arch/um/drivers/virtio_uml.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Virtio vhost-user driver
   4  *
   5  * Copyright(c) 2019 Intel Corporation
   6  *
   7  * This driver allows virtio devices to be used over a vhost-user socket.
   8  *
   9  * Guest devices can be instantiated by kernel module or command line
  10  * parameters. One device will be created for each parameter. Syntax:
  11  *
  12  *              virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]
  13  * where:
  14  *              <socket>        := vhost-user socket path to connect
  15  *              <virtio_id>     := virtio device id (as in virtio_ids.h)
  16  *              <platform_id>   := (optional) platform device id
  17  *
  18  * example:
  19  *              virtio_uml.device=/var/uml.socket:1
  20  *
  21  * Based on Virtio MMIO driver by Pawel Moll, copyright 2011-2014, ARM Ltd.
  22  */
  23 #include <linux/module.h>
  24 #include <linux/platform_device.h>
  25 #include <linux/slab.h>
  26 #include <linux/virtio.h>
  27 #include <linux/virtio_config.h>
  28 #include <linux/virtio_ring.h>
  29 #include <shared/as-layout.h>
  30 #include <irq_kern.h>
  31 #include <init.h>
  32 #include <os.h>
  33 #include "vhost_user.h"
  34
  35 /* Workaround due to a conflict between irq_user.h and irqreturn.h */
  36 #ifdef IRQ_NONE
  37 #undef IRQ_NONE
  38 #endif
  39
  40 #define MAX_SUPPORTED_QUEUE_SIZE        256
  41
  42 #define to_virtio_uml_device(_vdev) \
  43         container_of(_vdev, struct virtio_uml_device, vdev)
  44
  45 struct virtio_uml_platform_data {
  46         u32 virtio_device_id;
  47         const char *socket_path;
  48         struct work_struct conn_broken_wk;
  49         struct platform_device *pdev;
  50 };
  51
  52 struct virtio_uml_device {
  53         struct virtio_device vdev;
  54         struct platform_device *pdev;
  55
  56         int sock, req_fd;
  57         u64 features;
  58         u64 protocol_features;
  59         u8 status;
  60         u8 registered:1;
  61 };
  62
  63 struct virtio_uml_vq_info {
  64         int kick_fd, call_fd;
  65         char name[32];
  66 };
  67
  68 extern unsigned long long physmem_size, highmem;
  69
  70 #define vu_err(vu_dev, ...)     dev_err(&(vu_dev)->pdev->dev, __VA_ARGS__)
  71
  72 /* Vhost-user protocol */
  73
  74 static int full_sendmsg_fds(int fd, const void *buf, unsigned int len,
  75                             const int *fds, unsigned int fds_num)
  76 {
  77         int rc;
  78
  79         do {
  80                 rc = os_sendmsg_fds(fd, buf, len, fds, fds_num);
  81                 if (rc > 0) {
  82                         buf += rc;
  83                         len -= rc;
  84                         fds = NULL;
  85                         fds_num = 0;
  86                 }
  87         } while (len && (rc >= 0 || rc == -EINTR));
  88
  89         if (rc < 0)
  90                 return rc;
  91         return 0;
  92 }
  93
  94 static int full_read(int fd, void *buf, int len, bool abortable)
  95 {
  96         int rc;
  97
  98         do {
  99                 rc = os_read_file(fd, buf, len);
 100                 if (rc > 0) {
 101                         buf += rc;
 102                         len -= rc;
 103                 }
 104         } while (len && (rc > 0 || rc == -EINTR || (!abortable && rc == -EAGAIN)));
 105
 106         if (rc < 0)
 107                 return rc;
 108         if (rc == 0)
 109                 return -ECONNRESET;
 110         return 0;
 111 }
 112
 113 static int vhost_user_recv_header(int fd, struct vhost_user_msg *msg)
 114 {
 115         return full_read(fd, msg, sizeof(msg->header), true);
 116 }
 117
 118 static int vhost_user_recv(struct virtio_uml_device *vu_dev,
 119                            int fd, struct vhost_user_msg *msg,
 120                            size_t max_payload_size)
 121 {
 122         size_t size;
 123         int rc = vhost_user_recv_header(fd, msg);
 124
 125         if (rc == -ECONNRESET && vu_dev->registered) {
 126                 struct virtio_uml_platform_data *pdata;
 127
 128                 pdata = vu_dev->pdev->dev.platform_data;
 129
 130                 virtio_break_device(&vu_dev->vdev);
 131                 schedule_work(&pdata->conn_broken_wk);
 132         }
 133         if (rc)
 134                 return rc;
 135         size = msg->header.size;
 136         if (size > max_payload_size)
 137                 return -EPROTO;
 138         return full_read(fd, &msg->payload, size, false);
 139 }
 140
 141 static int vhost_user_recv_resp(struct virtio_uml_device *vu_dev,
 142                                 struct vhost_user_msg *msg,
 143                                 size_t max_payload_size)
 144 {
 145         int rc = vhost_user_recv(vu_dev, vu_dev->sock, msg, max_payload_size);
 146
 147         if (rc)
 148                 return rc;
 149
 150         if (msg->header.flags != (VHOST_USER_FLAG_REPLY | VHOST_USER_VERSION))
 151                 return -EPROTO;
 152
 153         return 0;
 154 }
 155
 156 static int vhost_user_recv_u64(struct virtio_uml_device *vu_dev,
 157                                u64 *value)
 158 {
 159         struct vhost_user_msg msg;
 160         int rc = vhost_user_recv_resp(vu_dev, &msg,
 161                                       sizeof(msg.payload.integer));
 162
 163         if (rc)
 164                 return rc;
 165         if (msg.header.size != sizeof(msg.payload.integer))
 166                 return -EPROTO;
 167         *value = msg.payload.integer;
 168         return 0;
 169 }
 170
 171 static int vhost_user_recv_req(struct virtio_uml_device *vu_dev,
 172                                struct vhost_user_msg *msg,
 173                                size_t max_payload_size)
 174 {
 175         int rc = vhost_user_recv(vu_dev, vu_dev->req_fd, msg, max_payload_size);
 176
 177         if (rc)
 178                 return rc;
 179
 180         if ((msg->header.flags & ~VHOST_USER_FLAG_NEED_REPLY) !=
 181                         VHOST_USER_VERSION)
 182                 return -EPROTO;
 183
 184         return 0;
 185 }
 186
 187 static int vhost_user_send(struct virtio_uml_device *vu_dev,
 188                            bool need_response, struct vhost_user_msg *msg,
 189                            int *fds, size_t num_fds)
 190 {
 191         size_t size = sizeof(msg->header) + msg->header.size;
 192         bool request_ack;
 193         int rc;
 194
 195         msg->header.flags |= VHOST_USER_VERSION;
 196
 197         /*
 198          * The need_response flag indicates that we already need a response,
 199          * e.g. to read the features. In these cases, don't request an ACK as
 200          * it is meaningless. Also request an ACK only if supported.
 201          */
 202         request_ack = !need_response;
 203         if (!(vu_dev->protocol_features &
 204                         BIT_ULL(VHOST_USER_PROTOCOL_F_REPLY_ACK)))
 205                 request_ack = false;
 206
 207         if (request_ack)
 208                 msg->header.flags |= VHOST_USER_FLAG_NEED_REPLY;
 209
 210         rc = full_sendmsg_fds(vu_dev->sock, msg, size, fds, num_fds);
 211         if (rc < 0)
 212                 return rc;
 213
 214         if (request_ack) {
 215                 uint64_t status;
 216
 217                 rc = vhost_user_recv_u64(vu_dev, &status);
 218                 if (rc)
 219                         return rc;
 220
 221                 if (status) {
 222                         vu_err(vu_dev, "slave reports error: %llu\n", status);
 223                         return -EIO;
 224                 }
 225         }
 226
 227         return 0;
 228 }
 229
 230 static int vhost_user_send_no_payload(struct virtio_uml_device *vu_dev,
 231                                       bool need_response, u32 request)
 232 {
 233         struct vhost_user_msg msg = {
 234                 .header.request = request,
 235         };
 236
 237         return vhost_user_send(vu_dev, need_response, &msg, NULL, 0);
 238 }
 239
 240 static int vhost_user_send_no_payload_fd(struct virtio_uml_device *vu_dev,
 241                                          u32 request, int fd)
 242 {
 243         struct vhost_user_msg msg = {
 244                 .header.request = request,
 245         };
 246
 247         return vhost_user_send(vu_dev, false, &msg, &fd, 1);
 248 }
 249
 250 static int vhost_user_send_u64(struct virtio_uml_device *vu_dev,
 251                                u32 request, u64 value)
 252 {
 253         struct vhost_user_msg msg = {
 254                 .header.request = request,
 255                 .header.size = sizeof(msg.payload.integer),
 256                 .payload.integer = value,
 257         };
 258
 259         return vhost_user_send(vu_dev, false, &msg, NULL, 0);
 260 }
 261
 262 static int vhost_user_set_owner(struct virtio_uml_device *vu_dev)
 263 {
 264         return vhost_user_send_no_payload(vu_dev, false, VHOST_USER_SET_OWNER);
 265 }
 266
 267 static int vhost_user_get_features(struct virtio_uml_device *vu_dev,
 268                                    u64 *features)
 269 {
 270         int rc = vhost_user_send_no_payload(vu_dev, true,
 271                                             VHOST_USER_GET_FEATURES);
 272
 273         if (rc)
 274                 return rc;
 275         return vhost_user_recv_u64(vu_dev, features);
 276 }
 277
 278 static int vhost_user_set_features(struct virtio_uml_device *vu_dev,
 279                                    u64 features)
 280 {
 281         return vhost_user_send_u64(vu_dev, VHOST_USER_SET_FEATURES, features);
 282 }
 283
 284 static int vhost_user_get_protocol_features(struct virtio_uml_device *vu_dev,
 285                                             u64 *protocol_features)
 286 {
 287         int rc = vhost_user_send_no_payload(vu_dev, true,
 288                         VHOST_USER_GET_PROTOCOL_FEATURES);
 289
 290         if (rc)
 291                 return rc;
 292         return vhost_user_recv_u64(vu_dev, protocol_features);
 293 }
 294
 295 static int vhost_user_set_protocol_features(struct virtio_uml_device *vu_dev,
 296                                             u64 protocol_features)
 297 {
 298         return vhost_user_send_u64(vu_dev, VHOST_USER_SET_PROTOCOL_FEATURES,
 299                                    protocol_features);
 300 }
 301
 302 static void vhost_user_reply(struct virtio_uml_device *vu_dev,
 303                              struct vhost_user_msg *msg, int response)
 304 {
 305         struct vhost_user_msg reply = {
 306                 .payload.integer = response,
 307         };
 308         size_t size = sizeof(reply.header) + sizeof(reply.payload.integer);
 309         int rc;
 310
 311         reply.header = msg->header;
 312         reply.header.flags &= ~VHOST_USER_FLAG_NEED_REPLY;
 313         reply.header.flags |= VHOST_USER_FLAG_REPLY;
 314         reply.header.size = sizeof(reply.payload.integer);
 315
 316         rc = full_sendmsg_fds(vu_dev->req_fd, &reply, size, NULL, 0);
 317
 318         if (rc)
 319                 vu_err(vu_dev,
 320                        "sending reply to slave request failed: %d (size %zu)\n",
 321                        rc, size);
 322 }
 323
 324 static irqreturn_t vu_req_interrupt(int irq, void *data)
 325 {
 326         struct virtio_uml_device *vu_dev = data;
 327         int response = 1;
 328         struct {
 329                 struct vhost_user_msg msg;
 330                 u8 extra_payload[512];
 331         } msg;
 332         int rc;
 333
 334         rc = vhost_user_recv_req(vu_dev, &msg.msg,
 335                                  sizeof(msg.msg.payload) +
 336                                  sizeof(msg.extra_payload));
 337
 338         if (rc)
 339                 return IRQ_NONE;
 340
 341         switch (msg.msg.header.request) {
 342         case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG:
 343                 virtio_config_changed(&vu_dev->vdev);
 344                 response = 0;
 345                 break;
 346         case VHOST_USER_SLAVE_IOTLB_MSG:
 347                 /* not supported - VIRTIO_F_IOMMU_PLATFORM */
 348         case VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG:
 349                 /* not supported - VHOST_USER_PROTOCOL_F_HOST_NOTIFIER */
 350         default:
 351                 vu_err(vu_dev, "unexpected slave request %d\n",
 352                        msg.msg.header.request);
 353         }
 354
 355         if (msg.msg.header.flags & VHOST_USER_FLAG_NEED_REPLY)
 356                 vhost_user_reply(vu_dev, &msg.msg, response);
 357
 358         return IRQ_HANDLED;
 359 }
 360
 361 static int vhost_user_init_slave_req(struct virtio_uml_device *vu_dev)
 362 {
 363         int rc, req_fds[2];
 364
 365         /* Use a pipe for slave req fd, SIGIO is not supported for eventfd */
 366         rc = os_pipe(req_fds, true, true);
 367         if (rc < 0)
 368                 return rc;
 369         vu_dev->req_fd = req_fds[0];
 370
 371         rc = um_request_irq(VIRTIO_IRQ, vu_dev->req_fd, IRQ_READ,
 372                             vu_req_interrupt, IRQF_SHARED,
 373                             vu_dev->pdev->name, vu_dev);
 374         if (rc)
 375                 goto err_close;
 376
 377         rc = vhost_user_send_no_payload_fd(vu_dev, VHOST_USER_SET_SLAVE_REQ_FD,
 378                                            req_fds[1]);
 379         if (rc)
 380                 goto err_free_irq;
 381
 382         goto out;
 383
 384 err_free_irq:
 385         um_free_irq(VIRTIO_IRQ, vu_dev);
 386 err_close:
 387         os_close_file(req_fds[0]);
 388 out:
 389         /* Close unused write end of request fds */
 390         os_close_file(req_fds[1]);
 391         return rc;
 392 }
 393
 394 static int vhost_user_init(struct virtio_uml_device *vu_dev)
 395 {
 396         int rc = vhost_user_set_owner(vu_dev);
 397
 398         if (rc)
 399                 return rc;
 400         rc = vhost_user_get_features(vu_dev, &vu_dev->features);
 401         if (rc)
 402                 return rc;
 403
 404         if (vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)) {
 405                 rc = vhost_user_get_protocol_features(vu_dev,
 406                                 &vu_dev->protocol_features);
 407                 if (rc)
 408                         return rc;
 409                 vu_dev->protocol_features &= VHOST_USER_SUPPORTED_PROTOCOL_F;
 410                 rc = vhost_user_set_protocol_features(vu_dev,
 411                                 vu_dev->protocol_features);
 412                 if (rc)
 413                         return rc;
 414         }
 415
 416         if (vu_dev->protocol_features &
 417                         BIT_ULL(VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
 418                 rc = vhost_user_init_slave_req(vu_dev);
 419                 if (rc)
 420                         return rc;
 421         }
 422
 423         return 0;
 424 }
 425
 426 static void vhost_user_get_config(struct virtio_uml_device *vu_dev,
 427                                   u32 offset, void *buf, u32 len)
 428 {
 429         u32 cfg_size = offset + len;
 430         struct vhost_user_msg *msg;
 431         size_t payload_size = sizeof(msg->payload.config) + cfg_size;
 432         size_t msg_size = sizeof(msg->header) + payload_size;
 433         int rc;
 434
 435         if (!(vu_dev->protocol_features &
 436               BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
 437                 return;
 438
 439         msg = kzalloc(msg_size, GFP_KERNEL);
 440         if (!msg)
 441                 return;
 442         msg->header.request = VHOST_USER_GET_CONFIG;
 443         msg->header.size = payload_size;
 444         msg->payload.config.offset = 0;
 445         msg->payload.config.size = cfg_size;
 446
 447         rc = vhost_user_send(vu_dev, true, msg, NULL, 0);
 448         if (rc) {
 449                 vu_err(vu_dev, "sending VHOST_USER_GET_CONFIG failed: %d\n",
 450                        rc);
 451                 goto free;
 452         }
 453
 454         rc = vhost_user_recv_resp(vu_dev, msg, msg_size);
 455         if (rc) {
 456                 vu_err(vu_dev,
 457                        "receiving VHOST_USER_GET_CONFIG response failed: %d\n",
 458                        rc);
 459                 goto free;
 460         }
 461
 462         if (msg->header.size != payload_size ||
 463             msg->payload.config.size != cfg_size) {
 464                 rc = -EPROTO;
 465                 vu_err(vu_dev,
 466                        "Invalid VHOST_USER_GET_CONFIG sizes (payload %d expected %zu, config %u expected %u)\n",
 467                        msg->header.size, payload_size,
 468                        msg->payload.config.size, cfg_size);
 469                 goto free;
 470         }
 471         memcpy(buf, msg->payload.config.payload + offset, len);
 472
 473 free:
 474         kfree(msg);
 475 }
 476
 477 static void vhost_user_set_config(struct virtio_uml_device *vu_dev,
 478                                   u32 offset, const void *buf, u32 len)
 479 {
 480         struct vhost_user_msg *msg;
 481         size_t payload_size = sizeof(msg->payload.config) + len;
 482         size_t msg_size = sizeof(msg->header) + payload_size;
 483         int rc;
 484
 485         if (!(vu_dev->protocol_features &
 486               BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
 487                 return;
 488
 489         msg = kzalloc(msg_size, GFP_KERNEL);
 490         if (!msg)
 491                 return;
 492         msg->header.request = VHOST_USER_SET_CONFIG;
 493         msg->header.size = payload_size;
 494         msg->payload.config.offset = offset;
 495         msg->payload.config.size = len;
 496         memcpy(msg->payload.config.payload, buf, len);
 497
 498         rc = vhost_user_send(vu_dev, false, msg, NULL, 0);
 499         if (rc)
 500                 vu_err(vu_dev, "sending VHOST_USER_SET_CONFIG failed: %d\n",
 501                        rc);
 502
 503         kfree(msg);
 504 }
 505
 506 static int vhost_user_init_mem_region(u64 addr, u64 size, int *fd_out,
 507                                       struct vhost_user_mem_region *region_out)
 508 {
 509         unsigned long long mem_offset;
 510         int rc = phys_mapping(addr, &mem_offset);
 511
 512         if (WARN(rc < 0, "phys_mapping of 0x%llx returned %d\n", addr, rc))
 513                 return -EFAULT;
 514         *fd_out = rc;
 515         region_out->guest_addr = addr;
 516         region_out->user_addr = addr;
 517         region_out->size = size;
 518         region_out->mmap_offset = mem_offset;
 519
 520         /* Ensure mapping is valid for the entire region */
 521         rc = phys_mapping(addr + size - 1, &mem_offset);
 522         if (WARN(rc != *fd_out, "phys_mapping of 0x%llx failed: %d != %d\n",
 523                  addr + size - 1, rc, *fd_out))
 524                 return -EFAULT;
 525         return 0;
 526 }
 527
 528 static int vhost_user_set_mem_table(struct virtio_uml_device *vu_dev)
 529 {
 530         struct vhost_user_msg msg = {
 531                 .header.request = VHOST_USER_SET_MEM_TABLE,
 532                 .header.size = sizeof(msg.payload.mem_regions),
 533                 .payload.mem_regions.num = 1,
 534         };
 535         unsigned long reserved = uml_reserved - uml_physmem;
 536         int fds[2];
 537         int rc;
 538
 539         /*
 540          * This is a bit tricky, see also the comment with setup_physmem().
 541          *
 542          * Essentially, setup_physmem() uses a file to mmap() our physmem,
 543          * but the code and data we *already* have is omitted. To us, this
 544          * is no difference, since they both become part of our address
 545          * space and memory consumption. To somebody looking in from the
 546          * outside, however, it is different because the part of our memory
 547          * consumption that's already part of the binary (code/data) is not
 548          * mapped from the file, so it's not visible to another mmap from
 549          * the file descriptor.
 550          *
 551          * Thus, don't advertise this space to the vhost-user slave. This
 552          * means that the slave will likely abort or similar when we give
 553          * it an address from the hidden range, since it's not marked as
 554          * a valid address, but at least that way we detect the issue and
 555          * don't just have the slave read an all-zeroes buffer from the
 556          * shared memory file, or write something there that we can never
 557          * see (depending on the direction of the virtqueue traffic.)
 558          *
 559          * Since we usually don't want to use .text for virtio buffers,
 560          * this effectively means that you cannot use
 561          *  1) global variables, which are in the .bss and not in the shm
 562          *     file-backed memory
 563          *  2) the stack in some processes, depending on where they have
 564          *     their stack (or maybe only no interrupt stack?)
 565          *
 566          * The stack is already not typically valid for DMA, so this isn't
 567          * much of a restriction, but global variables might be encountered.
 568          *
 569          * It might be possible to fix it by copying around the data that's
 570          * between bss_start and where we map the file now, but it's not
 571          * something that you typically encounter with virtio drivers, so
 572          * it didn't seem worthwhile.
 573          */
 574         rc = vhost_user_init_mem_region(reserved, physmem_size - reserved,
 575                                         &fds[0],
 576                                         &msg.payload.mem_regions.regions[0]);
 577
 578         if (rc < 0)
 579                 return rc;
 580         if (highmem) {
 581                 msg.payload.mem_regions.num++;
 582                 rc = vhost_user_init_mem_region(__pa(end_iomem), highmem,
 583                                 &fds[1], &msg.payload.mem_regions.regions[1]);
 584                 if (rc < 0)
 585                         return rc;
 586         }
 587
 588         return vhost_user_send(vu_dev, false, &msg, fds,
 589                                msg.payload.mem_regions.num);
 590 }
 591
 592 static int vhost_user_set_vring_state(struct virtio_uml_device *vu_dev,
 593                                       u32 request, u32 index, u32 num)
 594 {
 595         struct vhost_user_msg msg = {
 596                 .header.request = request,
 597                 .header.size = sizeof(msg.payload.vring_state),
 598                 .payload.vring_state.index = index,
 599                 .payload.vring_state.num = num,
 600         };
 601
 602         return vhost_user_send(vu_dev, false, &msg, NULL, 0);
 603 }
 604
 605 static int vhost_user_set_vring_num(struct virtio_uml_device *vu_dev,
 606                                     u32 index, u32 num)
 607 {
 608         return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_NUM,
 609                                           index, num);
 610 }
 611
 612 static int vhost_user_set_vring_base(struct virtio_uml_device *vu_dev,
 613                                      u32 index, u32 offset)
 614 {
 615         return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_BASE,
 616                                           index, offset);
 617 }
 618
 619 static int vhost_user_set_vring_addr(struct virtio_uml_device *vu_dev,
 620                                      u32 index, u64 desc, u64 used, u64 avail,
 621                                      u64 log)
 622 {
 623         struct vhost_user_msg msg = {
 624                 .header.request = VHOST_USER_SET_VRING_ADDR,
 625                 .header.size = sizeof(msg.payload.vring_addr),
 626                 .payload.vring_addr.index = index,
 627                 .payload.vring_addr.desc = desc,
 628                 .payload.vring_addr.used = used,
 629                 .payload.vring_addr.avail = avail,
 630                 .payload.vring_addr.log = log,
 631         };
 632
 633         return vhost_user_send(vu_dev, false, &msg, NULL, 0);
 634 }
 635
 636 static int vhost_user_set_vring_fd(struct virtio_uml_device *vu_dev,
 637                                    u32 request, int index, int fd)
 638 {
 639         struct vhost_user_msg msg = {
 640                 .header.request = request,
 641                 .header.size = sizeof(msg.payload.integer),
 642                 .payload.integer = index,
 643         };
 644
 645         if (index & ~VHOST_USER_VRING_INDEX_MASK)
 646                 return -EINVAL;
 647         if (fd < 0) {
 648                 msg.payload.integer |= VHOST_USER_VRING_POLL_MASK;
 649                 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
 650         }
 651         return vhost_user_send(vu_dev, false, &msg, &fd, 1);
 652 }
 653
 654 static int vhost_user_set_vring_call(struct virtio_uml_device *vu_dev,
 655                                      int index, int fd)
 656 {
 657         return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_CALL,
 658                                        index, fd);
 659 }
 660
 661 static int vhost_user_set_vring_kick(struct virtio_uml_device *vu_dev,
 662                                      int index, int fd)
 663 {
 664         return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_KICK,
 665                                        index, fd);
 666 }
 667
 668 static int vhost_user_set_vring_enable(struct virtio_uml_device *vu_dev,
 669                                        u32 index, bool enable)
 670 {
 671         if (!(vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)))
 672                 return 0;
 673
 674         return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_ENABLE,
 675                                           index, enable);
 676 }
 677
 678
 679 /* Virtio interface */
 680
 681 static bool vu_notify(struct virtqueue *vq)
 682 {
 683         struct virtio_uml_vq_info *info = vq->priv;
 684         const uint64_t n = 1;
 685         int rc;
 686
 687         do {
 688                 rc = os_write_file(info->kick_fd, &n, sizeof(n));
 689         } while (rc == -EINTR);
 690         return !WARN(rc != sizeof(n), "write returned %d\n", rc);
 691 }
 692
 693 static irqreturn_t vu_interrupt(int irq, void *opaque)
 694 {
 695         struct virtqueue *vq = opaque;
 696         struct virtio_uml_vq_info *info = vq->priv;
 697         uint64_t n;
 698         int rc;
 699         irqreturn_t ret = IRQ_NONE;
 700
 701         do {
 702                 rc = os_read_file(info->call_fd, &n, sizeof(n));
 703                 if (rc == sizeof(n))
 704                         ret |= vring_interrupt(irq, vq);
 705         } while (rc == sizeof(n) || rc == -EINTR);
 706         WARN(rc != -EAGAIN, "read returned %d\n", rc);
 707         return ret;
 708 }
 709
 710
 711 static void vu_get(struct virtio_device *vdev, unsigned offset,
 712                    void *buf, unsigned len)
 713 {
 714         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 715
 716         vhost_user_get_config(vu_dev, offset, buf, len);
 717 }
 718
 719 static void vu_set(struct virtio_device *vdev, unsigned offset,
 720                    const void *buf, unsigned len)
 721 {
 722         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 723
 724         vhost_user_set_config(vu_dev, offset, buf, len);
 725 }
 726
 727 static u8 vu_get_status(struct virtio_device *vdev)
 728 {
 729         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 730
 731         return vu_dev->status;
 732 }
 733
 734 static void vu_set_status(struct virtio_device *vdev, u8 status)
 735 {
 736         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 737
 738         vu_dev->status = status;
 739 }
 740
 741 static void vu_reset(struct virtio_device *vdev)
 742 {
 743         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 744
 745         vu_dev->status = 0;
 746 }
 747
 748 static void vu_del_vq(struct virtqueue *vq)
 749 {
 750         struct virtio_uml_vq_info *info = vq->priv;
 751
 752         um_free_irq(VIRTIO_IRQ, vq);
 753
 754         os_close_file(info->call_fd);
 755         os_close_file(info->kick_fd);
 756
 757         vring_del_virtqueue(vq);
 758         kfree(info);
 759 }
 760
 761 static void vu_del_vqs(struct virtio_device *vdev)
 762 {
 763         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 764         struct virtqueue *vq, *n;
 765         u64 features;
 766
 767         /* Note: reverse order as a workaround to a decoding bug in snabb */
 768         list_for_each_entry_reverse(vq, &vdev->vqs, list)
 769                 WARN_ON(vhost_user_set_vring_enable(vu_dev, vq->index, false));
 770
 771         /* Ensure previous messages have been processed */
 772         WARN_ON(vhost_user_get_features(vu_dev, &features));
 773
 774         list_for_each_entry_safe(vq, n, &vdev->vqs, list)
 775                 vu_del_vq(vq);
 776 }
 777
 778 static int vu_setup_vq_call_fd(struct virtio_uml_device *vu_dev,
 779                                struct virtqueue *vq)
 780 {
 781         struct virtio_uml_vq_info *info = vq->priv;
 782         int call_fds[2];
 783         int rc;
 784
 785         /* Use a pipe for call fd, since SIGIO is not supported for eventfd */
 786         rc = os_pipe(call_fds, true, true);
 787         if (rc < 0)
 788                 return rc;
 789
 790         info->call_fd = call_fds[0];
 791         rc = um_request_irq(VIRTIO_IRQ, info->call_fd, IRQ_READ,
 792                             vu_interrupt, IRQF_SHARED, info->name, vq);
 793         if (rc)
 794                 goto close_both;
 795
 796         rc = vhost_user_set_vring_call(vu_dev, vq->index, call_fds[1]);
 797         if (rc)
 798                 goto release_irq;
 799
 800         goto out;
 801
 802 release_irq:
 803         um_free_irq(VIRTIO_IRQ, vq);
 804 close_both:
 805         os_close_file(call_fds[0]);
 806 out:
 807         /* Close (unused) write end of call fds */
 808         os_close_file(call_fds[1]);
 809
 810         return rc;
 811 }
 812
 813 static struct virtqueue *vu_setup_vq(struct virtio_device *vdev,
 814                                      unsigned index, vq_callback_t *callback,
 815                                      const char *name, bool ctx)
 816 {
 817         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 818         struct platform_device *pdev = vu_dev->pdev;
 819         struct virtio_uml_vq_info *info;
 820         struct virtqueue *vq;
 821         int num = MAX_SUPPORTED_QUEUE_SIZE;
 822         int rc;
 823
 824         info = kzalloc(sizeof(*info), GFP_KERNEL);
 825         if (!info) {
 826                 rc = -ENOMEM;
 827                 goto error_kzalloc;
 828         }
 829         snprintf(info->name, sizeof(info->name), "%s.%d-%s", pdev->name,
 830                  pdev->id, name);
 831
 832         vq = vring_create_virtqueue(index, num, PAGE_SIZE, vdev, true, true,
 833                                     ctx, vu_notify, callback, info->name);
 834         if (!vq) {
 835                 rc = -ENOMEM;
 836                 goto error_create;
 837         }
 838         vq->priv = info;
 839         num = virtqueue_get_vring_size(vq);
 840
 841         rc = os_eventfd(0, 0);
 842         if (rc < 0)
 843                 goto error_kick;
 844         info->kick_fd = rc;
 845
 846         rc = vu_setup_vq_call_fd(vu_dev, vq);
 847         if (rc)
 848                 goto error_call;
 849
 850         rc = vhost_user_set_vring_num(vu_dev, index, num);
 851         if (rc)
 852                 goto error_setup;
 853
 854         rc = vhost_user_set_vring_base(vu_dev, index, 0);
 855         if (rc)
 856                 goto error_setup;
 857
 858         rc = vhost_user_set_vring_addr(vu_dev, index,
 859                                        virtqueue_get_desc_addr(vq),
 860                                        virtqueue_get_used_addr(vq),
 861                                        virtqueue_get_avail_addr(vq),
 862                                        (u64) -1);
 863         if (rc)
 864                 goto error_setup;
 865
 866         return vq;
 867
 868 error_setup:
 869         um_free_irq(VIRTIO_IRQ, vq);
 870         os_close_file(info->call_fd);
 871 error_call:
 872         os_close_file(info->kick_fd);
 873 error_kick:
 874         vring_del_virtqueue(vq);
 875 error_create:
 876         kfree(info);
 877 error_kzalloc:
 878         return ERR_PTR(rc);
 879 }
 880
 881 static int vu_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 882                        struct virtqueue *vqs[], vq_callback_t *callbacks[],
 883                        const char * const names[], const bool *ctx,
 884                        struct irq_affinity *desc)
 885 {
 886         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 887         int i, queue_idx = 0, rc;
 888         struct virtqueue *vq;
 889
 890         rc = vhost_user_set_mem_table(vu_dev);
 891         if (rc)
 892                 return rc;
 893
 894         for (i = 0; i < nvqs; ++i) {
 895                 if (!names[i]) {
 896                         vqs[i] = NULL;
 897                         continue;
 898                 }
 899
 900                 vqs[i] = vu_setup_vq(vdev, queue_idx++, callbacks[i], names[i],
 901                                      ctx ? ctx[i] : false);
 902                 if (IS_ERR(vqs[i])) {
 903                         rc = PTR_ERR(vqs[i]);
 904                         goto error_setup;
 905                 }
 906         }
 907
 908         list_for_each_entry(vq, &vdev->vqs, list) {
 909                 struct virtio_uml_vq_info *info = vq->priv;
 910
 911                 rc = vhost_user_set_vring_kick(vu_dev, vq->index,
 912                                                info->kick_fd);
 913                 if (rc)
 914                         goto error_setup;
 915
 916                 rc = vhost_user_set_vring_enable(vu_dev, vq->index, true);
 917                 if (rc)
 918                         goto error_setup;
 919         }
 920
 921         return 0;
 922
 923 error_setup:
 924         vu_del_vqs(vdev);
 925         return rc;
 926 }
 927
 928 static u64 vu_get_features(struct virtio_device *vdev)
 929 {
 930         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 931
 932         return vu_dev->features;
 933 }
 934
 935 static int vu_finalize_features(struct virtio_device *vdev)
 936 {
 937         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 938         u64 supported = vdev->features & VHOST_USER_SUPPORTED_F;
 939
 940         vring_transport_features(vdev);
 941         vu_dev->features = vdev->features | supported;
 942
 943         return vhost_user_set_features(vu_dev, vu_dev->features);
 944 }
 945
 946 static const char *vu_bus_name(struct virtio_device *vdev)
 947 {
 948         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 949
 950         return vu_dev->pdev->name;
 951 }
 952
 953 static const struct virtio_config_ops virtio_uml_config_ops = {
 954         .get = vu_get,
 955         .set = vu_set,
 956         .get_status = vu_get_status,
 957         .set_status = vu_set_status,
 958         .reset = vu_reset,
 959         .find_vqs = vu_find_vqs,
 960         .del_vqs = vu_del_vqs,
 961         .get_features = vu_get_features,
 962         .finalize_features = vu_finalize_features,
 963         .bus_name = vu_bus_name,
 964 };
 965
 966 static void virtio_uml_release_dev(struct device *d)
 967 {
 968         struct virtio_device *vdev =
 969                         container_of(d, struct virtio_device, dev);
 970         struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
 971
 972         /* might not have been opened due to not negotiating the feature */
 973         if (vu_dev->req_fd >= 0) {
 974                 um_free_irq(VIRTIO_IRQ, vu_dev);
 975                 os_close_file(vu_dev->req_fd);
 976         }
 977
 978         os_close_file(vu_dev->sock);
 979 }
 980
 981 /* Platform device */
 982
 983 static int virtio_uml_probe(struct platform_device *pdev)
 984 {
 985         struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
 986         struct virtio_uml_device *vu_dev;
 987         int rc;
 988
 989         if (!pdata)
 990                 return -EINVAL;
 991
 992         vu_dev = devm_kzalloc(&pdev->dev, sizeof(*vu_dev), GFP_KERNEL);
 993         if (!vu_dev)
 994                 return -ENOMEM;
 995
 996         vu_dev->vdev.dev.parent = &pdev->dev;
 997         vu_dev->vdev.dev.release = virtio_uml_release_dev;
 998         vu_dev->vdev.config = &virtio_uml_config_ops;
 999         vu_dev->vdev.id.device = pdata->virtio_device_id;
1000         vu_dev->vdev.id.vendor = VIRTIO_DEV_ANY_ID;
1001         vu_dev->pdev = pdev;
1002         vu_dev->req_fd = -1;
1003
1004         do {
1005                 rc = os_connect_socket(pdata->socket_path);
1006         } while (rc == -EINTR);
1007         if (rc < 0)
1008                 return rc;
1009         vu_dev->sock = rc;
1010
1011         rc = vhost_user_init(vu_dev);
1012         if (rc)
1013                 goto error_init;
1014
1015         platform_set_drvdata(pdev, vu_dev);
1016
1017         rc = register_virtio_device(&vu_dev->vdev);
1018         if (rc)
1019                 put_device(&vu_dev->vdev.dev);
1020         vu_dev->registered = 1;
1021         return rc;
1022
1023 error_init:
1024         os_close_file(vu_dev->sock);
1025         return rc;
1026 }
1027
1028 static int virtio_uml_remove(struct platform_device *pdev)
1029 {
1030         struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1031
1032         unregister_virtio_device(&vu_dev->vdev);
1033         return 0;
1034 }
1035
1036 /* Command line device list */
1037
1038 static void vu_cmdline_release_dev(struct device *d)
1039 {
1040 }
1041
1042 static struct device vu_cmdline_parent = {
1043         .init_name = "virtio-uml-cmdline",
1044         .release = vu_cmdline_release_dev,
1045 };
1046
1047 static bool vu_cmdline_parent_registered;
1048 static int vu_cmdline_id;
1049
1050 static int vu_unregister_cmdline_device(struct device *dev, void *data)
1051 {
1052         struct platform_device *pdev = to_platform_device(dev);
1053         struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1054
1055         kfree(pdata->socket_path);
1056         platform_device_unregister(pdev);
1057         return 0;
1058 }
1059
1060 static void vu_conn_broken(struct work_struct *wk)
1061 {
1062         struct virtio_uml_platform_data *pdata;
1063
1064         pdata = container_of(wk, struct virtio_uml_platform_data, conn_broken_wk);
1065         vu_unregister_cmdline_device(&pdata->pdev->dev, NULL);
1066 }
1067
1068 static int vu_cmdline_set(const char *device, const struct kernel_param *kp)
1069 {
1070         const char *ids = strchr(device, ':');
1071         unsigned int virtio_device_id;
1072         int processed, consumed, err;
1073         char *socket_path;
1074         struct virtio_uml_platform_data pdata, *ppdata;
1075         struct platform_device *pdev;
1076
1077         if (!ids || ids == device)
1078                 return -EINVAL;
1079
1080         processed = sscanf(ids, ":%u%n:%d%n",
1081                            &virtio_device_id, &consumed,
1082                            &vu_cmdline_id, &consumed);
1083
1084         if (processed < 1 || ids[consumed])
1085                 return -EINVAL;
1086
1087         if (!vu_cmdline_parent_registered) {
1088                 err = device_register(&vu_cmdline_parent);
1089                 if (err) {
1090                         pr_err("Failed to register parent device!\n");
1091                         put_device(&vu_cmdline_parent);
1092                         return err;
1093                 }
1094                 vu_cmdline_parent_registered = true;
1095         }
1096
1097         socket_path = kmemdup_nul(device, ids - device, GFP_KERNEL);
1098         if (!socket_path)
1099                 return -ENOMEM;
1100
1101         pdata.virtio_device_id = (u32) virtio_device_id;
1102         pdata.socket_path = socket_path;
1103
1104         pr_info("Registering device virtio-uml.%d id=%d at %s\n",
1105                 vu_cmdline_id, virtio_device_id, socket_path);
1106
1107         pdev = platform_device_register_data(&vu_cmdline_parent, "virtio-uml",
1108                                              vu_cmdline_id++, &pdata,
1109                                              sizeof(pdata));
1110         err = PTR_ERR_OR_ZERO(pdev);
1111         if (err)
1112                 goto free;
1113
1114         ppdata = pdev->dev.platform_data;
1115         ppdata->pdev = pdev;
1116         INIT_WORK(&ppdata->conn_broken_wk, vu_conn_broken);
1117
1118         return 0;
1119
1120 free:
1121         kfree(socket_path);
1122         return err;
1123 }
1124
1125 static int vu_cmdline_get_device(struct device *dev, void *data)
1126 {
1127         struct platform_device *pdev = to_platform_device(dev);
1128         struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1129         char *buffer = data;
1130         unsigned int len = strlen(buffer);
1131
1132         snprintf(buffer + len, PAGE_SIZE - len, "%s:%d:%d\n",
1133                  pdata->socket_path, pdata->virtio_device_id, pdev->id);
1134         return 0;
1135 }
1136
1137 static int vu_cmdline_get(char *buffer, const struct kernel_param *kp)
1138 {
1139         buffer[0] = '\0';
1140         if (vu_cmdline_parent_registered)
1141                 device_for_each_child(&vu_cmdline_parent, buffer,
1142                                       vu_cmdline_get_device);
1143         return strlen(buffer) + 1;
1144 }
1145
1146 static const struct kernel_param_ops vu_cmdline_param_ops = {
1147         .set = vu_cmdline_set,
1148         .get = vu_cmdline_get,
1149 };
1150
1151 device_param_cb(device, &vu_cmdline_param_ops, NULL, S_IRUSR);
1152 __uml_help(vu_cmdline_param_ops,
1153 "virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]\n"
1154 "    Configure a virtio device over a vhost-user socket.\n"
1155 "    See virtio_ids.h for a list of possible virtio device id values.\n"
1156 "    Optionally use a specific platform_device id.\n\n"
1157 );
1158
1159
1160 static void vu_unregister_cmdline_devices(void)
1161 {
1162         if (vu_cmdline_parent_registered) {
1163                 device_for_each_child(&vu_cmdline_parent, NULL,
1164                                       vu_unregister_cmdline_device);
1165                 device_unregister(&vu_cmdline_parent);
1166                 vu_cmdline_parent_registered = false;
1167         }
1168 }
1169
1170 /* Platform driver */
1171
1172 static const struct of_device_id virtio_uml_match[] = {
1173         { .compatible = "virtio,uml", },
1174         { }
1175 };
1176 MODULE_DEVICE_TABLE(of, virtio_uml_match);
1177
1178 static struct platform_driver virtio_uml_driver = {
1179         .probe = virtio_uml_probe,
1180         .remove = virtio_uml_remove,
1181         .driver = {
1182                 .name = "virtio-uml",
1183                 .of_match_table = virtio_uml_match,
1184         },
1185 };
1186
1187 static int __init virtio_uml_init(void)
1188 {
1189         return platform_driver_register(&virtio_uml_driver);
1190 }
1191
1192 static void __exit virtio_uml_exit(void)
1193 {
1194         platform_driver_unregister(&virtio_uml_driver);
1195         vu_unregister_cmdline_devices();
1196 }
1197
1198 module_init(virtio_uml_init);
1199 module_exit(virtio_uml_exit);
1200 __uml_exitcall(virtio_uml_exit);
1201
1202 MODULE_DESCRIPTION("UML driver for vhost-user virtio devices");
1203 MODULE_LICENSE("GPL");