The previous hash was too short and caused evaluation-time errors like:
invalid SRI hash 'sha256-AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA='
Additionally, since the fact that this is broken implies that nobody
could have been using it, "SRI" is a bit of a vague and obscure term,
`fakeSriHash` would be somewhat of a mouthful, and the relevant fetcher
parameters are just called `hash`, rename it to `fakeHash`.
This will avoid breaking the build whenever a non-major kernel update
happens. In the update script, we map each kernel version to the latest
patch for the latest kernel version less than or equal to what we
have packaged.
7-Zip's RAR implementation is built on the non-free UnRAR source code;
DOC/License.txt says:
Licenses for files are:
1) CPP/7zip/Compress/Rar* files: GNU LGPL + unRAR restriction
2) All other files: GNU LGPL
The GNU LGPL + unRAR restriction means that you must follow both
GNU LGPL rules and unRAR restriction rules.
...
unRAR restriction
-----------------
The decompression engine for RAR archives was developed using source
code of unRAR program.
All copyrights to original unRAR code are owned by Alexander Roshal.
The license for original unRAR code has the following restriction:
The unRAR sources cannot be used to re-create the RAR compression algorithm,
which is proprietary. Distribution of modified unRAR sources in separate form
or as a part of other software is permitted, provided that it is clearly
stated in the documentation and source comments that the code may
not be used to develop a RAR (WinRAR) compatible archiver.
The unrar licensing is [infamously restrictive and non-free][fedora];
it's inappropriate for us to keep the RAR support while labelling the
package as free software (and indeed there's a commented-out line
pointing out that the current `meta.license` is false). Unfortunately,
the 7-Zip upstream seems uninterested in replacing the code with a
freely-licensed alternative (see [7-Zip ticket #1229][7zip]).
[fedora]: https://fedoraproject.org/wiki/Licensing:Unrar
[7zip]: https://sourceforge.net/p/sevenzip/feature-requests/1229/
An alternative solution would be to mark the p7zip package as non-free
instead; I decided not to because its other functionality (especially
`.7z` support) is freely-licensed and useful, and there are free
software alternatives for extracting RAR files (e.g. in nixpkgs there's
`archiver`, which is written in a memory-safe language, and `unar`,
which at least doesn't have two patches for CVEs that haven't been
addressed upstream...).
I checked that `7z(1)` fails gracefully on `.rar` files now:
emily@renko ~/tmp> curl -L -O https://www.philippwinterberg.com/download/example.rar
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 5715k 100 5715k 0 0 6716k 0 --:--:-- --:--:-- --:--:-- 6716k
emily@renko ~/tmp> 7z x example.rar
7-Zip [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_CA.UTF-8,Utf16=on,HugeFiles=on,64 bits,8 CPUs x64)
Scanning the drive for archives:
1 file, 5853119 bytes (5716 KiB)
Extracting archive: example.rar
ERROR: example.rar
Can not open the file as archive
Can't open as archive: 1
Files: 0
Size: 0
Compressed: 0
Shimming out the Let's Encrypt domain name to reuse client configuration
doesn't work properly (Pebble uses different endpoint URL formats), is
recommended against by upstream,[1] and is unnecessary now that the ACME
module supports specifying an ACME server. This commit changes the tests
to use the domain name acme.test instead, and renames the letsencrypt
node to acme to reflect that it has nothing to do with the ACME server
that Let's Encrypt runs. The imports are renamed for clarity:
* nixos/tests/common/{letsencrypt => acme}/{common.nix => client}
* nixos/tests/common/{letsencrypt => acme}/{default.nix => server}
The test's other domain names are also adjusted to use *.test for
consistency (and to avoid misuse of non-reserved domain names such
as standalone.com).
[1] https://github.com/letsencrypt/pebble/issues/283#issuecomment-545123242
Co-authored-by: Yegor Timoshenko <yegortimoshenko@riseup.net>
This was added in aade4e577b, but the
implementation of the ACME module has been entirely rewritten since
then, and the test seems to run fine on AArch64.
These now depend on an external patch set; add them to the release tests
to ensure that the build doesn't break silently as new kernel updates
are merged.
linux-hardened sets kernel.unprivileged_userns_clone=0 by default; see
anthraxx/linux-hardened@104f44058f.
This allows the Nix sandbox to function while reducing the attack
surface posed by user namespaces, which allow unprivileged code to
exercise lots of root-only code paths and have lead to privilege
escalation vulnerabilities in the past.
We can safely leave user namespaces on for privileged users, as root
already has root privileges, but if you're not running builds on your
machine and really want to minimize the kernel attack surface then you
can set security.allowUserNamespaces to false.
Note that Chrome's sandbox requires either unprivileged CLONE_NEWUSER or
setuid, and Firefox's silently reduces the security level if it isn't
allowed (see about:support), so desktop users may want to set:
boot.kernel.sysctl."kernel.unprivileged_userns_clone" = true;
As far as I can tell, this has never defaulted to on upstream, and our
common kernel configuration doesn't turn it on, so the attack surface
reduction here is somewhat homeopathic.
This is an updated version of the former upstream,
https://github.com/AndroidHardeningArchive/linux-hardened, and provides
a minimal set of additional hardening patches on top of upstream.
The patch already incorporates many of our hardened profile defaults,
and releases are timely (Linux 5.5.15 and 5.6.2 were released on
2020-04-02; linux-hardened patches for them came out on 2020-04-03 and
2020-04-04 respectively).
Dropbear lags behind OpenSSH significantly in both support for modern
key formats like `ssh-ed25519`, let alone the recently-introduced
U2F/FIDO2-based `sk-ssh-ed25519@openssh.com` (as I found when I switched
my `authorizedKeys` over to it and promptly locked myself out of my
server's initrd SSH, breaking reboots), as well as security features
like multiprocess isolation. Using the same SSH daemon for stage-1 and
the main system ensures key formats will always remain compatible, as
well as more conveniently allowing the sharing of configuration and
host keys.
The main reason to use Dropbear over OpenSSH would be initrd space
concerns, but NixOS initrds are already large (17 MiB currently on my
server), and the size difference between the two isn't huge (the test's
initrd goes from 9.7 MiB to 12 MiB with this change). If the size is
still a problem, then it would be easy to shrink sshd down to a few
hundred kilobytes by using an initrd-specific build that uses musl and
disables things like Kerberos support.
This passes the test and works on my server, but more rigorous testing
and review from people who use initrd SSH would be appreciated!
Previously, the NixOS ACME module defaulted to using P-384 for
TLS certificates. I believe that this is a mistake, and that we
should use P-256 instead, despite it being theoretically
cryptographically weaker.
The security margin of a 256-bit elliptic curve cipher is substantial;
beyond a certain level, more bits in the key serve more to slow things
down than add meaningful protection. It's much more likely that ECDSA
will be broken entirely, or some fatal flaw will be found in the NIST
curves that makes them all insecure, than that the security margin
will be reduced enough to put P-256 at risk but not P-384. It's also
inconsistent to target a curve with a 192-bit security margin when our
recommended nginx TLS configuration allows 128-bit AES. [This Stack
Exchange answer][pornin] by cryptographer Thomas Pornin conveys the
general attitude among experts:
> Use P-256 to minimize trouble. If you feel that your manhood is
> threatened by using a 256-bit curve where a 384-bit curve is
> available, then use P-384: it will increases your computational and
> network costs (a factor of about 3 for CPU, a few extra dozen bytes
> on the network) but this is likely to be negligible in practice (in a
> SSL-powered Web server, the heavy cost is in "Web", not "SSL").
[pornin]: https://security.stackexchange.com/a/78624
While the NIST curves have many flaws (see [SafeCurves][safecurves]),
P-256 and P-384 are no different in this respect; SafeCurves gives
them the same rating. The only NIST curve Bernstein [thinks better of,
P-521][bernstein] (see "Other standard primes"), isn't usable for Web
PKI (it's [not supported by BoringSSL by default][boringssl] and hence
[doesn't work in Chromium/Chrome][chromium], and Let's Encrypt [don't
support it either][letsencrypt]).
[safecurves]: https://safecurves.cr.yp.to/
[bernstein]: https://blog.cr.yp.to/20140323-ecdsa.html
[boringssl]: https://boringssl.googlesource.com/boringssl/+/e9fc3e547e557492316932b62881c3386973ceb2
[chromium]: https://bugs.chromium.org/p/chromium/issues/detail?id=478225
[letsencrypt]: https://letsencrypt.org/docs/integration-guide/#supported-key-algorithms
So there's no real benefit to using P-384; what's the cost? In the
Stack Exchange answer I linked, Pornin estimates a factor of 3×
CPU usage, which wouldn't be so bad; unfortunately, this is wildly
optimistic in practice, as P-256 is much more common and therefore
much better optimized. [This GitHub comment][openssl] measures the
performance differential for raw Diffie-Hellman operations with OpenSSL
1.1.1 at a whopping 14× (even P-521 fares better!); [Caddy disables
P-384 by default][caddy] due to Go's [lack of accelerated assembly
implementations][crypto/elliptic] for it, and the difference there seems
even more extreme: [this golang-nuts post][golang-nuts] measures the key
generation performance differential at 275×. It's unlikely to be the
bottleneck for anyone, but I still feel kind of bad for anyone having
lego generate hundreds of certificates and sign challenges with them
with performance like that...
[openssl]: https://github.com/mozilla/server-side-tls/issues/190#issuecomment-421831599
[caddy]: https://github.com/caddyserver/caddy/blob/2cab475ba516fa725d012f53ca417c3e039607de/modules/caddytls/values.go#L113-L124
[crypto/elliptic]: https://github.com/golang/go/tree/2910c5b4a01a573ebc97744890a07c1a3122c67a/src/crypto/elliptic
[golang-nuts]: https://groups.google.com/forum/#!topic/golang-nuts/nlnJkBMMyzk
In conclusion, there's no real reason to use P-384 in general: if you
don't care about Web PKI compatibility and want to use a nicer curve,
then Ed25519 or P-521 are better options; if you're a NIST-fearing
paranoiac, you should use good old RSA; but if you're a normal person
running a web server, then you're best served by just using P-256. Right
now, NixOS makes an arbitrary decision between two equally-mediocre
curves that just so happens to slow down ECDH key agreement for every
TLS connection by over an order of magnitude; this commit fixes that.
Unfortunately, it seems like existing P-384 certificates won't get
migrated automatically on renewal without manual intervention, but
that's a more general problem with the existing ACME module (see #81634;
I know @yegortimoshenko is working on this). To migrate your
certificates manually, run:
$ sudo find /var/lib/acme/.lego/certificates -type f -delete
$ sudo find /var/lib/acme -name '*.pem' -delete
$ sudo systemctl restart 'acme-*.service' nginx.service
(No warranty. If it breaks, you get to keep both pieces. But it worked
for me.)
The current weekly setting causes every NixOS server to try to renew
its certificate at midnight on the dot on Monday. This contributes to
the general problem of periodic load spikes for Let's Encrypt; NixOS
is probably not a major contributor to that problem, but we can lead by
example by picking good defaults here.
The values here were chosen after consulting with @yuriks, an SRE at
Let's Encrypt:
* Randomize the time certificates are renewed within a 24 hour period.
* Check for renewal every 24 hours, to ensure the certificate is always
renewed before an expiry notice is sent out.
* Increase the AccuracySec (thus lowering the accuracy(!)), so that
systemd can coalesce the renewal with other timers being run.
(You might be worried that this would defeat the purpose of the time
skewing, but systemd is documented as avoiding this by picking a
random time.)
* source-han-sans: 1.004R -> 2.001
* source-han-serif: switch to Super OTC
* source-han-mono: init at 1.002
The Source Han fonts now use shared package infrastructure, and the
Super OTC distributions, which unify the various scripts into a single
bundle file, improving automatic font selection and reducing overall
disk space usage. This also means that the Traditional
Chinese—Hong Kong language variant is now included.
The old package names including language are aliased to the Super OTC
bundle packages.
The primary motivation of this change was to allow third-party modules
to be used with OpenResty, but it also results in a significant
reduction of code duplication.
Emily