Do no isolate in case of emergency or severe problems
This patch changes local-fs.target and systemd-fsck to not use
"isolate" when going into emergency.
This fixes https://bugzilla.redhat.com/show_bug.cgi?id=810722
The motivation is, that when something wents wrong, we should
keep everything as it is, to let the user fix the problem. When
isolating we stop a lot of services and therefore change the
system heavily so that it gets harder for the user to fix.
An example is a crypted partition. When the fsck in a crypted
partition fails, it previously used "emergency/start/isolate"
which stops cryptsetup. Therefore if the user tries to fsck
e.g. /dev/mapper/luks-356c20ae-c7a2-4f1c-ae1d-1d290a91b691
as printed by the failing fsck, then it will not find this
device (because it got closed).
So please apply this patch to let the user see the failing
situation.
Thanks!
[zj: removed dead isolate param from start_target().]
code in src/shared/macro.h only defined MAX/MIN in case
they were not defined previously. however the MAX/MIN
macros implemented in glibc are not of the "safe" kind but defined
as:
shutdown: correctly wait for processes we killed in the killall spree
Previously we simply counted how many processes we killed and expected
as many waitpid() calls to succeed. That however is incorrect to do.
As we might kill processes that are not our immediate children, and as
there might be left-over processes in the waitpid() queue from earlier
the we might get more ore less waitpid() events that we expect.
Hence: keep precise track of the processes we kill, remove the ones we
get waitpid() for, and after each time we get SIGCHLD check if all
others still exist. We use getpgid() to check if a PID still exists.
This should fix issues with journald not setting journal files offline
correctly on shutdown, because we'd too quickly proceed from SIGTERM to
SIGKILL because some left-over process was in our waitpid() queue.
core/socket: log errors when starting socket for this socket
When showing an error like 'Socket service not loaded', the
error won't show up in the status for the socket, unless it is
marked as SYSTEMD_UNIT=*.socket. Marking it as SYSTEMD_UNIT=*.service,
when the service is non-existent, is not useful.
No need to call the heavy artillery, when the original array
is sorted. Reduces complexity from n² log n to n log n, where
n is the number of items in the array, not very large, but
still.
If the configured number of samples was close to MAXSAMPLES,
the samples buffer could be overrun:
- by 1, because of off-by-one in the condition (samples > arg_samples_len),
and
- by many in case of an overrun, because the number of samples to
capture was increased, instead of being decreased.
Simplify things by converting to a normal for-loop.
In store.c: change buffer size from 4095 to 4096. 4095 is a strange
number.
nss-myhostname: use _cleanup_ and split function into two
The triply nested loop is just too much. Let's split out the
middle loop's body, so the whole thing is easier to read. Also
modernize the style a bit, using structure initialization to
avoid memset and such.
bus: rename sd_bus_get_peer() to sd_bus_get_server_id()
This function always returns the server side ID. The name suggested it
was actually always the peer's ID, but that's not correct if the call is
called on a server bus context. Hence, let's correct the name a bit.
Kelly Anderson [Fri, 29 Mar 2013 23:23:35 +0000 (19:23 -0400)]
build-sys: force Python to write UTF-8
Here is a patch that fixes documentation with python 3.x in non utf-8
locales. Specifically in my locale latin-1 is the default setting for
output going to stdout, which causes it to fail. By writing directly
to file we are able to set the locale to utf-8.
build-sys,man: use XML entities to substite strings
This makes it easier to add substitutions to man pages,
avoiding the separate transformation step.
mkdir -p's are removed from the rule, because xsltproc will
will create directories on it's own.
All in all, two or three forks per man page are avoided,
which should make things marginally faster.
Unfortunately python parsers must too be tweaked to handle
entities. This isn't particularly easy: with lxml a custom
Resolver can be used, but the stdlib etree doesn't support
external entities *at all*. So when running without lxml,
the entities are just removed. Right now it doesn't matter,
since the entities are not indexed anyway. But I intend to
add indexing of filenames in the near future, and then the
index generated without lxml might be missing a few lines.
Oh well.
gcc thinks that errno might be negative, and functions could return
something positive on error (-errno). Should not matter in practice,
but makes an -O4 build much quieter.
In order to write tests for the catalog functions, they
are made non-static and start taking a 'database' parameter,
which is the name of a file with the preprocessed catalog
entries.
This makes it possible to make test-catalog part of the
normal test suite, since it now only operates on files
in /tmp.
Coverity complains: systemd-199/src/journal/catalog.c:126:
buffer_size_warning: Calling strncpy with a maximum size argument of
32 bytes on destination array "i->language" of size 32 bytes might
leave the destination string unterminated.
...and unfortunately it was right. The string was defined as a
fixed-size string in some parts of the code, and used a
null-terminated string in others (e.g. in log statements). There's no
point in conserving one byte, so just define the max language tag
length to 31 bytes, and use null terminated strings everywhere.
Also, wrap some lines, zero-fill less bytes, use '\0' instead of just
0 to be more explicit that this is one byte.
systemd-199/src/bootchart/store.c:289: buffer_size_warning: Calling
strncpy with a maximum size argument of 256 bytes on destination array
"ps->name" of size 256 bytes might leave the destination string
unterminated.
...and indeed, the string was used as NULL-terminated later on.
pid_cmdline_strncpy is renamed to pid_cmdline_strscpy to commemorate
the fact that it *does* properly terminate the string.
systemd-199/src/shared/utmp-wtmp.c:228: buffer_size_warning: Calling
strncpy with a maximum size argument of 32 bytes on destination array
"store.ut_line" of size 32 bytes might leave the destination string
unterminated.
The destination string is unterminated on purpose, but we must
remember that.
Martin Pitt [Wed, 27 Mar 2013 07:15:12 +0000 (08:15 +0100)]
keymap: Fix Touchpad Toggle on MSI Wind U90/U100
This key is handled by the hardware already, so handling it again in software
nullifies the effect. Newer kernels read the real state and send out a separate
KEY_TOUCHPAD_ON or KEY_TOUCHPAD_OFF event, so in both cases we need to ignore
that key.
To make the result more visible, special return value
is used to tell automake that the test was skipped. While
at it, use the same return value in other skipped tests.
The rules governing %s where just too complicated. First of
all, looking at $SHELL is dangerous. For systemd --system,
it usually wouldn't be set. But it could be set if the admin
first started a debug shell, let's say /sbin/sash, and then
launched systemd from it. This shouldn't influence how daemons
are started later on, so is better ignored. Similar reasoning
holds for session mode. Some shells set $SHELL, while other
set it only when it wasn't set previously (e.g. zsh). This
results in fragility that is better avoided by ignoring $SHELL
totally.
With $SHELL out of the way, simplify things by saying that
%s==/bin/sh for root, and the configured shell otherwise.
get_shell() is the only caller, so it can be inlined.
units: automatically order all mount units after network.target
Previously it was necessary to pull in remote-fs-pre.target to order the
mount units against network.target since the ordering was done
transitively via remote-fs-pre.target.
As network implementations shouldn't need to know about the specific
use-case of network mounts we instead now simply order network.target
against all mounts too. This should make it unnecessary for network
managing services to import remote-fs-pre.target explicitly, as
network.target will now suffice.
Auke Kok [Fri, 22 Mar 2013 22:09:45 +0000 (15:09 -0700)]
readahead: chunk on spinning media
Readahead has all sorts of bad side effects depending on your
storage media. On rotating disks, it may be degrading startup
performance if enough requests are queued spanning linearly
over all blocks early at boot, and mount, blkid and friends
want to insert reads to the start of these block devices after.
The end result is that on spinning disks with ext3/4 that udev
and mounts take a very long time, and nothing really happens until
readahead is completely finished.
This has the net effect that the CPU is almost entirely idle
for the entire period that readahead is working. We could have
finished starting up quite a lot of services in this time if
we were smarter at how we do readahead.
This patch sorts all requests into 2 second "chunks" and sub-sorts
each chunk by block. This adds a single cross-drive seek per "chunk"
but has the benefit that we will have a lot of the blocks we need
early on in the boot sequence loaded into memory faster.
For a comparison of how before/after bootcharts look (ext4 on a
mobile 5400rpm 250GB drive) please look at:
http://foo-projects.org/~sofar/blocked-tests/
There are bootcharts in the "before" and "after" folders where you
should be able to see that many low-level services finish 5-7
seconds earlier with the patch applied (after).