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_mallocz(sizeof(*req));
 109     req->dev = s;
 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 static void do_multiwrite(BlockDriverState *bs, BlockRequest *blkreq,
 227     int num_writes)
 228 {
 229     int i, ret;
 230     ret = bdrv_aio_multiwrite(bs, blkreq, num_writes);
 231
 232     if (ret != 0) {
 233         for (i = 0; i < num_writes; i++) {
 234             if (blkreq[i].error) {
 235                 virtio_blk_rw_complete(blkreq[i].opaque, -EIO);
 236             }
 237         }
 238     }
 239 }
 240
 241 static void virtio_blk_handle_flush(VirtIOBlockReq *req)
 242 {
 243     BlockDriverAIOCB *acb;
 244
 245     acb = bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 246     if (!acb) {
 247         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 248     }
 249 }
 250
 251 static void virtio_blk_handle_write(BlockRequest *blkreq, int *num_writes,
 252     VirtIOBlockReq *req, BlockDriverState **old_bs)
 253 {
 254     if (req->out->sector & req->dev->sector_mask) {
 255         virtio_blk_rw_complete(req, -EIO);
 256         return;
 257     }
 258
 259     if (req->dev->bs != *old_bs || *num_writes == 32) {
 260         if (*old_bs != NULL) {
 261             do_multiwrite(*old_bs, blkreq, *num_writes);
 262         }
 263         *num_writes = 0;
 264         *old_bs = req->dev->bs;
 265     }
 266
 267     blkreq[*num_writes].sector = req->out->sector;
 268     blkreq[*num_writes].nb_sectors = req->qiov.size / 512;
 269     blkreq[*num_writes].qiov = &req->qiov;
 270     blkreq[*num_writes].cb = virtio_blk_rw_complete;
 271     blkreq[*num_writes].opaque = req;
 272     blkreq[*num_writes].error = 0;
 273
 274     (*num_writes)++;
 275 }
 276
 277 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 278 {
 279     BlockDriverAIOCB *acb;
 280
 281     if (req->out->sector & req->dev->sector_mask) {
 282         virtio_blk_rw_complete(req, -EIO);
 283         return;
 284     }
 285
 286     acb = bdrv_aio_readv(req->dev->bs, req->out->sector, &req->qiov,
 287                          req->qiov.size / 512, virtio_blk_rw_complete, req);
 288     if (!acb) {
 289         virtio_blk_rw_complete(req, -EIO);
 290     }
 291 }
 292
 293 typedef struct MultiReqBuffer {
 294     BlockRequest        blkreq[32];
 295     int                 num_writes;
 296     BlockDriverState    *old_bs;
 297 } MultiReqBuffer;
 298
 299 static void virtio_blk_handle_request(VirtIOBlockReq *req,
 300     MultiReqBuffer *mrb)
 301 {
 302     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 303         fprintf(stderr, "virtio-blk missing headers\n");
 304         exit(1);
 305     }
 306
 307     if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 308         req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 309         fprintf(stderr, "virtio-blk header not in correct element\n");
 310         exit(1);
 311     }
 312
 313     req->out = (void *)req->elem.out_sg[0].iov_base;
 314     req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 315
 316     if (req->out->type & VIRTIO_BLK_T_FLUSH) {
 317         virtio_blk_handle_flush(req);
 318     } else if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
 319         virtio_blk_handle_scsi(req);
 320     } else if (req->out->type & VIRTIO_BLK_T_OUT) {
 321         qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 322                                  req->elem.out_num - 1);
 323         virtio_blk_handle_write(mrb->blkreq, &mrb->num_writes,
 324             req, &mrb->old_bs);
 325     } else {
 326         qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 327                                  req->elem.in_num - 1);
 328         virtio_blk_handle_read(req);
 329     }
 330 }
 331
 332 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 333 {
 334     VirtIOBlock *s = to_virtio_blk(vdev);
 335     VirtIOBlockReq *req;
 336     MultiReqBuffer mrb = {
 337         .num_writes = 0,
 338         .old_bs = NULL,
 339     };
 340
 341     while ((req = virtio_blk_get_request(s))) {
 342         virtio_blk_handle_request(req, &mrb);
 343     }
 344
 345     if (mrb.num_writes > 0) {
 346         do_multiwrite(mrb.old_bs, mrb.blkreq, mrb.num_writes);
 347     }
 348
 349     /*
 350      * FIXME: Want to check for completions before returning to guest mode,
 351      * so cached reads and writes are reported as quickly as possible. But
 352      * that should be done in the generic block layer.
 353      */
 354 }
 355
 356 static void virtio_blk_dma_restart_bh(void *opaque)
 357 {
 358     VirtIOBlock *s = opaque;
 359     VirtIOBlockReq *req = s->rq;
 360     MultiReqBuffer mrb = {
 361         .num_writes = 0,
 362         .old_bs = NULL,
 363     };
 364
 365     qemu_bh_delete(s->bh);
 366     s->bh = NULL;
 367
 368     s->rq = NULL;
 369
 370     while (req) {
 371         virtio_blk_handle_request(req, &mrb);
 372         req = req->next;
 373     }
 374
 375     if (mrb.num_writes > 0) {
 376         do_multiwrite(mrb.old_bs, mrb.blkreq, mrb.num_writes);
 377     }
 378 }
 379
 380 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
 381 {
 382     VirtIOBlock *s = opaque;
 383
 384     if (!running)
 385         return;
 386
 387     if (!s->bh) {
 388         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 389         qemu_bh_schedule(s->bh);
 390     }
 391 }
 392
 393 static void virtio_blk_reset(VirtIODevice *vdev)
 394 {
 395     /*
 396      * This should cancel pending requests, but can't do nicely until there
 397      * are per-device request lists.
 398      */
 399     qemu_aio_flush();
 400 }
 401
 402 /* coalesce internal state, copy to pci i/o region 0
 403  */
 404 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 405 {
 406     VirtIOBlock *s = to_virtio_blk(vdev);
 407     struct virtio_blk_config blkcfg;
 408     uint64_t capacity;
 409     int cylinders, heads, secs;
 410
 411     bdrv_get_geometry(s->bs, &capacity);
 412     bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
 413     memset(&blkcfg, 0, sizeof(blkcfg));
 414     stq_raw(&blkcfg.capacity, capacity);
 415     stl_raw(&blkcfg.seg_max, 128 - 2);
 416     stw_raw(&blkcfg.cylinders, cylinders);
 417     blkcfg.heads = heads;
 418     blkcfg.sectors = secs & ~s->sector_mask;
 419     blkcfg.blk_size = s->conf->logical_block_size;
 420     blkcfg.size_max = 0;
 421     blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
 422     blkcfg.alignment_offset = 0;
 423     blkcfg.min_io_size = s->conf->min_io_size / blkcfg.blk_size;
 424     blkcfg.opt_io_size = s->conf->opt_io_size / blkcfg.blk_size;
 425     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
 426 }
 427
 428 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 429 {
 430     VirtIOBlock *s = to_virtio_blk(vdev);
 431
 432     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 433     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 434     features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
 435     features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
 436
 437     if (bdrv_enable_write_cache(s->bs))
 438         features |= (1 << VIRTIO_BLK_F_WCACHE);
 439
 440     if (bdrv_is_read_only(s->bs))
 441         features |= 1 << VIRTIO_BLK_F_RO;
 442
 443     return features;
 444 }
 445
 446 static void virtio_blk_save(QEMUFile *f, void *opaque)
 447 {
 448     VirtIOBlock *s = opaque;
 449     VirtIOBlockReq *req = s->rq;
 450
 451     virtio_save(&s->vdev, f);
 452
 453     while (req) {
 454         qemu_put_sbyte(f, 1);
 455         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 456         req = req->next;
 457     }
 458     qemu_put_sbyte(f, 0);
 459 }
 460
 461 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 462 {
 463     VirtIOBlock *s = opaque;
 464
 465     if (version_id != 2)
 466         return -EINVAL;
 467
 468     virtio_load(&s->vdev, f);
 469     while (qemu_get_sbyte(f)) {
 470         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 471         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 472         req->next = s->rq;
 473         s->rq = req->next;
 474     }
 475
 476     return 0;
 477 }
 478
 479 VirtIODevice *virtio_blk_init(DeviceState *dev, BlockConf *conf)
 480 {
 481     VirtIOBlock *s;
 482     int cylinders, heads, secs;
 483     static int virtio_blk_id;
 484
 485     s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
 486                                           sizeof(struct virtio_blk_config),
 487                                           sizeof(VirtIOBlock));
 488
 489     s->vdev.get_config = virtio_blk_update_config;
 490     s->vdev.get_features = virtio_blk_get_features;
 491     s->vdev.reset = virtio_blk_reset;
 492     s->bs = conf->dinfo->bdrv;
 493     s->conf = conf;
 494     s->rq = NULL;
 495     s->sector_mask = (s->conf->logical_block_size / 512) - 1;
 496     bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
 497
 498     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 499
 500     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 501     register_savevm("virtio-blk", virtio_blk_id++, 2,
 502                     virtio_blk_save, virtio_blk_load, s);
 503
 504     return &s->vdev;
 505 }