proc: use pidfd_open(..., PIDFD_THREAD) if available

With luck this approach might possibly perhaps go upstream!

We still fall back to waitfd if using PIDFD_THREAD doesn't work.
Complicated somewhat by having to use SYS_syscall() if a glibc too old
to support pidfd_open() isn't available (i.e. all released ones: it'll
be in 2.36), and by the likelihood that PIDFD_THREAD isn't in the
header even if the header is present, but otherwise fairly trivial.

Signed-off-by: Nick Alcock <nick.alcock@oracle.com>

Makeconfig

@@ -103,3 +103,4 @@ $(eval $(call check-symbol-rule,LIBFUSE3,fuse_session_receive_buf,fuse3))
 endif
 $(eval $(call check-header-rule,FUSE_NUMA,fuse_set_numa,fuse/fuse_lowlevel))
 $(eval $(call check-header-symbol-rule,CLOSE_RANGE,close_range(3,~0U,0),c,unistd))
+$(eval $(call check-header-rule,SYS_PIDFD_H,pidfd_open,sys/pidfd))

include/port.h

@@ -40,6 +40,8 @@ unsigned long linux_version_code(void);
 #define elf_getshdrnum elf_getshnum
 #endif
 
+int pid_fd(pid_t pid);
+
 #ifndef HAVE_WAITFD
 int waitfd(int which, pid_t upid, int options, int flags);
 #endif

libdtrace/dt_proc.c

@@ -72,7 +72,6 @@
  *    (proxy_call()->dt_proc_loop(), wwith a special case for cleanup).
  */
 
-#include <sys/wait.h>
 #include <sys/eventfd.h>
 #include <string.h>
 #include <signal.h>
@@ -937,11 +936,11 @@ dt_proc_control(void *arg)
 	Pset_ptrace_wrapper(dpr->dpr_proc, proxy_ptrace);
 
 	/*
-	 * Make a waitfd to this process, and set up polling structures
-	 * appropriately.  WEXITED | WSTOPPED is what Pwait() waits for.
+	 * Make a pidfd or (failing that) a pidfd to this process, and set up
+	 * polling structures appropriately.
 	 */
-	if ((dpr->dpr_fd = waitfd(P_PID, dpr->dpr_pid, WEXITED | WSTOPPED, 0)) < 0) {
-		dt_proc_error(dtp, dpr, "failed to get waitfd() for pid %li: %s\n",
+	if ((dpr->dpr_fd = pid_fd(dpr->dpr_pid)) < 0) {
+		dt_proc_error(dtp, dpr, "failed to get pidfd or waitfd for pid %li: %s\n",
 		    (long)dpr->dpr_pid, strerror(errno));
 		/*
 		 * Demote this to a mandatorily noninvasive grab: if we
@@ -950,7 +949,8 @@ dt_proc_control(void *arg)
 		 * we can do but hope.
 		 *
 		 * The dt_consume_proc_exits() function, called by
-		 * dtrace_consume(), checks for termination of such processes			 * (since nothing else will).
+		 * dtrace_consume(), checks for termination of such processes
+		 * (since nothing else will).
 		 */
 		Prelease(dpr->dpr_proc, PS_RELEASE_NORMAL);
 		if ((dpr->dpr_proc = Pgrab(dpr->dpr_pid, 2, 0,
@@ -1056,7 +1056,7 @@ dt_proc_loop(dt_proc_t *dpr, int awaiting_continue)
 
 	/*
 	 * We always want to listen on the proxy pipe; we only want to listen on
-	 * the process's waitfd pipe sometimes.
+	 * the process's pidfd pipe sometimes.
 	 */
 
 	pfd[0].events = POLLIN;
@@ -1244,7 +1244,7 @@ dt_proc_loop(dt_proc_t *dpr, int awaiting_continue)
 
 		/*
 		 * The process needs attention. Pwait() for it (which will make
-		 * the waitfd transition back to empty).
+		 * the pidfd transition back to empty).
 		 */
 		if (pfd[0].revents != 0) {
 			dt_dprintf("%d: Handling a process state change\n",

libport/Build

@@ -8,10 +8,11 @@ LIBS += libport
 
 libport_TARGET = libport
 libport_DIR := $(current-dir)
-ifdef HAVE_CLOSE_RANGE
-libport_SOURCES = gmatch.c linux_version_code.c strlcat.c strlcpy.c p_online.c time.c daemonize.c $(ARCHINC)/waitfd.c
-else
-libport_SOURCES = gmatch.c linux_version_code.c strlcat.c strlcpy.c p_online.c time.c daemonize.c close_range.c $(ARCHINC)/waitfd.c
+libport_SOURCES = gmatch.c linux_version_code.c strlcat.c strlcpy.c p_online.c time.c daemonize.c pid_fd.c $(ARCHINC)/waitfd.c
+
+ifndef HAVE_CLOSE_RANGE
+libport_SOURCES += close_range.c
 endif
+
 libport_LIBSOURCES := libport
 libport_CPPFLAGS := -Ilibdtrace

libport/pid_fd.c

@@ -0,0 +1,40 @@
+/*
+ * Licensed under the Universal Permissive License v 1.0 as shown at
+ * http://oss.oracle.com/licenses/upl.
+ */
+
+#include <sys/syscall.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <port.h>				/* for waitfd, etc */
+#ifdef HAVE_SYS_PIDFD_H
+#include <sys/pidfd.h>
+#endif
+
+#ifndef __NR_pidfd_open
+#define __NR_pidfd_open 434
+#endif
+
+#ifndef PIDFD_THREAD
+/* An arbitrary value unlikely ever to be useful for pidfds.  */
+#define PIDFD_THREAD O_NOCTTY
+#endif
+
+int
+pid_fd(pid_t pid)
+{
+	int fd;
+#ifdef HAVE_SYS_PIDFD_H
+	if ((fd = pidfd_open(pid, PIDFD_THREAD)) >= 0)
+		return fd;
+#else
+	if ((fd = syscall(__NR_pidfd_open, pid, PIDFD_THREAD)) >= 0)
+		return fd;
+#endif
+
+	/*
+	 * WEXITED | WSTOPPED is what Pwait() waits for.
+	 */
+        return waitfd(P_PID, pid, WEXITED | WSTOPPED, 0);
+}