1 /* 2 * Linux native AIO support. 3 * 4 * Copyright (C) 2009 IBM, Corp. 5 * Copyright (C) 2009 Red Hat, Inc. 6 * 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. 9 */ 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" 15 16 #include <libaio.h> 17 18 /* 19 * Queue size (per-device). 20 * 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. 25 */ 26 #define MAX_EVENTS 128 27 28 #define MAX_QUEUED_IO 128 29 30 struct qemu_laiocb { 31 BlockDriverAIOCB common; 32 struct qemu_laio_state *ctx; 33 struct iocb iocb; 34 ssize_t ret; 35 size_t nbytes; 36 QEMUIOVector *qiov; 37 bool is_read; 38 QLIST_ENTRY(qemu_laiocb) node; 39 }; 40 41 typedef struct { 42 struct iocb *iocbs[MAX_QUEUED_IO]; 43 int plugged; 44 unsigned int size; 45 unsigned int idx; 46 } LaioQueue; 47 48 struct qemu_laio_state { 49 io_context_t ctx; 50 EventNotifier e; 51 52 /* io queue for submit at batch */ 53 LaioQueue io_q; 54 }; 55 56 static inline ssize_t io_event_ret(struct io_event *ev) 57 { 58 return (ssize_t)(((uint64_t)ev->res2 << 32) | ev->res); 59 } 60 61 /* 62 * Completes an AIO request (calls the callback and frees the ACB). 63 */ 64 static void qemu_laio_process_completion(struct qemu_laio_state *s, 65 struct qemu_laiocb *laiocb) 66 { 67 int ret; 68 69 ret = laiocb->ret; 70 if (ret != -ECANCELED) { 71 if (ret == laiocb->nbytes) { 72 ret = 0; 73 } else if (ret >= 0) { 74 /* Short reads mean EOF, pad with zeros. */ 75 if (laiocb->is_read) { 76 qemu_iovec_memset(laiocb->qiov, ret, 0, 77 laiocb->qiov->size - ret); 78 } else { 79 ret = -EINVAL; 80 } 81 } 82 83 laiocb->common.cb(laiocb->common.opaque, ret); 84 } 85 86 qemu_aio_release(laiocb); 87 } 88 89 static void qemu_laio_completion_cb(EventNotifier *e) 90 { 91 struct qemu_laio_state *s = container_of(e, struct qemu_laio_state, e); 92 93 while (event_notifier_test_and_clear(&s->e)) { 94 struct io_event events[MAX_EVENTS]; 95 struct timespec ts = { 0 }; 96 int nevents, i; 97 98 do { 99 nevents = io_getevents(s->ctx, MAX_EVENTS, MAX_EVENTS, events, &ts); 100 } while (nevents == -EINTR); 101 102 for (i = 0; i < nevents; i++) { 103 struct iocb *iocb = events[i].obj; 104 struct qemu_laiocb *laiocb = 105 container_of(iocb, struct qemu_laiocb, iocb); 106 107 laiocb->ret = io_event_ret(&events[i]); 108 qemu_laio_process_completion(s, laiocb); 109 } 110 } 111 } 112 113 static void laio_cancel(BlockDriverAIOCB *blockacb) 114 { 115 struct qemu_laiocb *laiocb = (struct qemu_laiocb *)blockacb; 116 struct io_event event; 117 int ret; 118 119 if (laiocb->ret != -EINPROGRESS) 120 return; 121 122 /* 123 * Note that as of Linux 2.6.31 neither the block device code nor any 124 * filesystem implements cancellation of AIO request. 125 * Thus the polling loop below is the normal code path. 126 */ 127 ret = io_cancel(laiocb->ctx->ctx, &laiocb->iocb, &event); 128 if (ret == 0) { 129 laiocb->ret = -ECANCELED; 130 return; 131 } 132 133 /* 134 * We have to wait for the iocb to finish. 135 * 136 * The only way to get the iocb status update is by polling the io context. 137 * We might be able to do this slightly more optimal by removing the 138 * O_NONBLOCK flag. 139 */ 140 while (laiocb->ret == -EINPROGRESS) { 141 qemu_laio_completion_cb(&laiocb->ctx->e); 142 } 143 } 144 145 static const AIOCBInfo laio_aiocb_info = { 146 .aiocb_size = sizeof(struct qemu_laiocb), 147 .cancel = laio_cancel, 148 }; 149 150 static void ioq_init(LaioQueue *io_q) 151 { 152 io_q->size = MAX_QUEUED_IO; 153 io_q->idx = 0; 154 io_q->plugged = 0; 155 } 156 157 static int ioq_submit(struct qemu_laio_state *s) 158 { 159 int ret, i = 0; 160 int len = s->io_q.idx; 161 162 do { 163 ret = io_submit(s->ctx, len, s->io_q.iocbs); 164 } while (i++ < 3 && ret == -EAGAIN); 165 166 /* empty io queue */ 167 s->io_q.idx = 0; 168 169 if (ret < 0) { 170 i = 0; 171 } else { 172 i = ret; 173 } 174 175 for (; i < len; i++) { 176 struct qemu_laiocb *laiocb = 177 container_of(s->io_q.iocbs[i], struct qemu_laiocb, iocb); 178 179 laiocb->ret = (ret < 0) ? ret : -EIO; 180 qemu_laio_process_completion(s, laiocb); 181 } 182 return ret; 183 } 184 185 static void ioq_enqueue(struct qemu_laio_state *s, struct iocb *iocb) 186 { 187 unsigned int idx = s->io_q.idx; 188 189 s->io_q.iocbs[idx++] = iocb; 190 s->io_q.idx = idx; 191 192 /* submit immediately if queue is full */ 193 if (idx == s->io_q.size) { 194 ioq_submit(s); 195 } 196 } 197 198 void laio_io_plug(BlockDriverState *bs, void *aio_ctx) 199 { 200 struct qemu_laio_state *s = aio_ctx; 201 202 s->io_q.plugged++; 203 } 204 205 int laio_io_unplug(BlockDriverState *bs, void *aio_ctx, bool unplug) 206 { 207 struct qemu_laio_state *s = aio_ctx; 208 int ret = 0; 209 210 assert(s->io_q.plugged > 0 || !unplug); 211 212 if (unplug && --s->io_q.plugged > 0) { 213 return 0; 214 } 215 216 if (s->io_q.idx > 0) { 217 ret = ioq_submit(s); 218 } 219 220 return ret; 221 } 222 223 BlockDriverAIOCB *laio_submit(BlockDriverState *bs, void *aio_ctx, int fd, 224 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, 225 BlockDriverCompletionFunc *cb, void *opaque, int type) 226 { 227 struct qemu_laio_state *s = aio_ctx; 228 struct qemu_laiocb *laiocb; 229 struct iocb *iocbs; 230 off_t offset = sector_num * 512; 231 232 laiocb = qemu_aio_get(&laio_aiocb_info, bs, cb, opaque); 233 laiocb->nbytes = nb_sectors * 512; 234 laiocb->ctx = s; 235 laiocb->ret = -EINPROGRESS; 236 laiocb->is_read = (type == QEMU_AIO_READ); 237 laiocb->qiov = qiov; 238 239 iocbs = &laiocb->iocb; 240 241 switch (type) { 242 case QEMU_AIO_WRITE: 243 io_prep_pwritev(iocbs, fd, qiov->iov, qiov->niov, offset); 244 break; 245 case QEMU_AIO_READ: 246 io_prep_preadv(iocbs, fd, qiov->iov, qiov->niov, offset); 247 break; 248 /* Currently Linux kernel does not support other operations */ 249 default: 250 fprintf(stderr, "%s: invalid AIO request type 0x%x.\n", 251 __func__, type); 252 goto out_free_aiocb; 253 } 254 io_set_eventfd(&laiocb->iocb, event_notifier_get_fd(&s->e)); 255 256 if (!s->io_q.plugged) { 257 if (io_submit(s->ctx, 1, &iocbs) < 0) { 258 goto out_free_aiocb; 259 } 260 } else { 261 ioq_enqueue(s, iocbs); 262 } 263 return &laiocb->common; 264 265 out_free_aiocb: 266 qemu_aio_release(laiocb); 267 return NULL; 268 } 269 270 void laio_detach_aio_context(void *s_, AioContext *old_context) 271 { 272 struct qemu_laio_state *s = s_; 273 274 aio_set_event_notifier(old_context, &s->e, NULL); 275 } 276 277 void laio_attach_aio_context(void *s_, AioContext *new_context) 278 { 279 struct qemu_laio_state *s = s_; 280 281 aio_set_event_notifier(new_context, &s->e, qemu_laio_completion_cb); 282 } 283 284 void *laio_init(void) 285 { 286 struct qemu_laio_state *s; 287 288 s = g_malloc0(sizeof(*s)); 289 if (event_notifier_init(&s->e, false) < 0) { 290 goto out_free_state; 291 } 292 293 if (io_setup(MAX_EVENTS, &s->ctx) != 0) { 294 goto out_close_efd; 295 } 296 297 ioq_init(&s->io_q); 298 299 return s; 300 301 out_close_efd: 302 event_notifier_cleanup(&s->e); 303 out_free_state: 304 g_free(s); 305 return NULL; 306 } 307 308 void laio_cleanup(void *s_) 309 { 310 struct qemu_laio_state *s = s_; 311 312 event_notifier_cleanup(&s->e); 313 314 if (io_destroy(s->ctx) != 0) { 315 fprintf(stderr, "%s: destroy AIO context %p failed\n", 316 __func__, &s->ctx); 317 } 318 g_free(s); 319 } 320