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/module.h>
7#include <linux/mutex.h>
8#include <linux/virtio.h>
9#include <linux/virtio_blk.h>
10#include <linux/scatterlist.h>
11#include <linux/string_helpers.h>
12#include <scsi/scsi_cmnd.h>
13#include <linux/idr.h>
14#include <linux/blk-mq.h>
15#include <linux/numa.h>
16
17#define PART_BITS 4
18
19static int major;
20static DEFINE_IDA(vd_index_ida);
21
22static struct workqueue_struct *virtblk_wq;
23
24struct virtio_blk
25{
26 struct virtio_device *vdev;
27 struct virtqueue *vq;
28 spinlock_t vq_lock;
29
30 /* The disk structure for the kernel. */
31 struct gendisk *disk;
32
33 /* Process context for config space updates */
34 struct work_struct config_work;
35
36 /* Lock for config space updates */
37 struct mutex config_lock;
38
39 /* enable config space updates */
40 bool config_enable;
41
42 /* What host tells us, plus 2 for header & tailer. */
43 unsigned int sg_elems;
44
45 /* Ida index - used to track minor number allocations. */
46 int index;
47};
48
49struct virtblk_req
50{
51 struct request *req;
52 struct virtio_blk_outhdr out_hdr;
53 struct virtio_scsi_inhdr in_hdr;
54 u8 status;
55 struct scatterlist sg[];
56};
57
58static inline int virtblk_result(struct virtblk_req *vbr)
59{
60 switch (vbr->status) {
61 case VIRTIO_BLK_S_OK:
62 return 0;
63 case VIRTIO_BLK_S_UNSUPP:
64 return -ENOTTY;
65 default:
66 return -EIO;
67 }
68}
69
70static int __virtblk_add_req(struct virtqueue *vq,
71 struct virtblk_req *vbr,
72 struct scatterlist *data_sg,
73 bool have_data)
74{
75 struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
76 unsigned int num_out = 0, num_in = 0;
77 int type = vbr->out_hdr.type & ~VIRTIO_BLK_T_OUT;
78
79 sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
80 sgs[num_out++] = &hdr;
81
82 /*
83 * If this is a packet command we need a couple of additional headers.
84 * Behind the normal outhdr we put a segment with the scsi command
85 * block, and before the normal inhdr we put the sense data and the
86 * inhdr with additional status information.
87 */
88 if (type == VIRTIO_BLK_T_SCSI_CMD) {
89 sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len);
90 sgs[num_out++] = &cmd;
91 }
92
93 if (have_data) {
94 if (vbr->out_hdr.type & VIRTIO_BLK_T_OUT)
95 sgs[num_out++] = data_sg;
96 else
97 sgs[num_out + num_in++] = data_sg;
98 }
99
100 if (type == VIRTIO_BLK_T_SCSI_CMD) {
101 sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
102 sgs[num_out + num_in++] = &sense;
103 sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
104 sgs[num_out + num_in++] = &inhdr;
105 }
106
107 sg_init_one(&status, &vbr->status, sizeof(vbr->status));
108 sgs[num_out + num_in++] = &status;
109
110 return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC);
111}
112
113static inline void virtblk_request_done(struct request *req)
114{
115 struct virtblk_req *vbr = req->special;
116 int error = virtblk_result(vbr);
117
118 if (req->cmd_type == REQ_TYPE_BLOCK_PC) {
119 req->resid_len = vbr->in_hdr.residual;
120 req->sense_len = vbr->in_hdr.sense_len;
121 req->errors = vbr->in_hdr.errors;
122 } else if (req->cmd_type == REQ_TYPE_SPECIAL) {
123 req->errors = (error != 0);
124 }
125
126 blk_mq_end_io(req, error);
127}
128
129static void virtblk_done(struct virtqueue *vq)
130{
131 struct virtio_blk *vblk = vq->vdev->priv;
132 bool req_done = false;
133 struct virtblk_req *vbr;
134 unsigned long flags;
135 unsigned int len;
136
137 spin_lock_irqsave(&vblk->vq_lock, flags);
138 do {
139 virtqueue_disable_cb(vq);
140 while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
141 blk_mq_complete_request(vbr->req);
142 req_done = true;
143 }
144 if (unlikely(virtqueue_is_broken(vq)))
145 break;
146 } while (!virtqueue_enable_cb(vq));
147
148 /* In case queue is stopped waiting for more buffers. */
149 if (req_done)
150 blk_mq_start_stopped_hw_queues(vblk->disk->queue);
151 spin_unlock_irqrestore(&vblk->vq_lock, flags);
152}
153
154static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req)
155{
156 struct virtio_blk *vblk = hctx->queue->queuedata;
157 struct virtblk_req *vbr = req->special;
158 unsigned long flags;
159 unsigned int num;
160 const bool last = (req->cmd_flags & REQ_END) != 0;
161 int err;
162
163 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
164
165 vbr->req = req;
166 if (req->cmd_flags & REQ_FLUSH) {
167 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
168 vbr->out_hdr.sector = 0;
169 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
170 } else {
171 switch (req->cmd_type) {
172 case REQ_TYPE_FS:
173 vbr->out_hdr.type = 0;
174 vbr->out_hdr.sector = blk_rq_pos(vbr->req);
175 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
176 break;
177 case REQ_TYPE_BLOCK_PC:
178 vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
179 vbr->out_hdr.sector = 0;
180 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
181 break;
182 case REQ_TYPE_SPECIAL:
183 vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
184 vbr->out_hdr.sector = 0;
185 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
186 break;
187 default:
188 /* We don't put anything else in the queue. */
189 BUG();
190 }
191 }
192
193 num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg);
194 if (num) {
195 if (rq_data_dir(vbr->req) == WRITE)
196 vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
197 else
198 vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
199 }
200
201 spin_lock_irqsave(&vblk->vq_lock, flags);
202 err = __virtblk_add_req(vblk->vq, vbr, vbr->sg, num);
203 if (err) {
204 virtqueue_kick(vblk->vq);
205 blk_mq_stop_hw_queue(hctx);
206 spin_unlock_irqrestore(&vblk->vq_lock, flags);
207 /* Out of mem doesn't actually happen, since we fall back
208 * to direct descriptors */
209 if (err == -ENOMEM || err == -ENOSPC)
210 return BLK_MQ_RQ_QUEUE_BUSY;
211 return BLK_MQ_RQ_QUEUE_ERROR;
212 }
213
214 if (last)
215 virtqueue_kick(vblk->vq);
216
217 spin_unlock_irqrestore(&vblk->vq_lock, flags);
218 return BLK_MQ_RQ_QUEUE_OK;
219}
220
221/* return id (s/n) string for *disk to *id_str
222 */
223static int virtblk_get_id(struct gendisk *disk, char *id_str)
224{
225 struct virtio_blk *vblk = disk->private_data;
226 struct request *req;
227 struct bio *bio;
228 int err;
229
230 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
231 GFP_KERNEL);
232 if (IS_ERR(bio))
233 return PTR_ERR(bio);
234
235 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
236 if (IS_ERR(req)) {
237 bio_put(bio);
238 return PTR_ERR(req);
239 }
240
241 req->cmd_type = REQ_TYPE_SPECIAL;
242 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
243 blk_put_request(req);
244
245 return err;
246}
247
248static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
249 unsigned int cmd, unsigned long data)
250{
251 struct gendisk *disk = bdev->bd_disk;
252 struct virtio_blk *vblk = disk->private_data;
253
254 /*
255 * Only allow the generic SCSI ioctls if the host can support it.
256 */
257 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
258 return -ENOTTY;
259
260 return scsi_cmd_blk_ioctl(bdev, mode, cmd,
261 (void __user *)data);
262}
263
264/* We provide getgeo only to please some old bootloader/partitioning tools */
265static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
266{
267 struct virtio_blk *vblk = bd->bd_disk->private_data;
268
269 /* see if the host passed in geometry config */
270 if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
271 virtio_cread(vblk->vdev, struct virtio_blk_config,
272 geometry.cylinders, &geo->cylinders);
273 virtio_cread(vblk->vdev, struct virtio_blk_config,
274 geometry.heads, &geo->heads);
275 virtio_cread(vblk->vdev, struct virtio_blk_config,
276 geometry.sectors, &geo->sectors);
277 } else {
278 /* some standard values, similar to sd */
279 geo->heads = 1 << 6;
280 geo->sectors = 1 << 5;
281 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
282 }
283 return 0;
284}
285
286static const struct block_device_operations virtblk_fops = {
287 .ioctl = virtblk_ioctl,
288 .owner = THIS_MODULE,
289 .getgeo = virtblk_getgeo,
290};
291
292static int index_to_minor(int index)
293{
294 return index << PART_BITS;
295}
296
297static int minor_to_index(int minor)
298{
299 return minor >> PART_BITS;
300}
301
302static ssize_t virtblk_serial_show(struct device *dev,
303 struct device_attribute *attr, char *buf)
304{
305 struct gendisk *disk = dev_to_disk(dev);
306 int err;
307
308 /* sysfs gives us a PAGE_SIZE buffer */
309 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
310
311 buf[VIRTIO_BLK_ID_BYTES] = '\0';
312 err = virtblk_get_id(disk, buf);
313 if (!err)
314 return strlen(buf);
315
316 if (err == -EIO) /* Unsupported? Make it empty. */
317 return 0;
318
319 return err;
320}
321DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
322
323static void virtblk_config_changed_work(struct work_struct *work)
324{
325 struct virtio_blk *vblk =
326 container_of(work, struct virtio_blk, config_work);
327 struct virtio_device *vdev = vblk->vdev;
328 struct request_queue *q = vblk->disk->queue;
329 char cap_str_2[10], cap_str_10[10];
330 char *envp[] = { "RESIZE=1", NULL };
331 u64 capacity, size;
332
333 mutex_lock(&vblk->config_lock);
334 if (!vblk->config_enable)
335 goto done;
336
337 /* Host must always specify the capacity. */
338 virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
339
340 /* If capacity is too big, truncate with warning. */
341 if ((sector_t)capacity != capacity) {
342 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
343 (unsigned long long)capacity);
344 capacity = (sector_t)-1;
345 }
346
347 size = capacity * queue_logical_block_size(q);
348 string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
349 string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
350
351 dev_notice(&vdev->dev,
352 "new size: %llu %d-byte logical blocks (%s/%s)\n",
353 (unsigned long long)capacity,
354 queue_logical_block_size(q),
355 cap_str_10, cap_str_2);
356
357 set_capacity(vblk->disk, capacity);
358 revalidate_disk(vblk->disk);
359 kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
360done:
361 mutex_unlock(&vblk->config_lock);
362}
363
364static void virtblk_config_changed(struct virtio_device *vdev)
365{
366 struct virtio_blk *vblk = vdev->priv;
367
368 queue_work(virtblk_wq, &vblk->config_work);
369}
370
371static int init_vq(struct virtio_blk *vblk)
372{
373 int err = 0;
374
375 /* We expect one virtqueue, for output. */
376 vblk->vq = virtio_find_single_vq(vblk->vdev, virtblk_done, "requests");
377 if (IS_ERR(vblk->vq))
378 err = PTR_ERR(vblk->vq);
379
380 return err;
381}
382
383/*
384 * Legacy naming scheme used for virtio devices. We are stuck with it for
385 * virtio blk but don't ever use it for any new driver.
386 */
387static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
388{
389 const int base = 'z' - 'a' + 1;
390 char *begin = buf + strlen(prefix);
391 char *end = buf + buflen;
392 char *p;
393 int unit;
394
395 p = end - 1;
396 *p = '\0';
397 unit = base;
398 do {
399 if (p == begin)
400 return -EINVAL;
401 *--p = 'a' + (index % unit);
402 index = (index / unit) - 1;
403 } while (index >= 0);
404
405 memmove(begin, p, end - p);
406 memcpy(buf, prefix, strlen(prefix));
407
408 return 0;
409}
410
411static int virtblk_get_cache_mode(struct virtio_device *vdev)
412{
413 u8 writeback;
414 int err;
415
416 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
417 struct virtio_blk_config, wce,
418 &writeback);
419 if (err)
420 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE);
421
422 return writeback;
423}
424
425static void virtblk_update_cache_mode(struct virtio_device *vdev)
426{
427 u8 writeback = virtblk_get_cache_mode(vdev);
428 struct virtio_blk *vblk = vdev->priv;
429
430 if (writeback)
431 blk_queue_flush(vblk->disk->queue, REQ_FLUSH);
432 else
433 blk_queue_flush(vblk->disk->queue, 0);
434
435 revalidate_disk(vblk->disk);
436}
437
438static const char *const virtblk_cache_types[] = {
439 "write through", "write back"
440};
441
442static ssize_t
443virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
444 const char *buf, size_t count)
445{
446 struct gendisk *disk = dev_to_disk(dev);
447 struct virtio_blk *vblk = disk->private_data;
448 struct virtio_device *vdev = vblk->vdev;
449 int i;
450
451 BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
452 for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
453 if (sysfs_streq(buf, virtblk_cache_types[i]))
454 break;
455
456 if (i < 0)
457 return -EINVAL;
458
459 virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
460 virtblk_update_cache_mode(vdev);
461 return count;
462}
463
464static ssize_t
465virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
466 char *buf)
467{
468 struct gendisk *disk = dev_to_disk(dev);
469 struct virtio_blk *vblk = disk->private_data;
470 u8 writeback = virtblk_get_cache_mode(vblk->vdev);
471
472 BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
473 return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
474}
475
476static const struct device_attribute dev_attr_cache_type_ro =
477 __ATTR(cache_type, S_IRUGO,
478 virtblk_cache_type_show, NULL);
479static const struct device_attribute dev_attr_cache_type_rw =
480 __ATTR(cache_type, S_IRUGO|S_IWUSR,
481 virtblk_cache_type_show, virtblk_cache_type_store);
482
483static struct blk_mq_ops virtio_mq_ops = {
484 .queue_rq = virtio_queue_rq,
485 .map_queue = blk_mq_map_queue,
486 .alloc_hctx = blk_mq_alloc_single_hw_queue,
487 .free_hctx = blk_mq_free_single_hw_queue,
488 .complete = virtblk_request_done,
489};
490
491static struct blk_mq_reg virtio_mq_reg = {
492 .ops = &virtio_mq_ops,
493 .nr_hw_queues = 1,
494 .queue_depth = 0, /* Set in virtblk_probe */
495 .numa_node = NUMA_NO_NODE,
496 .flags = BLK_MQ_F_SHOULD_MERGE,
497};
498module_param_named(queue_depth, virtio_mq_reg.queue_depth, uint, 0444);
499
500static int virtblk_init_vbr(void *data, struct blk_mq_hw_ctx *hctx,
501 struct request *rq, unsigned int nr)
502{
503 struct virtio_blk *vblk = data;
504 struct virtblk_req *vbr = rq->special;
505
506 sg_init_table(vbr->sg, vblk->sg_elems);
507 return 0;
508}
509
510static int virtblk_probe(struct virtio_device *vdev)
511{
512 struct virtio_blk *vblk;
513 struct request_queue *q;
514 int err, index;
515
516 u64 cap;
517 u32 v, blk_size, sg_elems, opt_io_size;
518 u16 min_io_size;
519 u8 physical_block_exp, alignment_offset;
520
521 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
522 GFP_KERNEL);
523 if (err < 0)
524 goto out;
525 index = err;
526
527 /* We need to know how many segments before we allocate. */
528 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
529 struct virtio_blk_config, seg_max,
530 &sg_elems);
531
532 /* We need at least one SG element, whatever they say. */
533 if (err || !sg_elems)
534 sg_elems = 1;
535
536 /* We need an extra sg elements at head and tail. */
537 sg_elems += 2;
538 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
539 if (!vblk) {
540 err = -ENOMEM;
541 goto out_free_index;
542 }
543
544 vblk->vdev = vdev;
545 vblk->sg_elems = sg_elems;
546 mutex_init(&vblk->config_lock);
547
548 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
549 vblk->config_enable = true;
550
551 err = init_vq(vblk);
552 if (err)
553 goto out_free_vblk;
554 spin_lock_init(&vblk->vq_lock);
555
556 /* FIXME: How many partitions? How long is a piece of string? */
557 vblk->disk = alloc_disk(1 << PART_BITS);
558 if (!vblk->disk) {
559 err = -ENOMEM;
560 goto out_free_vq;
561 }
562
563 /* Default queue sizing is to fill the ring. */
564 if (!virtio_mq_reg.queue_depth) {
565 virtio_mq_reg.queue_depth = vblk->vq->num_free;
566 /* ... but without indirect descs, we use 2 descs per req */
567 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
568 virtio_mq_reg.queue_depth /= 2;
569 }
570 virtio_mq_reg.cmd_size =
571 sizeof(struct virtblk_req) +
572 sizeof(struct scatterlist) * sg_elems;
573
574 q = vblk->disk->queue = blk_mq_init_queue(&virtio_mq_reg, vblk);
575 if (!q) {
576 err = -ENOMEM;
577 goto out_put_disk;
578 }
579
580 blk_mq_init_commands(q, virtblk_init_vbr, vblk);
581
582 q->queuedata = vblk;
583
584 virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
585
586 vblk->disk->major = major;
587 vblk->disk->first_minor = index_to_minor(index);
588 vblk->disk->private_data = vblk;
589 vblk->disk->fops = &virtblk_fops;
590 vblk->disk->driverfs_dev = &vdev->dev;
591 vblk->index = index;
592
593 /* configure queue flush support */
594 virtblk_update_cache_mode(vdev);
595
596 /* If disk is read-only in the host, the guest should obey */
597 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
598 set_disk_ro(vblk->disk, 1);
599
600 /* Host must always specify the capacity. */
601 virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
602
603 /* If capacity is too big, truncate with warning. */
604 if ((sector_t)cap != cap) {
605 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
606 (unsigned long long)cap);
607 cap = (sector_t)-1;
608 }
609 set_capacity(vblk->disk, cap);
610
611 /* We can handle whatever the host told us to handle. */
612 blk_queue_max_segments(q, vblk->sg_elems-2);
613
614 /* No need to bounce any requests */
615 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
616
617 /* No real sector limit. */
618 blk_queue_max_hw_sectors(q, -1U);
619
620 /* Host can optionally specify maximum segment size and number of
621 * segments. */
622 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
623 struct virtio_blk_config, size_max, &v);
624 if (!err)
625 blk_queue_max_segment_size(q, v);
626 else
627 blk_queue_max_segment_size(q, -1U);
628
629 /* Host can optionally specify the block size of the device */
630 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
631 struct virtio_blk_config, blk_size,
632 &blk_size);
633 if (!err)
634 blk_queue_logical_block_size(q, blk_size);
635 else
636 blk_size = queue_logical_block_size(q);
637
638 /* Use topology information if available */
639 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
640 struct virtio_blk_config, physical_block_exp,
641 &physical_block_exp);
642 if (!err && physical_block_exp)
643 blk_queue_physical_block_size(q,
644 blk_size * (1 << physical_block_exp));
645
646 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
647 struct virtio_blk_config, alignment_offset,
648 &alignment_offset);
649 if (!err && alignment_offset)
650 blk_queue_alignment_offset(q, blk_size * alignment_offset);
651
652 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
653 struct virtio_blk_config, min_io_size,
654 &min_io_size);
655 if (!err && min_io_size)
656 blk_queue_io_min(q, blk_size * min_io_size);
657
658 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
659 struct virtio_blk_config, opt_io_size,
660 &opt_io_size);
661 if (!err && opt_io_size)
662 blk_queue_io_opt(q, blk_size * opt_io_size);
663
664 add_disk(vblk->disk);
665 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
666 if (err)
667 goto out_del_disk;
668
669 if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
670 err = device_create_file(disk_to_dev(vblk->disk),
671 &dev_attr_cache_type_rw);
672 else
673 err = device_create_file(disk_to_dev(vblk->disk),
674 &dev_attr_cache_type_ro);
675 if (err)
676 goto out_del_disk;
677 return 0;
678
679out_del_disk:
680 del_gendisk(vblk->disk);
681 blk_cleanup_queue(vblk->disk->queue);
682out_put_disk:
683 put_disk(vblk->disk);
684out_free_vq:
685 vdev->config->del_vqs(vdev);
686out_free_vblk:
687 kfree(vblk);
688out_free_index:
689 ida_simple_remove(&vd_index_ida, index);
690out:
691 return err;
692}
693
694static void virtblk_remove(struct virtio_device *vdev)
695{
696 struct virtio_blk *vblk = vdev->priv;
697 int index = vblk->index;
698 int refc;
699
700 /* Prevent config work handler from accessing the device. */
701 mutex_lock(&vblk->config_lock);
702 vblk->config_enable = false;
703 mutex_unlock(&vblk->config_lock);
704
705 del_gendisk(vblk->disk);
706 blk_cleanup_queue(vblk->disk->queue);
707
708 /* Stop all the virtqueues. */
709 vdev->config->reset(vdev);
710
711 flush_work(&vblk->config_work);
712
713 refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount);
714 put_disk(vblk->disk);
715 vdev->config->del_vqs(vdev);
716 kfree(vblk);
717
718 /* Only free device id if we don't have any users */
719 if (refc == 1)
720 ida_simple_remove(&vd_index_ida, index);
721}
722
723#ifdef CONFIG_PM_SLEEP
724static int virtblk_freeze(struct virtio_device *vdev)
725{
726 struct virtio_blk *vblk = vdev->priv;
727
728 /* Ensure we don't receive any more interrupts */
729 vdev->config->reset(vdev);
730
731 /* Prevent config work handler from accessing the device. */
732 mutex_lock(&vblk->config_lock);
733 vblk->config_enable = false;
734 mutex_unlock(&vblk->config_lock);
735
736 flush_work(&vblk->config_work);
737
738 blk_mq_stop_hw_queues(vblk->disk->queue);
739
740 vdev->config->del_vqs(vdev);
741 return 0;
742}
743
744static int virtblk_restore(struct virtio_device *vdev)
745{
746 struct virtio_blk *vblk = vdev->priv;
747 int ret;
748
749 vblk->config_enable = true;
750 ret = init_vq(vdev->priv);
751 if (!ret)
752 blk_mq_start_stopped_hw_queues(vblk->disk->queue);
753
754 return ret;
755}
756#endif
757
758static const struct virtio_device_id id_table[] = {
759 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
760 { 0 },
761};
762
763static unsigned int features[] = {
764 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
765 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
766 VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE
767};
768
769static struct virtio_driver virtio_blk = {
770 .feature_table = features,
771 .feature_table_size = ARRAY_SIZE(features),
772 .driver.name = KBUILD_MODNAME,
773 .driver.owner = THIS_MODULE,
774 .id_table = id_table,
775 .probe = virtblk_probe,
776 .remove = virtblk_remove,
777 .config_changed = virtblk_config_changed,
778#ifdef CONFIG_PM_SLEEP
779 .freeze = virtblk_freeze,
780 .restore = virtblk_restore,
781#endif
782};
783
784static int __init init(void)
785{
786 int error;
787
788 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
789 if (!virtblk_wq)
790 return -ENOMEM;
791
792 major = register_blkdev(0, "virtblk");
793 if (major < 0) {
794 error = major;
795 goto out_destroy_workqueue;
796 }
797
798 error = register_virtio_driver(&virtio_blk);
799 if (error)
800 goto out_unregister_blkdev;
801 return 0;
802
803out_unregister_blkdev:
804 unregister_blkdev(major, "virtblk");
805out_destroy_workqueue:
806 destroy_workqueue(virtblk_wq);
807 return error;
808}
809
810static void __exit fini(void)
811{
812 unregister_blkdev(major, "virtblk");
813 unregister_virtio_driver(&virtio_blk);
814 destroy_workqueue(virtblk_wq);
815}
816module_init(init);
817module_exit(fini);
818
819MODULE_DEVICE_TABLE(virtio, id_table);
820MODULE_DESCRIPTION("Virtio block driver");
821MODULE_LICENSE("GPL");
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/module.h>
7#include <linux/mutex.h>
8#include <linux/virtio.h>
9#include <linux/virtio_blk.h>
10#include <linux/scatterlist.h>
11#include <linux/string_helpers.h>
12#include <scsi/scsi_cmnd.h>
13#include <linux/idr.h>
14#include <linux/blk-mq.h>
15#include <linux/numa.h>
16
17#define PART_BITS 4
18#define VQ_NAME_LEN 16
19
20static int major;
21static DEFINE_IDA(vd_index_ida);
22
23static struct workqueue_struct *virtblk_wq;
24
25struct virtio_blk_vq {
26 struct virtqueue *vq;
27 spinlock_t lock;
28 char name[VQ_NAME_LEN];
29} ____cacheline_aligned_in_smp;
30
31struct virtio_blk {
32 struct virtio_device *vdev;
33
34 /* The disk structure for the kernel. */
35 struct gendisk *disk;
36
37 /* Block layer tags. */
38 struct blk_mq_tag_set tag_set;
39
40 /* Process context for config space updates */
41 struct work_struct config_work;
42
43 /* What host tells us, plus 2 for header & tailer. */
44 unsigned int sg_elems;
45
46 /* Ida index - used to track minor number allocations. */
47 int index;
48
49 /* num of vqs */
50 int num_vqs;
51 struct virtio_blk_vq *vqs;
52};
53
54struct virtblk_req {
55 struct request *req;
56 struct virtio_blk_outhdr out_hdr;
57 struct virtio_scsi_inhdr in_hdr;
58 u8 status;
59 struct scatterlist sg[];
60};
61
62static inline int virtblk_result(struct virtblk_req *vbr)
63{
64 switch (vbr->status) {
65 case VIRTIO_BLK_S_OK:
66 return 0;
67 case VIRTIO_BLK_S_UNSUPP:
68 return -ENOTTY;
69 default:
70 return -EIO;
71 }
72}
73
74static int __virtblk_add_req(struct virtqueue *vq,
75 struct virtblk_req *vbr,
76 struct scatterlist *data_sg,
77 bool have_data)
78{
79 struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6];
80 unsigned int num_out = 0, num_in = 0;
81 __virtio32 type = vbr->out_hdr.type & ~cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT);
82
83 sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr));
84 sgs[num_out++] = &hdr;
85
86 /*
87 * If this is a packet command we need a couple of additional headers.
88 * Behind the normal outhdr we put a segment with the scsi command
89 * block, and before the normal inhdr we put the sense data and the
90 * inhdr with additional status information.
91 */
92 if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) {
93 sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len);
94 sgs[num_out++] = &cmd;
95 }
96
97 if (have_data) {
98 if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT))
99 sgs[num_out++] = data_sg;
100 else
101 sgs[num_out + num_in++] = data_sg;
102 }
103
104 if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) {
105 sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE);
106 sgs[num_out + num_in++] = &sense;
107 sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr));
108 sgs[num_out + num_in++] = &inhdr;
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 inline void virtblk_request_done(struct request *req)
118{
119 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
120 struct virtio_blk *vblk = req->q->queuedata;
121 int error = virtblk_result(vbr);
122
123 if (req->cmd_type == REQ_TYPE_BLOCK_PC) {
124 req->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual);
125 req->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len);
126 req->errors = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors);
127 } else if (req->cmd_type == REQ_TYPE_DRV_PRIV) {
128 req->errors = (error != 0);
129 }
130
131 blk_mq_end_request(req, error);
132}
133
134static void virtblk_done(struct virtqueue *vq)
135{
136 struct virtio_blk *vblk = vq->vdev->priv;
137 bool req_done = false;
138 int qid = vq->index;
139 struct virtblk_req *vbr;
140 unsigned long flags;
141 unsigned int len;
142
143 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
144 do {
145 virtqueue_disable_cb(vq);
146 while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) {
147 blk_mq_complete_request(vbr->req, vbr->req->errors);
148 req_done = true;
149 }
150 if (unlikely(virtqueue_is_broken(vq)))
151 break;
152 } while (!virtqueue_enable_cb(vq));
153
154 /* In case queue is stopped waiting for more buffers. */
155 if (req_done)
156 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
157 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
158}
159
160static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx,
161 const struct blk_mq_queue_data *bd)
162{
163 struct virtio_blk *vblk = hctx->queue->queuedata;
164 struct request *req = bd->rq;
165 struct virtblk_req *vbr = blk_mq_rq_to_pdu(req);
166 unsigned long flags;
167 unsigned int num;
168 int qid = hctx->queue_num;
169 int err;
170 bool notify = false;
171
172 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
173
174 vbr->req = req;
175 if (req->cmd_flags & REQ_FLUSH) {
176 vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_FLUSH);
177 vbr->out_hdr.sector = 0;
178 vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
179 } else {
180 switch (req->cmd_type) {
181 case REQ_TYPE_FS:
182 vbr->out_hdr.type = 0;
183 vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, blk_rq_pos(vbr->req));
184 vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
185 break;
186 case REQ_TYPE_BLOCK_PC:
187 vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_SCSI_CMD);
188 vbr->out_hdr.sector = 0;
189 vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
190 break;
191 case REQ_TYPE_DRV_PRIV:
192 vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID);
193 vbr->out_hdr.sector = 0;
194 vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req));
195 break;
196 default:
197 /* We don't put anything else in the queue. */
198 BUG();
199 }
200 }
201
202 blk_mq_start_request(req);
203
204 num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg);
205 if (num) {
206 if (rq_data_dir(vbr->req) == WRITE)
207 vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT);
208 else
209 vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN);
210 }
211
212 spin_lock_irqsave(&vblk->vqs[qid].lock, flags);
213 err = __virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num);
214 if (err) {
215 virtqueue_kick(vblk->vqs[qid].vq);
216 blk_mq_stop_hw_queue(hctx);
217 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
218 /* Out of mem doesn't actually happen, since we fall back
219 * to direct descriptors */
220 if (err == -ENOMEM || err == -ENOSPC)
221 return BLK_MQ_RQ_QUEUE_BUSY;
222 return BLK_MQ_RQ_QUEUE_ERROR;
223 }
224
225 if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq))
226 notify = true;
227 spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags);
228
229 if (notify)
230 virtqueue_notify(vblk->vqs[qid].vq);
231 return BLK_MQ_RQ_QUEUE_OK;
232}
233
234/* return id (s/n) string for *disk to *id_str
235 */
236static int virtblk_get_id(struct gendisk *disk, char *id_str)
237{
238 struct virtio_blk *vblk = disk->private_data;
239 struct request *req;
240 struct bio *bio;
241 int err;
242
243 bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
244 GFP_KERNEL);
245 if (IS_ERR(bio))
246 return PTR_ERR(bio);
247
248 req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
249 if (IS_ERR(req)) {
250 bio_put(bio);
251 return PTR_ERR(req);
252 }
253
254 req->cmd_type = REQ_TYPE_DRV_PRIV;
255 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
256 blk_put_request(req);
257
258 return err;
259}
260
261static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
262 unsigned int cmd, unsigned long data)
263{
264 struct gendisk *disk = bdev->bd_disk;
265 struct virtio_blk *vblk = disk->private_data;
266
267 /*
268 * Only allow the generic SCSI ioctls if the host can support it.
269 */
270 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
271 return -ENOTTY;
272
273 return scsi_cmd_blk_ioctl(bdev, mode, cmd,
274 (void __user *)data);
275}
276
277/* We provide getgeo only to please some old bootloader/partitioning tools */
278static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
279{
280 struct virtio_blk *vblk = bd->bd_disk->private_data;
281
282 /* see if the host passed in geometry config */
283 if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
284 virtio_cread(vblk->vdev, struct virtio_blk_config,
285 geometry.cylinders, &geo->cylinders);
286 virtio_cread(vblk->vdev, struct virtio_blk_config,
287 geometry.heads, &geo->heads);
288 virtio_cread(vblk->vdev, struct virtio_blk_config,
289 geometry.sectors, &geo->sectors);
290 } else {
291 /* some standard values, similar to sd */
292 geo->heads = 1 << 6;
293 geo->sectors = 1 << 5;
294 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
295 }
296 return 0;
297}
298
299static const struct block_device_operations virtblk_fops = {
300 .ioctl = virtblk_ioctl,
301 .owner = THIS_MODULE,
302 .getgeo = virtblk_getgeo,
303};
304
305static int index_to_minor(int index)
306{
307 return index << PART_BITS;
308}
309
310static int minor_to_index(int minor)
311{
312 return minor >> PART_BITS;
313}
314
315static ssize_t virtblk_serial_show(struct device *dev,
316 struct device_attribute *attr, char *buf)
317{
318 struct gendisk *disk = dev_to_disk(dev);
319 int err;
320
321 /* sysfs gives us a PAGE_SIZE buffer */
322 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
323
324 buf[VIRTIO_BLK_ID_BYTES] = '\0';
325 err = virtblk_get_id(disk, buf);
326 if (!err)
327 return strlen(buf);
328
329 if (err == -EIO) /* Unsupported? Make it empty. */
330 return 0;
331
332 return err;
333}
334
335static DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
336
337static void virtblk_config_changed_work(struct work_struct *work)
338{
339 struct virtio_blk *vblk =
340 container_of(work, struct virtio_blk, config_work);
341 struct virtio_device *vdev = vblk->vdev;
342 struct request_queue *q = vblk->disk->queue;
343 char cap_str_2[10], cap_str_10[10];
344 char *envp[] = { "RESIZE=1", NULL };
345 u64 capacity;
346
347 /* Host must always specify the capacity. */
348 virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
349
350 /* If capacity is too big, truncate with warning. */
351 if ((sector_t)capacity != capacity) {
352 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
353 (unsigned long long)capacity);
354 capacity = (sector_t)-1;
355 }
356
357 string_get_size(capacity, queue_logical_block_size(q),
358 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
359 string_get_size(capacity, queue_logical_block_size(q),
360 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
361
362 dev_notice(&vdev->dev,
363 "new size: %llu %d-byte logical blocks (%s/%s)\n",
364 (unsigned long long)capacity,
365 queue_logical_block_size(q),
366 cap_str_10, cap_str_2);
367
368 set_capacity(vblk->disk, capacity);
369 revalidate_disk(vblk->disk);
370 kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
371}
372
373static void virtblk_config_changed(struct virtio_device *vdev)
374{
375 struct virtio_blk *vblk = vdev->priv;
376
377 queue_work(virtblk_wq, &vblk->config_work);
378}
379
380static int init_vq(struct virtio_blk *vblk)
381{
382 int err = 0;
383 int i;
384 vq_callback_t **callbacks;
385 const char **names;
386 struct virtqueue **vqs;
387 unsigned short num_vqs;
388 struct virtio_device *vdev = vblk->vdev;
389
390 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
391 struct virtio_blk_config, num_queues,
392 &num_vqs);
393 if (err)
394 num_vqs = 1;
395
396 vblk->vqs = kmalloc(sizeof(*vblk->vqs) * num_vqs, GFP_KERNEL);
397 if (!vblk->vqs) {
398 err = -ENOMEM;
399 goto out;
400 }
401
402 names = kmalloc(sizeof(*names) * num_vqs, GFP_KERNEL);
403 if (!names)
404 goto err_names;
405
406 callbacks = kmalloc(sizeof(*callbacks) * num_vqs, GFP_KERNEL);
407 if (!callbacks)
408 goto err_callbacks;
409
410 vqs = kmalloc(sizeof(*vqs) * num_vqs, GFP_KERNEL);
411 if (!vqs)
412 goto err_vqs;
413
414 for (i = 0; i < num_vqs; i++) {
415 callbacks[i] = virtblk_done;
416 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
417 names[i] = vblk->vqs[i].name;
418 }
419
420 /* Discover virtqueues and write information to configuration. */
421 err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names);
422 if (err)
423 goto err_find_vqs;
424
425 for (i = 0; i < num_vqs; i++) {
426 spin_lock_init(&vblk->vqs[i].lock);
427 vblk->vqs[i].vq = vqs[i];
428 }
429 vblk->num_vqs = num_vqs;
430
431 err_find_vqs:
432 kfree(vqs);
433 err_vqs:
434 kfree(callbacks);
435 err_callbacks:
436 kfree(names);
437 err_names:
438 if (err)
439 kfree(vblk->vqs);
440 out:
441 return err;
442}
443
444/*
445 * Legacy naming scheme used for virtio devices. We are stuck with it for
446 * virtio blk but don't ever use it for any new driver.
447 */
448static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
449{
450 const int base = 'z' - 'a' + 1;
451 char *begin = buf + strlen(prefix);
452 char *end = buf + buflen;
453 char *p;
454 int unit;
455
456 p = end - 1;
457 *p = '\0';
458 unit = base;
459 do {
460 if (p == begin)
461 return -EINVAL;
462 *--p = 'a' + (index % unit);
463 index = (index / unit) - 1;
464 } while (index >= 0);
465
466 memmove(begin, p, end - p);
467 memcpy(buf, prefix, strlen(prefix));
468
469 return 0;
470}
471
472static int virtblk_get_cache_mode(struct virtio_device *vdev)
473{
474 u8 writeback;
475 int err;
476
477 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
478 struct virtio_blk_config, wce,
479 &writeback);
480
481 /*
482 * If WCE is not configurable and flush is not available,
483 * assume no writeback cache is in use.
484 */
485 if (err)
486 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
487
488 return writeback;
489}
490
491static void virtblk_update_cache_mode(struct virtio_device *vdev)
492{
493 u8 writeback = virtblk_get_cache_mode(vdev);
494 struct virtio_blk *vblk = vdev->priv;
495
496 if (writeback)
497 blk_queue_flush(vblk->disk->queue, REQ_FLUSH);
498 else
499 blk_queue_flush(vblk->disk->queue, 0);
500
501 revalidate_disk(vblk->disk);
502}
503
504static const char *const virtblk_cache_types[] = {
505 "write through", "write back"
506};
507
508static ssize_t
509virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
510 const char *buf, size_t count)
511{
512 struct gendisk *disk = dev_to_disk(dev);
513 struct virtio_blk *vblk = disk->private_data;
514 struct virtio_device *vdev = vblk->vdev;
515 int i;
516
517 BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
518 for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
519 if (sysfs_streq(buf, virtblk_cache_types[i]))
520 break;
521
522 if (i < 0)
523 return -EINVAL;
524
525 virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
526 virtblk_update_cache_mode(vdev);
527 return count;
528}
529
530static ssize_t
531virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
532 char *buf)
533{
534 struct gendisk *disk = dev_to_disk(dev);
535 struct virtio_blk *vblk = disk->private_data;
536 u8 writeback = virtblk_get_cache_mode(vblk->vdev);
537
538 BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
539 return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
540}
541
542static const struct device_attribute dev_attr_cache_type_ro =
543 __ATTR(cache_type, S_IRUGO,
544 virtblk_cache_type_show, NULL);
545static const struct device_attribute dev_attr_cache_type_rw =
546 __ATTR(cache_type, S_IRUGO|S_IWUSR,
547 virtblk_cache_type_show, virtblk_cache_type_store);
548
549static int virtblk_init_request(void *data, struct request *rq,
550 unsigned int hctx_idx, unsigned int request_idx,
551 unsigned int numa_node)
552{
553 struct virtio_blk *vblk = data;
554 struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
555
556 sg_init_table(vbr->sg, vblk->sg_elems);
557 return 0;
558}
559
560static struct blk_mq_ops virtio_mq_ops = {
561 .queue_rq = virtio_queue_rq,
562 .map_queue = blk_mq_map_queue,
563 .complete = virtblk_request_done,
564 .init_request = virtblk_init_request,
565};
566
567static unsigned int virtblk_queue_depth;
568module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
569
570static int virtblk_probe(struct virtio_device *vdev)
571{
572 struct virtio_blk *vblk;
573 struct request_queue *q;
574 int err, index;
575
576 u64 cap;
577 u32 v, blk_size, sg_elems, opt_io_size;
578 u16 min_io_size;
579 u8 physical_block_exp, alignment_offset;
580
581 if (!vdev->config->get) {
582 dev_err(&vdev->dev, "%s failure: config access disabled\n",
583 __func__);
584 return -EINVAL;
585 }
586
587 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
588 GFP_KERNEL);
589 if (err < 0)
590 goto out;
591 index = err;
592
593 /* We need to know how many segments before we allocate. */
594 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
595 struct virtio_blk_config, seg_max,
596 &sg_elems);
597
598 /* We need at least one SG element, whatever they say. */
599 if (err || !sg_elems)
600 sg_elems = 1;
601
602 /* We need an extra sg elements at head and tail. */
603 sg_elems += 2;
604 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
605 if (!vblk) {
606 err = -ENOMEM;
607 goto out_free_index;
608 }
609
610 vblk->vdev = vdev;
611 vblk->sg_elems = sg_elems;
612
613 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
614
615 err = init_vq(vblk);
616 if (err)
617 goto out_free_vblk;
618
619 /* FIXME: How many partitions? How long is a piece of string? */
620 vblk->disk = alloc_disk(1 << PART_BITS);
621 if (!vblk->disk) {
622 err = -ENOMEM;
623 goto out_free_vq;
624 }
625
626 /* Default queue sizing is to fill the ring. */
627 if (!virtblk_queue_depth) {
628 virtblk_queue_depth = vblk->vqs[0].vq->num_free;
629 /* ... but without indirect descs, we use 2 descs per req */
630 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
631 virtblk_queue_depth /= 2;
632 }
633
634 memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
635 vblk->tag_set.ops = &virtio_mq_ops;
636 vblk->tag_set.queue_depth = virtblk_queue_depth;
637 vblk->tag_set.numa_node = NUMA_NO_NODE;
638 vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
639 vblk->tag_set.cmd_size =
640 sizeof(struct virtblk_req) +
641 sizeof(struct scatterlist) * sg_elems;
642 vblk->tag_set.driver_data = vblk;
643 vblk->tag_set.nr_hw_queues = vblk->num_vqs;
644
645 err = blk_mq_alloc_tag_set(&vblk->tag_set);
646 if (err)
647 goto out_put_disk;
648
649 q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set);
650 if (IS_ERR(q)) {
651 err = -ENOMEM;
652 goto out_free_tags;
653 }
654
655 q->queuedata = vblk;
656
657 virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
658
659 vblk->disk->major = major;
660 vblk->disk->first_minor = index_to_minor(index);
661 vblk->disk->private_data = vblk;
662 vblk->disk->fops = &virtblk_fops;
663 vblk->disk->driverfs_dev = &vdev->dev;
664 vblk->disk->flags |= GENHD_FL_EXT_DEVT;
665 vblk->index = index;
666
667 /* configure queue flush support */
668 virtblk_update_cache_mode(vdev);
669
670 /* If disk is read-only in the host, the guest should obey */
671 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
672 set_disk_ro(vblk->disk, 1);
673
674 /* Host must always specify the capacity. */
675 virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
676
677 /* If capacity is too big, truncate with warning. */
678 if ((sector_t)cap != cap) {
679 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
680 (unsigned long long)cap);
681 cap = (sector_t)-1;
682 }
683 set_capacity(vblk->disk, cap);
684
685 /* We can handle whatever the host told us to handle. */
686 blk_queue_max_segments(q, vblk->sg_elems-2);
687
688 /* No need to bounce any requests */
689 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
690
691 /* No real sector limit. */
692 blk_queue_max_hw_sectors(q, -1U);
693
694 /* Host can optionally specify maximum segment size and number of
695 * segments. */
696 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
697 struct virtio_blk_config, size_max, &v);
698 if (!err)
699 blk_queue_max_segment_size(q, v);
700 else
701 blk_queue_max_segment_size(q, -1U);
702
703 /* Host can optionally specify the block size of the device */
704 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
705 struct virtio_blk_config, blk_size,
706 &blk_size);
707 if (!err)
708 blk_queue_logical_block_size(q, blk_size);
709 else
710 blk_size = queue_logical_block_size(q);
711
712 /* Use topology information if available */
713 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
714 struct virtio_blk_config, physical_block_exp,
715 &physical_block_exp);
716 if (!err && physical_block_exp)
717 blk_queue_physical_block_size(q,
718 blk_size * (1 << physical_block_exp));
719
720 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
721 struct virtio_blk_config, alignment_offset,
722 &alignment_offset);
723 if (!err && alignment_offset)
724 blk_queue_alignment_offset(q, blk_size * alignment_offset);
725
726 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
727 struct virtio_blk_config, min_io_size,
728 &min_io_size);
729 if (!err && min_io_size)
730 blk_queue_io_min(q, blk_size * min_io_size);
731
732 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
733 struct virtio_blk_config, opt_io_size,
734 &opt_io_size);
735 if (!err && opt_io_size)
736 blk_queue_io_opt(q, blk_size * opt_io_size);
737
738 virtio_device_ready(vdev);
739
740 add_disk(vblk->disk);
741 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
742 if (err)
743 goto out_del_disk;
744
745 if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
746 err = device_create_file(disk_to_dev(vblk->disk),
747 &dev_attr_cache_type_rw);
748 else
749 err = device_create_file(disk_to_dev(vblk->disk),
750 &dev_attr_cache_type_ro);
751 if (err)
752 goto out_del_disk;
753 return 0;
754
755out_del_disk:
756 del_gendisk(vblk->disk);
757 blk_cleanup_queue(vblk->disk->queue);
758out_free_tags:
759 blk_mq_free_tag_set(&vblk->tag_set);
760out_put_disk:
761 put_disk(vblk->disk);
762out_free_vq:
763 vdev->config->del_vqs(vdev);
764out_free_vblk:
765 kfree(vblk);
766out_free_index:
767 ida_simple_remove(&vd_index_ida, index);
768out:
769 return err;
770}
771
772static void virtblk_remove(struct virtio_device *vdev)
773{
774 struct virtio_blk *vblk = vdev->priv;
775 int index = vblk->index;
776 int refc;
777
778 /* Make sure no work handler is accessing the device. */
779 flush_work(&vblk->config_work);
780
781 del_gendisk(vblk->disk);
782 blk_cleanup_queue(vblk->disk->queue);
783
784 blk_mq_free_tag_set(&vblk->tag_set);
785
786 /* Stop all the virtqueues. */
787 vdev->config->reset(vdev);
788
789 refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount);
790 put_disk(vblk->disk);
791 vdev->config->del_vqs(vdev);
792 kfree(vblk->vqs);
793 kfree(vblk);
794
795 /* Only free device id if we don't have any users */
796 if (refc == 1)
797 ida_simple_remove(&vd_index_ida, index);
798}
799
800#ifdef CONFIG_PM_SLEEP
801static int virtblk_freeze(struct virtio_device *vdev)
802{
803 struct virtio_blk *vblk = vdev->priv;
804
805 /* Ensure we don't receive any more interrupts */
806 vdev->config->reset(vdev);
807
808 /* Make sure no work handler is accessing the device. */
809 flush_work(&vblk->config_work);
810
811 blk_mq_stop_hw_queues(vblk->disk->queue);
812
813 vdev->config->del_vqs(vdev);
814 return 0;
815}
816
817static int virtblk_restore(struct virtio_device *vdev)
818{
819 struct virtio_blk *vblk = vdev->priv;
820 int ret;
821
822 ret = init_vq(vdev->priv);
823 if (ret)
824 return ret;
825
826 virtio_device_ready(vdev);
827
828 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
829 return 0;
830}
831#endif
832
833static const struct virtio_device_id id_table[] = {
834 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
835 { 0 },
836};
837
838static unsigned int features_legacy[] = {
839 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
840 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
841 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
842 VIRTIO_BLK_F_MQ,
843}
844;
845static unsigned int features[] = {
846 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
847 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
848 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
849 VIRTIO_BLK_F_MQ,
850};
851
852static struct virtio_driver virtio_blk = {
853 .feature_table = features,
854 .feature_table_size = ARRAY_SIZE(features),
855 .feature_table_legacy = features_legacy,
856 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
857 .driver.name = KBUILD_MODNAME,
858 .driver.owner = THIS_MODULE,
859 .id_table = id_table,
860 .probe = virtblk_probe,
861 .remove = virtblk_remove,
862 .config_changed = virtblk_config_changed,
863#ifdef CONFIG_PM_SLEEP
864 .freeze = virtblk_freeze,
865 .restore = virtblk_restore,
866#endif
867};
868
869static int __init init(void)
870{
871 int error;
872
873 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
874 if (!virtblk_wq)
875 return -ENOMEM;
876
877 major = register_blkdev(0, "virtblk");
878 if (major < 0) {
879 error = major;
880 goto out_destroy_workqueue;
881 }
882
883 error = register_virtio_driver(&virtio_blk);
884 if (error)
885 goto out_unregister_blkdev;
886 return 0;
887
888out_unregister_blkdev:
889 unregister_blkdev(major, "virtblk");
890out_destroy_workqueue:
891 destroy_workqueue(virtblk_wq);
892 return error;
893}
894
895static void __exit fini(void)
896{
897 unregister_virtio_driver(&virtio_blk);
898 unregister_blkdev(major, "virtblk");
899 destroy_workqueue(virtblk_wq);
900}
901module_init(init);
902module_exit(fini);
903
904MODULE_DEVICE_TABLE(virtio, id_table);
905MODULE_DESCRIPTION("Virtio block driver");
906MODULE_LICENSE("GPL");