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 } VirtIOBlock;
  31
  32 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
  33 {
  34     return (VirtIOBlock *)vdev;
  35 }
  36
  37 /* store identify data in little endian format
  38  */
  39 static inline void put_le16(uint16_t *p, unsigned int v)
  40 {
  41     *p = cpu_to_le16(v);
  42 }
  43
  44 /* copy to *dst from *src, nul pad dst tail as needed to len bytes
  45  */
  46 static inline void padstr(char *dst, const char *src, int len)
  47 {
  48     while (len--)
  49         *dst++ = *src ? *src++ : '\0';
  50 }
  51
  52 /* setup simulated identify data as appropriate for virtio block device
  53  *
  54  * ref: AT Attachment 8 - ATA/ATAPI Command Set (ATA8-ACS)
  55  */
  56 static inline void virtio_identify_template(struct virtio_blk_config *bc)
  57 {
  58     uint16_t *p = &bc->identify[0];
  59     uint64_t lba_sectors = bc->capacity;
  60
  61     memset(p, 0, sizeof(bc->identify));
  62     put_le16(p + 0, 0x0);                            /* ATA device */
  63     padstr((char *)(p + 23), QEMU_VERSION, 8);       /* firmware revision */
  64     padstr((char *)(p + 27), "QEMU VIRT_BLK", 40);   /* model# */
  65     put_le16(p + 47, 0x80ff);                        /* max xfer 255 sectors */
  66     put_le16(p + 49, 0x0b00);                        /* support IORDY/LBA/DMA */
  67     put_le16(p + 59, 0x1ff);                         /* cur xfer 255 sectors */
  68     put_le16(p + 80, 0x1f0);                         /* support ATA8/7/6/5/4 */
  69     put_le16(p + 81, 0x16);
  70     put_le16(p + 82, 0x400);
  71     put_le16(p + 83, 0x400);
  72     put_le16(p + 100, lba_sectors);
  73     put_le16(p + 101, lba_sectors >> 16);
  74     put_le16(p + 102, lba_sectors >> 32);
  75     put_le16(p + 103, lba_sectors >> 48);
  76 }
  77
  78 typedef struct VirtIOBlockReq
  79 {
  80     VirtIOBlock *dev;
  81     VirtQueueElement elem;
  82     struct virtio_blk_inhdr *in;
  83     struct virtio_blk_outhdr *out;
  84     struct virtio_scsi_inhdr *scsi;
  85     QEMUIOVector qiov;
  86     struct VirtIOBlockReq *next;
  87 } VirtIOBlockReq;
  88
  89 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
  90 {
  91     VirtIOBlock *s = req->dev;
  92
  93     req->in->status = status;
  94     virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
  95     virtio_notify(&s->vdev, s->vq);
  96
  97     qemu_free(req);
  98 }
  99
 100 static int virtio_blk_handle_write_error(VirtIOBlockReq *req, int error)
 101 {
 102     BlockInterfaceErrorAction action = drive_get_onerror(req->dev->bs);
 103     VirtIOBlock *s = req->dev;
 104
 105     if (action == BLOCK_ERR_IGNORE)
 106         return 0;
 107
 108     if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
 109             || action == BLOCK_ERR_STOP_ANY) {
 110         req->next = s->rq;
 111         s->rq = req;
 112         vm_stop(0);
 113     } else {
 114         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 115     }
 116
 117     return 1;
 118 }
 119
 120 static void virtio_blk_rw_complete(void *opaque, int ret)
 121 {
 122     VirtIOBlockReq *req = opaque;
 123
 124     if (ret && (req->out->type & VIRTIO_BLK_T_OUT)) {
 125         if (virtio_blk_handle_write_error(req, -ret))
 126             return;
 127     }
 128
 129     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 130 }
 131
 132 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
 133 {
 134     VirtIOBlockReq *req = qemu_mallocz(sizeof(*req));
 135     req->dev = s;
 136     return req;
 137 }
 138
 139 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 140 {
 141     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 142
 143     if (req != NULL) {
 144         if (!virtqueue_pop(s->vq, &req->elem)) {
 145             qemu_free(req);
 146             return NULL;
 147         }
 148     }
 149
 150     return req;
 151 }
 152
 153 #ifdef __linux__
 154 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 155 {
 156     struct sg_io_hdr hdr;
 157     int ret, size = 0;
 158     int status;
 159     int i;
 160
 161     /*
 162      * We require at least one output segment each for the virtio_blk_outhdr
 163      * and the SCSI command block.
 164      *
 165      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 166      * and the sense buffer pointer in the input segments.
 167      */
 168     if (req->elem.out_num < 2 || req->elem.in_num < 3) {
 169         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 170         return;
 171     }
 172
 173     /*
 174      * No support for bidirection commands yet.
 175      */
 176     if (req->elem.out_num > 2 && req->elem.in_num > 3) {
 177         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 178         return;
 179     }
 180
 181     /*
 182      * The scsi inhdr is placed in the second-to-last input segment, just
 183      * before the regular inhdr.
 184      */
 185     req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
 186     size = sizeof(*req->in) + sizeof(*req->scsi);
 187
 188     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 189     hdr.interface_id = 'S';
 190     hdr.cmd_len = req->elem.out_sg[1].iov_len;
 191     hdr.cmdp = req->elem.out_sg[1].iov_base;
 192     hdr.dxfer_len = 0;
 193
 194     if (req->elem.out_num > 2) {
 195         /*
 196          * If there are more than the minimally required 2 output segments
 197          * there is write payload starting from the third iovec.
 198          */
 199         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 200         hdr.iovec_count = req->elem.out_num - 2;
 201
 202         for (i = 0; i < hdr.iovec_count; i++)
 203             hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
 204
 205         hdr.dxferp = req->elem.out_sg + 2;
 206
 207     } else if (req->elem.in_num > 3) {
 208         /*
 209          * If we have more than 3 input segments the guest wants to actually
 210          * read data.
 211          */
 212         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 213         hdr.iovec_count = req->elem.in_num - 3;
 214         for (i = 0; i < hdr.iovec_count; i++)
 215             hdr.dxfer_len += req->elem.in_sg[i].iov_len;
 216
 217         hdr.dxferp = req->elem.in_sg;
 218         size += hdr.dxfer_len;
 219     } else {
 220         /*
 221          * Some SCSI commands don't actually transfer any data.
 222          */
 223         hdr.dxfer_direction = SG_DXFER_NONE;
 224     }
 225
 226     hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
 227     hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
 228     size += hdr.mx_sb_len;
 229
 230     ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
 231     if (ret) {
 232         status = VIRTIO_BLK_S_UNSUPP;
 233         hdr.status = ret;
 234         hdr.resid = hdr.dxfer_len;
 235     } else if (hdr.status) {
 236         status = VIRTIO_BLK_S_IOERR;
 237     } else {
 238         status = VIRTIO_BLK_S_OK;
 239     }
 240
 241     req->scsi->errors = hdr.status;
 242     req->scsi->residual = hdr.resid;
 243     req->scsi->sense_len = hdr.sb_len_wr;
 244     req->scsi->data_len = hdr.dxfer_len;
 245
 246     virtio_blk_req_complete(req, status);
 247 }
 248 #else
 249 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 250 {
 251     virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 252 }
 253 #endif /* __linux__ */
 254
 255 static void do_multiwrite(BlockDriverState *bs, BlockRequest *blkreq,
 256     int num_writes)
 257 {
 258     int i, ret;
 259     ret = bdrv_aio_multiwrite(bs, blkreq, num_writes);
 260
 261     if (ret != 0) {
 262         for (i = 0; i < num_writes; i++) {
 263             if (blkreq[i].error) {
 264                 virtio_blk_req_complete(blkreq[i].opaque, VIRTIO_BLK_S_IOERR);
 265             }
 266         }
 267     }
 268 }
 269
 270 static void virtio_blk_handle_write(BlockRequest *blkreq, int *num_writes,
 271     VirtIOBlockReq *req, BlockDriverState **old_bs)
 272 {
 273     if (req->dev->bs != *old_bs || *num_writes == 32) {
 274         if (*old_bs != NULL) {
 275             do_multiwrite(*old_bs, blkreq, *num_writes);
 276         }
 277         *num_writes = 0;
 278         *old_bs = req->dev->bs;
 279     }
 280
 281     blkreq[*num_writes].sector = req->out->sector;
 282     blkreq[*num_writes].nb_sectors = req->qiov.size / 512;
 283     blkreq[*num_writes].qiov = &req->qiov;
 284     blkreq[*num_writes].cb = virtio_blk_rw_complete;
 285     blkreq[*num_writes].opaque = req;
 286     blkreq[*num_writes].error = 0;
 287
 288     (*num_writes)++;
 289 }
 290
 291 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 292 {
 293     BlockDriverAIOCB *acb;
 294
 295     acb = bdrv_aio_readv(req->dev->bs, req->out->sector, &req->qiov,
 296                          req->qiov.size / 512, virtio_blk_rw_complete, req);
 297     if (!acb) {
 298         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 299     }
 300 }
 301
 302 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 303 {
 304     VirtIOBlock *s = to_virtio_blk(vdev);
 305     VirtIOBlockReq *req;
 306     BlockRequest blkreq[32];
 307     int num_writes = 0;
 308     BlockDriverState *old_bs = NULL;
 309
 310     while ((req = virtio_blk_get_request(s))) {
 311         if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 312             fprintf(stderr, "virtio-blk missing headers\n");
 313             exit(1);
 314         }
 315
 316         if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 317             req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 318             fprintf(stderr, "virtio-blk header not in correct element\n");
 319             exit(1);
 320         }
 321
 322         req->out = (void *)req->elem.out_sg[0].iov_base;
 323         req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 324
 325         if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
 326             virtio_blk_handle_scsi(req);
 327         } else if (req->out->type & VIRTIO_BLK_T_OUT) {
 328             qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 329                                      req->elem.out_num - 1);
 330             virtio_blk_handle_write(blkreq, &num_writes, req, &old_bs);
 331         } else {
 332             qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 333                                      req->elem.in_num - 1);
 334             virtio_blk_handle_read(req);
 335         }
 336     }
 337
 338     if (num_writes > 0) {
 339         do_multiwrite(old_bs, blkreq, num_writes);
 340     }
 341
 342     /*
 343      * FIXME: Want to check for completions before returning to guest mode,
 344      * so cached reads and writes are reported as quickly as possible. But
 345      * that should be done in the generic block layer.
 346      */
 347 }
 348
 349 static void virtio_blk_dma_restart_bh(void *opaque)
 350 {
 351     VirtIOBlock *s = opaque;
 352     VirtIOBlockReq *req = s->rq;
 353
 354     qemu_bh_delete(s->bh);
 355     s->bh = NULL;
 356
 357     s->rq = NULL;
 358
 359     while (req) {
 360         bdrv_aio_writev(req->dev->bs, req->out->sector, &req->qiov,
 361             req->qiov.size / 512, virtio_blk_rw_complete, req);
 362         req = req->next;
 363     }
 364 }
 365
 366 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
 367 {
 368     VirtIOBlock *s = opaque;
 369
 370     if (!running)
 371         return;
 372
 373     if (!s->bh) {
 374         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 375         qemu_bh_schedule(s->bh);
 376     }
 377 }
 378
 379 static void virtio_blk_reset(VirtIODevice *vdev)
 380 {
 381     /*
 382      * This should cancel pending requests, but can't do nicely until there
 383      * are per-device request lists.
 384      */
 385     qemu_aio_flush();
 386 }
 387
 388 /* coalesce internal state, copy to pci i/o region 0
 389  */
 390 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 391 {
 392     VirtIOBlock *s = to_virtio_blk(vdev);
 393     struct virtio_blk_config blkcfg;
 394     uint64_t capacity;
 395     int cylinders, heads, secs;
 396
 397     bdrv_get_geometry(s->bs, &capacity);
 398     bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
 399     memset(&blkcfg, 0, sizeof(blkcfg));
 400     stq_raw(&blkcfg.capacity, capacity);
 401     stl_raw(&blkcfg.seg_max, 128 - 2);
 402     stw_raw(&blkcfg.cylinders, cylinders);
 403     blkcfg.heads = heads;
 404     blkcfg.sectors = secs;
 405     blkcfg.size_max = 0;
 406     virtio_identify_template(&blkcfg);
 407     memcpy(&blkcfg.identify[VIRTIO_BLK_ID_SN], s->serial_str,
 408         VIRTIO_BLK_ID_SN_BYTES);
 409     memcpy(config, &blkcfg, sizeof(blkcfg));
 410 }
 411
 412 static uint32_t virtio_blk_get_features(VirtIODevice *vdev)
 413 {
 414     VirtIOBlock *s = to_virtio_blk(vdev);
 415     uint32_t features = 0;
 416
 417     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 418     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 419 #ifdef __linux__
 420     features |= (1 << VIRTIO_BLK_F_SCSI);
 421 #endif
 422     if (strcmp(s->serial_str, "0"))
 423         features |= 1 << VIRTIO_BLK_F_IDENTIFY;
 424
 425     return features;
 426 }
 427
 428 static void virtio_blk_save(QEMUFile *f, void *opaque)
 429 {
 430     VirtIOBlock *s = opaque;
 431     VirtIOBlockReq *req = s->rq;
 432
 433     virtio_save(&s->vdev, f);
 434
 435     while (req) {
 436         qemu_put_sbyte(f, 1);
 437         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 438         req = req->next;
 439     }
 440     qemu_put_sbyte(f, 0);
 441 }
 442
 443 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 444 {
 445     VirtIOBlock *s = opaque;
 446
 447     if (version_id != 2)
 448         return -EINVAL;
 449
 450     virtio_load(&s->vdev, f);
 451     while (qemu_get_sbyte(f)) {
 452         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 453         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 454         req->next = s->rq;
 455         s->rq = req->next;
 456     }
 457
 458     return 0;
 459 }
 460
 461 VirtIODevice *virtio_blk_init(DeviceState *dev, DriveInfo *dinfo)
 462 {
 463     VirtIOBlock *s;
 464     int cylinders, heads, secs;
 465     static int virtio_blk_id;
 466     char *ps;
 467
 468     s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
 469                                           sizeof(struct virtio_blk_config),
 470                                           sizeof(VirtIOBlock));
 471
 472     s->vdev.get_config = virtio_blk_update_config;
 473     s->vdev.get_features = virtio_blk_get_features;
 474     s->vdev.reset = virtio_blk_reset;
 475     s->bs = dinfo->bdrv;
 476     s->rq = NULL;
 477     if (strlen(ps = (char *)drive_get_serial(s->bs)))
 478         strncpy(s->serial_str, ps, sizeof(s->serial_str));
 479     else
 480         snprintf(s->serial_str, sizeof(s->serial_str), "0");
 481     s->bs->private = dev;
 482     bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
 483     bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
 484
 485     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 486
 487     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 488     register_savevm("virtio-blk", virtio_blk_id++, 2,
 489                     virtio_blk_save, virtio_blk_load, s);
 490
 491     return &s->vdev;
 492 }