virtio_blk.c revision 1f0918d03ff4b5c94540c71ce889672abdbc2f4a 
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_ids.h>
7#include <linux/virtio_blk.h>
8#include <linux/scatterlist.h>
9
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	/* What host tells us, plus 2 for header & tailer. */
30	unsigned int sg_elems;
31
32	/* Scatterlist: can be too big for stack. */
33	struct scatterlist sg[/*sg_elems*/];
34};
35
36struct virtblk_req
37{
38	struct list_head list;
39	struct request *req;
40	struct virtio_blk_outhdr out_hdr;
41	struct virtio_scsi_inhdr in_hdr;
42	u8 status;
43};
44
45static void blk_done(struct virtqueue *vq)
46{
47	struct virtio_blk *vblk = vq->vdev->priv;
48	struct virtblk_req *vbr;
49	unsigned int len;
50	unsigned long flags;
51
52	spin_lock_irqsave(&vblk->lock, flags);
53	while ((vbr = vblk->vq->vq_ops->get_buf(vblk->vq, &len)) != NULL) {
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	switch (req->cmd_type) {
96	case REQ_TYPE_FS:
97		vbr->out_hdr.type = 0;
98		vbr->out_hdr.sector = blk_rq_pos(vbr->req);
99		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
100		break;
101	case REQ_TYPE_BLOCK_PC:
102		vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
103		vbr->out_hdr.sector = 0;
104		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
105		break;
106	case REQ_TYPE_LINUX_BLOCK:
107		if (req->cmd[0] == REQ_LB_OP_FLUSH) {
108			vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
109			vbr->out_hdr.sector = 0;
110			vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
111			break;
112		}
113		/*FALLTHRU*/
114	default:
115		/* We don't put anything else in the queue. */
116		BUG();
117	}
118
119	if (blk_barrier_rq(vbr->req))
120		vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
121
122	sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
123
124	/*
125	 * If this is a packet command we need a couple of additional headers.
126	 * Behind the normal outhdr we put a segment with the scsi command
127	 * block, and before the normal inhdr we put the sense data and the
128	 * inhdr with additional status information before the normal inhdr.
129	 */
130	if (blk_pc_request(vbr->req))
131		sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
132
133	num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
134
135	if (blk_pc_request(vbr->req)) {
136		sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
137		sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
138			   sizeof(vbr->in_hdr));
139	}
140
141	sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
142		   sizeof(vbr->status));
143
144	if (num) {
145		if (rq_data_dir(vbr->req) == WRITE) {
146			vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
147			out += num;
148		} else {
149			vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
150			in += num;
151		}
152	}
153
154	if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
155		mempool_free(vbr, vblk->pool);
156		return false;
157	}
158
159	list_add_tail(&vbr->list, &vblk->reqs);
160	return true;
161}
162
163static void do_virtblk_request(struct request_queue *q)
164{
165	struct virtio_blk *vblk = q->queuedata;
166	struct request *req;
167	unsigned int issued = 0;
168
169	while ((req = blk_peek_request(q)) != NULL) {
170		BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
171
172		/* If this request fails, stop queue and wait for something to
173		   finish to restart it. */
174		if (!do_req(q, vblk, req)) {
175			blk_stop_queue(q);
176			break;
177		}
178		blk_start_request(req);
179		issued++;
180	}
181
182	if (issued)
183		vblk->vq->vq_ops->kick(vblk->vq);
184}
185
186/* return ATA identify data
187 */
188static int virtblk_identify(struct gendisk *disk, void *argp)
189{
190	struct virtio_blk *vblk = disk->private_data;
191	void *opaque;
192	int err = -ENOMEM;
193
194	opaque = kmalloc(VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
195	if (!opaque)
196		goto out;
197
198	err = virtio_config_buf(vblk->vdev, VIRTIO_BLK_F_IDENTIFY,
199		offsetof(struct virtio_blk_config, identify), opaque,
200		VIRTIO_BLK_ID_BYTES);
201
202	if (err)
203		goto out_kfree;
204
205	if (copy_to_user(argp, opaque, VIRTIO_BLK_ID_BYTES))
206		err = -EFAULT;
207
208out_kfree:
209	kfree(opaque);
210out:
211	return err;
212}
213
214static void virtblk_prepare_flush(struct request_queue *q, struct request *req)
215{
216	req->cmd_type = REQ_TYPE_LINUX_BLOCK;
217	req->cmd[0] = REQ_LB_OP_FLUSH;
218}
219
220static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
221			 unsigned cmd, unsigned long data)
222{
223	struct gendisk *disk = bdev->bd_disk;
224	struct virtio_blk *vblk = disk->private_data;
225	void __user *argp = (void __user *)data;
226
227	if (cmd == HDIO_GET_IDENTITY)
228		return virtblk_identify(disk, argp);
229
230	/*
231	 * Only allow the generic SCSI ioctls if the host can support it.
232	 */
233	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
234		return -ENOTTY;
235
236	return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp);
237}
238
239/* We provide getgeo only to please some old bootloader/partitioning tools */
240static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
241{
242	struct virtio_blk *vblk = bd->bd_disk->private_data;
243	struct virtio_blk_geometry vgeo;
244	int err;
245
246	/* see if the host passed in geometry config */
247	err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
248				offsetof(struct virtio_blk_config, geometry),
249				&vgeo);
250
251	if (!err) {
252		geo->heads = vgeo.heads;
253		geo->sectors = vgeo.sectors;
254		geo->cylinders = vgeo.cylinders;
255	} else {
256		/* some standard values, similar to sd */
257		geo->heads = 1 << 6;
258		geo->sectors = 1 << 5;
259		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
260	}
261	return 0;
262}
263
264static const struct block_device_operations virtblk_fops = {
265	.locked_ioctl = virtblk_ioctl,
266	.owner  = THIS_MODULE,
267	.getgeo = virtblk_getgeo,
268};
269
270static int index_to_minor(int index)
271{
272	return index << PART_BITS;
273}
274
275static int __devinit virtblk_probe(struct virtio_device *vdev)
276{
277	struct virtio_blk *vblk;
278	int err;
279	u64 cap;
280	u32 v;
281	u32 blk_size, sg_elems;
282
283	if (index_to_minor(index) >= 1 << MINORBITS)
284		return -ENOSPC;
285
286	/* We need to know how many segments before we allocate. */
287	err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
288				offsetof(struct virtio_blk_config, seg_max),
289				&sg_elems);
290	if (err)
291		sg_elems = 1;
292
293	/* We need an extra sg elements at head and tail. */
294	sg_elems += 2;
295	vdev->priv = vblk = kmalloc(sizeof(*vblk) +
296				    sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
297	if (!vblk) {
298		err = -ENOMEM;
299		goto out;
300	}
301
302	INIT_LIST_HEAD(&vblk->reqs);
303	spin_lock_init(&vblk->lock);
304	vblk->vdev = vdev;
305	vblk->sg_elems = sg_elems;
306	sg_init_table(vblk->sg, vblk->sg_elems);
307
308	/* We expect one virtqueue, for output. */
309	vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
310	if (IS_ERR(vblk->vq)) {
311		err = PTR_ERR(vblk->vq);
312		goto out_free_vblk;
313	}
314
315	vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
316	if (!vblk->pool) {
317		err = -ENOMEM;
318		goto out_free_vq;
319	}
320
321	/* FIXME: How many partitions?  How long is a piece of string? */
322	vblk->disk = alloc_disk(1 << PART_BITS);
323	if (!vblk->disk) {
324		err = -ENOMEM;
325		goto out_mempool;
326	}
327
328	vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
329	if (!vblk->disk->queue) {
330		err = -ENOMEM;
331		goto out_put_disk;
332	}
333
334	vblk->disk->queue->queuedata = vblk;
335	queue_flag_set_unlocked(QUEUE_FLAG_VIRT, vblk->disk->queue);
336
337	if (index < 26) {
338		sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
339	} else if (index < (26 + 1) * 26) {
340		sprintf(vblk->disk->disk_name, "vd%c%c",
341			'a' + index / 26 - 1, 'a' + index % 26);
342	} else {
343		const unsigned int m1 = (index / 26 - 1) / 26 - 1;
344		const unsigned int m2 = (index / 26 - 1) % 26;
345		const unsigned int m3 =  index % 26;
346		sprintf(vblk->disk->disk_name, "vd%c%c%c",
347			'a' + m1, 'a' + m2, 'a' + m3);
348	}
349
350	vblk->disk->major = major;
351	vblk->disk->first_minor = index_to_minor(index);
352	vblk->disk->private_data = vblk;
353	vblk->disk->fops = &virtblk_fops;
354	vblk->disk->driverfs_dev = &vdev->dev;
355	index++;
356
357	/* If barriers are supported, tell block layer that queue is ordered */
358	if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
359		blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_DRAIN_FLUSH,
360				  virtblk_prepare_flush);
361	else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
362		blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL);
363
364	/* If disk is read-only in the host, the guest should obey */
365	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
366		set_disk_ro(vblk->disk, 1);
367
368	/* Host must always specify the capacity. */
369	vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
370			  &cap, sizeof(cap));
371
372	/* If capacity is too big, truncate with warning. */
373	if ((sector_t)cap != cap) {
374		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
375			 (unsigned long long)cap);
376		cap = (sector_t)-1;
377	}
378	set_capacity(vblk->disk, cap);
379
380	/* We can handle whatever the host told us to handle. */
381	blk_queue_max_phys_segments(vblk->disk->queue, vblk->sg_elems-2);
382	blk_queue_max_hw_segments(vblk->disk->queue, vblk->sg_elems-2);
383
384	/* No need to bounce any requests */
385	blk_queue_bounce_limit(vblk->disk->queue, BLK_BOUNCE_ANY);
386
387	/* No real sector limit. */
388	blk_queue_max_sectors(vblk->disk->queue, -1U);
389
390	/* Host can optionally specify maximum segment size and number of
391	 * segments. */
392	err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
393				offsetof(struct virtio_blk_config, size_max),
394				&v);
395	if (!err)
396		blk_queue_max_segment_size(vblk->disk->queue, v);
397	else
398		blk_queue_max_segment_size(vblk->disk->queue, -1U);
399
400	/* Host can optionally specify the block size of the device */
401	err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
402				offsetof(struct virtio_blk_config, blk_size),
403				&blk_size);
404	if (!err)
405		blk_queue_logical_block_size(vblk->disk->queue, blk_size);
406
407	add_disk(vblk->disk);
408	return 0;
409
410out_put_disk:
411	put_disk(vblk->disk);
412out_mempool:
413	mempool_destroy(vblk->pool);
414out_free_vq:
415	vdev->config->del_vqs(vdev);
416out_free_vblk:
417	kfree(vblk);
418out:
419	return err;
420}
421
422static void __devexit virtblk_remove(struct virtio_device *vdev)
423{
424	struct virtio_blk *vblk = vdev->priv;
425
426	/* Nothing should be pending. */
427	BUG_ON(!list_empty(&vblk->reqs));
428
429	/* Stop all the virtqueues. */
430	vdev->config->reset(vdev);
431
432	del_gendisk(vblk->disk);
433	blk_cleanup_queue(vblk->disk->queue);
434	put_disk(vblk->disk);
435	mempool_destroy(vblk->pool);
436	vdev->config->del_vqs(vdev);
437	kfree(vblk);
438}
439
440static struct virtio_device_id id_table[] = {
441	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
442	{ 0 },
443};
444
445static unsigned int features[] = {
446	VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
447	VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
448	VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_IDENTIFY, VIRTIO_BLK_F_FLUSH
449};
450
451/*
452 * virtio_blk causes spurious section mismatch warning by
453 * simultaneously referring to a __devinit and a __devexit function.
454 * Use __refdata to avoid this warning.
455 */
456static struct virtio_driver __refdata virtio_blk = {
457	.feature_table = features,
458	.feature_table_size = ARRAY_SIZE(features),
459	.driver.name =	KBUILD_MODNAME,
460	.driver.owner =	THIS_MODULE,
461	.id_table =	id_table,
462	.probe =	virtblk_probe,
463	.remove =	__devexit_p(virtblk_remove),
464};
465
466static int __init init(void)
467{
468	major = register_blkdev(0, "virtblk");
469	if (major < 0)
470		return major;
471	return register_virtio_driver(&virtio_blk);
472}
473
474static void __exit fini(void)
475{
476	unregister_blkdev(major, "virtblk");
477	unregister_virtio_driver(&virtio_blk);
478}
479module_init(init);
480module_exit(fini);
481
482MODULE_DEVICE_TABLE(virtio, id_table);
483MODULE_DESCRIPTION("Virtio block driver");
484MODULE_LICENSE("GPL");
485