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