dwarf-assembly/stack_walker/stack_walker.cpp

252 lines
6.6 KiB
C++

#include "stack_walker.hpp"
#include <ucontext.h>
#include <unistd.h>
#include <link.h>
#include <cstring>
#include <cassert>
#include <cctype>
#include <cstdio>
#include <string>
#include <map>
#define UNUSED(x) (void)(x)
typedef void* dl_handle_t;
/** Describes a line in the memory map (which SO is loaded where) */
struct MemoryMapEntry {
MemoryMapEntry():
beg(0), end(0), offset(0), obj_path(), eh_dl_handle(nullptr)
{}
uintptr_t beg, end;
int offset;
std::string obj_path;
dl_handle_t eh_dl_handle;
};
/** `MemoryMapEntry`es mapped by their `beg` */
typedef std::map<
uintptr_t,
MemoryMapEntry,
std::greater<uintptr_t> > MemoryMap;
// here, std::greater is used because it allows easy access to the
// corresponding segment by using lower_bound
static MemoryMap memory_map;
/** Equivalent to a shell command `readlink -f` */
std::string readlink_rec(const char* path) {
char buf[2][1024];
int parity = 1;
strcpy(buf[1], path);
do {
ssize_t rc = readlink(buf[parity], buf[1-parity], 1024);
parity = 1 - parity;
if(rc < 0)
break;
else
buf[parity][rc] = '\0';
} while(true);
return std::string(buf[1 - parity]);
}
/** Called by `dl_iterate_phdr` later, initializes `memory_map` */
int fill_memory_map_callback(
struct dl_phdr_info* info,
size_t /*size*/,
void* /*data*/)
{
for(int sec = 0; sec < info->dlpi_phnum; ++sec) {
const ElfW(Phdr)& cur_hdr = info->dlpi_phdr[sec];
if(cur_hdr.p_type != PT_LOAD || (cur_hdr.p_flags & PF_X) == 0)
continue;
if(std::string(info->dlpi_name).find("linux-vdso")
!= std::string::npos)
{
continue;
}
MemoryMapEntry entry;
entry.beg = info->dlpi_addr + cur_hdr.p_vaddr;
entry.obj_path = std::string(info->dlpi_name);
entry.offset = cur_hdr.p_offset;
entry.end = entry.beg + cur_hdr.p_memsz;
if(entry.obj_path.empty()) { // The source binary itself
entry.obj_path = readlink_rec("/proc/self/exe");
}
memory_map.insert(std::make_pair(entry.beg, entry));
}
return 0;
}
void stack_walker_close() {
for(auto& mmap_entry_pair: memory_map) {
auto& mmap_entry = mmap_entry_pair.second;
if(mmap_entry.eh_dl_handle != nullptr)
dlclose(mmap_entry.eh_dl_handle);
}
}
bool stack_walker_init() {
if(dl_iterate_phdr(&fill_memory_map_callback, nullptr) != 0) {
stack_walker_close();
return false;
}
for(const auto& mmap_entry: memory_map) {
printf("%012lx-%012lx %08x %s\n",
mmap_entry.second.beg,
mmap_entry.second.end,
mmap_entry.second.offset,
mmap_entry.second.obj_path.c_str());
}
// Call `dlopen` on the `eh_elf.so` maching every entry in `memory_map`
for(auto& mmap_entry_pair: memory_map) {
auto& mmap_entry = mmap_entry_pair.second;
// Find SO's basename
size_t last_slash = mmap_entry.obj_path.rfind("/");
if(last_slash == std::string::npos)
last_slash = 0;
else
last_slash++;
std::string basename(mmap_entry.obj_path, last_slash);
// Load the SO
std::string eh_elf_name = basename + ".eh_elf.so";
mmap_entry.eh_dl_handle = dlopen(eh_elf_name.c_str(), RTLD_LAZY);
if(mmap_entry.eh_dl_handle == nullptr) {
fprintf(stderr,
"Error: cannot load shared object %s.\ndlerror: %s\n",
eh_elf_name.c_str(),
dlerror());
stack_walker_close();
return false;
}
}
return true;
}
unwind_context_t get_context() {
unwind_context_t out;
ucontext_t uctx;
if(getcontext(&uctx) < 0) {
assert(0); // Cleaner code will come later — TODO
}
out.rip = uctx.uc_mcontext.gregs[REG_RIP];
out.rsp = uctx.uc_mcontext.gregs[REG_RSP];
out.rbp = uctx.uc_mcontext.gregs[REG_RBP];
if(!unwind_context(out)) {
memset(&out, 0, sizeof(unwind_context_t));
return out;
}
return out;
}
MemoryMapEntry* get_mmap_entry(uintptr_t pc) {
// Get the memory_map entry
auto mmap_entry_it = memory_map.lower_bound(pc);
if(mmap_entry_it == memory_map.end()) {
return nullptr;
}
MemoryMapEntry& mmap_entry = mmap_entry_it->second;
if(!(mmap_entry.beg <= pc && pc <= mmap_entry.end))
return nullptr;
return &mmap_entry;
}
/** Get the `fde_func_t` function handling the given program counter — it may
* be by calling a lookup function, or by directly looking into the ELF
* symbols, depending on the state of the experiment. This is an abstraction
* function. */
_fde_func_t fde_handler_for_pc(uintptr_t pc, MemoryMapEntry& mmap_entry) {
#ifdef SGP_SWITCH_PER_FUNC
// Get the lookup function
_fde_func_t (*lookup)(uintptr_t) =
(_fde_func_t (*)(uintptr_t)) (
dlsym(mmap_entry.eh_dl_handle, "_fde_lookup"));
if(lookup == nullptr)
return nullptr;
// Get the translated pc
uintptr_t tr_pc = pc - mmap_entry.beg;
// Get the actual function
_fde_func_t rfunc = lookup(tr_pc);
if(rfunc == nullptr)
return nullptr;
return rfunc;
#elif SGP_GLOBAL_SWITCH
UNUSED(pc);
_fde_func_t global_switch =
(_fde_func_t) (dlsym(mmap_entry.eh_dl_handle, "_eh_elf"));
if(global_switch == nullptr)
return nullptr;
return global_switch;
#else
UNUSED(pc);
UNUSED(mmap_entry);
assert(false); // Please compile with either -DSCP_SWITCH_PER_FUNC or
// -DSCP_GLOBAL_SWITCH
#endif
}
bool unwind_context(unwind_context_t& ctx) {
if(ctx.rbp == 0 || ctx.rip + 1 == 0)
return false;
MemoryMapEntry* mmap_entry = get_mmap_entry(ctx.rip);
if(mmap_entry == nullptr)
return false;
_fde_func_t fde_func = fde_handler_for_pc(ctx.rip, *mmap_entry);
if(fde_func == nullptr)
return false;
uintptr_t tr_pc = ctx.rip - mmap_entry->beg;
ctx = fde_func(ctx, tr_pc);
return true;
}
void walk_stack(const std::function<void(const unwind_context_t&)>& mapped) {
unwind_context_t ctx = get_context();
do {
mapped(ctx);
} while(unwind_context(ctx));
}
uintptr_t get_register(const unwind_context_t& ctx, StackWalkerRegisters reg) {
switch(reg) {
case SW_REG_RIP:
return ctx.rip;
case SW_REG_RSP:
return ctx.rsp;
case SW_REG_RBP:
return ctx.rbp;
}
assert(0);
}