1 /* 2 * Linux io_uring support. 3 * 4 * Copyright (C) 2009 IBM, Corp. 5 * Copyright (C) 2009 Red Hat, Inc. 6 * Copyright (C) 2019 Aarushi Mehta 7 * 8 * This work is licensed under the terms of the GNU GPL, version 2 or later. 9 * See the COPYING file in the top-level directory. 10 */ 11 #include "qemu/osdep.h" 12 #include <liburing.h> 13 #include "block/aio.h" 14 #include "qemu/queue.h" 15 #include "block/block.h" 16 #include "block/raw-aio.h" 17 #include "qemu/coroutine.h" 18 #include "qemu/defer-call.h" 19 #include "qapi/error.h" 20 #include "sysemu/block-backend.h" 21 #include "trace.h" 22 23 /* Only used for assertions. */ 24 #include "qemu/coroutine_int.h" 25 26 /* io_uring ring size */ 27 #define MAX_ENTRIES 128 28 29 typedef struct LuringAIOCB { 30 Coroutine *co; 31 struct io_uring_sqe sqeq; 32 ssize_t ret; 33 QEMUIOVector *qiov; 34 bool is_read; 35 QSIMPLEQ_ENTRY(LuringAIOCB) next; 36 37 /* 38 * Buffered reads may require resubmission, see 39 * luring_resubmit_short_read(). 40 */ 41 int total_read; 42 QEMUIOVector resubmit_qiov; 43 } LuringAIOCB; 44 45 typedef struct LuringQueue { 46 unsigned int in_queue; 47 unsigned int in_flight; 48 bool blocked; 49 QSIMPLEQ_HEAD(, LuringAIOCB) submit_queue; 50 } LuringQueue; 51 52 struct LuringState { 53 AioContext *aio_context; 54 55 struct io_uring ring; 56 57 /* No locking required, only accessed from AioContext home thread */ 58 LuringQueue io_q; 59 60 QEMUBH *completion_bh; 61 }; 62 63 /** 64 * luring_resubmit: 65 * 66 * Resubmit a request by appending it to submit_queue. The caller must ensure 67 * that ioq_submit() is called later so that submit_queue requests are started. 68 */ 69 static void luring_resubmit(LuringState *s, LuringAIOCB *luringcb) 70 { 71 QSIMPLEQ_INSERT_TAIL(&s->io_q.submit_queue, luringcb, next); 72 s->io_q.in_queue++; 73 } 74 75 /** 76 * luring_resubmit_short_read: 77 * 78 * Short reads are rare but may occur. The remaining read request needs to be 79 * resubmitted. 80 */ 81 static void luring_resubmit_short_read(LuringState *s, LuringAIOCB *luringcb, 82 int nread) 83 { 84 QEMUIOVector *resubmit_qiov; 85 size_t remaining; 86 87 trace_luring_resubmit_short_read(s, luringcb, nread); 88 89 /* Update read position */ 90 luringcb->total_read += nread; 91 remaining = luringcb->qiov->size - luringcb->total_read; 92 93 /* Shorten qiov */ 94 resubmit_qiov = &luringcb->resubmit_qiov; 95 if (resubmit_qiov->iov == NULL) { 96 qemu_iovec_init(resubmit_qiov, luringcb->qiov->niov); 97 } else { 98 qemu_iovec_reset(resubmit_qiov); 99 } 100 qemu_iovec_concat(resubmit_qiov, luringcb->qiov, luringcb->total_read, 101 remaining); 102 103 /* Update sqe */ 104 luringcb->sqeq.off += nread; 105 luringcb->sqeq.addr = (uintptr_t)luringcb->resubmit_qiov.iov; 106 luringcb->sqeq.len = luringcb->resubmit_qiov.niov; 107 108 luring_resubmit(s, luringcb); 109 } 110 111 /** 112 * luring_process_completions: 113 * @s: AIO state 114 * 115 * Fetches completed I/O requests, consumes cqes and invokes their callbacks 116 * The function is somewhat tricky because it supports nested event loops, for 117 * example when a request callback invokes aio_poll(). 118 * 119 * Function schedules BH completion so it can be called again in a nested 120 * event loop. When there are no events left to complete the BH is being 121 * canceled. 122 * 123 */ 124 static void luring_process_completions(LuringState *s) 125 { 126 struct io_uring_cqe *cqes; 127 int total_bytes; 128 129 defer_call_begin(); 130 131 /* 132 * Request completion callbacks can run the nested event loop. 133 * Schedule ourselves so the nested event loop will "see" remaining 134 * completed requests and process them. Without this, completion 135 * callbacks that wait for other requests using a nested event loop 136 * would hang forever. 137 * 138 * This workaround is needed because io_uring uses poll_wait, which 139 * is woken up when new events are added to the uring, thus polling on 140 * the same uring fd will block unless more events are received. 141 * 142 * Other leaf block drivers (drivers that access the data themselves) 143 * are networking based, so they poll sockets for data and run the 144 * correct coroutine. 145 */ 146 qemu_bh_schedule(s->completion_bh); 147 148 while (io_uring_peek_cqe(&s->ring, &cqes) == 0) { 149 LuringAIOCB *luringcb; 150 int ret; 151 152 if (!cqes) { 153 break; 154 } 155 156 luringcb = io_uring_cqe_get_data(cqes); 157 ret = cqes->res; 158 io_uring_cqe_seen(&s->ring, cqes); 159 cqes = NULL; 160 161 /* Change counters one-by-one because we can be nested. */ 162 s->io_q.in_flight--; 163 trace_luring_process_completion(s, luringcb, ret); 164 165 /* total_read is non-zero only for resubmitted read requests */ 166 total_bytes = ret + luringcb->total_read; 167 168 if (ret < 0) { 169 /* 170 * Only writev/readv/fsync requests on regular files or host block 171 * devices are submitted. Therefore -EAGAIN is not expected but it's 172 * known to happen sometimes with Linux SCSI. Submit again and hope 173 * the request completes successfully. 174 * 175 * For more information, see: 176 * https://lore.kernel.org/io-uring/20210727165811.284510-3-axboe@kernel.dk/T/#u 177 * 178 * If the code is changed to submit other types of requests in the 179 * future, then this workaround may need to be extended to deal with 180 * genuine -EAGAIN results that should not be resubmitted 181 * immediately. 182 */ 183 if (ret == -EINTR || ret == -EAGAIN) { 184 luring_resubmit(s, luringcb); 185 continue; 186 } 187 } else if (!luringcb->qiov) { 188 goto end; 189 } else if (total_bytes == luringcb->qiov->size) { 190 ret = 0; 191 /* Only read/write */ 192 } else { 193 /* Short Read/Write */ 194 if (luringcb->is_read) { 195 if (ret > 0) { 196 luring_resubmit_short_read(s, luringcb, ret); 197 continue; 198 } else { 199 /* Pad with zeroes */ 200 qemu_iovec_memset(luringcb->qiov, total_bytes, 0, 201 luringcb->qiov->size - total_bytes); 202 ret = 0; 203 } 204 } else { 205 ret = -ENOSPC; 206 } 207 } 208 end: 209 luringcb->ret = ret; 210 qemu_iovec_destroy(&luringcb->resubmit_qiov); 211 212 /* 213 * If the coroutine is already entered it must be in ioq_submit() 214 * and will notice luringcb->ret has been filled in when it 215 * eventually runs later. Coroutines cannot be entered recursively 216 * so avoid doing that! 217 */ 218 assert(luringcb->co->ctx == s->aio_context); 219 if (!qemu_coroutine_entered(luringcb->co)) { 220 aio_co_wake(luringcb->co); 221 } 222 } 223 224 qemu_bh_cancel(s->completion_bh); 225 226 defer_call_end(); 227 } 228 229 static int ioq_submit(LuringState *s) 230 { 231 int ret = 0; 232 LuringAIOCB *luringcb, *luringcb_next; 233 234 while (s->io_q.in_queue > 0) { 235 /* 236 * Try to fetch sqes from the ring for requests waiting in 237 * the overflow queue 238 */ 239 QSIMPLEQ_FOREACH_SAFE(luringcb, &s->io_q.submit_queue, next, 240 luringcb_next) { 241 struct io_uring_sqe *sqes = io_uring_get_sqe(&s->ring); 242 if (!sqes) { 243 break; 244 } 245 /* Prep sqe for submission */ 246 *sqes = luringcb->sqeq; 247 QSIMPLEQ_REMOVE_HEAD(&s->io_q.submit_queue, next); 248 } 249 ret = io_uring_submit(&s->ring); 250 trace_luring_io_uring_submit(s, ret); 251 /* Prevent infinite loop if submission is refused */ 252 if (ret <= 0) { 253 if (ret == -EAGAIN || ret == -EINTR) { 254 continue; 255 } 256 break; 257 } 258 s->io_q.in_flight += ret; 259 s->io_q.in_queue -= ret; 260 } 261 s->io_q.blocked = (s->io_q.in_queue > 0); 262 263 if (s->io_q.in_flight) { 264 /* 265 * We can try to complete something just right away if there are 266 * still requests in-flight. 267 */ 268 luring_process_completions(s); 269 } 270 return ret; 271 } 272 273 static void luring_process_completions_and_submit(LuringState *s) 274 { 275 luring_process_completions(s); 276 277 if (s->io_q.in_queue > 0) { 278 ioq_submit(s); 279 } 280 } 281 282 static void qemu_luring_completion_bh(void *opaque) 283 { 284 LuringState *s = opaque; 285 luring_process_completions_and_submit(s); 286 } 287 288 static void qemu_luring_completion_cb(void *opaque) 289 { 290 LuringState *s = opaque; 291 luring_process_completions_and_submit(s); 292 } 293 294 static bool qemu_luring_poll_cb(void *opaque) 295 { 296 LuringState *s = opaque; 297 298 return io_uring_cq_ready(&s->ring); 299 } 300 301 static void qemu_luring_poll_ready(void *opaque) 302 { 303 LuringState *s = opaque; 304 305 luring_process_completions_and_submit(s); 306 } 307 308 static void ioq_init(LuringQueue *io_q) 309 { 310 QSIMPLEQ_INIT(&io_q->submit_queue); 311 io_q->in_queue = 0; 312 io_q->in_flight = 0; 313 io_q->blocked = false; 314 } 315 316 static void luring_deferred_fn(void *opaque) 317 { 318 LuringState *s = opaque; 319 trace_luring_unplug_fn(s, s->io_q.blocked, s->io_q.in_queue, 320 s->io_q.in_flight); 321 if (!s->io_q.blocked && s->io_q.in_queue > 0) { 322 ioq_submit(s); 323 } 324 } 325 326 /** 327 * luring_do_submit: 328 * @fd: file descriptor for I/O 329 * @luringcb: AIO control block 330 * @s: AIO state 331 * @offset: offset for request 332 * @type: type of request 333 * 334 * Fetches sqes from ring, adds to pending queue and preps them 335 * 336 */ 337 static int luring_do_submit(int fd, LuringAIOCB *luringcb, LuringState *s, 338 uint64_t offset, int type) 339 { 340 int ret; 341 struct io_uring_sqe *sqes = &luringcb->sqeq; 342 343 switch (type) { 344 case QEMU_AIO_WRITE: 345 io_uring_prep_writev(sqes, fd, luringcb->qiov->iov, 346 luringcb->qiov->niov, offset); 347 break; 348 case QEMU_AIO_ZONE_APPEND: 349 io_uring_prep_writev(sqes, fd, luringcb->qiov->iov, 350 luringcb->qiov->niov, offset); 351 break; 352 case QEMU_AIO_READ: 353 io_uring_prep_readv(sqes, fd, luringcb->qiov->iov, 354 luringcb->qiov->niov, offset); 355 break; 356 case QEMU_AIO_FLUSH: 357 io_uring_prep_fsync(sqes, fd, IORING_FSYNC_DATASYNC); 358 break; 359 default: 360 fprintf(stderr, "%s: invalid AIO request type, aborting 0x%x.\n", 361 __func__, type); 362 abort(); 363 } 364 io_uring_sqe_set_data(sqes, luringcb); 365 366 QSIMPLEQ_INSERT_TAIL(&s->io_q.submit_queue, luringcb, next); 367 s->io_q.in_queue++; 368 trace_luring_do_submit(s, s->io_q.blocked, s->io_q.in_queue, 369 s->io_q.in_flight); 370 if (!s->io_q.blocked) { 371 if (s->io_q.in_flight + s->io_q.in_queue >= MAX_ENTRIES) { 372 ret = ioq_submit(s); 373 trace_luring_do_submit_done(s, ret); 374 return ret; 375 } 376 377 defer_call(luring_deferred_fn, s); 378 } 379 return 0; 380 } 381 382 int coroutine_fn luring_co_submit(BlockDriverState *bs, int fd, uint64_t offset, 383 QEMUIOVector *qiov, int type) 384 { 385 int ret; 386 AioContext *ctx = qemu_get_current_aio_context(); 387 LuringState *s = aio_get_linux_io_uring(ctx); 388 LuringAIOCB luringcb = { 389 .co = qemu_coroutine_self(), 390 .ret = -EINPROGRESS, 391 .qiov = qiov, 392 .is_read = (type == QEMU_AIO_READ), 393 }; 394 trace_luring_co_submit(bs, s, &luringcb, fd, offset, qiov ? qiov->size : 0, 395 type); 396 ret = luring_do_submit(fd, &luringcb, s, offset, type); 397 398 if (ret < 0) { 399 return ret; 400 } 401 402 if (luringcb.ret == -EINPROGRESS) { 403 qemu_coroutine_yield(); 404 } 405 return luringcb.ret; 406 } 407 408 void luring_detach_aio_context(LuringState *s, AioContext *old_context) 409 { 410 aio_set_fd_handler(old_context, s->ring.ring_fd, 411 NULL, NULL, NULL, NULL, s); 412 qemu_bh_delete(s->completion_bh); 413 s->aio_context = NULL; 414 } 415 416 void luring_attach_aio_context(LuringState *s, AioContext *new_context) 417 { 418 s->aio_context = new_context; 419 s->completion_bh = aio_bh_new(new_context, qemu_luring_completion_bh, s); 420 aio_set_fd_handler(s->aio_context, s->ring.ring_fd, 421 qemu_luring_completion_cb, NULL, 422 qemu_luring_poll_cb, qemu_luring_poll_ready, s); 423 } 424 425 LuringState *luring_init(Error **errp) 426 { 427 int rc; 428 LuringState *s = g_new0(LuringState, 1); 429 struct io_uring *ring = &s->ring; 430 431 trace_luring_init_state(s, sizeof(*s)); 432 433 rc = io_uring_queue_init(MAX_ENTRIES, ring, 0); 434 if (rc < 0) { 435 error_setg_errno(errp, errno, "failed to init linux io_uring ring"); 436 g_free(s); 437 return NULL; 438 } 439 440 ioq_init(&s->io_q); 441 return s; 442 443 } 444 445 void luring_cleanup(LuringState *s) 446 { 447 io_uring_queue_exit(&s->ring); 448 trace_luring_cleanup_state(s); 449 g_free(s); 450 } 451