manager.c

Minix3.11的源码。[MINIX 3是一个为高可靠性应用而设计的自由且简洁的类UNIX系统。]

/*
 * Changes:
 *   Jul 22, 2005:	Created  (Jorrit N. Herder)
 */

#include "inc.h"
#include <unistd.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <minix/dmap.h>

/* Allocate variables. */
struct rproc rproc[NR_SYS_PROCS];		/* system process table */
struct rproc *rproc_ptr[NR_PROCS];		/* mapping for fast access */
int nr_in_use; 					/* number of services */
extern int errno;				/* error status */

/* Prototypes for internal functions that do the hard work. */
FORWARD _PROTOTYPE( int start_service, (struct rproc *rp) );
FORWARD _PROTOTYPE( int stop_service, (struct rproc *rp,int how) );

PRIVATE int shutting_down = FALSE;

#define EXEC_FAILED	49			/* recognizable status */

/*===========================================================================*
 *					do_up				     *
 *===========================================================================*/
PUBLIC int do_up(m_ptr)
message *m_ptr;					/* request message pointer */
{
/* A request was made to start a new system service. Dismember the request 
 * message and gather all information needed to start the service. Starting
 * is done by a helper routine.
 */
  register struct rproc *rp;			/* system process table */
  int slot_nr;					/* local table entry */
  int arg_count;				/* number of arguments */
  char *cmd_ptr;				/* parse command string */
  enum dev_style dev_style;			/* device style */
  int s;					/* status variable */

  /* See if there is a free entry in the table with system processes. */
  if (nr_in_use >= NR_SYS_PROCS) return(EAGAIN); 
  for (slot_nr = 0; slot_nr < NR_SYS_PROCS; slot_nr++) {
      rp = &rproc[slot_nr];			/* get pointer to slot */
      if (! rp->r_flags & RS_IN_USE) 		/* check if available */
	  break;
  }
  nr_in_use ++;					/* update administration */

  /* Obtain command name and parameters. This is a space-separated string
   * that looks like "/sbin/service arg1 arg2 ...". Arguments are optional.
   */
  if (m_ptr->RS_CMD_LEN > MAX_COMMAND_LEN) return(E2BIG);
  if (OK!=(s=sys_datacopy(m_ptr->m_source, (vir_bytes) m_ptr->RS_CMD_ADDR, 
  	SELF, (vir_bytes) rp->r_cmd, m_ptr->RS_CMD_LEN))) return(s);
  rp->r_cmd[m_ptr->RS_CMD_LEN] = '\0';		/* ensure it is terminated */
  if (rp->r_cmd[0] != '/') return(EINVAL);	/* insist on absolute path */

  /* Build argument vector to be passed to execute call. The format of the
   * arguments vector is: path, arguments, NULL. 
   */
  arg_count = 0;				/* initialize arg count */
  rp->r_argv[arg_count++] = rp->r_cmd;		/* start with path */
  cmd_ptr = rp->r_cmd;				/* do some parsing */ 
  while(*cmd_ptr != '\0') {			/* stop at end of string */
      if (*cmd_ptr == ' ') {			/* next argument */
          *cmd_ptr = '\0';			/* terminate previous */
	  while (*++cmd_ptr == ' ') ; 		/* skip spaces */
	  if (*cmd_ptr == '\0') break;		/* no arg following */
	  if (arg_count>MAX_NR_ARGS+1) break;	/* arg vector full */
          rp->r_argv[arg_count++] = cmd_ptr;	/* add to arg vector */
      }
      cmd_ptr ++;				/* continue parsing */
  }
  rp->r_argv[arg_count] = NULL;			/* end with NULL pointer */
  rp->r_argc = arg_count;

  /* Initialize some fields. */
  rp->r_period = m_ptr->RS_PERIOD;
  rp->r_dev_nr = m_ptr->RS_DEV_MAJOR;
  rp->r_dev_style = STYLE_DEV; 
  rp->r_restarts = -1; 				/* will be incremented */
  
  /* All information was gathered. Now try to start the system service. */
  return(start_service(rp));
}


/*===========================================================================*
 *				do_down					     *
 *===========================================================================*/
PUBLIC int do_down(message *m_ptr)
{
  register struct rproc *rp;
  pid_t pid = (pid_t) m_ptr->RS_PID;

  for (rp=BEG_RPROC_ADDR; rp<END_RPROC_ADDR; rp++) {
      if (rp->r_flags & RS_IN_USE && rp->r_pid == pid) {
#if VERBOSE
	  printf("stopping %d (%d)\n", pid, m_ptr->RS_PID);
#endif
	  stop_service(rp,RS_EXITING);
	  return(OK);
      }
  }
#if VERBOSE
  printf("not found %d (%d)\n", pid, m_ptr->RS_PID);
#endif
  return(ESRCH);
}


/*===========================================================================*
 *				do_refresh				     *
 *===========================================================================*/
PUBLIC int do_refresh(message *m_ptr)
{
  register struct rproc *rp;
  pid_t pid = (pid_t) m_ptr->RS_PID;

  for (rp=BEG_RPROC_ADDR; rp<END_RPROC_ADDR; rp++) {
      if (rp->r_flags & RS_IN_USE && rp->r_pid == pid) {
#if VERBOSE
	  printf("refreshing %d (%d)\n", pid, m_ptr->RS_PID);
#endif
	  stop_service(rp,RS_REFRESHING);
	  return(OK);
      }
  }
#if VERBOSE
  printf("not found %d (%d)\n", pid, m_ptr->RS_PID);
#endif
  return(ESRCH);
}

/*===========================================================================*
 *				do_rescue				     *
 *===========================================================================*/
PUBLIC int do_rescue(message *m_ptr)
{
  char rescue_dir[MAX_RESCUE_DIR_LEN];
  int s;

  /* Copy rescue directory from user. */
  if (m_ptr->RS_CMD_LEN > MAX_RESCUE_DIR_LEN) return(E2BIG);
  if (OK!=(s=sys_datacopy(m_ptr->m_source, (vir_bytes) m_ptr->RS_CMD_ADDR, 
  	SELF, (vir_bytes) rescue_dir, m_ptr->RS_CMD_LEN))) return(s);
  rescue_dir[m_ptr->RS_CMD_LEN] = '\0';		/* ensure it is terminated */
  if (rescue_dir[0] != '/') return(EINVAL);	/* insist on absolute path */

  /* Change RS' directory to the rescue directory. Provided that the needed
   * binaries are in the rescue dir, this makes recovery possible even if the 
   * (root) file system is no longer available, because no directory lookups
   * are required. Thus if an absolute path fails, we can try to strip the 
   * path an see if the command is in the rescue dir. 
   */
  if (chdir(rescue_dir) != 0) return(errno);
  return(OK);
}

/*===========================================================================*
 *				do_shutdown				     *
 *===========================================================================*/
PUBLIC int do_shutdown(message *m_ptr)
{
  /* Set flag so that RS server knows services shouldn't be restarted. */
  shutting_down = TRUE;
  return(OK);
}

/*===========================================================================*
 *				do_exit					     *
 *===========================================================================*/
PUBLIC void do_exit(message *m_ptr)
{
  register struct rproc *rp;
  pid_t exit_pid;
  int exit_status;

#if VERBOSE
  printf("RS: got SIGCHLD signal, doing wait to get exited child.\n");
#endif

  /* See which child exited and what the exit status is. This is done in a
   * loop because multiple childs may have exited, all reported by one 
   * SIGCHLD signal. The WNOHANG options is used to prevent blocking if, 
   * somehow, no exited child can be found. 
   */
  while ( (exit_pid = waitpid(-1, &exit_status, WNOHANG)) != 0 ) {

#if VERBOSE
      printf("RS: proc %d, pid %d, ", rp->r_proc_nr, exit_pid); 
      if (WIFSIGNALED(exit_status)) {
          printf("killed, signal number %d\n", WTERMSIG(exit_status));
      } 
      else if (WIFEXITED(exit_status)) {
          printf("normal exit, status %d\n", WEXITSTATUS(exit_status));
      }
#endif

      /* Search the system process table to see who exited. 
       * This should always succeed. 
       */
      for (rp=BEG_RPROC_ADDR; rp<END_RPROC_ADDR; rp++) {
          if ((rp->r_flags & RS_IN_USE) && rp->r_pid == exit_pid) {

              rproc_ptr[rp->r_proc_nr] = NULL;		/* invalidate */

              if ((rp->r_flags & RS_EXITING) || shutting_down) {
		  rp->r_flags = 0;			/* release slot */
		  rproc_ptr[rp->r_proc_nr] = NULL;
	      }
	      else if(rp->r_flags & RS_REFRESHING) {
		      rp->r_restarts = -1;		/* reset counter */
		      start_service(rp);		/* direct restart */
	      }
              else if (WIFEXITED(exit_status) &&
		      WEXITSTATUS(exit_status) == EXEC_FAILED) {
		  rp->r_flags = 0;			/* release slot */
              }
	      else {
#if VERBOSE
		  printf("Unexpected exit. Restarting %s\n", rp->r_cmd);
#endif
                  /* Determine what to do. If this is the first unexpected 
		   * exit, immediately restart this service. Otherwise use