| Commit message (Collapse) | Author | Age |
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
|
| |
When SBCL fails to start up, such as when previously-loaded shared libraries
cannot be found while trying to reinvoke a dumped image, it exits 1. We must
avoid erroneously interpreting this as a successful attempt to make changes.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
|
| |
It's the fact we're changing to a different home directory that's part of the
same root filesystem, and with the same UID/GID ranges, that means we call
UPLOAD-ALL-PREREQUISITE-DATA with this connection object.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
| |
In particular, if the secondary propapps are applied but make no change, do
not return :NO-CHANGE as the overall result.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This change was prompted by the realisation that RECONNECTS nullifies any
enclosing SERVICE:WITHOUT-STARTING-SERVICES because *HOST* during RECONNECTS
is a distinct HOST value, computed at hostattrs time, which can't be affected
by an enclosing combinator that works by temporarily pushing hostattrs at
:APPLY time. So, for example, if we chroot and subsequently establish a
:SETUID connection, properties might try to start services as the non-root
user and there is no way for them to know they shouldn't.
We can avoid this problem by using a connattr instead because we have a
mechanism for propagating connattrs to subsequently established connections.
And as in most cases the reason for not starting services is solely due to the
way in which we're connecting to the host, a connattr is more appropriate.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
| |
This reverts and replaces commit b4eb20904aa8f4ea7ec0ef58eac7889b4b864277.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
|
| |
We've long required package-local nicknames but this was not mentioned. We
now additionally require multithreading, a C toolchain and development headers
for libacl.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
| |
Previously we waited for waitpid(2) to return before reading from the pipes.
If either pipe filled up before the child process exited, we would deadlock.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
| |
This simplifies the API.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
| |
This prevents multiple attempts to upload the same item of prerequisite data.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
|
|
| |
This avoids some cases of interference between the deployment of the chroot
and other system activity. For example, before this change, the
:ALWAYS-DEPLOYS option to LIBVIRT:KVM-BOOTS-CHROOT{,-FOR} could interact with
Virtiofs in such a way as to break the running VM.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
|
|
|
| |
CONNECTION-CONNATTR for :REMOTE-HOME avoids relying on the HOME environment
variable having the correct value; its docstring describes a case in which
HOME's value can be incorrect. As determining :XDG-CACHE-HOME depends in most
cases on having the correct value for the remote home directory, don't rely on
the HOME environment variable here either.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
| |
In particular, reloads initiated by the call to ASDF:LOAD-SYSTEM in
CONSFIGURATOR.DATA.ASDF::SYSTEM-VERSION-FILES.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
| |
POSIX lacks 'mktemp -d' but it does have mkdtemp(3).
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
| |
Per the docstring of EVAL-IN-GRANDCHILD.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
With this change we handle the situation in which we don't know the OS of a
host on which we need to install system packages more cleanly than before.
Also rely on the new PACKAGE:INSTALLED to install the OS bootstrapper in
INSTALLER:CLEANLY-INSTALLED-ONCE. This allows us to simplify usage of the
property by replacing the ORIGINAL-OS argument with ORIGINAL-OS-ARCHITECTURE.
Making this change does mean that we now have two ways to specify the
different names a package has on different OSs: (i) something like
OS:ETYPECASE where each branch applies a property which invokes an OS-specific
package manager; and (ii) the plists supplied to PACKAGE:INSTALLED.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We establish a SKIP-PROPERTY restart for each sequencing combinator, such that
in addition to skipping over individual property applications, it is possible
to abandon the whole sequence. However, that restart discards information
about whether or not a change was made by the property applications of the
abandoned sequence prior to the property application interrupted by an error.
The new SKIP-SEQUENCE restart preserves this information by returning from the
DOLIST in APPLY-AND-PRINT.
The implementation of the SKIP-SEQUENCE restart must take into account the
fact that the property application interrupted by the error might have made a
change prior to failing. In particular, the new restart must not cause the
sequencing combinator to return :NO-CHANGE unless it can infer that the
property application interrupted by an error made no change. To achieve this,
capture whether the interrupted property application made a change by
introducing a distinction between plain FAILED-CHANGE and a new condition
class, ABORTED-CHANGE.
These changes permit the implementation of a new combinator, ESEQPROPS-UNTIL,
which invokes SKIP-SEQUENCE when a given condition is signalled. The new
combinator is like SEQPROPS in that it allows for continuing the deployment
despite a signalling of FAILED-CHANGE, but it is like ESEQPROPS in not
attempting to apply succeeding propapps. It also offers finer-grained control
over what kinds of failures are to be tolerated than does SEQPROPS. When the
condition is ABORTED-CHANGE or a subclass, ESEQPROPS-UNTIL returns information
about whether or not a change was made by the property applications of the
abandoned sequence prior to the property application that was interrupted,
enabling useful combinations with ON-CHANGE.
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|
|
|
|
| |
Signed-off-by: Sean Whitton <spwhitton@spwhitton.name>
|