?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mm, page_alloc: set alloc_flags only once in slowpath
[linux/fpc-iii.git] / drivers / block / virtio_blk.c
blob1523e05c46fc95b29c47af3b51ebdc9f93af9029
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
20 static int major;
21 static DEFINE_IDA(vd_index_ida);
22
23 static struct workqueue_struct *virtblk_wq;
24
25 struct virtio_blk_vq {
26 struct virtqueue *vq;
27 spinlock_t lock;
28 char name[VQ_NAME_LEN];
29 } ____cacheline_aligned_in_smp;
30
31 struct 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
54 struct 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
62 static 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
74 static 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
117 static 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
134 static 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
160 static 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_op(req) == REQ_OP_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 */
236 static int virtblk_get_id(struct gendisk *disk, char *id_str)
237 {
238 struct virtio_blk *vblk = disk->private_data;
239 struct request_queue *q = vblk->disk->queue;
240 struct request *req;
241 int err;
242
243 req = blk_get_request(q, READ, GFP_KERNEL);
244 if (IS_ERR(req))
245 return PTR_ERR(req);
246 req->cmd_type = REQ_TYPE_DRV_PRIV;
247
248 err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL);
249 if (err)
250 goto out;
251
252 err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
253 out:
254 blk_put_request(req);
255 return err;
256 }
257
258 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
259 unsigned int cmd, unsigned long data)
260 {
261 struct gendisk *disk = bdev->bd_disk;
262 struct virtio_blk *vblk = disk->private_data;
263
264 /*
265 * Only allow the generic SCSI ioctls if the host can support it.
266 */
267 if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
268 return -ENOTTY;
269
270 return scsi_cmd_blk_ioctl(bdev, mode, cmd,
271 (void __user *)data);
272 }
273
274 /* We provide getgeo only to please some old bootloader/partitioning tools */
275 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
276 {
277 struct virtio_blk *vblk = bd->bd_disk->private_data;
278
279 /* see if the host passed in geometry config */
280 if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) {
281 virtio_cread(vblk->vdev, struct virtio_blk_config,
282 geometry.cylinders, &geo->cylinders);
283 virtio_cread(vblk->vdev, struct virtio_blk_config,
284 geometry.heads, &geo->heads);
285 virtio_cread(vblk->vdev, struct virtio_blk_config,
286 geometry.sectors, &geo->sectors);
287 } else {
288 /* some standard values, similar to sd */
289 geo->heads = 1 << 6;
290 geo->sectors = 1 << 5;
291 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
292 }
293 return 0;
294 }
295
296 static const struct block_device_operations virtblk_fops = {
297 .ioctl = virtblk_ioctl,
298 .owner = THIS_MODULE,
299 .getgeo = virtblk_getgeo,
300 };
301
302 static int index_to_minor(int index)
303 {
304 return index << PART_BITS;
305 }
306
307 static int minor_to_index(int minor)
308 {
309 return minor >> PART_BITS;
310 }
311
312 static ssize_t virtblk_serial_show(struct device *dev,
313 struct device_attribute *attr, char *buf)
314 {
315 struct gendisk *disk = dev_to_disk(dev);
316 int err;
317
318 /* sysfs gives us a PAGE_SIZE buffer */
319 BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
320
321 buf[VIRTIO_BLK_ID_BYTES] = '\0';
322 err = virtblk_get_id(disk, buf);
323 if (!err)
324 return strlen(buf);
325
326 if (err == -EIO) /* Unsupported? Make it empty. */
327 return 0;
328
329 return err;
330 }
331
332 static DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
333
334 static void virtblk_config_changed_work(struct work_struct *work)
335 {
336 struct virtio_blk *vblk =
337 container_of(work, struct virtio_blk, config_work);
338 struct virtio_device *vdev = vblk->vdev;
339 struct request_queue *q = vblk->disk->queue;
340 char cap_str_2[10], cap_str_10[10];
341 char *envp[] = { "RESIZE=1", NULL };
342 u64 capacity;
343
344 /* Host must always specify the capacity. */
345 virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
346
347 /* If capacity is too big, truncate with warning. */
348 if ((sector_t)capacity != capacity) {
349 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
350 (unsigned long long)capacity);
351 capacity = (sector_t)-1;
352 }
353
354 string_get_size(capacity, queue_logical_block_size(q),
355 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
356 string_get_size(capacity, queue_logical_block_size(q),
357 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
358
359 dev_notice(&vdev->dev,
360 "new size: %llu %d-byte logical blocks (%s/%s)\n",
361 (unsigned long long)capacity,
362 queue_logical_block_size(q),
363 cap_str_10, cap_str_2);
364
365 set_capacity(vblk->disk, capacity);
366 revalidate_disk(vblk->disk);
367 kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp);
368 }
369
370 static void virtblk_config_changed(struct virtio_device *vdev)
371 {
372 struct virtio_blk *vblk = vdev->priv;
373
374 queue_work(virtblk_wq, &vblk->config_work);
375 }
376
377 static int init_vq(struct virtio_blk *vblk)
378 {
379 int err = 0;
380 int i;
381 vq_callback_t **callbacks;
382 const char **names;
383 struct virtqueue **vqs;
384 unsigned short num_vqs;
385 struct virtio_device *vdev = vblk->vdev;
386
387 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ,
388 struct virtio_blk_config, num_queues,
389 &num_vqs);
390 if (err)
391 num_vqs = 1;
392
393 vblk->vqs = kmalloc(sizeof(*vblk->vqs) * num_vqs, GFP_KERNEL);
394 if (!vblk->vqs) {
395 err = -ENOMEM;
396 goto out;
397 }
398
399 names = kmalloc(sizeof(*names) * num_vqs, GFP_KERNEL);
400 if (!names)
401 goto err_names;
402
403 callbacks = kmalloc(sizeof(*callbacks) * num_vqs, GFP_KERNEL);
404 if (!callbacks)
405 goto err_callbacks;
406
407 vqs = kmalloc(sizeof(*vqs) * num_vqs, GFP_KERNEL);
408 if (!vqs)
409 goto err_vqs;
410
411 for (i = 0; i < num_vqs; i++) {
412 callbacks[i] = virtblk_done;
413 snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i);
414 names[i] = vblk->vqs[i].name;
415 }
416
417 /* Discover virtqueues and write information to configuration. */
418 err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names);
419 if (err)
420 goto err_find_vqs;
421
422 for (i = 0; i < num_vqs; i++) {
423 spin_lock_init(&vblk->vqs[i].lock);
424 vblk->vqs[i].vq = vqs[i];
425 }
426 vblk->num_vqs = num_vqs;
427
428 err_find_vqs:
429 kfree(vqs);
430 err_vqs:
431 kfree(callbacks);
432 err_callbacks:
433 kfree(names);
434 err_names:
435 if (err)
436 kfree(vblk->vqs);
437 out:
438 return err;
439 }
440
441 /*
442 * Legacy naming scheme used for virtio devices. We are stuck with it for
443 * virtio blk but don't ever use it for any new driver.
444 */
445 static int virtblk_name_format(char *prefix, int index, char *buf, int buflen)
446 {
447 const int base = 'z' - 'a' + 1;
448 char *begin = buf + strlen(prefix);
449 char *end = buf + buflen;
450 char *p;
451 int unit;
452
453 p = end - 1;
454 *p = '\0';
455 unit = base;
456 do {
457 if (p == begin)
458 return -EINVAL;
459 *--p = 'a' + (index % unit);
460 index = (index / unit) - 1;
461 } while (index >= 0);
462
463 memmove(begin, p, end - p);
464 memcpy(buf, prefix, strlen(prefix));
465
466 return 0;
467 }
468
469 static int virtblk_get_cache_mode(struct virtio_device *vdev)
470 {
471 u8 writeback;
472 int err;
473
474 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE,
475 struct virtio_blk_config, wce,
476 &writeback);
477
478 /*
479 * If WCE is not configurable and flush is not available,
480 * assume no writeback cache is in use.
481 */
482 if (err)
483 writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH);
484
485 return writeback;
486 }
487
488 static void virtblk_update_cache_mode(struct virtio_device *vdev)
489 {
490 u8 writeback = virtblk_get_cache_mode(vdev);
491 struct virtio_blk *vblk = vdev->priv;
492
493 blk_queue_write_cache(vblk->disk->queue, writeback, false);
494 revalidate_disk(vblk->disk);
495 }
496
497 static const char *const virtblk_cache_types[] = {
498 "write through", "write back"
499 };
500
501 static ssize_t
502 virtblk_cache_type_store(struct device *dev, struct device_attribute *attr,
503 const char *buf, size_t count)
504 {
505 struct gendisk *disk = dev_to_disk(dev);
506 struct virtio_blk *vblk = disk->private_data;
507 struct virtio_device *vdev = vblk->vdev;
508 int i;
509
510 BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE));
511 for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; )
512 if (sysfs_streq(buf, virtblk_cache_types[i]))
513 break;
514
515 if (i < 0)
516 return -EINVAL;
517
518 virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i);
519 virtblk_update_cache_mode(vdev);
520 return count;
521 }
522
523 static ssize_t
524 virtblk_cache_type_show(struct device *dev, struct device_attribute *attr,
525 char *buf)
526 {
527 struct gendisk *disk = dev_to_disk(dev);
528 struct virtio_blk *vblk = disk->private_data;
529 u8 writeback = virtblk_get_cache_mode(vblk->vdev);
530
531 BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types));
532 return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]);
533 }
534
535 static const struct device_attribute dev_attr_cache_type_ro =
536 __ATTR(cache_type, S_IRUGO,
537 virtblk_cache_type_show, NULL);
538 static const struct device_attribute dev_attr_cache_type_rw =
539 __ATTR(cache_type, S_IRUGO|S_IWUSR,
540 virtblk_cache_type_show, virtblk_cache_type_store);
541
542 static int virtblk_init_request(void *data, struct request *rq,
543 unsigned int hctx_idx, unsigned int request_idx,
544 unsigned int numa_node)
545 {
546 struct virtio_blk *vblk = data;
547 struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq);
548
549 sg_init_table(vbr->sg, vblk->sg_elems);
550 return 0;
551 }
552
553 static struct blk_mq_ops virtio_mq_ops = {
554 .queue_rq = virtio_queue_rq,
555 .map_queue = blk_mq_map_queue,
556 .complete = virtblk_request_done,
557 .init_request = virtblk_init_request,
558 };
559
560 static unsigned int virtblk_queue_depth;
561 module_param_named(queue_depth, virtblk_queue_depth, uint, 0444);
562
563 static int virtblk_probe(struct virtio_device *vdev)
564 {
565 struct virtio_blk *vblk;
566 struct request_queue *q;
567 int err, index;
568
569 u64 cap;
570 u32 v, blk_size, sg_elems, opt_io_size;
571 u16 min_io_size;
572 u8 physical_block_exp, alignment_offset;
573
574 if (!vdev->config->get) {
575 dev_err(&vdev->dev, "%s failure: config access disabled\n",
576 __func__);
577 return -EINVAL;
578 }
579
580 err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS),
581 GFP_KERNEL);
582 if (err < 0)
583 goto out;
584 index = err;
585
586 /* We need to know how many segments before we allocate. */
587 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX,
588 struct virtio_blk_config, seg_max,
589 &sg_elems);
590
591 /* We need at least one SG element, whatever they say. */
592 if (err || !sg_elems)
593 sg_elems = 1;
594
595 /* We need an extra sg elements at head and tail. */
596 sg_elems += 2;
597 vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
598 if (!vblk) {
599 err = -ENOMEM;
600 goto out_free_index;
601 }
602
603 vblk->vdev = vdev;
604 vblk->sg_elems = sg_elems;
605
606 INIT_WORK(&vblk->config_work, virtblk_config_changed_work);
607
608 err = init_vq(vblk);
609 if (err)
610 goto out_free_vblk;
611
612 /* FIXME: How many partitions? How long is a piece of string? */
613 vblk->disk = alloc_disk(1 << PART_BITS);
614 if (!vblk->disk) {
615 err = -ENOMEM;
616 goto out_free_vq;
617 }
618
619 /* Default queue sizing is to fill the ring. */
620 if (!virtblk_queue_depth) {
621 virtblk_queue_depth = vblk->vqs[0].vq->num_free;
622 /* ... but without indirect descs, we use 2 descs per req */
623 if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC))
624 virtblk_queue_depth /= 2;
625 }
626
627 memset(&vblk->tag_set, 0, sizeof(vblk->tag_set));
628 vblk->tag_set.ops = &virtio_mq_ops;
629 vblk->tag_set.queue_depth = virtblk_queue_depth;
630 vblk->tag_set.numa_node = NUMA_NO_NODE;
631 vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
632 vblk->tag_set.cmd_size =
633 sizeof(struct virtblk_req) +
634 sizeof(struct scatterlist) * sg_elems;
635 vblk->tag_set.driver_data = vblk;
636 vblk->tag_set.nr_hw_queues = vblk->num_vqs;
637
638 err = blk_mq_alloc_tag_set(&vblk->tag_set);
639 if (err)
640 goto out_put_disk;
641
642 q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set);
643 if (IS_ERR(q)) {
644 err = -ENOMEM;
645 goto out_free_tags;
646 }
647
648 q->queuedata = vblk;
649
650 virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN);
651
652 vblk->disk->major = major;
653 vblk->disk->first_minor = index_to_minor(index);
654 vblk->disk->private_data = vblk;
655 vblk->disk->fops = &virtblk_fops;
656 vblk->disk->flags |= GENHD_FL_EXT_DEVT;
657 vblk->index = index;
658
659 /* configure queue flush support */
660 virtblk_update_cache_mode(vdev);
661
662 /* If disk is read-only in the host, the guest should obey */
663 if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
664 set_disk_ro(vblk->disk, 1);
665
666 /* Host must always specify the capacity. */
667 virtio_cread(vdev, struct virtio_blk_config, capacity, &cap);
668
669 /* If capacity is too big, truncate with warning. */
670 if ((sector_t)cap != cap) {
671 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
672 (unsigned long long)cap);
673 cap = (sector_t)-1;
674 }
675 set_capacity(vblk->disk, cap);
676
677 /* We can handle whatever the host told us to handle. */
678 blk_queue_max_segments(q, vblk->sg_elems-2);
679
680 /* No need to bounce any requests */
681 blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
682
683 /* No real sector limit. */
684 blk_queue_max_hw_sectors(q, -1U);
685
686 /* Host can optionally specify maximum segment size and number of
687 * segments. */
688 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX,
689 struct virtio_blk_config, size_max, &v);
690 if (!err)
691 blk_queue_max_segment_size(q, v);
692 else
693 blk_queue_max_segment_size(q, -1U);
694
695 /* Host can optionally specify the block size of the device */
696 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE,
697 struct virtio_blk_config, blk_size,
698 &blk_size);
699 if (!err)
700 blk_queue_logical_block_size(q, blk_size);
701 else
702 blk_size = queue_logical_block_size(q);
703
704 /* Use topology information if available */
705 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
706 struct virtio_blk_config, physical_block_exp,
707 &physical_block_exp);
708 if (!err && physical_block_exp)
709 blk_queue_physical_block_size(q,
710 blk_size * (1 << physical_block_exp));
711
712 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
713 struct virtio_blk_config, alignment_offset,
714 &alignment_offset);
715 if (!err && alignment_offset)
716 blk_queue_alignment_offset(q, blk_size * alignment_offset);
717
718 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
719 struct virtio_blk_config, min_io_size,
720 &min_io_size);
721 if (!err && min_io_size)
722 blk_queue_io_min(q, blk_size * min_io_size);
723
724 err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY,
725 struct virtio_blk_config, opt_io_size,
726 &opt_io_size);
727 if (!err && opt_io_size)
728 blk_queue_io_opt(q, blk_size * opt_io_size);
729
730 virtio_device_ready(vdev);
731
732 device_add_disk(&vdev->dev, vblk->disk);
733 err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
734 if (err)
735 goto out_del_disk;
736
737 if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE))
738 err = device_create_file(disk_to_dev(vblk->disk),
739 &dev_attr_cache_type_rw);
740 else
741 err = device_create_file(disk_to_dev(vblk->disk),
742 &dev_attr_cache_type_ro);
743 if (err)
744 goto out_del_disk;
745 return 0;
746
747 out_del_disk:
748 del_gendisk(vblk->disk);
749 blk_cleanup_queue(vblk->disk->queue);
750 out_free_tags:
751 blk_mq_free_tag_set(&vblk->tag_set);
752 out_put_disk:
753 put_disk(vblk->disk);
754 out_free_vq:
755 vdev->config->del_vqs(vdev);
756 out_free_vblk:
757 kfree(vblk);
758 out_free_index:
759 ida_simple_remove(&vd_index_ida, index);
760 out:
761 return err;
762 }
763
764 static void virtblk_remove(struct virtio_device *vdev)
765 {
766 struct virtio_blk *vblk = vdev->priv;
767 int index = vblk->index;
768 int refc;
769
770 /* Make sure no work handler is accessing the device. */
771 flush_work(&vblk->config_work);
772
773 del_gendisk(vblk->disk);
774 blk_cleanup_queue(vblk->disk->queue);
775
776 blk_mq_free_tag_set(&vblk->tag_set);
777
778 /* Stop all the virtqueues. */
779 vdev->config->reset(vdev);
780
781 refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount);
782 put_disk(vblk->disk);
783 vdev->config->del_vqs(vdev);
784 kfree(vblk->vqs);
785 kfree(vblk);
786
787 /* Only free device id if we don't have any users */
788 if (refc == 1)
789 ida_simple_remove(&vd_index_ida, index);
790 }
791
792 #ifdef CONFIG_PM_SLEEP
793 static int virtblk_freeze(struct virtio_device *vdev)
794 {
795 struct virtio_blk *vblk = vdev->priv;
796
797 /* Ensure we don't receive any more interrupts */
798 vdev->config->reset(vdev);
799
800 /* Make sure no work handler is accessing the device. */
801 flush_work(&vblk->config_work);
802
803 blk_mq_stop_hw_queues(vblk->disk->queue);
804
805 vdev->config->del_vqs(vdev);
806 return 0;
807 }
808
809 static int virtblk_restore(struct virtio_device *vdev)
810 {
811 struct virtio_blk *vblk = vdev->priv;
812 int ret;
813
814 ret = init_vq(vdev->priv);
815 if (ret)
816 return ret;
817
818 virtio_device_ready(vdev);
819
820 blk_mq_start_stopped_hw_queues(vblk->disk->queue, true);
821 return 0;
822 }
823 #endif
824
825 static const struct virtio_device_id id_table[] = {
826 { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
827 { 0 },
828 };
829
830 static unsigned int features_legacy[] = {
831 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
832 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI,
833 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
834 VIRTIO_BLK_F_MQ,
835 }
836 ;
837 static unsigned int features[] = {
838 VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY,
839 VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
840 VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE,
841 VIRTIO_BLK_F_MQ,
842 };
843
844 static struct virtio_driver virtio_blk = {
845 .feature_table = features,
846 .feature_table_size = ARRAY_SIZE(features),
847 .feature_table_legacy = features_legacy,
848 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
849 .driver.name = KBUILD_MODNAME,
850 .driver.owner = THIS_MODULE,
851 .id_table = id_table,
852 .probe = virtblk_probe,
853 .remove = virtblk_remove,
854 .config_changed = virtblk_config_changed,
855 #ifdef CONFIG_PM_SLEEP
856 .freeze = virtblk_freeze,
857 .restore = virtblk_restore,
858 #endif
859 };
860
861 static int __init init(void)
862 {
863 int error;
864
865 virtblk_wq = alloc_workqueue("virtio-blk", 0, 0);
866 if (!virtblk_wq)
867 return -ENOMEM;
868
869 major = register_blkdev(0, "virtblk");
870 if (major < 0) {
871 error = major;
872 goto out_destroy_workqueue;
873 }
874
875 error = register_virtio_driver(&virtio_blk);
876 if (error)
877 goto out_unregister_blkdev;
878 return 0;
879
880 out_unregister_blkdev:
881 unregister_blkdev(major, "virtblk");
882 out_destroy_workqueue:
883 destroy_workqueue(virtblk_wq);
884 return error;
885 }
886
887 static void __exit fini(void)
888 {
889 unregister_virtio_driver(&virtio_blk);
890 unregister_blkdev(major, "virtblk");
891 destroy_workqueue(virtblk_wq);
892 }
893 module_init(init);
894 module_exit(fini);
895
896 MODULE_DEVICE_TABLE(virtio, id_table);
897 MODULE_DESCRIPTION("Virtio block driver");
898 MODULE_LICENSE("GPL");
Linux with added FPC-III support