2 * Linux native AIO support.
4 * Copyright (C) 2009 IBM, Corp.
5 * Copyright (C) 2009 Red Hat, Inc.
7 * This work is licensed under the terms of the GNU GPL, version 2 or later.
8 * See the COPYING file in the top-level directory.
10 #include "qemu-common.h"
11 #include "block/aio.h"
12 #include "qemu/queue.h"
13 #include "block/raw-aio.h"
14 #include "qemu/event_notifier.h"
19 * Queue size (per-device).
21 * XXX: eventually we need to communicate this to the guest and/or make it
22 * tunable by the guest. If we get more outstanding requests at a time
23 * than this we will get EAGAIN from io_submit which is communicated to
24 * the guest as an I/O error.
26 #define MAX_EVENTS 128
28 #define MAX_QUEUED_IO 128
32 struct qemu_laio_state *ctx;
38 QLIST_ENTRY(qemu_laiocb) node;
42 struct iocb *iocbs[MAX_QUEUED_IO];
48 struct qemu_laio_state {
52 /* io queue for submit at batch */
55 /* I/O completion processing */
56 QEMUBH *completion_bh;
57 struct io_event events[MAX_EVENTS];
62 static inline ssize_t io_event_ret(struct io_event *ev)
64 return (ssize_t)(((uint64_t)ev->res2 << 32) | ev->res);
68 * Completes an AIO request (calls the callback and frees the ACB).
70 static void qemu_laio_process_completion(struct qemu_laio_state *s,
71 struct qemu_laiocb *laiocb)
76 if (ret != -ECANCELED) {
77 if (ret == laiocb->nbytes) {
79 } else if (ret >= 0) {
80 /* Short reads mean EOF, pad with zeros. */
81 if (laiocb->is_read) {
82 qemu_iovec_memset(laiocb->qiov, ret, 0,
83 laiocb->qiov->size - ret);
89 laiocb->common.cb(laiocb->common.opaque, ret);
91 qemu_aio_unref(laiocb);
94 /* The completion BH fetches completed I/O requests and invokes their
97 * The function is somewhat tricky because it supports nested event loops, for
98 * example when a request callback invokes aio_poll(). In order to do this,
99 * the completion events array and index are kept in qemu_laio_state. The BH
100 * reschedules itself as long as there are completions pending so it will
101 * either be called again in a nested event loop or will be called after all
102 * events have been completed. When there are no events left to complete, the
103 * BH returns without rescheduling.
105 static void qemu_laio_completion_bh(void *opaque)
107 struct qemu_laio_state *s = opaque;
109 /* Fetch more completion events when empty */
110 if (s->event_idx == s->event_max) {
112 struct timespec ts = { 0 };
113 s->event_max = io_getevents(s->ctx, MAX_EVENTS, MAX_EVENTS,
115 } while (s->event_max == -EINTR);
118 if (s->event_max <= 0) {
120 return; /* no more events */
124 /* Reschedule so nested event loops see currently pending completions */
125 qemu_bh_schedule(s->completion_bh);
127 /* Process completion events */
128 while (s->event_idx < s->event_max) {
129 struct iocb *iocb = s->events[s->event_idx].obj;
130 struct qemu_laiocb *laiocb =
131 container_of(iocb, struct qemu_laiocb, iocb);
133 laiocb->ret = io_event_ret(&s->events[s->event_idx]);
136 qemu_laio_process_completion(s, laiocb);
140 static void qemu_laio_completion_cb(EventNotifier *e)
142 struct qemu_laio_state *s = container_of(e, struct qemu_laio_state, e);
144 if (event_notifier_test_and_clear(&s->e)) {
145 qemu_bh_schedule(s->completion_bh);
149 static void laio_cancel(BlockAIOCB *blockacb)
151 struct qemu_laiocb *laiocb = (struct qemu_laiocb *)blockacb;
152 struct io_event event;
155 if (laiocb->ret != -EINPROGRESS) {
158 ret = io_cancel(laiocb->ctx->ctx, &laiocb->iocb, &event);
159 laiocb->ret = -ECANCELED;
161 /* iocb is not cancelled, cb will be called by the event loop later */
165 laiocb->common.cb(laiocb->common.opaque, laiocb->ret);
168 static const AIOCBInfo laio_aiocb_info = {
169 .aiocb_size = sizeof(struct qemu_laiocb),
170 .cancel_async = laio_cancel,
173 static void ioq_init(LaioQueue *io_q)
175 io_q->size = MAX_QUEUED_IO;
180 static int ioq_submit(struct qemu_laio_state *s)
183 int len = s->io_q.idx;
186 ret = io_submit(s->ctx, len, s->io_q.iocbs);
187 } while (i++ < 3 && ret == -EAGAIN);
198 for (; i < len; i++) {
199 struct qemu_laiocb *laiocb =
200 container_of(s->io_q.iocbs[i], struct qemu_laiocb, iocb);
202 laiocb->ret = (ret < 0) ? ret : -EIO;
203 qemu_laio_process_completion(s, laiocb);
208 static void ioq_enqueue(struct qemu_laio_state *s, struct iocb *iocb)
210 unsigned int idx = s->io_q.idx;
212 s->io_q.iocbs[idx++] = iocb;
215 /* submit immediately if queue is full */
216 if (idx == s->io_q.size) {
221 void laio_io_plug(BlockDriverState *bs, void *aio_ctx)
223 struct qemu_laio_state *s = aio_ctx;
228 int laio_io_unplug(BlockDriverState *bs, void *aio_ctx, bool unplug)
230 struct qemu_laio_state *s = aio_ctx;
233 assert(s->io_q.plugged > 0 || !unplug);
235 if (unplug && --s->io_q.plugged > 0) {
239 if (s->io_q.idx > 0) {
246 BlockAIOCB *laio_submit(BlockDriverState *bs, void *aio_ctx, int fd,
247 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
248 BlockCompletionFunc *cb, void *opaque, int type)
250 struct qemu_laio_state *s = aio_ctx;
251 struct qemu_laiocb *laiocb;
253 off_t offset = sector_num * 512;
255 laiocb = qemu_aio_get(&laio_aiocb_info, bs, cb, opaque);
256 laiocb->nbytes = nb_sectors * 512;
258 laiocb->ret = -EINPROGRESS;
259 laiocb->is_read = (type == QEMU_AIO_READ);
262 iocbs = &laiocb->iocb;
266 io_prep_pwritev(iocbs, fd, qiov->iov, qiov->niov, offset);
269 io_prep_preadv(iocbs, fd, qiov->iov, qiov->niov, offset);
271 /* Currently Linux kernel does not support other operations */
273 fprintf(stderr, "%s: invalid AIO request type 0x%x.\n",
277 io_set_eventfd(&laiocb->iocb, event_notifier_get_fd(&s->e));
279 if (!s->io_q.plugged) {
280 if (io_submit(s->ctx, 1, &iocbs) < 0) {
284 ioq_enqueue(s, iocbs);
286 return &laiocb->common;
289 qemu_aio_unref(laiocb);
293 void laio_detach_aio_context(void *s_, AioContext *old_context)
295 struct qemu_laio_state *s = s_;
297 aio_set_event_notifier(old_context, &s->e, NULL);
298 qemu_bh_delete(s->completion_bh);
301 void laio_attach_aio_context(void *s_, AioContext *new_context)
303 struct qemu_laio_state *s = s_;
305 s->completion_bh = aio_bh_new(new_context, qemu_laio_completion_bh, s);
306 aio_set_event_notifier(new_context, &s->e, qemu_laio_completion_cb);
309 void *laio_init(void)
311 struct qemu_laio_state *s;
313 s = g_malloc0(sizeof(*s));
314 if (event_notifier_init(&s->e, false) < 0) {
318 if (io_setup(MAX_EVENTS, &s->ctx) != 0) {
327 event_notifier_cleanup(&s->e);
333 void laio_cleanup(void *s_)
335 struct qemu_laio_state *s = s_;
337 event_notifier_cleanup(&s->e);
339 if (io_destroy(s->ctx) != 0) {
340 fprintf(stderr, "%s: destroy AIO context %p failed\n",