James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 1 | #include "aos/starter/subprocess.h" |
| 2 | |
| 3 | #include <grp.h> |
| 4 | #include <pwd.h> |
| 5 | #include <sys/prctl.h> |
| 6 | #include <sys/types.h> |
| 7 | #include <sys/wait.h> |
| 8 | |
| 9 | #include "glog/logging.h" |
| 10 | |
James Kuszmaul | 8544c49 | 2023-07-31 15:00:38 -0700 | [diff] [blame] | 11 | #include "aos/flatbuffer_merge.h" |
| 12 | |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 13 | namespace aos::starter { |
| 14 | |
Austin Schuh | bbeb37e | 2022-08-17 16:19:27 -0700 | [diff] [blame] | 15 | // RAII class to become root and restore back to the original user and group |
| 16 | // afterwards. |
| 17 | class Sudo { |
| 18 | public: |
| 19 | Sudo() { |
| 20 | // Save what we were. |
| 21 | PCHECK(getresuid(&ruid_, &euid_, &suid_) == 0); |
| 22 | PCHECK(getresgid(&rgid_, &egid_, &sgid_) == 0); |
| 23 | |
| 24 | // Become root. |
| 25 | PCHECK(setresuid(/* ruid */ 0 /* root */, /* euid */ 0, /* suid */ 0) == 0) |
| 26 | << ": Failed to become root"; |
| 27 | PCHECK(setresgid(/* ruid */ 0 /* root */, /* euid */ 0, /* suid */ 0) == 0) |
| 28 | << ": Failed to become root"; |
| 29 | } |
| 30 | |
| 31 | ~Sudo() { |
| 32 | // And recover. |
| 33 | PCHECK(setresgid(rgid_, egid_, sgid_) == 0); |
| 34 | PCHECK(setresuid(ruid_, euid_, suid_) == 0); |
| 35 | } |
| 36 | |
| 37 | uid_t ruid_, euid_, suid_; |
| 38 | gid_t rgid_, egid_, sgid_; |
| 39 | }; |
| 40 | |
Austin Schuh | 77e20a3 | 2023-08-01 12:25:03 -0700 | [diff] [blame] | 41 | MemoryCGroup::MemoryCGroup(std::string_view name, Create should_create) |
| 42 | : cgroup_(absl::StrCat("/sys/fs/cgroup/memory/aos_", name)), |
| 43 | should_create_(should_create) { |
| 44 | if (should_create_ == Create::kDoCreate) { |
| 45 | Sudo sudo; |
| 46 | int ret = mkdir(cgroup_.c_str(), 0755); |
Austin Schuh | bbeb37e | 2022-08-17 16:19:27 -0700 | [diff] [blame] | 47 | |
Austin Schuh | 77e20a3 | 2023-08-01 12:25:03 -0700 | [diff] [blame] | 48 | if (ret != 0) { |
| 49 | if (errno == EEXIST) { |
| 50 | PCHECK(rmdir(cgroup_.c_str()) == 0) |
| 51 | << ": Failed to remove previous cgroup " << cgroup_; |
| 52 | ret = mkdir(cgroup_.c_str(), 0755); |
| 53 | } |
Austin Schuh | bbeb37e | 2022-08-17 16:19:27 -0700 | [diff] [blame] | 54 | } |
Austin Schuh | bbeb37e | 2022-08-17 16:19:27 -0700 | [diff] [blame] | 55 | |
Austin Schuh | 77e20a3 | 2023-08-01 12:25:03 -0700 | [diff] [blame] | 56 | if (ret != 0) { |
| 57 | PLOG(FATAL) << ": Failed to create cgroup aos_" << cgroup_ |
| 58 | << ", do you have permission?"; |
| 59 | } |
Austin Schuh | bbeb37e | 2022-08-17 16:19:27 -0700 | [diff] [blame] | 60 | } |
| 61 | } |
| 62 | |
| 63 | void MemoryCGroup::AddTid(pid_t pid) { |
| 64 | if (pid == 0) { |
| 65 | pid = getpid(); |
| 66 | } |
Austin Schuh | 77e20a3 | 2023-08-01 12:25:03 -0700 | [diff] [blame] | 67 | if (should_create_ == Create::kDoCreate) { |
| 68 | Sudo sudo; |
| 69 | util::WriteStringToFileOrDie(absl::StrCat(cgroup_, "/tasks").c_str(), |
| 70 | std::to_string(pid)); |
| 71 | } else { |
| 72 | util::WriteStringToFileOrDie(absl::StrCat(cgroup_, "/tasks").c_str(), |
| 73 | std::to_string(pid)); |
| 74 | } |
Austin Schuh | bbeb37e | 2022-08-17 16:19:27 -0700 | [diff] [blame] | 75 | } |
| 76 | |
| 77 | void MemoryCGroup::SetLimit(std::string_view limit_name, uint64_t limit_value) { |
Austin Schuh | 77e20a3 | 2023-08-01 12:25:03 -0700 | [diff] [blame] | 78 | if (should_create_ == Create::kDoCreate) { |
| 79 | Sudo sudo; |
| 80 | util::WriteStringToFileOrDie(absl::StrCat(cgroup_, "/", limit_name).c_str(), |
| 81 | std::to_string(limit_value)); |
| 82 | } else { |
| 83 | util::WriteStringToFileOrDie(absl::StrCat(cgroup_, "/", limit_name).c_str(), |
| 84 | std::to_string(limit_value)); |
| 85 | } |
Austin Schuh | bbeb37e | 2022-08-17 16:19:27 -0700 | [diff] [blame] | 86 | } |
| 87 | |
| 88 | MemoryCGroup::~MemoryCGroup() { |
Austin Schuh | 77e20a3 | 2023-08-01 12:25:03 -0700 | [diff] [blame] | 89 | if (should_create_ == Create::kDoCreate) { |
| 90 | Sudo sudo; |
| 91 | PCHECK(rmdir(absl::StrCat(cgroup_).c_str()) == 0); |
| 92 | } |
Austin Schuh | bbeb37e | 2022-08-17 16:19:27 -0700 | [diff] [blame] | 93 | } |
| 94 | |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 95 | SignalListener::SignalListener(aos::ShmEventLoop *loop, |
| 96 | std::function<void(signalfd_siginfo)> callback) |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 97 | : SignalListener(loop->epoll(), std::move(callback)) {} |
| 98 | |
| 99 | SignalListener::SignalListener(aos::internal::EPoll *epoll, |
| 100 | std::function<void(signalfd_siginfo)> callback) |
| 101 | : SignalListener(epoll, callback, |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 102 | {SIGHUP, SIGINT, SIGQUIT, SIGABRT, SIGFPE, SIGSEGV, |
| 103 | SIGPIPE, SIGTERM, SIGBUS, SIGXCPU, SIGCHLD}) {} |
| 104 | |
| 105 | SignalListener::SignalListener(aos::ShmEventLoop *loop, |
| 106 | std::function<void(signalfd_siginfo)> callback, |
| 107 | std::initializer_list<unsigned int> signals) |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 108 | : SignalListener(loop->epoll(), std::move(callback), std::move(signals)) {} |
| 109 | |
| 110 | SignalListener::SignalListener(aos::internal::EPoll *epoll, |
| 111 | std::function<void(signalfd_siginfo)> callback, |
| 112 | std::initializer_list<unsigned int> signals) |
| 113 | : epoll_(epoll), callback_(std::move(callback)), signalfd_(signals) { |
| 114 | epoll_->OnReadable(signalfd_.fd(), [this] { |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 115 | signalfd_siginfo info = signalfd_.Read(); |
| 116 | |
| 117 | if (info.ssi_signo == 0) { |
| 118 | LOG(WARNING) << "Could not read " << sizeof(signalfd_siginfo) << " bytes"; |
| 119 | return; |
| 120 | } |
| 121 | |
| 122 | callback_(info); |
| 123 | }); |
| 124 | } |
| 125 | |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 126 | SignalListener::~SignalListener() { epoll_->DeleteFd(signalfd_.fd()); } |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 127 | |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 128 | Application::Application(std::string_view name, |
| 129 | std::string_view executable_name, |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 130 | aos::EventLoop *event_loop, |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 131 | std::function<void()> on_change, |
| 132 | QuietLogging quiet_flag) |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 133 | : name_(name), |
| 134 | path_(executable_name), |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 135 | event_loop_(event_loop), |
| 136 | start_timer_(event_loop_->AddTimer([this] { |
| 137 | status_ = aos::starter::State::RUNNING; |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 138 | LOG_IF(INFO, quiet_flag_ == QuietLogging::kNo) |
| 139 | << "Started '" << name_ << "' pid: " << pid_; |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 140 | })), |
| 141 | restart_timer_(event_loop_->AddTimer([this] { DoStart(); })), |
| 142 | stop_timer_(event_loop_->AddTimer([this] { |
| 143 | if (kill(pid_, SIGKILL) == 0) { |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 144 | LOG_IF(WARNING, quiet_flag_ == QuietLogging::kNo) |
| 145 | << "Failed to stop, sending SIGKILL to '" << name_ |
| 146 | << "' pid: " << pid_; |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 147 | } |
| 148 | })), |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 149 | pipe_timer_(event_loop_->AddTimer([this]() { FetchOutputs(); })), |
| 150 | child_status_handler_( |
| 151 | event_loop_->AddTimer([this]() { MaybeHandleSignal(); })), |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 152 | on_change_({on_change}), |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 153 | quiet_flag_(quiet_flag) { |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 154 | event_loop_->OnRun([this]() { |
| 155 | // Every second poll to check if the child is dead. This is used as a |
| 156 | // default for the case where the user is not directly catching SIGCHLD and |
| 157 | // calling MaybeHandleSignal for us. |
Philipp Schrader | a671252 | 2023-07-05 20:25:11 -0700 | [diff] [blame] | 158 | child_status_handler_->Schedule(event_loop_->monotonic_now(), |
| 159 | std::chrono::seconds(1)); |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 160 | }); |
| 161 | } |
| 162 | |
| 163 | Application::Application(const aos::Application *application, |
| 164 | aos::EventLoop *event_loop, |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 165 | std::function<void()> on_change, |
| 166 | QuietLogging quiet_flag) |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 167 | : Application(application->name()->string_view(), |
| 168 | application->has_executable_name() |
| 169 | ? application->executable_name()->string_view() |
| 170 | : application->name()->string_view(), |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 171 | event_loop, on_change, quiet_flag) { |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 172 | user_name_ = application->has_user() ? application->user()->str() : ""; |
| 173 | user_ = application->has_user() ? FindUid(user_name_.c_str()) : std::nullopt; |
| 174 | group_ = application->has_user() ? FindPrimaryGidForUser(user_name_.c_str()) |
| 175 | : std::nullopt; |
| 176 | autostart_ = application->autostart(); |
| 177 | autorestart_ = application->autorestart(); |
| 178 | if (application->has_args()) { |
| 179 | set_args(*application->args()); |
| 180 | } |
Austin Schuh | bbeb37e | 2022-08-17 16:19:27 -0700 | [diff] [blame] | 181 | |
| 182 | if (application->has_memory_limit() && application->memory_limit() > 0) { |
| 183 | SetMemoryLimit(application->memory_limit()); |
| 184 | } |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 185 | } |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 186 | |
| 187 | void Application::DoStart() { |
| 188 | if (status_ != aos::starter::State::WAITING) { |
| 189 | return; |
| 190 | } |
| 191 | |
| 192 | start_timer_->Disable(); |
| 193 | restart_timer_->Disable(); |
| 194 | |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 195 | status_pipes_ = util::ScopedPipe::MakePipe(); |
| 196 | |
| 197 | if (capture_stdout_) { |
| 198 | stdout_pipes_ = util::ScopedPipe::MakePipe(); |
| 199 | stdout_.clear(); |
| 200 | } |
| 201 | if (capture_stderr_) { |
| 202 | stderr_pipes_ = util::ScopedPipe::MakePipe(); |
| 203 | stderr_.clear(); |
| 204 | } |
| 205 | |
Philipp Schrader | a671252 | 2023-07-05 20:25:11 -0700 | [diff] [blame] | 206 | pipe_timer_->Schedule(event_loop_->monotonic_now(), |
| 207 | std::chrono::milliseconds(100)); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 208 | |
| 209 | const pid_t pid = fork(); |
| 210 | |
| 211 | if (pid != 0) { |
| 212 | if (pid == -1) { |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 213 | PLOG_IF(WARNING, quiet_flag_ == QuietLogging::kNo) |
| 214 | << "Failed to fork '" << name_ << "'"; |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 215 | stop_reason_ = aos::starter::LastStopReason::FORK_ERR; |
| 216 | status_ = aos::starter::State::STOPPED; |
| 217 | } else { |
| 218 | pid_ = pid; |
| 219 | id_ = next_id_++; |
| 220 | start_time_ = event_loop_->monotonic_now(); |
| 221 | status_ = aos::starter::State::STARTING; |
James Kuszmaul | 8544c49 | 2023-07-31 15:00:38 -0700 | [diff] [blame] | 222 | latest_timing_report_version_.reset(); |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 223 | LOG_IF(INFO, quiet_flag_ == QuietLogging::kNo) |
| 224 | << "Starting '" << name_ << "' pid " << pid_; |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 225 | |
Philipp Schrader | a671252 | 2023-07-05 20:25:11 -0700 | [diff] [blame] | 226 | // Set up timer which moves application to RUNNING state if it is still |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 227 | // alive in 1 second. |
Philipp Schrader | a671252 | 2023-07-05 20:25:11 -0700 | [diff] [blame] | 228 | start_timer_->Schedule(event_loop_->monotonic_now() + |
| 229 | std::chrono::seconds(1)); |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 230 | // Since we are the parent process, clear our write-side of all the pipes. |
| 231 | status_pipes_.write.reset(); |
| 232 | stdout_pipes_.write.reset(); |
| 233 | stderr_pipes_.write.reset(); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 234 | } |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 235 | OnChange(); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 236 | return; |
| 237 | } |
| 238 | |
Austin Schuh | bbeb37e | 2022-08-17 16:19:27 -0700 | [diff] [blame] | 239 | if (memory_cgroup_) { |
| 240 | memory_cgroup_->AddTid(); |
| 241 | } |
| 242 | |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 243 | // Since we are the child process, clear our read-side of all the pipes. |
| 244 | status_pipes_.read.reset(); |
| 245 | stdout_pipes_.read.reset(); |
| 246 | stderr_pipes_.read.reset(); |
| 247 | |
| 248 | // The status pipe will not be needed if the execve succeeds. |
| 249 | status_pipes_.write->SetCloexec(); |
| 250 | |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 251 | // Clear out signal mask of parent so forked process receives all signals |
| 252 | // normally. |
| 253 | sigset_t empty_mask; |
| 254 | sigemptyset(&empty_mask); |
| 255 | sigprocmask(SIG_SETMASK, &empty_mask, nullptr); |
| 256 | |
| 257 | // Cleanup children if starter dies in a way that is not handled gracefully. |
| 258 | if (prctl(PR_SET_PDEATHSIG, SIGKILL) == -1) { |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 259 | status_pipes_.write->Write( |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 260 | static_cast<uint32_t>(aos::starter::LastStopReason::SET_PRCTL_ERR)); |
| 261 | PLOG(FATAL) << "Could not set PR_SET_PDEATHSIG to SIGKILL"; |
| 262 | } |
| 263 | |
| 264 | if (group_) { |
| 265 | CHECK(!user_name_.empty()); |
| 266 | // The manpage for setgroups says we just need CAP_SETGID, but empirically |
| 267 | // we also need the effective UID to be 0 to make it work. user_ must also |
| 268 | // be set so we change this effective UID back later. |
| 269 | CHECK(user_); |
| 270 | if (seteuid(0) == -1) { |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 271 | status_pipes_.write->Write( |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 272 | static_cast<uint32_t>(aos::starter::LastStopReason::SET_GRP_ERR)); |
| 273 | PLOG(FATAL) << "Could not seteuid(0) for " << name_ |
| 274 | << " in preparation for setting groups"; |
| 275 | } |
| 276 | if (initgroups(user_name_.c_str(), *group_) == -1) { |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 277 | status_pipes_.write->Write( |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 278 | static_cast<uint32_t>(aos::starter::LastStopReason::SET_GRP_ERR)); |
| 279 | PLOG(FATAL) << "Could not initialize normal groups for " << name_ |
| 280 | << " as " << user_name_ << " with " << *group_; |
| 281 | } |
| 282 | if (setgid(*group_) == -1) { |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 283 | status_pipes_.write->Write( |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 284 | static_cast<uint32_t>(aos::starter::LastStopReason::SET_GRP_ERR)); |
| 285 | PLOG(FATAL) << "Could not set group for " << name_ << " to " << *group_; |
| 286 | } |
| 287 | } |
| 288 | |
| 289 | if (user_) { |
| 290 | if (setuid(*user_) == -1) { |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 291 | status_pipes_.write->Write( |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 292 | static_cast<uint32_t>(aos::starter::LastStopReason::SET_USR_ERR)); |
| 293 | PLOG(FATAL) << "Could not set user for " << name_ << " to " << *user_; |
| 294 | } |
| 295 | } |
| 296 | |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 297 | if (capture_stdout_) { |
| 298 | PCHECK(STDOUT_FILENO == dup2(stdout_pipes_.write->fd(), STDOUT_FILENO)); |
| 299 | stdout_pipes_.write.reset(); |
| 300 | } |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 301 | |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 302 | if (capture_stderr_) { |
| 303 | PCHECK(STDERR_FILENO == dup2(stderr_pipes_.write->fd(), STDERR_FILENO)); |
| 304 | stderr_pipes_.write.reset(); |
| 305 | } |
| 306 | |
Sanjay Narayanan | 01a228f | 2022-04-26 14:19:30 -0700 | [diff] [blame] | 307 | if (run_as_sudo_) { |
Sarah Newman | 6d1e53b | 2022-08-09 14:38:08 -0700 | [diff] [blame] | 308 | // For sudo we must supply the actual path |
| 309 | args_.insert(args_.begin(), path_); |
Sanjay Narayanan | 01a228f | 2022-04-26 14:19:30 -0700 | [diff] [blame] | 310 | args_.insert(args_.begin(), kSudo); |
Sarah Newman | 6d1e53b | 2022-08-09 14:38:08 -0700 | [diff] [blame] | 311 | } else { |
| 312 | // argv[0] should be the program name |
| 313 | args_.insert(args_.begin(), name_); |
Sanjay Narayanan | 01a228f | 2022-04-26 14:19:30 -0700 | [diff] [blame] | 314 | } |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 315 | |
| 316 | std::vector<char *> cargs = CArgs(); |
Philipp Schrader | 790cb54 | 2023-07-05 21:06:52 -0700 | [diff] [blame] | 317 | const char *path = run_as_sudo_ ? kSudo : path_.c_str(); |
Sanjay Narayanan | 01a228f | 2022-04-26 14:19:30 -0700 | [diff] [blame] | 318 | execvp(path, cargs.data()); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 319 | |
| 320 | // If we got here, something went wrong |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 321 | status_pipes_.write->Write( |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 322 | static_cast<uint32_t>(aos::starter::LastStopReason::EXECV_ERR)); |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 323 | PLOG_IF(WARNING, quiet_flag_ == QuietLogging::kNo) |
| 324 | << "Could not execute " << name_ << " (" << path_ << ')'; |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 325 | |
| 326 | _exit(EXIT_FAILURE); |
| 327 | } |
| 328 | |
James Kuszmaul | 8544c49 | 2023-07-31 15:00:38 -0700 | [diff] [blame] | 329 | void Application::ObserveTimingReport( |
| 330 | const aos::monotonic_clock::time_point send_time, |
| 331 | const aos::timing::Report *msg) { |
| 332 | if (msg->name()->string_view() == name_ && msg->pid() == pid_ && |
| 333 | msg->has_version()) { |
| 334 | latest_timing_report_version_ = msg->version()->str(); |
| 335 | last_timing_report_ = send_time; |
| 336 | } |
| 337 | } |
| 338 | |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 339 | void Application::FetchOutputs() { |
| 340 | if (capture_stdout_) { |
| 341 | stdout_pipes_.read->Read(&stdout_); |
| 342 | } |
| 343 | if (capture_stderr_) { |
| 344 | stderr_pipes_.read->Read(&stderr_); |
| 345 | } |
| 346 | } |
| 347 | |
| 348 | const std::string &Application::GetStdout() { |
| 349 | CHECK(capture_stdout_); |
| 350 | FetchOutputs(); |
| 351 | return stdout_; |
| 352 | } |
| 353 | |
| 354 | const std::string &Application::GetStderr() { |
| 355 | CHECK(capture_stderr_); |
| 356 | FetchOutputs(); |
| 357 | return stderr_; |
| 358 | } |
| 359 | |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 360 | void Application::DoStop(bool restart) { |
| 361 | // If stop or restart received, the old state of these is no longer applicable |
| 362 | // so cancel both. |
| 363 | restart_timer_->Disable(); |
| 364 | start_timer_->Disable(); |
| 365 | |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 366 | FetchOutputs(); |
| 367 | |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 368 | switch (status_) { |
| 369 | case aos::starter::State::STARTING: |
| 370 | case aos::starter::State::RUNNING: { |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 371 | LOG_IF(INFO, quiet_flag_ == QuietLogging::kNo) |
| 372 | << "Stopping '" << name_ << "' pid: " << pid_ << " with signal " |
| 373 | << SIGINT; |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 374 | status_ = aos::starter::State::STOPPING; |
| 375 | |
| 376 | kill(pid_, SIGINT); |
| 377 | |
| 378 | // Watchdog timer to SIGKILL application if it is still running 1 second |
| 379 | // after SIGINT |
Philipp Schrader | a671252 | 2023-07-05 20:25:11 -0700 | [diff] [blame] | 380 | stop_timer_->Schedule(event_loop_->monotonic_now() + |
| 381 | std::chrono::seconds(1)); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 382 | queue_restart_ = restart; |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 383 | OnChange(); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 384 | break; |
| 385 | } |
| 386 | case aos::starter::State::WAITING: { |
| 387 | // If waiting to restart, and receives restart, skip the waiting period |
| 388 | // and restart immediately. If stop received, all we have to do is move |
| 389 | // to the STOPPED state. |
| 390 | if (restart) { |
| 391 | DoStart(); |
| 392 | } else { |
| 393 | status_ = aos::starter::State::STOPPED; |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 394 | OnChange(); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 395 | } |
| 396 | break; |
| 397 | } |
| 398 | case aos::starter::State::STOPPING: { |
| 399 | // If the application is already stopping, then we just need to update the |
| 400 | // restart flag to the most recent status. |
| 401 | queue_restart_ = restart; |
| 402 | break; |
| 403 | } |
| 404 | case aos::starter::State::STOPPED: { |
| 405 | // Restart immediately if the application is already stopped |
| 406 | if (restart) { |
| 407 | status_ = aos::starter::State::WAITING; |
| 408 | DoStart(); |
| 409 | } |
| 410 | break; |
| 411 | } |
| 412 | } |
| 413 | } |
| 414 | |
| 415 | void Application::QueueStart() { |
| 416 | status_ = aos::starter::State::WAITING; |
| 417 | |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 418 | LOG_IF(INFO, quiet_flag_ == QuietLogging::kNo) |
| 419 | << "Restarting " << name_ << " in 3 seconds"; |
Philipp Schrader | a671252 | 2023-07-05 20:25:11 -0700 | [diff] [blame] | 420 | restart_timer_->Schedule(event_loop_->monotonic_now() + |
| 421 | std::chrono::seconds(3)); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 422 | start_timer_->Disable(); |
| 423 | stop_timer_->Disable(); |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 424 | OnChange(); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 425 | } |
| 426 | |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 427 | std::vector<char *> Application::CArgs() { |
| 428 | std::vector<char *> cargs; |
| 429 | std::transform(args_.begin(), args_.end(), std::back_inserter(cargs), |
| 430 | [](std::string &str) { return str.data(); }); |
| 431 | cargs.push_back(nullptr); |
| 432 | return cargs; |
| 433 | } |
| 434 | |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 435 | void Application::set_args( |
| 436 | const flatbuffers::Vector<flatbuffers::Offset<flatbuffers::String>> &v) { |
| 437 | args_.clear(); |
| 438 | std::transform(v.begin(), v.end(), std::back_inserter(args_), |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 439 | [](const flatbuffers::String *str) { return str->str(); }); |
| 440 | } |
| 441 | |
| 442 | void Application::set_args(std::vector<std::string> args) { |
| 443 | args_ = std::move(args); |
| 444 | } |
| 445 | |
| 446 | void Application::set_capture_stdout(bool capture) { |
| 447 | capture_stdout_ = capture; |
| 448 | } |
| 449 | |
| 450 | void Application::set_capture_stderr(bool capture) { |
| 451 | capture_stderr_ = capture; |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 452 | } |
| 453 | |
| 454 | std::optional<uid_t> Application::FindUid(const char *name) { |
| 455 | // TODO(austin): Use the reentrant version. This should be safe. |
| 456 | struct passwd *user_data = getpwnam(name); |
| 457 | if (user_data != nullptr) { |
| 458 | return user_data->pw_uid; |
| 459 | } else { |
| 460 | LOG(FATAL) << "Could not find user " << name; |
| 461 | return std::nullopt; |
| 462 | } |
| 463 | } |
| 464 | |
| 465 | std::optional<gid_t> Application::FindPrimaryGidForUser(const char *name) { |
| 466 | // TODO(austin): Use the reentrant version. This should be safe. |
| 467 | struct passwd *user_data = getpwnam(name); |
| 468 | if (user_data != nullptr) { |
| 469 | return user_data->pw_gid; |
| 470 | } else { |
| 471 | LOG(FATAL) << "Could not find user " << name; |
| 472 | return std::nullopt; |
| 473 | } |
| 474 | } |
| 475 | |
| 476 | flatbuffers::Offset<aos::starter::ApplicationStatus> |
James Kuszmaul | 6295a64 | 2022-03-22 15:23:59 -0700 | [diff] [blame] | 477 | Application::PopulateStatus(flatbuffers::FlatBufferBuilder *builder, |
| 478 | util::Top *top) { |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 479 | CHECK_NOTNULL(builder); |
| 480 | auto name_fbs = builder->CreateString(name_); |
| 481 | |
James Kuszmaul | 6295a64 | 2022-03-22 15:23:59 -0700 | [diff] [blame] | 482 | const bool valid_pid = pid_ > 0 && status_ != aos::starter::State::STOPPED; |
| 483 | const flatbuffers::Offset<util::ProcessInfo> process_info = |
| 484 | valid_pid ? top->InfoForProcess(builder, pid_) |
| 485 | : flatbuffers::Offset<util::ProcessInfo>(); |
| 486 | |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 487 | aos::starter::ApplicationStatus::Builder status_builder(*builder); |
| 488 | status_builder.add_name(name_fbs); |
| 489 | status_builder.add_state(status_); |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 490 | if (exit_code_.has_value()) { |
| 491 | status_builder.add_last_exit_code(exit_code_.value()); |
| 492 | } |
James Kuszmaul | 8544c49 | 2023-07-31 15:00:38 -0700 | [diff] [blame] | 493 | status_builder.add_has_active_timing_report( |
| 494 | last_timing_report_ + |
| 495 | // Leave a bit of margin on the timing report receipt time, to allow |
| 496 | // for timing errors. |
| 497 | 3 * std::chrono::milliseconds(FLAGS_timing_report_ms) > |
| 498 | event_loop_->monotonic_now()); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 499 | status_builder.add_last_stop_reason(stop_reason_); |
| 500 | if (pid_ != -1) { |
| 501 | status_builder.add_pid(pid_); |
| 502 | status_builder.add_id(id_); |
| 503 | } |
James Kuszmaul | 6295a64 | 2022-03-22 15:23:59 -0700 | [diff] [blame] | 504 | // Note that even if process_info is null, calling add_process_info is fine. |
| 505 | status_builder.add_process_info(process_info); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 506 | status_builder.add_last_start_time(start_time_.time_since_epoch().count()); |
| 507 | return status_builder.Finish(); |
| 508 | } |
| 509 | |
| 510 | void Application::Terminate() { |
| 511 | stop_reason_ = aos::starter::LastStopReason::TERMINATE; |
| 512 | DoStop(false); |
| 513 | terminating_ = true; |
| 514 | } |
| 515 | |
| 516 | void Application::HandleCommand(aos::starter::Command cmd) { |
| 517 | switch (cmd) { |
| 518 | case aos::starter::Command::START: { |
| 519 | switch (status_) { |
| 520 | case aos::starter::State::WAITING: { |
| 521 | restart_timer_->Disable(); |
| 522 | DoStart(); |
| 523 | break; |
| 524 | } |
| 525 | case aos::starter::State::STARTING: { |
| 526 | break; |
| 527 | } |
| 528 | case aos::starter::State::RUNNING: { |
| 529 | break; |
| 530 | } |
| 531 | case aos::starter::State::STOPPING: { |
| 532 | queue_restart_ = true; |
| 533 | break; |
| 534 | } |
| 535 | case aos::starter::State::STOPPED: { |
| 536 | status_ = aos::starter::State::WAITING; |
| 537 | DoStart(); |
| 538 | break; |
| 539 | } |
| 540 | } |
| 541 | break; |
| 542 | } |
| 543 | case aos::starter::Command::STOP: { |
| 544 | stop_reason_ = aos::starter::LastStopReason::STOP_REQUESTED; |
| 545 | DoStop(false); |
| 546 | break; |
| 547 | } |
| 548 | case aos::starter::Command::RESTART: { |
| 549 | stop_reason_ = aos::starter::LastStopReason::RESTART_REQUESTED; |
| 550 | DoStop(true); |
| 551 | break; |
| 552 | } |
| 553 | } |
| 554 | } |
| 555 | |
| 556 | bool Application::MaybeHandleSignal() { |
| 557 | int status; |
| 558 | |
Sarah Newman | 21c5920 | 2022-06-16 12:36:33 -0700 | [diff] [blame] | 559 | if (status_ == aos::starter::State::WAITING || |
| 560 | status_ == aos::starter::State::STOPPED) { |
| 561 | // We can't possibly have received a signal meant for this process. |
| 562 | return false; |
| 563 | } |
| 564 | |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 565 | // Check if the status of this process has changed |
Sarah Newman | 21c5920 | 2022-06-16 12:36:33 -0700 | [diff] [blame] | 566 | // The PID won't be -1 if this application has ever been run successfully |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 567 | if (pid_ == -1 || waitpid(pid_, &status, WNOHANG) != pid_) { |
| 568 | return false; |
| 569 | } |
| 570 | |
| 571 | // Check that the event was the process exiting |
| 572 | if (!WIFEXITED(status) && !WIFSIGNALED(status)) { |
| 573 | return false; |
| 574 | } |
| 575 | |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 576 | start_timer_->Disable(); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 577 | exit_time_ = event_loop_->monotonic_now(); |
| 578 | exit_code_ = WIFEXITED(status) ? WEXITSTATUS(status) : WTERMSIG(status); |
| 579 | |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 580 | if (auto read_result = status_pipes_.read->Read()) { |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 581 | stop_reason_ = static_cast<aos::starter::LastStopReason>(*read_result); |
| 582 | } |
| 583 | |
| 584 | switch (status_) { |
| 585 | case aos::starter::State::STARTING: { |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 586 | if (exit_code_.value() == 0) { |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 587 | LOG_IF(INFO, quiet_flag_ == QuietLogging::kNo) |
| 588 | << "Application '" << name_ << "' pid " << pid_ |
| 589 | << " exited with status " << exit_code_.value(); |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 590 | } else { |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 591 | LOG_IF(WARNING, quiet_flag_ == QuietLogging::kNo) |
| 592 | << "Failed to start '" << name_ << "' on pid " << pid_ |
| 593 | << " : Exited with status " << exit_code_.value(); |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 594 | } |
James Kuszmaul | 6f10b38 | 2022-03-11 22:31:38 -0800 | [diff] [blame] | 595 | if (autorestart()) { |
| 596 | QueueStart(); |
| 597 | } else { |
| 598 | status_ = aos::starter::State::STOPPED; |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 599 | OnChange(); |
James Kuszmaul | 6f10b38 | 2022-03-11 22:31:38 -0800 | [diff] [blame] | 600 | } |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 601 | break; |
| 602 | } |
| 603 | case aos::starter::State::RUNNING: { |
James Kuszmaul | d42edb4 | 2022-01-07 18:00:16 -0800 | [diff] [blame] | 604 | if (exit_code_.value() == 0) { |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 605 | LOG_IF(INFO, quiet_flag_ == QuietLogging::kNo) |
| 606 | << "Application '" << name_ << "' pid " << pid_ |
| 607 | << " exited with status " << exit_code_.value(); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 608 | } else { |
James Kuszmaul | 8544c49 | 2023-07-31 15:00:38 -0700 | [diff] [blame] | 609 | if (quiet_flag_ == QuietLogging::kNo) { |
| 610 | std::string version_string = |
| 611 | latest_timing_report_version_.has_value() |
| 612 | ? absl::StrCat("'", latest_timing_report_version_.value(), |
| 613 | "'") |
| 614 | : "unknown"; |
| 615 | LOG_IF(WARNING, quiet_flag_ == QuietLogging::kNo) |
| 616 | << "Application '" << name_ << "' pid " << pid_ << " version " |
| 617 | << version_string << " exited unexpectedly with status " |
| 618 | << exit_code_.value(); |
| 619 | } |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 620 | } |
James Kuszmaul | 6f10b38 | 2022-03-11 22:31:38 -0800 | [diff] [blame] | 621 | if (autorestart()) { |
| 622 | QueueStart(); |
| 623 | } else { |
| 624 | status_ = aos::starter::State::STOPPED; |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 625 | OnChange(); |
James Kuszmaul | 6f10b38 | 2022-03-11 22:31:38 -0800 | [diff] [blame] | 626 | } |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 627 | break; |
| 628 | } |
| 629 | case aos::starter::State::STOPPING: { |
payton.rehl | 2841b1c | 2023-05-25 17:23:55 -0700 | [diff] [blame] | 630 | LOG_IF(INFO, quiet_flag_ == QuietLogging::kNo) |
| 631 | << "Successfully stopped '" << name_ << "' pid: " << pid_ |
| 632 | << " with status " << exit_code_.value(); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 633 | status_ = aos::starter::State::STOPPED; |
| 634 | |
| 635 | // Disable force stop timer since the process already died |
| 636 | stop_timer_->Disable(); |
| 637 | |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 638 | OnChange(); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 639 | if (terminating_) { |
| 640 | return true; |
| 641 | } |
| 642 | |
| 643 | if (queue_restart_) { |
| 644 | queue_restart_ = false; |
| 645 | status_ = aos::starter::State::WAITING; |
| 646 | DoStart(); |
| 647 | } |
| 648 | break; |
| 649 | } |
| 650 | case aos::starter::State::WAITING: |
| 651 | case aos::starter::State::STOPPED: { |
Sarah Newman | 21c5920 | 2022-06-16 12:36:33 -0700 | [diff] [blame] | 652 | __builtin_unreachable(); |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 653 | break; |
| 654 | } |
| 655 | } |
| 656 | |
| 657 | return false; |
| 658 | } |
| 659 | |
Austin Schuh | 1cea903 | 2023-07-10 11:56:40 -0700 | [diff] [blame] | 660 | void Application::OnChange() { |
| 661 | for (auto &fn : on_change_) { |
| 662 | fn(); |
| 663 | } |
| 664 | } |
| 665 | |
Adam Snaider | 70deaf2 | 2023-08-11 13:58:34 -0700 | [diff] [blame^] | 666 | Application::~Application() { |
| 667 | start_timer_->Disable(); |
| 668 | restart_timer_->Disable(); |
| 669 | stop_timer_->Disable(); |
| 670 | pipe_timer_->Disable(); |
| 671 | child_status_handler_->Disable(); |
| 672 | } |
| 673 | |
James Kuszmaul | 3224b8e | 2022-01-07 19:00:39 -0800 | [diff] [blame] | 674 | } // namespace aos::starter |