| Age | Commit message (Collapse) | Author |
|---|
|
This allows an image to skip the creation of kernel depmod
data. It is useful for creating an image that will run as a
container image inside a host with no knowledge of the parent's
kernel.
Signed-off-by: Dan McGregor <dan.mcgregor@usask.ca>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
Since the rewrite of the image construction code in python a few
releases ago, we remove a couple of packages from the image as one of
the final steps when constructing the image (notably update-rc.d and
run-postinsts). However, because of the order of operations, these
packages are still listed both in the buildhistory
installed_package*.txt files and in the manifest file created next to
the image, which is wrong.
There were two possible solutions to this: (1) change the order such
that the uninstallation occurs before calling ROOTFS_POSTPROCESS_COMMAND
or (2) add another hook variable in such that we can have the
package list collection code run at the right time. Because it's
currently possible (but very much not recommended) to install additional
packages within ROOTFS_POSTPROCESS_COMMAND, which may have postinstall
scripts and thus require the packages we would otherwise uninstall if we
were to take option 1, option 2 is really the least likely to cause
problems. Therefore, add ROOTFS_POSTUNINSTALL_COMMAND and make the image
and buildhistory classes use it.
Fixes [YOCTO #6479].
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
|
|
If you add an item to EXTRA_IMAGE_FEATURES in your local.conf that is
not supported by image.bbclass itself (such as "tools-sdk" which is
implemented in core-image.bbclass), it can be somewhat annoying to have
the parse fall over if you have a recipe that inherits image.bbclass
only. Change the error from bb.fatal to skip the recipe instead so that
you only see the error when attempting to build the recipe, plus add a
bit of logic to report if the feature is coming in via
EXTRA_IMAGE_FEATURES.
Fixes [YOCTO #5023].
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
|
|
In the daisy (1.6) timeframe, when we rewrote the image construction in
Python, we neglected to reimplement the support for the little used and
undocumented variable MACHINE_POSTPROCESS_COMMAND, and apparently nobody
noticed. We have a better method for implementing machine-specific image
formats that is in wider use (i.e. add a custom class which implements
the new image type, add the class to IMAGE_CLASSES and the type to
IMAGE_FSTYPES), and we now also have wic. Thus it makes more sense to
just call this variable unsupported now and drop the sole remaining
reference to it.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
|
|
We need the depmod data so that the kernel depmod command works successfully
at rootfs time. The fact this was working inconsistently is now highlighted
after the command was made to error out. A simple test case is:
bitbake virtual/kernel image
bitbake vrituak/kernel -c clean
bitbake image -c rootfs -f
We fix it by adding the missing dependency, the data is in PKGDATA_DIR and
hence we use packagedata.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
With the rpm package backend enabled, running:
bitbake <image>
bitbake virtual/kernel -c clean
bitbake <image> -c rootfs -f
results in an image with incorrect kernel module dependency information.
The problem is that the System.map and kernel-abiversion files are needed
for depmod and after the recent kernel changes, these are no longer in
sstate.
Its reasonable to require the kernel to unpack/build if you're
about to build a module against it. It is not reasonable to require this
just to build a rootfs.
Therefore stash the needed files specifically for depmod.
Also fix some STAGING_KERNEL_DIR references which were incorrect, found
whilst sorting through his change.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The set_image_autologin function is GPE Login specific and the package
is not available in OE-Core so the function should be added in the
meta-gpe layer, if necessary. Drop this from the OE-Core as it is
unused.
Signed-off-by: Otavio Salvador <otavio@ossystems.com.br>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
New version of systemd implements a new feature of updating /etc
or /var when needed at boot. For details, please see link below.
Opointer.de/blog/projects/stateless.html
For now, at boot time, the systemd-sysusers.service would update user
database files (/etc/passwd, /etc/group, etc.) according to the configuration
files under /usr/lib/sysusers.d. This step is necessary for other systemd
services to work correctly. Examples of such services are systemd-resolved
and systemd-tmpfiles-setup.
The problem is that on a read-only file system, that is, if /etc is read-only,
the user database files could not be updated, causing failures of services.
This patch fixes this problem by adding users/groups at rootfs time.
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
There is a race over the do_package_qa task and the do_rootfs task
since rootfs recreates a directory. This patch disables the task
(which isn't used for images) to avoid the race:
NOTE: recipe core-image-minimal-1.0-r0: task do_package_qa: Started
NOTE: recipe core-image-minimal-1.0-r0: task do_rootfs: Started
ERROR: Build of do_package_qa failed
ERROR: Traceback (most recent call last):
File "/home/pokybuild/yocto-autobuilder/yocto-worker/nightly-mips/build/bitbake/lib/bb/build.py", line 497, in exec_task
return _exec_task(fn, task, d, quieterr)
File "/home/pokybuild/yocto-autobuilder/yocto-worker/nightly-mips/build/bitbake/lib/bb/build.py", line 440, in _exec_task
exec_func(func, localdata)
File "/home/pokybuild/yocto-autobuilder/yocto-worker/nightly-mips/build/bitbake/lib/bb/build.py", line 212, in exec_func
exec_func_python(func, d, runfile, cwd=adir)
File "/home/pokybuild/yocto-autobuilder/yocto-worker/nightly-mips/build/bitbake/lib/bb/build.py", line 237, in exec_func_python
os.chdir(cwd)
OSError: [Errno 2] No such file or directory: '/home/pokybuild/yocto-autobuilder/yocto-worker/nightly-mips/build/build/tmp/work/qemumips-poky-linux/core-image-minimal/1.0-r0/core-image-minimal-1.0'
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Don't mix tabs and spaces
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
Should have been removed with
92f956d tweak read_only_rootfs_hook to also support systemd based systems
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
When pseudo is configured to disallow fallback to the build host
/etc/hosts and /etc/group, the selection of ${IMAGE_ROOT} for
PSEUDO_PASSWD is insufficient as the necessary files will not be
available until base-passwd has been installed and its pkg_postinst
script run. Fall back to the ${STAGING_DIR_NATIVE} version of those
files until the rootfs versions are available. (The native copies are
never modified by the build; the ones in ${STAGING_DIR_TARGET} are
updated and may contain settings not consistent with what would be
created by post-install useradd/groupadd commands invoked in the image
rootfs.
Signed-off-by: Peter A. Bigot <pab@pabigot.com>
|
|
In oe-core commit 03c5f39b4d7dd8c81e0a130b7d5884e5af039a24,
it removed obsolete codes about variable MULTILIB_VENDORS.
We clean up the rest obsolete codes related with
MULTILIB_VENDORS
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: Ross Burton <ross.burton@intel.com>
|
|
The image_types.bbclass is a must since we use this in image.bbclass:
[snip]
python () {
deps = " " + imagetypes_getdepends(d)
d.appendVarFlag('do_rootfs', 'depends', deps)
[snip]
The imagetypes_getdepends() is defined in image_types.bbclass. Use "+="
to replace "?=" since it is a must, so that the user can use
"IMAGE_CLASSES = foo.bbclass" in local.conf to add their own image
class.
NOTE: the IMAGE_CLASSES_append = " foo" doesn't work since we use this
in image.bbclass:
IMAGE_CLASSES += "image_types"
inherit ${IMAGE_CLASSES}
I think that it is because inherit takes effect before append?
Another way to fix the problem is:
IMAGE_CLASSES ?= ""
inherit image_types ${IMAGE_CLASSES}
But it seems that we need another name for IMAGE_CLASSES, for example
IMAGE_CLASSES_EXTRA, and also need update the doc, which would make it
complicated.
Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Modify the read_only_rootfs_hook function to make it also have effect
on systemd based systems.
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Added base variables and package backend specific variables to
rootfs[vardeps] in order for rootfs to rebuild when changes are made.
Set some variables as [func] to inform bitbake that they are shell
scripts, so that it invokes its shell dependency parsing. Without
marking them as functions, changes in the actual function body would
not trigger rootfs rebuilds.
[YOCTO #6502]
Signed-off-by: Roxana Ciobanu <roxana.ciobanu@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Currently, /etc/timestamp is saved with minutes accuracy. To increase
the accuracy, modify the save-rtc.sh and bootmisc.sh scripts to save and
read /etc/timestamp respectively with seconds accuracy.
Signed-off-by: Richard Tollerton <rich.tollerton@ni.com>
Signed-off-by: Ben Shelton <ben.shelton@ni.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
it does not grok glibc ldconfig format
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The base_contains is kept as a compatibility method and we ought to
not use it in OE-Core so we can remove it from base metadata in
future.
Signed-off-by: Otavio Salvador <otavio@ossystems.com.br>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The sed expression was also replacing documentation text containing
PermitRootLogin in the line so "PermitRootLogin yes" was specified
twice.
Signed-off-by: Jonathan Liu <net147@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
I'm sick of seeing people adding to EXPORT_FUNCTIONS in these classes
when they clearly have no idea what it does.
Worse, these uses of it are all broken, the naming is incorrect and
they do nothing. Lets remove them and try and preserve any remaining
part of my sanity.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This function disables the reverse DNS lookup on QEMU targets to reduce the
delay when using static IP address. By disabling DNS lookup we can save a great
deal of time during automated testing on the autobuilder (on the order of ~400
seconds per ssh tranaction). This is seen when using the testimage, there is a
delay getting logged-in from the server to target.
It's enabled for all qemu imgaes by default and can be overridden by setting
the SSH_DISABLE_DNS_LOOKUP variable.
[YOCTO #5954]
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Replace 'rootfs' with '/dev/root' in read_only_rootfs_hook function
to match the latest change in fstab file from the base-files recipe.
The related commit is as follows.
commit e8bc7a136a81a0d8df2d32dfba0920c1b2835141
base-files: use /dev/root in /etc/fstab for systemd support
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Since we now run depmod when building images (as the postinst that does
this is now on kernel-base instead of kernel-image) it is possible to
have module file differences between the two halves of the multilib image,
and the code that checks for such differences detects this and fails.
Whitelist this file to avoid the failure.
Specifically, modules.alias, modules.dep and modules.symbol can differ
along with their .bin counterparts.
Related to fix for [YOCTO #5392].
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
If you want passwordless logins to work with pam enabled, then you can't
have "nullok_secure" enabled on pam_unix entries. Add some
postprocessing to change these to "nullok" when debug-tweaks is in
IMAGE_FEATURES, in order to make passwordless logins with PAM work
again.
Fixes [YOCTO #5973].
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The old wrapper got renamed to image_list_installed_packages().
Signed-off-by: Laurentiu Palcu <laurentiu.palcu@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Since tasks were renamed to packagegroups some time ago, this variable
name implies that its usage is necessarily related to them which is not
the case. Rename the variable to more closely represent what it does
(whilst still providing backwards-compatibility with a warning for
PACKAGE_GROUP).
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Previously the "package-management" feature was only fully supported
when inheriting core-image.bbclass, which is not really ideal given that
this is the standard way of adding runtime packaging to an image in
OE-Core.
Part of the fix for [YOCTO #5424].
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Changes to BAD_RECOMMENDATIONS or NO_RECOMMENDATIONS require the rootfs
for the image to be rebuilt.
[YOCTO #5898]
Signed-off-by: Jonathan Liu <net147@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This commit whitelists some common directories, so the multilib sanity
checks pass and also fixes an issue in lib/oe/rootfs.py when the
compared files do not exist.
Signed-off-by: Laurentiu Palcu <laurentiu.palcu@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
install definition
The sdk complementary install operation was defined in image.bbclass,
but the sdk recipe (such as meta-toolchain.bb) didn't inherit this
bbclass but populate_sdk, and both of image and populate_sdk bbclass
inherited populate_sdk_base bbclass, so move the sdk complementary
install definition to populate_sdk_base bbclass fixed this issue.
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: Laurentiu Palcu <laurentiu.palcu@intel.com>
|
|
Since the list_installed_packages() function has refactored in
python, do the necessary adjustments to license_create_manifes-
t() and write_image_manifest() in license.bbclass and image.bb-
class respectively;
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: Laurentiu Palcu <laurentiu.palcu@intel.com>
|
|
Since rootfs_install_complementary() is now implemented in python for
each backend, remove it from POPULATE_SDK_POST_TARGET_COMMAND. Call it
directly in python.
Signed-off-by: Laurentiu Palcu <laurentiu.palcu@intel.com>
|
|
Remove the directory, manually, in the Rootfs.create() function.
Signed-off-by: Laurentiu Palcu <laurentiu.palcu@intel.com>
|
|
This commit cleans up the functions that were ported to python.
Signed-off-by: Laurentiu Palcu <laurentiu.palcu@intel.com>
|
|
This commit will:
* remove old bash code common to all backends;
* create a new do_rootfs() python function that will use the new
rootfs/image creation routines;
* allow creation of dpkg based images;
* fail for rpm/opkg (not implemented yet);
Signed-off-by: Laurentiu Palcu <laurentiu.palcu@intel.com>
|
|
read-only-rootfs
If the rootfs is read-only then we aren't going to be updating any
alternatives or modifying the password file and these binaries will
be redundant. In an ideal world we would be able to stop them from
being installed in the first place but this is non-trivial to arrange.
As a workaround in the meantime, let's just uninstall them once image
construction is finished.
Signed-off-by: Phil Blundell <pb@pbcl.net>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The run-postinsts script has been moved to ${sbindir}.
[YOCTO #5719]
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Add ability to set the default target for systemd images.
The default target for system is controlled by SYSTEMD_DEFAULT_TARGET.
The default value for this variable is derived from checking whether
IMAGE_FEATURES contains 'x11-base' or not. Each image could override
this value in its own recipe. For now, we don't need to do any change,
because all images that support graphical environment has 'x11-base'
in its IMAGE_FEATURES.
[YOCTO #3816]
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Previously, this function replaces the root password with '*' if
'debug-tweaks' is not in IMAGE_FEATURES. It not only zaps empty root
password, but also zaps non-empty root password. That means, if the
user uses a bbappend file for base-passwd to set the root password, he
would not be able to login as root; if the user uses 'EXTRA_USERS_PARAMS'
to set the root password, he would still not be able to login as root.
What we really want from this function is to disallow empty root password
if 'debug-tweaks' is not in IMAGE_FEATURES. This function should not remove
non-empty root password because that password is usually deliberately set
by the user.
This patch renames zap_root_password to zap_empty_root_password to
better reflect the intent of this function. It also modifies the code
to make this function work correctly.
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Default USE_DEVFS to "1" so that the `makedevs' command is not run
at rootfs time by default. There are mainly two reasons to do so.
1. This will fix a build failure with initramfs-kexecboot-klibc-image.
"makedevs: No entry for root in search list "
2. Most of our images use a filesystem over /dev. Most of the time, it's
just devtmpfs. So we actually are using a filesystem over /dev.
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
After 1b8e4abd2d9c0 [bitbake.conf/package: Collapse PKGDATA_DIR into
a single machine specific directory], oe-pkgdata-util does not use
target_suffix parameter, so do not need to loop the vendor
Signed-off-by: Roy Li <rongqing.li@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
|
|
Now that none of the packagegroups depend on virtual/kernel, we have the problem
that MACHINE=qemumips bitbake core-image-minimal doesn't put a kernel
into the deploy directory. This breaks many common usecases and
user expectations.
To avoid this, add a dependency on the kernel deploy to image do_build tasks.
This should avoid any circular dependency issues but equally ensure users
have their expectations met.
[YOCTO #5581]
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Write a list of installed packages to a .manifest file next to the
image, so we can find out what went into the image after it has been
constructed without necessarily having to have buildhistory enabled
(although that will provide more detail.) We can make use of this for
example in the testimage class associated code that checks for installed
packages for determining whether or not to run specific tests.
Note: this replaces the previous ipk-specific manifest code with
something that works for ipk, rpm and deb, and instead of a pruned
status file, packages are listed one per line, in the following format:
<packagename> <packagearch> <version>
Tests for all three backends have shown that the performance impact of
this change is negligible (about 1.5s max).
Implements [YOCTO #5410]
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
We don't actually need this dummy class; "inherit" can be used with an
expression that evaluates to nothing with current BitBake.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Changes to split live into iso and hddimg without
adding a new image type class.
This patch has only a visible effect on HOB and solves
part 2 of #3197
[YOCTO #3197]
Signed-off-by: Valentin Popa <valentin.popa@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Remove a package from rootfs only if it's already installed. Also,
if a package is uninstalled, remove it from installed_pkgs.txt.
[YOCTO #5169]
Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
Signed-off-by: Saul Wold <sgw@linux.intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Setting TEST_IMAGE = "1" alone will now automatically run tests on the
image immediately after the image is built instead of having to add
INHERIT += "testimage" and run bitbake -c testimage <image> manually
(but that will still work). This restores functionality that was
present in the older imagetest-qemu class with IMAGETEST.
Signed-off-by: Paul Eggleton <paul.eggleton@linux.intel.com>
|
|
directory
Currently we have a hierarchy of pkgdata directories and the code has to put together
a search path and look through each in turn until it finds the data it needs.
This has lead to a number of hardcoded paths and file globing which
is unpredictable and undesirable. Worse, certain tricks that should be
easy like a GL specific package architecture become problematic with the
curretn search paths.
With the modern sstate code, we can do better and construct a single pkgdata
directory for each machine in just the same way as we do for the sysroot. This
is already tried and well tested. With such a single directory, all the code that
iterated through multiple pkgdata directories and simply be removed and give
a significant simplification of the code. Even existing build directories adapt
to the change well since the package contents doesn't change, just the location
they're installed to and the stamp for them.
The only complication is the we need a different shlibs directory for each
multilib. These are only used by package.bbclass and the simple fix is to
add MLPREFIX to the shlib directory name. This means the multilib packages will
repackage and the sstate checksum will change but an existing build directory
will adapt to the changes safely.
It is close to release however I believe the benefits this patch give us
are worth consideration for inclusion and give us more options for dealing
with problems like the GL one. It also sets the ground work well for
shlibs improvements in 1.6.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
In the target sysroot of an SDK we can have target system absolute links
which don't make sense. This adds a script which fixes them up to become relative
paths instead.
[YOCTO #5020]
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
Dan McGregor
Paul Eggleton
Richard Purdie
Otavio Salvador
Chen Qi
Michael Jones
Peter A. Bigot
Hongxu Jia
Robert Yang
Roxana Ciobanu
Blair Elliott
Khem Raj
Jonathan Liu
Saul Wold
Laurentiu Palcu
Phil Blundell
Ross Burton
Roy Li
Valentin Popa