TRANSCRIÇÃO 🔗
WEBVTT
00:00:00.000 --> 00:00:03.319
GitHub will block pull request target in public
00:00:03.319 --> 00:00:07.700
repos by default. And the first brownout is this
00:00:07.700 --> 00:00:11.060
Monday. Attackers are racing WordPress sites
00:00:11.060 --> 00:00:16.179
during core updates. And CISA says the bug is
00:00:16.179 --> 00:00:20.379
being exploited. And a monitoring script drained
00:00:20.379 --> 00:00:25.559
about $2.7 million from Polymarket users after
00:00:25.559 --> 00:00:29.539
the vendor's DNS was hijacked at its registrar.
00:00:29.879 --> 00:00:32.280
I'm Brian Teller from Teller's Tech, and this
00:00:32.280 --> 00:00:33.665
is Ship It Weekly.
00:00:50.243 --> 00:00:51.920
Welcome back to Ship It Weekly,
00:00:52.159 --> 00:00:56.659
the show about the DevOps, SRE, cloud, platform,
00:00:56.899 --> 00:00:59.859
and security stories that matter when you are
00:00:59.859 --> 00:01:02.439
the person keeping the thing running at three
00:01:02.439 --> 00:01:04.959
in the morning. For the weekly story list and
00:01:04.959 --> 00:01:08.680
source links, check out oncallbrief.com. For
00:01:08.680 --> 00:01:12.459
past episodes and show notes, head over to shipitweekly.fm.
00:01:12.459 --> 00:01:16.640
This week, GitHub's pull request target
00:01:16.640 --> 00:01:21.079
brownouts, a WordPress update race on CISA's
00:01:21.079 --> 00:01:24.879
exploited list, a hijacked DebugBear script
00:01:24.879 --> 00:01:29.540
draining Polymarket wallets, and the GitLab upgrade
00:01:29.540 --> 00:01:33.280
that didn't fix Alpine. Then we have a quick
00:01:33.280 --> 00:01:36.420
lightning round and a human closer about proving
00:01:36.420 --> 00:01:39.343
something didn't happen. Let's get into it.
00:01:43.833 --> 00:01:47.319
First up, a follow-up. And this one comes with dates.
00:01:47.579 --> 00:01:51.579
In episode 67, we covered GitHub's workflow execution
00:01:51.579 --> 00:01:55.640
protections. GitHub said then that starting November
00:01:55.640 --> 00:01:59.579
2nd, public repos without their own event policy
00:01:59.579 --> 00:02:03.420
would have pull request target disabled by default.
00:02:03.579 --> 00:02:07.359
That date has moved. Permanent enforcement now
00:02:07.359 --> 00:02:10.960
starts December 2nd. Before that, there are two
00:02:10.960 --> 00:02:14.960
brownouts: Monday, November 9th, and Monday,
00:02:15.080 --> 00:02:20.319
November 16th, both from 4 to 6 p.m. UTC. That's
00:02:20.319 --> 00:02:23.719
11 a.m. to 1 p.m. Eastern. During those two
00:02:23.719 --> 00:02:27.639
hours, blocked runs fail with an error annotation,
00:02:27.740 --> 00:02:31.599
and the workflow doesn't execute. After each
00:02:31.599 --> 00:02:34.960
window, the rule goes back to evaluate mode,
00:02:35.099 --> 00:02:38.699
where workflows run with a warning. Quickly,
00:02:38.919 --> 00:02:42.199
what the trigger does. Pull request target runs
00:02:42.199 --> 00:02:46.159
in the context of the base repository with access
00:02:46.159 --> 00:02:49.719
to that repo's secrets. That's fine for labeling
00:02:49.719 --> 00:02:53.360
a pull request. It goes wrong when the workflow
00:02:53.360 --> 00:02:57.080
checks out the fork's code and runs it, because
00:02:57.080 --> 00:03:00.979
now a stranger's code is running with your secrets.
00:03:01.400 --> 00:03:04.740
It's public repos only. Private and internal
00:03:04.740 --> 00:03:08.500
repos aren't affected. And neither is GitHub's
00:03:08.500 --> 00:03:12.330
Enterprise Server. So before Monday, open Insights
00:03:12.330 --> 00:03:15.870
and see which runs the default rule would block.
00:03:16.129 --> 00:03:19.270
If you don't need the trigger, let the rule block
00:03:19.270 --> 00:03:23.509
it. If you do, allow it explicitly in an Actions
00:03:23.509 --> 00:03:27.569
event policy, and use workflow file targeting
00:03:27.569 --> 00:03:31.270
so only the workflows that need it get it. If
00:03:31.270 --> 00:03:34.389
your open source project has a bot that labels
00:03:34.389 --> 00:03:38.270
or comments on pull requests, Monday at 11 Eastern
00:03:38.270 --> 00:03:41.956
is when you find out which trigger it uses.
00:03:46.450 --> 00:03:50.460
Next, a WordPress bug where the dangerous moment is
00:03:50.460 --> 00:03:54.259
the update itself. On October 28th, WordPress
00:03:54.259 --> 00:03:59.879
shipped 7.1.4 and backported the fix to every
00:03:59.879 --> 00:04:04.240
branch back to 4.7. Every version from 4.7
00:04:04.240 --> 00:04:09.439
through 7.1.3 is affected. On Tuesday, CISA
00:04:09.439 --> 00:04:12.340
added it to its Known Exploited Vulnerabilities
00:04:12.340 --> 00:04:19.439
catalog as CVE-2026-119977, with a federal
00:04:19.439 --> 00:04:23.100
deadline of November 6th, the day this episode
00:04:23.100 --> 00:04:26.300
comes out. Here's how it works. When WordPress
00:04:26.300 --> 00:04:30.500
does a core update with a complete package, it
00:04:30.500 --> 00:04:34.079
unpacks the new files into a predictable folder
00:04:34.079 --> 00:04:39.519
under wp-content/upgrade. That copy includes
00:04:39.519 --> 00:04:43.250
the WordPress installer. And on a lot of servers,
00:04:43.410 --> 00:04:46.870
that folder is reachable from the web. So for
00:04:46.870 --> 00:04:50.129
a moment, a fresh installer is sitting at the
00:04:50.129 --> 00:04:54.050
URL anyone can request. An unauthenticated attacker
00:04:54.050 --> 00:04:58.069
who requests it inside that window can work their
00:04:58.069 --> 00:05:02.129
way to running PHP on your server. In lab testing,
00:05:02.430 --> 00:05:06.310
that window lasts only seconds. So attackers
00:05:06.310 --> 00:05:10.350
don't wait for it. They poll. Wordfence blocked
00:05:10.350 --> 00:05:15.009
almost 63,000 attacks on this in a single day
00:05:15.009 --> 00:05:17.949
last week. What makes this one uncomfortable
00:05:17.949 --> 00:05:21.610
is that it punishes the right habit. Keeping
00:05:21.610 --> 00:05:25.250
WordPress current is exactly
Load full transcript
what opens the window.
00:05:25.509 --> 00:05:30.629
So update to 7.1.4 or the patched release on
00:05:30.629 --> 00:05:34.389
your branch and confirm it actually landed on
00:05:34.389 --> 00:05:37.629
the sites you assume are auto-updating. Then
00:05:38.060 --> 00:05:42.079
check your logs for requests to PHP files under
00:05:42.079 --> 00:05:46.819
wp-content/upgrade. And look for admin
00:05:46.819 --> 00:05:51.660
accounts, plugins, or PHP files you don't recognize.
00:05:52.120 --> 00:05:55.420
And if you can, stop the web server from serving
00:05:55.420 --> 00:05:58.571
anything out of that upgrade folder at all.
00:06:03.069 --> 00:06:07.069
Third, a monitoring script that turned into a wallet
00:06:07.069 --> 00:06:11.069
drainer. Polymarket says that on October 29th,
00:06:11.230 --> 00:06:13.829
a third-party vendor's performance monitoring
00:06:13.829 --> 00:06:18.269
script, which loaded on polymarket.com, was
00:06:18.269 --> 00:06:21.649
replaced with one built to target its users.
00:06:21.930 --> 00:06:25.389
For anyone with the site open during a short
00:06:25.389 --> 00:06:28.430
window that afternoon, it used their session
00:06:28.430 --> 00:06:32.069
to sell their positions and move their funds
00:06:32.069 --> 00:06:35.949
out. Outside estimates put it at about $2.7
00:06:35.949 --> 00:06:40.810
million from around 700 wallets and Polymarket
00:06:40.810 --> 00:06:44.449
says it's refunding people. The vendor was DebugBear,
00:06:44.629 --> 00:06:48.110
which makes real user monitoring for page speed.
00:06:48.490 --> 00:06:52.209
DebugBear says its DNS records were compromised
00:06:52.209 --> 00:06:55.569
through its domain registrar. It says the attack
00:06:55.569 --> 00:06:59.529
targeted one specific customer. So DebugBear's
00:06:59.529 --> 00:07:03.089
code didn't have to change at all. Polymarket's
00:07:03.089 --> 00:07:07.370
pages loaded a script from DebugBear's hostname
00:07:07.370 --> 00:07:10.980
and whoever controls the DNS from that hostname
00:07:10.980 --> 00:07:14.920
decides what code comes back. That code then
00:07:14.920 --> 00:07:18.959
runs inside Polymarket's page with the user's session.
00:07:19.560 --> 00:07:23.379
HTTPS doesn't stop it either because whoever
00:07:23.379 --> 00:07:27.459
controls a domain's DNS can usually get a valid
00:07:27.459 --> 00:07:30.379
certificate from it. A content security policy
00:07:30.379 --> 00:07:34.319
that allows the vendor's domain still allows
00:07:34.319 --> 00:07:38.480
it. Subresource integrity does help. The browser
00:07:38.480 --> 00:07:42.279
checks the script against a hash and won't run
00:07:42.279 --> 00:07:46.199
it if the file changed, even at the same URL.
00:07:46.600 --> 00:07:49.319
The trade-off is that the vendor's legitimate
00:07:49.319 --> 00:07:52.540
updates get blocked too, until you update the
00:07:52.540 --> 00:07:55.959
hash, which is why teams pin a fixed version.
00:07:56.519 --> 00:08:00.360
Polymarket's fix includes explicit version pinning,
00:08:00.379 --> 00:08:03.680
along with a stronger content security policy
00:08:03.680 --> 00:08:07.889
and sandboxing. So, two things to check. First,
00:08:08.069 --> 00:08:11.910
list every third-party script on pages where
00:08:11.910 --> 00:08:15.470
your users are signing in. And ask whether a
00:08:15.470 --> 00:08:19.250
page speed script needs to be there at all. Second,
00:08:19.410 --> 00:08:23.529
look at your own registrar account. Turn on registrar
00:08:23.529 --> 00:08:28.069
lock, put strong MFA on it, and alert on DNS
00:08:28.069 --> 00:08:31.329
changes. It's often the least watched account
00:08:31.329 --> 00:08:34.400
that can redirect everything you run.
00:08:39.005 --> 00:08:42.899
Fourth, a follow-up that changes advice we gave. In
00:08:42.899 --> 00:08:50.399
episode 72, we covered CVE-2026-97229, the CVSS
00:08:50.399 --> 00:08:55.379
10 GitLab flaw in how CI configuration handles
00:08:55.379 --> 00:08:58.639
regular expressions. If you ran GitLab yourself,
00:08:58.639 --> 00:09:06.620
we said upgrade to 19.4.2, 19.3.5, or 19.2.8.
00:09:06.620 --> 00:09:10.870
Now Alpine Linux has published what happened
00:09:10.870 --> 00:09:14.809
when an attacker came after its GitLab and why
00:09:14.809 --> 00:09:18.669
upgrading wasn't enough there. The bug is actually
00:09:18.669 --> 00:09:22.809
in Onigmo, the regular expression engine built
00:09:22.809 --> 00:09:27.429
into Ruby. GitLab's October 12th fix is a set
00:09:27.429 --> 00:09:31.370
of patches to the Ruby that GitLab bundles in
00:09:31.370 --> 00:09:35.129
its own packages and container images. Upstream
00:09:35.129 --> 00:09:39.210
Ruby didn't ship the fix until October 23rd.
00:09:39.370 --> 00:09:42.769
Alpine runs GitLab from its own container image
00:09:42.769 --> 00:09:47.549
on standard Ruby. On October 20th and 21st, a
00:09:47.549 --> 00:09:52.250
newly registered account sent crafted CI configs
00:09:52.250 --> 00:09:56.450
that crashed GitLab workers eight times. In four
00:09:56.450 --> 00:09:59.870
of those crashes, the attacker got partway through
00:09:59.870 --> 00:10:03.330
the exploit. Alpine upgraded GitLab on October
00:10:03.330 --> 00:10:07.710
22nd. On the 26th, they found it was still vulnerable,
00:10:08.289 --> 00:10:11.789
because their image never had GitLab's Ruby patch.
00:10:12.169 --> 00:10:15.629
They took it offline, rebuilt it, reset every
00:10:15.629 --> 00:10:19.769
password, and reopened on October 29th. If you
00:10:19.769 --> 00:10:23.269
run GitLab from source or from a container image
00:10:23.269 --> 00:10:27.169
you build yourself, upgrading GitLab isn't enough.
00:10:27.409 --> 00:10:34.509
You also need Ruby 3.3.13, 3.4.12, 4.0.9,
00:10:34.610 --> 00:10:40.429
or later. GitLab's official packages and images already carry the patch.
00:10:40.769 --> 00:10:44.610
And, since this is a Ruby bug, anything else
00:10:44.610 --> 00:10:47.929
of yours that compiles regular
expressions from
00:10:47.929 --> 00:10:51.970
untrusted input needs the new Ruby too. A version
00:10:51.970 --> 00:10:55.210
number tells you what you installed. It doesn't
00:10:55.210 --> 00:10:57.745
always tell you what's inside it.
00:11:05.150 --> 00:11:10.730
Quick lightning round. First, Alpine again. On Thursday, OpenSSH
00:11:10.730 --> 00:11:16.509
shipped an emergency release, 10.6p2, for a
00:11:16.509 --> 00:11:21.009
login bypass on systems built on musl with
00:11:21.009 --> 00:11:25.710
PAM turned off, which is how Alpine's OpenSSH
00:11:25.710 --> 00:11:29.309
server package is built. With password logins
00:11:29.309 --> 00:11:33.309
on, the default, an attacker can get into an account
00:11:33.309 --> 00:11:37.929
with no password set, like a key-only user, by
00:11:37.929 --> 00:11:41.330
sending a very long password, as long as another
00:11:41.330 --> 00:11:45.490
account on the box has a real one. Update, then
00:11:45.490 --> 00:11:49.470
restart sshd, since the upgrade doesn't restart
00:11:49.470 --> 00:11:53.129
it, or turn password authentication off. Second,
00:11:53.289 --> 00:11:57.509
if you run Debian or Ubuntu, update APT itself.
00:11:57.929 --> 00:12:03.830
CVE-2026-109518 lets anyone who can tamper
00:12:03.830 --> 00:12:08.470
with the HTTP traffic between APT and the archive,
00:12:09.029 --> 00:12:13.070
like a malicious mirror or proxy, swap a package
00:12:13.070 --> 00:12:17.490
after APT has verified it. And packages install
00:12:17.490 --> 00:12:21.450
as root. It needs download retries, which are
00:12:21.450 --> 00:12:26.529
on by default from Debian 12 and Ubuntu 22.04.
00:12:26.970 --> 00:12:31.230
Fixes shipped Wednesday, so update APT on your
00:12:31.230 --> 00:12:34.950
host and in the images you build. And third,
00:12:35.190 --> 00:12:39.759
.NET 8 and .NET 9 reach end of support next Tuesday,
00:12:39.980 --> 00:12:43.460
November 10th. Last week, we told you to patch
00:12:43.460 --> 00:12:47.639
.NET for the Brotli bug. After Tuesday, the next
00:12:47.639 --> 00:12:52.120
one like it just stays open. .NET 10 is the long-term
00:12:52.120 --> 00:12:56.321
support release, supported through November 2028.
00:13:04.718 --> 00:13:07.309
The human closer this week comes back to
00:13:07.309 --> 00:13:10.870
Alpine, and one sentence in their write-up:
00:13:11.029 --> 00:13:14.090
"We have found no evidence that it succeeded,
00:13:14.700 --> 00:13:18.419
but we cannot rule it out." That's an honest sentence,
00:13:18.639 --> 00:13:22.519
and it's harder to write than it sounds. Alpine
00:13:22.519 --> 00:13:26.279
had crash logs. They knew the attacker got
00:13:26.279 --> 00:13:30.039
partway through the exploit four times. What
00:13:30.039 --> 00:13:33.860
they couldn't do was prove that some other, quieter
00:13:33.860 --> 00:13:37.980
attempt didn't work. A successful exploit would
00:13:37.980 --> 00:13:42.419
leave no crash and no trace in the web logs.
00:13:43.049 --> 00:13:46.210
Some logs from the day of the attack were already
00:13:46.210 --> 00:13:49.850
gone and nobody was logging outbound network
00:13:49.850 --> 00:13:53.730
connections, not Alpine and not their hosting
00:13:53.730 --> 00:13:58.090
provider. So they rebuilt from scratch and they
00:13:58.090 --> 00:14:01.529
checked their releases against records the attacker
00:14:01.529 --> 00:14:04.769
couldn't have touched, a developer's own clone,
00:14:04.990 --> 00:14:09.070
and all 45 of their build servers. Look at what
00:14:09.070 --> 00:14:12.860
they are changing afterward. Most of it is evidence.
00:14:12.860 --> 00:14:17.240
Log outbound connections. Keep logs longer than
00:14:17.240 --> 00:14:21.840
14 days. Sign release tags, so anyone can verify
00:14:21.840 --> 00:14:26.019
a release without trusting GitLab. Now hold that
00:14:26.019 --> 00:14:29.779
up against the rest of this week. If the WordPress
00:14:29.779 --> 00:14:33.159
race hit one of your sites in those few seconds,
00:14:33.159 --> 00:14:37.019
would your logs show it? If a third-party script
00:14:37.019 --> 00:14:41.419
on your login page changed for an hour on a Thursday
00:14:41.419 --> 00:14:45.340
afternoon, would anyone know? If someone changed
00:14:45.340 --> 00:14:49.100
your DNS at the registrar, how long until you
00:14:49.100 --> 00:14:52.600
noticed? None of those questions are about stopping
00:14:52.600 --> 00:14:55.860
the attack. They are about whether you can say
00:14:55.860 --> 00:14:59.580
with a straight face that it didn't happen. Even
00:14:59.580 --> 00:15:04.340
CISA's WordPress entry asks for forensic triage,
00:15:04.379 --> 00:15:07.940
not just a patch. That's it for this week of
00:15:07.940 --> 00:15:10.700
Ship It Weekly. Follow or subscribe wherever
00:15:10.700 --> 00:15:13.240
you are watching or listening. You can find the
00:15:13.240 --> 00:15:16.799
weekly story list and source links at oncallbrief.com
00:15:16.799 --> 00:15:21.379
and past episodes and show notes at shipitweekly.fm.
00:15:21.379 --> 00:15:25.960
In this episode, we covered GitHub's pull
00:15:25.960 --> 00:15:29.139
request target brownouts, the WordPress update
00:15:29.139 --> 00:15:33.960
race on CISA's exploited list, the hijacked DebugBear
00:15:33.960 --> 00:15:37.460
script that drained Polymarket wallets,
00:15:37.870 --> 00:15:41.850
and Alpine's GitLab upgrade that didn't include
00:15:41.850 --> 00:15:44.470
the fix. I'm Brian Teller from Teller's Tech.
00:15:44.610 --> 00:15:47.710
Thanks for listening. And remember, "no evidence"
00:15:47.710 --> 00:15:51.169
is only as good as the logs you kept.