.Dd October 23, 1993 .Dt LINK 5 .Os .Sh NAME .Nm link .Nd dynamic loader and link editor interface .Sh SYNOPSIS .Fd #include .Sh DESCRIPTION The include file .Aq Pa link.h declares several structures that are present in dynamically linked programs and libraries. The structures define the interface between several components of the link-editor and loader mechanism. The layout of a number of these structures within the binaries resembles the a.out format in many places as it serves such similar functions as symbol definitions (including the accompanying string table) and relocation records needed to resolve references to external entities. It also records a number of data structures unique to the dynamic loading and linking process. These include references to other objects that are required to complete the link-editing process and indirection tables to facilitate .Ev Position Independent Code (PIC for short) to improve sharing of code pages among different processes. The collection of data structures described here will be refered to as the .Ev Run-time Relocation Section (RRS) and is embedded in the standard text and data segments of the dynamically linked program or shared object image as the existing .Xr a.out format offers no room for it elsewhere. .Pp Several utilities cooperate to ensure that the task of getting a program ready to run can complete successfully in a way that optimizes the use of system resources. The compiler emits PIC code from which shared libraries can be build by .Ev ld. The compiler also includes size information of any initialized data items through the .size assembler directive. PIC code differs from conventional code in that it accesses data variables through an indirection table, the Global Offset Table, by convention accessable by the reserved name .Ev _GLOBAL_OFFSET_TABLE_. The exact mechanism used for this is machine dependent, usually a machine register is reserved for the purpose. The rational behind this construct is to generate code that is independent of the actual load address. Only the values contained in the Global Offset Table may need updating at run-time depending on the load addresses of the various shared objects in the address space. .Pp Likewise, procedure calls to globally defined functions are redirected through the Procedure Linkage Table (PLT) residing in the data segment of the core image. Again, this is done to avoid run-time modifications to the text segment. .Pp The linker-editor allocates the Global Offset Table and Procedure Linkage Table when combining PIC object files into an image suitable for mapping into the process address space. It also collects all symbols that may be needed by the run-time link-editor and stores these along with the image's text and data bits. Another reserved symbol, .Ev _DYNAMIC is used to indicate the presence of the run-time linker structures. Whenever _DYNAMIC is relocated to 0, there is no need to invoke the run-time link-editor. If this symbol is non-zero, it points at a data structure from which the location of the necessary relocation- and symbol information can be derived. This is most notably used by the start-up module, .Ev crt0. The _DYNAMIC structure is conventionally located at the start of the data segment of the image to which it pertains. .Pp .Sh DATA STRUCTURES The data structures supporting dynamic linking and run-time relocation reside both in the text and data segments of the image they apply to. The text segments contain read-only data such as symbols descriptions and names, while the data segments contain the tables that need to be modified by during the relocation process. .Pp The _DYNAMIC symbol references a .Fa link_dynamic structure: .Bd -literal -offset indent struct link_dynamic { int ld_version; struct ld_debug *ldd; union { struct link_dynamic_2 *ld_2; } ld_un; struct ld_entry *ld_entry; }; .Ed .Bl -tag -width ld_version .It Fa ld_version This field provides for different versions of the dynamic linking implementation. The current version numbers understood by ld and ld.so are .Ev LD_VERSION_SUN (3), which is used by the SunOS 4.x releases, and .Ev LD_VERSION_BSD (8), which is currently in use by NetBSD release 0.9a. .It Fa ld_un Refers to a .Ev ld_version dependent data structure. .It Fa ldd_debug this field provides debuggers with a hook to access symbol tables of shared objects loaded as a result of the actions of the run-time link-editor. .El .Pp The .Fa link_dynamic_2 structure is the main .Dq dispatcher table, containing offsets into the image's segments where various symbol and relocation information is located. .Bd -literal -offset indent struct link_dynamic_2 { struct link_map *ld_loaded; long ld_need; long ld_rules; long ld_got; long ld_plt; long ld_rel; long ld_hash; long ld_symbols; long (*ld_stab_hash)(); long ld_buckets; long ld_strings; long ld_str_sz; long ld_text_sz; long ld_plt_sz; }; .Ed .Pp .Bl -tag -width ld_stab_hash .It Fa ld_loaded A pointer to the first link map loaded (see below). This field is set by .Xr ld.so. .It Fa ld_need The start of a (linked) list of shared objects needed by .Ev this object. .It Fa ld_rules Depricated (used by SunOS to specify library search rules). .It Fa ld_got The location of the Global Offset Table within this image. .It Fa ld_plt The location of the Procedure Linkage Table within this image. .It Fa ld_rel The location of an array of .Fa relocation_info structures (see .Xr a.out 5) specifying run-time relocations. .It Fa ld_hash The location of the hash table for fast symbol lookup in this object's symbol table. .It Fa ld_symbols The location of the symbol table. .It Fa ld_stab_hash Currently unused. .It Fa ld_buckets The number of buckets in .Fa ld_hash .It Fa ld_strings The location of the symbol string table that goes with .Fa ld_symbols. .It Fa ld_str_sz The size of the string table. .It Fa ld_text_sz The size of the object's text segment. .It Fa ld_plt_sz The size of the Procedure Linkage Table. .El .Pp A .Fa link_object structure descibes a shared object that is needed to complete the link edit process of the object containing it. A list of such objects (chained through .Fa lo_next) is pointed at by the .Fa ld_need in the link_dynamic_2 structure. .Bd -literal -offset indent struct link_object { long lo_name; u_int lo_library : 1, lo_unused : 31; short lo_major; short lo_minor; long lo_next; }; .Ed .Pp .Bl -tag -width lo_library .It Fa lo_name The offset in the text segment of a string describing this link object. .It Fa lo_library If set, .Fa lo_name specifies a library that is to be searched for by ld.so. The path name is obtained by searching a set of directories (see .Xr ldconfig) for a shared object matching .Ev lib\&\&.so.n.m. If not set, .Fa lo_name should point at a full path name for the desired shared object. .It Fa lo_major Specifies the major version number of the shared object to load. .It Fa lo_minor Specifies the prefered minor version number of the shared object to load. .El .Pp The run-time link-editor maintains a list of link maps to keep track of all shared objects loaded into a process' address space. These structures are only used at run-time and do not occur within the text or data segment of an executable or shared library. .Bd -literal -offset indent struct link_map { caddr_t lm_addr; char *lm_name; struct link_map *lm_next; struct link_object *lm_lop; caddr_t lm_lob; u_int lm_rwt : 1; struct link_dynamic *lm_ld; caddr_t lm_lpd; }; .Bl -tag -width ld_addr .It Fa lm_addr The address at which the shared object associated with this link map has been loaded. .It Fa lm_name The full path name of the loaded object. .It Fa lm_next Pointer to the next link map. .It Fa lm_lop The .Fa link_object structure that was responsible for this shared object to get loaded. .It Fa lm_lob Depricated. .It Fa lm_rwt Set if this object's text segment is currently writable. .It Fa lm_ld Pointer to this object .Fa link_dynamic structure. .It Fa lm_lpd Hook for attaching private data maintained by the run-time link-editor. .El .Ed .Pp Symbol description with size. This is simply an .Fa nlist structure with one field ( .Fa nz_size ) added. Used to convey size information on items in the data segment of shared objects. An array of these lives in the shared object's text segment and is addressed by the .Fa ld_symbols field of .Fa link_dynamic_2. .Bd -literal -offset indent struct nzlist { struct nlist nlist; u_long nz_size; #define nz_un nlist.n_un #define nz_strx nlist.n_un.n_strx #define nz_name nlist.n_un.n_name #define nz_type nlist.n_type #define nz_value nlist.n_value #define nz_desc nlist.n_desc #define nz_other nlist.n_other }; .Ed .Bl -tag -width nz_size .It Fa nlist (see .Xr nlist 5 ). .It Fa nz_size The size of the data represented by this symbol. .El .Pp A hash table is included within the text segment of shared object to to facilitate quick lookup of symbols during run-time link-editing. The .Fa ld_hash field of the .Fa link_dynamic_2 structure points at an array of .Fa rrs_hash structures: .Bd -literal -offset indent struct rrs_hash { int rh_symbolnum; /* symbol number */ int rh_next; /* next hash entry */ }; .Ed .Pp .Bl -tag -width rh_symbolnum .It Fa rh_symbolnum The index of the symbol in the shared object's symbol table (as given by the .Fa ld_symbols field). .It Fa rh_next In case of collisions, this field is the offset of the next entry in this hash table bucket. It is zero for the last bucket element. .El The .Fa rt_symbol structure is used to keep track of run-time allocated commons and data items copied from shared objects. These items are kept on linked list and is exported through the .Fa ldd_cp field in the .Fa ld_debug structure (see below) for use by debuggers. .Bd -literal -offset indent struct rt_symbol { struct nzlist *rt_sp; struct rt_symbol *rt_next; struct rt_symbol *rt_link; caddr_t rt_srcaddr; }; .Ed .Pp .Bl -tag -width rt_scraddr .It Fa rt_sp The symbol description. .It Fa rt_next Virtual address of next rt_symbol. .It Fa rt_link Next in hash bucket. Used by internally by ld.so. .It Fa rt_srcaddr Location of the source of initialized data within a shared object. .El .Pp The .Fa ld_debug structure is used by debuggers to gain knowledge of any shared objects that have been loaded in the process's address space as a result of run-time link-editing. Since the run-time link-editor runs as a part of process initialization, a debugger that wishes to access symbols from shared objects can only do so after the link-editor has been called from crt0. A dynamically linked binary contains a .Fa ld_debug structure which can be located by means of the .Fa ldd field in .Fa link_dynamic. .Bd -literal -offset indent struct ld_debug { int ldd_version; int ldd_in_debugger; int ldd_sym_loaded; char *ldd_bp_addr; int ldd_bp_inst; struct rt_symbol *ldd_cp; }; .Ed .Pp .Bl -tag -width ldd_in_debugger .It Fa ldd_version Version number of this interface. .It Fa ldd_in_debugger Set by the debugger to indicate to ld.so that the program is run under control of a debugger. .It Fa ldd_sym_loaded Set by ld.so whenever it adds symbols by loading shared objects. .It Fa ldd_bp_addr The address were a breakpoint will be set by the ld.so to divert control to the debugger. This address is determined by the start-up module, .Ev crt0.o, to be some convenient place before the call to _main. .It Fa ldd_bp_inst Contains the original instruction that was at .Fa ldd_bp_addr. The debugger is expected to put this instruction back before continuing the program. .It Fa ldd_cp A pointer to the linked list of run-time allocated symbols that the debugger may interested in. .El .Pp .Bd -literal -offset indent The .Fa ld_entry structure defines a set of service routines within ld.so. See .Ev libdl.a for more information. struct ld_entry { int (*dlopen)(); int (*dlclose)(); int (*dlsym)(); }; .Ed .Bd -literal -offset indent The .Fa crt_ldso structure defines the interface between the start-up code in crt0 and ld.so. struct crt_ldso { int crt_ba; int crt_dzfd; int crt_ldfd; struct link_dynamic *crt_dp; char **crt_ep; caddr_t crt_bp; }; #define CRT_VERSION_SUN 1 #define CRT_VERSION_BSD 2 .Ed .Bl -tag -width crt_dzfd .It Fa crt_ba The virtual address at which ld.so was loaded by crt0. .It Fa crt_dzfd On SunOS systems, this field contains an open file descriptor to .Dq /dev/zero used to get demand paged zeroed pages. On NetBSD systems it contains -1. .It Fa crt_ldfd Contains an open file descriptor that was used by crt0 to load ld.so. .It Fa crt_dp A pointer to main's .Fa link_dynamic structure. .It Fa crt_ep A pointer to the environment strings. .It Fa crt_bp The address at which a breakpoint will be placed by ld.so if run by a debugger. See .Fa ld_debug .El .Pp The .Fa hints_header and .Fa hints_bucket structures define the layout of the library hints, normally found in .Dq /var/run/ld.so.hints, which is used by ld.so to quickly locate the shared object images in the filesystem. The organization of the hints file is not unlike that of an .Dq a.out object file, in that it contains a header determining the offset and size of a table of fixed sized hash buckets and a common string pool. .Bd -literal -offset indent struct hints_header { long hh_magic; #define HH_MAGIC 011421044151 long hh_version; #define LD_HINTS_VERSION_1 1 long hh_hashtab; long hh_nbucket; long hh_strtab; long hh_strtab_sz; long hh_ehints; }; .Ed .Bl -tag -width hh_strtab_sz .It Fa hh_magic Hints file magic number. .It Fa hh_version Interface version number. .It Fa hh_hashtab Offset of hash table. .It Fa hh_strtab Offset of string table. .It Fa hh_strtab_sz Size of strings. .It Fa hh_ehints Maximum usable offset in hints file. .El .Pp .Bd -literal -offset indent /* * Hash table element in hints file. */ struct hints_bucket { int hi_namex; int hi_pathx; int hi_dewey[MAXDEWEY]; int hi_ndewey; #define hi_major hi_dewey[0] #define hi_minor hi_dewey[1] int hi_next; }; .Ed .Bl -tag -width hi_ndewey .It Fa hi_namex Index of the string identifying the library. .It Fa hi_pathx Index of the string representing the full path name of the library. .It Fa hi_dewey The version numbers of the shared library. .It Fa hi_ndewey The number of valid entries in .Fa hi_dewey. .It Fa hi_next Next bucket in case of hashing collisions. .El .Sh CAVEATS Only the GNU C compiler currently supports the creation of shared libraries. Other programming languages can not be used.