diff options
author | Eric Andersen <andersen@codepoet.org> | 2001-04-23 17:43:54 +0000 |
---|---|---|
committer | Eric Andersen <andersen@codepoet.org> | 2001-04-23 17:43:54 +0000 |
commit | 66f269d2a51dae6a2cb10f1a9ae4aaeba815219b (patch) | |
tree | e2094832990caf6d849ba90e4b1a82a6264f86b3 /ldso/ldso | |
parent | c4a3f3f81ea90e3df93c352ac0e2161a4bfd3327 (diff) |
Initial checkin for ld.so. This is a combination of effort from Manuel Novoa
III and me. I've been working on stripping out arch dependant stuff and
replacing it with generic stuff whenever possible.
-Erik
Diffstat (limited to 'ldso/ldso')
37 files changed, 9201 insertions, 0 deletions
diff --git a/ldso/ldso/Makefile b/ldso/ldso/Makefile new file mode 100644 index 000000000..80f0643ea --- /dev/null +++ b/ldso/ldso/Makefile @@ -0,0 +1,39 @@ +TOPDIR=../../ +include $(TOPDIR)Rules.mak +include $(TOPDIR)/ld.so-1/Config.mk + +DIRS = $(TARGET_ARCH) libdl + +CFLAGS += -DNO_UNDERSCORE -DVERBOSE_DLINKER +CFLAGS += -DUSE_CACHE #-fPIC -D__PIC__ #-funroll-loops + +CSRC= boot1.c hash.c readelflib1.c vsprintf.c +COBJS=$(patsubst %.c,%.o, $(CSRC)) +OBJS=$(COBJS) + +DLINKER = ld-linux-uclibc.so + +ELF_LDFLAGS=--shared # using GNU ld + +ifneq ($(DIRS),) +lib realclean clean:: + @set -e; for i in $(DIRS); do \ + echo making $@ in $$i; \ + $(MAKE) -C $$i $@; \ + done; +endif + +lib:: $(OBJS) + $(LD) -e _dl_boot $(ELF_LDFLAGS) -o $(DLINKER).$(LDSO_VMAJOR) \ + -soname $(DLINKER).$(LDSO_VMAJOR) *.o + +$(COBJS): %.o : %.c + $(CC) -I. -I./$(TARGET_ARCH) $(CFLAGS) -c $< -o $@ + $(STRIPTOOL) -x -R .note -R .comment $*.o + +realclean:: + $(RM) -f .depend $(DLINKER) core *.o *.a *.s *.i tmp_make foo *~ + +clean:: + $(RM) -f $(DLINKER) core *.o *.a *.s *.i tmp_make foo *~ + diff --git a/ldso/ldso/boot1.c b/ldso/ldso/boot1.c new file mode 100644 index 000000000..6b0b864cb --- /dev/null +++ b/ldso/ldso/boot1.c @@ -0,0 +1,1005 @@ +/* Run an ELF binary on a linux system. + + Copyright (C) 1993-1996, Eric Youngdale. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ + + + +/* Program to load an ELF binary on a linux system, and run it. + * References to symbols in sharable libraries can be resolved by + * an ELF sharable library. */ + +/* Disclaimer: I have never seen any AT&T source code for SVr4, nor have + I ever taken any courses on internals. This program was developed using + information available through the book "UNIX SYSTEM V RELEASE 4, + Programmers guide: Ansi C and Programming Support Tools", which did + a more than adequate job of explaining everything required to get this + working. */ + +/* + * The main trick with this program is that initially, we ourselves are not + * dynamicly linked. This means that we cannot access any global variables + * since the GOT is initialized by the linker assuming a virtual address of 0, + * and we cannot call any functions since the PLT is not initialized at all + * (it will tend to want to call the dynamic linker + * + * There are further restrictions - we cannot use large switch statements, + * since the compiler generates tables of addresses and jumps through them. + * We can use inline functions, because these do not transfer control to + * a new address, but they must be static so that they are not exported + * from the modules. We cannot use normal syscall stubs, because these + * all reference the errno global variable which is not yet initialized. + * We can use all of the local stack variables that we want, since these + * are all referenced to %ebp or %esp. + * + * Life is further complicated by the fact that initially we do not want + * to do a complete dynamic linking. We want to allow the user to supply + * new functions replacing some of the library versions, and until we have + * the list of modules that we should search set up, we do not want to do + * any of this. Thus I have chosen to only perform the relocations for + * variables that start with "_dl_" since ANSI specifies that the user is + * not supposed to redefine any of these variables. + * + * Fortunately, the linker itself leaves a few clues lying around, and + * when the kernel starts the image, there are a few further clues. + * First of all, there is information buried on the stack that the kernel + * leaves, which includes information about the load address that the + * program interpreter was loaded at, the number of sections, the address + * the application was loaded at and so forth. Here this information + * is stored in the array dl_info, and the indicies are taken from the + * file /usr/include/sys/auxv.h on any SVr4 system. + * + * The linker itself leaves a pointer to the .dynamic section in the first + * slot of the GOT, and as it turns out, %ebx points to ghe GOT when + * you are using PIC code, so we just dereference this to get the address + * of the dynamic sections. + * + * Typically you must load all text pages as writable so that dynamic linking + * can succeed. The kernel under SVr4 loads these as R/O, so we must call + * mprotect to change the protections. Once we are done, we should set these + * back again, so the desired behavior is achieved. Under linux there is + * currently no mprotect function in the distribution kernel (although + * someone has alpha patches), so for now everything is loaded writable. + * + * We do not have access to malloc and friends at the initial stages of dynamic + * linking, and it would be handy to have some scratchpad memory available + * for use as we set things up. It is a bit of a kluge, but we mmap /dev/zero + * to get one page of scratchpad. A simpleminded _dl_malloc is provided so + * that we have some memory that can be used for this purpose. Typically + * we would not want to use the same memory pool as malloc anyway - the user + * might want to redefine malloc for example. + * + * Our first task is to perform a minimal linking so that we can call other + * portions of the dynamic linker. Once we have done this, we then build + * the list of modules that the application requires, using LD_LIBRARY_PATH + * if this is not a suid program (/usr/lib otherwise). Once this is done, + * we can do the dynamic linking as required (and we must omit the things + * we did to get the dynamic linker up and running in the first place. + * After we have done this, we just have a few housekeeping chores and we + * can transfer control to the user's application. + */ + +#include <stdarg.h> +#include <linux/types.h> +#include <linux/fcntl.h> +#include <linux/unistd.h> +#include <linux/elf.h> +#include <linux/mman.h> +#include "link.h" + +#include "sysdep.h" +#include "hash.h" +#include "linuxelf.h" +#include "syscall.h" +#include "string.h" + +#include "../config.h" + +#define ALLOW_ZERO_PLTGOT + +static char * _dl_malloc_addr, *_dl_mmap_zero; +char * _dl_library_path = 0; /* Where we look for libraries */ +char *_dl_preload = 0; /* Things to be loaded before the libs. */ +char *_dl_progname = "/lib/ld-linux-uclibc.so.1"; +static char * _dl_not_lazy = 0; +static char * _dl_warn = 0; /* Used by ldd */ +static char * _dl_trace_loaded_objects = 0; +static int (*_dl_elf_main)(int, char **, char**); + +static int (*_dl_elf_init)(void); + +void * (*_dl_malloc_function)(int size) = NULL; + +struct r_debug * _dl_debug_addr = NULL; + +unsigned int * _dl_brkp; + +unsigned int * _dl_envp; + +#define DL_MALLOC(SIZE) ((void *) (malloc_buffer += SIZE, malloc_buffer - SIZE)) +/* + * Make sure that the malloc buffer is aligned on 4 byte boundary. For 64 bit + * platforms we may need to increase this to 8, but this is good enough for + * now. This is typically called after DL_MALLOC. + */ +#define REALIGN() malloc_buffer = (char *) (((unsigned int) malloc_buffer + 3) & ~(3)) + + + +#define ELF_HASH(RESULT,NAME) { \ + unsigned long hash = 0; \ + unsigned long tmp; \ + char * name = NAME; \ + while (*name){ \ + hash = (hash << 4) + *name++; \ + if((tmp = hash & 0xf0000000)) hash ^= tmp >> 24; \ + hash &= ~tmp; \ + } \ + RESULT = hash; \ +} +extern int _dl_linux_resolve(void); +extern int _dl_interpreter_exit(int); +extern char * _dl_strdup(const char *); +extern char * _dl_getenv(char * symbol, char ** envp); +extern void _dl_unsetenv(char * symbol, char ** envp); +extern int _dl_fixup(struct elf_resolve * tpnt); + +/* + * Datatype of a relocation on this platform + */ +#ifdef ELF_USES_RELOCA +typedef struct elf32_rela ELF_RELOC; +#else +typedef struct elf32_rel ELF_RELOC; +#endif + +/* + * This stub function is used by some debuggers. The idea is that they + * can set an internal breakpoint on it, so that we are notified when the + * address mapping is changed in some way. + */ +void _dl_debug_state() +{ + return; +} + +void _dl_boot(int args); + +void _dl_boot(int args){ + unsigned int argc; + char ** argv, ** envp; + int status; + + unsigned int load_addr; + unsigned int * got; + unsigned int * aux_dat; + int goof = 0; + struct elfhdr * header; + struct elf_resolve * tpnt; + struct dyn_elf * rpnt; + struct elf_resolve * app_tpnt; + unsigned int brk_addr; + unsigned int dl_data[AT_EGID+1]; + unsigned char * malloc_buffer, *mmap_zero; + int (*_dl_atexit)(void *); + int * lpnt; + struct dynamic * dpnt; + unsigned int *hash_addr; + struct r_debug * debug_addr; + unsigned int *chains; + int indx; + int _dl_secure; + + /* First obtain the information on the stack that tells us more about + what binary is loaded, where it is loaded, etc, etc */ + + GET_ARGV(aux_dat, args); + argc = *(aux_dat - 1); + argv = (char **) aux_dat; + aux_dat += argc; /* Skip over the argv pointers */ + aux_dat++; /* Skip over NULL at end of argv */ + envp = (char **) aux_dat; + while(*aux_dat) aux_dat++; /* Skip over the envp pointers */ + aux_dat++; /* Skip over NULL at end of envp */ + dl_data[AT_UID] = -1; /* check later to see if it is changed */ + while(*aux_dat) + { + unsigned int * ad1; + ad1 = aux_dat + 1; + if( *aux_dat <= AT_EGID ) dl_data[*aux_dat] = *ad1; + aux_dat += 2; + } + + /* Next, locate the GOT */ + + load_addr = dl_data[AT_BASE]; + + GET_GOT(got); + dpnt = (struct dynamic *) (*got + load_addr); + + /* OK, time for another hack. Now call mmap to get a page of writable + memory that can be used for a temporary malloc. We do not know brk + yet, so we cannot use real malloc. */ + + { + /* This hack is to work around a suspected asm bug in gcc-2.7.0 */ + int zfileno; +#define ZFILENO ((-1 & (~zfileno)) | zfileno) +/*#define ZFILENO -1*/ + +#ifndef MAP_ANONYMOUS +#ifdef __sparc__ +#define MAP_ANONYMOUS 0x20 +#else +#error MAP_ANONYMOUS not defined and suplementary value not known +#endif +#endif + + /* See if we need to relocate this address */ + mmap_zero = malloc_buffer = (unsigned char *) _dl_mmap((void*) 0, 4096, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, ZFILENO, 0); + if(_dl_mmap_check_error(mmap_zero)) { + SEND_STDERR("dl_boot: mmap of /dev/zero failed!\n"); + _dl_exit(13); + } + } + + tpnt = DL_MALLOC(sizeof(struct elf_resolve)); + REALIGN(); + _dl_memset (tpnt, 0, sizeof (*tpnt)); + app_tpnt = DL_MALLOC(sizeof(struct elf_resolve)); + REALIGN(); + _dl_memset (app_tpnt, 0, sizeof (*app_tpnt)); + + /* + * This is used by gdb to locate the chain of shared libraries that are currently loaded. + */ + debug_addr = DL_MALLOC(sizeof(struct r_debug)); + REALIGN(); + _dl_memset (debug_addr, 0, sizeof (*debug_addr)); + + /* OK, that was easy. Next scan the DYNAMIC section of the image. + We are only doing ourself right now - we will have to do the rest later */ + + while(dpnt->d_tag) + { + tpnt->dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val; + if(dpnt->d_tag == DT_TEXTREL || + SVR4_BUGCOMPAT) tpnt->dynamic_info[DT_TEXTREL] = 1; + dpnt++; + } + + { + struct elf_phdr * ppnt; + int i; + + ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + for(i=0; i<dl_data[AT_PHNUM]; i++, ppnt++) + if(ppnt->p_type == PT_DYNAMIC) { + dpnt = (struct dynamic *) ppnt->p_vaddr; + while(dpnt->d_tag) + { + if(dpnt->d_tag > DT_JMPREL) {dpnt++; continue; } + app_tpnt->dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val; + if(dpnt->d_tag == DT_DEBUG) dpnt->d_un.d_val = (int) debug_addr; + if(dpnt->d_tag == DT_TEXTREL || + SVR4_BUGCOMPAT) app_tpnt->dynamic_info[DT_TEXTREL] = 1; + dpnt++; + } + } + } + + /* Get some more of the information that we will need to dynamicly link + this module to itself */ + + hash_addr = (unsigned int *) (tpnt->dynamic_info[DT_HASH]+load_addr); + tpnt->nbucket = *hash_addr++; + tpnt->nchain = *hash_addr++; + tpnt->elf_buckets = hash_addr; + hash_addr += tpnt->nbucket; + chains = hash_addr; + + /* Ugly, ugly. We need to call mprotect to change the protection of + the text pages so that we can do the dynamic linking. We can set the + protection back again once we are done */ + + { + struct elf_phdr * ppnt; + int i; + + /* First cover the shared library/dynamic linker. */ + if(tpnt->dynamic_info[DT_TEXTREL]) { + header = (struct elfhdr *) dl_data[AT_BASE]; + ppnt = (struct elf_phdr *) (dl_data[AT_BASE] + header->e_phoff); + for(i=0; i<header->e_phnum ; i++, ppnt++) { + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W)) + _dl_mprotect((void *) (load_addr + (ppnt->p_vaddr & 0xfffff000)), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + PROT_READ | PROT_WRITE | PROT_EXEC); + } + } + + /* Now cover the application program. */ + if(app_tpnt->dynamic_info[DT_TEXTREL]) { + ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + for(i=0; i<dl_data[AT_PHNUM]; i++, ppnt++) { + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W)) + _dl_mprotect((void *) (ppnt->p_vaddr & 0xfffff000), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + PROT_READ | PROT_WRITE | PROT_EXEC); + } + } + } + + /* OK, now do the relocations. We do not do a lazy binding here, so + that once we are done, we have considerably more flexibility. */ + + goof = 0; + for(indx=0; indx < 2; indx++) + { + int i; + ELF_RELOC * rpnt; + unsigned int * reloc_addr; + unsigned int symbol_addr; + int symtab_index; + unsigned int rel_addr, rel_size; + + +#ifdef ELF_USES_RELOCA + rel_addr = (indx ? tpnt->dynamic_info[DT_JMPREL] : tpnt->dynamic_info[DT_RELA]); + rel_size = (indx ? tpnt->dynamic_info[DT_PLTRELSZ] : tpnt->dynamic_info[DT_RELASZ]); +#else + rel_addr = (indx ? tpnt->dynamic_info[DT_JMPREL] : tpnt->dynamic_info[DT_REL]); + rel_size = (indx ? tpnt->dynamic_info[DT_PLTRELSZ] : tpnt->dynamic_info[DT_RELSZ]); +#endif + + + if(!rel_addr) continue; + + /* Now parse the relocation information */ + rpnt = (ELF_RELOC *) (rel_addr + load_addr); + for(i=0; i< rel_size; i+=sizeof(ELF_RELOC), rpnt++){ + reloc_addr = (int *) (load_addr + (int)rpnt->r_offset); + symtab_index = ELF32_R_SYM(rpnt->r_info); + symbol_addr = 0; + if(symtab_index) { + char * strtab; + struct elf32_sym * symtab; + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB]+load_addr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB]+load_addr); + + /* We only do a partial dynamic linking right now. The user + is not supposed to redefine any symbols that start with + a '_', so we can do this with confidence. */ + + if (!_dl_symbol(strtab + symtab[symtab_index].st_name)) continue; + + symbol_addr = load_addr + symtab[symtab_index].st_value; + + if(!symbol_addr) { + /* + * This will segfault - you cannot call a function until + * we have finished the relocations. + */ + SEND_STDERR("ELF dynamic loader - unable to self-bootstrap - symbol "); + SEND_STDERR(strtab + symtab[symtab_index].st_name); + SEND_STDERR(" undefined.\n"); + goof++; + } + } + /* + * Use this machine-specific macro to perform the actual relocation. + */ + PERFORM_BOOTSTRAP_RELOC(rpnt, reloc_addr, symbol_addr, load_addr); + } + } + + if (goof) _dl_exit(14); + + /* OK, at this point we have a crude malloc capability. Start to build + the tables of the modules that are required for this beast to run. + We start with the basic executable, and then go from there. Eventually + we will run across ourself, and we will need to properly deal with that + as well. */ + + _dl_malloc_addr = malloc_buffer; + + _dl_mmap_zero = mmap_zero; +/* tpnt = _dl_malloc(sizeof(struct elf_resolve)); */ + +/* Now we have done the mandatory linking of some things. We are now + free to start using global variables, since these things have all been + fixed up by now. Still no function calls outside of this library , + since the dynamic resolver is not yet ready. */ + + lpnt = (int *) (tpnt->dynamic_info[DT_PLTGOT] + load_addr); + INIT_GOT(lpnt, tpnt); + + /* OK, this was a big step, now we need to scan all of the user images + and load them properly. */ + + tpnt->next = 0; + tpnt->libname = 0; + tpnt->libtype = program_interpreter; + + { struct elfhdr * epnt; + struct elf_phdr * ppnt; + int i; + + epnt = (struct elfhdr *) dl_data[AT_BASE]; + tpnt->n_phent = epnt->e_phnum; + tpnt->ppnt = ppnt = (struct elf_phdr *) (load_addr + epnt->e_phoff); + for(i=0;i < epnt->e_phnum; i++, ppnt++){ + if(ppnt->p_type == PT_DYNAMIC) { + tpnt->dynamic_addr = ppnt->p_vaddr + load_addr; + tpnt->dynamic_size = ppnt->p_filesz; + } + } + } + + tpnt->chains = chains; + tpnt->loadaddr = (char *) load_addr; + + brk_addr = 0; + rpnt = NULL; + + /* At this point we are now free to examine the user application, + and figure out which libraries are supposed to be called. Until + we have this list, we will not be completely ready for dynamic linking */ + + { + struct elf_phdr * ppnt; + int i; + + ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + for(i=0; i<dl_data[AT_PHNUM]; i++, ppnt++) { + if(ppnt->p_type == PT_LOAD) { + if(ppnt->p_vaddr + ppnt->p_memsz > brk_addr) + brk_addr = ppnt->p_vaddr + ppnt->p_memsz; + } + if(ppnt->p_type == PT_DYNAMIC) { +#ifndef ALLOW_ZERO_PLTGOT + /* make sure it's really there. */ + if (app_tpnt->dynamic_info[DT_PLTGOT] == 0) continue; +#endif + /* OK, we have what we need - slip this one into the list. */ + app_tpnt = _dl_add_elf_hash_table("", 0, + app_tpnt->dynamic_info, ppnt->p_vaddr, ppnt->p_filesz); + _dl_loaded_modules->libtype = elf_executable; + _dl_loaded_modules->ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + _dl_loaded_modules->n_phent = dl_data[AT_PHNUM]; + _dl_symbol_tables = rpnt = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt, 0, sizeof (*rpnt)); + rpnt->dyn = _dl_loaded_modules; + app_tpnt->usage_count++; + app_tpnt->symbol_scope = _dl_symbol_tables; + lpnt = (int *) (app_tpnt->dynamic_info[DT_PLTGOT]); +#ifdef ALLOW_ZERO_PLTGOT + if (lpnt) +#endif + INIT_GOT(lpnt, _dl_loaded_modules); + } + if(ppnt->p_type == PT_INTERP) { /* OK, fill this in - we did not have + this before */ + tpnt->libname = _dl_strdup((char *) ppnt->p_offset +(dl_data[AT_PHDR] & 0xfffff000)); + } + } + } + + if (argv[0]) + _dl_progname = argv[0]; + + /* Now we need to figure out what kind of options are selected. + Note that for SUID programs we ignore the settings in LD_LIBRARY_PATH */ + { + _dl_not_lazy = _dl_getenv("LD_BIND_NOW",envp); + + if ( (dl_data[AT_UID] == -1 && _dl_suid_ok()) || + (dl_data[AT_UID] != -1 && dl_data[AT_UID] == dl_data[AT_EUID] && + dl_data[AT_GID] == dl_data[AT_EGID])) + { + _dl_secure = 0; + _dl_preload = _dl_getenv("LD_PRELOAD", envp); + _dl_library_path = _dl_getenv("LD_LIBRARY_PATH",envp); + } + else + { + _dl_secure = 1; + _dl_preload = _dl_getenv("LD_PRELOAD", envp); + _dl_unsetenv("LD_AOUT_PRELOAD", envp); + _dl_unsetenv("LD_LIBRARY_PATH", envp); + _dl_unsetenv("LD_AOUT_LIBRARY_PATH", envp); + _dl_library_path = NULL; + } + } + + _dl_trace_loaded_objects = _dl_getenv("LD_TRACE_LOADED_OBJECTS", envp); + + /* OK, we now have the application in the list, and we have some + basic stuff in place. Now search through the list for other shared + libraries that should be loaded, and insert them on the list in the + correct order. */ + +#ifdef USE_CACHE + _dl_map_cache(); +#endif + + { + struct elf_resolve *tcurr; + struct elf_resolve *tpnt1; + char *lpnt; + + if (_dl_preload) { + char c, *str, *str2; + + str = _dl_preload; + while (*str == ':' || *str == ' ' || *str == '\t') + str++; + while (*str) { + str2 = str; + while (*str2 && *str2 != ':' && *str2 != ' ' && *str2 != '\t') + str2++; + c = *str2; + *str2 = '\0'; + if (!_dl_secure || _dl_strchr(str, '/') == NULL) { + tpnt1 = _dl_load_shared_library(_dl_secure, NULL, str); + if (!tpnt1) { + if (_dl_trace_loaded_objects) + _dl_fdprintf(1, "\t%s => not found\n", str); + else { + _dl_fdprintf(2, "%s: can't load library '%s'\n", + _dl_progname, str); + _dl_exit(15); + } + } else { + if (_dl_trace_loaded_objects && !tpnt1->usage_count) { + /* this is a real hack to make ldd not print the + library itself when run on a library. */ + if (_dl_strcmp(_dl_progname, str) != 0) + _dl_fdprintf(1, "\t%s => %s (0x%x)\n", str, tpnt1->libname, + (unsigned)tpnt1->loadaddr); + } + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + tpnt1->usage_count++; + tpnt1->symbol_scope = _dl_symbol_tables; + tpnt1->libtype = elf_lib; + rpnt->dyn = tpnt1; + } + } + *str2 = c; + str = str2; + while (*str == ':' || *str == ' ' || *str == '\t') + str++; + } + } + + { + int fd; + struct kernel_stat st; + char *preload; + + if (!_dl_stat(LDSO_PRELOAD, &st)) { + if ((fd = _dl_open(LDSO_PRELOAD, O_RDONLY)) < 0) { + _dl_fdprintf(2, "%s: can't open file '%s'\n", _dl_progname, + LDSO_PRELOAD); + } else { + preload = (caddr_t)_dl_mmap(0, st.st_size+1, PROT_READ|PROT_WRITE, + MAP_PRIVATE, fd, 0); + _dl_close (fd); + if (preload == (caddr_t)-1) { + _dl_fdprintf(2, "%s: can't map file '%s'\n", _dl_progname, + LDSO_PRELOAD); + } else { + char c, *cp, *cp2; + + /* convert all separators and comments to spaces */ + for (cp = preload; *cp; /*nada*/) { + if (*cp == ':' || *cp == '\t' || *cp == '\n') { + *cp++ = ' '; + } else if (*cp == '#') { + do + *cp++ = ' '; + while (*cp != '\n' && *cp != '\0'); + } else { + cp++; + } + } + + /* find start of first library */ + for (cp = preload; *cp && *cp == ' '; cp++) + /*nada*/; + + while (*cp) { + /* find end of library */ + for (cp2 = cp; *cp && *cp != ' '; cp++) + /*nada*/; + c = *cp; + *cp = '\0'; + + tpnt1 = _dl_load_shared_library(0, NULL, cp2); + if (!tpnt1) { + if (_dl_trace_loaded_objects) + _dl_fdprintf(1, "\t%s => not found\n", cp2); + else { + _dl_fdprintf(2, "%s: can't load library '%s'\n", + _dl_progname, cp2); + _dl_exit(15); + } + } else { + if (_dl_trace_loaded_objects && !tpnt1->usage_count) + _dl_fdprintf(1, "\t%s => %s (0x%x)\n", cp2, tpnt1->libname, + (unsigned)tpnt1->loadaddr); + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + tpnt1->usage_count++; + tpnt1->symbol_scope = _dl_symbol_tables; + tpnt1->libtype = elf_lib; + rpnt->dyn = tpnt1; + } + + /* find start of next library */ + *cp = c; + for (/*nada*/; *cp && *cp == ' '; cp++) + /*nada*/; + } + + _dl_munmap(preload, st.st_size+1); + } + } + } + } + + for (tcurr = _dl_loaded_modules; tcurr; tcurr = tcurr->next) + { + for (dpnt = (struct dynamic *)tcurr->dynamic_addr; dpnt->d_tag; dpnt++) + { + if(dpnt->d_tag == DT_NEEDED) + { + lpnt = tcurr->loadaddr + tcurr->dynamic_info[DT_STRTAB] + + dpnt->d_un.d_val; + if (tpnt && _dl_strcmp(lpnt, tpnt->libname) == 0) + { + struct elf_resolve * ttmp; + ttmp = _dl_loaded_modules; + while (ttmp->next) + ttmp = ttmp->next; + ttmp->next = tpnt; + tpnt->prev = ttmp; + tpnt->next = NULL; + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + rpnt->dyn = tpnt; + tpnt->usage_count++; + tpnt->symbol_scope = _dl_symbol_tables; + tpnt = NULL; + continue; + } + if (!(tpnt1 = _dl_load_shared_library(0, tcurr, lpnt))) + { + if (_dl_trace_loaded_objects) + _dl_fdprintf(1, "\t%s => not found\n", lpnt); + else + { + _dl_fdprintf(2, "%s: can't load library '%s'\n", + _dl_progname, lpnt); + _dl_exit(16); + } + } + else + { + if (_dl_trace_loaded_objects && !tpnt1->usage_count) + _dl_fdprintf(1, "\t%s => %s (0x%x)\n", lpnt, tpnt1->libname, + (unsigned)tpnt1->loadaddr); + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + tpnt1->usage_count++; + tpnt1->symbol_scope = _dl_symbol_tables; + tpnt1->libtype = elf_lib; + rpnt->dyn = tpnt1; + } + } + } + } + } + +#ifdef USE_CACHE + _dl_unmap_cache(); +#endif + + /* ldd uses uses this. I am not sure how you pick up the other flags */ + if(_dl_trace_loaded_objects) + { + _dl_warn = _dl_getenv("LD_WARN", envp); + if (!_dl_warn) _dl_exit(0); + } + + /* + * If the program interpreter is not in the module chain, add it. This will + * be required for dlopen to be able to access the internal functions in the + * dynamic linker. + */ + if(tpnt) { + struct elf_resolve * tcurr; + + tcurr = _dl_loaded_modules; + if (tcurr) + while(tcurr->next) tcurr = tcurr->next; + tpnt->next = NULL; + tpnt->usage_count++; + + if (tcurr) { + tcurr->next = tpnt; + tpnt->prev = tcurr; + } + else { + _dl_loaded_modules = tpnt; + tpnt->prev = NULL; + } + if (rpnt) { + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + } else { + rpnt = (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt, 0, sizeof (*(rpnt->next))); + } + rpnt->dyn = tpnt; + tpnt = NULL; + } + + /* + * OK, now all of the kids are tucked into bed in their proper addresses. + * Now we go through and look for REL and RELA records that indicate fixups + * to the GOT tables. We need to do this in reverse order so that COPY + * directives work correctly */ + + + goof = _dl_loaded_modules ? _dl_fixup(_dl_loaded_modules) : 0; + + + /* Some flavors of SVr4 do not generate the R_*_COPY directive, + and we have to manually search for entries that require fixups. + Solaris gets this one right, from what I understand. */ + + + if (_dl_symbol_tables) + goof += _dl_copy_fixups(_dl_symbol_tables); + + if(goof || _dl_trace_loaded_objects) _dl_exit(0); + + /* OK, at this point things are pretty much ready to run. Now we + need to touch up a few items that are required, and then + we can let the user application have at it. Note that + the dynamic linker itself is not guaranteed to be fully + dynamicly linked if we are using ld.so.1, so we have to look + up each symbol individually. */ + + + _dl_brkp = (unsigned int *) _dl_find_hash("___brk_addr", NULL, 1, NULL, 0); + if (_dl_brkp) *_dl_brkp = brk_addr; + _dl_envp = (unsigned int *) _dl_find_hash("__environ", NULL, 1, NULL, 0); + + if (_dl_envp) *_dl_envp = (unsigned int) envp; + + { + int i; + struct elf_phdr * ppnt; + + /* We had to set the protections of all pages to R/W for dynamic linking. + Set text pages back to R/O */ + for(tpnt = _dl_loaded_modules; tpnt; tpnt = tpnt->next) + for(ppnt = tpnt->ppnt, i=0; i < tpnt->n_phent; i++, ppnt++) + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W) && + tpnt->dynamic_info[DT_TEXTREL]) + _dl_mprotect((void *) (tpnt->loadaddr + (ppnt->p_vaddr & 0xfffff000)), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + LXFLAGS(ppnt->p_flags)); + + } + + _dl_atexit = (int (*)(void *)) _dl_find_hash("atexit", NULL, 1, NULL, 0); + + /* + * OK, fix one more thing - set up the debug_addr structure to point + * to our chain. Later we may need to fill in more fields, but this + * should be enough for now. + */ + debug_addr->r_map = (struct link_map *) _dl_loaded_modules; + debug_addr->r_version = 1; + debug_addr->r_ldbase = load_addr; + debug_addr->r_brk = (unsigned long) &_dl_debug_state; + _dl_debug_addr = debug_addr; + debug_addr->r_state = RT_CONSISTENT; + /* This is written in this funny way to keep gcc from inlining the + function call. */ + ((void (*)(void))debug_addr->r_brk)(); + + for(tpnt = _dl_loaded_modules; tpnt; tpnt = tpnt->next) + { + /* Apparently crt1 for the application is responsible for handling this. + * We only need to run the init/fini for shared libraries + */ + if (tpnt->libtype == program_interpreter || + tpnt->libtype == elf_executable) continue; + if (tpnt->init_flag & INIT_FUNCS_CALLED) continue; + tpnt->init_flag |= INIT_FUNCS_CALLED; + + if(tpnt->dynamic_info[DT_INIT]) { + _dl_elf_init = (int (*)(void)) (tpnt->loadaddr + + tpnt->dynamic_info[DT_INIT]); + (*_dl_elf_init)(); + } + if(_dl_atexit && tpnt->dynamic_info[DT_FINI]) + { + (*_dl_atexit)(tpnt->loadaddr + tpnt->dynamic_info[DT_FINI]); + } +#undef DL_DEBUG +#ifdef DL_DEBUG + else + { + _dl_fdprintf(2, tpnt->libname); + _dl_fdprintf(2, ": "); + if (!_dl_atexit) + _dl_fdprintf(2, "The address is atexit () is 0x0."); + if (!tpnt->dynamic_info[DT_FINI]) + _dl_fdprintf(2, "Invalid .fini section."); + _dl_fdprintf(2, "\n"); + } +#endif +#undef DL_DEBUG + } + + /* OK we are done here. Turn out the lights, and lock up. */ + _dl_elf_main = (int (*)(int, char**, char**)) dl_data[AT_ENTRY]; + + + /* + * Transfer control to the application. + */ + START(); +} + +int _dl_fixup(struct elf_resolve * tpnt) +{ + int goof = 0; + if(tpnt->next) goof += _dl_fixup(tpnt->next); + + if(tpnt->dynamic_info[DT_REL]) { +#ifdef ELF_USES_RELOCA + _dl_fdprintf(2, "%s: can't handle REL relocation records\n", _dl_progname); + _dl_exit(17); +#else + if (tpnt->init_flag & RELOCS_DONE) return goof; + tpnt->init_flag |= RELOCS_DONE; + + goof += _dl_parse_relocation_information(tpnt, tpnt->dynamic_info[DT_REL], + tpnt->dynamic_info[DT_RELSZ], 0); +#endif + } + if(tpnt->dynamic_info[DT_RELA]) { +#ifdef ELF_USES_RELOCA + if (tpnt->init_flag & RELOCS_DONE) return goof; + tpnt->init_flag |= RELOCS_DONE; + + goof += _dl_parse_relocation_information(tpnt, tpnt->dynamic_info[DT_RELA], + tpnt->dynamic_info[DT_RELASZ], 0); +#else + _dl_fdprintf(2, "%s: can't handle RELA relocation records\n", _dl_progname); + _dl_exit(18); +#endif + } + if(tpnt->dynamic_info[DT_JMPREL]) + { + if (tpnt->init_flag & JMP_RELOCS_DONE) return goof; + tpnt->init_flag |= JMP_RELOCS_DONE; + + if(! _dl_not_lazy || *_dl_not_lazy == 0) + _dl_parse_lazy_relocation_information(tpnt, tpnt->dynamic_info[DT_JMPREL], + tpnt->dynamic_info[DT_PLTRELSZ], 0); + else + goof += _dl_parse_relocation_information(tpnt, + tpnt->dynamic_info[DT_JMPREL], + tpnt->dynamic_info[DT_PLTRELSZ], 0); + } + return goof; +} + +void * _dl_malloc(int size) { + void * retval; + + if(_dl_malloc_function) + return (*_dl_malloc_function)(size); + + if(_dl_malloc_addr-_dl_mmap_zero+size>4096) { + _dl_mmap_zero = _dl_malloc_addr = (unsigned char *) _dl_mmap((void*) 0, size, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + if(_dl_mmap_check_error(_dl_mmap_zero)) { + _dl_fdprintf(2, "%s: can't map '/dev/zero'\n", _dl_progname); + _dl_exit(20); + } + } + retval = _dl_malloc_addr; + _dl_malloc_addr += size; + + /* + * Align memory to 4 byte boundary. Some platforms require this, others + * simply get better performance. + */ + _dl_malloc_addr = (char *) (((unsigned int) _dl_malloc_addr + 3) & ~(3)); + return retval; +} + +char * _dl_getenv(char *symbol, char **envp) +{ + char *pnt; + char *pnt1; + while ((pnt = *envp++)) { + pnt1 = symbol; + while (*pnt && *pnt == *pnt1) + pnt1++, pnt++; + if (!*pnt || *pnt != '=' || *pnt1) + continue; + return pnt+1; + } + return 0; +} + +void _dl_unsetenv(char *symbol, char **envp) +{ + char *pnt; + char *pnt1; + char **newenvp = envp; + for (pnt = *envp; pnt; pnt = *++envp) { + pnt1 = symbol; + while (*pnt && *pnt == *pnt1) + pnt1++, pnt++; + if(!*pnt || *pnt != '=' || *pnt1) + *newenvp++ = *envp; + } + *newenvp++ = *envp; + return; +} + +char * _dl_strdup(const char * string){ + char * retval; + int len; + + len = _dl_strlen(string); + retval = _dl_malloc(len + 1); + _dl_strcpy(retval, string); + return retval; +} + +/* In principle we could do the .fini stuff here, but we already + registered this stuff with atexit */ +int _dl_interpreter_exit(int exitcode){ +/* _dl_fdprintf(2, "Hey, look where I am!\n"); */ + return 0; +} diff --git a/ldso/ldso/dl-elf.c b/ldso/ldso/dl-elf.c new file mode 100644 index 000000000..9d1cd0ff5 --- /dev/null +++ b/ldso/ldso/dl-elf.c @@ -0,0 +1,588 @@ +/* Load an ELF sharable library into memory. + + Copyright (C) 1993-1996, Eric Youngdale. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ + + + +/* This file contains the helper routines to load an ELF sharable + library into memory and add the symbol table info to the chain. */ + +#include <linux/types.h> +#include <linux/fcntl.h> +#include <linux/errno.h> +#include "string.h" +/*#include <stdlib.h>*/ +#include <linux/mman.h> +#include <linux/stat.h> +#include "hash.h" +#include "linuxelf.h" +#include "sysdep.h" +#include <linux/unistd.h> +#include "syscall.h" +#ifdef USE_CACHE +#include "../config.h" +#endif + +extern char *_dl_progname; + +#ifdef USE_CACHE + +static caddr_t _dl_cache_addr = NULL; +static size_t _dl_cache_size = 0; + +int _dl_map_cache(void) +{ + int fd; + struct kernel_stat st; + header_t *header; + libentry_t *libent; + int i, strtabsize; + + if (_dl_cache_addr == (caddr_t)-1) + return -1; + else if (_dl_cache_addr != NULL) + return 0; + + if (_dl_stat(LDSO_CACHE, &st) || (fd = _dl_open(LDSO_CACHE, O_RDONLY)) < 0) + { + _dl_fdprintf(2, "%s: can't open cache '%s'\n", _dl_progname, LDSO_CACHE); + _dl_cache_addr = (caddr_t)-1; /* so we won't try again */ + return -1; + } + + _dl_cache_size = st.st_size; + _dl_cache_addr = (caddr_t)_dl_mmap(0, _dl_cache_size, PROT_READ, + MAP_SHARED, fd, 0); + _dl_close (fd); + if (_dl_cache_addr == (caddr_t)-1) + { + _dl_fdprintf(2, "%s: can't map cache '%s'\n", _dl_progname, LDSO_CACHE); + return -1; + } + + header = (header_t *)_dl_cache_addr; + + if (_dl_cache_size < sizeof (header_t) || + _dl_memcmp(header->magic, LDSO_CACHE_MAGIC, LDSO_CACHE_MAGIC_LEN) || + _dl_memcmp(header->version, LDSO_CACHE_VER, LDSO_CACHE_VER_LEN) || + _dl_cache_size < + (sizeof (header_t) + header->nlibs * sizeof (libentry_t)) || + _dl_cache_addr[_dl_cache_size-1] != '\0') + { + _dl_fdprintf(2, "%s: cache '%s' is corrupt\n", _dl_progname, LDSO_CACHE); + goto fail; + } + + strtabsize = _dl_cache_size - sizeof (header_t) - + header->nlibs * sizeof (libentry_t); + libent = (libentry_t *)&header[1]; + + for (i = 0; i < header->nlibs; i++) + { + if (libent[i].sooffset >= strtabsize || + libent[i].liboffset >= strtabsize) + { + _dl_fdprintf(2, "%s: cache '%s' is corrupt\n", _dl_progname, LDSO_CACHE); + goto fail; + } + } + + return 0; + +fail: + _dl_munmap(_dl_cache_addr, _dl_cache_size); + _dl_cache_addr = (caddr_t)-1; + return -1; +} + +int _dl_unmap_cache(void) +{ + if (_dl_cache_addr == NULL || _dl_cache_addr == (caddr_t)-1) + return -1; + +#if 1 + _dl_munmap (_dl_cache_addr, _dl_cache_size); + _dl_cache_addr = NULL; +#endif + + return 0; +} + +#endif + +/* + * Used to return error codes back to dlopen et. al. + */ + +unsigned int _dl_error_number; +unsigned int _dl_internal_error_number; + +struct elf_resolve * _dl_load_shared_library(int secure, + struct elf_resolve * tpnt, char * full_libname) { + char * pnt, *pnt1, *pnt2; + struct elf_resolve *tpnt1 = NULL; + char mylibname[2050]; + char * libname; + + _dl_internal_error_number = 0; + + /* quick hack to ensure mylibname buffer doesn't overflow. don't + allow full_libname or any directory to be longer than 1024. */ + if (_dl_strlen(full_libname) > 1024) + goto goof; + + pnt = libname = full_libname; + while (*pnt) { + if(*pnt == '/') libname = pnt+1; + pnt++; + } + + /* If the filename has any '/', try it straight and leave it at that. + For IBCS2 compatibility under linux, we substitute the string + /usr/i486-sysv4/lib for /usr/lib in library names. */ + + if (libname != full_libname) { + tpnt1 = _dl_load_elf_shared_library(secure, full_libname, 0); + if (tpnt1) + return tpnt1; + goto goof; + } + + /* + * The ABI specifies that RPATH is searched before LD_*_PATH or + * the default path of /usr/lib. + * Check in rpath directories + */ + for (tpnt = _dl_loaded_modules; tpnt; tpnt = tpnt->next) { + if (tpnt->libtype == elf_executable) { + pnt1 = (char *)tpnt->dynamic_info[DT_RPATH]; + if(pnt1) { + pnt1 += (unsigned int) tpnt->loadaddr + tpnt->dynamic_info[DT_STRTAB]; + while(*pnt1){ + pnt2 = mylibname; + while(*pnt1 && *pnt1 != ':') { + if (pnt2 - mylibname < 1024) + *pnt2++ = *pnt1++; + else + pnt1++; + } + if (pnt2 - mylibname >= 1024) + break; + if(pnt2[-1] != '/') *pnt2++ = '/'; + pnt = libname; + while(*pnt) *pnt2++ = *pnt++; + *pnt2++ = 0; + tpnt1 = _dl_load_elf_shared_library(secure, mylibname, 0); + if(tpnt1) return tpnt1; + if(*pnt1 == ':') pnt1++; + } + } + } + } + + + /* Check in LD_{ELF_}LIBRARY_PATH, if specified and allowed */ + pnt1 = _dl_library_path; + if (pnt1 && *pnt1) { + while (*pnt1) { + pnt2 = mylibname; + while(*pnt1 && *pnt1 != ':' && *pnt1 != ';') { + if (pnt2 - mylibname < 1024) + *pnt2++ = *pnt1++; + else + pnt1++; + } + if (pnt2 - mylibname >= 1024) + break; + if(pnt2[-1] != '/') *pnt2++ = '/'; + pnt = libname; + while(*pnt) *pnt2++ = *pnt++; + *pnt2++ = 0; + tpnt1 = _dl_load_elf_shared_library(secure, mylibname, 0); + if(tpnt1) return tpnt1; + if(*pnt1 == ':' || *pnt1 == ';') pnt1++; + } + } + + + /* + * Where should the cache be searched? There is no such concept in the + * ABI, so we have some flexibility here. For now, search it before + * the default path of /usr/lib. + */ +#ifdef USE_CACHE + if (_dl_cache_addr != NULL && _dl_cache_addr != (caddr_t)-1) + { + int i; + header_t *header = (header_t *)_dl_cache_addr; + libentry_t *libent = (libentry_t *)&header[1]; + char *strs = (char *)&libent[header->nlibs]; + + for (i = 0; i < header->nlibs; i++) + { + if ((libent[i].flags == LIB_ELF || + libent[i].flags == LIB_ELF_LIBC5) && + _dl_strcmp(libname, strs+libent[i].sooffset) == 0 && + (tpnt1 = _dl_load_elf_shared_library(secure, strs+libent[i].liboffset, 0))) + return tpnt1; + } + } +#endif + + +#ifdef UCLIBC_DEVEL + + /* Check in /usr/<arch>-linux-uclibc/lib */ + pnt1 = UCLIBC_INSTALL_DIR"/lib"; + pnt = mylibname; + while(*pnt1) *pnt++ = *pnt1++; + pnt1 = libname; + while(*pnt1) *pnt++ = *pnt1++; + *pnt++ = 0; + tpnt1 = _dl_load_elf_shared_library(secure, mylibname, 0); + if (tpnt1) return tpnt1; + +#else /* UCLIBC_DEVEL */ + + /* Check in /usr/lib */ + pnt1 = "/usr/lib/"; + pnt = mylibname; + while(*pnt1) *pnt++ = *pnt1++; + pnt1 = libname; + while(*pnt1) *pnt++ = *pnt1++; + *pnt++ = 0; + tpnt1 = _dl_load_elf_shared_library(secure, mylibname, 0); + if (tpnt1) return tpnt1; + + /* Check in /lib */ + /* try "/lib/". */ + pnt1 = "/lib/"; + pnt = mylibname; + while(*pnt1) *pnt++ = *pnt1++; + pnt1 = libname; + while(*pnt1) *pnt++ = *pnt1++; + *pnt++ = 0; + tpnt1 = _dl_load_elf_shared_library(secure, mylibname, 0); + if (tpnt1) return tpnt1; +#endif /* UCLIBC_DEVEL */ + +goof: + /* Well, we shot our wad on that one. All we can do now is punt */ + if (_dl_internal_error_number) _dl_error_number = _dl_internal_error_number; + else _dl_error_number = DL_ERROR_NOFILE; + return NULL; +} + +/* + * Read one ELF library into memory, mmap it into the correct locations and + * add the symbol info to the symbol chain. Perform any relocations that + * are required. + */ + +//extern _elf_rtbndr(void); + +struct elf_resolve * _dl_load_elf_shared_library(int secure, + char * libname, int flag) { + struct elfhdr * epnt; + unsigned int dynamic_addr = 0; + unsigned int dynamic_size = 0; + struct dynamic * dpnt; + struct elf_resolve * tpnt; + struct elf_phdr * ppnt; + int piclib; + char * status; + int flags; + char header[4096]; + int dynamic_info[24]; + int * lpnt; + unsigned int libaddr; + unsigned int minvma=0xffffffff, maxvma=0; + + int i; + int infile; + + /* If this file is already loaded, skip this step */ + tpnt = _dl_check_hashed_files(libname); + if(tpnt) return tpnt; + + /* If we are in secure mode (i.e. a setu/gid binary using LD_PRELOAD), + we don't load the library if it isn't setuid. */ + + if (secure) { + struct kernel_stat st; + if (_dl_stat(libname, &st) || !(st.st_mode & S_ISUID)) + return NULL; + } + + libaddr = 0; + infile = _dl_open(libname, O_RDONLY); + if(infile < 0) + { +#if 0 + /* + * NO! When we open shared libraries we may search several paths. + * it is inappropriate to generate an error here. + */ + _dl_fdprintf(2, "%s: can't open '%s'\n", _dl_progname, libname); +#endif + _dl_internal_error_number = DL_ERROR_NOFILE; + return NULL; + } + + _dl_read(infile, header, sizeof(header)); + epnt = (struct elfhdr *) header; + if (epnt->e_ident[0] != 0x7f || + epnt->e_ident[1] != 'E' || + epnt->e_ident[2] != 'L' || + epnt->e_ident[3] != 'F') { + _dl_fdprintf(2, "%s: '%s' is not an ELF file\n", _dl_progname, libname); + _dl_internal_error_number = DL_ERROR_NOTELF; + _dl_close(infile); + return NULL; + }; + + if((epnt->e_type != ET_DYN) || + (epnt->e_machine != MAGIC1 +#ifdef MAGIC2 + && epnt->e_machine != MAGIC2 +#endif + )){ + _dl_internal_error_number = (epnt->e_type != ET_DYN ? DL_ERROR_NOTDYN : DL_ERROR_NOTMAGIC); + _dl_fdprintf(2, "%s: '%s' is not an ELF executable for " ELF_TARGET "\n", + _dl_progname, libname); + _dl_close(infile); + return NULL; + }; + + ppnt = (struct elf_phdr *) &header[epnt->e_phoff]; + + piclib = 1; + for(i=0;i < epnt->e_phnum; i++){ + + if(ppnt->p_type == PT_DYNAMIC) { + if (dynamic_addr) + _dl_fdprintf(2, "%s: '%s' has more than one dynamic section\n", + _dl_progname, libname); + dynamic_addr = ppnt->p_vaddr; + dynamic_size = ppnt->p_filesz; + }; + + if(ppnt->p_type == PT_LOAD) { + /* See if this is a PIC library. */ + if(i == 0 && ppnt->p_vaddr > 0x1000000) { + piclib = 0; + minvma=ppnt->p_vaddr; + } + if(piclib && ppnt->p_vaddr < minvma) { + minvma = ppnt->p_vaddr; + } + if(((unsigned int)ppnt->p_vaddr + ppnt->p_memsz) > maxvma) { + maxvma = ppnt->p_vaddr + ppnt->p_memsz; + } + } + ppnt++; + }; + + maxvma=(maxvma+0xfffU)&~0xfffU; + minvma=minvma&~0xffffU; + + flags = MAP_PRIVATE /*| MAP_DENYWRITE*/; + if(!piclib) flags |= MAP_FIXED; + + status = (char *) _dl_mmap((char *) (piclib?0:minvma), + maxvma-minvma, + PROT_NONE, + flags | MAP_ANONYMOUS, -1, + 0); + if(_dl_mmap_check_error(status)) { + _dl_fdprintf(2, "%s: can't map '/dev/zero'\n", _dl_progname); + _dl_internal_error_number = DL_ERROR_MMAP_FAILED; + _dl_close(infile); + return NULL; + }; + libaddr=(unsigned int)status; + flags|=MAP_FIXED; + + /* Get the memory to store the library */ + ppnt = (struct elf_phdr *) &header[epnt->e_phoff]; + + for(i=0;i < epnt->e_phnum; i++){ + if(ppnt->p_type == PT_LOAD) { + + /* See if this is a PIC library. */ + if(i == 0 && ppnt->p_vaddr > 0x1000000) { + piclib = 0; + /* flags |= MAP_FIXED; */ + } + + + + if(ppnt->p_flags & PF_W) { + unsigned int map_size; + char * cpnt; + + status = (char *) _dl_mmap((char *) ((piclib?libaddr:0) + + (ppnt->p_vaddr & 0xfffff000)), + (ppnt->p_vaddr & 0xfff) + ppnt->p_filesz, + LXFLAGS(ppnt->p_flags), + flags, infile, + ppnt->p_offset & 0x7ffff000); + + if(_dl_mmap_check_error(status)) { + _dl_fdprintf(2, "%s: can't map '%s'\n", _dl_progname, libname); + _dl_internal_error_number = DL_ERROR_MMAP_FAILED; + _dl_munmap((char *)libaddr, maxvma-minvma); + _dl_close(infile); + return NULL; + }; + + /* Pad the last page with zeroes. */ + cpnt =(char *) (status + (ppnt->p_vaddr & 0xfff) + ppnt->p_filesz); + while(((unsigned int) cpnt) & 0xfff) *cpnt++ = 0; + +/* I am not quite sure if this is completely correct to do or not, but + the basic way that we handle bss segments is that we mmap /dev/zero if + there are any pages left over that are not mapped as part of the file */ + + map_size = (ppnt->p_vaddr + ppnt->p_filesz + 0xfff) & 0xfffff000; + if(map_size < ppnt->p_vaddr + ppnt->p_memsz) + status = (char *) _dl_mmap((char *) map_size + (piclib?libaddr:0), + ppnt->p_vaddr + ppnt->p_memsz - map_size, + LXFLAGS(ppnt->p_flags), + flags | MAP_ANONYMOUS, -1, 0); + } else + status = (char *) _dl_mmap((char *) (ppnt->p_vaddr & 0xfffff000) + + (piclib?libaddr:0), + (ppnt->p_vaddr & 0xfff) + ppnt->p_filesz, + LXFLAGS(ppnt->p_flags), + flags, infile, + ppnt->p_offset & 0x7ffff000); + if(_dl_mmap_check_error(status)) { + _dl_fdprintf(2, "%s: can't map '%s'\n", _dl_progname, libname); + _dl_internal_error_number = DL_ERROR_MMAP_FAILED; + _dl_munmap((char *)libaddr, maxvma-minvma); + _dl_close(infile); + return NULL; + }; + + /* if(libaddr == 0 && piclib) { + libaddr = (unsigned int) status; + flags |= MAP_FIXED; + }; */ + }; + ppnt++; + }; + _dl_close(infile); + + /* For a non-PIC library, the addresses are all absolute */ + if(piclib) { + dynamic_addr += (unsigned int) libaddr; + } + + /* + * OK, the ELF library is now loaded into VM in the correct locations + * The next step is to go through and do the dynamic linking (if needed). + */ + + /* Start by scanning the dynamic section to get all of the pointers */ + + if(!dynamic_addr) { + _dl_internal_error_number = DL_ERROR_NODYNAMIC; + _dl_fdprintf(2, "%s: '%s' is missing a dynamic section\n", _dl_progname, libname); + return NULL; + } + + dpnt = (struct dynamic *) dynamic_addr; + + dynamic_size = dynamic_size / sizeof(struct dynamic); + _dl_memset(dynamic_info, 0, sizeof(dynamic_info)); + for(i=0; i< dynamic_size; i++){ + if( dpnt->d_tag > DT_JMPREL ) {dpnt++; continue; } + dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val; + if(dpnt->d_tag == DT_TEXTREL || + SVR4_BUGCOMPAT) dynamic_info[DT_TEXTREL] = 1; + dpnt++; + }; + + /* If the TEXTREL is set, this means that we need to make the pages + writable before we perform relocations. Do this now. They get set back + again later. */ + + if (dynamic_info[DT_TEXTREL]) { + ppnt = (struct elf_phdr *) &header[epnt->e_phoff]; + for(i=0;i < epnt->e_phnum; i++, ppnt++){ + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W)) + _dl_mprotect((void *) ((piclib?libaddr:0) + (ppnt->p_vaddr & 0xfffff000)), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + PROT_READ | PROT_WRITE | PROT_EXEC); + } + } + + + tpnt = _dl_add_elf_hash_table(libname, (char *) libaddr, dynamic_info, dynamic_addr, + dynamic_size); + + tpnt->ppnt = (struct elf_phdr *) (tpnt->loadaddr + epnt->e_phoff); + tpnt->n_phent = epnt->e_phnum; + + /* + * OK, the next thing we need to do is to insert the dynamic linker into + * the proper entry in the GOT so that the PLT symbols can be properly + * resolved. + */ + + lpnt = (int *) dynamic_info[DT_PLTGOT]; + + if(lpnt) { + lpnt = (int *) (dynamic_info[DT_PLTGOT] + ((int) libaddr)); + INIT_GOT(lpnt, tpnt); + }; + + return tpnt; +} + +/* Ugly, ugly. Some versions of the SVr4 linker fail to generate COPY + relocations for global variables that are present both in the image and + the shared library. Go through and do it manually. If the images + are guaranteed to be generated by a trustworthy linker, then this + step can be skipped. */ + +int _dl_copy_fixups(struct dyn_elf * rpnt) +{ + int goof = 0; + struct elf_resolve * tpnt; + + if(rpnt->next) goof += _dl_copy_fixups(rpnt->next); + else return 0; + + tpnt = rpnt->dyn; + + if (tpnt->init_flag & COPY_RELOCS_DONE) return goof; + tpnt->init_flag |= COPY_RELOCS_DONE; + +#ifdef ELF_USES_RELOCA + goof += _dl_parse_copy_information(rpnt, tpnt->dynamic_info[DT_RELA], + tpnt->dynamic_info[DT_RELASZ], 0); + +#else + goof += _dl_parse_copy_information(rpnt, tpnt->dynamic_info[DT_REL], + tpnt->dynamic_info[DT_RELSZ], 0); + +#endif + return goof; +} + diff --git a/ldso/ldso/dl-hash.c b/ldso/ldso/dl-hash.c new file mode 100644 index 000000000..4eab974a8 --- /dev/null +++ b/ldso/ldso/dl-hash.c @@ -0,0 +1,284 @@ +/* Run an ELF binary on a linux system. + + Copyright (C) 1993-1996, Eric Youngdale. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ + + + +/* Various symbol table handling functions, including symbol lookup */ + +/*#include <stdlib.h>*/ +#include "string.h" +#include <linux/unistd.h> +#include <linux/elf.h> + +#include "libdl/dlfcn.h" +#include "hash.h" +#include "linuxelf.h" +#include "syscall.h" +#include "string.h" +#include "sysdep.h" + +/* + * This is the start of the linked list that describes all of the files present + * in the system with pointers to all of the symbol, string, and hash tables, + * as well as all of the other good stuff in the binary. + */ + +struct elf_resolve * _dl_loaded_modules = NULL; + +/* + * This is the list of modules that are loaded when the image is first + * started. As we add more via dlopen, they get added into other + * chains. + */ +struct dyn_elf * _dl_symbol_tables = NULL; + +/* + * This is the list of modules that are loaded via dlopen. We may need + * to search these for RTLD_GLOBAL files. + */ +struct dyn_elf * _dl_handles = NULL; + + +/* + * This is the hash function that is used by the ELF linker to generate + * the hash table that each executable and library is required to + * have. We need it to decode the hash table. + */ + +unsigned long _dl_elf_hash(const char * name){ + unsigned long hash = 0; + unsigned long tmp; + + while (*name){ + hash = (hash << 4) + *name++; + if((tmp = hash & 0xf0000000)) hash ^= tmp >> 24; + hash &= ~tmp; + }; + return hash; +} + +/* + * Check to see if a library has already been added to the hash chain. + */ +struct elf_resolve * _dl_check_hashed_files(char * libname){ + struct elf_resolve * tpnt; + int len = _dl_strlen(libname); + + for (tpnt = _dl_loaded_modules; tpnt; tpnt = tpnt->next) { + if (_dl_strncmp(tpnt->libname, libname, len) == 0 && + (tpnt->libname[len] == '\0' || tpnt->libname[len] == '.')) + return tpnt; + } + + return NULL; +} + +/* + * We call this function when we have just read an ELF library or executable. + * We add the relevant info to the symbol chain, so that we can resolve all + * externals properly. + */ + +struct elf_resolve * _dl_add_elf_hash_table(char * libname, + char * loadaddr, + unsigned int * dynamic_info, + unsigned int dynamic_addr, + unsigned int dynamic_size){ + unsigned int * hash_addr; + struct elf_resolve * tpnt; + int i; + + if (!_dl_loaded_modules) { + tpnt = _dl_loaded_modules = + (struct elf_resolve *) _dl_malloc(sizeof(struct elf_resolve)); + _dl_memset (tpnt, 0, sizeof (*tpnt)); + } + else { + tpnt = _dl_loaded_modules; + while(tpnt->next) tpnt = tpnt->next; + tpnt->next = (struct elf_resolve *) _dl_malloc(sizeof(struct elf_resolve)); + _dl_memset (tpnt->next, 0, sizeof (*(tpnt->next))); + tpnt->next->prev = tpnt; + tpnt = tpnt->next; + }; + + tpnt->next = NULL; + tpnt->init_flag = 0; + tpnt->libname = _dl_strdup(libname); + tpnt->dynamic_addr = dynamic_addr; + tpnt->dynamic_size = dynamic_size; + tpnt->libtype = loaded_file; + + if( dynamic_info[DT_HASH] != 0 ) + { + hash_addr = (unsigned int *) (dynamic_info[DT_HASH] + loadaddr); + tpnt->nbucket = *hash_addr++; + tpnt->nchain = *hash_addr++; + tpnt->elf_buckets = hash_addr; + hash_addr += tpnt->nbucket; + tpnt->chains = hash_addr; + } + tpnt->loadaddr = loadaddr; + for(i=0; i<24; i++) tpnt->dynamic_info[i] = dynamic_info[i]; + return tpnt; +} + + +/* + * This function resolves externals, and this is either called when we process + * relocations or when we call an entry in the PLT table for the first time. + */ + +char * _dl_find_hash(char * name, struct dyn_elf * rpnt1, + unsigned int instr_addr, struct elf_resolve * f_tpnt, + int copyrel){ + struct elf_resolve * tpnt; + int si; + char * pnt; + int pass; + char * strtab; + struct elf32_sym * symtab; + unsigned int elf_hash_number, hn; + char * weak_result; + struct elf_resolve * first_def; + struct dyn_elf * rpnt, first; + char * data_result = 0; /* nakao */ + + weak_result = 0; + elf_hash_number = _dl_elf_hash(name); + + /* A quick little hack to make sure that any symbol in the executable + will be preferred to one in a shared library. This is necessary so + that any shared library data symbols referenced in the executable + will be seen at the same address by the executable, shared libraries + and dynamically loaded code. -Rob Ryan (robr@cmu.edu) */ + if(!copyrel && rpnt1) { + first=(*_dl_symbol_tables); + first.next=rpnt1; + rpnt1=(&first); + } + + /* + * The passes are so that we can first search the regular symbols + * for whatever module was specified, and then search anything + * loaded with RTLD_GLOBAL. When pass is 1, it means we are just + * starting the first dlopened module, and anything above that + * is just the next one in the chain. + */ + for(pass = 0; (1==1); pass++) + { + + /* + * If we are just starting to search for RTLD_GLOBAL, setup + * the pointer for the start of the search. + */ + if( pass == 1) { + rpnt1 = _dl_handles; + } + + /* + * Anything after this, we need to skip to the next module. + */ + else if( pass >= 2) { + rpnt1 = rpnt1->next_handle; + } + + /* + * Make sure we still have a module, and make sure that this + * module was loaded with RTLD_GLOBAL. + */ + if( pass != 0 ) + { + if( rpnt1 == NULL ) break; + if( (rpnt1->flags & RTLD_GLOBAL) == 0) continue; + } + + for(rpnt = (rpnt1 ? rpnt1 : _dl_symbol_tables); + rpnt; rpnt = rpnt->next) { + tpnt = rpnt->dyn; + + /* + * The idea here is that if we are using dlsym, we want to + * first search the entire chain loaded from dlopen, and + * return a result from that if we found anything. If this + * fails, then we continue the search into the stuff loaded + * when the image was activated. For normal lookups, we start + * with rpnt == NULL, so we should never hit this. + */ + if( tpnt->libtype == elf_executable + && weak_result != 0 ) + { + break; + } + + /* + * Avoid calling .urem here. + */ + do_rem(hn, elf_hash_number, tpnt->nbucket); + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + + tpnt->loadaddr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + /* + * This crap is required because the first instance of a + * symbol on the chain will be used for all symbol references. + * Thus this instance must be resolved to an address that + * contains the actual function, + */ + + first_def = NULL; + + for(si = tpnt->elf_buckets[hn]; si; si = tpnt->chains[si]){ + pnt = strtab + symtab[si].st_name; + + if(_dl_strcmp(pnt, name) == 0 && + (ELF32_ST_TYPE(symtab[si].st_info) == STT_FUNC || + ELF32_ST_TYPE(symtab[si].st_info) == STT_NOTYPE || + ELF32_ST_TYPE(symtab[si].st_info) == STT_OBJECT) && + symtab[si].st_value != 0) { + + /* Here we make sure that we find a module where the symbol is + * actually defined. + */ + + if(f_tpnt) { + if(!first_def) first_def = tpnt; + if(first_def == f_tpnt && symtab[si].st_shndx == 0) + continue; + } + + switch(ELF32_ST_BIND(symtab[si].st_info)){ + case STB_GLOBAL: + if ( tpnt->libtype != elf_executable + && ELF32_ST_TYPE(symtab[si].st_info) == STT_NOTYPE) { /* nakao */ + data_result = tpnt->loadaddr + symtab[si].st_value; /* nakao */ + break; /* nakao */ + } else /* nakao */ + return tpnt->loadaddr + symtab[si].st_value; + case STB_WEAK: + if (!weak_result) weak_result = tpnt->loadaddr + symtab[si].st_value; + break; + default: /* Do local symbols need to be examined? */ + break; + } + } + } + } + } + if (data_result) return data_result; /* nakao */ + return weak_result; +} diff --git a/ldso/ldso/hash.c b/ldso/ldso/hash.c new file mode 100644 index 000000000..4eab974a8 --- /dev/null +++ b/ldso/ldso/hash.c @@ -0,0 +1,284 @@ +/* Run an ELF binary on a linux system. + + Copyright (C) 1993-1996, Eric Youngdale. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ + + + +/* Various symbol table handling functions, including symbol lookup */ + +/*#include <stdlib.h>*/ +#include "string.h" +#include <linux/unistd.h> +#include <linux/elf.h> + +#include "libdl/dlfcn.h" +#include "hash.h" +#include "linuxelf.h" +#include "syscall.h" +#include "string.h" +#include "sysdep.h" + +/* + * This is the start of the linked list that describes all of the files present + * in the system with pointers to all of the symbol, string, and hash tables, + * as well as all of the other good stuff in the binary. + */ + +struct elf_resolve * _dl_loaded_modules = NULL; + +/* + * This is the list of modules that are loaded when the image is first + * started. As we add more via dlopen, they get added into other + * chains. + */ +struct dyn_elf * _dl_symbol_tables = NULL; + +/* + * This is the list of modules that are loaded via dlopen. We may need + * to search these for RTLD_GLOBAL files. + */ +struct dyn_elf * _dl_handles = NULL; + + +/* + * This is the hash function that is used by the ELF linker to generate + * the hash table that each executable and library is required to + * have. We need it to decode the hash table. + */ + +unsigned long _dl_elf_hash(const char * name){ + unsigned long hash = 0; + unsigned long tmp; + + while (*name){ + hash = (hash << 4) + *name++; + if((tmp = hash & 0xf0000000)) hash ^= tmp >> 24; + hash &= ~tmp; + }; + return hash; +} + +/* + * Check to see if a library has already been added to the hash chain. + */ +struct elf_resolve * _dl_check_hashed_files(char * libname){ + struct elf_resolve * tpnt; + int len = _dl_strlen(libname); + + for (tpnt = _dl_loaded_modules; tpnt; tpnt = tpnt->next) { + if (_dl_strncmp(tpnt->libname, libname, len) == 0 && + (tpnt->libname[len] == '\0' || tpnt->libname[len] == '.')) + return tpnt; + } + + return NULL; +} + +/* + * We call this function when we have just read an ELF library or executable. + * We add the relevant info to the symbol chain, so that we can resolve all + * externals properly. + */ + +struct elf_resolve * _dl_add_elf_hash_table(char * libname, + char * loadaddr, + unsigned int * dynamic_info, + unsigned int dynamic_addr, + unsigned int dynamic_size){ + unsigned int * hash_addr; + struct elf_resolve * tpnt; + int i; + + if (!_dl_loaded_modules) { + tpnt = _dl_loaded_modules = + (struct elf_resolve *) _dl_malloc(sizeof(struct elf_resolve)); + _dl_memset (tpnt, 0, sizeof (*tpnt)); + } + else { + tpnt = _dl_loaded_modules; + while(tpnt->next) tpnt = tpnt->next; + tpnt->next = (struct elf_resolve *) _dl_malloc(sizeof(struct elf_resolve)); + _dl_memset (tpnt->next, 0, sizeof (*(tpnt->next))); + tpnt->next->prev = tpnt; + tpnt = tpnt->next; + }; + + tpnt->next = NULL; + tpnt->init_flag = 0; + tpnt->libname = _dl_strdup(libname); + tpnt->dynamic_addr = dynamic_addr; + tpnt->dynamic_size = dynamic_size; + tpnt->libtype = loaded_file; + + if( dynamic_info[DT_HASH] != 0 ) + { + hash_addr = (unsigned int *) (dynamic_info[DT_HASH] + loadaddr); + tpnt->nbucket = *hash_addr++; + tpnt->nchain = *hash_addr++; + tpnt->elf_buckets = hash_addr; + hash_addr += tpnt->nbucket; + tpnt->chains = hash_addr; + } + tpnt->loadaddr = loadaddr; + for(i=0; i<24; i++) tpnt->dynamic_info[i] = dynamic_info[i]; + return tpnt; +} + + +/* + * This function resolves externals, and this is either called when we process + * relocations or when we call an entry in the PLT table for the first time. + */ + +char * _dl_find_hash(char * name, struct dyn_elf * rpnt1, + unsigned int instr_addr, struct elf_resolve * f_tpnt, + int copyrel){ + struct elf_resolve * tpnt; + int si; + char * pnt; + int pass; + char * strtab; + struct elf32_sym * symtab; + unsigned int elf_hash_number, hn; + char * weak_result; + struct elf_resolve * first_def; + struct dyn_elf * rpnt, first; + char * data_result = 0; /* nakao */ + + weak_result = 0; + elf_hash_number = _dl_elf_hash(name); + + /* A quick little hack to make sure that any symbol in the executable + will be preferred to one in a shared library. This is necessary so + that any shared library data symbols referenced in the executable + will be seen at the same address by the executable, shared libraries + and dynamically loaded code. -Rob Ryan (robr@cmu.edu) */ + if(!copyrel && rpnt1) { + first=(*_dl_symbol_tables); + first.next=rpnt1; + rpnt1=(&first); + } + + /* + * The passes are so that we can first search the regular symbols + * for whatever module was specified, and then search anything + * loaded with RTLD_GLOBAL. When pass is 1, it means we are just + * starting the first dlopened module, and anything above that + * is just the next one in the chain. + */ + for(pass = 0; (1==1); pass++) + { + + /* + * If we are just starting to search for RTLD_GLOBAL, setup + * the pointer for the start of the search. + */ + if( pass == 1) { + rpnt1 = _dl_handles; + } + + /* + * Anything after this, we need to skip to the next module. + */ + else if( pass >= 2) { + rpnt1 = rpnt1->next_handle; + } + + /* + * Make sure we still have a module, and make sure that this + * module was loaded with RTLD_GLOBAL. + */ + if( pass != 0 ) + { + if( rpnt1 == NULL ) break; + if( (rpnt1->flags & RTLD_GLOBAL) == 0) continue; + } + + for(rpnt = (rpnt1 ? rpnt1 : _dl_symbol_tables); + rpnt; rpnt = rpnt->next) { + tpnt = rpnt->dyn; + + /* + * The idea here is that if we are using dlsym, we want to + * first search the entire chain loaded from dlopen, and + * return a result from that if we found anything. If this + * fails, then we continue the search into the stuff loaded + * when the image was activated. For normal lookups, we start + * with rpnt == NULL, so we should never hit this. + */ + if( tpnt->libtype == elf_executable + && weak_result != 0 ) + { + break; + } + + /* + * Avoid calling .urem here. + */ + do_rem(hn, elf_hash_number, tpnt->nbucket); + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + + tpnt->loadaddr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + /* + * This crap is required because the first instance of a + * symbol on the chain will be used for all symbol references. + * Thus this instance must be resolved to an address that + * contains the actual function, + */ + + first_def = NULL; + + for(si = tpnt->elf_buckets[hn]; si; si = tpnt->chains[si]){ + pnt = strtab + symtab[si].st_name; + + if(_dl_strcmp(pnt, name) == 0 && + (ELF32_ST_TYPE(symtab[si].st_info) == STT_FUNC || + ELF32_ST_TYPE(symtab[si].st_info) == STT_NOTYPE || + ELF32_ST_TYPE(symtab[si].st_info) == STT_OBJECT) && + symtab[si].st_value != 0) { + + /* Here we make sure that we find a module where the symbol is + * actually defined. + */ + + if(f_tpnt) { + if(!first_def) first_def = tpnt; + if(first_def == f_tpnt && symtab[si].st_shndx == 0) + continue; + } + + switch(ELF32_ST_BIND(symtab[si].st_info)){ + case STB_GLOBAL: + if ( tpnt->libtype != elf_executable + && ELF32_ST_TYPE(symtab[si].st_info) == STT_NOTYPE) { /* nakao */ + data_result = tpnt->loadaddr + symtab[si].st_value; /* nakao */ + break; /* nakao */ + } else /* nakao */ + return tpnt->loadaddr + symtab[si].st_value; + case STB_WEAK: + if (!weak_result) weak_result = tpnt->loadaddr + symtab[si].st_value; + break; + default: /* Do local symbols need to be examined? */ + break; + } + } + } + } + } + if (data_result) return data_result; /* nakao */ + return weak_result; +} diff --git a/ldso/ldso/hash.h b/ldso/ldso/hash.h new file mode 100644 index 000000000..2eeda2d46 --- /dev/null +++ b/ldso/ldso/hash.h @@ -0,0 +1,113 @@ +#include "link.h" + +#ifndef RTLD_NEXT +#define RTLD_NEXT ((void*)-1) +#endif + +struct dyn_elf{ + unsigned int flags; + struct elf_resolve * dyn; + struct dyn_elf * next_handle; /* Used by dlopen et al. */ + struct dyn_elf * next; +}; + +struct elf_resolve{ + /* These entries must be in this order to be compatible with the interface used + by gdb to obtain the list of symbols. */ + char * loadaddr; + char * libname; + unsigned int dynamic_addr; + struct elf_resolve * next; + struct elf_resolve * prev; + /* Nothing after this address is used by gdb. */ + enum {elf_lib, elf_executable,program_interpreter, loaded_file} libtype; + struct dyn_elf * symbol_scope; + unsigned short usage_count; + unsigned short int init_flag; + unsigned int nbucket; + unsigned int * elf_buckets; + /* + * These are only used with ELF style shared libraries + */ + unsigned int nchain; + unsigned int * chains; + unsigned int dynamic_info[24]; + + unsigned int dynamic_size; + unsigned int n_phent; + struct elf_phdr * ppnt; +}; + +#if 0 +/* + * The DT_DEBUG entry in the .dynamic section is given the address of this structure. + * gdb can pick this up to obtain the correct list of loaded modules. + */ + +struct r_debug{ + int r_version; + struct elf_resolve * link_map; + unsigned long brk_fun; + enum {RT_CONSISTENT, RT_ADD, RT_DELETE}; + unsigned long ldbase; +}; +#endif + +#define COPY_RELOCS_DONE 1 +#define RELOCS_DONE 2 +#define JMP_RELOCS_DONE 4 +#define INIT_FUNCS_CALLED 8 + +extern struct dyn_elf * _dl_symbol_tables; +extern struct elf_resolve * _dl_loaded_modules; +extern struct dyn_elf * _dl_handles; + +extern struct elf_resolve * _dl_check_hashed_files(char * libname); +extern struct elf_resolve * _dl_add_elf_hash_table(char * libname, + char * loadaddr, + unsigned int * dynamic_info, + unsigned int dynamic_addr, + unsigned int dynamic_size); +extern char * _dl_find_hash(char * name, struct dyn_elf * rpnt1, + unsigned int instr_addr, + struct elf_resolve * f_tpnt, + int copyrel); + +extern int _dl_linux_dynamic_link(void); + +#ifdef __mc68000__ +/* On m68k constant strings are referenced through the GOT. */ +/* XXX Requires load_addr to be defined. */ +#define SEND_STDERR(X) \ + { const char *__s = (X); \ + if (__s < (const char *) load_addr) __s += load_addr; \ + _dl_write (2, __s, _dl_strlen (__s)); \ + } +#else +#define SEND_STDERR(X) _dl_write(2, X, _dl_strlen(X)); +#endif +extern int _dl_fdprintf(int, const char *, ...); +extern char * _dl_library_path; +extern char * _dl_not_lazy; +extern char * _dl_strdup(const char *); +extern inline int _dl_symbol(char * name); +unsigned long _dl_elf_hash(const char * name); + +extern inline int _dl_symbol(char * name) +{ + if(name[0] != '_' || name[1] != 'd' || name[2] != 'l' || name[3] != '_') + return 0; + return 1; +} + +#define DL_ERROR_NOFILE 1 +#define DL_ERROR_NOZERO 2 +#define DL_ERROR_NOTELF 3 +#define DL_ERROR_NOTMAGIC 4 +#define DL_ERROR_NOTDYN 5 +#define DL_ERROR_MMAP_FAILED 6 +#define DL_ERROR_NODYNAMIC 7 +#define DL_WRONG_RELOCS 8 +#define DL_BAD_HANDLE 9 +#define DL_NO_SYMBOL 10 + diff --git a/ldso/ldso/i386/dl-sysdep.h b/ldso/ldso/i386/dl-sysdep.h new file mode 100644 index 000000000..9bbeef1fa --- /dev/null +++ b/ldso/ldso/i386/dl-sysdep.h @@ -0,0 +1,82 @@ + +/* + * Various assmbly language/system dependent hacks that are required + * so that we can minimize the amount of platform specific code. + */ + +/* + * Define this if the system uses RELOCA. + */ +#undef ELF_USES_RELOCA + +/* + * Get a pointer to the argv array. On many platforms this can be just + * the address if the first argument, on other platforms we need to + * do something a little more subtle here. + */ +#define GET_ARGV(ARGVP, ARGS) ARGVP = ((unsigned int*) & ARGS) +/* + * Get the address of the Global offset table. This must be absolute, not + * relative. + */ +#define GET_GOT(X) __asm__("\tmovl %%ebx,%0\n\t" : "=a" (X)) + +/* + * Initialization sequence for a GOT. + */ +#define INIT_GOT(GOT_BASE,MODULE) \ +{ \ + GOT_BASE[2] = (int) _dl_linux_resolve; \ + GOT_BASE[1] = (int) MODULE; \ +} + +/* + * Here is a macro to perform a relocation. This is only used when + * bootstrapping the dynamic loader. RELP is the relocation that we + * are performing, REL is the pointer to the address we are relocating. + * SYMBOL is the symbol involved in the relocation, and LOAD is the + * load address. + */ +#define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD) \ + switch(ELF32_R_TYPE((RELP)->r_info)){ \ + case R_386_32: \ + *REL += SYMBOL; \ + break; \ + case R_386_PC32: \ + *REL += SYMBOL - (unsigned int) REL; \ + break; \ + case R_386_GLOB_DAT: \ + case R_386_JMP_SLOT: \ + *REL = SYMBOL; \ + break; \ + case R_386_RELATIVE: \ + *REL += (unsigned int) LOAD; \ + break; \ + default: \ + _dl_exit(1); \ + } + + +/* + * Transfer control to the user's application, once the dynamic loader + * is done. + */ + +#define START() \ + __asm__ volatile ("leave\n\t" \ + "jmp *%%eax\n\t" \ + : "=a" (status) : \ + "d" (_dl_interpreter_exit), "a" (_dl_elf_main)) + + + +/* Here we define the magic numbers that this dynamic loader should accept */ + +#define MAGIC1 EM_386 +#define MAGIC2 EM_486 +/* Used for error messages */ +#define ELF_TARGET "386/486" + +extern unsigned int _dl_linux_resolver(int dummy, int i); + +#define do_rem(result, n, base) result = (n % base) diff --git a/ldso/ldso/i386/elfinterp.c b/ldso/ldso/i386/elfinterp.c new file mode 100644 index 000000000..3347882d6 --- /dev/null +++ b/ldso/ldso/i386/elfinterp.c @@ -0,0 +1,321 @@ +/* Run an ELF binary on a linux system. + + Copyright (C) 1993, Eric Youngdale. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ + +#ifndef VERBOSE_DLINKER +#define VERBOSE_DLINKER +#endif +#ifdef VERBOSE_DLINKER +static char * _dl_reltypes[] = {"R_386_NONE","R_386_32","R_386_PC32","R_386_GOT32", + "R_386_PLT32","R_386_COPY","R_386_GLOB_DAT", + "R_386_JMP_SLOT","R_386_RELATIVE","R_386_GOTOFF", + "R_386_GOTPC","R_386_NUM"}; +#endif + +/* Program to load an ELF binary on a linux system, and run it. +References to symbols in sharable libraries can be resolved by either +an ELF sharable library or a linux style of shared library. */ + +/* Disclaimer: I have never seen any AT&T source code for SVr4, nor have + I ever taken any courses on internals. This program was developed using + information available through the book "UNIX SYSTEM V RELEASE 4, + Programmers guide: Ansi C and Programming Support Tools", which did + a more than adequate job of explaining everything required to get this + working. */ + +#include <sys/types.h> +#include <errno.h> +#include <fcntl.h> +#include <linux/elf.h> + +#include "hash.h" +#include "linuxelf.h" +#include "../string.h" +#include "../syscall.h" + +#define SVR4_COMPATIBILITY + +extern char *_dl_progname; + +extern int _dl_linux_resolve(void); + +unsigned int _dl_linux_resolver(int dummy, int i) +{ + unsigned int * sp; + int reloc_entry; + int reloc_type; + struct elf32_rel * this_reloc; + char * strtab; + struct elf32_sym * symtab; + struct elf32_rel * rel_addr; + struct elf_resolve * tpnt; + int symtab_index; + char * new_addr; + char ** got_addr; + unsigned int instr_addr; + sp = &i; + reloc_entry = sp[1]; + tpnt = (struct elf_resolve *) sp[0]; + + rel_addr = (struct elf32_rel *) (tpnt->dynamic_info[DT_JMPREL] + + tpnt->loadaddr); + + this_reloc = rel_addr + (reloc_entry >> 3); + reloc_type = ELF32_R_TYPE(this_reloc->r_info); + symtab_index = ELF32_R_SYM(this_reloc->r_info); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + tpnt->loadaddr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + + if (reloc_type != R_386_JMP_SLOT) { + _dl_fdprintf(2, "%s: Incorrect relocation type in jump relocations\n", + _dl_progname); + _dl_exit(1); + }; + + /* Address of jump instruction to fix up */ + instr_addr = ((int)this_reloc->r_offset + (int)tpnt->loadaddr); + got_addr = (char **) instr_addr; + +#ifdef DEBUG + _dl_fdprintf(2, "Resolving symbol %s\n", + strtab + symtab[symtab_index].st_name); +#endif + + /* Get the address of the GOT entry */ + new_addr = _dl_find_hash(strtab + symtab[symtab_index].st_name, + tpnt->symbol_scope, (int) got_addr, tpnt, 0); + if(!new_addr) { + _dl_fdprintf(2, "%s: can't resolve symbol '%s'\n", + _dl_progname, strtab + symtab[symtab_index].st_name); + _dl_exit(1); + }; +/* #define DEBUG_LIBRARY */ +#ifdef DEBUG_LIBRARY + if((unsigned int) got_addr < 0x40000000) { + _dl_fdprintf(2, "Calling library function: %s\n", + strtab + symtab[symtab_index].st_name); + } else { + *got_addr = new_addr; + } +#else + *got_addr = new_addr; +#endif + return (unsigned int) new_addr; +} + +void _dl_parse_lazy_relocation_information(struct elf_resolve * tpnt, int rel_addr, + int rel_size, int type){ + int i; + char * strtab; + int reloc_type; + int symtab_index; + struct elf32_sym * symtab; + struct elf32_rel * rpnt; + unsigned int * reloc_addr; + + /* Now parse the relocation information */ + rpnt = (struct elf32_rel *) (rel_addr + tpnt->loadaddr); + rel_size = rel_size / sizeof(struct elf32_rel); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + tpnt->loadaddr); + strtab = ( char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + for(i=0; i< rel_size; i++, rpnt++){ + reloc_addr = (int *) (tpnt->loadaddr + (int)rpnt->r_offset); + reloc_type = ELF32_R_TYPE(rpnt->r_info); + symtab_index = ELF32_R_SYM(rpnt->r_info); + + /* When the dynamic linker bootstrapped itself, it resolved some symbols. + Make sure we do not do them again */ + if(!symtab_index && tpnt->libtype == program_interpreter) continue; + if(symtab_index && tpnt->libtype == program_interpreter && + _dl_symbol(strtab + symtab[symtab_index].st_name)) + continue; + + switch(reloc_type){ + case R_386_NONE: break; + case R_386_JMP_SLOT: + *reloc_addr += (unsigned int) tpnt->loadaddr; + break; + default: + _dl_fdprintf(2, "%s: (LAZY) can't handle reloc type ", _dl_progname); +#ifdef VERBOSE_DLINKER + _dl_fdprintf(2, "%s ", _dl_reltypes[reloc_type]); +#endif + if(symtab_index) _dl_fdprintf(2, "'%s'\n", + strtab + symtab[symtab_index].st_name); + _dl_exit(1); + }; + }; +} + +int _dl_parse_relocation_information(struct elf_resolve * tpnt, int rel_addr, + int rel_size, int type){ + int i; + char * strtab; + int reloc_type; + int goof = 0; + struct elf32_sym * symtab; + struct elf32_rel * rpnt; + unsigned int * reloc_addr; + unsigned int symbol_addr; + int symtab_index; + /* Now parse the relocation information */ + + rpnt = (struct elf32_rel *) (rel_addr + tpnt->loadaddr); + rel_size = rel_size / sizeof(struct elf32_rel); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + tpnt->loadaddr); + strtab = ( char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + for(i=0; i< rel_size; i++, rpnt++){ + reloc_addr = (int *) (tpnt->loadaddr + (int)rpnt->r_offset); + reloc_type = ELF32_R_TYPE(rpnt->r_info); + symtab_index = ELF32_R_SYM(rpnt->r_info); + symbol_addr = 0; + + if(!symtab_index && tpnt->libtype == program_interpreter) continue; + + if(symtab_index) { + + if(tpnt->libtype == program_interpreter && + _dl_symbol(strtab + symtab[symtab_index].st_name)) + continue; + + symbol_addr = (unsigned int) + _dl_find_hash(strtab + symtab[symtab_index].st_name, + tpnt->symbol_scope, (int) reloc_addr, + (reloc_type == R_386_JMP_SLOT ? tpnt : NULL), 0); + + /* + * We want to allow undefined references to weak symbols - this might + * have been intentional. We should not be linking local symbols + * here, so all bases should be covered. + */ + if(!symbol_addr && + ELF32_ST_BIND(symtab[symtab_index].st_info) == STB_GLOBAL) { + _dl_fdprintf(2, "%s: can't resolve symbol '%s'\n", + _dl_progname, strtab + symtab[symtab_index].st_name); + goof++; + } + } + switch(reloc_type){ + case R_386_NONE: + break; + case R_386_32: + *reloc_addr += symbol_addr; + break; + case R_386_PC32: + *reloc_addr += symbol_addr - (unsigned int) reloc_addr; + break; + case R_386_GLOB_DAT: + case R_386_JMP_SLOT: + *reloc_addr = symbol_addr; + break; + case R_386_RELATIVE: + *reloc_addr += (unsigned int) tpnt->loadaddr; + break; + case R_386_COPY: +#if 0 /* Do this later */ + _dl_fdprintf(2, "Doing copy for symbol "); + if(symtab_index) _dl_fdprintf(2, strtab + symtab[symtab_index].st_name); + _dl_fdprintf(2, "\n"); + _dl_memcpy((void *) symtab[symtab_index].st_value, + (void *) symbol_addr, + symtab[symtab_index].st_size); +#endif + break; + default: + _dl_fdprintf(2, "%s: can't handle reloc type ", _dl_progname); +#ifdef VERBOSE_DLINKER + _dl_fdprintf(2, "%s ", _dl_reltypes[reloc_type]); +#endif + if (symtab_index) + _dl_fdprintf(2, "'%s'\n", strtab + symtab[symtab_index].st_name); + _dl_exit(1); + }; + + }; + return goof; +} + + +/* This is done as a separate step, because there are cases where + information is first copied and later initialized. This results in + the wrong information being copied. Someone at Sun was complaining about + a bug in the handling of _COPY by SVr4, and this may in fact be what he + was talking about. Sigh. */ + +/* No, there are cases where the SVr4 linker fails to emit COPY relocs + at all */ + +int _dl_parse_copy_information(struct dyn_elf * xpnt, int rel_addr, + int rel_size, int type) +{ + int i; + char * strtab; + int reloc_type; + int goof = 0; + struct elf32_sym * symtab; + struct elf32_rel * rpnt; + unsigned int * reloc_addr; + unsigned int symbol_addr; + struct elf_resolve *tpnt; + int symtab_index; + /* Now parse the relocation information */ + + tpnt = xpnt->dyn; + + rpnt = (struct elf32_rel *) (rel_addr + tpnt->loadaddr); + rel_size = rel_size / sizeof(struct elf32_rel); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + tpnt->loadaddr); + strtab = ( char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + for(i=0; i< rel_size; i++, rpnt++){ + reloc_addr = (int *) (tpnt->loadaddr + (int)rpnt->r_offset); + reloc_type = ELF32_R_TYPE(rpnt->r_info); + if(reloc_type != R_386_COPY) continue; + symtab_index = ELF32_R_SYM(rpnt->r_info); + symbol_addr = 0; + if(!symtab_index && tpnt->libtype == program_interpreter) continue; + if(symtab_index) { + + if(tpnt->libtype == program_interpreter && + _dl_symbol(strtab + symtab[symtab_index].st_name)) + continue; + + symbol_addr = (unsigned int) + _dl_find_hash(strtab + symtab[symtab_index].st_name, + xpnt->next, (int) reloc_addr, NULL, 1); + if(!symbol_addr) { + _dl_fdprintf(2, "%s: can't resolve symbol '%s'\n", + _dl_progname, strtab + symtab[symtab_index].st_name); + goof++; + }; + }; + if (!goof) + _dl_memcpy((char *) symtab[symtab_index].st_value, + (char *) symbol_addr, + symtab[symtab_index].st_size); + }; + return goof; +} + + diff --git a/ldso/ldso/i386/ld_sysdep.h b/ldso/ldso/i386/ld_sysdep.h new file mode 100644 index 000000000..9bbeef1fa --- /dev/null +++ b/ldso/ldso/i386/ld_sysdep.h @@ -0,0 +1,82 @@ + +/* + * Various assmbly language/system dependent hacks that are required + * so that we can minimize the amount of platform specific code. + */ + +/* + * Define this if the system uses RELOCA. + */ +#undef ELF_USES_RELOCA + +/* + * Get a pointer to the argv array. On many platforms this can be just + * the address if the first argument, on other platforms we need to + * do something a little more subtle here. + */ +#define GET_ARGV(ARGVP, ARGS) ARGVP = ((unsigned int*) & ARGS) +/* + * Get the address of the Global offset table. This must be absolute, not + * relative. + */ +#define GET_GOT(X) __asm__("\tmovl %%ebx,%0\n\t" : "=a" (X)) + +/* + * Initialization sequence for a GOT. + */ +#define INIT_GOT(GOT_BASE,MODULE) \ +{ \ + GOT_BASE[2] = (int) _dl_linux_resolve; \ + GOT_BASE[1] = (int) MODULE; \ +} + +/* + * Here is a macro to perform a relocation. This is only used when + * bootstrapping the dynamic loader. RELP is the relocation that we + * are performing, REL is the pointer to the address we are relocating. + * SYMBOL is the symbol involved in the relocation, and LOAD is the + * load address. + */ +#define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD) \ + switch(ELF32_R_TYPE((RELP)->r_info)){ \ + case R_386_32: \ + *REL += SYMBOL; \ + break; \ + case R_386_PC32: \ + *REL += SYMBOL - (unsigned int) REL; \ + break; \ + case R_386_GLOB_DAT: \ + case R_386_JMP_SLOT: \ + *REL = SYMBOL; \ + break; \ + case R_386_RELATIVE: \ + *REL += (unsigned int) LOAD; \ + break; \ + default: \ + _dl_exit(1); \ + } + + +/* + * Transfer control to the user's application, once the dynamic loader + * is done. + */ + +#define START() \ + __asm__ volatile ("leave\n\t" \ + "jmp *%%eax\n\t" \ + : "=a" (status) : \ + "d" (_dl_interpreter_exit), "a" (_dl_elf_main)) + + + +/* Here we define the magic numbers that this dynamic loader should accept */ + +#define MAGIC1 EM_386 +#define MAGIC2 EM_486 +/* Used for error messages */ +#define ELF_TARGET "386/486" + +extern unsigned int _dl_linux_resolver(int dummy, int i); + +#define do_rem(result, n, base) result = (n % base) diff --git a/ldso/ldso/i386/resolve.S b/ldso/ldso/i386/resolve.S new file mode 100644 index 000000000..db22e7b82 --- /dev/null +++ b/ldso/ldso/i386/resolve.S @@ -0,0 +1,55 @@ +#if 0 +#include <sysdep.h> +#endif +/* + * These are various helper routines that are needed to run an ELF image. + */ +#ifndef ALIGN +#define ALIGN 4 +#endif + +#ifndef NO_UNDERSCORE +#define RUN _linux_run +#define RESOLVE __dl_linux_resolve +#define EXIT __interpreter_exit +#define RESOLVER __dl_linux_resolver +#define INIT ___loader_bootstrap +#else +#define RUN linux_run +#define RESOLVE _dl_linux_resolve +#define RESOLVER _dl_linux_resolver +#define EXIT _interpreter_exit +#define INIT __loader_bootstrap +#endif + +.text +.align ALIGN + .align 16 + +.globl RESOLVE + .type RESOLVE,@function +RESOLVE: + pusha /* preserve all regs */ + lea 0x20(%esp),%eax /* eax = tpnt and reloc_entry params */ + pushl 4(%eax) /* push copy of reloc_entry param */ + pushl (%eax) /* push copy of tpnt param */ + pushl %eax /* _dl_linux_resolver expects a dummy + * param - this could be removed */ +#ifdef __PIC__ + call .L24 +.L24: + popl %ebx + addl $_GLOBAL_OFFSET_TABLE_+[.-.L24],%ebx + movl RESOLVER@GOT(%ebx),%ebx /* eax = resolved func */ + call *%ebx +#else + call RESOLVER +#endif + movl %eax,0x2C(%esp) /* store func addr over original + * tpnt param */ + addl $0xC,%esp /* remove copy parameters */ + popa /* restore regs */ + ret $4 /* jump to func removing original + * reloc_entry param from stack */ +.LFE2: + .size RESOLVE,.LFE2-RESOLVE diff --git a/ldso/ldso/i386/sysdep.h b/ldso/ldso/i386/sysdep.h new file mode 100644 index 000000000..9bbeef1fa --- /dev/null +++ b/ldso/ldso/i386/sysdep.h @@ -0,0 +1,82 @@ + +/* + * Various assmbly language/system dependent hacks that are required + * so that we can minimize the amount of platform specific code. + */ + +/* + * Define this if the system uses RELOCA. + */ +#undef ELF_USES_RELOCA + +/* + * Get a pointer to the argv array. On many platforms this can be just + * the address if the first argument, on other platforms we need to + * do something a little more subtle here. + */ +#define GET_ARGV(ARGVP, ARGS) ARGVP = ((unsigned int*) & ARGS) +/* + * Get the address of the Global offset table. This must be absolute, not + * relative. + */ +#define GET_GOT(X) __asm__("\tmovl %%ebx,%0\n\t" : "=a" (X)) + +/* + * Initialization sequence for a GOT. + */ +#define INIT_GOT(GOT_BASE,MODULE) \ +{ \ + GOT_BASE[2] = (int) _dl_linux_resolve; \ + GOT_BASE[1] = (int) MODULE; \ +} + +/* + * Here is a macro to perform a relocation. This is only used when + * bootstrapping the dynamic loader. RELP is the relocation that we + * are performing, REL is the pointer to the address we are relocating. + * SYMBOL is the symbol involved in the relocation, and LOAD is the + * load address. + */ +#define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD) \ + switch(ELF32_R_TYPE((RELP)->r_info)){ \ + case R_386_32: \ + *REL += SYMBOL; \ + break; \ + case R_386_PC32: \ + *REL += SYMBOL - (unsigned int) REL; \ + break; \ + case R_386_GLOB_DAT: \ + case R_386_JMP_SLOT: \ + *REL = SYMBOL; \ + break; \ + case R_386_RELATIVE: \ + *REL += (unsigned int) LOAD; \ + break; \ + default: \ + _dl_exit(1); \ + } + + +/* + * Transfer control to the user's application, once the dynamic loader + * is done. + */ + +#define START() \ + __asm__ volatile ("leave\n\t" \ + "jmp *%%eax\n\t" \ + : "=a" (status) : \ + "d" (_dl_interpreter_exit), "a" (_dl_elf_main)) + + + +/* Here we define the magic numbers that this dynamic loader should accept */ + +#define MAGIC1 EM_386 +#define MAGIC2 EM_486 +/* Used for error messages */ +#define ELF_TARGET "386/486" + +extern unsigned int _dl_linux_resolver(int dummy, int i); + +#define do_rem(result, n, base) result = (n % base) diff --git a/ldso/ldso/ld-uClibc.c b/ldso/ldso/ld-uClibc.c new file mode 100644 index 000000000..6b0b864cb --- /dev/null +++ b/ldso/ldso/ld-uClibc.c @@ -0,0 +1,1005 @@ +/* Run an ELF binary on a linux system. + + Copyright (C) 1993-1996, Eric Youngdale. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ + + + +/* Program to load an ELF binary on a linux system, and run it. + * References to symbols in sharable libraries can be resolved by + * an ELF sharable library. */ + +/* Disclaimer: I have never seen any AT&T source code for SVr4, nor have + I ever taken any courses on internals. This program was developed using + information available through the book "UNIX SYSTEM V RELEASE 4, + Programmers guide: Ansi C and Programming Support Tools", which did + a more than adequate job of explaining everything required to get this + working. */ + +/* + * The main trick with this program is that initially, we ourselves are not + * dynamicly linked. This means that we cannot access any global variables + * since the GOT is initialized by the linker assuming a virtual address of 0, + * and we cannot call any functions since the PLT is not initialized at all + * (it will tend to want to call the dynamic linker + * + * There are further restrictions - we cannot use large switch statements, + * since the compiler generates tables of addresses and jumps through them. + * We can use inline functions, because these do not transfer control to + * a new address, but they must be static so that they are not exported + * from the modules. We cannot use normal syscall stubs, because these + * all reference the errno global variable which is not yet initialized. + * We can use all of the local stack variables that we want, since these + * are all referenced to %ebp or %esp. + * + * Life is further complicated by the fact that initially we do not want + * to do a complete dynamic linking. We want to allow the user to supply + * new functions replacing some of the library versions, and until we have + * the list of modules that we should search set up, we do not want to do + * any of this. Thus I have chosen to only perform the relocations for + * variables that start with "_dl_" since ANSI specifies that the user is + * not supposed to redefine any of these variables. + * + * Fortunately, the linker itself leaves a few clues lying around, and + * when the kernel starts the image, there are a few further clues. + * First of all, there is information buried on the stack that the kernel + * leaves, which includes information about the load address that the + * program interpreter was loaded at, the number of sections, the address + * the application was loaded at and so forth. Here this information + * is stored in the array dl_info, and the indicies are taken from the + * file /usr/include/sys/auxv.h on any SVr4 system. + * + * The linker itself leaves a pointer to the .dynamic section in the first + * slot of the GOT, and as it turns out, %ebx points to ghe GOT when + * you are using PIC code, so we just dereference this to get the address + * of the dynamic sections. + * + * Typically you must load all text pages as writable so that dynamic linking + * can succeed. The kernel under SVr4 loads these as R/O, so we must call + * mprotect to change the protections. Once we are done, we should set these + * back again, so the desired behavior is achieved. Under linux there is + * currently no mprotect function in the distribution kernel (although + * someone has alpha patches), so for now everything is loaded writable. + * + * We do not have access to malloc and friends at the initial stages of dynamic + * linking, and it would be handy to have some scratchpad memory available + * for use as we set things up. It is a bit of a kluge, but we mmap /dev/zero + * to get one page of scratchpad. A simpleminded _dl_malloc is provided so + * that we have some memory that can be used for this purpose. Typically + * we would not want to use the same memory pool as malloc anyway - the user + * might want to redefine malloc for example. + * + * Our first task is to perform a minimal linking so that we can call other + * portions of the dynamic linker. Once we have done this, we then build + * the list of modules that the application requires, using LD_LIBRARY_PATH + * if this is not a suid program (/usr/lib otherwise). Once this is done, + * we can do the dynamic linking as required (and we must omit the things + * we did to get the dynamic linker up and running in the first place. + * After we have done this, we just have a few housekeeping chores and we + * can transfer control to the user's application. + */ + +#include <stdarg.h> +#include <linux/types.h> +#include <linux/fcntl.h> +#include <linux/unistd.h> +#include <linux/elf.h> +#include <linux/mman.h> +#include "link.h" + +#include "sysdep.h" +#include "hash.h" +#include "linuxelf.h" +#include "syscall.h" +#include "string.h" + +#include "../config.h" + +#define ALLOW_ZERO_PLTGOT + +static char * _dl_malloc_addr, *_dl_mmap_zero; +char * _dl_library_path = 0; /* Where we look for libraries */ +char *_dl_preload = 0; /* Things to be loaded before the libs. */ +char *_dl_progname = "/lib/ld-linux-uclibc.so.1"; +static char * _dl_not_lazy = 0; +static char * _dl_warn = 0; /* Used by ldd */ +static char * _dl_trace_loaded_objects = 0; +static int (*_dl_elf_main)(int, char **, char**); + +static int (*_dl_elf_init)(void); + +void * (*_dl_malloc_function)(int size) = NULL; + +struct r_debug * _dl_debug_addr = NULL; + +unsigned int * _dl_brkp; + +unsigned int * _dl_envp; + +#define DL_MALLOC(SIZE) ((void *) (malloc_buffer += SIZE, malloc_buffer - SIZE)) +/* + * Make sure that the malloc buffer is aligned on 4 byte boundary. For 64 bit + * platforms we may need to increase this to 8, but this is good enough for + * now. This is typically called after DL_MALLOC. + */ +#define REALIGN() malloc_buffer = (char *) (((unsigned int) malloc_buffer + 3) & ~(3)) + + + +#define ELF_HASH(RESULT,NAME) { \ + unsigned long hash = 0; \ + unsigned long tmp; \ + char * name = NAME; \ + while (*name){ \ + hash = (hash << 4) + *name++; \ + if((tmp = hash & 0xf0000000)) hash ^= tmp >> 24; \ + hash &= ~tmp; \ + } \ + RESULT = hash; \ +} +extern int _dl_linux_resolve(void); +extern int _dl_interpreter_exit(int); +extern char * _dl_strdup(const char *); +extern char * _dl_getenv(char * symbol, char ** envp); +extern void _dl_unsetenv(char * symbol, char ** envp); +extern int _dl_fixup(struct elf_resolve * tpnt); + +/* + * Datatype of a relocation on this platform + */ +#ifdef ELF_USES_RELOCA +typedef struct elf32_rela ELF_RELOC; +#else +typedef struct elf32_rel ELF_RELOC; +#endif + +/* + * This stub function is used by some debuggers. The idea is that they + * can set an internal breakpoint on it, so that we are notified when the + * address mapping is changed in some way. + */ +void _dl_debug_state() +{ + return; +} + +void _dl_boot(int args); + +void _dl_boot(int args){ + unsigned int argc; + char ** argv, ** envp; + int status; + + unsigned int load_addr; + unsigned int * got; + unsigned int * aux_dat; + int goof = 0; + struct elfhdr * header; + struct elf_resolve * tpnt; + struct dyn_elf * rpnt; + struct elf_resolve * app_tpnt; + unsigned int brk_addr; + unsigned int dl_data[AT_EGID+1]; + unsigned char * malloc_buffer, *mmap_zero; + int (*_dl_atexit)(void *); + int * lpnt; + struct dynamic * dpnt; + unsigned int *hash_addr; + struct r_debug * debug_addr; + unsigned int *chains; + int indx; + int _dl_secure; + + /* First obtain the information on the stack that tells us more about + what binary is loaded, where it is loaded, etc, etc */ + + GET_ARGV(aux_dat, args); + argc = *(aux_dat - 1); + argv = (char **) aux_dat; + aux_dat += argc; /* Skip over the argv pointers */ + aux_dat++; /* Skip over NULL at end of argv */ + envp = (char **) aux_dat; + while(*aux_dat) aux_dat++; /* Skip over the envp pointers */ + aux_dat++; /* Skip over NULL at end of envp */ + dl_data[AT_UID] = -1; /* check later to see if it is changed */ + while(*aux_dat) + { + unsigned int * ad1; + ad1 = aux_dat + 1; + if( *aux_dat <= AT_EGID ) dl_data[*aux_dat] = *ad1; + aux_dat += 2; + } + + /* Next, locate the GOT */ + + load_addr = dl_data[AT_BASE]; + + GET_GOT(got); + dpnt = (struct dynamic *) (*got + load_addr); + + /* OK, time for another hack. Now call mmap to get a page of writable + memory that can be used for a temporary malloc. We do not know brk + yet, so we cannot use real malloc. */ + + { + /* This hack is to work around a suspected asm bug in gcc-2.7.0 */ + int zfileno; +#define ZFILENO ((-1 & (~zfileno)) | zfileno) +/*#define ZFILENO -1*/ + +#ifndef MAP_ANONYMOUS +#ifdef __sparc__ +#define MAP_ANONYMOUS 0x20 +#else +#error MAP_ANONYMOUS not defined and suplementary value not known +#endif +#endif + + /* See if we need to relocate this address */ + mmap_zero = malloc_buffer = (unsigned char *) _dl_mmap((void*) 0, 4096, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, ZFILENO, 0); + if(_dl_mmap_check_error(mmap_zero)) { + SEND_STDERR("dl_boot: mmap of /dev/zero failed!\n"); + _dl_exit(13); + } + } + + tpnt = DL_MALLOC(sizeof(struct elf_resolve)); + REALIGN(); + _dl_memset (tpnt, 0, sizeof (*tpnt)); + app_tpnt = DL_MALLOC(sizeof(struct elf_resolve)); + REALIGN(); + _dl_memset (app_tpnt, 0, sizeof (*app_tpnt)); + + /* + * This is used by gdb to locate the chain of shared libraries that are currently loaded. + */ + debug_addr = DL_MALLOC(sizeof(struct r_debug)); + REALIGN(); + _dl_memset (debug_addr, 0, sizeof (*debug_addr)); + + /* OK, that was easy. Next scan the DYNAMIC section of the image. + We are only doing ourself right now - we will have to do the rest later */ + + while(dpnt->d_tag) + { + tpnt->dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val; + if(dpnt->d_tag == DT_TEXTREL || + SVR4_BUGCOMPAT) tpnt->dynamic_info[DT_TEXTREL] = 1; + dpnt++; + } + + { + struct elf_phdr * ppnt; + int i; + + ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + for(i=0; i<dl_data[AT_PHNUM]; i++, ppnt++) + if(ppnt->p_type == PT_DYNAMIC) { + dpnt = (struct dynamic *) ppnt->p_vaddr; + while(dpnt->d_tag) + { + if(dpnt->d_tag > DT_JMPREL) {dpnt++; continue; } + app_tpnt->dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val; + if(dpnt->d_tag == DT_DEBUG) dpnt->d_un.d_val = (int) debug_addr; + if(dpnt->d_tag == DT_TEXTREL || + SVR4_BUGCOMPAT) app_tpnt->dynamic_info[DT_TEXTREL] = 1; + dpnt++; + } + } + } + + /* Get some more of the information that we will need to dynamicly link + this module to itself */ + + hash_addr = (unsigned int *) (tpnt->dynamic_info[DT_HASH]+load_addr); + tpnt->nbucket = *hash_addr++; + tpnt->nchain = *hash_addr++; + tpnt->elf_buckets = hash_addr; + hash_addr += tpnt->nbucket; + chains = hash_addr; + + /* Ugly, ugly. We need to call mprotect to change the protection of + the text pages so that we can do the dynamic linking. We can set the + protection back again once we are done */ + + { + struct elf_phdr * ppnt; + int i; + + /* First cover the shared library/dynamic linker. */ + if(tpnt->dynamic_info[DT_TEXTREL]) { + header = (struct elfhdr *) dl_data[AT_BASE]; + ppnt = (struct elf_phdr *) (dl_data[AT_BASE] + header->e_phoff); + for(i=0; i<header->e_phnum ; i++, ppnt++) { + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W)) + _dl_mprotect((void *) (load_addr + (ppnt->p_vaddr & 0xfffff000)), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + PROT_READ | PROT_WRITE | PROT_EXEC); + } + } + + /* Now cover the application program. */ + if(app_tpnt->dynamic_info[DT_TEXTREL]) { + ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + for(i=0; i<dl_data[AT_PHNUM]; i++, ppnt++) { + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W)) + _dl_mprotect((void *) (ppnt->p_vaddr & 0xfffff000), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + PROT_READ | PROT_WRITE | PROT_EXEC); + } + } + } + + /* OK, now do the relocations. We do not do a lazy binding here, so + that once we are done, we have considerably more flexibility. */ + + goof = 0; + for(indx=0; indx < 2; indx++) + { + int i; + ELF_RELOC * rpnt; + unsigned int * reloc_addr; + unsigned int symbol_addr; + int symtab_index; + unsigned int rel_addr, rel_size; + + +#ifdef ELF_USES_RELOCA + rel_addr = (indx ? tpnt->dynamic_info[DT_JMPREL] : tpnt->dynamic_info[DT_RELA]); + rel_size = (indx ? tpnt->dynamic_info[DT_PLTRELSZ] : tpnt->dynamic_info[DT_RELASZ]); +#else + rel_addr = (indx ? tpnt->dynamic_info[DT_JMPREL] : tpnt->dynamic_info[DT_REL]); + rel_size = (indx ? tpnt->dynamic_info[DT_PLTRELSZ] : tpnt->dynamic_info[DT_RELSZ]); +#endif + + + if(!rel_addr) continue; + + /* Now parse the relocation information */ + rpnt = (ELF_RELOC *) (rel_addr + load_addr); + for(i=0; i< rel_size; i+=sizeof(ELF_RELOC), rpnt++){ + reloc_addr = (int *) (load_addr + (int)rpnt->r_offset); + symtab_index = ELF32_R_SYM(rpnt->r_info); + symbol_addr = 0; + if(symtab_index) { + char * strtab; + struct elf32_sym * symtab; + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB]+load_addr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB]+load_addr); + + /* We only do a partial dynamic linking right now. The user + is not supposed to redefine any symbols that start with + a '_', so we can do this with confidence. */ + + if (!_dl_symbol(strtab + symtab[symtab_index].st_name)) continue; + + symbol_addr = load_addr + symtab[symtab_index].st_value; + + if(!symbol_addr) { + /* + * This will segfault - you cannot call a function until + * we have finished the relocations. + */ + SEND_STDERR("ELF dynamic loader - unable to self-bootstrap - symbol "); + SEND_STDERR(strtab + symtab[symtab_index].st_name); + SEND_STDERR(" undefined.\n"); + goof++; + } + } + /* + * Use this machine-specific macro to perform the actual relocation. + */ + PERFORM_BOOTSTRAP_RELOC(rpnt, reloc_addr, symbol_addr, load_addr); + } + } + + if (goof) _dl_exit(14); + + /* OK, at this point we have a crude malloc capability. Start to build + the tables of the modules that are required for this beast to run. + We start with the basic executable, and then go from there. Eventually + we will run across ourself, and we will need to properly deal with that + as well. */ + + _dl_malloc_addr = malloc_buffer; + + _dl_mmap_zero = mmap_zero; +/* tpnt = _dl_malloc(sizeof(struct elf_resolve)); */ + +/* Now we have done the mandatory linking of some things. We are now + free to start using global variables, since these things have all been + fixed up by now. Still no function calls outside of this library , + since the dynamic resolver is not yet ready. */ + + lpnt = (int *) (tpnt->dynamic_info[DT_PLTGOT] + load_addr); + INIT_GOT(lpnt, tpnt); + + /* OK, this was a big step, now we need to scan all of the user images + and load them properly. */ + + tpnt->next = 0; + tpnt->libname = 0; + tpnt->libtype = program_interpreter; + + { struct elfhdr * epnt; + struct elf_phdr * ppnt; + int i; + + epnt = (struct elfhdr *) dl_data[AT_BASE]; + tpnt->n_phent = epnt->e_phnum; + tpnt->ppnt = ppnt = (struct elf_phdr *) (load_addr + epnt->e_phoff); + for(i=0;i < epnt->e_phnum; i++, ppnt++){ + if(ppnt->p_type == PT_DYNAMIC) { + tpnt->dynamic_addr = ppnt->p_vaddr + load_addr; + tpnt->dynamic_size = ppnt->p_filesz; + } + } + } + + tpnt->chains = chains; + tpnt->loadaddr = (char *) load_addr; + + brk_addr = 0; + rpnt = NULL; + + /* At this point we are now free to examine the user application, + and figure out which libraries are supposed to be called. Until + we have this list, we will not be completely ready for dynamic linking */ + + { + struct elf_phdr * ppnt; + int i; + + ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + for(i=0; i<dl_data[AT_PHNUM]; i++, ppnt++) { + if(ppnt->p_type == PT_LOAD) { + if(ppnt->p_vaddr + ppnt->p_memsz > brk_addr) + brk_addr = ppnt->p_vaddr + ppnt->p_memsz; + } + if(ppnt->p_type == PT_DYNAMIC) { +#ifndef ALLOW_ZERO_PLTGOT + /* make sure it's really there. */ + if (app_tpnt->dynamic_info[DT_PLTGOT] == 0) continue; +#endif + /* OK, we have what we need - slip this one into the list. */ + app_tpnt = _dl_add_elf_hash_table("", 0, + app_tpnt->dynamic_info, ppnt->p_vaddr, ppnt->p_filesz); + _dl_loaded_modules->libtype = elf_executable; + _dl_loaded_modules->ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + _dl_loaded_modules->n_phent = dl_data[AT_PHNUM]; + _dl_symbol_tables = rpnt = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt, 0, sizeof (*rpnt)); + rpnt->dyn = _dl_loaded_modules; + app_tpnt->usage_count++; + app_tpnt->symbol_scope = _dl_symbol_tables; + lpnt = (int *) (app_tpnt->dynamic_info[DT_PLTGOT]); +#ifdef ALLOW_ZERO_PLTGOT + if (lpnt) +#endif + INIT_GOT(lpnt, _dl_loaded_modules); + } + if(ppnt->p_type == PT_INTERP) { /* OK, fill this in - we did not have + this before */ + tpnt->libname = _dl_strdup((char *) ppnt->p_offset +(dl_data[AT_PHDR] & 0xfffff000)); + } + } + } + + if (argv[0]) + _dl_progname = argv[0]; + + /* Now we need to figure out what kind of options are selected. + Note that for SUID programs we ignore the settings in LD_LIBRARY_PATH */ + { + _dl_not_lazy = _dl_getenv("LD_BIND_NOW",envp); + + if ( (dl_data[AT_UID] == -1 && _dl_suid_ok()) || + (dl_data[AT_UID] != -1 && dl_data[AT_UID] == dl_data[AT_EUID] && + dl_data[AT_GID] == dl_data[AT_EGID])) + { + _dl_secure = 0; + _dl_preload = _dl_getenv("LD_PRELOAD", envp); + _dl_library_path = _dl_getenv("LD_LIBRARY_PATH",envp); + } + else + { + _dl_secure = 1; + _dl_preload = _dl_getenv("LD_PRELOAD", envp); + _dl_unsetenv("LD_AOUT_PRELOAD", envp); + _dl_unsetenv("LD_LIBRARY_PATH", envp); + _dl_unsetenv("LD_AOUT_LIBRARY_PATH", envp); + _dl_library_path = NULL; + } + } + + _dl_trace_loaded_objects = _dl_getenv("LD_TRACE_LOADED_OBJECTS", envp); + + /* OK, we now have the application in the list, and we have some + basic stuff in place. Now search through the list for other shared + libraries that should be loaded, and insert them on the list in the + correct order. */ + +#ifdef USE_CACHE + _dl_map_cache(); +#endif + + { + struct elf_resolve *tcurr; + struct elf_resolve *tpnt1; + char *lpnt; + + if (_dl_preload) { + char c, *str, *str2; + + str = _dl_preload; + while (*str == ':' || *str == ' ' || *str == '\t') + str++; + while (*str) { + str2 = str; + while (*str2 && *str2 != ':' && *str2 != ' ' && *str2 != '\t') + str2++; + c = *str2; + *str2 = '\0'; + if (!_dl_secure || _dl_strchr(str, '/') == NULL) { + tpnt1 = _dl_load_shared_library(_dl_secure, NULL, str); + if (!tpnt1) { + if (_dl_trace_loaded_objects) + _dl_fdprintf(1, "\t%s => not found\n", str); + else { + _dl_fdprintf(2, "%s: can't load library '%s'\n", + _dl_progname, str); + _dl_exit(15); + } + } else { + if (_dl_trace_loaded_objects && !tpnt1->usage_count) { + /* this is a real hack to make ldd not print the + library itself when run on a library. */ + if (_dl_strcmp(_dl_progname, str) != 0) + _dl_fdprintf(1, "\t%s => %s (0x%x)\n", str, tpnt1->libname, + (unsigned)tpnt1->loadaddr); + } + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + tpnt1->usage_count++; + tpnt1->symbol_scope = _dl_symbol_tables; + tpnt1->libtype = elf_lib; + rpnt->dyn = tpnt1; + } + } + *str2 = c; + str = str2; + while (*str == ':' || *str == ' ' || *str == '\t') + str++; + } + } + + { + int fd; + struct kernel_stat st; + char *preload; + + if (!_dl_stat(LDSO_PRELOAD, &st)) { + if ((fd = _dl_open(LDSO_PRELOAD, O_RDONLY)) < 0) { + _dl_fdprintf(2, "%s: can't open file '%s'\n", _dl_progname, + LDSO_PRELOAD); + } else { + preload = (caddr_t)_dl_mmap(0, st.st_size+1, PROT_READ|PROT_WRITE, + MAP_PRIVATE, fd, 0); + _dl_close (fd); + if (preload == (caddr_t)-1) { + _dl_fdprintf(2, "%s: can't map file '%s'\n", _dl_progname, + LDSO_PRELOAD); + } else { + char c, *cp, *cp2; + + /* convert all separators and comments to spaces */ + for (cp = preload; *cp; /*nada*/) { + if (*cp == ':' || *cp == '\t' || *cp == '\n') { + *cp++ = ' '; + } else if (*cp == '#') { + do + *cp++ = ' '; + while (*cp != '\n' && *cp != '\0'); + } else { + cp++; + } + } + + /* find start of first library */ + for (cp = preload; *cp && *cp == ' '; cp++) + /*nada*/; + + while (*cp) { + /* find end of library */ + for (cp2 = cp; *cp && *cp != ' '; cp++) + /*nada*/; + c = *cp; + *cp = '\0'; + + tpnt1 = _dl_load_shared_library(0, NULL, cp2); + if (!tpnt1) { + if (_dl_trace_loaded_objects) + _dl_fdprintf(1, "\t%s => not found\n", cp2); + else { + _dl_fdprintf(2, "%s: can't load library '%s'\n", + _dl_progname, cp2); + _dl_exit(15); + } + } else { + if (_dl_trace_loaded_objects && !tpnt1->usage_count) + _dl_fdprintf(1, "\t%s => %s (0x%x)\n", cp2, tpnt1->libname, + (unsigned)tpnt1->loadaddr); + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + tpnt1->usage_count++; + tpnt1->symbol_scope = _dl_symbol_tables; + tpnt1->libtype = elf_lib; + rpnt->dyn = tpnt1; + } + + /* find start of next library */ + *cp = c; + for (/*nada*/; *cp && *cp == ' '; cp++) + /*nada*/; + } + + _dl_munmap(preload, st.st_size+1); + } + } + } + } + + for (tcurr = _dl_loaded_modules; tcurr; tcurr = tcurr->next) + { + for (dpnt = (struct dynamic *)tcurr->dynamic_addr; dpnt->d_tag; dpnt++) + { + if(dpnt->d_tag == DT_NEEDED) + { + lpnt = tcurr->loadaddr + tcurr->dynamic_info[DT_STRTAB] + + dpnt->d_un.d_val; + if (tpnt && _dl_strcmp(lpnt, tpnt->libname) == 0) + { + struct elf_resolve * ttmp; + ttmp = _dl_loaded_modules; + while (ttmp->next) + ttmp = ttmp->next; + ttmp->next = tpnt; + tpnt->prev = ttmp; + tpnt->next = NULL; + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + rpnt->dyn = tpnt; + tpnt->usage_count++; + tpnt->symbol_scope = _dl_symbol_tables; + tpnt = NULL; + continue; + } + if (!(tpnt1 = _dl_load_shared_library(0, tcurr, lpnt))) + { + if (_dl_trace_loaded_objects) + _dl_fdprintf(1, "\t%s => not found\n", lpnt); + else + { + _dl_fdprintf(2, "%s: can't load library '%s'\n", + _dl_progname, lpnt); + _dl_exit(16); + } + } + else + { + if (_dl_trace_loaded_objects && !tpnt1->usage_count) + _dl_fdprintf(1, "\t%s => %s (0x%x)\n", lpnt, tpnt1->libname, + (unsigned)tpnt1->loadaddr); + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + tpnt1->usage_count++; + tpnt1->symbol_scope = _dl_symbol_tables; + tpnt1->libtype = elf_lib; + rpnt->dyn = tpnt1; + } + } + } + } + } + +#ifdef USE_CACHE + _dl_unmap_cache(); +#endif + + /* ldd uses uses this. I am not sure how you pick up the other flags */ + if(_dl_trace_loaded_objects) + { + _dl_warn = _dl_getenv("LD_WARN", envp); + if (!_dl_warn) _dl_exit(0); + } + + /* + * If the program interpreter is not in the module chain, add it. This will + * be required for dlopen to be able to access the internal functions in the + * dynamic linker. + */ + if(tpnt) { + struct elf_resolve * tcurr; + + tcurr = _dl_loaded_modules; + if (tcurr) + while(tcurr->next) tcurr = tcurr->next; + tpnt->next = NULL; + tpnt->usage_count++; + + if (tcurr) { + tcurr->next = tpnt; + tpnt->prev = tcurr; + } + else { + _dl_loaded_modules = tpnt; + tpnt->prev = NULL; + } + if (rpnt) { + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + } else { + rpnt = (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt, 0, sizeof (*(rpnt->next))); + } + rpnt->dyn = tpnt; + tpnt = NULL; + } + + /* + * OK, now all of the kids are tucked into bed in their proper addresses. + * Now we go through and look for REL and RELA records that indicate fixups + * to the GOT tables. We need to do this in reverse order so that COPY + * directives work correctly */ + + + goof = _dl_loaded_modules ? _dl_fixup(_dl_loaded_modules) : 0; + + + /* Some flavors of SVr4 do not generate the R_*_COPY directive, + and we have to manually search for entries that require fixups. + Solaris gets this one right, from what I understand. */ + + + if (_dl_symbol_tables) + goof += _dl_copy_fixups(_dl_symbol_tables); + + if(goof || _dl_trace_loaded_objects) _dl_exit(0); + + /* OK, at this point things are pretty much ready to run. Now we + need to touch up a few items that are required, and then + we can let the user application have at it. Note that + the dynamic linker itself is not guaranteed to be fully + dynamicly linked if we are using ld.so.1, so we have to look + up each symbol individually. */ + + + _dl_brkp = (unsigned int *) _dl_find_hash("___brk_addr", NULL, 1, NULL, 0); + if (_dl_brkp) *_dl_brkp = brk_addr; + _dl_envp = (unsigned int *) _dl_find_hash("__environ", NULL, 1, NULL, 0); + + if (_dl_envp) *_dl_envp = (unsigned int) envp; + + { + int i; + struct elf_phdr * ppnt; + + /* We had to set the protections of all pages to R/W for dynamic linking. + Set text pages back to R/O */ + for(tpnt = _dl_loaded_modules; tpnt; tpnt = tpnt->next) + for(ppnt = tpnt->ppnt, i=0; i < tpnt->n_phent; i++, ppnt++) + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W) && + tpnt->dynamic_info[DT_TEXTREL]) + _dl_mprotect((void *) (tpnt->loadaddr + (ppnt->p_vaddr & 0xfffff000)), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + LXFLAGS(ppnt->p_flags)); + + } + + _dl_atexit = (int (*)(void *)) _dl_find_hash("atexit", NULL, 1, NULL, 0); + + /* + * OK, fix one more thing - set up the debug_addr structure to point + * to our chain. Later we may need to fill in more fields, but this + * should be enough for now. + */ + debug_addr->r_map = (struct link_map *) _dl_loaded_modules; + debug_addr->r_version = 1; + debug_addr->r_ldbase = load_addr; + debug_addr->r_brk = (unsigned long) &_dl_debug_state; + _dl_debug_addr = debug_addr; + debug_addr->r_state = RT_CONSISTENT; + /* This is written in this funny way to keep gcc from inlining the + function call. */ + ((void (*)(void))debug_addr->r_brk)(); + + for(tpnt = _dl_loaded_modules; tpnt; tpnt = tpnt->next) + { + /* Apparently crt1 for the application is responsible for handling this. + * We only need to run the init/fini for shared libraries + */ + if (tpnt->libtype == program_interpreter || + tpnt->libtype == elf_executable) continue; + if (tpnt->init_flag & INIT_FUNCS_CALLED) continue; + tpnt->init_flag |= INIT_FUNCS_CALLED; + + if(tpnt->dynamic_info[DT_INIT]) { + _dl_elf_init = (int (*)(void)) (tpnt->loadaddr + + tpnt->dynamic_info[DT_INIT]); + (*_dl_elf_init)(); + } + if(_dl_atexit && tpnt->dynamic_info[DT_FINI]) + { + (*_dl_atexit)(tpnt->loadaddr + tpnt->dynamic_info[DT_FINI]); + } +#undef DL_DEBUG +#ifdef DL_DEBUG + else + { + _dl_fdprintf(2, tpnt->libname); + _dl_fdprintf(2, ": "); + if (!_dl_atexit) + _dl_fdprintf(2, "The address is atexit () is 0x0."); + if (!tpnt->dynamic_info[DT_FINI]) + _dl_fdprintf(2, "Invalid .fini section."); + _dl_fdprintf(2, "\n"); + } +#endif +#undef DL_DEBUG + } + + /* OK we are done here. Turn out the lights, and lock up. */ + _dl_elf_main = (int (*)(int, char**, char**)) dl_data[AT_ENTRY]; + + + /* + * Transfer control to the application. + */ + START(); +} + +int _dl_fixup(struct elf_resolve * tpnt) +{ + int goof = 0; + if(tpnt->next) goof += _dl_fixup(tpnt->next); + + if(tpnt->dynamic_info[DT_REL]) { +#ifdef ELF_USES_RELOCA + _dl_fdprintf(2, "%s: can't handle REL relocation records\n", _dl_progname); + _dl_exit(17); +#else + if (tpnt->init_flag & RELOCS_DONE) return goof; + tpnt->init_flag |= RELOCS_DONE; + + goof += _dl_parse_relocation_information(tpnt, tpnt->dynamic_info[DT_REL], + tpnt->dynamic_info[DT_RELSZ], 0); +#endif + } + if(tpnt->dynamic_info[DT_RELA]) { +#ifdef ELF_USES_RELOCA + if (tpnt->init_flag & RELOCS_DONE) return goof; + tpnt->init_flag |= RELOCS_DONE; + + goof += _dl_parse_relocation_information(tpnt, tpnt->dynamic_info[DT_RELA], + tpnt->dynamic_info[DT_RELASZ], 0); +#else + _dl_fdprintf(2, "%s: can't handle RELA relocation records\n", _dl_progname); + _dl_exit(18); +#endif + } + if(tpnt->dynamic_info[DT_JMPREL]) + { + if (tpnt->init_flag & JMP_RELOCS_DONE) return goof; + tpnt->init_flag |= JMP_RELOCS_DONE; + + if(! _dl_not_lazy || *_dl_not_lazy == 0) + _dl_parse_lazy_relocation_information(tpnt, tpnt->dynamic_info[DT_JMPREL], + tpnt->dynamic_info[DT_PLTRELSZ], 0); + else + goof += _dl_parse_relocation_information(tpnt, + tpnt->dynamic_info[DT_JMPREL], + tpnt->dynamic_info[DT_PLTRELSZ], 0); + } + return goof; +} + +void * _dl_malloc(int size) { + void * retval; + + if(_dl_malloc_function) + return (*_dl_malloc_function)(size); + + if(_dl_malloc_addr-_dl_mmap_zero+size>4096) { + _dl_mmap_zero = _dl_malloc_addr = (unsigned char *) _dl_mmap((void*) 0, size, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + if(_dl_mmap_check_error(_dl_mmap_zero)) { + _dl_fdprintf(2, "%s: can't map '/dev/zero'\n", _dl_progname); + _dl_exit(20); + } + } + retval = _dl_malloc_addr; + _dl_malloc_addr += size; + + /* + * Align memory to 4 byte boundary. Some platforms require this, others + * simply get better performance. + */ + _dl_malloc_addr = (char *) (((unsigned int) _dl_malloc_addr + 3) & ~(3)); + return retval; +} + +char * _dl_getenv(char *symbol, char **envp) +{ + char *pnt; + char *pnt1; + while ((pnt = *envp++)) { + pnt1 = symbol; + while (*pnt && *pnt == *pnt1) + pnt1++, pnt++; + if (!*pnt || *pnt != '=' || *pnt1) + continue; + return pnt+1; + } + return 0; +} + +void _dl_unsetenv(char *symbol, char **envp) +{ + char *pnt; + char *pnt1; + char **newenvp = envp; + for (pnt = *envp; pnt; pnt = *++envp) { + pnt1 = symbol; + while (*pnt && *pnt == *pnt1) + pnt1++, pnt++; + if(!*pnt || *pnt != '=' || *pnt1) + *newenvp++ = *envp; + } + *newenvp++ = *envp; + return; +} + +char * _dl_strdup(const char * string){ + char * retval; + int len; + + len = _dl_strlen(string); + retval = _dl_malloc(len + 1); + _dl_strcpy(retval, string); + return retval; +} + +/* In principle we could do the .fini stuff here, but we already + registered this stuff with atexit */ +int _dl_interpreter_exit(int exitcode){ +/* _dl_fdprintf(2, "Hey, look where I am!\n"); */ + return 0; +} diff --git a/ldso/ldso/ld_hash.h b/ldso/ldso/ld_hash.h new file mode 100644 index 000000000..2eeda2d46 --- /dev/null +++ b/ldso/ldso/ld_hash.h @@ -0,0 +1,113 @@ +#include "link.h" + +#ifndef RTLD_NEXT +#define RTLD_NEXT ((void*)-1) +#endif + +struct dyn_elf{ + unsigned int flags; + struct elf_resolve * dyn; + struct dyn_elf * next_handle; /* Used by dlopen et al. */ + struct dyn_elf * next; +}; + +struct elf_resolve{ + /* These entries must be in this order to be compatible with the interface used + by gdb to obtain the list of symbols. */ + char * loadaddr; + char * libname; + unsigned int dynamic_addr; + struct elf_resolve * next; + struct elf_resolve * prev; + /* Nothing after this address is used by gdb. */ + enum {elf_lib, elf_executable,program_interpreter, loaded_file} libtype; + struct dyn_elf * symbol_scope; + unsigned short usage_count; + unsigned short int init_flag; + unsigned int nbucket; + unsigned int * elf_buckets; + /* + * These are only used with ELF style shared libraries + */ + unsigned int nchain; + unsigned int * chains; + unsigned int dynamic_info[24]; + + unsigned int dynamic_size; + unsigned int n_phent; + struct elf_phdr * ppnt; +}; + +#if 0 +/* + * The DT_DEBUG entry in the .dynamic section is given the address of this structure. + * gdb can pick this up to obtain the correct list of loaded modules. + */ + +struct r_debug{ + int r_version; + struct elf_resolve * link_map; + unsigned long brk_fun; + enum {RT_CONSISTENT, RT_ADD, RT_DELETE}; + unsigned long ldbase; +}; +#endif + +#define COPY_RELOCS_DONE 1 +#define RELOCS_DONE 2 +#define JMP_RELOCS_DONE 4 +#define INIT_FUNCS_CALLED 8 + +extern struct dyn_elf * _dl_symbol_tables; +extern struct elf_resolve * _dl_loaded_modules; +extern struct dyn_elf * _dl_handles; + +extern struct elf_resolve * _dl_check_hashed_files(char * libname); +extern struct elf_resolve * _dl_add_elf_hash_table(char * libname, + char * loadaddr, + unsigned int * dynamic_info, + unsigned int dynamic_addr, + unsigned int dynamic_size); +extern char * _dl_find_hash(char * name, struct dyn_elf * rpnt1, + unsigned int instr_addr, + struct elf_resolve * f_tpnt, + int copyrel); + +extern int _dl_linux_dynamic_link(void); + +#ifdef __mc68000__ +/* On m68k constant strings are referenced through the GOT. */ +/* XXX Requires load_addr to be defined. */ +#define SEND_STDERR(X) \ + { const char *__s = (X); \ + if (__s < (const char *) load_addr) __s += load_addr; \ + _dl_write (2, __s, _dl_strlen (__s)); \ + } +#else +#define SEND_STDERR(X) _dl_write(2, X, _dl_strlen(X)); +#endif +extern int _dl_fdprintf(int, const char *, ...); +extern char * _dl_library_path; +extern char * _dl_not_lazy; +extern char * _dl_strdup(const char *); +extern inline int _dl_symbol(char * name); +unsigned long _dl_elf_hash(const char * name); + +extern inline int _dl_symbol(char * name) +{ + if(name[0] != '_' || name[1] != 'd' || name[2] != 'l' || name[3] != '_') + return 0; + return 1; +} + +#define DL_ERROR_NOFILE 1 +#define DL_ERROR_NOZERO 2 +#define DL_ERROR_NOTELF 3 +#define DL_ERROR_NOTMAGIC 4 +#define DL_ERROR_NOTDYN 5 +#define DL_ERROR_MMAP_FAILED 6 +#define DL_ERROR_NODYNAMIC 7 +#define DL_WRONG_RELOCS 8 +#define DL_BAD_HANDLE 9 +#define DL_NO_SYMBOL 10 + diff --git a/ldso/ldso/ld_string.h b/ldso/ldso/ld_string.h new file mode 100644 index 000000000..1ea8fd7ae --- /dev/null +++ b/ldso/ldso/ld_string.h @@ -0,0 +1,112 @@ +#ifndef _LINUX_STRING_H_ +#define _LINUX_STRING_H_ + +#include <linux/types.h> /* for size_t */ + +#ifndef NULL +#define NULL ((void *) 0) +#endif + +extern inline char * _dl_strcpy(char * dst,const char *src) +{ + register char *ptr = dst; + + while (*src) + *dst++ = *src++; + *dst = '\0'; + + return ptr; +} + +extern inline int _dl_strcmp(const char * s1,const char * s2) +{ + unsigned register char c1, c2; + + do { + c1 = (unsigned char) *s1++; + c2 = (unsigned char) *s2++; + if (c1 == '\0') + return c1 - c2; + } + while (c1 == c2); + + return c1 - c2; +} + +extern inline int _dl_strncmp(const char * s1,const char * s2,size_t len) +{ + unsigned register char c1 = '\0'; + unsigned register char c2 = '\0'; + + while (len > 0) { + c1 = (unsigned char) *s1++; + c2 = (unsigned char) *s2++; + if (c1 == '\0' || c1 != c2) + return c1 - c2; + len--; + } + + return c1 - c2; +} + +extern inline char * _dl_strchr(const char * str,int c) +{ + register char ch; + + do { + if ((ch = *str) == c) + return (char *) str; + str++; + } + while (ch); + + return 0; +} + + +extern inline size_t _dl_strlen(const char * str) +{ + register char *ptr = (char *) str; + + while (*ptr) + ptr++; + return (ptr - str); +} + +extern inline void * _dl_memcpy(void * dst, const void * src, size_t len) +{ + register char *a = dst; + register const char *b = src; + + while (len--) + *a++ = *b++; + + return dst; +} + + +extern inline int _dl_memcmp(const void * s1,const void * s2,size_t len) +{ + unsigned char *c1 = (unsigned char *)s1; + unsigned char *c2 = (unsigned char *)s2; + + while (len--) { + if (*c1 != *c2) + return *c1 - *c2; + c1++; + c2++; + } + return 0; +} + +extern inline void * _dl_memset(void * str,int c,size_t len) +{ + register char *a = str; + + while (len--) + *a++ = c; + + return str; +} + +#endif diff --git a/ldso/ldso/ld_syscall.h b/ldso/ldso/ld_syscall.h new file mode 100644 index 000000000..3cf244338 --- /dev/null +++ b/ldso/ldso/ld_syscall.h @@ -0,0 +1,108 @@ +#include <linux/types.h> +#include <asm/unistd.h> + +#ifndef _dl_MAX_ERRNO +#define _dl_MAX_ERRNO 4096 +#endif + +#define _dl_mmap_check_error(__res) \ + (((int)__res) < 0 && ((int)__res) >= -_dl_MAX_ERRNO) + + +/* Here are the definitions for some syscalls that are used + by the dynamic linker. The idea is that we want to be able + to call these before the errno symbol is dynamicly linked, so + we use our own version here. Note that we cannot assume any + dynamic linking at all, so we cannot return any error codes. + We just punt if there is an error. */ + +/* Do not include unistd.h, so gcc doesn't whine about + * _exit returning. It really doesn't return... */ +#define __NR__dl_exit __NR_exit +static inline _syscall1(void, _dl_exit, int, status); + + +#define __NR__dl_close __NR_close +static inline _syscall1(int, _dl_close, int, fd); + + +#define __NR__dl_mmap_real __NR_mmap +static inline _syscall1(void *, _dl_mmap_real, unsigned long *, buffer); + +static inline void * _dl_mmap(void * addr, unsigned long size, int prot, + int flags, int fd, unsigned long offset) +{ + unsigned long buffer[6]; + + buffer[0] = (unsigned long) addr; + buffer[1] = (unsigned long) size; + buffer[2] = (unsigned long) prot; + buffer[3] = (unsigned long) flags; + buffer[4] = (unsigned long) fd; + buffer[5] = (unsigned long) offset; + return (void *) _dl_mmap_real(buffer); +} + +#define __NR__dl_open __NR_open +static inline _syscall2(int, _dl_open, const char *, fn, int, flags); + +#define __NR__dl_write __NR_write +static inline _syscall3(unsigned long, _dl_write, int, fd, + const void *, buf, unsigned long, count); + + +#define __NR__dl_read __NR_read +static inline _syscall3(unsigned long, _dl_read, int, fd, + const void *, buf, unsigned long, count); + +#define __NR__dl_mprotect __NR_mprotect +static inline _syscall3(int, _dl_mprotect, const void *, addr, unsigned long, len, int, prot); + + + +/* Pull in whatever this particular arch's kernel thinks the kernel version of + * struct stat should look like. It turns out that each arch has a different + * opinion on the subject, and different kernel revs use different names... */ +#define __NR__dl_stat __NR_stat +#define stat kernel_stat +#define new_stat kernel_stat +#include <asm/stat.h> +#undef new_stat +#undef stat +static inline _syscall2(int, _dl_stat, const char *, file_name, struct kernel_stat *, buf); + + +#define __NR__dl_munmap __NR_munmap +static inline _syscall2(int, _dl_munmap, void *, start, unsigned long, length); + +#define __NR__dl_getuid __NR_getuid +static inline _syscall0(gid_t, _dl_getuid); + +#define __NR__dl_geteuid __NR_geteuid +static inline _syscall0(uid_t, _dl_geteuid); + +#define __NR__dl_getgid __NR_getgid +static inline _syscall0(gid_t, _dl_getgid); + +#define __NR__dl_getegid __NR_getegid +static inline _syscall0(gid_t, _dl_getegid); + +/* + * Not an actual syscall, but we need something in assembly to say whether + * this is OK or not. + */ +extern inline int _dl_suid_ok(void) +{ + uid_t uid, euid, gid, egid; + + uid = _dl_getuid(); + euid = _dl_geteuid(); + gid = _dl_getgid(); + egid = _dl_getegid(); + + if(uid == euid && gid == egid) + return 1; + else + return 0; +} + diff --git a/ldso/ldso/ldso.c b/ldso/ldso/ldso.c new file mode 100644 index 000000000..6b0b864cb --- /dev/null +++ b/ldso/ldso/ldso.c @@ -0,0 +1,1005 @@ +/* Run an ELF binary on a linux system. + + Copyright (C) 1993-1996, Eric Youngdale. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ + + + +/* Program to load an ELF binary on a linux system, and run it. + * References to symbols in sharable libraries can be resolved by + * an ELF sharable library. */ + +/* Disclaimer: I have never seen any AT&T source code for SVr4, nor have + I ever taken any courses on internals. This program was developed using + information available through the book "UNIX SYSTEM V RELEASE 4, + Programmers guide: Ansi C and Programming Support Tools", which did + a more than adequate job of explaining everything required to get this + working. */ + +/* + * The main trick with this program is that initially, we ourselves are not + * dynamicly linked. This means that we cannot access any global variables + * since the GOT is initialized by the linker assuming a virtual address of 0, + * and we cannot call any functions since the PLT is not initialized at all + * (it will tend to want to call the dynamic linker + * + * There are further restrictions - we cannot use large switch statements, + * since the compiler generates tables of addresses and jumps through them. + * We can use inline functions, because these do not transfer control to + * a new address, but they must be static so that they are not exported + * from the modules. We cannot use normal syscall stubs, because these + * all reference the errno global variable which is not yet initialized. + * We can use all of the local stack variables that we want, since these + * are all referenced to %ebp or %esp. + * + * Life is further complicated by the fact that initially we do not want + * to do a complete dynamic linking. We want to allow the user to supply + * new functions replacing some of the library versions, and until we have + * the list of modules that we should search set up, we do not want to do + * any of this. Thus I have chosen to only perform the relocations for + * variables that start with "_dl_" since ANSI specifies that the user is + * not supposed to redefine any of these variables. + * + * Fortunately, the linker itself leaves a few clues lying around, and + * when the kernel starts the image, there are a few further clues. + * First of all, there is information buried on the stack that the kernel + * leaves, which includes information about the load address that the + * program interpreter was loaded at, the number of sections, the address + * the application was loaded at and so forth. Here this information + * is stored in the array dl_info, and the indicies are taken from the + * file /usr/include/sys/auxv.h on any SVr4 system. + * + * The linker itself leaves a pointer to the .dynamic section in the first + * slot of the GOT, and as it turns out, %ebx points to ghe GOT when + * you are using PIC code, so we just dereference this to get the address + * of the dynamic sections. + * + * Typically you must load all text pages as writable so that dynamic linking + * can succeed. The kernel under SVr4 loads these as R/O, so we must call + * mprotect to change the protections. Once we are done, we should set these + * back again, so the desired behavior is achieved. Under linux there is + * currently no mprotect function in the distribution kernel (although + * someone has alpha patches), so for now everything is loaded writable. + * + * We do not have access to malloc and friends at the initial stages of dynamic + * linking, and it would be handy to have some scratchpad memory available + * for use as we set things up. It is a bit of a kluge, but we mmap /dev/zero + * to get one page of scratchpad. A simpleminded _dl_malloc is provided so + * that we have some memory that can be used for this purpose. Typically + * we would not want to use the same memory pool as malloc anyway - the user + * might want to redefine malloc for example. + * + * Our first task is to perform a minimal linking so that we can call other + * portions of the dynamic linker. Once we have done this, we then build + * the list of modules that the application requires, using LD_LIBRARY_PATH + * if this is not a suid program (/usr/lib otherwise). Once this is done, + * we can do the dynamic linking as required (and we must omit the things + * we did to get the dynamic linker up and running in the first place. + * After we have done this, we just have a few housekeeping chores and we + * can transfer control to the user's application. + */ + +#include <stdarg.h> +#include <linux/types.h> +#include <linux/fcntl.h> +#include <linux/unistd.h> +#include <linux/elf.h> +#include <linux/mman.h> +#include "link.h" + +#include "sysdep.h" +#include "hash.h" +#include "linuxelf.h" +#include "syscall.h" +#include "string.h" + +#include "../config.h" + +#define ALLOW_ZERO_PLTGOT + +static char * _dl_malloc_addr, *_dl_mmap_zero; +char * _dl_library_path = 0; /* Where we look for libraries */ +char *_dl_preload = 0; /* Things to be loaded before the libs. */ +char *_dl_progname = "/lib/ld-linux-uclibc.so.1"; +static char * _dl_not_lazy = 0; +static char * _dl_warn = 0; /* Used by ldd */ +static char * _dl_trace_loaded_objects = 0; +static int (*_dl_elf_main)(int, char **, char**); + +static int (*_dl_elf_init)(void); + +void * (*_dl_malloc_function)(int size) = NULL; + +struct r_debug * _dl_debug_addr = NULL; + +unsigned int * _dl_brkp; + +unsigned int * _dl_envp; + +#define DL_MALLOC(SIZE) ((void *) (malloc_buffer += SIZE, malloc_buffer - SIZE)) +/* + * Make sure that the malloc buffer is aligned on 4 byte boundary. For 64 bit + * platforms we may need to increase this to 8, but this is good enough for + * now. This is typically called after DL_MALLOC. + */ +#define REALIGN() malloc_buffer = (char *) (((unsigned int) malloc_buffer + 3) & ~(3)) + + + +#define ELF_HASH(RESULT,NAME) { \ + unsigned long hash = 0; \ + unsigned long tmp; \ + char * name = NAME; \ + while (*name){ \ + hash = (hash << 4) + *name++; \ + if((tmp = hash & 0xf0000000)) hash ^= tmp >> 24; \ + hash &= ~tmp; \ + } \ + RESULT = hash; \ +} +extern int _dl_linux_resolve(void); +extern int _dl_interpreter_exit(int); +extern char * _dl_strdup(const char *); +extern char * _dl_getenv(char * symbol, char ** envp); +extern void _dl_unsetenv(char * symbol, char ** envp); +extern int _dl_fixup(struct elf_resolve * tpnt); + +/* + * Datatype of a relocation on this platform + */ +#ifdef ELF_USES_RELOCA +typedef struct elf32_rela ELF_RELOC; +#else +typedef struct elf32_rel ELF_RELOC; +#endif + +/* + * This stub function is used by some debuggers. The idea is that they + * can set an internal breakpoint on it, so that we are notified when the + * address mapping is changed in some way. + */ +void _dl_debug_state() +{ + return; +} + +void _dl_boot(int args); + +void _dl_boot(int args){ + unsigned int argc; + char ** argv, ** envp; + int status; + + unsigned int load_addr; + unsigned int * got; + unsigned int * aux_dat; + int goof = 0; + struct elfhdr * header; + struct elf_resolve * tpnt; + struct dyn_elf * rpnt; + struct elf_resolve * app_tpnt; + unsigned int brk_addr; + unsigned int dl_data[AT_EGID+1]; + unsigned char * malloc_buffer, *mmap_zero; + int (*_dl_atexit)(void *); + int * lpnt; + struct dynamic * dpnt; + unsigned int *hash_addr; + struct r_debug * debug_addr; + unsigned int *chains; + int indx; + int _dl_secure; + + /* First obtain the information on the stack that tells us more about + what binary is loaded, where it is loaded, etc, etc */ + + GET_ARGV(aux_dat, args); + argc = *(aux_dat - 1); + argv = (char **) aux_dat; + aux_dat += argc; /* Skip over the argv pointers */ + aux_dat++; /* Skip over NULL at end of argv */ + envp = (char **) aux_dat; + while(*aux_dat) aux_dat++; /* Skip over the envp pointers */ + aux_dat++; /* Skip over NULL at end of envp */ + dl_data[AT_UID] = -1; /* check later to see if it is changed */ + while(*aux_dat) + { + unsigned int * ad1; + ad1 = aux_dat + 1; + if( *aux_dat <= AT_EGID ) dl_data[*aux_dat] = *ad1; + aux_dat += 2; + } + + /* Next, locate the GOT */ + + load_addr = dl_data[AT_BASE]; + + GET_GOT(got); + dpnt = (struct dynamic *) (*got + load_addr); + + /* OK, time for another hack. Now call mmap to get a page of writable + memory that can be used for a temporary malloc. We do not know brk + yet, so we cannot use real malloc. */ + + { + /* This hack is to work around a suspected asm bug in gcc-2.7.0 */ + int zfileno; +#define ZFILENO ((-1 & (~zfileno)) | zfileno) +/*#define ZFILENO -1*/ + +#ifndef MAP_ANONYMOUS +#ifdef __sparc__ +#define MAP_ANONYMOUS 0x20 +#else +#error MAP_ANONYMOUS not defined and suplementary value not known +#endif +#endif + + /* See if we need to relocate this address */ + mmap_zero = malloc_buffer = (unsigned char *) _dl_mmap((void*) 0, 4096, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, ZFILENO, 0); + if(_dl_mmap_check_error(mmap_zero)) { + SEND_STDERR("dl_boot: mmap of /dev/zero failed!\n"); + _dl_exit(13); + } + } + + tpnt = DL_MALLOC(sizeof(struct elf_resolve)); + REALIGN(); + _dl_memset (tpnt, 0, sizeof (*tpnt)); + app_tpnt = DL_MALLOC(sizeof(struct elf_resolve)); + REALIGN(); + _dl_memset (app_tpnt, 0, sizeof (*app_tpnt)); + + /* + * This is used by gdb to locate the chain of shared libraries that are currently loaded. + */ + debug_addr = DL_MALLOC(sizeof(struct r_debug)); + REALIGN(); + _dl_memset (debug_addr, 0, sizeof (*debug_addr)); + + /* OK, that was easy. Next scan the DYNAMIC section of the image. + We are only doing ourself right now - we will have to do the rest later */ + + while(dpnt->d_tag) + { + tpnt->dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val; + if(dpnt->d_tag == DT_TEXTREL || + SVR4_BUGCOMPAT) tpnt->dynamic_info[DT_TEXTREL] = 1; + dpnt++; + } + + { + struct elf_phdr * ppnt; + int i; + + ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + for(i=0; i<dl_data[AT_PHNUM]; i++, ppnt++) + if(ppnt->p_type == PT_DYNAMIC) { + dpnt = (struct dynamic *) ppnt->p_vaddr; + while(dpnt->d_tag) + { + if(dpnt->d_tag > DT_JMPREL) {dpnt++; continue; } + app_tpnt->dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val; + if(dpnt->d_tag == DT_DEBUG) dpnt->d_un.d_val = (int) debug_addr; + if(dpnt->d_tag == DT_TEXTREL || + SVR4_BUGCOMPAT) app_tpnt->dynamic_info[DT_TEXTREL] = 1; + dpnt++; + } + } + } + + /* Get some more of the information that we will need to dynamicly link + this module to itself */ + + hash_addr = (unsigned int *) (tpnt->dynamic_info[DT_HASH]+load_addr); + tpnt->nbucket = *hash_addr++; + tpnt->nchain = *hash_addr++; + tpnt->elf_buckets = hash_addr; + hash_addr += tpnt->nbucket; + chains = hash_addr; + + /* Ugly, ugly. We need to call mprotect to change the protection of + the text pages so that we can do the dynamic linking. We can set the + protection back again once we are done */ + + { + struct elf_phdr * ppnt; + int i; + + /* First cover the shared library/dynamic linker. */ + if(tpnt->dynamic_info[DT_TEXTREL]) { + header = (struct elfhdr *) dl_data[AT_BASE]; + ppnt = (struct elf_phdr *) (dl_data[AT_BASE] + header->e_phoff); + for(i=0; i<header->e_phnum ; i++, ppnt++) { + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W)) + _dl_mprotect((void *) (load_addr + (ppnt->p_vaddr & 0xfffff000)), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + PROT_READ | PROT_WRITE | PROT_EXEC); + } + } + + /* Now cover the application program. */ + if(app_tpnt->dynamic_info[DT_TEXTREL]) { + ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + for(i=0; i<dl_data[AT_PHNUM]; i++, ppnt++) { + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W)) + _dl_mprotect((void *) (ppnt->p_vaddr & 0xfffff000), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + PROT_READ | PROT_WRITE | PROT_EXEC); + } + } + } + + /* OK, now do the relocations. We do not do a lazy binding here, so + that once we are done, we have considerably more flexibility. */ + + goof = 0; + for(indx=0; indx < 2; indx++) + { + int i; + ELF_RELOC * rpnt; + unsigned int * reloc_addr; + unsigned int symbol_addr; + int symtab_index; + unsigned int rel_addr, rel_size; + + +#ifdef ELF_USES_RELOCA + rel_addr = (indx ? tpnt->dynamic_info[DT_JMPREL] : tpnt->dynamic_info[DT_RELA]); + rel_size = (indx ? tpnt->dynamic_info[DT_PLTRELSZ] : tpnt->dynamic_info[DT_RELASZ]); +#else + rel_addr = (indx ? tpnt->dynamic_info[DT_JMPREL] : tpnt->dynamic_info[DT_REL]); + rel_size = (indx ? tpnt->dynamic_info[DT_PLTRELSZ] : tpnt->dynamic_info[DT_RELSZ]); +#endif + + + if(!rel_addr) continue; + + /* Now parse the relocation information */ + rpnt = (ELF_RELOC *) (rel_addr + load_addr); + for(i=0; i< rel_size; i+=sizeof(ELF_RELOC), rpnt++){ + reloc_addr = (int *) (load_addr + (int)rpnt->r_offset); + symtab_index = ELF32_R_SYM(rpnt->r_info); + symbol_addr = 0; + if(symtab_index) { + char * strtab; + struct elf32_sym * symtab; + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB]+load_addr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB]+load_addr); + + /* We only do a partial dynamic linking right now. The user + is not supposed to redefine any symbols that start with + a '_', so we can do this with confidence. */ + + if (!_dl_symbol(strtab + symtab[symtab_index].st_name)) continue; + + symbol_addr = load_addr + symtab[symtab_index].st_value; + + if(!symbol_addr) { + /* + * This will segfault - you cannot call a function until + * we have finished the relocations. + */ + SEND_STDERR("ELF dynamic loader - unable to self-bootstrap - symbol "); + SEND_STDERR(strtab + symtab[symtab_index].st_name); + SEND_STDERR(" undefined.\n"); + goof++; + } + } + /* + * Use this machine-specific macro to perform the actual relocation. + */ + PERFORM_BOOTSTRAP_RELOC(rpnt, reloc_addr, symbol_addr, load_addr); + } + } + + if (goof) _dl_exit(14); + + /* OK, at this point we have a crude malloc capability. Start to build + the tables of the modules that are required for this beast to run. + We start with the basic executable, and then go from there. Eventually + we will run across ourself, and we will need to properly deal with that + as well. */ + + _dl_malloc_addr = malloc_buffer; + + _dl_mmap_zero = mmap_zero; +/* tpnt = _dl_malloc(sizeof(struct elf_resolve)); */ + +/* Now we have done the mandatory linking of some things. We are now + free to start using global variables, since these things have all been + fixed up by now. Still no function calls outside of this library , + since the dynamic resolver is not yet ready. */ + + lpnt = (int *) (tpnt->dynamic_info[DT_PLTGOT] + load_addr); + INIT_GOT(lpnt, tpnt); + + /* OK, this was a big step, now we need to scan all of the user images + and load them properly. */ + + tpnt->next = 0; + tpnt->libname = 0; + tpnt->libtype = program_interpreter; + + { struct elfhdr * epnt; + struct elf_phdr * ppnt; + int i; + + epnt = (struct elfhdr *) dl_data[AT_BASE]; + tpnt->n_phent = epnt->e_phnum; + tpnt->ppnt = ppnt = (struct elf_phdr *) (load_addr + epnt->e_phoff); + for(i=0;i < epnt->e_phnum; i++, ppnt++){ + if(ppnt->p_type == PT_DYNAMIC) { + tpnt->dynamic_addr = ppnt->p_vaddr + load_addr; + tpnt->dynamic_size = ppnt->p_filesz; + } + } + } + + tpnt->chains = chains; + tpnt->loadaddr = (char *) load_addr; + + brk_addr = 0; + rpnt = NULL; + + /* At this point we are now free to examine the user application, + and figure out which libraries are supposed to be called. Until + we have this list, we will not be completely ready for dynamic linking */ + + { + struct elf_phdr * ppnt; + int i; + + ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + for(i=0; i<dl_data[AT_PHNUM]; i++, ppnt++) { + if(ppnt->p_type == PT_LOAD) { + if(ppnt->p_vaddr + ppnt->p_memsz > brk_addr) + brk_addr = ppnt->p_vaddr + ppnt->p_memsz; + } + if(ppnt->p_type == PT_DYNAMIC) { +#ifndef ALLOW_ZERO_PLTGOT + /* make sure it's really there. */ + if (app_tpnt->dynamic_info[DT_PLTGOT] == 0) continue; +#endif + /* OK, we have what we need - slip this one into the list. */ + app_tpnt = _dl_add_elf_hash_table("", 0, + app_tpnt->dynamic_info, ppnt->p_vaddr, ppnt->p_filesz); + _dl_loaded_modules->libtype = elf_executable; + _dl_loaded_modules->ppnt = (struct elf_phdr *) dl_data[AT_PHDR]; + _dl_loaded_modules->n_phent = dl_data[AT_PHNUM]; + _dl_symbol_tables = rpnt = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt, 0, sizeof (*rpnt)); + rpnt->dyn = _dl_loaded_modules; + app_tpnt->usage_count++; + app_tpnt->symbol_scope = _dl_symbol_tables; + lpnt = (int *) (app_tpnt->dynamic_info[DT_PLTGOT]); +#ifdef ALLOW_ZERO_PLTGOT + if (lpnt) +#endif + INIT_GOT(lpnt, _dl_loaded_modules); + } + if(ppnt->p_type == PT_INTERP) { /* OK, fill this in - we did not have + this before */ + tpnt->libname = _dl_strdup((char *) ppnt->p_offset +(dl_data[AT_PHDR] & 0xfffff000)); + } + } + } + + if (argv[0]) + _dl_progname = argv[0]; + + /* Now we need to figure out what kind of options are selected. + Note that for SUID programs we ignore the settings in LD_LIBRARY_PATH */ + { + _dl_not_lazy = _dl_getenv("LD_BIND_NOW",envp); + + if ( (dl_data[AT_UID] == -1 && _dl_suid_ok()) || + (dl_data[AT_UID] != -1 && dl_data[AT_UID] == dl_data[AT_EUID] && + dl_data[AT_GID] == dl_data[AT_EGID])) + { + _dl_secure = 0; + _dl_preload = _dl_getenv("LD_PRELOAD", envp); + _dl_library_path = _dl_getenv("LD_LIBRARY_PATH",envp); + } + else + { + _dl_secure = 1; + _dl_preload = _dl_getenv("LD_PRELOAD", envp); + _dl_unsetenv("LD_AOUT_PRELOAD", envp); + _dl_unsetenv("LD_LIBRARY_PATH", envp); + _dl_unsetenv("LD_AOUT_LIBRARY_PATH", envp); + _dl_library_path = NULL; + } + } + + _dl_trace_loaded_objects = _dl_getenv("LD_TRACE_LOADED_OBJECTS", envp); + + /* OK, we now have the application in the list, and we have some + basic stuff in place. Now search through the list for other shared + libraries that should be loaded, and insert them on the list in the + correct order. */ + +#ifdef USE_CACHE + _dl_map_cache(); +#endif + + { + struct elf_resolve *tcurr; + struct elf_resolve *tpnt1; + char *lpnt; + + if (_dl_preload) { + char c, *str, *str2; + + str = _dl_preload; + while (*str == ':' || *str == ' ' || *str == '\t') + str++; + while (*str) { + str2 = str; + while (*str2 && *str2 != ':' && *str2 != ' ' && *str2 != '\t') + str2++; + c = *str2; + *str2 = '\0'; + if (!_dl_secure || _dl_strchr(str, '/') == NULL) { + tpnt1 = _dl_load_shared_library(_dl_secure, NULL, str); + if (!tpnt1) { + if (_dl_trace_loaded_objects) + _dl_fdprintf(1, "\t%s => not found\n", str); + else { + _dl_fdprintf(2, "%s: can't load library '%s'\n", + _dl_progname, str); + _dl_exit(15); + } + } else { + if (_dl_trace_loaded_objects && !tpnt1->usage_count) { + /* this is a real hack to make ldd not print the + library itself when run on a library. */ + if (_dl_strcmp(_dl_progname, str) != 0) + _dl_fdprintf(1, "\t%s => %s (0x%x)\n", str, tpnt1->libname, + (unsigned)tpnt1->loadaddr); + } + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + tpnt1->usage_count++; + tpnt1->symbol_scope = _dl_symbol_tables; + tpnt1->libtype = elf_lib; + rpnt->dyn = tpnt1; + } + } + *str2 = c; + str = str2; + while (*str == ':' || *str == ' ' || *str == '\t') + str++; + } + } + + { + int fd; + struct kernel_stat st; + char *preload; + + if (!_dl_stat(LDSO_PRELOAD, &st)) { + if ((fd = _dl_open(LDSO_PRELOAD, O_RDONLY)) < 0) { + _dl_fdprintf(2, "%s: can't open file '%s'\n", _dl_progname, + LDSO_PRELOAD); + } else { + preload = (caddr_t)_dl_mmap(0, st.st_size+1, PROT_READ|PROT_WRITE, + MAP_PRIVATE, fd, 0); + _dl_close (fd); + if (preload == (caddr_t)-1) { + _dl_fdprintf(2, "%s: can't map file '%s'\n", _dl_progname, + LDSO_PRELOAD); + } else { + char c, *cp, *cp2; + + /* convert all separators and comments to spaces */ + for (cp = preload; *cp; /*nada*/) { + if (*cp == ':' || *cp == '\t' || *cp == '\n') { + *cp++ = ' '; + } else if (*cp == '#') { + do + *cp++ = ' '; + while (*cp != '\n' && *cp != '\0'); + } else { + cp++; + } + } + + /* find start of first library */ + for (cp = preload; *cp && *cp == ' '; cp++) + /*nada*/; + + while (*cp) { + /* find end of library */ + for (cp2 = cp; *cp && *cp != ' '; cp++) + /*nada*/; + c = *cp; + *cp = '\0'; + + tpnt1 = _dl_load_shared_library(0, NULL, cp2); + if (!tpnt1) { + if (_dl_trace_loaded_objects) + _dl_fdprintf(1, "\t%s => not found\n", cp2); + else { + _dl_fdprintf(2, "%s: can't load library '%s'\n", + _dl_progname, cp2); + _dl_exit(15); + } + } else { + if (_dl_trace_loaded_objects && !tpnt1->usage_count) + _dl_fdprintf(1, "\t%s => %s (0x%x)\n", cp2, tpnt1->libname, + (unsigned)tpnt1->loadaddr); + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + tpnt1->usage_count++; + tpnt1->symbol_scope = _dl_symbol_tables; + tpnt1->libtype = elf_lib; + rpnt->dyn = tpnt1; + } + + /* find start of next library */ + *cp = c; + for (/*nada*/; *cp && *cp == ' '; cp++) + /*nada*/; + } + + _dl_munmap(preload, st.st_size+1); + } + } + } + } + + for (tcurr = _dl_loaded_modules; tcurr; tcurr = tcurr->next) + { + for (dpnt = (struct dynamic *)tcurr->dynamic_addr; dpnt->d_tag; dpnt++) + { + if(dpnt->d_tag == DT_NEEDED) + { + lpnt = tcurr->loadaddr + tcurr->dynamic_info[DT_STRTAB] + + dpnt->d_un.d_val; + if (tpnt && _dl_strcmp(lpnt, tpnt->libname) == 0) + { + struct elf_resolve * ttmp; + ttmp = _dl_loaded_modules; + while (ttmp->next) + ttmp = ttmp->next; + ttmp->next = tpnt; + tpnt->prev = ttmp; + tpnt->next = NULL; + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + rpnt->dyn = tpnt; + tpnt->usage_count++; + tpnt->symbol_scope = _dl_symbol_tables; + tpnt = NULL; + continue; + } + if (!(tpnt1 = _dl_load_shared_library(0, tcurr, lpnt))) + { + if (_dl_trace_loaded_objects) + _dl_fdprintf(1, "\t%s => not found\n", lpnt); + else + { + _dl_fdprintf(2, "%s: can't load library '%s'\n", + _dl_progname, lpnt); + _dl_exit(16); + } + } + else + { + if (_dl_trace_loaded_objects && !tpnt1->usage_count) + _dl_fdprintf(1, "\t%s => %s (0x%x)\n", lpnt, tpnt1->libname, + (unsigned)tpnt1->loadaddr); + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + tpnt1->usage_count++; + tpnt1->symbol_scope = _dl_symbol_tables; + tpnt1->libtype = elf_lib; + rpnt->dyn = tpnt1; + } + } + } + } + } + +#ifdef USE_CACHE + _dl_unmap_cache(); +#endif + + /* ldd uses uses this. I am not sure how you pick up the other flags */ + if(_dl_trace_loaded_objects) + { + _dl_warn = _dl_getenv("LD_WARN", envp); + if (!_dl_warn) _dl_exit(0); + } + + /* + * If the program interpreter is not in the module chain, add it. This will + * be required for dlopen to be able to access the internal functions in the + * dynamic linker. + */ + if(tpnt) { + struct elf_resolve * tcurr; + + tcurr = _dl_loaded_modules; + if (tcurr) + while(tcurr->next) tcurr = tcurr->next; + tpnt->next = NULL; + tpnt->usage_count++; + + if (tcurr) { + tcurr->next = tpnt; + tpnt->prev = tcurr; + } + else { + _dl_loaded_modules = tpnt; + tpnt->prev = NULL; + } + if (rpnt) { + rpnt->next = + (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt->next, 0, sizeof (*(rpnt->next))); + rpnt = rpnt->next; + } else { + rpnt = (struct dyn_elf *) _dl_malloc(sizeof(struct dyn_elf)); + _dl_memset (rpnt, 0, sizeof (*(rpnt->next))); + } + rpnt->dyn = tpnt; + tpnt = NULL; + } + + /* + * OK, now all of the kids are tucked into bed in their proper addresses. + * Now we go through and look for REL and RELA records that indicate fixups + * to the GOT tables. We need to do this in reverse order so that COPY + * directives work correctly */ + + + goof = _dl_loaded_modules ? _dl_fixup(_dl_loaded_modules) : 0; + + + /* Some flavors of SVr4 do not generate the R_*_COPY directive, + and we have to manually search for entries that require fixups. + Solaris gets this one right, from what I understand. */ + + + if (_dl_symbol_tables) + goof += _dl_copy_fixups(_dl_symbol_tables); + + if(goof || _dl_trace_loaded_objects) _dl_exit(0); + + /* OK, at this point things are pretty much ready to run. Now we + need to touch up a few items that are required, and then + we can let the user application have at it. Note that + the dynamic linker itself is not guaranteed to be fully + dynamicly linked if we are using ld.so.1, so we have to look + up each symbol individually. */ + + + _dl_brkp = (unsigned int *) _dl_find_hash("___brk_addr", NULL, 1, NULL, 0); + if (_dl_brkp) *_dl_brkp = brk_addr; + _dl_envp = (unsigned int *) _dl_find_hash("__environ", NULL, 1, NULL, 0); + + if (_dl_envp) *_dl_envp = (unsigned int) envp; + + { + int i; + struct elf_phdr * ppnt; + + /* We had to set the protections of all pages to R/W for dynamic linking. + Set text pages back to R/O */ + for(tpnt = _dl_loaded_modules; tpnt; tpnt = tpnt->next) + for(ppnt = tpnt->ppnt, i=0; i < tpnt->n_phent; i++, ppnt++) + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W) && + tpnt->dynamic_info[DT_TEXTREL]) + _dl_mprotect((void *) (tpnt->loadaddr + (ppnt->p_vaddr & 0xfffff000)), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + LXFLAGS(ppnt->p_flags)); + + } + + _dl_atexit = (int (*)(void *)) _dl_find_hash("atexit", NULL, 1, NULL, 0); + + /* + * OK, fix one more thing - set up the debug_addr structure to point + * to our chain. Later we may need to fill in more fields, but this + * should be enough for now. + */ + debug_addr->r_map = (struct link_map *) _dl_loaded_modules; + debug_addr->r_version = 1; + debug_addr->r_ldbase = load_addr; + debug_addr->r_brk = (unsigned long) &_dl_debug_state; + _dl_debug_addr = debug_addr; + debug_addr->r_state = RT_CONSISTENT; + /* This is written in this funny way to keep gcc from inlining the + function call. */ + ((void (*)(void))debug_addr->r_brk)(); + + for(tpnt = _dl_loaded_modules; tpnt; tpnt = tpnt->next) + { + /* Apparently crt1 for the application is responsible for handling this. + * We only need to run the init/fini for shared libraries + */ + if (tpnt->libtype == program_interpreter || + tpnt->libtype == elf_executable) continue; + if (tpnt->init_flag & INIT_FUNCS_CALLED) continue; + tpnt->init_flag |= INIT_FUNCS_CALLED; + + if(tpnt->dynamic_info[DT_INIT]) { + _dl_elf_init = (int (*)(void)) (tpnt->loadaddr + + tpnt->dynamic_info[DT_INIT]); + (*_dl_elf_init)(); + } + if(_dl_atexit && tpnt->dynamic_info[DT_FINI]) + { + (*_dl_atexit)(tpnt->loadaddr + tpnt->dynamic_info[DT_FINI]); + } +#undef DL_DEBUG +#ifdef DL_DEBUG + else + { + _dl_fdprintf(2, tpnt->libname); + _dl_fdprintf(2, ": "); + if (!_dl_atexit) + _dl_fdprintf(2, "The address is atexit () is 0x0."); + if (!tpnt->dynamic_info[DT_FINI]) + _dl_fdprintf(2, "Invalid .fini section."); + _dl_fdprintf(2, "\n"); + } +#endif +#undef DL_DEBUG + } + + /* OK we are done here. Turn out the lights, and lock up. */ + _dl_elf_main = (int (*)(int, char**, char**)) dl_data[AT_ENTRY]; + + + /* + * Transfer control to the application. + */ + START(); +} + +int _dl_fixup(struct elf_resolve * tpnt) +{ + int goof = 0; + if(tpnt->next) goof += _dl_fixup(tpnt->next); + + if(tpnt->dynamic_info[DT_REL]) { +#ifdef ELF_USES_RELOCA + _dl_fdprintf(2, "%s: can't handle REL relocation records\n", _dl_progname); + _dl_exit(17); +#else + if (tpnt->init_flag & RELOCS_DONE) return goof; + tpnt->init_flag |= RELOCS_DONE; + + goof += _dl_parse_relocation_information(tpnt, tpnt->dynamic_info[DT_REL], + tpnt->dynamic_info[DT_RELSZ], 0); +#endif + } + if(tpnt->dynamic_info[DT_RELA]) { +#ifdef ELF_USES_RELOCA + if (tpnt->init_flag & RELOCS_DONE) return goof; + tpnt->init_flag |= RELOCS_DONE; + + goof += _dl_parse_relocation_information(tpnt, tpnt->dynamic_info[DT_RELA], + tpnt->dynamic_info[DT_RELASZ], 0); +#else + _dl_fdprintf(2, "%s: can't handle RELA relocation records\n", _dl_progname); + _dl_exit(18); +#endif + } + if(tpnt->dynamic_info[DT_JMPREL]) + { + if (tpnt->init_flag & JMP_RELOCS_DONE) return goof; + tpnt->init_flag |= JMP_RELOCS_DONE; + + if(! _dl_not_lazy || *_dl_not_lazy == 0) + _dl_parse_lazy_relocation_information(tpnt, tpnt->dynamic_info[DT_JMPREL], + tpnt->dynamic_info[DT_PLTRELSZ], 0); + else + goof += _dl_parse_relocation_information(tpnt, + tpnt->dynamic_info[DT_JMPREL], + tpnt->dynamic_info[DT_PLTRELSZ], 0); + } + return goof; +} + +void * _dl_malloc(int size) { + void * retval; + + if(_dl_malloc_function) + return (*_dl_malloc_function)(size); + + if(_dl_malloc_addr-_dl_mmap_zero+size>4096) { + _dl_mmap_zero = _dl_malloc_addr = (unsigned char *) _dl_mmap((void*) 0, size, + PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + if(_dl_mmap_check_error(_dl_mmap_zero)) { + _dl_fdprintf(2, "%s: can't map '/dev/zero'\n", _dl_progname); + _dl_exit(20); + } + } + retval = _dl_malloc_addr; + _dl_malloc_addr += size; + + /* + * Align memory to 4 byte boundary. Some platforms require this, others + * simply get better performance. + */ + _dl_malloc_addr = (char *) (((unsigned int) _dl_malloc_addr + 3) & ~(3)); + return retval; +} + +char * _dl_getenv(char *symbol, char **envp) +{ + char *pnt; + char *pnt1; + while ((pnt = *envp++)) { + pnt1 = symbol; + while (*pnt && *pnt == *pnt1) + pnt1++, pnt++; + if (!*pnt || *pnt != '=' || *pnt1) + continue; + return pnt+1; + } + return 0; +} + +void _dl_unsetenv(char *symbol, char **envp) +{ + char *pnt; + char *pnt1; + char **newenvp = envp; + for (pnt = *envp; pnt; pnt = *++envp) { + pnt1 = symbol; + while (*pnt && *pnt == *pnt1) + pnt1++, pnt++; + if(!*pnt || *pnt != '=' || *pnt1) + *newenvp++ = *envp; + } + *newenvp++ = *envp; + return; +} + +char * _dl_strdup(const char * string){ + char * retval; + int len; + + len = _dl_strlen(string); + retval = _dl_malloc(len + 1); + _dl_strcpy(retval, string); + return retval; +} + +/* In principle we could do the .fini stuff here, but we already + registered this stuff with atexit */ +int _dl_interpreter_exit(int exitcode){ +/* _dl_fdprintf(2, "Hey, look where I am!\n"); */ + return 0; +} diff --git a/ldso/ldso/link.h b/ldso/ldso/link.h new file mode 100644 index 000000000..adaa20e4a --- /dev/null +++ b/ldso/ldso/link.h @@ -0,0 +1,37 @@ +#ifndef _LINK_H +#define _LINK_H + +#include "elf.h" + +/* Header file that describes the internal data structures used by the + * ELF dynamic linker. */ + +struct link_map +{ + /* These entries must be in this order to be compatible with the + * interface used by gdb to obtain the list of symbols. */ + unsigned long l_addr; /* address at which object is mapped */ + char *l_name; /* full name of loaded object */ + Elf32_Dyn *l_ld; /* dynamic structure of object */ + struct link_map *l_next; + struct link_map *l_prev; +}; + +/* The DT_DEBUG entry in the .dynamic section is given the address of + * this structure. gdb can pick this up to obtain the correct list of + * loaded modules. */ +struct r_debug +{ + int r_version; /* debugging info version no */ + struct link_map *r_map; /* address of link_map */ + unsigned long r_brk; /* address of update routine */ + enum + { + RT_CONSISTENT, + RT_ADD, + RT_DELETE + } r_state; + unsigned long r_ldbase; /* base addr of ld.so */ +}; + +#endif /* _LINK_H */ diff --git a/ldso/ldso/linuxelf.h b/ldso/ldso/linuxelf.h new file mode 100644 index 000000000..cc016f6b7 --- /dev/null +++ b/ldso/ldso/linuxelf.h @@ -0,0 +1,137 @@ +/* This should eventually appear in the distribution version of linux/elf.h */ +#ifndef R_SPARC_NONE +#define R_SPARC_NONE 0 +#define R_SPARC_8 1 +#define R_SPARC_16 2 +#define R_SPARC_32 3 +#define R_SPARC_DISP8 4 +#define R_SPARC_DISP16 5 +#define R_SPARC_DISP32 6 +#define R_SPARC_WDISP30 7 +#define R_SPARC_WDISP22 8 +#define R_SPARC_HI22 9 +#define R_SPARC_22 10 +#define R_SPARC_13 11 +#define R_SPARC_LO10 12 +#define R_SPARC_GOT10 13 +#define R_SPARC_GOT13 14 +#define R_SPARC_GOT22 15 +#define R_SPARC_PC10 16 +#define R_SPARC_PC22 17 +#define R_SPARC_WPLT30 18 +#define R_SPARC_COPY 19 +#define R_SPARC_GLOB_DAT 20 +#define R_SPARC_JMP_SLOT 21 +#define R_SPARC_RELATIVE 22 +#define R_SPARC_UA32 23 +#endif + +#ifndef R_68K_NONE +#define R_68K_NONE 0 +#define R_68K_32 1 +#define R_68K_16 2 +#define R_68K_8 3 +#define R_68K_PC32 4 +#define R_68K_PC16 5 +#define R_68K_PC8 6 +#define R_68K_GOT32 7 +#define R_68K_GOT16 8 +#define R_68K_GOT8 9 +#define R_68K_GOT32O 10 +#define R_68K_GOT16O 11 +#define R_68K_GOT8O 12 +#define R_68K_PLT32 13 +#define R_68K_PLT16 14 +#define R_68K_PLT8 15 +#define R_68K_PLT32O 16 +#define R_68K_PLT16O 17 +#define R_68K_PLT8O 18 +#define R_68K_COPY 19 +#define R_68K_GLOB_DAT 20 +#define R_68K_JMP_SLOT 21 +#define R_68K_RELATIVE 22 +#define R_68K_NUM 23 +#endif + +/* + * These constants define the elements of the auxiliary vector used to + * pass additional information from the kernel to an ELF application. + */ + +#ifndef AT_NULL +typedef struct +{ + int a_type; + union{ + long a_val; + void *p_ptr; + void (*a_fcn)(); + } a_un; +} auxv_t; + +/* + * Values of a_type... These often appear in the file /usr/include/sys/auxv.h + * on SVr4 systems. + */ +#define AT_NULL 0 +#define AT_IGNORE 1 +#define AT_EXECFD 2 +#define AT_PHDR 3 +#define AT_PHENT 4 +#define AT_PHNUM 5 +#define AT_PAGESZ 6 +#define AT_BASE 7 +#define AT_FLAGS 8 +#define AT_ENTRY 9 +#endif +#ifndef AT_NOTELF +#define AT_NOTELF 10 /* program is not ELF */ +#define AT_UID 11 /* real uid */ +#define AT_EUID 12 /* effective uid */ +#define AT_GID 13 /* real gid */ +#define AT_EGID 14 /* effective gid */ +#endif + +extern int _dl_linux_resolve(void); +extern struct elf_resolve * _dl_load_shared_library(int secure, + struct elf_resolve *, char * libname); +extern void * _dl_malloc(int size); +extern int _dl_map_cache(void); +extern int _dl_unmap_cache(void); + +extern struct elf_resolve * _dl_load_elf_shared_library(int secure, + char * libname, int); +int _dl_copy_fixups(struct dyn_elf * tpnt); + +extern int linux_run(int argc, char * argv[]); + +extern void _dl_parse_lazy_relocation_information(struct elf_resolve * tpnt, int rel_addr, + int rel_size, int type); + +extern int _dl_parse_relocation_information(struct elf_resolve * tpnt, int rel_addr, + int rel_size, int type); +extern int _dl_parse_copy_information(struct dyn_elf * rpnt, int rel_addr, + int rel_size, int type); + + +/* This means that we may be forced to manually search for copy fixups + which were omitted by the linker. We cannot depend upon the DT_TEXTREL + to tell us whether there are fixups in a text section or not. */ + +#ifndef SVR4_BUGCOMPAT +#define SVR4_BUGCOMPAT 1 +#endif + +#ifndef PF_R +#define PF_R 4 +#define PF_W 2 +#define PF_X 1 +#endif + +/* Convert between the Linux flags for page protections and the + ones specified in the ELF standard. */ + +#define LXFLAGS(X) ( (((X) & PF_R) ? PROT_READ : 0) | \ + (((X) & PF_W) ? PROT_WRITE : 0) | \ + (((X) & PF_X) ? PROT_EXEC : 0)) + diff --git a/ldso/ldso/m68k/dl-sysdep.h b/ldso/ldso/m68k/dl-sysdep.h new file mode 100644 index 000000000..24af47ca0 --- /dev/null +++ b/ldso/ldso/m68k/dl-sysdep.h @@ -0,0 +1,87 @@ + +/* Various assmbly language/system dependent hacks that are required + so that we can minimize the amount of platform specific code. */ + +/* Define this if the system uses RELOCA. */ +#define ELF_USES_RELOCA + +/* Get a pointer to the argv array. On many platforms this can be + just the address if the first argument, on other platforms we need + to do something a little more subtle here. */ +#define GET_ARGV(ARGVP, ARGS) ((ARGVP) = ((unsigned int *) &(ARGS))) + +/* Get the address of the Global offset table. This must be absolute, + not relative. */ +#define GET_GOT(X) __asm__ ("movel %%a5,%0" : "=g" (X)) + +/* Initialization sequence for a GOT. */ +#define INIT_GOT(GOT_BASE,MODULE) \ +{ \ + GOT_BASE[2] = (int) _dl_linux_resolve; \ + GOT_BASE[1] = (int) (MODULE); \ +} + +/* Here is a macro to perform a relocation. This is only used when + bootstrapping the dynamic loader. RELP is the relocation that we + are performing, REL is the pointer to the address we are + relocating. SYMBOL is the symbol involved in the relocation, and + LOAD is the load address. */ +#define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD) \ + switch (ELF32_R_TYPE ((RELP)->r_info)) \ + { \ + case R_68K_8: \ + *(char *) (REL) = (SYMBOL) + (RELP)->r_addend; \ + break; \ + case R_68K_16: \ + *(short *) (REL) = (SYMBOL) + (RELP)->r_addend; \ + break; \ + case R_68K_32: \ + *(REL) = (SYMBOL) + (RELP)->r_addend; \ + break; \ + case R_68K_PC8: \ + *(char *) (REL) = ((SYMBOL) + (RELP)->r_addend \ + - (unsigned int) (REL)); \ + break; \ + case R_68K_PC16: \ + *(short *) (REL) = ((SYMBOL) + (RELP)->r_addend \ + - (unsigned int) (REL)); \ + break; \ + case R_68K_PC32: \ + *(REL) = ((SYMBOL) + (RELP)->r_addend \ + - (unsigned int) (REL)); \ + break; \ + case R_68K_GLOB_DAT: \ + case R_68K_JMP_SLOT: \ + *(REL) = (SYMBOL); \ + break; \ + case R_68K_RELATIVE: /* Compatibility kludge */ \ + *(REL) = ((unsigned int) (LOAD) + ((RELP)->r_addend ? : *(REL))); \ + break; \ + default: \ + _dl_exit (1); \ + } + + +/* Transfer control to the user's application, once the dynamic loader + is done. */ + +#define START() \ + __asm__ volatile ("unlk %%a6\n\t" \ + "jmp %0@" \ + : : "a" (_dl_elf_main)); + + + +/* Here we define the magic numbers that this dynamic loader should accept */ + +#define MAGIC1 EM_68K +#undef MAGIC2 +/* Used for error messages */ +#define ELF_TARGET "m68k" + +struct elf_resolve; +extern unsigned int _dl_linux_resolver (int, int, struct elf_resolve *, int); + +/* Define this because we do not want to call .udiv in the library. + Not needed for m68k. */ +#define do_rem(result, n, base) ((result) = (n) % (base)) diff --git a/ldso/ldso/m68k/elfinterp.c b/ldso/ldso/m68k/elfinterp.c new file mode 100644 index 000000000..9b73efd3f --- /dev/null +++ b/ldso/ldso/m68k/elfinterp.c @@ -0,0 +1,358 @@ +/* Run an ELF binary on a linux system. + + Copyright (C) 1993, Eric Youngdale. + Copyright (C) 1995, Andreas Schwab. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ + +/* Adapted to ELF/68k by Andreas Schwab. */ + +#ifndef VERBOSE_DLINKER +#define VERBOSE_DLINKER +#endif +#ifdef VERBOSE_DLINKER +static char *_dl_reltypes[] = +{ + "R_68K_NONE", + "R_68K_32", "R_68K_16", "R_68K_8", + "R_68K_PC32", "R_68K_PC16", "R_68K_PC8", + "R_68K_GOT32", "R_68K_GOT16", "R_68K_GOT8", + "R_68K_GOT32O", "R_68K_GOT16O", "R_68K_GOT8O", + "R_68K_PLT32", "R_68K_PLT16", "R_68K_PLT8", + "R_68K_PLT32O", "R_68K_PLT16O", "R_68K_PLT8O", + "R_68K_COPY", "R_68K_GLOB_DAT", "R_68K_JMP_SLOT", "R_68K_RELATIVE", + "R_68K_NUM" +}; +#endif + +/* Program to load an ELF binary on a linux system, and run it. + References to symbols in sharable libraries can be resolved by either + an ELF sharable library or a linux style of shared library. */ + +/* Disclaimer: I have never seen any AT&T source code for SVr4, nor have + I ever taken any courses on internals. This program was developed using + information available through the book "UNIX SYSTEM V RELEASE 4, + Programmers guide: Ansi C and Programming Support Tools", which did + a more than adequate job of explaining everything required to get this + working. */ + +#include <linux/types.h> +#include <linux/errno.h> +#include <linux/unistd.h> +/*#include <stdlib.h>*/ +#include "string.h" +#include <linux/unistd.h> +#include <linux/fcntl.h> +#include <linux/elf.h> + +#include "hash.h" +#include "linuxelf.h" +#include "sysdep.h" +#include "../syscall.h" +#include "../string.h" + +extern char *_dl_progname; + +unsigned int +_dl_linux_resolver (int dummy1, int dummy2, + struct elf_resolve *tpnt, int reloc_entry) +{ + int reloc_type; + struct elf32_rela *this_reloc; + char *strtab; + struct elf32_sym *symtab; + char *rel_addr; + int symtab_index; + char *new_addr; + char **got_addr; + unsigned int instr_addr; + + rel_addr = tpnt->loadaddr + tpnt->dynamic_info[DT_JMPREL]; + this_reloc = (struct elf32_rela *) (rel_addr + reloc_entry); + reloc_type = ELF32_R_TYPE (this_reloc->r_info); + symtab_index = ELF32_R_SYM (this_reloc->r_info); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + + tpnt->loadaddr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + + if (reloc_type != R_68K_JMP_SLOT) + { + _dl_fdprintf (2, "%s: incorrect relocation type in jump relocations\n", + _dl_progname); + _dl_exit (1); + } + + /* Address of jump instruction to fix up. */ + instr_addr = (int) this_reloc->r_offset + (int) tpnt->loadaddr; + got_addr = (char **) instr_addr; + +#ifdef DEBUG + _dl_fdprintf (2, "Resolving symbol %s\n", + strtab + symtab[symtab_index].st_name); +#endif + + /* Get the address of the GOT entry. */ + new_addr = _dl_find_hash (strtab + symtab[symtab_index].st_name, + tpnt->symbol_scope, (int) got_addr, tpnt, 0); + if (!new_addr) + { + _dl_fdprintf (2, "%s: can't resolve symbol '%s'\n", + _dl_progname, strtab + symtab[symtab_index].st_name); + _dl_exit (1); + } +/* #define DEBUG_LIBRARY */ +#ifdef DEBUG_LIBRARY + if ((unsigned int) got_addr < 0x40000000) + _dl_fdprintf (2, "Calling library function: %s\n", + strtab + symtab[symtab_index].st_name); + else +#endif + *got_addr = new_addr; + return (unsigned int) new_addr; +} + +void +_dl_parse_lazy_relocation_information (struct elf_resolve *tpnt, int rel_addr, + int rel_size, int type) +{ + int i; + char *strtab; + int reloc_type; + int symtab_index; + struct elf32_sym *symtab; + struct elf32_rela *rpnt; + unsigned int *reloc_addr; + + /* Now parse the relocation information. */ + rpnt = (struct elf32_rela *) (rel_addr + tpnt->loadaddr); + rel_size = rel_size / sizeof (struct elf32_rela); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + + tpnt->loadaddr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + for (i = 0; i < rel_size; i++, rpnt++) + { + reloc_addr = (int *) (tpnt->loadaddr + (int) rpnt->r_offset); + reloc_type = ELF32_R_TYPE (rpnt->r_info); + symtab_index = ELF32_R_SYM (rpnt->r_info); + + /* When the dynamic linker bootstrapped itself, it resolved some symbols. + Make sure we do not do them again. */ + if (tpnt->libtype == program_interpreter + && (!symtab_index + || _dl_symbol (strtab + symtab[symtab_index].st_name))) + continue; + + switch (reloc_type) + { + case R_68K_NONE: + break; + case R_68K_JMP_SLOT: + *reloc_addr += (unsigned int) tpnt->loadaddr; + break; + default: + _dl_fdprintf (2, "%s: (LAZY) can't handle reloc type ", _dl_progname); +#ifdef VERBOSE_DLINKER + _dl_fdprintf (2, "%s ", _dl_reltypes[reloc_type]); +#endif + if (symtab_index) + _dl_fdprintf (2, "'%s'", strtab + symtab[symtab_index].st_name); + _dl_fdprintf (2, "\n"); + _dl_exit (1); + } + } +} + +int +_dl_parse_relocation_information (struct elf_resolve *tpnt, int rel_addr, + int rel_size, int type) +{ + int i; + char *strtab; + int reloc_type; + int goof = 0; + struct elf32_sym *symtab; + struct elf32_rela *rpnt; + unsigned int *reloc_addr; + unsigned int symbol_addr; + int symtab_index; + /* Now parse the relocation information */ + + rpnt = (struct elf32_rela *) (rel_addr + tpnt->loadaddr); + rel_size = rel_size / sizeof (struct elf32_rela); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + + tpnt->loadaddr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + for (i = 0; i < rel_size; i++, rpnt++) + { + reloc_addr = (int *) (tpnt->loadaddr + (int) rpnt->r_offset); + reloc_type = ELF32_R_TYPE (rpnt->r_info); + symtab_index = ELF32_R_SYM (rpnt->r_info); + symbol_addr = 0; + + if (tpnt->libtype == program_interpreter + && (!symtab_index + || _dl_symbol (strtab + symtab[symtab_index].st_name))) + continue; + + if (symtab_index) + { + symbol_addr = (unsigned int) + _dl_find_hash (strtab + symtab[symtab_index].st_name, + tpnt->symbol_scope, (int) reloc_addr, + reloc_type == R_68K_JMP_SLOT ? tpnt : NULL, 0); + + /* We want to allow undefined references to weak symbols - + this might have been intentional. We should not be + linking local symbols here, so all bases should be + covered. */ + if (!symbol_addr + && ELF32_ST_BIND (symtab[symtab_index].st_info) == STB_GLOBAL) + { + _dl_fdprintf (2, "%s: can't resolve symbol '%s'\n", + _dl_progname, strtab + symtab[symtab_index].st_name); + goof++; + } + } + switch (reloc_type) + { + case R_68K_NONE: + break; + case R_68K_8: + *(char *) reloc_addr = symbol_addr + rpnt->r_addend; + break; + case R_68K_16: + *(short *) reloc_addr = symbol_addr + rpnt->r_addend; + break; + case R_68K_32: + *reloc_addr = symbol_addr + rpnt->r_addend; + break; + case R_68K_PC8: + *(char *) reloc_addr = (symbol_addr + rpnt->r_addend + - (unsigned int) reloc_addr); + break; + case R_68K_PC16: + *(short *) reloc_addr = (symbol_addr + rpnt->r_addend + - (unsigned int) reloc_addr); + break; + case R_68K_PC32: + *reloc_addr = (symbol_addr + rpnt->r_addend + - (unsigned int) reloc_addr); + break; + case R_68K_GLOB_DAT: + case R_68K_JMP_SLOT: + *reloc_addr = symbol_addr; + break; + case R_68K_RELATIVE: + *reloc_addr = ((unsigned int) tpnt->loadaddr + /* Compatibility kludge. */ + + (rpnt->r_addend ? : *reloc_addr)); + break; + case R_68K_COPY: +#if 0 /* Do this later. */ + _dl_fdprintf (2, "Doing copy"); + if (symtab_index) + _dl_fdprintf (2, " for symbol %s", + strtab + symtab[symtab_index].st_name); + _dl_fdprintf (2, "\n"); + _dl_memcpy ((void *) symtab[symtab_index].st_value, + (void *) symbol_addr, + symtab[symtab_index].st_size); +#endif + break; + default: + _dl_fdprintf (2, "%s: can't handle reloc type ", _dl_progname); +#ifdef VERBOSE_DLINKER + _dl_fdprintf (2, "%s ", _dl_reltypes[reloc_type]); +#endif + if (symtab_index) + _dl_fdprintf (2, "'%s'", strtab + symtab[symtab_index].st_name); + _dl_fdprintf (2, "\n"); + _dl_exit (1); + } + + } + return goof; +} + +/* This is done as a separate step, because there are cases where + information is first copied and later initialized. This results in + the wrong information being copied. Someone at Sun was complaining about + a bug in the handling of _COPY by SVr4, and this may in fact be what he + was talking about. Sigh. */ + +/* No, there are cases where the SVr4 linker fails to emit COPY relocs + at all. */ + +int +_dl_parse_copy_information (struct dyn_elf *xpnt, int rel_addr, + int rel_size, int type) +{ + int i; + char *strtab; + int reloc_type; + int goof = 0; + struct elf32_sym *symtab; + struct elf32_rela *rpnt; + unsigned int *reloc_addr; + unsigned int symbol_addr; + struct elf_resolve *tpnt; + int symtab_index; + /* Now parse the relocation information */ + + tpnt = xpnt->dyn; + + rpnt = (struct elf32_rela *) (rel_addr + tpnt->loadaddr); + rel_size = rel_size / sizeof (struct elf32_rela); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + + tpnt->loadaddr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + for (i = 0; i < rel_size; i++, rpnt++) + { + reloc_addr = (int *) (tpnt->loadaddr + (int) rpnt->r_offset); + reloc_type = ELF32_R_TYPE (rpnt->r_info); + if (reloc_type != R_68K_COPY) + continue; + symtab_index = ELF32_R_SYM (rpnt->r_info); + symbol_addr = 0; + if (tpnt->libtype == program_interpreter + && (!symtab_index + || _dl_symbol (strtab + symtab[symtab_index].st_name))) + continue; + if (symtab_index) + { + symbol_addr = (unsigned int) + _dl_find_hash (strtab + symtab[symtab_index].st_name, + xpnt->next, (int) reloc_addr, NULL, 1); + if (!symbol_addr) + { + _dl_fdprintf (2, "%s: can't resolve symbol '%s'\n", + _dl_progname, strtab + symtab[symtab_index].st_name); + goof++; + } + } + if (!goof) + _dl_memcpy ((void *) symtab[symtab_index].st_value, (void *) symbol_addr, + symtab[symtab_index].st_size); + } + return goof; +} diff --git a/ldso/ldso/m68k/ld_sysdep.h b/ldso/ldso/m68k/ld_sysdep.h new file mode 100644 index 000000000..24af47ca0 --- /dev/null +++ b/ldso/ldso/m68k/ld_sysdep.h @@ -0,0 +1,87 @@ + +/* Various assmbly language/system dependent hacks that are required + so that we can minimize the amount of platform specific code. */ + +/* Define this if the system uses RELOCA. */ +#define ELF_USES_RELOCA + +/* Get a pointer to the argv array. On many platforms this can be + just the address if the first argument, on other platforms we need + to do something a little more subtle here. */ +#define GET_ARGV(ARGVP, ARGS) ((ARGVP) = ((unsigned int *) &(ARGS))) + +/* Get the address of the Global offset table. This must be absolute, + not relative. */ +#define GET_GOT(X) __asm__ ("movel %%a5,%0" : "=g" (X)) + +/* Initialization sequence for a GOT. */ +#define INIT_GOT(GOT_BASE,MODULE) \ +{ \ + GOT_BASE[2] = (int) _dl_linux_resolve; \ + GOT_BASE[1] = (int) (MODULE); \ +} + +/* Here is a macro to perform a relocation. This is only used when + bootstrapping the dynamic loader. RELP is the relocation that we + are performing, REL is the pointer to the address we are + relocating. SYMBOL is the symbol involved in the relocation, and + LOAD is the load address. */ +#define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD) \ + switch (ELF32_R_TYPE ((RELP)->r_info)) \ + { \ + case R_68K_8: \ + *(char *) (REL) = (SYMBOL) + (RELP)->r_addend; \ + break; \ + case R_68K_16: \ + *(short *) (REL) = (SYMBOL) + (RELP)->r_addend; \ + break; \ + case R_68K_32: \ + *(REL) = (SYMBOL) + (RELP)->r_addend; \ + break; \ + case R_68K_PC8: \ + *(char *) (REL) = ((SYMBOL) + (RELP)->r_addend \ + - (unsigned int) (REL)); \ + break; \ + case R_68K_PC16: \ + *(short *) (REL) = ((SYMBOL) + (RELP)->r_addend \ + - (unsigned int) (REL)); \ + break; \ + case R_68K_PC32: \ + *(REL) = ((SYMBOL) + (RELP)->r_addend \ + - (unsigned int) (REL)); \ + break; \ + case R_68K_GLOB_DAT: \ + case R_68K_JMP_SLOT: \ + *(REL) = (SYMBOL); \ + break; \ + case R_68K_RELATIVE: /* Compatibility kludge */ \ + *(REL) = ((unsigned int) (LOAD) + ((RELP)->r_addend ? : *(REL))); \ + break; \ + default: \ + _dl_exit (1); \ + } + + +/* Transfer control to the user's application, once the dynamic loader + is done. */ + +#define START() \ + __asm__ volatile ("unlk %%a6\n\t" \ + "jmp %0@" \ + : : "a" (_dl_elf_main)); + + + +/* Here we define the magic numbers that this dynamic loader should accept */ + +#define MAGIC1 EM_68K +#undef MAGIC2 +/* Used for error messages */ +#define ELF_TARGET "m68k" + +struct elf_resolve; +extern unsigned int _dl_linux_resolver (int, int, struct elf_resolve *, int); + +/* Define this because we do not want to call .udiv in the library. + Not needed for m68k. */ +#define do_rem(result, n, base) ((result) = (n) % (base)) diff --git a/ldso/ldso/m68k/resolve.S b/ldso/ldso/m68k/resolve.S new file mode 100644 index 000000000..9b1a24c68 --- /dev/null +++ b/ldso/ldso/m68k/resolve.S @@ -0,0 +1,29 @@ +#if 0 +#include <sysdep.h> +#endif +/* + * These are various helper routines that are needed to run an ELF image. + */ + +#ifdef NO_UNDERSCORE +#define __dl_linux_resolve _dl_linux_resolve +#define __dl_linux_resolver _dl_linux_resolver +#endif + +.text +.even + +.globl __dl_linux_resolve + .type __dl_linux_resolve,@function +__dl_linux_resolve: + moveml %a0/%a1,%sp@- +#ifdef __PIC__ + bsrl __dl_linux_resolver@PLTPC +#else + jbsr __dl_linux_resolver +#endif + moveml %sp@+,%a0/%a1 + addql #8,%sp + jmp @(%d0) +.LFE2: + .size __dl_linux_resolve,.LFE2-__dl_linux_resolve diff --git a/ldso/ldso/m68k/sysdep.h b/ldso/ldso/m68k/sysdep.h new file mode 100644 index 000000000..24af47ca0 --- /dev/null +++ b/ldso/ldso/m68k/sysdep.h @@ -0,0 +1,87 @@ + +/* Various assmbly language/system dependent hacks that are required + so that we can minimize the amount of platform specific code. */ + +/* Define this if the system uses RELOCA. */ +#define ELF_USES_RELOCA + +/* Get a pointer to the argv array. On many platforms this can be + just the address if the first argument, on other platforms we need + to do something a little more subtle here. */ +#define GET_ARGV(ARGVP, ARGS) ((ARGVP) = ((unsigned int *) &(ARGS))) + +/* Get the address of the Global offset table. This must be absolute, + not relative. */ +#define GET_GOT(X) __asm__ ("movel %%a5,%0" : "=g" (X)) + +/* Initialization sequence for a GOT. */ +#define INIT_GOT(GOT_BASE,MODULE) \ +{ \ + GOT_BASE[2] = (int) _dl_linux_resolve; \ + GOT_BASE[1] = (int) (MODULE); \ +} + +/* Here is a macro to perform a relocation. This is only used when + bootstrapping the dynamic loader. RELP is the relocation that we + are performing, REL is the pointer to the address we are + relocating. SYMBOL is the symbol involved in the relocation, and + LOAD is the load address. */ +#define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD) \ + switch (ELF32_R_TYPE ((RELP)->r_info)) \ + { \ + case R_68K_8: \ + *(char *) (REL) = (SYMBOL) + (RELP)->r_addend; \ + break; \ + case R_68K_16: \ + *(short *) (REL) = (SYMBOL) + (RELP)->r_addend; \ + break; \ + case R_68K_32: \ + *(REL) = (SYMBOL) + (RELP)->r_addend; \ + break; \ + case R_68K_PC8: \ + *(char *) (REL) = ((SYMBOL) + (RELP)->r_addend \ + - (unsigned int) (REL)); \ + break; \ + case R_68K_PC16: \ + *(short *) (REL) = ((SYMBOL) + (RELP)->r_addend \ + - (unsigned int) (REL)); \ + break; \ + case R_68K_PC32: \ + *(REL) = ((SYMBOL) + (RELP)->r_addend \ + - (unsigned int) (REL)); \ + break; \ + case R_68K_GLOB_DAT: \ + case R_68K_JMP_SLOT: \ + *(REL) = (SYMBOL); \ + break; \ + case R_68K_RELATIVE: /* Compatibility kludge */ \ + *(REL) = ((unsigned int) (LOAD) + ((RELP)->r_addend ? : *(REL))); \ + break; \ + default: \ + _dl_exit (1); \ + } + + +/* Transfer control to the user's application, once the dynamic loader + is done. */ + +#define START() \ + __asm__ volatile ("unlk %%a6\n\t" \ + "jmp %0@" \ + : : "a" (_dl_elf_main)); + + + +/* Here we define the magic numbers that this dynamic loader should accept */ + +#define MAGIC1 EM_68K +#undef MAGIC2 +/* Used for error messages */ +#define ELF_TARGET "m68k" + +struct elf_resolve; +extern unsigned int _dl_linux_resolver (int, int, struct elf_resolve *, int); + +/* Define this because we do not want to call .udiv in the library. + Not needed for m68k. */ +#define do_rem(result, n, base) ((result) = (n) % (base)) diff --git a/ldso/ldso/readelflib1.c b/ldso/ldso/readelflib1.c new file mode 100644 index 000000000..9d1cd0ff5 --- /dev/null +++ b/ldso/ldso/readelflib1.c @@ -0,0 +1,588 @@ +/* Load an ELF sharable library into memory. + + Copyright (C) 1993-1996, Eric Youngdale. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ + + + +/* This file contains the helper routines to load an ELF sharable + library into memory and add the symbol table info to the chain. */ + +#include <linux/types.h> +#include <linux/fcntl.h> +#include <linux/errno.h> +#include "string.h" +/*#include <stdlib.h>*/ +#include <linux/mman.h> +#include <linux/stat.h> +#include "hash.h" +#include "linuxelf.h" +#include "sysdep.h" +#include <linux/unistd.h> +#include "syscall.h" +#ifdef USE_CACHE +#include "../config.h" +#endif + +extern char *_dl_progname; + +#ifdef USE_CACHE + +static caddr_t _dl_cache_addr = NULL; +static size_t _dl_cache_size = 0; + +int _dl_map_cache(void) +{ + int fd; + struct kernel_stat st; + header_t *header; + libentry_t *libent; + int i, strtabsize; + + if (_dl_cache_addr == (caddr_t)-1) + return -1; + else if (_dl_cache_addr != NULL) + return 0; + + if (_dl_stat(LDSO_CACHE, &st) || (fd = _dl_open(LDSO_CACHE, O_RDONLY)) < 0) + { + _dl_fdprintf(2, "%s: can't open cache '%s'\n", _dl_progname, LDSO_CACHE); + _dl_cache_addr = (caddr_t)-1; /* so we won't try again */ + return -1; + } + + _dl_cache_size = st.st_size; + _dl_cache_addr = (caddr_t)_dl_mmap(0, _dl_cache_size, PROT_READ, + MAP_SHARED, fd, 0); + _dl_close (fd); + if (_dl_cache_addr == (caddr_t)-1) + { + _dl_fdprintf(2, "%s: can't map cache '%s'\n", _dl_progname, LDSO_CACHE); + return -1; + } + + header = (header_t *)_dl_cache_addr; + + if (_dl_cache_size < sizeof (header_t) || + _dl_memcmp(header->magic, LDSO_CACHE_MAGIC, LDSO_CACHE_MAGIC_LEN) || + _dl_memcmp(header->version, LDSO_CACHE_VER, LDSO_CACHE_VER_LEN) || + _dl_cache_size < + (sizeof (header_t) + header->nlibs * sizeof (libentry_t)) || + _dl_cache_addr[_dl_cache_size-1] != '\0') + { + _dl_fdprintf(2, "%s: cache '%s' is corrupt\n", _dl_progname, LDSO_CACHE); + goto fail; + } + + strtabsize = _dl_cache_size - sizeof (header_t) - + header->nlibs * sizeof (libentry_t); + libent = (libentry_t *)&header[1]; + + for (i = 0; i < header->nlibs; i++) + { + if (libent[i].sooffset >= strtabsize || + libent[i].liboffset >= strtabsize) + { + _dl_fdprintf(2, "%s: cache '%s' is corrupt\n", _dl_progname, LDSO_CACHE); + goto fail; + } + } + + return 0; + +fail: + _dl_munmap(_dl_cache_addr, _dl_cache_size); + _dl_cache_addr = (caddr_t)-1; + return -1; +} + +int _dl_unmap_cache(void) +{ + if (_dl_cache_addr == NULL || _dl_cache_addr == (caddr_t)-1) + return -1; + +#if 1 + _dl_munmap (_dl_cache_addr, _dl_cache_size); + _dl_cache_addr = NULL; +#endif + + return 0; +} + +#endif + +/* + * Used to return error codes back to dlopen et. al. + */ + +unsigned int _dl_error_number; +unsigned int _dl_internal_error_number; + +struct elf_resolve * _dl_load_shared_library(int secure, + struct elf_resolve * tpnt, char * full_libname) { + char * pnt, *pnt1, *pnt2; + struct elf_resolve *tpnt1 = NULL; + char mylibname[2050]; + char * libname; + + _dl_internal_error_number = 0; + + /* quick hack to ensure mylibname buffer doesn't overflow. don't + allow full_libname or any directory to be longer than 1024. */ + if (_dl_strlen(full_libname) > 1024) + goto goof; + + pnt = libname = full_libname; + while (*pnt) { + if(*pnt == '/') libname = pnt+1; + pnt++; + } + + /* If the filename has any '/', try it straight and leave it at that. + For IBCS2 compatibility under linux, we substitute the string + /usr/i486-sysv4/lib for /usr/lib in library names. */ + + if (libname != full_libname) { + tpnt1 = _dl_load_elf_shared_library(secure, full_libname, 0); + if (tpnt1) + return tpnt1; + goto goof; + } + + /* + * The ABI specifies that RPATH is searched before LD_*_PATH or + * the default path of /usr/lib. + * Check in rpath directories + */ + for (tpnt = _dl_loaded_modules; tpnt; tpnt = tpnt->next) { + if (tpnt->libtype == elf_executable) { + pnt1 = (char *)tpnt->dynamic_info[DT_RPATH]; + if(pnt1) { + pnt1 += (unsigned int) tpnt->loadaddr + tpnt->dynamic_info[DT_STRTAB]; + while(*pnt1){ + pnt2 = mylibname; + while(*pnt1 && *pnt1 != ':') { + if (pnt2 - mylibname < 1024) + *pnt2++ = *pnt1++; + else + pnt1++; + } + if (pnt2 - mylibname >= 1024) + break; + if(pnt2[-1] != '/') *pnt2++ = '/'; + pnt = libname; + while(*pnt) *pnt2++ = *pnt++; + *pnt2++ = 0; + tpnt1 = _dl_load_elf_shared_library(secure, mylibname, 0); + if(tpnt1) return tpnt1; + if(*pnt1 == ':') pnt1++; + } + } + } + } + + + /* Check in LD_{ELF_}LIBRARY_PATH, if specified and allowed */ + pnt1 = _dl_library_path; + if (pnt1 && *pnt1) { + while (*pnt1) { + pnt2 = mylibname; + while(*pnt1 && *pnt1 != ':' && *pnt1 != ';') { + if (pnt2 - mylibname < 1024) + *pnt2++ = *pnt1++; + else + pnt1++; + } + if (pnt2 - mylibname >= 1024) + break; + if(pnt2[-1] != '/') *pnt2++ = '/'; + pnt = libname; + while(*pnt) *pnt2++ = *pnt++; + *pnt2++ = 0; + tpnt1 = _dl_load_elf_shared_library(secure, mylibname, 0); + if(tpnt1) return tpnt1; + if(*pnt1 == ':' || *pnt1 == ';') pnt1++; + } + } + + + /* + * Where should the cache be searched? There is no such concept in the + * ABI, so we have some flexibility here. For now, search it before + * the default path of /usr/lib. + */ +#ifdef USE_CACHE + if (_dl_cache_addr != NULL && _dl_cache_addr != (caddr_t)-1) + { + int i; + header_t *header = (header_t *)_dl_cache_addr; + libentry_t *libent = (libentry_t *)&header[1]; + char *strs = (char *)&libent[header->nlibs]; + + for (i = 0; i < header->nlibs; i++) + { + if ((libent[i].flags == LIB_ELF || + libent[i].flags == LIB_ELF_LIBC5) && + _dl_strcmp(libname, strs+libent[i].sooffset) == 0 && + (tpnt1 = _dl_load_elf_shared_library(secure, strs+libent[i].liboffset, 0))) + return tpnt1; + } + } +#endif + + +#ifdef UCLIBC_DEVEL + + /* Check in /usr/<arch>-linux-uclibc/lib */ + pnt1 = UCLIBC_INSTALL_DIR"/lib"; + pnt = mylibname; + while(*pnt1) *pnt++ = *pnt1++; + pnt1 = libname; + while(*pnt1) *pnt++ = *pnt1++; + *pnt++ = 0; + tpnt1 = _dl_load_elf_shared_library(secure, mylibname, 0); + if (tpnt1) return tpnt1; + +#else /* UCLIBC_DEVEL */ + + /* Check in /usr/lib */ + pnt1 = "/usr/lib/"; + pnt = mylibname; + while(*pnt1) *pnt++ = *pnt1++; + pnt1 = libname; + while(*pnt1) *pnt++ = *pnt1++; + *pnt++ = 0; + tpnt1 = _dl_load_elf_shared_library(secure, mylibname, 0); + if (tpnt1) return tpnt1; + + /* Check in /lib */ + /* try "/lib/". */ + pnt1 = "/lib/"; + pnt = mylibname; + while(*pnt1) *pnt++ = *pnt1++; + pnt1 = libname; + while(*pnt1) *pnt++ = *pnt1++; + *pnt++ = 0; + tpnt1 = _dl_load_elf_shared_library(secure, mylibname, 0); + if (tpnt1) return tpnt1; +#endif /* UCLIBC_DEVEL */ + +goof: + /* Well, we shot our wad on that one. All we can do now is punt */ + if (_dl_internal_error_number) _dl_error_number = _dl_internal_error_number; + else _dl_error_number = DL_ERROR_NOFILE; + return NULL; +} + +/* + * Read one ELF library into memory, mmap it into the correct locations and + * add the symbol info to the symbol chain. Perform any relocations that + * are required. + */ + +//extern _elf_rtbndr(void); + +struct elf_resolve * _dl_load_elf_shared_library(int secure, + char * libname, int flag) { + struct elfhdr * epnt; + unsigned int dynamic_addr = 0; + unsigned int dynamic_size = 0; + struct dynamic * dpnt; + struct elf_resolve * tpnt; + struct elf_phdr * ppnt; + int piclib; + char * status; + int flags; + char header[4096]; + int dynamic_info[24]; + int * lpnt; + unsigned int libaddr; + unsigned int minvma=0xffffffff, maxvma=0; + + int i; + int infile; + + /* If this file is already loaded, skip this step */ + tpnt = _dl_check_hashed_files(libname); + if(tpnt) return tpnt; + + /* If we are in secure mode (i.e. a setu/gid binary using LD_PRELOAD), + we don't load the library if it isn't setuid. */ + + if (secure) { + struct kernel_stat st; + if (_dl_stat(libname, &st) || !(st.st_mode & S_ISUID)) + return NULL; + } + + libaddr = 0; + infile = _dl_open(libname, O_RDONLY); + if(infile < 0) + { +#if 0 + /* + * NO! When we open shared libraries we may search several paths. + * it is inappropriate to generate an error here. + */ + _dl_fdprintf(2, "%s: can't open '%s'\n", _dl_progname, libname); +#endif + _dl_internal_error_number = DL_ERROR_NOFILE; + return NULL; + } + + _dl_read(infile, header, sizeof(header)); + epnt = (struct elfhdr *) header; + if (epnt->e_ident[0] != 0x7f || + epnt->e_ident[1] != 'E' || + epnt->e_ident[2] != 'L' || + epnt->e_ident[3] != 'F') { + _dl_fdprintf(2, "%s: '%s' is not an ELF file\n", _dl_progname, libname); + _dl_internal_error_number = DL_ERROR_NOTELF; + _dl_close(infile); + return NULL; + }; + + if((epnt->e_type != ET_DYN) || + (epnt->e_machine != MAGIC1 +#ifdef MAGIC2 + && epnt->e_machine != MAGIC2 +#endif + )){ + _dl_internal_error_number = (epnt->e_type != ET_DYN ? DL_ERROR_NOTDYN : DL_ERROR_NOTMAGIC); + _dl_fdprintf(2, "%s: '%s' is not an ELF executable for " ELF_TARGET "\n", + _dl_progname, libname); + _dl_close(infile); + return NULL; + }; + + ppnt = (struct elf_phdr *) &header[epnt->e_phoff]; + + piclib = 1; + for(i=0;i < epnt->e_phnum; i++){ + + if(ppnt->p_type == PT_DYNAMIC) { + if (dynamic_addr) + _dl_fdprintf(2, "%s: '%s' has more than one dynamic section\n", + _dl_progname, libname); + dynamic_addr = ppnt->p_vaddr; + dynamic_size = ppnt->p_filesz; + }; + + if(ppnt->p_type == PT_LOAD) { + /* See if this is a PIC library. */ + if(i == 0 && ppnt->p_vaddr > 0x1000000) { + piclib = 0; + minvma=ppnt->p_vaddr; + } + if(piclib && ppnt->p_vaddr < minvma) { + minvma = ppnt->p_vaddr; + } + if(((unsigned int)ppnt->p_vaddr + ppnt->p_memsz) > maxvma) { + maxvma = ppnt->p_vaddr + ppnt->p_memsz; + } + } + ppnt++; + }; + + maxvma=(maxvma+0xfffU)&~0xfffU; + minvma=minvma&~0xffffU; + + flags = MAP_PRIVATE /*| MAP_DENYWRITE*/; + if(!piclib) flags |= MAP_FIXED; + + status = (char *) _dl_mmap((char *) (piclib?0:minvma), + maxvma-minvma, + PROT_NONE, + flags | MAP_ANONYMOUS, -1, + 0); + if(_dl_mmap_check_error(status)) { + _dl_fdprintf(2, "%s: can't map '/dev/zero'\n", _dl_progname); + _dl_internal_error_number = DL_ERROR_MMAP_FAILED; + _dl_close(infile); + return NULL; + }; + libaddr=(unsigned int)status; + flags|=MAP_FIXED; + + /* Get the memory to store the library */ + ppnt = (struct elf_phdr *) &header[epnt->e_phoff]; + + for(i=0;i < epnt->e_phnum; i++){ + if(ppnt->p_type == PT_LOAD) { + + /* See if this is a PIC library. */ + if(i == 0 && ppnt->p_vaddr > 0x1000000) { + piclib = 0; + /* flags |= MAP_FIXED; */ + } + + + + if(ppnt->p_flags & PF_W) { + unsigned int map_size; + char * cpnt; + + status = (char *) _dl_mmap((char *) ((piclib?libaddr:0) + + (ppnt->p_vaddr & 0xfffff000)), + (ppnt->p_vaddr & 0xfff) + ppnt->p_filesz, + LXFLAGS(ppnt->p_flags), + flags, infile, + ppnt->p_offset & 0x7ffff000); + + if(_dl_mmap_check_error(status)) { + _dl_fdprintf(2, "%s: can't map '%s'\n", _dl_progname, libname); + _dl_internal_error_number = DL_ERROR_MMAP_FAILED; + _dl_munmap((char *)libaddr, maxvma-minvma); + _dl_close(infile); + return NULL; + }; + + /* Pad the last page with zeroes. */ + cpnt =(char *) (status + (ppnt->p_vaddr & 0xfff) + ppnt->p_filesz); + while(((unsigned int) cpnt) & 0xfff) *cpnt++ = 0; + +/* I am not quite sure if this is completely correct to do or not, but + the basic way that we handle bss segments is that we mmap /dev/zero if + there are any pages left over that are not mapped as part of the file */ + + map_size = (ppnt->p_vaddr + ppnt->p_filesz + 0xfff) & 0xfffff000; + if(map_size < ppnt->p_vaddr + ppnt->p_memsz) + status = (char *) _dl_mmap((char *) map_size + (piclib?libaddr:0), + ppnt->p_vaddr + ppnt->p_memsz - map_size, + LXFLAGS(ppnt->p_flags), + flags | MAP_ANONYMOUS, -1, 0); + } else + status = (char *) _dl_mmap((char *) (ppnt->p_vaddr & 0xfffff000) + + (piclib?libaddr:0), + (ppnt->p_vaddr & 0xfff) + ppnt->p_filesz, + LXFLAGS(ppnt->p_flags), + flags, infile, + ppnt->p_offset & 0x7ffff000); + if(_dl_mmap_check_error(status)) { + _dl_fdprintf(2, "%s: can't map '%s'\n", _dl_progname, libname); + _dl_internal_error_number = DL_ERROR_MMAP_FAILED; + _dl_munmap((char *)libaddr, maxvma-minvma); + _dl_close(infile); + return NULL; + }; + + /* if(libaddr == 0 && piclib) { + libaddr = (unsigned int) status; + flags |= MAP_FIXED; + }; */ + }; + ppnt++; + }; + _dl_close(infile); + + /* For a non-PIC library, the addresses are all absolute */ + if(piclib) { + dynamic_addr += (unsigned int) libaddr; + } + + /* + * OK, the ELF library is now loaded into VM in the correct locations + * The next step is to go through and do the dynamic linking (if needed). + */ + + /* Start by scanning the dynamic section to get all of the pointers */ + + if(!dynamic_addr) { + _dl_internal_error_number = DL_ERROR_NODYNAMIC; + _dl_fdprintf(2, "%s: '%s' is missing a dynamic section\n", _dl_progname, libname); + return NULL; + } + + dpnt = (struct dynamic *) dynamic_addr; + + dynamic_size = dynamic_size / sizeof(struct dynamic); + _dl_memset(dynamic_info, 0, sizeof(dynamic_info)); + for(i=0; i< dynamic_size; i++){ + if( dpnt->d_tag > DT_JMPREL ) {dpnt++; continue; } + dynamic_info[dpnt->d_tag] = dpnt->d_un.d_val; + if(dpnt->d_tag == DT_TEXTREL || + SVR4_BUGCOMPAT) dynamic_info[DT_TEXTREL] = 1; + dpnt++; + }; + + /* If the TEXTREL is set, this means that we need to make the pages + writable before we perform relocations. Do this now. They get set back + again later. */ + + if (dynamic_info[DT_TEXTREL]) { + ppnt = (struct elf_phdr *) &header[epnt->e_phoff]; + for(i=0;i < epnt->e_phnum; i++, ppnt++){ + if(ppnt->p_type == PT_LOAD && !(ppnt->p_flags & PF_W)) + _dl_mprotect((void *) ((piclib?libaddr:0) + (ppnt->p_vaddr & 0xfffff000)), + (ppnt->p_vaddr & 0xfff) + (unsigned int) ppnt->p_filesz, + PROT_READ | PROT_WRITE | PROT_EXEC); + } + } + + + tpnt = _dl_add_elf_hash_table(libname, (char *) libaddr, dynamic_info, dynamic_addr, + dynamic_size); + + tpnt->ppnt = (struct elf_phdr *) (tpnt->loadaddr + epnt->e_phoff); + tpnt->n_phent = epnt->e_phnum; + + /* + * OK, the next thing we need to do is to insert the dynamic linker into + * the proper entry in the GOT so that the PLT symbols can be properly + * resolved. + */ + + lpnt = (int *) dynamic_info[DT_PLTGOT]; + + if(lpnt) { + lpnt = (int *) (dynamic_info[DT_PLTGOT] + ((int) libaddr)); + INIT_GOT(lpnt, tpnt); + }; + + return tpnt; +} + +/* Ugly, ugly. Some versions of the SVr4 linker fail to generate COPY + relocations for global variables that are present both in the image and + the shared library. Go through and do it manually. If the images + are guaranteed to be generated by a trustworthy linker, then this + step can be skipped. */ + +int _dl_copy_fixups(struct dyn_elf * rpnt) +{ + int goof = 0; + struct elf_resolve * tpnt; + + if(rpnt->next) goof += _dl_copy_fixups(rpnt->next); + else return 0; + + tpnt = rpnt->dyn; + + if (tpnt->init_flag & COPY_RELOCS_DONE) return goof; + tpnt->init_flag |= COPY_RELOCS_DONE; + +#ifdef ELF_USES_RELOCA + goof += _dl_parse_copy_information(rpnt, tpnt->dynamic_info[DT_RELA], + tpnt->dynamic_info[DT_RELASZ], 0); + +#else + goof += _dl_parse_copy_information(rpnt, tpnt->dynamic_info[DT_REL], + tpnt->dynamic_info[DT_RELSZ], 0); + +#endif + return goof; +} + diff --git a/ldso/ldso/sparc/DEFS.h b/ldso/ldso/sparc/DEFS.h new file mode 100644 index 000000000..4b9abccfd --- /dev/null +++ b/ldso/ldso/sparc/DEFS.h @@ -0,0 +1,5 @@ +#define FUNC(name) \ + .global name; \ + .type name,@function; \ + .align 4; \ + name: diff --git a/ldso/ldso/sparc/dl-sysdep.h b/ldso/ldso/sparc/dl-sysdep.h new file mode 100644 index 000000000..1d4c0354f --- /dev/null +++ b/ldso/ldso/sparc/dl-sysdep.h @@ -0,0 +1,131 @@ + +/* + * Various assmbly language/system dependent hacks that are required + * so that we can minimize the amount of platform specific code. + */ +#define LINUXBIN + +/* + * Define this if the system uses RELOCA. + */ +#define ELF_USES_RELOCA + +/* + * Get the address of the Global offset table. This must be absolute, not + * relative. + */ +#define GET_GOT(X) __asm__("\tmov %%l7,%0\n\t" : "=r" (X)) + +/* + * Get a pointer to the argv array. On many platforms this can be just + * the address if the first argument, on other platforms we need to + * do something a little more subtle here. We assume that argc is stored + * at the word just below the argvp that we return here. + */ +#define GET_ARGV(ARGVP, ARGS) __asm__("\tadd %%fp,68,%0\n" : "=r" (ARGVP)); + +/* + * Initialization sequence for a GOT. For the Sparc, this points to the + * PLT, and we need to initialize a couple of the slots. The PLT should + * look like: + * + * save %sp, -64, %sp + * call _dl_linux_resolve + * nop + * .word implementation_dependent + */ +#define INIT_GOT(GOT_BASE,MODULE) \ +{ \ + GOT_BASE[0] = 0x9de3bfc0; /* save %sp, -64, %sp */ \ + GOT_BASE[1] = 0x40000000 | (((unsigned int) _dl_linux_resolve - (unsigned int) GOT_BASE - 4) >> 2); \ + GOT_BASE[2] = 0x01000000; /* nop */ \ + GOT_BASE[3] = (int) MODULE; \ +} + +/* + * Here is a macro to perform a relocation. This is only used when + * bootstrapping the dynamic loader. + */ +#define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD) \ + switch(ELF32_R_TYPE((RELP)->r_info)) { \ + case R_SPARC_32: \ + *REL = SYMBOL + (RELP)->r_addend; \ + break; \ + case R_SPARC_GLOB_DAT: \ + *REL = SYMBOL + (RELP)->r_addend; \ + break; \ + case R_SPARC_JMP_SLOT: \ + REL[1] = 0x03000000 | ((SYMBOL >> 10) & 0x3fffff); \ + REL[2] = 0x81c06000 | (SYMBOL & 0x3ff); \ + break; \ + case R_SPARC_NONE: \ + break; \ + case R_SPARC_WDISP30: \ + break; \ + case R_SPARC_RELATIVE: \ + *REL += (unsigned int) LOAD + (RELP)->r_addend; \ + break; \ + default: \ + _dl_exit(1); \ + } + + +/* + * Transfer control to the user's application, once the dynamic loader + * is done. The crt calls atexit with $g1 if not null, so we need to + * ensure that it contains NULL. + */ + +#define START() \ + __asm__ volatile ( \ + "add %%g0,%%g0,%%g1\n\t" \ + "jmpl %0, %%o7\n\t" \ + "restore %%g0,%%g0,%%g0\n\t" \ + : /*"=r" (status) */ : \ + "r" (_dl_elf_main): "g1", "o0", "o1") + + + +/* Here we define the magic numbers that this dynamic loader should accept */ + +#define MAGIC1 EM_SPARC +#undef MAGIC2 +/* Used for error messages */ +#define ELF_TARGET "Sparc" + +#ifndef COMPILE_ASM +extern unsigned int _dl_linux_resolver(unsigned int reloc_entry, + unsigned int * i); +#endif + +/* + * Define this if you want a dynamic loader that works on Solaris. + */ +#define SOLARIS_COMPATIBLE + +/* + * Define this because we do not want to call .udiv in the library. + * Change on the plans -miguel: + * We just statically link against .udiv. This is required + * if we want to be able to run on Sun4c machines. + */ + +/* We now link .urem against this one */ +#ifdef USE_V8 +#define do_rem(result,n,base) ({ \ +volatile int __res; \ +__asm__("mov %%g0,%%Y\n\t" \ + "sdiv %2,%3,%%l6\n\t" \ + "smul %%l6,%3,%%l6\n\t" \ + "sub %2,%%l6,%0\n\t" \ + :"=r" (result),"=r" (__res):"r" (n),"r"(base) : "l6" ); __res; }) +#else +#define do_rem(a,b,c) a = _dl_urem (b,c); +#endif +/* + * dbx wants the binder to have a specific name. Mustn't disappoint it. + */ +#ifdef SOLARIS_COMPATIBLE +#define _dl_linux_resolve _elf_rtbndr +#endif + diff --git a/ldso/ldso/sparc/elfinterp.c b/ldso/ldso/sparc/elfinterp.c new file mode 100644 index 000000000..6f0d9f8fd --- /dev/null +++ b/ldso/ldso/sparc/elfinterp.c @@ -0,0 +1,355 @@ +/* Run an ELF binary on a linux system. + + Copyright (C) 1995, Eric Youngdale. + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 2, or (at your option) + any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ + +#ifndef VERBOSE_DLINKER +#define VERBOSE_DLINKER +#endif +#ifdef VERBOSE_DLINKER +static char * _dl_reltypes[] = { "R_SPARC_NONE", "R_SPARC_8", + "R_SPARC_16", "R_SPARC_32", "R_SPARC_DISP8", "R_SPARC_DISP16", + "R_SPARC_DISP32", "R_SPARC_WDISP30", "R_SPARC_WDISP22", + "R_SPARC_HI22", "R_SPARC_22", "R_SPARC_13", "R_SPARC_LO10", + "R_SPARC_GOT10", "R_SPARC_GOT13", "R_SPARC_GOT22", "R_SPARC_PC10", + "R_SPARC_PC22", "R_SPARC_WPLT30", "R_SPARC_COPY", + "R_SPARC_GLOB_DAT", "R_SPARC_JMP_SLOT", "R_SPARC_RELATIVE", + "R_SPARC_UA32"}; +#endif + +/* Program to load an ELF binary on a linux system, and run it. +References to symbols in sharable libraries can be resolved by either +an ELF sharable library or a linux style of shared library. */ + +/* Disclaimer: I have never seen any AT&T source code for SVr4, nor have + I ever taken any courses on internals. This program was developed using + information available through the book "UNIX SYSTEM V RELEASE 4, + Programmers guide: Ansi C and Programming Support Tools", which did + a more than adequate job of explaining everything required to get this + working. */ + +#include <linux/types.h> +#include <linux/errno.h> +#include <linux/unistd.h> +/*#include <stdlib.h>*/ +#include "string.h" +#include <linux/unistd.h> +#include <linux/fcntl.h> +#include "hash.h" +#include "linuxelf.h" +#include "sysdep.h" +#include "../syscall.h" +#include "../string.h" + +#define SVR4_COMPATIBILITY + +extern char *_dl_progname; + +extern _dl_linux_resolve(void); + +unsigned int _dl_linux_resolver(unsigned int reloc_entry, unsigned int * plt) +{ + int reloc_type; + struct elf32_rela * this_reloc; + char * strtab; + struct elf32_sym * symtab; + struct elf32_rela * rel_addr; + struct elf_resolve * tpnt; + int symtab_index; + char * new_addr; + char ** got_addr; + unsigned int instr_addr; + tpnt = (struct elf_resolve *) plt[2]; + + rel_addr = (struct elf32_rela *) (tpnt->dynamic_info[DT_JMPREL] + + tpnt->loadaddr); + + /* + * Generate the correct relocation index into the .rela.plt section. + */ + reloc_entry = (reloc_entry >> 12) - 0xc; + + this_reloc = (struct elf32_rela *) ((char *) rel_addr + reloc_entry); + + reloc_type = ELF32_R_TYPE(this_reloc->r_info); + symtab_index = ELF32_R_SYM(this_reloc->r_info); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + tpnt->loadaddr); + strtab = (char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + _dl_fdprintf(2, "tpnt = %x\n", tpnt); + _dl_fdprintf(2, "reloc = %x\n", this_reloc); + _dl_fdprintf(2, "symtab = %x\n", symtab); + _dl_fdprintf(2, "strtab = %x\n", strtab); + + + if (reloc_type != R_SPARC_JMP_SLOT) { + _dl_fdprintf(2, "%s: incorrect relocation type in jump relocations (%d)\n", + _dl_progname, reloc_type); + _dl_exit(30); + }; + + /* Address of jump instruction to fix up */ + instr_addr = ((int)this_reloc->r_offset + (int)tpnt->loadaddr); + got_addr = (char **) instr_addr; + + _dl_fdprintf(2, "symtab_index %d\n", symtab_index); + +#ifdef DEBUG + _dl_fdprintf(2, "Resolving symbol %s\n", + strtab + symtab[symtab_index].st_name); +#endif + + /* Get the address of the GOT entry */ + new_addr = _dl_find_hash(strtab + symtab[symtab_index].st_name, + tpnt->symbol_scope, (int) got_addr, tpnt, 0); + if(!new_addr) { + _dl_fdprintf(2, "%s: can't resolve symbol '%s'\n", + _dl_progname, strtab + symtab[symtab_index].st_name); + _dl_exit(31); + }; +/* #define DEBUG_LIBRARY */ +#ifdef DEBUG_LIBRARY + if((unsigned int) got_addr < 0x40000000) { + _dl_fdprintf(2, "Calling library function: %s\n", + strtab + symtab[symtab_index].st_name); + } else { + got_addr[1] = (char *) (0x03000000 | (((unsigned int) new_addr >> 10) & 0x3fffff)); + got_addr[2] = (char *) (0x81c06000 | ((unsigned int) new_addr & 0x3ff)); + } +#else + got_addr[1] = (char *) (0x03000000 | (((unsigned int) new_addr >> 10) & 0x3fffff)); + got_addr[2] = (char *) (0x81c06000 | ((unsigned int) new_addr & 0x3ff)); +#endif + _dl_fdprintf(2, "Address = %x\n",new_addr); + _dl_exit(32); + + return (unsigned int) new_addr; +} + +void _dl_parse_lazy_relocation_information(struct elf_resolve * tpnt, int rel_addr, + int rel_size, int type){ + int i; + char * strtab; + int reloc_type; + int symtab_index; + struct elf32_sym * symtab; + struct elf32_rela * rpnt; + unsigned int * reloc_addr; + + /* Now parse the relocation information */ + rpnt = (struct elf32_rela *) (rel_addr + tpnt->loadaddr); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + tpnt->loadaddr); + strtab = ( char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + for(i=0; i< rel_size; i += sizeof(struct elf32_rela), rpnt++){ + reloc_addr = (int *) (tpnt->loadaddr + (int)rpnt->r_offset); + reloc_type = ELF32_R_TYPE(rpnt->r_info); + symtab_index = ELF32_R_SYM(rpnt->r_info); + + /* When the dynamic linker bootstrapped itself, it resolved some symbols. + Make sure we do not do them again */ + if(!symtab_index && tpnt->libtype == program_interpreter) continue; + if(symtab_index && tpnt->libtype == program_interpreter && + _dl_symbol(strtab + symtab[symtab_index].st_name)) + continue; + + switch(reloc_type){ + case R_SPARC_NONE: + break; + case R_SPARC_JMP_SLOT: + break; + default: + _dl_fdprintf(2, "%s: (LAZY) can't handle reloc type ", _dl_progname); +#ifdef VERBOSE_DLINKER + _dl_fdprintf(2, "%s ", _dl_reltypes[reloc_type]); +#endif + if(symtab_index) _dl_fdprintf(2, "'%s'\n", + strtab + symtab[symtab_index].st_name); + _dl_exit(33); + }; + }; +} + +int _dl_parse_relocation_information(struct elf_resolve * tpnt, int rel_addr, + int rel_size, int type){ + int i; + char * strtab; + int reloc_type; + int goof = 0; + struct elf32_sym * symtab; + struct elf32_rela * rpnt; + unsigned int * reloc_addr; + unsigned int symbol_addr; + int symtab_index; + /* Now parse the relocation information */ + + rpnt = (struct elf32_rela *) (rel_addr + tpnt->loadaddr); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + tpnt->loadaddr); + strtab = ( char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + for(i=0; i< rel_size; i+= sizeof(struct elf32_rela), rpnt++){ + reloc_addr = (int *) (tpnt->loadaddr + (int)rpnt->r_offset); + reloc_type = ELF32_R_TYPE(rpnt->r_info); + symtab_index = ELF32_R_SYM(rpnt->r_info); + symbol_addr = 0; + + if(!symtab_index && tpnt->libtype == program_interpreter) continue; + + if(symtab_index) { + + if(tpnt->libtype == program_interpreter && + _dl_symbol(strtab + symtab[symtab_index].st_name)) + continue; + + symbol_addr = (unsigned int) + _dl_find_hash(strtab + symtab[symtab_index].st_name, + tpnt->symbol_scope, (int) reloc_addr, + (reloc_type == R_SPARC_JMP_SLOT ? tpnt : NULL), 0); + + if(!symbol_addr && + ELF32_ST_BIND(symtab [symtab_index].st_info) == STB_GLOBAL) { + _dl_fdprintf(2, "%s: can't resolve symbol '%s'\n", + _dl_progname, strtab + symtab[symtab_index].st_name); + goof++; + }; + }; + switch(reloc_type){ + case R_SPARC_NONE: + break; + case R_SPARC_32: + *reloc_addr = symbol_addr + rpnt->r_addend; + break; + case R_SPARC_DISP32: + *reloc_addr = symbol_addr + rpnt->r_addend - (unsigned int) reloc_addr; + break; + case R_SPARC_GLOB_DAT: + *reloc_addr = symbol_addr + rpnt->r_addend; + break; + case R_SPARC_JMP_SLOT: + reloc_addr[1] = 0x03000000 | ((symbol_addr >> 10) & 0x3fffff); + reloc_addr[2] = 0x81c06000 | (symbol_addr & 0x3ff); + break; + case R_SPARC_RELATIVE: + *reloc_addr += (unsigned int) tpnt->loadaddr + rpnt->r_addend; + break; + case R_SPARC_HI22: + if (!symbol_addr) + symbol_addr = tpnt->loadaddr + rpnt->r_addend; + else + symbol_addr += rpnt->r_addend; + *reloc_addr = (*reloc_addr & 0xffc00000)|(symbol_addr >> 10); + break; + case R_SPARC_LO10: + if (!symbol_addr) + symbol_addr = tpnt->loadaddr + rpnt->r_addend; + else + symbol_addr += rpnt->r_addend; + *reloc_addr = (*reloc_addr & ~0x3ff)|(symbol_addr & 0x3ff); + break; + case R_SPARC_WDISP30: + *reloc_addr = (*reloc_addr & 0xc0000000)| + ((symbol_addr - (unsigned int) reloc_addr) >> 2); + break; + case R_SPARC_COPY: +#if 0 /* This one is done later */ + _dl_fdprintf(2, "Doing copy for symbol "); + if(symtab_index) _dl_fdprintf(2, strtab + symtab[symtab_index].st_name); + _dl_fdprintf(2, "\n"); + _dl_memcpy((void *) symtab[symtab_index].st_value, + (void *) symbol_addr, + symtab[symtab_index].st_size); +#endif + break; + default: + _dl_fdprintf(2, "%s: can't handle reloc type ", _dl_progname); +#ifdef VERBOSE_DLINKER + _dl_fdprintf(2, "%s ", _dl_reltypes[reloc_type]); +#endif + if (symtab_index) + _dl_fdprintf(2, "'%s'\n", strtab + symtab[symtab_index].st_name); + _dl_exit(34); + }; + + }; + return goof; +} + + +/* This is done as a separate step, because there are cases where + information is first copied and later initialized. This results in + the wrong information being copied. Someone at Sun was complaining about + a bug in the handling of _COPY by SVr4, and this may in fact be what he + was talking about. Sigh. */ + +/* No, there are cases where the SVr4 linker fails to emit COPY relocs + at all */ + +int _dl_parse_copy_information(struct dyn_elf * xpnt, int rel_addr, + int rel_size, int type) +{ + int i; + char * strtab; + int reloc_type; + int goof = 0; + struct elf32_sym * symtab; + struct elf32_rela * rpnt; + unsigned int * reloc_addr; + unsigned int symbol_addr; + struct elf_resolve *tpnt; + int symtab_index; + /* Now parse the relocation information */ + + tpnt = xpnt->dyn; + + rpnt = (struct elf32_rela *) (rel_addr + tpnt->loadaddr); + + symtab = (struct elf32_sym *) (tpnt->dynamic_info[DT_SYMTAB] + tpnt->loadaddr); + strtab = ( char *) (tpnt->dynamic_info[DT_STRTAB] + tpnt->loadaddr); + + for(i=0; i< rel_size; i+= sizeof(struct elf32_rela), rpnt++){ + reloc_addr = (int *) (tpnt->loadaddr + (int)rpnt->r_offset); + reloc_type = ELF32_R_TYPE(rpnt->r_info); + if(reloc_type != R_SPARC_COPY) continue; + symtab_index = ELF32_R_SYM(rpnt->r_info); + symbol_addr = 0; + if(!symtab_index && tpnt->libtype == program_interpreter) continue; + if(symtab_index) { + + if(tpnt->libtype == program_interpreter && + _dl_symbol(strtab + symtab[symtab_index].st_name)) + continue; + + symbol_addr = (unsigned int) + _dl_find_hash(strtab + symtab[symtab_index].st_name, + xpnt->next, (int) reloc_addr, NULL, 1); + if(!symbol_addr) { + _dl_fdprintf(2, "%s: can't resolve symbol '%s'\n", + _dl_progname, strtab + symtab[symtab_index].st_name); + goof++; + }; + }; + if (!goof) + _dl_memcpy((char *) symtab[symtab_index].st_value, + (char *) symbol_addr, + symtab[symtab_index].st_size); + }; + return goof; +} + + diff --git a/ldso/ldso/sparc/ld_sysdep.h b/ldso/ldso/sparc/ld_sysdep.h new file mode 100644 index 000000000..1d4c0354f --- /dev/null +++ b/ldso/ldso/sparc/ld_sysdep.h @@ -0,0 +1,131 @@ + +/* + * Various assmbly language/system dependent hacks that are required + * so that we can minimize the amount of platform specific code. + */ +#define LINUXBIN + +/* + * Define this if the system uses RELOCA. + */ +#define ELF_USES_RELOCA + +/* + * Get the address of the Global offset table. This must be absolute, not + * relative. + */ +#define GET_GOT(X) __asm__("\tmov %%l7,%0\n\t" : "=r" (X)) + +/* + * Get a pointer to the argv array. On many platforms this can be just + * the address if the first argument, on other platforms we need to + * do something a little more subtle here. We assume that argc is stored + * at the word just below the argvp that we return here. + */ +#define GET_ARGV(ARGVP, ARGS) __asm__("\tadd %%fp,68,%0\n" : "=r" (ARGVP)); + +/* + * Initialization sequence for a GOT. For the Sparc, this points to the + * PLT, and we need to initialize a couple of the slots. The PLT should + * look like: + * + * save %sp, -64, %sp + * call _dl_linux_resolve + * nop + * .word implementation_dependent + */ +#define INIT_GOT(GOT_BASE,MODULE) \ +{ \ + GOT_BASE[0] = 0x9de3bfc0; /* save %sp, -64, %sp */ \ + GOT_BASE[1] = 0x40000000 | (((unsigned int) _dl_linux_resolve - (unsigned int) GOT_BASE - 4) >> 2); \ + GOT_BASE[2] = 0x01000000; /* nop */ \ + GOT_BASE[3] = (int) MODULE; \ +} + +/* + * Here is a macro to perform a relocation. This is only used when + * bootstrapping the dynamic loader. + */ +#define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD) \ + switch(ELF32_R_TYPE((RELP)->r_info)) { \ + case R_SPARC_32: \ + *REL = SYMBOL + (RELP)->r_addend; \ + break; \ + case R_SPARC_GLOB_DAT: \ + *REL = SYMBOL + (RELP)->r_addend; \ + break; \ + case R_SPARC_JMP_SLOT: \ + REL[1] = 0x03000000 | ((SYMBOL >> 10) & 0x3fffff); \ + REL[2] = 0x81c06000 | (SYMBOL & 0x3ff); \ + break; \ + case R_SPARC_NONE: \ + break; \ + case R_SPARC_WDISP30: \ + break; \ + case R_SPARC_RELATIVE: \ + *REL += (unsigned int) LOAD + (RELP)->r_addend; \ + break; \ + default: \ + _dl_exit(1); \ + } + + +/* + * Transfer control to the user's application, once the dynamic loader + * is done. The crt calls atexit with $g1 if not null, so we need to + * ensure that it contains NULL. + */ + +#define START() \ + __asm__ volatile ( \ + "add %%g0,%%g0,%%g1\n\t" \ + "jmpl %0, %%o7\n\t" \ + "restore %%g0,%%g0,%%g0\n\t" \ + : /*"=r" (status) */ : \ + "r" (_dl_elf_main): "g1", "o0", "o1") + + + +/* Here we define the magic numbers that this dynamic loader should accept */ + +#define MAGIC1 EM_SPARC +#undef MAGIC2 +/* Used for error messages */ +#define ELF_TARGET "Sparc" + +#ifndef COMPILE_ASM +extern unsigned int _dl_linux_resolver(unsigned int reloc_entry, + unsigned int * i); +#endif + +/* + * Define this if you want a dynamic loader that works on Solaris. + */ +#define SOLARIS_COMPATIBLE + +/* + * Define this because we do not want to call .udiv in the library. + * Change on the plans -miguel: + * We just statically link against .udiv. This is required + * if we want to be able to run on Sun4c machines. + */ + +/* We now link .urem against this one */ +#ifdef USE_V8 +#define do_rem(result,n,base) ({ \ +volatile int __res; \ +__asm__("mov %%g0,%%Y\n\t" \ + "sdiv %2,%3,%%l6\n\t" \ + "smul %%l6,%3,%%l6\n\t" \ + "sub %2,%%l6,%0\n\t" \ + :"=r" (result),"=r" (__res):"r" (n),"r"(base) : "l6" ); __res; }) +#else +#define do_rem(a,b,c) a = _dl_urem (b,c); +#endif +/* + * dbx wants the binder to have a specific name. Mustn't disappoint it. + */ +#ifdef SOLARIS_COMPATIBLE +#define _dl_linux_resolve _elf_rtbndr +#endif + diff --git a/ldso/ldso/sparc/resolve.S b/ldso/ldso/sparc/resolve.S new file mode 100644 index 000000000..ea985b5c8 --- /dev/null +++ b/ldso/ldso/sparc/resolve.S @@ -0,0 +1,25 @@ +/* + * These are various helper routines that are needed to run an ELF image. + */ +#define COMPILE_ASM +#include "sysdep.h" + +.text + .align 16 + +.globl _dl_linux_resolve +_dl_linux_resolve: + /* + * Call the resolver - pass the address of the PLT so that we can + * figure out which module we are in. + */ + mov %o7,%o1 + call _dl_linux_resolver + mov %g1,%o0 + + jmpl %o0,%o7 + restore +.LFE2: + + .type _dl_linux_resolve,#function + .size _dl_linux_resolve,.LFE2-_dl_linux_resolve diff --git a/ldso/ldso/sparc/sdiv.S b/ldso/ldso/sparc/sdiv.S new file mode 100644 index 000000000..5e52e1959 --- /dev/null +++ b/ldso/ldso/sparc/sdiv.S @@ -0,0 +1,369 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .div name of function to generate + * div div=div => %o0 / %o1; div=rem => %o0 % %o1 + * true true=true => signed; true=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#include "DEFS.h" +#ifndef __linux__ +#ifdef __svr4__ +#include <sys/trap.h> +#else +#include "/usr/include/machine/trap.h" +#endif +#else +#include <asm/traps.h> +#endif + +FUNC(_dl_div) + ! compute sign of result; if neither is negative, no problem + orcc %o1, %o0, %g0 ! either negative? + bge 2f ! no, go do the divide + xor %o1, %o0, %g6 ! compute sign in any case + tst %o1 + bge 1f + tst %o0 + ! %o1 is definitely negative; %o0 might also be negative + bge 2f ! if %o0 not negative... + sub %g0, %o1, %o1 ! in any case, make %o1 nonneg +1: ! %o0 is negative, %o1 is nonnegative + sub %g0, %o0, %o0 ! make %o0 nonnegative +2: + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + sub %o2, 1, %o2 + + +Lgot_result: + ! check to see if answer should be < 0 + tst %g6 + bl,a 1f + sub %g0, %o2, %o2 +1: + retl + mov %o2, %o0 diff --git a/ldso/ldso/sparc/sysdep.h b/ldso/ldso/sparc/sysdep.h new file mode 100644 index 000000000..1d4c0354f --- /dev/null +++ b/ldso/ldso/sparc/sysdep.h @@ -0,0 +1,131 @@ + +/* + * Various assmbly language/system dependent hacks that are required + * so that we can minimize the amount of platform specific code. + */ +#define LINUXBIN + +/* + * Define this if the system uses RELOCA. + */ +#define ELF_USES_RELOCA + +/* + * Get the address of the Global offset table. This must be absolute, not + * relative. + */ +#define GET_GOT(X) __asm__("\tmov %%l7,%0\n\t" : "=r" (X)) + +/* + * Get a pointer to the argv array. On many platforms this can be just + * the address if the first argument, on other platforms we need to + * do something a little more subtle here. We assume that argc is stored + * at the word just below the argvp that we return here. + */ +#define GET_ARGV(ARGVP, ARGS) __asm__("\tadd %%fp,68,%0\n" : "=r" (ARGVP)); + +/* + * Initialization sequence for a GOT. For the Sparc, this points to the + * PLT, and we need to initialize a couple of the slots. The PLT should + * look like: + * + * save %sp, -64, %sp + * call _dl_linux_resolve + * nop + * .word implementation_dependent + */ +#define INIT_GOT(GOT_BASE,MODULE) \ +{ \ + GOT_BASE[0] = 0x9de3bfc0; /* save %sp, -64, %sp */ \ + GOT_BASE[1] = 0x40000000 | (((unsigned int) _dl_linux_resolve - (unsigned int) GOT_BASE - 4) >> 2); \ + GOT_BASE[2] = 0x01000000; /* nop */ \ + GOT_BASE[3] = (int) MODULE; \ +} + +/* + * Here is a macro to perform a relocation. This is only used when + * bootstrapping the dynamic loader. + */ +#define PERFORM_BOOTSTRAP_RELOC(RELP,REL,SYMBOL,LOAD) \ + switch(ELF32_R_TYPE((RELP)->r_info)) { \ + case R_SPARC_32: \ + *REL = SYMBOL + (RELP)->r_addend; \ + break; \ + case R_SPARC_GLOB_DAT: \ + *REL = SYMBOL + (RELP)->r_addend; \ + break; \ + case R_SPARC_JMP_SLOT: \ + REL[1] = 0x03000000 | ((SYMBOL >> 10) & 0x3fffff); \ + REL[2] = 0x81c06000 | (SYMBOL & 0x3ff); \ + break; \ + case R_SPARC_NONE: \ + break; \ + case R_SPARC_WDISP30: \ + break; \ + case R_SPARC_RELATIVE: \ + *REL += (unsigned int) LOAD + (RELP)->r_addend; \ + break; \ + default: \ + _dl_exit(1); \ + } + + +/* + * Transfer control to the user's application, once the dynamic loader + * is done. The crt calls atexit with $g1 if not null, so we need to + * ensure that it contains NULL. + */ + +#define START() \ + __asm__ volatile ( \ + "add %%g0,%%g0,%%g1\n\t" \ + "jmpl %0, %%o7\n\t" \ + "restore %%g0,%%g0,%%g0\n\t" \ + : /*"=r" (status) */ : \ + "r" (_dl_elf_main): "g1", "o0", "o1") + + + +/* Here we define the magic numbers that this dynamic loader should accept */ + +#define MAGIC1 EM_SPARC +#undef MAGIC2 +/* Used for error messages */ +#define ELF_TARGET "Sparc" + +#ifndef COMPILE_ASM +extern unsigned int _dl_linux_resolver(unsigned int reloc_entry, + unsigned int * i); +#endif + +/* + * Define this if you want a dynamic loader that works on Solaris. + */ +#define SOLARIS_COMPATIBLE + +/* + * Define this because we do not want to call .udiv in the library. + * Change on the plans -miguel: + * We just statically link against .udiv. This is required + * if we want to be able to run on Sun4c machines. + */ + +/* We now link .urem against this one */ +#ifdef USE_V8 +#define do_rem(result,n,base) ({ \ +volatile int __res; \ +__asm__("mov %%g0,%%Y\n\t" \ + "sdiv %2,%3,%%l6\n\t" \ + "smul %%l6,%3,%%l6\n\t" \ + "sub %2,%%l6,%0\n\t" \ + :"=r" (result),"=r" (__res):"r" (n),"r"(base) : "l6" ); __res; }) +#else +#define do_rem(a,b,c) a = _dl_urem (b,c); +#endif +/* + * dbx wants the binder to have a specific name. Mustn't disappoint it. + */ +#ifdef SOLARIS_COMPATIBLE +#define _dl_linux_resolve _elf_rtbndr +#endif + diff --git a/ldso/ldso/sparc/udiv.S b/ldso/ldso/sparc/udiv.S new file mode 100644 index 000000000..df4e5385e --- /dev/null +++ b/ldso/ldso/sparc/udiv.S @@ -0,0 +1,351 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .udiv name of function to generate + * div div=div => %o0 / %o1; div=rem => %o0 % %o1 + * false false=true => signed; false=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + +#include "DEFS.h" +#ifndef __linux__ +#ifdef __svr4__ +#include <sys/trap.h> +#else +#include "/usr/include/machine/trap.h" +#endif +#else +#include <asm/traps.h> +#endif + +FUNC(_dl_udiv) + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + sub %o2, 1, %o2 + + +Lgot_result: + + retl + mov %o2, %o0 diff --git a/ldso/ldso/sparc/umul.S b/ldso/ldso/sparc/umul.S new file mode 100644 index 000000000..7a26c295c --- /dev/null +++ b/ldso/ldso/sparc/umul.S @@ -0,0 +1,153 @@ +/* + * Unsigned multiply. Returns %o0 * %o1 in %o1%o0 (i.e., %o1 holds the + * upper 32 bits of the 64-bit product). + * + * This code optimizes short (less than 13-bit) multiplies. Short + * multiplies require 25 instruction cycles, and long ones require + * 45 instruction cycles. + * + * On return, overflow has occurred (%o1 is not zero) if and only if + * the Z condition code is clear, allowing, e.g., the following: + * + * call .umul + * nop + * bnz overflow (or tnz) + */ + +#include "DEFS.h" +FUNC(.umul) + or %o0, %o1, %o4 + mov %o0, %y ! multiplier -> Y + andncc %o4, 0xfff, %g0 ! test bits 12..31 of *both* args + be Lmul_shortway ! if zero, can do it the short way + andcc %g0, %g0, %o4 ! zero the partial product and clear N and V + + /* + * Long multiply. 32 steps, followed by a final shift step. + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %o1, %o4 ! 13 + mulscc %o4, %o1, %o4 ! 14 + mulscc %o4, %o1, %o4 ! 15 + mulscc %o4, %o1, %o4 ! 16 + mulscc %o4, %o1, %o4 ! 17 + mulscc %o4, %o1, %o4 ! 18 + mulscc %o4, %o1, %o4 ! 19 + mulscc %o4, %o1, %o4 ! 20 + mulscc %o4, %o1, %o4 ! 21 + mulscc %o4, %o1, %o4 ! 22 + mulscc %o4, %o1, %o4 ! 23 + mulscc %o4, %o1, %o4 ! 24 + mulscc %o4, %o1, %o4 ! 25 + mulscc %o4, %o1, %o4 ! 26 + mulscc %o4, %o1, %o4 ! 27 + mulscc %o4, %o1, %o4 ! 28 + mulscc %o4, %o1, %o4 ! 29 + mulscc %o4, %o1, %o4 ! 30 + mulscc %o4, %o1, %o4 ! 31 + mulscc %o4, %o1, %o4 ! 32 + mulscc %o4, %g0, %o4 ! final shift + + + /* + * Normally, with the shift-and-add approach, if both numbers are + * positive you get the correct result. With 32-bit two's-complement + * numbers, -x is represented as + * + * x 32 + * ( 2 - ------ ) mod 2 * 2 + * 32 + * 2 + * + * (the `mod 2' subtracts 1 from 1.bbbb). To avoid lots of 2^32s, + * we can treat this as if the radix point were just to the left + * of the sign bit (multiply by 2^32), and get + * + * -x = (2 - x) mod 2 + * + * Then, ignoring the `mod 2's for convenience: + * + * x * y = xy + * -x * y = 2y - xy + * x * -y = 2x - xy + * -x * -y = 4 - 2x - 2y + xy + * + * For signed multiplies, we subtract (x << 32) from the partial + * product to fix this problem for negative multipliers (see mul.s). + * Because of the way the shift into the partial product is calculated + * (N xor V), this term is automatically removed for the multiplicand, + * so we don't have to adjust. + * + * But for unsigned multiplies, the high order bit wasn't a sign bit, + * and the correction is wrong. So for unsigned multiplies where the + * high order bit is one, we end up with xy - (y << 32). To fix it + * we add y << 32. + */ +#if 0 + tst %o1 + bl,a 1f ! if %o1 < 0 (high order bit = 1), + add %o4, %o0, %o4 ! %o4 += %o0 (add y to upper half) +1: rd %y, %o0 ! get lower half of product + retl + addcc %o4, %g0, %o1 ! put upper half in place and set Z for %o1==0 +#else + /* Faster code from tege@sics.se. */ + sra %o1, 31, %o2 ! make mask from sign bit + and %o0, %o2, %o2 ! %o2 = 0 or %o0, depending on sign of %o1 + rd %y, %o0 ! get lower half of product + retl + addcc %o4, %o2, %o1 ! add compensation and put upper half in place +#endif + +Lmul_shortway: + /* + * Short multiply. 12 steps, followed by a final shift step. + * The resulting bits are off by 12 and (32-12) = 20 bit positions, + * but there is no problem with %o0 being negative (unlike above), + * and overflow is impossible (the answer is at most 24 bits long). + */ + mulscc %o4, %o1, %o4 ! 1 + mulscc %o4, %o1, %o4 ! 2 + mulscc %o4, %o1, %o4 ! 3 + mulscc %o4, %o1, %o4 ! 4 + mulscc %o4, %o1, %o4 ! 5 + mulscc %o4, %o1, %o4 ! 6 + mulscc %o4, %o1, %o4 ! 7 + mulscc %o4, %o1, %o4 ! 8 + mulscc %o4, %o1, %o4 ! 9 + mulscc %o4, %o1, %o4 ! 10 + mulscc %o4, %o1, %o4 ! 11 + mulscc %o4, %o1, %o4 ! 12 + mulscc %o4, %g0, %o4 ! final shift + + /* + * %o4 has 20 of the bits that should be in the result; %y has + * the bottom 12 (as %y's top 12). That is: + * + * %o4 %y + * +----------------+----------------+ + * | -12- | -20- | -12- | -20- | + * +------(---------+------)---------+ + * -----result----- + * + * The 12 bits of %o4 left of the `result' area are all zero; + * in fact, all top 20 bits of %o4 are zero. + */ + + rd %y, %o5 + sll %o4, 12, %o0 ! shift middle bits left 12 + srl %o5, 20, %o5 ! shift low bits right 20 + or %o5, %o0, %o0 + retl + addcc %g0, %g0, %o1 ! %o1 = zero, and set Z diff --git a/ldso/ldso/sparc/urem.S b/ldso/ldso/sparc/urem.S new file mode 100644 index 000000000..8d304038b --- /dev/null +++ b/ldso/ldso/sparc/urem.S @@ -0,0 +1,352 @@ + /* This file is generated from divrem.m4; DO NOT EDIT! */ +/* + * Division and remainder, from Appendix E of the Sparc Version 8 + * Architecture Manual, with fixes from Gordon Irlam. + */ + +/* + * Input: dividend and divisor in %o0 and %o1 respectively. + * + * m4 parameters: + * .urem name of function to generate + * rem rem=div => %o0 / %o1; rem=rem => %o0 % %o1 + * false false=true => signed; false=false => unsigned + * + * Algorithm parameters: + * N how many bits per iteration we try to get (4) + * WORDSIZE total number of bits (32) + * + * Derived constants: + * TOPBITS number of bits in the top decade of a number + * + * Important variables: + * Q the partial quotient under development (initially 0) + * R the remainder so far, initially the dividend + * ITER number of main division loop iterations required; + * equal to ceil(log2(quotient) / N). Note that this + * is the log base (2^N) of the quotient. + * V the current comparand, initially divisor*2^(ITER*N-1) + * + * Cost: + * Current estimate for non-large dividend is + * ceil(log2(quotient) / N) * (10 + 7N/2) + C + * A large dividend is one greater than 2^(31-TOPBITS) and takes a + * different path, as the upper bits of the quotient must be developed + * one bit at a time. + */ + + + +#include "DEFS.h" +#ifdef __linux__ +#include <asm/traps.h> +#else +#ifdef __svr4__ +#include <sys/trap.h> +#else +#include "/usr/include/machine/trap.h" +#endif +#endif + +FUNC(_dl_urem) + + ! Ready to divide. Compute size of quotient; scale comparand. + orcc %o1, %g0, %o5 + bne 1f + mov %o0, %o3 + + ! Divide by zero trap. If it returns, return 0 (about as + ! wrong as possible, but that is what SunOS does...). + ta ST_DIV0 + retl + clr %o0 + +1: + cmp %o3, %o5 ! if %o1 exceeds %o0, done + blu Lgot_result ! (and algorithm fails otherwise) + clr %o2 + sethi %hi(1 << (32 - 4 - 1)), %g1 + cmp %o3, %g1 + blu Lnot_really_big + clr %o4 + + ! Here the dividend is >= 2**(31-N) or so. We must be careful here, + ! as our usual N-at-a-shot divide step will cause overflow and havoc. + ! The number of bits in the result here is N*ITER+SC, where SC <= N. + ! Compute ITER in an unorthodox manner: know we need to shift V into + ! the top decade: so do not even bother to compare to R. + 1: + cmp %o5, %g1 + bgeu 3f + mov 1, %g7 + sll %o5, 4, %o5 + b 1b + add %o4, 1, %o4 + + ! Now compute %g7. + 2: addcc %o5, %o5, %o5 + bcc Lnot_too_big + add %g7, 1, %g7 + + ! We get here if the %o1 overflowed while shifting. + ! This means that %o3 has the high-order bit set. + ! Restore %o5 and subtract from %o3. + sll %g1, 4, %g1 ! high order bit + srl %o5, 1, %o5 ! rest of %o5 + add %o5, %g1, %o5 + b Ldo_single_div + sub %g7, 1, %g7 + + Lnot_too_big: + 3: cmp %o5, %o3 + blu 2b + nop + be Ldo_single_div + nop + /* NB: these are commented out in the V8-Sparc manual as well */ + /* (I do not understand this) */ + ! %o5 > %o3: went too far: back up 1 step + ! srl %o5, 1, %o5 + ! dec %g7 + ! do single-bit divide steps + ! + ! We have to be careful here. We know that %o3 >= %o5, so we can do the + ! first divide step without thinking. BUT, the others are conditional, + ! and are only done if %o3 >= 0. Because both %o3 and %o5 may have the high- + ! order bit set in the first step, just falling into the regular + ! division loop will mess up the first time around. + ! So we unroll slightly... + Ldo_single_div: + subcc %g7, 1, %g7 + bl Lend_regular_divide + nop + sub %o3, %o5, %o3 + mov 1, %o2 + b Lend_single_divloop + nop + Lsingle_divloop: + sll %o2, 1, %o2 + bl 1f + srl %o5, 1, %o5 + ! %o3 >= 0 + sub %o3, %o5, %o3 + b 2f + add %o2, 1, %o2 + 1: ! %o3 < 0 + add %o3, %o5, %o3 + sub %o2, 1, %o2 + 2: + Lend_single_divloop: + subcc %g7, 1, %g7 + bge Lsingle_divloop + tst %o3 + b,a Lend_regular_divide + +Lnot_really_big: +1: + sll %o5, 4, %o5 + cmp %o5, %o3 + bleu 1b + addcc %o4, 1, %o4 + be Lgot_result + sub %o4, 1, %o4 + + tst %o3 ! set up for initial iteration +Ldivloop: + sll %o2, 4, %o2 + ! depth 1, accumulated bits 0 + bl L.1.16 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 2, accumulated bits 1 + bl L.2.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits 3 + bl L.3.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 7 + bl L.4.23 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (7*2+1), %o2 + +L.4.23: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (7*2-1), %o2 + + +L.3.19: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 5 + bl L.4.21 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (5*2+1), %o2 + +L.4.21: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (5*2-1), %o2 + + + +L.2.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits 1 + bl L.3.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits 3 + bl L.4.19 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (3*2+1), %o2 + +L.4.19: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (3*2-1), %o2 + + +L.3.17: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits 1 + bl L.4.17 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (1*2+1), %o2 + +L.4.17: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (1*2-1), %o2 + + + + +L.1.16: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 2, accumulated bits -1 + bl L.2.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 3, accumulated bits -1 + bl L.3.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -1 + bl L.4.15 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2+1), %o2 + +L.4.15: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-1*2-1), %o2 + + +L.3.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -3 + bl L.4.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2+1), %o2 + +L.4.13: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-3*2-1), %o2 + + + +L.2.15: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 3, accumulated bits -3 + bl L.3.13 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + ! depth 4, accumulated bits -5 + bl L.4.11 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2+1), %o2 + +L.4.11: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-5*2-1), %o2 + + +L.3.13: + ! remainder is negative + addcc %o3,%o5,%o3 + ! depth 4, accumulated bits -7 + bl L.4.9 + srl %o5,1,%o5 + ! remainder is positive + subcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2+1), %o2 + +L.4.9: + ! remainder is negative + addcc %o3,%o5,%o3 + b 9f + add %o2, (-7*2-1), %o2 + + + + + 9: +Lend_regular_divide: + subcc %o4, 1, %o4 + bge Ldivloop + tst %o3 + bl,a Lgot_result + ! non-restoring fixup here (one instruction only!) + add %o3, %o1, %o3 + + +Lgot_result: + + retl + mov %o3, %o0 diff --git a/ldso/ldso/string.h b/ldso/ldso/string.h new file mode 100644 index 000000000..1ea8fd7ae --- /dev/null +++ b/ldso/ldso/string.h @@ -0,0 +1,112 @@ +#ifndef _LINUX_STRING_H_ +#define _LINUX_STRING_H_ + +#include <linux/types.h> /* for size_t */ + +#ifndef NULL +#define NULL ((void *) 0) +#endif + +extern inline char * _dl_strcpy(char * dst,const char *src) +{ + register char *ptr = dst; + + while (*src) + *dst++ = *src++; + *dst = '\0'; + + return ptr; +} + +extern inline int _dl_strcmp(const char * s1,const char * s2) +{ + unsigned register char c1, c2; + + do { + c1 = (unsigned char) *s1++; + c2 = (unsigned char) *s2++; + if (c1 == '\0') + return c1 - c2; + } + while (c1 == c2); + + return c1 - c2; +} + +extern inline int _dl_strncmp(const char * s1,const char * s2,size_t len) +{ + unsigned register char c1 = '\0'; + unsigned register char c2 = '\0'; + + while (len > 0) { + c1 = (unsigned char) *s1++; + c2 = (unsigned char) *s2++; + if (c1 == '\0' || c1 != c2) + return c1 - c2; + len--; + } + + return c1 - c2; +} + +extern inline char * _dl_strchr(const char * str,int c) +{ + register char ch; + + do { + if ((ch = *str) == c) + return (char *) str; + str++; + } + while (ch); + + return 0; +} + + +extern inline size_t _dl_strlen(const char * str) +{ + register char *ptr = (char *) str; + + while (*ptr) + ptr++; + return (ptr - str); +} + +extern inline void * _dl_memcpy(void * dst, const void * src, size_t len) +{ + register char *a = dst; + register const char *b = src; + + while (len--) + *a++ = *b++; + + return dst; +} + + +extern inline int _dl_memcmp(const void * s1,const void * s2,size_t len) +{ + unsigned char *c1 = (unsigned char *)s1; + unsigned char *c2 = (unsigned char *)s2; + + while (len--) { + if (*c1 != *c2) + return *c1 - *c2; + c1++; + c2++; + } + return 0; +} + +extern inline void * _dl_memset(void * str,int c,size_t len) +{ + register char *a = str; + + while (len--) + *a++ = c; + + return str; +} + +#endif diff --git a/ldso/ldso/syscall.h b/ldso/ldso/syscall.h new file mode 100644 index 000000000..3cf244338 --- /dev/null +++ b/ldso/ldso/syscall.h @@ -0,0 +1,108 @@ +#include <linux/types.h> +#include <asm/unistd.h> + +#ifndef _dl_MAX_ERRNO +#define _dl_MAX_ERRNO 4096 +#endif + +#define _dl_mmap_check_error(__res) \ + (((int)__res) < 0 && ((int)__res) >= -_dl_MAX_ERRNO) + + +/* Here are the definitions for some syscalls that are used + by the dynamic linker. The idea is that we want to be able + to call these before the errno symbol is dynamicly linked, so + we use our own version here. Note that we cannot assume any + dynamic linking at all, so we cannot return any error codes. + We just punt if there is an error. */ + +/* Do not include unistd.h, so gcc doesn't whine about + * _exit returning. It really doesn't return... */ +#define __NR__dl_exit __NR_exit +static inline _syscall1(void, _dl_exit, int, status); + + +#define __NR__dl_close __NR_close +static inline _syscall1(int, _dl_close, int, fd); + + +#define __NR__dl_mmap_real __NR_mmap +static inline _syscall1(void *, _dl_mmap_real, unsigned long *, buffer); + +static inline void * _dl_mmap(void * addr, unsigned long size, int prot, + int flags, int fd, unsigned long offset) +{ + unsigned long buffer[6]; + + buffer[0] = (unsigned long) addr; + buffer[1] = (unsigned long) size; + buffer[2] = (unsigned long) prot; + buffer[3] = (unsigned long) flags; + buffer[4] = (unsigned long) fd; + buffer[5] = (unsigned long) offset; + return (void *) _dl_mmap_real(buffer); +} + +#define __NR__dl_open __NR_open +static inline _syscall2(int, _dl_open, const char *, fn, int, flags); + +#define __NR__dl_write __NR_write +static inline _syscall3(unsigned long, _dl_write, int, fd, + const void *, buf, unsigned long, count); + + +#define __NR__dl_read __NR_read +static inline _syscall3(unsigned long, _dl_read, int, fd, + const void *, buf, unsigned long, count); + +#define __NR__dl_mprotect __NR_mprotect +static inline _syscall3(int, _dl_mprotect, const void *, addr, unsigned long, len, int, prot); + + + +/* Pull in whatever this particular arch's kernel thinks the kernel version of + * struct stat should look like. It turns out that each arch has a different + * opinion on the subject, and different kernel revs use different names... */ +#define __NR__dl_stat __NR_stat +#define stat kernel_stat +#define new_stat kernel_stat +#include <asm/stat.h> +#undef new_stat +#undef stat +static inline _syscall2(int, _dl_stat, const char *, file_name, struct kernel_stat *, buf); + + +#define __NR__dl_munmap __NR_munmap +static inline _syscall2(int, _dl_munmap, void *, start, unsigned long, length); + +#define __NR__dl_getuid __NR_getuid +static inline _syscall0(gid_t, _dl_getuid); + +#define __NR__dl_geteuid __NR_geteuid +static inline _syscall0(uid_t, _dl_geteuid); + +#define __NR__dl_getgid __NR_getgid +static inline _syscall0(gid_t, _dl_getgid); + +#define __NR__dl_getegid __NR_getegid +static inline _syscall0(gid_t, _dl_getegid); + +/* + * Not an actual syscall, but we need something in assembly to say whether + * this is OK or not. + */ +extern inline int _dl_suid_ok(void) +{ + uid_t uid, euid, gid, egid; + + uid = _dl_getuid(); + euid = _dl_geteuid(); + gid = _dl_getgid(); + egid = _dl_getegid(); + + if(uid == euid && gid == egid) + return 1; + else + return 0; +} + diff --git a/ldso/ldso/vsprintf.c b/ldso/ldso/vsprintf.c new file mode 100644 index 000000000..48c44e38f --- /dev/null +++ b/ldso/ldso/vsprintf.c @@ -0,0 +1,290 @@ +/* + * vsprintf.c + * + * Copyright (C) 1991-1996 Linus Torvalds + */ + +/* vsprintf.c -- Lars Wirzenius & Linus Torvalds. */ +/* + * Wirzenius wrote this portably, Torvalds fucked it up :-) + */ + +#include <stdarg.h> +#include "string.h" +#include "hash.h" +#include <linux/unistd.h> +#include "syscall.h" + +/* we use this so that we can do without the ctype library */ +#define is_digit(c) ((c) >= '0' && (c) <= '9') + +static int skip_atoi(const char **s) +{ + int i=0; + + while (is_digit(**s)) + i = i*10 + *((*s)++) - '0'; + return i; +} + +#define ZEROPAD 1 /* pad with zero */ +#define SIGN 2 /* unsigned/signed long */ +#define PLUS 4 /* show plus */ +#define SPACE 8 /* space if plus */ +#define LEFT 16 /* left justified */ +#define SPECIAL 32 /* 0x */ +#define LARGE 64 /* use 'ABCDEF' instead of 'abcdef' */ + +#ifndef __sparc__ +#define do_div(n,base) ({ \ +int __res; \ +__res = ((unsigned long) n) % (unsigned) base; \ +n = ((unsigned long) n) / (unsigned) base; \ +__res; }) +#else +#define do_div(n,base) _dl_div ((n)/(base)) +#define do_div(n,base) ({ \ +int __res; \ +__res = _dl_urem(((unsigned long) n),(unsigned) base); \ +n = _dl_udiv(((unsigned long) n),(unsigned) base); \ +__res; }) +#endif + +#define ADD_CHAR(s,n,c) ( ((n) > 1) ? *(s)++ = (c), (n)-- : (c) ) + +static char * number(char * str, int *bufsize, long num, int base, int size, int precision + ,int type) +{ + char c,sign,tmp[66]; + const char *digits="0123456789abcdefghijklmnopqrstuvwxyz"; + int i; + + if (type & LARGE) + digits = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"; + if (type & LEFT) + type &= ~ZEROPAD; + if (base < 2 || base > 36) + return 0; + c = (type & ZEROPAD) ? '0' : ' '; + sign = 0; + if (type & SIGN) { + if (num < 0) { + sign = '-'; + num = -num; + size--; + } else if (type & PLUS) { + sign = '+'; + size--; + } else if (type & SPACE) { + sign = ' '; + size--; + } + } + if (type & SPECIAL) { + if (base == 16) + size -= 2; + else if (base == 8) + size--; + } + i = 0; + if (num == 0) + tmp[i++]='0'; + else while (num != 0) + tmp[i++] = digits[do_div(num,base)]; + if (i > precision) + precision = i; + size -= precision; + if (!(type&(ZEROPAD+LEFT))) + while(size-->0) + ADD_CHAR(str, *bufsize, ' '); + if (sign) + ADD_CHAR(str, *bufsize, sign); + if (type & SPECIAL) { + if (base==8) + ADD_CHAR(str, *bufsize, '0'); + else if (base==16) { + ADD_CHAR(str, *bufsize, '0'); + ADD_CHAR(str, *bufsize, digits[33]); + } + } + if (!(type & LEFT)) + while (size-- > 0) + ADD_CHAR(str, *bufsize, c); + while (i < precision--) + ADD_CHAR(str, *bufsize, '0'); + while (i-- > 0) + ADD_CHAR(str, *bufsize, tmp[i]); + while (size-- > 0) + ADD_CHAR(str, *bufsize, ' '); + return str; +} + +int _dl_fdprintf(int fd, const char *fmt, ...) +{ + int len; + unsigned long num; + int i, base; + char * str; + const char *s; + + int flags; /* flags to number() */ + + int field_width; /* width of output field */ + int precision; /* min. # of digits for integers; max + number of chars for from string */ + int qualifier; /* 'h', 'l', or 'L' for integer fields */ + + int bufsize; + char buf[2048]; + va_list(args); + + va_start(args, fmt); + + for (str=buf, bufsize=sizeof buf ; *fmt ; ++fmt) { + if (*fmt != '%') { + ADD_CHAR(str, bufsize, *fmt); + continue; + } + + /* process flags */ + flags = 0; + repeat: + ++fmt; /* this also skips first '%' */ + switch (*fmt) { + case '-': flags |= LEFT; goto repeat; + case '+': flags |= PLUS; goto repeat; + case ' ': flags |= SPACE; goto repeat; + case '#': flags |= SPECIAL; goto repeat; + case '0': flags |= ZEROPAD; goto repeat; + } + + /* get field width */ + field_width = -1; + if (is_digit(*fmt)) + field_width = skip_atoi(&fmt); + else if (*fmt == '*') { + ++fmt; + /* it's the next argument */ + field_width = va_arg(args, int); + if (field_width < 0) { + field_width = -field_width; + flags |= LEFT; + } + } + + /* get the precision */ + precision = -1; + if (*fmt == '.') { + ++fmt; + if (is_digit(*fmt)) + precision = skip_atoi(&fmt); + else if (*fmt == '*') { + ++fmt; + /* it's the next argument */ + precision = va_arg(args, int); + } + if (precision < 0) + precision = 0; + } + + /* get the conversion qualifier */ + qualifier = -1; + if (*fmt == 'h' || *fmt == 'l' || *fmt == 'L') { + qualifier = *fmt; + ++fmt; + } + + /* default base */ + base = 10; + + switch (*fmt) { + case 'c': + if (!(flags & LEFT)) + while (--field_width > 0) + ADD_CHAR(str, bufsize, ' '); + ADD_CHAR(str, bufsize, (unsigned char) va_arg(args, int)); + while (--field_width > 0) + ADD_CHAR(str, bufsize, ' '); + continue; + + case 's': + s = va_arg(args, char *); + if (!s) + s = "<NULL>"; + + len = _dl_strlen(s); + + if (!(flags & LEFT)) + while (len < field_width--) + ADD_CHAR(str, bufsize, ' '); + for (i = 0; i < len; ++i) + ADD_CHAR(str, bufsize, *s++); + while (len < field_width--) + ADD_CHAR(str, bufsize, ' '); + continue; + + case 'p': + if (field_width == -1) { + field_width = 2*sizeof(void *); + flags |= ZEROPAD; + } + str = number(str, &bufsize, + (unsigned long) va_arg(args, void *), 16, + field_width, precision, flags); + continue; + + + case 'n': + if (qualifier == 'l') { + long * ip = va_arg(args, long *); + *ip = (str - buf); + } else { + int * ip = va_arg(args, int *); + *ip = (str - buf); + } + continue; + + /* integer number formats - set up the flags and "break" */ + case 'o': + base = 8; + break; + + case 'X': + flags |= LARGE; + case 'x': + base = 16; + break; + + case 'd': + case 'i': + flags |= SIGN; + case 'u': + break; + + default: + if (*fmt != '%') + ADD_CHAR(str, bufsize, '%'); + if (*fmt) + ADD_CHAR(str, bufsize, *fmt); + else + --fmt; + continue; + } + if (qualifier == 'l') + num = va_arg(args, unsigned long); + else if (qualifier == 'h') + if (flags & SIGN) + num = va_arg(args, short); + else + num = va_arg(args, unsigned short); + else if (flags & SIGN) + num = va_arg(args, int); + else + num = va_arg(args, unsigned int); + str = number(str, &bufsize, num, base, field_width, precision, flags); + } + *str = '\0'; + _dl_write(fd, buf, str-buf); + return str-buf; +} + |