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     QEMUBH *bh;
  29     BlockConf *conf;
  30 } VirtIOBlock;
  31
  32 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
  33 {
  34     return (VirtIOBlock *)vdev;
  35 }
  36
  37 typedef struct VirtIOBlockReq
  38 {
  39     VirtIOBlock *dev;
  40     VirtQueueElement elem;
  41     struct virtio_blk_inhdr *in;
  42     struct virtio_blk_outhdr *out;
  43     struct virtio_scsi_inhdr *scsi;
  44     QEMUIOVector qiov;
  45     struct VirtIOBlockReq *next;
  46 } VirtIOBlockReq;
  47
  48 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
  49 {
  50     VirtIOBlock *s = req->dev;
  51
  52     req->in->status = status;
  53     virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
  54     virtio_notify(&s->vdev, s->vq);
  55
  56     qemu_free(req);
  57 }
  58
  59 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
  60     int is_read)
  61 {
  62     BlockInterfaceErrorAction action =
  63         drive_get_on_error(req->dev->bs, is_read);
  64     VirtIOBlock *s = req->dev;
  65
  66     if (action == BLOCK_ERR_IGNORE) {
  67         bdrv_mon_event(req->dev->bs, BDRV_ACTION_IGNORE, is_read);
  68         return 0;
  69     }
  70
  71     if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
  72             || action == BLOCK_ERR_STOP_ANY) {
  73         req->next = s->rq;
  74         s->rq = req;
  75         vm_stop(0);
  76         bdrv_mon_event(req->dev->bs, BDRV_ACTION_STOP, is_read);
  77     } else {
  78         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
  79         bdrv_mon_event(req->dev->bs, BDRV_ACTION_REPORT, is_read);
  80     }
  81
  82     return 1;
  83 }
  84
  85 static void virtio_blk_rw_complete(void *opaque, int ret)
  86 {
  87     VirtIOBlockReq *req = opaque;
  88
  89     if (ret) {
  90         int is_read = !(req->out->type & VIRTIO_BLK_T_OUT);
  91         if (virtio_blk_handle_rw_error(req, -ret, is_read))
  92             return;
  93     }
  94
  95     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
  96 }
  97
  98 static void virtio_blk_flush_complete(void *opaque, int ret)
  99 {
 100     VirtIOBlockReq *req = opaque;
 101
 102     virtio_blk_req_complete(req, ret ? VIRTIO_BLK_S_IOERR : VIRTIO_BLK_S_OK);
 103 }
 104
 105 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
 106 {
 107     VirtIOBlockReq *req = qemu_mallocz(sizeof(*req));
 108     req->dev = s;
 109     return req;
 110 }
 111
 112 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 113 {
 114     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 115
 116     if (req != NULL) {
 117         if (!virtqueue_pop(s->vq, &req->elem)) {
 118             qemu_free(req);
 119             return NULL;
 120         }
 121     }
 122
 123     return req;
 124 }
 125
 126 #ifdef __linux__
 127 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 128 {
 129     struct sg_io_hdr hdr;
 130     int ret;
 131     int status;
 132     int i;
 133
 134     /*
 135      * We require at least one output segment each for the virtio_blk_outhdr
 136      * and the SCSI command block.
 137      *
 138      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 139      * and the sense buffer pointer in the input segments.
 140      */
 141     if (req->elem.out_num < 2 || req->elem.in_num < 3) {
 142         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 143         return;
 144     }
 145
 146     /*
 147      * No support for bidirection commands yet.
 148      */
 149     if (req->elem.out_num > 2 && req->elem.in_num > 3) {
 150         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 151         return;
 152     }
 153
 154     /*
 155      * The scsi inhdr is placed in the second-to-last input segment, just
 156      * before the regular inhdr.
 157      */
 158     req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
 159
 160     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 161     hdr.interface_id = 'S';
 162     hdr.cmd_len = req->elem.out_sg[1].iov_len;
 163     hdr.cmdp = req->elem.out_sg[1].iov_base;
 164     hdr.dxfer_len = 0;
 165
 166     if (req->elem.out_num > 2) {
 167         /*
 168          * If there are more than the minimally required 2 output segments
 169          * there is write payload starting from the third iovec.
 170          */
 171         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 172         hdr.iovec_count = req->elem.out_num - 2;
 173
 174         for (i = 0; i < hdr.iovec_count; i++)
 175             hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
 176
 177         hdr.dxferp = req->elem.out_sg + 2;
 178
 179     } else if (req->elem.in_num > 3) {
 180         /*
 181          * If we have more than 3 input segments the guest wants to actually
 182          * read data.
 183          */
 184         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 185         hdr.iovec_count = req->elem.in_num - 3;
 186         for (i = 0; i < hdr.iovec_count; i++)
 187             hdr.dxfer_len += req->elem.in_sg[i].iov_len;
 188
 189         hdr.dxferp = req->elem.in_sg;
 190     } else {
 191         /*
 192          * Some SCSI commands don't actually transfer any data.
 193          */
 194         hdr.dxfer_direction = SG_DXFER_NONE;
 195     }
 196
 197     hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
 198     hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
 199
 200     ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
 201     if (ret) {
 202         status = VIRTIO_BLK_S_UNSUPP;
 203         hdr.status = ret;
 204         hdr.resid = hdr.dxfer_len;
 205     } else if (hdr.status) {
 206         status = VIRTIO_BLK_S_IOERR;
 207     } else {
 208         status = VIRTIO_BLK_S_OK;
 209     }
 210
 211     req->scsi->errors = hdr.status;
 212     req->scsi->residual = hdr.resid;
 213     req->scsi->sense_len = hdr.sb_len_wr;
 214     req->scsi->data_len = hdr.dxfer_len;
 215
 216     virtio_blk_req_complete(req, status);
 217 }
 218 #else
 219 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 220 {
 221     virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 222 }
 223 #endif /* __linux__ */
 224
 225 static void do_multiwrite(BlockDriverState *bs, BlockRequest *blkreq,
 226     int num_writes)
 227 {
 228     int i, ret;
 229     ret = bdrv_aio_multiwrite(bs, blkreq, num_writes);
 230
 231     if (ret != 0) {
 232         for (i = 0; i < num_writes; i++) {
 233             if (blkreq[i].error) {
 234                 virtio_blk_rw_complete(blkreq[i].opaque, -EIO);
 235             }
 236         }
 237     }
 238 }
 239
 240 static void virtio_blk_handle_flush(VirtIOBlockReq *req)
 241 {
 242     BlockDriverAIOCB *acb;
 243
 244     acb = bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 245     if (!acb) {
 246         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 247     }
 248 }
 249
 250 static void virtio_blk_handle_write(BlockRequest *blkreq, int *num_writes,
 251     VirtIOBlockReq *req, BlockDriverState **old_bs)
 252 {
 253     if (req->dev->bs != *old_bs || *num_writes == 32) {
 254         if (*old_bs != NULL) {
 255             do_multiwrite(*old_bs, blkreq, *num_writes);
 256         }
 257         *num_writes = 0;
 258         *old_bs = req->dev->bs;
 259     }
 260
 261     blkreq[*num_writes].sector = req->out->sector;
 262     blkreq[*num_writes].nb_sectors = req->qiov.size / 512;
 263     blkreq[*num_writes].qiov = &req->qiov;
 264     blkreq[*num_writes].cb = virtio_blk_rw_complete;
 265     blkreq[*num_writes].opaque = req;
 266     blkreq[*num_writes].error = 0;
 267
 268     (*num_writes)++;
 269 }
 270
 271 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 272 {
 273     BlockDriverAIOCB *acb;
 274
 275     acb = bdrv_aio_readv(req->dev->bs, req->out->sector, &req->qiov,
 276                          req->qiov.size / 512, virtio_blk_rw_complete, req);
 277     if (!acb) {
 278         virtio_blk_rw_complete(req, -EIO);
 279     }
 280 }
 281
 282 typedef struct MultiReqBuffer {
 283     BlockRequest        blkreq[32];
 284     int                 num_writes;
 285     BlockDriverState    *old_bs;
 286 } MultiReqBuffer;
 287
 288 static void virtio_blk_handle_request(VirtIOBlockReq *req,
 289     MultiReqBuffer *mrb)
 290 {
 291     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 292         fprintf(stderr, "virtio-blk missing headers\n");
 293         exit(1);
 294     }
 295
 296     if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 297         req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 298         fprintf(stderr, "virtio-blk header not in correct element\n");
 299         exit(1);
 300     }
 301
 302     req->out = (void *)req->elem.out_sg[0].iov_base;
 303     req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 304
 305     if (req->out->type & VIRTIO_BLK_T_FLUSH) {
 306         virtio_blk_handle_flush(req);
 307     } else if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
 308         virtio_blk_handle_scsi(req);
 309     } else if (req->out->type & VIRTIO_BLK_T_OUT) {
 310         qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 311                                  req->elem.out_num - 1);
 312         virtio_blk_handle_write(mrb->blkreq, &mrb->num_writes,
 313             req, &mrb->old_bs);
 314     } else {
 315         qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 316                                  req->elem.in_num - 1);
 317         virtio_blk_handle_read(req);
 318     }
 319 }
 320
 321 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 322 {
 323     VirtIOBlock *s = to_virtio_blk(vdev);
 324     VirtIOBlockReq *req;
 325     MultiReqBuffer mrb = {
 326         .num_writes = 0,
 327         .old_bs = NULL,
 328     };
 329
 330     while ((req = virtio_blk_get_request(s))) {
 331         virtio_blk_handle_request(req, &mrb);
 332     }
 333
 334     if (mrb.num_writes > 0) {
 335         do_multiwrite(mrb.old_bs, mrb.blkreq, mrb.num_writes);
 336     }
 337
 338     /*
 339      * FIXME: Want to check for completions before returning to guest mode,
 340      * so cached reads and writes are reported as quickly as possible. But
 341      * that should be done in the generic block layer.
 342      */
 343 }
 344
 345 static void virtio_blk_dma_restart_bh(void *opaque)
 346 {
 347     VirtIOBlock *s = opaque;
 348     VirtIOBlockReq *req = s->rq;
 349     MultiReqBuffer mrb = {
 350         .num_writes = 0,
 351         .old_bs = NULL,
 352     };
 353
 354     qemu_bh_delete(s->bh);
 355     s->bh = NULL;
 356
 357     s->rq = NULL;
 358
 359     while (req) {
 360         virtio_blk_handle_request(req, &mrb);
 361         req = req->next;
 362     }
 363
 364     if (mrb.num_writes > 0) {
 365         do_multiwrite(mrb.old_bs, mrb.blkreq, mrb.num_writes);
 366     }
 367 }
 368
 369 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
 370 {
 371     VirtIOBlock *s = opaque;
 372
 373     if (!running)
 374         return;
 375
 376     if (!s->bh) {
 377         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 378         qemu_bh_schedule(s->bh);
 379     }
 380 }
 381
 382 static void virtio_blk_reset(VirtIODevice *vdev)
 383 {
 384     /*
 385      * This should cancel pending requests, but can't do nicely until there
 386      * are per-device request lists.
 387      */
 388     qemu_aio_flush();
 389 }
 390
 391 /* coalesce internal state, copy to pci i/o region 0
 392  */
 393 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 394 {
 395     VirtIOBlock *s = to_virtio_blk(vdev);
 396     struct virtio_blk_config blkcfg;
 397     uint64_t capacity;
 398     int cylinders, heads, secs;
 399
 400     bdrv_get_geometry(s->bs, &capacity);
 401     bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
 402     memset(&blkcfg, 0, sizeof(blkcfg));
 403     stq_raw(&blkcfg.capacity, capacity);
 404     stl_raw(&blkcfg.seg_max, 128 - 2);
 405     stw_raw(&blkcfg.cylinders, cylinders);
 406     blkcfg.heads = heads;
 407     blkcfg.sectors = secs;
 408     blkcfg.size_max = 0;
 409     blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
 410     blkcfg.alignment_offset = 0;
 411     blkcfg.min_io_size = s->conf->min_io_size / 512;
 412     blkcfg.opt_io_size = s->conf->opt_io_size / 512;
 413     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
 414 }
 415
 416 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 417 {
 418     VirtIOBlock *s = to_virtio_blk(vdev);
 419
 420     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 421     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 422     features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
 423
 424     if (bdrv_enable_write_cache(s->bs))
 425         features |= (1 << VIRTIO_BLK_F_WCACHE);
 426
 427     if (bdrv_is_read_only(s->bs))
 428         features |= 1 << VIRTIO_BLK_F_RO;
 429
 430     return features;
 431 }
 432
 433 static void virtio_blk_save(QEMUFile *f, void *opaque)
 434 {
 435     VirtIOBlock *s = opaque;
 436     VirtIOBlockReq *req = s->rq;
 437
 438     virtio_save(&s->vdev, f);
 439
 440     while (req) {
 441         qemu_put_sbyte(f, 1);
 442         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 443         req = req->next;
 444     }
 445     qemu_put_sbyte(f, 0);
 446 }
 447
 448 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 449 {
 450     VirtIOBlock *s = opaque;
 451
 452     if (version_id != 2)
 453         return -EINVAL;
 454
 455     virtio_load(&s->vdev, f);
 456     while (qemu_get_sbyte(f)) {
 457         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 458         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 459         req->next = s->rq;
 460         s->rq = req->next;
 461     }
 462
 463     return 0;
 464 }
 465
 466 VirtIODevice *virtio_blk_init(DeviceState *dev, BlockConf *conf)
 467 {
 468     VirtIOBlock *s;
 469     int cylinders, heads, secs;
 470     static int virtio_blk_id;
 471
 472     s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
 473                                           sizeof(struct virtio_blk_config),
 474                                           sizeof(VirtIOBlock));
 475
 476     s->vdev.get_config = virtio_blk_update_config;
 477     s->vdev.get_features = virtio_blk_get_features;
 478     s->vdev.reset = virtio_blk_reset;
 479     s->bs = conf->dinfo->bdrv;
 480     s->conf = conf;
 481     s->rq = NULL;
 482     bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
 483     bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
 484
 485     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 486
 487     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 488     register_savevm("virtio-blk", virtio_blk_id++, 2,
 489                     virtio_blk_save, virtio_blk_load, s);
 490
 491     return &s->vdev;
 492 }