Florian Forster [Thu, 3 Dec 2015 12:05:03 +0000 (13:05 +0100)]
ceph plugin: Rewrite handling of JSON state.
The previous code didn't handle non-numeric map values correctly,
leaking state and resulting in ridiculously long key strings. This
rewrite fixes this and adds a unit test to ensure that this is actually
working as intended.
Ruben Kerkhof [Tue, 4 Aug 2015 13:28:03 +0000 (15:28 +0200)]
gcrypt: remove exec-prefix check
libgcrypt-config --libs already includes the correct library include path.
Also, we hardcoded it to $exec-prefix/lib, which is just plain wrong on
64bit architectures.
Florian Forster [Mon, 30 Nov 2015 10:25:48 +0000 (11:25 +0100)]
ceph plugin: Rewrite parse_keys() and compact_ds_name().
This rewrite removes the logic to add the original length to the end of
the buffer in case of truncation. This logic was guarded by:
tmp[DATA_MAX_NAME_LEN - 1] = '\0';
if(strlen(tmp) > DATA_MAX_NAME_LEN - 1) { ... }
and was therefore dead code.
Florian Forster [Mon, 30 Nov 2015 09:15:06 +0000 (10:15 +0100)]
ceph plugin: Add unit test for parse_keys().
This is used to demonstrate a buffer overflow: when the first part of a
key is >63 characters, key_chars_remaining underflows and causes a buffer
overflow in the following iteration.
Florian Forster [Wed, 25 Nov 2015 10:51:04 +0000 (11:51 +0100)]
statsd plugin: Correctly free metrics at shutdown.
The previous shutdown handler leaked "timer" and "set" metrics. This is
not a huge problem in itself, as the daemon is exiting shortly after
anyway, but it makes debugging of memory leaks very hard.
Marc Fournier [Sun, 15 Nov 2015 10:49:06 +0000 (11:49 +0100)]
apache: parse whole response, not only the first 16 lines
mod_status recently started reporting more statistics. According to
http://www.apache.org/dist/httpd/CHANGES_2.4 `mod_proxy, mod_ssl,
mod_cache_socache, mod_socache_*: Support machine readable server-status
produced when using the "?auto" query string.` for version 2.4.13
onwards.
As we can't assume everything we need will be included in the first 16
lines of output anymore, we now simply scan through the whole output.
Benjamin Drung [Tue, 27 Oct 2015 09:23:46 +0000 (10:23 +0100)]
Add utils_ignorelist back to daemon (#1323)
commit 216c6246b73645ac093de15b87aedc9abc6ebc80 moves the collectd
daemon files into the src/daemon/ directory, but also silently drops
utils_ignorelist. Dropping utils_ignorelist from the daemon has the
drawback that it cannot be used by plugins any more and that 16 plugins
need to bundle ignorelist.
Thus add utils_ignorelist back to the collectd daemon.
Jan Andres [Wed, 21 Oct 2015 11:50:43 +0000 (13:50 +0200)]
Start write threads after plugin initialization.
Some plugins such as "network" create own threads from within their
init callbacks which can then start submitting data to the queue
right away, even if the read threads haven't been started yet.
If write threads are started before plugin initialization, this can
result in a race where a plugin's write callback gets called before
that plugin's init callback has completed.
To fix this, delay starting the write threads until after all plugins
have been initialized.
Marc Fournier [Wed, 4 Nov 2015 20:12:36 +0000 (21:12 +0100)]
varnish: isolate varnish v2 code
Segregating v2-specific code will allow reworking the v3 and v4 parts,
while ensuring backwards compatibility with v2.
The downside is that this leads to a large chunk of mostly duplicate
code. That said, my suggestion would be to drop support for varnish v2
in collectd 5.6.
Also correct the maximum length of the node name.
The callback name is "write_redis/%s" so the maximum
length of a node name is DATA_MAX_NAME_LEN - strlen("write_redis") -1.
Marc Fournier [Sun, 13 Sep 2015 22:03:44 +0000 (00:03 +0200)]
gmond: fix linking against libganglia at build time
Current versions of libganglia don't ship the ganglia-config utility
anymore. Change the configure script to not expect it to be in $PATH and
provide sensible defaults instead.
Ruben Kerkhof [Sun, 16 Aug 2015 13:07:03 +0000 (15:07 +0200)]
dns: fix compilation on OpenBSD
OpenBSD doesn't have pcap-bpf.h
pcap.h has been including pcap/bpf.h since 2006.
Since we require a pcap which has PCAP_ERROR_IFACE_NOT_UP, introduced in 2008
this shouldn't break anything.
Marc Fournier [Wed, 5 Aug 2015 16:05:02 +0000 (18:05 +0200)]
routeros: fix a small build error pointed out by clang
routeros.c:334:35: error: 'memset' call operates on objects of type 'cr_data_t' (aka 'struct cr_data_s') while the size is based on a different type 'cr_data_t *' (aka 'struct cr_data_s *') [-Werror,-Wsizeof-pointer-memaccess]
memset (router_data, 0, sizeof (router_data));
~~~~~~~~~~~ ^~~~~~~~~~~
routeros.c:334:35: note: did you mean to dereference the argument to 'sizeof' (and multiply it by the number of elements)?
memset (router_data, 0, sizeof (router_data));
^~~~~~~~~~~