sfd: Move service fds to separate file

It has nothing to do with utils but
rather a separate service engine.
Signed-off-by: Cyrill Gorcunov <gorcunov@gmail.com>
Signed-off-by: Andrei Vagin <avagin@gmail.com>

criu/Makefile.crtools

@@ -84,6 +84,7 @@ obj-y			+= autofs.o
 obj-y			+= fdstore.o
 obj-y			+= uffd.o
 obj-y			+= config.o
+obj-y			+= servicefd.o
 
 ifeq ($(VDSO),y)
 obj-y			+= pie-util-vdso.o

criu/include/servicefd.h

@@ -8,17 +8,17 @@ enum sfd_type {
 
 	LOG_FD_OFF,
 	IMG_FD_OFF,
-	PROC_FD_OFF,	/* fd with /proc for all proc_ calls */
+	PROC_FD_OFF,		/* fd with /proc for all proc_ calls */
 	PROC_PID_FD_OFF,
-	CR_PROC_FD_OFF, /* some other's proc fd.
-			 *  For dump -- target ns' proc
-			 *  For restore -- CRIU ns' proc
-			 */
-	ROOT_FD_OFF,	/* Root of the namespace we dump/restore */
+	CR_PROC_FD_OFF,		/* some other's proc fd:
+				 *  - For dump -- target ns' proc
+				 *  - For restore -- CRIU ns' proc
+				 */
+	ROOT_FD_OFF,		/* Root of the namespace we dump/restore */
 	CGROUP_YARD,
-	USERNSD_SK,	/* Socket for usernsd */
-	NS_FD_OFF,	/* Node's net namespace fd */
-	TRANSPORT_FD_OFF, /* to transfer file descriptors */
+	USERNSD_SK,		/* Socket for usernsd */
+	NS_FD_OFF,		/* Node's net namespace fd */
+	TRANSPORT_FD_OFF,	/* to transfer file descriptors */
 	RPC_SK_OFF,
 	FDSTORE_SK_OFF,
 
@@ -28,14 +28,13 @@ enum sfd_type {
 struct pstree_item;
 extern bool sfds_protected;
 
-extern void set_proc_self_fd(int fd);
-extern int clone_service_fd(struct pstree_item *me);
 extern int init_service_fd(void);
 extern int get_service_fd(enum sfd_type type);
+extern bool is_any_service_fd(int fd);
+extern bool is_service_fd(int fd, enum sfd_type type);
+extern int service_fd_min_fd(struct pstree_item *item);
 extern int install_service_fd(enum sfd_type type, int fd);
 extern int close_service_fd(enum sfd_type type);
-extern bool is_service_fd(int fd, enum sfd_type type);
-extern bool is_any_service_fd(int fd);
-extern int service_fd_min_fd(struct pstree_item *);
+extern int clone_service_fd(struct pstree_item *me);
 
 #endif /* __CR_SERVICE_FD_H__ */

criu/include/util.h

@@ -180,6 +180,8 @@ extern int cr_daemon(int nochdir, int noclose, int *keep_fd, int close_fd);
 extern int close_status_fd(void);
 extern int is_root_user(void);
 
+extern void set_proc_self_fd(int fd);
+
 static inline bool dir_dots(const struct dirent *de)
 {
 	return !strcmp(de->d_name, ".") || !strcmp(de->d_name, "..");

criu/plugin.c

@@ -10,6 +10,7 @@
 #include "common/compiler.h"
 #include "xmalloc.h"
 #include "plugin.h"
+#include "servicefd.h"
 #include "common/list.h"
 #include "log.h"
 
@@ -90,6 +91,11 @@ static int verify_plugin(cr_plugin_desc_t *d)
 	return 0;
 }
 
+int criu_get_image_dir(void)
+{
+	return get_service_fd(IMG_FD_OFF);
+}
+
 static int cr_lib_load(int stage, char *path)
 {
 	cr_plugin_desc_t *d;

criu/servicefd.c

+#include <stdio.h>
+#include <errno.h>
+#include <unistd.h>
+#include <sched.h>
+
+#include <sys/time.h>
+#include <sys/syscall.h>
+#include <sys/resource.h>
+
+#include "common/compiler.h"
+#include "common/list.h"
+
+#include "criu-log.h"
+
+#include "util.h"
+#include "bitops.h"
+#include "pstree.h"
+#include "files.h"
+#include "rst_info.h"
+#include "servicefd.h"
+
+#undef	LOG_PREFIX
+#define LOG_PREFIX "sfd: "
+
+/* Max potentially possible fd to be open by criu process */
+int service_fd_rlim_cur;
+
+/* Base of current process service fds set */
+static int service_fd_base;
+
+/* Id of current process in shared fdt */
+static int service_fd_id = 0;
+
+static DECLARE_BITMAP(sfd_map, SERVICE_FD_MAX);
+static int sfd_arr[SERVICE_FD_MAX];
+/*
+ * Variable for marking areas of code, where service fds modifications
+ * are prohibited. It's used to safe them from reusing their numbers
+ * by ordinary files. See install_service_fd() and close_service_fd().
+ */
+bool sfds_protected = false;
+
+int init_service_fd(void)
+{
+	struct rlimit64 rlimit;
+
+	/*
+	 * Service FDs are those that most likely won't
+	 * conflict with any 'real-life' ones
+	 */
+
+	if (syscall(__NR_prlimit64, getpid(), RLIMIT_NOFILE, NULL, &rlimit)) {
+		pr_perror("Can't get rlimit");
+		return -1;
+	}
+
+	service_fd_rlim_cur = (int)rlimit.rlim_cur;
+	return 0;
+}
+
+static int __get_service_fd(enum sfd_type type, int service_fd_id)
+{
+	return service_fd_base - type - SERVICE_FD_MAX * service_fd_id;
+}
+
+int get_service_fd(enum sfd_type type)
+{
+	BUG_ON((int)type <= SERVICE_FD_MIN || (int)type >= SERVICE_FD_MAX);
+
+	if (!test_bit(type, sfd_map))
+		return -1;
+
+	if (service_fd_base == 0)
+		return sfd_arr[type];
+
+	return __get_service_fd(type, service_fd_id);
+}
+
+bool is_any_service_fd(int fd)
+{
+	int sfd_min_fd = __get_service_fd(SERVICE_FD_MAX, service_fd_id);
+	int sfd_max_fd = __get_service_fd(SERVICE_FD_MIN, service_fd_id);
+
+	if (fd > sfd_min_fd && fd < sfd_max_fd) {
+		int type = SERVICE_FD_MAX - (fd - sfd_min_fd);
+		if (type > SERVICE_FD_MIN && type < SERVICE_FD_MAX)
+			return !!test_bit(type, sfd_map);
+	}
+
+	return false;
+}
+
+bool is_service_fd(int fd, enum sfd_type type)
+{
+	return fd == get_service_fd(type);
+}
+
+int service_fd_min_fd(struct pstree_item *item)
+{
+	struct fdt *fdt = rsti(item)->fdt;
+	int id = 0;
+
+	if (fdt)
+		id = fdt->nr - 1;
+	return service_fd_rlim_cur - (SERVICE_FD_MAX - 1) - SERVICE_FD_MAX * id;
+}
+
+static void sfds_protection_bug(enum sfd_type type)
+{
+	pr_err("Service fd %u is being modified in protected context\n", type);
+	print_stack_trace(current ? vpid(current) : 0);
+	BUG();
+}
+
+int install_service_fd(enum sfd_type type, int fd)
+{
+	int sfd = __get_service_fd(type, service_fd_id);
+
+	BUG_ON((int)type <= SERVICE_FD_MIN || (int)type >= SERVICE_FD_MAX);
+	if (sfds_protected && !test_bit(type, sfd_map))
+		sfds_protection_bug(type);
+
+	if (service_fd_base == 0) {
+		if (test_bit(type, sfd_map))
+			close(sfd_arr[type]);
+		sfd_arr[type] = fd;
+		set_bit(type, sfd_map);
+		return fd;
+	}
+
+	if (dup3(fd, sfd, O_CLOEXEC) != sfd) {
+		pr_perror("Dup %d -> %d failed", fd, sfd);
+		close(fd);
+		return -1;
+	}
+
+	set_bit(type, sfd_map);
+	close(fd);
+	return sfd;
+}
+
+int close_service_fd(enum sfd_type type)
+{
+	int fd;
+
+	if (sfds_protected)
+		sfds_protection_bug(type);
+
+	fd = get_service_fd(type);
+	if (fd < 0)
+		return 0;
+
+	if (close_safe(&fd))
+		return -1;
+
+	clear_bit(type, sfd_map);
+	return 0;
+}
+
+static void move_service_fd(struct pstree_item *me, int type, int new_id, int new_base)
+{
+	int old = get_service_fd(type);
+	int new = new_base - type - SERVICE_FD_MAX * new_id;
+	int ret;
+
+	if (old < 0)
+		return;
+
+	ret = dup2(old, new);
+	if (ret == -1) {
+		if (errno != EBADF)
+			pr_perror("Unable to clone %d->%d", old, new);
+	} else if (!(rsti(me)->clone_flags & CLONE_FILES))
+		close(old);
+}
+
+static int choose_service_fd_base(struct pstree_item *me)
+{
+	int nr, real_nr, fdt_nr = 1, id = rsti(me)->service_fd_id;
+
+	if (rsti(me)->fdt) {
+		/* The base is set by owner of fdt (id 0) */
+		if (id != 0)
+			return service_fd_base;
+		fdt_nr = rsti(me)->fdt->nr;
+	}
+	/* Now find process's max used fd number */
+	if (!list_empty(&rsti(me)->fds))
+		nr = list_entry(rsti(me)->fds.prev,
+				struct fdinfo_list_entry, ps_list)->fe->fd;
+	else
+		nr = -1;
+
+	nr = max(nr, inh_fd_max);
+	/*
+	 * Service fds go after max fd near right border of alignment:
+	 *
+	 * ...|max_fd|max_fd+1|...|sfd first|...|sfd last (aligned)|
+	 *
+	 * So, they take maximum numbers of area allocated by kernel.
+	 * See linux alloc_fdtable() for details.
+	 */
+	nr += (SERVICE_FD_MAX - SERVICE_FD_MIN) * fdt_nr;
+	nr += 16; /* Safety pad */
+	real_nr = nr;
+
+	nr /= (1024 / sizeof(void *));
+	if (nr)
+		nr = 1 << (32 - __builtin_clz(nr));
+	else
+		nr = 1;
+	nr *= (1024 / sizeof(void *));
+
+	if (nr > service_fd_rlim_cur) {
+		/* Right border is bigger, than rlim. OK, then just aligned value is enough */
+		nr = round_down(service_fd_rlim_cur, (1024 / sizeof(void *)));
+		if (nr < real_nr) {
+			pr_err("Can't chose service_fd_base: %d %d\n", nr, real_nr);
+			return -1;
+		}
+	}
+
+	return nr;
+}
+
+int clone_service_fd(struct pstree_item *me)
+{
+	int id, new_base, i, ret = -1;
+
+	new_base = choose_service_fd_base(me);
+	id = rsti(me)->service_fd_id;
+
+	if (new_base == -1)
+		return -1;
+	if (service_fd_base == new_base && service_fd_id == id)
+		return 0;
+
+	/* Dup sfds in memmove() style: they may overlap */
+	if (get_service_fd(LOG_FD_OFF) < new_base - LOG_FD_OFF - SERVICE_FD_MAX * id)
+		for (i = SERVICE_FD_MIN + 1; i < SERVICE_FD_MAX; i++)
+			move_service_fd(me, i, id, new_base);
+	else
+		for (i = SERVICE_FD_MAX - 1; i > SERVICE_FD_MIN; i--)
+			move_service_fd(me, i, id, new_base);
+
+	service_fd_base = new_base;
+	service_fd_id = id;
+	ret = 0;
+
+	return ret;
+}

criu/util.c

@@ -5,7 +5,6 @@
 #include <string.h>
 #include <errno.h>
 #include <stdbool.h>
-#include <limits.h>
 #include <signal.h>
 #include <unistd.h>
 #include <dirent.h>
@@ -17,8 +16,6 @@
 #include <sys/param.h>
 #include <sys/types.h>
 #include <sys/ptrace.h>
-#include <sys/time.h>
-#include <sys/resource.h>
 #include <sys/stat.h>
 #include <sys/mman.h>
 #include <sys/wait.h>
@@ -29,12 +26,8 @@
 #include <sched.h>
 #include <ctype.h>
 
-#include "bitops.h"
 #include "page.h"
-#include "common/compiler.h"
-#include "common/list.h"
 #include "util.h"
-#include "rst-malloc.h"
 #include "image.h"
 #include "vma.h"
 #include "mem.h"
@@ -43,7 +36,6 @@
 
 #include "clone-noasan.h"
 #include "cr_options.h"
-#include "servicefd.h"
 #include "cr-service.h"
 #include "files.h"
 #include "pstree.h"
@@ -422,237 +414,6 @@ int do_open_proc(pid_t pid, int flags, const char *fmt, ...)
 	return openat(dirfd, path, flags);
 }
 
-/* Max potentially possible fd to be open by criu process */
-int service_fd_rlim_cur;
-/* Base of current process service fds set */
-static int service_fd_base;
-/* Id of current process in shared fdt */
-static int service_fd_id = 0;
-
-int init_service_fd(void)
-{
-	struct rlimit64 rlimit;
-
-	/*
-	 * Service FDs are those that most likely won't
-	 * conflict with any 'real-life' ones
-	 */
-
-	if (syscall(__NR_prlimit64, getpid(), RLIMIT_NOFILE, NULL, &rlimit)) {
-		pr_perror("Can't get rlimit");
-		return -1;
-	}
-
-	service_fd_rlim_cur = (int)rlimit.rlim_cur;
-
-	return 0;
-}
-
-static int __get_service_fd(enum sfd_type type, int service_fd_id)
-{
-	return service_fd_base - type - SERVICE_FD_MAX * service_fd_id;
-}
-
-int service_fd_min_fd(struct pstree_item *item)
-{
-	struct fdt *fdt = rsti(item)->fdt;
-	int id = 0;
-
-	if (fdt)
-		id = fdt->nr - 1;
-	return service_fd_rlim_cur - (SERVICE_FD_MAX - 1) - SERVICE_FD_MAX * id;
-}
-
-static DECLARE_BITMAP(sfd_map, SERVICE_FD_MAX);
-static int sfd_arr[SERVICE_FD_MAX];
-/*
- * Variable for marking areas of code, where service fds modifications
- * are prohibited. It's used to safe them from reusing their numbers
- * by ordinary files. See install_service_fd() and close_service_fd().
- */
-bool sfds_protected = false;
-
-static void sfds_protection_bug(enum sfd_type type)
-{
-	pr_err("Service fd %u is being modified in protected context\n", type);
-	print_stack_trace(current ? vpid(current) : 0);
-	BUG();
-}
-
-int install_service_fd(enum sfd_type type, int fd)
-{
-	int sfd = __get_service_fd(type, service_fd_id);
-
-	BUG_ON((int)type <= SERVICE_FD_MIN || (int)type >= SERVICE_FD_MAX);
-	if (sfds_protected && !test_bit(type, sfd_map))
-		sfds_protection_bug(type);
-
-	if (service_fd_base == 0) {
-		if (test_bit(type, sfd_map))
-			close(sfd_arr[type]);
-		sfd_arr[type] = fd;
-		set_bit(type, sfd_map);
-		return fd;
-	}
-
-	if (dup3(fd, sfd, O_CLOEXEC) != sfd) {
-		pr_perror("Dup %d -> %d failed", fd, sfd);
-		close(fd);
-		return -1;
-	}
-
-	set_bit(type, sfd_map);
-	close(fd);
-	return sfd;
-}
-
-int get_service_fd(enum sfd_type type)
-{
-	BUG_ON((int)type <= SERVICE_FD_MIN || (int)type >= SERVICE_FD_MAX);
-
-	if (!test_bit(type, sfd_map))
-		return -1;
-
-	if (service_fd_base == 0)
-		return sfd_arr[type];
-
-	return __get_service_fd(type, service_fd_id);
-}
-
-int criu_get_image_dir(void)
-{
-	return get_service_fd(IMG_FD_OFF);
-}
-
-int close_service_fd(enum sfd_type type)
-{
-	int fd;
-
-	if (sfds_protected)
-		sfds_protection_bug(type);
-
-	fd = get_service_fd(type);
-	if (fd < 0)
-		return 0;
-
-	if (close_safe(&fd))
-		return -1;
-
-	clear_bit(type, sfd_map);
-	return 0;
-}
-
-static void move_service_fd(struct pstree_item *me, int type, int new_id, int new_base)
-{
-	int old = get_service_fd(type);
-	int new = new_base - type - SERVICE_FD_MAX * new_id;
-	int ret;
-
-	if (old < 0)
-		return;
-	ret = dup2(old, new);
-	if (ret == -1) {
-		if (errno != EBADF)
-			pr_perror("Unable to clone %d->%d", old, new);
-	} else if (!(rsti(me)->clone_flags & CLONE_FILES))
-		close(old);
-}
-
-static int choose_service_fd_base(struct pstree_item *me)
-{
-	int nr, real_nr, fdt_nr = 1, id = rsti(me)->service_fd_id;
-
-	if (rsti(me)->fdt) {
-		/* The base is set by owner of fdt (id 0) */
-		if (id != 0)
-			return service_fd_base;
-		fdt_nr = rsti(me)->fdt->nr;
-	}
-	/* Now find process's max used fd number */
-	if (!list_empty(&rsti(me)->fds))
-		nr = list_entry(rsti(me)->fds.prev,
-				struct fdinfo_list_entry, ps_list)->fe->fd;
-	else
-		nr = -1;
-
-	nr = max(nr, inh_fd_max);
-	/*
-	 * Service fds go after max fd near right border of alignment:
-	 *
-	 * ...|max_fd|max_fd+1|...|sfd first|...|sfd last (aligned)|
-	 *
-	 * So, they take maximum numbers of area allocated by kernel.
-	 * See linux alloc_fdtable() for details.
-	 */
-	nr += (SERVICE_FD_MAX - SERVICE_FD_MIN) * fdt_nr;
-	nr += 16; /* Safety pad */
-	real_nr = nr;
-
-	nr /= (1024 / sizeof(void *));
-	if (nr)
-		nr = 1 << (32 - __builtin_clz(nr));
-	else
-		nr = 1;
-	nr *= (1024 / sizeof(void *));
-
-	if (nr > service_fd_rlim_cur) {
-		/* Right border is bigger, than rlim. OK, then just aligned value is enough */
-		nr = round_down(service_fd_rlim_cur, (1024 / sizeof(void *)));
-		if (nr < real_nr) {
-			pr_err("Can't chose service_fd_base: %d %d\n", nr, real_nr);
-			return -1;
-		}
-	}
-
-	return nr;
-}
-
-int clone_service_fd(struct pstree_item *me)
-{
-	int id, new_base, i, ret = -1;
-
-	new_base = choose_service_fd_base(me);
-	id = rsti(me)->service_fd_id;
-
-	if (new_base == -1)
-		return -1;
-	if (service_fd_base == new_base && service_fd_id == id)
-		return 0;
-
-	/* Dup sfds in memmove() style: they may overlap */
-	if (get_service_fd(LOG_FD_OFF) < new_base - LOG_FD_OFF - SERVICE_FD_MAX * id)
-		for (i = SERVICE_FD_MIN + 1; i < SERVICE_FD_MAX; i++)
-			move_service_fd(me, i, id, new_base);
-	else
-		for (i = SERVICE_FD_MAX - 1; i > SERVICE_FD_MIN; i--)
-			move_service_fd(me, i, id, new_base);
-
-	service_fd_base = new_base;
-	service_fd_id = id;
-	ret = 0;
-
-	return ret;
-}
-
-bool is_any_service_fd(int fd)
-{
-	int sfd_min_fd = __get_service_fd(SERVICE_FD_MAX, service_fd_id);
-	int sfd_max_fd = __get_service_fd(SERVICE_FD_MIN, service_fd_id);
-
-	if (fd > sfd_min_fd && fd < sfd_max_fd) {
-		int type = SERVICE_FD_MAX - (fd - sfd_min_fd);
-		if (type > SERVICE_FD_MIN && type < SERVICE_FD_MAX)
-			return !!test_bit(type, sfd_map);
-	}
-
-	return false;
-}
-
-bool is_service_fd(int fd, enum sfd_type type)
-{
-	return fd == get_service_fd(type);
-}
-
 int copy_file(int fd_in, int fd_out, size_t bytes)
 {
 	ssize_t written = 0;