Merge pull request #2798 from grahamc/diff-hook

build: run diff-hook under --check and document diff-hook

6 files changed, 338 insertions, 17 deletions

diff --git a/doc/manual/advanced-topics/advanced-topics.xml b/doc/manual/advanced-topics/advanced-topics.xml
index b710f9f2b518..c304367aaf8a 100644
--- a/doc/manual/advanced-topics/advanced-topics.xml
+++ b/doc/manual/advanced-topics/advanced-topics.xml
@@ -7,5 +7,6 @@
 <title>Advanced Topics</title>
 
 <xi:include href="distributed-builds.xml" />
+<xi:include href="diff-hook.xml" />
 
 </part>
diff --git a/doc/manual/advanced-topics/diff-hook.xml b/doc/manual/advanced-topics/diff-hook.xml
new file mode 100644
index 000000000000..fb4bf819f94b
--- /dev/null
+++ b/doc/manual/advanced-topics/diff-hook.xml
@@ -0,0 +1,205 @@
+<chapter xmlns="http://docbook.org/ns/docbook"
+      xmlns:xlink="http://www.w3.org/1999/xlink"
+      xmlns:xi="http://www.w3.org/2001/XInclude"
+      xml:id="chap-diff-hook"
+      version="5.0"
+      >
+
+<title>Verifying Build Reproducibility with <option linkend="conf-diff-hook">diff-hook</option></title>
+
+<subtitle>Check build reproducibility by running builds multiple times
+and comparing their results.</subtitle>
+
+<para>Specify a program with Nix's <xref linkend="conf-diff-hook" /> to
+compare build results when two builds produce different results. Note:
+this hook is only executed if the results are not the same, this hook
+is not used for determining if the results are the same.</para>
+
+<para>For purposes of demonstration, we'll use the following Nix file,
+<filename>deterministic.nix</filename> for testing:</para>
+
+<programlisting>
+let
+  inherit (import &lt;nixpkgs&gt; {}) runCommand;
+in {
+  stable = runCommand "stable" {} ''
+    touch $out
+  '';
+
+  unstable = runCommand "unstable" {} ''
+    echo $RANDOM > $out
+  '';
+}
+</programlisting>
+
+<para>Additionally, <filename>nix.conf</filename> contains:
+
+<programlisting>
+diff-hook = /etc/nix/my-diff-hook
+run-diff-hook = true
+</programlisting>
+
+where <filename>/etc/nix/my-diff-hook</filename> is an executable
+file containing:
+
+<programlisting>
+#!/bin/sh
+exec &gt;&amp;2
+echo "For derivation $3:"
+/run/current-system/sw/bin/diff -r "$1" "$2"
+</programlisting>
+
+</para>
+
+<para>The diff hook is executed by the same user and group who ran the
+build. However, the diff hook does not have write access to the store
+path just built.</para>
+
+<section>
+  <title>
+    Spot-Checking Build Determinism
+  </title>
+
+  <para>
+    Verify a path which already exists in the Nix store by passing
+    <option>--check</option> to the build command.
+  </para>
+
+  <para>If the build passes and is deterministic, Nix will exit with a
+  status code of 0:</para>
+
+  <screen>
+$ nix-build ./deterministic.nix -A stable
+these derivations will be built:
+  /nix/store/z98fasz2jqy9gs0xbvdj939p27jwda38-stable.drv
+building '/nix/store/z98fasz2jqy9gs0xbvdj939p27jwda38-stable.drv'...
+/nix/store/yyxlzw3vqaas7wfp04g0b1xg51f2czgq-stable
+
+$ nix-build ./deterministic.nix -A stable --check
+checking outputs of '/nix/store/z98fasz2jqy9gs0xbvdj939p27jwda38-stable.drv'...
+/nix/store/yyxlzw3vqaas7wfp04g0b1xg51f2czgq-stable
+</screen>
+
+  <para>If the build is not deterministic, Nix will exit with a status
+  code of 1:</para>
+
+  <screen>
+$ nix-build ./deterministic.nix -A unstable
+these derivations will be built:
+  /nix/store/cgl13lbj1w368r5z8gywipl1ifli7dhk-unstable.drv
+building '/nix/store/cgl13lbj1w368r5z8gywipl1ifli7dhk-unstable.drv'...
+/nix/store/krpqk0l9ib0ibi1d2w52z293zw455cap-unstable
+
+$ nix-build ./deterministic.nix -A unstable --check
+checking outputs of '/nix/store/cgl13lbj1w368r5z8gywipl1ifli7dhk-unstable.drv'...
+error: derivation '/nix/store/cgl13lbj1w368r5z8gywipl1ifli7dhk-unstable.drv' may not be deterministic: output '/nix/store/krpqk0l9ib0ibi1d2w52z293zw455cap-unstable' differs
+</screen>
+
+<para>In the Nix daemon's log, we will now see:
+<screen>
+For derivation /nix/store/cgl13lbj1w368r5z8gywipl1ifli7dhk-unstable.drv:
+1c1
+&lt; 8108
+---
+&gt; 30204
+</screen>
+</para>
+
+  <para>Using <option>--check</option> with <option>--keep-failed</option>
+  will cause Nix to keep the second build's output in a special,
+  <literal>.check</literal> path:</para>
+
+  <screen>
+$ nix-build ./deterministic.nix -A unstable --check --keep-failed
+checking outputs of '/nix/store/cgl13lbj1w368r5z8gywipl1ifli7dhk-unstable.drv'...
+note: keeping build directory '/tmp/nix-build-unstable.drv-0'
+error: derivation '/nix/store/cgl13lbj1w368r5z8gywipl1ifli7dhk-unstable.drv' may not be deterministic: output '/nix/store/krpqk0l9ib0ibi1d2w52z293zw455cap-unstable' differs from '/nix/store/krpqk0l9ib0ibi1d2w52z293zw455cap-unstable.check'
+</screen>
+
+  <para>In particular, notice the
+  <literal>/nix/store/krpqk0l9ib0ibi1d2w52z293zw455cap-unstable.check</literal>
+  output. Nix has copied the build results to that directory where you
+  can examine it.</para>
+
+  <note xml:id="check-dirs-are-unregistered">
+    <title><literal>.check</literal> paths are not registered store paths</title>
+
+    <para>Check paths are not protected against garbage collection,
+    and this path will be deleted on the next garbage collection.</para>
+
+    <para>The path is guaranteed to be alive for the duration of
+    <xref linkend="conf-diff-hook" />'s execution, but may be deleted
+    any time after.</para>
+
+    <para>If the comparison is performed as part of automated tooling,
+    please use the diff-hook or author your tooling to handle the case
+    where the build was not deterministic and also a check path does
+    not exist.</para>
+  </note>
+
+  <para>
+    <option>--check</option> is only usable if the derivation has
+    been built on the system already. If the derivation has not been
+    built Nix will fail with the error:
+    <screen>
+error: some outputs of '/nix/store/hzi1h60z2qf0nb85iwnpvrai3j2w7rr6-unstable.drv' are not valid, so checking is not possible
+</screen>
+
+    Run the build without <option>--check</option>, and then try with
+    <option>--check</option> again.
+  </para>
+</section>
+
+<section>
+  <title>
+    Automatic and Optionally Enforced Determinism Verification
+  </title>
+
+  <para>
+    Automatically verify every build at build time by executing the
+    build multiple times.
+  </para>
+
+  <para>
+    Setting <xref linkend="conf-repeat" /> and
+    <xref linkend="conf-enforce-determinism" /> in your
+    <filename>nix.conf</filename> permits the automated verification
+    of every build Nix performs.
+  </para>
+
+  <para>
+    The following configuration will run each build three times, and
+    will require the build to be deterministic:
+
+    <programlisting>
+enforce-determinism = true
+repeat = 2
+</programlisting>
+  </para>
+
+  <para>
+    Setting <xref linkend="conf-enforce-determinism" /> to false as in
+    the following configuration will run the build multiple times,
+    execute the build hook, but will allow the build to succeed even
+    if it does not build reproducibly:
+
+    <programlisting>
+enforce-determinism = false
+repeat = 1
+</programlisting>
+  </para>
+
+  <para>
+    An example output of this configuration:
+    <screen>
+$ nix-build ./test.nix -A unstable
+these derivations will be built:
+  /nix/store/ch6llwpr2h8c3jmnf3f2ghkhx59aa97f-unstable.drv
+building '/nix/store/ch6llwpr2h8c3jmnf3f2ghkhx59aa97f-unstable.drv' (round 1/2)...
+building '/nix/store/ch6llwpr2h8c3jmnf3f2ghkhx59aa97f-unstable.drv' (round 2/2)...
+output '/nix/store/6xg356v9gl03hpbbg8gws77n19qanh02-unstable' of '/nix/store/ch6llwpr2h8c3jmnf3f2ghkhx59aa97f-unstable.drv' differs from '/nix/store/6xg356v9gl03hpbbg8gws77n19qanh02-unstable.check' from previous round
+/nix/store/6xg356v9gl03hpbbg8gws77n19qanh02-unstable
+</screen>
+  </para>
+</section>
+</chapter>
diff --git a/doc/manual/command-ref/conf-file.xml b/doc/manual/command-ref/conf-file.xml
index f0da1f612fee..24fbf28cff25 100644
--- a/doc/manual/command-ref/conf-file.xml
+++ b/doc/manual/command-ref/conf-file.xml
@@ -1,7 +1,9 @@
+<?xml version="1.0" encoding="utf-8"?>
 <refentry xmlns="http://docbook.org/ns/docbook"
           xmlns:xlink="http://www.w3.org/1999/xlink"
           xmlns:xi="http://www.w3.org/2001/XInclude"
-          xml:id="sec-conf-file">
+          xml:id="sec-conf-file"
+          version="5">
 
 <refmeta>
   <refentrytitle>nix.conf</refentrytitle>
@@ -240,6 +242,71 @@ false</literal>.</para>
 
   </varlistentry>
 
+  <varlistentry xml:id="conf-diff-hook"><term><literal>diff-hook</literal></term>
+  <listitem>
+    <para>
+      Absolute path to an executable capable of diffing build results.
+      The hook executes if <xref linkend="conf-run-diff-hook" /> is
+      true, and the output of a build is known to not be the same.
+      This program is not executed to determine if two results are the
+      same.
+    </para>
+
+    <para>
+      The diff hook is executed by the same user and group who ran the
+      build. However, the diff hook does not have write access to the
+      store path just built.
+    </para>
+
+    <para>The diff hook program receives three parameters:</para>
+
+    <orderedlist>
+      <listitem>
+        <para>
+          A path to the previous build's results
+        </para>
+      </listitem>
+
+      <listitem>
+        <para>
+          A path to the current build's results
+        </para>
+      </listitem>
+
+      <listitem>
+        <para>
+          The path to the build's derivation
+        </para>
+      </listitem>
+
+      <listitem>
+        <para>
+          The path to the build's scratch directory. This directory
+          will exist only if the build was run with
+          <option>--keep-failed</option>.
+        </para>
+      </listitem>
+    </orderedlist>
+
+    <para>
+      The stderr and stdout output from the diff hook will not be
+      displayed to the user. Instead, it will print to the nix-daemon's
+      log.
+    </para>
+
+    <para>When using the Nix daemon, <literal>diff-hook</literal> must
+    be set in the <filename>nix.conf</filename> configuration file, and
+    cannot be passed at the command line.
+    </para>
+  </listitem>
+  </varlistentry>
+
+  <varlistentry xml:id="conf-enforce-determinism">
+    <term><literal>enforce-determinism</literal></term>
+
+    <listitem><para>See <xref linkend="conf-repeat" />.</para></listitem>
+  </varlistentry>
+
   <varlistentry xml:id="conf-extra-sandbox-paths">
     <term><literal>extra-sandbox-paths</literal></term>
 
@@ -595,9 +662,9 @@ password <replaceable>my-password</replaceable>
     they are deterministic. The default value is 0. If the value is
     non-zero, every build is repeated the specified number of
     times. If the contents of any of the runs differs from the
-    previous ones, the build is rejected and the resulting store paths
-    are not registered as “valid” in Nix’s database.</para></listitem>
-
+    previous ones and <xref linkend="conf-enforce-determinism" /> is
+    true, the build is rejected and the resulting store paths are not
+    registered as “valid” in Nix’s database.</para></listitem>
   </varlistentry>
 
   <varlistentry xml:id="conf-require-sigs"><term><literal>require-sigs</literal></term>
@@ -628,6 +695,19 @@ password <replaceable>my-password</replaceable>
 
   </varlistentry>
 
+  <varlistentry xml:id="conf-run-diff-hook"><term><literal>run-diff-hook</literal></term>
+  <listitem>
+    <para>
+      If true, enable the execution of <xref linkend="conf-diff-hook" />.
+    </para>
+
+    <para>
+      When using the Nix daemon, <literal>run-diff-hook</literal> must
+      be set in the <filename>nix.conf</filename> configuration file,
+      and cannot be passed at the command line.
+    </para>
+  </listitem>
+  </varlistentry>
 
   <varlistentry xml:id="conf-sandbox"><term><literal>sandbox</literal></term>
 
diff --git a/src/libstore/build.cc b/src/libstore/build.cc
index 91eb97dfb873..b07461013cc2 100644
--- a/src/libstore/build.cc
+++ b/src/libstore/build.cc
@@ -461,6 +461,28 @@ static void commonChildInit(Pipe & logPipe)
     close(fdDevNull);
 }
 
+void handleDiffHook(uid_t uid, uid_t gid, Path tryA, Path tryB, Path drvPath, Path tmpDir)
+{
+    auto diffHook = settings.diffHook;
+    if (diffHook != "" && settings.runDiffHook) {
+        try {
+            RunOptions diffHookOptions(diffHook,{tryA, tryB, drvPath, tmpDir});
+            diffHookOptions.searchPath = true;
+            diffHookOptions.uid = uid;
+            diffHookOptions.gid = gid;
+            diffHookOptions.chdir = "/";
+
+            auto diffRes = runProgram(diffHookOptions);
+            if (!statusOk(diffRes.first))
+                throw ExecError(diffRes.first, fmt("diff-hook program '%1%' %2%", diffHook, statusToString(diffRes.first)));
+
+            if (diffRes.second != "")
+                printError(chomp(diffRes.second));
+        } catch (Error & error) {
+            printError("diff hook execution failed: %s", error.what());
+        }
+    }
+}
 
 //////////////////////////////////////////////////////////////////////
 
@@ -3039,8 +3061,7 @@ void DerivationGoal::registerOutputs()
     InodesSeen inodesSeen;
 
     Path checkSuffix = ".check";
-    bool runDiffHook = settings.runDiffHook;
-    bool keepPreviousRound = settings.keepFailed || runDiffHook;
+    bool keepPreviousRound = settings.keepFailed || settings.runDiffHook;
 
     std::exception_ptr delayedException;
 
@@ -3185,11 +3206,17 @@ void DerivationGoal::registerOutputs()
             if (!worker.store.isValidPath(path)) continue;
             auto info = *worker.store.queryPathInfo(path);
             if (hash.first != info.narHash) {
-                if (settings.keepFailed) {
+                if (settings.runDiffHook || settings.keepFailed) {
                     Path dst = worker.store.toRealPath(path + checkSuffix);
                     deletePath(dst);
                     if (rename(actualPath.c_str(), dst.c_str()))
                         throw SysError(format("renaming '%1%' to '%2%'") % actualPath % dst);
+
+                    handleDiffHook(
+                        buildUser ? buildUser->getUID() : getuid(),
+                        buildUser ? buildUser->getGID() : getgid(),
+                        path, dst, drvPath, tmpDir);
+
                     throw Error(format("derivation '%1%' may not be deterministic: output '%2%' differs from '%3%'")
                         % drvPath % path % dst);
                 } else
@@ -3254,16 +3281,10 @@ void DerivationGoal::registerOutputs()
                     ? fmt("output '%1%' of '%2%' differs from '%3%' from previous round", i->second.path, drvPath, prev)
                     : fmt("output '%1%' of '%2%' differs from previous round", i->second.path, drvPath);
 
-                auto diffHook = settings.diffHook;
-                if (prevExists && diffHook != "" && runDiffHook) {
-                    try {
-                        auto diff = runProgram(diffHook, true, {prev, i->second.path});
-                        if (diff != "")
-                            printError(chomp(diff));
-                    } catch (Error & error) {
-                        printError("diff hook execution failed: %s", error.what());
-                    }
-                }
+                handleDiffHook(
+                    buildUser ? buildUser->getUID() : getuid(),
+                    buildUser ? buildUser->getGID() : getgid(),
+                    prev, i->second.path, drvPath, tmpDir);
 
                 if (settings.enforceDeterminism)
                     throw NotDeterministic(msg);
diff --git a/src/libutil/util.cc b/src/libutil/util.cc
index a7170566533e..17aee2d5c3d0 100644
--- a/src/libutil/util.cc
+++ b/src/libutil/util.cc
@@ -16,6 +16,7 @@
 #include <future>
 
 #include <fcntl.h>
+#include <grp.h>
 #include <limits.h>
 #include <pwd.h>
 #include <sys/ioctl.h>
@@ -1025,6 +1026,16 @@ void runProgram2(const RunOptions & options)
         if (source && dup2(in.readSide.get(), STDIN_FILENO) == -1)
             throw SysError("dupping stdin");
 
+        if (options.chdir && chdir((*options.chdir).c_str()) == -1)
+            throw SysError("chdir failed");
+        if (options.gid && setgid(*options.gid) == -1)
+            throw SysError("setgid failed");
+        /* Drop all other groups if we're setgid. */
+        if (options.gid && setgroups(0, 0) == -1)
+            throw SysError("setgroups failed");
+        if (options.uid && setuid(*options.uid) == -1)
+            throw SysError("setuid failed");
+
         Strings args_(options.args);
         args_.push_front(options.program);
 
diff --git a/src/libutil/util.hh b/src/libutil/util.hh
index 54936a5cb10b..7c57d0afad98 100644
--- a/src/libutil/util.hh
+++ b/src/libutil/util.hh
@@ -267,6 +267,9 @@ string runProgram(Path program, bool searchPath = false,
 
 struct RunOptions
 {
+    std::optional<uid_t> uid;
+    std::optional<uid_t> gid;
+    std::optional<Path> chdir;
     Path program;
     bool searchPath = true;
     Strings args;