Improving `kmod` (#3254)

* -

* improving kmod

* bug fix

* added helpers for finding offsets

* improved helpers for determining offsets

* improving kmod helpers

* refactoring + handling kallsyms

* recovering from stash

* doc + test

* fixes based on coments

* improvements

docs/commands/kernel/kmod.md

@@ -2,7 +2,7 @@
 # kmod
 
 ```text
-usage: kmod [-h] [module_name]
+usage: kmod [-h] [-l PATH] [module_name]
 
 ```
 
@@ -18,6 +18,7 @@ Displays the loaded Linux kernel modules.
 |Short|Long|Help|
 | :--- | :--- | :--- |
 |-h|--help|show this help message and exit|
+|-l|--load|the path of the module to load|
 
 <!-- END OF AUTOGENERATED PART. Do not modify this line or the line below, they mark the end of the auto-generated part of the file. If you want to extend the documentation in a way which cannot easily be done by adding to the command help description, write below the following line. -->
 <!-- ------------\>8---- ----\>8---- ----\>8------------ -->

pwndbg/aglib/kernel/__init__.py

@@ -648,18 +648,24 @@ def num_numa_nodes() -> int:
 
 
 def node_data() -> pwndbg.dbg_mod.Value:
-    if arch_symbols() is not None:
-        return arch_symbols().node_data()
+    if (syms := arch_symbols()) is not None:
+        return syms.node_data()
     return None
 
 
 def slab_caches() -> pwndbg.dbg_mod.Value:
-    if arch_symbols() is not None:
-        return arch_symbols().slab_caches()
+    if (syms := arch_symbols()) is not None:
+        return syms.slab_caches()
     return None
 
 
 def per_cpu_offset() -> pwndbg.dbg_mod.Value:
-    if arch_symbols() is not None:
-        return arch_symbols().per_cpu_offset()
+    if (syms := arch_symbols()) is not None:
+        return syms.per_cpu_offset()
+    return None
+
+
+def modules() -> pwndbg.dbg_mod.Value:
+    if (syms := arch_symbols()) is not None:
+        return syms.modules()
     return None

pwndbg/aglib/kernel/kallsyms.py

@@ -12,6 +12,7 @@ from pwnlib.util.packing import u64
 
 import pwndbg.aglib
 import pwndbg.aglib.kernel
+import pwndbg.aglib.kernel.kmod
 import pwndbg.aglib.memory
 import pwndbg.color.message as M
 import pwndbg.commands
@@ -80,6 +81,8 @@ class Kallsyms:
         self.names = self.find_names()
         self.kernel_addresses = self.get_kernel_addresses()
         self.parse_symbol_table()
+        for sym_name, sym_addr, sym_type in pwndbg.aglib.kernel.kmod.all_modules_kallsyms():
+            self.kallsyms[sym_name] = (sym_addr, sym_type)
         print(M.info(f"Found {len(self.kallsyms)} ksymbols"))
 
     def find_token_table(self) -> int:

pwndbg/aglib/kernel/kmod.py

@@ -0,0 +1,221 @@
+from __future__ import annotations
+
+from enum import Enum
+from typing import List
+from typing import Tuple
+
+import pwndbg
+import pwndbg.color.message as M
+import pwndbg.commands
+from pwndbg.aglib.kernel.macros import for_each_entry
+
+
+class mod_mem_type(Enum):
+    # Calculate runtime memory footprint by summing sizes of MOD_TEXT, MOD_DATA, MOD_RODATA, MOD_RO_AFTER_INIT,
+    # which excludes initialization sections that are freed after the module load. See `enum mod_mem_type` in kernel source.
+    MOD_TEXT = 0
+    MOD_DATA = 1
+    MOD_RODATA = 2
+    MOD_RO_AFTER_INIT = 3  # might be empty
+    # MOD_INIT_TEXT,
+    # MOD_INIT_DATA,
+    # MOD_INIT_RODATA,
+    MOD_MEM_NUM_TYPES = 4
+
+
+# TODO: handle potential negative offsets when CONFIG_RANDSTRUCT=y
+@pwndbg.lib.cache.cache_until("stop")
+def module_name_offset():
+    modules = pwndbg.aglib.kernel.modules()
+    if modules is None:
+        print(M.warn("Could not find modules"))
+        return None
+    module = pwndbg.aglib.memory.read_pointer_width(int(modules))
+    for i in range(0x100):
+        offset = i * pwndbg.aglib.arch.ptrsize
+        try:
+            bs = pwndbg.aglib.memory.string(module + offset).decode("ascii")
+            if len(bs) < 2:
+                continue
+            return offset
+        except Exception:
+            pass
+    print(M.warn("Could not find module->name"))
+    return None
+
+
+@pwndbg.lib.cache.cache_until("stop")
+def module_mem_offset() -> Tuple[int | None, int | None, int | None]:
+    modules = pwndbg.aglib.kernel.modules()
+    if modules is None:
+        print(M.warn("Could not find modules"))
+        return None, None, None
+    module = pwndbg.aglib.memory.read_pointer_width(int(modules))
+    for i in range(0x100):
+        offset = i * pwndbg.aglib.arch.ptrsize
+        min_size = 0x10
+        if pwndbg.aglib.kernel.krelease() >= (6, 13):
+            min_size += 0x8
+        for module_memory_size in (
+            min_size,
+            min_size + 0x38,
+        ):
+            found = True
+            for mem_type in range(mod_mem_type.MOD_RO_AFTER_INIT.value):
+                mem_ptr = module + offset + mem_type * module_memory_size
+                if pwndbg.aglib.memory.peek(mem_ptr) is None:
+                    found = False
+                    break
+                base = pwndbg.aglib.memory.read_pointer_width(mem_ptr)
+                if base == 0 or ((base & 0xFFF) != 0):
+                    found = False
+                    break
+                size_offset = pwndbg.aglib.arch.ptrsize
+                if pwndbg.aglib.kernel.krelease() >= (6, 13):
+                    # https://elixir.bootlin.com/linux/v6.13/source/include/linux/module.h#L368
+                    # additional fields were added
+                    size_offset += pwndbg.aglib.arch.ptrsize + 4
+                size = pwndbg.aglib.memory.u32(mem_ptr + size_offset)
+                if not 0 < size < 0x100000:
+                    found = False
+                    break
+            if found:
+                return offset, module_memory_size, size_offset
+    print(M.warn("Could not find module->mem"))
+    return None, None, None
+
+
+@pwndbg.lib.cache.cache_until("stop")
+def module_layout_offset() -> Tuple[int | None, int | None]:
+    modules = pwndbg.aglib.kernel.modules()
+    if modules is None:
+        print(M.warn("Could not find modules"))
+        return None, None
+    module = pwndbg.aglib.memory.read_pointer_width(int(modules))
+    for i in range(0x100):  # enough to search through the struct
+        offset = i * pwndbg.aglib.arch.ptrsize
+        ptr = module + offset + pwndbg.aglib.arch.ptrsize
+        if pwndbg.aglib.memory.peek(ptr) is None:
+            continue
+        base = pwndbg.aglib.memory.read_pointer_width(ptr)
+        if base == 0 or ((base & 0xFFF) != 0):
+            continue
+        valid = True
+        for i in range(4):
+            size = pwndbg.aglib.memory.u32(ptr + 4 * i)
+            if not 0 < size < 0x100000:
+                valid = False
+                break
+        if valid:
+            return offset, offset + pwndbg.aglib.arch.ptrsize
+    print(M.warn("Could not find module->init_layout"))
+    return None, None
+
+
+@pwndbg.lib.cache.cache_until("stop")
+def module_kallsyms_offset():
+    modules = pwndbg.aglib.kernel.modules()
+    if modules is None:
+        print(M.warn("Could not find modules"))
+        return None, None
+    module = pwndbg.aglib.memory.read_pointer_width(int(modules))
+    for i in range(0x100):
+        offset = i * pwndbg.aglib.arch.ptrsize
+        ptr = module + offset
+        if pwndbg.aglib.memory.peek(ptr) is None:
+            continue
+        kallsyms = pwndbg.aglib.memory.read_pointer_width(ptr)
+        if pwndbg.aglib.memory.peek(kallsyms) is None or kallsyms == 0:
+            continue
+        symtab = pwndbg.aglib.memory.read_pointer_width(kallsyms)
+        if pwndbg.aglib.memory.peek(symtab) is None:
+            continue
+        num_symtab = pwndbg.aglib.memory.read_pointer_width(kallsyms + pwndbg.aglib.arch.ptrsize)
+        if pwndbg.aglib.memory.peek(num_symtab) is not None or num_symtab == 0:
+            continue
+        strtab = pwndbg.aglib.memory.read_pointer_width(kallsyms + pwndbg.aglib.arch.ptrsize * 2)
+        if pwndbg.aglib.memory.peek(strtab) is None:
+            continue
+        if pwndbg.aglib.kernel.krelease() >= (5, 2):
+            typetab = pwndbg.aglib.memory.read_pointer_width(
+                kallsyms + pwndbg.aglib.arch.ptrsize * 3
+            )
+            if pwndbg.aglib.memory.peek(typetab) is None:
+                continue
+        return offset
+    print(M.warn("Could not find module->kallsyms"))
+    return None
+
+
+@pwndbg.lib.cache.cache_until("stop")
+def module_list_with_typeinfo() -> Tuple[pwndbg.dbg_mod.Value, ...]:
+    modules = pwndbg.aglib.kernel.modules()
+    if modules is None:
+        print(M.warn("Could not find modules"))
+        return ()
+    result = []
+    head = pwndbg.aglib.memory.get_typed_pointer_value("struct list_head", modules)
+    for module in for_each_entry(head, "struct module", "list"):
+        result.append(module)
+    # each entry if pointing to hte start of the module
+    return tuple(result)
+
+
+@pwndbg.lib.cache.cache_until("stop")
+def module_list() -> Tuple[int, ...]:
+    modules = pwndbg.aglib.kernel.modules()
+    if modules is None:
+        print(M.warn("Could not find modules"))
+        return ()
+    modules = int(modules)
+    result = []
+    cur = pwndbg.aglib.memory.read_pointer_width(modules)
+    while cur != modules:
+        result.append(cur)
+        cur = pwndbg.aglib.memory.read_pointer_width(cur)
+    # each entry is pointing to the module->next
+    return tuple(result)
+
+
+def parse_module_kallsyms(kallsyms: int) -> List[Tuple[str, int, str]]:
+    is_64bit = pwndbg.aglib.arch.ptrsize == 8
+    sizeof_symtab_entry = 24 if is_64bit else 16
+    result = []
+    symtab = pwndbg.aglib.memory.read_pointer_width(kallsyms)
+    num_symtab = pwndbg.aglib.memory.read_pointer_width(kallsyms + pwndbg.aglib.arch.ptrsize)
+    strtab = pwndbg.aglib.memory.read_pointer_width(kallsyms + pwndbg.aglib.arch.ptrsize * 2)
+    typetab = None
+    if pwndbg.aglib.kernel.krelease() >= (5, 2):
+        typetab = pwndbg.aglib.memory.read_pointer_width(kallsyms + pwndbg.aglib.arch.ptrsize * 3)
+    strtab_offset = 0
+    for i in range(num_symtab):
+        sym_name = pwndbg.aglib.memory.string(strtab + strtab_offset).decode("utf-8")
+        strtab_offset += len(sym_name) + 1
+        if len(sym_name) == 0:
+            continue
+        sym_addr = pwndbg.aglib.memory.read_pointer_width(
+            int(symtab) + sizeof_symtab_entry * i + pwndbg.aglib.arch.ptrsize
+        )
+        sym_type = None
+        if pwndbg.aglib.kernel.krelease() >= (5, 2):
+            sym_type = chr(pwndbg.aglib.memory.u8(typetab + i))
+        else:
+            sym_type = chr(
+                pwndbg.aglib.memory.u8(symtab + sizeof_symtab_entry * i + 16 if is_64bit else 8)
+            )
+        result.append((sym_name, sym_addr, sym_type))
+    return result
+
+
+def all_modules_kallsyms() -> List[Tuple[str, int, str]]:
+    result = []
+    if pwndbg.aglib.typeinfo.load("struct module") is not None:
+        for module in module_list_with_typeinfo():
+            if module.type.has_field("kallsyms"):
+                kallsyms = int(module["kallsyms"])
+                result += parse_module_kallsyms(kallsyms)
+    elif module_kallsyms_offset() is not None:
+        for module in module_list():
+            kallsyms = pwndbg.aglib.memory.read_pointer_width(module + module_kallsyms_offset())
+            result += parse_module_kallsyms(kallsyms)
+    return result

pwndbg/aglib/kernel/symbol.py

@@ -296,7 +296,19 @@ class ArchSymbols:
         per_cpu_offset = pwndbg.aglib.symbol.lookup_symbol("__per_cpu_offset")
         if per_cpu_offset is not None:
             return per_cpu_offset
-        return self._per_cpu_offset()
+        per_cpu_offset = self._per_cpu_offset()
+        if per_cpu_offset is None:
+            return None
+        return pwndbg.aglib.memory.get_typed_pointer("unsigned long", per_cpu_offset)
+
+    def modules(self):
+        modules = pwndbg.aglib.symbol.lookup_symbol("modules")
+        if modules:
+            return modules
+        modules = self._modules()
+        if modules is None:
+            return None
+        return pwndbg.aglib.memory.get_typed_pointer("unsigned long", modules)
 
     def _node_data(self):
         raise NotImplementedError()
@@ -307,6 +319,9 @@ class ArchSymbols:
     def _per_cpu_offset(self):
         raise NotImplementedError()
 
+    def _modules(self):
+        raise NotImplementedError()
+
 
 class x86_64Symbols(ArchSymbols):
     # mov reg, [... - 0x...]
@@ -364,6 +379,10 @@ class x86_64Symbols(ArchSymbols):
             return result
         return self.qword_mov_reg_ripoff(disass)
 
+    def _modules(self):
+        disass = self.disass("find_module_all")
+        return self.qword_mov_reg_ripoff(disass)
+
 
 class Aarch64Symbols(ArchSymbols):
     # adrp x?, <kernel address>
@@ -406,8 +425,26 @@ class Aarch64Symbols(ArchSymbols):
         m = pattern.search(disass)
         if m is None:
             return None
-        return sum([int(m.group(i), 16) for i in [2, 3, 4]])
+        return sum(int(m.group(i), 16) for i in [2, 3, 4])
 
     def _per_cpu_offset(self):
         disass = self.disass("nr_iowait_cpu")
         return self.qword_adrp_add_const(disass)
+
+    def _modules(self):
+        disass = self.disass("find_module_all")
+        # adrp x<num>, 0x....
+        # ...
+        # add x<num>, x<num>, #0x...
+        # ...
+        # ldr x?, [x<num>, #0x]!...
+        pattern = re.compile(
+            r"adrp\s+x(\d+),\s+0x([0-9a-fA-F]+).*?\n"
+            r".*?add\s+x\1,\s+x\1,\s+#0x([0-9a-fA-F]+).*?\n"
+            r".*?ldr\s+x\d+,\s+\[x\1,\s+#0x([0-9a-fA-F]+)\]!",
+            re.DOTALL,
+        )
+        m = pattern.search(disass)
+        if m is None:
+            return None
+        return sum(int(m.group(i), 16) for i in [2, 3, 4])

pwndbg/commands/kbase.py

@@ -9,10 +9,6 @@ import pwndbg.dbg
 from pwndbg import config
 from pwndbg.commands import CommandCategory
 
-if pwndbg.dbg.is_gdblib_available():
-    import gdb
-
-
 parser = argparse.ArgumentParser(description="Finds the kernel virtual base address.")
 
 parser.add_argument("-r", "--rebase", action="store_true", help="rebase loaded symbol file")
@@ -40,9 +36,6 @@ def kbase(rebase=False) -> None:
     symbol_file = pwndbg.aglib.proc.exe
 
     if symbol_file:
-        if pwndbg.dbg.is_gdblib_available():
-            gdb.execute(f"add-symbol-file {symbol_file} {hex(base)}")
-        else:
-            print(M.error("Adding symbol not supported in LLDB yet"))
+        pwndbg.dbg.selected_inferior().add_symbol_file(symbol_file, base)
     else:
         print(M.error("No symbol file is currently loaded"))

pwndbg/commands/kmod.py

@@ -9,53 +9,85 @@ import argparse
 
 from tabulate import tabulate
 
+import pwndbg
+import pwndbg.aglib.kernel.kmod
+import pwndbg.color.message as M
 import pwndbg.commands
-from pwndbg.aglib.kernel.macros import for_each_entry
 
 parser = argparse.ArgumentParser(description="Displays the loaded Linux kernel modules.")
 parser.add_argument(
     "module_name", nargs="?", type=str, help="A module name substring to filter for"
 )
+parser.add_argument("-l", "--load", dest="path", type=str, help="the path of the module to load")
 
 
 @pwndbg.commands.Command(parser, category=pwndbg.commands.CommandCategory.KERNEL)
 @pwndbg.commands.OnlyWhenQemuKernel
 @pwndbg.commands.OnlyWhenPagingEnabled
-@pwndbg.commands.OnlyWithKernelDebugInfo
-def kmod(module_name=None) -> None:
+@pwndbg.commands.OnlyWithKernelDebugSymbols
+def kmod(module_name=None, path=None) -> None:
     # Look up the address of the `modules` symbol, containing the head of the linked list of kernel modules
-    modules_head = pwndbg.aglib.symbol.lookup_symbol_addr("modules")
+    modules_head = pwndbg.aglib.kernel.modules()
     if modules_head is None:
         print(
             "The modules symbol was not found. This may indicate that the symbol is not available in the current build."
         )
         return
 
-    print(f"Kernel modules address found at {modules_head:#x}.\n")
-
-    try:
-        table = []
-        headers = ["Address", "Name", "Size", "Used by"]
-        head = pwndbg.aglib.memory.get_typed_pointer_value("struct list_head", modules_head)
+    print(f"Kernel modules address found at {int(modules_head):#x}.\n")
 
+    table = []
+    headers = ["Address", "Name", "Size", "Used by"]
+    if pwndbg.aglib.typeinfo.load("struct module") is not None:
         # Iterate through the linked list of modules using for_each_entry
-        for module in for_each_entry(head, "struct module", "list"):
-            addr = int(module["mem"][0]["base"])
+        for module in pwndbg.aglib.kernel.kmod.module_list_with_typeinfo():
             name = pwndbg.aglib.memory.string(int(module["name"].address)).decode(
                 "utf-8", errors="ignore"
             )
-
-            # Calculate runtime memory footprint by summing sizes of MOD_TEXT, MOD_DATA, MOD_RODATA, MOD_RO_AFTER_INIT,
-            # which excludes initialization sections that are freed after the module load. See `enum mod_mem_type` in kernel source.
-            size = sum(int(module["mem"][i]["size"]) for i in range(4))
+            addr, size = None, None
+            if pwndbg.aglib.kernel.krelease() >= (6, 4):
+                addr = int(module["mem"][0]["base"])
+                size = sum(
+                    int(module["mem"][i]["size"])
+                    for i in range(pwndbg.aglib.kernel.kmod.mod_mem_type.MOD_MEM_NUM_TYPES.value)
+                )
+            else:
+                addr = int(module["init_layout"]["addr"])
+                size = module["init_layout"]["size"]
             uses = int(module["refcnt"]["counter"]) - 1
 
             # If module_name is provided, filter modules by name substring
             if not module_name or module_name in name:
                 table.append([f"{addr:#x}", name, size, uses])
+    else:
+        cur = pwndbg.aglib.memory.read_pointer_width(int(modules_head))
+        name_offset = pwndbg.aglib.kernel.kmod.module_name_offset()
+        for cur in pwndbg.aglib.kernel.kmod.module_list():
+            name = pwndbg.aglib.memory.string(cur + name_offset).decode()
+            if pwndbg.aglib.kernel.krelease() >= (6, 4):
+                mem_offset, module_memory_size, size_offset = (
+                    pwndbg.aglib.kernel.kmod.module_mem_offset()
+                )
+                addr = pwndbg.aglib.memory.read_pointer_width(cur + mem_offset)
+                size = 0
+                for i in range(pwndbg.aglib.kernel.kmod.mod_mem_type.MOD_MEM_NUM_TYPES.value):
+                    ptr = cur + mem_offset + module_memory_size * i
+                    size += pwndbg.aglib.memory.u32(ptr + size_offset)
+            else:
+                addr_offset, size_offset = pwndbg.aglib.kernel.kmod.module_layout_offset()
+                addr = pwndbg.aglib.memory.read_pointer_width(cur + addr_offset)
+                size = pwndbg.aglib.memory.u32(cur + size_offset)
 
-        print(tabulate(table, headers=headers, tablefmt="simple"))
-    except Exception as e:
-        print(
-            f"An error occurred while retrieving kernel modules. It may not be supported by your kernel version or debug symbols: {e}"
-        )
+            if not module_name or module_name in name:
+                table.append([f"{addr:#x}", name, size, "-"])
+    if path is not None:
+        if len(table) == 1:
+            pwndbg.dbg.selected_inferior().add_symbol_file(path, table[0][0])
+            return
+        if len(table) > 1:
+            print(M.warn("Multiple modules detected with the given filter"))
+        else:
+            print(M.warn("No modules detected with the given filter."))
+        return
+
+    print(tabulate(table, headers=headers, tablefmt="simple"))

pwndbg/dbg/__init__.py

@@ -626,6 +626,12 @@ class Process:
         """
         raise NotImplementedError()
 
+    def add_symbol_file(self, path, base):
+        """
+        Adds a symbol file at base
+        """
+        raise NotImplementedError()
+
 
 class TypeCode(Enum):
     """

pwndbg/dbg/gdb/__init__.py

@@ -976,6 +976,10 @@ class GDBProcess(pwndbg.dbg_mod.Process):
                 # We're done.
                 break
 
+    @override
+    def add_symbol_file(self, path, base):
+        gdb.execute(f"add-symbol-file {path} {base}")
+
 
 class GDBExecutionController(pwndbg.dbg_mod.ExecutionController):
     @override

tests/library/qemu_system/tests/test_commands_kernel.py

@@ -49,9 +49,8 @@ def test_command_kdmesg():
 
 
 def test_command_kmod():
-    if not pwndbg.aglib.kernel.has_debug_info():
+    if not pwndbg.aglib.kernel.has_debug_symbols("find_module_all"):
         res = gdb.execute("kmod", to_string=True)
-        assert "may only be run when debugging a Linux kernel with debug" in res
         return
 
     res = gdb.execute("kmod", to_string=True)