xref: /openbmc/linux/drivers/block/virtio_blk.c (revision 81de3bf3)
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 <scsi/scsi_cmnd.h>
15 #include <linux/idr.h>
16 #include <linux/blk-mq.h>
17 #include <linux/blk-mq-virtio.h>
18 #include <linux/numa.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 	struct virtio_device *vdev;
37 
38 	/* The disk structure for the kernel. */
39 	struct gendisk *disk;
40 
41 	/* Block layer tags. */
42 	struct blk_mq_tag_set tag_set;
43 
44 	/* Process context for config space updates */
45 	struct work_struct config_work;
46 
47 	/* What host tells us, plus 2 for header & tailer. */
48 	unsigned int sg_elems;
49 
50 	/* Ida index - used to track minor number allocations. */
51 	int index;
52 
53 	/* num of vqs */
54 	int num_vqs;
55 	struct virtio_blk_vq *vqs;
56 };
57 
58 struct virtblk_req {
59 #ifdef CONFIG_VIRTIO_BLK_SCSI
60 	struct scsi_request sreq;	/* for SCSI passthrough, must be first */
61 	u8 sense[SCSI_SENSE_BUFFERSIZE];
62 	struct virtio_scsi_inhdr in_hdr;
63 #endif
64 	struct virtio_blk_outhdr out_hdr;
65 	u8 status;
66 	struct scatterlist sg[];
67 };
68 
69 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
70 {
71 	switch (vbr->status) {
72 	case VIRTIO_BLK_S_OK:
73 		return BLK_STS_OK;
74 	case VIRTIO_BLK_S_UNSUPP:
75 		return BLK_STS_NOTSUPP;
76 	default:
77 		return BLK_STS_IOERR;
78 	}
79 }
80 
81 /*
82  * If this is a packet command we need a couple of additional headers.  Behind
83  * the normal outhdr we put a segment with the scsi command block, and before
84  * the normal inhdr we put the sense data and the inhdr with additional status
85  * information.
86  */
87 #ifdef CONFIG_VIRTIO_BLK_SCSI
88 static int virtblk_add_req_scsi(struct virtqueue *vq, struct virtblk_req *vbr,
89 		struct scatterlist *data_sg, bool have_data)
90 {
91 	struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
92 	unsigned int num_out = 0, num_in = 0;
93 
94 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
95 	sgs[num_out++] = &hdr;
96 	sg_init_one(&cmd, vbr->sreq.cmd, vbr->sreq.cmd_len);
97 	sgs[num_out++] = &cmd;
98 
99 	if (have_data) {
100 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
101 			sgs[num_out++] = data_sg;
102 		else
103 			sgs[num_out + num_in++] = data_sg;
104 	}
105 
106 	sg_init_one(&sense, vbr->sense, SCSI_SENSE_BUFFERSIZE);
107 	sgs[num_out + num_in++] = &sense;
108 	sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
109 	sgs[num_out + num_in++] = &inhdr;
110 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
111 	sgs[num_out + num_in++] = &status;
112 
113 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
114 }
115 
116 static inline void virtblk_scsi_request_done(struct request *req)
117 {
118 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
119 	struct virtio_blk *vblk = req->q->queuedata;
120 	struct scsi_request *sreq = &vbr->sreq;
121 
122 	sreq->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual);
123 	sreq->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len);
124 	sreq->result = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors);
125 }
126 
127 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
128 			     unsigned int cmd, unsigned long data)
129 {
130 	struct gendisk *disk = bdev->bd_disk;
131 	struct virtio_blk *vblk = disk->private_data;
132 
133 	/*
134 	 * Only allow the generic SCSI ioctls if the host can support it.
135 	 */
136 	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
137 		return -ENOTTY;
138 
139 	return scsi_cmd_blk_ioctl(bdev, mode, cmd,
140 				  (void __user *)data);
141 }
142 #else
143 static inline int virtblk_add_req_scsi(struct virtqueue *vq,
144 		struct virtblk_req *vbr, struct scatterlist *data_sg,
145 		bool have_data)
146 {
147 	return -EIO;
148 }
149 static inline void virtblk_scsi_request_done(struct request *req)
150 {
151 }
152 #define virtblk_ioctl	NULL
153 #endif /* CONFIG_VIRTIO_BLK_SCSI */
154 
155 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
156 		struct scatterlist *data_sg, bool have_data)
157 {
158 	struct scatterlist hdr, status, *sgs[3];
159 	unsigned int num_out = 0, num_in = 0;
160 
161 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
162 	sgs[num_out++] = &hdr;
163 
164 	if (have_data) {
165 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
166 			sgs[num_out++] = data_sg;
167 		else
168 			sgs[num_out + num_in++] = data_sg;
169 	}
170 
171 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
172 	sgs[num_out + num_in++] = &status;
173 
174 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
175 }
176 
177 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
178 {
179 	unsigned short segments = blk_rq_nr_discard_segments(req);
180 	unsigned short n = 0;
181 	struct virtio_blk_discard_write_zeroes *range;
182 	struct bio *bio;
183 	u32 flags = 0;
184 
185 	if (unmap)
186 		flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
187 
188 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
189 	if (!range)
190 		return -ENOMEM;
191 
192 	__rq_for_each_bio(bio, req) {
193 		u64 sector = bio->bi_iter.bi_sector;
194 		u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
195 
196 		range[n].flags = cpu_to_le32(flags);
197 		range[n].num_sectors = cpu_to_le32(num_sectors);
198 		range[n].sector = cpu_to_le64(sector);
199 		n++;
200 	}
201 
202 	req->special_vec.bv_page = virt_to_page(range);
203 	req->special_vec.bv_offset = offset_in_page(range);
204 	req->special_vec.bv_len = sizeof(*range) * segments;
205 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
206 
207 	return 0;
208 }
209 
210 static inline void virtblk_request_done(struct request *req)
211 {
212 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
213 
214 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
215 		kfree(page_address(req->special_vec.bv_page) +
216 		      req->special_vec.bv_offset);
217 	}
218 
219 	switch (req_op(req)) {
220 	case REQ_OP_SCSI_IN:
221 	case REQ_OP_SCSI_OUT:
222 		virtblk_scsi_request_done(req);
223 		break;
224 	}
225 
226 	blk_mq_end_request(req, virtblk_result(vbr));
227 }
228 
229 static void virtblk_done(struct virtqueue *vq)
230 {
231 	struct virtio_blk *vblk = vq->vdev->priv;
232 	bool req_done = false;
233 	int qid = vq->index;
234 	struct virtblk_req *vbr;
235 	unsigned long flags;
236 	unsigned int len;
237 
238 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
239 	do {
240 		virtqueue_disable_cb(vq);
241 		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
242 			struct request *req = blk_mq_rq_from_pdu(vbr);
243 
244 			blk_mq_complete_request(req);
245 			req_done = true;
246 		}
247 		if (unlikely(virtqueue_is_broken(vq)))
248 			break;
249 	} while (!virtqueue_enable_cb(vq));
250 
251 	/* In case queue is stopped waiting for more buffers. */
252 	if (req_done)
253 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
254 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
255 }
256 
257 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
258 {
259 	struct virtio_blk *vblk = hctx->queue->queuedata;
260 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
261 	bool kick;
262 
263 	spin_lock_irq(&vq->lock);
264 	kick = virtqueue_kick_prepare(vq->vq);
265 	spin_unlock_irq(&vq->lock);
266 
267 	if (kick)
268 		virtqueue_notify(vq->vq);
269 }
270 
271 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
272 			   const struct blk_mq_queue_data *bd)
273 {
274 	struct virtio_blk *vblk = hctx->queue->queuedata;
275 	struct request *req = bd->rq;
276 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
277 	unsigned long flags;
278 	unsigned int num;
279 	int qid = hctx->queue_num;
280 	int err;
281 	bool notify = false;
282 	bool unmap = false;
283 	u32 type;
284 
285 	BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
286 
287 	switch (req_op(req)) {
288 	case REQ_OP_READ:
289 	case REQ_OP_WRITE:
290 		type = 0;
291 		break;
292 	case REQ_OP_FLUSH:
293 		type = VIRTIO_BLK_T_FLUSH;
294 		break;
295 	case REQ_OP_DISCARD:
296 		type = VIRTIO_BLK_T_DISCARD;
297 		break;
298 	case REQ_OP_WRITE_ZEROES:
299 		type = VIRTIO_BLK_T_WRITE_ZEROES;
300 		unmap = !(req->cmd_flags & REQ_NOUNMAP);
301 		break;
302 	case REQ_OP_SCSI_IN:
303 	case REQ_OP_SCSI_OUT:
304 		type = VIRTIO_BLK_T_SCSI_CMD;
305 		break;
306 	case REQ_OP_DRV_IN:
307 		type = VIRTIO_BLK_T_GET_ID;
308 		break;
309 	default:
310 		WARN_ON_ONCE(1);
311 		return BLK_STS_IOERR;
312 	}
313 
314 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
315 	vbr->out_hdr.sector = type ?
316 		0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
317 	vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
318 
319 	blk_mq_start_request(req);
320 
321 	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
322 		err = virtblk_setup_discard_write_zeroes(req, unmap);
323 		if (err)
324 			return BLK_STS_RESOURCE;
325 	}
326 
327 	num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
328 	if (num) {
329 		if (rq_data_dir(req) == WRITE)
330 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
331 		else
332 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
333 	}
334 
335 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
336 	if (blk_rq_is_scsi(req))
337 		err = virtblk_add_req_scsi(vblk->vqs[qid].vq, vbr, vbr->sg, num);
338 	else
339 		err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
340 	if (err) {
341 		virtqueue_kick(vblk->vqs[qid].vq);
342 		blk_mq_stop_hw_queue(hctx);
343 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
344 		/* Out of mem doesn't actually happen, since we fall back
345 		 * to direct descriptors */
346 		if (err == -ENOMEM || err == -ENOSPC)
347 			return BLK_STS_DEV_RESOURCE;
348 		return BLK_STS_IOERR;
349 	}
350 
351 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
352 		notify = true;
353 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
354 
355 	if (notify)
356 		virtqueue_notify(vblk->vqs[qid].vq);
357 	return BLK_STS_OK;
358 }
359 
360 /* return id (s/n) string for *disk to *id_str
361  */
362 static int virtblk_get_id(struct gendisk *disk, char *id_str)
363 {
364 	struct virtio_blk *vblk = disk->private_data;
365 	struct request_queue *q = vblk->disk->queue;
366 	struct request *req;
367 	int err;
368 
369 	req = blk_get_request(q, REQ_OP_DRV_IN, 0);
370 	if (IS_ERR(req))
371 		return PTR_ERR(req);
372 
373 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
374 	if (err)
375 		goto out;
376 
377 	blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
378 	err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
379 out:
380 	blk_put_request(req);
381 	return err;
382 }
383 
384 /* We provide getgeo only to please some old bootloader/partitioning tools */
385 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
386 {
387 	struct virtio_blk *vblk = bd->bd_disk->private_data;
388 
389 	/* see if the host passed in geometry config */
390 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
391 		virtio_cread(vblk->vdev, struct virtio_blk_config,
392 			     geometry.cylinders, &geo->cylinders);
393 		virtio_cread(vblk->vdev, struct virtio_blk_config,
394 			     geometry.heads, &geo->heads);
395 		virtio_cread(vblk->vdev, struct virtio_blk_config,
396 			     geometry.sectors, &geo->sectors);
397 	} else {
398 		/* some standard values, similar to sd */
399 		geo->heads = 1 << 6;
400 		geo->sectors = 1 << 5;
401 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
402 	}
403 	return 0;
404 }
405 
406 static const struct block_device_operations virtblk_fops = {
407 	.ioctl  = virtblk_ioctl,
408 	.owner  = THIS_MODULE,
409 	.getgeo = virtblk_getgeo,
410 };
411 
412 static int index_to_minor(int index)
413 {
414 	return index << PART_BITS;
415 }
416 
417 static int minor_to_index(int minor)
418 {
419 	return minor >> PART_BITS;
420 }
421 
422 static ssize_t serial_show(struct device *dev,
423 			   struct device_attribute *attr, char *buf)
424 {
425 	struct gendisk *disk = dev_to_disk(dev);
426 	int err;
427 
428 	/* sysfs gives us a PAGE_SIZE buffer */
429 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
430 
431 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
432 	err = virtblk_get_id(disk, buf);
433 	if (!err)
434 		return strlen(buf);
435 
436 	if (err == -EIO) /* Unsupported? Make it empty. */
437 		return 0;
438 
439 	return err;
440 }
441 
442 static DEVICE_ATTR_RO(serial);
443 
444 /* The queue's logical block size must be set before calling this */
445 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
446 {
447 	struct virtio_device *vdev = vblk->vdev;
448 	struct request_queue *q = vblk->disk->queue;
449 	char cap_str_2[10], cap_str_10[10];
450 	unsigned long long nblocks;
451 	u64 capacity;
452 
453 	/* Host must always specify the capacity. */
454 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
455 
456 	/* If capacity is too big, truncate with warning. */
457 	if ((sector_t)capacity != capacity) {
458 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
459 			 (unsigned long long)capacity);
460 		capacity = (sector_t)-1;
461 	}
462 
463 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
464 
465 	string_get_size(nblocks, queue_logical_block_size(q),
466 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
467 	string_get_size(nblocks, queue_logical_block_size(q),
468 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
469 
470 	dev_notice(&vdev->dev,
471 		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
472 		   vblk->disk->disk_name,
473 		   resize ? "new size: " : "",
474 		   nblocks,
475 		   queue_logical_block_size(q),
476 		   cap_str_10,
477 		   cap_str_2);
478 
479 	set_capacity(vblk->disk, capacity);
480 }
481 
482 static void virtblk_config_changed_work(struct work_struct *work)
483 {
484 	struct virtio_blk *vblk =
485 		container_of(work, struct virtio_blk, config_work);
486 	char *envp[] = { "RESIZE=1", NULL };
487 
488 	virtblk_update_capacity(vblk, true);
489 	revalidate_disk(vblk->disk);
490 	kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
491 }
492 
493 static void virtblk_config_changed(struct virtio_device *vdev)
494 {
495 	struct virtio_blk *vblk = vdev->priv;
496 
497 	queue_work(virtblk_wq, &vblk->config_work);
498 }
499 
500 static int init_vq(struct virtio_blk *vblk)
501 {
502 	int err;
503 	int i;
504 	vq_callback_t **callbacks;
505 	const char **names;
506 	struct virtqueue **vqs;
507 	unsigned short num_vqs;
508 	struct virtio_device *vdev = vblk->vdev;
509 	struct irq_affinity desc = { 0, };
510 
511 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
512 				   struct virtio_blk_config, num_queues,
513 				   &num_vqs);
514 	if (err)
515 		num_vqs = 1;
516 
517 	num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
518 
519 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
520 	if (!vblk->vqs)
521 		return -ENOMEM;
522 
523 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
524 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
525 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
526 	if (!names || !callbacks || !vqs) {
527 		err = -ENOMEM;
528 		goto out;
529 	}
530 
531 	for (i = 0; i < num_vqs; i++) {
532 		callbacks[i] = virtblk_done;
533 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
534 		names[i] = vblk->vqs[i].name;
535 	}
536 
537 	/* Discover virtqueues and write information to configuration.  */
538 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
539 	if (err)
540 		goto out;
541 
542 	for (i = 0; i < num_vqs; i++) {
543 		spin_lock_init(&vblk->vqs[i].lock);
544 		vblk->vqs[i].vq = vqs[i];
545 	}
546 	vblk->num_vqs = num_vqs;
547 
548 out:
549 	kfree(vqs);
550 	kfree(callbacks);
551 	kfree(names);
552 	if (err)
553 		kfree(vblk->vqs);
554 	return err;
555 }
556 
557 /*
558  * Legacy naming scheme used for virtio devices.  We are stuck with it for
559  * virtio blk but don't ever use it for any new driver.
560  */
561 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
562 {
563 	const int base = 'z' - 'a' + 1;
564 	char *begin = buf + strlen(prefix);
565 	char *end = buf + buflen;
566 	char *p;
567 	int unit;
568 
569 	p = end - 1;
570 	*p = '\0';
571 	unit = base;
572 	do {
573 		if (p == begin)
574 			return -EINVAL;
575 		*--p = 'a' + (index % unit);
576 		index = (index / unit) - 1;
577 	} while (index >= 0);
578 
579 	memmove(begin, p, end - p);
580 	memcpy(buf, prefix, strlen(prefix));
581 
582 	return 0;
583 }
584 
585 static int virtblk_get_cache_mode(struct virtio_device *vdev)
586 {
587 	u8 writeback;
588 	int err;
589 
590 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
591 				   struct virtio_blk_config, wce,
592 				   &writeback);
593 
594 	/*
595 	 * If WCE is not configurable and flush is not available,
596 	 * assume no writeback cache is in use.
597 	 */
598 	if (err)
599 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
600 
601 	return writeback;
602 }
603 
604 static void virtblk_update_cache_mode(struct virtio_device *vdev)
605 {
606 	u8 writeback = virtblk_get_cache_mode(vdev);
607 	struct virtio_blk *vblk = vdev->priv;
608 
609 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
610 	revalidate_disk(vblk->disk);
611 }
612 
613 static const char *const virtblk_cache_types[] = {
614 	"write through", "write back"
615 };
616 
617 static ssize_t
618 cache_type_store(struct device *dev, struct device_attribute *attr,
619 		 const char *buf, size_t count)
620 {
621 	struct gendisk *disk = dev_to_disk(dev);
622 	struct virtio_blk *vblk = disk->private_data;
623 	struct virtio_device *vdev = vblk->vdev;
624 	int i;
625 
626 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
627 	i = sysfs_match_string(virtblk_cache_types, buf);
628 	if (i < 0)
629 		return i;
630 
631 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
632 	virtblk_update_cache_mode(vdev);
633 	return count;
634 }
635 
636 static ssize_t
637 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
638 {
639 	struct gendisk *disk = dev_to_disk(dev);
640 	struct virtio_blk *vblk = disk->private_data;
641 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
642 
643 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
644 	return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
645 }
646 
647 static DEVICE_ATTR_RW(cache_type);
648 
649 static struct attribute *virtblk_attrs[] = {
650 	&dev_attr_serial.attr,
651 	&dev_attr_cache_type.attr,
652 	NULL,
653 };
654 
655 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
656 		struct attribute *a, int n)
657 {
658 	struct device *dev = container_of(kobj, struct device, kobj);
659 	struct gendisk *disk = dev_to_disk(dev);
660 	struct virtio_blk *vblk = disk->private_data;
661 	struct virtio_device *vdev = vblk->vdev;
662 
663 	if (a == &dev_attr_cache_type.attr &&
664 	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
665 		return S_IRUGO;
666 
667 	return a->mode;
668 }
669 
670 static const struct attribute_group virtblk_attr_group = {
671 	.attrs = virtblk_attrs,
672 	.is_visible = virtblk_attrs_are_visible,
673 };
674 
675 static const struct attribute_group *virtblk_attr_groups[] = {
676 	&virtblk_attr_group,
677 	NULL,
678 };
679 
680 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
681 		unsigned int hctx_idx, unsigned int numa_node)
682 {
683 	struct virtio_blk *vblk = set->driver_data;
684 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
685 
686 #ifdef CONFIG_VIRTIO_BLK_SCSI
687 	vbr->sreq.sense = vbr->sense;
688 #endif
689 	sg_init_table(vbr->sg, vblk->sg_elems);
690 	return 0;
691 }
692 
693 static int virtblk_map_queues(struct blk_mq_tag_set *set)
694 {
695 	struct virtio_blk *vblk = set->driver_data;
696 
697 	return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
698 					vblk->vdev, 0);
699 }
700 
701 #ifdef CONFIG_VIRTIO_BLK_SCSI
702 static void virtblk_initialize_rq(struct request *req)
703 {
704 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
705 
706 	scsi_req_init(&vbr->sreq);
707 }
708 #endif
709 
710 static const struct blk_mq_ops virtio_mq_ops = {
711 	.queue_rq	= virtio_queue_rq,
712 	.commit_rqs	= virtio_commit_rqs,
713 	.complete	= virtblk_request_done,
714 	.init_request	= virtblk_init_request,
715 #ifdef CONFIG_VIRTIO_BLK_SCSI
716 	.initialize_rq_fn = virtblk_initialize_rq,
717 #endif
718 	.map_queues	= virtblk_map_queues,
719 };
720 
721 static unsigned int virtblk_queue_depth;
722 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
723 
724 static int virtblk_probe(struct virtio_device *vdev)
725 {
726 	struct virtio_blk *vblk;
727 	struct request_queue *q;
728 	int err, index;
729 
730 	u32 v, blk_size, max_size, sg_elems, opt_io_size;
731 	u16 min_io_size;
732 	u8 physical_block_exp, alignment_offset;
733 
734 	if (!vdev->config->get) {
735 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
736 			__func__);
737 		return -EINVAL;
738 	}
739 
740 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
741 			     GFP_KERNEL);
742 	if (err < 0)
743 		goto out;
744 	index = err;
745 
746 	/* We need to know how many segments before we allocate. */
747 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
748 				   struct virtio_blk_config, seg_max,
749 				   &sg_elems);
750 
751 	/* We need at least one SG element, whatever they say. */
752 	if (err || !sg_elems)
753 		sg_elems = 1;
754 
755 	/* We need an extra sg elements at head and tail. */
756 	sg_elems += 2;
757 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
758 	if (!vblk) {
759 		err = -ENOMEM;
760 		goto out_free_index;
761 	}
762 
763 	vblk->vdev = vdev;
764 	vblk->sg_elems = sg_elems;
765 
766 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
767 
768 	err = init_vq(vblk);
769 	if (err)
770 		goto out_free_vblk;
771 
772 	/* FIXME: How many partitions?  How long is a piece of string? */
773 	vblk->disk = alloc_disk(1 << PART_BITS);
774 	if (!vblk->disk) {
775 		err = -ENOMEM;
776 		goto out_free_vq;
777 	}
778 
779 	/* Default queue sizing is to fill the ring. */
780 	if (!virtblk_queue_depth) {
781 		virtblk_queue_depth = vblk->vqs[0].vq->num_free;
782 		/* ... but without indirect descs, we use 2 descs per req */
783 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
784 			virtblk_queue_depth /= 2;
785 	}
786 
787 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
788 	vblk->tag_set.ops = &virtio_mq_ops;
789 	vblk->tag_set.queue_depth = virtblk_queue_depth;
790 	vblk->tag_set.numa_node = NUMA_NO_NODE;
791 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
792 	vblk->tag_set.cmd_size =
793 		sizeof(struct virtblk_req) +
794 		sizeof(struct scatterlist) * sg_elems;
795 	vblk->tag_set.driver_data = vblk;
796 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
797 
798 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
799 	if (err)
800 		goto out_put_disk;
801 
802 	q = blk_mq_init_queue(&vblk->tag_set);
803 	if (IS_ERR(q)) {
804 		err = -ENOMEM;
805 		goto out_free_tags;
806 	}
807 	vblk->disk->queue = q;
808 
809 	q->queuedata = vblk;
810 
811 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
812 
813 	vblk->disk->major = major;
814 	vblk->disk->first_minor = index_to_minor(index);
815 	vblk->disk->private_data = vblk;
816 	vblk->disk->fops = &virtblk_fops;
817 	vblk->disk->flags |= GENHD_FL_EXT_DEVT;
818 	vblk->index = index;
819 
820 	/* configure queue flush support */
821 	virtblk_update_cache_mode(vdev);
822 
823 	/* If disk is read-only in the host, the guest should obey */
824 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
825 		set_disk_ro(vblk->disk, 1);
826 
827 	/* We can handle whatever the host told us to handle. */
828 	blk_queue_max_segments(q, vblk->sg_elems-2);
829 
830 	/* No real sector limit. */
831 	blk_queue_max_hw_sectors(q, -1U);
832 
833 	max_size = virtio_max_dma_size(vdev);
834 
835 	/* Host can optionally specify maximum segment size and number of
836 	 * segments. */
837 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
838 				   struct virtio_blk_config, size_max, &v);
839 	if (!err)
840 		max_size = min(max_size, v);
841 
842 	blk_queue_max_segment_size(q, max_size);
843 
844 	/* Host can optionally specify the block size of the device */
845 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
846 				   struct virtio_blk_config, blk_size,
847 				   &blk_size);
848 	if (!err)
849 		blk_queue_logical_block_size(q, blk_size);
850 	else
851 		blk_size = queue_logical_block_size(q);
852 
853 	/* Use topology information if available */
854 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
855 				   struct virtio_blk_config, physical_block_exp,
856 				   &physical_block_exp);
857 	if (!err && physical_block_exp)
858 		blk_queue_physical_block_size(q,
859 				blk_size * (1 << physical_block_exp));
860 
861 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
862 				   struct virtio_blk_config, alignment_offset,
863 				   &alignment_offset);
864 	if (!err && alignment_offset)
865 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
866 
867 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
868 				   struct virtio_blk_config, min_io_size,
869 				   &min_io_size);
870 	if (!err && min_io_size)
871 		blk_queue_io_min(q, blk_size * min_io_size);
872 
873 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
874 				   struct virtio_blk_config, opt_io_size,
875 				   &opt_io_size);
876 	if (!err && opt_io_size)
877 		blk_queue_io_opt(q, blk_size * opt_io_size);
878 
879 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
880 		q->limits.discard_granularity = blk_size;
881 
882 		virtio_cread(vdev, struct virtio_blk_config,
883 			     discard_sector_alignment, &v);
884 		q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
885 
886 		virtio_cread(vdev, struct virtio_blk_config,
887 			     max_discard_sectors, &v);
888 		blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
889 
890 		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
891 			     &v);
892 		blk_queue_max_discard_segments(q,
893 					       min_not_zero(v,
894 							    MAX_DISCARD_SEGMENTS));
895 
896 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
897 	}
898 
899 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
900 		virtio_cread(vdev, struct virtio_blk_config,
901 			     max_write_zeroes_sectors, &v);
902 		blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
903 	}
904 
905 	virtblk_update_capacity(vblk, false);
906 	virtio_device_ready(vdev);
907 
908 	device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
909 	return 0;
910 
911 out_free_tags:
912 	blk_mq_free_tag_set(&vblk->tag_set);
913 out_put_disk:
914 	put_disk(vblk->disk);
915 out_free_vq:
916 	vdev->config->del_vqs(vdev);
917 out_free_vblk:
918 	kfree(vblk);
919 out_free_index:
920 	ida_simple_remove(&vd_index_ida, index);
921 out:
922 	return err;
923 }
924 
925 static void virtblk_remove(struct virtio_device *vdev)
926 {
927 	struct virtio_blk *vblk = vdev->priv;
928 	int index = vblk->index;
929 	int refc;
930 
931 	/* Make sure no work handler is accessing the device. */
932 	flush_work(&vblk->config_work);
933 
934 	del_gendisk(vblk->disk);
935 	blk_cleanup_queue(vblk->disk->queue);
936 
937 	blk_mq_free_tag_set(&vblk->tag_set);
938 
939 	/* Stop all the virtqueues. */
940 	vdev->config->reset(vdev);
941 
942 	refc = kref_read(&disk_to_dev(vblk->disk)->kobj.kref);
943 	put_disk(vblk->disk);
944 	vdev->config->del_vqs(vdev);
945 	kfree(vblk->vqs);
946 	kfree(vblk);
947 
948 	/* Only free device id if we don't have any users */
949 	if (refc == 1)
950 		ida_simple_remove(&vd_index_ida, index);
951 }
952 
953 #ifdef CONFIG_PM_SLEEP
954 static int virtblk_freeze(struct virtio_device *vdev)
955 {
956 	struct virtio_blk *vblk = vdev->priv;
957 
958 	/* Ensure we don't receive any more interrupts */
959 	vdev->config->reset(vdev);
960 
961 	/* Make sure no work handler is accessing the device. */
962 	flush_work(&vblk->config_work);
963 
964 	blk_mq_quiesce_queue(vblk->disk->queue);
965 
966 	vdev->config->del_vqs(vdev);
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 #ifdef CONFIG_VIRTIO_BLK_SCSI
995 	VIRTIO_BLK_F_SCSI,
996 #endif
997 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
998 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
999 }
1000 ;
1001 static unsigned int features[] = {
1002 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1003 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1004 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1005 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1006 };
1007 
1008 static struct virtio_driver virtio_blk = {
1009 	.feature_table			= features,
1010 	.feature_table_size		= ARRAY_SIZE(features),
1011 	.feature_table_legacy		= features_legacy,
1012 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
1013 	.driver.name			= KBUILD_MODNAME,
1014 	.driver.owner			= THIS_MODULE,
1015 	.id_table			= id_table,
1016 	.probe				= virtblk_probe,
1017 	.remove				= virtblk_remove,
1018 	.config_changed			= virtblk_config_changed,
1019 #ifdef CONFIG_PM_SLEEP
1020 	.freeze				= virtblk_freeze,
1021 	.restore			= virtblk_restore,
1022 #endif
1023 };
1024 
1025 static int __init init(void)
1026 {
1027 	int error;
1028 
1029 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1030 	if (!virtblk_wq)
1031 		return -ENOMEM;
1032 
1033 	major = register_blkdev(0, "virtblk");
1034 	if (major < 0) {
1035 		error = major;
1036 		goto out_destroy_workqueue;
1037 	}
1038 
1039 	error = register_virtio_driver(&virtio_blk);
1040 	if (error)
1041 		goto out_unregister_blkdev;
1042 	return 0;
1043 
1044 out_unregister_blkdev:
1045 	unregister_blkdev(major, "virtblk");
1046 out_destroy_workqueue:
1047 	destroy_workqueue(virtblk_wq);
1048 	return error;
1049 }
1050 
1051 static void __exit fini(void)
1052 {
1053 	unregister_virtio_driver(&virtio_blk);
1054 	unregister_blkdev(major, "virtblk");
1055 	destroy_workqueue(virtblk_wq);
1056 }
1057 module_init(init);
1058 module_exit(fini);
1059 
1060 MODULE_DEVICE_TABLE(virtio, id_table);
1061 MODULE_DESCRIPTION("Virtio block driver");
1062 MODULE_LICENSE("GPL");
1063