ELF Concepts

ELF objects have an associated “ELF class” which denotes the natural machine word size for the architecture the object is associated with. Objects for 32 bit architectures have an ELF class of ELFCLASS32. Objects for 64 bit architectures have an ELF class of ELFCLASS64.

ELF objects also have an associated “endianness” which denotes the endianness of the machine architecture associated with the object. This may be ELFDATA2LSB for little-endian architectures and ELFDATA2MSB for big-endian architectures.

ELF objects are also associated with an API version number. This version number determines the layout of the individual components of an ELF file and the semantics associated with these.

Data Representation And Translation

An application would work with ELF data in its “native” representation, i.e., using the native byteorder and alignment mandated by the processor the application is running on. The “file” representation of the same data could use a different byte ordering and follow different constraints on object alignment than these native constraints.

Accordingly, the ELF(3) library offers translation facilities (elf32_xlatetof(3), elf32_xlatetom(3), elf64_xlatetof(3) and elf64_xlatetom(3)) to and from these representations and also provides higher-level APIs that retrieve and store data from the ELF object in a transparent manner.

Library Working Version

• The ELF version that the application was compiled against. This version determines the ABI expected by the application.
• The ELF version of the ELF object being manipulated by the application through the ELF library.
• The ELF version (or set of versions) supported by the ELF library itself.

In order to facilitate working with ELF objects of differing versions, the ELF library requires the application to call the elf_version() function before invoking many of its operations, in order to inform the library of the application's desired working version.

In the current implementation, all three versions have to be EV_CURRENT.

Namespace use

elf_

Used for class-independent functions.

elf32_

Used for functions working with 32 bit ELF objects.

elf64_

Used for functions working with 64 bit ELF objects.

Elf_

Used for class-independent data types.

ELF_C_

Used for command values used in a few functions. These symbols are defined as members of the Elf_Cmd enumeration.

ELF_E_

Used for error numbers.

ELF_F_

Used for flags.

ELF_K_

These constants define the kind of file associated with an ELF descriptor. See elf_kind(3). The symbols are defined by the Elf_Kind enumeration.

ELF_T_

These values are defined by the Elf_Type enumeration, and denote the types of ELF data structures that can be present in an ELF object.

In addition, the library uses symbols with prefixes _ELF and _libelf for its internal use.

Descriptors

Elf

An Elf descriptor represents an ELF object or an ar(1) archive. It is allocated using one of the elf_begin() or elf_memory() functions. An Elf descriptor can be used to read and write data to an ELF file. An Elf descriptor can be associated with zero or more Elf_Scn section descriptors.

Given an ELF descriptor, the application may retrieve the ELF object's class-dependent “Executable Header” structures using the elf32_getehdr() or elf64_getehdr() functions. A new Ehdr structure may be allocated using the elf64_newehdr() or elf64_newehdr() functions.

The “Program Header Table” associated with an ELF descriptor may be allocated using the elf32_getphdr() or elf64_getphdr() functions. A new program header table may be allocated or an existing table resized using the elf32_newphdr() or elf64_newphdr() functions.

The Elf structure is opaque and has no members visible to the application.

Elf_Data

An Elf_Data data structure describes an individual chunk of a ELF file as represented in memory. It has the following application-visible members:

uint64_t d_align

The in-file alignment of the data buffer within its containing ELF section. This value must be non-zero and a power of two.

void *d_buf

A pointer to data in memory.

uint64_t d_off

The offset within the containing section where this descriptor's data would be placed. This field will be computed by the library unless the application requests full control of the ELF object's layout.

uint64_t d_size

The number of bytes of data in this descriptor.

Elf_Type d_type

The ELF type (see below) of the data in this descriptor.

unsigned int d_version

The operating version for the data in this buffer.

Elf_Data descriptors are usually associated with Elf_Scn descriptors. Existing data descriptors associated with an ELF section may be structures are retrieved using the elf_getdata() and elf_rawdata() functions. The elf_newdata() function may be used to attach new data descriptors to an ELF section.

Elf_Scn

Elf_Scn descriptors represent a section in an ELF object.

They are retrieved using the elf_getscn() function. An application may iterate through the existing sections of an ELF object using the elf_nextscn() function. New sections may be allocated using the elf_newscn() function.

The Elf_Scn descriptor is opaque and contains no application modifiable fields.

Supported Elf Types

ELF_T_ADDR

Machine addresses.

ELF_T_BYTE

Byte data. The library will not attempt to translate byte data.

ELF_T_CAP

Software and hardware capability records.

ELF_T_DYN

Records used in a section of type SHT_DYNAMIC.

ELF_T_EHDR

ELF executable header.

ELF_T_GNUHASH

GNU-style hash tables.

ELF_T_HALF

16-bit unsigned words.

ELF_T_LWORD

64 bit unsigned words.

ELF_T_MOVE

ELF Move records.

ELF_T_NOTE

ELF Note structures.

ELF_T_OFF

File offsets.

ELF_T_PHDR

ELF program header table entries.

ELF_T_REL

ELF relocation entries.

ELF_T_RELA

ELF relocation entries with addends.

ELF_T_SHDR

ELF section header entries.

ELF_T_SWORD

Signed 32-bit words.

ELF_T_SXWORD

Signed 64-bit words.

ELF_T_SYMINFO

ELF symbol information.

ELF_T_SYM

ELF symbol table entries.

ELF_T_VDEF

Symbol version definition records.

ELF_T_VNEED

Symbol version requirement records.

ELF_T_WORD

Unsigned 32-bit words.

ELF_T_XWORD

Unsigned 64-bit words.

The symbol ELF_T_NUM denotes the number of Elf types known to the library.

The following table shows the mapping between ELF section types defined in elf(5) and the types supported by the library.

Functional Grouping

Archive Access

elf_getarsym()

Retrieve the archive symbol table.

elf_getarhdr()

Retrieve the archive header for an object.

elf_getbase()

Retrieve the offset of a member inside an archive.

elf_next()

Iterate through an ar(1) archive.

elf_rand()

Random access inside an ar(1) archive.

Data Structures

elf_getdata()

Retrieve translated data for an ELF section.

elf_getscn()

Retrieve the section descriptor for a named section.

elf_ndxscn()

Retrieve the index for a section.

elf_newdata()

Add a new Elf_Data descriptor to an ELF section.

elf_newscn()

Add a new section descriptor to an ELF descriptor.

elf_nextscn()

Iterate through the sections in an ELF object.

elf_rawdata()

Retrieve untranslated data for an ELF section.

elf_rawfile()

Return a pointer to the untranslated file contents for an ELF object.

elf32_getehdr(), elf64_getehdr()

Retrieve the Executable Header in an ELF object.

elf32_getphdr(), elf64_getphdr()

Retrieve the Program Header Table in an ELF object.

elf32_getshdr(), elf64_getshdr()

Retrieve the ELF section header associated with an Elf_Scn descriptor.

elf32_newehdr(), elf64_newehdr()

Allocate an Executable Header in an ELF object.

elf32_newphdr(), elf64_newphdr()

Allocate or resize the Program Header Table in an ELF object.

Data Translation

elf32_xlatetof(), elf64_xlatetof()

Translate an ELF data structure from its native representation to its file representation.

elf32_xlatetom(), elf64_xlatetom()

Translate an ELF data structure from its file representation to a native representation.

Error Reporting

elf_errno()

Retrieve the current error.

elf_errmsg()

Retrieve a human readable description of the current error.

Initialization

elf_begin()

Opens an ar(1) archive or ELF object given a file descriptor.

elf_end()

Close an ELF descriptor and release all its resources.

elf_memory()

Opens an ar(1) archive or ELF object present in a memory arena.

elf_version()

Sets the operating version.

IO Control

elf_cntl()

Manage the association between and ELF descriptor and its underlying file.

elf_flagdata()

Mark an Elf_Data descriptor as dirty.

elf_flagehdr()

Mark the ELF Executable Header in an ELF descriptor as dirty.

elf_flagphdr()

Mark the ELF Program Header Table in an ELF descriptor as dirty.

elf_flagscn()

Mark an Elf_Scn descriptor as dirty.

elf_flagshdr()

Mark an ELF Section Header as dirty.

elf_setshstrndx()

Set the index of the section name string table for the ELF object.

elf_update()

Recompute ELF object layout and optionally write the modified object back to the underlying file.

Queries

elf32_checksum(), elf64_checkum()

Compute checksum of an ELF object.

elf_getident()

Retrieve the identification bytes for an ELF object.

elf_getshnum()

Retrieve the number of sections in an ELF object.

elf_getshstrndx()

Retrieve the section index of the section name string table in an ELF object.

elf_hash()

Compute the ELF hash value of a string.

elf_kind()

Query the kind of object associated with an ELF descriptor.

elf32_fsize(), elf64_fsize()

Return the size of the file representation of an ELF type.

Controlling ELF Object Layout

However, if the application wishes to take complete charge of the layout of the ELF file, it may set the ELF_F_LAYOUT flag on an ELF descriptor using elf_flagelf(3), following which the library will use the data offsets and alignments specified by the application when laying out the file. Application control of file layout is described further in the elf_update(3) manual page.

Gaps in between sections will be filled with the fill character set by function elf_fill().

Error Handling

Memory Management Rules

Conversely the library will not free data that it has not allocated. As an example, an application may call elf_newdata(3) to allocate a new Elf_Data descriptor and can set the d_off member of the descriptor to point to a region of memory allocated using malloc(3). It is the applications responsibility to free this arena, though the library will reclaim the space used by the Elf_Data descriptor itself.