When an UDP inetd connection terminates we must make sure to unblock the
parent, which is (currently) blocked during the transaction.
This patch is a fix to a regression introduced in 2c904b5, for v3.1
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
- Don't warn of missing PID, even if only _d(), perfectly normal
- Allow stopping restart timer of non-services (e.g. task/run)
- Allow stopping services with no PID rn, i.e. stop libuEv timer
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
Finit tries to restart crashing services. If a service is removed
before Finit has given up restarting it, then the background timer
responsible the restart may dereference a free():d svc_t.
Hence, make sure to properly stop (and step) a service and its timer
callbacks in service_unregister() before calling svc_del().
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
In service_stop() we call svc_del() which, as of Finit v3.1, free():s
the svc_t, so we don't want any lingering restart timers to wake up
after free() and start dereferencing the svc_t.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
This is the first patch in a series to add support for
log:/path/to/logfile,priority:facility.level,tag:ident
In this patch support for 'priority:' and 'tag:', log file support
coming later.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
This patch adds support for creating, and removing, a PID file for
programs (services) that for some reason does not create one. With
Finit this is very useful since it allows us to use our condition
support even for "broken" daemons.
To enable this for a service, use the following syntax:
service pid[:[/path/to/]file[.pid]] /path/to/daemon
I.e., `pid` is a required argument if you want to enable the feature,
but the rest is optional:
pid --> /run/`basename $CMD`.pid
pid:file --> /run/file.pid
pid:file.dip --> /run/file.dip
pid:/mnt/k --> /mnt/k.pid
pid:/mnt/k.foo --> /mnt/k.foo
There is an additional feature with this patch. For daemons that run
with a different identiy one usually wants to create the PID file with
that identity, e.g.:
service inadyn -I dyn -- Dyn.com DDNS client
service inadyn -I freedns -- FreeDNS DDNS client
Here inadyn will look for /etc/dyn.conf and create /run/dyn.pid, and
/etc/freedns.conf and create /run/freedns.pid, respetively. While Finit
will in both cases assume the PID file is /run/inadyn.pid ... err fail!
To get out of this mess the `pid` argument can be used to tell Finit
what the real PID file the daemon creates is called:
service pid:!dyn inadyn -I dyn -- Dyn.com DDNS client
service pid:!freedns inadyn -I freedns -- FreeDNS DDNS client
The same composition rules listed above applies also to the `!` variant.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
The service_step() function may call svc_del() on an svc from the
service_monitor(), so we need a way to tell the latter that it no
longer can dereference the pointer.
This regression is limited to v3.1, due to the introduction of the
new non-shared-memory store of svc_t.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
This is a quite intrusive change. In addition to the already reserved
runlevels 0 and 6, halt and reboot, respectively, this patch reserves
runlevel 1 as *no networking*, like the traditional SysV init did. One
of the reasons for this change is the new 'single' user mode, introduced
earlier. Most users when entering 'S' or 'single' on the kernel cmdline
expect the single user mode (runlevel 1) to be without networking.
This leaves 2, 3, 4, 5, and 7, 8, 9 as fully user-configurable runlevels
with networking.
Of course, networking can still be enabled in runlevel 1 using ifconfig
or ifupdown, or change runlevel.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
This patch makes it possible to control the default redirection of
(misbehaving) services output to /dev/null. With this disabled,
a service may write to its stdout/stderr and mess up the console.
By default all service/run/tasks stdout/stderr is redirected.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
In f3669c7 the difference between static and dynamic services was
removed. This led to a regression in handling internally created
services for, e.g., dbus-daemon and udevd, mentioned in issue #90.
This patch introduces a "protected" flag for svc_t to prevent the
mark-and-sweep handling of regularly loaded services.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
Until now Finit has used mtime to determine if a file has been changed,
or simply touched to request reload, when `initctl reload` is called.
Using mtime for this purpose is a monumentally bad idea since time can
be changed at any point: a user may adjust the time, NTP continously
tunes the clock, and time stamps are stored as the walltime. So if we
compare timestamps against the last time we (booted or) did reload, we
would miss .conf file changes.
This patch replaces the mtime mistake with an inotify watcher for the
following files: /etc/finit.d/*.conf, /etc/finit.d/available/*.conf,
and /etc/finit.conf. All file changes between (bootstraps and) calls
to `initctl reload` are tracked, like the mtime backend it replaces.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
This patch fixes the problem with run/task jobs with conditions, run at
runlevel S (bootstrap), not being given enough time to actually start
and run even once.
The new svc counter 'once' is now inspected before transitioning away
from runlevel S to the configured runlevel. A global (currently hard
coded) timeout of 10 sec is given to all run/tasks.
As a spin-off, this patch starts up the event loop much earlier than
before. Meaning the initial runlevel change at boot is now done in
a finalize() callback when all run/tasks have completed, or when the
global timeout has expired. Starting the event loop earlier gives us
the possibility to rely on the pidfile plugin to provide our run/tasks
with the conditions necessary to start.
TODO: What remains is to add similar handling to all runlevel
transitions, not just the initial one at boot.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
Make svc state transition to RUNNING *before* service_start(), because
in that function we may call service_step() as well, which would start
the svc again (since it was still in READY).
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
Only run HOOK_SVC_START on successful start, and *after* having printed
the result of the started service, otherwise any print-outs from plugins
or tasks we start may overwrite the started service's progress.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
This patch adds a 'once' counter to the svc_t for run/task jobs. This
counter is incremented when Finit calls service_start() and is reset on
runlevel changes.
Also, at the end of a runlevel each run/task have their state reset to
SVC_HALTED_STATE so they can be launched again for the new runlevel, if
so declared.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
This patch fixes a rather nasty bug in the SIGTERM+SIGKILL logic
employed by Finit when stopping tasks, e.g. at a runlevel change.
When a process terminates, after Finit having sent it SIGTERM, and the
SIGKILL timer expires *before* the timer has been cancelled, then the
timer callback would cause kill(0, SIGKILL).
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
This patch extends the existing rlimit implementation to support setting
limits per service, run/task, inetd, and tty.
Use rlimit in /etc/finit.conf to change the global setting, which is
then inherited to each /etc/finit.d/*.conf. For each .conf file the
rlimit is reinitialized to the global finit.conf settings.
Also, add `unlimited` keyword, to replace the now deprecated `infinity`
keyword. The latter is however kept, for compatibility with previous
releases, for the foreseeable future.
Signed-off-by: Joachim Nilsson <troglobit@gmail.com>
Joachim Nilsson