guile/libguile/loader.c

/* Copyright 2001,2009-2015,2017-2018,2021
     Free Software Foundation, Inc.

   This file is part of Guile.

   Guile is free software: you can redistribute it and/or modify it
   under the terms of the GNU Lesser General Public License as published
   by the Free Software Foundation, either version 3 of the License, or
   (at your option) any later version.

   Guile is distributed in the hope that it will be useful, but WITHOUT
   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public
   License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with Guile.  If not, see
   <https://www.gnu.org/licenses/>.  */

#if HAVE_CONFIG_H
#  include <config.h>
#endif

#include <alignof.h>
#include <assert.h>
#include <byteswap.h>
#include <errno.h>
#include <fcntl.h>
#include <full-read.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
#include <verify.h>

#ifdef HAVE_SYS_MMAN_H
#include <sys/mman.h>
#endif

#include "boolean.h"
#include "bytevectors.h"
#include "elf.h"
#include "eval.h"
#include "extensions.h"
#include "gsubr.h"
#include "list.h"
#include "pairs.h"
#include "programs.h"
#include "strings.h"
#include "threads.h"
#include "version.h"

#include "loader.h"

#ifdef __MINGW32__
#include "posix-w32.h"
#endif

/* This file contains the loader for Guile's on-disk format: ELF with
   some custom tags in the dynamic segment.  */

#if SIZEOF_UINTPTR_T == 4
#define Elf_Half Elf32_Half
#define Elf_Word Elf32_Word
#define Elf_Ehdr Elf32_Ehdr
#define ELFCLASS ELFCLASS32
#define Elf_Phdr Elf32_Phdr
#define Elf_Dyn Elf32_Dyn
#elif SIZEOF_UINTPTR_T == 8
#define Elf_Half Elf64_Half
#define Elf_Word Elf64_Word
#define Elf_Ehdr Elf64_Ehdr
#define ELFCLASS ELFCLASS64
#define Elf_Phdr Elf64_Phdr
#define Elf_Dyn Elf64_Dyn
#else
#error
#endif

#define DT_LOGUILE          0x37146000  /* Start of Guile-specific */
#define DT_GUILE_GC_ROOT    0x37146000  /* Offset of GC roots */
#define DT_GUILE_GC_ROOT_SZ 0x37146001  /* Size in machine words of GC
                                           roots */
#define DT_GUILE_ENTRY      0x37146002  /* Address of entry thunk */
#define DT_GUILE_VM_VERSION 0x37146003  /* Bytecode version */
#define DT_GUILE_FRAME_MAPS 0x37146004  /* Frame maps */
#define DT_HIGUILE          0x37146fff  /* End of Guile-specific */

#ifdef WORDS_BIGENDIAN
#define ELFDATA ELFDATA2MSB
#else
#define ELFDATA ELFDATA2LSB
#endif

/* The page size.  */
static size_t page_size;

static void register_elf (char *data, size_t len, char *frame_maps);

enum bytecode_kind
  {
    BYTECODE_KIND_NONE,
    BYTECODE_KIND_GUILE_3_0
  };

static SCM
pointer_to_procedure (enum bytecode_kind bytecode_kind, char *ptr)
{
  switch (bytecode_kind)
    {
    case BYTECODE_KIND_GUILE_3_0:
      {
        return scm_i_make_program ((uint32_t *) ptr);
      }
    case BYTECODE_KIND_NONE:
    default:
      abort ();
    }
}

static const char*
check_elf_header (const Elf_Ehdr *header)
{
  if (!(header->e_ident[EI_MAG0] == ELFMAG0
        && header->e_ident[EI_MAG1] == ELFMAG1
        && header->e_ident[EI_MAG2] == ELFMAG2
        && header->e_ident[EI_MAG3] == ELFMAG3))
    return "not an ELF file";

  if (header->e_ident[EI_CLASS] != ELFCLASS)
    return "ELF file does not have native word size";

  if (header->e_ident[EI_DATA] != ELFDATA)
    return "ELF file does not have native byte order";

  if (header->e_ident[EI_VERSION] != EV_CURRENT)
    return "bad ELF version";

  if (header->e_ident[EI_OSABI] != ELFOSABI_STANDALONE)
    return "unexpected OS ABI";

  if (header->e_ident[EI_ABIVERSION] != 0)
    return "unexpected ABI version";

  if (header->e_type != ET_DYN)
    return "unexpected ELF type";

  if (header->e_machine != EM_NONE)
    return "unexpected machine";

  if (header->e_version != EV_CURRENT)
    return "unexpected ELF version";

  if (header->e_ehsize != sizeof *header)
    return "unexpected header size";

  if (header->e_phentsize != sizeof (Elf_Phdr))
    return "unexpected program header size";

  return NULL;
}

#define IS_ALIGNED(offset, alignment) \
  (!((offset) & ((alignment) - 1)))
#define ALIGN(offset, alignment) \
  ((offset + (alignment - 1)) & ~(alignment - 1))

/* Return the alignment required by the ELF at DATA, of LEN bytes.  */
static size_t
elf_alignment (const char *data, size_t len)
{
  Elf_Ehdr *header;
  int i;
  size_t alignment = 8;

  if (len < sizeof(Elf_Ehdr))
    return alignment;
  header = (Elf_Ehdr *) data;
  if (header->e_phoff + header->e_phnum * header->e_phentsize >= len)
    return alignment;
  for (i = 0; i < header->e_phnum; i++)
    {
      Elf_Phdr *phdr;
      const char *phdr_addr = data + header->e_phoff + i * header->e_phentsize;

      if (!IS_ALIGNED ((uintptr_t) phdr_addr, alignof_type (Elf_Phdr)))
        return alignment;
      phdr = (Elf_Phdr *) phdr_addr;

      if (phdr->p_align & (phdr->p_align - 1))
        return alignment;

      if (phdr->p_align > alignment)
        alignment = phdr->p_align;
    }

  return alignment;
}

/* This function leaks the memory that it allocates.  */
static char*
alloc_aligned (size_t len, unsigned alignment)
{
  char *ret;

  if (alignment == 8)
    {
      /* FIXME: Assert that we actually have an 8-byte-aligned malloc.  */
      ret = malloc (len);
    }
#if defined(HAVE_SYS_MMAN_H) && defined(HAVE_MAP_ANONYMOUS)
  else if (alignment == page_size)
    {
      ret = mmap (NULL, len, PROT_READ | PROT_WRITE,
                  MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
      if (ret == MAP_FAILED)
        scm_syserror ("load-thunk-from-memory");
    }
#endif
  else
    {
      if (len + alignment < len)
        abort ();

      ret = malloc (len + alignment - 1);
      if (!ret)
        abort ();
      ret = (char *) ALIGN ((uintptr_t) ret, (uintptr_t) alignment);
    }

  return ret;
}

static char*
copy_and_align_elf_data (const char *data, size_t len)
{
  size_t alignment;
  char *copy;

  alignment = elf_alignment (data, len);
  copy = alloc_aligned (len, alignment);
  memcpy(copy, data, len);

  return copy;
}

#ifdef HAVE_SYS_MMAN_H
static int
segment_flags_to_prot (Elf_Word flags)
{
  int prot = 0;

  if (flags & PF_X)
    prot |= PROT_EXEC;
  if (flags & PF_W)
    prot |= PROT_WRITE;
  if (flags & PF_R)
    prot |= PROT_READ;

  return prot;
}
#endif

static char*
process_dynamic_segment (char *base, Elf_Phdr *dyn_phdr,
                         SCM *init_out, SCM *entry_out, char **frame_maps_out)
{
  char *dyn_addr = base + dyn_phdr->p_vaddr;
  Elf_Dyn *dyn = (Elf_Dyn *) dyn_addr;
  size_t i, dyn_size = dyn_phdr->p_memsz / sizeof (Elf_Dyn);
  char *init = 0, *gc_root = 0, *entry = 0, *frame_maps = 0;
  ptrdiff_t gc_root_size = 0;
  enum bytecode_kind bytecode_kind = BYTECODE_KIND_NONE;

  for (i = 0; i < dyn_size; i++)
    {
      if (dyn[i].d_tag == DT_NULL)
        break;

      switch (dyn[i].d_tag)
        {
        case DT_INIT:
          if (init)
            return "duplicate DT_INIT";
          init = base + dyn[i].d_un.d_val;
          break;
        case DT_GUILE_GC_ROOT:
          if (gc_root)
            return "duplicate DT_GUILE_GC_ROOT";
          gc_root = base + dyn[i].d_un.d_val;
          break;
        case DT_GUILE_GC_ROOT_SZ:
          if (gc_root_size)
            return "duplicate DT_GUILE_GC_ROOT_SZ";
          gc_root_size = dyn[i].d_un.d_val;
          break;
        case DT_GUILE_ENTRY:
          if (entry)
            return "duplicate DT_GUILE_ENTRY";
          entry = base + dyn[i].d_un.d_val;
          break;
        case DT_GUILE_VM_VERSION:
          if (bytecode_kind != BYTECODE_KIND_NONE)
            return "duplicate DT_GUILE_VM_VERSION";
          {
            uint16_t major = dyn[i].d_un.d_val >> 16;
            uint16_t minor = dyn[i].d_un.d_val & 0xffff;
            switch (major)
              {
              case 0x0300:
                bytecode_kind = BYTECODE_KIND_GUILE_3_0;
                if (minor < SCM_OBJCODE_MINIMUM_MINOR_VERSION)
                  return "incompatible bytecode version";
                if (minor > SCM_OBJCODE_MINOR_VERSION)
                  return "incompatible bytecode version";
                break;
              default:
                return "incompatible bytecode kind";
              }
            break;
          }
        case DT_GUILE_FRAME_MAPS:
          if (frame_maps)
            return "duplicate DT_GUILE_FRAME_MAPS";
          frame_maps = base + dyn[i].d_un.d_val;
          break;
        }
    }

  if (!entry)
    return "missing DT_GUILE_ENTRY";

  switch (bytecode_kind)
    {
    case BYTECODE_KIND_GUILE_3_0:
      if ((uintptr_t) init % 4)
        return "unaligned DT_INIT";
      if ((uintptr_t) entry % 4)
        return "unaligned DT_GUILE_ENTRY";
      break;
    case BYTECODE_KIND_NONE:
    default:
      return "missing DT_GUILE_VM_VERSION";
    }

  if (gc_root)
    GC_add_roots (gc_root, gc_root + gc_root_size);

  *init_out = init ? pointer_to_procedure (bytecode_kind, init) : SCM_BOOL_F;
  *entry_out = pointer_to_procedure (bytecode_kind, entry);
  *frame_maps_out = frame_maps;

  return NULL;
}

#define ABORT(msg) do { err_msg = msg; errno = 0; goto cleanup; } while (0)

static SCM
load_thunk_from_memory (char *data, size_t len, int is_read_only)
#define FUNC_NAME "load-thunk-from-memory"
{
  Elf_Ehdr *header;
  Elf_Phdr *ph;
  const char *err_msg = 0;
  size_t n, alignment = 8;
  int i;
  int dynamic_segment = -1;
  SCM init = SCM_BOOL_F, entry = SCM_BOOL_F;
  char *frame_maps = 0;

  errno = 0;

  if (len < sizeof *header)
    ABORT ("object file too small");

  header = (Elf_Ehdr*) data;
  
  if ((err_msg = check_elf_header (header)))
    {
      errno = 0;				  /* not an OS error */
      goto cleanup;
    }

  if (header->e_phnum == 0)
    ABORT ("no loadable segments");
  n = header->e_phnum;

  if (len < header->e_phoff + n * sizeof (Elf_Phdr))
    ABORT ("object file too small");

  ph = (Elf_Phdr*) (data + header->e_phoff);

  /* Check that the segment table is sane.  */
  for (i = 0; i < n; i++)
    {
      if (ph[i].p_filesz != ph[i].p_memsz)
        ABORT ("expected p_filesz == p_memsz");

      if (!ph[i].p_flags)
        ABORT ("expected nonzero segment flags");

      if (ph[i].p_align < alignment)
        {
          if (ph[i].p_align % alignment)
            ABORT ("expected new alignment to be multiple of old");
          alignment = ph[i].p_align;
        }

      if (ph[i].p_type == PT_DYNAMIC)
        {
          if (dynamic_segment >= 0)
            ABORT ("expected only one PT_DYNAMIC segment");
          dynamic_segment = i;
          continue;
        }

      if (ph[i].p_type != PT_LOAD)
        ABORT ("unknown segment type");

      if (i == 0)
        {
          if (ph[i].p_vaddr != 0)
            ABORT ("first loadable vaddr is not 0");
        }
      else
        {
          if (ph[i].p_vaddr < ph[i-1].p_vaddr + ph[i-1].p_memsz)
            ABORT ("overlapping segments");

          if (ph[i].p_offset + ph[i].p_filesz > len)
            ABORT ("segment beyond end of byte array");
        }
    }

  if (dynamic_segment < 0)
    ABORT ("no PT_DYNAMIC segment");

  /* The ELF images that Guile currently emits have segments that are
     aligned on 64 KB boundaries, which might be larger than the actual
     page size (usually 4 KB).  However Guile doesn't actually use the
     absolute addresses at all.  All Guile needs is for the loaded image
     to be able to make the data section writable (for the mmap path),
     and for that the segment just needs to be page-aligned, and a page
     is always bigger than Guile's minimum alignment.  Since we know
     (for the mmap path) that the base _is_ page-aligned, we proceed
     ahead even if the image alignment is greater than the page
     size.  */
  if (!IS_ALIGNED ((uintptr_t) data, alignment)
      && !IS_ALIGNED (alignment, page_size))
    ABORT ("incorrectly aligned base");

  /* Allow writes to writable pages.  */
  if (is_read_only)
    {
#ifdef HAVE_SYS_MMAN_H
      for (i = 0; i < n; i++)
        {
          if (ph[i].p_type != PT_LOAD)
            continue;
          if (ph[i].p_flags == PF_R)
            continue;
          if (ph[i].p_align < page_size)
            continue;

          if (mprotect (data + ph[i].p_vaddr,
                        ph[i].p_memsz,
                        segment_flags_to_prot (ph[i].p_flags)))
            goto cleanup;
        }
#else
      ABORT ("expected writable pages");
#endif
    }

  if ((err_msg = process_dynamic_segment (data, &ph[dynamic_segment],
                                          &init, &entry, &frame_maps)))
    {
      errno = 0;				  /* not an OS error */
      goto cleanup;
    }

  if (scm_is_true (init))
    scm_call_0 (init);

  register_elf (data, len, frame_maps);

  /* Finally!  Return the thunk.  */
  return entry;

 cleanup:
  {
    if (errno)
      SCM_SYSERROR;
    scm_misc_error (FUNC_NAME, err_msg ? err_msg : "error loading ELF file",
                    SCM_EOL);
  }
}
#undef FUNC_NAME

static char*
map_file_contents (int fd, size_t len, int *is_read_only)
#define FUNC_NAME "load-thunk-from-file"
{
  char *data;

#ifdef HAVE_SYS_MMAN_H
  data = mmap (NULL, len, PROT_READ, MAP_PRIVATE, fd, 0);
  if (data == MAP_FAILED)
    SCM_SYSERROR;
  *is_read_only = 1;
#else
  if (lseek (fd, 0, SEEK_SET) < 0)
    {
      int errno_save = errno;
      (void) close (fd);
      errno = errno_save;
      SCM_SYSERROR;
    }

  /* Given that we are using the read fallback, optimistically assume
     that the .go files were made with 8-byte alignment.
     alignment.  */
  data = malloc (len);
  if (!data)
    {
      (void) close (fd);
      scm_misc_error (FUNC_NAME, "failed to allocate ~A bytes",
                      scm_list_1 (scm_from_size_t (len)));
    }

  if (full_read (fd, data, len) != len)
    {
      int errno_save = errno;
      (void) close (fd);
      errno = errno_save;
      if (errno)
        SCM_SYSERROR;
      scm_misc_error (FUNC_NAME, "short read while loading objcode",
                      SCM_EOL);
    }

  /* If our optimism failed, fall back.  */
  {
    unsigned alignment = elf_alignment (data, len);

    if (alignment != 8)
      {
        char *copy = copy_and_align_elf_data (data, len);
        free (data);
        data = copy;
      }
  }

  *is_read_only = 0;
#endif

  return data;
}
#undef FUNC_NAME

SCM_DEFINE (scm_load_thunk_from_file, "load-thunk-from-file", 1, 0, 0,
	    (SCM filename),
	    "")
#define FUNC_NAME s_scm_load_thunk_from_file
{
  char *c_filename;
  int fd, is_read_only;
  off_t end;
  char *data;

  SCM_VALIDATE_STRING (1, filename);

  c_filename = scm_to_locale_string (filename);
  fd = open (c_filename, O_RDONLY | O_BINARY | O_CLOEXEC);
  free (c_filename);
  if (fd < 0) SCM_SYSERROR;

  end = lseek (fd, 0, SEEK_END);
  if (end < 0)
    SCM_SYSERROR;

  data = map_file_contents (fd, end, &is_read_only);

  (void) close (fd);

  return load_thunk_from_memory (data, end, is_read_only);
}
#undef FUNC_NAME

SCM_DEFINE (scm_load_thunk_from_memory, "load-thunk-from-memory", 1, 0, 0,
	    (SCM bv),
	    "")
#define FUNC_NAME s_scm_load_thunk_from_memory
{
  char *data;
  size_t len;

  SCM_VALIDATE_BYTEVECTOR (1, bv);

  data = (char *) SCM_BYTEVECTOR_CONTENTS (bv);
  len = SCM_BYTEVECTOR_LENGTH (bv);

  /* Copy data in order to align it, to trace its GC roots and
     writable sections, and to keep it in memory.  */

  data = copy_and_align_elf_data (data, len);

  return load_thunk_from_memory (data, len, 0);
}
#undef FUNC_NAME



struct mapped_elf_image
{
  char *start;
  char *end;
  char *frame_maps;
};

static struct mapped_elf_image *mapped_elf_images = NULL;
static size_t mapped_elf_images_count = 0;
static size_t mapped_elf_images_allocated = 0;

static size_t
find_mapped_elf_insertion_index (char *ptr)
{
  /* "mapped_elf_images_count" must never be dereferenced.  */
  size_t start = 0, end = mapped_elf_images_count;

  while (start < end)
    {
      size_t n = start + (end - start) / 2;

      if (ptr < mapped_elf_images[n].end)
        end = n;
      else
        start = n + 1;
    }

  return start;
}

static void
register_elf (char *data, size_t len, char *frame_maps)
{
  scm_i_pthread_mutex_lock (&scm_i_misc_mutex);
  {
    /* My kingdom for a generic growable sorted vector library.  */
    if (mapped_elf_images_count == mapped_elf_images_allocated)
      {
        struct mapped_elf_image *prev;
        size_t n;

        if (mapped_elf_images_allocated)
          mapped_elf_images_allocated *= 2;
        else
          mapped_elf_images_allocated = 16;

        prev = mapped_elf_images;
        mapped_elf_images =
          scm_gc_malloc_pointerless (sizeof (*mapped_elf_images)
                                     * mapped_elf_images_allocated,
                                     "mapped elf images");

        for (n = 0; n < mapped_elf_images_count; n++)
          {
            mapped_elf_images[n].start = prev[n].start;
            mapped_elf_images[n].end = prev[n].end;
            mapped_elf_images[n].frame_maps = prev[n].frame_maps;
          }
      }

    {
      size_t end;
      size_t n = find_mapped_elf_insertion_index (data);

      for (end = mapped_elf_images_count; n < end; end--)
        {
          const struct mapped_elf_image *prev = &mapped_elf_images[end - 1];
          mapped_elf_images[end].start = prev->start;
          mapped_elf_images[end].end = prev->end;
          mapped_elf_images[end].frame_maps = prev->frame_maps;
        }
      mapped_elf_images_count++;

      mapped_elf_images[n].start = data;
      mapped_elf_images[n].end = data + len;
      mapped_elf_images[n].frame_maps = frame_maps;
    }
  }
  scm_i_pthread_mutex_unlock (&scm_i_misc_mutex);
}

static struct mapped_elf_image *
find_mapped_elf_image_unlocked (char *ptr)
{
  size_t n = find_mapped_elf_insertion_index ((char *) ptr);

  if (n < mapped_elf_images_count
      && mapped_elf_images[n].start <= ptr
      && ptr < mapped_elf_images[n].end)
    return &mapped_elf_images[n];

  return NULL;
}

static int
find_mapped_elf_image (char *ptr, struct mapped_elf_image *image)
{
  int result;

  scm_i_pthread_mutex_lock (&scm_i_misc_mutex);
  {
    struct mapped_elf_image *img = find_mapped_elf_image_unlocked (ptr);
    if (img)
      {
        memcpy (image, img, sizeof (*image));
        result = 1;
      }
    else
      result = 0;
  }
  scm_i_pthread_mutex_unlock (&scm_i_misc_mutex);

  return result;
}

static SCM
scm_find_mapped_elf_image (SCM ip)
{
  struct mapped_elf_image image;

  if (find_mapped_elf_image ((char *) scm_to_uintptr_t (ip), &image))
    {
      signed char *data = (signed char *) image.start;
      size_t len = image.end - image.start;

      return scm_c_take_gc_bytevector (data, len, SCM_BOOL_F);
    }

  return SCM_BOOL_F;
}

static SCM
scm_all_mapped_elf_images (void)
{
  SCM result = SCM_EOL;

  scm_i_pthread_mutex_lock (&scm_i_misc_mutex);
  {
    size_t n;
    for (n = 0; n < mapped_elf_images_count; n++)
      {
        signed char *data = (signed char *) mapped_elf_images[n].start;
        size_t len = mapped_elf_images[n].end - mapped_elf_images[n].start;
        result = scm_cons (scm_c_take_gc_bytevector (data, len, SCM_BOOL_F),
                           result);
      }
  }
  scm_i_pthread_mutex_unlock (&scm_i_misc_mutex);

  return result;
}

struct frame_map_prefix
{
  uint32_t text_offset;
  uint32_t maps_offset;
};

struct frame_map_header
{
  uint32_t addr;
  uint32_t map_offset;
};

verify (sizeof (struct frame_map_prefix) == 8);
verify (sizeof (struct frame_map_header) == 8);

const uint8_t *
scm_find_slot_map_unlocked (const uint32_t *ip)
{
  struct mapped_elf_image *image;
  char *base;
  struct frame_map_prefix *prefix;
  struct frame_map_header *headers;
  uintptr_t addr = (uintptr_t) ip;
  size_t start, end;

  image = find_mapped_elf_image_unlocked ((char *) ip);
  if (!image || !image->frame_maps)
    return NULL;

  base = image->frame_maps;
  prefix = (struct frame_map_prefix *) base;
  headers = (struct frame_map_header *) (base + sizeof (*prefix));

  if (addr < ((uintptr_t) image->start) + prefix->text_offset)
    return NULL;
  addr -= ((uintptr_t) image->start) + prefix->text_offset;

  start = 0;
  end = (prefix->maps_offset - sizeof (*prefix)) / sizeof (*headers);

  if (end == 0 || addr > headers[end - 1].addr)
    return NULL;

  while (start < end)
    {
      size_t n = start + (end - start) / 2;

      if (addr == headers[n].addr)
        return (const uint8_t*) (base + headers[n].map_offset);
      else if (addr < headers[n].addr)
        end = n;
      else
        start = n + 1;
    }

  return NULL;
}


void
scm_bootstrap_loader (void)
{
#ifndef __MINGW32__
  page_size = getpagesize ();
#else
  page_size = getpagesize_w32 ();
#endif
  /* page_size should be a power of two.  */
  if (page_size & (page_size - 1))
    abort ();

  scm_c_register_extension ("libguile-" SCM_EFFECTIVE_VERSION,
                            "scm_init_loader",
                            (scm_t_extension_init_func)scm_init_loader, NULL);
}

void
scm_init_loader (void)
{
#ifndef SCM_MAGIC_SNARFER
#include "loader.x"
#endif

  scm_c_define_gsubr ("find-mapped-elf-image", 1, 0, 0,
                      (scm_t_subr) scm_find_mapped_elf_image);
  scm_c_define_gsubr ("all-mapped-elf-images", 0, 0, 0,
                      (scm_t_subr) scm_all_mapped_elf_images);
}