- the *mount* module inserts a rootMountPoint without trailing /
into global storage, so we can't assume that here. On the other
hand, the paths passed in to the Worker functions are absolute
paths -- adjust the tests to follow that.
- The code in Workers.cpp assumes that rootMountPoint ends in a /
so that it can have filenames appended easily; make the tests
fit that assumption, but still need to check that it is so in
production.
- refactor running the command into a helper function,
to deal with the regular if-command-failed-then-complain pattern.
- mark parameters as unused.
- move distinction about kind of DBus file up into the MachineIdJob
and remove the enum that marked it.
- Testing some of the functionality that's been added just now:
- copyfile fails, buggy implementation
- poolsize fails, buggy implementation
- removefile not tested
- read-urandom or copy-existing-file are implemented
- fairly chatty on failure
- needs tests (probably the implementation should be moved to
a separate file and unit-tested)
- keep the rootMountPoint and the path-with-random-data separate
instead of concatenating them at the beginning. Then we can
use the "clean" names also within the host system.
- remove existing files for each kind of random-generation
that is enabled. There's a helper function for the case that
Cala is no longer setuid and needs help to remove those files
from the target (e.g. a setuid helper).
- Just use the existing rsync code, which can do both
files and directory trees.
- The existing code assumed we were always copying directories.
Now double-check beforehand.
- if a default DE is configured but the executable doesn't exist,
believe the .desktop file. Then use that, and warn if the
whole thing can not be found.
- for a DE entry which has a bad executable setting,
update the entry from the .desktop file using TryExec.
This assumes that the TryExec command is actually something
you might want to run.
- Sessions can be X11-sessions (living in xsessions) or Wayland-
(living in wayland-sessions). Look in both places.
- Refactor code a little to make it nicer to read.
- Do the async GeoIP checking in the async requirements-checking phase
- Do not return any requirements results -- we just need the async bit
- Drop the waiting widget, since it's not needed (done by the
requirements phase)
- If there is an item with id "" (empty), it is used as the
"no-package-selected" placeholder text.
- Existing code iterated over the abstract model and used the
name and description at the time the model was set -- but
by getting the name and description from the model, only
a single string was obtained instead of the full range
of translations.
- Therefore, when arriving on the page, the "no-package-selected"
information was displayed from the translation that was active
when the model was set.
Instead, extend the non-abstract model so we can find the no-package-
selected item and pass that explicitly to the page.
FIXES#1241
- Since the package chooser might be used more than once, or for
more specific items than "Packages", introduce a way to provide
specific strings for display.
- The only string needed is the ViewStep name, since the item with
id "" can be used for the no-selection item.
- This helps give meaningful names to code chunks
- Gives clang-format something to work with
- Reduces indentation depth
I think this is a bit of a code-style opinion issue: big complicated
lambdas doing UI things just don't seem like a good idea.
- hangs unpredictably during testing
- replace with the Calamares process-invocation runCommand(), which is also
synchronous but doesn't hang (or, hasn't, in testing so far)
If we don't have/need an image for the rootfs, we might want to
configure the `/` directory as a source for unpackfs. Unfortunately,
this raises an error:
- unpackfs first creates a temporary directory
- it then creates a subdirectory for each source, using the source
path's basename
- when the source is `/`, the basename is an empty string, therefore
the module tries to create an already existing directory
In order to prevent this error, we use the `os.makedirs` function with
parameter `exist_ok=True` instead of `os.mkdir`.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- AppData and AppStream can be disabled independently of finding
their requirements (possibly useful if you want to ignore
AppStream even when it's installed in your build environment).
- Add a little top-level documentation about WITH_
- These don't have to be static methods of PackageItem, a free
function is more convenient.
- Since it's not API of PackageItem anymore, need to
- update tests not to use API
- do API-not-available warnings in consumers
- The smallest size image of the default (or, if there is no
default, the first) screenshot is used.
- Remote URLs are not supported by QPixmap, so most will not
load anyway.
- Use *appstream* as key in one of the items for the package-
chooser to load data from the AppStream cache in the system.
- Usable for some applications; for DE-selection not so much.
- Currently unimplemented.
- Put the implementation entirely in a separate file, keep the
not-supported one in PackageModel.cpp (but only in an #ifdef).
- Makes the various optional-data-sources more similar.
- Simplify the iteration by first determining which partitions
are mountable (at all).
- This guards against the very rare case that a partition
does not have a mountPoint at all (the if guarded against that)
where the lambda passed to sort() would get a KeyError.
Instead of having a special case for extra mounts to be processed right
after the rootfs, a better approach is to add them to the partitions
list, and then sort the list by mount point.
This way, we make sure every partition is mounted right when it is
needed: `/` is obviously mounted first, `/run` is mounted before
`/run/udev`, and so on.
The overall process is therefore more generic and should suit all
use-cases.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- the "Select language" tooltip was applied to the form, so it would
show up inappropriately all over the place
- the buttons didn't have useful tooltips.
- having show*Url and donateUrl seems inconsistent, although
the show*Url settings were originally boolean-only.
- add "show" to the Donate button setting, to make them
all consistent (putting a boolean there will generate a
warning and hide the button, that's all).
- the generic (enum-based) setupButton() can handle all four
of the buttons, so setupLinks() can go away. Only the
(re)translation of the text on the button needs to be
done, so move that to the main RETRANSLATE.
- Handle buttons and their URL-opening in a more
general way with an enum; drop existing three-boot
method and special setupDonateButton()
- Doesn't compile because consumers haven't changed.
As the config files integer are now of type `QVariant::LongLong` instead
of `QVariant::Int`, requirements relying on this type were not parsed
correctly.
This patch fixes this, and adds an option to the python conversion to
take into account `QVariant::LongLong` types.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
When the rootfs partition is read-only, mount points for the other
partitions cannot be created, therefore they need to be created in a
tmpfs, already mounted somewhere in `/`.
However, the extra mounts are only mounted at the end, which causes an
error as no tmpfs is currently mounted.
This patch makes sure all extra mounts are mounted right after the `/`
partition, allowing the use of a read-only rootfs.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
This variable is declared in `if m:`. Of course if this codepath doesn't
run, the variable is not declared an Python doesn't like this kind of
surprise...
Signed-off-by: Arnaud Rebillout <arnaud.rebillout@collabora.com>
- If KPMcore is found -- it requires some other KDE Frameworks but
at least in pre-4.0 versions doesn't check very well for them --
then missing its dependencies is no cause for CMake failure.
Instead, log it nicely and suppress the module.
- calamares_automoc() sets AUTOMOC, but also adds some flags
to avoid compilation warnings from the generated MOC code.
- drop weird hard-coded include paths
- KDE and GNOME selection images were drawn by me for the bogus
package model, and should not be used.
- Keep Calamares logo though, even if it doesn't make much sense
to use it in package selection.
- Keep the no-selection image since it might be used more often,
but it's not very good.
- The ID and Screenshot entries might be weird in AppData (in particular,
a remove URL) so put those back under the control of Calamares even
when using AppData as the source of descriptions.
- Document all the static inline methods that do the work
- Fill up a QVariantMap from <name>, <summary> and <description><p>
elements, and use that to initialize the PackageItem.
- Doing a manual read of the XML, since existing appdata libraries
don't seem to have a convenient entry for what I need.
- Expand tests to loading AppData (currently, they fail).
- Add a FALLTHRU macro to annotate fallthrough situations in both
Clang and GCC,
- Annotate intentional fallthroughs.
- Add missing break which meant that the selection mode was
always multiple-selection.
- Using id's as keys in a map orders them indeterminately -- in
practice, alphabetically by key. Switch to a list form so that
the products stick to the order they have in the config file
(which means distro's can list "preferred" versions at top).
Package chooser is a **low density** package selector -- unlike
netinstall which offers a high density tree view -- for picking
zero, one, or more items from a small collection of packages.
This can be used, e.g., for "pick exactly one desktop environment",
"pick zero or more text editors" which can then be installed
by another module. The UI is big and shiny (rather than netinstall's
text-based tree view) and isn't suitable for more than a dozen or
so items.
- preservefiles generally needs to have the target filesystems
mounted, so that it can preserve to them; but you can also
configure it such that there is no need for mounted filesystems
(e.g. in OEM setup).
- Add an example line in CMakeLists.txt to show how that would be done.
- The mount module must happen before unpackfs because that (mount)
module sets up the root mount point (in /tmp) and some other
variables needed later.
Added new configuration "efiMountOptions" to fstab.conf
When generating the fstab entry for the ESP, take the mount options from
the new configuration or fall back to "mountOptions".
- The mitigations are slightly intrusive, and may clash
with other, similar mitigations (especially for initramfs,
the recommended solution is to configure the system with
the snippet outside of Calamares).
- These tests exercise the createTargetFile() logic,
which is essential for creating a safe initramfs
configuration snippet.
- Could be moved into libcalamares instead, since the tests
are not really initramfs specific.
- This is a simple variation on the theme of things-that-call-a-
initramfs-updater, so the code is mostly a copy of initramfs/
module. I didn't even bother to strip out the configuration-
handling (I figure it might be good for *something*) so now
"" and "$uname" are valid kernel names as well.
- Fixes security issue where the initramfs ends up readable
by all, and that includes the cryptfile for LUKS.
SEE #1190
- Rename classes and functions to be more descriptive
(a LuksDevice is .. information for a LUKS device, for instance).
- Move the smarts of unpacking a QVariantMap to LuksDevice.
- Apply code formatting
- Use 120 seconds for update-initramfs, instead of 10. Previous
Python code had no timeout at all, which wasn't so hot either.
10 seconds, though, is too short for slow CPU & slow disk.
- new implementation handles blank (maps to "all") configuration,
- allows specifying "$uname" as kernel name, to use `uname -r`,
- allows specifying a specific kernel.
- Just read /proc/partitions and process it; split into columns,
add relevant bits.
- This implementation supports devices named "name", which the other
didn't (but that would be really weird).
The tests now pass.
- This is a tiny bit of TDD to replace the existing implementation
(a shell pipeline) with processing inside Calamares.
- The test fails right now, since the implementations are not
the same.
- provide complete information for feature_summary
- set the right API version when building libcalamares
- report the beta version number when it's wrong
- Starting to centralize utility code for partitioning into
libcalamares instead of scattered and weirdly shared between
modules.
- This particular commit breaks compiling the modules, though.
- after model resets, restore what was previously selected.
- This avoids having an **empty** combobox, SEE #1141 but does
not actually set it back to the value the user had previously
picked (e.g. changing swap settings **still** breaks the
selection).
- The sub-directories under libcalamares (e.g. Utils, ..)
all live in namespace CalamaresUtils (well, except for Logger).
The services (e.g. subdirs other than utils/) live in their
own nested namespace, so partitioning should go into
CalamaresUtils::Partition for consistency.
src/modules/bootloader/main.py (install_secureboot): Run the configured
grubMkconfig command (should be `grub-mkconfig` or `grub2-mkconfig`) to
create `/boot/efi/EFI/$efi_bootloader_id/grub.cfg`. The sb-shim is just
a chainloader to GRUB 2, which expects a grub.cfg in that location, so
something has to create it or the installed system will not boot beyond
the GRUB rescue shell.
(install_grub): Fix misleading comment above the grubMkconfig call: it
is not the file specified in grubCfg that should be already filled out
by the grubcfg job module, that file is written by `grub*-mkconfig`
using `/etc/default/grub` as the input file. It is that input file
`/etc/default/grub` that should already be filled out by the grubcfg job
module. (The same input file is used in install_secureboot.)
- The output of subprocess is a bytes object, which needs to
be decoded so we can use it like a regular string (alternatively,
we could have changed more code to manipulate bytes, but eventually
we need a string to pass to a subsequent command anyway).
- Centralize the sanitizer so that it's consistent in different
environments.
- While here, add () to the sanitizer to avoid some distro's with
parenthesized names from creating weird EFI dirs.
- FIXES#934
- Whether this is really wanted depends on the distro, and I'm not
100% convinced the likely tags from Unicode are correct (or it'd
take a lot more data). In any case, starting Calamares in "NL"
gets me "nl_NL" as translation; presumably starting it in "BE"
will get me that as well (what about Les Wallons?)
- This also shows off that it's a real hack to have so much program
logic in the *widget* parts of each ViewStep. Longer-term,
a lot of functionality should go to the ViewStep itself, which
will then control the UI.
- Which translations are available is a global property
of Calamares itself, not of the plugins, so getting
the model of available translations should live there.
Move the relevant code (which is simple) from the
Welcome module.
- Use namespace CalamaresUtils::Locale consistently for this service.
- Move locale-related non-GUI support code from the Welcome module
to libcalamares; these are generally useful. Both Label (naming a locale)
and LabelModel (managing a bunch of those Labels) have been moved.
On systems with SELinux enabled, we have to create the directories on
top of which we mount another partition or virtual file system (e.g.,
/dev) with the correct SELinux context, BEFORE we mount the other
partition. Otherwise, SELinux will get really confused when systemd
tries to recreate the mount tree for a private file system namespace for
a service. And unfortunately, even an autorelabel does not fix it
because it runs when /dev etc. are already mounted.
Without this fix, on Fedora >= 30, the system installed with Calamares
would fail to start the dbus-broker system bus, leading to several
important pieces of functionality not working (e.g., shutdown as
non-root).
On systems without SELinux enabled, chcon (which is part of coreutils)
will just print a warning and do nothing, so this should always be safe.
Instead of relying on a module-specific implementation, use the new
PartitionSize class for storing partition sizes.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
Instead of relying on a module-specific implementation, use the new
PartitionSize class for storing partition sizes.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
In order to prepare for future refactoring of the PartSize class, move
the bytesToSectors() function to libcalamares in the CalamaresUtils
namespace.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- This small header file contained a few unrelated typedefs.
Move those typedefs to the classes they relate to. This
**does** mean that some consumers need to #include something
else instead.
- Use type names more consistently.
Editorial: why are **pages** responsible for creating the jobs?
- Remove (heavy-handed) top-level include_directories, in favor
of more focused ones; this helps to make sure that the dependencies
ordering is correct.
src/modules/partition/jobs/ClearMountsJob.cpp
(ClearMountsJob::getCryptoDevices): Skip not only `/dev/mapper/control`,
but also `/dev/mapper/live-*`. Fedora live images use
`/dev/mapper/live-*` internally. We must not unmount those devices,
because they are used by the live image and because we need
`/dev/mapper/live-base` in the `unpackfs` module.
src/modules/unpackfs/main.py (UnpackOperation.mount_image): Check
whether entry.source is a regular file or a device and only use
`-o loop` on regular files, not devices.
At least on Fedora >= 29, `-o loop` fails on the read-only device
`/dev/mapper/live-base` (though `-o loop,ro` would be accepted).
- Use a named enum instead of a collection of booleans
- Support old-style configuration but complain about it
- Update AppImage config as well
The new setup allows four different restart modes: never,
always, user-unchecked and user-checked. The user-modes
are interactive and give the user a choice (defaulting to
unchecked-don't-restart and checked-do-restart respectively).
The non-interactive versions vary in how they are
displayed.
- deprecate the old entries
- use a geoip sub-map for GeoIP configuration
- polish up documentation
- drop mention of blank and "legacy" styles for GeoIP config,
just update your URLs already.
- If KPMCore is not found, don't require the KF5 components
that it would depend on.
- If ECM is found, use KDEInstallDirs always, not just when
the partitioning module is used.
- OEMID is a module for configuring phase-0 things for an OEM,
like batch-ID. This is just a stub.
- Currently planned functionality is limited to just batch-ID.
- The only remaining functions in the file are string-related, so
rename to match their purpose.
- Drop this include file from most places, since they don't actually
use the string functionality at all.
- Remaining modules [networkcfg] [openrcdmcryptcfg] [rawfs] with
code that throws on bad configuration. Replace with meaningful
error messages, to better check cases of SyntaxError &c.
- One might argue whether an empty list of partitions to mount is
a bad thing. It suggests that the partition module wasn't used,
and so we're in an OEM situation -- but then everything should
already be mounted anyway. That's why I choose empty -> bail.
- [initcpio] remove superfluous inner function
- [initcpio] catch errors from mkinitcpio itself and report them in a nice
readable format.
- Save translators the effort of doing a dozen messages
with just the name of the module changed. All of these modules
bail out on bad configurations with a meaningful message.
- [initcpiocfg]
- [fstab]
- [initramfscfg]
- [localecfg]
- [luksbootkeyfile]
- [luksopenswaphookcfg]
- [machineid] Warn on bad config. It's conceivable that this is run
with an empty rootMountPoint (i.e. "") to modify the running system,
so only bail on None.
- Load full text, toggle display
- Swap Up and Down arrow semantics on button, to match usual
text-editor display (down means it's expanded, displayed)
- If all of the licenses are optional, you should be able to
continue without accepting. Refactor to a single visible
slot to check the conditions.
- Always set the globalsettings value; to "false" on entry
to make sure it's there.
- When setting the list of entries, check the conditions
(because if the list is empty, or all of them are optional,
then it's ok to continue).
FIXES#1124FIXES#1125
- QVector is a better match with passing in QStringList, otherwise
you end up dealing with Qt's int indexes vs. std::vector's uint
indexes everywhere.
- Introduce find()
- Use only utils/YamlUtils.h to pull in yaml-cpp and supporting code.
- When compiling with clang, turn off warnings that the system header
for yaml-cpp would generate.
This commit adds several checks while reading the configuration of the
`partition` module, in case the partition layout configuration is
misformed. If an error is encountered, an message is printed to the
console and the module reverts to the default partition layout.
Checks are also added when implementing the partition layout, in case a
problem occurs that couldn't be anticipated (for example, when a
partition size is in %, checking its absolute value require knowing the
total device size, which is not the case when the configuration is
being read).
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
Every call of `ParseStringSize` is replaced by using an instance of the
`PartUtils::PartSize` class.
This commit also removes the now-unused previous size parsing functions.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
In order to maintain consistency, and make use, create a new PartSize
class in the PartUtils namespace, which inherits from NamedSuffix for
easier parsing and handling of size strings.
The switch to using this class instead of the previous functions will be
done in a follow-up commit.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
Fixed 'prepend' terminology in comments, restored old loop as it seems to work for me now in overwriting the file as one overall multiline instead of overwriting the file with each line at a time, code simplification based on review comments... credits-adding will come in the next commit.
This change fixes a few issues and adds a few improvements to the LightDM Autologin configuration process:
Fixes:
- Fixes malforming of configuration file out of the box, as without `[SeatDefaults]`, `[Seat:*]` or similar in the configuration file LightDM will break on a lot of distributions using LightDM
- Preserves the intended lightdm.conf file settings outside of `autologin-user` if the distribution has an /etc/lightdm/lightdm.conf file of its own
Misc. changes:
- Small spelling fix
- KDE neon ships a post-3.3.0 KPMCore, with deprecations, but not yet
the KPMCore 4 API, so add another API-version check to handle the
deprecations. Keeps warnings down.
- Don't use this if we don't need it (QObject::tr is static).
- C++14 allows (copy) binding to arbitrary expresstions in lambda's,
so detach from this.
When using a custom partition layout with partition sizes in %, it can
be useful to set an upper limit to the partition size.
For instance, using a 20% size for the `/` partition will create a 24G
partition on a 120GB drive, but a 200GB partition on a 1TB drive, which
is not useful, and could be avoided by setting a maximum partition size.
This commit adds the `maxSize` parameter (with a default value of 100%).
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- When the manual partitioning page exists, it reacts to
changes in a bunch of models; these models can be changed
repeatedly from the choice page.
- the manual partitioning page really only needs to deal with
the relevant selections at the moment it is instantiated.
- rsync reports its own progress, and reports on files that
find -type f doesn't. This meant that the numbers didn't
match what was stored in entry.total
- The ir-phase adds files to be handled; to-phase happens once
ir-phase is over and the remaining files are processed.
By adding the to-phase files, percentages over 100% were
reported (in part because the number of files doesn't match).
- Update expected entries total from rsync output.
- Re-jig computation of how done everything is: tally it
up in integers, and do only one global progress percentage.
- The mismatch between "ir-chk" and the comment "to-check" led me
to check (ha!) the output of rsync, and it outputs "to-chk"
during small transfers; make sure the comment reflects what
is actually being used to track progress (which is "ir-chk").
- After the BootLoader model is reset, if a bootloader location
has been selected before, try to find it in the (now-reset)
model to preserve the selection.
- clear() signals modelReset(), which is true, but inconvenient
when we do a bunch of changes afterwards. Block signals,
and rely on own signaling when all of the changes are done.
- Keep blocking signals while updating the model, since the row
appends otherwise trigger a change in the connected combo box.
- For unsafe installations (compile-time option), make sure
things fail before partitions are actually written, unless
the other option is also turned off.
- This is a compile-time choice, and off by default. This may be useful
for developers that need to get through installation to a different
partition on their root drive.
- Add an option to avoid actually doing unsafe things. This is an extra
safeguard; you need to turn on one and turn off the other option to
really be unsafe.
- This was commented out to combat the crash in device->type()
- I believe the crash was caused by double-deletion, which was
fixed in 2092ec3c9a by not re-
parenting an immutable copy of something.
- Restore the button-fix, since we need that to keep the
*create* button in-sync with the selected partition.
FIXES#1097
As requested, this commit adds a new configuration option to the
partition.conf file, name `efiSystemPartitionSize`.
When this option is absent, the default size of 300MiB will be used.
Fixes#1090
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
In order to allow the use of these functions across the whole partition
module and keep all partition size-related functions in the same
namespace, this commit moves them to PartUtils.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
This commit creates a _KiB operator for future use by the partition
module.
It also fixes a typo in one instance of MiBtoBytes(), requiring a couple
extra fixes.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
In order to be able to parse partition size strings using the same
functions across the partition module, the parseSizeString() function is
exported to the PartUtils namespace.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- Make some methods that are called mostly as slots, actual slots,
instead of going through extra lambdas.
- Use QOverload<>::of for disambiguation instead of homebrew casts.
- Provide a convenience method that names a Partition* with the
best human-readable name we can find (worst-case, spit out a
pointer representation which will at least help figure out
the identity of the Partition*).
- These methods are used for multi-page view-steps, which are rare.
For all the others, just drop the empty implementation and defer
to the base class.
- Next was enabled early; presumably to cover the case that no requirements
were checked and the requirements checker never emitted an update signal.
Drop that since the module manager is now responsible for doing that checking.
- Can't re-parent across threads easily
- If device is made by immutableDeviceCopy(), then it's still owned by the
PartitionCoreModule; giving it away to the widget is not a good idea.
- The NAM is being created from a method call on the GeneralRequirements
object in the requirements-checking thread, while the GR object itself
was created in a different thread. This cross-thread parenting
produces a warning, and we don't need the parent relationship here
anyway.
- If a distro provides an install-scenario that doesn't provide a DM,
(e.g. via netinstall) then that should be ok; if there **is** a DM
it should be configured.
FIXES: #1095
Due to changes to the FileSsytem::typeForName() function, more
processing is needed to deal with locales and different cases.
This is done by refactoring the findFS() function, initially located in
the PartitionViewStep class, and making it available to the whole module.
Additionnally, more checks have been implemented regarding the use of
global storage in the PartitionLayout class, and the filesystem types
now use the correct FileSystem::Type, as requested.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- The default window title in the designer file wasn't a good string
to translate. Use one of the titles instanced from elsewhere.
- The window titles set in subclasses were not translatable.
FIXES#1092
When using the default partition layout (only a `/` partition), the
filesystem used was ext4, ignoring the `defaultFileSystemType`
configuration option.
This commit fixes this bug, so that any supported filesystem can now be
used for the default partitioning scheme.
Fixes#1093
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- Strings like "{} the {} with {}" are terrible for translators:
- no context
- no possibility to re-order grammatical units
- substituting in English parts-of-speech is going to make a mess
- Write the strings out with explicitly named substitutions,
no part-of-speech substitution, and better formatting.
- Add a (superfluous, since they don't have their own signals or slots)
Q_OBJECT macro to the VG jobs, to silence a Transifex warning (this
does make sure that the tool knows about the context for the translated
messages)
- Part of the tests checks that the example configuration
is not empty. So uncomment the example. (Distro's should
not be installing the example configs, and this one in
particular needs to be customized).
- Store changes to the selected swap-choice when the combobox changes
- Use that member instead of dereferencing the combobox
This avoids nullptr crashes when the combobox isn't even created
(e.g. when there is only one swap choice).
- drop the localized comparisons; that's just confusing
- warn when no default FS is set (then use ext4)
- fix case-insensitive fallback; it used fsType, which was
set to Unknown in the for loop.
- Make the explanations about 3 times as wide as the text-boxes
that they are explaining. This is partly moot because the
text-boxes have fixed pixel sizes in the designer file, but keep
it flexible for now.
- Using the assignment-operator just generates blank lines.
- Using QLog with a log-level avoids the cDebug()-style special
handling of warnings and errors (useless here, but may as well
fix code style).
When using the `rawfs` module for copying data, it may be useful to
save the source device used for later checks or actions. This commit
therefore adds a `source` field to each corresponding partition entry in
global storage, so that this information can be retrieved later during
the installation process.
Another small improvement is that global storage is now modified only
once (it was previously modified as many times as there were entries
processed by the `rawfs` module).
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
When choosing `systemd-boot` as the bootloader, numerous problems
occurred:
- the kernel and initrd were not copied to the EFI System Partition,
and therefore could not be reached by the bootloader
- the fallback entry used the default initramfs image instead of the
fallback image
`systemd-boot` provides the `kernel-install` utility, which
automatically copies the kernel + initramfs to the EFI partition, and
creates the corresponding bootloader entry.
Unfortunately, `kernel-install` cannot be used here as the module is not
executed in a chroot. As setting up one only for running a single
command would be overkill, this patch re-creates what `kernel-install`
usually does:
- copy the kernel and initramfs to their own subdirectory at the root of
the EFI partition
- create the corresponding entry configuration file
To this end, the `systemd-boot` installation code in the `bootloader`
module has been largely refactored, including removing a few duplicate
LOCs.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
In its current state, the 'rawfs' module requires the source partition
to be identified either by its mount point or device name, but using a
symlink to either one (e.g '/dev/disk/by-uuid/...') would fail.
This patch fetches the real path of source partition, allowing the use
of symlinks.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
When setting the size of a partition without indicating the unit, two
problems occur:
- the size is parsed as an integer, not as a string, hence the
configuration parsing fails
- the size parser doesn't recognize the fact that the size has no units
and defaults to 100%
This patch fixes the configuration parsing as well as the size string
parsing.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
When a partition doesn't have a minimum size in the partition layout
configuration, it defaults to using 100% of the available space.
This patch fixes this error by setting the minimum partition size to 0
when the attribute has been omitted.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- Add global- and job-configurations for test runs.
- Add a driver script that sets up some assumptions on the host
system so that the tests can complete.
- The idea is that these tests together get a decent code-coverage
for the module.
- This is a driver script for running testmodule.py multiple times
with different global- and job-configurations.
- Usage: testpythonrun.sh <modulename>
- Run the script from the build-directory. It uses files from the
tests/ (source) subdirectory to drive the test runs.
- Only need to get the list of supported filesystems *once*,
not for each and every filesystem that is going to be unpacked.
- Be more Python-idiomatic.
In some cases, we might want to copy a filesystem as if we were using a
simple 'dd' command, in order to create an exact copy, down to the block
level.
This can be useful in particular when working with dm-verity for
checking the rootfs integrity: that way, we can make a direct copy of
the rootfs and its verity partition and keep the system usable.
This patch adds a new 'rawfs' module to calamares, making possible to
block-copy a filesystem to a block device.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- Use locale "C" for checking filesystem names
- Also check other possibilities and case-insensitive, to
be more forgiving of weird configurations (and localizations)
- The language and BCP need to be in-sync
- Existing code was inconsistent in setting things, which is why
you could get through the locale page without setting a locale
(at all) or it would keep English in spite of picking Germand on
the welcome page.
- Patch tests to use that API.
Due to a computation error when calculating the total drive space and
each partition's last sector, the last partition's last sector was out
of boundaries, leading to an error creating this partition.
This patch fixes the computation algorithm to get rid of this error.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- Allow split-setting of the language and formats
- Test new constructors
- Since fromLanguageAndLocation can handle empty localeGen
lists just fine, skip all the weird checks that return
invalid guessed locale configurations.
- Replace createDefault() with a constructor that takes a
locale name; use it with en_US.UTF-8 in those places where
createDefault was previously used.
Some code was copied and adapted from PartitionActions.cpp. For full
compliance, it is best to copy the copyright holders list from this file
to PartitionLayout.cpp.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
When chosing the "Replace partition" option, free space is not handled
like any partition. In order to apply the custom partition layout in
that case too, we have to modify the code where "replace free space" is
handled.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
In some cases where a custom partition layout is used, use of this
layout is mandatory (this can be the case when using a read-only rootfs
which is updated by block-cpying an image file to it).
For these cases, the user must not be able to change the partition
layout, therefore we have to disable manual partitioning.
In order to stay consistent with current behaviour, manual partitioning
is still enabled by default. It will only be disabled if the partition
module's config file contains the corresponding option set to "false".
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
When choosing "Install alongside another system", the custom partition
layout is applied to the space freed by resizing the selected partition.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
This patches add new methods to both PartitionLayout and
PartitionCoreModule classes which apply the partition layout to the
available drive space.
In addition, the partition creation code from PartitioinActions is
removed to call the newly created methods instead, thus applying the
custom partition layout when the "Erase whole disk" and "Replace
partition" choices are selected.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
As we move some of the partition creation code away from
PartitionActions, we will need the bytesToSectors function. Rather than
copying it, we export it in the PartitionActions namespace, so that
other classes can use it.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
In order to keep the partition layout during calamares' execution, we
add a PartitionLayout object instance to PartitionCoreModule. This class
will therefore be used to initialize the PartitionLayout object and
interact with it thoughout the program's execution.
When no partition layout is present in the config file, we initialize
the layout with a single ext4 partition mounted on '/', as it was
previously done.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
In order to use a custom partition layout in the partition module, we
need to write this layout in the module's config file, and store it into
a dedicated object.
As it doesn't look appropriate to extend an existing class with layout
information, we create a new PartitionLayout class, which will be used
to parse the layout from the config file and keep it in memory.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
This basically means we talk about localization in the respective
localized variant. e.g. "German (Germany)" ➡ "Deutsch (Deutschland)".
If geoip lookup failed or isn't configured for whatever reason it's a
stretch to expect the user to know english enough to find their own
language. Preferring the localized strings resolves this issue.
Additionally this happens to bypass #712 respectively
https://bugreports.qt.io/browse/QTBUG-34287
as the native names are properly spelled. So, as long as Qt has localized
names the names will also be properly spelled.
- If a key is missing from mount.conf, don't raise KeyError
- If both keys are missing, suggest that mount.conf might
be missing instead (a consequence of INSTALL_CONFIG=OFF, for
instance).
- Simplify code a bit.
- Don't bother returning None explicitly.
- for remove and localInstall, add support for pre- and post-
scripts like there already was for install.
This feels like there's code duplication going on, but I haven't
thought of an elegant way to distinguish the available operations
so that I can pass around functions instead.
- The noload option prevents journal re-play (so it's an extra-
strong read-only) but is only applicable to ext3 and ext4.
Check the FS type before mounting; other FS types don't
accept -o noload and will fail to mount.
- This only applies to legacy (non-EFI) BIOS systems, and
adds the FlagBoot to whatever is already set for the root
filesystem, and only when autopartitioning the device.
Submitted by aliveafter1000.
FIXES: #1046CLOSES: #1049
Suggested by aliveafter1000: having a default value, and then
filling in the default in one place it is used and not others,
is weird. Instead of dropping the one use, remove the default
value: partition flags are important enough to be explicit.
- Handle legacy and modern config, mixed-configs,
- Translate strings to enum values,
- Default and warn as appropriate.
- Doesn't **do** anything with the config, though.
- Running lsblk and mount for debugging purposes can be
skipped when the debugging is going to be suppressed anyway.
This will speed things up just a little for regular users.
- While winnowing devices, the zram and nullptr cases
were mixed together; split them, for the sake of
logging more accurately.
- While here, fix up some coding-style issues.
- Distinguish size and atleast; in percentages they mean different
things (atleast is a bit weird as a percentage, but hey).
- Fix bug in percentage calculation.
- Avoid percentage above 100.
- Add documentation in config-file.
- Q_ASSERT doesn't work in constexpr functions because it's not
- May as well calculate bytes at compile-time, no need to give
the runaround via number-of-MiB
- Test the parsing of configuration values, that it gets the right
values when given correct(-ish) configurations. Doesn't test
any situations with bad config.
- Make RelativeSize public so we can use it in non-member functions
- Make a template out of matching the string suffixes; this is
safer because the length of the suffix can be computed at compile-time
(+1 for the trailing NUL) rather than writing it out in boilerplate.
- Non-functional as yet, WIP
- Intended for use on SBC images and similar, which need to resize
the root partition to the size of the SD card they're on.
FIXES#921
- The choice of swap needs to be handled in more places,
so make the enum available in the partition module core instead
of just inside the choice page.
- Q_ASSERT doesn't work in constexpr functions because it's not
- May as well calculate bytes at compile-time, no need to give
the runaround via number-of-MiB
This fixes the crash by calling the model-reset first, then
refreshing. Previously, the destructors that do the work
were still being called in the wrong order.
FIXES#1019
- The ResetHelper only finalized changes to the module on
destruction, but calls to refresh() assumed it was already
done. This leads to crashes when refresh() uses an intermediate
state of the model.
Introduce extra helpers, and rename refresh() to avoid calling the
old implementation from any code. The new helper just creates and
destroys a ResetHelper, before creating and destroying an object
that calls the new refreshAfterModelChange().
FIXES#1019
ESP == boot. at best this is duplicated information, at worst kpmcore may
implode if you try to set a boot flag since that is technically an MBR
type flag and means nothing within the context of GPT where ESP is the flag
to set.
having ESP as active flag AND then trying to set ESP means nothing is
set since kpmcore will think ESP is already set (it is listed as active
after all). this ultimately meant that nothing was set since there was
no delta between the requested flags and the already active flags.
- The parameter list was wrong, and could never have worked
(apparently noone mixes LightDM with basic setup, even though
it's one of the few with basic_setup()).
- Shorten some lines.
- Move variable initializations closer to where they are first used
- Also warn if no implementations are available
- Handle sysconfig as any other DM; there's no real reason
to treat it specially.
- Add (empty) implementations of all the abstract methods that
are not needed (or supported) by various DMs.
- Order the abstract methods by calling order
- Fix up have_dm
- Make root_mount_point a parameter of the DM classes
- Create instances as needed, then check if they're available
- Keep instances that actually need to be configured
- Clean up remaining cruft from removing old setup_autologin()
with all the string-comparison magic.
- Some DMs allow setting up the default DE. Factor that
out into a method like basic_setup() and setup_autologin()
and add it to the configuration chain.
- For all the DMs that have some kind of basic setup,
introduce a method in the DM class for doing just that.
- The Python code now doesn't call basic setup anywhere,
so this specific revision isn't going to work properly.
- Split the entire body of setup_autologin() to a method
per implementation class.
- Make the check for presence of a DM a class-method,
since if it fails, instantiation is going to be useless.
If displaymanagers is not set in the job config or globally,
it wasn't set at all, leading to a runtime error. Set to
None, so that the regular error message will be triggered.
FIXES#1002
- Reported by Bill Auger (I think), a 15GiB disk wouldn't hold
a 8.9GiB root plus 4GiB swap -- due to 10% overprovisioning
of swap, plus the 2.1GiB fudge factor.
- Calculating first free sector had an off-by-one so that
partitioning would start at 2049.
- EFI boot partition grew 1 sector larger than desired.
- While here, align everything to 1MiB boundaries as well.
FIXES#1008
- Make *mandatory* optional (and default to false); this
allows shorter lists of entries
- Allow degenerate entries which are just a name
(which have *mandatory* set to false as well).
SEE #992
- Allow just a name entry, instead of requiring an object
entry; this makes "foo" equal to { name: "foo", runlevel: "default" }
and simplifies more for the straightfoward case of #974.
- Based on comments from #974, follow the configuration
scheme from services-systemd, so with separate lists
"services" and "disable". This ties it **slightly**
less closely to the commands passed to rc-config.
- If runlevel isn't set (at all) then use "default". For
most systems that do not use multiple runlevels, this
simplifies the configuration to just a list of service names
to add or delete.
- Document the functions some more
- Only "state" (i.e. action) "add" and "del" make sense,
avoid calling rc-update for other keys (e.g. typo's).
This matches the documentation, although there might be
other actions that make sense (see also services-systemd,
with its enable, disable and mask actions).
- With refactored code, introducing new kinds of actions
is very few lines of code. Allow disabling targets
(services was already possible). Allow masking units,
but as a special case require the complete name.
FIXES#975
- The three steps of modifying services in the target
system do basically the same thing, so factor out
the loops and logging into a systemctl() function.
- Log to warning() instead of just debugging, on failure.
If USE_<foo> is given a value that doesn't match **anything**,
then bail out. Since USE_* is an explicit distro choice for a
specific implementation, it's an error if that implementation
is not there.
When there are multiple modules doing a thing and it really only
makes sense to have one of them in a given Calamares compilation,
the USE_<foo> variables allow you to select one, while ignoring
all the other implementations. If USE_<foo> is not set, all
implementations are included (as usual).
- A valid line (as explained in the comments at the top of
the locale.gen file) is <locale> <encoding> (two fields),
so lines with more than two fields can't be valid locale-
listing lines. For them, pretend they name locale "",
which won't be matched.
- Improved debug-logging
- Fix the actual problem of listing locales more than once,
by listing them all, uniqified, at the end, with an explanitory
comment in the generated file.
- Be more accepting of what constitutes a locale-line; this allows
spaces before and after the `#` comment sign, but because we're
uniquifying, this doesn't cause duplicates.
- Because we write the enabled locales at the end, the full file
comment-header is retained un-mangled (instead of accidentally
enabling a locale mentioned as an example there).
Testing for existence of a file in the live system, and then
copying it in the target system, is not a recipe for success.
- Fix the restore-from-backup part.
- Document that your live and target system must both have
/etc/locale.gen if you want this to work at all.
Also make install for yum and dnf follow the documented syntax: options
(-y) before the command (install), even though yum and dnf also accept
the other order. This also makes it consistent with remove.
Untangle the shortcuts; Create and Cancel had an overlap.
Skip 'r' (Revert all changes) and 'e' (Edit) and settle on
'a' (which might also mean "Add").
FIXES#977
Introduce the notion of emergency modules and emergency jobs.
Initial use will probably center around the preservefiles module,
and possibly umount.
FIXES#928
A potentially emergency module is one that has EMERGENCY
(in CMake) or emergency: true (in module.desc) set.
Any such module must also set emergency: true in the
configuration of the module. This is to allow for
instances of a module that **don't** run as emergency
modules, alongside actual emergency ones.
