Work around a network whose IPv6 is advertised but broken
The cited issue #12049 does not support the cause. Its reporter had native IPv6 that worked generally (archlinux.org over IPv6 returned 200, small IPv6 responses were fine), and the resets were intermittent, only on large files from pkgs.omarchy.org behind Cloudflare with cf-cache-status DYNAMIC, and the same curl -6 succeeded 20 minutes later. The reporter suspected the server edge, not a broken local path. The record presents it as evidence of a broken ISP or CPE path. Cause corrected to cover both a broken path on your side and a fault on the server's IPv6 side, with a diagnostic that tells them apart. A connection reset is an active RST, which a blackholing path does not produce, so the corrected cause no longer attributes resets to a blackhole. The workaround in option 3 is the one #12049 used verbatim, and the pacman Server line https://pkgs.omarchy.org/stable/$arch was confirmed in /etc/pacman.conf here. The gai.conf line is the Arch wiki's 'Prefer IPv4 over IPv6' line and is present commented in /etc/gai.conf (glibc). Quoting added because Omarchy package file names contain colons. The danger named CGNAT, which is IPv4 and is not broken by turning IPv6 off, so it is corrected to IPv6-only, 464XLAT and DS-Lite networks. #9610 supports the portal IPv6 blackhole remark. Not exercised: no broken IPv6 network was available.
Disabling IPv6 on a connection makes an IPv6-only network, or one that provides IPv4 only through 464XLAT or DS-Lite (some mobile carriers and ISPs), lose connectivity entirely. Use it per profile, never globally, and undo it with nmcli connection modify "<profile>" ipv6.method auto.
Symptom
Some sites take 20 seconds or more to start loading while others are instant. Large downloads fail part way with Connection reset by peer, including omarchy update:
error: failed retrieving file '...pkg.tar.zst' from pkgs.omarchy.org : Recv failure: Connection reset by peer
curl -4 against the same URL works every time, curl -6 stalls or resets. The laptop has a global IPv6 address and an IPv6 default route.
Cause
Applications try IPv6 first whenever the machine has a global IPv6 address and an IPv6 default route, which a router advertisement provides. Two different faults then look the same from the laptop.
- The IPv6 path on your side is broken or lossy: a misconfigured ISP tunnel, a CPE that drops large or fragmented packets, or a captive portal that blackholes IPv6 instead of redirecting it. Happy Eyeballs in browsers and libcurl hides a path that fails on connect, but a connection that opens and then stalls mid-transfer is not retried, so large downloads fail while small pages load.
- Your IPv6 works, but one server's IPv6 side does not. Issue #12049 is this case: large downloads from
pkgs.omarchy.org(behind Cloudflare) reset intermittently over IPv6 only, while other IPv6 sites worked and the same download over IPv4, or over IPv6 twenty minutes later, completed. The reporter suspected the CDN edge, and no cause was confirmed.
In both cases the workaround is to send that traffic over IPv4. This is distinct from the resolver problem in ipv6-dns-timeout-stub-not-on-localhost6: DNS answers fine here, and it is the IPv6 traffic itself that fails.
Fix
Confirm it is IPv6, then tell the two causes apart. If curl -6 fails against a large file on one host but works against other IPv6 sites, the fault is on that server's side and only a per-download workaround makes sense. If every IPv6 test stalls, the path on your side is broken:
ip -6 route show default
curl -6 -sS -o /dev/null -w '%{http_code} %{time_total}s\n' https://archlinux.org/
curl -4 -sS -o /dev/null -w '%{http_code} %{time_total}s\n' https://archlinux.org/
ping -6 -c 4 2606:4700:4700::1111
Prefer, in order of how much they change:
- Prefer IPv4 for this machine without turning IPv6 off. Add the Arch wiki's line to
/etc/gai.conf(it is already there, commented out), which reorders getaddrinfo results so programs that use them, such as curl and pacman, try IPv4 first:
echo 'precedence ::ffff:0:0/96 100' | sudo tee -a /etc/gai.conf
- Disable IPv6 only on the one network that is broken (the local path case):
nmcli -t -f NAME,DEVICE connection show --active
nmcli connection modify "<profile>" ipv6.method disabled
nmcli connection up "<profile>"
- To finish a single stuck package download now, fetch it over IPv4 into the pacman cache, then rerun the update, which verifies the cached file. Quote the name, because Omarchy package file names can contain a colon:
sudo rm -f "/var/cache/pacman/pkg/<file>.part"
sudo curl -4 -o "/var/cache/pacman/pkg/<file>" "https://pkgs.omarchy.org/stable/x86_64/<file>"
omarchy update
Avoid the kernel parameter ipv6.disable=1. It removes IPv6 everywhere, breaks networks where IPv6 works, and on Omarchy 4 it would have to go in /etc/limine-entry-tool.d/*.conf and needs a UKI rebuild.
Verify
nmcli -g ipv6.method connection show "<profile>" # disabled, if you used option 2
ip -6 addr show scope global # empty on that link after option 2
getent ahosts archlinux.org | head -3 # IPv4 first, after option 1
Re-run the download that failed. It should complete.