runtime: add PoC demos

CMakeLists.txt

@@ -12,3 +12,4 @@ add_subdirectory(examples)
 add_subdirectory(rewriter)
 add_subdirectory(partition-alloc)
 add_subdirectory(pad-tls)
+add_subdirectory(runtime)

runtime/CMakeLists.txt

@@ -0,0 +1,25 @@
+cmake_minimum_required(VERSION 3.12)
+project(IA2Runtime)
+
+add_executable(read-pkru
+  read_pkru_demo.c
+  get_inferior_pkru.c
+)
+
+add_executable(track-memory-map
+  memory_map.c
+  track_memory_map_demo.c
+  track_memory_map.c
+  mmap_event.c
+  get_inferior_pkru.c
+)
+
+add_executable(seccomp-filter
+  seccomp_filter.c
+)
+
+add_executable(landlock
+  landlock.c
+  strv.c
+  forbid_paths.c
+)

runtime/forbid_paths.c

@@ -0,0 +1,164 @@
+#define _GNU_SOURCE
+#include <ftw.h>
+#include <linux/landlock.h>
+#include <linux/prctl.h>
+
+#include <errno.h>
+#include <fcntl.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "forbid_paths.h"
+#include "landlock.h"
+#include "strv.h"
+
+static int allow_path(const char *path, const int ruleset_fd, bool shallow) {
+  struct landlock_path_beneath_attr path_beneath = {
+      .parent_fd = -1,
+  };
+
+  struct stat statbuf;
+
+  path_beneath.parent_fd = open(path, O_PATH | O_CLOEXEC);
+  if (path_beneath.parent_fd < 0) {
+    fprintf(stderr, "Failed to open \"%s\": %s\n", path, strerror(errno));
+    return -1;
+  }
+  if (fstat(path_beneath.parent_fd, &statbuf)) {
+    close(path_beneath.parent_fd);
+    return -1;
+  }
+  path_beneath.allowed_access =
+      ACCESS_FS_ROUGHLY_READ | ACCESS_FS_ROUGHLY_WRITE;
+
+  /* limit non-directory files to access flags relevant for regular files */
+  if (!S_ISDIR(statbuf.st_mode)) {
+    path_beneath.allowed_access &= ACCESS_FILE;
+  } else {
+    /* if we only shallowly allow this directory, only allow READ_DIR */
+    if (shallow)
+      path_beneath.allowed_access = LANDLOCK_ACCESS_FS_READ_DIR;
+  }
+
+  /* add rule */
+  if (landlock_add_rule(ruleset_fd, LANDLOCK_RULE_PATH_BENEATH, &path_beneath,
+                        0)) {
+    fprintf(stderr, "Failed to update the ruleset with \"%s\": %s\n", path,
+            strerror(errno));
+    close(path_beneath.parent_fd);
+    return -1;
+  }
+  close(path_beneath.parent_fd);
+
+  return 0;
+}
+
+/* does path have the given prefix as a prefix of its path segments? */
+static bool path_has_segment_prefix(const char *path, const char *prefix) {
+  if (!strcmp(prefix, "/")) {
+    return true;
+  }
+
+  while (*path == *prefix && *path && *prefix) {
+    path++;
+    prefix++;
+  }
+
+  if (*prefix == '\0' && (*path == '/' || *path == '\0')) {
+    return true;
+  }
+  return false;
+}
+
+/*
+Forbid access to the given paths, a null-terminated string vector.
+
+Because landlock is an allowlist-based system, we need to allow everything other
+than the specified paths. This is done by walking the directory tree downward
+from / toward each path. At each level, directories that do not (transitively)
+contain any forbidden files are recursively allowed, and directories that do are
+shallowly allowed.
+
+When the forbidden path is reached, its siblings are allowed by the same process
+(which thereby avoids allowing other forbidden files).
+*/
+
+static struct forbid_ctx {
+  const char **forbidden_paths;
+  int ruleset_fd;
+  int error;
+} ctx;
+
+/* At each level, directories that do not (transitively) contain any forbidden
+files are recursively allowed, and directories that do are shallowly allowed. */
+static int forbid_deep_or_shallow(const char *fpath, const struct stat *sb,
+                                  int typeflag, struct FTW *ftwbuf) {
+  const char *forbidden_path;
+
+  bool contains_forbidden_path = false;
+  bool is_forbidden_path = false;
+  /* is any forbidden path a descendant of this dir? */
+  for (int i = 0; (forbidden_path = ctx.forbidden_paths[i]); i++) {
+    /* is forbidden_path a descendant of this dir? */
+    bool prefix_matches = path_has_segment_prefix(forbidden_path, fpath);
+    if (prefix_matches) {
+      contains_forbidden_path = true;
+    }
+    is_forbidden_path |= !strcmp(forbidden_path, fpath);
+    if (is_forbidden_path) {
+      break;
+    }
+  }
+
+  /* if this is a forbidden path, do not allow it or children, just move on */
+  if (is_forbidden_path) {
+    return FTW_SKIP_SUBTREE;
+  }
+
+  /* do not allow or continue beneath symbolic links */
+  if (typeflag == FTW_SL) {
+    return FTW_SKIP_SUBTREE;
+  }
+
+  /* if contains forbidden path, allow shallowly and process children */
+  if (contains_forbidden_path) {
+    int ret = allow_path(fpath, ctx.ruleset_fd, true);
+    if (ret < 0) {
+      ctx.error = ret;
+      return FTW_STOP;
+    }
+
+    return FTW_CONTINUE;
+  }
+
+  /* allow whole dir or file */
+  int ret = allow_path(fpath, ctx.ruleset_fd, false);
+  if (ret < 0) {
+    ctx.error = ret;
+    return FTW_STOP;
+  }
+
+  /* do not inspect individual children if this is an allowed dir */
+  if (typeflag == FTW_D || typeflag == FTW_DNR) {
+    return FTW_SKIP_SUBTREE;
+  } else {
+    return FTW_CONTINUE;
+  }
+}
+
+/* one-argument wrapper for realpath */
+static char *realpath_alloc(const char *path) { return realpath(path, NULL); }
+
+int forbid_paths(const char **paths, const int ruleset_fd) {
+  /* normalize away symbolic links in the forbidden paths. if multiple routes to
+   * a file were allowed to exist (as permitted by links), we would allow the
+   * one not mentioned by name as a forbidden path, contrary to our intent. */
+  char **real_paths = strvmap(paths, realpath_alloc);
+  ctx.forbidden_paths = (const char **)real_paths;
+  ctx.ruleset_fd = ruleset_fd;
+  ctx.error = 0;
+  nftw("/", forbid_deep_or_shallow, 512, FTW_PHYS | FTW_ACTIONRETVAL);
+  strvfree(real_paths);
+  return ctx.error;
+}

runtime/forbid_paths.h

@@ -0,0 +1,3 @@
+#pragma once
+
+int forbid_paths(const char **paths, const int ruleset_fd);

runtime/get_inferior_pkru.c

@@ -0,0 +1,42 @@
+#include <linux/elf.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ptrace.h>
+#include <sys/uio.h>
+
+#include "get_inferior_pkru.h"
+
+/* this is largely copped from gdb and pared down to just what we need. it would
+ * be much more complex if we had to deal with the compacted xsave area. */
+
+#define X86_XSTATE_PKRU_SIZE 2696
+#define X86_XSTATE_MAX_SIZE 2696
+
+/* offset to the location of the PKRU register data structure used by the
+ * "xsave" instruction */
+static int xsave_pkeys_offset =
+    2688 + 0 * 8; /* %pkru (64 bits in XSTATE, 32-bit actually used by
+                     instructions and applications). */
+
+bool get_inferior_pkru(pid_t pid, uint32_t *pkru_out) {
+  char xstateregs[X86_XSTATE_MAX_SIZE];
+  struct iovec iov;
+
+  /* Pre-4.14 kernels have a bug (fixed by commit 0852b374173b
+     "x86/fpu: Add FPU state copying quirk to handle XRSTOR failure on
+     Intel Skylake CPUs") that sometimes causes the mxcsr location in
+     xstateregs not to be copied by PTRACE_GETREGSET.  Make sure that
+     the location is at least initialized with a defined value.  */
+  memset(xstateregs, 0, sizeof(xstateregs));
+  iov.iov_base = xstateregs;
+  iov.iov_len = sizeof(xstateregs);
+  if (ptrace(PTRACE_GETREGSET, pid, (unsigned int)NT_X86_XSTATE, (long)&iov) <
+      0) {
+    perror("could not read xstate registers");
+    return false;
+  }
+
+  memcpy(pkru_out, &xstateregs[xsave_pkeys_offset], sizeof(*pkru_out));
+  return true;
+}

runtime/get_inferior_pkru.h

@@ -0,0 +1,5 @@
+#include <stdbool.h>
+#include <stdint.h>
+#include <sys/types.h>
+
+bool get_inferior_pkru(pid_t pid, uint32_t *pkru_out);

runtime/landlock.c

@@ -0,0 +1,123 @@
+#include "landlock.h"
+#include "forbid_paths.h"
+#include "strv.h"
+
+#include <errno.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+int main(const int argc, char *const argv[], char *const *const envp) {
+  const char *cmd_path;
+  char *const *cmd_argv;
+  int ruleset_fd, abi_ver;
+  __u64 access_fs_rw = ACCESS_FS_ROUGHLY_READ | ACCESS_FS_ROUGHLY_WRITE;
+  struct landlock_ruleset_attr ruleset_attr = {
+      .handled_access_fs = access_fs_rw,
+  };
+
+  if (argc < 2) {
+    fprintf(stderr, "usage: DENY_PATH=... %s <command>\n\n", basename(argv[0]));
+    fprintf(stderr, "built against landlock ABI version <= %d\n",
+            LANDLOCK_ABI_LAST);
+    return 1;
+  }
+
+  abi_ver = landlock_create_ruleset(NULL, 0, LANDLOCK_CREATE_RULESET_VERSION);
+  if (abi_ver < 0) {
+    const int err = errno;
+
+    perror("Failed to check Landlock compatibility");
+    switch (err) {
+    case ENOSYS:
+      fprintf(stderr,
+              "Hint: Landlock is not supported by the current kernel. "
+              "To support it, build the kernel with "
+              "CONFIG_SECURITY_LANDLOCK=y and prepend "
+              "\"landlock,\" to the content of CONFIG_LSM.\n");
+      break;
+    case EOPNOTSUPP:
+      fprintf(stderr,
+              "Hint: Landlock is currently disabled. "
+              "It can be enabled in the kernel configuration by "
+              "prepending \"landlock,\" to the content of CONFIG_LSM, "
+              "or at boot time by setting the same content to the "
+              "\"lsm\" kernel parameter.\n");
+      break;
+    }
+    return 1;
+  }
+
+  switch (abi_ver) {
+  case 1:
+    /*
+     * Removes LANDLOCK_ACCESS_FS_REFER for ABI < 2
+     *
+     * Note: The "refer" operations (file renaming and linking
+     * across different directories) are always forbidden when using
+     * Landlock with ABI 1.
+     *
+     * If only ABI 1 is available, this sandboxer knowingly forbids
+     * refer operations.
+     *
+     * If a program *needs* to do refer operations after enabling
+     * Landlock, it can not use Landlock at ABI level 1.  To be
+     * compatible with different kernel versions, such programs
+     * should then fall back to not restrict themselves at all if
+     * the running kernel only supports ABI 1.
+     */
+    ruleset_attr.handled_access_fs &= ~LANDLOCK_ACCESS_FS_REFER_OR_0;
+    __attribute__((fallthrough));
+  case 2:
+    /* Removes LANDLOCK_ACCESS_FS_TRUNCATE for ABI < 3 */
+    ruleset_attr.handled_access_fs &= ~LANDLOCK_ACCESS_FS_TRUNCATE_OR_0;
+
+    fprintf(stderr,
+            "Hint: You should update the running kernel "
+            "to leverage Landlock features "
+            "provided by ABI version %d (instead of %d).\n",
+            LANDLOCK_ABI_LAST, abi_ver);
+    __attribute__((fallthrough));
+  case LANDLOCK_ABI_LAST:
+    break;
+  default:
+    fprintf(
+        stderr,
+        "rebuild sandboxer to use features from ABI version %d instead of %d\n",
+        abi_ver, LANDLOCK_ABI_LAST);
+  }
+  access_fs_rw &= ruleset_attr.handled_access_fs;
+
+  ruleset_fd = landlock_create_ruleset(&ruleset_attr, sizeof(ruleset_attr), 0);
+  if (ruleset_fd < 0) {
+    perror("Failed to create ruleset");
+    return 1;
+  }
+  const char *path = getenv("DENY_PATH");
+  const char *paths[] = {path, NULL};
+  if (forbid_paths(paths, ruleset_fd) < 0) {
+    fprintf(stderr, "Failed to set up path allowlist\n");
+    return 1;
+  }
+
+  if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
+    perror("Failed to restrict privileges");
+    goto err_close_ruleset;
+  }
+  if (landlock_restrict_self(ruleset_fd, 0)) {
+    perror("Failed to enforce ruleset");
+    goto err_close_ruleset;
+  }
+  close(ruleset_fd);
+
+  cmd_path = argv[1];
+  cmd_argv = argv + 1;
+  execvpe(cmd_path, cmd_argv, envp);
+  perror("execvpe");
+  return 1;
+
+err_close_ruleset:
+  close(ruleset_fd);
+  return 1;
+}