xref: /openbmc/qemu/hw/block/virtio-blk.c (revision be355a44)
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 "qemu/defer-call.h"
16 #include "qapi/error.h"
17 #include "qemu/iov.h"
18 #include "qemu/module.h"
19 #include "qemu/error-report.h"
20 #include "qemu/main-loop.h"
21 #include "block/block_int.h"
22 #include "trace.h"
23 #include "hw/block/block.h"
24 #include "hw/qdev-properties.h"
25 #include "sysemu/blockdev.h"
26 #include "sysemu/block-ram-registrar.h"
27 #include "sysemu/sysemu.h"
28 #include "sysemu/runstate.h"
29 #include "hw/virtio/virtio-blk.h"
30 #include "scsi/constants.h"
31 #ifdef __linux__
32 # include <scsi/sg.h>
33 #endif
34 #include "hw/virtio/virtio-bus.h"
35 #include "migration/qemu-file-types.h"
36 #include "hw/virtio/virtio-access.h"
37 #include "hw/virtio/virtio-blk-common.h"
38 #include "qemu/coroutine.h"
39 
40 static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq,
41                                     VirtIOBlockReq *req)
42 {
43     req->dev = s;
44     req->vq = vq;
45     req->qiov.size = 0;
46     req->in_len = 0;
47     req->next = NULL;
48     req->mr_next = NULL;
49 }
50 
51 static void virtio_blk_free_request(VirtIOBlockReq *req)
52 {
53     g_free(req);
54 }
55 
56 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
57 {
58     VirtIOBlock *s = req->dev;
59     VirtIODevice *vdev = VIRTIO_DEVICE(s);
60 
61     trace_virtio_blk_req_complete(vdev, req, status);
62 
63     stb_p(&req->in->status, status);
64     iov_discard_undo(&req->inhdr_undo);
65     iov_discard_undo(&req->outhdr_undo);
66     virtqueue_push(req->vq, &req->elem, req->in_len);
67     if (s->ioeventfd_started && !s->ioeventfd_disabled) {
68         virtio_notify_irqfd(vdev, req->vq);
69     } else {
70         virtio_notify(vdev, req->vq);
71     }
72 }
73 
74 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
75     bool is_read, bool acct_failed)
76 {
77     VirtIOBlock *s = req->dev;
78     BlockErrorAction action = blk_get_error_action(s->blk, is_read, error);
79 
80     if (action == BLOCK_ERROR_ACTION_STOP) {
81         /* Break the link as the next request is going to be parsed from the
82          * ring again. Otherwise we may end up doing a double completion! */
83         req->mr_next = NULL;
84 
85         WITH_QEMU_LOCK_GUARD(&s->rq_lock) {
86             req->next = s->rq;
87             s->rq = req;
88         }
89     } else if (action == BLOCK_ERROR_ACTION_REPORT) {
90         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
91         if (acct_failed) {
92             block_acct_failed(blk_get_stats(s->blk), &req->acct);
93         }
94         virtio_blk_free_request(req);
95     }
96 
97     blk_error_action(s->blk, action, is_read, error);
98     return action != BLOCK_ERROR_ACTION_IGNORE;
99 }
100 
101 static void virtio_blk_rw_complete(void *opaque, int ret)
102 {
103     VirtIOBlockReq *next = opaque;
104     VirtIOBlock *s = next->dev;
105     VirtIODevice *vdev = VIRTIO_DEVICE(s);
106 
107     while (next) {
108         VirtIOBlockReq *req = next;
109         next = req->mr_next;
110         trace_virtio_blk_rw_complete(vdev, req, ret);
111 
112         if (req->qiov.nalloc != -1) {
113             /* If nalloc is != -1 req->qiov is a local copy of the original
114              * external iovec. It was allocated in submit_requests to be
115              * able to merge requests. */
116             qemu_iovec_destroy(&req->qiov);
117         }
118 
119         if (ret) {
120             int p = virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type);
121             bool is_read = !(p & VIRTIO_BLK_T_OUT);
122             /* Note that memory may be dirtied on read failure.  If the
123              * virtio request is not completed here, as is the case for
124              * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
125              * correctly during live migration.  While this is ugly,
126              * it is acceptable because the device is free to write to
127              * the memory until the request is completed (which will
128              * happen on the other side of the migration).
129              */
130             if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) {
131                 continue;
132             }
133         }
134 
135         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
136         block_acct_done(blk_get_stats(s->blk), &req->acct);
137         virtio_blk_free_request(req);
138     }
139 }
140 
141 static void virtio_blk_flush_complete(void *opaque, int ret)
142 {
143     VirtIOBlockReq *req = opaque;
144     VirtIOBlock *s = req->dev;
145 
146     if (ret && virtio_blk_handle_rw_error(req, -ret, 0, true)) {
147         return;
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 
155 static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret)
156 {
157     VirtIOBlockReq *req = opaque;
158     VirtIOBlock *s = req->dev;
159     bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) &
160                             ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES;
161 
162     if (ret && virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) {
163         return;
164     }
165 
166     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
167     if (is_write_zeroes) {
168         block_acct_done(blk_get_stats(s->blk), &req->acct);
169     }
170     virtio_blk_free_request(req);
171 }
172 
173 #ifdef __linux__
174 
175 typedef struct {
176     VirtIOBlockReq *req;
177     struct sg_io_hdr hdr;
178 } VirtIOBlockIoctlReq;
179 
180 static void virtio_blk_ioctl_complete(void *opaque, int status)
181 {
182     VirtIOBlockIoctlReq *ioctl_req = opaque;
183     VirtIOBlockReq *req = ioctl_req->req;
184     VirtIOBlock *s = req->dev;
185     VirtIODevice *vdev = VIRTIO_DEVICE(s);
186     struct virtio_scsi_inhdr *scsi;
187     struct sg_io_hdr *hdr;
188 
189     scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
190 
191     if (status) {
192         status = VIRTIO_BLK_S_UNSUPP;
193         virtio_stl_p(vdev, &scsi->errors, 255);
194         goto out;
195     }
196 
197     hdr = &ioctl_req->hdr;
198     /*
199      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
200      * clear the masked_status field [hence status gets cleared too, see
201      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
202      * status has occurred.  However they do set DRIVER_SENSE in driver_status
203      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
204      */
205     if (hdr->status == 0 && hdr->sb_len_wr > 0) {
206         hdr->status = CHECK_CONDITION;
207     }
208 
209     virtio_stl_p(vdev, &scsi->errors,
210                  hdr->status | (hdr->msg_status << 8) |
211                  (hdr->host_status << 16) | (hdr->driver_status << 24));
212     virtio_stl_p(vdev, &scsi->residual, hdr->resid);
213     virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr);
214     virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
215 
216 out:
217     virtio_blk_req_complete(req, status);
218     virtio_blk_free_request(req);
219     g_free(ioctl_req);
220 }
221 
222 #endif
223 
224 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq)
225 {
226     VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq));
227 
228     if (req) {
229         virtio_blk_init_request(s, vq, req);
230     }
231     return req;
232 }
233 
234 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req)
235 {
236     int status = VIRTIO_BLK_S_OK;
237     struct virtio_scsi_inhdr *scsi = NULL;
238     VirtIOBlock *blk = req->dev;
239     VirtIODevice *vdev = VIRTIO_DEVICE(blk);
240     VirtQueueElement *elem = &req->elem;
241 
242 #ifdef __linux__
243     int i;
244     VirtIOBlockIoctlReq *ioctl_req;
245     BlockAIOCB *acb;
246 #endif
247 
248     /*
249      * We require at least one output segment each for the virtio_blk_outhdr
250      * and the SCSI command block.
251      *
252      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
253      * and the sense buffer pointer in the input segments.
254      */
255     if (elem->out_num < 2 || elem->in_num < 3) {
256         status = VIRTIO_BLK_S_IOERR;
257         goto fail;
258     }
259 
260     /*
261      * The scsi inhdr is placed in the second-to-last input segment, just
262      * before the regular inhdr.
263      */
264     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
265 
266     if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) {
267         status = VIRTIO_BLK_S_UNSUPP;
268         goto fail;
269     }
270 
271     /*
272      * No support for bidirection commands yet.
273      */
274     if (elem->out_num > 2 && elem->in_num > 3) {
275         status = VIRTIO_BLK_S_UNSUPP;
276         goto fail;
277     }
278 
279 #ifdef __linux__
280     ioctl_req = g_new0(VirtIOBlockIoctlReq, 1);
281     ioctl_req->req = req;
282     ioctl_req->hdr.interface_id = 'S';
283     ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len;
284     ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base;
285     ioctl_req->hdr.dxfer_len = 0;
286 
287     if (elem->out_num > 2) {
288         /*
289          * If there are more than the minimally required 2 output segments
290          * there is write payload starting from the third iovec.
291          */
292         ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV;
293         ioctl_req->hdr.iovec_count = elem->out_num - 2;
294 
295         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
296             ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
297         }
298 
299         ioctl_req->hdr.dxferp = elem->out_sg + 2;
300 
301     } else if (elem->in_num > 3) {
302         /*
303          * If we have more than 3 input segments the guest wants to actually
304          * read data.
305          */
306         ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV;
307         ioctl_req->hdr.iovec_count = elem->in_num - 3;
308         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
309             ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len;
310         }
311 
312         ioctl_req->hdr.dxferp = elem->in_sg;
313     } else {
314         /*
315          * Some SCSI commands don't actually transfer any data.
316          */
317         ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE;
318     }
319 
320     ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
321     ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
322 
323     acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr,
324                         virtio_blk_ioctl_complete, ioctl_req);
325     if (!acb) {
326         g_free(ioctl_req);
327         status = VIRTIO_BLK_S_UNSUPP;
328         goto fail;
329     }
330     return -EINPROGRESS;
331 #else
332     abort();
333 #endif
334 
335 fail:
336     /* Just put anything nonzero so that the ioctl fails in the guest.  */
337     if (scsi) {
338         virtio_stl_p(vdev, &scsi->errors, 255);
339     }
340     return status;
341 }
342 
343 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
344 {
345     int status;
346 
347     status = virtio_blk_handle_scsi_req(req);
348     if (status != -EINPROGRESS) {
349         virtio_blk_req_complete(req, status);
350         virtio_blk_free_request(req);
351     }
352 }
353 
354 static inline void submit_requests(VirtIOBlock *s, MultiReqBuffer *mrb,
355                                    int start, int num_reqs, int niov)
356 {
357     BlockBackend *blk = s->blk;
358     QEMUIOVector *qiov = &mrb->reqs[start]->qiov;
359     int64_t sector_num = mrb->reqs[start]->sector_num;
360     bool is_write = mrb->is_write;
361     BdrvRequestFlags flags = 0;
362 
363     if (num_reqs > 1) {
364         int i;
365         struct iovec *tmp_iov = qiov->iov;
366         int tmp_niov = qiov->niov;
367 
368         /* mrb->reqs[start]->qiov was initialized from external so we can't
369          * modify it here. We need to initialize it locally and then add the
370          * external iovecs. */
371         qemu_iovec_init(qiov, niov);
372 
373         for (i = 0; i < tmp_niov; i++) {
374             qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len);
375         }
376 
377         for (i = start + 1; i < start + num_reqs; i++) {
378             qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0,
379                               mrb->reqs[i]->qiov.size);
380             mrb->reqs[i - 1]->mr_next = mrb->reqs[i];
381         }
382 
383         trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev),
384                                          mrb, start, num_reqs,
385                                          sector_num << BDRV_SECTOR_BITS,
386                                          qiov->size, is_write);
387         block_acct_merge_done(blk_get_stats(blk),
388                               is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ,
389                               num_reqs - 1);
390     }
391 
392     if (blk_ram_registrar_ok(&s->blk_ram_registrar)) {
393         flags |= BDRV_REQ_REGISTERED_BUF;
394     }
395 
396     if (is_write) {
397         blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov,
398                         flags, virtio_blk_rw_complete,
399                         mrb->reqs[start]);
400     } else {
401         blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov,
402                        flags, virtio_blk_rw_complete,
403                        mrb->reqs[start]);
404     }
405 }
406 
407 static int multireq_compare(const void *a, const void *b)
408 {
409     const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a,
410                          *req2 = *(VirtIOBlockReq **)b;
411 
412     /*
413      * Note that we can't simply subtract sector_num1 from sector_num2
414      * here as that could overflow the return value.
415      */
416     if (req1->sector_num > req2->sector_num) {
417         return 1;
418     } else if (req1->sector_num < req2->sector_num) {
419         return -1;
420     } else {
421         return 0;
422     }
423 }
424 
425 static void virtio_blk_submit_multireq(VirtIOBlock *s, MultiReqBuffer *mrb)
426 {
427     int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0;
428     uint32_t max_transfer;
429     int64_t sector_num = 0;
430 
431     if (mrb->num_reqs == 1) {
432         submit_requests(s, mrb, 0, 1, -1);
433         mrb->num_reqs = 0;
434         return;
435     }
436 
437     max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk);
438 
439     qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs),
440           &multireq_compare);
441 
442     for (i = 0; i < mrb->num_reqs; i++) {
443         VirtIOBlockReq *req = mrb->reqs[i];
444         if (num_reqs > 0) {
445             /*
446              * NOTE: We cannot merge the requests in below situations:
447              * 1. requests are not sequential
448              * 2. merge would exceed maximum number of IOVs
449              * 3. merge would exceed maximum transfer length of backend device
450              */
451             if (sector_num + nb_sectors != req->sector_num ||
452                 niov > blk_get_max_iov(s->blk) - req->qiov.niov ||
453                 req->qiov.size > max_transfer ||
454                 nb_sectors > (max_transfer -
455                               req->qiov.size) / BDRV_SECTOR_SIZE) {
456                 submit_requests(s, mrb, start, num_reqs, niov);
457                 num_reqs = 0;
458             }
459         }
460 
461         if (num_reqs == 0) {
462             sector_num = req->sector_num;
463             nb_sectors = niov = 0;
464             start = i;
465         }
466 
467         nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE;
468         niov += req->qiov.niov;
469         num_reqs++;
470     }
471 
472     submit_requests(s, mrb, start, num_reqs, niov);
473     mrb->num_reqs = 0;
474 }
475 
476 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
477 {
478     VirtIOBlock *s = req->dev;
479 
480     block_acct_start(blk_get_stats(s->blk), &req->acct, 0,
481                      BLOCK_ACCT_FLUSH);
482 
483     /*
484      * Make sure all outstanding writes are posted to the backing device.
485      */
486     if (mrb->is_write && mrb->num_reqs > 0) {
487         virtio_blk_submit_multireq(s, mrb);
488     }
489     blk_aio_flush(s->blk, virtio_blk_flush_complete, req);
490 }
491 
492 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
493                                      uint64_t sector, size_t size)
494 {
495     uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
496     uint64_t total_sectors;
497 
498     if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
499         return false;
500     }
501     if (sector & dev->sector_mask) {
502         return false;
503     }
504     if (size % dev->conf.conf.logical_block_size) {
505         return false;
506     }
507     blk_get_geometry(dev->blk, &total_sectors);
508     if (sector > total_sectors || nb_sectors > total_sectors - sector) {
509         return false;
510     }
511     return true;
512 }
513 
514 static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req,
515     struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes)
516 {
517     VirtIOBlock *s = req->dev;
518     VirtIODevice *vdev = VIRTIO_DEVICE(s);
519     uint64_t sector;
520     uint32_t num_sectors, flags, max_sectors;
521     uint8_t err_status;
522     int bytes;
523 
524     sector = virtio_ldq_p(vdev, &dwz_hdr->sector);
525     num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors);
526     flags = virtio_ldl_p(vdev, &dwz_hdr->flags);
527     max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors :
528                   s->conf.max_discard_sectors;
529 
530     /*
531      * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check
532      * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in
533      * the integer variable.
534      */
535     if (unlikely(num_sectors > max_sectors)) {
536         err_status = VIRTIO_BLK_S_IOERR;
537         goto err;
538     }
539 
540     bytes = num_sectors << BDRV_SECTOR_BITS;
541 
542     if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) {
543         err_status = VIRTIO_BLK_S_IOERR;
544         goto err;
545     }
546 
547     /*
548      * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard
549      * and write zeroes commands if any unknown flag is set.
550      */
551     if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
552         err_status = VIRTIO_BLK_S_UNSUPP;
553         goto err;
554     }
555 
556     if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */
557         int blk_aio_flags = 0;
558 
559         if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) {
560             blk_aio_flags |= BDRV_REQ_MAY_UNMAP;
561         }
562 
563         block_acct_start(blk_get_stats(s->blk), &req->acct, bytes,
564                          BLOCK_ACCT_WRITE);
565 
566         blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS,
567                               bytes, blk_aio_flags,
568                               virtio_blk_discard_write_zeroes_complete, req);
569     } else { /* VIRTIO_BLK_T_DISCARD */
570         /*
571          * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for
572          * discard commands if the unmap flag is set.
573          */
574         if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) {
575             err_status = VIRTIO_BLK_S_UNSUPP;
576             goto err;
577         }
578 
579         blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes,
580                          virtio_blk_discard_write_zeroes_complete, req);
581     }
582 
583     return VIRTIO_BLK_S_OK;
584 
585 err:
586     if (is_write_zeroes) {
587         block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE);
588     }
589     return err_status;
590 }
591 
592 typedef struct ZoneCmdData {
593     VirtIOBlockReq *req;
594     struct iovec *in_iov;
595     unsigned in_num;
596     union {
597         struct {
598             unsigned int nr_zones;
599             BlockZoneDescriptor *zones;
600         } zone_report_data;
601         struct {
602             int64_t offset;
603         } zone_append_data;
604     };
605 } ZoneCmdData;
606 
607 /*
608  * check zoned_request: error checking before issuing requests. If all checks
609  * passed, return true.
610  * append: true if only zone append requests issued.
611  */
612 static bool check_zoned_request(VirtIOBlock *s, int64_t offset, int64_t len,
613                              bool append, uint8_t *status) {
614     BlockDriverState *bs = blk_bs(s->blk);
615     int index;
616 
617     if (!virtio_has_feature(s->host_features, VIRTIO_BLK_F_ZONED)) {
618         *status = VIRTIO_BLK_S_UNSUPP;
619         return false;
620     }
621 
622     if (offset < 0 || len < 0 || len > (bs->total_sectors << BDRV_SECTOR_BITS)
623         || offset > (bs->total_sectors << BDRV_SECTOR_BITS) - len) {
624         *status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
625         return false;
626     }
627 
628     if (append) {
629         if (bs->bl.write_granularity) {
630             if ((offset % bs->bl.write_granularity) != 0) {
631                 *status = VIRTIO_BLK_S_ZONE_UNALIGNED_WP;
632                 return false;
633             }
634         }
635 
636         index = offset / bs->bl.zone_size;
637         if (BDRV_ZT_IS_CONV(bs->wps->wp[index])) {
638             *status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
639             return false;
640         }
641 
642         if (len / 512 > bs->bl.max_append_sectors) {
643             if (bs->bl.max_append_sectors == 0) {
644                 *status = VIRTIO_BLK_S_UNSUPP;
645             } else {
646                 *status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
647             }
648             return false;
649         }
650     }
651     return true;
652 }
653 
654 static void virtio_blk_zone_report_complete(void *opaque, int ret)
655 {
656     ZoneCmdData *data = opaque;
657     VirtIOBlockReq *req = data->req;
658     VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
659     struct iovec *in_iov = data->in_iov;
660     unsigned in_num = data->in_num;
661     int64_t zrp_size, n, j = 0;
662     int64_t nz = data->zone_report_data.nr_zones;
663     int8_t err_status = VIRTIO_BLK_S_OK;
664 
665     trace_virtio_blk_zone_report_complete(vdev, req, nz, ret);
666     if (ret) {
667         err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
668         goto out;
669     }
670 
671     struct virtio_blk_zone_report zrp_hdr = (struct virtio_blk_zone_report) {
672         .nr_zones = cpu_to_le64(nz),
673     };
674     zrp_size = sizeof(struct virtio_blk_zone_report)
675                + sizeof(struct virtio_blk_zone_descriptor) * nz;
676     n = iov_from_buf(in_iov, in_num, 0, &zrp_hdr, sizeof(zrp_hdr));
677     if (n != sizeof(zrp_hdr)) {
678         virtio_error(vdev, "Driver provided input buffer that is too small!");
679         err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
680         goto out;
681     }
682 
683     for (size_t i = sizeof(zrp_hdr); i < zrp_size;
684         i += sizeof(struct virtio_blk_zone_descriptor), ++j) {
685         struct virtio_blk_zone_descriptor desc =
686             (struct virtio_blk_zone_descriptor) {
687                 .z_start = cpu_to_le64(data->zone_report_data.zones[j].start
688                     >> BDRV_SECTOR_BITS),
689                 .z_cap = cpu_to_le64(data->zone_report_data.zones[j].cap
690                     >> BDRV_SECTOR_BITS),
691                 .z_wp = cpu_to_le64(data->zone_report_data.zones[j].wp
692                     >> BDRV_SECTOR_BITS),
693         };
694 
695         switch (data->zone_report_data.zones[j].type) {
696         case BLK_ZT_CONV:
697             desc.z_type = VIRTIO_BLK_ZT_CONV;
698             break;
699         case BLK_ZT_SWR:
700             desc.z_type = VIRTIO_BLK_ZT_SWR;
701             break;
702         case BLK_ZT_SWP:
703             desc.z_type = VIRTIO_BLK_ZT_SWP;
704             break;
705         default:
706             g_assert_not_reached();
707         }
708 
709         switch (data->zone_report_data.zones[j].state) {
710         case BLK_ZS_RDONLY:
711             desc.z_state = VIRTIO_BLK_ZS_RDONLY;
712             break;
713         case BLK_ZS_OFFLINE:
714             desc.z_state = VIRTIO_BLK_ZS_OFFLINE;
715             break;
716         case BLK_ZS_EMPTY:
717             desc.z_state = VIRTIO_BLK_ZS_EMPTY;
718             break;
719         case BLK_ZS_CLOSED:
720             desc.z_state = VIRTIO_BLK_ZS_CLOSED;
721             break;
722         case BLK_ZS_FULL:
723             desc.z_state = VIRTIO_BLK_ZS_FULL;
724             break;
725         case BLK_ZS_EOPEN:
726             desc.z_state = VIRTIO_BLK_ZS_EOPEN;
727             break;
728         case BLK_ZS_IOPEN:
729             desc.z_state = VIRTIO_BLK_ZS_IOPEN;
730             break;
731         case BLK_ZS_NOT_WP:
732             desc.z_state = VIRTIO_BLK_ZS_NOT_WP;
733             break;
734         default:
735             g_assert_not_reached();
736         }
737 
738         /* TODO: it takes O(n^2) time complexity. Optimizations required. */
739         n = iov_from_buf(in_iov, in_num, i, &desc, sizeof(desc));
740         if (n != sizeof(desc)) {
741             virtio_error(vdev, "Driver provided input buffer "
742                                "for descriptors that is too small!");
743             err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
744         }
745     }
746 
747 out:
748     virtio_blk_req_complete(req, err_status);
749     virtio_blk_free_request(req);
750     g_free(data->zone_report_data.zones);
751     g_free(data);
752 }
753 
754 static void virtio_blk_handle_zone_report(VirtIOBlockReq *req,
755                                          struct iovec *in_iov,
756                                          unsigned in_num)
757 {
758     VirtIOBlock *s = req->dev;
759     VirtIODevice *vdev = VIRTIO_DEVICE(s);
760     unsigned int nr_zones;
761     ZoneCmdData *data;
762     int64_t zone_size, offset;
763     uint8_t err_status;
764 
765     if (req->in_len < sizeof(struct virtio_blk_inhdr) +
766             sizeof(struct virtio_blk_zone_report) +
767             sizeof(struct virtio_blk_zone_descriptor)) {
768         virtio_error(vdev, "in buffer too small for zone report");
769         return;
770     }
771 
772     /* start byte offset of the zone report */
773     offset = virtio_ldq_p(vdev, &req->out.sector) << BDRV_SECTOR_BITS;
774     if (!check_zoned_request(s, offset, 0, false, &err_status)) {
775         goto out;
776     }
777     nr_zones = (req->in_len - sizeof(struct virtio_blk_inhdr) -
778                 sizeof(struct virtio_blk_zone_report)) /
779                sizeof(struct virtio_blk_zone_descriptor);
780     trace_virtio_blk_handle_zone_report(vdev, req,
781                                         offset >> BDRV_SECTOR_BITS, nr_zones);
782 
783     zone_size = sizeof(BlockZoneDescriptor) * nr_zones;
784     data = g_malloc(sizeof(ZoneCmdData));
785     data->req = req;
786     data->in_iov = in_iov;
787     data->in_num = in_num;
788     data->zone_report_data.nr_zones = nr_zones;
789     data->zone_report_data.zones = g_malloc(zone_size),
790 
791     blk_aio_zone_report(s->blk, offset, &data->zone_report_data.nr_zones,
792                         data->zone_report_data.zones,
793                         virtio_blk_zone_report_complete, data);
794     return;
795 out:
796     virtio_blk_req_complete(req, err_status);
797     virtio_blk_free_request(req);
798 }
799 
800 static void virtio_blk_zone_mgmt_complete(void *opaque, int ret)
801 {
802     VirtIOBlockReq *req = opaque;
803     VirtIOBlock *s = req->dev;
804     VirtIODevice *vdev = VIRTIO_DEVICE(s);
805     int8_t err_status = VIRTIO_BLK_S_OK;
806     trace_virtio_blk_zone_mgmt_complete(vdev, req,ret);
807 
808     if (ret) {
809         err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
810     }
811 
812     virtio_blk_req_complete(req, err_status);
813     virtio_blk_free_request(req);
814 }
815 
816 static int virtio_blk_handle_zone_mgmt(VirtIOBlockReq *req, BlockZoneOp op)
817 {
818     VirtIOBlock *s = req->dev;
819     VirtIODevice *vdev = VIRTIO_DEVICE(s);
820     BlockDriverState *bs = blk_bs(s->blk);
821     int64_t offset = virtio_ldq_p(vdev, &req->out.sector) << BDRV_SECTOR_BITS;
822     uint64_t len;
823     uint64_t capacity = bs->total_sectors << BDRV_SECTOR_BITS;
824     uint8_t err_status = VIRTIO_BLK_S_OK;
825 
826     uint32_t type = virtio_ldl_p(vdev, &req->out.type);
827     if (type == VIRTIO_BLK_T_ZONE_RESET_ALL) {
828         /* Entire drive capacity */
829         offset = 0;
830         len = capacity;
831         trace_virtio_blk_handle_zone_reset_all(vdev, req, 0,
832                                                bs->total_sectors);
833     } else {
834         if (bs->bl.zone_size > capacity - offset) {
835             /* The zoned device allows the last smaller zone. */
836             len = capacity - bs->bl.zone_size * (bs->bl.nr_zones - 1);
837         } else {
838             len = bs->bl.zone_size;
839         }
840         trace_virtio_blk_handle_zone_mgmt(vdev, req, op,
841                                           offset >> BDRV_SECTOR_BITS,
842                                           len >> BDRV_SECTOR_BITS);
843     }
844 
845     if (!check_zoned_request(s, offset, len, false, &err_status)) {
846         goto out;
847     }
848 
849     blk_aio_zone_mgmt(s->blk, op, offset, len,
850                       virtio_blk_zone_mgmt_complete, req);
851 
852     return 0;
853 out:
854     virtio_blk_req_complete(req, err_status);
855     virtio_blk_free_request(req);
856     return err_status;
857 }
858 
859 static void virtio_blk_zone_append_complete(void *opaque, int ret)
860 {
861     ZoneCmdData *data = opaque;
862     VirtIOBlockReq *req = data->req;
863     VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
864     int64_t append_sector, n;
865     uint8_t err_status = VIRTIO_BLK_S_OK;
866 
867     if (ret) {
868         err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
869         goto out;
870     }
871 
872     virtio_stq_p(vdev, &append_sector,
873                  data->zone_append_data.offset >> BDRV_SECTOR_BITS);
874     n = iov_from_buf(data->in_iov, data->in_num, 0, &append_sector,
875                      sizeof(append_sector));
876     if (n != sizeof(append_sector)) {
877         virtio_error(vdev, "Driver provided input buffer less than size of "
878                            "append_sector");
879         err_status = VIRTIO_BLK_S_ZONE_INVALID_CMD;
880         goto out;
881     }
882     trace_virtio_blk_zone_append_complete(vdev, req, append_sector, ret);
883 
884 out:
885     virtio_blk_req_complete(req, err_status);
886     virtio_blk_free_request(req);
887     g_free(data);
888 }
889 
890 static int virtio_blk_handle_zone_append(VirtIOBlockReq *req,
891                                          struct iovec *out_iov,
892                                          struct iovec *in_iov,
893                                          uint64_t out_num,
894                                          unsigned in_num) {
895     VirtIOBlock *s = req->dev;
896     VirtIODevice *vdev = VIRTIO_DEVICE(s);
897     uint8_t err_status = VIRTIO_BLK_S_OK;
898 
899     int64_t offset = virtio_ldq_p(vdev, &req->out.sector) << BDRV_SECTOR_BITS;
900     int64_t len = iov_size(out_iov, out_num);
901 
902     trace_virtio_blk_handle_zone_append(vdev, req, offset >> BDRV_SECTOR_BITS);
903     if (!check_zoned_request(s, offset, len, true, &err_status)) {
904         goto out;
905     }
906 
907     ZoneCmdData *data = g_malloc(sizeof(ZoneCmdData));
908     data->req = req;
909     data->in_iov = in_iov;
910     data->in_num = in_num;
911     data->zone_append_data.offset = offset;
912     qemu_iovec_init_external(&req->qiov, out_iov, out_num);
913 
914     block_acct_start(blk_get_stats(s->blk), &req->acct, len,
915                      BLOCK_ACCT_ZONE_APPEND);
916 
917     blk_aio_zone_append(s->blk, &data->zone_append_data.offset, &req->qiov, 0,
918                         virtio_blk_zone_append_complete, data);
919     return 0;
920 
921 out:
922     virtio_blk_req_complete(req, err_status);
923     virtio_blk_free_request(req);
924     return err_status;
925 }
926 
927 static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
928 {
929     uint32_t type;
930     struct iovec *in_iov = req->elem.in_sg;
931     struct iovec *out_iov = req->elem.out_sg;
932     unsigned in_num = req->elem.in_num;
933     unsigned out_num = req->elem.out_num;
934     VirtIOBlock *s = req->dev;
935     VirtIODevice *vdev = VIRTIO_DEVICE(s);
936 
937     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
938         virtio_error(vdev, "virtio-blk missing headers");
939         return -1;
940     }
941 
942     if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out,
943                             sizeof(req->out)) != sizeof(req->out))) {
944         virtio_error(vdev, "virtio-blk request outhdr too short");
945         return -1;
946     }
947 
948     iov_discard_front_undoable(&out_iov, &out_num, sizeof(req->out),
949                                &req->outhdr_undo);
950 
951     if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
952         virtio_error(vdev, "virtio-blk request inhdr too short");
953         iov_discard_undo(&req->outhdr_undo);
954         return -1;
955     }
956 
957     /* We always touch the last byte, so just see how big in_iov is.  */
958     req->in_len = iov_size(in_iov, in_num);
959     req->in = (void *)in_iov[in_num - 1].iov_base
960               + in_iov[in_num - 1].iov_len
961               - sizeof(struct virtio_blk_inhdr);
962     iov_discard_back_undoable(in_iov, &in_num, sizeof(struct virtio_blk_inhdr),
963                               &req->inhdr_undo);
964 
965     type = virtio_ldl_p(vdev, &req->out.type);
966 
967     /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
968      * is an optional flag. Although a guest should not send this flag if
969      * not negotiated we ignored it in the past. So keep ignoring it. */
970     switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) {
971     case VIRTIO_BLK_T_IN:
972     {
973         bool is_write = type & VIRTIO_BLK_T_OUT;
974         req->sector_num = virtio_ldq_p(vdev, &req->out.sector);
975 
976         if (is_write) {
977             qemu_iovec_init_external(&req->qiov, out_iov, out_num);
978             trace_virtio_blk_handle_write(vdev, req, req->sector_num,
979                                           req->qiov.size / BDRV_SECTOR_SIZE);
980         } else {
981             qemu_iovec_init_external(&req->qiov, in_iov, in_num);
982             trace_virtio_blk_handle_read(vdev, req, req->sector_num,
983                                          req->qiov.size / BDRV_SECTOR_SIZE);
984         }
985 
986         if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) {
987             virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
988             block_acct_invalid(blk_get_stats(s->blk),
989                                is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
990             virtio_blk_free_request(req);
991             return 0;
992         }
993 
994         block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size,
995                          is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
996 
997         /* merge would exceed maximum number of requests or IO direction
998          * changes */
999         if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS ||
1000                                   is_write != mrb->is_write ||
1001                                   !s->conf.request_merging)) {
1002             virtio_blk_submit_multireq(s, mrb);
1003         }
1004 
1005         assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS);
1006         mrb->reqs[mrb->num_reqs++] = req;
1007         mrb->is_write = is_write;
1008         break;
1009     }
1010     case VIRTIO_BLK_T_FLUSH:
1011         virtio_blk_handle_flush(req, mrb);
1012         break;
1013     case VIRTIO_BLK_T_ZONE_REPORT:
1014         virtio_blk_handle_zone_report(req, in_iov, in_num);
1015         break;
1016     case VIRTIO_BLK_T_ZONE_OPEN:
1017         virtio_blk_handle_zone_mgmt(req, BLK_ZO_OPEN);
1018         break;
1019     case VIRTIO_BLK_T_ZONE_CLOSE:
1020         virtio_blk_handle_zone_mgmt(req, BLK_ZO_CLOSE);
1021         break;
1022     case VIRTIO_BLK_T_ZONE_FINISH:
1023         virtio_blk_handle_zone_mgmt(req, BLK_ZO_FINISH);
1024         break;
1025     case VIRTIO_BLK_T_ZONE_RESET:
1026         virtio_blk_handle_zone_mgmt(req, BLK_ZO_RESET);
1027         break;
1028     case VIRTIO_BLK_T_ZONE_RESET_ALL:
1029         virtio_blk_handle_zone_mgmt(req, BLK_ZO_RESET);
1030         break;
1031     case VIRTIO_BLK_T_SCSI_CMD:
1032         virtio_blk_handle_scsi(req);
1033         break;
1034     case VIRTIO_BLK_T_GET_ID:
1035     {
1036         /*
1037          * NB: per existing s/n string convention the string is
1038          * terminated by '\0' only when shorter than buffer.
1039          */
1040         const char *serial = s->conf.serial ? s->conf.serial : "";
1041         size_t size = MIN(strlen(serial) + 1,
1042                           MIN(iov_size(in_iov, in_num),
1043                               VIRTIO_BLK_ID_BYTES));
1044         iov_from_buf(in_iov, in_num, 0, serial, size);
1045         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
1046         virtio_blk_free_request(req);
1047         break;
1048     }
1049     case VIRTIO_BLK_T_ZONE_APPEND & ~VIRTIO_BLK_T_OUT:
1050         /*
1051          * Passing out_iov/out_num and in_iov/in_num is not safe
1052          * to access req->elem.out_sg directly because it may be
1053          * modified by virtio_blk_handle_request().
1054          */
1055         virtio_blk_handle_zone_append(req, out_iov, in_iov, out_num, in_num);
1056         break;
1057     /*
1058      * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with
1059      * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement,
1060      * so we must mask it for these requests, then we will check if it is set.
1061      */
1062     case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT:
1063     case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT:
1064     {
1065         struct virtio_blk_discard_write_zeroes dwz_hdr;
1066         size_t out_len = iov_size(out_iov, out_num);
1067         bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) ==
1068                                VIRTIO_BLK_T_WRITE_ZEROES;
1069         uint8_t err_status;
1070 
1071         /*
1072          * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains
1073          * more than one segment.
1074          */
1075         if (unlikely(!(type & VIRTIO_BLK_T_OUT) ||
1076                      out_len > sizeof(dwz_hdr))) {
1077             virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
1078             virtio_blk_free_request(req);
1079             return 0;
1080         }
1081 
1082         if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr,
1083                                 sizeof(dwz_hdr)) != sizeof(dwz_hdr))) {
1084             iov_discard_undo(&req->inhdr_undo);
1085             iov_discard_undo(&req->outhdr_undo);
1086             virtio_error(vdev, "virtio-blk discard/write_zeroes header"
1087                          " too short");
1088             return -1;
1089         }
1090 
1091         err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr,
1092                                                             is_write_zeroes);
1093         if (err_status != VIRTIO_BLK_S_OK) {
1094             virtio_blk_req_complete(req, err_status);
1095             virtio_blk_free_request(req);
1096         }
1097 
1098         break;
1099     }
1100     default:
1101         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
1102         virtio_blk_free_request(req);
1103     }
1104     return 0;
1105 }
1106 
1107 void virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
1108 {
1109     VirtIOBlockReq *req;
1110     MultiReqBuffer mrb = {};
1111     bool suppress_notifications = virtio_queue_get_notification(vq);
1112 
1113     defer_call_begin();
1114 
1115     do {
1116         if (suppress_notifications) {
1117             virtio_queue_set_notification(vq, 0);
1118         }
1119 
1120         while ((req = virtio_blk_get_request(s, vq))) {
1121             if (virtio_blk_handle_request(req, &mrb)) {
1122                 virtqueue_detach_element(req->vq, &req->elem, 0);
1123                 virtio_blk_free_request(req);
1124                 break;
1125             }
1126         }
1127 
1128         if (suppress_notifications) {
1129             virtio_queue_set_notification(vq, 1);
1130         }
1131     } while (!virtio_queue_empty(vq));
1132 
1133     if (mrb.num_reqs) {
1134         virtio_blk_submit_multireq(s, &mrb);
1135     }
1136 
1137     defer_call_end();
1138 }
1139 
1140 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
1141 {
1142     VirtIOBlock *s = (VirtIOBlock *)vdev;
1143 
1144     if (!s->ioeventfd_disabled && !s->ioeventfd_started) {
1145         /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
1146          * ioeventfd here instead of waiting for .set_status().
1147          */
1148         virtio_device_start_ioeventfd(vdev);
1149         if (!s->ioeventfd_disabled) {
1150             return;
1151         }
1152     }
1153 
1154     virtio_blk_handle_vq(s, vq);
1155 }
1156 
1157 static void virtio_blk_dma_restart_bh(void *opaque)
1158 {
1159     VirtIOBlockReq *req = opaque;
1160     VirtIOBlock *s = req->dev; /* we're called with at least one request */
1161 
1162     MultiReqBuffer mrb = {};
1163 
1164     while (req) {
1165         VirtIOBlockReq *next = req->next;
1166         if (virtio_blk_handle_request(req, &mrb)) {
1167             /* Device is now broken and won't do any processing until it gets
1168              * reset. Already queued requests will be lost: let's purge them.
1169              */
1170             while (req) {
1171                 next = req->next;
1172                 virtqueue_detach_element(req->vq, &req->elem, 0);
1173                 virtio_blk_free_request(req);
1174                 req = next;
1175             }
1176             break;
1177         }
1178         req = next;
1179     }
1180 
1181     if (mrb.num_reqs) {
1182         virtio_blk_submit_multireq(s, &mrb);
1183     }
1184 
1185     /* Paired with inc in virtio_blk_dma_restart_cb() */
1186     blk_dec_in_flight(s->conf.conf.blk);
1187 }
1188 
1189 static void virtio_blk_dma_restart_cb(void *opaque, bool running,
1190                                       RunState state)
1191 {
1192     VirtIOBlock *s = opaque;
1193     uint16_t num_queues = s->conf.num_queues;
1194 
1195     if (!running) {
1196         return;
1197     }
1198 
1199     /* Split the device-wide s->rq request list into per-vq request lists */
1200     g_autofree VirtIOBlockReq **vq_rq = g_new0(VirtIOBlockReq *, num_queues);
1201     VirtIOBlockReq *rq;
1202 
1203     WITH_QEMU_LOCK_GUARD(&s->rq_lock) {
1204         rq = s->rq;
1205         s->rq = NULL;
1206     }
1207 
1208     while (rq) {
1209         VirtIOBlockReq *next = rq->next;
1210         uint16_t idx = virtio_get_queue_index(rq->vq);
1211 
1212         rq->next = vq_rq[idx];
1213         vq_rq[idx] = rq;
1214         rq = next;
1215     }
1216 
1217     /* Schedule a BH to submit the requests in each vq's AioContext */
1218     for (uint16_t i = 0; i < num_queues; i++) {
1219         if (!vq_rq[i]) {
1220             continue;
1221         }
1222 
1223         /* Paired with dec in virtio_blk_dma_restart_bh() */
1224         blk_inc_in_flight(s->conf.conf.blk);
1225 
1226         aio_bh_schedule_oneshot(s->vq_aio_context[i],
1227                                 virtio_blk_dma_restart_bh,
1228                                 vq_rq[i]);
1229     }
1230 }
1231 
1232 static void virtio_blk_reset(VirtIODevice *vdev)
1233 {
1234     VirtIOBlock *s = VIRTIO_BLK(vdev);
1235     VirtIOBlockReq *req;
1236 
1237     /* Dataplane has stopped... */
1238     assert(!s->ioeventfd_started);
1239 
1240     /* ...but requests may still be in flight. */
1241     blk_drain(s->blk);
1242 
1243     /* We drop queued requests after blk_drain() because blk_drain() itself can
1244      * produce them. */
1245     WITH_QEMU_LOCK_GUARD(&s->rq_lock) {
1246         while (s->rq) {
1247             req = s->rq;
1248             s->rq = req->next;
1249 
1250             /* No other threads can access req->vq here */
1251             virtqueue_detach_element(req->vq, &req->elem, 0);
1252 
1253             virtio_blk_free_request(req);
1254         }
1255     }
1256 
1257     blk_set_enable_write_cache(s->blk, s->original_wce);
1258 }
1259 
1260 /* coalesce internal state, copy to pci i/o region 0
1261  */
1262 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
1263 {
1264     VirtIOBlock *s = VIRTIO_BLK(vdev);
1265     BlockConf *conf = &s->conf.conf;
1266     BlockDriverState *bs = blk_bs(s->blk);
1267     struct virtio_blk_config blkcfg;
1268     uint64_t capacity;
1269     int64_t length;
1270     int blk_size = conf->logical_block_size;
1271 
1272     blk_get_geometry(s->blk, &capacity);
1273     memset(&blkcfg, 0, sizeof(blkcfg));
1274     virtio_stq_p(vdev, &blkcfg.capacity, capacity);
1275     virtio_stl_p(vdev, &blkcfg.seg_max,
1276                  s->conf.seg_max_adjust ? s->conf.queue_size - 2 : 128 - 2);
1277     virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls);
1278     virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
1279     virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size);
1280     virtio_stl_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size);
1281     blkcfg.geometry.heads = conf->heads;
1282     /*
1283      * We must ensure that the block device capacity is a multiple of
1284      * the logical block size. If that is not the case, let's use
1285      * sector_mask to adopt the geometry to have a correct picture.
1286      * For those devices where the capacity is ok for the given geometry
1287      * we don't touch the sector value of the geometry, since some devices
1288      * (like s390 dasd) need a specific value. Here the capacity is already
1289      * cyls*heads*secs*blk_size and the sector value is not block size
1290      * divided by 512 - instead it is the amount of blk_size blocks
1291      * per track (cylinder).
1292      */
1293     length = blk_getlength(s->blk);
1294     if (length > 0 && length / conf->heads / conf->secs % blk_size) {
1295         blkcfg.geometry.sectors = conf->secs & ~s->sector_mask;
1296     } else {
1297         blkcfg.geometry.sectors = conf->secs;
1298     }
1299     blkcfg.size_max = 0;
1300     blkcfg.physical_block_exp = get_physical_block_exp(conf);
1301     blkcfg.alignment_offset = 0;
1302     blkcfg.wce = blk_enable_write_cache(s->blk);
1303     virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues);
1304     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) {
1305         uint32_t discard_granularity = conf->discard_granularity;
1306         if (discard_granularity == -1 || !s->conf.report_discard_granularity) {
1307             discard_granularity = blk_size;
1308         }
1309         virtio_stl_p(vdev, &blkcfg.max_discard_sectors,
1310                      s->conf.max_discard_sectors);
1311         virtio_stl_p(vdev, &blkcfg.discard_sector_alignment,
1312                      discard_granularity >> BDRV_SECTOR_BITS);
1313         /*
1314          * We support only one segment per request since multiple segments
1315          * are not widely used and there are no userspace APIs that allow
1316          * applications to submit multiple segments in a single call.
1317          */
1318         virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1);
1319     }
1320     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) {
1321         virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors,
1322                      s->conf.max_write_zeroes_sectors);
1323         blkcfg.write_zeroes_may_unmap = 1;
1324         virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1);
1325     }
1326     if (bs->bl.zoned != BLK_Z_NONE) {
1327         switch (bs->bl.zoned) {
1328         case BLK_Z_HM:
1329             blkcfg.zoned.model = VIRTIO_BLK_Z_HM;
1330             break;
1331         case BLK_Z_HA:
1332             blkcfg.zoned.model = VIRTIO_BLK_Z_HA;
1333             break;
1334         default:
1335             g_assert_not_reached();
1336         }
1337 
1338         virtio_stl_p(vdev, &blkcfg.zoned.zone_sectors,
1339                      bs->bl.zone_size / 512);
1340         virtio_stl_p(vdev, &blkcfg.zoned.max_active_zones,
1341                      bs->bl.max_active_zones);
1342         virtio_stl_p(vdev, &blkcfg.zoned.max_open_zones,
1343                      bs->bl.max_open_zones);
1344         virtio_stl_p(vdev, &blkcfg.zoned.write_granularity, blk_size);
1345         virtio_stl_p(vdev, &blkcfg.zoned.max_append_sectors,
1346                      bs->bl.max_append_sectors);
1347     } else {
1348         blkcfg.zoned.model = VIRTIO_BLK_Z_NONE;
1349     }
1350     memcpy(config, &blkcfg, s->config_size);
1351 }
1352 
1353 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
1354 {
1355     VirtIOBlock *s = VIRTIO_BLK(vdev);
1356     struct virtio_blk_config blkcfg;
1357 
1358     memcpy(&blkcfg, config, s->config_size);
1359 
1360     blk_set_enable_write_cache(s->blk, blkcfg.wce != 0);
1361 }
1362 
1363 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features,
1364                                         Error **errp)
1365 {
1366     VirtIOBlock *s = VIRTIO_BLK(vdev);
1367 
1368     /* Firstly sync all virtio-blk possible supported features */
1369     features |= s->host_features;
1370 
1371     virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
1372     virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
1373     virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
1374     virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
1375     if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) {
1376         if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) {
1377             error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
1378             return 0;
1379         }
1380     } else {
1381         virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT);
1382         virtio_add_feature(&features, VIRTIO_BLK_F_SCSI);
1383     }
1384 
1385     if (blk_enable_write_cache(s->blk) ||
1386         (s->conf.x_enable_wce_if_config_wce &&
1387          virtio_has_feature(features, VIRTIO_BLK_F_CONFIG_WCE))) {
1388         virtio_add_feature(&features, VIRTIO_BLK_F_WCE);
1389     }
1390     if (!blk_is_writable(s->blk)) {
1391         virtio_add_feature(&features, VIRTIO_BLK_F_RO);
1392     }
1393     if (s->conf.num_queues > 1) {
1394         virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
1395     }
1396 
1397     return features;
1398 }
1399 
1400 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
1401 {
1402     VirtIOBlock *s = VIRTIO_BLK(vdev);
1403 
1404     if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) {
1405         assert(!s->ioeventfd_started);
1406     }
1407 
1408     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
1409         return;
1410     }
1411 
1412     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
1413      * cache flushes.  Thus, the "auto writethrough" behavior is never
1414      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
1415      * Leaving it enabled would break the following sequence:
1416      *
1417      *     Guest started with "-drive cache=writethrough"
1418      *     Guest sets status to 0
1419      *     Guest sets DRIVER bit in status field
1420      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
1421      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
1422      *     Guest writes 1 to the WCE configuration field (writeback mode)
1423      *     Guest sets DRIVER_OK bit in status field
1424      *
1425      * s->blk would erroneously be placed in writethrough mode.
1426      */
1427     if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) {
1428         blk_set_enable_write_cache(s->blk,
1429                                    virtio_vdev_has_feature(vdev,
1430                                                            VIRTIO_BLK_F_WCE));
1431     }
1432 }
1433 
1434 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
1435 {
1436     VirtIOBlock *s = VIRTIO_BLK(vdev);
1437 
1438     WITH_QEMU_LOCK_GUARD(&s->rq_lock) {
1439         VirtIOBlockReq *req = s->rq;
1440 
1441         while (req) {
1442             qemu_put_sbyte(f, 1);
1443 
1444             if (s->conf.num_queues > 1) {
1445                 qemu_put_be32(f, virtio_get_queue_index(req->vq));
1446             }
1447 
1448             qemu_put_virtqueue_element(vdev, f, &req->elem);
1449             req = req->next;
1450         }
1451     }
1452 
1453     qemu_put_sbyte(f, 0);
1454 }
1455 
1456 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
1457                                   int version_id)
1458 {
1459     VirtIOBlock *s = VIRTIO_BLK(vdev);
1460 
1461     while (qemu_get_sbyte(f)) {
1462         unsigned nvqs = s->conf.num_queues;
1463         unsigned vq_idx = 0;
1464         VirtIOBlockReq *req;
1465 
1466         if (nvqs > 1) {
1467             vq_idx = qemu_get_be32(f);
1468 
1469             if (vq_idx >= nvqs) {
1470                 error_report("Invalid virtqueue index in request list: %#x",
1471                              vq_idx);
1472                 return -EINVAL;
1473             }
1474         }
1475 
1476         req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq));
1477         virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req);
1478 
1479         WITH_QEMU_LOCK_GUARD(&s->rq_lock) {
1480             req->next = s->rq;
1481             s->rq = req;
1482         }
1483     }
1484 
1485     return 0;
1486 }
1487 
1488 static bool
1489 validate_iothread_vq_mapping_list(IOThreadVirtQueueMappingList *list,
1490         uint16_t num_queues, Error **errp)
1491 {
1492     g_autofree unsigned long *vqs = bitmap_new(num_queues);
1493     g_autoptr(GHashTable) iothreads =
1494         g_hash_table_new(g_str_hash, g_str_equal);
1495 
1496     for (IOThreadVirtQueueMappingList *node = list; node; node = node->next) {
1497         const char *name = node->value->iothread;
1498         uint16List *vq;
1499 
1500         if (!iothread_by_id(name)) {
1501             error_setg(errp, "IOThread \"%s\" object does not exist", name);
1502             return false;
1503         }
1504 
1505         if (!g_hash_table_add(iothreads, (gpointer)name)) {
1506             error_setg(errp,
1507                     "duplicate IOThread name \"%s\" in iothread-vq-mapping",
1508                     name);
1509             return false;
1510         }
1511 
1512         if (node != list) {
1513             if (!!node->value->vqs != !!list->value->vqs) {
1514                 error_setg(errp, "either all items in iothread-vq-mapping "
1515                                  "must have vqs or none of them must have it");
1516                 return false;
1517             }
1518         }
1519 
1520         for (vq = node->value->vqs; vq; vq = vq->next) {
1521             if (vq->value >= num_queues) {
1522                 error_setg(errp, "vq index %u for IOThread \"%s\" must be "
1523                         "less than num_queues %u in iothread-vq-mapping",
1524                         vq->value, name, num_queues);
1525                 return false;
1526             }
1527 
1528             if (test_and_set_bit(vq->value, vqs)) {
1529                 error_setg(errp, "cannot assign vq %u to IOThread \"%s\" "
1530                         "because it is already assigned", vq->value, name);
1531                 return false;
1532             }
1533         }
1534     }
1535 
1536     if (list->value->vqs) {
1537         for (uint16_t i = 0; i < num_queues; i++) {
1538             if (!test_bit(i, vqs)) {
1539                 error_setg(errp,
1540                         "missing vq %u IOThread assignment in iothread-vq-mapping",
1541                         i);
1542                 return false;
1543             }
1544         }
1545     }
1546 
1547     return true;
1548 }
1549 
1550 static void virtio_resize_cb(void *opaque)
1551 {
1552     VirtIODevice *vdev = opaque;
1553 
1554     assert(qemu_get_current_aio_context() == qemu_get_aio_context());
1555     virtio_notify_config(vdev);
1556 }
1557 
1558 static void virtio_blk_resize(void *opaque)
1559 {
1560     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
1561 
1562     /*
1563      * virtio_notify_config() needs to acquire the BQL,
1564      * so it can't be called from an iothread. Instead, schedule
1565      * it to be run in the main context BH.
1566      */
1567     aio_bh_schedule_oneshot(qemu_get_aio_context(), virtio_resize_cb, vdev);
1568 }
1569 
1570 static void virtio_blk_ioeventfd_detach(VirtIOBlock *s)
1571 {
1572     VirtIODevice *vdev = VIRTIO_DEVICE(s);
1573 
1574     for (uint16_t i = 0; i < s->conf.num_queues; i++) {
1575         VirtQueue *vq = virtio_get_queue(vdev, i);
1576         virtio_queue_aio_detach_host_notifier(vq, s->vq_aio_context[i]);
1577     }
1578 }
1579 
1580 static void virtio_blk_ioeventfd_attach(VirtIOBlock *s)
1581 {
1582     VirtIODevice *vdev = VIRTIO_DEVICE(s);
1583 
1584     for (uint16_t i = 0; i < s->conf.num_queues; i++) {
1585         VirtQueue *vq = virtio_get_queue(vdev, i);
1586         virtio_queue_aio_attach_host_notifier(vq, s->vq_aio_context[i]);
1587     }
1588 }
1589 
1590 /* Suspend virtqueue ioeventfd processing during drain */
1591 static void virtio_blk_drained_begin(void *opaque)
1592 {
1593     VirtIOBlock *s = opaque;
1594 
1595     if (s->ioeventfd_started) {
1596         virtio_blk_ioeventfd_detach(s);
1597     }
1598 }
1599 
1600 /* Resume virtqueue ioeventfd processing after drain */
1601 static void virtio_blk_drained_end(void *opaque)
1602 {
1603     VirtIOBlock *s = opaque;
1604 
1605     if (s->ioeventfd_started) {
1606         virtio_blk_ioeventfd_attach(s);
1607     }
1608 }
1609 
1610 static const BlockDevOps virtio_block_ops = {
1611     .resize_cb     = virtio_blk_resize,
1612     .drained_begin = virtio_blk_drained_begin,
1613     .drained_end   = virtio_blk_drained_end,
1614 };
1615 
1616 /* Generate vq:AioContext mappings from a validated iothread-vq-mapping list */
1617 static void
1618 apply_vq_mapping(IOThreadVirtQueueMappingList *iothread_vq_mapping_list,
1619                  AioContext **vq_aio_context, uint16_t num_queues)
1620 {
1621     IOThreadVirtQueueMappingList *node;
1622     size_t num_iothreads = 0;
1623     size_t cur_iothread = 0;
1624 
1625     for (node = iothread_vq_mapping_list; node; node = node->next) {
1626         num_iothreads++;
1627     }
1628 
1629     for (node = iothread_vq_mapping_list; node; node = node->next) {
1630         IOThread *iothread = iothread_by_id(node->value->iothread);
1631         AioContext *ctx = iothread_get_aio_context(iothread);
1632 
1633         /* Released in virtio_blk_vq_aio_context_cleanup() */
1634         object_ref(OBJECT(iothread));
1635 
1636         if (node->value->vqs) {
1637             uint16List *vq;
1638 
1639             /* Explicit vq:IOThread assignment */
1640             for (vq = node->value->vqs; vq; vq = vq->next) {
1641                 vq_aio_context[vq->value] = ctx;
1642             }
1643         } else {
1644             /* Round-robin vq:IOThread assignment */
1645             for (unsigned i = cur_iothread; i < num_queues;
1646                  i += num_iothreads) {
1647                 vq_aio_context[i] = ctx;
1648             }
1649         }
1650 
1651         cur_iothread++;
1652     }
1653 }
1654 
1655 /* Context: BQL held */
1656 static bool virtio_blk_vq_aio_context_init(VirtIOBlock *s, Error **errp)
1657 {
1658     VirtIODevice *vdev = VIRTIO_DEVICE(s);
1659     VirtIOBlkConf *conf = &s->conf;
1660     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev)));
1661     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1662 
1663     if (conf->iothread || conf->iothread_vq_mapping_list) {
1664         if (!k->set_guest_notifiers || !k->ioeventfd_assign) {
1665             error_setg(errp,
1666                        "device is incompatible with iothread "
1667                        "(transport does not support notifiers)");
1668             return false;
1669         }
1670         if (!virtio_device_ioeventfd_enabled(vdev)) {
1671             error_setg(errp, "ioeventfd is required for iothread");
1672             return false;
1673         }
1674 
1675         /*
1676          * If ioeventfd is (re-)enabled while the guest is running there could
1677          * be block jobs that can conflict.
1678          */
1679         if (blk_op_is_blocked(conf->conf.blk, BLOCK_OP_TYPE_DATAPLANE, errp)) {
1680             error_prepend(errp, "cannot start virtio-blk ioeventfd: ");
1681             return false;
1682         }
1683     }
1684 
1685     s->vq_aio_context = g_new(AioContext *, conf->num_queues);
1686 
1687     if (conf->iothread_vq_mapping_list) {
1688         apply_vq_mapping(conf->iothread_vq_mapping_list, s->vq_aio_context,
1689                          conf->num_queues);
1690     } else if (conf->iothread) {
1691         AioContext *ctx = iothread_get_aio_context(conf->iothread);
1692         for (unsigned i = 0; i < conf->num_queues; i++) {
1693             s->vq_aio_context[i] = ctx;
1694         }
1695 
1696         /* Released in virtio_blk_vq_aio_context_cleanup() */
1697         object_ref(OBJECT(conf->iothread));
1698     } else {
1699         AioContext *ctx = qemu_get_aio_context();
1700         for (unsigned i = 0; i < conf->num_queues; i++) {
1701             s->vq_aio_context[i] = ctx;
1702         }
1703     }
1704 
1705     return true;
1706 }
1707 
1708 /* Context: BQL held */
1709 static void virtio_blk_vq_aio_context_cleanup(VirtIOBlock *s)
1710 {
1711     VirtIOBlkConf *conf = &s->conf;
1712 
1713     assert(!s->ioeventfd_started);
1714 
1715     if (conf->iothread_vq_mapping_list) {
1716         IOThreadVirtQueueMappingList *node;
1717 
1718         for (node = conf->iothread_vq_mapping_list; node; node = node->next) {
1719             IOThread *iothread = iothread_by_id(node->value->iothread);
1720             object_unref(OBJECT(iothread));
1721         }
1722     }
1723 
1724     if (conf->iothread) {
1725         object_unref(OBJECT(conf->iothread));
1726     }
1727 
1728     g_free(s->vq_aio_context);
1729     s->vq_aio_context = NULL;
1730 }
1731 
1732 /* Context: BQL held */
1733 static int virtio_blk_start_ioeventfd(VirtIODevice *vdev)
1734 {
1735     VirtIOBlock *s = VIRTIO_BLK(vdev);
1736     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(s)));
1737     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1738     unsigned i;
1739     unsigned nvqs = s->conf.num_queues;
1740     Error *local_err = NULL;
1741     int r;
1742 
1743     if (s->ioeventfd_started || s->ioeventfd_starting) {
1744         return 0;
1745     }
1746 
1747     s->ioeventfd_starting = true;
1748 
1749     /* Set up guest notifier (irq) */
1750     r = k->set_guest_notifiers(qbus->parent, nvqs, true);
1751     if (r != 0) {
1752         error_report("virtio-blk failed to set guest notifier (%d), "
1753                      "ensure -accel kvm is set.", r);
1754         goto fail_guest_notifiers;
1755     }
1756 
1757     /*
1758      * Batch all the host notifiers in a single transaction to avoid
1759      * quadratic time complexity in address_space_update_ioeventfds().
1760      */
1761     memory_region_transaction_begin();
1762 
1763     /* Set up virtqueue notify */
1764     for (i = 0; i < nvqs; i++) {
1765         r = virtio_bus_set_host_notifier(VIRTIO_BUS(qbus), i, true);
1766         if (r != 0) {
1767             int j = i;
1768 
1769             fprintf(stderr, "virtio-blk failed to set host notifier (%d)\n", r);
1770             while (i--) {
1771                 virtio_bus_set_host_notifier(VIRTIO_BUS(qbus), i, false);
1772             }
1773 
1774             /*
1775              * The transaction expects the ioeventfds to be open when it
1776              * commits. Do it now, before the cleanup loop.
1777              */
1778             memory_region_transaction_commit();
1779 
1780             while (j--) {
1781                 virtio_bus_cleanup_host_notifier(VIRTIO_BUS(qbus), j);
1782             }
1783             goto fail_host_notifiers;
1784         }
1785     }
1786 
1787     memory_region_transaction_commit();
1788 
1789     /*
1790      * Try to change the AioContext so that block jobs and other operations can
1791      * co-locate their activity in the same AioContext. If it fails, nevermind.
1792      */
1793     r = blk_set_aio_context(s->conf.conf.blk, s->vq_aio_context[0],
1794                             &local_err);
1795     if (r < 0) {
1796         warn_report_err(local_err);
1797     }
1798 
1799     /*
1800      * These fields must be visible to the IOThread when it processes the
1801      * virtqueue, otherwise it will think ioeventfd has not started yet.
1802      *
1803      * Make sure ->ioeventfd_started is false when blk_set_aio_context() is
1804      * called above so that draining does not cause the host notifier to be
1805      * detached/attached prematurely.
1806      */
1807     s->ioeventfd_starting = false;
1808     s->ioeventfd_started = true;
1809     smp_wmb(); /* paired with aio_notify_accept() on the read side */
1810 
1811     /* Get this show started by hooking up our callbacks */
1812     for (i = 0; i < nvqs; i++) {
1813         VirtQueue *vq = virtio_get_queue(vdev, i);
1814         AioContext *ctx = s->vq_aio_context[i];
1815 
1816         /* Kick right away to begin processing requests already in vring */
1817         event_notifier_set(virtio_queue_get_host_notifier(vq));
1818 
1819         if (!blk_in_drain(s->conf.conf.blk)) {
1820             virtio_queue_aio_attach_host_notifier(vq, ctx);
1821         }
1822     }
1823     return 0;
1824 
1825   fail_host_notifiers:
1826     k->set_guest_notifiers(qbus->parent, nvqs, false);
1827   fail_guest_notifiers:
1828     s->ioeventfd_disabled = true;
1829     s->ioeventfd_starting = false;
1830     return -ENOSYS;
1831 }
1832 
1833 /* Stop notifications for new requests from guest.
1834  *
1835  * Context: BH in IOThread
1836  */
1837 static void virtio_blk_ioeventfd_stop_vq_bh(void *opaque)
1838 {
1839     VirtQueue *vq = opaque;
1840     EventNotifier *host_notifier = virtio_queue_get_host_notifier(vq);
1841 
1842     virtio_queue_aio_detach_host_notifier(vq, qemu_get_current_aio_context());
1843 
1844     /*
1845      * Test and clear notifier after disabling event, in case poll callback
1846      * didn't have time to run.
1847      */
1848     virtio_queue_host_notifier_read(host_notifier);
1849 }
1850 
1851 /* Context: BQL held */
1852 static void virtio_blk_stop_ioeventfd(VirtIODevice *vdev)
1853 {
1854     VirtIOBlock *s = VIRTIO_BLK(vdev);
1855     BusState *qbus = qdev_get_parent_bus(DEVICE(s));
1856     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
1857     unsigned i;
1858     unsigned nvqs = s->conf.num_queues;
1859 
1860     if (!s->ioeventfd_started || s->ioeventfd_stopping) {
1861         return;
1862     }
1863 
1864     /* Better luck next time. */
1865     if (s->ioeventfd_disabled) {
1866         s->ioeventfd_disabled = false;
1867         s->ioeventfd_started = false;
1868         return;
1869     }
1870     s->ioeventfd_stopping = true;
1871 
1872     if (!blk_in_drain(s->conf.conf.blk)) {
1873         for (i = 0; i < nvqs; i++) {
1874             VirtQueue *vq = virtio_get_queue(vdev, i);
1875             AioContext *ctx = s->vq_aio_context[i];
1876 
1877             aio_wait_bh_oneshot(ctx, virtio_blk_ioeventfd_stop_vq_bh, vq);
1878         }
1879     }
1880 
1881     /*
1882      * Batch all the host notifiers in a single transaction to avoid
1883      * quadratic time complexity in address_space_update_ioeventfds().
1884      */
1885     memory_region_transaction_begin();
1886 
1887     for (i = 0; i < nvqs; i++) {
1888         virtio_bus_set_host_notifier(VIRTIO_BUS(qbus), i, false);
1889     }
1890 
1891     /*
1892      * The transaction expects the ioeventfds to be open when it
1893      * commits. Do it now, before the cleanup loop.
1894      */
1895     memory_region_transaction_commit();
1896 
1897     for (i = 0; i < nvqs; i++) {
1898         virtio_bus_cleanup_host_notifier(VIRTIO_BUS(qbus), i);
1899     }
1900 
1901     /*
1902      * Set ->ioeventfd_started to false before draining so that host notifiers
1903      * are not detached/attached anymore.
1904      */
1905     s->ioeventfd_started = false;
1906 
1907     /* Wait for virtio_blk_dma_restart_bh() and in flight I/O to complete */
1908     blk_drain(s->conf.conf.blk);
1909 
1910     /*
1911      * Try to switch bs back to the QEMU main loop. If other users keep the
1912      * BlockBackend in the iothread, that's ok
1913      */
1914     blk_set_aio_context(s->conf.conf.blk, qemu_get_aio_context(), NULL);
1915 
1916     /* Clean up guest notifier (irq) */
1917     k->set_guest_notifiers(qbus->parent, nvqs, false);
1918 
1919     s->ioeventfd_stopping = false;
1920 }
1921 
1922 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
1923 {
1924     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
1925     VirtIOBlock *s = VIRTIO_BLK(dev);
1926     VirtIOBlkConf *conf = &s->conf;
1927     Error *err = NULL;
1928     unsigned i;
1929 
1930     if (!conf->conf.blk) {
1931         error_setg(errp, "drive property not set");
1932         return;
1933     }
1934     if (!blk_is_inserted(conf->conf.blk)) {
1935         error_setg(errp, "Device needs media, but drive is empty");
1936         return;
1937     }
1938     if (conf->num_queues == VIRTIO_BLK_AUTO_NUM_QUEUES) {
1939         conf->num_queues = 1;
1940     }
1941     if (!conf->num_queues) {
1942         error_setg(errp, "num-queues property must be larger than 0");
1943         return;
1944     }
1945     if (conf->queue_size <= 2) {
1946         error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1947                    "must be > 2", conf->queue_size);
1948         return;
1949     }
1950     if (!is_power_of_2(conf->queue_size) ||
1951         conf->queue_size > VIRTQUEUE_MAX_SIZE) {
1952         error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
1953                    "must be a power of 2 (max %d)",
1954                    conf->queue_size, VIRTQUEUE_MAX_SIZE);
1955         return;
1956     }
1957 
1958     if (!blkconf_apply_backend_options(&conf->conf,
1959                                        !blk_supports_write_perm(conf->conf.blk),
1960                                        true, errp)) {
1961         return;
1962     }
1963     s->original_wce = blk_enable_write_cache(conf->conf.blk);
1964     if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) {
1965         return;
1966     }
1967 
1968     if (!blkconf_blocksizes(&conf->conf, errp)) {
1969         return;
1970     }
1971 
1972     BlockDriverState *bs = blk_bs(conf->conf.blk);
1973     if (bs->bl.zoned != BLK_Z_NONE) {
1974         virtio_add_feature(&s->host_features, VIRTIO_BLK_F_ZONED);
1975         if (bs->bl.zoned == BLK_Z_HM) {
1976             virtio_clear_feature(&s->host_features, VIRTIO_BLK_F_DISCARD);
1977         }
1978     }
1979 
1980     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) &&
1981         (!conf->max_discard_sectors ||
1982          conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1983         error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")"
1984                    ", must be between 1 and %d",
1985                    conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS);
1986         return;
1987     }
1988 
1989     if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) &&
1990         (!conf->max_write_zeroes_sectors ||
1991          conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) {
1992         error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32
1993                    "), must be between 1 and %d",
1994                    conf->max_write_zeroes_sectors,
1995                    (int)BDRV_REQUEST_MAX_SECTORS);
1996         return;
1997     }
1998 
1999     if (conf->iothread_vq_mapping_list) {
2000         if (conf->iothread) {
2001             error_setg(errp, "iothread and iothread-vq-mapping properties "
2002                              "cannot be set at the same time");
2003             return;
2004         }
2005 
2006         if (!validate_iothread_vq_mapping_list(conf->iothread_vq_mapping_list,
2007                                                conf->num_queues, errp)) {
2008             return;
2009         }
2010     }
2011 
2012     s->config_size = virtio_get_config_size(&virtio_blk_cfg_size_params,
2013                                             s->host_features);
2014     virtio_init(vdev, VIRTIO_ID_BLOCK, s->config_size);
2015 
2016     qemu_mutex_init(&s->rq_lock);
2017 
2018     s->blk = conf->conf.blk;
2019     s->rq = NULL;
2020     s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1;
2021 
2022     for (i = 0; i < conf->num_queues; i++) {
2023         virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output);
2024     }
2025     qemu_coroutine_inc_pool_size(conf->num_queues * conf->queue_size / 2);
2026 
2027     /* Don't start ioeventfd if transport does not support notifiers. */
2028     if (!virtio_device_ioeventfd_enabled(vdev)) {
2029         s->ioeventfd_disabled = true;
2030     }
2031 
2032     virtio_blk_vq_aio_context_init(s, &err);
2033     if (err != NULL) {
2034         error_propagate(errp, err);
2035         for (i = 0; i < conf->num_queues; i++) {
2036             virtio_del_queue(vdev, i);
2037         }
2038         virtio_cleanup(vdev);
2039         return;
2040     }
2041 
2042     /*
2043      * This must be after virtio_init() so virtio_blk_dma_restart_cb() gets
2044      * called after ->start_ioeventfd() has already set blk's AioContext.
2045      */
2046     s->change =
2047         qdev_add_vm_change_state_handler(dev, virtio_blk_dma_restart_cb, s);
2048 
2049     blk_ram_registrar_init(&s->blk_ram_registrar, s->blk);
2050     blk_set_dev_ops(s->blk, &virtio_block_ops, s);
2051 
2052     blk_iostatus_enable(s->blk);
2053 
2054     add_boot_device_lchs(dev, "/disk@0,0",
2055                          conf->conf.lcyls,
2056                          conf->conf.lheads,
2057                          conf->conf.lsecs);
2058 }
2059 
2060 static void virtio_blk_device_unrealize(DeviceState *dev)
2061 {
2062     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
2063     VirtIOBlock *s = VIRTIO_BLK(dev);
2064     VirtIOBlkConf *conf = &s->conf;
2065     unsigned i;
2066 
2067     blk_drain(s->blk);
2068     del_boot_device_lchs(dev, "/disk@0,0");
2069     virtio_blk_vq_aio_context_cleanup(s);
2070     for (i = 0; i < conf->num_queues; i++) {
2071         virtio_del_queue(vdev, i);
2072     }
2073     qemu_coroutine_dec_pool_size(conf->num_queues * conf->queue_size / 2);
2074     qemu_mutex_destroy(&s->rq_lock);
2075     blk_ram_registrar_destroy(&s->blk_ram_registrar);
2076     qemu_del_vm_change_state_handler(s->change);
2077     blockdev_mark_auto_del(s->blk);
2078     virtio_cleanup(vdev);
2079 }
2080 
2081 static void virtio_blk_instance_init(Object *obj)
2082 {
2083     VirtIOBlock *s = VIRTIO_BLK(obj);
2084 
2085     device_add_bootindex_property(obj, &s->conf.conf.bootindex,
2086                                   "bootindex", "/disk@0,0",
2087                                   DEVICE(obj));
2088 }
2089 
2090 static const VMStateDescription vmstate_virtio_blk = {
2091     .name = "virtio-blk",
2092     .minimum_version_id = 2,
2093     .version_id = 2,
2094     .fields = (const VMStateField[]) {
2095         VMSTATE_VIRTIO_DEVICE,
2096         VMSTATE_END_OF_LIST()
2097     },
2098 };
2099 
2100 static Property virtio_blk_properties[] = {
2101     DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf),
2102     DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf),
2103     DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf),
2104     DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial),
2105     DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features,
2106                       VIRTIO_BLK_F_CONFIG_WCE, true),
2107 #ifdef __linux__
2108     DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features,
2109                       VIRTIO_BLK_F_SCSI, false),
2110 #endif
2111     DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
2112                     true),
2113     DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues,
2114                        VIRTIO_BLK_AUTO_NUM_QUEUES),
2115     DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 256),
2116     DEFINE_PROP_BOOL("seg-max-adjust", VirtIOBlock, conf.seg_max_adjust, true),
2117     DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD,
2118                      IOThread *),
2119     DEFINE_PROP_IOTHREAD_VQ_MAPPING_LIST("iothread-vq-mapping", VirtIOBlock,
2120                                          conf.iothread_vq_mapping_list),
2121     DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features,
2122                       VIRTIO_BLK_F_DISCARD, true),
2123     DEFINE_PROP_BOOL("report-discard-granularity", VirtIOBlock,
2124                      conf.report_discard_granularity, true),
2125     DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features,
2126                       VIRTIO_BLK_F_WRITE_ZEROES, true),
2127     DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock,
2128                        conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS),
2129     DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock,
2130                        conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS),
2131     DEFINE_PROP_BOOL("x-enable-wce-if-config-wce", VirtIOBlock,
2132                      conf.x_enable_wce_if_config_wce, true),
2133     DEFINE_PROP_END_OF_LIST(),
2134 };
2135 
2136 static void virtio_blk_class_init(ObjectClass *klass, void *data)
2137 {
2138     DeviceClass *dc = DEVICE_CLASS(klass);
2139     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
2140 
2141     device_class_set_props(dc, virtio_blk_properties);
2142     dc->vmsd = &vmstate_virtio_blk;
2143     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
2144     vdc->realize = virtio_blk_device_realize;
2145     vdc->unrealize = virtio_blk_device_unrealize;
2146     vdc->get_config = virtio_blk_update_config;
2147     vdc->set_config = virtio_blk_set_config;
2148     vdc->get_features = virtio_blk_get_features;
2149     vdc->set_status = virtio_blk_set_status;
2150     vdc->reset = virtio_blk_reset;
2151     vdc->save = virtio_blk_save_device;
2152     vdc->load = virtio_blk_load_device;
2153     vdc->start_ioeventfd = virtio_blk_start_ioeventfd;
2154     vdc->stop_ioeventfd = virtio_blk_stop_ioeventfd;
2155 }
2156 
2157 static const TypeInfo virtio_blk_info = {
2158     .name = TYPE_VIRTIO_BLK,
2159     .parent = TYPE_VIRTIO_DEVICE,
2160     .instance_size = sizeof(VirtIOBlock),
2161     .instance_init = virtio_blk_instance_init,
2162     .class_init = virtio_blk_class_init,
2163 };
2164 
2165 static void virtio_register_types(void)
2166 {
2167     type_register_static(&virtio_blk_info);
2168 }
2169 
2170 type_init(virtio_register_types)
2171