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
  19 typedef struct VirtIOBlock
  20 {
  21     VirtIODevice vdev;
  22     BlockDriverState *bs;
  23     VirtQueue *vq;
  24     void *rq;
  25 } VirtIOBlock;
  26
  27 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
  28 {
  29     return (VirtIOBlock *)vdev;
  30 }
  31
  32 typedef struct VirtIOBlockReq
  33 {
  34     VirtIOBlock *dev;
  35     VirtQueueElement elem;
  36     struct virtio_blk_inhdr *in;
  37     struct virtio_blk_outhdr *out;
  38     size_t size;
  39     uint8_t *buffer;
  40     struct VirtIOBlockReq *next;
  41 } VirtIOBlockReq;
  42
  43 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
  44 {
  45     VirtIOBlock *s = req->dev;
  46
  47     req->in->status = status;
  48     virtqueue_push(s->vq, &req->elem, req->size + sizeof(*req->in));
  49     virtio_notify(&s->vdev, s->vq);
  50
  51     qemu_free(req->buffer);
  52     qemu_free(req);
  53 }
  54
  55 static int virtio_blk_handle_write_error(VirtIOBlockReq *req, int error)
  56 {
  57     BlockInterfaceErrorAction action = drive_get_onerror(req->dev->bs);
  58     VirtIOBlock *s = req->dev;
  59
  60     if (action == BLOCK_ERR_IGNORE)
  61         return 0;
  62
  63     if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
  64             || action == BLOCK_ERR_STOP_ANY) {
  65         req->next = s->rq;
  66         s->rq = req;
  67         vm_stop(0);
  68     } else {
  69         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
  70     }
  71
  72     return 1;
  73 }
  74
  75 static void virtio_blk_rw_complete(void *opaque, int ret)
  76 {
  77     VirtIOBlockReq *req = opaque;
  78
  79     /* Copy read data to the guest */
  80     if (!ret && !(req->out->type & VIRTIO_BLK_T_OUT)) {
  81         size_t offset = 0;
  82         int i;
  83
  84         for (i = 0; i < req->elem.in_num - 1; i++) {
  85             size_t len;
  86
  87             /* Be pretty defensive wrt malicious guests */
  88             len = MIN(req->elem.in_sg[i].iov_len,
  89                       req->size - offset);
  90
  91             memcpy(req->elem.in_sg[i].iov_base,
  92                    req->buffer + offset,
  93                    len);
  94             offset += len;
  95         }
  96     } else if (ret && (req->out->type & VIRTIO_BLK_T_OUT)) {
  97         if (virtio_blk_handle_write_error(req, -ret))
  98             return;
  99     }
 100
 101     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 102 }
 103
 104 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
 105 {
 106     VirtIOBlockReq *req = qemu_mallocz(sizeof(*req));
 107     if (req != NULL)
 108         req->dev = s;
 109     return req;
 110 }
 111
 112 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 113 {
 114     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 115
 116     if (req != NULL) {
 117         if (!virtqueue_pop(s->vq, &req->elem)) {
 118             qemu_free(req);
 119             return NULL;
 120         }
 121     }
 122
 123     return req;
 124 }
 125
 126 static int virtio_blk_handle_write(VirtIOBlockReq *req)
 127 {
 128     if (!req->buffer) {
 129         size_t offset = 0;
 130         int i;
 131
 132         for (i = 1; i < req->elem.out_num; i++)
 133             req->size += req->elem.out_sg[i].iov_len;
 134
 135         req->buffer = qemu_memalign(512, req->size);
 136         if (req->buffer == NULL) {
 137             qemu_free(req);
 138             return -1;
 139         }
 140
 141         /* We copy the data from the SG list to avoid splitting up the request.
 142            This helps performance a lot until we can pass full sg lists as AIO
 143            operations */
 144         for (i = 1; i < req->elem.out_num; i++) {
 145             size_t len;
 146
 147             len = MIN(req->elem.out_sg[i].iov_len,
 148                     req->size - offset);
 149             memcpy(req->buffer + offset,
 150                     req->elem.out_sg[i].iov_base,
 151                     len);
 152             offset += len;
 153         }
 154     }
 155
 156     bdrv_aio_write(req->dev->bs, req->out->sector, req->buffer, req->size / 512,
 157             virtio_blk_rw_complete, req);
 158     return 0;
 159 }
 160
 161 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 162 {
 163     VirtIOBlock *s = to_virtio_blk(vdev);
 164     VirtIOBlockReq *req;
 165
 166     while ((req = virtio_blk_get_request(s))) {
 167         int i;
 168
 169         if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 170             fprintf(stderr, "virtio-blk missing headers\n");
 171             exit(1);
 172         }
 173
 174         if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 175             req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 176             fprintf(stderr, "virtio-blk header not in correct element\n");
 177             exit(1);
 178         }
 179
 180         req->out = (void *)req->elem.out_sg[0].iov_base;
 181         req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 182
 183         if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
 184             unsigned int len = sizeof(*req->in);
 185
 186             req->in->status = VIRTIO_BLK_S_UNSUPP;
 187             virtqueue_push(vq, &req->elem, len);
 188             virtio_notify(vdev, vq);
 189             qemu_free(req);
 190         } else if (req->out->type & VIRTIO_BLK_T_OUT) {
 191             if (virtio_blk_handle_write(req) < 0)
 192                 break;
 193         } else {
 194             for (i = 0; i < req->elem.in_num - 1; i++)
 195                 req->size += req->elem.in_sg[i].iov_len;
 196
 197             req->buffer = qemu_memalign(512, req->size);
 198             if (req->buffer == NULL) {
 199                 qemu_free(req);
 200                 break;
 201             }
 202
 203             bdrv_aio_read(s->bs, req->out->sector,
 204                           req->buffer,
 205                           req->size / 512,
 206                           virtio_blk_rw_complete,
 207                           req);
 208         }
 209     }
 210     /*
 211      * FIXME: Want to check for completions before returning to guest mode,
 212      * so cached reads and writes are reported as quickly as possible. But
 213      * that should be done in the generic block layer.
 214      */
 215 }
 216
 217 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
 218 {
 219     VirtIOBlock *s = opaque;
 220     VirtIOBlockReq *req = s->rq;
 221
 222     if (!running)
 223         return;
 224
 225     s->rq = NULL;
 226
 227     while (req) {
 228         virtio_blk_handle_write(req);
 229         req = req->next;
 230     }
 231 }
 232
 233 static void virtio_blk_reset(VirtIODevice *vdev)
 234 {
 235     /*
 236      * This should cancel pending requests, but can't do nicely until there
 237      * are per-device request lists.
 238      */
 239     qemu_aio_flush();
 240 }
 241
 242 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 243 {
 244     VirtIOBlock *s = to_virtio_blk(vdev);
 245     struct virtio_blk_config blkcfg;
 246     uint64_t capacity;
 247     int cylinders, heads, secs;
 248
 249     bdrv_get_geometry(s->bs, &capacity);
 250     bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
 251     stq_raw(&blkcfg.capacity, capacity);
 252     stl_raw(&blkcfg.seg_max, 128 - 2);
 253     stw_raw(&blkcfg.cylinders, cylinders);
 254     blkcfg.heads = heads;
 255     blkcfg.sectors = secs;
 256     memcpy(config, &blkcfg, sizeof(blkcfg));
 257 }
 258
 259 static uint32_t virtio_blk_get_features(VirtIODevice *vdev)
 260 {
 261     return (1 << VIRTIO_BLK_F_SEG_MAX | 1 << VIRTIO_BLK_F_GEOMETRY);
 262 }
 263
 264 static void virtio_blk_save(QEMUFile *f, void *opaque)
 265 {
 266     VirtIOBlock *s = opaque;
 267     VirtIOBlockReq *req = s->rq;
 268
 269     virtio_save(&s->vdev, f);
 270
 271     while (req) {
 272         qemu_put_sbyte(f, 1);
 273         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 274         req = req->next;
 275     }
 276     qemu_put_sbyte(f, 0);
 277 }
 278
 279 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 280 {
 281     VirtIOBlock *s = opaque;
 282
 283     if (version_id != 2)
 284         return -EINVAL;
 285
 286     virtio_load(&s->vdev, f);
 287     while (qemu_get_sbyte(f)) {
 288         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 289         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 290         req->next = s->rq;
 291         s->rq = req->next;
 292     }
 293
 294     return 0;
 295 }
 296
 297 void *virtio_blk_init(PCIBus *bus, BlockDriverState *bs)
 298 {
 299     VirtIOBlock *s;
 300     int cylinders, heads, secs;
 301     static int virtio_blk_id;
 302
 303     s = (VirtIOBlock *)virtio_init_pci(bus, "virtio-blk",
 304                                        PCI_VENDOR_ID_REDHAT_QUMRANET,
 305                                        PCI_DEVICE_ID_VIRTIO_BLOCK,
 306                                        PCI_VENDOR_ID_REDHAT_QUMRANET,
 307                                        VIRTIO_ID_BLOCK,
 308                                        0x01, 0x80, 0x00,
 309                                        sizeof(struct virtio_blk_config), sizeof(VirtIOBlock));
 310     if (!s)
 311         return NULL;
 312
 313     s->vdev.get_config = virtio_blk_update_config;
 314     s->vdev.get_features = virtio_blk_get_features;
 315     s->vdev.reset = virtio_blk_reset;
 316     s->bs = bs;
 317     s->rq = NULL;
 318     bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
 319     bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
 320
 321     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 322
 323     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 324     register_savevm("virtio-blk", virtio_blk_id++, 2,
 325                     virtio_blk_save, virtio_blk_load, s);
 326
 327     return s;
 328 }