lib/librte_vhost/vhost_user/virtio-net-user.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdint.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <unistd.h>
  38 #include <sys/mman.h>
  39 #include <sys/types.h>
  40 #include <sys/stat.h>
  41 #include <unistd.h>
  42
  43 #include <rte_common.h>
  44 #include <rte_log.h>
  45
  46 #include "virtio-net.h"
  47 #include "virtio-net-user.h"
  48 #include "vhost-net-user.h"
  49 #include "vhost-net.h"
  50
  51 struct orig_region_map {
  52         int fd;
  53         uint64_t mapped_address;
  54         uint64_t mapped_size;
  55         uint64_t blksz;
  56 };
  57
  58 #define orig_region(ptr, nregions) \
  59         ((struct orig_region_map *)RTE_PTR_ADD((ptr), \
  60                 sizeof(struct virtio_memory) + \
  61                 sizeof(struct virtio_memory_regions) * (nregions)))
  62
  63 static uint64_t
  64 get_blk_size(int fd)
  65 {
  66         struct stat stat;
  67
  68         fstat(fd, &stat);
  69         return (uint64_t)stat.st_blksize;
  70 }
  71
  72 static void
  73 free_mem_region(struct virtio_net *dev)
  74 {
  75         struct orig_region_map *region;
  76         unsigned int idx;
  77         uint64_t alignment;
  78
  79         if (!dev || !dev->mem)
  80                 return;
  81
  82         region = orig_region(dev->mem, dev->mem->nregions);
  83         for (idx = 0; idx < dev->mem->nregions; idx++) {
  84                 if (region[idx].mapped_address) {
  85                         alignment = region[idx].blksz;
  86                         munmap((void *)(uintptr_t)
  87                                 RTE_ALIGN_FLOOR(
  88                                         region[idx].mapped_address, alignment),
  89                                 RTE_ALIGN_CEIL(
  90                                         region[idx].mapped_size, alignment));
  91                         close(region[idx].fd);
  92                 }
  93         }
  94 }
  95
  96 int
  97 user_set_mem_table(struct vhost_device_ctx ctx, struct VhostUserMsg *pmsg)
  98 {
  99         struct VhostUserMemory memory = pmsg->payload.memory;
 100         struct virtio_memory_regions *pregion;
 101         uint64_t mapped_address, mapped_size;
 102         struct virtio_net *dev;
 103         unsigned int idx = 0;
 104         struct orig_region_map *pregion_orig;
 105         uint64_t alignment;
 106
 107         /* unmap old memory regions one by one*/
 108         dev = get_device(ctx);
 109         if (dev == NULL)
 110                 return -1;
 111
 112         /* Remove from the data plane. */
 113         if (dev->flags & VIRTIO_DEV_RUNNING)
 114                 notify_ops->destroy_device(dev);
 115
 116         if (dev->mem) {
 117                 free_mem_region(dev);
 118                 free(dev->mem);
 119                 dev->mem = NULL;
 120         }
 121
 122         dev->mem = calloc(1,
 123                 sizeof(struct virtio_memory) +
 124                 sizeof(struct virtio_memory_regions) * memory.nregions +
 125                 sizeof(struct orig_region_map) * memory.nregions);
 126         if (dev->mem == NULL) {
 127                 RTE_LOG(ERR, VHOST_CONFIG,
 128                         "(%"PRIu64") Failed to allocate memory for dev->mem\n",
 129                         dev->device_fh);
 130                 return -1;
 131         }
 132         dev->mem->nregions = memory.nregions;
 133
 134         pregion_orig = orig_region(dev->mem, memory.nregions);
 135         for (idx = 0; idx < memory.nregions; idx++) {
 136                 pregion = &dev->mem->regions[idx];
 137                 pregion->guest_phys_address =
 138                         memory.regions[idx].guest_phys_addr;
 139                 pregion->guest_phys_address_end =
 140                         memory.regions[idx].guest_phys_addr +
 141                         memory.regions[idx].memory_size;
 142                 pregion->memory_size =
 143                         memory.regions[idx].memory_size;
 144                 pregion->userspace_address =
 145                         memory.regions[idx].userspace_addr;
 146
 147                 /* This is ugly */
 148                 mapped_size = memory.regions[idx].memory_size +
 149                         memory.regions[idx].mmap_offset;
 150                 mapped_address = (uint64_t)(uintptr_t)mmap(NULL,
 151                         mapped_size,
 152                         PROT_READ | PROT_WRITE, MAP_SHARED,
 153                         pmsg->fds[idx],
 154                         0);
 155
 156                 RTE_LOG(INFO, VHOST_CONFIG,
 157                         "mapped region %d fd:%d to %p sz:0x%"PRIx64" off:0x%"PRIx64"\n",
 158                         idx, pmsg->fds[idx], (void *)(uintptr_t)mapped_address,
 159                         mapped_size, memory.regions[idx].mmap_offset);
 160
 161                 if (mapped_address == (uint64_t)(uintptr_t)MAP_FAILED) {
 162                         RTE_LOG(ERR, VHOST_CONFIG,
 163                                 "mmap qemu guest failed.\n");
 164                         goto err_mmap;
 165                 }
 166
 167                 pregion_orig[idx].mapped_address = mapped_address;
 168                 pregion_orig[idx].mapped_size = mapped_size;
 169                 pregion_orig[idx].blksz = get_blk_size(pmsg->fds[idx]);
 170                 pregion_orig[idx].fd = pmsg->fds[idx];
 171
 172                 mapped_address +=  memory.regions[idx].mmap_offset;
 173
 174                 pregion->address_offset = mapped_address -
 175                         pregion->guest_phys_address;
 176
 177                 if (memory.regions[idx].guest_phys_addr == 0) {
 178                         dev->mem->base_address =
 179                                 memory.regions[idx].userspace_addr;
 180                         dev->mem->mapped_address =
 181                                 pregion->address_offset;
 182                 }
 183
 184                 LOG_DEBUG(VHOST_CONFIG,
 185                         "REGION: %u GPA: %p QEMU VA: %p SIZE (%"PRIu64")\n",
 186                         idx,
 187                         (void *)(uintptr_t)pregion->guest_phys_address,
 188                         (void *)(uintptr_t)pregion->userspace_address,
 189                          pregion->memory_size);
 190         }
 191
 192         return 0;
 193
 194 err_mmap:
 195         while (idx--) {
 196                 alignment = pregion_orig[idx].blksz;
 197                 munmap((void *)(uintptr_t)RTE_ALIGN_FLOOR(
 198                         pregion_orig[idx].mapped_address, alignment),
 199                         RTE_ALIGN_CEIL(pregion_orig[idx].mapped_size,
 200                                         alignment));
 201                 close(pregion_orig[idx].fd);
 202         }
 203         free(dev->mem);
 204         dev->mem = NULL;
 205         return -1;
 206 }
 207
 208 static int
 209 vq_is_ready(struct vhost_virtqueue *vq)
 210 {
 211         return vq && vq->desc   &&
 212                vq->kickfd != -1 &&
 213                vq->callfd != -1;
 214 }
 215
 216 static int
 217 virtio_is_ready(struct virtio_net *dev)
 218 {
 219         struct vhost_virtqueue *rvq, *tvq;
 220         uint32_t i;
 221
 222         for (i = 0; i < dev->virt_qp_nb; i++) {
 223                 rvq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_RXQ];
 224                 tvq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_TXQ];
 225
 226                 if (!vq_is_ready(rvq) || !vq_is_ready(tvq)) {
 227                         RTE_LOG(INFO, VHOST_CONFIG,
 228                                 "virtio is not ready for processing.\n");
 229                         return 0;
 230                 }
 231         }
 232
 233         RTE_LOG(INFO, VHOST_CONFIG,
 234                 "virtio is now ready for processing.\n");
 235         return 1;
 236 }
 237
 238 void
 239 user_set_vring_call(struct vhost_device_ctx ctx, struct VhostUserMsg *pmsg)
 240 {
 241         struct vhost_vring_file file;
 242
 243         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 244         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 245                 file.fd = -1;
 246         else
 247                 file.fd = pmsg->fds[0];
 248         RTE_LOG(INFO, VHOST_CONFIG,
 249                 "vring call idx:%d file:%d\n", file.index, file.fd);
 250         ops->set_vring_call(ctx, &file);
 251 }
 252
 253
 254 /*
 255  *  In vhost-user, when we receive kick message, will test whether virtio
 256  *  device is ready for packet processing.
 257  */
 258 void
 259 user_set_vring_kick(struct vhost_device_ctx ctx, struct VhostUserMsg *pmsg)
 260 {
 261         struct vhost_vring_file file;
 262         struct virtio_net *dev = get_device(ctx);
 263
 264         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 265         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 266                 file.fd = -1;
 267         else
 268                 file.fd = pmsg->fds[0];
 269         RTE_LOG(INFO, VHOST_CONFIG,
 270                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 271         ops->set_vring_kick(ctx, &file);
 272
 273         if (virtio_is_ready(dev) &&
 274                 !(dev->flags & VIRTIO_DEV_RUNNING))
 275                         notify_ops->new_device(dev);
 276 }
 277
 278 /*
 279  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 280  */
 281 int
 282 user_get_vring_base(struct vhost_device_ctx ctx,
 283         struct vhost_vring_state *state)
 284 {
 285         struct virtio_net *dev = get_device(ctx);
 286
 287         if (dev == NULL)
 288                 return -1;
 289         /* We have to stop the queue (virtio) if it is running. */
 290         if (dev->flags & VIRTIO_DEV_RUNNING)
 291                 notify_ops->destroy_device(dev);
 292
 293         /* Here we are safe to get the last used index */
 294         ops->get_vring_base(ctx, state->index, state);
 295
 296         RTE_LOG(INFO, VHOST_CONFIG,
 297                 "vring base idx:%d file:%d\n", state->index, state->num);
 298         /*
 299          * Based on current qemu vhost-user implementation, this message is
 300          * sent and only sent in vhost_vring_stop.
 301          * TODO: cleanup the vring, it isn't usable since here.
 302          */
 303         if (dev->virtqueue[state->index]->kickfd >= 0) {
 304                 close(dev->virtqueue[state->index]->kickfd);
 305                 dev->virtqueue[state->index]->kickfd = -1;
 306         }
 307
 308         return 0;
 309 }
 310
 311 /*
 312  * when virtio queues are ready to work, qemu will send us to
 313  * enable the virtio queue pair.
 314  */
 315 int
 316 user_set_vring_enable(struct vhost_device_ctx ctx,
 317                       struct vhost_vring_state *state)
 318 {
 319         struct virtio_net *dev = get_device(ctx);
 320         uint16_t base_idx = state->index;
 321         int enable = (int)state->num;
 322
 323         RTE_LOG(INFO, VHOST_CONFIG,
 324                 "set queue enable: %d to qp idx: %d\n",
 325                 enable, state->index);
 326
 327         if (notify_ops->vring_state_changed) {
 328                 notify_ops->vring_state_changed(dev, base_idx / VIRTIO_QNUM,
 329                                                 enable);
 330         }
 331
 332         dev->virtqueue[base_idx + VIRTIO_RXQ]->enabled = enable;
 333         dev->virtqueue[base_idx + VIRTIO_TXQ]->enabled = enable;
 334
 335         return 0;
 336 }
 337
 338 void
 339 user_destroy_device(struct vhost_device_ctx ctx)
 340 {
 341         struct virtio_net *dev = get_device(ctx);
 342
 343         if (dev && (dev->flags & VIRTIO_DEV_RUNNING))
 344                 notify_ops->destroy_device(dev);
 345
 346         if (dev && dev->mem) {
 347                 free_mem_region(dev);
 348                 free(dev->mem);
 349                 dev->mem = NULL;
 350         }
 351 }
 352
 353 void
 354 user_set_protocol_features(struct vhost_device_ctx ctx,
 355                            uint64_t protocol_features)
 356 {
 357         struct virtio_net *dev;
 358
 359         dev = get_device(ctx);
 360         if (dev == NULL || protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 361                 return;
 362
 363         dev->protocol_features = protocol_features;
 364 }