Add check for Linux minidump ending on bad write for exploitability rating.

src/processor/exploitability_linux.cc

Index: src/processor/exploitability_linux.cc
===================================================================
--- src/processor/exploitability_linux.cc	(revision 1491)
+++ src/processor/exploitability_linux.cc	(working copy)
@@ -36,6 +36,16 @@
 
 #include "processor/exploitability_linux.h"
 
+#include <regex.h>
+#include <stdio.h>
+#include <string.h>
+#include <sysexits.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+
+#include <sstream>
+#include <iterator>
+
 #include "google_breakpad/common/minidump_exception_linux.h"
 #include "google_breakpad/processor/call_stack.h"
 #include "google_breakpad/processor/process_state.h"
@@ -42,6 +52,10 @@
 #include "google_breakpad/processor/stack_frame.h"
 #include "processor/logging.h"
 
+#define MAX_INSTRUCTION_LEN 15
+// This is the buffer size for objdump's output.
+#define MAX_OBJDUMP_BUFFER_LEN 4096
+
 namespace {
 
 // This function in libc is called if the program was compiled with
@@ -53,6 +67,17 @@
 // can determine that the call would overflow the target buffer.
 const char kBoundsCheckFailureFunction[] = "__chk_fail";
 
+#ifndef _WIN32
+// PID of child process that is forked to call objdump.
+pid_t child_pid = -1;
+
+// Function to kill the child process used to call objdump in case the child
+// process hangs.
+void kill_child(int signal) {
+  kill(child_pid, SIGKILL);
+}
+#endif  // _WIN32
+
 }  // namespace
 
 namespace google_breakpad {
@@ -122,10 +147,12 @@
     return EXPLOITABILITY_ERR_PROCESSING;
   }
 
-  // Checking for the instruction pointer in a valid instruction region.
+  // Checking for the instruction pointer in a valid instruction region
+  // or if the crash resulted during an illegal write.
   if (!this->InstructionPointerInCode(instruction_ptr) ||
        this->StackPointerOffStack(stack_ptr) ||
-       this->ExecutableStackOrHeap()) {
+       this->ExecutableStackOrHeap() ||
+       this->EndedOnIllegalWrite(instruction_ptr)) {
     return EXPLOITABILITY_HIGH;
   }
 
@@ -134,6 +161,389 @@
   return EXPLOITABILITY_INTERESTING;
 }
 
+bool ExploitabilityLinux::EndedOnIllegalWrite(uint64_t instruction_ptr) {

[ivanpe, 2015/08/20 05:37:44]

I think that this functionality should be disabled by default.  You should add a
method to MinidumpProcessor to enable the disassemble functionality, if the user
of MinidumpProcessor desires.

[liuandrew, 2015/08/20 17:11:51]

Done.

+#ifdef _WIN32
+  BPLOG(INFO) << "MinGW does not support fork and exec. Terminating method.";
+#else  // _WIN32
+  // Get memory region containing instruction pointer.
+  MinidumpMemoryList *memory_list = dump_->GetMemoryList();
+  MinidumpMemoryRegion *memory_region =
+      memory_list ?
+      memory_list->GetMemoryRegionForAddress(instruction_ptr) : NULL;
+  if (!memory_region) {
+    BPLOG(ERROR) << "No memory region around instruction pointer.";

[ivanpe, 2015/08/20 05:37:44]

Please, change this to INFO.  In general we want to use ERROR only in cases
where we can file a bug as soon as we see such a message.  No memory region
around instruction seems like a valid scenario - not a bug in code.

[liuandrew, 2015/08/20 17:11:51]

Done.

+    return false;
+  }
+
+  // Get exception data to find architecture.
+  string architecture = "";
+  MinidumpException *exception = dump_->GetException();
+  // This should never evaluate to true, since this should not be reachable
+  // without checking for exception data earlier.
+  if (!exception) {
+    BPLOG(INFO) << "No exception data.";
+    return false;
+  }
+  const MDRawExceptionStream *raw_exception_stream = exception->exception();
+  const MinidumpContext *context = exception->GetContext();
+  // This should not evaluate to true, for the same reason mentioned above.
+  if (!raw_exception_stream || !context) {
+    BPLOG(INFO) << "No exception or architecture data.";
+    return false;
+  }
+  // Check architecture and set architecture variable to corresponding flag
+  // in objdump.
+  switch (context->GetContextCPU()) {
+    case MD_CONTEXT_X86:
+      architecture = "i386";
+      break;
+    case MD_CONTEXT_AMD64:
+      architecture = "i386:x86-64";
+      break;
+    default:
+      // Unsupported architecture. Note that ARM architectures are not
+      // supported because objdump does not support ARM.
+      return false;
+      break;
+  }
+
+  // Get memory region around instruction pointer and the number of bytes
+  // before and after the instruction pointer in the memory region.
+  const uint8_t *raw_memory = memory_region->GetMemory();
+  const uint64_t base = memory_region->GetBase();
+  if (base > instruction_ptr) {
+    BPLOG(ERROR) << "Memory region base value exceeds instruction pointer.";
+    return false;
+  }
+  const uint64_t offset = instruction_ptr - base;
+  if (memory_region->GetSize() < MAX_INSTRUCTION_LEN + offset) {
+    BPLOG(ERROR) << "Not enough bytes left to guarantee complete instruction.";

[ivanpe, 2015/08/20 05:37:43]

Please, replace with INFO

[liuandrew, 2015/08/20 17:11:51]

Done.

+    return false;
+  }
+
+  // Convert bytes into objdump output.
+  char objdump_output_buffer[MAX_OBJDUMP_BUFFER_LEN] = {0};
+  DisassembleBytes(architecture,
+                   raw_memory + offset,
+                   MAX_OBJDUMP_BUFFER_LEN,
+                   objdump_output_buffer);
+
+  // Set up regular expression to catch first instruction from objdump.
+  // The line with the instruction will begin with "0:".
+  regex_t regex;
+  regcomp(&regex, "0:", REG_EXTENDED | REG_NOSUB);

[ivanpe, 2015/08/20 05:37:44]

1. You are not checking the return value.

2. Isn't there a more lightweight way to find the first instruction, like
string::find ?

[liuandrew, 2015/08/20 17:11:51]

Done.

+
+  // Put buffer data into stream to output line-by-line.
+  std::stringstream objdump_stream;
+  objdump_stream.str(string(objdump_output_buffer));
+  string line;
+
+  // Pipe each output line into the string until the string contains
+  // the first instruction from objdump.
+  // Loop until the line matches the regex or there are no lines left.
+  do {
+    if (!getline(objdump_stream, line)) {
+      BPLOG(ERROR) << "Objdump instructions not found";

[ivanpe, 2015/08/20 05:37:44]

please change this to LOG

[liuandrew, 2015/08/20 17:11:51]

Done.

+      return false;

[ivanpe, 2015/08/20 05:37:44]

Don't you need to free regex?

[liuandrew, 2015/08/20 17:11:51]

Removed since I'm now using string::find.

+    }
+  } while (regexec(&regex, line.c_str(), 0, NULL, 0));
+  regfree(&regex);  // Free regex data.
+
+  // Convert objdump instruction line into the operation and operands.
+  string instruction = "";
+  string dest = "";
+  string src = "";
+  TokenizeObjdumpInstruction(line, &instruction, &dest, &src);
+

[ivanpe, 2015/08/20 05:37:44]

Please use a single empty line here.

[liuandrew, 2015/08/20 17:11:51]

Done.

+
+  // Check if the operation is a write to memory. First, the instruction
+  // must one that can write to memory. Second, the write destination
+  // must be a spot in memory rather than a register. Since there are no
+  // symbols from objdump, the destination will be enclosed by brackets.
+  if (dest.size() > 2 && dest.at(0) == '[' && dest.at(dest.size() - 1) == ']' &&
+      (!instruction.compare("mov") || !instruction.compare("inc") ||
+       !instruction.compare("dec") || !instruction.compare("and") ||
+       !instruction.compare("or") || !instruction.compare("xor") ||
+       !instruction.compare("not") || !instruction.compare("neg") ||
+       !instruction.compare("add") || !instruction.compare("sub") ||
+       !instruction.compare("shl") || !instruction.compare("shr"))) {
+    // Strip away enclosing brackets from the destination address.
+    dest = dest.substr(1, dest.size() - 2);
+    uint64_t write_address = 0;
+    CalculateAddress(dest, *context, &write_address);
+
+    // If the program crashed as a result of a write, the destination of
+    // the write must have been an address that did not permit writing.
+    // However, if the address is under 4k, due to program protections,
+    // the crash does not suggest exploitability for writes with such a
+    // low target address.
+    return write_address > 4096;
+  }
+#endif  // _WIN32
+  return false;
+}
+
+#ifndef _WIN32
+bool ExploitabilityLinux::CalculateAddress(const string &address_expression,
+                                           const DumpContext &context,
+                                           uint64_t *write_address) {
+  // The destination should be the format reg+a or reg-a, where reg
+  // is a register and a is a hexadecimal constant. Although more complex
+  // expressions can make valid instructions, objdump's disassembly outputs
+  // it in this simpler format.
+  // TODO(liuandrew): Handle more complex formats, should they arise.
+
+  if (!write_address) {
+    BPLOG(ERROR) << "Null parameter.";
+    return false;
+  }
+
+  // Clone parameter into a non-const string.
+  string expression = address_expression;
+
+  // Parse out the constant that is added to the address (if it exists).
+  size_t delim = expression.find('+');
+  bool positive_add_constant = true;
+  // Check if constant is subtracted instead of added.
+  if (delim == string::npos) {
+    positive_add_constant = false;
+    delim = expression.find('-');
+  }
+  uint32_t add_constant = 0;
+  // Save constant and remove it from the expression.
+  if (delim != string::npos) {
+    if (!sscanf(expression.substr(delim + 1).c_str(), "%x", &add_constant)) {
+      BPLOG(ERROR) << "Failed to scan constant.";
+      return false;
+    }
+    expression = expression.substr(0, delim);
+  }
+
+  // Set the the write address to the corresponding register.
+  // TODO(liuandrew): Add support for partial registers, such as
+  // the rax/eax/ax/ah/al chain.
+  switch (context.GetContextCPU()) {
+    case MD_CONTEXT_X86:
+      if (!expression.compare("eax")) {
+        *write_address = context.GetContextX86()->eax;
+      } else if (!expression.compare("ebx")) {
+        *write_address = context.GetContextX86()->ebx;
+      } else if (!expression.compare("ecx")) {
+        *write_address = context.GetContextX86()->ecx;
+      } else if (!expression.compare("edx")) {
+        *write_address = context.GetContextX86()->edx;
+      } else if (!expression.compare("edi")) {
+        *write_address = context.GetContextX86()->edi;
+      } else if (!expression.compare("esi")) {
+        *write_address = context.GetContextX86()->esi;
+      } else if (!expression.compare("ebp")) {
+        *write_address = context.GetContextX86()->ebp;
+      } else if (!expression.compare("esp")) {
+        *write_address = context.GetContextX86()->esp;
+      } else if (!expression.compare("eip")) {
+        *write_address = context.GetContextX86()->eip;
+      } else {
+        BPLOG(ERROR) << "Unsupported register";
+        return false;
+      }
+      break;
+    case MD_CONTEXT_AMD64:
+      if (!expression.compare("rax")) {
+        *write_address = context.GetContextAMD64()->rax;
+      } else if (!expression.compare("rbx")) {
+        *write_address = context.GetContextAMD64()->rbx;
+      } else if (!expression.compare("rcx")) {
+        *write_address = context.GetContextAMD64()->rcx;
+      } else if (!expression.compare("rdx")) {
+        *write_address = context.GetContextAMD64()->rdx;
+      } else if (!expression.compare("rdi")) {
+        *write_address = context.GetContextAMD64()->rdi;
+      } else if (!expression.compare("rsi")) {
+        *write_address = context.GetContextAMD64()->rsi;
+      } else if (!expression.compare("rbp")) {
+        *write_address = context.GetContextAMD64()->rbp;
+      } else if (!expression.compare("rsp")) {
+        *write_address = context.GetContextAMD64()->rsp;
+      } else if (!expression.compare("rip")) {
+        *write_address = context.GetContextAMD64()->rip;
+      } else if (!expression.compare("r8")) {
+        *write_address = context.GetContextAMD64()->r8;
+      } else if (!expression.compare("r9")) {
+        *write_address = context.GetContextAMD64()->r9;
+      } else if (!expression.compare("r10")) {
+        *write_address = context.GetContextAMD64()->r10;
+      } else if (!expression.compare("r11")) {
+        *write_address = context.GetContextAMD64()->r11;
+      } else if (!expression.compare("r12")) {
+        *write_address = context.GetContextAMD64()->r12;
+      } else if (!expression.compare("r13")) {
+        *write_address = context.GetContextAMD64()->r13;
+      } else if (!expression.compare("r14")) {
+        *write_address = context.GetContextAMD64()->r14;
+      } else if (!expression.compare("r15")) {
+        *write_address = context.GetContextAMD64()->r15;
+      } else {
+        BPLOG(ERROR) << "Unsupported register";
+        return false;
+      }
+      break;
+    default:
+      // This should not occur since the same switch condition
+      // should have terminated this method.
+      return false;
+      break;
+  }
+
+  // Add or subtract constant from write address (if applicable).
+  *write_address =
+      positive_add_constant ?
+      *write_address + add_constant : *write_address - add_constant;
+
+  return true;
+}
+
+bool ExploitabilityLinux::TokenizeObjdumpInstruction(const string &line,
+                                                     string *operation,
+                                                     string *dest,
+                                                     string *src) {
+  std::cout << "LINE: " << line << std::endl;

[ivanpe, 2015/08/20 05:37:44]

Please, do not output to console.

[liuandrew, 2015/08/20 17:11:51]

Done.

My bad, I accidentally left it here while debugging.

+  if (!operation || !dest || !src) {
+    BPLOG(ERROR) << "Null parameters passed.";
+    return false;
+  }
+
+  // Set all pointer values to empty strings.
+  *operation = "";
+  *dest = "";
+  *src = "";
+
+  // Tokenize the objdump line.
+  vector<string> tokens;
+  std::istringstream line_stream(line);
+  copy(std::istream_iterator<string>(line_stream),
+       std::istream_iterator<string>(),
+       std::back_inserter(tokens));
+
+  // Regex for the data in hex form. Each byte is two hex digits.
+  regex_t regex;
+  regcomp(&regex, "^[[:xdigit:]]{2}$", REG_EXTENDED | REG_NOSUB);
+
+  // Find and set the location of the operator. The operator appears
+  // directly after the chain of bytes that define the instruction. The
+  // operands will be the last token, given that the instruction has operands.
+  // If not, the operator is the last token. The loop skips the first token
+  // because the first token is the instruction number (namely "0:").
+  string operands = "";
+  for (size_t i = 1; i < tokens.size(); i++) {
+    // Check if current token no longer is in byte format.
+    if (regexec(&regex, tokens[i].c_str(), 0, NULL, 0)) {
+      // instruction = tokens[i];
+      *operation = tokens[i];
+      // If the operator is the last token, there are no operands.
+      if (i != tokens.size() - 1) {
+        operands = tokens[tokens.size() - 1];
+      }
+      break;
+    }
+  }
+  regfree(&regex);
+
+  if (operation->empty()) {
+    BPLOG(ERROR) << "Failed to parse out operation from objdump instruction.";
+    return false;
+  }
+
+  // Split operands into source and destination (if applicable).
+  if (!operands.empty()) {
+    size_t delim = operands.find(',');
+    if (delim == string::npos) {
+      *dest = operands;
+    } else {
+      *dest = operands.substr(0, delim);
+      *src = operands.substr(delim + 1);
+    }
+  }
+  return true;
+}
+
+bool ExploitabilityLinux::DisassembleBytes(const string &architecture,
+                                           const uint8_t *raw_bytes,
+                                           const unsigned int buffer_len,
+                                           char *objdump_output_buffer) {
+  int pipe_fd[2];
+
+  if (!raw_bytes || !objdump_output_buffer) {
+    BPLOG(ERROR) << "Bad input parameters.";
+    return false;
+  }
+
+  // Write raw bytes around instruction pointer to a temporary file to
+  // pass as an argument to objdump.
+  char raw_bytes_tmpfile[] = "/tmp/breakpad_mem_region-raw_bytes-XXXXXX";
+  int raw_bytes_fd = mkstemp(raw_bytes_tmpfile);
+  if (raw_bytes_fd < 0) {
+    BPLOG(ERROR) << "Failed to create tempfile.";
+    unlink(raw_bytes_tmpfile);
+    return false;
+  }
+  if (write(raw_bytes_fd, raw_bytes, MAX_INSTRUCTION_LEN)
+      != MAX_INSTRUCTION_LEN) {
+    BPLOG(ERROR) << "Writing of raw bytes failed.";
+    unlink(raw_bytes_tmpfile);
+    return false;
+  }
+
+  // Open pipe between STDOUT and the objdump output buffer.
+  if (pipe(pipe_fd) < 0) {
+    BPLOG(ERROR) << "Failed to pipe.";
+    return false;
+  }
+
+  // Fork process to call objdump.
+  if ((child_pid = fork()) < 0) {
+    BPLOG(ERROR) << "Forking failed.";
+    return false;
+  }
+
+  if (child_pid) {  // Parent code.
+    close(pipe_fd[1]);
+
+    // Set alarm to kill child process after one second in case it hangs.
+    signal(SIGALRM, (void (*)(int))kill_child);
+    alarm(1);
+
+    // Read piped output from objdump.
+    ssize_t bytes_read = read(pipe_fd[0],
+                              objdump_output_buffer,
+                              buffer_len);
+    int status;
+    wait(&status);  // Wait for child process to run objdump.
+    alarm(0);  // Disable alarm.
+    if (WEXITSTATUS(status)) {
+      BPLOG(ERROR) << "Child process encountered error.";
+      unlink(raw_bytes_tmpfile);
+      return false;
+    }
+    if (bytes_read < 0) {
+      BPLOG(ERROR) << "Failed to read objdump output.";
+      unlink(raw_bytes_tmpfile);
+      return false;
+    }
+    close(pipe_fd[0]);
+  } else {  // Child code.
+    close(pipe_fd[0]);
+    dup2(pipe_fd[1], STDOUT_FILENO);  // Send objdump output across pipe.
+    // Exec objdump.
+    execlp("objdump", "objdump", "-D", "-b", "binary", "-M", "intel",
+           "-m", architecture.c_str(), raw_bytes_tmpfile, NULL);
+    BPLOG(ERROR) << "Exec failed.";
+    exit(EX_OSERR);
+  }
+  unlink(raw_bytes_tmpfile);
+  return true;
+}
+#endif  // _WIN32
+
 bool ExploitabilityLinux::StackPointerOffStack(uint64_t stack_ptr) {
   MinidumpLinuxMapsList *linux_maps_list = dump_->GetLinuxMapsList();
   // Inconclusive if there are no mappings available.