Différences

Ci-dessous, les différences entre deux révisions de la page.

--- tuto:monitoring:graylog_to_crowdsec [05/03/2021 15:20]
dani [g2cs]
+++ tuto:monitoring:graylog_to_crowdsec [05/03/2021 19:19] (Version actuelle)
rv [Background]
@@ Ligne 3: / Ligne 3: @@
 ===== Background =====
-Crowdsec's architecture allows running several agents, each parsing the local logs on the server it's running, and sending events to a local API. While this approach works and is flexible, it might not be the most efficient. In my case, all my server are already sending their logs to a Graylog instance. Running one crowdsec agent on all of those VM would be a waist :
+Crowdsec's architecture allows running several agents, each parsing the local logs on the server it's running, and sending events to a local API. While this approach works and is flexible, it might not be the most efficient. In my case, all my servers are already sending their logs to a Graylog instance. Running one crowdsec agent on all of those VM would be a waste :
   * I really like the idea behind the Journal (systemd-journald), it's very conveniant. But it has a major drawback : it's slow has hell ! Better have everything on SSD, or reading the journal will slows everything down. As I already have journalbeat collecting logs from the Journal, I prefer not adding another Journal reader, which will slows things down even further
   * Let's assume we have 40 VM on which we'd like crowdsec agent running. This means something like 40x80MB = 3.2GB of RAM, just for crowdsec
@@ Ligne 12: / Ligne 12: @@
 ===== Send logs from Graylog to ? =====
-As I already have all my logs in Graylog, it'd be better to send this stream of logs to a single crowdsec installation. But, for now, crowdsec doesn't have network logs input, it can only reads files and the Journal (I've opened a [[https://github.com/crowdsecurity/crowdsec/issues/621|ticket]] for this). So, the idea is to somehow forward the logs I want fro Graylog to a small daemon, which would write logs for crowdsec to consume.
+As I already have all my logs in Graylog, it'd be better to send this stream of logs to a single crowdsec installation. But, for now, crowdsec doesn't have network logs input, it can only reads files and the Journal (I've opened a [[https://github.com/crowdsecurity/crowdsec/issues/621|ticket]] for this). So, the idea is to somehow forward the logs I want from Graylog to a small daemon, which would write logs for crowdsec to consume.
+Here's the global flow
+{{ :tuto:monitoring:graylog_to_crowdsec.png |}}
 ===== g2cs =====
@@ Ligne 28: / Ligne 31: @@
       * error.log
     * httpd
-      * access
+      * access.log
+      * error.log
+    * zimbra
+      * mailbox.log
+  * maxlines is the number of lines each file will get before being truncated. As those log files are only to feed crowdsec, the g2cs daemon will truncate them if they reach this number of lines, so they do not grow indefinitely
+<note tip>You can choose a directory on a tmpfs filesystem to improve performance, as you do not need those logs to be persistent</note>
+<note>We could probably do the same thing with rsyslog, but it's configuration is arcane to me, so, it was easier to write the small g2cs daemon instead</note>
+===== Configure crowdsec =====
+Now that we have our g2cs daemon running, you can configure crowdsec acquisition to read these files. Something like
+<code yaml>
+---
+filenames:
+- /run/g2cs/logs/syslog.log
+labels:
+    type: syslog
+---
+filenames:
+- /run/g2cs/logs/nginx/access.log
+- /run/g2cs/logs/nginx/error.log
+labels:
+    type: nginx
+---
+filenames:
+- /run/g2cs/logs/httpd/access.log
+- /run/g2cs/logs/httpd/error.log
+labels:
+    type: apache2
+---
+filenames:
+- /run/g2cs/logs/zimbra/mailbox.log
+labels:
+    type: zimbra
+</code>
+===== Install the syslog output plugin on Graylog =====
+OK, now that we have crowdsec and g2cs ready, we need to send our logs from Graylog to g2cs. For this, we'll use the [[https://github.com/wizecore/graylog2-output-syslog/|syslog output plugin]]. Just download the jar from github, place it in your Graylog plugin dir (this depends on how you have installed graylog), and restart graylog-server.
+===== Create a syslog output =====
+Now in Graylog, you can create a new output. Go in System -> Outputs. Select the "Syslog output" and click launch new output
+{{ :tuto:monitoring:graylog_output.png |}}
+Now, configure your Syslog output like this :
+{{ :tuto:monitoring:graylog_crowdsec_output.png |}}
+  * Choose the UDP protocol
+  * Enter the DNS name or IP address of your server running g2cs
+  * Choose the port on which g2cs bind
+  * Choose the CEF message format
+===== Assign output to streams =====
+Now, you can assign in Graylog your new output to the streams you want. Go in the Stream menu, then, "Manage outputs"
+{{ :tuto:monitoring:graylog_stream_manage_output.png |}}
+And assign your Syslog output
+{{ :tuto:monitoring:graylog_assign_output_to_stream.png |}}
+You should now see logs flowing from Graylog, to crowdsec. I'm using this on a small graylog setup, ingesting about 400msg/sec, out of which ~200msg/sec are parsed by my single crowdsec install. I just have to install the bouncers where I want to react to all the agressive IP collected on all my servers
+===== Run crowdsec with less privileges =====
+Bonus point : as crowdsec only access logs from the g2cs daemon, you can run both as a less privileged user, instead of root. First, create an unprivileged user
+<code bash>
+useradd -r -s /sbin/nologin g2cs
+</code>
+Now adapt the systemd unit for crowdsec, eg in /etc/systemd/system/crowdsec.service.d/user.conf
+<code ini>
+Service]
+User=g2cs
+Group=g2cs
+</code>
+And create a systemd unit for g2cs itself, /etc/systemd/system/g2cs.service
+<code ini>
+[Unit]
+Description=Graylog to Crowdsec syslog daemon
+After=syslog.target
+[Service]
+Type=simple
+ExecStart=/usr/local/bin/g2cs --port=514 --logdir=/run/g2cs/logs
+User=g2cs
+Group=g2cs
+Restart=always
+PrivateTmp=yes
+PrivateDevices=yes
+ProtectSystem=full
+ProtectHome=yes
+NoNewPrivileges=yes
+SyslogIdentifier=g2cs
+# Allow binding on privileged ports
+CapabilityBoundingSet=CAP_NET_BIND_SERVICE
+AmbientCapabilities=CAP_NET_BIND_SERVICE
+[Install]
+WantedBy=multi-user.target
+</code>