virtio_blk.c revision 1cde26f928863d90e9e7c1217880c8450464d305 
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		unsigned int nr_bytes;
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
74		__blk_end_request_all(vbr->req, error);
75		list_del(&vbr->list);
76		mempool_free(vbr, vblk->pool);
77	}
78	/* In case queue is stopped waiting for more buffers. */
79	blk_start_queue(vblk->disk->queue);
80	spin_unlock_irqrestore(&vblk->lock, flags);
81}
82
83static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
84		   struct request *req)
85{
86	unsigned long num, out = 0, in = 0;
87	struct virtblk_req *vbr;
88
89	vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
90	if (!vbr)
91		/* When another request finishes we'll try again. */
92		return false;
93
94	vbr->req = req;
95	if (blk_fs_request(vbr->req)) {
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	} else if (blk_pc_request(vbr->req)) {
100		vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
101		vbr->out_hdr.sector = 0;
102		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
103	} else {
104		/* We don't put anything else in the queue. */
105		BUG();
106	}
107
108	if (blk_barrier_rq(vbr->req))
109		vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
110
111	sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
112
113	/*
114	 * If this is a packet command we need a couple of additional headers.
115	 * Behind the normal outhdr we put a segment with the scsi command
116	 * block, and before the normal inhdr we put the sense data and the
117	 * inhdr with additional status information before the normal inhdr.
118	 */
119	if (blk_pc_request(vbr->req))
120		sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
121
122	num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
123
124	if (blk_pc_request(vbr->req)) {
125		sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
126		sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
127			   sizeof(vbr->in_hdr));
128	}
129
130	sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
131		   sizeof(vbr->status));
132
133	if (num) {
134		if (rq_data_dir(vbr->req) == WRITE) {
135			vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
136			out += num;
137		} else {
138			vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
139			in += num;
140		}
141	}
142
143	if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr)) {
144		mempool_free(vbr, vblk->pool);
145		return false;
146	}
147
148	list_add_tail(&vbr->list, &vblk->reqs);
149	return true;
150}
151
152static void do_virtblk_request(struct request_queue *q)
153{
154	struct virtio_blk *vblk = q->queuedata;
155	struct request *req;
156	unsigned int issued = 0;
157
158	while ((req = blk_peek_request(q)) != NULL) {
159		BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
160
161		/* If this request fails, stop queue and wait for something to
162		   finish to restart it. */
163		if (!do_req(q, vblk, req)) {
164			blk_stop_queue(q);
165			break;
166		}
167		blk_start_request(req);
168		issued++;
169	}
170
171	if (issued)
172		vblk->vq->vq_ops->kick(vblk->vq);
173}
174
175static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
176			 unsigned cmd, unsigned long data)
177{
178	struct gendisk *disk = bdev->bd_disk;
179	struct virtio_blk *vblk = disk->private_data;
180
181	/*
182	 * Only allow the generic SCSI ioctls if the host can support it.
183	 */
184	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
185		return -ENOIOCTLCMD;
186
187	return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
188			      (void __user *)data);
189}
190
191/* We provide getgeo only to please some old bootloader/partitioning tools */
192static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
193{
194	struct virtio_blk *vblk = bd->bd_disk->private_data;
195	struct virtio_blk_geometry vgeo;
196	int err;
197
198	/* see if the host passed in geometry config */
199	err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
200				offsetof(struct virtio_blk_config, geometry),
201				&vgeo);
202
203	if (!err) {
204		geo->heads = vgeo.heads;
205		geo->sectors = vgeo.sectors;
206		geo->cylinders = vgeo.cylinders;
207	} else {
208		/* some standard values, similar to sd */
209		geo->heads = 1 << 6;
210		geo->sectors = 1 << 5;
211		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
212	}
213	return 0;
214}
215
216static struct block_device_operations virtblk_fops = {
217	.locked_ioctl = virtblk_ioctl,
218	.owner  = THIS_MODULE,
219	.getgeo = virtblk_getgeo,
220};
221
222static int index_to_minor(int index)
223{
224	return index << PART_BITS;
225}
226
227static int virtblk_probe(struct virtio_device *vdev)
228{
229	struct virtio_blk *vblk;
230	int err;
231	u64 cap;
232	u32 v;
233	u32 blk_size, sg_elems;
234
235	if (index_to_minor(index) >= 1 << MINORBITS)
236		return -ENOSPC;
237
238	/* We need to know how many segments before we allocate. */
239	err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
240				offsetof(struct virtio_blk_config, seg_max),
241				&sg_elems);
242	if (err)
243		sg_elems = 1;
244
245	/* We need an extra sg elements at head and tail. */
246	sg_elems += 2;
247	vdev->priv = vblk = kmalloc(sizeof(*vblk) +
248				    sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
249	if (!vblk) {
250		err = -ENOMEM;
251		goto out;
252	}
253
254	INIT_LIST_HEAD(&vblk->reqs);
255	spin_lock_init(&vblk->lock);
256	vblk->vdev = vdev;
257	vblk->sg_elems = sg_elems;
258	sg_init_table(vblk->sg, vblk->sg_elems);
259
260	/* We expect one virtqueue, for output. */
261	vblk->vq = vdev->config->find_vq(vdev, 0, blk_done);
262	if (IS_ERR(vblk->vq)) {
263		err = PTR_ERR(vblk->vq);
264		goto out_free_vblk;
265	}
266
267	vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
268	if (!vblk->pool) {
269		err = -ENOMEM;
270		goto out_free_vq;
271	}
272
273	/* FIXME: How many partitions?  How long is a piece of string? */
274	vblk->disk = alloc_disk(1 << PART_BITS);
275	if (!vblk->disk) {
276		err = -ENOMEM;
277		goto out_mempool;
278	}
279
280	vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
281	if (!vblk->disk->queue) {
282		err = -ENOMEM;
283		goto out_put_disk;
284	}
285
286	vblk->disk->queue->queuedata = vblk;
287	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, vblk->disk->queue);
288
289	if (index < 26) {
290		sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
291	} else if (index < (26 + 1) * 26) {
292		sprintf(vblk->disk->disk_name, "vd%c%c",
293			'a' + index / 26 - 1, 'a' + index % 26);
294	} else {
295		const unsigned int m1 = (index / 26 - 1) / 26 - 1;
296		const unsigned int m2 = (index / 26 - 1) % 26;
297		const unsigned int m3 =  index % 26;
298		sprintf(vblk->disk->disk_name, "vd%c%c%c",
299			'a' + m1, 'a' + m2, 'a' + m3);
300	}
301
302	vblk->disk->major = major;
303	vblk->disk->first_minor = index_to_minor(index);
304	vblk->disk->private_data = vblk;
305	vblk->disk->fops = &virtblk_fops;
306	vblk->disk->driverfs_dev = &vdev->dev;
307	index++;
308
309	/* If barriers are supported, tell block layer that queue is ordered */
310	if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
311		blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL);
312
313	/* If disk is read-only in the host, the guest should obey */
314	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
315		set_disk_ro(vblk->disk, 1);
316
317	/* Host must always specify the capacity. */
318	vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
319			  &cap, sizeof(cap));
320
321	/* If capacity is too big, truncate with warning. */
322	if ((sector_t)cap != cap) {
323		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
324			 (unsigned long long)cap);
325		cap = (sector_t)-1;
326	}
327	set_capacity(vblk->disk, cap);
328
329	/* We can handle whatever the host told us to handle. */
330	blk_queue_max_phys_segments(vblk->disk->queue, vblk->sg_elems-2);
331	blk_queue_max_hw_segments(vblk->disk->queue, vblk->sg_elems-2);
332
333	/* No real sector limit. */
334	blk_queue_max_sectors(vblk->disk->queue, -1U);
335
336	/* Host can optionally specify maximum segment size and number of
337	 * segments. */
338	err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
339				offsetof(struct virtio_blk_config, size_max),
340				&v);
341	if (!err)
342		blk_queue_max_segment_size(vblk->disk->queue, v);
343	else
344		blk_queue_max_segment_size(vblk->disk->queue, -1U);
345
346	/* Host can optionally specify the block size of the device */
347	err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
348				offsetof(struct virtio_blk_config, blk_size),
349				&blk_size);
350	if (!err)
351		blk_queue_hardsect_size(vblk->disk->queue, blk_size);
352
353	add_disk(vblk->disk);
354	return 0;
355
356out_put_disk:
357	put_disk(vblk->disk);
358out_mempool:
359	mempool_destroy(vblk->pool);
360out_free_vq:
361	vdev->config->del_vq(vblk->vq);
362out_free_vblk:
363	kfree(vblk);
364out:
365	return err;
366}
367
368static void virtblk_remove(struct virtio_device *vdev)
369{
370	struct virtio_blk *vblk = vdev->priv;
371
372	/* Nothing should be pending. */
373	BUG_ON(!list_empty(&vblk->reqs));
374
375	/* Stop all the virtqueues. */
376	vdev->config->reset(vdev);
377
378	del_gendisk(vblk->disk);
379	blk_cleanup_queue(vblk->disk->queue);
380	put_disk(vblk->disk);
381	mempool_destroy(vblk->pool);
382	vdev->config->del_vq(vblk->vq);
383	kfree(vblk);
384}
385
386static struct virtio_device_id id_table[] = {
387	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
388	{ 0 },
389};
390
391static unsigned int features[] = {
392	VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
393	VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
394	VIRTIO_BLK_F_SCSI,
395};
396
397static struct virtio_driver virtio_blk = {
398	.feature_table = features,
399	.feature_table_size = ARRAY_SIZE(features),
400	.driver.name =	KBUILD_MODNAME,
401	.driver.owner =	THIS_MODULE,
402	.id_table =	id_table,
403	.probe =	virtblk_probe,
404	.remove =	__devexit_p(virtblk_remove),
405};
406
407static int __init init(void)
408{
409	major = register_blkdev(0, "virtblk");
410	if (major < 0)
411		return major;
412	return register_virtio_driver(&virtio_blk);
413}
414
415static void __exit fini(void)
416{
417	unregister_blkdev(major, "virtblk");
418	unregister_virtio_driver(&virtio_blk);
419}
420module_init(init);
421module_exit(fini);
422
423MODULE_DEVICE_TABLE(virtio, id_table);
424MODULE_DESCRIPTION("Virtio block driver");
425MODULE_LICENSE("GPL");
426