dfhack/library/DFProcessManager.cpp

527 lines
17 KiB
C++

/*
www.sourceforge.net/projects/dfhack
Copyright (c) 2009 Petr Mrázek (peterix), Kenneth Ferland (Impaler[WrG]), dorf
This software is provided 'as-is', without any express or implied
warranty. In no event will the authors be held liable for any
damages arising from the use of this software.
Permission is granted to anyone to use this software for any
purpose, including commercial applications, and to alter it and
redistribute it freely, subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must
not claim that you wrote the original software. If you use this
software in a product, an acknowledgment in the product documentation
would be appreciated but is not required.
2. Altered source versions must be plainly marked as such, and
must not be misrepresented as being the original software.
3. This notice may not be removed or altered from any source
distribution.
*/
#include "DFCommon.h"
#include "DFDataModel.h"
#include "DFMemInfo.h"
#include "tinyxml/tinyxml.h"
#include <iostream>
/// HACK: global variables (only one process can be attached at the same time.)
Process * g_pProcess; ///< current process. non-NULL when picked
ProcessHandle g_ProcessHandle; ///< cache of handle to current process. used for speed reasons
FILE * g_ProcessMemFile; ///< opened /proc/PID/mem, valid when attached
#ifdef LINUX_BUILD
/*
* LINUX version of the process finder.
*/
#include "md5/md5wrapper.h"
Process* ProcessManager::addProcess(const string & exe,ProcessHandle PH, const string & memFile)
{
md5wrapper md5;
// get hash of the running DF process
string hash = md5.getHashFromFile(exe);
vector<memory_info>::iterator it;
// iterate over the list of memory locations
for ( it=meminfo.begin() ; it < meminfo.end(); it++ )
{
if(hash == (*it).getString("md5")) // are the md5 hashes the same?
{
memory_info * m = &*it;
Process * ret;
//cout <<"Found process " << PH << ". It's DF version " << m->getVersion() << "." << endl;
// df can run under wine on Linux
if(memory_info::OS_WINDOWS == (*it).getOS())
{
ret= new Process(new DMWindows40d(),m,PH, PH);
}
else if (memory_info::OS_LINUX == (*it).getOS())
{
ret= new Process(new DMLinux40d(),m,PH, PH);
}
else
{
// some error happened, continue with next process
continue;
}
// tell Process about the /proc/PID/mem file
ret->setMemFile(memFile);
processes.push_back(ret);
return ret;
}
}
return NULL;
}
bool ProcessManager::findProcessess()
{
DIR *dir_p;
struct dirent *dir_entry_p;
string dir_name;
string exe_link;
string cwd_link;
string cmdline_path;
string cmdline;
// ALERT: buffer overrun potential
char target_name[1024];
int target_result;
int errorcount;
int result;
errorcount=0;
result=0;
// Open /proc/ directory
dir_p = opendir("/proc/");
// Reading /proc/ entries
while(NULL != (dir_entry_p = readdir(dir_p)))
{
// Only PID folders (numbers)
if (strspn(dir_entry_p->d_name, "0123456789") != strlen(dir_entry_p->d_name))
{
continue;
}
// string manipulation - get /proc/PID/exe link and /proc/PID/mem names
dir_name = "/proc/";
dir_name += dir_entry_p->d_name;
dir_name += "/";
exe_link = dir_name + "exe";
string mem_name = dir_name + "mem";
// resolve /proc/PID/exe link
target_result = readlink(exe_link.c_str(), target_name, sizeof(target_name)-1);
if (target_result == -1)
{
// bad result from link resolution, continue with another processed
continue;
}
// make sure we have a null terminated string...
target_name[target_result] = 0;
// is this the regular linux DF?
if (strstr(target_name, "dwarfort.exe") != NULL)
{
exe_link = target_name;
// get PID
result = atoi(dir_entry_p->d_name);
// create linux process, add it to the vector
addProcess(exe_link,result,mem_name);
// continue with next process
continue;
}
// FIXME: this fails when the wine process isn't started from the 'current working directory'. strip path data from cmdline
// DF in wine?
if(strstr(target_name, "wine-preloader")!= NULL)
{
// get working directory
cwd_link = dir_name + "cwd";
target_result = readlink(cwd_link.c_str(), target_name, sizeof(target_name)-1);
target_name[target_result] = 0;
// got path to executable, do the same for its name
cmdline_path = dir_name + "cmdline";
ifstream ifs ( cmdline_path.c_str() , ifstream::in );
getline(ifs,cmdline);
if (cmdline.find("dwarfort.exe") != string::npos || cmdline.find("Dwarf Fortress.exe") != string::npos)
{
// put executable name and path together
exe_link = target_name;
exe_link += "/";
exe_link += cmdline;
// get PID
result = atoi(dir_entry_p->d_name);
// create wine process, add it to the vector
addProcess(exe_link,result,mem_name);
}
}
}
closedir(dir_p);
// return value depends on if we found some DF processes
if(processes.size())
{
return true;
}
return false;
}
#else
// some magic - will come in handy when we start doing debugger stuff on Windows
bool EnableDebugPriv()
{
bool bRET = FALSE;
TOKEN_PRIVILEGES tp;
HANDLE hToken;
if (LookupPrivilegeValue(NULL, SE_DEBUG_NAME, &tp.Privileges[0].Luid))
{
if (OpenProcessToken(GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES, &hToken))
{
if (hToken != INVALID_HANDLE_VALUE)
{
tp.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED;
tp.PrivilegeCount = 1;
if (AdjustTokenPrivileges(hToken, FALSE, &tp, 0, 0, 0))
{
bRET = TRUE;
}
CloseHandle(hToken);
}
}
}
return bRET;
}
// WINDOWS version of the process finder
bool ProcessManager::findProcessess()
{
// Get the list of process identifiers.
//TODO: make this dynamic. (call first to get the array size and second to really get process handles)
DWORD ProcArray[512], memoryNeeded, numProccesses;
HMODULE hmod = NULL;
DWORD junk;
HANDLE hProcess;
bool found = false;
IMAGE_NT_HEADERS32 pe_header;
IMAGE_SECTION_HEADER sections[16];
EnableDebugPriv();
if ( !EnumProcesses( ProcArray, sizeof(ProcArray), &memoryNeeded ) )
{
return false;
}
// Calculate how many process identifiers were returned.
numProccesses = memoryNeeded / sizeof(DWORD);
// iterate through processes
for ( int i = 0; i < numProccesses; i++ )
{
found = false;
// open process
hProcess = OpenProcess( PROCESS_ALL_ACCESS, FALSE, ProcArray[i] );
if (NULL == hProcess)
continue;
// we've got some process, look at its first module
if(EnumProcessModules(hProcess, &hmod, 1 * sizeof(HMODULE), &junk))
{
// TODO: check module filename to verify that it's DF!
// got base ;)
uint32_t base = (uint32_t)hmod;
// read from this process
g_ProcessHandle = hProcess;
uint32_t pe_offset = MreadDWord(base+0x3C);
Mread(base + pe_offset , sizeof(pe_header), (uint8_t *)&pe_header);
Mread(base + pe_offset+ sizeof(pe_header), sizeof(sections) , (uint8_t *)&sections );
// see if there's a version entry that matches this process
vector<memory_info>::iterator it;
for ( it=meminfo.begin() ; it < meminfo.end(); it++ )
{
// filter by OS
if(memory_info::OS_WINDOWS == (*it).getOS())
{
uint32_t pe_timestamp = (*it).getHexValue("pe_timestamp");
if (pe_timestamp == pe_header.FileHeader.TimeDateStamp)
{
printf("Match found! Using version %s.\n", (*it).getVersion().c_str());
// give the process a data model and memory layout fixed for the base of first module
memory_info *m = new memory_info(*it);
m->RebaseAll(base);
// keep track of created memory_info objects so we can destroy them later
destroy_meminfo.push_back(m);
// process is responsible for destroying its data model
Process *ret= new Process(new DMWindows40d(),m,hProcess, ProcArray[i]);
processes.push_back(ret);
found = true;
break; // break the iterator loop
}
}
}
// close handle of processes that aren't DF
if(!found)
{
CloseHandle(hProcess);
}
}
}
if(processes.size())
return true;
return false;
}
#endif
void ProcessManager::ParseVTable(TiXmlElement* vtable, memory_info& mem)
{
TiXmlElement* pClassEntry;
TiXmlElement* pClassSubEntry;
const char * rebase = vtable->Attribute("rebase");
if(rebase)
{
int32_t rebase_offset = strtol(rebase, NULL, 16);
mem.RebaseVTable(rebase_offset);
}
pClassEntry = vtable->FirstChildElement();
for(;pClassEntry;pClassEntry=pClassEntry->NextSiblingElement())
{
string type = pClassEntry->Value();
const char *cstr_name = pClassEntry->Attribute("name");
const char *cstr_vtable = pClassEntry->Attribute("vtable");
if(type== "class")
{
mem.setClass(cstr_name, cstr_vtable);
}
else if (type == "multiclass")
{
const char *cstr_typeoffset = pClassEntry->Attribute("typeoffset");
int mclass = mem.setMultiClass(cstr_name, cstr_vtable, cstr_typeoffset);
pClassSubEntry = pClassEntry->FirstChildElement();
for(;pClassSubEntry;pClassSubEntry=pClassSubEntry->NextSiblingElement())
{
type = pClassSubEntry->Value();
if(type== "class")
{
cstr_name = pClassSubEntry->Attribute("name");
const char *cstr_value = pClassSubEntry->Attribute("type");
mem.setMultiClassChild(mclass,cstr_name,cstr_value);
}
}
}
}
}
void ProcessManager::ParseEntry (TiXmlElement* entry, memory_info& mem, map <string ,TiXmlElement *>& knownEntries)
{
TiXmlElement* pMemEntry;
const char *cstr_version = entry->Attribute("version");
const char *cstr_os = entry->Attribute("os");
const char *cstr_base = entry->Attribute("base");
const char *cstr_rebase = entry->Attribute("rebase");
if(cstr_base)
{
string base = cstr_base;
ParseEntry(knownEntries[base], mem, knownEntries);
}
// mandatory attributes missing?
if(!(cstr_version && cstr_os))
{
cerr << "Bad entry in memory.xml detected, version or os attribute is missing.";
// skip if we don't have valid attributes
return;
}
string os = cstr_os;
mem.setVersion(cstr_version);
mem.setOS(cstr_os);
// offset inherited addresses by 'rebase'.
int32_t rebase = 0;
if(cstr_rebase)
{
rebase = mem.getBase() + strtol(cstr_rebase, NULL, 16);
mem.RebaseAddresses(rebase);
}
//set base to default, we're overwriting this because the previous rebase could cause havoc on Vista/7
if(os == "windows")
{
// set default image base. this is fixed for base relocation later
mem.setBase(0x400000);
}
else if(os == "linux")
{
// this is wrong... I'm not going to do base image relocation on linux though.
// users are free to use a sane kernel that doesn't do this kind of **** by default
mem.setBase(0x0);
}
else
{
cerr << "unknown operating system " << os << endl;
return;
}
// process additional entries
//cout << "Entry " << cstr_version << " " << cstr_os << endl;
pMemEntry = entry->FirstChildElement()->ToElement();
for(;pMemEntry;pMemEntry=pMemEntry->NextSiblingElement())
{
// only elements get processed
const char *cstr_type = pMemEntry->Value();
const char *cstr_name = pMemEntry->Attribute("name");
const char *cstr_value = pMemEntry->GetText();
// check for missing parts
string type, name, value;
type = cstr_type;
if(type == "VTable")
{
ParseVTable(pMemEntry, mem);
continue;
}
if( !(cstr_name && cstr_value))
{
cerr << "underspecified MemInfo entry" << endl;
continue;
}
name = cstr_name;
value = cstr_value;
if (type == "HexValue")
{
mem.setHexValue(name, value);
}
else if (type == "Address")
{
mem.setAddress(name, value);
}
else if (type == "Offset")
{
mem.setOffset(name, value);
}
else if (type == "String")
{
mem.setString(name, value);
}
else
{
cerr << "Unknown MemInfo type: " << type << endl;
}
} // for
} // method
// load the XML file with offsets
bool ProcessManager::loadDescriptors(string path_to_xml)
{
TiXmlDocument doc( path_to_xml.c_str() );
bool loadOkay = doc.LoadFile();
TiXmlHandle hDoc(&doc);
TiXmlElement* pElem;
TiXmlHandle hRoot(0);
memory_info mem;
if ( loadOkay )
{
// block: name
{
pElem=hDoc.FirstChildElement().Element();
// should always have a valid root but handle gracefully if it does
if (!pElem)
{
cerr << "no pElem found" << endl;
return false;
}
string m_name=pElem->Value();
if(m_name != "DFExtractor")
{
cerr << "DFExtractor != " << m_name << endl;
return false;
}
//cout << "got DFExtractor XML!" << endl;
// save this for later
hRoot=TiXmlHandle(pElem);
}
// transform elements
{
// trash existing list
meminfo.clear();
TiXmlElement* pMemInfo=hRoot.FirstChild( "MemoryDescriptors" ).FirstChild( "Entry" ).Element();
map <string ,TiXmlElement *> map_pNamedEntries;
vector <TiXmlElement *> v_pEntries;
for( ; pMemInfo; pMemInfo=pMemInfo->NextSiblingElement("Entry"))
{
v_pEntries.push_back(pMemInfo);
const char *id;
if(id= pMemInfo->Attribute("id"))
{
string str_id = id;
map_pNamedEntries[str_id] = pMemInfo;
}
}
for(uint32_t i = 0; i< v_pEntries.size();i++)
{
memory_info mem;
//FIXME: add a set of entries processed in a step of this cycle, use it to check for infinite loops
/* recursive */ParseEntry( v_pEntries[i] , mem , map_pNamedEntries);
meminfo.push_back(mem);
}
// process found things here
}
return true;
}
else
{
// load failed
cerr << "Can't load memory offsets from memory.xml" << endl;
return false;
}
}
uint32_t ProcessManager::size()
{
return processes.size();
};
Process * ProcessManager::operator[](uint32_t index)
{
assert(index < processes.size());
return processes[index];
};
ProcessManager::ProcessManager( string path_to_xml )
{
currentProcess = NULL;
currentProcessHandle = 0;
loadDescriptors( path_to_xml );
}
ProcessManager::~ProcessManager()
{
// delete all processes
for(uint32_t i = 0;i < processes.size();i++)
{
delete processes[i];
}
//delete all generated memory_info stuff
for(uint32_t i = 0;i < destroy_meminfo.size();i++)
{
delete destroy_meminfo[i];
}
}