Loading...
1//#define DEBUG
2#include <linux/spinlock.h>
3#include <linux/slab.h>
4#include <linux/blkdev.h>
5#include <linux/hdreg.h>
6#include <linux/virtio.h>
7#include <linux/virtio_blk.h>
8#include <linux/scatterlist.h>
9#include <linux/string_helpers.h>
10#include <scsi/scsi_cmnd.h>
11
12#define PART_BITS 4
13
14static int major, index;
15struct workqueue_struct *virtblk_wq;
16
17struct virtio_blk
18{
19 spinlock_t lock;
20
21 struct virtio_device *vdev;
22 struct virtqueue *vq;
23
24 /* The disk structure for the kernel. */
25 struct gendisk *disk;
26
27 /* Request tracking. */
28 struct list_head reqs;
29
30 mempool_t *pool;
31
32 /* Process context for config space updates */
33 struct work_struct config_work;
34
35 /* What host tells us, plus 2 for header & tailer. */
36 unsigned int sg_elems;
37
38 /* Scatterlist: can be too big for stack. */
39 struct scatterlist sg[/*sg_elems*/];
40};
41
42struct virtblk_req
43{
44 struct list_head list;
45 struct request *req;
46 struct virtio_blk_outhdr out_hdr;
47 struct virtio_scsi_inhdr in_hdr;
48 u8 status;
49};
50
51static void blk_done(struct virtqueue *vq)
52{
53 struct virtio_blk *vblk = vq->vdev->priv;
54 struct virtblk_req *vbr;
55 unsigned int len;
56 unsigned long flags;
57
58 spin_lock_irqsave(&vblk->lock, flags);
59 while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
60 int error;
61
62 switch (vbr->status) {
63 case VIRTIO_BLK_S_OK:
64 error = 0;
65 break;
66 case VIRTIO_BLK_S_UNSUPP:
67 error = -ENOTTY;
68 break;
69 default:
70 error = -EIO;
71 break;
72 }
73
74 switch (vbr->req->cmd_type) {
75 case REQ_TYPE_BLOCK_PC:
76 vbr->req->resid_len = vbr->in_hdr.residual;
77 vbr->req->sense_len = vbr->in_hdr.sense_len;
78 vbr->req->errors = vbr->in_hdr.errors;
79 break;
80 case REQ_TYPE_SPECIAL:
81 vbr->req->errors = (error != 0);
82 break;
83 default:
84 break;
85 }
86
87 __blk_end_request_all(vbr->req, error);
88 list_del(&vbr->list);
89 mempool_free(vbr, vblk->pool);
90 }
91 /* In case queue is stopped waiting for more buffers. */
92 blk_start_queue(vblk->disk->queue);
93 spin_unlock_irqrestore(&vblk->lock, flags);
94}
95
96static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
97 struct request *req)
98{
99 unsigned long num, out = 0, in = 0;
100 struct virtblk_req *vbr;
101
102 vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
103 if (!vbr)
104 /* When another request finishes we'll try again. */
105 return false;
106
107 vbr->req = req;
108
109 if (req->cmd_flags & REQ_FLUSH) {
110 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
111 vbr->out_hdr.sector = 0;
112 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
113 } else {
114 switch (req->cmd_type) {
115 case REQ_TYPE_FS:
116 vbr->out_hdr.type = 0;
117 vbr->out_hdr.sector = blk_rq_pos(vbr->req);
118 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
119 break;
120 case REQ_TYPE_BLOCK_PC:
121 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
122 vbr->out_hdr.sector = 0;
123 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
124 break;
125 case REQ_TYPE_SPECIAL:
126 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
127 vbr->out_hdr.sector = 0;
128 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
129 break;
130 default:
131 /* We don't put anything else in the queue. */
132 BUG();
133 }
134 }
135
136 sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
137
138 /*
139 * If this is a packet command we need a couple of additional headers.
140 * Behind the normal outhdr we put a segment with the scsi command
141 * block, and before the normal inhdr we put the sense data and the
142 * inhdr with additional status information before the normal inhdr.
143 */
144 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
145 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
146
147 num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
148
149 if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
150 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
151 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
152 sizeof(vbr->in_hdr));
153 }
154
155 sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
156 sizeof(vbr->status));
157
158 if (num) {
159 if (rq_data_dir(vbr->req) == WRITE) {
160 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
161 out += num;
162 } else {
163 vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
164 in += num;
165 }
166 }
167
168 if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
169 mempool_free(vbr, vblk->pool);
170 return false;
171 }
172
173 list_add_tail(&vbr->list, &vblk->reqs);
174 return true;
175}
176
177static void do_virtblk_request(struct request_queue *q)
178{
179 struct virtio_blk *vblk = q->queuedata;
180 struct request *req;
181 unsigned int issued = 0;
182
183 while ((req = blk_peek_request(q)) != NULL) {
184 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
185
186 /* If this request fails, stop queue and wait for something to
187 finish to restart it. */
188 if (!do_req(q, vblk, req)) {
189 blk_stop_queue(q);
190 break;
191 }
192 blk_start_request(req);
193 issued++;
194 }
195
196 if (issued)
197 virtqueue_kick(vblk->vq);
198}
199
200/* return id (s/n) string for *disk to *id_str
201 */
202static int virtblk_get_id(struct gendisk *disk, char *id_str)
203{
204 struct virtio_blk *vblk = disk->private_data;
205 struct request *req;
206 struct bio *bio;
207 int err;
208
209 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
210 GFP_KERNEL);
211 if (IS_ERR(bio))
212 return PTR_ERR(bio);
213
214 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
215 if (IS_ERR(req)) {
216 bio_put(bio);
217 return PTR_ERR(req);
218 }
219
220 req->cmd_type = REQ_TYPE_SPECIAL;
221 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
222 blk_put_request(req);
223
224 return err;
225}
226
227static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
228 unsigned int cmd, unsigned long data)
229{
230 struct gendisk *disk = bdev->bd_disk;
231 struct virtio_blk *vblk = disk->private_data;
232
233 /*
234 * Only allow the generic SCSI ioctls if the host can support it.
235 */
236 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
237 return -ENOTTY;
238
239 return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
240 (void __user *)data);
241}
242
243/* We provide getgeo only to please some old bootloader/partitioning tools */
244static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
245{
246 struct virtio_blk *vblk = bd->bd_disk->private_data;
247 struct virtio_blk_geometry vgeo;
248 int err;
249
250 /* see if the host passed in geometry config */
251 err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
252 offsetof(struct virtio_blk_config, geometry),
253 &vgeo);
254
255 if (!err) {
256 geo->heads = vgeo.heads;
257 geo->sectors = vgeo.sectors;
258 geo->cylinders = vgeo.cylinders;
259 } else {
260 /* some standard values, similar to sd */
261 geo->heads = 1 << 6;
262 geo->sectors = 1 << 5;
263 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
264 }
265 return 0;
266}
267
268static const struct block_device_operations virtblk_fops = {
269 .ioctl = virtblk_ioctl,
270 .owner = THIS_MODULE,
271 .getgeo = virtblk_getgeo,
272};
273
274static int index_to_minor(int index)
275{
276 return index << PART_BITS;
277}
278
279static ssize_t virtblk_serial_show(struct device *dev,
280 struct device_attribute *attr, char *buf)
281{
282 struct gendisk *disk = dev_to_disk(dev);
283 int err;
284
285 /* sysfs gives us a PAGE_SIZE buffer */
286 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
287
288 buf[VIRTIO_BLK_ID_BYTES] = '\0';
289 err = virtblk_get_id(disk, buf);
290 if (!err)
291 return strlen(buf);
292
293 if (err == -EIO) /* Unsupported? Make it empty. */
294 return 0;
295
296 return err;
297}
298DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
299
300static void virtblk_config_changed_work(struct work_struct *work)
301{
302 struct virtio_blk *vblk =
303 container_of(work, struct virtio_blk, config_work);
304 struct virtio_device *vdev = vblk->vdev;
305 struct request_queue *q = vblk->disk->queue;
306 char cap_str_2[10], cap_str_10[10];
307 u64 capacity, size;
308
309 /* Host must always specify the capacity. */
310 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
311 &capacity, sizeof(capacity));
312
313 /* If capacity is too big, truncate with warning. */
314 if ((sector_t)capacity != capacity) {
315 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
316 (unsigned long long)capacity);
317 capacity = (sector_t)-1;
318 }
319
320 size = capacity * queue_logical_block_size(q);
321 string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
322 string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
323
324 dev_notice(&vdev->dev,
325 "new size: %llu %d-byte logical blocks (%s/%s)\n",
326 (unsigned long long)capacity,
327 queue_logical_block_size(q),
328 cap_str_10, cap_str_2);
329
330 set_capacity(vblk->disk, capacity);
331}
332
333static void virtblk_config_changed(struct virtio_device *vdev)
334{
335 struct virtio_blk *vblk = vdev->priv;
336
337 queue_work(virtblk_wq, &vblk->config_work);
338}
339
340static int __devinit virtblk_probe(struct virtio_device *vdev)
341{
342 struct virtio_blk *vblk;
343 struct request_queue *q;
344 int err;
345 u64 cap;
346 u32 v, blk_size, sg_elems, opt_io_size;
347 u16 min_io_size;
348 u8 physical_block_exp, alignment_offset;
349
350 if (index_to_minor(index) >= 1 << MINORBITS)
351 return -ENOSPC;
352
353 /* We need to know how many segments before we allocate. */
354 err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
355 offsetof(struct virtio_blk_config, seg_max),
356 &sg_elems);
357
358 /* We need at least one SG element, whatever they say. */
359 if (err || !sg_elems)
360 sg_elems = 1;
361
362 /* We need an extra sg elements at head and tail. */
363 sg_elems += 2;
364 vdev->priv = vblk = kmalloc(sizeof(*vblk) +
365 sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
366 if (!vblk) {
367 err = -ENOMEM;
368 goto out;
369 }
370
371 INIT_LIST_HEAD(&vblk->reqs);
372 spin_lock_init(&vblk->lock);
373 vblk->vdev = vdev;
374 vblk->sg_elems = sg_elems;
375 sg_init_table(vblk->sg, vblk->sg_elems);
376 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
377
378 /* We expect one virtqueue, for output. */
379 vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
380 if (IS_ERR(vblk->vq)) {
381 err = PTR_ERR(vblk->vq);
382 goto out_free_vblk;
383 }
384
385 vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
386 if (!vblk->pool) {
387 err = -ENOMEM;
388 goto out_free_vq;
389 }
390
391 /* FIXME: How many partitions? How long is a piece of string? */
392 vblk->disk = alloc_disk(1 << PART_BITS);
393 if (!vblk->disk) {
394 err = -ENOMEM;
395 goto out_mempool;
396 }
397
398 q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
399 if (!q) {
400 err = -ENOMEM;
401 goto out_put_disk;
402 }
403
404 q->queuedata = vblk;
405
406 if (index < 26) {
407 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
408 } else if (index < (26 + 1) * 26) {
409 sprintf(vblk->disk->disk_name, "vd%c%c",
410 'a' + index / 26 - 1, 'a' + index % 26);
411 } else {
412 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
413 const unsigned int m2 = (index / 26 - 1) % 26;
414 const unsigned int m3 = index % 26;
415 sprintf(vblk->disk->disk_name, "vd%c%c%c",
416 'a' + m1, 'a' + m2, 'a' + m3);
417 }
418
419 vblk->disk->major = major;
420 vblk->disk->first_minor = index_to_minor(index);
421 vblk->disk->private_data = vblk;
422 vblk->disk->fops = &virtblk_fops;
423 vblk->disk->driverfs_dev = &vdev->dev;
424 index++;
425
426 /* configure queue flush support */
427 if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH))
428 blk_queue_flush(q, REQ_FLUSH);
429
430 /* If disk is read-only in the host, the guest should obey */
431 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
432 set_disk_ro(vblk->disk, 1);
433
434 /* Host must always specify the capacity. */
435 vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
436 &cap, sizeof(cap));
437
438 /* If capacity is too big, truncate with warning. */
439 if ((sector_t)cap != cap) {
440 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
441 (unsigned long long)cap);
442 cap = (sector_t)-1;
443 }
444 set_capacity(vblk->disk, cap);
445
446 /* We can handle whatever the host told us to handle. */
447 blk_queue_max_segments(q, vblk->sg_elems-2);
448
449 /* No need to bounce any requests */
450 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
451
452 /* No real sector limit. */
453 blk_queue_max_hw_sectors(q, -1U);
454
455 /* Host can optionally specify maximum segment size and number of
456 * segments. */
457 err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
458 offsetof(struct virtio_blk_config, size_max),
459 &v);
460 if (!err)
461 blk_queue_max_segment_size(q, v);
462 else
463 blk_queue_max_segment_size(q, -1U);
464
465 /* Host can optionally specify the block size of the device */
466 err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
467 offsetof(struct virtio_blk_config, blk_size),
468 &blk_size);
469 if (!err)
470 blk_queue_logical_block_size(q, blk_size);
471 else
472 blk_size = queue_logical_block_size(q);
473
474 /* Use topology information if available */
475 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
476 offsetof(struct virtio_blk_config, physical_block_exp),
477 &physical_block_exp);
478 if (!err && physical_block_exp)
479 blk_queue_physical_block_size(q,
480 blk_size * (1 << physical_block_exp));
481
482 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
483 offsetof(struct virtio_blk_config, alignment_offset),
484 &alignment_offset);
485 if (!err && alignment_offset)
486 blk_queue_alignment_offset(q, blk_size * alignment_offset);
487
488 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
489 offsetof(struct virtio_blk_config, min_io_size),
490 &min_io_size);
491 if (!err && min_io_size)
492 blk_queue_io_min(q, blk_size * min_io_size);
493
494 err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
495 offsetof(struct virtio_blk_config, opt_io_size),
496 &opt_io_size);
497 if (!err && opt_io_size)
498 blk_queue_io_opt(q, blk_size * opt_io_size);
499
500
501 add_disk(vblk->disk);
502 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
503 if (err)
504 goto out_del_disk;
505
506 return 0;
507
508out_del_disk:
509 del_gendisk(vblk->disk);
510 blk_cleanup_queue(vblk->disk->queue);
511out_put_disk:
512 put_disk(vblk->disk);
513out_mempool:
514 mempool_destroy(vblk->pool);
515out_free_vq:
516 vdev->config->del_vqs(vdev);
517out_free_vblk:
518 kfree(vblk);
519out:
520 return err;
521}
522
523static void __devexit virtblk_remove(struct virtio_device *vdev)
524{
525 struct virtio_blk *vblk = vdev->priv;
526
527 flush_work(&vblk->config_work);
528
529 /* Nothing should be pending. */
530 BUG_ON(!list_empty(&vblk->reqs));
531
532 /* Stop all the virtqueues. */
533 vdev->config->reset(vdev);
534
535 del_gendisk(vblk->disk);
536 blk_cleanup_queue(vblk->disk->queue);
537 put_disk(vblk->disk);
538 mempool_destroy(vblk->pool);
539 vdev->config->del_vqs(vdev);
540 kfree(vblk);
541}
542
543static const struct virtio_device_id id_table[] = {
544 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
545 { 0 },
546};
547
548static unsigned int features[] = {
549 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
550 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
551 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
552};
553
554/*
555 * virtio_blk causes spurious section mismatch warning by
556 * simultaneously referring to a __devinit and a __devexit function.
557 * Use __refdata to avoid this warning.
558 */
559static struct virtio_driver __refdata virtio_blk = {
560 .feature_table = features,
561 .feature_table_size = ARRAY_SIZE(features),
562 .driver.name = KBUILD_MODNAME,
563 .driver.owner = THIS_MODULE,
564 .id_table = id_table,
565 .probe = virtblk_probe,
566 .remove = __devexit_p(virtblk_remove),
567 .config_changed = virtblk_config_changed,
568};
569
570static int __init init(void)
571{
572 int error;
573
574 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
575 if (!virtblk_wq)
576 return -ENOMEM;
577
578 major = register_blkdev(0, "virtblk");
579 if (major < 0) {
580 error = major;
581 goto out_destroy_workqueue;
582 }
583
584 error = register_virtio_driver(&virtio_blk);
585 if (error)
586 goto out_unregister_blkdev;
587 return 0;
588
589out_unregister_blkdev:
590 unregister_blkdev(major, "virtblk");
591out_destroy_workqueue:
592 destroy_workqueue(virtblk_wq);
593 return error;
594}
595
596static void __exit fini(void)
597{
598 unregister_blkdev(major, "virtblk");
599 unregister_virtio_driver(&virtio_blk);
600 destroy_workqueue(virtblk_wq);
601}
602module_init(init);
603module_exit(fini);
604
605MODULE_DEVICE_TABLE(virtio, id_table);
606MODULE_DESCRIPTION("Virtio block driver");
607MODULE_LICENSE("GPL");
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
27static int major;
28static DEFINE_IDA(vd_index_ida);
29
30static struct workqueue_struct *virtblk_wq;
31
32struct virtio_blk_vq {
33 struct virtqueue *vq;
34 spinlock_t lock;
35 char name[VQ_NAME_LEN];
36} ____cacheline_aligned_in_smp;
37
38struct 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
77struct virtblk_req {
78 struct virtio_blk_outhdr out_hdr;
79 u8 status;
80 struct scatterlist sg[];
81};
82
83static 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
95static 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
117static 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
165static 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(page_address(req->special_vec.bv_page) +
171 req->special_vec.bv_offset);
172 }
173
174 blk_mq_end_request(req, virtblk_result(vbr));
175}
176
177static void virtblk_done(struct virtqueue *vq)
178{
179 struct virtio_blk *vblk = vq->vdev->priv;
180 bool req_done = false;
181 int qid = vq->index;
182 struct virtblk_req *vbr;
183 unsigned long flags;
184 unsigned int len;
185
186 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
187 do {
188 virtqueue_disable_cb(vq);
189 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
190 struct request *req = blk_mq_rq_from_pdu(vbr);
191
192 if (likely(!blk_should_fake_timeout(req->q)))
193 blk_mq_complete_request(req);
194 req_done = true;
195 }
196 if (unlikely(virtqueue_is_broken(vq)))
197 break;
198 } while (!virtqueue_enable_cb(vq));
199
200 /* In case queue is stopped waiting for more buffers. */
201 if (req_done)
202 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
203 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
204}
205
206static void virtio_commit_rqs(struct blk_mq_hw_ctx *hctx)
207{
208 struct virtio_blk *vblk = hctx->queue->queuedata;
209 struct virtio_blk_vq *vq = &vblk->vqs[hctx->queue_num];
210 bool kick;
211
212 spin_lock_irq(&vq->lock);
213 kick = virtqueue_kick_prepare(vq->vq);
214 spin_unlock_irq(&vq->lock);
215
216 if (kick)
217 virtqueue_notify(vq->vq);
218}
219
220static blk_status_t virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
221 const struct blk_mq_queue_data *bd)
222{
223 struct virtio_blk *vblk = hctx->queue->queuedata;
224 struct request *req = bd->rq;
225 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
226 unsigned long flags;
227 unsigned int num;
228 int qid = hctx->queue_num;
229 int err;
230 bool notify = false;
231 bool unmap = false;
232 u32 type;
233
234 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
235
236 switch (req_op(req)) {
237 case REQ_OP_READ:
238 case REQ_OP_WRITE:
239 type = 0;
240 break;
241 case REQ_OP_FLUSH:
242 type = VIRTIO_BLK_T_FLUSH;
243 break;
244 case REQ_OP_DISCARD:
245 type = VIRTIO_BLK_T_DISCARD;
246 break;
247 case REQ_OP_WRITE_ZEROES:
248 type = VIRTIO_BLK_T_WRITE_ZEROES;
249 unmap = !(req->cmd_flags & REQ_NOUNMAP);
250 break;
251 case REQ_OP_DRV_IN:
252 type = VIRTIO_BLK_T_GET_ID;
253 break;
254 default:
255 WARN_ON_ONCE(1);
256 return BLK_STS_IOERR;
257 }
258
259 vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, type);
260 vbr->out_hdr.sector = type ?
261 0 : cpu_to_virtio64(vblk->vdev, blk_rq_pos(req));
262 vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(req));
263
264 blk_mq_start_request(req);
265
266 if (type == VIRTIO_BLK_T_DISCARD || type == VIRTIO_BLK_T_WRITE_ZEROES) {
267 err = virtblk_setup_discard_write_zeroes(req, unmap);
268 if (err)
269 return BLK_STS_RESOURCE;
270 }
271
272 num = blk_rq_map_sg(hctx->queue, req, vbr->sg);
273 if (num) {
274 if (rq_data_dir(req) == WRITE)
275 vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
276 else
277 vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
278 }
279
280 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
281 err = virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
282 if (err) {
283 virtqueue_kick(vblk->vqs[qid].vq);
284 /* Don't stop the queue if -ENOMEM: we may have failed to
285 * bounce the buffer due to global resource outage.
286 */
287 if (err == -ENOSPC)
288 blk_mq_stop_hw_queue(hctx);
289 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
290 switch (err) {
291 case -ENOSPC:
292 return BLK_STS_DEV_RESOURCE;
293 case -ENOMEM:
294 return BLK_STS_RESOURCE;
295 default:
296 return BLK_STS_IOERR;
297 }
298 }
299
300 if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
301 notify = true;
302 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
303
304 if (notify)
305 virtqueue_notify(vblk->vqs[qid].vq);
306 return BLK_STS_OK;
307}
308
309/* return id (s/n) string for *disk to *id_str
310 */
311static int virtblk_get_id(struct gendisk *disk, char *id_str)
312{
313 struct virtio_blk *vblk = disk->private_data;
314 struct request_queue *q = vblk->disk->queue;
315 struct request *req;
316 int err;
317
318 req = blk_get_request(q, REQ_OP_DRV_IN, 0);
319 if (IS_ERR(req))
320 return PTR_ERR(req);
321
322 err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
323 if (err)
324 goto out;
325
326 blk_execute_rq(vblk->disk, req, false);
327 err = blk_status_to_errno(virtblk_result(blk_mq_rq_to_pdu(req)));
328out:
329 blk_put_request(req);
330 return err;
331}
332
333static void virtblk_get(struct virtio_blk *vblk)
334{
335 refcount_inc(&vblk->refs);
336}
337
338static void virtblk_put(struct virtio_blk *vblk)
339{
340 if (refcount_dec_and_test(&vblk->refs)) {
341 ida_simple_remove(&vd_index_ida, vblk->index);
342 mutex_destroy(&vblk->vdev_mutex);
343 kfree(vblk);
344 }
345}
346
347static int virtblk_open(struct block_device *bd, fmode_t mode)
348{
349 struct virtio_blk *vblk = bd->bd_disk->private_data;
350 int ret = 0;
351
352 mutex_lock(&vblk->vdev_mutex);
353
354 if (vblk->vdev)
355 virtblk_get(vblk);
356 else
357 ret = -ENXIO;
358
359 mutex_unlock(&vblk->vdev_mutex);
360 return ret;
361}
362
363static void virtblk_release(struct gendisk *disk, fmode_t mode)
364{
365 struct virtio_blk *vblk = disk->private_data;
366
367 virtblk_put(vblk);
368}
369
370/* We provide getgeo only to please some old bootloader/partitioning tools */
371static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
372{
373 struct virtio_blk *vblk = bd->bd_disk->private_data;
374 int ret = 0;
375
376 mutex_lock(&vblk->vdev_mutex);
377
378 if (!vblk->vdev) {
379 ret = -ENXIO;
380 goto out;
381 }
382
383 /* see if the host passed in geometry config */
384 if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
385 virtio_cread(vblk->vdev, struct virtio_blk_config,
386 geometry.cylinders, &geo->cylinders);
387 virtio_cread(vblk->vdev, struct virtio_blk_config,
388 geometry.heads, &geo->heads);
389 virtio_cread(vblk->vdev, struct virtio_blk_config,
390 geometry.sectors, &geo->sectors);
391 } else {
392 /* some standard values, similar to sd */
393 geo->heads = 1 << 6;
394 geo->sectors = 1 << 5;
395 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
396 }
397out:
398 mutex_unlock(&vblk->vdev_mutex);
399 return ret;
400}
401
402static const struct block_device_operations virtblk_fops = {
403 .owner = THIS_MODULE,
404 .open = virtblk_open,
405 .release = virtblk_release,
406 .getgeo = virtblk_getgeo,
407};
408
409static int index_to_minor(int index)
410{
411 return index << PART_BITS;
412}
413
414static int minor_to_index(int minor)
415{
416 return minor >> PART_BITS;
417}
418
419static ssize_t serial_show(struct device *dev,
420 struct device_attribute *attr, char *buf)
421{
422 struct gendisk *disk = dev_to_disk(dev);
423 int err;
424
425 /* sysfs gives us a PAGE_SIZE buffer */
426 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
427
428 buf[VIRTIO_BLK_ID_BYTES] = '\0';
429 err = virtblk_get_id(disk, buf);
430 if (!err)
431 return strlen(buf);
432
433 if (err == -EIO) /* Unsupported? Make it empty. */
434 return 0;
435
436 return err;
437}
438
439static DEVICE_ATTR_RO(serial);
440
441/* The queue's logical block size must be set before calling this */
442static void virtblk_update_capacity(struct virtio_blk *vblk, bool resize)
443{
444 struct virtio_device *vdev = vblk->vdev;
445 struct request_queue *q = vblk->disk->queue;
446 char cap_str_2[10], cap_str_10[10];
447 unsigned long long nblocks;
448 u64 capacity;
449
450 /* Host must always specify the capacity. */
451 virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
452
453 nblocks = DIV_ROUND_UP_ULL(capacity, queue_logical_block_size(q) >> 9);
454
455 string_get_size(nblocks, queue_logical_block_size(q),
456 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
457 string_get_size(nblocks, queue_logical_block_size(q),
458 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
459
460 dev_notice(&vdev->dev,
461 "[%s] %s%llu %d-byte logical blocks (%s/%s)\n",
462 vblk->disk->disk_name,
463 resize ? "new size: " : "",
464 nblocks,
465 queue_logical_block_size(q),
466 cap_str_10,
467 cap_str_2);
468
469 set_capacity_and_notify(vblk->disk, capacity);
470}
471
472static void virtblk_config_changed_work(struct work_struct *work)
473{
474 struct virtio_blk *vblk =
475 container_of(work, struct virtio_blk, config_work);
476
477 virtblk_update_capacity(vblk, true);
478}
479
480static void virtblk_config_changed(struct virtio_device *vdev)
481{
482 struct virtio_blk *vblk = vdev->priv;
483
484 queue_work(virtblk_wq, &vblk->config_work);
485}
486
487static int init_vq(struct virtio_blk *vblk)
488{
489 int err;
490 int i;
491 vq_callback_t **callbacks;
492 const char **names;
493 struct virtqueue **vqs;
494 unsigned short num_vqs;
495 struct virtio_device *vdev = vblk->vdev;
496 struct irq_affinity desc = { 0, };
497
498 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
499 struct virtio_blk_config, num_queues,
500 &num_vqs);
501 if (err)
502 num_vqs = 1;
503
504 num_vqs = min_t(unsigned int, nr_cpu_ids, num_vqs);
505
506 vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL);
507 if (!vblk->vqs)
508 return -ENOMEM;
509
510 names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL);
511 callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL);
512 vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL);
513 if (!names || !callbacks || !vqs) {
514 err = -ENOMEM;
515 goto out;
516 }
517
518 for (i = 0; i < num_vqs; i++) {
519 callbacks[i] = virtblk_done;
520 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
521 names[i] = vblk->vqs[i].name;
522 }
523
524 /* Discover virtqueues and write information to configuration. */
525 err = virtio_find_vqs(vdev, num_vqs, vqs, callbacks, names, &desc);
526 if (err)
527 goto out;
528
529 for (i = 0; i < num_vqs; i++) {
530 spin_lock_init(&vblk->vqs[i].lock);
531 vblk->vqs[i].vq = vqs[i];
532 }
533 vblk->num_vqs = num_vqs;
534
535out:
536 kfree(vqs);
537 kfree(callbacks);
538 kfree(names);
539 if (err)
540 kfree(vblk->vqs);
541 return err;
542}
543
544/*
545 * Legacy naming scheme used for virtio devices. We are stuck with it for
546 * virtio blk but don't ever use it for any new driver.
547 */
548static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
549{
550 const int base = 'z' - 'a' + 1;
551 char *begin = buf + strlen(prefix);
552 char *end = buf + buflen;
553 char *p;
554 int unit;
555
556 p = end - 1;
557 *p = '\0';
558 unit = base;
559 do {
560 if (p == begin)
561 return -EINVAL;
562 *--p = 'a' + (index % unit);
563 index = (index / unit) - 1;
564 } while (index >= 0);
565
566 memmove(begin, p, end - p);
567 memcpy(buf, prefix, strlen(prefix));
568
569 return 0;
570}
571
572static int virtblk_get_cache_mode(struct virtio_device *vdev)
573{
574 u8 writeback;
575 int err;
576
577 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
578 struct virtio_blk_config, wce,
579 &writeback);
580
581 /*
582 * If WCE is not configurable and flush is not available,
583 * assume no writeback cache is in use.
584 */
585 if (err)
586 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
587
588 return writeback;
589}
590
591static void virtblk_update_cache_mode(struct virtio_device *vdev)
592{
593 u8 writeback = virtblk_get_cache_mode(vdev);
594 struct virtio_blk *vblk = vdev->priv;
595
596 blk_queue_write_cache(vblk->disk->queue, writeback, false);
597}
598
599static const char *const virtblk_cache_types[] = {
600 "write through", "write back"
601};
602
603static ssize_t
604cache_type_store(struct device *dev, struct device_attribute *attr,
605 const char *buf, size_t count)
606{
607 struct gendisk *disk = dev_to_disk(dev);
608 struct virtio_blk *vblk = disk->private_data;
609 struct virtio_device *vdev = vblk->vdev;
610 int i;
611
612 BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
613 i = sysfs_match_string(virtblk_cache_types, buf);
614 if (i < 0)
615 return i;
616
617 virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
618 virtblk_update_cache_mode(vdev);
619 return count;
620}
621
622static ssize_t
623cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
624{
625 struct gendisk *disk = dev_to_disk(dev);
626 struct virtio_blk *vblk = disk->private_data;
627 u8 writeback = virtblk_get_cache_mode(vblk->vdev);
628
629 BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
630 return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
631}
632
633static DEVICE_ATTR_RW(cache_type);
634
635static struct attribute *virtblk_attrs[] = {
636 &dev_attr_serial.attr,
637 &dev_attr_cache_type.attr,
638 NULL,
639};
640
641static umode_t virtblk_attrs_are_visible(struct kobject *kobj,
642 struct attribute *a, int n)
643{
644 struct device *dev = kobj_to_dev(kobj);
645 struct gendisk *disk = dev_to_disk(dev);
646 struct virtio_blk *vblk = disk->private_data;
647 struct virtio_device *vdev = vblk->vdev;
648
649 if (a == &dev_attr_cache_type.attr &&
650 !virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
651 return S_IRUGO;
652
653 return a->mode;
654}
655
656static const struct attribute_group virtblk_attr_group = {
657 .attrs = virtblk_attrs,
658 .is_visible = virtblk_attrs_are_visible,
659};
660
661static const struct attribute_group *virtblk_attr_groups[] = {
662 &virtblk_attr_group,
663 NULL,
664};
665
666static int virtblk_init_request(struct blk_mq_tag_set *set, struct request *rq,
667 unsigned int hctx_idx, unsigned int numa_node)
668{
669 struct virtio_blk *vblk = set->driver_data;
670 struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
671
672 sg_init_table(vbr->sg, vblk->sg_elems);
673 return 0;
674}
675
676static int virtblk_map_queues(struct blk_mq_tag_set *set)
677{
678 struct virtio_blk *vblk = set->driver_data;
679
680 return blk_mq_virtio_map_queues(&set->map[HCTX_TYPE_DEFAULT],
681 vblk->vdev, 0);
682}
683
684static const struct blk_mq_ops virtio_mq_ops = {
685 .queue_rq = virtio_queue_rq,
686 .commit_rqs = virtio_commit_rqs,
687 .complete = virtblk_request_done,
688 .init_request = virtblk_init_request,
689 .map_queues = virtblk_map_queues,
690};
691
692static unsigned int virtblk_queue_depth;
693module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
694
695static int virtblk_probe(struct virtio_device *vdev)
696{
697 struct virtio_blk *vblk;
698 struct request_queue *q;
699 int err, index;
700
701 u32 v, blk_size, max_size, sg_elems, opt_io_size;
702 u16 min_io_size;
703 u8 physical_block_exp, alignment_offset;
704 unsigned int queue_depth;
705
706 if (!vdev->config->get) {
707 dev_err(&vdev->dev, "%s failure: config access disabled\n",
708 __func__);
709 return -EINVAL;
710 }
711
712 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
713 GFP_KERNEL);
714 if (err < 0)
715 goto out;
716 index = err;
717
718 /* We need to know how many segments before we allocate. */
719 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
720 struct virtio_blk_config, seg_max,
721 &sg_elems);
722
723 /* We need at least one SG element, whatever they say. */
724 if (err || !sg_elems)
725 sg_elems = 1;
726
727 /* Prevent integer overflows and honor max vq size */
728 sg_elems = min_t(u32, sg_elems, VIRTIO_BLK_MAX_SG_ELEMS - 2);
729
730 /* We need extra sg elements at head and tail. */
731 sg_elems += 2;
732 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
733 if (!vblk) {
734 err = -ENOMEM;
735 goto out_free_index;
736 }
737
738 /* This reference is dropped in virtblk_remove(). */
739 refcount_set(&vblk->refs, 1);
740 mutex_init(&vblk->vdev_mutex);
741
742 vblk->vdev = vdev;
743 vblk->sg_elems = sg_elems;
744
745 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
746
747 err = init_vq(vblk);
748 if (err)
749 goto out_free_vblk;
750
751 /* Default queue sizing is to fill the ring. */
752 if (!virtblk_queue_depth) {
753 queue_depth = vblk->vqs[0].vq->num_free;
754 /* ... but without indirect descs, we use 2 descs per req */
755 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
756 queue_depth /= 2;
757 } else {
758 queue_depth = virtblk_queue_depth;
759 }
760
761 memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
762 vblk->tag_set.ops = &virtio_mq_ops;
763 vblk->tag_set.queue_depth = queue_depth;
764 vblk->tag_set.numa_node = NUMA_NO_NODE;
765 vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
766 vblk->tag_set.cmd_size =
767 sizeof(struct virtblk_req) +
768 sizeof(struct scatterlist) * sg_elems;
769 vblk->tag_set.driver_data = vblk;
770 vblk->tag_set.nr_hw_queues = vblk->num_vqs;
771
772 err = blk_mq_alloc_tag_set(&vblk->tag_set);
773 if (err)
774 goto out_free_vq;
775
776 vblk->disk = blk_mq_alloc_disk(&vblk->tag_set, vblk);
777 if (IS_ERR(vblk->disk)) {
778 err = PTR_ERR(vblk->disk);
779 goto out_free_tags;
780 }
781 q = vblk->disk->queue;
782
783 virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
784
785 vblk->disk->major = major;
786 vblk->disk->first_minor = index_to_minor(index);
787 vblk->disk->minors = 1 << PART_BITS;
788 vblk->disk->private_data = vblk;
789 vblk->disk->fops = &virtblk_fops;
790 vblk->disk->flags |= GENHD_FL_EXT_DEVT;
791 vblk->index = index;
792
793 /* configure queue flush support */
794 virtblk_update_cache_mode(vdev);
795
796 /* If disk is read-only in the host, the guest should obey */
797 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
798 set_disk_ro(vblk->disk, 1);
799
800 /* We can handle whatever the host told us to handle. */
801 blk_queue_max_segments(q, vblk->sg_elems-2);
802
803 /* No real sector limit. */
804 blk_queue_max_hw_sectors(q, -1U);
805
806 max_size = virtio_max_dma_size(vdev);
807
808 /* Host can optionally specify maximum segment size and number of
809 * segments. */
810 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
811 struct virtio_blk_config, size_max, &v);
812 if (!err)
813 max_size = min(max_size, v);
814
815 blk_queue_max_segment_size(q, max_size);
816
817 /* Host can optionally specify the block size of the device */
818 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
819 struct virtio_blk_config, blk_size,
820 &blk_size);
821 if (!err)
822 blk_queue_logical_block_size(q, blk_size);
823 else
824 blk_size = queue_logical_block_size(q);
825
826 /* Use topology information if available */
827 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
828 struct virtio_blk_config, physical_block_exp,
829 &physical_block_exp);
830 if (!err && physical_block_exp)
831 blk_queue_physical_block_size(q,
832 blk_size * (1 << physical_block_exp));
833
834 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
835 struct virtio_blk_config, alignment_offset,
836 &alignment_offset);
837 if (!err && alignment_offset)
838 blk_queue_alignment_offset(q, blk_size * alignment_offset);
839
840 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
841 struct virtio_blk_config, min_io_size,
842 &min_io_size);
843 if (!err && min_io_size)
844 blk_queue_io_min(q, blk_size * min_io_size);
845
846 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
847 struct virtio_blk_config, opt_io_size,
848 &opt_io_size);
849 if (!err && opt_io_size)
850 blk_queue_io_opt(q, blk_size * opt_io_size);
851
852 if (virtio_has_feature(vdev, VIRTIO_BLK_F_DISCARD)) {
853 q->limits.discard_granularity = blk_size;
854
855 virtio_cread(vdev, struct virtio_blk_config,
856 discard_sector_alignment, &v);
857 q->limits.discard_alignment = v ? v << SECTOR_SHIFT : 0;
858
859 virtio_cread(vdev, struct virtio_blk_config,
860 max_discard_sectors, &v);
861 blk_queue_max_discard_sectors(q, v ? v : UINT_MAX);
862
863 virtio_cread(vdev, struct virtio_blk_config, max_discard_seg,
864 &v);
865 blk_queue_max_discard_segments(q,
866 min_not_zero(v,
867 MAX_DISCARD_SEGMENTS));
868
869 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
870 }
871
872 if (virtio_has_feature(vdev, VIRTIO_BLK_F_WRITE_ZEROES)) {
873 virtio_cread(vdev, struct virtio_blk_config,
874 max_write_zeroes_sectors, &v);
875 blk_queue_max_write_zeroes_sectors(q, v ? v : UINT_MAX);
876 }
877
878 virtblk_update_capacity(vblk, false);
879 virtio_device_ready(vdev);
880
881 device_add_disk(&vdev->dev, vblk->disk, virtblk_attr_groups);
882 return 0;
883
884out_free_tags:
885 blk_mq_free_tag_set(&vblk->tag_set);
886out_free_vq:
887 vdev->config->del_vqs(vdev);
888 kfree(vblk->vqs);
889out_free_vblk:
890 kfree(vblk);
891out_free_index:
892 ida_simple_remove(&vd_index_ida, index);
893out:
894 return err;
895}
896
897static void virtblk_remove(struct virtio_device *vdev)
898{
899 struct virtio_blk *vblk = vdev->priv;
900
901 /* Make sure no work handler is accessing the device. */
902 flush_work(&vblk->config_work);
903
904 del_gendisk(vblk->disk);
905 blk_cleanup_disk(vblk->disk);
906 blk_mq_free_tag_set(&vblk->tag_set);
907
908 mutex_lock(&vblk->vdev_mutex);
909
910 /* Stop all the virtqueues. */
911 vdev->config->reset(vdev);
912
913 /* Virtqueues are stopped, nothing can use vblk->vdev anymore. */
914 vblk->vdev = NULL;
915
916 vdev->config->del_vqs(vdev);
917 kfree(vblk->vqs);
918
919 mutex_unlock(&vblk->vdev_mutex);
920
921 virtblk_put(vblk);
922}
923
924#ifdef CONFIG_PM_SLEEP
925static int virtblk_freeze(struct virtio_device *vdev)
926{
927 struct virtio_blk *vblk = vdev->priv;
928
929 /* Ensure we don't receive any more interrupts */
930 vdev->config->reset(vdev);
931
932 /* Make sure no work handler is accessing the device. */
933 flush_work(&vblk->config_work);
934
935 blk_mq_quiesce_queue(vblk->disk->queue);
936
937 vdev->config->del_vqs(vdev);
938 kfree(vblk->vqs);
939
940 return 0;
941}
942
943static int virtblk_restore(struct virtio_device *vdev)
944{
945 struct virtio_blk *vblk = vdev->priv;
946 int ret;
947
948 ret = init_vq(vdev->priv);
949 if (ret)
950 return ret;
951
952 virtio_device_ready(vdev);
953
954 blk_mq_unquiesce_queue(vblk->disk->queue);
955 return 0;
956}
957#endif
958
959static const struct virtio_device_id id_table[] = {
960 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
961 { 0 },
962};
963
964static unsigned int features_legacy[] = {
965 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
966 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
967 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
968 VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
969}
970;
971static unsigned int features[] = {
972 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
973 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
974 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
975 VIRTIO_BLK_F_MQ, VIRTIO_BLK_F_DISCARD, VIRTIO_BLK_F_WRITE_ZEROES,
976};
977
978static struct virtio_driver virtio_blk = {
979 .feature_table = features,
980 .feature_table_size = ARRAY_SIZE(features),
981 .feature_table_legacy = features_legacy,
982 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
983 .driver.name = KBUILD_MODNAME,
984 .driver.owner = THIS_MODULE,
985 .id_table = id_table,
986 .probe = virtblk_probe,
987 .remove = virtblk_remove,
988 .config_changed = virtblk_config_changed,
989#ifdef CONFIG_PM_SLEEP
990 .freeze = virtblk_freeze,
991 .restore = virtblk_restore,
992#endif
993};
994
995static int __init init(void)
996{
997 int error;
998
999 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
1000 if (!virtblk_wq)
1001 return -ENOMEM;
1002
1003 major = register_blkdev(0, "virtblk");
1004 if (major < 0) {
1005 error = major;
1006 goto out_destroy_workqueue;
1007 }
1008
1009 error = register_virtio_driver(&virtio_blk);
1010 if (error)
1011 goto out_unregister_blkdev;
1012 return 0;
1013
1014out_unregister_blkdev:
1015 unregister_blkdev(major, "virtblk");
1016out_destroy_workqueue:
1017 destroy_workqueue(virtblk_wq);
1018 return error;
1019}
1020
1021static void __exit fini(void)
1022{
1023 unregister_virtio_driver(&virtio_blk);
1024 unregister_blkdev(major, "virtblk");
1025 destroy_workqueue(virtblk_wq);
1026}
1027module_init(init);
1028module_exit(fini);
1029
1030MODULE_DEVICE_TABLE(virtio, id_table);
1031MODULE_DESCRIPTION("Virtio block driver");
1032MODULE_LICENSE("GPL");