Make the receiving thread non blocking

[lttng-tools.git] / liblttkconsumerd / liblttkconsumerd.c

/*
 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
 *                      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; only version 2
 * of the License.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
 */

#define _GNU_SOURCE
#include <fcntl.h>
#include <poll.h>
#include <pthread.h>
#include <stdlib.h>
#include <string.h>
#include <sys/mman.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <unistd.h>
#include <urcu/list.h>

#include "libkernelctl.h"
#include "liblttkconsumerd.h"
#include "lttngerr.h"

static
struct kconsumerd_global_data {
        /*
         * kconsumerd_data.lock protects kconsumerd_data.fd_list,
         * kconsumerd_data.fds_count, and kconsumerd_data.need_update.  It
         * ensures the count matches the number of items in the fd_list.
         * It ensures the list updates *always* trigger an fd_array
         * update (therefore need to make list update vs
         * kconsumerd_data.need_update flag update atomic, and also flag
         * read, fd array and flag clear atomic).
         */
        pthread_mutex_t lock;
        /*
         * Number of element for the list below.  Protected by
         * kconsumerd_data.lock.
         */
        unsigned int fds_count;
        /*
         * List of FDs.  Protected by kconsumerd_data.lock.
         */
        struct kconsumerd_fd_list fd_list;
        /*
         * Flag specifying if the local array of FDs needs update in the
         * poll function.  Protected by kconsumerd_data.lock.
         */
        unsigned int need_update;
} kconsumerd_data = {
        .fd_list.head = CDS_LIST_HEAD_INIT(kconsumerd_data.fd_list.head),
};

/* communication with splice */
static int kconsumerd_thread_pipe[2];

/* pipe to wake the poll thread when necessary */
static int kconsumerd_poll_pipe[2];

/* to let the signal handler wake up the fd receiver thread */
static int kconsumerd_should_quit[2];

/* timeout parameter, to control the polling thread grace period */
static int kconsumerd_poll_timeout = -1;

/* socket to communicate errors with sessiond */
static int kconsumerd_error_socket;

/* socket to exchange commands with sessiond */
static char *kconsumerd_command_sock_path;

/*
 * flag to inform the polling thread to quit when all fd hung up.
 * Updated by the kconsumerd_thread_receive_fds when it notices that all
 * fds has hung up. Also updated by the signal handler
 * (kconsumerd_should_exit()). Read by the polling threads.
 */
static volatile int kconsumerd_quit = 0;

/*
 * kconsumerd_set_error_socket
 *
 * Set the error socket
 */
void kconsumerd_set_error_socket(int sock)
{
        kconsumerd_error_socket = sock;
}

/*
 * kconsumerd_set_command_socket_path
 *
 * Set the command socket path
 */
void kconsumerd_set_command_socket_path(char *sock)
{
        kconsumerd_command_sock_path = sock;
}

/*
 * kconsumerd_find_session_fd
 *
 * Find a session fd in the global list.
 * The kconsumerd_data.lock must be locked during this call
 *
 * Return 1 if found else 0
 */
static int kconsumerd_find_session_fd(int fd)
{
        struct kconsumerd_fd *iter;

        cds_list_for_each_entry(iter, &kconsumerd_data.fd_list.head, list) {
                if (iter->sessiond_fd == fd) {
                        DBG("Duplicate session fd %d", fd);
                        pthread_mutex_unlock(&kconsumerd_data.lock);
                        return 1;
                }
        }

        return 0;
}

/*
 * kconsumerd_del_fd
 *
 * Remove a fd from the global list protected by a mutex
 */
static void kconsumerd_del_fd(struct kconsumerd_fd *lcf)
{
        pthread_mutex_lock(&kconsumerd_data.lock);
        cds_list_del(&lcf->list);
        if (kconsumerd_data.fds_count > 0) {
                kconsumerd_data.fds_count--;
                if (lcf != NULL) {
                        close(lcf->out_fd);
                        close(lcf->consumerd_fd);
                        free(lcf);
                        lcf = NULL;
                }
        }
        kconsumerd_data.need_update = 1;
        pthread_mutex_unlock(&kconsumerd_data.lock);
}

/*
 * kconsumerd_add_fd
 *
 * Add a fd to the global list protected by a mutex
 */
static int kconsumerd_add_fd(struct lttcomm_kconsumerd_msg *buf, int consumerd_fd)
{
        int ret;
        struct kconsumerd_fd *tmp_fd;

        pthread_mutex_lock(&kconsumerd_data.lock);
        /* Check if already exist */
        ret = kconsumerd_find_session_fd(buf->fd);
        if (ret == 1) {
                goto end;
        }

        tmp_fd = malloc(sizeof(struct kconsumerd_fd));
        tmp_fd->sessiond_fd = buf->fd;
        tmp_fd->consumerd_fd = consumerd_fd;
        tmp_fd->state = buf->state;
        tmp_fd->max_sb_size = buf->max_sb_size;
        strncpy(tmp_fd->path_name, buf->path_name, PATH_MAX);

        /* Opening the tracefile in write mode */
        ret = open(tmp_fd->path_name,
                        O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO);
        if (ret < 0) {
                ERR("Opening %s", tmp_fd->path_name);
                perror("open");
                goto end;
        }
        tmp_fd->out_fd = ret;
        tmp_fd->out_fd_offset = 0;

        DBG("Adding %s (%d, %d, %d)", tmp_fd->path_name,
                        tmp_fd->sessiond_fd, tmp_fd->consumerd_fd, tmp_fd->out_fd);

        cds_list_add(&tmp_fd->list, &kconsumerd_data.fd_list.head);
        kconsumerd_data.fds_count++;
        kconsumerd_data.need_update = 1;
end:
        pthread_mutex_unlock(&kconsumerd_data.lock);
        return ret;
}

/*
 * kconsumerd_change_fd_state
 *
 * Update a fd according to what we just received
 */
static void kconsumerd_change_fd_state(int sessiond_fd,
                enum kconsumerd_fd_state state)
{
        struct kconsumerd_fd *iter;

        pthread_mutex_lock(&kconsumerd_data.lock);
        cds_list_for_each_entry(iter, &kconsumerd_data.fd_list.head, list) {
                if (iter->sessiond_fd == sessiond_fd) {
                        iter->state = state;
                        break;
                }
        }
        kconsumerd_data.need_update = 1;
        pthread_mutex_unlock(&kconsumerd_data.lock);
}

/*
 * kconsumerd_update_poll_array
 *
 * Allocate the pollfd structure and the local view of the out fds
 * to avoid doing a lookup in the linked list and concurrency issues
 * when writing is needed.
 * Returns the number of fds in the structures
 * Called with kconsumerd_data.lock held.
 */
static int kconsumerd_update_poll_array(struct pollfd **pollfd,
                struct kconsumerd_fd **local_kconsumerd_fd)
{
        struct kconsumerd_fd *iter;
        int i = 0;

        DBG("Updating poll fd array");

        cds_list_for_each_entry(iter, &kconsumerd_data.fd_list.head, list) {
                DBG("Inside for each");
                if (iter->state == ACTIVE_FD) {
                        DBG("Active FD %d", iter->consumerd_fd);
                        (*pollfd)[i].fd = iter->consumerd_fd;
                        (*pollfd)[i].events = POLLIN | POLLPRI;
                        local_kconsumerd_fd[i] = iter;
                        i++;
                }
        }

        /*
         * insert the kconsumerd_poll_pipe at the end of the array and don't
         * increment i so nb_fd is the number of real FD
         */
        (*pollfd)[i].fd = kconsumerd_poll_pipe[0];
        (*pollfd)[i].events = POLLIN;
        return i;
}


/*
 * kconsumerd_on_read_subbuffer_mmap
 *
 * mmap the ring buffer, read it and write the data to the tracefile.
 * Returns the number of bytes written
 */
static int kconsumerd_on_read_subbuffer_mmap(
                struct kconsumerd_fd *kconsumerd_fd, unsigned long len)
{
        unsigned long mmap_len, mmap_offset, padded_len, padding_len;
        char *mmap_base;
        char *padding = NULL;
        long ret = 0;
        off_t orig_offset = kconsumerd_fd->out_fd_offset;
        int fd = kconsumerd_fd->consumerd_fd;
        int outfd = kconsumerd_fd->out_fd;

        /* get the padded subbuffer size to know the padding required */
        ret = kernctl_get_padded_subbuf_size(fd, &padded_len);
        if (ret != 0) {
                ret = errno;
                perror("kernctl_get_padded_subbuf_size");
                goto end;
        }
        padding_len = padded_len - len;
        padding = malloc(padding_len * sizeof(char));
        memset(padding, '\0', padding_len);

        /* get the len of the mmap region */
        ret = kernctl_get_mmap_len(fd, &mmap_len);
        if (ret != 0) {
                ret = errno;
                perror("kernctl_get_mmap_len");
                goto end;
        }

        /* get the offset inside the fd to mmap */
        ret = kernctl_get_mmap_read_offset(fd, &mmap_offset);
        if (ret != 0) {
                ret = errno;
                perror("kernctl_get_mmap_read_offset");
                goto end;
        }

        mmap_base = mmap(NULL, mmap_len, PROT_READ, MAP_PRIVATE, fd, mmap_offset);
        if (mmap_base == MAP_FAILED) {
                perror("Error mmaping");
                ret = -1;
                goto end;
        }

        while (len > 0) {
                ret = write(outfd, mmap_base, len);
                if (ret >= len) {
                        len = 0;
                } else if (ret < 0) {
                        ret = errno;
                        perror("Error in file write");
                        goto end;
                }
                /* This won't block, but will start writeout asynchronously */
                sync_file_range(outfd, kconsumerd_fd->out_fd_offset, ret,
                                SYNC_FILE_RANGE_WRITE);
                kconsumerd_fd->out_fd_offset += ret;
        }

        /* once all the data is written, write the padding to disk */
        ret = write(outfd, padding, padding_len);
        if (ret < 0) {
                ret = errno;
                perror("Error writing padding to file");
                goto end;
        }

        /*
         * This does a blocking write-and-wait on any page that belongs to the
         * subbuffer prior to the one we just wrote.
         * Don't care about error values, as these are just hints and ways to
         * limit the amount of page cache used.
         */
        if (orig_offset >= kconsumerd_fd->max_sb_size) {
                sync_file_range(outfd, orig_offset - kconsumerd_fd->max_sb_size,
                                kconsumerd_fd->max_sb_size,
                                SYNC_FILE_RANGE_WAIT_BEFORE
                                | SYNC_FILE_RANGE_WRITE
                                | SYNC_FILE_RANGE_WAIT_AFTER);

                /*
                 * Give hints to the kernel about how we access the file:
                 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
                 * we write it.
                 *
                 * We need to call fadvise again after the file grows because the
                 * kernel does not seem to apply fadvise to non-existing parts of the
                 * file.
                 *
                 * Call fadvise _after_ having waited for the page writeback to
                 * complete because the dirty page writeback semantic is not well
                 * defined. So it can be expected to lead to lower throughput in
                 * streaming.
                 */
                posix_fadvise(outfd, orig_offset - kconsumerd_fd->max_sb_size,
                                kconsumerd_fd->max_sb_size, POSIX_FADV_DONTNEED);
        }
        goto end;

end:
        if (padding != NULL) {
                free(padding);
        }
        return ret;
}

/*
 * kconsumerd_on_read_subbuffer
 *
 * Splice the data from the ring buffer to the tracefile.
 * Returns the number of bytes spliced
 */
static int kconsumerd_on_read_subbuffer(
                struct kconsumerd_fd *kconsumerd_fd, unsigned long len)
{
        long ret = 0;
        loff_t offset = 0;
        off_t orig_offset = kconsumerd_fd->out_fd_offset;
        int fd = kconsumerd_fd->consumerd_fd;
        int outfd = kconsumerd_fd->out_fd;

        while (len > 0) {
                DBG("splice chan to pipe offset %lu (fd : %d)",
                                (unsigned long)offset, fd);
                ret = splice(fd, &offset, kconsumerd_thread_pipe[1], NULL, len,
                                SPLICE_F_MOVE | SPLICE_F_MORE);
                DBG("splice chan to pipe ret %ld", ret);
                if (ret < 0) {
                        ret = errno;
                        perror("Error in relay splice");
                        goto splice_error;
                }

                ret = splice(kconsumerd_thread_pipe[0], NULL, outfd, NULL, ret,
                                SPLICE_F_MOVE | SPLICE_F_MORE);
                DBG("splice pipe to file %ld", ret);
                if (ret < 0) {
                        ret = errno;
                        perror("Error in file splice");
                        goto splice_error;
                }
                if (ret >= len) {
                        len = 0;
                }
                /* This won't block, but will start writeout asynchronously */
                sync_file_range(outfd, kconsumerd_fd->out_fd_offset, ret,
                                SYNC_FILE_RANGE_WRITE);
                kconsumerd_fd->out_fd_offset += ret;
        }

        /*
         * This does a blocking write-and-wait on any page that belongs to the
         * subbuffer prior to the one we just wrote.
         * Don't care about error values, as these are just hints and ways to
         * limit the amount of page cache used.
         */
        if (orig_offset >= kconsumerd_fd->max_sb_size) {
                sync_file_range(outfd, orig_offset - kconsumerd_fd->max_sb_size,
                                kconsumerd_fd->max_sb_size,
                                SYNC_FILE_RANGE_WAIT_BEFORE
                                | SYNC_FILE_RANGE_WRITE
                                | SYNC_FILE_RANGE_WAIT_AFTER);
                /*
                 * Give hints to the kernel about how we access the file:
                 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
                 * we write it.
                 *
                 * We need to call fadvise again after the file grows because the
                 * kernel does not seem to apply fadvise to non-existing parts of the
                 * file.
                 *
                 * Call fadvise _after_ having waited for the page writeback to
                 * complete because the dirty page writeback semantic is not well
                 * defined. So it can be expected to lead to lower throughput in
                 * streaming.
                 */
                posix_fadvise(outfd, orig_offset - kconsumerd_fd->max_sb_size,
                                kconsumerd_fd->max_sb_size, POSIX_FADV_DONTNEED);
        }
        goto end;

splice_error:
        /* send the appropriate error description to sessiond */
        switch(ret) {
        case EBADF:
                kconsumerd_send_error(KCONSUMERD_SPLICE_EBADF);
                break;
        case EINVAL:
                kconsumerd_send_error(KCONSUMERD_SPLICE_EINVAL);
                break;
        case ENOMEM:
                kconsumerd_send_error(KCONSUMERD_SPLICE_ENOMEM);
                break;
        case ESPIPE:
                kconsumerd_send_error(KCONSUMERD_SPLICE_ESPIPE);
                break;
        }

end:
        return ret;
}

/*
 * kconsumerd_read_subbuffer
 *
 * Consume data on a file descriptor and write it on a trace file
 */
static int kconsumerd_read_subbuffer(struct kconsumerd_fd *kconsumerd_fd)
{
        unsigned long len;
        int err;
        long ret = 0;
        int infd = kconsumerd_fd->consumerd_fd;

        DBG("In kconsumerd_read_subbuffer (infd : %d)", infd);
        /* Get the next subbuffer */
        err = kernctl_get_next_subbuf(infd);
        if (err != 0) {
                ret = errno;
                perror("Reserving sub buffer failed (everything is normal, "
                                "it is due to concurrency)");
                goto end;
        }

        switch (DEFAULT_KERNEL_CHANNEL_OUTPUT) {
        case LTTNG_EVENT_SPLICE:
                /* read the whole subbuffer */
                err = kernctl_get_padded_subbuf_size(infd, &len);
                if (err != 0) {
                        ret = errno;
                        perror("Getting sub-buffer len failed.");
                        goto end;
                }

                /* splice the subbuffer to the tracefile */
                ret = kconsumerd_on_read_subbuffer(kconsumerd_fd, len);
                if (ret < 0) {
                        /*
                         * display the error but continue processing to try
                         * to release the subbuffer
                         */
                        ERR("Error splicing to tracefile");
                }
                break;
        case LTTNG_EVENT_MMAP:
                /* read the used subbuffer size */
                err = kernctl_get_subbuf_size(infd, &len);
                if (err != 0) {
                        ret = errno;
                        perror("Getting sub-buffer len failed.");
                        goto end;
                }
                /* write the subbuffer to the tracefile */
                ret = kconsumerd_on_read_subbuffer_mmap(kconsumerd_fd, len);
                if (ret < 0) {
                        /*
                         * display the error but continue processing to try
                         * to release the subbuffer
                         */
                        ERR("Error writing to tracefile");
                }
                break;
        default:
                ERR("Unknown output method");
                ret = -1;
        }

        err = kernctl_put_next_subbuf(infd);
        if (err != 0) {
                ret = errno;
                if (errno == EFAULT) {
                        perror("Error in unreserving sub buffer\n");
                } else if (errno == EIO) {
                        /* Should never happen with newer LTTng versions */
                        perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
                }
                goto end;
        }

end:
        return ret;
}

/*
 * kconsumerd_poll_socket
 *
 * Poll on the should_quit pipe and the command socket
 * return -1 on error and should exit, 0 if data is
 * available on the command socket
 */
int kconsumerd_poll_socket(struct pollfd *kconsumerd_sockpoll)
{
        int num_rdy;

        num_rdy = poll(kconsumerd_sockpoll, 2, -1);
        if (num_rdy == -1) {
                perror("Poll error");
                goto exit;
        }
        if (kconsumerd_sockpoll[0].revents == POLLIN) {
                DBG("kconsumerd_should_quit wake up");
                goto exit;
        }
        return 0;

exit:
        return -1;
}

/*
 * kconsumerd_consumerd_recv_fd
 *
 * Receives an array of file descriptors and the associated
 * structures describing each fd (path name).
 * Returns the size of received data
 */
static int kconsumerd_consumerd_recv_fd(int sfd,
                struct pollfd *kconsumerd_sockpoll, int size,
                enum kconsumerd_command cmd_type)
{
        struct msghdr msg;
        struct iovec iov[1];
        int ret = 0, i, tmp2;
        struct cmsghdr *cmsg;
        int nb_fd;
        char recv_fd[CMSG_SPACE(sizeof(int))];
        struct lttcomm_kconsumerd_msg lkm;

        /* the number of fds we are about to receive */
        nb_fd = size / sizeof(struct lttcomm_kconsumerd_msg);

        for (i = 0; i < nb_fd; i++) {
                memset(&msg, 0, sizeof(msg));

                /* Prepare to receive the structures */
                iov[0].iov_base = &lkm;
                iov[0].iov_len = sizeof(lkm);
                msg.msg_iov = iov;
                msg.msg_iovlen = 1;

                msg.msg_control = recv_fd;
                msg.msg_controllen = sizeof(recv_fd);

                DBG("Waiting to receive fd");
                if (kconsumerd_poll_socket(kconsumerd_sockpoll) < 0) {
                        goto end;
                }

                if ((ret = recvmsg(sfd, &msg, 0)) < 0) {
                        perror("recvmsg");
                        continue;
                }

                if (ret != (size / nb_fd)) {
                        ERR("Received only %d, expected %d", ret, size);
                        kconsumerd_send_error(KCONSUMERD_ERROR_RECV_FD);
                        goto end;
                }

                cmsg = CMSG_FIRSTHDR(&msg);
                if (!cmsg) {
                        ERR("Invalid control message header");
                        ret = -1;
                        kconsumerd_send_error(KCONSUMERD_ERROR_RECV_FD);
                        goto end;
                }
                /* if we received fds */
                if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
                        switch (cmd_type) {
                        case ADD_STREAM:
                                DBG("kconsumerd_add_fd %s (%d)", lkm.path_name, (CMSG_DATA(cmsg)[0]));
                                ret = kconsumerd_add_fd(&lkm, (CMSG_DATA(cmsg)[0]));
                                if (ret < 0) {
                                        kconsumerd_send_error(KCONSUMERD_OUTFD_ERROR);
                                        goto end;
                                }
                                break;
                        case UPDATE_STREAM:
                                kconsumerd_change_fd_state(lkm.fd, lkm.state);
                                break;
                        default:
                                break;
                        }
                        /* signal the poll thread */
                        tmp2 = write(kconsumerd_poll_pipe[1], "4", 1);
                } else {
                        ERR("Didn't received any fd");
                        kconsumerd_send_error(KCONSUMERD_ERROR_RECV_FD);
                        ret = -1;
                        goto end;
                }
        }

end:
        return ret;
}

/*
 *  kconsumerd_thread_poll_fds
 *
 *  This thread polls the fds in the ltt_fd_list to consume the data
 *  and write it to tracefile if necessary.
 */
void *kconsumerd_thread_poll_fds(void *data)
{
        int num_rdy, num_hup, high_prio, ret, i;
        struct pollfd *pollfd = NULL;
        /* local view of the fds */
        struct kconsumerd_fd **local_kconsumerd_fd = NULL;
        /* local view of kconsumerd_data.fds_count */
        int nb_fd = 0;
        char tmp;
        int tmp2;

        ret = pipe(kconsumerd_thread_pipe);
        if (ret < 0) {
                perror("Error creating pipe");
                goto end;
        }

        local_kconsumerd_fd = malloc(sizeof(struct kconsumerd_fd));

        while (1) {
                high_prio = 0;
                num_hup = 0;

                /*
                 * the ltt_fd_list has been updated, we need to update our
                 * local array as well
                 */
                pthread_mutex_lock(&kconsumerd_data.lock);
                if (kconsumerd_data.need_update) {
                        if (pollfd != NULL) {
                                free(pollfd);
                                pollfd = NULL;
                        }
                        if (local_kconsumerd_fd != NULL) {
                                free(local_kconsumerd_fd);
                                local_kconsumerd_fd = NULL;
                        }

                        /* allocate for all fds + 1 for the kconsumerd_poll_pipe */
                        pollfd = malloc((kconsumerd_data.fds_count + 1) * sizeof(struct pollfd));
                        if (pollfd == NULL) {
                                perror("pollfd malloc");
                                pthread_mutex_unlock(&kconsumerd_data.lock);
                                goto end;
                        }

                        /* allocate for all fds + 1 for the kconsumerd_poll_pipe */
                        local_kconsumerd_fd = malloc((kconsumerd_data.fds_count + 1) *
                                        sizeof(struct kconsumerd_fd));
                        if (local_kconsumerd_fd == NULL) {
                                perror("local_kconsumerd_fd malloc");
                                pthread_mutex_unlock(&kconsumerd_data.lock);
                                goto end;
                        }
                        ret = kconsumerd_update_poll_array(&pollfd, local_kconsumerd_fd);
                        if (ret < 0) {
                                ERR("Error in allocating pollfd or local_outfds");
                                kconsumerd_send_error(KCONSUMERD_POLL_ERROR);
                                pthread_mutex_unlock(&kconsumerd_data.lock);
                                goto end;
                        }
                        nb_fd = ret;
                        kconsumerd_data.need_update = 0;
                }
                pthread_mutex_unlock(&kconsumerd_data.lock);

                /* poll on the array of fds */
                DBG("polling on %d fd", nb_fd + 1);
                num_rdy = poll(pollfd, nb_fd + 1, kconsumerd_poll_timeout);
                DBG("poll num_rdy : %d", num_rdy);
                if (num_rdy == -1) {
                        perror("Poll error");
                        kconsumerd_send_error(KCONSUMERD_POLL_ERROR);
                        goto end;
                } else if (num_rdy == 0) {
                        DBG("Polling thread timed out");
                        goto end;
                }

                /* No FDs and kconsumerd_quit, kconsumerd_cleanup the thread */
                if (nb_fd == 0 && kconsumerd_quit == 1) {
                        goto end;
                }

                /*
                 * If the kconsumerd_poll_pipe triggered poll go
                 * directly to the beginning of the loop to update the
                 * array. We want to prioritize array update over
                 * low-priority reads.
                 */
                if (pollfd[nb_fd].revents == POLLIN) {
                        DBG("kconsumerd_poll_pipe wake up");
                        tmp2 = read(kconsumerd_poll_pipe[0], &tmp, 1);
                        continue;
                }

                /* Take care of high priority channels first. */
                for (i = 0; i < nb_fd; i++) {
                        switch(pollfd[i].revents) {
                        case POLLERR:
                                ERR("Error returned in polling fd %d.", pollfd[i].fd);
                                kconsumerd_del_fd(local_kconsumerd_fd[i]);
                                num_hup++;
                                break;
                        case POLLHUP:
                                DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
                                kconsumerd_del_fd(local_kconsumerd_fd[i]);
                                num_hup++;
                                break;
                        case POLLNVAL:
                                ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
                                kconsumerd_del_fd(local_kconsumerd_fd[i]);
                                num_hup++;
                                break;
                        case POLLPRI:
                                DBG("Urgent read on fd %d", pollfd[i].fd);
                                high_prio = 1;
                                ret = kconsumerd_read_subbuffer(local_kconsumerd_fd[i]);
                                /* it's ok to have an unavailable sub-buffer */
                                if (ret == EAGAIN) {
                                        ret = 0;
                                }
                                break;
                        }
                }

                /* If every buffer FD has hung up, we end the read loop here */
                if (nb_fd > 0 && num_hup == nb_fd) {
                        DBG("every buffer FD has hung up\n");
                        if (kconsumerd_quit == 1) {
                                goto end;
                        }
                        continue;
                }

                /* Take care of low priority channels. */
                if (high_prio == 0) {
                        for (i = 0; i < nb_fd; i++) {
                                if (pollfd[i].revents == POLLIN) {
                                        DBG("Normal read on fd %d", pollfd[i].fd);
                                        ret = kconsumerd_read_subbuffer(local_kconsumerd_fd[i]);
                                        /* it's ok to have an unavailable subbuffer */
                                        if (ret == EAGAIN) {
                                                ret = 0;
                                        }
                                }
                        }
                }
        }
end:
        DBG("polling thread exiting");
        if (pollfd != NULL) {
                free(pollfd);
                pollfd = NULL;
        }
        if (local_kconsumerd_fd != NULL) {
                free(local_kconsumerd_fd);
                local_kconsumerd_fd = NULL;
        }
        return NULL;
}

/*
 * kconsumerd_init(void)
 *
 * initialise the necessary environnement :
 * - inform the polling thread to update the polling array
 * - create the poll_pipe
 * - create the should_quit pipe (for signal handler)
 */
int kconsumerd_init(void)
{
        int ret;

        /* need to update the polling array at init time */
        kconsumerd_data.need_update = 1;

        ret = pipe(kconsumerd_poll_pipe);
        if (ret < 0) {
                perror("Error creating poll pipe");
                goto end;
        }

        ret = pipe(kconsumerd_should_quit);
        if (ret < 0) {
                perror("Error creating recv pipe");
                goto end;
        }

end:
        return ret;
}

/*
 *  kconsumerd_thread_receive_fds
 *
 *  This thread listens on the consumerd socket and
 *  receives the file descriptors from ltt-sessiond
 */
void *kconsumerd_thread_receive_fds(void *data)
{
        int sock, client_socket, ret;
        struct lttcomm_kconsumerd_header tmp;
        /*
         * structure to poll for incoming data on communication socket
         * avoids making blocking sockets
         */
        struct pollfd kconsumerd_sockpoll[2];


        DBG("Creating command socket %s", kconsumerd_command_sock_path);
        unlink(kconsumerd_command_sock_path);
        client_socket = lttcomm_create_unix_sock(kconsumerd_command_sock_path);
        if (client_socket < 0) {
                ERR("Cannot create command socket");
                goto end;
        }

        ret = lttcomm_listen_unix_sock(client_socket);
        if (ret < 0) {
                goto end;
        }

        DBG("Sending ready command to ltt-sessiond");
        ret = kconsumerd_send_error(KCONSUMERD_COMMAND_SOCK_READY);
        if (ret < 0) {
                ERR("Error sending ready command to ltt-sessiond");
                goto end;
        }

        ret = fcntl(client_socket, F_SETFL, O_NONBLOCK);
        if (ret < 0) {
                perror("fcntl O_NONBLOCK");
                goto end;
        }

        /* prepare the FDs to poll : to client socket and the should_quit pipe */
        kconsumerd_sockpoll[0].fd = kconsumerd_should_quit[0];
        kconsumerd_sockpoll[0].events = POLLIN | POLLPRI;
        kconsumerd_sockpoll[1].fd = client_socket;
        kconsumerd_sockpoll[1].events = POLLIN | POLLPRI;

        if (kconsumerd_poll_socket(kconsumerd_sockpoll) < 0) {
                goto end;
        }
        DBG("Connection on client_socket");

        /* Blocking call, waiting for transmission */
        sock = lttcomm_accept_unix_sock(client_socket);
        if (sock <= 0) {
                WARN("On accept");
                goto end;
        }
        ret = fcntl(sock, F_SETFL, O_NONBLOCK);
        if (ret < 0) {
                perror("fcntl O_NONBLOCK");
                goto end;
        }

        /* update the polling structure to poll on the established socket */
        kconsumerd_sockpoll[1].fd = sock;
        kconsumerd_sockpoll[1].events = POLLIN | POLLPRI;

        while (1) {
                if (kconsumerd_poll_socket(kconsumerd_sockpoll) < 0) {
                        goto end;
                }
                DBG("Incoming fds on sock");

                /* We first get the number of fd we are about to receive */
                ret = lttcomm_recv_unix_sock(sock, &tmp,
                                sizeof(struct lttcomm_kconsumerd_header));
                if (ret <= 0) {
                        ERR("Communication interrupted on command socket");
                        goto end;
                }
                if (tmp.cmd_type == STOP) {
                        DBG("Received STOP command");
                        goto end;
                }
                if (kconsumerd_quit) {
                        DBG("kconsumerd_thread_receive_fds received quit from signal");
                        goto end;
                }

                /* we received a command to add or update fds */
                ret = kconsumerd_consumerd_recv_fd(sock, kconsumerd_sockpoll,
                                tmp.payload_size, tmp.cmd_type);
                if (ret <= 0) {
                        ERR("Receiving the FD, exiting");
                        goto end;
                }
                DBG("received fds on sock");
        }

end:
        DBG("kconsumerd_thread_receive_fds exiting");

        /*
         * when all fds have hung up, the polling thread
         * can exit cleanly
         */
        kconsumerd_quit = 1;

        /*
         * 2s of grace period, if no polling events occur during
         * this period, the polling thread will exit even if there
         * are still open FDs (should not happen, but safety mechanism).
         */
        kconsumerd_poll_timeout = KCONSUMERD_POLL_GRACE_PERIOD;

        /* wake up the polling thread */
        ret = write(kconsumerd_poll_pipe[1], "4", 1);
        if (ret < 0) {
                perror("poll pipe write");
        }
        return NULL;
}

/*
 *  kconsumerd_cleanup
 *
 *  Cleanup the daemon's socket on exit
 */
void kconsumerd_cleanup(void)
{
        struct kconsumerd_fd *iter;

        /* remove the socket file */
        unlink(kconsumerd_command_sock_path);

        /*
         * close all outfd. Called when there are no more threads
         * running (after joining on the threads), no need to protect
         * list iteration with mutex.
         */
        cds_list_for_each_entry(iter, &kconsumerd_data.fd_list.head, list) {
                kconsumerd_del_fd(iter);
        }
}

/*
 * kconsumerd_should_exit
 *
 * Called from signal handler.
 */
void kconsumerd_should_exit(void)
{
        int ret;
        kconsumerd_quit = 1;
        ret = write(kconsumerd_should_quit[1], "4", 1);
}

/*
 * kconsumerd_send_error
 *
 * send return code to ltt-sessiond
 */
int kconsumerd_send_error(enum lttcomm_return_code cmd)
{
        if (kconsumerd_error_socket > 0) {
                return lttcomm_send_unix_sock(kconsumerd_error_socket, &cmd,
                                sizeof(enum lttcomm_sessiond_command));
        }

        return 0;
}