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 *)
  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         if (dev->mem) {
 113                 free_mem_region(dev);
 114                 free(dev->mem);
 115                 dev->mem = NULL;
 116         }
 117
 118         dev->mem = calloc(1,
 119                 sizeof(struct virtio_memory) +
 120                 sizeof(struct virtio_memory_regions) * memory.nregions +
 121                 sizeof(struct orig_region_map) * memory.nregions);
 122         if (dev->mem == NULL) {
 123                 RTE_LOG(ERR, VHOST_CONFIG,
 124                         "(%"PRIu64") Failed to allocate memory for dev->mem\n",
 125                         dev->device_fh);
 126                 return -1;
 127         }
 128         dev->mem->nregions = memory.nregions;
 129
 130         pregion_orig = orig_region(dev->mem, memory.nregions);
 131         for (idx = 0; idx < memory.nregions; idx++) {
 132                 pregion = &dev->mem->regions[idx];
 133                 pregion->guest_phys_address =
 134                         memory.regions[idx].guest_phys_addr;
 135                 pregion->guest_phys_address_end =
 136                         memory.regions[idx].guest_phys_addr +
 137                         memory.regions[idx].memory_size;
 138                 pregion->memory_size =
 139                         memory.regions[idx].memory_size;
 140                 pregion->userspace_address =
 141                         memory.regions[idx].userspace_addr;
 142
 143                 /* This is ugly */
 144                 mapped_size = memory.regions[idx].memory_size +
 145                         memory.regions[idx].mmap_offset;
 146                 mapped_address = (uint64_t)(uintptr_t)mmap(NULL,
 147                         mapped_size,
 148                         PROT_READ | PROT_WRITE, MAP_SHARED,
 149                         pmsg->fds[idx],
 150                         0);
 151
 152                 RTE_LOG(INFO, VHOST_CONFIG,
 153                         "mapped region %d fd:%d to %p sz:0x%"PRIx64" off:0x%"PRIx64"\n",
 154                         idx, pmsg->fds[idx], (void *)mapped_address,
 155                         mapped_size, memory.regions[idx].mmap_offset);
 156
 157                 if (mapped_address == (uint64_t)(uintptr_t)MAP_FAILED) {
 158                         RTE_LOG(ERR, VHOST_CONFIG,
 159                                 "mmap qemu guest failed.\n");
 160                         goto err_mmap;
 161                 }
 162
 163                 pregion_orig[idx].mapped_address = mapped_address;
 164                 pregion_orig[idx].mapped_size = mapped_size;
 165                 pregion_orig[idx].blksz = get_blk_size(pmsg->fds[idx]);
 166                 pregion_orig[idx].fd = pmsg->fds[idx];
 167
 168                 mapped_address +=  memory.regions[idx].mmap_offset;
 169
 170                 pregion->address_offset = mapped_address -
 171                         pregion->guest_phys_address;
 172
 173                 if (memory.regions[idx].guest_phys_addr == 0) {
 174                         dev->mem->base_address =
 175                                 memory.regions[idx].userspace_addr;
 176                         dev->mem->mapped_address =
 177                                 pregion->address_offset;
 178                 }
 179
 180                 LOG_DEBUG(VHOST_CONFIG,
 181                         "REGION: %u GPA: %p QEMU VA: %p SIZE (%"PRIu64")\n",
 182                         idx,
 183                         (void *)(uintptr_t)pregion->guest_phys_address,
 184                         (void *)(uintptr_t)pregion->userspace_address,
 185                          pregion->memory_size);
 186         }
 187
 188         return 0;
 189
 190 err_mmap:
 191         while (idx--) {
 192                 alignment = pregion_orig[idx].blksz;
 193                 munmap((void *)RTE_ALIGN_FLOOR(
 194                         pregion_orig[idx].mapped_address, alignment),
 195                         RTE_ALIGN_CEIL(pregion_orig[idx].mapped_size,
 196                                         alignment));
 197                 close(pregion_orig[idx].fd);
 198         }
 199         free(dev->mem);
 200         dev->mem = NULL;
 201         return -1;
 202 }
 203
 204 static int
 205 virtio_is_ready(struct virtio_net *dev)
 206 {
 207         struct vhost_virtqueue *rvq, *tvq;
 208
 209         /* mq support in future.*/
 210         rvq = dev->virtqueue[VIRTIO_RXQ];
 211         tvq = dev->virtqueue[VIRTIO_TXQ];
 212         if (rvq && tvq && rvq->desc && tvq->desc &&
 213                 (rvq->kickfd != (eventfd_t)-1) &&
 214                 (rvq->callfd != (eventfd_t)-1) &&
 215                 (tvq->kickfd != (eventfd_t)-1) &&
 216                 (tvq->callfd != (eventfd_t)-1)) {
 217                 RTE_LOG(INFO, VHOST_CONFIG,
 218                         "virtio is now ready for processing.\n");
 219                 return 1;
 220         }
 221         RTE_LOG(INFO, VHOST_CONFIG,
 222                 "virtio isn't ready for processing.\n");
 223         return 0;
 224 }
 225
 226 void
 227 user_set_vring_call(struct vhost_device_ctx ctx, struct VhostUserMsg *pmsg)
 228 {
 229         struct vhost_vring_file file;
 230
 231         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 232         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 233                 file.fd = -1;
 234         else
 235                 file.fd = pmsg->fds[0];
 236         RTE_LOG(INFO, VHOST_CONFIG,
 237                 "vring call idx:%d file:%d\n", file.index, file.fd);
 238         ops->set_vring_call(ctx, &file);
 239 }
 240
 241
 242 /*
 243  *  In vhost-user, when we receive kick message, will test whether virtio
 244  *  device is ready for packet processing.
 245  */
 246 void
 247 user_set_vring_kick(struct vhost_device_ctx ctx, struct VhostUserMsg *pmsg)
 248 {
 249         struct vhost_vring_file file;
 250         struct virtio_net *dev = get_device(ctx);
 251
 252         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 253         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 254                 file.fd = -1;
 255         else
 256                 file.fd = pmsg->fds[0];
 257         RTE_LOG(INFO, VHOST_CONFIG,
 258                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 259         ops->set_vring_kick(ctx, &file);
 260
 261         if (virtio_is_ready(dev) &&
 262                 !(dev->flags & VIRTIO_DEV_RUNNING))
 263                         notify_ops->new_device(dev);
 264 }
 265
 266 /*
 267  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 268  */
 269 int
 270 user_get_vring_base(struct vhost_device_ctx ctx,
 271         struct vhost_vring_state *state)
 272 {
 273         struct virtio_net *dev = get_device(ctx);
 274
 275         /* We have to stop the queue (virtio) if it is running. */
 276         if (dev->flags & VIRTIO_DEV_RUNNING)
 277                 notify_ops->destroy_device(dev);
 278
 279         /* Here we are safe to get the last used index */
 280         ops->get_vring_base(ctx, state->index, state);
 281
 282         RTE_LOG(INFO, VHOST_CONFIG,
 283                 "vring base idx:%d file:%d\n", state->index, state->num);
 284         /*
 285          * Based on current qemu vhost-user implementation, this message is
 286          * sent and only sent in vhost_vring_stop.
 287          * TODO: cleanup the vring, it isn't usable since here.
 288          */
 289         if (((int)dev->virtqueue[VIRTIO_RXQ]->callfd) >= 0) {
 290                 close(dev->virtqueue[VIRTIO_RXQ]->callfd);
 291                 dev->virtqueue[VIRTIO_RXQ]->callfd = (eventfd_t)-1;
 292         }
 293         if (((int)dev->virtqueue[VIRTIO_TXQ]->callfd) >= 0) {
 294                 close(dev->virtqueue[VIRTIO_TXQ]->callfd);
 295                 dev->virtqueue[VIRTIO_TXQ]->callfd = (eventfd_t)-1;
 296         }
 297
 298         return 0;
 299 }
 300
 301 void
 302 user_destroy_device(struct vhost_device_ctx ctx)
 303 {
 304         struct virtio_net *dev = get_device(ctx);
 305
 306         if (dev && (dev->flags & VIRTIO_DEV_RUNNING))
 307                 notify_ops->destroy_device(dev);
 308
 309         if (dev && dev->mem) {
 310                 free_mem_region(dev);
 311                 free(dev->mem);
 312                 dev->mem = NULL;
 313         }
 314 }