ld, ld.so - link editor, dynamic link editor

---

SYNOPSIS

DESCRIPTION

ld combines object programs to create an executable file or another object program suitable for further ld processing (with the -r option). The object modules on which ld operates are specified on the command line, and can be:

• simple object files, which typically end in the .o suffix, and are referred to as ``dot-oh'' files
• ar.1v library archives (.a), or ``libraries''
• dynamically-bound, sharable object files (.so), are also referred to as ``shared libraries,'' which are created from previous ld executions.

Unless an output file is specified, ld produces a file named a.out. This file contains the object files given as input, appropriately combined to form an executable file.

OPTIONS

Options should appear before filenames, except for abbreviated library names specified with -l options, and some binding control options specified by -B (which can appear anywhere in the line).

-align datum

Force the global uninitialized data symbol datum (usually a FORTRAN common block) to be page-aligned. Increase its size to a whole number of pages, and place its first byte at the start of a page.

-assert assertion-keyword

Check an assertion about the link editing being performed. The assertion desired is specified by the assertion-keyword string. ld is silent if the assertion holds, else it yields a diagnostic and aborts. Valid assertion-keyword's and their interpretations are:

definitions

If the resulting program were run now, there would be no run-time undefined symbol diagnostics. This assertion is set by default.

nodefinitions

Permit the successful linking of a program with unresolved references.

nosymbolic

There are no symbolic relocation items remaining to be resolved.

pure-text

The resulting load has no relocation items remaining in its text.

-A name

Incremental loading: linking is to be done in a manner so that the resulting object may be read into an already executing program. name is the name of a file whose symbol table is taken as a basis on which to define additional symbols. Only newly linked material is entered into the text and data portions of a.out, but the new symbol table will reflect all symbols defined before and after the incremental load. This argument must appear before any other object file in the argument list. One or both of the -T options may be used as well, and will be taken to mean that the newly linked segment will commence at the corresponding addresses (which must be a multiple of the page size). The default value is the old value of _end.

-Bbinding-keyword

Specify allowed binding times for the items which follow. Allowed values of binding-keyword are:

dynamic

Allow dynamic binding: do not resolve symbolic references, allow creation of run-time symbol and relocation environment. -Bdynamic is the default. When -Bdynamic is in effect, all sharable objects encountered until a succeeding -Bstatic may be added dynamically to the object being linked. Non-sharable objects are bound statically.

nosymbolic

Do not perform symbolic relocation, even if other options imply it.

static

Bind statically. Opposite of -Bdynamic. Implied when either -n or -N is specified. Influences handling of all objects following its specification on a command line until the next -Bdynamic.

symbolic

Force symbolic relocation. Normally implied if an entry point has been specified with -e, or if dynamic loading is in effect.

-d

Force common storage for uninitialized variables and other common symbols to be allocated in the current ld run, even when the -r flag is present (which would otherwise postpone this binding until the final linking phase).

-dc

Do -d, but also copy initialized data referenced by this program from shared objects.

-dp

Force an alias definition of undefined procedure entry points. Used with dynamic binding to improve sharing and the locality of run-time relocations.

-D hex

Pad the data segment with zero-valued bytes to make it hex bytes long.

-e entry

Define the entry point: the entry argument is made the name of the entry point of the loaded program. Implies -Bsymbolic.

-lx[.v]

This option is an abbreviation for the library name libx.a, where x is a string. ld searches for libraries first in any directories specified with -L options, then in the standard directories /lib, /usr/lib, and /usr/local/lib. A library is searched when its name is encountered, so the placement of a -l is significant. If a dynamically loadable object is found, and -Bdynamic is in effect at that point on the command line, then ld prepares to access the object for relocation at run-time. In such a case, the optional .v suffix can be used to indicate a specific library version.

-Ldir

Add dir to the list of directories in which to search for libraries. Directories specified with -L are searched before the standard directories, /lib, /usr/lib, and /usr/local/lib. When building a program in which one or more objects are loaded when -Bdynamic is in effect, those directories specified by -L options will be ``remembered'' for use at execution time. This permits the construction of software that uses shared objects as libraries not residing in the standard locations and avoids requiring the specification of the LD_LIBRARY_PATH environment variable in order to execute such software. Note that such directories are retained in exactly the form specified in the option, which means that relative directory specifications (i.e., not beginning with ``/'') will be evaluated relative to the current directory when the program is run, not just during the operation of ld.

-M

Produce a primitive load map, listing the names of the files which will be loaded.

-n

Arrange (by giving the output file a 0410 ``magic number'') that when the output file is executed, the text portion will be read-only with the data areas placed at the beginning of the next address boundary following the end of the text. Implies -Bstatic.

-N

Do not make the text portion read-only. (Use ``magic number'' 0407.) Implies -Bstatic.

-o name

name is made the name of the ld output file, instead of a.out.

-p

Arrange for the data segment to begin on a page boundary, even if the text is not shared (with the -N option).

-r

Generate relocation bits in the output file so that it can be the subject of another ld run. This flag also prevents final definitions from being given to common symbols, and suppresses the ``undefined symbol'' diagnostics.

-s

Strip the output, that is, remove the symbol table and relocation bits to save space (but impair the usefulness of the debuggers). This information can also be removed by strip.1

-S

Strip the output by removing all symbols except locals and globals.

-t

Trace: display the name of each file as it is processed.

-T[text]hex

Start the text segment at location hex. Specifying -T is the same as using the -Ttext option.

-Tdata hex

Start the data segment at location hex. This option is only of use to programmers wishing to write code for PROMs, since the resulting code cannot be executed by the system.

-u name

Enter name as an undefined symbol. This is useful for loading wholly from a library, since initially the symbol table is empty and an unresolved reference is needed to force the loading of the first routine.

-x

Preserve only global (non-.globl) symbols in the output symbol table; only enter external symbols. This option saves some space in the output file.

-X

Record local symbols, except for those whose names begin with L. This option is used by cc to discard internally generated labels while retaining symbols local to routines.

-ysym

Display each file in which sym appears, its type and whether the file defines or references it. Many such options may be given to trace many symbols. It is usually necessary to begin sym with an `_', as external C, FORTRAN and Pascal variables begin with underscores.

-z

Arrange for the process demand paged from the resulting executable file (0413 ``magic number''). This is the default. Results in a (32-byte) header on the output file followed by text and data segments, each of which has a multiple of page-size bytes (being padded out with null characters in the file if necessary). With this format the first few BSS segment symbols may actually end up in the data segment; this is to avoid wasting the space resulting from rounding the data segment size. Implies -Bdynamic.

USAGE

Command Line Processing

In general, options should appear ahead of the list of files to process. Unless otherwise specified, the effect of an option covers all of ld operations, independent of that option's placement on the command line. Exceptions to this rule include some of the binding control options specified by `-B' and the abbreviated library-names specified by `-l'. These may appear anywhere, and their influence is dependent upon their location. Some options may be obtained from environment variables, such options are interpreted before any on the command line (see ENVIRONMENT, below).

Object File Processing

The files specified on the command line are processed in the order listed. Information is extracted from each file, and concatenated to form the output. The specific processing performed on a given file depends upon whether it is a simple object file, a library archive, or a shared library.

Simple object (.o) files are concatenated to the output as they are encountered.

Library archive (.a) files are searched exactly once each, as each is encountered; only those archive entries matching an unresolved external reference are extracted and concatenated to the output. If a member of an archive references a symbol defined by another member of that same archive, the member making the reference must appear before the member containing the definition.

On Sun386i, a library contains a dictionary of symbols, On other Sun systems, processing library archives through ranlib.1 provides this dictionary. In addition, you can use lorder.1 in combination with tsort.1 to place library members in calling order (see lorder.1 for details), or both (for fastest symbol lookup). The first member of an archived processed by ranlib has the reserved name of __.SYMDEF, which ld takes to be the dictionary of all symbols defined by members of the archive.

Sharable objects (.so) are scanned for symbol definitions and references, but are not normally included in the output from ld, except in cases where a shared library exports initialized data structures and the -dc option is in effect. However, the occurrence of each sharable object file in the ld command line is noted in the resulting executable file; this notation is utilized by an execution-time variant of ld, ld.so, for deferred and dynamic loading and binding during execution. See Execution-Time Loading, below, for details.

The -l option specifies a short name for an object file or archive used as a library. The full name of the object file is derived by adding the prefix lib and a suffix of either .a or .so[.v] to indicate an ar.1v archive or a shared library, respectively. The specific suffix used is determined through rules discussed in Binding and Relocation Semantics, below.

ld searches for the desired object file through a list of directories specified by -L options, the environment variable LD_LIBRARY_PATH, and finally, the built-in list of standard library directories: /lib, /usr/lib, and /usr/local/lib.

Binding and Relocation Semantics

The manner in which ld processes a given object file is dependent in part upon the ``binding mode'' in which it is operating at the time the file is encountered. This binding mode is specified by the -B flag, which takes the keyword arguments:

dynamic

Allow dynamic binding, do not resolve symbolic references, and allow creation of execution-time symbol and relocation information. This is the default setting.

static

Force static binding, implied by options that generate non-sharable executable formats.

-Bdynamic and -Bstatic may be specified several times, and may be used to toggle each other on and off. Like -l, the influence of each depends upon its location within the command line. When -Bdynamic is in effect, -l searches may be satisfied by the first occurrence of either form of library (.so or .a), but if both are encountered, the .so form is preferred. When -Bstatic is in effect, ld refuses to use any .so libraries it encounters; it continue searching for the .a form. Furthermore, an explicit request to load a .so file is treated as an error.

After ld has processed all input files and command line options, the form of the output it produces is based on the information provided in both. ld first tries to reduce all symbolic references to relative numerical offsets within the executable it is building. To perform this ``symbolic reduction,'' ld must be able to determine that:

• all information relating to the program has been provided, in particular, no .so is to be added at execution time; and/or
• the program has an entry point, and symbolic reduction can be performed for all symbols having definitions existing in the material provided.

It should be noted that uninitialized ``common'' areas (for example, uninitialized C globals) are allocated by the link editor after it has collected all references. In particular, this allocation can not occur in a program that still requires the addition of information contained in a .so file, as the missing information may affect the allocation process. Initialized ``commons'' however, are allocated within the executable in which their definition appears.

After ld has performed all the symbolic reductions it can, it attempts to transform all relative references to absolute addresses. ld is able to perform this ``relative reduction'' only if it has been provided some absolute address, either implicitly through the specification of an entry point, or explicitly through ld command-line options. If, after performing all the reductions it can, there are no further relocations or definitions to perform, then ld has produced a completely linked executable.

Execution-Time Loading

In the event that one or more reductions can not be completed, the executable will require further link editing at execution time in order to be usable. Such executables contain an data structure identified with the symbol __DYNAMIC. An incompletely linked ``main'' program should be linked with a ``bootstrap'' routine that invokes ld.so, which uses the information contained in the main program's __DYNAMIC to assemble the rest of the executables constituting the entire program. A standard Sun compilation driver (such as cc.1v automatically includes such a module in each ``main'' executable.

When ld.so is given control on program startup, it finds all .so files specified when the program was constructed (and all .so's on which they depend), and loads them into the address space. The algorithm by which such files are found mimics that used when ld is run, and like ld, can be influenced by the setting of LD_LIBRARY_PATH and any -L options specified to ld when the program was built. ld.so then completes all remaining relocations, with the exception of procedure call relocations; failure to resolve a given non-procedural relocation results in termination of the program with an appropriate diagnostic.

Procedure relocations are resolved when the referencing instruction is first executed. It should be noted that it is possible for ``undefined symbol'' diagnostics to be produced during program execution if a given target is not defined when referenced.

Although it is possible for binding errors to occur at execution-time, such an occurrence generally indicates something wrong in the maintenance of shared objects. ld's -assert nodefinitions function (on by default) checks at ld-time whether or not an execution-time binding error would occur.

Version Handling for Shared Libraries

To allow the independent evolution of .so's used as libraries and the programs which use them, ld's handling of .so files found through -l options involves the retention and management of version control information. The .so files used as such ``shared libraries'' are post-fixed with a Dewey-decimal format string describing the version of the ``library'' contained in the file.

The first decimal component is called the library's ``major version'' number, and the second component its ``minor version'' number. When ld records a .so used as a library, it also records these two numbers in the database used by ld.so at execution time. In turn, ld.so uses these numbers to decide which of multiple versions of a given library is ``best'' or whether any of the available versions are acceptable. The rules are:

• Major Versions Identical: the major version used at execution time must exactly match the version found at ld-time. Failure to find an instance of the library with a matching major version causes a diagnostic to be issued and the program's execution to be terminated.
• Highest Minor Version: in the presence of multiple instances of libraries that match the desired major version, ld.so uses the highest minor version it finds. However, if the highest minor version found at execution time is less than the version found at ld-time, a warning diagnostic is issued; program execution continues.

The semantics of version numbers are such that major version numbers should be changed whenever interfaces are changed. Minor versions should be changed to reflect compatible updates to libraries, and programs will silently favor the highest compatible version they can obtain.

Special Symbols

A number of symbols have special meanings to ld and programs should not define these symbols. The symbols described below are those actually seen by ld. Note: C and several other languages prepend symbols they use with `_'.

_etext

The first location after the text of the program.

_edata

The first location after initialized data.

_end

The first location after all data.

__DYNAMIC

Identifies an ld-produced data structure. It is defined with a non-zero value in executables which require execution-time link editing. By convention, if defined, it is the first symbol in the symbol table associated with an a.out file.

__GLOBAL_OFFSET_TABLE_

A position-independent reference to an ld-constructed table of addresses. This table is constructed from ``position-independent'' data references occurring in objects that have been assembled with the assembler's -k flag (invoked on behalf of C compilations performed with the -pic flag). A related table (for which no symbol is currently defined) contains a series of transfer instructions and is created from ``position-independent'' procedure calls or, if -dp is specified to ld, a list of undefined symbols.

Symbols in object files beginning with the letter L are taken to be local symbols and unless otherwise specified are purged from ld output files.

ENVIRONMENT

LD_LIBRARY_PATH

A colon-separated list of directories in which to search for libraries specified with the -l option. Similar to the PATH environment variable. LD_LIBRARY_PATH also affects library searching during execution-time loading, causing the search to use first those directories found in the environment variable, then any directories specified by -L options, and finally the standard directories /usr/lib and /usr/local/lib. NOTE: when running a set-user- or set-group-ID program, ld.so will only search for libraries in directories it ``trusts'', which are /usr/lib, /usr/5lib, /usr/local/lib, and any directories specified within the executable as a result of -L options given when the executable was constructed.

LD_OPTIONS

A default set of options to ld. LD_OPTIONS is interpreted by ld just as though its value had been placed on the command line, immediately following the name used to invoke ld, as in:

example% ld $LD_OPTIONS ... other-arguments ...

Note: Environment variable-names beginning with the characters `LD_' are reserved for possible future enhancements to ld.

FILES

/usr/lib/lib*.a

libraries

lib*.so.v

shared libraries

lib*.sa.v

exported, initialized shared library data

/usr/lib/ld.so

execution-time ld

/usr/lib/*crt*.o

default program bootstraps

a.out

output file

/usr/local/lib

SEE ALSO

BUGS

Options are being overloaded and are an inappropriate vehicle for describing to ld the wide variety of things it can do. There needs to be a link-editing language which can be used in the more complex situations.

The -r option does not properly handle programs assembled with the -k (position-independent) flag, invoked from cc with -pic or -PIC.

Created by unroff & hp-tools. © somebody (See intro for details). All Rights Reserved. Last modified 11/5/97