The following information is taken from the feature_test_macros man
page:
<features.h> is a Linux/glibc-specific header file. Other systems have
an analogous file, but typically with a different name. This header
file is automatically included by other header files as required: it is
not necessary to explicitly include it in order to employ feature test
macros.
Reported-by: Tibor Vago <tibor.vago@gmail.com>
X-Gentoo-Bug: 399635
X-Gentoo-Bug-URL: http://bugs.gentoo.org/show_bug.cgi?id=399635
This adds a new function that allows us to override some specific rc.conf
options via the kernel cmdline aka /proc/cmdline.
This makes esp. sense in cases where a rc_parallel enabled system has boot
issues. Simply disable rc_parallel by appending rc_parallel=no to your kernel
cmdline.
This commit provides the checkpath applet with feature parity to
systemd's tmpfiles.c create_item function.
Very similarly to the systemd function, it does NOT do any of the
cleanup work in this function.
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org>
The old behaviour was to exit(EXIT_SUCCESS) in case the service has been stopped
already, even if further commands has been passed to the init script
(like zap, start).
So using for example /etc/init.d/foo stop zap start would abort immediately
after "stop" if the service has been stopped already. Though there may be cases
were we need it to proceed with the remaining commands, zap and start in this
case.
This patch fixes the behaviour to continue and proceed with the remaining
commands whenever necessary.
X-Gentoo-Bug: 371845
X-Gentoo-Bug-URL: https://bugs.gentoo.org/371845
During the sysinit and shutdown runlevels the logfile destination may be
read-only. Skip the error messages in this case.
X-Gentoo-Bug: 390645
X-Gentoo-Bug-URL: https://bugs.gentoo.org/390645
The logfile or its basedir may be read-only during shutdown because the
directory may be umounted or read-only remounted already. In this case we simply
skip this error.
This is related to a comment in bug 390645 but the initial bug is not fixed
through this commit.
X-Gentoo-Bug: 390645
X-Gentoo-Bug-URL: https://bugs.gentoo.org/390645
A variable used in the previous commit was actually removed in another
commit; that is why I didn't catch it.
Reported-by: Duncan < 1i5t5.duncan@cox.net>
X-Gentoo-Bug: 394369
X-Gentoo-Bug-URL: http://bugs.gentoo.org/show_bug.cgi?id=394369
The ewarn() function was affected by the EINFO_QUIET environment
variable which lead to warning messages being suppressed. Warnings
should not be suppressed.
Reported-by: Hanno Boeck <hanno@gentoo.org>
X-Gentoo-Bug: 380073
X-Gentoo-Bug-URL: http://bugs.gentoo.org/show_bug.cgi?id=380073
symbolic links should not be followed in an attempt to work out the name
of the service we are running. Also, @sysconfdir@/conf.d should be tried
as a backup directory for configuration files.
I would like to thank Robin Johnson for his input on this change.
X-Gentoo-Bug: 350910
X-Gentoo-Bug-URL: http://bugs.gentoo.org/show_bug.cgi?id=350910
As introduced in bug 372547 using service_get_value() in the init scripts
or using rc_service_value_get() directly will only return one line.
This patch fixes it by using the new rc_getfile() function, it returns even
multiple lines. We're still using a char *, so the lines will be appended
instead of added into new element.
X-Gentoo-Bug: 372547
X-Gentoo-Bug-URL: http://bugs.gentoo.org/372547
<snip>
Read the entire @file into the buffer and set @len to the
size of the buffer when finished. For C strings, this will
be strlen(buffer) + 1.
Don't forget to free the buffer afterwards!
</snip>
We also fix bug 374899 by adding this new function.
X-Gentoo-Bug: 374899
X-Gentoo-Bug-URL: http://bugs.gentoo.org/374899
OpenRC goes through the config and checks each option for duplicates.
Lets say we're on "rc_logger" currently and its the last option in the config
file and we previously defined rc_logger_path.
It now goes through all previous config options and compares those against the
current one "rc_logger" *but* it compares only the first N bytes, in this
case strlen("rc_logger"). So it strips the _path from "rc_logger_path" which
ends up into "rc_logger" and it compares that against the current one (also
"rc_logger"), it would then simply override the previous definition.
This patch fixes this behaviour to always compare the full option / variable
names.
Add "e" (O_CLOEXEC) to all fopen() calls.
Remove system() call and replace it by fopen() and friends.
This also fixes bug 368933
X-Gentoo-Bug: 368933
X-Gentoo-Bug-URL: https://bugs.gentoo.org/368933
Rather than listing explicit object files, ignore all of them in
the whole tree.
Also ignore patch/gdb related files throughout.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Since we always want debug.mk whenever we include cc.mk, move the include
out of the Makefiles and into cc.mk itself. This also fixes an include
order bug in rc/Makefile where debug.mk is included before cc.mk and
breaks the default CFLAGS setup in cc.mk.
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
runscript will try to get the dir and basename of a file/link in case it
contains at least one slash. This patch gives a temporary copy of the path to
the dirname() function since dirname() can modify its argument.
If there were no extra commands, rc_service_extra_commands returned a
list containing a single empty string. This changes that to return an
empty list, which is more consistent with what you would expect.
X-Gentoo-Bug: 360013
X-Gentoo-Bug-URL: http://bugs.gentoo.org/show_bug.cgi?id=360013
This function was returning the setting of the variable $opts, which is
not correct. $opts was used in baselayout-1, but it is replaced by
$extra_commands and $extra_started_commands in openrc.
This does not appear to break backward compatibility since this function
does not appear to be used anywhere in the openrc code.
Thanks to James Le Cuirot for the original patch.
X-Gentoo-Bug: 360013
X-Gentoo-Bug-URL: http://bugs.gentoo.org/show_bug.cgi?id=360013
This reworks the implementation of the --applet option so that it is
processed in run_applets() and does not require two calls to the
getopts_long() function. It is based on code by Robin Johnson and Chris
Richards.
X-Gentoo-Bug: 351712
X-Gentoo-Bug-URL: http://bugs.gentoo.org/show_bug.cgi?id=351712
Parisc generates a few relocs against internal symbols which are OK.
X-Gentoo-Bug: 258913
X-Gentoo-Bug-URL: http://bugs.gentoo.org/258913
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
The target for the symlinks in ${SBINDIR} (the same directory as the rc
binary) was "rc". This is not correct; the target should be
${SBINDIR}/rc instead.
This allows error checking inside the included files instead of either
including the appropriate file or a blank file.
Also the blank file named .mk gets removed by this change.
The main makefile, init.d/Makefile and src/librc/Makefile all contain
several shell calls which can be handled as make conditionals. This
switches them to conditionals.
This reworks the shell calls in the makefiles to use immediate
evaluation and should improve parallel building.
X-Gentoo-Bug: 289264
X-Gentoo-Bug-URL: http://bugs.gentoo.org/show_bug.cgi?id=289264
Refactor rc_conf_value into librc for use in library context.
Also requires moving:
- rc_conf internal static
- Defines: PROFILE_ENV, SYS_WHITELIST, USR_WHITELIST, RC_PATH_PREFIX
moved to rc.h with new RC_ prefix added.
- Defines: RC_CONF, RC_CONF_OLD moved to rc.h.
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org>
In addition to detecting what multicall applet we want via argv[0],
provide an explicit override mode with a --applet initial argument.
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org>
* status on a stopped service now has a return code of 3 (was 1)
* starting an already started service now has a return code of 0 (was 1)
* stopping an already stopped service now has a return code of 0 (was 1)
Any system using cgroups was being detected as an LXC system. This was
triggering on OpenVZ under RHEL6 as well as the "automated per tty task
groups" as discussed on the LKML.
All LXC users should now switch to the new rc_sys variable introduced in
the previous patch.
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org>
- Fixes bugs #347583, #349389, both of which were triggered by cgroups
being detected as the LXC subsystem type.
- Makes it much easier to select "prefix" type.
- "rc -S" will now print a warning if you have not configured rc_sys in
/etc/rc.conf
- All other semantics of rc_sys are unchanged in this patch.
Signed-off-by: Robin H. Johnson <robbat2@gentoo.org>
Under normal conditions, runscript creates one child and waits for its
termination, which is signaled by a pipe write from the SIGCHLD
sighandler.
When running killprocs however more than one SIGHCLD signal is generated, at
least on all of my amd64 boxes running on real hardware and in vmware.
When the first SIGCHLD occurs svc_exec leaves the loop and closes the pipe.
Subsequent SIGCHLDs during the close can lead to a race condition and create an
EBADF error in the pipe write (pipe is closed but the file handle is still !=
-1).
We avoid this by blocking SIGHCHLD during the pipe close.
Gentoo has always used the sytle:
* moo ... [ ok ]
Latest openrc now does:
* moo... [ok]
Realistically, 4 spaces out of 80 on reduced terminals doesn't make any
sort of realistic difference and it's been just fine for the last 10
years, so keep the default behavior.
the mtime of a file. It saves the shutdown time to this file also.
This is handy for systems without a working RTC chip.
Based on an idea by Michael A. Smith <michael@smith-li.com>.
Fixes Gentoo #272073.
OpenRC version is now stored as plaintext in /libexec/rc/version
Plugins (cursplash, splashutils) will have to be re-compiled to pickup
the new directories. State data needs to be moved from /lib/rc/init.d
to /libexec/rc/init.d as well.
This enables the code to include the daemons interpreter when matching
process name which fixes bug #164.
Existing init scripts that rely on the old behaviour will need to be
fixed.
New option so you can set the number of seconds to wait for and
check the daemon is still running.
The old code that periodically polled wasn't that reliable - in this
case simpler == better.
Fixes#160
