virtio_blk.c revision 577ebb374c78314ac4617242f509e2f5e7156649
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/module.h> 7#include <linux/mutex.h> 8#include <linux/virtio.h> 9#include <linux/virtio_blk.h> 10#include <linux/scatterlist.h> 11#include <linux/string_helpers.h> 12#include <scsi/scsi_cmnd.h> 13#include <linux/idr.h> 14 15#define PART_BITS 4 16 17static int major; 18static DEFINE_IDA(vd_index_ida); 19 20struct workqueue_struct *virtblk_wq; 21 22struct virtio_blk 23{ 24 spinlock_t lock; 25 26 struct virtio_device *vdev; 27 struct virtqueue *vq; 28 29 /* The disk structure for the kernel. */ 30 struct gendisk *disk; 31 32 /* Request tracking. */ 33 struct list_head reqs; 34 35 mempool_t *pool; 36 37 /* Process context for config space updates */ 38 struct work_struct config_work; 39 40 /* Lock for config space updates */ 41 struct mutex config_lock; 42 43 /* enable config space updates */ 44 bool config_enable; 45 46 /* What host tells us, plus 2 for header & tailer. */ 47 unsigned int sg_elems; 48 49 /* Ida index - used to track minor number allocations. */ 50 int index; 51 52 /* Scatterlist: can be too big for stack. */ 53 struct scatterlist sg[/*sg_elems*/]; 54}; 55 56struct virtblk_req 57{ 58 struct list_head list; 59 struct request *req; 60 struct virtio_blk_outhdr out_hdr; 61 struct virtio_scsi_inhdr in_hdr; 62 u8 status; 63}; 64 65static void blk_done(struct virtqueue *vq) 66{ 67 struct virtio_blk *vblk = vq->vdev->priv; 68 struct virtblk_req *vbr; 69 unsigned int len; 70 unsigned long flags; 71 72 spin_lock_irqsave(&vblk->lock, flags); 73 while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) { 74 int error; 75 76 switch (vbr->status) { 77 case VIRTIO_BLK_S_OK: 78 error = 0; 79 break; 80 case VIRTIO_BLK_S_UNSUPP: 81 error = -ENOTTY; 82 break; 83 default: 84 error = -EIO; 85 break; 86 } 87 88 switch (vbr->req->cmd_type) { 89 case REQ_TYPE_BLOCK_PC: 90 vbr->req->resid_len = vbr->in_hdr.residual; 91 vbr->req->sense_len = vbr->in_hdr.sense_len; 92 vbr->req->errors = vbr->in_hdr.errors; 93 break; 94 case REQ_TYPE_SPECIAL: 95 vbr->req->errors = (error != 0); 96 break; 97 default: 98 break; 99 } 100 101 __blk_end_request_all(vbr->req, error); 102 list_del(&vbr->list); 103 mempool_free(vbr, vblk->pool); 104 } 105 /* In case queue is stopped waiting for more buffers. */ 106 blk_start_queue(vblk->disk->queue); 107 spin_unlock_irqrestore(&vblk->lock, flags); 108} 109 110static bool do_req(struct request_queue *q, struct virtio_blk *vblk, 111 struct request *req) 112{ 113 unsigned long num, out = 0, in = 0; 114 struct virtblk_req *vbr; 115 116 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC); 117 if (!vbr) 118 /* When another request finishes we'll try again. */ 119 return false; 120 121 vbr->req = req; 122 123 if (req->cmd_flags & REQ_FLUSH) { 124 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH; 125 vbr->out_hdr.sector = 0; 126 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 127 } else { 128 switch (req->cmd_type) { 129 case REQ_TYPE_FS: 130 vbr->out_hdr.type = 0; 131 vbr->out_hdr.sector = blk_rq_pos(vbr->req); 132 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 133 break; 134 case REQ_TYPE_BLOCK_PC: 135 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; 136 vbr->out_hdr.sector = 0; 137 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 138 break; 139 case REQ_TYPE_SPECIAL: 140 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID; 141 vbr->out_hdr.sector = 0; 142 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 143 break; 144 default: 145 /* We don't put anything else in the queue. */ 146 BUG(); 147 } 148 } 149 150 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr)); 151 152 /* 153 * If this is a packet command we need a couple of additional headers. 154 * Behind the normal outhdr we put a segment with the scsi command 155 * block, and before the normal inhdr we put the sense data and the 156 * inhdr with additional status information before the normal inhdr. 157 */ 158 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) 159 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len); 160 161 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out); 162 163 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) { 164 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, SCSI_SENSE_BUFFERSIZE); 165 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr, 166 sizeof(vbr->in_hdr)); 167 } 168 169 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status, 170 sizeof(vbr->status)); 171 172 if (num) { 173 if (rq_data_dir(vbr->req) == WRITE) { 174 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; 175 out += num; 176 } else { 177 vbr->out_hdr.type |= VIRTIO_BLK_T_IN; 178 in += num; 179 } 180 } 181 182 if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr, GFP_ATOMIC)<0) { 183 mempool_free(vbr, vblk->pool); 184 return false; 185 } 186 187 list_add_tail(&vbr->list, &vblk->reqs); 188 return true; 189} 190 191static void do_virtblk_request(struct request_queue *q) 192{ 193 struct virtio_blk *vblk = q->queuedata; 194 struct request *req; 195 unsigned int issued = 0; 196 197 while ((req = blk_peek_request(q)) != NULL) { 198 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); 199 200 /* If this request fails, stop queue and wait for something to 201 finish to restart it. */ 202 if (!do_req(q, vblk, req)) { 203 blk_stop_queue(q); 204 break; 205 } 206 blk_start_request(req); 207 issued++; 208 } 209 210 if (issued) 211 virtqueue_kick(vblk->vq); 212} 213 214/* return id (s/n) string for *disk to *id_str 215 */ 216static int virtblk_get_id(struct gendisk *disk, char *id_str) 217{ 218 struct virtio_blk *vblk = disk->private_data; 219 struct request *req; 220 struct bio *bio; 221 int err; 222 223 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES, 224 GFP_KERNEL); 225 if (IS_ERR(bio)) 226 return PTR_ERR(bio); 227 228 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL); 229 if (IS_ERR(req)) { 230 bio_put(bio); 231 return PTR_ERR(req); 232 } 233 234 req->cmd_type = REQ_TYPE_SPECIAL; 235 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); 236 blk_put_request(req); 237 238 return err; 239} 240 241static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, 242 unsigned int cmd, unsigned long data) 243{ 244 struct gendisk *disk = bdev->bd_disk; 245 struct virtio_blk *vblk = disk->private_data; 246 247 /* 248 * Only allow the generic SCSI ioctls if the host can support it. 249 */ 250 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) 251 return -ENOTTY; 252 253 return scsi_cmd_blk_ioctl(bdev, mode, cmd, 254 (void __user *)data); 255} 256 257/* We provide getgeo only to please some old bootloader/partitioning tools */ 258static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) 259{ 260 struct virtio_blk *vblk = bd->bd_disk->private_data; 261 struct virtio_blk_geometry vgeo; 262 int err; 263 264 /* see if the host passed in geometry config */ 265 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY, 266 offsetof(struct virtio_blk_config, geometry), 267 &vgeo); 268 269 if (!err) { 270 geo->heads = vgeo.heads; 271 geo->sectors = vgeo.sectors; 272 geo->cylinders = vgeo.cylinders; 273 } else { 274 /* some standard values, similar to sd */ 275 geo->heads = 1 << 6; 276 geo->sectors = 1 << 5; 277 geo->cylinders = get_capacity(bd->bd_disk) >> 11; 278 } 279 return 0; 280} 281 282static const struct block_device_operations virtblk_fops = { 283 .ioctl = virtblk_ioctl, 284 .owner = THIS_MODULE, 285 .getgeo = virtblk_getgeo, 286}; 287 288static int index_to_minor(int index) 289{ 290 return index << PART_BITS; 291} 292 293static int minor_to_index(int minor) 294{ 295 return minor >> PART_BITS; 296} 297 298static ssize_t virtblk_serial_show(struct device *dev, 299 struct device_attribute *attr, char *buf) 300{ 301 struct gendisk *disk = dev_to_disk(dev); 302 int err; 303 304 /* sysfs gives us a PAGE_SIZE buffer */ 305 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); 306 307 buf[VIRTIO_BLK_ID_BYTES] = '\0'; 308 err = virtblk_get_id(disk, buf); 309 if (!err) 310 return strlen(buf); 311 312 if (err == -EIO) /* Unsupported? Make it empty. */ 313 return 0; 314 315 return err; 316} 317DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL); 318 319static void virtblk_config_changed_work(struct work_struct *work) 320{ 321 struct virtio_blk *vblk = 322 container_of(work, struct virtio_blk, config_work); 323 struct virtio_device *vdev = vblk->vdev; 324 struct request_queue *q = vblk->disk->queue; 325 char cap_str_2[10], cap_str_10[10]; 326 u64 capacity, size; 327 328 mutex_lock(&vblk->config_lock); 329 if (!vblk->config_enable) 330 goto done; 331 332 /* Host must always specify the capacity. */ 333 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), 334 &capacity, sizeof(capacity)); 335 336 /* If capacity is too big, truncate with warning. */ 337 if ((sector_t)capacity != capacity) { 338 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", 339 (unsigned long long)capacity); 340 capacity = (sector_t)-1; 341 } 342 343 size = capacity * queue_logical_block_size(q); 344 string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); 345 string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); 346 347 dev_notice(&vdev->dev, 348 "new size: %llu %d-byte logical blocks (%s/%s)\n", 349 (unsigned long long)capacity, 350 queue_logical_block_size(q), 351 cap_str_10, cap_str_2); 352 353 set_capacity(vblk->disk, capacity); 354done: 355 mutex_unlock(&vblk->config_lock); 356} 357 358static void virtblk_config_changed(struct virtio_device *vdev) 359{ 360 struct virtio_blk *vblk = vdev->priv; 361 362 queue_work(virtblk_wq, &vblk->config_work); 363} 364 365static int init_vq(struct virtio_blk *vblk) 366{ 367 int err = 0; 368 369 /* We expect one virtqueue, for output. */ 370 vblk->vq = virtio_find_single_vq(vblk->vdev, blk_done, "requests"); 371 if (IS_ERR(vblk->vq)) 372 err = PTR_ERR(vblk->vq); 373 374 return err; 375} 376 377static int __devinit virtblk_probe(struct virtio_device *vdev) 378{ 379 struct virtio_blk *vblk; 380 struct request_queue *q; 381 int err, index; 382 u64 cap; 383 u32 v, blk_size, sg_elems, opt_io_size; 384 u16 min_io_size; 385 u8 physical_block_exp, alignment_offset; 386 387 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS), 388 GFP_KERNEL); 389 if (err < 0) 390 goto out; 391 index = err; 392 393 /* We need to know how many segments before we allocate. */ 394 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX, 395 offsetof(struct virtio_blk_config, seg_max), 396 &sg_elems); 397 398 /* We need at least one SG element, whatever they say. */ 399 if (err || !sg_elems) 400 sg_elems = 1; 401 402 /* We need an extra sg elements at head and tail. */ 403 sg_elems += 2; 404 vdev->priv = vblk = kmalloc(sizeof(*vblk) + 405 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL); 406 if (!vblk) { 407 err = -ENOMEM; 408 goto out_free_index; 409 } 410 411 INIT_LIST_HEAD(&vblk->reqs); 412 spin_lock_init(&vblk->lock); 413 vblk->vdev = vdev; 414 vblk->sg_elems = sg_elems; 415 sg_init_table(vblk->sg, vblk->sg_elems); 416 mutex_init(&vblk->config_lock); 417 INIT_WORK(&vblk->config_work, virtblk_config_changed_work); 418 vblk->config_enable = true; 419 420 err = init_vq(vblk); 421 if (err) 422 goto out_free_vblk; 423 424 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req)); 425 if (!vblk->pool) { 426 err = -ENOMEM; 427 goto out_free_vq; 428 } 429 430 /* FIXME: How many partitions? How long is a piece of string? */ 431 vblk->disk = alloc_disk(1 << PART_BITS); 432 if (!vblk->disk) { 433 err = -ENOMEM; 434 goto out_mempool; 435 } 436 437 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock); 438 if (!q) { 439 err = -ENOMEM; 440 goto out_put_disk; 441 } 442 443 q->queuedata = vblk; 444 445 if (index < 26) { 446 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26); 447 } else if (index < (26 + 1) * 26) { 448 sprintf(vblk->disk->disk_name, "vd%c%c", 449 'a' + index / 26 - 1, 'a' + index % 26); 450 } else { 451 const unsigned int m1 = (index / 26 - 1) / 26 - 1; 452 const unsigned int m2 = (index / 26 - 1) % 26; 453 const unsigned int m3 = index % 26; 454 sprintf(vblk->disk->disk_name, "vd%c%c%c", 455 'a' + m1, 'a' + m2, 'a' + m3); 456 } 457 458 vblk->disk->major = major; 459 vblk->disk->first_minor = index_to_minor(index); 460 vblk->disk->private_data = vblk; 461 vblk->disk->fops = &virtblk_fops; 462 vblk->disk->driverfs_dev = &vdev->dev; 463 vblk->index = index; 464 465 /* configure queue flush support */ 466 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH)) 467 blk_queue_flush(q, REQ_FLUSH); 468 469 /* If disk is read-only in the host, the guest should obey */ 470 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) 471 set_disk_ro(vblk->disk, 1); 472 473 /* Host must always specify the capacity. */ 474 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), 475 &cap, sizeof(cap)); 476 477 /* If capacity is too big, truncate with warning. */ 478 if ((sector_t)cap != cap) { 479 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", 480 (unsigned long long)cap); 481 cap = (sector_t)-1; 482 } 483 set_capacity(vblk->disk, cap); 484 485 /* We can handle whatever the host told us to handle. */ 486 blk_queue_max_segments(q, vblk->sg_elems-2); 487 488 /* No need to bounce any requests */ 489 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); 490 491 /* No real sector limit. */ 492 blk_queue_max_hw_sectors(q, -1U); 493 494 /* Host can optionally specify maximum segment size and number of 495 * segments. */ 496 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX, 497 offsetof(struct virtio_blk_config, size_max), 498 &v); 499 if (!err) 500 blk_queue_max_segment_size(q, v); 501 else 502 blk_queue_max_segment_size(q, -1U); 503 504 /* Host can optionally specify the block size of the device */ 505 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE, 506 offsetof(struct virtio_blk_config, blk_size), 507 &blk_size); 508 if (!err) 509 blk_queue_logical_block_size(q, blk_size); 510 else 511 blk_size = queue_logical_block_size(q); 512 513 /* Use topology information if available */ 514 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 515 offsetof(struct virtio_blk_config, physical_block_exp), 516 &physical_block_exp); 517 if (!err && physical_block_exp) 518 blk_queue_physical_block_size(q, 519 blk_size * (1 << physical_block_exp)); 520 521 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 522 offsetof(struct virtio_blk_config, alignment_offset), 523 &alignment_offset); 524 if (!err && alignment_offset) 525 blk_queue_alignment_offset(q, blk_size * alignment_offset); 526 527 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 528 offsetof(struct virtio_blk_config, min_io_size), 529 &min_io_size); 530 if (!err && min_io_size) 531 blk_queue_io_min(q, blk_size * min_io_size); 532 533 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 534 offsetof(struct virtio_blk_config, opt_io_size), 535 &opt_io_size); 536 if (!err && opt_io_size) 537 blk_queue_io_opt(q, blk_size * opt_io_size); 538 539 540 add_disk(vblk->disk); 541 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial); 542 if (err) 543 goto out_del_disk; 544 545 return 0; 546 547out_del_disk: 548 del_gendisk(vblk->disk); 549 blk_cleanup_queue(vblk->disk->queue); 550out_put_disk: 551 put_disk(vblk->disk); 552out_mempool: 553 mempool_destroy(vblk->pool); 554out_free_vq: 555 vdev->config->del_vqs(vdev); 556out_free_vblk: 557 kfree(vblk); 558out_free_index: 559 ida_simple_remove(&vd_index_ida, index); 560out: 561 return err; 562} 563 564static void __devexit virtblk_remove(struct virtio_device *vdev) 565{ 566 struct virtio_blk *vblk = vdev->priv; 567 int index = vblk->index; 568 569 /* Prevent config work handler from accessing the device. */ 570 mutex_lock(&vblk->config_lock); 571 vblk->config_enable = false; 572 mutex_unlock(&vblk->config_lock); 573 574 /* Nothing should be pending. */ 575 BUG_ON(!list_empty(&vblk->reqs)); 576 577 /* Stop all the virtqueues. */ 578 vdev->config->reset(vdev); 579 580 flush_work(&vblk->config_work); 581 582 del_gendisk(vblk->disk); 583 blk_cleanup_queue(vblk->disk->queue); 584 put_disk(vblk->disk); 585 mempool_destroy(vblk->pool); 586 vdev->config->del_vqs(vdev); 587 kfree(vblk); 588 ida_simple_remove(&vd_index_ida, index); 589} 590 591#ifdef CONFIG_PM 592static int virtblk_freeze(struct virtio_device *vdev) 593{ 594 struct virtio_blk *vblk = vdev->priv; 595 596 /* Ensure we don't receive any more interrupts */ 597 vdev->config->reset(vdev); 598 599 /* Prevent config work handler from accessing the device. */ 600 mutex_lock(&vblk->config_lock); 601 vblk->config_enable = false; 602 mutex_unlock(&vblk->config_lock); 603 604 flush_work(&vblk->config_work); 605 606 spin_lock_irq(vblk->disk->queue->queue_lock); 607 blk_stop_queue(vblk->disk->queue); 608 spin_unlock_irq(vblk->disk->queue->queue_lock); 609 blk_sync_queue(vblk->disk->queue); 610 611 vdev->config->del_vqs(vdev); 612 return 0; 613} 614 615static int virtblk_restore(struct virtio_device *vdev) 616{ 617 struct virtio_blk *vblk = vdev->priv; 618 int ret; 619 620 vblk->config_enable = true; 621 ret = init_vq(vdev->priv); 622 if (!ret) { 623 spin_lock_irq(vblk->disk->queue->queue_lock); 624 blk_start_queue(vblk->disk->queue); 625 spin_unlock_irq(vblk->disk->queue->queue_lock); 626 } 627 return ret; 628} 629#endif 630 631static const struct virtio_device_id id_table[] = { 632 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, 633 { 0 }, 634}; 635 636static unsigned int features[] = { 637 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, 638 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI, 639 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY 640}; 641 642/* 643 * virtio_blk causes spurious section mismatch warning by 644 * simultaneously referring to a __devinit and a __devexit function. 645 * Use __refdata to avoid this warning. 646 */ 647static struct virtio_driver __refdata virtio_blk = { 648 .feature_table = features, 649 .feature_table_size = ARRAY_SIZE(features), 650 .driver.name = KBUILD_MODNAME, 651 .driver.owner = THIS_MODULE, 652 .id_table = id_table, 653 .probe = virtblk_probe, 654 .remove = __devexit_p(virtblk_remove), 655 .config_changed = virtblk_config_changed, 656#ifdef CONFIG_PM 657 .freeze = virtblk_freeze, 658 .restore = virtblk_restore, 659#endif 660}; 661 662static int __init init(void) 663{ 664 int error; 665 666 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0); 667 if (!virtblk_wq) 668 return -ENOMEM; 669 670 major = register_blkdev(0, "virtblk"); 671 if (major < 0) { 672 error = major; 673 goto out_destroy_workqueue; 674 } 675 676 error = register_virtio_driver(&virtio_blk); 677 if (error) 678 goto out_unregister_blkdev; 679 return 0; 680 681out_unregister_blkdev: 682 unregister_blkdev(major, "virtblk"); 683out_destroy_workqueue: 684 destroy_workqueue(virtblk_wq); 685 return error; 686} 687 688static void __exit fini(void) 689{ 690 unregister_blkdev(major, "virtblk"); 691 unregister_virtio_driver(&virtio_blk); 692 destroy_workqueue(virtblk_wq); 693} 694module_init(init); 695module_exit(fini); 696 697MODULE_DEVICE_TABLE(virtio, id_table); 698MODULE_DESCRIPTION("Virtio block driver"); 699MODULE_LICENSE("GPL"); 700