| 1 | /* |
| 2 | * vhost_scsi host device |
| 3 | * |
| 4 | * Copyright IBM, Corp. 2011 |
| 5 | * |
| 6 | * Authors: |
| 7 | * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com> |
| 8 | * |
| 9 | * Changes for QEMU mainline + tcm_vhost kernel upstream: |
| 10 | * Nicholas Bellinger <nab@risingtidesystems.com> |
| 11 | * |
| 12 | * This work is licensed under the terms of the GNU LGPL, version 2 or later. |
| 13 | * See the COPYING.LIB file in the top-level directory. |
| 14 | * |
| 15 | */ |
| 16 | |
| 17 | #include "qemu/osdep.h" |
| 18 | #include "standard-headers/linux/vhost_types.h" |
| 19 | #include <linux/vhost.h> |
| 20 | #include <sys/ioctl.h> |
| 21 | #include "qapi/error.h" |
| 22 | #include "qemu/error-report.h" |
| 23 | #include "qemu/module.h" |
| 24 | #include "monitor/monitor.h" |
| 25 | #include "migration/blocker.h" |
| 26 | #include "hw/virtio/vhost-scsi.h" |
| 27 | #include "hw/virtio/vhost.h" |
| 28 | #include "hw/virtio/virtio-scsi.h" |
| 29 | #include "hw/virtio/virtio-bus.h" |
| 30 | #include "hw/core/fw-path-provider.h" |
| 31 | #include "hw/core/qdev-properties.h" |
| 32 | #include "qemu/cutils.h" |
| 33 | #include "system/system.h" |
| 34 | |
| 35 | /* Features supported by host kernel. */ |
| 36 | static const int kernel_feature_bits[] = { |
| 37 | VIRTIO_F_NOTIFY_ON_EMPTY, |
| 38 | VIRTIO_RING_F_INDIRECT_DESC, |
| 39 | VIRTIO_RING_F_EVENT_IDX, |
| 40 | VIRTIO_SCSI_F_HOTPLUG, |
| 41 | VIRTIO_F_RING_RESET, |
| 42 | VIRTIO_F_IN_ORDER, |
| 43 | VIRTIO_F_NOTIFICATION_DATA, |
| 44 | VHOST_INVALID_FEATURE_BIT |
| 45 | }; |
| 46 | |
| 47 | static int vhost_scsi_set_endpoint(VHostSCSI *s) |
| 48 | { |
| 49 | VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s); |
| 50 | VHostSCSICommon *vsc = VHOST_SCSI_COMMON(s); |
| 51 | const VhostOps *vhost_ops = vsc->dev.vhost_ops; |
| 52 | struct vhost_scsi_target backend; |
| 53 | int ret; |
| 54 | |
| 55 | memset(&backend, 0, sizeof(backend)); |
| 56 | pstrcpy(backend.vhost_wwpn, sizeof(backend.vhost_wwpn), vs->conf.wwpn); |
| 57 | ret = vhost_ops->vhost_scsi_set_endpoint(&vsc->dev, &backend); |
| 58 | if (ret < 0) { |
| 59 | return -errno; |
| 60 | } |
| 61 | return 0; |
| 62 | } |
| 63 | |
| 64 | static void vhost_scsi_clear_endpoint(VHostSCSI *s) |
| 65 | { |
| 66 | VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(s); |
| 67 | VHostSCSICommon *vsc = VHOST_SCSI_COMMON(s); |
| 68 | struct vhost_scsi_target backend; |
| 69 | const VhostOps *vhost_ops = vsc->dev.vhost_ops; |
| 70 | |
| 71 | memset(&backend, 0, sizeof(backend)); |
| 72 | pstrcpy(backend.vhost_wwpn, sizeof(backend.vhost_wwpn), vs->conf.wwpn); |
| 73 | vhost_ops->vhost_scsi_clear_endpoint(&vsc->dev, &backend); |
| 74 | } |
| 75 | |
| 76 | static int vhost_scsi_start(VHostSCSI *s) |
| 77 | { |
| 78 | int ret, abi_version; |
| 79 | VHostSCSICommon *vsc = VHOST_SCSI_COMMON(s); |
| 80 | const VhostOps *vhost_ops = vsc->dev.vhost_ops; |
| 81 | Error *local_err = NULL; |
| 82 | |
| 83 | ret = vhost_ops->vhost_scsi_get_abi_version(&vsc->dev, &abi_version); |
| 84 | if (ret < 0) { |
| 85 | return -errno; |
| 86 | } |
| 87 | if (abi_version > VHOST_SCSI_ABI_VERSION) { |
| 88 | error_report("vhost-scsi: The running tcm_vhost kernel abi_version:" |
| 89 | " %d is greater than vhost_scsi userspace supports: %d," |
| 90 | " please upgrade your version of QEMU", abi_version, |
| 91 | VHOST_SCSI_ABI_VERSION); |
| 92 | return -ENOSYS; |
| 93 | } |
| 94 | |
| 95 | ret = vhost_scsi_common_start(vsc, &local_err); |
| 96 | if (ret < 0) { |
| 97 | error_reportf_err(local_err, "Error starting vhost-scsi: "); |
| 98 | return ret; |
| 99 | } |
| 100 | |
| 101 | ret = vhost_scsi_set_endpoint(s); |
| 102 | if (ret < 0) { |
| 103 | error_report("Error setting vhost-scsi endpoint"); |
| 104 | vhost_scsi_common_stop(vsc); |
| 105 | } |
| 106 | |
| 107 | return ret; |
| 108 | } |
| 109 | |
| 110 | static void vhost_scsi_stop(VHostSCSI *s) |
| 111 | { |
| 112 | VHostSCSICommon *vsc = VHOST_SCSI_COMMON(s); |
| 113 | |
| 114 | vhost_scsi_clear_endpoint(s); |
| 115 | vhost_scsi_common_stop(vsc); |
| 116 | } |
| 117 | |
| 118 | static int vhost_scsi_set_status(VirtIODevice *vdev, uint8_t val) |
| 119 | { |
| 120 | VHostSCSI *s = VHOST_SCSI(vdev); |
| 121 | VHostSCSICommon *vsc = VHOST_SCSI_COMMON(s); |
| 122 | bool start = (val & VIRTIO_CONFIG_S_DRIVER_OK); |
| 123 | |
| 124 | if (!vdev->vm_running) { |
| 125 | start = false; |
| 126 | } |
| 127 | |
| 128 | if (vhost_dev_is_started(&vsc->dev) == start) { |
| 129 | return 0; |
| 130 | } |
| 131 | |
| 132 | if (start) { |
| 133 | int ret; |
| 134 | |
| 135 | ret = vhost_scsi_start(s); |
| 136 | if (ret < 0) { |
| 137 | error_report("unable to start vhost-scsi: %s", strerror(-ret)); |
| 138 | exit(1); |
| 139 | } |
| 140 | } else { |
| 141 | vhost_scsi_stop(s); |
| 142 | } |
| 143 | return 0; |
| 144 | } |
| 145 | |
| 146 | static void vhost_dummy_handle_output(VirtIODevice *vdev, VirtQueue *vq) |
| 147 | { |
| 148 | } |
| 149 | |
| 150 | static int vhost_scsi_pre_save(void *opaque) |
| 151 | { |
| 152 | VHostSCSICommon *vsc = opaque; |
| 153 | |
| 154 | /* At this point, backend must be stopped, otherwise |
| 155 | * it might keep writing to memory. */ |
| 156 | assert(!vhost_dev_is_started(&vsc->dev)); |
| 157 | |
| 158 | return 0; |
| 159 | } |
| 160 | |
| 161 | static const VMStateDescription vmstate_virtio_vhost_scsi = { |
| 162 | .name = "virtio-vhost_scsi", |
| 163 | .minimum_version_id = 1, |
| 164 | .version_id = 1, |
| 165 | .fields = (const VMStateField[]) { |
| 166 | VMSTATE_VIRTIO_DEVICE, |
| 167 | VMSTATE_END_OF_LIST() |
| 168 | }, |
| 169 | .pre_save = vhost_scsi_pre_save, |
| 170 | }; |
| 171 | |
| 172 | static int vhost_scsi_set_workers(VHostSCSICommon *vsc, bool per_virtqueue) |
| 173 | { |
| 174 | struct vhost_dev *dev = &vsc->dev; |
| 175 | struct vhost_vring_worker vq_worker; |
| 176 | struct vhost_worker_state worker; |
| 177 | int i, ret = 0; |
| 178 | |
| 179 | /* Use default worker */ |
| 180 | if (!per_virtqueue || dev->nvqs == VHOST_SCSI_VQ_NUM_FIXED + 1) { |
| 181 | return 0; |
| 182 | } |
| 183 | |
| 184 | /* |
| 185 | * ctl/evt share the first worker since it will be rare for them |
| 186 | * to send cmds while IO is running. |
| 187 | */ |
| 188 | for (i = VHOST_SCSI_VQ_NUM_FIXED + 1; i < dev->nvqs; i++) { |
| 189 | memset(&worker, 0, sizeof(worker)); |
| 190 | |
| 191 | ret = dev->vhost_ops->vhost_new_worker(dev, &worker); |
| 192 | if (ret == -ENOTTY) { |
| 193 | /* |
| 194 | * worker ioctls are not implemented so just ignore and |
| 195 | * and continue device setup. |
| 196 | */ |
| 197 | warn_report("vhost-scsi: Backend supports a single worker. " |
| 198 | "Ignoring worker_per_virtqueue=true setting."); |
| 199 | ret = 0; |
| 200 | break; |
| 201 | } else if (ret) { |
| 202 | break; |
| 203 | } |
| 204 | |
| 205 | memset(&vq_worker, 0, sizeof(vq_worker)); |
| 206 | vq_worker.worker_id = worker.worker_id; |
| 207 | vq_worker.index = i; |
| 208 | |
| 209 | ret = dev->vhost_ops->vhost_attach_vring_worker(dev, &vq_worker); |
| 210 | if (ret == -ENOTTY) { |
| 211 | /* |
| 212 | * It's a bug for the kernel to have supported the worker creation |
| 213 | * ioctl but not attach. |
| 214 | */ |
| 215 | dev->vhost_ops->vhost_free_worker(dev, &worker); |
| 216 | break; |
| 217 | } else if (ret) { |
| 218 | break; |
| 219 | } |
| 220 | } |
| 221 | |
| 222 | return ret; |
| 223 | } |
| 224 | |
| 225 | static void vhost_scsi_realize(DeviceState *dev, Error **errp) |
| 226 | { |
| 227 | ERRP_GUARD(); |
| 228 | VirtIOSCSICommon *vs = VIRTIO_SCSI_COMMON(dev); |
| 229 | VHostSCSICommon *vsc = VHOST_SCSI_COMMON(dev); |
| 230 | Error *err = NULL; |
| 231 | int vhostfd = -1; |
| 232 | int ret; |
| 233 | struct vhost_virtqueue *vqs = NULL; |
| 234 | |
| 235 | if (!vs->conf.wwpn) { |
| 236 | error_setg(errp, "vhost-scsi: missing wwpn"); |
| 237 | return; |
| 238 | } |
| 239 | |
| 240 | if (vs->conf.vhostfd) { |
| 241 | vhostfd = monitor_fd_param(monitor_cur(), vs->conf.vhostfd, errp); |
| 242 | if (vhostfd == -1) { |
| 243 | error_prepend(errp, "vhost-scsi: unable to parse vhostfd: "); |
| 244 | return; |
| 245 | } |
| 246 | } else { |
| 247 | vhostfd = open("/dev/vhost-scsi", O_RDWR); |
| 248 | if (vhostfd < 0) { |
| 249 | error_setg_file_open(errp, errno, "/dev/vhost-scsi"); |
| 250 | return; |
| 251 | } |
| 252 | } |
| 253 | |
| 254 | virtio_scsi_common_realize(dev, |
| 255 | vhost_dummy_handle_output, |
| 256 | vhost_dummy_handle_output, |
| 257 | vhost_dummy_handle_output, |
| 258 | &err); |
| 259 | if (err != NULL) { |
| 260 | error_propagate(errp, err); |
| 261 | goto close_fd; |
| 262 | } |
| 263 | |
| 264 | if (!vsc->migratable) { |
| 265 | error_setg(&vsc->migration_blocker, |
| 266 | "vhost-scsi does not support migration in all cases. " |
| 267 | "When external environment supports it (Orchestrator migrates " |
| 268 | "target SCSI device state or use shared storage over network), " |
| 269 | "set 'migratable' property to true to enable migration."); |
| 270 | if (migrate_add_blocker_normal(&vsc->migration_blocker, errp) < 0) { |
| 271 | goto free_virtio; |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | vsc->dev.nvqs = VHOST_SCSI_VQ_NUM_FIXED + vs->conf.num_queues; |
| 276 | vqs = g_new0(struct vhost_virtqueue, vsc->dev.nvqs); |
| 277 | vsc->dev.vqs = vqs; |
| 278 | vsc->dev.vq_index = 0; |
| 279 | |
| 280 | ret = vhost_dev_init(&vsc->dev, (void *)(uintptr_t)vhostfd, |
| 281 | VHOST_BACKEND_TYPE_KERNEL, 0, errp); |
| 282 | if (ret < 0) { |
| 283 | /* |
| 284 | * vhost_dev_init calls vhost_dev_cleanup on error, which closes |
| 285 | * vhostfd, don't double close it. |
| 286 | */ |
| 287 | vhostfd = -1; |
| 288 | goto free_vqs; |
| 289 | } |
| 290 | |
| 291 | ret = vhost_scsi_set_workers(vsc, vs->conf.worker_per_virtqueue); |
| 292 | if (ret < 0) { |
| 293 | error_setg(errp, "vhost-scsi: vhost worker setup failed: %s", |
| 294 | strerror(-ret)); |
| 295 | goto free_vqs; |
| 296 | } |
| 297 | |
| 298 | /* At present, channel and lun both are 0 for bootable vhost-scsi disk */ |
| 299 | vsc->channel = 0; |
| 300 | vsc->lun = 0; |
| 301 | /* Note: we can also get the minimum tpgt from kernel */ |
| 302 | vsc->target = vs->conf.boot_tpgt; |
| 303 | |
| 304 | return; |
| 305 | |
| 306 | free_vqs: |
| 307 | g_free(vqs); |
| 308 | if (!vsc->migratable) { |
| 309 | migrate_del_blocker(&vsc->migration_blocker); |
| 310 | } |
| 311 | free_virtio: |
| 312 | virtio_scsi_common_unrealize(dev); |
| 313 | close_fd: |
| 314 | if (vhostfd >= 0) { |
| 315 | close(vhostfd); |
| 316 | } |
| 317 | } |
| 318 | |
| 319 | static void vhost_scsi_unrealize(DeviceState *dev) |
| 320 | { |
| 321 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
| 322 | VHostSCSICommon *vsc = VHOST_SCSI_COMMON(dev); |
| 323 | struct vhost_virtqueue *vqs = vsc->dev.vqs; |
| 324 | |
| 325 | if (!vsc->migratable) { |
| 326 | migrate_del_blocker(&vsc->migration_blocker); |
| 327 | } |
| 328 | |
| 329 | /* This will stop vhost backend. */ |
| 330 | vhost_scsi_set_status(vdev, 0); |
| 331 | |
| 332 | vhost_dev_cleanup(&vsc->dev); |
| 333 | g_free(vqs); |
| 334 | |
| 335 | virtio_scsi_common_unrealize(dev); |
| 336 | } |
| 337 | |
| 338 | static struct vhost_dev *vhost_scsi_get_vhost(VirtIODevice *vdev) |
| 339 | { |
| 340 | VHostSCSI *s = VHOST_SCSI(vdev); |
| 341 | VHostSCSICommon *vsc = VHOST_SCSI_COMMON(s); |
| 342 | return &vsc->dev; |
| 343 | } |
| 344 | |
| 345 | static const Property vhost_scsi_properties[] = { |
| 346 | DEFINE_PROP_STRING("vhostfd", VirtIOSCSICommon, conf.vhostfd), |
| 347 | DEFINE_PROP_STRING("wwpn", VirtIOSCSICommon, conf.wwpn), |
| 348 | DEFINE_PROP_UINT32("boot_tpgt", VirtIOSCSICommon, conf.boot_tpgt, 0), |
| 349 | DEFINE_PROP_UINT32("num_queues", VirtIOSCSICommon, conf.num_queues, |
| 350 | VIRTIO_SCSI_AUTO_NUM_QUEUES), |
| 351 | DEFINE_PROP_UINT32("virtqueue_size", VirtIOSCSICommon, conf.virtqueue_size, |
| 352 | 128), |
| 353 | DEFINE_PROP_BOOL("seg_max_adjust", VirtIOSCSICommon, conf.seg_max_adjust, |
| 354 | true), |
| 355 | DEFINE_PROP_UINT32("max_sectors", VirtIOSCSICommon, conf.max_sectors, |
| 356 | 0xFFFF), |
| 357 | DEFINE_PROP_UINT32("cmd_per_lun", VirtIOSCSICommon, conf.cmd_per_lun, 128), |
| 358 | DEFINE_PROP_BIT64("t10_pi", VHostSCSICommon, host_features, |
| 359 | VIRTIO_SCSI_F_T10_PI, |
| 360 | false), |
| 361 | DEFINE_PROP_BIT64("hotplug", VHostSCSICommon, host_features, |
| 362 | VIRTIO_SCSI_F_HOTPLUG, |
| 363 | false), |
| 364 | DEFINE_PROP_BOOL("migratable", VHostSCSICommon, migratable, false), |
| 365 | DEFINE_PROP_BOOL("worker_per_virtqueue", VirtIOSCSICommon, |
| 366 | conf.worker_per_virtqueue, false), |
| 367 | }; |
| 368 | |
| 369 | static void vhost_scsi_class_init(ObjectClass *klass, const void *data) |
| 370 | { |
| 371 | DeviceClass *dc = DEVICE_CLASS(klass); |
| 372 | VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); |
| 373 | FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(klass); |
| 374 | |
| 375 | device_class_set_props(dc, vhost_scsi_properties); |
| 376 | dc->vmsd = &vmstate_virtio_vhost_scsi; |
| 377 | set_bit(DEVICE_CATEGORY_STORAGE, dc->categories); |
| 378 | vdc->realize = vhost_scsi_realize; |
| 379 | vdc->unrealize = vhost_scsi_unrealize; |
| 380 | vdc->get_features = vhost_scsi_common_get_features; |
| 381 | vdc->set_config = vhost_scsi_common_set_config; |
| 382 | vdc->set_status = vhost_scsi_set_status; |
| 383 | vdc->get_vhost = vhost_scsi_get_vhost; |
| 384 | fwc->get_dev_path = vhost_scsi_common_get_fw_dev_path; |
| 385 | } |
| 386 | |
| 387 | static void vhost_scsi_instance_init(Object *obj) |
| 388 | { |
| 389 | VHostSCSICommon *vsc = VHOST_SCSI_COMMON(obj); |
| 390 | |
| 391 | vsc->feature_bits = kernel_feature_bits; |
| 392 | |
| 393 | device_add_bootindex_property(obj, &vsc->bootindex, "bootindex", NULL, |
| 394 | DEVICE(vsc)); |
| 395 | } |
| 396 | |
| 397 | static const TypeInfo vhost_scsi_info = { |
| 398 | .name = TYPE_VHOST_SCSI, |
| 399 | .parent = TYPE_VHOST_SCSI_COMMON, |
| 400 | .instance_size = sizeof(VHostSCSI), |
| 401 | .class_init = vhost_scsi_class_init, |
| 402 | .instance_init = vhost_scsi_instance_init, |
| 403 | .interfaces = (const InterfaceInfo[]) { |
| 404 | { TYPE_FW_PATH_PROVIDER }, |
| 405 | { } |
| 406 | }, |
| 407 | }; |
| 408 | |
| 409 | static void virtio_register_types(void) |
| 410 | { |
| 411 | type_register_static(&vhost_scsi_info); |
| 412 | } |
| 413 | |
| 414 | type_init(virtio_register_types) |