Firefox network race condition

loveydovey

loveydovey

I think this is a Firefox issue, not FreeBSD networking stack or anything, but am I the only one having this?

Two of websites intermittently open up as if I am not logged in (even though I am logged in)....then I refresh the page, and it opens up like it should with me being logged in. It's literally only gemini.google.com and x.com/grok.com that do this.

AI suggested there's a race condition there somewhere. Apparently, TLS 1.3 that is heavily used by x.com/google.com uses 1 roundtrip handshake, meaning it can send a request with cookies before the handshake is complete or something like that? Anyway, I tinkered with some firefox settings, but I'm still getting this problem once or twice a day. It's not a big issue, but I'm curious.

AI also suggested FreeBSD network stack is so robust that it causes things like this to happen LOL.

I'm on Firefox 138. This was happening on FF 137 on Linux too, so I'm pretty sure it's a FF issue. But the interesting thing it was happening a lot more on FF 137 on Linux, versus FreeBSD, so it MIGHT be related to the OS.
 
I'm now trying to disable Enhanced Tracking Protection for these problematic domains. When it's on, Firefox isolates cookies. When cookies are isolated, when you open a website on cold start with a related domain cookie, it might block it but then allow it after the page is fully active.
 
Shit happens. Not just with these two sites. I often notice some unusual things at my regional level, too. For example, I couldn't make a payment due to an incomprehensible logout from the payment system - I start typing a template, enter my card details, and then... bam, the site says my session has expired, even though I was clearly active and there was no downtime (timeout). Then the site wrote some more crap. I don't really pay attention to that anymore. I don't have time to waste. On the 7th or 8th try, I managed to make a payment. Another time, a Tor browser accessed RuTube but couldn't play a video. The same thing happened in Linux. But for some reason, this happened a little more often in FreeBSD. That is, If I changed the torus chain to a new one, the chances of playing the video in Linux (in this case, Alpine Linux) were significantly higher. It was also possible to achieve results in FreeBSD, but for some reason this "jump rope" worked better in Linux. I don't have time to examine this situation under a microscope. But that's the reality. At least, that's what I encountered. There are quirks everywhere.
I also read somewhere about FreeBSD's "super-secure" network stack. They adopted it, I think, into Apple... same old story.
 
No, these problems for me are very consistent and very specific to two entities. Everything else works like clockwork. So I am sure this is not "shit happens" type of a thing.

Cookie containerization wasn't the issue, it seems. I'm now gonna try to cap my FF config to TLS1.2 version and see if that solves it.
 
I switched my webserver to TLS 1.3 a while back (exclusive for a while but later alongside 1.2; FF still defaulted 1.3) and haven't seen any issues any OS.

On my router with OpenWRT I think I had oddities if I messed with hardware offload or dropping invalid packets in Firewall.
 
Apparently, if the cause is bad site caching you can't completely fix it because it's a bug on the website. You can disable caching on the browser, but who would do that? Anyway, I'm also wondering, as cracauer@ mentioned, the extension issue. I have ublock origin and umatrix ... and umatrix does block service workers by default.
 
Thank you, cracauer@...I think the problem was the disabled service workers. I wouldn't have thought _that_ would cause that behavior because everything else works with them turned off. I'll mark this solved for the time being.
 
Ok, I just got the same problem with gemini.google.com . Service worker enabling does seem to make this issue less frequent, but it doesn't solve it. So weird.
 
You could try waterfox, which is a derivative of firefox but with a lot of things like telemetry taken out. It might be interesting to test if the bug you are seeing happens in waterfox as well. There is a waterfox package.
 
You must log in or register to reply here.
Back
Top