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