Commit | Line | Data |
---|---|---|
e467cde2 RR |
1 | //#define DEBUG |
2 | #include <linux/spinlock.h> | |
5a0e3ad6 | 3 | #include <linux/slab.h> |
e467cde2 RR |
4 | #include <linux/blkdev.h> |
5 | #include <linux/hdreg.h> | |
0c8d44f2 | 6 | #include <linux/module.h> |
4678d6f9 | 7 | #include <linux/mutex.h> |
e467cde2 RR |
8 | #include <linux/virtio.h> |
9 | #include <linux/virtio_blk.h> | |
3d1266c7 | 10 | #include <linux/scatterlist.h> |
7a7c924c | 11 | #include <linux/string_helpers.h> |
6917f83f | 12 | #include <scsi/scsi_cmnd.h> |
5087a50e | 13 | #include <linux/idr.h> |
3d1266c7 | 14 | |
4f3bf19c | 15 | #define PART_BITS 4 |
e467cde2 | 16 | |
a98755c5 AH |
17 | static bool use_bio; |
18 | module_param(use_bio, bool, S_IRUGO); | |
19 | ||
5087a50e MT |
20 | static int major; |
21 | static DEFINE_IDA(vd_index_ida); | |
22 | ||
7a7c924c | 23 | struct workqueue_struct *virtblk_wq; |
4f3bf19c | 24 | |
e467cde2 RR |
25 | struct virtio_blk |
26 | { | |
e467cde2 RR |
27 | struct virtio_device *vdev; |
28 | struct virtqueue *vq; | |
a98755c5 | 29 | wait_queue_head_t queue_wait; |
e467cde2 RR |
30 | |
31 | /* The disk structure for the kernel. */ | |
32 | struct gendisk *disk; | |
33 | ||
e467cde2 RR |
34 | mempool_t *pool; |
35 | ||
7a7c924c CH |
36 | /* Process context for config space updates */ |
37 | struct work_struct config_work; | |
38 | ||
4678d6f9 MT |
39 | /* Lock for config space updates */ |
40 | struct mutex config_lock; | |
41 | ||
42 | /* enable config space updates */ | |
43 | bool config_enable; | |
44 | ||
0864b79a RR |
45 | /* What host tells us, plus 2 for header & tailer. */ |
46 | unsigned int sg_elems; | |
47 | ||
5087a50e MT |
48 | /* Ida index - used to track minor number allocations. */ |
49 | int index; | |
50 | ||
e467cde2 | 51 | /* Scatterlist: can be too big for stack. */ |
0864b79a | 52 | struct scatterlist sg[/*sg_elems*/]; |
e467cde2 RR |
53 | }; |
54 | ||
55 | struct virtblk_req | |
56 | { | |
e467cde2 | 57 | struct request *req; |
a98755c5 | 58 | struct bio *bio; |
e467cde2 | 59 | struct virtio_blk_outhdr out_hdr; |
1cde26f9 | 60 | struct virtio_scsi_inhdr in_hdr; |
cb38fa23 | 61 | u8 status; |
a98755c5 | 62 | struct scatterlist sg[]; |
e467cde2 RR |
63 | }; |
64 | ||
a98755c5 AH |
65 | static inline int virtblk_result(struct virtblk_req *vbr) |
66 | { | |
67 | switch (vbr->status) { | |
68 | case VIRTIO_BLK_S_OK: | |
69 | return 0; | |
70 | case VIRTIO_BLK_S_UNSUPP: | |
71 | return -ENOTTY; | |
72 | default: | |
73 | return -EIO; | |
74 | } | |
75 | } | |
76 | ||
77 | static inline void virtblk_request_done(struct virtio_blk *vblk, | |
78 | struct virtblk_req *vbr) | |
79 | { | |
80 | struct request *req = vbr->req; | |
81 | int error = virtblk_result(vbr); | |
82 | ||
83 | if (req->cmd_type == REQ_TYPE_BLOCK_PC) { | |
84 | req->resid_len = vbr->in_hdr.residual; | |
85 | req->sense_len = vbr->in_hdr.sense_len; | |
86 | req->errors = vbr->in_hdr.errors; | |
87 | } else if (req->cmd_type == REQ_TYPE_SPECIAL) { | |
88 | req->errors = (error != 0); | |
89 | } | |
90 | ||
91 | __blk_end_request_all(req, error); | |
92 | mempool_free(vbr, vblk->pool); | |
93 | } | |
94 | ||
95 | static inline void virtblk_bio_done(struct virtio_blk *vblk, | |
96 | struct virtblk_req *vbr) | |
97 | { | |
98 | bio_endio(vbr->bio, virtblk_result(vbr)); | |
99 | mempool_free(vbr, vblk->pool); | |
100 | } | |
101 | ||
102 | static void virtblk_done(struct virtqueue *vq) | |
e467cde2 RR |
103 | { |
104 | struct virtio_blk *vblk = vq->vdev->priv; | |
a98755c5 | 105 | unsigned long bio_done = 0, req_done = 0; |
e467cde2 | 106 | struct virtblk_req *vbr; |
e467cde2 | 107 | unsigned long flags; |
a98755c5 | 108 | unsigned int len; |
e467cde2 | 109 | |
2c95a329 | 110 | spin_lock_irqsave(vblk->disk->queue->queue_lock, flags); |
09ec6b69 | 111 | while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) { |
a98755c5 AH |
112 | if (vbr->bio) { |
113 | virtblk_bio_done(vblk, vbr); | |
114 | bio_done++; | |
115 | } else { | |
116 | virtblk_request_done(vblk, vbr); | |
117 | req_done++; | |
33659ebb | 118 | } |
e467cde2 RR |
119 | } |
120 | /* In case queue is stopped waiting for more buffers. */ | |
a98755c5 AH |
121 | if (req_done) |
122 | blk_start_queue(vblk->disk->queue); | |
2c95a329 | 123 | spin_unlock_irqrestore(vblk->disk->queue->queue_lock, flags); |
a98755c5 AH |
124 | |
125 | if (bio_done) | |
126 | wake_up(&vblk->queue_wait); | |
127 | } | |
128 | ||
129 | static inline struct virtblk_req *virtblk_alloc_req(struct virtio_blk *vblk, | |
130 | gfp_t gfp_mask) | |
131 | { | |
132 | struct virtblk_req *vbr; | |
133 | ||
134 | vbr = mempool_alloc(vblk->pool, gfp_mask); | |
135 | if (vbr && use_bio) | |
136 | sg_init_table(vbr->sg, vblk->sg_elems); | |
137 | ||
138 | return vbr; | |
e467cde2 RR |
139 | } |
140 | ||
141 | static bool do_req(struct request_queue *q, struct virtio_blk *vblk, | |
142 | struct request *req) | |
143 | { | |
1cde26f9 | 144 | unsigned long num, out = 0, in = 0; |
e467cde2 RR |
145 | struct virtblk_req *vbr; |
146 | ||
a98755c5 | 147 | vbr = virtblk_alloc_req(vblk, GFP_ATOMIC); |
e467cde2 RR |
148 | if (!vbr) |
149 | /* When another request finishes we'll try again. */ | |
150 | return false; | |
151 | ||
152 | vbr->req = req; | |
a98755c5 | 153 | vbr->bio = NULL; |
dd40e456 FT |
154 | if (req->cmd_flags & REQ_FLUSH) { |
155 | vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH; | |
4cb2ea28 | 156 | vbr->out_hdr.sector = 0; |
157 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
dd40e456 FT |
158 | } else { |
159 | switch (req->cmd_type) { | |
160 | case REQ_TYPE_FS: | |
161 | vbr->out_hdr.type = 0; | |
162 | vbr->out_hdr.sector = blk_rq_pos(vbr->req); | |
163 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
164 | break; | |
165 | case REQ_TYPE_BLOCK_PC: | |
166 | vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; | |
f1b0ef06 CH |
167 | vbr->out_hdr.sector = 0; |
168 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
169 | break; | |
dd40e456 FT |
170 | case REQ_TYPE_SPECIAL: |
171 | vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID; | |
172 | vbr->out_hdr.sector = 0; | |
173 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
174 | break; | |
175 | default: | |
176 | /* We don't put anything else in the queue. */ | |
177 | BUG(); | |
f1b0ef06 | 178 | } |
e467cde2 RR |
179 | } |
180 | ||
1cde26f9 | 181 | sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr)); |
e467cde2 | 182 | |
1cde26f9 HR |
183 | /* |
184 | * If this is a packet command we need a couple of additional headers. | |
185 | * Behind the normal outhdr we put a segment with the scsi command | |
186 | * block, and before the normal inhdr we put the sense data and the | |
187 | * inhdr with additional status information before the normal inhdr. | |
188 | */ | |
33659ebb | 189 | if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) |
1cde26f9 HR |
190 | sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len); |
191 | ||
192 | num = blk_rq_map_sg(q, vbr->req, vblk->sg + out); | |
193 | ||
33659ebb | 194 | if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) { |
6917f83f | 195 | sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, SCSI_SENSE_BUFFERSIZE); |
1cde26f9 HR |
196 | sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr, |
197 | sizeof(vbr->in_hdr)); | |
198 | } | |
199 | ||
200 | sg_set_buf(&vblk->sg[num + out + in++], &vbr->status, | |
201 | sizeof(vbr->status)); | |
202 | ||
203 | if (num) { | |
204 | if (rq_data_dir(vbr->req) == WRITE) { | |
205 | vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; | |
206 | out += num; | |
207 | } else { | |
208 | vbr->out_hdr.type |= VIRTIO_BLK_T_IN; | |
209 | in += num; | |
210 | } | |
e467cde2 RR |
211 | } |
212 | ||
a98755c5 AH |
213 | if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr, |
214 | GFP_ATOMIC) < 0) { | |
e467cde2 RR |
215 | mempool_free(vbr, vblk->pool); |
216 | return false; | |
217 | } | |
218 | ||
e467cde2 RR |
219 | return true; |
220 | } | |
221 | ||
a98755c5 | 222 | static void virtblk_request(struct request_queue *q) |
e467cde2 | 223 | { |
6c3b46f7 | 224 | struct virtio_blk *vblk = q->queuedata; |
e467cde2 RR |
225 | struct request *req; |
226 | unsigned int issued = 0; | |
227 | ||
9934c8c0 | 228 | while ((req = blk_peek_request(q)) != NULL) { |
0864b79a | 229 | BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); |
e467cde2 RR |
230 | |
231 | /* If this request fails, stop queue and wait for something to | |
232 | finish to restart it. */ | |
233 | if (!do_req(q, vblk, req)) { | |
234 | blk_stop_queue(q); | |
235 | break; | |
236 | } | |
9934c8c0 | 237 | blk_start_request(req); |
e467cde2 RR |
238 | issued++; |
239 | } | |
240 | ||
241 | if (issued) | |
09ec6b69 | 242 | virtqueue_kick(vblk->vq); |
e467cde2 RR |
243 | } |
244 | ||
a98755c5 AH |
245 | static void virtblk_add_buf_wait(struct virtio_blk *vblk, |
246 | struct virtblk_req *vbr, | |
247 | unsigned long out, | |
248 | unsigned long in) | |
249 | { | |
250 | DEFINE_WAIT(wait); | |
251 | ||
252 | for (;;) { | |
253 | prepare_to_wait_exclusive(&vblk->queue_wait, &wait, | |
254 | TASK_UNINTERRUPTIBLE); | |
255 | ||
256 | spin_lock_irq(vblk->disk->queue->queue_lock); | |
257 | if (virtqueue_add_buf(vblk->vq, vbr->sg, out, in, vbr, | |
258 | GFP_ATOMIC) < 0) { | |
259 | spin_unlock_irq(vblk->disk->queue->queue_lock); | |
260 | io_schedule(); | |
261 | } else { | |
262 | virtqueue_kick(vblk->vq); | |
263 | spin_unlock_irq(vblk->disk->queue->queue_lock); | |
264 | break; | |
265 | } | |
266 | ||
267 | } | |
268 | ||
269 | finish_wait(&vblk->queue_wait, &wait); | |
270 | } | |
271 | ||
272 | static void virtblk_make_request(struct request_queue *q, struct bio *bio) | |
273 | { | |
274 | struct virtio_blk *vblk = q->queuedata; | |
275 | unsigned int num, out = 0, in = 0; | |
276 | struct virtblk_req *vbr; | |
277 | ||
278 | BUG_ON(bio->bi_phys_segments + 2 > vblk->sg_elems); | |
279 | BUG_ON(bio->bi_rw & (REQ_FLUSH | REQ_FUA)); | |
280 | ||
281 | vbr = virtblk_alloc_req(vblk, GFP_NOIO); | |
282 | if (!vbr) { | |
283 | bio_endio(bio, -ENOMEM); | |
284 | return; | |
285 | } | |
286 | ||
287 | vbr->bio = bio; | |
288 | vbr->req = NULL; | |
289 | vbr->out_hdr.type = 0; | |
290 | vbr->out_hdr.sector = bio->bi_sector; | |
291 | vbr->out_hdr.ioprio = bio_prio(bio); | |
292 | ||
293 | sg_set_buf(&vbr->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr)); | |
294 | ||
295 | num = blk_bio_map_sg(q, bio, vbr->sg + out); | |
296 | ||
297 | sg_set_buf(&vbr->sg[num + out + in++], &vbr->status, | |
298 | sizeof(vbr->status)); | |
299 | ||
300 | if (num) { | |
301 | if (bio->bi_rw & REQ_WRITE) { | |
302 | vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; | |
303 | out += num; | |
304 | } else { | |
305 | vbr->out_hdr.type |= VIRTIO_BLK_T_IN; | |
306 | in += num; | |
307 | } | |
308 | } | |
309 | ||
310 | spin_lock_irq(vblk->disk->queue->queue_lock); | |
311 | if (unlikely(virtqueue_add_buf(vblk->vq, vbr->sg, out, in, vbr, | |
312 | GFP_ATOMIC) < 0)) { | |
313 | spin_unlock_irq(vblk->disk->queue->queue_lock); | |
314 | virtblk_add_buf_wait(vblk, vbr, out, in); | |
315 | return; | |
316 | } | |
317 | virtqueue_kick(vblk->vq); | |
318 | spin_unlock_irq(vblk->disk->queue->queue_lock); | |
319 | } | |
320 | ||
4cb2ea28 | 321 | /* return id (s/n) string for *disk to *id_str |
322 | */ | |
323 | static int virtblk_get_id(struct gendisk *disk, char *id_str) | |
324 | { | |
325 | struct virtio_blk *vblk = disk->private_data; | |
326 | struct request *req; | |
327 | struct bio *bio; | |
e4c4776d | 328 | int err; |
4cb2ea28 | 329 | |
330 | bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES, | |
331 | GFP_KERNEL); | |
332 | if (IS_ERR(bio)) | |
333 | return PTR_ERR(bio); | |
334 | ||
335 | req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL); | |
336 | if (IS_ERR(req)) { | |
337 | bio_put(bio); | |
338 | return PTR_ERR(req); | |
339 | } | |
340 | ||
341 | req->cmd_type = REQ_TYPE_SPECIAL; | |
e4c4776d MS |
342 | err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); |
343 | blk_put_request(req); | |
344 | ||
345 | return err; | |
4cb2ea28 | 346 | } |
347 | ||
fe5a50a1 CH |
348 | static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, |
349 | unsigned int cmd, unsigned long data) | |
e467cde2 | 350 | { |
1cde26f9 HR |
351 | struct gendisk *disk = bdev->bd_disk; |
352 | struct virtio_blk *vblk = disk->private_data; | |
353 | ||
354 | /* | |
355 | * Only allow the generic SCSI ioctls if the host can support it. | |
356 | */ | |
357 | if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) | |
d9ecdea7 | 358 | return -ENOTTY; |
1cde26f9 | 359 | |
577ebb37 PB |
360 | return scsi_cmd_blk_ioctl(bdev, mode, cmd, |
361 | (void __user *)data); | |
e467cde2 RR |
362 | } |
363 | ||
135da0b0 CB |
364 | /* We provide getgeo only to please some old bootloader/partitioning tools */ |
365 | static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
366 | { | |
48e4043d RH |
367 | struct virtio_blk *vblk = bd->bd_disk->private_data; |
368 | struct virtio_blk_geometry vgeo; | |
369 | int err; | |
370 | ||
371 | /* see if the host passed in geometry config */ | |
372 | err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY, | |
373 | offsetof(struct virtio_blk_config, geometry), | |
374 | &vgeo); | |
375 | ||
376 | if (!err) { | |
377 | geo->heads = vgeo.heads; | |
378 | geo->sectors = vgeo.sectors; | |
379 | geo->cylinders = vgeo.cylinders; | |
380 | } else { | |
381 | /* some standard values, similar to sd */ | |
382 | geo->heads = 1 << 6; | |
383 | geo->sectors = 1 << 5; | |
384 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
385 | } | |
135da0b0 CB |
386 | return 0; |
387 | } | |
388 | ||
83d5cde4 | 389 | static const struct block_device_operations virtblk_fops = { |
8a6cfeb6 | 390 | .ioctl = virtblk_ioctl, |
135da0b0 CB |
391 | .owner = THIS_MODULE, |
392 | .getgeo = virtblk_getgeo, | |
e467cde2 RR |
393 | }; |
394 | ||
d50ed907 CB |
395 | static int index_to_minor(int index) |
396 | { | |
397 | return index << PART_BITS; | |
398 | } | |
399 | ||
5087a50e MT |
400 | static int minor_to_index(int minor) |
401 | { | |
402 | return minor >> PART_BITS; | |
403 | } | |
404 | ||
a5eb9e4f RH |
405 | static ssize_t virtblk_serial_show(struct device *dev, |
406 | struct device_attribute *attr, char *buf) | |
407 | { | |
408 | struct gendisk *disk = dev_to_disk(dev); | |
409 | int err; | |
410 | ||
411 | /* sysfs gives us a PAGE_SIZE buffer */ | |
412 | BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); | |
413 | ||
414 | buf[VIRTIO_BLK_ID_BYTES] = '\0'; | |
415 | err = virtblk_get_id(disk, buf); | |
416 | if (!err) | |
417 | return strlen(buf); | |
418 | ||
419 | if (err == -EIO) /* Unsupported? Make it empty. */ | |
420 | return 0; | |
421 | ||
422 | return err; | |
423 | } | |
424 | DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL); | |
425 | ||
7a7c924c CH |
426 | static void virtblk_config_changed_work(struct work_struct *work) |
427 | { | |
428 | struct virtio_blk *vblk = | |
429 | container_of(work, struct virtio_blk, config_work); | |
430 | struct virtio_device *vdev = vblk->vdev; | |
431 | struct request_queue *q = vblk->disk->queue; | |
432 | char cap_str_2[10], cap_str_10[10]; | |
433 | u64 capacity, size; | |
434 | ||
4678d6f9 MT |
435 | mutex_lock(&vblk->config_lock); |
436 | if (!vblk->config_enable) | |
437 | goto done; | |
438 | ||
7a7c924c CH |
439 | /* Host must always specify the capacity. */ |
440 | vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), | |
441 | &capacity, sizeof(capacity)); | |
442 | ||
443 | /* If capacity is too big, truncate with warning. */ | |
444 | if ((sector_t)capacity != capacity) { | |
445 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
446 | (unsigned long long)capacity); | |
447 | capacity = (sector_t)-1; | |
448 | } | |
449 | ||
450 | size = capacity * queue_logical_block_size(q); | |
451 | string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); | |
452 | string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); | |
453 | ||
454 | dev_notice(&vdev->dev, | |
455 | "new size: %llu %d-byte logical blocks (%s/%s)\n", | |
456 | (unsigned long long)capacity, | |
457 | queue_logical_block_size(q), | |
458 | cap_str_10, cap_str_2); | |
459 | ||
460 | set_capacity(vblk->disk, capacity); | |
e9986f30 | 461 | revalidate_disk(vblk->disk); |
4678d6f9 MT |
462 | done: |
463 | mutex_unlock(&vblk->config_lock); | |
7a7c924c CH |
464 | } |
465 | ||
466 | static void virtblk_config_changed(struct virtio_device *vdev) | |
467 | { | |
468 | struct virtio_blk *vblk = vdev->priv; | |
469 | ||
470 | queue_work(virtblk_wq, &vblk->config_work); | |
471 | } | |
472 | ||
6abd6e5a AS |
473 | static int init_vq(struct virtio_blk *vblk) |
474 | { | |
475 | int err = 0; | |
476 | ||
477 | /* We expect one virtqueue, for output. */ | |
a98755c5 | 478 | vblk->vq = virtio_find_single_vq(vblk->vdev, virtblk_done, "requests"); |
6abd6e5a AS |
479 | if (IS_ERR(vblk->vq)) |
480 | err = PTR_ERR(vblk->vq); | |
481 | ||
482 | return err; | |
483 | } | |
484 | ||
c0aa3e09 RM |
485 | /* |
486 | * Legacy naming scheme used for virtio devices. We are stuck with it for | |
487 | * virtio blk but don't ever use it for any new driver. | |
488 | */ | |
489 | static int virtblk_name_format(char *prefix, int index, char *buf, int buflen) | |
490 | { | |
491 | const int base = 'z' - 'a' + 1; | |
492 | char *begin = buf + strlen(prefix); | |
493 | char *end = buf + buflen; | |
494 | char *p; | |
495 | int unit; | |
496 | ||
497 | p = end - 1; | |
498 | *p = '\0'; | |
499 | unit = base; | |
500 | do { | |
501 | if (p == begin) | |
502 | return -EINVAL; | |
503 | *--p = 'a' + (index % unit); | |
504 | index = (index / unit) - 1; | |
505 | } while (index >= 0); | |
506 | ||
507 | memmove(begin, p, end - p); | |
508 | memcpy(buf, prefix, strlen(prefix)); | |
509 | ||
510 | return 0; | |
511 | } | |
512 | ||
cd5d5038 PB |
513 | static int virtblk_get_cache_mode(struct virtio_device *vdev) |
514 | { | |
515 | u8 writeback; | |
516 | int err; | |
517 | ||
518 | err = virtio_config_val(vdev, VIRTIO_BLK_F_CONFIG_WCE, | |
519 | offsetof(struct virtio_blk_config, wce), | |
520 | &writeback); | |
521 | if (err) | |
522 | writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE); | |
523 | ||
524 | return writeback; | |
525 | } | |
526 | ||
527 | static void virtblk_update_cache_mode(struct virtio_device *vdev) | |
528 | { | |
529 | u8 writeback = virtblk_get_cache_mode(vdev); | |
530 | struct virtio_blk *vblk = vdev->priv; | |
531 | ||
a98755c5 | 532 | if (writeback && !use_bio) |
cd5d5038 PB |
533 | blk_queue_flush(vblk->disk->queue, REQ_FLUSH); |
534 | else | |
535 | blk_queue_flush(vblk->disk->queue, 0); | |
536 | ||
537 | revalidate_disk(vblk->disk); | |
538 | } | |
539 | ||
540 | static const char *const virtblk_cache_types[] = { | |
541 | "write through", "write back" | |
542 | }; | |
543 | ||
544 | static ssize_t | |
545 | virtblk_cache_type_store(struct device *dev, struct device_attribute *attr, | |
546 | const char *buf, size_t count) | |
547 | { | |
548 | struct gendisk *disk = dev_to_disk(dev); | |
549 | struct virtio_blk *vblk = disk->private_data; | |
550 | struct virtio_device *vdev = vblk->vdev; | |
551 | int i; | |
552 | u8 writeback; | |
553 | ||
554 | BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE)); | |
555 | for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; ) | |
556 | if (sysfs_streq(buf, virtblk_cache_types[i])) | |
557 | break; | |
558 | ||
559 | if (i < 0) | |
560 | return -EINVAL; | |
561 | ||
562 | writeback = i; | |
563 | vdev->config->set(vdev, | |
564 | offsetof(struct virtio_blk_config, wce), | |
565 | &writeback, sizeof(writeback)); | |
566 | ||
567 | virtblk_update_cache_mode(vdev); | |
568 | return count; | |
569 | } | |
570 | ||
571 | static ssize_t | |
572 | virtblk_cache_type_show(struct device *dev, struct device_attribute *attr, | |
573 | char *buf) | |
574 | { | |
575 | struct gendisk *disk = dev_to_disk(dev); | |
576 | struct virtio_blk *vblk = disk->private_data; | |
577 | u8 writeback = virtblk_get_cache_mode(vblk->vdev); | |
578 | ||
579 | BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types)); | |
580 | return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]); | |
581 | } | |
582 | ||
583 | static const struct device_attribute dev_attr_cache_type_ro = | |
584 | __ATTR(cache_type, S_IRUGO, | |
585 | virtblk_cache_type_show, NULL); | |
586 | static const struct device_attribute dev_attr_cache_type_rw = | |
587 | __ATTR(cache_type, S_IRUGO|S_IWUSR, | |
588 | virtblk_cache_type_show, virtblk_cache_type_store); | |
589 | ||
98e94444 | 590 | static int __devinit virtblk_probe(struct virtio_device *vdev) |
e467cde2 RR |
591 | { |
592 | struct virtio_blk *vblk; | |
69740c8b | 593 | struct request_queue *q; |
5087a50e | 594 | int err, index; |
a98755c5 AH |
595 | int pool_size; |
596 | ||
e467cde2 | 597 | u64 cap; |
69740c8b CH |
598 | u32 v, blk_size, sg_elems, opt_io_size; |
599 | u16 min_io_size; | |
600 | u8 physical_block_exp, alignment_offset; | |
e467cde2 | 601 | |
5087a50e MT |
602 | err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS), |
603 | GFP_KERNEL); | |
604 | if (err < 0) | |
605 | goto out; | |
606 | index = err; | |
4f3bf19c | 607 | |
0864b79a RR |
608 | /* We need to know how many segments before we allocate. */ |
609 | err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX, | |
610 | offsetof(struct virtio_blk_config, seg_max), | |
611 | &sg_elems); | |
a5b365a6 CH |
612 | |
613 | /* We need at least one SG element, whatever they say. */ | |
614 | if (err || !sg_elems) | |
0864b79a RR |
615 | sg_elems = 1; |
616 | ||
617 | /* We need an extra sg elements at head and tail. */ | |
618 | sg_elems += 2; | |
619 | vdev->priv = vblk = kmalloc(sizeof(*vblk) + | |
620 | sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL); | |
e467cde2 RR |
621 | if (!vblk) { |
622 | err = -ENOMEM; | |
5087a50e | 623 | goto out_free_index; |
e467cde2 RR |
624 | } |
625 | ||
a98755c5 | 626 | init_waitqueue_head(&vblk->queue_wait); |
e467cde2 | 627 | vblk->vdev = vdev; |
0864b79a RR |
628 | vblk->sg_elems = sg_elems; |
629 | sg_init_table(vblk->sg, vblk->sg_elems); | |
4678d6f9 | 630 | mutex_init(&vblk->config_lock); |
a98755c5 | 631 | |
7a7c924c | 632 | INIT_WORK(&vblk->config_work, virtblk_config_changed_work); |
4678d6f9 | 633 | vblk->config_enable = true; |
e467cde2 | 634 | |
6abd6e5a AS |
635 | err = init_vq(vblk); |
636 | if (err) | |
e467cde2 | 637 | goto out_free_vblk; |
e467cde2 | 638 | |
a98755c5 AH |
639 | pool_size = sizeof(struct virtblk_req); |
640 | if (use_bio) | |
641 | pool_size += sizeof(struct scatterlist) * sg_elems; | |
642 | vblk->pool = mempool_create_kmalloc_pool(1, pool_size); | |
e467cde2 RR |
643 | if (!vblk->pool) { |
644 | err = -ENOMEM; | |
645 | goto out_free_vq; | |
646 | } | |
647 | ||
e467cde2 | 648 | /* FIXME: How many partitions? How long is a piece of string? */ |
4f3bf19c | 649 | vblk->disk = alloc_disk(1 << PART_BITS); |
e467cde2 RR |
650 | if (!vblk->disk) { |
651 | err = -ENOMEM; | |
4f3bf19c | 652 | goto out_mempool; |
e467cde2 RR |
653 | } |
654 | ||
a98755c5 | 655 | q = vblk->disk->queue = blk_init_queue(virtblk_request, NULL); |
69740c8b | 656 | if (!q) { |
e467cde2 RR |
657 | err = -ENOMEM; |
658 | goto out_put_disk; | |
659 | } | |
660 | ||
a98755c5 AH |
661 | if (use_bio) |
662 | blk_queue_make_request(q, virtblk_make_request); | |
69740c8b | 663 | q->queuedata = vblk; |
7d116b62 | 664 | |
c0aa3e09 | 665 | virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN); |
d50ed907 | 666 | |
e467cde2 | 667 | vblk->disk->major = major; |
d50ed907 | 668 | vblk->disk->first_minor = index_to_minor(index); |
e467cde2 RR |
669 | vblk->disk->private_data = vblk; |
670 | vblk->disk->fops = &virtblk_fops; | |
c4839346 | 671 | vblk->disk->driverfs_dev = &vdev->dev; |
5087a50e | 672 | vblk->index = index; |
4f3bf19c | 673 | |
02c42b7a | 674 | /* configure queue flush support */ |
cd5d5038 | 675 | virtblk_update_cache_mode(vdev); |
e467cde2 | 676 | |
3ef53609 CB |
677 | /* If disk is read-only in the host, the guest should obey */ |
678 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) | |
679 | set_disk_ro(vblk->disk, 1); | |
680 | ||
a586d4f6 | 681 | /* Host must always specify the capacity. */ |
72e61eb4 RR |
682 | vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity), |
683 | &cap, sizeof(cap)); | |
e467cde2 RR |
684 | |
685 | /* If capacity is too big, truncate with warning. */ | |
686 | if ((sector_t)cap != cap) { | |
687 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
688 | (unsigned long long)cap); | |
689 | cap = (sector_t)-1; | |
690 | } | |
691 | set_capacity(vblk->disk, cap); | |
692 | ||
0864b79a | 693 | /* We can handle whatever the host told us to handle. */ |
ee714f2d | 694 | blk_queue_max_segments(q, vblk->sg_elems-2); |
0864b79a | 695 | |
4eff3cae | 696 | /* No need to bounce any requests */ |
69740c8b | 697 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
4eff3cae | 698 | |
4b7f7e20 | 699 | /* No real sector limit. */ |
ee714f2d | 700 | blk_queue_max_hw_sectors(q, -1U); |
4b7f7e20 | 701 | |
a586d4f6 RR |
702 | /* Host can optionally specify maximum segment size and number of |
703 | * segments. */ | |
704 | err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX, | |
705 | offsetof(struct virtio_blk_config, size_max), | |
706 | &v); | |
e467cde2 | 707 | if (!err) |
69740c8b | 708 | blk_queue_max_segment_size(q, v); |
4b7f7e20 | 709 | else |
69740c8b | 710 | blk_queue_max_segment_size(q, -1U); |
e467cde2 | 711 | |
066f4d82 CB |
712 | /* Host can optionally specify the block size of the device */ |
713 | err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE, | |
714 | offsetof(struct virtio_blk_config, blk_size), | |
715 | &blk_size); | |
716 | if (!err) | |
69740c8b CH |
717 | blk_queue_logical_block_size(q, blk_size); |
718 | else | |
719 | blk_size = queue_logical_block_size(q); | |
720 | ||
721 | /* Use topology information if available */ | |
722 | err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, | |
723 | offsetof(struct virtio_blk_config, physical_block_exp), | |
724 | &physical_block_exp); | |
725 | if (!err && physical_block_exp) | |
726 | blk_queue_physical_block_size(q, | |
727 | blk_size * (1 << physical_block_exp)); | |
728 | ||
729 | err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, | |
730 | offsetof(struct virtio_blk_config, alignment_offset), | |
731 | &alignment_offset); | |
732 | if (!err && alignment_offset) | |
733 | blk_queue_alignment_offset(q, blk_size * alignment_offset); | |
734 | ||
735 | err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, | |
736 | offsetof(struct virtio_blk_config, min_io_size), | |
737 | &min_io_size); | |
738 | if (!err && min_io_size) | |
739 | blk_queue_io_min(q, blk_size * min_io_size); | |
740 | ||
741 | err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY, | |
742 | offsetof(struct virtio_blk_config, opt_io_size), | |
743 | &opt_io_size); | |
744 | if (!err && opt_io_size) | |
745 | blk_queue_io_opt(q, blk_size * opt_io_size); | |
746 | ||
e467cde2 | 747 | add_disk(vblk->disk); |
a5eb9e4f RH |
748 | err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial); |
749 | if (err) | |
750 | goto out_del_disk; | |
751 | ||
cd5d5038 PB |
752 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) |
753 | err = device_create_file(disk_to_dev(vblk->disk), | |
754 | &dev_attr_cache_type_rw); | |
755 | else | |
756 | err = device_create_file(disk_to_dev(vblk->disk), | |
757 | &dev_attr_cache_type_ro); | |
758 | if (err) | |
759 | goto out_del_disk; | |
e467cde2 RR |
760 | return 0; |
761 | ||
a5eb9e4f RH |
762 | out_del_disk: |
763 | del_gendisk(vblk->disk); | |
764 | blk_cleanup_queue(vblk->disk->queue); | |
e467cde2 RR |
765 | out_put_disk: |
766 | put_disk(vblk->disk); | |
e467cde2 RR |
767 | out_mempool: |
768 | mempool_destroy(vblk->pool); | |
769 | out_free_vq: | |
d2a7ddda | 770 | vdev->config->del_vqs(vdev); |
e467cde2 RR |
771 | out_free_vblk: |
772 | kfree(vblk); | |
5087a50e MT |
773 | out_free_index: |
774 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
775 | out: |
776 | return err; | |
777 | } | |
778 | ||
98e94444 | 779 | static void __devexit virtblk_remove(struct virtio_device *vdev) |
e467cde2 RR |
780 | { |
781 | struct virtio_blk *vblk = vdev->priv; | |
5087a50e | 782 | int index = vblk->index; |
e467cde2 | 783 | |
4678d6f9 MT |
784 | /* Prevent config work handler from accessing the device. */ |
785 | mutex_lock(&vblk->config_lock); | |
786 | vblk->config_enable = false; | |
787 | mutex_unlock(&vblk->config_lock); | |
7a7c924c | 788 | |
02e2b124 | 789 | del_gendisk(vblk->disk); |
483001c7 | 790 | blk_cleanup_queue(vblk->disk->queue); |
02e2b124 | 791 | |
6e5aa7ef RR |
792 | /* Stop all the virtqueues. */ |
793 | vdev->config->reset(vdev); | |
794 | ||
4678d6f9 MT |
795 | flush_work(&vblk->config_work); |
796 | ||
e467cde2 | 797 | put_disk(vblk->disk); |
e467cde2 | 798 | mempool_destroy(vblk->pool); |
d2a7ddda | 799 | vdev->config->del_vqs(vdev); |
e467cde2 | 800 | kfree(vblk); |
5087a50e | 801 | ida_simple_remove(&vd_index_ida, index); |
e467cde2 RR |
802 | } |
803 | ||
f8fb5bc2 AS |
804 | #ifdef CONFIG_PM |
805 | static int virtblk_freeze(struct virtio_device *vdev) | |
806 | { | |
807 | struct virtio_blk *vblk = vdev->priv; | |
808 | ||
809 | /* Ensure we don't receive any more interrupts */ | |
810 | vdev->config->reset(vdev); | |
811 | ||
812 | /* Prevent config work handler from accessing the device. */ | |
813 | mutex_lock(&vblk->config_lock); | |
814 | vblk->config_enable = false; | |
815 | mutex_unlock(&vblk->config_lock); | |
816 | ||
817 | flush_work(&vblk->config_work); | |
818 | ||
819 | spin_lock_irq(vblk->disk->queue->queue_lock); | |
820 | blk_stop_queue(vblk->disk->queue); | |
821 | spin_unlock_irq(vblk->disk->queue->queue_lock); | |
822 | blk_sync_queue(vblk->disk->queue); | |
823 | ||
824 | vdev->config->del_vqs(vdev); | |
825 | return 0; | |
826 | } | |
827 | ||
828 | static int virtblk_restore(struct virtio_device *vdev) | |
829 | { | |
830 | struct virtio_blk *vblk = vdev->priv; | |
831 | int ret; | |
832 | ||
833 | vblk->config_enable = true; | |
834 | ret = init_vq(vdev->priv); | |
835 | if (!ret) { | |
836 | spin_lock_irq(vblk->disk->queue->queue_lock); | |
837 | blk_start_queue(vblk->disk->queue); | |
838 | spin_unlock_irq(vblk->disk->queue->queue_lock); | |
839 | } | |
840 | return ret; | |
841 | } | |
842 | #endif | |
843 | ||
47483e25 | 844 | static const struct virtio_device_id id_table[] = { |
e467cde2 RR |
845 | { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, |
846 | { 0 }, | |
847 | }; | |
848 | ||
c45a6816 | 849 | static unsigned int features[] = { |
02c42b7a TH |
850 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, |
851 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI, | |
cd5d5038 | 852 | VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE |
c45a6816 RR |
853 | }; |
854 | ||
4fbfff76 RM |
855 | /* |
856 | * virtio_blk causes spurious section mismatch warning by | |
857 | * simultaneously referring to a __devinit and a __devexit function. | |
858 | * Use __refdata to avoid this warning. | |
859 | */ | |
860 | static struct virtio_driver __refdata virtio_blk = { | |
7a7c924c CH |
861 | .feature_table = features, |
862 | .feature_table_size = ARRAY_SIZE(features), | |
863 | .driver.name = KBUILD_MODNAME, | |
864 | .driver.owner = THIS_MODULE, | |
865 | .id_table = id_table, | |
866 | .probe = virtblk_probe, | |
867 | .remove = __devexit_p(virtblk_remove), | |
868 | .config_changed = virtblk_config_changed, | |
f8fb5bc2 AS |
869 | #ifdef CONFIG_PM |
870 | .freeze = virtblk_freeze, | |
871 | .restore = virtblk_restore, | |
872 | #endif | |
e467cde2 RR |
873 | }; |
874 | ||
875 | static int __init init(void) | |
876 | { | |
7a7c924c CH |
877 | int error; |
878 | ||
879 | virtblk_wq = alloc_workqueue("virtio-blk", 0, 0); | |
880 | if (!virtblk_wq) | |
881 | return -ENOMEM; | |
882 | ||
4f3bf19c | 883 | major = register_blkdev(0, "virtblk"); |
7a7c924c CH |
884 | if (major < 0) { |
885 | error = major; | |
886 | goto out_destroy_workqueue; | |
887 | } | |
888 | ||
889 | error = register_virtio_driver(&virtio_blk); | |
890 | if (error) | |
891 | goto out_unregister_blkdev; | |
892 | return 0; | |
893 | ||
894 | out_unregister_blkdev: | |
895 | unregister_blkdev(major, "virtblk"); | |
896 | out_destroy_workqueue: | |
897 | destroy_workqueue(virtblk_wq); | |
898 | return error; | |
e467cde2 RR |
899 | } |
900 | ||
901 | static void __exit fini(void) | |
902 | { | |
4f3bf19c | 903 | unregister_blkdev(major, "virtblk"); |
e467cde2 | 904 | unregister_virtio_driver(&virtio_blk); |
7a7c924c | 905 | destroy_workqueue(virtblk_wq); |
e467cde2 RR |
906 | } |
907 | module_init(init); | |
908 | module_exit(fini); | |
909 | ||
910 | MODULE_DEVICE_TABLE(virtio, id_table); | |
911 | MODULE_DESCRIPTION("Virtio block driver"); | |
912 | MODULE_LICENSE("GPL"); |