xref: /openbmc/qemu/hw/block/vhost-user-blk.c (revision d177892d)
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/vhost.h"
27 #include "hw/virtio/vhost-user-blk.h"
28 #include "hw/virtio/virtio.h"
29 #include "hw/virtio/virtio-bus.h"
30 #include "hw/virtio/virtio-access.h"
31 #include "sysemu/sysemu.h"
32 #include "sysemu/runstate.h"
33 
34 static const int user_feature_bits[] = {
35     VIRTIO_BLK_F_SIZE_MAX,
36     VIRTIO_BLK_F_SEG_MAX,
37     VIRTIO_BLK_F_GEOMETRY,
38     VIRTIO_BLK_F_BLK_SIZE,
39     VIRTIO_BLK_F_TOPOLOGY,
40     VIRTIO_BLK_F_MQ,
41     VIRTIO_BLK_F_RO,
42     VIRTIO_BLK_F_FLUSH,
43     VIRTIO_BLK_F_CONFIG_WCE,
44     VIRTIO_BLK_F_DISCARD,
45     VIRTIO_BLK_F_WRITE_ZEROES,
46     VIRTIO_F_VERSION_1,
47     VIRTIO_RING_F_INDIRECT_DESC,
48     VIRTIO_RING_F_EVENT_IDX,
49     VIRTIO_F_NOTIFY_ON_EMPTY,
50     VHOST_INVALID_FEATURE_BIT
51 };
52 
53 static void vhost_user_blk_update_config(VirtIODevice *vdev, uint8_t *config)
54 {
55     VHostUserBlk *s = VHOST_USER_BLK(vdev);
56 
57     /* Our num_queues overrides the device backend */
58     virtio_stw_p(vdev, &s->blkcfg.num_queues, s->num_queues);
59 
60     memcpy(config, &s->blkcfg, sizeof(struct virtio_blk_config));
61 }
62 
63 static void vhost_user_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
64 {
65     VHostUserBlk *s = VHOST_USER_BLK(vdev);
66     struct virtio_blk_config *blkcfg = (struct virtio_blk_config *)config;
67     int ret;
68 
69     if (blkcfg->wce == s->blkcfg.wce) {
70         return;
71     }
72 
73     ret = vhost_dev_set_config(&s->dev, &blkcfg->wce,
74                                offsetof(struct virtio_blk_config, wce),
75                                sizeof(blkcfg->wce),
76                                VHOST_SET_CONFIG_TYPE_MASTER);
77     if (ret) {
78         error_report("set device config space failed");
79         return;
80     }
81 
82     s->blkcfg.wce = blkcfg->wce;
83 }
84 
85 static int vhost_user_blk_handle_config_change(struct vhost_dev *dev)
86 {
87     int ret;
88     struct virtio_blk_config blkcfg;
89     VHostUserBlk *s = VHOST_USER_BLK(dev->vdev);
90 
91     ret = vhost_dev_get_config(dev, (uint8_t *)&blkcfg,
92                                sizeof(struct virtio_blk_config));
93     if (ret < 0) {
94         error_report("get config space failed");
95         return -1;
96     }
97 
98     /* valid for resize only */
99     if (blkcfg.capacity != s->blkcfg.capacity) {
100         s->blkcfg.capacity = blkcfg.capacity;
101         memcpy(dev->vdev->config, &s->blkcfg, sizeof(struct virtio_blk_config));
102         virtio_notify_config(dev->vdev);
103     }
104 
105     return 0;
106 }
107 
108 const VhostDevConfigOps blk_ops = {
109     .vhost_dev_config_notifier = vhost_user_blk_handle_config_change,
110 };
111 
112 static int vhost_user_blk_start(VirtIODevice *vdev)
113 {
114     VHostUserBlk *s = VHOST_USER_BLK(vdev);
115     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev)));
116     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
117     int i, ret;
118 
119     if (!k->set_guest_notifiers) {
120         error_report("binding does not support guest notifiers");
121         return -ENOSYS;
122     }
123 
124     ret = vhost_dev_enable_notifiers(&s->dev, vdev);
125     if (ret < 0) {
126         error_report("Error enabling host notifiers: %d", -ret);
127         return ret;
128     }
129 
130     ret = k->set_guest_notifiers(qbus->parent, s->dev.nvqs, true);
131     if (ret < 0) {
132         error_report("Error binding guest notifier: %d", -ret);
133         goto err_host_notifiers;
134     }
135 
136     s->dev.acked_features = vdev->guest_features;
137 
138     ret = vhost_dev_prepare_inflight(&s->dev, vdev);
139     if (ret < 0) {
140         error_report("Error set inflight format: %d", -ret);
141         goto err_guest_notifiers;
142     }
143 
144     if (!s->inflight->addr) {
145         ret = vhost_dev_get_inflight(&s->dev, s->queue_size, s->inflight);
146         if (ret < 0) {
147             error_report("Error get inflight: %d", -ret);
148             goto err_guest_notifiers;
149         }
150     }
151 
152     ret = vhost_dev_set_inflight(&s->dev, s->inflight);
153     if (ret < 0) {
154         error_report("Error set inflight: %d", -ret);
155         goto err_guest_notifiers;
156     }
157 
158     ret = vhost_dev_start(&s->dev, vdev);
159     if (ret < 0) {
160         error_report("Error starting vhost: %d", -ret);
161         goto err_guest_notifiers;
162     }
163     s->started_vu = true;
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     return ret;
174 
175 err_guest_notifiers:
176     k->set_guest_notifiers(qbus->parent, s->dev.nvqs, false);
177 err_host_notifiers:
178     vhost_dev_disable_notifiers(&s->dev, vdev);
179     return ret;
180 }
181 
182 static void vhost_user_blk_stop(VirtIODevice *vdev)
183 {
184     VHostUserBlk *s = VHOST_USER_BLK(vdev);
185     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev)));
186     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
187     int ret;
188 
189     if (!s->started_vu) {
190         return;
191     }
192     s->started_vu = false;
193 
194     if (!k->set_guest_notifiers) {
195         return;
196     }
197 
198     vhost_dev_stop(&s->dev, vdev);
199 
200     ret = k->set_guest_notifiers(qbus->parent, s->dev.nvqs, false);
201     if (ret < 0) {
202         error_report("vhost guest notifier cleanup failed: %d", ret);
203         return;
204     }
205 
206     vhost_dev_disable_notifiers(&s->dev, vdev);
207 }
208 
209 static void vhost_user_blk_set_status(VirtIODevice *vdev, uint8_t status)
210 {
211     VHostUserBlk *s = VHOST_USER_BLK(vdev);
212     bool should_start = virtio_device_started(vdev, status);
213     int ret;
214 
215     if (!vdev->vm_running) {
216         should_start = false;
217     }
218 
219     if (!s->connected) {
220         return;
221     }
222 
223     if (s->dev.started == should_start) {
224         return;
225     }
226 
227     if (should_start) {
228         ret = vhost_user_blk_start(vdev);
229         if (ret < 0) {
230             error_report("vhost-user-blk: vhost start failed: %s",
231                          strerror(-ret));
232             qemu_chr_fe_disconnect(&s->chardev);
233         }
234     } else {
235         vhost_user_blk_stop(vdev);
236     }
237 
238 }
239 
240 static uint64_t vhost_user_blk_get_features(VirtIODevice *vdev,
241                                             uint64_t features,
242                                             Error **errp)
243 {
244     VHostUserBlk *s = VHOST_USER_BLK(vdev);
245 
246     /* Turn on pre-defined features */
247     virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
248     virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
249     virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
250     virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
251     virtio_add_feature(&features, VIRTIO_BLK_F_FLUSH);
252     virtio_add_feature(&features, VIRTIO_BLK_F_RO);
253     virtio_add_feature(&features, VIRTIO_BLK_F_DISCARD);
254     virtio_add_feature(&features, VIRTIO_BLK_F_WRITE_ZEROES);
255 
256     if (s->config_wce) {
257         virtio_add_feature(&features, VIRTIO_BLK_F_CONFIG_WCE);
258     }
259     if (s->num_queues > 1) {
260         virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
261     }
262 
263     return vhost_get_features(&s->dev, user_feature_bits, features);
264 }
265 
266 static void vhost_user_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
267 {
268     VHostUserBlk *s = VHOST_USER_BLK(vdev);
269     int i, ret;
270 
271     if (!vdev->start_on_kick) {
272         return;
273     }
274 
275     if (!s->connected) {
276         return;
277     }
278 
279     if (s->dev.started) {
280         return;
281     }
282 
283     /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
284      * vhost here instead of waiting for .set_status().
285      */
286     ret = vhost_user_blk_start(vdev);
287     if (ret < 0) {
288         error_report("vhost-user-blk: vhost start failed: %s",
289                      strerror(-ret));
290         qemu_chr_fe_disconnect(&s->chardev);
291         return;
292     }
293 
294     /* Kick right away to begin processing requests already in vring */
295     for (i = 0; i < s->dev.nvqs; i++) {
296         VirtQueue *kick_vq = virtio_get_queue(vdev, i);
297 
298         if (!virtio_queue_get_desc_addr(vdev, i)) {
299             continue;
300         }
301         event_notifier_set(virtio_queue_get_host_notifier(kick_vq));
302     }
303 }
304 
305 static void vhost_user_blk_reset(VirtIODevice *vdev)
306 {
307     VHostUserBlk *s = VHOST_USER_BLK(vdev);
308 
309     vhost_dev_free_inflight(s->inflight);
310 }
311 
312 static int vhost_user_blk_connect(DeviceState *dev)
313 {
314     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
315     VHostUserBlk *s = VHOST_USER_BLK(vdev);
316     int ret = 0;
317 
318     if (s->connected) {
319         return 0;
320     }
321     s->connected = true;
322 
323     s->dev.nvqs = s->num_queues;
324     s->dev.vqs = s->vhost_vqs;
325     s->dev.vq_index = 0;
326     s->dev.backend_features = 0;
327 
328     vhost_dev_set_config_notifier(&s->dev, &blk_ops);
329 
330     ret = vhost_dev_init(&s->dev, &s->vhost_user, VHOST_BACKEND_TYPE_USER, 0);
331     if (ret < 0) {
332         error_report("vhost-user-blk: vhost initialization failed: %s",
333                      strerror(-ret));
334         return ret;
335     }
336 
337     /* restore vhost state */
338     if (virtio_device_started(vdev, vdev->status)) {
339         ret = vhost_user_blk_start(vdev);
340         if (ret < 0) {
341             error_report("vhost-user-blk: vhost start failed: %s",
342                          strerror(-ret));
343             return ret;
344         }
345     }
346 
347     return 0;
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 
365 static void vhost_user_blk_event(void *opaque, QEMUChrEvent event,
366                                  bool realized);
367 
368 static void vhost_user_blk_event_realize(void *opaque, QEMUChrEvent event)
369 {
370     vhost_user_blk_event(opaque, event, false);
371 }
372 
373 static void vhost_user_blk_event_oper(void *opaque, QEMUChrEvent event)
374 {
375     vhost_user_blk_event(opaque, event, true);
376 }
377 
378 static void vhost_user_blk_chr_closed_bh(void *opaque)
379 {
380     DeviceState *dev = opaque;
381     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
382     VHostUserBlk *s = VHOST_USER_BLK(vdev);
383 
384     vhost_user_blk_disconnect(dev);
385     qemu_chr_fe_set_handlers(&s->chardev, NULL, NULL,
386             vhost_user_blk_event_oper, NULL, opaque, NULL, true);
387 }
388 
389 static void vhost_user_blk_event(void *opaque, QEMUChrEvent event,
390                                  bool realized)
391 {
392     DeviceState *dev = opaque;
393     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
394     VHostUserBlk *s = VHOST_USER_BLK(vdev);
395 
396     switch (event) {
397     case CHR_EVENT_OPENED:
398         if (vhost_user_blk_connect(dev) < 0) {
399             qemu_chr_fe_disconnect(&s->chardev);
400             return;
401         }
402         break;
403     case CHR_EVENT_CLOSED:
404         /*
405          * Closing the connection should happen differently on device
406          * initialization and operation stages.
407          * On initalization, we want to re-start vhost_dev initialization
408          * from the very beginning right away when the connection is closed,
409          * so we clean up vhost_dev on each connection closing.
410          * On operation, we want to postpone vhost_dev cleanup to let the
411          * other code perform its own cleanup sequence using vhost_dev data
412          * (e.g. vhost_dev_set_log).
413          */
414         if (realized && !runstate_check(RUN_STATE_SHUTDOWN)) {
415             /*
416              * A close event may happen during a read/write, but vhost
417              * code assumes the vhost_dev remains setup, so delay the
418              * stop & clear.
419              */
420             AioContext *ctx = qemu_get_current_aio_context();
421 
422             qemu_chr_fe_set_handlers(&s->chardev, NULL, NULL, NULL, NULL,
423                     NULL, NULL, false);
424             aio_bh_schedule_oneshot(ctx, vhost_user_blk_chr_closed_bh, opaque);
425 
426             /*
427              * Move vhost device to the stopped state. The vhost-user device
428              * will be clean up and disconnected in BH. This can be useful in
429              * the vhost migration code. If disconnect was caught there is an
430              * option for the general vhost code to get the dev state without
431              * knowing its type (in this case vhost-user).
432              */
433             s->dev.started = false;
434         } else {
435             vhost_user_blk_disconnect(dev);
436         }
437         break;
438     case CHR_EVENT_BREAK:
439     case CHR_EVENT_MUX_IN:
440     case CHR_EVENT_MUX_OUT:
441         /* Ignore */
442         break;
443     }
444 }
445 
446 static void vhost_user_blk_device_realize(DeviceState *dev, Error **errp)
447 {
448     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
449     VHostUserBlk *s = VHOST_USER_BLK(vdev);
450     Error *err = NULL;
451     int i, ret;
452 
453     if (!s->chardev.chr) {
454         error_setg(errp, "vhost-user-blk: chardev is mandatory");
455         return;
456     }
457 
458     if (s->num_queues == VHOST_USER_BLK_AUTO_NUM_QUEUES) {
459         s->num_queues = 1;
460     }
461     if (!s->num_queues || s->num_queues > VIRTIO_QUEUE_MAX) {
462         error_setg(errp, "vhost-user-blk: invalid number of IO queues");
463         return;
464     }
465 
466     if (!s->queue_size) {
467         error_setg(errp, "vhost-user-blk: queue size must be non-zero");
468         return;
469     }
470 
471     if (!vhost_user_init(&s->vhost_user, &s->chardev, errp)) {
472         return;
473     }
474 
475     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK,
476                 sizeof(struct virtio_blk_config));
477 
478     s->virtqs = g_new(VirtQueue *, s->num_queues);
479     for (i = 0; i < s->num_queues; i++) {
480         s->virtqs[i] = virtio_add_queue(vdev, s->queue_size,
481                                         vhost_user_blk_handle_output);
482     }
483 
484     s->inflight = g_new0(struct vhost_inflight, 1);
485     s->vhost_vqs = g_new0(struct vhost_virtqueue, s->num_queues);
486     s->connected = false;
487 
488     qemu_chr_fe_set_handlers(&s->chardev,  NULL, NULL,
489                              vhost_user_blk_event_realize, NULL, (void *)dev,
490                              NULL, true);
491 
492 reconnect:
493     if (qemu_chr_fe_wait_connected(&s->chardev, &err) < 0) {
494         error_report_err(err);
495         goto virtio_err;
496     }
497 
498     /* check whether vhost_user_blk_connect() failed or not */
499     if (!s->connected) {
500         goto reconnect;
501     }
502 
503     ret = vhost_dev_get_config(&s->dev, (uint8_t *)&s->blkcfg,
504                                sizeof(struct virtio_blk_config));
505     if (ret < 0) {
506         error_report("vhost-user-blk: get block config failed");
507         goto reconnect;
508     }
509 
510     /* we're fully initialized, now we can operate, so change the handler */
511     qemu_chr_fe_set_handlers(&s->chardev,  NULL, NULL,
512                              vhost_user_blk_event_oper, NULL, (void *)dev,
513                              NULL, true);
514     return;
515 
516 virtio_err:
517     g_free(s->vhost_vqs);
518     s->vhost_vqs = NULL;
519     g_free(s->inflight);
520     s->inflight = NULL;
521     for (i = 0; i < s->num_queues; i++) {
522         virtio_delete_queue(s->virtqs[i]);
523     }
524     g_free(s->virtqs);
525     virtio_cleanup(vdev);
526     vhost_user_cleanup(&s->vhost_user);
527 }
528 
529 static void vhost_user_blk_device_unrealize(DeviceState *dev)
530 {
531     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
532     VHostUserBlk *s = VHOST_USER_BLK(dev);
533     int i;
534 
535     virtio_set_status(vdev, 0);
536     qemu_chr_fe_set_handlers(&s->chardev,  NULL, NULL, NULL,
537                              NULL, NULL, NULL, false);
538     vhost_dev_cleanup(&s->dev);
539     vhost_dev_free_inflight(s->inflight);
540     g_free(s->vhost_vqs);
541     s->vhost_vqs = NULL;
542     g_free(s->inflight);
543     s->inflight = NULL;
544 
545     for (i = 0; i < s->num_queues; i++) {
546         virtio_delete_queue(s->virtqs[i]);
547     }
548     g_free(s->virtqs);
549     virtio_cleanup(vdev);
550     vhost_user_cleanup(&s->vhost_user);
551 }
552 
553 static void vhost_user_blk_instance_init(Object *obj)
554 {
555     VHostUserBlk *s = VHOST_USER_BLK(obj);
556 
557     device_add_bootindex_property(obj, &s->bootindex, "bootindex",
558                                   "/disk@0,0", DEVICE(obj));
559 }
560 
561 static const VMStateDescription vmstate_vhost_user_blk = {
562     .name = "vhost-user-blk",
563     .minimum_version_id = 1,
564     .version_id = 1,
565     .fields = (VMStateField[]) {
566         VMSTATE_VIRTIO_DEVICE,
567         VMSTATE_END_OF_LIST()
568     },
569 };
570 
571 static Property vhost_user_blk_properties[] = {
572     DEFINE_PROP_CHR("chardev", VHostUserBlk, chardev),
573     DEFINE_PROP_UINT16("num-queues", VHostUserBlk, num_queues,
574                        VHOST_USER_BLK_AUTO_NUM_QUEUES),
575     DEFINE_PROP_UINT32("queue-size", VHostUserBlk, queue_size, 128),
576     DEFINE_PROP_BIT("config-wce", VHostUserBlk, config_wce, 0, true),
577     DEFINE_PROP_END_OF_LIST(),
578 };
579 
580 static void vhost_user_blk_class_init(ObjectClass *klass, void *data)
581 {
582     DeviceClass *dc = DEVICE_CLASS(klass);
583     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
584 
585     device_class_set_props(dc, vhost_user_blk_properties);
586     dc->vmsd = &vmstate_vhost_user_blk;
587     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
588     vdc->realize = vhost_user_blk_device_realize;
589     vdc->unrealize = vhost_user_blk_device_unrealize;
590     vdc->get_config = vhost_user_blk_update_config;
591     vdc->set_config = vhost_user_blk_set_config;
592     vdc->get_features = vhost_user_blk_get_features;
593     vdc->set_status = vhost_user_blk_set_status;
594     vdc->reset = vhost_user_blk_reset;
595 }
596 
597 static const TypeInfo vhost_user_blk_info = {
598     .name = TYPE_VHOST_USER_BLK,
599     .parent = TYPE_VIRTIO_DEVICE,
600     .instance_size = sizeof(VHostUserBlk),
601     .instance_init = vhost_user_blk_instance_init,
602     .class_init = vhost_user_blk_class_init,
603 };
604 
605 static void virtio_register_types(void)
606 {
607     type_register_static(&vhost_user_blk_info);
608 }
609 
610 type_init(virtio_register_types)
611