virtio_blk.c revision e95646c3ec33c8ec0693992da4332a6b32eb7e31 
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 PART_BITS 4
10
11static int major, index;
12
13struct virtio_blk
14{
15	spinlock_t lock;
16
17	struct virtio_device *vdev;
18	struct virtqueue *vq;
19
20	/* The disk structure for the kernel. */
21	struct gendisk *disk;
22
23	/* Request tracking. */
24	struct list_head reqs;
25
26	mempool_t *pool;
27
28	/* What host tells us, plus 2 for header & tailer. */
29	unsigned int sg_elems;
30
31	/* Scatterlist: can be too big for stack. */
32	struct scatterlist sg[/*sg_elems*/];
33};
34
35struct virtblk_req
36{
37	struct list_head list;
38	struct request *req;
39	struct virtio_blk_outhdr out_hdr;
40	struct virtio_scsi_inhdr in_hdr;
41	u8 status;
42};
43
44static void blk_done(struct virtqueue *vq)
45{
46	struct virtio_blk *vblk = vq->vdev->priv;
47	struct virtblk_req *vbr;
48	unsigned int len;
49	unsigned long flags;
50
51	spin_lock_irqsave(&vblk->lock, flags);
52	while ((vbr = vblk->vq->vq_ops->get_buf(vblk->vq, &len)) != NULL) {
53		int error;
54
55		switch (vbr->status) {
56		case VIRTIO_BLK_S_OK:
57			error = 0;
58			break;
59		case VIRTIO_BLK_S_UNSUPP:
60			error = -ENOTTY;
61			break;
62		default:
63			error = -EIO;
64			break;
65		}
66
67		if (blk_pc_request(vbr->req)) {
68			vbr->req->resid_len = vbr->in_hdr.residual;
69			vbr->req->sense_len = vbr->in_hdr.sense_len;
70			vbr->req->errors = vbr->in_hdr.errors;
71		}
72
73		__blk_end_request_all(vbr->req, error);
74		list_del(&vbr->list);
75		mempool_free(vbr, vblk->pool);
76	}
77	/* In case queue is stopped waiting for more buffers. */
78	blk_start_queue(vblk->disk->queue);
79	spin_unlock_irqrestore(&vblk->lock, flags);
80}
81
82static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
83		   struct request *req)
84{
85	unsigned long num, out = 0, in = 0;
86	struct virtblk_req *vbr;
87
88	vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
89	if (!vbr)
90		/* When another request finishes we'll try again. */
91		return false;
92
93	vbr->req = req;
94	switch (req->cmd_type) {
95	case REQ_TYPE_FS:
96		vbr->out_hdr.type = 0;
97		vbr->out_hdr.sector = blk_rq_pos(vbr->req);
98		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
99		break;
100	case REQ_TYPE_BLOCK_PC:
101		vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
102		vbr->out_hdr.sector = 0;
103		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
104		break;
105	case REQ_TYPE_LINUX_BLOCK:
106		if (req->cmd[0] == REQ_LB_OP_FLUSH) {
107			vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
108			vbr->out_hdr.sector = 0;
109			vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
110			break;
111		}
112		/*FALLTHRU*/
113	default:
114		/* We don't put anything else in the queue. */
115		BUG();
116	}
117
118	if (blk_barrier_rq(vbr->req))
119		vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
120
121	sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
122
123	/*
124	 * If this is a packet command we need a couple of additional headers.
125	 * Behind the normal outhdr we put a segment with the scsi command
126	 * block, and before the normal inhdr we put the sense data and the
127	 * inhdr with additional status information before the normal inhdr.
128	 */
129	if (blk_pc_request(vbr->req))
130		sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
131
132	num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
133
134	if (blk_pc_request(vbr->req)) {
135		sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
136		sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
137			   sizeof(vbr->in_hdr));
138	}
139
140	sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
141		   sizeof(vbr->status));
142
143	if (num) {
144		if (rq_data_dir(vbr->req) == WRITE) {
145			vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
146			out += num;
147		} else {
148			vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
149			in += num;
150		}
151	}
152
153	if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
154		mempool_free(vbr, vblk->pool);
155		return false;
156	}
157
158	list_add_tail(&vbr->list, &vblk->reqs);
159	return true;
160}
161
162static void do_virtblk_request(struct request_queue *q)
163{
164	struct virtio_blk *vblk = q->queuedata;
165	struct request *req;
166	unsigned int issued = 0;
167
168	while ((req = blk_peek_request(q)) != NULL) {
169		BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
170
171		/* If this request fails, stop queue and wait for something to
172		   finish to restart it. */
173		if (!do_req(q, vblk, req)) {
174			blk_stop_queue(q);
175			break;
176		}
177		blk_start_request(req);
178		issued++;
179	}
180
181	if (issued)
182		vblk->vq->vq_ops->kick(vblk->vq);
183}
184
185/* return ATA identify data
186 */
187static int virtblk_identify(struct gendisk *disk, void *argp)
188{
189	struct virtio_blk *vblk = disk->private_data;
190	void *opaque;
191	int err = -ENOMEM;
192
193	opaque = kmalloc(VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
194	if (!opaque)
195		goto out;
196
197	err = virtio_config_buf(vblk->vdev, VIRTIO_BLK_F_IDENTIFY,
198		offsetof(struct virtio_blk_config, identify), opaque,
199		VIRTIO_BLK_ID_BYTES);
200
201	if (err)
202		goto out_kfree;
203
204	if (copy_to_user(argp, opaque, VIRTIO_BLK_ID_BYTES))
205		err = -EFAULT;
206
207out_kfree:
208	kfree(opaque);
209out:
210	return err;
211}
212
213static void virtblk_prepare_flush(struct request_queue *q, struct request *req)
214{
215	req->cmd_type = REQ_TYPE_LINUX_BLOCK;
216	req->cmd[0] = REQ_LB_OP_FLUSH;
217}
218
219static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
220			 unsigned cmd, unsigned long data)
221{
222	struct gendisk *disk = bdev->bd_disk;
223	struct virtio_blk *vblk = disk->private_data;
224	void __user *argp = (void __user *)data;
225
226	if (cmd == HDIO_GET_IDENTITY)
227		return virtblk_identify(disk, argp);
228
229	/*
230	 * Only allow the generic SCSI ioctls if the host can support it.
231	 */
232	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
233		return -ENOTTY;
234
235	return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
236}
237
238/* We provide getgeo only to please some old bootloader/partitioning tools */
239static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
240{
241	struct virtio_blk *vblk = bd->bd_disk->private_data;
242	struct virtio_blk_geometry vgeo;
243	int err;
244
245	/* see if the host passed in geometry config */
246	err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
247				offsetof(struct virtio_blk_config, geometry),
248				&vgeo);
249
250	if (!err) {
251		geo->heads = vgeo.heads;
252		geo->sectors = vgeo.sectors;
253		geo->cylinders = vgeo.cylinders;
254	} else {
255		/* some standard values, similar to sd */
256		geo->heads = 1 << 6;
257		geo->sectors = 1 << 5;
258		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
259	}
260	return 0;
261}
262
263static const struct block_device_operations virtblk_fops = {
264	.locked_ioctl = virtblk_ioctl,
265	.owner  = THIS_MODULE,
266	.getgeo = virtblk_getgeo,
267};
268
269static int index_to_minor(int index)
270{
271	return index << PART_BITS;
272}
273
274static int __devinit virtblk_probe(struct virtio_device *vdev)
275{
276	struct virtio_blk *vblk;
277	int err;
278	u64 cap;
279	u32 v;
280	u32 blk_size, sg_elems;
281
282	if (index_to_minor(index) >= 1 << MINORBITS)
283		return -ENOSPC;
284
285	/* We need to know how many segments before we allocate. */
286	err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
287				offsetof(struct virtio_blk_config, seg_max),
288				&sg_elems);
289	if (err)
290		sg_elems = 1;
291
292	/* We need an extra sg elements at head and tail. */
293	sg_elems += 2;
294	vdev->priv = vblk = kmalloc(sizeof(*vblk) +
295				    sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
296	if (!vblk) {
297		err = -ENOMEM;
298		goto out;
299	}
300
301	INIT_LIST_HEAD(&vblk->reqs);
302	spin_lock_init(&vblk->lock);
303	vblk->vdev = vdev;
304	vblk->sg_elems = sg_elems;
305	sg_init_table(vblk->sg, vblk->sg_elems);
306
307	/* We expect one virtqueue, for output. */
308	vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
309	if (IS_ERR(vblk->vq)) {
310		err = PTR_ERR(vblk->vq);
311		goto out_free_vblk;
312	}
313
314	vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
315	if (!vblk->pool) {
316		err = -ENOMEM;
317		goto out_free_vq;
318	}
319
320	/* FIXME: How many partitions?  How long is a piece of string? */
321	vblk->disk = alloc_disk(1 << PART_BITS);
322	if (!vblk->disk) {
323		err = -ENOMEM;
324		goto out_mempool;
325	}
326
327	vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
328	if (!vblk->disk->queue) {
329		err = -ENOMEM;
330		goto out_put_disk;
331	}
332
333	vblk->disk->queue->queuedata = vblk;
334
335	if (index < 26) {
336		sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
337	} else if (index < (26 + 1) * 26) {
338		sprintf(vblk->disk->disk_name, "vd%c%c",
339			'a' + index / 26 - 1, 'a' + index % 26);
340	} else {
341		const unsigned int m1 = (index / 26 - 1) / 26 - 1;
342		const unsigned int m2 = (index / 26 - 1) % 26;
343		const unsigned int m3 =  index % 26;
344		sprintf(vblk->disk->disk_name, "vd%c%c%c",
345			'a' + m1, 'a' + m2, 'a' + m3);
346	}
347
348	vblk->disk->major = major;
349	vblk->disk->first_minor = index_to_minor(index);
350	vblk->disk->private_data = vblk;
351	vblk->disk->fops = &virtblk_fops;
352	vblk->disk->driverfs_dev = &vdev->dev;
353	index++;
354
355	/* If barriers are supported, tell block layer that queue is ordered */
356	if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
357		blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_DRAIN_FLUSH,
358				  virtblk_prepare_flush);
359	else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
360		blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL);
361
362	/* If disk is read-only in the host, the guest should obey */
363	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
364		set_disk_ro(vblk->disk, 1);
365
366	/* Host must always specify the capacity. */
367	vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
368			  &cap, sizeof(cap));
369
370	/* If capacity is too big, truncate with warning. */
371	if ((sector_t)cap != cap) {
372		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
373			 (unsigned long long)cap);
374		cap = (sector_t)-1;
375	}
376	set_capacity(vblk->disk, cap);
377
378	/* We can handle whatever the host told us to handle. */
379	blk_queue_max_phys_segments(vblk->disk->queue, vblk->sg_elems-2);
380	blk_queue_max_hw_segments(vblk->disk->queue, vblk->sg_elems-2);
381
382	/* No need to bounce any requests */
383	blk_queue_bounce_limit(vblk->disk->queue, BLK_BOUNCE_ANY);
384
385	/* No real sector limit. */
386	blk_queue_max_sectors(vblk->disk->queue, -1U);
387
388	/* Host can optionally specify maximum segment size and number of
389	 * segments. */
390	err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
391				offsetof(struct virtio_blk_config, size_max),
392				&v);
393	if (!err)
394		blk_queue_max_segment_size(vblk->disk->queue, v);
395	else
396		blk_queue_max_segment_size(vblk->disk->queue, -1U);
397
398	/* Host can optionally specify the block size of the device */
399	err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
400				offsetof(struct virtio_blk_config, blk_size),
401				&blk_size);
402	if (!err)
403		blk_queue_logical_block_size(vblk->disk->queue, blk_size);
404
405	add_disk(vblk->disk);
406	return 0;
407
408out_put_disk:
409	put_disk(vblk->disk);
410out_mempool:
411	mempool_destroy(vblk->pool);
412out_free_vq:
413	vdev->config->del_vqs(vdev);
414out_free_vblk:
415	kfree(vblk);
416out:
417	return err;
418}
419
420static void __devexit virtblk_remove(struct virtio_device *vdev)
421{
422	struct virtio_blk *vblk = vdev->priv;
423
424	/* Nothing should be pending. */
425	BUG_ON(!list_empty(&vblk->reqs));
426
427	/* Stop all the virtqueues. */
428	vdev->config->reset(vdev);
429
430	del_gendisk(vblk->disk);
431	blk_cleanup_queue(vblk->disk->queue);
432	put_disk(vblk->disk);
433	mempool_destroy(vblk->pool);
434	vdev->config->del_vqs(vdev);
435	kfree(vblk);
436}
437
438static struct virtio_device_id id_table[] = {
439	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
440	{ 0 },
441};
442
443static unsigned int features[] = {
444	VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
445	VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
446	VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_IDENTIFY, VIRTIO_BLK_F_FLUSH
447};
448
449/*
450 * virtio_blk causes spurious section mismatch warning by
451 * simultaneously referring to a __devinit and a __devexit function.
452 * Use __refdata to avoid this warning.
453 */
454static struct virtio_driver __refdata virtio_blk = {
455	.feature_table = features,
456	.feature_table_size = ARRAY_SIZE(features),
457	.driver.name =	KBUILD_MODNAME,
458	.driver.owner =	THIS_MODULE,
459	.id_table =	id_table,
460	.probe =	virtblk_probe,
461	.remove =	__devexit_p(virtblk_remove),
462};
463
464static int __init init(void)
465{
466	major = register_blkdev(0, "virtblk");
467	if (major < 0)
468		return major;
469	return register_virtio_driver(&virtio_blk);
470}
471
472static void __exit fini(void)
473{
474	unregister_blkdev(major, "virtblk");
475	unregister_virtio_driver(&virtio_blk);
476}
477module_init(init);
478module_exit(fini);
479
480MODULE_DEVICE_TABLE(virtio, id_table);
481MODULE_DESCRIPTION("Virtio block driver");
482MODULE_LICENSE("GPL");
483