lix/package.nix

383 lines
11 KiB
Nix
Raw Normal View History

{
pkgs,
lib,
stdenv,
aws-sdk-cpp,
# If the patched version of Boehm isn't passed, then patch it based off of
# pkgs.boehmgc. This allows `callPackage`ing this file without needing to
# to implement behavior that this package flat out doesn't build without
# anyway, but also allows easily overriding the patch logic.
boehmgc-nix ? __forDefaults.boehmgc-nix,
boehmgc,
nlohmann_json,
bison,
build-release-notes,
boost,
brotli,
bzip2,
cmake,
curl,
doxygen,
editline,
libstore/local-derivation-goal: prohibit creating setuid/setgid binaries With Linux kernel >=6.6 & glibc 2.39 a `fchmodat2(2)` is available that isn't filtered away by the libseccomp sandbox. Being able to use this to bypass that restriction has surprising results for some builds such as lxc[1]: > With kernel ≥6.6 and glibc 2.39, lxc's install phase uses fchmodat2, > which slips through https://github.com/NixOS/nix/blob/9b88e5284608116b7db0dbd3d5dd7a33b90d52d7/src/libstore/build/local-derivation-goal.cc#L1650-L1663. > The fixupPhase then uses fchmodat, which fails. > With older kernel or glibc, setting the suid bit fails in the > install phase, which is not treated as fatal, and then the > fixup phase does not try to set it again. Please note that there are still ways to bypass this sandbox[2] and this is mostly a fix for the breaking builds. This change works by creating a syscall filter for the `fchmodat2` syscall (number 452 on most systems). The problem is that glibc 2.39 is needed to have the correct syscall number available via `__NR_fchmodat2` / `__SNR_fchmodat2`, but this flake is still on nixpkgs 23.11. To have this change everywhere and not dependent on the glibc this package is built against, I added a header "fchmodat2-compat.hh" that sets the syscall number based on the architecture. On most platforms its 452 according to glibc with a few exceptions: $ rg --pcre2 'define __NR_fchmodat2 (?!452)' sysdeps/unix/sysv/linux/x86_64/x32/arch-syscall.h 58:#define __NR_fchmodat2 1073742276 sysdeps/unix/sysv/linux/mips/mips64/n32/arch-syscall.h 67:#define __NR_fchmodat2 6452 sysdeps/unix/sysv/linux/mips/mips64/n64/arch-syscall.h 62:#define __NR_fchmodat2 5452 sysdeps/unix/sysv/linux/mips/mips32/arch-syscall.h 70:#define __NR_fchmodat2 4452 sysdeps/unix/sysv/linux/alpha/arch-syscall.h 59:#define __NR_fchmodat2 562 I added a small regression-test to the setuid integration-test that attempts to set the suid bit on a file using the fchmodat2 syscall. I confirmed that the test fails without the change in local-derivation-goal. Additionally, we require libseccomp 2.5.5 or greater now: as it turns out, libseccomp maintains an internal syscall table and validates each rule against it. This means that when using libseccomp 2.5.4 or older, one may pass `452` as syscall number against it, but since it doesn't exist in the internal structure, `libseccomp` will refuse to create a filter for that. This happens with nixpkgs-23.11, i.e. on stable NixOS and when building Lix against the project's flake. To work around that * a backport of libseccomp 2.5.5 on upstream nixpkgs has been scheduled[3]. * the package now uses libseccomp 2.5.5 on its own already. This is to provide a quick fix since the correct fix for 23.11 is still a staging cycle away. We still need the compat header though since `SCMP_SYS(fchmodat2)` internally transforms this into `__SNR_fchmodat2` which points to `__NR_fchmodat2` from glibc 2.39, so it wouldn't build on glibc 2.38. The updated syscall table from libseccomp 2.5.5 is NOT used for that step, but used later, so we need both, our compat header and their syscall table 🤷 Relevant PRs in CppNix: * https://github.com/NixOS/nix/pull/10591 * https://github.com/NixOS/nix/pull/10501 [1] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2031073804 [2] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2030844251 [3] https://github.com/NixOS/nixpkgs/pull/306070 (cherry picked from commit ba6804518772e6afb403dd55478365d4b863c854) Change-Id: I6921ab5a363188c6bff617750d00bb517276b7fe
2024-04-14 14:10:23 +02:00
fetchurl,
flex,
git,
gtest,
jq,
libarchive,
libcpuid,
libstore/local-derivation-goal: prohibit creating setuid/setgid binaries With Linux kernel >=6.6 & glibc 2.39 a `fchmodat2(2)` is available that isn't filtered away by the libseccomp sandbox. Being able to use this to bypass that restriction has surprising results for some builds such as lxc[1]: > With kernel ≥6.6 and glibc 2.39, lxc's install phase uses fchmodat2, > which slips through https://github.com/NixOS/nix/blob/9b88e5284608116b7db0dbd3d5dd7a33b90d52d7/src/libstore/build/local-derivation-goal.cc#L1650-L1663. > The fixupPhase then uses fchmodat, which fails. > With older kernel or glibc, setting the suid bit fails in the > install phase, which is not treated as fatal, and then the > fixup phase does not try to set it again. Please note that there are still ways to bypass this sandbox[2] and this is mostly a fix for the breaking builds. This change works by creating a syscall filter for the `fchmodat2` syscall (number 452 on most systems). The problem is that glibc 2.39 is needed to have the correct syscall number available via `__NR_fchmodat2` / `__SNR_fchmodat2`, but this flake is still on nixpkgs 23.11. To have this change everywhere and not dependent on the glibc this package is built against, I added a header "fchmodat2-compat.hh" that sets the syscall number based on the architecture. On most platforms its 452 according to glibc with a few exceptions: $ rg --pcre2 'define __NR_fchmodat2 (?!452)' sysdeps/unix/sysv/linux/x86_64/x32/arch-syscall.h 58:#define __NR_fchmodat2 1073742276 sysdeps/unix/sysv/linux/mips/mips64/n32/arch-syscall.h 67:#define __NR_fchmodat2 6452 sysdeps/unix/sysv/linux/mips/mips64/n64/arch-syscall.h 62:#define __NR_fchmodat2 5452 sysdeps/unix/sysv/linux/mips/mips32/arch-syscall.h 70:#define __NR_fchmodat2 4452 sysdeps/unix/sysv/linux/alpha/arch-syscall.h 59:#define __NR_fchmodat2 562 I added a small regression-test to the setuid integration-test that attempts to set the suid bit on a file using the fchmodat2 syscall. I confirmed that the test fails without the change in local-derivation-goal. Additionally, we require libseccomp 2.5.5 or greater now: as it turns out, libseccomp maintains an internal syscall table and validates each rule against it. This means that when using libseccomp 2.5.4 or older, one may pass `452` as syscall number against it, but since it doesn't exist in the internal structure, `libseccomp` will refuse to create a filter for that. This happens with nixpkgs-23.11, i.e. on stable NixOS and when building Lix against the project's flake. To work around that * a backport of libseccomp 2.5.5 on upstream nixpkgs has been scheduled[3]. * the package now uses libseccomp 2.5.5 on its own already. This is to provide a quick fix since the correct fix for 23.11 is still a staging cycle away. We still need the compat header though since `SCMP_SYS(fchmodat2)` internally transforms this into `__SNR_fchmodat2` which points to `__NR_fchmodat2` from glibc 2.39, so it wouldn't build on glibc 2.38. The updated syscall table from libseccomp 2.5.5 is NOT used for that step, but used later, so we need both, our compat header and their syscall table 🤷 Relevant PRs in CppNix: * https://github.com/NixOS/nix/pull/10591 * https://github.com/NixOS/nix/pull/10501 [1] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2031073804 [2] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2030844251 [3] https://github.com/NixOS/nixpkgs/pull/306070 (cherry picked from commit ba6804518772e6afb403dd55478365d4b863c854) Change-Id: I6921ab5a363188c6bff617750d00bb517276b7fe
2024-04-14 14:10:23 +02:00
libseccomp-nix ? __forDefaults.libseccomp-nix,
libseccomp,
libsodium,
lsof,
lowdown,
mdbook,
mdbook-linkcheck,
mercurial,
meson,
ninja,
openssl,
pkg-config,
python3,
rapidcheck,
sqlite,
toml11,
util-linuxMinimal ? utillinuxMinimal,
utillinuxMinimal ? null,
xz,
busybox-sandbox-shell,
# internal fork of nix-doc providing :doc in the repl
lix-doc ? __forDefaults.lix-doc,
pname ? "nix",
versionSuffix ? "",
officialRelease ? true,
# Set to true to build the release notes for the next release.
buildUnreleasedNotes ? false,
internalApiDocs ? false,
# Avoid setting things that would interfere with a functioning devShell
forDevShell ? false,
# Not a real argument, just the only way to approximate let-binding some
# stuff for argument defaults.
__forDefaults ? {
canRunInstalled = stdenv.buildPlatform.canExecute stdenv.hostPlatform;
boehmgc-nix = (boehmgc.override { enableLargeConfig = true; }).overrideAttrs {
patches = [
# We do *not* include prev.patches (which doesn't exist in normal pkgs.boehmgc anyway)
# because if the caller of this package passed a patched boehm as `boehmgc` instead of
# `boehmgc-nix` then this will almost certainly have duplicate patches, which means
# the patches won't apply and we'll get a build failure.
./boehmgc-coroutine-sp-fallback.diff
# https://github.com/ivmai/bdwgc/pull/586
./boehmgc-traceable_allocator-public.diff
];
};
lix-doc = pkgs.callPackage ./lix-doc/package.nix { };
libstore/local-derivation-goal: prohibit creating setuid/setgid binaries With Linux kernel >=6.6 & glibc 2.39 a `fchmodat2(2)` is available that isn't filtered away by the libseccomp sandbox. Being able to use this to bypass that restriction has surprising results for some builds such as lxc[1]: > With kernel ≥6.6 and glibc 2.39, lxc's install phase uses fchmodat2, > which slips through https://github.com/NixOS/nix/blob/9b88e5284608116b7db0dbd3d5dd7a33b90d52d7/src/libstore/build/local-derivation-goal.cc#L1650-L1663. > The fixupPhase then uses fchmodat, which fails. > With older kernel or glibc, setting the suid bit fails in the > install phase, which is not treated as fatal, and then the > fixup phase does not try to set it again. Please note that there are still ways to bypass this sandbox[2] and this is mostly a fix for the breaking builds. This change works by creating a syscall filter for the `fchmodat2` syscall (number 452 on most systems). The problem is that glibc 2.39 is needed to have the correct syscall number available via `__NR_fchmodat2` / `__SNR_fchmodat2`, but this flake is still on nixpkgs 23.11. To have this change everywhere and not dependent on the glibc this package is built against, I added a header "fchmodat2-compat.hh" that sets the syscall number based on the architecture. On most platforms its 452 according to glibc with a few exceptions: $ rg --pcre2 'define __NR_fchmodat2 (?!452)' sysdeps/unix/sysv/linux/x86_64/x32/arch-syscall.h 58:#define __NR_fchmodat2 1073742276 sysdeps/unix/sysv/linux/mips/mips64/n32/arch-syscall.h 67:#define __NR_fchmodat2 6452 sysdeps/unix/sysv/linux/mips/mips64/n64/arch-syscall.h 62:#define __NR_fchmodat2 5452 sysdeps/unix/sysv/linux/mips/mips32/arch-syscall.h 70:#define __NR_fchmodat2 4452 sysdeps/unix/sysv/linux/alpha/arch-syscall.h 59:#define __NR_fchmodat2 562 I added a small regression-test to the setuid integration-test that attempts to set the suid bit on a file using the fchmodat2 syscall. I confirmed that the test fails without the change in local-derivation-goal. Additionally, we require libseccomp 2.5.5 or greater now: as it turns out, libseccomp maintains an internal syscall table and validates each rule against it. This means that when using libseccomp 2.5.4 or older, one may pass `452` as syscall number against it, but since it doesn't exist in the internal structure, `libseccomp` will refuse to create a filter for that. This happens with nixpkgs-23.11, i.e. on stable NixOS and when building Lix against the project's flake. To work around that * a backport of libseccomp 2.5.5 on upstream nixpkgs has been scheduled[3]. * the package now uses libseccomp 2.5.5 on its own already. This is to provide a quick fix since the correct fix for 23.11 is still a staging cycle away. We still need the compat header though since `SCMP_SYS(fchmodat2)` internally transforms this into `__SNR_fchmodat2` which points to `__NR_fchmodat2` from glibc 2.39, so it wouldn't build on glibc 2.38. The updated syscall table from libseccomp 2.5.5 is NOT used for that step, but used later, so we need both, our compat header and their syscall table 🤷 Relevant PRs in CppNix: * https://github.com/NixOS/nix/pull/10591 * https://github.com/NixOS/nix/pull/10501 [1] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2031073804 [2] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2030844251 [3] https://github.com/NixOS/nixpkgs/pull/306070 (cherry picked from commit ba6804518772e6afb403dd55478365d4b863c854) Change-Id: I6921ab5a363188c6bff617750d00bb517276b7fe
2024-04-14 14:10:23 +02:00
# FIXME remove when we have libsecomp 2.5.5 (currently in staging-23.11)
libseccomp-nix = libseccomp.overrideAttrs (_: rec {
version = "2.5.5";
src = fetchurl {
url = "https://github.com/seccomp/libseccomp/releases/download/v${version}/libseccomp-${version}.tar.gz";
hash = "sha256-JIosik2bmFiqa69ScSw0r+/PnJ6Ut23OAsHJqiX7M3U=";
};
});
},
}:
let
inherit (__forDefaults) canRunInstalled;
inherit (lib) fileset;
version = lib.fileContents ./.version + versionSuffix;
aws-sdk-cpp-nix = aws-sdk-cpp.override {
apis = [
"s3"
"transfer"
];
customMemoryManagement = false;
};
# Reimplementation of Nixpkgs' Meson cross file, with some additions to make
# it actually work.
mesonCrossFile =
let
cpuFamily =
platform:
with platform;
if isAarch32 then
"arm"
else if isx86_32 then
"x86"
else
platform.uname.processor;
in
builtins.toFile "lix-cross-file.conf" ''
[properties]
# Meson is convinced that if !buildPlatform.canExecute hostPlatform then we cannot
# build anything at all, which is not at all correct. If we can't execute the host
# platform, we'll just disable tests and doc gen.
needs_exe_wrapper = false
[binaries]
# Meson refuses to consider any CMake binary during cross compilation if it's
# not explicitly specified here, in the cross file.
# https://github.com/mesonbuild/meson/blob/0ed78cf6fa6d87c0738f67ae43525e661b50a8a2/mesonbuild/cmake/executor.py#L72
cmake = 'cmake'
'';
# The internal API docs need these for the build, but if we're not building
# Nix itself, then these don't need to be propagated.
maybePropagatedInputs = [
boehmgc-nix
nlohmann_json
];
# .gitignore has already been processed, so any changes in it are irrelevant
# at this point. It is not represented verbatim for test purposes because
# that would interfere with repo semantics.
baseFiles = fileset.fileFilter (f: f.name != ".gitignore") ./.;
configureFiles = fileset.unions [ ./.version ];
topLevelBuildFiles = fileset.unions ([
./meson.build
./meson.options
./meson
./scripts/meson.build
]);
functionalTestFiles = fileset.unions [
./tests/functional
./tests/unit
(fileset.fileFilter (f: lib.strings.hasPrefix "nix-profile" f.name) ./scripts)
];
in
stdenv.mkDerivation (finalAttrs: {
inherit pname version;
src = fileset.toSource {
root = ./.;
fileset = fileset.intersection baseFiles (
fileset.unions (
[
configureFiles
topLevelBuildFiles
functionalTestFiles
]
++ lib.optionals (!finalAttrs.dontBuild || internalApiDocs) [
./boehmgc-coroutine-sp-fallback.diff
./doc
./misc
./precompiled-headers.h
./src
./COPYING
]
)
);
};
VERSION_SUFFIX = versionSuffix;
outputs =
[ "out" ]
++ lib.optionals (!finalAttrs.dontBuild) [
"dev"
"doc"
];
dontBuild = false;
mesonFlags =
lib.optionals stdenv.hostPlatform.isLinux [
# You'd think meson could just find this in PATH, but busybox is in buildInputs,
# which don't actually get added to PATH. And buildInputs is correct over
# nativeBuildInputs since this should be a busybox executable on the host.
"-Dsandbox-shell=${lib.getExe' busybox-sandbox-shell "busybox"}"
]
++ lib.optional stdenv.hostPlatform.isStatic "-Denable-embedded-sandbox-shell=true"
++ lib.optional (finalAttrs.dontBuild) "-Denable-build=false"
++ [
# mesonConfigurePhase automatically passes -Dauto_features=enabled,
# so we must explicitly enable or disable features that we are not passing
# dependencies for.
(lib.mesonEnable "internal-api-docs" internalApiDocs)
(lib.mesonBool "enable-tests" finalAttrs.doCheck)
(lib.mesonBool "enable-docs" canRunInstalled)
]
++ lib.optional (stdenv.hostPlatform != stdenv.buildPlatform) "--cross-file=${mesonCrossFile}";
# We only include CMake so that Meson can locate toml11, which only ships CMake dependency metadata.
dontUseCmakeConfigure = true;
nativeBuildInputs =
[
bison
flex
python3
meson
ninja
cmake
]
++ [
(lib.getBin lowdown)
mdbook
mdbook-linkcheck
]
++ [
pkg-config
# Tests
git
mercurial
jq
lsof
]
++ lib.optional stdenv.hostPlatform.isLinux util-linuxMinimal
++ lib.optional (!officialRelease && buildUnreleasedNotes) build-release-notes
++ lib.optional (internalApiDocs || forDevShell) doxygen;
buildInputs =
[
curl
bzip2
xz
brotli
editline
openssl
sqlite
libarchive
boost
lowdown
libsodium
toml11
lix-doc
]
++ lib.optionals stdenv.hostPlatform.isLinux [
libstore/local-derivation-goal: prohibit creating setuid/setgid binaries With Linux kernel >=6.6 & glibc 2.39 a `fchmodat2(2)` is available that isn't filtered away by the libseccomp sandbox. Being able to use this to bypass that restriction has surprising results for some builds such as lxc[1]: > With kernel ≥6.6 and glibc 2.39, lxc's install phase uses fchmodat2, > which slips through https://github.com/NixOS/nix/blob/9b88e5284608116b7db0dbd3d5dd7a33b90d52d7/src/libstore/build/local-derivation-goal.cc#L1650-L1663. > The fixupPhase then uses fchmodat, which fails. > With older kernel or glibc, setting the suid bit fails in the > install phase, which is not treated as fatal, and then the > fixup phase does not try to set it again. Please note that there are still ways to bypass this sandbox[2] and this is mostly a fix for the breaking builds. This change works by creating a syscall filter for the `fchmodat2` syscall (number 452 on most systems). The problem is that glibc 2.39 is needed to have the correct syscall number available via `__NR_fchmodat2` / `__SNR_fchmodat2`, but this flake is still on nixpkgs 23.11. To have this change everywhere and not dependent on the glibc this package is built against, I added a header "fchmodat2-compat.hh" that sets the syscall number based on the architecture. On most platforms its 452 according to glibc with a few exceptions: $ rg --pcre2 'define __NR_fchmodat2 (?!452)' sysdeps/unix/sysv/linux/x86_64/x32/arch-syscall.h 58:#define __NR_fchmodat2 1073742276 sysdeps/unix/sysv/linux/mips/mips64/n32/arch-syscall.h 67:#define __NR_fchmodat2 6452 sysdeps/unix/sysv/linux/mips/mips64/n64/arch-syscall.h 62:#define __NR_fchmodat2 5452 sysdeps/unix/sysv/linux/mips/mips32/arch-syscall.h 70:#define __NR_fchmodat2 4452 sysdeps/unix/sysv/linux/alpha/arch-syscall.h 59:#define __NR_fchmodat2 562 I added a small regression-test to the setuid integration-test that attempts to set the suid bit on a file using the fchmodat2 syscall. I confirmed that the test fails without the change in local-derivation-goal. Additionally, we require libseccomp 2.5.5 or greater now: as it turns out, libseccomp maintains an internal syscall table and validates each rule against it. This means that when using libseccomp 2.5.4 or older, one may pass `452` as syscall number against it, but since it doesn't exist in the internal structure, `libseccomp` will refuse to create a filter for that. This happens with nixpkgs-23.11, i.e. on stable NixOS and when building Lix against the project's flake. To work around that * a backport of libseccomp 2.5.5 on upstream nixpkgs has been scheduled[3]. * the package now uses libseccomp 2.5.5 on its own already. This is to provide a quick fix since the correct fix for 23.11 is still a staging cycle away. We still need the compat header though since `SCMP_SYS(fchmodat2)` internally transforms this into `__SNR_fchmodat2` which points to `__NR_fchmodat2` from glibc 2.39, so it wouldn't build on glibc 2.38. The updated syscall table from libseccomp 2.5.5 is NOT used for that step, but used later, so we need both, our compat header and their syscall table 🤷 Relevant PRs in CppNix: * https://github.com/NixOS/nix/pull/10591 * https://github.com/NixOS/nix/pull/10501 [1] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2031073804 [2] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2030844251 [3] https://github.com/NixOS/nixpkgs/pull/306070 (cherry picked from commit ba6804518772e6afb403dd55478365d4b863c854) Change-Id: I6921ab5a363188c6bff617750d00bb517276b7fe
2024-04-14 14:10:23 +02:00
libseccomp-nix
busybox-sandbox-shell
]
++ lib.optional internalApiDocs rapidcheck
++ lib.optional stdenv.hostPlatform.isx86_64 libcpuid
# There have been issues building these dependencies
++ lib.optional (stdenv.hostPlatform == stdenv.buildPlatform) aws-sdk-cpp-nix
++ lib.optionals (finalAttrs.dontBuild) maybePropagatedInputs;
checkInputs = [
gtest
rapidcheck
];
propagatedBuildInputs = lib.optionals (!finalAttrs.dontBuild) maybePropagatedInputs;
disallowedReferences = [ boost ];
# Needed for Meson to find Boost.
# https://github.com/NixOS/nixpkgs/issues/86131.
env = {
BOOST_INCLUDEDIR = "${lib.getDev boost}/include";
BOOST_LIBRARYDIR = "${lib.getLib boost}/lib";
};
preConfigure =
lib.optionalString (!finalAttrs.dontBuild && !stdenv.hostPlatform.isStatic) ''
# Copy libboost_context so we don't get all of Boost in our closure.
# https://github.com/NixOS/nixpkgs/issues/45462
mkdir -p $out/lib
cp -pd ${boost}/lib/{libboost_context*,libboost_thread*,libboost_system*} $out/lib
rm -f $out/lib/*.a
''
+ lib.optionalString (!finalAttrs.dontBuild && stdenv.hostPlatform.isLinux) ''
chmod u+w $out/lib/*.so.*
patchelf --set-rpath $out/lib:${stdenv.cc.cc.lib}/lib $out/lib/libboost_thread.so.*
''
+ lib.optionalString (!finalAttrs.dontBuild && stdenv.hostPlatform.isDarwin) ''
for LIB in $out/lib/*.dylib; do
chmod u+w $LIB
install_name_tool -id $LIB $LIB
install_name_tool -delete_rpath ${boost}/lib/ $LIB || true
done
install_name_tool -change ${boost}/lib/libboost_system.dylib $out/lib/libboost_system.dylib $out/lib/libboost_thread.dylib
''
+ ''
# Workaround https://github.com/NixOS/nixpkgs/issues/294890.
if [[ -n "''${doCheck:-}" ]]; then
appendToVar configureFlags "--enable-tests"
else
appendToVar configureFlags "--disable-tests"
fi
'';
mesonBuildType = "debugoptimized";
installTargets = lib.optional internalApiDocs "internal-api-html";
enableParallelBuilding = true;
doCheck = canRunInstalled;
mesonCheckFlags = [ "--suite=check" ];
# Make sure the internal API docs are already built, because mesonInstallPhase
# won't let us build them there. They would normally be built in buildPhase,
# but the internal API docs are conventionally built with doBuild = false.
preInstall = lib.optional internalApiDocs ''
meson ''${mesonBuildFlags:-} compile "$installTargets"
'';
postInstall =
lib.optionalString (!finalAttrs.dontBuild) ''
mkdir -p $doc/nix-support
echo "doc manual $doc/share/doc/nix/manual" >> $doc/nix-support/hydra-build-products
''
+ lib.optionalString stdenv.hostPlatform.isStatic ''
mkdir -p $out/nix-support
echo "file binary-dist $out/bin/nix" >> $out/nix-support/hydra-build-products
''
+ lib.optionalString stdenv.isDarwin ''
for lib in libnixutil.dylib libnixexpr.dylib; do
install_name_tool \
-change "${lib.getLib boost}/lib/libboost_context.dylib" \
"$out/lib/libboost_context.dylib" \
"$out/lib/$lib"
done
''
+ lib.optionalString internalApiDocs ''
mkdir -p $out/nix-support
echo "doc internal-api-docs $out/share/doc/nix/internal-api/html" >> "$out/nix-support/hydra-build-products"
'';
doInstallCheck = finalAttrs.doCheck;
mesonInstallCheckFlags = [ "--suite=installcheck" ];
installCheckPhase = ''
runHook preInstallCheck
flagsArray=($mesonInstallCheckFlags "''${mesonInstallCheckFlagsArray[@]}")
meson test --no-rebuild "''${flagsArray[@]}"
runHook postInstallCheck
'';
separateDebugInfo = !stdenv.hostPlatform.isStatic && !finalAttrs.dontBuild;
strictDeps = true;
Build with traps on signed overflow This is UB, we should not be doing it, and we can cheaply turn it into crashes reliably. We would much rather have crashes than the program doing something silly. Benchmarks, but i wonder if they are nonsense because they get identical times across compilers?! | Command | Mean [ms] | Min [ms] | Max [ms] | Relative | |:---|---:|---:|---:|---:| | `result-clang/bin/nix eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix` | 375.5 ± 24.0 | 353.8 | 408.8 | 1.00 | | `result-gcc/bin/nix eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix` | 407.9 ± 26.0 | 385.1 | 449.5 | 1.09 ± 0.10 | | `result-clangsan/bin/nix eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix` | 382.2 ± 26.6 | 354.9 | 419.0 | 1.02 ± 0.10 | | `result-gccsan/bin/nix eval -f ../nixpkgs/pkgs/development/haskell-modules/hackage-packages.nix` | 408.6 ± 24.6 | 384.5 | 441.9 | 1.09 ± 0.10 | | Command | Mean [s] | Min [s] | Max [s] | Relative | |:---|---:|---:|---:|---:| | `result-clang/bin/nix search --no-eval-cache github:nixos/nixpkgs/e1fa12d4f6c6fe19ccb59cac54b5b3f25e160870 hello` | 17.199 ± 0.167 | 16.930 | 17.499 | 1.01 ± 0.01 | | `result-gcc/bin/nix search --no-eval-cache github:nixos/nixpkgs/e1fa12d4f6c6fe19ccb59cac54b5b3f25e160870 hello` | 17.409 ± 0.126 | 17.242 | 17.633 | 1.02 ± 0.01 | | `result-clangsan/bin/nix search --no-eval-cache github:nixos/nixpkgs/e1fa12d4f6c6fe19ccb59cac54b5b3f25e160870 hello` | 17.080 ± 0.137 | 16.879 | 17.350 | 1.00 | | `result-gccsan/bin/nix search --no-eval-cache github:nixos/nixpkgs/e1fa12d4f6c6fe19ccb59cac54b5b3f25e160870 hello` | 17.396 ± 0.160 | 17.131 | 17.660 | 1.02 ± 0.01 | | Command | Mean [s] | Min [s] | Max [s] | Relative | |:---|---:|---:|---:|---:| | `result-clang/bin/nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 6.267 ± 0.069 | 6.197 | 6.415 | 1.02 ± 0.01 | | `result-gcc/bin/nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 6.232 ± 0.045 | 6.180 | 6.311 | 1.01 ± 0.01 | | `result-clangsan/bin/nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 6.162 ± 0.020 | 6.133 | 6.196 | 1.00 | | `result-gccsan/bin/nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 6.229 ± 0.031 | 6.199 | 6.289 | 1.01 ± 0.01 | | Command | Mean [s] | Min [s] | Max [s] | Relative | |:---|---:|---:|---:|---:| | `GC_INITIAL_HEAP_SIZE=10g result-clang/bin/nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 4.683 ± 0.044 | 4.630 | 4.761 | 1.00 | | `GC_INITIAL_HEAP_SIZE=10g result-gcc/bin/nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 4.750 ± 0.041 | 4.680 | 4.812 | 1.01 ± 0.01 | | `GC_INITIAL_HEAP_SIZE=10g result-clangsan/bin/nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 4.703 ± 0.040 | 4.640 | 4.760 | 1.00 ± 0.01 | | `GC_INITIAL_HEAP_SIZE=10g result-gccsan/bin/nix eval --raw --impure --expr 'with import <nixpkgs/nixos> {}; system'` | 4.766 ± 0.037 | 4.727 | 4.844 | 1.02 ± 0.01 | Change-Id: I616ca3eab670317587d47b41870d8ac963c019ae
2024-03-25 07:45:25 +01:00
# strictoverflow is disabled because we trap on signed overflow instead
hardeningDisable = [ "strictoverflow" ] ++ lib.optional stdenv.hostPlatform.isStatic "pie";
meta.platforms = lib.platforms.unix;
passthru.perl-bindings = pkgs.callPackage ./perl { inherit fileset stdenv; };
libstore/local-derivation-goal: prohibit creating setuid/setgid binaries With Linux kernel >=6.6 & glibc 2.39 a `fchmodat2(2)` is available that isn't filtered away by the libseccomp sandbox. Being able to use this to bypass that restriction has surprising results for some builds such as lxc[1]: > With kernel ≥6.6 and glibc 2.39, lxc's install phase uses fchmodat2, > which slips through https://github.com/NixOS/nix/blob/9b88e5284608116b7db0dbd3d5dd7a33b90d52d7/src/libstore/build/local-derivation-goal.cc#L1650-L1663. > The fixupPhase then uses fchmodat, which fails. > With older kernel or glibc, setting the suid bit fails in the > install phase, which is not treated as fatal, and then the > fixup phase does not try to set it again. Please note that there are still ways to bypass this sandbox[2] and this is mostly a fix for the breaking builds. This change works by creating a syscall filter for the `fchmodat2` syscall (number 452 on most systems). The problem is that glibc 2.39 is needed to have the correct syscall number available via `__NR_fchmodat2` / `__SNR_fchmodat2`, but this flake is still on nixpkgs 23.11. To have this change everywhere and not dependent on the glibc this package is built against, I added a header "fchmodat2-compat.hh" that sets the syscall number based on the architecture. On most platforms its 452 according to glibc with a few exceptions: $ rg --pcre2 'define __NR_fchmodat2 (?!452)' sysdeps/unix/sysv/linux/x86_64/x32/arch-syscall.h 58:#define __NR_fchmodat2 1073742276 sysdeps/unix/sysv/linux/mips/mips64/n32/arch-syscall.h 67:#define __NR_fchmodat2 6452 sysdeps/unix/sysv/linux/mips/mips64/n64/arch-syscall.h 62:#define __NR_fchmodat2 5452 sysdeps/unix/sysv/linux/mips/mips32/arch-syscall.h 70:#define __NR_fchmodat2 4452 sysdeps/unix/sysv/linux/alpha/arch-syscall.h 59:#define __NR_fchmodat2 562 I added a small regression-test to the setuid integration-test that attempts to set the suid bit on a file using the fchmodat2 syscall. I confirmed that the test fails without the change in local-derivation-goal. Additionally, we require libseccomp 2.5.5 or greater now: as it turns out, libseccomp maintains an internal syscall table and validates each rule against it. This means that when using libseccomp 2.5.4 or older, one may pass `452` as syscall number against it, but since it doesn't exist in the internal structure, `libseccomp` will refuse to create a filter for that. This happens with nixpkgs-23.11, i.e. on stable NixOS and when building Lix against the project's flake. To work around that * a backport of libseccomp 2.5.5 on upstream nixpkgs has been scheduled[3]. * the package now uses libseccomp 2.5.5 on its own already. This is to provide a quick fix since the correct fix for 23.11 is still a staging cycle away. We still need the compat header though since `SCMP_SYS(fchmodat2)` internally transforms this into `__SNR_fchmodat2` which points to `__NR_fchmodat2` from glibc 2.39, so it wouldn't build on glibc 2.38. The updated syscall table from libseccomp 2.5.5 is NOT used for that step, but used later, so we need both, our compat header and their syscall table 🤷 Relevant PRs in CppNix: * https://github.com/NixOS/nix/pull/10591 * https://github.com/NixOS/nix/pull/10501 [1] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2031073804 [2] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2030844251 [3] https://github.com/NixOS/nixpkgs/pull/306070 (cherry picked from commit ba6804518772e6afb403dd55478365d4b863c854) Change-Id: I6921ab5a363188c6bff617750d00bb517276b7fe
2024-04-14 14:10:23 +02:00
# Export the patched version of boehmgc & libseccomp.
# flake.nix exports that into its overlay.
libstore/local-derivation-goal: prohibit creating setuid/setgid binaries With Linux kernel >=6.6 & glibc 2.39 a `fchmodat2(2)` is available that isn't filtered away by the libseccomp sandbox. Being able to use this to bypass that restriction has surprising results for some builds such as lxc[1]: > With kernel ≥6.6 and glibc 2.39, lxc's install phase uses fchmodat2, > which slips through https://github.com/NixOS/nix/blob/9b88e5284608116b7db0dbd3d5dd7a33b90d52d7/src/libstore/build/local-derivation-goal.cc#L1650-L1663. > The fixupPhase then uses fchmodat, which fails. > With older kernel or glibc, setting the suid bit fails in the > install phase, which is not treated as fatal, and then the > fixup phase does not try to set it again. Please note that there are still ways to bypass this sandbox[2] and this is mostly a fix for the breaking builds. This change works by creating a syscall filter for the `fchmodat2` syscall (number 452 on most systems). The problem is that glibc 2.39 is needed to have the correct syscall number available via `__NR_fchmodat2` / `__SNR_fchmodat2`, but this flake is still on nixpkgs 23.11. To have this change everywhere and not dependent on the glibc this package is built against, I added a header "fchmodat2-compat.hh" that sets the syscall number based on the architecture. On most platforms its 452 according to glibc with a few exceptions: $ rg --pcre2 'define __NR_fchmodat2 (?!452)' sysdeps/unix/sysv/linux/x86_64/x32/arch-syscall.h 58:#define __NR_fchmodat2 1073742276 sysdeps/unix/sysv/linux/mips/mips64/n32/arch-syscall.h 67:#define __NR_fchmodat2 6452 sysdeps/unix/sysv/linux/mips/mips64/n64/arch-syscall.h 62:#define __NR_fchmodat2 5452 sysdeps/unix/sysv/linux/mips/mips32/arch-syscall.h 70:#define __NR_fchmodat2 4452 sysdeps/unix/sysv/linux/alpha/arch-syscall.h 59:#define __NR_fchmodat2 562 I added a small regression-test to the setuid integration-test that attempts to set the suid bit on a file using the fchmodat2 syscall. I confirmed that the test fails without the change in local-derivation-goal. Additionally, we require libseccomp 2.5.5 or greater now: as it turns out, libseccomp maintains an internal syscall table and validates each rule against it. This means that when using libseccomp 2.5.4 or older, one may pass `452` as syscall number against it, but since it doesn't exist in the internal structure, `libseccomp` will refuse to create a filter for that. This happens with nixpkgs-23.11, i.e. on stable NixOS and when building Lix against the project's flake. To work around that * a backport of libseccomp 2.5.5 on upstream nixpkgs has been scheduled[3]. * the package now uses libseccomp 2.5.5 on its own already. This is to provide a quick fix since the correct fix for 23.11 is still a staging cycle away. We still need the compat header though since `SCMP_SYS(fchmodat2)` internally transforms this into `__SNR_fchmodat2` which points to `__NR_fchmodat2` from glibc 2.39, so it wouldn't build on glibc 2.38. The updated syscall table from libseccomp 2.5.5 is NOT used for that step, but used later, so we need both, our compat header and their syscall table 🤷 Relevant PRs in CppNix: * https://github.com/NixOS/nix/pull/10591 * https://github.com/NixOS/nix/pull/10501 [1] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2031073804 [2] https://github.com/NixOS/nixpkgs/issues/300635#issuecomment-2030844251 [3] https://github.com/NixOS/nixpkgs/pull/306070 (cherry picked from commit ba6804518772e6afb403dd55478365d4b863c854) Change-Id: I6921ab5a363188c6bff617750d00bb517276b7fe
2024-04-14 14:10:23 +02:00
passthru = {
inherit (__forDefaults) boehmgc-nix libseccomp-nix;
};
})