| 1 | /* |
| 2 | * Vhost-user filesystem virtio device |
| 3 | * |
| 4 | * Copyright 2018-2019 Red Hat, Inc. |
| 5 | * |
| 6 | * Authors: |
| 7 | * Stefan Hajnoczi <stefanha@redhat.com> |
| 8 | * |
| 9 | * This work is licensed under the terms of the GNU GPL, version 2 or |
| 10 | * (at your option) any later version. See the COPYING file in the |
| 11 | * top-level directory. |
| 12 | */ |
| 13 | |
| 14 | #include "qemu/osdep.h" |
| 15 | #include <sys/ioctl.h> |
| 16 | #include "standard-headers/linux/virtio_fs.h" |
| 17 | #include "qapi/error.h" |
| 18 | #include "hw/core/qdev-properties.h" |
| 19 | #include "hw/core/qdev-properties-system.h" |
| 20 | #include "hw/virtio/virtio-bus.h" |
| 21 | #include "hw/virtio/virtio-access.h" |
| 22 | #include "qemu/error-report.h" |
| 23 | #include "hw/virtio/vhost.h" |
| 24 | #include "hw/virtio/vhost-user-fs.h" |
| 25 | #include "monitor/monitor.h" |
| 26 | #include "system/system.h" |
| 27 | |
| 28 | static const int user_feature_bits[] = { |
| 29 | VIRTIO_F_VERSION_1, |
| 30 | VIRTIO_RING_F_INDIRECT_DESC, |
| 31 | VIRTIO_RING_F_EVENT_IDX, |
| 32 | VIRTIO_F_NOTIFY_ON_EMPTY, |
| 33 | VIRTIO_F_RING_PACKED, |
| 34 | VIRTIO_F_IOMMU_PLATFORM, |
| 35 | VIRTIO_F_RING_RESET, |
| 36 | VIRTIO_F_IN_ORDER, |
| 37 | VIRTIO_F_NOTIFICATION_DATA, |
| 38 | VHOST_INVALID_FEATURE_BIT |
| 39 | }; |
| 40 | |
| 41 | static void vuf_get_config(VirtIODevice *vdev, uint8_t *config) |
| 42 | { |
| 43 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 44 | struct virtio_fs_config fscfg = {}; |
| 45 | |
| 46 | memcpy((char *)fscfg.tag, fs->conf.tag, |
| 47 | MIN(strlen(fs->conf.tag) + 1, sizeof(fscfg.tag))); |
| 48 | |
| 49 | virtio_stl_p(vdev, &fscfg.num_request_queues, fs->conf.num_request_queues); |
| 50 | |
| 51 | memcpy(config, &fscfg, sizeof(fscfg)); |
| 52 | } |
| 53 | |
| 54 | static void vuf_start(VirtIODevice *vdev) |
| 55 | { |
| 56 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 57 | BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev))); |
| 58 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); |
| 59 | int ret; |
| 60 | int i; |
| 61 | |
| 62 | if (!k->set_guest_notifiers) { |
| 63 | error_report("binding does not support guest notifiers"); |
| 64 | return; |
| 65 | } |
| 66 | |
| 67 | ret = vhost_dev_enable_notifiers(&fs->vhost_dev, vdev); |
| 68 | if (ret < 0) { |
| 69 | error_report("Error enabling host notifiers: %d", -ret); |
| 70 | return; |
| 71 | } |
| 72 | |
| 73 | ret = k->set_guest_notifiers(qbus->parent, fs->vhost_dev.nvqs, true); |
| 74 | if (ret < 0) { |
| 75 | error_report("Error binding guest notifier: %d", -ret); |
| 76 | goto err_host_notifiers; |
| 77 | } |
| 78 | |
| 79 | fs->vhost_dev.acked_features = vdev->guest_features; |
| 80 | ret = vhost_dev_start(&fs->vhost_dev, vdev, true); |
| 81 | if (ret < 0) { |
| 82 | error_report("Error starting vhost: %d", -ret); |
| 83 | goto err_guest_notifiers; |
| 84 | } |
| 85 | |
| 86 | /* |
| 87 | * guest_notifier_mask/pending not used yet, so just unmask |
| 88 | * everything here. virtio-pci will do the right thing by |
| 89 | * enabling/disabling irqfd. |
| 90 | */ |
| 91 | for (i = 0; i < fs->vhost_dev.nvqs; i++) { |
| 92 | vhost_virtqueue_mask(&fs->vhost_dev, vdev, i, false); |
| 93 | } |
| 94 | |
| 95 | return; |
| 96 | |
| 97 | err_guest_notifiers: |
| 98 | k->set_guest_notifiers(qbus->parent, fs->vhost_dev.nvqs, false); |
| 99 | err_host_notifiers: |
| 100 | vhost_dev_disable_notifiers(&fs->vhost_dev, vdev); |
| 101 | } |
| 102 | |
| 103 | static int vuf_stop(VirtIODevice *vdev) |
| 104 | { |
| 105 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 106 | BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(vdev))); |
| 107 | VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus); |
| 108 | int ret, err; |
| 109 | |
| 110 | if (!k->set_guest_notifiers) { |
| 111 | return 0; |
| 112 | } |
| 113 | |
| 114 | ret = vhost_dev_stop(&fs->vhost_dev, vdev, true); |
| 115 | |
| 116 | err = k->set_guest_notifiers(qbus->parent, fs->vhost_dev.nvqs, false); |
| 117 | if (err < 0) { |
| 118 | error_report("vhost guest notifier cleanup failed: %d", err); |
| 119 | return err; |
| 120 | } |
| 121 | |
| 122 | vhost_dev_disable_notifiers(&fs->vhost_dev, vdev); |
| 123 | return ret; |
| 124 | } |
| 125 | |
| 126 | static int vuf_set_status(VirtIODevice *vdev, uint8_t status) |
| 127 | { |
| 128 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 129 | bool should_start = virtio_device_should_start(vdev, status); |
| 130 | |
| 131 | if (vhost_dev_is_started(&fs->vhost_dev) == should_start) { |
| 132 | return 0; |
| 133 | } |
| 134 | |
| 135 | if (should_start) { |
| 136 | vuf_start(vdev); |
| 137 | } else { |
| 138 | int ret; |
| 139 | ret = vuf_stop(vdev); |
| 140 | if (ret < 0) { |
| 141 | return ret; |
| 142 | } |
| 143 | } |
| 144 | return 0; |
| 145 | } |
| 146 | |
| 147 | static uint64_t vuf_get_features(VirtIODevice *vdev, |
| 148 | uint64_t features, |
| 149 | Error **errp) |
| 150 | { |
| 151 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 152 | |
| 153 | return vhost_get_features(&fs->vhost_dev, user_feature_bits, features); |
| 154 | } |
| 155 | |
| 156 | static void vuf_handle_output(VirtIODevice *vdev, VirtQueue *vq) |
| 157 | { |
| 158 | /* |
| 159 | * Not normally called; it's the daemon that handles the queue; |
| 160 | * however virtio's cleanup path can call this. |
| 161 | */ |
| 162 | } |
| 163 | |
| 164 | static void vuf_guest_notifier_mask(VirtIODevice *vdev, int idx, |
| 165 | bool mask) |
| 166 | { |
| 167 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 168 | |
| 169 | /* |
| 170 | * Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1 |
| 171 | * as the macro of configure interrupt's IDX, If this driver does not |
| 172 | * support, the function will return |
| 173 | */ |
| 174 | |
| 175 | if (idx == VIRTIO_CONFIG_IRQ_IDX) { |
| 176 | return; |
| 177 | } |
| 178 | vhost_virtqueue_mask(&fs->vhost_dev, vdev, idx, mask); |
| 179 | } |
| 180 | |
| 181 | static bool vuf_guest_notifier_pending(VirtIODevice *vdev, int idx) |
| 182 | { |
| 183 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 184 | |
| 185 | /* |
| 186 | * Add the check for configure interrupt, Use VIRTIO_CONFIG_IRQ_IDX -1 |
| 187 | * as the macro of configure interrupt's IDX, If this driver does not |
| 188 | * support, the function will return |
| 189 | */ |
| 190 | |
| 191 | if (idx == VIRTIO_CONFIG_IRQ_IDX) { |
| 192 | return false; |
| 193 | } |
| 194 | return vhost_virtqueue_pending(&fs->vhost_dev, idx); |
| 195 | } |
| 196 | |
| 197 | static void vuf_device_realize(DeviceState *dev, Error **errp) |
| 198 | { |
| 199 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
| 200 | VHostUserFS *fs = VHOST_USER_FS(dev); |
| 201 | unsigned int i; |
| 202 | size_t len; |
| 203 | int ret; |
| 204 | |
| 205 | if (!fs->conf.chardev.chr) { |
| 206 | error_setg(errp, "missing chardev"); |
| 207 | return; |
| 208 | } |
| 209 | |
| 210 | if (!fs->conf.tag) { |
| 211 | error_setg(errp, "missing tag property"); |
| 212 | return; |
| 213 | } |
| 214 | len = strlen(fs->conf.tag); |
| 215 | if (len == 0) { |
| 216 | error_setg(errp, "tag property cannot be empty"); |
| 217 | return; |
| 218 | } |
| 219 | if (len > sizeof_field(struct virtio_fs_config, tag)) { |
| 220 | error_setg(errp, "tag property must be %zu bytes or less", |
| 221 | sizeof_field(struct virtio_fs_config, tag)); |
| 222 | return; |
| 223 | } |
| 224 | |
| 225 | if (fs->conf.num_request_queues == 0) { |
| 226 | error_setg(errp, "num-request-queues property must be larger than 0"); |
| 227 | return; |
| 228 | } |
| 229 | |
| 230 | if (!is_power_of_2(fs->conf.queue_size)) { |
| 231 | error_setg(errp, "queue-size property must be a power of 2"); |
| 232 | return; |
| 233 | } |
| 234 | |
| 235 | if (fs->conf.queue_size > VIRTQUEUE_MAX_SIZE) { |
| 236 | error_setg(errp, "queue-size property must be %u or smaller", |
| 237 | VIRTQUEUE_MAX_SIZE); |
| 238 | return; |
| 239 | } |
| 240 | |
| 241 | if (!vhost_user_init(&fs->vhost_user, &fs->conf.chardev, errp)) { |
| 242 | return; |
| 243 | } |
| 244 | |
| 245 | virtio_init(vdev, VIRTIO_ID_FS, sizeof(struct virtio_fs_config)); |
| 246 | |
| 247 | /* Hiprio queue */ |
| 248 | fs->hiprio_vq = virtio_add_queue(vdev, fs->conf.queue_size, vuf_handle_output); |
| 249 | |
| 250 | /* Request queues */ |
| 251 | fs->req_vqs = g_new(VirtQueue *, fs->conf.num_request_queues); |
| 252 | for (i = 0; i < fs->conf.num_request_queues; i++) { |
| 253 | fs->req_vqs[i] = virtio_add_queue(vdev, fs->conf.queue_size, vuf_handle_output); |
| 254 | } |
| 255 | |
| 256 | /* 1 high prio queue, plus the number configured */ |
| 257 | fs->vhost_dev.nvqs = 1 + fs->conf.num_request_queues; |
| 258 | fs->vhost_dev.vqs = g_new0(struct vhost_virtqueue, fs->vhost_dev.nvqs); |
| 259 | ret = vhost_dev_init(&fs->vhost_dev, &fs->vhost_user, |
| 260 | VHOST_BACKEND_TYPE_USER, 0, errp); |
| 261 | if (ret < 0) { |
| 262 | goto err_virtio; |
| 263 | } |
| 264 | |
| 265 | return; |
| 266 | |
| 267 | err_virtio: |
| 268 | vhost_user_cleanup(&fs->vhost_user); |
| 269 | virtio_delete_queue(fs->hiprio_vq); |
| 270 | for (i = 0; i < fs->conf.num_request_queues; i++) { |
| 271 | virtio_delete_queue(fs->req_vqs[i]); |
| 272 | } |
| 273 | g_free(fs->req_vqs); |
| 274 | virtio_cleanup(vdev); |
| 275 | g_free(fs->vhost_dev.vqs); |
| 276 | } |
| 277 | |
| 278 | static void vuf_device_unrealize(DeviceState *dev) |
| 279 | { |
| 280 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
| 281 | VHostUserFS *fs = VHOST_USER_FS(dev); |
| 282 | struct vhost_virtqueue *vhost_vqs = fs->vhost_dev.vqs; |
| 283 | int i; |
| 284 | |
| 285 | /* This will stop vhost backend if appropriate. */ |
| 286 | vuf_set_status(vdev, 0); |
| 287 | |
| 288 | vhost_dev_cleanup(&fs->vhost_dev); |
| 289 | |
| 290 | vhost_user_cleanup(&fs->vhost_user); |
| 291 | |
| 292 | virtio_delete_queue(fs->hiprio_vq); |
| 293 | for (i = 0; i < fs->conf.num_request_queues; i++) { |
| 294 | virtio_delete_queue(fs->req_vqs[i]); |
| 295 | } |
| 296 | g_free(fs->req_vqs); |
| 297 | virtio_cleanup(vdev); |
| 298 | g_free(vhost_vqs); |
| 299 | } |
| 300 | |
| 301 | static struct vhost_dev *vuf_get_vhost(VirtIODevice *vdev) |
| 302 | { |
| 303 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 304 | return &fs->vhost_dev; |
| 305 | } |
| 306 | |
| 307 | /** |
| 308 | * Fetch the internal state from virtiofsd and save it to `f`. |
| 309 | */ |
| 310 | static int vuf_save_state(QEMUFile *f, void *pv, size_t size, |
| 311 | const VMStateField *field, JSONWriter *vmdesc) |
| 312 | { |
| 313 | VirtIODevice *vdev = pv; |
| 314 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 315 | Error *local_error = NULL; |
| 316 | int ret; |
| 317 | |
| 318 | ret = vhost_save_backend_state(&fs->vhost_dev, f, &local_error); |
| 319 | if (ret < 0) { |
| 320 | error_reportf_err(local_error, |
| 321 | "Error saving back-end state of %s device %s " |
| 322 | "(tag: \"%s\"): ", |
| 323 | vdev->name, vdev->parent_obj.canonical_path, |
| 324 | fs->conf.tag ?: "<none>"); |
| 325 | return ret; |
| 326 | } |
| 327 | |
| 328 | return 0; |
| 329 | } |
| 330 | |
| 331 | /** |
| 332 | * Load virtiofsd's internal state from `f` and send it over to virtiofsd. |
| 333 | */ |
| 334 | static int vuf_load_state(QEMUFile *f, void *pv, size_t size, |
| 335 | const VMStateField *field) |
| 336 | { |
| 337 | VirtIODevice *vdev = pv; |
| 338 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 339 | Error *local_error = NULL; |
| 340 | int ret; |
| 341 | |
| 342 | ret = vhost_load_backend_state(&fs->vhost_dev, f, &local_error); |
| 343 | if (ret < 0) { |
| 344 | error_reportf_err(local_error, |
| 345 | "Error loading back-end state of %s device %s " |
| 346 | "(tag: \"%s\"): ", |
| 347 | vdev->name, vdev->parent_obj.canonical_path, |
| 348 | fs->conf.tag ?: "<none>"); |
| 349 | return ret; |
| 350 | } |
| 351 | |
| 352 | return 0; |
| 353 | } |
| 354 | |
| 355 | static bool vuf_is_internal_migration(void *opaque) |
| 356 | { |
| 357 | /* TODO: Return false when an external migration is requested */ |
| 358 | return true; |
| 359 | } |
| 360 | |
| 361 | static int vuf_check_migration_support(void *opaque) |
| 362 | { |
| 363 | VirtIODevice *vdev = opaque; |
| 364 | VHostUserFS *fs = VHOST_USER_FS(vdev); |
| 365 | |
| 366 | if (!vhost_supports_device_state(&fs->vhost_dev)) { |
| 367 | error_report("Back-end of %s device %s (tag: \"%s\") does not support " |
| 368 | "migration through qemu", |
| 369 | vdev->name, vdev->parent_obj.canonical_path, |
| 370 | fs->conf.tag ?: "<none>"); |
| 371 | return -ENOTSUP; |
| 372 | } |
| 373 | |
| 374 | return 0; |
| 375 | } |
| 376 | |
| 377 | static const VMStateDescription vuf_backend_vmstate; |
| 378 | |
| 379 | static const VMStateDescription vuf_vmstate = { |
| 380 | .name = "vhost-user-fs", |
| 381 | .version_id = 0, |
| 382 | .fields = (const VMStateField[]) { |
| 383 | VMSTATE_VIRTIO_DEVICE, |
| 384 | VMSTATE_END_OF_LIST() |
| 385 | }, |
| 386 | .subsections = (const VMStateDescription * const []) { |
| 387 | &vuf_backend_vmstate, |
| 388 | NULL, |
| 389 | } |
| 390 | }; |
| 391 | |
| 392 | static const VMStateDescription vuf_backend_vmstate = { |
| 393 | .name = "vhost-user-fs-backend", |
| 394 | .version_id = 0, |
| 395 | .needed = vuf_is_internal_migration, |
| 396 | .pre_load = vuf_check_migration_support, |
| 397 | .pre_save = vuf_check_migration_support, |
| 398 | .fields = (const VMStateField[]) { |
| 399 | { |
| 400 | .name = "back-end", |
| 401 | .info = &(const VMStateInfo) { |
| 402 | .name = "virtio-fs back-end state", |
| 403 | .get = vuf_load_state, |
| 404 | .put = vuf_save_state, |
| 405 | }, |
| 406 | }, |
| 407 | VMSTATE_END_OF_LIST() |
| 408 | }, |
| 409 | }; |
| 410 | |
| 411 | static const Property vuf_properties[] = { |
| 412 | DEFINE_PROP_CHR("chardev", VHostUserFS, conf.chardev), |
| 413 | DEFINE_PROP_STRING("tag", VHostUserFS, conf.tag), |
| 414 | DEFINE_PROP_UINT16("num-request-queues", VHostUserFS, |
| 415 | conf.num_request_queues, 1), |
| 416 | DEFINE_PROP_UINT16("queue-size", VHostUserFS, conf.queue_size, 128), |
| 417 | }; |
| 418 | |
| 419 | static void vuf_instance_init(Object *obj) |
| 420 | { |
| 421 | VHostUserFS *fs = VHOST_USER_FS(obj); |
| 422 | |
| 423 | device_add_bootindex_property(obj, &fs->bootindex, "bootindex", |
| 424 | "/filesystem@0", DEVICE(obj)); |
| 425 | } |
| 426 | |
| 427 | static void vuf_class_init(ObjectClass *klass, const void *data) |
| 428 | { |
| 429 | DeviceClass *dc = DEVICE_CLASS(klass); |
| 430 | VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); |
| 431 | |
| 432 | device_class_set_props(dc, vuf_properties); |
| 433 | dc->vmsd = &vuf_vmstate; |
| 434 | set_bit(DEVICE_CATEGORY_STORAGE, dc->categories); |
| 435 | vdc->realize = vuf_device_realize; |
| 436 | vdc->unrealize = vuf_device_unrealize; |
| 437 | vdc->get_features = vuf_get_features; |
| 438 | vdc->get_config = vuf_get_config; |
| 439 | vdc->set_status = vuf_set_status; |
| 440 | vdc->guest_notifier_mask = vuf_guest_notifier_mask; |
| 441 | vdc->guest_notifier_pending = vuf_guest_notifier_pending; |
| 442 | vdc->get_vhost = vuf_get_vhost; |
| 443 | } |
| 444 | |
| 445 | static const TypeInfo vuf_info = { |
| 446 | .name = TYPE_VHOST_USER_FS, |
| 447 | .parent = TYPE_VIRTIO_DEVICE, |
| 448 | .instance_size = sizeof(VHostUserFS), |
| 449 | .instance_init = vuf_instance_init, |
| 450 | .class_init = vuf_class_init, |
| 451 | }; |
| 452 | |
| 453 | static void vuf_register_types(void) |
| 454 | { |
| 455 | type_register_static(&vuf_info); |
| 456 | } |
| 457 | |
| 458 | type_init(vuf_register_types) |