Let's look at a code example that makes use of ptrace to create a debugger program:
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <errno.h>
#include <signal.h>
#include <elf.h>
#include <sys/types.h>
#include <sys/user.h>
#include <sys/stat.h>
#include <sys/ptrace.h>
#include <sys/mman.h>
typedef struct handle {
Elf64_Ehdr *ehdr;
Elf64_Phdr *phdr;
Elf64_Shdr *shdr;
uint8_t *mem;
char *symname;
Elf64_Addr symaddr;
struct user_regs_struct pt_reg;
char *exec;
} handle_t;
Elf64_Addr lookup_symbol(handle_t *, const char *);
int main(int argc, char **argv, char **envp)
{
int fd;
handle_t h;
struct stat st;
long trap, orig;
int status, pid;
char * args[2];
if (argc < 3) {
printf("Usage: %s <program> <function>\n", argv[0]);
exit(0);
}
if ((h.exec = strdup(argv[1])) == NULL) {
perror("strdup");
exit(-1);
}
args[0] = h.exec;
args[1] = NULL;
if ((h.symname = strdup(argv[2])) == NULL) {
perror("strdup");
exit(-1);
}
if ((fd = open(argv[1], O_RDONLY)) < 0) {
perror("open");
exit(-1);
}
if (fstat(fd, &st) < 0) {
perror("fstat");
exit(-1);
}
h.mem = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
if (h.mem == MAP_FAILED) {
perror("mmap");
exit(-1);
}
h.ehdr = (Elf64_Ehdr *)h.mem;
h.phdr = (Elf64_Phdr *)(h.mem + h.ehdr->e_phoff);
h.shdr = (Elf64_Shdr *)(h.mem + h.ehdr->e_shoff);
if+ (h.mem[0] != 0x7f || strcmp((char *)&h.mem[1], "ELF")) {
printf("%s is not an ELF file\n",h.exec);
exit(-1);
}
if (h.ehdr->e_type != ET_EXEC) {
printf("%s is not an ELF executable\n", h.exec);
exit(-1);
}
if (h.ehdr->e_shstrndx == 0 || h.ehdr->e_shoff == 0 || h.ehdr->e_shnum == 0) {
printf("Section header table not found\n");
exit(-1);
}
if ((h.symaddr = lookup_symbol(&h, h.symname)) == 0) {
printf("Unable to find symbol: %s not found in executable\n", h.symname);
exit(-1);
}
close(fd);
if ((pid = fork()) < 0) {
perror("fork");
exit(-1);
}
if (pid == 0) {
if (ptrace(PTRACE_TRACEME, pid, NULL, NULL) < 0) {
perror("PTRACE_TRACEME");
exit(-1);
}
execve(h.exec, args, envp);
exit(0);
}
wait(&status);
printf("Beginning analysis of pid: %d at %lx\n", pid, h.symaddr);
if ((orig = ptrace(PTRACE_PEEKTEXT, pid, h.symaddr, NULL)) < 0) {
perror("PTRACE_PEEKTEXT");
exit(-1);
}
trap = (orig & ~0xff) | 0xcc;
if (ptrace(PTRACE_POKETEXT, pid, h.symaddr, trap) < 0) {
perror("PTRACE_POKETEXT");
exit(-1);
}
trace:
if (ptrace(PTRACE_CONT, pid, NULL, NULL) < 0) {
perror("PTRACE_CONT");
exit(-1);
}
wait(&status);
if (WIFSTOPPED(status) && WSTOPSIG(status) == SIGTRAP) {
if (ptrace(PTRACE_GETREGS, pid, NULL, &h.pt_reg) < 0) {
perror("PTRACE_GETREGS");
exit(-1);
}
printf("\nExecutable %s (pid: %d) has hit breakpoint 0x%lx\n",
h.exec, pid, h.symaddr);
printf("%%rcx: %llx\n%%rdx: %llx\n%%rbx: %llx\n"
"%%rax: %llx\n%%rdi: %llx\n%%rsi: %llx\n"
"%%r8: %llx\n%%r9: %llx\n%%r10: %llx\n"
"%%r11: %llx\n%%r12 %llx\n%%r13 %llx\n"
"%%r14: %llx\n%%r15: %llx\n%%rsp: %llx",
h.pt_reg.rcx, h.pt_reg.rdx, h.pt_reg.rbx,
h.pt_reg.rax, h.pt_reg.rdi, h.pt_reg.rsi,
h.pt_reg.r8, h.pt_reg.r9, h.pt_reg.r10,
h.pt_reg.r11, h.pt_reg.r12, h.pt_reg.r13,
h.pt_reg.r14, h.pt_reg.r15, h.pt_reg.rsp);
printf("\nPlease hit any key to continue: ");
getchar();
if (ptrace(PTRACE_POKETEXT, pid, h.symaddr, orig) < 0) {
perror("PTRACE_POKETEXT");
exit(-1);
}
h.pt_reg.rip = h.pt_reg.rip - 1;
if (ptrace(PTRACE_SETREGS, pid, NULL, &h.pt_reg) < 0) {
perror("PTRACE_SETREGS");
exit(-1);
}
if (ptrace(PTRACE_SINGLESTEP, pid, NULL, NULL) < 0) {
perror("PTRACE_SINGLESTEP");
exit(-1);
}
wait(NULL);
if (ptrace(PTRACE_POKETEXT, pid, h.symaddr, trap) < 0) {
perror("PTRACE_POKETEXT");
exit(-1);
}
goto trace;
}
if (WIFEXITED(status))
printf("Completed tracing pid: %d\n", pid);
exit(0);
}
Elf64_Addr lookup_symbol(handle_t *h, const char *symname)
{
int i, j;
char *strtab;
Elf64_Sym *symtab;
for (i = 0; i < h->ehdr->e_shnum; i++) {
if (h->shdr[i].sh_type == SHT_SYMTAB) {
strtab = (char *)&h->mem[h->shdr[h->shdr[i].sh_link].sh_offset];
symtab = (Elf64_Sym *)&h->mem[h->shdr[i].sh_offset];
for (j = 0; j < h->shdr[i].sh_size/sizeof(Elf64_Sym); j++) {
if(strcmp(&strtab[symtab->st_name], symname) == 0)
return (symtab->st_value);
symtab++;
}
}
}
return 0;
}
}To compile the preceding source code, use this:
gcc tracer.c –o tracer
Keep in mind that tracer.c locates the symbol table by finding and referencing the SHT_SYMTAB type section header, so it will not work on executables that have been stripped of the SHT_SYMTAB symbol table (although they may have SHT_DYNSYM). This actually makes sense, because usually we are debugging programs that are still in their development phase, so they usually do have a complete symbol table.
The other limitation is that it doesn't allow you to pass arguments to the program you are executing and tracing. So, it wouldn't do well in a real debugging situation, where you may need to pass switches or command-line options to your program that is being debugged.
As an example of the ./tracer program that we designed, let's try it on a very simple program that calls a function called print_string(char *) twice, and passes to it the Hello 1 string on the first round and Hello 2 on the second.
Here's an example of using the ./tracer code:
$ ./tracer ./test print_string Beginning analysis of pid: 6297 at 40057d Executable ./test (pid: 6297) has hit breakpoint 0x40057d %rcx: 0 %rdx: 7fff4accbf18 %rbx: 0 %rax: 400597 %rdi: 400644 %rsi: 7fff4accbf08 %r8: 7fd4f09efe80 %r9: 7fd4f0a05560 %r10: 7fff4accbcb0 %r11: 7fd4f0650dd0 %r12 400490 %r13 7fff4accbf00 %r14: 0 %r15: 0 %rsp: 7fff4accbe18 Please hit any key to continue: c Hello 1 Executable ./test (pid: 6297) has hit breakpoint 0x40057d %rcx: ffffffffffffffff %rdx: 7fd4f09f09e0 %rbx: 0 %rax: 9 %rdi: 40064d %rsi: 7fd4f0c14000 %r8: ffffffff %r9: 0 %r10: 22 %r11: 246 %r12 400490 %r13 7fff4accbf00 %r14: 0 %r15: 0 %rsp: 7fff4accbe18 Hello 2 Please hit any key to continue: Completed tracing pid: 6297
As you can see, a breakpoint was set on print_string, and each time the function was called, our ./tracer program caught the trap, printed the register values, and then continued executing after we hit a character. The ./tracer program is a good example of how a debugger such as gdb works. Although it is much simpler, it demonstrates process tracing, breakpoints, and symbol lookup.
This program works great if you want to execute a program and trace it all at once. But what about tracing a process that is already running? In such a case, we would want to attach to the process image with PTRACE_ATTACH. This request sends a SIGSTOP to the process we are attaching to, so we use wait or waitpid to wait for the process to stop.