projects/linux·blob master · refs

1733 lines · 43.5 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2//#define DEBUG
3#include <linux/spinlock.h>
4#include <linux/slab.h>
5#include <linux/blkdev.h>
6#include <linux/hdreg.h>
7#include <linux/module.h>
8#include <linux/mutex.h>
9#include <linux/interrupt.h>
10#include <linux/virtio.h>
11#include <linux/virtio_blk.h>
12#include <linux/scatterlist.h>
13#include <linux/string_helpers.h>
14#include <linux/idr.h>
15#include <linux/blk-mq.h>
16#include <linux/numa.h>
17#include <linux/vmalloc.h>
18#include <uapi/linux/virtio_ring.h>
19
20#define PART_BITS 4
21#define VQ_NAME_LEN 16
22#define MAX_DISCARD_SEGMENTS 256u
23
24/* The maximum number of sg elements that fit into a virtqueue */
25#define VIRTIO_BLK_MAX_SG_ELEMS 32768
26
27#ifdef CONFIG_ARCH_NO_SG_CHAIN
28#define VIRTIO_BLK_INLINE_SG_CNT 0
29#else
30#define VIRTIO_BLK_INLINE_SG_CNT 2
31#endif
32
33static unsigned int num_request_queues;
34module_param(num_request_queues, uint, 0644);
35MODULE_PARM_DESC(num_request_queues,
36 "Limit the number of request queues to use for blk device. "
37 "0 for no limit. "
38 "Values > nr_cpu_ids truncated to nr_cpu_ids.");
39
40static unsigned int poll_queues;
41module_param(poll_queues, uint, 0644);
42MODULE_PARM_DESC(poll_queues, "The number of dedicated virtqueues for polling I/O");
43
44static int major;
45static DEFINE_IDA(vd_index_ida);
46
47static struct workqueue_struct *virtblk_wq;
48
49struct virtio_blk_vq {
50 struct virtqueue *vq;
51 spinlock_t lock;
52 char name[VQ_NAME_LEN];
53} ____cacheline_aligned_in_smp;
54
55struct virtio_blk {
56 /*
57 * This mutex must be held by anything that may run after
58 * virtblk_remove() sets vblk->vdev to NULL.
59 *
60 * blk-mq, virtqueue processing, and sysfs attribute code paths are
61 * shut down before vblk->vdev is set to NULL and therefore do not need
62 * to hold this mutex.
63 */
64 struct mutex vdev_mutex;
65 struct virtio_device *vdev;
66
67 /* The disk structure for the kernel. */
68 struct gendisk *disk;
69
70 /* Block layer tags. */
71 struct blk_mq_tag_set tag_set;
72
73 /* Process context for config space updates */
74 struct work_struct config_work;
75
76 /* Ida index - used to track minor number allocations. */
77 int index;
78
79 /* num of vqs */
80 int num_vqs;
81 int io_queues[HCTX_MAX_TYPES];
82 struct virtio_blk_vq *vqs;
83
84 /* For zoned device */
85 unsigned int zone_sectors;
86};
87
88struct virtblk_req {
89 /* Out header */
90 struct virtio_blk_outhdr out_hdr;
91
92 /* In header */
93 union {
94 u8 status;
95
96 /*
97 * The zone append command has an extended in header.
98 * The status field in zone_append_in_hdr must always
99 * be the last byte.
100 */
101 struct {
102 __virtio64 sector;
103 u8 status;
104 } zone_append;
105 } in_hdr;
106
107 size_t in_hdr_len;
108
109 struct sg_table sg_table;
110 struct scatterlist sg[];
111};
112
113static inline blk_status_t virtblk_result(u8 status)
114{
115 switch (status) {
116 case VIRTIO_BLK_S_OK:
117 return BLK_STS_OK;
118 case VIRTIO_BLK_S_UNSUPP:
119 return BLK_STS_NOTSUPP;
120 case VIRTIO_BLK_S_ZONE_OPEN_RESOURCE:
121 return BLK_STS_ZONE_OPEN_RESOURCE;
122 case VIRTIO_BLK_S_ZONE_ACTIVE_RESOURCE:
123 return BLK_STS_ZONE_ACTIVE_RESOURCE;
124 case VIRTIO_BLK_S_IOERR:
125 case VIRTIO_BLK_S_ZONE_UNALIGNED_WP:
126 default:
127 return BLK_STS_IOERR;
128 }
129}
130
131static inline struct virtio_blk_vq *get_virtio_blk_vq(struct blk_mq_hw_ctx *hctx)
132{
133 struct virtio_blk *vblk = hctx->queue->queuedata;
134 struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
135
136 return vq;
137}
138
139static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr)
140{
141 struct scatterlist out_hdr, in_hdr, *sgs[3];
142 unsigned int num_out = 0, num_in = 0;
143
144 sg_init_one(&out_hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
145 sgs[num_out++] = &out_hdr;
146
147 if (vbr->sg_table.nents) {
148 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
149 sgs[num_out++] = vbr->sg_table.sgl;
150 else
151 sgs[num_out + num_in++] = vbr->sg_table.sgl;
152 }
153
154 sg_init_one(&in_hdr, &vbr->in_hdr.status, vbr->in_hdr_len);
155 sgs[num_out + num_in++] = &in_hdr;
156
157 return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
158}
159
160static int virtblk_setup_discard_write_zeroes_erase(struct request *req, bool unmap)
161{
162 unsigned short segments = blk_rq_nr_discard_segments(req);
163 unsigned short n = 0;
164 struct virtio_blk_discard_write_zeroes *range;
165 struct bio *bio;
166 u32 flags = 0;
167
168 if (unmap)
169 flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
170
171 range = kmalloc_objs(*range, segments, GFP_ATOMIC);
172 if (!range)
173 return -ENOMEM;
174
175 /*
176 * Single max discard segment means multi-range discard isn't
177 * supported, and block layer only runs contiguity merge like
178 * normal RW request. So we can't reply on bio for retrieving
179 * each range info.
180 */
181 if (queue_max_discard_segments(req->q) == 1) {
182 range[0].flags = cpu_to_le32(flags);
183 range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
184 range[0].sector = cpu_to_le64(blk_rq_pos(req));
185 n = 1;
186 } else {
187 __rq_for_each_bio(bio, req) {
188 u64 sector = bio->bi_iter.bi_sector;
189 u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
190
191 range[n].flags = cpu_to_le32(flags);
192 range[n].num_sectors = cpu_to_le32(num_sectors);
193 range[n].sector = cpu_to_le64(sector);
194 n++;
195 }
196 }
197
198 WARN_ON_ONCE(n != segments);
199
200 bvec_set_virt(&req->special_vec, range, sizeof(*range) * segments);
201 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
202
203 return 0;
204}
205
206static void virtblk_unmap_data(struct request *req, struct virtblk_req *vbr)
207{
208 if (blk_rq_nr_phys_segments(req))
209 sg_free_table_chained(&vbr->sg_table,
210 VIRTIO_BLK_INLINE_SG_CNT);
211}
212
213static int virtblk_map_data(struct blk_mq_hw_ctx *hctx, struct request *req,
214 struct virtblk_req *vbr)
215{
216 int err;
217
218 if (!blk_rq_nr_phys_segments(req))
219 return 0;
220
221 vbr->sg_table.sgl = vbr->sg;
222 err = sg_alloc_table_chained(&vbr->sg_table,
223 blk_rq_nr_phys_segments(req),
224 vbr->sg_table.sgl,
225 VIRTIO_BLK_INLINE_SG_CNT);
226 if (unlikely(err))
227 return -ENOMEM;
228
229 return blk_rq_map_sg(req, vbr->sg_table.sgl);
230}
231
232static void virtblk_cleanup_cmd(struct request *req)
233{
234 if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
235 kfree(bvec_virt(&req->special_vec));
236}
237
238static blk_status_t virtblk_setup_cmd(struct virtio_device *vdev,
239 struct request *req,
240 struct virtblk_req *vbr)
241{
242 size_t in_hdr_len = sizeof(vbr->in_hdr.status);
243 bool unmap = false;
244 u32 type;
245 u64 sector = 0;
246
247 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && op_is_zone_mgmt(req_op(req)))
248 return BLK_STS_NOTSUPP;
249
250 /* Set fields for all request types */
251 vbr->out_hdr.ioprio = cpu_to_virtio32(vdev, req_get_ioprio(req));
252
253 switch (req_op(req)) {
254 case REQ_OP_READ:
255 type = VIRTIO_BLK_T_IN;
256 sector = blk_rq_pos(req);
257 break;
258 case REQ_OP_WRITE:
259 type = VIRTIO_BLK_T_OUT;
260 sector = blk_rq_pos(req);
261 break;
262 case REQ_OP_FLUSH:
263 type = VIRTIO_BLK_T_FLUSH;
264 break;
265 case REQ_OP_DISCARD:
266 type = VIRTIO_BLK_T_DISCARD;
267 break;
268 case REQ_OP_WRITE_ZEROES:
269 type = VIRTIO_BLK_T_WRITE_ZEROES;
270 unmap = !(req->cmd_flags & REQ_NOUNMAP);
271 break;
272 case REQ_OP_SECURE_ERASE:
273 type = VIRTIO_BLK_T_SECURE_ERASE;
274 break;
275 case REQ_OP_ZONE_OPEN:
276 type = VIRTIO_BLK_T_ZONE_OPEN;
277 sector = blk_rq_pos(req);
278 break;
279 case REQ_OP_ZONE_CLOSE:
280 type = VIRTIO_BLK_T_ZONE_CLOSE;
281 sector = blk_rq_pos(req);
282 break;
283 case REQ_OP_ZONE_FINISH:
284 type = VIRTIO_BLK_T_ZONE_FINISH;
285 sector = blk_rq_pos(req);
286 break;
287 case REQ_OP_ZONE_APPEND:
288 type = VIRTIO_BLK_T_ZONE_APPEND;
289 sector = blk_rq_pos(req);
290 in_hdr_len = sizeof(vbr->in_hdr.zone_append);
291 break;
292 case REQ_OP_ZONE_RESET:
293 type = VIRTIO_BLK_T_ZONE_RESET;
294 sector = blk_rq_pos(req);
295 break;
296 case REQ_OP_ZONE_RESET_ALL:
297 type = VIRTIO_BLK_T_ZONE_RESET_ALL;
298 break;
299 case REQ_OP_DRV_IN:
300 /*
301 * Out header has already been prepared by the caller (virtblk_get_id()
302 * or virtblk_submit_zone_report()), nothing to do here.
303 */
304 return 0;
305 default:
306 WARN_ON_ONCE(1);
307 return BLK_STS_IOERR;
308 }
309
310 /* Set fields for non-REQ_OP_DRV_IN request types */
311 vbr->in_hdr_len = in_hdr_len;
312 vbr->out_hdr.type = cpu_to_virtio32(vdev, type);
313 vbr->out_hdr.sector = cpu_to_virtio64(vdev, sector);
314
315 if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES ||
316 type == VIRTIO_BLK_T_SECURE_ERASE) {
317 if (virtblk_setup_discard_write_zeroes_erase(req, unmap))
318 return BLK_STS_RESOURCE;
319 }
320
321 return 0;
322}
323
324/*
325 * The status byte is always the last byte of the virtblk request
326 * in-header. This helper fetches its value for all in-header formats
327 * that are currently defined.
328 */
329static inline u8 virtblk_vbr_status(struct virtblk_req *vbr)
330{
331 return *((u8 *)&vbr->in_hdr + vbr->in_hdr_len - 1);
332}
333
334static inline void virtblk_request_done(struct request *req)
335{
336 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
337 blk_status_t status = virtblk_result(virtblk_vbr_status(vbr));
338 struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
339
340 virtblk_unmap_data(req, vbr);
341 virtblk_cleanup_cmd(req);
342
343 if (req_op(req) == REQ_OP_ZONE_APPEND)
344 req->__sector = virtio64_to_cpu(vblk->vdev,
345 vbr->in_hdr.zone_append.sector);
346
347 blk_mq_end_request(req, status);
348}
349
350static void virtblk_done(struct virtqueue *vq)
351{
352 struct virtio_blk *vblk = vq->vdev->priv;
353 bool req_done = false;
354 int qid = vq->index;
355 struct virtblk_req *vbr;
356 unsigned long flags;
357 unsigned int len;
358
359 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
360 do {
361 virtqueue_disable_cb(vq);
362 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
363 struct request *req = blk_mq_rq_from_pdu(vbr);
364
365 if (likely(!blk_should_fake_timeout(req->q)))
366 blk_mq_complete_request(req);
367 req_done = true;
368 }
369 } while (!virtqueue_enable_cb(vq));
370
371 /* In case queue is stopped waiting for more buffers. */
372 if (req_done)
373 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
374 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
375}
376
377static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
378{
379 struct virtio_blk *vblk = hctx->queue->queuedata;
380 struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
381 bool kick;
382
383 spin_lock_irq(&vq->lock);
384 kick = virtqueue_kick_prepare(vq->vq);
385 spin_unlock_irq(&vq->lock);
386
387 if (kick)
388 virtqueue_notify(vq->vq);
389}
390
391static blk_status_t virtblk_fail_to_queue(struct request *req, int rc)
392{
393 virtblk_cleanup_cmd(req);
394 switch (rc) {
395 case -ENOSPC:
396 return BLK_STS_DEV_RESOURCE;
397 case -ENOMEM:
398 return BLK_STS_RESOURCE;
399 default:
400 return BLK_STS_IOERR;
401 }
402}
403
404static blk_status_t virtblk_prep_rq(struct blk_mq_hw_ctx *hctx,
405 struct virtio_blk *vblk,
406 struct request *req,
407 struct virtblk_req *vbr)
408{
409 blk_status_t status;
410 int num;
411
412 status = virtblk_setup_cmd(vblk->vdev, req, vbr);
413 if (unlikely(status))
414 return status;
415
416 num = virtblk_map_data(hctx, req, vbr);
417 if (unlikely(num < 0))
418 return virtblk_fail_to_queue(req, -ENOMEM);
419 vbr->sg_table.nents = num;
420
421 blk_mq_start_request(req);
422
423 return BLK_STS_OK;
424}
425
426static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
427 const struct blk_mq_queue_data *bd)
428{
429 struct virtio_blk *vblk = hctx->queue->queuedata;
430 struct request *req = bd->rq;
431 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
432 unsigned long flags;
433 int qid = hctx->queue_num;
434 bool notify = false;
435 blk_status_t status;
436 int err;
437
438 status = virtblk_prep_rq(hctx, vblk, req, vbr);
439 if (unlikely(status))
440 return status;
441
442 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
443 err = virtblk_add_req(vblk->vqs[qid].vq, vbr);
444 if (err) {
445 virtqueue_kick(vblk->vqs[qid].vq);
446 /* Don't stop the queue if -ENOMEM: we may have failed to
447 * bounce the buffer due to global resource outage.
448 */
449 if (err == -ENOSPC)
450 blk_mq_stop_hw_queue(hctx);
451 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
452 virtblk_unmap_data(req, vbr);
453 return virtblk_fail_to_queue(req, err);
454 }
455
456 if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
457 notify = true;
458 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
459
460 if (notify)
461 virtqueue_notify(vblk->vqs[qid].vq);
462 return BLK_STS_OK;
463}
464
465static bool virtblk_prep_rq_batch(struct request *req)
466{
467 struct virtio_blk *vblk = req->mq_hctx->queue->queuedata;
468 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
469
470 return virtblk_prep_rq(req->mq_hctx, vblk, req, vbr) == BLK_STS_OK;
471}
472
473static void virtblk_add_req_batch(struct virtio_blk_vq *vq,
474 struct rq_list *rqlist)
475{
476 struct request *req;
477 unsigned long flags;
478 bool kick;
479
480 spin_lock_irqsave(&vq->lock, flags);
481
482 while ((req = rq_list_pop(rqlist))) {
483 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
484 int err;
485
486 err = virtblk_add_req(vq->vq, vbr);
487 if (err) {
488 virtblk_unmap_data(req, vbr);
489 virtblk_cleanup_cmd(req);
490 blk_mq_requeue_request(req, true);
491 }
492 }
493
494 kick = virtqueue_kick_prepare(vq->vq);
495 spin_unlock_irqrestore(&vq->lock, flags);
496
497 if (kick)
498 virtqueue_notify(vq->vq);
499}
500
501static void virtio_queue_rqs(struct rq_list *rqlist)
502{
503 struct rq_list submit_list = { };
504 struct rq_list requeue_list = { };
505 struct virtio_blk_vq *vq = NULL;
506 struct request *req;
507
508 while ((req = rq_list_pop(rqlist))) {
509 struct virtio_blk_vq *this_vq = get_virtio_blk_vq(req->mq_hctx);
510
511 if (vq && vq != this_vq)
512 virtblk_add_req_batch(vq, &submit_list);
513 vq = this_vq;
514
515 if (virtblk_prep_rq_batch(req))
516 rq_list_add_tail(&submit_list, req);
517 else
518 rq_list_add_tail(&requeue_list, req);
519 }
520
521 if (vq)
522 virtblk_add_req_batch(vq, &submit_list);
523 *rqlist = requeue_list;
524}
525
526#ifdef CONFIG_BLK_DEV_ZONED
527static void *virtblk_alloc_report_buffer(struct virtio_blk *vblk,
528 unsigned int nr_zones,
529 size_t *buflen)
530{
531 struct request_queue *q = vblk->disk->queue;
532 size_t bufsize;
533 void *buf;
534
535 nr_zones = min_t(unsigned int, nr_zones,
536 get_capacity(vblk->disk) >> ilog2(vblk->zone_sectors));
537
538 bufsize = sizeof(struct virtio_blk_zone_report) +
539 nr_zones * sizeof(struct virtio_blk_zone_descriptor);
540 bufsize = min_t(size_t, bufsize,
541 queue_max_hw_sectors(q) << SECTOR_SHIFT);
542 bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
543
544 while (bufsize >= sizeof(struct virtio_blk_zone_report)) {
545 buf = __vmalloc(bufsize, GFP_KERNEL | __GFP_NORETRY);
546 if (buf) {
547 *buflen = bufsize;
548 return buf;
549 }
550 bufsize >>= 1;
551 }
552
553 return NULL;
554}
555
556static int virtblk_submit_zone_report(struct virtio_blk *vblk,
557 char *report_buf, size_t report_len,
558 sector_t sector)
559{
560 struct request_queue *q = vblk->disk->queue;
561 struct request *req;
562 struct virtblk_req *vbr;
563 int err;
564
565 req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
566 if (IS_ERR(req))
567 return PTR_ERR(req);
568
569 vbr = blk_mq_rq_to_pdu(req);
570 vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
571 vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_ZONE_REPORT);
572 vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, sector);
573
574 err = blk_rq_map_kern(req, report_buf, report_len, GFP_KERNEL);
575 if (err)
576 goto out;
577
578 blk_execute_rq(req, false);
579 err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
580out:
581 blk_mq_free_request(req);
582 return err;
583}
584
585static int virtblk_parse_zone(struct virtio_blk *vblk,
586 struct virtio_blk_zone_descriptor *entry,
587 unsigned int idx,
588 struct blk_report_zones_args *args)
589{
590 struct blk_zone zone = { };
591
592 zone.start = virtio64_to_cpu(vblk->vdev, entry->z_start);
593 if (zone.start + vblk->zone_sectors <= get_capacity(vblk->disk))
594 zone.len = vblk->zone_sectors;
595 else
596 zone.len = get_capacity(vblk->disk) - zone.start;
597 zone.capacity = virtio64_to_cpu(vblk->vdev, entry->z_cap);
598 zone.wp = virtio64_to_cpu(vblk->vdev, entry->z_wp);
599
600 switch (entry->z_type) {
601 case VIRTIO_BLK_ZT_SWR:
602 zone.type = BLK_ZONE_TYPE_SEQWRITE_REQ;
603 break;
604 case VIRTIO_BLK_ZT_SWP:
605 zone.type = BLK_ZONE_TYPE_SEQWRITE_PREF;
606 break;
607 case VIRTIO_BLK_ZT_CONV:
608 zone.type = BLK_ZONE_TYPE_CONVENTIONAL;
609 break;
610 default:
611 dev_err(&vblk->vdev->dev, "zone %llu: invalid type %#x\n",
612 zone.start, entry->z_type);
613 return -EIO;
614 }
615
616 switch (entry->z_state) {
617 case VIRTIO_BLK_ZS_EMPTY:
618 zone.cond = BLK_ZONE_COND_EMPTY;
619 break;
620 case VIRTIO_BLK_ZS_CLOSED:
621 zone.cond = BLK_ZONE_COND_CLOSED;
622 break;
623 case VIRTIO_BLK_ZS_FULL:
624 zone.cond = BLK_ZONE_COND_FULL;
625 zone.wp = zone.start + zone.len;
626 break;
627 case VIRTIO_BLK_ZS_EOPEN:
628 zone.cond = BLK_ZONE_COND_EXP_OPEN;
629 break;
630 case VIRTIO_BLK_ZS_IOPEN:
631 zone.cond = BLK_ZONE_COND_IMP_OPEN;
632 break;
633 case VIRTIO_BLK_ZS_NOT_WP:
634 zone.cond = BLK_ZONE_COND_NOT_WP;
635 break;
636 case VIRTIO_BLK_ZS_RDONLY:
637 zone.cond = BLK_ZONE_COND_READONLY;
638 zone.wp = ULONG_MAX;
639 break;
640 case VIRTIO_BLK_ZS_OFFLINE:
641 zone.cond = BLK_ZONE_COND_OFFLINE;
642 zone.wp = ULONG_MAX;
643 break;
644 default:
645 dev_err(&vblk->vdev->dev, "zone %llu: invalid condition %#x\n",
646 zone.start, entry->z_state);
647 return -EIO;
648 }
649
650 /*
651 * The callback below checks the validity of the reported
652 * entry data, no need to further validate it here.
653 */
654 return disk_report_zone(vblk->disk, &zone, idx, args);
655}
656
657static int virtblk_report_zones(struct gendisk *disk, sector_t sector,
658 unsigned int nr_zones,
659 struct blk_report_zones_args *args)
660{
661 struct virtio_blk *vblk = disk->private_data;
662 struct virtio_blk_zone_report *report;
663 unsigned long long nz, i;
664 size_t buflen;
665 unsigned int zone_idx = 0;
666 int ret;
667
668 if (WARN_ON_ONCE(!vblk->zone_sectors))
669 return -EOPNOTSUPP;
670
671 report = virtblk_alloc_report_buffer(vblk, nr_zones, &buflen);
672 if (!report)
673 return -ENOMEM;
674
675 mutex_lock(&vblk->vdev_mutex);
676
677 if (!vblk->vdev) {
678 ret = -ENXIO;
679 goto fail_report;
680 }
681
682 while (zone_idx < nr_zones && sector < get_capacity(vblk->disk)) {
683 memset(report, 0, buflen);
684
685 ret = virtblk_submit_zone_report(vblk, (char *)report,
686 buflen, sector);
687 if (ret)
688 goto fail_report;
689
690 nz = min_t(u64, virtio64_to_cpu(vblk->vdev, report->nr_zones),
691 nr_zones);
692 nz = min_t(u64, nz,
693 (buflen - sizeof(*report)) / sizeof(report->zones[0]));
694 if (!nz)
695 break;
696
697 for (i = 0; i < nz && zone_idx < nr_zones; i++) {
698 ret = virtblk_parse_zone(vblk, &report->zones[i],
699 zone_idx, args);
700 if (ret)
701 goto fail_report;
702
703 sector = virtio64_to_cpu(vblk->vdev,
704 report->zones[i].z_start) +
705 vblk->zone_sectors;
706 zone_idx++;
707 }
708 }
709
710 if (zone_idx > 0)
711 ret = zone_idx;
712 else
713 ret = -EINVAL;
714fail_report:
715 mutex_unlock(&vblk->vdev_mutex);
716 kvfree(report);
717 return ret;
718}
719
720static int virtblk_read_zoned_limits(struct virtio_blk *vblk,
721 struct queue_limits *lim)
722{
723 struct virtio_device *vdev = vblk->vdev;
724 u32 v, wg;
725
726 dev_dbg(&vdev->dev, "probing host-managed zoned device\n");
727
728 lim->features |= BLK_FEAT_ZONED;
729
730 virtio_cread(vdev, struct virtio_blk_config,
731 zoned.max_open_zones, &v);
732 lim->max_open_zones = v;
733 dev_dbg(&vdev->dev, "max open zones = %u\n", v);
734
735 virtio_cread(vdev, struct virtio_blk_config,
736 zoned.max_active_zones, &v);
737 lim->max_active_zones = v;
738 dev_dbg(&vdev->dev, "max active zones = %u\n", v);
739
740 virtio_cread(vdev, struct virtio_blk_config,
741 zoned.write_granularity, &wg);
742 if (!wg) {
743 dev_warn(&vdev->dev, "zero write granularity reported\n");
744 return -ENODEV;
745 }
746 lim->zone_write_granularity = wg;
747 lim->physical_block_size = wg;
748 lim->io_min = wg;
749
750 dev_dbg(&vdev->dev, "write granularity = %u\n", wg);
751
752 /*
753 * virtio ZBD specification doesn't require zones to be a power of
754 * two sectors in size, but the code in this driver expects that.
755 */
756 virtio_cread(vdev, struct virtio_blk_config, zoned.zone_sectors,
757 &vblk->zone_sectors);
758 if (vblk->zone_sectors == 0 || !is_power_of_2(vblk->zone_sectors)) {
759 dev_err(&vdev->dev,
760 "zoned device with non power of two zone size %u\n",
761 vblk->zone_sectors);
762 return -ENODEV;
763 }
764 lim->chunk_sectors = vblk->zone_sectors;
765 dev_dbg(&vdev->dev, "zone sectors = %u\n", vblk->zone_sectors);
766
767 if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
768 dev_warn(&vblk->vdev->dev,
769 "ignoring negotiated F_DISCARD for zoned device\n");
770 lim->max_hw_discard_sectors = 0;
771 }
772
773 virtio_cread(vdev, struct virtio_blk_config,
774 zoned.max_append_sectors, &v);
775 if (!v) {
776 dev_warn(&vdev->dev, "zero max_append_sectors reported\n");
777 return -ENODEV;
778 }
779 if ((v << SECTOR_SHIFT) < wg) {
780 dev_err(&vdev->dev,
781 "write granularity %u exceeds max_append_sectors %u limit\n",
782 wg, v);
783 return -ENODEV;
784 }
785 lim->max_hw_zone_append_sectors = v;
786 dev_dbg(&vdev->dev, "max append sectors = %u\n", v);
787
788 return 0;
789}
790#else
791/*
792 * Zoned block device support is not configured in this kernel, host-managed
793 * zoned devices can't be supported.
794 */
795#define virtblk_report_zones NULL
796static inline int virtblk_read_zoned_limits(struct virtio_blk *vblk,
797 struct queue_limits *lim)
798{
799 dev_err(&vblk->vdev->dev,
800 "virtio_blk: zoned devices are not supported");
801 return -EOPNOTSUPP;
802}
803#endif /* CONFIG_BLK_DEV_ZONED */
804
805/* return id (s/n) string for *disk to *id_str
806 */
807static int virtblk_get_id(struct gendisk *disk, char *id_str)
808{
809 struct virtio_blk *vblk = disk->private_data;
810 struct request_queue *q = vblk->disk->queue;
811 struct request *req;
812 struct virtblk_req *vbr;
813 int err;
814
815 req = blk_mq_alloc_request(q, REQ_OP_DRV_IN, 0);
816 if (IS_ERR(req))
817 return PTR_ERR(req);
818
819 vbr = blk_mq_rq_to_pdu(req);
820 vbr->in_hdr_len = sizeof(vbr->in_hdr.status);
821 vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
822 vbr->out_hdr.sector = 0;
823
824 err = blk_rq_map_kern(req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
825 if (err)
826 goto out;
827
828 blk_execute_rq(req, false);
829 err = blk_status_to_errno(virtblk_result(vbr->in_hdr.status));
830out:
831 blk_mq_free_request(req);
832 return err;
833}
834
835/* We provide getgeo only to please some old bootloader/partitioning tools */
836static int virtblk_getgeo(struct gendisk *disk, struct hd_geometry *geo)
837{
838 struct virtio_blk *vblk = disk->private_data;
839 int ret = 0;
840
841 mutex_lock(&vblk->vdev_mutex);
842
843 if (!vblk->vdev) {
844 ret = -ENXIO;
845 goto out;
846 }
847
848 /* see if the host passed in geometry config */
849 if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
850 virtio_cread(vblk->vdev, struct virtio_blk_config,
851 geometry.cylinders, &geo->cylinders);
852 virtio_cread(vblk->vdev, struct virtio_blk_config,
853 geometry.heads, &geo->heads);
854 virtio_cread(vblk->vdev, struct virtio_blk_config,
855 geometry.sectors, &geo->sectors);
856 } else {
857 /* some standard values, similar to sd */
858 geo->heads = 1 << 6;
859 geo->sectors = 1 << 5;
860 geo->cylinders = get_capacity(disk) >> 11;
861 }
862out:
863 mutex_unlock(&vblk->vdev_mutex);
864 return ret;
865}
866
867static void virtblk_free_disk(struct gendisk *disk)
868{
869 struct virtio_blk *vblk = disk->private_data;
870
871 ida_free(&vd_index_ida, vblk->index);
872 mutex_destroy(&vblk->vdev_mutex);
873 kfree(vblk);
874}
875
876static const struct block_device_operations virtblk_fops = {
877 .owner = THIS_MODULE,
878 .getgeo = virtblk_getgeo,
879 .free_disk = virtblk_free_disk,
880 .report_zones = virtblk_report_zones,
881};
882
883static int index_to_minor(int index)
884{
885 return index << PART_BITS;
886}
887
888static int minor_to_index(int minor)
889{
890 return minor >> PART_BITS;
891}
892
893static ssize_t serial_show(struct device *dev,
894 struct device_attribute *attr, char *buf)
895{
896 struct gendisk *disk = dev_to_disk(dev);
897 int err;
898
899 /* sysfs gives us a PAGE_SIZE buffer */
900 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
901
902 buf[VIRTIO_BLK_ID_BYTES] = '\0';
903 err = virtblk_get_id(disk, buf);
904 if (!err)
905 return strlen(buf);
906
907 if (err == -EIO) /* Unsupported? Make it empty. */
908 return 0;
909
910 return err;
911}
912
913static DEVICE_ATTR_RO(serial);
914
915/* The queue's logical block size must be set before calling this */
916static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
917{
918 struct virtio_device *vdev = vblk->vdev;
919 struct request_queue *q = vblk->disk->queue;
920 char cap_str_2[10], cap_str_10[10];
921 unsigned long long nblocks;
922 u64 capacity;
923
924 /* Host must always specify the capacity. */
925 virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
926
927 nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
928
929 string_get_size(nblocks, queue_logical_block_size(q),
930 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
931 string_get_size(nblocks, queue_logical_block_size(q),
932 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
933
934 dev_notice(&vdev->dev,
935 "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
936 vblk->disk->disk_name,
937 resize ? "new size: " : "",
938 nblocks,
939 queue_logical_block_size(q),
940 cap_str_10,
941 cap_str_2);
942
943 set_capacity_and_notify(vblk->disk, capacity);
944}
945
946static void virtblk_config_changed_work(struct work_struct *work)
947{
948 struct virtio_blk *vblk =
949 container_of(work, struct virtio_blk, config_work);
950
951 virtblk_update_capacity(vblk, true);
952}
953
954static void virtblk_config_changed(struct virtio_device *vdev)
955{
956 struct virtio_blk *vblk = vdev->priv;
957
958 queue_work(virtblk_wq, &vblk->config_work);
959}
960
961static int init_vq(struct virtio_blk *vblk)
962{
963 int err;
964 unsigned short i;
965 struct virtqueue_info *vqs_info;
966 struct virtqueue **vqs;
967 unsigned short num_vqs;
968 unsigned short num_poll_vqs;
969 struct virtio_device *vdev = vblk->vdev;
970 struct irq_affinity desc = { 0, };
971
972 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
973 struct virtio_blk_config, num_queues,
974 &num_vqs);
975 if (err)
976 num_vqs = 1;
977
978 if (!err && !num_vqs) {
979 dev_err(&vdev->dev, "MQ advertised but zero queues reported\n");
980 return -EINVAL;
981 }
982
983 num_vqs = blk_mq_num_possible_queues(
984 min_not_zero(num_request_queues, num_vqs));
985
986 num_poll_vqs = min_t(unsigned int, poll_queues, num_vqs - 1);
987
988 vblk->io_queues[HCTX_TYPE_DEFAULT] = num_vqs - num_poll_vqs;
989 vblk->io_queues[HCTX_TYPE_READ] = 0;
990 vblk->io_queues[HCTX_TYPE_POLL] = num_poll_vqs;
991
992 dev_info(&vdev->dev, "%d/%d/%d default/read/poll queues\n",
993 vblk->io_queues[HCTX_TYPE_DEFAULT],
994 vblk->io_queues[HCTX_TYPE_READ],
995 vblk->io_queues[HCTX_TYPE_POLL]);
996
997 vblk->vqs = kmalloc_objs(*vblk->vqs, num_vqs);
998 if (!vblk->vqs)
999 return -ENOMEM;
1000
1001 vqs_info = kzalloc_objs(*vqs_info, num_vqs);
1002 vqs = kmalloc_objs(*vqs, num_vqs);
1003 if (!vqs_info || !vqs) {
1004 err = -ENOMEM;
1005 goto out;
1006 }
1007
1008 for (i = 0; i < num_vqs - num_poll_vqs; i++) {
1009 vqs_info[i].callback = virtblk_done;
1010 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%u", i);
1011 vqs_info[i].name = vblk->vqs[i].name;
1012 }
1013
1014 for (; i < num_vqs; i++) {
1015 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req_poll.%u", i);
1016 vqs_info[i].name = vblk->vqs[i].name;
1017 }
1018
1019 /* Discover virtqueues and write information to configuration. */
1020 err = virtio_find_vqs(vdev, num_vqs, vqs, vqs_info, &desc);
1021 if (err)
1022 goto out;
1023
1024 for (i = 0; i < num_vqs; i++) {
1025 spin_lock_init(&vblk->vqs[i].lock);
1026 vblk->vqs[i].vq = vqs[i];
1027 }
1028 vblk->num_vqs = num_vqs;
1029
1030out:
1031 kfree(vqs);
1032 kfree(vqs_info);
1033 if (err) {
1034 kfree(vblk->vqs);
1035 /*
1036 * Set to NULL to prevent freeing vqs again during freezing.
1037 */
1038 vblk->vqs = NULL;
1039 }
1040 return err;
1041}
1042
1043/*
1044 * Legacy naming scheme used for virtio devices. We are stuck with it for
1045 * virtio blk but don't ever use it for any new driver.
1046 */
1047static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
1048{
1049 const int base = 'z' - 'a' + 1;
1050 char *begin = buf + strlen(prefix);
1051 char *end = buf + buflen;
1052 char *p;
1053 int unit;
1054
1055 p = end - 1;
1056 *p = '\0';
1057 unit = base;
1058 do {
1059 if (p == begin)
1060 return -EINVAL;
1061 *--p = 'a' + (index % unit);
1062 index = (index / unit) - 1;
1063 } while (index >= 0);
1064
1065 memmove(begin, p, end - p);
1066 memcpy(buf, prefix, strlen(prefix));
1067
1068 return 0;
1069}
1070
1071static int virtblk_get_cache_mode(struct virtio_device *vdev)
1072{
1073 u8 writeback;
1074 int err;
1075
1076 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
1077 struct virtio_blk_config, wce,
1078 &writeback);
1079
1080 /*
1081 * If WCE is not configurable and flush is not available,
1082 * assume no writeback cache is in use.
1083 */
1084 if (err)
1085 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
1086
1087 return writeback;
1088}
1089
1090static const char *const virtblk_cache_types[] = {
1091 "write through", "write back"
1092};
1093
1094static ssize_t
1095cache_type_store(struct device *dev, struct device_attribute *attr,
1096 const char *buf, size_t count)
1097{
1098 struct gendisk *disk = dev_to_disk(dev);
1099 struct virtio_blk *vblk = disk->private_data;
1100 struct virtio_device *vdev = vblk->vdev;
1101 struct queue_limits lim;
1102 int i;
1103
1104 BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
1105 i = sysfs_match_string(virtblk_cache_types, buf);
1106 if (i < 0)
1107 return i;
1108
1109 virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
1110
1111 lim = queue_limits_start_update(disk->queue);
1112 if (virtblk_get_cache_mode(vdev))
1113 lim.features |= BLK_FEAT_WRITE_CACHE;
1114 else
1115 lim.features &= ~BLK_FEAT_WRITE_CACHE;
1116 i = queue_limits_commit_update_frozen(disk->queue, &lim);
1117 if (i)
1118 return i;
1119 return count;
1120}
1121
1122static ssize_t
1123cache_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
1133static DEVICE_ATTR_RW(cache_type);
1134
1135static struct attribute *virtblk_attrs[] = {
1136 &dev_attr_serial.attr,
1137 &dev_attr_cache_type.attr,
1138 NULL,
1139};
1140
1141static 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
1156static const struct attribute_group virtblk_attr_group = {
1157 .attrs = virtblk_attrs,
1158 .is_visible = virtblk_attrs_are_visible,
1159};
1160
1161static const struct attribute_group *virtblk_attr_groups[] = {
1162 &virtblk_attr_group,
1163 NULL,
1164};
1165
1166static 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_map_hw_queues(&set->map[i],
1190 &vblk->vdev->dev, 0);
1191 }
1192}
1193
1194static void virtblk_complete_batch(struct io_comp_batch *iob)
1195{
1196 struct request *req;
1197
1198 rq_list_for_each(&iob->req_list, req) {
1199 virtblk_unmap_data(req, blk_mq_rq_to_pdu(req));
1200 virtblk_cleanup_cmd(req);
1201 }
1202 blk_mq_end_request_batch(iob);
1203}
1204
1205static int virtblk_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1206{
1207 struct virtio_blk *vblk = hctx->queue->queuedata;
1208 struct virtio_blk_vq *vq = get_virtio_blk_vq(hctx);
1209 struct virtblk_req *vbr;
1210 unsigned long flags;
1211 unsigned int len;
1212 int found = 0;
1213
1214 spin_lock_irqsave(&vq->lock, flags);
1215
1216 while ((vbr = virtqueue_get_buf(vq->vq, &len)) != NULL) {
1217 struct request *req = blk_mq_rq_from_pdu(vbr);
1218 u8 status = virtblk_vbr_status(vbr);
1219
1220 found++;
1221 if (!blk_mq_complete_request_remote(req) &&
1222 !blk_mq_add_to_batch(req, iob, status != VIRTIO_BLK_S_OK,
1223 virtblk_complete_batch))
1224 virtblk_request_done(req);
1225 }
1226
1227 if (found)
1228 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
1229
1230 spin_unlock_irqrestore(&vq->lock, flags);
1231
1232 return found;
1233}
1234
1235static const struct blk_mq_ops virtio_mq_ops = {
1236 .queue_rq = virtio_queue_rq,
1237 .queue_rqs = virtio_queue_rqs,
1238 .commit_rqs = virtio_commit_rqs,
1239 .complete = virtblk_request_done,
1240 .map_queues = virtblk_map_queues,
1241 .poll = virtblk_poll,
1242};
1243
1244static unsigned int virtblk_queue_depth;
1245module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
1246
1247static int virtblk_read_limits(struct virtio_blk *vblk,
1248 struct queue_limits *lim)
1249{
1250 struct virtio_device *vdev = vblk->vdev;
1251 u32 v, max_size, sg_elems, opt_io_size;
1252 u32 max_discard_segs = 0;
1253 u32 discard_granularity = 0;
1254 u16 min_io_size;
1255 u8 physical_block_exp, alignment_offset;
1256 size_t max_dma_size;
1257 int err;
1258
1259 /* We need to know how many segments before we allocate. */
1260 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
1261 struct virtio_blk_config, seg_max,
1262 &sg_elems);
1263
1264 /* We need at least one SG element, whatever they say. */
1265 if (err || !sg_elems)
1266 sg_elems = 1;
1267
1268 /* Prevent integer overflows and honor max vq size */
1269 sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
1270
1271 /* We can handle whatever the host told us to handle. */
1272 lim->max_segments = sg_elems;
1273
1274 /* No real sector limit. */
1275 lim->max_hw_sectors = UINT_MAX;
1276
1277 max_dma_size = virtio_max_dma_size(vdev);
1278 max_size = max_dma_size > U32_MAX ? U32_MAX : max_dma_size;
1279
1280 /* Host can optionally specify maximum segment size and number of
1281 * segments. */
1282 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
1283 struct virtio_blk_config, size_max, &v);
1284 if (!err)
1285 max_size = min(max_size, v);
1286
1287 lim->max_segment_size = max_size;
1288
1289 /* Host can optionally specify the block size of the device */
1290 virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
1291 struct virtio_blk_config, blk_size,
1292 &lim->logical_block_size);
1293
1294 /* Use topology information if available */
1295 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1296 struct virtio_blk_config, physical_block_exp,
1297 &physical_block_exp);
1298 if (!err && physical_block_exp)
1299 lim->physical_block_size =
1300 lim->logical_block_size * (1 << physical_block_exp);
1301
1302 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1303 struct virtio_blk_config, alignment_offset,
1304 &alignment_offset);
1305 if (!err && alignment_offset)
1306 lim->alignment_offset =
1307 lim->logical_block_size * alignment_offset;
1308
1309 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1310 struct virtio_blk_config, min_io_size,
1311 &min_io_size);
1312 if (!err && min_io_size)
1313 lim->io_min = lim->logical_block_size * min_io_size;
1314
1315 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
1316 struct virtio_blk_config, opt_io_size,
1317 &opt_io_size);
1318 if (!err && opt_io_size)
1319 lim->io_opt = lim->logical_block_size * opt_io_size;
1320
1321 if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
1322 virtio_cread(vdev, struct virtio_blk_config,
1323 discard_sector_alignment, &discard_granularity);
1324
1325 virtio_cread(vdev, struct virtio_blk_config,
1326 max_discard_sectors, &v);
1327 lim->max_hw_discard_sectors = v ? v : UINT_MAX;
1328
1329 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
1330 &max_discard_segs);
1331 }
1332
1333 if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
1334 virtio_cread(vdev, struct virtio_blk_config,
1335 max_write_zeroes_sectors, &v);
1336 lim->max_write_zeroes_sectors = v ? v : UINT_MAX;
1337 }
1338
1339 /* The discard and secure erase limits are combined since the Linux
1340 * block layer uses the same limit for both commands.
1341 *
1342 * If both VIRTIO_BLK_F_SECURE_ERASE and VIRTIO_BLK_F_DISCARD features
1343 * are negotiated, we will use the minimum between the limits.
1344 *
1345 * discard sector alignment is set to the minimum between discard_sector_alignment
1346 * and secure_erase_sector_alignment.
1347 *
1348 * max discard sectors is set to the minimum between max_discard_seg and
1349 * max_secure_erase_seg.
1350 */
1351 if (virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1352
1353 virtio_cread(vdev, struct virtio_blk_config,
1354 secure_erase_sector_alignment, &v);
1355
1356 /* secure_erase_sector_alignment should not be zero, the device should set a
1357 * valid number of sectors.
1358 */
1359 if (!v) {
1360 dev_err(&vdev->dev,
1361 "virtio_blk: secure_erase_sector_alignment can't be 0\n");
1362 return -EINVAL;
1363 }
1364
1365 discard_granularity = min_not_zero(discard_granularity, v);
1366
1367 virtio_cread(vdev, struct virtio_blk_config,
1368 max_secure_erase_sectors, &v);
1369
1370 /* max_secure_erase_sectors should not be zero, the device should set a
1371 * valid number of sectors.
1372 */
1373 if (!v) {
1374 dev_err(&vdev->dev,
1375 "virtio_blk: max_secure_erase_sectors can't be 0\n");
1376 return -EINVAL;
1377 }
1378
1379 lim->max_secure_erase_sectors = v;
1380
1381 virtio_cread(vdev, struct virtio_blk_config,
1382 max_secure_erase_seg, &v);
1383
1384 /* max_secure_erase_seg should not be zero, the device should set a
1385 * valid number of segments
1386 */
1387 if (!v) {
1388 dev_err(&vdev->dev,
1389 "virtio_blk: max_secure_erase_seg can't be 0\n");
1390 return -EINVAL;
1391 }
1392
1393 max_discard_segs = min_not_zero(max_discard_segs, v);
1394 }
1395
1396 if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD) ||
1397 virtio_has_feature(vdev, VIRTIO_BLK_F_SECURE_ERASE)) {
1398 /* max_discard_seg and discard_granularity will be 0 only
1399 * if max_discard_seg and discard_sector_alignment fields in the virtio
1400 * config are 0 and VIRTIO_BLK_F_SECURE_ERASE feature is not negotiated.
1401 * In this case, we use default values.
1402 */
1403 if (!max_discard_segs)
1404 max_discard_segs = sg_elems;
1405
1406 lim->max_discard_segments =
1407 min(max_discard_segs, MAX_DISCARD_SEGMENTS);
1408
1409 if (discard_granularity)
1410 lim->discard_granularity =
1411 discard_granularity << SECTOR_SHIFT;
1412 else
1413 lim->discard_granularity = lim->logical_block_size;
1414 }
1415
1416 if (virtio_has_feature(vdev, VIRTIO_BLK_F_ZONED)) {
1417 u8 model;
1418
1419 virtio_cread(vdev, struct virtio_blk_config, zoned.model, &model);
1420 switch (model) {
1421 case VIRTIO_BLK_Z_NONE:
1422 case VIRTIO_BLK_Z_HA:
1423 /* treat host-aware devices as non-zoned */
1424 return 0;
1425 case VIRTIO_BLK_Z_HM:
1426 err = virtblk_read_zoned_limits(vblk, lim);
1427 if (err)
1428 return err;
1429 break;
1430 default:
1431 dev_err(&vdev->dev, "unsupported zone model %d\n", model);
1432 return -EINVAL;
1433 }
1434 }
1435
1436 return 0;
1437}
1438
1439static int virtblk_probe(struct virtio_device *vdev)
1440{
1441 struct virtio_blk *vblk;
1442 struct queue_limits lim = {
1443 .features = BLK_FEAT_ROTATIONAL,
1444 .logical_block_size = SECTOR_SIZE,
1445 };
1446 int err, index;
1447 unsigned int queue_depth;
1448
1449 if (!vdev->config->get) {
1450 dev_err(&vdev->dev, "%s failure: config access disabled\n",
1451 __func__);
1452 return -EINVAL;
1453 }
1454
1455 err = ida_alloc_range(&vd_index_ida, 0,
1456 minor_to_index(1 << MINORBITS) - 1, GFP_KERNEL);
1457 if (err < 0)
1458 goto out;
1459 index = err;
1460
1461 vdev->priv = vblk = kmalloc_obj(*vblk);
1462 if (!vblk) {
1463 err = -ENOMEM;
1464 goto out_free_index;
1465 }
1466
1467 mutex_init(&vblk->vdev_mutex);
1468
1469 vblk->vdev = vdev;
1470
1471 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
1472
1473 err = init_vq(vblk);
1474 if (err)
1475 goto out_free_vblk;
1476
1477 /* Default queue sizing is to fill the ring. */
1478 if (!virtblk_queue_depth) {
1479 queue_depth = vblk->vqs[0].vq->num_free;
1480 /* ... but without indirect descs, we use 2 descs per req */
1481 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
1482 queue_depth /= 2;
1483 } else {
1484 queue_depth = virtblk_queue_depth;
1485 }
1486
1487 memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
1488 vblk->tag_set.ops = &virtio_mq_ops;
1489 vblk->tag_set.queue_depth = queue_depth;
1490 vblk->tag_set.numa_node = NUMA_NO_NODE;
1491 vblk->tag_set.cmd_size =
1492 sizeof(struct virtblk_req) +
1493 sizeof(struct scatterlist) * VIRTIO_BLK_INLINE_SG_CNT;
1494 vblk->tag_set.driver_data = vblk;
1495 vblk->tag_set.nr_hw_queues = vblk->num_vqs;
1496 vblk->tag_set.nr_maps = 1;
1497 if (vblk->io_queues[HCTX_TYPE_POLL])
1498 vblk->tag_set.nr_maps = 3;
1499
1500 err = blk_mq_alloc_tag_set(&vblk->tag_set);
1501 if (err)
1502 goto out_free_vq;
1503
1504 err = virtblk_read_limits(vblk, &lim);
1505 if (err)
1506 goto out_free_tags;
1507
1508 if (virtblk_get_cache_mode(vdev))
1509 lim.features |= BLK_FEAT_WRITE_CACHE;
1510
1511 vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, &lim, vblk);
1512 if (IS_ERR(vblk->disk)) {
1513 err = PTR_ERR(vblk->disk);
1514 goto out_free_tags;
1515 }
1516
1517 virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
1518
1519 vblk->disk->major = major;
1520 vblk->disk->first_minor = index_to_minor(index);
1521 vblk->disk->minors = 1 << PART_BITS;
1522 vblk->disk->private_data = vblk;
1523 vblk->disk->fops = &virtblk_fops;
1524 vblk->index = index;
1525
1526 /* If disk is read-only in the host, the guest should obey */
1527 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
1528 set_disk_ro(vblk->disk, 1);
1529
1530 virtblk_update_capacity(vblk, false);
1531 virtio_device_ready(vdev);
1532
1533 /*
1534 * All steps that follow use the VQs therefore they need to be
1535 * placed after the virtio_device_ready() call above.
1536 */
1537 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
1538 (lim.features & BLK_FEAT_ZONED)) {
1539 err = blk_revalidate_disk_zones(vblk->disk);
1540 if (err)
1541 goto out_cleanup_disk;
1542 }
1543
1544 err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
1545 if (err)
1546 goto out_cleanup_disk;
1547
1548 return 0;
1549
1550out_cleanup_disk:
1551 put_disk(vblk->disk);
1552out_free_tags:
1553 blk_mq_free_tag_set(&vblk->tag_set);
1554out_free_vq:
1555 vdev->config->del_vqs(vdev);
1556 kfree(vblk->vqs);
1557out_free_vblk:
1558 kfree(vblk);
1559out_free_index:
1560 ida_free(&vd_index_ida, index);
1561out:
1562 return err;
1563}
1564
1565static void virtblk_remove(struct virtio_device *vdev)
1566{
1567 struct virtio_blk *vblk = vdev->priv;
1568
1569 /* Make sure no work handler is accessing the device. */
1570 flush_work(&vblk->config_work);
1571
1572 del_gendisk(vblk->disk);
1573 blk_mq_free_tag_set(&vblk->tag_set);
1574
1575 mutex_lock(&vblk->vdev_mutex);
1576
1577 /* Stop all the virtqueues. */
1578 virtio_reset_device(vdev);
1579
1580 /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
1581 vblk->vdev = NULL;
1582
1583 vdev->config->del_vqs(vdev);
1584 kfree(vblk->vqs);
1585
1586 mutex_unlock(&vblk->vdev_mutex);
1587
1588 put_disk(vblk->disk);
1589}
1590
1591static int virtblk_freeze_priv(struct virtio_device *vdev)
1592{
1593 struct virtio_blk *vblk = vdev->priv;
1594 struct request_queue *q = vblk->disk->queue;
1595 unsigned int memflags;
1596
1597 /* Ensure no requests in virtqueues before deleting vqs. */
1598 memflags = blk_mq_freeze_queue(q);
1599 blk_mq_quiesce_queue_nowait(q);
1600 blk_mq_unfreeze_queue(q, memflags);
1601
1602 /* Ensure we don't receive any more interrupts */
1603 virtio_reset_device(vdev);
1604
1605 /* Make sure no work handler is accessing the device. */
1606 flush_work(&vblk->config_work);
1607
1608 vdev->config->del_vqs(vdev);
1609 kfree(vblk->vqs);
1610 /*
1611 * Set to NULL to prevent freeing vqs again after a failed vqs
1612 * allocation during resume. Note that kfree() already handles NULL
1613 * pointers safely.
1614 */
1615 vblk->vqs = NULL;
1616
1617 return 0;
1618}
1619
1620static int virtblk_restore_priv(struct virtio_device *vdev)
1621{
1622 struct virtio_blk *vblk = vdev->priv;
1623 int ret;
1624
1625 ret = init_vq(vdev->priv);
1626 if (ret)
1627 return ret;
1628
1629 virtio_device_ready(vdev);
1630 blk_mq_unquiesce_queue(vblk->disk->queue);
1631
1632 return 0;
1633}
1634
1635#ifdef CONFIG_PM_SLEEP
1636static int virtblk_freeze(struct virtio_device *vdev)
1637{
1638 return virtblk_freeze_priv(vdev);
1639}
1640
1641static int virtblk_restore(struct virtio_device *vdev)
1642{
1643 return virtblk_restore_priv(vdev);
1644}
1645#endif
1646
1647static int virtblk_reset_prepare(struct virtio_device *vdev)
1648{
1649 return virtblk_freeze_priv(vdev);
1650}
1651
1652static int virtblk_reset_done(struct virtio_device *vdev)
1653{
1654 return virtblk_restore_priv(vdev);
1655}
1656
1657static const struct virtio_device_id id_table[] = {
1658 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
1659 { 0 },
1660};
1661
1662static unsigned int features_legacy[] = {
1663 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1664 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1665 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1666 VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1667 VIRTIO_BLK_F_SECURE_ERASE,
1668}
1669;
1670static unsigned int features[] = {
1671 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
1672 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
1673 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
1674 VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
1675 VIRTIO_BLK_F_SECURE_ERASE, VIRTIO_BLK_F_ZONED,
1676};
1677
1678static struct virtio_driver virtio_blk = {
1679 .feature_table = features,
1680 .feature_table_size = ARRAY_SIZE(features),
1681 .feature_table_legacy = features_legacy,
1682 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
1683 .driver.name = KBUILD_MODNAME,
1684 .id_table = id_table,
1685 .probe = virtblk_probe,
1686 .remove = virtblk_remove,
1687 .config_changed = virtblk_config_changed,
1688#ifdef CONFIG_PM_SLEEP
1689 .freeze = virtblk_freeze,
1690 .restore = virtblk_restore,
1691#endif
1692 .reset_prepare = virtblk_reset_prepare,
1693 .reset_done = virtblk_reset_done,
1694};
1695
1696static int __init virtio_blk_init(void)
1697{
1698 int error;
1699
1700 virtblk_wq = alloc_workqueue("virtio-blk", WQ_PERCPU, 0);
1701 if (!virtblk_wq)
1702 return -ENOMEM;
1703
1704 major = register_blkdev(0, "virtblk");
1705 if (major < 0) {
1706 error = major;
1707 goto out_destroy_workqueue;
1708 }
1709
1710 error = register_virtio_driver(&virtio_blk);
1711 if (error)
1712 goto out_unregister_blkdev;
1713 return 0;
1714
1715out_unregister_blkdev:
1716 unregister_blkdev(major, "virtblk");
1717out_destroy_workqueue:
1718 destroy_workqueue(virtblk_wq);
1719 return error;
1720}
1721
1722static void __exit virtio_blk_fini(void)
1723{
1724 unregister_virtio_driver(&virtio_blk);
1725 unregister_blkdev(major, "virtblk");
1726 destroy_workqueue(virtblk_wq);
1727}
1728module_init(virtio_blk_init);
1729module_exit(virtio_blk_fini);
1730
1731MODULE_DEVICE_TABLE(virtio, id_table);
1732MODULE_DESCRIPTION("Virtio block driver");
1733MODULE_LICENSE("GPL");