githubmirror/liburing

src/queue.c
403 строки10 KB
Jens Axboe
src/queue: don't flush SQ ring for new wait interface
09 авг 2021, 06:44
09 авг 2021, 06:440ea4ccd
/* SPDX-License-Identifier: MIT */
#define _POSIX_C_SOURCE 200112L

#include <sys/types.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <unistd.h>
#include <errno.h>
#include <string.h>
#include <stdbool.h>

#include "liburing/compat.h"
#include "liburing/io_uring.h"
#include "liburing.h"
#include "liburing/barrier.h"

#include "syscall.h"

/*
 * Returns true if we're not using SQ thread (thus nobody submits but us)
 * or if IORING_SQ_NEED_WAKEUP is set, so submit thread must be explicitly
 * awakened. For the latter case, we set the thread wakeup flag.
 */
static inline bool sq_ring_needs_enter(struct io_uring *ring, unsigned *flags)
{
	if (!(ring->flags & IORING_SETUP_SQPOLL))
		return true;

	if (uring_unlikely(IO_URING_READ_ONCE(*ring->sq.kflags) &
			   IORING_SQ_NEED_WAKEUP)) {
		*flags |= IORING_ENTER_SQ_WAKEUP;
		return true;
	}

	return false;
}

static inline bool cq_ring_needs_flush(struct io_uring *ring)
{
	return IO_URING_READ_ONCE(*ring->sq.kflags) & IORING_SQ_CQ_OVERFLOW;
}

static int __io_uring_peek_cqe(struct io_uring *ring,
			       struct io_uring_cqe **cqe_ptr,
			       unsigned *nr_available)
{
	struct io_uring_cqe *cqe;
	int err = 0;
	unsigned available;
	unsigned mask = *ring->cq.kring_mask;

	do {
		unsigned tail = io_uring_smp_load_acquire(ring->cq.ktail);
		unsigned head = *ring->cq.khead;

		cqe = NULL;
		available = tail - head;
		if (!available)
			break;

		cqe = &ring->cq.cqes[head & mask];
		if (!(ring->features & IORING_FEAT_EXT_ARG) &&
				cqe->user_data == LIBURING_UDATA_TIMEOUT) {
			if (cqe->res < 0)
				err = cqe->res;
			io_uring_cq_advance(ring, 1);
			if (!err)
				continue;
			cqe = NULL;
		}

		break;
	} while (1);

	*cqe_ptr = cqe;
	*nr_available = available;
	return err;
}

struct get_data {
	unsigned submit;
	unsigned wait_nr;
	unsigned get_flags;
	int sz;
	void *arg;
};

static int _io_uring_get_cqe(struct io_uring *ring, struct io_uring_cqe **cqe_ptr,
			     struct get_data *data)
{
	struct io_uring_cqe *cqe = NULL;
	int err;

	do {
		bool need_enter = false;
		bool cq_overflow_flush = false;
		unsigned flags = 0;
		unsigned nr_available;
		int ret;

		err = __io_uring_peek_cqe(ring, &cqe, &nr_available);
		if (err)
			break;
		if (!cqe && !data->wait_nr && !data->submit) {
			if (!cq_ring_needs_flush(ring)) {
				err = -EAGAIN;
				break;
			}
			cq_overflow_flush = true;
		}
		if (data->wait_nr > nr_available || cq_overflow_flush) {
			flags = IORING_ENTER_GETEVENTS | data->get_flags;
			need_enter = true;
		}
		if (data->submit) {
			sq_ring_needs_enter(ring, &flags);
			need_enter = true;
		}
		if (!need_enter)
			break;

		ret = __sys_io_uring_enter2(ring->ring_fd, data->submit,
				data->wait_nr, flags, data->arg,
				data->sz);
		if (ret < 0) {
			err = -errno;
			break;
		}

		data->submit -= ret;
		if (cqe)
			break;
	} while (1);

	*cqe_ptr = cqe;
	return err;
}

int __io_uring_get_cqe(struct io_uring *ring, struct io_uring_cqe **cqe_ptr,
		       unsigned submit, unsigned wait_nr, sigset_t *sigmask)
{
	struct get_data data = {
		.submit		= submit,
		.wait_nr 	= wait_nr,
		.get_flags	= 0,
		.sz		= _NSIG / 8,
		.arg		= sigmask,
	};

	return _io_uring_get_cqe(ring, cqe_ptr, &data);
}

/*
 * Fill in an array of IO completions up to count, if any are available.
 * Returns the amount of IO completions filled.
 */
unsigned io_uring_peek_batch_cqe(struct io_uring *ring,
				 struct io_uring_cqe **cqes, unsigned count)
{
	unsigned ready;
	bool overflow_checked = false;

again:
	ready = io_uring_cq_ready(ring);
	if (ready) {
		unsigned head = *ring->cq.khead;
		unsigned mask = *ring->cq.kring_mask;
		unsigned last;
		int i = 0;

		count = count > ready ? ready : count;
		last = head + count;
		for (;head != last; head++, i++)
			cqes[i] = &ring->cq.cqes[head & mask];

		return count;
	}

	if (overflow_checked)
		goto done;

	if (cq_ring_needs_flush(ring)) {
		__sys_io_uring_enter(ring->ring_fd, 0, 0,
				     IORING_ENTER_GETEVENTS, NULL);
		overflow_checked = true;
		goto again;
	}

done:
	return 0;
}

/*
 * Sync internal state with kernel ring state on the SQ side. Returns the
 * number of pending items in the SQ ring, for the shared ring.
 */
int __io_uring_flush_sq(struct io_uring *ring)
{
	struct io_uring_sq *sq = &ring->sq;
	const unsigned mask = *sq->kring_mask;
	unsigned ktail = *sq->ktail;
	unsigned to_submit = sq->sqe_tail - sq->sqe_head;

	if (!to_submit)
		goto out;

	/*
	 * Fill in sqes that we have queued up, adding them to the kernel ring
	 */
	do {
		sq->array[ktail & mask] = sq->sqe_head & mask;
		ktail++;
		sq->sqe_head++;
	} while (--to_submit);

	/*
	 * Ensure that the kernel sees the SQE updates before it sees the tail
	 * update.
	 */
	io_uring_smp_store_release(sq->ktail, ktail);
out:
	/*
	 * This _may_ look problematic, as we're not supposed to be reading
	 * SQ->head without acquire semantics. When we're in SQPOLL mode, the
	 * kernel submitter could be updating this right now. For non-SQPOLL,
	 * task itself does it, and there's no potential race. But even for
	 * SQPOLL, the load is going to be potentially out-of-date the very
	 * instant it's done, regardless or whether or not it's done
	 * atomically. Worst case, we're going to be over-estimating what
	 * we can submit. The point is, we need to be able to deal with this
	 * situation regardless of any perceived atomicity.
	 */
	return ktail - *sq->khead;
}

/*
 * If we have kernel support for IORING_ENTER_EXT_ARG, then we can use that
 * more efficiently than queueing an internal timeout command.
 */
static int io_uring_wait_cqes_new(struct io_uring *ring,
				  struct io_uring_cqe **cqe_ptr,
				  unsigned wait_nr, struct __kernel_timespec *ts,
				  sigset_t *sigmask)
{
	struct io_uring_getevents_arg arg = {
		.sigmask	= (unsigned long) sigmask,
		.sigmask_sz	= _NSIG / 8,
		.ts		= (unsigned long) ts
	};
	struct get_data data = {
		.wait_nr	= wait_nr,
		.get_flags	= IORING_ENTER_EXT_ARG,
		.sz		= sizeof(arg),
		.arg		= &arg
	};

	return _io_uring_get_cqe(ring, cqe_ptr, &data);
}

/*
 * Like io_uring_wait_cqe(), except it accepts a timeout value as well. Note
 * that an sqe is used internally to handle the timeout. For kernel doesn't
 * support IORING_FEAT_EXT_ARG, applications using this function must never
 * set sqe->user_data to LIBURING_UDATA_TIMEOUT!
 *
 * For kernels without IORING_FEAT_EXT_ARG (5.10 and older), if 'ts' is
 * specified, the application need not call io_uring_submit() before
 * calling this function, as we will do that on its behalf. From this it also
 * follows that this function isn't safe to use for applications that split SQ
 * and CQ handling between two threads and expect that to work without
 * synchronization, as this function manipulates both the SQ and CQ side.
 *
 * For kernels with IORING_FEAT_EXT_ARG, no implicit submission is done and
 * hence this function is safe to use for applications that split SQ and CQ
 * handling between two threads.
 */
int io_uring_wait_cqes(struct io_uring *ring, struct io_uring_cqe **cqe_ptr,
		       unsigned wait_nr, struct __kernel_timespec *ts,
		       sigset_t *sigmask)
{
	unsigned to_submit = 0;

	if (ts) {
		struct io_uring_sqe *sqe;
		int ret;

		if (ring->features & IORING_FEAT_EXT_ARG)
			return io_uring_wait_cqes_new(ring, cqe_ptr, wait_nr,
							ts, sigmask);

		/*
		 * If the SQ ring is full, we may need to submit IO first
		 */
		sqe = io_uring_get_sqe(ring);
		if (!sqe) {
			ret = io_uring_submit(ring);
			if (ret < 0)
				return ret;
			sqe = io_uring_get_sqe(ring);
			if (!sqe)
				return -EAGAIN;
		}
		io_uring_prep_timeout(sqe, ts, wait_nr, 0);
		sqe->user_data = LIBURING_UDATA_TIMEOUT;
		to_submit = __io_uring_flush_sq(ring);
	}

	return __io_uring_get_cqe(ring, cqe_ptr, to_submit, wait_nr, sigmask);
}

/*
 * See io_uring_wait_cqes() - this function is the same, it just always uses
 * '1' as the wait_nr.
 */
int io_uring_wait_cqe_timeout(struct io_uring *ring,
			      struct io_uring_cqe **cqe_ptr,
			      struct __kernel_timespec *ts)
{
	return io_uring_wait_cqes(ring, cqe_ptr, 1, ts, NULL);
}

/*
 * Submit sqes acquired from io_uring_get_sqe() to the kernel.
 *
 * Returns number of sqes submitted
 */
static int __io_uring_submit(struct io_uring *ring, unsigned submitted,
			     unsigned wait_nr)
{
	unsigned flags;
	int ret;

	flags = 0;
	if (sq_ring_needs_enter(ring, &flags) || wait_nr) {
		if (wait_nr || (ring->flags & IORING_SETUP_IOPOLL))
			flags |= IORING_ENTER_GETEVENTS;

		ret = __sys_io_uring_enter(ring->ring_fd, submitted, wait_nr,
						flags, NULL);
		if (ret < 0)
			return -errno;
	} else
		ret = submitted;

	return ret;
}

static int __io_uring_submit_and_wait(struct io_uring *ring, unsigned wait_nr)
{
	return __io_uring_submit(ring, __io_uring_flush_sq(ring), wait_nr);
}

/*
 * Submit sqes acquired from io_uring_get_sqe() to the kernel.
 *
 * Returns number of sqes submitted
 */
int io_uring_submit(struct io_uring *ring)
{
	return __io_uring_submit_and_wait(ring, 0);
}

/*
 * Like io_uring_submit(), but allows waiting for events as well.
 *
 * Returns number of sqes submitted
 */
int io_uring_submit_and_wait(struct io_uring *ring, unsigned wait_nr)
{
	return __io_uring_submit_and_wait(ring, wait_nr);
}

/*
 * 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 int head = io_uring_smp_load_acquire(sq->khead);
	unsigned int next = sq->sqe_tail + 1;
	struct io_uring_sqe *sqe = NULL;

	if (next - head <= *sq->kring_entries) {
		sqe = &sq->sqes[sq->sqe_tail & *sq->kring_mask];
		sq->sqe_tail = next;
	}
	return sqe;
}

int __io_uring_sqring_wait(struct io_uring *ring)
{
	int ret;

	ret = __sys_io_uring_enter(ring->ring_fd, 0, 0, IORING_ENTER_SQ_WAIT,
					NULL);
	if (ret < 0)
		ret = -errno;
	return ret;
}