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 
 *
 */

#ifndef INIT_HEADERS
#include "init-libc-headers.h"
#else
#include INIT_HEADERS
#endif

#include "config-stage1.h"

#if defined(__powerpc__)
#define TIOCSCTTY     0x540
#endif

char * env[] = {
	"PATH=/usr/bin:/bin:/sbin:/usr/sbin:/usr/X11R6/bin",
	"LD_LIBRARY_PATH=/lib:/usr/lib:/usr/X11R6/lib",
	"HOME=/",
	"TERM=linux",
	"TERMINFO=/etc/terminfo",
	NULL
};

char ** myenv = NULL;

/* 
 * 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 
 */

int klog_pid;


void fatal_error(const char *msg)
{
	printf("FATAL ERROR IN INIT: %s : %s\n\nI can't recover from this, please reboot manually and send bugreport.\n", msg,strerror(errno));
	while (1);
}

void print_error(char *msg)
{
	printf("E: %s\n", msg);
}

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

/* 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)
 */
void doklog()
{
	int in, out, i, ii;
	int log;
	char buf[1024];

	/* open kernel message logger */
	in = open("/proc/kmsg", O_RDONLY,0);
	if (in < 0) {
		print_error("could not open /proc/kmsg");
		return;
	}

	if ((log = open("/tmp/syslog", O_WRONLY | O_CREAT, 0644)) < 0) {
		perror("/tmp/syslog");
		print_error("error opening /tmp/syslog");
		sleep(5);
		return;
	}

	if ((klog_pid = fork())) {
		close(in);
		close(log);
		return;
	} else {
		close(0); 
		close(1);
		close(2);
	}
	
	out = open("/dev/tty4", O_WRONLY, 0);
	if (out < 0) {
		perror("can't open /dev/tty4");
		print_warning("couldn't open tty for syslog -- still using /tmp/syslog\n");
	}

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

	while (1) {
		i = read(in, buf, sizeof(buf));
		if (i > 0) {
			if (out >= 0)
				ii = write(out, buf, i);
			ii = write(log, buf, i);
		}
	}
}

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

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

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

	if ((ep = myenv = malloc(sizeof(char *) * 32)) == NULL) {
		print_error("can't malloc env");
		myenv = env;
		return;
	}
	
	do {
		*ep++ = p;
		p += strlen(p);
	} while(++p < buf+i);
	*ep = NULL;
}

#define LOOP_CLR_FD	0x4C01

void del_loop(char *device) 
{
	int fd;
	if ((fd = open(device, O_RDONLY, 0)) < 0) {
		printf("del_loop open failed\n");
		return;
	}

	if (ioctl(fd, LOOP_CLR_FD, 0) < 0) {
		printf("del_loop ioctl failed");
		return;
	}

	close(fd);
}

/*
  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];
	int xlen;
	struct stat st;
	
	xlen = snprintf(path, bytes, "%s/%s", dir, name);
	/* assert(xlen < bytes); */
	
	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 */
static 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_error(what);
	}
	return 0;
}

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

/* attempt to unmount all filesystems in /proc/mounts */
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) {
		print_error("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) {
		printf("failed to umount some filesystems\n");
		while (1);
	}
}

#define MAJOR(dev)      ((dev)>>8)
#define MINOR(dev)      ((dev) & 0xff)
#define MKDEV(ma,mi)    ((ma)<<8 | (mi))

#define DEV_PERM 00400|00200|00040|00020
#define CHR_DEV 0020000|DEV_PERM
#define BLK_DEV 0060000|DEV_PERM

#define RAMFS_MAGIC 0x858458f6
#define TMPFS_MAGIC 0x01021994

int main(int argc, char **argv)
{
	struct stat rst, cst, ist;
	struct statfs sfs;
	pid_t installpid, childpid;
	int wait_status;
	int fd;
	int fds[2];
	int end_stage1 = 0;
	
	if (mount("/proc", "/proc", "proc", 0, NULL))
		fatal_error("Unable to mount proc filesystem");
	if (mount("/sysfs", "/sys", "sysfs", 0, NULL))
		fatal_error("Unable to mount sysfs filesystem");
	
	/* ignore Control-C and keyboard stop signals */
	signal(SIGINT, SIG_IGN);
	signal(SIGTSTP, SIG_IGN);

	fd = open("/dev/console", O_RDWR, 0);
	if (fd < 0)
		fatal_error("failed to open /dev/console");
		
	dup2(fd, 0);
	dup2(fd, 1);
	dup2(fd, 2);
	close(fd);

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

	char my_hostname[] = "localhost.localdomain";
	if (sethostname(my_hostname, sizeof(my_hostname)) < 0)
		print_error("could not set hostname");
		
	/* the default domainname (as of 2.0.35) is "(none)", which confuses glibc */
	if (setdomainname("", 0) < 0)
		print_error("could not set domainname");

	doklog();

	/* 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_error("failed to create env pipe");

	fcntl(fds[0], F_SETFD, 1);
	fcntl(fds[1], F_SETFD, 0);
	
	if (!(installpid = fork())) {
		/* child */
		char * child_argv[2];
		child_argv[0] = "/sbin/stage1";
		child_argv[1] = NULL;
		execve(child_argv[0], child_argv, env);
		printf("error in exec of stage1 :-(\n");
		return 0;
	}

	close(fds[1]);
	
	while (!end_stage1) {
		childpid = wait4(-1, &wait_status, 0, NULL);
		if (childpid == installpid)
			end_stage1 = 1;
	}

	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");
		while (1);
	}

	grab_env(fds[0]);

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

	printf("Spawning init ...");

	/* rest was seamlessy stolen from klibc */
	/* First, change to the new root directory */
	if (chdir(STAGE2_LOCATION))
		fatal_error("chdir to new root");

	/* 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_error("stat");

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

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

	/* Make sure we're on a ramfs */
	if ( statfs("/", &sfs) )
		fatal_error("statfs /");
	if ( sfs.f_type != RAMFS_MAGIC && sfs.f_type != TMPFS_MAGIC )
		fatal_error("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_error("statfs "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_error("overmounting image location");
		/* test for nested mount: disk or nfs with iso image */
		if ( statfs(IMAGE_LOCATION, &sfs) )
			fatal_error("nested statfs "IMAGE_LOCATION);
		if ( sfs.f_type != RAMFS_MAGIC && sfs.f_type != TMPFS_MAGIC )
			if ( mount(IMAGE_LOCATION, "." IMAGE_LOCATION "/isolinux", NULL, MS_MOVE, NULL) )
				fatal_error("overmounting nested image location");
	}

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

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

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

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

	/* Open /dev/console */
	if ((fd = open("/dev/console", O_RDWR)) != -1) {
		dup2(fd, 0);
		dup2(fd, 1);
		dup2(fd, 2);
		close(fd);
	}

	/* Check for given init */
	if (stat(STAGE2_BINNAME, &ist) || !S_ISREG(ist.st_mode))
		fatal_error("can't find init on root fs");

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

	argv[0] = STAGE2_BINNAME;
	execve(argv[0], argv, myenv);
	fatal_error("stage2");			/* Failed to spawn init */
	return 0;
}