xref: /linux/drivers/block/virtio_blk.c (revision 8eecf1c9929aef24e9e75280a39ed1ba3c64fb71)
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 #ifdef CONFIG_ARCH_NO_SG_CHAIN
28 #define VIRTIO_BLK_INLINE_SG_CNT	0
29 #else
30 #define VIRTIO_BLK_INLINE_SG_CNT	2
31 #endif
32 
33 static unsigned int num_request_queues;
34 module_param(num_request_queues, uint, 0644);
35 MODULE_PARM_DESC(num_request_queues,
36 		 "Limit the number of request queues to use for blk device. "
37 		 "0 for no limit. "
38 		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
39 
40 static int major;
41 static DEFINE_IDA(vd_index_ida);
42 
43 static struct workqueue_struct *virtblk_wq;
44 
45 struct virtio_blk_vq {
46 	struct virtqueue *vq;
47 	spinlock_t lock;
48 	char name[VQ_NAME_LEN];
49 } ____cacheline_aligned_in_smp;
50 
51 struct virtio_blk {
52 	/*
53 	 * This mutex must be held by anything that may run after
54 	 * virtblk_remove() sets vblk->vdev to NULL.
55 	 *
56 	 * blk-mq, virtqueue processing, and sysfs attribute code paths are
57 	 * shut down before vblk->vdev is set to NULL and therefore do not need
58 	 * to hold this mutex.
59 	 */
60 	struct mutex vdev_mutex;
61 	struct virtio_device *vdev;
62 
63 	/* The disk structure for the kernel. */
64 	struct gendisk *disk;
65 
66 	/* Block layer tags. */
67 	struct blk_mq_tag_set tag_set;
68 
69 	/* Process context for config space updates */
70 	struct work_struct config_work;
71 
72 	/* Ida index - used to track minor number allocations. */
73 	int index;
74 
75 	/* num of vqs */
76 	int num_vqs;
77 	struct virtio_blk_vq *vqs;
78 };
79 
80 struct virtblk_req {
81 	struct virtio_blk_outhdr out_hdr;
82 	u8 status;
83 	struct sg_table sg_table;
84 	struct scatterlist sg[];
85 };
86 
87 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
88 {
89 	switch (vbr->status) {
90 	case VIRTIO_BLK_S_OK:
91 		return BLK_STS_OK;
92 	case VIRTIO_BLK_S_UNSUPP:
93 		return BLK_STS_NOTSUPP;
94 	default:
95 		return BLK_STS_IOERR;
96 	}
97 }
98 
99 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
100 		struct scatterlist *data_sg, bool have_data)
101 {
102 	struct scatterlist hdr, status, *sgs[3];
103 	unsigned int num_out = 0, num_in = 0;
104 
105 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
106 	sgs[num_out++] = &hdr;
107 
108 	if (have_data) {
109 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
110 			sgs[num_out++] = data_sg;
111 		else
112 			sgs[num_out + num_in++] = data_sg;
113 	}
114 
115 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
116 	sgs[num_out + num_in++] = &status;
117 
118 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
119 }
120 
121 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
122 {
123 	unsigned short segments = blk_rq_nr_discard_segments(req);
124 	unsigned short n = 0;
125 	struct virtio_blk_discard_write_zeroes *range;
126 	struct bio *bio;
127 	u32 flags = 0;
128 
129 	if (unmap)
130 		flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
131 
132 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
133 	if (!range)
134 		return -ENOMEM;
135 
136 	/*
137 	 * Single max discard segment means multi-range discard isn't
138 	 * supported, and block layer only runs contiguity merge like
139 	 * normal RW request. So we can't reply on bio for retrieving
140 	 * each range info.
141 	 */
142 	if (queue_max_discard_segments(req->q) == 1) {
143 		range[0].flags = cpu_to_le32(flags);
144 		range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
145 		range[0].sector = cpu_to_le64(blk_rq_pos(req));
146 		n = 1;
147 	} else {
148 		__rq_for_each_bio(bio, req) {
149 			u64 sector = bio->bi_iter.bi_sector;
150 			u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
151 
152 			range[n].flags = cpu_to_le32(flags);
153 			range[n].num_sectors = cpu_to_le32(num_sectors);
154 			range[n].sector = cpu_to_le64(sector);
155 			n++;
156 		}
157 	}
158 
159 	WARN_ON_ONCE(n != segments);
160 
161 	req->special_vec.bv_page = virt_to_page(range);
162 	req->special_vec.bv_offset = offset_in_page(range);
163 	req->special_vec.bv_len = sizeof(*range) * segments;
164 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
165 
166 	return 0;
167 }
168 
169 static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
170 {
171 	if (blk_rq_nr_phys_segments(req))
172 		sg_free_table_chained(&vbr->sg_table,
173 				      VIRTIO_BLK_INLINE_SG_CNT);
174 }
175 
176 static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
177 		struct virtblk_req *vbr)
178 {
179 	int err;
180 
181 	if (!blk_rq_nr_phys_segments(req))
182 		return 0;
183 
184 	vbr->sg_table.sgl = vbr->sg;
185 	err = sg_alloc_table_chained(&vbr->sg_table,
186 				     blk_rq_nr_phys_segments(req),
187 				     vbr->sg_table.sgl,
188 				     VIRTIO_BLK_INLINE_SG_CNT);
189 	if (unlikely(err))
190 		return -ENOMEM;
191 
192 	return blk_rq_map_sg(hctx->queue, req, vbr->sg_table.sgl);
193 }
194 
195 static void virtblk_cleanup_cmd(struct request *req)
196 {
197 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
198 		kfree(bvec_virt(&req->special_vec));
199 }
200 
201 static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
202 				      struct request *req,
203 				      struct virtblk_req *vbr)
204 {
205 	bool unmap = false;
206 	u32 type;
207 
208 	vbr->out_hdr.sector = 0;
209 
210 	switch (req_op(req)) {
211 	case REQ_OP_READ:
212 		type = VIRTIO_BLK_T_IN;
213 		vbr->out_hdr.sector = cpu_to_virtio64(vdev,
214 						      blk_rq_pos(req));
215 		break;
216 	case REQ_OP_WRITE:
217 		type = VIRTIO_BLK_T_OUT;
218 		vbr->out_hdr.sector = cpu_to_virtio64(vdev,
219 						      blk_rq_pos(req));
220 		break;
221 	case REQ_OP_FLUSH:
222 		type = VIRTIO_BLK_T_FLUSH;
223 		break;
224 	case REQ_OP_DISCARD:
225 		type = VIRTIO_BLK_T_DISCARD;
226 		break;
227 	case REQ_OP_WRITE_ZEROES:
228 		type = VIRTIO_BLK_T_WRITE_ZEROES;
229 		unmap = !(req->cmd_flags & REQ_NOUNMAP);
230 		break;
231 	case REQ_OP_DRV_IN:
232 		type = VIRTIO_BLK_T_GET_ID;
233 		break;
234 	default:
235 		WARN_ON_ONCE(1);
236 		return BLK_STS_IOERR;
237 	}
238 
239 	vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
240 	vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
241 
242 	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
243 		if (virtblk_setup_discard_write_zeroes(req, unmap))
244 			return BLK_STS_RESOURCE;
245 	}
246 
247 	return 0;
248 }
249 
250 static inline void virtblk_request_done(struct request *req)
251 {
252 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
253 
254 	virtblk_unmap_data(req, vbr);
255 	virtblk_cleanup_cmd(req);
256 	blk_mq_end_request(req, virtblk_result(vbr));
257 }
258 
259 static void virtblk_done(struct virtqueue *vq)
260 {
261 	struct virtio_blk *vblk = vq->vdev->priv;
262 	bool req_done = false;
263 	int qid = vq->index;
264 	struct virtblk_req *vbr;
265 	unsigned long flags;
266 	unsigned int len;
267 
268 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
269 	do {
270 		virtqueue_disable_cb(vq);
271 		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
272 			struct request *req = blk_mq_rq_from_pdu(vbr);
273 
274 			if (likely(!blk_should_fake_timeout(req->q)))
275 				blk_mq_complete_request(req);
276 			req_done = true;
277 		}
278 		if (unlikely(virtqueue_is_broken(vq)))
279 			break;
280 	} while (!virtqueue_enable_cb(vq));
281 
282 	/* In case queue is stopped waiting for more buffers. */
283 	if (req_done)
284 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
285 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
286 }
287 
288 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
289 {
290 	struct virtio_blk *vblk = hctx->queue->queuedata;
291 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
292 	bool kick;
293 
294 	spin_lock_irq(&vq->lock);
295 	kick = virtqueue_kick_prepare(vq->vq);
296 	spin_unlock_irq(&vq->lock);
297 
298 	if (kick)
299 		virtqueue_notify(vq->vq);
300 }
301 
302 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
303 			   const struct blk_mq_queue_data *bd)
304 {
305 	struct virtio_blk *vblk = hctx->queue->queuedata;
306 	struct request *req = bd->rq;
307 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
308 	unsigned long flags;
309 	int num;
310 	int qid = hctx->queue_num;
311 	bool notify = false;
312 	blk_status_t status;
313 	int err;
314 
315 	status = virtblk_setup_cmd(vblk->vdev, req, vbr);
316 	if (unlikely(status))
317 		return status;
318 
319 	blk_mq_start_request(req);
320 
321 	num = virtblk_map_data(hctx, req, vbr);
322 	if (unlikely(num < 0)) {
323 		virtblk_cleanup_cmd(req);
324 		return BLK_STS_RESOURCE;
325 	}
326 
327 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
328 	err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg_table.sgl, num);
329 	if (err) {
330 		virtqueue_kick(vblk->vqs[qid].vq);
331 		/* Don't stop the queue if -ENOMEM: we may have failed to
332 		 * bounce the buffer due to global resource outage.
333 		 */
334 		if (err == -ENOSPC)
335 			blk_mq_stop_hw_queue(hctx);
336 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
337 		virtblk_unmap_data(req, vbr);
338 		virtblk_cleanup_cmd(req);
339 		switch (err) {
340 		case -ENOSPC:
341 			return BLK_STS_DEV_RESOURCE;
342 		case -ENOMEM:
343 			return BLK_STS_RESOURCE;
344 		default:
345 			return BLK_STS_IOERR;
346 		}
347 	}
348 
349 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
350 		notify = true;
351 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
352 
353 	if (notify)
354 		virtqueue_notify(vblk->vqs[qid].vq);
355 	return BLK_STS_OK;
356 }
357 
358 /* return id (s/n) string for *disk to *id_str
359  */
360 static int virtblk_get_id(struct gendisk *disk, char *id_str)
361 {
362 	struct virtio_blk *vblk = disk->private_data;
363 	struct request_queue *q = vblk->disk->queue;
364 	struct request *req;
365 	int err;
366 
367 	req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
368 	if (IS_ERR(req))
369 		return PTR_ERR(req);
370 
371 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
372 	if (err)
373 		goto out;
374 
375 	blk_execute_rq(req, false);
376 	err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
377 out:
378 	blk_mq_free_request(req);
379 	return err;
380 }
381 
382 /* We provide getgeo only to please some old bootloader/partitioning tools */
383 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
384 {
385 	struct virtio_blk *vblk = bd->bd_disk->private_data;
386 	int ret = 0;
387 
388 	mutex_lock(&vblk->vdev_mutex);
389 
390 	if (!vblk->vdev) {
391 		ret = -ENXIO;
392 		goto out;
393 	}
394 
395 	/* see if the host passed in geometry config */
396 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
397 		virtio_cread(vblk->vdev, struct virtio_blk_config,
398 			     geometry.cylinders, &geo->cylinders);
399 		virtio_cread(vblk->vdev, struct virtio_blk_config,
400 			     geometry.heads, &geo->heads);
401 		virtio_cread(vblk->vdev, struct virtio_blk_config,
402 			     geometry.sectors, &geo->sectors);
403 	} else {
404 		/* some standard values, similar to sd */
405 		geo->heads = 1 << 6;
406 		geo->sectors = 1 << 5;
407 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
408 	}
409 out:
410 	mutex_unlock(&vblk->vdev_mutex);
411 	return ret;
412 }
413 
414 static void virtblk_free_disk(struct gendisk *disk)
415 {
416 	struct virtio_blk *vblk = disk->private_data;
417 
418 	ida_simple_remove(&vd_index_ida, vblk->index);
419 	mutex_destroy(&vblk->vdev_mutex);
420 	kfree(vblk);
421 }
422 
423 static const struct block_device_operations virtblk_fops = {
424 	.owner  	= THIS_MODULE,
425 	.getgeo		= virtblk_getgeo,
426 	.free_disk	= virtblk_free_disk,
427 };
428 
429 static int index_to_minor(int index)
430 {
431 	return index << PART_BITS;
432 }
433 
434 static int minor_to_index(int minor)
435 {
436 	return minor >> PART_BITS;
437 }
438 
439 static ssize_t serial_show(struct device *dev,
440 			   struct device_attribute *attr, char *buf)
441 {
442 	struct gendisk *disk = dev_to_disk(dev);
443 	int err;
444 
445 	/* sysfs gives us a PAGE_SIZE buffer */
446 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
447 
448 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
449 	err = virtblk_get_id(disk, buf);
450 	if (!err)
451 		return strlen(buf);
452 
453 	if (err == -EIO) /* Unsupported? Make it empty. */
454 		return 0;
455 
456 	return err;
457 }
458 
459 static DEVICE_ATTR_RO(serial);
460 
461 /* The queue's logical block size must be set before calling this */
462 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
463 {
464 	struct virtio_device *vdev = vblk->vdev;
465 	struct request_queue *q = vblk->disk->queue;
466 	char cap_str_2[10], cap_str_10[10];
467 	unsigned long long nblocks;
468 	u64 capacity;
469 
470 	/* Host must always specify the capacity. */
471 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
472 
473 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
474 
475 	string_get_size(nblocks, queue_logical_block_size(q),
476 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
477 	string_get_size(nblocks, queue_logical_block_size(q),
478 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
479 
480 	dev_notice(&vdev->dev,
481 		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
482 		   vblk->disk->disk_name,
483 		   resize ? "new size: " : "",
484 		   nblocks,
485 		   queue_logical_block_size(q),
486 		   cap_str_10,
487 		   cap_str_2);
488 
489 	set_capacity_and_notify(vblk->disk, capacity);
490 }
491 
492 static void virtblk_config_changed_work(struct work_struct *work)
493 {
494 	struct virtio_blk *vblk =
495 		container_of(work, struct virtio_blk, config_work);
496 
497 	virtblk_update_capacity(vblk, true);
498 }
499 
500 static void virtblk_config_changed(struct virtio_device *vdev)
501 {
502 	struct virtio_blk *vblk = vdev->priv;
503 
504 	queue_work(virtblk_wq, &vblk->config_work);
505 }
506 
507 static int init_vq(struct virtio_blk *vblk)
508 {
509 	int err;
510 	int i;
511 	vq_callback_t **callbacks;
512 	const char **names;
513 	struct virtqueue **vqs;
514 	unsigned short num_vqs;
515 	struct virtio_device *vdev = vblk->vdev;
516 	struct irq_affinity desc = { 0, };
517 
518 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
519 				   struct virtio_blk_config, num_queues,
520 				   &num_vqs);
521 	if (err)
522 		num_vqs = 1;
523 	if (!err && !num_vqs) {
524 		dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
525 		return -EINVAL;
526 	}
527 
528 	num_vqs = min_t(unsigned int,
529 			min_not_zero(num_request_queues, nr_cpu_ids),
530 			num_vqs);
531 
532 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
533 	if (!vblk->vqs)
534 		return -ENOMEM;
535 
536 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
537 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
538 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
539 	if (!names || !callbacks || !vqs) {
540 		err = -ENOMEM;
541 		goto out;
542 	}
543 
544 	for (i = 0; i < num_vqs; i++) {
545 		callbacks[i] = virtblk_done;
546 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
547 		names[i] = vblk->vqs[i].name;
548 	}
549 
550 	/* Discover virtqueues and write information to configuration.  */
551 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
552 	if (err)
553 		goto out;
554 
555 	for (i = 0; i < num_vqs; i++) {
556 		spin_lock_init(&vblk->vqs[i].lock);
557 		vblk->vqs[i].vq = vqs[i];
558 	}
559 	vblk->num_vqs = num_vqs;
560 
561 out:
562 	kfree(vqs);
563 	kfree(callbacks);
564 	kfree(names);
565 	if (err)
566 		kfree(vblk->vqs);
567 	return err;
568 }
569 
570 /*
571  * Legacy naming scheme used for virtio devices.  We are stuck with it for
572  * virtio blk but don't ever use it for any new driver.
573  */
574 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
575 {
576 	const int base = 'z' - 'a' + 1;
577 	char *begin = buf + strlen(prefix);
578 	char *end = buf + buflen;
579 	char *p;
580 	int unit;
581 
582 	p = end - 1;
583 	*p = '\0';
584 	unit = base;
585 	do {
586 		if (p == begin)
587 			return -EINVAL;
588 		*--p = 'a' + (index % unit);
589 		index = (index / unit) - 1;
590 	} while (index >= 0);
591 
592 	memmove(begin, p, end - p);
593 	memcpy(buf, prefix, strlen(prefix));
594 
595 	return 0;
596 }
597 
598 static int virtblk_get_cache_mode(struct virtio_device *vdev)
599 {
600 	u8 writeback;
601 	int err;
602 
603 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
604 				   struct virtio_blk_config, wce,
605 				   &writeback);
606 
607 	/*
608 	 * If WCE is not configurable and flush is not available,
609 	 * assume no writeback cache is in use.
610 	 */
611 	if (err)
612 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
613 
614 	return writeback;
615 }
616 
617 static void virtblk_update_cache_mode(struct virtio_device *vdev)
618 {
619 	u8 writeback = virtblk_get_cache_mode(vdev);
620 	struct virtio_blk *vblk = vdev->priv;
621 
622 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
623 }
624 
625 static const char *const virtblk_cache_types[] = {
626 	"write through", "write back"
627 };
628 
629 static ssize_t
630 cache_type_store(struct device *dev, struct device_attribute *attr,
631 		 const char *buf, size_t count)
632 {
633 	struct gendisk *disk = dev_to_disk(dev);
634 	struct virtio_blk *vblk = disk->private_data;
635 	struct virtio_device *vdev = vblk->vdev;
636 	int i;
637 
638 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
639 	i = sysfs_match_string(virtblk_cache_types, buf);
640 	if (i < 0)
641 		return i;
642 
643 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
644 	virtblk_update_cache_mode(vdev);
645 	return count;
646 }
647 
648 static ssize_t
649 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
650 {
651 	struct gendisk *disk = dev_to_disk(dev);
652 	struct virtio_blk *vblk = disk->private_data;
653 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
654 
655 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
656 	return sysfs_emit(buf, "%s\n", virtblk_cache_types[writeback]);
657 }
658 
659 static DEVICE_ATTR_RW(cache_type);
660 
661 static struct attribute *virtblk_attrs[] = {
662 	&dev_attr_serial.attr,
663 	&dev_attr_cache_type.attr,
664 	NULL,
665 };
666 
667 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
668 		struct attribute *a, int n)
669 {
670 	struct device *dev = kobj_to_dev(kobj);
671 	struct gendisk *disk = dev_to_disk(dev);
672 	struct virtio_blk *vblk = disk->private_data;
673 	struct virtio_device *vdev = vblk->vdev;
674 
675 	if (a == &dev_attr_cache_type.attr &&
676 	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
677 		return S_IRUGO;
678 
679 	return a->mode;
680 }
681 
682 static const struct attribute_group virtblk_attr_group = {
683 	.attrs = virtblk_attrs,
684 	.is_visible = virtblk_attrs_are_visible,
685 };
686 
687 static const struct attribute_group *virtblk_attr_groups[] = {
688 	&virtblk_attr_group,
689 	NULL,
690 };
691 
692 static int virtblk_map_queues(struct blk_mq_tag_set *set)
693 {
694 	struct virtio_blk *vblk = set->driver_data;
695 
696 	return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
697 					vblk->vdev, 0);
698 }
699 
700 static const struct blk_mq_ops virtio_mq_ops = {
701 	.queue_rq	= virtio_queue_rq,
702 	.commit_rqs	= virtio_commit_rqs,
703 	.complete	= virtblk_request_done,
704 	.map_queues	= virtblk_map_queues,
705 };
706 
707 static unsigned int virtblk_queue_depth;
708 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
709 
710 static int virtblk_probe(struct virtio_device *vdev)
711 {
712 	struct virtio_blk *vblk;
713 	struct request_queue *q;
714 	int err, index;
715 
716 	u32 v, blk_size, max_size, sg_elems, opt_io_size;
717 	u16 min_io_size;
718 	u8 physical_block_exp, alignment_offset;
719 	unsigned int queue_depth;
720 
721 	if (!vdev->config->get) {
722 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
723 			__func__);
724 		return -EINVAL;
725 	}
726 
727 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
728 			     GFP_KERNEL);
729 	if (err < 0)
730 		goto out;
731 	index = err;
732 
733 	/* We need to know how many segments before we allocate. */
734 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
735 				   struct virtio_blk_config, seg_max,
736 				   &sg_elems);
737 
738 	/* We need at least one SG element, whatever they say. */
739 	if (err || !sg_elems)
740 		sg_elems = 1;
741 
742 	/* Prevent integer overflows and honor max vq size */
743 	sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
744 
745 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
746 	if (!vblk) {
747 		err = -ENOMEM;
748 		goto out_free_index;
749 	}
750 
751 	mutex_init(&vblk->vdev_mutex);
752 
753 	vblk->vdev = vdev;
754 
755 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
756 
757 	err = init_vq(vblk);
758 	if (err)
759 		goto out_free_vblk;
760 
761 	/* Default queue sizing is to fill the ring. */
762 	if (!virtblk_queue_depth) {
763 		queue_depth = vblk->vqs[0].vq->num_free;
764 		/* ... but without indirect descs, we use 2 descs per req */
765 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
766 			queue_depth /= 2;
767 	} else {
768 		queue_depth = virtblk_queue_depth;
769 	}
770 
771 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
772 	vblk->tag_set.ops = &virtio_mq_ops;
773 	vblk->tag_set.queue_depth = queue_depth;
774 	vblk->tag_set.numa_node = NUMA_NO_NODE;
775 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
776 	vblk->tag_set.cmd_size =
777 		sizeof(struct virtblk_req) +
778 		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
779 	vblk->tag_set.driver_data = vblk;
780 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
781 
782 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
783 	if (err)
784 		goto out_free_vq;
785 
786 	vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
787 	if (IS_ERR(vblk->disk)) {
788 		err = PTR_ERR(vblk->disk);
789 		goto out_free_tags;
790 	}
791 	q = vblk->disk->queue;
792 
793 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
794 
795 	vblk->disk->major = major;
796 	vblk->disk->first_minor = index_to_minor(index);
797 	vblk->disk->minors = 1 << PART_BITS;
798 	vblk->disk->private_data = vblk;
799 	vblk->disk->fops = &virtblk_fops;
800 	vblk->index = index;
801 
802 	/* configure queue flush support */
803 	virtblk_update_cache_mode(vdev);
804 
805 	/* If disk is read-only in the host, the guest should obey */
806 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
807 		set_disk_ro(vblk->disk, 1);
808 
809 	/* We can handle whatever the host told us to handle. */
810 	blk_queue_max_segments(q, sg_elems);
811 
812 	/* No real sector limit. */
813 	blk_queue_max_hw_sectors(q, -1U);
814 
815 	max_size = virtio_max_dma_size(vdev);
816 
817 	/* Host can optionally specify maximum segment size and number of
818 	 * segments. */
819 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
820 				   struct virtio_blk_config, size_max, &v);
821 	if (!err)
822 		max_size = min(max_size, v);
823 
824 	blk_queue_max_segment_size(q, max_size);
825 
826 	/* Host can optionally specify the block size of the device */
827 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
828 				   struct virtio_blk_config, blk_size,
829 				   &blk_size);
830 	if (!err) {
831 		err = blk_validate_block_size(blk_size);
832 		if (err) {
833 			dev_err(&vdev->dev,
834 				"virtio_blk: invalid block size: 0x%x\n",
835 				blk_size);
836 			goto out_cleanup_disk;
837 		}
838 
839 		blk_queue_logical_block_size(q, blk_size);
840 	} else
841 		blk_size = queue_logical_block_size(q);
842 
843 	/* Use topology information if available */
844 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
845 				   struct virtio_blk_config, physical_block_exp,
846 				   &physical_block_exp);
847 	if (!err && physical_block_exp)
848 		blk_queue_physical_block_size(q,
849 				blk_size * (1 << physical_block_exp));
850 
851 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
852 				   struct virtio_blk_config, alignment_offset,
853 				   &alignment_offset);
854 	if (!err && alignment_offset)
855 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
856 
857 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
858 				   struct virtio_blk_config, min_io_size,
859 				   &min_io_size);
860 	if (!err && min_io_size)
861 		blk_queue_io_min(q, blk_size * min_io_size);
862 
863 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
864 				   struct virtio_blk_config, opt_io_size,
865 				   &opt_io_size);
866 	if (!err && opt_io_size)
867 		blk_queue_io_opt(q, blk_size * opt_io_size);
868 
869 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
870 		virtio_cread(vdev, struct virtio_blk_config,
871 			     discard_sector_alignment, &v);
872 		if (v)
873 			q->limits.discard_granularity = v << SECTOR_SHIFT;
874 		else
875 			q->limits.discard_granularity = blk_size;
876 
877 		virtio_cread(vdev, struct virtio_blk_config,
878 			     max_discard_sectors, &v);
879 		blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
880 
881 		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
882 			     &v);
883 
884 		/*
885 		 * max_discard_seg == 0 is out of spec but we always
886 		 * handled it.
887 		 */
888 		if (!v)
889 			v = sg_elems;
890 		blk_queue_max_discard_segments(q,
891 					       min(v, MAX_DISCARD_SEGMENTS));
892 	}
893 
894 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
895 		virtio_cread(vdev, struct virtio_blk_config,
896 			     max_write_zeroes_sectors, &v);
897 		blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
898 	}
899 
900 	virtblk_update_capacity(vblk, false);
901 	virtio_device_ready(vdev);
902 
903 	err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
904 	if (err)
905 		goto out_cleanup_disk;
906 
907 	return 0;
908 
909 out_cleanup_disk:
910 	blk_cleanup_disk(vblk->disk);
911 out_free_tags:
912 	blk_mq_free_tag_set(&vblk->tag_set);
913 out_free_vq:
914 	vdev->config->del_vqs(vdev);
915 	kfree(vblk->vqs);
916 out_free_vblk:
917 	kfree(vblk);
918 out_free_index:
919 	ida_simple_remove(&vd_index_ida, index);
920 out:
921 	return err;
922 }
923 
924 static void virtblk_remove(struct virtio_device *vdev)
925 {
926 	struct virtio_blk *vblk = vdev->priv;
927 
928 	/* Make sure no work handler is accessing the device. */
929 	flush_work(&vblk->config_work);
930 
931 	del_gendisk(vblk->disk);
932 	blk_cleanup_queue(vblk->disk->queue);
933 	blk_mq_free_tag_set(&vblk->tag_set);
934 
935 	mutex_lock(&vblk->vdev_mutex);
936 
937 	/* Stop all the virtqueues. */
938 	virtio_reset_device(vdev);
939 
940 	/* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
941 	vblk->vdev = NULL;
942 
943 	vdev->config->del_vqs(vdev);
944 	kfree(vblk->vqs);
945 
946 	mutex_unlock(&vblk->vdev_mutex);
947 
948 	put_disk(vblk->disk);
949 }
950 
951 #ifdef CONFIG_PM_SLEEP
952 static int virtblk_freeze(struct virtio_device *vdev)
953 {
954 	struct virtio_blk *vblk = vdev->priv;
955 
956 	/* Ensure we don't receive any more interrupts */
957 	virtio_reset_device(vdev);
958 
959 	/* Make sure no work handler is accessing the device. */
960 	flush_work(&vblk->config_work);
961 
962 	blk_mq_quiesce_queue(vblk->disk->queue);
963 
964 	vdev->config->del_vqs(vdev);
965 	kfree(vblk->vqs);
966 
967 	return 0;
968 }
969 
970 static int virtblk_restore(struct virtio_device *vdev)
971 {
972 	struct virtio_blk *vblk = vdev->priv;
973 	int ret;
974 
975 	ret = init_vq(vdev->priv);
976 	if (ret)
977 		return ret;
978 
979 	virtio_device_ready(vdev);
980 
981 	blk_mq_unquiesce_queue(vblk->disk->queue);
982 	return 0;
983 }
984 #endif
985 
986 static const struct virtio_device_id id_table[] = {
987 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
988 	{ 0 },
989 };
990 
991 static unsigned int features_legacy[] = {
992 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
993 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
994 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
995 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
996 }
997 ;
998 static unsigned int features[] = {
999 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1000 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1001 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1002 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1003 };
1004 
1005 static struct virtio_driver virtio_blk = {
1006 	.feature_table			= features,
1007 	.feature_table_size		= ARRAY_SIZE(features),
1008 	.feature_table_legacy		= features_legacy,
1009 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
1010 	.driver.name			= KBUILD_MODNAME,
1011 	.driver.owner			= THIS_MODULE,
1012 	.id_table			= id_table,
1013 	.probe				= virtblk_probe,
1014 	.remove				= virtblk_remove,
1015 	.config_changed			= virtblk_config_changed,
1016 #ifdef CONFIG_PM_SLEEP
1017 	.freeze				= virtblk_freeze,
1018 	.restore			= virtblk_restore,
1019 #endif
1020 };
1021 
1022 static int __init virtio_blk_init(void)
1023 {
1024 	int error;
1025 
1026 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1027 	if (!virtblk_wq)
1028 		return -ENOMEM;
1029 
1030 	major = register_blkdev(0, "virtblk");
1031 	if (major < 0) {
1032 		error = major;
1033 		goto out_destroy_workqueue;
1034 	}
1035 
1036 	error = register_virtio_driver(&virtio_blk);
1037 	if (error)
1038 		goto out_unregister_blkdev;
1039 	return 0;
1040 
1041 out_unregister_blkdev:
1042 	unregister_blkdev(major, "virtblk");
1043 out_destroy_workqueue:
1044 	destroy_workqueue(virtblk_wq);
1045 	return error;
1046 }
1047 
1048 static void __exit virtio_blk_fini(void)
1049 {
1050 	unregister_virtio_driver(&virtio_blk);
1051 	unregister_blkdev(major, "virtblk");
1052 	destroy_workqueue(virtblk_wq);
1053 }
1054 module_init(virtio_blk_init);
1055 module_exit(virtio_blk_fini);
1056 
1057 MODULE_DEVICE_TABLE(virtio, id_table);
1058 MODULE_DESCRIPTION("Virtio block driver");
1059 MODULE_LICENSE("GPL");
1060