#include <stdio.h>
#include <fcntl.h>
#include <unistd.h>
#include <stdlib.h>
#include <inttypes.h>
#include <sys/time.h>
#include <sys/uio.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <sys/sysmacros.h>
#include <dirent.h>
#include <string.h>
#include <pthread.h>
#include <errno.h>
#include <numa.h>

#define RWF_UNCACHED	0x80

static int random_io = 0;

static unsigned long long mtime_since(const struct timeval *s,
				      const struct timeval *e)
{
	long long sec, usec;

	sec = e->tv_sec - s->tv_sec;
	usec = (e->tv_usec - s->tv_usec);
	if (sec > 0 && usec < 0) {
		sec--;
		usec += 1000000;
	}

	sec *= 1000;
	usec /= 1000;
	return sec + usec;
}

static unsigned long long mtime_since_now(struct timeval *tv)
{
	struct timeval end;

	gettimeofday(&end, NULL);
	return mtime_since(tv, &end);
}

struct thread_data {
	pthread_t thread;
	pthread_barrier_t barrier;
	int fd;
	volatile int done;
	int rwf_flags;
	int bs;
	int reading;
	void *buf;
	unsigned long max_blocks;
	unsigned long bytes;
};

static off_t next_off(struct thread_data *d, off_t last_off, int bytes)
{
	if (random_io) {
		long off = lrand48();
		long block = off % d->max_blocks;
		return block * d->bs;
	}
	return last_off + bytes;
}

static void *thread_fn(void *__data)
{
	struct thread_data *d = __data;
	struct iovec iov = { .iov_len = d->bs };
	const char *msg = d->reading ? "preadv2" : "pwritev2";
	off_t offset = 0;
	struct stat sb;

	if (fstat(d->fd, &sb) < 0) {
		perror("fstat");
		goto done;
	}
	if (random_io)
		d->max_blocks = sb.st_size / d->bs;

	if (posix_memalign(&d->buf, 4096, d->bs))
		goto done;
	iov.iov_base = d->buf;

	pthread_barrier_wait(&d->barrier);

	for (;;) {
		int ret;

		if (d->reading)
			ret = preadv2(d->fd, &iov, 1, offset, d->rwf_flags);
		else
			ret = pwritev2(d->fd, &iov, 1, offset, d->rwf_flags);
		if (ret < 0) {
			perror(msg);
			break;
		} else if (!ret) {
			break;
		}

		d->bytes += ret;
		offset = next_off(d, offset, ret);
	}
done:
	d->done = 1;
	return NULL;
}

int main(int argc, char *argv[])
{
	struct timeval tv, start;
	unsigned int bs;
	unsigned long last_bytes = 0, bytes = 0;
	struct thread_data *td;
	int rwf_flags = 0;
	int reading = 1;
	int nr, i;

	if (argc < 5) {
		printf("%s: bs uncached reading <files>\n", argv[0]);
		return 1;
	}

	bs = atoi(argv[1]);
	if (!!atoi(argv[2]))
		rwf_flags |= RWF_UNCACHED;
	reading = !!atoi(argv[3]);

	nr = argc - 4;
	if (!nr) {
		fprintf(stderr, "Need files\n");
		return 1;
	}

	if (!reading)
		random_io = 0;

	td = calloc(nr, sizeof(struct thread_data));

	for (i = 0; i < nr; i++) {
		int flags;

		if (reading)
			flags = O_RDONLY;
		else
			flags = O_WRONLY;
again:
		td[i].fd = open(argv[4 + i], flags, 0644);
		if (td[i].fd < 0) {
			if (!reading && errno == ENOENT) {
				flags |= O_CREAT;
				goto again;
			}
			perror("open");
			return 1;
		}
		posix_fadvise(td[i].fd, 0, (off_t) -1, POSIX_FADV_RANDOM);
		posix_fadvise(td[i].fd, 0, (off_t) -1, POSIX_FADV_DONTNEED);
		td[i].rwf_flags = rwf_flags;
		td[i].bs = bs;
		td[i].reading = reading;
	}

	printf("Starting %d threads\n", nr);
	for (i = 0; i < nr; i++) {
		pthread_barrier_init(&td[i].barrier, NULL, 2);
		pthread_create(&td[i].thread, NULL, thread_fn, &td[i]);
		pthread_barrier_wait(&td[i].barrier);
	}

	printf("%sing bs %u, uncached %d\n", reading ? "read" : "writ", bs, (rwf_flags & RWF_UNCACHED) != 0);
	gettimeofday(&tv, NULL);
	start = tv;
	do {
		unsigned long mb, msec;
		int done = 0;

		usleep(100000);
		msec = mtime_since_now(&tv);
		bytes = 0;
		for (i = 0; i < nr; i++) {
			bytes += td[i].bytes;
			done += td[i].done;
		}
		if (msec < 1000 && (done != nr))
			continue;

		mb = (bytes - last_bytes) >> 10;
		last_bytes = bytes;
		gettimeofday(&tv, NULL);
		printf("%3llds: %luMB/sec, MB=%lu\n", mtime_since_now(&start) / 1000, mb / msec, (bytes / msec) >> 10);
		if (done == nr)
			break;
	} while (1);

	return 0;
}