propagator/init.c

/*
 * Guillaume Cottenceau (gc@mandrakesoft.com)
 *
 * Copyright 2000 MandrakeSoft
 *
 * This software may be freely redistributed under the terms of the GNU
 * public license.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 *
 */

/*
 * Portions from Erik Troan (ewt@redhat.com)
 *
 * Copyright 1996 Red Hat Software
 *
 */

#include <errno.h>
#include <string.h>
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <dirent.h>
#include <fcntl.h>
#include <limits.h>
#include <sys/klog.h>
#include <sys/mount.h>
#include <sys/stat.h>
#include <sys/vfs.h>
#include <sys/wait.h>
#include <linux/vt.h>

#include "config-stage1.h"
#include "lomount.h"
#include "tools.h"
#include "common.h"
#include "udev.h"

#define RAMFS_MAGIC 0x858458f6
#define TMPFS_MAGIC 0x01021994

#ifndef MNT_DETACH /* sys/mount.h still doesn't define it */
#define MNT_DETACH 2
#endif

/*
 * This need to safe symbols export
 */
#define fatal(m)	init_fatal(m)
#define warn(m)		init_warn(m)

char * const env[] = {
	"PATH=/usr/bin:/bin:/sbin:/usr/sbin",
	"HOME=/",
	"RUN_INITRD=1",
	"TERM=linux",
	"TERMINFO=/etc/terminfo",
	NULL
};

char ** myenv = NULL;

/*
 * Need to cooperate with make-inird scripts?
 *
 * 0=no, -1=yes.
 *
 * It is also turn on lazy initialization mode (after
 * /dev, /sys and /proc mounts, after udevd started).
 * This mode require make-initrd generated initramfs
 * image and patched make-initrd-propagator with hook
 * /scripts/pre/prepare/060-propagator. By default,
 * propagator is master to udev control.
 */
int cooperate_mode = 0;

char *init_top[] = {"/sbin/init-top", "init-top", NULL};
char *init_premount[] = {"/sbin/init-premount", "init-premount", NULL};
char *init_bottom[] = {"/sbin/init-bottom", "init-bottom", NULL};

extern void stage1();

/*
 * this needs to handle the following cases:
 *
 *  1) run from a CD root filesystem
 *  2) run from a read only nfs rooted filesystem
 *  3) run from a floppy
 *  4) run from a floppy that's been loaded into a ramdisk
 */

void init_fatal(const char *msg)
{
	printf("FATAL ERROR IN INIT: %s\nI can't recover from this, "
			"please reboot manually and send bugreport.\n", msg);
	for(;;) sleep(600);
}

void init_warn(const char *msg)
{
	printf("W: %s\n", msg);
}

static int _mknod(const char *pathname, mode_t mode, dev_t dev)
{
	int rc;

	rc = mknod(pathname, mode, dev);
	if (rc < 0 && errno == EEXIST)
		rc = 0;

	return rc;
}

/* fork to:
 *   (1) watch /proc/kmsg and copy the stuff to /dev/tty4
 *   (2) listens to /dev/log and copy also this stuff (log from programs)
 */
static pid_t doklog()
{
	int in, out, i;
	pid_t pid;
	char buf[1024];

	/* open kernel message logger */
	if ((in = open("/proc/kmsg", O_RDONLY, 0)) < 0)
		fatal("failed to open /proc/kmsg");

	if (_mknod("/dev/tty4", S_IFCHR, MKDEV(4, 4)) < 0 ||
		(out = open("/dev/tty4", O_WRONLY, 0)) < 0)
		fatal("failed to open /dev/tty4");

	if ((pid = fork())) {
		if (pid < 0) fatal("doklog");
		close(in);
		close(out);
		return pid;
	}

	/* child */
	close(0);
	close(1);
	close(2);

	/* disable on-console syslog output */
	klogctl(8, NULL, 1);

	for (;;) {
		if ((i = read(in, buf, sizeof(buf))) > 0)
			i = write(out, buf, i);
	}
}

pid_t spawn(char * const av[])
{
	pid_t pid;

	if ((pid = fork())) {
		if (pid < 0) fatal(av[0]);
		return pid;
	} else {
		execve(av[0], &av[1], env);
		perror(av[0]);
		exit(1);
	}
}

static void spawn_hook(char *av[])
{
	struct stat st;
	int status;

	if (stat(av[0], &st) || !S_ISREG(st.st_mode) || !(st.st_mode & S_IXUSR)) return;
	if (waitpid(spawn(av), &status, 0) < 0 || !(WIFEXITED(status))) fatal(av[0]);
}

static void take_env(int fd)
{
	static char buf[PIPE_BUF];
	char *p = buf;
	char **ep;
	int i;

	if ((i = read(fd, buf, sizeof(buf))) < 0) {
		warn("failed to read env from pipe");
		myenv = (char **) env;
		return;
	}

	close(fd);
	buf[i] = '\0';

	if ((ep = myenv = (char **) malloc(sizeof(char *) * 32)) == NULL) {
		warn("can't malloc env");
		myenv = (char **) env;
		return;
	}

	do {
		*ep++ = p;
		p += strlen(p);
	} while(++p < buf+i);
	*ep = NULL;
}

/*
  initramfs cleaner
*/
static int nuke(const char *what);

static int nuke_dirent(int len, const char *dir, const char *name, dev_t me)
{
	int bytes = len+strlen(name)+2;
	char path[bytes];
	struct stat st;

#if 0 /* This is a way to suppress warning message about unused xlen */
	int xlen = snprintf(path, bytes, "%s/%s", dir, name);
	assert(xlen < bytes);
#else
	snprintf(path, bytes, "%s/%s", dir, name);
#endif

	if ( lstat(path, &st) )
		return ENOENT;		/* Return 0 since already gone? */

	if ( st.st_dev != me )
		return 0;		/* DO NOT recurse down mount points!!!!! */

	return nuke(path);
}

/* Wipe the contents of a directory, but not the directory itself */
int nuke_dir(const char *what)
{
	int len = strlen(what);
	DIR *dir;
	struct dirent *d;
	int err = 0;
	struct stat st;

	if ( lstat(what, &st) )
		return errno;

	if ( !S_ISDIR(st.st_mode) )
		return ENOTDIR;

	if ( !(dir = opendir(what)) ) {
		/* EACCES means we can't read it.  Might be empty and removable;
		   if not, the rmdir() in nuke() will trigger an error. */
		return (errno == EACCES) ? 0 : errno;
	}

	while ( (d = readdir(dir)) ) {
		/* Skip . and .. */
		if ( d->d_name[0] == '.' &&
			 (d->d_name[1] == '\0' ||
			  (d->d_name[1] == '.' && d->d_name[2] == '\0')) )
			continue;

		err = nuke_dirent(len, what, d->d_name, st.st_dev);
		if ( err ) {
			closedir(dir);
			return err;
		}
	}

	closedir(dir);

	return 0;
}

static int nuke(const char *what)
{
	int rv;
	int err = 0;

	rv = unlink(what);
	if ( rv < 0 ) {
		if ( errno == EISDIR ) {
			/* It's a directory. */
			err = nuke_dir(what);
			if ( !err ) err = rmdir(what) ? errno : err;
		} else {
			err = errno;
		}
	}

	if ( err ) {
		errno = err;
		fatal(what);
	}
	return 0;
}

struct filesystem
{
	char * dev;
	char * name;
	char * fs;
	int mounted;
};

/* attempt to unmount all filesystems in /proc/mounts */
static void unmount_filesystems(void)
{
	int fd, size;
	char buf[65535];	/* this should be big enough */
	char *p;
	struct filesystem fs[500];
	int numfs = 0;
	int i, nb;

	printf("unmounting filesystems...\n");

	fd = open("/proc/mounts", O_RDONLY, 0);
	if (fd < 1) {
		warn("failed to open /proc/mounts");
		sleep(2);
		return;
	}

	size = read(fd, buf, sizeof(buf) - 1);
	buf[size] = '\0';

	close(fd);

	p = buf;
	while (*p) {
		fs[numfs].mounted = 1;
		fs[numfs].dev = p;
		while (*p != ' ') p++;
		*p++ = '\0';
		fs[numfs].name = p;
		while (*p != ' ') p++;
		*p++ = '\0';
		fs[numfs].fs = p;
		while (*p != ' ') p++;
		*p++ = '\0';
		while (*p != '\n') p++;
		p++;
		if (strcmp(fs[numfs].name, "/") != 0) numfs++; /* skip if root, no need to take initrd root in account */
	}

	/* Pixel's ultra-optimized sorting algorithm:
	   multiple passes trying to umount everything until nothing moves
	   anymore (a.k.a holy shotgun method) */
	do {
		nb = 0;
		for (i = 0; i < numfs; i++) {
			/*printf("trying with %s\n", fs[i].name);*/
			if (fs[i].mounted && umount(fs[i].name) == 0) {
				if (strncmp(fs[i].dev + sizeof("/dev/") - 1, "loop",
					    sizeof("loop") - 1) == 0)
					del_loop(fs[i].dev);

				printf("\t%s\n", fs[i].name);
				fs[i].mounted = 0;
				nb++;
			}
		}
	} while (nb);

	for (i = nb = 0; i < numfs; i++)
		if (fs[i].mounted) {
			printf("\t%s umount failed\n", fs[i].name);
			if (strcmp(fs[i].fs, "ext2") == 0) nb++; /* don't count not-ext2 umount failed */
		}

	if (nb) {
		fatal("failed to umount some filesystems\n");
	}
}

int main(int argc, char **argv)
{
	static const char localhost[] = "localhost.localdomain";
	struct stat rst, cst, ist, pst;
	struct statfs sfs;
	pid_t pid, klogpid;
	sigset_t sig;
	int wait_status;
	int fd, fds[2];
	char *init = NULL;

	/* auto-detect lazy initialization mode */
	fd = open("/proc/mounts", O_RDONLY, 0);
	if (fd >= 0) {
		close(fd);
		cooperate_mode = -1;
	}
	fd = -1;

	if (!cooperate_mode) {
		if (mount("/proc", "/proc", "proc", 0, NULL))
			fatal("failed to mount proc filesystem");
		if (mount("sysfs", "/sys", "sysfs", 0, NULL))
			fatal("failed to mount sysfs filesystem");
		if (mount("udevfs", "/dev", "devtmpfs", 0, "size=8M,mode=0755")) {
			switch (errno) {
			case ENODEV:
				/* There is no devtmpfs for current kernel, try mount tmpfs */
				if (mount("udev", "/dev", "tmpfs", 0, "size=8M,mode=0755"))
					fatal("failed to mount tmpfs filesystem");
			case EBUSY:
				/* Don't mount /dev if it is already mounted */
				break;
			default:
				fatal("failed to mount devtmpfs filesystem");
			}
		}
	}

	/* ignore Control-C and keyboard stop signals */
	sigemptyset(&sig);
	sigaddset(&sig, SIGINT);
	sigaddset(&sig, SIGTSTP);
	sigprocmask(SIG_BLOCK, &sig, NULL);

	if (!cooperate_mode) {
		if (_mknod("/dev/console", S_IFCHR, MKDEV(5, 1)) < 0 ||
			(fd = open("/dev/console", O_RDWR, 0)) < 0)
		{
			fatal("failed to open /dev/console");
		}

		dup2(fd, 0);
		dup2(fd, 1);
		dup2(fd, 2);
		close(fd);

		if (_mknod("/dev/null", S_IFCHR, MKDEV(1, 3)) < 0)
			fatal("failed to create /dev/null");
	}

	/* I set me up as session leader (probably not necessary?) */
	setsid();
	if (ioctl(0, TIOCSCTTY, NULL)) {
		perror("TIOCSCTTY");
		warn("could not set new controlling tty");
	}

	if (sethostname(localhost, sizeof(localhost)) < 0)
		warn("could not set hostname");

	/* the default domainname (as of 2.0.35) is "(none)", which confuses glibc */
	if (setdomainname("", 0) < 0)
		warn("could not set domainname");

	if (mkdirs_dev(".initramfs", "pts", "shm", NULL) < 0)
		fatal("mkdir\n");

	klogpid = doklog();
	spawn_hook(init_top);
	udev_start();
	spawn_hook(init_premount);

	/* Go into normal init mode - keep going, and then do a orderly shutdown
	   when:

	   1) install exits
	   2) we receive a SIGHUP
	*/

	printf("Running stage1...\n");

	/* create a pipe for env passing */
	if (pipe(fds) < 0)
		fatal("failed to create env pipe");

	fcntl(fds[0], F_SETFD, 1);
	fcntl(fds[1], F_SETFD, 0);

	if ((pid = fork())) {
		if (pid < 0) fatal("Failed to spawn stage1");
		close(fds[1]);
		while (pid != wait(&wait_status));
	} else {
		stage1();
	}

	if (!(WIFEXITED(wait_status))) {
		/* something went wrong */

		printf("wait_status: %i, install exited abnormally ", wait_status);
		if (WIFSIGNALED(wait_status))
			printf("-- received signal %d", WTERMSIG(wait_status));
		printf("\n");

		sync(); sync();

		printf("sending termination signals...");
		kill(-1, 15);
		sleep(2);
		printf("done\n");

		printf("sending kill signals...");
		kill(-1, 9);
		sleep(2);
		printf("done\n");

		unmount_filesystems();

		printf("you may safely reboot your system\n");
		for (;;) sleep(600);
	}

	take_env(fds[0]);
	udev_stop();

	spawn_hook(init_bottom);

	kill(klogpid, 9);
	waitpid(klogpid, &wait_status, 0);

	/* deallocate all unused consoles */
	ioctl(0, VT_DISALLOCATE, 0);

	printf("Spawning init ...");

	/* rest was seamlessy stolen from klibc */
	/* First, change to the new root directory */
	if (chdir(STAGE2_LOCATION))
		fatal("chdir to new root");
	if (stat("/bin/plymouth", &pst) == 0 ) {
		char * plymouth[] =   { "/bin/plymouth", "plymouth", "--newroot=/root", NULL };
		spawn(plymouth);
	}

	/* This is a potentially highly destructive program.  Take some
	   extra precautions. */

	/* Make sure the current directory is not on the same filesystem
	   as the root directory */
	if (stat("/", &rst) || stat(".", &cst))
		fatal("stat");

	if ( rst.st_dev == cst.st_dev )
		fatal("current directory on the same filesystem as the root");

	/* The initramfs should have /init */
	if ( stat("/init", &ist) || !S_ISREG(ist.st_mode) )
		fatal("can't find /init on initramfs");

	/* Make sure we're on a ramfs */
	if ( statfs("/", &sfs) )
		fatal("statfs /");
	if ( sfs.f_type != RAMFS_MAGIC && sfs.f_type != TMPFS_MAGIC )
		fatal("rootfs not a ramfs or tmpfs");

	/* Okay, I think we should be safe... */

	/* overmount image under new root if needed */
	if ( statfs(IMAGE_LOCATION, &sfs) )
		fatal("statfs() on "IMAGE_LOCATION);
	/* if something is mounted under IMAGE_LOCATION ? */
	if ( sfs.f_type != RAMFS_MAGIC && sfs.f_type != TMPFS_MAGIC ) {
		if ( mount(IMAGE_LOCATION, "." IMAGE_LOCATION, NULL, MS_MOVE, NULL) )
			fatal("overmounting "IMAGE_LOCATION);
		/* test for nested mount: disk or nfs with iso image */
		if ( statfs(IMAGE_LOCATION, &sfs) )
			fatal("statfs() on nested "IMAGE_LOCATION);
		if ( sfs.f_type != RAMFS_MAGIC && sfs.f_type != TMPFS_MAGIC )
			if (umount2(IMAGE_LOCATION, MNT_DETACH))
				fatal("lazy umounting nested "IMAGE_LOCATION);
	}

	umount("/sys");
	umount("/proc/bus/usb");
	umount("/proc");

	if (mount("/dev", "./dev", NULL, MS_MOVE, NULL))
		fatal("overmounting /dev");

	/* Delete rootfs contents */
	if (nuke_dir("/"))
		fatal("nuking initramfs contents");

	/* Overmount the root */
	if (mount(".", "/", NULL, MS_MOVE, NULL))
		fatal("overmounting root");

	/* chroot, chdir */
	if (chroot(".") || chdir("/"))
		fatal("chroot");

	/* Check for given init */
	init = get_from_env("INIT", (const char* const*) myenv);
	if (init == NULL) init = STAGE2_BINNAME;

	if (stat(init, &ist) || !S_ISREG(ist.st_mode) || !(ist.st_mode & S_IXUSR))
		fatal("can't find init on root fs");

	/* Spawn init */
	printf(" done.\n");

	/* unblock signals */
	sigprocmask(SIG_UNBLOCK, &sig, NULL);

	argv[0] = init;
	execve(argv[0], argv, myenv);
	fatal("stage2");			/* Failed to spawn init */
	return 0;
}