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