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