xref: /openbmc/qemu/hw/block/virtio-blk.c (revision 93e0932b)
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/block-ram-registrar.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/runstate.h"
27 #include "hw/virtio/virtio-blk.h"
28 #include "dataplane/virtio-blk.h"
29 #include "scsi/constants.h"
30 #ifdef __linux__
31 # include <scsi/sg.h>
32 #endif
33 #include "hw/virtio/virtio-bus.h"
34 #include "migration/qemu-file-types.h"
35 #include "hw/virtio/virtio-access.h"
36 #include "hw/virtio/virtio-blk-common.h"
37 #include "qemu/coroutine.h"
38 
39 static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq,
40                                     VirtIOBlockReq *req)
41 {
42     req->dev = s;
43     req->vq = vq;
44     req->qiov.size = 0;
45     req->in_len = 0;
46     req->next = NULL;
47     req->mr_next = NULL;
48 }
49 
50 static void virtio_blk_free_request(VirtIOBlockReq *req)
51 {
52     g_free(req);
53 }
54 
55 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
56 {
57     VirtIOBlock *s = req->dev;
58     VirtIODevice *vdev = VIRTIO_DEVICE(s);
59 
60     trace_virtio_blk_req_complete(vdev, req, status);
61 
62     stb_p(&req->in->status, status);
63     iov_discard_undo(&req->inhdr_undo);
64     iov_discard_undo(&req->outhdr_undo);
65     virtqueue_push(req->vq, &req->elem, req->in_len);
66     if (s->dataplane_started && !s->dataplane_disabled) {
67         virtio_blk_data_plane_notify(s->dataplane, req->vq);
68     } else {
69         virtio_notify(vdev, req->vq);
70     }
71 }
72 
73 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
74     bool is_read, bool acct_failed)
75 {
76     VirtIOBlock *s = req->dev;
77     BlockErrorAction action = blk_get_error_action(s->blk, is_read, error);
78 
79     if (action == BLOCK_ERROR_ACTION_STOP) {
80         /* Break the link as the next request is going to be parsed from the
81          * ring again. Otherwise we may end up doing a double completion! */
82         req->mr_next = NULL;
83         req->next = s->rq;
84         s->rq = req;
85     } else if (action == BLOCK_ERROR_ACTION_REPORT) {
86         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
87         if (acct_failed) {
88             block_acct_failed(blk_get_stats(s->blk), &req->acct);
89         }
90         virtio_blk_free_request(req);
91     }
92 
93     blk_error_action(s->blk, action, is_read, error);
94     return action != BLOCK_ERROR_ACTION_IGNORE;
95 }
96 
97 static void virtio_blk_rw_complete(void *opaque, int ret)
98 {
99     VirtIOBlockReq *next = opaque;
100     VirtIOBlock *s = next->dev;
101     VirtIODevice *vdev = VIRTIO_DEVICE(s);
102 
103     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
104     while (next) {
105         VirtIOBlockReq *req = next;
106         next = req->mr_next;
107         trace_virtio_blk_rw_complete(vdev, req, ret);
108 
109         if (req->qiov.nalloc != -1) {
110             /* If nalloc is != -1 req->qiov is a local copy of the original
111              * external iovec. It was allocated in submit_requests to be
112              * able to merge requests. */
113             qemu_iovec_destroy(&req->qiov);
114         }
115 
116         if (ret) {
117             int p = virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type);
118             bool is_read = !(p & VIRTIO_BLK_T_OUT);
119             /* Note that memory may be dirtied on read failure.  If the
120              * virtio request is not completed here, as is the case for
121              * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
122              * correctly during live migration.  While this is ugly,
123              * it is acceptable because the device is free to write to
124              * the memory until the request is completed (which will
125              * happen on the other side of the migration).
126              */
127             if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) {
128                 continue;
129             }
130         }
131 
132         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
133         block_acct_done(blk_get_stats(s->blk), &req->acct);
134         virtio_blk_free_request(req);
135     }
136     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
137 }
138 
139 static void virtio_blk_flush_complete(void *opaque, int ret)
140 {
141     VirtIOBlockReq *req = opaque;
142     VirtIOBlock *s = req->dev;
143 
144     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
145     if (ret) {
146         if (virtio_blk_handle_rw_error(req, -ret, 0, true)) {
147             goto out;
148         }
149     }
150 
151     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
152     block_acct_done(blk_get_stats(s->blk), &req->acct);
153     virtio_blk_free_request(req);
154 
155 out:
156     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
157 }
158 
159 static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret)
160 {
161     VirtIOBlockReq *req = opaque;
162     VirtIOBlock *s = req->dev;
163     bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) &
164                             ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES;
165 
166     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
167     if (ret) {
168         if (virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) {
169             goto out;
170         }
171     }
172 
173     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
174     if (is_write_zeroes) {
175         block_acct_done(blk_get_stats(s->blk), &req->acct);
176     }
177     virtio_blk_free_request(req);
178 
179 out:
180     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
181 }
182 
183 #ifdef __linux__
184 
185 typedef struct {
186     VirtIOBlockReq *req;
187     struct sg_io_hdr hdr;
188 } VirtIOBlockIoctlReq;
189 
190 static void virtio_blk_ioctl_complete(void *opaque, int status)
191 {
192     VirtIOBlockIoctlReq *ioctl_req = opaque;
193     VirtIOBlockReq *req = ioctl_req->req;
194     VirtIOBlock *s = req->dev;
195     VirtIODevice *vdev = VIRTIO_DEVICE(s);
196     struct virtio_scsi_inhdr *scsi;
197     struct sg_io_hdr *hdr;
198 
199     scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
200 
201     if (status) {
202         status = VIRTIO_BLK_S_UNSUPP;
203         virtio_stl_p(vdev, &scsi->errors, 255);
204         goto out;
205     }
206 
207     hdr = &ioctl_req->hdr;
208     /*
209      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
210      * clear the masked_status field [hence status gets cleared too, see
211      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
212      * status has occurred.  However they do set DRIVER_SENSE in driver_status
213      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
214      */
215     if (hdr->status == 0 && hdr->sb_len_wr > 0) {
216         hdr->status = CHECK_CONDITION;
217     }
218 
219     virtio_stl_p(vdev, &scsi->errors,
220                  hdr->status | (hdr->msg_status << 8) |
221                  (hdr->host_status << 16) | (hdr->driver_status << 24));
222     virtio_stl_p(vdev, &scsi->residual, hdr->resid);
223     virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr);
224     virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
225 
226 out:
227     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
228     virtio_blk_req_complete(req, status);
229     virtio_blk_free_request(req);
230     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
231     g_free(ioctl_req);
232 }
233 
234 #endif
235 
236 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq)
237 {
238     VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq));
239 
240     if (req) {
241         virtio_blk_init_request(s, vq, req);
242     }
243     return req;
244 }
245 
246 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req)
247 {
248     int status = VIRTIO_BLK_S_OK;
249     struct virtio_scsi_inhdr *scsi = NULL;
250     VirtIOBlock *blk = req->dev;
251     VirtIODevice *vdev = VIRTIO_DEVICE(blk);
252     VirtQueueElement *elem = &req->elem;
253 
254 #ifdef __linux__
255     int i;
256     VirtIOBlockIoctlReq *ioctl_req;
257     BlockAIOCB *acb;
258 #endif
259 
260     /*
261      * We require at least one output segment each for the virtio_blk_outhdr
262      * and the SCSI command block.
263      *
264      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
265      * and the sense buffer pointer in the input segments.
266      */
267     if (elem->out_num < 2 || elem->in_num < 3) {
268         status = VIRTIO_BLK_S_IOERR;
269         goto fail;
270     }
271 
272     /*
273      * The scsi inhdr is placed in the second-to-last input segment, just
274      * before the regular inhdr.
275      */
276     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
277 
278     if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) {
279         status = VIRTIO_BLK_S_UNSUPP;
280         goto fail;
281     }
282 
283     /*
284      * No support for bidirection commands yet.
285      */
286     if (elem->out_num > 2 && elem->in_num > 3) {
287         status = VIRTIO_BLK_S_UNSUPP;
288         goto fail;
289     }
290 
291 #ifdef __linux__
292     ioctl_req = g_new0(VirtIOBlockIoctlReq, 1);
293     ioctl_req->req = req;
294     ioctl_req->hdr.interface_id = 'S';
295     ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len;
296     ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base;
297     ioctl_req->hdr.dxfer_len = 0;
298 
299     if (elem->out_num > 2) {
300         /*
301          * If there are more than the minimally required 2 output segments
302          * there is write payload starting from the third iovec.
303          */
304         ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV;
305         ioctl_req->hdr.iovec_count = elem->out_num - 2;
306 
307         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
308             ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
309         }
310 
311         ioctl_req->hdr.dxferp = elem->out_sg + 2;
312 
313     } else if (elem->in_num > 3) {
314         /*
315          * If we have more than 3 input segments the guest wants to actually
316          * read data.
317          */
318         ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV;
319         ioctl_req->hdr.iovec_count = elem->in_num - 3;
320         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
321             ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len;
322         }
323 
324         ioctl_req->hdr.dxferp = elem->in_sg;
325     } else {
326         /*
327          * Some SCSI commands don't actually transfer any data.
328          */
329         ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE;
330     }
331 
332     ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
333     ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
334 
335     acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr,
336                         virtio_blk_ioctl_complete, ioctl_req);
337     if (!acb) {
338         g_free(ioctl_req);
339         status = VIRTIO_BLK_S_UNSUPP;
340         goto fail;
341     }
342     return -EINPROGRESS;
343 #else
344     abort();
345 #endif
346 
347 fail:
348     /* Just put anything nonzero so that the ioctl fails in the guest.  */
349     if (scsi) {
350         virtio_stl_p(vdev, &scsi->errors, 255);
351     }
352     return status;
353 }
354 
355 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
356 {
357     int status;
358 
359     status = virtio_blk_handle_scsi_req(req);
360     if (status != -EINPROGRESS) {
361         virtio_blk_req_complete(req, status);
362         virtio_blk_free_request(req);
363     }
364 }
365 
366 static inline void submit_requests(VirtIOBlock *s, MultiReqBuffer *mrb,
367                                    int start, int num_reqs, int niov)
368 {
369     BlockBackend *blk = s->blk;
370     QEMUIOVector *qiov = &mrb->reqs[start]->qiov;
371     int64_t sector_num = mrb->reqs[start]->sector_num;
372     bool is_write = mrb->is_write;
373     BdrvRequestFlags flags = 0;
374 
375     if (num_reqs > 1) {
376         int i;
377         struct iovec *tmp_iov = qiov->iov;
378         int tmp_niov = qiov->niov;
379 
380         /* mrb->reqs[start]->qiov was initialized from external so we can't
381          * modify it here. We need to initialize it locally and then add the
382          * external iovecs. */
383         qemu_iovec_init(qiov, niov);
384 
385         for (i = 0; i < tmp_niov; i++) {
386             qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len);
387         }
388 
389         for (i = start + 1; i < start + num_reqs; i++) {
390             qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0,
391                               mrb->reqs[i]->qiov.size);
392             mrb->reqs[i - 1]->mr_next = mrb->reqs[i];
393         }
394 
395         trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev),
396                                          mrb, start, num_reqs,
397                                          sector_num << BDRV_SECTOR_BITS,
398                                          qiov->size, is_write);
399         block_acct_merge_done(blk_get_stats(blk),
400                               is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ,
401                               num_reqs - 1);
402     }
403 
404     if (blk_ram_registrar_ok(&s->blk_ram_registrar)) {
405         flags |= BDRV_REQ_REGISTERED_BUF;
406     }
407 
408     if (is_write) {
409         blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov,
410                         flags, virtio_blk_rw_complete,
411                         mrb->reqs[start]);
412     } else {
413         blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov,
414                        flags, virtio_blk_rw_complete,
415                        mrb->reqs[start]);
416     }
417 }
418 
419 static int multireq_compare(const void *a, const void *b)
420 {
421     const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a,
422                          *req2 = *(VirtIOBlockReq **)b;
423 
424     /*
425      * Note that we can't simply subtract sector_num1 from sector_num2
426      * here as that could overflow the return value.
427      */
428     if (req1->sector_num > req2->sector_num) {
429         return 1;
430     } else if (req1->sector_num < req2->sector_num) {
431         return -1;
432     } else {
433         return 0;
434     }
435 }
436 
437 static void virtio_blk_submit_multireq(VirtIOBlock *s, MultiReqBuffer *mrb)
438 {
439     int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0;
440     uint32_t max_transfer;
441     int64_t sector_num = 0;
442 
443     if (mrb->num_reqs == 1) {
444         submit_requests(s, mrb, 0, 1, -1);
445         mrb->num_reqs = 0;
446         return;
447     }
448 
449     max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk);
450 
451     qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs),
452           &multireq_compare);
453 
454     for (i = 0; i < mrb->num_reqs; i++) {
455         VirtIOBlockReq *req = mrb->reqs[i];
456         if (num_reqs > 0) {
457             /*
458              * NOTE: We cannot merge the requests in below situations:
459              * 1. requests are not sequential
460              * 2. merge would exceed maximum number of IOVs
461              * 3. merge would exceed maximum transfer length of backend device
462              */
463             if (sector_num + nb_sectors != req->sector_num ||
464                 niov > blk_get_max_iov(s->blk) - req->qiov.niov ||
465                 req->qiov.size > max_transfer ||
466                 nb_sectors > (max_transfer -
467                               req->qiov.size) / BDRV_SECTOR_SIZE) {
468                 submit_requests(s, mrb, start, num_reqs, niov);
469                 num_reqs = 0;
470             }
471         }
472 
473         if (num_reqs == 0) {
474             sector_num = req->sector_num;
475             nb_sectors = niov = 0;
476             start = i;
477         }
478 
479         nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE;
480         niov += req->qiov.niov;
481         num_reqs++;
482     }
483 
484     submit_requests(s, mrb, start, num_reqs, niov);
485     mrb->num_reqs = 0;
486 }
487 
488 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
489 {
490     VirtIOBlock *s = req->dev;
491 
492     block_acct_start(blk_get_stats(s->blk), &req->acct, 0,
493                      BLOCK_ACCT_FLUSH);
494 
495     /*
496      * Make sure all outstanding writes are posted to the backing device.
497      */
498     if (mrb->is_write && mrb->num_reqs > 0) {
499         virtio_blk_submit_multireq(s, mrb);
500     }
501     blk_aio_flush(s->blk, virtio_blk_flush_complete, req);
502 }
503 
504 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
505                                      uint64_t sector, size_t size)
506 {
507     uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
508     uint64_t total_sectors;
509 
510     if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
511         return false;
512     }
513     if (sector & dev->sector_mask) {
514         return false;
515     }
516     if (size % dev->conf.conf.logical_block_size) {
517         return false;
518     }
519     blk_get_geometry(dev->blk, &total_sectors);
520     if (sector > total_sectors || nb_sectors > total_sectors - sector) {
521         return false;
522     }
523     return true;
524 }
525 
526 static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req,
527     struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes)
528 {
529     VirtIOBlock *s = req->dev;
530     VirtIODevice *vdev = VIRTIO_DEVICE(s);
531     uint64_t sector;
532     uint32_t num_sectors, flags, max_sectors;
533     uint8_t err_status;
534     int bytes;
535 
536     sector = virtio_ldq_p(vdev, &dwz_hdr->sector);
537     num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors);
538     flags = virtio_ldl_p(vdev, &dwz_hdr->flags);
539     max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors :
540                   s->conf.max_discard_sectors;
541 
542     /*
543      * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check
544      * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in
545      * the integer variable.
546      */
547     if (unlikely(num_sectors > max_sectors)) {
548         err_status = VIRTIO_BLK_S_IOERR;
549         goto err;
550     }
551 
552     bytes = num_sectors << BDRV_SECTOR_BITS;
553 
554     if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) {
555         err_status = VIRTIO_BLK_S_IOERR;
556         goto err;
557     }
558 
559     /*
560      * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard
561      * and write zeroes commands if any unknown flag is set.
562      */
563     if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
564         err_status = VIRTIO_BLK_S_UNSUPP;
565         goto err;
566     }
567 
568     if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */
569         int blk_aio_flags = 0;
570 
571         if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) {
572             blk_aio_flags |= BDRV_REQ_MAY_UNMAP;
573         }
574 
575         block_acct_start(blk_get_stats(s->blk), &req->acct, bytes,
576                          BLOCK_ACCT_WRITE);
577 
578         blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS,
579                               bytes, blk_aio_flags,
580                               virtio_blk_discard_write_zeroes_complete, req);
581     } else { /* VIRTIO_BLK_T_DISCARD */
582         /*
583          * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for
584          * discard commands if the unmap flag is set.
585          */
586         if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
587             err_status = VIRTIO_BLK_S_UNSUPP;
588             goto err;
589         }
590 
591         blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes,
592                          virtio_blk_discard_write_zeroes_complete, req);
593     }
594 
595     return VIRTIO_BLK_S_OK;
596 
597 err:
598     if (is_write_zeroes) {
599         block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE);
600     }
601     return err_status;
602 }
603 
604 static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
605 {
606     uint32_t type;
607     struct iovec *in_iov = req->elem.in_sg;
608     struct iovec *out_iov = req->elem.out_sg;
609     unsigned in_num = req->elem.in_num;
610     unsigned out_num = req->elem.out_num;
611     VirtIOBlock *s = req->dev;
612     VirtIODevice *vdev = VIRTIO_DEVICE(s);
613 
614     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
615         virtio_error(vdev, "virtio-blk missing headers");
616         return -1;
617     }
618 
619     if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out,
620                             sizeof(req->out)) != sizeof(req->out))) {
621         virtio_error(vdev, "virtio-blk request outhdr too short");
622         return -1;
623     }
624 
625     iov_discard_front_undoable(&out_iov, &out_num, sizeof(req->out),
626                                &req->outhdr_undo);
627 
628     if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
629         virtio_error(vdev, "virtio-blk request inhdr too short");
630         iov_discard_undo(&req->outhdr_undo);
631         return -1;
632     }
633 
634     /* We always touch the last byte, so just see how big in_iov is.  */
635     req->in_len = iov_size(in_iov, in_num);
636     req->in = (void *)in_iov[in_num - 1].iov_base
637               + in_iov[in_num - 1].iov_len
638               - sizeof(struct virtio_blk_inhdr);
639     iov_discard_back_undoable(in_iov, &in_num, sizeof(struct virtio_blk_inhdr),
640                               &req->inhdr_undo);
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, 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             iov_discard_undo(&req->inhdr_undo);
736             iov_discard_undo(&req->outhdr_undo);
737             virtio_error(vdev, "virtio-blk discard/write_zeroes header"
738                          " too short");
739             return -1;
740         }
741 
742         err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr,
743                                                             is_write_zeroes);
744         if (err_status != VIRTIO_BLK_S_OK) {
745             virtio_blk_req_complete(req, err_status);
746             virtio_blk_free_request(req);
747         }
748 
749         break;
750     }
751     default:
752         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
753         virtio_blk_free_request(req);
754     }
755     return 0;
756 }
757 
758 void virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
759 {
760     VirtIOBlockReq *req;
761     MultiReqBuffer mrb = {};
762     bool suppress_notifications = virtio_queue_get_notification(vq);
763 
764     aio_context_acquire(blk_get_aio_context(s->blk));
765     blk_io_plug(s->blk);
766 
767     do {
768         if (suppress_notifications) {
769             virtio_queue_set_notification(vq, 0);
770         }
771 
772         while ((req = virtio_blk_get_request(s, vq))) {
773             if (virtio_blk_handle_request(req, &mrb)) {
774                 virtqueue_detach_element(req->vq, &req->elem, 0);
775                 virtio_blk_free_request(req);
776                 break;
777             }
778         }
779 
780         if (suppress_notifications) {
781             virtio_queue_set_notification(vq, 1);
782         }
783     } while (!virtio_queue_empty(vq));
784 
785     if (mrb.num_reqs) {
786         virtio_blk_submit_multireq(s, &mrb);
787     }
788 
789     blk_io_unplug(s->blk);
790     aio_context_release(blk_get_aio_context(s->blk));
791 }
792 
793 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
794 {
795     VirtIOBlock *s = (VirtIOBlock *)vdev;
796 
797     if (s->dataplane && !s->dataplane_started) {
798         /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
799          * dataplane here instead of waiting for .set_status().
800          */
801         virtio_device_start_ioeventfd(vdev);
802         if (!s->dataplane_disabled) {
803             return;
804         }
805     }
806     virtio_blk_handle_vq(s, vq);
807 }
808 
809 static void virtio_blk_dma_restart_bh(void *opaque)
810 {
811     VirtIOBlock *s = opaque;
812 
813     VirtIOBlockReq *req = s->rq;
814     MultiReqBuffer mrb = {};
815 
816     s->rq = NULL;
817 
818     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
819     while (req) {
820         VirtIOBlockReq *next = req->next;
821         if (virtio_blk_handle_request(req, &mrb)) {
822             /* Device is now broken and won't do any processing until it gets
823              * reset. Already queued requests will be lost: let's purge them.
824              */
825             while (req) {
826                 next = req->next;
827                 virtqueue_detach_element(req->vq, &req->elem, 0);
828                 virtio_blk_free_request(req);
829                 req = next;
830             }
831             break;
832         }
833         req = next;
834     }
835 
836     if (mrb.num_reqs) {
837         virtio_blk_submit_multireq(s, &mrb);
838     }
839 
840     /* Paired with inc in virtio_blk_dma_restart_cb() */
841     blk_dec_in_flight(s->conf.conf.blk);
842 
843     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
844 }
845 
846 static void virtio_blk_dma_restart_cb(void *opaque, bool running,
847                                       RunState state)
848 {
849     VirtIOBlock *s = opaque;
850 
851     if (!running) {
852         return;
853     }
854 
855     /* Paired with dec in virtio_blk_dma_restart_bh() */
856     blk_inc_in_flight(s->conf.conf.blk);
857 
858     aio_bh_schedule_oneshot(blk_get_aio_context(s->conf.conf.blk),
859             virtio_blk_dma_restart_bh, s);
860 }
861 
862 static void virtio_blk_reset(VirtIODevice *vdev)
863 {
864     VirtIOBlock *s = VIRTIO_BLK(vdev);
865     AioContext *ctx;
866     VirtIOBlockReq *req;
867 
868     ctx = blk_get_aio_context(s->blk);
869     aio_context_acquire(ctx);
870     blk_drain(s->blk);
871 
872     /* We drop queued requests after blk_drain() because blk_drain() itself can
873      * produce them. */
874     while (s->rq) {
875         req = s->rq;
876         s->rq = req->next;
877         virtqueue_detach_element(req->vq, &req->elem, 0);
878         virtio_blk_free_request(req);
879     }
880 
881     aio_context_release(ctx);
882 
883     assert(!s->dataplane_started);
884     blk_set_enable_write_cache(s->blk, s->original_wce);
885 }
886 
887 /* coalesce internal state, copy to pci i/o region 0
888  */
889 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
890 {
891     VirtIOBlock *s = VIRTIO_BLK(vdev);
892     BlockConf *conf = &s->conf.conf;
893     struct virtio_blk_config blkcfg;
894     uint64_t capacity;
895     int64_t length;
896     int blk_size = conf->logical_block_size;
897     AioContext *ctx;
898 
899     ctx = blk_get_aio_context(s->blk);
900     aio_context_acquire(ctx);
901 
902     blk_get_geometry(s->blk, &capacity);
903     memset(&blkcfg, 0, sizeof(blkcfg));
904     virtio_stq_p(vdev, &blkcfg.capacity, capacity);
905     virtio_stl_p(vdev, &blkcfg.seg_max,
906                  s->conf.seg_max_adjust ? s->conf.queue_size - 2 : 128 - 2);
907     virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls);
908     virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
909     virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size);
910     virtio_stl_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size);
911     blkcfg.geometry.heads = conf->heads;
912     /*
913      * We must ensure that the block device capacity is a multiple of
914      * the logical block size. If that is not the case, let's use
915      * sector_mask to adopt the geometry to have a correct picture.
916      * For those devices where the capacity is ok for the given geometry
917      * we don't touch the sector value of the geometry, since some devices
918      * (like s390 dasd) need a specific value. Here the capacity is already
919      * cyls*heads*secs*blk_size and the sector value is not block size
920      * divided by 512 - instead it is the amount of blk_size blocks
921      * per track (cylinder).
922      */
923     length = blk_getlength(s->blk);
924     aio_context_release(ctx);
925     if (length > 0 && length / conf->heads / conf->secs % blk_size) {
926         blkcfg.geometry.sectors = conf->secs & ~s->sector_mask;
927     } else {
928         blkcfg.geometry.sectors = conf->secs;
929     }
930     blkcfg.size_max = 0;
931     blkcfg.physical_block_exp = get_physical_block_exp(conf);
932     blkcfg.alignment_offset = 0;
933     blkcfg.wce = blk_enable_write_cache(s->blk);
934     virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues);
935     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) {
936         uint32_t discard_granularity = conf->discard_granularity;
937         if (discard_granularity == -1 || !s->conf.report_discard_granularity) {
938             discard_granularity = blk_size;
939         }
940         virtio_stl_p(vdev, &blkcfg.max_discard_sectors,
941                      s->conf.max_discard_sectors);
942         virtio_stl_p(vdev, &blkcfg.discard_sector_alignment,
943                      discard_granularity >> BDRV_SECTOR_BITS);
944         /*
945          * We support only one segment per request since multiple segments
946          * are not widely used and there are no userspace APIs that allow
947          * applications to submit multiple segments in a single call.
948          */
949         virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1);
950     }
951     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) {
952         virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors,
953                      s->conf.max_write_zeroes_sectors);
954         blkcfg.write_zeroes_may_unmap = 1;
955         virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1);
956     }
957     memcpy(config, &blkcfg, s->config_size);
958 }
959 
960 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
961 {
962     VirtIOBlock *s = VIRTIO_BLK(vdev);
963     struct virtio_blk_config blkcfg;
964 
965     memcpy(&blkcfg, config, s->config_size);
966 
967     aio_context_acquire(blk_get_aio_context(s->blk));
968     blk_set_enable_write_cache(s->blk, blkcfg.wce != 0);
969     aio_context_release(blk_get_aio_context(s->blk));
970 }
971 
972 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features,
973                                         Error **errp)
974 {
975     VirtIOBlock *s = VIRTIO_BLK(vdev);
976 
977     /* Firstly sync all virtio-blk possible supported features */
978     features |= s->host_features;
979 
980     virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
981     virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
982     virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
983     virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
984     if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) {
985         if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) {
986             error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
987             return 0;
988         }
989     } else {
990         virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT);
991         virtio_add_feature(&features, VIRTIO_BLK_F_SCSI);
992     }
993 
994     if (blk_enable_write_cache(s->blk) ||
995         (s->conf.x_enable_wce_if_config_wce &&
996          virtio_has_feature(features, VIRTIO_BLK_F_CONFIG_WCE))) {
997         virtio_add_feature(&features, VIRTIO_BLK_F_WCE);
998     }
999     if (!blk_is_writable(s->blk)) {
1000         virtio_add_feature(&features, VIRTIO_BLK_F_RO);
1001     }
1002     if (s->conf.num_queues > 1) {
1003         virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
1004     }
1005 
1006     return features;
1007 }
1008 
1009 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
1010 {
1011     VirtIOBlock *s = VIRTIO_BLK(vdev);
1012 
1013     if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) {
1014         assert(!s->dataplane_started);
1015     }
1016 
1017     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1018         return;
1019     }
1020 
1021     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
1022      * cache flushes.  Thus, the "auto writethrough" behavior is never
1023      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
1024      * Leaving it enabled would break the following sequence:
1025      *
1026      *     Guest started with "-drive cache=writethrough"
1027      *     Guest sets status to 0
1028      *     Guest sets DRIVER bit in status field
1029      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
1030      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
1031      *     Guest writes 1 to the WCE configuration field (writeback mode)
1032      *     Guest sets DRIVER_OK bit in status field
1033      *
1034      * s->blk would erroneously be placed in writethrough mode.
1035      */
1036     if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) {
1037         aio_context_acquire(blk_get_aio_context(s->blk));
1038         blk_set_enable_write_cache(s->blk,
1039                                    virtio_vdev_has_feature(vdev,
1040                                                            VIRTIO_BLK_F_WCE));
1041         aio_context_release(blk_get_aio_context(s->blk));
1042     }
1043 }
1044 
1045 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
1046 {
1047     VirtIOBlock *s = VIRTIO_BLK(vdev);
1048     VirtIOBlockReq *req = s->rq;
1049 
1050     while (req) {
1051         qemu_put_sbyte(f, 1);
1052 
1053         if (s->conf.num_queues > 1) {
1054             qemu_put_be32(f, virtio_get_queue_index(req->vq));
1055         }
1056 
1057         qemu_put_virtqueue_element(vdev, f, &req->elem);
1058         req = req->next;
1059     }
1060     qemu_put_sbyte(f, 0);
1061 }
1062 
1063 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
1064                                   int version_id)
1065 {
1066     VirtIOBlock *s = VIRTIO_BLK(vdev);
1067 
1068     while (qemu_get_sbyte(f)) {
1069         unsigned nvqs = s->conf.num_queues;
1070         unsigned vq_idx = 0;
1071         VirtIOBlockReq *req;
1072 
1073         if (nvqs > 1) {
1074             vq_idx = qemu_get_be32(f);
1075 
1076             if (vq_idx >= nvqs) {
1077                 error_report("Invalid virtqueue index in request list: %#x",
1078                              vq_idx);
1079                 return -EINVAL;
1080             }
1081         }
1082 
1083         req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq));
1084         virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req);
1085         req->next = s->rq;
1086         s->rq = req;
1087     }
1088 
1089     return 0;
1090 }
1091 
1092 static void virtio_resize_cb(void *opaque)
1093 {
1094     VirtIODevice *vdev = opaque;
1095 
1096     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1097     virtio_notify_config(vdev);
1098 }
1099 
1100 static void virtio_blk_resize(void *opaque)
1101 {
1102     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
1103 
1104     /*
1105      * virtio_notify_config() needs to acquire the global mutex,
1106      * so it can't be called from an iothread. Instead, schedule
1107      * it to be run in the main context BH.
1108      */
1109     aio_bh_schedule_oneshot(qemu_get_aio_context(), virtio_resize_cb, vdev);
1110 }
1111 
1112 static const BlockDevOps virtio_block_ops = {
1113     .resize_cb = virtio_blk_resize,
1114 };
1115 
1116 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
1117 {
1118     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1119     VirtIOBlock *s = VIRTIO_BLK(dev);
1120     VirtIOBlkConf *conf = &s->conf;
1121     Error *err = NULL;
1122     unsigned i;
1123 
1124     if (!conf->conf.blk) {
1125         error_setg(errp, "drive property not set");
1126         return;
1127     }
1128     if (!blk_is_inserted(conf->conf.blk)) {
1129         error_setg(errp, "Device needs media, but drive is empty");
1130         return;
1131     }
1132     if (conf->num_queues == VIRTIO_BLK_AUTO_NUM_QUEUES) {
1133         conf->num_queues = 1;
1134     }
1135     if (!conf->num_queues) {
1136         error_setg(errp, "num-queues property must be larger than 0");
1137         return;
1138     }
1139     if (conf->queue_size <= 2) {
1140         error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1141                    "must be > 2", conf->queue_size);
1142         return;
1143     }
1144     if (!is_power_of_2(conf->queue_size) ||
1145         conf->queue_size > VIRTQUEUE_MAX_SIZE) {
1146         error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1147                    "must be a power of 2 (max %d)",
1148                    conf->queue_size, VIRTQUEUE_MAX_SIZE);
1149         return;
1150     }
1151 
1152     if (!blkconf_apply_backend_options(&conf->conf,
1153                                        !blk_supports_write_perm(conf->conf.blk),
1154                                        true, errp)) {
1155         return;
1156     }
1157     s->original_wce = blk_enable_write_cache(conf->conf.blk);
1158     if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) {
1159         return;
1160     }
1161 
1162     if (!blkconf_blocksizes(&conf->conf, errp)) {
1163         return;
1164     }
1165 
1166     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) &&
1167         (!conf->max_discard_sectors ||
1168          conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1169         error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")"
1170                    ", must be between 1 and %d",
1171                    conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS);
1172         return;
1173     }
1174 
1175     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) &&
1176         (!conf->max_write_zeroes_sectors ||
1177          conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1178         error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32
1179                    "), must be between 1 and %d",
1180                    conf->max_write_zeroes_sectors,
1181                    (int)BDRV_REQUEST_MAX_SECTORS);
1182         return;
1183     }
1184 
1185     s->config_size = virtio_get_config_size(&virtio_blk_cfg_size_params,
1186                                             s->host_features);
1187     virtio_init(vdev, VIRTIO_ID_BLOCK, s->config_size);
1188 
1189     s->blk = conf->conf.blk;
1190     s->rq = NULL;
1191     s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1;
1192 
1193     for (i = 0; i < conf->num_queues; i++) {
1194         virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output);
1195     }
1196     qemu_coroutine_inc_pool_size(conf->num_queues * conf->queue_size / 2);
1197     virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err);
1198     if (err != NULL) {
1199         error_propagate(errp, err);
1200         for (i = 0; i < conf->num_queues; i++) {
1201             virtio_del_queue(vdev, i);
1202         }
1203         virtio_cleanup(vdev);
1204         return;
1205     }
1206 
1207     /*
1208      * This must be after virtio_init() so virtio_blk_dma_restart_cb() gets
1209      * called after ->start_ioeventfd() has already set blk's AioContext.
1210      */
1211     s->change =
1212         qdev_add_vm_change_state_handler(dev, virtio_blk_dma_restart_cb, s);
1213 
1214     blk_ram_registrar_init(&s->blk_ram_registrar, s->blk);
1215     blk_set_dev_ops(s->blk, &virtio_block_ops, s);
1216 
1217     blk_iostatus_enable(s->blk);
1218 
1219     add_boot_device_lchs(dev, "/disk@0,0",
1220                          conf->conf.lcyls,
1221                          conf->conf.lheads,
1222                          conf->conf.lsecs);
1223 }
1224 
1225 static void virtio_blk_device_unrealize(DeviceState *dev)
1226 {
1227     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1228     VirtIOBlock *s = VIRTIO_BLK(dev);
1229     VirtIOBlkConf *conf = &s->conf;
1230     unsigned i;
1231 
1232     blk_drain(s->blk);
1233     del_boot_device_lchs(dev, "/disk@0,0");
1234     virtio_blk_data_plane_destroy(s->dataplane);
1235     s->dataplane = NULL;
1236     for (i = 0; i < conf->num_queues; i++) {
1237         virtio_del_queue(vdev, i);
1238     }
1239     qemu_coroutine_dec_pool_size(conf->num_queues * conf->queue_size / 2);
1240     blk_ram_registrar_destroy(&s->blk_ram_registrar);
1241     qemu_del_vm_change_state_handler(s->change);
1242     blockdev_mark_auto_del(s->blk);
1243     virtio_cleanup(vdev);
1244 }
1245 
1246 static void virtio_blk_instance_init(Object *obj)
1247 {
1248     VirtIOBlock *s = VIRTIO_BLK(obj);
1249 
1250     device_add_bootindex_property(obj, &s->conf.conf.bootindex,
1251                                   "bootindex", "/disk@0,0",
1252                                   DEVICE(obj));
1253 }
1254 
1255 static const VMStateDescription vmstate_virtio_blk = {
1256     .name = "virtio-blk",
1257     .minimum_version_id = 2,
1258     .version_id = 2,
1259     .fields = (VMStateField[]) {
1260         VMSTATE_VIRTIO_DEVICE,
1261         VMSTATE_END_OF_LIST()
1262     },
1263 };
1264 
1265 static Property virtio_blk_properties[] = {
1266     DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf),
1267     DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf),
1268     DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf),
1269     DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial),
1270     DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features,
1271                       VIRTIO_BLK_F_CONFIG_WCE, true),
1272 #ifdef __linux__
1273     DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features,
1274                       VIRTIO_BLK_F_SCSI, false),
1275 #endif
1276     DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
1277                     true),
1278     DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues,
1279                        VIRTIO_BLK_AUTO_NUM_QUEUES),
1280     DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 256),
1281     DEFINE_PROP_BOOL("seg-max-adjust", VirtIOBlock, conf.seg_max_adjust, true),
1282     DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD,
1283                      IOThread *),
1284     DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features,
1285                       VIRTIO_BLK_F_DISCARD, true),
1286     DEFINE_PROP_BOOL("report-discard-granularity", VirtIOBlock,
1287                      conf.report_discard_granularity, true),
1288     DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features,
1289                       VIRTIO_BLK_F_WRITE_ZEROES, true),
1290     DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock,
1291                        conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS),
1292     DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock,
1293                        conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS),
1294     DEFINE_PROP_BOOL("x-enable-wce-if-config-wce", VirtIOBlock,
1295                      conf.x_enable_wce_if_config_wce, true),
1296     DEFINE_PROP_END_OF_LIST(),
1297 };
1298 
1299 static void virtio_blk_class_init(ObjectClass *klass, void *data)
1300 {
1301     DeviceClass *dc = DEVICE_CLASS(klass);
1302     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1303 
1304     device_class_set_props(dc, virtio_blk_properties);
1305     dc->vmsd = &vmstate_virtio_blk;
1306     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1307     vdc->realize = virtio_blk_device_realize;
1308     vdc->unrealize = virtio_blk_device_unrealize;
1309     vdc->get_config = virtio_blk_update_config;
1310     vdc->set_config = virtio_blk_set_config;
1311     vdc->get_features = virtio_blk_get_features;
1312     vdc->set_status = virtio_blk_set_status;
1313     vdc->reset = virtio_blk_reset;
1314     vdc->save = virtio_blk_save_device;
1315     vdc->load = virtio_blk_load_device;
1316     vdc->start_ioeventfd = virtio_blk_data_plane_start;
1317     vdc->stop_ioeventfd = virtio_blk_data_plane_stop;
1318 }
1319 
1320 static const TypeInfo virtio_blk_info = {
1321     .name = TYPE_VIRTIO_BLK,
1322     .parent = TYPE_VIRTIO_DEVICE,
1323     .instance_size = sizeof(VirtIOBlock),
1324     .instance_init = virtio_blk_instance_init,
1325     .class_init = virtio_blk_class_init,
1326 };
1327 
1328 static void virtio_register_types(void)
1329 {
1330     type_register_static(&virtio_blk_info);
1331 }
1332 
1333 type_init(virtio_register_types)
1334