• Home
  • History
  • Annotate
  • only in this directory
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/virtio.h>
9#include <linux/virtio_blk.h>
10#include <linux/scatterlist.h>
11#include <linux/string_helpers.h>
12#include <scsi/scsi_cmnd.h>
13#include <linux/idr.h>
14#include <linux/blk-mq.h>
15#include <linux/numa.h>
16
17#define PART_BITS 4
18#define VQ_NAME_LEN 16
19
20static int major;
21static DEFINE_IDA(vd_index_ida);
22
23static struct workqueue_struct *virtblk_wq;
24
25struct virtio_blk_vq {
26	struct virtqueue *vq;
27	spinlock_t lock;
28	char name[VQ_NAME_LEN];
29} ____cacheline_aligned_in_smp;
30
31struct virtio_blk {
32	struct virtio_device *vdev;
33
34	/* The disk structure for the kernel. */
35	struct gendisk *disk;
36
37	/* Block layer tags. */
38	struct blk_mq_tag_set tag_set;
39
40	/* Process context for config space updates */
41	struct work_struct config_work;
42
43	/* What host tells us, plus 2 for header & tailer. */
44	unsigned int sg_elems;
45
46	/* Ida index - used to track minor number allocations. */
47	int index;
48
49	/* num of vqs */
50	int num_vqs;
51	struct virtio_blk_vq *vqs;
52};
53
54struct virtblk_req {
55	struct request *req;
56	struct virtio_blk_outhdr out_hdr;
57	struct virtio_scsi_inhdr in_hdr;
58	u8 status;
59	struct scatterlist sg[];
60};
61
62static inline int virtblk_result(struct virtblk_req *vbr)
63{
64	switch (vbr->status) {
65	case VIRTIO_BLK_S_OK:
66		return 0;
67	case VIRTIO_BLK_S_UNSUPP:
68		return -ENOTTY;
69	default:
70		return -EIO;
71	}
72}
73
74static int __virtblk_add_req(struct virtqueue *vq,
75			     struct virtblk_req *vbr,
76			     struct scatterlist *data_sg,
77			     bool have_data)
78{
79	struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
80	unsigned int num_out = 0, num_in = 0;
81	__virtio32 type = vbr->out_hdr.type & ~cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT);
82
83	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
84	sgs[num_out++] = &hdr;
85
86	/*
87	 * If this is a packet command we need a couple of additional headers.
88	 * Behind the normal outhdr we put a segment with the scsi command
89	 * block, and before the normal inhdr we put the sense data and the
90	 * inhdr with additional status information.
91	 */
92	if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) {
93		sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len);
94		sgs[num_out++] = &cmd;
95	}
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	if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) {
105		sg_init_one(&sense, vbr->req->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	}
110
111	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
112	sgs[num_out + num_in++] = &status;
113
114	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
115}
116
117static inline void virtblk_request_done(struct request *req)
118{
119	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
120	struct virtio_blk *vblk = req->q->queuedata;
121	int error = virtblk_result(vbr);
122
123	if (req->cmd_type == REQ_TYPE_BLOCK_PC) {
124		req->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual);
125		req->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len);
126		req->errors = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors);
127	} else if (req->cmd_type == REQ_TYPE_SPECIAL) {
128		req->errors = (error != 0);
129	}
130
131	blk_mq_end_request(req, error);
132}
133
134static void virtblk_done(struct virtqueue *vq)
135{
136	struct virtio_blk *vblk = vq->vdev->priv;
137	bool req_done = false;
138	int qid = vq->index;
139	struct virtblk_req *vbr;
140	unsigned long flags;
141	unsigned int len;
142
143	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
144	do {
145		virtqueue_disable_cb(vq);
146		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
147			blk_mq_complete_request(vbr->req);
148			req_done = true;
149		}
150		if (unlikely(virtqueue_is_broken(vq)))
151			break;
152	} while (!virtqueue_enable_cb(vq));
153
154	/* In case queue is stopped waiting for more buffers. */
155	if (req_done)
156		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
157	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
158}
159
160static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
161			   const struct blk_mq_queue_data *bd)
162{
163	struct virtio_blk *vblk = hctx->queue->queuedata;
164	struct request *req = bd->rq;
165	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
166	unsigned long flags;
167	unsigned int num;
168	int qid = hctx->queue_num;
169	int err;
170	bool notify = false;
171
172	BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
173
174	vbr->req = req;
175	if (req->cmd_flags & REQ_FLUSH) {
176		vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_FLUSH);
177		vbr->out_hdr.sector = 0;
178		vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
179	} else {
180		switch (req->cmd_type) {
181		case REQ_TYPE_FS:
182			vbr->out_hdr.type = 0;
183			vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, blk_rq_pos(vbr->req));
184			vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
185			break;
186		case REQ_TYPE_BLOCK_PC:
187			vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_SCSI_CMD);
188			vbr->out_hdr.sector = 0;
189			vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
190			break;
191		case REQ_TYPE_SPECIAL:
192			vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
193			vbr->out_hdr.sector = 0;
194			vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
195			break;
196		default:
197			/* We don't put anything else in the queue. */
198			BUG();
199		}
200	}
201
202	blk_mq_start_request(req);
203
204	num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg);
205	if (num) {
206		if (rq_data_dir(vbr->req) == WRITE)
207			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
208		else
209			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
210	}
211
212	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
213	err = __virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
214	if (err) {
215		virtqueue_kick(vblk->vqs[qid].vq);
216		blk_mq_stop_hw_queue(hctx);
217		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
218		/* Out of mem doesn't actually happen, since we fall back
219		 * to direct descriptors */
220		if (err == -ENOMEM || err == -ENOSPC)
221			return BLK_MQ_RQ_QUEUE_BUSY;
222		return BLK_MQ_RQ_QUEUE_ERROR;
223	}
224
225	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
226		notify = true;
227	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
228
229	if (notify)
230		virtqueue_notify(vblk->vqs[qid].vq);
231	return BLK_MQ_RQ_QUEUE_OK;
232}
233
234/* return id (s/n) string for *disk to *id_str
235 */
236static int virtblk_get_id(struct gendisk *disk, char *id_str)
237{
238	struct virtio_blk *vblk = disk->private_data;
239	struct request *req;
240	struct bio *bio;
241	int err;
242
243	bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
244			   GFP_KERNEL);
245	if (IS_ERR(bio))
246		return PTR_ERR(bio);
247
248	req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
249	if (IS_ERR(req)) {
250		bio_put(bio);
251		return PTR_ERR(req);
252	}
253
254	req->cmd_type = REQ_TYPE_SPECIAL;
255	err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
256	blk_put_request(req);
257
258	return err;
259}
260
261static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
262			     unsigned int cmd, unsigned long data)
263{
264	struct gendisk *disk = bdev->bd_disk;
265	struct virtio_blk *vblk = disk->private_data;
266
267	/*
268	 * Only allow the generic SCSI ioctls if the host can support it.
269	 */
270	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
271		return -ENOTTY;
272
273	return scsi_cmd_blk_ioctl(bdev, mode, cmd,
274				  (void __user *)data);
275}
276
277/* We provide getgeo only to please some old bootloader/partitioning tools */
278static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
279{
280	struct virtio_blk *vblk = bd->bd_disk->private_data;
281
282	/* see if the host passed in geometry config */
283	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
284		virtio_cread(vblk->vdev, struct virtio_blk_config,
285			     geometry.cylinders, &geo->cylinders);
286		virtio_cread(vblk->vdev, struct virtio_blk_config,
287			     geometry.heads, &geo->heads);
288		virtio_cread(vblk->vdev, struct virtio_blk_config,
289			     geometry.sectors, &geo->sectors);
290	} else {
291		/* some standard values, similar to sd */
292		geo->heads = 1 << 6;
293		geo->sectors = 1 << 5;
294		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
295	}
296	return 0;
297}
298
299static const struct block_device_operations virtblk_fops = {
300	.ioctl  = virtblk_ioctl,
301	.owner  = THIS_MODULE,
302	.getgeo = virtblk_getgeo,
303};
304
305static int index_to_minor(int index)
306{
307	return index << PART_BITS;
308}
309
310static int minor_to_index(int minor)
311{
312	return minor >> PART_BITS;
313}
314
315static ssize_t virtblk_serial_show(struct device *dev,
316				struct device_attribute *attr, char *buf)
317{
318	struct gendisk *disk = dev_to_disk(dev);
319	int err;
320
321	/* sysfs gives us a PAGE_SIZE buffer */
322	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
323
324	buf[VIRTIO_BLK_ID_BYTES] = '\0';
325	err = virtblk_get_id(disk, buf);
326	if (!err)
327		return strlen(buf);
328
329	if (err == -EIO) /* Unsupported? Make it empty. */
330		return 0;
331
332	return err;
333}
334
335static DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
336
337static void virtblk_config_changed_work(struct work_struct *work)
338{
339	struct virtio_blk *vblk =
340		container_of(work, struct virtio_blk, config_work);
341	struct virtio_device *vdev = vblk->vdev;
342	struct request_queue *q = vblk->disk->queue;
343	char cap_str_2[10], cap_str_10[10];
344	char *envp[] = { "RESIZE=1", NULL };
345	u64 capacity;
346
347	/* Host must always specify the capacity. */
348	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
349
350	/* If capacity is too big, truncate with warning. */
351	if ((sector_t)capacity != capacity) {
352		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
353			 (unsigned long long)capacity);
354		capacity = (sector_t)-1;
355	}
356
357	string_get_size(capacity, queue_logical_block_size(q),
358			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
359	string_get_size(capacity, queue_logical_block_size(q),
360			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
361
362	dev_notice(&vdev->dev,
363		  "new size: %llu %d-byte logical blocks (%s/%s)\n",
364		  (unsigned long long)capacity,
365		  queue_logical_block_size(q),
366		  cap_str_10, cap_str_2);
367
368	set_capacity(vblk->disk, capacity);
369	revalidate_disk(vblk->disk);
370	kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
371}
372
373static void virtblk_config_changed(struct virtio_device *vdev)
374{
375	struct virtio_blk *vblk = vdev->priv;
376
377	queue_work(virtblk_wq, &vblk->config_work);
378}
379
380static int init_vq(struct virtio_blk *vblk)
381{
382	int err = 0;
383	int i;
384	vq_callback_t **callbacks;
385	const char **names;
386	struct virtqueue **vqs;
387	unsigned short num_vqs;
388	struct virtio_device *vdev = vblk->vdev;
389
390	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
391				   struct virtio_blk_config, num_queues,
392				   &num_vqs);
393	if (err)
394		num_vqs = 1;
395
396	vblk->vqs = kmalloc(sizeof(*vblk->vqs) * num_vqs, GFP_KERNEL);
397	if (!vblk->vqs) {
398		err = -ENOMEM;
399		goto out;
400	}
401
402	names = kmalloc(sizeof(*names) * num_vqs, GFP_KERNEL);
403	if (!names)
404		goto err_names;
405
406	callbacks = kmalloc(sizeof(*callbacks) * num_vqs, GFP_KERNEL);
407	if (!callbacks)
408		goto err_callbacks;
409
410	vqs = kmalloc(sizeof(*vqs) * num_vqs, GFP_KERNEL);
411	if (!vqs)
412		goto err_vqs;
413
414	for (i = 0; i < num_vqs; i++) {
415		callbacks[i] = virtblk_done;
416		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
417		names[i] = vblk->vqs[i].name;
418	}
419
420	/* Discover virtqueues and write information to configuration.  */
421	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names);
422	if (err)
423		goto err_find_vqs;
424
425	for (i = 0; i < num_vqs; i++) {
426		spin_lock_init(&vblk->vqs[i].lock);
427		vblk->vqs[i].vq = vqs[i];
428	}
429	vblk->num_vqs = num_vqs;
430
431 err_find_vqs:
432	kfree(vqs);
433 err_vqs:
434	kfree(callbacks);
435 err_callbacks:
436	kfree(names);
437 err_names:
438	if (err)
439		kfree(vblk->vqs);
440 out:
441	return err;
442}
443
444/*
445 * Legacy naming scheme used for virtio devices.  We are stuck with it for
446 * virtio blk but don't ever use it for any new driver.
447 */
448static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
449{
450	const int base = 'z' - 'a' + 1;
451	char *begin = buf + strlen(prefix);
452	char *end = buf + buflen;
453	char *p;
454	int unit;
455
456	p = end - 1;
457	*p = '\0';
458	unit = base;
459	do {
460		if (p == begin)
461			return -EINVAL;
462		*--p = 'a' + (index % unit);
463		index = (index / unit) - 1;
464	} while (index >= 0);
465
466	memmove(begin, p, end - p);
467	memcpy(buf, prefix, strlen(prefix));
468
469	return 0;
470}
471
472static int virtblk_get_cache_mode(struct virtio_device *vdev)
473{
474	u8 writeback;
475	int err;
476
477	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
478				   struct virtio_blk_config, wce,
479				   &writeback);
480	if (err)
481		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE) ||
482		            virtio_has_feature(vdev, VIRTIO_F_VERSION_1);
483
484	return writeback;
485}
486
487static void virtblk_update_cache_mode(struct virtio_device *vdev)
488{
489	u8 writeback = virtblk_get_cache_mode(vdev);
490	struct virtio_blk *vblk = vdev->priv;
491
492	if (writeback)
493		blk_queue_flush(vblk->disk->queue, REQ_FLUSH);
494	else
495		blk_queue_flush(vblk->disk->queue, 0);
496
497	revalidate_disk(vblk->disk);
498}
499
500static const char *const virtblk_cache_types[] = {
501	"write through", "write back"
502};
503
504static ssize_t
505virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
506			 const char *buf, size_t count)
507{
508	struct gendisk *disk = dev_to_disk(dev);
509	struct virtio_blk *vblk = disk->private_data;
510	struct virtio_device *vdev = vblk->vdev;
511	int i;
512
513	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
514	for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
515		if (sysfs_streq(buf, virtblk_cache_types[i]))
516			break;
517
518	if (i < 0)
519		return -EINVAL;
520
521	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
522	virtblk_update_cache_mode(vdev);
523	return count;
524}
525
526static ssize_t
527virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
528			 char *buf)
529{
530	struct gendisk *disk = dev_to_disk(dev);
531	struct virtio_blk *vblk = disk->private_data;
532	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
533
534	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
535	return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
536}
537
538static const struct device_attribute dev_attr_cache_type_ro =
539	__ATTR(cache_type, S_IRUGO,
540	       virtblk_cache_type_show, NULL);
541static const struct device_attribute dev_attr_cache_type_rw =
542	__ATTR(cache_type, S_IRUGO|S_IWUSR,
543	       virtblk_cache_type_show, virtblk_cache_type_store);
544
545static int virtblk_init_request(void *data, struct request *rq,
546		unsigned int hctx_idx, unsigned int request_idx,
547		unsigned int numa_node)
548{
549	struct virtio_blk *vblk = data;
550	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
551
552	sg_init_table(vbr->sg, vblk->sg_elems);
553	return 0;
554}
555
556static struct blk_mq_ops virtio_mq_ops = {
557	.queue_rq	= virtio_queue_rq,
558	.map_queue	= blk_mq_map_queue,
559	.complete	= virtblk_request_done,
560	.init_request	= virtblk_init_request,
561};
562
563static unsigned int virtblk_queue_depth;
564module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
565
566static int virtblk_probe(struct virtio_device *vdev)
567{
568	struct virtio_blk *vblk;
569	struct request_queue *q;
570	int err, index;
571
572	u64 cap;
573	u32 v, blk_size, sg_elems, opt_io_size;
574	u16 min_io_size;
575	u8 physical_block_exp, alignment_offset;
576
577	if (!vdev->config->get) {
578		dev_err(&vdev->dev, "%s failure: config access disabled\n",
579			__func__);
580		return -EINVAL;
581	}
582
583	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
584			     GFP_KERNEL);
585	if (err < 0)
586		goto out;
587	index = err;
588
589	/* We need to know how many segments before we allocate. */
590	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
591				   struct virtio_blk_config, seg_max,
592				   &sg_elems);
593
594	/* We need at least one SG element, whatever they say. */
595	if (err || !sg_elems)
596		sg_elems = 1;
597
598	/* We need an extra sg elements at head and tail. */
599	sg_elems += 2;
600	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
601	if (!vblk) {
602		err = -ENOMEM;
603		goto out_free_index;
604	}
605
606	vblk->vdev = vdev;
607	vblk->sg_elems = sg_elems;
608
609	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
610
611	err = init_vq(vblk);
612	if (err)
613		goto out_free_vblk;
614
615	/* FIXME: How many partitions?  How long is a piece of string? */
616	vblk->disk = alloc_disk(1 << PART_BITS);
617	if (!vblk->disk) {
618		err = -ENOMEM;
619		goto out_free_vq;
620	}
621
622	/* Default queue sizing is to fill the ring. */
623	if (!virtblk_queue_depth) {
624		virtblk_queue_depth = vblk->vqs[0].vq->num_free;
625		/* ... but without indirect descs, we use 2 descs per req */
626		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
627			virtblk_queue_depth /= 2;
628	}
629
630	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
631	vblk->tag_set.ops = &virtio_mq_ops;
632	vblk->tag_set.queue_depth = virtblk_queue_depth;
633	vblk->tag_set.numa_node = NUMA_NO_NODE;
634	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
635	vblk->tag_set.cmd_size =
636		sizeof(struct virtblk_req) +
637		sizeof(struct scatterlist) * sg_elems;
638	vblk->tag_set.driver_data = vblk;
639	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
640
641	err = blk_mq_alloc_tag_set(&vblk->tag_set);
642	if (err)
643		goto out_put_disk;
644
645	q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set);
646	if (IS_ERR(q)) {
647		err = -ENOMEM;
648		goto out_free_tags;
649	}
650
651	q->queuedata = vblk;
652
653	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
654
655	vblk->disk->major = major;
656	vblk->disk->first_minor = index_to_minor(index);
657	vblk->disk->private_data = vblk;
658	vblk->disk->fops = &virtblk_fops;
659	vblk->disk->driverfs_dev = &vdev->dev;
660	vblk->index = index;
661
662	/* configure queue flush support */
663	virtblk_update_cache_mode(vdev);
664
665	/* If disk is read-only in the host, the guest should obey */
666	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
667		set_disk_ro(vblk->disk, 1);
668
669	/* Host must always specify the capacity. */
670	virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
671
672	/* If capacity is too big, truncate with warning. */
673	if ((sector_t)cap != cap) {
674		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
675			 (unsigned long long)cap);
676		cap = (sector_t)-1;
677	}
678	set_capacity(vblk->disk, cap);
679
680	/* We can handle whatever the host told us to handle. */
681	blk_queue_max_segments(q, vblk->sg_elems-2);
682
683	/* No need to bounce any requests */
684	blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
685
686	/* No real sector limit. */
687	blk_queue_max_hw_sectors(q, -1U);
688
689	/* Host can optionally specify maximum segment size and number of
690	 * segments. */
691	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
692				   struct virtio_blk_config, size_max, &v);
693	if (!err)
694		blk_queue_max_segment_size(q, v);
695	else
696		blk_queue_max_segment_size(q, -1U);
697
698	/* Host can optionally specify the block size of the device */
699	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
700				   struct virtio_blk_config, blk_size,
701				   &blk_size);
702	if (!err)
703		blk_queue_logical_block_size(q, blk_size);
704	else
705		blk_size = queue_logical_block_size(q);
706
707	/* Use topology information if available */
708	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
709				   struct virtio_blk_config, physical_block_exp,
710				   &physical_block_exp);
711	if (!err && physical_block_exp)
712		blk_queue_physical_block_size(q,
713				blk_size * (1 << physical_block_exp));
714
715	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
716				   struct virtio_blk_config, alignment_offset,
717				   &alignment_offset);
718	if (!err && alignment_offset)
719		blk_queue_alignment_offset(q, blk_size * alignment_offset);
720
721	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
722				   struct virtio_blk_config, min_io_size,
723				   &min_io_size);
724	if (!err && min_io_size)
725		blk_queue_io_min(q, blk_size * min_io_size);
726
727	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
728				   struct virtio_blk_config, opt_io_size,
729				   &opt_io_size);
730	if (!err && opt_io_size)
731		blk_queue_io_opt(q, blk_size * opt_io_size);
732
733	virtio_device_ready(vdev);
734
735	add_disk(vblk->disk);
736	err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
737	if (err)
738		goto out_del_disk;
739
740	if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
741		err = device_create_file(disk_to_dev(vblk->disk),
742					 &dev_attr_cache_type_rw);
743	else
744		err = device_create_file(disk_to_dev(vblk->disk),
745					 &dev_attr_cache_type_ro);
746	if (err)
747		goto out_del_disk;
748	return 0;
749
750out_del_disk:
751	del_gendisk(vblk->disk);
752	blk_cleanup_queue(vblk->disk->queue);
753out_free_tags:
754	blk_mq_free_tag_set(&vblk->tag_set);
755out_put_disk:
756	put_disk(vblk->disk);
757out_free_vq:
758	vdev->config->del_vqs(vdev);
759out_free_vblk:
760	kfree(vblk);
761out_free_index:
762	ida_simple_remove(&vd_index_ida, index);
763out:
764	return err;
765}
766
767static void virtblk_remove(struct virtio_device *vdev)
768{
769	struct virtio_blk *vblk = vdev->priv;
770	int index = vblk->index;
771	int refc;
772
773	/* Make sure no work handler is accessing the device. */
774	flush_work(&vblk->config_work);
775
776	del_gendisk(vblk->disk);
777	blk_cleanup_queue(vblk->disk->queue);
778
779	blk_mq_free_tag_set(&vblk->tag_set);
780
781	/* Stop all the virtqueues. */
782	vdev->config->reset(vdev);
783
784	refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount);
785	put_disk(vblk->disk);
786	vdev->config->del_vqs(vdev);
787	kfree(vblk->vqs);
788	kfree(vblk);
789
790	/* Only free device id if we don't have any users */
791	if (refc == 1)
792		ida_simple_remove(&vd_index_ida, index);
793}
794
795#ifdef CONFIG_PM_SLEEP
796static int virtblk_freeze(struct virtio_device *vdev)
797{
798	struct virtio_blk *vblk = vdev->priv;
799
800	/* Ensure we don't receive any more interrupts */
801	vdev->config->reset(vdev);
802
803	/* Make sure no work handler is accessing the device. */
804	flush_work(&vblk->config_work);
805
806	blk_mq_stop_hw_queues(vblk->disk->queue);
807
808	vdev->config->del_vqs(vdev);
809	return 0;
810}
811
812static int virtblk_restore(struct virtio_device *vdev)
813{
814	struct virtio_blk *vblk = vdev->priv;
815	int ret;
816
817	ret = init_vq(vdev->priv);
818	if (ret)
819		return ret;
820
821	virtio_device_ready(vdev);
822
823	blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
824	return 0;
825}
826#endif
827
828static const struct virtio_device_id id_table[] = {
829	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
830	{ 0 },
831};
832
833static unsigned int features_legacy[] = {
834	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
835	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
836	VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
837	VIRTIO_BLK_F_MQ,
838}
839;
840static unsigned int features[] = {
841	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
842	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
843	VIRTIO_BLK_F_TOPOLOGY,
844	VIRTIO_BLK_F_MQ,
845};
846
847static struct virtio_driver virtio_blk = {
848	.feature_table			= features,
849	.feature_table_size		= ARRAY_SIZE(features),
850	.feature_table_legacy		= features_legacy,
851	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
852	.driver.name			= KBUILD_MODNAME,
853	.driver.owner			= THIS_MODULE,
854	.id_table			= id_table,
855	.probe				= virtblk_probe,
856	.remove				= virtblk_remove,
857	.config_changed			= virtblk_config_changed,
858#ifdef CONFIG_PM_SLEEP
859	.freeze				= virtblk_freeze,
860	.restore			= virtblk_restore,
861#endif
862};
863
864static int __init init(void)
865{
866	int error;
867
868	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
869	if (!virtblk_wq)
870		return -ENOMEM;
871
872	major = register_blkdev(0, "virtblk");
873	if (major < 0) {
874		error = major;
875		goto out_destroy_workqueue;
876	}
877
878	error = register_virtio_driver(&virtio_blk);
879	if (error)
880		goto out_unregister_blkdev;
881	return 0;
882
883out_unregister_blkdev:
884	unregister_blkdev(major, "virtblk");
885out_destroy_workqueue:
886	destroy_workqueue(virtblk_wq);
887	return error;
888}
889
890static void __exit fini(void)
891{
892	unregister_virtio_driver(&virtio_blk);
893	unregister_blkdev(major, "virtblk");
894	destroy_workqueue(virtblk_wq);
895}
896module_init(init);
897module_exit(fini);
898
899MODULE_DEVICE_TABLE(virtio, id_table);
900MODULE_DESCRIPTION("Virtio block driver");
901MODULE_LICENSE("GPL");
902