SmartData@PoliTO · Network measurements · October 2026
Service continuity, Internet scanning and web probing during a critical security response.
The PHP2SHELL vulnerabilities prompted Politecnico di Torino to restrict external access to selected web services, apply updates and progressively restore access. We combine operational records with passive network measurements to understand what changed for our services—and which external clients kept contacting the network.
One incident. Three questions.
A security response changes both service availability and the traffic we observe. We examine the incident from two perspectives: the university services receiving connections and the external sources initiating them.
- Service continuity. How did access restrictions, courtesy pages, mitigations and patching affect the services used by students and staff?
- Horizontal scanning. How are newly observed sources organised when they contact large portions of the campus address space? Do their targets overlap, or do they divide the work?
- Concentrated web probing. What distinguishes clients making many connections to a few web services, and what evidence can link activity across source addresses?
- 535
- PHP websites in the university’s operational update scope
- 67.7 million
- TCP/443 connection and attempt records in the analysed export
- 3.56 million
- distinct external source IP addresses in that export
The website count describes the operational response. The traffic counts describe external-to-campus TCP/443 observations; they are not counts of attacks, HTTP requests or individual users.
The main take-home messages
Three complementary accounts connect the operational response with the traffic seen at the network boundary. Each part provides the detailed timelines, figures and methodology.
The services · Continuity and recovery
A responding server is not necessarily a functioning website.
Services followed different recovery paths. Campus, Wi-Fi and VPN access remained available, while selected services faced external restrictions. The main Teaching Portal, didattica.polito.it, remained open throughout and is distinct from the teaching application affected by the restrictions.
- Restrictions can amplify retries. Incomplete connections to app.didattica.polito.it increased sharply across two equal-duration comparison windows.
- Response volume adds context. The main website continued answering with a small courtesy page, then returned substantially operational with mitigations before its full update.
11.4× more incomplete records for the teaching application in the restriction window than in the reference window. This is a transport observation, not an attack count.
The dark side · Horizontal scanning
New scanner groups keep exploring the network.
Groups of external IP addresses appear throughout the observation period and contact large portions of the campus address space. Some repeatedly contact the same destinations; others have almost disjoint target sets, consistent with a division of work.
- Infrastructure has several dimensions. Many source addresses or /24 prefixes can still belong to a single Autonomous System.
- Timing does not establish causality. The observations do not establish a disclosure-driven increase or identify PHP2SHELL-specific exploitation.
2,186 newly observed high-fanout sources. One cohort of 92 IPs contacts 92.1% of the reference address space with almost no overlap between source target sets.
The dark side · Probing web services
A few sources concentrate many exchanges on a few services.
A separate analysis selects high-volume clients contacting the university’s web servers. Twelve AWS addresses dominate this selected sample. Five present out-of-band security-testing names with identifiers sharing the same embedded machine field—a link more specific than a common cloud provider.
- Different clients have different patterns. A China Telecom source is slower and longer-lived, with no observed identifier link to the AWS group.
- A shared identifier is a correlation signal. It does not identify a physical machine, an operator, a successful callback or a compromise.
12 AWS IPs generate 83,044 bidirectional-data flows: 86.9% of the flows from the 17 selected clients, not of all campus web traffic.
The common lesson: interpreting traffic during an incident requires service-state information, both complete and incomplete connection records, and analysis across multiple source addresses.
The measurements reveal how services recover and how external clients behave. They do not, on their own, demonstrate exploitation of the disclosed vulnerabilities.
What the evidence can tell us
The study uses logs produced by Tstat, the passive monitoring tool used to collect network and transport statistics at Politecnico di Torino, together with operational communications and the service-restoration account. The traffic subset analysed here covers TCP port 443. It does not describe every protocol or every route used to access university services.
Connections are not web requests. A complete flow or an exchange with data in both directions does not prove that an application worked or that a request succeeded. Conversely, incomplete records can include legitimate retries and background scanning.
New means new within the capture. The short observation history does not establish that a source had never visited before. The detailed parts also distinguish selected populations and keep logger timestamps separate from operational times; exact reaction times to public disclosure should not be inferred without validated clock alignment.
Correlation is not exploitation evidence. Broad scans, short repetitive responses and security-testing identifiers help characterise behaviour. Application and host logs would be needed to establish specific exploit attempts or their outcomes. The absence of such evidence here does not establish the absence of attacks.
- M. Mellia, R. Lo Cigno and F. Neri. Measuring IP and TCP behavior on edge nodes with Tstat. Computer Networks, 47(1), 1–21, 2005.
- A. Finamore, M. Mellia, M. Meo, M. M. Munafò and D. Rossi. Experiences of Internet Traffic Monitoring with Tstat. IEEE Network, 25(3), 8–14, 2011.
Study overview and traffic analyses: Marco Mellia and Kai Huang, Politecnico di Torino · SmartData@PoliTO.
Operational response and service reconstruction: Giorgio Santiano, Andrea Garzena, Mauro Innocenti, Andrea Tomè, Gianluca Oglietti and Paolo Francese, ISIAD. Full authorship and contributions are listed in the detailed parts.