eive-obsw/watchdog/Watchdog.cpp

#include "Watchdog.h"
#include "watchdogConf.h"

#include <errno.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <poll.h>
#include <unistd.h>
#include <fcntl.h>

#include <iostream>
#include <fstream>
#include <thread>
#include <cstring>
#include <filesystem>


WatchdogTask::WatchdogTask (): fd(0) {
    int result = 0;
    // Only create the FIFO if it does not exist yet
    if(not std::filesystem::exists(watchdog::FIFO_NAME)) {
        // Permission  666 or rw-rw-rw-
        mode_t mode = DEFFILEMODE;
        result = mkfifo(watchdog::FIFO_NAME.c_str(), mode);
        if(result != 0) {
            std::cerr << "eive-watchdog: Could not created named pipe at " <<
                    watchdog::FIFO_NAME << ", error " << errno << ": " << strerror(errno) <<
                    std::endl;
            throw std::runtime_error("eive-watchdog: FIFO creation failed");
        }
#if WATCHDOG_VERBOSE_LEVEL >= 1
        std::cout << "eive-watchdog: Pipe at " << watchdog::FIFO_NAME <<
                " created successfully" << std::endl;
#endif
    }
}

WatchdogTask::~WatchdogTask() {

}

int WatchdogTask::performOperation() {
    // Open FIFO read only and non-blocking
    fd = open(watchdog::FIFO_NAME.c_str(), O_RDONLY | O_NONBLOCK);
    if(fd < 0) {
        std::cerr << "eive-watchdog: Opening pipe " << watchdog::FIFO_NAME <<
                "read-only failed with " << errno << ": " << strerror(errno) << std::endl;
    }
    state = States::RUNNING;

    while(true) {
        WatchdogTask::LoopResult loopResult = watchdogLoop();
        switch(loopResult) {
        case(LoopResult::OK): {
            performRunningOperation();
            break;
        }
        case(LoopResult::CANCEL_RQ): {
            std::cout << "eive-watchdog: Received cancel request, closing watchdog.." << std::endl;
            return 0;
        }
        case(LoopResult::SUSPEND_RQ): {
            performSuspendOperation();
            break;
        }
        case(LoopResult::TIMEOUT): {
            performNotRunningOperation(loopResult);
            break;
        }
        case(LoopResult::HUNG_UP): {
            performNotRunningOperation(loopResult);
            break;
        }
        case(LoopResult::RESTART_RQ): {
            if(state == States::SUSPENDED or state == States::FAULTY) {
                performRunningOperation();
            }
            break;
        }
        case(LoopResult::FAULT): {
            using namespace std::chrono_literals;
            // Configuration error
            std::cerr << "Fault has occured in watchdog loop" << std::endl;
            // Prevent spam
            std::this_thread::sleep_for(2000ms);

        }
        }
    }
    if (close(fd) < 0) {
        std::cerr << "eive-watchdog: Closing named pipe at " << watchdog::FIFO_NAME <<
                "failed, error " << errno << ": " << strerror(errno) << std::endl;
    }
    std::cout << "eive-watchdog: Finished" << std::endl;
    return 0;
}

WatchdogTask::LoopResult WatchdogTask::watchdogLoop() {
    using namespace std::chrono_literals;
    struct pollfd waiter = {};
    waiter.fd = fd;
    waiter.events = POLLIN;

    switch(state) {
    case(States::SUSPENDED): {
        // Sleep, then check whether a restart request was received
        std::this_thread::sleep_for(1000ms);
        break;
    }
    case(States::RUNNING): {
        // Continue as usual
        break;
    }
    case(States::NOT_STARTED): {
        // This should not happen
        std::cerr << "eive-watchdog: State is NOT_STARTED, configuration error" << std::endl;
        break;
    }
    case(States::FAULTY): {
        // TODO: Not sure what to do yet. Continue for now
        break;
    }
    }

    // 10 seconds timeout, only poll one file descriptor
    switch(poll(&waiter, 1, watchdog::TIMEOUT_MS)) {
    case(0): {
        return LoopResult::TIMEOUT;
    }
    case(1): {
        return pollEvent(waiter);
    }
    default: {
        std::cerr << "eive-watchdog: Unknown poll error at " << watchdog::FIFO_NAME << ", error " <<
                errno << ": " << strerror(errno) << std::endl;
        break;
    }
    }
    return LoopResult::OK;
}

WatchdogTask::LoopResult WatchdogTask::pollEvent(struct pollfd& waiter) {
    if (waiter.revents & POLLIN) {
        ssize_t readLen = read(fd, buf.data(), buf.size());
        if (readLen < 0) {
            std::cerr << "eive-watchdog: Read error on pipe " << watchdog::FIFO_NAME <<
                    ", error " << errno << ": " << strerror(errno) << std::endl;
            return LoopResult::OK;
        }
#if WATCHDOG_VERBOSE_LEVEL == 2
        std::cout << "Read " << readLen << " byte(s) on the pipe " << FIFO_NAME
                << std::endl;
#endif
        else if(readLen >= 1) {
            return parseCommandByte(readLen);
        }

    }
    else if(waiter.revents & POLLERR) {
        std::cerr << "eive-watchdog: Poll error error on pipe " << watchdog::FIFO_NAME <<
                std::endl;
        return LoopResult::FAULT;
    }
    else if (waiter.revents & POLLHUP) {
        // Writer closed its end
        return LoopResult::HUNG_UP;
    }
    return LoopResult::FAULT;
}

WatchdogTask::LoopResult WatchdogTask::parseCommandByte(ssize_t readLen) {
    for(ssize_t idx = 0; idx < readLen; idx++) {
        char readChar = buf[idx];
        // Cancel request
        if(readChar == watchdog::CANCEL_CHAR) {
            return LoopResult::CANCEL_RQ;
        }
        // Begin request. Does not work if the operation was not suspended before
        else if(readChar ==  watchdog::RESTART_CHAR) {
            return LoopResult::RESTART_RQ;
        }
        // Suspend request
        else if(readChar == watchdog::SUSPEND_CHAR) {
            return LoopResult::SUSPEND_RQ;
        }
        // Everything else: All working as expected
    }
    return LoopResult::OK;
}

int WatchdogTask::performRunningOperation() {
    if(state != States::RUNNING) {
        state = States::RUNNING;
    }

    if(not obswRunning) {
        if(printNotRunningLatch) {
            // Reset latch so user can see timeouts
            printNotRunningLatch = false;
        }

        obswRunning = true;
        std::cout << "eive-watchdog: Running OBSW detected.." << std::endl;
#if WATCHDOG_CREATE_FILE_IF_RUNNING == 1
        std::cout << "eive-watchdog: Creating " << watchdog::RUNNING_FILE_NAME << std::endl;
        if (not std::filesystem::exists(watchdog::RUNNING_FILE_NAME)) {
            std::ofstream obswRunningFile(watchdog::RUNNING_FILE_NAME);
            if(not obswRunningFile.good()) {
                std::cerr << "Creating file " << watchdog::RUNNING_FILE_NAME << " failed"
                        << std::endl;
            }
        }
#endif
    }
    return 0;
}

int WatchdogTask::performNotRunningOperation(LoopResult type) {
    // Latch prevents spam on console
    if(not printNotRunningLatch) {
        if(type == LoopResult::HUNG_UP) {
            std::cout << "eive-watchdog: FIFO writer hung up!" << std::endl;
        }
        else {
            std::cout << "eive-watchdog: The FIFO timed out!" << std::endl;
        }
        printNotRunningLatch = true;
    }

    if(obswRunning) {
#if WATCHDOG_CREATE_FILE_IF_RUNNING == 1
        if (std::filesystem::exists(watchdog::RUNNING_FILE_NAME)) {
            int result = std::remove(watchdog::RUNNING_FILE_NAME.c_str());
            if(result != 0) {
                std::cerr << "Removing " << watchdog::RUNNING_FILE_NAME << " failed with code " <<
                        errno << ": " << strerror(errno) << std::endl;
            }
        }
#endif
        obswRunning = false;
    }
    if(type == LoopResult::HUNG_UP) {
        using namespace std::chrono_literals;
        // Prevent spam
        std::this_thread::sleep_for(2000ms);
    }
    return 0;
}

int WatchdogTask::performSuspendOperation() {
    if(state == States::RUNNING or state == States::FAULTY) {
        std::cout << "eive-watchdog: Suspending watchdog operations" << std::endl;
        watchdogRunning = false;
        state = States::SUSPENDED;
    }
    return 0;
}