hw/virtio-blk.c

   1 /*
   2  * Virtio Block Device
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include <qemu-common.h>
  15 #include <sysemu.h>
  16 #include "virtio-blk.h"
  17 #include "block_int.h"
  18 #ifdef __linux__
  19 # include <scsi/sg.h>
  20 #endif
  21
  22 typedef struct VirtIOBlock
  23 {
  24     VirtIODevice vdev;
  25     BlockDriverState *bs;
  26     VirtQueue *vq;
  27     void *rq;
  28 } VirtIOBlock;
  29
  30 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
  31 {
  32     return (VirtIOBlock *)vdev;
  33 }
  34
  35 typedef struct VirtIOBlockReq
  36 {
  37     VirtIOBlock *dev;
  38     VirtQueueElement elem;
  39     struct virtio_blk_inhdr *in;
  40     struct virtio_blk_outhdr *out;
  41     struct virtio_scsi_inhdr *scsi;
  42     QEMUIOVector qiov;
  43     struct VirtIOBlockReq *next;
  44 } VirtIOBlockReq;
  45
  46 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
  47 {
  48     VirtIOBlock *s = req->dev;
  49
  50     req->in->status = status;
  51     virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
  52     virtio_notify(&s->vdev, s->vq);
  53
  54     qemu_free(req);
  55 }
  56
  57 static int virtio_blk_handle_write_error(VirtIOBlockReq *req, int error)
  58 {
  59     BlockInterfaceErrorAction action = drive_get_onerror(req->dev->bs);
  60     VirtIOBlock *s = req->dev;
  61
  62     if (action == BLOCK_ERR_IGNORE)
  63         return 0;
  64
  65     if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
  66             || action == BLOCK_ERR_STOP_ANY) {
  67         req->next = s->rq;
  68         s->rq = req;
  69         vm_stop(0);
  70     } else {
  71         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
  72     }
  73
  74     return 1;
  75 }
  76
  77 static void virtio_blk_rw_complete(void *opaque, int ret)
  78 {
  79     VirtIOBlockReq *req = opaque;
  80
  81     if (ret && (req->out->type & VIRTIO_BLK_T_OUT)) {
  82         if (virtio_blk_handle_write_error(req, -ret))
  83             return;
  84     }
  85
  86     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
  87 }
  88
  89 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
  90 {
  91     VirtIOBlockReq *req = qemu_mallocz(sizeof(*req));
  92     req->dev = s;
  93     return req;
  94 }
  95
  96 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
  97 {
  98     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
  99
 100     if (req != NULL) {
 101         if (!virtqueue_pop(s->vq, &req->elem)) {
 102             qemu_free(req);
 103             return NULL;
 104         }
 105     }
 106
 107     return req;
 108 }
 109
 110 #ifdef __linux__
 111 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 112 {
 113     struct sg_io_hdr hdr;
 114     int ret, size = 0;
 115     int status;
 116     int i;
 117
 118     /*
 119      * We require at least one output segment each for the virtio_blk_outhdr
 120      * and the SCSI command block.
 121      *
 122      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 123      * and the sense buffer pointer in the input segments.
 124      */
 125     if (req->elem.out_num < 2 || req->elem.in_num < 3) {
 126         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 127         return;
 128     }
 129
 130     /*
 131      * No support for bidirection commands yet.
 132      */
 133     if (req->elem.out_num > 2 && req->elem.in_num > 3) {
 134         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 135         return;
 136     }
 137
 138     /*
 139      * The scsi inhdr is placed in the second-to-last input segment, just
 140      * before the regular inhdr.
 141      */
 142     req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
 143     size = sizeof(*req->in) + sizeof(*req->scsi);
 144
 145     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 146     hdr.interface_id = 'S';
 147     hdr.cmd_len = req->elem.out_sg[1].iov_len;
 148     hdr.cmdp = req->elem.out_sg[1].iov_base;
 149     hdr.dxfer_len = 0;
 150
 151     if (req->elem.out_num > 2) {
 152         /*
 153          * If there are more than the minimally required 2 output segments
 154          * there is write payload starting from the third iovec.
 155          */
 156         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 157         hdr.iovec_count = req->elem.out_num - 2;
 158
 159         for (i = 0; i < hdr.iovec_count; i++)
 160             hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
 161
 162         hdr.dxferp = req->elem.out_sg + 2;
 163
 164     } else if (req->elem.in_num > 3) {
 165         /*
 166          * If we have more than 3 input segments the guest wants to actually
 167          * read data.
 168          */
 169         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 170         hdr.iovec_count = req->elem.in_num - 3;
 171         for (i = 0; i < hdr.iovec_count; i++)
 172             hdr.dxfer_len += req->elem.in_sg[i].iov_len;
 173
 174         hdr.dxferp = req->elem.in_sg;
 175         size += hdr.dxfer_len;
 176     } else {
 177         /*
 178          * Some SCSI commands don't actually transfer any data.
 179          */
 180         hdr.dxfer_direction = SG_DXFER_NONE;
 181     }
 182
 183     hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
 184     hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
 185     size += hdr.mx_sb_len;
 186
 187     ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
 188     if (ret) {
 189         status = VIRTIO_BLK_S_UNSUPP;
 190         hdr.status = ret;
 191         hdr.resid = hdr.dxfer_len;
 192     } else if (hdr.status) {
 193         status = VIRTIO_BLK_S_IOERR;
 194     } else {
 195         status = VIRTIO_BLK_S_OK;
 196     }
 197
 198     req->scsi->errors = hdr.status;
 199     req->scsi->residual = hdr.resid;
 200     req->scsi->sense_len = hdr.sb_len_wr;
 201     req->scsi->data_len = hdr.dxfer_len;
 202
 203     virtio_blk_req_complete(req, status);
 204 }
 205 #else
 206 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 207 {
 208     virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 209 }
 210 #endif /* __linux__ */
 211
 212 static void virtio_blk_handle_write(VirtIOBlockReq *req)
 213 {
 214     bdrv_aio_writev(req->dev->bs, req->out->sector, &req->qiov,
 215                     req->qiov.size / 512, virtio_blk_rw_complete, req);
 216 }
 217
 218 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 219 {
 220     bdrv_aio_readv(req->dev->bs, req->out->sector, &req->qiov,
 221                    req->qiov.size / 512, virtio_blk_rw_complete, req);
 222 }
 223
 224 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 225 {
 226     VirtIOBlock *s = to_virtio_blk(vdev);
 227     VirtIOBlockReq *req;
 228
 229     while ((req = virtio_blk_get_request(s))) {
 230         if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 231             fprintf(stderr, "virtio-blk missing headers\n");
 232             exit(1);
 233         }
 234
 235         if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 236             req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 237             fprintf(stderr, "virtio-blk header not in correct element\n");
 238             exit(1);
 239         }
 240
 241         req->out = (void *)req->elem.out_sg[0].iov_base;
 242         req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 243
 244         if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
 245             virtio_blk_handle_scsi(req);
 246         } else if (req->out->type & VIRTIO_BLK_T_OUT) {
 247             qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 248                                      req->elem.out_num - 1);
 249             virtio_blk_handle_write(req);
 250         } else {
 251             qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 252                                      req->elem.in_num - 1);
 253             virtio_blk_handle_read(req);
 254         }
 255     }
 256     /*
 257      * FIXME: Want to check for completions before returning to guest mode,
 258      * so cached reads and writes are reported as quickly as possible. But
 259      * that should be done in the generic block layer.
 260      */
 261 }
 262
 263 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
 264 {
 265     VirtIOBlock *s = opaque;
 266     VirtIOBlockReq *req = s->rq;
 267
 268     if (!running)
 269         return;
 270
 271     s->rq = NULL;
 272
 273     while (req) {
 274         virtio_blk_handle_write(req);
 275         req = req->next;
 276     }
 277 }
 278
 279 static void virtio_blk_reset(VirtIODevice *vdev)
 280 {
 281     /*
 282      * This should cancel pending requests, but can't do nicely until there
 283      * are per-device request lists.
 284      */
 285     qemu_aio_flush();
 286 }
 287
 288 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 289 {
 290     VirtIOBlock *s = to_virtio_blk(vdev);
 291     struct virtio_blk_config blkcfg;
 292     uint64_t capacity;
 293     int cylinders, heads, secs;
 294
 295     bdrv_get_geometry(s->bs, &capacity);
 296     bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
 297     stq_raw(&blkcfg.capacity, capacity);
 298     stl_raw(&blkcfg.seg_max, 128 - 2);
 299     stw_raw(&blkcfg.cylinders, cylinders);
 300     blkcfg.heads = heads;
 301     blkcfg.sectors = secs;
 302     memcpy(config, &blkcfg, sizeof(blkcfg));
 303 }
 304
 305 static uint32_t virtio_blk_get_features(VirtIODevice *vdev)
 306 {
 307     uint32_t features = 0;
 308
 309     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 310     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 311 #ifdef __linux__
 312     features |= (1 << VIRTIO_BLK_F_SCSI);
 313 #endif
 314
 315     return features;
 316 }
 317
 318 static void virtio_blk_save(QEMUFile *f, void *opaque)
 319 {
 320     VirtIOBlock *s = opaque;
 321     VirtIOBlockReq *req = s->rq;
 322
 323     virtio_save(&s->vdev, f);
 324
 325     while (req) {
 326         qemu_put_sbyte(f, 1);
 327         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 328         req = req->next;
 329     }
 330     qemu_put_sbyte(f, 0);
 331 }
 332
 333 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 334 {
 335     VirtIOBlock *s = opaque;
 336
 337     if (version_id != 2)
 338         return -EINVAL;
 339
 340     virtio_load(&s->vdev, f);
 341     while (qemu_get_sbyte(f)) {
 342         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 343         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 344         req->next = s->rq;
 345         s->rq = req->next;
 346     }
 347
 348     return 0;
 349 }
 350
 351 VirtIODevice *virtio_blk_init(DeviceState *dev)
 352 {
 353     VirtIOBlock *s;
 354     int cylinders, heads, secs;
 355     static int virtio_blk_id;
 356     BlockDriverState *bs;
 357
 358     s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
 359                                           sizeof(struct virtio_blk_config),
 360                                           sizeof(VirtIOBlock));
 361
 362     bs = qdev_init_bdrv(dev, IF_VIRTIO);
 363     s->vdev.get_config = virtio_blk_update_config;
 364     s->vdev.get_features = virtio_blk_get_features;
 365     s->vdev.reset = virtio_blk_reset;
 366     s->bs = bs;
 367     s->rq = NULL;
 368     bs->private = dev;
 369     bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
 370     bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
 371
 372     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 373
 374     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 375     register_savevm("virtio-blk", virtio_blk_id++, 2,
 376                     virtio_blk_save, virtio_blk_load, s);
 377
 378     return &s->vdev;
 379 }