hw/virtio/vhost-user.c

   1 /*
   2  * vhost-user
   3  *
   4  * Copyright (c) 2013 Virtual Open Systems Sarl.
   5  *
   6  * This work is licensed under the terms of the GNU GPL, version 2 or later.
   7  * See the COPYING file in the top-level directory.
   8  *
   9  */
  10
  11 #include "qemu/osdep.h"
  12 #include "qapi/error.h"
  13 #include "hw/virtio/vhost.h"
  14 #include "hw/virtio/vhost-user.h"
  15 #include "hw/virtio/vhost-backend.h"
  16 #include "hw/virtio/virtio.h"
  17 #include "hw/virtio/virtio-net.h"
  18 #include "chardev/char-fe.h"
  19 #include "sysemu/kvm.h"
  20 #include "qemu/error-report.h"
  21 #include "qemu/main-loop.h"
  22 #include "qemu/sockets.h"
  23 #include "sysemu/cryptodev.h"
  24 #include "migration/migration.h"
  25 #include "migration/postcopy-ram.h"
  26 #include "trace.h"
  27
  28 #include <sys/ioctl.h>
  29 #include <sys/socket.h>
  30 #include <sys/un.h>
  31
  32 #include "standard-headers/linux/vhost_types.h"
  33
  34 #ifdef CONFIG_LINUX
  35 #include <linux/userfaultfd.h>
  36 #endif
  37
  38 #define VHOST_MEMORY_BASELINE_NREGIONS    8
  39 #define VHOST_USER_F_PROTOCOL_FEATURES 30
  40 #define VHOST_USER_SLAVE_MAX_FDS     8
  41
  42 /*
  43  * Set maximum number of RAM slots supported to
  44  * the maximum number supported by the target
  45  * hardware plaform.
  46  */
  47 #if defined(TARGET_X86) || defined(TARGET_X86_64) || \
  48     defined(TARGET_ARM) || defined(TARGET_ARM_64)
  49 #include "hw/acpi/acpi.h"
  50 #define VHOST_USER_MAX_RAM_SLOTS ACPI_MAX_RAM_SLOTS
  51
  52 #elif defined(TARGET_PPC) || defined(TARGET_PPC_64)
  53 #include "hw/ppc/spapr.h"
  54 #define VHOST_USER_MAX_RAM_SLOTS SPAPR_MAX_RAM_SLOTS
  55
  56 #else
  57 #define VHOST_USER_MAX_RAM_SLOTS 512
  58 #endif
  59
  60 /*
  61  * Maximum size of virtio device config space
  62  */
  63 #define VHOST_USER_MAX_CONFIG_SIZE 256
  64
  65 enum VhostUserProtocolFeature {
  66     VHOST_USER_PROTOCOL_F_MQ = 0,
  67     VHOST_USER_PROTOCOL_F_LOG_SHMFD = 1,
  68     VHOST_USER_PROTOCOL_F_RARP = 2,
  69     VHOST_USER_PROTOCOL_F_REPLY_ACK = 3,
  70     VHOST_USER_PROTOCOL_F_NET_MTU = 4,
  71     VHOST_USER_PROTOCOL_F_SLAVE_REQ = 5,
  72     VHOST_USER_PROTOCOL_F_CROSS_ENDIAN = 6,
  73     VHOST_USER_PROTOCOL_F_CRYPTO_SESSION = 7,
  74     VHOST_USER_PROTOCOL_F_PAGEFAULT = 8,
  75     VHOST_USER_PROTOCOL_F_CONFIG = 9,
  76     VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD = 10,
  77     VHOST_USER_PROTOCOL_F_HOST_NOTIFIER = 11,
  78     VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD = 12,
  79     VHOST_USER_PROTOCOL_F_RESET_DEVICE = 13,
  80     /* Feature 14 reserved for VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS. */
  81     VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS = 15,
  82     VHOST_USER_PROTOCOL_F_MAX
  83 };
  84
  85 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
  86
  87 typedef enum VhostUserRequest {
  88     VHOST_USER_NONE = 0,
  89     VHOST_USER_GET_FEATURES = 1,
  90     VHOST_USER_SET_FEATURES = 2,
  91     VHOST_USER_SET_OWNER = 3,
  92     VHOST_USER_RESET_OWNER = 4,
  93     VHOST_USER_SET_MEM_TABLE = 5,
  94     VHOST_USER_SET_LOG_BASE = 6,
  95     VHOST_USER_SET_LOG_FD = 7,
  96     VHOST_USER_SET_VRING_NUM = 8,
  97     VHOST_USER_SET_VRING_ADDR = 9,
  98     VHOST_USER_SET_VRING_BASE = 10,
  99     VHOST_USER_GET_VRING_BASE = 11,
 100     VHOST_USER_SET_VRING_KICK = 12,
 101     VHOST_USER_SET_VRING_CALL = 13,
 102     VHOST_USER_SET_VRING_ERR = 14,
 103     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
 104     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
 105     VHOST_USER_GET_QUEUE_NUM = 17,
 106     VHOST_USER_SET_VRING_ENABLE = 18,
 107     VHOST_USER_SEND_RARP = 19,
 108     VHOST_USER_NET_SET_MTU = 20,
 109     VHOST_USER_SET_SLAVE_REQ_FD = 21,
 110     VHOST_USER_IOTLB_MSG = 22,
 111     VHOST_USER_SET_VRING_ENDIAN = 23,
 112     VHOST_USER_GET_CONFIG = 24,
 113     VHOST_USER_SET_CONFIG = 25,
 114     VHOST_USER_CREATE_CRYPTO_SESSION = 26,
 115     VHOST_USER_CLOSE_CRYPTO_SESSION = 27,
 116     VHOST_USER_POSTCOPY_ADVISE  = 28,
 117     VHOST_USER_POSTCOPY_LISTEN  = 29,
 118     VHOST_USER_POSTCOPY_END     = 30,
 119     VHOST_USER_GET_INFLIGHT_FD = 31,
 120     VHOST_USER_SET_INFLIGHT_FD = 32,
 121     VHOST_USER_GPU_SET_SOCKET = 33,
 122     VHOST_USER_RESET_DEVICE = 34,
 123     /* Message number 35 reserved for VHOST_USER_VRING_KICK. */
 124     VHOST_USER_GET_MAX_MEM_SLOTS = 36,
 125     VHOST_USER_ADD_MEM_REG = 37,
 126     VHOST_USER_REM_MEM_REG = 38,
 127     VHOST_USER_MAX
 128 } VhostUserRequest;
 129
 130 typedef enum VhostUserSlaveRequest {
 131     VHOST_USER_SLAVE_NONE = 0,
 132     VHOST_USER_SLAVE_IOTLB_MSG = 1,
 133     VHOST_USER_SLAVE_CONFIG_CHANGE_MSG = 2,
 134     VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG = 3,
 135     VHOST_USER_SLAVE_MAX
 136 }  VhostUserSlaveRequest;
 137
 138 typedef struct VhostUserMemoryRegion {
 139     uint64_t guest_phys_addr;
 140     uint64_t memory_size;
 141     uint64_t userspace_addr;
 142     uint64_t mmap_offset;
 143 } VhostUserMemoryRegion;
 144
 145 typedef struct VhostUserMemory {
 146     uint32_t nregions;
 147     uint32_t padding;
 148     VhostUserMemoryRegion regions[VHOST_MEMORY_BASELINE_NREGIONS];
 149 } VhostUserMemory;
 150
 151 typedef struct VhostUserMemRegMsg {
 152     uint64_t padding;
 153     VhostUserMemoryRegion region;
 154 } VhostUserMemRegMsg;
 155
 156 typedef struct VhostUserLog {
 157     uint64_t mmap_size;
 158     uint64_t mmap_offset;
 159 } VhostUserLog;
 160
 161 typedef struct VhostUserConfig {
 162     uint32_t offset;
 163     uint32_t size;
 164     uint32_t flags;
 165     uint8_t region[VHOST_USER_MAX_CONFIG_SIZE];
 166 } VhostUserConfig;
 167
 168 #define VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN    512
 169 #define VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN  64
 170
 171 typedef struct VhostUserCryptoSession {
 172     /* session id for success, -1 on errors */
 173     int64_t session_id;
 174     CryptoDevBackendSymSessionInfo session_setup_data;
 175     uint8_t key[VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN];
 176     uint8_t auth_key[VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN];
 177 } VhostUserCryptoSession;
 178
 179 static VhostUserConfig c __attribute__ ((unused));
 180 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
 181                                    + sizeof(c.size) \
 182                                    + sizeof(c.flags))
 183
 184 typedef struct VhostUserVringArea {
 185     uint64_t u64;
 186     uint64_t size;
 187     uint64_t offset;
 188 } VhostUserVringArea;
 189
 190 typedef struct VhostUserInflight {
 191     uint64_t mmap_size;
 192     uint64_t mmap_offset;
 193     uint16_t num_queues;
 194     uint16_t queue_size;
 195 } VhostUserInflight;
 196
 197 typedef struct {
 198     VhostUserRequest request;
 199
 200 #define VHOST_USER_VERSION_MASK     (0x3)
 201 #define VHOST_USER_REPLY_MASK       (0x1<<2)
 202 #define VHOST_USER_NEED_REPLY_MASK  (0x1 << 3)
 203     uint32_t flags;
 204     uint32_t size; /* the following payload size */
 205 } QEMU_PACKED VhostUserHeader;
 206
 207 typedef union {
 208 #define VHOST_USER_VRING_IDX_MASK   (0xff)
 209 #define VHOST_USER_VRING_NOFD_MASK  (0x1<<8)
 210         uint64_t u64;
 211         struct vhost_vring_state state;
 212         struct vhost_vring_addr addr;
 213         VhostUserMemory memory;
 214         VhostUserMemRegMsg mem_reg;
 215         VhostUserLog log;
 216         struct vhost_iotlb_msg iotlb;
 217         VhostUserConfig config;
 218         VhostUserCryptoSession session;
 219         VhostUserVringArea area;
 220         VhostUserInflight inflight;
 221 } VhostUserPayload;
 222
 223 typedef struct VhostUserMsg {
 224     VhostUserHeader hdr;
 225     VhostUserPayload payload;
 226 } QEMU_PACKED VhostUserMsg;
 227
 228 static VhostUserMsg m __attribute__ ((unused));
 229 #define VHOST_USER_HDR_SIZE (sizeof(VhostUserHeader))
 230
 231 #define VHOST_USER_PAYLOAD_SIZE (sizeof(VhostUserPayload))
 232
 233 /* The version of the protocol we support */
 234 #define VHOST_USER_VERSION    (0x1)
 235
 236 struct vhost_user {
 237     struct vhost_dev *dev;
 238     /* Shared between vhost devs of the same virtio device */
 239     VhostUserState *user;
 240     int slave_fd;
 241     NotifierWithReturn postcopy_notifier;
 242     struct PostCopyFD  postcopy_fd;
 243     uint64_t           postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
 244     /* Length of the region_rb and region_rb_offset arrays */
 245     size_t             region_rb_len;
 246     /* RAMBlock associated with a given region */
 247     RAMBlock         **region_rb;
 248     /* The offset from the start of the RAMBlock to the start of the
 249      * vhost region.
 250      */
 251     ram_addr_t        *region_rb_offset;
 252
 253     /* True once we've entered postcopy_listen */
 254     bool               postcopy_listen;
 255
 256     /* Our current regions */
 257     int num_shadow_regions;
 258     struct vhost_memory_region shadow_regions[VHOST_USER_MAX_RAM_SLOTS];
 259 };
 260
 261 struct scrub_regions {
 262     struct vhost_memory_region *region;
 263     int reg_idx;
 264     int fd_idx;
 265 };
 266
 267 static bool ioeventfd_enabled(void)
 268 {
 269     return !kvm_enabled() || kvm_eventfds_enabled();
 270 }
 271
 272 static int vhost_user_read_header(struct vhost_dev *dev, VhostUserMsg *msg)
 273 {
 274     struct vhost_user *u = dev->opaque;
 275     CharBackend *chr = u->user->chr;
 276     uint8_t *p = (uint8_t *) msg;
 277     int r, size = VHOST_USER_HDR_SIZE;
 278
 279     r = qemu_chr_fe_read_all(chr, p, size);
 280     if (r != size) {
 281         error_report("Failed to read msg header. Read %d instead of %d."
 282                      " Original request %d.", r, size, msg->hdr.request);
 283         return -1;
 284     }
 285
 286     /* validate received flags */
 287     if (msg->hdr.flags != (VHOST_USER_REPLY_MASK | VHOST_USER_VERSION)) {
 288         error_report("Failed to read msg header."
 289                 " Flags 0x%x instead of 0x%x.", msg->hdr.flags,
 290                 VHOST_USER_REPLY_MASK | VHOST_USER_VERSION);
 291         return -1;
 292     }
 293
 294     return 0;
 295 }
 296
 297 static int vhost_user_read(struct vhost_dev *dev, VhostUserMsg *msg)
 298 {
 299     struct vhost_user *u = dev->opaque;
 300     CharBackend *chr = u->user->chr;
 301     uint8_t *p = (uint8_t *) msg;
 302     int r, size;
 303
 304     if (vhost_user_read_header(dev, msg) < 0) {
 305         return -1;
 306     }
 307
 308     /* validate message size is sane */
 309     if (msg->hdr.size > VHOST_USER_PAYLOAD_SIZE) {
 310         error_report("Failed to read msg header."
 311                 " Size %d exceeds the maximum %zu.", msg->hdr.size,
 312                 VHOST_USER_PAYLOAD_SIZE);
 313         return -1;
 314     }
 315
 316     if (msg->hdr.size) {
 317         p += VHOST_USER_HDR_SIZE;
 318         size = msg->hdr.size;
 319         r = qemu_chr_fe_read_all(chr, p, size);
 320         if (r != size) {
 321             error_report("Failed to read msg payload."
 322                          " Read %d instead of %d.", r, msg->hdr.size);
 323             return -1;
 324         }
 325     }
 326
 327     return 0;
 328 }
 329
 330 static int process_message_reply(struct vhost_dev *dev,
 331                                  const VhostUserMsg *msg)
 332 {
 333     VhostUserMsg msg_reply;
 334
 335     if ((msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) == 0) {
 336         return 0;
 337     }
 338
 339     if (vhost_user_read(dev, &msg_reply) < 0) {
 340         return -1;
 341     }
 342
 343     if (msg_reply.hdr.request != msg->hdr.request) {
 344         error_report("Received unexpected msg type."
 345                      "Expected %d received %d",
 346                      msg->hdr.request, msg_reply.hdr.request);
 347         return -1;
 348     }
 349
 350     return msg_reply.payload.u64 ? -1 : 0;
 351 }
 352
 353 static bool vhost_user_one_time_request(VhostUserRequest request)
 354 {
 355     switch (request) {
 356     case VHOST_USER_SET_OWNER:
 357     case VHOST_USER_RESET_OWNER:
 358     case VHOST_USER_SET_MEM_TABLE:
 359     case VHOST_USER_GET_QUEUE_NUM:
 360     case VHOST_USER_NET_SET_MTU:
 361         return true;
 362     default:
 363         return false;
 364     }
 365 }
 366
 367 /* most non-init callers ignore the error */
 368 static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
 369                             int *fds, int fd_num)
 370 {
 371     struct vhost_user *u = dev->opaque;
 372     CharBackend *chr = u->user->chr;
 373     int ret, size = VHOST_USER_HDR_SIZE + msg->hdr.size;
 374
 375     /*
 376      * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
 377      * we just need send it once in the first time. For later such
 378      * request, we just ignore it.
 379      */
 380     if (vhost_user_one_time_request(msg->hdr.request) && dev->vq_index != 0) {
 381         msg->hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
 382         return 0;
 383     }
 384
 385     if (qemu_chr_fe_set_msgfds(chr, fds, fd_num) < 0) {
 386         error_report("Failed to set msg fds.");
 387         return -1;
 388     }
 389
 390     ret = qemu_chr_fe_write_all(chr, (const uint8_t *) msg, size);
 391     if (ret != size) {
 392         error_report("Failed to write msg."
 393                      " Wrote %d instead of %d.", ret, size);
 394         return -1;
 395     }
 396
 397     return 0;
 398 }
 399
 400 int vhost_user_gpu_set_socket(struct vhost_dev *dev, int fd)
 401 {
 402     VhostUserMsg msg = {
 403         .hdr.request = VHOST_USER_GPU_SET_SOCKET,
 404         .hdr.flags = VHOST_USER_VERSION,
 405     };
 406
 407     return vhost_user_write(dev, &msg, &fd, 1);
 408 }
 409
 410 static int vhost_user_set_log_base(struct vhost_dev *dev, uint64_t base,
 411                                    struct vhost_log *log)
 412 {
 413     int fds[VHOST_USER_MAX_RAM_SLOTS];
 414     size_t fd_num = 0;
 415     bool shmfd = virtio_has_feature(dev->protocol_features,
 416                                     VHOST_USER_PROTOCOL_F_LOG_SHMFD);
 417     VhostUserMsg msg = {
 418         .hdr.request = VHOST_USER_SET_LOG_BASE,
 419         .hdr.flags = VHOST_USER_VERSION,
 420         .payload.log.mmap_size = log->size * sizeof(*(log->log)),
 421         .payload.log.mmap_offset = 0,
 422         .hdr.size = sizeof(msg.payload.log),
 423     };
 424
 425     if (shmfd && log->fd != -1) {
 426         fds[fd_num++] = log->fd;
 427     }
 428
 429     if (vhost_user_write(dev, &msg, fds, fd_num) < 0) {
 430         return -1;
 431     }
 432
 433     if (shmfd) {
 434         msg.hdr.size = 0;
 435         if (vhost_user_read(dev, &msg) < 0) {
 436             return -1;
 437         }
 438
 439         if (msg.hdr.request != VHOST_USER_SET_LOG_BASE) {
 440             error_report("Received unexpected msg type. "
 441                          "Expected %d received %d",
 442                          VHOST_USER_SET_LOG_BASE, msg.hdr.request);
 443             return -1;
 444         }
 445     }
 446
 447     return 0;
 448 }
 449
 450 static MemoryRegion *vhost_user_get_mr_data(uint64_t addr, ram_addr_t *offset,
 451                                             int *fd)
 452 {
 453     MemoryRegion *mr;
 454
 455     assert((uintptr_t)addr == addr);
 456     mr = memory_region_from_host((void *)(uintptr_t)addr, offset);
 457     *fd = memory_region_get_fd(mr);
 458
 459     return mr;
 460 }
 461
 462 static void vhost_user_fill_msg_region(VhostUserMemoryRegion *dst,
 463                                        struct vhost_memory_region *src,
 464                                        uint64_t mmap_offset)
 465 {
 466     assert(src != NULL && dst != NULL);
 467     dst->userspace_addr = src->userspace_addr;
 468     dst->memory_size = src->memory_size;
 469     dst->guest_phys_addr = src->guest_phys_addr;
 470     dst->mmap_offset = mmap_offset;
 471 }
 472
 473 static int vhost_user_fill_set_mem_table_msg(struct vhost_user *u,
 474                                              struct vhost_dev *dev,
 475                                              VhostUserMsg *msg,
 476                                              int *fds, size_t *fd_num,
 477                                              bool track_ramblocks)
 478 {
 479     int i, fd;
 480     ram_addr_t offset;
 481     MemoryRegion *mr;
 482     struct vhost_memory_region *reg;
 483     VhostUserMemoryRegion region_buffer;
 484
 485     msg->hdr.request = VHOST_USER_SET_MEM_TABLE;
 486
 487     for (i = 0; i < dev->mem->nregions; ++i) {
 488         reg = dev->mem->regions + i;
 489
 490         mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
 491         if (fd > 0) {
 492             if (track_ramblocks) {
 493                 assert(*fd_num < VHOST_MEMORY_BASELINE_NREGIONS);
 494                 trace_vhost_user_set_mem_table_withfd(*fd_num, mr->name,
 495                                                       reg->memory_size,
 496                                                       reg->guest_phys_addr,
 497                                                       reg->userspace_addr,
 498                                                       offset);
 499                 u->region_rb_offset[i] = offset;
 500                 u->region_rb[i] = mr->ram_block;
 501             } else if (*fd_num == VHOST_MEMORY_BASELINE_NREGIONS) {
 502                 error_report("Failed preparing vhost-user memory table msg");
 503                 return -1;
 504             }
 505             vhost_user_fill_msg_region(&region_buffer, reg, offset);
 506             msg->payload.memory.regions[*fd_num] = region_buffer;
 507             fds[(*fd_num)++] = fd;
 508         } else if (track_ramblocks) {
 509             u->region_rb_offset[i] = 0;
 510             u->region_rb[i] = NULL;
 511         }
 512     }
 513
 514     msg->payload.memory.nregions = *fd_num;
 515
 516     if (!*fd_num) {
 517         error_report("Failed initializing vhost-user memory map, "
 518                      "consider using -object memory-backend-file share=on");
 519         return -1;
 520     }
 521
 522     msg->hdr.size = sizeof(msg->payload.memory.nregions);
 523     msg->hdr.size += sizeof(msg->payload.memory.padding);
 524     msg->hdr.size += *fd_num * sizeof(VhostUserMemoryRegion);
 525
 526     return 1;
 527 }
 528
 529 static inline bool reg_equal(struct vhost_memory_region *shadow_reg,
 530                              struct vhost_memory_region *vdev_reg)
 531 {
 532     return shadow_reg->guest_phys_addr == vdev_reg->guest_phys_addr &&
 533         shadow_reg->userspace_addr == vdev_reg->userspace_addr &&
 534         shadow_reg->memory_size == vdev_reg->memory_size;
 535 }
 536
 537 static void scrub_shadow_regions(struct vhost_dev *dev,
 538                                  struct scrub_regions *add_reg,
 539                                  int *nr_add_reg,
 540                                  struct scrub_regions *rem_reg,
 541                                  int *nr_rem_reg, uint64_t *shadow_pcb,
 542                                  bool track_ramblocks)
 543 {
 544     struct vhost_user *u = dev->opaque;
 545     bool found[VHOST_USER_MAX_RAM_SLOTS] = {};
 546     struct vhost_memory_region *reg, *shadow_reg;
 547     int i, j, fd, add_idx = 0, rm_idx = 0, fd_num = 0;
 548     ram_addr_t offset;
 549     MemoryRegion *mr;
 550     bool matching;
 551
 552     /*
 553      * Find memory regions present in our shadow state which are not in
 554      * the device's current memory state.
 555      *
 556      * Mark regions in both the shadow and device state as "found".
 557      */
 558     for (i = 0; i < u->num_shadow_regions; i++) {
 559         shadow_reg = &u->shadow_regions[i];
 560         matching = false;
 561
 562         for (j = 0; j < dev->mem->nregions; j++) {
 563             reg = &dev->mem->regions[j];
 564
 565             mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
 566
 567             if (reg_equal(shadow_reg, reg)) {
 568                 matching = true;
 569                 found[j] = true;
 570                 if (track_ramblocks) {
 571                     /*
 572                      * Reset postcopy client bases, region_rb, and
 573                      * region_rb_offset in case regions are removed.
 574                      */
 575                     if (fd > 0) {
 576                         u->region_rb_offset[j] = offset;
 577                         u->region_rb[j] = mr->ram_block;
 578                         shadow_pcb[j] = u->postcopy_client_bases[i];
 579                     } else {
 580                         u->region_rb_offset[j] = 0;
 581                         u->region_rb[j] = NULL;
 582                     }
 583                 }
 584                 break;
 585             }
 586         }
 587
 588         /*
 589          * If the region was not found in the current device memory state
 590          * create an entry for it in the removed list.
 591          */
 592         if (!matching) {
 593             rem_reg[rm_idx].region = shadow_reg;
 594             rem_reg[rm_idx++].reg_idx = i;
 595         }
 596     }
 597
 598     /*
 599      * For regions not marked "found", create entries in the added list.
 600      *
 601      * Note their indexes in the device memory state and the indexes of their
 602      * file descriptors.
 603      */
 604     for (i = 0; i < dev->mem->nregions; i++) {
 605         reg = &dev->mem->regions[i];
 606         vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
 607         if (fd > 0) {
 608             ++fd_num;
 609         }
 610
 611         /*
 612          * If the region was in both the shadow and device state we don't
 613          * need to send a VHOST_USER_ADD_MEM_REG message for it.
 614          */
 615         if (found[i]) {
 616             continue;
 617         }
 618
 619         add_reg[add_idx].region = reg;
 620         add_reg[add_idx].reg_idx = i;
 621         add_reg[add_idx++].fd_idx = fd_num;
 622     }
 623     *nr_rem_reg = rm_idx;
 624     *nr_add_reg = add_idx;
 625
 626     return;
 627 }
 628
 629 static int send_remove_regions(struct vhost_dev *dev,
 630                                struct scrub_regions *remove_reg,
 631                                int nr_rem_reg, VhostUserMsg *msg,
 632                                bool reply_supported)
 633 {
 634     struct vhost_user *u = dev->opaque;
 635     struct vhost_memory_region *shadow_reg;
 636     int i, fd, shadow_reg_idx, ret;
 637     ram_addr_t offset;
 638     VhostUserMemoryRegion region_buffer;
 639
 640     /*
 641      * The regions in remove_reg appear in the same order they do in the
 642      * shadow table. Therefore we can minimize memory copies by iterating
 643      * through remove_reg backwards.
 644      */
 645     for (i = nr_rem_reg - 1; i >= 0; i--) {
 646         shadow_reg = remove_reg[i].region;
 647         shadow_reg_idx = remove_reg[i].reg_idx;
 648
 649         vhost_user_get_mr_data(shadow_reg->userspace_addr, &offset, &fd);
 650
 651         if (fd > 0) {
 652             msg->hdr.request = VHOST_USER_REM_MEM_REG;
 653             vhost_user_fill_msg_region(&region_buffer, shadow_reg, 0);
 654             msg->payload.mem_reg.region = region_buffer;
 655
 656             if (vhost_user_write(dev, msg, &fd, 1) < 0) {
 657                 return -1;
 658             }
 659
 660             if (reply_supported) {
 661                 ret = process_message_reply(dev, msg);
 662                 if (ret) {
 663                     return ret;
 664                 }
 665             }
 666         }
 667
 668         /*
 669          * At this point we know the backend has unmapped the region. It is now
 670          * safe to remove it from the shadow table.
 671          */
 672         memmove(&u->shadow_regions[shadow_reg_idx],
 673                 &u->shadow_regions[shadow_reg_idx + 1],
 674                 sizeof(struct vhost_memory_region) *
 675                 (u->num_shadow_regions - shadow_reg_idx - 1));
 676         u->num_shadow_regions--;
 677     }
 678
 679     return 0;
 680 }
 681
 682 static int send_add_regions(struct vhost_dev *dev,
 683                             struct scrub_regions *add_reg, int nr_add_reg,
 684                             VhostUserMsg *msg, uint64_t *shadow_pcb,
 685                             bool reply_supported, bool track_ramblocks)
 686 {
 687     struct vhost_user *u = dev->opaque;
 688     int i, fd, ret, reg_idx, reg_fd_idx;
 689     struct vhost_memory_region *reg;
 690     MemoryRegion *mr;
 691     ram_addr_t offset;
 692     VhostUserMsg msg_reply;
 693     VhostUserMemoryRegion region_buffer;
 694
 695     for (i = 0; i < nr_add_reg; i++) {
 696         reg = add_reg[i].region;
 697         reg_idx = add_reg[i].reg_idx;
 698         reg_fd_idx = add_reg[i].fd_idx;
 699
 700         mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
 701
 702         if (fd > 0) {
 703             if (track_ramblocks) {
 704                 trace_vhost_user_set_mem_table_withfd(reg_fd_idx, mr->name,
 705                                                       reg->memory_size,
 706                                                       reg->guest_phys_addr,
 707                                                       reg->userspace_addr,
 708                                                       offset);
 709                 u->region_rb_offset[reg_idx] = offset;
 710                 u->region_rb[reg_idx] = mr->ram_block;
 711             }
 712             msg->hdr.request = VHOST_USER_ADD_MEM_REG;
 713             vhost_user_fill_msg_region(&region_buffer, reg, offset);
 714             msg->payload.mem_reg.region = region_buffer;
 715
 716             if (vhost_user_write(dev, msg, &fd, 1) < 0) {
 717                 return -1;
 718             }
 719
 720             if (track_ramblocks) {
 721                 uint64_t reply_gpa;
 722
 723                 if (vhost_user_read(dev, &msg_reply) < 0) {
 724                     return -1;
 725                 }
 726
 727                 reply_gpa = msg_reply.payload.mem_reg.region.guest_phys_addr;
 728
 729                 if (msg_reply.hdr.request != VHOST_USER_ADD_MEM_REG) {
 730                     error_report("%s: Received unexpected msg type."
 731                                  "Expected %d received %d", __func__,
 732                                  VHOST_USER_ADD_MEM_REG,
 733                                  msg_reply.hdr.request);
 734                     return -1;
 735                 }
 736
 737                 /*
 738                  * We're using the same structure, just reusing one of the
 739                  * fields, so it should be the same size.
 740                  */
 741                 if (msg_reply.hdr.size != msg->hdr.size) {
 742                     error_report("%s: Unexpected size for postcopy reply "
 743                                  "%d vs %d", __func__, msg_reply.hdr.size,
 744                                  msg->hdr.size);
 745                     return -1;
 746                 }
 747
 748                 /* Get the postcopy client base from the backend's reply. */
 749                 if (reply_gpa == dev->mem->regions[reg_idx].guest_phys_addr) {
 750                     shadow_pcb[reg_idx] =
 751                         msg_reply.payload.mem_reg.region.userspace_addr;
 752                     trace_vhost_user_set_mem_table_postcopy(
 753                         msg_reply.payload.mem_reg.region.userspace_addr,
 754                         msg->payload.mem_reg.region.userspace_addr,
 755                         reg_fd_idx, reg_idx);
 756                 } else {
 757                     error_report("%s: invalid postcopy reply for region. "
 758                                  "Got guest physical address %" PRIX64 ", expected "
 759                                  "%" PRIX64, __func__, reply_gpa,
 760                                  dev->mem->regions[reg_idx].guest_phys_addr);
 761                     return -1;
 762                 }
 763             } else if (reply_supported) {
 764                 ret = process_message_reply(dev, msg);
 765                 if (ret) {
 766                     return ret;
 767                 }
 768             }
 769         } else if (track_ramblocks) {
 770             u->region_rb_offset[reg_idx] = 0;
 771             u->region_rb[reg_idx] = NULL;
 772         }
 773
 774         /*
 775          * At this point, we know the backend has mapped in the new
 776          * region, if the region has a valid file descriptor.
 777          *
 778          * The region should now be added to the shadow table.
 779          */
 780         u->shadow_regions[u->num_shadow_regions].guest_phys_addr =
 781             reg->guest_phys_addr;
 782         u->shadow_regions[u->num_shadow_regions].userspace_addr =
 783             reg->userspace_addr;
 784         u->shadow_regions[u->num_shadow_regions].memory_size =
 785             reg->memory_size;
 786         u->num_shadow_regions++;
 787     }
 788
 789     return 0;
 790 }
 791
 792 static int vhost_user_add_remove_regions(struct vhost_dev *dev,
 793                                          VhostUserMsg *msg,
 794                                          bool reply_supported,
 795                                          bool track_ramblocks)
 796 {
 797     struct vhost_user *u = dev->opaque;
 798     struct scrub_regions add_reg[VHOST_USER_MAX_RAM_SLOTS];
 799     struct scrub_regions rem_reg[VHOST_USER_MAX_RAM_SLOTS];
 800     uint64_t shadow_pcb[VHOST_USER_MAX_RAM_SLOTS] = {};
 801     int nr_add_reg, nr_rem_reg;
 802
 803     msg->hdr.size = sizeof(msg->payload.mem_reg);
 804
 805     /* Find the regions which need to be removed or added. */
 806     scrub_shadow_regions(dev, add_reg, &nr_add_reg, rem_reg, &nr_rem_reg,
 807                          shadow_pcb, track_ramblocks);
 808
 809     if (nr_rem_reg && send_remove_regions(dev, rem_reg, nr_rem_reg, msg,
 810                 reply_supported) < 0)
 811     {
 812         goto err;
 813     }
 814
 815     if (nr_add_reg && send_add_regions(dev, add_reg, nr_add_reg, msg,
 816                 shadow_pcb, reply_supported, track_ramblocks) < 0)
 817     {
 818         goto err;
 819     }
 820
 821     if (track_ramblocks) {
 822         memcpy(u->postcopy_client_bases, shadow_pcb,
 823                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
 824         /*
 825          * Now we've registered this with the postcopy code, we ack to the
 826          * client, because now we're in the position to be able to deal with
 827          * any faults it generates.
 828          */
 829         /* TODO: Use this for failure cases as well with a bad value. */
 830         msg->hdr.size = sizeof(msg->payload.u64);
 831         msg->payload.u64 = 0; /* OK */
 832
 833         if (vhost_user_write(dev, msg, NULL, 0) < 0) {
 834             return -1;
 835         }
 836     }
 837
 838     return 0;
 839
 840 err:
 841     if (track_ramblocks) {
 842         memcpy(u->postcopy_client_bases, shadow_pcb,
 843                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
 844     }
 845
 846     return -1;
 847 }
 848
 849 static int vhost_user_set_mem_table_postcopy(struct vhost_dev *dev,
 850                                              struct vhost_memory *mem,
 851                                              bool reply_supported,
 852                                              bool config_mem_slots)
 853 {
 854     struct vhost_user *u = dev->opaque;
 855     int fds[VHOST_MEMORY_BASELINE_NREGIONS];
 856     size_t fd_num = 0;
 857     VhostUserMsg msg_reply;
 858     int region_i, msg_i;
 859
 860     VhostUserMsg msg = {
 861         .hdr.flags = VHOST_USER_VERSION,
 862     };
 863
 864     if (u->region_rb_len < dev->mem->nregions) {
 865         u->region_rb = g_renew(RAMBlock*, u->region_rb, dev->mem->nregions);
 866         u->region_rb_offset = g_renew(ram_addr_t, u->region_rb_offset,
 867                                       dev->mem->nregions);
 868         memset(&(u->region_rb[u->region_rb_len]), '\0',
 869                sizeof(RAMBlock *) * (dev->mem->nregions - u->region_rb_len));
 870         memset(&(u->region_rb_offset[u->region_rb_len]), '\0',
 871                sizeof(ram_addr_t) * (dev->mem->nregions - u->region_rb_len));
 872         u->region_rb_len = dev->mem->nregions;
 873     }
 874
 875     if (config_mem_slots) {
 876         if (vhost_user_add_remove_regions(dev, &msg, reply_supported,
 877                                           true) < 0) {
 878             return -1;
 879         }
 880     } else {
 881         if (vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
 882                                               true) < 0) {
 883             return -1;
 884         }
 885
 886         if (vhost_user_write(dev, &msg, fds, fd_num) < 0) {
 887             return -1;
 888         }
 889
 890         if (vhost_user_read(dev, &msg_reply) < 0) {
 891             return -1;
 892         }
 893
 894         if (msg_reply.hdr.request != VHOST_USER_SET_MEM_TABLE) {
 895             error_report("%s: Received unexpected msg type."
 896                          "Expected %d received %d", __func__,
 897                          VHOST_USER_SET_MEM_TABLE, msg_reply.hdr.request);
 898             return -1;
 899         }
 900
 901         /*
 902          * We're using the same structure, just reusing one of the
 903          * fields, so it should be the same size.
 904          */
 905         if (msg_reply.hdr.size != msg.hdr.size) {
 906             error_report("%s: Unexpected size for postcopy reply "
 907                          "%d vs %d", __func__, msg_reply.hdr.size,
 908                          msg.hdr.size);
 909             return -1;
 910         }
 911
 912         memset(u->postcopy_client_bases, 0,
 913                sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
 914
 915         /*
 916          * They're in the same order as the regions that were sent
 917          * but some of the regions were skipped (above) if they
 918          * didn't have fd's
 919          */
 920         for (msg_i = 0, region_i = 0;
 921              region_i < dev->mem->nregions;
 922              region_i++) {
 923             if (msg_i < fd_num &&
 924                 msg_reply.payload.memory.regions[msg_i].guest_phys_addr ==
 925                 dev->mem->regions[region_i].guest_phys_addr) {
 926                 u->postcopy_client_bases[region_i] =
 927                     msg_reply.payload.memory.regions[msg_i].userspace_addr;
 928                 trace_vhost_user_set_mem_table_postcopy(
 929                     msg_reply.payload.memory.regions[msg_i].userspace_addr,
 930                     msg.payload.memory.regions[msg_i].userspace_addr,
 931                     msg_i, region_i);
 932                 msg_i++;
 933             }
 934         }
 935         if (msg_i != fd_num) {
 936             error_report("%s: postcopy reply not fully consumed "
 937                          "%d vs %zd",
 938                          __func__, msg_i, fd_num);
 939             return -1;
 940         }
 941
 942         /*
 943          * Now we've registered this with the postcopy code, we ack to the
 944          * client, because now we're in the position to be able to deal
 945          * with any faults it generates.
 946          */
 947         /* TODO: Use this for failure cases as well with a bad value. */
 948         msg.hdr.size = sizeof(msg.payload.u64);
 949         msg.payload.u64 = 0; /* OK */
 950         if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
 951             return -1;
 952         }
 953     }
 954
 955     return 0;
 956 }
 957
 958 static int vhost_user_set_mem_table(struct vhost_dev *dev,
 959                                     struct vhost_memory *mem)
 960 {
 961     struct vhost_user *u = dev->opaque;
 962     int fds[VHOST_MEMORY_BASELINE_NREGIONS];
 963     size_t fd_num = 0;
 964     bool do_postcopy = u->postcopy_listen && u->postcopy_fd.handler;
 965     bool reply_supported = virtio_has_feature(dev->protocol_features,
 966                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
 967     bool config_mem_slots =
 968         virtio_has_feature(dev->protocol_features,
 969                            VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS);
 970
 971     if (do_postcopy) {
 972         /*
 973          * Postcopy has enough differences that it's best done in it's own
 974          * version
 975          */
 976         return vhost_user_set_mem_table_postcopy(dev, mem, reply_supported,
 977                                                  config_mem_slots);
 978     }
 979
 980     VhostUserMsg msg = {
 981         .hdr.flags = VHOST_USER_VERSION,
 982     };
 983
 984     if (reply_supported) {
 985         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
 986     }
 987
 988     if (config_mem_slots) {
 989         if (vhost_user_add_remove_regions(dev, &msg, reply_supported,
 990                                           false) < 0) {
 991             return -1;
 992         }
 993     } else {
 994         if (vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
 995                                               false) < 0) {
 996             return -1;
 997         }
 998         if (vhost_user_write(dev, &msg, fds, fd_num) < 0) {
 999             return -1;
1000         }
1001
1002         if (reply_supported) {
1003             return process_message_reply(dev, &msg);
1004         }
1005     }
1006
1007     return 0;
1008 }
1009
1010 static int vhost_user_set_vring_addr(struct vhost_dev *dev,
1011                                      struct vhost_vring_addr *addr)
1012 {
1013     VhostUserMsg msg = {
1014         .hdr.request = VHOST_USER_SET_VRING_ADDR,
1015         .hdr.flags = VHOST_USER_VERSION,
1016         .payload.addr = *addr,
1017         .hdr.size = sizeof(msg.payload.addr),
1018     };
1019
1020     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1021         return -1;
1022     }
1023
1024     return 0;
1025 }
1026
1027 static int vhost_user_set_vring_endian(struct vhost_dev *dev,
1028                                        struct vhost_vring_state *ring)
1029 {
1030     bool cross_endian = virtio_has_feature(dev->protocol_features,
1031                                            VHOST_USER_PROTOCOL_F_CROSS_ENDIAN);
1032     VhostUserMsg msg = {
1033         .hdr.request = VHOST_USER_SET_VRING_ENDIAN,
1034         .hdr.flags = VHOST_USER_VERSION,
1035         .payload.state = *ring,
1036         .hdr.size = sizeof(msg.payload.state),
1037     };
1038
1039     if (!cross_endian) {
1040         error_report("vhost-user trying to send unhandled ioctl");
1041         return -1;
1042     }
1043
1044     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1045         return -1;
1046     }
1047
1048     return 0;
1049 }
1050
1051 static int vhost_set_vring(struct vhost_dev *dev,
1052                            unsigned long int request,
1053                            struct vhost_vring_state *ring)
1054 {
1055     VhostUserMsg msg = {
1056         .hdr.request = request,
1057         .hdr.flags = VHOST_USER_VERSION,
1058         .payload.state = *ring,
1059         .hdr.size = sizeof(msg.payload.state),
1060     };
1061
1062     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1063         return -1;
1064     }
1065
1066     return 0;
1067 }
1068
1069 static int vhost_user_set_vring_num(struct vhost_dev *dev,
1070                                     struct vhost_vring_state *ring)
1071 {
1072     return vhost_set_vring(dev, VHOST_USER_SET_VRING_NUM, ring);
1073 }
1074
1075 static void vhost_user_host_notifier_restore(struct vhost_dev *dev,
1076                                              int queue_idx)
1077 {
1078     struct vhost_user *u = dev->opaque;
1079     VhostUserHostNotifier *n = &u->user->notifier[queue_idx];
1080     VirtIODevice *vdev = dev->vdev;
1081
1082     if (n->addr && !n->set) {
1083         virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, true);
1084         n->set = true;
1085     }
1086 }
1087
1088 static void vhost_user_host_notifier_remove(struct vhost_dev *dev,
1089                                             int queue_idx)
1090 {
1091     struct vhost_user *u = dev->opaque;
1092     VhostUserHostNotifier *n = &u->user->notifier[queue_idx];
1093     VirtIODevice *vdev = dev->vdev;
1094
1095     if (n->addr && n->set) {
1096         virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, false);
1097         n->set = false;
1098     }
1099 }
1100
1101 static int vhost_user_set_vring_base(struct vhost_dev *dev,
1102                                      struct vhost_vring_state *ring)
1103 {
1104     vhost_user_host_notifier_restore(dev, ring->index);
1105
1106     return vhost_set_vring(dev, VHOST_USER_SET_VRING_BASE, ring);
1107 }
1108
1109 static int vhost_user_set_vring_enable(struct vhost_dev *dev, int enable)
1110 {
1111     int i;
1112
1113     if (!virtio_has_feature(dev->features, VHOST_USER_F_PROTOCOL_FEATURES)) {
1114         return -1;
1115     }
1116
1117     for (i = 0; i < dev->nvqs; ++i) {
1118         struct vhost_vring_state state = {
1119             .index = dev->vq_index + i,
1120             .num   = enable,
1121         };
1122
1123         vhost_set_vring(dev, VHOST_USER_SET_VRING_ENABLE, &state);
1124     }
1125
1126     return 0;
1127 }
1128
1129 static int vhost_user_get_vring_base(struct vhost_dev *dev,
1130                                      struct vhost_vring_state *ring)
1131 {
1132     VhostUserMsg msg = {
1133         .hdr.request = VHOST_USER_GET_VRING_BASE,
1134         .hdr.flags = VHOST_USER_VERSION,
1135         .payload.state = *ring,
1136         .hdr.size = sizeof(msg.payload.state),
1137     };
1138
1139     vhost_user_host_notifier_remove(dev, ring->index);
1140
1141     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1142         return -1;
1143     }
1144
1145     if (vhost_user_read(dev, &msg) < 0) {
1146         return -1;
1147     }
1148
1149     if (msg.hdr.request != VHOST_USER_GET_VRING_BASE) {
1150         error_report("Received unexpected msg type. Expected %d received %d",
1151                      VHOST_USER_GET_VRING_BASE, msg.hdr.request);
1152         return -1;
1153     }
1154
1155     if (msg.hdr.size != sizeof(msg.payload.state)) {
1156         error_report("Received bad msg size.");
1157         return -1;
1158     }
1159
1160     *ring = msg.payload.state;
1161
1162     return 0;
1163 }
1164
1165 static int vhost_set_vring_file(struct vhost_dev *dev,
1166                                 VhostUserRequest request,
1167                                 struct vhost_vring_file *file)
1168 {
1169     int fds[VHOST_USER_MAX_RAM_SLOTS];
1170     size_t fd_num = 0;
1171     VhostUserMsg msg = {
1172         .hdr.request = request,
1173         .hdr.flags = VHOST_USER_VERSION,
1174         .payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK,
1175         .hdr.size = sizeof(msg.payload.u64),
1176     };
1177
1178     if (ioeventfd_enabled() && file->fd > 0) {
1179         fds[fd_num++] = file->fd;
1180     } else {
1181         msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
1182     }
1183
1184     if (vhost_user_write(dev, &msg, fds, fd_num) < 0) {
1185         return -1;
1186     }
1187
1188     return 0;
1189 }
1190
1191 static int vhost_user_set_vring_kick(struct vhost_dev *dev,
1192                                      struct vhost_vring_file *file)
1193 {
1194     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_KICK, file);
1195 }
1196
1197 static int vhost_user_set_vring_call(struct vhost_dev *dev,
1198                                      struct vhost_vring_file *file)
1199 {
1200     return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_CALL, file);
1201 }
1202
1203 static int vhost_user_set_u64(struct vhost_dev *dev, int request, uint64_t u64)
1204 {
1205     VhostUserMsg msg = {
1206         .hdr.request = request,
1207         .hdr.flags = VHOST_USER_VERSION,
1208         .payload.u64 = u64,
1209         .hdr.size = sizeof(msg.payload.u64),
1210     };
1211
1212     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1213         return -1;
1214     }
1215
1216     return 0;
1217 }
1218
1219 static int vhost_user_set_features(struct vhost_dev *dev,
1220                                    uint64_t features)
1221 {
1222     return vhost_user_set_u64(dev, VHOST_USER_SET_FEATURES, features);
1223 }
1224
1225 static int vhost_user_set_protocol_features(struct vhost_dev *dev,
1226                                             uint64_t features)
1227 {
1228     return vhost_user_set_u64(dev, VHOST_USER_SET_PROTOCOL_FEATURES, features);
1229 }
1230
1231 static int vhost_user_get_u64(struct vhost_dev *dev, int request, uint64_t *u64)
1232 {
1233     VhostUserMsg msg = {
1234         .hdr.request = request,
1235         .hdr.flags = VHOST_USER_VERSION,
1236     };
1237
1238     if (vhost_user_one_time_request(request) && dev->vq_index != 0) {
1239         return 0;
1240     }
1241
1242     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1243         return -1;
1244     }
1245
1246     if (vhost_user_read(dev, &msg) < 0) {
1247         return -1;
1248     }
1249
1250     if (msg.hdr.request != request) {
1251         error_report("Received unexpected msg type. Expected %d received %d",
1252                      request, msg.hdr.request);
1253         return -1;
1254     }
1255
1256     if (msg.hdr.size != sizeof(msg.payload.u64)) {
1257         error_report("Received bad msg size.");
1258         return -1;
1259     }
1260
1261     *u64 = msg.payload.u64;
1262
1263     return 0;
1264 }
1265
1266 static int vhost_user_get_features(struct vhost_dev *dev, uint64_t *features)
1267 {
1268     return vhost_user_get_u64(dev, VHOST_USER_GET_FEATURES, features);
1269 }
1270
1271 static int vhost_user_set_owner(struct vhost_dev *dev)
1272 {
1273     VhostUserMsg msg = {
1274         .hdr.request = VHOST_USER_SET_OWNER,
1275         .hdr.flags = VHOST_USER_VERSION,
1276     };
1277
1278     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1279         return -1;
1280     }
1281
1282     return 0;
1283 }
1284
1285 static int vhost_user_get_max_memslots(struct vhost_dev *dev,
1286                                        uint64_t *max_memslots)
1287 {
1288     uint64_t backend_max_memslots;
1289     int err;
1290
1291     err = vhost_user_get_u64(dev, VHOST_USER_GET_MAX_MEM_SLOTS,
1292                              &backend_max_memslots);
1293     if (err < 0) {
1294         return err;
1295     }
1296
1297     *max_memslots = backend_max_memslots;
1298
1299     return 0;
1300 }
1301
1302 static int vhost_user_reset_device(struct vhost_dev *dev)
1303 {
1304     VhostUserMsg msg = {
1305         .hdr.flags = VHOST_USER_VERSION,
1306     };
1307
1308     msg.hdr.request = virtio_has_feature(dev->protocol_features,
1309                                          VHOST_USER_PROTOCOL_F_RESET_DEVICE)
1310         ? VHOST_USER_RESET_DEVICE
1311         : VHOST_USER_RESET_OWNER;
1312
1313     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1314         return -1;
1315     }
1316
1317     return 0;
1318 }
1319
1320 static int vhost_user_slave_handle_config_change(struct vhost_dev *dev)
1321 {
1322     int ret = -1;
1323
1324     if (!dev->config_ops) {
1325         return -1;
1326     }
1327
1328     if (dev->config_ops->vhost_dev_config_notifier) {
1329         ret = dev->config_ops->vhost_dev_config_notifier(dev);
1330     }
1331
1332     return ret;
1333 }
1334
1335 static int vhost_user_slave_handle_vring_host_notifier(struct vhost_dev *dev,
1336                                                        VhostUserVringArea *area,
1337                                                        int fd)
1338 {
1339     int queue_idx = area->u64 & VHOST_USER_VRING_IDX_MASK;
1340     size_t page_size = qemu_real_host_page_size;
1341     struct vhost_user *u = dev->opaque;
1342     VhostUserState *user = u->user;
1343     VirtIODevice *vdev = dev->vdev;
1344     VhostUserHostNotifier *n;
1345     void *addr;
1346     char *name;
1347
1348     if (!virtio_has_feature(dev->protocol_features,
1349                             VHOST_USER_PROTOCOL_F_HOST_NOTIFIER) ||
1350         vdev == NULL || queue_idx >= virtio_get_num_queues(vdev)) {
1351         return -1;
1352     }
1353
1354     n = &user->notifier[queue_idx];
1355
1356     if (n->addr) {
1357         virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, false);
1358         object_unparent(OBJECT(&n->mr));
1359         munmap(n->addr, page_size);
1360         n->addr = NULL;
1361     }
1362
1363     if (area->u64 & VHOST_USER_VRING_NOFD_MASK) {
1364         return 0;
1365     }
1366
1367     /* Sanity check. */
1368     if (area->size != page_size) {
1369         return -1;
1370     }
1371
1372     addr = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1373                 fd, area->offset);
1374     if (addr == MAP_FAILED) {
1375         return -1;
1376     }
1377
1378     name = g_strdup_printf("vhost-user/host-notifier@%p mmaps[%d]",
1379                            user, queue_idx);
1380     memory_region_init_ram_device_ptr(&n->mr, OBJECT(vdev), name,
1381                                       page_size, addr);
1382     g_free(name);
1383
1384     if (virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, true)) {
1385         munmap(addr, page_size);
1386         return -1;
1387     }
1388
1389     n->addr = addr;
1390     n->set = true;
1391
1392     return 0;
1393 }
1394
1395 static void slave_read(void *opaque)
1396 {
1397     struct vhost_dev *dev = opaque;
1398     struct vhost_user *u = dev->opaque;
1399     VhostUserHeader hdr = { 0, };
1400     VhostUserPayload payload = { 0, };
1401     int size, ret = 0;
1402     struct iovec iov;
1403     struct msghdr msgh;
1404     int fd[VHOST_USER_SLAVE_MAX_FDS];
1405     char control[CMSG_SPACE(sizeof(fd))];
1406     struct cmsghdr *cmsg;
1407     int i, fdsize = 0;
1408
1409     memset(&msgh, 0, sizeof(msgh));
1410     msgh.msg_iov = &iov;
1411     msgh.msg_iovlen = 1;
1412     msgh.msg_control = control;
1413     msgh.msg_controllen = sizeof(control);
1414
1415     memset(fd, -1, sizeof(fd));
1416
1417     /* Read header */
1418     iov.iov_base = &hdr;
1419     iov.iov_len = VHOST_USER_HDR_SIZE;
1420
1421     do {
1422         size = recvmsg(u->slave_fd, &msgh, 0);
1423     } while (size < 0 && (errno == EINTR || errno == EAGAIN));
1424
1425     if (size != VHOST_USER_HDR_SIZE) {
1426         error_report("Failed to read from slave.");
1427         goto err;
1428     }
1429
1430     if (msgh.msg_flags & MSG_CTRUNC) {
1431         error_report("Truncated message.");
1432         goto err;
1433     }
1434
1435     for (cmsg = CMSG_FIRSTHDR(&msgh); cmsg != NULL;
1436          cmsg = CMSG_NXTHDR(&msgh, cmsg)) {
1437             if (cmsg->cmsg_level == SOL_SOCKET &&
1438                 cmsg->cmsg_type == SCM_RIGHTS) {
1439                     fdsize = cmsg->cmsg_len - CMSG_LEN(0);
1440                     memcpy(fd, CMSG_DATA(cmsg), fdsize);
1441                     break;
1442             }
1443     }
1444
1445     if (hdr.size > VHOST_USER_PAYLOAD_SIZE) {
1446         error_report("Failed to read msg header."
1447                 " Size %d exceeds the maximum %zu.", hdr.size,
1448                 VHOST_USER_PAYLOAD_SIZE);
1449         goto err;
1450     }
1451
1452     /* Read payload */
1453     do {
1454         size = read(u->slave_fd, &payload, hdr.size);
1455     } while (size < 0 && (errno == EINTR || errno == EAGAIN));
1456
1457     if (size != hdr.size) {
1458         error_report("Failed to read payload from slave.");
1459         goto err;
1460     }
1461
1462     switch (hdr.request) {
1463     case VHOST_USER_SLAVE_IOTLB_MSG:
1464         ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
1465         break;
1466     case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG :
1467         ret = vhost_user_slave_handle_config_change(dev);
1468         break;
1469     case VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG:
1470         ret = vhost_user_slave_handle_vring_host_notifier(dev, &payload.area,
1471                                                           fd[0]);
1472         break;
1473     default:
1474         error_report("Received unexpected msg type: %d.", hdr.request);
1475         ret = -EINVAL;
1476     }
1477
1478     /* Close the remaining file descriptors. */
1479     for (i = 0; i < fdsize; i++) {
1480         if (fd[i] != -1) {
1481             close(fd[i]);
1482         }
1483     }
1484
1485     /*
1486      * REPLY_ACK feature handling. Other reply types has to be managed
1487      * directly in their request handlers.
1488      */
1489     if (hdr.flags & VHOST_USER_NEED_REPLY_MASK) {
1490         struct iovec iovec[2];
1491
1492
1493         hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
1494         hdr.flags |= VHOST_USER_REPLY_MASK;
1495
1496         payload.u64 = !!ret;
1497         hdr.size = sizeof(payload.u64);
1498
1499         iovec[0].iov_base = &hdr;
1500         iovec[0].iov_len = VHOST_USER_HDR_SIZE;
1501         iovec[1].iov_base = &payload;
1502         iovec[1].iov_len = hdr.size;
1503
1504         do {
1505             size = writev(u->slave_fd, iovec, ARRAY_SIZE(iovec));
1506         } while (size < 0 && (errno == EINTR || errno == EAGAIN));
1507
1508         if (size != VHOST_USER_HDR_SIZE + hdr.size) {
1509             error_report("Failed to send msg reply to slave.");
1510             goto err;
1511         }
1512     }
1513
1514     return;
1515
1516 err:
1517     qemu_set_fd_handler(u->slave_fd, NULL, NULL, NULL);
1518     close(u->slave_fd);
1519     u->slave_fd = -1;
1520     for (i = 0; i < fdsize; i++) {
1521         if (fd[i] != -1) {
1522             close(fd[i]);
1523         }
1524     }
1525     return;
1526 }
1527
1528 static int vhost_setup_slave_channel(struct vhost_dev *dev)
1529 {
1530     VhostUserMsg msg = {
1531         .hdr.request = VHOST_USER_SET_SLAVE_REQ_FD,
1532         .hdr.flags = VHOST_USER_VERSION,
1533     };
1534     struct vhost_user *u = dev->opaque;
1535     int sv[2], ret = 0;
1536     bool reply_supported = virtio_has_feature(dev->protocol_features,
1537                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
1538
1539     if (!virtio_has_feature(dev->protocol_features,
1540                             VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
1541         return 0;
1542     }
1543
1544     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
1545         error_report("socketpair() failed");
1546         return -1;
1547     }
1548
1549     u->slave_fd = sv[0];
1550     qemu_set_fd_handler(u->slave_fd, slave_read, NULL, dev);
1551
1552     if (reply_supported) {
1553         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1554     }
1555
1556     ret = vhost_user_write(dev, &msg, &sv[1], 1);
1557     if (ret) {
1558         goto out;
1559     }
1560
1561     if (reply_supported) {
1562         ret = process_message_reply(dev, &msg);
1563     }
1564
1565 out:
1566     close(sv[1]);
1567     if (ret) {
1568         qemu_set_fd_handler(u->slave_fd, NULL, NULL, NULL);
1569         close(u->slave_fd);
1570         u->slave_fd = -1;
1571     }
1572
1573     return ret;
1574 }
1575
1576 #ifdef CONFIG_LINUX
1577 /*
1578  * Called back from the postcopy fault thread when a fault is received on our
1579  * ufd.
1580  * TODO: This is Linux specific
1581  */
1582 static int vhost_user_postcopy_fault_handler(struct PostCopyFD *pcfd,
1583                                              void *ufd)
1584 {
1585     struct vhost_dev *dev = pcfd->data;
1586     struct vhost_user *u = dev->opaque;
1587     struct uffd_msg *msg = ufd;
1588     uint64_t faultaddr = msg->arg.pagefault.address;
1589     RAMBlock *rb = NULL;
1590     uint64_t rb_offset;
1591     int i;
1592
1593     trace_vhost_user_postcopy_fault_handler(pcfd->idstr, faultaddr,
1594                                             dev->mem->nregions);
1595     for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1596         trace_vhost_user_postcopy_fault_handler_loop(i,
1597                 u->postcopy_client_bases[i], dev->mem->regions[i].memory_size);
1598         if (faultaddr >= u->postcopy_client_bases[i]) {
1599             /* Ofset of the fault address in the vhost region */
1600             uint64_t region_offset = faultaddr - u->postcopy_client_bases[i];
1601             if (region_offset < dev->mem->regions[i].memory_size) {
1602                 rb_offset = region_offset + u->region_rb_offset[i];
1603                 trace_vhost_user_postcopy_fault_handler_found(i,
1604                         region_offset, rb_offset);
1605                 rb = u->region_rb[i];
1606                 return postcopy_request_shared_page(pcfd, rb, faultaddr,
1607                                                     rb_offset);
1608             }
1609         }
1610     }
1611     error_report("%s: Failed to find region for fault %" PRIx64,
1612                  __func__, faultaddr);
1613     return -1;
1614 }
1615
1616 static int vhost_user_postcopy_waker(struct PostCopyFD *pcfd, RAMBlock *rb,
1617                                      uint64_t offset)
1618 {
1619     struct vhost_dev *dev = pcfd->data;
1620     struct vhost_user *u = dev->opaque;
1621     int i;
1622
1623     trace_vhost_user_postcopy_waker(qemu_ram_get_idstr(rb), offset);
1624
1625     if (!u) {
1626         return 0;
1627     }
1628     /* Translate the offset into an address in the clients address space */
1629     for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1630         if (u->region_rb[i] == rb &&
1631             offset >= u->region_rb_offset[i] &&
1632             offset < (u->region_rb_offset[i] +
1633                       dev->mem->regions[i].memory_size)) {
1634             uint64_t client_addr = (offset - u->region_rb_offset[i]) +
1635                                    u->postcopy_client_bases[i];
1636             trace_vhost_user_postcopy_waker_found(client_addr);
1637             return postcopy_wake_shared(pcfd, client_addr, rb);
1638         }
1639     }
1640
1641     trace_vhost_user_postcopy_waker_nomatch(qemu_ram_get_idstr(rb), offset);
1642     return 0;
1643 }
1644 #endif
1645
1646 /*
1647  * Called at the start of an inbound postcopy on reception of the
1648  * 'advise' command.
1649  */
1650 static int vhost_user_postcopy_advise(struct vhost_dev *dev, Error **errp)
1651 {
1652 #ifdef CONFIG_LINUX
1653     struct vhost_user *u = dev->opaque;
1654     CharBackend *chr = u->user->chr;
1655     int ufd;
1656     VhostUserMsg msg = {
1657         .hdr.request = VHOST_USER_POSTCOPY_ADVISE,
1658         .hdr.flags = VHOST_USER_VERSION,
1659     };
1660
1661     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1662         error_setg(errp, "Failed to send postcopy_advise to vhost");
1663         return -1;
1664     }
1665
1666     if (vhost_user_read(dev, &msg) < 0) {
1667         error_setg(errp, "Failed to get postcopy_advise reply from vhost");
1668         return -1;
1669     }
1670
1671     if (msg.hdr.request != VHOST_USER_POSTCOPY_ADVISE) {
1672         error_setg(errp, "Unexpected msg type. Expected %d received %d",
1673                      VHOST_USER_POSTCOPY_ADVISE, msg.hdr.request);
1674         return -1;
1675     }
1676
1677     if (msg.hdr.size) {
1678         error_setg(errp, "Received bad msg size.");
1679         return -1;
1680     }
1681     ufd = qemu_chr_fe_get_msgfd(chr);
1682     if (ufd < 0) {
1683         error_setg(errp, "%s: Failed to get ufd", __func__);
1684         return -1;
1685     }
1686     qemu_set_nonblock(ufd);
1687
1688     /* register ufd with userfault thread */
1689     u->postcopy_fd.fd = ufd;
1690     u->postcopy_fd.data = dev;
1691     u->postcopy_fd.handler = vhost_user_postcopy_fault_handler;
1692     u->postcopy_fd.waker = vhost_user_postcopy_waker;
1693     u->postcopy_fd.idstr = "vhost-user"; /* Need to find unique name */
1694     postcopy_register_shared_ufd(&u->postcopy_fd);
1695     return 0;
1696 #else
1697     error_setg(errp, "Postcopy not supported on non-Linux systems");
1698     return -1;
1699 #endif
1700 }
1701
1702 /*
1703  * Called at the switch to postcopy on reception of the 'listen' command.
1704  */
1705 static int vhost_user_postcopy_listen(struct vhost_dev *dev, Error **errp)
1706 {
1707     struct vhost_user *u = dev->opaque;
1708     int ret;
1709     VhostUserMsg msg = {
1710         .hdr.request = VHOST_USER_POSTCOPY_LISTEN,
1711         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
1712     };
1713     u->postcopy_listen = true;
1714     trace_vhost_user_postcopy_listen();
1715     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1716         error_setg(errp, "Failed to send postcopy_listen to vhost");
1717         return -1;
1718     }
1719
1720     ret = process_message_reply(dev, &msg);
1721     if (ret) {
1722         error_setg(errp, "Failed to receive reply to postcopy_listen");
1723         return ret;
1724     }
1725
1726     return 0;
1727 }
1728
1729 /*
1730  * Called at the end of postcopy
1731  */
1732 static int vhost_user_postcopy_end(struct vhost_dev *dev, Error **errp)
1733 {
1734     VhostUserMsg msg = {
1735         .hdr.request = VHOST_USER_POSTCOPY_END,
1736         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
1737     };
1738     int ret;
1739     struct vhost_user *u = dev->opaque;
1740
1741     trace_vhost_user_postcopy_end_entry();
1742     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
1743         error_setg(errp, "Failed to send postcopy_end to vhost");
1744         return -1;
1745     }
1746
1747     ret = process_message_reply(dev, &msg);
1748     if (ret) {
1749         error_setg(errp, "Failed to receive reply to postcopy_end");
1750         return ret;
1751     }
1752     postcopy_unregister_shared_ufd(&u->postcopy_fd);
1753     close(u->postcopy_fd.fd);
1754     u->postcopy_fd.handler = NULL;
1755
1756     trace_vhost_user_postcopy_end_exit();
1757
1758     return 0;
1759 }
1760
1761 static int vhost_user_postcopy_notifier(NotifierWithReturn *notifier,
1762                                         void *opaque)
1763 {
1764     struct PostcopyNotifyData *pnd = opaque;
1765     struct vhost_user *u = container_of(notifier, struct vhost_user,
1766                                          postcopy_notifier);
1767     struct vhost_dev *dev = u->dev;
1768
1769     switch (pnd->reason) {
1770     case POSTCOPY_NOTIFY_PROBE:
1771         if (!virtio_has_feature(dev->protocol_features,
1772                                 VHOST_USER_PROTOCOL_F_PAGEFAULT)) {
1773             /* TODO: Get the device name into this error somehow */
1774             error_setg(pnd->errp,
1775                        "vhost-user backend not capable of postcopy");
1776             return -ENOENT;
1777         }
1778         break;
1779
1780     case POSTCOPY_NOTIFY_INBOUND_ADVISE:
1781         return vhost_user_postcopy_advise(dev, pnd->errp);
1782
1783     case POSTCOPY_NOTIFY_INBOUND_LISTEN:
1784         return vhost_user_postcopy_listen(dev, pnd->errp);
1785
1786     case POSTCOPY_NOTIFY_INBOUND_END:
1787         return vhost_user_postcopy_end(dev, pnd->errp);
1788
1789     default:
1790         /* We ignore notifications we don't know */
1791         break;
1792     }
1793
1794     return 0;
1795 }
1796
1797 static int vhost_user_backend_init(struct vhost_dev *dev, void *opaque)
1798 {
1799     uint64_t features, protocol_features, ram_slots;
1800     struct vhost_user *u;
1801     int err;
1802
1803     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
1804
1805     u = g_new0(struct vhost_user, 1);
1806     u->user = opaque;
1807     u->slave_fd = -1;
1808     u->dev = dev;
1809     dev->opaque = u;
1810
1811     err = vhost_user_get_features(dev, &features);
1812     if (err < 0) {
1813         return err;
1814     }
1815
1816     if (virtio_has_feature(features, VHOST_USER_F_PROTOCOL_FEATURES)) {
1817         dev->backend_features |= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
1818
1819         err = vhost_user_get_u64(dev, VHOST_USER_GET_PROTOCOL_FEATURES,
1820                                  &protocol_features);
1821         if (err < 0) {
1822             return err;
1823         }
1824
1825         dev->protocol_features =
1826             protocol_features & VHOST_USER_PROTOCOL_FEATURE_MASK;
1827
1828         if (!dev->config_ops || !dev->config_ops->vhost_dev_config_notifier) {
1829             /* Don't acknowledge CONFIG feature if device doesn't support it */
1830             dev->protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_CONFIG);
1831         } else if (!(protocol_features &
1832                     (1ULL << VHOST_USER_PROTOCOL_F_CONFIG))) {
1833             error_report("Device expects VHOST_USER_PROTOCOL_F_CONFIG "
1834                     "but backend does not support it.");
1835             return -1;
1836         }
1837
1838         err = vhost_user_set_protocol_features(dev, dev->protocol_features);
1839         if (err < 0) {
1840             return err;
1841         }
1842
1843         /* query the max queues we support if backend supports Multiple Queue */
1844         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_MQ)) {
1845             err = vhost_user_get_u64(dev, VHOST_USER_GET_QUEUE_NUM,
1846                                      &dev->max_queues);
1847             if (err < 0) {
1848                 return err;
1849             }
1850         }
1851
1852         if (virtio_has_feature(features, VIRTIO_F_IOMMU_PLATFORM) &&
1853                 !(virtio_has_feature(dev->protocol_features,
1854                     VHOST_USER_PROTOCOL_F_SLAVE_REQ) &&
1855                  virtio_has_feature(dev->protocol_features,
1856                     VHOST_USER_PROTOCOL_F_REPLY_ACK))) {
1857             error_report("IOMMU support requires reply-ack and "
1858                          "slave-req protocol features.");
1859             return -1;
1860         }
1861
1862         /* get max memory regions if backend supports configurable RAM slots */
1863         if (!virtio_has_feature(dev->protocol_features,
1864                                 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS)) {
1865             u->user->memory_slots = VHOST_MEMORY_BASELINE_NREGIONS;
1866         } else {
1867             err = vhost_user_get_max_memslots(dev, &ram_slots);
1868             if (err < 0) {
1869                 return err;
1870             }
1871
1872             if (ram_slots < u->user->memory_slots) {
1873                 error_report("The backend specified a max ram slots limit "
1874                              "of %" PRIu64", when the prior validated limit was %d. "
1875                              "This limit should never decrease.", ram_slots,
1876                              u->user->memory_slots);
1877                 return -1;
1878             }
1879
1880             u->user->memory_slots = MIN(ram_slots, VHOST_USER_MAX_RAM_SLOTS);
1881         }
1882     }
1883
1884     if (dev->migration_blocker == NULL &&
1885         !virtio_has_feature(dev->protocol_features,
1886                             VHOST_USER_PROTOCOL_F_LOG_SHMFD)) {
1887         error_setg(&dev->migration_blocker,
1888                    "Migration disabled: vhost-user backend lacks "
1889                    "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
1890     }
1891
1892     if (dev->vq_index == 0) {
1893         err = vhost_setup_slave_channel(dev);
1894         if (err < 0) {
1895             return err;
1896         }
1897     }
1898
1899     u->postcopy_notifier.notify = vhost_user_postcopy_notifier;
1900     postcopy_add_notifier(&u->postcopy_notifier);
1901
1902     return 0;
1903 }
1904
1905 static int vhost_user_backend_cleanup(struct vhost_dev *dev)
1906 {
1907     struct vhost_user *u;
1908
1909     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
1910
1911     u = dev->opaque;
1912     if (u->postcopy_notifier.notify) {
1913         postcopy_remove_notifier(&u->postcopy_notifier);
1914         u->postcopy_notifier.notify = NULL;
1915     }
1916     u->postcopy_listen = false;
1917     if (u->postcopy_fd.handler) {
1918         postcopy_unregister_shared_ufd(&u->postcopy_fd);
1919         close(u->postcopy_fd.fd);
1920         u->postcopy_fd.handler = NULL;
1921     }
1922     if (u->slave_fd >= 0) {
1923         qemu_set_fd_handler(u->slave_fd, NULL, NULL, NULL);
1924         close(u->slave_fd);
1925         u->slave_fd = -1;
1926     }
1927     g_free(u->region_rb);
1928     u->region_rb = NULL;
1929     g_free(u->region_rb_offset);
1930     u->region_rb_offset = NULL;
1931     u->region_rb_len = 0;
1932     g_free(u);
1933     dev->opaque = 0;
1934
1935     return 0;
1936 }
1937
1938 static int vhost_user_get_vq_index(struct vhost_dev *dev, int idx)
1939 {
1940     assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
1941
1942     return idx;
1943 }
1944
1945 static int vhost_user_memslots_limit(struct vhost_dev *dev)
1946 {
1947     struct vhost_user *u = dev->opaque;
1948
1949     return u->user->memory_slots;
1950 }
1951
1952 static bool vhost_user_requires_shm_log(struct vhost_dev *dev)
1953 {
1954     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
1955
1956     return virtio_has_feature(dev->protocol_features,
1957                               VHOST_USER_PROTOCOL_F_LOG_SHMFD);
1958 }
1959
1960 static int vhost_user_migration_done(struct vhost_dev *dev, char* mac_addr)
1961 {
1962     VhostUserMsg msg = { };
1963
1964     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
1965
1966     /* If guest supports GUEST_ANNOUNCE do nothing */
1967     if (virtio_has_feature(dev->acked_features, VIRTIO_NET_F_GUEST_ANNOUNCE)) {
1968         return 0;
1969     }
1970
1971     /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
1972     if (virtio_has_feature(dev->protocol_features,
1973                            VHOST_USER_PROTOCOL_F_RARP)) {
1974         msg.hdr.request = VHOST_USER_SEND_RARP;
1975         msg.hdr.flags = VHOST_USER_VERSION;
1976         memcpy((char *)&msg.payload.u64, mac_addr, 6);
1977         msg.hdr.size = sizeof(msg.payload.u64);
1978
1979         return vhost_user_write(dev, &msg, NULL, 0);
1980     }
1981     return -1;
1982 }
1983
1984 static bool vhost_user_can_merge(struct vhost_dev *dev,
1985                                  uint64_t start1, uint64_t size1,
1986                                  uint64_t start2, uint64_t size2)
1987 {
1988     ram_addr_t offset;
1989     int mfd, rfd;
1990
1991     (void)vhost_user_get_mr_data(start1, &offset, &mfd);
1992     (void)vhost_user_get_mr_data(start2, &offset, &rfd);
1993
1994     return mfd == rfd;
1995 }
1996
1997 static int vhost_user_net_set_mtu(struct vhost_dev *dev, uint16_t mtu)
1998 {
1999     VhostUserMsg msg;
2000     bool reply_supported = virtio_has_feature(dev->protocol_features,
2001                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
2002
2003     if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU))) {
2004         return 0;
2005     }
2006
2007     msg.hdr.request = VHOST_USER_NET_SET_MTU;
2008     msg.payload.u64 = mtu;
2009     msg.hdr.size = sizeof(msg.payload.u64);
2010     msg.hdr.flags = VHOST_USER_VERSION;
2011     if (reply_supported) {
2012         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2013     }
2014
2015     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2016         return -1;
2017     }
2018
2019     /* If reply_ack supported, slave has to ack specified MTU is valid */
2020     if (reply_supported) {
2021         return process_message_reply(dev, &msg);
2022     }
2023
2024     return 0;
2025 }
2026
2027 static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
2028                                             struct vhost_iotlb_msg *imsg)
2029 {
2030     VhostUserMsg msg = {
2031         .hdr.request = VHOST_USER_IOTLB_MSG,
2032         .hdr.size = sizeof(msg.payload.iotlb),
2033         .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
2034         .payload.iotlb = *imsg,
2035     };
2036
2037     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2038         return -EFAULT;
2039     }
2040
2041     return process_message_reply(dev, &msg);
2042 }
2043
2044
2045 static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
2046 {
2047     /* No-op as the receive channel is not dedicated to IOTLB messages. */
2048 }
2049
2050 static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
2051                                  uint32_t config_len)
2052 {
2053     VhostUserMsg msg = {
2054         .hdr.request = VHOST_USER_GET_CONFIG,
2055         .hdr.flags = VHOST_USER_VERSION,
2056         .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + config_len,
2057     };
2058
2059     if (!virtio_has_feature(dev->protocol_features,
2060                 VHOST_USER_PROTOCOL_F_CONFIG)) {
2061         return -1;
2062     }
2063
2064     if (config_len > VHOST_USER_MAX_CONFIG_SIZE) {
2065         return -1;
2066     }
2067
2068     msg.payload.config.offset = 0;
2069     msg.payload.config.size = config_len;
2070     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2071         return -1;
2072     }
2073
2074     if (vhost_user_read(dev, &msg) < 0) {
2075         return -1;
2076     }
2077
2078     if (msg.hdr.request != VHOST_USER_GET_CONFIG) {
2079         error_report("Received unexpected msg type. Expected %d received %d",
2080                      VHOST_USER_GET_CONFIG, msg.hdr.request);
2081         return -1;
2082     }
2083
2084     if (msg.hdr.size != VHOST_USER_CONFIG_HDR_SIZE + config_len) {
2085         error_report("Received bad msg size.");
2086         return -1;
2087     }
2088
2089     memcpy(config, msg.payload.config.region, config_len);
2090
2091     return 0;
2092 }
2093
2094 static int vhost_user_set_config(struct vhost_dev *dev, const uint8_t *data,
2095                                  uint32_t offset, uint32_t size, uint32_t flags)
2096 {
2097     uint8_t *p;
2098     bool reply_supported = virtio_has_feature(dev->protocol_features,
2099                                               VHOST_USER_PROTOCOL_F_REPLY_ACK);
2100
2101     VhostUserMsg msg = {
2102         .hdr.request = VHOST_USER_SET_CONFIG,
2103         .hdr.flags = VHOST_USER_VERSION,
2104         .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + size,
2105     };
2106
2107     if (!virtio_has_feature(dev->protocol_features,
2108                 VHOST_USER_PROTOCOL_F_CONFIG)) {
2109         return -1;
2110     }
2111
2112     if (reply_supported) {
2113         msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2114     }
2115
2116     if (size > VHOST_USER_MAX_CONFIG_SIZE) {
2117         return -1;
2118     }
2119
2120     msg.payload.config.offset = offset,
2121     msg.payload.config.size = size,
2122     msg.payload.config.flags = flags,
2123     p = msg.payload.config.region;
2124     memcpy(p, data, size);
2125
2126     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2127         return -1;
2128     }
2129
2130     if (reply_supported) {
2131         return process_message_reply(dev, &msg);
2132     }
2133
2134     return 0;
2135 }
2136
2137 static int vhost_user_crypto_create_session(struct vhost_dev *dev,
2138                                             void *session_info,
2139                                             uint64_t *session_id)
2140 {
2141     bool crypto_session = virtio_has_feature(dev->protocol_features,
2142                                        VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2143     CryptoDevBackendSymSessionInfo *sess_info = session_info;
2144     VhostUserMsg msg = {
2145         .hdr.request = VHOST_USER_CREATE_CRYPTO_SESSION,
2146         .hdr.flags = VHOST_USER_VERSION,
2147         .hdr.size = sizeof(msg.payload.session),
2148     };
2149
2150     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2151
2152     if (!crypto_session) {
2153         error_report("vhost-user trying to send unhandled ioctl");
2154         return -1;
2155     }
2156
2157     memcpy(&msg.payload.session.session_setup_data, sess_info,
2158               sizeof(CryptoDevBackendSymSessionInfo));
2159     if (sess_info->key_len) {
2160         memcpy(&msg.payload.session.key, sess_info->cipher_key,
2161                sess_info->key_len);
2162     }
2163     if (sess_info->auth_key_len > 0) {
2164         memcpy(&msg.payload.session.auth_key, sess_info->auth_key,
2165                sess_info->auth_key_len);
2166     }
2167     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2168         error_report("vhost_user_write() return -1, create session failed");
2169         return -1;
2170     }
2171
2172     if (vhost_user_read(dev, &msg) < 0) {
2173         error_report("vhost_user_read() return -1, create session failed");
2174         return -1;
2175     }
2176
2177     if (msg.hdr.request != VHOST_USER_CREATE_CRYPTO_SESSION) {
2178         error_report("Received unexpected msg type. Expected %d received %d",
2179                      VHOST_USER_CREATE_CRYPTO_SESSION, msg.hdr.request);
2180         return -1;
2181     }
2182
2183     if (msg.hdr.size != sizeof(msg.payload.session)) {
2184         error_report("Received bad msg size.");
2185         return -1;
2186     }
2187
2188     if (msg.payload.session.session_id < 0) {
2189         error_report("Bad session id: %" PRId64 "",
2190                               msg.payload.session.session_id);
2191         return -1;
2192     }
2193     *session_id = msg.payload.session.session_id;
2194
2195     return 0;
2196 }
2197
2198 static int
2199 vhost_user_crypto_close_session(struct vhost_dev *dev, uint64_t session_id)
2200 {
2201     bool crypto_session = virtio_has_feature(dev->protocol_features,
2202                                        VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2203     VhostUserMsg msg = {
2204         .hdr.request = VHOST_USER_CLOSE_CRYPTO_SESSION,
2205         .hdr.flags = VHOST_USER_VERSION,
2206         .hdr.size = sizeof(msg.payload.u64),
2207     };
2208     msg.payload.u64 = session_id;
2209
2210     if (!crypto_session) {
2211         error_report("vhost-user trying to send unhandled ioctl");
2212         return -1;
2213     }
2214
2215     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2216         error_report("vhost_user_write() return -1, close session failed");
2217         return -1;
2218     }
2219
2220     return 0;
2221 }
2222
2223 static bool vhost_user_mem_section_filter(struct vhost_dev *dev,
2224                                           MemoryRegionSection *section)
2225 {
2226     bool result;
2227
2228     result = memory_region_get_fd(section->mr) >= 0;
2229
2230     return result;
2231 }
2232
2233 static int vhost_user_get_inflight_fd(struct vhost_dev *dev,
2234                                       uint16_t queue_size,
2235                                       struct vhost_inflight *inflight)
2236 {
2237     void *addr;
2238     int fd;
2239     struct vhost_user *u = dev->opaque;
2240     CharBackend *chr = u->user->chr;
2241     VhostUserMsg msg = {
2242         .hdr.request = VHOST_USER_GET_INFLIGHT_FD,
2243         .hdr.flags = VHOST_USER_VERSION,
2244         .payload.inflight.num_queues = dev->nvqs,
2245         .payload.inflight.queue_size = queue_size,
2246         .hdr.size = sizeof(msg.payload.inflight),
2247     };
2248
2249     if (!virtio_has_feature(dev->protocol_features,
2250                             VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2251         return 0;
2252     }
2253
2254     if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
2255         return -1;
2256     }
2257
2258     if (vhost_user_read(dev, &msg) < 0) {
2259         return -1;
2260     }
2261
2262     if (msg.hdr.request != VHOST_USER_GET_INFLIGHT_FD) {
2263         error_report("Received unexpected msg type. "
2264                      "Expected %d received %d",
2265                      VHOST_USER_GET_INFLIGHT_FD, msg.hdr.request);
2266         return -1;
2267     }
2268
2269     if (msg.hdr.size != sizeof(msg.payload.inflight)) {
2270         error_report("Received bad msg size.");
2271         return -1;
2272     }
2273
2274     if (!msg.payload.inflight.mmap_size) {
2275         return 0;
2276     }
2277
2278     fd = qemu_chr_fe_get_msgfd(chr);
2279     if (fd < 0) {
2280         error_report("Failed to get mem fd");
2281         return -1;
2282     }
2283
2284     addr = mmap(0, msg.payload.inflight.mmap_size, PROT_READ | PROT_WRITE,
2285                 MAP_SHARED, fd, msg.payload.inflight.mmap_offset);
2286
2287     if (addr == MAP_FAILED) {
2288         error_report("Failed to mmap mem fd");
2289         close(fd);
2290         return -1;
2291     }
2292
2293     inflight->addr = addr;
2294     inflight->fd = fd;
2295     inflight->size = msg.payload.inflight.mmap_size;
2296     inflight->offset = msg.payload.inflight.mmap_offset;
2297     inflight->queue_size = queue_size;
2298
2299     return 0;
2300 }
2301
2302 static int vhost_user_set_inflight_fd(struct vhost_dev *dev,
2303                                       struct vhost_inflight *inflight)
2304 {
2305     VhostUserMsg msg = {
2306         .hdr.request = VHOST_USER_SET_INFLIGHT_FD,
2307         .hdr.flags = VHOST_USER_VERSION,
2308         .payload.inflight.mmap_size = inflight->size,
2309         .payload.inflight.mmap_offset = inflight->offset,
2310         .payload.inflight.num_queues = dev->nvqs,
2311         .payload.inflight.queue_size = inflight->queue_size,
2312         .hdr.size = sizeof(msg.payload.inflight),
2313     };
2314
2315     if (!virtio_has_feature(dev->protocol_features,
2316                             VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2317         return 0;
2318     }
2319
2320     if (vhost_user_write(dev, &msg, &inflight->fd, 1) < 0) {
2321         return -1;
2322     }
2323
2324     return 0;
2325 }
2326
2327 bool vhost_user_init(VhostUserState *user, CharBackend *chr, Error **errp)
2328 {
2329     if (user->chr) {
2330         error_setg(errp, "Cannot initialize vhost-user state");
2331         return false;
2332     }
2333     user->chr = chr;
2334     user->memory_slots = 0;
2335     return true;
2336 }
2337
2338 void vhost_user_cleanup(VhostUserState *user)
2339 {
2340     int i;
2341
2342     if (!user->chr) {
2343         return;
2344     }
2345
2346     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2347         if (user->notifier[i].addr) {
2348             object_unparent(OBJECT(&user->notifier[i].mr));
2349             munmap(user->notifier[i].addr, qemu_real_host_page_size);
2350             user->notifier[i].addr = NULL;
2351         }
2352     }
2353     user->chr = NULL;
2354 }
2355
2356 const VhostOps user_ops = {
2357         .backend_type = VHOST_BACKEND_TYPE_USER,
2358         .vhost_backend_init = vhost_user_backend_init,
2359         .vhost_backend_cleanup = vhost_user_backend_cleanup,
2360         .vhost_backend_memslots_limit = vhost_user_memslots_limit,
2361         .vhost_set_log_base = vhost_user_set_log_base,
2362         .vhost_set_mem_table = vhost_user_set_mem_table,
2363         .vhost_set_vring_addr = vhost_user_set_vring_addr,
2364         .vhost_set_vring_endian = vhost_user_set_vring_endian,
2365         .vhost_set_vring_num = vhost_user_set_vring_num,
2366         .vhost_set_vring_base = vhost_user_set_vring_base,
2367         .vhost_get_vring_base = vhost_user_get_vring_base,
2368         .vhost_set_vring_kick = vhost_user_set_vring_kick,
2369         .vhost_set_vring_call = vhost_user_set_vring_call,
2370         .vhost_set_features = vhost_user_set_features,
2371         .vhost_get_features = vhost_user_get_features,
2372         .vhost_set_owner = vhost_user_set_owner,
2373         .vhost_reset_device = vhost_user_reset_device,
2374         .vhost_get_vq_index = vhost_user_get_vq_index,
2375         .vhost_set_vring_enable = vhost_user_set_vring_enable,
2376         .vhost_requires_shm_log = vhost_user_requires_shm_log,
2377         .vhost_migration_done = vhost_user_migration_done,
2378         .vhost_backend_can_merge = vhost_user_can_merge,
2379         .vhost_net_set_mtu = vhost_user_net_set_mtu,
2380         .vhost_set_iotlb_callback = vhost_user_set_iotlb_callback,
2381         .vhost_send_device_iotlb_msg = vhost_user_send_device_iotlb_msg,
2382         .vhost_get_config = vhost_user_get_config,
2383         .vhost_set_config = vhost_user_set_config,
2384         .vhost_crypto_create_session = vhost_user_crypto_create_session,
2385         .vhost_crypto_close_session = vhost_user_crypto_close_session,
2386         .vhost_backend_mem_section_filter = vhost_user_mem_section_filter,
2387         .vhost_get_inflight_fd = vhost_user_get_inflight_fd,
2388         .vhost_set_inflight_fd = vhost_user_set_inflight_fd,
2389 };