- Consistent use of $VAR vs ${VAR}
- Consistent use of && vs 'if'
- Add error checking to some places
- Consistent error messages ("Can't" vs "Cannot", etc.)
- Function declarations at the top
- Miscellaneous adjustments
Colin Walters [Wed, 24 Apr 2013 22:19:04 +0000 (18:19 -0400)]
coredump: use realloc() loop instead of malloc(768M)
I typically run VMs with 1024MiB allocated; systemd is unable to write
coredumps in this scenario at all because the default kernel
configuration will only overcommit 50% of available RAM.
Distributions may have selinux but not sushell or might
need to set a custom debug shell.
Defaults to /sbin/sushell if selinux is enabled, /bin/sh if not.
[zj: Renamed --with-debugshelltty to --with-debug-tty, and
added a line in output showing DEBUGSHELL and DEBUGTTY.
I figure that debug shell is pretty useful, and I hope
the extra line in configure status will draw attention
to it.]
clang emits warnings about unused attribute _saved_errno_, which drown
out other—potentially useful—warnings. gcc documentation is not exactly
verbose about the effects of __attribute__((unused)) on variables, but
let's assume that it works if the unit test passes.
It is imperative that open source code be well attributed.
Sprinkle attribute((alloc_size)) here and there, telling gcc
how much memory we are actually allocating.
According to gcc documentation, returned pointer "cannot alias any
other pointer valid when the function returns" and "the memory has
undefined content". This second part is (hopefully) untrue for all
those functions.
systemd-python: attach fields to JournalHandler, add SYSLOG_IDENTIFIER
Arbitrary fields can be attached at the level of the handler,
and they'll be sent with all messages from this handler.
This facility is used to attach SYSLOG_IDENTIFIER to all messages,
since otherwise journald attaches SYSLOG_IDENTIFIER=python or
something similar, which is completely useless.
When a trigger unit wants to know if a stop is queued for it, we should
just check precisely that and do not check whether it is actually
stopped already. This is because we use these checks usually from state
change calls where the state variables are not updated yet.
This change splits unit_pending_inactive() into two calls
unit_inactive_or_pending() and unit_stop_pending(). The former checks
state and pending jobs, the latter only pending jobs.
nss-myhostname: resolve 'localhost' so that /etc/hosts becomes optional
This makes sure nss-myhostname not only resolves the local host name to
127.0.0.2/::1 but also the host name 'localhost: to 127.0.0.1/::1. This
makes installation of /etc/passwd optional, as it usually only includes
a mapping for 'localhost'.
This change also resolves ::1 to the local hostname (as before), but
also lists 'localhost' as an alias. This means look-ups are now fully
reversible, even though they are 1:n mappings.
Finally, the module will no longer erroneously claim that local IP
addresses which aren't on the loopback device were.
Kay Sievers [Wed, 24 Apr 2013 17:12:44 +0000 (19:12 +0200)]
do not create /dev/rtc symlink, let systemd search for it if needed
The export of the RTCs hctosys flag is uneccesary, the kernel takes care
of the persistemt clock management itself, without any need for:
CONFIG_RTC_HCTOSYS=y
CONFIG_RTC_HCTOSYS_DEVICE="rtc0"
"Chaotic hardware platforms" without native kernel persistent clock
support will find the proper RTC with the logic rtc_open() without
the need for a custom symlink.
readahead: be more verbose about creation failures
systemd-readahead reports "Failed to create shared memory segment:
No such file or directory", but it's unclear how it can happen. Be
more verbose about failures.
systemd: fall back to mounting /sys/fs/cgroup sans xattr
xattrs on cgroup fs were added back in v3.6-rc3-3-g03b1cde. But we
support kernels >= 2.6.39, and we should also support kernels compiled
w/o xattr support, even if systemd is compiled with xattr support.
Fall back to mounting without xattr support.
Instead of having explicit type-specific callbacks that inform the
triggering unit when a triggered unit changes state, make this generic
so that state changes are forwarded betwee any triggered and triggering
unit.
Also, get rid of UnitRef references from automount, timer, path units,
to the units they trigger and rely exclsuively on UNIT_TRIGGER type
dendencies.
cgroup: make sure all our cgroup objects have a suffix and are properly escaped
Session objects will now get the .session suffix, user objects the .user
suffix, nspawn containers the .nspawn suffix.
This also changes the user cgroups to be named after the numeric UID
rather than the username, since this allows us the parse these paths
standalone without requiring access to the cgroup file system.
This also changes the mapping of instanced units to cgroups. Instead of
mapping foo@bar.service to the cgroup path /user/foo@.service/bar we
will now map it to /user/foo@.service/foo@bar.service, in order to
ensure that all our objects are properly suffixed in the tree.
As discussed with Dan Berrange it's a good idea to suffix all objects in
the cgroup tree with ".something", so that when the system is
partitioned using a resource management tool we can drop objects of
different types into the same partition directory without generate
namespace conflicts.
We'l add this to the Pax Control Group document as soon as write access
to the fdo wiki is restored.
systemd,nspawn: use extended attributes to store metadata
All attributes are stored as text, since root_directory is already
text, and it seems easier to have all of them in text format.
Attributes are written in the trusted. namespace, because the kernel
currently does not allow user. attributes on cgroups. This is a PITA,
and CAP_SYS_ADMIN is required to *read* the attributes. Alas.
A second pipe is opened for the child to signal the parent that the
cgroup hierarchy has been set up.
Because statfs.t_type can be int on some architecures, we have to cast
the const magic to the type, otherwise the compiler warns about
signed/unsigned comparison, because the magic can be 32 bit unsigned.
statfs(2) man page is also wrong on some systems, because
f_type is not __SWORD_TYPE on some architecures.
The following program:
int main(int argc, char**argv)
{
struct statfs s;
statfs(argv[1], &s);
nspawn: create empty /etc/resolv.conf if necessary
nspawn will overmount resolv.conf if it exists. Since e.g.
default install with yum doesn't create /etc/resolv.conf,
a container created with yum will not have network. This
seems undesirable, and since we overmount the file anyway,
let's create it too.
Also, mounting a read-write /etc/resolv.conf in the container
is treated as a failure, since it makes it possible to
modify hosts /etc/resolv.conf from inside the container.
systemd-logind: Fix linking by reordering libraries in LDADD
libsystemd-audit needs functions from libsystemd-shared, so
libsystemd-audit needs to appear first. Otherwise:
CCLD systemd-logind
./.libs/libsystemd-audit.a(audit.o): In function `audit_session_from_pid':
/home/josh/src/systemd/src/shared/audit.c:50: undefined reference to `detect_container'
buildsys: Add --disable-tests to avoid building tests
This patch adds --disable-tests to configure. It is based on a patch
posted by Thierry Reding in 2010. The motivation for adding it is that
some tests fail link-time when cross-compiling.
The patch adds a new Makefile variable -- manual_tests -- and uses
that instead of noinst_PROGRAMS. However, if ENABLE_TESTS is true,
the former is added to the latter. It also renames noinst_tests to
simply tests.
Harald Hoyer [Thu, 18 Apr 2013 08:15:25 +0000 (10:15 +0200)]
fileio.c: do not parse comments after non-whitespace chars
systemd does not want to understand comments after the first
non-whitespace char occured.
key=foo #comment will result into key == "foo #comment"
key="foo" #comment will result into key == "foo#comment"
"key= #comment" will result into key == "#comment"
"key #comment" is an invalid line
Harald Hoyer [Thu, 18 Apr 2013 05:34:25 +0000 (07:34 +0200)]
Add ugly CMP_F_TYPE() macro
On some architectures (like s390x) the kernel has the type int for
f_type, but long in userspace.
Assigning the 32 bit magic constants from linux/magic.h to the 31 bit
signed f_type in the kernel, causes f_type to be negative for some
constants.
glibc extends the int to long for those architecures in 64 bit mode, so
the negative int becomes a negative long, which cannot be simply
compared to the original magic constant, because the compiler would
automatically cast the constant to long.
To workaround this issue, we also compare to the (int)MAGIC value in a
macro. Of course, we could do #ifdef with the architecure, but it has to
be maintained, and the magic constants are 32 bit anyway.
Someday, when the int is unsigned or long for all architectures, we can
remove this macro again. Until then, keep it as simple as it can be.