nix-super/src/libstore/daemon.cc

1118 lines
38 KiB
C++
Raw Normal View History

#include "daemon.hh"
#include "monitor-fd.hh"
#include "worker-protocol.hh"
#include "worker-protocol-impl.hh"
#include "build-result.hh"
#include "store-api.hh"
#include "store-cast.hh"
#include "gc-store.hh"
#include "log-store.hh"
#include "path-with-outputs.hh"
#include "finally.hh"
#include "archive.hh"
#include "derivations.hh"
#include "args.hh"
namespace nix::daemon {
Sink & operator << (Sink & sink, const Logger::Fields & fields)
{
sink << fields.size();
for (auto & f : fields) {
sink << f.type;
if (f.type == Logger::Field::tInt)
sink << f.i;
else if (f.type == Logger::Field::tString)
sink << f.s;
else abort();
}
return sink;
}
/* Logger that forwards log messages to the client, *if* we're in a
state where the protocol allows it (i.e., when canSendStderr is
true). */
struct TunnelLogger : public Logger
{
FdSink & to;
struct State
{
bool canSendStderr = false;
std::vector<std::string> pendingMsgs;
};
Sync<State> state_;
unsigned int clientVersion;
TunnelLogger(FdSink & to, unsigned int clientVersion)
: to(to), clientVersion(clientVersion) { }
void enqueueMsg(const std::string & s)
{
auto state(state_.lock());
if (state->canSendStderr) {
assert(state->pendingMsgs.empty());
try {
to(s);
to.flush();
} catch (...) {
/* Write failed; that means that the other side is
gone. */
state->canSendStderr = false;
throw;
}
} else
state->pendingMsgs.push_back(s);
}
void log(Verbosity lvl, std::string_view s) override
{
if (lvl > verbosity) return;
StringSink buf;
buf << STDERR_NEXT << (s + "\n");
enqueueMsg(buf.s);
}
void logEI(const ErrorInfo & ei) override
{
2020-04-20 02:16:51 +03:00
if (ei.level > verbosity) return;
2020-06-15 15:12:39 +03:00
std::stringstream oss;
2020-06-29 19:20:51 +03:00
showErrorInfo(oss, ei, false);
2020-04-20 02:16:51 +03:00
StringSink buf;
buf << STDERR_NEXT << oss.str();
enqueueMsg(buf.s);
}
/* startWork() means that we're starting an operation for which we
want to send out stderr to the client. */
void startWork()
{
auto state(state_.lock());
state->canSendStderr = true;
for (auto & msg : state->pendingMsgs)
to(msg);
state->pendingMsgs.clear();
to.flush();
}
/* stopWork() means that we're done; stop sending stderr to the
client. */
void stopWork(const Error * ex = nullptr)
{
auto state(state_.lock());
state->canSendStderr = false;
if (!ex)
to << STDERR_LAST;
else {
if (GET_PROTOCOL_MINOR(clientVersion) >= 26) {
to << STDERR_ERROR << *ex;
} else {
to << STDERR_ERROR << ex->what() << ex->status;
}
}
}
void startActivity(ActivityId act, Verbosity lvl, ActivityType type,
const std::string & s, const Fields & fields, ActivityId parent) override
{
if (GET_PROTOCOL_MINOR(clientVersion) < 20) {
if (!s.empty())
log(lvl, s + "...");
return;
}
StringSink buf;
buf << STDERR_START_ACTIVITY << act << lvl << type << s << fields << parent;
enqueueMsg(buf.s);
}
void stopActivity(ActivityId act) override
{
if (GET_PROTOCOL_MINOR(clientVersion) < 20) return;
StringSink buf;
buf << STDERR_STOP_ACTIVITY << act;
enqueueMsg(buf.s);
}
void result(ActivityId act, ResultType type, const Fields & fields) override
{
if (GET_PROTOCOL_MINOR(clientVersion) < 20) return;
StringSink buf;
buf << STDERR_RESULT << act << type << fields;
enqueueMsg(buf.s);
}
};
struct TunnelSink : Sink
{
Sink & to;
TunnelSink(Sink & to) : to(to) { }
2020-12-02 15:00:43 +02:00
void operator () (std::string_view data)
{
to << STDERR_WRITE;
2020-12-02 15:00:43 +02:00
writeString(data, to);
}
};
struct TunnelSource : BufferedSource
{
Source & from;
BufferedSink & to;
TunnelSource(Source & from, BufferedSink & to) : from(from), to(to) { }
size_t readUnbuffered(char * data, size_t len) override
{
to << STDERR_READ << len;
to.flush();
size_t n = readString(data, len, from);
if (n == 0) throw EndOfFile("unexpected end-of-file");
return n;
}
};
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
struct ClientSettings
{
bool keepFailed;
bool keepGoing;
bool tryFallback;
Verbosity verbosity;
unsigned int maxBuildJobs;
time_t maxSilentTime;
bool verboseBuild;
unsigned int buildCores;
bool useSubstitutes;
StringMap overrides;
void apply(TrustedFlag trusted)
{
settings.keepFailed = keepFailed;
settings.keepGoing = keepGoing;
settings.tryFallback = tryFallback;
nix::verbosity = verbosity;
settings.maxBuildJobs.assign(maxBuildJobs);
settings.maxSilentTime = maxSilentTime;
settings.verboseBuild = verboseBuild;
settings.buildCores = buildCores;
settings.useSubstitutes = useSubstitutes;
for (auto & i : overrides) {
auto & name(i.first);
auto & value(i.second);
auto setSubstituters = [&](Setting<Strings> & res) {
if (name != res.name && res.aliases.count(name) == 0)
return false;
StringSet trusted = settings.trustedSubstituters;
for (auto & s : settings.substituters.get())
trusted.insert(s);
Strings subs;
auto ss = tokenizeString<Strings>(value);
for (auto & s : ss)
if (trusted.count(s))
subs.push_back(s);
else if (!hasSuffix(s, "/") && trusted.count(s + "/"))
subs.push_back(s + "/");
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
else
warn("ignoring untrusted substituter '%s', you are not a trusted user.\n"
"Run `man nix.conf` for more information on the `substituters` configuration option.", s);
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
res = subs;
return true;
};
try {
if (name == "ssh-auth-sock") // obsolete
;
else if (name == experimentalFeatureSettings.experimentalFeatures.name) {
// We dont want to forward the experimental features to
// the daemon, as that could cause some pretty weird stuff
if (parseFeatures(tokenizeString<StringSet>(value)) != experimentalFeatureSettings.experimentalFeatures.get())
debug("Ignoring the client-specified experimental features");
} else if (name == settings.pluginFiles.name) {
if (tokenizeString<Paths>(value) != settings.pluginFiles.get())
warn("Ignoring the client-specified plugin-files.\n"
"The client specifying plugins to the daemon never made sense, and was removed in Nix >=2.14.");
}
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
else if (trusted
|| name == settings.buildTimeout.name
|| name == settings.maxSilentTime.name
|| name == settings.pollInterval.name
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
|| name == "connect-timeout"
|| (name == "builders" && value == ""))
settings.set(name, value);
else if (setSubstituters(settings.substituters))
;
else
debug("ignoring the client-specified setting '%s', because it is a restricted setting and you are not a trusted user", name);
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
} catch (UsageError & e) {
warn(e.what());
}
}
}
};
static std::vector<DerivedPath> readDerivedPaths(Store & store, unsigned int clientVersion, WorkerProto::ReadConn conn)
{
2021-04-05 16:48:18 +03:00
std::vector<DerivedPath> reqs;
if (GET_PROTOCOL_MINOR(clientVersion) >= 30) {
reqs = WorkerProto::Serialise<std::vector<DerivedPath>>::read(store, conn);
} else {
for (auto & s : readStrings<Strings>(conn.from))
2021-04-05 16:48:18 +03:00
reqs.push_back(parsePathWithOutputs(store, s).toDerivedPath());
}
return reqs;
}
static void performOp(TunnelLogger * logger, ref<Store> store,
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
TrustedFlag trusted, RecursiveFlag recursive, unsigned int clientVersion,
Source & from, BufferedSink & to, WorkerProto::Op op)
{
WorkerProto::ReadConn rconn { .from = from };
WorkerProto::WriteConn wconn { .to = to };
switch (op) {
case WorkerProto::Op::IsValidPath: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
bool result = store->isValidPath(path);
logger->stopWork();
to << result;
break;
}
case WorkerProto::Op::QueryValidPaths: {
auto paths = WorkerProto::Serialise<StorePathSet>::read(*store, rconn);
SubstituteFlag substitute = NoSubstitute;
if (GET_PROTOCOL_MINOR(clientVersion) >= 27) {
substitute = readInt(from) ? Substitute : NoSubstitute;
}
logger->startWork();
if (substitute) {
store->substitutePaths(paths);
}
auto res = store->queryValidPaths(paths, substitute);
logger->stopWork();
WorkerProto::write(*store, wconn, res);
break;
}
case WorkerProto::Op::HasSubstitutes: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
StorePathSet paths; // FIXME
paths.insert(path);
auto res = store->querySubstitutablePaths(paths);
logger->stopWork();
to << (res.count(path) != 0);
break;
}
case WorkerProto::Op::QuerySubstitutablePaths: {
auto paths = WorkerProto::Serialise<StorePathSet>::read(*store, rconn);
logger->startWork();
auto res = store->querySubstitutablePaths(paths);
logger->stopWork();
WorkerProto::write(*store, wconn, res);
break;
}
case WorkerProto::Op::QueryPathHash: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
auto hash = store->queryPathInfo(path)->narHash;
logger->stopWork();
to << hash.to_string(Base16, false);
break;
}
case WorkerProto::Op::QueryReferences:
case WorkerProto::Op::QueryReferrers:
case WorkerProto::Op::QueryValidDerivers:
case WorkerProto::Op::QueryDerivationOutputs: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
StorePathSet paths;
if (op == WorkerProto::Op::QueryReferences)
for (auto & i : store->queryPathInfo(path)->references)
paths.insert(i);
else if (op == WorkerProto::Op::QueryReferrers)
store->queryReferrers(path, paths);
else if (op == WorkerProto::Op::QueryValidDerivers)
paths = store->queryValidDerivers(path);
else paths = store->queryDerivationOutputs(path);
logger->stopWork();
WorkerProto::write(*store, wconn, paths);
break;
}
case WorkerProto::Op::QueryDerivationOutputNames: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
auto names = store->readDerivation(path).outputNames();
logger->stopWork();
to << names;
break;
}
case WorkerProto::Op::QueryDerivationOutputMap: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
auto outputs = store->queryPartialDerivationOutputMap(path);
logger->stopWork();
WorkerProto::write(*store, wconn, outputs);
break;
}
case WorkerProto::Op::QueryDeriver: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
auto info = store->queryPathInfo(path);
logger->stopWork();
to << (info->deriver ? store->printStorePath(*info->deriver) : "");
break;
}
case WorkerProto::Op::QueryPathFromHashPart: {
auto hashPart = readString(from);
logger->startWork();
auto path = store->queryPathFromHashPart(hashPart);
logger->stopWork();
to << (path ? store->printStorePath(*path) : "");
break;
}
case WorkerProto::Op::AddToStore: {
2020-09-17 20:27:11 +03:00
if (GET_PROTOCOL_MINOR(clientVersion) >= 25) {
auto name = readString(from);
auto camStr = readString(from);
auto refs = WorkerProto::Serialise<StorePathSet>::read(*store, rconn);
2020-09-18 11:06:34 +03:00
bool repairBool;
from >> repairBool;
auto repair = RepairFlag{repairBool};
2020-09-17 20:27:11 +03:00
logger->startWork();
auto pathInfo = [&]() {
2020-09-17 20:27:11 +03:00
// NB: FramedSource must be out of scope before logger->stopWork();
auto [contentAddressMethod, hashType_] = ContentAddressMethod::parse(camStr);
auto hashType = hashType_; // work around clang bug
2020-09-17 20:27:11 +03:00
FramedSource source(from);
// TODO this is essentially RemoteStore::addCAToStore. Move it up to Store.
2020-09-17 20:27:11 +03:00
return std::visit(overloaded {
[&](const TextIngestionMethod &) {
if (hashType != htSHA256)
2023-05-09 20:24:53 +03:00
throw UnimplementedError("When adding text-hashed data called '%s', only SHA-256 is supported but '%s' was given",
name, printHashType(hashType));
2020-09-17 20:27:11 +03:00
// We could stream this by changing Store
std::string contents = source.drain();
auto path = store->addTextToStore(name, contents, refs, repair);
return store->queryPathInfo(path);
2020-09-17 20:27:11 +03:00
},
[&](const FileIngestionMethod & fim) {
auto path = store->addToStoreFromDump(source, name, fim, hashType, repair, refs);
return store->queryPathInfo(path);
2020-09-17 20:27:11 +03:00
},
}, contentAddressMethod.raw);
2020-09-17 20:27:11 +03:00
}();
logger->stopWork();
pathInfo->write(to, *store, GET_PROTOCOL_MINOR(clientVersion));
2020-09-17 20:27:11 +03:00
} else {
HashType hashAlgo;
std::string baseName;
FileIngestionMethod method;
{
bool fixed;
uint8_t recursive;
std::string hashAlgoRaw;
from >> baseName >> fixed /* obsolete */ >> recursive >> hashAlgoRaw;
if (recursive > (uint8_t) FileIngestionMethod::Recursive)
throw Error("unsupported FileIngestionMethod with value of %i; you may need to upgrade nix-daemon", recursive);
method = FileIngestionMethod { recursive };
/* Compatibility hack. */
if (!fixed) {
hashAlgoRaw = "sha256";
method = FileIngestionMethod::Recursive;
}
hashAlgo = parseHashType(hashAlgoRaw);
}
auto dumpSource = sinkToSource([&](Sink & saved) {
if (method == FileIngestionMethod::Recursive) {
/* We parse the NAR dump through into `saved` unmodified,
so why all this extra work? We still parse the NAR so
that we aren't sending arbitrary data to `saved`
unwittingly`, and we know when the NAR ends so we don't
consume the rest of `from` and can't parse another
command. (We don't trust `addToStoreFromDump` to not
eagerly consume the entire stream it's given, past the
length of the Nar. */
TeeSource savedNARSource(from, saved);
ParseSink sink; /* null sink; just parse the NAR */
parseDump(sink, savedNARSource);
} else {
/* Incrementally parse the NAR file, stripping the
metadata, and streaming the sole file we expect into
`saved`. */
RetrieveRegularNARSink savedRegular { saved };
parseDump(savedRegular, from);
if (!savedRegular.regular) throw Error("regular file expected");
}
});
2020-09-17 20:27:11 +03:00
logger->startWork();
auto path = store->addToStoreFromDump(*dumpSource, baseName, method, hashAlgo);
2020-09-17 20:27:11 +03:00
logger->stopWork();
2020-09-17 20:27:11 +03:00
to << store->printStorePath(path);
}
break;
}
case WorkerProto::Op::AddMultipleToStore: {
bool repair, dontCheckSigs;
from >> repair >> dontCheckSigs;
if (!trusted && dontCheckSigs)
dontCheckSigs = false;
logger->startWork();
{
FramedSource source(from);
store->addMultipleToStore(source,
RepairFlag{repair},
dontCheckSigs ? NoCheckSigs : CheckSigs);
}
logger->stopWork();
break;
}
case WorkerProto::Op::AddTextToStore: {
std::string suffix = readString(from);
std::string s = readString(from);
auto refs = WorkerProto::Serialise<StorePathSet>::read(*store, rconn);
logger->startWork();
auto path = store->addTextToStore(suffix, s, refs, NoRepair);
logger->stopWork();
to << store->printStorePath(path);
break;
}
case WorkerProto::Op::ExportPath: {
auto path = store->parseStorePath(readString(from));
readInt(from); // obsolete
logger->startWork();
TunnelSink sink(to);
store->exportPath(path, sink);
logger->stopWork();
to << 1;
break;
}
case WorkerProto::Op::ImportPaths: {
logger->startWork();
TunnelSource source(from, to);
auto paths = store->importPaths(source,
trusted ? NoCheckSigs : CheckSigs);
logger->stopWork();
Strings paths2;
for (auto & i : paths) paths2.push_back(store->printStorePath(i));
to << paths2;
break;
}
case WorkerProto::Op::BuildPaths: {
auto drvs = readDerivedPaths(*store, clientVersion, rconn);
BuildMode mode = bmNormal;
if (GET_PROTOCOL_MINOR(clientVersion) >= 15) {
mode = (BuildMode) readInt(from);
/* Repairing is not atomic, so disallowed for "untrusted"
Move `trustedUsers` and `allowedUsers` to separate config struct These settings are not needed for libstore at all, they are just used by the nix daemon *command* for authorization on unix domain sockets. My moving them to a new configuration struct just in that file, we avoid them leaking anywhere else. Also, it is good to break up the mammoth `Settings` struct in general. Issue #5638 tracks this. The message is not changed because I do not want to regress in convenience to the user. Just saying "this connection is not trusted" doesn't tell them out to fix the issue. The ideal thing to do would be to somehow parameterize `processCommand` on how the error should be displayed, so different sorts of connections can display different information to the user based on how authentication is performed for the connection in question. This, however, is a good bit more work, so it is left for the future. This came up with me thinking about the tcp:// store (#5265). The larger project is not TCP *per se*, but the idea that it should be possible for something else to manage access control to services like the Nix Daemon, and those services simply trust or trust the incoming connection as they are told. This is a more capability-oriented way of thinking about trust than "every server implements its own auth separately" as we are used to today. Its very great that libstore itself already implements just this model, and so via this refactor I basically want to "enshrine" that so it continues to be the case.
2023-02-02 20:20:44 +02:00
clients.
FIXME: layer violation in this message: the daemon code (i.e.
this file) knows whether a client/connection is trusted, but it
does not how how the client was authenticated. The mechanism
need not be getting the UID of the other end of a Unix Domain
Socket.
*/
if (mode == bmRepair && !trusted)
throw Error("repairing is not allowed because you are not in 'trusted-users'");
}
logger->startWork();
store->buildPaths(drvs, mode);
logger->stopWork();
to << 1;
break;
}
case WorkerProto::Op::BuildPathsWithResults: {
auto drvs = readDerivedPaths(*store, clientVersion, rconn);
BuildMode mode = bmNormal;
mode = (BuildMode) readInt(from);
/* Repairing is not atomic, so disallowed for "untrusted"
Move `trustedUsers` and `allowedUsers` to separate config struct These settings are not needed for libstore at all, they are just used by the nix daemon *command* for authorization on unix domain sockets. My moving them to a new configuration struct just in that file, we avoid them leaking anywhere else. Also, it is good to break up the mammoth `Settings` struct in general. Issue #5638 tracks this. The message is not changed because I do not want to regress in convenience to the user. Just saying "this connection is not trusted" doesn't tell them out to fix the issue. The ideal thing to do would be to somehow parameterize `processCommand` on how the error should be displayed, so different sorts of connections can display different information to the user based on how authentication is performed for the connection in question. This, however, is a good bit more work, so it is left for the future. This came up with me thinking about the tcp:// store (#5265). The larger project is not TCP *per se*, but the idea that it should be possible for something else to manage access control to services like the Nix Daemon, and those services simply trust or trust the incoming connection as they are told. This is a more capability-oriented way of thinking about trust than "every server implements its own auth separately" as we are used to today. Its very great that libstore itself already implements just this model, and so via this refactor I basically want to "enshrine" that so it continues to be the case.
2023-02-02 20:20:44 +02:00
clients.
FIXME: layer violation; see above. */
if (mode == bmRepair && !trusted)
throw Error("repairing is not allowed because you are not in 'trusted-users'");
logger->startWork();
auto results = store->buildPathsWithResults(drvs, mode);
logger->stopWork();
WorkerProto::write(*store, wconn, results);
break;
}
case WorkerProto::Op::BuildDerivation: {
auto drvPath = store->parseStorePath(readString(from));
BasicDerivation drv;
readDerivation(from, *store, drv, Derivation::nameFromPath(drvPath));
BuildMode buildMode = (BuildMode) readInt(from);
logger->startWork();
auto drvType = drv.type();
/* Content-addressed derivations are trustless because their output paths
are verified by their content alone, so any derivation is free to
try to produce such a path.
Input-addressed derivation output paths, however, are calculated
from the derivation closure that produced them---even knowing the
root derivation is not enough. That the output data actually came
from those derivations is fundamentally unverifiable, but the daemon
trusts itself on that matter. The question instead is whether the
submitted plan has rights to the output paths it wants to fill, and
at least the derivation closure proves that.
It would have been nice if input-address algorithm merely depended
on the build time closure, rather than depending on the derivation
closure. That would mean input-addressed paths used at build time
would just be trusted and not need their own evidence. This is in
fact fine as the same guarantees would hold *inductively*: either
the remote builder has those paths and already trusts them, or it
needs to build them too and thus their evidence must be provided in
turn. The advantage of this variant algorithm is that the evidence
for input-addressed paths which the remote builder already has
doesn't need to be sent again.
That said, now that we have floating CA derivations, it is better
that people just migrate to those which also solve this problem, and
others. It's the same migration difficulty with strictly more
benefit.
Lastly, do note that when we parse fixed-output content-addressed
derivations, we throw out the precomputed output paths and just
store the hashes, so there aren't two competing sources of truth an
attacker could exploit. */
if (!(drvType.isCA() || trusted))
throw Error("you are not privileged to build input-addressed derivations");
/* Make sure that the non-input-addressed derivations that got this far
are in fact content-addressed if we don't trust them. */
assert(drvType.isCA() || trusted);
/* Recompute the derivation path when we cannot trust the original. */
if (!trusted) {
/* Recomputing the derivation path for input-address derivations
makes it harder to audit them after the fact, since we need the
original not-necessarily-resolved derivation to verify the drv
derivation as adequate claim to the input-addressed output
paths. */
assert(drvType.isCA());
Derivation drv2;
static_cast<BasicDerivation &>(drv2) = drv;
drvPath = writeDerivation(*store, Derivation { drv2 });
}
auto res = store->buildDerivation(drvPath, drv, buildMode);
logger->stopWork();
to << res.status << res.errorMsg;
if (GET_PROTOCOL_MINOR(clientVersion) >= 29) {
2021-03-22 17:18:48 +02:00
to << res.timesBuilt << res.isNonDeterministic << res.startTime << res.stopTime;
}
if (GET_PROTOCOL_MINOR(clientVersion) >= 28) {
DrvOutputs builtOutputs;
for (auto & [output, realisation] : res.builtOutputs)
builtOutputs.insert_or_assign(realisation.id, realisation);
WorkerProto::write(*store, wconn, builtOutputs);
}
break;
}
case WorkerProto::Op::EnsurePath: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
store->ensurePath(path);
logger->stopWork();
to << 1;
break;
}
case WorkerProto::Op::AddTempRoot: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
store->addTempRoot(path);
logger->stopWork();
to << 1;
break;
}
case WorkerProto::Op::AddIndirectRoot: {
Path path = absPath(readString(from));
logger->startWork();
auto & gcStore = require<GcStore>(*store);
gcStore.addIndirectRoot(path);
logger->stopWork();
to << 1;
break;
}
2021-08-17 19:53:14 +03:00
// Obsolete.
case WorkerProto::Op::SyncWithGC: {
logger->startWork();
logger->stopWork();
to << 1;
break;
}
case WorkerProto::Op::FindRoots: {
logger->startWork();
auto & gcStore = require<GcStore>(*store);
Roots roots = gcStore.findRoots(!trusted);
logger->stopWork();
size_t size = 0;
for (auto & i : roots)
size += i.second.size();
to << size;
for (auto & [target, links] : roots)
for (auto & link : links)
to << link << store->printStorePath(target);
break;
}
case WorkerProto::Op::CollectGarbage: {
GCOptions options;
options.action = (GCOptions::GCAction) readInt(from);
options.pathsToDelete = WorkerProto::Serialise<StorePathSet>::read(*store, rconn);
from >> options.ignoreLiveness >> options.maxFreed;
// obsolete fields
readInt(from);
readInt(from);
readInt(from);
GCResults results;
logger->startWork();
if (options.ignoreLiveness)
throw Error("you are not allowed to ignore liveness");
auto & gcStore = require<GcStore>(*store);
gcStore.collectGarbage(options, results);
logger->stopWork();
to << results.paths << results.bytesFreed << 0 /* obsolete */;
break;
}
case WorkerProto::Op::SetOptions: {
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
ClientSettings clientSettings;
clientSettings.keepFailed = readInt(from);
clientSettings.keepGoing = readInt(from);
clientSettings.tryFallback = readInt(from);
clientSettings.verbosity = (Verbosity) readInt(from);
clientSettings.maxBuildJobs = readInt(from);
clientSettings.maxSilentTime = readInt(from);
readInt(from); // obsolete useBuildHook
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
clientSettings.verboseBuild = lvlError == (Verbosity) readInt(from);
readInt(from); // obsolete logType
readInt(from); // obsolete printBuildTrace
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
clientSettings.buildCores = readInt(from);
clientSettings.useSubstitutes = readInt(from);
if (GET_PROTOCOL_MINOR(clientVersion) >= 12) {
unsigned int n = readInt(from);
for (unsigned int i = 0; i < n; i++) {
auto name = readString(from);
auto value = readString(from);
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
clientSettings.overrides.emplace(name, value);
}
}
logger->startWork();
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
// FIXME: use some setting in recursive mode. Will need to use
// non-global variables.
if (!recursive)
clientSettings.apply(trusted);
logger->stopWork();
break;
}
case WorkerProto::Op::QuerySubstitutablePathInfo: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
SubstitutablePathInfos infos;
store->querySubstitutablePathInfos({{path, std::nullopt}}, infos);
logger->stopWork();
auto i = infos.find(path);
if (i == infos.end())
to << 0;
else {
to << 1
<< (i->second.deriver ? store->printStorePath(*i->second.deriver) : "");
WorkerProto::write(*store, wconn, i->second.references);
to << i->second.downloadSize
<< i->second.narSize;
}
break;
}
case WorkerProto::Op::QuerySubstitutablePathInfos: {
SubstitutablePathInfos infos;
StorePathCAMap pathsMap = {};
if (GET_PROTOCOL_MINOR(clientVersion) < 22) {
auto paths = WorkerProto::Serialise<StorePathSet>::read(*store, rconn);
for (auto & path : paths)
pathsMap.emplace(path, std::nullopt);
} else
pathsMap = WorkerProto::Serialise<StorePathCAMap>::read(*store, rconn);
logger->startWork();
store->querySubstitutablePathInfos(pathsMap, infos);
logger->stopWork();
to << infos.size();
for (auto & i : infos) {
to << store->printStorePath(i.first)
<< (i.second.deriver ? store->printStorePath(*i.second.deriver) : "");
WorkerProto::write(*store, wconn, i.second.references);
to << i.second.downloadSize << i.second.narSize;
}
break;
}
case WorkerProto::Op::QueryAllValidPaths: {
logger->startWork();
auto paths = store->queryAllValidPaths();
logger->stopWork();
WorkerProto::write(*store, wconn, paths);
break;
}
case WorkerProto::Op::QueryPathInfo: {
auto path = store->parseStorePath(readString(from));
std::shared_ptr<const ValidPathInfo> info;
logger->startWork();
try {
info = store->queryPathInfo(path);
} catch (InvalidPath &) {
if (GET_PROTOCOL_MINOR(clientVersion) < 17) throw;
}
logger->stopWork();
if (info) {
if (GET_PROTOCOL_MINOR(clientVersion) >= 17)
to << 1;
info->write(to, *store, GET_PROTOCOL_MINOR(clientVersion), false);
} else {
assert(GET_PROTOCOL_MINOR(clientVersion) >= 17);
to << 0;
}
break;
}
case WorkerProto::Op::OptimiseStore:
logger->startWork();
store->optimiseStore();
logger->stopWork();
to << 1;
break;
case WorkerProto::Op::VerifyStore: {
bool checkContents, repair;
from >> checkContents >> repair;
logger->startWork();
if (repair && !trusted)
throw Error("you are not privileged to repair paths");
bool errors = store->verifyStore(checkContents, (RepairFlag) repair);
logger->stopWork();
to << errors;
break;
}
case WorkerProto::Op::AddSignatures: {
auto path = store->parseStorePath(readString(from));
StringSet sigs = readStrings<StringSet>(from);
logger->startWork();
store->addSignatures(path, sigs);
logger->stopWork();
to << 1;
break;
}
case WorkerProto::Op::NarFromPath: {
auto path = store->parseStorePath(readString(from));
logger->startWork();
logger->stopWork();
2020-08-15 00:59:31 +03:00
dumpPath(store->toRealPath(path), to);
break;
}
case WorkerProto::Op::AddToStoreNar: {
bool repair, dontCheckSigs;
2020-08-06 21:31:48 +03:00
auto path = store->parseStorePath(readString(from));
auto deriver = readString(from);
2020-08-06 21:31:48 +03:00
auto narHash = Hash::parseAny(readString(from), htSHA256);
ValidPathInfo info { path, narHash };
if (deriver != "")
info.deriver = store->parseStorePath(deriver);
info.references = WorkerProto::Serialise<StorePathSet>::read(*store, rconn);
from >> info.registrationTime >> info.narSize >> info.ultimate;
info.sigs = readStrings<StringSet>(from);
info.ca = ContentAddress::parseOpt(readString(from));
2020-06-02 03:37:43 +03:00
from >> repair >> dontCheckSigs;
if (!trusted && dontCheckSigs)
dontCheckSigs = false;
if (!trusted)
info.ultimate = false;
if (GET_PROTOCOL_MINOR(clientVersion) >= 23) {
logger->startWork();
{
FramedSource source(from);
store->addToStore(info, source, (RepairFlag) repair,
dontCheckSigs ? NoCheckSigs : CheckSigs);
}
logger->stopWork();
}
else {
std::unique_ptr<Source> source;
StringSink saved;
if (GET_PROTOCOL_MINOR(clientVersion) >= 21)
source = std::make_unique<TunnelSource>(from, to);
else {
TeeSource tee { from, saved };
ParseSink ether;
parseDump(ether, tee);
source = std::make_unique<StringSource>(saved.s);
}
logger->startWork();
// FIXME: race if addToStore doesn't read source?
store->addToStore(info, *source, (RepairFlag) repair,
dontCheckSigs ? NoCheckSigs : CheckSigs);
logger->stopWork();
}
break;
}
case WorkerProto::Op::QueryMissing: {
auto targets = readDerivedPaths(*store, clientVersion, rconn);
logger->startWork();
StorePathSet willBuild, willSubstitute, unknown;
2020-07-30 14:10:49 +03:00
uint64_t downloadSize, narSize;
store->queryMissing(targets, willBuild, willSubstitute, unknown, downloadSize, narSize);
logger->stopWork();
WorkerProto::write(*store, wconn, willBuild);
WorkerProto::write(*store, wconn, willSubstitute);
WorkerProto::write(*store, wconn, unknown);
to << downloadSize << narSize;
break;
}
case WorkerProto::Op::RegisterDrvOutput: {
logger->startWork();
if (GET_PROTOCOL_MINOR(clientVersion) < 31) {
auto outputId = DrvOutput::parse(readString(from));
auto outputPath = StorePath(readString(from));
store->registerDrvOutput(Realisation{
.id = outputId, .outPath = outputPath});
} else {
auto realisation = WorkerProto::Serialise<Realisation>::read(*store, rconn);
store->registerDrvOutput(realisation);
}
logger->stopWork();
break;
}
case WorkerProto::Op::QueryRealisation: {
logger->startWork();
auto outputId = DrvOutput::parse(readString(from));
auto info = store->queryRealisation(outputId);
logger->stopWork();
if (GET_PROTOCOL_MINOR(clientVersion) < 31) {
std::set<StorePath> outPaths;
if (info) outPaths.insert(info->outPath);
WorkerProto::write(*store, wconn, outPaths);
} else {
std::set<Realisation> realisations;
if (info) realisations.insert(*info);
WorkerProto::write(*store, wconn, realisations);
}
break;
}
case WorkerProto::Op::AddBuildLog: {
StorePath path{readString(from)};
logger->startWork();
if (!trusted)
throw Error("you are not privileged to add logs");
auto & logStore = require<LogStore>(*store);
{
FramedSource source(from);
StringSink sink;
source.drainInto(sink);
logStore.addBuildLog(path, sink.s);
}
logger->stopWork();
to << 1;
break;
}
case WorkerProto::Op::QueryFailedPaths:
case WorkerProto::Op::ClearFailedPaths:
throw Error("Removed operation %1%", op);
default:
throw Error("invalid operation %1%", op);
}
}
void processConnection(
ref<Store> store,
FdSource & from,
FdSink & to,
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
TrustedFlag trusted,
RecursiveFlag recursive)
{
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
auto monitor = !recursive ? std::make_unique<MonitorFdHup>(from.fd) : nullptr;
/* Exchange the greeting. */
unsigned int magic = readInt(from);
if (magic != WORKER_MAGIC_1) throw Error("protocol mismatch");
to << WORKER_MAGIC_2 << PROTOCOL_VERSION;
to.flush();
unsigned int clientVersion = readInt(from);
if (clientVersion < 0x10a)
throw Error("the Nix client version is too old");
auto tunnelLogger = new TunnelLogger(to, clientVersion);
auto prevLogger = nix::logger;
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
// FIXME
if (!recursive)
logger = tunnelLogger;
unsigned int opCount = 0;
Finally finally([&]() {
_isInterrupted = false;
printMsgUsing(prevLogger, lvlDebug, "%d operations", opCount);
});
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
if (GET_PROTOCOL_MINOR(clientVersion) >= 14 && readInt(from)) {
// Obsolete CPU affinity.
readInt(from);
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
}
if (GET_PROTOCOL_MINOR(clientVersion) >= 11)
readInt(from); // obsolete reserveSpace
if (GET_PROTOCOL_MINOR(clientVersion) >= 33)
to << nixVersion;
if (GET_PROTOCOL_MINOR(clientVersion) >= 35) {
// We and the underlying store both need to trust the client for
// it to be trusted.
auto temp = trusted
? store->isTrustedClient()
: std::optional { NotTrusted };
WorkerProto::WriteConn wconn { .to = to };
WorkerProto::write(*store, wconn, temp);
}
/* Send startup error messages to the client. */
tunnelLogger->startWork();
try {
tunnelLogger->stopWork();
to.flush();
/* Process client requests. */
while (true) {
WorkerProto::Op op;
try {
op = (enum WorkerProto::Op) readInt(from);
} catch (Interrupted & e) {
break;
} catch (EndOfFile & e) {
break;
}
printMsgUsing(prevLogger, lvlDebug, "received daemon op %d", op);
opCount++;
debug("performing daemon worker op: %d", op);
try {
Recursive Nix support This allows Nix builders to call Nix to build derivations, with some limitations. Example: let nixpkgs = fetchTarball channel:nixos-18.03; in with import <nixpkgs> {}; runCommand "foo" { buildInputs = [ nix jq ]; NIX_PATH = "nixpkgs=${nixpkgs}"; } '' hello=$(nix-build -E '(import <nixpkgs> {}).hello.overrideDerivation (args: { name = "hello-3.5"; })') $hello/bin/hello mkdir -p $out/bin ln -s $hello/bin/hello $out/bin/hello nix path-info -r --json $hello | jq . '' This derivation makes a recursive Nix call to build GNU Hello and symlinks it from its $out, i.e. # ll ./result/bin/ lrwxrwxrwx 1 root root 63 Jan 1 1970 hello -> /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5/bin/hello # nix-store -qR ./result /nix/store/hwwqshlmazzjzj7yhrkyjydxamvvkfd3-glibc-2.26-131 /nix/store/s0awxrs71gickhaqdwxl506hzccb30y5-hello-3.5 /nix/store/sgmvvyw8vhfqdqb619bxkcpfn9lvd8ss-foo This is implemented as follows: * Before running the outer builder, Nix creates a Unix domain socket '.nix-socket' in the builder's temporary directory and sets $NIX_REMOTE to point to it. It starts a thread to process connections to this socket. (Thus you don't need to have nix-daemon running.) * The daemon thread uses a wrapper store (RestrictedStore) to keep track of paths added through recursive Nix calls, to implement some restrictions (see below), and to do some censorship (e.g. for purity, queryPathInfo() won't return impure information such as signatures and timestamps). * After the build finishes, the output paths are scanned for references to the paths added through recursive Nix calls (in addition to the inputs closure). Thus, in the example above, $out has a reference to $hello. The main restriction on recursive Nix calls is that they cannot do arbitrary substitutions. For example, doing nix-store -r /nix/store/kmwd1hq55akdb9sc7l3finr175dajlby-hello-2.10 is forbidden unless /nix/store/kmwd... is in the inputs closure or previously built by a recursive Nix call. This is to prevent irreproducible derivations that have hidden dependencies on substituters or the current store contents. Building a derivation is fine, however, and Nix will use substitutes if available. In other words, the builder has to present proof that it knows how to build a desired store path from scratch by constructing a derivation graph for that path. Probably we should also disallow instantiating/building fixed-output derivations (specifically, those that access the network, but currently we have no way to mark fixed-output derivations that don't access the network). Otherwise sandboxed derivations can bypass sandbox restrictions and access the network. When sandboxing is enabled, we make paths appear in the sandbox of the builder by entering the mount namespace of the builder and bind-mounting each path. This is tricky because we do a pivot_root() in the builder to change the root directory of its mount namespace, and thus the host /nix/store is not visible in the mount namespace of the builder. To get around this, just before doing pivot_root(), we branch a second mount namespace that shares its /nix/store mountpoint with the parent. Recursive Nix currently doesn't work on macOS in sandboxed mode (because we can't change the sandbox policy of a running build) and in non-root mode (because setns() barfs).
2018-10-02 17:01:26 +03:00
performOp(tunnelLogger, store, trusted, recursive, clientVersion, from, to, op);
} catch (Error & e) {
/* If we're not in a state where we can send replies, then
something went wrong processing the input of the
client. This can happen especially if I/O errors occur
during addTextToStore() / importPath(). If that
happens, just send the error message and exit. */
bool errorAllowed = tunnelLogger->state_.lock()->canSendStderr;
tunnelLogger->stopWork(&e);
if (!errorAllowed) throw;
} catch (std::bad_alloc & e) {
auto ex = Error("Nix daemon out of memory");
tunnelLogger->stopWork(&ex);
throw;
}
to.flush();
assert(!tunnelLogger->state_.lock()->canSendStderr);
};
} catch (Error & e) {
tunnelLogger->stopWork(&e);
to.flush();
return;
} catch (std::exception & e) {
auto ex = Error(e.what());
tunnelLogger->stopWork(&ex);
to.flush();
return;
}
}
}