xref: /openbmc/linux/drivers/block/virtio_blk.c (revision adb19164)
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 always
100 		 * be the last byte.
101 		 */
102 		struct {
103 			__virtio64 sector;
104 			u8 status;
105 		} zone_append;
106 	} in_hdr;
107 
108 	size_t in_hdr_len;
109 
110 	struct sg_table sg_table;
111 	struct scatterlist sg[];
112 };
113 
114 static inline blk_status_t virtblk_result(u8 status)
115 {
116 	switch (status) {
117 	case VIRTIO_BLK_S_OK:
118 		return BLK_STS_OK;
119 	case VIRTIO_BLK_S_UNSUPP:
120 		return BLK_STS_NOTSUPP;
121 	case VIRTIO_BLK_S_ZONE_OPEN_RESOURCE:
122 		return BLK_STS_ZONE_OPEN_RESOURCE;
123 	case VIRTIO_BLK_S_ZONE_ACTIVE_RESOURCE:
124 		return BLK_STS_ZONE_ACTIVE_RESOURCE;
125 	case VIRTIO_BLK_S_IOERR:
126 	case VIRTIO_BLK_S_ZONE_UNALIGNED_WP:
127 	default:
128 		return BLK_STS_IOERR;
129 	}
130 }
131 
132 static inline struct virtio_blk_vq *get_virtio_blk_vq(struct blk_mq_hw_ctx *hctx)
133 {
134 	struct virtio_blk *vblk = hctx->queue->queuedata;
135 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
136 
137 	return vq;
138 }
139 
140 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr)
141 {
142 	struct scatterlist out_hdr, in_hdr, *sgs[3];
143 	unsigned int num_out = 0, num_in = 0;
144 
145 	sg_init_one(&out_hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
146 	sgs[num_out++] = &out_hdr;
147 
148 	if (vbr->sg_table.nents) {
149 		if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
150 			sgs[num_out++] = vbr->sg_table.sgl;
151 		else
152 			sgs[num_out + num_in++] = vbr->sg_table.sgl;
153 	}
154 
155 	sg_init_one(&in_hdr, &vbr->in_hdr.status, vbr->in_hdr_len);
156 	sgs[num_out + num_in++] = &in_hdr;
157 
158 	return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
159 }
160 
161 static int virtblk_setup_discard_write_zeroes_erase(struct request *req, bool unmap)
162 {
163 	unsigned short segments = blk_rq_nr_discard_segments(req);
164 	unsigned short n = 0;
165 	struct virtio_blk_discard_write_zeroes *range;
166 	struct bio *bio;
167 	u32 flags = 0;
168 
169 	if (unmap)
170 		flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
171 
172 	range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
173 	if (!range)
174 		return -ENOMEM;
175 
176 	/*
177 	 * Single max discard segment means multi-range discard isn't
178 	 * supported, and block layer only runs contiguity merge like
179 	 * normal RW request. So we can't reply on bio for retrieving
180 	 * each range info.
181 	 */
182 	if (queue_max_discard_segments(req->q) == 1) {
183 		range[0].flags = cpu_to_le32(flags);
184 		range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
185 		range[0].sector = cpu_to_le64(blk_rq_pos(req));
186 		n = 1;
187 	} else {
188 		__rq_for_each_bio(bio, req) {
189 			u64 sector = bio->bi_iter.bi_sector;
190 			u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
191 
192 			range[n].flags = cpu_to_le32(flags);
193 			range[n].num_sectors = cpu_to_le32(num_sectors);
194 			range[n].sector = cpu_to_le64(sector);
195 			n++;
196 		}
197 	}
198 
199 	WARN_ON_ONCE(n != segments);
200 
201 	bvec_set_virt(&req->special_vec, range, sizeof(*range) * segments);
202 	req->rq_flags |= RQF_SPECIAL_PAYLOAD;
203 
204 	return 0;
205 }
206 
207 static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
208 {
209 	if (blk_rq_nr_phys_segments(req))
210 		sg_free_table_chained(&vbr->sg_table,
211 				      VIRTIO_BLK_INLINE_SG_CNT);
212 }
213 
214 static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
215 		struct virtblk_req *vbr)
216 {
217 	int err;
218 
219 	if (!blk_rq_nr_phys_segments(req))
220 		return 0;
221 
222 	vbr->sg_table.sgl = vbr->sg;
223 	err = sg_alloc_table_chained(&vbr->sg_table,
224 				     blk_rq_nr_phys_segments(req),
225 				     vbr->sg_table.sgl,
226 				     VIRTIO_BLK_INLINE_SG_CNT);
227 	if (unlikely(err))
228 		return -ENOMEM;
229 
230 	return blk_rq_map_sg(hctx->queue, req, vbr->sg_table.sgl);
231 }
232 
233 static void virtblk_cleanup_cmd(struct request *req)
234 {
235 	if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
236 		kfree(bvec_virt(&req->special_vec));
237 }
238 
239 static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
240 				      struct request *req,
241 				      struct virtblk_req *vbr)
242 {
243 	size_t in_hdr_len = sizeof(vbr->in_hdr.status);
244 	bool unmap = false;
245 	u32 type;
246 	u64 sector = 0;
247 
248 	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && op_is_zone_mgmt(req_op(req)))
249 		return BLK_STS_NOTSUPP;
250 
251 	/* Set fields for all request types */
252 	vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
253 
254 	switch (req_op(req)) {
255 	case REQ_OP_READ:
256 		type = VIRTIO_BLK_T_IN;
257 		sector = blk_rq_pos(req);
258 		break;
259 	case REQ_OP_WRITE:
260 		type = VIRTIO_BLK_T_OUT;
261 		sector = blk_rq_pos(req);
262 		break;
263 	case REQ_OP_FLUSH:
264 		type = VIRTIO_BLK_T_FLUSH;
265 		break;
266 	case REQ_OP_DISCARD:
267 		type = VIRTIO_BLK_T_DISCARD;
268 		break;
269 	case REQ_OP_WRITE_ZEROES:
270 		type = VIRTIO_BLK_T_WRITE_ZEROES;
271 		unmap = !(req->cmd_flags & REQ_NOUNMAP);
272 		break;
273 	case REQ_OP_SECURE_ERASE:
274 		type = VIRTIO_BLK_T_SECURE_ERASE;
275 		break;
276 	case REQ_OP_ZONE_OPEN:
277 		type = VIRTIO_BLK_T_ZONE_OPEN;
278 		sector = blk_rq_pos(req);
279 		break;
280 	case REQ_OP_ZONE_CLOSE:
281 		type = VIRTIO_BLK_T_ZONE_CLOSE;
282 		sector = blk_rq_pos(req);
283 		break;
284 	case REQ_OP_ZONE_FINISH:
285 		type = VIRTIO_BLK_T_ZONE_FINISH;
286 		sector = blk_rq_pos(req);
287 		break;
288 	case REQ_OP_ZONE_APPEND:
289 		type = VIRTIO_BLK_T_ZONE_APPEND;
290 		sector = blk_rq_pos(req);
291 		in_hdr_len = sizeof(vbr->in_hdr.zone_append);
292 		break;
293 	case REQ_OP_ZONE_RESET:
294 		type = VIRTIO_BLK_T_ZONE_RESET;
295 		sector = blk_rq_pos(req);
296 		break;
297 	case REQ_OP_ZONE_RESET_ALL:
298 		type = VIRTIO_BLK_T_ZONE_RESET_ALL;
299 		break;
300 	case REQ_OP_DRV_IN:
301 		/*
302 		 * Out header has already been prepared by the caller (virtblk_get_id()
303 		 * or virtblk_submit_zone_report()), nothing to do here.
304 		 */
305 		return 0;
306 	default:
307 		WARN_ON_ONCE(1);
308 		return BLK_STS_IOERR;
309 	}
310 
311 	/* Set fields for non-REQ_OP_DRV_IN request types */
312 	vbr->in_hdr_len = in_hdr_len;
313 	vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
314 	vbr->out_hdr.sector = cpu_to_virtio64(vdev, sector);
315 
316 	if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES ||
317 	    type == VIRTIO_BLK_T_SECURE_ERASE) {
318 		if (virtblk_setup_discard_write_zeroes_erase(req, unmap))
319 			return BLK_STS_RESOURCE;
320 	}
321 
322 	return 0;
323 }
324 
325 /*
326  * The status byte is always the last byte of the virtblk request
327  * in-header. This helper fetches its value for all in-header formats
328  * that are currently defined.
329  */
330 static inline u8 virtblk_vbr_status(struct virtblk_req *vbr)
331 {
332 	return *((u8 *)&vbr->in_hdr + vbr->in_hdr_len - 1);
333 }
334 
335 static inline void virtblk_request_done(struct request *req)
336 {
337 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
338 	blk_status_t status = virtblk_result(virtblk_vbr_status(vbr));
339 	struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
340 
341 	virtblk_unmap_data(req, vbr);
342 	virtblk_cleanup_cmd(req);
343 
344 	if (req_op(req) == REQ_OP_ZONE_APPEND)
345 		req->__sector = virtio64_to_cpu(vblk->vdev,
346 						vbr->in_hdr.zone_append.sector);
347 
348 	blk_mq_end_request(req, status);
349 }
350 
351 static void virtblk_done(struct virtqueue *vq)
352 {
353 	struct virtio_blk *vblk = vq->vdev->priv;
354 	bool req_done = false;
355 	int qid = vq->index;
356 	struct virtblk_req *vbr;
357 	unsigned long flags;
358 	unsigned int len;
359 
360 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
361 	do {
362 		virtqueue_disable_cb(vq);
363 		while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
364 			struct request *req = blk_mq_rq_from_pdu(vbr);
365 
366 			if (likely(!blk_should_fake_timeout(req->q)))
367 				blk_mq_complete_request(req);
368 			req_done = true;
369 		}
370 		if (unlikely(virtqueue_is_broken(vq)))
371 			break;
372 	} while (!virtqueue_enable_cb(vq));
373 
374 	/* In case queue is stopped waiting for more buffers. */
375 	if (req_done)
376 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
377 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
378 }
379 
380 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
381 {
382 	struct virtio_blk *vblk = hctx->queue->queuedata;
383 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
384 	bool kick;
385 
386 	spin_lock_irq(&vq->lock);
387 	kick = virtqueue_kick_prepare(vq->vq);
388 	spin_unlock_irq(&vq->lock);
389 
390 	if (kick)
391 		virtqueue_notify(vq->vq);
392 }
393 
394 static blk_status_t virtblk_fail_to_queue(struct request *req, int rc)
395 {
396 	virtblk_cleanup_cmd(req);
397 	switch (rc) {
398 	case -ENOSPC:
399 		return BLK_STS_DEV_RESOURCE;
400 	case -ENOMEM:
401 		return BLK_STS_RESOURCE;
402 	default:
403 		return BLK_STS_IOERR;
404 	}
405 }
406 
407 static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
408 					struct virtio_blk *vblk,
409 					struct request *req,
410 					struct virtblk_req *vbr)
411 {
412 	blk_status_t status;
413 	int num;
414 
415 	status = virtblk_setup_cmd(vblk->vdev, req, vbr);
416 	if (unlikely(status))
417 		return status;
418 
419 	num = virtblk_map_data(hctx, req, vbr);
420 	if (unlikely(num < 0))
421 		return virtblk_fail_to_queue(req, -ENOMEM);
422 	vbr->sg_table.nents = num;
423 
424 	blk_mq_start_request(req);
425 
426 	return BLK_STS_OK;
427 }
428 
429 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
430 			   const struct blk_mq_queue_data *bd)
431 {
432 	struct virtio_blk *vblk = hctx->queue->queuedata;
433 	struct request *req = bd->rq;
434 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
435 	unsigned long flags;
436 	int qid = hctx->queue_num;
437 	bool notify = false;
438 	blk_status_t status;
439 	int err;
440 
441 	status = virtblk_prep_rq(hctx, vblk, req, vbr);
442 	if (unlikely(status))
443 		return status;
444 
445 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
446 	err = virtblk_add_req(vblk->vqs[qid].vq, vbr);
447 	if (err) {
448 		virtqueue_kick(vblk->vqs[qid].vq);
449 		/* Don't stop the queue if -ENOMEM: we may have failed to
450 		 * bounce the buffer due to global resource outage.
451 		 */
452 		if (err == -ENOSPC)
453 			blk_mq_stop_hw_queue(hctx);
454 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
455 		virtblk_unmap_data(req, vbr);
456 		return virtblk_fail_to_queue(req, err);
457 	}
458 
459 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
460 		notify = true;
461 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
462 
463 	if (notify)
464 		virtqueue_notify(vblk->vqs[qid].vq);
465 	return BLK_STS_OK;
466 }
467 
468 static bool virtblk_prep_rq_batch(struct request *req)
469 {
470 	struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
471 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
472 
473 	req->mq_hctx->tags->rqs[req->tag] = req;
474 
475 	return virtblk_prep_rq(req->mq_hctx, vblk, req, vbr) == BLK_STS_OK;
476 }
477 
478 static bool virtblk_add_req_batch(struct virtio_blk_vq *vq,
479 					struct request **rqlist)
480 {
481 	unsigned long flags;
482 	int err;
483 	bool kick;
484 
485 	spin_lock_irqsave(&vq->lock, flags);
486 
487 	while (!rq_list_empty(*rqlist)) {
488 		struct request *req = rq_list_pop(rqlist);
489 		struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
490 
491 		err = virtblk_add_req(vq->vq, vbr);
492 		if (err) {
493 			virtblk_unmap_data(req, vbr);
494 			virtblk_cleanup_cmd(req);
495 			blk_mq_requeue_request(req, true);
496 		}
497 	}
498 
499 	kick = virtqueue_kick_prepare(vq->vq);
500 	spin_unlock_irqrestore(&vq->lock, flags);
501 
502 	return kick;
503 }
504 
505 static void virtio_queue_rqs(struct request **rqlist)
506 {
507 	struct request *req, *next, *prev = NULL;
508 	struct request *requeue_list = NULL;
509 
510 	rq_list_for_each_safe(rqlist, req, next) {
511 		struct virtio_blk_vq *vq = get_virtio_blk_vq(req->mq_hctx);
512 		bool kick;
513 
514 		if (!virtblk_prep_rq_batch(req)) {
515 			rq_list_move(rqlist, &requeue_list, req, prev);
516 			req = prev;
517 			if (!req)
518 				continue;
519 		}
520 
521 		if (!next || req->mq_hctx != next->mq_hctx) {
522 			req->rq_next = NULL;
523 			kick = virtblk_add_req_batch(vq, rqlist);
524 			if (kick)
525 				virtqueue_notify(vq->vq);
526 
527 			*rqlist = next;
528 			prev = NULL;
529 		} else
530 			prev = req;
531 	}
532 
533 	*rqlist = requeue_list;
534 }
535 
536 #ifdef CONFIG_BLK_DEV_ZONED
537 static void *virtblk_alloc_report_buffer(struct virtio_blk *vblk,
538 					  unsigned int nr_zones,
539 					  size_t *buflen)
540 {
541 	struct request_queue *q = vblk->disk->queue;
542 	size_t bufsize;
543 	void *buf;
544 
545 	nr_zones = min_t(unsigned int, nr_zones,
546 			 get_capacity(vblk->disk) >> ilog2(vblk->zone_sectors));
547 
548 	bufsize = sizeof(struct virtio_blk_zone_report) +
549 		nr_zones * sizeof(struct virtio_blk_zone_descriptor);
550 	bufsize = min_t(size_t, bufsize,
551 			queue_max_hw_sectors(q) << SECTOR_SHIFT);
552 	bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
553 
554 	while (bufsize >= sizeof(struct virtio_blk_zone_report)) {
555 		buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
556 		if (buf) {
557 			*buflen = bufsize;
558 			return buf;
559 		}
560 		bufsize >>= 1;
561 	}
562 
563 	return NULL;
564 }
565 
566 static int virtblk_submit_zone_report(struct virtio_blk *vblk,
567 				       char *report_buf, size_t report_len,
568 				       sector_t sector)
569 {
570 	struct request_queue *q = vblk->disk->queue;
571 	struct request *req;
572 	struct virtblk_req *vbr;
573 	int err;
574 
575 	req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
576 	if (IS_ERR(req))
577 		return PTR_ERR(req);
578 
579 	vbr = blk_mq_rq_to_pdu(req);
580 	vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
581 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_ZONE_REPORT);
582 	vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, sector);
583 
584 	err = blk_rq_map_kern(q, req, report_buf, report_len, GFP_KERNEL);
585 	if (err)
586 		goto out;
587 
588 	blk_execute_rq(req, false);
589 	err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
590 out:
591 	blk_mq_free_request(req);
592 	return err;
593 }
594 
595 static int virtblk_parse_zone(struct virtio_blk *vblk,
596 			       struct virtio_blk_zone_descriptor *entry,
597 			       unsigned int idx, report_zones_cb cb, void *data)
598 {
599 	struct blk_zone zone = { };
600 
601 	zone.start = virtio64_to_cpu(vblk->vdev, entry->z_start);
602 	if (zone.start + vblk->zone_sectors <= get_capacity(vblk->disk))
603 		zone.len = vblk->zone_sectors;
604 	else
605 		zone.len = get_capacity(vblk->disk) - zone.start;
606 	zone.capacity = virtio64_to_cpu(vblk->vdev, entry->z_cap);
607 	zone.wp = virtio64_to_cpu(vblk->vdev, entry->z_wp);
608 
609 	switch (entry->z_type) {
610 	case VIRTIO_BLK_ZT_SWR:
611 		zone.type = BLK_ZONE_TYPE_SEQWRITE_REQ;
612 		break;
613 	case VIRTIO_BLK_ZT_SWP:
614 		zone.type = BLK_ZONE_TYPE_SEQWRITE_PREF;
615 		break;
616 	case VIRTIO_BLK_ZT_CONV:
617 		zone.type = BLK_ZONE_TYPE_CONVENTIONAL;
618 		break;
619 	default:
620 		dev_err(&vblk->vdev->dev, "zone %llu: invalid type %#x\n",
621 			zone.start, entry->z_type);
622 		return -EIO;
623 	}
624 
625 	switch (entry->z_state) {
626 	case VIRTIO_BLK_ZS_EMPTY:
627 		zone.cond = BLK_ZONE_COND_EMPTY;
628 		break;
629 	case VIRTIO_BLK_ZS_CLOSED:
630 		zone.cond = BLK_ZONE_COND_CLOSED;
631 		break;
632 	case VIRTIO_BLK_ZS_FULL:
633 		zone.cond = BLK_ZONE_COND_FULL;
634 		zone.wp = zone.start + zone.len;
635 		break;
636 	case VIRTIO_BLK_ZS_EOPEN:
637 		zone.cond = BLK_ZONE_COND_EXP_OPEN;
638 		break;
639 	case VIRTIO_BLK_ZS_IOPEN:
640 		zone.cond = BLK_ZONE_COND_IMP_OPEN;
641 		break;
642 	case VIRTIO_BLK_ZS_NOT_WP:
643 		zone.cond = BLK_ZONE_COND_NOT_WP;
644 		break;
645 	case VIRTIO_BLK_ZS_RDONLY:
646 		zone.cond = BLK_ZONE_COND_READONLY;
647 		zone.wp = ULONG_MAX;
648 		break;
649 	case VIRTIO_BLK_ZS_OFFLINE:
650 		zone.cond = BLK_ZONE_COND_OFFLINE;
651 		zone.wp = ULONG_MAX;
652 		break;
653 	default:
654 		dev_err(&vblk->vdev->dev, "zone %llu: invalid condition %#x\n",
655 			zone.start, entry->z_state);
656 		return -EIO;
657 	}
658 
659 	/*
660 	 * The callback below checks the validity of the reported
661 	 * entry data, no need to further validate it here.
662 	 */
663 	return cb(&zone, idx, data);
664 }
665 
666 static int virtblk_report_zones(struct gendisk *disk, sector_t sector,
667 				 unsigned int nr_zones, report_zones_cb cb,
668 				 void *data)
669 {
670 	struct virtio_blk *vblk = disk->private_data;
671 	struct virtio_blk_zone_report *report;
672 	unsigned long long nz, i;
673 	size_t buflen;
674 	unsigned int zone_idx = 0;
675 	int ret;
676 
677 	if (WARN_ON_ONCE(!vblk->zone_sectors))
678 		return -EOPNOTSUPP;
679 
680 	report = virtblk_alloc_report_buffer(vblk, nr_zones, &buflen);
681 	if (!report)
682 		return -ENOMEM;
683 
684 	mutex_lock(&vblk->vdev_mutex);
685 
686 	if (!vblk->vdev) {
687 		ret = -ENXIO;
688 		goto fail_report;
689 	}
690 
691 	while (zone_idx < nr_zones && sector < get_capacity(vblk->disk)) {
692 		memset(report, 0, buflen);
693 
694 		ret = virtblk_submit_zone_report(vblk, (char *)report,
695 						 buflen, sector);
696 		if (ret)
697 			goto fail_report;
698 
699 		nz = min_t(u64, virtio64_to_cpu(vblk->vdev, report->nr_zones),
700 			   nr_zones);
701 		if (!nz)
702 			break;
703 
704 		for (i = 0; i < nz && zone_idx < nr_zones; i++) {
705 			ret = virtblk_parse_zone(vblk, &report->zones[i],
706 						 zone_idx, cb, data);
707 			if (ret)
708 				goto fail_report;
709 
710 			sector = virtio64_to_cpu(vblk->vdev,
711 						 report->zones[i].z_start) +
712 				 vblk->zone_sectors;
713 			zone_idx++;
714 		}
715 	}
716 
717 	if (zone_idx > 0)
718 		ret = zone_idx;
719 	else
720 		ret = -EINVAL;
721 fail_report:
722 	mutex_unlock(&vblk->vdev_mutex);
723 	kvfree(report);
724 	return ret;
725 }
726 
727 static void virtblk_revalidate_zones(struct virtio_blk *vblk)
728 {
729 	u8 model;
730 
731 	virtio_cread(vblk->vdev, struct virtio_blk_config,
732 		     zoned.model, &model);
733 	switch (model) {
734 	default:
735 		dev_err(&vblk->vdev->dev, "unknown zone model %d\n", model);
736 		fallthrough;
737 	case VIRTIO_BLK_Z_NONE:
738 	case VIRTIO_BLK_Z_HA:
739 		disk_set_zoned(vblk->disk, BLK_ZONED_NONE);
740 		return;
741 	case VIRTIO_BLK_Z_HM:
742 		WARN_ON_ONCE(!vblk->zone_sectors);
743 		if (!blk_revalidate_disk_zones(vblk->disk, NULL))
744 			set_capacity_and_notify(vblk->disk, 0);
745 	}
746 }
747 
748 static int virtblk_probe_zoned_device(struct virtio_device *vdev,
749 				       struct virtio_blk *vblk,
750 				       struct request_queue *q)
751 {
752 	u32 v, wg;
753 	u8 model;
754 
755 	virtio_cread(vdev, struct virtio_blk_config,
756 		     zoned.model, &model);
757 
758 	switch (model) {
759 	case VIRTIO_BLK_Z_NONE:
760 	case VIRTIO_BLK_Z_HA:
761 		/* Present the host-aware device as non-zoned */
762 		return 0;
763 	case VIRTIO_BLK_Z_HM:
764 		break;
765 	default:
766 		dev_err(&vdev->dev, "unsupported zone model %d\n", model);
767 		return -EINVAL;
768 	}
769 
770 	dev_dbg(&vdev->dev, "probing host-managed zoned device\n");
771 
772 	disk_set_zoned(vblk->disk, BLK_ZONED_HM);
773 	blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
774 
775 	virtio_cread(vdev, struct virtio_blk_config,
776 		     zoned.max_open_zones, &v);
777 	disk_set_max_open_zones(vblk->disk, v);
778 	dev_dbg(&vdev->dev, "max open zones = %u\n", v);
779 
780 	virtio_cread(vdev, struct virtio_blk_config,
781 		     zoned.max_active_zones, &v);
782 	disk_set_max_active_zones(vblk->disk, v);
783 	dev_dbg(&vdev->dev, "max active zones = %u\n", v);
784 
785 	virtio_cread(vdev, struct virtio_blk_config,
786 		     zoned.write_granularity, &wg);
787 	if (!wg) {
788 		dev_warn(&vdev->dev, "zero write granularity reported\n");
789 		return -ENODEV;
790 	}
791 	blk_queue_physical_block_size(q, wg);
792 	blk_queue_io_min(q, wg);
793 
794 	dev_dbg(&vdev->dev, "write granularity = %u\n", wg);
795 
796 	/*
797 	 * virtio ZBD specification doesn't require zones to be a power of
798 	 * two sectors in size, but the code in this driver expects that.
799 	 */
800 	virtio_cread(vdev, struct virtio_blk_config, zoned.zone_sectors,
801 		     &vblk->zone_sectors);
802 	if (vblk->zone_sectors == 0 || !is_power_of_2(vblk->zone_sectors)) {
803 		dev_err(&vdev->dev,
804 			"zoned device with non power of two zone size %u\n",
805 			vblk->zone_sectors);
806 		return -ENODEV;
807 	}
808 	blk_queue_chunk_sectors(q, vblk->zone_sectors);
809 	dev_dbg(&vdev->dev, "zone sectors = %u\n", vblk->zone_sectors);
810 
811 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
812 		dev_warn(&vblk->vdev->dev,
813 			 "ignoring negotiated F_DISCARD for zoned device\n");
814 		blk_queue_max_discard_sectors(q, 0);
815 	}
816 
817 	virtio_cread(vdev, struct virtio_blk_config,
818 		     zoned.max_append_sectors, &v);
819 	if (!v) {
820 		dev_warn(&vdev->dev, "zero max_append_sectors reported\n");
821 		return -ENODEV;
822 	}
823 	if ((v << SECTOR_SHIFT) < wg) {
824 		dev_err(&vdev->dev,
825 			"write granularity %u exceeds max_append_sectors %u limit\n",
826 			wg, v);
827 		return -ENODEV;
828 	}
829 	blk_queue_max_zone_append_sectors(q, v);
830 	dev_dbg(&vdev->dev, "max append sectors = %u\n", v);
831 
832 	return blk_revalidate_disk_zones(vblk->disk, NULL);
833 }
834 
835 #else
836 
837 /*
838  * Zoned block device support is not configured in this kernel.
839  * Host-managed zoned devices can't be supported, but others are
840  * good to go as regular block devices.
841  */
842 #define virtblk_report_zones       NULL
843 
844 static inline void virtblk_revalidate_zones(struct virtio_blk *vblk)
845 {
846 }
847 
848 static inline int virtblk_probe_zoned_device(struct virtio_device *vdev,
849 			struct virtio_blk *vblk, struct request_queue *q)
850 {
851 	u8 model;
852 
853 	virtio_cread(vdev, struct virtio_blk_config, zoned.model, &model);
854 	if (model == VIRTIO_BLK_Z_HM) {
855 		dev_err(&vdev->dev,
856 			"virtio_blk: zoned devices are not supported");
857 		return -EOPNOTSUPP;
858 	}
859 
860 	return 0;
861 }
862 #endif /* CONFIG_BLK_DEV_ZONED */
863 
864 /* return id (s/n) string for *disk to *id_str
865  */
866 static int virtblk_get_id(struct gendisk *disk, char *id_str)
867 {
868 	struct virtio_blk *vblk = disk->private_data;
869 	struct request_queue *q = vblk->disk->queue;
870 	struct request *req;
871 	struct virtblk_req *vbr;
872 	int err;
873 
874 	req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
875 	if (IS_ERR(req))
876 		return PTR_ERR(req);
877 
878 	vbr = blk_mq_rq_to_pdu(req);
879 	vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
880 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
881 	vbr->out_hdr.sector = 0;
882 
883 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
884 	if (err)
885 		goto out;
886 
887 	blk_execute_rq(req, false);
888 	err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
889 out:
890 	blk_mq_free_request(req);
891 	return err;
892 }
893 
894 /* We provide getgeo only to please some old bootloader/partitioning tools */
895 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
896 {
897 	struct virtio_blk *vblk = bd->bd_disk->private_data;
898 	int ret = 0;
899 
900 	mutex_lock(&vblk->vdev_mutex);
901 
902 	if (!vblk->vdev) {
903 		ret = -ENXIO;
904 		goto out;
905 	}
906 
907 	/* see if the host passed in geometry config */
908 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
909 		virtio_cread(vblk->vdev, struct virtio_blk_config,
910 			     geometry.cylinders, &geo->cylinders);
911 		virtio_cread(vblk->vdev, struct virtio_blk_config,
912 			     geometry.heads, &geo->heads);
913 		virtio_cread(vblk->vdev, struct virtio_blk_config,
914 			     geometry.sectors, &geo->sectors);
915 	} else {
916 		/* some standard values, similar to sd */
917 		geo->heads = 1 << 6;
918 		geo->sectors = 1 << 5;
919 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
920 	}
921 out:
922 	mutex_unlock(&vblk->vdev_mutex);
923 	return ret;
924 }
925 
926 static void virtblk_free_disk(struct gendisk *disk)
927 {
928 	struct virtio_blk *vblk = disk->private_data;
929 
930 	ida_free(&vd_index_ida, vblk->index);
931 	mutex_destroy(&vblk->vdev_mutex);
932 	kfree(vblk);
933 }
934 
935 static const struct block_device_operations virtblk_fops = {
936 	.owner  	= THIS_MODULE,
937 	.getgeo		= virtblk_getgeo,
938 	.free_disk	= virtblk_free_disk,
939 	.report_zones	= virtblk_report_zones,
940 };
941 
942 static int index_to_minor(int index)
943 {
944 	return index << PART_BITS;
945 }
946 
947 static int minor_to_index(int minor)
948 {
949 	return minor >> PART_BITS;
950 }
951 
952 static ssize_t serial_show(struct device *dev,
953 			   struct device_attribute *attr, char *buf)
954 {
955 	struct gendisk *disk = dev_to_disk(dev);
956 	int err;
957 
958 	/* sysfs gives us a PAGE_SIZE buffer */
959 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
960 
961 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
962 	err = virtblk_get_id(disk, buf);
963 	if (!err)
964 		return strlen(buf);
965 
966 	if (err == -EIO) /* Unsupported? Make it empty. */
967 		return 0;
968 
969 	return err;
970 }
971 
972 static DEVICE_ATTR_RO(serial);
973 
974 /* The queue's logical block size must be set before calling this */
975 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
976 {
977 	struct virtio_device *vdev = vblk->vdev;
978 	struct request_queue *q = vblk->disk->queue;
979 	char cap_str_2[10], cap_str_10[10];
980 	unsigned long long nblocks;
981 	u64 capacity;
982 
983 	/* Host must always specify the capacity. */
984 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
985 
986 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
987 
988 	string_get_size(nblocks, queue_logical_block_size(q),
989 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
990 	string_get_size(nblocks, queue_logical_block_size(q),
991 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
992 
993 	dev_notice(&vdev->dev,
994 		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
995 		   vblk->disk->disk_name,
996 		   resize ? "new size: " : "",
997 		   nblocks,
998 		   queue_logical_block_size(q),
999 		   cap_str_10,
1000 		   cap_str_2);
1001 
1002 	set_capacity_and_notify(vblk->disk, capacity);
1003 }
1004 
1005 static void virtblk_config_changed_work(struct work_struct *work)
1006 {
1007 	struct virtio_blk *vblk =
1008 		container_of(work, struct virtio_blk, config_work);
1009 
1010 	virtblk_revalidate_zones(vblk);
1011 	virtblk_update_capacity(vblk, true);
1012 }
1013 
1014 static void virtblk_config_changed(struct virtio_device *vdev)
1015 {
1016 	struct virtio_blk *vblk = vdev->priv;
1017 
1018 	queue_work(virtblk_wq, &vblk->config_work);
1019 }
1020 
1021 static int init_vq(struct virtio_blk *vblk)
1022 {
1023 	int err;
1024 	unsigned short i;
1025 	vq_callback_t **callbacks;
1026 	const char **names;
1027 	struct virtqueue **vqs;
1028 	unsigned short num_vqs;
1029 	unsigned short num_poll_vqs;
1030 	struct virtio_device *vdev = vblk->vdev;
1031 	struct irq_affinity desc = { 0, };
1032 
1033 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
1034 				   struct virtio_blk_config, num_queues,
1035 				   &num_vqs);
1036 	if (err)
1037 		num_vqs = 1;
1038 
1039 	if (!err && !num_vqs) {
1040 		dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
1041 		return -EINVAL;
1042 	}
1043 
1044 	num_vqs = min_t(unsigned int,
1045 			min_not_zero(num_request_queues, nr_cpu_ids),
1046 			num_vqs);
1047 
1048 	num_poll_vqs = min_t(unsigned int, poll_queues, num_vqs - 1);
1049 
1050 	vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
1051 	vblk->io_queues[HCTX_TYPE_READ] = 0;
1052 	vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
1053 
1054 	dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n",
1055 				vblk->io_queues[HCTX_TYPE_DEFAULT],
1056 				vblk->io_queues[HCTX_TYPE_READ],
1057 				vblk->io_queues[HCTX_TYPE_POLL]);
1058 
1059 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
1060 	if (!vblk->vqs)
1061 		return -ENOMEM;
1062 
1063 	names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
1064 	callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
1065 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
1066 	if (!names || !callbacks || !vqs) {
1067 		err = -ENOMEM;
1068 		goto out;
1069 	}
1070 
1071 	for (i = 0; i < num_vqs - num_poll_vqs; i++) {
1072 		callbacks[i] = virtblk_done;
1073 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%u", i);
1074 		names[i] = vblk->vqs[i].name;
1075 	}
1076 
1077 	for (; i < num_vqs; i++) {
1078 		callbacks[i] = NULL;
1079 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%u", i);
1080 		names[i] = vblk->vqs[i].name;
1081 	}
1082 
1083 	/* Discover virtqueues and write information to configuration.  */
1084 	err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
1085 	if (err)
1086 		goto out;
1087 
1088 	for (i = 0; i < num_vqs; i++) {
1089 		spin_lock_init(&vblk->vqs[i].lock);
1090 		vblk->vqs[i].vq = vqs[i];
1091 	}
1092 	vblk->num_vqs = num_vqs;
1093 
1094 out:
1095 	kfree(vqs);
1096 	kfree(callbacks);
1097 	kfree(names);
1098 	if (err)
1099 		kfree(vblk->vqs);
1100 	return err;
1101 }
1102 
1103 /*
1104  * Legacy naming scheme used for virtio devices.  We are stuck with it for
1105  * virtio blk but don't ever use it for any new driver.
1106  */
1107 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
1108 {
1109 	const int base = 'z' - 'a' + 1;
1110 	char *begin = buf + strlen(prefix);
1111 	char *end = buf + buflen;
1112 	char *p;
1113 	int unit;
1114 
1115 	p = end - 1;
1116 	*p = '\0';
1117 	unit = base;
1118 	do {
1119 		if (p == begin)
1120 			return -EINVAL;
1121 		*--p = 'a' + (index % unit);
1122 		index = (index / unit) - 1;
1123 	} while (index >= 0);
1124 
1125 	memmove(begin, p, end - p);
1126 	memcpy(buf, prefix, strlen(prefix));
1127 
1128 	return 0;
1129 }
1130 
1131 static int virtblk_get_cache_mode(struct virtio_device *vdev)
1132 {
1133 	u8 writeback;
1134 	int err;
1135 
1136 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
1137 				   struct virtio_blk_config, wce,
1138 				   &writeback);
1139 
1140 	/*
1141 	 * If WCE is not configurable and flush is not available,
1142 	 * assume no writeback cache is in use.
1143 	 */
1144 	if (err)
1145 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
1146 
1147 	return writeback;
1148 }
1149 
1150 static void virtblk_update_cache_mode(struct virtio_device *vdev)
1151 {
1152 	u8 writeback = virtblk_get_cache_mode(vdev);
1153 	struct virtio_blk *vblk = vdev->priv;
1154 
1155 	blk_queue_write_cache(vblk->disk->queue, writeback, false);
1156 }
1157 
1158 static const char *const virtblk_cache_types[] = {
1159 	"write through", "write back"
1160 };
1161 
1162 static ssize_t
1163 cache_type_store(struct device *dev, struct device_attribute *attr,
1164 		 const char *buf, size_t count)
1165 {
1166 	struct gendisk *disk = dev_to_disk(dev);
1167 	struct virtio_blk *vblk = disk->private_data;
1168 	struct virtio_device *vdev = vblk->vdev;
1169 	int i;
1170 
1171 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
1172 	i = sysfs_match_string(virtblk_cache_types, buf);
1173 	if (i < 0)
1174 		return i;
1175 
1176 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
1177 	virtblk_update_cache_mode(vdev);
1178 	return count;
1179 }
1180 
1181 static ssize_t
1182 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
1183 {
1184 	struct gendisk *disk = dev_to_disk(dev);
1185 	struct virtio_blk *vblk = disk->private_data;
1186 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
1187 
1188 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
1189 	return sysfs_emit(buf, "%s\n", virtblk_cache_types[writeback]);
1190 }
1191 
1192 static DEVICE_ATTR_RW(cache_type);
1193 
1194 static struct attribute *virtblk_attrs[] = {
1195 	&dev_attr_serial.attr,
1196 	&dev_attr_cache_type.attr,
1197 	NULL,
1198 };
1199 
1200 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
1201 		struct attribute *a, int n)
1202 {
1203 	struct device *dev = kobj_to_dev(kobj);
1204 	struct gendisk *disk = dev_to_disk(dev);
1205 	struct virtio_blk *vblk = disk->private_data;
1206 	struct virtio_device *vdev = vblk->vdev;
1207 
1208 	if (a == &dev_attr_cache_type.attr &&
1209 	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
1210 		return S_IRUGO;
1211 
1212 	return a->mode;
1213 }
1214 
1215 static const struct attribute_group virtblk_attr_group = {
1216 	.attrs = virtblk_attrs,
1217 	.is_visible = virtblk_attrs_are_visible,
1218 };
1219 
1220 static const struct attribute_group *virtblk_attr_groups[] = {
1221 	&virtblk_attr_group,
1222 	NULL,
1223 };
1224 
1225 static void virtblk_map_queues(struct blk_mq_tag_set *set)
1226 {
1227 	struct virtio_blk *vblk = set->driver_data;
1228 	int i, qoff;
1229 
1230 	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1231 		struct blk_mq_queue_map *map = &set->map[i];
1232 
1233 		map->nr_queues = vblk->io_queues[i];
1234 		map->queue_offset = qoff;
1235 		qoff += map->nr_queues;
1236 
1237 		if (map->nr_queues == 0)
1238 			continue;
1239 
1240 		/*
1241 		 * Regular queues have interrupts and hence CPU affinity is
1242 		 * defined by the core virtio code, but polling queues have
1243 		 * no interrupts so we let the block layer assign CPU affinity.
1244 		 */
1245 		if (i == HCTX_TYPE_POLL)
1246 			blk_mq_map_queues(&set->map[i]);
1247 		else
1248 			blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
1249 	}
1250 }
1251 
1252 static void virtblk_complete_batch(struct io_comp_batch *iob)
1253 {
1254 	struct request *req;
1255 
1256 	rq_list_for_each(&iob->req_list, req) {
1257 		virtblk_unmap_data(req, blk_mq_rq_to_pdu(req));
1258 		virtblk_cleanup_cmd(req);
1259 	}
1260 	blk_mq_end_request_batch(iob);
1261 }
1262 
1263 static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1264 {
1265 	struct virtio_blk *vblk = hctx->queue->queuedata;
1266 	struct virtio_blk_vq *vq = get_virtio_blk_vq(hctx);
1267 	struct virtblk_req *vbr;
1268 	unsigned long flags;
1269 	unsigned int len;
1270 	int found = 0;
1271 
1272 	spin_lock_irqsave(&vq->lock, flags);
1273 
1274 	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
1275 		struct request *req = blk_mq_rq_from_pdu(vbr);
1276 
1277 		found++;
1278 		if (!blk_mq_complete_request_remote(req) &&
1279 		    !blk_mq_add_to_batch(req, iob, virtblk_vbr_status(vbr),
1280 						virtblk_complete_batch))
1281 			virtblk_request_done(req);
1282 	}
1283 
1284 	if (found)
1285 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
1286 
1287 	spin_unlock_irqrestore(&vq->lock, flags);
1288 
1289 	return found;
1290 }
1291 
1292 static const struct blk_mq_ops virtio_mq_ops = {
1293 	.queue_rq	= virtio_queue_rq,
1294 	.queue_rqs	= virtio_queue_rqs,
1295 	.commit_rqs	= virtio_commit_rqs,
1296 	.complete	= virtblk_request_done,
1297 	.map_queues	= virtblk_map_queues,
1298 	.poll		= virtblk_poll,
1299 };
1300 
1301 static unsigned int virtblk_queue_depth;
1302 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
1303 
1304 static int virtblk_probe(struct virtio_device *vdev)
1305 {
1306 	struct virtio_blk *vblk;
1307 	struct request_queue *q;
1308 	int err, index;
1309 
1310 	u32 v, blk_size, max_size, sg_elems, opt_io_size;
1311 	u32 max_discard_segs = 0;
1312 	u32 discard_granularity = 0;
1313 	u16 min_io_size;
1314 	u8 physical_block_exp, alignment_offset;
1315 	unsigned int queue_depth;
1316 	size_t max_dma_size;
1317 
1318 	if (!vdev->config->get) {
1319 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
1320 			__func__);
1321 		return -EINVAL;
1322 	}
1323 
1324 	err = ida_alloc_range(&vd_index_ida, 0,
1325 			      minor_to_index(1 << MINORBITS) - 1, GFP_KERNEL);
1326 	if (err < 0)
1327 		goto out;
1328 	index = err;
1329 
1330 	/* We need to know how many segments before we allocate. */
1331 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
1332 				   struct virtio_blk_config, seg_max,
1333 				   &sg_elems);
1334 
1335 	/* We need at least one SG element, whatever they say. */
1336 	if (err || !sg_elems)
1337 		sg_elems = 1;
1338 
1339 	/* Prevent integer overflows and honor max vq size */
1340 	sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
1341 
1342 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
1343 	if (!vblk) {
1344 		err = -ENOMEM;
1345 		goto out_free_index;
1346 	}
1347 
1348 	mutex_init(&vblk->vdev_mutex);
1349 
1350 	vblk->vdev = vdev;
1351 
1352 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
1353 
1354 	err = init_vq(vblk);
1355 	if (err)
1356 		goto out_free_vblk;
1357 
1358 	/* Default queue sizing is to fill the ring. */
1359 	if (!virtblk_queue_depth) {
1360 		queue_depth = vblk->vqs[0].vq->num_free;
1361 		/* ... but without indirect descs, we use 2 descs per req */
1362 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
1363 			queue_depth /= 2;
1364 	} else {
1365 		queue_depth = virtblk_queue_depth;
1366 	}
1367 
1368 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
1369 	vblk->tag_set.ops = &virtio_mq_ops;
1370 	vblk->tag_set.queue_depth = queue_depth;
1371 	vblk->tag_set.numa_node = NUMA_NO_NODE;
1372 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1373 	vblk->tag_set.cmd_size =
1374 		sizeof(struct virtblk_req) +
1375 		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
1376 	vblk->tag_set.driver_data = vblk;
1377 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
1378 	vblk->tag_set.nr_maps = 1;
1379 	if (vblk->io_queues[HCTX_TYPE_POLL])
1380 		vblk->tag_set.nr_maps = 3;
1381 
1382 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
1383 	if (err)
1384 		goto out_free_vq;
1385 
1386 	vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
1387 	if (IS_ERR(vblk->disk)) {
1388 		err = PTR_ERR(vblk->disk);
1389 		goto out_free_tags;
1390 	}
1391 	q = vblk->disk->queue;
1392 
1393 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
1394 
1395 	vblk->disk->major = major;
1396 	vblk->disk->first_minor = index_to_minor(index);
1397 	vblk->disk->minors = 1 << PART_BITS;
1398 	vblk->disk->private_data = vblk;
1399 	vblk->disk->fops = &virtblk_fops;
1400 	vblk->index = index;
1401 
1402 	/* configure queue flush support */
1403 	virtblk_update_cache_mode(vdev);
1404 
1405 	/* If disk is read-only in the host, the guest should obey */
1406 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
1407 		set_disk_ro(vblk->disk, 1);
1408 
1409 	/* We can handle whatever the host told us to handle. */
1410 	blk_queue_max_segments(q, sg_elems);
1411 
1412 	/* No real sector limit. */
1413 	blk_queue_max_hw_sectors(q, UINT_MAX);
1414 
1415 	max_dma_size = virtio_max_dma_size(vdev);
1416 	max_size = max_dma_size > U32_MAX ? U32_MAX : max_dma_size;
1417 
1418 	/* Host can optionally specify maximum segment size and number of
1419 	 * segments. */
1420 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
1421 				   struct virtio_blk_config, size_max, &v);
1422 	if (!err)
1423 		max_size = min(max_size, v);
1424 
1425 	blk_queue_max_segment_size(q, max_size);
1426 
1427 	/* Host can optionally specify the block size of the device */
1428 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
1429 				   struct virtio_blk_config, blk_size,
1430 				   &blk_size);
1431 	if (!err) {
1432 		err = blk_validate_block_size(blk_size);
1433 		if (err) {
1434 			dev_err(&vdev->dev,
1435 				"virtio_blk: invalid block size: 0x%x\n",
1436 				blk_size);
1437 			goto out_cleanup_disk;
1438 		}
1439 
1440 		blk_queue_logical_block_size(q, blk_size);
1441 	} else
1442 		blk_size = queue_logical_block_size(q);
1443 
1444 	/* Use topology information if available */
1445 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1446 				   struct virtio_blk_config, physical_block_exp,
1447 				   &physical_block_exp);
1448 	if (!err && physical_block_exp)
1449 		blk_queue_physical_block_size(q,
1450 				blk_size * (1 << physical_block_exp));
1451 
1452 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1453 				   struct virtio_blk_config, alignment_offset,
1454 				   &alignment_offset);
1455 	if (!err && alignment_offset)
1456 		blk_queue_alignment_offset(q, blk_size * alignment_offset);
1457 
1458 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1459 				   struct virtio_blk_config, min_io_size,
1460 				   &min_io_size);
1461 	if (!err && min_io_size)
1462 		blk_queue_io_min(q, blk_size * min_io_size);
1463 
1464 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1465 				   struct virtio_blk_config, opt_io_size,
1466 				   &opt_io_size);
1467 	if (!err && opt_io_size)
1468 		blk_queue_io_opt(q, blk_size * opt_io_size);
1469 
1470 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
1471 		virtio_cread(vdev, struct virtio_blk_config,
1472 			     discard_sector_alignment, &discard_granularity);
1473 
1474 		virtio_cread(vdev, struct virtio_blk_config,
1475 			     max_discard_sectors, &v);
1476 		blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
1477 
1478 		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
1479 			     &max_discard_segs);
1480 	}
1481 
1482 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
1483 		virtio_cread(vdev, struct virtio_blk_config,
1484 			     max_write_zeroes_sectors, &v);
1485 		blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
1486 	}
1487 
1488 	/* The discard and secure erase limits are combined since the Linux
1489 	 * block layer uses the same limit for both commands.
1490 	 *
1491 	 * If both VIRTIO_BLK_F_SECURE_ERASE and VIRTIO_BLK_F_DISCARD features
1492 	 * are negotiated, we will use the minimum between the limits.
1493 	 *
1494 	 * discard sector alignment is set to the minimum between discard_sector_alignment
1495 	 * and secure_erase_sector_alignment.
1496 	 *
1497 	 * max discard sectors is set to the minimum between max_discard_seg and
1498 	 * max_secure_erase_seg.
1499 	 */
1500 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1501 
1502 		virtio_cread(vdev, struct virtio_blk_config,
1503 			     secure_erase_sector_alignment, &v);
1504 
1505 		/* secure_erase_sector_alignment should not be zero, the device should set a
1506 		 * valid number of sectors.
1507 		 */
1508 		if (!v) {
1509 			dev_err(&vdev->dev,
1510 				"virtio_blk: secure_erase_sector_alignment can't be 0\n");
1511 			err = -EINVAL;
1512 			goto out_cleanup_disk;
1513 		}
1514 
1515 		discard_granularity = min_not_zero(discard_granularity, v);
1516 
1517 		virtio_cread(vdev, struct virtio_blk_config,
1518 			     max_secure_erase_sectors, &v);
1519 
1520 		/* max_secure_erase_sectors should not be zero, the device should set a
1521 		 * valid number of sectors.
1522 		 */
1523 		if (!v) {
1524 			dev_err(&vdev->dev,
1525 				"virtio_blk: max_secure_erase_sectors can't be 0\n");
1526 			err = -EINVAL;
1527 			goto out_cleanup_disk;
1528 		}
1529 
1530 		blk_queue_max_secure_erase_sectors(q, v);
1531 
1532 		virtio_cread(vdev, struct virtio_blk_config,
1533 			     max_secure_erase_seg, &v);
1534 
1535 		/* max_secure_erase_seg should not be zero, the device should set a
1536 		 * valid number of segments
1537 		 */
1538 		if (!v) {
1539 			dev_err(&vdev->dev,
1540 				"virtio_blk: max_secure_erase_seg can't be 0\n");
1541 			err = -EINVAL;
1542 			goto out_cleanup_disk;
1543 		}
1544 
1545 		max_discard_segs = min_not_zero(max_discard_segs, v);
1546 	}
1547 
1548 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD) ||
1549 	    virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1550 		/* max_discard_seg and discard_granularity will be 0 only
1551 		 * if max_discard_seg and discard_sector_alignment fields in the virtio
1552 		 * config are 0 and VIRTIO_BLK_F_SECURE_ERASE feature is not negotiated.
1553 		 * In this case, we use default values.
1554 		 */
1555 		if (!max_discard_segs)
1556 			max_discard_segs = sg_elems;
1557 
1558 		blk_queue_max_discard_segments(q,
1559 					       min(max_discard_segs, MAX_DISCARD_SEGMENTS));
1560 
1561 		if (discard_granularity)
1562 			q->limits.discard_granularity = discard_granularity << SECTOR_SHIFT;
1563 		else
1564 			q->limits.discard_granularity = blk_size;
1565 	}
1566 
1567 	virtblk_update_capacity(vblk, false);
1568 	virtio_device_ready(vdev);
1569 
1570 	/*
1571 	 * All steps that follow use the VQs therefore they need to be
1572 	 * placed after the virtio_device_ready() call above.
1573 	 */
1574 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_ZONED)) {
1575 		err = virtblk_probe_zoned_device(vdev, vblk, q);
1576 		if (err)
1577 			goto out_cleanup_disk;
1578 	}
1579 
1580 	err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
1581 	if (err)
1582 		goto out_cleanup_disk;
1583 
1584 	return 0;
1585 
1586 out_cleanup_disk:
1587 	put_disk(vblk->disk);
1588 out_free_tags:
1589 	blk_mq_free_tag_set(&vblk->tag_set);
1590 out_free_vq:
1591 	vdev->config->del_vqs(vdev);
1592 	kfree(vblk->vqs);
1593 out_free_vblk:
1594 	kfree(vblk);
1595 out_free_index:
1596 	ida_free(&vd_index_ida, index);
1597 out:
1598 	return err;
1599 }
1600 
1601 static void virtblk_remove(struct virtio_device *vdev)
1602 {
1603 	struct virtio_blk *vblk = vdev->priv;
1604 
1605 	/* Make sure no work handler is accessing the device. */
1606 	flush_work(&vblk->config_work);
1607 
1608 	del_gendisk(vblk->disk);
1609 	blk_mq_free_tag_set(&vblk->tag_set);
1610 
1611 	mutex_lock(&vblk->vdev_mutex);
1612 
1613 	/* Stop all the virtqueues. */
1614 	virtio_reset_device(vdev);
1615 
1616 	/* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
1617 	vblk->vdev = NULL;
1618 
1619 	vdev->config->del_vqs(vdev);
1620 	kfree(vblk->vqs);
1621 
1622 	mutex_unlock(&vblk->vdev_mutex);
1623 
1624 	put_disk(vblk->disk);
1625 }
1626 
1627 #ifdef CONFIG_PM_SLEEP
1628 static int virtblk_freeze(struct virtio_device *vdev)
1629 {
1630 	struct virtio_blk *vblk = vdev->priv;
1631 
1632 	/* Ensure no requests in virtqueues before deleting vqs. */
1633 	blk_mq_freeze_queue(vblk->disk->queue);
1634 
1635 	/* Ensure we don't receive any more interrupts */
1636 	virtio_reset_device(vdev);
1637 
1638 	/* Make sure no work handler is accessing the device. */
1639 	flush_work(&vblk->config_work);
1640 
1641 	vdev->config->del_vqs(vdev);
1642 	kfree(vblk->vqs);
1643 
1644 	return 0;
1645 }
1646 
1647 static int virtblk_restore(struct virtio_device *vdev)
1648 {
1649 	struct virtio_blk *vblk = vdev->priv;
1650 	int ret;
1651 
1652 	ret = init_vq(vdev->priv);
1653 	if (ret)
1654 		return ret;
1655 
1656 	virtio_device_ready(vdev);
1657 
1658 	blk_mq_unfreeze_queue(vblk->disk->queue);
1659 	return 0;
1660 }
1661 #endif
1662 
1663 static const struct virtio_device_id id_table[] = {
1664 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1665 	{ 0 },
1666 };
1667 
1668 static unsigned int features_legacy[] = {
1669 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1670 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1671 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1672 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1673 	VIRTIO_BLK_F_SECURE_ERASE,
1674 }
1675 ;
1676 static unsigned int features[] = {
1677 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1678 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1679 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1680 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1681 	VIRTIO_BLK_F_SECURE_ERASE, VIRTIO_BLK_F_ZONED,
1682 };
1683 
1684 static struct virtio_driver virtio_blk = {
1685 	.feature_table			= features,
1686 	.feature_table_size		= ARRAY_SIZE(features),
1687 	.feature_table_legacy		= features_legacy,
1688 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
1689 	.driver.name			= KBUILD_MODNAME,
1690 	.driver.owner			= THIS_MODULE,
1691 	.id_table			= id_table,
1692 	.probe				= virtblk_probe,
1693 	.remove				= virtblk_remove,
1694 	.config_changed			= virtblk_config_changed,
1695 #ifdef CONFIG_PM_SLEEP
1696 	.freeze				= virtblk_freeze,
1697 	.restore			= virtblk_restore,
1698 #endif
1699 };
1700 
1701 static int __init virtio_blk_init(void)
1702 {
1703 	int error;
1704 
1705 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1706 	if (!virtblk_wq)
1707 		return -ENOMEM;
1708 
1709 	major = register_blkdev(0, "virtblk");
1710 	if (major < 0) {
1711 		error = major;
1712 		goto out_destroy_workqueue;
1713 	}
1714 
1715 	error = register_virtio_driver(&virtio_blk);
1716 	if (error)
1717 		goto out_unregister_blkdev;
1718 	return 0;
1719 
1720 out_unregister_blkdev:
1721 	unregister_blkdev(major, "virtblk");
1722 out_destroy_workqueue:
1723 	destroy_workqueue(virtblk_wq);
1724 	return error;
1725 }
1726 
1727 static void __exit virtio_blk_fini(void)
1728 {
1729 	unregister_virtio_driver(&virtio_blk);
1730 	unregister_blkdev(major, "virtblk");
1731 	destroy_workqueue(virtblk_wq);
1732 }
1733 module_init(virtio_blk_init);
1734 module_exit(virtio_blk_fini);
1735 
1736 MODULE_DEVICE_TABLE(virtio, id_table);
1737 MODULE_DESCRIPTION("Virtio block driver");
1738 MODULE_LICENSE("GPL");
1739