virtio_blk.c revision 69740c8ba878f58bc3c71f74618fc2cd1da990da 
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
185static void virtblk_prepare_flush(struct request_queue *q, struct request *req)
186{
187	req->cmd_type = REQ_TYPE_LINUX_BLOCK;
188	req->cmd[0] = REQ_LB_OP_FLUSH;
189}
190
191static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
192			 unsigned cmd, unsigned long data)
193{
194	struct gendisk *disk = bdev->bd_disk;
195	struct virtio_blk *vblk = disk->private_data;
196
197	/*
198	 * Only allow the generic SCSI ioctls if the host can support it.
199	 */
200	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
201		return -ENOTTY;
202
203	return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
204			      (void __user *)data);
205}
206
207/* We provide getgeo only to please some old bootloader/partitioning tools */
208static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
209{
210	struct virtio_blk *vblk = bd->bd_disk->private_data;
211	struct virtio_blk_geometry vgeo;
212	int err;
213
214	/* see if the host passed in geometry config */
215	err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
216				offsetof(struct virtio_blk_config, geometry),
217				&vgeo);
218
219	if (!err) {
220		geo->heads = vgeo.heads;
221		geo->sectors = vgeo.sectors;
222		geo->cylinders = vgeo.cylinders;
223	} else {
224		/* some standard values, similar to sd */
225		geo->heads = 1 << 6;
226		geo->sectors = 1 << 5;
227		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
228	}
229	return 0;
230}
231
232static const struct block_device_operations virtblk_fops = {
233	.locked_ioctl = virtblk_ioctl,
234	.owner  = THIS_MODULE,
235	.getgeo = virtblk_getgeo,
236};
237
238static int index_to_minor(int index)
239{
240	return index << PART_BITS;
241}
242
243static int __devinit virtblk_probe(struct virtio_device *vdev)
244{
245	struct virtio_blk *vblk;
246	struct request_queue *q;
247	int err;
248	u64 cap;
249	u32 v, blk_size, sg_elems, opt_io_size;
250	u16 min_io_size;
251	u8 physical_block_exp, alignment_offset;
252
253	if (index_to_minor(index) >= 1 << MINORBITS)
254		return -ENOSPC;
255
256	/* We need to know how many segments before we allocate. */
257	err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
258				offsetof(struct virtio_blk_config, seg_max),
259				&sg_elems);
260	if (err)
261		sg_elems = 1;
262
263	/* We need an extra sg elements at head and tail. */
264	sg_elems += 2;
265	vdev->priv = vblk = kmalloc(sizeof(*vblk) +
266				    sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
267	if (!vblk) {
268		err = -ENOMEM;
269		goto out;
270	}
271
272	INIT_LIST_HEAD(&vblk->reqs);
273	spin_lock_init(&vblk->lock);
274	vblk->vdev = vdev;
275	vblk->sg_elems = sg_elems;
276	sg_init_table(vblk->sg, vblk->sg_elems);
277
278	/* We expect one virtqueue, for output. */
279	vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
280	if (IS_ERR(vblk->vq)) {
281		err = PTR_ERR(vblk->vq);
282		goto out_free_vblk;
283	}
284
285	vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
286	if (!vblk->pool) {
287		err = -ENOMEM;
288		goto out_free_vq;
289	}
290
291	/* FIXME: How many partitions?  How long is a piece of string? */
292	vblk->disk = alloc_disk(1 << PART_BITS);
293	if (!vblk->disk) {
294		err = -ENOMEM;
295		goto out_mempool;
296	}
297
298	q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
299	if (!q) {
300		err = -ENOMEM;
301		goto out_put_disk;
302	}
303
304	q->queuedata = vblk;
305
306	if (index < 26) {
307		sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
308	} else if (index < (26 + 1) * 26) {
309		sprintf(vblk->disk->disk_name, "vd%c%c",
310			'a' + index / 26 - 1, 'a' + index % 26);
311	} else {
312		const unsigned int m1 = (index / 26 - 1) / 26 - 1;
313		const unsigned int m2 = (index / 26 - 1) % 26;
314		const unsigned int m3 =  index % 26;
315		sprintf(vblk->disk->disk_name, "vd%c%c%c",
316			'a' + m1, 'a' + m2, 'a' + m3);
317	}
318
319	vblk->disk->major = major;
320	vblk->disk->first_minor = index_to_minor(index);
321	vblk->disk->private_data = vblk;
322	vblk->disk->fops = &virtblk_fops;
323	vblk->disk->driverfs_dev = &vdev->dev;
324	index++;
325
326	/* If barriers are supported, tell block layer that queue is ordered */
327	if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
328		blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH,
329				  virtblk_prepare_flush);
330	else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
331		blk_queue_ordered(q, QUEUE_ORDERED_TAG, NULL);
332
333	/* If disk is read-only in the host, the guest should obey */
334	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
335		set_disk_ro(vblk->disk, 1);
336
337	/* Host must always specify the capacity. */
338	vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
339			  &cap, sizeof(cap));
340
341	/* If capacity is too big, truncate with warning. */
342	if ((sector_t)cap != cap) {
343		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
344			 (unsigned long long)cap);
345		cap = (sector_t)-1;
346	}
347	set_capacity(vblk->disk, cap);
348
349	/* We can handle whatever the host told us to handle. */
350	blk_queue_max_phys_segments(q, vblk->sg_elems-2);
351	blk_queue_max_hw_segments(q, vblk->sg_elems-2);
352
353	/* No need to bounce any requests */
354	blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
355
356	/* No real sector limit. */
357	blk_queue_max_sectors(q, -1U);
358
359	/* Host can optionally specify maximum segment size and number of
360	 * segments. */
361	err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
362				offsetof(struct virtio_blk_config, size_max),
363				&v);
364	if (!err)
365		blk_queue_max_segment_size(q, v);
366	else
367		blk_queue_max_segment_size(q, -1U);
368
369	/* Host can optionally specify the block size of the device */
370	err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
371				offsetof(struct virtio_blk_config, blk_size),
372				&blk_size);
373	if (!err)
374		blk_queue_logical_block_size(q, blk_size);
375	else
376		blk_size = queue_logical_block_size(q);
377
378	/* Use topology information if available */
379	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
380			offsetof(struct virtio_blk_config, physical_block_exp),
381			&physical_block_exp);
382	if (!err && physical_block_exp)
383		blk_queue_physical_block_size(q,
384				blk_size * (1 << physical_block_exp));
385
386	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
387			offsetof(struct virtio_blk_config, alignment_offset),
388			&alignment_offset);
389	if (!err && alignment_offset)
390		blk_queue_alignment_offset(q, blk_size * alignment_offset);
391
392	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
393			offsetof(struct virtio_blk_config, min_io_size),
394			&min_io_size);
395	if (!err && min_io_size)
396		blk_queue_io_min(q, blk_size * min_io_size);
397
398	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
399			offsetof(struct virtio_blk_config, opt_io_size),
400			&opt_io_size);
401	if (!err && opt_io_size)
402		blk_queue_io_opt(q, blk_size * opt_io_size);
403
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_FLUSH, VIRTIO_BLK_F_TOPOLOGY
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