os_dep.c

Boost provides free peer-reviewed portable C++ source libraries. We emphasize libraries that work

/*
 * Copyright 1988, 1989 Hans-J. Boehm, Alan J. Demers
 * Copyright (c) 1991-1995 by Xerox Corporation.  All rights reserved.
 * Copyright (c) 1996-1999 by Silicon Graphics.  All rights reserved.
 * Copyright (c) 1999 by Hewlett-Packard Company.  All rights reserved.
 *
 * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
 * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
 *
 * Permission is hereby granted to use or copy this program
 * for any purpose,  provided the above notices are retained on all copies.
 * Permission to modify the code and to distribute modified code is granted,
 * provided the above notices are retained, and a notice that the code was
 * modified is included with the above copyright notice.
 */

# include "private/gc_priv.h"
# ifdef THREADS
#   include "atomic_ops.h"
# endif

# if defined(LINUX) && !defined(POWERPC)
#   include <linux/version.h>
#   if (LINUX_VERSION_CODE <= 0x10400)
      /* Ugly hack to get struct sigcontext_struct definition.  Required      */
      /* for some early 1.3.X releases.  Will hopefully go away soon. */
      /* in some later Linux releases, asm/sigcontext.h may have to   */
      /* be included instead.                                         */
#     define __KERNEL__
#     include <asm/signal.h>
#     undef __KERNEL__
#   else
      /* Kernels prior to 2.1.1 defined struct sigcontext_struct instead of */
      /* struct sigcontext.  libc6 (glibc2) uses "struct sigcontext" in     */
      /* prototypes, so we have to include the top-level sigcontext.h to    */
      /* make sure the former gets defined to be the latter if appropriate. */
#     include <features.h>
#     if 2 <= __GLIBC__
#       if 2 == __GLIBC__ && 0 == __GLIBC_MINOR__
	  /* glibc 2.1 no longer has sigcontext.h.  But signal.h	*/
	  /* has the right declaration for glibc 2.1.			*/
#         include <sigcontext.h>
#       endif /* 0 == __GLIBC_MINOR__ */
#     else /* not 2 <= __GLIBC__ */
        /* libc5 doesn't have <sigcontext.h>: go directly with the kernel   */
        /* one.  Check LINUX_VERSION_CODE to see which we should reference. */
#       include <asm/sigcontext.h>
#     endif /* 2 <= __GLIBC__ */
#   endif
# endif
# if !defined(OS2) && !defined(PCR) && !defined(AMIGA) && !defined(MACOS) \
    && !defined(MSWINCE)
#   include <sys/types.h>
#   if !defined(MSWIN32)
#   	include <unistd.h>
#   endif
# endif

# include <stdio.h>
# if defined(MSWINCE)
#   define SIGSEGV 0 /* value is irrelevant */
# else
#   include <signal.h>
# endif

#ifdef UNIX_LIKE
# include <fcntl.h>
#endif

#if defined(LINUX) || defined(LINUX_STACKBOTTOM)
# include <ctype.h>
#endif

/* Blatantly OS dependent routines, except for those that are related 	*/
/* to dynamic loading.							*/

#ifdef AMIGA
# define GC_AMIGA_DEF
# include "AmigaOS.c"
# undef GC_AMIGA_DEF
#endif

#if defined(MSWIN32) || defined(MSWINCE) || defined(CYGWIN32)
# define WIN32_LEAN_AND_MEAN
# define NOSERVICE
# include <windows.h>
  /* It's not clear this is completely kosher under Cygwin.  But it	*/
  /* allows us to get a working GC_get_stack_base.			*/
#endif

#ifdef MACOS
# include <Processes.h>
#endif

#ifdef IRIX5
# include <sys/uio.h>
# include <malloc.h>   /* for locking */
#endif

#if defined(LINUX) || defined(FREEBSD) || defined(SOLARIS) || defined(IRIX5) \
	|| defined(USE_MMAP) || defined(USE_MUNMAP)
# define MMAP_SUPPORTED
#endif

#if defined(MMAP_SUPPORTED) || defined(ADD_HEAP_GUARD_PAGES)
# if defined(USE_MUNMAP) && !defined(USE_MMAP)
    --> USE_MUNMAP requires USE_MMAP
# endif
# include <sys/types.h>
# include <sys/mman.h>
# include <sys/stat.h>
# include <errno.h>
#endif

#ifdef DARWIN
/* for get_etext and friends */
#include <mach-o/getsect.h>
#endif

#ifdef DJGPP
  /* Apparently necessary for djgpp 2.01.  May cause problems with	*/
  /* other versions.							*/
  typedef long unsigned int caddr_t;
#endif

#ifdef PCR
# include "il/PCR_IL.h"
# include "th/PCR_ThCtl.h"
# include "mm/PCR_MM.h"
#endif

#if !defined(NO_EXECUTE_PERMISSION)
# define OPT_PROT_EXEC PROT_EXEC
#else
# define OPT_PROT_EXEC 0
#endif

#if defined(LINUX) && \
    (defined(USE_PROC_FOR_LIBRARIES) || defined(IA64) || !defined(SMALL_CONFIG))
# define NEED_PROC_MAPS
#endif

#ifdef NEED_PROC_MAPS
/* We need to parse /proc/self/maps, either to find dynamic libraries,	*/
/* and/or to find the register backing store base (IA64).  Do it once	*/
/* here.								*/

#define READ read

/* Repeatedly perform a read call until the buffer is filled or	*/
/* we encounter EOF.						*/
ssize_t GC_repeat_read(int fd, char *buf, size_t count)
{
    ssize_t num_read = 0;
    ssize_t result;
    
    while (num_read < count) {
	result = READ(fd, buf + num_read, count - num_read);
	if (result < 0) return result;
	if (result == 0) break;
	num_read += result;
    }
    return num_read;
}

/* Determine the length of a file by incrementally reading it into a 	*/
/* This would be sily to use on a file supporting lseek, but Linux	*/
/* /proc files usually do not.						*/
size_t GC_get_file_len(int f)
{
    size_t total = 0;
    ssize_t result;
#   define GET_FILE_LEN_BUF_SZ 500
    char buf[GET_FILE_LEN_BUF_SZ];

    do {
	result = read(f, buf, GET_FILE_LEN_BUF_SZ);
	if (result == -1) return 0;
	total += result;
    } while (result > 0);
    return total;
}

size_t GC_get_maps_len(void)
{
    int f = open("/proc/self/maps", O_RDONLY);
    size_t result = GC_get_file_len(f);
    close(f);
    return result;
}

/*
 * Copy the contents of /proc/self/maps to a buffer in our address space.
 * Return the address of the buffer, or zero on failure.
 * This code could be simplified if we could determine its size
 * ahead of time.
 */
char * GC_get_maps(void)
{
    int f;
    int result;
    static char init_buf[1];
    static char *maps_buf = init_buf;
    static size_t maps_buf_sz = 1;
    size_t maps_size, old_maps_size = 0;

    /* The buffer is essentially static, so there must be a single client. */
    GC_ASSERT(I_HOLD_LOCK());

    /* Note that in the presence of threads, the maps file can	*/
    /* essentially shrink asynchronously and unexpectedly as 	*/
    /* threads that we already think of as dead release their	*/
    /* stacks.  And there is no easy way to read the entire	*/
    /* file atomically.  This is arguably a misfeature of the	*/
    /* /proc/.../maps interface.				*/

    /* Since we dont believe the file can grow			*/
    /* asynchronously, it should suffice to first determine	*/
    /* the size (using lseek or read), and then to reread the	*/
    /* file.  If the size is inconsistent we have to retry.	*/
    /* This only matters with threads enabled, and if we use 	*/
    /* this to locate roots (not the default).			*/

    /* Determine the initial size of /proc/self/maps.		*/
    /* Note that lseek doesn't work, at least as of 2.6.15.	*/
#   ifdef THREADS
	maps_size = GC_get_maps_len();
	if (0 == maps_size) return 0;
#   else
	maps_size = 4000;	/* Guess */
#   endif

    /* Read /proc/self/maps, growing maps_buf as necessary.	*/
    /* Note that we may not allocate conventionally, and	*/
    /* thus can't use stdio.					*/
	do {
	    while (maps_size >= maps_buf_sz) {
	      /* Grow only by powers of 2, since we leak "too small" buffers. */
	      while (maps_size >= maps_buf_sz) maps_buf_sz *= 2;
	      maps_buf = GC_scratch_alloc(maps_buf_sz);
#	      ifdef THREADS
	        /* Recompute initial length, since we allocated.	*/
	        /* This can only happen a few times per program		*/
	        /* execution.						*/
	        maps_size = GC_get_maps_len();
	        if (0 == maps_size) return 0;
#	      endif
	      if (maps_buf == 0) return 0;
	    }
	    GC_ASSERT(maps_buf_sz >= maps_size + 1);
	    f = open("/proc/self/maps", O_RDONLY);
	    if (-1 == f) return 0;
#	    ifdef THREADS
	      old_maps_size = maps_size;
#	    endif
	    maps_size = 0;
	    do {
	        result = GC_repeat_read(f, maps_buf, maps_buf_sz-1);
	        if (result <= 0) return 0;
	        maps_size += result;
	    } while (result == maps_buf_sz-1);
	    close(f);
#	    ifdef THREADS
	      if (maps_size > old_maps_size) {
		GC_err_printf("Old maps size = %d, new maps size = %d\n",
			      old_maps_size, maps_size);
		ABORT("Unexpected asynchronous /proc/self/maps growth: "
		      "Unregistered thread?");
	      }
#	    endif
	} while (maps_size >= maps_buf_sz || maps_size < old_maps_size);
		/* In the single-threaded case, the second clause is false.	*/
        maps_buf[maps_size] = '\0';
	
    /* Apply fn to result. */
	return maps_buf;
}

//
//  GC_parse_map_entry parses an entry from /proc/self/maps so we can
//  locate all writable data segments that belong to shared libraries.
//  The format of one of these entries and the fields we care about
//  is as follows:
//  XXXXXXXX-XXXXXXXX r-xp 00000000 30:05 260537     name of mapping...\n
//  ^^^^^^^^ ^^^^^^^^ ^^^^          ^^
//  start    end      prot          maj_dev
//
//  Note that since about august 2003 kernels, the columns no longer have
//  fixed offsets on 64-bit kernels.  Hence we no longer rely on fixed offsets
//  anywhere, which is safer anyway.
//

/*
 * Assign various fields of the first line in buf_ptr to *start, *end,
 * *prot, *maj_dev and *mapping_name.  Mapping_name may be NULL.
 * *prot and *mapping_name are assigned pointers into the original
 * buffer.
 */
char *GC_parse_map_entry(char *buf_ptr, ptr_t *start, ptr_t *end,
                                char **prot, unsigned int *maj_dev,
				char **mapping_name)
{
    char *start_start, *end_start, *maj_dev_start;
    char *p;
    char *endp;

    if (buf_ptr == NULL || *buf_ptr == '\0') {
        return NULL;
    }

    p = buf_ptr;
    while (isspace(*p)) ++p;
    start_start = p;
    GC_ASSERT(isxdigit(*start_start));
    *start = (ptr_t)strtoul(start_start, &endp, 16); p = endp;
    GC_ASSERT(*p=='-');

    ++p;
    end_start = p;
    GC_ASSERT(isxdigit(*end_start));
    *end = (ptr_t)strtoul(end_start, &endp, 16); p = endp;
    GC_ASSERT(isspace(*p));

    while (isspace(*p)) ++p;
    GC_ASSERT(*p == 'r' || *p == '-');
    *prot = p;
    /* Skip past protection field to offset field */
       while (!isspace(*p)) ++p; while (isspace(*p)) ++p;
    GC_ASSERT(isxdigit(*p));
    /* Skip past offset field, which we ignore */
          while (!isspace(*p)) ++p; while (isspace(*p)) ++p;
    maj_dev_start = p;
    GC_ASSERT(isxdigit(*maj_dev_start));
    *maj_dev = strtoul(maj_dev_start, NULL, 16);

    if (mapping_name == 0) {
      while (*p && *p++ != '\n');
    } else {
      while (*p && *p != '\n' && *p != '/' && *p != '[') p++;
      *mapping_name = p;
      while (*p && *p++ != '\n');
    }

    return p;
}

/* Try to read the backing store base from /proc/self/maps.		*/
/* Return the bounds of the writable mapping with a 0 major device,	*/
/* which includes the address passed as data.				*/
/* Return FALSE if there is no such mapping.				*/
GC_bool GC_enclosing_mapping(ptr_t addr, ptr_t *startp, ptr_t *endp)
{
  char *prot;
  ptr_t my_start, my_end;
  unsigned int maj_dev;
  char *maps = GC_get_maps();
  char *buf_ptr = maps;
  
  if (0 == maps) return(FALSE);
  for (;;) {
    buf_ptr = GC_parse_map_entry(buf_ptr, &my_start, &my_end,
		    	         &prot, &maj_dev, 0);

    if (buf_ptr == NULL) return FALSE;
    if (prot[1] == 'w' && maj_dev == 0) {
        if (my_end > addr && my_start <= addr) {
    	  *startp = my_start;
	  *endp = my_end;
	  return TRUE;
	}
    }
  }
  return FALSE;
}

/* Find the text(code) mapping for the library whose name starts with nm. */
GC_bool GC_text_mapping(char *nm, ptr_t *startp, ptr_t *endp)
{
  size_t nm_len = strlen(nm);
  char *prot;
  char *map_path;
  ptr_t my_start, my_end;
  unsigned int maj_dev;
  char *maps = GC_get_maps();
  char *buf_ptr = maps;
  
  if (0 == maps) return(FALSE);
  for (;;) {
    buf_ptr = GC_parse_map_entry(buf_ptr, &my_start, &my_end,
		    	         &prot, &maj_dev, &map_path);

    if (buf_ptr == NULL) return FALSE;
    if (prot[0] == 'r' && prot[1] == '-' && prot[2] == 'x' &&
	strncmp(nm, map_path, nm_len) == 0) {
    	  *startp = my_start;
	  *endp = my_end;
	  return TRUE;
    }
  }
  return FALSE;
}

#ifdef IA64
static ptr_t backing_store_base_from_proc(void)
{
    ptr_t my_start, my_end;
    if (!GC_enclosing_mapping(GC_save_regs_in_stack(), &my_start, &my_end)) {
	if (GC_print_stats) {
	    GC_log_printf("Failed to find backing store base from /proc\n");
	}
	return 0;
    }
    return my_start;
}
#endif

#endif /* NEED_PROC_MAPS */	

#if defined(SEARCH_FOR_DATA_START)
  /* The I386 case can be handled without a search.  The Alpha case	*/
  /* used to be handled differently as well, but the rules changed	*/
  /* for recent Linux versions.  This seems to be the easiest way to	*/
  /* cover all versions.						*/

# if defined(LINUX) || defined(HURD)
    /* Some Linux distributions arrange to define __data_start.  Some	*/
    /* define data_start as a weak symbol.  The latter is technically	*/
    /* broken, since the user program may define data_start, in which	*/
    /* case we lose.  Nonetheless, we try both, prefering __data_start.	*/
    /* We assume gcc-compatible pragmas.	*/
#   pragma weak __data_start
    extern int __data_start[];
#   pragma weak data_start
    extern int data_start[];
# endif /* LINUX */
  extern int _end[];

  ptr_t GC_data_start;