xref: /openbmc/linux/drivers/block/virtio_blk.c (revision 3213486f)
1 //#define DEBUG
2 #include <linux/spinlock.h>
3 #include <linux/slab.h>
4 #include <linux/blkdev.h>
5 #include <linux/hdreg.h>
6 #include <linux/module.h>
7 #include <linux/mutex.h>
8 #include <linux/interrupt.h>
9 #include <linux/virtio.h>
10 #include <linux/virtio_blk.h>
11 #include <linux/scatterlist.h>
12 #include <linux/string_helpers.h>
13 #include <scsi/scsi_cmnd.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 
19 #define PART_BITS 4
20 #define VQ_NAME_LEN 16
21 #define MAX_DISCARD_SEGMENTS 256u
22 
23 static int major;
24 static DEFINE_IDA(vd_index_ida);
25 
26 static struct workqueue_struct *virtblk_wq;
27 
28 struct virtio_blk_vq {
29 	struct virtqueue *vq;
30 	spinlock_t lock;
31 	char name[VQ_NAME_LEN];
32 } ____cacheline_aligned_in_smp;
33 
34 struct virtio_blk {
35 	struct virtio_device *vdev;
36 
37 	/* The disk structure for the kernel. */
38 	struct gendisk *disk;
39 
40 	/* Block layer tags. */
41 	struct blk_mq_tag_set tag_set;
42 
43 	/* Process context for config space updates */
44 	struct work_struct config_work;
45 
46 	/* What host tells us, plus 2 for header & tailer. */
47 	unsigned int sg_elems;
48 
49 	/* Ida index - used to track minor number allocations. */
50 	int index;
51 
52 	/* num of vqs */
53 	int num_vqs;
54 	struct virtio_blk_vq *vqs;
55 };
56 
57 struct virtblk_req {
58 #ifdef CONFIG_VIRTIO_BLK_SCSI
59 	struct scsi_request sreq;	/* for SCSI passthrough, must be first */
60 	u8 sense[SCSI_SENSE_BUFFERSIZE];
61 	struct virtio_scsi_inhdr in_hdr;
62 #endif
63 	struct virtio_blk_outhdr out_hdr;
64 	u8 status;
65 	struct scatterlist sg[];
66 };
67 
68 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
69 {
70 	switch (vbr->status) {
71 	case VIRTIO_BLK_S_OK:
72 		return BLK_STS_OK;
73 	case VIRTIO_BLK_S_UNSUPP:
74 		return BLK_STS_NOTSUPP;
75 	default:
76 		return BLK_STS_IOERR;
77 	}
78 }
79 
80 /*
81  * If this is a packet command we need a couple of additional headers.  Behind
82  * the normal outhdr we put a segment with the scsi command block, and before
83  * the normal inhdr we put the sense data and the inhdr with additional status
84  * information.
85  */
86 #ifdef CONFIG_VIRTIO_BLK_SCSI
87 static int virtblk_add_req_scsi(struct virtqueue *vq, struct virtblk_req *vbr,
88 		struct scatterlist *data_sg, bool have_data)
89 {
90 	struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
91 	unsigned int num_out = 0, num_in = 0;
92 
93 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
94 	sgs[num_out++] = &hdr;
95 	sg_init_one(&cmd, vbr->sreq.cmd, vbr->sreq.cmd_len);
96 	sgs[num_out++] = &cmd;
97 
98 	if (have_data) {
99 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
100 			sgs[num_out++] = data_sg;
101 		else
102 			sgs[num_out + num_in++] = data_sg;
103 	}
104 
105 	sg_init_one(&sense, vbr->sense, SCSI_SENSE_BUFFERSIZE);
106 	sgs[num_out + num_in++] = &sense;
107 	sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
108 	sgs[num_out + num_in++] = &inhdr;
109 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
110 	sgs[num_out + num_in++] = &status;
111 
112 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
113 }
114 
115 static inline void virtblk_scsi_request_done(struct request *req)
116 {
117 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
118 	struct virtio_blk *vblk = req->q->queuedata;
119 	struct scsi_request *sreq = &vbr->sreq;
120 
121 	sreq->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual);
122 	sreq->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len);
123 	sreq->result = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors);
124 }
125 
126 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
127 			     unsigned int cmd, unsigned long data)
128 {
129 	struct gendisk *disk = bdev->bd_disk;
130 	struct virtio_blk *vblk = disk->private_data;
131 
132 	/*
133 	 * Only allow the generic SCSI ioctls if the host can support it.
134 	 */
135 	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
136 		return -ENOTTY;
137 
138 	return scsi_cmd_blk_ioctl(bdev, mode, cmd,
139 				  (void __user *)data);
140 }
141 #else
142 static inline int virtblk_add_req_scsi(struct virtqueue *vq,
143 		struct virtblk_req *vbr, struct scatterlist *data_sg,
144 		bool have_data)
145 {
146 	return -EIO;
147 }
148 static inline void virtblk_scsi_request_done(struct request *req)
149 {
150 }
151 #define virtblk_ioctl	NULL
152 #endif /* CONFIG_VIRTIO_BLK_SCSI */
153 
154 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
155 		struct scatterlist *data_sg, bool have_data)
156 {
157 	struct scatterlist hdr, status, *sgs[3];
158 	unsigned int num_out = 0, num_in = 0;
159 
160 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
161 	sgs[num_out++] = &hdr;
162 
163 	if (have_data) {
164 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
165 			sgs[num_out++] = data_sg;
166 		else
167 			sgs[num_out + num_in++] = data_sg;
168 	}
169 
170 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
171 	sgs[num_out + num_in++] = &status;
172 
173 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
174 }
175 
176 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
177 {
178 	unsigned short segments = blk_rq_nr_discard_segments(req);
179 	unsigned short n = 0;
180 	struct virtio_blk_discard_write_zeroes *range;
181 	struct bio *bio;
182 	u32 flags = 0;
183 
184 	if (unmap)
185 		flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
186 
187 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
188 	if (!range)
189 		return -ENOMEM;
190 
191 	__rq_for_each_bio(bio, req) {
192 		u64 sector = bio->bi_iter.bi_sector;
193 		u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
194 
195 		range[n].flags = cpu_to_le32(flags);
196 		range[n].num_sectors = cpu_to_le32(num_sectors);
197 		range[n].sector = cpu_to_le64(sector);
198 		n++;
199 	}
200 
201 	req->special_vec.bv_page = virt_to_page(range);
202 	req->special_vec.bv_offset = offset_in_page(range);
203 	req->special_vec.bv_len = sizeof(*range) * segments;
204 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
205 
206 	return 0;
207 }
208 
209 static inline void virtblk_request_done(struct request *req)
210 {
211 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
212 
213 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
214 		kfree(page_address(req->special_vec.bv_page) +
215 		      req->special_vec.bv_offset);
216 	}
217 
218 	switch (req_op(req)) {
219 	case REQ_OP_SCSI_IN:
220 	case REQ_OP_SCSI_OUT:
221 		virtblk_scsi_request_done(req);
222 		break;
223 	}
224 
225 	blk_mq_end_request(req, virtblk_result(vbr));
226 }
227 
228 static void virtblk_done(struct virtqueue *vq)
229 {
230 	struct virtio_blk *vblk = vq->vdev->priv;
231 	bool req_done = false;
232 	int qid = vq->index;
233 	struct virtblk_req *vbr;
234 	unsigned long flags;
235 	unsigned int len;
236 
237 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
238 	do {
239 		virtqueue_disable_cb(vq);
240 		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
241 			struct request *req = blk_mq_rq_from_pdu(vbr);
242 
243 			blk_mq_complete_request(req);
244 			req_done = true;
245 		}
246 		if (unlikely(virtqueue_is_broken(vq)))
247 			break;
248 	} while (!virtqueue_enable_cb(vq));
249 
250 	/* In case queue is stopped waiting for more buffers. */
251 	if (req_done)
252 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
253 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
254 }
255 
256 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
257 {
258 	struct virtio_blk *vblk = hctx->queue->queuedata;
259 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
260 	bool kick;
261 
262 	spin_lock_irq(&vq->lock);
263 	kick = virtqueue_kick_prepare(vq->vq);
264 	spin_unlock_irq(&vq->lock);
265 
266 	if (kick)
267 		virtqueue_notify(vq->vq);
268 }
269 
270 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
271 			   const struct blk_mq_queue_data *bd)
272 {
273 	struct virtio_blk *vblk = hctx->queue->queuedata;
274 	struct request *req = bd->rq;
275 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
276 	unsigned long flags;
277 	unsigned int num;
278 	int qid = hctx->queue_num;
279 	int err;
280 	bool notify = false;
281 	bool unmap = false;
282 	u32 type;
283 
284 	BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
285 
286 	switch (req_op(req)) {
287 	case REQ_OP_READ:
288 	case REQ_OP_WRITE:
289 		type = 0;
290 		break;
291 	case REQ_OP_FLUSH:
292 		type = VIRTIO_BLK_T_FLUSH;
293 		break;
294 	case REQ_OP_DISCARD:
295 		type = VIRTIO_BLK_T_DISCARD;
296 		break;
297 	case REQ_OP_WRITE_ZEROES:
298 		type = VIRTIO_BLK_T_WRITE_ZEROES;
299 		unmap = !(req->cmd_flags & REQ_NOUNMAP);
300 		break;
301 	case REQ_OP_SCSI_IN:
302 	case REQ_OP_SCSI_OUT:
303 		type = VIRTIO_BLK_T_SCSI_CMD;
304 		break;
305 	case REQ_OP_DRV_IN:
306 		type = VIRTIO_BLK_T_GET_ID;
307 		break;
308 	default:
309 		WARN_ON_ONCE(1);
310 		return BLK_STS_IOERR;
311 	}
312 
313 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
314 	vbr->out_hdr.sector = type ?
315 		0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
316 	vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
317 
318 	blk_mq_start_request(req);
319 
320 	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
321 		err = virtblk_setup_discard_write_zeroes(req, unmap);
322 		if (err)
323 			return BLK_STS_RESOURCE;
324 	}
325 
326 	num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
327 	if (num) {
328 		if (rq_data_dir(req) == WRITE)
329 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
330 		else
331 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
332 	}
333 
334 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
335 	if (blk_rq_is_scsi(req))
336 		err = virtblk_add_req_scsi(vblk->vqs[qid].vq, vbr, vbr->sg, num);
337 	else
338 		err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
339 	if (err) {
340 		virtqueue_kick(vblk->vqs[qid].vq);
341 		blk_mq_stop_hw_queue(hctx);
342 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
343 		/* Out of mem doesn't actually happen, since we fall back
344 		 * to direct descriptors */
345 		if (err == -ENOMEM || err == -ENOSPC)
346 			return BLK_STS_DEV_RESOURCE;
347 		return BLK_STS_IOERR;
348 	}
349 
350 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
351 		notify = true;
352 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
353 
354 	if (notify)
355 		virtqueue_notify(vblk->vqs[qid].vq);
356 	return BLK_STS_OK;
357 }
358 
359 /* return id (s/n) string for *disk to *id_str
360  */
361 static int virtblk_get_id(struct gendisk *disk, char *id_str)
362 {
363 	struct virtio_blk *vblk = disk->private_data;
364 	struct request_queue *q = vblk->disk->queue;
365 	struct request *req;
366 	int err;
367 
368 	req = blk_get_request(q, REQ_OP_DRV_IN, 0);
369 	if (IS_ERR(req))
370 		return PTR_ERR(req);
371 
372 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
373 	if (err)
374 		goto out;
375 
376 	blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
377 	err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
378 out:
379 	blk_put_request(req);
380 	return err;
381 }
382 
383 /* We provide getgeo only to please some old bootloader/partitioning tools */
384 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
385 {
386 	struct virtio_blk *vblk = bd->bd_disk->private_data;
387 
388 	/* see if the host passed in geometry config */
389 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
390 		virtio_cread(vblk->vdev, struct virtio_blk_config,
391 			     geometry.cylinders, &geo->cylinders);
392 		virtio_cread(vblk->vdev, struct virtio_blk_config,
393 			     geometry.heads, &geo->heads);
394 		virtio_cread(vblk->vdev, struct virtio_blk_config,
395 			     geometry.sectors, &geo->sectors);
396 	} else {
397 		/* some standard values, similar to sd */
398 		geo->heads = 1 << 6;
399 		geo->sectors = 1 << 5;
400 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
401 	}
402 	return 0;
403 }
404 
405 static const struct block_device_operations virtblk_fops = {
406 	.ioctl  = virtblk_ioctl,
407 	.owner  = THIS_MODULE,
408 	.getgeo = virtblk_getgeo,
409 };
410 
411 static int index_to_minor(int index)
412 {
413 	return index << PART_BITS;
414 }
415 
416 static int minor_to_index(int minor)
417 {
418 	return minor >> PART_BITS;
419 }
420 
421 static ssize_t serial_show(struct device *dev,
422 			   struct device_attribute *attr, char *buf)
423 {
424 	struct gendisk *disk = dev_to_disk(dev);
425 	int err;
426 
427 	/* sysfs gives us a PAGE_SIZE buffer */
428 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
429 
430 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
431 	err = virtblk_get_id(disk, buf);
432 	if (!err)
433 		return strlen(buf);
434 
435 	if (err == -EIO) /* Unsupported? Make it empty. */
436 		return 0;
437 
438 	return err;
439 }
440 
441 static DEVICE_ATTR_RO(serial);
442 
443 /* The queue's logical block size must be set before calling this */
444 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
445 {
446 	struct virtio_device *vdev = vblk->vdev;
447 	struct request_queue *q = vblk->disk->queue;
448 	char cap_str_2[10], cap_str_10[10];
449 	unsigned long long nblocks;
450 	u64 capacity;
451 
452 	/* Host must always specify the capacity. */
453 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
454 
455 	/* If capacity is too big, truncate with warning. */
456 	if ((sector_t)capacity != capacity) {
457 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
458 			 (unsigned long long)capacity);
459 		capacity = (sector_t)-1;
460 	}
461 
462 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
463 
464 	string_get_size(nblocks, queue_logical_block_size(q),
465 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
466 	string_get_size(nblocks, queue_logical_block_size(q),
467 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
468 
469 	dev_notice(&vdev->dev,
470 		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
471 		   vblk->disk->disk_name,
472 		   resize ? "new size: " : "",
473 		   nblocks,
474 		   queue_logical_block_size(q),
475 		   cap_str_10,
476 		   cap_str_2);
477 
478 	set_capacity(vblk->disk, capacity);
479 }
480 
481 static void virtblk_config_changed_work(struct work_struct *work)
482 {
483 	struct virtio_blk *vblk =
484 		container_of(work, struct virtio_blk, config_work);
485 	char *envp[] = { "RESIZE=1", NULL };
486 
487 	virtblk_update_capacity(vblk, true);
488 	revalidate_disk(vblk->disk);
489 	kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
490 }
491 
492 static void virtblk_config_changed(struct virtio_device *vdev)
493 {
494 	struct virtio_blk *vblk = vdev->priv;
495 
496 	queue_work(virtblk_wq, &vblk->config_work);
497 }
498 
499 static int init_vq(struct virtio_blk *vblk)
500 {
501 	int err;
502 	int i;
503 	vq_callback_t **callbacks;
504 	const char **names;
505 	struct virtqueue **vqs;
506 	unsigned short num_vqs;
507 	struct virtio_device *vdev = vblk->vdev;
508 	struct irq_affinity desc = { 0, };
509 
510 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
511 				   struct virtio_blk_config, num_queues,
512 				   &num_vqs);
513 	if (err)
514 		num_vqs = 1;
515 
516 	num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
517 
518 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
519 	if (!vblk->vqs)
520 		return -ENOMEM;
521 
522 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
523 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
524 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
525 	if (!names || !callbacks || !vqs) {
526 		err = -ENOMEM;
527 		goto out;
528 	}
529 
530 	for (i = 0; i < num_vqs; i++) {
531 		callbacks[i] = virtblk_done;
532 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
533 		names[i] = vblk->vqs[i].name;
534 	}
535 
536 	/* Discover virtqueues and write information to configuration.  */
537 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
538 	if (err)
539 		goto out;
540 
541 	for (i = 0; i < num_vqs; i++) {
542 		spin_lock_init(&vblk->vqs[i].lock);
543 		vblk->vqs[i].vq = vqs[i];
544 	}
545 	vblk->num_vqs = num_vqs;
546 
547 out:
548 	kfree(vqs);
549 	kfree(callbacks);
550 	kfree(names);
551 	if (err)
552 		kfree(vblk->vqs);
553 	return err;
554 }
555 
556 /*
557  * Legacy naming scheme used for virtio devices.  We are stuck with it for
558  * virtio blk but don't ever use it for any new driver.
559  */
560 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
561 {
562 	const int base = 'z' - 'a' + 1;
563 	char *begin = buf + strlen(prefix);
564 	char *end = buf + buflen;
565 	char *p;
566 	int unit;
567 
568 	p = end - 1;
569 	*p = '\0';
570 	unit = base;
571 	do {
572 		if (p == begin)
573 			return -EINVAL;
574 		*--p = 'a' + (index % unit);
575 		index = (index / unit) - 1;
576 	} while (index >= 0);
577 
578 	memmove(begin, p, end - p);
579 	memcpy(buf, prefix, strlen(prefix));
580 
581 	return 0;
582 }
583 
584 static int virtblk_get_cache_mode(struct virtio_device *vdev)
585 {
586 	u8 writeback;
587 	int err;
588 
589 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
590 				   struct virtio_blk_config, wce,
591 				   &writeback);
592 
593 	/*
594 	 * If WCE is not configurable and flush is not available,
595 	 * assume no writeback cache is in use.
596 	 */
597 	if (err)
598 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
599 
600 	return writeback;
601 }
602 
603 static void virtblk_update_cache_mode(struct virtio_device *vdev)
604 {
605 	u8 writeback = virtblk_get_cache_mode(vdev);
606 	struct virtio_blk *vblk = vdev->priv;
607 
608 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
609 	revalidate_disk(vblk->disk);
610 }
611 
612 static const char *const virtblk_cache_types[] = {
613 	"write through", "write back"
614 };
615 
616 static ssize_t
617 cache_type_store(struct device *dev, struct device_attribute *attr,
618 		 const char *buf, size_t count)
619 {
620 	struct gendisk *disk = dev_to_disk(dev);
621 	struct virtio_blk *vblk = disk->private_data;
622 	struct virtio_device *vdev = vblk->vdev;
623 	int i;
624 
625 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
626 	i = sysfs_match_string(virtblk_cache_types, buf);
627 	if (i < 0)
628 		return i;
629 
630 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
631 	virtblk_update_cache_mode(vdev);
632 	return count;
633 }
634 
635 static ssize_t
636 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
637 {
638 	struct gendisk *disk = dev_to_disk(dev);
639 	struct virtio_blk *vblk = disk->private_data;
640 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
641 
642 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
643 	return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
644 }
645 
646 static DEVICE_ATTR_RW(cache_type);
647 
648 static struct attribute *virtblk_attrs[] = {
649 	&dev_attr_serial.attr,
650 	&dev_attr_cache_type.attr,
651 	NULL,
652 };
653 
654 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
655 		struct attribute *a, int n)
656 {
657 	struct device *dev = container_of(kobj, struct device, kobj);
658 	struct gendisk *disk = dev_to_disk(dev);
659 	struct virtio_blk *vblk = disk->private_data;
660 	struct virtio_device *vdev = vblk->vdev;
661 
662 	if (a == &dev_attr_cache_type.attr &&
663 	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
664 		return S_IRUGO;
665 
666 	return a->mode;
667 }
668 
669 static const struct attribute_group virtblk_attr_group = {
670 	.attrs = virtblk_attrs,
671 	.is_visible = virtblk_attrs_are_visible,
672 };
673 
674 static const struct attribute_group *virtblk_attr_groups[] = {
675 	&virtblk_attr_group,
676 	NULL,
677 };
678 
679 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
680 		unsigned int hctx_idx, unsigned int numa_node)
681 {
682 	struct virtio_blk *vblk = set->driver_data;
683 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
684 
685 #ifdef CONFIG_VIRTIO_BLK_SCSI
686 	vbr->sreq.sense = vbr->sense;
687 #endif
688 	sg_init_table(vbr->sg, vblk->sg_elems);
689 	return 0;
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[0], vblk->vdev, 0);
697 }
698 
699 #ifdef CONFIG_VIRTIO_BLK_SCSI
700 static void virtblk_initialize_rq(struct request *req)
701 {
702 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
703 
704 	scsi_req_init(&vbr->sreq);
705 }
706 #endif
707 
708 static const struct blk_mq_ops virtio_mq_ops = {
709 	.queue_rq	= virtio_queue_rq,
710 	.commit_rqs	= virtio_commit_rqs,
711 	.complete	= virtblk_request_done,
712 	.init_request	= virtblk_init_request,
713 #ifdef CONFIG_VIRTIO_BLK_SCSI
714 	.initialize_rq_fn = virtblk_initialize_rq,
715 #endif
716 	.map_queues	= virtblk_map_queues,
717 };
718 
719 static unsigned int virtblk_queue_depth;
720 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
721 
722 static int virtblk_probe(struct virtio_device *vdev)
723 {
724 	struct virtio_blk *vblk;
725 	struct request_queue *q;
726 	int err, index;
727 
728 	u32 v, blk_size, max_size, sg_elems, opt_io_size;
729 	u16 min_io_size;
730 	u8 physical_block_exp, alignment_offset;
731 
732 	if (!vdev->config->get) {
733 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
734 			__func__);
735 		return -EINVAL;
736 	}
737 
738 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
739 			     GFP_KERNEL);
740 	if (err < 0)
741 		goto out;
742 	index = err;
743 
744 	/* We need to know how many segments before we allocate. */
745 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
746 				   struct virtio_blk_config, seg_max,
747 				   &sg_elems);
748 
749 	/* We need at least one SG element, whatever they say. */
750 	if (err || !sg_elems)
751 		sg_elems = 1;
752 
753 	/* We need an extra sg elements at head and tail. */
754 	sg_elems += 2;
755 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
756 	if (!vblk) {
757 		err = -ENOMEM;
758 		goto out_free_index;
759 	}
760 
761 	vblk->vdev = vdev;
762 	vblk->sg_elems = sg_elems;
763 
764 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
765 
766 	err = init_vq(vblk);
767 	if (err)
768 		goto out_free_vblk;
769 
770 	/* FIXME: How many partitions?  How long is a piece of string? */
771 	vblk->disk = alloc_disk(1 << PART_BITS);
772 	if (!vblk->disk) {
773 		err = -ENOMEM;
774 		goto out_free_vq;
775 	}
776 
777 	/* Default queue sizing is to fill the ring. */
778 	if (!virtblk_queue_depth) {
779 		virtblk_queue_depth = vblk->vqs[0].vq->num_free;
780 		/* ... but without indirect descs, we use 2 descs per req */
781 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
782 			virtblk_queue_depth /= 2;
783 	}
784 
785 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
786 	vblk->tag_set.ops = &virtio_mq_ops;
787 	vblk->tag_set.queue_depth = virtblk_queue_depth;
788 	vblk->tag_set.numa_node = NUMA_NO_NODE;
789 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
790 	vblk->tag_set.cmd_size =
791 		sizeof(struct virtblk_req) +
792 		sizeof(struct scatterlist) * sg_elems;
793 	vblk->tag_set.driver_data = vblk;
794 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
795 
796 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
797 	if (err)
798 		goto out_put_disk;
799 
800 	q = blk_mq_init_queue(&vblk->tag_set);
801 	if (IS_ERR(q)) {
802 		err = -ENOMEM;
803 		goto out_free_tags;
804 	}
805 	vblk->disk->queue = q;
806 
807 	q->queuedata = vblk;
808 
809 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
810 
811 	vblk->disk->major = major;
812 	vblk->disk->first_minor = index_to_minor(index);
813 	vblk->disk->private_data = vblk;
814 	vblk->disk->fops = &virtblk_fops;
815 	vblk->disk->flags |= GENHD_FL_EXT_DEVT;
816 	vblk->index = index;
817 
818 	/* configure queue flush support */
819 	virtblk_update_cache_mode(vdev);
820 
821 	/* If disk is read-only in the host, the guest should obey */
822 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
823 		set_disk_ro(vblk->disk, 1);
824 
825 	/* We can handle whatever the host told us to handle. */
826 	blk_queue_max_segments(q, vblk->sg_elems-2);
827 
828 	/* No real sector limit. */
829 	blk_queue_max_hw_sectors(q, -1U);
830 
831 	max_size = virtio_max_dma_size(vdev);
832 
833 	/* Host can optionally specify maximum segment size and number of
834 	 * segments. */
835 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
836 				   struct virtio_blk_config, size_max, &v);
837 	if (!err)
838 		max_size = min(max_size, v);
839 
840 	blk_queue_max_segment_size(q, max_size);
841 
842 	/* Host can optionally specify the block size of the device */
843 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
844 				   struct virtio_blk_config, blk_size,
845 				   &blk_size);
846 	if (!err)
847 		blk_queue_logical_block_size(q, blk_size);
848 	else
849 		blk_size = queue_logical_block_size(q);
850 
851 	/* Use topology information if available */
852 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
853 				   struct virtio_blk_config, physical_block_exp,
854 				   &physical_block_exp);
855 	if (!err && physical_block_exp)
856 		blk_queue_physical_block_size(q,
857 				blk_size * (1 << physical_block_exp));
858 
859 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
860 				   struct virtio_blk_config, alignment_offset,
861 				   &alignment_offset);
862 	if (!err && alignment_offset)
863 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
864 
865 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
866 				   struct virtio_blk_config, min_io_size,
867 				   &min_io_size);
868 	if (!err && min_io_size)
869 		blk_queue_io_min(q, blk_size * min_io_size);
870 
871 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
872 				   struct virtio_blk_config, opt_io_size,
873 				   &opt_io_size);
874 	if (!err && opt_io_size)
875 		blk_queue_io_opt(q, blk_size * opt_io_size);
876 
877 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
878 		q->limits.discard_granularity = blk_size;
879 
880 		virtio_cread(vdev, struct virtio_blk_config,
881 			     discard_sector_alignment, &v);
882 		q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
883 
884 		virtio_cread(vdev, struct virtio_blk_config,
885 			     max_discard_sectors, &v);
886 		blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
887 
888 		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
889 			     &v);
890 		blk_queue_max_discard_segments(q,
891 					       min_not_zero(v,
892 							    MAX_DISCARD_SEGMENTS));
893 
894 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
895 	}
896 
897 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
898 		virtio_cread(vdev, struct virtio_blk_config,
899 			     max_write_zeroes_sectors, &v);
900 		blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
901 	}
902 
903 	virtblk_update_capacity(vblk, false);
904 	virtio_device_ready(vdev);
905 
906 	device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
907 	return 0;
908 
909 out_free_tags:
910 	blk_mq_free_tag_set(&vblk->tag_set);
911 out_put_disk:
912 	put_disk(vblk->disk);
913 out_free_vq:
914 	vdev->config->del_vqs(vdev);
915 out_free_vblk:
916 	kfree(vblk);
917 out_free_index:
918 	ida_simple_remove(&vd_index_ida, index);
919 out:
920 	return err;
921 }
922 
923 static void virtblk_remove(struct virtio_device *vdev)
924 {
925 	struct virtio_blk *vblk = vdev->priv;
926 	int index = vblk->index;
927 	int refc;
928 
929 	/* Make sure no work handler is accessing the device. */
930 	flush_work(&vblk->config_work);
931 
932 	del_gendisk(vblk->disk);
933 	blk_cleanup_queue(vblk->disk->queue);
934 
935 	blk_mq_free_tag_set(&vblk->tag_set);
936 
937 	/* Stop all the virtqueues. */
938 	vdev->config->reset(vdev);
939 
940 	refc = kref_read(&disk_to_dev(vblk->disk)->kobj.kref);
941 	put_disk(vblk->disk);
942 	vdev->config->del_vqs(vdev);
943 	kfree(vblk->vqs);
944 	kfree(vblk);
945 
946 	/* Only free device id if we don't have any users */
947 	if (refc == 1)
948 		ida_simple_remove(&vd_index_ida, index);
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 	vdev->config->reset(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 	return 0;
966 }
967 
968 static int virtblk_restore(struct virtio_device *vdev)
969 {
970 	struct virtio_blk *vblk = vdev->priv;
971 	int ret;
972 
973 	ret = init_vq(vdev->priv);
974 	if (ret)
975 		return ret;
976 
977 	virtio_device_ready(vdev);
978 
979 	blk_mq_unquiesce_queue(vblk->disk->queue);
980 	return 0;
981 }
982 #endif
983 
984 static const struct virtio_device_id id_table[] = {
985 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
986 	{ 0 },
987 };
988 
989 static unsigned int features_legacy[] = {
990 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
991 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
992 #ifdef CONFIG_VIRTIO_BLK_SCSI
993 	VIRTIO_BLK_F_SCSI,
994 #endif
995 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
996 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
997 }
998 ;
999 static unsigned int features[] = {
1000 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1001 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1002 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1003 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1004 };
1005 
1006 static struct virtio_driver virtio_blk = {
1007 	.feature_table			= features,
1008 	.feature_table_size		= ARRAY_SIZE(features),
1009 	.feature_table_legacy		= features_legacy,
1010 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
1011 	.driver.name			= KBUILD_MODNAME,
1012 	.driver.owner			= THIS_MODULE,
1013 	.id_table			= id_table,
1014 	.probe				= virtblk_probe,
1015 	.remove				= virtblk_remove,
1016 	.config_changed			= virtblk_config_changed,
1017 #ifdef CONFIG_PM_SLEEP
1018 	.freeze				= virtblk_freeze,
1019 	.restore			= virtblk_restore,
1020 #endif
1021 };
1022 
1023 static int __init init(void)
1024 {
1025 	int error;
1026 
1027 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1028 	if (!virtblk_wq)
1029 		return -ENOMEM;
1030 
1031 	major = register_blkdev(0, "virtblk");
1032 	if (major < 0) {
1033 		error = major;
1034 		goto out_destroy_workqueue;
1035 	}
1036 
1037 	error = register_virtio_driver(&virtio_blk);
1038 	if (error)
1039 		goto out_unregister_blkdev;
1040 	return 0;
1041 
1042 out_unregister_blkdev:
1043 	unregister_blkdev(major, "virtblk");
1044 out_destroy_workqueue:
1045 	destroy_workqueue(virtblk_wq);
1046 	return error;
1047 }
1048 
1049 static void __exit fini(void)
1050 {
1051 	unregister_virtio_driver(&virtio_blk);
1052 	unregister_blkdev(major, "virtblk");
1053 	destroy_workqueue(virtblk_wq);
1054 }
1055 module_init(init);
1056 module_exit(fini);
1057 
1058 MODULE_DEVICE_TABLE(virtio, id_table);
1059 MODULE_DESCRIPTION("Virtio block driver");
1060 MODULE_LICENSE("GPL");
1061