terminating the process early from log subroutine") Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Thomas Lamprecht [Sat, 24 Feb 2024 17:17:10 +0000 (18:17 +0100)]
tests: code-style and error handling fixes for ui2stdio
use modern calling style, avoid duplicate use of Test::More module,
handle fork error more visible, handle pipe creation errors and do
that all in one commit as it's just a test and I don't care.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Thomas Lamprecht [Sat, 24 Feb 2024 16:56:26 +0000 (17:56 +0100)]
stdio connected UI: drop perl prototype definitions
The prototypes where completely circumvented by calling those two
methods by reference via &, and that probably happened as the send_msg
one was just wrong, it forced scalar context for the second parameter,
while that was a list (or well hash, but the difference there can be
blurry).
Anyhow, prototypes are not always of help, and can be a PITA with
side-effects too, and especially for such small modules it has not
that much use to declare them for privately-scoped methods, so just
drop them and fix the calling style.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Thomas Lamprecht [Fri, 23 Feb 2024 16:19:15 +0000 (17:19 +0100)]
fqdn comparison: expand test scope
Add some negative tests to ensure a `return true` (exaggerated)
refactoring won't pass the suite, and add one test where a and b is
the same, just to be sure.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Christoph Heiss [Thu, 15 Feb 2024 12:39:38 +0000 (13:39 +0100)]
fix #5230: sys: net: properly escape FQDN regex
Due to interpolation, the \. sequence must be double-escaped.
Previously, this would result in a non-escaped dot, thus matching much
more liberally than it should.
Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>
[ TL: fix bug # reference in code comments ] Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Christoph Heiss [Tue, 13 Feb 2024 15:14:03 +0000 (16:14 +0100)]
fix #4872: run env: use run_command() for country detection
This fixes a rather longstanding issue [0][1] with the country
detection, in that it might get completely stuck and thus hangs the
installation.
This is due how Perl, signals and line reading interacts.
A minimal reproducer, how the installer currently works, looks like
this:
```
#!/usr/bin/env perl
use strict;
use warnings;
open (my $fh, '-|', 'sleep', '1000') or die;
my $prev = alarm(2);
eval {
local $SIG{ALRM} = sub { die "timed out!\n" };
my $line;
while (defined ($line = <$fh>)) {
print "line: $line";
}
};
alarm($prev);
close($fh);
```
One might expect that this times out after 2 seconds, as specified in
`alarm(2)`. The thruth is that `$line = <$fh>` apparently prevents the
signal to go through. This then causes the installer to hang there
indefinitely, if `traceroute` never progresses - which seems to happen
on lots of (weird) networks, as evidently can be seen in the forum [1].
Proxmox::Sys::Command::run_command() handles of these weird cases, takes
care of the nitty-gritty details and - most importantly - interacts
properly with SIGALRM, so just use that instead.
This _should_ really fix that issue, but reproducing it 1:1 as part of
the installation process is _very_ hard, basically pure luck. But
rewriting the reproducer using run_command (in the exact same way that
this patch rewrites detect_country_tracing_to()) fixes the issue there,
so it's the best we can probably do.
NB: This causes that the traceroute command is now printed to the log
(as run_command() logs that by default), which we could also hide e.g.
through another parameter if wanted.
Christoph Heiss [Tue, 13 Feb 2024 15:14:01 +0000 (16:14 +0100)]
sys: command: allow terminating the process early from log subroutine
If the logging subroutine $func returns CMD_FINISHED after processing a
line, the running subprocess is killed early.
This mechanism can be used when e.g. only a certain part of the output
of a (long-running) command is needed, avoiding the extra time it would
take the command to finish properly.
This is done in a entirely backwards-compatible way, i.e. existing
usages don't need any modification.
Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>
Christoph Heiss [Tue, 13 Feb 2024 15:14:00 +0000 (16:14 +0100)]
sys: command: handle EINTR in run_command()
Previously, the I/O loop would continue endlessly until the subprocess
exited.
This explicit handling allows run_command() to be used with e.g.
alarm().
Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>
Christoph Heiss [Tue, 13 Feb 2024 15:13:59 +0000 (16:13 +0100)]
sys: command: factor out kill() + waitpid() from run_command()
This moves the kill() + waitpid() combo into a separate subroutine,
avoiding open-coding that sequence. wait_for_process() also handles
properly unkillable process (e.g. in D-state) and avoids completely
locking up the installer in such cases. See [0].
For the latter case, a timeout exists (with a default of 5 seconds) in
which to wait for the process to exit after sending an optional
TERM/KILL signal.
Also while at it, add a few basic tests for run_command().
Christoph Heiss [Tue, 13 Feb 2024 15:13:58 +0000 (16:13 +0100)]
low-level: initialize UI backend for 'dump-env' subcommand too
Some detection routines might try to log things and call some
Proxmox::Ui functions all the way down, so just initialize it with the
stdio backend to avoid errors.
Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>
For consistency sake, all colons and trailing spaces in labels that were
followed with an entry were removed, this matches other panels such as
the password and country/timezone panels.
Signed-off-by: Maximiliano Sandoval <m.sandoval@proxmox.com> Reviewed-by: Christoph Heiss <c.heiss@proxmox.com> Tested-by: Christoph Heiss <c.heiss@proxmox.com> Tested-by: Tested-by: Lukas Wagner <l.wagner@proxmox.com>
Using boxes causes the labels to not align correctly in certain
circumstances. In the following commits we replace the use of boxes with
grids and set the margins and spacing directly on the respective grid.
Signed-off-by: Maximiliano Sandoval <m.sandoval@proxmox.com> Reviewed-by: Christoph Heiss <c.heiss@proxmox.com> Tested-by: Christoph Heiss <c.heiss@proxmox.com> Tested-by: Tested-by: Lukas Wagner <l.wagner@proxmox.com>
Previously the grids were inserted in a succession of boxes each with
its own set of margins and spacing. We define the margins now
exclusively in the grid and account for previous values.
Note that we match the top and bottom margins of the 'Target Harddisk'
panel which does not need to use a grid.
Signed-off-by: Maximiliano Sandoval <m.sandoval@proxmox.com> Reviewed-by: Christoph Heiss <c.heiss@proxmox.com> Tested-by: Christoph Heiss <c.heiss@proxmox.com> Tested-by: Tested-by: Lukas Wagner <l.wagner@proxmox.com>
Christoph Heiss [Tue, 6 Feb 2024 13:12:26 +0000 (14:12 +0100)]
proxinstall, common: remove "off" as zfs checksum option
See also the thread at [0] for the initial discussion/idea.
Disabling checksums is considered an "extraordinarily bad idea" [1] (for
pretty obvious reason) and nobody should really ever use it.
Thus remove the option completely; just so that users cannot simply
disable checksum "for performance reasons" without knowing about the
implications of this.
As pointed out by Thomas, it can still be set to "off" after the
installation using the `zfs` tool, if really wanted.
Christoph Heiss [Fri, 22 Dec 2023 10:52:24 +0000 (11:52 +0100)]
proxinstall, common: remove deprecated fletcher2 as zfs checksum algorithm
Fletcher-2 has long been deprecated and should not be used anymore
[0][1], so we probably should not offer it anymore too. It's been
deprecated since at least over 3 years, beyond that it's hard to find
an exact date.
Stoiko Ivanov [Tue, 21 Nov 2023 13:11:18 +0000 (14:11 +0100)]
zfs: create dataset var-lib-vz for /var/lib/vz
Creating rpool/var/lib/vz and all intermediate datasets causes a
service-failure of `var.mount` upon shutdown.
creating the dataset for /var/lib/vz directly at the rpool and setting
its mountpoint property seems the most robust way to address this.
The alternative approach of setting `canmount=off` on the `var`
dataset seems a bit dangerous (users setting a zfs property and
suddenly hiding their /var contents).
The only small downside to this approach is that the setting of the
mountpoint happens quite a bit after extracting the data - but this
would probably be better addressed with a refactoring of the
lowlevel-installer code (setting the zfs-pool up under /target and
getting rid of a few special cases)
Stoiko Ivanov [Tue, 21 Nov 2023 11:09:59 +0000 (12:09 +0100)]
serial installer: add serial config for grub to target system
Matching if a serial will be needed for grub is based on the target
commandline - the speed is also read from there. The unit is based
on the ttyS device - although I'd assume that this might not always
match up.
Stoiko Ivanov [Fri, 17 Nov 2023 17:30:23 +0000 (18:30 +0100)]
tui: fix interface sort order
Currently, when multiple NICs are present in a system the TUI
sometimes selects the wrong interface (not the one that has the
default gateway/dhcp lease)
I assume this is due to HashMap's values yielding an iterator in
arbitrary order
Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com> Co-authored-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
[ TL: avoid intermediate vector, reuse the SelectView's iter()] Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Stefan Sterz [Fri, 17 Nov 2023 16:32:18 +0000 (17:32 +0100)]
tui: bootdisk zfs config: add a maximum value to the `copies` option
according to `man zfsprops` the copies option can only be 1, 2, or 3.
limit the field to 3 just like we do for the GTK based UI, as setting
higher options can't work anyway.
Signed-off-by: Stefan Sterz <s.sterz@proxmox.com>
[ TL: fleece in note that we already limit this in the GTK UI ] Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Christoph Heiss [Fri, 17 Nov 2023 13:45:27 +0000 (14:45 +0100)]
tui: install progress: use ok/cancel as button text for installer prompt
The GTK installer/UI module in the low-level installer does the same.
Messages used with this are worded for this, using yes/no instead can be
quite confusing (e.g.
Proxmox::Install::ask_existing_vg_rename_or_abort())
Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>
Christoph Heiss [Fri, 17 Nov 2023 12:12:18 +0000 (13:12 +0100)]
tui: fix changing between non-LVM and LVM filesystem in bootdisk chooser
Happens due to a force-unwrap() under the false assumption that the
disk for LVM configurations always exists when switching to a LVM
filesystem.
This fails spectacularly with a panic when switching from e.g. Btrfs to
ext4 in the filesystem chooser.
Fixes: eda9fa0 ("fix #4856: tui: bootdisk: use correct defaults in advanced dialog") Reported-by: Christian Ebner <c.ebner@proxmox.com> Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>
Stoiko Ivanov [Thu, 16 Nov 2023 19:59:47 +0000 (20:59 +0100)]
run env: do not store emtpy hostname
without this patch the hostname ends up as the empty string in
run-env-info.json, which results in a parse-error in the TUI code
(an empty string is not None, but still too short as hostname)
Minimally tested on a VM.
Fixes: bda1cdf ("run env: retrieve and store hostname from DHCP lease
if available") Signed-off-by: Stoiko Ivanov <s.ivanov@proxmox.com>
Stoiko Ivanov [Thu, 16 Nov 2023 15:00:41 +0000 (16:00 +0100)]
zfs: set acltype=posix for root-dataset
journald as a core component tries setting a ACL on the journal files
for (non-root) users and fails on our ZFS installs.
Resulting in dmesg being spammed with messages from journald upon each
journal-rotation for each user upon their first login.
This is also suggested by OpenZFS in their Debian guide for root on
ZFS:
https://openzfs.github.io/openzfs-docs/Getting%20Started/Debian/Debian%20Bookworm%20Root%20on%20ZFS.html
Tested by setting this on a machine of mine, where this has been
bugging for quite a while.
Stoiko Ivanov [Thu, 16 Nov 2023 15:00:38 +0000 (16:00 +0100)]
fix #4747: pass kernel cmdline parameters to target system
Parameters needed for booting during installation are best preserved
in the target cmdline as well - e.g. if you need a particular
cmdline switch for your system to boot at all - not having to add it
for the first boot of the installed system and manually adding it to
the bootloader config is an improvement.
This additionally enables us to drop the console parameter handling
for serial consoles (it is just one of the parameters to pass along).
Finally it fixes the regular expressions for the installer settings we
read from the cmdline (swapsize, maxroot,...) which were broken if
added as last entry.
Stoiko Ivanov [Tue, 14 Nov 2023 17:31:48 +0000 (18:31 +0100)]
serial install: fix console parameter parsing
The regex matching in Proxmox::Install::Config was blindly copied from
above - so the other parameters are also likely to not get recognized
if they are the last on the cmdline
Stoiko Ivanov [Thu, 22 Jun 2023 20:08:48 +0000 (22:08 +0200)]
pass optional console parameter from installer to target
If an installation needs to provide a dedicated console parameter
(e.g. because it runs on the serial console) the target system most
likely will need the parameter too.
This patch adds the parameter to the kernel-commandline (in case zfs
is used for both grub and systemd)
Christoph Heiss [Fri, 10 Nov 2023 14:17:26 +0000 (15:17 +0100)]
low-level, tui: count down auto-reboot timeout
The GUI installer already has the same functionality, with this the TUI
installer gains the same. It is a nice touch anyway, primarily to
indicate to the user that the installer is not frozen or similar.
Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>
Christoph Heiss [Thu, 9 Nov 2023 09:40:56 +0000 (10:40 +0100)]
fix #4856: tui: bootdisk: use correct defaults in advanced dialog
The size of the install disk was set to the size of the first disk,
regardless of what disk was selected. This only happened if the advanced
options dialog was never opened, and only a disk was selected in the
main bootdisk dialog.
Properly solving this involved restructuring the LVM advanced bootdisk
dialog, to also hold the selected disks, like the ZFS and Btrfs dialogs.
In addition to that, the `BootdiskOptionsRef` needs quite some passing
around, to cover all the cases, since the dialog also needs to be
"reentrant-safe".
I tested (among other things):
* Only select disk, don't open the advanced dialog, go to summary,
then back to the bootdisk dialog -> selected disk should be kept
* Select disk, open advanced dialog but leave everything as is, go to
summary, then go back again -> selected disk should be kept
* Same as previous, but change the "Total size" for the disk, go to
summary and back -> selected disk and size should be kept
* Same as previous, but additionally change filesystem to XFS -> disk,
filesystem and size should be kept
* Same as previous, but then create a ZFS RAID, go to summary & back,
ZFS RAID should be kept with all parameters
* etc ..
Further I also verified that the correct disk size(s) get written into
the setup structure for the low-level installer.
Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>
Christoph Heiss [Thu, 9 Nov 2023 09:47:56 +0000 (10:47 +0100)]
low level: testmode: take path to disk image instead of using /dev/null
.. in exactly the same way GUI and TUI installer do, streamlining them.
Up until now, testing the TUI installer often involved hand-editing the
`run-env-info.json` to put some proper disk sizes > 0 in place. This
makes this process a lot easier.
Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>
install: install correct grub metapackage for the current boot-mode
grub packages in debian split between:
* meta-packages, which handles (among other things) the reinstalling
grub to the actual device/ESP in case of a version upgrade (grub-pc,
grub-efi-amd64)
* bin-packages, which contain the actual boot-loaders
The bin-packages can coexist on a system, but the meta-package
conflict with each other (didn't check why, but I don't see a hard
conflict on a quick glance)
Currently our ISO installs grub-pc unconditionally (and both bin
packages, since we install the legacy bootloader also on uefi-booted
systems). This results in uefi-systems not getting a new grub
installed automatically upon upgrade.
Reported in our community-forum from users who upgraded to PVE 8.0,
and still run into an issue fixed in grub for bookworm:
https://forum.proxmox.com/threads/.123512/
Reproduced and analyzed by Friedrich.
This patch changes the installer, to install the meta-package fitting
for the boot-mode.
We do not set the debconf variable install_devices, because the
'install_devices' variable is only defined for 'grub-pc', and thus
(still) only set for that package/namespace.
Christoph Heiss [Tue, 31 Oct 2023 12:10:54 +0000 (13:10 +0100)]
tui: views: add optional suffix label for `NumericEditView`s
Most of the churn here is due to changing the inner view from an
`EditView` to a `LinearLayout`. Also prompted the introduction of two
small helpers .inner() and .inner_mut() to simplify things everywhere
else in the view.
Signed-off-by: Christoph Heiss <c.heiss@proxmox.com>