[ML] Add system call restrictions to the ML processes (elastic#98)

Restrict the ability of the autodetect, autoconfig, categorize and normalize
programs to make system calls. Implemented for Linux (Seccomp BPF),
 macOS (sandbox) and Windows (Job Groups)

Author: davidkyle

bin/autoconfig/Main.cc

@@ -28,6 +28,8 @@
 #include <config/CAutoconfigurerParams.h>
 #include <config/CReportWriter.h>
 
+#include <seccomp/CSystemCallFilter.h>
+
 #include "CCmdLineParser.h"
 
 #include <boost/bind.hpp>
@@ -76,6 +78,8 @@ int main(int argc, char** argv) {
 
     ml::core::CProcessPriority::reducePriority();
 
+    ml::seccomp::CSystemCallFilter::installSystemCallFilter();
+
     if (ioMgr.initIo() == false) {
         LOG_FATAL(<< "Failed to initialise IO");
         return EXIT_FAILURE;

bin/autodetect/Main.cc

@@ -42,6 +42,8 @@
 #include <api/CSingleStreamSearcher.h>
 #include <api/CStateRestoreStreamFilter.h>
 
+#include <seccomp/CSystemCallFilter.h>
+
 #include "CCmdLineParser.h"
 
 #include <boost/bind.hpp>
@@ -120,6 +122,8 @@ int main(int argc, char** argv) {
 
     ml::core::CProcessPriority::reducePriority();
 
+    ml::seccomp::CSystemCallFilter::installSystemCallFilter();
+
     if (ioMgr.initIo() == false) {
         LOG_FATAL(<< "Failed to initialise IO");
         return EXIT_FAILURE;

bin/categorize/Main.cc

@@ -38,6 +38,8 @@
 #include <api/CSingleStreamSearcher.h>
 #include <api/CStateRestoreStreamFilter.h>
 
+#include <seccomp/CSystemCallFilter.h>
+
 #include "CCmdLineParser.h"
 
 #include <boost/bind.hpp>
@@ -91,6 +93,8 @@ int main(int argc, char** argv) {
 
     ml::core::CProcessPriority::reducePriority();
 
+    ml::seccomp::CSystemCallFilter::installSystemCallFilter();
+
     if (ioMgr.initIo() == false) {
         LOG_FATAL(<< "Failed to initialise IO");
         return EXIT_FAILURE;

bin/normalize/Main.cc

@@ -28,6 +28,8 @@
 #include <api/CLineifiedJsonOutputWriter.h>
 #include <api/CResultNormalizer.h>
 
+#include <seccomp/CSystemCallFilter.h>
+
 #include "CCmdLineParser.h"
 
 #include <boost/bind.hpp>
@@ -78,6 +80,8 @@ int main(int argc, char** argv) {
 
     ml::core::CProcessPriority::reducePriority();
 
+    ml::seccomp::CSystemCallFilter::installSystemCallFilter();
+
     if (ioMgr.initIo() == false) {
         LOG_FATAL(<< "Failed to initialise IO");
         return EXIT_FAILURE;

docs/CHANGELOG.asciidoc

@@ -11,7 +11,7 @@
 
 === Deprecations
 
-=== New Features 
+=== New Features
 
 === Enhancements
 
@@ -24,6 +24,10 @@ Improve partition analysis memory usage ({pull}97[#97])
 Forecasting of Machine Learning job time series is now supported for large jobs by temporarily storing
 model state on disk ({pull}89[#89])
 
+Secure the ML processes by preventing system calls such as fork and exec. The Linux implemenation uses
+Seccomp BPF to intercept system calls and is available in kernels since 3.5. On Windows Job Objects prevent
+new processes being created and macOS uses the sandbox functionality ({pull}98[#98])
+
 === Bug Fixes
 
 Age seasonal components in proportion to the fraction of values with which they're updated ({pull}88[#88])

include/seccomp/CSystemCallFilter.h

@@ -0,0 +1,43 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the Elastic License;
+ * you may not use this file except in compliance with the Elastic License.
+ */
+#ifndef INCLUDED_ml_seccomp_CSystemCallFilter_h
+#define INCLUDED_ml_seccomp_CSystemCallFilter_h
+
+#include <core/CNonInstantiatable.h>
+
+namespace ml {
+namespace seccomp {
+
+//! \brief
+//! Installs secure computing modes for Linux, macOs and Windows
+//!
+//! DESCRIPTION:\n
+//! ML processes require a subset of system calls to function correctly.
+//! These are create a named pipe, connect to a named pipe, read and write
+//! no other system calls are necessary and should be resticted to prevent
+//! malicious actions.
+//!
+//! IMPLEMENTATION DECISIONS:\n
+//! Implementations are platform specific more details can be found in the
+//! particular .cc files.
+//!
+//! Linux:
+//! Seccomp BPF is used to restrict system calls on kernels since 3.5.
+//!
+//! macOs:
+//! The sandbox facility is used to restict access to system resources.
+//!
+//! Windows:
+//! Job Objects prevent the process spawning another.
+//!
+class CSystemCallFilter : private core::CNonInstantiatable {
+public:
+    static void installSystemCallFilter();
+};
+}
+}
+
+#endif // INCLUDED_ml_seccomp_CSystemCallFilter_h

lib/Makefile

@@ -18,6 +18,8 @@ COMPONENTS= \
             test \
             api \
             config \
+            seccomp \
+
 
 
 include $(CPP_SRC_HOME)/mk/toplevel.mk

lib/seccomp/.objs/.gitignore

@@ -0,0 +1,151 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the Elastic License;
+ * you may not use this file except in compliance with the Elastic License.
+ */
+#include "seccomp/CSystemCallFilter.h"
+
+#include <core/CLogger.h>
+
+#include <linux/audit.h>
+#include <linux/filter.h>
+#include <sys/prctl.h>
+#include <sys/syscall.h>
+
+#include <cerrno>
+#include <cstdint>
+#include <cstring>
+
+namespace ml {
+namespace seccomp {
+
+namespace {
+// The old x32 ABI always has bit 30 set in the sys call numbers.
+// The x64 architecture should fail these calls
+const std::uint32_t UPPER_NR_LIMIT = 0x3FFFFFFF;
+
+// Offset to the nr field in struct seccomp_data
+const std::uint32_t SECCOMP_DATA_NR_OFFSET = 0x00;
+// Offset to the arch field in struct seccomp_data
+const std::uint32_t SECCOMP_DATA_ARCH_OFFSET = 0x04;
+
+// Copied from seccomp.h
+// seccomp.h cannot be included as it was added in Linux kernel 3.17
+// and this must build on older versions.
+// TODO: remove on the minumum build kernel version supports seccomp
+#define SECCOMP_MODE_FILTER 2
+#define SECCOMP_RET_ERRNO 0x00050000U
+#define SECCOMP_RET_ALLOW 0x7fff0000U
+#define SECCOMP_RET_DATA 0x0000ffffU
+
+// Added in Linux 3.5
+#ifndef PR_SET_NO_NEW_PRIVS
+#define PR_SET_NO_NEW_PRIVS 38
+#endif
+
+const struct sock_filter FILTER[] = {
+    // Load architecture from 'seccomp_data' buffer into accumulator
+    BPF_STMT(BPF_LD | BPF_W | BPF_ABS, SECCOMP_DATA_ARCH_OFFSET),
+    // Jump to disallow if architecture is not X86_64
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, AUDIT_ARCH_X86_64, 0, 5),
+    // Load the system call number into accumulator
+    BPF_STMT(BPF_LD | BPF_W | BPF_ABS, SECCOMP_DATA_NR_OFFSET),
+    // Only applies to X86_64 arch. Jump to disallow for calls using the x32 ABI
+    BPF_JUMP(BPF_JMP | BPF_JGT | BPF_K, UPPER_NR_LIMIT, 34, 0),
+    // Allowed sys calls, jump to return allow on match
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_read, 34, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_write, 33, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_writev, 32, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_lseek, 31, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_lstat, 30, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_readlink, 29, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_stat, 28, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_fstat, 27, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_open, 26, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_close, 25, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_connect, 24, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_clone, 23, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_statfs, 22, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_dup2, 21, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_rmdir, 20, 0), // for forecast temp storage
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_getdents, 19, 0), // for forecast temp storage
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_openat, 18, 0), // for forecast temp storage
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_tgkill, 17, 0), // for the crash handler
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_rt_sigaction, 16, 0), // for the crash handler
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_rt_sigreturn, 15, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_futex, 14, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_madvise, 13, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_unlink, 12, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_mknod, 11, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_nanosleep, 10, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_set_robust_list, 9, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_mprotect, 8, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_munmap, 7, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_mmap, 6, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_getuid, 5, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_exit_group, 4, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_access, 3, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_brk, 2, 0),
+    BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_exit, 1, 0),
+    // Disallow call with error code EACCES
+    BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ERRNO | (EACCES & SECCOMP_RET_DATA)),
+    // Allow call
+    BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW)};
+
+bool canUseSeccompBpf() {
+    // This call is expected to fail due to the nullptr argument
+    // but the failure mode informs us if the kernel was configured
+    // with CONFIG_SECCOMP_FILTER
+    // http://man7.org/linux/man-pages/man2/prctl.2.html
+    int result = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, nullptr);
+    int configError = errno;
+    if (result != -1) {
+        LOG_ERROR(<< "prctl set seccomp with null argument should have failed");
+        return false;
+    }
+
+    // If the kernel is not configured with CONFIG_SECCOMP_FILTER
+    // or CONFIG_SECCOMP the error is EINVAL. EFAULT indicates the
+    // seccomp filters are enabled but the 3rd argument (nullptr)
+    // was invalid.
+    return configError == EFAULT;
+}
+}
+
+void CSystemCallFilter::installSystemCallFilter() {
+    if (canUseSeccompBpf()) {
+        LOG_DEBUG(<< "Seccomp BPF filters available");
+
+        // Ensure more permissive privileges cannot be set in future.
+        // This must be set before installing the filter.
+        // PR_SET_NO_NEW_PRIVS was aded in kernel 3.5
+        if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
+            LOG_ERROR(<< "prctl PR_SET_NO_NEW_PRIVS failed: " << std::strerror(errno));
+            return;
+        }
+
+        struct sock_fprog prog = {
+            .len = static_cast<unsigned short>(sizeof(FILTER) / sizeof(FILTER[0])),
+            .filter = const_cast<sock_filter*>(FILTER)};
+
+        // Install the filter.
+        // prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, filter) was introduced
+        // in kernel 3.5. This is functionally equivalent to
+        // seccomp(SECCOMP_SET_MODE_FILTER, 0, filter) which was added in
+        // kernel 3.17. We choose the older more compatible function.
+        // Note this precludes the use of calling seccomp() with the
+        // SECCOMP_FILTER_FLAG_TSYNC which is acceptable if the filter
+        // is installed by the main thread before any other threads are
+        // spawned.
+        if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog)) {
+            LOG_ERROR(<< "Unable to install Seccomp BPF: " << std::strerror(errno));
+        } else {
+            LOG_DEBUG(<< "Seccomp BPF installed");
+        }
+
+    } else {
+        LOG_DEBUG(<< "Seccomp BPF not available");
+    }
+}
+}
+}

lib/seccomp/CSystemCallFilter_Linux.cc

@@ -0,0 +1,92 @@
+/*
+ * Copyright Elasticsearch B.V. and/or licensed to Elasticsearch B.V. under one
+ * or more contributor license agreements. Licensed under the Elastic License;
+ * you may not use this file except in compliance with the Elastic License.
+ */
+#include "seccomp/CSystemCallFilter.h"
+
+#include <core/CLogger.h>
+
+#include <paths.h>
+#include <sandbox.h>
+#include <unistd.h>
+
+#include <cerrno>
+#include <cstring>
+
+namespace ml {
+namespace seccomp {
+
+namespace {
+// The Sandbox rules deny all actions apart from creating fifos,
+// opening files, reading and writing.
+// (allow file-write*) is required for mkfifo and that permission
+// can not be set using the more granular controls.
+const std::string SANDBOX_RULES("\
+    (version 1) \
+    (deny default) \
+    (allow file-read*) \
+    (allow file-read-data) \
+    (allow file-write*) \
+    (allow file-write-data)");
+
+// mkstemps will replace the Xs with random characters
+const std::string FILE_NAME_TEMPLATE("ml.XXXXXX.sb");
+// The length of the suffix '.sb'
+const int FILE_NAME_TEMPLATE_SUFFIX_LEN = 3;
+
+std::string getTempDir() {
+    // Prefer to use the temporary directory set by the Elasticsearch JVM
+    const char* tmpDir(::getenv("TMPDIR"));
+
+    // If TMPDIR is not set use _PATH_VARTMP
+    std::string path((tmpDir == nullptr) ? _PATH_VARTMP : tmpDir);
+    // Make sure path ends with a slash so it's ready to have a file name appended
+    if (path[path.length() - 1] != '/') {
+        path += '/';
+    }
+    return path;
+}
+
+std::string writeTempRulesFile() {
+    std::string profileFilename = getTempDir() + FILE_NAME_TEMPLATE;
+
+    // Create and open a temporary file with a random name
+    // profileFilename is updated with the new filename.
+    int fd = mkstemps(&profileFilename[0], FILE_NAME_TEMPLATE_SUFFIX_LEN);
+    if (fd == -1) {
+        LOG_ERROR(<< "Opening a temporary file with mkstemps failed: "
+                  << std::strerror(errno));
+        return std::string();
+    }
+    write(fd, SANDBOX_RULES.c_str(), SANDBOX_RULES.size());
+    close(fd);
+
+    return profileFilename;
+}
+}
+
+void CSystemCallFilter::installSystemCallFilter() {
+    std::string profileFilename = writeTempRulesFile();
+    if (profileFilename.empty()) {
+        LOG_WARN(<< "Cannot write sandbox rules. macOS sandbox will not be initialized");
+        return;
+    }
+
+    char* errorbuf = nullptr;
+    if (sandbox_init(profileFilename.c_str(), SANDBOX_NAMED, &errorbuf) != 0) {
+        std::string msg("Error initializing macOS sandbox");
+        if (errorbuf != nullptr) {
+            msg += ": ";
+            msg += errorbuf;
+            sandbox_free_error(errorbuf);
+        }
+        LOG_ERROR(<< msg);
+    } else {
+        LOG_DEBUG(<< "macOS sandbox initialized");
+    }
+
+    std::remove(profileFilename.c_str());
+}
+}
+}