dfhack/tools/supported/SegmentedFinder.h

538 lines
14 KiB
C++

#ifndef SEGMENTED_FINDER_H
#define SEGMENTED_FINDER_H
#include <malloc.h>
#include <iosfwd>
#include <iterator>
class SegmentedFinder;
class SegmentFinder
{
public:
SegmentFinder(DFHack::t_memrange & mr, DFHack::Context * DF, SegmentedFinder * SF)
{
_DF = DF;
mr_ = mr;
if(mr.valid)
{
mr_.buffer = (uint8_t *)malloc (mr_.end - mr_.start);
_SF = SF;
try
{
DF->ReadRaw(mr_.start,(mr_.end - mr_.start),mr_.buffer);
valid = true;
}
catch (DFHack::Error::MemoryAccessDenied &)
{
free(mr_.buffer);
valid = false;
mr.valid = false; // mark the range passed in as bad
cout << "Range 0x" << hex << mr_.start << " - 0x" << mr_.end << dec << " not readable." << endl;
}
}
}
~SegmentFinder()
{
if(valid)
free(mr_.buffer);
}
bool isValid()
{
return valid;
}
template <class needleType, class hayType, typename comparator >
bool Find (needleType needle, const uint8_t increment , vector <uint64_t> &newfound, comparator oper)
{
if(!valid) return !newfound.empty();
//loop
for(uint64_t offset = 0; offset < (mr_.end - mr_.start) - sizeof(hayType); offset += increment)
{
if( oper(_SF,(hayType *)(mr_.buffer + offset), needle) )
newfound.push_back(mr_.start + offset);
}
return !newfound.empty();
}
template < class needleType, class hayType, typename comparator >
uint64_t FindInRange (needleType needle, comparator oper, uint64_t start, uint64_t length)
{
if(!valid) return 0;
uint64_t stopper = min((mr_.end - mr_.start) - sizeof(hayType), (start - mr_.start) - sizeof(hayType) + length);
//loop
for(uint64_t offset = start - mr_.start; offset < stopper; offset +=1)
{
if( oper(_SF,(hayType *)(mr_.buffer + offset), needle) )
return mr_.start + offset;
}
return 0;
}
template <class needleType, class hayType, typename comparator >
bool Filter (needleType needle, vector <uint64_t> &found, vector <uint64_t> &newfound, comparator oper)
{
if(!valid) return !newfound.empty();
for( uint64_t i = 0; i < found.size(); i++)
{
if(mr_.isInRange(found[i]))
{
uint64_t corrected = found[i] - mr_.start;
if( oper(_SF,(hayType *)(mr_.buffer + corrected), needle) )
newfound.push_back(found[i]);
}
}
return !newfound.empty();
}
private:
friend class SegmentedFinder;
SegmentedFinder * _SF;
DFHack::Context * _DF;
DFHack::t_memrange mr_;
bool valid;
};
class SegmentedFinder
{
public:
SegmentedFinder(vector <DFHack::t_memrange>& ranges, DFHack::Context * DF)
{
_DF = DF;
for(int i = 0; i < ranges.size(); i++)
{
segments.push_back(new SegmentFinder(ranges[i], DF, this));
}
}
~SegmentedFinder()
{
for(int i = 0; i < segments.size(); i++)
{
delete segments[i];
}
}
SegmentFinder * getSegmentForAddress (uint64_t addr)
{
for(int i = 0; i < segments.size(); i++)
{
if(segments[i]->mr_.isInRange(addr))
{
return segments[i];
}
}
return 0;
}
template <class needleType, class hayType, typename comparator >
bool Find (const needleType needle, const uint8_t increment, vector <uint64_t> &found, comparator oper)
{
found.clear();
for(int i = 0; i < segments.size(); i++)
{
segments[i]->Find<needleType,hayType,comparator>(needle, increment, found, oper);
}
return !(found.empty());
}
template < class needleType, class hayType, typename comparator >
uint64_t FindInRange (needleType needle, comparator oper, uint64_t start, uint64_t length)
{
SegmentFinder * sf = getSegmentForAddress(start);
if(sf)
{
return sf->FindInRange<needleType,hayType,comparator>(needle, oper, start, length);
}
return 0;
}
template <class needleType, class hayType, typename comparator >
bool Filter (const needleType needle, vector <uint64_t> &found, comparator oper)
{
vector <uint64_t> newfound;
for(int i = 0; i < segments.size(); i++)
{
segments[i]->Filter<needleType,hayType,comparator>(needle, found, newfound, oper);
}
found.clear();
found = newfound;
return !(found.empty());
}
template <class needleType, class hayType, typename comparator >
bool Incremental (needleType needle, const uint8_t increment ,vector <uint64_t> &found, comparator oper)
{
if(found.empty())
{
return Find <needleType, hayType, comparator>(needle,increment,found,oper);
}
else
{
return Filter <needleType, hayType, comparator>(needle,found,oper);
}
}
template <typename T>
T * Translate(uint64_t address)
{
for(int i = 0; i < segments.size(); i++)
{
if(segments[i]->mr_.isInRange(address))
{
return (T *) (segments[i]->mr_.buffer + address - segments[i]->mr_.start);
}
}
return 0;
}
template <typename T>
T Read(uint64_t address)
{
return *Translate<T>(address);
}
template <typename T>
bool Read(uint64_t address, T& target)
{
T * test = Translate<T>(address);
if(test)
{
target = *test;
return true;
}
return false;
}
private:
DFHack::Context * _DF;
vector <SegmentFinder *> segments;
};
template <typename T>
bool equalityP (SegmentedFinder* s, T *x, T y)
{
return (*x) == y;
}
struct vecTriplet
{
uint32_t start;
uint32_t finish;
uint32_t alloc_finish;
};
template <typename Needle>
bool vectorLength (SegmentedFinder* s, vecTriplet *x, Needle &y)
{
if(x->start <= x->finish && x->finish <= x->alloc_finish)
if((x->finish - x->start) == y)
return true;
return false;
}
// find a vector of 32bit pointers, where an object pointed to has a string 'y' as the first member
bool vectorString (SegmentedFinder* s, vecTriplet *x, const char *y)
{
uint32_t object_ptr;
uint32_t idx = x->start;
// iterate over vector of pointers
for(uint32_t idx = x->start; idx < x->finish; idx += sizeof(uint32_t))
{
// deref ptr idx, get ptr to object
if(!s->Read(idx,object_ptr))
{
return false;
}
// deref ptr to first object, get ptr to string
uint32_t string_ptr;
if(!s->Read(object_ptr,string_ptr))
return false;
// get string location in our local cache
char * str = s->Translate<char>(string_ptr);
if(!str)
return false;
if(strcmp(y, str) == 0)
return true;
}
return false;
}
// find a vector of 32bit pointers, where the first object pointed to has a string 'y' as the first member
bool vectorStringFirst (SegmentedFinder* s, vecTriplet *x, const char *y)
{
uint32_t object_ptr;
uint32_t idx = x->start;
// deref ptr idx, get ptr to object
if(!s->Read(idx,object_ptr))
{
return false;
}
// deref ptr to first object, get ptr to string
uint32_t string_ptr;
if(!s->Read(object_ptr,string_ptr))
return false;
// get string location in our local cache
char * str = s->Translate<char>(string_ptr);
if(!str)
return false;
if(strcmp(y, str) == 0)
return true;
return false;
}
// test if the address is between vector.start and vector.finish
// not very useful alone, but could be a good step to filter some things
bool vectorAddrWithin (SegmentedFinder* s, vecTriplet *x, uint32_t address)
{
if(address < x->finish && address >= x->start)
return true;
return false;
}
// test if an object address is within the vector of pointers
//
bool vectorOfPtrWithin (SegmentedFinder* s, vecTriplet *x, uint32_t address)
{
uint32_t object_ptr;
uint32_t idx = x->start;
for(uint32_t idx = x->start; idx < x->finish; idx += sizeof(uint32_t))
{
if(!s->Read(idx,object_ptr))
{
return false;
}
if(object_ptr == address)
return true;
}
return false;
}
bool vectorAll (SegmentedFinder* s, vecTriplet *x, int )
{
if(x->start <= x->finish && x->finish <= x->alloc_finish)
{
if(s->getSegmentForAddress(x->start) == s->getSegmentForAddress(x->finish)
&& s->getSegmentForAddress(x->finish) == s->getSegmentForAddress(x->alloc_finish))
return true;
}
return false;
}
class Bytestreamdata
{
public:
void * object;
uint32_t length;
uint32_t allocated;
uint32_t n_used;
};
class Bytestream
{
public:
Bytestream(void * obj, uint32_t len, bool alloc = false)
{
d = new Bytestreamdata();
d->allocated = alloc;
d->object = obj;
d->length = len;
d->n_used = 1;
constant = false;
}
Bytestream()
{
d = new Bytestreamdata();
d->allocated = false;
d->object = 0;
d->length = 0;
d->n_used = 1;
constant = false;
}
Bytestream( Bytestream & bs)
{
d =bs.d;
d->n_used++;
constant = false;
}
Bytestream( const Bytestream & bs)
{
d =bs.d;
d->n_used++;
constant = true;
}
~Bytestream()
{
d->n_used --;
if(d->allocated && d->object && d->n_used == 0)
{
free (d->object);
free (d);
}
}
bool Allocate(size_t bytes)
{
if(constant)
return false;
if(d->allocated)
{
d->object = realloc(d->object, bytes);
}
else
{
d->object = malloc( bytes );
}
if(d->object)
{
d->allocated = bytes;
return true;
}
else
{
d->allocated = 0;
return false;
}
}
template < class T >
bool insert( T what )
{
if(constant)
return false;
if(d->length+sizeof(T) >= d->allocated)
Allocate((d->length+sizeof(T)) * 2);
(*(T *)( (uint64_t)d->object + d->length)) = what;
d->length += sizeof(T);
return true;
}
Bytestreamdata * d;
bool constant;
};
std::ostream& operator<< ( std::ostream& out, Bytestream& bs )
{
if(bs.d->object)
{
out << "bytestream " << dec << bs.d->length << "/" << bs.d->allocated << " bytes" << endl;
for(int i = 0; i < bs.d->length; i++)
{
out << hex << (int) ((uint8_t *) bs.d->object)[i] << " ";
}
out << endl;
}
else
{
out << "empty bytestresm" << endl;
}
return out;
}
std::istream& operator>> ( std::istream& out, Bytestream& bs )
{
string read;
while(!out.eof())
{
string tmp;
out >> tmp;
read.append(tmp);
}
cout << read << endl;
bs.d->length = 0;
size_t first = read.find_first_of("\"");
size_t last = read.find_last_of("\"");
size_t start = first + 1;
if(first == read.npos)
{
std::transform(read.begin(), read.end(), read.begin(), (int(*)(int)) tolower);
bs.Allocate(read.size()); // overkill. size / 2 should be good, but this is safe
int state = 0;
char big = 0;
char small = 0;
string::iterator it = read.begin();
// iterate through string, construct a bytestream out of 00-FF bytes
while(it != read.end())
{
char reads = *it;
if((reads >='0' && reads <= '9'))
{
if(state == 0)
{
big = reads - '0';
state = 1;
}
else if(state == 1)
{
small = reads - '0';
state = 0;
bs.insert<char>(big*16 + small);
}
}
if((reads >= 'a' && reads <= 'f'))
{
if(state == 0)
{
big = reads - 'a' + 10;
state = 1;
}
else if(state == 1)
{
small = reads - 'a' + 10;
state = 0;
bs.insert<char>(big*16 + small);
}
}
it++;
}
// we end in state= 1. should we add or should we trim... or throw errors?
// I decided on adding
if (state == 1)
{
small = 0;
bs.insert<char>(big*16 + small);
}
}
else
{
if(last == first)
{
// only one "
last = read.size();
}
size_t length = last - start;
// construct bytestream out of stuff between ""
bs.d->length = length;
if(length)
{
// todo: Bytestream should be able to handle this without external code
bs.Allocate(length);
bs.d->length = length;
const char* strstart = read.c_str();
memcpy(bs.d->object, strstart + start, length);
}
else
{
bs.d->object = 0;
}
}
cout << bs;
return out;
}
bool findBytestream (SegmentedFinder* s, void *addr, Bytestream compare )
{
if(memcmp(addr, compare.d->object, compare.d->length) == 0)
return true;
return false;
}
bool findString (SegmentedFinder* s, uint32_t *addr, const char * compare )
{
// read string pointer, translate to local scheme
char *str = s->Translate<char>(*addr);
// verify
if(!str)
return false;
if(strcmp(str, compare) == 0)
return true;
return false;
}
bool findStrBuffer (SegmentedFinder* s, uint32_t *addr, const char * compare )
{
if(strcmp((const char *)addr, compare) == 0)
return true;
return false;
}
#endif // SEGMENTED_FINDER_H