xref: /openbmc/qemu/hw/block/virtio-blk.c (revision 2d7ac0af)
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 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     virtqueue_push(req->vq, &req->elem, req->in_len);
84     if (s->dataplane_started && !s->dataplane_disabled) {
85         virtio_blk_data_plane_notify(s->dataplane, req->vq);
86     } else {
87         virtio_notify(vdev, req->vq);
88     }
89 }
90 
91 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
92     bool is_read, bool acct_failed)
93 {
94     VirtIOBlock *s = req->dev;
95     BlockErrorAction action = blk_get_error_action(s->blk, is_read, error);
96 
97     if (action == BLOCK_ERROR_ACTION_STOP) {
98         /* Break the link as the next request is going to be parsed from the
99          * ring again. Otherwise we may end up doing a double completion! */
100         req->mr_next = NULL;
101         req->next = s->rq;
102         s->rq = req;
103     } else if (action == BLOCK_ERROR_ACTION_REPORT) {
104         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
105         if (acct_failed) {
106             block_acct_failed(blk_get_stats(s->blk), &req->acct);
107         }
108         virtio_blk_free_request(req);
109     }
110 
111     blk_error_action(s->blk, action, is_read, error);
112     return action != BLOCK_ERROR_ACTION_IGNORE;
113 }
114 
115 static void virtio_blk_rw_complete(void *opaque, int ret)
116 {
117     VirtIOBlockReq *next = opaque;
118     VirtIOBlock *s = next->dev;
119     VirtIODevice *vdev = VIRTIO_DEVICE(s);
120 
121     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
122     while (next) {
123         VirtIOBlockReq *req = next;
124         next = req->mr_next;
125         trace_virtio_blk_rw_complete(vdev, req, ret);
126 
127         if (req->qiov.nalloc != -1) {
128             /* If nalloc is != -1 req->qiov is a local copy of the original
129              * external iovec. It was allocated in submit_requests to be
130              * able to merge requests. */
131             qemu_iovec_destroy(&req->qiov);
132         }
133 
134         if (ret) {
135             int p = virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type);
136             bool is_read = !(p & VIRTIO_BLK_T_OUT);
137             /* Note that memory may be dirtied on read failure.  If the
138              * virtio request is not completed here, as is the case for
139              * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
140              * correctly during live migration.  While this is ugly,
141              * it is acceptable because the device is free to write to
142              * the memory until the request is completed (which will
143              * happen on the other side of the migration).
144              */
145             if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) {
146                 continue;
147             }
148         }
149 
150         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
151         block_acct_done(blk_get_stats(s->blk), &req->acct);
152         virtio_blk_free_request(req);
153     }
154     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
155 }
156 
157 static void virtio_blk_flush_complete(void *opaque, int ret)
158 {
159     VirtIOBlockReq *req = opaque;
160     VirtIOBlock *s = req->dev;
161 
162     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
163     if (ret) {
164         if (virtio_blk_handle_rw_error(req, -ret, 0, true)) {
165             goto out;
166         }
167     }
168 
169     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
170     block_acct_done(blk_get_stats(s->blk), &req->acct);
171     virtio_blk_free_request(req);
172 
173 out:
174     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
175 }
176 
177 static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret)
178 {
179     VirtIOBlockReq *req = opaque;
180     VirtIOBlock *s = req->dev;
181     bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) &
182                             ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES;
183 
184     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
185     if (ret) {
186         if (virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) {
187             goto out;
188         }
189     }
190 
191     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
192     if (is_write_zeroes) {
193         block_acct_done(blk_get_stats(s->blk), &req->acct);
194     }
195     virtio_blk_free_request(req);
196 
197 out:
198     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
199 }
200 
201 #ifdef __linux__
202 
203 typedef struct {
204     VirtIOBlockReq *req;
205     struct sg_io_hdr hdr;
206 } VirtIOBlockIoctlReq;
207 
208 static void virtio_blk_ioctl_complete(void *opaque, int status)
209 {
210     VirtIOBlockIoctlReq *ioctl_req = opaque;
211     VirtIOBlockReq *req = ioctl_req->req;
212     VirtIOBlock *s = req->dev;
213     VirtIODevice *vdev = VIRTIO_DEVICE(s);
214     struct virtio_scsi_inhdr *scsi;
215     struct sg_io_hdr *hdr;
216 
217     scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
218 
219     if (status) {
220         status = VIRTIO_BLK_S_UNSUPP;
221         virtio_stl_p(vdev, &scsi->errors, 255);
222         goto out;
223     }
224 
225     hdr = &ioctl_req->hdr;
226     /*
227      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
228      * clear the masked_status field [hence status gets cleared too, see
229      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
230      * status has occurred.  However they do set DRIVER_SENSE in driver_status
231      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
232      */
233     if (hdr->status == 0 && hdr->sb_len_wr > 0) {
234         hdr->status = CHECK_CONDITION;
235     }
236 
237     virtio_stl_p(vdev, &scsi->errors,
238                  hdr->status | (hdr->msg_status << 8) |
239                  (hdr->host_status << 16) | (hdr->driver_status << 24));
240     virtio_stl_p(vdev, &scsi->residual, hdr->resid);
241     virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr);
242     virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
243 
244 out:
245     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
246     virtio_blk_req_complete(req, status);
247     virtio_blk_free_request(req);
248     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
249     g_free(ioctl_req);
250 }
251 
252 #endif
253 
254 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq)
255 {
256     VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq));
257 
258     if (req) {
259         virtio_blk_init_request(s, vq, req);
260     }
261     return req;
262 }
263 
264 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req)
265 {
266     int status = VIRTIO_BLK_S_OK;
267     struct virtio_scsi_inhdr *scsi = NULL;
268     VirtIOBlock *blk = req->dev;
269     VirtIODevice *vdev = VIRTIO_DEVICE(blk);
270     VirtQueueElement *elem = &req->elem;
271 
272 #ifdef __linux__
273     int i;
274     VirtIOBlockIoctlReq *ioctl_req;
275     BlockAIOCB *acb;
276 #endif
277 
278     /*
279      * We require at least one output segment each for the virtio_blk_outhdr
280      * and the SCSI command block.
281      *
282      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
283      * and the sense buffer pointer in the input segments.
284      */
285     if (elem->out_num < 2 || elem->in_num < 3) {
286         status = VIRTIO_BLK_S_IOERR;
287         goto fail;
288     }
289 
290     /*
291      * The scsi inhdr is placed in the second-to-last input segment, just
292      * before the regular inhdr.
293      */
294     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
295 
296     if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) {
297         status = VIRTIO_BLK_S_UNSUPP;
298         goto fail;
299     }
300 
301     /*
302      * No support for bidirection commands yet.
303      */
304     if (elem->out_num > 2 && elem->in_num > 3) {
305         status = VIRTIO_BLK_S_UNSUPP;
306         goto fail;
307     }
308 
309 #ifdef __linux__
310     ioctl_req = g_new0(VirtIOBlockIoctlReq, 1);
311     ioctl_req->req = req;
312     ioctl_req->hdr.interface_id = 'S';
313     ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len;
314     ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base;
315     ioctl_req->hdr.dxfer_len = 0;
316 
317     if (elem->out_num > 2) {
318         /*
319          * If there are more than the minimally required 2 output segments
320          * there is write payload starting from the third iovec.
321          */
322         ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV;
323         ioctl_req->hdr.iovec_count = elem->out_num - 2;
324 
325         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
326             ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
327         }
328 
329         ioctl_req->hdr.dxferp = elem->out_sg + 2;
330 
331     } else if (elem->in_num > 3) {
332         /*
333          * If we have more than 3 input segments the guest wants to actually
334          * read data.
335          */
336         ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV;
337         ioctl_req->hdr.iovec_count = elem->in_num - 3;
338         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
339             ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len;
340         }
341 
342         ioctl_req->hdr.dxferp = elem->in_sg;
343     } else {
344         /*
345          * Some SCSI commands don't actually transfer any data.
346          */
347         ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE;
348     }
349 
350     ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
351     ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
352 
353     acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr,
354                         virtio_blk_ioctl_complete, ioctl_req);
355     if (!acb) {
356         g_free(ioctl_req);
357         status = VIRTIO_BLK_S_UNSUPP;
358         goto fail;
359     }
360     return -EINPROGRESS;
361 #else
362     abort();
363 #endif
364 
365 fail:
366     /* Just put anything nonzero so that the ioctl fails in the guest.  */
367     if (scsi) {
368         virtio_stl_p(vdev, &scsi->errors, 255);
369     }
370     return status;
371 }
372 
373 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
374 {
375     int status;
376 
377     status = virtio_blk_handle_scsi_req(req);
378     if (status != -EINPROGRESS) {
379         virtio_blk_req_complete(req, status);
380         virtio_blk_free_request(req);
381     }
382 }
383 
384 static inline void submit_requests(BlockBackend *blk, MultiReqBuffer *mrb,
385                                    int start, int num_reqs, int niov)
386 {
387     QEMUIOVector *qiov = &mrb->reqs[start]->qiov;
388     int64_t sector_num = mrb->reqs[start]->sector_num;
389     bool is_write = mrb->is_write;
390 
391     if (num_reqs > 1) {
392         int i;
393         struct iovec *tmp_iov = qiov->iov;
394         int tmp_niov = qiov->niov;
395 
396         /* mrb->reqs[start]->qiov was initialized from external so we can't
397          * modify it here. We need to initialize it locally and then add the
398          * external iovecs. */
399         qemu_iovec_init(qiov, niov);
400 
401         for (i = 0; i < tmp_niov; i++) {
402             qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len);
403         }
404 
405         for (i = start + 1; i < start + num_reqs; i++) {
406             qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0,
407                               mrb->reqs[i]->qiov.size);
408             mrb->reqs[i - 1]->mr_next = mrb->reqs[i];
409         }
410 
411         trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev),
412                                          mrb, start, num_reqs,
413                                          sector_num << BDRV_SECTOR_BITS,
414                                          qiov->size, is_write);
415         block_acct_merge_done(blk_get_stats(blk),
416                               is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ,
417                               num_reqs - 1);
418     }
419 
420     if (is_write) {
421         blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
422                         virtio_blk_rw_complete, mrb->reqs[start]);
423     } else {
424         blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
425                        virtio_blk_rw_complete, mrb->reqs[start]);
426     }
427 }
428 
429 static int multireq_compare(const void *a, const void *b)
430 {
431     const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a,
432                          *req2 = *(VirtIOBlockReq **)b;
433 
434     /*
435      * Note that we can't simply subtract sector_num1 from sector_num2
436      * here as that could overflow the return value.
437      */
438     if (req1->sector_num > req2->sector_num) {
439         return 1;
440     } else if (req1->sector_num < req2->sector_num) {
441         return -1;
442     } else {
443         return 0;
444     }
445 }
446 
447 static void virtio_blk_submit_multireq(BlockBackend *blk, MultiReqBuffer *mrb)
448 {
449     int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0;
450     uint32_t max_transfer;
451     int64_t sector_num = 0;
452 
453     if (mrb->num_reqs == 1) {
454         submit_requests(blk, mrb, 0, 1, -1);
455         mrb->num_reqs = 0;
456         return;
457     }
458 
459     max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk);
460 
461     qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs),
462           &multireq_compare);
463 
464     for (i = 0; i < mrb->num_reqs; i++) {
465         VirtIOBlockReq *req = mrb->reqs[i];
466         if (num_reqs > 0) {
467             /*
468              * NOTE: We cannot merge the requests in below situations:
469              * 1. requests are not sequential
470              * 2. merge would exceed maximum number of IOVs
471              * 3. merge would exceed maximum transfer length of backend device
472              */
473             if (sector_num + nb_sectors != req->sector_num ||
474                 niov > blk_get_max_iov(blk) - req->qiov.niov ||
475                 req->qiov.size > max_transfer ||
476                 nb_sectors > (max_transfer -
477                               req->qiov.size) / BDRV_SECTOR_SIZE) {
478                 submit_requests(blk, mrb, start, num_reqs, niov);
479                 num_reqs = 0;
480             }
481         }
482 
483         if (num_reqs == 0) {
484             sector_num = req->sector_num;
485             nb_sectors = niov = 0;
486             start = i;
487         }
488 
489         nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE;
490         niov += req->qiov.niov;
491         num_reqs++;
492     }
493 
494     submit_requests(blk, mrb, start, num_reqs, niov);
495     mrb->num_reqs = 0;
496 }
497 
498 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
499 {
500     VirtIOBlock *s = req->dev;
501 
502     block_acct_start(blk_get_stats(s->blk), &req->acct, 0,
503                      BLOCK_ACCT_FLUSH);
504 
505     /*
506      * Make sure all outstanding writes are posted to the backing device.
507      */
508     if (mrb->is_write && mrb->num_reqs > 0) {
509         virtio_blk_submit_multireq(s->blk, mrb);
510     }
511     blk_aio_flush(s->blk, virtio_blk_flush_complete, req);
512 }
513 
514 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
515                                      uint64_t sector, size_t size)
516 {
517     uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
518     uint64_t total_sectors;
519 
520     if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
521         return false;
522     }
523     if (sector & dev->sector_mask) {
524         return false;
525     }
526     if (size % dev->conf.conf.logical_block_size) {
527         return false;
528     }
529     blk_get_geometry(dev->blk, &total_sectors);
530     if (sector > total_sectors || nb_sectors > total_sectors - sector) {
531         return false;
532     }
533     return true;
534 }
535 
536 static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req,
537     struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes)
538 {
539     VirtIOBlock *s = req->dev;
540     VirtIODevice *vdev = VIRTIO_DEVICE(s);
541     uint64_t sector;
542     uint32_t num_sectors, flags, max_sectors;
543     uint8_t err_status;
544     int bytes;
545 
546     sector = virtio_ldq_p(vdev, &dwz_hdr->sector);
547     num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors);
548     flags = virtio_ldl_p(vdev, &dwz_hdr->flags);
549     max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors :
550                   s->conf.max_discard_sectors;
551 
552     /*
553      * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check
554      * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in
555      * the integer variable.
556      */
557     if (unlikely(num_sectors > max_sectors)) {
558         err_status = VIRTIO_BLK_S_IOERR;
559         goto err;
560     }
561 
562     bytes = num_sectors << BDRV_SECTOR_BITS;
563 
564     if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) {
565         err_status = VIRTIO_BLK_S_IOERR;
566         goto err;
567     }
568 
569     /*
570      * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard
571      * and write zeroes commands if any unknown flag is set.
572      */
573     if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
574         err_status = VIRTIO_BLK_S_UNSUPP;
575         goto err;
576     }
577 
578     if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */
579         int blk_aio_flags = 0;
580 
581         if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) {
582             blk_aio_flags |= BDRV_REQ_MAY_UNMAP;
583         }
584 
585         block_acct_start(blk_get_stats(s->blk), &req->acct, bytes,
586                          BLOCK_ACCT_WRITE);
587 
588         blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS,
589                               bytes, blk_aio_flags,
590                               virtio_blk_discard_write_zeroes_complete, req);
591     } else { /* VIRTIO_BLK_T_DISCARD */
592         /*
593          * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for
594          * discard commands if the unmap flag is set.
595          */
596         if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
597             err_status = VIRTIO_BLK_S_UNSUPP;
598             goto err;
599         }
600 
601         blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes,
602                          virtio_blk_discard_write_zeroes_complete, req);
603     }
604 
605     return VIRTIO_BLK_S_OK;
606 
607 err:
608     if (is_write_zeroes) {
609         block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE);
610     }
611     return err_status;
612 }
613 
614 static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
615 {
616     uint32_t type;
617     struct iovec *in_iov = req->elem.in_sg;
618     struct iovec *out_iov = req->elem.out_sg;
619     unsigned in_num = req->elem.in_num;
620     unsigned out_num = req->elem.out_num;
621     VirtIOBlock *s = req->dev;
622     VirtIODevice *vdev = VIRTIO_DEVICE(s);
623 
624     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
625         virtio_error(vdev, "virtio-blk missing headers");
626         return -1;
627     }
628 
629     if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out,
630                             sizeof(req->out)) != sizeof(req->out))) {
631         virtio_error(vdev, "virtio-blk request outhdr too short");
632         return -1;
633     }
634 
635     iov_discard_front(&out_iov, &out_num, sizeof(req->out));
636 
637     if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
638         virtio_error(vdev, "virtio-blk request inhdr too short");
639         return -1;
640     }
641 
642     /* We always touch the last byte, so just see how big in_iov is.  */
643     req->in_len = iov_size(in_iov, in_num);
644     req->in = (void *)in_iov[in_num - 1].iov_base
645               + in_iov[in_num - 1].iov_len
646               - sizeof(struct virtio_blk_inhdr);
647     iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr));
648 
649     type = virtio_ldl_p(vdev, &req->out.type);
650 
651     /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
652      * is an optional flag. Although a guest should not send this flag if
653      * not negotiated we ignored it in the past. So keep ignoring it. */
654     switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) {
655     case VIRTIO_BLK_T_IN:
656     {
657         bool is_write = type & VIRTIO_BLK_T_OUT;
658         req->sector_num = virtio_ldq_p(vdev, &req->out.sector);
659 
660         if (is_write) {
661             qemu_iovec_init_external(&req->qiov, out_iov, out_num);
662             trace_virtio_blk_handle_write(vdev, req, req->sector_num,
663                                           req->qiov.size / BDRV_SECTOR_SIZE);
664         } else {
665             qemu_iovec_init_external(&req->qiov, in_iov, in_num);
666             trace_virtio_blk_handle_read(vdev, req, req->sector_num,
667                                          req->qiov.size / BDRV_SECTOR_SIZE);
668         }
669 
670         if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) {
671             virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
672             block_acct_invalid(blk_get_stats(s->blk),
673                                is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
674             virtio_blk_free_request(req);
675             return 0;
676         }
677 
678         block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size,
679                          is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
680 
681         /* merge would exceed maximum number of requests or IO direction
682          * changes */
683         if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS ||
684                                   is_write != mrb->is_write ||
685                                   !s->conf.request_merging)) {
686             virtio_blk_submit_multireq(s->blk, mrb);
687         }
688 
689         assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS);
690         mrb->reqs[mrb->num_reqs++] = req;
691         mrb->is_write = is_write;
692         break;
693     }
694     case VIRTIO_BLK_T_FLUSH:
695         virtio_blk_handle_flush(req, mrb);
696         break;
697     case VIRTIO_BLK_T_SCSI_CMD:
698         virtio_blk_handle_scsi(req);
699         break;
700     case VIRTIO_BLK_T_GET_ID:
701     {
702         /*
703          * NB: per existing s/n string convention the string is
704          * terminated by '\0' only when shorter than buffer.
705          */
706         const char *serial = s->conf.serial ? s->conf.serial : "";
707         size_t size = MIN(strlen(serial) + 1,
708                           MIN(iov_size(in_iov, in_num),
709                               VIRTIO_BLK_ID_BYTES));
710         iov_from_buf(in_iov, in_num, 0, serial, size);
711         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
712         virtio_blk_free_request(req);
713         break;
714     }
715     /*
716      * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with
717      * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement,
718      * so we must mask it for these requests, then we will check if it is set.
719      */
720     case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT:
721     case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT:
722     {
723         struct virtio_blk_discard_write_zeroes dwz_hdr;
724         size_t out_len = iov_size(out_iov, out_num);
725         bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) ==
726                                VIRTIO_BLK_T_WRITE_ZEROES;
727         uint8_t err_status;
728 
729         /*
730          * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains
731          * more than one segment.
732          */
733         if (unlikely(!(type & VIRTIO_BLK_T_OUT) ||
734                      out_len > sizeof(dwz_hdr))) {
735             virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
736             virtio_blk_free_request(req);
737             return 0;
738         }
739 
740         if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr,
741                                 sizeof(dwz_hdr)) != sizeof(dwz_hdr))) {
742             virtio_error(vdev, "virtio-blk discard/write_zeroes header"
743                          " too short");
744             return -1;
745         }
746 
747         err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr,
748                                                             is_write_zeroes);
749         if (err_status != VIRTIO_BLK_S_OK) {
750             virtio_blk_req_complete(req, err_status);
751             virtio_blk_free_request(req);
752         }
753 
754         break;
755     }
756     default:
757         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
758         virtio_blk_free_request(req);
759     }
760     return 0;
761 }
762 
763 bool virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
764 {
765     VirtIOBlockReq *req;
766     MultiReqBuffer mrb = {};
767     bool suppress_notifications = virtio_queue_get_notification(vq);
768     bool progress = false;
769 
770     aio_context_acquire(blk_get_aio_context(s->blk));
771     blk_io_plug(s->blk);
772 
773     do {
774         if (suppress_notifications) {
775             virtio_queue_set_notification(vq, 0);
776         }
777 
778         while ((req = virtio_blk_get_request(s, vq))) {
779             progress = true;
780             if (virtio_blk_handle_request(req, &mrb)) {
781                 virtqueue_detach_element(req->vq, &req->elem, 0);
782                 virtio_blk_free_request(req);
783                 break;
784             }
785         }
786 
787         if (suppress_notifications) {
788             virtio_queue_set_notification(vq, 1);
789         }
790     } while (!virtio_queue_empty(vq));
791 
792     if (mrb.num_reqs) {
793         virtio_blk_submit_multireq(s->blk, &mrb);
794     }
795 
796     blk_io_unplug(s->blk);
797     aio_context_release(blk_get_aio_context(s->blk));
798     return progress;
799 }
800 
801 static void virtio_blk_handle_output_do(VirtIOBlock *s, VirtQueue *vq)
802 {
803     virtio_blk_handle_vq(s, vq);
804 }
805 
806 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
807 {
808     VirtIOBlock *s = (VirtIOBlock *)vdev;
809 
810     if (s->dataplane) {
811         /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
812          * dataplane here instead of waiting for .set_status().
813          */
814         virtio_device_start_ioeventfd(vdev);
815         if (!s->dataplane_disabled) {
816             return;
817         }
818     }
819     virtio_blk_handle_output_do(s, vq);
820 }
821 
822 void virtio_blk_process_queued_requests(VirtIOBlock *s, bool is_bh)
823 {
824     VirtIOBlockReq *req = s->rq;
825     MultiReqBuffer mrb = {};
826 
827     s->rq = NULL;
828 
829     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
830     while (req) {
831         VirtIOBlockReq *next = req->next;
832         if (virtio_blk_handle_request(req, &mrb)) {
833             /* Device is now broken and won't do any processing until it gets
834              * reset. Already queued requests will be lost: let's purge them.
835              */
836             while (req) {
837                 next = req->next;
838                 virtqueue_detach_element(req->vq, &req->elem, 0);
839                 virtio_blk_free_request(req);
840                 req = next;
841             }
842             break;
843         }
844         req = next;
845     }
846 
847     if (mrb.num_reqs) {
848         virtio_blk_submit_multireq(s->blk, &mrb);
849     }
850     if (is_bh) {
851         blk_dec_in_flight(s->conf.conf.blk);
852     }
853     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
854 }
855 
856 static void virtio_blk_dma_restart_bh(void *opaque)
857 {
858     VirtIOBlock *s = opaque;
859 
860     qemu_bh_delete(s->bh);
861     s->bh = NULL;
862 
863     virtio_blk_process_queued_requests(s, true);
864 }
865 
866 static void virtio_blk_dma_restart_cb(void *opaque, int running,
867                                       RunState state)
868 {
869     VirtIOBlock *s = opaque;
870     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(s)));
871     VirtioBusState *bus = VIRTIO_BUS(qbus);
872 
873     if (!running) {
874         return;
875     }
876 
877     /*
878      * If ioeventfd is enabled, don't schedule the BH here as queued
879      * requests will be processed while starting the data plane.
880      */
881     if (!s->bh && !virtio_bus_ioeventfd_enabled(bus)) {
882         s->bh = aio_bh_new(blk_get_aio_context(s->conf.conf.blk),
883                            virtio_blk_dma_restart_bh, s);
884         blk_inc_in_flight(s->conf.conf.blk);
885         qemu_bh_schedule(s->bh);
886     }
887 }
888 
889 static void virtio_blk_reset(VirtIODevice *vdev)
890 {
891     VirtIOBlock *s = VIRTIO_BLK(vdev);
892     AioContext *ctx;
893     VirtIOBlockReq *req;
894 
895     ctx = blk_get_aio_context(s->blk);
896     aio_context_acquire(ctx);
897     blk_drain(s->blk);
898 
899     /* We drop queued requests after blk_drain() because blk_drain() itself can
900      * produce them. */
901     while (s->rq) {
902         req = s->rq;
903         s->rq = req->next;
904         virtqueue_detach_element(req->vq, &req->elem, 0);
905         virtio_blk_free_request(req);
906     }
907 
908     aio_context_release(ctx);
909 
910     assert(!s->dataplane_started);
911     blk_set_enable_write_cache(s->blk, s->original_wce);
912 }
913 
914 /* coalesce internal state, copy to pci i/o region 0
915  */
916 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
917 {
918     VirtIOBlock *s = VIRTIO_BLK(vdev);
919     BlockConf *conf = &s->conf.conf;
920     struct virtio_blk_config blkcfg;
921     uint64_t capacity;
922     int64_t length;
923     int blk_size = conf->logical_block_size;
924 
925     blk_get_geometry(s->blk, &capacity);
926     memset(&blkcfg, 0, sizeof(blkcfg));
927     virtio_stq_p(vdev, &blkcfg.capacity, capacity);
928     virtio_stl_p(vdev, &blkcfg.seg_max,
929                  s->conf.seg_max_adjust ? s->conf.queue_size - 2 : 128 - 2);
930     virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls);
931     virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
932     virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size);
933     virtio_stl_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size);
934     blkcfg.geometry.heads = conf->heads;
935     /*
936      * We must ensure that the block device capacity is a multiple of
937      * the logical block size. If that is not the case, let's use
938      * sector_mask to adopt the geometry to have a correct picture.
939      * For those devices where the capacity is ok for the given geometry
940      * we don't touch the sector value of the geometry, since some devices
941      * (like s390 dasd) need a specific value. Here the capacity is already
942      * cyls*heads*secs*blk_size and the sector value is not block size
943      * divided by 512 - instead it is the amount of blk_size blocks
944      * per track (cylinder).
945      */
946     length = blk_getlength(s->blk);
947     if (length > 0 && length / conf->heads / conf->secs % blk_size) {
948         blkcfg.geometry.sectors = conf->secs & ~s->sector_mask;
949     } else {
950         blkcfg.geometry.sectors = conf->secs;
951     }
952     blkcfg.size_max = 0;
953     blkcfg.physical_block_exp = get_physical_block_exp(conf);
954     blkcfg.alignment_offset = 0;
955     blkcfg.wce = blk_enable_write_cache(s->blk);
956     virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues);
957     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) {
958         virtio_stl_p(vdev, &blkcfg.max_discard_sectors,
959                      s->conf.max_discard_sectors);
960         virtio_stl_p(vdev, &blkcfg.discard_sector_alignment,
961                      blk_size >> BDRV_SECTOR_BITS);
962         /*
963          * We support only one segment per request since multiple segments
964          * are not widely used and there are no userspace APIs that allow
965          * applications to submit multiple segments in a single call.
966          */
967         virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1);
968     }
969     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) {
970         virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors,
971                      s->conf.max_write_zeroes_sectors);
972         blkcfg.write_zeroes_may_unmap = 1;
973         virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1);
974     }
975     memcpy(config, &blkcfg, s->config_size);
976 }
977 
978 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
979 {
980     VirtIOBlock *s = VIRTIO_BLK(vdev);
981     struct virtio_blk_config blkcfg;
982 
983     memcpy(&blkcfg, config, s->config_size);
984 
985     aio_context_acquire(blk_get_aio_context(s->blk));
986     blk_set_enable_write_cache(s->blk, blkcfg.wce != 0);
987     aio_context_release(blk_get_aio_context(s->blk));
988 }
989 
990 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features,
991                                         Error **errp)
992 {
993     VirtIOBlock *s = VIRTIO_BLK(vdev);
994 
995     /* Firstly sync all virtio-blk possible supported features */
996     features |= s->host_features;
997 
998     virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
999     virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
1000     virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
1001     virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
1002     if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) {
1003         if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) {
1004             error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
1005             return 0;
1006         }
1007     } else {
1008         virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT);
1009         virtio_add_feature(&features, VIRTIO_BLK_F_SCSI);
1010     }
1011 
1012     if (blk_enable_write_cache(s->blk) ||
1013         (s->conf.x_enable_wce_if_config_wce &&
1014          virtio_has_feature(features, VIRTIO_BLK_F_CONFIG_WCE))) {
1015         virtio_add_feature(&features, VIRTIO_BLK_F_WCE);
1016     }
1017     if (blk_is_read_only(s->blk)) {
1018         virtio_add_feature(&features, VIRTIO_BLK_F_RO);
1019     }
1020     if (s->conf.num_queues > 1) {
1021         virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
1022     }
1023 
1024     return features;
1025 }
1026 
1027 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
1028 {
1029     VirtIOBlock *s = VIRTIO_BLK(vdev);
1030 
1031     if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) {
1032         assert(!s->dataplane_started);
1033     }
1034 
1035     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1036         return;
1037     }
1038 
1039     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
1040      * cache flushes.  Thus, the "auto writethrough" behavior is never
1041      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
1042      * Leaving it enabled would break the following sequence:
1043      *
1044      *     Guest started with "-drive cache=writethrough"
1045      *     Guest sets status to 0
1046      *     Guest sets DRIVER bit in status field
1047      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
1048      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
1049      *     Guest writes 1 to the WCE configuration field (writeback mode)
1050      *     Guest sets DRIVER_OK bit in status field
1051      *
1052      * s->blk would erroneously be placed in writethrough mode.
1053      */
1054     if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) {
1055         aio_context_acquire(blk_get_aio_context(s->blk));
1056         blk_set_enable_write_cache(s->blk,
1057                                    virtio_vdev_has_feature(vdev,
1058                                                            VIRTIO_BLK_F_WCE));
1059         aio_context_release(blk_get_aio_context(s->blk));
1060     }
1061 }
1062 
1063 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
1064 {
1065     VirtIOBlock *s = VIRTIO_BLK(vdev);
1066     VirtIOBlockReq *req = s->rq;
1067 
1068     while (req) {
1069         qemu_put_sbyte(f, 1);
1070 
1071         if (s->conf.num_queues > 1) {
1072             qemu_put_be32(f, virtio_get_queue_index(req->vq));
1073         }
1074 
1075         qemu_put_virtqueue_element(vdev, f, &req->elem);
1076         req = req->next;
1077     }
1078     qemu_put_sbyte(f, 0);
1079 }
1080 
1081 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
1082                                   int version_id)
1083 {
1084     VirtIOBlock *s = VIRTIO_BLK(vdev);
1085 
1086     while (qemu_get_sbyte(f)) {
1087         unsigned nvqs = s->conf.num_queues;
1088         unsigned vq_idx = 0;
1089         VirtIOBlockReq *req;
1090 
1091         if (nvqs > 1) {
1092             vq_idx = qemu_get_be32(f);
1093 
1094             if (vq_idx >= nvqs) {
1095                 error_report("Invalid virtqueue index in request list: %#x",
1096                              vq_idx);
1097                 return -EINVAL;
1098             }
1099         }
1100 
1101         req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq));
1102         virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req);
1103         req->next = s->rq;
1104         s->rq = req;
1105     }
1106 
1107     return 0;
1108 }
1109 
1110 static void virtio_resize_cb(void *opaque)
1111 {
1112     VirtIODevice *vdev = opaque;
1113 
1114     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1115     virtio_notify_config(vdev);
1116 }
1117 
1118 static void virtio_blk_resize(void *opaque)
1119 {
1120     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
1121 
1122     /*
1123      * virtio_notify_config() needs to acquire the global mutex,
1124      * so it can't be called from an iothread. Instead, schedule
1125      * it to be run in the main context BH.
1126      */
1127     aio_bh_schedule_oneshot(qemu_get_aio_context(), virtio_resize_cb, vdev);
1128 }
1129 
1130 static const BlockDevOps virtio_block_ops = {
1131     .resize_cb = virtio_blk_resize,
1132 };
1133 
1134 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
1135 {
1136     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1137     VirtIOBlock *s = VIRTIO_BLK(dev);
1138     VirtIOBlkConf *conf = &s->conf;
1139     Error *err = NULL;
1140     unsigned i;
1141 
1142     if (!conf->conf.blk) {
1143         error_setg(errp, "drive property not set");
1144         return;
1145     }
1146     if (!blk_is_inserted(conf->conf.blk)) {
1147         error_setg(errp, "Device needs media, but drive is empty");
1148         return;
1149     }
1150     if (!conf->num_queues) {
1151         error_setg(errp, "num-queues property must be larger than 0");
1152         return;
1153     }
1154     if (conf->queue_size <= 2) {
1155         error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1156                    "must be > 2", conf->queue_size);
1157         return;
1158     }
1159     if (!is_power_of_2(conf->queue_size) ||
1160         conf->queue_size > VIRTQUEUE_MAX_SIZE) {
1161         error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1162                    "must be a power of 2 (max %d)",
1163                    conf->queue_size, VIRTQUEUE_MAX_SIZE);
1164         return;
1165     }
1166 
1167     if (!blkconf_apply_backend_options(&conf->conf,
1168                                        blk_is_read_only(conf->conf.blk), true,
1169                                        errp)) {
1170         return;
1171     }
1172     s->original_wce = blk_enable_write_cache(conf->conf.blk);
1173     if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) {
1174         return;
1175     }
1176 
1177     if (!blkconf_blocksizes(&conf->conf, errp)) {
1178         return;
1179     }
1180 
1181     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) &&
1182         (!conf->max_discard_sectors ||
1183          conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1184         error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")"
1185                    ", must be between 1 and %d",
1186                    conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS);
1187         return;
1188     }
1189 
1190     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) &&
1191         (!conf->max_write_zeroes_sectors ||
1192          conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1193         error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32
1194                    "), must be between 1 and %d",
1195                    conf->max_write_zeroes_sectors,
1196                    (int)BDRV_REQUEST_MAX_SECTORS);
1197         return;
1198     }
1199 
1200     virtio_blk_set_config_size(s, s->host_features);
1201 
1202     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK, s->config_size);
1203 
1204     s->blk = conf->conf.blk;
1205     s->rq = NULL;
1206     s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1;
1207 
1208     for (i = 0; i < conf->num_queues; i++) {
1209         virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output);
1210     }
1211     virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err);
1212     if (err != NULL) {
1213         error_propagate(errp, err);
1214         for (i = 0; i < conf->num_queues; i++) {
1215             virtio_del_queue(vdev, i);
1216         }
1217         virtio_cleanup(vdev);
1218         return;
1219     }
1220 
1221     s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
1222     blk_set_dev_ops(s->blk, &virtio_block_ops, s);
1223     blk_set_guest_block_size(s->blk, s->conf.conf.logical_block_size);
1224 
1225     blk_iostatus_enable(s->blk);
1226 
1227     add_boot_device_lchs(dev, "/disk@0,0",
1228                          conf->conf.lcyls,
1229                          conf->conf.lheads,
1230                          conf->conf.lsecs);
1231 }
1232 
1233 static void virtio_blk_device_unrealize(DeviceState *dev)
1234 {
1235     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1236     VirtIOBlock *s = VIRTIO_BLK(dev);
1237     VirtIOBlkConf *conf = &s->conf;
1238     unsigned i;
1239 
1240     blk_drain(s->blk);
1241     del_boot_device_lchs(dev, "/disk@0,0");
1242     virtio_blk_data_plane_destroy(s->dataplane);
1243     s->dataplane = NULL;
1244     for (i = 0; i < conf->num_queues; i++) {
1245         virtio_del_queue(vdev, i);
1246     }
1247     qemu_del_vm_change_state_handler(s->change);
1248     blockdev_mark_auto_del(s->blk);
1249     virtio_cleanup(vdev);
1250 }
1251 
1252 static void virtio_blk_instance_init(Object *obj)
1253 {
1254     VirtIOBlock *s = VIRTIO_BLK(obj);
1255 
1256     device_add_bootindex_property(obj, &s->conf.conf.bootindex,
1257                                   "bootindex", "/disk@0,0",
1258                                   DEVICE(obj));
1259 }
1260 
1261 static const VMStateDescription vmstate_virtio_blk = {
1262     .name = "virtio-blk",
1263     .minimum_version_id = 2,
1264     .version_id = 2,
1265     .fields = (VMStateField[]) {
1266         VMSTATE_VIRTIO_DEVICE,
1267         VMSTATE_END_OF_LIST()
1268     },
1269 };
1270 
1271 static Property virtio_blk_properties[] = {
1272     DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf),
1273     DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf),
1274     DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf),
1275     DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial),
1276     DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features,
1277                       VIRTIO_BLK_F_CONFIG_WCE, true),
1278 #ifdef __linux__
1279     DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features,
1280                       VIRTIO_BLK_F_SCSI, false),
1281 #endif
1282     DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
1283                     true),
1284     DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 1),
1285     DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 256),
1286     DEFINE_PROP_BOOL("seg-max-adjust", VirtIOBlock, conf.seg_max_adjust, true),
1287     DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD,
1288                      IOThread *),
1289     DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features,
1290                       VIRTIO_BLK_F_DISCARD, true),
1291     DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features,
1292                       VIRTIO_BLK_F_WRITE_ZEROES, true),
1293     DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock,
1294                        conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS),
1295     DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock,
1296                        conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS),
1297     DEFINE_PROP_BOOL("x-enable-wce-if-config-wce", VirtIOBlock,
1298                      conf.x_enable_wce_if_config_wce, true),
1299     DEFINE_PROP_END_OF_LIST(),
1300 };
1301 
1302 static void virtio_blk_class_init(ObjectClass *klass, void *data)
1303 {
1304     DeviceClass *dc = DEVICE_CLASS(klass);
1305     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1306 
1307     device_class_set_props(dc, virtio_blk_properties);
1308     dc->vmsd = &vmstate_virtio_blk;
1309     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1310     vdc->realize = virtio_blk_device_realize;
1311     vdc->unrealize = virtio_blk_device_unrealize;
1312     vdc->get_config = virtio_blk_update_config;
1313     vdc->set_config = virtio_blk_set_config;
1314     vdc->get_features = virtio_blk_get_features;
1315     vdc->set_status = virtio_blk_set_status;
1316     vdc->reset = virtio_blk_reset;
1317     vdc->save = virtio_blk_save_device;
1318     vdc->load = virtio_blk_load_device;
1319     vdc->start_ioeventfd = virtio_blk_data_plane_start;
1320     vdc->stop_ioeventfd = virtio_blk_data_plane_stop;
1321 }
1322 
1323 static const TypeInfo virtio_blk_info = {
1324     .name = TYPE_VIRTIO_BLK,
1325     .parent = TYPE_VIRTIO_DEVICE,
1326     .instance_size = sizeof(VirtIOBlock),
1327     .instance_init = virtio_blk_instance_init,
1328     .class_init = virtio_blk_class_init,
1329 };
1330 
1331 static void virtio_register_types(void)
1332 {
1333     type_register_static(&virtio_blk_info);
1334 }
1335 
1336 type_init(virtio_register_types)
1337