This replaces the mess of buildEnvs with a single Rust binary that
spits out a mostly-complete root filesystem for an fhsenv.
The main goal is to have includeClosures, as we want all of the
dependencies to be in the fhsenv to avoid Steam's (and others')
LD_LIBRARY_PATH shenanigans, but without 32-bit libraries leaking
into lib64 when a 64-bit package like mangohud depends on a 32-bit
version of itself.
We "fix" this by actually looking at the files and explicitly moving
32-bit stuff to $out/lib32. This could be avoided if we had recursive
Nix, or at least system info in exportReferencesGraph, but alas.
For some reason this also shrinks the fhsenvs massively, even though
there's currently no layout optimization (e.g. a package with paths
like lib/foo/{bar,baz} will produce two symlinks in the output, even
when it's more optimal to symlink lib/foo to $out/lib/foo directly).
`makeWrapper` is often used in these with `source "${makeWrapper}/nix-support/setup-hook"`
which causes `error: makeWrapper/makeShellWrapper must be in nativeBuildInputs` on cross.
Prior to this commit it was not possible to modify e.g. the list of ignored directories at all, however given that `buildFHSEnvBubblewrap` effectively uses a sandboxing tool (*bwrap*) I feel like this is a missed opportunity.
The code in nixpkgs already covers all the knobs that are required to get *Nix* itself to run inside bubblewrap, so why not allow users to make that additional modification?
While additional `ro_mounts` and such can be *added* to the bubblewrap invocation, the already mounted directories cannot be removed, and even if shadowed by e.g. a tmpfs mount, this would still allow something inside the sandbox to potentially unmount the tmpfs and access the data.
So what this change does is moving the snippet where custom code can be injected down by four lines so that users can actually modify those variables e.g. using `ignored+=( /home /srv /mnt /boot )`.
The only cases in which this would break is:
- someone using those variable names in `extraPreBwrapCmds` already and relying on them being overwritten; I would consider that chance slim, and the fix would be easy enough
- someone using a construct like `false && \` to disable the `ignored` initialisation and effectively working around this limitation; again the chances are slim (even though I know I'd be affected), and the fix would be easy enough (as this change makes the workaround needless anyway so it's an improvement)
Signed-off-by: benaryorg <binary@benary.org>
In preparation for the deprecation of `stdenv.isX`.
These shorthands are not conducive to cross-compilation because they
hide the platforms.
Darwin might get cross-compilation for which the continued usage of `stdenv.isDarwin` will get in the way
One example of why this is bad and especially affects compiler packages
https://www.github.com/NixOS/nixpkgs/pull/343059
There are too many files to go through manually but a treewide should
get users thinking when they see a `hostPlatform.isX` in a place where it
doesn't make sense.
```
fd --type f "\.nix" | xargs sd --fixed-strings "stdenv.is" "stdenv.hostPlatform.is"
fd --type f "\.nix" | xargs sd --fixed-strings "stdenv'.is" "stdenv'.hostPlatform.is"
fd --type f "\.nix" | xargs sd --fixed-strings "clangStdenv.is" "clangStdenv.hostPlatform.is"
fd --type f "\.nix" | xargs sd --fixed-strings "gccStdenv.is" "gccStdenv.hostPlatform.is"
fd --type f "\.nix" | xargs sd --fixed-strings "stdenvNoCC.is" "stdenvNoCC.hostPlatform.is"
fd --type f "\.nix" | xargs sd --fixed-strings "inherit (stdenv) is" "inherit (stdenv.hostPlatform) is"
fd --type f "\.nix" | xargs sd --fixed-strings "buildStdenv.is" "buildStdenv.hostPlatform.is"
fd --type f "\.nix" | xargs sd --fixed-strings "effectiveStdenv.is" "effectiveStdenv.hostPlatform.is"
fd --type f "\.nix" | xargs sd --fixed-strings "originalStdenv.is" "originalStdenv.hostPlatform.is"
```
GStreamer searches for plugins relative to its binary's location.
However, since bd97973ce0, it uses its *real* binary location, breaking the FHS.
Fixes#311004 (tested on Heroic and Lutris too).
The $DISPLAY variable has a format of [host]:num[.screen]. Previously,
the number would only be extracted properly if it had the form :num.
Allow all forms but correctly discard the unused parts.
`multiPaths` is defined via invalid operation on `null` value for
non-multilib environments. Noticed on `notesnook` evan failure as:
nix-repl> notesnook.fhsenv.multiPaths
253| passthru = {
254| inherit args baseTargetPaths targetPaths baseMultiPaths multiPaths ldconfig;
| ^
255| };
error: attempt to call something which is not a function but null
The change makes `multiPaths` a private variable.
when bubblewraps tries to link all the required files in etc from the
host to the fhs environment, it will re-create the /etc directory.
It will do so with `0700` permissions. This causes permissions issues
with non-root programs when they need to access configuration in the
environment /etc.
By mounting /etc as a tmpfs early, bwrap will make the directory `0755`
as expected.
It's useful to have access to these attributes from packages built with
buildFHSEnvBubblewrap, and it reduces the difference between FHS and
non-FHS packages.
'name' is already handled by runCommandLocal.
`buildFHSEnvBubblewrap { pname = ...; }` currently results in eval error
because args.name doesn't exist then. Fix it by only using args.name if
it exists.
It's useful to be able to introspect all packages which are available
in the fhsenv. I've renamed basePkgs and baseMultiPkgs to be
consistent with the naming scheme used for the bits that were
previously public — names ending in "Pkgs" are for functions, and
names ending in "Paths" are the results of those functions.
The implicit contract of buildFHSUserEnv was that it allows to run
software built for a typical GNU/Linux distribution (not NixOS) without
patching it (patchelf, autoPatchelfHook, etc.). Note that this does not
inherently imply running untrusted programs.
buildFHSUserEnv was implemented by using chroot and assembling a
standard-compliant FHS environment in the new root. As expected, this
did not provide any kind of isolation between the system and the
programs.
However, when it was later reimplemented using bubblewrap
(PR #225748), which *is* a security tool, several isolation features
involving detaches Linux namespaces were turned on by default.
This decision has introduced a number of breakages that are very
difficult to debug and trace back to this change.
For example: `unshareIPC` breaks software audio mixing in programs using
ALSA (dmix) and `unsharePID` breaks gdb,
Since:
1. the security features were enable without any clear threat model;
2. `buildFHSEnvBubblewrap` is supposed to be a drop-in replacement of
`buildFHSEnvChrootenv` (see the release notes for NixOS 23.05);
3. the change is breaking in several common cases (security does not
come for free);
4. the contract was not changed, or at least communicated in a clear
way to the users;
all security features should be turned off by default.
P.S. It would be useful to create a variant of buildFHSEnv that does
provide some isolation. This could unshare some namespaces and mount
only limited parts of the filesystem.
Note that buildFHSEnv mounts every directory in / under the new root, so
again, very little is gained by unsharing alone.
Before the change the following command did not work:
$ nix develop -i --impure --expr 'with import <nixpkgs> { system = "i686-linux"; }; (buildFHSUserEnv { name = "t"; targetPkgs = ps: with ps; [ libmpc stdenv.cc ]; }).env'
$ ld -lmpc -o a
ld: cannot find -lmpc: No such file or directory
It is expected to work as `NIX_LDFLAGS` does contain valid values:
$ echo $NIX_LDFLAGS
-L/usr/lib -L/usr/lib32
Note that for `gcc` it does work:
$ printf "int main(){}" | gcc -x c - -lmpc -o a
It happens because `HOST` role is enabled for `cc`:
$ echo $NIX_CC_WRAPPER_TARGET_HOST_i686_unknown_linux_gnu
1
But not for `BINTOOLS`:
$ echo $NIX_BINTOOLS_WRAPPER_TARGET_HOST_i686_unknown_linux_gnu
<empty>
The change adds BINTOOLS role and fixes linking:
$ nix develop -i --impure --expr 'with import ~/nm { system = "i686-linux"; }; (buildFHSUserEnv { name = "t"; targetPkgs = ps: with ps; [ libmpc stdenv.cc ]; }).env'
$ printf "int main(){}" | gcc -x c - -lmpc -o a
$ ld -lmpc -o a
ld: warning: cannot find entry symbol _start; not setting start address
Most FHSEnv-wrapped packages in Nixpkgs wrap a x86_64-linux binary; making
multiArch unnecessary bloat closure size. Saves about 1G in anki-bin.
This makes multiArch FHSEnvs the exception rather than the rule.
This reverts commit 840f2e0ac5, reversing
changes made to d3ed0402e5.
This breaks appimage which puts args into the runScript and we don't provide a
good way to pass thru additional args.
The actual bug was in nix-alien which should escape paths; providing a valid
runScript is the responsibility of the caller.
