Initial implementation of landlock calls

Author: Riolku

README.md

@@ -108,7 +108,7 @@ $ pip install -e .
 Several environment variables can be specified to control the compilation of the sandbox:
 
 * `DMOJ_USE_SECCOMP`; set it to `no` if you're building on a pre-Linux 3.5 kernel to disable `seccomp` filtering in favour of pure `ptrace` (slower).
-   This flag has no effect when building outside of Linux.
+* `DMOJ_USE_LANDLOCK`; set it to `yes` if you're building on Linux after `5.13` to enable `landlock` sandboxing, which is faster.
 * `DMOJ_TARGET_ARCH`; use it to override the default architecture specified for compiling the sandbox (via `-march`).
    Usually this is `native`, but will not be specified on ARM unless `DMOJ_TARGET_ARCH` is set (a generic, slow build will be compiled instead).
 

dmoj/cptbox/_cptbox.pyx

@@ -1,5 +1,6 @@
 # cython: language_level=3
 from cpython.exc cimport PyErr_NoMemory, PyErr_SetFromErrno
+from cpython.ref cimport PyObject
 from libc.stdio cimport FILE, fopen, fclose, fgets, sprintf
 from libc.stdlib cimport malloc, free, strtoul
 from libc.string cimport strncmp, strlen
@@ -68,9 +69,12 @@ cdef extern from 'ptbox.h' nogil:
         bint was_initialized()
         bool use_seccomp()
         bool use_seccomp(bool enabled)
+        bool use_landlock()
+        bool use_landlock(bool enabled)
 
     cdef bint PTBOX_FREEBSD
     cdef bint PTBOX_SECCOMP
+    cdef bint PTBOX_LANDLOCK
     cdef int MAX_SYSCALL
 
     cdef int PTBOX_EVENT_ATTACH
@@ -121,6 +125,13 @@ cdef extern from 'helper.h' nogil:
         bool use_seccomp
         int abi_for_seccomp
         int *seccomp_handlers
+        bool use_landlock
+        char** read_exact_files
+        char** read_exact_dirs
+        char** read_recursive_dirs
+        char** write_exact_files
+        char** write_exact_dirs
+        char** write_recursive_dirs
 
     void cptbox_closefrom(int lowfd)
     int cptbox_child_run(child_config *)
@@ -510,6 +521,16 @@ cdef class Process:
                 for i in range(MAX_SYSCALL):
                     config.seccomp_handlers[i] = handlers[i]
 
+            config.use_landlock = self._use_landlock()
+            if config.use_landlock:
+                config.read_exact_files = alloc_byte_array(self.read_exact_files)
+                config.read_exact_dirs = alloc_byte_array(self.read_exact_dirs)
+                config.read_recursive_dirs = alloc_byte_array(self.read_recursive_dirs)
+                config.write_exact_files = alloc_byte_array(self.write_exact_files)
+                config.write_exact_dirs = alloc_byte_array(self.write_exact_dirs)
+                config.write_recursive_dirs = alloc_byte_array(self.write_recursive_dirs)
+
+
             if self.process.spawn(pt_child, &config):
                 raise RuntimeError('failed to spawn child')
         finally:
@@ -537,6 +558,18 @@ cdef class Process:
         if not self.process.use_seccomp(enabled):
             raise RuntimeError("Can't change whether seccomp is used after process is created.")
 
+    cdef inline bool _use_landlock(self):
+        return self.process.use_landlock()
+
+    @property
+    def use_landlock(self):
+        return self.process.use_landlock()
+
+    @use_landlock.setter
+    def use_landlock(self, bool enabled):
+        if not self.process.use_landlock(enabled):
+            raise RuntimeError("Can't change whether landlock is used after process is created.")
+
     @property
     def was_initialized(self):
         return self.process.was_initialized()

dmoj/cptbox/helper.cpp

@@ -30,6 +30,10 @@
 #   define FD_DIR "/proc/self/fd"
 #endif
 
+#if PTBOX_LANDLOCK
+#   include "landlock_helpers.h"
+#endif
+
 inline void setrlimit2(int resource, rlim_t cur, rlim_t max) {
     rlimit limit;
     limit.rlim_cur = cur;
@@ -117,6 +121,55 @@ int cptbox_child_run(const struct child_config *config) {
     }
 #endif
 
+#if PTBOX_LANDLOCK
+    int ruleset_fd;
+    struct landlock_ruleset_attr ruleset_attr = {
+        .handled_access_fs =
+            LANDLOCK_ACCESS_FS_EXECUTE |
+            LANDLOCK_ACCESS_FS_WRITE_FILE |
+            LANDLOCK_ACCESS_FS_READ_FILE |
+            LANDLOCK_ACCESS_FS_READ_DIR |
+            LANDLOCK_ACCESS_FS_REMOVE_DIR |
+            LANDLOCK_ACCESS_FS_REMOVE_FILE |
+            LANDLOCK_ACCESS_FS_MAKE_CHAR |
+            LANDLOCK_ACCESS_FS_MAKE_DIR |
+            LANDLOCK_ACCESS_FS_MAKE_REG |
+            LANDLOCK_ACCESS_FS_MAKE_SOCK |
+            LANDLOCK_ACCESS_FS_MAKE_FIFO |
+            LANDLOCK_ACCESS_FS_MAKE_BLOCK |
+            LANDLOCK_ACCESS_FS_MAKE_SYM,
+    };
+
+    ruleset_fd = landlock_create_ruleset(&ruleset_attr, sizeof(ruleset_attr), 0);
+    if (ruleset_fd < 0) {
+        perror("Failed to create a ruleset");
+        return PTBOX_SPAWN_FAIL_LANDLOCK;
+    }
+
+#define READ_EXACT_FILE_RULE      LANDLOCK_ACCESS_FS_EXECUTE | LANDLOCK_ACCESS_FS_READ_FILE
+#define READ_EXACT_DIR_RULE       LANDLOCK_ACCESS_FS_READ_DIR
+#define READ_RECURSIVE_DIR_RULE   READ_EXACT_FILE_RULE | READ_EXACT_DIR_RULE
+#define WRITE_EXACT_FILE_RULE     LANDLOCK_ACCESS_FS_WRITE_FILE
+#define WRITE_EXACT_DIR_RULE      LANDLOCK_ACCESS_FS_READ_DIR
+#define WRITE_RECURSIVE_DIR_RULE  WRITE_EXACT_FILE_RULE | WRITE_EXACT_DIR_RULE
+
+    if(
+        add_rules(ruleset_fd, config->read_exact_files,     READ_EXACT_FILE_RULE) ||
+        add_rules(ruleset_fd, config->read_exact_dirs,      READ_EXACT_DIR_RULE) ||
+        add_rules(ruleset_fd, config->read_recursive_dirs,  READ_RECURSIVE_DIR_RULE) ||
+        add_rules(ruleset_fd, config->write_exact_files,    WRITE_EXACT_FILE_RULE) ||
+        add_rules(ruleset_fd, config->write_exact_dirs,     WRITE_EXACT_DIR_RULE) ||
+        add_rules(ruleset_fd, config->write_recursive_dirs, WRITE_RECURSIVE_DIR_RULE)
+    ) {
+        return PTBOX_SPAWN_FAIL_LANDLOCK;
+    }
+
+    if(landlock_restrict_self(ruleset_fd, 0)) {
+    		perror("Failed to enforce ruleset");
+    		return PTBOX_SPAWN_FAIL_LANDLOCK;
+  	}
+#endif
+
     // All these limits should be dropped after initializing seccomp, since seccomp allocates
     // memory, and if an arena isn't sufficiently free it could force seccomp into an OOM
     // situation where we'd fail to initialize.

dmoj/cptbox/helper.h

@@ -6,6 +6,7 @@
 #define PTBOX_SPAWN_FAIL_SECCOMP        203
 #define PTBOX_SPAWN_FAIL_TRACEME        204
 #define PTBOX_SPAWN_FAIL_EXECVE         205
+#define PTBOX_SPAWN_FAIL_LANDLOCK       206
 
 struct child_config {
     unsigned long memory;
@@ -23,6 +24,13 @@ struct child_config {
     int stderr_;
     bool use_seccomp;
     int *seccomp_handlers;
+    bool use_landlock;
+    char** read_exact_files;
+    char** read_exact_dirs;
+    char** read_recursive_dirs;
+    char** write_exact_files;
+    char** write_exact_dirs;
+    char** write_recursive_dirs;
 };
 
 void cptbox_closefrom(int lowfd);

dmoj/cptbox/landlock.h

@@ -0,0 +1,58 @@
+#include "landlock_helpers.h"
+
+#if PTBOX_LANDLOCK
+    #include <linux/landlock.h>
+
+    int add_rules(const int ruleset_fd, const char* const* const paths, __u64 access_rule) {
+        struct landlock_path_beneath_attr path_beneath = {
+            .parent_fd = -1,
+            .allowed_access = access_rule,
+        };
+
+        for(const char* const* pathptr = paths; *pathptr; pathptr++) {
+            path_beneath.parent_fd = open(*pathptr, O_PATH | O_CLOEXEC);
+            if(path_beneath.parent_fd < 0) {
+                perror("Failed to open path for rule");
+                return -1;
+            }
+            if(landlock_add_rule(ruleset_fd, LANDLOCK_RULE_PATH_BENEATH, &path_beneath, 0)) {
+                perror("Failed to add rule to ruleset");
+                return -1;
+            }
+            if(close(path_beneath.parent_fd)) {
+                // Not Fatal: we have a CLOEXEC flag
+                perror("Failed to close path for rule");
+            }
+        }
+
+        return 0;
+    }
+
+    #ifndef landlock_create_ruleset
+    int landlock_create_ruleset(
+    		const struct landlock_ruleset_attr *const attr,
+    		const size_t size, const __u32 flags)
+    {
+    	return syscall(__NR_landlock_create_ruleset, attr, size, flags);
+    }
+    #endif
+
+    #ifndef landlock_add_rule
+    int landlock_add_rule(const int ruleset_fd,
+    		const enum landlock_rule_type rule_type,
+    		const void *const rule_attr, const __u32 flags)
+    {
+    	return syscall(__NR_landlock_add_rule, ruleset_fd, rule_type,
+    			rule_attr, flags);
+    }
+    #endif
+
+    #ifndef landlock_restrict_self
+    int landlock_restrict_self(const int ruleset_fd,
+    		const __u32 flags)
+    {
+    	return syscall(__NR_landlock_restrict_self, ruleset_fd, flags);
+    }
+    #endif
+
+#endif

dmoj/cptbox/landlock_helper.c

@@ -0,0 +1,3 @@
+#include <linux/types.h>
+
+int add_rules(const int ruleset_fd, const char* const* const paths, __u64 access_rule);

dmoj/cptbox/landlock_helpers.h

@@ -29,6 +29,12 @@
 #include <seccomp.h>
 #endif
 
+#ifdef PTBOX_USE_LANDLOCK
+#define PTBOX_LANDLOCK 1
+#else
+#define PTBOX_LANDLOCK 0
+#endif
+
 #if PTBOX_FREEBSD
 #include "ext_freebsd.h"
 #else
@@ -125,6 +131,8 @@ class pt_process {
     bool was_initialized() { return _initialized; }
     bool use_seccomp() { return _use_seccomp; }
     bool use_seccomp(bool enable);
+    bool use_landlock() { return _use_landlock; }
+    bool use_landlock(bool enable);
 protected:
     int dispatch(int event, unsigned long param);
     int protection_fault(int syscall, int type = PTBOX_EVENT_PROTECTION);
@@ -141,6 +149,7 @@ class pt_process {
     bool _trace_syscalls;
     bool _initialized;
     bool _use_seccomp;
+    bool _use_landlock;
 };
 
 class pt_debugger {

dmoj/cptbox/ptbox.h

@@ -105,6 +105,15 @@ bool pt_process::use_seccomp(bool enabled) {
     return true;
 }
 
+bool pt_process::use_landlock(bool enabled) {
+    if (pid) {
+        // Do not allow updates after the process is spawned.
+        return false;
+    }
+    _use_landlock = PTBOX_LANDLOCK && enabled;
+    return true;
+}
+
 int pt_process::monitor() {
     bool in_syscall = false, first = true, spawned = false;
     struct timespec start, end, delta;

dmoj/cptbox/ptproc.cpp

@@ -9,6 +9,7 @@
 from typing import Callable, Dict, List, Optional, Tuple, Type
 
 from dmoj.cptbox._cptbox import *
+from dmoj.cptbox.filesystem_policies import ExactDir, ExactFile, FilesystemAccessRule, RecursiveDir
 from dmoj.cptbox.handlers import ALLOW, DISALLOW, ErrnoHandlerCallback, _CALLBACK
 from dmoj.cptbox.syscalls import SYSCALL_COUNT, by_id, sys_exit, sys_exit_group, sys_getpid, translator
 from dmoj.utils.communicate import safe_communicate as _safe_communicate
@@ -30,6 +31,7 @@
 
 FREEBSD = sys.platform.startswith('freebsd')
 BAD_SECCOMP = sys.platform == 'linux' and tuple(map(int, os.uname().release.partition('-')[0].split('.'))) < (4, 8)
+BAD_LANDLOCK = sys.platform == 'linux' and tuple(map(int, os.uname().release.partition('-')[0].split('.'))) < (5, 13)
 
 _address_bits = {
     PTBOX_ABI_X86: 32,
@@ -98,8 +100,11 @@ def __init__(
         self,
         args: List[bytes],
         *,
+        read_fs: List[FilesystemAccessRule] = [],
+        write_fs: List[FilesystemAccessRule] = [],
         executable: bytes,
         avoid_seccomp: bool = False,
+        avoid_landlock: bool = False,
         security=None,
         time: int = 0,
         memory: int = 0,
@@ -117,11 +122,16 @@ def __init__(
     ) -> None:
         self._executable = executable
         self.use_seccomp = security is not None and not avoid_seccomp
+        self.use_landlock = security is not None and not avoid_landlock
 
         if self.use_seccomp and BAD_SECCOMP:
             log.warning('Requires Linux 4.8+ to use seccomp, you have: %s', os.uname().release)
             self.use_seccomp = False
 
+        if self.use_landlock and BAD_LANDLOCK:
+            log.warning(f'Requires Linux 5.13+ to use landlock, you have: {os.uname().release}')
+            self.use_landlock = False
+
         self._args = args
         self._chdir = cwd
         self._env = [
@@ -164,6 +174,9 @@ def __init__(
                             handler = _CALLBACK
                         self._handler(abi, call, handler)
 
+        if self.use_landlock:
+            self.load_files(read_fs, write_fs)
+
         self._died = threading.Event()
         self._spawned_or_errored = threading.Event()
         self._spawn_error = None
@@ -201,6 +214,19 @@ def _get_seccomp_handlers(self) -> List[int]:
                     handlers[call] = handler.errno
         return handlers
 
+
+    def load_files(self, read_fs, write_fs) -> None:
+        def filter(source: List[FilesystemAccessRule], type) -> List[type]:
+            return [rule for rule in source if isinstance(rule, type)]
+
+        self.read_exact_files = filter(read_fs, ExactFile)
+        self.read_exact_dirs = filter(read_fs, ExactDir)
+        self.read_recursive_dirs = filter(read_fs, RecursiveDir)
+        
+        self.write_exact_files = filter(write_fs, ExactFile)
+        self.write_exact_dirs = filter(write_fs, ExactDir)
+        self.write_recursive_dirs = filter(write_fs, RecursiveDir)
+
     def wait(self) -> int:
         self._died.wait()
         assert self.returncode is not None

dmoj/cptbox/tracer.py

@@ -158,6 +158,8 @@ def launch(self, *args, **kwargs):
 
         return TracedPopen(
             [utf8bytes(a) for a in self.get_cmdline(**kwargs) + list(args)],
+            read_fs=self.get_fs(),
+            write_fs=self.get_write_fs(),
             executable=utf8bytes(self.get_executable()),
             security=self.get_security(launch_kwargs=kwargs),
             address_grace=self.get_address_grace(),

dmoj/executors/mixins.py

@@ -24,6 +24,8 @@
 
 # Allow manually disabling seccomp on old kernels. WSL 1 doesn't have seccomp.
 has_seccomp = sys.platform.startswith('linux') and not is_wsl1 and os.environ.get('DMOJ_USE_SECCOMP') != 'no'
+# Must explicitly enable landlock
+has_landlock = sys.platform.startswith('linux') and os.environ.get('DMOJ_USE_LANDLOCK') == 'yes'
 try:
     parallel = int(os.environ['DMOJ_PARALLEL'])
 except (KeyError, ValueError):
@@ -169,6 +171,9 @@ def unavailable(self, e):
     print('*' * 79)
     macros.append(('PTBOX_NO_SECCOMP', None))
 
+if has_landlock:
+    macros.append(('PTBOX_USE_LANDLOCK', None))
+
 extensions = [
     Extension('dmoj.checkers._checker', sources=['dmoj/checkers/_checker.c']),
     Extension('dmoj.cptbox._cptbox', sources=cptbox_sources, language='c++', libraries=libs, define_macros=macros),