xref: /openbmc/qemu/hw/block/virtio-blk.c (revision 754cb9c0)
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-common.h"
17 #include "qemu/iov.h"
18 #include "qemu/error-report.h"
19 #include "trace.h"
20 #include "hw/block/block.h"
21 #include "sysemu/blockdev.h"
22 #include "hw/virtio/virtio-blk.h"
23 #include "dataplane/virtio-blk.h"
24 #include "scsi/constants.h"
25 #ifdef __linux__
26 # include <scsi/sg.h>
27 #endif
28 #include "hw/virtio/virtio-bus.h"
29 #include "hw/virtio/virtio-access.h"
30 
31 /* Config size before the discard support (hide associated config fields) */
32 #define VIRTIO_BLK_CFG_SIZE offsetof(struct virtio_blk_config, \
33                                      max_discard_sectors)
34 /*
35  * Starting from the discard feature, we can use this array to properly
36  * set the config size depending on the features enabled.
37  */
38 static VirtIOFeature feature_sizes[] = {
39     {.flags = 1ULL << VIRTIO_BLK_F_DISCARD,
40      .end = virtio_endof(struct virtio_blk_config, discard_sector_alignment)},
41     {.flags = 1ULL << VIRTIO_BLK_F_WRITE_ZEROES,
42      .end = virtio_endof(struct virtio_blk_config, write_zeroes_may_unmap)},
43     {}
44 };
45 
46 static void virtio_blk_set_config_size(VirtIOBlock *s, uint64_t host_features)
47 {
48     s->config_size = MAX(VIRTIO_BLK_CFG_SIZE,
49         virtio_feature_get_config_size(feature_sizes, host_features));
50 
51     assert(s->config_size <= sizeof(struct virtio_blk_config));
52 }
53 
54 static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq,
55                                     VirtIOBlockReq *req)
56 {
57     req->dev = s;
58     req->vq = vq;
59     req->qiov.size = 0;
60     req->in_len = 0;
61     req->next = NULL;
62     req->mr_next = NULL;
63 }
64 
65 static void virtio_blk_free_request(VirtIOBlockReq *req)
66 {
67     g_free(req);
68 }
69 
70 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
71 {
72     VirtIOBlock *s = req->dev;
73     VirtIODevice *vdev = VIRTIO_DEVICE(s);
74 
75     trace_virtio_blk_req_complete(vdev, req, status);
76 
77     stb_p(&req->in->status, status);
78     virtqueue_push(req->vq, &req->elem, req->in_len);
79     if (s->dataplane_started && !s->dataplane_disabled) {
80         virtio_blk_data_plane_notify(s->dataplane, req->vq);
81     } else {
82         virtio_notify(vdev, req->vq);
83     }
84 }
85 
86 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
87     bool is_read, bool acct_failed)
88 {
89     VirtIOBlock *s = req->dev;
90     BlockErrorAction action = blk_get_error_action(s->blk, is_read, error);
91 
92     if (action == BLOCK_ERROR_ACTION_STOP) {
93         /* Break the link as the next request is going to be parsed from the
94          * ring again. Otherwise we may end up doing a double completion! */
95         req->mr_next = NULL;
96         req->next = s->rq;
97         s->rq = req;
98     } else if (action == BLOCK_ERROR_ACTION_REPORT) {
99         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
100         if (acct_failed) {
101             block_acct_failed(blk_get_stats(s->blk), &req->acct);
102         }
103         virtio_blk_free_request(req);
104     }
105 
106     blk_error_action(s->blk, action, is_read, error);
107     return action != BLOCK_ERROR_ACTION_IGNORE;
108 }
109 
110 static void virtio_blk_rw_complete(void *opaque, int ret)
111 {
112     VirtIOBlockReq *next = opaque;
113     VirtIOBlock *s = next->dev;
114     VirtIODevice *vdev = VIRTIO_DEVICE(s);
115 
116     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
117     while (next) {
118         VirtIOBlockReq *req = next;
119         next = req->mr_next;
120         trace_virtio_blk_rw_complete(vdev, req, ret);
121 
122         if (req->qiov.nalloc != -1) {
123             /* If nalloc is != -1 req->qiov is a local copy of the original
124              * external iovec. It was allocated in submit_requests to be
125              * able to merge requests. */
126             qemu_iovec_destroy(&req->qiov);
127         }
128 
129         if (ret) {
130             int p = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type);
131             bool is_read = !(p & VIRTIO_BLK_T_OUT);
132             /* Note that memory may be dirtied on read failure.  If the
133              * virtio request is not completed here, as is the case for
134              * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
135              * correctly during live migration.  While this is ugly,
136              * it is acceptable because the device is free to write to
137              * the memory until the request is completed (which will
138              * happen on the other side of the migration).
139              */
140             if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) {
141                 continue;
142             }
143         }
144 
145         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
146         block_acct_done(blk_get_stats(req->dev->blk), &req->acct);
147         virtio_blk_free_request(req);
148     }
149     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
150 }
151 
152 static void virtio_blk_flush_complete(void *opaque, int ret)
153 {
154     VirtIOBlockReq *req = opaque;
155     VirtIOBlock *s = req->dev;
156 
157     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
158     if (ret) {
159         if (virtio_blk_handle_rw_error(req, -ret, 0, true)) {
160             goto out;
161         }
162     }
163 
164     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
165     block_acct_done(blk_get_stats(s->blk), &req->acct);
166     virtio_blk_free_request(req);
167 
168 out:
169     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
170 }
171 
172 static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret)
173 {
174     VirtIOBlockReq *req = opaque;
175     VirtIOBlock *s = req->dev;
176     bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) &
177                             ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES;
178 
179     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
180     if (ret) {
181         if (virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) {
182             goto out;
183         }
184     }
185 
186     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
187     if (is_write_zeroes) {
188         block_acct_done(blk_get_stats(s->blk), &req->acct);
189     }
190     virtio_blk_free_request(req);
191 
192 out:
193     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
194 }
195 
196 #ifdef __linux__
197 
198 typedef struct {
199     VirtIOBlockReq *req;
200     struct sg_io_hdr hdr;
201 } VirtIOBlockIoctlReq;
202 
203 static void virtio_blk_ioctl_complete(void *opaque, int status)
204 {
205     VirtIOBlockIoctlReq *ioctl_req = opaque;
206     VirtIOBlockReq *req = ioctl_req->req;
207     VirtIOBlock *s = req->dev;
208     VirtIODevice *vdev = VIRTIO_DEVICE(s);
209     struct virtio_scsi_inhdr *scsi;
210     struct sg_io_hdr *hdr;
211 
212     scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
213 
214     if (status) {
215         status = VIRTIO_BLK_S_UNSUPP;
216         virtio_stl_p(vdev, &scsi->errors, 255);
217         goto out;
218     }
219 
220     hdr = &ioctl_req->hdr;
221     /*
222      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
223      * clear the masked_status field [hence status gets cleared too, see
224      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
225      * status has occurred.  However they do set DRIVER_SENSE in driver_status
226      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
227      */
228     if (hdr->status == 0 && hdr->sb_len_wr > 0) {
229         hdr->status = CHECK_CONDITION;
230     }
231 
232     virtio_stl_p(vdev, &scsi->errors,
233                  hdr->status | (hdr->msg_status << 8) |
234                  (hdr->host_status << 16) | (hdr->driver_status << 24));
235     virtio_stl_p(vdev, &scsi->residual, hdr->resid);
236     virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr);
237     virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
238 
239 out:
240     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
241     virtio_blk_req_complete(req, status);
242     virtio_blk_free_request(req);
243     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
244     g_free(ioctl_req);
245 }
246 
247 #endif
248 
249 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq)
250 {
251     VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq));
252 
253     if (req) {
254         virtio_blk_init_request(s, vq, req);
255     }
256     return req;
257 }
258 
259 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req)
260 {
261     int status = VIRTIO_BLK_S_OK;
262     struct virtio_scsi_inhdr *scsi = NULL;
263     VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
264     VirtQueueElement *elem = &req->elem;
265     VirtIOBlock *blk = req->dev;
266 
267 #ifdef __linux__
268     int i;
269     VirtIOBlockIoctlReq *ioctl_req;
270     BlockAIOCB *acb;
271 #endif
272 
273     /*
274      * We require at least one output segment each for the virtio_blk_outhdr
275      * and the SCSI command block.
276      *
277      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
278      * and the sense buffer pointer in the input segments.
279      */
280     if (elem->out_num < 2 || elem->in_num < 3) {
281         status = VIRTIO_BLK_S_IOERR;
282         goto fail;
283     }
284 
285     /*
286      * The scsi inhdr is placed in the second-to-last input segment, just
287      * before the regular inhdr.
288      */
289     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
290 
291     if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) {
292         status = VIRTIO_BLK_S_UNSUPP;
293         goto fail;
294     }
295 
296     /*
297      * No support for bidirection commands yet.
298      */
299     if (elem->out_num > 2 && elem->in_num > 3) {
300         status = VIRTIO_BLK_S_UNSUPP;
301         goto fail;
302     }
303 
304 #ifdef __linux__
305     ioctl_req = g_new0(VirtIOBlockIoctlReq, 1);
306     ioctl_req->req = req;
307     ioctl_req->hdr.interface_id = 'S';
308     ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len;
309     ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base;
310     ioctl_req->hdr.dxfer_len = 0;
311 
312     if (elem->out_num > 2) {
313         /*
314          * If there are more than the minimally required 2 output segments
315          * there is write payload starting from the third iovec.
316          */
317         ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV;
318         ioctl_req->hdr.iovec_count = elem->out_num - 2;
319 
320         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
321             ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
322         }
323 
324         ioctl_req->hdr.dxferp = elem->out_sg + 2;
325 
326     } else if (elem->in_num > 3) {
327         /*
328          * If we have more than 3 input segments the guest wants to actually
329          * read data.
330          */
331         ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV;
332         ioctl_req->hdr.iovec_count = elem->in_num - 3;
333         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
334             ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len;
335         }
336 
337         ioctl_req->hdr.dxferp = elem->in_sg;
338     } else {
339         /*
340          * Some SCSI commands don't actually transfer any data.
341          */
342         ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE;
343     }
344 
345     ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
346     ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
347 
348     acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr,
349                         virtio_blk_ioctl_complete, ioctl_req);
350     if (!acb) {
351         g_free(ioctl_req);
352         status = VIRTIO_BLK_S_UNSUPP;
353         goto fail;
354     }
355     return -EINPROGRESS;
356 #else
357     abort();
358 #endif
359 
360 fail:
361     /* Just put anything nonzero so that the ioctl fails in the guest.  */
362     if (scsi) {
363         virtio_stl_p(vdev, &scsi->errors, 255);
364     }
365     return status;
366 }
367 
368 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
369 {
370     int status;
371 
372     status = virtio_blk_handle_scsi_req(req);
373     if (status != -EINPROGRESS) {
374         virtio_blk_req_complete(req, status);
375         virtio_blk_free_request(req);
376     }
377 }
378 
379 static inline void submit_requests(BlockBackend *blk, MultiReqBuffer *mrb,
380                                    int start, int num_reqs, int niov)
381 {
382     QEMUIOVector *qiov = &mrb->reqs[start]->qiov;
383     int64_t sector_num = mrb->reqs[start]->sector_num;
384     bool is_write = mrb->is_write;
385 
386     if (num_reqs > 1) {
387         int i;
388         struct iovec *tmp_iov = qiov->iov;
389         int tmp_niov = qiov->niov;
390 
391         /* mrb->reqs[start]->qiov was initialized from external so we can't
392          * modify it here. We need to initialize it locally and then add the
393          * external iovecs. */
394         qemu_iovec_init(qiov, niov);
395 
396         for (i = 0; i < tmp_niov; i++) {
397             qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len);
398         }
399 
400         for (i = start + 1; i < start + num_reqs; i++) {
401             qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0,
402                               mrb->reqs[i]->qiov.size);
403             mrb->reqs[i - 1]->mr_next = mrb->reqs[i];
404         }
405 
406         trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev),
407                                          mrb, start, num_reqs,
408                                          sector_num << BDRV_SECTOR_BITS,
409                                          qiov->size, is_write);
410         block_acct_merge_done(blk_get_stats(blk),
411                               is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ,
412                               num_reqs - 1);
413     }
414 
415     if (is_write) {
416         blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
417                         virtio_blk_rw_complete, mrb->reqs[start]);
418     } else {
419         blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
420                        virtio_blk_rw_complete, mrb->reqs[start]);
421     }
422 }
423 
424 static int multireq_compare(const void *a, const void *b)
425 {
426     const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a,
427                          *req2 = *(VirtIOBlockReq **)b;
428 
429     /*
430      * Note that we can't simply subtract sector_num1 from sector_num2
431      * here as that could overflow the return value.
432      */
433     if (req1->sector_num > req2->sector_num) {
434         return 1;
435     } else if (req1->sector_num < req2->sector_num) {
436         return -1;
437     } else {
438         return 0;
439     }
440 }
441 
442 static void virtio_blk_submit_multireq(BlockBackend *blk, MultiReqBuffer *mrb)
443 {
444     int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0;
445     uint32_t max_transfer;
446     int64_t sector_num = 0;
447 
448     if (mrb->num_reqs == 1) {
449         submit_requests(blk, mrb, 0, 1, -1);
450         mrb->num_reqs = 0;
451         return;
452     }
453 
454     max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk);
455 
456     qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs),
457           &multireq_compare);
458 
459     for (i = 0; i < mrb->num_reqs; i++) {
460         VirtIOBlockReq *req = mrb->reqs[i];
461         if (num_reqs > 0) {
462             /*
463              * NOTE: We cannot merge the requests in below situations:
464              * 1. requests are not sequential
465              * 2. merge would exceed maximum number of IOVs
466              * 3. merge would exceed maximum transfer length of backend device
467              */
468             if (sector_num + nb_sectors != req->sector_num ||
469                 niov > blk_get_max_iov(blk) - req->qiov.niov ||
470                 req->qiov.size > max_transfer ||
471                 nb_sectors > (max_transfer -
472                               req->qiov.size) / BDRV_SECTOR_SIZE) {
473                 submit_requests(blk, mrb, start, num_reqs, niov);
474                 num_reqs = 0;
475             }
476         }
477 
478         if (num_reqs == 0) {
479             sector_num = req->sector_num;
480             nb_sectors = niov = 0;
481             start = i;
482         }
483 
484         nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE;
485         niov += req->qiov.niov;
486         num_reqs++;
487     }
488 
489     submit_requests(blk, mrb, start, num_reqs, niov);
490     mrb->num_reqs = 0;
491 }
492 
493 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
494 {
495     block_acct_start(blk_get_stats(req->dev->blk), &req->acct, 0,
496                      BLOCK_ACCT_FLUSH);
497 
498     /*
499      * Make sure all outstanding writes are posted to the backing device.
500      */
501     if (mrb->is_write && mrb->num_reqs > 0) {
502         virtio_blk_submit_multireq(req->dev->blk, mrb);
503     }
504     blk_aio_flush(req->dev->blk, virtio_blk_flush_complete, req);
505 }
506 
507 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
508                                      uint64_t sector, size_t size)
509 {
510     uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
511     uint64_t total_sectors;
512 
513     if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
514         return false;
515     }
516     if (sector & dev->sector_mask) {
517         return false;
518     }
519     if (size % dev->conf.conf.logical_block_size) {
520         return false;
521     }
522     blk_get_geometry(dev->blk, &total_sectors);
523     if (sector > total_sectors || nb_sectors > total_sectors - sector) {
524         return false;
525     }
526     return true;
527 }
528 
529 static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req,
530     struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes)
531 {
532     VirtIOBlock *s = req->dev;
533     VirtIODevice *vdev = VIRTIO_DEVICE(s);
534     uint64_t sector;
535     uint32_t num_sectors, flags, max_sectors;
536     uint8_t err_status;
537     int bytes;
538 
539     sector = virtio_ldq_p(vdev, &dwz_hdr->sector);
540     num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors);
541     flags = virtio_ldl_p(vdev, &dwz_hdr->flags);
542     max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors :
543                   s->conf.max_discard_sectors;
544 
545     /*
546      * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check
547      * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in
548      * the integer variable.
549      */
550     if (unlikely(num_sectors > max_sectors)) {
551         err_status = VIRTIO_BLK_S_IOERR;
552         goto err;
553     }
554 
555     bytes = num_sectors << BDRV_SECTOR_BITS;
556 
557     if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) {
558         err_status = VIRTIO_BLK_S_IOERR;
559         goto err;
560     }
561 
562     /*
563      * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard
564      * and write zeroes commands if any unknown flag is set.
565      */
566     if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
567         err_status = VIRTIO_BLK_S_UNSUPP;
568         goto err;
569     }
570 
571     if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */
572         int blk_aio_flags = 0;
573 
574         if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) {
575             blk_aio_flags |= BDRV_REQ_MAY_UNMAP;
576         }
577 
578         block_acct_start(blk_get_stats(s->blk), &req->acct, bytes,
579                          BLOCK_ACCT_WRITE);
580 
581         blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS,
582                               bytes, blk_aio_flags,
583                               virtio_blk_discard_write_zeroes_complete, req);
584     } else { /* VIRTIO_BLK_T_DISCARD */
585         /*
586          * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for
587          * discard commands if the unmap flag is set.
588          */
589         if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
590             err_status = VIRTIO_BLK_S_UNSUPP;
591             goto err;
592         }
593 
594         blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes,
595                          virtio_blk_discard_write_zeroes_complete, req);
596     }
597 
598     return VIRTIO_BLK_S_OK;
599 
600 err:
601     if (is_write_zeroes) {
602         block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE);
603     }
604     return err_status;
605 }
606 
607 static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
608 {
609     uint32_t type;
610     struct iovec *in_iov = req->elem.in_sg;
611     struct iovec *out_iov = req->elem.out_sg;
612     unsigned in_num = req->elem.in_num;
613     unsigned out_num = req->elem.out_num;
614     VirtIOBlock *s = req->dev;
615     VirtIODevice *vdev = VIRTIO_DEVICE(s);
616 
617     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
618         virtio_error(vdev, "virtio-blk missing headers");
619         return -1;
620     }
621 
622     if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out,
623                             sizeof(req->out)) != sizeof(req->out))) {
624         virtio_error(vdev, "virtio-blk request outhdr too short");
625         return -1;
626     }
627 
628     iov_discard_front(&out_iov, &out_num, sizeof(req->out));
629 
630     if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
631         virtio_error(vdev, "virtio-blk request inhdr too short");
632         return -1;
633     }
634 
635     /* We always touch the last byte, so just see how big in_iov is.  */
636     req->in_len = iov_size(in_iov, in_num);
637     req->in = (void *)in_iov[in_num - 1].iov_base
638               + in_iov[in_num - 1].iov_len
639               - sizeof(struct virtio_blk_inhdr);
640     iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr));
641 
642     type = virtio_ldl_p(vdev, &req->out.type);
643 
644     /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
645      * is an optional flag. Although a guest should not send this flag if
646      * not negotiated we ignored it in the past. So keep ignoring it. */
647     switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) {
648     case VIRTIO_BLK_T_IN:
649     {
650         bool is_write = type & VIRTIO_BLK_T_OUT;
651         req->sector_num = virtio_ldq_p(vdev, &req->out.sector);
652 
653         if (is_write) {
654             qemu_iovec_init_external(&req->qiov, out_iov, out_num);
655             trace_virtio_blk_handle_write(vdev, req, req->sector_num,
656                                           req->qiov.size / BDRV_SECTOR_SIZE);
657         } else {
658             qemu_iovec_init_external(&req->qiov, in_iov, in_num);
659             trace_virtio_blk_handle_read(vdev, req, req->sector_num,
660                                          req->qiov.size / BDRV_SECTOR_SIZE);
661         }
662 
663         if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) {
664             virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
665             block_acct_invalid(blk_get_stats(s->blk),
666                                is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
667             virtio_blk_free_request(req);
668             return 0;
669         }
670 
671         block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size,
672                          is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
673 
674         /* merge would exceed maximum number of requests or IO direction
675          * changes */
676         if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS ||
677                                   is_write != mrb->is_write ||
678                                   !s->conf.request_merging)) {
679             virtio_blk_submit_multireq(s->blk, mrb);
680         }
681 
682         assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS);
683         mrb->reqs[mrb->num_reqs++] = req;
684         mrb->is_write = is_write;
685         break;
686     }
687     case VIRTIO_BLK_T_FLUSH:
688         virtio_blk_handle_flush(req, mrb);
689         break;
690     case VIRTIO_BLK_T_SCSI_CMD:
691         virtio_blk_handle_scsi(req);
692         break;
693     case VIRTIO_BLK_T_GET_ID:
694     {
695         /*
696          * NB: per existing s/n string convention the string is
697          * terminated by '\0' only when shorter than buffer.
698          */
699         const char *serial = s->conf.serial ? s->conf.serial : "";
700         size_t size = MIN(strlen(serial) + 1,
701                           MIN(iov_size(in_iov, in_num),
702                               VIRTIO_BLK_ID_BYTES));
703         iov_from_buf(in_iov, in_num, 0, serial, size);
704         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
705         virtio_blk_free_request(req);
706         break;
707     }
708     /*
709      * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with
710      * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement,
711      * so we must mask it for these requests, then we will check if it is set.
712      */
713     case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT:
714     case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT:
715     {
716         struct virtio_blk_discard_write_zeroes dwz_hdr;
717         size_t out_len = iov_size(out_iov, out_num);
718         bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) ==
719                                VIRTIO_BLK_T_WRITE_ZEROES;
720         uint8_t err_status;
721 
722         /*
723          * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains
724          * more than one segment.
725          */
726         if (unlikely(!(type & VIRTIO_BLK_T_OUT) ||
727                      out_len > sizeof(dwz_hdr))) {
728             virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
729             virtio_blk_free_request(req);
730             return 0;
731         }
732 
733         if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr,
734                                 sizeof(dwz_hdr)) != sizeof(dwz_hdr))) {
735             virtio_error(vdev, "virtio-blk discard/write_zeroes header"
736                          " too short");
737             return -1;
738         }
739 
740         err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr,
741                                                             is_write_zeroes);
742         if (err_status != VIRTIO_BLK_S_OK) {
743             virtio_blk_req_complete(req, err_status);
744             virtio_blk_free_request(req);
745         }
746 
747         break;
748     }
749     default:
750         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
751         virtio_blk_free_request(req);
752     }
753     return 0;
754 }
755 
756 bool virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
757 {
758     VirtIOBlockReq *req;
759     MultiReqBuffer mrb = {};
760     bool progress = false;
761 
762     aio_context_acquire(blk_get_aio_context(s->blk));
763     blk_io_plug(s->blk);
764 
765     do {
766         virtio_queue_set_notification(vq, 0);
767 
768         while ((req = virtio_blk_get_request(s, vq))) {
769             progress = true;
770             if (virtio_blk_handle_request(req, &mrb)) {
771                 virtqueue_detach_element(req->vq, &req->elem, 0);
772                 virtio_blk_free_request(req);
773                 break;
774             }
775         }
776 
777         virtio_queue_set_notification(vq, 1);
778     } while (!virtio_queue_empty(vq));
779 
780     if (mrb.num_reqs) {
781         virtio_blk_submit_multireq(s->blk, &mrb);
782     }
783 
784     blk_io_unplug(s->blk);
785     aio_context_release(blk_get_aio_context(s->blk));
786     return progress;
787 }
788 
789 static void virtio_blk_handle_output_do(VirtIOBlock *s, VirtQueue *vq)
790 {
791     virtio_blk_handle_vq(s, vq);
792 }
793 
794 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
795 {
796     VirtIOBlock *s = (VirtIOBlock *)vdev;
797 
798     if (s->dataplane) {
799         /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
800          * dataplane here instead of waiting for .set_status().
801          */
802         virtio_device_start_ioeventfd(vdev);
803         if (!s->dataplane_disabled) {
804             return;
805         }
806     }
807     virtio_blk_handle_output_do(s, vq);
808 }
809 
810 static void virtio_blk_dma_restart_bh(void *opaque)
811 {
812     VirtIOBlock *s = opaque;
813     VirtIOBlockReq *req = s->rq;
814     MultiReqBuffer mrb = {};
815 
816     qemu_bh_delete(s->bh);
817     s->bh = NULL;
818 
819     s->rq = NULL;
820 
821     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
822     while (req) {
823         VirtIOBlockReq *next = req->next;
824         if (virtio_blk_handle_request(req, &mrb)) {
825             /* Device is now broken and won't do any processing until it gets
826              * reset. Already queued requests will be lost: let's purge them.
827              */
828             while (req) {
829                 next = req->next;
830                 virtqueue_detach_element(req->vq, &req->elem, 0);
831                 virtio_blk_free_request(req);
832                 req = next;
833             }
834             break;
835         }
836         req = next;
837     }
838 
839     if (mrb.num_reqs) {
840         virtio_blk_submit_multireq(s->blk, &mrb);
841     }
842     blk_dec_in_flight(s->conf.conf.blk);
843     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
844 }
845 
846 static void virtio_blk_dma_restart_cb(void *opaque, int running,
847                                       RunState state)
848 {
849     VirtIOBlock *s = opaque;
850 
851     if (!running) {
852         return;
853     }
854 
855     if (!s->bh) {
856         /* FIXME The data plane is not started yet, so these requests are
857          * processed in the main thread. */
858         s->bh = aio_bh_new(blk_get_aio_context(s->conf.conf.blk),
859                            virtio_blk_dma_restart_bh, s);
860         blk_inc_in_flight(s->conf.conf.blk);
861         qemu_bh_schedule(s->bh);
862     }
863 }
864 
865 static void virtio_blk_reset(VirtIODevice *vdev)
866 {
867     VirtIOBlock *s = VIRTIO_BLK(vdev);
868     AioContext *ctx;
869     VirtIOBlockReq *req;
870 
871     ctx = blk_get_aio_context(s->blk);
872     aio_context_acquire(ctx);
873     blk_drain(s->blk);
874 
875     /* We drop queued requests after blk_drain() because blk_drain() itself can
876      * produce them. */
877     while (s->rq) {
878         req = s->rq;
879         s->rq = req->next;
880         virtqueue_detach_element(req->vq, &req->elem, 0);
881         virtio_blk_free_request(req);
882     }
883 
884     aio_context_release(ctx);
885 
886     assert(!s->dataplane_started);
887     blk_set_enable_write_cache(s->blk, s->original_wce);
888 }
889 
890 /* coalesce internal state, copy to pci i/o region 0
891  */
892 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
893 {
894     VirtIOBlock *s = VIRTIO_BLK(vdev);
895     BlockConf *conf = &s->conf.conf;
896     struct virtio_blk_config blkcfg;
897     uint64_t capacity;
898     int64_t length;
899     int blk_size = conf->logical_block_size;
900 
901     blk_get_geometry(s->blk, &capacity);
902     memset(&blkcfg, 0, sizeof(blkcfg));
903     virtio_stq_p(vdev, &blkcfg.capacity, capacity);
904     virtio_stl_p(vdev, &blkcfg.seg_max, 128 - 2);
905     virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls);
906     virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
907     virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size);
908     virtio_stw_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size);
909     blkcfg.geometry.heads = conf->heads;
910     /*
911      * We must ensure that the block device capacity is a multiple of
912      * the logical block size. If that is not the case, let's use
913      * sector_mask to adopt the geometry to have a correct picture.
914      * For those devices where the capacity is ok for the given geometry
915      * we don't touch the sector value of the geometry, since some devices
916      * (like s390 dasd) need a specific value. Here the capacity is already
917      * cyls*heads*secs*blk_size and the sector value is not block size
918      * divided by 512 - instead it is the amount of blk_size blocks
919      * per track (cylinder).
920      */
921     length = blk_getlength(s->blk);
922     if (length > 0 && length / conf->heads / conf->secs % blk_size) {
923         blkcfg.geometry.sectors = conf->secs & ~s->sector_mask;
924     } else {
925         blkcfg.geometry.sectors = conf->secs;
926     }
927     blkcfg.size_max = 0;
928     blkcfg.physical_block_exp = get_physical_block_exp(conf);
929     blkcfg.alignment_offset = 0;
930     blkcfg.wce = blk_enable_write_cache(s->blk);
931     virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues);
932     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) {
933         virtio_stl_p(vdev, &blkcfg.max_discard_sectors,
934                      s->conf.max_discard_sectors);
935         virtio_stl_p(vdev, &blkcfg.discard_sector_alignment,
936                      blk_size >> BDRV_SECTOR_BITS);
937         /*
938          * We support only one segment per request since multiple segments
939          * are not widely used and there are no userspace APIs that allow
940          * applications to submit multiple segments in a single call.
941          */
942         virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1);
943     }
944     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) {
945         virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors,
946                      s->conf.max_write_zeroes_sectors);
947         blkcfg.write_zeroes_may_unmap = 1;
948         virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1);
949     }
950     memcpy(config, &blkcfg, s->config_size);
951 }
952 
953 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
954 {
955     VirtIOBlock *s = VIRTIO_BLK(vdev);
956     struct virtio_blk_config blkcfg;
957 
958     memcpy(&blkcfg, config, s->config_size);
959 
960     aio_context_acquire(blk_get_aio_context(s->blk));
961     blk_set_enable_write_cache(s->blk, blkcfg.wce != 0);
962     aio_context_release(blk_get_aio_context(s->blk));
963 }
964 
965 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features,
966                                         Error **errp)
967 {
968     VirtIOBlock *s = VIRTIO_BLK(vdev);
969 
970     /* Firstly sync all virtio-blk possible supported features */
971     features |= s->host_features;
972 
973     virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
974     virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
975     virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
976     virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
977     if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) {
978         if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) {
979             error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
980             return 0;
981         }
982     } else {
983         virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT);
984         virtio_add_feature(&features, VIRTIO_BLK_F_SCSI);
985     }
986 
987     if (blk_enable_write_cache(s->blk)) {
988         virtio_add_feature(&features, VIRTIO_BLK_F_WCE);
989     }
990     if (blk_is_read_only(s->blk)) {
991         virtio_add_feature(&features, VIRTIO_BLK_F_RO);
992     }
993     if (s->conf.num_queues > 1) {
994         virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
995     }
996 
997     return features;
998 }
999 
1000 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
1001 {
1002     VirtIOBlock *s = VIRTIO_BLK(vdev);
1003 
1004     if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) {
1005         assert(!s->dataplane_started);
1006     }
1007 
1008     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1009         return;
1010     }
1011 
1012     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
1013      * cache flushes.  Thus, the "auto writethrough" behavior is never
1014      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
1015      * Leaving it enabled would break the following sequence:
1016      *
1017      *     Guest started with "-drive cache=writethrough"
1018      *     Guest sets status to 0
1019      *     Guest sets DRIVER bit in status field
1020      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
1021      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
1022      *     Guest writes 1 to the WCE configuration field (writeback mode)
1023      *     Guest sets DRIVER_OK bit in status field
1024      *
1025      * s->blk would erroneously be placed in writethrough mode.
1026      */
1027     if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) {
1028         aio_context_acquire(blk_get_aio_context(s->blk));
1029         blk_set_enable_write_cache(s->blk,
1030                                    virtio_vdev_has_feature(vdev,
1031                                                            VIRTIO_BLK_F_WCE));
1032         aio_context_release(blk_get_aio_context(s->blk));
1033     }
1034 }
1035 
1036 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
1037 {
1038     VirtIOBlock *s = VIRTIO_BLK(vdev);
1039     VirtIOBlockReq *req = s->rq;
1040 
1041     while (req) {
1042         qemu_put_sbyte(f, 1);
1043 
1044         if (s->conf.num_queues > 1) {
1045             qemu_put_be32(f, virtio_get_queue_index(req->vq));
1046         }
1047 
1048         qemu_put_virtqueue_element(f, &req->elem);
1049         req = req->next;
1050     }
1051     qemu_put_sbyte(f, 0);
1052 }
1053 
1054 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
1055                                   int version_id)
1056 {
1057     VirtIOBlock *s = VIRTIO_BLK(vdev);
1058 
1059     while (qemu_get_sbyte(f)) {
1060         unsigned nvqs = s->conf.num_queues;
1061         unsigned vq_idx = 0;
1062         VirtIOBlockReq *req;
1063 
1064         if (nvqs > 1) {
1065             vq_idx = qemu_get_be32(f);
1066 
1067             if (vq_idx >= nvqs) {
1068                 error_report("Invalid virtqueue index in request list: %#x",
1069                              vq_idx);
1070                 return -EINVAL;
1071             }
1072         }
1073 
1074         req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq));
1075         virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req);
1076         req->next = s->rq;
1077         s->rq = req;
1078     }
1079 
1080     return 0;
1081 }
1082 
1083 static void virtio_blk_resize(void *opaque)
1084 {
1085     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
1086 
1087     virtio_notify_config(vdev);
1088 }
1089 
1090 static const BlockDevOps virtio_block_ops = {
1091     .resize_cb = virtio_blk_resize,
1092 };
1093 
1094 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
1095 {
1096     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1097     VirtIOBlock *s = VIRTIO_BLK(dev);
1098     VirtIOBlkConf *conf = &s->conf;
1099     Error *err = NULL;
1100     unsigned i;
1101 
1102     if (!conf->conf.blk) {
1103         error_setg(errp, "drive property not set");
1104         return;
1105     }
1106     if (!blk_is_inserted(conf->conf.blk)) {
1107         error_setg(errp, "Device needs media, but drive is empty");
1108         return;
1109     }
1110     if (!conf->num_queues) {
1111         error_setg(errp, "num-queues property must be larger than 0");
1112         return;
1113     }
1114     if (!is_power_of_2(conf->queue_size) ||
1115         conf->queue_size > VIRTQUEUE_MAX_SIZE) {
1116         error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1117                    "must be a power of 2 (max %d)",
1118                    conf->queue_size, VIRTQUEUE_MAX_SIZE);
1119         return;
1120     }
1121 
1122     if (!blkconf_apply_backend_options(&conf->conf,
1123                                        blk_is_read_only(conf->conf.blk), true,
1124                                        errp)) {
1125         return;
1126     }
1127     s->original_wce = blk_enable_write_cache(conf->conf.blk);
1128     if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) {
1129         return;
1130     }
1131 
1132     blkconf_blocksizes(&conf->conf);
1133 
1134     if (conf->conf.logical_block_size >
1135         conf->conf.physical_block_size) {
1136         error_setg(errp,
1137                    "logical_block_size > physical_block_size not supported");
1138         return;
1139     }
1140 
1141     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) &&
1142         (!conf->max_discard_sectors ||
1143          conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1144         error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")"
1145                    ", must be between 1 and %d",
1146                    conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS);
1147         return;
1148     }
1149 
1150     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) &&
1151         (!conf->max_write_zeroes_sectors ||
1152          conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1153         error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32
1154                    "), must be between 1 and %d",
1155                    conf->max_write_zeroes_sectors,
1156                    (int)BDRV_REQUEST_MAX_SECTORS);
1157         return;
1158     }
1159 
1160     virtio_blk_set_config_size(s, s->host_features);
1161 
1162     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK, s->config_size);
1163 
1164     s->blk = conf->conf.blk;
1165     s->rq = NULL;
1166     s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1;
1167 
1168     for (i = 0; i < conf->num_queues; i++) {
1169         virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output);
1170     }
1171     virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err);
1172     if (err != NULL) {
1173         error_propagate(errp, err);
1174         virtio_cleanup(vdev);
1175         return;
1176     }
1177 
1178     s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
1179     blk_set_dev_ops(s->blk, &virtio_block_ops, s);
1180     blk_set_guest_block_size(s->blk, s->conf.conf.logical_block_size);
1181 
1182     blk_iostatus_enable(s->blk);
1183 }
1184 
1185 static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp)
1186 {
1187     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1188     VirtIOBlock *s = VIRTIO_BLK(dev);
1189 
1190     virtio_blk_data_plane_destroy(s->dataplane);
1191     s->dataplane = NULL;
1192     qemu_del_vm_change_state_handler(s->change);
1193     blockdev_mark_auto_del(s->blk);
1194     virtio_cleanup(vdev);
1195 }
1196 
1197 static void virtio_blk_instance_init(Object *obj)
1198 {
1199     VirtIOBlock *s = VIRTIO_BLK(obj);
1200 
1201     device_add_bootindex_property(obj, &s->conf.conf.bootindex,
1202                                   "bootindex", "/disk@0,0",
1203                                   DEVICE(obj), NULL);
1204 }
1205 
1206 static const VMStateDescription vmstate_virtio_blk = {
1207     .name = "virtio-blk",
1208     .minimum_version_id = 2,
1209     .version_id = 2,
1210     .fields = (VMStateField[]) {
1211         VMSTATE_VIRTIO_DEVICE,
1212         VMSTATE_END_OF_LIST()
1213     },
1214 };
1215 
1216 static Property virtio_blk_properties[] = {
1217     DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf),
1218     DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf),
1219     DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf),
1220     DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial),
1221     DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features,
1222                       VIRTIO_BLK_F_CONFIG_WCE, true),
1223 #ifdef __linux__
1224     DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features,
1225                       VIRTIO_BLK_F_SCSI, false),
1226 #endif
1227     DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
1228                     true),
1229     DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 1),
1230     DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 128),
1231     DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD,
1232                      IOThread *),
1233     DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features,
1234                       VIRTIO_BLK_F_DISCARD, true),
1235     DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features,
1236                       VIRTIO_BLK_F_WRITE_ZEROES, true),
1237     DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock,
1238                        conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS),
1239     DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock,
1240                        conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS),
1241     DEFINE_PROP_END_OF_LIST(),
1242 };
1243 
1244 static void virtio_blk_class_init(ObjectClass *klass, void *data)
1245 {
1246     DeviceClass *dc = DEVICE_CLASS(klass);
1247     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1248 
1249     dc->props = virtio_blk_properties;
1250     dc->vmsd = &vmstate_virtio_blk;
1251     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1252     vdc->realize = virtio_blk_device_realize;
1253     vdc->unrealize = virtio_blk_device_unrealize;
1254     vdc->get_config = virtio_blk_update_config;
1255     vdc->set_config = virtio_blk_set_config;
1256     vdc->get_features = virtio_blk_get_features;
1257     vdc->set_status = virtio_blk_set_status;
1258     vdc->reset = virtio_blk_reset;
1259     vdc->save = virtio_blk_save_device;
1260     vdc->load = virtio_blk_load_device;
1261     vdc->start_ioeventfd = virtio_blk_data_plane_start;
1262     vdc->stop_ioeventfd = virtio_blk_data_plane_stop;
1263 }
1264 
1265 static const TypeInfo virtio_blk_info = {
1266     .name = TYPE_VIRTIO_BLK,
1267     .parent = TYPE_VIRTIO_DEVICE,
1268     .instance_size = sizeof(VirtIOBlock),
1269     .instance_init = virtio_blk_instance_init,
1270     .class_init = virtio_blk_class_init,
1271 };
1272 
1273 static void virtio_register_types(void)
1274 {
1275     type_register_static(&virtio_blk_info);
1276 }
1277 
1278 type_init(virtio_register_types)
1279