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     char serial_str[BLOCK_SERIAL_STRLEN + 1];
  29     QEMUBH *bh;
  30     size_t config_size;
  31 } VirtIOBlock;
  32
  33 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
  34 {
  35     return (VirtIOBlock *)vdev;
  36 }
  37
  38 /* store identify data in little endian format
  39  */
  40 static inline void put_le16(uint16_t *p, unsigned int v)
  41 {
  42     *p = cpu_to_le16(v);
  43 }
  44
  45 /* copy to *dst from *src, nul pad dst tail as needed to len bytes
  46  */
  47 static inline void padstr(char *dst, const char *src, int len)
  48 {
  49     while (len--)
  50         *dst++ = *src ? *src++ : '\0';
  51 }
  52
  53 /* setup simulated identify data as appropriate for virtio block device
  54  *
  55  * ref: AT Attachment 8 - ATA/ATAPI Command Set (ATA8-ACS)
  56  */
  57 static inline void virtio_identify_template(struct virtio_blk_config *bc)
  58 {
  59     uint16_t *p = &bc->identify[0];
  60     uint64_t lba_sectors = bc->capacity;
  61
  62     memset(p, 0, sizeof(bc->identify));
  63     put_le16(p + 0, 0x0);                            /* ATA device */
  64     padstr((char *)(p + 23), QEMU_VERSION, 8);       /* firmware revision */
  65     padstr((char *)(p + 27), "QEMU VIRT_BLK", 40);   /* model# */
  66     put_le16(p + 47, 0x80ff);                        /* max xfer 255 sectors */
  67     put_le16(p + 49, 0x0b00);                        /* support IORDY/LBA/DMA */
  68     put_le16(p + 59, 0x1ff);                         /* cur xfer 255 sectors */
  69     put_le16(p + 80, 0x1f0);                         /* support ATA8/7/6/5/4 */
  70     put_le16(p + 81, 0x16);
  71     put_le16(p + 82, 0x400);
  72     put_le16(p + 83, 0x400);
  73     put_le16(p + 100, lba_sectors);
  74     put_le16(p + 101, lba_sectors >> 16);
  75     put_le16(p + 102, lba_sectors >> 32);
  76     put_le16(p + 103, lba_sectors >> 48);
  77 }
  78
  79 typedef struct VirtIOBlockReq
  80 {
  81     VirtIOBlock *dev;
  82     VirtQueueElement elem;
  83     struct virtio_blk_inhdr *in;
  84     struct virtio_blk_outhdr *out;
  85     struct virtio_scsi_inhdr *scsi;
  86     QEMUIOVector qiov;
  87     struct VirtIOBlockReq *next;
  88 } VirtIOBlockReq;
  89
  90 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
  91 {
  92     VirtIOBlock *s = req->dev;
  93
  94     req->in->status = status;
  95     virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
  96     virtio_notify(&s->vdev, s->vq);
  97
  98     qemu_free(req);
  99 }
 100
 101 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
 102     int is_read)
 103 {
 104     BlockInterfaceErrorAction action =
 105         drive_get_on_error(req->dev->bs, is_read);
 106     VirtIOBlock *s = req->dev;
 107
 108     if (action == BLOCK_ERR_IGNORE)
 109         return 0;
 110
 111     if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
 112             || action == BLOCK_ERR_STOP_ANY) {
 113         req->next = s->rq;
 114         s->rq = req;
 115         vm_stop(0);
 116     } else {
 117         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 118     }
 119
 120     return 1;
 121 }
 122
 123 static void virtio_blk_rw_complete(void *opaque, int ret)
 124 {
 125     VirtIOBlockReq *req = opaque;
 126
 127     if (ret) {
 128         int is_read = !(req->out->type & VIRTIO_BLK_T_OUT);
 129         if (virtio_blk_handle_rw_error(req, -ret, is_read))
 130             return;
 131     }
 132
 133     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 134 }
 135
 136 static void virtio_blk_flush_complete(void *opaque, int ret)
 137 {
 138     VirtIOBlockReq *req = opaque;
 139
 140     virtio_blk_req_complete(req, ret ? VIRTIO_BLK_S_IOERR : VIRTIO_BLK_S_OK);
 141 }
 142
 143 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
 144 {
 145     VirtIOBlockReq *req = qemu_mallocz(sizeof(*req));
 146     req->dev = s;
 147     return req;
 148 }
 149
 150 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 151 {
 152     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 153
 154     if (req != NULL) {
 155         if (!virtqueue_pop(s->vq, &req->elem)) {
 156             qemu_free(req);
 157             return NULL;
 158         }
 159     }
 160
 161     return req;
 162 }
 163
 164 #ifdef __linux__
 165 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 166 {
 167     struct sg_io_hdr hdr;
 168     int ret, size = 0;
 169     int status;
 170     int i;
 171
 172     /*
 173      * We require at least one output segment each for the virtio_blk_outhdr
 174      * and the SCSI command block.
 175      *
 176      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 177      * and the sense buffer pointer in the input segments.
 178      */
 179     if (req->elem.out_num < 2 || req->elem.in_num < 3) {
 180         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 181         return;
 182     }
 183
 184     /*
 185      * No support for bidirection commands yet.
 186      */
 187     if (req->elem.out_num > 2 && req->elem.in_num > 3) {
 188         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 189         return;
 190     }
 191
 192     /*
 193      * The scsi inhdr is placed in the second-to-last input segment, just
 194      * before the regular inhdr.
 195      */
 196     req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
 197     size = sizeof(*req->in) + sizeof(*req->scsi);
 198
 199     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 200     hdr.interface_id = 'S';
 201     hdr.cmd_len = req->elem.out_sg[1].iov_len;
 202     hdr.cmdp = req->elem.out_sg[1].iov_base;
 203     hdr.dxfer_len = 0;
 204
 205     if (req->elem.out_num > 2) {
 206         /*
 207          * If there are more than the minimally required 2 output segments
 208          * there is write payload starting from the third iovec.
 209          */
 210         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 211         hdr.iovec_count = req->elem.out_num - 2;
 212
 213         for (i = 0; i < hdr.iovec_count; i++)
 214             hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
 215
 216         hdr.dxferp = req->elem.out_sg + 2;
 217
 218     } else if (req->elem.in_num > 3) {
 219         /*
 220          * If we have more than 3 input segments the guest wants to actually
 221          * read data.
 222          */
 223         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 224         hdr.iovec_count = req->elem.in_num - 3;
 225         for (i = 0; i < hdr.iovec_count; i++)
 226             hdr.dxfer_len += req->elem.in_sg[i].iov_len;
 227
 228         hdr.dxferp = req->elem.in_sg;
 229         size += hdr.dxfer_len;
 230     } else {
 231         /*
 232          * Some SCSI commands don't actually transfer any data.
 233          */
 234         hdr.dxfer_direction = SG_DXFER_NONE;
 235     }
 236
 237     hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
 238     hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
 239     size += hdr.mx_sb_len;
 240
 241     ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
 242     if (ret) {
 243         status = VIRTIO_BLK_S_UNSUPP;
 244         hdr.status = ret;
 245         hdr.resid = hdr.dxfer_len;
 246     } else if (hdr.status) {
 247         status = VIRTIO_BLK_S_IOERR;
 248     } else {
 249         status = VIRTIO_BLK_S_OK;
 250     }
 251
 252     req->scsi->errors = hdr.status;
 253     req->scsi->residual = hdr.resid;
 254     req->scsi->sense_len = hdr.sb_len_wr;
 255     req->scsi->data_len = hdr.dxfer_len;
 256
 257     virtio_blk_req_complete(req, status);
 258 }
 259 #else
 260 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 261 {
 262     virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 263 }
 264 #endif /* __linux__ */
 265
 266 static void do_multiwrite(BlockDriverState *bs, BlockRequest *blkreq,
 267     int num_writes)
 268 {
 269     int i, ret;
 270     ret = bdrv_aio_multiwrite(bs, blkreq, num_writes);
 271
 272     if (ret != 0) {
 273         for (i = 0; i < num_writes; i++) {
 274             if (blkreq[i].error) {
 275                 virtio_blk_req_complete(blkreq[i].opaque, VIRTIO_BLK_S_IOERR);
 276             }
 277         }
 278     }
 279 }
 280
 281 static void virtio_blk_handle_flush(VirtIOBlockReq *req)
 282 {
 283     BlockDriverAIOCB *acb;
 284
 285     acb = bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 286     if (!acb) {
 287         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 288     }
 289 }
 290
 291 static void virtio_blk_handle_write(BlockRequest *blkreq, int *num_writes,
 292     VirtIOBlockReq *req, BlockDriverState **old_bs)
 293 {
 294     if (req->dev->bs != *old_bs || *num_writes == 32) {
 295         if (*old_bs != NULL) {
 296             do_multiwrite(*old_bs, blkreq, *num_writes);
 297         }
 298         *num_writes = 0;
 299         *old_bs = req->dev->bs;
 300     }
 301
 302     blkreq[*num_writes].sector = req->out->sector;
 303     blkreq[*num_writes].nb_sectors = req->qiov.size / 512;
 304     blkreq[*num_writes].qiov = &req->qiov;
 305     blkreq[*num_writes].cb = virtio_blk_rw_complete;
 306     blkreq[*num_writes].opaque = req;
 307     blkreq[*num_writes].error = 0;
 308
 309     (*num_writes)++;
 310 }
 311
 312 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 313 {
 314     BlockDriverAIOCB *acb;
 315
 316     acb = bdrv_aio_readv(req->dev->bs, req->out->sector, &req->qiov,
 317                          req->qiov.size / 512, virtio_blk_rw_complete, req);
 318     if (!acb) {
 319         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 320     }
 321 }
 322
 323 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 324 {
 325     VirtIOBlock *s = to_virtio_blk(vdev);
 326     VirtIOBlockReq *req;
 327     BlockRequest blkreq[32];
 328     int num_writes = 0;
 329     BlockDriverState *old_bs = NULL;
 330
 331     while ((req = virtio_blk_get_request(s))) {
 332         if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 333             fprintf(stderr, "virtio-blk missing headers\n");
 334             exit(1);
 335         }
 336
 337         if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 338             req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 339             fprintf(stderr, "virtio-blk header not in correct element\n");
 340             exit(1);
 341         }
 342
 343         req->out = (void *)req->elem.out_sg[0].iov_base;
 344         req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 345
 346         if (req->out->type & VIRTIO_BLK_T_FLUSH) {
 347             virtio_blk_handle_flush(req);
 348         } else if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
 349             virtio_blk_handle_scsi(req);
 350         } else if (req->out->type & VIRTIO_BLK_T_OUT) {
 351             qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 352                                      req->elem.out_num - 1);
 353             virtio_blk_handle_write(blkreq, &num_writes, req, &old_bs);
 354         } else {
 355             qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 356                                      req->elem.in_num - 1);
 357             virtio_blk_handle_read(req);
 358         }
 359     }
 360
 361     if (num_writes > 0) {
 362         do_multiwrite(old_bs, blkreq, num_writes);
 363     }
 364
 365     /*
 366      * FIXME: Want to check for completions before returning to guest mode,
 367      * so cached reads and writes are reported as quickly as possible. But
 368      * that should be done in the generic block layer.
 369      */
 370 }
 371
 372 static void virtio_blk_dma_restart_bh(void *opaque)
 373 {
 374     VirtIOBlock *s = opaque;
 375     VirtIOBlockReq *req = s->rq;
 376
 377     qemu_bh_delete(s->bh);
 378     s->bh = NULL;
 379
 380     s->rq = NULL;
 381
 382     while (req) {
 383         bdrv_aio_writev(req->dev->bs, req->out->sector, &req->qiov,
 384             req->qiov.size / 512, virtio_blk_rw_complete, req);
 385         req = req->next;
 386     }
 387 }
 388
 389 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
 390 {
 391     VirtIOBlock *s = opaque;
 392
 393     if (!running)
 394         return;
 395
 396     if (!s->bh) {
 397         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 398         qemu_bh_schedule(s->bh);
 399     }
 400 }
 401
 402 static void virtio_blk_reset(VirtIODevice *vdev)
 403 {
 404     /*
 405      * This should cancel pending requests, but can't do nicely until there
 406      * are per-device request lists.
 407      */
 408     qemu_aio_flush();
 409 }
 410
 411 /* coalesce internal state, copy to pci i/o region 0
 412  */
 413 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 414 {
 415     VirtIOBlock *s = to_virtio_blk(vdev);
 416     struct virtio_blk_config blkcfg;
 417     uint64_t capacity;
 418     int cylinders, heads, secs;
 419
 420     bdrv_get_geometry(s->bs, &capacity);
 421     bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
 422     memset(&blkcfg, 0, sizeof(blkcfg));
 423     stq_raw(&blkcfg.capacity, capacity);
 424     stl_raw(&blkcfg.seg_max, 128 - 2);
 425     stw_raw(&blkcfg.cylinders, cylinders);
 426     blkcfg.heads = heads;
 427     blkcfg.sectors = secs;
 428     blkcfg.size_max = 0;
 429     virtio_identify_template(&blkcfg);
 430     memcpy(&blkcfg.identify[VIRTIO_BLK_ID_SN], s->serial_str,
 431         VIRTIO_BLK_ID_SN_BYTES);
 432     memcpy(config, &blkcfg, s->config_size);
 433 }
 434
 435 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 436 {
 437     VirtIOBlock *s = to_virtio_blk(vdev);
 438
 439     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 440     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 441
 442     if (bdrv_enable_write_cache(s->bs))
 443         features |= (1 << VIRTIO_BLK_F_WCACHE);
 444     if (strcmp(s->serial_str, "0"))
 445         features |= 1 << VIRTIO_BLK_F_IDENTIFY;
 446
 447     if (bdrv_is_read_only(s->bs))
 448         features |= 1 << VIRTIO_BLK_F_RO;
 449
 450     return features;
 451 }
 452
 453 static void virtio_blk_save(QEMUFile *f, void *opaque)
 454 {
 455     VirtIOBlock *s = opaque;
 456     VirtIOBlockReq *req = s->rq;
 457
 458     virtio_save(&s->vdev, f);
 459
 460     while (req) {
 461         qemu_put_sbyte(f, 1);
 462         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 463         req = req->next;
 464     }
 465     qemu_put_sbyte(f, 0);
 466 }
 467
 468 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 469 {
 470     VirtIOBlock *s = opaque;
 471
 472     if (version_id != 2)
 473         return -EINVAL;
 474
 475     virtio_load(&s->vdev, f);
 476     while (qemu_get_sbyte(f)) {
 477         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 478         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 479         req->next = s->rq;
 480         s->rq = req->next;
 481     }
 482
 483     return 0;
 484 }
 485
 486 VirtIODevice *virtio_blk_init(DeviceState *dev, DriveInfo *dinfo)
 487 {
 488     VirtIOBlock *s;
 489     int cylinders, heads, secs;
 490     static int virtio_blk_id;
 491     char *ps = (char *)drive_get_serial(dinfo->bdrv);
 492     size_t size = strlen(ps) ? sizeof(struct virtio_blk_config) :
 493             offsetof(struct virtio_blk_config, _blk_size);
 494
 495     s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
 496                                           size,
 497                                           sizeof(VirtIOBlock));
 498
 499     s->config_size = size;
 500     s->vdev.get_config = virtio_blk_update_config;
 501     s->vdev.get_features = virtio_blk_get_features;
 502     s->vdev.reset = virtio_blk_reset;
 503     s->bs = dinfo->bdrv;
 504     s->rq = NULL;
 505     if (strlen(ps))
 506         strncpy(s->serial_str, ps, sizeof(s->serial_str));
 507     else
 508         snprintf(s->serial_str, sizeof(s->serial_str), "0");
 509     bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
 510     bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
 511
 512     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 513
 514     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 515     register_savevm("virtio-blk", virtio_blk_id++, 2,
 516                     virtio_blk_save, virtio_blk_load, s);
 517
 518     return &s->vdev;
 519 }