compel: plugins -- Add fds plugin

This is fd passing via unix sockets (scm creds) suitable for use
by parasite code.
Signed-off-by: Dmitry Safonov <dsafonov@virtuozzo.com>
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>
Signed-off-by: Andrei Vagin <avagin@virtuozzo.com>

compel/include/compiler.h

+../../criu/include/compiler.h
\ No newline at end of file

compel/plugins/Makefile

@@ -23,6 +23,11 @@ asflags-y		+= -iquote $(SRC_DIR)/$(ARCH_DIR)
 asflags-y		+= -Wstrict-prototypes -Wa,--noexecstack
 asflags-y		+= -D__ASSEMBLY__ -nostdlib -fomit-frame-pointer
 
+#
+# Fds plugin
+target			+= fds
+fds-obj-y		+= fds/fds.o
+
 #
 # Shmem plugin
 target			+= shmem

compel/plugins/fds/fds.c

+#include "uapi/plugins.h"
+
+#include "uapi/std/syscall.h"
+#include "uapi/std/string.h"
+#include "uapi/plugin-fds.h"
+
+#include "std-priv.h"
+
+#include "compiler.h"
+
+#define __sys(foo)	sys_##foo
+#define __std(foo)	std_##foo
+
+#include "../../src/shared/fds.c"
+
+int fds_send(int *fds, int nr_fds)
+{
+	return fds_send_via(std_ctl_sock(), fds, nr_fds);
+}
+
+int fds_recv(int *fds, int nr_fds)
+{
+	return fds_recv_via(std_ctl_sock(), fds, nr_fds);
+}
+
+PLUGIN_REGISTER_DUMMY(fds)

compel/plugins/include/uapi/plugin-fds.h

+/*
+ * plugin-fds.h -- API for fds compel plugin
+ */
+
+#ifndef __COMPEL_PLUGIN_FDS_H__
+#define __COMPEL_PLUGIN_FDS_H__
+
+extern int fds_send(int *fds, int nr_fds);
+extern int fds_recv(int *fds, int nr_fds);
+
+static inline int fds_send_one(int fd)
+{
+	return fds_send(&fd, 1);
+}
+
+static inline int fds_recv_one(void)
+{
+	int fd, ret;
+
+	ret = fds_recv(&fd, 1);
+	if (ret)
+		fd = -1;
+
+	return fd;
+}
+
+#endif /* __COMPEL_PLUGIN_FDS_H__ */

compel/src/shared/fds.c

+#include <sys/socket.h>
+#include <sys/un.h>
+
+/*
+ * Because of kernel doing kmalloc for user data passed
+ * in SCM messages, and there is kernel's SCM_MAX_FD as a limit
+ * for descriptors passed at once we're trying to reduce
+ * the pressue on kernel memory manager and use predefined
+ * known to work well size of the message buffer.
+ */
+#define CR_SCM_MSG_SIZE		(1024)
+#define CR_SCM_MAX_FD		(252)
+
+struct scm_fdset {
+	struct msghdr	hdr;
+	struct iovec	iov;
+	char		msg_buf[CR_SCM_MSG_SIZE];
+	char		f;
+};
+
+static void scm_fdset_init_chunk(struct scm_fdset *fdset, int nr_fds)
+{
+	struct cmsghdr *cmsg;
+
+	fdset->hdr.msg_controllen = CMSG_LEN(sizeof(int) * nr_fds);
+
+	cmsg		= CMSG_FIRSTHDR(&fdset->hdr);
+	cmsg->cmsg_len	= fdset->hdr.msg_controllen;
+}
+
+static int *scm_fdset_init(struct scm_fdset *fdset)
+{
+	struct cmsghdr *cmsg;
+
+	BUILD_BUG_ON(sizeof(fdset->msg_buf) < (CMSG_SPACE(sizeof(int) * CR_SCM_MAX_FD)));
+
+	fdset->iov.iov_base		= &fdset->f;
+	fdset->iov.iov_len		= 1;
+
+	fdset->hdr.msg_iov		= &fdset->iov;
+	fdset->hdr.msg_iovlen		= 1;
+	fdset->hdr.msg_name		= NULL;
+	fdset->hdr.msg_namelen		= 0;
+
+	fdset->hdr.msg_control		= &fdset->msg_buf;
+	fdset->hdr.msg_controllen	= CMSG_LEN(sizeof(int) * CR_SCM_MAX_FD);
+
+	cmsg				= CMSG_FIRSTHDR(&fdset->hdr);
+	cmsg->cmsg_len			= fdset->hdr.msg_controllen;
+	cmsg->cmsg_level		= SOL_SOCKET;
+	cmsg->cmsg_type			= SCM_RIGHTS;
+
+	return (int *)CMSG_DATA(cmsg);
+}
+
+int fds_send_via(int sock, int *fds, int nr_fds)
+{
+	struct scm_fdset fdset;
+	int i, min_fd, ret;
+	int *cmsg_data;
+
+	cmsg_data = scm_fdset_init(&fdset);
+
+	for (i = 0; i < nr_fds; i += min_fd) {
+		min_fd = min(CR_SCM_MAX_FD, nr_fds - i);
+		scm_fdset_init_chunk(&fdset, min_fd);
+		__std(memcpy(cmsg_data, &fds[i], sizeof(int) * min_fd));
+		ret = __sys(sendmsg(sock, &fdset.hdr, 0));
+		if (ret <= 0)
+			return ret ? : -1;
+	}
+
+	return 0;
+}
+
+int fds_recv_via(int sock, int *fds, int nr_fds)
+{
+	struct scm_fdset fdset;
+	struct cmsghdr *cmsg;
+	int *cmsg_data;
+	int ret;
+	int i, min_fd;
+
+	cmsg_data = scm_fdset_init(&fdset);
+	for (i = 0; i < nr_fds; i += min_fd) {
+		min_fd = min(CR_SCM_MAX_FD, nr_fds - i);
+		scm_fdset_init_chunk(&fdset, min_fd);
+
+		ret = __sys(recvmsg(sock, &fdset.hdr, 0));
+		if (ret <= 0)
+			return ret ? : -1;
+
+		cmsg = CMSG_FIRSTHDR(&fdset.hdr);
+		if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS)
+			return -1;
+		if (fdset.hdr.msg_flags & MSG_CTRUNC)
+			return -2;
+
+		min_fd = (cmsg->cmsg_len - sizeof(struct cmsghdr)) / sizeof(int);
+		/*
+		 * In case if kernel screwed the recepient, most probably
+		 * the caller stack frame will be overwriten, just scream
+		 * and exit.
+		 */
+		if (unlikely(min_fd > CR_SCM_MAX_FD))
+			*(volatile unsigned long *)NULL = 0xdead0000 + __LINE__;
+		if (unlikely(min_fd <= 0))
+			return -1;
+		__std(memcpy(&fds[i], cmsg_data, sizeof(int) * min_fd));
+	}
+
+	return 0;
+}