xref: /openbmc/linux/drivers/block/virtio_blk.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
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 
12 static int major, index;
13 
14 struct 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 
36 struct 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 
45 static 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 		default:
78 			break;
79 		}
80 
81 		__blk_end_request_all(vbr->req, error);
82 		list_del(&vbr->list);
83 		mempool_free(vbr, vblk->pool);
84 	}
85 	/* In case queue is stopped waiting for more buffers. */
86 	blk_start_queue(vblk->disk->queue);
87 	spin_unlock_irqrestore(&vblk->lock, flags);
88 }
89 
90 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
91 		   struct request *req)
92 {
93 	unsigned long num, out = 0, in = 0;
94 	struct virtblk_req *vbr;
95 
96 	vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
97 	if (!vbr)
98 		/* When another request finishes we'll try again. */
99 		return false;
100 
101 	vbr->req = req;
102 
103 	if (req->cmd_flags & REQ_FLUSH) {
104 		vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
105 		vbr->out_hdr.sector = 0;
106 		vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
107 	} else {
108 		switch (req->cmd_type) {
109 		case REQ_TYPE_FS:
110 			vbr->out_hdr.type = 0;
111 			vbr->out_hdr.sector = blk_rq_pos(vbr->req);
112 			vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
113 			break;
114 		case REQ_TYPE_BLOCK_PC:
115 			vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
116 			vbr->out_hdr.sector = 0;
117 			vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
118 			break;
119 		case REQ_TYPE_SPECIAL:
120 			vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
121 			vbr->out_hdr.sector = 0;
122 			vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
123 			break;
124 		default:
125 			/* We don't put anything else in the queue. */
126 			BUG();
127 		}
128 	}
129 
130 	sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
131 
132 	/*
133 	 * If this is a packet command we need a couple of additional headers.
134 	 * Behind the normal outhdr we put a segment with the scsi command
135 	 * block, and before the normal inhdr we put the sense data and the
136 	 * inhdr with additional status information before the normal inhdr.
137 	 */
138 	if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
139 		sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
140 
141 	num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
142 
143 	if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
144 		sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
145 		sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
146 			   sizeof(vbr->in_hdr));
147 	}
148 
149 	sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
150 		   sizeof(vbr->status));
151 
152 	if (num) {
153 		if (rq_data_dir(vbr->req) == WRITE) {
154 			vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
155 			out += num;
156 		} else {
157 			vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
158 			in += num;
159 		}
160 	}
161 
162 	if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
163 		mempool_free(vbr, vblk->pool);
164 		return false;
165 	}
166 
167 	list_add_tail(&vbr->list, &vblk->reqs);
168 	return true;
169 }
170 
171 static void do_virtblk_request(struct request_queue *q)
172 {
173 	struct virtio_blk *vblk = q->queuedata;
174 	struct request *req;
175 	unsigned int issued = 0;
176 
177 	while ((req = blk_peek_request(q)) != NULL) {
178 		BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
179 
180 		/* If this request fails, stop queue and wait for something to
181 		   finish to restart it. */
182 		if (!do_req(q, vblk, req)) {
183 			blk_stop_queue(q);
184 			break;
185 		}
186 		blk_start_request(req);
187 		issued++;
188 	}
189 
190 	if (issued)
191 		virtqueue_kick(vblk->vq);
192 }
193 
194 /* return id (s/n) string for *disk to *id_str
195  */
196 static int virtblk_get_id(struct gendisk *disk, char *id_str)
197 {
198 	struct virtio_blk *vblk = disk->private_data;
199 	struct request *req;
200 	struct bio *bio;
201 	int err;
202 
203 	bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
204 			   GFP_KERNEL);
205 	if (IS_ERR(bio))
206 		return PTR_ERR(bio);
207 
208 	req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
209 	if (IS_ERR(req)) {
210 		bio_put(bio);
211 		return PTR_ERR(req);
212 	}
213 
214 	req->cmd_type = REQ_TYPE_SPECIAL;
215 	err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
216 	blk_put_request(req);
217 
218 	return err;
219 }
220 
221 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
222 			     unsigned int cmd, unsigned long data)
223 {
224 	struct gendisk *disk = bdev->bd_disk;
225 	struct virtio_blk *vblk = disk->private_data;
226 
227 	/*
228 	 * Only allow the generic SCSI ioctls if the host can support it.
229 	 */
230 	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
231 		return -ENOTTY;
232 
233 	return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
234 			      (void __user *)data);
235 }
236 
237 /* We provide getgeo only to please some old bootloader/partitioning tools */
238 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
239 {
240 	struct virtio_blk *vblk = bd->bd_disk->private_data;
241 	struct virtio_blk_geometry vgeo;
242 	int err;
243 
244 	/* see if the host passed in geometry config */
245 	err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
246 				offsetof(struct virtio_blk_config, geometry),
247 				&vgeo);
248 
249 	if (!err) {
250 		geo->heads = vgeo.heads;
251 		geo->sectors = vgeo.sectors;
252 		geo->cylinders = vgeo.cylinders;
253 	} else {
254 		/* some standard values, similar to sd */
255 		geo->heads = 1 << 6;
256 		geo->sectors = 1 << 5;
257 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
258 	}
259 	return 0;
260 }
261 
262 static const struct block_device_operations virtblk_fops = {
263 	.ioctl  = virtblk_ioctl,
264 	.owner  = THIS_MODULE,
265 	.getgeo = virtblk_getgeo,
266 };
267 
268 static int index_to_minor(int index)
269 {
270 	return index << PART_BITS;
271 }
272 
273 static ssize_t virtblk_serial_show(struct device *dev,
274 				struct device_attribute *attr, char *buf)
275 {
276 	struct gendisk *disk = dev_to_disk(dev);
277 	int err;
278 
279 	/* sysfs gives us a PAGE_SIZE buffer */
280 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
281 
282 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
283 	err = virtblk_get_id(disk, buf);
284 	if (!err)
285 		return strlen(buf);
286 
287 	if (err == -EIO) /* Unsupported? Make it empty. */
288 		return 0;
289 
290 	return err;
291 }
292 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
293 
294 static int __devinit virtblk_probe(struct virtio_device *vdev)
295 {
296 	struct virtio_blk *vblk;
297 	struct request_queue *q;
298 	int err;
299 	u64 cap;
300 	u32 v, blk_size, sg_elems, opt_io_size;
301 	u16 min_io_size;
302 	u8 physical_block_exp, alignment_offset;
303 
304 	if (index_to_minor(index) >= 1 << MINORBITS)
305 		return -ENOSPC;
306 
307 	/* We need to know how many segments before we allocate. */
308 	err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
309 				offsetof(struct virtio_blk_config, seg_max),
310 				&sg_elems);
311 
312 	/* We need at least one SG element, whatever they say. */
313 	if (err || !sg_elems)
314 		sg_elems = 1;
315 
316 	/* We need an extra sg elements at head and tail. */
317 	sg_elems += 2;
318 	vdev->priv = vblk = kmalloc(sizeof(*vblk) +
319 				    sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
320 	if (!vblk) {
321 		err = -ENOMEM;
322 		goto out;
323 	}
324 
325 	INIT_LIST_HEAD(&vblk->reqs);
326 	spin_lock_init(&vblk->lock);
327 	vblk->vdev = vdev;
328 	vblk->sg_elems = sg_elems;
329 	sg_init_table(vblk->sg, vblk->sg_elems);
330 
331 	/* We expect one virtqueue, for output. */
332 	vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
333 	if (IS_ERR(vblk->vq)) {
334 		err = PTR_ERR(vblk->vq);
335 		goto out_free_vblk;
336 	}
337 
338 	vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
339 	if (!vblk->pool) {
340 		err = -ENOMEM;
341 		goto out_free_vq;
342 	}
343 
344 	/* FIXME: How many partitions?  How long is a piece of string? */
345 	vblk->disk = alloc_disk(1 << PART_BITS);
346 	if (!vblk->disk) {
347 		err = -ENOMEM;
348 		goto out_mempool;
349 	}
350 
351 	q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
352 	if (!q) {
353 		err = -ENOMEM;
354 		goto out_put_disk;
355 	}
356 
357 	q->queuedata = vblk;
358 
359 	if (index < 26) {
360 		sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
361 	} else if (index < (26 + 1) * 26) {
362 		sprintf(vblk->disk->disk_name, "vd%c%c",
363 			'a' + index / 26 - 1, 'a' + index % 26);
364 	} else {
365 		const unsigned int m1 = (index / 26 - 1) / 26 - 1;
366 		const unsigned int m2 = (index / 26 - 1) % 26;
367 		const unsigned int m3 =  index % 26;
368 		sprintf(vblk->disk->disk_name, "vd%c%c%c",
369 			'a' + m1, 'a' + m2, 'a' + m3);
370 	}
371 
372 	vblk->disk->major = major;
373 	vblk->disk->first_minor = index_to_minor(index);
374 	vblk->disk->private_data = vblk;
375 	vblk->disk->fops = &virtblk_fops;
376 	vblk->disk->driverfs_dev = &vdev->dev;
377 	index++;
378 
379 	/* configure queue flush support */
380 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
381 		blk_queue_flush(q, REQ_FLUSH);
382 
383 	/* If disk is read-only in the host, the guest should obey */
384 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
385 		set_disk_ro(vblk->disk, 1);
386 
387 	/* Host must always specify the capacity. */
388 	vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
389 			  &cap, sizeof(cap));
390 
391 	/* If capacity is too big, truncate with warning. */
392 	if ((sector_t)cap != cap) {
393 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
394 			 (unsigned long long)cap);
395 		cap = (sector_t)-1;
396 	}
397 	set_capacity(vblk->disk, cap);
398 
399 	/* We can handle whatever the host told us to handle. */
400 	blk_queue_max_segments(q, vblk->sg_elems-2);
401 
402 	/* No need to bounce any requests */
403 	blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
404 
405 	/* No real sector limit. */
406 	blk_queue_max_hw_sectors(q, -1U);
407 
408 	/* Host can optionally specify maximum segment size and number of
409 	 * segments. */
410 	err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
411 				offsetof(struct virtio_blk_config, size_max),
412 				&v);
413 	if (!err)
414 		blk_queue_max_segment_size(q, v);
415 	else
416 		blk_queue_max_segment_size(q, -1U);
417 
418 	/* Host can optionally specify the block size of the device */
419 	err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
420 				offsetof(struct virtio_blk_config, blk_size),
421 				&blk_size);
422 	if (!err)
423 		blk_queue_logical_block_size(q, blk_size);
424 	else
425 		blk_size = queue_logical_block_size(q);
426 
427 	/* Use topology information if available */
428 	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
429 			offsetof(struct virtio_blk_config, physical_block_exp),
430 			&physical_block_exp);
431 	if (!err && physical_block_exp)
432 		blk_queue_physical_block_size(q,
433 				blk_size * (1 << physical_block_exp));
434 
435 	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
436 			offsetof(struct virtio_blk_config, alignment_offset),
437 			&alignment_offset);
438 	if (!err && alignment_offset)
439 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
440 
441 	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
442 			offsetof(struct virtio_blk_config, min_io_size),
443 			&min_io_size);
444 	if (!err && min_io_size)
445 		blk_queue_io_min(q, blk_size * min_io_size);
446 
447 	err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
448 			offsetof(struct virtio_blk_config, opt_io_size),
449 			&opt_io_size);
450 	if (!err && opt_io_size)
451 		blk_queue_io_opt(q, blk_size * opt_io_size);
452 
453 
454 	add_disk(vblk->disk);
455 	err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
456 	if (err)
457 		goto out_del_disk;
458 
459 	return 0;
460 
461 out_del_disk:
462 	del_gendisk(vblk->disk);
463 	blk_cleanup_queue(vblk->disk->queue);
464 out_put_disk:
465 	put_disk(vblk->disk);
466 out_mempool:
467 	mempool_destroy(vblk->pool);
468 out_free_vq:
469 	vdev->config->del_vqs(vdev);
470 out_free_vblk:
471 	kfree(vblk);
472 out:
473 	return err;
474 }
475 
476 static void __devexit virtblk_remove(struct virtio_device *vdev)
477 {
478 	struct virtio_blk *vblk = vdev->priv;
479 
480 	/* Nothing should be pending. */
481 	BUG_ON(!list_empty(&vblk->reqs));
482 
483 	/* Stop all the virtqueues. */
484 	vdev->config->reset(vdev);
485 
486 	del_gendisk(vblk->disk);
487 	blk_cleanup_queue(vblk->disk->queue);
488 	put_disk(vblk->disk);
489 	mempool_destroy(vblk->pool);
490 	vdev->config->del_vqs(vdev);
491 	kfree(vblk);
492 }
493 
494 static const struct virtio_device_id id_table[] = {
495 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
496 	{ 0 },
497 };
498 
499 static unsigned int features[] = {
500 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
501 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
502 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
503 };
504 
505 /*
506  * virtio_blk causes spurious section mismatch warning by
507  * simultaneously referring to a __devinit and a __devexit function.
508  * Use __refdata to avoid this warning.
509  */
510 static struct virtio_driver __refdata virtio_blk = {
511 	.feature_table = features,
512 	.feature_table_size = ARRAY_SIZE(features),
513 	.driver.name =	KBUILD_MODNAME,
514 	.driver.owner =	THIS_MODULE,
515 	.id_table =	id_table,
516 	.probe =	virtblk_probe,
517 	.remove =	__devexit_p(virtblk_remove),
518 };
519 
520 static int __init init(void)
521 {
522 	major = register_blkdev(0, "virtblk");
523 	if (major < 0)
524 		return major;
525 	return register_virtio_driver(&virtio_blk);
526 }
527 
528 static void __exit fini(void)
529 {
530 	unregister_blkdev(major, "virtblk");
531 	unregister_virtio_driver(&virtio_blk);
532 }
533 module_init(init);
534 module_exit(fini);
535 
536 MODULE_DEVICE_TABLE(virtio, id_table);
537 MODULE_DESCRIPTION("Virtio block driver");
538 MODULE_LICENSE("GPL");
539