virtio_blk.c revision 33659ebbae262228eef4e0fe990f393d1f0ed941 
1//#define DEBUG
2#include <linux/spinlock.h>
3#include <linux/slab.h>
4#include <linux/blkdev.h>
5#include <linux/hdreg.h>
6#include <linux/virtio.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 = virtqueue_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		switch (vbr->req->cmd_type) {
69		case REQ_TYPE_BLOCK_PC:
70			vbr->req->resid_len = vbr->in_hdr.residual;
71			vbr->req->sense_len = vbr->in_hdr.sense_len;
72			vbr->req->errors = vbr->in_hdr.errors;
73			break;
74		case REQ_TYPE_SPECIAL:
75			vbr->req->errors = (error != 0);
76			break;
77		}
78
79		__blk_end_request_all(vbr->req, error);
80		list_del(&vbr->list);
81		mempool_free(vbr, vblk->pool);
82	}
83	/* In case queue is stopped waiting for more buffers. */
84	blk_start_queue(vblk->disk->queue);
85	spin_unlock_irqrestore(&vblk->lock, flags);
86}
87
88static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
89		   struct request *req)
90{
91	unsigned long num, out = 0, in = 0;
92	struct virtblk_req *vbr;
93
94	vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
95	if (!vbr)
96		/* When another request finishes we'll try again. */
97		return false;
98
99	vbr->req = req;
100	switch (req->cmd_type) {
101	case REQ_TYPE_FS:
102		vbr->out_hdr.type = 0;
103		vbr->out_hdr.sector = blk_rq_pos(vbr->req);
104		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
105		break;
106	case REQ_TYPE_BLOCK_PC:
107		vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
108		vbr->out_hdr.sector = 0;
109		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
110		break;
111	case REQ_TYPE_SPECIAL:
112		vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
113		vbr->out_hdr.sector = 0;
114		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
115		break;
116	case REQ_TYPE_LINUX_BLOCK:
117		if (req->cmd[0] == REQ_LB_OP_FLUSH) {
118			vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
119			vbr->out_hdr.sector = 0;
120			vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
121			break;
122		}
123		/*FALLTHRU*/
124	default:
125		/* We don't put anything else in the queue. */
126		BUG();
127	}
128
129	if (vbr->req->cmd_flags & REQ_HARDBARRIER)
130		vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
131
132	sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
133
134	/*
135	 * If this is a packet command we need a couple of additional headers.
136	 * Behind the normal outhdr we put a segment with the scsi command
137	 * block, and before the normal inhdr we put the sense data and the
138	 * inhdr with additional status information before the normal inhdr.
139	 */
140	if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
141		sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
142
143	num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
144
145	if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
146		sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
147		sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
148			   sizeof(vbr->in_hdr));
149	}
150
151	sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
152		   sizeof(vbr->status));
153
154	if (num) {
155		if (rq_data_dir(vbr->req) == WRITE) {
156			vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
157			out += num;
158		} else {
159			vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
160			in += num;
161		}
162	}
163
164	if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
165		mempool_free(vbr, vblk->pool);
166		return false;
167	}
168
169	list_add_tail(&vbr->list, &vblk->reqs);
170	return true;
171}
172
173static void do_virtblk_request(struct request_queue *q)
174{
175	struct virtio_blk *vblk = q->queuedata;
176	struct request *req;
177	unsigned int issued = 0;
178
179	while ((req = blk_peek_request(q)) != NULL) {
180		BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
181
182		/* If this request fails, stop queue and wait for something to
183		   finish to restart it. */
184		if (!do_req(q, vblk, req)) {
185			blk_stop_queue(q);
186			break;
187		}
188		blk_start_request(req);
189		issued++;
190	}
191
192	if (issued)
193		virtqueue_kick(vblk->vq);
194}
195
196static void virtblk_prepare_flush(struct request_queue *q, struct request *req)
197{
198	req->cmd_type = REQ_TYPE_LINUX_BLOCK;
199	req->cmd[0] = REQ_LB_OP_FLUSH;
200}
201
202/* return id (s/n) string for *disk to *id_str
203 */
204static int virtblk_get_id(struct gendisk *disk, char *id_str)
205{
206	struct virtio_blk *vblk = disk->private_data;
207	struct request *req;
208	struct bio *bio;
209
210	bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
211			   GFP_KERNEL);
212	if (IS_ERR(bio))
213		return PTR_ERR(bio);
214
215	req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
216	if (IS_ERR(req)) {
217		bio_put(bio);
218		return PTR_ERR(req);
219	}
220
221	req->cmd_type = REQ_TYPE_SPECIAL;
222	return blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
223}
224
225static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
226			 unsigned cmd, unsigned long data)
227{
228	struct gendisk *disk = bdev->bd_disk;
229	struct virtio_blk *vblk = disk->private_data;
230
231	if (cmd == 0x56424944) { /* 'VBID' */
232		void __user *usr_data = (void __user *)data;
233		char id_str[VIRTIO_BLK_ID_BYTES];
234		int err;
235
236		err = virtblk_get_id(disk, id_str);
237		if (!err && copy_to_user(usr_data, id_str, VIRTIO_BLK_ID_BYTES))
238			err = -EFAULT;
239		return err;
240	}
241	/*
242	 * Only allow the generic SCSI ioctls if the host can support it.
243	 */
244	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
245		return -ENOTTY;
246
247	return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
248			      (void __user *)data);
249}
250
251/* We provide getgeo only to please some old bootloader/partitioning tools */
252static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
253{
254	struct virtio_blk *vblk = bd->bd_disk->private_data;
255	struct virtio_blk_geometry vgeo;
256	int err;
257
258	/* see if the host passed in geometry config */
259	err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
260				offsetof(struct virtio_blk_config, geometry),
261				&vgeo);
262
263	if (!err) {
264		geo->heads = vgeo.heads;
265		geo->sectors = vgeo.sectors;
266		geo->cylinders = vgeo.cylinders;
267	} else {
268		/* some standard values, similar to sd */
269		geo->heads = 1 << 6;
270		geo->sectors = 1 << 5;
271		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
272	}
273	return 0;
274}
275
276static const struct block_device_operations virtblk_fops = {
277	.locked_ioctl = virtblk_ioctl,
278	.owner  = THIS_MODULE,
279	.getgeo = virtblk_getgeo,
280};
281
282static int index_to_minor(int index)
283{
284	return index << PART_BITS;
285}
286
287static int __devinit virtblk_probe(struct virtio_device *vdev)
288{
289	struct virtio_blk *vblk;
290	struct request_queue *q;
291	int err;
292	u64 cap;
293	u32 v, blk_size, sg_elems, opt_io_size;
294	u16 min_io_size;
295	u8 physical_block_exp, alignment_offset;
296
297	if (index_to_minor(index) >= 1 << MINORBITS)
298		return -ENOSPC;
299
300	/* We need to know how many segments before we allocate. */
301	err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
302				offsetof(struct virtio_blk_config, seg_max),
303				&sg_elems);
304
305	/* We need at least one SG element, whatever they say. */
306	if (err || !sg_elems)
307		sg_elems = 1;
308
309	/* We need an extra sg elements at head and tail. */
310	sg_elems += 2;
311	vdev->priv = vblk = kmalloc(sizeof(*vblk) +
312				    sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
313	if (!vblk) {
314		err = -ENOMEM;
315		goto out;
316	}
317
318	INIT_LIST_HEAD(&vblk->reqs);
319	spin_lock_init(&vblk->lock);
320	vblk->vdev = vdev;
321	vblk->sg_elems = sg_elems;
322	sg_init_table(vblk->sg, vblk->sg_elems);
323
324	/* We expect one virtqueue, for output. */
325	vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
326	if (IS_ERR(vblk->vq)) {
327		err = PTR_ERR(vblk->vq);
328		goto out_free_vblk;
329	}
330
331	vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
332	if (!vblk->pool) {
333		err = -ENOMEM;
334		goto out_free_vq;
335	}
336
337	/* FIXME: How many partitions?  How long is a piece of string? */
338	vblk->disk = alloc_disk(1 << PART_BITS);
339	if (!vblk->disk) {
340		err = -ENOMEM;
341		goto out_mempool;
342	}
343
344	q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
345	if (!q) {
346		err = -ENOMEM;
347		goto out_put_disk;
348	}
349
350	q->queuedata = vblk;
351
352	if (index < 26) {
353		sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
354	} else if (index < (26 + 1) * 26) {
355		sprintf(vblk->disk->disk_name, "vd%c%c",
356			'a' + index / 26 - 1, 'a' + index % 26);
357	} else {
358		const unsigned int m1 = (index / 26 - 1) / 26 - 1;
359		const unsigned int m2 = (index / 26 - 1) % 26;
360		const unsigned int m3 =  index % 26;
361		sprintf(vblk->disk->disk_name, "vd%c%c%c",
362			'a' + m1, 'a' + m2, 'a' + m3);
363	}
364
365	vblk->disk->major = major;
366	vblk->disk->first_minor = index_to_minor(index);
367	vblk->disk->private_data = vblk;
368	vblk->disk->fops = &virtblk_fops;
369	vblk->disk->driverfs_dev = &vdev->dev;
370	index++;
371
372	/* If barriers are supported, tell block layer that queue is ordered */
373	if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
374		blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH,
375				  virtblk_prepare_flush);
376	else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
377		blk_queue_ordered(q, QUEUE_ORDERED_TAG, NULL);
378
379	/* If disk is read-only in the host, the guest should obey */
380	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
381		set_disk_ro(vblk->disk, 1);
382
383	/* Host must always specify the capacity. */
384	vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
385			  &cap, sizeof(cap));
386
387	/* If capacity is too big, truncate with warning. */
388	if ((sector_t)cap != cap) {
389		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
390			 (unsigned long long)cap);
391		cap = (sector_t)-1;
392	}
393	set_capacity(vblk->disk, cap);
394
395	/* We can handle whatever the host told us to handle. */
396	blk_queue_max_segments(q, vblk->sg_elems-2);
397
398	/* No need to bounce any requests */
399	blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
400
401	/* No real sector limit. */
402	blk_queue_max_hw_sectors(q, -1U);
403
404	/* Host can optionally specify maximum segment size and number of
405	 * segments. */
406	err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
407				offsetof(struct virtio_blk_config, size_max),
408				&v);
409	if (!err)
410		blk_queue_max_segment_size(q, v);
411	else
412		blk_queue_max_segment_size(q, -1U);
413
414	/* Host can optionally specify the block size of the device */
415	err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
416				offsetof(struct virtio_blk_config, blk_size),
417				&blk_size);
418	if (!err)
419		blk_queue_logical_block_size(q, blk_size);
420	else
421		blk_size = queue_logical_block_size(q);
422
423	/* Use topology information if available */
424	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
425			offsetof(struct virtio_blk_config, physical_block_exp),
426			&physical_block_exp);
427	if (!err && physical_block_exp)
428		blk_queue_physical_block_size(q,
429				blk_size * (1 << physical_block_exp));
430
431	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
432			offsetof(struct virtio_blk_config, alignment_offset),
433			&alignment_offset);
434	if (!err && alignment_offset)
435		blk_queue_alignment_offset(q, blk_size * alignment_offset);
436
437	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
438			offsetof(struct virtio_blk_config, min_io_size),
439			&min_io_size);
440	if (!err && min_io_size)
441		blk_queue_io_min(q, blk_size * min_io_size);
442
443	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
444			offsetof(struct virtio_blk_config, opt_io_size),
445			&opt_io_size);
446	if (!err && opt_io_size)
447		blk_queue_io_opt(q, blk_size * opt_io_size);
448
449
450	add_disk(vblk->disk);
451	return 0;
452
453out_put_disk:
454	put_disk(vblk->disk);
455out_mempool:
456	mempool_destroy(vblk->pool);
457out_free_vq:
458	vdev->config->del_vqs(vdev);
459out_free_vblk:
460	kfree(vblk);
461out:
462	return err;
463}
464
465static void __devexit virtblk_remove(struct virtio_device *vdev)
466{
467	struct virtio_blk *vblk = vdev->priv;
468
469	/* Nothing should be pending. */
470	BUG_ON(!list_empty(&vblk->reqs));
471
472	/* Stop all the virtqueues. */
473	vdev->config->reset(vdev);
474
475	del_gendisk(vblk->disk);
476	blk_cleanup_queue(vblk->disk->queue);
477	put_disk(vblk->disk);
478	mempool_destroy(vblk->pool);
479	vdev->config->del_vqs(vdev);
480	kfree(vblk);
481}
482
483static const struct virtio_device_id id_table[] = {
484	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
485	{ 0 },
486};
487
488static unsigned int features[] = {
489	VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
490	VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
491	VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
492};
493
494/*
495 * virtio_blk causes spurious section mismatch warning by
496 * simultaneously referring to a __devinit and a __devexit function.
497 * Use __refdata to avoid this warning.
498 */
499static struct virtio_driver __refdata virtio_blk = {
500	.feature_table = features,
501	.feature_table_size = ARRAY_SIZE(features),
502	.driver.name =	KBUILD_MODNAME,
503	.driver.owner =	THIS_MODULE,
504	.id_table =	id_table,
505	.probe =	virtblk_probe,
506	.remove =	__devexit_p(virtblk_remove),
507};
508
509static int __init init(void)
510{
511	major = register_blkdev(0, "virtblk");
512	if (major < 0)
513		return major;
514	return register_virtio_driver(&virtio_blk);
515}
516
517static void __exit fini(void)
518{
519	unregister_blkdev(major, "virtblk");
520	unregister_virtio_driver(&virtio_blk);
521}
522module_init(init);
523module_exit(fini);
524
525MODULE_DEVICE_TABLE(virtio, id_table);
526MODULE_DESCRIPTION("Virtio block driver");
527MODULE_LICENSE("GPL");
528