| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
| |
There's no code left in these files that I want to move anywhere, so the
remaining parts can go.
|
| | |
|
| | |
|
| |
|
|
|
| |
These decisions will now be made in anaconda proper. There's no need to
involve three separate processes in the discussion.
|
| | |
|
| |
|
|
|
|
| |
This gets rid of all the argument parsing code in init, loader support for
reading args in some sort of test mode, and hopefully the last of the bogl
support.
|
| | |
|
| |
|
|
|
|
|
| |
* We don't need to populate /dev.
* We don't need to mount filesystems.
* We don't need to start dbus or udevd.
* We don't need to start a shell in multiple places.
|
| |
|
|
|
|
|
|
| |
Previously killall5 omitted the init's session (which includes the anaconda
process group) and anaconda and everything spawned from there (including
the vnc server) survived both the killall5 calls.
Related: rhbz#679397
|
| |
|
|
|
|
|
| |
Change loader/init.c so it doesn't bother checking the return value
of mount(). Instead it just checks to see if the mount functions as
expected - for example, after mounting /proc it checks to see if
/proc/cmdline exists.
|
| |
|
|
|
|
|
| |
As long as we can write to a few places (e.g. /etc, /var, /tmp) it's OK
if the root filesystem is readonly. Remove the fatal error if the mount
fails. Also clean up the mount call a bit - MS_MGC_VAL hasn't been
required since kernel 2.4...
|
| |
|
|
|
|
|
|
| |
This is done by a narrowly aimed 'udev trigger' call that shouldn't
retrigger anything else. No related modules need to be loaded manually
now, they're builtin.
Resolves: rhbz#672527
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
| |
There were two kinds of unused but set variables:
- the really pointless ones
- the ones whose only purpose was to silence the warnings
about ignoring the return values from functions declared with
__attribute__((__warn_unused_result__))
Since gcc 4.5 the warnings can be disabled using -Wno-unused-result,
so the variables are no longer necessary.
Then -Wno-unused-but-set-variable is an overkill, so drop it.
|
| | |
|
| |
|
|
|
|
|
|
|
| |
Seriously overengineer the initialization of the env variables we pass
to loader so that we can set MALLOC_PERTURB_ to a random number. Is
this worth it - who can say?
You could be listening to the score of "The Magnificent Seven" right
now, but you're not.
|
| |
|
|
|
| |
Setting HOME to /root didn't work right because I invented two
mechanisms to do the same thing. Get rid of one of them.
|
| |
|
|
|
|
| |
Not working right since 88958d9d84f48cfa0f3ca7ff94326b49c252ebe1.
Related: rhbz#576439
|
| | |
|
| |
|
|
|
| |
If chdir() does not return 0, display an error of some sort and don't do
what we were about to do.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
| |
Discussed with jlaska and amit. Removes the virtiolog= command line
parameter again, istead we will always look for
/dev/virtio-ports/org.fedoraproject.anaconda.log.0 and forward logs to it
if it exists. In theory, with some cooperation from the libvirt team, this
could ensure every F14 installation has remote logging set up, ready to be
picked on the host.
Related: rhbz#576439
|
| |
|
|
|
|
|
| |
ssh was already droping you there (though I'm not entirely sure where
the directory is coming from in that case), and you'd not have
.bash_history and .profile and .gdbinit. So now everything uses /root ,
and you get the stuff you might expect.
|
| | |
|
| |
|
|
| |
This gets rid of ugly checking for this in /proc/cmdline in loader and init checks if devel is actually a single word to avoid conflicts
|
| |
|
|
| |
- also make sure the terminal is in usable state when back in init
|
| |
|
|
|
| |
- when development ('devel' argument) mode is requested, init enables
core dumping and starts shell in the case of unexpected exit
|
| |
|
|
|
|
| |
If we wait only on udevd's pid, it means we've got zombie processes
hanging out. So instead, wait on all pids, and then only do the udevd
stuff if we actually see udevd exit.
|
| | |
|
| |
|
|
| |
make sure sparc64 has lib64 paths set that same as other 64 bit arches that use lib64
|
| |
|
|
|
|
|
|
| |
Stop the kernel correctly on HALT (so that we dont see 'kernel panic, they
killed init' on i386 and on s390 one gets easy access to manual IPL).
Introduce a new reboot method that does allow us to see the backtrace and
doesn't scroll the screen up with useless unmount info.
|
| |
|
|
| |
The error could be triggered, e.g. by appending 'stage2=hd:label="fedora"' to the boot line.
|
| |
|
|
|
|
|
|
| |
dcantrell's recent change to the shutdown mechanism exposed a situation when
shutdown is correctly handled but the init's exit handler still gets called and
dumps its callstack, that's not desirable because (in case of an exception in
python code for example) part of the useful information on the screen is
scrolled out of view.
|
| |
|
|
|
|
|
| |
A proper syslog daemon allows for remote logging that includes installed
system's syslog.
Removes the init.c code that simulated syslog activity until now.
|
| | |
|
| |
|
|
|
|
| |
Wrong check for the architecture resulted in having no lib64 directories
in LD_LIBRARY_PATH, and that's why anaconda couldn't find the libudev
library.
|
| |
|
|
|
|
|
|
|
|
| |
Be default mount makes tmpfs mounts 1/2 RAM size, on 256MB machines, where
part of the RAM is taken up by a shared framebuffer, this results in a
/tmp which is to small to hold install.img, resulting in install failure
when trying to do for example a network installation.
This patch fixes this by specifying a size for /tmp, it uses 250MB which
should be plenty for now and the forseeable future.
|
| |
|
|
|
| |
All previous comments about test mode being untested apply to loader, but
doubly so.
|
| |
|
|
|
| |
We've got a lot of dead code here. This removes a bunch of it. Also,
it fixes gettimespecofday's math to yield the correct result.
|
| |
|
|
|
|
| |
Since we don't want init to exit when we call udevadm, fork first. But
before we do that, may as well wait for our forked udevd to exit, so
there's some chance it's listening for connections.
|
| |
|
|
|
|
|
|
|
|
|
| |
Currently if init exits unexpectedly, we get "Kernel panic - not
syncing: Attempted to kill init!" and then a useless kernel traceback
(hey, it went through do_exit()!, who'd'a thunk?) but nothing meaningful
about what went wrong in init. This patch attempts to rectify this
situation by adding the traceback handler from loader to init, as well as
triggering it (in both loader and init) if we exit without going through a
path we've specifically marked, by using doExit() instead of exit(3) or
_exit(2), as not needing a traceback.
|
| |
|
|
|
|
| |
This fixes the vim magic to be something it'll actually do something
with. I probably got it wrong the first time, but hey, better usable
than not.
|
| |
|
|
| |
This snuck in as part of another commit when it shouldn't have.
|
| | |
|
| |
|
|
|
|
|
|
| |
The livecd includes anaconda, for obvious reasons. However this means that
70-anaconda.rules will be installed on the finished system and running our
rules has some unusual side effects. While we slowly push our udev rules
changes into upstream packages, this should make them a no-op on the
installed system.
|
| |
|
|
|
|
|
|
|
|
| |
Touch /dev/.in_sysinit, as that stops /lib/udev/rules.d/65-md-incremental.rules
from messing with mdraid sets.
This patch adds the touching twice, once to our own init, for when running
as standalone installer, and once in python for when running from a livecd,
to stop the udev trigger "block" we do will cause
/lib/udev/rules.d/65-md-incremental.rules to trigger in the livecd case.
|
| | |
|
| |
|
|
|
|
| |
The new reboot code didn't take into account the fact that there's a
different sigint handler while anaconda is running vs when we're done
installing. This patch fixes that bad assumption.
|
| |
|
|
|
|
| |
Rework shutDown() so that we don't do unmounts until we're going to
reboot or power off the machine. Also some minor cleanups like taking
out all the negative-phrasing of "noKill" in the code itself.
|
| | |
|
