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;
 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
 198     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 199     hdr.interface_id = 'S';
 200     hdr.cmd_len = req->elem.out_sg[1].iov_len;
 201     hdr.cmdp = req->elem.out_sg[1].iov_base;
 202     hdr.dxfer_len = 0;
 203
 204     if (req->elem.out_num > 2) {
 205         /*
 206          * If there are more than the minimally required 2 output segments
 207          * there is write payload starting from the third iovec.
 208          */
 209         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 210         hdr.iovec_count = req->elem.out_num - 2;
 211
 212         for (i = 0; i < hdr.iovec_count; i++)
 213             hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
 214
 215         hdr.dxferp = req->elem.out_sg + 2;
 216
 217     } else if (req->elem.in_num > 3) {
 218         /*
 219          * If we have more than 3 input segments the guest wants to actually
 220          * read data.
 221          */
 222         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 223         hdr.iovec_count = req->elem.in_num - 3;
 224         for (i = 0; i < hdr.iovec_count; i++)
 225             hdr.dxfer_len += req->elem.in_sg[i].iov_len;
 226
 227         hdr.dxferp = req->elem.in_sg;
 228     } else {
 229         /*
 230          * Some SCSI commands don't actually transfer any data.
 231          */
 232         hdr.dxfer_direction = SG_DXFER_NONE;
 233     }
 234
 235     hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
 236     hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_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_rw_complete(blkreq[i].opaque, -EIO);
 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_rw_complete(req, -EIO);
 317     }
 318 }
 319
 320 typedef struct MultiReqBuffer {
 321     BlockRequest        blkreq[32];
 322     int                 num_writes;
 323     BlockDriverState    *old_bs;
 324 } MultiReqBuffer;
 325
 326 static void virtio_blk_handle_request(VirtIOBlockReq *req,
 327     MultiReqBuffer *mrb)
 328 {
 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(mrb->blkreq, &mrb->num_writes,
 351             req, &mrb->old_bs);
 352     } else {
 353         qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 354                                  req->elem.in_num - 1);
 355         virtio_blk_handle_read(req);
 356     }
 357 }
 358
 359 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 360 {
 361     VirtIOBlock *s = to_virtio_blk(vdev);
 362     VirtIOBlockReq *req;
 363     MultiReqBuffer mrb = {
 364         .num_writes = 0,
 365         .old_bs = NULL,
 366     };
 367
 368     while ((req = virtio_blk_get_request(s))) {
 369         virtio_blk_handle_request(req, &mrb);
 370     }
 371
 372     if (mrb.num_writes > 0) {
 373         do_multiwrite(mrb.old_bs, mrb.blkreq, mrb.num_writes);
 374     }
 375
 376     /*
 377      * FIXME: Want to check for completions before returning to guest mode,
 378      * so cached reads and writes are reported as quickly as possible. But
 379      * that should be done in the generic block layer.
 380      */
 381 }
 382
 383 static void virtio_blk_dma_restart_bh(void *opaque)
 384 {
 385     VirtIOBlock *s = opaque;
 386     VirtIOBlockReq *req = s->rq;
 387     MultiReqBuffer mrb = {
 388         .num_writes = 0,
 389         .old_bs = NULL,
 390     };
 391
 392     qemu_bh_delete(s->bh);
 393     s->bh = NULL;
 394
 395     s->rq = NULL;
 396
 397     while (req) {
 398         virtio_blk_handle_request(req, &mrb);
 399         req = req->next;
 400     }
 401
 402     if (mrb.num_writes > 0) {
 403         do_multiwrite(mrb.old_bs, mrb.blkreq, mrb.num_writes);
 404     }
 405 }
 406
 407 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
 408 {
 409     VirtIOBlock *s = opaque;
 410
 411     if (!running)
 412         return;
 413
 414     if (!s->bh) {
 415         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 416         qemu_bh_schedule(s->bh);
 417     }
 418 }
 419
 420 static void virtio_blk_reset(VirtIODevice *vdev)
 421 {
 422     /*
 423      * This should cancel pending requests, but can't do nicely until there
 424      * are per-device request lists.
 425      */
 426     qemu_aio_flush();
 427 }
 428
 429 /* coalesce internal state, copy to pci i/o region 0
 430  */
 431 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 432 {
 433     VirtIOBlock *s = to_virtio_blk(vdev);
 434     struct virtio_blk_config blkcfg;
 435     uint64_t capacity;
 436     int cylinders, heads, secs;
 437
 438     bdrv_get_geometry(s->bs, &capacity);
 439     bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
 440     memset(&blkcfg, 0, sizeof(blkcfg));
 441     stq_raw(&blkcfg.capacity, capacity);
 442     stl_raw(&blkcfg.seg_max, 128 - 2);
 443     stw_raw(&blkcfg.cylinders, cylinders);
 444     blkcfg.heads = heads;
 445     blkcfg.sectors = secs;
 446     blkcfg.size_max = 0;
 447     virtio_identify_template(&blkcfg);
 448     memcpy(&blkcfg.identify[VIRTIO_BLK_ID_SN], s->serial_str,
 449         VIRTIO_BLK_ID_SN_BYTES);
 450     memcpy(config, &blkcfg, s->config_size);
 451 }
 452
 453 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 454 {
 455     VirtIOBlock *s = to_virtio_blk(vdev);
 456
 457     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 458     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 459
 460     if (bdrv_enable_write_cache(s->bs))
 461         features |= (1 << VIRTIO_BLK_F_WCACHE);
 462     if (strcmp(s->serial_str, "0"))
 463         features |= 1 << VIRTIO_BLK_F_IDENTIFY;
 464
 465     if (bdrv_is_read_only(s->bs))
 466         features |= 1 << VIRTIO_BLK_F_RO;
 467
 468     return features;
 469 }
 470
 471 static void virtio_blk_save(QEMUFile *f, void *opaque)
 472 {
 473     VirtIOBlock *s = opaque;
 474     VirtIOBlockReq *req = s->rq;
 475
 476     virtio_save(&s->vdev, f);
 477
 478     while (req) {
 479         qemu_put_sbyte(f, 1);
 480         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 481         req = req->next;
 482     }
 483     qemu_put_sbyte(f, 0);
 484 }
 485
 486 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 487 {
 488     VirtIOBlock *s = opaque;
 489
 490     if (version_id != 2)
 491         return -EINVAL;
 492
 493     virtio_load(&s->vdev, f);
 494     while (qemu_get_sbyte(f)) {
 495         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 496         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 497         req->next = s->rq;
 498         s->rq = req->next;
 499     }
 500
 501     return 0;
 502 }
 503
 504 VirtIODevice *virtio_blk_init(DeviceState *dev, DriveInfo *dinfo)
 505 {
 506     VirtIOBlock *s;
 507     int cylinders, heads, secs;
 508     static int virtio_blk_id;
 509     char *ps = (char *)drive_get_serial(dinfo->bdrv);
 510     size_t size = strlen(ps) ? sizeof(struct virtio_blk_config) :
 511             offsetof(struct virtio_blk_config, _blk_size);
 512
 513     s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
 514                                           size,
 515                                           sizeof(VirtIOBlock));
 516
 517     s->config_size = size;
 518     s->vdev.get_config = virtio_blk_update_config;
 519     s->vdev.get_features = virtio_blk_get_features;
 520     s->vdev.reset = virtio_blk_reset;
 521     s->bs = dinfo->bdrv;
 522     s->rq = NULL;
 523     if (strlen(ps))
 524         strncpy(s->serial_str, ps, sizeof(s->serial_str));
 525     else
 526         snprintf(s->serial_str, sizeof(s->serial_str), "0");
 527     bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
 528     bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
 529
 530     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 531
 532     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 533     register_savevm("virtio-blk", virtio_blk_id++, 2,
 534                     virtio_blk_save, virtio_blk_load, s);
 535
 536     return &s->vdev;
 537 }