Commit | Line | Data |
---|---|---|
e467cde2 RR |
1 | //#define DEBUG |
2 | #include <linux/spinlock.h> | |
3 | #include <linux/blkdev.h> | |
4 | #include <linux/hdreg.h> | |
5 | #include <linux/virtio.h> | |
6 | #include <linux/virtio_blk.h> | |
3d1266c7 JA |
7 | #include <linux/scatterlist.h> |
8 | ||
4f3bf19c | 9 | #define PART_BITS 4 |
e467cde2 | 10 | |
d50ed907 | 11 | static int major, index; |
4f3bf19c | 12 | |
e467cde2 RR |
13 | struct virtio_blk |
14 | { | |
15 | spinlock_t lock; | |
16 | ||
17 | struct virtio_device *vdev; | |
18 | struct virtqueue *vq; | |
19 | ||
20 | /* The disk structure for the kernel. */ | |
21 | struct gendisk *disk; | |
22 | ||
23 | /* Request tracking. */ | |
24 | struct list_head reqs; | |
25 | ||
26 | mempool_t *pool; | |
27 | ||
0864b79a RR |
28 | /* What host tells us, plus 2 for header & tailer. */ |
29 | unsigned int sg_elems; | |
30 | ||
e467cde2 | 31 | /* Scatterlist: can be too big for stack. */ |
0864b79a | 32 | struct scatterlist sg[/*sg_elems*/]; |
e467cde2 RR |
33 | }; |
34 | ||
35 | struct virtblk_req | |
36 | { | |
37 | struct list_head list; | |
38 | struct request *req; | |
39 | struct virtio_blk_outhdr out_hdr; | |
1cde26f9 | 40 | struct virtio_scsi_inhdr in_hdr; |
cb38fa23 | 41 | u8 status; |
e467cde2 RR |
42 | }; |
43 | ||
18445c4d | 44 | static void blk_done(struct virtqueue *vq) |
e467cde2 RR |
45 | { |
46 | struct virtio_blk *vblk = vq->vdev->priv; | |
47 | struct virtblk_req *vbr; | |
48 | unsigned int len; | |
49 | unsigned long flags; | |
50 | ||
51 | spin_lock_irqsave(&vblk->lock, flags); | |
52 | while ((vbr = vblk->vq->vq_ops->get_buf(vblk->vq, &len)) != NULL) { | |
8316982a | 53 | int error; |
1cde26f9 | 54 | |
cb38fa23 | 55 | switch (vbr->status) { |
e467cde2 | 56 | case VIRTIO_BLK_S_OK: |
8316982a | 57 | error = 0; |
e467cde2 RR |
58 | break; |
59 | case VIRTIO_BLK_S_UNSUPP: | |
8316982a | 60 | error = -ENOTTY; |
e467cde2 RR |
61 | break; |
62 | default: | |
8316982a | 63 | error = -EIO; |
e467cde2 RR |
64 | break; |
65 | } | |
66 | ||
1cde26f9 HR |
67 | if (blk_pc_request(vbr->req)) { |
68 | vbr->req->resid_len = vbr->in_hdr.residual; | |
69 | vbr->req->sense_len = vbr->in_hdr.sense_len; | |
70 | vbr->req->errors = vbr->in_hdr.errors; | |
71 | } | |
72 | ||
40cbbb78 | 73 | __blk_end_request_all(vbr->req, error); |
e467cde2 RR |
74 | list_del(&vbr->list); |
75 | mempool_free(vbr, vblk->pool); | |
76 | } | |
77 | /* In case queue is stopped waiting for more buffers. */ | |
78 | blk_start_queue(vblk->disk->queue); | |
79 | spin_unlock_irqrestore(&vblk->lock, flags); | |
e467cde2 RR |
80 | } |
81 | ||
82 | static bool do_req(struct request_queue *q, struct virtio_blk *vblk, | |
83 | struct request *req) | |
84 | { | |
1cde26f9 | 85 | unsigned long num, out = 0, in = 0; |
e467cde2 RR |
86 | struct virtblk_req *vbr; |
87 | ||
88 | vbr = mempool_alloc(vblk->pool, GFP_ATOMIC); | |
89 | if (!vbr) | |
90 | /* When another request finishes we'll try again. */ | |
91 | return false; | |
92 | ||
93 | vbr->req = req; | |
94 | if (blk_fs_request(vbr->req)) { | |
95 | vbr->out_hdr.type = 0; | |
83096ebf | 96 | vbr->out_hdr.sector = blk_rq_pos(vbr->req); |
766ca442 | 97 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); |
e467cde2 RR |
98 | } else if (blk_pc_request(vbr->req)) { |
99 | vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; | |
100 | vbr->out_hdr.sector = 0; | |
766ca442 | 101 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); |
e467cde2 RR |
102 | } else { |
103 | /* We don't put anything else in the queue. */ | |
104 | BUG(); | |
105 | } | |
106 | ||
107 | if (blk_barrier_rq(vbr->req)) | |
108 | vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER; | |
109 | ||
1cde26f9 | 110 | sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr)); |
e467cde2 | 111 | |
1cde26f9 HR |
112 | /* |
113 | * If this is a packet command we need a couple of additional headers. | |
114 | * Behind the normal outhdr we put a segment with the scsi command | |
115 | * block, and before the normal inhdr we put the sense data and the | |
116 | * inhdr with additional status information before the normal inhdr. | |
117 | */ | |
118 | if (blk_pc_request(vbr->req)) | |
119 | sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len); | |
120 | ||
121 | num = blk_rq_map_sg(q, vbr->req, vblk->sg + out); | |
122 | ||
123 | if (blk_pc_request(vbr->req)) { | |
124 | sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96); | |
125 | sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr, | |
126 | sizeof(vbr->in_hdr)); | |
127 | } | |
128 | ||
129 | sg_set_buf(&vblk->sg[num + out + in++], &vbr->status, | |
130 | sizeof(vbr->status)); | |
131 | ||
132 | if (num) { | |
133 | if (rq_data_dir(vbr->req) == WRITE) { | |
134 | vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; | |
135 | out += num; | |
136 | } else { | |
137 | vbr->out_hdr.type |= VIRTIO_BLK_T_IN; | |
138 | in += num; | |
139 | } | |
e467cde2 RR |
140 | } |
141 | ||
142 | if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr)) { | |
143 | mempool_free(vbr, vblk->pool); | |
144 | return false; | |
145 | } | |
146 | ||
147 | list_add_tail(&vbr->list, &vblk->reqs); | |
148 | return true; | |
149 | } | |
150 | ||
151 | static void do_virtblk_request(struct request_queue *q) | |
152 | { | |
6c3b46f7 | 153 | struct virtio_blk *vblk = q->queuedata; |
e467cde2 RR |
154 | struct request *req; |
155 | unsigned int issued = 0; | |
156 | ||
9934c8c0 | 157 | while ((req = blk_peek_request(q)) != NULL) { |
0864b79a | 158 | BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); |
e467cde2 RR |
159 | |
160 | /* If this request fails, stop queue and wait for something to | |
161 | finish to restart it. */ | |
162 | if (!do_req(q, vblk, req)) { | |
163 | blk_stop_queue(q); | |
164 | break; | |
165 | } | |
9934c8c0 | 166 | blk_start_request(req); |
e467cde2 RR |
167 | issued++; |
168 | } | |
169 | ||
170 | if (issued) | |
171 | vblk->vq->vq_ops->kick(vblk->vq); | |
172 | } | |
173 | ||
1d589bb1 | 174 | /* return ATA identify data |
175 | */ | |
176 | static int virtblk_identify(struct gendisk *disk, void *argp) | |
177 | { | |
178 | struct virtio_blk *vblk = disk->private_data; | |
179 | void *opaque; | |
180 | int err = -ENOMEM; | |
181 | ||
182 | opaque = kmalloc(VIRTIO_BLK_ID_BYTES, GFP_KERNEL); | |
183 | if (!opaque) | |
184 | goto out; | |
185 | ||
186 | err = virtio_config_buf(vblk->vdev, VIRTIO_BLK_F_IDENTIFY, | |
187 | offsetof(struct virtio_blk_config, identify), opaque, | |
188 | VIRTIO_BLK_ID_BYTES); | |
189 | ||
190 | if (err) | |
191 | goto out_kfree; | |
192 | ||
193 | if (copy_to_user(argp, opaque, VIRTIO_BLK_ID_BYTES)) | |
194 | err = -EFAULT; | |
195 | ||
196 | out_kfree: | |
197 | kfree(opaque); | |
198 | out: | |
199 | return err; | |
200 | } | |
201 | ||
4e109852 | 202 | static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, |
e467cde2 RR |
203 | unsigned cmd, unsigned long data) |
204 | { | |
1cde26f9 HR |
205 | struct gendisk *disk = bdev->bd_disk; |
206 | struct virtio_blk *vblk = disk->private_data; | |
1d589bb1 | 207 | void __user *argp = (void __user *)data; |
208 | ||
209 | if (cmd == HDIO_GET_IDENTITY) | |
210 | return virtblk_identify(disk, argp); | |
1cde26f9 HR |
211 | |
212 | /* | |
213 | * Only allow the generic SCSI ioctls if the host can support it. | |
214 | */ | |
215 | if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) | |
216 | return -ENOIOCTLCMD; | |
217 | ||
1d589bb1 | 218 | return scsi_cmd_ioctl(disk->queue, disk, mode, cmd, argp); |
e467cde2 RR |
219 | } |
220 | ||
135da0b0 CB |
221 | /* We provide getgeo only to please some old bootloader/partitioning tools */ |
222 | static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
223 | { | |
48e4043d RH |
224 | struct virtio_blk *vblk = bd->bd_disk->private_data; |
225 | struct virtio_blk_geometry vgeo; | |
226 | int err; | |
227 | ||
228 | /* see if the host passed in geometry config */ | |
229 | err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY, | |
230 | offsetof(struct virtio_blk_config, geometry), | |
231 | &vgeo); | |
232 | ||
233 | if (!err) { | |
234 | geo->heads = vgeo.heads; | |
235 | geo->sectors = vgeo.sectors; | |
236 | geo->cylinders = vgeo.cylinders; | |
237 | } else { | |
238 | /* some standard values, similar to sd */ | |
239 | geo->heads = 1 << 6; | |
240 | geo->sectors = 1 << 5; | |
241 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
242 | } | |
135da0b0 CB |
243 | return 0; |
244 | } | |
245 | ||
e467cde2 | 246 | static struct block_device_operations virtblk_fops = { |
4e109852 | 247 | .locked_ioctl = virtblk_ioctl, |
135da0b0 CB |
248 | .owner = THIS_MODULE, |
249 | .getgeo = virtblk_getgeo, | |
e467cde2 RR |
250 | }; |
251 | ||
d50ed907 CB |
252 | static int index_to_minor(int index) |
253 | { | |
254 | return index << PART_BITS; | |
255 | } | |
256 | ||
98e94444 | 257 | static int __devinit virtblk_probe(struct virtio_device *vdev) |
e467cde2 RR |
258 | { |
259 | struct virtio_blk *vblk; | |
4f3bf19c | 260 | int err; |
e467cde2 RR |
261 | u64 cap; |
262 | u32 v; | |
0864b79a | 263 | u32 blk_size, sg_elems; |
e467cde2 | 264 | |
d50ed907 | 265 | if (index_to_minor(index) >= 1 << MINORBITS) |
4f3bf19c CB |
266 | return -ENOSPC; |
267 | ||
0864b79a RR |
268 | /* We need to know how many segments before we allocate. */ |
269 | err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX, | |
270 | offsetof(struct virtio_blk_config, seg_max), | |
271 | &sg_elems); | |
272 | if (err) | |
273 | sg_elems = 1; | |
274 | ||
275 | /* We need an extra sg elements at head and tail. */ | |
276 | sg_elems += 2; | |
277 | vdev->priv = vblk = kmalloc(sizeof(*vblk) + | |
278 | sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL); | |
e467cde2 RR |
279 | if (!vblk) { |
280 | err = -ENOMEM; | |
281 | goto out; | |
282 | } | |
283 | ||
284 | INIT_LIST_HEAD(&vblk->reqs); | |
285 | spin_lock_init(&vblk->lock); | |
286 | vblk->vdev = vdev; | |
0864b79a RR |
287 | vblk->sg_elems = sg_elems; |
288 | sg_init_table(vblk->sg, vblk->sg_elems); | |
e467cde2 RR |
289 | |
290 | /* We expect one virtqueue, for output. */ | |
d2a7ddda | 291 | vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests"); |
e467cde2 RR |
292 | if (IS_ERR(vblk->vq)) { |
293 | err = PTR_ERR(vblk->vq); | |
294 | goto out_free_vblk; | |
295 | } | |
296 | ||
297 | vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req)); | |
298 | if (!vblk->pool) { | |
299 | err = -ENOMEM; | |
300 | goto out_free_vq; | |
301 | } | |
302 | ||
e467cde2 | 303 | /* FIXME: How many partitions? How long is a piece of string? */ |
4f3bf19c | 304 | vblk->disk = alloc_disk(1 << PART_BITS); |
e467cde2 RR |
305 | if (!vblk->disk) { |
306 | err = -ENOMEM; | |
4f3bf19c | 307 | goto out_mempool; |
e467cde2 RR |
308 | } |
309 | ||
310 | vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock); | |
311 | if (!vblk->disk->queue) { | |
312 | err = -ENOMEM; | |
313 | goto out_put_disk; | |
314 | } | |
315 | ||
6c3b46f7 | 316 | vblk->disk->queue->queuedata = vblk; |
7d116b62 FLVC |
317 | queue_flag_set_unlocked(QUEUE_FLAG_VIRT, vblk->disk->queue); |
318 | ||
d50ed907 CB |
319 | if (index < 26) { |
320 | sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26); | |
321 | } else if (index < (26 + 1) * 26) { | |
322 | sprintf(vblk->disk->disk_name, "vd%c%c", | |
323 | 'a' + index / 26 - 1, 'a' + index % 26); | |
324 | } else { | |
325 | const unsigned int m1 = (index / 26 - 1) / 26 - 1; | |
326 | const unsigned int m2 = (index / 26 - 1) % 26; | |
327 | const unsigned int m3 = index % 26; | |
328 | sprintf(vblk->disk->disk_name, "vd%c%c%c", | |
329 | 'a' + m1, 'a' + m2, 'a' + m3); | |
330 | } | |
331 | ||
e467cde2 | 332 | vblk->disk->major = major; |
d50ed907 | 333 | vblk->disk->first_minor = index_to_minor(index); |
e467cde2 RR |
334 | vblk->disk->private_data = vblk; |
335 | vblk->disk->fops = &virtblk_fops; | |
c4839346 | 336 | vblk->disk->driverfs_dev = &vdev->dev; |
d50ed907 | 337 | index++; |
4f3bf19c | 338 | |
e467cde2 | 339 | /* If barriers are supported, tell block layer that queue is ordered */ |
c45a6816 | 340 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER)) |
e467cde2 RR |
341 | blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL); |
342 | ||
3ef53609 CB |
343 | /* If disk is read-only in the host, the guest should obey */ |
344 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) | |
345 | set_disk_ro(vblk->disk, 1); | |
346 | ||
a586d4f6 | 347 | /* Host must always specify the capacity. */ |
72e61eb4 RR |
348 | vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), |
349 | &cap, sizeof(cap)); | |
e467cde2 RR |
350 | |
351 | /* If capacity is too big, truncate with warning. */ | |
352 | if ((sector_t)cap != cap) { | |
353 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
354 | (unsigned long long)cap); | |
355 | cap = (sector_t)-1; | |
356 | } | |
357 | set_capacity(vblk->disk, cap); | |
358 | ||
0864b79a RR |
359 | /* We can handle whatever the host told us to handle. */ |
360 | blk_queue_max_phys_segments(vblk->disk->queue, vblk->sg_elems-2); | |
361 | blk_queue_max_hw_segments(vblk->disk->queue, vblk->sg_elems-2); | |
362 | ||
4b7f7e20 RR |
363 | /* No real sector limit. */ |
364 | blk_queue_max_sectors(vblk->disk->queue, -1U); | |
365 | ||
a586d4f6 RR |
366 | /* Host can optionally specify maximum segment size and number of |
367 | * segments. */ | |
368 | err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX, | |
369 | offsetof(struct virtio_blk_config, size_max), | |
370 | &v); | |
e467cde2 RR |
371 | if (!err) |
372 | blk_queue_max_segment_size(vblk->disk->queue, v); | |
4b7f7e20 | 373 | else |
b194aee9 | 374 | blk_queue_max_segment_size(vblk->disk->queue, -1U); |
e467cde2 | 375 | |
066f4d82 CB |
376 | /* Host can optionally specify the block size of the device */ |
377 | err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE, | |
378 | offsetof(struct virtio_blk_config, blk_size), | |
379 | &blk_size); | |
380 | if (!err) | |
e1defc4f | 381 | blk_queue_logical_block_size(vblk->disk->queue, blk_size); |
066f4d82 | 382 | |
e467cde2 RR |
383 | add_disk(vblk->disk); |
384 | return 0; | |
385 | ||
386 | out_put_disk: | |
387 | put_disk(vblk->disk); | |
e467cde2 RR |
388 | out_mempool: |
389 | mempool_destroy(vblk->pool); | |
390 | out_free_vq: | |
d2a7ddda | 391 | vdev->config->del_vqs(vdev); |
e467cde2 RR |
392 | out_free_vblk: |
393 | kfree(vblk); | |
394 | out: | |
395 | return err; | |
396 | } | |
397 | ||
98e94444 | 398 | static void __devexit virtblk_remove(struct virtio_device *vdev) |
e467cde2 RR |
399 | { |
400 | struct virtio_blk *vblk = vdev->priv; | |
e467cde2 | 401 | |
6e5aa7ef | 402 | /* Nothing should be pending. */ |
e467cde2 | 403 | BUG_ON(!list_empty(&vblk->reqs)); |
6e5aa7ef RR |
404 | |
405 | /* Stop all the virtqueues. */ | |
406 | vdev->config->reset(vdev); | |
407 | ||
ac9d463a | 408 | del_gendisk(vblk->disk); |
e467cde2 RR |
409 | blk_cleanup_queue(vblk->disk->queue); |
410 | put_disk(vblk->disk); | |
e467cde2 | 411 | mempool_destroy(vblk->pool); |
d2a7ddda | 412 | vdev->config->del_vqs(vdev); |
e467cde2 RR |
413 | kfree(vblk); |
414 | } | |
415 | ||
416 | static struct virtio_device_id id_table[] = { | |
417 | { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, | |
418 | { 0 }, | |
419 | }; | |
420 | ||
c45a6816 RR |
421 | static unsigned int features[] = { |
422 | VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, | |
066f4d82 | 423 | VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, |
1d589bb1 | 424 | VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_IDENTIFY |
c45a6816 RR |
425 | }; |
426 | ||
e467cde2 | 427 | static struct virtio_driver virtio_blk = { |
c45a6816 RR |
428 | .feature_table = features, |
429 | .feature_table_size = ARRAY_SIZE(features), | |
e467cde2 RR |
430 | .driver.name = KBUILD_MODNAME, |
431 | .driver.owner = THIS_MODULE, | |
432 | .id_table = id_table, | |
433 | .probe = virtblk_probe, | |
434 | .remove = __devexit_p(virtblk_remove), | |
435 | }; | |
436 | ||
437 | static int __init init(void) | |
438 | { | |
4f3bf19c CB |
439 | major = register_blkdev(0, "virtblk"); |
440 | if (major < 0) | |
441 | return major; | |
e467cde2 RR |
442 | return register_virtio_driver(&virtio_blk); |
443 | } | |
444 | ||
445 | static void __exit fini(void) | |
446 | { | |
4f3bf19c | 447 | unregister_blkdev(major, "virtblk"); |
e467cde2 RR |
448 | unregister_virtio_driver(&virtio_blk); |
449 | } | |
450 | module_init(init); | |
451 | module_exit(fini); | |
452 | ||
453 | MODULE_DEVICE_TABLE(virtio, id_table); | |
454 | MODULE_DESCRIPTION("Virtio block driver"); | |
455 | MODULE_LICENSE("GPL"); |