xref: /linux/drivers/block/virtio_blk.c (revision 60684c2bd35064043360e6f716d1b7c20e967b7d)
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 <linux/vmalloc.h>
19 #include <uapi/linux/virtio_ring.h>
20 
21 #define PART_BITS 4
22 #define VQ_NAME_LEN 16
23 #define MAX_DISCARD_SEGMENTS 256u
24 
25 /* The maximum number of sg elements that fit into a virtqueue */
26 #define VIRTIO_BLK_MAX_SG_ELEMS 32768
27 
28 #ifdef CONFIG_ARCH_NO_SG_CHAIN
29 #define VIRTIO_BLK_INLINE_SG_CNT	0
30 #else
31 #define VIRTIO_BLK_INLINE_SG_CNT	2
32 #endif
33 
34 static unsigned int num_request_queues;
35 module_param(num_request_queues, uint, 0644);
36 MODULE_PARM_DESC(num_request_queues,
37 		 "Limit the number of request queues to use for blk device. "
38 		 "0 for no limit. "
39 		 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
40 
41 static unsigned int poll_queues;
42 module_param(poll_queues, uint, 0644);
43 MODULE_PARM_DESC(poll_queues, "The number of dedicated virtqueues for polling I/O");
44 
45 static int major;
46 static DEFINE_IDA(vd_index_ida);
47 
48 static struct workqueue_struct *virtblk_wq;
49 
50 struct virtio_blk_vq {
51 	struct virtqueue *vq;
52 	spinlock_t lock;
53 	char name[VQ_NAME_LEN];
54 } ____cacheline_aligned_in_smp;
55 
56 struct virtio_blk {
57 	/*
58 	 * This mutex must be held by anything that may run after
59 	 * virtblk_remove() sets vblk->vdev to NULL.
60 	 *
61 	 * blk-mq, virtqueue processing, and sysfs attribute code paths are
62 	 * shut down before vblk->vdev is set to NULL and therefore do not need
63 	 * to hold this mutex.
64 	 */
65 	struct mutex vdev_mutex;
66 	struct virtio_device *vdev;
67 
68 	/* The disk structure for the kernel. */
69 	struct gendisk *disk;
70 
71 	/* Block layer tags. */
72 	struct blk_mq_tag_set tag_set;
73 
74 	/* Process context for config space updates */
75 	struct work_struct config_work;
76 
77 	/* Ida index - used to track minor number allocations. */
78 	int index;
79 
80 	/* num of vqs */
81 	int num_vqs;
82 	int io_queues[HCTX_MAX_TYPES];
83 	struct virtio_blk_vq *vqs;
84 
85 	/* For zoned device */
86 	unsigned int zone_sectors;
87 };
88 
89 struct virtblk_req {
90 	/* Out header */
91 	struct virtio_blk_outhdr out_hdr;
92 
93 	/* In header */
94 	union {
95 		u8 status;
96 
97 		/*
98 		 * The zone append command has an extended in header.
99 		 * The status field in zone_append_in_hdr must have
100 		 * the same offset in virtblk_req as the non-zoned
101 		 * status field above.
102 		 */
103 		struct {
104 			u8 status;
105 			u8 reserved[7];
106 			__le64 append_sector;
107 		} zone_append_in_hdr;
108 	};
109 
110 	size_t in_hdr_len;
111 
112 	struct sg_table sg_table;
113 	struct scatterlist sg[];
114 };
115 
116 static inline blk_status_t virtblk_result(u8 status)
117 {
118 	switch (status) {
119 	case VIRTIO_BLK_S_OK:
120 		return BLK_STS_OK;
121 	case VIRTIO_BLK_S_UNSUPP:
122 		return BLK_STS_NOTSUPP;
123 	case VIRTIO_BLK_S_ZONE_OPEN_RESOURCE:
124 		return BLK_STS_ZONE_OPEN_RESOURCE;
125 	case VIRTIO_BLK_S_ZONE_ACTIVE_RESOURCE:
126 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
127 	case VIRTIO_BLK_S_IOERR:
128 	case VIRTIO_BLK_S_ZONE_UNALIGNED_WP:
129 	default:
130 		return BLK_STS_IOERR;
131 	}
132 }
133 
134 static inline struct virtio_blk_vq *get_virtio_blk_vq(struct blk_mq_hw_ctx *hctx)
135 {
136 	struct virtio_blk *vblk = hctx->queue->queuedata;
137 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
138 
139 	return vq;
140 }
141 
142 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr)
143 {
144 	struct scatterlist out_hdr, in_hdr, *sgs[3];
145 	unsigned int num_out = 0, num_in = 0;
146 
147 	sg_init_one(&out_hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
148 	sgs[num_out++] = &out_hdr;
149 
150 	if (vbr->sg_table.nents) {
151 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
152 			sgs[num_out++] = vbr->sg_table.sgl;
153 		else
154 			sgs[num_out + num_in++] = vbr->sg_table.sgl;
155 	}
156 
157 	sg_init_one(&in_hdr, &vbr->status, vbr->in_hdr_len);
158 	sgs[num_out + num_in++] = &in_hdr;
159 
160 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
161 }
162 
163 static int virtblk_setup_discard_write_zeroes_erase(struct request *req, bool unmap)
164 {
165 	unsigned short segments = blk_rq_nr_discard_segments(req);
166 	unsigned short n = 0;
167 	struct virtio_blk_discard_write_zeroes *range;
168 	struct bio *bio;
169 	u32 flags = 0;
170 
171 	if (unmap)
172 		flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
173 
174 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
175 	if (!range)
176 		return -ENOMEM;
177 
178 	/*
179 	 * Single max discard segment means multi-range discard isn't
180 	 * supported, and block layer only runs contiguity merge like
181 	 * normal RW request. So we can't reply on bio for retrieving
182 	 * each range info.
183 	 */
184 	if (queue_max_discard_segments(req->q) == 1) {
185 		range[0].flags = cpu_to_le32(flags);
186 		range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
187 		range[0].sector = cpu_to_le64(blk_rq_pos(req));
188 		n = 1;
189 	} else {
190 		__rq_for_each_bio(bio, req) {
191 			u64 sector = bio->bi_iter.bi_sector;
192 			u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
193 
194 			range[n].flags = cpu_to_le32(flags);
195 			range[n].num_sectors = cpu_to_le32(num_sectors);
196 			range[n].sector = cpu_to_le64(sector);
197 			n++;
198 		}
199 	}
200 
201 	WARN_ON_ONCE(n != segments);
202 
203 	bvec_set_virt(&req->special_vec, range, sizeof(*range) * segments);
204 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
205 
206 	return 0;
207 }
208 
209 static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
210 {
211 	if (blk_rq_nr_phys_segments(req))
212 		sg_free_table_chained(&vbr->sg_table,
213 				      VIRTIO_BLK_INLINE_SG_CNT);
214 }
215 
216 static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
217 		struct virtblk_req *vbr)
218 {
219 	int err;
220 
221 	if (!blk_rq_nr_phys_segments(req))
222 		return 0;
223 
224 	vbr->sg_table.sgl = vbr->sg;
225 	err = sg_alloc_table_chained(&vbr->sg_table,
226 				     blk_rq_nr_phys_segments(req),
227 				     vbr->sg_table.sgl,
228 				     VIRTIO_BLK_INLINE_SG_CNT);
229 	if (unlikely(err))
230 		return -ENOMEM;
231 
232 	return blk_rq_map_sg(hctx->queue, req, vbr->sg_table.sgl);
233 }
234 
235 static void virtblk_cleanup_cmd(struct request *req)
236 {
237 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
238 		kfree(bvec_virt(&req->special_vec));
239 }
240 
241 static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
242 				      struct request *req,
243 				      struct virtblk_req *vbr)
244 {
245 	size_t in_hdr_len = sizeof(vbr->status);
246 	bool unmap = false;
247 	u32 type;
248 	u64 sector = 0;
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->zone_append_in_hdr);
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 		/* Out header already filled in, nothing to do */
301 		return 0;
302 	default:
303 		WARN_ON_ONCE(1);
304 		return BLK_STS_IOERR;
305 	}
306 
307 	/* Set fields for non-REQ_OP_DRV_IN request types */
308 	vbr->in_hdr_len = in_hdr_len;
309 	vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
310 	vbr->out_hdr.sector = cpu_to_virtio64(vdev, sector);
311 
312 	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES ||
313 	    type == VIRTIO_BLK_T_SECURE_ERASE) {
314 		if (virtblk_setup_discard_write_zeroes_erase(req, unmap))
315 			return BLK_STS_RESOURCE;
316 	}
317 
318 	return 0;
319 }
320 
321 static inline void virtblk_request_done(struct request *req)
322 {
323 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
324 	blk_status_t status = virtblk_result(vbr->status);
325 
326 	virtblk_unmap_data(req, vbr);
327 	virtblk_cleanup_cmd(req);
328 
329 	if (req_op(req) == REQ_OP_ZONE_APPEND)
330 		req->__sector = le64_to_cpu(vbr->zone_append_in_hdr.append_sector);
331 
332 	blk_mq_end_request(req, status);
333 }
334 
335 static void virtblk_complete_batch(struct io_comp_batch *iob)
336 {
337 	struct request *req;
338 
339 	rq_list_for_each(&iob->req_list, req) {
340 		virtblk_unmap_data(req, blk_mq_rq_to_pdu(req));
341 		virtblk_cleanup_cmd(req);
342 	}
343 	blk_mq_end_request_batch(iob);
344 }
345 
346 static int virtblk_handle_req(struct virtio_blk_vq *vq,
347 			      struct io_comp_batch *iob)
348 {
349 	struct virtblk_req *vbr;
350 	int req_done = 0;
351 	unsigned int len;
352 
353 	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
354 		struct request *req = blk_mq_rq_from_pdu(vbr);
355 
356 		if (likely(!blk_should_fake_timeout(req->q)) &&
357 		    !blk_mq_complete_request_remote(req) &&
358 		    !blk_mq_add_to_batch(req, iob, vbr->status,
359 					 virtblk_complete_batch))
360 			virtblk_request_done(req);
361 		req_done++;
362 	}
363 
364 	return req_done;
365 }
366 
367 static void virtblk_done(struct virtqueue *vq)
368 {
369 	struct virtio_blk *vblk = vq->vdev->priv;
370 	struct virtio_blk_vq *vblk_vq = &vblk->vqs[vq->index];
371 	int req_done = 0;
372 	unsigned long flags;
373 	DEFINE_IO_COMP_BATCH(iob);
374 
375 	spin_lock_irqsave(&vblk_vq->lock, flags);
376 	do {
377 		virtqueue_disable_cb(vq);
378 		req_done += virtblk_handle_req(vblk_vq, &iob);
379 
380 		if (unlikely(virtqueue_is_broken(vq)))
381 			break;
382 	} while (!virtqueue_enable_cb(vq));
383 
384 	if (req_done) {
385 		if (!rq_list_empty(iob.req_list))
386 			iob.complete(&iob);
387 
388 		/* In case queue is stopped waiting for more buffers. */
389 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
390 	}
391 	spin_unlock_irqrestore(&vblk_vq->lock, flags);
392 }
393 
394 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
395 {
396 	struct virtio_blk *vblk = hctx->queue->queuedata;
397 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
398 	bool kick;
399 
400 	spin_lock_irq(&vq->lock);
401 	kick = virtqueue_kick_prepare(vq->vq);
402 	spin_unlock_irq(&vq->lock);
403 
404 	if (kick)
405 		virtqueue_notify(vq->vq);
406 }
407 
408 static blk_status_t virtblk_fail_to_queue(struct request *req, int rc)
409 {
410 	virtblk_cleanup_cmd(req);
411 	switch (rc) {
412 	case -ENOSPC:
413 		return BLK_STS_DEV_RESOURCE;
414 	case -ENOMEM:
415 		return BLK_STS_RESOURCE;
416 	default:
417 		return BLK_STS_IOERR;
418 	}
419 }
420 
421 static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
422 					struct virtio_blk *vblk,
423 					struct request *req,
424 					struct virtblk_req *vbr)
425 {
426 	blk_status_t status;
427 	int num;
428 
429 	status = virtblk_setup_cmd(vblk->vdev, req, vbr);
430 	if (unlikely(status))
431 		return status;
432 
433 	num = virtblk_map_data(hctx, req, vbr);
434 	if (unlikely(num < 0))
435 		return virtblk_fail_to_queue(req, -ENOMEM);
436 	vbr->sg_table.nents = num;
437 
438 	blk_mq_start_request(req);
439 
440 	return BLK_STS_OK;
441 }
442 
443 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
444 			   const struct blk_mq_queue_data *bd)
445 {
446 	struct virtio_blk *vblk = hctx->queue->queuedata;
447 	struct request *req = bd->rq;
448 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
449 	unsigned long flags;
450 	int qid = hctx->queue_num;
451 	bool notify = false;
452 	blk_status_t status;
453 	int err;
454 
455 	status = virtblk_prep_rq(hctx, vblk, req, vbr);
456 	if (unlikely(status))
457 		return status;
458 
459 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
460 	err = virtblk_add_req(vblk->vqs[qid].vq, vbr);
461 	if (err) {
462 		virtqueue_kick(vblk->vqs[qid].vq);
463 		/* Don't stop the queue if -ENOMEM: we may have failed to
464 		 * bounce the buffer due to global resource outage.
465 		 */
466 		if (err == -ENOSPC)
467 			blk_mq_stop_hw_queue(hctx);
468 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
469 		virtblk_unmap_data(req, vbr);
470 		return virtblk_fail_to_queue(req, err);
471 	}
472 
473 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
474 		notify = true;
475 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
476 
477 	if (notify)
478 		virtqueue_notify(vblk->vqs[qid].vq);
479 	return BLK_STS_OK;
480 }
481 
482 static bool virtblk_prep_rq_batch(struct request *req)
483 {
484 	struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
485 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
486 
487 	req->mq_hctx->tags->rqs[req->tag] = req;
488 
489 	return virtblk_prep_rq(req->mq_hctx, vblk, req, vbr) == BLK_STS_OK;
490 }
491 
492 static bool virtblk_add_req_batch(struct virtio_blk_vq *vq,
493 					struct request **rqlist)
494 {
495 	unsigned long flags;
496 	int err;
497 	bool kick;
498 
499 	spin_lock_irqsave(&vq->lock, flags);
500 
501 	while (!rq_list_empty(*rqlist)) {
502 		struct request *req = rq_list_pop(rqlist);
503 		struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
504 
505 		err = virtblk_add_req(vq->vq, vbr);
506 		if (err) {
507 			virtblk_unmap_data(req, vbr);
508 			virtblk_cleanup_cmd(req);
509 			blk_mq_requeue_request(req, true);
510 		}
511 	}
512 
513 	kick = virtqueue_kick_prepare(vq->vq);
514 	spin_unlock_irqrestore(&vq->lock, flags);
515 
516 	return kick;
517 }
518 
519 static void virtio_queue_rqs(struct request **rqlist)
520 {
521 	struct request *req, *next, *prev = NULL;
522 	struct request *requeue_list = NULL;
523 
524 	rq_list_for_each_safe(rqlist, req, next) {
525 		struct virtio_blk_vq *vq = get_virtio_blk_vq(req->mq_hctx);
526 		bool kick;
527 
528 		if (!virtblk_prep_rq_batch(req)) {
529 			rq_list_move(rqlist, &requeue_list, req, prev);
530 			req = prev;
531 			if (!req)
532 				continue;
533 		}
534 
535 		if (!next || req->mq_hctx != next->mq_hctx) {
536 			req->rq_next = NULL;
537 			kick = virtblk_add_req_batch(vq, rqlist);
538 			if (kick)
539 				virtqueue_notify(vq->vq);
540 
541 			*rqlist = next;
542 			prev = NULL;
543 		} else
544 			prev = req;
545 	}
546 
547 	*rqlist = requeue_list;
548 }
549 
550 #ifdef CONFIG_BLK_DEV_ZONED
551 static void *virtblk_alloc_report_buffer(struct virtio_blk *vblk,
552 					  unsigned int nr_zones,
553 					  unsigned int zone_sectors,
554 					  size_t *buflen)
555 {
556 	struct request_queue *q = vblk->disk->queue;
557 	size_t bufsize;
558 	void *buf;
559 
560 	nr_zones = min_t(unsigned int, nr_zones,
561 			 get_capacity(vblk->disk) >> ilog2(zone_sectors));
562 
563 	bufsize = sizeof(struct virtio_blk_zone_report) +
564 		nr_zones * sizeof(struct virtio_blk_zone_descriptor);
565 	bufsize = min_t(size_t, bufsize,
566 			queue_max_hw_sectors(q) << SECTOR_SHIFT);
567 	bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
568 
569 	while (bufsize >= sizeof(struct virtio_blk_zone_report)) {
570 		buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
571 		if (buf) {
572 			*buflen = bufsize;
573 			return buf;
574 		}
575 		bufsize >>= 1;
576 	}
577 
578 	return NULL;
579 }
580 
581 static int virtblk_submit_zone_report(struct virtio_blk *vblk,
582 				       char *report_buf, size_t report_len,
583 				       sector_t sector)
584 {
585 	struct request_queue *q = vblk->disk->queue;
586 	struct request *req;
587 	struct virtblk_req *vbr;
588 	int err;
589 
590 	req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
591 	if (IS_ERR(req))
592 		return PTR_ERR(req);
593 
594 	vbr = blk_mq_rq_to_pdu(req);
595 	vbr->in_hdr_len = sizeof(vbr->status);
596 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_ZONE_REPORT);
597 	vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, sector);
598 
599 	err = blk_rq_map_kern(q, req, report_buf, report_len, GFP_KERNEL);
600 	if (err)
601 		goto out;
602 
603 	blk_execute_rq(req, false);
604 	err = blk_status_to_errno(virtblk_result(vbr->status));
605 out:
606 	blk_mq_free_request(req);
607 	return err;
608 }
609 
610 static int virtblk_parse_zone(struct virtio_blk *vblk,
611 			       struct virtio_blk_zone_descriptor *entry,
612 			       unsigned int idx, unsigned int zone_sectors,
613 			       report_zones_cb cb, void *data)
614 {
615 	struct blk_zone zone = { };
616 
617 	if (entry->z_type != VIRTIO_BLK_ZT_SWR &&
618 	    entry->z_type != VIRTIO_BLK_ZT_SWP &&
619 	    entry->z_type != VIRTIO_BLK_ZT_CONV) {
620 		dev_err(&vblk->vdev->dev, "invalid zone type %#x\n",
621 			entry->z_type);
622 		return -EINVAL;
623 	}
624 
625 	zone.type = entry->z_type;
626 	zone.cond = entry->z_state;
627 	zone.len = zone_sectors;
628 	zone.capacity = le64_to_cpu(entry->z_cap);
629 	zone.start = le64_to_cpu(entry->z_start);
630 	if (zone.cond == BLK_ZONE_COND_FULL)
631 		zone.wp = zone.start + zone.len;
632 	else
633 		zone.wp = le64_to_cpu(entry->z_wp);
634 
635 	return cb(&zone, idx, data);
636 }
637 
638 static int virtblk_report_zones(struct gendisk *disk, sector_t sector,
639 				 unsigned int nr_zones, report_zones_cb cb,
640 				 void *data)
641 {
642 	struct virtio_blk *vblk = disk->private_data;
643 	struct virtio_blk_zone_report *report;
644 	unsigned int zone_sectors = vblk->zone_sectors;
645 	unsigned int nz, i;
646 	int ret, zone_idx = 0;
647 	size_t buflen;
648 
649 	if (WARN_ON_ONCE(!vblk->zone_sectors))
650 		return -EOPNOTSUPP;
651 
652 	report = virtblk_alloc_report_buffer(vblk, nr_zones,
653 					     zone_sectors, &buflen);
654 	if (!report)
655 		return -ENOMEM;
656 
657 	while (zone_idx < nr_zones && sector < get_capacity(vblk->disk)) {
658 		memset(report, 0, buflen);
659 
660 		ret = virtblk_submit_zone_report(vblk, (char *)report,
661 						 buflen, sector);
662 		if (ret) {
663 			if (ret > 0)
664 				ret = -EIO;
665 			goto out_free;
666 		}
667 		nz = min((unsigned int)le64_to_cpu(report->nr_zones), nr_zones);
668 		if (!nz)
669 			break;
670 
671 		for (i = 0; i < nz && zone_idx < nr_zones; i++) {
672 			ret = virtblk_parse_zone(vblk, &report->zones[i],
673 						 zone_idx, zone_sectors, cb, data);
674 			if (ret)
675 				goto out_free;
676 			sector = le64_to_cpu(report->zones[i].z_start) + zone_sectors;
677 			zone_idx++;
678 		}
679 	}
680 
681 	if (zone_idx > 0)
682 		ret = zone_idx;
683 	else
684 		ret = -EINVAL;
685 out_free:
686 	kvfree(report);
687 	return ret;
688 }
689 
690 static void virtblk_revalidate_zones(struct virtio_blk *vblk)
691 {
692 	u8 model;
693 
694 	if (!vblk->zone_sectors)
695 		return;
696 
697 	virtio_cread(vblk->vdev, struct virtio_blk_config,
698 		     zoned.model, &model);
699 	if (!blk_revalidate_disk_zones(vblk->disk, NULL))
700 		set_capacity_and_notify(vblk->disk, 0);
701 }
702 
703 static int virtblk_probe_zoned_device(struct virtio_device *vdev,
704 				       struct virtio_blk *vblk,
705 				       struct request_queue *q)
706 {
707 	u32 v;
708 	u8 model;
709 	int ret;
710 
711 	virtio_cread(vdev, struct virtio_blk_config,
712 		     zoned.model, &model);
713 
714 	switch (model) {
715 	case VIRTIO_BLK_Z_NONE:
716 		return 0;
717 	case VIRTIO_BLK_Z_HM:
718 		break;
719 	case VIRTIO_BLK_Z_HA:
720 		/*
721 		 * Present the host-aware device as a regular drive.
722 		 * TODO It is possible to add an option to make it appear
723 		 * in the system as a zoned drive.
724 		 */
725 		return 0;
726 	default:
727 		dev_err(&vdev->dev, "unsupported zone model %d\n", model);
728 		return -EINVAL;
729 	}
730 
731 	dev_dbg(&vdev->dev, "probing host-managed zoned device\n");
732 
733 	disk_set_zoned(vblk->disk, BLK_ZONED_HM);
734 	blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
735 
736 	virtio_cread(vdev, struct virtio_blk_config,
737 		     zoned.max_open_zones, &v);
738 	disk_set_max_open_zones(vblk->disk, le32_to_cpu(v));
739 
740 	dev_dbg(&vdev->dev, "max open zones = %u\n", le32_to_cpu(v));
741 
742 	virtio_cread(vdev, struct virtio_blk_config,
743 		     zoned.max_active_zones, &v);
744 	disk_set_max_active_zones(vblk->disk, le32_to_cpu(v));
745 	dev_dbg(&vdev->dev, "max active zones = %u\n", le32_to_cpu(v));
746 
747 	virtio_cread(vdev, struct virtio_blk_config,
748 		     zoned.write_granularity, &v);
749 	if (!v) {
750 		dev_warn(&vdev->dev, "zero write granularity reported\n");
751 		return -ENODEV;
752 	}
753 	blk_queue_physical_block_size(q, le32_to_cpu(v));
754 	blk_queue_io_min(q, le32_to_cpu(v));
755 
756 	dev_dbg(&vdev->dev, "write granularity = %u\n", le32_to_cpu(v));
757 
758 	/*
759 	 * virtio ZBD specification doesn't require zones to be a power of
760 	 * two sectors in size, but the code in this driver expects that.
761 	 */
762 	virtio_cread(vdev, struct virtio_blk_config, zoned.zone_sectors, &v);
763 	vblk->zone_sectors = le32_to_cpu(v);
764 	if (vblk->zone_sectors == 0 || !is_power_of_2(vblk->zone_sectors)) {
765 		dev_err(&vdev->dev,
766 			"zoned device with non power of two zone size %u\n",
767 			vblk->zone_sectors);
768 		return -ENODEV;
769 	}
770 	dev_dbg(&vdev->dev, "zone sectors = %u\n", vblk->zone_sectors);
771 
772 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
773 		dev_warn(&vblk->vdev->dev,
774 			 "ignoring negotiated F_DISCARD for zoned device\n");
775 		blk_queue_max_discard_sectors(q, 0);
776 	}
777 
778 	ret = blk_revalidate_disk_zones(vblk->disk, NULL);
779 	if (!ret) {
780 		virtio_cread(vdev, struct virtio_blk_config,
781 			     zoned.max_append_sectors, &v);
782 		if (!v) {
783 			dev_warn(&vdev->dev, "zero max_append_sectors reported\n");
784 			return -ENODEV;
785 		}
786 		blk_queue_max_zone_append_sectors(q, le32_to_cpu(v));
787 		dev_dbg(&vdev->dev, "max append sectors = %u\n", le32_to_cpu(v));
788 	}
789 
790 	return ret;
791 }
792 
793 static inline bool virtblk_has_zoned_feature(struct virtio_device *vdev)
794 {
795 	return virtio_has_feature(vdev, VIRTIO_BLK_F_ZONED);
796 }
797 #else
798 
799 /*
800  * Zoned block device support is not configured in this kernel.
801  * We only need to define a few symbols to avoid compilation errors.
802  */
803 #define virtblk_report_zones       NULL
804 static inline void virtblk_revalidate_zones(struct virtio_blk *vblk)
805 {
806 }
807 static inline int virtblk_probe_zoned_device(struct virtio_device *vdev,
808 			struct virtio_blk *vblk, struct request_queue *q)
809 {
810 	return -EOPNOTSUPP;
811 }
812 
813 static inline bool virtblk_has_zoned_feature(struct virtio_device *vdev)
814 {
815 	return false;
816 }
817 #endif /* CONFIG_BLK_DEV_ZONED */
818 
819 /* return id (s/n) string for *disk to *id_str
820  */
821 static int virtblk_get_id(struct gendisk *disk, char *id_str)
822 {
823 	struct virtio_blk *vblk = disk->private_data;
824 	struct request_queue *q = vblk->disk->queue;
825 	struct request *req;
826 	struct virtblk_req *vbr;
827 	int err;
828 
829 	req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
830 	if (IS_ERR(req))
831 		return PTR_ERR(req);
832 
833 	vbr = blk_mq_rq_to_pdu(req);
834 	vbr->in_hdr_len = sizeof(vbr->status);
835 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
836 	vbr->out_hdr.sector = 0;
837 
838 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
839 	if (err)
840 		goto out;
841 
842 	blk_execute_rq(req, false);
843 	err = blk_status_to_errno(virtblk_result(vbr->status));
844 out:
845 	blk_mq_free_request(req);
846 	return err;
847 }
848 
849 /* We provide getgeo only to please some old bootloader/partitioning tools */
850 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
851 {
852 	struct virtio_blk *vblk = bd->bd_disk->private_data;
853 	int ret = 0;
854 
855 	mutex_lock(&vblk->vdev_mutex);
856 
857 	if (!vblk->vdev) {
858 		ret = -ENXIO;
859 		goto out;
860 	}
861 
862 	/* see if the host passed in geometry config */
863 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
864 		virtio_cread(vblk->vdev, struct virtio_blk_config,
865 			     geometry.cylinders, &geo->cylinders);
866 		virtio_cread(vblk->vdev, struct virtio_blk_config,
867 			     geometry.heads, &geo->heads);
868 		virtio_cread(vblk->vdev, struct virtio_blk_config,
869 			     geometry.sectors, &geo->sectors);
870 	} else {
871 		/* some standard values, similar to sd */
872 		geo->heads = 1 << 6;
873 		geo->sectors = 1 << 5;
874 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
875 	}
876 out:
877 	mutex_unlock(&vblk->vdev_mutex);
878 	return ret;
879 }
880 
881 static void virtblk_free_disk(struct gendisk *disk)
882 {
883 	struct virtio_blk *vblk = disk->private_data;
884 
885 	ida_free(&vd_index_ida, vblk->index);
886 	mutex_destroy(&vblk->vdev_mutex);
887 	kfree(vblk);
888 }
889 
890 static const struct block_device_operations virtblk_fops = {
891 	.owner  	= THIS_MODULE,
892 	.getgeo		= virtblk_getgeo,
893 	.free_disk	= virtblk_free_disk,
894 	.report_zones	= virtblk_report_zones,
895 };
896 
897 static int index_to_minor(int index)
898 {
899 	return index << PART_BITS;
900 }
901 
902 static int minor_to_index(int minor)
903 {
904 	return minor >> PART_BITS;
905 }
906 
907 static ssize_t serial_show(struct device *dev,
908 			   struct device_attribute *attr, char *buf)
909 {
910 	struct gendisk *disk = dev_to_disk(dev);
911 	int err;
912 
913 	/* sysfs gives us a PAGE_SIZE buffer */
914 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
915 
916 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
917 	err = virtblk_get_id(disk, buf);
918 	if (!err)
919 		return strlen(buf);
920 
921 	if (err == -EIO) /* Unsupported? Make it empty. */
922 		return 0;
923 
924 	return err;
925 }
926 
927 static DEVICE_ATTR_RO(serial);
928 
929 /* The queue's logical block size must be set before calling this */
930 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
931 {
932 	struct virtio_device *vdev = vblk->vdev;
933 	struct request_queue *q = vblk->disk->queue;
934 	char cap_str_2[10], cap_str_10[10];
935 	unsigned long long nblocks;
936 	u64 capacity;
937 
938 	/* Host must always specify the capacity. */
939 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
940 
941 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
942 
943 	string_get_size(nblocks, queue_logical_block_size(q),
944 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
945 	string_get_size(nblocks, queue_logical_block_size(q),
946 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
947 
948 	dev_notice(&vdev->dev,
949 		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
950 		   vblk->disk->disk_name,
951 		   resize ? "new size: " : "",
952 		   nblocks,
953 		   queue_logical_block_size(q),
954 		   cap_str_10,
955 		   cap_str_2);
956 
957 	set_capacity_and_notify(vblk->disk, capacity);
958 }
959 
960 static void virtblk_config_changed_work(struct work_struct *work)
961 {
962 	struct virtio_blk *vblk =
963 		container_of(work, struct virtio_blk, config_work);
964 
965 	virtblk_revalidate_zones(vblk);
966 	virtblk_update_capacity(vblk, true);
967 }
968 
969 static void virtblk_config_changed(struct virtio_device *vdev)
970 {
971 	struct virtio_blk *vblk = vdev->priv;
972 
973 	queue_work(virtblk_wq, &vblk->config_work);
974 }
975 
976 static int init_vq(struct virtio_blk *vblk)
977 {
978 	int err;
979 	int i;
980 	vq_callback_t **callbacks;
981 	const char **names;
982 	struct virtqueue **vqs;
983 	unsigned short num_vqs;
984 	unsigned int num_poll_vqs;
985 	struct virtio_device *vdev = vblk->vdev;
986 	struct irq_affinity desc = { 0, };
987 
988 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
989 				   struct virtio_blk_config, num_queues,
990 				   &num_vqs);
991 	if (err)
992 		num_vqs = 1;
993 
994 	if (!err && !num_vqs) {
995 		dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
996 		return -EINVAL;
997 	}
998 
999 	num_vqs = min_t(unsigned int,
1000 			min_not_zero(num_request_queues, nr_cpu_ids),
1001 			num_vqs);
1002 
1003 	num_poll_vqs = min_t(unsigned int, poll_queues, num_vqs - 1);
1004 
1005 	vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
1006 	vblk->io_queues[HCTX_TYPE_READ] = 0;
1007 	vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
1008 
1009 	dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n",
1010 				vblk->io_queues[HCTX_TYPE_DEFAULT],
1011 				vblk->io_queues[HCTX_TYPE_READ],
1012 				vblk->io_queues[HCTX_TYPE_POLL]);
1013 
1014 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
1015 	if (!vblk->vqs)
1016 		return -ENOMEM;
1017 
1018 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
1019 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
1020 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
1021 	if (!names || !callbacks || !vqs) {
1022 		err = -ENOMEM;
1023 		goto out;
1024 	}
1025 
1026 	for (i = 0; i < num_vqs - num_poll_vqs; i++) {
1027 		callbacks[i] = virtblk_done;
1028 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
1029 		names[i] = vblk->vqs[i].name;
1030 	}
1031 
1032 	for (; i < num_vqs; i++) {
1033 		callbacks[i] = NULL;
1034 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%d", i);
1035 		names[i] = vblk->vqs[i].name;
1036 	}
1037 
1038 	/* Discover virtqueues and write information to configuration.  */
1039 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
1040 	if (err)
1041 		goto out;
1042 
1043 	for (i = 0; i < num_vqs; i++) {
1044 		spin_lock_init(&vblk->vqs[i].lock);
1045 		vblk->vqs[i].vq = vqs[i];
1046 	}
1047 	vblk->num_vqs = num_vqs;
1048 
1049 out:
1050 	kfree(vqs);
1051 	kfree(callbacks);
1052 	kfree(names);
1053 	if (err)
1054 		kfree(vblk->vqs);
1055 	return err;
1056 }
1057 
1058 /*
1059  * Legacy naming scheme used for virtio devices.  We are stuck with it for
1060  * virtio blk but don't ever use it for any new driver.
1061  */
1062 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
1063 {
1064 	const int base = 'z' - 'a' + 1;
1065 	char *begin = buf + strlen(prefix);
1066 	char *end = buf + buflen;
1067 	char *p;
1068 	int unit;
1069 
1070 	p = end - 1;
1071 	*p = '\0';
1072 	unit = base;
1073 	do {
1074 		if (p == begin)
1075 			return -EINVAL;
1076 		*--p = 'a' + (index % unit);
1077 		index = (index / unit) - 1;
1078 	} while (index >= 0);
1079 
1080 	memmove(begin, p, end - p);
1081 	memcpy(buf, prefix, strlen(prefix));
1082 
1083 	return 0;
1084 }
1085 
1086 static int virtblk_get_cache_mode(struct virtio_device *vdev)
1087 {
1088 	u8 writeback;
1089 	int err;
1090 
1091 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
1092 				   struct virtio_blk_config, wce,
1093 				   &writeback);
1094 
1095 	/*
1096 	 * If WCE is not configurable and flush is not available,
1097 	 * assume no writeback cache is in use.
1098 	 */
1099 	if (err)
1100 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
1101 
1102 	return writeback;
1103 }
1104 
1105 static void virtblk_update_cache_mode(struct virtio_device *vdev)
1106 {
1107 	u8 writeback = virtblk_get_cache_mode(vdev);
1108 	struct virtio_blk *vblk = vdev->priv;
1109 
1110 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
1111 }
1112 
1113 static const char *const virtblk_cache_types[] = {
1114 	"write through", "write back"
1115 };
1116 
1117 static ssize_t
1118 cache_type_store(struct device *dev, struct device_attribute *attr,
1119 		 const char *buf, size_t count)
1120 {
1121 	struct gendisk *disk = dev_to_disk(dev);
1122 	struct virtio_blk *vblk = disk->private_data;
1123 	struct virtio_device *vdev = vblk->vdev;
1124 	int i;
1125 
1126 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
1127 	i = sysfs_match_string(virtblk_cache_types, buf);
1128 	if (i < 0)
1129 		return i;
1130 
1131 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
1132 	virtblk_update_cache_mode(vdev);
1133 	return count;
1134 }
1135 
1136 static ssize_t
1137 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
1138 {
1139 	struct gendisk *disk = dev_to_disk(dev);
1140 	struct virtio_blk *vblk = disk->private_data;
1141 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
1142 
1143 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
1144 	return sysfs_emit(buf, "%s\n", virtblk_cache_types[writeback]);
1145 }
1146 
1147 static DEVICE_ATTR_RW(cache_type);
1148 
1149 static struct attribute *virtblk_attrs[] = {
1150 	&dev_attr_serial.attr,
1151 	&dev_attr_cache_type.attr,
1152 	NULL,
1153 };
1154 
1155 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
1156 		struct attribute *a, int n)
1157 {
1158 	struct device *dev = kobj_to_dev(kobj);
1159 	struct gendisk *disk = dev_to_disk(dev);
1160 	struct virtio_blk *vblk = disk->private_data;
1161 	struct virtio_device *vdev = vblk->vdev;
1162 
1163 	if (a == &dev_attr_cache_type.attr &&
1164 	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
1165 		return S_IRUGO;
1166 
1167 	return a->mode;
1168 }
1169 
1170 static const struct attribute_group virtblk_attr_group = {
1171 	.attrs = virtblk_attrs,
1172 	.is_visible = virtblk_attrs_are_visible,
1173 };
1174 
1175 static const struct attribute_group *virtblk_attr_groups[] = {
1176 	&virtblk_attr_group,
1177 	NULL,
1178 };
1179 
1180 static void virtblk_map_queues(struct blk_mq_tag_set *set)
1181 {
1182 	struct virtio_blk *vblk = set->driver_data;
1183 	int i, qoff;
1184 
1185 	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1186 		struct blk_mq_queue_map *map = &set->map[i];
1187 
1188 		map->nr_queues = vblk->io_queues[i];
1189 		map->queue_offset = qoff;
1190 		qoff += map->nr_queues;
1191 
1192 		if (map->nr_queues == 0)
1193 			continue;
1194 
1195 		/*
1196 		 * Regular queues have interrupts and hence CPU affinity is
1197 		 * defined by the core virtio code, but polling queues have
1198 		 * no interrupts so we let the block layer assign CPU affinity.
1199 		 */
1200 		if (i == HCTX_TYPE_POLL)
1201 			blk_mq_map_queues(&set->map[i]);
1202 		else
1203 			blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
1204 	}
1205 }
1206 
1207 static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1208 {
1209 	struct virtio_blk *vblk = hctx->queue->queuedata;
1210 	struct virtio_blk_vq *vq = get_virtio_blk_vq(hctx);
1211 	unsigned long flags;
1212 	int found = 0;
1213 
1214 	spin_lock_irqsave(&vq->lock, flags);
1215 	found = virtblk_handle_req(vq, iob);
1216 
1217 	if (found)
1218 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
1219 
1220 	spin_unlock_irqrestore(&vq->lock, flags);
1221 
1222 	return found;
1223 }
1224 
1225 static const struct blk_mq_ops virtio_mq_ops = {
1226 	.queue_rq	= virtio_queue_rq,
1227 	.queue_rqs	= virtio_queue_rqs,
1228 	.commit_rqs	= virtio_commit_rqs,
1229 	.complete	= virtblk_request_done,
1230 	.map_queues	= virtblk_map_queues,
1231 	.poll		= virtblk_poll,
1232 };
1233 
1234 static unsigned int virtblk_queue_depth;
1235 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
1236 
1237 static int virtblk_probe(struct virtio_device *vdev)
1238 {
1239 	struct virtio_blk *vblk;
1240 	struct request_queue *q;
1241 	int err, index;
1242 
1243 	u32 v, blk_size, max_size, sg_elems, opt_io_size;
1244 	u32 max_discard_segs = 0;
1245 	u32 discard_granularity = 0;
1246 	u16 min_io_size;
1247 	u8 physical_block_exp, alignment_offset;
1248 	unsigned int queue_depth;
1249 
1250 	if (!vdev->config->get) {
1251 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
1252 			__func__);
1253 		return -EINVAL;
1254 	}
1255 
1256 	err = ida_alloc_range(&vd_index_ida, 0,
1257 			      minor_to_index(1 << MINORBITS) - 1, GFP_KERNEL);
1258 	if (err < 0)
1259 		goto out;
1260 	index = err;
1261 
1262 	/* We need to know how many segments before we allocate. */
1263 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
1264 				   struct virtio_blk_config, seg_max,
1265 				   &sg_elems);
1266 
1267 	/* We need at least one SG element, whatever they say. */
1268 	if (err || !sg_elems)
1269 		sg_elems = 1;
1270 
1271 	/* Prevent integer overflows and honor max vq size */
1272 	sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
1273 
1274 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
1275 	if (!vblk) {
1276 		err = -ENOMEM;
1277 		goto out_free_index;
1278 	}
1279 
1280 	mutex_init(&vblk->vdev_mutex);
1281 
1282 	vblk->vdev = vdev;
1283 
1284 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
1285 
1286 	err = init_vq(vblk);
1287 	if (err)
1288 		goto out_free_vblk;
1289 
1290 	/* Default queue sizing is to fill the ring. */
1291 	if (!virtblk_queue_depth) {
1292 		queue_depth = vblk->vqs[0].vq->num_free;
1293 		/* ... but without indirect descs, we use 2 descs per req */
1294 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
1295 			queue_depth /= 2;
1296 	} else {
1297 		queue_depth = virtblk_queue_depth;
1298 	}
1299 
1300 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
1301 	vblk->tag_set.ops = &virtio_mq_ops;
1302 	vblk->tag_set.queue_depth = queue_depth;
1303 	vblk->tag_set.numa_node = NUMA_NO_NODE;
1304 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1305 	vblk->tag_set.cmd_size =
1306 		sizeof(struct virtblk_req) +
1307 		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
1308 	vblk->tag_set.driver_data = vblk;
1309 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
1310 	vblk->tag_set.nr_maps = 1;
1311 	if (vblk->io_queues[HCTX_TYPE_POLL])
1312 		vblk->tag_set.nr_maps = 3;
1313 
1314 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
1315 	if (err)
1316 		goto out_free_vq;
1317 
1318 	vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
1319 	if (IS_ERR(vblk->disk)) {
1320 		err = PTR_ERR(vblk->disk);
1321 		goto out_free_tags;
1322 	}
1323 	q = vblk->disk->queue;
1324 
1325 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
1326 
1327 	vblk->disk->major = major;
1328 	vblk->disk->first_minor = index_to_minor(index);
1329 	vblk->disk->minors = 1 << PART_BITS;
1330 	vblk->disk->private_data = vblk;
1331 	vblk->disk->fops = &virtblk_fops;
1332 	vblk->index = index;
1333 
1334 	/* configure queue flush support */
1335 	virtblk_update_cache_mode(vdev);
1336 
1337 	/* If disk is read-only in the host, the guest should obey */
1338 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
1339 		set_disk_ro(vblk->disk, 1);
1340 
1341 	/* We can handle whatever the host told us to handle. */
1342 	blk_queue_max_segments(q, sg_elems);
1343 
1344 	/* No real sector limit. */
1345 	blk_queue_max_hw_sectors(q, UINT_MAX);
1346 
1347 	max_size = virtio_max_dma_size(vdev);
1348 
1349 	/* Host can optionally specify maximum segment size and number of
1350 	 * segments. */
1351 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
1352 				   struct virtio_blk_config, size_max, &v);
1353 	if (!err)
1354 		max_size = min(max_size, v);
1355 
1356 	blk_queue_max_segment_size(q, max_size);
1357 
1358 	/* Host can optionally specify the block size of the device */
1359 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
1360 				   struct virtio_blk_config, blk_size,
1361 				   &blk_size);
1362 	if (!err) {
1363 		err = blk_validate_block_size(blk_size);
1364 		if (err) {
1365 			dev_err(&vdev->dev,
1366 				"virtio_blk: invalid block size: 0x%x\n",
1367 				blk_size);
1368 			goto out_cleanup_disk;
1369 		}
1370 
1371 		blk_queue_logical_block_size(q, blk_size);
1372 	} else
1373 		blk_size = queue_logical_block_size(q);
1374 
1375 	/* Use topology information if available */
1376 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1377 				   struct virtio_blk_config, physical_block_exp,
1378 				   &physical_block_exp);
1379 	if (!err && physical_block_exp)
1380 		blk_queue_physical_block_size(q,
1381 				blk_size * (1 << physical_block_exp));
1382 
1383 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1384 				   struct virtio_blk_config, alignment_offset,
1385 				   &alignment_offset);
1386 	if (!err && alignment_offset)
1387 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
1388 
1389 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1390 				   struct virtio_blk_config, min_io_size,
1391 				   &min_io_size);
1392 	if (!err && min_io_size)
1393 		blk_queue_io_min(q, blk_size * min_io_size);
1394 
1395 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1396 				   struct virtio_blk_config, opt_io_size,
1397 				   &opt_io_size);
1398 	if (!err && opt_io_size)
1399 		blk_queue_io_opt(q, blk_size * opt_io_size);
1400 
1401 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
1402 		virtio_cread(vdev, struct virtio_blk_config,
1403 			     discard_sector_alignment, &discard_granularity);
1404 
1405 		virtio_cread(vdev, struct virtio_blk_config,
1406 			     max_discard_sectors, &v);
1407 		blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
1408 
1409 		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
1410 			     &max_discard_segs);
1411 	}
1412 
1413 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
1414 		virtio_cread(vdev, struct virtio_blk_config,
1415 			     max_write_zeroes_sectors, &v);
1416 		blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
1417 	}
1418 
1419 	/* The discard and secure erase limits are combined since the Linux
1420 	 * block layer uses the same limit for both commands.
1421 	 *
1422 	 * If both VIRTIO_BLK_F_SECURE_ERASE and VIRTIO_BLK_F_DISCARD features
1423 	 * are negotiated, we will use the minimum between the limits.
1424 	 *
1425 	 * discard sector alignment is set to the minimum between discard_sector_alignment
1426 	 * and secure_erase_sector_alignment.
1427 	 *
1428 	 * max discard sectors is set to the minimum between max_discard_seg and
1429 	 * max_secure_erase_seg.
1430 	 */
1431 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1432 
1433 		virtio_cread(vdev, struct virtio_blk_config,
1434 			     secure_erase_sector_alignment, &v);
1435 
1436 		/* secure_erase_sector_alignment should not be zero, the device should set a
1437 		 * valid number of sectors.
1438 		 */
1439 		if (!v) {
1440 			dev_err(&vdev->dev,
1441 				"virtio_blk: secure_erase_sector_alignment can't be 0\n");
1442 			err = -EINVAL;
1443 			goto out_cleanup_disk;
1444 		}
1445 
1446 		discard_granularity = min_not_zero(discard_granularity, v);
1447 
1448 		virtio_cread(vdev, struct virtio_blk_config,
1449 			     max_secure_erase_sectors, &v);
1450 
1451 		/* max_secure_erase_sectors should not be zero, the device should set a
1452 		 * valid number of sectors.
1453 		 */
1454 		if (!v) {
1455 			dev_err(&vdev->dev,
1456 				"virtio_blk: max_secure_erase_sectors can't be 0\n");
1457 			err = -EINVAL;
1458 			goto out_cleanup_disk;
1459 		}
1460 
1461 		blk_queue_max_secure_erase_sectors(q, v);
1462 
1463 		virtio_cread(vdev, struct virtio_blk_config,
1464 			     max_secure_erase_seg, &v);
1465 
1466 		/* max_secure_erase_seg should not be zero, the device should set a
1467 		 * valid number of segments
1468 		 */
1469 		if (!v) {
1470 			dev_err(&vdev->dev,
1471 				"virtio_blk: max_secure_erase_seg can't be 0\n");
1472 			err = -EINVAL;
1473 			goto out_cleanup_disk;
1474 		}
1475 
1476 		max_discard_segs = min_not_zero(max_discard_segs, v);
1477 	}
1478 
1479 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD) ||
1480 	    virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1481 		/* max_discard_seg and discard_granularity will be 0 only
1482 		 * if max_discard_seg and discard_sector_alignment fields in the virtio
1483 		 * config are 0 and VIRTIO_BLK_F_SECURE_ERASE feature is not negotiated.
1484 		 * In this case, we use default values.
1485 		 */
1486 		if (!max_discard_segs)
1487 			max_discard_segs = sg_elems;
1488 
1489 		blk_queue_max_discard_segments(q,
1490 					       min(max_discard_segs, MAX_DISCARD_SEGMENTS));
1491 
1492 		if (discard_granularity)
1493 			q->limits.discard_granularity = discard_granularity << SECTOR_SHIFT;
1494 		else
1495 			q->limits.discard_granularity = blk_size;
1496 	}
1497 
1498 	virtblk_update_capacity(vblk, false);
1499 	virtio_device_ready(vdev);
1500 
1501 	if (virtblk_has_zoned_feature(vdev)) {
1502 		err = virtblk_probe_zoned_device(vdev, vblk, q);
1503 		if (err)
1504 			goto out_cleanup_disk;
1505 	}
1506 
1507 	dev_info(&vdev->dev, "blk config size: %zu\n",
1508 		sizeof(struct virtio_blk_config));
1509 
1510 	err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
1511 	if (err)
1512 		goto out_cleanup_disk;
1513 
1514 	return 0;
1515 
1516 out_cleanup_disk:
1517 	put_disk(vblk->disk);
1518 out_free_tags:
1519 	blk_mq_free_tag_set(&vblk->tag_set);
1520 out_free_vq:
1521 	vdev->config->del_vqs(vdev);
1522 	kfree(vblk->vqs);
1523 out_free_vblk:
1524 	kfree(vblk);
1525 out_free_index:
1526 	ida_free(&vd_index_ida, index);
1527 out:
1528 	return err;
1529 }
1530 
1531 static void virtblk_remove(struct virtio_device *vdev)
1532 {
1533 	struct virtio_blk *vblk = vdev->priv;
1534 
1535 	/* Make sure no work handler is accessing the device. */
1536 	flush_work(&vblk->config_work);
1537 
1538 	del_gendisk(vblk->disk);
1539 	blk_mq_free_tag_set(&vblk->tag_set);
1540 
1541 	mutex_lock(&vblk->vdev_mutex);
1542 
1543 	/* Stop all the virtqueues. */
1544 	virtio_reset_device(vdev);
1545 
1546 	/* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
1547 	vblk->vdev = NULL;
1548 
1549 	vdev->config->del_vqs(vdev);
1550 	kfree(vblk->vqs);
1551 
1552 	mutex_unlock(&vblk->vdev_mutex);
1553 
1554 	put_disk(vblk->disk);
1555 }
1556 
1557 #ifdef CONFIG_PM_SLEEP
1558 static int virtblk_freeze(struct virtio_device *vdev)
1559 {
1560 	struct virtio_blk *vblk = vdev->priv;
1561 
1562 	/* Ensure we don't receive any more interrupts */
1563 	virtio_reset_device(vdev);
1564 
1565 	/* Make sure no work handler is accessing the device. */
1566 	flush_work(&vblk->config_work);
1567 
1568 	blk_mq_quiesce_queue(vblk->disk->queue);
1569 
1570 	vdev->config->del_vqs(vdev);
1571 	kfree(vblk->vqs);
1572 
1573 	return 0;
1574 }
1575 
1576 static int virtblk_restore(struct virtio_device *vdev)
1577 {
1578 	struct virtio_blk *vblk = vdev->priv;
1579 	int ret;
1580 
1581 	ret = init_vq(vdev->priv);
1582 	if (ret)
1583 		return ret;
1584 
1585 	virtio_device_ready(vdev);
1586 
1587 	blk_mq_unquiesce_queue(vblk->disk->queue);
1588 	return 0;
1589 }
1590 #endif
1591 
1592 static const struct virtio_device_id id_table[] = {
1593 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1594 	{ 0 },
1595 };
1596 
1597 static unsigned int features_legacy[] = {
1598 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1599 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1600 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1601 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1602 	VIRTIO_BLK_F_SECURE_ERASE,
1603 }
1604 ;
1605 static unsigned int features[] = {
1606 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1607 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1608 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1609 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1610 	VIRTIO_BLK_F_SECURE_ERASE,
1611 #ifdef CONFIG_BLK_DEV_ZONED
1612 	VIRTIO_BLK_F_ZONED,
1613 #endif /* CONFIG_BLK_DEV_ZONED */
1614 };
1615 
1616 static struct virtio_driver virtio_blk = {
1617 	.feature_table			= features,
1618 	.feature_table_size		= ARRAY_SIZE(features),
1619 	.feature_table_legacy		= features_legacy,
1620 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
1621 	.driver.name			= KBUILD_MODNAME,
1622 	.driver.owner			= THIS_MODULE,
1623 	.id_table			= id_table,
1624 	.probe				= virtblk_probe,
1625 	.remove				= virtblk_remove,
1626 	.config_changed			= virtblk_config_changed,
1627 #ifdef CONFIG_PM_SLEEP
1628 	.freeze				= virtblk_freeze,
1629 	.restore			= virtblk_restore,
1630 #endif
1631 };
1632 
1633 static int __init virtio_blk_init(void)
1634 {
1635 	int error;
1636 
1637 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1638 	if (!virtblk_wq)
1639 		return -ENOMEM;
1640 
1641 	major = register_blkdev(0, "virtblk");
1642 	if (major < 0) {
1643 		error = major;
1644 		goto out_destroy_workqueue;
1645 	}
1646 
1647 	error = register_virtio_driver(&virtio_blk);
1648 	if (error)
1649 		goto out_unregister_blkdev;
1650 	return 0;
1651 
1652 out_unregister_blkdev:
1653 	unregister_blkdev(major, "virtblk");
1654 out_destroy_workqueue:
1655 	destroy_workqueue(virtblk_wq);
1656 	return error;
1657 }
1658 
1659 static void __exit virtio_blk_fini(void)
1660 {
1661 	unregister_virtio_driver(&virtio_blk);
1662 	unregister_blkdev(major, "virtblk");
1663 	destroy_workqueue(virtblk_wq);
1664 }
1665 module_init(virtio_blk_init);
1666 module_exit(virtio_blk_fini);
1667 
1668 MODULE_DEVICE_TABLE(virtio, id_table);
1669 MODULE_DESCRIPTION("Virtio block driver");
1670 MODULE_LICENSE("GPL");
1671