ELF(3E) ELF(3E)
elf - object file access library
cc [flag ...] file ... -lelf [library ...]
#include <libelf.h>
Functions in the ELF access library let a program manipulate ELF
(Executable and Linking Format) object files, archive files, and archive
members. The header file provides type and function declarations for all
library services.
Programs communicate with many of the higher-level routines using an ELF
descriptor. That is, when the program starts working with a file,
elf_begin creates an ELF descriptor through which the program manipulates
the structures and information in the file. These ELF descriptors can be
used both to read and to write files. After the program establishes an
ELF descriptor for a file, it may then obtain section descriptors to
manipulate the sections of the file [see elf_getscn(3E)]. Sections hold
the bulk of an object file's real information, such as text, data, the
symbol table, and so on. A section descriptor ``belongs'' to a
particular ELF descriptor, just as a section belongs to a file. Finally,
data descriptors are available through section descriptors, allowing the
program to manipulate the information associated with a section. A data
descriptor ``belongs'' to a section descriptor.
Descriptors provide private handles to a file and its pieces. In other
words, a data descriptor is associated with one section descriptor, which
is associated with one ELF descriptor, which is associated with one file.
Although descriptors are private, they give access to data that may be
shared. Consider programs that combine input files, using incoming data
to create or update another file. Such a program might get data
descriptors for an input and an output section. It then could update the
output descriptor to reuse the input descriptor's data. That is, the
descriptors are distinct, but they could share the associated data bytes.
This sharing avoids the space overhead for duplicate buffers and the
performance overhead for copying data unnecessarily.
ELF provides a framework in which to define a family of object files,
supporting multiple processors and architectures. An important
distinction among object files is the class, or capacity, of the file.
The 32-bit class supports architectures in which a 32-bit object can
represent addresses, file sizes, etc., as in the following.
Name Purpose
_______________________________________
Elf32_Addr | Unsigned address
Elf32_Half | Unsigned medium integer
Elf32_Off | Unsigned file offset
Page 1
ELF(3E) ELF(3E)
Elf32_Sword Signed large integer
Elf32_Word | Unsigned large integer
unsigned char | Unsigned small integer
______________|________________________
Other classes will be defined as necessary, to support larger (or
smaller) machines. Some library services deal only with data objects for
a specific class, while others are class-independent. To make this
distinction clear, library function names reflect their status, as
described below.
Conceptually, two parallel sets of objects support cross compilation
environments. One set corresponds to file contents, while the other set
corresponds to the native memory image of the program manipulating the
file. Type definitions supplied by the header files work on the native
machine, which may have different data encodings (size, byte order, etc.)
than the target machine. Although native memory objects should be at
least as big as the file objects (to avoid information loss), they may be
bigger if that is more natural for the host machine.
Translation facilities exist to convert between file and memory
representations. Some library routines convert data automatically, while
others leave conversion as the program's responsibility. Either way,
programs that create object files must write file-typed objects to those
files; programs that read object files must take a similar view. See
elf_xlatelf_begin and related functions let
a program deal with the native memory types, converting between memory
objects and their file equivalents automatically when reading or writing
an object file.
Object file versions allow ELF to adapt to new requirements. Threeindependent-versions
can be important to a program. First, an
application program knows about a particular version by virtue of being
compiled with certain header files. Second, the access library similarly
is compiled with header files that control what versions it understands.
Third, an ELF object file holds a value identifying its version,
determined by the ELF version known by the file's creator. Ideally, all
three versions would be the same, but they may differ.
If a program's version is newer than the access library, the program
might use information unknown to the library. Translation routines
might not work properly, leading to undefined behavior. This
condition merits installing a new library.
Page 2
ELF(3E) ELF(3E)
The library's version might be newer than the program's and the
file's. The library understands old versions, thus avoiding
compatibility problems in this case.
Finally, a file's version might be newer than either the program or
the library understands. The program might or might not be able to
process the file properly, depending on whether the file has extra
information and whether that information can be safely ignored.
Again, the safe alternative is to install a new library that
understands the file's version.
To accommodate these differences, a program must use elf_version to pass
its version to the library, thus establishing the working version for the
process. Using this, the library accepts data from and presents data to
the program in the proper representations. When the library reads object
files, it uses each file's version to interpret the data. When writing
files or converting memory types to the file equivalents, the library
uses the program's working version for the file data.
As mentioned above, elf_begin and related routines provide a higher-level
interface to ELF files, performing input and output on behalf of the
application program. These routines assume a program can hold entire
files in memory, without explicitly using temporary files. When reading
a file, the library routines bring the data into memory and perform
subsequent operations on the memory copy. Programs that wish to read or
write large object files with this model must execute on a machine with a
large process virtual address space. If the underlying operating system
limits the number of open files, a program can use elf_cntl to retrieve
all necessary data from the file, allowing the program to close the file
descriptor and reuse it.
Although the elf_begin interfaces are convenient and efficient for many
programs, they might be inappropriate for some. In those cases, an
application may invoke the elf_xlate data translation routines directly.
These routines perform no input or output, leaving that as the
application's responsibility. By assuming a larger share of the job, an
application controls its input and output model.
LIBRARY NAMES
Names associated with the library take several forms.
elf_NAME These class-independent names perform some service, NAME,
for the program.
elf32_NAME Service names with an embedded class, 32 here, indicate
they work only for the designated class of files.
Elf_Type Data types can be class-independent as well,
distinguished by Type.
Page 3
ELF(3E) ELF(3E)
Elf32_Type Class-dependent data types have an embedded class name,
32 here.
ELF_C_CMD Several functions take commands that control their
actions. These values are members of the Elf_Cmd
enumeration; they range from zero through ELF_C_NUM-1.
ELF_F_FLAG Several functions take flags that control library status
and/or actions. Flags are bits that may be combined.
ELF32_FSZ_TYPE These constants give the file sizes in bytes of the basic
ELF types for the 32-bit class of files. See elf_fsize
for more information.
ELF_K_KIND The function elf_kind identifies the KIND of file
associated with an ELF descriptor. These values are
members of the Elf_Kind enumeration; they range from zero
through ELF_K_NUM-1.
ELF_T_TYPE When a service function, such as elf_xlate, deals with
multiple types, names of this form specify the desired
TYPE. Thus, for example, ELF_T_EHDR is directly related
to Elf32_Ehdr. These values are members of the Elf_Type
enumeration; they range from zero through ELF_T_NUM-1.
elf_begin(3E), elf_cntl(3E), elf_end(3E), elf_error(3E), elf_fill(3E),
elf_flag(3E), elf_fsize(3E), elf_getarhdr(3E), elf_getarsym(3E),
elf_getbase(3E), elf_getdata(3E), elf_getehdr(3E), elf_getident(3E),
elf_getphdr(3E), elf_getscn(3E), elf_getshdr(3E), elf_hash(3E),
elf_kind(3E), elf_next(3E), elf_rand(3E), elf_rawfile(3E),
elf_strptr(3E), elf_update(3E), elf_version(3E), elf_xlate(3E), a.out(4)
ar(4)
The chapter ``Object Files'' in UNIX SYSTEM V RELEASE 4 Programmer's
Guide: ANSI C and Programming Support Tools, published by Prentice Hall,
ISBN 0-13-933706-7.
The standard SVR4 elf man page mentions processor-dependent header files
with names of the form <sys/elf_NAME<b>.h> where NAME is a processor name in
a table, such as M32. There are no such header files in IRIX.
A 32-bit application can construct 64-bit binaries using functions
defined on the above-mentioned man pages. However the man pages and
certain books published on SVR4 specifically document fields in the
Elf_Data structure as long. This restricts the generated object files
(even 64-bit object files) to have 32-bit values at most when constructed
by a 32-bit application. The resulting object file is always 64-bit clean
(the documented 32-bit fields are only too small during construction of
the object file, not too small in the object file itself). This matters,
for example, in the Fortran compiler where the bss section might need to
be greater than 32-bits. See /usr/include/libelf.h for the definition(s)
Page 4
ELF(3E) ELF(3E)
of Elf_Data.
To overcome this difficulty, the generating application should be
compiled with the preprocessor flag _LIBELF_XTND_64 defined to all
compilation units and should link to -lelf_xtnd and -ldwarf_xtnd instead
of -lelf and -ldwarf. If this is done, the definition of the fields in
the Elf_Data changes to 64 bits. This change permeates the definition of
an Elf * (even though the definition of Elf* is opaque to the
application), requiring the application build to be completely consistent
and define _LIBELF_XTND_64 everywhere in the application build and to
link with -lelf_xtnd and -ldwarf_xtnd instead of -lelf and -ldwarf.
It is just possible that by compiling with _LIBELF_XTND_64 visible in
some compilation units but not others an application could manage to pass
an Elf * or other libelf structure created without _LIBELF_XTND_64 into a
libelf function call compiled with _LIBELF_XTND_64. Or vice-versa. The
result will surely be chaos.
_LIBELF_XTND_64 is irrelevant to any 64-bit application and the
-lelf_xtnd and -ldwarf_xtnd are not needed, since 64-bit applications can
build true 64-bit object files without defining _LIBELF_XTND_64.
Applications which only read 64-bit object files need not use
_LIBELF_XTND_64 since the section headers, program headers, and data of
the file have 64-bit fields without _LIBELF_XTND_64 being defined.
PPPPaaaaggggeeee 5555 [ Back ]
|
