xref: /linux/drivers/block/virtio_blk.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
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/module.h>
7 #include <linux/mutex.h>
8 #include <linux/virtio.h>
9 #include <linux/virtio_blk.h>
10 #include <linux/scatterlist.h>
11 #include <linux/string_helpers.h>
12 #include <scsi/scsi_cmnd.h>
13 #include <linux/idr.h>
14 #include <linux/blk-mq.h>
15 #include <linux/numa.h>
16 
17 #define PART_BITS 4
18 #define VQ_NAME_LEN 16
19 
20 static int major;
21 static DEFINE_IDA(vd_index_ida);
22 
23 static struct workqueue_struct *virtblk_wq;
24 
25 struct virtio_blk_vq {
26 	struct virtqueue *vq;
27 	spinlock_t lock;
28 	char name[VQ_NAME_LEN];
29 } ____cacheline_aligned_in_smp;
30 
31 struct virtio_blk {
32 	struct virtio_device *vdev;
33 
34 	/* The disk structure for the kernel. */
35 	struct gendisk *disk;
36 
37 	/* Block layer tags. */
38 	struct blk_mq_tag_set tag_set;
39 
40 	/* Process context for config space updates */
41 	struct work_struct config_work;
42 
43 	/* What host tells us, plus 2 for header & tailer. */
44 	unsigned int sg_elems;
45 
46 	/* Ida index - used to track minor number allocations. */
47 	int index;
48 
49 	/* num of vqs */
50 	int num_vqs;
51 	struct virtio_blk_vq *vqs;
52 };
53 
54 struct virtblk_req {
55 	struct request *req;
56 	struct virtio_blk_outhdr out_hdr;
57 	struct virtio_scsi_inhdr in_hdr;
58 	u8 status;
59 	struct scatterlist sg[];
60 };
61 
62 static inline int virtblk_result(struct virtblk_req *vbr)
63 {
64 	switch (vbr->status) {
65 	case VIRTIO_BLK_S_OK:
66 		return 0;
67 	case VIRTIO_BLK_S_UNSUPP:
68 		return -ENOTTY;
69 	default:
70 		return -EIO;
71 	}
72 }
73 
74 static int __virtblk_add_req(struct virtqueue *vq,
75 			     struct virtblk_req *vbr,
76 			     struct scatterlist *data_sg,
77 			     bool have_data)
78 {
79 	struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
80 	unsigned int num_out = 0, num_in = 0;
81 	__virtio32 type = vbr->out_hdr.type & ~cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT);
82 
83 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
84 	sgs[num_out++] = &hdr;
85 
86 	/*
87 	 * If this is a packet command we need a couple of additional headers.
88 	 * Behind the normal outhdr we put a segment with the scsi command
89 	 * block, and before the normal inhdr we put the sense data and the
90 	 * inhdr with additional status information.
91 	 */
92 	if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) {
93 		sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len);
94 		sgs[num_out++] = &cmd;
95 	}
96 
97 	if (have_data) {
98 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
99 			sgs[num_out++] = data_sg;
100 		else
101 			sgs[num_out + num_in++] = data_sg;
102 	}
103 
104 	if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) {
105 		sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
106 		sgs[num_out + num_in++] = &sense;
107 		sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
108 		sgs[num_out + num_in++] = &inhdr;
109 	}
110 
111 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
112 	sgs[num_out + num_in++] = &status;
113 
114 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
115 }
116 
117 static inline void virtblk_request_done(struct request *req)
118 {
119 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
120 	struct virtio_blk *vblk = req->q->queuedata;
121 	int error = virtblk_result(vbr);
122 
123 	if (req->cmd_type == REQ_TYPE_BLOCK_PC) {
124 		req->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual);
125 		req->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len);
126 		req->errors = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors);
127 	} else if (req->cmd_type == REQ_TYPE_SPECIAL) {
128 		req->errors = (error != 0);
129 	}
130 
131 	blk_mq_end_request(req, error);
132 }
133 
134 static void virtblk_done(struct virtqueue *vq)
135 {
136 	struct virtio_blk *vblk = vq->vdev->priv;
137 	bool req_done = false;
138 	int qid = vq->index;
139 	struct virtblk_req *vbr;
140 	unsigned long flags;
141 	unsigned int len;
142 
143 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
144 	do {
145 		virtqueue_disable_cb(vq);
146 		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
147 			blk_mq_complete_request(vbr->req);
148 			req_done = true;
149 		}
150 		if (unlikely(virtqueue_is_broken(vq)))
151 			break;
152 	} while (!virtqueue_enable_cb(vq));
153 
154 	/* In case queue is stopped waiting for more buffers. */
155 	if (req_done)
156 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
157 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
158 }
159 
160 static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
161 			   const struct blk_mq_queue_data *bd)
162 {
163 	struct virtio_blk *vblk = hctx->queue->queuedata;
164 	struct request *req = bd->rq;
165 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
166 	unsigned long flags;
167 	unsigned int num;
168 	int qid = hctx->queue_num;
169 	int err;
170 	bool notify = false;
171 
172 	BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
173 
174 	vbr->req = req;
175 	if (req->cmd_flags & REQ_FLUSH) {
176 		vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_FLUSH);
177 		vbr->out_hdr.sector = 0;
178 		vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
179 	} else {
180 		switch (req->cmd_type) {
181 		case REQ_TYPE_FS:
182 			vbr->out_hdr.type = 0;
183 			vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, blk_rq_pos(vbr->req));
184 			vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
185 			break;
186 		case REQ_TYPE_BLOCK_PC:
187 			vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_SCSI_CMD);
188 			vbr->out_hdr.sector = 0;
189 			vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
190 			break;
191 		case REQ_TYPE_SPECIAL:
192 			vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
193 			vbr->out_hdr.sector = 0;
194 			vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
195 			break;
196 		default:
197 			/* We don't put anything else in the queue. */
198 			BUG();
199 		}
200 	}
201 
202 	blk_mq_start_request(req);
203 
204 	num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg);
205 	if (num) {
206 		if (rq_data_dir(vbr->req) == WRITE)
207 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
208 		else
209 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
210 	}
211 
212 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
213 	err = __virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
214 	if (err) {
215 		virtqueue_kick(vblk->vqs[qid].vq);
216 		blk_mq_stop_hw_queue(hctx);
217 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
218 		/* Out of mem doesn't actually happen, since we fall back
219 		 * to direct descriptors */
220 		if (err == -ENOMEM || err == -ENOSPC)
221 			return BLK_MQ_RQ_QUEUE_BUSY;
222 		return BLK_MQ_RQ_QUEUE_ERROR;
223 	}
224 
225 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
226 		notify = true;
227 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
228 
229 	if (notify)
230 		virtqueue_notify(vblk->vqs[qid].vq);
231 	return BLK_MQ_RQ_QUEUE_OK;
232 }
233 
234 /* return id (s/n) string for *disk to *id_str
235  */
236 static int virtblk_get_id(struct gendisk *disk, char *id_str)
237 {
238 	struct virtio_blk *vblk = disk->private_data;
239 	struct request *req;
240 	struct bio *bio;
241 	int err;
242 
243 	bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
244 			   GFP_KERNEL);
245 	if (IS_ERR(bio))
246 		return PTR_ERR(bio);
247 
248 	req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
249 	if (IS_ERR(req)) {
250 		bio_put(bio);
251 		return PTR_ERR(req);
252 	}
253 
254 	req->cmd_type = REQ_TYPE_SPECIAL;
255 	err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
256 	blk_put_request(req);
257 
258 	return err;
259 }
260 
261 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
262 			     unsigned int cmd, unsigned long data)
263 {
264 	struct gendisk *disk = bdev->bd_disk;
265 	struct virtio_blk *vblk = disk->private_data;
266 
267 	/*
268 	 * Only allow the generic SCSI ioctls if the host can support it.
269 	 */
270 	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
271 		return -ENOTTY;
272 
273 	return scsi_cmd_blk_ioctl(bdev, mode, cmd,
274 				  (void __user *)data);
275 }
276 
277 /* We provide getgeo only to please some old bootloader/partitioning tools */
278 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
279 {
280 	struct virtio_blk *vblk = bd->bd_disk->private_data;
281 
282 	/* see if the host passed in geometry config */
283 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
284 		virtio_cread(vblk->vdev, struct virtio_blk_config,
285 			     geometry.cylinders, &geo->cylinders);
286 		virtio_cread(vblk->vdev, struct virtio_blk_config,
287 			     geometry.heads, &geo->heads);
288 		virtio_cread(vblk->vdev, struct virtio_blk_config,
289 			     geometry.sectors, &geo->sectors);
290 	} else {
291 		/* some standard values, similar to sd */
292 		geo->heads = 1 << 6;
293 		geo->sectors = 1 << 5;
294 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
295 	}
296 	return 0;
297 }
298 
299 static const struct block_device_operations virtblk_fops = {
300 	.ioctl  = virtblk_ioctl,
301 	.owner  = THIS_MODULE,
302 	.getgeo = virtblk_getgeo,
303 };
304 
305 static int index_to_minor(int index)
306 {
307 	return index << PART_BITS;
308 }
309 
310 static int minor_to_index(int minor)
311 {
312 	return minor >> PART_BITS;
313 }
314 
315 static ssize_t virtblk_serial_show(struct device *dev,
316 				struct device_attribute *attr, char *buf)
317 {
318 	struct gendisk *disk = dev_to_disk(dev);
319 	int err;
320 
321 	/* sysfs gives us a PAGE_SIZE buffer */
322 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
323 
324 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
325 	err = virtblk_get_id(disk, buf);
326 	if (!err)
327 		return strlen(buf);
328 
329 	if (err == -EIO) /* Unsupported? Make it empty. */
330 		return 0;
331 
332 	return err;
333 }
334 
335 static DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
336 
337 static void virtblk_config_changed_work(struct work_struct *work)
338 {
339 	struct virtio_blk *vblk =
340 		container_of(work, struct virtio_blk, config_work);
341 	struct virtio_device *vdev = vblk->vdev;
342 	struct request_queue *q = vblk->disk->queue;
343 	char cap_str_2[10], cap_str_10[10];
344 	char *envp[] = { "RESIZE=1", NULL };
345 	u64 capacity, size;
346 
347 	/* Host must always specify the capacity. */
348 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
349 
350 	/* If capacity is too big, truncate with warning. */
351 	if ((sector_t)capacity != capacity) {
352 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
353 			 (unsigned long long)capacity);
354 		capacity = (sector_t)-1;
355 	}
356 
357 	size = capacity * queue_logical_block_size(q);
358 	string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
359 	string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
360 
361 	dev_notice(&vdev->dev,
362 		  "new size: %llu %d-byte logical blocks (%s/%s)\n",
363 		  (unsigned long long)capacity,
364 		  queue_logical_block_size(q),
365 		  cap_str_10, cap_str_2);
366 
367 	set_capacity(vblk->disk, capacity);
368 	revalidate_disk(vblk->disk);
369 	kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
370 }
371 
372 static void virtblk_config_changed(struct virtio_device *vdev)
373 {
374 	struct virtio_blk *vblk = vdev->priv;
375 
376 	queue_work(virtblk_wq, &vblk->config_work);
377 }
378 
379 static int init_vq(struct virtio_blk *vblk)
380 {
381 	int err = 0;
382 	int i;
383 	vq_callback_t **callbacks;
384 	const char **names;
385 	struct virtqueue **vqs;
386 	unsigned short num_vqs;
387 	struct virtio_device *vdev = vblk->vdev;
388 
389 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
390 				   struct virtio_blk_config, num_queues,
391 				   &num_vqs);
392 	if (err)
393 		num_vqs = 1;
394 
395 	vblk->vqs = kmalloc(sizeof(*vblk->vqs) * num_vqs, GFP_KERNEL);
396 	if (!vblk->vqs) {
397 		err = -ENOMEM;
398 		goto out;
399 	}
400 
401 	names = kmalloc(sizeof(*names) * num_vqs, GFP_KERNEL);
402 	if (!names)
403 		goto err_names;
404 
405 	callbacks = kmalloc(sizeof(*callbacks) * num_vqs, GFP_KERNEL);
406 	if (!callbacks)
407 		goto err_callbacks;
408 
409 	vqs = kmalloc(sizeof(*vqs) * num_vqs, GFP_KERNEL);
410 	if (!vqs)
411 		goto err_vqs;
412 
413 	for (i = 0; i < num_vqs; i++) {
414 		callbacks[i] = virtblk_done;
415 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
416 		names[i] = vblk->vqs[i].name;
417 	}
418 
419 	/* Discover virtqueues and write information to configuration.  */
420 	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names);
421 	if (err)
422 		goto err_find_vqs;
423 
424 	for (i = 0; i < num_vqs; i++) {
425 		spin_lock_init(&vblk->vqs[i].lock);
426 		vblk->vqs[i].vq = vqs[i];
427 	}
428 	vblk->num_vqs = num_vqs;
429 
430  err_find_vqs:
431 	kfree(vqs);
432  err_vqs:
433 	kfree(callbacks);
434  err_callbacks:
435 	kfree(names);
436  err_names:
437 	if (err)
438 		kfree(vblk->vqs);
439  out:
440 	return err;
441 }
442 
443 /*
444  * Legacy naming scheme used for virtio devices.  We are stuck with it for
445  * virtio blk but don't ever use it for any new driver.
446  */
447 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
448 {
449 	const int base = 'z' - 'a' + 1;
450 	char *begin = buf + strlen(prefix);
451 	char *end = buf + buflen;
452 	char *p;
453 	int unit;
454 
455 	p = end - 1;
456 	*p = '\0';
457 	unit = base;
458 	do {
459 		if (p == begin)
460 			return -EINVAL;
461 		*--p = 'a' + (index % unit);
462 		index = (index / unit) - 1;
463 	} while (index >= 0);
464 
465 	memmove(begin, p, end - p);
466 	memcpy(buf, prefix, strlen(prefix));
467 
468 	return 0;
469 }
470 
471 static int virtblk_get_cache_mode(struct virtio_device *vdev)
472 {
473 	u8 writeback;
474 	int err;
475 
476 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
477 				   struct virtio_blk_config, wce,
478 				   &writeback);
479 	if (err)
480 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE) ||
481 		            virtio_has_feature(vdev, VIRTIO_F_VERSION_1);
482 
483 	return writeback;
484 }
485 
486 static void virtblk_update_cache_mode(struct virtio_device *vdev)
487 {
488 	u8 writeback = virtblk_get_cache_mode(vdev);
489 	struct virtio_blk *vblk = vdev->priv;
490 
491 	if (writeback)
492 		blk_queue_flush(vblk->disk->queue, REQ_FLUSH);
493 	else
494 		blk_queue_flush(vblk->disk->queue, 0);
495 
496 	revalidate_disk(vblk->disk);
497 }
498 
499 static const char *const virtblk_cache_types[] = {
500 	"write through", "write back"
501 };
502 
503 static ssize_t
504 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
505 			 const char *buf, size_t count)
506 {
507 	struct gendisk *disk = dev_to_disk(dev);
508 	struct virtio_blk *vblk = disk->private_data;
509 	struct virtio_device *vdev = vblk->vdev;
510 	int i;
511 
512 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
513 	for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
514 		if (sysfs_streq(buf, virtblk_cache_types[i]))
515 			break;
516 
517 	if (i < 0)
518 		return -EINVAL;
519 
520 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
521 	virtblk_update_cache_mode(vdev);
522 	return count;
523 }
524 
525 static ssize_t
526 virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
527 			 char *buf)
528 {
529 	struct gendisk *disk = dev_to_disk(dev);
530 	struct virtio_blk *vblk = disk->private_data;
531 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
532 
533 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
534 	return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
535 }
536 
537 static const struct device_attribute dev_attr_cache_type_ro =
538 	__ATTR(cache_type, S_IRUGO,
539 	       virtblk_cache_type_show, NULL);
540 static const struct device_attribute dev_attr_cache_type_rw =
541 	__ATTR(cache_type, S_IRUGO|S_IWUSR,
542 	       virtblk_cache_type_show, virtblk_cache_type_store);
543 
544 static int virtblk_init_request(void *data, struct request *rq,
545 		unsigned int hctx_idx, unsigned int request_idx,
546 		unsigned int numa_node)
547 {
548 	struct virtio_blk *vblk = data;
549 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
550 
551 	sg_init_table(vbr->sg, vblk->sg_elems);
552 	return 0;
553 }
554 
555 static struct blk_mq_ops virtio_mq_ops = {
556 	.queue_rq	= virtio_queue_rq,
557 	.map_queue	= blk_mq_map_queue,
558 	.complete	= virtblk_request_done,
559 	.init_request	= virtblk_init_request,
560 };
561 
562 static unsigned int virtblk_queue_depth;
563 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
564 
565 static int virtblk_probe(struct virtio_device *vdev)
566 {
567 	struct virtio_blk *vblk;
568 	struct request_queue *q;
569 	int err, index;
570 
571 	u64 cap;
572 	u32 v, blk_size, sg_elems, opt_io_size;
573 	u16 min_io_size;
574 	u8 physical_block_exp, alignment_offset;
575 
576 	if (!vdev->config->get) {
577 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
578 			__func__);
579 		return -EINVAL;
580 	}
581 
582 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
583 			     GFP_KERNEL);
584 	if (err < 0)
585 		goto out;
586 	index = err;
587 
588 	/* We need to know how many segments before we allocate. */
589 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
590 				   struct virtio_blk_config, seg_max,
591 				   &sg_elems);
592 
593 	/* We need at least one SG element, whatever they say. */
594 	if (err || !sg_elems)
595 		sg_elems = 1;
596 
597 	/* We need an extra sg elements at head and tail. */
598 	sg_elems += 2;
599 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
600 	if (!vblk) {
601 		err = -ENOMEM;
602 		goto out_free_index;
603 	}
604 
605 	vblk->vdev = vdev;
606 	vblk->sg_elems = sg_elems;
607 
608 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
609 
610 	err = init_vq(vblk);
611 	if (err)
612 		goto out_free_vblk;
613 
614 	/* FIXME: How many partitions?  How long is a piece of string? */
615 	vblk->disk = alloc_disk(1 << PART_BITS);
616 	if (!vblk->disk) {
617 		err = -ENOMEM;
618 		goto out_free_vq;
619 	}
620 
621 	/* Default queue sizing is to fill the ring. */
622 	if (!virtblk_queue_depth) {
623 		virtblk_queue_depth = vblk->vqs[0].vq->num_free;
624 		/* ... but without indirect descs, we use 2 descs per req */
625 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
626 			virtblk_queue_depth /= 2;
627 	}
628 
629 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
630 	vblk->tag_set.ops = &virtio_mq_ops;
631 	vblk->tag_set.queue_depth = virtblk_queue_depth;
632 	vblk->tag_set.numa_node = NUMA_NO_NODE;
633 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
634 	vblk->tag_set.cmd_size =
635 		sizeof(struct virtblk_req) +
636 		sizeof(struct scatterlist) * sg_elems;
637 	vblk->tag_set.driver_data = vblk;
638 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
639 
640 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
641 	if (err)
642 		goto out_put_disk;
643 
644 	q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set);
645 	if (IS_ERR(q)) {
646 		err = -ENOMEM;
647 		goto out_free_tags;
648 	}
649 
650 	q->queuedata = vblk;
651 
652 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
653 
654 	vblk->disk->major = major;
655 	vblk->disk->first_minor = index_to_minor(index);
656 	vblk->disk->private_data = vblk;
657 	vblk->disk->fops = &virtblk_fops;
658 	vblk->disk->driverfs_dev = &vdev->dev;
659 	vblk->index = index;
660 
661 	/* configure queue flush support */
662 	virtblk_update_cache_mode(vdev);
663 
664 	/* If disk is read-only in the host, the guest should obey */
665 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
666 		set_disk_ro(vblk->disk, 1);
667 
668 	/* Host must always specify the capacity. */
669 	virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
670 
671 	/* If capacity is too big, truncate with warning. */
672 	if ((sector_t)cap != cap) {
673 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
674 			 (unsigned long long)cap);
675 		cap = (sector_t)-1;
676 	}
677 	set_capacity(vblk->disk, cap);
678 
679 	/* We can handle whatever the host told us to handle. */
680 	blk_queue_max_segments(q, vblk->sg_elems-2);
681 
682 	/* No need to bounce any requests */
683 	blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
684 
685 	/* No real sector limit. */
686 	blk_queue_max_hw_sectors(q, -1U);
687 
688 	/* Host can optionally specify maximum segment size and number of
689 	 * segments. */
690 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
691 				   struct virtio_blk_config, size_max, &v);
692 	if (!err)
693 		blk_queue_max_segment_size(q, v);
694 	else
695 		blk_queue_max_segment_size(q, -1U);
696 
697 	/* Host can optionally specify the block size of the device */
698 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
699 				   struct virtio_blk_config, blk_size,
700 				   &blk_size);
701 	if (!err)
702 		blk_queue_logical_block_size(q, blk_size);
703 	else
704 		blk_size = queue_logical_block_size(q);
705 
706 	/* Use topology information if available */
707 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
708 				   struct virtio_blk_config, physical_block_exp,
709 				   &physical_block_exp);
710 	if (!err && physical_block_exp)
711 		blk_queue_physical_block_size(q,
712 				blk_size * (1 << physical_block_exp));
713 
714 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
715 				   struct virtio_blk_config, alignment_offset,
716 				   &alignment_offset);
717 	if (!err && alignment_offset)
718 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
719 
720 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
721 				   struct virtio_blk_config, min_io_size,
722 				   &min_io_size);
723 	if (!err && min_io_size)
724 		blk_queue_io_min(q, blk_size * min_io_size);
725 
726 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
727 				   struct virtio_blk_config, opt_io_size,
728 				   &opt_io_size);
729 	if (!err && opt_io_size)
730 		blk_queue_io_opt(q, blk_size * opt_io_size);
731 
732 	virtio_device_ready(vdev);
733 
734 	add_disk(vblk->disk);
735 	err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
736 	if (err)
737 		goto out_del_disk;
738 
739 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
740 		err = device_create_file(disk_to_dev(vblk->disk),
741 					 &dev_attr_cache_type_rw);
742 	else
743 		err = device_create_file(disk_to_dev(vblk->disk),
744 					 &dev_attr_cache_type_ro);
745 	if (err)
746 		goto out_del_disk;
747 	return 0;
748 
749 out_del_disk:
750 	del_gendisk(vblk->disk);
751 	blk_cleanup_queue(vblk->disk->queue);
752 out_free_tags:
753 	blk_mq_free_tag_set(&vblk->tag_set);
754 out_put_disk:
755 	put_disk(vblk->disk);
756 out_free_vq:
757 	vdev->config->del_vqs(vdev);
758 out_free_vblk:
759 	kfree(vblk);
760 out_free_index:
761 	ida_simple_remove(&vd_index_ida, index);
762 out:
763 	return err;
764 }
765 
766 static void virtblk_remove(struct virtio_device *vdev)
767 {
768 	struct virtio_blk *vblk = vdev->priv;
769 	int index = vblk->index;
770 	int refc;
771 
772 	/* Make sure no work handler is accessing the device. */
773 	flush_work(&vblk->config_work);
774 
775 	del_gendisk(vblk->disk);
776 	blk_cleanup_queue(vblk->disk->queue);
777 
778 	blk_mq_free_tag_set(&vblk->tag_set);
779 
780 	/* Stop all the virtqueues. */
781 	vdev->config->reset(vdev);
782 
783 	refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount);
784 	put_disk(vblk->disk);
785 	vdev->config->del_vqs(vdev);
786 	kfree(vblk->vqs);
787 	kfree(vblk);
788 
789 	/* Only free device id if we don't have any users */
790 	if (refc == 1)
791 		ida_simple_remove(&vd_index_ida, index);
792 }
793 
794 #ifdef CONFIG_PM_SLEEP
795 static int virtblk_freeze(struct virtio_device *vdev)
796 {
797 	struct virtio_blk *vblk = vdev->priv;
798 
799 	/* Ensure we don't receive any more interrupts */
800 	vdev->config->reset(vdev);
801 
802 	/* Make sure no work handler is accessing the device. */
803 	flush_work(&vblk->config_work);
804 
805 	blk_mq_stop_hw_queues(vblk->disk->queue);
806 
807 	vdev->config->del_vqs(vdev);
808 	return 0;
809 }
810 
811 static int virtblk_restore(struct virtio_device *vdev)
812 {
813 	struct virtio_blk *vblk = vdev->priv;
814 	int ret;
815 
816 	ret = init_vq(vdev->priv);
817 	if (ret)
818 		return ret;
819 
820 	virtio_device_ready(vdev);
821 
822 	blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
823 	return 0;
824 }
825 #endif
826 
827 static const struct virtio_device_id id_table[] = {
828 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
829 	{ 0 },
830 };
831 
832 static unsigned int features_legacy[] = {
833 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
834 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
835 	VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
836 	VIRTIO_BLK_F_MQ,
837 }
838 ;
839 static unsigned int features[] = {
840 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
841 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
842 	VIRTIO_BLK_F_TOPOLOGY,
843 	VIRTIO_BLK_F_MQ,
844 };
845 
846 static struct virtio_driver virtio_blk = {
847 	.feature_table			= features,
848 	.feature_table_size		= ARRAY_SIZE(features),
849 	.feature_table_legacy		= features_legacy,
850 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
851 	.driver.name			= KBUILD_MODNAME,
852 	.driver.owner			= THIS_MODULE,
853 	.id_table			= id_table,
854 	.probe				= virtblk_probe,
855 	.remove				= virtblk_remove,
856 	.config_changed			= virtblk_config_changed,
857 #ifdef CONFIG_PM_SLEEP
858 	.freeze				= virtblk_freeze,
859 	.restore			= virtblk_restore,
860 #endif
861 };
862 
863 static int __init init(void)
864 {
865 	int error;
866 
867 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
868 	if (!virtblk_wq)
869 		return -ENOMEM;
870 
871 	major = register_blkdev(0, "virtblk");
872 	if (major < 0) {
873 		error = major;
874 		goto out_destroy_workqueue;
875 	}
876 
877 	error = register_virtio_driver(&virtio_blk);
878 	if (error)
879 		goto out_unregister_blkdev;
880 	return 0;
881 
882 out_unregister_blkdev:
883 	unregister_blkdev(major, "virtblk");
884 out_destroy_workqueue:
885 	destroy_workqueue(virtblk_wq);
886 	return error;
887 }
888 
889 static void __exit fini(void)
890 {
891 	unregister_virtio_driver(&virtio_blk);
892 	unregister_blkdev(major, "virtblk");
893 	destroy_workqueue(virtblk_wq);
894 }
895 module_init(init);
896 module_exit(fini);
897 
898 MODULE_DEVICE_TABLE(virtio, id_table);
899 MODULE_DESCRIPTION("Virtio block driver");
900 MODULE_LICENSE("GPL");
901