// process segments from interpreter (ld.so or ld-linux.so)
static bool read_interp_segments(struct ps_prochandle* ph) {
ELF_EHDR interp_ehdr;
if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) {
print_debug("interpreter is not a valid ELF file\n");
return false;
}
if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) {
print_debug("can't read segments of interpreter\n");
return false;
}
return true;
}
int open_elf(const char *filename, elf_bin_t *bin)
{
int err;
bin->fd = -1;
if ( (err = open_exec(filename)) < 0 )
return err;
bin->fd = err;
if ( (err = read_elf_header(bin->fd, &bin->hdr)) < 0 )
return err;
return 0;
}
/* Main entry point of elf-parser */
int32_t main(int32_t argc, char *argv[])
{
int32_t fd;
Elf32_Ehdr eh; /* elf-header is fixed size */
if(argc!=2) {
printf("Usage: elf-parser <ELF-file>\n");
return 0;
}
fd = open(argv[1], O_RDONLY|O_SYNC);
if(fd<0) {
printf("Error %d Unable to open %s\n", fd, argv[1]);
return 0;
}
/* ELF header : at start of file */
read_elf_header(fd, &eh);
if(!is_ELF(eh)) {
return 0;
}
if(is64Bit(eh)){
Elf64_Ehdr eh64; /* elf-header is fixed size */
Elf64_Shdr* sh_tbl; /* section-header table is variable size */
read_elf_header64(fd, &eh64);
print_elf_header64(eh64);
/* Section header table : */
sh_tbl = malloc(eh64.e_shentsize * eh64.e_shnum);
if(!sh_tbl) {
printf("Failed to allocate %d bytes\n",
(eh64.e_shentsize * eh64.e_shnum));
}
read_section_header_table64(fd, eh64, sh_tbl);
print_section_headers64(fd, eh64, sh_tbl);
/* Symbol tables :
* sh_tbl[i].sh_type
* |`- SHT_SYMTAB
* `- SHT_DYNSYM
*/
print_symbols64(fd, eh64, sh_tbl);
/* Save .text section as text.S
*/
save_text_section64(fd, eh64, sh_tbl);
/* Disassemble .text section
* Logs asm instructions to stdout
* Currently supports ARMv7
*/
disassemble64(fd, eh64, sh_tbl);
} else{
Elf32_Shdr* sh_tbl; /* section-header table is variable size */
print_elf_header(eh);
/* Section header table : */
sh_tbl = malloc(eh.e_shentsize * eh.e_shnum);
if(!sh_tbl) {
printf("Failed to allocate %d bytes\n",
(eh.e_shentsize * eh.e_shnum));
}
read_section_header_table(fd, eh, sh_tbl);
print_section_headers(fd, eh, sh_tbl);
/* Symbol tables :
* sh_tbl[i].sh_type
* |`- SHT_SYMTAB
* `- SHT_DYNSYM
*/
print_symbols(fd, eh, sh_tbl);
/* Save .text section as text.S
*/
save_text_section(fd, eh, sh_tbl);
/* Disassemble .text section
* Logs asm instructions to stdout
* Currently supports ARMv7
*/
disassemble(fd, eh, sh_tbl);
}
return 0;
}
// the one and only one exposed stuff from this file
struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
ELF_EHDR core_ehdr;
ELF_EHDR exec_ehdr;
ELF_EHDR lib_ehdr;
struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
if (ph == NULL) {
print_debug("can't allocate ps_prochandle\n");
return NULL;
}
if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
free(ph);
print_debug("can't allocate ps_prochandle\n");
return NULL;
}
// initialize ph
ph->ops = &core_ops;
ph->core->core_fd = -1;
ph->core->exec_fd = -1;
ph->core->interp_fd = -1;
// open the core file
if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
print_debug("can't open core file\n");
goto err;
}
// read core file ELF header
if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) {
print_debug("core file is not a valid ELF ET_CORE file\n");
goto err;
}
if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
print_debug("can't open executable file\n");
goto err;
}
if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true || exec_ehdr.e_type != ET_EXEC) {
print_debug("executable file is not a valid ELF ET_EXEC file\n");
goto err;
}
// process core file segments
if (read_core_segments(ph, &core_ehdr) != true)
goto err;
// process exec file segments
if (read_exec_segments(ph, &exec_ehdr) != true)
goto err;
// exec file is also treated like a shared object for symbol search
if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd,
(uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL)
goto err;
// allocate and sort maps into map_array, we need to do this
// here because read_shared_lib_info needs to read from debuggee
// address space
if (sort_map_array(ph) != true)
goto err;
if (read_shared_lib_info(ph) != true)
goto err;
// sort again because we have added more mappings from shared objects
if (sort_map_array(ph) != true)
goto err;
if (init_classsharing_workaround(ph) != true)
goto err;
return ph;
err:
Prelease(ph);
return NULL;
}
// read shared library info from runtime linker's data structures.
// This work is done by librtlb_db in Solaris
static bool read_shared_lib_info(struct ps_prochandle* ph) {
uintptr_t addr = ph->core->dynamic_addr;
uintptr_t debug_base;
uintptr_t first_link_map_addr;
uintptr_t ld_base_addr;
uintptr_t link_map_addr;
uintptr_t lib_base_diff;
uintptr_t lib_base;
uintptr_t lib_name_addr;
char lib_name[BUF_SIZE];
ELF_DYN dyn;
ELF_EHDR elf_ehdr;
int lib_fd;
// _DYNAMIC has information of the form
// [tag] [data] [tag] [data] .....
// Both tag and data are pointer sized.
// We look for dynamic info with DT_DEBUG. This has shared object info.
// refer to struct r_debug in link.h
dyn.d_tag = DT_NULL;
while (dyn.d_tag != DT_DEBUG) {
if (ps_pdread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) {
print_debug("can't read debug info from _DYNAMIC\n");
return false;
}
addr += sizeof(ELF_DYN);
}
// we have got Dyn entry with DT_DEBUG
debug_base = dyn.d_un.d_ptr;
// at debug_base we have struct r_debug. This has first link map in r_map field
if (ps_pdread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET,
&first_link_map_addr, sizeof(uintptr_t)) != PS_OK) {
print_debug("can't read first link map address\n");
return false;
}
// read ld_base address from struct r_debug
if (ps_pdread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr,
sizeof(uintptr_t)) != PS_OK) {
print_debug("can't read ld base address\n");
return false;
}
ph->core->ld_base_addr = ld_base_addr;
print_debug("interpreter base address is 0x%lx\n", ld_base_addr);
// now read segments from interp (i.e ld.so or ld-linux.so)
if (read_interp_segments(ph) != true)
return false;
// after adding interpreter (ld.so) mappings sort again
if (sort_map_array(ph) != true)
return false;
print_debug("first link map is at 0x%lx\n", first_link_map_addr);
link_map_addr = first_link_map_addr;
while (link_map_addr != 0) {
// read library base address of the .so. Note that even though <sys/link.h> calls
// link_map->l_addr as "base address", this is * not * really base virtual
// address of the shared object. This is actually the difference b/w the virtual
// address mentioned in shared object and the actual virtual base where runtime
// linker loaded it. We use "base diff" in read_lib_segments call below.
if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET,
&lib_base_diff, sizeof(uintptr_t)) != PS_OK) {
print_debug("can't read shared object base address diff\n");
return false;
}
// read address of the name
if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET,
&lib_name_addr, sizeof(uintptr_t)) != PS_OK) {
print_debug("can't read address of shared object name\n");
return false;
}
// read name of the shared object
lib_name[0] = '\0';
if (lib_name_addr != 0 &&
read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) {
print_debug("can't read shared object name\n");
// don't let failure to read the name stop opening the file. If something is really wrong
// it will fail later.
}
if (lib_name[0] != '\0') {
// ignore empty lib names
lib_fd = pathmap_open(lib_name);
if (lib_fd < 0) {
print_debug("can't open shared object %s\n", lib_name);
// continue with other libraries...
} else {
if (read_elf_header(lib_fd, &elf_ehdr)) {
lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr);
print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n",
lib_name, lib_base, lib_base_diff);
// while adding library mappings we need to use "base difference".
if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) {
print_debug("can't read shared object's segments\n");
close(lib_fd);
return false;
}
add_lib_info_fd(ph, lib_name, lib_fd, lib_base);
// Map info is added for the library (lib_name) so
// we need to re-sort it before calling the p_pdread.
if (sort_map_array(ph) != true)
return false;
} else {
print_debug("can't read ELF header for shared object %s\n", lib_name);
close(lib_fd);
// continue with other libraries...
}
}
}
// read next link_map address
if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET,
&link_map_addr, sizeof(uintptr_t)) != PS_OK) {
print_debug("can't read next link in link_map\n");
return false;
}
}
return true;
}
/*
Map the built-in lisp core sections.
NOTE! We need to do this without using malloc because the memory layout
is not set until some time after this is done.
*/
void
map_core_sections(const char *exec_name)
{
int exec_fd;
Elf_Shdr sh; /* A section header entry. */
Elf_Shdr strsecent;
char nambuf[10];
int soff;
int strsecoff; /* File offset to string table section. */
int sections_remaining = 3;
int i, j;
extern int
image_dynamic_space_size,
image_static_space_size,
image_read_only_space_size;
if (!(exec_fd = open(exec_name, O_RDONLY))) {
perror("Can't open executable!");
exit(-1);
}
read_elf_header(exec_fd, &eh);
/* Find the section name string section. Save its file offset. */
soff = eh.e_shoff + eh.e_shstrndx * eh.e_shentsize;
elseek(exec_fd, soff, __func__);
read_section_header_entry(exec_fd, &strsecent);
strsecoff = strsecent.sh_offset;
for (i = 0; i < eh.e_shnum && sections_remaining > 0; i++) {
/* Read an entry from the section header table. */
elseek(exec_fd, eh.e_shoff + i * eh.e_shentsize, __func__);
read_section_header_entry(exec_fd, &sh);
/* Read the name from the string section. */
elseek(exec_fd, strsecoff + sh.sh_name, __func__);
eread(exec_fd, nambuf, 6, __func__);
if (sh.sh_type == SHT_PROGBITS) {
/* See if this section is one of the lisp core sections. */
for (j = 0; j < 3; j++) {
if (!strncmp(nambuf, section_names[j], 6)) {
os_vm_address_t addr;
/*
* Found a core section. Map it!
*
* Although the segment may contain the correct
* address for the start of the segment, we don't
* care. We infer the address from the segment
* name. (The names better be unique!!!!) This
* approach allows for a possibly simpler linking
* operation because we don't have to figure out
* how to get the linker to give segments the
* correct address.
*/
addr = section_addr[j];
if ((os_vm_address_t) os_map(exec_fd, sh.sh_offset,
addr, sh.sh_size)
== (os_vm_address_t) -1) {
fprintf(stderr, "%s: Can't map section %s\n", __func__, section_names[j]);
exit(-1);
}
switch(j) {
case 0: /* Dynamic space. */
/* Dynamic space variables are set in lisp.c. */
image_dynamic_space_size = sh.sh_size;
break;
case 1: /* Static space. */
image_static_space_size = sh.sh_size;
break;
case 2: /* Read-only space. */
image_read_only_space_size = sh.sh_size;
break;
default:
/* Should never get here. */
abort();
break;
}
sections_remaining--;
/* Found a core section, don't check the other core section names. */
break;
}
}
}
}
close(exec_fd);
if (sections_remaining != 0) {
fprintf(stderr, "Couldn't map all core sections! Exiting!\n");
exit(-1);
}
}
// read symbol table from given fd.
struct symtab* build_symtab(int fd) {
ELF_EHDR ehdr;
struct symtab* symtab = NULL;
// Reading of elf header
struct elf_section *scn_cache = NULL;
int cnt = 0;
ELF_SHDR* shbuf = NULL;
ELF_SHDR* cursct = NULL;
ELF_PHDR* phbuf = NULL;
int symtab_found = 0;
int dynsym_found = 0;
uint32_t symsection = SHT_SYMTAB;
uintptr_t baseaddr = (uintptr_t)-1;
lseek(fd, (off_t)0L, SEEK_SET);
if (! read_elf_header(fd, &ehdr)) {
// not an elf
return NULL;
}
// read ELF header
if ((shbuf = read_section_header_table(fd, &ehdr)) == NULL) {
goto quit;
}
baseaddr = find_base_address(fd, &ehdr);
scn_cache = calloc(ehdr.e_shnum, sizeof(*scn_cache));
if (scn_cache == NULL) {
goto quit;
}
for (cursct = shbuf, cnt = 0; cnt < ehdr.e_shnum; cnt++) {
scn_cache[cnt].c_shdr = cursct;
if (cursct->sh_type == SHT_SYMTAB ||
cursct->sh_type == SHT_STRTAB ||
cursct->sh_type == SHT_DYNSYM) {
if ( (scn_cache[cnt].c_data = read_section_data(fd, &ehdr, cursct)) == NULL) {
goto quit;
}
}
if (cursct->sh_type == SHT_SYMTAB)
symtab_found++;
if (cursct->sh_type == SHT_DYNSYM)
dynsym_found++;
cursct++;
}
if (!symtab_found && dynsym_found)
symsection = SHT_DYNSYM;
for (cnt = 1; cnt < ehdr.e_shnum; cnt++) {
ELF_SHDR *shdr = scn_cache[cnt].c_shdr;
if (shdr->sh_type == symsection) {
ELF_SYM *syms;
int j, n;
size_t size;
// FIXME: there could be multiple data buffers associated with the
// same ELF section. Here we can handle only one buffer. See man page
// for elf_getdata on Solaris.
// guarantee(symtab == NULL, "multiple symtab");
symtab = calloc(1, sizeof(*symtab));
if (symtab == NULL) {
goto quit;
}
// the symbol table
syms = (ELF_SYM *)scn_cache[cnt].c_data;
// number of symbols
n = shdr->sh_size / shdr->sh_entsize;
// create hash table, we use berkeley db to
// manipulate the hash table.
symtab->hash_table = dbopen(NULL, O_CREAT | O_RDWR, 0600, DB_HASH, NULL);
// guarantee(symtab->hash_table, "unexpected failure: dbopen");
if (symtab->hash_table == NULL)
goto bad;
// shdr->sh_link points to the section that contains the actual strings
// for symbol names. the st_name field in ELF_SYM is just the
// string table index. we make a copy of the string table so the
// strings will not be destroyed by elf_end.
size = scn_cache[shdr->sh_link].c_shdr->sh_size;
symtab->strs = malloc(size);
if (symtab->strs == NULL)
goto bad;
memcpy(symtab->strs, scn_cache[shdr->sh_link].c_data, size);
// allocate memory for storing symbol offset and size;
symtab->num_symbols = n;
symtab->symbols = calloc(n , sizeof(*symtab->symbols));
if (symtab->symbols == NULL)
goto bad;
// copy symbols info our symtab and enter them info the hash table
for (j = 0; j < n; j++, syms++) {
DBT key, value;
char *sym_name = symtab->strs + syms->st_name;
// skip non-object and non-function symbols
int st_type = ELF_ST_TYPE(syms->st_info);
if ( st_type != STT_FUNC && st_type != STT_OBJECT)
continue;
// skip empty strings and undefined symbols
if (*sym_name == '\0' || syms->st_shndx == SHN_UNDEF) continue;
symtab->symbols[j].name = sym_name;
symtab->symbols[j].offset = syms->st_value - baseaddr;
symtab->symbols[j].size = syms->st_size;
key.data = sym_name;
key.size = strlen(sym_name) + 1;
value.data = &(symtab->symbols[j]);
value.size = sizeof(symtab_symbol);
(*symtab->hash_table->put)(symtab->hash_table, &key, &value, 0);
}
}
}
goto quit;
bad:
destroy_symtab(symtab);
symtab = NULL;
quit:
if (shbuf) free(shbuf);
if (phbuf) free(phbuf);
if (scn_cache) {
for (cnt = 0; cnt < ehdr.e_shnum; cnt++) {
if (scn_cache[cnt].c_data != NULL) {
free(scn_cache[cnt].c_data);
}
}
free(scn_cache);
}
return symtab;
}
void
elf_mod_symload(int strtablen)
{
Elf_Ehdr ehdr;
char *shstrtab;
struct elf_section *head, *s;
char *symbuf, *strbuf;
/*
* Seek to the text offset to start loading...
*/
if (lseek(modfd, 0, SEEK_SET) == -1)
err(12, "lseek");
if (read_elf_header(modfd, &ehdr) < 0)
return;
shstrtab = read_shstring_table(modfd, &ehdr);
read_sections(modfd, &ehdr, shstrtab, &head);
for (s = head; s; s = s->next) {
struct elf_section *p = s;
if ((p->type == SHT_SYMTAB) || (p->type == SHT_DYNSYM)) {
if (debug)
fprintf(stderr, "loading `%s': addr = %p, "
"size = %#lx\n",
s->name, s->addr, (u_long)s->size);
/*
* Seek to the file offset to start loading it...
*/
if (lseek(modfd, p->offset, SEEK_SET) == -1)
err(12, "lseek");
symbuf = malloc(p->size);
if (symbuf == 0)
err(13, "malloc");
if (read(modfd, symbuf, p->size) != p->size)
err(14, "read");
loadsym(symbuf, p->size);
free(symbuf);
}
}
for (s = head; s; s = s->next) {
struct elf_section *p = s;
if ((p->type == SHT_STRTAB) &&
(strcmp(p->name, ".strtab") == 0 )) {
if (debug)
fprintf(stderr, "loading `%s': addr = %p, "
"size = %#lx\n",
s->name, s->addr, (u_long)s->size);
/*
* Seek to the file offset to start loading it...
*/
if (lseek(modfd, p->offset, SEEK_SET) == -1)
err(12, "lseek");
strbuf = malloc(p->size);
if (strbuf == 0)
err(13, "malloc");
if (read(modfd, strbuf, p->size) != p->size)
err(14, "read");
loadsym(strbuf, p->size);
free(strbuf);
}
}
free(shstrtab);
free_sections(head);
return;
}
/* load a prelinked module; returns entry point */
void *
elf_mod_load(int fd)
{
Elf_Ehdr ehdr;
size_t zero_size = 0;
size_t b;
ssize_t n;
char *shstrtab;
struct elf_section *head, *s;
char buf[10 * BUFSIZ];
void *addr = NULL;
if (read_elf_header(fd, &ehdr) < 0)
return NULL;
shstrtab = read_shstring_table(fd, &ehdr);
read_sections(fd, &ehdr, shstrtab, &head);
for (s = head; s; s = s->next) {
if (s->type != SHT_STRTAB && s->type != SHT_SYMTAB &&
s->type != SHT_DYNSYM) {
if (debug)
fprintf(stderr, "loading `%s': addr = %p, "
"size = %#lx\n",
s->name, s->addr, (u_long)s->size);
if (s->type == SHT_NOBITS) {
/* skip some space */
zero_size += s->size;
} else {
if (addr != NULL) {
/*
* if there is a gap in the prelinked
* module, transfer some empty space.
*/
zero_size += (char*)s->addr -
(char*)addr;
}
if (zero_size) {
loadspace(zero_size);
zero_size = 0;
}
b = s->size;
if (lseek(fd, s->offset, SEEK_SET) == -1)
err(1, "lseek");
while (b) {
n = read(fd, buf, MIN(b, sizeof(buf)));
if (n == 0)
errx(1, "unexpected EOF");
if (n < 0)
err(1, "read");
loadbuf(buf, n);
b -= n;
}
addr = (char*)s->addr + s->size;
}
}
}
if (zero_size)
loadspace(zero_size);
free_sections(head);
free(shstrtab);
return (void *)ehdr.e_entry;
}
/* return size needed by the module */
int
elf_mod_sizes(int fd, size_t *modsize, int *strtablen,
struct lmc_resrv *resrvp, struct stat *sp)
{
Elf_Ehdr ehdr;
ssize_t off = 0;
size_t data_hole = 0;
char *shstrtab, *strtab;
struct elf_section *head, *s, *stab;
if (read_elf_header(fd, &ehdr) < 0)
return -1;
shstrtab = read_shstring_table(fd, &ehdr);
read_sections(fd, &ehdr, shstrtab, &head);
for (s = head; s; s = s->next) {
/* XXX impossible! */
if (s->type == SHT_STRTAB && s->type == SHT_SYMTAB &&
s->type == SHT_DYNSYM)
continue;
if (debug)
fprintf(stderr,
"%s: addr = %p size = %#lx align = %#lx\n",
s->name, s->addr, (u_long)s->size, (u_long)s->align);
/*
* XXX try to get rid of the hole before the data
* section that GNU-ld likes to put there
*/
if (strcmp(s->name, ".data") == 0 && s->addr > (void *)off) {
data_offset = roundup(off, s->align);
if (debug)
fprintf(stderr, ".data section forced to "
"offset %p (was %p)\n",
(void *)data_offset, s->addr);
/* later remove size of compressed hole from off */
data_hole = (ssize_t)s->addr - data_offset;
}
off = (ssize_t)s->addr + s->size;
}
off -= data_hole;
/* XXX round to pagesize? */
*modsize = roundup(off, sysconf(_SC_PAGESIZE));
/* get string table length */
strtab = read_string_table(fd, head, strtablen);
free(shstrtab);
free(strtab);
/* get symbol table sections */
get_symtab(&head);
stab = head;
resrvp->sym_symsize = 0;
while (stab) {
resrvp->sym_symsize += stab->size;
stab = stab->next;
}
resrvp->sym_size = resrvp->sym_symsize + *strtablen;
free_sections(head);
return (0);
}
