This allows builds to call setuid binaries. This was previously
possible until we started using seccomp. Turns out that seccomp by
default disallows processes from acquiring new privileges. Generally,
any use of setuid binaries (except those created by the builder
itself) is by definition impure, but some people were relying on this
ability for certain tests.
Example:
$ nix build '(with import <nixpkgs> {}; runCommand "foo" {} "/run/wrappers/bin/ping -c 1 8.8.8.8; exit 1")' --no-allow-new-privileges
builder for ‘/nix/store/j0nd8kv85hd6r4kxgnwzvr0k65ykf6fv-foo.drv’ failed with exit code 1; last 2 log lines:
cannot raise the capability into the Ambient set
: Operation not permitted
$ nix build '(with import <nixpkgs> {}; runCommand "foo" {} "/run/wrappers/bin/ping -c 1 8.8.8.8; exit 1")' --allow-new-privileges
builder for ‘/nix/store/j0nd8kv85hd6r4kxgnwzvr0k65ykf6fv-foo.drv’ failed with exit code 1; last 6 log lines:
PING 8.8.8.8 (8.8.8.8) 56(84) bytes of data.
64 bytes from 8.8.8.8: icmp_seq=1 ttl=46 time=15.2 ms
Fixes#1429.
Functions like copyClosure() had 3 bool arguments, which creates a
severe risk of mixing up arguments.
Also, implement copyClosure() using copyPaths().
Sandboxes cannot be nested, so if Nix's build runs inside a sandbox,
it cannot use a sandbox itself. I don't see a clean way to detect
whether we're in a sandbox, so use a test-specific hack.
https://github.com/NixOS/nix/issues/1413
Even with "build-use-sandbox = false", we now use sandboxing with a
permissive profile that allows everything except the creation of
setuid/setgid binaries.
Also, add rules to allow fixed-output derivations to access the
network.
These rules are sufficient to build stdenvDarwin without any
__sandboxProfile magic.
The filename used was not unique and owned by the build user, so
builds could fail with
error: while setting up the build environment: cannot unlink ‘/nix/store/99i210ihnsjacajaw8r33fmgjvzpg6nr-bison-3.0.4.drv.sb’: Permission denied
Fixes
src/libstore/build.cc:2321:45: error: non-constant-expression cannot be narrowed from type 'int' to 'scmp_datum_t' (aka 'unsigned long') in initializer list [-Wc++11-narrowing]
This prevents builders from setting the S_ISUID or S_ISGID bits,
preventing users from using a nixbld* user to create a setuid/setgid
binary to interfere with subsequent builds under the same nixbld* uid.
This is based on aszlig's seccomp code
(47f587700d).
Reported by Linus Heckemann.
This is useful when we're using a diverted store (e.g. "--store
local?root=/tmp/nix") in conjunction with a statically-linked sh from
the host store (e.g. "sandbox-paths =/bin/sh=/nix/store/.../bin/busybox").
There is a security issue when a build accidentally stores its $TMPDIR
in some critical place, such as an RPATH. If
TMPDIR=/tmp/nix-build-..., then any user on the system can recreate
that directory and inject libraries into the RPATH of programs
executed by other users. Since /build probably doesn't exist (or isn't
world-writable), this mitigates the issue.
This is useful for one-off situations where you want to specify a
builder on the command line instead of having to mess with
nix.machines. E.g.
$ nix-build -A hello --argstr system x86_64-darwin \
--option builders 'root@macstadium1 x86_64-darwin'
will perform the specified build on "macstadium1".
It also removes the need for a separate nix.machines file since you
can specify builders in nix.conf directly. (In fact nix.machines is
yet another hack that predates the general nix.conf configuration
file, IIRC.)
Note: this option is supported by the daemon for trusted users. The
fact that this allows trusted users to specify paths to SSH keys to
which they don't normally have access is maybe a bit too much trust...
For backwards compatibility, if the URI is just a hostname, ssh://
(i.e. LegacySSHStore) is prepended automatically.
Also, all fields except the URI are now optional. For example, this is
a valid nix.machines file:
local?root=/tmp/nix
This is useful for testing the remote build machinery since you don't
have to mess around with ssh.
This is to simplify remote build configuration. These environment
variables predate nix.conf.
The build hook now has a sensible default (namely build-remote).
The current load is kept in the Nix state directory now.
Since build-remote uses buildDerivation() now, we don't need to copy
the .drv file anymore. This greatly reduces the set of input paths
copied to the remote side (e.g. from 392 to 51 store paths for GNU
hello on x86_64-darwin).
The typical use is to inherit Config and add Setting<T> members:
class MyClass : private Config
{
Setting<int> foo{this, 123, "foo", "the number of foos to use"};
Setting<std::string> bar{this, "blabla", "bar", "the name of the bar"};
MyClass() : Config(readConfigFile("/etc/my-app.conf"))
{
std::cout << foo << "\n"; // will print 123 unless overriden
}
};
Currently, this is used by Store and its subclasses for store
parameters. You now get a warning if you specify a non-existant store
parameter in a store URI.
This fixes "No such file or directory" when opening /dev/ptmx
(e.g. http://hydra.nixos.org/build/51094249).
The reason appears to be some changes to /dev/ptmx / /dev/pts handling
between Linux 4.4 and 4.9. See
https://patchwork.kernel.org/patch/7832531/.
The fix is to go back to mounting a proper /dev/pts instance inside
the sandbox. Happily, this now works inside user namespaces, even for
unprivileged users. So
NIX_REMOTE=local?root=/tmp/nix nix-build \
'<nixpkgs/nixos/tests/misc.nix>' -A test
works for non-root users.
The downside is that the fix breaks sandbox builds on older kernels
(probably pre-4.6), since mounting a devpts fails inside user
namespaces for some reason I've never been able to figure out. Builds
on those systems will fail with
error: while setting up the build environment: mounting /dev/pts: Invalid argument
Ah well.
This allows various Store implementations to provide different ways to
get build logs. For example, BinaryCacheStore can get the build logs
from the binary cache.
Also, remove the log-servers option since we can use substituters for
this.
Previously, the Settings class allowed other code to query for string
properties, which led to a proliferation of code all over the place making
up new options without any sort of central registry of valid options. This
commit pulls all those options back into the central Settings class and
removes the public get() methods, to discourage future abuses like that.
Furthermore, because we know the full set of options ahead of time, we
now fail loudly if someone enters an unrecognized option, thus preventing
subtle typos. With some template fun, we could probably also dump the full
set of options (with documentation, defaults, etc.) to the command line,
but I'm not doing that yet here.
This allows <nix/fetchurl.nix> to fetch private Git/Mercurial
repositories, e.g.
import <nix/fetchurl.nix> {
url = https://edolstra@bitbucket.org/edolstra/my-private-repo/get/80a14018daed.tar.bz2;
sha256 = "1mgqzn7biqkq3hf2697b0jc4wabkqhmzq2srdymjfa6sb9zb6qs7";
}
where /etc/nix/netrc contains:
machine bitbucket.org
login edolstra
password blabla...
This works even when sandboxing is enabled.
To do: add unpacking support (i.e. fetchzip functionality).
Because config.h can #define things like _FILE_OFFSET_BITS=64 and not
every compilation unit includes config.h, we currently compile half of
Nix with _FILE_OFFSET_BITS=64 and other half with _FILE_OFFSET_BITS
unset. This causes major havoc with the Settings class on e.g. 32-bit ARM,
where different compilation units disagree with the struct layout.
E.g.:
diff --git a/src/libstore/globals.cc b/src/libstore/globals.cc
@@ -166,6 +166,8 @@ void Settings::update()
_get(useSubstitutes, "build-use-substitutes");
+ fprintf(stderr, "at Settings::update(): &useSubstitutes = %p\n", &nix::settings.useSubstitutes);
_get(buildUsersGroup, "build-users-group");
diff --git a/src/libstore/remote-store.cc b/src/libstore/remote-store.cc
+++ b/src/libstore/remote-store.cc
@@ -138,6 +138,8 @@ void RemoteStore::initConnection(Connection & conn)
void RemoteStore::setOptions(Connection & conn)
{
+ fprintf(stderr, "at RemoteStore::setOptions(): &useSubstitutes = %p\n", &nix::settings.useSubstitutes);
conn.to << wopSetOptions
Gave me:
at Settings::update(): &useSubstitutes = 0xb6e5c5cb
at RemoteStore::setOptions(): &useSubstitutes = 0xb6e5c5c7
That was not a fun one to debug!
This writes info about every path in the closure in the same format as
‘nix path-info --json’. Thus it also includes NAR hashes and sizes.
Example:
[
{
"path": "/nix/store/10h6li26i7g6z3mdpvra09yyf10mmzdr-hello-2.10",
"narHash": "sha256:0ckdc4z20kkmpqdilx0wl6cricxv90lh85xpv2qljppcmz6vzcxl",
"narSize": 197648,
"references": [
"/nix/store/10h6li26i7g6z3mdpvra09yyf10mmzdr-hello-2.10",
"/nix/store/27binbdy296qvjycdgr1535v8872vz3z-glibc-2.24"
],
"closureSize": 20939776
},
{
"path": "/nix/store/27binbdy296qvjycdgr1535v8872vz3z-glibc-2.24",
"narHash": "sha256:1nfn3m3p98y1c0kd0brp80dn9n5mycwgrk183j17rajya0h7gax3",
"narSize": 20742128,
"references": [
"/nix/store/27binbdy296qvjycdgr1535v8872vz3z-glibc-2.24"
],
"closureSize": 20742128
}
]
Fixes#1134.
Previously, all derivation attributes had to be coerced into strings
so that they could be passed via the environment. This is lossy
(e.g. lists get flattened, necessitating configureFlags
vs. configureFlagsArray, of which the latter cannot be specified as an
attribute), doesn't support attribute sets at all, and has size
limitations (necessitating hacks like passAsFile).
This patch adds a new mode for passing attributes to builders, namely
encoded as a JSON file ".attrs.json" in the current directory of the
builder. This mode is activated via the special attribute
__structuredAttrs = true;
(The idea is that one day we can set this in stdenv.mkDerivation.)
For example,
stdenv.mkDerivation {
__structuredAttrs = true;
name = "foo";
buildInputs = [ pkgs.hello pkgs.cowsay ];
doCheck = true;
hardening.format = false;
}
results in a ".attrs.json" file containing (sans the indentation):
{
"buildInputs": [],
"builder": "/nix/store/ygl61ycpr2vjqrx775l1r2mw1g2rb754-bash-4.3-p48/bin/bash",
"configureFlags": [
"--with-foo",
"--with-bar=1 2"
],
"doCheck": true,
"hardening": {
"format": false
},
"name": "foo",
"nativeBuildInputs": [
"/nix/store/10h6li26i7g6z3mdpvra09yyf10mmzdr-hello-2.10",
"/nix/store/4jnvjin0r6wp6cv1hdm5jbkx3vinlcvk-cowsay-3.03"
],
"propagatedBuildInputs": [],
"propagatedNativeBuildInputs": [],
"stdenv": "/nix/store/f3hw3p8armnzy6xhd4h8s7anfjrs15n2-stdenv",
"system": "x86_64-linux"
}
"passAsFile" is ignored in this mode because it's not needed - large
strings are included directly in the JSON representation.
It is up to the builder to do something with the JSON
representation. For example, in bash-based builders, lists/attrsets of
string values could be mapped to bash (associative) arrays.
This closes a long-time bug that allowed builds to hang Nix
indefinitely (regardless of timeouts) simply by doing
exec > /dev/null 2>&1; while true; do true; done
Now, on EOF, we just send SIGKILL to the child to make sure it's
really gone.
That is, when build-repeat > 0, and the output of two rounds differ,
then print a warning rather than fail the build. This is primarily to
let Hydra check reproducibility of all packages.
These syscalls are only available in 32bit architectures, but libseccomp
should handle them correctly even if we're on native architectures that
do not have these syscalls.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
Commands such as "cp -p" also use fsetxattr() in addition to fchown(),
so we need to make sure these syscalls always return successful as well
in order to avoid nasty "Invalid value" errors.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
What we basically want is a seccomp mode 2 BPF program like this but for
every architecture:
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, offsetof(struct seccomp_data, nr)),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_chown, 4, 0),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_fchown, 3, 0),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_fchownat, 2, 0),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_lchown, 1, 0),
BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ERRNO)
However, on 32 bit architectures we do have chown32, lchown32 and
fchown32, so we'd need to add all the architecture blurb which
libseccomp handles for us.
So we only need to make sure that we add the 32bit seccomp arch while
we're on x86_64 and otherwise we just stay at the native architecture
which was set during seccomp_init(), which more or less replicates
setting 32bit personality during runChild().
The FORCE_SUCCESS() macro here could be a bit less ugly but I think
repeating the seccomp_rule_add() all over the place is way uglier.
Another way would have been to create a vector of syscalls to iterate
over, but that would make error messages uglier because we can either
only print the (libseccomp-internal) syscall number or use
seccomp_syscall_resolve_num_arch() to get the name or even make the
vector a pair number/name, essentially duplicating everything again.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
This reverts commit ff0c0b645c.
We're going to use seccomp to allow "cp -p" and force chown-related
syscalls to always return 0.
Signed-off-by: aszlig <aszlig@redmoonstudios.org>
For example, you can now set
build-sandbox-paths = /dev/nvidiactl?
to specify that /dev/nvidiactl should only be mounted in the sandbox
if it exists in the host filesystem. This is useful e.g. for EC2
images that should support both CUDA and non-CUDA instances.
We can now write
throw Error("file '%s' not found", path);
instead of
throw Error(format("file '%s' not found") % path);
and similarly
printError("file '%s' not found", path);
instead of
printMsg(lvlError, format("file '%s' not found") % path);
This largely reverts c68e5913c7. Running
builds as root breaks "cp -p", since when running as root, "cp -p"
assumes that it can succesfully chown() files. But that's not actually
the case since the user namespace doesn't provide a complete uid
mapping. So it barfs with a fatal error message ("cp: failed to
preserve ownership for 'foo': Invalid argument").
This fixes an assertion failure in "assert(goal);" in
Worker::waitForInput() after a substitution goal is cancelled by the
termination of another goal. The problem was the line
//worker.childTerminated(shared_from_this()); // FIXME
in the SubstitutionGoal destructor. This was disabled because
shared_from_this() obviously doesn't work from a destructor. So we now
use a real pointer for object identity.
For example, you can now say:
configureFlags = "--prefix=${placeholder "out"} --includedir=${placeholder "dev"}";
The strings returned by the ‘placeholder’ builtin are replaced at
build time by the actual store paths corresponding to the specified
outputs.
Previously, you had to work around the inability to self-reference by doing stuff like:
preConfigure = ''
configureFlags+=" --prefix $out --includedir=$dev"
'';
or rely on ad-hoc variable interpolation semantics in Autoconf or Make
(e.g. --prefix=\$(out)), which doesn't always work.
This makes it easier to create a diverted store, i.e.
NIX_REMOTE="local?root=/tmp/root"
instead of
NIX_REMOTE="local?real=/tmp/root/nix/store&state=/tmp/root/nix/var/nix" NIX_LOG_DIR=/tmp/root/nix/var/log
