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 <uapi/linux/virtio_ring.h>
19
20 #define PART_BITS 4
21 #define VQ_NAME_LEN 16
22 #define MAX_DISCARD_SEGMENTS 256u
23
24 /* The maximum number of sg elements that fit into a virtqueue */
25 #define VIRTIO_BLK_MAX_SG_ELEMS 32768
26
27 static int major;
28 static DEFINE_IDA(vd_index_ida);
29
30 static struct workqueue_struct *virtblk_wq;
31
32 struct virtio_blk_vq {
33 struct virtqueue *vq;
34 spinlock_t lock;
35 char name[VQ_NAME_LEN];
36 } ____cacheline_aligned_in_smp;
37
38 struct virtio_blk {
39 /*
40 * This mutex must be held by anything that may run after
41 * virtblk_remove() sets vblk->vdev to NULL.
42 *
43 * blk-mq, virtqueue processing, and sysfs attribute code paths are
44 * shut down before vblk->vdev is set to NULL and therefore do not need
45 * to hold this mutex.
46 */
47 struct mutex vdev_mutex;
48 struct virtio_device *vdev;
49
50 /* The disk structure for the kernel. */
51 struct gendisk *disk;
52
53 /* Block layer tags. */
54 struct blk_mq_tag_set tag_set;
55
56 /* Process context for config space updates */
57 struct work_struct config_work;
58
59 /*
60 * Tracks references from block_device_operations open/release and
61 * virtio_driver probe/remove so this object can be freed once no
62 * longer in use.
63 */
64 refcount_t refs;
65
66 /* What host tells us, plus 2 for header & tailer. */
67 unsigned int sg_elems;
68
69 /* Ida index - used to track minor number allocations. */
70 int index;
71
72 /* num of vqs */
73 int num_vqs;
74 struct virtio_blk_vq *vqs;
75 };
76
77 struct virtblk_req {
78 struct virtio_blk_outhdr out_hdr;
79 u8 status;
80 struct scatterlist sg[];
81 };
82
virtblk_result(struct virtblk_req * vbr)83 static inline blk_status_t virtblk_result(struct virtblk_req *vbr)
84 {
85 switch (vbr->status) {
86 case VIRTIO_BLK_S_OK:
87 return BLK_STS_OK;
88 case VIRTIO_BLK_S_UNSUPP:
89 return BLK_STS_NOTSUPP;
90 default:
91 return BLK_STS_IOERR;
92 }
93 }
94
virtblk_add_req(struct virtqueue * vq,struct virtblk_req * vbr,struct scatterlist * data_sg,bool have_data)95 static int virtblk_add_req(struct virtqueue *vq, struct virtblk_req *vbr,
96 struct scatterlist *data_sg, bool have_data)
97 {
98 struct scatterlist hdr, status, *sgs[3];
99 unsigned int num_out = 0, num_in = 0;
100
101 sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
102 sgs[num_out++] = &hdr;
103
104 if (have_data) {
105 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
106 sgs[num_out++] = data_sg;
107 else
108 sgs[num_out + num_in++] = data_sg;
109 }
110
111 sg_init_one(&status, &vbr->status, sizeof(vbr->status));
112 sgs[num_out + num_in++] = &status;
113
114 return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
115 }
116
virtblk_setup_discard_write_zeroes(struct request * req,bool unmap)117 static int virtblk_setup_discard_write_zeroes(struct request *req, bool unmap)
118 {
119 unsigned short segments = blk_rq_nr_discard_segments(req);
120 unsigned short n = 0;
121 struct virtio_blk_discard_write_zeroes *range;
122 struct bio *bio;
123 u32 flags = 0;
124
125 if (unmap)
126 flags |= VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP;
127
128 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
129 if (!range)
130 return -ENOMEM;
131
132 /*
133 * Single max discard segment means multi-range discard isn't
134 * supported, and block layer only runs contiguity merge like
135 * normal RW request. So we can't reply on bio for retrieving
136 * each range info.
137 */
138 if (queue_max_discard_segments(req->q) == 1) {
139 range[0].flags = cpu_to_le32(flags);
140 range[0].num_sectors = cpu_to_le32(blk_rq_sectors(req));
141 range[0].sector = cpu_to_le64(blk_rq_pos(req));
142 n = 1;
143 } else {
144 __rq_for_each_bio(bio, req) {
145 u64 sector = bio->bi_iter.bi_sector;
146 u32 num_sectors = bio->bi_iter.bi_size >> SECTOR_SHIFT;
147
148 range[n].flags = cpu_to_le32(flags);
149 range[n].num_sectors = cpu_to_le32(num_sectors);
150 range[n].sector = cpu_to_le64(sector);
151 n++;
152 }
153 }
154
155 WARN_ON_ONCE(n != segments);
156
157 req->special_vec.bv_page = virt_to_page(range);
158 req->special_vec.bv_offset = offset_in_page(range);
159 req->special_vec.bv_len = sizeof(*range) * segments;
160 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
161
162 return 0;
163 }
164
virtblk_request_done(struct request * req)165 static inline void virtblk_request_done(struct request *req)
166 {
167 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
168
169 if (req->rq_flags & RQF_SPECIAL_PAYLOAD)
170 kfree(bvec_virt(&req->special_vec));
171 blk_mq_end_request(req, virtblk_result(vbr));
172 }
173
virtblk_done(struct virtqueue * vq)174 static void virtblk_done(struct virtqueue *vq)
175 {
176 struct virtio_blk *vblk = vq->vdev->priv;
177 bool req_done = false;
178 int qid = vq->index;
179 struct virtblk_req *vbr;
180 unsigned long flags;
181 unsigned int len;
182
183 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
184 do {
185 virtqueue_disable_cb(vq);
186 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
187 struct request *req = blk_mq_rq_from_pdu(vbr);
188
189 if (likely(!blk_should_fake_timeout(req->q)))
190 blk_mq_complete_request(req);
191 req_done = true;
192 }
193 if (unlikely(virtqueue_is_broken(vq)))
194 break;
195 } while (!virtqueue_enable_cb(vq));
196
197 /* In case queue is stopped waiting for more buffers. */
198 if (req_done)
199 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
200 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
201 }
202
virtio_commit_rqs(struct blk_mq_hw_ctx * hctx)203 static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
204 {
205 struct virtio_blk *vblk = hctx->queue->queuedata;
206 struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
207 bool kick;
208
209 spin_lock_irq(&vq->lock);
210 kick = virtqueue_kick_prepare(vq->vq);
211 spin_unlock_irq(&vq->lock);
212
213 if (kick)
214 virtqueue_notify(vq->vq);
215 }
216
virtio_queue_rq(struct blk_mq_hw_ctx * hctx,const struct blk_mq_queue_data * bd)217 static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
218 const struct blk_mq_queue_data *bd)
219 {
220 struct virtio_blk *vblk = hctx->queue->queuedata;
221 struct request *req = bd->rq;
222 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
223 unsigned long flags;
224 unsigned int num;
225 int qid = hctx->queue_num;
226 int err;
227 bool notify = false;
228 bool unmap = false;
229 u32 type;
230
231 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
232
233 switch (req_op(req)) {
234 case REQ_OP_READ:
235 case REQ_OP_WRITE:
236 type = 0;
237 break;
238 case REQ_OP_FLUSH:
239 type = VIRTIO_BLK_T_FLUSH;
240 break;
241 case REQ_OP_DISCARD:
242 type = VIRTIO_BLK_T_DISCARD;
243 break;
244 case REQ_OP_WRITE_ZEROES:
245 type = VIRTIO_BLK_T_WRITE_ZEROES;
246 unmap = !(req->cmd_flags & REQ_NOUNMAP);
247 break;
248 case REQ_OP_DRV_IN:
249 type = VIRTIO_BLK_T_GET_ID;
250 break;
251 default:
252 WARN_ON_ONCE(1);
253 return BLK_STS_IOERR;
254 }
255
256 vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
257 vbr->out_hdr.sector = type ?
258 0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
259 vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
260
261 blk_mq_start_request(req);
262
263 if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
264 err = virtblk_setup_discard_write_zeroes(req, unmap);
265 if (err)
266 return BLK_STS_RESOURCE;
267 }
268
269 num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
270 if (num) {
271 if (rq_data_dir(req) == WRITE)
272 vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
273 else
274 vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
275 }
276
277 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
278 err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
279 if (err) {
280 virtqueue_kick(vblk->vqs[qid].vq);
281 /* Don't stop the queue if -ENOMEM: we may have failed to
282 * bounce the buffer due to global resource outage.
283 */
284 if (err == -ENOSPC)
285 blk_mq_stop_hw_queue(hctx);
286 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
287 switch (err) {
288 case -ENOSPC:
289 return BLK_STS_DEV_RESOURCE;
290 case -ENOMEM:
291 return BLK_STS_RESOURCE;
292 default:
293 return BLK_STS_IOERR;
294 }
295 }
296
297 if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
298 notify = true;
299 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
300
301 if (notify)
302 virtqueue_notify(vblk->vqs[qid].vq);
303 return BLK_STS_OK;
304 }
305
306 /* return id (s/n) string for *disk to *id_str
307 */
virtblk_get_id(struct gendisk * disk,char * id_str)308 static int virtblk_get_id(struct gendisk *disk, char *id_str)
309 {
310 struct virtio_blk *vblk = disk->private_data;
311 struct request_queue *q = vblk->disk->queue;
312 struct request *req;
313 int err;
314
315 req = blk_get_request(q, REQ_OP_DRV_IN, 0);
316 if (IS_ERR(req))
317 return PTR_ERR(req);
318
319 err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
320 if (err)
321 goto out;
322
323 blk_execute_rq(vblk->disk, req, false);
324 err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
325 out:
326 blk_put_request(req);
327 return err;
328 }
329
virtblk_get(struct virtio_blk * vblk)330 static void virtblk_get(struct virtio_blk *vblk)
331 {
332 refcount_inc(&vblk->refs);
333 }
334
virtblk_put(struct virtio_blk * vblk)335 static void virtblk_put(struct virtio_blk *vblk)
336 {
337 if (refcount_dec_and_test(&vblk->refs)) {
338 ida_simple_remove(&vd_index_ida, vblk->index);
339 mutex_destroy(&vblk->vdev_mutex);
340 kfree(vblk);
341 }
342 }
343
virtblk_open(struct block_device * bd,fmode_t mode)344 static int virtblk_open(struct block_device *bd, fmode_t mode)
345 {
346 struct virtio_blk *vblk = bd->bd_disk->private_data;
347 int ret = 0;
348
349 mutex_lock(&vblk->vdev_mutex);
350
351 if (vblk->vdev)
352 virtblk_get(vblk);
353 else
354 ret = -ENXIO;
355
356 mutex_unlock(&vblk->vdev_mutex);
357 return ret;
358 }
359
virtblk_release(struct gendisk * disk,fmode_t mode)360 static void virtblk_release(struct gendisk *disk, fmode_t mode)
361 {
362 struct virtio_blk *vblk = disk->private_data;
363
364 virtblk_put(vblk);
365 }
366
367 /* We provide getgeo only to please some old bootloader/partitioning tools */
virtblk_getgeo(struct block_device * bd,struct hd_geometry * geo)368 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
369 {
370 struct virtio_blk *vblk = bd->bd_disk->private_data;
371 int ret = 0;
372
373 mutex_lock(&vblk->vdev_mutex);
374
375 if (!vblk->vdev) {
376 ret = -ENXIO;
377 goto out;
378 }
379
380 /* see if the host passed in geometry config */
381 if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
382 virtio_cread(vblk->vdev, struct virtio_blk_config,
383 geometry.cylinders, &geo->cylinders);
384 virtio_cread(vblk->vdev, struct virtio_blk_config,
385 geometry.heads, &geo->heads);
386 virtio_cread(vblk->vdev, struct virtio_blk_config,
387 geometry.sectors, &geo->sectors);
388 } else {
389 /* some standard values, similar to sd */
390 geo->heads = 1 << 6;
391 geo->sectors = 1 << 5;
392 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
393 }
394 out:
395 mutex_unlock(&vblk->vdev_mutex);
396 return ret;
397 }
398
399 static const struct block_device_operations virtblk_fops = {
400 .owner = THIS_MODULE,
401 .open = virtblk_open,
402 .release = virtblk_release,
403 .getgeo = virtblk_getgeo,
404 };
405
index_to_minor(int index)406 static int index_to_minor(int index)
407 {
408 return index << PART_BITS;
409 }
410
minor_to_index(int minor)411 static int minor_to_index(int minor)
412 {
413 return minor >> PART_BITS;
414 }
415
serial_show(struct device * dev,struct device_attribute * attr,char * buf)416 static ssize_t serial_show(struct device *dev,
417 struct device_attribute *attr, char *buf)
418 {
419 struct gendisk *disk = dev_to_disk(dev);
420 int err;
421
422 /* sysfs gives us a PAGE_SIZE buffer */
423 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
424
425 buf[VIRTIO_BLK_ID_BYTES] = '\0';
426 err = virtblk_get_id(disk, buf);
427 if (!err)
428 return strlen(buf);
429
430 if (err == -EIO) /* Unsupported? Make it empty. */
431 return 0;
432
433 return err;
434 }
435
436 static DEVICE_ATTR_RO(serial);
437
438 /* The queue's logical block size must be set before calling this */
virtblk_update_capacity(struct virtio_blk * vblk,bool resize)439 static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
440 {
441 struct virtio_device *vdev = vblk->vdev;
442 struct request_queue *q = vblk->disk->queue;
443 char cap_str_2[10], cap_str_10[10];
444 unsigned long long nblocks;
445 u64 capacity;
446
447 /* Host must always specify the capacity. */
448 virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
449
450 nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
451
452 string_get_size(nblocks, queue_logical_block_size(q),
453 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
454 string_get_size(nblocks, queue_logical_block_size(q),
455 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
456
457 dev_notice(&vdev->dev,
458 "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
459 vblk->disk->disk_name,
460 resize ? "new size: " : "",
461 nblocks,
462 queue_logical_block_size(q),
463 cap_str_10,
464 cap_str_2);
465
466 set_capacity_and_notify(vblk->disk, capacity);
467 }
468
virtblk_config_changed_work(struct work_struct * work)469 static void virtblk_config_changed_work(struct work_struct *work)
470 {
471 struct virtio_blk *vblk =
472 container_of(work, struct virtio_blk, config_work);
473
474 virtblk_update_capacity(vblk, true);
475 }
476
virtblk_config_changed(struct virtio_device * vdev)477 static void virtblk_config_changed(struct virtio_device *vdev)
478 {
479 struct virtio_blk *vblk = vdev->priv;
480
481 queue_work(virtblk_wq, &vblk->config_work);
482 }
483
init_vq(struct virtio_blk * vblk)484 static int init_vq(struct virtio_blk *vblk)
485 {
486 int err;
487 int i;
488 vq_callback_t **callbacks;
489 const char **names;
490 struct virtqueue **vqs;
491 unsigned short num_vqs;
492 struct virtio_device *vdev = vblk->vdev;
493 struct irq_affinity desc = { 0, };
494
495 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
496 struct virtio_blk_config, num_queues,
497 &num_vqs);
498 if (err)
499 num_vqs = 1;
500
501 num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
502
503 vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
504 if (!vblk->vqs)
505 return -ENOMEM;
506
507 names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
508 callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
509 vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
510 if (!names || !callbacks || !vqs) {
511 err = -ENOMEM;
512 goto out;
513 }
514
515 for (i = 0; i < num_vqs; i++) {
516 callbacks[i] = virtblk_done;
517 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
518 names[i] = vblk->vqs[i].name;
519 }
520
521 /* Discover virtqueues and write information to configuration. */
522 err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
523 if (err)
524 goto out;
525
526 for (i = 0; i < num_vqs; i++) {
527 spin_lock_init(&vblk->vqs[i].lock);
528 vblk->vqs[i].vq = vqs[i];
529 }
530 vblk->num_vqs = num_vqs;
531
532 out:
533 kfree(vqs);
534 kfree(callbacks);
535 kfree(names);
536 if (err)
537 kfree(vblk->vqs);
538 return err;
539 }
540
541 /*
542 * Legacy naming scheme used for virtio devices. We are stuck with it for
543 * virtio blk but don't ever use it for any new driver.
544 */
virtblk_name_format(char * prefix,int index,char * buf,int buflen)545 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
546 {
547 const int base = 'z' - 'a' + 1;
548 char *begin = buf + strlen(prefix);
549 char *end = buf + buflen;
550 char *p;
551 int unit;
552
553 p = end - 1;
554 *p = '\0';
555 unit = base;
556 do {
557 if (p == begin)
558 return -EINVAL;
559 *--p = 'a' + (index % unit);
560 index = (index / unit) - 1;
561 } while (index >= 0);
562
563 memmove(begin, p, end - p);
564 memcpy(buf, prefix, strlen(prefix));
565
566 return 0;
567 }
568
virtblk_get_cache_mode(struct virtio_device * vdev)569 static int virtblk_get_cache_mode(struct virtio_device *vdev)
570 {
571 u8 writeback;
572 int err;
573
574 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
575 struct virtio_blk_config, wce,
576 &writeback);
577
578 /*
579 * If WCE is not configurable and flush is not available,
580 * assume no writeback cache is in use.
581 */
582 if (err)
583 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
584
585 return writeback;
586 }
587
virtblk_update_cache_mode(struct virtio_device * vdev)588 static void virtblk_update_cache_mode(struct virtio_device *vdev)
589 {
590 u8 writeback = virtblk_get_cache_mode(vdev);
591 struct virtio_blk *vblk = vdev->priv;
592
593 blk_queue_write_cache(vblk->disk->queue, writeback, false);
594 }
595
596 static const char *const virtblk_cache_types[] = {
597 "write through", "write back"
598 };
599
600 static ssize_t
cache_type_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)601 cache_type_store(struct device *dev, struct device_attribute *attr,
602 const char *buf, size_t count)
603 {
604 struct gendisk *disk = dev_to_disk(dev);
605 struct virtio_blk *vblk = disk->private_data;
606 struct virtio_device *vdev = vblk->vdev;
607 int i;
608
609 BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
610 i = sysfs_match_string(virtblk_cache_types, buf);
611 if (i < 0)
612 return i;
613
614 virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
615 virtblk_update_cache_mode(vdev);
616 return count;
617 }
618
619 static ssize_t
cache_type_show(struct device * dev,struct device_attribute * attr,char * buf)620 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
621 {
622 struct gendisk *disk = dev_to_disk(dev);
623 struct virtio_blk *vblk = disk->private_data;
624 u8 writeback = virtblk_get_cache_mode(vblk->vdev);
625
626 BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
627 return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
628 }
629
630 static DEVICE_ATTR_RW(cache_type);
631
632 static struct attribute *virtblk_attrs[] = {
633 &dev_attr_serial.attr,
634 &dev_attr_cache_type.attr,
635 NULL,
636 };
637
virtblk_attrs_are_visible(struct kobject * kobj,struct attribute * a,int n)638 static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
639 struct attribute *a, int n)
640 {
641 struct device *dev = kobj_to_dev(kobj);
642 struct gendisk *disk = dev_to_disk(dev);
643 struct virtio_blk *vblk = disk->private_data;
644 struct virtio_device *vdev = vblk->vdev;
645
646 if (a == &dev_attr_cache_type.attr &&
647 !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
648 return S_IRUGO;
649
650 return a->mode;
651 }
652
653 static const struct attribute_group virtblk_attr_group = {
654 .attrs = virtblk_attrs,
655 .is_visible = virtblk_attrs_are_visible,
656 };
657
658 static const struct attribute_group *virtblk_attr_groups[] = {
659 &virtblk_attr_group,
660 NULL,
661 };
662
virtblk_init_request(struct blk_mq_tag_set * set,struct request * rq,unsigned int hctx_idx,unsigned int numa_node)663 static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
664 unsigned int hctx_idx, unsigned int numa_node)
665 {
666 struct virtio_blk *vblk = set->driver_data;
667 struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
668
669 sg_init_table(vbr->sg, vblk->sg_elems);
670 return 0;
671 }
672
virtblk_map_queues(struct blk_mq_tag_set * set)673 static int virtblk_map_queues(struct blk_mq_tag_set *set)
674 {
675 struct virtio_blk *vblk = set->driver_data;
676
677 return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
678 vblk->vdev, 0);
679 }
680
681 static const struct blk_mq_ops virtio_mq_ops = {
682 .queue_rq = virtio_queue_rq,
683 .commit_rqs = virtio_commit_rqs,
684 .complete = virtblk_request_done,
685 .init_request = virtblk_init_request,
686 .map_queues = virtblk_map_queues,
687 };
688
689 static unsigned int virtblk_queue_depth;
690 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
691
virtblk_probe(struct virtio_device * vdev)692 static int virtblk_probe(struct virtio_device *vdev)
693 {
694 struct virtio_blk *vblk;
695 struct request_queue *q;
696 int err, index;
697
698 u32 v, blk_size, max_size, sg_elems, opt_io_size;
699 u16 min_io_size;
700 u8 physical_block_exp, alignment_offset;
701 unsigned int queue_depth;
702
703 if (!vdev->config->get) {
704 dev_err(&vdev->dev, "%s failure: config access disabled\n",
705 __func__);
706 return -EINVAL;
707 }
708
709 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
710 GFP_KERNEL);
711 if (err < 0)
712 goto out;
713 index = err;
714
715 /* We need to know how many segments before we allocate. */
716 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
717 struct virtio_blk_config, seg_max,
718 &sg_elems);
719
720 /* We need at least one SG element, whatever they say. */
721 if (err || !sg_elems)
722 sg_elems = 1;
723
724 /* Prevent integer overflows and honor max vq size */
725 sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
726
727 /* We need extra sg elements at head and tail. */
728 sg_elems += 2;
729 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
730 if (!vblk) {
731 err = -ENOMEM;
732 goto out_free_index;
733 }
734
735 /* This reference is dropped in virtblk_remove(). */
736 refcount_set(&vblk->refs, 1);
737 mutex_init(&vblk->vdev_mutex);
738
739 vblk->vdev = vdev;
740 vblk->sg_elems = sg_elems;
741
742 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
743
744 err = init_vq(vblk);
745 if (err)
746 goto out_free_vblk;
747
748 /* Default queue sizing is to fill the ring. */
749 if (!virtblk_queue_depth) {
750 queue_depth = vblk->vqs[0].vq->num_free;
751 /* ... but without indirect descs, we use 2 descs per req */
752 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
753 queue_depth /= 2;
754 } else {
755 queue_depth = virtblk_queue_depth;
756 }
757
758 memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
759 vblk->tag_set.ops = &virtio_mq_ops;
760 vblk->tag_set.queue_depth = queue_depth;
761 vblk->tag_set.numa_node = NUMA_NO_NODE;
762 vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
763 vblk->tag_set.cmd_size =
764 sizeof(struct virtblk_req) +
765 sizeof(struct scatterlist) * sg_elems;
766 vblk->tag_set.driver_data = vblk;
767 vblk->tag_set.nr_hw_queues = vblk->num_vqs;
768
769 err = blk_mq_alloc_tag_set(&vblk->tag_set);
770 if (err)
771 goto out_free_vq;
772
773 vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
774 if (IS_ERR(vblk->disk)) {
775 err = PTR_ERR(vblk->disk);
776 goto out_free_tags;
777 }
778 q = vblk->disk->queue;
779
780 virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
781
782 vblk->disk->major = major;
783 vblk->disk->first_minor = index_to_minor(index);
784 vblk->disk->minors = 1 << PART_BITS;
785 vblk->disk->private_data = vblk;
786 vblk->disk->fops = &virtblk_fops;
787 vblk->disk->flags |= GENHD_FL_EXT_DEVT;
788 vblk->index = index;
789
790 /* configure queue flush support */
791 virtblk_update_cache_mode(vdev);
792
793 /* If disk is read-only in the host, the guest should obey */
794 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
795 set_disk_ro(vblk->disk, 1);
796
797 /* We can handle whatever the host told us to handle. */
798 blk_queue_max_segments(q, vblk->sg_elems-2);
799
800 /* No real sector limit. */
801 blk_queue_max_hw_sectors(q, -1U);
802
803 max_size = virtio_max_dma_size(vdev);
804
805 /* Host can optionally specify maximum segment size and number of
806 * segments. */
807 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
808 struct virtio_blk_config, size_max, &v);
809 if (!err)
810 max_size = min(max_size, v);
811
812 blk_queue_max_segment_size(q, max_size);
813
814 /* Host can optionally specify the block size of the device */
815 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
816 struct virtio_blk_config, blk_size,
817 &blk_size);
818 if (!err)
819 blk_queue_logical_block_size(q, blk_size);
820 else
821 blk_size = queue_logical_block_size(q);
822
823 /* Use topology information if available */
824 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
825 struct virtio_blk_config, physical_block_exp,
826 &physical_block_exp);
827 if (!err && physical_block_exp)
828 blk_queue_physical_block_size(q,
829 blk_size * (1 << physical_block_exp));
830
831 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
832 struct virtio_blk_config, alignment_offset,
833 &alignment_offset);
834 if (!err && alignment_offset)
835 blk_queue_alignment_offset(q, blk_size * alignment_offset);
836
837 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
838 struct virtio_blk_config, min_io_size,
839 &min_io_size);
840 if (!err && min_io_size)
841 blk_queue_io_min(q, blk_size * min_io_size);
842
843 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
844 struct virtio_blk_config, opt_io_size,
845 &opt_io_size);
846 if (!err && opt_io_size)
847 blk_queue_io_opt(q, blk_size * opt_io_size);
848
849 if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
850 q->limits.discard_granularity = blk_size;
851
852 virtio_cread(vdev, struct virtio_blk_config,
853 discard_sector_alignment, &v);
854 q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
855
856 virtio_cread(vdev, struct virtio_blk_config,
857 max_discard_sectors, &v);
858 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
859
860 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
861 &v);
862 blk_queue_max_discard_segments(q,
863 min_not_zero(v,
864 MAX_DISCARD_SEGMENTS));
865
866 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
867 }
868
869 if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
870 virtio_cread(vdev, struct virtio_blk_config,
871 max_write_zeroes_sectors, &v);
872 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
873 }
874
875 virtblk_update_capacity(vblk, false);
876 virtio_device_ready(vdev);
877
878 err = device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
879 if (err)
880 goto out_cleanup_disk;
881
882 return 0;
883
884 out_cleanup_disk:
885 blk_cleanup_disk(vblk->disk);
886 out_free_tags:
887 blk_mq_free_tag_set(&vblk->tag_set);
888 out_free_vq:
889 vdev->config->del_vqs(vdev);
890 kfree(vblk->vqs);
891 out_free_vblk:
892 kfree(vblk);
893 out_free_index:
894 ida_simple_remove(&vd_index_ida, index);
895 out:
896 return err;
897 }
898
virtblk_remove(struct virtio_device * vdev)899 static void virtblk_remove(struct virtio_device *vdev)
900 {
901 struct virtio_blk *vblk = vdev->priv;
902
903 /* Make sure no work handler is accessing the device. */
904 flush_work(&vblk->config_work);
905
906 del_gendisk(vblk->disk);
907 blk_cleanup_disk(vblk->disk);
908 blk_mq_free_tag_set(&vblk->tag_set);
909
910 mutex_lock(&vblk->vdev_mutex);
911
912 /* Stop all the virtqueues. */
913 vdev->config->reset(vdev);
914
915 /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
916 vblk->vdev = NULL;
917
918 vdev->config->del_vqs(vdev);
919 kfree(vblk->vqs);
920
921 mutex_unlock(&vblk->vdev_mutex);
922
923 virtblk_put(vblk);
924 }
925
926 #ifdef CONFIG_PM_SLEEP
virtblk_freeze(struct virtio_device * vdev)927 static int virtblk_freeze(struct virtio_device *vdev)
928 {
929 struct virtio_blk *vblk = vdev->priv;
930
931 /* Ensure we don't receive any more interrupts */
932 vdev->config->reset(vdev);
933
934 /* Make sure no work handler is accessing the device. */
935 flush_work(&vblk->config_work);
936
937 blk_mq_quiesce_queue(vblk->disk->queue);
938
939 vdev->config->del_vqs(vdev);
940 kfree(vblk->vqs);
941
942 return 0;
943 }
944
virtblk_restore(struct virtio_device * vdev)945 static int virtblk_restore(struct virtio_device *vdev)
946 {
947 struct virtio_blk *vblk = vdev->priv;
948 int ret;
949
950 ret = init_vq(vdev->priv);
951 if (ret)
952 return ret;
953
954 virtio_device_ready(vdev);
955
956 blk_mq_unquiesce_queue(vblk->disk->queue);
957 return 0;
958 }
959 #endif
960
961 static const struct virtio_device_id id_table[] = {
962 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
963 { 0 },
964 };
965
966 static unsigned int features_legacy[] = {
967 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
968 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
969 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
970 VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
971 }
972 ;
973 static unsigned int features[] = {
974 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
975 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
976 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
977 VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
978 };
979
980 static struct virtio_driver virtio_blk = {
981 .feature_table = features,
982 .feature_table_size = ARRAY_SIZE(features),
983 .feature_table_legacy = features_legacy,
984 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
985 .driver.name = KBUILD_MODNAME,
986 .driver.owner = THIS_MODULE,
987 .id_table = id_table,
988 .probe = virtblk_probe,
989 .remove = virtblk_remove,
990 .config_changed = virtblk_config_changed,
991 #ifdef CONFIG_PM_SLEEP
992 .freeze = virtblk_freeze,
993 .restore = virtblk_restore,
994 #endif
995 };
996
init(void)997 static int __init init(void)
998 {
999 int error;
1000
1001 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1002 if (!virtblk_wq)
1003 return -ENOMEM;
1004
1005 major = register_blkdev(0, "virtblk");
1006 if (major < 0) {
1007 error = major;
1008 goto out_destroy_workqueue;
1009 }
1010
1011 error = register_virtio_driver(&virtio_blk);
1012 if (error)
1013 goto out_unregister_blkdev;
1014 return 0;
1015
1016 out_unregister_blkdev:
1017 unregister_blkdev(major, "virtblk");
1018 out_destroy_workqueue:
1019 destroy_workqueue(virtblk_wq);
1020 return error;
1021 }
1022
fini(void)1023 static void __exit fini(void)
1024 {
1025 unregister_virtio_driver(&virtio_blk);
1026 unregister_blkdev(major, "virtblk");
1027 destroy_workqueue(virtblk_wq);
1028 }
1029 module_init(init);
1030 module_exit(fini);
1031
1032 MODULE_DEVICE_TABLE(virtio, id_table);
1033 MODULE_DESCRIPTION("Virtio block driver");
1034 MODULE_LICENSE("GPL");
1035
