1 //#define DEBUG 2 #include <linux/spinlock.h> 3 #include <linux/blkdev.h> 4 #include <linux/hdreg.h> 5 #include <linux/virtio.h> 6 #include <linux/virtio_blk.h> 7 #include <linux/scatterlist.h> 8 9 #define VIRTIO_MAX_SG (3+MAX_PHYS_SEGMENTS) 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 /* Scatterlist: can be too big for stack. */ 30 struct scatterlist sg[VIRTIO_MAX_SG]; 31 }; 32 33 struct virtblk_req 34 { 35 struct list_head list; 36 struct request *req; 37 struct virtio_blk_outhdr out_hdr; 38 u8 status; 39 }; 40 41 static void blk_done(struct virtqueue *vq) 42 { 43 struct virtio_blk *vblk = vq->vdev->priv; 44 struct virtblk_req *vbr; 45 unsigned int len; 46 unsigned long flags; 47 48 spin_lock_irqsave(&vblk->lock, flags); 49 while ((vbr = vblk->vq->vq_ops->get_buf(vblk->vq, &len)) != NULL) { 50 int uptodate; 51 switch (vbr->status) { 52 case VIRTIO_BLK_S_OK: 53 uptodate = 1; 54 break; 55 case VIRTIO_BLK_S_UNSUPP: 56 uptodate = -ENOTTY; 57 break; 58 default: 59 uptodate = 0; 60 break; 61 } 62 63 end_dequeued_request(vbr->req, uptodate); 64 list_del(&vbr->list); 65 mempool_free(vbr, vblk->pool); 66 } 67 /* In case queue is stopped waiting for more buffers. */ 68 blk_start_queue(vblk->disk->queue); 69 spin_unlock_irqrestore(&vblk->lock, flags); 70 } 71 72 static bool do_req(struct request_queue *q, struct virtio_blk *vblk, 73 struct request *req) 74 { 75 unsigned long num, out, in; 76 struct virtblk_req *vbr; 77 78 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC); 79 if (!vbr) 80 /* When another request finishes we'll try again. */ 81 return false; 82 83 vbr->req = req; 84 if (blk_fs_request(vbr->req)) { 85 vbr->out_hdr.type = 0; 86 vbr->out_hdr.sector = vbr->req->sector; 87 vbr->out_hdr.ioprio = vbr->req->ioprio; 88 } else if (blk_pc_request(vbr->req)) { 89 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; 90 vbr->out_hdr.sector = 0; 91 vbr->out_hdr.ioprio = vbr->req->ioprio; 92 } else { 93 /* We don't put anything else in the queue. */ 94 BUG(); 95 } 96 97 if (blk_barrier_rq(vbr->req)) 98 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER; 99 100 /* This init could be done at vblk creation time */ 101 sg_init_table(vblk->sg, VIRTIO_MAX_SG); 102 sg_set_buf(&vblk->sg[0], &vbr->out_hdr, sizeof(vbr->out_hdr)); 103 num = blk_rq_map_sg(q, vbr->req, vblk->sg+1); 104 sg_set_buf(&vblk->sg[num+1], &vbr->status, sizeof(vbr->status)); 105 106 if (rq_data_dir(vbr->req) == WRITE) { 107 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; 108 out = 1 + num; 109 in = 1; 110 } else { 111 vbr->out_hdr.type |= VIRTIO_BLK_T_IN; 112 out = 1; 113 in = 1 + num; 114 } 115 116 if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr)) { 117 mempool_free(vbr, vblk->pool); 118 return false; 119 } 120 121 list_add_tail(&vbr->list, &vblk->reqs); 122 return true; 123 } 124 125 static void do_virtblk_request(struct request_queue *q) 126 { 127 struct virtio_blk *vblk = NULL; 128 struct request *req; 129 unsigned int issued = 0; 130 131 while ((req = elv_next_request(q)) != NULL) { 132 vblk = req->rq_disk->private_data; 133 BUG_ON(req->nr_phys_segments > ARRAY_SIZE(vblk->sg)); 134 135 /* If this request fails, stop queue and wait for something to 136 finish to restart it. */ 137 if (!do_req(q, vblk, req)) { 138 blk_stop_queue(q); 139 break; 140 } 141 blkdev_dequeue_request(req); 142 issued++; 143 } 144 145 if (issued) 146 vblk->vq->vq_ops->kick(vblk->vq); 147 } 148 149 static int virtblk_ioctl(struct inode *inode, struct file *filp, 150 unsigned cmd, unsigned long data) 151 { 152 return scsi_cmd_ioctl(filp, inode->i_bdev->bd_disk->queue, 153 inode->i_bdev->bd_disk, cmd, 154 (void __user *)data); 155 } 156 157 /* We provide getgeo only to please some old bootloader/partitioning tools */ 158 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) 159 { 160 struct virtio_blk *vblk = bd->bd_disk->private_data; 161 struct virtio_blk_geometry vgeo; 162 int err; 163 164 /* see if the host passed in geometry config */ 165 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY, 166 offsetof(struct virtio_blk_config, geometry), 167 &vgeo); 168 169 if (!err) { 170 geo->heads = vgeo.heads; 171 geo->sectors = vgeo.sectors; 172 geo->cylinders = vgeo.cylinders; 173 } else { 174 /* some standard values, similar to sd */ 175 geo->heads = 1 << 6; 176 geo->sectors = 1 << 5; 177 geo->cylinders = get_capacity(bd->bd_disk) >> 11; 178 } 179 return 0; 180 } 181 182 static struct block_device_operations virtblk_fops = { 183 .ioctl = virtblk_ioctl, 184 .owner = THIS_MODULE, 185 .getgeo = virtblk_getgeo, 186 }; 187 188 static int index_to_minor(int index) 189 { 190 return index << PART_BITS; 191 } 192 193 static int virtblk_probe(struct virtio_device *vdev) 194 { 195 struct virtio_blk *vblk; 196 int err; 197 u64 cap; 198 u32 v; 199 200 if (index_to_minor(index) >= 1 << MINORBITS) 201 return -ENOSPC; 202 203 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL); 204 if (!vblk) { 205 err = -ENOMEM; 206 goto out; 207 } 208 209 INIT_LIST_HEAD(&vblk->reqs); 210 spin_lock_init(&vblk->lock); 211 vblk->vdev = vdev; 212 213 /* We expect one virtqueue, for output. */ 214 vblk->vq = vdev->config->find_vq(vdev, 0, blk_done); 215 if (IS_ERR(vblk->vq)) { 216 err = PTR_ERR(vblk->vq); 217 goto out_free_vblk; 218 } 219 220 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req)); 221 if (!vblk->pool) { 222 err = -ENOMEM; 223 goto out_free_vq; 224 } 225 226 /* FIXME: How many partitions? How long is a piece of string? */ 227 vblk->disk = alloc_disk(1 << PART_BITS); 228 if (!vblk->disk) { 229 err = -ENOMEM; 230 goto out_mempool; 231 } 232 233 vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock); 234 if (!vblk->disk->queue) { 235 err = -ENOMEM; 236 goto out_put_disk; 237 } 238 239 if (index < 26) { 240 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26); 241 } else if (index < (26 + 1) * 26) { 242 sprintf(vblk->disk->disk_name, "vd%c%c", 243 'a' + index / 26 - 1, 'a' + index % 26); 244 } else { 245 const unsigned int m1 = (index / 26 - 1) / 26 - 1; 246 const unsigned int m2 = (index / 26 - 1) % 26; 247 const unsigned int m3 = index % 26; 248 sprintf(vblk->disk->disk_name, "vd%c%c%c", 249 'a' + m1, 'a' + m2, 'a' + m3); 250 } 251 252 vblk->disk->major = major; 253 vblk->disk->first_minor = index_to_minor(index); 254 vblk->disk->private_data = vblk; 255 vblk->disk->fops = &virtblk_fops; 256 vblk->disk->driverfs_dev = &vdev->dev; 257 index++; 258 259 /* If barriers are supported, tell block layer that queue is ordered */ 260 if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER)) 261 blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL); 262 263 /* Host must always specify the capacity. */ 264 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), 265 &cap, sizeof(cap)); 266 267 /* If capacity is too big, truncate with warning. */ 268 if ((sector_t)cap != cap) { 269 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", 270 (unsigned long long)cap); 271 cap = (sector_t)-1; 272 } 273 set_capacity(vblk->disk, cap); 274 275 /* Host can optionally specify maximum segment size and number of 276 * segments. */ 277 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX, 278 offsetof(struct virtio_blk_config, size_max), 279 &v); 280 if (!err) 281 blk_queue_max_segment_size(vblk->disk->queue, v); 282 283 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX, 284 offsetof(struct virtio_blk_config, seg_max), 285 &v); 286 if (!err) 287 blk_queue_max_hw_segments(vblk->disk->queue, v); 288 289 add_disk(vblk->disk); 290 return 0; 291 292 out_put_disk: 293 put_disk(vblk->disk); 294 out_mempool: 295 mempool_destroy(vblk->pool); 296 out_free_vq: 297 vdev->config->del_vq(vblk->vq); 298 out_free_vblk: 299 kfree(vblk); 300 out: 301 return err; 302 } 303 304 static void virtblk_remove(struct virtio_device *vdev) 305 { 306 struct virtio_blk *vblk = vdev->priv; 307 308 /* Nothing should be pending. */ 309 BUG_ON(!list_empty(&vblk->reqs)); 310 311 /* Stop all the virtqueues. */ 312 vdev->config->reset(vdev); 313 314 blk_cleanup_queue(vblk->disk->queue); 315 put_disk(vblk->disk); 316 mempool_destroy(vblk->pool); 317 vdev->config->del_vq(vblk->vq); 318 kfree(vblk); 319 } 320 321 static struct virtio_device_id id_table[] = { 322 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, 323 { 0 }, 324 }; 325 326 static unsigned int features[] = { 327 VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, 328 VIRTIO_BLK_F_GEOMETRY, 329 }; 330 331 static struct virtio_driver virtio_blk = { 332 .feature_table = features, 333 .feature_table_size = ARRAY_SIZE(features), 334 .driver.name = KBUILD_MODNAME, 335 .driver.owner = THIS_MODULE, 336 .id_table = id_table, 337 .probe = virtblk_probe, 338 .remove = __devexit_p(virtblk_remove), 339 }; 340 341 static int __init init(void) 342 { 343 major = register_blkdev(0, "virtblk"); 344 if (major < 0) 345 return major; 346 return register_virtio_driver(&virtio_blk); 347 } 348 349 static void __exit fini(void) 350 { 351 unregister_blkdev(major, "virtblk"); 352 unregister_virtio_driver(&virtio_blk); 353 } 354 module_init(init); 355 module_exit(fini); 356 357 MODULE_DEVICE_TABLE(virtio, id_table); 358 MODULE_DESCRIPTION("Virtio block driver"); 359 MODULE_LICENSE("GPL"); 360