xref: /linux/drivers/block/virtio_blk.c (revision d4b996f9ef1fe83d9ce9ad5c1ca0bd8231638ce5)
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 /* The maximum number of sg elements that fit into a virtqueue */
25 #define VIRTIO_BLK_MAX_SG_ELEMS 32768
26 
27 static int major;
28 static DEFINE_IDA(vd_index_ida);
29 
30 static struct workqueue_struct *virtblk_wq;
31 
32 struct virtio_blk_vq {
33 	struct virtqueue *vq;
34 	spinlock_t lock;
35 	char name[VQ_NAME_LEN];
36 } ____cacheline_aligned_in_smp;
37 
38 struct virtio_blk {
39 	/*
40 	 * This mutex must be held by anything that may run after
41 	 * virtblk_remove() sets vblk->vdev to NULL.
42 	 *
43 	 * blk-mq, virtqueue processing, and sysfs attribute code paths are
44 	 * shut down before vblk->vdev is set to NULL and therefore do not need
45 	 * to hold this mutex.
46 	 */
47 	struct mutex vdev_mutex;
48 	struct virtio_device *vdev;
49 
50 	/* The disk structure for the kernel. */
51 	struct gendisk *disk;
52 
53 	/* Block layer tags. */
54 	struct blk_mq_tag_set tag_set;
55 
56 	/* Process context for config space updates */
57 	struct work_struct config_work;
58 
59 	/*
60 	 * Tracks references from block_device_operations open/release and
61 	 * virtio_driver probe/remove so this object can be freed once no
62 	 * longer in use.
63 	 */
64 	refcount_t refs;
65 
66 	/* What host tells us, plus 2 for header & tailer. */
67 	unsigned int sg_elems;
68 
69 	/* Ida index - used to track minor number allocations. */
70 	int index;
71 
72 	/* num of vqs */
73 	int num_vqs;
74 	struct virtio_blk_vq *vqs;
75 };
76 
77 struct virtblk_req {
78 	struct virtio_blk_outhdr out_hdr;
79 	u8 status;
80 	struct scatterlist sg[];
81 };
82 
83 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
84 {
85 	switch (vbr->status) {
86 	case VIRTIO_BLK_S_OK:
87 		return BLK_STS_OK;
88 	case VIRTIO_BLK_S_UNSUPP:
89 		return BLK_STS_NOTSUPP;
90 	default:
91 		return BLK_STS_IOERR;
92 	}
93 }
94 
95 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
96 		struct scatterlist *data_sg, bool have_data)
97 {
98 	struct scatterlist hdr, status, *sgs[3];
99 	unsigned int num_out = 0, num_in = 0;
100 
101 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
102 	sgs[num_out++] = &hdr;
103 
104 	if (have_data) {
105 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
106 			sgs[num_out++] = data_sg;
107 		else
108 			sgs[num_out + num_in++] = data_sg;
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 int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
118 {
119 	unsigned short segments = blk_rq_nr_discard_segments(req);
120 	unsigned short n = 0;
121 	struct virtio_blk_discard_write_zeroes *range;
122 	struct bio *bio;
123 	u32 flags = 0;
124 
125 	if (unmap)
126 		flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
127 
128 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
129 	if (!range)
130 		return -ENOMEM;
131 
132 	/*
133 	 * Single max discard segment means multi-range discard isn't
134 	 * supported, and block layer only runs contiguity merge like
135 	 * normal RW request. So we can't reply on bio for retrieving
136 	 * each range info.
137 	 */
138 	if (queue_max_discard_segments(req->q) == 1) {
139 		range[0].flags = cpu_to_le32(flags);
140 		range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
141 		range[0].sector = cpu_to_le64(blk_rq_pos(req));
142 		n = 1;
143 	} else {
144 		__rq_for_each_bio(bio, req) {
145 			u64 sector = bio->bi_iter.bi_sector;
146 			u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
147 
148 			range[n].flags = cpu_to_le32(flags);
149 			range[n].num_sectors = cpu_to_le32(num_sectors);
150 			range[n].sector = cpu_to_le64(sector);
151 			n++;
152 		}
153 	}
154 
155 	WARN_ON_ONCE(n != segments);
156 
157 	req->special_vec.bv_page = virt_to_page(range);
158 	req->special_vec.bv_offset = offset_in_page(range);
159 	req->special_vec.bv_len = sizeof(*range) * segments;
160 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
161 
162 	return 0;
163 }
164 
165 static inline void virtblk_request_done(struct request *req)
166 {
167 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
168 
169 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
170 		kfree(page_address(req->special_vec.bv_page) +
171 		      req->special_vec.bv_offset);
172 	}
173 
174 	blk_mq_end_request(req, virtblk_result(vbr));
175 }
176 
177 static void virtblk_done(struct virtqueue *vq)
178 {
179 	struct virtio_blk *vblk = vq->vdev->priv;
180 	bool req_done = false;
181 	int qid = vq->index;
182 	struct virtblk_req *vbr;
183 	unsigned long flags;
184 	unsigned int len;
185 
186 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
187 	do {
188 		virtqueue_disable_cb(vq);
189 		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
190 			struct request *req = blk_mq_rq_from_pdu(vbr);
191 
192 			if (likely(!blk_should_fake_timeout(req->q)))
193 				blk_mq_complete_request(req);
194 			req_done = true;
195 		}
196 		if (unlikely(virtqueue_is_broken(vq)))
197 			break;
198 	} while (!virtqueue_enable_cb(vq));
199 
200 	/* In case queue is stopped waiting for more buffers. */
201 	if (req_done)
202 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
203 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
204 }
205 
206 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
207 {
208 	struct virtio_blk *vblk = hctx->queue->queuedata;
209 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
210 	bool kick;
211 
212 	spin_lock_irq(&vq->lock);
213 	kick = virtqueue_kick_prepare(vq->vq);
214 	spin_unlock_irq(&vq->lock);
215 
216 	if (kick)
217 		virtqueue_notify(vq->vq);
218 }
219 
220 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
221 			   const struct blk_mq_queue_data *bd)
222 {
223 	struct virtio_blk *vblk = hctx->queue->queuedata;
224 	struct request *req = bd->rq;
225 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
226 	unsigned long flags;
227 	unsigned int num;
228 	int qid = hctx->queue_num;
229 	int err;
230 	bool notify = false;
231 	bool unmap = false;
232 	u32 type;
233 
234 	BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
235 
236 	switch (req_op(req)) {
237 	case REQ_OP_READ:
238 	case REQ_OP_WRITE:
239 		type = 0;
240 		break;
241 	case REQ_OP_FLUSH:
242 		type = VIRTIO_BLK_T_FLUSH;
243 		break;
244 	case REQ_OP_DISCARD:
245 		type = VIRTIO_BLK_T_DISCARD;
246 		break;
247 	case REQ_OP_WRITE_ZEROES:
248 		type = VIRTIO_BLK_T_WRITE_ZEROES;
249 		unmap = !(req->cmd_flags & REQ_NOUNMAP);
250 		break;
251 	case REQ_OP_DRV_IN:
252 		type = VIRTIO_BLK_T_GET_ID;
253 		break;
254 	default:
255 		WARN_ON_ONCE(1);
256 		return BLK_STS_IOERR;
257 	}
258 
259 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
260 	vbr->out_hdr.sector = type ?
261 		0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
262 	vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
263 
264 	blk_mq_start_request(req);
265 
266 	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
267 		err = virtblk_setup_discard_write_zeroes(req, unmap);
268 		if (err)
269 			return BLK_STS_RESOURCE;
270 	}
271 
272 	num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
273 	if (num) {
274 		if (rq_data_dir(req) == WRITE)
275 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
276 		else
277 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
278 	}
279 
280 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
281 	err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
282 	if (err) {
283 		virtqueue_kick(vblk->vqs[qid].vq);
284 		/* Don't stop the queue if -ENOMEM: we may have failed to
285 		 * bounce the buffer due to global resource outage.
286 		 */
287 		if (err == -ENOSPC)
288 			blk_mq_stop_hw_queue(hctx);
289 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
290 		switch (err) {
291 		case -ENOSPC:
292 			return BLK_STS_DEV_RESOURCE;
293 		case -ENOMEM:
294 			return BLK_STS_RESOURCE;
295 		default:
296 			return BLK_STS_IOERR;
297 		}
298 	}
299 
300 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
301 		notify = true;
302 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
303 
304 	if (notify)
305 		virtqueue_notify(vblk->vqs[qid].vq);
306 	return BLK_STS_OK;
307 }
308 
309 /* return id (s/n) string for *disk to *id_str
310  */
311 static int virtblk_get_id(struct gendisk *disk, char *id_str)
312 {
313 	struct virtio_blk *vblk = disk->private_data;
314 	struct request_queue *q = vblk->disk->queue;
315 	struct request *req;
316 	int err;
317 
318 	req = blk_get_request(q, REQ_OP_DRV_IN, 0);
319 	if (IS_ERR(req))
320 		return PTR_ERR(req);
321 
322 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
323 	if (err)
324 		goto out;
325 
326 	blk_execute_rq(vblk->disk, req, false);
327 	err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
328 out:
329 	blk_put_request(req);
330 	return err;
331 }
332 
333 static void virtblk_get(struct virtio_blk *vblk)
334 {
335 	refcount_inc(&vblk->refs);
336 }
337 
338 static void virtblk_put(struct virtio_blk *vblk)
339 {
340 	if (refcount_dec_and_test(&vblk->refs)) {
341 		ida_simple_remove(&vd_index_ida, vblk->index);
342 		mutex_destroy(&vblk->vdev_mutex);
343 		kfree(vblk);
344 	}
345 }
346 
347 static int virtblk_open(struct block_device *bd, fmode_t mode)
348 {
349 	struct virtio_blk *vblk = bd->bd_disk->private_data;
350 	int ret = 0;
351 
352 	mutex_lock(&vblk->vdev_mutex);
353 
354 	if (vblk->vdev)
355 		virtblk_get(vblk);
356 	else
357 		ret = -ENXIO;
358 
359 	mutex_unlock(&vblk->vdev_mutex);
360 	return ret;
361 }
362 
363 static void virtblk_release(struct gendisk *disk, fmode_t mode)
364 {
365 	struct virtio_blk *vblk = disk->private_data;
366 
367 	virtblk_put(vblk);
368 }
369 
370 /* We provide getgeo only to please some old bootloader/partitioning tools */
371 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
372 {
373 	struct virtio_blk *vblk = bd->bd_disk->private_data;
374 	int ret = 0;
375 
376 	mutex_lock(&vblk->vdev_mutex);
377 
378 	if (!vblk->vdev) {
379 		ret = -ENXIO;
380 		goto out;
381 	}
382 
383 	/* see if the host passed in geometry config */
384 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
385 		virtio_cread(vblk->vdev, struct virtio_blk_config,
386 			     geometry.cylinders, &geo->cylinders);
387 		virtio_cread(vblk->vdev, struct virtio_blk_config,
388 			     geometry.heads, &geo->heads);
389 		virtio_cread(vblk->vdev, struct virtio_blk_config,
390 			     geometry.sectors, &geo->sectors);
391 	} else {
392 		/* some standard values, similar to sd */
393 		geo->heads = 1 << 6;
394 		geo->sectors = 1 << 5;
395 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
396 	}
397 out:
398 	mutex_unlock(&vblk->vdev_mutex);
399 	return ret;
400 }
401 
402 static const struct block_device_operations virtblk_fops = {
403 	.owner  = THIS_MODULE,
404 	.open = virtblk_open,
405 	.release = virtblk_release,
406 	.getgeo = virtblk_getgeo,
407 };
408 
409 static int index_to_minor(int index)
410 {
411 	return index << PART_BITS;
412 }
413 
414 static int minor_to_index(int minor)
415 {
416 	return minor >> PART_BITS;
417 }
418 
419 static ssize_t serial_show(struct device *dev,
420 			   struct device_attribute *attr, char *buf)
421 {
422 	struct gendisk *disk = dev_to_disk(dev);
423 	int err;
424 
425 	/* sysfs gives us a PAGE_SIZE buffer */
426 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
427 
428 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
429 	err = virtblk_get_id(disk, buf);
430 	if (!err)
431 		return strlen(buf);
432 
433 	if (err == -EIO) /* Unsupported? Make it empty. */
434 		return 0;
435 
436 	return err;
437 }
438 
439 static DEVICE_ATTR_RO(serial);
440 
441 /* The queue's logical block size must be set before calling this */
442 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
443 {
444 	struct virtio_device *vdev = vblk->vdev;
445 	struct request_queue *q = vblk->disk->queue;
446 	char cap_str_2[10], cap_str_10[10];
447 	unsigned long long nblocks;
448 	u64 capacity;
449 
450 	/* Host must always specify the capacity. */
451 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
452 
453 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
454 
455 	string_get_size(nblocks, queue_logical_block_size(q),
456 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
457 	string_get_size(nblocks, queue_logical_block_size(q),
458 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
459 
460 	dev_notice(&vdev->dev,
461 		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
462 		   vblk->disk->disk_name,
463 		   resize ? "new size: " : "",
464 		   nblocks,
465 		   queue_logical_block_size(q),
466 		   cap_str_10,
467 		   cap_str_2);
468 
469 	set_capacity_and_notify(vblk->disk, capacity);
470 }
471 
472 static void virtblk_config_changed_work(struct work_struct *work)
473 {
474 	struct virtio_blk *vblk =
475 		container_of(work, struct virtio_blk, config_work);
476 
477 	virtblk_update_capacity(vblk, true);
478 }
479 
480 static void virtblk_config_changed(struct virtio_device *vdev)
481 {
482 	struct virtio_blk *vblk = vdev->priv;
483 
484 	queue_work(virtblk_wq, &vblk->config_work);
485 }
486 
487 static int init_vq(struct virtio_blk *vblk)
488 {
489 	int err;
490 	int i;
491 	vq_callback_t **callbacks;
492 	const char **names;
493 	struct virtqueue **vqs;
494 	unsigned short num_vqs;
495 	struct virtio_device *vdev = vblk->vdev;
496 	struct irq_affinity desc = { 0, };
497 
498 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
499 				   struct virtio_blk_config, num_queues,
500 				   &num_vqs);
501 	if (err)
502 		num_vqs = 1;
503 
504 	num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
505 
506 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
507 	if (!vblk->vqs)
508 		return -ENOMEM;
509 
510 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
511 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
512 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
513 	if (!names || !callbacks || !vqs) {
514 		err = -ENOMEM;
515 		goto out;
516 	}
517 
518 	for (i = 0; i < num_vqs; i++) {
519 		callbacks[i] = virtblk_done;
520 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
521 		names[i] = vblk->vqs[i].name;
522 	}
523 
524 	/* Discover virtqueues and write information to configuration.  */
525 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
526 	if (err)
527 		goto out;
528 
529 	for (i = 0; i < num_vqs; i++) {
530 		spin_lock_init(&vblk->vqs[i].lock);
531 		vblk->vqs[i].vq = vqs[i];
532 	}
533 	vblk->num_vqs = num_vqs;
534 
535 out:
536 	kfree(vqs);
537 	kfree(callbacks);
538 	kfree(names);
539 	if (err)
540 		kfree(vblk->vqs);
541 	return err;
542 }
543 
544 /*
545  * Legacy naming scheme used for virtio devices.  We are stuck with it for
546  * virtio blk but don't ever use it for any new driver.
547  */
548 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
549 {
550 	const int base = 'z' - 'a' + 1;
551 	char *begin = buf + strlen(prefix);
552 	char *end = buf + buflen;
553 	char *p;
554 	int unit;
555 
556 	p = end - 1;
557 	*p = '\0';
558 	unit = base;
559 	do {
560 		if (p == begin)
561 			return -EINVAL;
562 		*--p = 'a' + (index % unit);
563 		index = (index / unit) - 1;
564 	} while (index >= 0);
565 
566 	memmove(begin, p, end - p);
567 	memcpy(buf, prefix, strlen(prefix));
568 
569 	return 0;
570 }
571 
572 static int virtblk_get_cache_mode(struct virtio_device *vdev)
573 {
574 	u8 writeback;
575 	int err;
576 
577 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
578 				   struct virtio_blk_config, wce,
579 				   &writeback);
580 
581 	/*
582 	 * If WCE is not configurable and flush is not available,
583 	 * assume no writeback cache is in use.
584 	 */
585 	if (err)
586 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
587 
588 	return writeback;
589 }
590 
591 static void virtblk_update_cache_mode(struct virtio_device *vdev)
592 {
593 	u8 writeback = virtblk_get_cache_mode(vdev);
594 	struct virtio_blk *vblk = vdev->priv;
595 
596 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
597 }
598 
599 static const char *const virtblk_cache_types[] = {
600 	"write through", "write back"
601 };
602 
603 static ssize_t
604 cache_type_store(struct device *dev, struct device_attribute *attr,
605 		 const char *buf, size_t count)
606 {
607 	struct gendisk *disk = dev_to_disk(dev);
608 	struct virtio_blk *vblk = disk->private_data;
609 	struct virtio_device *vdev = vblk->vdev;
610 	int i;
611 
612 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
613 	i = sysfs_match_string(virtblk_cache_types, buf);
614 	if (i < 0)
615 		return i;
616 
617 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
618 	virtblk_update_cache_mode(vdev);
619 	return count;
620 }
621 
622 static ssize_t
623 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
624 {
625 	struct gendisk *disk = dev_to_disk(dev);
626 	struct virtio_blk *vblk = disk->private_data;
627 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
628 
629 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
630 	return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
631 }
632 
633 static DEVICE_ATTR_RW(cache_type);
634 
635 static struct attribute *virtblk_attrs[] = {
636 	&dev_attr_serial.attr,
637 	&dev_attr_cache_type.attr,
638 	NULL,
639 };
640 
641 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
642 		struct attribute *a, int n)
643 {
644 	struct device *dev = kobj_to_dev(kobj);
645 	struct gendisk *disk = dev_to_disk(dev);
646 	struct virtio_blk *vblk = disk->private_data;
647 	struct virtio_device *vdev = vblk->vdev;
648 
649 	if (a == &dev_attr_cache_type.attr &&
650 	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
651 		return S_IRUGO;
652 
653 	return a->mode;
654 }
655 
656 static const struct attribute_group virtblk_attr_group = {
657 	.attrs = virtblk_attrs,
658 	.is_visible = virtblk_attrs_are_visible,
659 };
660 
661 static const struct attribute_group *virtblk_attr_groups[] = {
662 	&virtblk_attr_group,
663 	NULL,
664 };
665 
666 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
667 		unsigned int hctx_idx, unsigned int numa_node)
668 {
669 	struct virtio_blk *vblk = set->driver_data;
670 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
671 
672 	sg_init_table(vbr->sg, vblk->sg_elems);
673 	return 0;
674 }
675 
676 static int virtblk_map_queues(struct blk_mq_tag_set *set)
677 {
678 	struct virtio_blk *vblk = set->driver_data;
679 
680 	return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
681 					vblk->vdev, 0);
682 }
683 
684 static const struct blk_mq_ops virtio_mq_ops = {
685 	.queue_rq	= virtio_queue_rq,
686 	.commit_rqs	= virtio_commit_rqs,
687 	.complete	= virtblk_request_done,
688 	.init_request	= virtblk_init_request,
689 	.map_queues	= virtblk_map_queues,
690 };
691 
692 static unsigned int virtblk_queue_depth;
693 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
694 
695 static int virtblk_probe(struct virtio_device *vdev)
696 {
697 	struct virtio_blk *vblk;
698 	struct request_queue *q;
699 	int err, index;
700 
701 	u32 v, blk_size, max_size, sg_elems, opt_io_size;
702 	u16 min_io_size;
703 	u8 physical_block_exp, alignment_offset;
704 	unsigned int queue_depth;
705 
706 	if (!vdev->config->get) {
707 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
708 			__func__);
709 		return -EINVAL;
710 	}
711 
712 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
713 			     GFP_KERNEL);
714 	if (err < 0)
715 		goto out;
716 	index = err;
717 
718 	/* We need to know how many segments before we allocate. */
719 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
720 				   struct virtio_blk_config, seg_max,
721 				   &sg_elems);
722 
723 	/* We need at least one SG element, whatever they say. */
724 	if (err || !sg_elems)
725 		sg_elems = 1;
726 
727 	/* Prevent integer overflows and honor max vq size */
728 	sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
729 
730 	/* We need extra sg elements at head and tail. */
731 	sg_elems += 2;
732 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
733 	if (!vblk) {
734 		err = -ENOMEM;
735 		goto out_free_index;
736 	}
737 
738 	/* This reference is dropped in virtblk_remove(). */
739 	refcount_set(&vblk->refs, 1);
740 	mutex_init(&vblk->vdev_mutex);
741 
742 	vblk->vdev = vdev;
743 	vblk->sg_elems = sg_elems;
744 
745 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
746 
747 	err = init_vq(vblk);
748 	if (err)
749 		goto out_free_vblk;
750 
751 	/* Default queue sizing is to fill the ring. */
752 	if (likely(!virtblk_queue_depth)) {
753 		queue_depth = vblk->vqs[0].vq->num_free;
754 		/* ... but without indirect descs, we use 2 descs per req */
755 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
756 			queue_depth /= 2;
757 	} else {
758 		queue_depth = virtblk_queue_depth;
759 	}
760 
761 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
762 	vblk->tag_set.ops = &virtio_mq_ops;
763 	vblk->tag_set.queue_depth = queue_depth;
764 	vblk->tag_set.numa_node = NUMA_NO_NODE;
765 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
766 	vblk->tag_set.cmd_size =
767 		sizeof(struct virtblk_req) +
768 		sizeof(struct scatterlist) * sg_elems;
769 	vblk->tag_set.driver_data = vblk;
770 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
771 
772 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
773 	if (err)
774 		goto out_free_vq;
775 
776 	vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
777 	if (IS_ERR(vblk->disk)) {
778 		err = PTR_ERR(vblk->disk);
779 		goto out_free_tags;
780 	}
781 	q = vblk->disk->queue;
782 
783 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
784 
785 	vblk->disk->major = major;
786 	vblk->disk->first_minor = index_to_minor(index);
787 	vblk->disk->minors = 1 << PART_BITS;
788 	vblk->disk->private_data = vblk;
789 	vblk->disk->fops = &virtblk_fops;
790 	vblk->disk->flags |= GENHD_FL_EXT_DEVT;
791 	vblk->index = index;
792 
793 	/* configure queue flush support */
794 	virtblk_update_cache_mode(vdev);
795 
796 	/* If disk is read-only in the host, the guest should obey */
797 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
798 		set_disk_ro(vblk->disk, 1);
799 
800 	/* We can handle whatever the host told us to handle. */
801 	blk_queue_max_segments(q, vblk->sg_elems-2);
802 
803 	/* No real sector limit. */
804 	blk_queue_max_hw_sectors(q, -1U);
805 
806 	max_size = virtio_max_dma_size(vdev);
807 
808 	/* Host can optionally specify maximum segment size and number of
809 	 * segments. */
810 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
811 				   struct virtio_blk_config, size_max, &v);
812 	if (!err)
813 		max_size = min(max_size, v);
814 
815 	blk_queue_max_segment_size(q, max_size);
816 
817 	/* Host can optionally specify the block size of the device */
818 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
819 				   struct virtio_blk_config, blk_size,
820 				   &blk_size);
821 	if (!err)
822 		blk_queue_logical_block_size(q, blk_size);
823 	else
824 		blk_size = queue_logical_block_size(q);
825 
826 	/* Use topology information if available */
827 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
828 				   struct virtio_blk_config, physical_block_exp,
829 				   &physical_block_exp);
830 	if (!err && physical_block_exp)
831 		blk_queue_physical_block_size(q,
832 				blk_size * (1 << physical_block_exp));
833 
834 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
835 				   struct virtio_blk_config, alignment_offset,
836 				   &alignment_offset);
837 	if (!err && alignment_offset)
838 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
839 
840 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
841 				   struct virtio_blk_config, min_io_size,
842 				   &min_io_size);
843 	if (!err && min_io_size)
844 		blk_queue_io_min(q, blk_size * min_io_size);
845 
846 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
847 				   struct virtio_blk_config, opt_io_size,
848 				   &opt_io_size);
849 	if (!err && opt_io_size)
850 		blk_queue_io_opt(q, blk_size * opt_io_size);
851 
852 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
853 		q->limits.discard_granularity = blk_size;
854 
855 		virtio_cread(vdev, struct virtio_blk_config,
856 			     discard_sector_alignment, &v);
857 		q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
858 
859 		virtio_cread(vdev, struct virtio_blk_config,
860 			     max_discard_sectors, &v);
861 		blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
862 
863 		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
864 			     &v);
865 		blk_queue_max_discard_segments(q,
866 					       min_not_zero(v,
867 							    MAX_DISCARD_SEGMENTS));
868 
869 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
870 	}
871 
872 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
873 		virtio_cread(vdev, struct virtio_blk_config,
874 			     max_write_zeroes_sectors, &v);
875 		blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
876 	}
877 
878 	virtblk_update_capacity(vblk, false);
879 	virtio_device_ready(vdev);
880 
881 	device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
882 	return 0;
883 
884 out_free_tags:
885 	blk_mq_free_tag_set(&vblk->tag_set);
886 out_free_vq:
887 	vdev->config->del_vqs(vdev);
888 	kfree(vblk->vqs);
889 out_free_vblk:
890 	kfree(vblk);
891 out_free_index:
892 	ida_simple_remove(&vd_index_ida, index);
893 out:
894 	return err;
895 }
896 
897 static void virtblk_remove(struct virtio_device *vdev)
898 {
899 	struct virtio_blk *vblk = vdev->priv;
900 
901 	/* Make sure no work handler is accessing the device. */
902 	flush_work(&vblk->config_work);
903 
904 	del_gendisk(vblk->disk);
905 	blk_cleanup_disk(vblk->disk);
906 	blk_mq_free_tag_set(&vblk->tag_set);
907 
908 	mutex_lock(&vblk->vdev_mutex);
909 
910 	/* Stop all the virtqueues. */
911 	vdev->config->reset(vdev);
912 
913 	/* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
914 	vblk->vdev = NULL;
915 
916 	vdev->config->del_vqs(vdev);
917 	kfree(vblk->vqs);
918 
919 	mutex_unlock(&vblk->vdev_mutex);
920 
921 	virtblk_put(vblk);
922 }
923 
924 #ifdef CONFIG_PM_SLEEP
925 static int virtblk_freeze(struct virtio_device *vdev)
926 {
927 	struct virtio_blk *vblk = vdev->priv;
928 
929 	/* Ensure we don't receive any more interrupts */
930 	vdev->config->reset(vdev);
931 
932 	/* Make sure no work handler is accessing the device. */
933 	flush_work(&vblk->config_work);
934 
935 	blk_mq_quiesce_queue(vblk->disk->queue);
936 
937 	vdev->config->del_vqs(vdev);
938 	kfree(vblk->vqs);
939 
940 	return 0;
941 }
942 
943 static int virtblk_restore(struct virtio_device *vdev)
944 {
945 	struct virtio_blk *vblk = vdev->priv;
946 	int ret;
947 
948 	ret = init_vq(vdev->priv);
949 	if (ret)
950 		return ret;
951 
952 	virtio_device_ready(vdev);
953 
954 	blk_mq_unquiesce_queue(vblk->disk->queue);
955 	return 0;
956 }
957 #endif
958 
959 static const struct virtio_device_id id_table[] = {
960 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
961 	{ 0 },
962 };
963 
964 static unsigned int features_legacy[] = {
965 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
966 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
967 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
968 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
969 }
970 ;
971 static unsigned int features[] = {
972 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
973 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
974 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
975 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
976 };
977 
978 static struct virtio_driver virtio_blk = {
979 	.feature_table			= features,
980 	.feature_table_size		= ARRAY_SIZE(features),
981 	.feature_table_legacy		= features_legacy,
982 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
983 	.driver.name			= KBUILD_MODNAME,
984 	.driver.owner			= THIS_MODULE,
985 	.id_table			= id_table,
986 	.probe				= virtblk_probe,
987 	.remove				= virtblk_remove,
988 	.config_changed			= virtblk_config_changed,
989 #ifdef CONFIG_PM_SLEEP
990 	.freeze				= virtblk_freeze,
991 	.restore			= virtblk_restore,
992 #endif
993 };
994 
995 static int __init init(void)
996 {
997 	int error;
998 
999 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1000 	if (!virtblk_wq)
1001 		return -ENOMEM;
1002 
1003 	major = register_blkdev(0, "virtblk");
1004 	if (major < 0) {
1005 		error = major;
1006 		goto out_destroy_workqueue;
1007 	}
1008 
1009 	error = register_virtio_driver(&virtio_blk);
1010 	if (error)
1011 		goto out_unregister_blkdev;
1012 	return 0;
1013 
1014 out_unregister_blkdev:
1015 	unregister_blkdev(major, "virtblk");
1016 out_destroy_workqueue:
1017 	destroy_workqueue(virtblk_wq);
1018 	return error;
1019 }
1020 
1021 static void __exit fini(void)
1022 {
1023 	unregister_virtio_driver(&virtio_blk);
1024 	unregister_blkdev(major, "virtblk");
1025 	destroy_workqueue(virtblk_wq);
1026 }
1027 module_init(init);
1028 module_exit(fini);
1029 
1030 MODULE_DEVICE_TABLE(virtio, id_table);
1031 MODULE_DESCRIPTION("Virtio block driver");
1032 MODULE_LICENSE("GPL");
1033