xref: /openbmc/qemu/hw/block/virtio-blk.c (revision 39164c13)
1 /*
2  * Virtio Block Device
3  *
4  * Copyright IBM, Corp. 2007
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "qapi/error.h"
16 #include "qemu-common.h"
17 #include "qemu/iov.h"
18 #include "qemu/error-report.h"
19 #include "trace.h"
20 #include "hw/block/block.h"
21 #include "sysemu/block-backend.h"
22 #include "sysemu/blockdev.h"
23 #include "hw/virtio/virtio-blk.h"
24 #include "dataplane/virtio-blk.h"
25 #include "block/scsi.h"
26 #ifdef __linux__
27 # include <scsi/sg.h>
28 #endif
29 #include "hw/virtio/virtio-bus.h"
30 #include "hw/virtio/virtio-access.h"
31 
32 static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq,
33                                     VirtIOBlockReq *req)
34 {
35     req->dev = s;
36     req->vq = vq;
37     req->qiov.size = 0;
38     req->in_len = 0;
39     req->next = NULL;
40     req->mr_next = NULL;
41 }
42 
43 static void virtio_blk_free_request(VirtIOBlockReq *req)
44 {
45     if (req) {
46         g_free(req);
47     }
48 }
49 
50 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
51 {
52     VirtIOBlock *s = req->dev;
53     VirtIODevice *vdev = VIRTIO_DEVICE(s);
54 
55     trace_virtio_blk_req_complete(req, status);
56 
57     stb_p(&req->in->status, status);
58     virtqueue_push(req->vq, &req->elem, req->in_len);
59     if (s->dataplane_started && !s->dataplane_disabled) {
60         virtio_blk_data_plane_notify(s->dataplane, req->vq);
61     } else {
62         virtio_notify(vdev, req->vq);
63     }
64 }
65 
66 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
67     bool is_read)
68 {
69     BlockErrorAction action = blk_get_error_action(req->dev->blk,
70                                                    is_read, error);
71     VirtIOBlock *s = req->dev;
72 
73     if (action == BLOCK_ERROR_ACTION_STOP) {
74         /* Break the link as the next request is going to be parsed from the
75          * ring again. Otherwise we may end up doing a double completion! */
76         req->mr_next = NULL;
77         req->next = s->rq;
78         s->rq = req;
79     } else if (action == BLOCK_ERROR_ACTION_REPORT) {
80         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
81         block_acct_failed(blk_get_stats(s->blk), &req->acct);
82         virtio_blk_free_request(req);
83     }
84 
85     blk_error_action(s->blk, action, is_read, error);
86     return action != BLOCK_ERROR_ACTION_IGNORE;
87 }
88 
89 static void virtio_blk_rw_complete(void *opaque, int ret)
90 {
91     VirtIOBlockReq *next = opaque;
92     VirtIOBlock *s = next->dev;
93 
94     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
95     while (next) {
96         VirtIOBlockReq *req = next;
97         next = req->mr_next;
98         trace_virtio_blk_rw_complete(req, ret);
99 
100         if (req->qiov.nalloc != -1) {
101             /* If nalloc is != 1 req->qiov is a local copy of the original
102              * external iovec. It was allocated in submit_merged_requests
103              * to be able to merge requests. */
104             qemu_iovec_destroy(&req->qiov);
105         }
106 
107         if (ret) {
108             int p = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type);
109             bool is_read = !(p & VIRTIO_BLK_T_OUT);
110             /* Note that memory may be dirtied on read failure.  If the
111              * virtio request is not completed here, as is the case for
112              * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
113              * correctly during live migration.  While this is ugly,
114              * it is acceptable because the device is free to write to
115              * the memory until the request is completed (which will
116              * happen on the other side of the migration).
117              */
118             if (virtio_blk_handle_rw_error(req, -ret, is_read)) {
119                 continue;
120             }
121         }
122 
123         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
124         block_acct_done(blk_get_stats(req->dev->blk), &req->acct);
125         virtio_blk_free_request(req);
126     }
127     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
128 }
129 
130 static void virtio_blk_flush_complete(void *opaque, int ret)
131 {
132     VirtIOBlockReq *req = opaque;
133     VirtIOBlock *s = req->dev;
134 
135     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
136     if (ret) {
137         if (virtio_blk_handle_rw_error(req, -ret, 0)) {
138             goto out;
139         }
140     }
141 
142     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
143     block_acct_done(blk_get_stats(req->dev->blk), &req->acct);
144     virtio_blk_free_request(req);
145 
146 out:
147     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
148 }
149 
150 #ifdef __linux__
151 
152 typedef struct {
153     VirtIOBlockReq *req;
154     struct sg_io_hdr hdr;
155 } VirtIOBlockIoctlReq;
156 
157 static void virtio_blk_ioctl_complete(void *opaque, int status)
158 {
159     VirtIOBlockIoctlReq *ioctl_req = opaque;
160     VirtIOBlockReq *req = ioctl_req->req;
161     VirtIOBlock *s = req->dev;
162     VirtIODevice *vdev = VIRTIO_DEVICE(s);
163     struct virtio_scsi_inhdr *scsi;
164     struct sg_io_hdr *hdr;
165 
166     scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
167 
168     if (status) {
169         status = VIRTIO_BLK_S_UNSUPP;
170         virtio_stl_p(vdev, &scsi->errors, 255);
171         goto out;
172     }
173 
174     hdr = &ioctl_req->hdr;
175     /*
176      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
177      * clear the masked_status field [hence status gets cleared too, see
178      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
179      * status has occurred.  However they do set DRIVER_SENSE in driver_status
180      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
181      */
182     if (hdr->status == 0 && hdr->sb_len_wr > 0) {
183         hdr->status = CHECK_CONDITION;
184     }
185 
186     virtio_stl_p(vdev, &scsi->errors,
187                  hdr->status | (hdr->msg_status << 8) |
188                  (hdr->host_status << 16) | (hdr->driver_status << 24));
189     virtio_stl_p(vdev, &scsi->residual, hdr->resid);
190     virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr);
191     virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
192 
193 out:
194     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
195     virtio_blk_req_complete(req, status);
196     virtio_blk_free_request(req);
197     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
198     g_free(ioctl_req);
199 }
200 
201 #endif
202 
203 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq)
204 {
205     VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq));
206 
207     if (req) {
208         virtio_blk_init_request(s, vq, req);
209     }
210     return req;
211 }
212 
213 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req)
214 {
215     int status = VIRTIO_BLK_S_OK;
216     struct virtio_scsi_inhdr *scsi = NULL;
217     VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
218     VirtQueueElement *elem = &req->elem;
219     VirtIOBlock *blk = req->dev;
220 
221 #ifdef __linux__
222     int i;
223     VirtIOBlockIoctlReq *ioctl_req;
224     BlockAIOCB *acb;
225 #endif
226 
227     /*
228      * We require at least one output segment each for the virtio_blk_outhdr
229      * and the SCSI command block.
230      *
231      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
232      * and the sense buffer pointer in the input segments.
233      */
234     if (elem->out_num < 2 || elem->in_num < 3) {
235         status = VIRTIO_BLK_S_IOERR;
236         goto fail;
237     }
238 
239     /*
240      * The scsi inhdr is placed in the second-to-last input segment, just
241      * before the regular inhdr.
242      */
243     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
244 
245     if (!blk->conf.scsi) {
246         status = VIRTIO_BLK_S_UNSUPP;
247         goto fail;
248     }
249 
250     /*
251      * No support for bidirection commands yet.
252      */
253     if (elem->out_num > 2 && elem->in_num > 3) {
254         status = VIRTIO_BLK_S_UNSUPP;
255         goto fail;
256     }
257 
258 #ifdef __linux__
259     ioctl_req = g_new0(VirtIOBlockIoctlReq, 1);
260     ioctl_req->req = req;
261     ioctl_req->hdr.interface_id = 'S';
262     ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len;
263     ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base;
264     ioctl_req->hdr.dxfer_len = 0;
265 
266     if (elem->out_num > 2) {
267         /*
268          * If there are more than the minimally required 2 output segments
269          * there is write payload starting from the third iovec.
270          */
271         ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV;
272         ioctl_req->hdr.iovec_count = elem->out_num - 2;
273 
274         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
275             ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
276         }
277 
278         ioctl_req->hdr.dxferp = elem->out_sg + 2;
279 
280     } else if (elem->in_num > 3) {
281         /*
282          * If we have more than 3 input segments the guest wants to actually
283          * read data.
284          */
285         ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV;
286         ioctl_req->hdr.iovec_count = elem->in_num - 3;
287         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
288             ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len;
289         }
290 
291         ioctl_req->hdr.dxferp = elem->in_sg;
292     } else {
293         /*
294          * Some SCSI commands don't actually transfer any data.
295          */
296         ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE;
297     }
298 
299     ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
300     ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
301 
302     acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr,
303                         virtio_blk_ioctl_complete, ioctl_req);
304     if (!acb) {
305         g_free(ioctl_req);
306         status = VIRTIO_BLK_S_UNSUPP;
307         goto fail;
308     }
309     return -EINPROGRESS;
310 #else
311     abort();
312 #endif
313 
314 fail:
315     /* Just put anything nonzero so that the ioctl fails in the guest.  */
316     if (scsi) {
317         virtio_stl_p(vdev, &scsi->errors, 255);
318     }
319     return status;
320 }
321 
322 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
323 {
324     int status;
325 
326     status = virtio_blk_handle_scsi_req(req);
327     if (status != -EINPROGRESS) {
328         virtio_blk_req_complete(req, status);
329         virtio_blk_free_request(req);
330     }
331 }
332 
333 static inline void submit_requests(BlockBackend *blk, MultiReqBuffer *mrb,
334                                    int start, int num_reqs, int niov)
335 {
336     QEMUIOVector *qiov = &mrb->reqs[start]->qiov;
337     int64_t sector_num = mrb->reqs[start]->sector_num;
338     bool is_write = mrb->is_write;
339 
340     if (num_reqs > 1) {
341         int i;
342         struct iovec *tmp_iov = qiov->iov;
343         int tmp_niov = qiov->niov;
344 
345         /* mrb->reqs[start]->qiov was initialized from external so we can't
346          * modify it here. We need to initialize it locally and then add the
347          * external iovecs. */
348         qemu_iovec_init(qiov, niov);
349 
350         for (i = 0; i < tmp_niov; i++) {
351             qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len);
352         }
353 
354         for (i = start + 1; i < start + num_reqs; i++) {
355             qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0,
356                               mrb->reqs[i]->qiov.size);
357             mrb->reqs[i - 1]->mr_next = mrb->reqs[i];
358         }
359 
360         trace_virtio_blk_submit_multireq(mrb, start, num_reqs,
361                                          sector_num << BDRV_SECTOR_BITS,
362                                          qiov->size, is_write);
363         block_acct_merge_done(blk_get_stats(blk),
364                               is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ,
365                               num_reqs - 1);
366     }
367 
368     if (is_write) {
369         blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
370                         virtio_blk_rw_complete, mrb->reqs[start]);
371     } else {
372         blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
373                        virtio_blk_rw_complete, mrb->reqs[start]);
374     }
375 }
376 
377 static int multireq_compare(const void *a, const void *b)
378 {
379     const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a,
380                          *req2 = *(VirtIOBlockReq **)b;
381 
382     /*
383      * Note that we can't simply subtract sector_num1 from sector_num2
384      * here as that could overflow the return value.
385      */
386     if (req1->sector_num > req2->sector_num) {
387         return 1;
388     } else if (req1->sector_num < req2->sector_num) {
389         return -1;
390     } else {
391         return 0;
392     }
393 }
394 
395 static void virtio_blk_submit_multireq(BlockBackend *blk, MultiReqBuffer *mrb)
396 {
397     int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0;
398     uint32_t max_transfer;
399     int64_t sector_num = 0;
400 
401     if (mrb->num_reqs == 1) {
402         submit_requests(blk, mrb, 0, 1, -1);
403         mrb->num_reqs = 0;
404         return;
405     }
406 
407     max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk);
408 
409     qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs),
410           &multireq_compare);
411 
412     for (i = 0; i < mrb->num_reqs; i++) {
413         VirtIOBlockReq *req = mrb->reqs[i];
414         if (num_reqs > 0) {
415             /*
416              * NOTE: We cannot merge the requests in below situations:
417              * 1. requests are not sequential
418              * 2. merge would exceed maximum number of IOVs
419              * 3. merge would exceed maximum transfer length of backend device
420              */
421             if (sector_num + nb_sectors != req->sector_num ||
422                 niov > blk_get_max_iov(blk) - req->qiov.niov ||
423                 req->qiov.size > max_transfer ||
424                 nb_sectors > (max_transfer -
425                               req->qiov.size) / BDRV_SECTOR_SIZE) {
426                 submit_requests(blk, mrb, start, num_reqs, niov);
427                 num_reqs = 0;
428             }
429         }
430 
431         if (num_reqs == 0) {
432             sector_num = req->sector_num;
433             nb_sectors = niov = 0;
434             start = i;
435         }
436 
437         nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE;
438         niov += req->qiov.niov;
439         num_reqs++;
440     }
441 
442     submit_requests(blk, mrb, start, num_reqs, niov);
443     mrb->num_reqs = 0;
444 }
445 
446 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
447 {
448     block_acct_start(blk_get_stats(req->dev->blk), &req->acct, 0,
449                      BLOCK_ACCT_FLUSH);
450 
451     /*
452      * Make sure all outstanding writes are posted to the backing device.
453      */
454     if (mrb->is_write && mrb->num_reqs > 0) {
455         virtio_blk_submit_multireq(req->dev->blk, mrb);
456     }
457     blk_aio_flush(req->dev->blk, virtio_blk_flush_complete, req);
458 }
459 
460 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
461                                      uint64_t sector, size_t size)
462 {
463     uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
464     uint64_t total_sectors;
465 
466     if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
467         return false;
468     }
469     if (sector & dev->sector_mask) {
470         return false;
471     }
472     if (size % dev->conf.conf.logical_block_size) {
473         return false;
474     }
475     blk_get_geometry(dev->blk, &total_sectors);
476     if (sector > total_sectors || nb_sectors > total_sectors - sector) {
477         return false;
478     }
479     return true;
480 }
481 
482 static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
483 {
484     uint32_t type;
485     struct iovec *in_iov = req->elem.in_sg;
486     struct iovec *iov = req->elem.out_sg;
487     unsigned in_num = req->elem.in_num;
488     unsigned out_num = req->elem.out_num;
489     VirtIOBlock *s = req->dev;
490     VirtIODevice *vdev = VIRTIO_DEVICE(s);
491 
492     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
493         virtio_error(vdev, "virtio-blk missing headers");
494         return -1;
495     }
496 
497     if (unlikely(iov_to_buf(iov, out_num, 0, &req->out,
498                             sizeof(req->out)) != sizeof(req->out))) {
499         virtio_error(vdev, "virtio-blk request outhdr too short");
500         return -1;
501     }
502 
503     iov_discard_front(&iov, &out_num, sizeof(req->out));
504 
505     if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
506         virtio_error(vdev, "virtio-blk request inhdr too short");
507         return -1;
508     }
509 
510     /* We always touch the last byte, so just see how big in_iov is.  */
511     req->in_len = iov_size(in_iov, in_num);
512     req->in = (void *)in_iov[in_num - 1].iov_base
513               + in_iov[in_num - 1].iov_len
514               - sizeof(struct virtio_blk_inhdr);
515     iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr));
516 
517     type = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type);
518 
519     /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
520      * is an optional flag. Although a guest should not send this flag if
521      * not negotiated we ignored it in the past. So keep ignoring it. */
522     switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) {
523     case VIRTIO_BLK_T_IN:
524     {
525         bool is_write = type & VIRTIO_BLK_T_OUT;
526         req->sector_num = virtio_ldq_p(VIRTIO_DEVICE(req->dev),
527                                        &req->out.sector);
528 
529         if (is_write) {
530             qemu_iovec_init_external(&req->qiov, iov, out_num);
531             trace_virtio_blk_handle_write(req, req->sector_num,
532                                           req->qiov.size / BDRV_SECTOR_SIZE);
533         } else {
534             qemu_iovec_init_external(&req->qiov, in_iov, in_num);
535             trace_virtio_blk_handle_read(req, req->sector_num,
536                                          req->qiov.size / BDRV_SECTOR_SIZE);
537         }
538 
539         if (!virtio_blk_sect_range_ok(req->dev, req->sector_num,
540                                       req->qiov.size)) {
541             virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
542             block_acct_invalid(blk_get_stats(req->dev->blk),
543                                is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
544             virtio_blk_free_request(req);
545             return 0;
546         }
547 
548         block_acct_start(blk_get_stats(req->dev->blk),
549                          &req->acct, req->qiov.size,
550                          is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
551 
552         /* merge would exceed maximum number of requests or IO direction
553          * changes */
554         if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS ||
555                                   is_write != mrb->is_write ||
556                                   !req->dev->conf.request_merging)) {
557             virtio_blk_submit_multireq(req->dev->blk, mrb);
558         }
559 
560         assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS);
561         mrb->reqs[mrb->num_reqs++] = req;
562         mrb->is_write = is_write;
563         break;
564     }
565     case VIRTIO_BLK_T_FLUSH:
566         virtio_blk_handle_flush(req, mrb);
567         break;
568     case VIRTIO_BLK_T_SCSI_CMD:
569         virtio_blk_handle_scsi(req);
570         break;
571     case VIRTIO_BLK_T_GET_ID:
572     {
573         VirtIOBlock *s = req->dev;
574 
575         /*
576          * NB: per existing s/n string convention the string is
577          * terminated by '\0' only when shorter than buffer.
578          */
579         const char *serial = s->conf.serial ? s->conf.serial : "";
580         size_t size = MIN(strlen(serial) + 1,
581                           MIN(iov_size(in_iov, in_num),
582                               VIRTIO_BLK_ID_BYTES));
583         iov_from_buf(in_iov, in_num, 0, serial, size);
584         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
585         virtio_blk_free_request(req);
586         break;
587     }
588     default:
589         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
590         virtio_blk_free_request(req);
591     }
592     return 0;
593 }
594 
595 bool virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
596 {
597     VirtIOBlockReq *req;
598     MultiReqBuffer mrb = {};
599     bool progress = false;
600 
601     aio_context_acquire(blk_get_aio_context(s->blk));
602     blk_io_plug(s->blk);
603 
604     do {
605         virtio_queue_set_notification(vq, 0);
606 
607         while ((req = virtio_blk_get_request(s, vq))) {
608             progress = true;
609             if (virtio_blk_handle_request(req, &mrb)) {
610                 virtqueue_detach_element(req->vq, &req->elem, 0);
611                 virtio_blk_free_request(req);
612                 break;
613             }
614         }
615 
616         virtio_queue_set_notification(vq, 1);
617     } while (!virtio_queue_empty(vq));
618 
619     if (mrb.num_reqs) {
620         virtio_blk_submit_multireq(s->blk, &mrb);
621     }
622 
623     blk_io_unplug(s->blk);
624     aio_context_release(blk_get_aio_context(s->blk));
625     return progress;
626 }
627 
628 static void virtio_blk_handle_output_do(VirtIOBlock *s, VirtQueue *vq)
629 {
630     virtio_blk_handle_vq(s, vq);
631 }
632 
633 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
634 {
635     VirtIOBlock *s = (VirtIOBlock *)vdev;
636 
637     if (s->dataplane) {
638         /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
639          * dataplane here instead of waiting for .set_status().
640          */
641         virtio_device_start_ioeventfd(vdev);
642         if (!s->dataplane_disabled) {
643             return;
644         }
645     }
646     virtio_blk_handle_output_do(s, vq);
647 }
648 
649 static void virtio_blk_dma_restart_bh(void *opaque)
650 {
651     VirtIOBlock *s = opaque;
652     VirtIOBlockReq *req = s->rq;
653     MultiReqBuffer mrb = {};
654 
655     qemu_bh_delete(s->bh);
656     s->bh = NULL;
657 
658     s->rq = NULL;
659 
660     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
661     while (req) {
662         VirtIOBlockReq *next = req->next;
663         if (virtio_blk_handle_request(req, &mrb)) {
664             /* Device is now broken and won't do any processing until it gets
665              * reset. Already queued requests will be lost: let's purge them.
666              */
667             while (req) {
668                 next = req->next;
669                 virtqueue_detach_element(req->vq, &req->elem, 0);
670                 virtio_blk_free_request(req);
671                 req = next;
672             }
673             break;
674         }
675         req = next;
676     }
677 
678     if (mrb.num_reqs) {
679         virtio_blk_submit_multireq(s->blk, &mrb);
680     }
681     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
682 }
683 
684 static void virtio_blk_dma_restart_cb(void *opaque, int running,
685                                       RunState state)
686 {
687     VirtIOBlock *s = opaque;
688 
689     if (!running) {
690         return;
691     }
692 
693     if (!s->bh) {
694         s->bh = aio_bh_new(blk_get_aio_context(s->conf.conf.blk),
695                            virtio_blk_dma_restart_bh, s);
696         qemu_bh_schedule(s->bh);
697     }
698 }
699 
700 static void virtio_blk_reset(VirtIODevice *vdev)
701 {
702     VirtIOBlock *s = VIRTIO_BLK(vdev);
703     AioContext *ctx;
704     VirtIOBlockReq *req;
705 
706     ctx = blk_get_aio_context(s->blk);
707     aio_context_acquire(ctx);
708     blk_drain(s->blk);
709 
710     /* We drop queued requests after blk_drain() because blk_drain() itself can
711      * produce them. */
712     while (s->rq) {
713         req = s->rq;
714         s->rq = req->next;
715         virtqueue_detach_element(req->vq, &req->elem, 0);
716         virtio_blk_free_request(req);
717     }
718 
719     aio_context_release(ctx);
720 
721     assert(!s->dataplane_started);
722     blk_set_enable_write_cache(s->blk, s->original_wce);
723 }
724 
725 /* coalesce internal state, copy to pci i/o region 0
726  */
727 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
728 {
729     VirtIOBlock *s = VIRTIO_BLK(vdev);
730     BlockConf *conf = &s->conf.conf;
731     struct virtio_blk_config blkcfg;
732     uint64_t capacity;
733     int blk_size = conf->logical_block_size;
734 
735     blk_get_geometry(s->blk, &capacity);
736     memset(&blkcfg, 0, sizeof(blkcfg));
737     virtio_stq_p(vdev, &blkcfg.capacity, capacity);
738     virtio_stl_p(vdev, &blkcfg.seg_max, 128 - 2);
739     virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls);
740     virtio_stl_p(vdev, &blkcfg.blk_size, blk_size);
741     virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size);
742     virtio_stw_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size);
743     blkcfg.geometry.heads = conf->heads;
744     /*
745      * We must ensure that the block device capacity is a multiple of
746      * the logical block size. If that is not the case, let's use
747      * sector_mask to adopt the geometry to have a correct picture.
748      * For those devices where the capacity is ok for the given geometry
749      * we don't touch the sector value of the geometry, since some devices
750      * (like s390 dasd) need a specific value. Here the capacity is already
751      * cyls*heads*secs*blk_size and the sector value is not block size
752      * divided by 512 - instead it is the amount of blk_size blocks
753      * per track (cylinder).
754      */
755     if (blk_getlength(s->blk) /  conf->heads / conf->secs % blk_size) {
756         blkcfg.geometry.sectors = conf->secs & ~s->sector_mask;
757     } else {
758         blkcfg.geometry.sectors = conf->secs;
759     }
760     blkcfg.size_max = 0;
761     blkcfg.physical_block_exp = get_physical_block_exp(conf);
762     blkcfg.alignment_offset = 0;
763     blkcfg.wce = blk_enable_write_cache(s->blk);
764     virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues);
765     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
766 }
767 
768 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
769 {
770     VirtIOBlock *s = VIRTIO_BLK(vdev);
771     struct virtio_blk_config blkcfg;
772 
773     memcpy(&blkcfg, config, sizeof(blkcfg));
774 
775     aio_context_acquire(blk_get_aio_context(s->blk));
776     blk_set_enable_write_cache(s->blk, blkcfg.wce != 0);
777     aio_context_release(blk_get_aio_context(s->blk));
778 }
779 
780 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features,
781                                         Error **errp)
782 {
783     VirtIOBlock *s = VIRTIO_BLK(vdev);
784 
785     virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
786     virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
787     virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
788     virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
789     if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) {
790         if (s->conf.scsi) {
791             error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
792             return 0;
793         }
794     } else {
795         virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT);
796         virtio_add_feature(&features, VIRTIO_BLK_F_SCSI);
797     }
798 
799     if (s->conf.config_wce) {
800         virtio_add_feature(&features, VIRTIO_BLK_F_CONFIG_WCE);
801     }
802     if (blk_enable_write_cache(s->blk)) {
803         virtio_add_feature(&features, VIRTIO_BLK_F_WCE);
804     }
805     if (blk_is_read_only(s->blk)) {
806         virtio_add_feature(&features, VIRTIO_BLK_F_RO);
807     }
808     if (s->conf.num_queues > 1) {
809         virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
810     }
811 
812     return features;
813 }
814 
815 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
816 {
817     VirtIOBlock *s = VIRTIO_BLK(vdev);
818 
819     if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) {
820         assert(!s->dataplane_started);
821     }
822 
823     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
824         return;
825     }
826 
827     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
828      * cache flushes.  Thus, the "auto writethrough" behavior is never
829      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
830      * Leaving it enabled would break the following sequence:
831      *
832      *     Guest started with "-drive cache=writethrough"
833      *     Guest sets status to 0
834      *     Guest sets DRIVER bit in status field
835      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
836      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
837      *     Guest writes 1 to the WCE configuration field (writeback mode)
838      *     Guest sets DRIVER_OK bit in status field
839      *
840      * s->blk would erroneously be placed in writethrough mode.
841      */
842     if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) {
843         aio_context_acquire(blk_get_aio_context(s->blk));
844         blk_set_enable_write_cache(s->blk,
845                                    virtio_vdev_has_feature(vdev,
846                                                            VIRTIO_BLK_F_WCE));
847         aio_context_release(blk_get_aio_context(s->blk));
848     }
849 }
850 
851 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
852 {
853     VirtIOBlock *s = VIRTIO_BLK(vdev);
854     VirtIOBlockReq *req = s->rq;
855 
856     while (req) {
857         qemu_put_sbyte(f, 1);
858 
859         if (s->conf.num_queues > 1) {
860             qemu_put_be32(f, virtio_get_queue_index(req->vq));
861         }
862 
863         qemu_put_virtqueue_element(f, &req->elem);
864         req = req->next;
865     }
866     qemu_put_sbyte(f, 0);
867 }
868 
869 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
870                                   int version_id)
871 {
872     VirtIOBlock *s = VIRTIO_BLK(vdev);
873 
874     while (qemu_get_sbyte(f)) {
875         unsigned nvqs = s->conf.num_queues;
876         unsigned vq_idx = 0;
877         VirtIOBlockReq *req;
878 
879         if (nvqs > 1) {
880             vq_idx = qemu_get_be32(f);
881 
882             if (vq_idx >= nvqs) {
883                 error_report("Invalid virtqueue index in request list: %#x",
884                              vq_idx);
885                 return -EINVAL;
886             }
887         }
888 
889         req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq));
890         virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req);
891         req->next = s->rq;
892         s->rq = req;
893     }
894 
895     return 0;
896 }
897 
898 static void virtio_blk_resize(void *opaque)
899 {
900     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
901 
902     virtio_notify_config(vdev);
903 }
904 
905 static const BlockDevOps virtio_block_ops = {
906     .resize_cb = virtio_blk_resize,
907 };
908 
909 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
910 {
911     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
912     VirtIOBlock *s = VIRTIO_BLK(dev);
913     VirtIOBlkConf *conf = &s->conf;
914     Error *err = NULL;
915     unsigned i;
916 
917     if (!conf->conf.blk) {
918         error_setg(errp, "drive property not set");
919         return;
920     }
921     if (!blk_is_inserted(conf->conf.blk)) {
922         error_setg(errp, "Device needs media, but drive is empty");
923         return;
924     }
925     if (!conf->num_queues) {
926         error_setg(errp, "num-queues property must be larger than 0");
927         return;
928     }
929 
930     blkconf_serial(&conf->conf, &conf->serial);
931     blkconf_apply_backend_options(&conf->conf);
932     s->original_wce = blk_enable_write_cache(conf->conf.blk);
933     blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, &err);
934     if (err) {
935         error_propagate(errp, err);
936         return;
937     }
938     blkconf_blocksizes(&conf->conf);
939 
940     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK,
941                 sizeof(struct virtio_blk_config));
942 
943     s->blk = conf->conf.blk;
944     s->rq = NULL;
945     s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1;
946 
947     for (i = 0; i < conf->num_queues; i++) {
948         virtio_add_queue(vdev, 128, virtio_blk_handle_output);
949     }
950     virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err);
951     if (err != NULL) {
952         error_propagate(errp, err);
953         virtio_cleanup(vdev);
954         return;
955     }
956 
957     s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
958     blk_set_dev_ops(s->blk, &virtio_block_ops, s);
959     blk_set_guest_block_size(s->blk, s->conf.conf.logical_block_size);
960 
961     blk_iostatus_enable(s->blk);
962 }
963 
964 static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp)
965 {
966     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
967     VirtIOBlock *s = VIRTIO_BLK(dev);
968 
969     virtio_blk_data_plane_destroy(s->dataplane);
970     s->dataplane = NULL;
971     qemu_del_vm_change_state_handler(s->change);
972     blockdev_mark_auto_del(s->blk);
973     virtio_cleanup(vdev);
974 }
975 
976 static void virtio_blk_instance_init(Object *obj)
977 {
978     VirtIOBlock *s = VIRTIO_BLK(obj);
979 
980     object_property_add_link(obj, "iothread", TYPE_IOTHREAD,
981                              (Object **)&s->conf.iothread,
982                              qdev_prop_allow_set_link_before_realize,
983                              OBJ_PROP_LINK_UNREF_ON_RELEASE, NULL);
984     device_add_bootindex_property(obj, &s->conf.conf.bootindex,
985                                   "bootindex", "/disk@0,0",
986                                   DEVICE(obj), NULL);
987 }
988 
989 static const VMStateDescription vmstate_virtio_blk = {
990     .name = "virtio-blk",
991     .minimum_version_id = 2,
992     .version_id = 2,
993     .fields = (VMStateField[]) {
994         VMSTATE_VIRTIO_DEVICE,
995         VMSTATE_END_OF_LIST()
996     },
997 };
998 
999 static Property virtio_blk_properties[] = {
1000     DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf),
1001     DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf),
1002     DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf),
1003     DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial),
1004     DEFINE_PROP_BIT("config-wce", VirtIOBlock, conf.config_wce, 0, true),
1005 #ifdef __linux__
1006     DEFINE_PROP_BIT("scsi", VirtIOBlock, conf.scsi, 0, false),
1007 #endif
1008     DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
1009                     true),
1010     DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 1),
1011     DEFINE_PROP_END_OF_LIST(),
1012 };
1013 
1014 static void virtio_blk_class_init(ObjectClass *klass, void *data)
1015 {
1016     DeviceClass *dc = DEVICE_CLASS(klass);
1017     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1018 
1019     dc->props = virtio_blk_properties;
1020     dc->vmsd = &vmstate_virtio_blk;
1021     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1022     vdc->realize = virtio_blk_device_realize;
1023     vdc->unrealize = virtio_blk_device_unrealize;
1024     vdc->get_config = virtio_blk_update_config;
1025     vdc->set_config = virtio_blk_set_config;
1026     vdc->get_features = virtio_blk_get_features;
1027     vdc->set_status = virtio_blk_set_status;
1028     vdc->reset = virtio_blk_reset;
1029     vdc->save = virtio_blk_save_device;
1030     vdc->load = virtio_blk_load_device;
1031     vdc->start_ioeventfd = virtio_blk_data_plane_start;
1032     vdc->stop_ioeventfd = virtio_blk_data_plane_stop;
1033 }
1034 
1035 static const TypeInfo virtio_blk_info = {
1036     .name = TYPE_VIRTIO_BLK,
1037     .parent = TYPE_VIRTIO_DEVICE,
1038     .instance_size = sizeof(VirtIOBlock),
1039     .instance_init = virtio_blk_instance_init,
1040     .class_init = virtio_blk_class_init,
1041 };
1042 
1043 static void virtio_register_types(void)
1044 {
1045     type_register_static(&virtio_blk_info);
1046 }
1047 
1048 type_init(virtio_register_types)
1049