Allow users to enable sub-reaping

This allows users that don't control PID 1 in their container to still
use Tini to reap zombies.

ci/run_build.sh

@@ -14,6 +14,11 @@ export DIST_DIR="$(readlink -f "${DIST_DIR}")"
 export BUILD_DIR="$(readlink -f "${BUILD_DIR}")"
 
 
+# Our build platform doesn't have those newer Linux flags, but we want Tini to have subreaper support
+# We also use those in our tests
+export CFLAGS="-DPR_SET_CHILD_SUBREAPER=36 -DPR_GET_CHILD_SUBREAPER=37"
+
+
 # Ensure Python output is not buffered (to make tests output clearer)
 export PYTHONUNBUFFERED=1
 
@@ -65,20 +70,7 @@ virtualenv "${VENV}"
 export PATH="${VENV}/bin:${PATH}"
 
 # Install test dependencies
-
-# We need a patched version because Travis only gives us Ubuntu Precise
-# (whose Linux headers don't include PR_SET_CHILD_SUBREAPER), but actually
-# runs a newer Linux Kernel (because we're actually in Docker) that has the
-# PR_SET_CHILD_SUBREAPER prctl call.
-pushd /tmp
-pip install python-prctl==1.6.1 --download="."
-tar -xvf /tmp/python-prctl-1.6.1.tar.gz
-cd python-prctl-1.6.1
-patch -p1 < "${SOURCE_DIR}/test/0001-Add-PR_SET_CHILD_SUBREAPER.patch"
-python setup.py install
-popd
-
-pip install psutil
+pip install psutil python-prctl
 
 # Run tests
 python "${SOURCE_DIR}/test/run_inner_tests.py"

src/tini.c

@@ -3,6 +3,7 @@
 
 #include <sys/types.h>
 #include <sys/wait.h>
+#include <sys/prctl.h>
 
 #include <errno.h>
 #include <signal.h>
@@ -24,6 +25,18 @@
 #define ARRAY_LEN(x)  (sizeof(x) / sizeof(x[0]))
 
 
+#ifdef PR_SET_CHILD_SUBREAPER
+#define HAS_SUBREAPER 1
+#define OPT_STRING "hsv"
+#else
+#define HAS_SUBREAPER 0
+#define OPT_STRING "hv"
+#endif
+
+
+#if HAS_SUBREAPER
+static int subreaper = 0;
+#endif
 static int verbosity = 0;
 static struct timespec ts = { .tv_sec = 1, .tv_nsec = 0 };
 
@@ -58,6 +71,9 @@ void print_usage(char* const name, FILE* const file) {
 	fprintf(file, "Usage: %s [OPTIONS] PROGRAM -- [ARGS]\n\n", basename(name));
 	fprintf(file, "Execute a program under the supervision of a valid init process (%s)\n\n", basename(name));
 	fprintf(file, "  -h: Show this help message and exit.\n");
+#if HAS_SUBREAPER
+	fprintf(file, "  -s: Register as a process subreaper (requires Linux >= 3.4).\n");
+#endif
 	fprintf(file, "  -v: Generate more verbose output. Repeat up to 4 times.\n");
 	fprintf(file, "\n");
 }
@@ -67,13 +83,18 @@ int parse_args(const int argc, char* const argv[], char* (**child_args_ptr_ptr)[
 	char* name = argv[0];
 
 	int c;
-	while ((c = getopt (argc, argv, "hv")) != -1) {
+	while ((c = getopt(argc, argv, OPT_STRING)) != -1) {
 		switch (c) {
 			case 'h':
 				/* TODO - Shouldn't cause exit with -1 ..*/
 				print_usage(name, stdout);
 				*parse_fail_exitcode_ptr = 0;
 				return 1;
+#if HAS_SUBREAPER
+			case 's':
+				subreaper++;
+				break;
+#endif
 			case 'v':
 				verbosity++;
 				break;
@@ -107,6 +128,50 @@ int parse_args(const int argc, char* const argv[], char* (**child_args_ptr_ptr)[
 	return 0;
 }
 
+
+#if HAS_SUBREAPER
+int register_subreaper () {
+	if (subreaper > 0) {
+		if (prctl(PR_SET_CHILD_SUBREAPER)) {
+			if (errno == EINVAL) {
+				PRINT_FATAL("PR_SET_CHILD_SUBREAPER is unavailable on this platform. Are you using Linux >= 3.4?")
+			} else {
+				PRINT_FATAL("Failed to register as child subreaper: %s", strerror(errno))
+			}
+			return 1;
+		} else {
+			PRINT_TRACE("Registered as child subreaper");
+		}
+	}
+	return 0;
+}
+#endif
+
+
+void reaper_check () {
+	/* Check that we can properly reap zombies */
+#if HAS_SUBREAPER
+	int bit = 0;
+#endif
+
+	if (getpid() == 1) {
+		return;
+	}
+
+#if HAS_SUBREAPER
+	if (prctl(PR_GET_CHILD_SUBREAPER, &bit)) {
+		PRINT_DEBUG("Failed to read child subreaper attribute: %s", strerror(errno));
+	} else if (bit == 1) {
+		return;
+	}
+#endif
+
+	PRINT_WARNING("Tini is not running as PID 1 and isn't registered as a child subreaper.\n\
+        Zombie processes will not be re-parented to Tini, so zombie reaping won't work.\n\
+        Use -s or run Tini as PID 1 to fix the problem.");
+}
+
+
 int prepare_sigmask(sigset_t* const parent_sigset_ptr, sigset_t* const child_sigset_ptr) {
 	/* Prepare signals to block; make sure we don't block program error signals. */
 	if (sigfillset(parent_sigset_ptr)) {
@@ -182,14 +247,14 @@ int reap_zombies(const pid_t child_pid, int* const child_exitcode_ptr) {
 
 			case -1:
 				if (errno == ECHILD) {
-					PRINT_TRACE("No child to wait.");
+					PRINT_TRACE("No child to wait");
 					break;
 				}
 				PRINT_FATAL("Error while waiting for pids: '%s'", strerror(errno));
 				return 1;
 
 			case 0:
-				PRINT_TRACE("No child to reap.");
+				PRINT_TRACE("No child to reap");
 				break;
 
 			default:
@@ -209,7 +274,7 @@ int reap_zombies(const pid_t child_pid, int* const child_exitcode_ptr) {
 						PRINT_INFO("Main child exited with signal (with signal '%s')", strsignal(WTERMSIG(current_status)));
 						*child_exitcode_ptr = 128 + WTERMSIG(current_status);
 					} else {
-						PRINT_FATAL("Main child exited for unknown reason!");
+						PRINT_FATAL("Main child exited for unknown reason");
 						return 1;
 					}
 				}
@@ -233,6 +298,13 @@ int main(int argc, char *argv[]) {
 	int child_exitcode = -1;  // This isn't a valid exitcode, and lets us tell whether the child has exited.
 	int parse_exitcode = 1;   // By default, we exit with 1 if parsing fails.
 
+	/* Parse command line arguments */
+	char* (*child_args_ptr)[];
+	int parse_args_ret = parse_args(argc, argv, &child_args_ptr, &parse_exitcode);
+	if (parse_args_ret) {
+		return parse_exitcode;
+	}
+
 	/* Prepare sigmask */
 	sigset_t parent_sigset;
 	sigset_t child_sigset;
@@ -240,13 +312,17 @@ int main(int argc, char *argv[]) {
 		return 1;
 	}
 
-	/* Parse command line arguments */
-	char* (*child_args_ptr)[];
-	int parse_args_ret = parse_args(argc, argv, &child_args_ptr, &parse_exitcode);
-	if (parse_args_ret) {
-		return parse_exitcode;
-	}
+#if HAS_SUBREAPER
+	/* If available and requested, register as a subreaper */
+	if (register_subreaper()) {
+		return 1;
+	};
+#endif
+
+	/* Are we going to reap zombies properly? If not, warn. */
+	reaper_check();
 
+	/* Go on */
 	if (spawn(&child_sigset, *child_args_ptr, &child_pid)) {
 		return 1;
 	}
@@ -264,7 +340,7 @@ int main(int argc, char *argv[]) {
 		}
 
 		if (child_exitcode != -1) {
-			PRINT_TRACE("Child has exited. Exiting");
+			PRINT_TRACE("Exiting: child has exited");
 			return child_exitcode;
 		}
 	}

test/0001-Add-PR_SET_CHILD_SUBREAPER.patch

@@ -1,35 +0,0 @@
-From b8c6ccd4575837e3901bbdee7b219ef951dc2065 Mon Sep 17 00:00:00 2001
-From: Thomas Orozco <thomas@orozco.fr>
-Date: Sun, 28 Jun 2015 15:25:37 +0200
-Subject: [PATCH] Add PR_SET_CHILD_SUBREAPER
-
----
- _prctlmodule.c | 12 ++++++++++++
- 1 file changed, 12 insertions(+)
-
-diff --git a/_prctlmodule.c b/_prctlmodule.c
-index 14121c3..19ad141 100644
---- a/_prctlmodule.c
-+++ b/_prctlmodule.c
-@@ -15,6 +15,18 @@
- #include <sys/prctl.h>
- #include <sys/signal.h>
- 
-+/* Our builds run in a Docker environment that has those, but they are
-+ * not in the kernel headers. Add them.
-+ */
-+
-+#ifndef PR_SET_CHILD_SUBREAPER
-+#define PR_SET_CHILD_SUBREAPER 36
-+#endif
-+
-+#ifndef PR_GET_CHILD_SUBREAPER
-+#define PR_GET_CHILD_SUBREAPER 37
-+#endif
-+
- /* New in 2.6.32, but named and implemented inconsistently. The linux
-  * implementation has two ways of setting the policy to the default, and thus
-  * needs an extra argument. We ignore the first argument and always call
--- 
-2.4.3
-