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 = virtqueue_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 switch (vbr->req->cmd_type) { 69 case REQ_TYPE_BLOCK_PC: 70 vbr->req->resid_len = vbr->in_hdr.residual; 71 vbr->req->sense_len = vbr->in_hdr.sense_len; 72 vbr->req->errors = vbr->in_hdr.errors; 73 break; 74 case REQ_TYPE_SPECIAL: 75 vbr->req->errors = (error != 0); 76 break; 77 default: 78 break; 79 } 80 81 __blk_end_request_all(vbr->req, error); 82 list_del(&vbr->list); 83 mempool_free(vbr, vblk->pool); 84 } 85 /* In case queue is stopped waiting for more buffers. */ 86 blk_start_queue(vblk->disk->queue); 87 spin_unlock_irqrestore(&vblk->lock, flags); 88 } 89 90 static bool do_req(struct request_queue *q, struct virtio_blk *vblk, 91 struct request *req) 92 { 93 unsigned long num, out = 0, in = 0; 94 struct virtblk_req *vbr; 95 96 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC); 97 if (!vbr) 98 /* When another request finishes we'll try again. */ 99 return false; 100 101 vbr->req = req; 102 103 if (req->cmd_flags & REQ_FLUSH) { 104 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH; 105 vbr->out_hdr.sector = 0; 106 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 107 } else { 108 switch (req->cmd_type) { 109 case REQ_TYPE_FS: 110 vbr->out_hdr.type = 0; 111 vbr->out_hdr.sector = blk_rq_pos(vbr->req); 112 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 113 break; 114 case REQ_TYPE_BLOCK_PC: 115 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; 116 vbr->out_hdr.sector = 0; 117 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 118 break; 119 case REQ_TYPE_SPECIAL: 120 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID; 121 vbr->out_hdr.sector = 0; 122 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 123 break; 124 default: 125 /* We don't put anything else in the queue. */ 126 BUG(); 127 } 128 } 129 130 if (vbr->req->cmd_flags & REQ_HARDBARRIER) 131 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER; 132 133 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr)); 134 135 /* 136 * If this is a packet command we need a couple of additional headers. 137 * Behind the normal outhdr we put a segment with the scsi command 138 * block, and before the normal inhdr we put the sense data and the 139 * inhdr with additional status information before the normal inhdr. 140 */ 141 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) 142 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len); 143 144 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out); 145 146 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) { 147 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96); 148 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr, 149 sizeof(vbr->in_hdr)); 150 } 151 152 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status, 153 sizeof(vbr->status)); 154 155 if (num) { 156 if (rq_data_dir(vbr->req) == WRITE) { 157 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; 158 out += num; 159 } else { 160 vbr->out_hdr.type |= VIRTIO_BLK_T_IN; 161 in += num; 162 } 163 } 164 165 if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) { 166 mempool_free(vbr, vblk->pool); 167 return false; 168 } 169 170 list_add_tail(&vbr->list, &vblk->reqs); 171 return true; 172 } 173 174 static void do_virtblk_request(struct request_queue *q) 175 { 176 struct virtio_blk *vblk = q->queuedata; 177 struct request *req; 178 unsigned int issued = 0; 179 180 while ((req = blk_peek_request(q)) != NULL) { 181 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); 182 183 /* If this request fails, stop queue and wait for something to 184 finish to restart it. */ 185 if (!do_req(q, vblk, req)) { 186 blk_stop_queue(q); 187 break; 188 } 189 blk_start_request(req); 190 issued++; 191 } 192 193 if (issued) 194 virtqueue_kick(vblk->vq); 195 } 196 197 /* return id (s/n) string for *disk to *id_str 198 */ 199 static int virtblk_get_id(struct gendisk *disk, char *id_str) 200 { 201 struct virtio_blk *vblk = disk->private_data; 202 struct request *req; 203 struct bio *bio; 204 int err; 205 206 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES, 207 GFP_KERNEL); 208 if (IS_ERR(bio)) 209 return PTR_ERR(bio); 210 211 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL); 212 if (IS_ERR(req)) { 213 bio_put(bio); 214 return PTR_ERR(req); 215 } 216 217 req->cmd_type = REQ_TYPE_SPECIAL; 218 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); 219 blk_put_request(req); 220 221 return err; 222 } 223 224 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, 225 unsigned int cmd, unsigned long data) 226 { 227 struct gendisk *disk = bdev->bd_disk; 228 struct virtio_blk *vblk = disk->private_data; 229 230 /* 231 * Only allow the generic SCSI ioctls if the host can support it. 232 */ 233 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) 234 return -ENOTTY; 235 236 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, 237 (void __user *)data); 238 } 239 240 /* We provide getgeo only to please some old bootloader/partitioning tools */ 241 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) 242 { 243 struct virtio_blk *vblk = bd->bd_disk->private_data; 244 struct virtio_blk_geometry vgeo; 245 int err; 246 247 /* see if the host passed in geometry config */ 248 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY, 249 offsetof(struct virtio_blk_config, geometry), 250 &vgeo); 251 252 if (!err) { 253 geo->heads = vgeo.heads; 254 geo->sectors = vgeo.sectors; 255 geo->cylinders = vgeo.cylinders; 256 } else { 257 /* some standard values, similar to sd */ 258 geo->heads = 1 << 6; 259 geo->sectors = 1 << 5; 260 geo->cylinders = get_capacity(bd->bd_disk) >> 11; 261 } 262 return 0; 263 } 264 265 static const struct block_device_operations virtblk_fops = { 266 .ioctl = virtblk_ioctl, 267 .owner = THIS_MODULE, 268 .getgeo = virtblk_getgeo, 269 }; 270 271 static int index_to_minor(int index) 272 { 273 return index << PART_BITS; 274 } 275 276 static ssize_t virtblk_serial_show(struct device *dev, 277 struct device_attribute *attr, char *buf) 278 { 279 struct gendisk *disk = dev_to_disk(dev); 280 int err; 281 282 /* sysfs gives us a PAGE_SIZE buffer */ 283 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); 284 285 buf[VIRTIO_BLK_ID_BYTES] = '\0'; 286 err = virtblk_get_id(disk, buf); 287 if (!err) 288 return strlen(buf); 289 290 if (err == -EIO) /* Unsupported? Make it empty. */ 291 return 0; 292 293 return err; 294 } 295 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL); 296 297 static int __devinit virtblk_probe(struct virtio_device *vdev) 298 { 299 struct virtio_blk *vblk; 300 struct request_queue *q; 301 int err; 302 u64 cap; 303 u32 v, blk_size, sg_elems, opt_io_size; 304 u16 min_io_size; 305 u8 physical_block_exp, alignment_offset; 306 307 if (index_to_minor(index) >= 1 << MINORBITS) 308 return -ENOSPC; 309 310 /* We need to know how many segments before we allocate. */ 311 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX, 312 offsetof(struct virtio_blk_config, seg_max), 313 &sg_elems); 314 315 /* We need at least one SG element, whatever they say. */ 316 if (err || !sg_elems) 317 sg_elems = 1; 318 319 /* We need an extra sg elements at head and tail. */ 320 sg_elems += 2; 321 vdev->priv = vblk = kmalloc(sizeof(*vblk) + 322 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL); 323 if (!vblk) { 324 err = -ENOMEM; 325 goto out; 326 } 327 328 INIT_LIST_HEAD(&vblk->reqs); 329 spin_lock_init(&vblk->lock); 330 vblk->vdev = vdev; 331 vblk->sg_elems = sg_elems; 332 sg_init_table(vblk->sg, vblk->sg_elems); 333 334 /* We expect one virtqueue, for output. */ 335 vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests"); 336 if (IS_ERR(vblk->vq)) { 337 err = PTR_ERR(vblk->vq); 338 goto out_free_vblk; 339 } 340 341 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req)); 342 if (!vblk->pool) { 343 err = -ENOMEM; 344 goto out_free_vq; 345 } 346 347 /* FIXME: How many partitions? How long is a piece of string? */ 348 vblk->disk = alloc_disk(1 << PART_BITS); 349 if (!vblk->disk) { 350 err = -ENOMEM; 351 goto out_mempool; 352 } 353 354 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock); 355 if (!q) { 356 err = -ENOMEM; 357 goto out_put_disk; 358 } 359 360 q->queuedata = vblk; 361 362 if (index < 26) { 363 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26); 364 } else if (index < (26 + 1) * 26) { 365 sprintf(vblk->disk->disk_name, "vd%c%c", 366 'a' + index / 26 - 1, 'a' + index % 26); 367 } else { 368 const unsigned int m1 = (index / 26 - 1) / 26 - 1; 369 const unsigned int m2 = (index / 26 - 1) % 26; 370 const unsigned int m3 = index % 26; 371 sprintf(vblk->disk->disk_name, "vd%c%c%c", 372 'a' + m1, 'a' + m2, 'a' + m3); 373 } 374 375 vblk->disk->major = major; 376 vblk->disk->first_minor = index_to_minor(index); 377 vblk->disk->private_data = vblk; 378 vblk->disk->fops = &virtblk_fops; 379 vblk->disk->driverfs_dev = &vdev->dev; 380 index++; 381 382 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH)) { 383 /* 384 * If the FLUSH feature is supported we do have support for 385 * flushing a volatile write cache on the host. Use that 386 * to implement write barrier support. 387 */ 388 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH); 389 } else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER)) { 390 /* 391 * If the BARRIER feature is supported the host expects us 392 * to order request by tags. This implies there is not 393 * volatile write cache on the host, and that the host 394 * never re-orders outstanding I/O. This feature is not 395 * useful for real life scenarious and deprecated. 396 */ 397 blk_queue_ordered(q, QUEUE_ORDERED_TAG); 398 } else { 399 /* 400 * If the FLUSH feature is not supported we must assume that 401 * the host does not perform any kind of volatile write 402 * caching. We still need to drain the queue to provider 403 * proper barrier semantics. 404 */ 405 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN); 406 } 407 408 /* If disk is read-only in the host, the guest should obey */ 409 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) 410 set_disk_ro(vblk->disk, 1); 411 412 /* Host must always specify the capacity. */ 413 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), 414 &cap, sizeof(cap)); 415 416 /* If capacity is too big, truncate with warning. */ 417 if ((sector_t)cap != cap) { 418 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", 419 (unsigned long long)cap); 420 cap = (sector_t)-1; 421 } 422 set_capacity(vblk->disk, cap); 423 424 /* We can handle whatever the host told us to handle. */ 425 blk_queue_max_segments(q, vblk->sg_elems-2); 426 427 /* No need to bounce any requests */ 428 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); 429 430 /* No real sector limit. */ 431 blk_queue_max_hw_sectors(q, -1U); 432 433 /* Host can optionally specify maximum segment size and number of 434 * segments. */ 435 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX, 436 offsetof(struct virtio_blk_config, size_max), 437 &v); 438 if (!err) 439 blk_queue_max_segment_size(q, v); 440 else 441 blk_queue_max_segment_size(q, -1U); 442 443 /* Host can optionally specify the block size of the device */ 444 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE, 445 offsetof(struct virtio_blk_config, blk_size), 446 &blk_size); 447 if (!err) 448 blk_queue_logical_block_size(q, blk_size); 449 else 450 blk_size = queue_logical_block_size(q); 451 452 /* Use topology information if available */ 453 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 454 offsetof(struct virtio_blk_config, physical_block_exp), 455 &physical_block_exp); 456 if (!err && physical_block_exp) 457 blk_queue_physical_block_size(q, 458 blk_size * (1 << physical_block_exp)); 459 460 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 461 offsetof(struct virtio_blk_config, alignment_offset), 462 &alignment_offset); 463 if (!err && alignment_offset) 464 blk_queue_alignment_offset(q, blk_size * alignment_offset); 465 466 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 467 offsetof(struct virtio_blk_config, min_io_size), 468 &min_io_size); 469 if (!err && min_io_size) 470 blk_queue_io_min(q, blk_size * min_io_size); 471 472 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 473 offsetof(struct virtio_blk_config, opt_io_size), 474 &opt_io_size); 475 if (!err && opt_io_size) 476 blk_queue_io_opt(q, blk_size * opt_io_size); 477 478 479 add_disk(vblk->disk); 480 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial); 481 if (err) 482 goto out_del_disk; 483 484 return 0; 485 486 out_del_disk: 487 del_gendisk(vblk->disk); 488 blk_cleanup_queue(vblk->disk->queue); 489 out_put_disk: 490 put_disk(vblk->disk); 491 out_mempool: 492 mempool_destroy(vblk->pool); 493 out_free_vq: 494 vdev->config->del_vqs(vdev); 495 out_free_vblk: 496 kfree(vblk); 497 out: 498 return err; 499 } 500 501 static void __devexit virtblk_remove(struct virtio_device *vdev) 502 { 503 struct virtio_blk *vblk = vdev->priv; 504 505 /* Nothing should be pending. */ 506 BUG_ON(!list_empty(&vblk->reqs)); 507 508 /* Stop all the virtqueues. */ 509 vdev->config->reset(vdev); 510 511 del_gendisk(vblk->disk); 512 blk_cleanup_queue(vblk->disk->queue); 513 put_disk(vblk->disk); 514 mempool_destroy(vblk->pool); 515 vdev->config->del_vqs(vdev); 516 kfree(vblk); 517 } 518 519 static const struct virtio_device_id id_table[] = { 520 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, 521 { 0 }, 522 }; 523 524 static unsigned int features[] = { 525 VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, 526 VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, 527 VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY 528 }; 529 530 /* 531 * virtio_blk causes spurious section mismatch warning by 532 * simultaneously referring to a __devinit and a __devexit function. 533 * Use __refdata to avoid this warning. 534 */ 535 static struct virtio_driver __refdata virtio_blk = { 536 .feature_table = features, 537 .feature_table_size = ARRAY_SIZE(features), 538 .driver.name = KBUILD_MODNAME, 539 .driver.owner = THIS_MODULE, 540 .id_table = id_table, 541 .probe = virtblk_probe, 542 .remove = __devexit_p(virtblk_remove), 543 }; 544 545 static int __init init(void) 546 { 547 major = register_blkdev(0, "virtblk"); 548 if (major < 0) 549 return major; 550 return register_virtio_driver(&virtio_blk); 551 } 552 553 static void __exit fini(void) 554 { 555 unregister_blkdev(major, "virtblk"); 556 unregister_virtio_driver(&virtio_blk); 557 } 558 module_init(init); 559 module_exit(fini); 560 561 MODULE_DEVICE_TABLE(virtio, id_table); 562 MODULE_DESCRIPTION("Virtio block driver"); 563 MODULE_LICENSE("GPL"); 564