virtio_blk.c revision 10bc310c27af1ed358e62351e7ac1d0110c3da27
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 12static int major, index; 13 14struct 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 36struct 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 45static 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 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 if (blk_special_request(vbr->req)) 74 vbr->req->errors = (error != 0); 75 76 __blk_end_request_all(vbr->req, error); 77 list_del(&vbr->list); 78 mempool_free(vbr, vblk->pool); 79 } 80 /* In case queue is stopped waiting for more buffers. */ 81 blk_start_queue(vblk->disk->queue); 82 spin_unlock_irqrestore(&vblk->lock, flags); 83} 84 85static bool do_req(struct request_queue *q, struct virtio_blk *vblk, 86 struct request *req) 87{ 88 unsigned long num, out = 0, in = 0; 89 struct virtblk_req *vbr; 90 91 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC); 92 if (!vbr) 93 /* When another request finishes we'll try again. */ 94 return false; 95 96 vbr->req = req; 97 switch (req->cmd_type) { 98 case REQ_TYPE_FS: 99 vbr->out_hdr.type = 0; 100 vbr->out_hdr.sector = blk_rq_pos(vbr->req); 101 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 102 break; 103 case REQ_TYPE_BLOCK_PC: 104 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; 105 vbr->out_hdr.sector = 0; 106 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 107 break; 108 case REQ_TYPE_SPECIAL: 109 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID; 110 vbr->out_hdr.sector = 0; 111 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 112 break; 113 case REQ_TYPE_LINUX_BLOCK: 114 if (req->cmd[0] == REQ_LB_OP_FLUSH) { 115 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH; 116 vbr->out_hdr.sector = 0; 117 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); 118 break; 119 } 120 /*FALLTHRU*/ 121 default: 122 /* We don't put anything else in the queue. */ 123 BUG(); 124 } 125 126 if (blk_barrier_rq(vbr->req)) 127 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER; 128 129 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr)); 130 131 /* 132 * If this is a packet command we need a couple of additional headers. 133 * Behind the normal outhdr we put a segment with the scsi command 134 * block, and before the normal inhdr we put the sense data and the 135 * inhdr with additional status information before the normal inhdr. 136 */ 137 if (blk_pc_request(vbr->req)) 138 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len); 139 140 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out); 141 142 if (blk_pc_request(vbr->req)) { 143 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96); 144 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr, 145 sizeof(vbr->in_hdr)); 146 } 147 148 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status, 149 sizeof(vbr->status)); 150 151 if (num) { 152 if (rq_data_dir(vbr->req) == WRITE) { 153 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; 154 out += num; 155 } else { 156 vbr->out_hdr.type |= VIRTIO_BLK_T_IN; 157 in += num; 158 } 159 } 160 161 if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) { 162 mempool_free(vbr, vblk->pool); 163 return false; 164 } 165 166 list_add_tail(&vbr->list, &vblk->reqs); 167 return true; 168} 169 170static void do_virtblk_request(struct request_queue *q) 171{ 172 struct virtio_blk *vblk = q->queuedata; 173 struct request *req; 174 unsigned int issued = 0; 175 176 while ((req = blk_peek_request(q)) != NULL) { 177 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); 178 179 /* If this request fails, stop queue and wait for something to 180 finish to restart it. */ 181 if (!do_req(q, vblk, req)) { 182 blk_stop_queue(q); 183 break; 184 } 185 blk_start_request(req); 186 issued++; 187 } 188 189 if (issued) 190 virtqueue_kick(vblk->vq); 191} 192 193static void virtblk_prepare_flush(struct request_queue *q, struct request *req) 194{ 195 req->cmd_type = REQ_TYPE_LINUX_BLOCK; 196 req->cmd[0] = REQ_LB_OP_FLUSH; 197} 198 199/* return id (s/n) string for *disk to *id_str 200 */ 201static int virtblk_get_id(struct gendisk *disk, char *id_str) 202{ 203 struct virtio_blk *vblk = disk->private_data; 204 struct request *req; 205 struct bio *bio; 206 207 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES, 208 GFP_KERNEL); 209 if (IS_ERR(bio)) 210 return PTR_ERR(bio); 211 212 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL); 213 if (IS_ERR(req)) { 214 bio_put(bio); 215 return PTR_ERR(req); 216 } 217 218 req->cmd_type = REQ_TYPE_SPECIAL; 219 return blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); 220} 221 222static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, 223 unsigned cmd, unsigned long data) 224{ 225 struct gendisk *disk = bdev->bd_disk; 226 struct virtio_blk *vblk = disk->private_data; 227 228 if (cmd == 0x56424944) { /* 'VBID' */ 229 void __user *usr_data = (void __user *)data; 230 char id_str[VIRTIO_BLK_ID_BYTES]; 231 int err; 232 233 err = virtblk_get_id(disk, id_str); 234 if (!err && copy_to_user(usr_data, id_str, VIRTIO_BLK_ID_BYTES)) 235 err = -EFAULT; 236 return err; 237 } 238 /* 239 * Only allow the generic SCSI ioctls if the host can support it. 240 */ 241 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) 242 return -ENOTTY; 243 244 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, 245 (void __user *)data); 246} 247 248/* We provide getgeo only to please some old bootloader/partitioning tools */ 249static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) 250{ 251 struct virtio_blk *vblk = bd->bd_disk->private_data; 252 struct virtio_blk_geometry vgeo; 253 int err; 254 255 /* see if the host passed in geometry config */ 256 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY, 257 offsetof(struct virtio_blk_config, geometry), 258 &vgeo); 259 260 if (!err) { 261 geo->heads = vgeo.heads; 262 geo->sectors = vgeo.sectors; 263 geo->cylinders = vgeo.cylinders; 264 } else { 265 /* some standard values, similar to sd */ 266 geo->heads = 1 << 6; 267 geo->sectors = 1 << 5; 268 geo->cylinders = get_capacity(bd->bd_disk) >> 11; 269 } 270 return 0; 271} 272 273static const struct block_device_operations virtblk_fops = { 274 .locked_ioctl = virtblk_ioctl, 275 .owner = THIS_MODULE, 276 .getgeo = virtblk_getgeo, 277}; 278 279static int index_to_minor(int index) 280{ 281 return index << PART_BITS; 282} 283 284static int __devinit virtblk_probe(struct virtio_device *vdev) 285{ 286 struct virtio_blk *vblk; 287 struct request_queue *q; 288 int err; 289 u64 cap; 290 u32 v, blk_size, sg_elems, opt_io_size; 291 u16 min_io_size; 292 u8 physical_block_exp, alignment_offset; 293 294 if (index_to_minor(index) >= 1 << MINORBITS) 295 return -ENOSPC; 296 297 /* We need to know how many segments before we allocate. */ 298 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX, 299 offsetof(struct virtio_blk_config, seg_max), 300 &sg_elems); 301 302 /* We need at least one SG element, whatever they say. */ 303 if (err || !sg_elems) 304 sg_elems = 1; 305 306 /* We need an extra sg elements at head and tail. */ 307 sg_elems += 2; 308 vdev->priv = vblk = kmalloc(sizeof(*vblk) + 309 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL); 310 if (!vblk) { 311 err = -ENOMEM; 312 goto out; 313 } 314 315 INIT_LIST_HEAD(&vblk->reqs); 316 spin_lock_init(&vblk->lock); 317 vblk->vdev = vdev; 318 vblk->sg_elems = sg_elems; 319 sg_init_table(vblk->sg, vblk->sg_elems); 320 321 /* We expect one virtqueue, for output. */ 322 vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests"); 323 if (IS_ERR(vblk->vq)) { 324 err = PTR_ERR(vblk->vq); 325 goto out_free_vblk; 326 } 327 328 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req)); 329 if (!vblk->pool) { 330 err = -ENOMEM; 331 goto out_free_vq; 332 } 333 334 /* FIXME: How many partitions? How long is a piece of string? */ 335 vblk->disk = alloc_disk(1 << PART_BITS); 336 if (!vblk->disk) { 337 err = -ENOMEM; 338 goto out_mempool; 339 } 340 341 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock); 342 if (!q) { 343 err = -ENOMEM; 344 goto out_put_disk; 345 } 346 347 q->queuedata = vblk; 348 349 if (index < 26) { 350 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26); 351 } else if (index < (26 + 1) * 26) { 352 sprintf(vblk->disk->disk_name, "vd%c%c", 353 'a' + index / 26 - 1, 'a' + index % 26); 354 } else { 355 const unsigned int m1 = (index / 26 - 1) / 26 - 1; 356 const unsigned int m2 = (index / 26 - 1) % 26; 357 const unsigned int m3 = index % 26; 358 sprintf(vblk->disk->disk_name, "vd%c%c%c", 359 'a' + m1, 'a' + m2, 'a' + m3); 360 } 361 362 vblk->disk->major = major; 363 vblk->disk->first_minor = index_to_minor(index); 364 vblk->disk->private_data = vblk; 365 vblk->disk->fops = &virtblk_fops; 366 vblk->disk->driverfs_dev = &vdev->dev; 367 index++; 368 369 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH)) { 370 /* 371 * If the FLUSH feature is supported we do have support for 372 * flushing a volatile write cache on the host. Use that 373 * to implement write barrier support. 374 */ 375 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH, 376 virtblk_prepare_flush); 377 } else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER)) { 378 /* 379 * If the BARRIER feature is supported the host expects us 380 * to order request by tags. This implies there is not 381 * volatile write cache on the host, and that the host 382 * never re-orders outstanding I/O. This feature is not 383 * useful for real life scenarious and deprecated. 384 */ 385 blk_queue_ordered(q, QUEUE_ORDERED_TAG, NULL); 386 } else { 387 /* 388 * If the FLUSH feature is not supported we must assume that 389 * the host does not perform any kind of volatile write 390 * caching. We still need to drain the queue to provider 391 * proper barrier semantics. 392 */ 393 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN, NULL); 394 } 395 396 /* If disk is read-only in the host, the guest should obey */ 397 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) 398 set_disk_ro(vblk->disk, 1); 399 400 /* Host must always specify the capacity. */ 401 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), 402 &cap, sizeof(cap)); 403 404 /* If capacity is too big, truncate with warning. */ 405 if ((sector_t)cap != cap) { 406 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", 407 (unsigned long long)cap); 408 cap = (sector_t)-1; 409 } 410 set_capacity(vblk->disk, cap); 411 412 /* We can handle whatever the host told us to handle. */ 413 blk_queue_max_segments(q, vblk->sg_elems-2); 414 415 /* No need to bounce any requests */ 416 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); 417 418 /* No real sector limit. */ 419 blk_queue_max_hw_sectors(q, -1U); 420 421 /* Host can optionally specify maximum segment size and number of 422 * segments. */ 423 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX, 424 offsetof(struct virtio_blk_config, size_max), 425 &v); 426 if (!err) 427 blk_queue_max_segment_size(q, v); 428 else 429 blk_queue_max_segment_size(q, -1U); 430 431 /* Host can optionally specify the block size of the device */ 432 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE, 433 offsetof(struct virtio_blk_config, blk_size), 434 &blk_size); 435 if (!err) 436 blk_queue_logical_block_size(q, blk_size); 437 else 438 blk_size = queue_logical_block_size(q); 439 440 /* Use topology information if available */ 441 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 442 offsetof(struct virtio_blk_config, physical_block_exp), 443 &physical_block_exp); 444 if (!err && physical_block_exp) 445 blk_queue_physical_block_size(q, 446 blk_size * (1 << physical_block_exp)); 447 448 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 449 offsetof(struct virtio_blk_config, alignment_offset), 450 &alignment_offset); 451 if (!err && alignment_offset) 452 blk_queue_alignment_offset(q, blk_size * alignment_offset); 453 454 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 455 offsetof(struct virtio_blk_config, min_io_size), 456 &min_io_size); 457 if (!err && min_io_size) 458 blk_queue_io_min(q, blk_size * min_io_size); 459 460 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, 461 offsetof(struct virtio_blk_config, opt_io_size), 462 &opt_io_size); 463 if (!err && opt_io_size) 464 blk_queue_io_opt(q, blk_size * opt_io_size); 465 466 467 add_disk(vblk->disk); 468 return 0; 469 470out_put_disk: 471 put_disk(vblk->disk); 472out_mempool: 473 mempool_destroy(vblk->pool); 474out_free_vq: 475 vdev->config->del_vqs(vdev); 476out_free_vblk: 477 kfree(vblk); 478out: 479 return err; 480} 481 482static void __devexit virtblk_remove(struct virtio_device *vdev) 483{ 484 struct virtio_blk *vblk = vdev->priv; 485 486 /* Nothing should be pending. */ 487 BUG_ON(!list_empty(&vblk->reqs)); 488 489 /* Stop all the virtqueues. */ 490 vdev->config->reset(vdev); 491 492 del_gendisk(vblk->disk); 493 blk_cleanup_queue(vblk->disk->queue); 494 put_disk(vblk->disk); 495 mempool_destroy(vblk->pool); 496 vdev->config->del_vqs(vdev); 497 kfree(vblk); 498} 499 500static const struct virtio_device_id id_table[] = { 501 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, 502 { 0 }, 503}; 504 505static unsigned int features[] = { 506 VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, 507 VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, 508 VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY 509}; 510 511/* 512 * virtio_blk causes spurious section mismatch warning by 513 * simultaneously referring to a __devinit and a __devexit function. 514 * Use __refdata to avoid this warning. 515 */ 516static struct virtio_driver __refdata virtio_blk = { 517 .feature_table = features, 518 .feature_table_size = ARRAY_SIZE(features), 519 .driver.name = KBUILD_MODNAME, 520 .driver.owner = THIS_MODULE, 521 .id_table = id_table, 522 .probe = virtblk_probe, 523 .remove = __devexit_p(virtblk_remove), 524}; 525 526static int __init init(void) 527{ 528 major = register_blkdev(0, "virtblk"); 529 if (major < 0) 530 return major; 531 return register_virtio_driver(&virtio_blk); 532} 533 534static void __exit fini(void) 535{ 536 unregister_blkdev(major, "virtblk"); 537 unregister_virtio_driver(&virtio_blk); 538} 539module_init(init); 540module_exit(fini); 541 542MODULE_DEVICE_TABLE(virtio, id_table); 543MODULE_DESCRIPTION("Virtio block driver"); 544MODULE_LICENSE("GPL"); 545