1 //#define DEBUG 2 #include <linux/spinlock.h> 3 #include <linux/slab.h> 4 #include <linux/blkdev.h> 5 #include <linux/hdreg.h> 6 #include <linux/virtio.h> 7 #include <linux/virtio_blk.h> 8 #include <linux/scatterlist.h> 9 10 #define PART_BITS 4 11 12 static int major, index; 13 14 struct virtio_blk 15 { 16 spinlock_t lock; 17 18 struct virtio_device *vdev; 19 struct virtqueue *vq; 20 21 /* The disk structure for the kernel. */ 22 struct gendisk *disk; 23 24 /* Request tracking. */ 25 struct list_head reqs; 26 27 mempool_t *pool; 28 29 /* What host tells us, plus 2 for header & tailer. */ 30 unsigned int sg_elems; 31 32 /* Scatterlist: can be too big for stack. */ 33 struct scatterlist sg[/*sg_elems*/]; 34 }; 35 36 struct virtblk_req 37 { 38 struct list_head list; 39 struct request *req; 40 struct virtio_blk_outhdr out_hdr; 41 struct virtio_scsi_inhdr in_hdr; 42 u8 status; 43 }; 44 45 static void blk_done(struct virtqueue *vq) 46 { 47 struct virtio_blk *vblk = vq->vdev->priv; 48 struct virtblk_req *vbr; 49 unsigned int len; 50 unsigned long flags; 51 52 spin_lock_irqsave(&vblk->lock, flags); 53 while ((vbr = vblk->vq->vq_ops->get_buf(vblk->vq, &len)) != NULL) { 54 int error; 55 56 switch (vbr->status) { 57 case VIRTIO_BLK_S_OK: 58 error = 0; 59 break; 60 case VIRTIO_BLK_S_UNSUPP: 61 error = -ENOTTY; 62 break; 63 default: 64 error = -EIO; 65 break; 66 } 67 68 if (blk_pc_request(vbr->req)) { 69 vbr->req->resid_len = vbr->in_hdr.residual; 70 vbr->req->sense_len = vbr->in_hdr.sense_len; 71 vbr->req->errors = vbr->in_hdr.errors; 72 } 73 74 __blk_end_request_all(vbr->req, error); 75 list_del(&vbr->list); 76 mempool_free(vbr, vblk->pool); 77 } 78 /* In case queue is stopped waiting for more buffers. */ 79 blk_start_queue(vblk->disk->queue); 80 spin_unlock_irqrestore(&vblk->lock, flags); 81 } 82 83 static bool do_req(struct request_queue *q, struct virtio_blk *vblk, 84 struct request *req) 85 { 86 unsigned long num, out = 0, in = 0; 87 struct virtblk_req *vbr; 88 89 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC); 90 if (!vbr) 91 /* When another request finishes we'll try again. */ 92 return false; 93 94 vbr->req = req; 95 switch (req->cmd_type) { 96 case REQ_TYPE_FS: 97 vbr->out_hdr.type = 0; 98 vbr->out_hdr.sector = blk_rq_pos(vbr->req); 99 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 100 break; 101 case REQ_TYPE_BLOCK_PC: 102 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; 103 vbr->out_hdr.sector = 0; 104 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 105 break; 106 case REQ_TYPE_LINUX_BLOCK: 107 if (req->cmd[0] == REQ_LB_OP_FLUSH) { 108 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH; 109 vbr->out_hdr.sector = 0; 110 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 111 break; 112 } 113 /*FALLTHRU*/ 114 default: 115 /* We don't put anything else in the queue. */ 116 BUG(); 117 } 118 119 if (blk_barrier_rq(vbr->req)) 120 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER; 121 122 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr)); 123 124 /* 125 * If this is a packet command we need a couple of additional headers. 126 * Behind the normal outhdr we put a segment with the scsi command 127 * block, and before the normal inhdr we put the sense data and the 128 * inhdr with additional status information before the normal inhdr. 129 */ 130 if (blk_pc_request(vbr->req)) 131 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len); 132 133 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out); 134 135 if (blk_pc_request(vbr->req)) { 136 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96); 137 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr, 138 sizeof(vbr->in_hdr)); 139 } 140 141 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status, 142 sizeof(vbr->status)); 143 144 if (num) { 145 if (rq_data_dir(vbr->req) == WRITE) { 146 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; 147 out += num; 148 } else { 149 vbr->out_hdr.type |= VIRTIO_BLK_T_IN; 150 in += num; 151 } 152 } 153 154 if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) { 155 mempool_free(vbr, vblk->pool); 156 return false; 157 } 158 159 list_add_tail(&vbr->list, &vblk->reqs); 160 return true; 161 } 162 163 static void do_virtblk_request(struct request_queue *q) 164 { 165 struct virtio_blk *vblk = q->queuedata; 166 struct request *req; 167 unsigned int issued = 0; 168 169 while ((req = blk_peek_request(q)) != NULL) { 170 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); 171 172 /* If this request fails, stop queue and wait for something to 173 finish to restart it. */ 174 if (!do_req(q, vblk, req)) { 175 blk_stop_queue(q); 176 break; 177 } 178 blk_start_request(req); 179 issued++; 180 } 181 182 if (issued) 183 vblk->vq->vq_ops->kick(vblk->vq); 184 } 185 186 static void virtblk_prepare_flush(struct request_queue *q, struct request *req) 187 { 188 req->cmd_type = REQ_TYPE_LINUX_BLOCK; 189 req->cmd[0] = REQ_LB_OP_FLUSH; 190 } 191 192 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, 193 unsigned cmd, unsigned long data) 194 { 195 struct gendisk *disk = bdev->bd_disk; 196 struct virtio_blk *vblk = disk->private_data; 197 198 /* 199 * Only allow the generic SCSI ioctls if the host can support it. 200 */ 201 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) 202 return -ENOTTY; 203 204 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, 205 (void __user *)data); 206 } 207 208 /* We provide getgeo only to please some old bootloader/partitioning tools */ 209 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) 210 { 211 struct virtio_blk *vblk = bd->bd_disk->private_data; 212 struct virtio_blk_geometry vgeo; 213 int err; 214 215 /* see if the host passed in geometry config */ 216 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY, 217 offsetof(struct virtio_blk_config, geometry), 218 &vgeo); 219 220 if (!err) { 221 geo->heads = vgeo.heads; 222 geo->sectors = vgeo.sectors; 223 geo->cylinders = vgeo.cylinders; 224 } else { 225 /* some standard values, similar to sd */ 226 geo->heads = 1 << 6; 227 geo->sectors = 1 << 5; 228 geo->cylinders = get_capacity(bd->bd_disk) >> 11; 229 } 230 return 0; 231 } 232 233 static const struct block_device_operations virtblk_fops = { 234 .locked_ioctl = virtblk_ioctl, 235 .owner = THIS_MODULE, 236 .getgeo = virtblk_getgeo, 237 }; 238 239 static int index_to_minor(int index) 240 { 241 return index << PART_BITS; 242 } 243 244 static int __devinit virtblk_probe(struct virtio_device *vdev) 245 { 246 struct virtio_blk *vblk; 247 struct request_queue *q; 248 int err; 249 u64 cap; 250 u32 v, blk_size, sg_elems, opt_io_size; 251 u16 min_io_size; 252 u8 physical_block_exp, alignment_offset; 253 254 if (index_to_minor(index) >= 1 << MINORBITS) 255 return -ENOSPC; 256 257 /* We need to know how many segments before we allocate. */ 258 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX, 259 offsetof(struct virtio_blk_config, seg_max), 260 &sg_elems); 261 if (err) 262 sg_elems = 1; 263 264 /* We need an extra sg elements at head and tail. */ 265 sg_elems += 2; 266 vdev->priv = vblk = kmalloc(sizeof(*vblk) + 267 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL); 268 if (!vblk) { 269 err = -ENOMEM; 270 goto out; 271 } 272 273 INIT_LIST_HEAD(&vblk->reqs); 274 spin_lock_init(&vblk->lock); 275 vblk->vdev = vdev; 276 vblk->sg_elems = sg_elems; 277 sg_init_table(vblk->sg, vblk->sg_elems); 278 279 /* We expect one virtqueue, for output. */ 280 vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests"); 281 if (IS_ERR(vblk->vq)) { 282 err = PTR_ERR(vblk->vq); 283 goto out_free_vblk; 284 } 285 286 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req)); 287 if (!vblk->pool) { 288 err = -ENOMEM; 289 goto out_free_vq; 290 } 291 292 /* FIXME: How many partitions? How long is a piece of string? */ 293 vblk->disk = alloc_disk(1 << PART_BITS); 294 if (!vblk->disk) { 295 err = -ENOMEM; 296 goto out_mempool; 297 } 298 299 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock); 300 if (!q) { 301 err = -ENOMEM; 302 goto out_put_disk; 303 } 304 305 q->queuedata = vblk; 306 307 if (index < 26) { 308 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26); 309 } else if (index < (26 + 1) * 26) { 310 sprintf(vblk->disk->disk_name, "vd%c%c", 311 'a' + index / 26 - 1, 'a' + index % 26); 312 } else { 313 const unsigned int m1 = (index / 26 - 1) / 26 - 1; 314 const unsigned int m2 = (index / 26 - 1) % 26; 315 const unsigned int m3 = index % 26; 316 sprintf(vblk->disk->disk_name, "vd%c%c%c", 317 'a' + m1, 'a' + m2, 'a' + m3); 318 } 319 320 vblk->disk->major = major; 321 vblk->disk->first_minor = index_to_minor(index); 322 vblk->disk->private_data = vblk; 323 vblk->disk->fops = &virtblk_fops; 324 vblk->disk->driverfs_dev = &vdev->dev; 325 index++; 326 327 /* If barriers are supported, tell block layer that queue is ordered */ 328 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH)) 329 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH, 330 virtblk_prepare_flush); 331 else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER)) 332 blk_queue_ordered(q, QUEUE_ORDERED_TAG, NULL); 333 334 /* If disk is read-only in the host, the guest should obey */ 335 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) 336 set_disk_ro(vblk->disk, 1); 337 338 /* Host must always specify the capacity. */ 339 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), 340 &cap, sizeof(cap)); 341 342 /* If capacity is too big, truncate with warning. */ 343 if ((sector_t)cap != cap) { 344 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", 345 (unsigned long long)cap); 346 cap = (sector_t)-1; 347 } 348 set_capacity(vblk->disk, cap); 349 350 /* We can handle whatever the host told us to handle. */ 351 blk_queue_max_segments(q, vblk->sg_elems-2); 352 353 /* No need to bounce any requests */ 354 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); 355 356 /* No real sector limit. */ 357 blk_queue_max_hw_sectors(q, -1U); 358 359 /* Host can optionally specify maximum segment size and number of 360 * segments. */ 361 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX, 362 offsetof(struct virtio_blk_config, size_max), 363 &v); 364 if (!err) 365 blk_queue_max_segment_size(q, v); 366 else 367 blk_queue_max_segment_size(q, -1U); 368 369 /* Host can optionally specify the block size of the device */ 370 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE, 371 offsetof(struct virtio_blk_config, blk_size), 372 &blk_size); 373 if (!err) 374 blk_queue_logical_block_size(q, blk_size); 375 else 376 blk_size = queue_logical_block_size(q); 377 378 /* Use topology information if available */ 379 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 380 offsetof(struct virtio_blk_config, physical_block_exp), 381 &physical_block_exp); 382 if (!err && physical_block_exp) 383 blk_queue_physical_block_size(q, 384 blk_size * (1 << physical_block_exp)); 385 386 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 387 offsetof(struct virtio_blk_config, alignment_offset), 388 &alignment_offset); 389 if (!err && alignment_offset) 390 blk_queue_alignment_offset(q, blk_size * alignment_offset); 391 392 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 393 offsetof(struct virtio_blk_config, min_io_size), 394 &min_io_size); 395 if (!err && min_io_size) 396 blk_queue_io_min(q, blk_size * min_io_size); 397 398 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 399 offsetof(struct virtio_blk_config, opt_io_size), 400 &opt_io_size); 401 if (!err && opt_io_size) 402 blk_queue_io_opt(q, blk_size * opt_io_size); 403 404 405 add_disk(vblk->disk); 406 return 0; 407 408 out_put_disk: 409 put_disk(vblk->disk); 410 out_mempool: 411 mempool_destroy(vblk->pool); 412 out_free_vq: 413 vdev->config->del_vqs(vdev); 414 out_free_vblk: 415 kfree(vblk); 416 out: 417 return err; 418 } 419 420 static void __devexit virtblk_remove(struct virtio_device *vdev) 421 { 422 struct virtio_blk *vblk = vdev->priv; 423 424 /* Nothing should be pending. */ 425 BUG_ON(!list_empty(&vblk->reqs)); 426 427 /* Stop all the virtqueues. */ 428 vdev->config->reset(vdev); 429 430 del_gendisk(vblk->disk); 431 blk_cleanup_queue(vblk->disk->queue); 432 put_disk(vblk->disk); 433 mempool_destroy(vblk->pool); 434 vdev->config->del_vqs(vdev); 435 kfree(vblk); 436 } 437 438 static const struct virtio_device_id id_table[] = { 439 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, 440 { 0 }, 441 }; 442 443 static unsigned int features[] = { 444 VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, 445 VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, 446 VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY 447 }; 448 449 /* 450 * virtio_blk causes spurious section mismatch warning by 451 * simultaneously referring to a __devinit and a __devexit function. 452 * Use __refdata to avoid this warning. 453 */ 454 static struct virtio_driver __refdata virtio_blk = { 455 .feature_table = features, 456 .feature_table_size = ARRAY_SIZE(features), 457 .driver.name = KBUILD_MODNAME, 458 .driver.owner = THIS_MODULE, 459 .id_table = id_table, 460 .probe = virtblk_probe, 461 .remove = __devexit_p(virtblk_remove), 462 }; 463 464 static int __init init(void) 465 { 466 major = register_blkdev(0, "virtblk"); 467 if (major < 0) 468 return major; 469 return register_virtio_driver(&virtio_blk); 470 } 471 472 static void __exit fini(void) 473 { 474 unregister_blkdev(major, "virtblk"); 475 unregister_virtio_driver(&virtio_blk); 476 } 477 module_init(init); 478 module_exit(fini); 479 480 MODULE_DEVICE_TABLE(virtio, id_table); 481 MODULE_DESCRIPTION("Virtio block driver"); 482 MODULE_LICENSE("GPL"); 483