1// SPDX-License-Identifier: GPL-2.0-only
2//#define DEBUG
3#include <linux/spinlock.h>
4#include <linux/slab.h>
5#include <linux/blkdev.h>
6#include <linux/hdreg.h>
7#include <linux/module.h>
8#include <linux/mutex.h>
9#include <linux/interrupt.h>
10#include <linux/virtio.h>
11#include <linux/virtio_blk.h>
12#include <linux/scatterlist.h>
13#include <linux/string_helpers.h>
14#include <linux/idr.h>
15#include <linux/blk-mq.h>
16#include <linux/numa.h>
17#include <linux/vmalloc.h>
18#include <uapi/linux/virtio_ring.h>
19
20#define PART_BITS 4
21#define VQ_NAME_LEN 16
22#define MAX_DISCARD_SEGMENTS 256u
23
24/* The maximum number of sg elements that fit into a virtqueue */
25#define VIRTIO_BLK_MAX_SG_ELEMS 32768
26
27#ifdef CONFIG_ARCH_NO_SG_CHAIN
28#define VIRTIO_BLK_INLINE_SG_CNT 0
29#else
30#define VIRTIO_BLK_INLINE_SG_CNT 2
31#endif
32
33static unsigned int num_request_queues;
34module_param(num_request_queues, uint, 0644);
35MODULE_PARM_DESC(num_request_queues,
36 "Limit the number of request queues to use for blk device. "
37 "0 for no limit. "
38 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
39
40static unsigned int poll_queues;
41module_param(poll_queues, uint, 0644);
42MODULE_PARM_DESC(poll_queues, "The number of dedicated virtqueues for polling I/O");
43
44static int major;
45static DEFINE_IDA(vd_index_ida);
46
47static struct workqueue_struct *virtblk_wq;
48
49struct virtio_blk_vq {
50 struct virtqueue *vq;
51 spinlock_t lock;
52 char name[VQ_NAME_LEN];
53} ____cacheline_aligned_in_smp;
54
55struct virtio_blk {
56 /*
57 * This mutex must be held by anything that may run after
58 * virtblk_remove() sets vblk->vdev to NULL.
59 *
60 * blk-mq, virtqueue processing, and sysfs attribute code paths are
61 * shut down before vblk->vdev is set to NULL and therefore do not need
62 * to hold this mutex.
63 */
64 struct mutex vdev_mutex;
65 struct virtio_device *vdev;
66
67 /* The disk structure for the kernel. */
68 struct gendisk *disk;
69
70 /* Block layer tags. */
71 struct blk_mq_tag_set tag_set;
72
73 /* Process context for config space updates */
74 struct work_struct config_work;
75
76 /* Ida index - used to track minor number allocations. */
77 int index;
78
79 /* num of vqs */
80 int num_vqs;
81 int io_queues[HCTX_MAX_TYPES];
82 struct virtio_blk_vq *vqs;
83
84 /* For zoned device */
85 unsigned int zone_sectors;
86};
87
88struct virtblk_req {
89 /* Out header */
90 struct virtio_blk_outhdr out_hdr;
91
92 /* In header */
93 union {
94 u8 status;
95
96 /*
97 * The zone append command has an extended in header.
98 * The status field in zone_append_in_hdr must always
99 * be the last byte.
100 */
101 struct {
102 __virtio64 sector;
103 u8 status;
104 } zone_append;
105 } in_hdr;
106
107 size_t in_hdr_len;
108
109 struct sg_table sg_table;
110 struct scatterlist sg[];
111};
112
113static inline blk_status_t virtblk_result(u8 status)
114{
115 switch (status) {
116 case VIRTIO_BLK_S_OK:
117 return BLK_STS_OK;
118 case VIRTIO_BLK_S_UNSUPP:
119 return BLK_STS_NOTSUPP;
120 case VIRTIO_BLK_S_ZONE_OPEN_RESOURCE:
121 return BLK_STS_ZONE_OPEN_RESOURCE;
122 case VIRTIO_BLK_S_ZONE_ACTIVE_RESOURCE:
123 return BLK_STS_ZONE_ACTIVE_RESOURCE;
124 case VIRTIO_BLK_S_IOERR:
125 case VIRTIO_BLK_S_ZONE_UNALIGNED_WP:
126 default:
127 return BLK_STS_IOERR;
128 }
129}
130
131static inline struct virtio_blk_vq *get_virtio_blk_vq(struct blk_mq_hw_ctx *hctx)
132{
133 struct virtio_blk *vblk = hctx->queue->queuedata;
134 struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
135
136 return vq;
137}
138
139static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr)
140{
141 struct scatterlist out_hdr, in_hdr, *sgs[3];
142 unsigned int num_out = 0, num_in = 0;
143
144 sg_init_one(&out_hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
145 sgs[num_out++] = &out_hdr;
146
147 if (vbr->sg_table.nents) {
148 if (vbr->out_hdr.type & cpu_to_virtio32(vdev: vq->vdev, VIRTIO_BLK_T_OUT))
149 sgs[num_out++] = vbr->sg_table.sgl;
150 else
151 sgs[num_out + num_in++] = vbr->sg_table.sgl;
152 }
153
154 sg_init_one(&in_hdr, &vbr->in_hdr.status, vbr->in_hdr_len);
155 sgs[num_out + num_in++] = &in_hdr;
156
157 return virtqueue_add_sgs(vq, sgs, out_sgs: num_out, in_sgs: num_in, data: vbr, GFP_ATOMIC);
158}
159
160static int virtblk_setup_discard_write_zeroes_erase(struct request *req, bool unmap)
161{
162 unsigned short segments = blk_rq_nr_discard_segments(rq: req);
163 unsigned short n = 0;
164 struct virtio_blk_discard_write_zeroes *range;
165 struct bio *bio;
166 u32 flags = 0;
167
168 if (unmap)
169 flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
170
171 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
172 if (!range)
173 return -ENOMEM;
174
175 /*
176 * Single max discard segment means multi-range discard isn't
177 * supported, and block layer only runs contiguity merge like
178 * normal RW request. So we can't reply on bio for retrieving
179 * each range info.
180 */
181 if (queue_max_discard_segments(q: req->q) == 1) {
182 range[0].flags = cpu_to_le32(flags);
183 range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
184 range[0].sector = cpu_to_le64(blk_rq_pos(req));
185 n = 1;
186 } else {
187 __rq_for_each_bio(bio, req) {
188 u64 sector = bio->bi_iter.bi_sector;
189 u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
190
191 range[n].flags = cpu_to_le32(flags);
192 range[n].num_sectors = cpu_to_le32(num_sectors);
193 range[n].sector = cpu_to_le64(sector);
194 n++;
195 }
196 }
197
198 WARN_ON_ONCE(n != segments);
199
200 bvec_set_virt(bv: &req->special_vec, vaddr: range, len: sizeof(*range) * segments);
201 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
202
203 return 0;
204}
205
206static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
207{
208 if (blk_rq_nr_phys_segments(rq: req))
209 sg_free_table_chained(table: &vbr->sg_table,
210 VIRTIO_BLK_INLINE_SG_CNT);
211}
212
213static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
214 struct virtblk_req *vbr)
215{
216 int err;
217
218 if (!blk_rq_nr_phys_segments(rq: req))
219 return 0;
220
221 vbr->sg_table.sgl = vbr->sg;
222 err = sg_alloc_table_chained(table: &vbr->sg_table,
223 nents: blk_rq_nr_phys_segments(rq: req),
224 first_chunk: vbr->sg_table.sgl,
225 VIRTIO_BLK_INLINE_SG_CNT);
226 if (unlikely(err))
227 return -ENOMEM;
228
229 return blk_rq_map_sg(rq: req, sglist: vbr->sg_table.sgl);
230}
231
232static void virtblk_cleanup_cmd(struct request *req)
233{
234 if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
235 kfree(objp: bvec_virt(bvec: &req->special_vec));
236}
237
238static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
239 struct request *req,
240 struct virtblk_req *vbr)
241{
242 size_t in_hdr_len = sizeof(vbr->in_hdr.status);
243 bool unmap = false;
244 u32 type;
245 u64 sector = 0;
246
247 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && op_is_zone_mgmt(op: req_op(req)))
248 return BLK_STS_NOTSUPP;
249
250 /* Set fields for all request types */
251 vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, val: req_get_ioprio(req));
252
253 switch (req_op(req)) {
254 case REQ_OP_READ:
255 type = VIRTIO_BLK_T_IN;
256 sector = blk_rq_pos(rq: req);
257 break;
258 case REQ_OP_WRITE:
259 type = VIRTIO_BLK_T_OUT;
260 sector = blk_rq_pos(rq: req);
261 break;
262 case REQ_OP_FLUSH:
263 type = VIRTIO_BLK_T_FLUSH;
264 break;
265 case REQ_OP_DISCARD:
266 type = VIRTIO_BLK_T_DISCARD;
267 break;
268 case REQ_OP_WRITE_ZEROES:
269 type = VIRTIO_BLK_T_WRITE_ZEROES;
270 unmap = !(req->cmd_flags & REQ_NOUNMAP);
271 break;
272 case REQ_OP_SECURE_ERASE:
273 type = VIRTIO_BLK_T_SECURE_ERASE;
274 break;
275 case REQ_OP_ZONE_OPEN:
276 type = VIRTIO_BLK_T_ZONE_OPEN;
277 sector = blk_rq_pos(rq: req);
278 break;
279 case REQ_OP_ZONE_CLOSE:
280 type = VIRTIO_BLK_T_ZONE_CLOSE;
281 sector = blk_rq_pos(rq: req);
282 break;
283 case REQ_OP_ZONE_FINISH:
284 type = VIRTIO_BLK_T_ZONE_FINISH;
285 sector = blk_rq_pos(rq: req);
286 break;
287 case REQ_OP_ZONE_APPEND:
288 type = VIRTIO_BLK_T_ZONE_APPEND;
289 sector = blk_rq_pos(rq: req);
290 in_hdr_len = sizeof(vbr->in_hdr.zone_append);
291 break;
292 case REQ_OP_ZONE_RESET:
293 type = VIRTIO_BLK_T_ZONE_RESET;
294 sector = blk_rq_pos(rq: req);
295 break;
296 case REQ_OP_ZONE_RESET_ALL:
297 type = VIRTIO_BLK_T_ZONE_RESET_ALL;
298 break;
299 case REQ_OP_DRV_IN:
300 /*
301 * Out header has already been prepared by the caller (virtblk_get_id()
302 * or virtblk_submit_zone_report()), nothing to do here.
303 */
304 return 0;
305 default:
306 WARN_ON_ONCE(1);
307 return BLK_STS_IOERR;
308 }
309
310 /* Set fields for non-REQ_OP_DRV_IN request types */
311 vbr->in_hdr_len = in_hdr_len;
312 vbr->out_hdr.type = cpu_to_virtio32(vdev, val: type);
313 vbr->out_hdr.sector = cpu_to_virtio64(vdev, val: sector);
314
315 if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES ||
316 type == VIRTIO_BLK_T_SECURE_ERASE) {
317 if (virtblk_setup_discard_write_zeroes_erase(req, unmap))
318 return BLK_STS_RESOURCE;
319 }
320
321 return 0;
322}
323
324/*
325 * The status byte is always the last byte of the virtblk request
326 * in-header. This helper fetches its value for all in-header formats
327 * that are currently defined.
328 */
329static inline u8 virtblk_vbr_status(struct virtblk_req *vbr)
330{
331 return *((u8 *)&vbr->in_hdr + vbr->in_hdr_len - 1);
332}
333
334static inline void virtblk_request_done(struct request *req)
335{
336 struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq: req);
337 blk_status_t status = virtblk_result(status: virtblk_vbr_status(vbr));
338 struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
339
340 virtblk_unmap_data(req, vbr);
341 virtblk_cleanup_cmd(req);
342
343 if (req_op(req) == REQ_OP_ZONE_APPEND)
344 req->__sector = virtio64_to_cpu(vdev: vblk->vdev,
345 val: vbr->in_hdr.zone_append.sector);
346
347 blk_mq_end_request(rq: req, error: status);
348}
349
350static void virtblk_done(struct virtqueue *vq)
351{
352 struct virtio_blk *vblk = vq->vdev->priv;
353 bool req_done = false;
354 int qid = vq->index;
355 struct virtblk_req *vbr;
356 unsigned long flags;
357 unsigned int len;
358
359 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
360 do {
361 virtqueue_disable_cb(vq);
362 while ((vbr = virtqueue_get_buf(vq: vblk->vqs[qid].vq, len: &len)) != NULL) {
363 struct request *req = blk_mq_rq_from_pdu(pdu: vbr);
364
365 if (likely(!blk_should_fake_timeout(req->q)))
366 blk_mq_complete_request(rq: req);
367 req_done = true;
368 }
369 } while (!virtqueue_enable_cb(vq));
370
371 /* In case queue is stopped waiting for more buffers. */
372 if (req_done)
373 blk_mq_start_stopped_hw_queues(q: vblk->disk->queue, async: true);
374 spin_unlock_irqrestore(lock: &vblk->vqs[qid].lock, flags);
375}
376
377static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
378{
379 struct virtio_blk *vblk = hctx->queue->queuedata;
380 struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
381 bool kick;
382
383 spin_lock_irq(lock: &vq->lock);
384 kick = virtqueue_kick_prepare(vq: vq->vq);
385 spin_unlock_irq(lock: &vq->lock);
386
387 if (kick)
388 virtqueue_notify(vq: vq->vq);
389}
390
391static blk_status_t virtblk_fail_to_queue(struct request *req, int rc)
392{
393 virtblk_cleanup_cmd(req);
394 switch (rc) {
395 case -ENOSPC:
396 return BLK_STS_DEV_RESOURCE;
397 case -ENOMEM:
398 return BLK_STS_RESOURCE;
399 default:
400 return BLK_STS_IOERR;
401 }
402}
403
404static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
405 struct virtio_blk *vblk,
406 struct request *req,
407 struct virtblk_req *vbr)
408{
409 blk_status_t status;
410 int num;
411
412 status = virtblk_setup_cmd(vdev: vblk->vdev, req, vbr);
413 if (unlikely(status))
414 return status;
415
416 num = virtblk_map_data(hctx, req, vbr);
417 if (unlikely(num < 0))
418 return virtblk_fail_to_queue(req, rc: -ENOMEM);
419 vbr->sg_table.nents = num;
420
421 blk_mq_start_request(rq: req);
422
423 return BLK_STS_OK;
424}
425
426static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
427 const struct blk_mq_queue_data *bd)
428{
429 struct virtio_blk *vblk = hctx->queue->queuedata;
430 struct request *req = bd->rq;
431 struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq: req);
432 unsigned long flags;
433 int qid = hctx->queue_num;
434 bool notify = false;
435 blk_status_t status;
436 int err;
437
438 status = virtblk_prep_rq(hctx, vblk, req, vbr);
439 if (unlikely(status))
440 return status;
441
442 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
443 err = virtblk_add_req(vq: vblk->vqs[qid].vq, vbr);
444 if (err) {
445 virtqueue_kick(vq: vblk->vqs[qid].vq);
446 /* Don't stop the queue if -ENOMEM: we may have failed to
447 * bounce the buffer due to global resource outage.
448 */
449 if (err == -ENOSPC)
450 blk_mq_stop_hw_queue(hctx);
451 spin_unlock_irqrestore(lock: &vblk->vqs[qid].lock, flags);
452 virtblk_unmap_data(req, vbr);
453 return virtblk_fail_to_queue(req, rc: err);
454 }
455
456 if (bd->last && virtqueue_kick_prepare(vq: vblk->vqs[qid].vq))
457 notify = true;
458 spin_unlock_irqrestore(lock: &vblk->vqs[qid].lock, flags);
459
460 if (notify)
461 virtqueue_notify(vq: vblk->vqs[qid].vq);
462 return BLK_STS_OK;
463}
464
465static bool virtblk_prep_rq_batch(struct request *req)
466{
467 struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
468 struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq: req);
469
470 return virtblk_prep_rq(hctx: req->mq_hctx, vblk, req, vbr) == BLK_STS_OK;
471}
472
473static void virtblk_add_req_batch(struct virtio_blk_vq *vq,
474 struct rq_list *rqlist)
475{
476 struct request *req;
477 unsigned long flags;
478 bool kick;
479
480 spin_lock_irqsave(&vq->lock, flags);
481
482 while ((req = rq_list_pop(rl: rqlist))) {
483 struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq: req);
484 int err;
485
486 err = virtblk_add_req(vq: vq->vq, vbr);
487 if (err) {
488 virtblk_unmap_data(req, vbr);
489 virtblk_cleanup_cmd(req);
490 blk_mq_requeue_request(rq: req, kick_requeue_list: true);
491 }
492 }
493
494 kick = virtqueue_kick_prepare(vq: vq->vq);
495 spin_unlock_irqrestore(lock: &vq->lock, flags);
496
497 if (kick)
498 virtqueue_notify(vq: vq->vq);
499}
500
501static void virtio_queue_rqs(struct rq_list *rqlist)
502{
503 struct rq_list submit_list = { };
504 struct rq_list requeue_list = { };
505 struct virtio_blk_vq *vq = NULL;
506 struct request *req;
507
508 while ((req = rq_list_pop(rl: rqlist))) {
509 struct virtio_blk_vq *this_vq = get_virtio_blk_vq(hctx: req->mq_hctx);
510
511 if (vq && vq != this_vq)
512 virtblk_add_req_batch(vq, rqlist: &submit_list);
513 vq = this_vq;
514
515 if (virtblk_prep_rq_batch(req))
516 rq_list_add_tail(rl: &submit_list, rq: req);
517 else
518 rq_list_add_tail(rl: &requeue_list, rq: req);
519 }
520
521 if (vq)
522 virtblk_add_req_batch(vq, rqlist: &submit_list);
523 *rqlist = requeue_list;
524}
525
526#ifdef CONFIG_BLK_DEV_ZONED
527static void *virtblk_alloc_report_buffer(struct virtio_blk *vblk,
528 unsigned int nr_zones,
529 size_t *buflen)
530{
531 struct request_queue *q = vblk->disk->queue;
532 size_t bufsize;
533 void *buf;
534
535 nr_zones = min_t(unsigned int, nr_zones,
536 get_capacity(vblk->disk) >> ilog2(vblk->zone_sectors));
537
538 bufsize = sizeof(struct virtio_blk_zone_report) +
539 nr_zones * sizeof(struct virtio_blk_zone_descriptor);
540 bufsize = min_t(size_t, bufsize,
541 queue_max_hw_sectors(q) << SECTOR_SHIFT);
542 bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
543
544 while (bufsize >= sizeof(struct virtio_blk_zone_report)) {
545 buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
546 if (buf) {
547 *buflen = bufsize;
548 return buf;
549 }
550 bufsize >>= 1;
551 }
552
553 return NULL;
554}
555
556static int virtblk_submit_zone_report(struct virtio_blk *vblk,
557 char *report_buf, size_t report_len,
558 sector_t sector)
559{
560 struct request_queue *q = vblk->disk->queue;
561 struct request *req;
562 struct virtblk_req *vbr;
563 int err;
564
565 req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
566 if (IS_ERR(req))
567 return PTR_ERR(req);
568
569 vbr = blk_mq_rq_to_pdu(req);
570 vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
571 vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_ZONE_REPORT);
572 vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, sector);
573
574 err = blk_rq_map_kern(req, report_buf, report_len, GFP_KERNEL);
575 if (err)
576 goto out;
577
578 blk_execute_rq(req, false);
579 err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
580out:
581 blk_mq_free_request(req);
582 return err;
583}
584
585static int virtblk_parse_zone(struct virtio_blk *vblk,
586 struct virtio_blk_zone_descriptor *entry,
587 unsigned int idx, report_zones_cb cb, void *data)
588{
589 struct blk_zone zone = { };
590
591 zone.start = virtio64_to_cpu(vblk->vdev, entry->z_start);
592 if (zone.start + vblk->zone_sectors <= get_capacity(vblk->disk))
593 zone.len = vblk->zone_sectors;
594 else
595 zone.len = get_capacity(vblk->disk) - zone.start;
596 zone.capacity = virtio64_to_cpu(vblk->vdev, entry->z_cap);
597 zone.wp = virtio64_to_cpu(vblk->vdev, entry->z_wp);
598
599 switch (entry->z_type) {
600 case VIRTIO_BLK_ZT_SWR:
601 zone.type = BLK_ZONE_TYPE_SEQWRITE_REQ;
602 break;
603 case VIRTIO_BLK_ZT_SWP:
604 zone.type = BLK_ZONE_TYPE_SEQWRITE_PREF;
605 break;
606 case VIRTIO_BLK_ZT_CONV:
607 zone.type = BLK_ZONE_TYPE_CONVENTIONAL;
608 break;
609 default:
610 dev_err(&vblk->vdev->dev, "zone %llu: invalid type %#x\n",
611 zone.start, entry->z_type);
612 return -EIO;
613 }
614
615 switch (entry->z_state) {
616 case VIRTIO_BLK_ZS_EMPTY:
617 zone.cond = BLK_ZONE_COND_EMPTY;
618 break;
619 case VIRTIO_BLK_ZS_CLOSED:
620 zone.cond = BLK_ZONE_COND_CLOSED;
621 break;
622 case VIRTIO_BLK_ZS_FULL:
623 zone.cond = BLK_ZONE_COND_FULL;
624 zone.wp = zone.start + zone.len;
625 break;
626 case VIRTIO_BLK_ZS_EOPEN:
627 zone.cond = BLK_ZONE_COND_EXP_OPEN;
628 break;
629 case VIRTIO_BLK_ZS_IOPEN:
630 zone.cond = BLK_ZONE_COND_IMP_OPEN;
631 break;
632 case VIRTIO_BLK_ZS_NOT_WP:
633 zone.cond = BLK_ZONE_COND_NOT_WP;
634 break;
635 case VIRTIO_BLK_ZS_RDONLY:
636 zone.cond = BLK_ZONE_COND_READONLY;
637 zone.wp = ULONG_MAX;
638 break;
639 case VIRTIO_BLK_ZS_OFFLINE:
640 zone.cond = BLK_ZONE_COND_OFFLINE;
641 zone.wp = ULONG_MAX;
642 break;
643 default:
644 dev_err(&vblk->vdev->dev, "zone %llu: invalid condition %#x\n",
645 zone.start, entry->z_state);
646 return -EIO;
647 }
648
649 /*
650 * The callback below checks the validity of the reported
651 * entry data, no need to further validate it here.
652 */
653 return cb(&zone, idx, data);
654}
655
656static int virtblk_report_zones(struct gendisk *disk, sector_t sector,
657 unsigned int nr_zones, report_zones_cb cb,
658 void *data)
659{
660 struct virtio_blk *vblk = disk->private_data;
661 struct virtio_blk_zone_report *report;
662 unsigned long long nz, i;
663 size_t buflen;
664 unsigned int zone_idx = 0;
665 int ret;
666
667 if (WARN_ON_ONCE(!vblk->zone_sectors))
668 return -EOPNOTSUPP;
669
670 report = virtblk_alloc_report_buffer(vblk, nr_zones, &buflen);
671 if (!report)
672 return -ENOMEM;
673
674 mutex_lock(&vblk->vdev_mutex);
675
676 if (!vblk->vdev) {
677 ret = -ENXIO;
678 goto fail_report;
679 }
680
681 while (zone_idx < nr_zones && sector < get_capacity(vblk->disk)) {
682 memset(report, 0, buflen);
683
684 ret = virtblk_submit_zone_report(vblk, (char *)report,
685 buflen, sector);
686 if (ret)
687 goto fail_report;
688
689 nz = min_t(u64, virtio64_to_cpu(vblk->vdev, report->nr_zones),
690 nr_zones);
691 if (!nz)
692 break;
693
694 for (i = 0; i < nz && zone_idx < nr_zones; i++) {
695 ret = virtblk_parse_zone(vblk, &report->zones[i],
696 zone_idx, cb, data);
697 if (ret)
698 goto fail_report;
699
700 sector = virtio64_to_cpu(vblk->vdev,
701 report->zones[i].z_start) +
702 vblk->zone_sectors;
703 zone_idx++;
704 }
705 }
706
707 if (zone_idx > 0)
708 ret = zone_idx;
709 else
710 ret = -EINVAL;
711fail_report:
712 mutex_unlock(&vblk->vdev_mutex);
713 kvfree(report);
714 return ret;
715}
716
717static int virtblk_read_zoned_limits(struct virtio_blk *vblk,
718 struct queue_limits *lim)
719{
720 struct virtio_device *vdev = vblk->vdev;
721 u32 v, wg;
722
723 dev_dbg(&vdev->dev, "probing host-managed zoned device\n");
724
725 lim->features |= BLK_FEAT_ZONED;
726
727 virtio_cread(vdev, struct virtio_blk_config,
728 zoned.max_open_zones, &v);
729 lim->max_open_zones = v;
730 dev_dbg(&vdev->dev, "max open zones = %u\n", v);
731
732 virtio_cread(vdev, struct virtio_blk_config,
733 zoned.max_active_zones, &v);
734 lim->max_active_zones = v;
735 dev_dbg(&vdev->dev, "max active zones = %u\n", v);
736
737 virtio_cread(vdev, struct virtio_blk_config,
738 zoned.write_granularity, &wg);
739 if (!wg) {
740 dev_warn(&vdev->dev, "zero write granularity reported\n");
741 return -ENODEV;
742 }
743 lim->physical_block_size = wg;
744 lim->io_min = wg;
745
746 dev_dbg(&vdev->dev, "write granularity = %u\n", wg);
747
748 /*
749 * virtio ZBD specification doesn't require zones to be a power of
750 * two sectors in size, but the code in this driver expects that.
751 */
752 virtio_cread(vdev, struct virtio_blk_config, zoned.zone_sectors,
753 &vblk->zone_sectors);
754 if (vblk->zone_sectors == 0 || !is_power_of_2(vblk->zone_sectors)) {
755 dev_err(&vdev->dev,
756 "zoned device with non power of two zone size %u\n",
757 vblk->zone_sectors);
758 return -ENODEV;
759 }
760 lim->chunk_sectors = vblk->zone_sectors;
761 dev_dbg(&vdev->dev, "zone sectors = %u\n", vblk->zone_sectors);
762
763 if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
764 dev_warn(&vblk->vdev->dev,
765 "ignoring negotiated F_DISCARD for zoned device\n");
766 lim->max_hw_discard_sectors = 0;
767 }
768
769 virtio_cread(vdev, struct virtio_blk_config,
770 zoned.max_append_sectors, &v);
771 if (!v) {
772 dev_warn(&vdev->dev, "zero max_append_sectors reported\n");
773 return -ENODEV;
774 }
775 if ((v << SECTOR_SHIFT) < wg) {
776 dev_err(&vdev->dev,
777 "write granularity %u exceeds max_append_sectors %u limit\n",
778 wg, v);
779 return -ENODEV;
780 }
781 lim->max_hw_zone_append_sectors = v;
782 dev_dbg(&vdev->dev, "max append sectors = %u\n", v);
783
784 return 0;
785}
786#else
787/*
788 * Zoned block device support is not configured in this kernel, host-managed
789 * zoned devices can't be supported.
790 */
791#define virtblk_report_zones NULL
792static inline int virtblk_read_zoned_limits(struct virtio_blk *vblk,
793 struct queue_limits *lim)
794{
795 dev_err(&vblk->vdev->dev,
796 "virtio_blk: zoned devices are not supported");
797 return -EOPNOTSUPP;
798}
799#endif /* CONFIG_BLK_DEV_ZONED */
800
801/* return id (s/n) string for *disk to *id_str
802 */
803static int virtblk_get_id(struct gendisk *disk, char *id_str)
804{
805 struct virtio_blk *vblk = disk->private_data;
806 struct request_queue *q = vblk->disk->queue;
807 struct request *req;
808 struct virtblk_req *vbr;
809 int err;
810
811 req = blk_mq_alloc_request(q, opf: REQ_OP_DRV_IN, flags: 0);
812 if (IS_ERR(ptr: req))
813 return PTR_ERR(ptr: req);
814
815 vbr = blk_mq_rq_to_pdu(rq: req);
816 vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
817 vbr->out_hdr.type = cpu_to_virtio32(vdev: vblk->vdev, VIRTIO_BLK_T_GET_ID);
818 vbr->out_hdr.sector = 0;
819
820 err = blk_rq_map_kern(rq: req, kbuf: id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
821 if (err)
822 goto out;
823
824 blk_execute_rq(rq: req, at_head: false);
825 err = blk_status_to_errno(status: virtblk_result(status: vbr->in_hdr.status));
826out:
827 blk_mq_free_request(rq: req);
828 return err;
829}
830
831/* We provide getgeo only to please some old bootloader/partitioning tools */
832static int virtblk_getgeo(struct gendisk *disk, struct hd_geometry *geo)
833{
834 struct virtio_blk *vblk = disk->private_data;
835 int ret = 0;
836
837 mutex_lock(lock: &vblk->vdev_mutex);
838
839 if (!vblk->vdev) {
840 ret = -ENXIO;
841 goto out;
842 }
843
844 /* see if the host passed in geometry config */
845 if (virtio_has_feature(vdev: vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
846 virtio_cread(vblk->vdev, struct virtio_blk_config,
847 geometry.cylinders, &geo->cylinders);
848 virtio_cread(vblk->vdev, struct virtio_blk_config,
849 geometry.heads, &geo->heads);
850 virtio_cread(vblk->vdev, struct virtio_blk_config,
851 geometry.sectors, &geo->sectors);
852 } else {
853 /* some standard values, similar to sd */
854 geo->heads = 1 << 6;
855 geo->sectors = 1 << 5;
856 geo->cylinders = get_capacity(disk) >> 11;
857 }
858out:
859 mutex_unlock(lock: &vblk->vdev_mutex);
860 return ret;
861}
862
863static void virtblk_free_disk(struct gendisk *disk)
864{
865 struct virtio_blk *vblk = disk->private_data;
866
867 ida_free(&vd_index_ida, id: vblk->index);
868 mutex_destroy(lock: &vblk->vdev_mutex);
869 kfree(objp: vblk);
870}
871
872static const struct block_device_operations virtblk_fops = {
873 .owner = THIS_MODULE,
874 .getgeo = virtblk_getgeo,
875 .free_disk = virtblk_free_disk,
876 .report_zones = virtblk_report_zones,
877};
878
879static int index_to_minor(int index)
880{
881 return index << PART_BITS;
882}
883
884static int minor_to_index(int minor)
885{
886 return minor >> PART_BITS;
887}
888
889static ssize_t serial_show(struct device *dev,
890 struct device_attribute *attr, char *buf)
891{
892 struct gendisk *disk = dev_to_disk(dev);
893 int err;
894
895 /* sysfs gives us a PAGE_SIZE buffer */
896 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
897
898 buf[VIRTIO_BLK_ID_BYTES] = '\0';
899 err = virtblk_get_id(disk, id_str: buf);
900 if (!err)
901 return strlen(buf);
902
903 if (err == -EIO) /* Unsupported? Make it empty. */
904 return 0;
905
906 return err;
907}
908
909static DEVICE_ATTR_RO(serial);
910
911/* The queue's logical block size must be set before calling this */
912static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
913{
914 struct virtio_device *vdev = vblk->vdev;
915 struct request_queue *q = vblk->disk->queue;
916 char cap_str_2[10], cap_str_10[10];
917 unsigned long long nblocks;
918 u64 capacity;
919
920 /* Host must always specify the capacity. */
921 virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
922
923 nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
924
925 string_get_size(size: nblocks, blk_size: queue_logical_block_size(q),
926 units: STRING_UNITS_2, buf: cap_str_2, len: sizeof(cap_str_2));
927 string_get_size(size: nblocks, blk_size: queue_logical_block_size(q),
928 units: STRING_UNITS_10, buf: cap_str_10, len: sizeof(cap_str_10));
929
930 dev_notice(&vdev->dev,
931 "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
932 vblk->disk->disk_name,
933 resize ? "new size: " : "",
934 nblocks,
935 queue_logical_block_size(q),
936 cap_str_10,
937 cap_str_2);
938
939 set_capacity_and_notify(disk: vblk->disk, size: capacity);
940}
941
942static void virtblk_config_changed_work(struct work_struct *work)
943{
944 struct virtio_blk *vblk =
945 container_of(work, struct virtio_blk, config_work);
946
947 virtblk_update_capacity(vblk, resize: true);
948}
949
950static void virtblk_config_changed(struct virtio_device *vdev)
951{
952 struct virtio_blk *vblk = vdev->priv;
953
954 queue_work(wq: virtblk_wq, work: &vblk->config_work);
955}
956
957static int init_vq(struct virtio_blk *vblk)
958{
959 int err;
960 unsigned short i;
961 struct virtqueue_info *vqs_info;
962 struct virtqueue **vqs;
963 unsigned short num_vqs;
964 unsigned short num_poll_vqs;
965 struct virtio_device *vdev = vblk->vdev;
966 struct irq_affinity desc = { 0, };
967
968 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
969 struct virtio_blk_config, num_queues,
970 &num_vqs);
971 if (err)
972 num_vqs = 1;
973
974 if (!err && !num_vqs) {
975 dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
976 return -EINVAL;
977 }
978
979 num_vqs = blk_mq_num_possible_queues(
980 min_not_zero(num_request_queues, num_vqs));
981
982 num_poll_vqs = min_t(unsigned int, poll_queues, num_vqs - 1);
983
984 vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
985 vblk->io_queues[HCTX_TYPE_READ] = 0;
986 vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
987
988 dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n",
989 vblk->io_queues[HCTX_TYPE_DEFAULT],
990 vblk->io_queues[HCTX_TYPE_READ],
991 vblk->io_queues[HCTX_TYPE_POLL]);
992
993 vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
994 if (!vblk->vqs)
995 return -ENOMEM;
996
997 vqs_info = kcalloc(num_vqs, sizeof(*vqs_info), GFP_KERNEL);
998 vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
999 if (!vqs_info || !vqs) {
1000 err = -ENOMEM;
1001 goto out;
1002 }
1003
1004 for (i = 0; i < num_vqs - num_poll_vqs; i++) {
1005 vqs_info[i].callback = virtblk_done;
1006 snprintf(buf: vblk->vqs[i].name, VQ_NAME_LEN, fmt: "req.%u", i);
1007 vqs_info[i].name = vblk->vqs[i].name;
1008 }
1009
1010 for (; i < num_vqs; i++) {
1011 snprintf(buf: vblk->vqs[i].name, VQ_NAME_LEN, fmt: "req_poll.%u", i);
1012 vqs_info[i].name = vblk->vqs[i].name;
1013 }
1014
1015 /* Discover virtqueues and write information to configuration. */
1016 err = virtio_find_vqs(vdev, nvqs: num_vqs, vqs, vqs_info, desc: &desc);
1017 if (err)
1018 goto out;
1019
1020 for (i = 0; i < num_vqs; i++) {
1021 spin_lock_init(&vblk->vqs[i].lock);
1022 vblk->vqs[i].vq = vqs[i];
1023 }
1024 vblk->num_vqs = num_vqs;
1025
1026out:
1027 kfree(objp: vqs);
1028 kfree(objp: vqs_info);
1029 if (err)
1030 kfree(objp: vblk->vqs);
1031 return err;
1032}
1033
1034/*
1035 * Legacy naming scheme used for virtio devices. We are stuck with it for
1036 * virtio blk but don't ever use it for any new driver.
1037 */
1038static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
1039{
1040 const int base = 'z' - 'a' + 1;
1041 char *begin = buf + strlen(prefix);
1042 char *end = buf + buflen;
1043 char *p;
1044 int unit;
1045
1046 p = end - 1;
1047 *p = '\0';
1048 unit = base;
1049 do {
1050 if (p == begin)
1051 return -EINVAL;
1052 *--p = 'a' + (index % unit);
1053 index = (index / unit) - 1;
1054 } while (index >= 0);
1055
1056 memmove(dest: begin, src: p, count: end - p);
1057 memcpy(to: buf, from: prefix, len: strlen(prefix));
1058
1059 return 0;
1060}
1061
1062static int virtblk_get_cache_mode(struct virtio_device *vdev)
1063{
1064 u8 writeback;
1065 int err;
1066
1067 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
1068 struct virtio_blk_config, wce,
1069 &writeback);
1070
1071 /*
1072 * If WCE is not configurable and flush is not available,
1073 * assume no writeback cache is in use.
1074 */
1075 if (err)
1076 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
1077
1078 return writeback;
1079}
1080
1081static const char *const virtblk_cache_types[] = {
1082 "write through", "write back"
1083};
1084
1085static ssize_t
1086cache_type_store(struct device *dev, struct device_attribute *attr,
1087 const char *buf, size_t count)
1088{
1089 struct gendisk *disk = dev_to_disk(dev);
1090 struct virtio_blk *vblk = disk->private_data;
1091 struct virtio_device *vdev = vblk->vdev;
1092 struct queue_limits lim;
1093 int i;
1094
1095 BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
1096 i = sysfs_match_string(virtblk_cache_types, buf);
1097 if (i < 0)
1098 return i;
1099
1100 virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), val: i);
1101
1102 lim = queue_limits_start_update(q: disk->queue);
1103 if (virtblk_get_cache_mode(vdev))
1104 lim.features |= BLK_FEAT_WRITE_CACHE;
1105 else
1106 lim.features &= ~BLK_FEAT_WRITE_CACHE;
1107 i = queue_limits_commit_update_frozen(q: disk->queue, lim: &lim);
1108 if (i)
1109 return i;
1110 return count;
1111}
1112
1113static ssize_t
1114cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
1115{
1116 struct gendisk *disk = dev_to_disk(dev);
1117 struct virtio_blk *vblk = disk->private_data;
1118 u8 writeback = virtblk_get_cache_mode(vdev: vblk->vdev);
1119
1120 BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
1121 return sysfs_emit(buf, fmt: "%s\n", virtblk_cache_types[writeback]);
1122}
1123
1124static DEVICE_ATTR_RW(cache_type);
1125
1126static struct attribute *virtblk_attrs[] = {
1127 &dev_attr_serial.attr,
1128 &dev_attr_cache_type.attr,
1129 NULL,
1130};
1131
1132static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
1133 struct attribute *a, int n)
1134{
1135 struct device *dev = kobj_to_dev(kobj);
1136 struct gendisk *disk = dev_to_disk(dev);
1137 struct virtio_blk *vblk = disk->private_data;
1138 struct virtio_device *vdev = vblk->vdev;
1139
1140 if (a == &dev_attr_cache_type.attr &&
1141 !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
1142 return S_IRUGO;
1143
1144 return a->mode;
1145}
1146
1147static const struct attribute_group virtblk_attr_group = {
1148 .attrs = virtblk_attrs,
1149 .is_visible = virtblk_attrs_are_visible,
1150};
1151
1152static const struct attribute_group *virtblk_attr_groups[] = {
1153 &virtblk_attr_group,
1154 NULL,
1155};
1156
1157static void virtblk_map_queues(struct blk_mq_tag_set *set)
1158{
1159 struct virtio_blk *vblk = set->driver_data;
1160 int i, qoff;
1161
1162 for (i = 0, qoff = 0; i < set->nr_maps; i++) {
1163 struct blk_mq_queue_map *map = &set->map[i];
1164
1165 map->nr_queues = vblk->io_queues[i];
1166 map->queue_offset = qoff;
1167 qoff += map->nr_queues;
1168
1169 if (map->nr_queues == 0)
1170 continue;
1171
1172 /*
1173 * Regular queues have interrupts and hence CPU affinity is
1174 * defined by the core virtio code, but polling queues have
1175 * no interrupts so we let the block layer assign CPU affinity.
1176 */
1177 if (i == HCTX_TYPE_POLL)
1178 blk_mq_map_queues(qmap: &set->map[i]);
1179 else
1180 blk_mq_map_hw_queues(qmap: &set->map[i],
1181 dev: &vblk->vdev->dev, offset: 0);
1182 }
1183}
1184
1185static void virtblk_complete_batch(struct io_comp_batch *iob)
1186{
1187 struct request *req;
1188
1189 rq_list_for_each(&iob->req_list, req) {
1190 virtblk_unmap_data(req, vbr: blk_mq_rq_to_pdu(rq: req));
1191 virtblk_cleanup_cmd(req);
1192 }
1193 blk_mq_end_request_batch(ib: iob);
1194}
1195
1196static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1197{
1198 struct virtio_blk *vblk = hctx->queue->queuedata;
1199 struct virtio_blk_vq *vq = get_virtio_blk_vq(hctx);
1200 struct virtblk_req *vbr;
1201 unsigned long flags;
1202 unsigned int len;
1203 int found = 0;
1204
1205 spin_lock_irqsave(&vq->lock, flags);
1206
1207 while ((vbr = virtqueue_get_buf(vq: vq->vq, len: &len)) != NULL) {
1208 struct request *req = blk_mq_rq_from_pdu(pdu: vbr);
1209 u8 status = virtblk_vbr_status(vbr);
1210
1211 found++;
1212 if (!blk_mq_complete_request_remote(rq: req) &&
1213 !blk_mq_add_to_batch(req, iob, is_error: status != VIRTIO_BLK_S_OK,
1214 complete: virtblk_complete_batch))
1215 virtblk_request_done(req);
1216 }
1217
1218 if (found)
1219 blk_mq_start_stopped_hw_queues(q: vblk->disk->queue, async: true);
1220
1221 spin_unlock_irqrestore(lock: &vq->lock, flags);
1222
1223 return found;
1224}
1225
1226static const struct blk_mq_ops virtio_mq_ops = {
1227 .queue_rq = virtio_queue_rq,
1228 .queue_rqs = virtio_queue_rqs,
1229 .commit_rqs = virtio_commit_rqs,
1230 .complete = virtblk_request_done,
1231 .map_queues = virtblk_map_queues,
1232 .poll = virtblk_poll,
1233};
1234
1235static unsigned int virtblk_queue_depth;
1236module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
1237
1238static int virtblk_read_limits(struct virtio_blk *vblk,
1239 struct queue_limits *lim)
1240{
1241 struct virtio_device *vdev = vblk->vdev;
1242 u32 v, max_size, sg_elems, opt_io_size;
1243 u32 max_discard_segs = 0;
1244 u32 discard_granularity = 0;
1245 u16 min_io_size;
1246 u8 physical_block_exp, alignment_offset;
1247 size_t max_dma_size;
1248 int err;
1249
1250 /* We need to know how many segments before we allocate. */
1251 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
1252 struct virtio_blk_config, seg_max,
1253 &sg_elems);
1254
1255 /* We need at least one SG element, whatever they say. */
1256 if (err || !sg_elems)
1257 sg_elems = 1;
1258
1259 /* Prevent integer overflows and honor max vq size */
1260 sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
1261
1262 /* We can handle whatever the host told us to handle. */
1263 lim->max_segments = sg_elems;
1264
1265 /* No real sector limit. */
1266 lim->max_hw_sectors = UINT_MAX;
1267
1268 max_dma_size = virtio_max_dma_size(vdev);
1269 max_size = max_dma_size > U32_MAX ? U32_MAX : max_dma_size;
1270
1271 /* Host can optionally specify maximum segment size and number of
1272 * segments. */
1273 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
1274 struct virtio_blk_config, size_max, &v);
1275 if (!err)
1276 max_size = min(max_size, v);
1277
1278 lim->max_segment_size = max_size;
1279
1280 /* Host can optionally specify the block size of the device */
1281 virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
1282 struct virtio_blk_config, blk_size,
1283 &lim->logical_block_size);
1284
1285 /* Use topology information if available */
1286 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1287 struct virtio_blk_config, physical_block_exp,
1288 &physical_block_exp);
1289 if (!err && physical_block_exp)
1290 lim->physical_block_size =
1291 lim->logical_block_size * (1 << physical_block_exp);
1292
1293 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1294 struct virtio_blk_config, alignment_offset,
1295 &alignment_offset);
1296 if (!err && alignment_offset)
1297 lim->alignment_offset =
1298 lim->logical_block_size * alignment_offset;
1299
1300 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1301 struct virtio_blk_config, min_io_size,
1302 &min_io_size);
1303 if (!err && min_io_size)
1304 lim->io_min = lim->logical_block_size * min_io_size;
1305
1306 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1307 struct virtio_blk_config, opt_io_size,
1308 &opt_io_size);
1309 if (!err && opt_io_size)
1310 lim->io_opt = lim->logical_block_size * opt_io_size;
1311
1312 if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
1313 virtio_cread(vdev, struct virtio_blk_config,
1314 discard_sector_alignment, &discard_granularity);
1315
1316 virtio_cread(vdev, struct virtio_blk_config,
1317 max_discard_sectors, &v);
1318 lim->max_hw_discard_sectors = v ? v : UINT_MAX;
1319
1320 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
1321 &max_discard_segs);
1322 }
1323
1324 if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
1325 virtio_cread(vdev, struct virtio_blk_config,
1326 max_write_zeroes_sectors, &v);
1327 lim->max_write_zeroes_sectors = v ? v : UINT_MAX;
1328 }
1329
1330 /* The discard and secure erase limits are combined since the Linux
1331 * block layer uses the same limit for both commands.
1332 *
1333 * If both VIRTIO_BLK_F_SECURE_ERASE and VIRTIO_BLK_F_DISCARD features
1334 * are negotiated, we will use the minimum between the limits.
1335 *
1336 * discard sector alignment is set to the minimum between discard_sector_alignment
1337 * and secure_erase_sector_alignment.
1338 *
1339 * max discard sectors is set to the minimum between max_discard_seg and
1340 * max_secure_erase_seg.
1341 */
1342 if (virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1343
1344 virtio_cread(vdev, struct virtio_blk_config,
1345 secure_erase_sector_alignment, &v);
1346
1347 /* secure_erase_sector_alignment should not be zero, the device should set a
1348 * valid number of sectors.
1349 */
1350 if (!v) {
1351 dev_err(&vdev->dev,
1352 "virtio_blk: secure_erase_sector_alignment can't be 0\n");
1353 return -EINVAL;
1354 }
1355
1356 discard_granularity = min_not_zero(discard_granularity, v);
1357
1358 virtio_cread(vdev, struct virtio_blk_config,
1359 max_secure_erase_sectors, &v);
1360
1361 /* max_secure_erase_sectors should not be zero, the device should set a
1362 * valid number of sectors.
1363 */
1364 if (!v) {
1365 dev_err(&vdev->dev,
1366 "virtio_blk: max_secure_erase_sectors can't be 0\n");
1367 return -EINVAL;
1368 }
1369
1370 lim->max_secure_erase_sectors = v;
1371
1372 virtio_cread(vdev, struct virtio_blk_config,
1373 max_secure_erase_seg, &v);
1374
1375 /* max_secure_erase_seg should not be zero, the device should set a
1376 * valid number of segments
1377 */
1378 if (!v) {
1379 dev_err(&vdev->dev,
1380 "virtio_blk: max_secure_erase_seg can't be 0\n");
1381 return -EINVAL;
1382 }
1383
1384 max_discard_segs = min_not_zero(max_discard_segs, v);
1385 }
1386
1387 if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD) ||
1388 virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1389 /* max_discard_seg and discard_granularity will be 0 only
1390 * if max_discard_seg and discard_sector_alignment fields in the virtio
1391 * config are 0 and VIRTIO_BLK_F_SECURE_ERASE feature is not negotiated.
1392 * In this case, we use default values.
1393 */
1394 if (!max_discard_segs)
1395 max_discard_segs = sg_elems;
1396
1397 lim->max_discard_segments =
1398 min(max_discard_segs, MAX_DISCARD_SEGMENTS);
1399
1400 if (discard_granularity)
1401 lim->discard_granularity =
1402 discard_granularity << SECTOR_SHIFT;
1403 else
1404 lim->discard_granularity = lim->logical_block_size;
1405 }
1406
1407 if (virtio_has_feature(vdev, VIRTIO_BLK_F_ZONED)) {
1408 u8 model;
1409
1410 virtio_cread(vdev, struct virtio_blk_config, zoned.model, &model);
1411 switch (model) {
1412 case VIRTIO_BLK_Z_NONE:
1413 case VIRTIO_BLK_Z_HA:
1414 /* treat host-aware devices as non-zoned */
1415 return 0;
1416 case VIRTIO_BLK_Z_HM:
1417 err = virtblk_read_zoned_limits(vblk, lim);
1418 if (err)
1419 return err;
1420 break;
1421 default:
1422 dev_err(&vdev->dev, "unsupported zone model %d\n", model);
1423 return -EINVAL;
1424 }
1425 }
1426
1427 return 0;
1428}
1429
1430static int virtblk_probe(struct virtio_device *vdev)
1431{
1432 struct virtio_blk *vblk;
1433 struct queue_limits lim = {
1434 .features = BLK_FEAT_ROTATIONAL,
1435 .logical_block_size = SECTOR_SIZE,
1436 };
1437 int err, index;
1438 unsigned int queue_depth;
1439
1440 if (!vdev->config->get) {
1441 dev_err(&vdev->dev, "%s failure: config access disabled\n",
1442 __func__);
1443 return -EINVAL;
1444 }
1445
1446 err = ida_alloc_range(&vd_index_ida, min: 0,
1447 max: minor_to_index(minor: 1 << MINORBITS) - 1, GFP_KERNEL);
1448 if (err < 0)
1449 goto out;
1450 index = err;
1451
1452 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
1453 if (!vblk) {
1454 err = -ENOMEM;
1455 goto out_free_index;
1456 }
1457
1458 mutex_init(&vblk->vdev_mutex);
1459
1460 vblk->vdev = vdev;
1461
1462 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
1463
1464 err = init_vq(vblk);
1465 if (err)
1466 goto out_free_vblk;
1467
1468 /* Default queue sizing is to fill the ring. */
1469 if (!virtblk_queue_depth) {
1470 queue_depth = vblk->vqs[0].vq->num_free;
1471 /* ... but without indirect descs, we use 2 descs per req */
1472 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
1473 queue_depth /= 2;
1474 } else {
1475 queue_depth = virtblk_queue_depth;
1476 }
1477
1478 memset(s: &vblk->tag_set, c: 0, n: sizeof(vblk->tag_set));
1479 vblk->tag_set.ops = &virtio_mq_ops;
1480 vblk->tag_set.queue_depth = queue_depth;
1481 vblk->tag_set.numa_node = NUMA_NO_NODE;
1482 vblk->tag_set.cmd_size =
1483 sizeof(struct virtblk_req) +
1484 sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
1485 vblk->tag_set.driver_data = vblk;
1486 vblk->tag_set.nr_hw_queues = vblk->num_vqs;
1487 vblk->tag_set.nr_maps = 1;
1488 if (vblk->io_queues[HCTX_TYPE_POLL])
1489 vblk->tag_set.nr_maps = 3;
1490
1491 err = blk_mq_alloc_tag_set(set: &vblk->tag_set);
1492 if (err)
1493 goto out_free_vq;
1494
1495 err = virtblk_read_limits(vblk, lim: &lim);
1496 if (err)
1497 goto out_free_tags;
1498
1499 if (virtblk_get_cache_mode(vdev))
1500 lim.features |= BLK_FEAT_WRITE_CACHE;
1501
1502 vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, &lim, vblk);
1503 if (IS_ERR(ptr: vblk->disk)) {
1504 err = PTR_ERR(ptr: vblk->disk);
1505 goto out_free_tags;
1506 }
1507
1508 virtblk_name_format(prefix: "vd", index, buf: vblk->disk->disk_name, DISK_NAME_LEN);
1509
1510 vblk->disk->major = major;
1511 vblk->disk->first_minor = index_to_minor(index);
1512 vblk->disk->minors = 1 << PART_BITS;
1513 vblk->disk->private_data = vblk;
1514 vblk->disk->fops = &virtblk_fops;
1515 vblk->index = index;
1516
1517 /* If disk is read-only in the host, the guest should obey */
1518 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
1519 set_disk_ro(disk: vblk->disk, read_only: 1);
1520
1521 virtblk_update_capacity(vblk, resize: false);
1522 virtio_device_ready(dev: vdev);
1523
1524 /*
1525 * All steps that follow use the VQs therefore they need to be
1526 * placed after the virtio_device_ready() call above.
1527 */
1528 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
1529 (lim.features & BLK_FEAT_ZONED)) {
1530 err = blk_revalidate_disk_zones(disk: vblk->disk);
1531 if (err)
1532 goto out_cleanup_disk;
1533 }
1534
1535 err = device_add_disk(parent: &vdev->dev, disk: vblk->disk, groups: virtblk_attr_groups);
1536 if (err)
1537 goto out_cleanup_disk;
1538
1539 return 0;
1540
1541out_cleanup_disk:
1542 put_disk(disk: vblk->disk);
1543out_free_tags:
1544 blk_mq_free_tag_set(set: &vblk->tag_set);
1545out_free_vq:
1546 vdev->config->del_vqs(vdev);
1547 kfree(objp: vblk->vqs);
1548out_free_vblk:
1549 kfree(objp: vblk);
1550out_free_index:
1551 ida_free(&vd_index_ida, id: index);
1552out:
1553 return err;
1554}
1555
1556static void virtblk_remove(struct virtio_device *vdev)
1557{
1558 struct virtio_blk *vblk = vdev->priv;
1559
1560 /* Make sure no work handler is accessing the device. */
1561 flush_work(work: &vblk->config_work);
1562
1563 del_gendisk(gp: vblk->disk);
1564 blk_mq_free_tag_set(set: &vblk->tag_set);
1565
1566 mutex_lock(lock: &vblk->vdev_mutex);
1567
1568 /* Stop all the virtqueues. */
1569 virtio_reset_device(dev: vdev);
1570
1571 /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
1572 vblk->vdev = NULL;
1573
1574 vdev->config->del_vqs(vdev);
1575 kfree(objp: vblk->vqs);
1576
1577 mutex_unlock(lock: &vblk->vdev_mutex);
1578
1579 put_disk(disk: vblk->disk);
1580}
1581
1582static int virtblk_freeze_priv(struct virtio_device *vdev)
1583{
1584 struct virtio_blk *vblk = vdev->priv;
1585 struct request_queue *q = vblk->disk->queue;
1586 unsigned int memflags;
1587
1588 /* Ensure no requests in virtqueues before deleting vqs. */
1589 memflags = blk_mq_freeze_queue(q);
1590 blk_mq_quiesce_queue_nowait(q);
1591 blk_mq_unfreeze_queue(q, memflags);
1592
1593 /* Ensure we don't receive any more interrupts */
1594 virtio_reset_device(dev: vdev);
1595
1596 /* Make sure no work handler is accessing the device. */
1597 flush_work(work: &vblk->config_work);
1598
1599 vdev->config->del_vqs(vdev);
1600 kfree(objp: vblk->vqs);
1601
1602 return 0;
1603}
1604
1605static int virtblk_restore_priv(struct virtio_device *vdev)
1606{
1607 struct virtio_blk *vblk = vdev->priv;
1608 int ret;
1609
1610 ret = init_vq(vblk: vdev->priv);
1611 if (ret)
1612 return ret;
1613
1614 virtio_device_ready(dev: vdev);
1615 blk_mq_unquiesce_queue(q: vblk->disk->queue);
1616
1617 return 0;
1618}
1619
1620#ifdef CONFIG_PM_SLEEP
1621static int virtblk_freeze(struct virtio_device *vdev)
1622{
1623 return virtblk_freeze_priv(vdev);
1624}
1625
1626static int virtblk_restore(struct virtio_device *vdev)
1627{
1628 return virtblk_restore_priv(vdev);
1629}
1630#endif
1631
1632static int virtblk_reset_prepare(struct virtio_device *vdev)
1633{
1634 return virtblk_freeze_priv(vdev);
1635}
1636
1637static int virtblk_reset_done(struct virtio_device *vdev)
1638{
1639 return virtblk_restore_priv(vdev);
1640}
1641
1642static const struct virtio_device_id id_table[] = {
1643 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1644 { 0 },
1645};
1646
1647static unsigned int features_legacy[] = {
1648 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1649 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1650 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1651 VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1652 VIRTIO_BLK_F_SECURE_ERASE,
1653}
1654;
1655static unsigned int features[] = {
1656 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1657 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1658 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1659 VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1660 VIRTIO_BLK_F_SECURE_ERASE, VIRTIO_BLK_F_ZONED,
1661};
1662
1663static struct virtio_driver virtio_blk = {
1664 .feature_table = features,
1665 .feature_table_size = ARRAY_SIZE(features),
1666 .feature_table_legacy = features_legacy,
1667 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
1668 .driver.name = KBUILD_MODNAME,
1669 .id_table = id_table,
1670 .probe = virtblk_probe,
1671 .remove = virtblk_remove,
1672 .config_changed = virtblk_config_changed,
1673#ifdef CONFIG_PM_SLEEP
1674 .freeze = virtblk_freeze,
1675 .restore = virtblk_restore,
1676#endif
1677 .reset_prepare = virtblk_reset_prepare,
1678 .reset_done = virtblk_reset_done,
1679};
1680
1681static int __init virtio_blk_init(void)
1682{
1683 int error;
1684
1685 virtblk_wq = alloc_workqueue("virtio-blk", WQ_PERCPU, 0);
1686 if (!virtblk_wq)
1687 return -ENOMEM;
1688
1689 major = register_blkdev(0, "virtblk");
1690 if (major < 0) {
1691 error = major;
1692 goto out_destroy_workqueue;
1693 }
1694
1695 error = register_virtio_driver(&virtio_blk);
1696 if (error)
1697 goto out_unregister_blkdev;
1698 return 0;
1699
1700out_unregister_blkdev:
1701 unregister_blkdev(major, name: "virtblk");
1702out_destroy_workqueue:
1703 destroy_workqueue(wq: virtblk_wq);
1704 return error;
1705}
1706
1707static void __exit virtio_blk_fini(void)
1708{
1709 unregister_virtio_driver(drv: &virtio_blk);
1710 unregister_blkdev(major, name: "virtblk");
1711 destroy_workqueue(wq: virtblk_wq);
1712}
1713module_init(virtio_blk_init);
1714module_exit(virtio_blk_fini);
1715
1716MODULE_DEVICE_TABLE(virtio, id_table);
1717MODULE_DESCRIPTION("Virtio block driver");
1718MODULE_LICENSE("GPL");
1719