xref: /linux/drivers/block/virtio_blk.c (revision 37744feebc086908fd89760650f458ab19071750)
1 // SPDX-License-Identifier: GPL-2.0-only
2 //#define DEBUG
3 #include <linux/spinlock.h>
4 #include <linux/slab.h>
5 #include <linux/blkdev.h>
6 #include <linux/hdreg.h>
7 #include <linux/module.h>
8 #include <linux/mutex.h>
9 #include <linux/interrupt.h>
10 #include <linux/virtio.h>
11 #include <linux/virtio_blk.h>
12 #include <linux/scatterlist.h>
13 #include <linux/string_helpers.h>
14 #include <linux/idr.h>
15 #include <linux/blk-mq.h>
16 #include <linux/blk-mq-virtio.h>
17 #include <linux/numa.h>
18 #include <uapi/linux/virtio_ring.h>
19 
20 #define PART_BITS 4
21 #define VQ_NAME_LEN 16
22 #define MAX_DISCARD_SEGMENTS 256u
23 
24 static int major;
25 static DEFINE_IDA(vd_index_ida);
26 
27 static struct workqueue_struct *virtblk_wq;
28 
29 struct virtio_blk_vq {
30 	struct virtqueue *vq;
31 	spinlock_t lock;
32 	char name[VQ_NAME_LEN];
33 } ____cacheline_aligned_in_smp;
34 
35 struct virtio_blk {
36 	/*
37 	 * This mutex must be held by anything that may run after
38 	 * virtblk_remove() sets vblk->vdev to NULL.
39 	 *
40 	 * blk-mq, virtqueue processing, and sysfs attribute code paths are
41 	 * shut down before vblk->vdev is set to NULL and therefore do not need
42 	 * to hold this mutex.
43 	 */
44 	struct mutex vdev_mutex;
45 	struct virtio_device *vdev;
46 
47 	/* The disk structure for the kernel. */
48 	struct gendisk *disk;
49 
50 	/* Block layer tags. */
51 	struct blk_mq_tag_set tag_set;
52 
53 	/* Process context for config space updates */
54 	struct work_struct config_work;
55 
56 	/*
57 	 * Tracks references from block_device_operations open/release and
58 	 * virtio_driver probe/remove so this object can be freed once no
59 	 * longer in use.
60 	 */
61 	refcount_t refs;
62 
63 	/* What host tells us, plus 2 for header & tailer. */
64 	unsigned int sg_elems;
65 
66 	/* Ida index - used to track minor number allocations. */
67 	int index;
68 
69 	/* num of vqs */
70 	int num_vqs;
71 	struct virtio_blk_vq *vqs;
72 };
73 
74 struct virtblk_req {
75 	struct virtio_blk_outhdr out_hdr;
76 	u8 status;
77 	struct scatterlist sg[];
78 };
79 
80 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
81 {
82 	switch (vbr->status) {
83 	case VIRTIO_BLK_S_OK:
84 		return BLK_STS_OK;
85 	case VIRTIO_BLK_S_UNSUPP:
86 		return BLK_STS_NOTSUPP;
87 	default:
88 		return BLK_STS_IOERR;
89 	}
90 }
91 
92 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
93 		struct scatterlist *data_sg, bool have_data)
94 {
95 	struct scatterlist hdr, status, *sgs[3];
96 	unsigned int num_out = 0, num_in = 0;
97 
98 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
99 	sgs[num_out++] = &hdr;
100 
101 	if (have_data) {
102 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
103 			sgs[num_out++] = data_sg;
104 		else
105 			sgs[num_out + num_in++] = data_sg;
106 	}
107 
108 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
109 	sgs[num_out + num_in++] = &status;
110 
111 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
112 }
113 
114 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
115 {
116 	unsigned short segments = blk_rq_nr_discard_segments(req);
117 	unsigned short n = 0;
118 	struct virtio_blk_discard_write_zeroes *range;
119 	struct bio *bio;
120 	u32 flags = 0;
121 
122 	if (unmap)
123 		flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
124 
125 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
126 	if (!range)
127 		return -ENOMEM;
128 
129 	__rq_for_each_bio(bio, req) {
130 		u64 sector = bio->bi_iter.bi_sector;
131 		u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
132 
133 		range[n].flags = cpu_to_le32(flags);
134 		range[n].num_sectors = cpu_to_le32(num_sectors);
135 		range[n].sector = cpu_to_le64(sector);
136 		n++;
137 	}
138 
139 	req->special_vec.bv_page = virt_to_page(range);
140 	req->special_vec.bv_offset = offset_in_page(range);
141 	req->special_vec.bv_len = sizeof(*range) * segments;
142 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
143 
144 	return 0;
145 }
146 
147 static inline void virtblk_request_done(struct request *req)
148 {
149 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
150 
151 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
152 		kfree(page_address(req->special_vec.bv_page) +
153 		      req->special_vec.bv_offset);
154 	}
155 
156 	blk_mq_end_request(req, virtblk_result(vbr));
157 }
158 
159 static void virtblk_done(struct virtqueue *vq)
160 {
161 	struct virtio_blk *vblk = vq->vdev->priv;
162 	bool req_done = false;
163 	int qid = vq->index;
164 	struct virtblk_req *vbr;
165 	unsigned long flags;
166 	unsigned int len;
167 
168 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
169 	do {
170 		virtqueue_disable_cb(vq);
171 		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
172 			struct request *req = blk_mq_rq_from_pdu(vbr);
173 
174 			blk_mq_complete_request(req);
175 			req_done = true;
176 		}
177 		if (unlikely(virtqueue_is_broken(vq)))
178 			break;
179 	} while (!virtqueue_enable_cb(vq));
180 
181 	/* In case queue is stopped waiting for more buffers. */
182 	if (req_done)
183 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
184 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
185 }
186 
187 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
188 {
189 	struct virtio_blk *vblk = hctx->queue->queuedata;
190 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
191 	bool kick;
192 
193 	spin_lock_irq(&vq->lock);
194 	kick = virtqueue_kick_prepare(vq->vq);
195 	spin_unlock_irq(&vq->lock);
196 
197 	if (kick)
198 		virtqueue_notify(vq->vq);
199 }
200 
201 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
202 			   const struct blk_mq_queue_data *bd)
203 {
204 	struct virtio_blk *vblk = hctx->queue->queuedata;
205 	struct request *req = bd->rq;
206 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
207 	unsigned long flags;
208 	unsigned int num;
209 	int qid = hctx->queue_num;
210 	int err;
211 	bool notify = false;
212 	bool unmap = false;
213 	u32 type;
214 
215 	BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
216 
217 	switch (req_op(req)) {
218 	case REQ_OP_READ:
219 	case REQ_OP_WRITE:
220 		type = 0;
221 		break;
222 	case REQ_OP_FLUSH:
223 		type = VIRTIO_BLK_T_FLUSH;
224 		break;
225 	case REQ_OP_DISCARD:
226 		type = VIRTIO_BLK_T_DISCARD;
227 		break;
228 	case REQ_OP_WRITE_ZEROES:
229 		type = VIRTIO_BLK_T_WRITE_ZEROES;
230 		unmap = !(req->cmd_flags & REQ_NOUNMAP);
231 		break;
232 	case REQ_OP_DRV_IN:
233 		type = VIRTIO_BLK_T_GET_ID;
234 		break;
235 	default:
236 		WARN_ON_ONCE(1);
237 		return BLK_STS_IOERR;
238 	}
239 
240 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
241 	vbr->out_hdr.sector = type ?
242 		0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
243 	vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
244 
245 	blk_mq_start_request(req);
246 
247 	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
248 		err = virtblk_setup_discard_write_zeroes(req, unmap);
249 		if (err)
250 			return BLK_STS_RESOURCE;
251 	}
252 
253 	num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
254 	if (num) {
255 		if (rq_data_dir(req) == WRITE)
256 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
257 		else
258 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
259 	}
260 
261 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
262 	err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
263 	if (err) {
264 		virtqueue_kick(vblk->vqs[qid].vq);
265 		/* Don't stop the queue if -ENOMEM: we may have failed to
266 		 * bounce the buffer due to global resource outage.
267 		 */
268 		if (err == -ENOSPC)
269 			blk_mq_stop_hw_queue(hctx);
270 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
271 		switch (err) {
272 		case -ENOSPC:
273 			return BLK_STS_DEV_RESOURCE;
274 		case -ENOMEM:
275 			return BLK_STS_RESOURCE;
276 		default:
277 			return BLK_STS_IOERR;
278 		}
279 	}
280 
281 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
282 		notify = true;
283 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
284 
285 	if (notify)
286 		virtqueue_notify(vblk->vqs[qid].vq);
287 	return BLK_STS_OK;
288 }
289 
290 /* return id (s/n) string for *disk to *id_str
291  */
292 static int virtblk_get_id(struct gendisk *disk, char *id_str)
293 {
294 	struct virtio_blk *vblk = disk->private_data;
295 	struct request_queue *q = vblk->disk->queue;
296 	struct request *req;
297 	int err;
298 
299 	req = blk_get_request(q, REQ_OP_DRV_IN, 0);
300 	if (IS_ERR(req))
301 		return PTR_ERR(req);
302 
303 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
304 	if (err)
305 		goto out;
306 
307 	blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
308 	err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
309 out:
310 	blk_put_request(req);
311 	return err;
312 }
313 
314 static void virtblk_get(struct virtio_blk *vblk)
315 {
316 	refcount_inc(&vblk->refs);
317 }
318 
319 static void virtblk_put(struct virtio_blk *vblk)
320 {
321 	if (refcount_dec_and_test(&vblk->refs)) {
322 		ida_simple_remove(&vd_index_ida, vblk->index);
323 		mutex_destroy(&vblk->vdev_mutex);
324 		kfree(vblk);
325 	}
326 }
327 
328 static int virtblk_open(struct block_device *bd, fmode_t mode)
329 {
330 	struct virtio_blk *vblk = bd->bd_disk->private_data;
331 	int ret = 0;
332 
333 	mutex_lock(&vblk->vdev_mutex);
334 
335 	if (vblk->vdev)
336 		virtblk_get(vblk);
337 	else
338 		ret = -ENXIO;
339 
340 	mutex_unlock(&vblk->vdev_mutex);
341 	return ret;
342 }
343 
344 static void virtblk_release(struct gendisk *disk, fmode_t mode)
345 {
346 	struct virtio_blk *vblk = disk->private_data;
347 
348 	virtblk_put(vblk);
349 }
350 
351 /* We provide getgeo only to please some old bootloader/partitioning tools */
352 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
353 {
354 	struct virtio_blk *vblk = bd->bd_disk->private_data;
355 	int ret = 0;
356 
357 	mutex_lock(&vblk->vdev_mutex);
358 
359 	if (!vblk->vdev) {
360 		ret = -ENXIO;
361 		goto out;
362 	}
363 
364 	/* see if the host passed in geometry config */
365 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
366 		virtio_cread(vblk->vdev, struct virtio_blk_config,
367 			     geometry.cylinders, &geo->cylinders);
368 		virtio_cread(vblk->vdev, struct virtio_blk_config,
369 			     geometry.heads, &geo->heads);
370 		virtio_cread(vblk->vdev, struct virtio_blk_config,
371 			     geometry.sectors, &geo->sectors);
372 	} else {
373 		/* some standard values, similar to sd */
374 		geo->heads = 1 << 6;
375 		geo->sectors = 1 << 5;
376 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
377 	}
378 out:
379 	mutex_unlock(&vblk->vdev_mutex);
380 	return ret;
381 }
382 
383 static const struct block_device_operations virtblk_fops = {
384 	.owner  = THIS_MODULE,
385 	.open = virtblk_open,
386 	.release = virtblk_release,
387 	.getgeo = virtblk_getgeo,
388 };
389 
390 static int index_to_minor(int index)
391 {
392 	return index << PART_BITS;
393 }
394 
395 static int minor_to_index(int minor)
396 {
397 	return minor >> PART_BITS;
398 }
399 
400 static ssize_t serial_show(struct device *dev,
401 			   struct device_attribute *attr, char *buf)
402 {
403 	struct gendisk *disk = dev_to_disk(dev);
404 	int err;
405 
406 	/* sysfs gives us a PAGE_SIZE buffer */
407 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
408 
409 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
410 	err = virtblk_get_id(disk, buf);
411 	if (!err)
412 		return strlen(buf);
413 
414 	if (err == -EIO) /* Unsupported? Make it empty. */
415 		return 0;
416 
417 	return err;
418 }
419 
420 static DEVICE_ATTR_RO(serial);
421 
422 /* The queue's logical block size must be set before calling this */
423 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
424 {
425 	struct virtio_device *vdev = vblk->vdev;
426 	struct request_queue *q = vblk->disk->queue;
427 	char cap_str_2[10], cap_str_10[10];
428 	unsigned long long nblocks;
429 	u64 capacity;
430 
431 	/* Host must always specify the capacity. */
432 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
433 
434 	/* If capacity is too big, truncate with warning. */
435 	if ((sector_t)capacity != capacity) {
436 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
437 			 (unsigned long long)capacity);
438 		capacity = (sector_t)-1;
439 	}
440 
441 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
442 
443 	string_get_size(nblocks, queue_logical_block_size(q),
444 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
445 	string_get_size(nblocks, queue_logical_block_size(q),
446 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
447 
448 	dev_notice(&vdev->dev,
449 		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
450 		   vblk->disk->disk_name,
451 		   resize ? "new size: " : "",
452 		   nblocks,
453 		   queue_logical_block_size(q),
454 		   cap_str_10,
455 		   cap_str_2);
456 
457 	set_capacity_revalidate_and_notify(vblk->disk, capacity, true);
458 }
459 
460 static void virtblk_config_changed_work(struct work_struct *work)
461 {
462 	struct virtio_blk *vblk =
463 		container_of(work, struct virtio_blk, config_work);
464 
465 	virtblk_update_capacity(vblk, true);
466 }
467 
468 static void virtblk_config_changed(struct virtio_device *vdev)
469 {
470 	struct virtio_blk *vblk = vdev->priv;
471 
472 	queue_work(virtblk_wq, &vblk->config_work);
473 }
474 
475 static int init_vq(struct virtio_blk *vblk)
476 {
477 	int err;
478 	int i;
479 	vq_callback_t **callbacks;
480 	const char **names;
481 	struct virtqueue **vqs;
482 	unsigned short num_vqs;
483 	struct virtio_device *vdev = vblk->vdev;
484 	struct irq_affinity desc = { 0, };
485 
486 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
487 				   struct virtio_blk_config, num_queues,
488 				   &num_vqs);
489 	if (err)
490 		num_vqs = 1;
491 
492 	num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
493 
494 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
495 	if (!vblk->vqs)
496 		return -ENOMEM;
497 
498 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
499 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
500 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
501 	if (!names || !callbacks || !vqs) {
502 		err = -ENOMEM;
503 		goto out;
504 	}
505 
506 	for (i = 0; i < num_vqs; i++) {
507 		callbacks[i] = virtblk_done;
508 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
509 		names[i] = vblk->vqs[i].name;
510 	}
511 
512 	/* Discover virtqueues and write information to configuration.  */
513 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
514 	if (err)
515 		goto out;
516 
517 	for (i = 0; i < num_vqs; i++) {
518 		spin_lock_init(&vblk->vqs[i].lock);
519 		vblk->vqs[i].vq = vqs[i];
520 	}
521 	vblk->num_vqs = num_vqs;
522 
523 out:
524 	kfree(vqs);
525 	kfree(callbacks);
526 	kfree(names);
527 	if (err)
528 		kfree(vblk->vqs);
529 	return err;
530 }
531 
532 /*
533  * Legacy naming scheme used for virtio devices.  We are stuck with it for
534  * virtio blk but don't ever use it for any new driver.
535  */
536 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
537 {
538 	const int base = 'z' - 'a' + 1;
539 	char *begin = buf + strlen(prefix);
540 	char *end = buf + buflen;
541 	char *p;
542 	int unit;
543 
544 	p = end - 1;
545 	*p = '\0';
546 	unit = base;
547 	do {
548 		if (p == begin)
549 			return -EINVAL;
550 		*--p = 'a' + (index % unit);
551 		index = (index / unit) - 1;
552 	} while (index >= 0);
553 
554 	memmove(begin, p, end - p);
555 	memcpy(buf, prefix, strlen(prefix));
556 
557 	return 0;
558 }
559 
560 static int virtblk_get_cache_mode(struct virtio_device *vdev)
561 {
562 	u8 writeback;
563 	int err;
564 
565 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
566 				   struct virtio_blk_config, wce,
567 				   &writeback);
568 
569 	/*
570 	 * If WCE is not configurable and flush is not available,
571 	 * assume no writeback cache is in use.
572 	 */
573 	if (err)
574 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
575 
576 	return writeback;
577 }
578 
579 static void virtblk_update_cache_mode(struct virtio_device *vdev)
580 {
581 	u8 writeback = virtblk_get_cache_mode(vdev);
582 	struct virtio_blk *vblk = vdev->priv;
583 
584 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
585 	revalidate_disk(vblk->disk);
586 }
587 
588 static const char *const virtblk_cache_types[] = {
589 	"write through", "write back"
590 };
591 
592 static ssize_t
593 cache_type_store(struct device *dev, struct device_attribute *attr,
594 		 const char *buf, size_t count)
595 {
596 	struct gendisk *disk = dev_to_disk(dev);
597 	struct virtio_blk *vblk = disk->private_data;
598 	struct virtio_device *vdev = vblk->vdev;
599 	int i;
600 
601 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
602 	i = sysfs_match_string(virtblk_cache_types, buf);
603 	if (i < 0)
604 		return i;
605 
606 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
607 	virtblk_update_cache_mode(vdev);
608 	return count;
609 }
610 
611 static ssize_t
612 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
613 {
614 	struct gendisk *disk = dev_to_disk(dev);
615 	struct virtio_blk *vblk = disk->private_data;
616 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
617 
618 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
619 	return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
620 }
621 
622 static DEVICE_ATTR_RW(cache_type);
623 
624 static struct attribute *virtblk_attrs[] = {
625 	&dev_attr_serial.attr,
626 	&dev_attr_cache_type.attr,
627 	NULL,
628 };
629 
630 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
631 		struct attribute *a, int n)
632 {
633 	struct device *dev = container_of(kobj, struct device, kobj);
634 	struct gendisk *disk = dev_to_disk(dev);
635 	struct virtio_blk *vblk = disk->private_data;
636 	struct virtio_device *vdev = vblk->vdev;
637 
638 	if (a == &dev_attr_cache_type.attr &&
639 	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
640 		return S_IRUGO;
641 
642 	return a->mode;
643 }
644 
645 static const struct attribute_group virtblk_attr_group = {
646 	.attrs = virtblk_attrs,
647 	.is_visible = virtblk_attrs_are_visible,
648 };
649 
650 static const struct attribute_group *virtblk_attr_groups[] = {
651 	&virtblk_attr_group,
652 	NULL,
653 };
654 
655 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
656 		unsigned int hctx_idx, unsigned int numa_node)
657 {
658 	struct virtio_blk *vblk = set->driver_data;
659 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
660 
661 	sg_init_table(vbr->sg, vblk->sg_elems);
662 	return 0;
663 }
664 
665 static int virtblk_map_queues(struct blk_mq_tag_set *set)
666 {
667 	struct virtio_blk *vblk = set->driver_data;
668 
669 	return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
670 					vblk->vdev, 0);
671 }
672 
673 static const struct blk_mq_ops virtio_mq_ops = {
674 	.queue_rq	= virtio_queue_rq,
675 	.commit_rqs	= virtio_commit_rqs,
676 	.complete	= virtblk_request_done,
677 	.init_request	= virtblk_init_request,
678 	.map_queues	= virtblk_map_queues,
679 };
680 
681 static unsigned int virtblk_queue_depth;
682 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
683 
684 static int virtblk_probe(struct virtio_device *vdev)
685 {
686 	struct virtio_blk *vblk;
687 	struct request_queue *q;
688 	int err, index;
689 
690 	u32 v, blk_size, max_size, sg_elems, opt_io_size;
691 	u16 min_io_size;
692 	u8 physical_block_exp, alignment_offset;
693 
694 	if (!vdev->config->get) {
695 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
696 			__func__);
697 		return -EINVAL;
698 	}
699 
700 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
701 			     GFP_KERNEL);
702 	if (err < 0)
703 		goto out;
704 	index = err;
705 
706 	/* We need to know how many segments before we allocate. */
707 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
708 				   struct virtio_blk_config, seg_max,
709 				   &sg_elems);
710 
711 	/* We need at least one SG element, whatever they say. */
712 	if (err || !sg_elems)
713 		sg_elems = 1;
714 
715 	/* We need an extra sg elements at head and tail. */
716 	sg_elems += 2;
717 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
718 	if (!vblk) {
719 		err = -ENOMEM;
720 		goto out_free_index;
721 	}
722 
723 	/* This reference is dropped in virtblk_remove(). */
724 	refcount_set(&vblk->refs, 1);
725 	mutex_init(&vblk->vdev_mutex);
726 
727 	vblk->vdev = vdev;
728 	vblk->sg_elems = sg_elems;
729 
730 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
731 
732 	err = init_vq(vblk);
733 	if (err)
734 		goto out_free_vblk;
735 
736 	/* FIXME: How many partitions?  How long is a piece of string? */
737 	vblk->disk = alloc_disk(1 << PART_BITS);
738 	if (!vblk->disk) {
739 		err = -ENOMEM;
740 		goto out_free_vq;
741 	}
742 
743 	/* Default queue sizing is to fill the ring. */
744 	if (!virtblk_queue_depth) {
745 		virtblk_queue_depth = vblk->vqs[0].vq->num_free;
746 		/* ... but without indirect descs, we use 2 descs per req */
747 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
748 			virtblk_queue_depth /= 2;
749 	}
750 
751 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
752 	vblk->tag_set.ops = &virtio_mq_ops;
753 	vblk->tag_set.queue_depth = virtblk_queue_depth;
754 	vblk->tag_set.numa_node = NUMA_NO_NODE;
755 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
756 	vblk->tag_set.cmd_size =
757 		sizeof(struct virtblk_req) +
758 		sizeof(struct scatterlist) * sg_elems;
759 	vblk->tag_set.driver_data = vblk;
760 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
761 
762 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
763 	if (err)
764 		goto out_put_disk;
765 
766 	q = blk_mq_init_queue(&vblk->tag_set);
767 	if (IS_ERR(q)) {
768 		err = -ENOMEM;
769 		goto out_free_tags;
770 	}
771 	vblk->disk->queue = q;
772 
773 	q->queuedata = vblk;
774 
775 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
776 
777 	vblk->disk->major = major;
778 	vblk->disk->first_minor = index_to_minor(index);
779 	vblk->disk->private_data = vblk;
780 	vblk->disk->fops = &virtblk_fops;
781 	vblk->disk->flags |= GENHD_FL_EXT_DEVT;
782 	vblk->index = index;
783 
784 	/* configure queue flush support */
785 	virtblk_update_cache_mode(vdev);
786 
787 	/* If disk is read-only in the host, the guest should obey */
788 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
789 		set_disk_ro(vblk->disk, 1);
790 
791 	/* We can handle whatever the host told us to handle. */
792 	blk_queue_max_segments(q, vblk->sg_elems-2);
793 
794 	/* No real sector limit. */
795 	blk_queue_max_hw_sectors(q, -1U);
796 
797 	max_size = virtio_max_dma_size(vdev);
798 
799 	/* Host can optionally specify maximum segment size and number of
800 	 * segments. */
801 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
802 				   struct virtio_blk_config, size_max, &v);
803 	if (!err)
804 		max_size = min(max_size, v);
805 
806 	blk_queue_max_segment_size(q, max_size);
807 
808 	/* Host can optionally specify the block size of the device */
809 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
810 				   struct virtio_blk_config, blk_size,
811 				   &blk_size);
812 	if (!err)
813 		blk_queue_logical_block_size(q, blk_size);
814 	else
815 		blk_size = queue_logical_block_size(q);
816 
817 	/* Use topology information if available */
818 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
819 				   struct virtio_blk_config, physical_block_exp,
820 				   &physical_block_exp);
821 	if (!err && physical_block_exp)
822 		blk_queue_physical_block_size(q,
823 				blk_size * (1 << physical_block_exp));
824 
825 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
826 				   struct virtio_blk_config, alignment_offset,
827 				   &alignment_offset);
828 	if (!err && alignment_offset)
829 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
830 
831 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
832 				   struct virtio_blk_config, min_io_size,
833 				   &min_io_size);
834 	if (!err && min_io_size)
835 		blk_queue_io_min(q, blk_size * min_io_size);
836 
837 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
838 				   struct virtio_blk_config, opt_io_size,
839 				   &opt_io_size);
840 	if (!err && opt_io_size)
841 		blk_queue_io_opt(q, blk_size * opt_io_size);
842 
843 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
844 		q->limits.discard_granularity = blk_size;
845 
846 		virtio_cread(vdev, struct virtio_blk_config,
847 			     discard_sector_alignment, &v);
848 		q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
849 
850 		virtio_cread(vdev, struct virtio_blk_config,
851 			     max_discard_sectors, &v);
852 		blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
853 
854 		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
855 			     &v);
856 		blk_queue_max_discard_segments(q,
857 					       min_not_zero(v,
858 							    MAX_DISCARD_SEGMENTS));
859 
860 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
861 	}
862 
863 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
864 		virtio_cread(vdev, struct virtio_blk_config,
865 			     max_write_zeroes_sectors, &v);
866 		blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
867 	}
868 
869 	virtblk_update_capacity(vblk, false);
870 	virtio_device_ready(vdev);
871 
872 	device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
873 	return 0;
874 
875 out_free_tags:
876 	blk_mq_free_tag_set(&vblk->tag_set);
877 out_put_disk:
878 	put_disk(vblk->disk);
879 out_free_vq:
880 	vdev->config->del_vqs(vdev);
881 out_free_vblk:
882 	kfree(vblk);
883 out_free_index:
884 	ida_simple_remove(&vd_index_ida, index);
885 out:
886 	return err;
887 }
888 
889 static void virtblk_remove(struct virtio_device *vdev)
890 {
891 	struct virtio_blk *vblk = vdev->priv;
892 
893 	/* Make sure no work handler is accessing the device. */
894 	flush_work(&vblk->config_work);
895 
896 	del_gendisk(vblk->disk);
897 	blk_cleanup_queue(vblk->disk->queue);
898 
899 	blk_mq_free_tag_set(&vblk->tag_set);
900 
901 	mutex_lock(&vblk->vdev_mutex);
902 
903 	/* Stop all the virtqueues. */
904 	vdev->config->reset(vdev);
905 
906 	/* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
907 	vblk->vdev = NULL;
908 
909 	put_disk(vblk->disk);
910 	vdev->config->del_vqs(vdev);
911 	kfree(vblk->vqs);
912 
913 	mutex_unlock(&vblk->vdev_mutex);
914 
915 	virtblk_put(vblk);
916 }
917 
918 #ifdef CONFIG_PM_SLEEP
919 static int virtblk_freeze(struct virtio_device *vdev)
920 {
921 	struct virtio_blk *vblk = vdev->priv;
922 
923 	/* Ensure we don't receive any more interrupts */
924 	vdev->config->reset(vdev);
925 
926 	/* Make sure no work handler is accessing the device. */
927 	flush_work(&vblk->config_work);
928 
929 	blk_mq_quiesce_queue(vblk->disk->queue);
930 
931 	vdev->config->del_vqs(vdev);
932 	return 0;
933 }
934 
935 static int virtblk_restore(struct virtio_device *vdev)
936 {
937 	struct virtio_blk *vblk = vdev->priv;
938 	int ret;
939 
940 	ret = init_vq(vdev->priv);
941 	if (ret)
942 		return ret;
943 
944 	virtio_device_ready(vdev);
945 
946 	blk_mq_unquiesce_queue(vblk->disk->queue);
947 	return 0;
948 }
949 #endif
950 
951 static const struct virtio_device_id id_table[] = {
952 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
953 	{ 0 },
954 };
955 
956 static unsigned int features_legacy[] = {
957 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
958 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
959 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
960 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
961 }
962 ;
963 static unsigned int features[] = {
964 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
965 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
966 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
967 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
968 };
969 
970 static struct virtio_driver virtio_blk = {
971 	.feature_table			= features,
972 	.feature_table_size		= ARRAY_SIZE(features),
973 	.feature_table_legacy		= features_legacy,
974 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
975 	.driver.name			= KBUILD_MODNAME,
976 	.driver.owner			= THIS_MODULE,
977 	.id_table			= id_table,
978 	.probe				= virtblk_probe,
979 	.remove				= virtblk_remove,
980 	.config_changed			= virtblk_config_changed,
981 #ifdef CONFIG_PM_SLEEP
982 	.freeze				= virtblk_freeze,
983 	.restore			= virtblk_restore,
984 #endif
985 };
986 
987 static int __init init(void)
988 {
989 	int error;
990 
991 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
992 	if (!virtblk_wq)
993 		return -ENOMEM;
994 
995 	major = register_blkdev(0, "virtblk");
996 	if (major < 0) {
997 		error = major;
998 		goto out_destroy_workqueue;
999 	}
1000 
1001 	error = register_virtio_driver(&virtio_blk);
1002 	if (error)
1003 		goto out_unregister_blkdev;
1004 	return 0;
1005 
1006 out_unregister_blkdev:
1007 	unregister_blkdev(major, "virtblk");
1008 out_destroy_workqueue:
1009 	destroy_workqueue(virtblk_wq);
1010 	return error;
1011 }
1012 
1013 static void __exit fini(void)
1014 {
1015 	unregister_virtio_driver(&virtio_blk);
1016 	unregister_blkdev(major, "virtblk");
1017 	destroy_workqueue(virtblk_wq);
1018 }
1019 module_init(init);
1020 module_exit(fini);
1021 
1022 MODULE_DEVICE_TABLE(virtio, id_table);
1023 MODULE_DESCRIPTION("Virtio block driver");
1024 MODULE_LICENSE("GPL");
1025