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 "hw/virtio/vhost.h"
  12 #include "hw/virtio/vhost-backend.h"
  13 #include "sysemu/char.h"
  14 #include "sysemu/kvm.h"
  15 #include "qemu/error-report.h"
  16 #include "qemu/sockets.h"
  17 #include "exec/ram_addr.h"
  18
  19 #include <fcntl.h>
  20 #include <unistd.h>
  21 #include <sys/ioctl.h>
  22 #include <sys/socket.h>
  23 #include <sys/un.h>
  24 #include <linux/vhost.h>
  25
  26 #define VHOST_MEMORY_MAX_NREGIONS    8
  27
  28 typedef enum VhostUserRequest {
  29     VHOST_USER_NONE = 0,
  30     VHOST_USER_GET_FEATURES = 1,
  31     VHOST_USER_SET_FEATURES = 2,
  32     VHOST_USER_SET_OWNER = 3,
  33     VHOST_USER_RESET_OWNER = 4,
  34     VHOST_USER_SET_MEM_TABLE = 5,
  35     VHOST_USER_SET_LOG_BASE = 6,
  36     VHOST_USER_SET_LOG_FD = 7,
  37     VHOST_USER_SET_VRING_NUM = 8,
  38     VHOST_USER_SET_VRING_ADDR = 9,
  39     VHOST_USER_SET_VRING_BASE = 10,
  40     VHOST_USER_GET_VRING_BASE = 11,
  41     VHOST_USER_SET_VRING_KICK = 12,
  42     VHOST_USER_SET_VRING_CALL = 13,
  43     VHOST_USER_SET_VRING_ERR = 14,
  44     VHOST_USER_MAX
  45 } VhostUserRequest;
  46
  47 typedef struct VhostUserMemoryRegion {
  48     uint64_t guest_phys_addr;
  49     uint64_t memory_size;
  50     uint64_t userspace_addr;
  51     uint64_t mmap_offset;
  52 } VhostUserMemoryRegion;
  53
  54 typedef struct VhostUserMemory {
  55     uint32_t nregions;
  56     uint32_t padding;
  57     VhostUserMemoryRegion regions[VHOST_MEMORY_MAX_NREGIONS];
  58 } VhostUserMemory;
  59
  60 typedef struct VhostUserMsg {
  61     VhostUserRequest request;
  62
  63 #define VHOST_USER_VERSION_MASK     (0x3)
  64 #define VHOST_USER_REPLY_MASK       (0x1<<2)
  65     uint32_t flags;
  66     uint32_t size; /* the following payload size */
  67     union {
  68 #define VHOST_USER_VRING_IDX_MASK   (0xff)
  69 #define VHOST_USER_VRING_NOFD_MASK  (0x1<<8)
  70         uint64_t u64;
  71         struct vhost_vring_state state;
  72         struct vhost_vring_addr addr;
  73         VhostUserMemory memory;
  74     };
  75 } QEMU_PACKED VhostUserMsg;
  76
  77 static VhostUserMsg m __attribute__ ((unused));
  78 #define VHOST_USER_HDR_SIZE (sizeof(m.request) \
  79                             + sizeof(m.flags) \
  80                             + sizeof(m.size))
  81
  82 #define VHOST_USER_PAYLOAD_SIZE (sizeof(m) - VHOST_USER_HDR_SIZE)
  83
  84 /* The version of the protocol we support */
  85 #define VHOST_USER_VERSION    (0x1)
  86
  87 static bool ioeventfd_enabled(void)
  88 {
  89     return kvm_enabled() && kvm_eventfds_enabled();
  90 }
  91
  92 static unsigned long int ioctl_to_vhost_user_request[VHOST_USER_MAX] = {
  93     -1,                     /* VHOST_USER_NONE */
  94     VHOST_GET_FEATURES,     /* VHOST_USER_GET_FEATURES */
  95     VHOST_SET_FEATURES,     /* VHOST_USER_SET_FEATURES */
  96     VHOST_SET_OWNER,        /* VHOST_USER_SET_OWNER */
  97     VHOST_RESET_OWNER,      /* VHOST_USER_RESET_OWNER */
  98     VHOST_SET_MEM_TABLE,    /* VHOST_USER_SET_MEM_TABLE */
  99     VHOST_SET_LOG_BASE,     /* VHOST_USER_SET_LOG_BASE */
 100     VHOST_SET_LOG_FD,       /* VHOST_USER_SET_LOG_FD */
 101     VHOST_SET_VRING_NUM,    /* VHOST_USER_SET_VRING_NUM */
 102     VHOST_SET_VRING_ADDR,   /* VHOST_USER_SET_VRING_ADDR */
 103     VHOST_SET_VRING_BASE,   /* VHOST_USER_SET_VRING_BASE */
 104     VHOST_GET_VRING_BASE,   /* VHOST_USER_GET_VRING_BASE */
 105     VHOST_SET_VRING_KICK,   /* VHOST_USER_SET_VRING_KICK */
 106     VHOST_SET_VRING_CALL,   /* VHOST_USER_SET_VRING_CALL */
 107     VHOST_SET_VRING_ERR     /* VHOST_USER_SET_VRING_ERR */
 108 };
 109
 110 static VhostUserRequest vhost_user_request_translate(unsigned long int request)
 111 {
 112     VhostUserRequest idx;
 113
 114     for (idx = 0; idx < VHOST_USER_MAX; idx++) {
 115         if (ioctl_to_vhost_user_request[idx] == request) {
 116             break;
 117         }
 118     }
 119
 120     return (idx == VHOST_USER_MAX) ? VHOST_USER_NONE : idx;
 121 }
 122
 123 static int vhost_user_read(struct vhost_dev *dev, VhostUserMsg *msg)
 124 {
 125     CharDriverState *chr = dev->opaque;
 126     uint8_t *p = (uint8_t *) msg;
 127     int r, size = VHOST_USER_HDR_SIZE;
 128
 129     r = qemu_chr_fe_read_all(chr, p, size);
 130     if (r != size) {
 131         error_report("Failed to read msg header. Read %d instead of %d.", r,
 132                 size);
 133         goto fail;
 134     }
 135
 136     /* validate received flags */
 137     if (msg->flags != (VHOST_USER_REPLY_MASK | VHOST_USER_VERSION)) {
 138         error_report("Failed to read msg header."
 139                 " Flags 0x%x instead of 0x%x.", msg->flags,
 140                 VHOST_USER_REPLY_MASK | VHOST_USER_VERSION);
 141         goto fail;
 142     }
 143
 144     /* validate message size is sane */
 145     if (msg->size > VHOST_USER_PAYLOAD_SIZE) {
 146         error_report("Failed to read msg header."
 147                 " Size %d exceeds the maximum %zu.", msg->size,
 148                 VHOST_USER_PAYLOAD_SIZE);
 149         goto fail;
 150     }
 151
 152     if (msg->size) {
 153         p += VHOST_USER_HDR_SIZE;
 154         size = msg->size;
 155         r = qemu_chr_fe_read_all(chr, p, size);
 156         if (r != size) {
 157             error_report("Failed to read msg payload."
 158                          " Read %d instead of %d.", r, msg->size);
 159             goto fail;
 160         }
 161     }
 162
 163     return 0;
 164
 165 fail:
 166     return -1;
 167 }
 168
 169 static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
 170                             int *fds, int fd_num)
 171 {
 172     CharDriverState *chr = dev->opaque;
 173     int size = VHOST_USER_HDR_SIZE + msg->size;
 174
 175     if (fd_num) {
 176         qemu_chr_fe_set_msgfds(chr, fds, fd_num);
 177     }
 178
 179     return qemu_chr_fe_write_all(chr, (const uint8_t *) msg, size) == size ?
 180             0 : -1;
 181 }
 182
 183 static int vhost_user_call(struct vhost_dev *dev, unsigned long int request,
 184         void *arg)
 185 {
 186     VhostUserMsg msg;
 187     VhostUserRequest msg_request;
 188     struct vhost_vring_file *file = 0;
 189     int need_reply = 0;
 190     int fds[VHOST_MEMORY_MAX_NREGIONS];
 191     int i, fd;
 192     size_t fd_num = 0;
 193
 194     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 195
 196     msg_request = vhost_user_request_translate(request);
 197     msg.request = msg_request;
 198     msg.flags = VHOST_USER_VERSION;
 199     msg.size = 0;
 200
 201     switch (request) {
 202     case VHOST_GET_FEATURES:
 203         need_reply = 1;
 204         break;
 205
 206     case VHOST_SET_FEATURES:
 207     case VHOST_SET_LOG_BASE:
 208         msg.u64 = *((__u64 *) arg);
 209         msg.size = sizeof(m.u64);
 210         break;
 211
 212     case VHOST_SET_OWNER:
 213     case VHOST_RESET_OWNER:
 214         break;
 215
 216     case VHOST_SET_MEM_TABLE:
 217         for (i = 0; i < dev->mem->nregions; ++i) {
 218             struct vhost_memory_region *reg = dev->mem->regions + i;
 219             ram_addr_t ram_addr;
 220
 221             assert((uintptr_t)reg->userspace_addr == reg->userspace_addr);
 222             qemu_ram_addr_from_host((void *)(uintptr_t)reg->userspace_addr, &ram_addr);
 223             fd = qemu_get_ram_fd(ram_addr);
 224             if (fd > 0) {
 225                 msg.memory.regions[fd_num].userspace_addr = reg->userspace_addr;
 226                 msg.memory.regions[fd_num].memory_size  = reg->memory_size;
 227                 msg.memory.regions[fd_num].guest_phys_addr = reg->guest_phys_addr;
 228                 msg.memory.regions[fd_num].mmap_offset = reg->userspace_addr -
 229                     (uintptr_t) qemu_get_ram_block_host_ptr(ram_addr);
 230                 assert(fd_num < VHOST_MEMORY_MAX_NREGIONS);
 231                 fds[fd_num++] = fd;
 232             }
 233         }
 234
 235         msg.memory.nregions = fd_num;
 236
 237         if (!fd_num) {
 238             error_report("Failed initializing vhost-user memory map, "
 239                     "consider using -object memory-backend-file share=on");
 240             return -1;
 241         }
 242
 243         msg.size = sizeof(m.memory.nregions);
 244         msg.size += sizeof(m.memory.padding);
 245         msg.size += fd_num * sizeof(VhostUserMemoryRegion);
 246
 247         break;
 248
 249     case VHOST_SET_LOG_FD:
 250         fds[fd_num++] = *((int *) arg);
 251         break;
 252
 253     case VHOST_SET_VRING_NUM:
 254     case VHOST_SET_VRING_BASE:
 255         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
 256         msg.size = sizeof(m.state);
 257         break;
 258
 259     case VHOST_GET_VRING_BASE:
 260         memcpy(&msg.state, arg, sizeof(struct vhost_vring_state));
 261         msg.size = sizeof(m.state);
 262         need_reply = 1;
 263         break;
 264
 265     case VHOST_SET_VRING_ADDR:
 266         memcpy(&msg.addr, arg, sizeof(struct vhost_vring_addr));
 267         msg.size = sizeof(m.addr);
 268         break;
 269
 270     case VHOST_SET_VRING_KICK:
 271     case VHOST_SET_VRING_CALL:
 272     case VHOST_SET_VRING_ERR:
 273         file = arg;
 274         msg.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
 275         msg.size = sizeof(m.u64);
 276         if (ioeventfd_enabled() && file->fd > 0) {
 277             fds[fd_num++] = file->fd;
 278         } else {
 279             msg.u64 |= VHOST_USER_VRING_NOFD_MASK;
 280         }
 281         break;
 282     default:
 283         error_report("vhost-user trying to send unhandled ioctl");
 284         return -1;
 285         break;
 286     }
 287
 288     if (vhost_user_write(dev, &msg, fds, fd_num) < 0) {
 289         return 0;
 290     }
 291
 292     if (need_reply) {
 293         if (vhost_user_read(dev, &msg) < 0) {
 294             return 0;
 295         }
 296
 297         if (msg_request != msg.request) {
 298             error_report("Received unexpected msg type."
 299                     " Expected %d received %d", msg_request, msg.request);
 300             return -1;
 301         }
 302
 303         switch (msg_request) {
 304         case VHOST_USER_GET_FEATURES:
 305             if (msg.size != sizeof(m.u64)) {
 306                 error_report("Received bad msg size.");
 307                 return -1;
 308             }
 309             *((__u64 *) arg) = msg.u64;
 310             break;
 311         case VHOST_USER_GET_VRING_BASE:
 312             if (msg.size != sizeof(m.state)) {
 313                 error_report("Received bad msg size.");
 314                 return -1;
 315             }
 316             memcpy(arg, &msg.state, sizeof(struct vhost_vring_state));
 317             break;
 318         default:
 319             error_report("Received unexpected msg type.");
 320             return -1;
 321             break;
 322         }
 323     }
 324
 325     return 0;
 326 }
 327
 328 static int vhost_user_init(struct vhost_dev *dev, void *opaque)
 329 {
 330     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 331
 332     dev->opaque = opaque;
 333
 334     return 0;
 335 }
 336
 337 static int vhost_user_cleanup(struct vhost_dev *dev)
 338 {
 339     assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 340
 341     dev->opaque = 0;
 342
 343     return 0;
 344 }
 345
 346 const VhostOps user_ops = {
 347         .backend_type = VHOST_BACKEND_TYPE_USER,
 348         .vhost_call = vhost_user_call,
 349         .vhost_backend_init = vhost_user_init,
 350         .vhost_backend_cleanup = vhost_user_cleanup
 351         };