@@PRODUCT@@

Choosing a security profile per server

Every server has one security model: nine modules, each with an owner, a profile that bundles the settings, and one list of settings that runs from the platform through the server, the reseller and the customer account down to a single webs

Written for: Administrator

Every server has one security model: nine modules, each with an owner, a profile that bundles the settings, and one list of settings that runs from the platform through the server, the reseller and the customer account down to a single website. This guide shows how an administrator puts a server on a profile, changes or locks a single setting, and checks where a value comes from. The panel screen arrives in a later step; for now everything goes through the panel's API and through corectl on the server itself.

Step 1: see what the server does now

Ask for the server. You see the profile, how many settings deviate from it, the modules with their owner and health, and whether the last push from the panel reached the server.

curl -s https://panel1.corecp.dev/api/v1/security/nodes/stck1.corecp.dev \
  | jq '{profile, profile_changes, sync: .sync.state,
         modules: [.modules[] | "\(.module)=\(.owner)"]}'

A server the panel sees for the first time gets the Standard profile. Whatever the administrator had already configured on that server — the mode of the website firewall, for example — stays as it is: that value is taken over and the profile only fills in what nobody had decided yet. Such a server then shows as "standard + 1 change".

On the server itself corectl security status gives the same picture, with a line saying when the panel last pushed anything.

The settings live in /etc/corecp/node.yaml on the server. If you or a tool of your own rewrites that file — a YAML library, for example — the moment of the last push may come back in a different spelling, such as 2026-11-12 20:39:53+00:00 instead of 2026-11-12T20:39:53Z. The server reads both, and writes its own form again at the next change.

Step 2: put the server on a profile

There are four profiles: Off, Basic, Standard and Strict. A profile is a bundle of values at server level: the scanner, the website firewall, the request limit, the wp-admin challenge, XML-RPC, wp-cron, the process monitor, suspending on malware and fleet reputation.

curl -s -X PUT https://panel1.corecp.dev/api/v1/security/nodes/stck1.corecp.dev/profile \
  -H 'Content-Type: application/json' -d '{"profile":"strict"}' \
  | jq '{profile, profile_changes, sync: .sync.state}'

The answer says synced when the server received the new values. If it says not_in_sync, the reason is next to it (the server is unreachable, for instance). The server keeps running on its previous values; the panel retries every five minutes, and you can force a retry with a POST on …/nodes/stck1.corecp.dev/push.

Several servers at once go through POST /api/v1/security/profile with a list of servers. Every server is applied on its own and reported on its own: a server that fails does not hold the others back.

Step 3: deviate from the profile, or lock a setting

A single setting is changed per key. The server then counts as "profile + n changes", so it stays visible that it deviates. With lock the value becomes binding for everything below the server: a reseller or customer who wants to change the same key gets the refusal setting_locked, naming the level that holds the lock.

curl -s -X PUT https://panel1.corecp.dev/api/v1/security/nodes/stck1.corecp.dev/settings/wp.admin.challenge \
  -H 'Content-Type: application/json' -d '{"value":"always","lock":true}' \
  | jq '.settings[] | select(.key=="wp.admin.challenge") | {value, locked, source_level}'

A DELETE on the same route removes the value (and the lock) again; the server then inherits from the platform once more.

Step 4: switch modules on or off

Every module has an owner: CoreCP, cPGuard or off. Any server administrator can switch a module off and give it back to CoreCP; choosing cPGuard is the platform administrator's. The firewall and hardening can never be given to cPGuard, and the firewall can never be off either: the packet filter is always CoreCP's. Reputation and brute force stay CoreCP's too: cPGuard's IP reputation feed and its log monitor's bans work only inside its own packet filter, which never runs beside CoreCP's (measured; the refusal says so).

Choosing cPGuard for a server (platform administrator): the button Choose cPGuard for this server above the modules installs the addon if needed, with the licence already stored under integrations, and hands cPGuard the scanner and the website firewall rules. It runs as a task on the server; the modules below show the outcome. Afterwards you move any module back to CoreCP on its own with its picker, or everything at once with Everything back to CoreCP. A switch stops the old owner first and then starts the new one; if that fails, the old owner is back. The findings of cPGuard's scanner appear in the same Findings tab as the native scanner's.

curl -s -X PUT https://panel1.corecp.dev/api/v1/security/nodes/stck1.corecp.dev/modules/scanner \
  -H 'Content-Type: application/json' -d '{"owner":"off"}' \
  | jq '[.modules[] | select(.module=="scanner")]'

Step 5: check where a value comes from

On the server, corectl security explain says for every key what applies and which level and which id it came from — for one account or one website as well. That is the same answer the panel gives, with the same origin.

corectl security explain wp.cron --domain shop.example.com
corectl security explain wp.xmlrpc --account customer01

When it says locked there, the value is locked at that level. A customer sees a padlock instead of a change button for such a setting in the panel.

What resellers and customers may do

A reseller sets defaults and locks for their own realm for the settings of accounts and websites (wp-cron, XML-RPC, the wp-admin challenge, the crawlers and the country rules), through /api/v1/security/realm. Server matters — the profile, the modules, the firewall — stay with the server administrator. A customer configures their own websites within what is not locked; the website firewall mode of a website remains a server administrator's decision.

Hardening and the real visitor address

The profile also decides which PHP functions are switched off on the server: Standard takes away what a website never needs (process control, changing user), Strict also takes away exec, system and their kin, which some plugins use for image previews or backups. Whatever the profile, PHP files are never run from wp-content/uploads, and websites send mail only through the server's own mail system. Check what the server runs now:

corectl security status

The hardening: line shows the PHP set and the uploads rule; the real ip: line shows which proxies the server believes. Behind Cloudflare that works by itself: logs, the website firewall and bans see your visitors, not Cloudflare. Behind a proxy or load balancer of your own, add its addresses:

corectl security set web.real_ip.trusted 198.51.100.10
corectl security set web.real_ip.header X-Forwarded-For

Add only proxies that set that header themselves; a visitor's own header from anywhere else is ignored.

The malware scanner: findings, quarantine and repair

The scanner module looks for infected files in the hosting accounts' homes without a paid licence. With the Standard profile it watches files as they are written and quarantines a known infection within the minute; Basic leaves out the watching. Every profile but Off scans each night the files that changed that day and runs a full scan once a month. See how the scanner is doing:

corectl scan status
corectl scan findings --status live

A file the rules call a virus is moved to the quarantine store, which only root can read, and the website keeps running without it. A suspicious file — an .htaccess that runs images as PHP, say — is reported and left in place, and so is a program file in a website folder (with Strict it is quarantined too). A link that points outside the account's own home — into another customer's home, or to a system file — is removed. Every finding records the path, the owner, the rule, the file's fingerprint and the time.

To repair, restore the file from the newest backup copy that scans clean, put it back if it was a false alarm, or delete the quarantined copy for good:

corectl scan act restore --ids f20261114201002-208301
corectl scan act release --ids f20261114201002-208301
corectl scan act delete --account customer01 --rule CORECP_VIRUS_PHP_Eval_Request_Input

A released file goes on the server's false-positive list and is not reported again until its content changes. When no backup holds a clean copy, the restore refuses and says so; the quarantined copy stays as evidence. With scanner.suspend_on_malware on, an account that has scanner.suspend_repeat virus findings within scanner.suspend_days days is suspended, with the scanner named as the reason in the account's state and its findings; you lift it as you would any suspension. When the engine or the rules are missing the scanner says unhealthy in corectl security status and quarantines nothing — it never reports a server clean that it could not scan.

Customers with shell access run the same commands as themselves and see only their own findings: they can start a scan of their own home, look at the results and restore or rescan a file; releasing and deleting stay with you, so a false alarm is something they ask you about.

WordPress: wp-cron, xmlrpc, verification and the blacklist

The WordPress module has settings that go down to the website and that a customer may choose as long as they are not locked: wp.cron (WordPress or system, with wp.cron.interval_minutes), wp.xmlrpc (open, known, closed) and web.trusted_ips. As administrator you keep two platform lists: the networks known lets through (wp.xmlrpc.networks, Jetpack's by default) and the plugin blacklist (wp.plugins.blacklist), deactivated on every site and kept off.

corectl wp modes
corectl security set wp.xmlrpc closed --domain customer.com
corectl wp checksum --all --dry-run
corectl wp dbscan --all

Every night the server puts back WordPress files that differ from the official release (after a restore point) and scans the databases; the customer is told. A database finding is never cleaned by itself.

The web layer: request limit, challenge page, login protection, country rules

From Security 5 every website is also protected at the HTTP layer, without an external service. The profile sets the numbers; you tune per website.

  1. Request limit. A visitor who makes more requests per second than the class allows (generous, normal or tight; off in the Off profile) is not refused but sent to a challenge page of the server itself. The browser solves a small puzzle in a fraction of a second and is through for three days. Nothing is sent to another service. Trusted addresses (Trusted addresses on the account or the website) are never limited.
  2. Login protection for WordPress (wp-admin challenge): off, only when suspicious (the Standard default: a visitor with failed logins or a bad reputation) or always. Trusted addresses are never challenged.
  3. Country rules (Country rules): none, the whole website or the admin and login pages only, with a list of two-letter country codes and whether the list is refused (deny) or the only one admitted (allow).
  4. AI and Meta crawlers: allowed by default; switch them off per website.

Open the website's security settings (Security ▸ the website) and set the mode; a level above may have locked it, and the page says so. On the server, corectl web status shows what runs and corectl security status the health of the web module. If the challenge page's service is down, the limit still holds and the admin pages stay reachable; the module shows as unhealthy until it is back.

Monitoring: processes, cron, system files and the server's own addresses

From Security 7 the monitor module watches what a file scanner misses. It runs on its own when CoreCP owns the module (the default).

  1. Processes. Every minute the server looks at the processes of the hosting accounts. A program that runs from a temporary directory (/tmp, /var/tmp, /dev/shm, the account's own ~/tmp), from a file that was deleted, or as a known crypto miner is stopped in the Standard and Strict profiles, and reported in Basic (Process monitor: off, notify, stop). Its program file goes into the malware scanner's quarantine, or is kept as evidence. An unknown process that keeps a whole processor busy for half an hour is only reported. Processes of root and of system users are never stopped.
  2. A false alarm goes on Allowed processes (the server) or Allowed processes of this account with the full path of the program; it applies from the next minute, nothing has to restart. A process name on the list only silences the "busy processor" report. Only an administrator or server administrator fills these lists.
  3. Cron. A new or changed line in an account's crontab that downloads something and runs it, or hides what it runs, is reported to the account's owner and on the server's advisor. It is never removed.
  4. System files. Every day the server checks each file of its installed packages against the package's checksum and looks for traces of known rootkits. A difference is reported, never repaired.
  5. The server's own addresses. Every hour they are looked up on the blocklists in Blocklists for the server's own addresses (Spamhaus ZEN by default). You are notified when an address is listed, and again when it is removed from the list.

On the server, corectl security monitor status shows the module's health and the last run of each check, and corectl security monitor findings what it found. A check that cannot run shows the module as unhealthy; nothing is then reported as clean.