Monday, May 19, 2014

In this post, we conclude how we can map raw function pointers to symbols offline. This means we no longer depend on debugging sessions to resolve a stack. When we encounter a stack trace for our executable that is not resolved. We simply pass the raw function pointers to our tool along with the load address of our executable and each of these function pointers are then looked up based on their RVA. The RVA is the relative address of the function pointer from the load address.
RVA =  eip  - load address
if the stack trace was already available in the format executable name + offset, the offset translates to RVA
the raw function pointers are easier to see in a crash log and hence and hence the tool takes the eip directly. Note that the eip points to the return address in a debugger, here we mention it as the function pointer
 When we reviewed the win32 process startup stages, we saw how the process was initialized and what constitutes its address space. This helped us calculate the offsets.
The PDB is laid out in contiguous memory regions identified by section numbers. Each section has an address range and the addresses are found as offset from a given section. This is different from the Relative Virtual Address which indicates a position from the load address. The function pointers in an unresolved stack is a pointer that is specifically an offset equal to RVA from the load address. We use this to find the symbols.  All symbols have a type information associated. We filter the result to only the type Function because that's what we are after.
Note that the symbols can be found by different methods such as the two lookups we have mentioned above. At the same time we can also iterate over all the symbols to dump the information based on tables. Since the PDB keeps track of all information regarding symbols in tables, we can exhaustively scan the tables for information on all symbols.
Dumping all the information on the symbols helps to get the information in text format which can then be searched for RVA, offset, section number or name of a function.
The latter approach is quite useful when we have heavy text parsing utilities.



No comments:

Post a Comment