libcmd,daemon: Reap children via a self-pipe trick

Somewhat better solution for https://github.com/NixOS/nix/issues/15394. Instead
of reaping immediately in the signal handler we instead punt that at the poll loop.
This way Pid destructor has a chance to properly run instead of being rudely interrupted
by the signal handler.

See: https://man7.org/tlpi/code/online/dist/altio/self_pipe.c.html

src/libcmd/include/nix/cmd/unix-socket-server.hh

@@ -53,6 +53,19 @@ struct ServeUnixSocketOptions
      * Mode for the created socket file.
      */
     mode_t socketMode = 0666;
+
+#ifndef _WIN32
+    /**
+     * Additional file descriptor to poll. Useful for doing a self-pipe trick
+     * https://cr.yp.to/docs/selfpipe.html.
+     */
+    Descriptor auxiliaryFd = INVALID_DESCRIPTOR;
+
+    /**
+     * Optional callback invoked on POLLIN event for auxiliaryFd.
+     */
+    std::function<void()> onAuxiliaryFdPollin = nullptr;
+#endif
 };
 
 /**

src/libcmd/unix/unix-socket-server.cc

@@ -9,6 +9,8 @@
 #include "nix/util/unix-domain-socket.hh"
 #include "nix/util/util.hh"
 
+#include <ranges>
+
 #include <sys/socket.h>
 #include <sys/un.h>
 #include <poll.h>
@@ -83,6 +85,9 @@ PeerInfo getPeerInfo(Descriptor remote)
     for (auto & i : listeningSockets)
         fds.push_back({.fd = i.get(), .events = POLLIN});
 
+    if (options.auxiliaryFd != INVALID_DESCRIPTOR)
+        fds.push_back({.fd = options.auxiliaryFd, .events = POLLIN});
+
     //  Loop accepting connections.
     while (1) {
         try {
@@ -95,7 +100,11 @@ PeerInfo getPeerInfo(Descriptor remote)
                 throw SysError("polling for incoming connections");
             }
 
-            for (auto & fd : fds) {
+            if (options.auxiliaryFd != INVALID_DESCRIPTOR && options.onAuxiliaryFdPollin && fds.back().revents & POLLIN)
+                /* Useful for reaping children. */
+                options.onAuxiliaryFdPollin();
+
+            for (auto & fd : std::views::take(fds, listeningSockets.size())) {
                 if (!fd.revents)
                     continue;
 

src/nix/unix/daemon.cc

@@ -120,13 +120,20 @@ static ssize_t splice(int fd_in, void * off_in, int fd_out, void * off_out, size
 }
 #endif
 
+static Pipe sigChldPipe;
+
 static void sigChldHandler(int sigNo)
 {
     // Ensure we don't modify errno of whatever we've interrupted
     auto saved_errno = errno;
-    //  Reap all dead children.
-    while (waitpid(-1, 0, WNOHANG) > 0)
-        ;
+    /* Write to the self-pipe that gets polled in the accept loop. Pipe
+       is non-blocking. https://man7.org/tlpi/code/online/dist/altio/self_pipe.c.html */
+    auto res = ::write(sigChldPipe.writeSide.get(), "x", 1);
+    if (res == -1 && errno != EAGAIN) {
+        /* Something is deeply wrong. Can't call std::terminate here because our terminate
+           handler is not safe for that. */
+        abort();
+    }
     errno = saved_errno;
 }
 
@@ -243,7 +250,13 @@ static void daemonLoop(ref<const StoreConfig> storeConfig, std::optional<Trusted
     if (chdir("/") == -1)
         throw SysError("cannot change current directory");
 
-    //  Get rid of children automatically; don't let them become zombies.
+    sigChldPipe.create();
+
+    for (auto fd : {sigChldPipe.readSide.get(), sigChldPipe.writeSide.get()})
+        if (::fcntl(fd, F_SETFL, O_NONBLOCK) == -1)
+            throw SysError("making self-pipe non-blocking");
+
+    // Get rid of children automatically; don't let them become zombies.
     setSigChldAction(true);
 
 #ifdef __linux__
@@ -272,6 +285,27 @@ static void daemonLoop(ref<const StoreConfig> storeConfig, std::optional<Trusted
             {
                 .socketPath = settings.nixDaemonSocketFile,
                 .socketMode = 0666,
+                .auxiliaryFd = sigChldPipe.readSide.get(),
+                .onAuxiliaryFdPollin =
+                    []() {
+                        std::array<char, 64> buf;
+
+                        /* Drain the self-pipe. */
+                        while (true) {
+                            if (::read(sigChldPipe.readSide.get(), buf.data(), buf.size()) == -1) {
+                                if (errno == EAGAIN)
+                                    break;
+                                else
+                                    throw SysError("reading from self-pipe for SIGCHLD");
+                            }
+                        }
+
+                        /* Reap all dead children. */
+                        pid_t pid = -1;
+                        int status;
+                        while (pid = ::waitpid(/*pid (any child process)=*/-1, &status, WNOHANG), pid > 0)
+                            printInfo("reaped child process %1%, status = %2%", pid, statusToString(status));
+                    },
             },
             [&](AutoCloseFD remote, std::function<void()> closeListeners) {
                 unix::closeOnExec(remote.get());