Added more descriptive Syscall argument types, and an API for introspecting arguments.

PiperOrigin-RevId: 612904089
Change-Id: Ia0ef7b0559f7eed923981b13fa8224bc891e8c37

sandboxed_api/sandbox2/BUILD.bazel

@@ -136,6 +136,7 @@ cc_library(
     deps = [
         ":util",
         "//sandboxed_api:config",
+        "//sandboxed_api/util:status",
         "@com_google_absl//absl/algorithm:container",
         "@com_google_absl//absl/status",
         "@com_google_absl//absl/status:statusor",
@@ -799,6 +800,8 @@ cc_library(
         "//sandboxed_api/util:file_helpers",
         "//sandboxed_api/util:fileops",
         "//sandboxed_api/util:raw_logging",
+        "//sandboxed_api/util:status",
+        "@com_google_absl//absl/algorithm:container",
         "@com_google_absl//absl/base:core_headers",
         "@com_google_absl//absl/status",
         "@com_google_absl//absl/status:statusor",

sandboxed_api/sandbox2/CMakeLists.txt

@@ -90,6 +90,7 @@ target_link_libraries(sandbox2_syscall
           absl::strings
           sandbox2::util
           sapi::base
+          sapi::status
   PUBLIC absl::log
 )
 
@@ -687,7 +688,8 @@ add_library(sandbox2_util ${SAPI_LIB_TYPE}
 )
 add_library(sandbox2::util ALIAS sandbox2_util)
 target_link_libraries(sandbox2_util
-  PRIVATE absl::core_headers
+  PRIVATE absl::algorithm_container
+          absl::core_headers
           absl::str_format
           absl::strings
           sapi::config
@@ -696,6 +698,7 @@ target_link_libraries(sandbox2_util
           sapi::fileops
           sapi::base
           sapi::raw_logging
+          sapi::status
   PUBLIC absl::status
          absl::statusor
 )

sandboxed_api/sandbox2/syscall.cc

@@ -74,6 +74,10 @@ std::string Syscall::GetName() const {
   return absl::StrFormat("UNKNOWN[%d/0x%x]", nr_, nr_);
 }
 
+std::vector<syscalls::ArgData> Syscall::GetArgumentsData() const {
+  return SyscallTable::get(arch_).GetArgumentsData(nr_, args_.data(), pid_);
+}
+
 std::vector<std::string> Syscall::GetArgumentsDescription() const {
   return SyscallTable::get(arch_).GetArgumentsDescription(nr_, args_.data(),
                                                           pid_);

sandboxed_api/sandbox2/syscall.h

@@ -27,6 +27,7 @@
 #include <vector>
 
 #include "sandboxed_api/config.h"  // IWYU pragma: export
+#include "sandboxed_api/sandbox2/syscall_defs.h"
 
 namespace sandbox2 {
 
@@ -59,7 +60,7 @@ class Syscall {
   uint64_t instruction_pointer() const { return ip_; }
 
   std::string GetName() const;
-
+  std::vector<syscalls::ArgData> GetArgumentsData() const;
   std::vector<std::string> GetArgumentsDescription() const;
   std::string GetDescription() const;
 

sandboxed_api/sandbox2/syscall_defs.h

@@ -4,36 +4,200 @@
 #include <sys/types.h>
 
 #include <array>
+#include <cstddef>
 #include <cstdint>
+#include <optional>
 #include <string>
+#include <type_traits>
 #include <vector>
 
+#include "absl/status/status.h"
 #include "absl/status/statusor.h"
+#include "absl/strings/str_format.h"
 #include "absl/strings/string_view.h"
 #include "absl/types/span.h"
 #include "sandboxed_api/config.h"
-#include "sandboxed_api/sandbox2/syscall.h"
+#include "sandboxed_api/sandbox2/util.h"
+#include "sandboxed_api/util/status_macros.h"
 
 namespace sandbox2 {
 namespace syscalls {
 
 constexpr int kMaxArgs = 6;
 
-// Type of a given syscall argument. Used with argument conversion routines.
 enum ArgType {
-  kGen = 1,
-  kInt,
   kPath,
-  kHex,
-  kOct,
-  kSocketCall,
-  kSocketCallPtr,
-  kSignal,
   kString,
+  kGen,
+  kStruct,
+  kPtr,
+  kArray,
+  kInt,
+  kFlags,
+  kResource,
+  kPid,
+  kSignal,
+  kIpcResource,
+  kSharedAddress,
+  kGid,
+  kUid,
+  // These kLenN types indicate that the argument is a length, and which
+  // other argument it applies to. I.e., if parameter 3 is type kPollFdArray,
+  // and parameter 4 has the length, then parameter 4 is of type kLen3.
+  kLen0,
+  kLen1,
+  kLen2,
+  kLen3,
+  kLen4,
+  kLen5,
   kAddressFamily,
+  kGidArray,
+  kPollFdArray,
   kSockaddr,
-  kSockmsghdr,
-  kCloneFlag,
+};
+
+constexpr absl::string_view ArgTypeToString(ArgType type) {
+  switch (type) {
+    case kPath:
+      return "path";
+    case kString:
+      return "string";
+    case kGen:
+      return "generic type";
+    case kStruct:
+      return "struct";
+    case kPtr:
+      return "pointer";
+    case kArray:
+      return "array";
+    case kInt:
+      return "int";
+    case kFlags:
+      return "flags";
+    case kResource:
+      return "resource";
+    case kPid:
+      return "pid";
+    case kSignal:
+      return "signal";
+    case kIpcResource:
+      return "ipc resource";
+    case kSharedAddress:
+      return "shared address";
+    case kGid:
+      return "gid";
+    case kUid:
+      return "uid";
+    case kLen0:
+      return "length of parameter 0";
+    case kLen1:
+      return "length of parameter 1";
+    case kLen2:
+      return "length of parameter 2";
+    case kLen3:
+      return "length of parameter 3";
+    case kLen4:
+      return "length of parameter 4";
+    case kLen5:
+      return "length of parameter 5";
+    case kAddressFamily:
+      return "address family";
+    case kGidArray:
+      return "gid array";
+    case kPollFdArray:
+      return "poll fd array";
+    case kSockaddr:
+      return "sockaddr struct";
+    default:
+      return "invalid type";
+  }
+}
+
+class ArgData {
+ public:
+  template <typename T>
+  struct StructArray {
+    std::vector<T> array;
+    bool truncated;
+  };
+
+  ArgData(syscalls::ArgType type, pid_t pid, uint64_t value,
+          std::optional<uint64_t> length = std::nullopt)
+      : type_(type), pid_(pid), value_(value), length_(length) {}
+
+  ArgType type() const { return type_; }
+  pid_t pid() const { return pid_; }
+  uint64_t value() const { return value_; }
+  std::optional<uint64_t> length() const { return length_; }
+
+  std::string GetDescription() const;
+
+  absl::StatusOr<std::string> ReadAsString() const {
+    return util::ReadCPathFromPid(pid_, value_);
+  }
+
+  template <typename T>
+  absl::StatusOr<T> ReadAsStruct() const {
+    if (length_.has_value() && *length_ < sizeof(T)) {
+      return absl::InternalError(absl::StrFormat(
+          "specified length [%llu] is not enough for to sizeof(%s) == %llu",
+          *length_, typeid(T).name(), sizeof(T)));
+    }
+    SAPI_ASSIGN_OR_RETURN(std::vector<uint8_t> b,
+                          util::ReadBytesFromPid(pid_, value_, sizeof(T)));
+    return BytesToStruct<T>(b);
+  }
+
+  template <typename T>
+  absl::StatusOr<StructArray<T>> ReadAsStructArray() const {
+    static uint64_t kMaxAllowedBytes = 1 << 20;  // 1MB
+
+    if (!length_.has_value()) {
+      return absl::InternalError("length is not set");
+    }
+
+    bool truncated = false;
+    uint64_t length = *length_ * sizeof(T);
+    if (length > kMaxAllowedBytes) {
+      truncated = true;
+      length = (kMaxAllowedBytes / sizeof(T)) * sizeof(T);
+    }
+
+    SAPI_ASSIGN_OR_RETURN(std::vector<uint8_t> b,
+                          util::ReadBytesFromPid(pid_, value_, length));
+    absl::Span<const uint8_t> bytes = absl::MakeSpan(b);
+    if (bytes.size() < length) {
+      return absl::InternalError("could not read full struct array");
+    }
+    std::vector<T> ret;
+    for (size_t i = 0; i < bytes.size(); i += sizeof(T)) {
+      SAPI_ASSIGN_OR_RETURN(T t, BytesToStruct<T>(bytes.subspan(i, sizeof(T))));
+      ret.push_back(t);
+    }
+
+    return StructArray<T>{std::move(ret), truncated};
+  }
+
+ private:
+  template <typename T>
+  static absl::StatusOr<T> BytesToStruct(absl::Span<const uint8_t> bytes) {
+    static_assert(std::is_pod<T>(), "Can only cast bytes to POD structs");
+    if (bytes.size() < sizeof(T)) {
+      return absl::InternalError(absl::StrFormat(
+          "bytes size [%llu] is not equal to sizeof(%s) == %llu", bytes.size(),
+          typeid(T).name(), sizeof(T)));
+    }
+    T t;
+    memcpy(&t, bytes.data(), sizeof(T));
+    return t;
+  }
+
+  absl::StatusOr<std::string> GetDescriptionImpl() const;
+
+  syscalls::ArgType type_;
+  pid_t pid_;
+  uint64_t value_;
+  std::optional<uint64_t> length_;
 };
 
 }  // namespace syscalls
@@ -50,10 +214,6 @@ class SyscallTable {
       return num_args;
     }
 
-    static std::string GetArgumentDescription(uint64_t value,
-                                              syscalls::ArgType type,
-                                              pid_t pid);
-
     static constexpr bool BySyscallNr(const SyscallTable::Entry& a,
                                       const SyscallTable::Entry& b) {
       return a.nr < b.nr;
@@ -72,6 +232,10 @@ class SyscallTable {
 
   absl::string_view GetName(int syscall) const;
 
+  std::vector<syscalls::ArgData> GetArgumentsData(int syscall,
+                                                  const uint64_t values[],
+                                                  pid_t pid) const;
+
   std::vector<std::string> GetArgumentsDescription(int syscall,
                                                    const uint64_t values[],
                                                    pid_t pid) const;

sandboxed_api/sandbox2/util.cc

@@ -14,15 +14,18 @@
 
 #include "sandboxed_api/sandbox2/util.h"
 
+#include <linux/limits.h>
 #include <sched.h>
 #include <spawn.h>
 #include <sys/ptrace.h>
 #include <sys/resource.h>
+#include <sys/socket.h>
 #include <sys/uio.h>
 #include <sys/wait.h>
 #include <syscall.h>
 #include <unistd.h>
 
+#include <algorithm>
 #include <cerrno>
 #include <csetjmp>
 #include <cstddef>
@@ -33,6 +36,7 @@
 #include <utility>
 #include <vector>
 
+#include "absl/algorithm/container.h"
 #include "absl/base/attributes.h"
 #include "absl/base/macros.h"
 #include "absl/base/optimization.h"
@@ -51,7 +55,7 @@
 #include "sandboxed_api/util/fileops.h"
 #include "sandboxed_api/util/path.h"
 #include "sandboxed_api/util/raw_logging.h"
-
+#include "sandboxed_api/util/status_macros.h"
 namespace sandbox2::util {
 
 namespace file = ::sapi::file;
@@ -137,6 +141,19 @@ std::string GetProgName(pid_t pid) {
   return file_util::fileops::Basename(file_util::fileops::ReadLink(fname));
 }
 
+absl::StatusOr<std::string> GetResolvedFdLink(pid_t pid, uint32_t fd) {
+  // The proc/PID/fd directory contains links for all of that process' file
+  // descriptors. They'll show up as more informative strings (paths, sockets).
+  std::string fd_path = absl::StrFormat("/proc/%u/fd/%u", pid, fd);
+  std::string result(PATH_MAX, '\0');
+  ssize_t size = readlink(fd_path.c_str(), &result[0], PATH_MAX);
+  if (size < 0) {
+    return absl::ErrnoToStatus(size, "failed to read link");
+  }
+  result.resize(size);
+  return result;
+}
+
 std::string GetCmdLine(pid_t pid) {
   std::string fname = file::JoinPath("/proc", absl::StrCat(pid), "cmdline");
   std::string cmdline;
@@ -326,6 +343,29 @@ std::string GetSignalName(int signo) {
   return absl::StrFormat("%s [%d]", kSignalNames[signo], signo);
 }
 
+std::string GetAddressFamily(int addr_family) {
+  // Taken from definitions in `socket.h`. Each family's index in the array is
+  // also its integer value.
+  constexpr absl::string_view kAddressFamilies[] = {
+      "AF_UNSPEC",     "AF_UNIX",      "AF_INET",     "AF_AX25",
+      "AF_IPX",        "AF_APPLETALK", "AF_NETROM",   "AF_BRIDGE",
+      "AF_ATMPVC",     "AF_X25",       "AF_INET6",    "AF_ROSE",
+      "AF_DECnet",     "AF_NETBEUI",   "AF_SECURITY", "AF_KEY",
+      "AF_NETLINK",    "AF_PACKET",    "AF_ASH",      "AF_ECONET",
+      "AF_ATMSVC",     "AF_RDS",       "AF_SNA",      "AF_IRDA",
+      "AF_PPPOX",      "AF_WANPIPE",   "AF_LLC",      "AF_IB",
+      "AF_MPLS",       "AF_CAN",       "AF_TIPC",     "AF_BLUETOOTH",
+      "AF_IUCV",       "AF_RXRPC",     "AF_ISDN",     "AF_PHONET",
+      "AF_IEEE802154", "AF_CAIF",      "AF_ALG",      "AF_NFC",
+      "AF_VSOCK",      "AF_KCM",       "AF_QIPCRTR",  "AF_SMC",
+      "AF_XDP",        "AF_MCTP"};
+
+  if (addr_family < 0 && addr_family >= ABSL_ARRAYSIZE(kAddressFamilies)) {
+    return absl::StrFormat("UNKNOWN_ADDRESS_FAMILY [%d]", addr_family);
+  }
+  return std::string(kAddressFamilies[addr_family]);
+}
+
 std::string GetRlimitName(int resource) {
   switch (resource) {
     case RLIMIT_AS:
@@ -370,45 +410,53 @@ std::string GetPtraceEventName(int event) {
   }
 }
 
-absl::StatusOr<std::string> ReadCPathFromPid(pid_t pid, uintptr_t ptr) {
-  std::string path(PATH_MAX, '\0');
-  iovec local_iov[] = {{&path[0], path.size()}};
-
+absl::StatusOr<std::vector<uint8_t>> ReadBytesFromPid(pid_t pid, uintptr_t ptr,
+                                                      uint64_t size) {
   static const uintptr_t page_size = getpagesize();
-  static const uintptr_t page_mask = ~(page_size - 1);
-  // See 'man process_vm_readv' for details on how to read NUL-terminated
-  // strings with this syscall.
-  size_t len1 = ((ptr + page_size) & page_mask) - ptr;
-  len1 = (len1 > path.size()) ? path.size() : len1;
-  size_t len2 = (path.size() <= len1) ? 0UL : path.size() - len1;
-  // Second iov is wrapping around to NULL ptr.
-  if ((ptr + len1) < ptr) {
-    len2 = 0UL;
+  static const uintptr_t page_mask = page_size - 1;
+
+  // Input sanity checks.
+  if (size == 0) {
+    return std::vector<uint8_t>();
   }
 
-  iovec remote_iov[] = {
-      {reinterpret_cast<void*>(ptr), len1},
-      {reinterpret_cast<void*>(ptr + len1), len2},
-  };
+  // Allocate enough bytes to hold the entire size.
+  std::vector<uint8_t> bytes(size, 0);
+  iovec local_iov[] = {{bytes.data(), bytes.size()}};
+  // Stores all the necessary iovecs to move memory.
+  std::vector<iovec> remote_iov;
+  // Each iovec should be contained to a single page.
+  size_t consumed = 0;
+  while (consumed < size) {
+    // Read till the end of the page, at most the remaining number of bytes.
+    size_t chunk_size =
+        std::min(size - consumed, page_size - ((ptr + consumed) & page_mask));
+    remote_iov.push_back({reinterpret_cast<void*>(ptr + consumed), chunk_size});
+    consumed += chunk_size;
+  }
 
-  SAPI_RAW_VLOG(4, "ReadCPathFromPid (iovec): len1: %zu, len2: %zu", len1,
-                len2);
-  if (process_vm_readv(pid, local_iov, ABSL_ARRAYSIZE(local_iov), remote_iov,
-                       ABSL_ARRAYSIZE(remote_iov), 0) < 0) {
+  ssize_t result = process_vm_readv(pid, local_iov, ABSL_ARRAYSIZE(local_iov),
+                                    remote_iov.data(), remote_iov.size(), 0);
+  if (result < 0) {
     return absl::ErrnoToStatus(
         errno,
         absl::StrFormat("process_vm_readv() failed for PID: %d at address: %#x",
-                        pid, reinterpret_cast<uintptr_t>(ptr)));
+                        pid, ptr));
   }
+  // Ensure only successfully read bytes are returned.
+  bytes.resize(result);
+  return bytes;
+}
 
-  // Check for whether there's a NUL byte in the buffer. If not, it's an
-  // incorrect path (or >PATH_MAX).
-  auto pos = path.find('\0');
-  if (pos == std::string::npos) {
-    return absl::FailedPreconditionError(absl::StrCat(
-        "No NUL-byte inside the C string '", absl::CHexEscape(path), "'"));
+absl::StatusOr<std::string> ReadCPathFromPid(pid_t pid, uintptr_t ptr) {
+  SAPI_ASSIGN_OR_RETURN(std::vector<uint8_t> bytes,
+                        ReadBytesFromPid(pid, ptr, PATH_MAX));
+  auto null_pos = absl::c_find(bytes, '\0');
+  std::string path(bytes.begin(), null_pos);
+  if (null_pos == bytes.end()) {
+    return absl::FailedPreconditionError(
+        absl::StrFormat("path '%s' is too long", absl::CHexEscape(path)));
   }
-  path.resize(pos);
   return path;
 }
 

sandboxed_api/sandbox2/util.h

@@ -63,6 +63,9 @@ inline void CharPtrArrToVecString(char* const* arr,
 // Returns the program name (via /proc/self/comm) for a given PID.
 std::string GetProgName(pid_t pid);
 
+// Given a resource descriptor FD and a PID, returns link of /proc/PID/fds/FD.
+absl::StatusOr<std::string> GetResolvedFdLink(pid_t pid, uint32_t fd);
+
 // Returns the command line (via /proc/self/cmdline) for a given PID. The
 // argument separators '\0' are converted to spaces.
 std::string GetCmdLine(pid_t pid);
@@ -95,12 +98,19 @@ absl::StatusOr<int> Communicate(const std::vector<std::string>& argv,
 // Returns signal description.
 std::string GetSignalName(int signo);
 
+// Returns the socket address family as a string ("AF_INET", ...)
+std::string GetAddressFamily(int addr_family);
+
 // Returns rlimit resource name
 std::string GetRlimitName(int resource);
 
 // Returns ptrace event name
 std::string GetPtraceEventName(int event);
 
+// Reads `size` bytes from the given `ptr` address, or returns an error.
+absl::StatusOr<std::vector<uint8_t>> ReadBytesFromPid(pid_t pid, uintptr_t ptr,
+                                                      uint64_t size);
+
 // Reads a path string (NUL-terminated, shorter than PATH_MAX) from another
 // process memory
 absl::StatusOr<std::string> ReadCPathFromPid(pid_t pid, uintptr_t ptr);