• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 //#define DEBUG
3 #include <linux/spinlock.h>
4 #include <linux/slab.h>
5 #include <linux/blkdev.h>
6 #include <linux/hdreg.h>
7 #include <linux/module.h>
8 #include <linux/mutex.h>
9 #include <linux/interrupt.h>
10 #include <linux/virtio.h>
11 #include <linux/virtio_blk.h>
12 #include <linux/scatterlist.h>
13 #include <linux/string_helpers.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 #include <uapi/linux/virtio_ring.h>
19 
20 #define PART_BITS 4
21 #define VQ_NAME_LEN 16
22 #define MAX_DISCARD_SEGMENTS 256u
23 
24 /* The maximum number of sg elements that fit into a virtqueue */
25 #define VIRTIO_BLK_MAX_SG_ELEMS 32768
26 
27 #ifdef CONFIG_ARCH_NO_SG_CHAIN
28 #define VIRTIO_BLK_INLINE_SG_CNT	0
29 #else
30 #define VIRTIO_BLK_INLINE_SG_CNT	2
31 #endif
32 
33 static int major;
34 static DEFINE_IDA(vd_index_ida);
35 
36 static struct workqueue_struct *virtblk_wq;
37 
38 struct virtio_blk_vq {
39 	struct virtqueue *vq;
40 	spinlock_t lock;
41 	char name[VQ_NAME_LEN];
42 } ____cacheline_aligned_in_smp;
43 
44 struct virtio_blk {
45 	/*
46 	 * This mutex must be held by anything that may run after
47 	 * virtblk_remove() sets vblk->vdev to NULL.
48 	 *
49 	 * blk-mq, virtqueue processing, and sysfs attribute code paths are
50 	 * shut down before vblk->vdev is set to NULL and therefore do not need
51 	 * to hold this mutex.
52 	 */
53 	struct mutex vdev_mutex;
54 	struct virtio_device *vdev;
55 
56 	/* The disk structure for the kernel. */
57 	struct gendisk *disk;
58 
59 	/* Block layer tags. */
60 	struct blk_mq_tag_set tag_set;
61 
62 	/* Process context for config space updates */
63 	struct work_struct config_work;
64 
65 	/*
66 	 * Tracks references from block_device_operations open/release and
67 	 * virtio_driver probe/remove so this object can be freed once no
68 	 * longer in use.
69 	 */
70 	refcount_t refs;
71 
72 	/* What host tells us, plus 2 for header & tailer. */
73 	unsigned int sg_elems;
74 
75 	/* Ida index - used to track minor number allocations. */
76 	int index;
77 
78 	/* num of vqs */
79 	int num_vqs;
80 	struct virtio_blk_vq *vqs;
81 };
82 
83 struct virtblk_req {
84 	struct virtio_blk_outhdr out_hdr;
85 	u8 status;
86 	struct sg_table sg_table;
87 	struct scatterlist sg[];
88 };
89 
virtblk_result(struct virtblk_req * vbr)90 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
91 {
92 	switch (vbr->status) {
93 	case VIRTIO_BLK_S_OK:
94 		return BLK_STS_OK;
95 	case VIRTIO_BLK_S_UNSUPP:
96 		return BLK_STS_NOTSUPP;
97 	default:
98 		return BLK_STS_IOERR;
99 	}
100 }
101 
virtblk_add_req(struct virtqueue * vq,struct virtblk_req * vbr,struct scatterlist * data_sg,bool have_data)102 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
103 		struct scatterlist *data_sg, bool have_data)
104 {
105 	struct scatterlist hdr, status, *sgs[3];
106 	unsigned int num_out = 0, num_in = 0;
107 
108 	sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
109 	sgs[num_out++] = &hdr;
110 
111 	if (have_data) {
112 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
113 			sgs[num_out++] = data_sg;
114 		else
115 			sgs[num_out + num_in++] = data_sg;
116 	}
117 
118 	sg_init_one(&status, &vbr->status, sizeof(vbr->status));
119 	sgs[num_out + num_in++] = &status;
120 
121 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
122 }
123 
virtblk_setup_discard_write_zeroes(struct request * req,bool unmap)124 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
125 {
126 	unsigned short segments = blk_rq_nr_discard_segments(req);
127 	unsigned short n = 0;
128 	struct virtio_blk_discard_write_zeroes *range;
129 	struct bio *bio;
130 	u32 flags = 0;
131 
132 	if (unmap)
133 		flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
134 
135 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
136 	if (!range)
137 		return -ENOMEM;
138 
139 	/*
140 	 * Single max discard segment means multi-range discard isn't
141 	 * supported, and block layer only runs contiguity merge like
142 	 * normal RW request. So we can't reply on bio for retrieving
143 	 * each range info.
144 	 */
145 	if (queue_max_discard_segments(req->q) == 1) {
146 		range[0].flags = cpu_to_le32(flags);
147 		range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
148 		range[0].sector = cpu_to_le64(blk_rq_pos(req));
149 		n = 1;
150 	} else {
151 		__rq_for_each_bio(bio, req) {
152 			u64 sector = bio->bi_iter.bi_sector;
153 			u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
154 
155 			range[n].flags = cpu_to_le32(flags);
156 			range[n].num_sectors = cpu_to_le32(num_sectors);
157 			range[n].sector = cpu_to_le64(sector);
158 			n++;
159 		}
160 	}
161 
162 	WARN_ON_ONCE(n != segments);
163 
164 	req->special_vec.bv_page = virt_to_page(range);
165 	req->special_vec.bv_offset = offset_in_page(range);
166 	req->special_vec.bv_len = sizeof(*range) * segments;
167 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
168 
169 	return 0;
170 }
171 
virtblk_unmap_data(struct request * req,struct virtblk_req * vbr)172 static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
173 {
174 	if (blk_rq_nr_phys_segments(req))
175 		sg_free_table_chained(&vbr->sg_table,
176 				      VIRTIO_BLK_INLINE_SG_CNT);
177 }
178 
virtblk_map_data(struct blk_mq_hw_ctx * hctx,struct request * req,struct virtblk_req * vbr)179 static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
180 		struct virtblk_req *vbr)
181 {
182 	int err;
183 
184 	if (!blk_rq_nr_phys_segments(req))
185 		return 0;
186 
187 	vbr->sg_table.sgl = vbr->sg;
188 	err = sg_alloc_table_chained(&vbr->sg_table,
189 				     blk_rq_nr_phys_segments(req),
190 				     vbr->sg_table.sgl,
191 				     VIRTIO_BLK_INLINE_SG_CNT);
192 	if (unlikely(err))
193 		return -ENOMEM;
194 
195 	return blk_rq_map_sg(hctx->queue, req, vbr->sg_table.sgl);
196 }
197 
virtblk_cleanup_cmd(struct request * req)198 static void virtblk_cleanup_cmd(struct request *req)
199 {
200 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
201 		kfree(bvec_virt(&req->special_vec));
202 }
203 
virtblk_setup_cmd(struct virtio_device * vdev,struct request * req,struct virtblk_req * vbr)204 static int virtblk_setup_cmd(struct virtio_device *vdev, struct request *req,
205 		struct virtblk_req *vbr)
206 {
207 	bool unmap = false;
208 	u32 type;
209 
210 	vbr->out_hdr.sector = 0;
211 
212 	switch (req_op(req)) {
213 	case REQ_OP_READ:
214 		type = VIRTIO_BLK_T_IN;
215 		vbr->out_hdr.sector = cpu_to_virtio64(vdev,
216 						      blk_rq_pos(req));
217 		break;
218 	case REQ_OP_WRITE:
219 		type = VIRTIO_BLK_T_OUT;
220 		vbr->out_hdr.sector = cpu_to_virtio64(vdev,
221 						      blk_rq_pos(req));
222 		break;
223 	case REQ_OP_FLUSH:
224 		type = VIRTIO_BLK_T_FLUSH;
225 		break;
226 	case REQ_OP_DISCARD:
227 		type = VIRTIO_BLK_T_DISCARD;
228 		break;
229 	case REQ_OP_WRITE_ZEROES:
230 		type = VIRTIO_BLK_T_WRITE_ZEROES;
231 		unmap = !(req->cmd_flags & REQ_NOUNMAP);
232 		break;
233 	case REQ_OP_DRV_IN:
234 		type = VIRTIO_BLK_T_GET_ID;
235 		break;
236 	default:
237 		WARN_ON_ONCE(1);
238 		return BLK_STS_IOERR;
239 	}
240 
241 	vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
242 	vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
243 
244 	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
245 		if (virtblk_setup_discard_write_zeroes(req, unmap))
246 			return BLK_STS_RESOURCE;
247 	}
248 
249 	return 0;
250 }
251 
virtblk_request_done(struct request * req)252 static inline void virtblk_request_done(struct request *req)
253 {
254 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
255 
256 	virtblk_unmap_data(req, vbr);
257 	virtblk_cleanup_cmd(req);
258 	blk_mq_end_request(req, virtblk_result(vbr));
259 }
260 
virtblk_done(struct virtqueue * vq)261 static void virtblk_done(struct virtqueue *vq)
262 {
263 	struct virtio_blk *vblk = vq->vdev->priv;
264 	bool req_done = false;
265 	int qid = vq->index;
266 	struct virtblk_req *vbr;
267 	unsigned long flags;
268 	unsigned int len;
269 
270 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
271 	do {
272 		virtqueue_disable_cb(vq);
273 		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
274 			struct request *req = blk_mq_rq_from_pdu(vbr);
275 
276 			if (likely(!blk_should_fake_timeout(req->q)))
277 				blk_mq_complete_request(req);
278 			req_done = true;
279 		}
280 		if (unlikely(virtqueue_is_broken(vq)))
281 			break;
282 	} while (!virtqueue_enable_cb(vq));
283 
284 	/* In case queue is stopped waiting for more buffers. */
285 	if (req_done)
286 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
287 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
288 }
289 
virtio_commit_rqs(struct blk_mq_hw_ctx * hctx)290 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
291 {
292 	struct virtio_blk *vblk = hctx->queue->queuedata;
293 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
294 	bool kick;
295 
296 	spin_lock_irq(&vq->lock);
297 	kick = virtqueue_kick_prepare(vq->vq);
298 	spin_unlock_irq(&vq->lock);
299 
300 	if (kick)
301 		virtqueue_notify(vq->vq);
302 }
303 
virtio_queue_rq(struct blk_mq_hw_ctx * hctx,const struct blk_mq_queue_data * bd)304 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
305 			   const struct blk_mq_queue_data *bd)
306 {
307 	struct virtio_blk *vblk = hctx->queue->queuedata;
308 	struct request *req = bd->rq;
309 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
310 	unsigned long flags;
311 	int num;
312 	int qid = hctx->queue_num;
313 	int err;
314 	bool notify = false;
315 
316 	err = virtblk_setup_cmd(vblk->vdev, req, vbr);
317 	if (unlikely(err))
318 		return err;
319 
320 	blk_mq_start_request(req);
321 
322 	num = virtblk_map_data(hctx, req, vbr);
323 	if (unlikely(num < 0)) {
324 		virtblk_cleanup_cmd(req);
325 		return BLK_STS_RESOURCE;
326 	}
327 
328 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
329 	err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg_table.sgl, num);
330 	if (err) {
331 		virtqueue_kick(vblk->vqs[qid].vq);
332 		/* Don't stop the queue if -ENOMEM: we may have failed to
333 		 * bounce the buffer due to global resource outage.
334 		 */
335 		if (err == -ENOSPC)
336 			blk_mq_stop_hw_queue(hctx);
337 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
338 		virtblk_unmap_data(req, vbr);
339 		virtblk_cleanup_cmd(req);
340 		switch (err) {
341 		case -ENOSPC:
342 			return BLK_STS_DEV_RESOURCE;
343 		case -ENOMEM:
344 			return BLK_STS_RESOURCE;
345 		default:
346 			return BLK_STS_IOERR;
347 		}
348 	}
349 
350 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
351 		notify = true;
352 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
353 
354 	if (notify)
355 		virtqueue_notify(vblk->vqs[qid].vq);
356 	return BLK_STS_OK;
357 }
358 
359 /* return id (s/n) string for *disk to *id_str
360  */
virtblk_get_id(struct gendisk * disk,char * id_str)361 static int virtblk_get_id(struct gendisk *disk, char *id_str)
362 {
363 	struct virtio_blk *vblk = disk->private_data;
364 	struct request_queue *q = vblk->disk->queue;
365 	struct request *req;
366 	int err;
367 
368 	req = blk_get_request(q, REQ_OP_DRV_IN, 0);
369 	if (IS_ERR(req))
370 		return PTR_ERR(req);
371 
372 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
373 	if (err)
374 		goto out;
375 
376 	blk_execute_rq(vblk->disk, req, false);
377 	err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
378 out:
379 	blk_put_request(req);
380 	return err;
381 }
382 
virtblk_get(struct virtio_blk * vblk)383 static void virtblk_get(struct virtio_blk *vblk)
384 {
385 	refcount_inc(&vblk->refs);
386 }
387 
virtblk_put(struct virtio_blk * vblk)388 static void virtblk_put(struct virtio_blk *vblk)
389 {
390 	if (refcount_dec_and_test(&vblk->refs)) {
391 		ida_simple_remove(&vd_index_ida, vblk->index);
392 		mutex_destroy(&vblk->vdev_mutex);
393 		kfree(vblk);
394 	}
395 }
396 
virtblk_open(struct block_device * bd,fmode_t mode)397 static int virtblk_open(struct block_device *bd, fmode_t mode)
398 {
399 	struct virtio_blk *vblk = bd->bd_disk->private_data;
400 	int ret = 0;
401 
402 	mutex_lock(&vblk->vdev_mutex);
403 
404 	if (vblk->vdev)
405 		virtblk_get(vblk);
406 	else
407 		ret = -ENXIO;
408 
409 	mutex_unlock(&vblk->vdev_mutex);
410 	return ret;
411 }
412 
virtblk_release(struct gendisk * disk,fmode_t mode)413 static void virtblk_release(struct gendisk *disk, fmode_t mode)
414 {
415 	struct virtio_blk *vblk = disk->private_data;
416 
417 	virtblk_put(vblk);
418 }
419 
420 /* We provide getgeo only to please some old bootloader/partitioning tools */
virtblk_getgeo(struct block_device * bd,struct hd_geometry * geo)421 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
422 {
423 	struct virtio_blk *vblk = bd->bd_disk->private_data;
424 	int ret = 0;
425 
426 	mutex_lock(&vblk->vdev_mutex);
427 
428 	if (!vblk->vdev) {
429 		ret = -ENXIO;
430 		goto out;
431 	}
432 
433 	/* see if the host passed in geometry config */
434 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
435 		virtio_cread(vblk->vdev, struct virtio_blk_config,
436 			     geometry.cylinders, &geo->cylinders);
437 		virtio_cread(vblk->vdev, struct virtio_blk_config,
438 			     geometry.heads, &geo->heads);
439 		virtio_cread(vblk->vdev, struct virtio_blk_config,
440 			     geometry.sectors, &geo->sectors);
441 	} else {
442 		/* some standard values, similar to sd */
443 		geo->heads = 1 << 6;
444 		geo->sectors = 1 << 5;
445 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
446 	}
447 out:
448 	mutex_unlock(&vblk->vdev_mutex);
449 	return ret;
450 }
451 
452 static const struct block_device_operations virtblk_fops = {
453 	.owner  = THIS_MODULE,
454 	.open = virtblk_open,
455 	.release = virtblk_release,
456 	.getgeo = virtblk_getgeo,
457 };
458 
index_to_minor(int index)459 static int index_to_minor(int index)
460 {
461 	return index << PART_BITS;
462 }
463 
minor_to_index(int minor)464 static int minor_to_index(int minor)
465 {
466 	return minor >> PART_BITS;
467 }
468 
serial_show(struct device * dev,struct device_attribute * attr,char * buf)469 static ssize_t serial_show(struct device *dev,
470 			   struct device_attribute *attr, char *buf)
471 {
472 	struct gendisk *disk = dev_to_disk(dev);
473 	int err;
474 
475 	/* sysfs gives us a PAGE_SIZE buffer */
476 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
477 
478 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
479 	err = virtblk_get_id(disk, buf);
480 	if (!err)
481 		return strlen(buf);
482 
483 	if (err == -EIO) /* Unsupported? Make it empty. */
484 		return 0;
485 
486 	return err;
487 }
488 
489 static DEVICE_ATTR_RO(serial);
490 
491 /* The queue's logical block size must be set before calling this */
virtblk_update_capacity(struct virtio_blk * vblk,bool resize)492 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
493 {
494 	struct virtio_device *vdev = vblk->vdev;
495 	struct request_queue *q = vblk->disk->queue;
496 	char cap_str_2[10], cap_str_10[10];
497 	unsigned long long nblocks;
498 	u64 capacity;
499 
500 	/* Host must always specify the capacity. */
501 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
502 
503 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
504 
505 	string_get_size(nblocks, queue_logical_block_size(q),
506 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
507 	string_get_size(nblocks, queue_logical_block_size(q),
508 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
509 
510 	dev_notice(&vdev->dev,
511 		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
512 		   vblk->disk->disk_name,
513 		   resize ? "new size: " : "",
514 		   nblocks,
515 		   queue_logical_block_size(q),
516 		   cap_str_10,
517 		   cap_str_2);
518 
519 	set_capacity_and_notify(vblk->disk, capacity);
520 }
521 
virtblk_config_changed_work(struct work_struct * work)522 static void virtblk_config_changed_work(struct work_struct *work)
523 {
524 	struct virtio_blk *vblk =
525 		container_of(work, struct virtio_blk, config_work);
526 
527 	virtblk_update_capacity(vblk, true);
528 }
529 
virtblk_config_changed(struct virtio_device * vdev)530 static void virtblk_config_changed(struct virtio_device *vdev)
531 {
532 	struct virtio_blk *vblk = vdev->priv;
533 
534 	queue_work(virtblk_wq, &vblk->config_work);
535 }
536 
init_vq(struct virtio_blk * vblk)537 static int init_vq(struct virtio_blk *vblk)
538 {
539 	int err;
540 	int i;
541 	vq_callback_t **callbacks;
542 	const char **names;
543 	struct virtqueue **vqs;
544 	unsigned short num_vqs;
545 	struct virtio_device *vdev = vblk->vdev;
546 	struct irq_affinity desc = { 0, };
547 
548 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
549 				   struct virtio_blk_config, num_queues,
550 				   &num_vqs);
551 	if (err)
552 		num_vqs = 1;
553 	if (!err && !num_vqs) {
554 		dev_err(&vdev->dev, "MQ advertisted but zero queues reported\n");
555 		return -EINVAL;
556 	}
557 
558 	num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
559 
560 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
561 	if (!vblk->vqs)
562 		return -ENOMEM;
563 
564 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
565 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
566 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
567 	if (!names || !callbacks || !vqs) {
568 		err = -ENOMEM;
569 		goto out;
570 	}
571 
572 	for (i = 0; i < num_vqs; i++) {
573 		callbacks[i] = virtblk_done;
574 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
575 		names[i] = vblk->vqs[i].name;
576 	}
577 
578 	/* Discover virtqueues and write information to configuration.  */
579 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
580 	if (err)
581 		goto out;
582 
583 	for (i = 0; i < num_vqs; i++) {
584 		spin_lock_init(&vblk->vqs[i].lock);
585 		vblk->vqs[i].vq = vqs[i];
586 	}
587 	vblk->num_vqs = num_vqs;
588 
589 out:
590 	kfree(vqs);
591 	kfree(callbacks);
592 	kfree(names);
593 	if (err)
594 		kfree(vblk->vqs);
595 	return err;
596 }
597 
598 /*
599  * Legacy naming scheme used for virtio devices.  We are stuck with it for
600  * virtio blk but don't ever use it for any new driver.
601  */
virtblk_name_format(char * prefix,int index,char * buf,int buflen)602 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
603 {
604 	const int base = 'z' - 'a' + 1;
605 	char *begin = buf + strlen(prefix);
606 	char *end = buf + buflen;
607 	char *p;
608 	int unit;
609 
610 	p = end - 1;
611 	*p = '\0';
612 	unit = base;
613 	do {
614 		if (p == begin)
615 			return -EINVAL;
616 		*--p = 'a' + (index % unit);
617 		index = (index / unit) - 1;
618 	} while (index >= 0);
619 
620 	memmove(begin, p, end - p);
621 	memcpy(buf, prefix, strlen(prefix));
622 
623 	return 0;
624 }
625 
virtblk_get_cache_mode(struct virtio_device * vdev)626 static int virtblk_get_cache_mode(struct virtio_device *vdev)
627 {
628 	u8 writeback;
629 	int err;
630 
631 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
632 				   struct virtio_blk_config, wce,
633 				   &writeback);
634 
635 	/*
636 	 * If WCE is not configurable and flush is not available,
637 	 * assume no writeback cache is in use.
638 	 */
639 	if (err)
640 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
641 
642 	return writeback;
643 }
644 
virtblk_update_cache_mode(struct virtio_device * vdev)645 static void virtblk_update_cache_mode(struct virtio_device *vdev)
646 {
647 	u8 writeback = virtblk_get_cache_mode(vdev);
648 	struct virtio_blk *vblk = vdev->priv;
649 
650 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
651 }
652 
653 static const char *const virtblk_cache_types[] = {
654 	"write through", "write back"
655 };
656 
657 static ssize_t
cache_type_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)658 cache_type_store(struct device *dev, struct device_attribute *attr,
659 		 const char *buf, size_t count)
660 {
661 	struct gendisk *disk = dev_to_disk(dev);
662 	struct virtio_blk *vblk = disk->private_data;
663 	struct virtio_device *vdev = vblk->vdev;
664 	int i;
665 
666 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
667 	i = sysfs_match_string(virtblk_cache_types, buf);
668 	if (i < 0)
669 		return i;
670 
671 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
672 	virtblk_update_cache_mode(vdev);
673 	return count;
674 }
675 
676 static ssize_t
cache_type_show(struct device * dev,struct device_attribute * attr,char * buf)677 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
678 {
679 	struct gendisk *disk = dev_to_disk(dev);
680 	struct virtio_blk *vblk = disk->private_data;
681 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
682 
683 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
684 	return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
685 }
686 
687 static DEVICE_ATTR_RW(cache_type);
688 
689 static struct attribute *virtblk_attrs[] = {
690 	&dev_attr_serial.attr,
691 	&dev_attr_cache_type.attr,
692 	NULL,
693 };
694 
virtblk_attrs_are_visible(struct kobject * kobj,struct attribute * a,int n)695 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
696 		struct attribute *a, int n)
697 {
698 	struct device *dev = kobj_to_dev(kobj);
699 	struct gendisk *disk = dev_to_disk(dev);
700 	struct virtio_blk *vblk = disk->private_data;
701 	struct virtio_device *vdev = vblk->vdev;
702 
703 	if (a == &dev_attr_cache_type.attr &&
704 	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
705 		return S_IRUGO;
706 
707 	return a->mode;
708 }
709 
710 static const struct attribute_group virtblk_attr_group = {
711 	.attrs = virtblk_attrs,
712 	.is_visible = virtblk_attrs_are_visible,
713 };
714 
715 static const struct attribute_group *virtblk_attr_groups[] = {
716 	&virtblk_attr_group,
717 	NULL,
718 };
719 
virtblk_map_queues(struct blk_mq_tag_set * set)720 static int virtblk_map_queues(struct blk_mq_tag_set *set)
721 {
722 	struct virtio_blk *vblk = set->driver_data;
723 
724 	return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
725 					vblk->vdev, 0);
726 }
727 
728 static const struct blk_mq_ops virtio_mq_ops = {
729 	.queue_rq	= virtio_queue_rq,
730 	.commit_rqs	= virtio_commit_rqs,
731 	.complete	= virtblk_request_done,
732 	.map_queues	= virtblk_map_queues,
733 };
734 
735 static unsigned int virtblk_queue_depth;
736 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
737 
virtblk_probe(struct virtio_device * vdev)738 static int virtblk_probe(struct virtio_device *vdev)
739 {
740 	struct virtio_blk *vblk;
741 	struct request_queue *q;
742 	int err, index;
743 
744 	u32 v, blk_size, max_size, sg_elems, opt_io_size;
745 	u16 min_io_size;
746 	u8 physical_block_exp, alignment_offset;
747 	unsigned int queue_depth;
748 	size_t max_dma_size;
749 
750 	if (!vdev->config->get) {
751 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
752 			__func__);
753 		return -EINVAL;
754 	}
755 
756 	err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
757 			     GFP_KERNEL);
758 	if (err < 0)
759 		goto out;
760 	index = err;
761 
762 	/* We need to know how many segments before we allocate. */
763 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
764 				   struct virtio_blk_config, seg_max,
765 				   &sg_elems);
766 
767 	/* We need at least one SG element, whatever they say. */
768 	if (err || !sg_elems)
769 		sg_elems = 1;
770 
771 	/* Prevent integer overflows and honor max vq size */
772 	sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
773 
774 	/* We need extra sg elements at head and tail. */
775 	sg_elems += 2;
776 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
777 	if (!vblk) {
778 		err = -ENOMEM;
779 		goto out_free_index;
780 	}
781 
782 	/* This reference is dropped in virtblk_remove(). */
783 	refcount_set(&vblk->refs, 1);
784 	mutex_init(&vblk->vdev_mutex);
785 
786 	vblk->vdev = vdev;
787 	vblk->sg_elems = sg_elems;
788 
789 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
790 
791 	err = init_vq(vblk);
792 	if (err)
793 		goto out_free_vblk;
794 
795 	/* Default queue sizing is to fill the ring. */
796 	if (!virtblk_queue_depth) {
797 		queue_depth = vblk->vqs[0].vq->num_free;
798 		/* ... but without indirect descs, we use 2 descs per req */
799 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
800 			queue_depth /= 2;
801 	} else {
802 		queue_depth = virtblk_queue_depth;
803 	}
804 
805 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
806 	vblk->tag_set.ops = &virtio_mq_ops;
807 	vblk->tag_set.queue_depth = queue_depth;
808 	vblk->tag_set.numa_node = NUMA_NO_NODE;
809 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
810 	vblk->tag_set.cmd_size =
811 		sizeof(struct virtblk_req) +
812 		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
813 	vblk->tag_set.driver_data = vblk;
814 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
815 
816 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
817 	if (err)
818 		goto out_free_vq;
819 
820 	vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
821 	if (IS_ERR(vblk->disk)) {
822 		err = PTR_ERR(vblk->disk);
823 		goto out_free_tags;
824 	}
825 	q = vblk->disk->queue;
826 
827 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
828 
829 	vblk->disk->major = major;
830 	vblk->disk->first_minor = index_to_minor(index);
831 	vblk->disk->minors = 1 << PART_BITS;
832 	vblk->disk->private_data = vblk;
833 	vblk->disk->fops = &virtblk_fops;
834 	vblk->disk->flags |= GENHD_FL_EXT_DEVT;
835 	vblk->index = index;
836 
837 	/* configure queue flush support */
838 	virtblk_update_cache_mode(vdev);
839 
840 	/* If disk is read-only in the host, the guest should obey */
841 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
842 		set_disk_ro(vblk->disk, 1);
843 
844 	/* We can handle whatever the host told us to handle. */
845 	blk_queue_max_segments(q, vblk->sg_elems-2);
846 
847 	/* No real sector limit. */
848 	blk_queue_max_hw_sectors(q, -1U);
849 
850 	max_dma_size = virtio_max_dma_size(vdev);
851 	max_size = max_dma_size > U32_MAX ? U32_MAX : max_dma_size;
852 
853 	/* Host can optionally specify maximum segment size and number of
854 	 * segments. */
855 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
856 				   struct virtio_blk_config, size_max, &v);
857 	if (!err)
858 		max_size = min(max_size, v);
859 
860 	blk_queue_max_segment_size(q, max_size);
861 
862 	/* Host can optionally specify the block size of the device */
863 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
864 				   struct virtio_blk_config, blk_size,
865 				   &blk_size);
866 	if (!err) {
867 		err = blk_validate_block_size(blk_size);
868 		if (err) {
869 			dev_err(&vdev->dev,
870 				"virtio_blk: invalid block size: 0x%x\n",
871 				blk_size);
872 			goto out_cleanup_disk;
873 		}
874 
875 		blk_queue_logical_block_size(q, blk_size);
876 	} else
877 		blk_size = queue_logical_block_size(q);
878 
879 	/* Use topology information if available */
880 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
881 				   struct virtio_blk_config, physical_block_exp,
882 				   &physical_block_exp);
883 	if (!err && physical_block_exp)
884 		blk_queue_physical_block_size(q,
885 				blk_size * (1 << physical_block_exp));
886 
887 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
888 				   struct virtio_blk_config, alignment_offset,
889 				   &alignment_offset);
890 	if (!err && alignment_offset)
891 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
892 
893 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
894 				   struct virtio_blk_config, min_io_size,
895 				   &min_io_size);
896 	if (!err && min_io_size)
897 		blk_queue_io_min(q, blk_size * min_io_size);
898 
899 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
900 				   struct virtio_blk_config, opt_io_size,
901 				   &opt_io_size);
902 	if (!err && opt_io_size)
903 		blk_queue_io_opt(q, blk_size * opt_io_size);
904 
905 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
906 		virtio_cread(vdev, struct virtio_blk_config,
907 			     discard_sector_alignment, &v);
908 		if (v)
909 			q->limits.discard_granularity = v << SECTOR_SHIFT;
910 		else
911 			q->limits.discard_granularity = blk_size;
912 
913 		virtio_cread(vdev, struct virtio_blk_config,
914 			     max_discard_sectors, &v);
915 		blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
916 
917 		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
918 			     &v);
919 
920 		/*
921 		 * max_discard_seg == 0 is out of spec but we always
922 		 * handled it.
923 		 */
924 		if (!v)
925 			v = sg_elems - 2;
926 		blk_queue_max_discard_segments(q,
927 					       min(v, MAX_DISCARD_SEGMENTS));
928 
929 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
930 	}
931 
932 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
933 		virtio_cread(vdev, struct virtio_blk_config,
934 			     max_write_zeroes_sectors, &v);
935 		blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
936 	}
937 
938 	virtblk_update_capacity(vblk, false);
939 	virtio_device_ready(vdev);
940 
941 	err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
942 	if (err)
943 		goto out_cleanup_disk;
944 
945 	return 0;
946 
947 out_cleanup_disk:
948 	blk_cleanup_disk(vblk->disk);
949 out_free_tags:
950 	blk_mq_free_tag_set(&vblk->tag_set);
951 out_free_vq:
952 	vdev->config->del_vqs(vdev);
953 	kfree(vblk->vqs);
954 out_free_vblk:
955 	kfree(vblk);
956 out_free_index:
957 	ida_simple_remove(&vd_index_ida, index);
958 out:
959 	return err;
960 }
961 
virtblk_remove(struct virtio_device * vdev)962 static void virtblk_remove(struct virtio_device *vdev)
963 {
964 	struct virtio_blk *vblk = vdev->priv;
965 
966 	/* Make sure no work handler is accessing the device. */
967 	flush_work(&vblk->config_work);
968 
969 	del_gendisk(vblk->disk);
970 	blk_cleanup_disk(vblk->disk);
971 	blk_mq_free_tag_set(&vblk->tag_set);
972 
973 	mutex_lock(&vblk->vdev_mutex);
974 
975 	/* Stop all the virtqueues. */
976 	vdev->config->reset(vdev);
977 
978 	/* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
979 	vblk->vdev = NULL;
980 
981 	vdev->config->del_vqs(vdev);
982 	kfree(vblk->vqs);
983 
984 	mutex_unlock(&vblk->vdev_mutex);
985 
986 	virtblk_put(vblk);
987 }
988 
989 #ifdef CONFIG_PM_SLEEP
virtblk_freeze(struct virtio_device * vdev)990 static int virtblk_freeze(struct virtio_device *vdev)
991 {
992 	struct virtio_blk *vblk = vdev->priv;
993 
994 	/* Ensure we don't receive any more interrupts */
995 	vdev->config->reset(vdev);
996 
997 	/* Make sure no work handler is accessing the device. */
998 	flush_work(&vblk->config_work);
999 
1000 	blk_mq_quiesce_queue(vblk->disk->queue);
1001 
1002 	vdev->config->del_vqs(vdev);
1003 	kfree(vblk->vqs);
1004 
1005 	return 0;
1006 }
1007 
virtblk_restore(struct virtio_device * vdev)1008 static int virtblk_restore(struct virtio_device *vdev)
1009 {
1010 	struct virtio_blk *vblk = vdev->priv;
1011 	int ret;
1012 
1013 	ret = init_vq(vdev->priv);
1014 	if (ret)
1015 		return ret;
1016 
1017 	virtio_device_ready(vdev);
1018 
1019 	blk_mq_unquiesce_queue(vblk->disk->queue);
1020 	return 0;
1021 }
1022 #endif
1023 
1024 static const struct virtio_device_id id_table[] = {
1025 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1026 	{ 0 },
1027 };
1028 
1029 static unsigned int features_legacy[] = {
1030 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1031 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1032 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1033 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1034 }
1035 ;
1036 static unsigned int features[] = {
1037 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1038 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1039 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1040 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1041 };
1042 
1043 static struct virtio_driver virtio_blk = {
1044 	.feature_table			= features,
1045 	.feature_table_size		= ARRAY_SIZE(features),
1046 	.feature_table_legacy		= features_legacy,
1047 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
1048 	.driver.name			= KBUILD_MODNAME,
1049 	.driver.owner			= THIS_MODULE,
1050 	.id_table			= id_table,
1051 	.probe				= virtblk_probe,
1052 	.remove				= virtblk_remove,
1053 	.config_changed			= virtblk_config_changed,
1054 #ifdef CONFIG_PM_SLEEP
1055 	.freeze				= virtblk_freeze,
1056 	.restore			= virtblk_restore,
1057 #endif
1058 };
1059 
init(void)1060 static int __init init(void)
1061 {
1062 	int error;
1063 
1064 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1065 	if (!virtblk_wq)
1066 		return -ENOMEM;
1067 
1068 	major = register_blkdev(0, "virtblk");
1069 	if (major < 0) {
1070 		error = major;
1071 		goto out_destroy_workqueue;
1072 	}
1073 
1074 	error = register_virtio_driver(&virtio_blk);
1075 	if (error)
1076 		goto out_unregister_blkdev;
1077 	return 0;
1078 
1079 out_unregister_blkdev:
1080 	unregister_blkdev(major, "virtblk");
1081 out_destroy_workqueue:
1082 	destroy_workqueue(virtblk_wq);
1083 	return error;
1084 }
1085 
fini(void)1086 static void __exit fini(void)
1087 {
1088 	unregister_virtio_driver(&virtio_blk);
1089 	unregister_blkdev(major, "virtblk");
1090 	destroy_workqueue(virtblk_wq);
1091 }
1092 module_init(init);
1093 module_exit(fini);
1094 
1095 MODULE_DEVICE_TABLE(virtio, id_table);
1096 MODULE_DESCRIPTION("Virtio block driver");
1097 MODULE_LICENSE("GPL");
1098