প্রতিলিপি 🔗
Tova Dvorin: Brief, a SafeBreach podcast. I'm Tova Dvorin,, your host, joined as always by my favorite cyber offensive cybersecurity engineer Adrian Culley.. This week, the PHP project released emergency fixes for two critical vulnerabilities, sending the whole internet into a tailspin, by the way. CERT-EU was already warning that some organizations might need to take websites offline, and now the NSA has given the whole thing a name, PHP2 Shell. Adrian, we're a few days past the patch. How concerned should people be?
Adrian: So it is serious Tova, however, cyber black October, vuln apocalypse, it is not. PHP sits underneath a huge share of the web. WordPress, Drupal, Joomla, Magento, plus intranets and vendor appliances nobody remembers even installing. There are two CVEs that are relevant, CVE 2026, 109618 and 109619. Both rated 9.2 on CVSS version four. PHP's own advisory says they can allow unauthenticated remote code execution. The patches exist, applying them everywhere is the hard part.
Tova Dvorin: Translate that for anyone who doesn't spend their day reading C VSS scores. What does on what does one why can't I speak today? What does unauthenticated remote code execution actually mean?
Adrian: So someone on the internet can run commands on your server without logging in, without a password, without tricking a user into clicking anything. They send a request and your server starts doing something it should never do. That's the sentence nobody wants to read out next to the name of their web server. The NSA's advisory stresses that these are flaws in PHP itself, not in the host of application or its configuration.
Tova Dvorin: And this isn't limited to one vulnerable website or plugin. Who exactly is exposed here?
Adrian: So it's very blunt over all PHP installations serving web requests are at risk. It named PHP dash FPM Apache with mod underscore PHP, PHP dash CGI, Franken PHP, which many of you may never have heard of before and Lightspeed. Then it ranks them. The risk is highest for mod underscore PHP where PHP says there is a demonstrated exploit in which an unauthenticated request can trigger the attack. before any application code runs.
Tova Dvorin: Hang on, before the application code even runs? That sounds like a pretty important distinction.
Adrian: Because most of what we call web application security assumes the attacker has to get through your application. Input validation, framework protections, code review, the authentication layer. If the floor fires while PHP is still handling the request before your code is even invoked, all of that is standing in a room that the attacker never enters.
Tova Dvorin: So you've essentially put an alarm on the front door and the burglar just comes through the wall. Fantastic. It's like the Kool Aid man, but evil.
Adrian: That's about right to be fair. Edge controls can still look at the request and vendors are shipping detection rules, which I'll come back to, but your application layer has nothing to say about it.
Tova Dvorin: Let's unpack the two vulnerabilities. What's going wrong under the hood without turning this into an instruction manual?
Adrian: Neither of us is going to describe exploitation and PHP's advisory doesn't either. At the level the project has published, CVE 2026.109618 is described as remote code execution via out of memory corruption. When PHP runs out of memory partway through a request, it can be interrupted while its internal request data is only partly set up. Cleanup then treats that half built data as if it were valid.
Tova Dvorin: So it's the cleanup process itself that breaks. PHP tries to recover from a problem and ends up creating a much bigger one.
Adrian: Exactly. And the second is the same theme from a different direction. CVE 2026 109619 is remote code execution via stale global state. Request owned internal state can survive after its memory has been freed and a later request on the same worker can use it before normal initialization sets in. That's why worker reuse matters. One workaround in the advisory is vendors have reproduced it is limiting each worker to a single request.
Tova Dvorin: Are both vulnerabilities equally dangerous across all of these different PHP setups?
Adrian: No, and the advisor is careful about what has been shown. From the out of memory floor, PHP's analysis says PHP-FPM is susceptible, but no complete exploit against FPM has been demonstrated. For the stale state floor, PHP says a session lifecycle route was validated against both mod underscore PHP and FPM with the FPM path depending on a suitable application operation. So mod underscore PHP is the sharp end. FPM is not a surf harbour though.
Tova Dvorin: Okay. How far back are we talking? Some of the early morning some of the early warnings mentioned PHP five.
Adrian: So early reporting, including from CertEU and the Australian Cyber Security Centre, said all versions from PHP 5 onwards could be affected. That was before details were published, so treat it as a hedge rather than a precise range. What's firm is the fix list. PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35, all released on the 21st of October.
Tova Dvorin: And what happens if you're still running an older version? Because I suspect there are quite a few PHP seven servers out there.
Adrian: There are PHP, as our own advisory says, 8.1 and older no longer receive updates from the project. But given the severity, the PHP security team published minimal source backports for 7.4, 8.0 and 8.1 aimed at people building from source or without a long-term support vendor. Some distributions and vendors are providing their own backports, so do check with yours.
Tova Dvorin: And source backport isn't exactly the same thing as getting a tested update throughout your normal package man throughout your normal package manager.
Adrian: Not remotely, you're running an end of life branch, you're taking on risk with every one of those options. And the honest answer is that you need
Load full transcript
an upgrade plan. One more date for the diary, 8.2 reaches end of life on the 31st of December this year. The release notes already say so.
Tova Dvorin: So what did the actual fix change? Can we explain that without getting lost in the internals?
Adrian: So Datadog security researchers published an analysis as they describe it, the fixed changes on memory exhaustion interrupts execution. A soft limit now defers the error until PHP reaches a safe point. A higher hard limit ends the worker immediately instead of running normal cleanup over data that might be invalid. The patch also resets request owned state before anything can reuse it. The project says it fixes the issue at the source.
Tova Dvorin: Now the timeline here is interesting because the warnings started before the patches were even available. Let's explore that a little bit.
Adrian: So on Tuesday the 20th, the PHP security team published an advance notice, an upcoming release, two critical vulnerabilities, both 9.2 scheduled for Wednesday at 2pm UTC. They also said they reserved the right to release earlier if details leaked or exploitation appeared. SirEU published advisory 202618 the same day and it named both CVE identifiers.
Tova Dvorin: What were organizations supposed to do in that window when they knew the problem was coming but couldn't patch yet?
Adrian: As reported three things, identify every PHP application in the estate and be ready to patch them all at once. If you can't patch, consider isolating internet facing applications from public networks. And if you can't patch quickly, consider taking websites offline.
Tova Dvorin: Telling organizations to take entire websites offline is a pretty serious recommendation. What did that signal to you?
Adrian: The people who had seen the advanced material thought the exposure was serious Tova and it wasn't only advice. Reporting at the time said some large Japanese sites went into maintenance mode ahead of the release window so they could patch the moment the fixes landed. That's operational disruption on the vulnerability with no confirmed victims and tells you how the people closest to it were reading the risk.
Tova Dvorin: Which brings us to the obvious question, how do we know whether anyone's actually exploiting this today?
Adrian: So we need to be very careful here. Here's what I can say with confidence. The Australian Cyber Security Centre said it was not aware of successful active exploitation within Australia. No threat actor has been attributed. As of last Wednesday, neither CVE was on CESA's known exploitive vulnerabilities list, though that may well have moved by the time the listeners hear this.
Tova Dvorin: Okay, so what are the signals we are seeing, at least at the time of this recording, even if they haven't been independently confirmed?
Adrian: Quite a lot. The SANS Internet Storm Center diary says Cloudflare is reporting exploit attempts and that proof of concept exploits and details have been published. Datadogs Write-Up cites AWS reporting attempts as early as the 17th, four days before disclosure. Those are secondary reports, so I'd call them credible signals rather than established fact. The sensible planning assumption is that somebody's already probing this vulnerability.
Tova Dvorin: That's difficult part, isn't it? Not having a confirmed victim doesn't mean that nobody's been compromised, especially if we're still figuring out what to look for.
Adrian: Same logic Tova, nobody has confirmed a compromise. My read that this is mostly because nobody is certain what a compromise would look like in their logs yet. A PHP worker getting corrupted isn't going to produce a tidy alert. You find it from what happens afterwards.
Tova Dvorin: And the NSA has gone as far as saying attackers are likely to weaponize PHP 2 shell. How much weight should we, the audience, put on that?
Adrian: It's an assessment, not an observation, and I want to keep those separate, but it's a significant assessment. The advisory asks organisations to inventory anything internet facing that depends on PHP, to add enhanced logging to replace end of life PHP services that can't be patched, and if an immediate update isn't feasible to take the system offline until it can be fully mitigated. That's consistent with what CertEU and the ACSC were saying.
Tova Dvorin: Let's turn that into something practical. If you're on a security team listening to this, where do you start?
Adrian: Inventory because you can't patch what you can't find. Not just the web servers you know about, content management systems, e-commerce platforms, intranets, containers built from old images, third-party hosted sites, and vendor appliances with PHP embedded. Internet facing first, the usual failure is that the asset list stops at the servers the infrastructure team owns, and the marketing site on the hosting provider's platform is nobody's.
Tova Dvorin: So first find everything, then patch it. Although I'm guessing even that isn't quite as straightforward as that sounds.
Adrian: Several, the fixed versions of the four I listed are your distribution of vendors packages. Linux distributions and long-term support vendors got the fixes before the public release, so check your own channel first. One example from the advisory, the volunteer maintained Remy repository, said it would not have updates on release day and PHP told people to switch to distribution packages or build from source. And in container estates, patching means rebuilding images and redeploying. not just updating a host.
Tova Dvorin: And the teams that can't patch immediately, what about the workarounds? How much protection do they actually buy you?
Adrian: So the project lists them and it's very clear about their limits. They reduce exposure to specific attack paths but do not make an installation safe. And the release notes say updating is the only safe choice. For the memory floor, that includes setting the memory limit to unlimited, which sounds
backwards. It removes the controllable failure PHP would otherwise hit, but the advisory warns an underlying allocation failure can still reach the vulnerable path.
Tova Dvorin: Sorry, unlimited memory. That sounds like the opposite of a security recommendation. Explain that one.
Adrian: Because the attack depends on PHP hitting its own memory limit mid-request. Remove that limit and you remove the easiest way to trigger it, but then you need something else to stop a single request eating the machine. So the advisory says use a container or cgroup limit as an availability boundary, ideally one that kills only the affected worker. It specifically warns against address space resource limits because those can cause a considerable failure themselves.
Tova Dvorin: What's the temporary fix for the second vulnerability?
Adrian: Switch off the session upload progress feature, restart every worker and set it somewhere that applies before request passing, not at runtime. On PHP 8.5, disable fatal error backtraces. The advisory is explicit that these are not with a variance. So these are turnicase. They're worth applying if you must, and then secure. So what comes next? The attackers goal is a foothold web server and foothold behavior is more predictable than the vulnerability. I'd expect a web shell in a web accessible directory. Mitre attack technique 1505.003. Command run by the web server process, technique 159 and outbound connections from that process, technique 1071.001. That's expected behavior, not something seen in the reported. More modified PHP files in web groups are worth hunting for on the way.
Tova Dvorin: servers make sound connections to just about anywhere.
Adrian: That's the question. In plenty of estates, yes, because nobody ever restricted eGrep from the web tier. If a web worker can't really reach the internet and attack it with code execution, it an easy second stage. Tight egress rules turn compromise into an incident you can contain. Vendors have published detection signatures and firewall rules too. Treat those as tripwires again and not fixes.
Tova Dvorin: Let's make this concrete. Let's say it's any weekday morning, you're the CISO and you've got five minutes with your team to be able to address this. What are the three questions you ask?
Adrian: So one, can you list every PHP runtime to operate or depend on? And what version do you want? Two, for the ones you can't patch this week, what is the compensating plan and who owns it? Three, and this is the uncomfortable one, if someone did get code execution on a web server tomorrow, would you know about it? Would the shell spawn alerts? Would the outbound connection be blocked and would it be on it in time?
Tova Dvorin: And that last question is the one I'd want to prove, not just get a reassuring answer to.
Adrian: It is the defenders I've worked with exercise the chain after the foothold, not just the rule on paper, a web shell landing commands running from the web process, data going out over HTTP. That's the kind of thing safe breach customers do with the hackers playbook. I'm deliberately not talking about product for these specific CVEs today. The principle holds on any platform though. When did you last validate that your controls stop a web tier foothold?
Tova Dvorin: Makes sense. Now Adrian, before we wrap up, what are you keeping an eye on next?
Adrian: So, four things. Whether CESA adds either CVE to the known exploitive vulnerabilities catalogue, which would turn probing into confirmed exploitation. Whether anyone demonstrates a complete exploit against the dash FPM, because that's where most modern stacks live. Whether the first public incident is attributed to a named actor and what happens on 31st of December when PHP 8.2 stops getting fixes. because there's a lot of people's estates we'll still be using 8.2.
Tova Dvorin: Is available, but that doesn't mean the risk has disappeared. Find the PHP instances you didn't know you had, get them updated, and make sure that you'd spot a compromise if one happened. Adrian, thank you so much for filling us in on this very big and very developing story today. And listeners, until next time, stay safe, stay safe with Safe Breach.