xref: /openbmc/linux/drivers/block/virtio_blk.c (revision e3d786a3)
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 
22 static int major;
23 static DEFINE_IDA(vd_index_ida);
24 
25 static struct workqueue_struct *virtblk_wq;
26 
27 struct virtio_blk_vq {
28 	struct virtqueue *vq;
29 	spinlock_t lock;
30 	char name[VQ_NAME_LEN];
31 } ____cacheline_aligned_in_smp;
32 
33 struct virtio_blk {
34 	struct virtio_device *vdev;
35 
36 	/* The disk structure for the kernel. */
37 	struct gendisk *disk;
38 
39 	/* Block layer tags. */
40 	struct blk_mq_tag_set tag_set;
41 
42 	/* Process context for config space updates */
43 	struct work_struct config_work;
44 
45 	/* What host tells us, plus 2 for header & tailer. */
46 	unsigned int sg_elems;
47 
48 	/* Ida index - used to track minor number allocations. */
49 	int index;
50 
51 	/* num of vqs */
52 	int num_vqs;
53 	struct virtio_blk_vq *vqs;
54 };
55 
56 struct virtblk_req {
57 #ifdef CONFIG_VIRTIO_BLK_SCSI
58 	struct scsi_request sreq;	/* for SCSI passthrough, must be first */
59 	u8 sense[SCSI_SENSE_BUFFERSIZE];
60 	struct virtio_scsi_inhdr in_hdr;
61 #endif
62 	struct virtio_blk_outhdr out_hdr;
63 	u8 status;
64 	struct scatterlist sg[];
65 };
66 
67 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
68 {
69 	switch (vbr->status) {
70 	case VIRTIO_BLK_S_OK:
71 		return BLK_STS_OK;
72 	case VIRTIO_BLK_S_UNSUPP:
73 		return BLK_STS_NOTSUPP;
74 	default:
75 		return BLK_STS_IOERR;
76 	}
77 }
78 
79 /*
80  * If this is a packet command we need a couple of additional headers.  Behind
81  * the normal outhdr we put a segment with the scsi command block, and before
82  * the normal inhdr we put the sense data and the inhdr with additional status
83  * information.
84  */
85 #ifdef CONFIG_VIRTIO_BLK_SCSI
86 static int virtblk_add_req_scsi(struct virtqueue *vq, struct virtblk_req *vbr,
87 		struct scatterlist *data_sg, bool have_data)
88 {
89 	struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
90 	unsigned int num_out = 0, num_in = 0;
91 
92 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
93 	sgs[num_out++] = &hdr;
94 	sg_init_one(&cmd, vbr->sreq.cmd, vbr->sreq.cmd_len);
95 	sgs[num_out++] = &cmd;
96 
97 	if (have_data) {
98 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
99 			sgs[num_out++] = data_sg;
100 		else
101 			sgs[num_out + num_in++] = data_sg;
102 	}
103 
104 	sg_init_one(&sense, vbr->sense, SCSI_SENSE_BUFFERSIZE);
105 	sgs[num_out + num_in++] = &sense;
106 	sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
107 	sgs[num_out + num_in++] = &inhdr;
108 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
109 	sgs[num_out + num_in++] = &status;
110 
111 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
112 }
113 
114 static inline void virtblk_scsi_request_done(struct request *req)
115 {
116 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
117 	struct virtio_blk *vblk = req->q->queuedata;
118 	struct scsi_request *sreq = &vbr->sreq;
119 
120 	sreq->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual);
121 	sreq->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len);
122 	sreq->result = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors);
123 }
124 
125 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
126 			     unsigned int cmd, unsigned long data)
127 {
128 	struct gendisk *disk = bdev->bd_disk;
129 	struct virtio_blk *vblk = disk->private_data;
130 
131 	/*
132 	 * Only allow the generic SCSI ioctls if the host can support it.
133 	 */
134 	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
135 		return -ENOTTY;
136 
137 	return scsi_cmd_blk_ioctl(bdev, mode, cmd,
138 				  (void __user *)data);
139 }
140 #else
141 static inline int virtblk_add_req_scsi(struct virtqueue *vq,
142 		struct virtblk_req *vbr, struct scatterlist *data_sg,
143 		bool have_data)
144 {
145 	return -EIO;
146 }
147 static inline void virtblk_scsi_request_done(struct request *req)
148 {
149 }
150 #define virtblk_ioctl	NULL
151 #endif /* CONFIG_VIRTIO_BLK_SCSI */
152 
153 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
154 		struct scatterlist *data_sg, bool have_data)
155 {
156 	struct scatterlist hdr, status, *sgs[3];
157 	unsigned int num_out = 0, num_in = 0;
158 
159 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
160 	sgs[num_out++] = &hdr;
161 
162 	if (have_data) {
163 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
164 			sgs[num_out++] = data_sg;
165 		else
166 			sgs[num_out + num_in++] = data_sg;
167 	}
168 
169 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
170 	sgs[num_out + num_in++] = &status;
171 
172 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
173 }
174 
175 static inline void virtblk_request_done(struct request *req)
176 {
177 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
178 
179 	switch (req_op(req)) {
180 	case REQ_OP_SCSI_IN:
181 	case REQ_OP_SCSI_OUT:
182 		virtblk_scsi_request_done(req);
183 		break;
184 	}
185 
186 	blk_mq_end_request(req, virtblk_result(vbr));
187 }
188 
189 static void virtblk_done(struct virtqueue *vq)
190 {
191 	struct virtio_blk *vblk = vq->vdev->priv;
192 	bool req_done = false;
193 	int qid = vq->index;
194 	struct virtblk_req *vbr;
195 	unsigned long flags;
196 	unsigned int len;
197 
198 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
199 	do {
200 		virtqueue_disable_cb(vq);
201 		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
202 			struct request *req = blk_mq_rq_from_pdu(vbr);
203 
204 			blk_mq_complete_request(req);
205 			req_done = true;
206 		}
207 		if (unlikely(virtqueue_is_broken(vq)))
208 			break;
209 	} while (!virtqueue_enable_cb(vq));
210 
211 	/* In case queue is stopped waiting for more buffers. */
212 	if (req_done)
213 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
214 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
215 }
216 
217 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
218 			   const struct blk_mq_queue_data *bd)
219 {
220 	struct virtio_blk *vblk = hctx->queue->queuedata;
221 	struct request *req = bd->rq;
222 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
223 	unsigned long flags;
224 	unsigned int num;
225 	int qid = hctx->queue_num;
226 	int err;
227 	bool notify = false;
228 	u32 type;
229 
230 	BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
231 
232 	switch (req_op(req)) {
233 	case REQ_OP_READ:
234 	case REQ_OP_WRITE:
235 		type = 0;
236 		break;
237 	case REQ_OP_FLUSH:
238 		type = VIRTIO_BLK_T_FLUSH;
239 		break;
240 	case REQ_OP_SCSI_IN:
241 	case REQ_OP_SCSI_OUT:
242 		type = VIRTIO_BLK_T_SCSI_CMD;
243 		break;
244 	case REQ_OP_DRV_IN:
245 		type = VIRTIO_BLK_T_GET_ID;
246 		break;
247 	default:
248 		WARN_ON_ONCE(1);
249 		return BLK_STS_IOERR;
250 	}
251 
252 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
253 	vbr->out_hdr.sector = type ?
254 		0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
255 	vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
256 
257 	blk_mq_start_request(req);
258 
259 	num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
260 	if (num) {
261 		if (rq_data_dir(req) == WRITE)
262 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
263 		else
264 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
265 	}
266 
267 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
268 	if (blk_rq_is_scsi(req))
269 		err = virtblk_add_req_scsi(vblk->vqs[qid].vq, vbr, vbr->sg, num);
270 	else
271 		err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
272 	if (err) {
273 		virtqueue_kick(vblk->vqs[qid].vq);
274 		blk_mq_stop_hw_queue(hctx);
275 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
276 		/* Out of mem doesn't actually happen, since we fall back
277 		 * to direct descriptors */
278 		if (err == -ENOMEM || err == -ENOSPC)
279 			return BLK_STS_DEV_RESOURCE;
280 		return BLK_STS_IOERR;
281 	}
282 
283 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
284 		notify = true;
285 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
286 
287 	if (notify)
288 		virtqueue_notify(vblk->vqs[qid].vq);
289 	return BLK_STS_OK;
290 }
291 
292 /* return id (s/n) string for *disk to *id_str
293  */
294 static int virtblk_get_id(struct gendisk *disk, char *id_str)
295 {
296 	struct virtio_blk *vblk = disk->private_data;
297 	struct request_queue *q = vblk->disk->queue;
298 	struct request *req;
299 	int err;
300 
301 	req = blk_get_request(q, REQ_OP_DRV_IN, 0);
302 	if (IS_ERR(req))
303 		return PTR_ERR(req);
304 
305 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
306 	if (err)
307 		goto out;
308 
309 	blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
310 	err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
311 out:
312 	blk_put_request(req);
313 	return err;
314 }
315 
316 /* We provide getgeo only to please some old bootloader/partitioning tools */
317 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
318 {
319 	struct virtio_blk *vblk = bd->bd_disk->private_data;
320 
321 	/* see if the host passed in geometry config */
322 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
323 		virtio_cread(vblk->vdev, struct virtio_blk_config,
324 			     geometry.cylinders, &geo->cylinders);
325 		virtio_cread(vblk->vdev, struct virtio_blk_config,
326 			     geometry.heads, &geo->heads);
327 		virtio_cread(vblk->vdev, struct virtio_blk_config,
328 			     geometry.sectors, &geo->sectors);
329 	} else {
330 		/* some standard values, similar to sd */
331 		geo->heads = 1 << 6;
332 		geo->sectors = 1 << 5;
333 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
334 	}
335 	return 0;
336 }
337 
338 static const struct block_device_operations virtblk_fops = {
339 	.ioctl  = virtblk_ioctl,
340 	.owner  = THIS_MODULE,
341 	.getgeo = virtblk_getgeo,
342 };
343 
344 static int index_to_minor(int index)
345 {
346 	return index << PART_BITS;
347 }
348 
349 static int minor_to_index(int minor)
350 {
351 	return minor >> PART_BITS;
352 }
353 
354 static ssize_t serial_show(struct device *dev,
355 			   struct device_attribute *attr, char *buf)
356 {
357 	struct gendisk *disk = dev_to_disk(dev);
358 	int err;
359 
360 	/* sysfs gives us a PAGE_SIZE buffer */
361 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
362 
363 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
364 	err = virtblk_get_id(disk, buf);
365 	if (!err)
366 		return strlen(buf);
367 
368 	if (err == -EIO) /* Unsupported? Make it empty. */
369 		return 0;
370 
371 	return err;
372 }
373 
374 static DEVICE_ATTR_RO(serial);
375 
376 /* The queue's logical block size must be set before calling this */
377 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
378 {
379 	struct virtio_device *vdev = vblk->vdev;
380 	struct request_queue *q = vblk->disk->queue;
381 	char cap_str_2[10], cap_str_10[10];
382 	unsigned long long nblocks;
383 	u64 capacity;
384 
385 	/* Host must always specify the capacity. */
386 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
387 
388 	/* If capacity is too big, truncate with warning. */
389 	if ((sector_t)capacity != capacity) {
390 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
391 			 (unsigned long long)capacity);
392 		capacity = (sector_t)-1;
393 	}
394 
395 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
396 
397 	string_get_size(nblocks, queue_logical_block_size(q),
398 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
399 	string_get_size(nblocks, queue_logical_block_size(q),
400 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
401 
402 	dev_notice(&vdev->dev,
403 		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
404 		   vblk->disk->disk_name,
405 		   resize ? "new size: " : "",
406 		   nblocks,
407 		   queue_logical_block_size(q),
408 		   cap_str_10,
409 		   cap_str_2);
410 
411 	set_capacity(vblk->disk, capacity);
412 }
413 
414 static void virtblk_config_changed_work(struct work_struct *work)
415 {
416 	struct virtio_blk *vblk =
417 		container_of(work, struct virtio_blk, config_work);
418 	char *envp[] = { "RESIZE=1", NULL };
419 
420 	virtblk_update_capacity(vblk, true);
421 	revalidate_disk(vblk->disk);
422 	kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
423 }
424 
425 static void virtblk_config_changed(struct virtio_device *vdev)
426 {
427 	struct virtio_blk *vblk = vdev->priv;
428 
429 	queue_work(virtblk_wq, &vblk->config_work);
430 }
431 
432 static int init_vq(struct virtio_blk *vblk)
433 {
434 	int err;
435 	int i;
436 	vq_callback_t **callbacks;
437 	const char **names;
438 	struct virtqueue **vqs;
439 	unsigned short num_vqs;
440 	struct virtio_device *vdev = vblk->vdev;
441 	struct irq_affinity desc = { 0, };
442 
443 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
444 				   struct virtio_blk_config, num_queues,
445 				   &num_vqs);
446 	if (err)
447 		num_vqs = 1;
448 
449 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
450 	if (!vblk->vqs)
451 		return -ENOMEM;
452 
453 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
454 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
455 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
456 	if (!names || !callbacks || !vqs) {
457 		err = -ENOMEM;
458 		goto out;
459 	}
460 
461 	for (i = 0; i < num_vqs; i++) {
462 		callbacks[i] = virtblk_done;
463 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
464 		names[i] = vblk->vqs[i].name;
465 	}
466 
467 	/* Discover virtqueues and write information to configuration.  */
468 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
469 	if (err)
470 		goto out;
471 
472 	for (i = 0; i < num_vqs; i++) {
473 		spin_lock_init(&vblk->vqs[i].lock);
474 		vblk->vqs[i].vq = vqs[i];
475 	}
476 	vblk->num_vqs = num_vqs;
477 
478 out:
479 	kfree(vqs);
480 	kfree(callbacks);
481 	kfree(names);
482 	if (err)
483 		kfree(vblk->vqs);
484 	return err;
485 }
486 
487 /*
488  * Legacy naming scheme used for virtio devices.  We are stuck with it for
489  * virtio blk but don't ever use it for any new driver.
490  */
491 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
492 {
493 	const int base = 'z' - 'a' + 1;
494 	char *begin = buf + strlen(prefix);
495 	char *end = buf + buflen;
496 	char *p;
497 	int unit;
498 
499 	p = end - 1;
500 	*p = '\0';
501 	unit = base;
502 	do {
503 		if (p == begin)
504 			return -EINVAL;
505 		*--p = 'a' + (index % unit);
506 		index = (index / unit) - 1;
507 	} while (index >= 0);
508 
509 	memmove(begin, p, end - p);
510 	memcpy(buf, prefix, strlen(prefix));
511 
512 	return 0;
513 }
514 
515 static int virtblk_get_cache_mode(struct virtio_device *vdev)
516 {
517 	u8 writeback;
518 	int err;
519 
520 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
521 				   struct virtio_blk_config, wce,
522 				   &writeback);
523 
524 	/*
525 	 * If WCE is not configurable and flush is not available,
526 	 * assume no writeback cache is in use.
527 	 */
528 	if (err)
529 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
530 
531 	return writeback;
532 }
533 
534 static void virtblk_update_cache_mode(struct virtio_device *vdev)
535 {
536 	u8 writeback = virtblk_get_cache_mode(vdev);
537 	struct virtio_blk *vblk = vdev->priv;
538 
539 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
540 	revalidate_disk(vblk->disk);
541 }
542 
543 static const char *const virtblk_cache_types[] = {
544 	"write through", "write back"
545 };
546 
547 static ssize_t
548 cache_type_store(struct device *dev, struct device_attribute *attr,
549 		 const char *buf, size_t count)
550 {
551 	struct gendisk *disk = dev_to_disk(dev);
552 	struct virtio_blk *vblk = disk->private_data;
553 	struct virtio_device *vdev = vblk->vdev;
554 	int i;
555 
556 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
557 	i = sysfs_match_string(virtblk_cache_types, buf);
558 	if (i < 0)
559 		return i;
560 
561 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
562 	virtblk_update_cache_mode(vdev);
563 	return count;
564 }
565 
566 static ssize_t
567 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
568 {
569 	struct gendisk *disk = dev_to_disk(dev);
570 	struct virtio_blk *vblk = disk->private_data;
571 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
572 
573 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
574 	return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
575 }
576 
577 static DEVICE_ATTR_RW(cache_type);
578 
579 static struct attribute *virtblk_attrs[] = {
580 	&dev_attr_serial.attr,
581 	&dev_attr_cache_type.attr,
582 	NULL,
583 };
584 
585 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
586 		struct attribute *a, int n)
587 {
588 	struct device *dev = container_of(kobj, struct device, kobj);
589 	struct gendisk *disk = dev_to_disk(dev);
590 	struct virtio_blk *vblk = disk->private_data;
591 	struct virtio_device *vdev = vblk->vdev;
592 
593 	if (a == &dev_attr_cache_type.attr &&
594 	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
595 		return S_IRUGO;
596 
597 	return a->mode;
598 }
599 
600 static const struct attribute_group virtblk_attr_group = {
601 	.attrs = virtblk_attrs,
602 	.is_visible = virtblk_attrs_are_visible,
603 };
604 
605 static const struct attribute_group *virtblk_attr_groups[] = {
606 	&virtblk_attr_group,
607 	NULL,
608 };
609 
610 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
611 		unsigned int hctx_idx, unsigned int numa_node)
612 {
613 	struct virtio_blk *vblk = set->driver_data;
614 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
615 
616 #ifdef CONFIG_VIRTIO_BLK_SCSI
617 	vbr->sreq.sense = vbr->sense;
618 #endif
619 	sg_init_table(vbr->sg, vblk->sg_elems);
620 	return 0;
621 }
622 
623 static int virtblk_map_queues(struct blk_mq_tag_set *set)
624 {
625 	struct virtio_blk *vblk = set->driver_data;
626 
627 	return blk_mq_virtio_map_queues(set, vblk->vdev, 0);
628 }
629 
630 #ifdef CONFIG_VIRTIO_BLK_SCSI
631 static void virtblk_initialize_rq(struct request *req)
632 {
633 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
634 
635 	scsi_req_init(&vbr->sreq);
636 }
637 #endif
638 
639 static const struct blk_mq_ops virtio_mq_ops = {
640 	.queue_rq	= virtio_queue_rq,
641 	.complete	= virtblk_request_done,
642 	.init_request	= virtblk_init_request,
643 #ifdef CONFIG_VIRTIO_BLK_SCSI
644 	.initialize_rq_fn = virtblk_initialize_rq,
645 #endif
646 	.map_queues	= virtblk_map_queues,
647 };
648 
649 static unsigned int virtblk_queue_depth;
650 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
651 
652 static int virtblk_probe(struct virtio_device *vdev)
653 {
654 	struct virtio_blk *vblk;
655 	struct request_queue *q;
656 	int err, index;
657 
658 	u32 v, blk_size, sg_elems, opt_io_size;
659 	u16 min_io_size;
660 	u8 physical_block_exp, alignment_offset;
661 
662 	if (!vdev->config->get) {
663 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
664 			__func__);
665 		return -EINVAL;
666 	}
667 
668 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
669 			     GFP_KERNEL);
670 	if (err < 0)
671 		goto out;
672 	index = err;
673 
674 	/* We need to know how many segments before we allocate. */
675 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
676 				   struct virtio_blk_config, seg_max,
677 				   &sg_elems);
678 
679 	/* We need at least one SG element, whatever they say. */
680 	if (err || !sg_elems)
681 		sg_elems = 1;
682 
683 	/* We need an extra sg elements at head and tail. */
684 	sg_elems += 2;
685 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
686 	if (!vblk) {
687 		err = -ENOMEM;
688 		goto out_free_index;
689 	}
690 
691 	vblk->vdev = vdev;
692 	vblk->sg_elems = sg_elems;
693 
694 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
695 
696 	err = init_vq(vblk);
697 	if (err)
698 		goto out_free_vblk;
699 
700 	/* FIXME: How many partitions?  How long is a piece of string? */
701 	vblk->disk = alloc_disk(1 << PART_BITS);
702 	if (!vblk->disk) {
703 		err = -ENOMEM;
704 		goto out_free_vq;
705 	}
706 
707 	/* Default queue sizing is to fill the ring. */
708 	if (!virtblk_queue_depth) {
709 		virtblk_queue_depth = vblk->vqs[0].vq->num_free;
710 		/* ... but without indirect descs, we use 2 descs per req */
711 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
712 			virtblk_queue_depth /= 2;
713 	}
714 
715 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
716 	vblk->tag_set.ops = &virtio_mq_ops;
717 	vblk->tag_set.queue_depth = virtblk_queue_depth;
718 	vblk->tag_set.numa_node = NUMA_NO_NODE;
719 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
720 	vblk->tag_set.cmd_size =
721 		sizeof(struct virtblk_req) +
722 		sizeof(struct scatterlist) * sg_elems;
723 	vblk->tag_set.driver_data = vblk;
724 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
725 
726 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
727 	if (err)
728 		goto out_put_disk;
729 
730 	q = blk_mq_init_queue(&vblk->tag_set);
731 	if (IS_ERR(q)) {
732 		err = -ENOMEM;
733 		goto out_free_tags;
734 	}
735 	vblk->disk->queue = q;
736 
737 	q->queuedata = vblk;
738 
739 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
740 
741 	vblk->disk->major = major;
742 	vblk->disk->first_minor = index_to_minor(index);
743 	vblk->disk->private_data = vblk;
744 	vblk->disk->fops = &virtblk_fops;
745 	vblk->disk->flags |= GENHD_FL_EXT_DEVT;
746 	vblk->index = index;
747 
748 	/* configure queue flush support */
749 	virtblk_update_cache_mode(vdev);
750 
751 	/* If disk is read-only in the host, the guest should obey */
752 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
753 		set_disk_ro(vblk->disk, 1);
754 
755 	/* We can handle whatever the host told us to handle. */
756 	blk_queue_max_segments(q, vblk->sg_elems-2);
757 
758 	/* No real sector limit. */
759 	blk_queue_max_hw_sectors(q, -1U);
760 
761 	/* Host can optionally specify maximum segment size and number of
762 	 * segments. */
763 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
764 				   struct virtio_blk_config, size_max, &v);
765 	if (!err)
766 		blk_queue_max_segment_size(q, v);
767 	else
768 		blk_queue_max_segment_size(q, -1U);
769 
770 	/* Host can optionally specify the block size of the device */
771 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
772 				   struct virtio_blk_config, blk_size,
773 				   &blk_size);
774 	if (!err)
775 		blk_queue_logical_block_size(q, blk_size);
776 	else
777 		blk_size = queue_logical_block_size(q);
778 
779 	/* Use topology information if available */
780 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
781 				   struct virtio_blk_config, physical_block_exp,
782 				   &physical_block_exp);
783 	if (!err && physical_block_exp)
784 		blk_queue_physical_block_size(q,
785 				blk_size * (1 << physical_block_exp));
786 
787 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
788 				   struct virtio_blk_config, alignment_offset,
789 				   &alignment_offset);
790 	if (!err && alignment_offset)
791 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
792 
793 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
794 				   struct virtio_blk_config, min_io_size,
795 				   &min_io_size);
796 	if (!err && min_io_size)
797 		blk_queue_io_min(q, blk_size * min_io_size);
798 
799 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
800 				   struct virtio_blk_config, opt_io_size,
801 				   &opt_io_size);
802 	if (!err && opt_io_size)
803 		blk_queue_io_opt(q, blk_size * opt_io_size);
804 
805 	virtblk_update_capacity(vblk, false);
806 	virtio_device_ready(vdev);
807 
808 	device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
809 	return 0;
810 
811 out_free_tags:
812 	blk_mq_free_tag_set(&vblk->tag_set);
813 out_put_disk:
814 	put_disk(vblk->disk);
815 out_free_vq:
816 	vdev->config->del_vqs(vdev);
817 out_free_vblk:
818 	kfree(vblk);
819 out_free_index:
820 	ida_simple_remove(&vd_index_ida, index);
821 out:
822 	return err;
823 }
824 
825 static void virtblk_remove(struct virtio_device *vdev)
826 {
827 	struct virtio_blk *vblk = vdev->priv;
828 	int index = vblk->index;
829 	int refc;
830 
831 	/* Make sure no work handler is accessing the device. */
832 	flush_work(&vblk->config_work);
833 
834 	del_gendisk(vblk->disk);
835 	blk_cleanup_queue(vblk->disk->queue);
836 
837 	blk_mq_free_tag_set(&vblk->tag_set);
838 
839 	/* Stop all the virtqueues. */
840 	vdev->config->reset(vdev);
841 
842 	refc = kref_read(&disk_to_dev(vblk->disk)->kobj.kref);
843 	put_disk(vblk->disk);
844 	vdev->config->del_vqs(vdev);
845 	kfree(vblk->vqs);
846 	kfree(vblk);
847 
848 	/* Only free device id if we don't have any users */
849 	if (refc == 1)
850 		ida_simple_remove(&vd_index_ida, index);
851 }
852 
853 #ifdef CONFIG_PM_SLEEP
854 static int virtblk_freeze(struct virtio_device *vdev)
855 {
856 	struct virtio_blk *vblk = vdev->priv;
857 
858 	/* Ensure we don't receive any more interrupts */
859 	vdev->config->reset(vdev);
860 
861 	/* Make sure no work handler is accessing the device. */
862 	flush_work(&vblk->config_work);
863 
864 	blk_mq_quiesce_queue(vblk->disk->queue);
865 
866 	vdev->config->del_vqs(vdev);
867 	return 0;
868 }
869 
870 static int virtblk_restore(struct virtio_device *vdev)
871 {
872 	struct virtio_blk *vblk = vdev->priv;
873 	int ret;
874 
875 	ret = init_vq(vdev->priv);
876 	if (ret)
877 		return ret;
878 
879 	virtio_device_ready(vdev);
880 
881 	blk_mq_unquiesce_queue(vblk->disk->queue);
882 	return 0;
883 }
884 #endif
885 
886 static const struct virtio_device_id id_table[] = {
887 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
888 	{ 0 },
889 };
890 
891 static unsigned int features_legacy[] = {
892 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
893 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
894 #ifdef CONFIG_VIRTIO_BLK_SCSI
895 	VIRTIO_BLK_F_SCSI,
896 #endif
897 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
898 	VIRTIO_BLK_F_MQ,
899 }
900 ;
901 static unsigned int features[] = {
902 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
903 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
904 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
905 	VIRTIO_BLK_F_MQ,
906 };
907 
908 static struct virtio_driver virtio_blk = {
909 	.feature_table			= features,
910 	.feature_table_size		= ARRAY_SIZE(features),
911 	.feature_table_legacy		= features_legacy,
912 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
913 	.driver.name			= KBUILD_MODNAME,
914 	.driver.owner			= THIS_MODULE,
915 	.id_table			= id_table,
916 	.probe				= virtblk_probe,
917 	.remove				= virtblk_remove,
918 	.config_changed			= virtblk_config_changed,
919 #ifdef CONFIG_PM_SLEEP
920 	.freeze				= virtblk_freeze,
921 	.restore			= virtblk_restore,
922 #endif
923 };
924 
925 static int __init init(void)
926 {
927 	int error;
928 
929 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
930 	if (!virtblk_wq)
931 		return -ENOMEM;
932 
933 	major = register_blkdev(0, "virtblk");
934 	if (major < 0) {
935 		error = major;
936 		goto out_destroy_workqueue;
937 	}
938 
939 	error = register_virtio_driver(&virtio_blk);
940 	if (error)
941 		goto out_unregister_blkdev;
942 	return 0;
943 
944 out_unregister_blkdev:
945 	unregister_blkdev(major, "virtblk");
946 out_destroy_workqueue:
947 	destroy_workqueue(virtblk_wq);
948 	return error;
949 }
950 
951 static void __exit fini(void)
952 {
953 	unregister_virtio_driver(&virtio_blk);
954 	unregister_blkdev(major, "virtblk");
955 	destroy_workqueue(virtblk_wq);
956 }
957 module_init(init);
958 module_exit(fini);
959 
960 MODULE_DEVICE_TABLE(virtio, id_table);
961 MODULE_DESCRIPTION("Virtio block driver");
962 MODULE_LICENSE("GPL");
963