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 "virtio-blk.h"
  16 #ifdef __linux__
  17 # include <scsi/sg.h>
  18 #endif
  19
  20 typedef struct VirtIOBlock
  21 {
  22     VirtIODevice vdev;
  23     BlockDriverState *bs;
  24     VirtQueue *vq;
  25     void *rq;
  26     QEMUBH *bh;
  27     BlockConf *conf;
  28     unsigned short sector_mask;
  29 } VirtIOBlock;
  30
  31 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
  32 {
  33     return (VirtIOBlock *)vdev;
  34 }
  35
  36 typedef struct VirtIOBlockReq
  37 {
  38     VirtIOBlock *dev;
  39     VirtQueueElement elem;
  40     struct virtio_blk_inhdr *in;
  41     struct virtio_blk_outhdr *out;
  42     struct virtio_scsi_inhdr *scsi;
  43     QEMUIOVector qiov;
  44     struct VirtIOBlockReq *next;
  45 } VirtIOBlockReq;
  46
  47 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
  48 {
  49     VirtIOBlock *s = req->dev;
  50
  51     req->in->status = status;
  52     virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
  53     virtio_notify(&s->vdev, s->vq);
  54
  55     qemu_free(req);
  56 }
  57
  58 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
  59     int is_read)
  60 {
  61     BlockInterfaceErrorAction action =
  62         drive_get_on_error(req->dev->bs, is_read);
  63     VirtIOBlock *s = req->dev;
  64
  65     if (action == BLOCK_ERR_IGNORE) {
  66         bdrv_mon_event(s->bs, BDRV_ACTION_IGNORE, is_read);
  67         return 0;
  68     }
  69
  70     if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
  71             || action == BLOCK_ERR_STOP_ANY) {
  72         req->next = s->rq;
  73         s->rq = req;
  74         bdrv_mon_event(s->bs, BDRV_ACTION_STOP, is_read);
  75         vm_stop(0);
  76     } else {
  77         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
  78         bdrv_mon_event(s->bs, BDRV_ACTION_REPORT, is_read);
  79     }
  80
  81     return 1;
  82 }
  83
  84 static void virtio_blk_rw_complete(void *opaque, int ret)
  85 {
  86     VirtIOBlockReq *req = opaque;
  87
  88     if (ret) {
  89         int is_read = !(req->out->type & VIRTIO_BLK_T_OUT);
  90         if (virtio_blk_handle_rw_error(req, -ret, is_read))
  91             return;
  92     }
  93
  94     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
  95 }
  96
  97 static void virtio_blk_flush_complete(void *opaque, int ret)
  98 {
  99     VirtIOBlockReq *req = opaque;
 100
 101     virtio_blk_req_complete(req, ret ? VIRTIO_BLK_S_IOERR : VIRTIO_BLK_S_OK);
 102 }
 103
 104 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
 105 {
 106     VirtIOBlockReq *req = qemu_malloc(sizeof(*req));
 107     req->dev = s;
 108     req->qiov.size = 0;
 109     req->next = NULL;
 110     return req;
 111 }
 112
 113 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 114 {
 115     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 116
 117     if (req != NULL) {
 118         if (!virtqueue_pop(s->vq, &req->elem)) {
 119             qemu_free(req);
 120             return NULL;
 121         }
 122     }
 123
 124     return req;
 125 }
 126
 127 #ifdef __linux__
 128 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 129 {
 130     struct sg_io_hdr hdr;
 131     int ret;
 132     int status;
 133     int i;
 134
 135     /*
 136      * We require at least one output segment each for the virtio_blk_outhdr
 137      * and the SCSI command block.
 138      *
 139      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 140      * and the sense buffer pointer in the input segments.
 141      */
 142     if (req->elem.out_num < 2 || req->elem.in_num < 3) {
 143         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 144         return;
 145     }
 146
 147     /*
 148      * No support for bidirection commands yet.
 149      */
 150     if (req->elem.out_num > 2 && req->elem.in_num > 3) {
 151         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 152         return;
 153     }
 154
 155     /*
 156      * The scsi inhdr is placed in the second-to-last input segment, just
 157      * before the regular inhdr.
 158      */
 159     req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
 160
 161     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 162     hdr.interface_id = 'S';
 163     hdr.cmd_len = req->elem.out_sg[1].iov_len;
 164     hdr.cmdp = req->elem.out_sg[1].iov_base;
 165     hdr.dxfer_len = 0;
 166
 167     if (req->elem.out_num > 2) {
 168         /*
 169          * If there are more than the minimally required 2 output segments
 170          * there is write payload starting from the third iovec.
 171          */
 172         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 173         hdr.iovec_count = req->elem.out_num - 2;
 174
 175         for (i = 0; i < hdr.iovec_count; i++)
 176             hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
 177
 178         hdr.dxferp = req->elem.out_sg + 2;
 179
 180     } else if (req->elem.in_num > 3) {
 181         /*
 182          * If we have more than 3 input segments the guest wants to actually
 183          * read data.
 184          */
 185         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 186         hdr.iovec_count = req->elem.in_num - 3;
 187         for (i = 0; i < hdr.iovec_count; i++)
 188             hdr.dxfer_len += req->elem.in_sg[i].iov_len;
 189
 190         hdr.dxferp = req->elem.in_sg;
 191     } else {
 192         /*
 193          * Some SCSI commands don't actually transfer any data.
 194          */
 195         hdr.dxfer_direction = SG_DXFER_NONE;
 196     }
 197
 198     hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
 199     hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
 200
 201     ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
 202     if (ret) {
 203         status = VIRTIO_BLK_S_UNSUPP;
 204         hdr.status = ret;
 205         hdr.resid = hdr.dxfer_len;
 206     } else if (hdr.status) {
 207         status = VIRTIO_BLK_S_IOERR;
 208     } else {
 209         status = VIRTIO_BLK_S_OK;
 210     }
 211
 212     req->scsi->errors = hdr.status;
 213     req->scsi->residual = hdr.resid;
 214     req->scsi->sense_len = hdr.sb_len_wr;
 215     req->scsi->data_len = hdr.dxfer_len;
 216
 217     virtio_blk_req_complete(req, status);
 218 }
 219 #else
 220 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 221 {
 222     virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 223 }
 224 #endif /* __linux__ */
 225
 226 typedef struct MultiReqBuffer {
 227     BlockRequest        blkreq[32];
 228     unsigned int        num_writes;
 229 } MultiReqBuffer;
 230
 231 static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
 232 {
 233     int i, ret;
 234
 235     if (!mrb->num_writes) {
 236         return;
 237     }
 238
 239     ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
 240     if (ret != 0) {
 241         for (i = 0; i < mrb->num_writes; i++) {
 242             if (mrb->blkreq[i].error) {
 243                 virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
 244             }
 245         }
 246     }
 247
 248     mrb->num_writes = 0;
 249 }
 250
 251 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 252 {
 253     BlockDriverAIOCB *acb;
 254
 255     /*
 256      * Make sure all outstanding writes are posted to the backing device.
 257      */
 258     virtio_submit_multiwrite(req->dev->bs, mrb);
 259
 260     acb = bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 261     if (!acb) {
 262         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 263     }
 264 }
 265
 266 static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 267 {
 268     BlockRequest *blkreq;
 269
 270     if (req->out->sector & req->dev->sector_mask) {
 271         virtio_blk_rw_complete(req, -EIO);
 272         return;
 273     }
 274
 275     if (mrb->num_writes == 32) {
 276         virtio_submit_multiwrite(req->dev->bs, mrb);
 277     }
 278
 279     blkreq = &mrb->blkreq[mrb->num_writes];
 280     blkreq->sector = req->out->sector;
 281     blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
 282     blkreq->qiov = &req->qiov;
 283     blkreq->cb = virtio_blk_rw_complete;
 284     blkreq->opaque = req;
 285     blkreq->error = 0;
 286
 287     mrb->num_writes++;
 288 }
 289
 290 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 291 {
 292     BlockDriverAIOCB *acb;
 293
 294     if (req->out->sector & req->dev->sector_mask) {
 295         virtio_blk_rw_complete(req, -EIO);
 296         return;
 297     }
 298
 299     acb = bdrv_aio_readv(req->dev->bs, req->out->sector, &req->qiov,
 300                          req->qiov.size / BDRV_SECTOR_SIZE,
 301                          virtio_blk_rw_complete, req);
 302     if (!acb) {
 303         virtio_blk_rw_complete(req, -EIO);
 304     }
 305 }
 306
 307 static void virtio_blk_handle_request(VirtIOBlockReq *req,
 308     MultiReqBuffer *mrb)
 309 {
 310     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 311         fprintf(stderr, "virtio-blk missing headers\n");
 312         exit(1);
 313     }
 314
 315     if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 316         req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 317         fprintf(stderr, "virtio-blk header not in correct element\n");
 318         exit(1);
 319     }
 320
 321     req->out = (void *)req->elem.out_sg[0].iov_base;
 322     req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 323
 324     if (req->out->type & VIRTIO_BLK_T_FLUSH) {
 325         virtio_blk_handle_flush(req, mrb);
 326     } else if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
 327         virtio_blk_handle_scsi(req);
 328     } else if (req->out->type & VIRTIO_BLK_T_OUT) {
 329         qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 330                                  req->elem.out_num - 1);
 331         virtio_blk_handle_write(req, mrb);
 332     } else {
 333         qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 334                                  req->elem.in_num - 1);
 335         virtio_blk_handle_read(req);
 336     }
 337 }
 338
 339 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 340 {
 341     VirtIOBlock *s = to_virtio_blk(vdev);
 342     VirtIOBlockReq *req;
 343     MultiReqBuffer mrb = {
 344         .num_writes = 0,
 345     };
 346
 347     while ((req = virtio_blk_get_request(s))) {
 348         virtio_blk_handle_request(req, &mrb);
 349     }
 350
 351     virtio_submit_multiwrite(s->bs, &mrb);
 352
 353     /*
 354      * FIXME: Want to check for completions before returning to guest mode,
 355      * so cached reads and writes are reported as quickly as possible. But
 356      * that should be done in the generic block layer.
 357      */
 358 }
 359
 360 static void virtio_blk_dma_restart_bh(void *opaque)
 361 {
 362     VirtIOBlock *s = opaque;
 363     VirtIOBlockReq *req = s->rq;
 364     MultiReqBuffer mrb = {
 365         .num_writes = 0,
 366     };
 367
 368     qemu_bh_delete(s->bh);
 369     s->bh = NULL;
 370
 371     s->rq = NULL;
 372
 373     while (req) {
 374         virtio_blk_handle_request(req, &mrb);
 375         req = req->next;
 376     }
 377
 378     virtio_submit_multiwrite(s->bs, &mrb);
 379 }
 380
 381 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
 382 {
 383     VirtIOBlock *s = opaque;
 384
 385     if (!running)
 386         return;
 387
 388     if (!s->bh) {
 389         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 390         qemu_bh_schedule(s->bh);
 391     }
 392 }
 393
 394 static void virtio_blk_reset(VirtIODevice *vdev)
 395 {
 396     /*
 397      * This should cancel pending requests, but can't do nicely until there
 398      * are per-device request lists.
 399      */
 400     qemu_aio_flush();
 401 }
 402
 403 /* coalesce internal state, copy to pci i/o region 0
 404  */
 405 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 406 {
 407     VirtIOBlock *s = to_virtio_blk(vdev);
 408     struct virtio_blk_config blkcfg;
 409     uint64_t capacity;
 410     int cylinders, heads, secs;
 411
 412     bdrv_get_geometry(s->bs, &capacity);
 413     bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
 414     memset(&blkcfg, 0, sizeof(blkcfg));
 415     stq_raw(&blkcfg.capacity, capacity);
 416     stl_raw(&blkcfg.seg_max, 128 - 2);
 417     stw_raw(&blkcfg.cylinders, cylinders);
 418     blkcfg.heads = heads;
 419     blkcfg.sectors = secs & ~s->sector_mask;
 420     blkcfg.blk_size = s->conf->logical_block_size;
 421     blkcfg.size_max = 0;
 422     blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
 423     blkcfg.alignment_offset = 0;
 424     blkcfg.min_io_size = s->conf->min_io_size / blkcfg.blk_size;
 425     blkcfg.opt_io_size = s->conf->opt_io_size / blkcfg.blk_size;
 426     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
 427 }
 428
 429 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 430 {
 431     VirtIOBlock *s = to_virtio_blk(vdev);
 432
 433     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 434     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 435     features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
 436     features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
 437
 438     if (bdrv_enable_write_cache(s->bs))
 439         features |= (1 << VIRTIO_BLK_F_WCACHE);
 440
 441     if (bdrv_is_read_only(s->bs))
 442         features |= 1 << VIRTIO_BLK_F_RO;
 443
 444     return features;
 445 }
 446
 447 static void virtio_blk_save(QEMUFile *f, void *opaque)
 448 {
 449     VirtIOBlock *s = opaque;
 450     VirtIOBlockReq *req = s->rq;
 451
 452     virtio_save(&s->vdev, f);
 453
 454     while (req) {
 455         qemu_put_sbyte(f, 1);
 456         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 457         req = req->next;
 458     }
 459     qemu_put_sbyte(f, 0);
 460 }
 461
 462 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 463 {
 464     VirtIOBlock *s = opaque;
 465
 466     if (version_id != 2)
 467         return -EINVAL;
 468
 469     virtio_load(&s->vdev, f);
 470     while (qemu_get_sbyte(f)) {
 471         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 472         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 473         req->next = s->rq;
 474         s->rq = req->next;
 475     }
 476
 477     return 0;
 478 }
 479
 480 VirtIODevice *virtio_blk_init(DeviceState *dev, BlockConf *conf)
 481 {
 482     VirtIOBlock *s;
 483     int cylinders, heads, secs;
 484     static int virtio_blk_id;
 485
 486     s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
 487                                           sizeof(struct virtio_blk_config),
 488                                           sizeof(VirtIOBlock));
 489
 490     s->vdev.get_config = virtio_blk_update_config;
 491     s->vdev.get_features = virtio_blk_get_features;
 492     s->vdev.reset = virtio_blk_reset;
 493     s->bs = conf->dinfo->bdrv;
 494     s->conf = conf;
 495     s->rq = NULL;
 496     s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
 497     bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
 498
 499     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 500
 501     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 502     register_savevm("virtio-blk", virtio_blk_id++, 2,
 503                     virtio_blk_save, virtio_blk_load, s);
 504
 505     return &s->vdev;
 506 }