Intro
A security report landed with an item about "request smuggling in nginx + PHP-FPM setups." The person in charge of the report (a non-engineer) sent me a short question:
"Does our site use nginx?"
The answer fits in one word. But that day I decided not to stop at one word, and instead showed, one line at a time, what layers the server actually stands on. As a result, the question that was bound to come next — "We already patched PHP, so why are we updating Linux again?" — got resolved in the same thread, and I even got the long-pending approval for the kernel-update reboot.
First, verify the facts: is nginx installed?
Before answering, I check with commands, not memory.
$ rpm -q nginx httpd package nginx is not installed httpd-2.4.71-1.amzn2023.x86_64 $ sudo ss -ltnp | grep -E ':80 |:443 ' LISTEN 0 511 *:80 *:* users:(("httpd",...)) LISTEN 0 511 *:443 *:* users:(("httpd",...)) nginx is not installed; what listens on 80/443 is Apache (httpd). Just to be safe, I also checked the configuration the vulnerability assumes (reusing proxied connections to FPM):
$ grep -ri 'enablereuse' /etc/httpd/conf /etc/httpd/conf.d (no matches) Conclusion: "not applicable to us." Five minutes total. But if I just reply "we don't use it," the next security item will trigger the exact same exchange.
Hand over a "layer map," one line each
So I attached these three lines to the reply, together with the conclusion:
[App layer] The site itself + PHP ← the recent "PHP patch" was here [Web layer] Apache (httpd) ← today's nginx item is here; we run Apache, so N/A [OS layer] Linux (kernel, etc.) ← updates take effect only AFTER a reboot A site looks like a single page, but it is really a stack of layers, and every patch notification is about exactly one of them. Keep this map fixed, and from now on a report can be one sentence: "this one is about the Web layer."
The expected question arrives: "PHP is done, so why update Linux again?"
Looking at the map, the person asked exactly what I had anticipated. Their mental model was "PHP patch = server patch" — patch once and the whole server is up to date.
With the layers separated, the answer writes itself:
- What we did recently was an app-layer (PHP) patch. The OS-layer kernel is a different thing
- A kernel update does not take effect just by installing the package. The old kernel keeps running in memory, so a reboot is required
- Therefore "PHP is done" and "a Linux reboot is still pending" are both true at the same time — no contradiction
I also reported one more thing I had never passed along. On the Web layer, Apache had shipped 2.4.71 with a Critical (remote code execution) fix on 11/12, and our nightly auto-update timer had applied it the same day. When automation runs quietly and well, nobody ever hears about it. A successful application like that is exactly what deserves a one-line mention on the layer map.
Takeaways
- For reports to non-engineers, keep one fixed "layer map" of the server. Every time a patch notification arrives, point at the layer it belongs to
- With layers separated, finished work and remaining work never get mixed up. Here: "PHP: done / Apache: auto-applied / kernel: awaiting reboot"
- A kernel patch isn't done until the reboot — that single line became the basis for the approval
- The more off-target a question looks, the bigger the chance to show the structure. One "do we use nginx?" led straight to the reboot approval that had been stuck
