#define errno fnord
#include <unistd.h>
#include <fcntl.h>
#include <sys/mman.h>
#include <elf.h>
#include <stdlib.h>
#include <stdint.h>
#undef errno
#if (__WORDSIZE == 64)
#define phdr Elf64_Phdr
#define ehdr Elf64_Ehdr
#define shdr Elf64_Shdr
#define sym Elf64_Sym
#define dyn Elf64_Dyn
#define rela Elf64_Rela
#define R_SYM ELF64_R_SYM
#define R_TYPE ELF64_R_TYPE
#else
#define phdr Elf32_Phdr
#define ehdr Elf32_Ehdr
#define shdr Elf32_Shdr
#define sym Elf32_Sym
#define dyn Elf32_Dyn
#define rela Elf32_Rela
#define R_SYM ELF32_R_SYM
#define R_TYPE ELF32_R_TYPE
#endif
static int errno;
__attribute__((visibility("hidden"))) int* __errno_location(void) { return &errno; }
static size_t _strlen(const char* s) {
size_t i;
for (i=0; s[i]; ++i);
return i;
}
static char* _stpcpy(char* dest,const char* src) {
size_t i;
for (i=0; src[i]; ++i)
dest[i]=src[i];
dest[i]=0;
return dest+i;
}
static char* _strchr(char* s,char c) {
size_t i;
for (i=0; s[i] && s[i]!=c; ++i) ;
if (s[i]==c) return s+i; else return NULL;
}
static int _strcmp(const void* str1,const void* str2) {
const unsigned char* a=str1;
const unsigned char* b=str2;
size_t i;
int r;
for (i=0; (r=(a[i]-b[i]))==0 && a[i]; ++i) ;
return r;
}
static int _memcmp(const void* dst,const void* src,size_t count) {
const unsigned char* a=dst;
const unsigned char* b=src;
size_t i;
int r;
for (i=0; i<count && (r=(a[i]-b[i]))==0; ++i) ;
return r;
}
static void* _memcpy(void* dst,const void* src,size_t len) {
char* a=dst;
const char* b=src;
size_t i;
for (i=0; i<len; ++i)
a[i]=b[i];
return dst;
}
static void _memset(void* dst,unsigned char c,size_t len) {
unsigned char* a=dst;
size_t i;
for (i=0; i<len; ++i) a[i]=c;
}
static ssize_t __write1(const char* s) {
write(1,s,_strlen(s));
}
static ssize_t __write2(const char* s) {
write(2,s,_strlen(s));
}
ssize_t write(int fd,const void* buf,size_t len) __attribute__((visibility("hidden")));
int open(const char* pathname,int flags, ...) __attribute__((visibility("hidden")));
ssize_t read(int fd,void* buf,size_t len) __attribute__((visibility("hidden")));
int close(int fd) __attribute__((visibility("hidden")));
ssize_t pread64(int fd, void *buf, size_t count, off64_t offset) __attribute__((visibility("hidden")));;
void *mmap(void *__addr, size_t __len, int __prot, int __flags, int __fd, off_t __offset) __attribute__((visibility("hidden")));
int munmap(void *__addr, size_t __len) __attribute__((visibility("hidden")));
int mprotect (void *__addr, size_t __len, int __prot) __attribute__((visibility("hidden")));
void exit(int res) __attribute__((visibility("hidden")));
static struct page {
struct page* next;
size_t a;
char data[4096-sizeof(size_t)-sizeof(struct page*)];
}* heap;
static void* _malloc(size_t l) {
struct page** p;
if (l>sizeof(heap->data)) return 0;
if (l%(sizeof(void*)*2)) {
l += sizeof(void*)*2;
l -= l%(sizeof(void*)*2);
}
for (p=&heap; *p && (*p)->a<l; p=&((*p)->next)) ;
if (!*p) {
void* tmp=mmap(0,4096,PROT_READ|PROT_WRITE,MAP_ANONYMOUS|MAP_PRIVATE,-1,0);
if (tmp==MAP_FAILED) return 0;
*p=tmp;
(*p)->a=sizeof(heap->data);
}
if (l <= (*p)->a) {
char* tmp=(*p)->data+sizeof((*p)->data)-(*p)->a;
(*p)->a-=l;
return tmp;
} else
return 0; // can't happen
}
static char path[100];
static char* ldlp;
static struct dll {
struct dll* next;
ehdr* e;
void* code,* data;
size_t codelen,datalen,codeplus;
char name[1]; // asciiz of library name
} *dlls, dllroot;
static int map_sections(int fd,const ehdr* e,const phdr* p,struct dll* D) {
size_t i;
uintptr_t codeplus=0;
for (i=0; i<e->e_phnum; ++i) {
if (p[i].p_type==PT_LOAD) {
size_t delta=p[i].p_offset%4096;
size_t maplen=p[i].p_filesz+delta;
size_t bssdiff=(p[i].p_filesz+delta)%4096;
char* c;
if ((p[i].p_flags&PF_W) && (p[i].p_flags&PF_X)) {
__write2("section is both executable and writable, aborting!\n");
return 1;
}
if (!(p[i].p_flags&PF_W)) {
/* code segment */
size_t ofs,len,rolen=0,nolen=0,rolen2=0,vaddr=p[i].p_vaddr,baseofs=0;
/* the first segment will be the code segment, and it will have
* either a fixed address or 0 if it's a shared library. We
* insist that the mapping start at file offset 0, and we extend
* the mapping so it includes the section table */
ofs=p[i].p_offset;
len=p[i].p_filesz;
if (ofs) {
__write2("can't happen error: ofs!=0\n");
exit(1);
if (vaddr)
vaddr-=ofs;
else
baseofs=ofs;
rolen=ofs;
len+=ofs;
ofs=0;
}
if (ofs+len < e->e_shoff+e->e_shnum*e->e_shentsize) {
size_t needed=e->e_shoff+e->e_shnum*e->e_shentsize;
/* if this mapping does not include the section table is not
* included, extend the mapping to include it */
rolen2=e->e_shnum*e->e_shentsize;
if (rolen2>needed-len)
/* we were almost there, part of the section table was
* already mapped */
rolen2=needed-len;
else
nolen=needed-len-rolen2;
/*
* +------------------------+
* | rolen e-> | if the mapping did not start at beginning of file, this is the beginning of file, PROT_READ
* +------------------------+
* | len base-> | this is the actual mapping, base points here, PROT_READ|PROT_EXEC
* +------------------------+
* | nolen | stuff we don't really need and will mprotect PROT_NONE
* +------------------------+
* | rolen2 e+e->shoff | the section header table, PROT_READ
* +------------------------+
*/
}
c=mmap((char*)vaddr,rolen+len+nolen+rolen2,
((p[i].p_flags&PF_R)?PROT_READ:0) |
((p[i].p_flags&PF_X)?PROT_EXEC:0),
MAP_SHARED|(vaddr?MAP_FIXED:0),
fd,0);
/* in case the can't happen branch ever happens */
D->e=(ehdr*)c;
D->code=c+rolen; D->codelen=len;
// D->s=(shdr*)(c+e->e_shoff);
if (rolen>=4096) /* if we extended the mapping in the front, remove exec permissions */
mprotect(c,rolen&~4095,PROT_READ);
if (!vaddr && !codeplus) codeplus=(uintptr_t)(c+rolen);
if (nolen) {
/* We mapped junk in the middle.
* If there are full pages in there, map them PROT_NONE */
char* start=c+rolen+len;
size_t len=nolen;
size_t diff=(-(uintptr_t)start & 4095);
if (diff < nolen) {
/* diff is the part at the beginning we need to skip because
* it's on a page we actually need to be executable.
* Now find out if we overshoot onto a page we want */
size_t removeatend=((uintptr_t)c+rolen+len)&4095;
mprotect(start+diff,len-diff-removeatend,PROT_NONE);
}
}
if (rolen2) {
/* Now we want to mprotect PROT_READ the section table.
* What makes this complex is that mprotect granularity is one
* page. First figure out the region we are interested in. */
char* start=c+rolen+len+nolen;
size_t len=rolen2;
/* we want to mprotect from start to start+len */
int diff=((uintptr_t)start)&4095;
/* This is the offset of start from the page start. We need
* to mprotect the whole page -- unless it also has code on
* it, in which case we need to leave it alone */
if (diff > nolen) {
size_t roundup=(-(uintptr_t)start)&4095;
/* we need to skip the first page. Let's see if there is
* anything left in that case */
start+=roundup;
if (rolen2>roundup)
len-=roundup;
else
len=0;
} else {
start-=diff;
len+=diff;
}
if (len) mprotect(start,len,PROT_READ);
}
} else if (p[i].p_flags&PF_W) { /* data segment */
/* We have already mapped the code segment to base.
* For programs, the base mapping of the code and data segment
* is absolute and we just map there with MAP_FIXED. For shared
* libraries however, the base mapping of the code segment is 0,
* and the vaddr of the data segment is not absolute but
* relative to the code segment */
/* The data segment may not start on a page boundary. Round
* start address if necessary. */
size_t addr,ofs,len,memsetstart=0,memsetend=0,additional=0;
addr=p[i].p_vaddr+codeplus;
ofs=p[i].p_offset;
len=p[i].p_filesz;
if ((memsetstart=(ofs%4096))) {
len+=memsetstart;
ofs-=memsetstart;
addr-=memsetstart;
}
/* The data segment consists of actual data, but a part of it is
* data initialized to 0, the .bss section. This section is not
* actually put in the file to save space, but the ELF loader
* (that's us) is expected to allocate that data at program
* start. */
memsetend=p[i].p_memsz-p[i].p_filesz;
if (memsetend) {
/* We have a .bss. We need to handle two cases. First: if
* the number of bytes is small enough to fit on the last page
* we already mapped for the data, all we need to do is memset
* it. Otherwise we needto memset the rest of that page and
* map some additional pages after it. */
size_t bytes_on_last_page=(-len)&4095;
if (memsetend<bytes_on_last_page)
len+=memsetend;
else {
len+=bytes_on_last_page;
additional=memsetend-bytes_on_last_page;
memsetend=bytes_on_last_page;
}
}
c=mmap((char*)addr,len,PROT_READ|PROT_WRITE,MAP_PRIVATE|MAP_FIXED,fd,ofs);
if (c==MAP_FAILED) {
__write2("mmap failed!\n");
return 1;
}
if (memsetstart) _memset(c,0,memsetstart);
if (memsetend) _memset(c+len-memsetend,0,memsetend);
if (additional) {
char* tmp=mmap(c+len,additional,PROT_READ|PROT_WRITE,MAP_PRIVATE|MAP_ANONYMOUS,-1,0);
if (tmp==MAP_FAILED) {
__write2("mmap failed!\n");
return 1;
}
}
D->data=c+memsetstart; D->datalen=len-memsetstart;
D->codeplus=codeplus;
} else {
__write2("can't happen error: LOAD segment that is neither code nor data.\n");
return 1;
}
}
}
return 0;
}
static int __loadlibrary(const char* fn) {
struct dll* D;
int fd;
char buf[1000];
ehdr* e;
phdr* p;
shdr* s;
size_t i;
char* code=0,* data=0;
char* base=0;
size_t codelen,datalen;
#if 0
__write1("trying ");
__write1(fn);
__write1("\n");
#endif
fd=open(fn,O_RDONLY);
if (fd==-1) return -1;
if (read(fd,buf,1000)<1000) {
kaputt:
close(fd);
return -1;
}
if (_memcmp(buf,"\177ELF",4)) goto kaputt;
e=(ehdr*)buf;
if (e->e_ident[4] != (sizeof(size_t)/4) || // wrong word size
e->e_type != ET_DYN || // not shared object
e->e_machine != // for different architecture
#if defined(__x86_64__)
EM_X86_64
#elif defined(__i386__)
EM_386
#else
#error architecture not recognized, please add
#endif
||
e->e_phentsize != sizeof(phdr) || // wrong program header entry size
e->e_phnum > 20 || // too many program header entries
e->e_shentsize !=sizeof(shdr) || // wrong section header entry size
e->e_shnum > 100) // too many sections
goto kaputt;
if (e->e_phoff>=1000 || e->e_phentsize*e->e_phnum>1000-e->e_phoff) {
size_t wanted=e->e_phentsize*e->e_phnum;
p=alloca(wanted);
if (pread64(fd,p,wanted,e->e_phoff)!=wanted)
goto kaputt;
} else
p=(phdr*)(buf+e->e_phoff);
struct dll dll;
if (map_sections(fd,e,p,&dll)) {
__write2("map_sections failed\n");
goto kaputt;
}
close(fd);
{
const char* tmp;
for (i=0; fn[i]; ++i)
if (fn[i]=='/') tmp=fn+i;
++tmp;
D=_malloc(sizeof(struct dll)+_strlen(tmp));
_stpcpy(D->name,tmp);
}
D->next=0;
D->code=dll.code; D->codelen=dll.codelen;
D->data=dll.data; D->datalen=dll.datalen;
// D->s=dll.s;
D->e=dll.e;
D->codeplus=dll.codeplus;
{
struct dll** x;
for (x=&dlls; *x; x=&(*x)->next) ;
*x=D;
}
return 0;
}
static int loadlibrary(const char* fn) {
char lp[200];
int r;
char* c;
const char* shortname=fn;
struct dll* d;
{
size_t i;
for (i=0; fn[i]; ++i)
if (fn[i]=='/') shortname=fn+i+1;
}
if (_strlen(fn)>50) return -1;
for (d=dlls; d; d=d->next)
if (!_strcmp(d->name,shortname))
return 0;
__write1("loadlibrary(\"");
__write1(fn);
__write1("\")\n");
if (fn[0]=='/') {
return __loadlibrary(fn);
}
c=_stpcpy(lp,path);
*c++='/';
_stpcpy(c,fn);
r=__loadlibrary(lp);
if (r==0) return r;
if (ldlp) {
size_t i;
char* d;
c=ldlp;
for (i=0; ; ++i) {
again:
if (c[i]==':' || c[i]==0) {
if (i<100) {
lp[i]='/';
_stpcpy(lp+i+1,fn);
r=__loadlibrary(lp);
if (r==0) return;
}
if (c[i]==0) break;
c+=i+1; i=0; goto again;
} else
if (i<100) lp[i]=c[i];
}
}
return r;
}
static int loadlibs(struct dll* D) {
size_t i;
phdr* p=(phdr*)((char*)D->e+D->e->e_phoff);
dyn* d;
size_t dnum,dynstrlen;
char* dynstr;
/* we know we have exactly one code and exactly one data segment,
* otherwise we wouldn't have gotten this far */
for (i=0; i<D->e->e_phnum; ++i)
if (p[i].p_type==PT_DYNAMIC) {
d=(dyn*)((char*)p[i].p_vaddr+D->codeplus);
dnum=p[i].p_memsz/sizeof(dyn);
break;
}
for (i=0; i<dnum; ++i)
if (d[i].d_tag==DT_STRTAB) {
dynstr=(char*)d[i].d_un.d_ptr+D->codeplus;
break;
} else if (d[i].d_tag==DT_NULL)
break;
/* we now have a dynamic section we can traverse */
for (i=0; i<dnum; ++i) {
if (d[i].d_tag==DT_NEEDED) {
if (loadlibrary(dynstr+d[i].d_un.d_val)) {
__write2("library ");
__write2(dynstr+d[i].d_un.d_val);
__write2(" not found!\n");
exit(2);
}
} else if (d[i].d_tag==DT_NULL)
break;
}
return 0;
}
static unsigned int elf_hash(const unsigned char *name) {
unsigned int h=0, g;
while (*name) {
h = (h<<4) + *(name++);
if ((g = h&0xf0000000)) h ^= g>>24;
h ^= g;
}
return h;
}
static uint_fast32_t gnu_hash(const unsigned char *s) {
unsigned char c;
uint_fast32_t h=5381;
for(c=*s;(c!='\0');c=*++s) {
// h=h*33+c;
h=((h<<5)+h)+c;
}
return (h&0xffffffff);
}
static char* dlsym_int(const char* symbol,struct dll* x) {
for (; x; x=x->next) {
size_t i;
dyn* d;
sym* sy;
phdr* p=(phdr*)(x->e->e_phoff+(char*)x->e);
const char* strtab;
size_t dnum;
int* hash=0;
for (i=0; i<x->e->e_phnum; ++i)
if (p[i].p_type==PT_DYNAMIC) {
d=(dyn*)(x->codeplus + p[i].p_vaddr);
dnum=p[i].p_memsz/sizeof(dyn);
break;
}
for (i=0; i<dnum; ++i) {
if (d[i].d_tag == DT_HASH)
hash=(int*)((char*)x->codeplus + d[i].d_un.d_ptr);
else if (d[i].d_tag == DT_SYMTAB)
sy=(sym*)((char*)(x->codeplus+d[i].d_un.d_ptr));
else if (d[i].d_tag == DT_STRTAB)
strtab=(char*)(x->codeplus+d[i].d_un.d_ptr);
}
/* hash[0] is the number of buckets. */
/* hash[1] is the hash chain length, not used here */
size_t bhash=elf_hash(symbol)%hash[0];
unsigned int* chain=hash+2+hash[0];
unsigned int index;
for (index=(hash+2)[bhash]; index; index=chain[index]) {
#if 0
__write1(strtab+sy[index].st_name); __write1("\n");
#endif
if (sy[index].st_value && sy[index].st_shndx!=SHN_UNDEF && !_strcmp(strtab+sy[index].st_name,symbol)) {
return (char*)x->codeplus+sy[index].st_value;
}
}
#if 0
if (x->next) {
__write1(" ... next: ");
__write1(x->next->name);
__write1("\n");
}
#endif
}
return 0;
}
static void* dlsym(const char* s) {
return dlsym_int(s,&dllroot);
}
static void* _dlsym(const char* s) {
void* x=dlsym(s);
if (!x) {
__write2("ld.so: lookup of symbol \"");
__write2(s);
__write2("\" failed.\n");
// exit(1);
}
return x;
}
static void resolve(struct dll* D) {
size_t i;
phdr* p=(phdr*)((char*)D->e+D->e->e_phoff);
dyn* d=0;
size_t dnum,dynstrlen,rnum=0;
char* dynstr=0, *pltgot=0, *pltrel=0;
rela* r=0;
sym* symtab=0;
/* we know we have exactly one code and exactly one data segment,
* otherwise we wouldn't have gotten this far */
for (i=0; i<D->e->e_phnum; ++i)
if (p[i].p_type==PT_DYNAMIC) {
d=(dyn*)((char*)p[i].p_vaddr+D->codeplus);
dnum=p[i].p_memsz/sizeof(dyn);
break;
}
for (i=0; i<dnum; ++i)
if (d[i].d_tag==DT_STRTAB)
dynstr=(char*)d[i].d_un.d_ptr+D->codeplus;
else if (d[i].d_tag==DT_RELA)
r=(rela*)((char*)d[i].d_un.d_ptr+D->codeplus);
else if (d[i].d_tag==DT_RELASZ)
rnum=d[i].d_un.d_val/sizeof(rela);
else if (d[i].d_tag==DT_SYMTAB)
symtab=(sym*)((char*)d[i].d_un.d_ptr+D->codeplus);
else if (d[i].d_tag==0)
break;
for (i=0; i<rnum; ++i) {
size_t* x=(size_t*)((char*)(r[i].r_offset+D->codeplus));
char* y;
size_t sym=R_SYM(r[i].r_info);
switch (R_TYPE(r[i].r_info)) {
#if defined(__x86_64__)
case R_X86_64_64:
*x=D->codeplus+symtab[sym].st_value;
break;
case R_X86_64_COPY:
y=dlsym_int(symtab[sym].st_name+dynstr,D->next);
if (!y && ELF32_ST_BIND(symtab[sym].st_info) != STB_WEAK) {
__write2("symbol lookup failed: ");
__write2(dynstr+symtab[sym].st_name);
__write2("\n");
exit(1);
}
_memcpy(x,y,symtab[sym].st_size);
break;
case R_X86_64_GLOB_DAT:
case R_X86_64_JUMP_SLOT:
y=dlsym(symtab[sym].st_name+dynstr);
if (!y && ELF32_ST_BIND(symtab[sym].st_info) != STB_WEAK) {
__write2("symbol lookup failed: ");
__write2(dynstr+symtab[sym].st_name);
__write2("\n");
exit(1);
}
*x=(uintptr_t)y;
break;
case R_X86_64_RELATIVE:
*x=r[i].r_addend+D->codeplus;
break;
case R_X86_64_32:
*(uint32_t*)x=*(uint32_t*)_dlsym(symtab[sym].st_name+dynstr)+r[i].r_addend;
break;
default:
__write2("unknown relocation type!\n");
exit(1);
break;
#else
#error fixme: add relocation types for your platform
#endif
}
}
return;
}
int main(int argc,char* argv[],char* envp[]) {
int fd;
size_t l;
char* m;
char buf[1000];
ehdr* e;
phdr* p;
phdr* code=0,* data=0;
shdr* s;
dyn* d;
size_t dnum;
char* dynstr;
size_t dynstrlen;
size_t i;
#if 0
{
fd=open("/proc/self/maps",O_RDONLY);
if (fd!=-1) {
size_t l;
do {
l=read(fd,buf,sizeof(buf));
write(1,buf,l);
} while (l==sizeof(buf));
close(fd);
}
}
#endif
if (argc<2) {
__write2("usage: ld.so /path/to/binary\n");
return 0;
}
fd=open("/etc/diet.ld.conf",O_RDONLY);
if (fd!=-1) {
int r=read(fd,path,99);
if (r>0) path[r]=0;
while (r>0 && path[r-1]=='\n') path[--r]=0;
close(fd);
}
for (i=0; envp[i]; ++i) {
if (_memcmp(envp[i],"LD_LIBRARY_PATH=",16)==0)
ldlp=envp[i]+16;
}
fd=open(argv[1],O_RDONLY);
if (fd==-1) {
__write2("could not open \"");
__write2(argv[1]);
__write2("\".\n");
return 1;
}
l=read(fd,buf,1000);
if (l<sizeof(ehdr) || _memcmp(buf,"\177ELF",4)) {
kaputt:
__write2("invalid ELF file \"");
close(fd);
return 1;
}
e=(ehdr*)buf;
if (e->e_ident[4] != (sizeof(size_t)/4)) {
__write2("wrong word size!\n");
return 1;
}
#if 0
if (e->e_ident[7] != ELFOSABI_LINUX) {
__write2("ABI not Linux!\n");
return 1;
}
#endif
if (e->e_type != ET_EXEC) {
__write2("not an executable!\n");
return 1;
}
if (e->e_machine !=
#if defined(__x86_64__)
EM_X86_64
#elif defined(__i386__)
EM_386
#else
#error architecture not recognized, please add
#endif
) {
__write2("invalid architecture!\n");
return 1;
}
if (e->e_phentsize != sizeof(phdr)) {
__write2("invalid phentsize!\n");
return 1;
}
if (e->e_phnum > 20) {
__write2("phnum too large!\n");
return 1;
}
if (e->e_shentsize != sizeof(shdr)) {
__write2("invalid shentsize!\n");
return 1;
}
if (e->e_shnum > 100) {
__write2("shnum too large!\n");
return 1;
}
if (e->e_phoff>=l || e->e_phentsize*e->e_phnum>l-e->e_phoff) {
size_t wanted=e->e_phentsize*e->e_phnum;
p=alloca(wanted);
if (pread64(fd,p,wanted,e->e_phoff)!=wanted)
goto kaputt;
} else
p=(phdr*)(buf+e->e_phoff);
dlls=&dllroot;
if (map_sections(fd,e,p,&dllroot)) {
__write2("map_sections failed!\n");
return 1;
}
close(fd);
loadlibs(&dllroot);
/* now load the prerequisites of the libraries we loaded */
{
struct dll* x;
for (x=dlls; x; x=x->next) {
loadlibs(x);
}
}
resolve(&dllroot);
__write2("jumping...\n");
{
int (*_init)(int argc,char* argv[],char* envp[])=(void*)(e->e_entry+dllroot.codeplus);
return _init(argc,argv,envp);
}
#if 0
{
char* x=dlsym("theint");
__write1("done\n");
}
#endif
#if 0
printf("jump to %p\n",e->e_entry);
for (i=0; i<16; ++i) {
printf("%02x ",((unsigned char*)e->e_entry)[i]);
}
#endif
return 0;
}
LinuxTV legacy CVS <linuxtv.org/cvs>
![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to ldso.c CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [DVB] / dietlibc