exec.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。]

/* This file handles the EXEC system call.  It performs the work as follows:
 *    - see if the permissions allow the file to be executed
 *    - read the header and extract the sizes
 *    - fetch the initial args and environment from the user space
 *    - allocate the memory for the new process
 *    - copy the initial stack from PM to the process
 *    - read in the text and data segments and copy to the process
 *    - take care of setuid and setgid bits
 *    - fix up 'mproc' table
 *    - tell kernel about EXEC
 *    - save offset to initial argc (for ps)
 *
 * The entry points into this file are:
 *   do_exec:	 perform the EXEC system call
 *   rw_seg:	 read or write a segment from or to a file
 *   find_share: find a process whose text segment can be shared
 */

#include "pm.h"
#include <sys/stat.h>
#include <minix/callnr.h>
#include <minix/com.h>
#include <a.out.h>
#include <signal.h>
#include <string.h>
#include "mproc.h"
#include "param.h"

FORWARD _PROTOTYPE( int new_mem, (struct mproc *sh_mp, vir_bytes text_bytes,
		vir_bytes data_bytes, vir_bytes bss_bytes,
		vir_bytes stk_bytes, phys_bytes tot_bytes)		);
FORWARD _PROTOTYPE( void patch_ptr, (char stack[ARG_MAX], vir_bytes base) );
FORWARD _PROTOTYPE( int insert_arg, (char stack[ARG_MAX],
		vir_bytes *stk_bytes, char *arg, int replace)		);
FORWARD _PROTOTYPE( char *patch_stack, (int fd, char stack[ARG_MAX],
		vir_bytes *stk_bytes, char *script)			);
FORWARD _PROTOTYPE( int read_header, (int fd, int *ft, vir_bytes *text_bytes,
		vir_bytes *data_bytes, vir_bytes *bss_bytes,
		phys_bytes *tot_bytes, long *sym_bytes, vir_clicks sc,
		vir_bytes *pc)						);

#define ESCRIPT	(-2000)	/* Returned by read_header for a #! script. */
#define PTRSIZE	sizeof(char *) /* Size of pointers in argv[] and envp[]. */

/*===========================================================================*
 *				do_exec					     *
 *===========================================================================*/
PUBLIC int do_exec()
{
/* Perform the execve(name, argv, envp) call.  The user library builds a
 * complete stack image, including pointers, args, environ, etc.  The stack
 * is copied to a buffer inside PM, and then to the new core image.
 */
  register struct mproc *rmp;
  struct mproc *sh_mp;
  int m, r, fd, ft, sn;
  static char mbuf[ARG_MAX];	/* buffer for stack and zeroes */
  static char name_buf[PATH_MAX]; /* the name of the file to exec */
  char *new_sp, *name, *basename;
  vir_bytes src, dst, text_bytes, data_bytes, bss_bytes, stk_bytes, vsp;
  phys_bytes tot_bytes;		/* total space for program, including gap */
  long sym_bytes;
  vir_clicks sc;
  struct stat s_buf[2], *s_p;
  vir_bytes pc;

  /* Do some validity checks. */
  rmp = mp;
  stk_bytes = (vir_bytes) m_in.stack_bytes;
  if (stk_bytes > ARG_MAX) return(ENOMEM);	/* stack too big */
  if (m_in.exec_len <= 0 || m_in.exec_len > PATH_MAX) return(EINVAL);

  /* Get the exec file name and see if the file is executable. */
  src = (vir_bytes) m_in.exec_name;
  dst = (vir_bytes) name_buf;
  r = sys_datacopy(who, (vir_bytes) src,
		PM_PROC_NR, (vir_bytes) dst, (phys_bytes) m_in.exec_len);
  if (r != OK) return(r);	/* file name not in user data segment */

  /* Fetch the stack from the user before destroying the old core image. */
  src = (vir_bytes) m_in.stack_ptr;
  dst = (vir_bytes) mbuf;
  r = sys_datacopy(who, (vir_bytes) src,
  			PM_PROC_NR, (vir_bytes) dst, (phys_bytes)stk_bytes);
  /* can't fetch stack (e.g. bad virtual addr) */
  if (r != OK) return(EACCES);	

  r = 0;	/* r = 0 (first attempt), or 1 (interpreted script) */
  name = name_buf;	/* name of file to exec. */
  do {
	s_p = &s_buf[r];
	tell_fs(CHDIR, who, FALSE, 0);  /* switch to the user's FS environ */
	fd = allowed(name, s_p, X_BIT);	/* is file executable? */
	if (fd < 0)  return(fd);		/* file was not executable */

	/* Read the file header and extract the segment sizes. */
	sc = (stk_bytes + CLICK_SIZE - 1) >> CLICK_SHIFT;

	m = read_header(fd, &ft, &text_bytes, &data_bytes, &bss_bytes, 
					&tot_bytes, &sym_bytes, sc, &pc);
	if (m != ESCRIPT || ++r > 1) break;
  } while ((name = patch_stack(fd, mbuf, &stk_bytes, name_buf)) != NULL);

  if (m < 0) {
	close(fd);		/* something wrong with header */
	return(stk_bytes > ARG_MAX ? ENOMEM : ENOEXEC);
  }

  /* Can the process' text be shared with that of one already running? */
  sh_mp = find_share(rmp, s_p->st_ino, s_p->st_dev, s_p->st_ctime);

  /* Allocate new memory and release old memory.  Fix map and tell kernel. */
  r = new_mem(sh_mp, text_bytes, data_bytes, bss_bytes, stk_bytes, tot_bytes);
  if (r != OK) {
	close(fd);		/* insufficient core or program too big */
	return(r);
  }

  /* Save file identification to allow it to be shared. */
  rmp->mp_ino = s_p->st_ino;
  rmp->mp_dev = s_p->st_dev;
  rmp->mp_ctime = s_p->st_ctime;

  /* Patch up stack and copy it from PM to new core image. */
  vsp = (vir_bytes) rmp->mp_seg[S].mem_vir << CLICK_SHIFT;
  vsp += (vir_bytes) rmp->mp_seg[S].mem_len << CLICK_SHIFT;
  vsp -= stk_bytes;
  patch_ptr(mbuf, vsp);
  src = (vir_bytes) mbuf;
  r = sys_datacopy(PM_PROC_NR, (vir_bytes) src,
  			who, (vir_bytes) vsp, (phys_bytes)stk_bytes);
  if (r != OK) panic(__FILE__,"do_exec stack copy err on", who);

  /* Read in text and data segments. */
  if (sh_mp != NULL) {
	lseek(fd, (off_t) text_bytes, SEEK_CUR);  /* shared: skip text */
  } else {
	rw_seg(0, fd, who, T, text_bytes);
  }
  rw_seg(0, fd, who, D, data_bytes);

  close(fd);			/* don't need exec file any more */

  /* Take care of setuid/setgid bits. */
  if ((rmp->mp_flags & TRACED) == 0) { /* suppress if tracing */
	if (s_buf[0].st_mode & I_SET_UID_BIT) {
		rmp->mp_effuid = s_buf[0].st_uid;
		tell_fs(SETUID,who, (int)rmp->mp_realuid, (int)rmp->mp_effuid);
	}
	if (s_buf[0].st_mode & I_SET_GID_BIT) {
		rmp->mp_effgid = s_buf[0].st_gid;
		tell_fs(SETGID,who, (int)rmp->mp_realgid, (int)rmp->mp_effgid);
	}
  }

  /* Save offset to initial argc (for ps) */
  rmp->mp_procargs = vsp;

  /* Fix 'mproc' fields, tell kernel that exec is done,  reset caught sigs. */
  for (sn = 1; sn <= _NSIG; sn++) {
	if (sigismember(&rmp->mp_catch, sn)) {
		sigdelset(&rmp->mp_catch, sn);
		rmp->mp_sigact[sn].sa_handler = SIG_DFL;
		sigemptyset(&rmp->mp_sigact[sn].sa_mask);
	}
  }

  rmp->mp_flags &= ~SEPARATE;	/* turn off SEPARATE bit */
  rmp->mp_flags |= ft;		/* turn it on for separate I & D files */
  new_sp = (char *) vsp;

  tell_fs(EXEC, who, 0, 0);	/* allow FS to handle FD_CLOEXEC files */

  /* System will save command line for debugging, ps(1) output, etc. */
  basename = strrchr(name, '/');
  if (basename == NULL) basename = name; else basename++;
  strncpy(rmp->mp_name, basename, PROC_NAME_LEN-1);
  rmp->mp_name[PROC_NAME_LEN] = '\0';
  sys_exec(who, new_sp, basename, pc);

  /* Cause a signal if this process is traced. */
  if (rmp->mp_flags & TRACED) check_sig(rmp->mp_pid, SIGTRAP);

  return(SUSPEND);		/* no reply, new program just runs */
}

/*===========================================================================*
 *				read_header				     *
 *===========================================================================*/
PRIVATE int read_header(fd, ft, text_bytes, data_bytes, bss_bytes, 
						tot_bytes, sym_bytes, sc, pc)
int fd;				/* file descriptor for reading exec file */
int *ft;			/* place to return ft number */
vir_bytes *text_bytes;		/* place to return text size */
vir_bytes *data_bytes;		/* place to return initialized data size */
vir_bytes *bss_bytes;		/* place to return bss size */
phys_bytes *tot_bytes;		/* place to return total size */
long *sym_bytes;		/* place to return symbol table size */
vir_clicks sc;			/* stack size in clicks */
vir_bytes *pc;			/* program entry point (initial PC) */
{
/* Read the header and extract the text, data, bss and total sizes from it. */

  int m, ct;
  vir_clicks tc, dc, s_vir, dvir;
  phys_clicks totc;
  struct exec hdr;		/* a.out header is read in here */

  /* Read the header and check the magic number.  The standard MINIX header 
   * is defined in <a.out.h>.  It consists of 8 chars followed by 6 longs.
   * Then come 4 more longs that are not used here.
   *	Byte 0: magic number 0x01
   *	Byte 1: magic number 0x03
   *	Byte 2: normal = 0x10 (not checked, 0 is OK), separate I/D = 0x20
   *	Byte 3: CPU type, Intel 16 bit = 0x04, Intel 32 bit = 0x10, 
   *            Motorola = 0x0B, Sun SPARC = 0x17
   *	Byte 4: Header length = 0x20
   *	Bytes 5-7 are not used.
   *
   *	Now come the 6 longs
   *	Bytes  8-11: size of text segments in bytes
   *	Bytes 12-15: size of initialized data segment in bytes
   *	Bytes 16-19: size of bss in bytes
   *	Bytes 20-23: program entry point
   *	Bytes 24-27: total memory allocated to program (text, data + stack)
   *	Bytes 28-31: size of symbol table in bytes
   * The longs are represented in a machine dependent order,
   * little-endian on the 8088, big-endian on the 68000.
   * The header is followed directly by the text and data segments, and the 
   * symbol table (if any). The sizes are given in the header. Only the 
   * text and data segments are copied into memory by exec. The header is 
   * used here only. The symbol table is for the benefit of a debugger and 
   * is ignored here.
   */

  if ((m= read(fd, &hdr, A_MINHDR)) < 2) return(ENOEXEC);

  /* Interpreted script? */
  if (((char *) &hdr)[0] == '#' && ((char *) &hdr)[1] == '!') return(ESCRIPT);

  if (m != A_MINHDR) return(ENOEXEC);

  /* Check magic number, cpu type, and flags. */
  if (BADMAG(hdr)) return(ENOEXEC);
#if (CHIP == INTEL && _WORD_SIZE == 2)
  if (hdr.a_cpu != A_I8086) return(ENOEXEC);
#endif
#if (CHIP == INTEL && _WORD_SIZE == 4)
  if (hdr.a_cpu != A_I80386) return(ENOEXEC);
#endif
  if ((hdr.a_flags & ~(A_NSYM | A_EXEC | A_SEP)) != 0) return(ENOEXEC);

  *ft = ( (hdr.a_flags & A_SEP) ? SEPARATE : 0);    /* separate I & D or not */

  /* Get text and data sizes. */
  *text_bytes = (vir_bytes) hdr.a_text;	/* text size in bytes */
  *data_bytes = (vir_bytes) hdr.a_data;	/* data size in bytes */
  *bss_bytes  = (vir_bytes) hdr.a_bss;	/* bss size in bytes */
  *tot_bytes  = hdr.a_total;		/* total bytes to allocate for prog */
  *sym_bytes  = hdr.a_syms;		/* symbol table size in bytes */
  if (*tot_bytes == 0) return(ENOEXEC);

  if (*ft != SEPARATE) {
	/* If I & D space is not separated, it is all considered data. Text=0*/
	*data_bytes += *text_bytes;
	*text_bytes = 0;
  }
  *pc = hdr.a_entry;	/* initial address to start execution */

  /* Check to see if segment sizes are feasible. */
  tc = ((unsigned long) *text_bytes + CLICK_SIZE - 1) >> CLICK_SHIFT;
  dc = (*data_bytes + *bss_bytes + CLICK_SIZE - 1) >> CLICK_SHIFT;
  totc = (*tot_bytes + CLICK_SIZE - 1) >> CLICK_SHIFT;
  if (dc >= totc) return(ENOEXEC);	/* stack must be at least 1 click */
  dvir = (*ft == SEPARATE ? 0 : tc);
  s_vir = dvir + (totc - sc);
#if (CHIP == INTEL && _WORD_SIZE == 2)
  m = size_ok(*ft, tc, dc, sc, dvir, s_vir);
#else
  m = (dvir + dc > s_vir) ? ENOMEM : OK;
#endif
  ct = hdr.a_hdrlen & BYTE;		/* header length */
  if (ct > A_MINHDR) lseek(fd, (off_t) ct, SEEK_SET); /* skip unused hdr */
  return(m);
}

/*===========================================================================*
 *				new_mem					     *
 *===========================================================================*/
PRIVATE int new_mem(sh_mp, text_bytes, data_bytes,
	bss_bytes,stk_bytes,tot_bytes)
struct mproc *sh_mp;		/* text can be shared with this process */
vir_bytes text_bytes;		/* text segment size in bytes */
vir_bytes data_bytes;		/* size of initialized data in bytes */
vir_bytes bss_bytes;		/* size of bss in bytes */
vir_bytes stk_bytes;		/* size of initial stack segment in bytes */
phys_bytes tot_bytes;		/* total memory to allocate, including gap */
{