virtio_blk.c revision 4cb2ea28c55cf5e5ef83aec535099ffce3c583df 
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 = 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		if (blk_special_request(vbr->req))
74			vbr->req->errors = (error != 0);
75
76		__blk_end_request_all(vbr->req, error);
77		list_del(&vbr->list);
78		mempool_free(vbr, vblk->pool);
79	}
80	/* In case queue is stopped waiting for more buffers. */
81	blk_start_queue(vblk->disk->queue);
82	spin_unlock_irqrestore(&vblk->lock, flags);
83}
84
85static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
86		   struct request *req)
87{
88	unsigned long num, out = 0, in = 0;
89	struct virtblk_req *vbr;
90
91	vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
92	if (!vbr)
93		/* When another request finishes we'll try again. */
94		return false;
95
96	vbr->req = req;
97	switch (req->cmd_type) {
98	case REQ_TYPE_FS:
99		vbr->out_hdr.type = 0;
100		vbr->out_hdr.sector = blk_rq_pos(vbr->req);
101		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
102		break;
103	case REQ_TYPE_BLOCK_PC:
104		vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
105		vbr->out_hdr.sector = 0;
106		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
107		break;
108	case REQ_TYPE_SPECIAL:
109		vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
110		vbr->out_hdr.sector = 0;
111		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
112		break;
113	case REQ_TYPE_LINUX_BLOCK:
114		if (req->cmd[0] == REQ_LB_OP_FLUSH) {
115			vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
116			vbr->out_hdr.sector = 0;
117			vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
118			break;
119		}
120		/*FALLTHRU*/
121	default:
122		/* We don't put anything else in the queue. */
123		BUG();
124	}
125
126	if (blk_barrier_rq(vbr->req))
127		vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
128
129	sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
130
131	/*
132	 * If this is a packet command we need a couple of additional headers.
133	 * Behind the normal outhdr we put a segment with the scsi command
134	 * block, and before the normal inhdr we put the sense data and the
135	 * inhdr with additional status information before the normal inhdr.
136	 */
137	if (blk_pc_request(vbr->req))
138		sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
139
140	num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
141
142	if (blk_pc_request(vbr->req)) {
143		sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
144		sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
145			   sizeof(vbr->in_hdr));
146	}
147
148	sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
149		   sizeof(vbr->status));
150
151	if (num) {
152		if (rq_data_dir(vbr->req) == WRITE) {
153			vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
154			out += num;
155		} else {
156			vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
157			in += num;
158		}
159	}
160
161	if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
162		mempool_free(vbr, vblk->pool);
163		return false;
164	}
165
166	list_add_tail(&vbr->list, &vblk->reqs);
167	return true;
168}
169
170static void do_virtblk_request(struct request_queue *q)
171{
172	struct virtio_blk *vblk = q->queuedata;
173	struct request *req;
174	unsigned int issued = 0;
175
176	while ((req = blk_peek_request(q)) != NULL) {
177		BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
178
179		/* If this request fails, stop queue and wait for something to
180		   finish to restart it. */
181		if (!do_req(q, vblk, req)) {
182			blk_stop_queue(q);
183			break;
184		}
185		blk_start_request(req);
186		issued++;
187	}
188
189	if (issued)
190		vblk->vq->vq_ops->kick(vblk->vq);
191}
192
193static void virtblk_prepare_flush(struct request_queue *q, struct request *req)
194{
195	req->cmd_type = REQ_TYPE_LINUX_BLOCK;
196	req->cmd[0] = REQ_LB_OP_FLUSH;
197}
198
199/* return id (s/n) string for *disk to *id_str
200 */
201static int virtblk_get_id(struct gendisk *disk, char *id_str)
202{
203	struct virtio_blk *vblk = disk->private_data;
204	struct request *req;
205	struct bio *bio;
206
207	bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
208			   GFP_KERNEL);
209	if (IS_ERR(bio))
210		return PTR_ERR(bio);
211
212	req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
213	if (IS_ERR(req)) {
214		bio_put(bio);
215		return PTR_ERR(req);
216	}
217
218	req->cmd_type = REQ_TYPE_SPECIAL;
219	return blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
220}
221
222static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
223			 unsigned cmd, unsigned long data)
224{
225	struct gendisk *disk = bdev->bd_disk;
226	struct virtio_blk *vblk = disk->private_data;
227
228	/*
229	 * Only allow the generic SCSI ioctls if the host can support it.
230	 */
231	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
232		return -ENOTTY;
233
234	return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
235			      (void __user *)data);
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	struct request_queue *q;
278	int err;
279	u64 cap;
280	u32 v, blk_size, sg_elems, opt_io_size;
281	u16 min_io_size;
282	u8 physical_block_exp, alignment_offset;
283
284	if (index_to_minor(index) >= 1 << MINORBITS)
285		return -ENOSPC;
286
287	/* We need to know how many segments before we allocate. */
288	err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
289				offsetof(struct virtio_blk_config, seg_max),
290				&sg_elems);
291	if (err)
292		sg_elems = 1;
293
294	/* We need an extra sg elements at head and tail. */
295	sg_elems += 2;
296	vdev->priv = vblk = kmalloc(sizeof(*vblk) +
297				    sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
298	if (!vblk) {
299		err = -ENOMEM;
300		goto out;
301	}
302
303	INIT_LIST_HEAD(&vblk->reqs);
304	spin_lock_init(&vblk->lock);
305	vblk->vdev = vdev;
306	vblk->sg_elems = sg_elems;
307	sg_init_table(vblk->sg, vblk->sg_elems);
308
309	/* We expect one virtqueue, for output. */
310	vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
311	if (IS_ERR(vblk->vq)) {
312		err = PTR_ERR(vblk->vq);
313		goto out_free_vblk;
314	}
315
316	vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
317	if (!vblk->pool) {
318		err = -ENOMEM;
319		goto out_free_vq;
320	}
321
322	/* FIXME: How many partitions?  How long is a piece of string? */
323	vblk->disk = alloc_disk(1 << PART_BITS);
324	if (!vblk->disk) {
325		err = -ENOMEM;
326		goto out_mempool;
327	}
328
329	q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
330	if (!q) {
331		err = -ENOMEM;
332		goto out_put_disk;
333	}
334
335	q->queuedata = vblk;
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(q, QUEUE_ORDERED_DRAIN_FLUSH,
360				  virtblk_prepare_flush);
361	else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER))
362		blk_queue_ordered(q, 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_segments(q, vblk->sg_elems-2);
382
383	/* No need to bounce any requests */
384	blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
385
386	/* No real sector limit. */
387	blk_queue_max_hw_sectors(q, -1U);
388
389	/* Host can optionally specify maximum segment size and number of
390	 * segments. */
391	err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
392				offsetof(struct virtio_blk_config, size_max),
393				&v);
394	if (!err)
395		blk_queue_max_segment_size(q, v);
396	else
397		blk_queue_max_segment_size(q, -1U);
398
399	/* Host can optionally specify the block size of the device */
400	err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
401				offsetof(struct virtio_blk_config, blk_size),
402				&blk_size);
403	if (!err)
404		blk_queue_logical_block_size(q, blk_size);
405	else
406		blk_size = queue_logical_block_size(q);
407
408	/* Use topology information if available */
409	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
410			offsetof(struct virtio_blk_config, physical_block_exp),
411			&physical_block_exp);
412	if (!err && physical_block_exp)
413		blk_queue_physical_block_size(q,
414				blk_size * (1 << physical_block_exp));
415
416	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
417			offsetof(struct virtio_blk_config, alignment_offset),
418			&alignment_offset);
419	if (!err && alignment_offset)
420		blk_queue_alignment_offset(q, blk_size * alignment_offset);
421
422	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
423			offsetof(struct virtio_blk_config, min_io_size),
424			&min_io_size);
425	if (!err && min_io_size)
426		blk_queue_io_min(q, blk_size * min_io_size);
427
428	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
429			offsetof(struct virtio_blk_config, opt_io_size),
430			&opt_io_size);
431	if (!err && opt_io_size)
432		blk_queue_io_opt(q, blk_size * opt_io_size);
433
434
435	add_disk(vblk->disk);
436	return 0;
437
438out_put_disk:
439	put_disk(vblk->disk);
440out_mempool:
441	mempool_destroy(vblk->pool);
442out_free_vq:
443	vdev->config->del_vqs(vdev);
444out_free_vblk:
445	kfree(vblk);
446out:
447	return err;
448}
449
450static void __devexit virtblk_remove(struct virtio_device *vdev)
451{
452	struct virtio_blk *vblk = vdev->priv;
453
454	/* Nothing should be pending. */
455	BUG_ON(!list_empty(&vblk->reqs));
456
457	/* Stop all the virtqueues. */
458	vdev->config->reset(vdev);
459
460	del_gendisk(vblk->disk);
461	blk_cleanup_queue(vblk->disk->queue);
462	put_disk(vblk->disk);
463	mempool_destroy(vblk->pool);
464	vdev->config->del_vqs(vdev);
465	kfree(vblk);
466}
467
468static const struct virtio_device_id id_table[] = {
469	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
470	{ 0 },
471};
472
473static unsigned int features[] = {
474	VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
475	VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
476	VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
477};
478
479/*
480 * virtio_blk causes spurious section mismatch warning by
481 * simultaneously referring to a __devinit and a __devexit function.
482 * Use __refdata to avoid this warning.
483 */
484static struct virtio_driver __refdata virtio_blk = {
485	.feature_table = features,
486	.feature_table_size = ARRAY_SIZE(features),
487	.driver.name =	KBUILD_MODNAME,
488	.driver.owner =	THIS_MODULE,
489	.id_table =	id_table,
490	.probe =	virtblk_probe,
491	.remove =	__devexit_p(virtblk_remove),
492};
493
494static int __init init(void)
495{
496	major = register_blkdev(0, "virtblk");
497	if (major < 0)
498		return major;
499	return register_virtio_driver(&virtio_blk);
500}
501
502static void __exit fini(void)
503{
504	unregister_blkdev(major, "virtblk");
505	unregister_virtio_driver(&virtio_blk);
506}
507module_init(init);
508module_exit(fini);
509
510MODULE_DEVICE_TABLE(virtio, id_table);
511MODULE_DESCRIPTION("Virtio block driver");
512MODULE_LICENSE("GPL");
513