virtio_blk.c revision 72e61eb40b55dd57031ec5971e810649f82b0259
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 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 /* Scatterlist: can be too big for stack. */ 30 struct scatterlist sg[VIRTIO_MAX_SG]; 31}; 32 33struct virtblk_req 34{ 35 struct list_head list; 36 struct request *req; 37 struct virtio_blk_outhdr out_hdr; 38 u8 status; 39}; 40 41static 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 72static 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 125static 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 149static 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 */ 158static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) 159{ 160 /* some standard values, similar to sd */ 161 geo->heads = 1 << 6; 162 geo->sectors = 1 << 5; 163 geo->cylinders = get_capacity(bd->bd_disk) >> 11; 164 return 0; 165} 166 167static struct block_device_operations virtblk_fops = { 168 .ioctl = virtblk_ioctl, 169 .owner = THIS_MODULE, 170 .getgeo = virtblk_getgeo, 171}; 172 173static int index_to_minor(int index) 174{ 175 return index << PART_BITS; 176} 177 178static int virtblk_probe(struct virtio_device *vdev) 179{ 180 struct virtio_blk *vblk; 181 int err; 182 u64 cap; 183 u32 v; 184 185 if (index_to_minor(index) >= 1 << MINORBITS) 186 return -ENOSPC; 187 188 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL); 189 if (!vblk) { 190 err = -ENOMEM; 191 goto out; 192 } 193 194 INIT_LIST_HEAD(&vblk->reqs); 195 spin_lock_init(&vblk->lock); 196 vblk->vdev = vdev; 197 198 /* We expect one virtqueue, for output. */ 199 vblk->vq = vdev->config->find_vq(vdev, 0, blk_done); 200 if (IS_ERR(vblk->vq)) { 201 err = PTR_ERR(vblk->vq); 202 goto out_free_vblk; 203 } 204 205 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req)); 206 if (!vblk->pool) { 207 err = -ENOMEM; 208 goto out_free_vq; 209 } 210 211 /* FIXME: How many partitions? How long is a piece of string? */ 212 vblk->disk = alloc_disk(1 << PART_BITS); 213 if (!vblk->disk) { 214 err = -ENOMEM; 215 goto out_mempool; 216 } 217 218 vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock); 219 if (!vblk->disk->queue) { 220 err = -ENOMEM; 221 goto out_put_disk; 222 } 223 224 if (index < 26) { 225 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26); 226 } else if (index < (26 + 1) * 26) { 227 sprintf(vblk->disk->disk_name, "vd%c%c", 228 'a' + index / 26 - 1, 'a' + index % 26); 229 } else { 230 const unsigned int m1 = (index / 26 - 1) / 26 - 1; 231 const unsigned int m2 = (index / 26 - 1) % 26; 232 const unsigned int m3 = index % 26; 233 sprintf(vblk->disk->disk_name, "vd%c%c%c", 234 'a' + m1, 'a' + m2, 'a' + m3); 235 } 236 237 vblk->disk->major = major; 238 vblk->disk->first_minor = index_to_minor(index); 239 vblk->disk->private_data = vblk; 240 vblk->disk->fops = &virtblk_fops; 241 vblk->disk->driverfs_dev = &vdev->dev; 242 index++; 243 244 /* If barriers are supported, tell block layer that queue is ordered */ 245 if (vdev->config->feature(vdev, VIRTIO_BLK_F_BARRIER)) 246 blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL); 247 248 /* Host must always specify the capacity. */ 249 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), 250 &cap, sizeof(cap)); 251 252 /* If capacity is too big, truncate with warning. */ 253 if ((sector_t)cap != cap) { 254 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", 255 (unsigned long long)cap); 256 cap = (sector_t)-1; 257 } 258 set_capacity(vblk->disk, cap); 259 260 /* Host can optionally specify maximum segment size and number of 261 * segments. */ 262 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX, 263 offsetof(struct virtio_blk_config, size_max), 264 &v); 265 if (!err) 266 blk_queue_max_segment_size(vblk->disk->queue, v); 267 268 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX, 269 offsetof(struct virtio_blk_config, seg_max), 270 &v); 271 if (!err) 272 blk_queue_max_hw_segments(vblk->disk->queue, v); 273 274 add_disk(vblk->disk); 275 return 0; 276 277out_put_disk: 278 put_disk(vblk->disk); 279out_mempool: 280 mempool_destroy(vblk->pool); 281out_free_vq: 282 vdev->config->del_vq(vblk->vq); 283out_free_vblk: 284 kfree(vblk); 285out: 286 return err; 287} 288 289static void virtblk_remove(struct virtio_device *vdev) 290{ 291 struct virtio_blk *vblk = vdev->priv; 292 293 /* Nothing should be pending. */ 294 BUG_ON(!list_empty(&vblk->reqs)); 295 296 /* Stop all the virtqueues. */ 297 vdev->config->reset(vdev); 298 299 blk_cleanup_queue(vblk->disk->queue); 300 put_disk(vblk->disk); 301 mempool_destroy(vblk->pool); 302 vdev->config->del_vq(vblk->vq); 303 kfree(vblk); 304} 305 306static struct virtio_device_id id_table[] = { 307 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, 308 { 0 }, 309}; 310 311static struct virtio_driver virtio_blk = { 312 .driver.name = KBUILD_MODNAME, 313 .driver.owner = THIS_MODULE, 314 .id_table = id_table, 315 .probe = virtblk_probe, 316 .remove = __devexit_p(virtblk_remove), 317}; 318 319static int __init init(void) 320{ 321 major = register_blkdev(0, "virtblk"); 322 if (major < 0) 323 return major; 324 return register_virtio_driver(&virtio_blk); 325} 326 327static void __exit fini(void) 328{ 329 unregister_blkdev(major, "virtblk"); 330 unregister_virtio_driver(&virtio_blk); 331} 332module_init(init); 333module_exit(fini); 334 335MODULE_DEVICE_TABLE(virtio, id_table); 336MODULE_DESCRIPTION("Virtio block driver"); 337MODULE_LICENSE("GPL"); 338