virtio_blk.c revision 5087a50e66bd51b6e72c60bce4757a42b93f6b2c
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#include <linux/string_helpers.h> 10#include <scsi/scsi_cmnd.h> 11#include <linux/idr.h> 12 13#define PART_BITS 4 14 15static int major; 16static DEFINE_IDA(vd_index_ida); 17 18struct workqueue_struct *virtblk_wq; 19 20struct virtio_blk 21{ 22 spinlock_t lock; 23 24 struct virtio_device *vdev; 25 struct virtqueue *vq; 26 27 /* The disk structure for the kernel. */ 28 struct gendisk *disk; 29 30 /* Request tracking. */ 31 struct list_head reqs; 32 33 mempool_t *pool; 34 35 /* Process context for config space updates */ 36 struct work_struct config_work; 37 38 /* What host tells us, plus 2 for header & tailer. */ 39 unsigned int sg_elems; 40 41 /* Ida index - used to track minor number allocations. */ 42 int index; 43 44 /* Scatterlist: can be too big for stack. */ 45 struct scatterlist sg[/*sg_elems*/]; 46}; 47 48struct virtblk_req 49{ 50 struct list_head list; 51 struct request *req; 52 struct virtio_blk_outhdr out_hdr; 53 struct virtio_scsi_inhdr in_hdr; 54 u8 status; 55}; 56 57static void blk_done(struct virtqueue *vq) 58{ 59 struct virtio_blk *vblk = vq->vdev->priv; 60 struct virtblk_req *vbr; 61 unsigned int len; 62 unsigned long flags; 63 64 spin_lock_irqsave(&vblk->lock, flags); 65 while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) { 66 int error; 67 68 switch (vbr->status) { 69 case VIRTIO_BLK_S_OK: 70 error = 0; 71 break; 72 case VIRTIO_BLK_S_UNSUPP: 73 error = -ENOTTY; 74 break; 75 default: 76 error = -EIO; 77 break; 78 } 79 80 switch (vbr->req->cmd_type) { 81 case REQ_TYPE_BLOCK_PC: 82 vbr->req->resid_len = vbr->in_hdr.residual; 83 vbr->req->sense_len = vbr->in_hdr.sense_len; 84 vbr->req->errors = vbr->in_hdr.errors; 85 break; 86 case REQ_TYPE_SPECIAL: 87 vbr->req->errors = (error != 0); 88 break; 89 default: 90 break; 91 } 92 93 __blk_end_request_all(vbr->req, error); 94 list_del(&vbr->list); 95 mempool_free(vbr, vblk->pool); 96 } 97 /* In case queue is stopped waiting for more buffers. */ 98 blk_start_queue(vblk->disk->queue); 99 spin_unlock_irqrestore(&vblk->lock, flags); 100} 101 102static bool do_req(struct request_queue *q, struct virtio_blk *vblk, 103 struct request *req) 104{ 105 unsigned long num, out = 0, in = 0; 106 struct virtblk_req *vbr; 107 108 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC); 109 if (!vbr) 110 /* When another request finishes we'll try again. */ 111 return false; 112 113 vbr->req = req; 114 115 if (req->cmd_flags & REQ_FLUSH) { 116 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH; 117 vbr->out_hdr.sector = 0; 118 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 119 } else { 120 switch (req->cmd_type) { 121 case REQ_TYPE_FS: 122 vbr->out_hdr.type = 0; 123 vbr->out_hdr.sector = blk_rq_pos(vbr->req); 124 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 125 break; 126 case REQ_TYPE_BLOCK_PC: 127 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; 128 vbr->out_hdr.sector = 0; 129 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 130 break; 131 case REQ_TYPE_SPECIAL: 132 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID; 133 vbr->out_hdr.sector = 0; 134 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 135 break; 136 default: 137 /* We don't put anything else in the queue. */ 138 BUG(); 139 } 140 } 141 142 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr)); 143 144 /* 145 * If this is a packet command we need a couple of additional headers. 146 * Behind the normal outhdr we put a segment with the scsi command 147 * block, and before the normal inhdr we put the sense data and the 148 * inhdr with additional status information before the normal inhdr. 149 */ 150 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) 151 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len); 152 153 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out); 154 155 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) { 156 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, SCSI_SENSE_BUFFERSIZE); 157 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr, 158 sizeof(vbr->in_hdr)); 159 } 160 161 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status, 162 sizeof(vbr->status)); 163 164 if (num) { 165 if (rq_data_dir(vbr->req) == WRITE) { 166 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; 167 out += num; 168 } else { 169 vbr->out_hdr.type |= VIRTIO_BLK_T_IN; 170 in += num; 171 } 172 } 173 174 if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) { 175 mempool_free(vbr, vblk->pool); 176 return false; 177 } 178 179 list_add_tail(&vbr->list, &vblk->reqs); 180 return true; 181} 182 183static void do_virtblk_request(struct request_queue *q) 184{ 185 struct virtio_blk *vblk = q->queuedata; 186 struct request *req; 187 unsigned int issued = 0; 188 189 while ((req = blk_peek_request(q)) != NULL) { 190 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); 191 192 /* If this request fails, stop queue and wait for something to 193 finish to restart it. */ 194 if (!do_req(q, vblk, req)) { 195 blk_stop_queue(q); 196 break; 197 } 198 blk_start_request(req); 199 issued++; 200 } 201 202 if (issued) 203 virtqueue_kick(vblk->vq); 204} 205 206/* return id (s/n) string for *disk to *id_str 207 */ 208static int virtblk_get_id(struct gendisk *disk, char *id_str) 209{ 210 struct virtio_blk *vblk = disk->private_data; 211 struct request *req; 212 struct bio *bio; 213 int err; 214 215 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES, 216 GFP_KERNEL); 217 if (IS_ERR(bio)) 218 return PTR_ERR(bio); 219 220 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL); 221 if (IS_ERR(req)) { 222 bio_put(bio); 223 return PTR_ERR(req); 224 } 225 226 req->cmd_type = REQ_TYPE_SPECIAL; 227 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); 228 blk_put_request(req); 229 230 return err; 231} 232 233static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, 234 unsigned int cmd, unsigned long data) 235{ 236 struct gendisk *disk = bdev->bd_disk; 237 struct virtio_blk *vblk = disk->private_data; 238 239 /* 240 * Only allow the generic SCSI ioctls if the host can support it. 241 */ 242 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) 243 return -ENOTTY; 244 245 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, 246 (void __user *)data); 247} 248 249/* We provide getgeo only to please some old bootloader/partitioning tools */ 250static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) 251{ 252 struct virtio_blk *vblk = bd->bd_disk->private_data; 253 struct virtio_blk_geometry vgeo; 254 int err; 255 256 /* see if the host passed in geometry config */ 257 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY, 258 offsetof(struct virtio_blk_config, geometry), 259 &vgeo); 260 261 if (!err) { 262 geo->heads = vgeo.heads; 263 geo->sectors = vgeo.sectors; 264 geo->cylinders = vgeo.cylinders; 265 } else { 266 /* some standard values, similar to sd */ 267 geo->heads = 1 << 6; 268 geo->sectors = 1 << 5; 269 geo->cylinders = get_capacity(bd->bd_disk) >> 11; 270 } 271 return 0; 272} 273 274static const struct block_device_operations virtblk_fops = { 275 .ioctl = virtblk_ioctl, 276 .owner = THIS_MODULE, 277 .getgeo = virtblk_getgeo, 278}; 279 280static int index_to_minor(int index) 281{ 282 return index << PART_BITS; 283} 284 285static int minor_to_index(int minor) 286{ 287 return minor >> PART_BITS; 288} 289 290static ssize_t virtblk_serial_show(struct device *dev, 291 struct device_attribute *attr, char *buf) 292{ 293 struct gendisk *disk = dev_to_disk(dev); 294 int err; 295 296 /* sysfs gives us a PAGE_SIZE buffer */ 297 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); 298 299 buf[VIRTIO_BLK_ID_BYTES] = '\0'; 300 err = virtblk_get_id(disk, buf); 301 if (!err) 302 return strlen(buf); 303 304 if (err == -EIO) /* Unsupported? Make it empty. */ 305 return 0; 306 307 return err; 308} 309DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL); 310 311static void virtblk_config_changed_work(struct work_struct *work) 312{ 313 struct virtio_blk *vblk = 314 container_of(work, struct virtio_blk, config_work); 315 struct virtio_device *vdev = vblk->vdev; 316 struct request_queue *q = vblk->disk->queue; 317 char cap_str_2[10], cap_str_10[10]; 318 u64 capacity, size; 319 320 /* Host must always specify the capacity. */ 321 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), 322 &capacity, sizeof(capacity)); 323 324 /* If capacity is too big, truncate with warning. */ 325 if ((sector_t)capacity != capacity) { 326 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", 327 (unsigned long long)capacity); 328 capacity = (sector_t)-1; 329 } 330 331 size = capacity * queue_logical_block_size(q); 332 string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); 333 string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); 334 335 dev_notice(&vdev->dev, 336 "new size: %llu %d-byte logical blocks (%s/%s)\n", 337 (unsigned long long)capacity, 338 queue_logical_block_size(q), 339 cap_str_10, cap_str_2); 340 341 set_capacity(vblk->disk, capacity); 342} 343 344static void virtblk_config_changed(struct virtio_device *vdev) 345{ 346 struct virtio_blk *vblk = vdev->priv; 347 348 queue_work(virtblk_wq, &vblk->config_work); 349} 350 351static int __devinit virtblk_probe(struct virtio_device *vdev) 352{ 353 struct virtio_blk *vblk; 354 struct request_queue *q; 355 int err, index; 356 u64 cap; 357 u32 v, blk_size, sg_elems, opt_io_size; 358 u16 min_io_size; 359 u8 physical_block_exp, alignment_offset; 360 361 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS), 362 GFP_KERNEL); 363 if (err < 0) 364 goto out; 365 index = err; 366 367 /* We need to know how many segments before we allocate. */ 368 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX, 369 offsetof(struct virtio_blk_config, seg_max), 370 &sg_elems); 371 372 /* We need at least one SG element, whatever they say. */ 373 if (err || !sg_elems) 374 sg_elems = 1; 375 376 /* We need an extra sg elements at head and tail. */ 377 sg_elems += 2; 378 vdev->priv = vblk = kmalloc(sizeof(*vblk) + 379 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL); 380 if (!vblk) { 381 err = -ENOMEM; 382 goto out_free_index; 383 } 384 385 INIT_LIST_HEAD(&vblk->reqs); 386 spin_lock_init(&vblk->lock); 387 vblk->vdev = vdev; 388 vblk->sg_elems = sg_elems; 389 sg_init_table(vblk->sg, vblk->sg_elems); 390 INIT_WORK(&vblk->config_work, virtblk_config_changed_work); 391 392 /* We expect one virtqueue, for output. */ 393 vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests"); 394 if (IS_ERR(vblk->vq)) { 395 err = PTR_ERR(vblk->vq); 396 goto out_free_vblk; 397 } 398 399 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req)); 400 if (!vblk->pool) { 401 err = -ENOMEM; 402 goto out_free_vq; 403 } 404 405 /* FIXME: How many partitions? How long is a piece of string? */ 406 vblk->disk = alloc_disk(1 << PART_BITS); 407 if (!vblk->disk) { 408 err = -ENOMEM; 409 goto out_mempool; 410 } 411 412 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock); 413 if (!q) { 414 err = -ENOMEM; 415 goto out_put_disk; 416 } 417 418 q->queuedata = vblk; 419 420 if (index < 26) { 421 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26); 422 } else if (index < (26 + 1) * 26) { 423 sprintf(vblk->disk->disk_name, "vd%c%c", 424 'a' + index / 26 - 1, 'a' + index % 26); 425 } else { 426 const unsigned int m1 = (index / 26 - 1) / 26 - 1; 427 const unsigned int m2 = (index / 26 - 1) % 26; 428 const unsigned int m3 = index % 26; 429 sprintf(vblk->disk->disk_name, "vd%c%c%c", 430 'a' + m1, 'a' + m2, 'a' + m3); 431 } 432 433 vblk->disk->major = major; 434 vblk->disk->first_minor = index_to_minor(index); 435 vblk->disk->private_data = vblk; 436 vblk->disk->fops = &virtblk_fops; 437 vblk->disk->driverfs_dev = &vdev->dev; 438 vblk->index = index; 439 440 /* configure queue flush support */ 441 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH)) 442 blk_queue_flush(q, REQ_FLUSH); 443 444 /* If disk is read-only in the host, the guest should obey */ 445 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) 446 set_disk_ro(vblk->disk, 1); 447 448 /* Host must always specify the capacity. */ 449 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), 450 &cap, sizeof(cap)); 451 452 /* If capacity is too big, truncate with warning. */ 453 if ((sector_t)cap != cap) { 454 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", 455 (unsigned long long)cap); 456 cap = (sector_t)-1; 457 } 458 set_capacity(vblk->disk, cap); 459 460 /* We can handle whatever the host told us to handle. */ 461 blk_queue_max_segments(q, vblk->sg_elems-2); 462 463 /* No need to bounce any requests */ 464 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); 465 466 /* No real sector limit. */ 467 blk_queue_max_hw_sectors(q, -1U); 468 469 /* Host can optionally specify maximum segment size and number of 470 * segments. */ 471 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX, 472 offsetof(struct virtio_blk_config, size_max), 473 &v); 474 if (!err) 475 blk_queue_max_segment_size(q, v); 476 else 477 blk_queue_max_segment_size(q, -1U); 478 479 /* Host can optionally specify the block size of the device */ 480 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE, 481 offsetof(struct virtio_blk_config, blk_size), 482 &blk_size); 483 if (!err) 484 blk_queue_logical_block_size(q, blk_size); 485 else 486 blk_size = queue_logical_block_size(q); 487 488 /* Use topology information if available */ 489 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 490 offsetof(struct virtio_blk_config, physical_block_exp), 491 &physical_block_exp); 492 if (!err && physical_block_exp) 493 blk_queue_physical_block_size(q, 494 blk_size * (1 << physical_block_exp)); 495 496 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 497 offsetof(struct virtio_blk_config, alignment_offset), 498 &alignment_offset); 499 if (!err && alignment_offset) 500 blk_queue_alignment_offset(q, blk_size * alignment_offset); 501 502 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 503 offsetof(struct virtio_blk_config, min_io_size), 504 &min_io_size); 505 if (!err && min_io_size) 506 blk_queue_io_min(q, blk_size * min_io_size); 507 508 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 509 offsetof(struct virtio_blk_config, opt_io_size), 510 &opt_io_size); 511 if (!err && opt_io_size) 512 blk_queue_io_opt(q, blk_size * opt_io_size); 513 514 515 add_disk(vblk->disk); 516 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial); 517 if (err) 518 goto out_del_disk; 519 520 return 0; 521 522out_del_disk: 523 del_gendisk(vblk->disk); 524 blk_cleanup_queue(vblk->disk->queue); 525out_put_disk: 526 put_disk(vblk->disk); 527out_mempool: 528 mempool_destroy(vblk->pool); 529out_free_vq: 530 vdev->config->del_vqs(vdev); 531out_free_vblk: 532 kfree(vblk); 533out_free_index: 534 ida_simple_remove(&vd_index_ida, index); 535out: 536 return err; 537} 538 539static void __devexit virtblk_remove(struct virtio_device *vdev) 540{ 541 struct virtio_blk *vblk = vdev->priv; 542 int index = vblk->index; 543 544 flush_work(&vblk->config_work); 545 546 /* Nothing should be pending. */ 547 BUG_ON(!list_empty(&vblk->reqs)); 548 549 /* Stop all the virtqueues. */ 550 vdev->config->reset(vdev); 551 552 del_gendisk(vblk->disk); 553 blk_cleanup_queue(vblk->disk->queue); 554 put_disk(vblk->disk); 555 mempool_destroy(vblk->pool); 556 vdev->config->del_vqs(vdev); 557 kfree(vblk); 558 ida_simple_remove(&vd_index_ida, index); 559} 560 561static const struct virtio_device_id id_table[] = { 562 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, 563 { 0 }, 564}; 565 566static unsigned int features[] = { 567 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, 568 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI, 569 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY 570}; 571 572/* 573 * virtio_blk causes spurious section mismatch warning by 574 * simultaneously referring to a __devinit and a __devexit function. 575 * Use __refdata to avoid this warning. 576 */ 577static struct virtio_driver __refdata virtio_blk = { 578 .feature_table = features, 579 .feature_table_size = ARRAY_SIZE(features), 580 .driver.name = KBUILD_MODNAME, 581 .driver.owner = THIS_MODULE, 582 .id_table = id_table, 583 .probe = virtblk_probe, 584 .remove = __devexit_p(virtblk_remove), 585 .config_changed = virtblk_config_changed, 586}; 587 588static int __init init(void) 589{ 590 int error; 591 592 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0); 593 if (!virtblk_wq) 594 return -ENOMEM; 595 596 major = register_blkdev(0, "virtblk"); 597 if (major < 0) { 598 error = major; 599 goto out_destroy_workqueue; 600 } 601 602 error = register_virtio_driver(&virtio_blk); 603 if (error) 604 goto out_unregister_blkdev; 605 return 0; 606 607out_unregister_blkdev: 608 unregister_blkdev(major, "virtblk"); 609out_destroy_workqueue: 610 destroy_workqueue(virtblk_wq); 611 return error; 612} 613 614static void __exit fini(void) 615{ 616 unregister_blkdev(major, "virtblk"); 617 unregister_virtio_driver(&virtio_blk); 618 destroy_workqueue(virtblk_wq); 619} 620module_init(init); 621module_exit(fini); 622 623MODULE_DEVICE_TABLE(virtio, id_table); 624MODULE_DESCRIPTION("Virtio block driver"); 625MODULE_LICENSE("GPL"); 626