xref: /openbmc/qemu/block/io_uring.c (revision e0091133)
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 "qapi/error.h"
19 #include "trace.h"
20 
21 /* io_uring ring size */
22 #define MAX_ENTRIES 128
23 
24 typedef struct LuringAIOCB {
25     Coroutine *co;
26     struct io_uring_sqe sqeq;
27     ssize_t ret;
28     QEMUIOVector *qiov;
29     bool is_read;
30     QSIMPLEQ_ENTRY(LuringAIOCB) next;
31 
32     /*
33      * Buffered reads may require resubmission, see
34      * luring_resubmit_short_read().
35      */
36     int total_read;
37     QEMUIOVector resubmit_qiov;
38 } LuringAIOCB;
39 
40 typedef struct LuringQueue {
41     int plugged;
42     unsigned int in_queue;
43     unsigned int in_flight;
44     bool blocked;
45     QSIMPLEQ_HEAD(, LuringAIOCB) submit_queue;
46 } LuringQueue;
47 
48 typedef struct LuringState {
49     AioContext *aio_context;
50 
51     struct io_uring ring;
52 
53     /* io queue for submit at batch.  Protected by AioContext lock. */
54     LuringQueue io_q;
55 
56     /* I/O completion processing.  Only runs in I/O thread.  */
57     QEMUBH *completion_bh;
58 } LuringState;
59 
60 /**
61  * luring_resubmit:
62  *
63  * Resubmit a request by appending it to submit_queue.  The caller must ensure
64  * that ioq_submit() is called later so that submit_queue requests are started.
65  */
66 static void luring_resubmit(LuringState *s, LuringAIOCB *luringcb)
67 {
68     QSIMPLEQ_INSERT_TAIL(&s->io_q.submit_queue, luringcb, next);
69     s->io_q.in_queue++;
70 }
71 
72 /**
73  * luring_resubmit_short_read:
74  *
75  * Short reads are rare but may occur. The remaining read request needs to be
76  * resubmitted.
77  */
78 static void luring_resubmit_short_read(LuringState *s, LuringAIOCB *luringcb,
79                                        int nread)
80 {
81     QEMUIOVector *resubmit_qiov;
82     size_t remaining;
83 
84     trace_luring_resubmit_short_read(s, luringcb, nread);
85 
86     /* Update read position */
87     luringcb->total_read += nread;
88     remaining = luringcb->qiov->size - luringcb->total_read;
89 
90     /* Shorten qiov */
91     resubmit_qiov = &luringcb->resubmit_qiov;
92     if (resubmit_qiov->iov == NULL) {
93         qemu_iovec_init(resubmit_qiov, luringcb->qiov->niov);
94     } else {
95         qemu_iovec_reset(resubmit_qiov);
96     }
97     qemu_iovec_concat(resubmit_qiov, luringcb->qiov, luringcb->total_read,
98                       remaining);
99 
100     /* Update sqe */
101     luringcb->sqeq.off += nread;
102     luringcb->sqeq.addr = (__u64)(uintptr_t)luringcb->resubmit_qiov.iov;
103     luringcb->sqeq.len = luringcb->resubmit_qiov.niov;
104 
105     luring_resubmit(s, luringcb);
106 }
107 
108 /**
109  * luring_process_completions:
110  * @s: AIO state
111  *
112  * Fetches completed I/O requests, consumes cqes and invokes their callbacks
113  * The function is somewhat tricky because it supports nested event loops, for
114  * example when a request callback invokes aio_poll().
115  *
116  * Function schedules BH completion so it  can be called again in a nested
117  * event loop.  When there are no events left  to complete the BH is being
118  * canceled.
119  *
120  */
121 static void luring_process_completions(LuringState *s)
122 {
123     struct io_uring_cqe *cqes;
124     int total_bytes;
125     /*
126      * Request completion callbacks can run the nested event loop.
127      * Schedule ourselves so the nested event loop will "see" remaining
128      * completed requests and process them.  Without this, completion
129      * callbacks that wait for other requests using a nested event loop
130      * would hang forever.
131      *
132      * This workaround is needed because io_uring uses poll_wait, which
133      * is woken up when new events are added to the uring, thus polling on
134      * the same uring fd will block unless more events are received.
135      *
136      * Other leaf block drivers (drivers that access the data themselves)
137      * are networking based, so they poll sockets for data and run the
138      * correct coroutine.
139      */
140     qemu_bh_schedule(s->completion_bh);
141 
142     while (io_uring_peek_cqe(&s->ring, &cqes) == 0) {
143         LuringAIOCB *luringcb;
144         int ret;
145 
146         if (!cqes) {
147             break;
148         }
149 
150         luringcb = io_uring_cqe_get_data(cqes);
151         ret = cqes->res;
152         io_uring_cqe_seen(&s->ring, cqes);
153         cqes = NULL;
154 
155         /* Change counters one-by-one because we can be nested. */
156         s->io_q.in_flight--;
157         trace_luring_process_completion(s, luringcb, ret);
158 
159         /* total_read is non-zero only for resubmitted read requests */
160         total_bytes = ret + luringcb->total_read;
161 
162         if (ret < 0) {
163             /*
164              * Only writev/readv/fsync requests on regular files or host block
165              * devices are submitted. Therefore -EAGAIN is not expected but it's
166              * known to happen sometimes with Linux SCSI. Submit again and hope
167              * the request completes successfully.
168              *
169              * For more information, see:
170              * https://lore.kernel.org/io-uring/20210727165811.284510-3-axboe@kernel.dk/T/#u
171              *
172              * If the code is changed to submit other types of requests in the
173              * future, then this workaround may need to be extended to deal with
174              * genuine -EAGAIN results that should not be resubmitted
175              * immediately.
176              */
177             if (ret == -EINTR || ret == -EAGAIN) {
178                 luring_resubmit(s, luringcb);
179                 continue;
180             }
181         } else if (!luringcb->qiov) {
182             goto end;
183         } else if (total_bytes == luringcb->qiov->size) {
184             ret = 0;
185         /* Only read/write */
186         } else {
187             /* Short Read/Write */
188             if (luringcb->is_read) {
189                 if (ret > 0) {
190                     luring_resubmit_short_read(s, luringcb, ret);
191                     continue;
192                 } else {
193                     /* Pad with zeroes */
194                     qemu_iovec_memset(luringcb->qiov, total_bytes, 0,
195                                       luringcb->qiov->size - total_bytes);
196                     ret = 0;
197                 }
198             } else {
199                 ret = -ENOSPC;
200             }
201         }
202 end:
203         luringcb->ret = ret;
204         qemu_iovec_destroy(&luringcb->resubmit_qiov);
205 
206         /*
207          * If the coroutine is already entered it must be in ioq_submit()
208          * and will notice luringcb->ret has been filled in when it
209          * eventually runs later. Coroutines cannot be entered recursively
210          * so avoid doing that!
211          */
212         if (!qemu_coroutine_entered(luringcb->co)) {
213             aio_co_wake(luringcb->co);
214         }
215     }
216     qemu_bh_cancel(s->completion_bh);
217 }
218 
219 static int ioq_submit(LuringState *s)
220 {
221     int ret = 0;
222     LuringAIOCB *luringcb, *luringcb_next;
223 
224     while (s->io_q.in_queue > 0) {
225         /*
226          * Try to fetch sqes from the ring for requests waiting in
227          * the overflow queue
228          */
229         QSIMPLEQ_FOREACH_SAFE(luringcb, &s->io_q.submit_queue, next,
230                               luringcb_next) {
231             struct io_uring_sqe *sqes = io_uring_get_sqe(&s->ring);
232             if (!sqes) {
233                 break;
234             }
235             /* Prep sqe for submission */
236             *sqes = luringcb->sqeq;
237             QSIMPLEQ_REMOVE_HEAD(&s->io_q.submit_queue, next);
238         }
239         ret = io_uring_submit(&s->ring);
240         trace_luring_io_uring_submit(s, ret);
241         /* Prevent infinite loop if submission is refused */
242         if (ret <= 0) {
243             if (ret == -EAGAIN || ret == -EINTR) {
244                 continue;
245             }
246             break;
247         }
248         s->io_q.in_flight += ret;
249         s->io_q.in_queue  -= ret;
250     }
251     s->io_q.blocked = (s->io_q.in_queue > 0);
252 
253     if (s->io_q.in_flight) {
254         /*
255          * We can try to complete something just right away if there are
256          * still requests in-flight.
257          */
258         luring_process_completions(s);
259     }
260     return ret;
261 }
262 
263 static void luring_process_completions_and_submit(LuringState *s)
264 {
265     aio_context_acquire(s->aio_context);
266     luring_process_completions(s);
267 
268     if (!s->io_q.plugged && s->io_q.in_queue > 0) {
269         ioq_submit(s);
270     }
271     aio_context_release(s->aio_context);
272 }
273 
274 static void qemu_luring_completion_bh(void *opaque)
275 {
276     LuringState *s = opaque;
277     luring_process_completions_and_submit(s);
278 }
279 
280 static void qemu_luring_completion_cb(void *opaque)
281 {
282     LuringState *s = opaque;
283     luring_process_completions_and_submit(s);
284 }
285 
286 static bool qemu_luring_poll_cb(void *opaque)
287 {
288     LuringState *s = opaque;
289 
290     return io_uring_cq_ready(&s->ring);
291 }
292 
293 static void qemu_luring_poll_ready(void *opaque)
294 {
295     LuringState *s = opaque;
296 
297     luring_process_completions_and_submit(s);
298 }
299 
300 static void ioq_init(LuringQueue *io_q)
301 {
302     QSIMPLEQ_INIT(&io_q->submit_queue);
303     io_q->plugged = 0;
304     io_q->in_queue = 0;
305     io_q->in_flight = 0;
306     io_q->blocked = false;
307 }
308 
309 void luring_io_plug(BlockDriverState *bs, LuringState *s)
310 {
311     trace_luring_io_plug(s);
312     s->io_q.plugged++;
313 }
314 
315 void luring_io_unplug(BlockDriverState *bs, LuringState *s)
316 {
317     assert(s->io_q.plugged);
318     trace_luring_io_unplug(s, s->io_q.blocked, s->io_q.plugged,
319                            s->io_q.in_queue, s->io_q.in_flight);
320     if (--s->io_q.plugged == 0 &&
321         !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_READ:
349         io_uring_prep_readv(sqes, fd, luringcb->qiov->iov,
350                             luringcb->qiov->niov, offset);
351         break;
352     case QEMU_AIO_FLUSH:
353         io_uring_prep_fsync(sqes, fd, IORING_FSYNC_DATASYNC);
354         break;
355     default:
356         fprintf(stderr, "%s: invalid AIO request type, aborting 0x%x.\n",
357                         __func__, type);
358         abort();
359     }
360     io_uring_sqe_set_data(sqes, luringcb);
361 
362     QSIMPLEQ_INSERT_TAIL(&s->io_q.submit_queue, luringcb, next);
363     s->io_q.in_queue++;
364     trace_luring_do_submit(s, s->io_q.blocked, s->io_q.plugged,
365                            s->io_q.in_queue, s->io_q.in_flight);
366     if (!s->io_q.blocked &&
367         (!s->io_q.plugged ||
368          s->io_q.in_flight + s->io_q.in_queue >= MAX_ENTRIES)) {
369         ret = ioq_submit(s);
370         trace_luring_do_submit_done(s, ret);
371         return ret;
372     }
373     return 0;
374 }
375 
376 int coroutine_fn luring_co_submit(BlockDriverState *bs, LuringState *s, int fd,
377                                   uint64_t offset, QEMUIOVector *qiov, int type)
378 {
379     int ret;
380     LuringAIOCB luringcb = {
381         .co         = qemu_coroutine_self(),
382         .ret        = -EINPROGRESS,
383         .qiov       = qiov,
384         .is_read    = (type == QEMU_AIO_READ),
385     };
386     trace_luring_co_submit(bs, s, &luringcb, fd, offset, qiov ? qiov->size : 0,
387                            type);
388     ret = luring_do_submit(fd, &luringcb, s, offset, type);
389 
390     if (ret < 0) {
391         return ret;
392     }
393 
394     if (luringcb.ret == -EINPROGRESS) {
395         qemu_coroutine_yield();
396     }
397     return luringcb.ret;
398 }
399 
400 void luring_detach_aio_context(LuringState *s, AioContext *old_context)
401 {
402     aio_set_fd_handler(old_context, s->ring.ring_fd, false,
403                        NULL, NULL, NULL, NULL, s);
404     qemu_bh_delete(s->completion_bh);
405     s->aio_context = NULL;
406 }
407 
408 void luring_attach_aio_context(LuringState *s, AioContext *new_context)
409 {
410     s->aio_context = new_context;
411     s->completion_bh = aio_bh_new(new_context, qemu_luring_completion_bh, s);
412     aio_set_fd_handler(s->aio_context, s->ring.ring_fd, false,
413                        qemu_luring_completion_cb, NULL,
414                        qemu_luring_poll_cb, qemu_luring_poll_ready, s);
415 }
416 
417 LuringState *luring_init(Error **errp)
418 {
419     int rc;
420     LuringState *s = g_new0(LuringState, 1);
421     struct io_uring *ring = &s->ring;
422 
423     trace_luring_init_state(s, sizeof(*s));
424 
425     rc = io_uring_queue_init(MAX_ENTRIES, ring, 0);
426     if (rc < 0) {
427         error_setg_errno(errp, errno, "failed to init linux io_uring ring");
428         g_free(s);
429         return NULL;
430     }
431 
432     ioq_init(&s->io_q);
433     return s;
434 
435 }
436 
437 void luring_cleanup(LuringState *s)
438 {
439     io_uring_queue_exit(&s->ring);
440     trace_luring_cleanup_state(s);
441     g_free(s);
442 }
443