Read a server log and point at the attacks SENTRY
The evidence of an attack is usually in a log nobody has time to read. This reads it in the browser, groups it by source, and points at the brute force, the scanning and the requests carrying a payload.
The log
What stands out
How it works
How to read this
Addresses and accounts are ordered worst first: a login after failures, a new administrator and requests carrying a payload sit above raw scanning. It reads volume and signature, not intent, so confirm before blocking - a busy legitimate client can look like a scanner, and a slow attack can stay under the line. Export the list to feed a block list, and keep the raw log with CHAIN if it may be evidence.
What it looks for
By address: 8 failed logins is noted and 20 is a brute force; a success after 5 failures is a possible compromise; failures against 10 or more different usernames is password spraying or credential stuffing; 5 non-existent usernames is enumeration; 12 requests to missing paths is noted and 30 is scanning; and any request carrying an attack payload or a scanner’s user-agent.
By account: failures from 5 or more addresses is one account tried from many places, and 20 is a distributed brute force; a success after failures from 5 addresses is a possible compromise. From sudo, su, useradd, usermod and gpasswd in the same log: sudo refused (not in sudoers), 3 wrong sudo passwords, 3 failed su to root, a root shell or su through sudo, commands that change accounts or permissions, an account first using sudo in the second half of a log whose first half shows others using it, an account created or given UID 0, and an account added to sudo, wheel, admin, root, docker or lxd.
Auth0 logs (a JSON array, or one event per line) count failed logins (f, fp, fu) and successes (s) the same way, and add Auth0’s blocks (limit_wc, limit_mu, limit_sul), breached passwords (pwd_leak), 3 failed or rejected MFA challenges, an MFA factor changed on an account with failed sign-ins, 5 failed Management API calls (fapi), and successful ones (sapi) from an address that failed sign-ins.
Syslog lines in either the RFC 3164 or the RFC 5424 format are read by the program named in them. A file is read a piece at a time, so a log of several hundred megabytes does not stall the page; only the counts are kept, with up to five example lines for each address and account. Where an address is on a map is not looked at.
Questions people ask about SENTRY
Which log formats can it read?
The SSH, sudo, su, useradd, usermod and gpasswd lines of auth.log or secure; syslog lines with an RFC 3164 or RFC 5424 header, read by the program named in them; Apache and nginx access logs in the Common or Combined format; and Auth0 tenant logs, as a JSON array or one event per line, including a log stream’s wrapped events. In SSH lines, rsyslog’s message repeated 12 times is counted as twelve.
It does not read web server error logs or Windows event logs, and a rotated log compressed as .gz has to be decompressed first.
What is the difference between brute force, password spraying and credential stuffing?
Brute force tries many passwords against one account or a few. Password spraying tries one or two common passwords against many accounts, staying under each account’s lockout limit. Credential stuffing tries username and password pairs leaked from another site’s breach.
In a log, spraying and stuffing look the same: one address failing against many different usernames. SENTRY flags that at 10 different usernames from one address and names both. It flags a brute force at 20 failed logins from one address, and one account failing from 20 different addresses as a distributed brute force.
Why is a successful login flagged as a possible compromise?
Because a success after a run of failures is the shape of a guessed password. SENTRY flags it when an address that has failed 5 or more times then signs in, and when an account signs in after failing from 5 or more different addresses.
It may be a person who mistyped and then got it right. Each address and account keeps up to five example lines, and the line that made the finding is always among them, so you can see the time and the account before deciding.
What counts as an attack payload in a web log?
The request path and query string, read both as written and URL-decoded, are checked for SQL injection (union select, ' or '1'='1, sleep(), path traversal (../, /etc/passwd), Log4Shell (${jndi:), cross-site scripting (<script, onerror=), command injection, template injection and server-side request forgery ({{ }}, 169.254.169.254), and probes for sensitive paths such as /.env, /.git, wp-login and phpmyadmin. A user-agent naming a scanner such as sqlmap, Nikto, Nmap or Nuclei is noted separately.
The status code is not considered, so a probe answered with a 404 is still listed: it shows the attack was tried, not that it worked. An access log does not record request bodies, so a payload sent in a POST body cannot be seen here. The table of requests keeps the first 500.
What do the sudo and su findings mean?
They show who became root and how. A root shell through sudo (sudo -i, sudo su, sudo bash) is noted, and is raised when the account was created during the log or first used sudo only in its second half. A refusal (user NOT in sudoers) is flagged as an account trying for rights it does not have, and 3 wrong sudo passwords or 3 failed su to root are noted.
An account created with, or given, UID 0 is a second root account. Being added to sudo, wheel, admin, root, docker or lxd is flagged as well, the last two because a member of either can mount the host’s root filesystem into a container.
Can it read a very large log?
Yes. The file is read a megabyte at a time and only counts are kept, so a log of several hundred megabytes does not stall the page. Past 200,000 addresses or accounts, new ones are counted but not tracked, and the page says so. A single line longer than 64 KB is not treated as a log line; it is passed over and counted.
Why is an address I trust flagged?
Because SENTRY reads volume and signature, not intent. A monitoring check or a crawler can request 30 missing pages, an office behind one public address can fail against 10 usernames on a Monday morning, and a client with an old saved password can fail 20 times on its own. A slow attack, equally, can stay under every threshold.
Confirm before blocking. The two CSV exports, of flagged addresses and of flagged accounts, carry the counts, the first and last time seen and the reasons for each, which is what you need to check a block list against.
Doing this often? The Sovereign Console reads logs and the rest of the evidence into one case and timeline, with Sigma rules to hunt across them. On its own, US$2 a month; with everything else, US$5 a month or US$48 a year.