Linux: Add support for threads in both lsof and sockstat plugins.

gcmoreira · gcmoreira · commit 867c8e3f1bde · 2024-09-13T15:33:19.000+10:00
- lsof plugin: source code refactored - lsof plugin: Added the 'device' column to complete the inode information. An inode number is specific to the filesystem/device it belongs to. - lsof/sockstat plugins: Add threads support. Threads may or may not share the file descriptor table with the thread group leader, depending on whether the CLONE_FILES flag is included in the clone() syscall. Also, once started, a thread can unshare the fd table with its parent. Refer to the unshare() libc syscall wrapper man page, unshare(2). Additionally, note that the Linux lsof command in user space includes thread listings by default as well. Now, there are two columns to identify the thread group ID (PID) and the task/thread ID (TID). - Added inode getters from both, the dentry and file structs. From kernels +3.9 the file struct cached the inode pointer. So, when possible, we get this value. - Improve smear protection in these plugins and various APIs (volatilityfoundation#1243)
diff --git a/volatility3/framework/plugins/linux/lsof.py b/volatility3/framework/plugins/linux/lsof.py
@@ -1,12 +1,12 @@
 # This file is Copyright 2024 Volatility Foundation and licensed under the Volatility Software License 1.0
 # which is available at https://www.volatilityfoundation.org/license/vsl-v1.0
 #
-"""A module containing a collection of plugins that produce data typically
-found in Linux's /proc file system."""
-import logging, datetime
+import logging
+from datetime import datetime
+from dataclasses import dataclass, astuple, field
 from typing import List, Callable
 
-from volatility3.framework import renderers, interfaces, constants, exceptions
+from volatility3.framework import renderers, interfaces, constants
 from volatility3.framework.configuration import requirements
 from volatility3.framework.interfaces import plugins
 from volatility3.framework.objects import utility
@@ -17,11 +17,94 @@
 vollog = logging.getLogger(__name__)
 
 
+@dataclass
+class FDUser:
+    """FD user representation, featuring augmented information and formatted fields.
+    This is the data the plugin will eventually display.
+    """
+
+    task_tgid: int
+    task_tid: int
+    task_comm: str
+    fd_num: int
+    full_path: str
+    device: str = field(default=renderers.NotAvailableValue())
+    inode_num: int = field(default=renderers.NotAvailableValue())
+    inode_type: str = field(default=renderers.NotAvailableValue())
+    file_mode: str = field(default=renderers.NotAvailableValue())
+    change_time: datetime = field(default=renderers.NotAvailableValue())
+    modification_time: datetime = field(default=renderers.NotAvailableValue())
+    access_time: datetime = field(default=renderers.NotAvailableValue())
+    inode_size: int = field(default=renderers.NotAvailableValue())
+
+
+@dataclass
+class FDInternal:
+    """FD internal representation containing only the core objects
+
+    Fields:
+        task: 'task_truct' object
+        fd_fields: FD fields as obtained from LinuxUtilities.files_descriptors_for_process()
+    """
+
+    task: interfaces.objects.ObjectInterface
+    fd_fields: tuple[int, int, str]
+
+    def to_user(self) -> FDUser:
+        """Augment the FD information to be presented to the user
+
+        Returns:
+            An InodeUser dataclass
+        """
+        # Ensure all types are atomic immutable. Otherwise, astuple() will take a long
+        # time doing a deepcopy of the Volatility objects.
+        task_tgid = int(self.task.tgid)
+        task_tid = int(self.task.pid)
+        task_comm = utility.array_to_string(self.task.comm)
+        fd_num, filp, full_path = self.fd_fields
+        fd_num = int(fd_num)
+        full_path = str(full_path)
+        inode = filp.get_inode()
+        if inode:
+            superblock_ptr = inode.i_sb
+            if superblock_ptr and superblock_ptr.is_readable():
+                device = f"{superblock_ptr.major}:{superblock_ptr.minor}"
+            else:
+                device = renderers.NotAvailableValue()
+
+            fd_user = FDUser(
+                task_tgid=task_tgid,
+                task_tid=task_tid,
+                task_comm=task_comm,
+                fd_num=fd_num,
+                full_path=full_path,
+                device=device,
+                inode_num=int(inode.i_ino),
+                inode_type=inode.get_inode_type() or renderers.UnparsableValue(),
+                file_mode=inode.get_file_mode(),
+                change_time=inode.get_change_time(),
+                modification_time=inode.get_modification_time(),
+                access_time=inode.get_access_time(),
+                inode_size=int(inode.i_size),
+            )
+        else:
+            # We use the dataclasses' default values
+            fd_user = FDUser(
+                task_tgid=task_tgid,
+                task_tid=task_tid,
+                task_comm=task_comm,
+                fd_num=fd_num,
+                full_path=full_path,
+            )
+
+        return fd_user
+
+
 class Lsof(plugins.PluginInterface, timeliner.TimeLinerInterface):
     """Lists open files for each processes."""
 
     _required_framework_version = (2, 0, 0)
-    _version = (1, 2, 0)
+    _version = (2, 0, 0)
 
     @classmethod
     def get_requirements(cls) -> List[interfaces.configuration.RequirementInterface]:
@@ -45,126 +128,86 @@ def get_requirements(cls) -> List[interfaces.configuration.RequirementInterface]
             ),
         ]
 
-    @classmethod
-    def get_inode_metadata(cls, filp: interfaces.objects.ObjectInterface):
-        try:
-            dentry = filp.get_dentry()
-            if dentry:
-                inode_object = dentry.d_inode
-                if inode_object and inode_object.is_valid():
-                    itype = (
-                        inode_object.get_inode_type() or renderers.NotAvailableValue()
-                    )
-                    return (
-                        inode_object.i_ino,
-                        itype,
-                        inode_object.i_size,
-                        inode_object.get_file_mode(),
-                        inode_object.get_change_time(),
-                        inode_object.get_modification_time(),
-                        inode_object.get_access_time(),
-                    )
-        except (exceptions.InvalidAddressException, AttributeError) as e:
-            vollog.warning(f"Can't get inode metadata: {e}")
-        return None
-
     @classmethod
     def list_fds(
         cls,
         context: interfaces.context.ContextInterface,
-        symbol_table: str,
+        vmlinux_module_name: str,
         filter_func: Callable[[int], bool] = lambda _: False,
-    ):
+    ) -> FDInternal:
+        """Enumerates open file descriptors in tasks
+
+        Args:
+            context: The context to retrieve required elements (layers, symbol tables) from
+            vmlinux_module_name: The name of the kernel module on which to operate
+            filter_func: A function which takes a process object and returns True if the process
+                should be ignored/filtered
+
+        Yields:
+            A FDInternal object
+        """
         linuxutils_symbol_table = None
-        for task in pslist.PsList.list_tasks(context, symbol_table, filter_func):
+        for task in pslist.PsList.list_tasks(
+            context, vmlinux_module_name, filter_func, include_threads=True
+        ):
             if linuxutils_symbol_table is None:
                 if constants.BANG not in task.vol.type_name:
                     raise ValueError("Task is not part of a symbol table")
                 linuxutils_symbol_table = task.vol.type_name.split(constants.BANG)[0]
 
-            task_comm = utility.array_to_string(task.comm)
-            pid = int(task.pid)
-
             fd_generator = linux.LinuxUtilities.files_descriptors_for_process(
                 context, linuxutils_symbol_table, task
             )
 
             for fd_fields in fd_generator:
-                yield pid, task_comm, task, fd_fields
+                yield FDInternal(task=task, fd_fields=fd_fields)
 
-    @classmethod
-    def list_fds_and_inodes(
-        cls,
-        context: interfaces.context.ContextInterface,
-        symbol_table: str,
-        filter_func: Callable[[int], bool] = lambda _: False,
-    ):
-        for pid, task_comm, task, (fd_num, filp, full_path) in cls.list_fds(
-            context, symbol_table, filter_func
-        ):
-            inode_metadata = cls.get_inode_metadata(filp)
-            if inode_metadata is None:
-                inode_metadata = tuple(
-                    interfaces.renderers.BaseAbsentValue() for _ in range(7)
-                )
-            yield pid, task_comm, task, fd_num, filp, full_path, inode_metadata
-
-    def _generator(self, pids, symbol_table):
+    def _generator(self, pids, vmlinux_module_name):
         filter_func = pslist.PsList.create_pid_filter(pids)
-        fds_generator = self.list_fds_and_inodes(
-            self.context, symbol_table, filter_func=filter_func
-        )
-
-        for (
-            pid,
-            task_comm,
-            task,
-            fd_num,
-            filp,
-            full_path,
-            inode_metadata,
-        ) in fds_generator:
-            inode_num, itype, file_size, imode, ctime, mtime, atime = inode_metadata
-            fields = (
-                pid,
-                task_comm,
-                fd_num,
-                full_path,
-                inode_num,
-                itype,
-                imode,
-                ctime,
-                mtime,
-                atime,
-                file_size,
-            )
-            yield (0, fields)
+        for fd_internal in self.list_fds(
+            self.context, vmlinux_module_name, filter_func=filter_func
+        ):
+            fd_user = fd_internal.to_user()
+            yield (0, astuple(fd_user))
 
     def run(self):
         pids = self.config.get("pid", None)
-        symbol_table = self.config["kernel"]
+        vmlinux_module_name = self.config["kernel"]
 
         tree_grid_args = [
             ("PID", int),
+            ("TID", int),
             ("Process", str),
             ("FD", int),
             ("Path", str),
+            ("Device", str),
             ("Inode", int),
             ("Type", str),
             ("Mode", str),
-            ("Changed", datetime.datetime),
-            ("Modified", datetime.datetime),
-            ("Accessed", datetime.datetime),
+            ("Changed", datetime),
+            ("Modified", datetime),
+            ("Accessed", datetime),
             ("Size", int),
         ]
-        return renderers.TreeGrid(tree_grid_args, self._generator(pids, symbol_table))
+        return renderers.TreeGrid(
+            tree_grid_args, self._generator(pids, vmlinux_module_name)
+        )
 
     def generate_timeline(self):
         pids = self.config.get("pid", None)
-        symbol_table = self.config["kernel"]
-        for row in self._generator(pids, symbol_table):
-            _depth, row_data = row
-            description = f'Process {row_data[1]} ({row_data[0]}) Open "{row_data[3]}"'
-            yield description, timeliner.TimeLinerType.CHANGED, row_data[7]
-            yield description, timeliner.TimeLinerType.MODIFIED, row_data[8]
-            yield description, timeliner.TimeLinerType.ACCESSED, row_data[9]
+        vmlinux_module_name = self.config["kernel"]
+
+        filter_func = pslist.PsList.create_pid_filter(pids)
+        for fd_internal in self.list_fds(
+            self.context, vmlinux_module_name, filter_func=filter_func
+        ):
+            fd_user = fd_internal.to_user()
+
+            description = (
+                f"Process {fd_user.task_comm} ({fd_user.task_tgid}/{fd_user.task_tid}) "
+                f"Open '{fd_user.full_path}'"
+            )
+
+            yield description, timeliner.TimeLinerType.CHANGED, fd_user.change_time
+            yield description, timeliner.TimeLinerType.MODIFIED, fd_user.modification_time
+            yield description, timeliner.TimeLinerType.ACCESSED, fd_user.access_time
diff --git a/volatility3/framework/plugins/linux/sockstat.py b/volatility3/framework/plugins/linux/sockstat.py
@@ -22,9 +22,10 @@ class SockHandlers(interfaces.configuration.VersionableInterface):
 
     _required_framework_version = (2, 0, 0)
 
-    _version = (1, 0, 0)
+    _version = (1, 0, 1)
 
-    def __init__(self, vmlinux, task):
+    def __init__(self, vmlinux, task, *args, **kwargs):
+        super().__init__(*args, **kwargs)
         self._vmlinux = vmlinux
         self._task = task
 
@@ -438,7 +439,7 @@ class Sockstat(plugins.PluginInterface):
 
     _required_framework_version = (2, 0, 0)
 
-    _version = (1, 0, 0)
+    _version = (2, 0, 0)
 
     @classmethod
     def get_requirements(cls):
@@ -452,7 +453,7 @@ def get_requirements(cls):
                 name="SockHandlers", component=SockHandlers, version=(1, 0, 0)
             ),
             requirements.PluginRequirement(
-                name="lsof", plugin=lsof.Lsof, version=(1, 1, 0)
+                name="lsof", plugin=lsof.Lsof, version=(2, 0, 0)
             ),
             requirements.VersionRequirement(
                 name="linuxutils", component=linux.LinuxUtilities, version=(2, 0, 0)
@@ -507,8 +508,9 @@ def list_sockets(
         dfop_addr = vmlinux.object_from_symbol("sockfs_dentry_operations").vol.offset
 
         fd_generator = lsof.Lsof.list_fds(context, vmlinux.name, filter_func)
-        for _pid, _task_comm, task, fd_fields in fd_generator:
-            fd_num, filp, _full_path = fd_fields
+        for fd_internal in fd_generator:
+            fd_num, filp, _full_path = fd_internal.fd_fields
+            task = fd_internal.task
 
             if filp.f_op not in (sfop_addr, dfop_addr):
                 continue
@@ -617,6 +619,7 @@ def _generator(self, pids: List[int], netns_id_arg: int, symbol_table: str):
 
             fields = (
                 netns_id,
+                task.tgid,
                 task.pid,
                 fd_num,
                 format_hints.Hex(sock.vol.offset),
@@ -636,7 +639,8 @@ def run(self):
 
         tree_grid_args = [
             ("NetNS", int),
-            ("Pid", int),
+            ("PID", int),
+            ("TID", int),
             ("FD", int),
             ("Sock Offset", format_hints.Hex),
             ("Family", str),
diff --git a/volatility3/framework/symbols/linux/__init__.py b/volatility3/framework/symbols/linux/__init__.py
@@ -256,7 +256,7 @@ def files_descriptors_for_process(
         task: interfaces.objects.ObjectInterface,
     ):
         # task.files can be null
-        if not task.files:
+        if not (task.files and task.files.is_readable()):
             return None
 
         fd_table = task.files.get_fds()
@@ -276,7 +276,7 @@ def files_descriptors_for_process(
         )
 
         for fd_num, filp in enumerate(fds):
-            if filp != 0:
+            if filp and filp.is_readable():
                 full_path = LinuxUtilities.path_for_file(context, task, filp)
 
                 yield fd_num, filp, full_path
diff --git a/volatility3/framework/symbols/linux/extensions/__init__.py b/volatility3/framework/symbols/linux/extensions/__init__.py
