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_rw_complete(blkreq[i].opaque, -EIO);
 276             }
 277         }
 278     }
 279 }
 280
 281 static void virtio_blk_handle_flush(BlockRequest *blkreq, int *num_writes,
 282     VirtIOBlockReq *req, BlockDriverState **old_bs)
 283 {
 284     BlockDriverAIOCB *acb;
 285
 286     /*
 287      * Make sure all outstanding writes are posted to the backing device.
 288      */
 289     if (*old_bs != NULL) {
 290         do_multiwrite(*old_bs, blkreq, *num_writes);
 291     }
 292     *num_writes = 0;
 293     *old_bs = req->dev->bs;
 294
 295     acb = bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 296     if (!acb) {
 297         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 298     }
 299 }
 300
 301 static void virtio_blk_handle_write(BlockRequest *blkreq, int *num_writes,
 302     VirtIOBlockReq *req, BlockDriverState **old_bs)
 303 {
 304     if (req->dev->bs != *old_bs || *num_writes == 32) {
 305         if (*old_bs != NULL) {
 306             do_multiwrite(*old_bs, blkreq, *num_writes);
 307         }
 308         *num_writes = 0;
 309         *old_bs = req->dev->bs;
 310     }
 311
 312     blkreq[*num_writes].sector = req->out->sector;
 313     blkreq[*num_writes].nb_sectors = req->qiov.size / 512;
 314     blkreq[*num_writes].qiov = &req->qiov;
 315     blkreq[*num_writes].cb = virtio_blk_rw_complete;
 316     blkreq[*num_writes].opaque = req;
 317     blkreq[*num_writes].error = 0;
 318
 319     (*num_writes)++;
 320 }
 321
 322 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 323 {
 324     BlockDriverAIOCB *acb;
 325
 326     acb = bdrv_aio_readv(req->dev->bs, req->out->sector, &req->qiov,
 327                          req->qiov.size / 512, virtio_blk_rw_complete, req);
 328     if (!acb) {
 329         virtio_blk_rw_complete(req, -EIO);
 330     }
 331 }
 332
 333 typedef struct MultiReqBuffer {
 334     BlockRequest        blkreq[32];
 335     int                 num_writes;
 336     BlockDriverState    *old_bs;
 337 } MultiReqBuffer;
 338
 339 static void virtio_blk_handle_request(VirtIOBlockReq *req,
 340     MultiReqBuffer *mrb)
 341 {
 342     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 343         fprintf(stderr, "virtio-blk missing headers\n");
 344         exit(1);
 345     }
 346
 347     if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 348         req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 349         fprintf(stderr, "virtio-blk header not in correct element\n");
 350         exit(1);
 351     }
 352
 353     req->out = (void *)req->elem.out_sg[0].iov_base;
 354     req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 355
 356     if (req->out->type & VIRTIO_BLK_T_FLUSH) {
 357         virtio_blk_handle_flush(mrb->blkreq, &mrb->num_writes,
 358             req, &mrb->old_bs);
 359     } else if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
 360         virtio_blk_handle_scsi(req);
 361     } else if (req->out->type & VIRTIO_BLK_T_OUT) {
 362         qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 363                                  req->elem.out_num - 1);
 364         virtio_blk_handle_write(mrb->blkreq, &mrb->num_writes,
 365             req, &mrb->old_bs);
 366     } else {
 367         qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 368                                  req->elem.in_num - 1);
 369         virtio_blk_handle_read(req);
 370     }
 371 }
 372
 373 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 374 {
 375     VirtIOBlock *s = to_virtio_blk(vdev);
 376     VirtIOBlockReq *req;
 377     MultiReqBuffer mrb = {
 378         .num_writes = 0,
 379         .old_bs = NULL,
 380     };
 381
 382     while ((req = virtio_blk_get_request(s))) {
 383         virtio_blk_handle_request(req, &mrb);
 384     }
 385
 386     if (mrb.num_writes > 0) {
 387         do_multiwrite(mrb.old_bs, mrb.blkreq, mrb.num_writes);
 388     }
 389
 390     /*
 391      * FIXME: Want to check for completions before returning to guest mode,
 392      * so cached reads and writes are reported as quickly as possible. But
 393      * that should be done in the generic block layer.
 394      */
 395 }
 396
 397 static void virtio_blk_dma_restart_bh(void *opaque)
 398 {
 399     VirtIOBlock *s = opaque;
 400     VirtIOBlockReq *req = s->rq;
 401     MultiReqBuffer mrb = {
 402         .num_writes = 0,
 403         .old_bs = NULL,
 404     };
 405
 406     qemu_bh_delete(s->bh);
 407     s->bh = NULL;
 408
 409     s->rq = NULL;
 410
 411     while (req) {
 412         virtio_blk_handle_request(req, &mrb);
 413         req = req->next;
 414     }
 415
 416     if (mrb.num_writes > 0) {
 417         do_multiwrite(mrb.old_bs, mrb.blkreq, mrb.num_writes);
 418     }
 419 }
 420
 421 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
 422 {
 423     VirtIOBlock *s = opaque;
 424
 425     if (!running)
 426         return;
 427
 428     if (!s->bh) {
 429         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 430         qemu_bh_schedule(s->bh);
 431     }
 432 }
 433
 434 static void virtio_blk_reset(VirtIODevice *vdev)
 435 {
 436     /*
 437      * This should cancel pending requests, but can't do nicely until there
 438      * are per-device request lists.
 439      */
 440     qemu_aio_flush();
 441 }
 442
 443 /* coalesce internal state, copy to pci i/o region 0
 444  */
 445 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 446 {
 447     VirtIOBlock *s = to_virtio_blk(vdev);
 448     struct virtio_blk_config blkcfg;
 449     uint64_t capacity;
 450     int cylinders, heads, secs;
 451
 452     bdrv_get_geometry(s->bs, &capacity);
 453     bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
 454     memset(&blkcfg, 0, sizeof(blkcfg));
 455     stq_raw(&blkcfg.capacity, capacity);
 456     stl_raw(&blkcfg.seg_max, 128 - 2);
 457     stw_raw(&blkcfg.cylinders, cylinders);
 458     blkcfg.heads = heads;
 459     blkcfg.sectors = secs;
 460     blkcfg.size_max = 0;
 461     virtio_identify_template(&blkcfg);
 462     memcpy(&blkcfg.identify[VIRTIO_BLK_ID_SN], s->serial_str,
 463         VIRTIO_BLK_ID_SN_BYTES);
 464     memcpy(config, &blkcfg, s->config_size);
 465 }
 466
 467 static uint32_t virtio_blk_get_features(VirtIODevice *vdev)
 468 {
 469     VirtIOBlock *s = to_virtio_blk(vdev);
 470     uint32_t features = 0;
 471
 472     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 473     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 474
 475     if (bdrv_enable_write_cache(s->bs))
 476         features |= (1 << VIRTIO_BLK_F_WCACHE);
 477 #ifdef __linux__
 478     features |= (1 << VIRTIO_BLK_F_SCSI);
 479 #endif
 480     if (strcmp(s->serial_str, "0"))
 481         features |= 1 << VIRTIO_BLK_F_IDENTIFY;
 482
 483     if (bdrv_is_read_only(s->bs))
 484         features |= 1 << VIRTIO_BLK_F_RO;
 485
 486     return features;
 487 }
 488
 489 static void virtio_blk_save(QEMUFile *f, void *opaque)
 490 {
 491     VirtIOBlock *s = opaque;
 492     VirtIOBlockReq *req = s->rq;
 493
 494     virtio_save(&s->vdev, f);
 495
 496     while (req) {
 497         qemu_put_sbyte(f, 1);
 498         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 499         req = req->next;
 500     }
 501     qemu_put_sbyte(f, 0);
 502 }
 503
 504 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 505 {
 506     VirtIOBlock *s = opaque;
 507
 508     if (version_id != 2)
 509         return -EINVAL;
 510
 511     virtio_load(&s->vdev, f);
 512     while (qemu_get_sbyte(f)) {
 513         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 514         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 515         req->next = s->rq;
 516         s->rq = req->next;
 517     }
 518
 519     return 0;
 520 }
 521
 522 VirtIODevice *virtio_blk_init(DeviceState *dev, DriveInfo *dinfo)
 523 {
 524     VirtIOBlock *s;
 525     int cylinders, heads, secs;
 526     static int virtio_blk_id;
 527     char *ps = (char *)drive_get_serial(dinfo->bdrv);
 528     size_t size = strlen(ps) ? sizeof(struct virtio_blk_config) :
 529             offsetof(struct virtio_blk_config, _blk_size);
 530
 531     s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
 532                                           size,
 533                                           sizeof(VirtIOBlock));
 534
 535     s->config_size = size;
 536     s->vdev.get_config = virtio_blk_update_config;
 537     s->vdev.get_features = virtio_blk_get_features;
 538     s->vdev.reset = virtio_blk_reset;
 539     s->bs = dinfo->bdrv;
 540     s->rq = NULL;
 541     if (strlen(ps))
 542         strncpy(s->serial_str, ps, sizeof(s->serial_str));
 543     else
 544         snprintf(s->serial_str, sizeof(s->serial_str), "0");
 545     bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
 546     bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
 547
 548     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 549
 550     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 551     register_savevm("virtio-blk", virtio_blk_id++, 2,
 552                     virtio_blk_save, virtio_blk_load, s);
 553
 554     return &s->vdev;
 555 }