Merge tag 'edgar/xilinx-next-2022-09-21.for-upstream' of https://github.com/edgarigl...
[qemu.git] / hw / block / virtio-blk.c
1 /*
2 * Virtio Block Device
3 *
4 * Copyright IBM, Corp. 2007
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 */
13
14 #include "qemu/osdep.h"
15 #include "qapi/error.h"
16 #include "qemu/iov.h"
17 #include "qemu/module.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "trace.h"
21 #include "hw/block/block.h"
22 #include "hw/qdev-properties.h"
23 #include "sysemu/blockdev.h"
24 #include "sysemu/sysemu.h"
25 #include "sysemu/runstate.h"
26 #include "hw/virtio/virtio-blk.h"
27 #include "dataplane/virtio-blk.h"
28 #include "scsi/constants.h"
29 #ifdef __linux__
30 # include <scsi/sg.h>
31 #endif
32 #include "hw/virtio/virtio-bus.h"
33 #include "migration/qemu-file-types.h"
34 #include "hw/virtio/virtio-access.h"
35
36 /* Config size before the discard support (hide associated config fields) */
37 #define VIRTIO_BLK_CFG_SIZE offsetof(struct virtio_blk_config, \
38 max_discard_sectors)
39 /*
40 * Starting from the discard feature, we can use this array to properly
41 * set the config size depending on the features enabled.
42 */
43 static const VirtIOFeature feature_sizes[] = {
44 {.flags = 1ULL << VIRTIO_BLK_F_DISCARD,
45 .end = endof(struct virtio_blk_config, discard_sector_alignment)},
46 {.flags = 1ULL << VIRTIO_BLK_F_WRITE_ZEROES,
47 .end = endof(struct virtio_blk_config, write_zeroes_may_unmap)},
48 {}
49 };
50
51 static void virtio_blk_set_config_size(VirtIOBlock *s, uint64_t host_features)
52 {
53 s->config_size = MAX(VIRTIO_BLK_CFG_SIZE,
54 virtio_feature_get_config_size(feature_sizes, host_features));
55
56 assert(s->config_size <= sizeof(struct virtio_blk_config));
57 }
58
59 static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq,
60 VirtIOBlockReq *req)
61 {
62 req->dev = s;
63 req->vq = vq;
64 req->qiov.size = 0;
65 req->in_len = 0;
66 req->next = NULL;
67 req->mr_next = NULL;
68 }
69
70 static void virtio_blk_free_request(VirtIOBlockReq *req)
71 {
72 g_free(req);
73 }
74
75 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
76 {
77 VirtIOBlock *s = req->dev;
78 VirtIODevice *vdev = VIRTIO_DEVICE(s);
79
80 trace_virtio_blk_req_complete(vdev, req, status);
81
82 stb_p(&req->in->status, status);
83 iov_discard_undo(&req->inhdr_undo);
84 iov_discard_undo(&req->outhdr_undo);
85 virtqueue_push(req->vq, &req->elem, req->in_len);
86 if (s->dataplane_started && !s->dataplane_disabled) {
87 virtio_blk_data_plane_notify(s->dataplane, req->vq);
88 } else {
89 virtio_notify(vdev, req->vq);
90 }
91 }
92
93 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
94 bool is_read, bool acct_failed)
95 {
96 VirtIOBlock *s = req->dev;
97 BlockErrorAction action = blk_get_error_action(s->blk, is_read, error);
98
99 if (action == BLOCK_ERROR_ACTION_STOP) {
100 /* Break the link as the next request is going to be parsed from the
101 * ring again. Otherwise we may end up doing a double completion! */
102 req->mr_next = NULL;
103 req->next = s->rq;
104 s->rq = req;
105 } else if (action == BLOCK_ERROR_ACTION_REPORT) {
106 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
107 if (acct_failed) {
108 block_acct_failed(blk_get_stats(s->blk), &req->acct);
109 }
110 virtio_blk_free_request(req);
111 }
112
113 blk_error_action(s->blk, action, is_read, error);
114 return action != BLOCK_ERROR_ACTION_IGNORE;
115 }
116
117 static void virtio_blk_rw_complete(void *opaque, int ret)
118 {
119 VirtIOBlockReq *next = opaque;
120 VirtIOBlock *s = next->dev;
121 VirtIODevice *vdev = VIRTIO_DEVICE(s);
122
123 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
124 while (next) {
125 VirtIOBlockReq *req = next;
126 next = req->mr_next;
127 trace_virtio_blk_rw_complete(vdev, req, ret);
128
129 if (req->qiov.nalloc != -1) {
130 /* If nalloc is != -1 req->qiov is a local copy of the original
131 * external iovec. It was allocated in submit_requests to be
132 * able to merge requests. */
133 qemu_iovec_destroy(&req->qiov);
134 }
135
136 if (ret) {
137 int p = virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type);
138 bool is_read = !(p & VIRTIO_BLK_T_OUT);
139 /* Note that memory may be dirtied on read failure. If the
140 * virtio request is not completed here, as is the case for
141 * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
142 * correctly during live migration. While this is ugly,
143 * it is acceptable because the device is free to write to
144 * the memory until the request is completed (which will
145 * happen on the other side of the migration).
146 */
147 if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) {
148 continue;
149 }
150 }
151
152 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
153 block_acct_done(blk_get_stats(s->blk), &req->acct);
154 virtio_blk_free_request(req);
155 }
156 aio_context_release(blk_get_aio_context(s->conf.conf.blk));
157 }
158
159 static void virtio_blk_flush_complete(void *opaque, int ret)
160 {
161 VirtIOBlockReq *req = opaque;
162 VirtIOBlock *s = req->dev;
163
164 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
165 if (ret) {
166 if (virtio_blk_handle_rw_error(req, -ret, 0, true)) {
167 goto out;
168 }
169 }
170
171 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
172 block_acct_done(blk_get_stats(s->blk), &req->acct);
173 virtio_blk_free_request(req);
174
175 out:
176 aio_context_release(blk_get_aio_context(s->conf.conf.blk));
177 }
178
179 static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret)
180 {
181 VirtIOBlockReq *req = opaque;
182 VirtIOBlock *s = req->dev;
183 bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) &
184 ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES;
185
186 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
187 if (ret) {
188 if (virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) {
189 goto out;
190 }
191 }
192
193 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
194 if (is_write_zeroes) {
195 block_acct_done(blk_get_stats(s->blk), &req->acct);
196 }
197 virtio_blk_free_request(req);
198
199 out:
200 aio_context_release(blk_get_aio_context(s->conf.conf.blk));
201 }
202
203 #ifdef __linux__
204
205 typedef struct {
206 VirtIOBlockReq *req;
207 struct sg_io_hdr hdr;
208 } VirtIOBlockIoctlReq;
209
210 static void virtio_blk_ioctl_complete(void *opaque, int status)
211 {
212 VirtIOBlockIoctlReq *ioctl_req = opaque;
213 VirtIOBlockReq *req = ioctl_req->req;
214 VirtIOBlock *s = req->dev;
215 VirtIODevice *vdev = VIRTIO_DEVICE(s);
216 struct virtio_scsi_inhdr *scsi;
217 struct sg_io_hdr *hdr;
218
219 scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
220
221 if (status) {
222 status = VIRTIO_BLK_S_UNSUPP;
223 virtio_stl_p(vdev, &scsi->errors, 255);
224 goto out;
225 }
226
227 hdr = &ioctl_req->hdr;
228 /*
229 * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
230 * clear the masked_status field [hence status gets cleared too, see
231 * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
232 * status has occurred. However they do set DRIVER_SENSE in driver_status
233 * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
234 */
235 if (hdr->status == 0 && hdr->sb_len_wr > 0) {
236 hdr->status = CHECK_CONDITION;
237 }
238
239 virtio_stl_p(vdev, &scsi->errors,
240 hdr->status | (hdr->msg_status << 8) |
241 (hdr->host_status << 16) | (hdr->driver_status << 24));
242 virtio_stl_p(vdev, &scsi->residual, hdr->resid);
243 virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr);
244 virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
245
246 out:
247 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
248 virtio_blk_req_complete(req, status);
249 virtio_blk_free_request(req);
250 aio_context_release(blk_get_aio_context(s->conf.conf.blk));
251 g_free(ioctl_req);
252 }
253
254 #endif
255
256 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq)
257 {
258 VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq));
259
260 if (req) {
261 virtio_blk_init_request(s, vq, req);
262 }
263 return req;
264 }
265
266 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req)
267 {
268 int status = VIRTIO_BLK_S_OK;
269 struct virtio_scsi_inhdr *scsi = NULL;
270 VirtIOBlock *blk = req->dev;
271 VirtIODevice *vdev = VIRTIO_DEVICE(blk);
272 VirtQueueElement *elem = &req->elem;
273
274 #ifdef __linux__
275 int i;
276 VirtIOBlockIoctlReq *ioctl_req;
277 BlockAIOCB *acb;
278 #endif
279
280 /*
281 * We require at least one output segment each for the virtio_blk_outhdr
282 * and the SCSI command block.
283 *
284 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
285 * and the sense buffer pointer in the input segments.
286 */
287 if (elem->out_num < 2 || elem->in_num < 3) {
288 status = VIRTIO_BLK_S_IOERR;
289 goto fail;
290 }
291
292 /*
293 * The scsi inhdr is placed in the second-to-last input segment, just
294 * before the regular inhdr.
295 */
296 scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
297
298 if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) {
299 status = VIRTIO_BLK_S_UNSUPP;
300 goto fail;
301 }
302
303 /*
304 * No support for bidirection commands yet.
305 */
306 if (elem->out_num > 2 && elem->in_num > 3) {
307 status = VIRTIO_BLK_S_UNSUPP;
308 goto fail;
309 }
310
311 #ifdef __linux__
312 ioctl_req = g_new0(VirtIOBlockIoctlReq, 1);
313 ioctl_req->req = req;
314 ioctl_req->hdr.interface_id = 'S';
315 ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len;
316 ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base;
317 ioctl_req->hdr.dxfer_len = 0;
318
319 if (elem->out_num > 2) {
320 /*
321 * If there are more than the minimally required 2 output segments
322 * there is write payload starting from the third iovec.
323 */
324 ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV;
325 ioctl_req->hdr.iovec_count = elem->out_num - 2;
326
327 for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
328 ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
329 }
330
331 ioctl_req->hdr.dxferp = elem->out_sg + 2;
332
333 } else if (elem->in_num > 3) {
334 /*
335 * If we have more than 3 input segments the guest wants to actually
336 * read data.
337 */
338 ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV;
339 ioctl_req->hdr.iovec_count = elem->in_num - 3;
340 for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
341 ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len;
342 }
343
344 ioctl_req->hdr.dxferp = elem->in_sg;
345 } else {
346 /*
347 * Some SCSI commands don't actually transfer any data.
348 */
349 ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE;
350 }
351
352 ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
353 ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
354
355 acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr,
356 virtio_blk_ioctl_complete, ioctl_req);
357 if (!acb) {
358 g_free(ioctl_req);
359 status = VIRTIO_BLK_S_UNSUPP;
360 goto fail;
361 }
362 return -EINPROGRESS;
363 #else
364 abort();
365 #endif
366
367 fail:
368 /* Just put anything nonzero so that the ioctl fails in the guest. */
369 if (scsi) {
370 virtio_stl_p(vdev, &scsi->errors, 255);
371 }
372 return status;
373 }
374
375 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
376 {
377 int status;
378
379 status = virtio_blk_handle_scsi_req(req);
380 if (status != -EINPROGRESS) {
381 virtio_blk_req_complete(req, status);
382 virtio_blk_free_request(req);
383 }
384 }
385
386 static inline void submit_requests(BlockBackend *blk, MultiReqBuffer *mrb,
387 int start, int num_reqs, int niov)
388 {
389 QEMUIOVector *qiov = &mrb->reqs[start]->qiov;
390 int64_t sector_num = mrb->reqs[start]->sector_num;
391 bool is_write = mrb->is_write;
392
393 if (num_reqs > 1) {
394 int i;
395 struct iovec *tmp_iov = qiov->iov;
396 int tmp_niov = qiov->niov;
397
398 /* mrb->reqs[start]->qiov was initialized from external so we can't
399 * modify it here. We need to initialize it locally and then add the
400 * external iovecs. */
401 qemu_iovec_init(qiov, niov);
402
403 for (i = 0; i < tmp_niov; i++) {
404 qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len);
405 }
406
407 for (i = start + 1; i < start + num_reqs; i++) {
408 qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0,
409 mrb->reqs[i]->qiov.size);
410 mrb->reqs[i - 1]->mr_next = mrb->reqs[i];
411 }
412
413 trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev),
414 mrb, start, num_reqs,
415 sector_num << BDRV_SECTOR_BITS,
416 qiov->size, is_write);
417 block_acct_merge_done(blk_get_stats(blk),
418 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ,
419 num_reqs - 1);
420 }
421
422 if (is_write) {
423 blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
424 virtio_blk_rw_complete, mrb->reqs[start]);
425 } else {
426 blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
427 virtio_blk_rw_complete, mrb->reqs[start]);
428 }
429 }
430
431 static int multireq_compare(const void *a, const void *b)
432 {
433 const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a,
434 *req2 = *(VirtIOBlockReq **)b;
435
436 /*
437 * Note that we can't simply subtract sector_num1 from sector_num2
438 * here as that could overflow the return value.
439 */
440 if (req1->sector_num > req2->sector_num) {
441 return 1;
442 } else if (req1->sector_num < req2->sector_num) {
443 return -1;
444 } else {
445 return 0;
446 }
447 }
448
449 static void virtio_blk_submit_multireq(BlockBackend *blk, MultiReqBuffer *mrb)
450 {
451 int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0;
452 uint32_t max_transfer;
453 int64_t sector_num = 0;
454
455 if (mrb->num_reqs == 1) {
456 submit_requests(blk, mrb, 0, 1, -1);
457 mrb->num_reqs = 0;
458 return;
459 }
460
461 max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk);
462
463 qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs),
464 &multireq_compare);
465
466 for (i = 0; i < mrb->num_reqs; i++) {
467 VirtIOBlockReq *req = mrb->reqs[i];
468 if (num_reqs > 0) {
469 /*
470 * NOTE: We cannot merge the requests in below situations:
471 * 1. requests are not sequential
472 * 2. merge would exceed maximum number of IOVs
473 * 3. merge would exceed maximum transfer length of backend device
474 */
475 if (sector_num + nb_sectors != req->sector_num ||
476 niov > blk_get_max_iov(blk) - req->qiov.niov ||
477 req->qiov.size > max_transfer ||
478 nb_sectors > (max_transfer -
479 req->qiov.size) / BDRV_SECTOR_SIZE) {
480 submit_requests(blk, mrb, start, num_reqs, niov);
481 num_reqs = 0;
482 }
483 }
484
485 if (num_reqs == 0) {
486 sector_num = req->sector_num;
487 nb_sectors = niov = 0;
488 start = i;
489 }
490
491 nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE;
492 niov += req->qiov.niov;
493 num_reqs++;
494 }
495
496 submit_requests(blk, mrb, start, num_reqs, niov);
497 mrb->num_reqs = 0;
498 }
499
500 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
501 {
502 VirtIOBlock *s = req->dev;
503
504 block_acct_start(blk_get_stats(s->blk), &req->acct, 0,
505 BLOCK_ACCT_FLUSH);
506
507 /*
508 * Make sure all outstanding writes are posted to the backing device.
509 */
510 if (mrb->is_write && mrb->num_reqs > 0) {
511 virtio_blk_submit_multireq(s->blk, mrb);
512 }
513 blk_aio_flush(s->blk, virtio_blk_flush_complete, req);
514 }
515
516 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
517 uint64_t sector, size_t size)
518 {
519 uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
520 uint64_t total_sectors;
521
522 if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
523 return false;
524 }
525 if (sector & dev->sector_mask) {
526 return false;
527 }
528 if (size % dev->conf.conf.logical_block_size) {
529 return false;
530 }
531 blk_get_geometry(dev->blk, &total_sectors);
532 if (sector > total_sectors || nb_sectors > total_sectors - sector) {
533 return false;
534 }
535 return true;
536 }
537
538 static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req,
539 struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes)
540 {
541 VirtIOBlock *s = req->dev;
542 VirtIODevice *vdev = VIRTIO_DEVICE(s);
543 uint64_t sector;
544 uint32_t num_sectors, flags, max_sectors;
545 uint8_t err_status;
546 int bytes;
547
548 sector = virtio_ldq_p(vdev, &dwz_hdr->sector);
549 num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors);
550 flags = virtio_ldl_p(vdev, &dwz_hdr->flags);
551 max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors :
552 s->conf.max_discard_sectors;
553
554 /*
555 * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check
556 * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in
557 * the integer variable.
558 */
559 if (unlikely(num_sectors > max_sectors)) {
560 err_status = VIRTIO_BLK_S_IOERR;
561 goto err;
562 }
563
564 bytes = num_sectors << BDRV_SECTOR_BITS;
565
566 if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) {
567 err_status = VIRTIO_BLK_S_IOERR;
568 goto err;
569 }
570
571 /*
572 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard
573 * and write zeroes commands if any unknown flag is set.
574 */
575 if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
576 err_status = VIRTIO_BLK_S_UNSUPP;
577 goto err;
578 }
579
580 if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */
581 int blk_aio_flags = 0;
582
583 if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) {
584 blk_aio_flags |= BDRV_REQ_MAY_UNMAP;
585 }
586
587 block_acct_start(blk_get_stats(s->blk), &req->acct, bytes,
588 BLOCK_ACCT_WRITE);
589
590 blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS,
591 bytes, blk_aio_flags,
592 virtio_blk_discard_write_zeroes_complete, req);
593 } else { /* VIRTIO_BLK_T_DISCARD */
594 /*
595 * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for
596 * discard commands if the unmap flag is set.
597 */
598 if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
599 err_status = VIRTIO_BLK_S_UNSUPP;
600 goto err;
601 }
602
603 blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes,
604 virtio_blk_discard_write_zeroes_complete, req);
605 }
606
607 return VIRTIO_BLK_S_OK;
608
609 err:
610 if (is_write_zeroes) {
611 block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE);
612 }
613 return err_status;
614 }
615
616 static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
617 {
618 uint32_t type;
619 struct iovec *in_iov = req->elem.in_sg;
620 struct iovec *out_iov = req->elem.out_sg;
621 unsigned in_num = req->elem.in_num;
622 unsigned out_num = req->elem.out_num;
623 VirtIOBlock *s = req->dev;
624 VirtIODevice *vdev = VIRTIO_DEVICE(s);
625
626 if (req->elem.out_num < 1 || req->elem.in_num < 1) {
627 virtio_error(vdev, "virtio-blk missing headers");
628 return -1;
629 }
630
631 if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out,
632 sizeof(req->out)) != sizeof(req->out))) {
633 virtio_error(vdev, "virtio-blk request outhdr too short");
634 return -1;
635 }
636
637 iov_discard_front_undoable(&out_iov, &out_num, sizeof(req->out),
638 &req->outhdr_undo);
639
640 if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
641 virtio_error(vdev, "virtio-blk request inhdr too short");
642 iov_discard_undo(&req->outhdr_undo);
643 return -1;
644 }
645
646 /* We always touch the last byte, so just see how big in_iov is. */
647 req->in_len = iov_size(in_iov, in_num);
648 req->in = (void *)in_iov[in_num - 1].iov_base
649 + in_iov[in_num - 1].iov_len
650 - sizeof(struct virtio_blk_inhdr);
651 iov_discard_back_undoable(in_iov, &in_num, sizeof(struct virtio_blk_inhdr),
652 &req->inhdr_undo);
653
654 type = virtio_ldl_p(vdev, &req->out.type);
655
656 /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
657 * is an optional flag. Although a guest should not send this flag if
658 * not negotiated we ignored it in the past. So keep ignoring it. */
659 switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) {
660 case VIRTIO_BLK_T_IN:
661 {
662 bool is_write = type & VIRTIO_BLK_T_OUT;
663 req->sector_num = virtio_ldq_p(vdev, &req->out.sector);
664
665 if (is_write) {
666 qemu_iovec_init_external(&req->qiov, out_iov, out_num);
667 trace_virtio_blk_handle_write(vdev, req, req->sector_num,
668 req->qiov.size / BDRV_SECTOR_SIZE);
669 } else {
670 qemu_iovec_init_external(&req->qiov, in_iov, in_num);
671 trace_virtio_blk_handle_read(vdev, req, req->sector_num,
672 req->qiov.size / BDRV_SECTOR_SIZE);
673 }
674
675 if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) {
676 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
677 block_acct_invalid(blk_get_stats(s->blk),
678 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
679 virtio_blk_free_request(req);
680 return 0;
681 }
682
683 block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size,
684 is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
685
686 /* merge would exceed maximum number of requests or IO direction
687 * changes */
688 if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS ||
689 is_write != mrb->is_write ||
690 !s->conf.request_merging)) {
691 virtio_blk_submit_multireq(s->blk, mrb);
692 }
693
694 assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS);
695 mrb->reqs[mrb->num_reqs++] = req;
696 mrb->is_write = is_write;
697 break;
698 }
699 case VIRTIO_BLK_T_FLUSH:
700 virtio_blk_handle_flush(req, mrb);
701 break;
702 case VIRTIO_BLK_T_SCSI_CMD:
703 virtio_blk_handle_scsi(req);
704 break;
705 case VIRTIO_BLK_T_GET_ID:
706 {
707 /*
708 * NB: per existing s/n string convention the string is
709 * terminated by '\0' only when shorter than buffer.
710 */
711 const char *serial = s->conf.serial ? s->conf.serial : "";
712 size_t size = MIN(strlen(serial) + 1,
713 MIN(iov_size(in_iov, in_num),
714 VIRTIO_BLK_ID_BYTES));
715 iov_from_buf(in_iov, in_num, 0, serial, size);
716 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
717 virtio_blk_free_request(req);
718 break;
719 }
720 /*
721 * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with
722 * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement,
723 * so we must mask it for these requests, then we will check if it is set.
724 */
725 case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT:
726 case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT:
727 {
728 struct virtio_blk_discard_write_zeroes dwz_hdr;
729 size_t out_len = iov_size(out_iov, out_num);
730 bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) ==
731 VIRTIO_BLK_T_WRITE_ZEROES;
732 uint8_t err_status;
733
734 /*
735 * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains
736 * more than one segment.
737 */
738 if (unlikely(!(type & VIRTIO_BLK_T_OUT) ||
739 out_len > sizeof(dwz_hdr))) {
740 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
741 virtio_blk_free_request(req);
742 return 0;
743 }
744
745 if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr,
746 sizeof(dwz_hdr)) != sizeof(dwz_hdr))) {
747 iov_discard_undo(&req->inhdr_undo);
748 iov_discard_undo(&req->outhdr_undo);
749 virtio_error(vdev, "virtio-blk discard/write_zeroes header"
750 " too short");
751 return -1;
752 }
753
754 err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr,
755 is_write_zeroes);
756 if (err_status != VIRTIO_BLK_S_OK) {
757 virtio_blk_req_complete(req, err_status);
758 virtio_blk_free_request(req);
759 }
760
761 break;
762 }
763 default:
764 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
765 virtio_blk_free_request(req);
766 }
767 return 0;
768 }
769
770 void virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
771 {
772 VirtIOBlockReq *req;
773 MultiReqBuffer mrb = {};
774 bool suppress_notifications = virtio_queue_get_notification(vq);
775
776 aio_context_acquire(blk_get_aio_context(s->blk));
777 blk_io_plug(s->blk);
778
779 do {
780 if (suppress_notifications) {
781 virtio_queue_set_notification(vq, 0);
782 }
783
784 while ((req = virtio_blk_get_request(s, vq))) {
785 if (virtio_blk_handle_request(req, &mrb)) {
786 virtqueue_detach_element(req->vq, &req->elem, 0);
787 virtio_blk_free_request(req);
788 break;
789 }
790 }
791
792 if (suppress_notifications) {
793 virtio_queue_set_notification(vq, 1);
794 }
795 } while (!virtio_queue_empty(vq));
796
797 if (mrb.num_reqs) {
798 virtio_blk_submit_multireq(s->blk, &mrb);
799 }
800
801 blk_io_unplug(s->blk);
802 aio_context_release(blk_get_aio_context(s->blk));
803 }
804
805 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
806 {
807 VirtIOBlock *s = (VirtIOBlock *)vdev;
808
809 if (s->dataplane && !s->dataplane_started) {
810 /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
811 * dataplane here instead of waiting for .set_status().
812 */
813 virtio_device_start_ioeventfd(vdev);
814 if (!s->dataplane_disabled) {
815 return;
816 }
817 }
818 virtio_blk_handle_vq(s, vq);
819 }
820
821 void virtio_blk_process_queued_requests(VirtIOBlock *s, bool is_bh)
822 {
823 VirtIOBlockReq *req = s->rq;
824 MultiReqBuffer mrb = {};
825
826 s->rq = NULL;
827
828 aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
829 while (req) {
830 VirtIOBlockReq *next = req->next;
831 if (virtio_blk_handle_request(req, &mrb)) {
832 /* Device is now broken and won't do any processing until it gets
833 * reset. Already queued requests will be lost: let's purge them.
834 */
835 while (req) {
836 next = req->next;
837 virtqueue_detach_element(req->vq, &req->elem, 0);
838 virtio_blk_free_request(req);
839 req = next;
840 }
841 break;
842 }
843 req = next;
844 }
845
846 if (mrb.num_reqs) {
847 virtio_blk_submit_multireq(s->blk, &mrb);
848 }
849 if (is_bh) {
850 blk_dec_in_flight(s->conf.conf.blk);
851 }
852 aio_context_release(blk_get_aio_context(s->conf.conf.blk));
853 }
854
855 static void virtio_blk_dma_restart_bh(void *opaque)
856 {
857 VirtIOBlock *s = opaque;
858
859 qemu_bh_delete(s->bh);
860 s->bh = NULL;
861
862 virtio_blk_process_queued_requests(s, true);
863 }
864
865 static void virtio_blk_dma_restart_cb(void *opaque, bool running,
866 RunState state)
867 {
868 VirtIOBlock *s = opaque;
869 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(s)));
870 VirtioBusState *bus = VIRTIO_BUS(qbus);
871
872 if (!running) {
873 return;
874 }
875
876 /*
877 * If ioeventfd is enabled, don't schedule the BH here as queued
878 * requests will be processed while starting the data plane.
879 */
880 if (!s->bh && !virtio_bus_ioeventfd_enabled(bus)) {
881 s->bh = aio_bh_new(blk_get_aio_context(s->conf.conf.blk),
882 virtio_blk_dma_restart_bh, s);
883 blk_inc_in_flight(s->conf.conf.blk);
884 qemu_bh_schedule(s->bh);
885 }
886 }
887
888 static void virtio_blk_reset(VirtIODevice *vdev)
889 {
890 VirtIOBlock *s = VIRTIO_BLK(vdev);
891 AioContext *ctx;
892 VirtIOBlockReq *req;
893
894 ctx = blk_get_aio_context(s->blk);
895 aio_context_acquire(ctx);
896 blk_drain(s->blk);
897
898 /* We drop queued requests after blk_drain() because blk_drain() itself can
899 * produce them. */
900 while (s->rq) {
901 req = s->rq;
902 s->rq = req->next;
903 virtqueue_detach_element(req->vq, &req->elem, 0);
904 virtio_blk_free_request(req);
905 }
906
907 aio_context_release(ctx);
908
909 assert(!s->dataplane_started);
910 blk_set_enable_write_cache(s->blk, s->original_wce);
911 }
912
913 /* coalesce internal state, copy to pci i/o region 0
914 */
915 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
916 {
917 VirtIOBlock *s = VIRTIO_BLK(vdev);
918 BlockConf *conf = &s->conf.conf;
919 struct virtio_blk_config blkcfg;
920 uint64_t capacity;
921 int64_t length;
922 int blk_size = conf->logical_block_size;
923
924 blk_get_geometry(s->blk, &capacity);
925 memset(&blkcfg, 0, sizeof(blkcfg));
926 virtio_stq_p(vdev, &blkcfg.capacity, capacity);
927 virtio_stl_p(vdev, &blkcfg.seg_max,
928 s->conf.seg_max_adjust ? s->conf.queue_size - 2 : 128 - 2);
929 virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls);
930 virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
931 virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size);
932 virtio_stl_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size);
933 blkcfg.geometry.heads = conf->heads;
934 /*
935 * We must ensure that the block device capacity is a multiple of
936 * the logical block size. If that is not the case, let's use
937 * sector_mask to adopt the geometry to have a correct picture.
938 * For those devices where the capacity is ok for the given geometry
939 * we don't touch the sector value of the geometry, since some devices
940 * (like s390 dasd) need a specific value. Here the capacity is already
941 * cyls*heads*secs*blk_size and the sector value is not block size
942 * divided by 512 - instead it is the amount of blk_size blocks
943 * per track (cylinder).
944 */
945 length = blk_getlength(s->blk);
946 if (length > 0 && length / conf->heads / conf->secs % blk_size) {
947 blkcfg.geometry.sectors = conf->secs & ~s->sector_mask;
948 } else {
949 blkcfg.geometry.sectors = conf->secs;
950 }
951 blkcfg.size_max = 0;
952 blkcfg.physical_block_exp = get_physical_block_exp(conf);
953 blkcfg.alignment_offset = 0;
954 blkcfg.wce = blk_enable_write_cache(s->blk);
955 virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues);
956 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) {
957 uint32_t discard_granularity = conf->discard_granularity;
958 if (discard_granularity == -1 || !s->conf.report_discard_granularity) {
959 discard_granularity = blk_size;
960 }
961 virtio_stl_p(vdev, &blkcfg.max_discard_sectors,
962 s->conf.max_discard_sectors);
963 virtio_stl_p(vdev, &blkcfg.discard_sector_alignment,
964 discard_granularity >> BDRV_SECTOR_BITS);
965 /*
966 * We support only one segment per request since multiple segments
967 * are not widely used and there are no userspace APIs that allow
968 * applications to submit multiple segments in a single call.
969 */
970 virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1);
971 }
972 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) {
973 virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors,
974 s->conf.max_write_zeroes_sectors);
975 blkcfg.write_zeroes_may_unmap = 1;
976 virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1);
977 }
978 memcpy(config, &blkcfg, s->config_size);
979 }
980
981 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
982 {
983 VirtIOBlock *s = VIRTIO_BLK(vdev);
984 struct virtio_blk_config blkcfg;
985
986 memcpy(&blkcfg, config, s->config_size);
987
988 aio_context_acquire(blk_get_aio_context(s->blk));
989 blk_set_enable_write_cache(s->blk, blkcfg.wce != 0);
990 aio_context_release(blk_get_aio_context(s->blk));
991 }
992
993 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features,
994 Error **errp)
995 {
996 VirtIOBlock *s = VIRTIO_BLK(vdev);
997
998 /* Firstly sync all virtio-blk possible supported features */
999 features |= s->host_features;
1000
1001 virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
1002 virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
1003 virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
1004 virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
1005 if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) {
1006 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) {
1007 error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
1008 return 0;
1009 }
1010 } else {
1011 virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT);
1012 virtio_add_feature(&features, VIRTIO_BLK_F_SCSI);
1013 }
1014
1015 if (blk_enable_write_cache(s->blk) ||
1016 (s->conf.x_enable_wce_if_config_wce &&
1017 virtio_has_feature(features, VIRTIO_BLK_F_CONFIG_WCE))) {
1018 virtio_add_feature(&features, VIRTIO_BLK_F_WCE);
1019 }
1020 if (!blk_is_writable(s->blk)) {
1021 virtio_add_feature(&features, VIRTIO_BLK_F_RO);
1022 }
1023 if (s->conf.num_queues > 1) {
1024 virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
1025 }
1026
1027 return features;
1028 }
1029
1030 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
1031 {
1032 VirtIOBlock *s = VIRTIO_BLK(vdev);
1033
1034 if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) {
1035 assert(!s->dataplane_started);
1036 }
1037
1038 if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1039 return;
1040 }
1041
1042 /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
1043 * cache flushes. Thus, the "auto writethrough" behavior is never
1044 * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
1045 * Leaving it enabled would break the following sequence:
1046 *
1047 * Guest started with "-drive cache=writethrough"
1048 * Guest sets status to 0
1049 * Guest sets DRIVER bit in status field
1050 * Guest reads host features (WCE=0, CONFIG_WCE=1)
1051 * Guest writes guest features (WCE=0, CONFIG_WCE=1)
1052 * Guest writes 1 to the WCE configuration field (writeback mode)
1053 * Guest sets DRIVER_OK bit in status field
1054 *
1055 * s->blk would erroneously be placed in writethrough mode.
1056 */
1057 if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) {
1058 aio_context_acquire(blk_get_aio_context(s->blk));
1059 blk_set_enable_write_cache(s->blk,
1060 virtio_vdev_has_feature(vdev,
1061 VIRTIO_BLK_F_WCE));
1062 aio_context_release(blk_get_aio_context(s->blk));
1063 }
1064 }
1065
1066 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
1067 {
1068 VirtIOBlock *s = VIRTIO_BLK(vdev);
1069 VirtIOBlockReq *req = s->rq;
1070
1071 while (req) {
1072 qemu_put_sbyte(f, 1);
1073
1074 if (s->conf.num_queues > 1) {
1075 qemu_put_be32(f, virtio_get_queue_index(req->vq));
1076 }
1077
1078 qemu_put_virtqueue_element(vdev, f, &req->elem);
1079 req = req->next;
1080 }
1081 qemu_put_sbyte(f, 0);
1082 }
1083
1084 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
1085 int version_id)
1086 {
1087 VirtIOBlock *s = VIRTIO_BLK(vdev);
1088
1089 while (qemu_get_sbyte(f)) {
1090 unsigned nvqs = s->conf.num_queues;
1091 unsigned vq_idx = 0;
1092 VirtIOBlockReq *req;
1093
1094 if (nvqs > 1) {
1095 vq_idx = qemu_get_be32(f);
1096
1097 if (vq_idx >= nvqs) {
1098 error_report("Invalid virtqueue index in request list: %#x",
1099 vq_idx);
1100 return -EINVAL;
1101 }
1102 }
1103
1104 req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq));
1105 virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req);
1106 req->next = s->rq;
1107 s->rq = req;
1108 }
1109
1110 return 0;
1111 }
1112
1113 static void virtio_resize_cb(void *opaque)
1114 {
1115 VirtIODevice *vdev = opaque;
1116
1117 assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1118 virtio_notify_config(vdev);
1119 }
1120
1121 static void virtio_blk_resize(void *opaque)
1122 {
1123 VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
1124
1125 /*
1126 * virtio_notify_config() needs to acquire the global mutex,
1127 * so it can't be called from an iothread. Instead, schedule
1128 * it to be run in the main context BH.
1129 */
1130 aio_bh_schedule_oneshot(qemu_get_aio_context(), virtio_resize_cb, vdev);
1131 }
1132
1133 static const BlockDevOps virtio_block_ops = {
1134 .resize_cb = virtio_blk_resize,
1135 };
1136
1137 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
1138 {
1139 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1140 VirtIOBlock *s = VIRTIO_BLK(dev);
1141 VirtIOBlkConf *conf = &s->conf;
1142 Error *err = NULL;
1143 unsigned i;
1144
1145 if (!conf->conf.blk) {
1146 error_setg(errp, "drive property not set");
1147 return;
1148 }
1149 if (!blk_is_inserted(conf->conf.blk)) {
1150 error_setg(errp, "Device needs media, but drive is empty");
1151 return;
1152 }
1153 if (conf->num_queues == VIRTIO_BLK_AUTO_NUM_QUEUES) {
1154 conf->num_queues = 1;
1155 }
1156 if (!conf->num_queues) {
1157 error_setg(errp, "num-queues property must be larger than 0");
1158 return;
1159 }
1160 if (conf->queue_size <= 2) {
1161 error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1162 "must be > 2", conf->queue_size);
1163 return;
1164 }
1165 if (!is_power_of_2(conf->queue_size) ||
1166 conf->queue_size > VIRTQUEUE_MAX_SIZE) {
1167 error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1168 "must be a power of 2 (max %d)",
1169 conf->queue_size, VIRTQUEUE_MAX_SIZE);
1170 return;
1171 }
1172
1173 if (!blkconf_apply_backend_options(&conf->conf,
1174 !blk_supports_write_perm(conf->conf.blk),
1175 true, errp)) {
1176 return;
1177 }
1178 s->original_wce = blk_enable_write_cache(conf->conf.blk);
1179 if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) {
1180 return;
1181 }
1182
1183 if (!blkconf_blocksizes(&conf->conf, errp)) {
1184 return;
1185 }
1186
1187 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) &&
1188 (!conf->max_discard_sectors ||
1189 conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1190 error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")"
1191 ", must be between 1 and %d",
1192 conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS);
1193 return;
1194 }
1195
1196 if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) &&
1197 (!conf->max_write_zeroes_sectors ||
1198 conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1199 error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32
1200 "), must be between 1 and %d",
1201 conf->max_write_zeroes_sectors,
1202 (int)BDRV_REQUEST_MAX_SECTORS);
1203 return;
1204 }
1205
1206 virtio_blk_set_config_size(s, s->host_features);
1207
1208 virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK, s->config_size);
1209
1210 s->blk = conf->conf.blk;
1211 s->rq = NULL;
1212 s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1;
1213
1214 for (i = 0; i < conf->num_queues; i++) {
1215 virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output);
1216 }
1217 virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err);
1218 if (err != NULL) {
1219 error_propagate(errp, err);
1220 for (i = 0; i < conf->num_queues; i++) {
1221 virtio_del_queue(vdev, i);
1222 }
1223 virtio_cleanup(vdev);
1224 return;
1225 }
1226
1227 s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
1228 blk_set_dev_ops(s->blk, &virtio_block_ops, s);
1229 blk_set_guest_block_size(s->blk, s->conf.conf.logical_block_size);
1230
1231 blk_iostatus_enable(s->blk);
1232
1233 add_boot_device_lchs(dev, "/disk@0,0",
1234 conf->conf.lcyls,
1235 conf->conf.lheads,
1236 conf->conf.lsecs);
1237 }
1238
1239 static void virtio_blk_device_unrealize(DeviceState *dev)
1240 {
1241 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1242 VirtIOBlock *s = VIRTIO_BLK(dev);
1243 VirtIOBlkConf *conf = &s->conf;
1244 unsigned i;
1245
1246 blk_drain(s->blk);
1247 del_boot_device_lchs(dev, "/disk@0,0");
1248 virtio_blk_data_plane_destroy(s->dataplane);
1249 s->dataplane = NULL;
1250 for (i = 0; i < conf->num_queues; i++) {
1251 virtio_del_queue(vdev, i);
1252 }
1253 qemu_del_vm_change_state_handler(s->change);
1254 blockdev_mark_auto_del(s->blk);
1255 virtio_cleanup(vdev);
1256 }
1257
1258 static void virtio_blk_instance_init(Object *obj)
1259 {
1260 VirtIOBlock *s = VIRTIO_BLK(obj);
1261
1262 device_add_bootindex_property(obj, &s->conf.conf.bootindex,
1263 "bootindex", "/disk@0,0",
1264 DEVICE(obj));
1265 }
1266
1267 static const VMStateDescription vmstate_virtio_blk = {
1268 .name = "virtio-blk",
1269 .minimum_version_id = 2,
1270 .version_id = 2,
1271 .fields = (VMStateField[]) {
1272 VMSTATE_VIRTIO_DEVICE,
1273 VMSTATE_END_OF_LIST()
1274 },
1275 };
1276
1277 static Property virtio_blk_properties[] = {
1278 DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf),
1279 DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf),
1280 DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf),
1281 DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial),
1282 DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features,
1283 VIRTIO_BLK_F_CONFIG_WCE, true),
1284 #ifdef __linux__
1285 DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features,
1286 VIRTIO_BLK_F_SCSI, false),
1287 #endif
1288 DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
1289 true),
1290 DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues,
1291 VIRTIO_BLK_AUTO_NUM_QUEUES),
1292 DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 256),
1293 DEFINE_PROP_BOOL("seg-max-adjust", VirtIOBlock, conf.seg_max_adjust, true),
1294 DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD,
1295 IOThread *),
1296 DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features,
1297 VIRTIO_BLK_F_DISCARD, true),
1298 DEFINE_PROP_BOOL("report-discard-granularity", VirtIOBlock,
1299 conf.report_discard_granularity, true),
1300 DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features,
1301 VIRTIO_BLK_F_WRITE_ZEROES, true),
1302 DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock,
1303 conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS),
1304 DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock,
1305 conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS),
1306 DEFINE_PROP_BOOL("x-enable-wce-if-config-wce", VirtIOBlock,
1307 conf.x_enable_wce_if_config_wce, true),
1308 DEFINE_PROP_END_OF_LIST(),
1309 };
1310
1311 static void virtio_blk_class_init(ObjectClass *klass, void *data)
1312 {
1313 DeviceClass *dc = DEVICE_CLASS(klass);
1314 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1315
1316 device_class_set_props(dc, virtio_blk_properties);
1317 dc->vmsd = &vmstate_virtio_blk;
1318 set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1319 vdc->realize = virtio_blk_device_realize;
1320 vdc->unrealize = virtio_blk_device_unrealize;
1321 vdc->get_config = virtio_blk_update_config;
1322 vdc->set_config = virtio_blk_set_config;
1323 vdc->get_features = virtio_blk_get_features;
1324 vdc->set_status = virtio_blk_set_status;
1325 vdc->reset = virtio_blk_reset;
1326 vdc->save = virtio_blk_save_device;
1327 vdc->load = virtio_blk_load_device;
1328 vdc->start_ioeventfd = virtio_blk_data_plane_start;
1329 vdc->stop_ioeventfd = virtio_blk_data_plane_stop;
1330 }
1331
1332 static const TypeInfo virtio_blk_info = {
1333 .name = TYPE_VIRTIO_BLK,
1334 .parent = TYPE_VIRTIO_DEVICE,
1335 .instance_size = sizeof(VirtIOBlock),
1336 .instance_init = virtio_blk_instance_init,
1337 .class_init = virtio_blk_class_init,
1338 };
1339
1340 static void virtio_register_types(void)
1341 {
1342 type_register_static(&virtio_blk_info);
1343 }
1344
1345 type_init(virtio_register_types)