What is attack and what are attack components?¶
Attack is a single hit or multiple hits that have the same attack type, parameter with the attack vector, and the address they are sent to. Hits may come from the same or different IP addresses and have different value of the attack vector within one attack type.
Hit is a serialized malicious request (original malicious request and metadata added by the filtering node).
Attack vector is a part of a malicious request containing the attack sign.
All attacks that can be detected by Wallarm are divided into groups:
Input validation attacks
Attack detection method depends on the attack group. To detect behavioral attacks, additional Wallarm node configuration is required.
Input validation attacks¶
Input validation attacks include SQL injection, cross‑site scripting, remote code execution, Path Traversal and other attack types. Each attack type are characterized by specific symbol (token) combinations sent in the requests. To detect input validation attacks, it is required to conduct syntax analysis of the requests - parse requests in order to detect specific symbol combinations.
Input validation attacks are detected by the filtering node using the listed tools.
Detection of input validation attacks is enabled for all clients by default.
Behavioral attacks include classes of brute‑force attacks: passwords and session identifiers brute‑forcing, files and directories forced browsing (dirbust), credential stuffing. Behavioral attacks can be characterized by a large number of requests with different forced parameter values sent to a typical URL for a limited timeframe.
For example, if an attacker forces password, many similar requests with different
password values can be sent to the user authentication URL:
To detect behavioral attacks, it is required to conduct syntax analysis of requests and correlation analysis of request number and time between requests. Correlation analysis is conducted when the threshold of request number sent to user authentication or resource file directory URL is exceeded. Request number threshold should be set to reduce the risk of legitimate request blocking (for example, when the user inputs incorrect password to his account several times).
Correlation analysis is conducted by the Wallarm postanalytics module.
Comparison of the received requests number and the threshold for the requests number, and blocking of requests is conducted in the Wallarm Cloud.
When behavioral attack is detected, request sources are blocked, namely the IP addresses the requests were sent from are added to the blacklist.
To protect the resource against behavioral attacks, it is required to set the threshold for correlation analysis and URLs that are vulnerable to behavioral attacks.
Brute force protection restrictions
When searching for brute‑force attack signs, Wallarm nodes analyze only HTTP requests that do not contain signs of other attack types. For example, the requests are not considered to be a part of brute-force attack in the following cases:
- These requests contain signs of input validation attacks.
- These requests match the regular expression specified in the rule Define a request as an attack based on a regular expression.
Types of protected resources¶
Wallarm nodes analyze HTTP and WebSocket traffic sent to the protected resources:
HTTP traffic analysis is enabled by default.
WebSocket traffic analysis should be enabled additionally via the directive
Wallarm nodes analyze WebSocket traffic only for input validation attacks.
Protected resource API can be designed on the basis of REST, gRPC, or GraphQL technologies.
Attack detection process¶
To detect attacks, Wallarm uses the following process:
Determine the request format and parse every request part as described in the document about request parsing.
Determine the endpoint the request is addressed to.
Apply custom detection rules determined in the LOM file.
Make a decision whether the request is malicious or not based on rules determined in proton.db and LOM.
Tools for attack detection¶
To detect malicious requests, Wallarm nodes analyze all requests sent to the protected resource using the following tools:
- Library libproton
Custom detection rules
The libproton library is a primary tool for detecting malicious requests. The library uses the component proton.db which determines different attack type signs as token sequences, for example:
union select for the SQL injection attack type. If the request contains a token sequence that matches the sequence from proton.db, this request is considered to be an attack of the corresponding type.
Wallarm regularly updates proton.db with token sequences for new attack types and for already described attack types.
The libdetection library additionally validates attacks detected by the library libproton as follows:
If libdetection confirms the attack signs detected by libproton, the attack is uploaded to the Wallarm Cloud and blocked (if the filtering node is working in the
If libdetection does not confirm the attack signs detected by libproton, the request is considered legitimate, the attack is not uploaded to the Wallarm Cloud and is not blocked (if the filtering node is working in the
Using libdetection ensures the double‑detection of attacks and reduces the number of false positives.
Attack types validated by the libdetection library
Currently, the library libdetection only validates SQL Injection attacks.
How libdetection works¶
The particular characteristic of libdetection is that it analyzes requests not only for token sequences specific for attack types, but also for context in which the token sequence was sent.
The library contains the character strings of different attack type syntaxes (SQL Injection for now). The string is named as the context. Example of the context for the SQL injection attack type:
SELECT example FROM table WHERE id=
The library conducts the attack syntax analysis for matching the contexts. If the attack does not match the contexts, then the request will not be defined as a malicious request and will not be blocked (if the filtering node is working in the
Analyzing requests with the libdetection library is disabled by default. To reduce the number of false positives, we recommend enabling analysis.
To enable the analysis:
- Set the value of the directive
on. The directive can be set inside the
locationblock of the NGINX configuration file.
- Set the value of the directive
onto allow analyzing the request body. The directive can be set inside the
locationblock of the NGINX configuration file.
Memory consumption increase
When analyzing attacks using the libdetection library, the amount of memory consumed by NGINX and Wallarm processes may increase by about 10%.
To check the operation of libdetection, you can send the following legitimate request to the protected resource:
curl "http://localhost/?id=1' UNION SELECT"
The library libproton will detect
UNION SELECTas the SQL Injection attack sign. Since
UNION SELECTwithout other commands is not a sign of the SQL Injection attack, libproton detects a false positive.
If analyzing of requests with the libdetection library is enabled, the SQL Injection attack sign will not be confirmed in the request. The request will be considered legitimate, the attack will not be uploaded to the Wallarm Cloud and will not be blocked (if the filtering node is working in the
Custom detection rules¶
Wallarm clients can set custom detection rules based on protected application specificities. There are the following types of custom detection rules:
Monitoring and blocking attacks¶
Wallarm can process attacks in the following modes:
Monitoring mode: detects attacks and displays information about attacks in the Wallarm Console.
Blocking mode: detects, blocks attacks and displays information about attacks in the Wallarm Console.
Wallarm ensures quality request analysis and low level of false positives. However each protected application has its own specificities, so we recommend analyzing the work of the Wallarm in the monitoring mode before enabling the blocking mode.
To control the filtration mode, the directive
wallarm_mode is used. More detailed information about filtration mode configuration is available within the link.
The filtration mode for behavioral attacks is configured separately via the particular trigger.
False positive occurs when attack signs are detected in the legitimate request or when legitimate entity is qualified as a vulnerability. More details on false positives in vulnerability scanning →
When analyzing requests for attacks, Wallarm uses the standard rule set that provides optimal application protection with ultra‑low false positives. Due to protected application specificities, standard rules may mistakenly recognize attack signs in legitimate requests. For example: SQL injection attack may be detected in the request adding a post with malicious SQL query description to the Database Administrator Forum.
In such cases, standard rules need to be adjusted to accommodate protected application specificities by using the following methods:
Analyze potential false positives (by filtering all attacks by the tag
!known) and if confirming false positives, mark particular attacks or hits appropriately. Wallarm will automatically create the rules disabling analysis of the same requests for detected attack signs.
Disable detection of certain attack types in particular requests.
Identifying and handling false positives is a part of fine‑tuning Wallarm API Security to protect your applications. We recommend to deploy the first Wallarm node in the monitoring mode and analyze detected attacks. If some attacks are mistakenly recognized as attacks, mark them as false positives and switch the filtering node to blocking mode.
Managing detected attacks¶
All detected attacks are displayed in the Wallarm Console → Events section by the filter
attacks. You can manage attacks through the interface as follows:
View and analyze attacks
Increase the priority of an attack in the recheck queue
Mark attacks or separate hits as false positives
Create the rules for custom processing of separate hits
For more information on managing attacks, see the instructions on working with attacks.