]>
Commit | Line | Data |
---|---|---|
fd66dbd4 SH |
1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
2 | /* | |
3 | * libblkio BlockDriver | |
4 | * | |
5 | * Copyright Red Hat, Inc. | |
6 | * | |
7 | * Author: | |
8 | * Stefan Hajnoczi <stefanha@redhat.com> | |
9 | */ | |
10 | ||
11 | #include "qemu/osdep.h" | |
12 | #include <blkio.h> | |
13 | #include "block/block_int.h" | |
14 | #include "qapi/error.h" | |
15 | #include "qapi/qmp/qdict.h" | |
16 | #include "qemu/module.h" | |
17 | ||
18 | /* | |
19 | * Keep the QEMU BlockDriver names identical to the libblkio driver names. | |
20 | * Using macros instead of typing out the string literals avoids typos. | |
21 | */ | |
22 | #define DRIVER_IO_URING "io_uring" | |
23 | #define DRIVER_NVME_IO_URING "nvme-io_uring" | |
24 | #define DRIVER_VIRTIO_BLK_VHOST_USER "virtio-blk-vhost-user" | |
25 | #define DRIVER_VIRTIO_BLK_VHOST_VDPA "virtio-blk-vhost-vdpa" | |
26 | ||
27 | /* | |
28 | * Allocated bounce buffers are kept in a list sorted by buffer address. | |
29 | */ | |
30 | typedef struct BlkioBounceBuf { | |
31 | QLIST_ENTRY(BlkioBounceBuf) next; | |
32 | ||
33 | /* The bounce buffer */ | |
34 | struct iovec buf; | |
35 | } BlkioBounceBuf; | |
36 | ||
37 | typedef struct { | |
38 | /* | |
39 | * libblkio is not thread-safe so this lock protects ->blkio and | |
40 | * ->blkioq. | |
41 | */ | |
42 | QemuMutex blkio_lock; | |
43 | struct blkio *blkio; | |
44 | struct blkioq *blkioq; /* make this multi-queue in the future... */ | |
45 | int completion_fd; | |
46 | ||
47 | /* | |
48 | * Polling fetches the next completion into this field. | |
49 | * | |
50 | * No lock is necessary since only one thread calls aio_poll() and invokes | |
51 | * fd and poll handlers. | |
52 | */ | |
53 | struct blkio_completion poll_completion; | |
54 | ||
55 | /* | |
56 | * Protects ->bounce_pool, ->bounce_bufs, ->bounce_available. | |
57 | * | |
58 | * Lock ordering: ->bounce_lock before ->blkio_lock. | |
59 | */ | |
60 | CoMutex bounce_lock; | |
61 | ||
62 | /* Bounce buffer pool */ | |
63 | struct blkio_mem_region bounce_pool; | |
64 | ||
65 | /* Sorted list of allocated bounce buffers */ | |
66 | QLIST_HEAD(, BlkioBounceBuf) bounce_bufs; | |
67 | ||
68 | /* Queue for coroutines waiting for bounce buffer space */ | |
69 | CoQueue bounce_available; | |
70 | ||
71 | /* The value of the "mem-region-alignment" property */ | |
72 | size_t mem_region_alignment; | |
73 | ||
74 | /* Can we skip adding/deleting blkio_mem_regions? */ | |
75 | bool needs_mem_regions; | |
76 | } BDRVBlkioState; | |
77 | ||
78 | /* Called with s->bounce_lock held */ | |
79 | static int blkio_resize_bounce_pool(BDRVBlkioState *s, int64_t bytes) | |
80 | { | |
81 | /* There can be no allocated bounce buffers during resize */ | |
82 | assert(QLIST_EMPTY(&s->bounce_bufs)); | |
83 | ||
84 | /* Pad size to reduce frequency of resize calls */ | |
85 | bytes += 128 * 1024; | |
86 | ||
87 | WITH_QEMU_LOCK_GUARD(&s->blkio_lock) { | |
88 | int ret; | |
89 | ||
90 | if (s->bounce_pool.addr) { | |
91 | blkio_unmap_mem_region(s->blkio, &s->bounce_pool); | |
92 | blkio_free_mem_region(s->blkio, &s->bounce_pool); | |
93 | memset(&s->bounce_pool, 0, sizeof(s->bounce_pool)); | |
94 | } | |
95 | ||
96 | /* Automatically freed when s->blkio is destroyed */ | |
97 | ret = blkio_alloc_mem_region(s->blkio, &s->bounce_pool, bytes); | |
98 | if (ret < 0) { | |
99 | return ret; | |
100 | } | |
101 | ||
102 | ret = blkio_map_mem_region(s->blkio, &s->bounce_pool); | |
103 | if (ret < 0) { | |
104 | blkio_free_mem_region(s->blkio, &s->bounce_pool); | |
105 | memset(&s->bounce_pool, 0, sizeof(s->bounce_pool)); | |
106 | return ret; | |
107 | } | |
108 | } | |
109 | ||
110 | return 0; | |
111 | } | |
112 | ||
113 | /* Called with s->bounce_lock held */ | |
114 | static bool | |
115 | blkio_do_alloc_bounce_buffer(BDRVBlkioState *s, BlkioBounceBuf *bounce, | |
116 | int64_t bytes) | |
117 | { | |
118 | void *addr = s->bounce_pool.addr; | |
119 | BlkioBounceBuf *cur = NULL; | |
120 | BlkioBounceBuf *prev = NULL; | |
121 | ptrdiff_t space; | |
122 | ||
123 | /* | |
124 | * This is just a linear search over the holes between requests. An | |
125 | * efficient allocator would be nice. | |
126 | */ | |
127 | QLIST_FOREACH(cur, &s->bounce_bufs, next) { | |
128 | space = cur->buf.iov_base - addr; | |
129 | if (bytes <= space) { | |
130 | QLIST_INSERT_BEFORE(cur, bounce, next); | |
131 | bounce->buf.iov_base = addr; | |
132 | bounce->buf.iov_len = bytes; | |
133 | return true; | |
134 | } | |
135 | ||
136 | addr = cur->buf.iov_base + cur->buf.iov_len; | |
137 | prev = cur; | |
138 | } | |
139 | ||
140 | /* Is there space after the last request? */ | |
141 | space = s->bounce_pool.addr + s->bounce_pool.len - addr; | |
142 | if (bytes > space) { | |
143 | return false; | |
144 | } | |
145 | if (prev) { | |
146 | QLIST_INSERT_AFTER(prev, bounce, next); | |
147 | } else { | |
148 | QLIST_INSERT_HEAD(&s->bounce_bufs, bounce, next); | |
149 | } | |
150 | bounce->buf.iov_base = addr; | |
151 | bounce->buf.iov_len = bytes; | |
152 | return true; | |
153 | } | |
154 | ||
155 | static int coroutine_fn | |
156 | blkio_alloc_bounce_buffer(BDRVBlkioState *s, BlkioBounceBuf *bounce, | |
157 | int64_t bytes) | |
158 | { | |
159 | /* | |
160 | * Ensure fairness: first time around we join the back of the queue, | |
161 | * subsequently we join the front so we don't lose our place. | |
162 | */ | |
163 | CoQueueWaitFlags wait_flags = 0; | |
164 | ||
165 | QEMU_LOCK_GUARD(&s->bounce_lock); | |
166 | ||
167 | /* Ensure fairness: don't even try if other requests are already waiting */ | |
168 | if (!qemu_co_queue_empty(&s->bounce_available)) { | |
169 | qemu_co_queue_wait_flags(&s->bounce_available, &s->bounce_lock, | |
170 | wait_flags); | |
171 | wait_flags = CO_QUEUE_WAIT_FRONT; | |
172 | } | |
173 | ||
174 | while (true) { | |
175 | if (blkio_do_alloc_bounce_buffer(s, bounce, bytes)) { | |
176 | /* Kick the next queued request since there may be space */ | |
177 | qemu_co_queue_next(&s->bounce_available); | |
178 | return 0; | |
179 | } | |
180 | ||
181 | /* | |
182 | * If there are no in-flight requests then the pool was simply too | |
183 | * small. | |
184 | */ | |
185 | if (QLIST_EMPTY(&s->bounce_bufs)) { | |
186 | bool ok; | |
187 | int ret; | |
188 | ||
189 | ret = blkio_resize_bounce_pool(s, bytes); | |
190 | if (ret < 0) { | |
191 | /* Kick the next queued request since that may fail too */ | |
192 | qemu_co_queue_next(&s->bounce_available); | |
193 | return ret; | |
194 | } | |
195 | ||
196 | ok = blkio_do_alloc_bounce_buffer(s, bounce, bytes); | |
197 | assert(ok); /* must have space this time */ | |
198 | return 0; | |
199 | } | |
200 | ||
201 | qemu_co_queue_wait_flags(&s->bounce_available, &s->bounce_lock, | |
202 | wait_flags); | |
203 | wait_flags = CO_QUEUE_WAIT_FRONT; | |
204 | } | |
205 | } | |
206 | ||
207 | static void coroutine_fn blkio_free_bounce_buffer(BDRVBlkioState *s, | |
208 | BlkioBounceBuf *bounce) | |
209 | { | |
210 | QEMU_LOCK_GUARD(&s->bounce_lock); | |
211 | ||
212 | QLIST_REMOVE(bounce, next); | |
213 | ||
214 | /* Wake up waiting coroutines since space may now be available */ | |
215 | qemu_co_queue_next(&s->bounce_available); | |
216 | } | |
217 | ||
218 | /* For async to .bdrv_co_*() conversion */ | |
219 | typedef struct { | |
220 | Coroutine *coroutine; | |
221 | int ret; | |
222 | } BlkioCoData; | |
223 | ||
224 | static void blkio_completion_fd_read(void *opaque) | |
225 | { | |
226 | BlockDriverState *bs = opaque; | |
227 | BDRVBlkioState *s = bs->opaque; | |
228 | uint64_t val; | |
229 | int ret; | |
230 | ||
231 | /* Polling may have already fetched a completion */ | |
232 | if (s->poll_completion.user_data != NULL) { | |
233 | BlkioCoData *cod = s->poll_completion.user_data; | |
234 | cod->ret = s->poll_completion.ret; | |
235 | ||
236 | /* Clear it in case aio_co_wake() enters a nested event loop */ | |
237 | s->poll_completion.user_data = NULL; | |
238 | ||
239 | aio_co_wake(cod->coroutine); | |
240 | } | |
241 | ||
242 | /* Reset completion fd status */ | |
243 | ret = read(s->completion_fd, &val, sizeof(val)); | |
244 | ||
245 | /* Ignore errors, there's nothing we can do */ | |
246 | (void)ret; | |
247 | ||
248 | /* | |
249 | * Reading one completion at a time makes nested event loop re-entrancy | |
250 | * simple. Change this loop to get multiple completions in one go if it | |
251 | * becomes a performance bottleneck. | |
252 | */ | |
253 | while (true) { | |
254 | struct blkio_completion completion; | |
255 | ||
256 | WITH_QEMU_LOCK_GUARD(&s->blkio_lock) { | |
257 | ret = blkioq_do_io(s->blkioq, &completion, 0, 1, NULL); | |
258 | } | |
259 | if (ret != 1) { | |
260 | break; | |
261 | } | |
262 | ||
263 | BlkioCoData *cod = completion.user_data; | |
264 | cod->ret = completion.ret; | |
265 | aio_co_wake(cod->coroutine); | |
266 | } | |
267 | } | |
268 | ||
269 | static bool blkio_completion_fd_poll(void *opaque) | |
270 | { | |
271 | BlockDriverState *bs = opaque; | |
272 | BDRVBlkioState *s = bs->opaque; | |
273 | int ret; | |
274 | ||
275 | /* Just in case we already fetched a completion */ | |
276 | if (s->poll_completion.user_data != NULL) { | |
277 | return true; | |
278 | } | |
279 | ||
280 | WITH_QEMU_LOCK_GUARD(&s->blkio_lock) { | |
281 | ret = blkioq_do_io(s->blkioq, &s->poll_completion, 0, 1, NULL); | |
282 | } | |
283 | return ret == 1; | |
284 | } | |
285 | ||
286 | static void blkio_completion_fd_poll_ready(void *opaque) | |
287 | { | |
288 | blkio_completion_fd_read(opaque); | |
289 | } | |
290 | ||
291 | static void blkio_attach_aio_context(BlockDriverState *bs, | |
292 | AioContext *new_context) | |
293 | { | |
294 | BDRVBlkioState *s = bs->opaque; | |
295 | ||
296 | aio_set_fd_handler(new_context, | |
297 | s->completion_fd, | |
298 | false, | |
299 | blkio_completion_fd_read, | |
300 | NULL, | |
301 | blkio_completion_fd_poll, | |
302 | blkio_completion_fd_poll_ready, | |
303 | bs); | |
304 | } | |
305 | ||
306 | static void blkio_detach_aio_context(BlockDriverState *bs) | |
307 | { | |
308 | BDRVBlkioState *s = bs->opaque; | |
309 | ||
310 | aio_set_fd_handler(bdrv_get_aio_context(bs), | |
311 | s->completion_fd, | |
312 | false, NULL, NULL, NULL, NULL, NULL); | |
313 | } | |
314 | ||
315 | /* Call with s->blkio_lock held to submit I/O after enqueuing a new request */ | |
316 | static void blkio_submit_io(BlockDriverState *bs) | |
317 | { | |
318 | if (qatomic_read(&bs->io_plugged) == 0) { | |
319 | BDRVBlkioState *s = bs->opaque; | |
320 | ||
321 | blkioq_do_io(s->blkioq, NULL, 0, 0, NULL); | |
322 | } | |
323 | } | |
324 | ||
325 | static int coroutine_fn | |
326 | blkio_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
327 | { | |
328 | BDRVBlkioState *s = bs->opaque; | |
329 | BlkioCoData cod = { | |
330 | .coroutine = qemu_coroutine_self(), | |
331 | }; | |
332 | ||
333 | WITH_QEMU_LOCK_GUARD(&s->blkio_lock) { | |
334 | blkioq_discard(s->blkioq, offset, bytes, &cod, 0); | |
335 | blkio_submit_io(bs); | |
336 | } | |
337 | ||
338 | qemu_coroutine_yield(); | |
339 | return cod.ret; | |
340 | } | |
341 | ||
342 | static int coroutine_fn | |
343 | blkio_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
344 | QEMUIOVector *qiov, BdrvRequestFlags flags) | |
345 | { | |
346 | BlkioCoData cod = { | |
347 | .coroutine = qemu_coroutine_self(), | |
348 | }; | |
349 | BDRVBlkioState *s = bs->opaque; | |
350 | bool use_bounce_buffer = s->needs_mem_regions; | |
351 | BlkioBounceBuf bounce; | |
352 | struct iovec *iov = qiov->iov; | |
353 | int iovcnt = qiov->niov; | |
354 | ||
355 | if (use_bounce_buffer) { | |
356 | int ret = blkio_alloc_bounce_buffer(s, &bounce, bytes); | |
357 | if (ret < 0) { | |
358 | return ret; | |
359 | } | |
360 | ||
361 | iov = &bounce.buf; | |
362 | iovcnt = 1; | |
363 | } | |
364 | ||
365 | WITH_QEMU_LOCK_GUARD(&s->blkio_lock) { | |
366 | blkioq_readv(s->blkioq, offset, iov, iovcnt, &cod, 0); | |
367 | blkio_submit_io(bs); | |
368 | } | |
369 | ||
370 | qemu_coroutine_yield(); | |
371 | ||
372 | if (use_bounce_buffer) { | |
373 | if (cod.ret == 0) { | |
374 | qemu_iovec_from_buf(qiov, 0, | |
375 | bounce.buf.iov_base, | |
376 | bounce.buf.iov_len); | |
377 | } | |
378 | ||
379 | blkio_free_bounce_buffer(s, &bounce); | |
380 | } | |
381 | ||
382 | return cod.ret; | |
383 | } | |
384 | ||
385 | static int coroutine_fn blkio_co_pwritev(BlockDriverState *bs, int64_t offset, | |
386 | int64_t bytes, QEMUIOVector *qiov, BdrvRequestFlags flags) | |
387 | { | |
388 | uint32_t blkio_flags = (flags & BDRV_REQ_FUA) ? BLKIO_REQ_FUA : 0; | |
389 | BlkioCoData cod = { | |
390 | .coroutine = qemu_coroutine_self(), | |
391 | }; | |
392 | BDRVBlkioState *s = bs->opaque; | |
393 | bool use_bounce_buffer = s->needs_mem_regions; | |
394 | BlkioBounceBuf bounce; | |
395 | struct iovec *iov = qiov->iov; | |
396 | int iovcnt = qiov->niov; | |
397 | ||
398 | if (use_bounce_buffer) { | |
399 | int ret = blkio_alloc_bounce_buffer(s, &bounce, bytes); | |
400 | if (ret < 0) { | |
401 | return ret; | |
402 | } | |
403 | ||
404 | qemu_iovec_to_buf(qiov, 0, bounce.buf.iov_base, bytes); | |
405 | iov = &bounce.buf; | |
406 | iovcnt = 1; | |
407 | } | |
408 | ||
409 | WITH_QEMU_LOCK_GUARD(&s->blkio_lock) { | |
410 | blkioq_writev(s->blkioq, offset, iov, iovcnt, &cod, blkio_flags); | |
411 | blkio_submit_io(bs); | |
412 | } | |
413 | ||
414 | qemu_coroutine_yield(); | |
415 | ||
416 | if (use_bounce_buffer) { | |
417 | blkio_free_bounce_buffer(s, &bounce); | |
418 | } | |
419 | ||
420 | return cod.ret; | |
421 | } | |
422 | ||
423 | static int coroutine_fn blkio_co_flush(BlockDriverState *bs) | |
424 | { | |
425 | BDRVBlkioState *s = bs->opaque; | |
426 | BlkioCoData cod = { | |
427 | .coroutine = qemu_coroutine_self(), | |
428 | }; | |
429 | ||
430 | WITH_QEMU_LOCK_GUARD(&s->blkio_lock) { | |
431 | blkioq_flush(s->blkioq, &cod, 0); | |
432 | blkio_submit_io(bs); | |
433 | } | |
434 | ||
435 | qemu_coroutine_yield(); | |
436 | return cod.ret; | |
437 | } | |
438 | ||
439 | static int coroutine_fn blkio_co_pwrite_zeroes(BlockDriverState *bs, | |
440 | int64_t offset, int64_t bytes, BdrvRequestFlags flags) | |
441 | { | |
442 | BDRVBlkioState *s = bs->opaque; | |
443 | BlkioCoData cod = { | |
444 | .coroutine = qemu_coroutine_self(), | |
445 | }; | |
446 | uint32_t blkio_flags = 0; | |
447 | ||
448 | if (flags & BDRV_REQ_FUA) { | |
449 | blkio_flags |= BLKIO_REQ_FUA; | |
450 | } | |
451 | if (!(flags & BDRV_REQ_MAY_UNMAP)) { | |
452 | blkio_flags |= BLKIO_REQ_NO_UNMAP; | |
453 | } | |
454 | if (flags & BDRV_REQ_NO_FALLBACK) { | |
455 | blkio_flags |= BLKIO_REQ_NO_FALLBACK; | |
456 | } | |
457 | ||
458 | WITH_QEMU_LOCK_GUARD(&s->blkio_lock) { | |
459 | blkioq_write_zeroes(s->blkioq, offset, bytes, &cod, blkio_flags); | |
460 | blkio_submit_io(bs); | |
461 | } | |
462 | ||
463 | qemu_coroutine_yield(); | |
464 | return cod.ret; | |
465 | } | |
466 | ||
467 | static void blkio_io_unplug(BlockDriverState *bs) | |
468 | { | |
469 | BDRVBlkioState *s = bs->opaque; | |
470 | ||
471 | WITH_QEMU_LOCK_GUARD(&s->blkio_lock) { | |
472 | blkio_submit_io(bs); | |
473 | } | |
474 | } | |
475 | ||
476 | static int blkio_io_uring_open(BlockDriverState *bs, QDict *options, int flags, | |
477 | Error **errp) | |
478 | { | |
479 | const char *filename = qdict_get_str(options, "filename"); | |
480 | BDRVBlkioState *s = bs->opaque; | |
481 | int ret; | |
482 | ||
483 | ret = blkio_set_str(s->blkio, "path", filename); | |
484 | qdict_del(options, "filename"); | |
485 | if (ret < 0) { | |
486 | error_setg_errno(errp, -ret, "failed to set path: %s", | |
487 | blkio_get_error_msg()); | |
488 | return ret; | |
489 | } | |
490 | ||
491 | if (flags & BDRV_O_NOCACHE) { | |
492 | ret = blkio_set_bool(s->blkio, "direct", true); | |
493 | if (ret < 0) { | |
494 | error_setg_errno(errp, -ret, "failed to set direct: %s", | |
495 | blkio_get_error_msg()); | |
496 | return ret; | |
497 | } | |
498 | } | |
499 | ||
500 | return 0; | |
501 | } | |
502 | ||
503 | static int blkio_nvme_io_uring(BlockDriverState *bs, QDict *options, int flags, | |
504 | Error **errp) | |
505 | { | |
506 | const char *filename = qdict_get_str(options, "filename"); | |
507 | BDRVBlkioState *s = bs->opaque; | |
508 | int ret; | |
509 | ||
510 | ret = blkio_set_str(s->blkio, "path", filename); | |
511 | qdict_del(options, "filename"); | |
512 | if (ret < 0) { | |
513 | error_setg_errno(errp, -ret, "failed to set path: %s", | |
514 | blkio_get_error_msg()); | |
515 | return ret; | |
516 | } | |
517 | ||
518 | if (!(flags & BDRV_O_NOCACHE)) { | |
519 | error_setg(errp, "cache.direct=off is not supported"); | |
520 | return -EINVAL; | |
521 | } | |
522 | ||
523 | return 0; | |
524 | } | |
525 | ||
526 | static int blkio_virtio_blk_common_open(BlockDriverState *bs, | |
527 | QDict *options, int flags, Error **errp) | |
528 | { | |
529 | const char *path = qdict_get_try_str(options, "path"); | |
530 | BDRVBlkioState *s = bs->opaque; | |
531 | int ret; | |
532 | ||
533 | if (!path) { | |
534 | error_setg(errp, "missing 'path' option"); | |
535 | return -EINVAL; | |
536 | } | |
537 | ||
538 | ret = blkio_set_str(s->blkio, "path", path); | |
539 | qdict_del(options, "path"); | |
540 | if (ret < 0) { | |
541 | error_setg_errno(errp, -ret, "failed to set path: %s", | |
542 | blkio_get_error_msg()); | |
543 | return ret; | |
544 | } | |
545 | ||
546 | if (!(flags & BDRV_O_NOCACHE)) { | |
547 | error_setg(errp, "cache.direct=off is not supported"); | |
548 | return -EINVAL; | |
549 | } | |
550 | return 0; | |
551 | } | |
552 | ||
553 | static int blkio_file_open(BlockDriverState *bs, QDict *options, int flags, | |
554 | Error **errp) | |
555 | { | |
556 | const char *blkio_driver = bs->drv->protocol_name; | |
557 | BDRVBlkioState *s = bs->opaque; | |
558 | int ret; | |
559 | ||
560 | ret = blkio_create(blkio_driver, &s->blkio); | |
561 | if (ret < 0) { | |
562 | error_setg_errno(errp, -ret, "blkio_create failed: %s", | |
563 | blkio_get_error_msg()); | |
564 | return ret; | |
565 | } | |
566 | ||
567 | if (strcmp(blkio_driver, DRIVER_IO_URING) == 0) { | |
568 | ret = blkio_io_uring_open(bs, options, flags, errp); | |
569 | } else if (strcmp(blkio_driver, DRIVER_NVME_IO_URING) == 0) { | |
570 | ret = blkio_nvme_io_uring(bs, options, flags, errp); | |
571 | } else if (strcmp(blkio_driver, DRIVER_VIRTIO_BLK_VHOST_USER) == 0) { | |
572 | ret = blkio_virtio_blk_common_open(bs, options, flags, errp); | |
573 | } else if (strcmp(blkio_driver, DRIVER_VIRTIO_BLK_VHOST_VDPA) == 0) { | |
574 | ret = blkio_virtio_blk_common_open(bs, options, flags, errp); | |
575 | } else { | |
576 | g_assert_not_reached(); | |
577 | } | |
578 | if (ret < 0) { | |
579 | blkio_destroy(&s->blkio); | |
580 | return ret; | |
581 | } | |
582 | ||
583 | if (!(flags & BDRV_O_RDWR)) { | |
584 | ret = blkio_set_bool(s->blkio, "read-only", true); | |
585 | if (ret < 0) { | |
586 | error_setg_errno(errp, -ret, "failed to set read-only: %s", | |
587 | blkio_get_error_msg()); | |
588 | blkio_destroy(&s->blkio); | |
589 | return ret; | |
590 | } | |
591 | } | |
592 | ||
593 | ret = blkio_connect(s->blkio); | |
594 | if (ret < 0) { | |
595 | error_setg_errno(errp, -ret, "blkio_connect failed: %s", | |
596 | blkio_get_error_msg()); | |
597 | blkio_destroy(&s->blkio); | |
598 | return ret; | |
599 | } | |
600 | ||
601 | ret = blkio_get_bool(s->blkio, | |
602 | "needs-mem-regions", | |
603 | &s->needs_mem_regions); | |
604 | if (ret < 0) { | |
605 | error_setg_errno(errp, -ret, | |
606 | "failed to get needs-mem-regions: %s", | |
607 | blkio_get_error_msg()); | |
608 | blkio_destroy(&s->blkio); | |
609 | return ret; | |
610 | } | |
611 | ||
612 | ret = blkio_get_uint64(s->blkio, | |
613 | "mem-region-alignment", | |
614 | &s->mem_region_alignment); | |
615 | if (ret < 0) { | |
616 | error_setg_errno(errp, -ret, | |
617 | "failed to get mem-region-alignment: %s", | |
618 | blkio_get_error_msg()); | |
619 | blkio_destroy(&s->blkio); | |
620 | return ret; | |
621 | } | |
622 | ||
623 | ret = blkio_start(s->blkio); | |
624 | if (ret < 0) { | |
625 | error_setg_errno(errp, -ret, "blkio_start failed: %s", | |
626 | blkio_get_error_msg()); | |
627 | blkio_destroy(&s->blkio); | |
628 | return ret; | |
629 | } | |
630 | ||
631 | bs->supported_write_flags = BDRV_REQ_FUA; | |
632 | bs->supported_zero_flags = BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | | |
633 | BDRV_REQ_NO_FALLBACK; | |
634 | ||
635 | qemu_mutex_init(&s->blkio_lock); | |
636 | qemu_co_mutex_init(&s->bounce_lock); | |
637 | qemu_co_queue_init(&s->bounce_available); | |
638 | QLIST_INIT(&s->bounce_bufs); | |
639 | s->blkioq = blkio_get_queue(s->blkio, 0); | |
640 | s->completion_fd = blkioq_get_completion_fd(s->blkioq); | |
641 | ||
642 | blkio_attach_aio_context(bs, bdrv_get_aio_context(bs)); | |
643 | return 0; | |
644 | } | |
645 | ||
646 | static void blkio_close(BlockDriverState *bs) | |
647 | { | |
648 | BDRVBlkioState *s = bs->opaque; | |
649 | ||
650 | /* There is no destroy() API for s->bounce_lock */ | |
651 | ||
652 | qemu_mutex_destroy(&s->blkio_lock); | |
653 | blkio_detach_aio_context(bs); | |
654 | blkio_destroy(&s->blkio); | |
655 | } | |
656 | ||
657 | static int64_t blkio_getlength(BlockDriverState *bs) | |
658 | { | |
659 | BDRVBlkioState *s = bs->opaque; | |
660 | uint64_t capacity; | |
661 | int ret; | |
662 | ||
663 | WITH_QEMU_LOCK_GUARD(&s->blkio_lock) { | |
664 | ret = blkio_get_uint64(s->blkio, "capacity", &capacity); | |
665 | } | |
666 | if (ret < 0) { | |
667 | return -ret; | |
668 | } | |
669 | ||
670 | return capacity; | |
671 | } | |
672 | ||
673 | static int blkio_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
674 | { | |
675 | return 0; | |
676 | } | |
677 | ||
678 | static void blkio_refresh_limits(BlockDriverState *bs, Error **errp) | |
679 | { | |
680 | BDRVBlkioState *s = bs->opaque; | |
681 | QEMU_LOCK_GUARD(&s->blkio_lock); | |
682 | int value; | |
683 | int ret; | |
684 | ||
685 | ret = blkio_get_int(s->blkio, "request-alignment", &value); | |
686 | if (ret < 0) { | |
687 | error_setg_errno(errp, -ret, "failed to get \"request-alignment\": %s", | |
688 | blkio_get_error_msg()); | |
689 | return; | |
690 | } | |
691 | bs->bl.request_alignment = value; | |
692 | if (bs->bl.request_alignment < 1 || | |
693 | bs->bl.request_alignment >= INT_MAX || | |
694 | !is_power_of_2(bs->bl.request_alignment)) { | |
695 | error_setg(errp, "invalid \"request-alignment\" value %" PRIu32 ", " | |
696 | "must be a power of 2 less than INT_MAX", | |
697 | bs->bl.request_alignment); | |
698 | return; | |
699 | } | |
700 | ||
701 | ret = blkio_get_int(s->blkio, "optimal-io-size", &value); | |
702 | if (ret < 0) { | |
703 | error_setg_errno(errp, -ret, "failed to get \"optimal-io-size\": %s", | |
704 | blkio_get_error_msg()); | |
705 | return; | |
706 | } | |
707 | bs->bl.opt_transfer = value; | |
708 | if (bs->bl.opt_transfer > INT_MAX || | |
709 | (bs->bl.opt_transfer % bs->bl.request_alignment)) { | |
710 | error_setg(errp, "invalid \"optimal-io-size\" value %" PRIu32 ", must " | |
711 | "be a multiple of %" PRIu32, bs->bl.opt_transfer, | |
712 | bs->bl.request_alignment); | |
713 | return; | |
714 | } | |
715 | ||
716 | ret = blkio_get_int(s->blkio, "max-transfer", &value); | |
717 | if (ret < 0) { | |
718 | error_setg_errno(errp, -ret, "failed to get \"max-transfer\": %s", | |
719 | blkio_get_error_msg()); | |
720 | return; | |
721 | } | |
722 | bs->bl.max_transfer = value; | |
723 | if ((bs->bl.max_transfer % bs->bl.request_alignment) || | |
724 | (bs->bl.opt_transfer && (bs->bl.max_transfer % bs->bl.opt_transfer))) { | |
725 | error_setg(errp, "invalid \"max-transfer\" value %" PRIu32 ", must be " | |
726 | "a multiple of %" PRIu32 " and %" PRIu32 " (if non-zero)", | |
727 | bs->bl.max_transfer, bs->bl.request_alignment, | |
728 | bs->bl.opt_transfer); | |
729 | return; | |
730 | } | |
731 | ||
732 | ret = blkio_get_int(s->blkio, "buf-alignment", &value); | |
733 | if (ret < 0) { | |
734 | error_setg_errno(errp, -ret, "failed to get \"buf-alignment\": %s", | |
735 | blkio_get_error_msg()); | |
736 | return; | |
737 | } | |
738 | if (value < 1) { | |
739 | error_setg(errp, "invalid \"buf-alignment\" value %d, must be " | |
740 | "positive", value); | |
741 | return; | |
742 | } | |
743 | bs->bl.min_mem_alignment = value; | |
744 | ||
745 | ret = blkio_get_int(s->blkio, "optimal-buf-alignment", &value); | |
746 | if (ret < 0) { | |
747 | error_setg_errno(errp, -ret, | |
748 | "failed to get \"optimal-buf-alignment\": %s", | |
749 | blkio_get_error_msg()); | |
750 | return; | |
751 | } | |
752 | if (value < 1) { | |
753 | error_setg(errp, "invalid \"optimal-buf-alignment\" value %d, " | |
754 | "must be positive", value); | |
755 | return; | |
756 | } | |
757 | bs->bl.opt_mem_alignment = value; | |
758 | ||
759 | ret = blkio_get_int(s->blkio, "max-segments", &value); | |
760 | if (ret < 0) { | |
761 | error_setg_errno(errp, -ret, "failed to get \"max-segments\": %s", | |
762 | blkio_get_error_msg()); | |
763 | return; | |
764 | } | |
765 | if (value < 1) { | |
766 | error_setg(errp, "invalid \"max-segments\" value %d, must be positive", | |
767 | value); | |
768 | return; | |
769 | } | |
770 | bs->bl.max_iov = value; | |
771 | } | |
772 | ||
773 | /* | |
774 | * TODO | |
775 | * Missing libblkio APIs: | |
776 | * - block_status | |
777 | * - co_invalidate_cache | |
778 | * | |
779 | * Out of scope? | |
780 | * - create | |
781 | * - truncate | |
782 | */ | |
783 | ||
784 | #define BLKIO_DRIVER(name, ...) \ | |
785 | { \ | |
786 | .format_name = name, \ | |
787 | .protocol_name = name, \ | |
788 | .instance_size = sizeof(BDRVBlkioState), \ | |
789 | .bdrv_file_open = blkio_file_open, \ | |
790 | .bdrv_close = blkio_close, \ | |
791 | .bdrv_getlength = blkio_getlength, \ | |
792 | .bdrv_get_info = blkio_get_info, \ | |
793 | .bdrv_attach_aio_context = blkio_attach_aio_context, \ | |
794 | .bdrv_detach_aio_context = blkio_detach_aio_context, \ | |
795 | .bdrv_co_pdiscard = blkio_co_pdiscard, \ | |
796 | .bdrv_co_preadv = blkio_co_preadv, \ | |
797 | .bdrv_co_pwritev = blkio_co_pwritev, \ | |
798 | .bdrv_co_flush_to_disk = blkio_co_flush, \ | |
799 | .bdrv_co_pwrite_zeroes = blkio_co_pwrite_zeroes, \ | |
800 | .bdrv_io_unplug = blkio_io_unplug, \ | |
801 | .bdrv_refresh_limits = blkio_refresh_limits, \ | |
802 | __VA_ARGS__ \ | |
803 | } | |
804 | ||
805 | static BlockDriver bdrv_io_uring = BLKIO_DRIVER( | |
806 | DRIVER_IO_URING, | |
807 | .bdrv_needs_filename = true, | |
808 | ); | |
809 | ||
810 | static BlockDriver bdrv_nvme_io_uring = BLKIO_DRIVER( | |
811 | DRIVER_NVME_IO_URING, | |
812 | .bdrv_needs_filename = true, | |
813 | ); | |
814 | ||
815 | static BlockDriver bdrv_virtio_blk_vhost_user = BLKIO_DRIVER( | |
816 | DRIVER_VIRTIO_BLK_VHOST_USER | |
817 | ); | |
818 | ||
819 | static BlockDriver bdrv_virtio_blk_vhost_vdpa = BLKIO_DRIVER( | |
820 | DRIVER_VIRTIO_BLK_VHOST_VDPA | |
821 | ); | |
822 | ||
823 | static void bdrv_blkio_init(void) | |
824 | { | |
825 | bdrv_register(&bdrv_io_uring); | |
826 | bdrv_register(&bdrv_nvme_io_uring); | |
827 | bdrv_register(&bdrv_virtio_blk_vhost_user); | |
828 | bdrv_register(&bdrv_virtio_blk_vhost_vdpa); | |
829 | } | |
830 | ||
831 | block_init(bdrv_blkio_init); |