blob: b4d65532792a4b754ca773a7e74dd794ed62abff [file] [log] [blame]
/*
* Copyright (C) 2011 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "exec_utils.h"
#include <poll.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <unistd.h>
#ifdef __BIONIC__
#include <sys/pidfd.h>
#endif
#include <chrono>
#include <climits>
#include <condition_variable>
#include <cstdint>
#include <cstring>
#include <ctime>
#include <mutex>
#include <optional>
#include <string>
#include <string_view>
#include <thread>
#include <vector>
#include "android-base/file.h"
#include "android-base/parseint.h"
#include "android-base/scopeguard.h"
#include "android-base/stringprintf.h"
#include "android-base/strings.h"
#include "android-base/unique_fd.h"
#include "base/macros.h"
#include "base/utils.h"
#include "runtime.h"
namespace art {
namespace {
using ::android::base::ParseInt;
using ::android::base::ReadFileToString;
using ::android::base::StringPrintf;
using ::android::base::unique_fd;
std::string ToCommandLine(const std::vector<std::string>& args) {
return android::base::Join(args, ' ');
}
// Fork and execute a command specified in a subprocess.
// If there is a runtime (Runtime::Current != nullptr) then the subprocess is created with the
// same environment that existed when the runtime was started.
// Returns the process id of the child process on success, -1 otherwise.
pid_t ExecWithoutWait(const std::vector<std::string>& arg_vector, std::string* error_msg) {
// Convert the args to char pointers.
const char* program = arg_vector[0].c_str();
std::vector<char*> args;
args.reserve(arg_vector.size() + 1);
for (const auto& arg : arg_vector) {
args.push_back(const_cast<char*>(arg.c_str()));
}
args.push_back(nullptr);
// fork and exec
pid_t pid = fork();
if (pid == 0) {
// no allocation allowed between fork and exec
// change process groups, so we don't get reaped by ProcessManager
setpgid(0, 0);
// (b/30160149): protect subprocesses from modifications to LD_LIBRARY_PATH, etc.
// Use the snapshot of the environment from the time the runtime was created.
char** envp = (Runtime::Current() == nullptr) ? nullptr : Runtime::Current()->GetEnvSnapshot();
if (envp == nullptr) {
execv(program, &args[0]);
} else {
execve(program, &args[0], envp);
}
// This should be regarded as a crash rather than a normal return.
PLOG(FATAL) << "Failed to execute (" << ToCommandLine(arg_vector) << ")";
UNREACHABLE();
} else if (pid == -1) {
*error_msg = StringPrintf("Failed to execute (%s) because fork failed: %s",
ToCommandLine(arg_vector).c_str(),
strerror(errno));
return -1;
} else {
return pid;
}
}
ExecResult WaitChild(pid_t pid,
const std::vector<std::string>& arg_vector,
bool no_wait,
std::string* error_msg) {
siginfo_t info;
// WNOWAIT leaves the child in a waitable state. The call is still blocking.
int options = WEXITED | (no_wait ? WNOWAIT : 0);
if (TEMP_FAILURE_RETRY(waitid(P_PID, pid, &info, options)) != 0) {
*error_msg = StringPrintf("waitid failed for (%s) pid %d: %s",
ToCommandLine(arg_vector).c_str(),
pid,
strerror(errno));
return {.status = ExecResult::kUnknown};
}
if (info.si_pid != pid) {
*error_msg = StringPrintf("waitid failed for (%s): wanted pid %d, got %d: %s",
ToCommandLine(arg_vector).c_str(),
pid,
info.si_pid,
strerror(errno));
return {.status = ExecResult::kUnknown};
}
if (info.si_code != CLD_EXITED) {
*error_msg =
StringPrintf("Failed to execute (%s) because the child process is terminated by signal %d",
ToCommandLine(arg_vector).c_str(),
info.si_status);
return {.status = ExecResult::kSignaled, .signal = info.si_status};
}
return {.status = ExecResult::kExited, .exit_code = info.si_status};
}
// A fallback implementation of `WaitChildWithTimeout` that creates a thread to wait instead of
// relying on `pidfd_open`.
ExecResult WaitChildWithTimeoutFallback(pid_t pid,
const std::vector<std::string>& arg_vector,
int timeout_ms,
std::string* error_msg) {
bool child_exited = false;
bool timed_out = false;
std::condition_variable cv;
std::mutex m;
std::thread wait_thread([&]() {
std::unique_lock<std::mutex> lock(m);
if (!cv.wait_for(lock, std::chrono::milliseconds(timeout_ms), [&] { return child_exited; })) {
timed_out = true;
kill(pid, SIGKILL);
}
});
ExecResult result = WaitChild(pid, arg_vector, /*no_wait=*/true, error_msg);
{
std::unique_lock<std::mutex> lock(m);
child_exited = true;
}
cv.notify_all();
wait_thread.join();
// The timeout error should have a higher priority than any other error.
if (timed_out) {
*error_msg =
StringPrintf("Failed to execute (%s) because the child process timed out after %dms",
ToCommandLine(arg_vector).c_str(),
timeout_ms);
return ExecResult{.status = ExecResult::kTimedOut};
}
return result;
}
// Waits for the child process to finish and leaves the child in a waitable state.
ExecResult WaitChildWithTimeout(pid_t pid,
unique_fd pidfd,
const std::vector<std::string>& arg_vector,
int timeout_ms,
std::string* error_msg) {
auto cleanup = android::base::make_scope_guard([&]() {
kill(pid, SIGKILL);
std::string ignored_error_msg;
WaitChild(pid, arg_vector, /*no_wait=*/true, &ignored_error_msg);
});
struct pollfd pfd;
pfd.fd = pidfd.get();
pfd.events = POLLIN;
int poll_ret = TEMP_FAILURE_RETRY(poll(&pfd, /*nfds=*/1, timeout_ms));
pidfd.reset();
if (poll_ret < 0) {
*error_msg = StringPrintf("poll failed for pid %d: %s", pid, strerror(errno));
return {.status = ExecResult::kUnknown};
}
if (poll_ret == 0) {
*error_msg =
StringPrintf("Failed to execute (%s) because the child process timed out after %dms",
ToCommandLine(arg_vector).c_str(),
timeout_ms);
return {.status = ExecResult::kTimedOut};
}
cleanup.Disable();
return WaitChild(pid, arg_vector, /*no_wait=*/true, error_msg);
}
bool ParseProcStat(const std::string& stat_content,
int64_t uptime_ms,
int64_t ticks_per_sec,
/*out*/ ProcessStat* stat) {
size_t pos = stat_content.rfind(") ");
if (pos == std::string::npos) {
return false;
}
std::vector<std::string> stat_fields;
// Skip the first two fields. The second field is the parenthesized process filename, which can
// contain anything, including spaces.
Split(std::string_view(stat_content).substr(pos + 2), ' ', &stat_fields);
constexpr int kSkippedFields = 2;
int64_t utime, stime, cutime, cstime, starttime;
if (stat_fields.size() < 22 - kSkippedFields ||
!ParseInt(stat_fields[13 - kSkippedFields], &utime) ||
!ParseInt(stat_fields[14 - kSkippedFields], &stime) ||
!ParseInt(stat_fields[15 - kSkippedFields], &cutime) ||
!ParseInt(stat_fields[16 - kSkippedFields], &cstime) ||
!ParseInt(stat_fields[21 - kSkippedFields], &starttime)) {
return false;
}
if (starttime == 0) {
// The start time is the time the process started after system boot, so it's not supposed to be
// zero unless the process is `init`.
return false;
}
stat->cpu_time_ms = (utime + stime + cutime + cstime) * 1000 / ticks_per_sec;
stat->wall_time_ms = uptime_ms - starttime * 1000 / ticks_per_sec;
return true;
}
} // namespace
int ExecUtils::ExecAndReturnCode(const std::vector<std::string>& arg_vector,
std::string* error_msg) const {
return ExecAndReturnResult(arg_vector, /*timeout_sec=*/-1, error_msg).exit_code;
}
ExecResult ExecUtils::ExecAndReturnResult(const std::vector<std::string>& arg_vector,
int timeout_sec,
std::string* error_msg) const {
return ExecAndReturnResult(arg_vector, timeout_sec, ExecCallbacks(), /*stat=*/nullptr, error_msg);
}
ExecResult ExecUtils::ExecAndReturnResult(const std::vector<std::string>& arg_vector,
int timeout_sec,
const ExecCallbacks& callbacks,
/*out*/ ProcessStat* stat,
/*out*/ std::string* error_msg) const {
if (timeout_sec > INT_MAX / 1000) {
*error_msg = "Timeout too large";
return {.status = ExecResult::kStartFailed};
}
// Start subprocess.
pid_t pid = ExecWithoutWait(arg_vector, error_msg);
if (pid == -1) {
return {.status = ExecResult::kStartFailed};
}
callbacks.on_start(pid);
// Wait for subprocess to finish.
ExecResult result;
if (timeout_sec >= 0) {
unique_fd pidfd = PidfdOpen(pid);
if (pidfd.get() >= 0) {
result =
WaitChildWithTimeout(pid, std::move(pidfd), arg_vector, timeout_sec * 1000, error_msg);
} else {
LOG(DEBUG) << StringPrintf(
"pidfd_open failed for pid %d: %s, falling back", pid, strerror(errno));
result = WaitChildWithTimeoutFallback(pid, arg_vector, timeout_sec * 1000, error_msg);
}
} else {
result = WaitChild(pid, arg_vector, /*no_wait=*/true, error_msg);
}
if (stat != nullptr) {
std::string local_error_msg;
if (!GetStat(pid, stat, &local_error_msg)) {
LOG(ERROR) << "Failed to get process stat: " << local_error_msg;
}
}
callbacks.on_end(pid);
std::string local_error_msg;
// TODO(jiakaiz): Use better logic to detect waitid failure.
if (WaitChild(pid, arg_vector, /*no_wait=*/false, &local_error_msg).status ==
ExecResult::kUnknown) {
LOG(ERROR) << "Failed to clean up child process '" << arg_vector[0] << "': " << local_error_msg;
}
return result;
}
bool ExecUtils::Exec(const std::vector<std::string>& arg_vector, std::string* error_msg) const {
int status = ExecAndReturnCode(arg_vector, error_msg);
if (status < 0) {
// Internal error. The error message is already set.
return false;
}
if (status > 0) {
*error_msg =
StringPrintf("Failed to execute (%s) because the child process returns non-zero exit code",
ToCommandLine(arg_vector).c_str());
return false;
}
return true;
}
unique_fd ExecUtils::PidfdOpen(pid_t pid) const {
#ifdef __BIONIC__
return unique_fd(pidfd_open(pid, /*flags=*/0));
#else
// There is no glibc wrapper for pidfd_open.
#ifndef SYS_pidfd_open
constexpr int SYS_pidfd_open = 434;
#endif
return unique_fd(syscall(SYS_pidfd_open, pid, /*flags=*/0));
#endif
}
std::string ExecUtils::GetProcStat(pid_t pid) const {
std::string stat_content;
if (!ReadFileToString(StringPrintf("/proc/%d/stat", pid), &stat_content)) {
stat_content = "";
}
return stat_content;
}
std::optional<int64_t> ExecUtils::GetUptimeMs(std::string* error_msg) const {
timespec t;
if (clock_gettime(CLOCK_MONOTONIC, &t) != 0) {
*error_msg = ART_FORMAT("Failed to get uptime: {}", strerror(errno));
return std::nullopt;
}
return t.tv_sec * 1000 + t.tv_nsec / 1000000;
}
int64_t ExecUtils::GetTicksPerSec() const { return sysconf(_SC_CLK_TCK); }
bool ExecUtils::GetStat(pid_t pid,
/*out*/ ProcessStat* stat,
/*out*/ std::string* error_msg) const {
std::optional<int64_t> uptime_ms = GetUptimeMs(error_msg);
if (!uptime_ms.has_value()) {
return false;
}
std::string stat_content = GetProcStat(pid);
if (stat_content.empty()) {
*error_msg = StringPrintf("Failed to read /proc/%d/stat: %s", pid, strerror(errno));
return false;
}
int64_t ticks_per_sec = GetTicksPerSec();
if (!ParseProcStat(stat_content, *uptime_ms, ticks_per_sec, stat)) {
*error_msg = StringPrintf("Failed to parse /proc/%d/stat '%s'", pid, stat_content.c_str());
return false;
}
return true;
}
} // namespace art