// quick sort: requires n>1
static void qsort_map_array(map_elem_t *a, uint32_t n) {
uint32_t i, j;
map_elem_t x, y;
// x = random pivot
i = random_uint(n);
x = a[i];
// swap a[i] and a[0]
a[i] = a[0];
a[0] = x;
i = 0;
j = n;
do { j--; } while (a[j].index > x.index);
do { i++; } while (i <= j && a[i].index < x.index);
while (i < j) {
// swap a[i] and a[j]
y = a[i]; a[i] = a[j]; a[j] = y;
do { j--; } while (a[j].index > x.index);
do { i++; } while (a[i].index < x.index);
}
// swap a[0] = x and a[j]
a[0] = a[j];
a[j] = x;
// recursive sort
sort_map_array(a, j);
j ++;
sort_map_array(a + j, n - j);
}
/*
* Normalize map:
* - sort all elements in increasing index order
*/
void normalize_map(map_t *map) {
sort_map_array(map->data, map->nelems);
}
// 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;
}
