xref: /openbmc/qemu/hw/block/vhost-user-blk.c (revision 3df4c288)
1 /*
2  * vhost-user-blk host device
3  *
4  * Copyright(C) 2017 Intel Corporation.
5  *
6  * Authors:
7  *  Changpeng Liu <changpeng.liu@intel.com>
8  *
9  * Largely based on the "vhost-user-scsi.c" and "vhost-scsi.c" implemented by:
10  * Felipe Franciosi <felipe@nutanix.com>
11  * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
12  * Nicholas Bellinger <nab@risingtidesystems.com>
13  *
14  * This work is licensed under the terms of the GNU LGPL, version 2 or later.
15  * See the COPYING.LIB file in the top-level directory.
16  *
17  */
18 
19 #include "qemu/osdep.h"
20 #include "qapi/error.h"
21 #include "qemu/error-report.h"
22 #include "qemu/cutils.h"
23 #include "hw/qdev-core.h"
24 #include "hw/qdev-properties.h"
25 #include "hw/qdev-properties-system.h"
26 #include "hw/virtio/virtio-blk-common.h"
27 #include "hw/virtio/vhost.h"
28 #include "hw/virtio/vhost-user-blk.h"
29 #include "hw/virtio/virtio.h"
30 #include "hw/virtio/virtio-bus.h"
31 #include "hw/virtio/virtio-access.h"
32 #include "sysemu/sysemu.h"
33 #include "sysemu/runstate.h"
34 
35 static const int user_feature_bits[] = {
36     VIRTIO_BLK_F_SIZE_MAX,
37     VIRTIO_BLK_F_SEG_MAX,
38     VIRTIO_BLK_F_GEOMETRY,
39     VIRTIO_BLK_F_BLK_SIZE,
40     VIRTIO_BLK_F_TOPOLOGY,
41     VIRTIO_BLK_F_MQ,
42     VIRTIO_BLK_F_RO,
43     VIRTIO_BLK_F_FLUSH,
44     VIRTIO_BLK_F_CONFIG_WCE,
45     VIRTIO_BLK_F_DISCARD,
46     VIRTIO_BLK_F_WRITE_ZEROES,
47     VIRTIO_F_VERSION_1,
48     VIRTIO_RING_F_INDIRECT_DESC,
49     VIRTIO_RING_F_EVENT_IDX,
50     VIRTIO_F_NOTIFY_ON_EMPTY,
51     VIRTIO_F_RING_PACKED,
52     VIRTIO_F_IOMMU_PLATFORM,
53     VIRTIO_F_RING_RESET,
54     VHOST_INVALID_FEATURE_BIT
55 };
56 
57 static void vhost_user_blk_event(void *opaque, QEMUChrEvent event);
58 
59 static void vhost_user_blk_update_config(VirtIODevice *vdev, uint8_t *config)
60 {
61     VHostUserBlk *s = VHOST_USER_BLK(vdev);
62 
63     /* Our num_queues overrides the device backend */
64     virtio_stw_p(vdev, &s->blkcfg.num_queues, s->num_queues);
65 
66     memcpy(config, &s->blkcfg, vdev->config_len);
67 }
68 
69 static void vhost_user_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
70 {
71     VHostUserBlk *s = VHOST_USER_BLK(vdev);
72     struct virtio_blk_config *blkcfg = (struct virtio_blk_config *)config;
73     int ret;
74 
75     if (blkcfg->wce == s->blkcfg.wce) {
76         return;
77     }
78 
79     ret = vhost_dev_set_config(&s->dev, &blkcfg->wce,
80                                offsetof(struct virtio_blk_config, wce),
81                                sizeof(blkcfg->wce),
82                                VHOST_SET_CONFIG_TYPE_FRONTEND);
83     if (ret) {
84         error_report("set device config space failed");
85         return;
86     }
87 
88     s->blkcfg.wce = blkcfg->wce;
89 }
90 
91 static int vhost_user_blk_handle_config_change(struct vhost_dev *dev)
92 {
93     int ret;
94     VirtIODevice *vdev = dev->vdev;
95     VHostUserBlk *s = VHOST_USER_BLK(dev->vdev);
96     Error *local_err = NULL;
97 
98     if (!dev->started) {
99         return 0;
100     }
101 
102     ret = vhost_dev_get_config(dev, (uint8_t *)&s->blkcfg,
103                                vdev->config_len, &local_err);
104     if (ret < 0) {
105         error_report_err(local_err);
106         return ret;
107     }
108 
109     memcpy(dev->vdev->config, &s->blkcfg, vdev->config_len);
110     virtio_notify_config(dev->vdev);
111 
112     return 0;
113 }
114 
115 const VhostDevConfigOps blk_ops = {
116     .vhost_dev_config_notifier = vhost_user_blk_handle_config_change,
117 };
118 
119 static int vhost_user_blk_start(VirtIODevice *vdev, Error **errp)
120 {
121     VHostUserBlk *s = VHOST_USER_BLK(vdev);
122     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev)));
123     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
124     int i, ret;
125 
126     if (!k->set_guest_notifiers) {
127         error_setg(errp, "binding does not support guest notifiers");
128         return -ENOSYS;
129     }
130 
131     ret = vhost_dev_enable_notifiers(&s->dev, vdev);
132     if (ret < 0) {
133         error_setg_errno(errp, -ret, "Error enabling host notifiers");
134         return ret;
135     }
136 
137     ret = k->set_guest_notifiers(qbus->parent, s->dev.nvqs, true);
138     if (ret < 0) {
139         error_setg_errno(errp, -ret, "Error binding guest notifier");
140         goto err_host_notifiers;
141     }
142 
143     s->dev.acked_features = vdev->guest_features;
144 
145     ret = vhost_dev_prepare_inflight(&s->dev, vdev);
146     if (ret < 0) {
147         error_setg_errno(errp, -ret, "Error setting inflight format");
148         goto err_guest_notifiers;
149     }
150 
151     if (!s->inflight->addr) {
152         ret = vhost_dev_get_inflight(&s->dev, s->queue_size, s->inflight);
153         if (ret < 0) {
154             error_setg_errno(errp, -ret, "Error getting inflight");
155             goto err_guest_notifiers;
156         }
157     }
158 
159     ret = vhost_dev_set_inflight(&s->dev, s->inflight);
160     if (ret < 0) {
161         error_setg_errno(errp, -ret, "Error setting inflight");
162         goto err_guest_notifiers;
163     }
164 
165     /* guest_notifier_mask/pending not used yet, so just unmask
166      * everything here. virtio-pci will do the right thing by
167      * enabling/disabling irqfd.
168      */
169     for (i = 0; i < s->dev.nvqs; i++) {
170         vhost_virtqueue_mask(&s->dev, vdev, i, false);
171     }
172 
173     s->dev.vq_index_end = s->dev.nvqs;
174     ret = vhost_dev_start(&s->dev, vdev, true);
175     if (ret < 0) {
176         error_setg_errno(errp, -ret, "Error starting vhost");
177         goto err_guest_notifiers;
178     }
179     s->started_vu = true;
180 
181     return ret;
182 
183 err_guest_notifiers:
184     for (i = 0; i < s->dev.nvqs; i++) {
185         vhost_virtqueue_mask(&s->dev, vdev, i, true);
186     }
187     k->set_guest_notifiers(qbus->parent, s->dev.nvqs, false);
188 err_host_notifiers:
189     vhost_dev_disable_notifiers(&s->dev, vdev);
190     return ret;
191 }
192 
193 static void vhost_user_blk_stop(VirtIODevice *vdev)
194 {
195     VHostUserBlk *s = VHOST_USER_BLK(vdev);
196     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev)));
197     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
198     int ret;
199 
200     if (!s->started_vu) {
201         return;
202     }
203     s->started_vu = false;
204 
205     if (!k->set_guest_notifiers) {
206         return;
207     }
208 
209     vhost_dev_stop(&s->dev, vdev, true);
210 
211     ret = k->set_guest_notifiers(qbus->parent, s->dev.nvqs, false);
212     if (ret < 0) {
213         error_report("vhost guest notifier cleanup failed: %d", ret);
214         return;
215     }
216 
217     vhost_dev_disable_notifiers(&s->dev, vdev);
218 }
219 
220 static void vhost_user_blk_set_status(VirtIODevice *vdev, uint8_t status)
221 {
222     VHostUserBlk *s = VHOST_USER_BLK(vdev);
223     bool should_start = virtio_device_should_start(vdev, status);
224     Error *local_err = NULL;
225     int ret;
226 
227     if (!s->connected) {
228         return;
229     }
230 
231     if (vhost_dev_is_started(&s->dev) == should_start) {
232         return;
233     }
234 
235     if (should_start) {
236         ret = vhost_user_blk_start(vdev, &local_err);
237         if (ret < 0) {
238             error_reportf_err(local_err, "vhost-user-blk: vhost start failed: ");
239             qemu_chr_fe_disconnect(&s->chardev);
240         }
241     } else {
242         vhost_user_blk_stop(vdev);
243     }
244 
245 }
246 
247 static uint64_t vhost_user_blk_get_features(VirtIODevice *vdev,
248                                             uint64_t features,
249                                             Error **errp)
250 {
251     VHostUserBlk *s = VHOST_USER_BLK(vdev);
252 
253     /* Turn on pre-defined features */
254     virtio_add_feature(&features, VIRTIO_BLK_F_SIZE_MAX);
255     virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
256     virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
257     virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
258     virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
259     virtio_add_feature(&features, VIRTIO_BLK_F_FLUSH);
260     virtio_add_feature(&features, VIRTIO_BLK_F_RO);
261 
262     if (s->num_queues > 1) {
263         virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
264     }
265 
266     return vhost_get_features(&s->dev, user_feature_bits, features);
267 }
268 
269 static void vhost_user_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
270 {
271     VHostUserBlk *s = VHOST_USER_BLK(vdev);
272     Error *local_err = NULL;
273     int i, ret;
274 
275     if (!vdev->start_on_kick) {
276         return;
277     }
278 
279     if (!s->connected) {
280         return;
281     }
282 
283     if (vhost_dev_is_started(&s->dev)) {
284         return;
285     }
286 
287     /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
288      * vhost here instead of waiting for .set_status().
289      */
290     ret = vhost_user_blk_start(vdev, &local_err);
291     if (ret < 0) {
292         error_reportf_err(local_err, "vhost-user-blk: vhost start failed: ");
293         qemu_chr_fe_disconnect(&s->chardev);
294         return;
295     }
296 
297     /* Kick right away to begin processing requests already in vring */
298     for (i = 0; i < s->dev.nvqs; i++) {
299         VirtQueue *kick_vq = virtio_get_queue(vdev, i);
300 
301         if (!virtio_queue_get_desc_addr(vdev, i)) {
302             continue;
303         }
304         event_notifier_set(virtio_queue_get_host_notifier(kick_vq));
305     }
306 }
307 
308 static void vhost_user_blk_reset(VirtIODevice *vdev)
309 {
310     VHostUserBlk *s = VHOST_USER_BLK(vdev);
311 
312     vhost_dev_free_inflight(s->inflight);
313 }
314 
315 static int vhost_user_blk_connect(DeviceState *dev, Error **errp)
316 {
317     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
318     VHostUserBlk *s = VHOST_USER_BLK(vdev);
319     int ret = 0;
320 
321     if (s->connected) {
322         return 0;
323     }
324 
325     s->dev.num_queues = s->num_queues;
326     s->dev.nvqs = s->num_queues;
327     s->dev.vqs = s->vhost_vqs;
328     s->dev.vq_index = 0;
329     s->dev.backend_features = 0;
330 
331     vhost_dev_set_config_notifier(&s->dev, &blk_ops);
332 
333     s->vhost_user.supports_config = true;
334     ret = vhost_dev_init(&s->dev, &s->vhost_user, VHOST_BACKEND_TYPE_USER, 0,
335                          errp);
336     if (ret < 0) {
337         return ret;
338     }
339 
340     s->connected = true;
341 
342     /* restore vhost state */
343     if (virtio_device_started(vdev, vdev->status)) {
344         ret = vhost_user_blk_start(vdev, errp);
345     }
346 
347     return ret;
348 }
349 
350 static void vhost_user_blk_disconnect(DeviceState *dev)
351 {
352     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
353     VHostUserBlk *s = VHOST_USER_BLK(vdev);
354 
355     if (!s->connected) {
356         return;
357     }
358     s->connected = false;
359 
360     vhost_user_blk_stop(vdev);
361 
362     vhost_dev_cleanup(&s->dev);
363 
364     /* Re-instate the event handler for new connections */
365     qemu_chr_fe_set_handlers(&s->chardev, NULL, NULL, vhost_user_blk_event,
366                              NULL, dev, NULL, true);
367 }
368 
369 static void vhost_user_blk_event(void *opaque, QEMUChrEvent event)
370 {
371     DeviceState *dev = opaque;
372     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
373     VHostUserBlk *s = VHOST_USER_BLK(vdev);
374     Error *local_err = NULL;
375 
376     switch (event) {
377     case CHR_EVENT_OPENED:
378         if (vhost_user_blk_connect(dev, &local_err) < 0) {
379             error_report_err(local_err);
380             qemu_chr_fe_disconnect(&s->chardev);
381             return;
382         }
383         break;
384     case CHR_EVENT_CLOSED:
385         /* defer close until later to avoid circular close */
386         vhost_user_async_close(dev, &s->chardev, &s->dev,
387                                vhost_user_blk_disconnect, vhost_user_blk_event);
388         break;
389     case CHR_EVENT_BREAK:
390     case CHR_EVENT_MUX_IN:
391     case CHR_EVENT_MUX_OUT:
392         /* Ignore */
393         break;
394     }
395 }
396 
397 static int vhost_user_blk_realize_connect(VHostUserBlk *s, Error **errp)
398 {
399     DeviceState *dev = DEVICE(s);
400     int ret;
401 
402     s->connected = false;
403 
404     ret = qemu_chr_fe_wait_connected(&s->chardev, errp);
405     if (ret < 0) {
406         return ret;
407     }
408 
409     ret = vhost_user_blk_connect(dev, errp);
410     if (ret < 0) {
411         qemu_chr_fe_disconnect(&s->chardev);
412         return ret;
413     }
414     assert(s->connected);
415 
416     ret = vhost_dev_get_config(&s->dev, (uint8_t *)&s->blkcfg,
417                                VIRTIO_DEVICE(s)->config_len, errp);
418     if (ret < 0) {
419         qemu_chr_fe_disconnect(&s->chardev);
420         vhost_dev_cleanup(&s->dev);
421         return ret;
422     }
423 
424     return 0;
425 }
426 
427 static void vhost_user_blk_device_realize(DeviceState *dev, Error **errp)
428 {
429     ERRP_GUARD();
430     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
431     VHostUserBlk *s = VHOST_USER_BLK(vdev);
432     size_t config_size;
433     int retries;
434     int i, ret;
435 
436     if (!s->chardev.chr) {
437         error_setg(errp, "chardev is mandatory");
438         return;
439     }
440 
441     if (s->num_queues == VHOST_USER_BLK_AUTO_NUM_QUEUES) {
442         s->num_queues = 1;
443     }
444     if (!s->num_queues || s->num_queues > VIRTIO_QUEUE_MAX) {
445         error_setg(errp, "invalid number of IO queues");
446         return;
447     }
448 
449     if (!s->queue_size) {
450         error_setg(errp, "queue size must be non-zero");
451         return;
452     }
453     if (s->queue_size > VIRTQUEUE_MAX_SIZE) {
454         error_setg(errp, "queue size must not exceed %d",
455                    VIRTQUEUE_MAX_SIZE);
456         return;
457     }
458 
459     if (!vhost_user_init(&s->vhost_user, &s->chardev, errp)) {
460         return;
461     }
462 
463     config_size = virtio_get_config_size(&virtio_blk_cfg_size_params,
464                                          vdev->host_features);
465     virtio_init(vdev, VIRTIO_ID_BLOCK, config_size);
466 
467     s->virtqs = g_new(VirtQueue *, s->num_queues);
468     for (i = 0; i < s->num_queues; i++) {
469         s->virtqs[i] = virtio_add_queue(vdev, s->queue_size,
470                                         vhost_user_blk_handle_output);
471     }
472 
473     s->inflight = g_new0(struct vhost_inflight, 1);
474     s->vhost_vqs = g_new0(struct vhost_virtqueue, s->num_queues);
475 
476     retries = VU_REALIZE_CONN_RETRIES;
477     assert(!*errp);
478     do {
479         if (*errp) {
480             error_prepend(errp, "Reconnecting after error: ");
481             error_report_err(*errp);
482             *errp = NULL;
483         }
484         ret = vhost_user_blk_realize_connect(s, errp);
485     } while (ret < 0 && retries--);
486 
487     if (ret < 0) {
488         goto virtio_err;
489     }
490 
491     /* we're fully initialized, now we can operate, so add the handler */
492     qemu_chr_fe_set_handlers(&s->chardev,  NULL, NULL,
493                              vhost_user_blk_event, NULL, (void *)dev,
494                              NULL, true);
495     return;
496 
497 virtio_err:
498     g_free(s->vhost_vqs);
499     s->vhost_vqs = NULL;
500     g_free(s->inflight);
501     s->inflight = NULL;
502     for (i = 0; i < s->num_queues; i++) {
503         virtio_delete_queue(s->virtqs[i]);
504     }
505     g_free(s->virtqs);
506     virtio_cleanup(vdev);
507     vhost_user_cleanup(&s->vhost_user);
508 }
509 
510 static void vhost_user_blk_device_unrealize(DeviceState *dev)
511 {
512     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
513     VHostUserBlk *s = VHOST_USER_BLK(dev);
514     int i;
515 
516     virtio_set_status(vdev, 0);
517     qemu_chr_fe_set_handlers(&s->chardev,  NULL, NULL, NULL,
518                              NULL, NULL, NULL, false);
519     vhost_dev_cleanup(&s->dev);
520     vhost_dev_free_inflight(s->inflight);
521     g_free(s->vhost_vqs);
522     s->vhost_vqs = NULL;
523     g_free(s->inflight);
524     s->inflight = NULL;
525 
526     for (i = 0; i < s->num_queues; i++) {
527         virtio_delete_queue(s->virtqs[i]);
528     }
529     g_free(s->virtqs);
530     virtio_cleanup(vdev);
531     vhost_user_cleanup(&s->vhost_user);
532 }
533 
534 static void vhost_user_blk_instance_init(Object *obj)
535 {
536     VHostUserBlk *s = VHOST_USER_BLK(obj);
537 
538     device_add_bootindex_property(obj, &s->bootindex, "bootindex",
539                                   "/disk@0,0", DEVICE(obj));
540 }
541 
542 static struct vhost_dev *vhost_user_blk_get_vhost(VirtIODevice *vdev)
543 {
544     VHostUserBlk *s = VHOST_USER_BLK(vdev);
545     return &s->dev;
546 }
547 
548 static const VMStateDescription vmstate_vhost_user_blk = {
549     .name = "vhost-user-blk",
550     .minimum_version_id = 1,
551     .version_id = 1,
552     .fields = (const VMStateField[]) {
553         VMSTATE_VIRTIO_DEVICE,
554         VMSTATE_END_OF_LIST()
555     },
556 };
557 
558 static Property vhost_user_blk_properties[] = {
559     DEFINE_PROP_CHR("chardev", VHostUserBlk, chardev),
560     DEFINE_PROP_UINT16("num-queues", VHostUserBlk, num_queues,
561                        VHOST_USER_BLK_AUTO_NUM_QUEUES),
562     DEFINE_PROP_UINT32("queue-size", VHostUserBlk, queue_size, 128),
563     DEFINE_PROP_BIT64("config-wce", VHostUserBlk, parent_obj.host_features,
564                       VIRTIO_BLK_F_CONFIG_WCE, true),
565     DEFINE_PROP_BIT64("discard", VHostUserBlk, parent_obj.host_features,
566                       VIRTIO_BLK_F_DISCARD, true),
567     DEFINE_PROP_BIT64("write-zeroes", VHostUserBlk, parent_obj.host_features,
568                       VIRTIO_BLK_F_WRITE_ZEROES, true),
569     DEFINE_PROP_END_OF_LIST(),
570 };
571 
572 static void vhost_user_blk_class_init(ObjectClass *klass, void *data)
573 {
574     DeviceClass *dc = DEVICE_CLASS(klass);
575     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
576 
577     device_class_set_props(dc, vhost_user_blk_properties);
578     dc->vmsd = &vmstate_vhost_user_blk;
579     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
580     vdc->realize = vhost_user_blk_device_realize;
581     vdc->unrealize = vhost_user_blk_device_unrealize;
582     vdc->get_config = vhost_user_blk_update_config;
583     vdc->set_config = vhost_user_blk_set_config;
584     vdc->get_features = vhost_user_blk_get_features;
585     vdc->set_status = vhost_user_blk_set_status;
586     vdc->reset = vhost_user_blk_reset;
587     vdc->get_vhost = vhost_user_blk_get_vhost;
588 }
589 
590 static const TypeInfo vhost_user_blk_info = {
591     .name = TYPE_VHOST_USER_BLK,
592     .parent = TYPE_VIRTIO_DEVICE,
593     .instance_size = sizeof(VHostUserBlk),
594     .instance_init = vhost_user_blk_instance_init,
595     .class_init = vhost_user_blk_class_init,
596 };
597 
598 static void virtio_register_types(void)
599 {
600     type_register_static(&vhost_user_blk_info);
601 }
602 
603 type_init(virtio_register_types)
604