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