xref: /linux/drivers/block/virtio_blk.c (revision 0d6ccfe6b319d56da63b7d7cfbcecd92780a680d)
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 	} while (!virtqueue_enable_cb(vq));
371 
372 	/* In case queue is stopped waiting for more buffers. */
373 	if (req_done)
374 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
375 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
376 }
377 
378 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
379 {
380 	struct virtio_blk *vblk = hctx->queue->queuedata;
381 	struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
382 	bool kick;
383 
384 	spin_lock_irq(&vq->lock);
385 	kick = virtqueue_kick_prepare(vq->vq);
386 	spin_unlock_irq(&vq->lock);
387 
388 	if (kick)
389 		virtqueue_notify(vq->vq);
390 }
391 
392 static blk_status_t virtblk_fail_to_queue(struct request *req, int rc)
393 {
394 	virtblk_cleanup_cmd(req);
395 	switch (rc) {
396 	case -ENOSPC:
397 		return BLK_STS_DEV_RESOURCE;
398 	case -ENOMEM:
399 		return BLK_STS_RESOURCE;
400 	default:
401 		return BLK_STS_IOERR;
402 	}
403 }
404 
405 static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
406 					struct virtio_blk *vblk,
407 					struct request *req,
408 					struct virtblk_req *vbr)
409 {
410 	blk_status_t status;
411 	int num;
412 
413 	status = virtblk_setup_cmd(vblk->vdev, req, vbr);
414 	if (unlikely(status))
415 		return status;
416 
417 	num = virtblk_map_data(hctx, req, vbr);
418 	if (unlikely(num < 0))
419 		return virtblk_fail_to_queue(req, -ENOMEM);
420 	vbr->sg_table.nents = num;
421 
422 	blk_mq_start_request(req);
423 
424 	return BLK_STS_OK;
425 }
426 
427 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
428 			   const struct blk_mq_queue_data *bd)
429 {
430 	struct virtio_blk *vblk = hctx->queue->queuedata;
431 	struct request *req = bd->rq;
432 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
433 	unsigned long flags;
434 	int qid = hctx->queue_num;
435 	bool notify = false;
436 	blk_status_t status;
437 	int err;
438 
439 	status = virtblk_prep_rq(hctx, vblk, req, vbr);
440 	if (unlikely(status))
441 		return status;
442 
443 	spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
444 	err = virtblk_add_req(vblk->vqs[qid].vq, vbr);
445 	if (err) {
446 		virtqueue_kick(vblk->vqs[qid].vq);
447 		/* Don't stop the queue if -ENOMEM: we may have failed to
448 		 * bounce the buffer due to global resource outage.
449 		 */
450 		if (err == -ENOSPC)
451 			blk_mq_stop_hw_queue(hctx);
452 		spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
453 		virtblk_unmap_data(req, vbr);
454 		return virtblk_fail_to_queue(req, err);
455 	}
456 
457 	if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
458 		notify = true;
459 	spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
460 
461 	if (notify)
462 		virtqueue_notify(vblk->vqs[qid].vq);
463 	return BLK_STS_OK;
464 }
465 
466 static bool virtblk_prep_rq_batch(struct request *req)
467 {
468 	struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
469 	struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
470 
471 	return virtblk_prep_rq(req->mq_hctx, vblk, req, vbr) == BLK_STS_OK;
472 }
473 
474 static bool virtblk_add_req_batch(struct virtio_blk_vq *vq,
475 					struct request **rqlist)
476 {
477 	unsigned long flags;
478 	int err;
479 	bool kick;
480 
481 	spin_lock_irqsave(&vq->lock, flags);
482 
483 	while (!rq_list_empty(*rqlist)) {
484 		struct request *req = rq_list_pop(rqlist);
485 		struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
486 
487 		err = virtblk_add_req(vq->vq, vbr);
488 		if (err) {
489 			virtblk_unmap_data(req, vbr);
490 			virtblk_cleanup_cmd(req);
491 			blk_mq_requeue_request(req, true);
492 		}
493 	}
494 
495 	kick = virtqueue_kick_prepare(vq->vq);
496 	spin_unlock_irqrestore(&vq->lock, flags);
497 
498 	return kick;
499 }
500 
501 static void virtio_queue_rqs(struct request **rqlist)
502 {
503 	struct request *req, *next, *prev = NULL;
504 	struct request *requeue_list = NULL;
505 
506 	rq_list_for_each_safe(rqlist, req, next) {
507 		struct virtio_blk_vq *vq = get_virtio_blk_vq(req->mq_hctx);
508 		bool kick;
509 
510 		if (!virtblk_prep_rq_batch(req)) {
511 			rq_list_move(rqlist, &requeue_list, req, prev);
512 			req = prev;
513 			if (!req)
514 				continue;
515 		}
516 
517 		if (!next || req->mq_hctx != next->mq_hctx) {
518 			req->rq_next = NULL;
519 			kick = virtblk_add_req_batch(vq, rqlist);
520 			if (kick)
521 				virtqueue_notify(vq->vq);
522 
523 			*rqlist = next;
524 			prev = NULL;
525 		} else
526 			prev = req;
527 	}
528 
529 	*rqlist = requeue_list;
530 }
531 
532 #ifdef CONFIG_BLK_DEV_ZONED
533 static void *virtblk_alloc_report_buffer(struct virtio_blk *vblk,
534 					  unsigned int nr_zones,
535 					  size_t *buflen)
536 {
537 	struct request_queue *q = vblk->disk->queue;
538 	size_t bufsize;
539 	void *buf;
540 
541 	nr_zones = min_t(unsigned int, nr_zones,
542 			 get_capacity(vblk->disk) >> ilog2(vblk->zone_sectors));
543 
544 	bufsize = sizeof(struct virtio_blk_zone_report) +
545 		nr_zones * sizeof(struct virtio_blk_zone_descriptor);
546 	bufsize = min_t(size_t, bufsize,
547 			queue_max_hw_sectors(q) << SECTOR_SHIFT);
548 	bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
549 
550 	while (bufsize >= sizeof(struct virtio_blk_zone_report)) {
551 		buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
552 		if (buf) {
553 			*buflen = bufsize;
554 			return buf;
555 		}
556 		bufsize >>= 1;
557 	}
558 
559 	return NULL;
560 }
561 
562 static int virtblk_submit_zone_report(struct virtio_blk *vblk,
563 				       char *report_buf, size_t report_len,
564 				       sector_t sector)
565 {
566 	struct request_queue *q = vblk->disk->queue;
567 	struct request *req;
568 	struct virtblk_req *vbr;
569 	int err;
570 
571 	req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
572 	if (IS_ERR(req))
573 		return PTR_ERR(req);
574 
575 	vbr = blk_mq_rq_to_pdu(req);
576 	vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
577 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_ZONE_REPORT);
578 	vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, sector);
579 
580 	err = blk_rq_map_kern(q, req, report_buf, report_len, GFP_KERNEL);
581 	if (err)
582 		goto out;
583 
584 	blk_execute_rq(req, false);
585 	err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
586 out:
587 	blk_mq_free_request(req);
588 	return err;
589 }
590 
591 static int virtblk_parse_zone(struct virtio_blk *vblk,
592 			       struct virtio_blk_zone_descriptor *entry,
593 			       unsigned int idx, report_zones_cb cb, void *data)
594 {
595 	struct blk_zone zone = { };
596 
597 	zone.start = virtio64_to_cpu(vblk->vdev, entry->z_start);
598 	if (zone.start + vblk->zone_sectors <= get_capacity(vblk->disk))
599 		zone.len = vblk->zone_sectors;
600 	else
601 		zone.len = get_capacity(vblk->disk) - zone.start;
602 	zone.capacity = virtio64_to_cpu(vblk->vdev, entry->z_cap);
603 	zone.wp = virtio64_to_cpu(vblk->vdev, entry->z_wp);
604 
605 	switch (entry->z_type) {
606 	case VIRTIO_BLK_ZT_SWR:
607 		zone.type = BLK_ZONE_TYPE_SEQWRITE_REQ;
608 		break;
609 	case VIRTIO_BLK_ZT_SWP:
610 		zone.type = BLK_ZONE_TYPE_SEQWRITE_PREF;
611 		break;
612 	case VIRTIO_BLK_ZT_CONV:
613 		zone.type = BLK_ZONE_TYPE_CONVENTIONAL;
614 		break;
615 	default:
616 		dev_err(&vblk->vdev->dev, "zone %llu: invalid type %#x\n",
617 			zone.start, entry->z_type);
618 		return -EIO;
619 	}
620 
621 	switch (entry->z_state) {
622 	case VIRTIO_BLK_ZS_EMPTY:
623 		zone.cond = BLK_ZONE_COND_EMPTY;
624 		break;
625 	case VIRTIO_BLK_ZS_CLOSED:
626 		zone.cond = BLK_ZONE_COND_CLOSED;
627 		break;
628 	case VIRTIO_BLK_ZS_FULL:
629 		zone.cond = BLK_ZONE_COND_FULL;
630 		zone.wp = zone.start + zone.len;
631 		break;
632 	case VIRTIO_BLK_ZS_EOPEN:
633 		zone.cond = BLK_ZONE_COND_EXP_OPEN;
634 		break;
635 	case VIRTIO_BLK_ZS_IOPEN:
636 		zone.cond = BLK_ZONE_COND_IMP_OPEN;
637 		break;
638 	case VIRTIO_BLK_ZS_NOT_WP:
639 		zone.cond = BLK_ZONE_COND_NOT_WP;
640 		break;
641 	case VIRTIO_BLK_ZS_RDONLY:
642 		zone.cond = BLK_ZONE_COND_READONLY;
643 		zone.wp = ULONG_MAX;
644 		break;
645 	case VIRTIO_BLK_ZS_OFFLINE:
646 		zone.cond = BLK_ZONE_COND_OFFLINE;
647 		zone.wp = ULONG_MAX;
648 		break;
649 	default:
650 		dev_err(&vblk->vdev->dev, "zone %llu: invalid condition %#x\n",
651 			zone.start, entry->z_state);
652 		return -EIO;
653 	}
654 
655 	/*
656 	 * The callback below checks the validity of the reported
657 	 * entry data, no need to further validate it here.
658 	 */
659 	return cb(&zone, idx, data);
660 }
661 
662 static int virtblk_report_zones(struct gendisk *disk, sector_t sector,
663 				 unsigned int nr_zones, report_zones_cb cb,
664 				 void *data)
665 {
666 	struct virtio_blk *vblk = disk->private_data;
667 	struct virtio_blk_zone_report *report;
668 	unsigned long long nz, i;
669 	size_t buflen;
670 	unsigned int zone_idx = 0;
671 	int ret;
672 
673 	if (WARN_ON_ONCE(!vblk->zone_sectors))
674 		return -EOPNOTSUPP;
675 
676 	report = virtblk_alloc_report_buffer(vblk, nr_zones, &buflen);
677 	if (!report)
678 		return -ENOMEM;
679 
680 	mutex_lock(&vblk->vdev_mutex);
681 
682 	if (!vblk->vdev) {
683 		ret = -ENXIO;
684 		goto fail_report;
685 	}
686 
687 	while (zone_idx < nr_zones && sector < get_capacity(vblk->disk)) {
688 		memset(report, 0, buflen);
689 
690 		ret = virtblk_submit_zone_report(vblk, (char *)report,
691 						 buflen, sector);
692 		if (ret)
693 			goto fail_report;
694 
695 		nz = min_t(u64, virtio64_to_cpu(vblk->vdev, report->nr_zones),
696 			   nr_zones);
697 		if (!nz)
698 			break;
699 
700 		for (i = 0; i < nz && zone_idx < nr_zones; i++) {
701 			ret = virtblk_parse_zone(vblk, &report->zones[i],
702 						 zone_idx, cb, data);
703 			if (ret)
704 				goto fail_report;
705 
706 			sector = virtio64_to_cpu(vblk->vdev,
707 						 report->zones[i].z_start) +
708 				 vblk->zone_sectors;
709 			zone_idx++;
710 		}
711 	}
712 
713 	if (zone_idx > 0)
714 		ret = zone_idx;
715 	else
716 		ret = -EINVAL;
717 fail_report:
718 	mutex_unlock(&vblk->vdev_mutex);
719 	kvfree(report);
720 	return ret;
721 }
722 
723 static int virtblk_read_zoned_limits(struct virtio_blk *vblk,
724 		struct queue_limits *lim)
725 {
726 	struct virtio_device *vdev = vblk->vdev;
727 	u32 v, wg;
728 
729 	dev_dbg(&vdev->dev, "probing host-managed zoned device\n");
730 
731 	lim->features |= BLK_FEAT_ZONED;
732 
733 	virtio_cread(vdev, struct virtio_blk_config,
734 		     zoned.max_open_zones, &v);
735 	lim->max_open_zones = v;
736 	dev_dbg(&vdev->dev, "max open zones = %u\n", v);
737 
738 	virtio_cread(vdev, struct virtio_blk_config,
739 		     zoned.max_active_zones, &v);
740 	lim->max_active_zones = v;
741 	dev_dbg(&vdev->dev, "max active zones = %u\n", v);
742 
743 	virtio_cread(vdev, struct virtio_blk_config,
744 		     zoned.write_granularity, &wg);
745 	if (!wg) {
746 		dev_warn(&vdev->dev, "zero write granularity reported\n");
747 		return -ENODEV;
748 	}
749 	lim->physical_block_size = wg;
750 	lim->io_min = wg;
751 
752 	dev_dbg(&vdev->dev, "write granularity = %u\n", wg);
753 
754 	/*
755 	 * virtio ZBD specification doesn't require zones to be a power of
756 	 * two sectors in size, but the code in this driver expects that.
757 	 */
758 	virtio_cread(vdev, struct virtio_blk_config, zoned.zone_sectors,
759 		     &vblk->zone_sectors);
760 	if (vblk->zone_sectors == 0 || !is_power_of_2(vblk->zone_sectors)) {
761 		dev_err(&vdev->dev,
762 			"zoned device with non power of two zone size %u\n",
763 			vblk->zone_sectors);
764 		return -ENODEV;
765 	}
766 	lim->chunk_sectors = vblk->zone_sectors;
767 	dev_dbg(&vdev->dev, "zone sectors = %u\n", vblk->zone_sectors);
768 
769 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
770 		dev_warn(&vblk->vdev->dev,
771 			 "ignoring negotiated F_DISCARD for zoned device\n");
772 		lim->max_hw_discard_sectors = 0;
773 	}
774 
775 	virtio_cread(vdev, struct virtio_blk_config,
776 		     zoned.max_append_sectors, &v);
777 	if (!v) {
778 		dev_warn(&vdev->dev, "zero max_append_sectors reported\n");
779 		return -ENODEV;
780 	}
781 	if ((v << SECTOR_SHIFT) < wg) {
782 		dev_err(&vdev->dev,
783 			"write granularity %u exceeds max_append_sectors %u limit\n",
784 			wg, v);
785 		return -ENODEV;
786 	}
787 	lim->max_zone_append_sectors = v;
788 	dev_dbg(&vdev->dev, "max append sectors = %u\n", v);
789 
790 	return 0;
791 }
792 #else
793 /*
794  * Zoned block device support is not configured in this kernel, host-managed
795  * zoned devices can't be supported.
796  */
797 #define virtblk_report_zones       NULL
798 static inline int virtblk_read_zoned_limits(struct virtio_blk *vblk,
799 		struct queue_limits *lim)
800 {
801 	dev_err(&vblk->vdev->dev,
802 		"virtio_blk: zoned devices are not supported");
803 	return -EOPNOTSUPP;
804 }
805 #endif /* CONFIG_BLK_DEV_ZONED */
806 
807 /* return id (s/n) string for *disk to *id_str
808  */
809 static int virtblk_get_id(struct gendisk *disk, char *id_str)
810 {
811 	struct virtio_blk *vblk = disk->private_data;
812 	struct request_queue *q = vblk->disk->queue;
813 	struct request *req;
814 	struct virtblk_req *vbr;
815 	int err;
816 
817 	req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
818 	if (IS_ERR(req))
819 		return PTR_ERR(req);
820 
821 	vbr = blk_mq_rq_to_pdu(req);
822 	vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
823 	vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
824 	vbr->out_hdr.sector = 0;
825 
826 	err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
827 	if (err)
828 		goto out;
829 
830 	blk_execute_rq(req, false);
831 	err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
832 out:
833 	blk_mq_free_request(req);
834 	return err;
835 }
836 
837 /* We provide getgeo only to please some old bootloader/partitioning tools */
838 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
839 {
840 	struct virtio_blk *vblk = bd->bd_disk->private_data;
841 	int ret = 0;
842 
843 	mutex_lock(&vblk->vdev_mutex);
844 
845 	if (!vblk->vdev) {
846 		ret = -ENXIO;
847 		goto out;
848 	}
849 
850 	/* see if the host passed in geometry config */
851 	if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
852 		virtio_cread(vblk->vdev, struct virtio_blk_config,
853 			     geometry.cylinders, &geo->cylinders);
854 		virtio_cread(vblk->vdev, struct virtio_blk_config,
855 			     geometry.heads, &geo->heads);
856 		virtio_cread(vblk->vdev, struct virtio_blk_config,
857 			     geometry.sectors, &geo->sectors);
858 	} else {
859 		/* some standard values, similar to sd */
860 		geo->heads = 1 << 6;
861 		geo->sectors = 1 << 5;
862 		geo->cylinders = get_capacity(bd->bd_disk) >> 11;
863 	}
864 out:
865 	mutex_unlock(&vblk->vdev_mutex);
866 	return ret;
867 }
868 
869 static void virtblk_free_disk(struct gendisk *disk)
870 {
871 	struct virtio_blk *vblk = disk->private_data;
872 
873 	ida_free(&vd_index_ida, vblk->index);
874 	mutex_destroy(&vblk->vdev_mutex);
875 	kfree(vblk);
876 }
877 
878 static const struct block_device_operations virtblk_fops = {
879 	.owner  	= THIS_MODULE,
880 	.getgeo		= virtblk_getgeo,
881 	.free_disk	= virtblk_free_disk,
882 	.report_zones	= virtblk_report_zones,
883 };
884 
885 static int index_to_minor(int index)
886 {
887 	return index << PART_BITS;
888 }
889 
890 static int minor_to_index(int minor)
891 {
892 	return minor >> PART_BITS;
893 }
894 
895 static ssize_t serial_show(struct device *dev,
896 			   struct device_attribute *attr, char *buf)
897 {
898 	struct gendisk *disk = dev_to_disk(dev);
899 	int err;
900 
901 	/* sysfs gives us a PAGE_SIZE buffer */
902 	BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
903 
904 	buf[VIRTIO_BLK_ID_BYTES] = '\0';
905 	err = virtblk_get_id(disk, buf);
906 	if (!err)
907 		return strlen(buf);
908 
909 	if (err == -EIO) /* Unsupported? Make it empty. */
910 		return 0;
911 
912 	return err;
913 }
914 
915 static DEVICE_ATTR_RO(serial);
916 
917 /* The queue's logical block size must be set before calling this */
918 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
919 {
920 	struct virtio_device *vdev = vblk->vdev;
921 	struct request_queue *q = vblk->disk->queue;
922 	char cap_str_2[10], cap_str_10[10];
923 	unsigned long long nblocks;
924 	u64 capacity;
925 
926 	/* Host must always specify the capacity. */
927 	virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
928 
929 	nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
930 
931 	string_get_size(nblocks, queue_logical_block_size(q),
932 			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
933 	string_get_size(nblocks, queue_logical_block_size(q),
934 			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
935 
936 	dev_notice(&vdev->dev,
937 		   "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
938 		   vblk->disk->disk_name,
939 		   resize ? "new size: " : "",
940 		   nblocks,
941 		   queue_logical_block_size(q),
942 		   cap_str_10,
943 		   cap_str_2);
944 
945 	set_capacity_and_notify(vblk->disk, capacity);
946 }
947 
948 static void virtblk_config_changed_work(struct work_struct *work)
949 {
950 	struct virtio_blk *vblk =
951 		container_of(work, struct virtio_blk, config_work);
952 
953 	virtblk_update_capacity(vblk, true);
954 }
955 
956 static void virtblk_config_changed(struct virtio_device *vdev)
957 {
958 	struct virtio_blk *vblk = vdev->priv;
959 
960 	queue_work(virtblk_wq, &vblk->config_work);
961 }
962 
963 static int init_vq(struct virtio_blk *vblk)
964 {
965 	int err;
966 	unsigned short i;
967 	struct virtqueue_info *vqs_info;
968 	struct virtqueue **vqs;
969 	unsigned short num_vqs;
970 	unsigned short num_poll_vqs;
971 	struct virtio_device *vdev = vblk->vdev;
972 	struct irq_affinity desc = { 0, };
973 
974 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
975 				   struct virtio_blk_config, num_queues,
976 				   &num_vqs);
977 	if (err)
978 		num_vqs = 1;
979 
980 	if (!err && !num_vqs) {
981 		dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
982 		return -EINVAL;
983 	}
984 
985 	num_vqs = min_t(unsigned int,
986 			min_not_zero(num_request_queues, nr_cpu_ids),
987 			num_vqs);
988 
989 	num_poll_vqs = min_t(unsigned int, poll_queues, num_vqs - 1);
990 
991 	vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
992 	vblk->io_queues[HCTX_TYPE_READ] = 0;
993 	vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
994 
995 	dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n",
996 				vblk->io_queues[HCTX_TYPE_DEFAULT],
997 				vblk->io_queues[HCTX_TYPE_READ],
998 				vblk->io_queues[HCTX_TYPE_POLL]);
999 
1000 	vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
1001 	if (!vblk->vqs)
1002 		return -ENOMEM;
1003 
1004 	vqs_info = kcalloc(num_vqs, sizeof(*vqs_info), GFP_KERNEL);
1005 	vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
1006 	if (!vqs_info || !vqs) {
1007 		err = -ENOMEM;
1008 		goto out;
1009 	}
1010 
1011 	for (i = 0; i < num_vqs - num_poll_vqs; i++) {
1012 		vqs_info[i].callback = virtblk_done;
1013 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%u", i);
1014 		vqs_info[i].name = vblk->vqs[i].name;
1015 	}
1016 
1017 	for (; i < num_vqs; i++) {
1018 		snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%u", i);
1019 		vqs_info[i].name = vblk->vqs[i].name;
1020 	}
1021 
1022 	/* Discover virtqueues and write information to configuration.  */
1023 	err = virtio_find_vqs(vdev, num_vqs, vqs, vqs_info, &desc);
1024 	if (err)
1025 		goto out;
1026 
1027 	for (i = 0; i < num_vqs; i++) {
1028 		spin_lock_init(&vblk->vqs[i].lock);
1029 		vblk->vqs[i].vq = vqs[i];
1030 	}
1031 	vblk->num_vqs = num_vqs;
1032 
1033 out:
1034 	kfree(vqs);
1035 	kfree(vqs_info);
1036 	if (err)
1037 		kfree(vblk->vqs);
1038 	return err;
1039 }
1040 
1041 /*
1042  * Legacy naming scheme used for virtio devices.  We are stuck with it for
1043  * virtio blk but don't ever use it for any new driver.
1044  */
1045 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
1046 {
1047 	const int base = 'z' - 'a' + 1;
1048 	char *begin = buf + strlen(prefix);
1049 	char *end = buf + buflen;
1050 	char *p;
1051 	int unit;
1052 
1053 	p = end - 1;
1054 	*p = '\0';
1055 	unit = base;
1056 	do {
1057 		if (p == begin)
1058 			return -EINVAL;
1059 		*--p = 'a' + (index % unit);
1060 		index = (index / unit) - 1;
1061 	} while (index >= 0);
1062 
1063 	memmove(begin, p, end - p);
1064 	memcpy(buf, prefix, strlen(prefix));
1065 
1066 	return 0;
1067 }
1068 
1069 static int virtblk_get_cache_mode(struct virtio_device *vdev)
1070 {
1071 	u8 writeback;
1072 	int err;
1073 
1074 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
1075 				   struct virtio_blk_config, wce,
1076 				   &writeback);
1077 
1078 	/*
1079 	 * If WCE is not configurable and flush is not available,
1080 	 * assume no writeback cache is in use.
1081 	 */
1082 	if (err)
1083 		writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
1084 
1085 	return writeback;
1086 }
1087 
1088 static const char *const virtblk_cache_types[] = {
1089 	"write through", "write back"
1090 };
1091 
1092 static ssize_t
1093 cache_type_store(struct device *dev, struct device_attribute *attr,
1094 		 const char *buf, size_t count)
1095 {
1096 	struct gendisk *disk = dev_to_disk(dev);
1097 	struct virtio_blk *vblk = disk->private_data;
1098 	struct virtio_device *vdev = vblk->vdev;
1099 	struct queue_limits lim;
1100 	int i;
1101 
1102 	BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
1103 	i = sysfs_match_string(virtblk_cache_types, buf);
1104 	if (i < 0)
1105 		return i;
1106 
1107 	virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
1108 
1109 	lim = queue_limits_start_update(disk->queue);
1110 	if (virtblk_get_cache_mode(vdev))
1111 		lim.features |= BLK_FEAT_WRITE_CACHE;
1112 	else
1113 		lim.features &= ~BLK_FEAT_WRITE_CACHE;
1114 	blk_mq_freeze_queue(disk->queue);
1115 	i = queue_limits_commit_update(disk->queue, &lim);
1116 	blk_mq_unfreeze_queue(disk->queue);
1117 	if (i)
1118 		return i;
1119 	return count;
1120 }
1121 
1122 static ssize_t
1123 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
1124 {
1125 	struct gendisk *disk = dev_to_disk(dev);
1126 	struct virtio_blk *vblk = disk->private_data;
1127 	u8 writeback = virtblk_get_cache_mode(vblk->vdev);
1128 
1129 	BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
1130 	return sysfs_emit(buf, "%s\n", virtblk_cache_types[writeback]);
1131 }
1132 
1133 static DEVICE_ATTR_RW(cache_type);
1134 
1135 static struct attribute *virtblk_attrs[] = {
1136 	&dev_attr_serial.attr,
1137 	&dev_attr_cache_type.attr,
1138 	NULL,
1139 };
1140 
1141 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
1142 		struct attribute *a, int n)
1143 {
1144 	struct device *dev = kobj_to_dev(kobj);
1145 	struct gendisk *disk = dev_to_disk(dev);
1146 	struct virtio_blk *vblk = disk->private_data;
1147 	struct virtio_device *vdev = vblk->vdev;
1148 
1149 	if (a == &dev_attr_cache_type.attr &&
1150 	    !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
1151 		return S_IRUGO;
1152 
1153 	return a->mode;
1154 }
1155 
1156 static const struct attribute_group virtblk_attr_group = {
1157 	.attrs = virtblk_attrs,
1158 	.is_visible = virtblk_attrs_are_visible,
1159 };
1160 
1161 static const struct attribute_group *virtblk_attr_groups[] = {
1162 	&virtblk_attr_group,
1163 	NULL,
1164 };
1165 
1166 static void virtblk_map_queues(struct blk_mq_tag_set *set)
1167 {
1168 	struct virtio_blk *vblk = set->driver_data;
1169 	int i, qoff;
1170 
1171 	for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1172 		struct blk_mq_queue_map *map = &set->map[i];
1173 
1174 		map->nr_queues = vblk->io_queues[i];
1175 		map->queue_offset = qoff;
1176 		qoff += map->nr_queues;
1177 
1178 		if (map->nr_queues == 0)
1179 			continue;
1180 
1181 		/*
1182 		 * Regular queues have interrupts and hence CPU affinity is
1183 		 * defined by the core virtio code, but polling queues have
1184 		 * no interrupts so we let the block layer assign CPU affinity.
1185 		 */
1186 		if (i == HCTX_TYPE_POLL)
1187 			blk_mq_map_queues(&set->map[i]);
1188 		else
1189 			blk_mq_virtio_map_queues(&set->map[i], vblk->vdev, 0);
1190 	}
1191 }
1192 
1193 static void virtblk_complete_batch(struct io_comp_batch *iob)
1194 {
1195 	struct request *req;
1196 
1197 	rq_list_for_each(&iob->req_list, req) {
1198 		virtblk_unmap_data(req, blk_mq_rq_to_pdu(req));
1199 		virtblk_cleanup_cmd(req);
1200 	}
1201 	blk_mq_end_request_batch(iob);
1202 }
1203 
1204 static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1205 {
1206 	struct virtio_blk *vblk = hctx->queue->queuedata;
1207 	struct virtio_blk_vq *vq = get_virtio_blk_vq(hctx);
1208 	struct virtblk_req *vbr;
1209 	unsigned long flags;
1210 	unsigned int len;
1211 	int found = 0;
1212 
1213 	spin_lock_irqsave(&vq->lock, flags);
1214 
1215 	while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
1216 		struct request *req = blk_mq_rq_from_pdu(vbr);
1217 
1218 		found++;
1219 		if (!blk_mq_complete_request_remote(req) &&
1220 		    !blk_mq_add_to_batch(req, iob, virtblk_vbr_status(vbr),
1221 						virtblk_complete_batch))
1222 			virtblk_request_done(req);
1223 	}
1224 
1225 	if (found)
1226 		blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
1227 
1228 	spin_unlock_irqrestore(&vq->lock, flags);
1229 
1230 	return found;
1231 }
1232 
1233 static const struct blk_mq_ops virtio_mq_ops = {
1234 	.queue_rq	= virtio_queue_rq,
1235 	.queue_rqs	= virtio_queue_rqs,
1236 	.commit_rqs	= virtio_commit_rqs,
1237 	.complete	= virtblk_request_done,
1238 	.map_queues	= virtblk_map_queues,
1239 	.poll		= virtblk_poll,
1240 };
1241 
1242 static unsigned int virtblk_queue_depth;
1243 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
1244 
1245 static int virtblk_read_limits(struct virtio_blk *vblk,
1246 		struct queue_limits *lim)
1247 {
1248 	struct virtio_device *vdev = vblk->vdev;
1249 	u32 v, max_size, sg_elems, opt_io_size;
1250 	u32 max_discard_segs = 0;
1251 	u32 discard_granularity = 0;
1252 	u16 min_io_size;
1253 	u8 physical_block_exp, alignment_offset;
1254 	size_t max_dma_size;
1255 	int err;
1256 
1257 	/* We need to know how many segments before we allocate. */
1258 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
1259 				   struct virtio_blk_config, seg_max,
1260 				   &sg_elems);
1261 
1262 	/* We need at least one SG element, whatever they say. */
1263 	if (err || !sg_elems)
1264 		sg_elems = 1;
1265 
1266 	/* Prevent integer overflows and honor max vq size */
1267 	sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
1268 
1269 	/* We can handle whatever the host told us to handle. */
1270 	lim->max_segments = sg_elems;
1271 
1272 	/* No real sector limit. */
1273 	lim->max_hw_sectors = UINT_MAX;
1274 
1275 	max_dma_size = virtio_max_dma_size(vdev);
1276 	max_size = max_dma_size > U32_MAX ? U32_MAX : max_dma_size;
1277 
1278 	/* Host can optionally specify maximum segment size and number of
1279 	 * segments. */
1280 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
1281 				   struct virtio_blk_config, size_max, &v);
1282 	if (!err)
1283 		max_size = min(max_size, v);
1284 
1285 	lim->max_segment_size = max_size;
1286 
1287 	/* Host can optionally specify the block size of the device */
1288 	virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
1289 				   struct virtio_blk_config, blk_size,
1290 				   &lim->logical_block_size);
1291 
1292 	/* Use topology information if available */
1293 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1294 				   struct virtio_blk_config, physical_block_exp,
1295 				   &physical_block_exp);
1296 	if (!err && physical_block_exp)
1297 		lim->physical_block_size =
1298 			lim->logical_block_size * (1 << physical_block_exp);
1299 
1300 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1301 				   struct virtio_blk_config, alignment_offset,
1302 				   &alignment_offset);
1303 	if (!err && alignment_offset)
1304 		lim->alignment_offset =
1305 			lim->logical_block_size * alignment_offset;
1306 
1307 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1308 				   struct virtio_blk_config, min_io_size,
1309 				   &min_io_size);
1310 	if (!err && min_io_size)
1311 		lim->io_min = lim->logical_block_size * min_io_size;
1312 
1313 	err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1314 				   struct virtio_blk_config, opt_io_size,
1315 				   &opt_io_size);
1316 	if (!err && opt_io_size)
1317 		lim->io_opt = lim->logical_block_size * opt_io_size;
1318 
1319 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
1320 		virtio_cread(vdev, struct virtio_blk_config,
1321 			     discard_sector_alignment, &discard_granularity);
1322 
1323 		virtio_cread(vdev, struct virtio_blk_config,
1324 			     max_discard_sectors, &v);
1325 		lim->max_hw_discard_sectors = v ? v : UINT_MAX;
1326 
1327 		virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
1328 			     &max_discard_segs);
1329 	}
1330 
1331 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
1332 		virtio_cread(vdev, struct virtio_blk_config,
1333 			     max_write_zeroes_sectors, &v);
1334 		lim->max_write_zeroes_sectors = v ? v : UINT_MAX;
1335 	}
1336 
1337 	/* The discard and secure erase limits are combined since the Linux
1338 	 * block layer uses the same limit for both commands.
1339 	 *
1340 	 * If both VIRTIO_BLK_F_SECURE_ERASE and VIRTIO_BLK_F_DISCARD features
1341 	 * are negotiated, we will use the minimum between the limits.
1342 	 *
1343 	 * discard sector alignment is set to the minimum between discard_sector_alignment
1344 	 * and secure_erase_sector_alignment.
1345 	 *
1346 	 * max discard sectors is set to the minimum between max_discard_seg and
1347 	 * max_secure_erase_seg.
1348 	 */
1349 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1350 
1351 		virtio_cread(vdev, struct virtio_blk_config,
1352 			     secure_erase_sector_alignment, &v);
1353 
1354 		/* secure_erase_sector_alignment should not be zero, the device should set a
1355 		 * valid number of sectors.
1356 		 */
1357 		if (!v) {
1358 			dev_err(&vdev->dev,
1359 				"virtio_blk: secure_erase_sector_alignment can't be 0\n");
1360 			return -EINVAL;
1361 		}
1362 
1363 		discard_granularity = min_not_zero(discard_granularity, v);
1364 
1365 		virtio_cread(vdev, struct virtio_blk_config,
1366 			     max_secure_erase_sectors, &v);
1367 
1368 		/* max_secure_erase_sectors should not be zero, the device should set a
1369 		 * valid number of sectors.
1370 		 */
1371 		if (!v) {
1372 			dev_err(&vdev->dev,
1373 				"virtio_blk: max_secure_erase_sectors can't be 0\n");
1374 			return -EINVAL;
1375 		}
1376 
1377 		lim->max_secure_erase_sectors = v;
1378 
1379 		virtio_cread(vdev, struct virtio_blk_config,
1380 			     max_secure_erase_seg, &v);
1381 
1382 		/* max_secure_erase_seg should not be zero, the device should set a
1383 		 * valid number of segments
1384 		 */
1385 		if (!v) {
1386 			dev_err(&vdev->dev,
1387 				"virtio_blk: max_secure_erase_seg can't be 0\n");
1388 			return -EINVAL;
1389 		}
1390 
1391 		max_discard_segs = min_not_zero(max_discard_segs, v);
1392 	}
1393 
1394 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD) ||
1395 	    virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1396 		/* max_discard_seg and discard_granularity will be 0 only
1397 		 * if max_discard_seg and discard_sector_alignment fields in the virtio
1398 		 * config are 0 and VIRTIO_BLK_F_SECURE_ERASE feature is not negotiated.
1399 		 * In this case, we use default values.
1400 		 */
1401 		if (!max_discard_segs)
1402 			max_discard_segs = sg_elems;
1403 
1404 		lim->max_discard_segments =
1405 			min(max_discard_segs, MAX_DISCARD_SEGMENTS);
1406 
1407 		if (discard_granularity)
1408 			lim->discard_granularity =
1409 				discard_granularity << SECTOR_SHIFT;
1410 		else
1411 			lim->discard_granularity = lim->logical_block_size;
1412 	}
1413 
1414 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_ZONED)) {
1415 		u8 model;
1416 
1417 		virtio_cread(vdev, struct virtio_blk_config, zoned.model, &model);
1418 		switch (model) {
1419 		case VIRTIO_BLK_Z_NONE:
1420 		case VIRTIO_BLK_Z_HA:
1421 			/* treat host-aware devices as non-zoned */
1422 			return 0;
1423 		case VIRTIO_BLK_Z_HM:
1424 			err = virtblk_read_zoned_limits(vblk, lim);
1425 			if (err)
1426 				return err;
1427 			break;
1428 		default:
1429 			dev_err(&vdev->dev, "unsupported zone model %d\n", model);
1430 			return -EINVAL;
1431 		}
1432 	}
1433 
1434 	return 0;
1435 }
1436 
1437 static int virtblk_probe(struct virtio_device *vdev)
1438 {
1439 	struct virtio_blk *vblk;
1440 	struct queue_limits lim = {
1441 		.features		= BLK_FEAT_ROTATIONAL,
1442 		.logical_block_size	= SECTOR_SIZE,
1443 	};
1444 	int err, index;
1445 	unsigned int queue_depth;
1446 
1447 	if (!vdev->config->get) {
1448 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
1449 			__func__);
1450 		return -EINVAL;
1451 	}
1452 
1453 	err = ida_alloc_range(&vd_index_ida, 0,
1454 			      minor_to_index(1 << MINORBITS) - 1, GFP_KERNEL);
1455 	if (err < 0)
1456 		goto out;
1457 	index = err;
1458 
1459 	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
1460 	if (!vblk) {
1461 		err = -ENOMEM;
1462 		goto out_free_index;
1463 	}
1464 
1465 	mutex_init(&vblk->vdev_mutex);
1466 
1467 	vblk->vdev = vdev;
1468 
1469 	INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
1470 
1471 	err = init_vq(vblk);
1472 	if (err)
1473 		goto out_free_vblk;
1474 
1475 	/* Default queue sizing is to fill the ring. */
1476 	if (!virtblk_queue_depth) {
1477 		queue_depth = vblk->vqs[0].vq->num_free;
1478 		/* ... but without indirect descs, we use 2 descs per req */
1479 		if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
1480 			queue_depth /= 2;
1481 	} else {
1482 		queue_depth = virtblk_queue_depth;
1483 	}
1484 
1485 	memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
1486 	vblk->tag_set.ops = &virtio_mq_ops;
1487 	vblk->tag_set.queue_depth = queue_depth;
1488 	vblk->tag_set.numa_node = NUMA_NO_NODE;
1489 	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1490 	vblk->tag_set.cmd_size =
1491 		sizeof(struct virtblk_req) +
1492 		sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
1493 	vblk->tag_set.driver_data = vblk;
1494 	vblk->tag_set.nr_hw_queues = vblk->num_vqs;
1495 	vblk->tag_set.nr_maps = 1;
1496 	if (vblk->io_queues[HCTX_TYPE_POLL])
1497 		vblk->tag_set.nr_maps = 3;
1498 
1499 	err = blk_mq_alloc_tag_set(&vblk->tag_set);
1500 	if (err)
1501 		goto out_free_vq;
1502 
1503 	err = virtblk_read_limits(vblk, &lim);
1504 	if (err)
1505 		goto out_free_tags;
1506 
1507 	if (virtblk_get_cache_mode(vdev))
1508 		lim.features |= BLK_FEAT_WRITE_CACHE;
1509 
1510 	vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, &lim, vblk);
1511 	if (IS_ERR(vblk->disk)) {
1512 		err = PTR_ERR(vblk->disk);
1513 		goto out_free_tags;
1514 	}
1515 
1516 	virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
1517 
1518 	vblk->disk->major = major;
1519 	vblk->disk->first_minor = index_to_minor(index);
1520 	vblk->disk->minors = 1 << PART_BITS;
1521 	vblk->disk->private_data = vblk;
1522 	vblk->disk->fops = &virtblk_fops;
1523 	vblk->index = index;
1524 
1525 	/* If disk is read-only in the host, the guest should obey */
1526 	if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
1527 		set_disk_ro(vblk->disk, 1);
1528 
1529 	virtblk_update_capacity(vblk, false);
1530 	virtio_device_ready(vdev);
1531 
1532 	/*
1533 	 * All steps that follow use the VQs therefore they need to be
1534 	 * placed after the virtio_device_ready() call above.
1535 	 */
1536 	if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
1537 	    (lim.features & BLK_FEAT_ZONED)) {
1538 		err = blk_revalidate_disk_zones(vblk->disk);
1539 		if (err)
1540 			goto out_cleanup_disk;
1541 	}
1542 
1543 	err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
1544 	if (err)
1545 		goto out_cleanup_disk;
1546 
1547 	return 0;
1548 
1549 out_cleanup_disk:
1550 	put_disk(vblk->disk);
1551 out_free_tags:
1552 	blk_mq_free_tag_set(&vblk->tag_set);
1553 out_free_vq:
1554 	vdev->config->del_vqs(vdev);
1555 	kfree(vblk->vqs);
1556 out_free_vblk:
1557 	kfree(vblk);
1558 out_free_index:
1559 	ida_free(&vd_index_ida, index);
1560 out:
1561 	return err;
1562 }
1563 
1564 static void virtblk_remove(struct virtio_device *vdev)
1565 {
1566 	struct virtio_blk *vblk = vdev->priv;
1567 
1568 	/* Make sure no work handler is accessing the device. */
1569 	flush_work(&vblk->config_work);
1570 
1571 	del_gendisk(vblk->disk);
1572 	blk_mq_free_tag_set(&vblk->tag_set);
1573 
1574 	mutex_lock(&vblk->vdev_mutex);
1575 
1576 	/* Stop all the virtqueues. */
1577 	virtio_reset_device(vdev);
1578 
1579 	/* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
1580 	vblk->vdev = NULL;
1581 
1582 	vdev->config->del_vqs(vdev);
1583 	kfree(vblk->vqs);
1584 
1585 	mutex_unlock(&vblk->vdev_mutex);
1586 
1587 	put_disk(vblk->disk);
1588 }
1589 
1590 #ifdef CONFIG_PM_SLEEP
1591 static int virtblk_freeze(struct virtio_device *vdev)
1592 {
1593 	struct virtio_blk *vblk = vdev->priv;
1594 
1595 	/* Ensure no requests in virtqueues before deleting vqs. */
1596 	blk_mq_freeze_queue(vblk->disk->queue);
1597 
1598 	/* Ensure we don't receive any more interrupts */
1599 	virtio_reset_device(vdev);
1600 
1601 	/* Make sure no work handler is accessing the device. */
1602 	flush_work(&vblk->config_work);
1603 
1604 	vdev->config->del_vqs(vdev);
1605 	kfree(vblk->vqs);
1606 
1607 	return 0;
1608 }
1609 
1610 static int virtblk_restore(struct virtio_device *vdev)
1611 {
1612 	struct virtio_blk *vblk = vdev->priv;
1613 	int ret;
1614 
1615 	ret = init_vq(vdev->priv);
1616 	if (ret)
1617 		return ret;
1618 
1619 	virtio_device_ready(vdev);
1620 
1621 	blk_mq_unfreeze_queue(vblk->disk->queue);
1622 	return 0;
1623 }
1624 #endif
1625 
1626 static const struct virtio_device_id id_table[] = {
1627 	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1628 	{ 0 },
1629 };
1630 
1631 static unsigned int features_legacy[] = {
1632 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1633 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1634 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1635 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1636 	VIRTIO_BLK_F_SECURE_ERASE,
1637 }
1638 ;
1639 static unsigned int features[] = {
1640 	VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1641 	VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1642 	VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1643 	VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1644 	VIRTIO_BLK_F_SECURE_ERASE, VIRTIO_BLK_F_ZONED,
1645 };
1646 
1647 static struct virtio_driver virtio_blk = {
1648 	.feature_table			= features,
1649 	.feature_table_size		= ARRAY_SIZE(features),
1650 	.feature_table_legacy		= features_legacy,
1651 	.feature_table_size_legacy	= ARRAY_SIZE(features_legacy),
1652 	.driver.name			= KBUILD_MODNAME,
1653 	.id_table			= id_table,
1654 	.probe				= virtblk_probe,
1655 	.remove				= virtblk_remove,
1656 	.config_changed			= virtblk_config_changed,
1657 #ifdef CONFIG_PM_SLEEP
1658 	.freeze				= virtblk_freeze,
1659 	.restore			= virtblk_restore,
1660 #endif
1661 };
1662 
1663 static int __init virtio_blk_init(void)
1664 {
1665 	int error;
1666 
1667 	virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1668 	if (!virtblk_wq)
1669 		return -ENOMEM;
1670 
1671 	major = register_blkdev(0, "virtblk");
1672 	if (major < 0) {
1673 		error = major;
1674 		goto out_destroy_workqueue;
1675 	}
1676 
1677 	error = register_virtio_driver(&virtio_blk);
1678 	if (error)
1679 		goto out_unregister_blkdev;
1680 	return 0;
1681 
1682 out_unregister_blkdev:
1683 	unregister_blkdev(major, "virtblk");
1684 out_destroy_workqueue:
1685 	destroy_workqueue(virtblk_wq);
1686 	return error;
1687 }
1688 
1689 static void __exit virtio_blk_fini(void)
1690 {
1691 	unregister_virtio_driver(&virtio_blk);
1692 	unregister_blkdev(major, "virtblk");
1693 	destroy_workqueue(virtblk_wq);
1694 }
1695 module_init(virtio_blk_init);
1696 module_exit(virtio_blk_fini);
1697 
1698 MODULE_DEVICE_TABLE(virtio, id_table);
1699 MODULE_DESCRIPTION("Virtio block driver");
1700 MODULE_LICENSE("GPL");
1701