xref: /openbmc/qemu/hw/block/virtio-blk.c (revision 5c81161f804144b146607f890e84613a4cbad95c)
1 /*
2  * Virtio Block Device
3  *
4  * Copyright IBM, Corp. 2007
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  */
13 
14 #include "qemu/osdep.h"
15 #include "qapi/error.h"
16 #include "qemu-common.h"
17 #include "qemu/iov.h"
18 #include "qemu/error-report.h"
19 #include "trace.h"
20 #include "hw/block/block.h"
21 #include "sysemu/blockdev.h"
22 #include "hw/virtio/virtio-blk.h"
23 #include "dataplane/virtio-blk.h"
24 #include "scsi/constants.h"
25 #ifdef __linux__
26 # include <scsi/sg.h>
27 #endif
28 #include "hw/virtio/virtio-bus.h"
29 #include "hw/virtio/virtio-access.h"
30 
31 /* We don't support discard yet, hide associated config fields. */
32 #define VIRTIO_BLK_CFG_SIZE offsetof(struct virtio_blk_config, \
33                                      max_discard_sectors)
34 
35 static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq,
36                                     VirtIOBlockReq *req)
37 {
38     req->dev = s;
39     req->vq = vq;
40     req->qiov.size = 0;
41     req->in_len = 0;
42     req->next = NULL;
43     req->mr_next = NULL;
44 }
45 
46 static void virtio_blk_free_request(VirtIOBlockReq *req)
47 {
48     g_free(req);
49 }
50 
51 static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status)
52 {
53     VirtIOBlock *s = req->dev;
54     VirtIODevice *vdev = VIRTIO_DEVICE(s);
55 
56     trace_virtio_blk_req_complete(vdev, req, status);
57 
58     stb_p(&req->in->status, status);
59     virtqueue_push(req->vq, &req->elem, req->in_len);
60     if (s->dataplane_started && !s->dataplane_disabled) {
61         virtio_blk_data_plane_notify(s->dataplane, req->vq);
62     } else {
63         virtio_notify(vdev, req->vq);
64     }
65 }
66 
67 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
68     bool is_read, bool acct_failed)
69 {
70     VirtIOBlock *s = req->dev;
71     BlockErrorAction action = blk_get_error_action(s->blk, is_read, error);
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         if (acct_failed) {
82             block_acct_failed(blk_get_stats(s->blk), &req->acct);
83         }
84         virtio_blk_free_request(req);
85     }
86 
87     blk_error_action(s->blk, action, is_read, error);
88     return action != BLOCK_ERROR_ACTION_IGNORE;
89 }
90 
91 static void virtio_blk_rw_complete(void *opaque, int ret)
92 {
93     VirtIOBlockReq *next = opaque;
94     VirtIOBlock *s = next->dev;
95     VirtIODevice *vdev = VIRTIO_DEVICE(s);
96 
97     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
98     while (next) {
99         VirtIOBlockReq *req = next;
100         next = req->mr_next;
101         trace_virtio_blk_rw_complete(vdev, req, ret);
102 
103         if (req->qiov.nalloc != -1) {
104             /* If nalloc is != -1 req->qiov is a local copy of the original
105              * external iovec. It was allocated in submit_requests to be
106              * able to merge requests. */
107             qemu_iovec_destroy(&req->qiov);
108         }
109 
110         if (ret) {
111             int p = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type);
112             bool is_read = !(p & VIRTIO_BLK_T_OUT);
113             /* Note that memory may be dirtied on read failure.  If the
114              * virtio request is not completed here, as is the case for
115              * BLOCK_ERROR_ACTION_STOP, the memory may not be copied
116              * correctly during live migration.  While this is ugly,
117              * it is acceptable because the device is free to write to
118              * the memory until the request is completed (which will
119              * happen on the other side of the migration).
120              */
121             if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) {
122                 continue;
123             }
124         }
125 
126         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
127         block_acct_done(blk_get_stats(req->dev->blk), &req->acct);
128         virtio_blk_free_request(req);
129     }
130     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
131 }
132 
133 static void virtio_blk_flush_complete(void *opaque, int ret)
134 {
135     VirtIOBlockReq *req = opaque;
136     VirtIOBlock *s = req->dev;
137 
138     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
139     if (ret) {
140         if (virtio_blk_handle_rw_error(req, -ret, 0, true)) {
141             goto out;
142         }
143     }
144 
145     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
146     block_acct_done(blk_get_stats(s->blk), &req->acct);
147     virtio_blk_free_request(req);
148 
149 out:
150     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
151 }
152 
153 #ifdef __linux__
154 
155 typedef struct {
156     VirtIOBlockReq *req;
157     struct sg_io_hdr hdr;
158 } VirtIOBlockIoctlReq;
159 
160 static void virtio_blk_ioctl_complete(void *opaque, int status)
161 {
162     VirtIOBlockIoctlReq *ioctl_req = opaque;
163     VirtIOBlockReq *req = ioctl_req->req;
164     VirtIOBlock *s = req->dev;
165     VirtIODevice *vdev = VIRTIO_DEVICE(s);
166     struct virtio_scsi_inhdr *scsi;
167     struct sg_io_hdr *hdr;
168 
169     scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
170 
171     if (status) {
172         status = VIRTIO_BLK_S_UNSUPP;
173         virtio_stl_p(vdev, &scsi->errors, 255);
174         goto out;
175     }
176 
177     hdr = &ioctl_req->hdr;
178     /*
179      * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi)
180      * clear the masked_status field [hence status gets cleared too, see
181      * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED
182      * status has occurred.  However they do set DRIVER_SENSE in driver_status
183      * field. Also a (sb_len_wr > 0) indicates there is a sense buffer.
184      */
185     if (hdr->status == 0 && hdr->sb_len_wr > 0) {
186         hdr->status = CHECK_CONDITION;
187     }
188 
189     virtio_stl_p(vdev, &scsi->errors,
190                  hdr->status | (hdr->msg_status << 8) |
191                  (hdr->host_status << 16) | (hdr->driver_status << 24));
192     virtio_stl_p(vdev, &scsi->residual, hdr->resid);
193     virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr);
194     virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len);
195 
196 out:
197     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
198     virtio_blk_req_complete(req, status);
199     virtio_blk_free_request(req);
200     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
201     g_free(ioctl_req);
202 }
203 
204 #endif
205 
206 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq)
207 {
208     VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq));
209 
210     if (req) {
211         virtio_blk_init_request(s, vq, req);
212     }
213     return req;
214 }
215 
216 static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req)
217 {
218     int status = VIRTIO_BLK_S_OK;
219     struct virtio_scsi_inhdr *scsi = NULL;
220     VirtIODevice *vdev = VIRTIO_DEVICE(req->dev);
221     VirtQueueElement *elem = &req->elem;
222     VirtIOBlock *blk = req->dev;
223 
224 #ifdef __linux__
225     int i;
226     VirtIOBlockIoctlReq *ioctl_req;
227     BlockAIOCB *acb;
228 #endif
229 
230     /*
231      * We require at least one output segment each for the virtio_blk_outhdr
232      * and the SCSI command block.
233      *
234      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
235      * and the sense buffer pointer in the input segments.
236      */
237     if (elem->out_num < 2 || elem->in_num < 3) {
238         status = VIRTIO_BLK_S_IOERR;
239         goto fail;
240     }
241 
242     /*
243      * The scsi inhdr is placed in the second-to-last input segment, just
244      * before the regular inhdr.
245      */
246     scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base;
247 
248     if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) {
249         status = VIRTIO_BLK_S_UNSUPP;
250         goto fail;
251     }
252 
253     /*
254      * No support for bidirection commands yet.
255      */
256     if (elem->out_num > 2 && elem->in_num > 3) {
257         status = VIRTIO_BLK_S_UNSUPP;
258         goto fail;
259     }
260 
261 #ifdef __linux__
262     ioctl_req = g_new0(VirtIOBlockIoctlReq, 1);
263     ioctl_req->req = req;
264     ioctl_req->hdr.interface_id = 'S';
265     ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len;
266     ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base;
267     ioctl_req->hdr.dxfer_len = 0;
268 
269     if (elem->out_num > 2) {
270         /*
271          * If there are more than the minimally required 2 output segments
272          * there is write payload starting from the third iovec.
273          */
274         ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV;
275         ioctl_req->hdr.iovec_count = elem->out_num - 2;
276 
277         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
278             ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len;
279         }
280 
281         ioctl_req->hdr.dxferp = elem->out_sg + 2;
282 
283     } else if (elem->in_num > 3) {
284         /*
285          * If we have more than 3 input segments the guest wants to actually
286          * read data.
287          */
288         ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV;
289         ioctl_req->hdr.iovec_count = elem->in_num - 3;
290         for (i = 0; i < ioctl_req->hdr.iovec_count; i++) {
291             ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len;
292         }
293 
294         ioctl_req->hdr.dxferp = elem->in_sg;
295     } else {
296         /*
297          * Some SCSI commands don't actually transfer any data.
298          */
299         ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE;
300     }
301 
302     ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base;
303     ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len;
304 
305     acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr,
306                         virtio_blk_ioctl_complete, ioctl_req);
307     if (!acb) {
308         g_free(ioctl_req);
309         status = VIRTIO_BLK_S_UNSUPP;
310         goto fail;
311     }
312     return -EINPROGRESS;
313 #else
314     abort();
315 #endif
316 
317 fail:
318     /* Just put anything nonzero so that the ioctl fails in the guest.  */
319     if (scsi) {
320         virtio_stl_p(vdev, &scsi->errors, 255);
321     }
322     return status;
323 }
324 
325 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
326 {
327     int status;
328 
329     status = virtio_blk_handle_scsi_req(req);
330     if (status != -EINPROGRESS) {
331         virtio_blk_req_complete(req, status);
332         virtio_blk_free_request(req);
333     }
334 }
335 
336 static inline void submit_requests(BlockBackend *blk, MultiReqBuffer *mrb,
337                                    int start, int num_reqs, int niov)
338 {
339     QEMUIOVector *qiov = &mrb->reqs[start]->qiov;
340     int64_t sector_num = mrb->reqs[start]->sector_num;
341     bool is_write = mrb->is_write;
342 
343     if (num_reqs > 1) {
344         int i;
345         struct iovec *tmp_iov = qiov->iov;
346         int tmp_niov = qiov->niov;
347 
348         /* mrb->reqs[start]->qiov was initialized from external so we can't
349          * modify it here. We need to initialize it locally and then add the
350          * external iovecs. */
351         qemu_iovec_init(qiov, niov);
352 
353         for (i = 0; i < tmp_niov; i++) {
354             qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len);
355         }
356 
357         for (i = start + 1; i < start + num_reqs; i++) {
358             qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0,
359                               mrb->reqs[i]->qiov.size);
360             mrb->reqs[i - 1]->mr_next = mrb->reqs[i];
361         }
362 
363         trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev),
364                                          mrb, start, num_reqs,
365                                          sector_num << BDRV_SECTOR_BITS,
366                                          qiov->size, is_write);
367         block_acct_merge_done(blk_get_stats(blk),
368                               is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ,
369                               num_reqs - 1);
370     }
371 
372     if (is_write) {
373         blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
374                         virtio_blk_rw_complete, mrb->reqs[start]);
375     } else {
376         blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0,
377                        virtio_blk_rw_complete, mrb->reqs[start]);
378     }
379 }
380 
381 static int multireq_compare(const void *a, const void *b)
382 {
383     const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a,
384                          *req2 = *(VirtIOBlockReq **)b;
385 
386     /*
387      * Note that we can't simply subtract sector_num1 from sector_num2
388      * here as that could overflow the return value.
389      */
390     if (req1->sector_num > req2->sector_num) {
391         return 1;
392     } else if (req1->sector_num < req2->sector_num) {
393         return -1;
394     } else {
395         return 0;
396     }
397 }
398 
399 static void virtio_blk_submit_multireq(BlockBackend *blk, MultiReqBuffer *mrb)
400 {
401     int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0;
402     uint32_t max_transfer;
403     int64_t sector_num = 0;
404 
405     if (mrb->num_reqs == 1) {
406         submit_requests(blk, mrb, 0, 1, -1);
407         mrb->num_reqs = 0;
408         return;
409     }
410 
411     max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk);
412 
413     qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs),
414           &multireq_compare);
415 
416     for (i = 0; i < mrb->num_reqs; i++) {
417         VirtIOBlockReq *req = mrb->reqs[i];
418         if (num_reqs > 0) {
419             /*
420              * NOTE: We cannot merge the requests in below situations:
421              * 1. requests are not sequential
422              * 2. merge would exceed maximum number of IOVs
423              * 3. merge would exceed maximum transfer length of backend device
424              */
425             if (sector_num + nb_sectors != req->sector_num ||
426                 niov > blk_get_max_iov(blk) - req->qiov.niov ||
427                 req->qiov.size > max_transfer ||
428                 nb_sectors > (max_transfer -
429                               req->qiov.size) / BDRV_SECTOR_SIZE) {
430                 submit_requests(blk, mrb, start, num_reqs, niov);
431                 num_reqs = 0;
432             }
433         }
434 
435         if (num_reqs == 0) {
436             sector_num = req->sector_num;
437             nb_sectors = niov = 0;
438             start = i;
439         }
440 
441         nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE;
442         niov += req->qiov.niov;
443         num_reqs++;
444     }
445 
446     submit_requests(blk, mrb, start, num_reqs, niov);
447     mrb->num_reqs = 0;
448 }
449 
450 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
451 {
452     block_acct_start(blk_get_stats(req->dev->blk), &req->acct, 0,
453                      BLOCK_ACCT_FLUSH);
454 
455     /*
456      * Make sure all outstanding writes are posted to the backing device.
457      */
458     if (mrb->is_write && mrb->num_reqs > 0) {
459         virtio_blk_submit_multireq(req->dev->blk, mrb);
460     }
461     blk_aio_flush(req->dev->blk, virtio_blk_flush_complete, req);
462 }
463 
464 static bool virtio_blk_sect_range_ok(VirtIOBlock *dev,
465                                      uint64_t sector, size_t size)
466 {
467     uint64_t nb_sectors = size >> BDRV_SECTOR_BITS;
468     uint64_t total_sectors;
469 
470     if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) {
471         return false;
472     }
473     if (sector & dev->sector_mask) {
474         return false;
475     }
476     if (size % dev->conf.conf.logical_block_size) {
477         return false;
478     }
479     blk_get_geometry(dev->blk, &total_sectors);
480     if (sector > total_sectors || nb_sectors > total_sectors - sector) {
481         return false;
482     }
483     return true;
484 }
485 
486 static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb)
487 {
488     uint32_t type;
489     struct iovec *in_iov = req->elem.in_sg;
490     struct iovec *out_iov = req->elem.out_sg;
491     unsigned in_num = req->elem.in_num;
492     unsigned out_num = req->elem.out_num;
493     VirtIOBlock *s = req->dev;
494     VirtIODevice *vdev = VIRTIO_DEVICE(s);
495 
496     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
497         virtio_error(vdev, "virtio-blk missing headers");
498         return -1;
499     }
500 
501     if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out,
502                             sizeof(req->out)) != sizeof(req->out))) {
503         virtio_error(vdev, "virtio-blk request outhdr too short");
504         return -1;
505     }
506 
507     iov_discard_front(&out_iov, &out_num, sizeof(req->out));
508 
509     if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
510         virtio_error(vdev, "virtio-blk request inhdr too short");
511         return -1;
512     }
513 
514     /* We always touch the last byte, so just see how big in_iov is.  */
515     req->in_len = iov_size(in_iov, in_num);
516     req->in = (void *)in_iov[in_num - 1].iov_base
517               + in_iov[in_num - 1].iov_len
518               - sizeof(struct virtio_blk_inhdr);
519     iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr));
520 
521     type = virtio_ldl_p(vdev, &req->out.type);
522 
523     /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER
524      * is an optional flag. Although a guest should not send this flag if
525      * not negotiated we ignored it in the past. So keep ignoring it. */
526     switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) {
527     case VIRTIO_BLK_T_IN:
528     {
529         bool is_write = type & VIRTIO_BLK_T_OUT;
530         req->sector_num = virtio_ldq_p(vdev, &req->out.sector);
531 
532         if (is_write) {
533             qemu_iovec_init_external(&req->qiov, out_iov, out_num);
534             trace_virtio_blk_handle_write(vdev, req, req->sector_num,
535                                           req->qiov.size / BDRV_SECTOR_SIZE);
536         } else {
537             qemu_iovec_init_external(&req->qiov, in_iov, in_num);
538             trace_virtio_blk_handle_read(vdev, req, req->sector_num,
539                                          req->qiov.size / BDRV_SECTOR_SIZE);
540         }
541 
542         if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) {
543             virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
544             block_acct_invalid(blk_get_stats(s->blk),
545                                is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
546             virtio_blk_free_request(req);
547             return 0;
548         }
549 
550         block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size,
551                          is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ);
552 
553         /* merge would exceed maximum number of requests or IO direction
554          * changes */
555         if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS ||
556                                   is_write != mrb->is_write ||
557                                   !s->conf.request_merging)) {
558             virtio_blk_submit_multireq(s->blk, mrb);
559         }
560 
561         assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS);
562         mrb->reqs[mrb->num_reqs++] = req;
563         mrb->is_write = is_write;
564         break;
565     }
566     case VIRTIO_BLK_T_FLUSH:
567         virtio_blk_handle_flush(req, mrb);
568         break;
569     case VIRTIO_BLK_T_SCSI_CMD:
570         virtio_blk_handle_scsi(req);
571         break;
572     case VIRTIO_BLK_T_GET_ID:
573     {
574         /*
575          * NB: per existing s/n string convention the string is
576          * terminated by '\0' only when shorter than buffer.
577          */
578         const char *serial = s->conf.serial ? s->conf.serial : "";
579         size_t size = MIN(strlen(serial) + 1,
580                           MIN(iov_size(in_iov, in_num),
581                               VIRTIO_BLK_ID_BYTES));
582         iov_from_buf(in_iov, in_num, 0, serial, size);
583         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
584         virtio_blk_free_request(req);
585         break;
586     }
587     default:
588         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
589         virtio_blk_free_request(req);
590     }
591     return 0;
592 }
593 
594 bool virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq)
595 {
596     VirtIOBlockReq *req;
597     MultiReqBuffer mrb = {};
598     bool progress = false;
599 
600     aio_context_acquire(blk_get_aio_context(s->blk));
601     blk_io_plug(s->blk);
602 
603     do {
604         virtio_queue_set_notification(vq, 0);
605 
606         while ((req = virtio_blk_get_request(s, vq))) {
607             progress = true;
608             if (virtio_blk_handle_request(req, &mrb)) {
609                 virtqueue_detach_element(req->vq, &req->elem, 0);
610                 virtio_blk_free_request(req);
611                 break;
612             }
613         }
614 
615         virtio_queue_set_notification(vq, 1);
616     } while (!virtio_queue_empty(vq));
617 
618     if (mrb.num_reqs) {
619         virtio_blk_submit_multireq(s->blk, &mrb);
620     }
621 
622     blk_io_unplug(s->blk);
623     aio_context_release(blk_get_aio_context(s->blk));
624     return progress;
625 }
626 
627 static void virtio_blk_handle_output_do(VirtIOBlock *s, VirtQueue *vq)
628 {
629     virtio_blk_handle_vq(s, vq);
630 }
631 
632 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
633 {
634     VirtIOBlock *s = (VirtIOBlock *)vdev;
635 
636     if (s->dataplane) {
637         /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
638          * dataplane here instead of waiting for .set_status().
639          */
640         virtio_device_start_ioeventfd(vdev);
641         if (!s->dataplane_disabled) {
642             return;
643         }
644     }
645     virtio_blk_handle_output_do(s, vq);
646 }
647 
648 static void virtio_blk_dma_restart_bh(void *opaque)
649 {
650     VirtIOBlock *s = opaque;
651     VirtIOBlockReq *req = s->rq;
652     MultiReqBuffer mrb = {};
653 
654     qemu_bh_delete(s->bh);
655     s->bh = NULL;
656 
657     s->rq = NULL;
658 
659     aio_context_acquire(blk_get_aio_context(s->conf.conf.blk));
660     while (req) {
661         VirtIOBlockReq *next = req->next;
662         if (virtio_blk_handle_request(req, &mrb)) {
663             /* Device is now broken and won't do any processing until it gets
664              * reset. Already queued requests will be lost: let's purge them.
665              */
666             while (req) {
667                 next = req->next;
668                 virtqueue_detach_element(req->vq, &req->elem, 0);
669                 virtio_blk_free_request(req);
670                 req = next;
671             }
672             break;
673         }
674         req = next;
675     }
676 
677     if (mrb.num_reqs) {
678         virtio_blk_submit_multireq(s->blk, &mrb);
679     }
680     aio_context_release(blk_get_aio_context(s->conf.conf.blk));
681 }
682 
683 static void virtio_blk_dma_restart_cb(void *opaque, int running,
684                                       RunState state)
685 {
686     VirtIOBlock *s = opaque;
687 
688     if (!running) {
689         return;
690     }
691 
692     if (!s->bh) {
693         s->bh = aio_bh_new(blk_get_aio_context(s->conf.conf.blk),
694                            virtio_blk_dma_restart_bh, s);
695         qemu_bh_schedule(s->bh);
696     }
697 }
698 
699 static void virtio_blk_reset(VirtIODevice *vdev)
700 {
701     VirtIOBlock *s = VIRTIO_BLK(vdev);
702     AioContext *ctx;
703     VirtIOBlockReq *req;
704 
705     ctx = blk_get_aio_context(s->blk);
706     aio_context_acquire(ctx);
707     blk_drain(s->blk);
708 
709     /* We drop queued requests after blk_drain() because blk_drain() itself can
710      * produce them. */
711     while (s->rq) {
712         req = s->rq;
713         s->rq = req->next;
714         virtqueue_detach_element(req->vq, &req->elem, 0);
715         virtio_blk_free_request(req);
716     }
717 
718     aio_context_release(ctx);
719 
720     assert(!s->dataplane_started);
721     blk_set_enable_write_cache(s->blk, s->original_wce);
722 }
723 
724 /* coalesce internal state, copy to pci i/o region 0
725  */
726 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
727 {
728     VirtIOBlock *s = VIRTIO_BLK(vdev);
729     BlockConf *conf = &s->conf.conf;
730     struct virtio_blk_config blkcfg;
731     uint64_t capacity;
732     int64_t length;
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     length = blk_getlength(s->blk);
756     if (length > 0 && length / conf->heads / conf->secs % blk_size) {
757         blkcfg.geometry.sectors = conf->secs & ~s->sector_mask;
758     } else {
759         blkcfg.geometry.sectors = conf->secs;
760     }
761     blkcfg.size_max = 0;
762     blkcfg.physical_block_exp = get_physical_block_exp(conf);
763     blkcfg.alignment_offset = 0;
764     blkcfg.wce = blk_enable_write_cache(s->blk);
765     virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues);
766     memcpy(config, &blkcfg, VIRTIO_BLK_CFG_SIZE);
767     QEMU_BUILD_BUG_ON(VIRTIO_BLK_CFG_SIZE > sizeof(blkcfg));
768 }
769 
770 static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
771 {
772     VirtIOBlock *s = VIRTIO_BLK(vdev);
773     struct virtio_blk_config blkcfg;
774 
775     memcpy(&blkcfg, config, VIRTIO_BLK_CFG_SIZE);
776     QEMU_BUILD_BUG_ON(VIRTIO_BLK_CFG_SIZE > sizeof(blkcfg));
777 
778     aio_context_acquire(blk_get_aio_context(s->blk));
779     blk_set_enable_write_cache(s->blk, blkcfg.wce != 0);
780     aio_context_release(blk_get_aio_context(s->blk));
781 }
782 
783 static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features,
784                                         Error **errp)
785 {
786     VirtIOBlock *s = VIRTIO_BLK(vdev);
787 
788     /* Firstly sync all virtio-blk possible supported features */
789     features |= s->host_features;
790 
791     virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
792     virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
793     virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
794     virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
795     if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) {
796         if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) {
797             error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0");
798             return 0;
799         }
800     } else {
801         virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT);
802         virtio_add_feature(&features, VIRTIO_BLK_F_SCSI);
803     }
804 
805     if (blk_enable_write_cache(s->blk)) {
806         virtio_add_feature(&features, VIRTIO_BLK_F_WCE);
807     }
808     if (blk_is_read_only(s->blk)) {
809         virtio_add_feature(&features, VIRTIO_BLK_F_RO);
810     }
811     if (s->conf.num_queues > 1) {
812         virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
813     }
814 
815     return features;
816 }
817 
818 static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status)
819 {
820     VirtIOBlock *s = VIRTIO_BLK(vdev);
821 
822     if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) {
823         assert(!s->dataplane_started);
824     }
825 
826     if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) {
827         return;
828     }
829 
830     /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send
831      * cache flushes.  Thus, the "auto writethrough" behavior is never
832      * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature.
833      * Leaving it enabled would break the following sequence:
834      *
835      *     Guest started with "-drive cache=writethrough"
836      *     Guest sets status to 0
837      *     Guest sets DRIVER bit in status field
838      *     Guest reads host features (WCE=0, CONFIG_WCE=1)
839      *     Guest writes guest features (WCE=0, CONFIG_WCE=1)
840      *     Guest writes 1 to the WCE configuration field (writeback mode)
841      *     Guest sets DRIVER_OK bit in status field
842      *
843      * s->blk would erroneously be placed in writethrough mode.
844      */
845     if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) {
846         aio_context_acquire(blk_get_aio_context(s->blk));
847         blk_set_enable_write_cache(s->blk,
848                                    virtio_vdev_has_feature(vdev,
849                                                            VIRTIO_BLK_F_WCE));
850         aio_context_release(blk_get_aio_context(s->blk));
851     }
852 }
853 
854 static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f)
855 {
856     VirtIOBlock *s = VIRTIO_BLK(vdev);
857     VirtIOBlockReq *req = s->rq;
858 
859     while (req) {
860         qemu_put_sbyte(f, 1);
861 
862         if (s->conf.num_queues > 1) {
863             qemu_put_be32(f, virtio_get_queue_index(req->vq));
864         }
865 
866         qemu_put_virtqueue_element(f, &req->elem);
867         req = req->next;
868     }
869     qemu_put_sbyte(f, 0);
870 }
871 
872 static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f,
873                                   int version_id)
874 {
875     VirtIOBlock *s = VIRTIO_BLK(vdev);
876 
877     while (qemu_get_sbyte(f)) {
878         unsigned nvqs = s->conf.num_queues;
879         unsigned vq_idx = 0;
880         VirtIOBlockReq *req;
881 
882         if (nvqs > 1) {
883             vq_idx = qemu_get_be32(f);
884 
885             if (vq_idx >= nvqs) {
886                 error_report("Invalid virtqueue index in request list: %#x",
887                              vq_idx);
888                 return -EINVAL;
889             }
890         }
891 
892         req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq));
893         virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req);
894         req->next = s->rq;
895         s->rq = req;
896     }
897 
898     return 0;
899 }
900 
901 static void virtio_blk_resize(void *opaque)
902 {
903     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
904 
905     virtio_notify_config(vdev);
906 }
907 
908 static const BlockDevOps virtio_block_ops = {
909     .resize_cb = virtio_blk_resize,
910 };
911 
912 static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
913 {
914     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
915     VirtIOBlock *s = VIRTIO_BLK(dev);
916     VirtIOBlkConf *conf = &s->conf;
917     Error *err = NULL;
918     unsigned i;
919 
920     if (!conf->conf.blk) {
921         error_setg(errp, "drive property not set");
922         return;
923     }
924     if (!blk_is_inserted(conf->conf.blk)) {
925         error_setg(errp, "Device needs media, but drive is empty");
926         return;
927     }
928     if (!conf->num_queues) {
929         error_setg(errp, "num-queues property must be larger than 0");
930         return;
931     }
932     if (!is_power_of_2(conf->queue_size) ||
933         conf->queue_size > VIRTQUEUE_MAX_SIZE) {
934         error_setg(errp, "invalid queue-size property (%" PRIu16 "), "
935                    "must be a power of 2 (max %d)",
936                    conf->queue_size, VIRTQUEUE_MAX_SIZE);
937         return;
938     }
939 
940     if (!blkconf_apply_backend_options(&conf->conf,
941                                        blk_is_read_only(conf->conf.blk), true,
942                                        errp)) {
943         return;
944     }
945     s->original_wce = blk_enable_write_cache(conf->conf.blk);
946     if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) {
947         return;
948     }
949 
950     blkconf_blocksizes(&conf->conf);
951 
952     if (conf->conf.logical_block_size >
953         conf->conf.physical_block_size) {
954         error_setg(errp,
955                    "logical_block_size > physical_block_size not supported");
956         return;
957     }
958 
959     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK, VIRTIO_BLK_CFG_SIZE);
960 
961     s->blk = conf->conf.blk;
962     s->rq = NULL;
963     s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1;
964 
965     for (i = 0; i < conf->num_queues; i++) {
966         virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output);
967     }
968     virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err);
969     if (err != NULL) {
970         error_propagate(errp, err);
971         virtio_cleanup(vdev);
972         return;
973     }
974 
975     s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
976     blk_set_dev_ops(s->blk, &virtio_block_ops, s);
977     blk_set_guest_block_size(s->blk, s->conf.conf.logical_block_size);
978 
979     blk_iostatus_enable(s->blk);
980 }
981 
982 static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp)
983 {
984     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
985     VirtIOBlock *s = VIRTIO_BLK(dev);
986 
987     virtio_blk_data_plane_destroy(s->dataplane);
988     s->dataplane = NULL;
989     qemu_del_vm_change_state_handler(s->change);
990     blockdev_mark_auto_del(s->blk);
991     virtio_cleanup(vdev);
992 }
993 
994 static void virtio_blk_instance_init(Object *obj)
995 {
996     VirtIOBlock *s = VIRTIO_BLK(obj);
997 
998     device_add_bootindex_property(obj, &s->conf.conf.bootindex,
999                                   "bootindex", "/disk@0,0",
1000                                   DEVICE(obj), NULL);
1001 }
1002 
1003 static const VMStateDescription vmstate_virtio_blk = {
1004     .name = "virtio-blk",
1005     .minimum_version_id = 2,
1006     .version_id = 2,
1007     .fields = (VMStateField[]) {
1008         VMSTATE_VIRTIO_DEVICE,
1009         VMSTATE_END_OF_LIST()
1010     },
1011 };
1012 
1013 static Property virtio_blk_properties[] = {
1014     DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf),
1015     DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf),
1016     DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf),
1017     DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial),
1018     DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features,
1019                       VIRTIO_BLK_F_CONFIG_WCE, true),
1020 #ifdef __linux__
1021     DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features,
1022                       VIRTIO_BLK_F_SCSI, false),
1023 #endif
1024     DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
1025                     true),
1026     DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 1),
1027     DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 128),
1028     DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD,
1029                      IOThread *),
1030     DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features,
1031                       VIRTIO_BLK_F_DISCARD, true),
1032     DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features,
1033                       VIRTIO_BLK_F_WRITE_ZEROES, true),
1034     DEFINE_PROP_END_OF_LIST(),
1035 };
1036 
1037 static void virtio_blk_class_init(ObjectClass *klass, void *data)
1038 {
1039     DeviceClass *dc = DEVICE_CLASS(klass);
1040     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1041 
1042     dc->props = virtio_blk_properties;
1043     dc->vmsd = &vmstate_virtio_blk;
1044     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
1045     vdc->realize = virtio_blk_device_realize;
1046     vdc->unrealize = virtio_blk_device_unrealize;
1047     vdc->get_config = virtio_blk_update_config;
1048     vdc->set_config = virtio_blk_set_config;
1049     vdc->get_features = virtio_blk_get_features;
1050     vdc->set_status = virtio_blk_set_status;
1051     vdc->reset = virtio_blk_reset;
1052     vdc->save = virtio_blk_save_device;
1053     vdc->load = virtio_blk_load_device;
1054     vdc->start_ioeventfd = virtio_blk_data_plane_start;
1055     vdc->stop_ioeventfd = virtio_blk_data_plane_stop;
1056 }
1057 
1058 static const TypeInfo virtio_blk_info = {
1059     .name = TYPE_VIRTIO_BLK,
1060     .parent = TYPE_VIRTIO_DEVICE,
1061     .instance_size = sizeof(VirtIOBlock),
1062     .instance_init = virtio_blk_instance_init,
1063     .class_init = virtio_blk_class_init,
1064 };
1065 
1066 static void virtio_register_types(void)
1067 {
1068     type_register_static(&virtio_blk_info);
1069 }
1070 
1071 type_init(virtio_register_types)
1072