• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
20 static int major;
21 static DEFINE_IDA(vd_index_ida);
22 
23 static struct workqueue_struct *virtblk_wq;
24 
25 struct virtio_blk_vq {
26 	struct virtqueue *vq;
27 	spinlock_t lock;
28 	char name[VQ_NAME_LEN];
29 } ____cacheline_aligned_in_smp;
30 
31 struct 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 
54 struct virtblk_req {
55 	struct request *req;
56 	struct virtio_blk_outhdr out_hdr;
57 	struct virtio_scsi_inhdr in_hdr;
58 	u8 status;
59 	u8 sense[SCSI_SENSE_BUFFERSIZE];
60 	struct scatterlist sg[];
61 };
62 
virtblk_result(struct virtblk_req * vbr)63 static inline int virtblk_result(struct virtblk_req *vbr)
64 {
65 	switch (vbr->status) {
66 	case VIRTIO_BLK_S_OK:
67 		return 0;
68 	case VIRTIO_BLK_S_UNSUPP:
69 		return -ENOTTY;
70 	default:
71 		return -EIO;
72 	}
73 }
74 
__virtblk_add_req(struct virtqueue * vq,struct virtblk_req * vbr,struct scatterlist * data_sg,bool have_data)75 static int __virtblk_add_req(struct virtqueue *vq,
76 			     struct virtblk_req *vbr,
77 			     struct scatterlist *data_sg,
78 			     bool have_data)
79 {
80 	struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
81 	unsigned int num_out = 0, num_in = 0;
82 	__virtio32 type = vbr->out_hdr.type & ~cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT);
83 
84 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
85 	sgs[num_out++] = &hdr;
86 
87 	/*
88 	 * If this is a packet command we need a couple of additional headers.
89 	 * Behind the normal outhdr we put a segment with the scsi command
90 	 * block, and before the normal inhdr we put the sense data and the
91 	 * inhdr with additional status information.
92 	 */
93 	if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) {
94 		sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len);
95 		sgs[num_out++] = &cmd;
96 	}
97 
98 	if (have_data) {
99 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
100 			sgs[num_out++] = data_sg;
101 		else
102 			sgs[num_out + num_in++] = data_sg;
103 	}
104 
105 	if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) {
106 		memcpy(vbr->sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
107 		sg_init_one(&sense, vbr->sense, SCSI_SENSE_BUFFERSIZE);
108 		sgs[num_out + num_in++] = &sense;
109 		sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
110 		sgs[num_out + num_in++] = &inhdr;
111 	}
112 
113 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
114 	sgs[num_out + num_in++] = &status;
115 
116 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
117 }
118 
virtblk_request_done(struct request * req)119 static inline void virtblk_request_done(struct request *req)
120 {
121 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
122 	struct virtio_blk *vblk = req->q->queuedata;
123 	int error = virtblk_result(vbr);
124 
125 	if (req->cmd_type == REQ_TYPE_BLOCK_PC) {
126 		req->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual);
127 		req->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len);
128 		req->errors = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors);
129 	} else if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
130 		req->errors = (error != 0);
131 	}
132 
133 	blk_mq_end_request(req, error);
134 }
135 
virtblk_done(struct virtqueue * vq)136 static void virtblk_done(struct virtqueue *vq)
137 {
138 	struct virtio_blk *vblk = vq->vdev->priv;
139 	bool req_done = false;
140 	int qid = vq->index;
141 	struct virtblk_req *vbr;
142 	unsigned long flags;
143 	unsigned int len;
144 
145 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
146 	do {
147 		virtqueue_disable_cb(vq);
148 		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
149 			blk_mq_complete_request(vbr->req, vbr->req->errors);
150 			req_done = true;
151 		}
152 		if (unlikely(virtqueue_is_broken(vq)))
153 			break;
154 	} while (!virtqueue_enable_cb(vq));
155 
156 	/* In case queue is stopped waiting for more buffers. */
157 	if (req_done)
158 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
159 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
160 }
161 
virtio_queue_rq(struct blk_mq_hw_ctx * hctx,const struct blk_mq_queue_data * bd)162 static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
163 			   const struct blk_mq_queue_data *bd)
164 {
165 	struct virtio_blk *vblk = hctx->queue->queuedata;
166 	struct request *req = bd->rq;
167 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
168 	unsigned long flags;
169 	unsigned int num;
170 	int qid = hctx->queue_num;
171 	int err;
172 	bool notify = false;
173 
174 	BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
175 
176 	vbr->req = req;
177 	if (req_op(req) == REQ_OP_FLUSH) {
178 		vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_FLUSH);
179 		vbr->out_hdr.sector = 0;
180 		vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
181 	} else {
182 		switch (req->cmd_type) {
183 		case REQ_TYPE_FS:
184 			vbr->out_hdr.type = 0;
185 			vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, blk_rq_pos(vbr->req));
186 			vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
187 			break;
188 		case REQ_TYPE_BLOCK_PC:
189 			vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_SCSI_CMD);
190 			vbr->out_hdr.sector = 0;
191 			vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
192 			break;
193 		case REQ_TYPE_DRV_PRIV:
194 			vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
195 			vbr->out_hdr.sector = 0;
196 			vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
197 			break;
198 		default:
199 			/* We don't put anything else in the queue. */
200 			BUG();
201 		}
202 	}
203 
204 	blk_mq_start_request(req);
205 
206 	num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg);
207 	if (num) {
208 		if (rq_data_dir(vbr->req) == WRITE)
209 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
210 		else
211 			vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
212 	}
213 
214 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
215 	err = __virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
216 	if (err) {
217 		virtqueue_kick(vblk->vqs[qid].vq);
218 		blk_mq_stop_hw_queue(hctx);
219 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
220 		/* Out of mem doesn't actually happen, since we fall back
221 		 * to direct descriptors */
222 		if (err == -ENOMEM || err == -ENOSPC)
223 			return BLK_MQ_RQ_QUEUE_BUSY;
224 		return BLK_MQ_RQ_QUEUE_ERROR;
225 	}
226 
227 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
228 		notify = true;
229 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
230 
231 	if (notify)
232 		virtqueue_notify(vblk->vqs[qid].vq);
233 	return BLK_MQ_RQ_QUEUE_OK;
234 }
235 
236 /* return id (s/n) string for *disk to *id_str
237  */
virtblk_get_id(struct gendisk * disk,char * id_str)238 static int virtblk_get_id(struct gendisk *disk, char *id_str)
239 {
240 	struct virtio_blk *vblk = disk->private_data;
241 	struct request_queue *q = vblk->disk->queue;
242 	struct request *req;
243 	int err;
244 
245 	req = blk_get_request(q, READ, GFP_KERNEL);
246 	if (IS_ERR(req))
247 		return PTR_ERR(req);
248 	req->cmd_type = REQ_TYPE_DRV_PRIV;
249 
250 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
251 	if (err)
252 		goto out;
253 
254 	err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
255 out:
256 	blk_put_request(req);
257 	return err;
258 }
259 
virtblk_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long data)260 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
261 			     unsigned int cmd, unsigned long data)
262 {
263 	struct gendisk *disk = bdev->bd_disk;
264 	struct virtio_blk *vblk = disk->private_data;
265 
266 	/*
267 	 * Only allow the generic SCSI ioctls if the host can support it.
268 	 */
269 	if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
270 		return -ENOTTY;
271 
272 	return scsi_cmd_blk_ioctl(bdev, mode, cmd,
273 				  (void __user *)data);
274 }
275 
276 /* We provide getgeo only to please some old bootloader/partitioning tools */
virtblk_getgeo(struct block_device * bd,struct hd_geometry * geo)277 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
278 {
279 	struct virtio_blk *vblk = bd->bd_disk->private_data;
280 
281 	/* see if the host passed in geometry config */
282 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
283 		virtio_cread(vblk->vdev, struct virtio_blk_config,
284 			     geometry.cylinders, &geo->cylinders);
285 		virtio_cread(vblk->vdev, struct virtio_blk_config,
286 			     geometry.heads, &geo->heads);
287 		virtio_cread(vblk->vdev, struct virtio_blk_config,
288 			     geometry.sectors, &geo->sectors);
289 	} else {
290 		/* some standard values, similar to sd */
291 		geo->heads = 1 << 6;
292 		geo->sectors = 1 << 5;
293 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
294 	}
295 	return 0;
296 }
297 
298 static const struct block_device_operations virtblk_fops = {
299 	.ioctl  = virtblk_ioctl,
300 	.owner  = THIS_MODULE,
301 	.getgeo = virtblk_getgeo,
302 };
303 
index_to_minor(int index)304 static int index_to_minor(int index)
305 {
306 	return index << PART_BITS;
307 }
308 
minor_to_index(int minor)309 static int minor_to_index(int minor)
310 {
311 	return minor >> PART_BITS;
312 }
313 
virtblk_serial_show(struct device * dev,struct device_attribute * attr,char * buf)314 static ssize_t virtblk_serial_show(struct device *dev,
315 				struct device_attribute *attr, char *buf)
316 {
317 	struct gendisk *disk = dev_to_disk(dev);
318 	int err;
319 
320 	/* sysfs gives us a PAGE_SIZE buffer */
321 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
322 
323 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
324 	err = virtblk_get_id(disk, buf);
325 	if (!err)
326 		return strlen(buf);
327 
328 	if (err == -EIO) /* Unsupported? Make it empty. */
329 		return 0;
330 
331 	return err;
332 }
333 
334 static DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
335 
virtblk_config_changed_work(struct work_struct * work)336 static void virtblk_config_changed_work(struct work_struct *work)
337 {
338 	struct virtio_blk *vblk =
339 		container_of(work, struct virtio_blk, config_work);
340 	struct virtio_device *vdev = vblk->vdev;
341 	struct request_queue *q = vblk->disk->queue;
342 	char cap_str_2[10], cap_str_10[10];
343 	char *envp[] = { "RESIZE=1", NULL };
344 	u64 capacity;
345 
346 	/* Host must always specify the capacity. */
347 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
348 
349 	/* If capacity is too big, truncate with warning. */
350 	if ((sector_t)capacity != capacity) {
351 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
352 			 (unsigned long long)capacity);
353 		capacity = (sector_t)-1;
354 	}
355 
356 	string_get_size(capacity, queue_logical_block_size(q),
357 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
358 	string_get_size(capacity, queue_logical_block_size(q),
359 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
360 
361 	dev_notice(&vdev->dev,
362 		  "new size: %llu %d-byte logical blocks (%s/%s)\n",
363 		  (unsigned long long)capacity,
364 		  queue_logical_block_size(q),
365 		  cap_str_10, cap_str_2);
366 
367 	set_capacity(vblk->disk, capacity);
368 	revalidate_disk(vblk->disk);
369 	kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
370 }
371 
virtblk_config_changed(struct virtio_device * vdev)372 static void virtblk_config_changed(struct virtio_device *vdev)
373 {
374 	struct virtio_blk *vblk = vdev->priv;
375 
376 	queue_work(virtblk_wq, &vblk->config_work);
377 }
378 
init_vq(struct virtio_blk * vblk)379 static int init_vq(struct virtio_blk *vblk)
380 {
381 	int err;
382 	int i;
383 	vq_callback_t **callbacks;
384 	const char **names;
385 	struct virtqueue **vqs;
386 	unsigned short num_vqs;
387 	struct virtio_device *vdev = vblk->vdev;
388 
389 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
390 				   struct virtio_blk_config, num_queues,
391 				   &num_vqs);
392 	if (err)
393 		num_vqs = 1;
394 
395 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
396 	if (!vblk->vqs)
397 		return -ENOMEM;
398 
399 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
400 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
401 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
402 	if (!names || !callbacks || !vqs) {
403 		err = -ENOMEM;
404 		goto out;
405 	}
406 
407 	for (i = 0; i < num_vqs; i++) {
408 		callbacks[i] = virtblk_done;
409 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
410 		names[i] = vblk->vqs[i].name;
411 	}
412 
413 	/* Discover virtqueues and write information to configuration.  */
414 	err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names);
415 	if (err)
416 		goto out;
417 
418 	for (i = 0; i < num_vqs; i++) {
419 		spin_lock_init(&vblk->vqs[i].lock);
420 		vblk->vqs[i].vq = vqs[i];
421 	}
422 	vblk->num_vqs = num_vqs;
423 
424 out:
425 	kfree(vqs);
426 	kfree(callbacks);
427 	kfree(names);
428 	if (err)
429 		kfree(vblk->vqs);
430 	return err;
431 }
432 
433 /*
434  * Legacy naming scheme used for virtio devices.  We are stuck with it for
435  * virtio blk but don't ever use it for any new driver.
436  */
virtblk_name_format(char * prefix,int index,char * buf,int buflen)437 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
438 {
439 	const int base = 'z' - 'a' + 1;
440 	char *begin = buf + strlen(prefix);
441 	char *end = buf + buflen;
442 	char *p;
443 	int unit;
444 
445 	p = end - 1;
446 	*p = '\0';
447 	unit = base;
448 	do {
449 		if (p == begin)
450 			return -EINVAL;
451 		*--p = 'a' + (index % unit);
452 		index = (index / unit) - 1;
453 	} while (index >= 0);
454 
455 	memmove(begin, p, end - p);
456 	memcpy(buf, prefix, strlen(prefix));
457 
458 	return 0;
459 }
460 
virtblk_get_cache_mode(struct virtio_device * vdev)461 static int virtblk_get_cache_mode(struct virtio_device *vdev)
462 {
463 	u8 writeback;
464 	int err;
465 
466 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
467 				   struct virtio_blk_config, wce,
468 				   &writeback);
469 
470 	/*
471 	 * If WCE is not configurable and flush is not available,
472 	 * assume no writeback cache is in use.
473 	 */
474 	if (err)
475 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
476 
477 	return writeback;
478 }
479 
virtblk_update_cache_mode(struct virtio_device * vdev)480 static void virtblk_update_cache_mode(struct virtio_device *vdev)
481 {
482 	u8 writeback = virtblk_get_cache_mode(vdev);
483 	struct virtio_blk *vblk = vdev->priv;
484 
485 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
486 	revalidate_disk(vblk->disk);
487 }
488 
489 static const char *const virtblk_cache_types[] = {
490 	"write through", "write back"
491 };
492 
493 static ssize_t
virtblk_cache_type_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)494 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
495 			 const char *buf, size_t count)
496 {
497 	struct gendisk *disk = dev_to_disk(dev);
498 	struct virtio_blk *vblk = disk->private_data;
499 	struct virtio_device *vdev = vblk->vdev;
500 	int i;
501 
502 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
503 	for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
504 		if (sysfs_streq(buf, virtblk_cache_types[i]))
505 			break;
506 
507 	if (i < 0)
508 		return -EINVAL;
509 
510 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
511 	virtblk_update_cache_mode(vdev);
512 	return count;
513 }
514 
515 static ssize_t
virtblk_cache_type_show(struct device * dev,struct device_attribute * attr,char * buf)516 virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
517 			 char *buf)
518 {
519 	struct gendisk *disk = dev_to_disk(dev);
520 	struct virtio_blk *vblk = disk->private_data;
521 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
522 
523 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
524 	return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
525 }
526 
527 static const struct device_attribute dev_attr_cache_type_ro =
528 	__ATTR(cache_type, S_IRUGO,
529 	       virtblk_cache_type_show, NULL);
530 static const struct device_attribute dev_attr_cache_type_rw =
531 	__ATTR(cache_type, S_IRUGO|S_IWUSR,
532 	       virtblk_cache_type_show, virtblk_cache_type_store);
533 
virtblk_init_request(void * data,struct request * rq,unsigned int hctx_idx,unsigned int request_idx,unsigned int numa_node)534 static int virtblk_init_request(void *data, struct request *rq,
535 		unsigned int hctx_idx, unsigned int request_idx,
536 		unsigned int numa_node)
537 {
538 	struct virtio_blk *vblk = data;
539 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
540 
541 	sg_init_table(vbr->sg, vblk->sg_elems);
542 	return 0;
543 }
544 
545 static struct blk_mq_ops virtio_mq_ops = {
546 	.queue_rq	= virtio_queue_rq,
547 	.complete	= virtblk_request_done,
548 	.init_request	= virtblk_init_request,
549 };
550 
551 static unsigned int virtblk_queue_depth;
552 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
553 
virtblk_probe(struct virtio_device * vdev)554 static int virtblk_probe(struct virtio_device *vdev)
555 {
556 	struct virtio_blk *vblk;
557 	struct request_queue *q;
558 	int err, index;
559 
560 	u64 cap;
561 	u32 v, blk_size, sg_elems, opt_io_size;
562 	u16 min_io_size;
563 	u8 physical_block_exp, alignment_offset;
564 
565 	if (!vdev->config->get) {
566 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
567 			__func__);
568 		return -EINVAL;
569 	}
570 
571 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
572 			     GFP_KERNEL);
573 	if (err < 0)
574 		goto out;
575 	index = err;
576 
577 	/* We need to know how many segments before we allocate. */
578 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
579 				   struct virtio_blk_config, seg_max,
580 				   &sg_elems);
581 
582 	/* We need at least one SG element, whatever they say. */
583 	if (err || !sg_elems)
584 		sg_elems = 1;
585 
586 	/* We need an extra sg elements at head and tail. */
587 	sg_elems += 2;
588 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
589 	if (!vblk) {
590 		err = -ENOMEM;
591 		goto out_free_index;
592 	}
593 
594 	vblk->vdev = vdev;
595 	vblk->sg_elems = sg_elems;
596 
597 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
598 
599 	err = init_vq(vblk);
600 	if (err)
601 		goto out_free_vblk;
602 
603 	/* FIXME: How many partitions?  How long is a piece of string? */
604 	vblk->disk = alloc_disk(1 << PART_BITS);
605 	if (!vblk->disk) {
606 		err = -ENOMEM;
607 		goto out_free_vq;
608 	}
609 
610 	/* Default queue sizing is to fill the ring. */
611 	if (!virtblk_queue_depth) {
612 		virtblk_queue_depth = vblk->vqs[0].vq->num_free;
613 		/* ... but without indirect descs, we use 2 descs per req */
614 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
615 			virtblk_queue_depth /= 2;
616 	}
617 
618 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
619 	vblk->tag_set.ops = &virtio_mq_ops;
620 	vblk->tag_set.queue_depth = virtblk_queue_depth;
621 	vblk->tag_set.numa_node = NUMA_NO_NODE;
622 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
623 	vblk->tag_set.cmd_size =
624 		sizeof(struct virtblk_req) +
625 		sizeof(struct scatterlist) * sg_elems;
626 	vblk->tag_set.driver_data = vblk;
627 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
628 
629 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
630 	if (err)
631 		goto out_put_disk;
632 
633 	q = blk_mq_init_queue(&vblk->tag_set);
634 	if (IS_ERR(q)) {
635 		err = -ENOMEM;
636 		goto out_free_tags;
637 	}
638 	vblk->disk->queue = q;
639 
640 	q->queuedata = vblk;
641 
642 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
643 
644 	vblk->disk->major = major;
645 	vblk->disk->first_minor = index_to_minor(index);
646 	vblk->disk->private_data = vblk;
647 	vblk->disk->fops = &virtblk_fops;
648 	vblk->disk->flags |= GENHD_FL_EXT_DEVT;
649 	vblk->index = index;
650 
651 	/* configure queue flush support */
652 	virtblk_update_cache_mode(vdev);
653 
654 	/* If disk is read-only in the host, the guest should obey */
655 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
656 		set_disk_ro(vblk->disk, 1);
657 
658 	/* Host must always specify the capacity. */
659 	virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
660 
661 	/* If capacity is too big, truncate with warning. */
662 	if ((sector_t)cap != cap) {
663 		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
664 			 (unsigned long long)cap);
665 		cap = (sector_t)-1;
666 	}
667 	set_capacity(vblk->disk, cap);
668 
669 	/* We can handle whatever the host told us to handle. */
670 	blk_queue_max_segments(q, vblk->sg_elems-2);
671 
672 	/* No need to bounce any requests */
673 	blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
674 
675 	/* No real sector limit. */
676 	blk_queue_max_hw_sectors(q, -1U);
677 
678 	/* Host can optionally specify maximum segment size and number of
679 	 * segments. */
680 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
681 				   struct virtio_blk_config, size_max, &v);
682 	if (!err)
683 		blk_queue_max_segment_size(q, v);
684 	else
685 		blk_queue_max_segment_size(q, -1U);
686 
687 	/* Host can optionally specify the block size of the device */
688 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
689 				   struct virtio_blk_config, blk_size,
690 				   &blk_size);
691 	if (!err)
692 		blk_queue_logical_block_size(q, blk_size);
693 	else
694 		blk_size = queue_logical_block_size(q);
695 
696 	/* Use topology information if available */
697 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
698 				   struct virtio_blk_config, physical_block_exp,
699 				   &physical_block_exp);
700 	if (!err && physical_block_exp)
701 		blk_queue_physical_block_size(q,
702 				blk_size * (1 << physical_block_exp));
703 
704 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
705 				   struct virtio_blk_config, alignment_offset,
706 				   &alignment_offset);
707 	if (!err && alignment_offset)
708 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
709 
710 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
711 				   struct virtio_blk_config, min_io_size,
712 				   &min_io_size);
713 	if (!err && min_io_size)
714 		blk_queue_io_min(q, blk_size * min_io_size);
715 
716 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
717 				   struct virtio_blk_config, opt_io_size,
718 				   &opt_io_size);
719 	if (!err && opt_io_size)
720 		blk_queue_io_opt(q, blk_size * opt_io_size);
721 
722 	virtio_device_ready(vdev);
723 
724 	device_add_disk(&vdev->dev, vblk->disk);
725 	err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
726 	if (err)
727 		goto out_del_disk;
728 
729 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
730 		err = device_create_file(disk_to_dev(vblk->disk),
731 					 &dev_attr_cache_type_rw);
732 	else
733 		err = device_create_file(disk_to_dev(vblk->disk),
734 					 &dev_attr_cache_type_ro);
735 	if (err)
736 		goto out_del_disk;
737 	return 0;
738 
739 out_del_disk:
740 	del_gendisk(vblk->disk);
741 	blk_cleanup_queue(vblk->disk->queue);
742 out_free_tags:
743 	blk_mq_free_tag_set(&vblk->tag_set);
744 out_put_disk:
745 	put_disk(vblk->disk);
746 out_free_vq:
747 	vdev->config->del_vqs(vdev);
748 out_free_vblk:
749 	kfree(vblk);
750 out_free_index:
751 	ida_simple_remove(&vd_index_ida, index);
752 out:
753 	return err;
754 }
755 
virtblk_remove(struct virtio_device * vdev)756 static void virtblk_remove(struct virtio_device *vdev)
757 {
758 	struct virtio_blk *vblk = vdev->priv;
759 	int index = vblk->index;
760 	int refc;
761 
762 	/* Make sure no work handler is accessing the device. */
763 	flush_work(&vblk->config_work);
764 
765 	del_gendisk(vblk->disk);
766 	blk_cleanup_queue(vblk->disk->queue);
767 
768 	blk_mq_free_tag_set(&vblk->tag_set);
769 
770 	/* Stop all the virtqueues. */
771 	vdev->config->reset(vdev);
772 
773 	refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount);
774 	put_disk(vblk->disk);
775 	vdev->config->del_vqs(vdev);
776 	kfree(vblk->vqs);
777 	kfree(vblk);
778 
779 	/* Only free device id if we don't have any users */
780 	if (refc == 1)
781 		ida_simple_remove(&vd_index_ida, index);
782 }
783 
784 #ifdef CONFIG_PM_SLEEP
virtblk_freeze(struct virtio_device * vdev)785 static int virtblk_freeze(struct virtio_device *vdev)
786 {
787 	struct virtio_blk *vblk = vdev->priv;
788 
789 	/* Ensure we don't receive any more interrupts */
790 	vdev->config->reset(vdev);
791 
792 	/* Make sure no work handler is accessing the device. */
793 	flush_work(&vblk->config_work);
794 
795 	blk_mq_stop_hw_queues(vblk->disk->queue);
796 
797 	vdev->config->del_vqs(vdev);
798 	return 0;
799 }
800 
virtblk_restore(struct virtio_device * vdev)801 static int virtblk_restore(struct virtio_device *vdev)
802 {
803 	struct virtio_blk *vblk = vdev->priv;
804 	int ret;
805 
806 	ret = init_vq(vdev->priv);
807 	if (ret)
808 		return ret;
809 
810 	virtio_device_ready(vdev);
811 
812 	blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
813 	return 0;
814 }
815 #endif
816 
817 static const struct virtio_device_id id_table[] = {
818 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
819 	{ 0 },
820 };
821 
822 static unsigned int features_legacy[] = {
823 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
824 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
825 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
826 	VIRTIO_BLK_F_MQ,
827 }
828 ;
829 static unsigned int features[] = {
830 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
831 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
832 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
833 	VIRTIO_BLK_F_MQ,
834 };
835 
836 static struct virtio_driver virtio_blk = {
837 	.feature_table			= features,
838 	.feature_table_size		= ARRAY_SIZE(features),
839 	.feature_table_legacy		= features_legacy,
840 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
841 	.driver.name			= KBUILD_MODNAME,
842 	.driver.owner			= THIS_MODULE,
843 	.id_table			= id_table,
844 	.probe				= virtblk_probe,
845 	.remove				= virtblk_remove,
846 	.config_changed			= virtblk_config_changed,
847 #ifdef CONFIG_PM_SLEEP
848 	.freeze				= virtblk_freeze,
849 	.restore			= virtblk_restore,
850 #endif
851 };
852 
init(void)853 static int __init init(void)
854 {
855 	int error;
856 
857 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
858 	if (!virtblk_wq)
859 		return -ENOMEM;
860 
861 	major = register_blkdev(0, "virtblk");
862 	if (major < 0) {
863 		error = major;
864 		goto out_destroy_workqueue;
865 	}
866 
867 	error = register_virtio_driver(&virtio_blk);
868 	if (error)
869 		goto out_unregister_blkdev;
870 	return 0;
871 
872 out_unregister_blkdev:
873 	unregister_blkdev(major, "virtblk");
874 out_destroy_workqueue:
875 	destroy_workqueue(virtblk_wq);
876 	return error;
877 }
878 
fini(void)879 static void __exit fini(void)
880 {
881 	unregister_virtio_driver(&virtio_blk);
882 	unregister_blkdev(major, "virtblk");
883 	destroy_workqueue(virtblk_wq);
884 }
885 module_init(init);
886 module_exit(fini);
887 
888 MODULE_DEVICE_TABLE(virtio, id_table);
889 MODULE_DESCRIPTION("Virtio block driver");
890 MODULE_LICENSE("GPL");
891