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 "qom/object.h" 24 #include "hw/qdev-core.h" 25 #include "hw/qdev-properties.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 memcpy(config, &s->blkcfg, sizeof(struct virtio_blk_config)); 58 } 59 60 static void vhost_user_blk_set_config(VirtIODevice *vdev, const uint8_t *config) 61 { 62 VHostUserBlk *s = VHOST_USER_BLK(vdev); 63 struct virtio_blk_config *blkcfg = (struct virtio_blk_config *)config; 64 int ret; 65 66 if (blkcfg->wce == s->blkcfg.wce) { 67 return; 68 } 69 70 ret = vhost_dev_set_config(&s->dev, &blkcfg->wce, 71 offsetof(struct virtio_blk_config, wce), 72 sizeof(blkcfg->wce), 73 VHOST_SET_CONFIG_TYPE_MASTER); 74 if (ret) { 75 error_report("set device config space failed"); 76 return; 77 } 78 79 s->blkcfg.wce = blkcfg->wce; 80 } 81 82 static int vhost_user_blk_handle_config_change(struct vhost_dev *dev) 83 { 84 int ret; 85 struct virtio_blk_config blkcfg; 86 VHostUserBlk *s = VHOST_USER_BLK(dev->vdev); 87 88 ret = vhost_dev_get_config(dev, (uint8_t *)&blkcfg, 89 sizeof(struct virtio_blk_config)); 90 if (ret < 0) { 91 error_report("get config space failed"); 92 return -1; 93 } 94 95 /* valid for resize only */ 96 if (blkcfg.capacity != s->blkcfg.capacity) { 97 s->blkcfg.capacity = blkcfg.capacity; 98 memcpy(dev->vdev->config, &s->blkcfg, sizeof(struct virtio_blk_config)); 99 virtio_notify_config(dev->vdev); 100 } 101 102 return 0; 103 } 104 105 const VhostDevConfigOps blk_ops = { 106 .vhost_dev_config_notifier = vhost_user_blk_handle_config_change, 107 }; 108 109 static int vhost_user_blk_start(VirtIODevice *vdev) 110 { 111 VHostUserBlk *s = VHOST_USER_BLK(vdev); 112 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev))); 113 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 114 int i, ret; 115 116 if (!k->set_guest_notifiers) { 117 error_report("binding does not support guest notifiers"); 118 return -ENOSYS; 119 } 120 121 ret = vhost_dev_enable_notifiers(&s->dev, vdev); 122 if (ret < 0) { 123 error_report("Error enabling host notifiers: %d", -ret); 124 return ret; 125 } 126 127 ret = k->set_guest_notifiers(qbus->parent, s->dev.nvqs, true); 128 if (ret < 0) { 129 error_report("Error binding guest notifier: %d", -ret); 130 goto err_host_notifiers; 131 } 132 133 s->dev.acked_features = vdev->guest_features; 134 135 if (!s->inflight->addr) { 136 ret = vhost_dev_get_inflight(&s->dev, s->queue_size, s->inflight); 137 if (ret < 0) { 138 error_report("Error get inflight: %d", -ret); 139 goto err_guest_notifiers; 140 } 141 } 142 143 ret = vhost_dev_set_inflight(&s->dev, s->inflight); 144 if (ret < 0) { 145 error_report("Error set inflight: %d", -ret); 146 goto err_guest_notifiers; 147 } 148 149 ret = vhost_dev_start(&s->dev, vdev); 150 if (ret < 0) { 151 error_report("Error starting vhost: %d", -ret); 152 goto err_guest_notifiers; 153 } 154 155 /* guest_notifier_mask/pending not used yet, so just unmask 156 * everything here. virtio-pci will do the right thing by 157 * enabling/disabling irqfd. 158 */ 159 for (i = 0; i < s->dev.nvqs; i++) { 160 vhost_virtqueue_mask(&s->dev, vdev, i, false); 161 } 162 163 return ret; 164 165 err_guest_notifiers: 166 k->set_guest_notifiers(qbus->parent, s->dev.nvqs, false); 167 err_host_notifiers: 168 vhost_dev_disable_notifiers(&s->dev, vdev); 169 return ret; 170 } 171 172 static void vhost_user_blk_stop(VirtIODevice *vdev) 173 { 174 VHostUserBlk *s = VHOST_USER_BLK(vdev); 175 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev))); 176 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); 177 int ret; 178 179 if (!k->set_guest_notifiers) { 180 return; 181 } 182 183 vhost_dev_stop(&s->dev, vdev); 184 185 ret = k->set_guest_notifiers(qbus->parent, s->dev.nvqs, false); 186 if (ret < 0) { 187 error_report("vhost guest notifier cleanup failed: %d", ret); 188 return; 189 } 190 191 vhost_dev_disable_notifiers(&s->dev, vdev); 192 } 193 194 static void vhost_user_blk_set_status(VirtIODevice *vdev, uint8_t status) 195 { 196 VHostUserBlk *s = VHOST_USER_BLK(vdev); 197 bool should_start = virtio_device_started(vdev, status); 198 int ret; 199 200 if (!vdev->vm_running) { 201 should_start = false; 202 } 203 204 if (!s->connected) { 205 return; 206 } 207 208 if (s->dev.started == should_start) { 209 return; 210 } 211 212 if (should_start) { 213 ret = vhost_user_blk_start(vdev); 214 if (ret < 0) { 215 error_report("vhost-user-blk: vhost start failed: %s", 216 strerror(-ret)); 217 qemu_chr_fe_disconnect(&s->chardev); 218 } 219 } else { 220 vhost_user_blk_stop(vdev); 221 } 222 223 } 224 225 static uint64_t vhost_user_blk_get_features(VirtIODevice *vdev, 226 uint64_t features, 227 Error **errp) 228 { 229 VHostUserBlk *s = VHOST_USER_BLK(vdev); 230 231 /* Turn on pre-defined features */ 232 virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX); 233 virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY); 234 virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY); 235 virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE); 236 virtio_add_feature(&features, VIRTIO_BLK_F_FLUSH); 237 virtio_add_feature(&features, VIRTIO_BLK_F_RO); 238 virtio_add_feature(&features, VIRTIO_BLK_F_DISCARD); 239 virtio_add_feature(&features, VIRTIO_BLK_F_WRITE_ZEROES); 240 241 if (s->config_wce) { 242 virtio_add_feature(&features, VIRTIO_BLK_F_CONFIG_WCE); 243 } 244 if (s->num_queues > 1) { 245 virtio_add_feature(&features, VIRTIO_BLK_F_MQ); 246 } 247 248 return vhost_get_features(&s->dev, user_feature_bits, features); 249 } 250 251 static void vhost_user_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq) 252 { 253 VHostUserBlk *s = VHOST_USER_BLK(vdev); 254 int i, ret; 255 256 if (!vdev->start_on_kick) { 257 return; 258 } 259 260 if (!s->connected) { 261 return; 262 } 263 264 if (s->dev.started) { 265 return; 266 } 267 268 /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start 269 * vhost here instead of waiting for .set_status(). 270 */ 271 ret = vhost_user_blk_start(vdev); 272 if (ret < 0) { 273 error_report("vhost-user-blk: vhost start failed: %s", 274 strerror(-ret)); 275 qemu_chr_fe_disconnect(&s->chardev); 276 return; 277 } 278 279 /* Kick right away to begin processing requests already in vring */ 280 for (i = 0; i < s->dev.nvqs; i++) { 281 VirtQueue *kick_vq = virtio_get_queue(vdev, i); 282 283 if (!virtio_queue_get_desc_addr(vdev, i)) { 284 continue; 285 } 286 event_notifier_set(virtio_queue_get_host_notifier(kick_vq)); 287 } 288 } 289 290 static void vhost_user_blk_reset(VirtIODevice *vdev) 291 { 292 VHostUserBlk *s = VHOST_USER_BLK(vdev); 293 294 vhost_dev_free_inflight(s->inflight); 295 } 296 297 static int vhost_user_blk_connect(DeviceState *dev) 298 { 299 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 300 VHostUserBlk *s = VHOST_USER_BLK(vdev); 301 int ret = 0; 302 303 if (s->connected) { 304 return 0; 305 } 306 s->connected = true; 307 308 s->dev.nvqs = s->num_queues; 309 s->dev.vqs = s->vhost_vqs; 310 s->dev.vq_index = 0; 311 s->dev.backend_features = 0; 312 313 vhost_dev_set_config_notifier(&s->dev, &blk_ops); 314 315 ret = vhost_dev_init(&s->dev, &s->vhost_user, VHOST_BACKEND_TYPE_USER, 0); 316 if (ret < 0) { 317 error_report("vhost-user-blk: vhost initialization failed: %s", 318 strerror(-ret)); 319 return ret; 320 } 321 322 /* restore vhost state */ 323 if (virtio_device_started(vdev, vdev->status)) { 324 ret = vhost_user_blk_start(vdev); 325 if (ret < 0) { 326 error_report("vhost-user-blk: vhost start failed: %s", 327 strerror(-ret)); 328 return ret; 329 } 330 } 331 332 return 0; 333 } 334 335 static void vhost_user_blk_disconnect(DeviceState *dev) 336 { 337 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 338 VHostUserBlk *s = VHOST_USER_BLK(vdev); 339 340 if (!s->connected) { 341 return; 342 } 343 s->connected = false; 344 345 if (s->dev.started) { 346 vhost_user_blk_stop(vdev); 347 } 348 349 vhost_dev_cleanup(&s->dev); 350 } 351 352 static void vhost_user_blk_event(void *opaque, QEMUChrEvent event); 353 354 static void vhost_user_blk_chr_closed_bh(void *opaque) 355 { 356 DeviceState *dev = opaque; 357 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 358 VHostUserBlk *s = VHOST_USER_BLK(vdev); 359 360 vhost_user_blk_disconnect(dev); 361 qemu_chr_fe_set_handlers(&s->chardev, NULL, NULL, vhost_user_blk_event, 362 NULL, opaque, NULL, true); 363 } 364 365 static void vhost_user_blk_event(void *opaque, QEMUChrEvent event) 366 { 367 DeviceState *dev = opaque; 368 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 369 VHostUserBlk *s = VHOST_USER_BLK(vdev); 370 371 switch (event) { 372 case CHR_EVENT_OPENED: 373 if (vhost_user_blk_connect(dev) < 0) { 374 qemu_chr_fe_disconnect(&s->chardev); 375 return; 376 } 377 break; 378 case CHR_EVENT_CLOSED: 379 /* 380 * A close event may happen during a read/write, but vhost 381 * code assumes the vhost_dev remains setup, so delay the 382 * stop & clear. There are two possible paths to hit this 383 * disconnect event: 384 * 1. When VM is in the RUN_STATE_PRELAUNCH state. The 385 * vhost_user_blk_device_realize() is a caller. 386 * 2. In tha main loop phase after VM start. 387 * 388 * For p2 the disconnect event will be delayed. We can't 389 * do the same for p1, because we are not running the loop 390 * at this moment. So just skip this step and perform 391 * disconnect in the caller function. 392 * 393 * TODO: maybe it is a good idea to make the same fix 394 * for other vhost-user devices. 395 */ 396 if (runstate_is_running()) { 397 AioContext *ctx = qemu_get_current_aio_context(); 398 399 qemu_chr_fe_set_handlers(&s->chardev, NULL, NULL, NULL, NULL, 400 NULL, NULL, false); 401 aio_bh_schedule_oneshot(ctx, vhost_user_blk_chr_closed_bh, opaque); 402 } 403 break; 404 case CHR_EVENT_BREAK: 405 case CHR_EVENT_MUX_IN: 406 case CHR_EVENT_MUX_OUT: 407 /* Ignore */ 408 break; 409 } 410 } 411 412 static void vhost_user_blk_device_realize(DeviceState *dev, Error **errp) 413 { 414 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 415 VHostUserBlk *s = VHOST_USER_BLK(vdev); 416 Error *err = NULL; 417 int i, ret; 418 419 if (!s->chardev.chr) { 420 error_setg(errp, "vhost-user-blk: chardev is mandatory"); 421 return; 422 } 423 424 if (!s->num_queues || s->num_queues > VIRTIO_QUEUE_MAX) { 425 error_setg(errp, "vhost-user-blk: invalid number of IO queues"); 426 return; 427 } 428 429 if (!s->queue_size) { 430 error_setg(errp, "vhost-user-blk: queue size must be non-zero"); 431 return; 432 } 433 434 if (!vhost_user_init(&s->vhost_user, &s->chardev, errp)) { 435 return; 436 } 437 438 virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK, 439 sizeof(struct virtio_blk_config)); 440 441 s->virtqs = g_new(VirtQueue *, s->num_queues); 442 for (i = 0; i < s->num_queues; i++) { 443 s->virtqs[i] = virtio_add_queue(vdev, s->queue_size, 444 vhost_user_blk_handle_output); 445 } 446 447 s->inflight = g_new0(struct vhost_inflight, 1); 448 s->vhost_vqs = g_new0(struct vhost_virtqueue, s->num_queues); 449 s->connected = false; 450 451 qemu_chr_fe_set_handlers(&s->chardev, NULL, NULL, vhost_user_blk_event, 452 NULL, (void *)dev, NULL, true); 453 454 reconnect: 455 if (qemu_chr_fe_wait_connected(&s->chardev, &err) < 0) { 456 error_report_err(err); 457 goto virtio_err; 458 } 459 460 /* check whether vhost_user_blk_connect() failed or not */ 461 if (!s->connected) { 462 goto reconnect; 463 } 464 465 ret = vhost_dev_get_config(&s->dev, (uint8_t *)&s->blkcfg, 466 sizeof(struct virtio_blk_config)); 467 if (ret < 0) { 468 error_report("vhost-user-blk: get block config failed"); 469 goto reconnect; 470 } 471 472 if (s->blkcfg.num_queues != s->num_queues) { 473 s->blkcfg.num_queues = s->num_queues; 474 } 475 476 return; 477 478 virtio_err: 479 g_free(s->vhost_vqs); 480 s->vhost_vqs = NULL; 481 g_free(s->inflight); 482 s->inflight = NULL; 483 for (i = 0; i < s->num_queues; i++) { 484 virtio_delete_queue(s->virtqs[i]); 485 } 486 g_free(s->virtqs); 487 virtio_cleanup(vdev); 488 vhost_user_cleanup(&s->vhost_user); 489 } 490 491 static void vhost_user_blk_device_unrealize(DeviceState *dev) 492 { 493 VirtIODevice *vdev = VIRTIO_DEVICE(dev); 494 VHostUserBlk *s = VHOST_USER_BLK(dev); 495 int i; 496 497 virtio_set_status(vdev, 0); 498 qemu_chr_fe_set_handlers(&s->chardev, NULL, NULL, NULL, 499 NULL, NULL, NULL, false); 500 vhost_dev_cleanup(&s->dev); 501 vhost_dev_free_inflight(s->inflight); 502 g_free(s->vhost_vqs); 503 s->vhost_vqs = NULL; 504 g_free(s->inflight); 505 s->inflight = NULL; 506 507 for (i = 0; i < s->num_queues; i++) { 508 virtio_delete_queue(s->virtqs[i]); 509 } 510 g_free(s->virtqs); 511 virtio_cleanup(vdev); 512 vhost_user_cleanup(&s->vhost_user); 513 } 514 515 static void vhost_user_blk_instance_init(Object *obj) 516 { 517 VHostUserBlk *s = VHOST_USER_BLK(obj); 518 519 device_add_bootindex_property(obj, &s->bootindex, "bootindex", 520 "/disk@0,0", DEVICE(obj)); 521 } 522 523 static const VMStateDescription vmstate_vhost_user_blk = { 524 .name = "vhost-user-blk", 525 .minimum_version_id = 1, 526 .version_id = 1, 527 .fields = (VMStateField[]) { 528 VMSTATE_VIRTIO_DEVICE, 529 VMSTATE_END_OF_LIST() 530 }, 531 }; 532 533 static Property vhost_user_blk_properties[] = { 534 DEFINE_PROP_CHR("chardev", VHostUserBlk, chardev), 535 DEFINE_PROP_UINT16("num-queues", VHostUserBlk, num_queues, 1), 536 DEFINE_PROP_UINT32("queue-size", VHostUserBlk, queue_size, 128), 537 DEFINE_PROP_BIT("config-wce", VHostUserBlk, config_wce, 0, true), 538 DEFINE_PROP_END_OF_LIST(), 539 }; 540 541 static void vhost_user_blk_class_init(ObjectClass *klass, void *data) 542 { 543 DeviceClass *dc = DEVICE_CLASS(klass); 544 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); 545 546 device_class_set_props(dc, vhost_user_blk_properties); 547 dc->vmsd = &vmstate_vhost_user_blk; 548 set_bit(DEVICE_CATEGORY_STORAGE, dc->categories); 549 vdc->realize = vhost_user_blk_device_realize; 550 vdc->unrealize = vhost_user_blk_device_unrealize; 551 vdc->get_config = vhost_user_blk_update_config; 552 vdc->set_config = vhost_user_blk_set_config; 553 vdc->get_features = vhost_user_blk_get_features; 554 vdc->set_status = vhost_user_blk_set_status; 555 vdc->reset = vhost_user_blk_reset; 556 } 557 558 static const TypeInfo vhost_user_blk_info = { 559 .name = TYPE_VHOST_USER_BLK, 560 .parent = TYPE_VIRTIO_DEVICE, 561 .instance_size = sizeof(VHostUserBlk), 562 .instance_init = vhost_user_blk_instance_init, 563 .class_init = vhost_user_blk_class_init, 564 }; 565 566 static void virtio_register_types(void) 567 { 568 type_register_static(&vhost_user_blk_info); 569 } 570 571 type_init(virtio_register_types) 572