xref: /openbmc/linux/drivers/block/virtio_blk.c (revision 1c2dd16a)
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 int virtblk_result(struct virtblk_req *vbr)
68 {
69 	switch (vbr->status) {
70 	case VIRTIO_BLK_S_OK:
71 		return 0;
72 	case VIRTIO_BLK_S_UNSUPP:
73 		return -ENOTTY;
74 	default:
75 		return -EIO;
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 int 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_MQ_RQ_QUEUE_ERROR;
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 (req_op(req) == REQ_OP_SCSI_IN || req_op(req) == REQ_OP_SCSI_OUT)
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_MQ_RQ_QUEUE_BUSY;
280 		return BLK_MQ_RQ_QUEUE_ERROR;
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_MQ_RQ_QUEUE_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 = 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 	u64 capacity;
385 
386 	/* Host must always specify the capacity. */
387 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
388 
389 	/* If capacity is too big, truncate with warning. */
390 	if ((sector_t)capacity != capacity) {
391 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
392 			 (unsigned long long)capacity);
393 		capacity = (sector_t)-1;
394 	}
395 
396 	string_get_size(capacity, queue_logical_block_size(q),
397 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
398 	string_get_size(capacity, queue_logical_block_size(q),
399 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
400 
401 	dev_notice(&vdev->dev,
402 		  "new size: %llu %d-byte logical blocks (%s/%s)\n",
403 		  (unsigned long long)capacity,
404 		  queue_logical_block_size(q),
405 		  cap_str_10, cap_str_2);
406 
407 	set_capacity(vblk->disk, capacity);
408 	revalidate_disk(vblk->disk);
409 	kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
410 }
411 
412 static void virtblk_config_changed(struct virtio_device *vdev)
413 {
414 	struct virtio_blk *vblk = vdev->priv;
415 
416 	queue_work(virtblk_wq, &vblk->config_work);
417 }
418 
419 static int init_vq(struct virtio_blk *vblk)
420 {
421 	int err;
422 	int i;
423 	vq_callback_t **callbacks;
424 	const char **names;
425 	struct virtqueue **vqs;
426 	unsigned short num_vqs;
427 	struct virtio_device *vdev = vblk->vdev;
428 	struct irq_affinity desc = { 0, };
429 
430 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
431 				   struct virtio_blk_config, num_queues,
432 				   &num_vqs);
433 	if (err)
434 		num_vqs = 1;
435 
436 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
437 	if (!vblk->vqs)
438 		return -ENOMEM;
439 
440 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
441 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
442 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
443 	if (!names || !callbacks || !vqs) {
444 		err = -ENOMEM;
445 		goto out;
446 	}
447 
448 	for (i = 0; i < num_vqs; i++) {
449 		callbacks[i] = virtblk_done;
450 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
451 		names[i] = vblk->vqs[i].name;
452 	}
453 
454 	/* Discover virtqueues and write information to configuration.  */
455 	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names,
456 			&desc);
457 	if (err)
458 		goto out;
459 
460 	for (i = 0; i < num_vqs; i++) {
461 		spin_lock_init(&vblk->vqs[i].lock);
462 		vblk->vqs[i].vq = vqs[i];
463 	}
464 	vblk->num_vqs = num_vqs;
465 
466 out:
467 	kfree(vqs);
468 	kfree(callbacks);
469 	kfree(names);
470 	if (err)
471 		kfree(vblk->vqs);
472 	return err;
473 }
474 
475 /*
476  * Legacy naming scheme used for virtio devices.  We are stuck with it for
477  * virtio blk but don't ever use it for any new driver.
478  */
479 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
480 {
481 	const int base = 'z' - 'a' + 1;
482 	char *begin = buf + strlen(prefix);
483 	char *end = buf + buflen;
484 	char *p;
485 	int unit;
486 
487 	p = end - 1;
488 	*p = '\0';
489 	unit = base;
490 	do {
491 		if (p == begin)
492 			return -EINVAL;
493 		*--p = 'a' + (index % unit);
494 		index = (index / unit) - 1;
495 	} while (index >= 0);
496 
497 	memmove(begin, p, end - p);
498 	memcpy(buf, prefix, strlen(prefix));
499 
500 	return 0;
501 }
502 
503 static int virtblk_get_cache_mode(struct virtio_device *vdev)
504 {
505 	u8 writeback;
506 	int err;
507 
508 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
509 				   struct virtio_blk_config, wce,
510 				   &writeback);
511 
512 	/*
513 	 * If WCE is not configurable and flush is not available,
514 	 * assume no writeback cache is in use.
515 	 */
516 	if (err)
517 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
518 
519 	return writeback;
520 }
521 
522 static void virtblk_update_cache_mode(struct virtio_device *vdev)
523 {
524 	u8 writeback = virtblk_get_cache_mode(vdev);
525 	struct virtio_blk *vblk = vdev->priv;
526 
527 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
528 	revalidate_disk(vblk->disk);
529 }
530 
531 static const char *const virtblk_cache_types[] = {
532 	"write through", "write back"
533 };
534 
535 static ssize_t
536 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
537 			 const char *buf, size_t count)
538 {
539 	struct gendisk *disk = dev_to_disk(dev);
540 	struct virtio_blk *vblk = disk->private_data;
541 	struct virtio_device *vdev = vblk->vdev;
542 	int i;
543 
544 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
545 	for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
546 		if (sysfs_streq(buf, virtblk_cache_types[i]))
547 			break;
548 
549 	if (i < 0)
550 		return -EINVAL;
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(void *data, struct request *rq,
577 		unsigned int hctx_idx, unsigned int request_idx,
578 		unsigned int numa_node)
579 {
580 	struct virtio_blk *vblk = data;
581 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
582 
583 #ifdef CONFIG_VIRTIO_BLK_SCSI
584 	vbr->sreq.sense = vbr->sense;
585 #endif
586 	sg_init_table(vbr->sg, vblk->sg_elems);
587 	return 0;
588 }
589 
590 static int virtblk_map_queues(struct blk_mq_tag_set *set)
591 {
592 	struct virtio_blk *vblk = set->driver_data;
593 
594 	return blk_mq_virtio_map_queues(set, vblk->vdev, 0);
595 }
596 
597 static const struct blk_mq_ops virtio_mq_ops = {
598 	.queue_rq	= virtio_queue_rq,
599 	.complete	= virtblk_request_done,
600 	.init_request	= virtblk_init_request,
601 	.map_queues	= virtblk_map_queues,
602 };
603 
604 static unsigned int virtblk_queue_depth;
605 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
606 
607 static int virtblk_probe(struct virtio_device *vdev)
608 {
609 	struct virtio_blk *vblk;
610 	struct request_queue *q;
611 	int err, index;
612 
613 	u64 cap;
614 	u32 v, blk_size, sg_elems, opt_io_size;
615 	u16 min_io_size;
616 	u8 physical_block_exp, alignment_offset;
617 
618 	if (!vdev->config->get) {
619 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
620 			__func__);
621 		return -EINVAL;
622 	}
623 
624 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
625 			     GFP_KERNEL);
626 	if (err < 0)
627 		goto out;
628 	index = err;
629 
630 	/* We need to know how many segments before we allocate. */
631 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
632 				   struct virtio_blk_config, seg_max,
633 				   &sg_elems);
634 
635 	/* We need at least one SG element, whatever they say. */
636 	if (err || !sg_elems)
637 		sg_elems = 1;
638 
639 	/* We need an extra sg elements at head and tail. */
640 	sg_elems += 2;
641 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
642 	if (!vblk) {
643 		err = -ENOMEM;
644 		goto out_free_index;
645 	}
646 
647 	vblk->vdev = vdev;
648 	vblk->sg_elems = sg_elems;
649 
650 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
651 
652 	err = init_vq(vblk);
653 	if (err)
654 		goto out_free_vblk;
655 
656 	/* FIXME: How many partitions?  How long is a piece of string? */
657 	vblk->disk = alloc_disk(1 << PART_BITS);
658 	if (!vblk->disk) {
659 		err = -ENOMEM;
660 		goto out_free_vq;
661 	}
662 
663 	/* Default queue sizing is to fill the ring. */
664 	if (!virtblk_queue_depth) {
665 		virtblk_queue_depth = vblk->vqs[0].vq->num_free;
666 		/* ... but without indirect descs, we use 2 descs per req */
667 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
668 			virtblk_queue_depth /= 2;
669 	}
670 
671 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
672 	vblk->tag_set.ops = &virtio_mq_ops;
673 	vblk->tag_set.queue_depth = virtblk_queue_depth;
674 	vblk->tag_set.numa_node = NUMA_NO_NODE;
675 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
676 	vblk->tag_set.cmd_size =
677 		sizeof(struct virtblk_req) +
678 		sizeof(struct scatterlist) * sg_elems;
679 	vblk->tag_set.driver_data = vblk;
680 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
681 
682 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
683 	if (err)
684 		goto out_put_disk;
685 
686 	q = blk_mq_init_queue(&vblk->tag_set);
687 	if (IS_ERR(q)) {
688 		err = -ENOMEM;
689 		goto out_free_tags;
690 	}
691 	vblk->disk->queue = q;
692 
693 	q->queuedata = vblk;
694 
695 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
696 
697 	vblk->disk->major = major;
698 	vblk->disk->first_minor = index_to_minor(index);
699 	vblk->disk->private_data = vblk;
700 	vblk->disk->fops = &virtblk_fops;
701 	vblk->disk->flags |= GENHD_FL_EXT_DEVT;
702 	vblk->index = index;
703 
704 	/* configure queue flush support */
705 	virtblk_update_cache_mode(vdev);
706 
707 	/* If disk is read-only in the host, the guest should obey */
708 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
709 		set_disk_ro(vblk->disk, 1);
710 
711 	/* Host must always specify the capacity. */
712 	virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
713 
714 	/* If capacity is too big, truncate with warning. */
715 	if ((sector_t)cap != cap) {
716 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
717 			 (unsigned long long)cap);
718 		cap = (sector_t)-1;
719 	}
720 	set_capacity(vblk->disk, cap);
721 
722 	/* We can handle whatever the host told us to handle. */
723 	blk_queue_max_segments(q, vblk->sg_elems-2);
724 
725 	/* No need to bounce any requests */
726 	blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
727 
728 	/* No real sector limit. */
729 	blk_queue_max_hw_sectors(q, -1U);
730 
731 	/* Host can optionally specify maximum segment size and number of
732 	 * segments. */
733 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
734 				   struct virtio_blk_config, size_max, &v);
735 	if (!err)
736 		blk_queue_max_segment_size(q, v);
737 	else
738 		blk_queue_max_segment_size(q, -1U);
739 
740 	/* Host can optionally specify the block size of the device */
741 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
742 				   struct virtio_blk_config, blk_size,
743 				   &blk_size);
744 	if (!err)
745 		blk_queue_logical_block_size(q, blk_size);
746 	else
747 		blk_size = queue_logical_block_size(q);
748 
749 	/* Use topology information if available */
750 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
751 				   struct virtio_blk_config, physical_block_exp,
752 				   &physical_block_exp);
753 	if (!err && physical_block_exp)
754 		blk_queue_physical_block_size(q,
755 				blk_size * (1 << physical_block_exp));
756 
757 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
758 				   struct virtio_blk_config, alignment_offset,
759 				   &alignment_offset);
760 	if (!err && alignment_offset)
761 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
762 
763 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
764 				   struct virtio_blk_config, min_io_size,
765 				   &min_io_size);
766 	if (!err && min_io_size)
767 		blk_queue_io_min(q, blk_size * min_io_size);
768 
769 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
770 				   struct virtio_blk_config, opt_io_size,
771 				   &opt_io_size);
772 	if (!err && opt_io_size)
773 		blk_queue_io_opt(q, blk_size * opt_io_size);
774 
775 	virtio_device_ready(vdev);
776 
777 	device_add_disk(&vdev->dev, vblk->disk);
778 	err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
779 	if (err)
780 		goto out_del_disk;
781 
782 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
783 		err = device_create_file(disk_to_dev(vblk->disk),
784 					 &dev_attr_cache_type_rw);
785 	else
786 		err = device_create_file(disk_to_dev(vblk->disk),
787 					 &dev_attr_cache_type_ro);
788 	if (err)
789 		goto out_del_disk;
790 	return 0;
791 
792 out_del_disk:
793 	del_gendisk(vblk->disk);
794 	blk_cleanup_queue(vblk->disk->queue);
795 out_free_tags:
796 	blk_mq_free_tag_set(&vblk->tag_set);
797 out_put_disk:
798 	put_disk(vblk->disk);
799 out_free_vq:
800 	vdev->config->del_vqs(vdev);
801 out_free_vblk:
802 	kfree(vblk);
803 out_free_index:
804 	ida_simple_remove(&vd_index_ida, index);
805 out:
806 	return err;
807 }
808 
809 static void virtblk_remove(struct virtio_device *vdev)
810 {
811 	struct virtio_blk *vblk = vdev->priv;
812 	int index = vblk->index;
813 	int refc;
814 
815 	/* Make sure no work handler is accessing the device. */
816 	flush_work(&vblk->config_work);
817 
818 	del_gendisk(vblk->disk);
819 	blk_cleanup_queue(vblk->disk->queue);
820 
821 	blk_mq_free_tag_set(&vblk->tag_set);
822 
823 	/* Stop all the virtqueues. */
824 	vdev->config->reset(vdev);
825 
826 	refc = kref_read(&disk_to_dev(vblk->disk)->kobj.kref);
827 	put_disk(vblk->disk);
828 	vdev->config->del_vqs(vdev);
829 	kfree(vblk->vqs);
830 	kfree(vblk);
831 
832 	/* Only free device id if we don't have any users */
833 	if (refc == 1)
834 		ida_simple_remove(&vd_index_ida, index);
835 }
836 
837 #ifdef CONFIG_PM_SLEEP
838 static int virtblk_freeze(struct virtio_device *vdev)
839 {
840 	struct virtio_blk *vblk = vdev->priv;
841 
842 	/* Ensure we don't receive any more interrupts */
843 	vdev->config->reset(vdev);
844 
845 	/* Make sure no work handler is accessing the device. */
846 	flush_work(&vblk->config_work);
847 
848 	blk_mq_stop_hw_queues(vblk->disk->queue);
849 
850 	vdev->config->del_vqs(vdev);
851 	return 0;
852 }
853 
854 static int virtblk_restore(struct virtio_device *vdev)
855 {
856 	struct virtio_blk *vblk = vdev->priv;
857 	int ret;
858 
859 	ret = init_vq(vdev->priv);
860 	if (ret)
861 		return ret;
862 
863 	virtio_device_ready(vdev);
864 
865 	blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
866 	return 0;
867 }
868 #endif
869 
870 static const struct virtio_device_id id_table[] = {
871 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
872 	{ 0 },
873 };
874 
875 static unsigned int features_legacy[] = {
876 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
877 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
878 #ifdef CONFIG_VIRTIO_BLK_SCSI
879 	VIRTIO_BLK_F_SCSI,
880 #endif
881 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
882 	VIRTIO_BLK_F_MQ,
883 }
884 ;
885 static unsigned int features[] = {
886 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
887 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
888 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
889 	VIRTIO_BLK_F_MQ,
890 };
891 
892 static struct virtio_driver virtio_blk = {
893 	.feature_table			= features,
894 	.feature_table_size		= ARRAY_SIZE(features),
895 	.feature_table_legacy		= features_legacy,
896 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
897 	.driver.name			= KBUILD_MODNAME,
898 	.driver.owner			= THIS_MODULE,
899 	.id_table			= id_table,
900 	.probe				= virtblk_probe,
901 	.remove				= virtblk_remove,
902 	.config_changed			= virtblk_config_changed,
903 #ifdef CONFIG_PM_SLEEP
904 	.freeze				= virtblk_freeze,
905 	.restore			= virtblk_restore,
906 #endif
907 };
908 
909 static int __init init(void)
910 {
911 	int error;
912 
913 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
914 	if (!virtblk_wq)
915 		return -ENOMEM;
916 
917 	major = register_blkdev(0, "virtblk");
918 	if (major < 0) {
919 		error = major;
920 		goto out_destroy_workqueue;
921 	}
922 
923 	error = register_virtio_driver(&virtio_blk);
924 	if (error)
925 		goto out_unregister_blkdev;
926 	return 0;
927 
928 out_unregister_blkdev:
929 	unregister_blkdev(major, "virtblk");
930 out_destroy_workqueue:
931 	destroy_workqueue(virtblk_wq);
932 	return error;
933 }
934 
935 static void __exit fini(void)
936 {
937 	unregister_virtio_driver(&virtio_blk);
938 	unregister_blkdev(major, "virtblk");
939 	destroy_workqueue(virtblk_wq);
940 }
941 module_init(init);
942 module_exit(fini);
943 
944 MODULE_DEVICE_TABLE(virtio, id_table);
945 MODULE_DESCRIPTION("Virtio block driver");
946 MODULE_LICENSE("GPL");
947