virtio_blk.c

   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/numa.h>
  17#include <linux/vmalloc.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
  33static unsigned int num_request_queues;
  34module_param(num_request_queues, uint, 0644);
  35MODULE_PARM_DESC(num_request_queues,
  36		 "Limit the number of request queues to use for blk device. "
  37		 "0 for no limit. "
  38		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
  39
  40static unsigned int poll_queues;
  41module_param(poll_queues, uint, 0644);
  42MODULE_PARM_DESC(poll_queues, "The number of dedicated virtqueues for polling I/O");
  43
  44static int major;
  45static DEFINE_IDA(vd_index_ida);
  46
  47static struct workqueue_struct *virtblk_wq;
  48
  49struct virtio_blk_vq {
  50	struct virtqueue *vq;
  51	spinlock_t lock;
  52	char name[VQ_NAME_LEN];
  53} ____cacheline_aligned_in_smp;
  54
  55struct virtio_blk {
  56	/*
  57	 * This mutex must be held by anything that may run after
  58	 * virtblk_remove() sets vblk->vdev to NULL.
  59	 *
  60	 * blk-mq, virtqueue processing, and sysfs attribute code paths are
  61	 * shut down before vblk->vdev is set to NULL and therefore do not need
  62	 * to hold this mutex.
  63	 */
  64	struct mutex vdev_mutex;
  65	struct virtio_device *vdev;
  66
  67	/* The disk structure for the kernel. */
  68	struct gendisk *disk;
  69
  70	/* Block layer tags. */
  71	struct blk_mq_tag_set tag_set;
  72
  73	/* Process context for config space updates */
  74	struct work_struct config_work;
  75
  76	/* Ida index - used to track minor number allocations. */
  77	int index;
  78
  79	/* num of vqs */
  80	int num_vqs;
  81	int io_queues[HCTX_MAX_TYPES];
  82	struct virtio_blk_vq *vqs;
  83
  84	/* For zoned device */
  85	unsigned int zone_sectors;
  86};
  87
  88struct virtblk_req {
  89	/* Out header */
  90	struct virtio_blk_outhdr out_hdr;
  91
  92	/* In header */
  93	union {
  94		u8 status;
  95
  96		/*
  97		 * The zone append command has an extended in header.
  98		 * The status field in zone_append_in_hdr must always
  99		 * be the last byte.
 100		 */
 101		struct {
 102			__virtio64 sector;
 103			u8 status;
 104		} zone_append;
 105	} in_hdr;
 106
 107	size_t in_hdr_len;
 108
 109	struct sg_table sg_table;
 110	struct scatterlist sg[];
 111};
 112
 113static inline blk_status_t virtblk_result(u8 status)
 114{
 115	switch (status) {
 116	case VIRTIO_BLK_S_OK:
 117		return BLK_STS_OK;
 118	case VIRTIO_BLK_S_UNSUPP:
 119		return BLK_STS_NOTSUPP;
 120	case VIRTIO_BLK_S_ZONE_OPEN_RESOURCE:
 121		return BLK_STS_ZONE_OPEN_RESOURCE;
 122	case VIRTIO_BLK_S_ZONE_ACTIVE_RESOURCE:
 123		return BLK_STS_ZONE_ACTIVE_RESOURCE;
 124	case VIRTIO_BLK_S_IOERR:
 125	case VIRTIO_BLK_S_ZONE_UNALIGNED_WP:
 126	default:
 127		return BLK_STS_IOERR;
 128	}
 129}
 130
 131static inline struct virtio_blk_vq *get_virtio_blk_vq(struct blk_mq_hw_ctx *hctx)
 132{
 133	struct virtio_blk *vblk = hctx->queue->queuedata;
 134	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 135
 136	return vq;
 137}
 138
 139static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr)
 140{
 141	struct scatterlist out_hdr, in_hdr, *sgs[3];
 142	unsigned int num_out = 0, num_in = 0;
 143
 144	sg_init_one(&out_hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
 145	sgs[num_out++] = &out_hdr;
 146
 147	if (vbr->sg_table.nents) {
 148		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
 149			sgs[num_out++] = vbr->sg_table.sgl;
 150		else
 151			sgs[num_out + num_in++] = vbr->sg_table.sgl;
 152	}
 153
 154	sg_init_one(&in_hdr, &vbr->in_hdr.status, vbr->in_hdr_len);
 155	sgs[num_out + num_in++] = &in_hdr;
 156
 157	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
 158}
 159
 160static int virtblk_setup_discard_write_zeroes_erase(struct request *req, bool unmap)
 161{
 162	unsigned short segments = blk_rq_nr_discard_segments(req);
 163	unsigned short n = 0;
 164	struct virtio_blk_discard_write_zeroes *range;
 165	struct bio *bio;
 166	u32 flags = 0;
 167
 168	if (unmap)
 169		flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
 170
 171	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 172	if (!range)
 173		return -ENOMEM;
 174
 175	/*
 176	 * Single max discard segment means multi-range discard isn't
 177	 * supported, and block layer only runs contiguity merge like
 178	 * normal RW request. So we can't reply on bio for retrieving
 179	 * each range info.
 180	 */
 181	if (queue_max_discard_segments(req->q) == 1) {
 182		range[0].flags = cpu_to_le32(flags);
 183		range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
 184		range[0].sector = cpu_to_le64(blk_rq_pos(req));
 185		n = 1;
 186	} else {
 187		__rq_for_each_bio(bio, req) {
 188			u64 sector = bio->bi_iter.bi_sector;
 189			u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
 190
 191			range[n].flags = cpu_to_le32(flags);
 192			range[n].num_sectors = cpu_to_le32(num_sectors);
 193			range[n].sector = cpu_to_le64(sector);
 194			n++;
 195		}
 196	}
 197
 198	WARN_ON_ONCE(n != segments);
 199
 200	bvec_set_virt(&req->special_vec, range, sizeof(*range) * segments);
 201	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 202
 203	return 0;
 204}
 205
 206static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
 207{
 208	if (blk_rq_nr_phys_segments(req))
 209		sg_free_table_chained(&vbr->sg_table,
 210				      VIRTIO_BLK_INLINE_SG_CNT);
 211}
 212
 213static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
 214		struct virtblk_req *vbr)
 215{
 216	int err;
 217
 218	if (!blk_rq_nr_phys_segments(req))
 219		return 0;
 220
 221	vbr->sg_table.sgl = vbr->sg;
 222	err = sg_alloc_table_chained(&vbr->sg_table,
 223				     blk_rq_nr_phys_segments(req),
 224				     vbr->sg_table.sgl,
 225				     VIRTIO_BLK_INLINE_SG_CNT);
 226	if (unlikely(err))
 227		return -ENOMEM;
 228
 229	return blk_rq_map_sg(req, vbr->sg_table.sgl);
 230}
 231
 232static void virtblk_cleanup_cmd(struct request *req)
 233{
 234	if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
 235		kfree(bvec_virt(&req->special_vec));
 236}
 237
 238static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
 239				      struct request *req,
 240				      struct virtblk_req *vbr)
 241{
 242	size_t in_hdr_len = sizeof(vbr->in_hdr.status);
 243	bool unmap = false;
 244	u32 type;
 245	u64 sector = 0;
 246
 247	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && op_is_zone_mgmt(req_op(req)))
 248		return BLK_STS_NOTSUPP;
 249
 250	/* Set fields for all request types */
 251	vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
 252
 253	switch (req_op(req)) {
 254	case REQ_OP_READ:
 255		type = VIRTIO_BLK_T_IN;
 256		sector = blk_rq_pos(req);
 257		break;
 258	case REQ_OP_WRITE:
 259		type = VIRTIO_BLK_T_OUT;
 260		sector = blk_rq_pos(req);
 261		break;
 262	case REQ_OP_FLUSH:
 263		type = VIRTIO_BLK_T_FLUSH;
 264		break;
 265	case REQ_OP_DISCARD:
 266		type = VIRTIO_BLK_T_DISCARD;
 267		break;
 268	case REQ_OP_WRITE_ZEROES:
 269		type = VIRTIO_BLK_T_WRITE_ZEROES;
 270		unmap = !(req->cmd_flags & REQ_NOUNMAP);
 271		break;
 272	case REQ_OP_SECURE_ERASE:
 273		type = VIRTIO_BLK_T_SECURE_ERASE;
 274		break;
 275	case REQ_OP_ZONE_OPEN:
 276		type = VIRTIO_BLK_T_ZONE_OPEN;
 277		sector = blk_rq_pos(req);
 278		break;
 279	case REQ_OP_ZONE_CLOSE:
 280		type = VIRTIO_BLK_T_ZONE_CLOSE;
 281		sector = blk_rq_pos(req);
 282		break;
 283	case REQ_OP_ZONE_FINISH:
 284		type = VIRTIO_BLK_T_ZONE_FINISH;
 285		sector = blk_rq_pos(req);
 286		break;
 287	case REQ_OP_ZONE_APPEND:
 288		type = VIRTIO_BLK_T_ZONE_APPEND;
 289		sector = blk_rq_pos(req);
 290		in_hdr_len = sizeof(vbr->in_hdr.zone_append);
 291		break;
 292	case REQ_OP_ZONE_RESET:
 293		type = VIRTIO_BLK_T_ZONE_RESET;
 294		sector = blk_rq_pos(req);
 295		break;
 296	case REQ_OP_ZONE_RESET_ALL:
 297		type = VIRTIO_BLK_T_ZONE_RESET_ALL;
 298		break;
 299	case REQ_OP_DRV_IN:
 300		/*
 301		 * Out header has already been prepared by the caller (virtblk_get_id()
 302		 * or virtblk_submit_zone_report()), nothing to do here.
 303		 */
 304		return 0;
 305	default:
 306		WARN_ON_ONCE(1);
 307		return BLK_STS_IOERR;
 308	}
 309
 310	/* Set fields for non-REQ_OP_DRV_IN request types */
 311	vbr->in_hdr_len = in_hdr_len;
 312	vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
 313	vbr->out_hdr.sector = cpu_to_virtio64(vdev, sector);
 314
 315	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES ||
 316	    type == VIRTIO_BLK_T_SECURE_ERASE) {
 317		if (virtblk_setup_discard_write_zeroes_erase(req, unmap))
 318			return BLK_STS_RESOURCE;
 319	}
 320
 321	return 0;
 322}
 323
 324/*
 325 * The status byte is always the last byte of the virtblk request
 326 * in-header. This helper fetches its value for all in-header formats
 327 * that are currently defined.
 328 */
 329static inline u8 virtblk_vbr_status(struct virtblk_req *vbr)
 330{
 331	return *((u8 *)&vbr->in_hdr + vbr->in_hdr_len - 1);
 332}
 333
 334static inline void virtblk_request_done(struct request *req)
 335{
 336	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 337	blk_status_t status = virtblk_result(virtblk_vbr_status(vbr));
 338	struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
 339
 340	virtblk_unmap_data(req, vbr);
 341	virtblk_cleanup_cmd(req);
 342
 343	if (req_op(req) == REQ_OP_ZONE_APPEND)
 344		req->__sector = virtio64_to_cpu(vblk->vdev,
 345						vbr->in_hdr.zone_append.sector);
 346
 347	blk_mq_end_request(req, status);
 348}
 349
 350static void virtblk_done(struct virtqueue *vq)
 351{
 352	struct virtio_blk *vblk = vq->vdev->priv;
 353	bool req_done = false;
 354	int qid = vq->index;
 355	struct virtblk_req *vbr;
 356	unsigned long flags;
 357	unsigned int len;
 358
 359	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 360	do {
 361		virtqueue_disable_cb(vq);
 362		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
 363			struct request *req = blk_mq_rq_from_pdu(vbr);
 364
 365			if (likely(!blk_should_fake_timeout(req->q)))
 366				blk_mq_complete_request(req);
 367			req_done = true;
 368		}
 369	} while (!virtqueue_enable_cb(vq));
 370
 371	/* In case queue is stopped waiting for more buffers. */
 372	if (req_done)
 373		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
 374	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 375}
 376
 377static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
 378{
 379	struct virtio_blk *vblk = hctx->queue->queuedata;
 380	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
 381	bool kick;
 382
 383	spin_lock_irq(&vq->lock);
 384	kick = virtqueue_kick_prepare(vq->vq);
 385	spin_unlock_irq(&vq->lock);
 386
 387	if (kick)
 388		virtqueue_notify(vq->vq);
 389}
 390
 391static blk_status_t virtblk_fail_to_queue(struct request *req, int rc)
 392{
 393	virtblk_cleanup_cmd(req);
 394	switch (rc) {
 395	case -ENOSPC:
 396		return BLK_STS_DEV_RESOURCE;
 397	case -ENOMEM:
 398		return BLK_STS_RESOURCE;
 399	default:
 400		return BLK_STS_IOERR;
 401	}
 402}
 403
 404static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
 405					struct virtio_blk *vblk,
 406					struct request *req,
 407					struct virtblk_req *vbr)
 408{
 409	blk_status_t status;
 410	int num;
 411
 412	status = virtblk_setup_cmd(vblk->vdev, req, vbr);
 413	if (unlikely(status))
 414		return status;
 415
 416	num = virtblk_map_data(hctx, req, vbr);
 417	if (unlikely(num < 0))
 418		return virtblk_fail_to_queue(req, -ENOMEM);
 419	vbr->sg_table.nents = num;
 420
 421	blk_mq_start_request(req);
 422
 423	return BLK_STS_OK;
 424}
 425
 426static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
 427			   const struct blk_mq_queue_data *bd)
 428{
 429	struct virtio_blk *vblk = hctx->queue->queuedata;
 430	struct request *req = bd->rq;
 431	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 432	unsigned long flags;
 433	int qid = hctx->queue_num;
 434	bool notify = false;
 435	blk_status_t status;
 436	int err;
 437
 438	status = virtblk_prep_rq(hctx, vblk, req, vbr);
 439	if (unlikely(status))
 440		return status;
 441
 442	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
 443	err = virtblk_add_req(vblk->vqs[qid].vq, vbr);
 444	if (err) {
 445		virtqueue_kick(vblk->vqs[qid].vq);
 446		/* Don't stop the queue if -ENOMEM: we may have failed to
 447		 * bounce the buffer due to global resource outage.
 448		 */
 449		if (err == -ENOSPC)
 450			blk_mq_stop_hw_queue(hctx);
 451		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 452		virtblk_unmap_data(req, vbr);
 453		return virtblk_fail_to_queue(req, err);
 454	}
 455
 456	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
 457		notify = true;
 458	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
 459
 460	if (notify)
 461		virtqueue_notify(vblk->vqs[qid].vq);
 462	return BLK_STS_OK;
 463}
 464
 465static bool virtblk_prep_rq_batch(struct request *req)
 466{
 467	struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
 468	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 469
 470	return virtblk_prep_rq(req->mq_hctx, vblk, req, vbr) == BLK_STS_OK;
 471}
 472
 473static void virtblk_add_req_batch(struct virtio_blk_vq *vq,
 474		struct rq_list *rqlist)
 475{
 476	struct request *req;
 477	unsigned long flags;
 478	bool kick;
 479
 480	spin_lock_irqsave(&vq->lock, flags);
 481
 482	while ((req = rq_list_pop(rqlist))) {
 483		struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
 484		int err;
 485
 486		err = virtblk_add_req(vq->vq, vbr);
 487		if (err) {
 488			virtblk_unmap_data(req, vbr);
 489			virtblk_cleanup_cmd(req);
 490			blk_mq_requeue_request(req, true);
 491		}
 492	}
 493
 494	kick = virtqueue_kick_prepare(vq->vq);
 495	spin_unlock_irqrestore(&vq->lock, flags);
 496
 497	if (kick)
 498		virtqueue_notify(vq->vq);
 499}
 500
 501static void virtio_queue_rqs(struct rq_list *rqlist)
 502{
 503	struct rq_list submit_list = { };
 504	struct rq_list requeue_list = { };
 505	struct virtio_blk_vq *vq = NULL;
 506	struct request *req;
 507
 508	while ((req = rq_list_pop(rqlist))) {
 509		struct virtio_blk_vq *this_vq = get_virtio_blk_vq(req->mq_hctx);
 510
 511		if (vq && vq != this_vq)
 512			virtblk_add_req_batch(vq, &submit_list);
 513		vq = this_vq;
 514
 515		if (virtblk_prep_rq_batch(req))
 516			rq_list_add_tail(&submit_list, req);
 517		else
 518			rq_list_add_tail(&requeue_list, req);
 519	}
 520
 521	if (vq)
 522		virtblk_add_req_batch(vq, &submit_list);
 523	*rqlist = requeue_list;
 524}
 525
 526#ifdef CONFIG_BLK_DEV_ZONED
 527static void *virtblk_alloc_report_buffer(struct virtio_blk *vblk,
 528					  unsigned int nr_zones,
 529					  size_t *buflen)
 530{
 531	struct request_queue *q = vblk->disk->queue;
 532	size_t bufsize;
 533	void *buf;
 534
 535	nr_zones = min_t(unsigned int, nr_zones,
 536			 get_capacity(vblk->disk) >> ilog2(vblk->zone_sectors));
 537
 538	bufsize = sizeof(struct virtio_blk_zone_report) +
 539		nr_zones * sizeof(struct virtio_blk_zone_descriptor);
 540	bufsize = min_t(size_t, bufsize,
 541			queue_max_hw_sectors(q) << SECTOR_SHIFT);
 542	bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
 543
 544	while (bufsize >= sizeof(struct virtio_blk_zone_report)) {
 545		buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
 546		if (buf) {
 547			*buflen = bufsize;
 548			return buf;
 549		}
 550		bufsize >>= 1;
 551	}
 552
 553	return NULL;
 554}
 555
 556static int virtblk_submit_zone_report(struct virtio_blk *vblk,
 557				       char *report_buf, size_t report_len,
 558				       sector_t sector)
 559{
 560	struct request_queue *q = vblk->disk->queue;
 561	struct request *req;
 562	struct virtblk_req *vbr;
 563	int err;
 564
 565	req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
 566	if (IS_ERR(req))
 567		return PTR_ERR(req);
 568
 569	vbr = blk_mq_rq_to_pdu(req);
 570	vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
 571	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_ZONE_REPORT);
 572	vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, sector);
 573
 574	err = blk_rq_map_kern(req, report_buf, report_len, GFP_KERNEL);
 575	if (err)
 576		goto out;
 577
 578	blk_execute_rq(req, false);
 579	err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
 580out:
 581	blk_mq_free_request(req);
 582	return err;
 583}
 584
 585static int virtblk_parse_zone(struct virtio_blk *vblk,
 586			       struct virtio_blk_zone_descriptor *entry,
 587			       unsigned int idx,
 588			       struct blk_report_zones_args *args)
 589{
 590	struct blk_zone zone = { };
 591
 592	zone.start = virtio64_to_cpu(vblk->vdev, entry->z_start);
 593	if (zone.start + vblk->zone_sectors <= get_capacity(vblk->disk))
 594		zone.len = vblk->zone_sectors;
 595	else
 596		zone.len = get_capacity(vblk->disk) - zone.start;
 597	zone.capacity = virtio64_to_cpu(vblk->vdev, entry->z_cap);
 598	zone.wp = virtio64_to_cpu(vblk->vdev, entry->z_wp);
 599
 600	switch (entry->z_type) {
 601	case VIRTIO_BLK_ZT_SWR:
 602		zone.type = BLK_ZONE_TYPE_SEQWRITE_REQ;
 603		break;
 604	case VIRTIO_BLK_ZT_SWP:
 605		zone.type = BLK_ZONE_TYPE_SEQWRITE_PREF;
 606		break;
 607	case VIRTIO_BLK_ZT_CONV:
 608		zone.type = BLK_ZONE_TYPE_CONVENTIONAL;
 609		break;
 610	default:
 611		dev_err(&vblk->vdev->dev, "zone %llu: invalid type %#x\n",
 612			zone.start, entry->z_type);
 613		return -EIO;
 614	}
 615
 616	switch (entry->z_state) {
 617	case VIRTIO_BLK_ZS_EMPTY:
 618		zone.cond = BLK_ZONE_COND_EMPTY;
 619		break;
 620	case VIRTIO_BLK_ZS_CLOSED:
 621		zone.cond = BLK_ZONE_COND_CLOSED;
 622		break;
 623	case VIRTIO_BLK_ZS_FULL:
 624		zone.cond = BLK_ZONE_COND_FULL;
 625		zone.wp = zone.start + zone.len;
 626		break;
 627	case VIRTIO_BLK_ZS_EOPEN:
 628		zone.cond = BLK_ZONE_COND_EXP_OPEN;
 629		break;
 630	case VIRTIO_BLK_ZS_IOPEN:
 631		zone.cond = BLK_ZONE_COND_IMP_OPEN;
 632		break;
 633	case VIRTIO_BLK_ZS_NOT_WP:
 634		zone.cond = BLK_ZONE_COND_NOT_WP;
 635		break;
 636	case VIRTIO_BLK_ZS_RDONLY:
 637		zone.cond = BLK_ZONE_COND_READONLY;
 638		zone.wp = ULONG_MAX;
 639		break;
 640	case VIRTIO_BLK_ZS_OFFLINE:
 641		zone.cond = BLK_ZONE_COND_OFFLINE;
 642		zone.wp = ULONG_MAX;
 643		break;
 644	default:
 645		dev_err(&vblk->vdev->dev, "zone %llu: invalid condition %#x\n",
 646			zone.start, entry->z_state);
 647		return -EIO;
 648	}
 649
 650	/*
 651	 * The callback below checks the validity of the reported
 652	 * entry data, no need to further validate it here.
 653	 */
 654	return disk_report_zone(vblk->disk, &zone, idx, args);
 655}
 656
 657static int virtblk_report_zones(struct gendisk *disk, sector_t sector,
 658				 unsigned int nr_zones,
 659				 struct blk_report_zones_args *args)
 660{
 661	struct virtio_blk *vblk = disk->private_data;
 662	struct virtio_blk_zone_report *report;
 663	unsigned long long nz, i;
 664	size_t buflen;
 665	unsigned int zone_idx = 0;
 666	int ret;
 667
 668	if (WARN_ON_ONCE(!vblk->zone_sectors))
 669		return -EOPNOTSUPP;
 670
 671	report = virtblk_alloc_report_buffer(vblk, nr_zones, &buflen);
 672	if (!report)
 673		return -ENOMEM;
 674
 675	mutex_lock(&vblk->vdev_mutex);
 676
 677	if (!vblk->vdev) {
 678		ret = -ENXIO;
 679		goto fail_report;
 680	}
 681
 682	while (zone_idx < nr_zones && sector < get_capacity(vblk->disk)) {
 683		memset(report, 0, buflen);
 684
 685		ret = virtblk_submit_zone_report(vblk, (char *)report,
 686						 buflen, sector);
 687		if (ret)
 688			goto fail_report;
 689
 690		nz = min_t(u64, virtio64_to_cpu(vblk->vdev, report->nr_zones),
 691			   nr_zones);
 692		if (!nz)
 693			break;
 694
 695		for (i = 0; i < nz && zone_idx < nr_zones; i++) {
 696			ret = virtblk_parse_zone(vblk, &report->zones[i],
 697						 zone_idx, args);
 698			if (ret)
 699				goto fail_report;
 700
 701			sector = virtio64_to_cpu(vblk->vdev,
 702						 report->zones[i].z_start) +
 703				 vblk->zone_sectors;
 704			zone_idx++;
 705		}
 706	}
 707
 708	if (zone_idx > 0)
 709		ret = zone_idx;
 710	else
 711		ret = -EINVAL;
 712fail_report:
 713	mutex_unlock(&vblk->vdev_mutex);
 714	kvfree(report);
 715	return ret;
 716}
 717
 718static int virtblk_read_zoned_limits(struct virtio_blk *vblk,
 719		struct queue_limits *lim)
 720{
 721	struct virtio_device *vdev = vblk->vdev;
 722	u32 v, wg;
 723
 724	dev_dbg(&vdev->dev, "probing host-managed zoned device\n");
 725
 726	lim->features |= BLK_FEAT_ZONED;
 727
 728	virtio_cread(vdev, struct virtio_blk_config,
 729		     zoned.max_open_zones, &v);
 730	lim->max_open_zones = v;
 731	dev_dbg(&vdev->dev, "max open zones = %u\n", v);
 732
 733	virtio_cread(vdev, struct virtio_blk_config,
 734		     zoned.max_active_zones, &v);
 735	lim->max_active_zones = v;
 736	dev_dbg(&vdev->dev, "max active zones = %u\n", v);
 737
 738	virtio_cread(vdev, struct virtio_blk_config,
 739		     zoned.write_granularity, &wg);
 740	if (!wg) {
 741		dev_warn(&vdev->dev, "zero write granularity reported\n");
 742		return -ENODEV;
 743	}
 744	lim->physical_block_size = wg;
 745	lim->io_min = wg;
 746
 747	dev_dbg(&vdev->dev, "write granularity = %u\n", wg);
 748
 749	/*
 750	 * virtio ZBD specification doesn't require zones to be a power of
 751	 * two sectors in size, but the code in this driver expects that.
 752	 */
 753	virtio_cread(vdev, struct virtio_blk_config, zoned.zone_sectors,
 754		     &vblk->zone_sectors);
 755	if (vblk->zone_sectors == 0 || !is_power_of_2(vblk->zone_sectors)) {
 756		dev_err(&vdev->dev,
 757			"zoned device with non power of two zone size %u\n",
 758			vblk->zone_sectors);
 759		return -ENODEV;
 760	}
 761	lim->chunk_sectors = vblk->zone_sectors;
 762	dev_dbg(&vdev->dev, "zone sectors = %u\n", vblk->zone_sectors);
 763
 764	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
 765		dev_warn(&vblk->vdev->dev,
 766			 "ignoring negotiated F_DISCARD for zoned device\n");
 767		lim->max_hw_discard_sectors = 0;
 768	}
 769
 770	virtio_cread(vdev, struct virtio_blk_config,
 771		     zoned.max_append_sectors, &v);
 772	if (!v) {
 773		dev_warn(&vdev->dev, "zero max_append_sectors reported\n");
 774		return -ENODEV;
 775	}
 776	if ((v << SECTOR_SHIFT) < wg) {
 777		dev_err(&vdev->dev,
 778			"write granularity %u exceeds max_append_sectors %u limit\n",
 779			wg, v);
 780		return -ENODEV;
 781	}
 782	lim->max_hw_zone_append_sectors = v;
 783	dev_dbg(&vdev->dev, "max append sectors = %u\n", v);
 784
 785	return 0;
 786}
 787#else
 788/*
 789 * Zoned block device support is not configured in this kernel, host-managed
 790 * zoned devices can't be supported.
 791 */
 792#define virtblk_report_zones       NULL
 793static inline int virtblk_read_zoned_limits(struct virtio_blk *vblk,
 794		struct queue_limits *lim)
 795{
 796	dev_err(&vblk->vdev->dev,
 797		"virtio_blk: zoned devices are not supported");
 798	return -EOPNOTSUPP;
 799}
 800#endif /* CONFIG_BLK_DEV_ZONED */
 801
 802/* return id (s/n) string for *disk to *id_str
 803 */
 804static int virtblk_get_id(struct gendisk *disk, char *id_str)
 805{
 806	struct virtio_blk *vblk = disk->private_data;
 807	struct request_queue *q = vblk->disk->queue;
 808	struct request *req;
 809	struct virtblk_req *vbr;
 810	int err;
 811
 812	req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
 813	if (IS_ERR(req))
 814		return PTR_ERR(req);
 815
 816	vbr = blk_mq_rq_to_pdu(req);
 817	vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
 818	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
 819	vbr->out_hdr.sector = 0;
 820
 821	err = blk_rq_map_kern(req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
 822	if (err)
 823		goto out;
 824
 825	blk_execute_rq(req, false);
 826	err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
 827out:
 828	blk_mq_free_request(req);
 829	return err;
 830}
 831
 832/* We provide getgeo only to please some old bootloader/partitioning tools */
 833static int virtblk_getgeo(struct gendisk *disk, struct hd_geometry *geo)
 834{
 835	struct virtio_blk *vblk = disk->private_data;
 836	int ret = 0;
 837
 838	mutex_lock(&vblk->vdev_mutex);
 839
 840	if (!vblk->vdev) {
 841		ret = -ENXIO;
 842		goto out;
 843	}
 844
 845	/* see if the host passed in geometry config */
 846	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
 847		virtio_cread(vblk->vdev, struct virtio_blk_config,
 848			     geometry.cylinders, &geo->cylinders);
 849		virtio_cread(vblk->vdev, struct virtio_blk_config,
 850			     geometry.heads, &geo->heads);
 851		virtio_cread(vblk->vdev, struct virtio_blk_config,
 852			     geometry.sectors, &geo->sectors);
 853	} else {
 854		/* some standard values, similar to sd */
 855		geo->heads = 1 << 6;
 856		geo->sectors = 1 << 5;
 857		geo->cylinders = get_capacity(disk) >> 11;
 858	}
 859out:
 860	mutex_unlock(&vblk->vdev_mutex);
 861	return ret;
 862}
 863
 864static void virtblk_free_disk(struct gendisk *disk)
 865{
 866	struct virtio_blk *vblk = disk->private_data;
 867
 868	ida_free(&vd_index_ida, vblk->index);
 869	mutex_destroy(&vblk->vdev_mutex);
 870	kfree(vblk);
 871}
 872
 873static const struct block_device_operations virtblk_fops = {
 874	.owner  	= THIS_MODULE,
 875	.getgeo		= virtblk_getgeo,
 876	.free_disk	= virtblk_free_disk,
 877	.report_zones	= virtblk_report_zones,
 878};
 879
 880static int index_to_minor(int index)
 881{
 882	return index << PART_BITS;
 883}
 884
 885static int minor_to_index(int minor)
 886{
 887	return minor >> PART_BITS;
 888}
 889
 890static ssize_t serial_show(struct device *dev,
 891			   struct device_attribute *attr, char *buf)
 892{
 893	struct gendisk *disk = dev_to_disk(dev);
 894	int err;
 895
 896	/* sysfs gives us a PAGE_SIZE buffer */
 897	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 898
 899	buf[VIRTIO_BLK_ID_BYTES] = '\0';
 900	err = virtblk_get_id(disk, buf);
 901	if (!err)
 902		return strlen(buf);
 903
 904	if (err == -EIO) /* Unsupported? Make it empty. */
 905		return 0;
 906
 907	return err;
 908}
 909
 910static DEVICE_ATTR_RO(serial);
 911
 912/* The queue's logical block size must be set before calling this */
 913static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
 914{
 915	struct virtio_device *vdev = vblk->vdev;
 916	struct request_queue *q = vblk->disk->queue;
 917	char cap_str_2[10], cap_str_10[10];
 918	unsigned long long nblocks;
 919	u64 capacity;
 920
 921	/* Host must always specify the capacity. */
 922	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
 923
 924	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
 925
 926	string_get_size(nblocks, queue_logical_block_size(q),
 927			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
 928	string_get_size(nblocks, queue_logical_block_size(q),
 929			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
 930
 931	dev_notice(&vdev->dev,
 932		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
 933		   vblk->disk->disk_name,
 934		   resize ? "new size: " : "",
 935		   nblocks,
 936		   queue_logical_block_size(q),
 937		   cap_str_10,
 938		   cap_str_2);
 939
 940	set_capacity_and_notify(vblk->disk, capacity);
 941}
 942
 943static void virtblk_config_changed_work(struct work_struct *work)
 944{
 945	struct virtio_blk *vblk =
 946		container_of(work, struct virtio_blk, config_work);
 947
 948	virtblk_update_capacity(vblk, true);
 949}
 950
 951static void virtblk_config_changed(struct virtio_device *vdev)
 952{
 953	struct virtio_blk *vblk = vdev->priv;
 954
 955	queue_work(virtblk_wq, &vblk->config_work);
 956}
 957
 958static int init_vq(struct virtio_blk *vblk)
 959{
 960	int err;
 961	unsigned short i;
 962	struct virtqueue_info *vqs_info;
 963	struct virtqueue **vqs;
 964	unsigned short num_vqs;
 965	unsigned short num_poll_vqs;
 966	struct virtio_device *vdev = vblk->vdev;
 967	struct irq_affinity desc = { 0, };
 968
 969	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
 970				   struct virtio_blk_config, num_queues,
 971				   &num_vqs);
 972	if (err)
 973		num_vqs = 1;
 974
 975	if (!err && !num_vqs) {
 976		dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
 977		return -EINVAL;
 978	}
 979
 980	num_vqs = blk_mq_num_possible_queues(
 981			min_not_zero(num_request_queues, num_vqs));
 982
 983	num_poll_vqs = min_t(unsigned int, poll_queues, num_vqs - 1);
 984
 985	vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
 986	vblk->io_queues[HCTX_TYPE_READ] = 0;
 987	vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
 988
 989	dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n",
 990				vblk->io_queues[HCTX_TYPE_DEFAULT],
 991				vblk->io_queues[HCTX_TYPE_READ],
 992				vblk->io_queues[HCTX_TYPE_POLL]);
 993
 994	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
 995	if (!vblk->vqs)
 996		return -ENOMEM;
 997
 998	vqs_info = kcalloc(num_vqs, sizeof(*vqs_info), GFP_KERNEL);
 999	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
1000	if (!vqs_info || !vqs) {
1001		err = -ENOMEM;
1002		goto out;
1003	}
1004
1005	for (i = 0; i < num_vqs - num_poll_vqs; i++) {
1006		vqs_info[i].callback = virtblk_done;
1007		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%u", i);
1008		vqs_info[i].name = vblk->vqs[i].name;
1009	}
1010
1011	for (; i < num_vqs; i++) {
1012		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%u", i);
1013		vqs_info[i].name = vblk->vqs[i].name;
1014	}
1015
1016	/* Discover virtqueues and write information to configuration.  */
1017	err = virtio_find_vqs(vdev, num_vqs, vqs, vqs_info, &desc);
1018	if (err)
1019		goto out;
1020
1021	for (i = 0; i < num_vqs; i++) {
1022		spin_lock_init(&vblk->vqs[i].lock);
1023		vblk->vqs[i].vq = vqs[i];
1024	}
1025	vblk->num_vqs = num_vqs;
1026
1027out:
1028	kfree(vqs);
1029	kfree(vqs_info);
1030	if (err) {
1031		kfree(vblk->vqs);
1032		/*
1033		 * Set to NULL to prevent freeing vqs again during freezing.
1034		 */
1035		vblk->vqs = NULL;
1036	}
1037	return err;
1038}
1039
1040/*
1041 * Legacy naming scheme used for virtio devices.  We are stuck with it for
1042 * virtio blk but don't ever use it for any new driver.
1043 */
1044static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
1045{
1046	const int base = 'z' - 'a' + 1;
1047	char *begin = buf + strlen(prefix);
1048	char *end = buf + buflen;
1049	char *p;
1050	int unit;
1051
1052	p = end - 1;
1053	*p = '\0';
1054	unit = base;
1055	do {
1056		if (p == begin)
1057			return -EINVAL;
1058		*--p = 'a' + (index % unit);
1059		index = (index / unit) - 1;
1060	} while (index >= 0);
1061
1062	memmove(begin, p, end - p);
1063	memcpy(buf, prefix, strlen(prefix));
1064
1065	return 0;
1066}
1067
1068static int virtblk_get_cache_mode(struct virtio_device *vdev)
1069{
1070	u8 writeback;
1071	int err;
1072
1073	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
1074				   struct virtio_blk_config, wce,
1075				   &writeback);
1076
1077	/*
1078	 * If WCE is not configurable and flush is not available,
1079	 * assume no writeback cache is in use.
1080	 */
1081	if (err)
1082		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
1083
1084	return writeback;
1085}
1086
1087static const char *const virtblk_cache_types[] = {
1088	"write through", "write back"
1089};
1090
1091static ssize_t
1092cache_type_store(struct device *dev, struct device_attribute *attr,
1093		 const char *buf, size_t count)
1094{
1095	struct gendisk *disk = dev_to_disk(dev);
1096	struct virtio_blk *vblk = disk->private_data;
1097	struct virtio_device *vdev = vblk->vdev;
1098	struct queue_limits lim;
1099	int i;
1100
1101	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
1102	i = sysfs_match_string(virtblk_cache_types, buf);
1103	if (i < 0)
1104		return i;
1105
1106	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
1107
1108	lim = queue_limits_start_update(disk->queue);
1109	if (virtblk_get_cache_mode(vdev))
1110		lim.features |= BLK_FEAT_WRITE_CACHE;
1111	else
1112		lim.features &= ~BLK_FEAT_WRITE_CACHE;
1113	i = queue_limits_commit_update_frozen(disk->queue, &lim);
1114	if (i)
1115		return i;
1116	return count;
1117}
1118
1119static ssize_t
1120cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
1121{
1122	struct gendisk *disk = dev_to_disk(dev);
1123	struct virtio_blk *vblk = disk->private_data;
1124	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
1125
1126	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
1127	return sysfs_emit(buf, "%s\n", virtblk_cache_types[writeback]);
1128}
1129
1130static DEVICE_ATTR_RW(cache_type);
1131
1132static struct attribute *virtblk_attrs[] = {
1133	&dev_attr_serial.attr,
1134	&dev_attr_cache_type.attr,
1135	NULL,
1136};
1137
1138static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
1139		struct attribute *a, int n)
1140{
1141	struct device *dev = kobj_to_dev(kobj);
1142	struct gendisk *disk = dev_to_disk(dev);
1143	struct virtio_blk *vblk = disk->private_data;
1144	struct virtio_device *vdev = vblk->vdev;
1145
1146	if (a == &dev_attr_cache_type.attr &&
1147	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
1148		return S_IRUGO;
1149
1150	return a->mode;
1151}
1152
1153static const struct attribute_group virtblk_attr_group = {
1154	.attrs = virtblk_attrs,
1155	.is_visible = virtblk_attrs_are_visible,
1156};
1157
1158static const struct attribute_group *virtblk_attr_groups[] = {
1159	&virtblk_attr_group,
1160	NULL,
1161};
1162
1163static void virtblk_map_queues(struct blk_mq_tag_set *set)
1164{
1165	struct virtio_blk *vblk = set->driver_data;
1166	int i, qoff;
1167
1168	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1169		struct blk_mq_queue_map *map = &set->map[i];
1170
1171		map->nr_queues = vblk->io_queues[i];
1172		map->queue_offset = qoff;
1173		qoff += map->nr_queues;
1174
1175		if (map->nr_queues == 0)
1176			continue;
1177
1178		/*
1179		 * Regular queues have interrupts and hence CPU affinity is
1180		 * defined by the core virtio code, but polling queues have
1181		 * no interrupts so we let the block layer assign CPU affinity.
1182		 */
1183		if (i == HCTX_TYPE_POLL)
1184			blk_mq_map_queues(&set->map[i]);
1185		else
1186			blk_mq_map_hw_queues(&set->map[i],
1187					     &vblk->vdev->dev, 0);
1188	}
1189}
1190
1191static void virtblk_complete_batch(struct io_comp_batch *iob)
1192{
1193	struct request *req;
1194
1195	rq_list_for_each(&iob->req_list, req) {
1196		virtblk_unmap_data(req, blk_mq_rq_to_pdu(req));
1197		virtblk_cleanup_cmd(req);
1198	}
1199	blk_mq_end_request_batch(iob);
1200}
1201
1202static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1203{
1204	struct virtio_blk *vblk = hctx->queue->queuedata;
1205	struct virtio_blk_vq *vq = get_virtio_blk_vq(hctx);
1206	struct virtblk_req *vbr;
1207	unsigned long flags;
1208	unsigned int len;
1209	int found = 0;
1210
1211	spin_lock_irqsave(&vq->lock, flags);
1212
1213	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
1214		struct request *req = blk_mq_rq_from_pdu(vbr);
1215		u8 status = virtblk_vbr_status(vbr);
1216
1217		found++;
1218		if (!blk_mq_complete_request_remote(req) &&
1219		    !blk_mq_add_to_batch(req, iob, status != VIRTIO_BLK_S_OK,
1220					 virtblk_complete_batch))
1221			virtblk_request_done(req);
1222	}
1223
1224	if (found)
1225		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
1226
1227	spin_unlock_irqrestore(&vq->lock, flags);
1228
1229	return found;
1230}
1231
1232static const struct blk_mq_ops virtio_mq_ops = {
1233	.queue_rq	= virtio_queue_rq,
1234	.queue_rqs	= virtio_queue_rqs,
1235	.commit_rqs	= virtio_commit_rqs,
1236	.complete	= virtblk_request_done,
1237	.map_queues	= virtblk_map_queues,
1238	.poll		= virtblk_poll,
1239};
1240
1241static unsigned int virtblk_queue_depth;
1242module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
1243
1244static int virtblk_read_limits(struct virtio_blk *vblk,
1245		struct queue_limits *lim)
1246{
1247	struct virtio_device *vdev = vblk->vdev;
1248	u32 v, max_size, sg_elems, opt_io_size;
1249	u32 max_discard_segs = 0;
1250	u32 discard_granularity = 0;
1251	u16 min_io_size;
1252	u8 physical_block_exp, alignment_offset;
1253	size_t max_dma_size;
1254	int err;
1255
1256	/* We need to know how many segments before we allocate. */
1257	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
1258				   struct virtio_blk_config, seg_max,
1259				   &sg_elems);
1260
1261	/* We need at least one SG element, whatever they say. */
1262	if (err || !sg_elems)
1263		sg_elems = 1;
1264
1265	/* Prevent integer overflows and honor max vq size */
1266	sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
1267
1268	/* We can handle whatever the host told us to handle. */
1269	lim->max_segments = sg_elems;
1270
1271	/* No real sector limit. */
1272	lim->max_hw_sectors = UINT_MAX;
1273
1274	max_dma_size = virtio_max_dma_size(vdev);
1275	max_size = max_dma_size > U32_MAX ? U32_MAX : max_dma_size;
1276
1277	/* Host can optionally specify maximum segment size and number of
1278	 * segments. */
1279	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
1280				   struct virtio_blk_config, size_max, &v);
1281	if (!err)
1282		max_size = min(max_size, v);
1283
1284	lim->max_segment_size = max_size;
1285
1286	/* Host can optionally specify the block size of the device */
1287	virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
1288				   struct virtio_blk_config, blk_size,
1289				   &lim->logical_block_size);
1290
1291	/* Use topology information if available */
1292	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1293				   struct virtio_blk_config, physical_block_exp,
1294				   &physical_block_exp);
1295	if (!err && physical_block_exp)
1296		lim->physical_block_size =
1297			lim->logical_block_size * (1 << physical_block_exp);
1298
1299	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1300				   struct virtio_blk_config, alignment_offset,
1301				   &alignment_offset);
1302	if (!err && alignment_offset)
1303		lim->alignment_offset =
1304			lim->logical_block_size * alignment_offset;
1305
1306	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1307				   struct virtio_blk_config, min_io_size,
1308				   &min_io_size);
1309	if (!err && min_io_size)
1310		lim->io_min = lim->logical_block_size * min_io_size;
1311
1312	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1313				   struct virtio_blk_config, opt_io_size,
1314				   &opt_io_size);
1315	if (!err && opt_io_size)
1316		lim->io_opt = lim->logical_block_size * opt_io_size;
1317
1318	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
1319		virtio_cread(vdev, struct virtio_blk_config,
1320			     discard_sector_alignment, &discard_granularity);
1321
1322		virtio_cread(vdev, struct virtio_blk_config,
1323			     max_discard_sectors, &v);
1324		lim->max_hw_discard_sectors = v ? v : UINT_MAX;
1325
1326		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
1327			     &max_discard_segs);
1328	}
1329
1330	if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
1331		virtio_cread(vdev, struct virtio_blk_config,
1332			     max_write_zeroes_sectors, &v);
1333		lim->max_write_zeroes_sectors = v ? v : UINT_MAX;
1334	}
1335
1336	/* The discard and secure erase limits are combined since the Linux
1337	 * block layer uses the same limit for both commands.
1338	 *
1339	 * If both VIRTIO_BLK_F_SECURE_ERASE and VIRTIO_BLK_F_DISCARD features
1340	 * are negotiated, we will use the minimum between the limits.
1341	 *
1342	 * discard sector alignment is set to the minimum between discard_sector_alignment
1343	 * and secure_erase_sector_alignment.
1344	 *
1345	 * max discard sectors is set to the minimum between max_discard_seg and
1346	 * max_secure_erase_seg.
1347	 */
1348	if (virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1349
1350		virtio_cread(vdev, struct virtio_blk_config,
1351			     secure_erase_sector_alignment, &v);
1352
1353		/* secure_erase_sector_alignment should not be zero, the device should set a
1354		 * valid number of sectors.
1355		 */
1356		if (!v) {
1357			dev_err(&vdev->dev,
1358				"virtio_blk: secure_erase_sector_alignment can't be 0\n");
1359			return -EINVAL;
1360		}
1361
1362		discard_granularity = min_not_zero(discard_granularity, v);
1363
1364		virtio_cread(vdev, struct virtio_blk_config,
1365			     max_secure_erase_sectors, &v);
1366
1367		/* max_secure_erase_sectors should not be zero, the device should set a
1368		 * valid number of sectors.
1369		 */
1370		if (!v) {
1371			dev_err(&vdev->dev,
1372				"virtio_blk: max_secure_erase_sectors can't be 0\n");
1373			return -EINVAL;
1374		}
1375
1376		lim->max_secure_erase_sectors = v;
1377
1378		virtio_cread(vdev, struct virtio_blk_config,
1379			     max_secure_erase_seg, &v);
1380
1381		/* max_secure_erase_seg should not be zero, the device should set a
1382		 * valid number of segments
1383		 */
1384		if (!v) {
1385			dev_err(&vdev->dev,
1386				"virtio_blk: max_secure_erase_seg can't be 0\n");
1387			return -EINVAL;
1388		}
1389
1390		max_discard_segs = min_not_zero(max_discard_segs, v);
1391	}
1392
1393	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD) ||
1394	    virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1395		/* max_discard_seg and discard_granularity will be 0 only
1396		 * if max_discard_seg and discard_sector_alignment fields in the virtio
1397		 * config are 0 and VIRTIO_BLK_F_SECURE_ERASE feature is not negotiated.
1398		 * In this case, we use default values.
1399		 */
1400		if (!max_discard_segs)
1401			max_discard_segs = sg_elems;
1402
1403		lim->max_discard_segments =
1404			min(max_discard_segs, MAX_DISCARD_SEGMENTS);
1405
1406		if (discard_granularity)
1407			lim->discard_granularity =
1408				discard_granularity << SECTOR_SHIFT;
1409		else
1410			lim->discard_granularity = lim->logical_block_size;
1411	}
1412
1413	if (virtio_has_feature(vdev, VIRTIO_BLK_F_ZONED)) {
1414		u8 model;
1415
1416		virtio_cread(vdev, struct virtio_blk_config, zoned.model, &model);
1417		switch (model) {
1418		case VIRTIO_BLK_Z_NONE:
1419		case VIRTIO_BLK_Z_HA:
1420			/* treat host-aware devices as non-zoned */
1421			return 0;
1422		case VIRTIO_BLK_Z_HM:
1423			err = virtblk_read_zoned_limits(vblk, lim);
1424			if (err)
1425				return err;
1426			break;
1427		default:
1428			dev_err(&vdev->dev, "unsupported zone model %d\n", model);
1429			return -EINVAL;
1430		}
1431	}
1432
1433	return 0;
1434}
1435
1436static int virtblk_probe(struct virtio_device *vdev)
1437{
1438	struct virtio_blk *vblk;
1439	struct queue_limits lim = {
1440		.features		= BLK_FEAT_ROTATIONAL,
1441		.logical_block_size	= SECTOR_SIZE,
1442	};
1443	int err, index;
1444	unsigned int queue_depth;
1445
1446	if (!vdev->config->get) {
1447		dev_err(&vdev->dev, "%s failure: config access disabled\n",
1448			__func__);
1449		return -EINVAL;
1450	}
1451
1452	err = ida_alloc_range(&vd_index_ida, 0,
1453			      minor_to_index(1 << MINORBITS) - 1, GFP_KERNEL);
1454	if (err < 0)
1455		goto out;
1456	index = err;
1457
1458	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
1459	if (!vblk) {
1460		err = -ENOMEM;
1461		goto out_free_index;
1462	}
1463
1464	mutex_init(&vblk->vdev_mutex);
1465
1466	vblk->vdev = vdev;
1467
1468	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
1469
1470	err = init_vq(vblk);
1471	if (err)
1472		goto out_free_vblk;
1473
1474	/* Default queue sizing is to fill the ring. */
1475	if (!virtblk_queue_depth) {
1476		queue_depth = vblk->vqs[0].vq->num_free;
1477		/* ... but without indirect descs, we use 2 descs per req */
1478		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
1479			queue_depth /= 2;
1480	} else {
1481		queue_depth = virtblk_queue_depth;
1482	}
1483
1484	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
1485	vblk->tag_set.ops = &virtio_mq_ops;
1486	vblk->tag_set.queue_depth = queue_depth;
1487	vblk->tag_set.numa_node = NUMA_NO_NODE;
1488	vblk->tag_set.cmd_size =
1489		sizeof(struct virtblk_req) +
1490		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
1491	vblk->tag_set.driver_data = vblk;
1492	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
1493	vblk->tag_set.nr_maps = 1;
1494	if (vblk->io_queues[HCTX_TYPE_POLL])
1495		vblk->tag_set.nr_maps = 3;
1496
1497	err = blk_mq_alloc_tag_set(&vblk->tag_set);
1498	if (err)
1499		goto out_free_vq;
1500
1501	err = virtblk_read_limits(vblk, &lim);
1502	if (err)
1503		goto out_free_tags;
1504
1505	if (virtblk_get_cache_mode(vdev))
1506		lim.features |= BLK_FEAT_WRITE_CACHE;
1507
1508	vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, &lim, vblk);
1509	if (IS_ERR(vblk->disk)) {
1510		err = PTR_ERR(vblk->disk);
1511		goto out_free_tags;
1512	}
1513
1514	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
1515
1516	vblk->disk->major = major;
1517	vblk->disk->first_minor = index_to_minor(index);
1518	vblk->disk->minors = 1 << PART_BITS;
1519	vblk->disk->private_data = vblk;
1520	vblk->disk->fops = &virtblk_fops;
1521	vblk->index = index;
1522
1523	/* If disk is read-only in the host, the guest should obey */
1524	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
1525		set_disk_ro(vblk->disk, 1);
1526
1527	virtblk_update_capacity(vblk, false);
1528	virtio_device_ready(vdev);
1529
1530	/*
1531	 * All steps that follow use the VQs therefore they need to be
1532	 * placed after the virtio_device_ready() call above.
1533	 */
1534	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
1535	    (lim.features & BLK_FEAT_ZONED)) {
1536		err = blk_revalidate_disk_zones(vblk->disk);
1537		if (err)
1538			goto out_cleanup_disk;
1539	}
1540
1541	err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
1542	if (err)
1543		goto out_cleanup_disk;
1544
1545	return 0;
1546
1547out_cleanup_disk:
1548	put_disk(vblk->disk);
1549out_free_tags:
1550	blk_mq_free_tag_set(&vblk->tag_set);
1551out_free_vq:
1552	vdev->config->del_vqs(vdev);
1553	kfree(vblk->vqs);
1554out_free_vblk:
1555	kfree(vblk);
1556out_free_index:
1557	ida_free(&vd_index_ida, index);
1558out:
1559	return err;
1560}
1561
1562static void virtblk_remove(struct virtio_device *vdev)
1563{
1564	struct virtio_blk *vblk = vdev->priv;
1565
1566	/* Make sure no work handler is accessing the device. */
1567	flush_work(&vblk->config_work);
1568
1569	del_gendisk(vblk->disk);
1570	blk_mq_free_tag_set(&vblk->tag_set);
1571
1572	mutex_lock(&vblk->vdev_mutex);
1573
1574	/* Stop all the virtqueues. */
1575	virtio_reset_device(vdev);
1576
1577	/* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
1578	vblk->vdev = NULL;
1579
1580	vdev->config->del_vqs(vdev);
1581	kfree(vblk->vqs);
1582
1583	mutex_unlock(&vblk->vdev_mutex);
1584
1585	put_disk(vblk->disk);
1586}
1587
1588static int virtblk_freeze_priv(struct virtio_device *vdev)
1589{
1590	struct virtio_blk *vblk = vdev->priv;
1591	struct request_queue *q = vblk->disk->queue;
1592	unsigned int memflags;
1593
1594	/* Ensure no requests in virtqueues before deleting vqs. */
1595	memflags = blk_mq_freeze_queue(q);
1596	blk_mq_quiesce_queue_nowait(q);
1597	blk_mq_unfreeze_queue(q, memflags);
1598
1599	/* Ensure we don't receive any more interrupts */
1600	virtio_reset_device(vdev);
1601
1602	/* Make sure no work handler is accessing the device. */
1603	flush_work(&vblk->config_work);
1604
1605	vdev->config->del_vqs(vdev);
1606	kfree(vblk->vqs);
1607	/*
1608	 * Set to NULL to prevent freeing vqs again after a failed vqs
1609	 * allocation during resume. Note that kfree() already handles NULL
1610	 * pointers safely.
1611	 */
1612	vblk->vqs = NULL;
1613
1614	return 0;
1615}
1616
1617static int virtblk_restore_priv(struct virtio_device *vdev)
1618{
1619	struct virtio_blk *vblk = vdev->priv;
1620	int ret;
1621
1622	ret = init_vq(vdev->priv);
1623	if (ret)
1624		return ret;
1625
1626	virtio_device_ready(vdev);
1627	blk_mq_unquiesce_queue(vblk->disk->queue);
1628
1629	return 0;
1630}
1631
1632#ifdef CONFIG_PM_SLEEP
1633static int virtblk_freeze(struct virtio_device *vdev)
1634{
1635	return virtblk_freeze_priv(vdev);
1636}
1637
1638static int virtblk_restore(struct virtio_device *vdev)
1639{
1640	return virtblk_restore_priv(vdev);
1641}
1642#endif
1643
1644static int virtblk_reset_prepare(struct virtio_device *vdev)
1645{
1646	return virtblk_freeze_priv(vdev);
1647}
1648
1649static int virtblk_reset_done(struct virtio_device *vdev)
1650{
1651	return virtblk_restore_priv(vdev);
1652}
1653
1654static const struct virtio_device_id id_table[] = {
1655	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1656	{ 0 },
1657};
1658
1659static unsigned int features_legacy[] = {
1660	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1661	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1662	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1663	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1664	VIRTIO_BLK_F_SECURE_ERASE,
1665}
1666;
1667static unsigned int features[] = {
1668	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1669	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1670	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1671	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1672	VIRTIO_BLK_F_SECURE_ERASE, VIRTIO_BLK_F_ZONED,
1673};
1674
1675static struct virtio_driver virtio_blk = {
1676	.feature_table			= features,
1677	.feature_table_size		= ARRAY_SIZE(features),
1678	.feature_table_legacy		= features_legacy,
1679	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
1680	.driver.name			= KBUILD_MODNAME,
1681	.id_table			= id_table,
1682	.probe				= virtblk_probe,
1683	.remove				= virtblk_remove,
1684	.config_changed			= virtblk_config_changed,
1685#ifdef CONFIG_PM_SLEEP
1686	.freeze				= virtblk_freeze,
1687	.restore			= virtblk_restore,
1688#endif
1689	.reset_prepare			= virtblk_reset_prepare,
1690	.reset_done			= virtblk_reset_done,
1691};
1692
1693static int __init virtio_blk_init(void)
1694{
1695	int error;
1696
1697	virtblk_wq = alloc_workqueue("virtio-blk", WQ_PERCPU, 0);
1698	if (!virtblk_wq)
1699		return -ENOMEM;
1700
1701	major = register_blkdev(0, "virtblk");
1702	if (major < 0) {
1703		error = major;
1704		goto out_destroy_workqueue;
1705	}
1706
1707	error = register_virtio_driver(&virtio_blk);
1708	if (error)
1709		goto out_unregister_blkdev;
1710	return 0;
1711
1712out_unregister_blkdev:
1713	unregister_blkdev(major, "virtblk");
1714out_destroy_workqueue:
1715	destroy_workqueue(virtblk_wq);
1716	return error;
1717}
1718
1719static void __exit virtio_blk_fini(void)
1720{
1721	unregister_virtio_driver(&virtio_blk);
1722	unregister_blkdev(major, "virtblk");
1723	destroy_workqueue(virtblk_wq);
1724}
1725module_init(virtio_blk_init);
1726module_exit(virtio_blk_fini);
1727
1728MODULE_DEVICE_TABLE(virtio, id_table);
1729MODULE_DESCRIPTION("Virtio block driver");
1730MODULE_LICENSE("GPL");