xref: /titanic_51/usr/src/cmd/sgs/libelf/common/README.LFS (revision bebb829deac32e16136b725d421267d3dceb6cfd)

#
# CDDL HEADER START
#
# The contents of this file are subject to the terms of the
# Common Development and Distribution License (the "License").
# You may not use this file except in compliance with the License.
#
# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
# or http://www.opensolaris.org/os/licensing.
# See the License for the specific language governing permissions
# and limitations under the License.
#
# When distributing Covered Code, include this CDDL HEADER in each
# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
# If applicable, add the following below this CDDL HEADER, with the
# fields enclosed by brackets "[]" replaced with your own identifying
# information: Portions Copyright [yyyy] [name of copyright owner]
#
# CDDL HEADER END
#

#
# Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
# Use is subject to license terms.
#


Why 32-bit libelf is not Large File Aware
-----------------------------------------

The ELF format uses unsigned 32-bit integers for offsets, so the
theoretical limit on a 32-bit ELF object is 4GB. However, libelf
imposes a 2GB limit on the objects it can create. The Solaris
link-editor and related tools are all based on libelf, so the
32-bit version of the link-editor also has a 2GB limit, despite
the theoretical limit of 4GB.

Large file support (LFS) is a half step between the 32 and 64-bit
worlds, in which an otherwise 32-bit limited process is allowed to
read and write data to a file that can be larger than 2GB (the extent
of a signed 32-bit integer, as represented by the system type off_t).
LFS is useful if the program only needs to access a small subset of
the file data at any given time (e.g. /usr/bin/cat). It is less useful
if the program needs to access a large amount of data at once --- having
been freed from the file limit, the program will simply hit the virtual
memory limit (4GB).

In particular, the link-editor generally requires twice as much
memory as the size of the output object, half to hold the input
objects, and half to hold the result. This means that a 32-bit
link-editor process will hit the 2GB file size limit and the 4GB
address space limit at roughly the same time. As a result, a
large file aware 32-bit version of libelf has no significant value.
Despite this, the question of what it would take to make libelf
large file aware comes up from time to time.

The first step would be to provide alternative versions of
all public data structures that involve the off_t data type.
These structs, found in /usr/include/libelf.h, are:

	/*
	 * Archive member header
	 */
	typedef struct {
		char		*ar_name;
		time_t		ar_date;
		uid_t		ar_uid;
		gid_t 		ar_gid;
		mode_t		ar_mode;
		off_t		ar_size;
		char 		*ar_rawname;
	} Elf_Arhdr;


	/*
	 * Data descriptor
	 */
	typedef struct {
		Elf_Void	*d_buf;
		Elf_Type	d_type;
		size_t		d_size;
		off_t		d_off;		/* offset into section */
		size_t		d_align;	/* alignment in section */
		unsigned	d_version;	/* elf version */
	} Elf_Data;

As off_t is a signed type, these alternative versions would have to use
an off64_t type instead.

In addition to providing alternative large file aware Elf_Arhdr and
Elf_Data types, it would be necessary to implement large file aware
versions of the public functions that use them, also found in
/usr/include/libelf.h:

	/*
	 * Function declarations
	 */
	unsigned  elf_flagdata(Elf_Data *, Elf_Cmd, unsigned);
	Elf_Arhdr *elf_getarhdr(Elf *);
	off_t	  elf_getbase(Elf *);
	Elf_Data  *elf_getdata(Elf_Scn *, Elf_Data *);
	Elf_Data  *elf_newdata(Elf_Scn *);
	Elf_Data  *elf_rawdata(Elf_Scn *, Elf_Data *);
	off_t	  elf_update(Elf *, Elf_Cmd);
	Elf_Data  *elf32_xlatetof(Elf_Data *, const Elf_Data *, unsigned);
	Elf_Data  *elf32_xlatetom(Elf_Data *, const Elf_Data *, unsigned);
	Elf_Data  *elf64_xlatetof(Elf_Data *, const Elf_Data *, unsigned);
	Elf_Data  *elf64_xlatetom(Elf_Data *, const Elf_Data *, unsigned);

It is important to note that these new versions cannot replace the
original definitions. Those must continue to be available to support
non-large-file-aware programs. These new types and functions would be in
addition to the pre-existing versions.

When you make code like this large file aware, it is necessary to undertake
a careful analysis of the code to ensure that all the surrounding code uses
variable types large enough to handle the increased range. Hence, this work
is more complicated than simply supplying variants that use a bigger
off_t and rebuilding --- that is just the first step.

There are two standard preprocessor definitions used to control
large file support:

	_LARGEFILE64_SOURCE
	_FILE_OFFSET_BITS

These preprocessor definitions would be used to determine whether
a given program linked against libelf would see the regular, or
the large file aware versions of the above types and routines.
This is the same approach used in other large file capable software,
such as libc.

Finally, all the applications that rely on libelf would need to be made
large file aware. As with libelf itself, there is more to such an effort
than recompiling with preprocessor macros set. The code in these
applications would need to be examined carefully. Some of these programs
are very old, and were not originally written with such type portability
in mind. Such code can be difficult to transition.

To work around the 2GB limit in 32-bit libelf:

    - The fundamental limits of a 32-bit address space mean
      that a program this large should be 64-bit. Only a 64-bit
      address space has enough room for that much code, plus the
      stack and heap needed to do useful work with it.

    - The 64-bit version of libelf is also able to process
      32-bit objects, and does not have a 2GB file size limit.
      Therefore, the 64-bit link-editor can be used to build a 32-bit
      executable which is >2GB. The resulting program will consume over
      half the available address space just to start running. However,
      there may be enough address space left for it to do useful work.

      Note that the 32-bit limit for sharable objects remains at
      2GB --- imposed by the runtime linker, which is also not large
      file aware.