👋 right on! I actually also have used containers as a key to my security layout before, but yeah you miss out on all the benefits of portage.

I was doing something crazy and actually running Gentoo inside each one! It was very difficult to stay up-to-date. But I basically had my host as barebones as possible and used LibVirt containers for everything, attempting to make a few templates that I could keep updated and base other VMs on. I was able to keep this up for about two years then I had to relax (was my main PC). But it was really secure, and it does work.

The benefit of encapsulation is that you have a lot of freedom inside each container, like install a different distro if you need to. Also as long as they are isolated you don’t need to worry as much about their individual security. But it’s still good to. I ran SELinux on the host and non-SELinux (but hardened) in the guests.

SELinux has a lot of advantages over users/groups, but I think the latter can be just as secure if you know what you’re doing. For example with SELinux you can prevent certain applications from accessing the network, or restrict access to certain ports, etc. It’s also useful for desktop environments where a lot of GUI apps run under one user- e.g. neither my main user nor any other program can access my keepassxc directory, only the keepassxc process (and root) can (even though the application is running under my main user). You can also restrict root quite a bit, especially if you compile in the option to prevent disabling SELinux at boot (I need to recompile my kernel to disable it).

But again while it is fun to learn, it is quite a pain and I’ve relaxed the setup on my new computer to use a different user for everything (including gui apps), which I think is secure enough for me. But this style relies on my ability to adhere to it, whereas with SELinux you can set it up to where you’re forced to

Like others have mentioned, SELinux could be a great addition. It can be a massive pain, but it’s really effective at locking things down (if configured properly).

However, the difficulty will depend on the distro. I use it with Gentoo, which has plenty of support/docs for it and provides policies for many packages. Although (when running strict policy types) I usually end up needing to adjust them or write my own.

Obviously Red Hat would be another good choice, but I haven’t tried it. Fedora also has good support, but I’ve only ever used the OOTB targeted policies.

That said, I’ve started relying on users/groups more often lately, since it really gets in the way of everything.

Ah, nice idea. I’ve tried a few different ways of doing this, and I think what you’re seeing is a discrepancy in how the compiler handles member access into incomplete types. It seems that, in your examples, the compiler is allowing -> decltype(f.private_msg) within the class, but I think it’s not selecting do_something outside of it because it uses decltype(t.private_msg). In my case, I’m not even able to do that within the class.

For example, since I’m not able to use decltype(f.private_msg) inside the class, I’m using decltype(private_msg) instead, which causes an error at the do_something declaration related to incomplete type (presumably because of the t.private_msg usage):

// candidate template ignored; member access into incomplete type
template 〈class T〉 auto do_something(T ＆t) -> decltype(t.private_msg);
class Foo {
        const char *private_msg = "You can't touch me!";
        friend auto do_something〈〉(Foo ＆f) -> decltype(private_msg);
};
template 〈〉 auto do_something(Foo ＆f) -> decltype(f.private_msg) {
        return f.private_msg;
}

My reasoning is that removing the t.private_msg from the declaration works:

template 〈class Ret, class T〉 auto do_something(T ＆t) -> Ret;
class Foo {
        const char *private_msg = "You can't touch me!";
        friend auto do_something〈〉(Foo ＆f) -> decltype(private_msg);
};
template 〈〉 auto do_something(Foo ＆f) -> decltype(f.private_msg) {
        return f.private_msg;
}
static Foo foo{};
// this works, but Ret cannot be deduced and must be specified somehow:
static auto something = do_something〈const char*〉(foo);

The reason your second example works is because the friend template inside the class acts as a template declaration rather than a specialization, which isn’t specialized until after Foo is complete:

// the do_something inside Foo is a declaration, meaning this isn't used
// template 〈class T〉
// auto do_something(T ＆t) -> decltype(t.private_msg);
class Foo {
        const char *private_msg = "You can't touch me!";
        template 〈class T〉 // t.private_msg is allowed because T is not Foo yet
        friend auto do_something(T ＆t) -> decltype(t.private_msg);
};
template 〈〉 auto do_something(Foo ＆f) -> decltype(f.private_msg) {
        return f.private_msg;
}

I think the issue is that Foo is incomplete when you’re declaring the friend, so I think it’s impossible. I just tried it and g++ ignores the target candidate due to “member access into incomplete type”, which makes sense since std::begin is already defined and calls .begin(). The closest you can get is to use another friend to expose arr and overload std::begin manually, but that’s a bit silly 😅

I usually use Awk to do the heavy lifting within my Bash scripts (e.g. arg parsing, filtering, stream transformation), or I’ll embed a Node.JS script for anything more advanced. In some cases, I’ll use eval to process generated bash syntax, or I’ll pipe into sh (which can be a good way to set up multiprocessing). I’ve also wanted to try zx, but I generally just stick to inlining since it saves a dependency.

I started by writing small scripts to automate things, but really got into it after learning how fun it can be to make the computer do stuff. I also see it as a kind of creative outlet, but in general I just want to learn how to fix anything in software if I’m not satisfied with how it works.

I use LUKS-encrypted LVM volumes to store everything (and transfer via SSH or HTTPS), but would use GPG if I needed to encrypt individual files.

Ah good to know! Will try that if I ever run into issues, thanks

I’ve been using it for years and I think it’s great. Currently on a 6 Pro. It’s true that some apps don’t work without Google Play services, but GrapheneOS has the option to install the google stuff in a sandbox, so you shouldn’t run into any issues if you do that. Personally, I don’t use Play services unless I need to, and use Aurora store for any apps that aren’t on F-Droid.

In any case, you can always revert to stock or try another OS

Edit: as faede has pointed out, it appears that Google Wallet has issues. Also, the usage docs mention issues with banking apps in general, so that’s something to consider

np! Hope it helps; it’s a big pain but I do think it’s pretty secure if configured correctly

Awesome! Here are a few things that come to mind:

Make sure you have some aliases/functions for common operations:

• audit2allow -a to view audit violations (or -d for dmesg audits)
  • also -r to add a requires statement for module construction
• restorecon -Rv to recursively apply file contexts from policy (or -FRv to also apply user context)
• rm -f /var/log/audit/audit.log.*; >/var/log/audit/audit.log to clear audit logs
  • note: sometimes lots of logfiles (audit.log.1, etc.) collect, slowing down audit2allow
• chown -R user:user PATH; chcon -R -u user_u PATH to recursively change labels to user
  • could be generalized for arbitrary Linux/SELinux users
• semanage fcontext -a -t TYPE PATH -s $SEUSER to add a custom file context to the policy
  • e.g. semanage fcontext -a -t "user_secrets_t" "/home/[^/]+/.secrets(/.*)?" -s user_u
  • I’ve had better luck with this approach than the standard method of creating a .fc file, but in any case a custom policy is needed to create custom types
• semanage fcontext -d PATH to remove a custom file context
• semanage fcontext -lC to list custom file contexts
• semodule -DB to rebuild policy with all dontaudit rules disabled
  • often, something will not work, but audit2allow doesn’t show anything
• semodule -B to rebuild policy (with dontaudit rules)
• semodule -i MODULE.pp to install a module
• semodule -r MODULE to remove a module

Also a few scripts for policy creation and management are essential. There are two basic approaches to policy creation: modules and policy modules.

Modules: can be used to modify AVC rules and are pretty simple

# a violation has occurred that you want to allow or dontaudit
echo "module my_allow 1.0;" > my_allow.te
audit2allow -ar >> my_allow.te

# verify that my_allow.te has what you expect
cat my_allow.te

# build and install the module (replace mcs with whatever policy you are using)
make -f /usr/share/selinux/mcs/include/Makefile my_allow.pp
semodule -i my_allow.pp

# clear audit logs
rm -f /var/log/audit/audit.log.*; >/var/log/audit/audit.log

Policy modules: can do anything, but are complicated, and the tools for creating them are mostly based on Red Hat.

Creating a new type:

# generate foo.fc, foo.if, and foo.te
sepolicy generate --newtype -t foo_var_lib_t -n foo

# note: see sepolicy-generate(8); sepolicy generate only supports the following
#       type suffixes, but its output files can be adapted to your use case
# _tmp_t
# _unit_file_t
# _var_cache_t
# _var_lib_t
# _var_log_t
# _var_run_t
# _var_spool_t
# _port_t

# modify the .fc file with the desired file contexts, for example (with s0 for mcs)
# /path/to/context/target	--	gen_context(system_u:object_r:type_t,s0)
#
# note: the "--" matches regular files, -d for directories, -c for character
#       devices, -l for symbolic links, -b for block devices, or can be omitted
#       to match anything. also, as mentioned before, I often have better luck
#       with `semanage fcontext`, especially for user directories
vi foo.fc

# build and install the policy module
make -f /usr/share/selinux/mcs/include/Makefile foo.pp
semodule -i foo.pp

# use restorecon to adjust the file contexts of any paths you have 

# by default, all operations involving this type will be denied
# (and are sometimes not audited)
semodule -DB # --disable_dontaudit
# ... use the type, collect violations ...
audit2allow -ar >> foo.te
# if dontaudit is disabled, you'll likely have a lot things to remove from here
vi foo.te

# ... repeat until rules regarding type are fully defined

Creating a new application type:

# sepolicy-generate is made for Red Hat,
# but you can use --application to get started

# creates a bunch of files that define bar_t and bar_exec_t
sepolicy generate --application -n bar [-u USER] CMD

# remove the line making the app permissive (up to you, but
# I prefer using audit violations to define the permissions)
perl -i -00 -pe 's/^permissive bar_t;\n\n//g' bar.te

# ensure that the file bar_exec_t file context points to the right bin:
vi bar.fc

# build and install the policy module
make -f /usr/share/selinux/mcs/include/Makefile bar.pp
semodule -i bar.pp

# ... use the application, update AVC rules, repeat ...

If your target application is interpreted, you’ll need to write a custom C program that launches the interpreter in a specific context, then write your policy around that application. For example, you should execv something like this: /usr/bin/runcon -u user_u -t my_script_t /bin/bash PROG.

For vegetables I throw everything into a big stew with a lot of different things (kale, broccoli, cauliflower, onion, potato, mushrooms, tofu, garlic, beans), lots of hot sauce, seasoning, olive oil, etc. and eat the same thing every day, for the most part. I don’t eat enough fruit but I do have a handful of dried fruit with oats every day

Totally, props on taking it on as your first distro! Haha, yeah a week of pain sounds about right. My last Gentoo setup took an entire month (off and on), but I was doing something crazy (Qubes-like, every application in its own Gentoo VM, strict SELinux on host and guests)… ended up ditching that because I got comfortable enough with SELinux to write stronger policies for everything important, which is good enough for me.

I had the benefit of using other distros before trying Gentoo, so my first attempt at it wasn’t so bad (but still took two full days). It’s definitely taught me way more than any other distro, including Arch (although Arch was a very good stepping stone). I don’t think I could go back to anything else at this point

Yep! Gotta love the flexibility of it

Ah gotcha, just asking because I’ve never used it before. Good to know that Gentoo supports hardening it

Oh good to know! Thanks for the tips. What do you like about musl over glibc?

I would look into Gentoo’s Hardened + SELinux profile if you want good security in a standard system, but as others have mentioned QubesOS is probably the most secure option OOTB (but it is very limiting). SELinux is pretty difficult to use but it’s really effective, and there is good information about it on the Gentoo wiki. Not sure what exactly goes into their hardened profile but I know it implements at least some of the suggestions listed on that site (like hardened compilation flags). Also it’s probably more vulnerable to 0-day attacks than Qubes, since it uses up-to-date software. But it’s really flexible, and learning SELinux is useful

deleted by creator

Haha yeah, nicely put. I do enjoy the content, mostly because I’ve been following these creators for some time, and it’s hard to find a replacement for it… there is a lot of great content there, but it makes me feel gross using it. And same, I had no problem finding an alternative for Reddit (this), probably because I was not very attached to individual creators there.

I’m hoping a decentralized solution gains traction, but in the meantime I’ve been trying to limit the amount of information I share with the platform. I’m not actively trying to restrict my usage (most of that was achieved when I stopped using an account), but maybe it’s a good idea to do so. I mostly use it when eating or going to sleep, and there are better ways to occupy that time.

Major bugs usually get fixed pretty quickly- I always check the GitHub to make sure I have the latest version when I have issues. And Invidious can work as an alternative most of the time, but some instances work better than others