xref: /openbmc/linux/drivers/block/virtio_blk.c (revision 8cb5d748)
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_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, GFP_KERNEL);
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 virtblk_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(serial, S_IRUGO, virtblk_serial_show, NULL);
375 
376 static void virtblk_config_changed_work(struct work_struct *work)
377 {
378 	struct virtio_blk *vblk =
379 		container_of(work, struct virtio_blk, config_work);
380 	struct virtio_device *vdev = vblk->vdev;
381 	struct request_queue *q = vblk->disk->queue;
382 	char cap_str_2[10], cap_str_10[10];
383 	char *envp[] = { "RESIZE=1", NULL };
384 	unsigned long long nblocks;
385 	u64 capacity;
386 
387 	/* Host must always specify the capacity. */
388 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
389 
390 	/* If capacity is too big, truncate with warning. */
391 	if ((sector_t)capacity != capacity) {
392 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
393 			 (unsigned long long)capacity);
394 		capacity = (sector_t)-1;
395 	}
396 
397 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
398 
399 	string_get_size(nblocks, queue_logical_block_size(q),
400 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
401 	string_get_size(nblocks, queue_logical_block_size(q),
402 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
403 
404 	dev_notice(&vdev->dev,
405 		   "new size: %llu %d-byte logical blocks (%s/%s)\n",
406 		   nblocks,
407 		   queue_logical_block_size(q),
408 		   cap_str_10,
409 		   cap_str_2);
410 
411 	set_capacity(vblk->disk, capacity);
412 	revalidate_disk(vblk->disk);
413 	kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
414 }
415 
416 static void virtblk_config_changed(struct virtio_device *vdev)
417 {
418 	struct virtio_blk *vblk = vdev->priv;
419 
420 	queue_work(virtblk_wq, &vblk->config_work);
421 }
422 
423 static int init_vq(struct virtio_blk *vblk)
424 {
425 	int err;
426 	int i;
427 	vq_callback_t **callbacks;
428 	const char **names;
429 	struct virtqueue **vqs;
430 	unsigned short num_vqs;
431 	struct virtio_device *vdev = vblk->vdev;
432 	struct irq_affinity desc = { 0, };
433 
434 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
435 				   struct virtio_blk_config, num_queues,
436 				   &num_vqs);
437 	if (err)
438 		num_vqs = 1;
439 
440 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
441 	if (!vblk->vqs)
442 		return -ENOMEM;
443 
444 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
445 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
446 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
447 	if (!names || !callbacks || !vqs) {
448 		err = -ENOMEM;
449 		goto out;
450 	}
451 
452 	for (i = 0; i < num_vqs; i++) {
453 		callbacks[i] = virtblk_done;
454 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
455 		names[i] = vblk->vqs[i].name;
456 	}
457 
458 	/* Discover virtqueues and write information to configuration.  */
459 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
460 	if (err)
461 		goto out;
462 
463 	for (i = 0; i < num_vqs; i++) {
464 		spin_lock_init(&vblk->vqs[i].lock);
465 		vblk->vqs[i].vq = vqs[i];
466 	}
467 	vblk->num_vqs = num_vqs;
468 
469 out:
470 	kfree(vqs);
471 	kfree(callbacks);
472 	kfree(names);
473 	if (err)
474 		kfree(vblk->vqs);
475 	return err;
476 }
477 
478 /*
479  * Legacy naming scheme used for virtio devices.  We are stuck with it for
480  * virtio blk but don't ever use it for any new driver.
481  */
482 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
483 {
484 	const int base = 'z' - 'a' + 1;
485 	char *begin = buf + strlen(prefix);
486 	char *end = buf + buflen;
487 	char *p;
488 	int unit;
489 
490 	p = end - 1;
491 	*p = '\0';
492 	unit = base;
493 	do {
494 		if (p == begin)
495 			return -EINVAL;
496 		*--p = 'a' + (index % unit);
497 		index = (index / unit) - 1;
498 	} while (index >= 0);
499 
500 	memmove(begin, p, end - p);
501 	memcpy(buf, prefix, strlen(prefix));
502 
503 	return 0;
504 }
505 
506 static int virtblk_get_cache_mode(struct virtio_device *vdev)
507 {
508 	u8 writeback;
509 	int err;
510 
511 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
512 				   struct virtio_blk_config, wce,
513 				   &writeback);
514 
515 	/*
516 	 * If WCE is not configurable and flush is not available,
517 	 * assume no writeback cache is in use.
518 	 */
519 	if (err)
520 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
521 
522 	return writeback;
523 }
524 
525 static void virtblk_update_cache_mode(struct virtio_device *vdev)
526 {
527 	u8 writeback = virtblk_get_cache_mode(vdev);
528 	struct virtio_blk *vblk = vdev->priv;
529 
530 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
531 	revalidate_disk(vblk->disk);
532 }
533 
534 static const char *const virtblk_cache_types[] = {
535 	"write through", "write back"
536 };
537 
538 static ssize_t
539 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
540 			 const char *buf, size_t count)
541 {
542 	struct gendisk *disk = dev_to_disk(dev);
543 	struct virtio_blk *vblk = disk->private_data;
544 	struct virtio_device *vdev = vblk->vdev;
545 	int i;
546 
547 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
548 	i = sysfs_match_string(virtblk_cache_types, buf);
549 	if (i < 0)
550 		return i;
551 
552 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
553 	virtblk_update_cache_mode(vdev);
554 	return count;
555 }
556 
557 static ssize_t
558 virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
559 			 char *buf)
560 {
561 	struct gendisk *disk = dev_to_disk(dev);
562 	struct virtio_blk *vblk = disk->private_data;
563 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
564 
565 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
566 	return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
567 }
568 
569 static const struct device_attribute dev_attr_cache_type_ro =
570 	__ATTR(cache_type, S_IRUGO,
571 	       virtblk_cache_type_show, NULL);
572 static const struct device_attribute dev_attr_cache_type_rw =
573 	__ATTR(cache_type, S_IRUGO|S_IWUSR,
574 	       virtblk_cache_type_show, virtblk_cache_type_store);
575 
576 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
577 		unsigned int hctx_idx, unsigned int numa_node)
578 {
579 	struct virtio_blk *vblk = set->driver_data;
580 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
581 
582 #ifdef CONFIG_VIRTIO_BLK_SCSI
583 	vbr->sreq.sense = vbr->sense;
584 #endif
585 	sg_init_table(vbr->sg, vblk->sg_elems);
586 	return 0;
587 }
588 
589 static int virtblk_map_queues(struct blk_mq_tag_set *set)
590 {
591 	struct virtio_blk *vblk = set->driver_data;
592 
593 	return blk_mq_virtio_map_queues(set, vblk->vdev, 0);
594 }
595 
596 static const struct blk_mq_ops virtio_mq_ops = {
597 	.queue_rq	= virtio_queue_rq,
598 	.complete	= virtblk_request_done,
599 	.init_request	= virtblk_init_request,
600 	.map_queues	= virtblk_map_queues,
601 };
602 
603 static unsigned int virtblk_queue_depth;
604 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
605 
606 static int virtblk_probe(struct virtio_device *vdev)
607 {
608 	struct virtio_blk *vblk;
609 	struct request_queue *q;
610 	int err, index;
611 
612 	u64 cap;
613 	u32 v, blk_size, sg_elems, opt_io_size;
614 	u16 min_io_size;
615 	u8 physical_block_exp, alignment_offset;
616 
617 	if (!vdev->config->get) {
618 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
619 			__func__);
620 		return -EINVAL;
621 	}
622 
623 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
624 			     GFP_KERNEL);
625 	if (err < 0)
626 		goto out;
627 	index = err;
628 
629 	/* We need to know how many segments before we allocate. */
630 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
631 				   struct virtio_blk_config, seg_max,
632 				   &sg_elems);
633 
634 	/* We need at least one SG element, whatever they say. */
635 	if (err || !sg_elems)
636 		sg_elems = 1;
637 
638 	/* We need an extra sg elements at head and tail. */
639 	sg_elems += 2;
640 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
641 	if (!vblk) {
642 		err = -ENOMEM;
643 		goto out_free_index;
644 	}
645 
646 	vblk->vdev = vdev;
647 	vblk->sg_elems = sg_elems;
648 
649 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
650 
651 	err = init_vq(vblk);
652 	if (err)
653 		goto out_free_vblk;
654 
655 	/* FIXME: How many partitions?  How long is a piece of string? */
656 	vblk->disk = alloc_disk(1 << PART_BITS);
657 	if (!vblk->disk) {
658 		err = -ENOMEM;
659 		goto out_free_vq;
660 	}
661 
662 	/* Default queue sizing is to fill the ring. */
663 	if (!virtblk_queue_depth) {
664 		virtblk_queue_depth = vblk->vqs[0].vq->num_free;
665 		/* ... but without indirect descs, we use 2 descs per req */
666 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
667 			virtblk_queue_depth /= 2;
668 	}
669 
670 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
671 	vblk->tag_set.ops = &virtio_mq_ops;
672 	vblk->tag_set.queue_depth = virtblk_queue_depth;
673 	vblk->tag_set.numa_node = NUMA_NO_NODE;
674 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
675 	vblk->tag_set.cmd_size =
676 		sizeof(struct virtblk_req) +
677 		sizeof(struct scatterlist) * sg_elems;
678 	vblk->tag_set.driver_data = vblk;
679 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
680 
681 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
682 	if (err)
683 		goto out_put_disk;
684 
685 	q = blk_mq_init_queue(&vblk->tag_set);
686 	if (IS_ERR(q)) {
687 		err = -ENOMEM;
688 		goto out_free_tags;
689 	}
690 	vblk->disk->queue = q;
691 
692 	q->queuedata = vblk;
693 
694 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
695 
696 	vblk->disk->major = major;
697 	vblk->disk->first_minor = index_to_minor(index);
698 	vblk->disk->private_data = vblk;
699 	vblk->disk->fops = &virtblk_fops;
700 	vblk->disk->flags |= GENHD_FL_EXT_DEVT;
701 	vblk->index = index;
702 
703 	/* configure queue flush support */
704 	virtblk_update_cache_mode(vdev);
705 
706 	/* If disk is read-only in the host, the guest should obey */
707 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
708 		set_disk_ro(vblk->disk, 1);
709 
710 	/* Host must always specify the capacity. */
711 	virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
712 
713 	/* If capacity is too big, truncate with warning. */
714 	if ((sector_t)cap != cap) {
715 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
716 			 (unsigned long long)cap);
717 		cap = (sector_t)-1;
718 	}
719 	set_capacity(vblk->disk, cap);
720 
721 	/* We can handle whatever the host told us to handle. */
722 	blk_queue_max_segments(q, vblk->sg_elems-2);
723 
724 	/* No real sector limit. */
725 	blk_queue_max_hw_sectors(q, -1U);
726 
727 	/* Host can optionally specify maximum segment size and number of
728 	 * segments. */
729 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
730 				   struct virtio_blk_config, size_max, &v);
731 	if (!err)
732 		blk_queue_max_segment_size(q, v);
733 	else
734 		blk_queue_max_segment_size(q, -1U);
735 
736 	/* Host can optionally specify the block size of the device */
737 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
738 				   struct virtio_blk_config, blk_size,
739 				   &blk_size);
740 	if (!err)
741 		blk_queue_logical_block_size(q, blk_size);
742 	else
743 		blk_size = queue_logical_block_size(q);
744 
745 	/* Use topology information if available */
746 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
747 				   struct virtio_blk_config, physical_block_exp,
748 				   &physical_block_exp);
749 	if (!err && physical_block_exp)
750 		blk_queue_physical_block_size(q,
751 				blk_size * (1 << physical_block_exp));
752 
753 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
754 				   struct virtio_blk_config, alignment_offset,
755 				   &alignment_offset);
756 	if (!err && alignment_offset)
757 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
758 
759 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
760 				   struct virtio_blk_config, min_io_size,
761 				   &min_io_size);
762 	if (!err && min_io_size)
763 		blk_queue_io_min(q, blk_size * min_io_size);
764 
765 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
766 				   struct virtio_blk_config, opt_io_size,
767 				   &opt_io_size);
768 	if (!err && opt_io_size)
769 		blk_queue_io_opt(q, blk_size * opt_io_size);
770 
771 	virtio_device_ready(vdev);
772 
773 	device_add_disk(&vdev->dev, vblk->disk);
774 	err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
775 	if (err)
776 		goto out_del_disk;
777 
778 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
779 		err = device_create_file(disk_to_dev(vblk->disk),
780 					 &dev_attr_cache_type_rw);
781 	else
782 		err = device_create_file(disk_to_dev(vblk->disk),
783 					 &dev_attr_cache_type_ro);
784 	if (err)
785 		goto out_del_disk;
786 	return 0;
787 
788 out_del_disk:
789 	del_gendisk(vblk->disk);
790 	blk_cleanup_queue(vblk->disk->queue);
791 out_free_tags:
792 	blk_mq_free_tag_set(&vblk->tag_set);
793 out_put_disk:
794 	put_disk(vblk->disk);
795 out_free_vq:
796 	vdev->config->del_vqs(vdev);
797 out_free_vblk:
798 	kfree(vblk);
799 out_free_index:
800 	ida_simple_remove(&vd_index_ida, index);
801 out:
802 	return err;
803 }
804 
805 static void virtblk_remove(struct virtio_device *vdev)
806 {
807 	struct virtio_blk *vblk = vdev->priv;
808 	int index = vblk->index;
809 	int refc;
810 
811 	/* Make sure no work handler is accessing the device. */
812 	flush_work(&vblk->config_work);
813 
814 	del_gendisk(vblk->disk);
815 	blk_cleanup_queue(vblk->disk->queue);
816 
817 	blk_mq_free_tag_set(&vblk->tag_set);
818 
819 	/* Stop all the virtqueues. */
820 	vdev->config->reset(vdev);
821 
822 	refc = kref_read(&disk_to_dev(vblk->disk)->kobj.kref);
823 	put_disk(vblk->disk);
824 	vdev->config->del_vqs(vdev);
825 	kfree(vblk->vqs);
826 	kfree(vblk);
827 
828 	/* Only free device id if we don't have any users */
829 	if (refc == 1)
830 		ida_simple_remove(&vd_index_ida, index);
831 }
832 
833 #ifdef CONFIG_PM_SLEEP
834 static int virtblk_freeze(struct virtio_device *vdev)
835 {
836 	struct virtio_blk *vblk = vdev->priv;
837 
838 	/* Ensure we don't receive any more interrupts */
839 	vdev->config->reset(vdev);
840 
841 	/* Make sure no work handler is accessing the device. */
842 	flush_work(&vblk->config_work);
843 
844 	blk_mq_quiesce_queue(vblk->disk->queue);
845 
846 	vdev->config->del_vqs(vdev);
847 	return 0;
848 }
849 
850 static int virtblk_restore(struct virtio_device *vdev)
851 {
852 	struct virtio_blk *vblk = vdev->priv;
853 	int ret;
854 
855 	ret = init_vq(vdev->priv);
856 	if (ret)
857 		return ret;
858 
859 	virtio_device_ready(vdev);
860 
861 	blk_mq_unquiesce_queue(vblk->disk->queue);
862 	return 0;
863 }
864 #endif
865 
866 static const struct virtio_device_id id_table[] = {
867 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
868 	{ 0 },
869 };
870 
871 static unsigned int features_legacy[] = {
872 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
873 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
874 #ifdef CONFIG_VIRTIO_BLK_SCSI
875 	VIRTIO_BLK_F_SCSI,
876 #endif
877 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
878 	VIRTIO_BLK_F_MQ,
879 }
880 ;
881 static unsigned int features[] = {
882 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
883 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
884 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
885 	VIRTIO_BLK_F_MQ,
886 };
887 
888 static struct virtio_driver virtio_blk = {
889 	.feature_table			= features,
890 	.feature_table_size		= ARRAY_SIZE(features),
891 	.feature_table_legacy		= features_legacy,
892 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
893 	.driver.name			= KBUILD_MODNAME,
894 	.driver.owner			= THIS_MODULE,
895 	.id_table			= id_table,
896 	.probe				= virtblk_probe,
897 	.remove				= virtblk_remove,
898 	.config_changed			= virtblk_config_changed,
899 #ifdef CONFIG_PM_SLEEP
900 	.freeze				= virtblk_freeze,
901 	.restore			= virtblk_restore,
902 #endif
903 };
904 
905 static int __init init(void)
906 {
907 	int error;
908 
909 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
910 	if (!virtblk_wq)
911 		return -ENOMEM;
912 
913 	major = register_blkdev(0, "virtblk");
914 	if (major < 0) {
915 		error = major;
916 		goto out_destroy_workqueue;
917 	}
918 
919 	error = register_virtio_driver(&virtio_blk);
920 	if (error)
921 		goto out_unregister_blkdev;
922 	return 0;
923 
924 out_unregister_blkdev:
925 	unregister_blkdev(major, "virtblk");
926 out_destroy_workqueue:
927 	destroy_workqueue(virtblk_wq);
928 	return error;
929 }
930 
931 static void __exit fini(void)
932 {
933 	unregister_virtio_driver(&virtio_blk);
934 	unregister_blkdev(major, "virtblk");
935 	destroy_workqueue(virtblk_wq);
936 }
937 module_init(init);
938 module_exit(fini);
939 
940 MODULE_DEVICE_TABLE(virtio, id_table);
941 MODULE_DESCRIPTION("Virtio block driver");
942 MODULE_LICENSE("GPL");
943