/*

 * Copyright (c) 2014 DeNA Co., Ltd.

 *

 * Permission is hereby granted, free of charge, to any person obtaining a copy

 * of this software and associated documentation files (the "Software"), to

 * deal in the Software without restriction, including without limitation the

 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or

 * sell copies of the Software, and to permit persons to whom the Software is

 * furnished to do so, subject to the following conditions:

 *

 * The above copyright notice and this permission notice shall be included in

 * all copies or substantial portions of the Software.

 *

 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE

 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER

 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING

 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS

 * IN THE SOFTWARE.

 */

#include <assert.h>

#include <limits.h>

#include <stdio.h>

#include <sys/epoll.h>

#if H2O_USE_IO_URING

#include "h2o/io_uring.h"

#endif

#if 0

#define DEBUG_LOG(...) h2o_error_printf(__VA_ARGS__)

#else

#define DEBUG_LOG(...)

#endif

struct st_h2o_evloop_epoll_t {

    h2o_evloop_t super;

    int ep;

#if H2O_USE_IO_URING

    h2o_io_uring_t io_uring;

#endif

};

static int change_epoll_mode(struct st_h2o_evloop_socket_t *sock, uint32_t events)

{

    struct st_h2o_evloop_epoll_t *loop = (void *)sock->loop;

    struct epoll_event ev = {.events = events, .data = {.ptr = sock}};

    int op, ret;

    if ((sock->_flags & H2O_SOCKET_FLAG__EPOLL_IS_REGISTERED) == 0) {

        op = EPOLL_CTL_ADD;

        sock->_flags |= H2O_SOCKET_FLAG__EPOLL_IS_REGISTERED;

    } else {

        op = EPOLL_CTL_MOD;

    }

    while ((ret = epoll_ctl(loop->ep, op, sock->fd, &ev)) != 0 && errno == EINTR)

        ;

    return ret == 0;

}

static int delete_from_epoll_on_close(struct st_h2o_evloop_socket_t *sock)

{

    struct st_h2o_evloop_epoll_t *loop = (void *)sock->loop;

    int ret;

    if ((sock->_flags & H2O_SOCKET_FLAG__EPOLL_IS_REGISTERED) == 0)

        return 1;

    while ((ret = epoll_ctl(loop->ep, EPOLL_CTL_DEL, sock->fd, NULL)) != 0 && errno == EINTR)

        ;

    return ret == 0;

}

static int handle_zerocopy_notification(struct st_h2o_evloop_socket_t *sock)

{

#if H2O_USE_MSG_ZEROCOPY

    int made_progress = 0;

    /* Read the completion events and release buffers. `recvmmsg` with two entries is used as a cheap way of making sure that all

     * notifications are read from the queue (this requirement comes from the us eof edge trigger once the socket is closed). */

    while (1) {

        struct mmsghdr msg[2];

        char cbuf[2][CMSG_SPACE(sizeof(struct sock_extended_err))];

        for (size_t i = 0; i < PTLS_ELEMENTSOF(msg); ++i)

            msg[i] = (struct mmsghdr){.msg_hdr = {.msg_control = cbuf[i], .msg_controllen = sizeof(cbuf[i])}};

        struct timespec timeout = {0};

        ssize_t ret;

        while ((ret = recvmmsg(sock->fd, msg, PTLS_ELEMENTSOF(msg), MSG_ERRQUEUE, &timeout)) == -1 && errno == EINTR)

            ;

        if (ret == -1)

            break;

        for (size_t i = 0; i < ret; ++i) {

            struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg[i].msg_hdr);

            if (cmsg != NULL) {

                struct sock_extended_err *ee = (void *)CMSG_DATA(cmsg);

                if (ee->ee_errno == 0 && ee->ee_origin == SO_EE_ORIGIN_ZEROCOPY) {

                    /* for each range being obtained, convert the wrapped value to 64-bit, then release the memory */

                    for (uint32_t c32 = ee->ee_info; c32 <= ee->ee_data; ++c32) {

                        uint64_t c64 = (sock->super._zerocopy->first_counter & 0xffffffff00000000) | c32;

                        if (c64 < sock->super._zerocopy->first_counter)

                            c64 += 0x100000000;

                        void *p = zerocopy_buffers_release(sock->super._zerocopy, c64);

                        if (p != NULL) {

                            if (sock->super.ssl != NULL && p == sock->super.ssl->output.buf.base) {

                                /* buffer being released from zerocopy still has some pending data to be written */

                                assert(sock->super.ssl->output.zerocopy_owned);

                                sock->super.ssl->output.zerocopy_owned = 0;

                            } else {

                                h2o_mem_free_recycle(&h2o_socket_zerocopy_buffer_allocator, p);

                            }

                            --h2o_socket_num_zerocopy_buffers_inflight;

                        }

                    }

                }

            }

            made_progress = 1;

        }

        /* partial read means that the notification queue has become empty */

        if (ret < PTLS_ELEMENTSOF(msg))

            break;

    }

    /* if the socket has been shut down and zerocopy buffer has become empty, link the socket so that it would be destroyed */

    if (made_progress && (sock->_flags & H2O_SOCKET_FLAG_IS_DISPOSED) != 0 && zerocopy_buffers_is_empty(sock->super._zerocopy))

        link_to_statechanged(sock);

    return made_progress;

#else

    return 0;

#endif

}

static int update_status(struct st_h2o_evloop_epoll_t *loop)

{

    while (loop->super._statechanged.head != NULL) {

        /* detach the top */

        struct st_h2o_evloop_socket_t *sock = loop->super._statechanged.head;

        loop->super._statechanged.head = sock->_next_statechanged;

        sock->_next_statechanged = sock;

        /* update the state */

        if ((sock->_flags & H2O_SOCKET_FLAG_IS_DISPOSED) != 0) {

            if (sock->super._zerocopy == NULL || zerocopy_buffers_is_empty(sock->super._zerocopy)) {

                /* Call close (2) and destroy, now that all zero copy buffers have been reclaimed. */

                if (sock->super._zerocopy != NULL) {

                    zerocopy_buffers_dispose(sock->super._zerocopy);

                    free(sock->super._zerocopy);

                }

                if (sock->fd != -1) {

                    if (!delete_from_epoll_on_close(sock))

                        h2o_error_printf("update_status: epoll(DEL) returned error %d (fd=%d)\n", errno, sock->fd);

                    close(sock->fd);

                }

                free(sock);

            }

        } else {

            uint32_t events = 0;

            int changed = 0;

            if (h2o_socket_is_reading(&sock->super)) {

                events |= EPOLLIN;

                if ((sock->_flags & H2O_SOCKET_FLAG_IS_POLLED_FOR_READ) == 0) {

                    sock->_flags |= H2O_SOCKET_FLAG_IS_POLLED_FOR_READ;

                    changed = 1;

                }

            } else {

                if ((sock->_flags & H2O_SOCKET_FLAG_IS_POLLED_FOR_READ) != 0) {

                    sock->_flags &= ~H2O_SOCKET_FLAG_IS_POLLED_FOR_READ;

                    changed = 1;

                }

            }

            if (h2o_socket_is_writing(&sock->super)) {

                events |= EPOLLOUT;

                if ((sock->_flags & H2O_SOCKET_FLAG_IS_POLLED_FOR_WRITE) == 0) {

                    sock->_flags |= H2O_SOCKET_FLAG_IS_POLLED_FOR_WRITE;

                    changed = 1;

                }

            } else {

                if ((sock->_flags & H2O_SOCKET_FLAG_IS_POLLED_FOR_WRITE) != 0) {

                    sock->_flags &= ~H2O_SOCKET_FLAG_IS_POLLED_FOR_WRITE;

                    changed = 1;

                }

            }

            if (changed) {

                if (!change_epoll_mode(sock, events))

                    return -1;

            }

        }

    }

    loop->super._statechanged.tail_ref = &loop->super._statechanged.head;

    return 0;

}

int evloop_do_proceed(h2o_evloop_t *_loop, int32_t max_wait)

{

    struct st_h2o_evloop_epoll_t *loop = (struct st_h2o_evloop_epoll_t *)_loop;

    struct epoll_event events[256];

    int nevents, i;

    /* collect (and update) status */

    if (update_status(loop) != 0)

        return -1;

    /* poll */

    max_wait = adjust_max_wait(&loop->super, max_wait);

    nevents = epoll_wait(loop->ep, events, sizeof(events) / sizeof(events[0]), max_wait);

    update_now(&loop->super);

    if (nevents == -1)

        return -1;

    if (nevents != 0) {

        h2o_sliding_counter_start(&loop->super.exec_time_nanosec_counter, loop->super._now_nanosec);

    }

    /* update readable flags, perform writes */

    for (i = 0; i != nevents; ++i) {

        struct st_h2o_evloop_socket_t *sock = events[i].data.ptr;

        int notified = 0;

        /* When receiving HUP (indicating reset) while the socket is polled neither for read nor write, unregister the socket from

         * epoll, otherwise epoll_wait() would continue raising the HUP event. This problem cannot be avoided by using edge trigger.

         * The application will eventually try to read or write to the socket and at that point close the socket, detecting that it

         * has become unusable. */

        if ((events[i].events & EPOLLHUP) != 0 &&

            (sock->_flags & (H2O_SOCKET_FLAG_IS_POLLED_FOR_READ | H2O_SOCKET_FLAG_IS_POLLED_FOR_WRITE)) == 0 &&

            !(sock->super._zerocopy != NULL && (sock->_flags & H2O_SOCKET_FLAG_IS_DISPOSED) != 0)) {

            assert((sock->_flags & H2O_SOCKET_FLAG__EPOLL_IS_REGISTERED) != 0);

            int ret;

            while ((ret = epoll_ctl(loop->ep, EPOLL_CTL_DEL, sock->fd, NULL)) != 0 && errno == EINTR)

                ;

            if (ret != 0)

                h2o_error_printf("failed to unregister socket (fd:%d) that raised HUP; errno=%d\n", sock->fd, errno);

            sock->_flags &= ~H2O_SOCKET_FLAG__EPOLL_IS_REGISTERED;

            notified = 1;

        }

        /* If the error event was a zerocopy notification, hide the error notification to application. Doing so is fine because

         * level-triggered interface is used while the socket is open. If there is another type of pending error event, it would be

         * notified once we run out of zerocopy notifications. */

        if ((events[i].events & EPOLLERR) != 0 && sock->super._zerocopy != NULL && handle_zerocopy_notification(sock)) {

            events[i].events &= ~EPOLLERR;

            notified = 1;

        }

        /* Handle read and write events. */

        if ((events[i].events & (EPOLLIN | EPOLLHUP | EPOLLERR)) != 0) {

            if ((sock->_flags & H2O_SOCKET_FLAG_IS_POLLED_FOR_READ) != 0) {

                sock->_flags |= H2O_SOCKET_FLAG_IS_READ_READY;

                link_to_pending(sock);

                notified = 1;

            }

        }

        if ((events[i].events & (EPOLLOUT | EPOLLHUP | EPOLLERR)) != 0) {

            if ((sock->_flags & H2O_SOCKET_FLAG_IS_POLLED_FOR_WRITE) != 0) {

                write_pending(sock);

                notified = 1;

            }

        }

        /* Report events that could be notified, as that would help us debug issues. This mechanism is disabled once the socket is

         * closed, as there will be misfires due to the nature of edge triggers (race between us draining between events queued up).

         */

        if (!notified && (sock->_flags & H2O_SOCKET_FLAG_IS_DISPOSED) == 0) {

            static pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER;

            static time_t last_reported = 0;

            time_t now = time(NULL);

            pthread_mutex_lock(&lock);

            if (last_reported + 60 < now) {

                last_reported = now;

                h2o_error_printf("ignoring epoll event (fd:%d,event:0x%x,flags:0x%x)\n", sock->fd, (int)events[i].events,

                                 sock->_flags);

            }

            pthread_mutex_unlock(&lock);

        }

    }

    return 0;

}

static void evloop_do_on_socket_create(struct st_h2o_evloop_socket_t *sock)

{

}

static int evloop_do_on_socket_close(struct st_h2o_evloop_socket_t *sock)

{

    int ret;

    /* Nothing to do if fd has been detached already. */

    if (sock->fd == -1)

        return 0;

    /* If zero copy is in action, disconnect using shutdown(). Then, poll the socket until all zero copy buffers are reclaimed, at

     * which point we dispose of the socket. Edge trigger is used, as in level trigger EPOLLHUP will be notified continuously. */

    if (sock->super._zerocopy != NULL && !zerocopy_buffers_is_empty(sock->super._zerocopy)) {

        while ((ret = shutdown(sock->fd, SHUT_RDWR)) == -1 && errno == EINTR)

            ;

        if (ret != 0 && errno != ENOTCONN)

            h2o_error_printf("socket_close: shutdown(SHUT_RDWR) failed; errno=%d, fd=%d\n", errno, sock->fd);

        if (!change_epoll_mode(sock, EPOLLET))

            h2o_fatal("socket_close: epoll_ctl(MOD) failed; errno=%d, fd=%d\n", errno, sock->fd);

        /* drain error notifications after registering the edge trigger, otherwise there's chance of stall */

        handle_zerocopy_notification(sock);

        return 1;

    }

    /* Unregister from epoll. */

    if (!delete_from_epoll_on_close(sock))

        h2o_error_printf("socket_close: epoll(DEL) returned error %d (fd=%d)\n", errno, sock->fd);

    return 0;

}

static void evloop_do_on_socket_export(struct st_h2o_evloop_socket_t *sock)

{

    if (!delete_from_epoll_on_close(sock))

        h2o_error_printf("socket_export: epoll(DEL) returned error %d (fd=%d)\n", errno, sock->fd);

}

static void evloop_do_dispose(h2o_evloop_t *_loop)

{

    struct st_h2o_evloop_epoll_t *loop = (struct st_h2o_evloop_epoll_t *)_loop;

    close(loop->ep);

}

h2o_evloop_t *h2o_evloop_create(void)

{

    struct st_h2o_evloop_epoll_t *loop = (struct st_h2o_evloop_epoll_t *)create_evloop(sizeof(*loop));

    if ((loop->ep = epoll_create1(EPOLL_CLOEXEC)) == -1) {

        char buf[128];

        h2o_fatal("h2o_evloop_create: epoll_create1 failed:%d:%s\n", errno, h2o_strerror_r(errno, buf, sizeof(buf)));

    }

#if H2O_USE_IO_URING

    h2o_io_uring_init(&loop->super);

#endif

    return &loop->super;

}

#if H2O_USE_IO_URING

struct st_h2o_io_uring_t *h2o_evloop__io_uring(h2o_evloop_t *_loop)

{

    struct st_h2o_evloop_epoll_t *loop = (struct st_h2o_evloop_epoll_t *)_loop;

    return &loop->io_uring;

}

#endif