xref: /openbmc/qemu/hw/block/vhost-user-blk.c (revision fe1127da)
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/virtio/vhost.h"
26 #include "hw/virtio/vhost-user-blk.h"
27 #include "hw/virtio/virtio.h"
28 #include "hw/virtio/virtio-bus.h"
29 #include "hw/virtio/virtio-access.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/runstate.h"
32 
33 static const int user_feature_bits[] = {
34     VIRTIO_BLK_F_SIZE_MAX,
35     VIRTIO_BLK_F_SEG_MAX,
36     VIRTIO_BLK_F_GEOMETRY,
37     VIRTIO_BLK_F_BLK_SIZE,
38     VIRTIO_BLK_F_TOPOLOGY,
39     VIRTIO_BLK_F_MQ,
40     VIRTIO_BLK_F_RO,
41     VIRTIO_BLK_F_FLUSH,
42     VIRTIO_BLK_F_CONFIG_WCE,
43     VIRTIO_BLK_F_DISCARD,
44     VIRTIO_BLK_F_WRITE_ZEROES,
45     VIRTIO_F_VERSION_1,
46     VIRTIO_RING_F_INDIRECT_DESC,
47     VIRTIO_RING_F_EVENT_IDX,
48     VIRTIO_F_NOTIFY_ON_EMPTY,
49     VHOST_INVALID_FEATURE_BIT
50 };
51 
52 static void vhost_user_blk_update_config(VirtIODevice *vdev, uint8_t *config)
53 {
54     VHostUserBlk *s = VHOST_USER_BLK(vdev);
55 
56     memcpy(config, &s->blkcfg, sizeof(struct virtio_blk_config));
57 }
58 
59 static void vhost_user_blk_set_config(VirtIODevice *vdev, const uint8_t *config)
60 {
61     VHostUserBlk *s = VHOST_USER_BLK(vdev);
62     struct virtio_blk_config *blkcfg = (struct virtio_blk_config *)config;
63     int ret;
64 
65     if (blkcfg->wce == s->blkcfg.wce) {
66         return;
67     }
68 
69     ret = vhost_dev_set_config(&s->dev, &blkcfg->wce,
70                                offsetof(struct virtio_blk_config, wce),
71                                sizeof(blkcfg->wce),
72                                VHOST_SET_CONFIG_TYPE_MASTER);
73     if (ret) {
74         error_report("set device config space failed");
75         return;
76     }
77 
78     s->blkcfg.wce = blkcfg->wce;
79 }
80 
81 static int vhost_user_blk_handle_config_change(struct vhost_dev *dev)
82 {
83     int ret;
84     struct virtio_blk_config blkcfg;
85     VHostUserBlk *s = VHOST_USER_BLK(dev->vdev);
86 
87     ret = vhost_dev_get_config(dev, (uint8_t *)&blkcfg,
88                                sizeof(struct virtio_blk_config));
89     if (ret < 0) {
90         error_report("get config space failed");
91         return -1;
92     }
93 
94     /* valid for resize only */
95     if (blkcfg.capacity != s->blkcfg.capacity) {
96         s->blkcfg.capacity = blkcfg.capacity;
97         memcpy(dev->vdev->config, &s->blkcfg, sizeof(struct virtio_blk_config));
98         virtio_notify_config(dev->vdev);
99     }
100 
101     return 0;
102 }
103 
104 const VhostDevConfigOps blk_ops = {
105     .vhost_dev_config_notifier = vhost_user_blk_handle_config_change,
106 };
107 
108 static int vhost_user_blk_start(VirtIODevice *vdev)
109 {
110     VHostUserBlk *s = VHOST_USER_BLK(vdev);
111     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev)));
112     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
113     int i, ret;
114 
115     if (!k->set_guest_notifiers) {
116         error_report("binding does not support guest notifiers");
117         return -ENOSYS;
118     }
119 
120     ret = vhost_dev_enable_notifiers(&s->dev, vdev);
121     if (ret < 0) {
122         error_report("Error enabling host notifiers: %d", -ret);
123         return ret;
124     }
125 
126     ret = k->set_guest_notifiers(qbus->parent, s->dev.nvqs, true);
127     if (ret < 0) {
128         error_report("Error binding guest notifier: %d", -ret);
129         goto err_host_notifiers;
130     }
131 
132     s->dev.acked_features = vdev->guest_features;
133 
134     ret = vhost_dev_prepare_inflight(&s->dev, vdev);
135     if (ret < 0) {
136         error_report("Error set inflight format: %d", -ret);
137         goto err_guest_notifiers;
138     }
139 
140     if (!s->inflight->addr) {
141         ret = vhost_dev_get_inflight(&s->dev, s->queue_size, s->inflight);
142         if (ret < 0) {
143             error_report("Error get inflight: %d", -ret);
144             goto err_guest_notifiers;
145         }
146     }
147 
148     ret = vhost_dev_set_inflight(&s->dev, s->inflight);
149     if (ret < 0) {
150         error_report("Error set inflight: %d", -ret);
151         goto err_guest_notifiers;
152     }
153 
154     ret = vhost_dev_start(&s->dev, vdev);
155     if (ret < 0) {
156         error_report("Error starting vhost: %d", -ret);
157         goto err_guest_notifiers;
158     }
159     s->started_vu = true;
160 
161     /* guest_notifier_mask/pending not used yet, so just unmask
162      * everything here. virtio-pci will do the right thing by
163      * enabling/disabling irqfd.
164      */
165     for (i = 0; i < s->dev.nvqs; i++) {
166         vhost_virtqueue_mask(&s->dev, vdev, i, false);
167     }
168 
169     return ret;
170 
171 err_guest_notifiers:
172     k->set_guest_notifiers(qbus->parent, s->dev.nvqs, false);
173 err_host_notifiers:
174     vhost_dev_disable_notifiers(&s->dev, vdev);
175     return ret;
176 }
177 
178 static void vhost_user_blk_stop(VirtIODevice *vdev)
179 {
180     VHostUserBlk *s = VHOST_USER_BLK(vdev);
181     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev)));
182     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
183     int ret;
184 
185     if (!s->started_vu) {
186         return;
187     }
188     s->started_vu = false;
189 
190     if (!k->set_guest_notifiers) {
191         return;
192     }
193 
194     vhost_dev_stop(&s->dev, vdev);
195 
196     ret = k->set_guest_notifiers(qbus->parent, s->dev.nvqs, false);
197     if (ret < 0) {
198         error_report("vhost guest notifier cleanup failed: %d", ret);
199         return;
200     }
201 
202     vhost_dev_disable_notifiers(&s->dev, vdev);
203 }
204 
205 static void vhost_user_blk_set_status(VirtIODevice *vdev, uint8_t status)
206 {
207     VHostUserBlk *s = VHOST_USER_BLK(vdev);
208     bool should_start = virtio_device_started(vdev, status);
209     int ret;
210 
211     if (!vdev->vm_running) {
212         should_start = false;
213     }
214 
215     if (!s->connected) {
216         return;
217     }
218 
219     if (s->dev.started == should_start) {
220         return;
221     }
222 
223     if (should_start) {
224         ret = vhost_user_blk_start(vdev);
225         if (ret < 0) {
226             error_report("vhost-user-blk: vhost start failed: %s",
227                          strerror(-ret));
228             qemu_chr_fe_disconnect(&s->chardev);
229         }
230     } else {
231         vhost_user_blk_stop(vdev);
232     }
233 
234 }
235 
236 static uint64_t vhost_user_blk_get_features(VirtIODevice *vdev,
237                                             uint64_t features,
238                                             Error **errp)
239 {
240     VHostUserBlk *s = VHOST_USER_BLK(vdev);
241 
242     /* Turn on pre-defined features */
243     virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX);
244     virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY);
245     virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY);
246     virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE);
247     virtio_add_feature(&features, VIRTIO_BLK_F_FLUSH);
248     virtio_add_feature(&features, VIRTIO_BLK_F_RO);
249     virtio_add_feature(&features, VIRTIO_BLK_F_DISCARD);
250     virtio_add_feature(&features, VIRTIO_BLK_F_WRITE_ZEROES);
251 
252     if (s->config_wce) {
253         virtio_add_feature(&features, VIRTIO_BLK_F_CONFIG_WCE);
254     }
255     if (s->num_queues > 1) {
256         virtio_add_feature(&features, VIRTIO_BLK_F_MQ);
257     }
258 
259     return vhost_get_features(&s->dev, user_feature_bits, features);
260 }
261 
262 static void vhost_user_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
263 {
264     VHostUserBlk *s = VHOST_USER_BLK(vdev);
265     int i, ret;
266 
267     if (!vdev->start_on_kick) {
268         return;
269     }
270 
271     if (!s->connected) {
272         return;
273     }
274 
275     if (s->dev.started) {
276         return;
277     }
278 
279     /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start
280      * vhost here instead of waiting for .set_status().
281      */
282     ret = vhost_user_blk_start(vdev);
283     if (ret < 0) {
284         error_report("vhost-user-blk: vhost start failed: %s",
285                      strerror(-ret));
286         qemu_chr_fe_disconnect(&s->chardev);
287         return;
288     }
289 
290     /* Kick right away to begin processing requests already in vring */
291     for (i = 0; i < s->dev.nvqs; i++) {
292         VirtQueue *kick_vq = virtio_get_queue(vdev, i);
293 
294         if (!virtio_queue_get_desc_addr(vdev, i)) {
295             continue;
296         }
297         event_notifier_set(virtio_queue_get_host_notifier(kick_vq));
298     }
299 }
300 
301 static void vhost_user_blk_reset(VirtIODevice *vdev)
302 {
303     VHostUserBlk *s = VHOST_USER_BLK(vdev);
304 
305     vhost_dev_free_inflight(s->inflight);
306 }
307 
308 static int vhost_user_blk_connect(DeviceState *dev)
309 {
310     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
311     VHostUserBlk *s = VHOST_USER_BLK(vdev);
312     int ret = 0;
313 
314     if (s->connected) {
315         return 0;
316     }
317     s->connected = true;
318 
319     s->dev.nvqs = s->num_queues;
320     s->dev.vqs = s->vhost_vqs;
321     s->dev.vq_index = 0;
322     s->dev.backend_features = 0;
323 
324     vhost_dev_set_config_notifier(&s->dev, &blk_ops);
325 
326     ret = vhost_dev_init(&s->dev, &s->vhost_user, VHOST_BACKEND_TYPE_USER, 0);
327     if (ret < 0) {
328         error_report("vhost-user-blk: vhost initialization failed: %s",
329                      strerror(-ret));
330         return ret;
331     }
332 
333     /* restore vhost state */
334     if (virtio_device_started(vdev, vdev->status)) {
335         ret = vhost_user_blk_start(vdev);
336         if (ret < 0) {
337             error_report("vhost-user-blk: vhost start failed: %s",
338                          strerror(-ret));
339             return ret;
340         }
341     }
342 
343     return 0;
344 }
345 
346 static void vhost_user_blk_disconnect(DeviceState *dev)
347 {
348     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
349     VHostUserBlk *s = VHOST_USER_BLK(vdev);
350 
351     if (!s->connected) {
352         return;
353     }
354     s->connected = false;
355 
356     vhost_user_blk_stop(vdev);
357 
358     vhost_dev_cleanup(&s->dev);
359 }
360 
361 static void vhost_user_blk_event(void *opaque, QEMUChrEvent event);
362 
363 static void vhost_user_blk_chr_closed_bh(void *opaque)
364 {
365     DeviceState *dev = opaque;
366     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
367     VHostUserBlk *s = VHOST_USER_BLK(vdev);
368 
369     vhost_user_blk_disconnect(dev);
370     qemu_chr_fe_set_handlers(&s->chardev, NULL, NULL, vhost_user_blk_event,
371             NULL, opaque, NULL, true);
372 }
373 
374 static void vhost_user_blk_event(void *opaque, QEMUChrEvent event)
375 {
376     DeviceState *dev = opaque;
377     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
378     VHostUserBlk *s = VHOST_USER_BLK(vdev);
379 
380     switch (event) {
381     case CHR_EVENT_OPENED:
382         if (vhost_user_blk_connect(dev) < 0) {
383             qemu_chr_fe_disconnect(&s->chardev);
384             return;
385         }
386         break;
387     case CHR_EVENT_CLOSED:
388         /*
389          * A close event may happen during a read/write, but vhost
390          * code assumes the vhost_dev remains setup, so delay the
391          * stop & clear. There are two possible paths to hit this
392          * disconnect event:
393          * 1. When VM is in the RUN_STATE_PRELAUNCH state. The
394          * vhost_user_blk_device_realize() is a caller.
395          * 2. In tha main loop phase after VM start.
396          *
397          * For p2 the disconnect event will be delayed. We can't
398          * do the same for p1, because we are not running the loop
399          * at this moment. So just skip this step and perform
400          * disconnect in the caller function.
401          *
402          * TODO: maybe it is a good idea to make the same fix
403          * for other vhost-user devices.
404          */
405         if (runstate_is_running()) {
406             AioContext *ctx = qemu_get_current_aio_context();
407 
408             qemu_chr_fe_set_handlers(&s->chardev, NULL, NULL, NULL, NULL,
409                     NULL, NULL, false);
410             aio_bh_schedule_oneshot(ctx, vhost_user_blk_chr_closed_bh, opaque);
411         }
412 
413         /*
414          * Move vhost device to the stopped state. The vhost-user device
415          * will be clean up and disconnected in BH. This can be useful in
416          * the vhost migration code. If disconnect was caught there is an
417          * option for the general vhost code to get the dev state without
418          * knowing its type (in this case vhost-user).
419          */
420         s->dev.started = false;
421         break;
422     case CHR_EVENT_BREAK:
423     case CHR_EVENT_MUX_IN:
424     case CHR_EVENT_MUX_OUT:
425         /* Ignore */
426         break;
427     }
428 }
429 
430 static void vhost_user_blk_device_realize(DeviceState *dev, Error **errp)
431 {
432     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
433     VHostUserBlk *s = VHOST_USER_BLK(vdev);
434     Error *err = NULL;
435     int i, ret;
436 
437     if (!s->chardev.chr) {
438         error_setg(errp, "vhost-user-blk: chardev is mandatory");
439         return;
440     }
441 
442     if (s->num_queues == VHOST_USER_BLK_AUTO_NUM_QUEUES) {
443         s->num_queues = 1;
444     }
445     if (!s->num_queues || s->num_queues > VIRTIO_QUEUE_MAX) {
446         error_setg(errp, "vhost-user-blk: invalid number of IO queues");
447         return;
448     }
449 
450     if (!s->queue_size) {
451         error_setg(errp, "vhost-user-blk: queue size must be non-zero");
452         return;
453     }
454 
455     if (!vhost_user_init(&s->vhost_user, &s->chardev, errp)) {
456         return;
457     }
458 
459     virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK,
460                 sizeof(struct virtio_blk_config));
461 
462     s->virtqs = g_new(VirtQueue *, s->num_queues);
463     for (i = 0; i < s->num_queues; i++) {
464         s->virtqs[i] = virtio_add_queue(vdev, s->queue_size,
465                                         vhost_user_blk_handle_output);
466     }
467 
468     s->inflight = g_new0(struct vhost_inflight, 1);
469     s->vhost_vqs = g_new0(struct vhost_virtqueue, s->num_queues);
470     s->connected = false;
471 
472     qemu_chr_fe_set_handlers(&s->chardev,  NULL, NULL, vhost_user_blk_event,
473                              NULL, (void *)dev, NULL, true);
474 
475 reconnect:
476     if (qemu_chr_fe_wait_connected(&s->chardev, &err) < 0) {
477         error_report_err(err);
478         goto virtio_err;
479     }
480 
481     /* check whether vhost_user_blk_connect() failed or not */
482     if (!s->connected) {
483         goto reconnect;
484     }
485 
486     ret = vhost_dev_get_config(&s->dev, (uint8_t *)&s->blkcfg,
487                                sizeof(struct virtio_blk_config));
488     if (ret < 0) {
489         error_report("vhost-user-blk: get block config failed");
490         goto reconnect;
491     }
492 
493     if (s->blkcfg.num_queues != s->num_queues) {
494         s->blkcfg.num_queues = s->num_queues;
495     }
496 
497     return;
498 
499 virtio_err:
500     g_free(s->vhost_vqs);
501     s->vhost_vqs = NULL;
502     g_free(s->inflight);
503     s->inflight = NULL;
504     for (i = 0; i < s->num_queues; i++) {
505         virtio_delete_queue(s->virtqs[i]);
506     }
507     g_free(s->virtqs);
508     virtio_cleanup(vdev);
509     vhost_user_cleanup(&s->vhost_user);
510 }
511 
512 static void vhost_user_blk_device_unrealize(DeviceState *dev)
513 {
514     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
515     VHostUserBlk *s = VHOST_USER_BLK(dev);
516     int i;
517 
518     virtio_set_status(vdev, 0);
519     qemu_chr_fe_set_handlers(&s->chardev,  NULL, NULL, NULL,
520                              NULL, NULL, NULL, false);
521     vhost_dev_cleanup(&s->dev);
522     vhost_dev_free_inflight(s->inflight);
523     g_free(s->vhost_vqs);
524     s->vhost_vqs = NULL;
525     g_free(s->inflight);
526     s->inflight = NULL;
527 
528     for (i = 0; i < s->num_queues; i++) {
529         virtio_delete_queue(s->virtqs[i]);
530     }
531     g_free(s->virtqs);
532     virtio_cleanup(vdev);
533     vhost_user_cleanup(&s->vhost_user);
534 }
535 
536 static void vhost_user_blk_instance_init(Object *obj)
537 {
538     VHostUserBlk *s = VHOST_USER_BLK(obj);
539 
540     device_add_bootindex_property(obj, &s->bootindex, "bootindex",
541                                   "/disk@0,0", DEVICE(obj));
542 }
543 
544 static const VMStateDescription vmstate_vhost_user_blk = {
545     .name = "vhost-user-blk",
546     .minimum_version_id = 1,
547     .version_id = 1,
548     .fields = (VMStateField[]) {
549         VMSTATE_VIRTIO_DEVICE,
550         VMSTATE_END_OF_LIST()
551     },
552 };
553 
554 static Property vhost_user_blk_properties[] = {
555     DEFINE_PROP_CHR("chardev", VHostUserBlk, chardev),
556     DEFINE_PROP_UINT16("num-queues", VHostUserBlk, num_queues,
557                        VHOST_USER_BLK_AUTO_NUM_QUEUES),
558     DEFINE_PROP_UINT32("queue-size", VHostUserBlk, queue_size, 128),
559     DEFINE_PROP_BIT("config-wce", VHostUserBlk, config_wce, 0, true),
560     DEFINE_PROP_END_OF_LIST(),
561 };
562 
563 static void vhost_user_blk_class_init(ObjectClass *klass, void *data)
564 {
565     DeviceClass *dc = DEVICE_CLASS(klass);
566     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
567 
568     device_class_set_props(dc, vhost_user_blk_properties);
569     dc->vmsd = &vmstate_vhost_user_blk;
570     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
571     vdc->realize = vhost_user_blk_device_realize;
572     vdc->unrealize = vhost_user_blk_device_unrealize;
573     vdc->get_config = vhost_user_blk_update_config;
574     vdc->set_config = vhost_user_blk_set_config;
575     vdc->get_features = vhost_user_blk_get_features;
576     vdc->set_status = vhost_user_blk_set_status;
577     vdc->reset = vhost_user_blk_reset;
578 }
579 
580 static const TypeInfo vhost_user_blk_info = {
581     .name = TYPE_VHOST_USER_BLK,
582     .parent = TYPE_VIRTIO_DEVICE,
583     .instance_size = sizeof(VHostUserBlk),
584     .instance_init = vhost_user_blk_instance_init,
585     .class_init = vhost_user_blk_class_init,
586 };
587 
588 static void virtio_register_types(void)
589 {
590     type_register_static(&vhost_user_blk_info);
591 }
592 
593 type_init(virtio_register_types)
594