util: use close_range when it's supported

close_range is faster than reading /proc/self/fd and closing descriptors
one by one.

Signed-off-by: Andrei Vagin <avagin@gmail.com>

compel/arch/arm/plugins/std/syscalls/syscall.def

@@ -118,6 +118,7 @@ fsopen				430	430	(char *fsname, unsigned int flags)
 fsconfig			431	431	(int fd, unsigned int cmd, const char *key, const char *value, int aux)
 fsmount				432	432	(int fd, unsigned int flags, unsigned int attr_flags)
 clone3				435	435	(struct clone_args *uargs, size_t size)
+close_range			436	436	(unsigned int fd, unsigned int max_fd, unsigned int flags)
 pidfd_open			434	434	(pid_t pid, unsigned int flags)
 openat2				437	437	(int dirfd, char *pathname, struct open_how *how, size_t size)
 pidfd_getfd			438	438	(int pidfd, int targetfd, unsigned int flags)

compel/arch/mips/plugins/std/syscalls/syscall_64.tbl

@@ -115,6 +115,7 @@ __NR_fsopen			5430		sys_fsopen		(char *fsname, unsigned int flags)
 __NR_fsconfig			5431		sys_fsconfig		(int fd, unsigned int cmd, const char *key, const char *value, int aux)
 __NR_fsmount			5432		sys_fsmount		(int fd, unsigned int flags, unsigned int attr_flags)
 __NR_clone3			5435		sys_clone3		(struct clone_args *uargs, size_t size)
+__NR_close_range		5436		sys_close_range		(unsigned int fd, unsigned int max_fd, unsigned int flags)
 __NR_pidfd_open			5434		sys_pidfd_open		(pid_t pid, unsigned int flags)
 __NR_openat2			5437		sys_openat2		(int dirfd, char *pathname, struct open_how *how, size_t size)
 __NR_pidfd_getfd		5438		sys_pidfd_getfd		(int pidfd, int targetfd, unsigned int flags)

compel/arch/ppc64/plugins/std/syscalls/syscall-ppc64.tbl

@@ -114,6 +114,7 @@ __NR_fsopen		430		sys_fsopen		(char *fsname, unsigned int flags)
 __NR_fsconfig		431		sys_fsconfig		(int fd, unsigned int cmd, const char *key, const char *value, int aux)
 __NR_fsmount		432		sys_fsmount		(int fd, unsigned int flags, unsigned int attr_flags)
 __NR_clone3		435		sys_clone3		(struct clone_args *uargs, size_t size)
+__NR_close_range	436		sys_close_range		(unsigned int fd, unsigned int max_fd, unsigned int flags)
 __NR_pidfd_open		434		sys_pidfd_open		(pid_t pid, unsigned int flags)
 __NR_openat2		437		sys_openat2		(int dirfd, char *pathname, struct open_how *how, size_t size)
 __NR_pidfd_getfd	438		sys_pidfd_getfd		(int pidfd, int targetfd, unsigned int flags)

compel/arch/s390/plugins/std/syscalls/syscall-s390.tbl

@@ -114,6 +114,7 @@ __NR_fsopen		430		sys_fsopen		(char *fsname, unsigned int flags)
 __NR_fsconfig		431		sys_fsconfig		(int fd, unsigned int cmd, const char *key, const char *value, int aux)
 __NR_fsmount		432		sys_fsmount		(int fd, unsigned int flags, unsigned int attr_flags)
 __NR_clone3		435		sys_clone3		(struct clone_args *uargs, size_t size)
+__NR_close_range	436		sys_close_range		(unsigned int fd, unsigned int max_fd, unsigned int flags)
 __NR_pidfd_open		434		sys_pidfd_open		(pid_t pid, unsigned int flags)
 __NR_openat2		437		sys_openat2		(int dirfd, char *pathname, struct open_how *how, size_t size)
 __NR_pidfd_getfd	438		sys_pidfd_getfd		(int pidfd, int targetfd, unsigned int flags)

compel/arch/x86/plugins/std/syscalls/syscall_32.tbl

@@ -102,6 +102,7 @@ __NR_fsopen		430		sys_fsopen		(char *fsname, unsigned int flags)
 __NR_fsconfig		431		sys_fsconfig		(int fd, unsigned int cmd, const char *key, const char *value, int aux)
 __NR_fsmount		432		sys_fsmount		(int fd, unsigned int flags, unsigned int attr_flags)
 __NR_clone3		435		sys_clone3		(struct clone_args *uargs, size_t size)
+__NR_close_range	436		sys_close_range		(unsigned int fd, unsigned int max_fd, unsigned int flags)
 __NR_pidfd_open		434		sys_pidfd_open		(pid_t pid, unsigned int flags)
 __NR_openat2		437		sys_openat2		(int dirfd, char *pathname, struct open_how *how, size_t size)
 __NR_pidfd_getfd	438		sys_pidfd_getfd		(int pidfd, int targetfd, unsigned int flags)

compel/arch/x86/plugins/std/syscalls/syscall_64.tbl

@@ -113,6 +113,7 @@ __NR_fsopen			430		sys_fsopen		(char *fsname, unsigned int flags)
 __NR_fsconfig			431		sys_fsconfig		(int fd, unsigned int cmd, const char *key, const char *value, int aux)
 __NR_fsmount			432		sys_fsmount		(int fd, unsigned int flags, unsigned int attr_flags)
 __NR_clone3			435		sys_clone3		(struct clone_args *uargs, size_t size)
+__NR_close_range		436		sys_close_range		(unsigned int fd, unsigned int max_fd, unsigned int flags)
 __NR_pidfd_open			434		sys_pidfd_open		(pid_t pid, unsigned int flags)
 __NR_openat2		437		sys_openat2		(int dirfd, char *pathname, struct open_how *how, size_t size)
 __NR_pidfd_getfd		438		sys_pidfd_getfd		(int pidfd, int targetfd, unsigned int flags)

criu/include/kerndat.h

@@ -88,6 +88,7 @@ struct kerndat_s {
 	bool has_membarrier_get_registrations;
 	bool has_pagemap_scan;
 	bool has_shstk;
+	bool has_close_range;
 };
 
 extern struct kerndat_s kdat;

criu/include/util.h

@@ -411,4 +411,6 @@ extern void util_init(void);
 
 extern char *resolve_mountpoint(char *path);
 
+extern int cr_close_range(unsigned int fd, unsigned int max_fd, unsigned int flags);
+
 #endif /* __CR_UTIL_H__ */

criu/kerndat.c

@@ -1699,6 +1699,27 @@ static int kerndat_has_membarrier_get_registrations(void)
 	return 0;
 }
 
+static int kerndat_has_close_range(void)
+{
+	/* fd is greater than max_fd, so close_range should return EINVAL. */
+	if (cr_close_range(2, 1, 0) == 0) {
+		pr_err("close_range succeeded unexpectedly\n");
+		return -1;
+	}
+
+	if (errno == ENOSYS) {
+		pr_debug("close_range isn't supported\n");
+		return 0;
+	}
+	if (errno != EINVAL) {
+		pr_perror("close_range returned unexpected error code");
+		return -1;
+	}
+
+	kdat.has_close_range = true;
+	return 0;
+}
+
 /*
  * Some features depend on resource that can be dynamically changed
  * at the OS runtime. There are cases that we cannot determine the
@@ -1956,6 +1977,10 @@ int kerndat_init(void)
 		pr_err("kerndat_has_shstk failed when initializing kerndat.\n");
 		ret = -1;
 	}
+	if (!ret && kerndat_has_close_range()) {
+		pr_err("kerndat_has_close_range has failed when initializing kerndat.\n");
+		ret = -1;
+	}
 
 	kerndat_lsm();
 	kerndat_mmap_min_addr();

criu/util.c

@@ -54,6 +54,7 @@
 #include "action-scripts.h"
 
 #include "compel/infect-util.h"
+#include <compel/plugins/std/syscall-codes.h>
 
 #define VMA_OPT_LEN 128
 
@@ -518,12 +519,25 @@ int cr_system(int in, int out, int err, char *cmd, char *const argv[], unsigned
 	return cr_system_userns(in, out, err, cmd, argv, flags, -1);
 }
 
+int cr_close_range(unsigned int fd, unsigned int max_fd, unsigned int flags)
+{
+	return syscall(__NR_close_range, fd, max_fd, flags);
+}
+
 static int close_fds(int minfd)
 {
 	DIR *dir;
 	struct dirent *de;
 	int fd, ret, dfd;
 
+	if (kdat.has_close_range) {
+		if (cr_close_range(minfd, ~0, 0)) {
+			pr_perror("close_range failed");
+			return -1;
+		}
+		return 0;
+	}
+
 	dir = opendir("/proc/self/fd");
 	if (dir == NULL) {
 		pr_perror("Can't open /proc/self/fd");