io_uring: add a few test tools

This adds two test programs in tools/io_uring/ that demonstrate both
the raw io_uring API (and all features) through a small benchmark
app, io_uring-bench, and the liburing exposed API in a simplified
cp(1) implementation through io_uring-cp.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

tools/io_uring/Makefile

@@ -0,0 +1,18 @@
+# SPDX-License-Identifier: GPL-2.0
+# Makefile for io_uring test tools
+CFLAGS += -Wall -Wextra -g -D_GNU_SOURCE
+LDLIBS += -lpthread
+
+all: io_uring-cp io_uring-bench
+%: %.c
+	$(CC) $(CFLAGS) -o $@ $^
+
+io_uring-bench: syscall.o io_uring-bench.o
+	$(CC) $(CFLAGS) $(LDLIBS) -o $@ $^
+
+io_uring-cp: setup.o syscall.o queue.o
+
+clean:
+	$(RM) io_uring-cp io_uring-bench *.o
+
+.PHONY: all clean

tools/io_uring/README

@@ -0,0 +1,29 @@
+This directory includes a few programs that demonstrate how to use io_uring
+in an application. The examples are:
+
+io_uring-cp
+	A very basic io_uring implementation of cp(1). It takes two
+	arguments, copies the first argument to the second. This example
+	is part of liburing, and hence uses the simplified liburing API
+	for setting up an io_uring instance, submitting IO, completing IO,
+	etc. The support functions in queue.c and setup.c are straight
+	out of liburing.
+
+io_uring-bench
+	Benchmark program that does random reads on a number of files. This
+	app demonstrates the various features of io_uring, like fixed files,
+	fixed buffers, and polled IO. There are options in the program to
+	control which features to use. Arguments is the file (or files) that
+	io_uring-bench should operate on. This uses the raw io_uring
+	interface.
+
+liburing can be cloned with git here:
+
+	git://git.kernel.dk/liburing
+
+and contains a number of unit tests as well for testing io_uring. It also
+comes with man pages for the three system calls.
+
+Fio includes an io_uring engine, you can clone fio here:
+
+	git://git.kernel.dk/fio

tools/io_uring/barrier.h

@@ -0,0 +1,16 @@
+#ifndef LIBURING_BARRIER_H
+#define LIBURING_BARRIER_H
+
+#if defined(__x86_64) || defined(__i386__)
+#define read_barrier()	__asm__ __volatile__("":::"memory")
+#define write_barrier()	__asm__ __volatile__("":::"memory")
+#else
+/*
+ * Add arch appropriate definitions. Be safe and use full barriers for
+ * archs we don't have support for.
+ */
+#define read_barrier()	__sync_synchronize()
+#define write_barrier()	__sync_synchronize()
+#endif
+
+#endif

tools/io_uring/io_uring-bench.c

@@ -0,0 +1,616 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Simple benchmark program that uses the various features of io_uring
+ * to provide fast random access to a device/file. It has various
+ * options that are control how we use io_uring, see the OPTIONS section
+ * below. This uses the raw io_uring interface.
+ *
+ * Copyright (C) 2018-2019 Jens Axboe
+ */
+#include <stdio.h>
+#include <errno.h>
+#include <assert.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <signal.h>
+#include <inttypes.h>
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/ioctl.h>
+#include <sys/syscall.h>
+#include <sys/resource.h>
+#include <sys/mman.h>
+#include <sys/uio.h>
+#include <linux/fs.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <string.h>
+#include <pthread.h>
+#include <sched.h>
+
+#include "liburing.h"
+#include "barrier.h"
+
+#ifndef IOCQE_FLAG_CACHEHIT
+#define IOCQE_FLAG_CACHEHIT	(1U << 0)
+#endif
+
+#define min(a, b)		((a < b) ? (a) : (b))
+
+struct io_sq_ring {
+	unsigned *head;
+	unsigned *tail;
+	unsigned *ring_mask;
+	unsigned *ring_entries;
+	unsigned *flags;
+	unsigned *array;
+};
+
+struct io_cq_ring {
+	unsigned *head;
+	unsigned *tail;
+	unsigned *ring_mask;
+	unsigned *ring_entries;
+	struct io_uring_cqe *cqes;
+};
+
+#define DEPTH			128
+
+#define BATCH_SUBMIT		32
+#define BATCH_COMPLETE		32
+
+#define BS			4096
+
+#define MAX_FDS			16
+
+static unsigned sq_ring_mask, cq_ring_mask;
+
+struct file {
+	unsigned long max_blocks;
+	unsigned pending_ios;
+	int real_fd;
+	int fixed_fd;
+};
+
+struct submitter {
+	pthread_t thread;
+	int ring_fd;
+	struct drand48_data rand;
+	struct io_sq_ring sq_ring;
+	struct io_uring_sqe *sqes;
+	struct iovec iovecs[DEPTH];
+	struct io_cq_ring cq_ring;
+	int inflight;
+	unsigned long reaps;
+	unsigned long done;
+	unsigned long calls;
+	unsigned long cachehit, cachemiss;
+	volatile int finish;
+
+	__s32 *fds;
+
+	struct file files[MAX_FDS];
+	unsigned nr_files;
+	unsigned cur_file;
+};
+
+static struct submitter submitters[1];
+static volatile int finish;
+
+/*
+ * OPTIONS: Set these to test the various features of io_uring.
+ */
+static int polled = 1;		/* use IO polling */
+static int fixedbufs = 1;	/* use fixed user buffers */
+static int register_files = 1;	/* use fixed files */
+static int buffered = 0;	/* use buffered IO, not O_DIRECT */
+static int sq_thread_poll = 0;	/* use kernel submission/poller thread */
+static int sq_thread_cpu = -1;	/* pin above thread to this CPU */
+static int do_nop = 0;		/* no-op SQ ring commands */
+
+static int io_uring_register_buffers(struct submitter *s)
+{
+	if (do_nop)
+		return 0;
+
+	return io_uring_register(s->ring_fd, IORING_REGISTER_BUFFERS, s->iovecs,
+					DEPTH);
+}
+
+static int io_uring_register_files(struct submitter *s)
+{
+	unsigned i;
+
+	if (do_nop)
+		return 0;
+
+	s->fds = calloc(s->nr_files, sizeof(__s32));
+	for (i = 0; i < s->nr_files; i++) {
+		s->fds[i] = s->files[i].real_fd;
+		s->files[i].fixed_fd = i;
+	}
+
+	return io_uring_register(s->ring_fd, IORING_REGISTER_FILES, s->fds,
+					s->nr_files);
+}
+
+static int gettid(void)
+{
+	return syscall(__NR_gettid);
+}
+
+static unsigned file_depth(struct submitter *s)
+{
+	return (DEPTH + s->nr_files - 1) / s->nr_files;
+}
+
+static void init_io(struct submitter *s, unsigned index)
+{
+	struct io_uring_sqe *sqe = &s->sqes[index];
+	unsigned long offset;
+	struct file *f;
+	long r;
+
+	if (do_nop) {
+		sqe->opcode = IORING_OP_NOP;
+		return;
+	}
+
+	if (s->nr_files == 1) {
+		f = &s->files[0];
+	} else {
+		f = &s->files[s->cur_file];
+		if (f->pending_ios >= file_depth(s)) {
+			s->cur_file++;
+			if (s->cur_file == s->nr_files)
+				s->cur_file = 0;
+			f = &s->files[s->cur_file];
+		}
+	}
+	f->pending_ios++;
+
+	lrand48_r(&s->rand, &r);
+	offset = (r % (f->max_blocks - 1)) * BS;
+
+	if (register_files) {
+		sqe->flags = IOSQE_FIXED_FILE;
+		sqe->fd = f->fixed_fd;
+	} else {
+		sqe->flags = 0;
+		sqe->fd = f->real_fd;
+	}
+	if (fixedbufs) {
+		sqe->opcode = IORING_OP_READ_FIXED;
+		sqe->addr = (unsigned long) s->iovecs[index].iov_base;
+		sqe->len = BS;
+		sqe->buf_index = index;
+	} else {
+		sqe->opcode = IORING_OP_READV;
+		sqe->addr = (unsigned long) &s->iovecs[index];
+		sqe->len = 1;
+		sqe->buf_index = 0;
+	}
+	sqe->ioprio = 0;
+	sqe->off = offset;
+	sqe->user_data = (unsigned long) f;
+}
+
+static int prep_more_ios(struct submitter *s, unsigned max_ios)
+{
+	struct io_sq_ring *ring = &s->sq_ring;
+	unsigned index, tail, next_tail, prepped = 0;
+
+	next_tail = tail = *ring->tail;
+	do {
+		next_tail++;
+		read_barrier();
+		if (next_tail == *ring->head)
+			break;
+
+		index = tail & sq_ring_mask;
+		init_io(s, index);
+		ring->array[index] = index;
+		prepped++;
+		tail = next_tail;
+	} while (prepped < max_ios);
+
+	if (*ring->tail != tail) {
+		/* order tail store with writes to sqes above */
+		write_barrier();
+		*ring->tail = tail;
+		write_barrier();
+	}
+	return prepped;
+}
+
+static int get_file_size(struct file *f)
+{
+	struct stat st;
+
+	if (fstat(f->real_fd, &st) < 0)
+		return -1;
+	if (S_ISBLK(st.st_mode)) {
+		unsigned long long bytes;
+
+		if (ioctl(f->real_fd, BLKGETSIZE64, &bytes) != 0)
+			return -1;
+
+		f->max_blocks = bytes / BS;
+		return 0;
+	} else if (S_ISREG(st.st_mode)) {
+		f->max_blocks = st.st_size / BS;
+		return 0;
+	}
+
+	return -1;
+}
+
+static int reap_events(struct submitter *s)
+{
+	struct io_cq_ring *ring = &s->cq_ring;
+	struct io_uring_cqe *cqe;
+	unsigned head, reaped = 0;
+
+	head = *ring->head;
+	do {
+		struct file *f;
+
+		read_barrier();
+		if (head == *ring->tail)
+			break;
+		cqe = &ring->cqes[head & cq_ring_mask];
+		if (!do_nop) {
+			f = (struct file *) (uintptr_t) cqe->user_data;
+			f->pending_ios--;
+			if (cqe->res != BS) {
+				printf("io: unexpected ret=%d\n", cqe->res);
+				if (polled && cqe->res == -EOPNOTSUPP)
+					printf("Your filesystem doesn't support poll\n");
+				return -1;
+			}
+		}
+		if (cqe->flags & IOCQE_FLAG_CACHEHIT)
+			s->cachehit++;
+		else
+			s->cachemiss++;
+		reaped++;
+		head++;
+	} while (1);
+
+	s->inflight -= reaped;
+	*ring->head = head;
+	write_barrier();
+	return reaped;
+}
+
+static void *submitter_fn(void *data)
+{
+	struct submitter *s = data;
+	struct io_sq_ring *ring = &s->sq_ring;
+	int ret, prepped;
+
+	printf("submitter=%d\n", gettid());
+
+	srand48_r(pthread_self(), &s->rand);
+
+	prepped = 0;
+	do {
+		int to_wait, to_submit, this_reap, to_prep;
+
+		if (!prepped && s->inflight < DEPTH) {
+			to_prep = min(DEPTH - s->inflight, BATCH_SUBMIT);
+			prepped = prep_more_ios(s, to_prep);
+		}
+		s->inflight += prepped;
+submit_more:
+		to_submit = prepped;
+submit:
+		if (to_submit && (s->inflight + to_submit <= DEPTH))
+			to_wait = 0;
+		else
+			to_wait = min(s->inflight + to_submit, BATCH_COMPLETE);
+
+		/*
+		 * Only need to call io_uring_enter if we're not using SQ thread
+		 * poll, or if IORING_SQ_NEED_WAKEUP is set.
+		 */
+		if (!sq_thread_poll || (*ring->flags & IORING_SQ_NEED_WAKEUP)) {
+			unsigned flags = 0;
+
+			if (to_wait)
+				flags = IORING_ENTER_GETEVENTS;
+			if ((*ring->flags & IORING_SQ_NEED_WAKEUP))
+				flags |= IORING_ENTER_SQ_WAKEUP;
+			ret = io_uring_enter(s->ring_fd, to_submit, to_wait,
+						flags, NULL);
+			s->calls++;
+		}
+
+		/*
+		 * For non SQ thread poll, we already got the events we needed
+		 * through the io_uring_enter() above. For SQ thread poll, we
+		 * need to loop here until we find enough events.
+		 */
+		this_reap = 0;
+		do {
+			int r;
+			r = reap_events(s);
+			if (r == -1) {
+				s->finish = 1;
+				break;
+			} else if (r > 0)
+				this_reap += r;
+		} while (sq_thread_poll && this_reap < to_wait);
+		s->reaps += this_reap;
+
+		if (ret >= 0) {
+			if (!ret) {
+				to_submit = 0;
+				if (s->inflight)
+					goto submit;
+				continue;
+			} else if (ret < to_submit) {
+				int diff = to_submit - ret;
+
+				s->done += ret;
+				prepped -= diff;
+				goto submit_more;
+			}
+			s->done += ret;
+			prepped = 0;
+			continue;
+		} else if (ret < 0) {
+			if (errno == EAGAIN) {
+				if (s->finish)
+					break;
+				if (this_reap)
+					goto submit;
+				to_submit = 0;
+				goto submit;
+			}
+			printf("io_submit: %s\n", strerror(errno));
+			break;
+		}
+	} while (!s->finish);
+
+	finish = 1;
+	return NULL;
+}
+
+static void sig_int(int sig)
+{
+	printf("Exiting on signal %d\n", sig);
+	submitters[0].finish = 1;
+	finish = 1;
+}
+
+static void arm_sig_int(void)
+{
+	struct sigaction act;
+
+	memset(&act, 0, sizeof(act));
+	act.sa_handler = sig_int;
+	act.sa_flags = SA_RESTART;
+	sigaction(SIGINT, &act, NULL);
+}
+
+static int setup_ring(struct submitter *s)
+{
+	struct io_sq_ring *sring = &s->sq_ring;
+	struct io_cq_ring *cring = &s->cq_ring;
+	struct io_uring_params p;
+	int ret, fd;
+	void *ptr;
+
+	memset(&p, 0, sizeof(p));
+
+	if (polled && !do_nop)
+		p.flags |= IORING_SETUP_IOPOLL;
+	if (sq_thread_poll) {
+		p.flags |= IORING_SETUP_SQPOLL;
+		if (sq_thread_cpu != -1) {
+			p.flags |= IORING_SETUP_SQ_AFF;
+			p.sq_thread_cpu = sq_thread_cpu;
+		}
+	}
+
+	fd = io_uring_setup(DEPTH, &p);
+	if (fd < 0) {
+		perror("io_uring_setup");
+		return 1;
+	}
+	s->ring_fd = fd;
+
+	if (fixedbufs) {
+		ret = io_uring_register_buffers(s);
+		if (ret < 0) {
+			perror("io_uring_register_buffers");
+			return 1;
+		}
+	}
+
+	if (register_files) {
+		ret = io_uring_register_files(s);
+		if (ret < 0) {
+			perror("io_uring_register_files");
+			return 1;
+		}
+	}
+
+	ptr = mmap(0, p.sq_off.array + p.sq_entries * sizeof(__u32),
+			PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd,
+			IORING_OFF_SQ_RING);
+	printf("sq_ring ptr = 0x%p\n", ptr);
+	sring->head = ptr + p.sq_off.head;
+	sring->tail = ptr + p.sq_off.tail;
+	sring->ring_mask = ptr + p.sq_off.ring_mask;
+	sring->ring_entries = ptr + p.sq_off.ring_entries;
+	sring->flags = ptr + p.sq_off.flags;
+	sring->array = ptr + p.sq_off.array;
+	sq_ring_mask = *sring->ring_mask;
+
+	s->sqes = mmap(0, p.sq_entries * sizeof(struct io_uring_sqe),
+			PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd,
+			IORING_OFF_SQES);
+	printf("sqes ptr    = 0x%p\n", s->sqes);
+
+	ptr = mmap(0, p.cq_off.cqes + p.cq_entries * sizeof(struct io_uring_cqe),
+			PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd,
+			IORING_OFF_CQ_RING);
+	printf("cq_ring ptr = 0x%p\n", ptr);
+	cring->head = ptr + p.cq_off.head;
+	cring->tail = ptr + p.cq_off.tail;
+	cring->ring_mask = ptr + p.cq_off.ring_mask;
+	cring->ring_entries = ptr + p.cq_off.ring_entries;
+	cring->cqes = ptr + p.cq_off.cqes;
+	cq_ring_mask = *cring->ring_mask;
+	return 0;
+}
+
+static void file_depths(char *buf)
+{
+	struct submitter *s = &submitters[0];
+	unsigned i;
+	char *p;
+
+	buf[0] = '\0';
+	p = buf;
+	for (i = 0; i < s->nr_files; i++) {
+		struct file *f = &s->files[i];
+
+		if (i + 1 == s->nr_files)
+			p += sprintf(p, "%d", f->pending_ios);
+		else
+			p += sprintf(p, "%d, ", f->pending_ios);
+	}
+}
+
+int main(int argc, char *argv[])
+{
+	struct submitter *s = &submitters[0];
+	unsigned long done, calls, reap, cache_hit, cache_miss;
+	int err, i, flags, fd;
+	char *fdepths;
+	void *ret;
+
+	if (!do_nop && argc < 2) {
+		printf("%s: filename\n", argv[0]);
+		return 1;
+	}
+
+	flags = O_RDONLY | O_NOATIME;
+	if (!buffered)
+		flags |= O_DIRECT;
+
+	i = 1;
+	while (!do_nop && i < argc) {
+		struct file *f;
+
+		if (s->nr_files == MAX_FDS) {
+			printf("Max number of files (%d) reached\n", MAX_FDS);
+			break;
+		}
+		fd = open(argv[i], flags);
+		if (fd < 0) {
+			perror("open");
+			return 1;
+		}
+
+		f = &s->files[s->nr_files];
+		f->real_fd = fd;
+		if (get_file_size(f)) {
+			printf("failed getting size of device/file\n");
+			return 1;
+		}
+		if (f->max_blocks <= 1) {
+			printf("Zero file/device size?\n");
+			return 1;
+		}
+		f->max_blocks--;
+
+		printf("Added file %s\n", argv[i]);
+		s->nr_files++;
+		i++;
+	}
+
+	if (fixedbufs) {
+		struct rlimit rlim;
+
+		rlim.rlim_cur = RLIM_INFINITY;
+		rlim.rlim_max = RLIM_INFINITY;
+		if (setrlimit(RLIMIT_MEMLOCK, &rlim) < 0) {
+			perror("setrlimit");
+			return 1;
+		}
+	}
+
+	arm_sig_int();
+
+	for (i = 0; i < DEPTH; i++) {
+		void *buf;
+
+		if (posix_memalign(&buf, BS, BS)) {
+			printf("failed alloc\n");
+			return 1;
+		}
+		s->iovecs[i].iov_base = buf;
+		s->iovecs[i].iov_len = BS;
+	}
+
+	err = setup_ring(s);
+	if (err) {
+		printf("ring setup failed: %s, %d\n", strerror(errno), err);
+		return 1;
+	}
+	printf("polled=%d, fixedbufs=%d, buffered=%d", polled, fixedbufs, buffered);
+	printf(" QD=%d, sq_ring=%d, cq_ring=%d\n", DEPTH, *s->sq_ring.ring_entries, *s->cq_ring.ring_entries);
+
+	pthread_create(&s->thread, NULL, submitter_fn, s);
+
+	fdepths = malloc(8 * s->nr_files);
+	cache_hit = cache_miss = reap = calls = done = 0;
+	do {
+		unsigned long this_done = 0;
+		unsigned long this_reap = 0;
+		unsigned long this_call = 0;
+		unsigned long this_cache_hit = 0;
+		unsigned long this_cache_miss = 0;
+		unsigned long rpc = 0, ipc = 0;
+		double hit = 0.0;
+
+		sleep(1);
+		this_done += s->done;
+		this_call += s->calls;
+		this_reap += s->reaps;
+		this_cache_hit += s->cachehit;
+		this_cache_miss += s->cachemiss;
+		if (this_cache_hit && this_cache_miss) {
+			unsigned long hits, total;
+
+			hits = this_cache_hit - cache_hit;
+			total = hits + this_cache_miss - cache_miss;
+			hit = (double) hits / (double) total;
+			hit *= 100.0;
+		}
+		if (this_call - calls) {
+			rpc = (this_done - done) / (this_call - calls);
+			ipc = (this_reap - reap) / (this_call - calls);
+		} else
+			rpc = ipc = -1;
+		file_depths(fdepths);
+		printf("IOPS=%lu, IOS/call=%ld/%ld, inflight=%u (%s), Cachehit=%0.2f%%\n",
+				this_done - done, rpc, ipc, s->inflight,
+				fdepths, hit);
+		done = this_done;
+		calls = this_call;
+		reap = this_reap;
+		cache_hit = s->cachehit;
+		cache_miss = s->cachemiss;
+	} while (!finish);
+
+	pthread_join(s->thread, &ret);
+	close(s->ring_fd);
+	free(fdepths);
+	return 0;
+}

tools/io_uring/io_uring-cp.c

@@ -0,0 +1,251 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Simple test program that demonstrates a file copy through io_uring. This
+ * uses the API exposed by liburing.
+ *
+ * Copyright (C) 2018-2019 Jens Axboe
+ */
+#include <stdio.h>
+#include <fcntl.h>
+#include <string.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <assert.h>
+#include <errno.h>
+#include <inttypes.h>
+#include <sys/stat.h>
+#include <sys/ioctl.h>
+
+#include "liburing.h"
+
+#define QD	64
+#define BS	(32*1024)
+
+static int infd, outfd;
+
+struct io_data {
+	int read;
+	off_t first_offset, offset;
+	size_t first_len;
+	struct iovec iov;
+};
+
+static int setup_context(unsigned entries, struct io_uring *ring)
+{
+	int ret;
+
+	ret = io_uring_queue_init(entries, ring, 0);
+	if (ret < 0) {
+		fprintf(stderr, "queue_init: %s\n", strerror(-ret));
+		return -1;
+	}
+
+	return 0;
+}
+
+static int get_file_size(int fd, off_t *size)
+{
+	struct stat st;
+
+	if (fstat(fd, &st) < 0)
+		return -1;
+	if (S_ISREG(st.st_mode)) {
+		*size = st.st_size;
+		return 0;
+	} else if (S_ISBLK(st.st_mode)) {
+		unsigned long long bytes;
+
+		if (ioctl(fd, BLKGETSIZE64, &bytes) != 0)
+			return -1;
+
+		*size = bytes;
+		return 0;
+	}
+
+	return -1;
+}
+
+static void queue_prepped(struct io_uring *ring, struct io_data *data)
+{
+	struct io_uring_sqe *sqe;
+
+	sqe = io_uring_get_sqe(ring);
+	assert(sqe);
+
+	if (data->read)
+		io_uring_prep_readv(sqe, infd, &data->iov, 1, data->offset);
+	else
+		io_uring_prep_writev(sqe, outfd, &data->iov, 1, data->offset);
+
+	io_uring_sqe_set_data(sqe, data);
+}
+
+static int queue_read(struct io_uring *ring, off_t size, off_t offset)
+{
+	struct io_uring_sqe *sqe;
+	struct io_data *data;
+
+	sqe = io_uring_get_sqe(ring);
+	if (!sqe)
+		return 1;
+
+	data = malloc(size + sizeof(*data));
+	data->read = 1;
+	data->offset = data->first_offset = offset;
+
+	data->iov.iov_base = data + 1;
+	data->iov.iov_len = size;
+	data->first_len = size;
+
+	io_uring_prep_readv(sqe, infd, &data->iov, 1, offset);
+	io_uring_sqe_set_data(sqe, data);
+	return 0;
+}
+
+static void queue_write(struct io_uring *ring, struct io_data *data)
+{
+	data->read = 0;
+	data->offset = data->first_offset;
+
+	data->iov.iov_base = data + 1;
+	data->iov.iov_len = data->first_len;
+
+	queue_prepped(ring, data);
+	io_uring_submit(ring);
+}
+
+static int copy_file(struct io_uring *ring, off_t insize)
+{
+	unsigned long reads, writes;
+	struct io_uring_cqe *cqe;
+	off_t write_left, offset;
+	int ret;
+
+	write_left = insize;
+	writes = reads = offset = 0;
+
+	while (insize || write_left) {
+		unsigned long had_reads;
+		int got_comp;
+
+		/*
+		 * Queue up as many reads as we can
+		 */
+		had_reads = reads;
+		while (insize) {
+			off_t this_size = insize;
+
+			if (reads + writes >= QD)
+				break;
+			if (this_size > BS)
+				this_size = BS;
+			else if (!this_size)
+				break;
+
+			if (queue_read(ring, this_size, offset))
+				break;
+
+			insize -= this_size;
+			offset += this_size;
+			reads++;
+		}
+
+		if (had_reads != reads) {
+			ret = io_uring_submit(ring);
+			if (ret < 0) {
+				fprintf(stderr, "io_uring_submit: %s\n", strerror(-ret));
+				break;
+			}
+		}
+
+		/*
+		 * Queue is full at this point. Find at least one completion.
+		 */
+		got_comp = 0;
+		while (write_left) {
+			struct io_data *data;
+
+			if (!got_comp) {
+				ret = io_uring_wait_completion(ring, &cqe);
+				got_comp = 1;
+			} else
+				ret = io_uring_get_completion(ring, &cqe);
+			if (ret < 0) {
+				fprintf(stderr, "io_uring_get_completion: %s\n",
+							strerror(-ret));
+				return 1;
+			}
+			if (!cqe)
+				break;
+
+			data = (struct io_data *) (uintptr_t) cqe->user_data;
+			if (cqe->res < 0) {
+				if (cqe->res == -EAGAIN) {
+					queue_prepped(ring, data);
+					continue;
+				}
+				fprintf(stderr, "cqe failed: %s\n",
+						strerror(-cqe->res));
+				return 1;
+			} else if ((size_t) cqe->res != data->iov.iov_len) {
+				/* Short read/write, adjust and requeue */
+				data->iov.iov_base += cqe->res;
+				data->iov.iov_len -= cqe->res;
+				data->offset += cqe->res;
+				queue_prepped(ring, data);
+				continue;
+			}
+
+			/*
+			 * All done. if write, nothing else to do. if read,
+			 * queue up corresponding write.
+			 */
+			if (data->read) {
+				queue_write(ring, data);
+				write_left -= data->first_len;
+				reads--;
+				writes++;
+			} else {
+				free(data);
+				writes--;
+			}
+		}
+	}
+
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	struct io_uring ring;
+	off_t insize;
+	int ret;
+
+	if (argc < 3) {
+		printf("%s: infile outfile\n", argv[0]);
+		return 1;
+	}
+
+	infd = open(argv[1], O_RDONLY);
+	if (infd < 0) {
+		perror("open infile");
+		return 1;
+	}
+	outfd = open(argv[2], O_WRONLY | O_CREAT | O_TRUNC, 0644);
+	if (outfd < 0) {
+		perror("open outfile");
+		return 1;
+	}
+
+	if (setup_context(QD, &ring))
+		return 1;
+	if (get_file_size(infd, &insize))
+		return 1;
+
+	ret = copy_file(&ring, insize);
+
+	close(infd);
+	close(outfd);
+	io_uring_queue_exit(&ring);
+	return ret;
+}

tools/io_uring/liburing.h

@@ -0,0 +1,143 @@
+#ifndef LIB_URING_H
+#define LIB_URING_H
+
+#include <sys/uio.h>
+#include <signal.h>
+#include <string.h>
+#include "../../include/uapi/linux/io_uring.h"
+
+/*
+ * Library interface to io_uring
+ */
+struct io_uring_sq {
+	unsigned *khead;
+	unsigned *ktail;
+	unsigned *kring_mask;
+	unsigned *kring_entries;
+	unsigned *kflags;
+	unsigned *kdropped;
+	unsigned *array;
+	struct io_uring_sqe *sqes;
+
+	unsigned sqe_head;
+	unsigned sqe_tail;
+
+	size_t ring_sz;
+};
+
+struct io_uring_cq {
+	unsigned *khead;
+	unsigned *ktail;
+	unsigned *kring_mask;
+	unsigned *kring_entries;
+	unsigned *koverflow;
+	struct io_uring_cqe *cqes;
+
+	size_t ring_sz;
+};
+
+struct io_uring {
+	struct io_uring_sq sq;
+	struct io_uring_cq cq;
+	int ring_fd;
+};
+
+/*
+ * System calls
+ */
+extern int io_uring_setup(unsigned entries, struct io_uring_params *p);
+extern int io_uring_enter(unsigned fd, unsigned to_submit,
+	unsigned min_complete, unsigned flags, sigset_t *sig);
+extern int io_uring_register(int fd, unsigned int opcode, void *arg,
+	unsigned int nr_args);
+
+/*
+ * Library interface
+ */
+extern int io_uring_queue_init(unsigned entries, struct io_uring *ring,
+	unsigned flags);
+extern int io_uring_queue_mmap(int fd, struct io_uring_params *p,
+	struct io_uring *ring);
+extern void io_uring_queue_exit(struct io_uring *ring);
+extern int io_uring_get_completion(struct io_uring *ring,
+	struct io_uring_cqe **cqe_ptr);
+extern int io_uring_wait_completion(struct io_uring *ring,
+	struct io_uring_cqe **cqe_ptr);
+extern int io_uring_submit(struct io_uring *ring);
+extern struct io_uring_sqe *io_uring_get_sqe(struct io_uring *ring);
+
+/*
+ * Command prep helpers
+ */
+static inline void io_uring_sqe_set_data(struct io_uring_sqe *sqe, void *data)
+{
+	sqe->user_data = (unsigned long) data;
+}
+
+static inline void io_uring_prep_rw(int op, struct io_uring_sqe *sqe, int fd,
+				    void *addr, unsigned len, off_t offset)
+{
+	memset(sqe, 0, sizeof(*sqe));
+	sqe->opcode = op;
+	sqe->fd = fd;
+	sqe->off = offset;
+	sqe->addr = (unsigned long) addr;
+	sqe->len = len;
+}
+
+static inline void io_uring_prep_readv(struct io_uring_sqe *sqe, int fd,
+				       struct iovec *iovecs, unsigned nr_vecs,
+				       off_t offset)
+{
+	io_uring_prep_rw(IORING_OP_READV, sqe, fd, iovecs, nr_vecs, offset);
+}
+
+static inline void io_uring_prep_read_fixed(struct io_uring_sqe *sqe, int fd,
+					    void *buf, unsigned nbytes,
+					    off_t offset)
+{
+	io_uring_prep_rw(IORING_OP_READ_FIXED, sqe, fd, buf, nbytes, offset);
+}
+
+static inline void io_uring_prep_writev(struct io_uring_sqe *sqe, int fd,
+				        struct iovec *iovecs, unsigned nr_vecs,
+					off_t offset)
+{
+	io_uring_prep_rw(IORING_OP_WRITEV, sqe, fd, iovecs, nr_vecs, offset);
+}
+
+static inline void io_uring_prep_write_fixed(struct io_uring_sqe *sqe, int fd,
+					     void *buf, unsigned nbytes,
+					     off_t offset)
+{
+	io_uring_prep_rw(IORING_OP_WRITE_FIXED, sqe, fd, buf, nbytes, offset);
+}
+
+static inline void io_uring_prep_poll_add(struct io_uring_sqe *sqe, int fd,
+					  short poll_mask)
+{
+	memset(sqe, 0, sizeof(*sqe));
+	sqe->opcode = IORING_OP_POLL_ADD;
+	sqe->fd = fd;
+	sqe->poll_events = poll_mask;
+}
+
+static inline void io_uring_prep_poll_remove(struct io_uring_sqe *sqe,
+					     void *user_data)
+{
+	memset(sqe, 0, sizeof(*sqe));
+	sqe->opcode = IORING_OP_POLL_REMOVE;
+	sqe->addr = (unsigned long) user_data;
+}
+
+static inline void io_uring_prep_fsync(struct io_uring_sqe *sqe, int fd,
+				       int datasync)
+{
+	memset(sqe, 0, sizeof(*sqe));
+	sqe->opcode = IORING_OP_FSYNC;
+	sqe->fd = fd;
+	if (datasync)
+		sqe->fsync_flags = IORING_FSYNC_DATASYNC;
+}
+
+#endif

tools/io_uring/queue.c

@@ -0,0 +1,164 @@
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/mman.h>
+#include <unistd.h>
+#include <errno.h>
+#include <string.h>
+
+#include "liburing.h"
+#include "barrier.h"
+
+static int __io_uring_get_completion(struct io_uring *ring,
+				     struct io_uring_cqe **cqe_ptr, int wait)
+{
+	struct io_uring_cq *cq = &ring->cq;
+	const unsigned mask = *cq->kring_mask;
+	unsigned head;
+	int ret;
+
+	*cqe_ptr = NULL;
+	head = *cq->khead;
+	do {
+		/*
+		 * It's necessary to use a read_barrier() before reading
+		 * the CQ tail, since the kernel updates it locklessly. The
+		 * kernel has the matching store barrier for the update. The
+		 * kernel also ensures that previous stores to CQEs are ordered
+		 * with the tail update.
+		 */
+		read_barrier();
+		if (head != *cq->ktail) {
+			*cqe_ptr = &cq->cqes[head & mask];
+			break;
+		}
+		if (!wait)
+			break;
+		ret = io_uring_enter(ring->ring_fd, 0, 1,
+					IORING_ENTER_GETEVENTS, NULL);
+		if (ret < 0)
+			return -errno;
+	} while (1);
+
+	if (*cqe_ptr) {
+		*cq->khead = head + 1;
+		/*
+		 * Ensure that the kernel sees our new head, the kernel has
+		 * the matching read barrier.
+		 */
+		write_barrier();
+	}
+
+	return 0;
+}
+
+/*
+ * Return an IO completion, if one is readily available
+ */
+int io_uring_get_completion(struct io_uring *ring,
+			    struct io_uring_cqe **cqe_ptr)
+{
+	return __io_uring_get_completion(ring, cqe_ptr, 0);
+}
+
+/*
+ * Return an IO completion, waiting for it if necessary
+ */
+int io_uring_wait_completion(struct io_uring *ring,
+			     struct io_uring_cqe **cqe_ptr)
+{
+	return __io_uring_get_completion(ring, cqe_ptr, 1);
+}
+
+/*
+ * Submit sqes acquired from io_uring_get_sqe() to the kernel.
+ *
+ * Returns number of sqes submitted
+ */
+int io_uring_submit(struct io_uring *ring)
+{
+	struct io_uring_sq *sq = &ring->sq;
+	const unsigned mask = *sq->kring_mask;
+	unsigned ktail, ktail_next, submitted;
+	int ret;
+
+	/*
+	 * If we have pending IO in the kring, submit it first. We need a
+	 * read barrier here to match the kernels store barrier when updating
+	 * the SQ head.
+	 */
+	read_barrier();
+	if (*sq->khead != *sq->ktail) {
+		submitted = *sq->kring_entries;
+		goto submit;
+	}
+
+	if (sq->sqe_head == sq->sqe_tail)
+		return 0;
+
+	/*
+	 * Fill in sqes that we have queued up, adding them to the kernel ring
+	 */
+	submitted = 0;
+	ktail = ktail_next = *sq->ktail;
+	while (sq->sqe_head < sq->sqe_tail) {
+		ktail_next++;
+		read_barrier();
+
+		sq->array[ktail & mask] = sq->sqe_head & mask;
+		ktail = ktail_next;
+
+		sq->sqe_head++;
+		submitted++;
+	}
+
+	if (!submitted)
+		return 0;
+
+	if (*sq->ktail != ktail) {
+		/*
+		 * First write barrier ensures that the SQE stores are updated
+		 * with the tail update. This is needed so that the kernel
+		 * will never see a tail update without the preceeding sQE
+		 * stores being done.
+		 */
+		write_barrier();
+		*sq->ktail = ktail;
+		/*
+		 * The kernel has the matching read barrier for reading the
+		 * SQ tail.
+		 */
+		write_barrier();
+	}
+
+submit:
+	ret = io_uring_enter(ring->ring_fd, submitted, 0,
+				IORING_ENTER_GETEVENTS, NULL);
+	if (ret < 0)
+		return -errno;
+
+	return 0;
+}
+
+/*
+ * Return an sqe to fill. Application must later call io_uring_submit()
+ * when it's ready to tell the kernel about it. The caller may call this
+ * function multiple times before calling io_uring_submit().
+ *
+ * Returns a vacant sqe, or NULL if we're full.
+ */
+struct io_uring_sqe *io_uring_get_sqe(struct io_uring *ring)
+{
+	struct io_uring_sq *sq = &ring->sq;
+	unsigned next = sq->sqe_tail + 1;
+	struct io_uring_sqe *sqe;
+
+	/*
+	 * All sqes are used
+	 */
+	if (next - sq->sqe_head > *sq->kring_entries)
+		return NULL;
+
+	sqe = &sq->sqes[sq->sqe_tail & *sq->kring_mask];
+	sq->sqe_tail = next;
+	return sqe;
+}

tools/io_uring/setup.c

@@ -0,0 +1,103 @@
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/mman.h>
+#include <unistd.h>
+#include <errno.h>
+#include <string.h>
+
+#include "liburing.h"
+
+static int io_uring_mmap(int fd, struct io_uring_params *p,
+			 struct io_uring_sq *sq, struct io_uring_cq *cq)
+{
+	size_t size;
+	void *ptr;
+	int ret;
+
+	sq->ring_sz = p->sq_off.array + p->sq_entries * sizeof(unsigned);
+	ptr = mmap(0, sq->ring_sz, PROT_READ | PROT_WRITE,
+			MAP_SHARED | MAP_POPULATE, fd, IORING_OFF_SQ_RING);
+	if (ptr == MAP_FAILED)
+		return -errno;
+	sq->khead = ptr + p->sq_off.head;
+	sq->ktail = ptr + p->sq_off.tail;
+	sq->kring_mask = ptr + p->sq_off.ring_mask;
+	sq->kring_entries = ptr + p->sq_off.ring_entries;
+	sq->kflags = ptr + p->sq_off.flags;
+	sq->kdropped = ptr + p->sq_off.dropped;
+	sq->array = ptr + p->sq_off.array;
+
+	size = p->sq_entries * sizeof(struct io_uring_sqe),
+	sq->sqes = mmap(0, size, PROT_READ | PROT_WRITE,
+				MAP_SHARED | MAP_POPULATE, fd,
+				IORING_OFF_SQES);
+	if (sq->sqes == MAP_FAILED) {
+		ret = -errno;
+err:
+		munmap(sq->khead, sq->ring_sz);
+		return ret;
+	}
+
+	cq->ring_sz = p->cq_off.cqes + p->cq_entries * sizeof(struct io_uring_cqe);
+	ptr = mmap(0, cq->ring_sz, PROT_READ | PROT_WRITE,
+			MAP_SHARED | MAP_POPULATE, fd, IORING_OFF_CQ_RING);
+	if (ptr == MAP_FAILED) {
+		ret = -errno;
+		munmap(sq->sqes, p->sq_entries * sizeof(struct io_uring_sqe));
+		goto err;
+	}
+	cq->khead = ptr + p->cq_off.head;
+	cq->ktail = ptr + p->cq_off.tail;
+	cq->kring_mask = ptr + p->cq_off.ring_mask;
+	cq->kring_entries = ptr + p->cq_off.ring_entries;
+	cq->koverflow = ptr + p->cq_off.overflow;
+	cq->cqes = ptr + p->cq_off.cqes;
+	return 0;
+}
+
+/*
+ * For users that want to specify sq_thread_cpu or sq_thread_idle, this
+ * interface is a convenient helper for mmap()ing the rings.
+ * Returns -1 on error, or zero on success.  On success, 'ring'
+ * contains the necessary information to read/write to the rings.
+ */
+int io_uring_queue_mmap(int fd, struct io_uring_params *p, struct io_uring *ring)
+{
+	int ret;
+
+	memset(ring, 0, sizeof(*ring));
+	ret = io_uring_mmap(fd, p, &ring->sq, &ring->cq);
+	if (!ret)
+		ring->ring_fd = fd;
+	return ret;
+}
+
+/*
+ * Returns -1 on error, or zero on success. On success, 'ring'
+ * contains the necessary information to read/write to the rings.
+ */
+int io_uring_queue_init(unsigned entries, struct io_uring *ring, unsigned flags)
+{
+	struct io_uring_params p;
+	int fd;
+
+	memset(&p, 0, sizeof(p));
+	p.flags = flags;
+
+	fd = io_uring_setup(entries, &p);
+	if (fd < 0)
+		return fd;
+
+	return io_uring_queue_mmap(fd, &p, ring);
+}
+
+void io_uring_queue_exit(struct io_uring *ring)
+{
+	struct io_uring_sq *sq = &ring->sq;
+	struct io_uring_cq *cq = &ring->cq;
+
+	munmap(sq->sqes, *sq->kring_entries * sizeof(struct io_uring_sqe));
+	munmap(sq->khead, sq->ring_sz);
+	munmap(cq->khead, cq->ring_sz);
+	close(ring->ring_fd);
+}

tools/io_uring/syscall.c

@@ -0,0 +1,40 @@
+/*
+ * Will go away once libc support is there
+ */
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <sys/uio.h>
+#include <signal.h>
+#include "liburing.h"
+
+#if defined(__x86_64) || defined(__i386__)
+#ifndef __NR_sys_io_uring_setup
+#define __NR_sys_io_uring_setup		425
+#endif
+#ifndef __NR_sys_io_uring_enter
+#define __NR_sys_io_uring_enter		426
+#endif
+#ifndef __NR_sys_io_uring_register
+#define __NR_sys_io_uring_register	427
+#endif
+#else
+#error "Arch not supported yet"
+#endif
+
+int io_uring_register(int fd, unsigned int opcode, void *arg,
+		      unsigned int nr_args)
+{
+	return syscall(__NR_sys_io_uring_register, fd, opcode, arg, nr_args);
+}
+
+int io_uring_setup(unsigned entries, struct io_uring_params *p)
+{
+	return syscall(__NR_sys_io_uring_setup, entries, p);
+}
+
+int io_uring_enter(unsigned fd, unsigned to_submit, unsigned min_complete,
+		   unsigned flags, sigset_t *sig)
+{
+	return syscall(__NR_sys_io_uring_enter, fd, to_submit, min_complete,
+			flags, sig, _NSIG / 8);
+}