change regex to filesystem policy matcher, first step of DMOJ#871

.freebsd.test.py

@@ -2,6 +2,7 @@
 
 from dmoj import judgeenv
 from dmoj.citest import ci_test
+from dmoj.executors.filesystem_policies import FilesystemPolicy, ExactFile
 
 EXECUTORS = [
     'AWK',
@@ -31,13 +32,16 @@
 
 def main():
     judgeenv.env['runtime'] = {}
-    judgeenv.env['extra_fs'] = {'PERL': ['/dev/dtrace/helper$'], 'RUBY2': ['/dev/dtrace/helper$']}
+    judgeenv.env['extra_fs'] = {
+        'PERL': FilesystemPolicy([ExactFile("/dev/dtrace/helper")]),
+        'RUBY2': FilesystemPolicy([ExactFile("/dev/dtrace/helper")])
+    }
 
     logging.basicConfig(level=logging.INFO)
 
     print('Using extra allowed filesystems:')
     for lang, fs in judgeenv.env['extra_fs'].iteritems():
-        print('%-6s: %s' % (lang, '|'.join(fs)))
+        print('%-6s: %s' % (lang, '|'.join([x.path for x in fs.rules])))
     print()
 
     print('Testing executors...')

dmoj/cptbox/isolate.py

@@ -1,14 +1,13 @@
 import logging
 import os
-import re
 import sys
 
 from dmoj.cptbox._cptbox import AT_FDCWD, bsd_get_proc_cwd, bsd_get_proc_fdno
 from dmoj.cptbox.tracer import MaxLengthExceeded
 from dmoj.cptbox.handlers import ACCESS_EACCES, ACCESS_ENOENT, ACCESS_EPERM, ALLOW
-
 from dmoj.cptbox.syscalls import *
 from dmoj.utils.unicode import utf8text
+from dmoj.executors.filesystem_policies import FilesystemPolicy
 
 log = logging.getLogger('dmoj.security')
 open_write_flags = [os.O_WRONLY, os.O_RDWR, os.O_TRUNC, os.O_CREAT, os.O_EXCL]
@@ -185,11 +184,12 @@ def __init__(self, read_fs, write_fs=None, writable=(1, 2)):
 
     def _compile_fs_jail(self, fs):
         if fs:
-            fs_re = '|'.join(fs)
+            fs.build()
+            return fs
         else:
-            fs_re = '(?!)'  # Disallow accessing everything by default.
-
-        return re.compile(fs_re)
+            dummy = FilesystemPolicy([])
+            dummy.build()
+            return dummy
 
     def is_write_flags(self, open_flags):
         for flag in open_write_flags:
@@ -258,9 +258,7 @@ def _file_access_check(self, rel_file, debugger, is_open, flag_reg=1, dirfd=AT_F
 
         is_write = is_open and self.is_write_flags(getattr(debugger, 'uarg%d' % flag_reg))
         fs_jail = self.write_fs_jail if is_write else self.read_fs_jail
-        if fs_jail.match(file) is None:
-            return file, False
-        return file, True
+        return file, fs_jail.check(file)
 
     def get_full_path(self, debugger, file, dirfd=AT_FDCWD):
         dirfd = (dirfd & 0x7FFFFFFF) - (dirfd & 0x80000000)

dmoj/executors/COFFEE.py

@@ -1,6 +1,7 @@
 import os
 
 from dmoj.executors.script_executor import ScriptExecutor
+from dmoj.executors.filesystem_policies import FilesystemPolicy, ExactFile
 
 
 class Executor(ScriptExecutor):
@@ -38,7 +39,10 @@ def get_cmdline(self, **kwargs):
         return [self.get_command(), self.runtime_dict['coffee'], self._code]
 
     def get_fs(self):
-        return super().get_fs() + [self.runtime_dict['coffee'], self._code]
+        return super().get_fs() + FilesystemPolicy([
+            ExactFile(self.runtime_dict['coffee']),
+            ExactFile(self._code)
+        ])
 
     @classmethod
     def get_versionable_commands(cls):

dmoj/executors/DART.py

@@ -16,7 +16,6 @@ class Executor(CompiledExecutor):
     address_grace = 128 * 1024
 
     syscalls = ['epoll_create', 'epoll_ctl', 'epoll_wait', 'timerfd_settime', 'memfd_create', 'ftruncate']
-    fs = ['.*/vm-service$']
 
     def get_compile_args(self):
         return [self.get_command(), '--snapshot=%s' % self.get_compiled_file(), self._code]

dmoj/executors/FORTH.py

@@ -1,4 +1,5 @@
 from dmoj.executors.script_executor import ScriptExecutor
+from dmoj.executors.filesystem_policies import FilesystemPolicy, ExactFile
 
 
 class Executor(ScriptExecutor):
@@ -10,7 +11,9 @@ class Executor(ScriptExecutor):
 
 HELLO
 '''
-    fs = [r'/\.gforth-history$']
+    fs = FilesystemPolicy([
+        ExactFile('/.gforth-history')
+    ])
 
     def get_cmdline(self, **kwargs):
         return [self.get_command(), self._code, '-e', 'bye']

dmoj/executors/PERL.py

@@ -1,11 +1,14 @@
 from dmoj.executors.script_executor import ScriptExecutor
+from dmoj.executors.filesystem_policies import FilesystemPolicy, RecursiveDir
 
 
 class Executor(ScriptExecutor):
     ext = 'pl'
     name = 'PERL'
     command = 'perl'
-    fs = ['/etc/perl/.*?']
+    fs = FilesystemPolicy([
+        RecursiveDir('/etc/perl')
+    ])
     test_program = 'print<>'
     syscalls = ['umtx_op']
 

dmoj/executors/PHP.py

@@ -7,8 +7,6 @@ class Executor(ScriptExecutor):
     command = 'php'
     command_paths = ['php7', 'php5', 'php']
 
-    fs = [r'.*/php[\w-]*\.ini$', r'.*/conf.d/.*\.ini$']
-
     test_program = '<?php while($f = fgets(STDIN)) echo $f;'
 
     def get_cmdline(self, **kwargs):

dmoj/executors/RKT.py

@@ -1,13 +1,14 @@
-import os
-
 from dmoj.executors.compiled_executor import CompiledExecutor
+from dmoj.executors.filesystem_policies import FilesystemPolicy, RecursiveDir, ExactFile
 
 
 class Executor(CompiledExecutor):
     ext = 'rkt'
     name = 'RKT'
-    fs = [os.path.expanduser(r'~/\.racket/'), os.path.expanduser(r'~/\.local/share/racket/'),
-          '/etc/racket/.*?', '/etc/passwd$']
+    fs = FilesystemPolicy([
+        RecursiveDir("/etc/racket"),
+        ExactFile("/etc/passwd")
+    ])
 
     command = 'racket'
 

dmoj/executors/RUBY2.py

@@ -2,6 +2,7 @@
 import re
 
 from dmoj.executors.script_executor import ScriptExecutor
+from dmoj.executors.filesystem_policies import FilesystemPolicy, ExactFile
 
 
 class Executor(ScriptExecutor):
@@ -12,7 +13,9 @@ class Executor(ScriptExecutor):
     nproc = -1
     command_paths = ['ruby2.%d' % i for i in reversed(range(0, 8))] + ['ruby2%d' % i for i in reversed(range(0, 8))]
     syscalls = ['thr_set_name', 'eventfd2']
-    fs = ['/proc/self/loginuid$']
+    fs = FilesystemPolicy([
+        ExactFile('/proc/self/loginuid')
+    ])
 
     def get_fs(self):
         fs = super().get_fs()

dmoj/executors/SWIFT.py

@@ -5,7 +5,6 @@ class Executor(CompiledExecutor):
     ext = 'swift'
     name = 'SWIFT'
     command = 'swiftc'
-    fs = ['/lib']
     test_program = 'print(readLine()!)'
 
     def get_compile_args(self):

dmoj/executors/TUR.py

@@ -2,6 +2,7 @@
 
 from dmoj.executors.compiled_executor import CompiledExecutor
 from dmoj.judgeenv import env
+from dmoj.executors.filesystem_policies import FilesystemPolicy, ExactFile
 
 
 class Executor(CompiledExecutor):
@@ -15,7 +16,9 @@ class Executor(CompiledExecutor):
 '''
 
     def get_fs(self):
-        return super().get_fs() + [self._code + 'bc']
+        return super().get_fs() + FilesystemPolicy([
+            ExactFile(self._code + 'bc')
+        ])
 
     def get_compile_args(self):
         tprologc = self.runtime_dict['tprologc']

dmoj/executors/filesystem_policies.py

@@ -0,0 +1,136 @@
+from typing import List, Union
+
+
+class ACCESS_MODE:
+    NONE = 0
+    EXACT = 1
+    RECURSIVE = 2
+
+
+class Dir:
+    def __init__(self):
+        self.access_mode = ACCESS_MODE.NONE
+
+        self.map = {}
+
+
+class File:
+    pass
+
+
+class ExactFile:
+    def __init__(self, path: str):
+        self.path = path
+
+
+class ExactDir:
+    access_mode = ACCESS_MODE.EXACT
+
+    def __init__(self, path: str):
+        self.path = path
+
+
+class RecursiveDir:
+    access_mode = ACCESS_MODE.RECURSIVE
+
+    def __init__(self, path: str):
+        self.path = path
+
+
+Rule = Union[ExactFile, ExactDir, RecursiveDir]
+
+
+class FilesystemPolicy:
+    def __init__(self, rules: List[Rule]):
+        self.rules = rules
+
+        self.built = False
+
+    def __add__(self, other):
+        assert not self.built and not other.built, "Refusing to merge built policies: merge before building"
+
+        return FilesystemPolicy(self.rules + other.rules)
+
+    def build(self):
+        self.root = Dir()
+
+        for rule in self.rules:
+            self._add_rule(rule)
+
+        self.built = True
+
+    def _add_rule(self, rule: Rule):
+        if rule.path == "/":
+            return self._finalize_root_rule(rule)
+
+        node = self.root
+        path = rule.path
+
+        assert path[0] == '/', "Rule must specify an absolute path to rule"
+
+        *directory_path, final_component = path[1:].split("/")
+
+        for component in directory_path:
+            assert component != "", "Must not have empty components in rule to add"
+
+            new_node = node.map.setdefault(component, Dir())
+
+            assert isinstance(new_node, Dir), "Cannot descend into non-directory"
+
+            node = new_node
+
+        self._finalize_rule(node, final_component, rule)
+
+    def _finalize_root_rule(self, rule: Rule):
+        assert not isinstance(rule, ExactFile), "Root is not a file"
+
+        self._finalize_directory_rule(self.root, rule)
+
+    def _finalize_rule(self, node: Dir, final_component: str, rule: Rule):
+        assert final_component != "", "Must not have trailing slashes in rule path"
+
+        if isinstance(rule, ExactFile):
+            new_node = node.map.setdefault(final_component, File())
+
+            assert isinstance(new_node, File), "Can't add ExactFile: Dir rule exists"
+
+        else:
+            new_node = node.map.setdefault(final_component, Dir())
+
+            assert isinstance(new_node, Dir), "Can't add rule: File rule exists"
+
+            self._finalize_directory_rule(new_node, rule)
+
+    def _finalize_directory_rule(self, node: Dir, rule: Union[ExactDir, RecursiveDir]):
+        node.access_mode = max(node.access_mode, rule.access_mode)  # Allow the more permissive rule
+
+    # `path` should be a canonical path. The output of `realpath` is appropriate
+    def check(self, path: str) -> bool:
+        if path == "/":
+            return self._check_final_node(self.root)
+
+        assert self.built, "Must build the policy object before checking"
+        assert path[0] == '/', "Must pass an absolute path to check"
+
+        path_split = path[1:].split("/")
+        node = self.root
+
+        for component in path_split:
+            assert component != "", "Must not have empty components in path to check"
+
+            if isinstance(node, File):
+                return False
+
+            elif node.access_mode == ACCESS_MODE.RECURSIVE:
+                return True
+
+            else:
+                node = node.map.get(component)
+
+                if node is None:
+                    return False
+
+        return self._check_final_node(node)
+
+    def _check_final_node(self, node: Union[Dir, File]) -> bool:
+        return isinstance(node, File) or node.access_mode != ACCESS_MODE.NONE

dmoj/executors/java_executor.py

@@ -11,6 +11,7 @@
 from dmoj.executors.mixins import SingleDigitVersionMixin
 from dmoj.judgeenv import skip_self_test
 from dmoj.utils.unicode import utf8bytes, utf8text
+from dmoj.executors.filesystem_policies import FilesystemPolicy, ExactFile
 
 recomment = re.compile(r'/\*.*?\*/', re.DOTALL | re.U)
 restring = re.compile(r''''(?:\\.|[^'\\])'|"(?:\\.|[^"\\])*"''', re.DOTALL | re.U)
@@ -81,10 +82,14 @@ def get_executable(self):
         return self.get_vm()
 
     def get_fs(self):
-        return super().get_fs() + [self._agent_file]
+        return super().get_fs() + FilesystemPolicy([
+            ExactFile(self._agent_file)
+        ])
 
     def get_write_fs(self):
-        return super().get_write_fs() + [os.path.join(self._dir, 'submission_jvm_crash.log')]
+        return super().get_write_fs() + FilesystemPolicy([
+            ExactFile(os.path.join(self._dir, 'submission_jvm_crash.log'))
+        ])
 
     def get_agent_flag(self):
         agent_flag = '-javaagent:%s=policy:%s' % (self._agent_file, self._policy_file)