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> |
1cf7e9c6 JA |
14 | #include <linux/blk-mq.h> |
15 | #include <linux/numa.h> | |
3d1266c7 | 16 | |
4f3bf19c | 17 | #define PART_BITS 4 |
e467cde2 | 18 | |
5087a50e MT |
19 | static int major; |
20 | static DEFINE_IDA(vd_index_ida); | |
21 | ||
2a647bfe | 22 | static struct workqueue_struct *virtblk_wq; |
4f3bf19c | 23 | |
e467cde2 RR |
24 | struct virtio_blk |
25 | { | |
e467cde2 RR |
26 | struct virtio_device *vdev; |
27 | struct virtqueue *vq; | |
1cf7e9c6 | 28 | spinlock_t vq_lock; |
e467cde2 RR |
29 | |
30 | /* The disk structure for the kernel. */ | |
31 | struct gendisk *disk; | |
32 | ||
7a7c924c CH |
33 | /* Process context for config space updates */ |
34 | struct work_struct config_work; | |
35 | ||
4678d6f9 MT |
36 | /* Lock for config space updates */ |
37 | struct mutex config_lock; | |
38 | ||
39 | /* enable config space updates */ | |
40 | bool config_enable; | |
41 | ||
0864b79a RR |
42 | /* What host tells us, plus 2 for header & tailer. */ |
43 | unsigned int sg_elems; | |
44 | ||
5087a50e MT |
45 | /* Ida index - used to track minor number allocations. */ |
46 | int index; | |
e467cde2 RR |
47 | }; |
48 | ||
49 | struct virtblk_req | |
50 | { | |
e467cde2 RR |
51 | struct request *req; |
52 | struct virtio_blk_outhdr out_hdr; | |
1cde26f9 | 53 | struct virtio_scsi_inhdr in_hdr; |
cb38fa23 | 54 | u8 status; |
a98755c5 | 55 | struct scatterlist sg[]; |
e467cde2 RR |
56 | }; |
57 | ||
a98755c5 AH |
58 | static 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 | ||
8f39db9d | 70 | static int __virtblk_add_req(struct virtqueue *vq, |
20af3cfd PB |
71 | struct virtblk_req *vbr, |
72 | struct scatterlist *data_sg, | |
0a11cc36 | 73 | bool have_data) |
c85a1f91 | 74 | { |
20af3cfd | 75 | struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6]; |
8f39db9d | 76 | unsigned int num_out = 0, num_in = 0; |
20af3cfd | 77 | int type = vbr->out_hdr.type & ~VIRTIO_BLK_T_OUT; |
8f39db9d PB |
78 | |
79 | sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr)); | |
80 | sgs[num_out++] = &hdr; | |
81 | ||
20af3cfd PB |
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 | ||
0a11cc36 | 93 | if (have_data) { |
8f39db9d | 94 | if (vbr->out_hdr.type & VIRTIO_BLK_T_OUT) |
20af3cfd | 95 | sgs[num_out++] = data_sg; |
8f39db9d | 96 | else |
20af3cfd PB |
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; | |
8f39db9d PB |
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); | |
5ee21a52 PB |
111 | } |
112 | ||
c85a1f91 | 113 | static inline void virtblk_request_done(struct virtblk_req *vbr) |
a98755c5 AH |
114 | { |
115 | struct request *req = vbr->req; | |
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 | ||
1cf7e9c6 | 126 | blk_mq_end_io(req, error); |
a98755c5 AH |
127 | } |
128 | ||
129 | static void virtblk_done(struct virtqueue *vq) | |
e467cde2 RR |
130 | { |
131 | struct virtio_blk *vblk = vq->vdev->priv; | |
1cf7e9c6 | 132 | bool req_done = false; |
e467cde2 | 133 | struct virtblk_req *vbr; |
e467cde2 | 134 | unsigned long flags; |
a98755c5 | 135 | unsigned int len; |
e467cde2 | 136 | |
1cf7e9c6 | 137 | spin_lock_irqsave(&vblk->vq_lock, flags); |
bb811108 AH |
138 | do { |
139 | virtqueue_disable_cb(vq); | |
140 | while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) { | |
1cf7e9c6 JA |
141 | virtblk_request_done(vbr); |
142 | req_done = true; | |
33659ebb | 143 | } |
7f03b17d HG |
144 | if (unlikely(virtqueue_is_broken(vq))) |
145 | break; | |
bb811108 | 146 | } while (!virtqueue_enable_cb(vq)); |
1cf7e9c6 JA |
147 | spin_unlock_irqrestore(&vblk->vq_lock, flags); |
148 | ||
e467cde2 | 149 | /* In case queue is stopped waiting for more buffers. */ |
a98755c5 | 150 | if (req_done) |
1cf7e9c6 | 151 | blk_mq_start_stopped_hw_queues(vblk->disk->queue); |
a98755c5 AH |
152 | } |
153 | ||
1cf7e9c6 | 154 | static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req) |
e467cde2 | 155 | { |
1cf7e9c6 JA |
156 | struct virtio_blk *vblk = hctx->queue->queuedata; |
157 | struct virtblk_req *vbr = req->special; | |
158 | unsigned long flags; | |
20af3cfd | 159 | unsigned int num; |
1cf7e9c6 | 160 | const bool last = (req->cmd_flags & REQ_END) != 0; |
e467cde2 | 161 | |
1cf7e9c6 | 162 | BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); |
e467cde2 RR |
163 | |
164 | vbr->req = req; | |
dd40e456 FT |
165 | if (req->cmd_flags & REQ_FLUSH) { |
166 | vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH; | |
4cb2ea28 | 167 | vbr->out_hdr.sector = 0; |
168 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
dd40e456 FT |
169 | } else { |
170 | switch (req->cmd_type) { | |
171 | case REQ_TYPE_FS: | |
172 | vbr->out_hdr.type = 0; | |
173 | vbr->out_hdr.sector = blk_rq_pos(vbr->req); | |
174 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
175 | break; | |
176 | case REQ_TYPE_BLOCK_PC: | |
177 | vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; | |
f1b0ef06 CH |
178 | vbr->out_hdr.sector = 0; |
179 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
180 | break; | |
dd40e456 FT |
181 | case REQ_TYPE_SPECIAL: |
182 | vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID; | |
183 | vbr->out_hdr.sector = 0; | |
184 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
185 | break; | |
186 | default: | |
187 | /* We don't put anything else in the queue. */ | |
188 | BUG(); | |
f1b0ef06 | 189 | } |
e467cde2 RR |
190 | } |
191 | ||
1cf7e9c6 | 192 | num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg); |
1cde26f9 | 193 | if (num) { |
20af3cfd | 194 | if (rq_data_dir(vbr->req) == WRITE) |
1cde26f9 | 195 | vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; |
20af3cfd | 196 | else |
1cde26f9 | 197 | vbr->out_hdr.type |= VIRTIO_BLK_T_IN; |
e467cde2 RR |
198 | } |
199 | ||
1cf7e9c6 JA |
200 | spin_lock_irqsave(&vblk->vq_lock, flags); |
201 | if (__virtblk_add_req(vblk->vq, vbr, vbr->sg, num) < 0) { | |
f02b9ac3 | 202 | virtqueue_kick(vblk->vq); |
1cf7e9c6 JA |
203 | spin_unlock_irqrestore(&vblk->vq_lock, flags); |
204 | blk_mq_stop_hw_queue(hctx); | |
1cf7e9c6 | 205 | return BLK_MQ_RQ_QUEUE_BUSY; |
a98755c5 AH |
206 | } |
207 | ||
1cf7e9c6 JA |
208 | if (last) |
209 | virtqueue_kick(vblk->vq); | |
f02b9ac3 SL |
210 | |
211 | spin_unlock_irqrestore(&vblk->vq_lock, flags); | |
1cf7e9c6 | 212 | return BLK_MQ_RQ_QUEUE_OK; |
a98755c5 AH |
213 | } |
214 | ||
4cb2ea28 | 215 | /* return id (s/n) string for *disk to *id_str |
216 | */ | |
217 | static int virtblk_get_id(struct gendisk *disk, char *id_str) | |
218 | { | |
219 | struct virtio_blk *vblk = disk->private_data; | |
220 | struct request *req; | |
221 | struct bio *bio; | |
e4c4776d | 222 | int err; |
4cb2ea28 | 223 | |
224 | bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES, | |
225 | GFP_KERNEL); | |
226 | if (IS_ERR(bio)) | |
227 | return PTR_ERR(bio); | |
228 | ||
229 | req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL); | |
230 | if (IS_ERR(req)) { | |
231 | bio_put(bio); | |
232 | return PTR_ERR(req); | |
233 | } | |
234 | ||
235 | req->cmd_type = REQ_TYPE_SPECIAL; | |
e4c4776d MS |
236 | err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); |
237 | blk_put_request(req); | |
238 | ||
239 | return err; | |
4cb2ea28 | 240 | } |
241 | ||
fe5a50a1 CH |
242 | static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, |
243 | unsigned int cmd, unsigned long data) | |
e467cde2 | 244 | { |
1cde26f9 HR |
245 | struct gendisk *disk = bdev->bd_disk; |
246 | struct virtio_blk *vblk = disk->private_data; | |
247 | ||
248 | /* | |
249 | * Only allow the generic SCSI ioctls if the host can support it. | |
250 | */ | |
251 | if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) | |
d9ecdea7 | 252 | return -ENOTTY; |
1cde26f9 | 253 | |
577ebb37 PB |
254 | return scsi_cmd_blk_ioctl(bdev, mode, cmd, |
255 | (void __user *)data); | |
e467cde2 RR |
256 | } |
257 | ||
135da0b0 CB |
258 | /* We provide getgeo only to please some old bootloader/partitioning tools */ |
259 | static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
260 | { | |
48e4043d | 261 | struct virtio_blk *vblk = bd->bd_disk->private_data; |
48e4043d RH |
262 | |
263 | /* see if the host passed in geometry config */ | |
855e0c52 RR |
264 | if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) { |
265 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
266 | geometry.cylinders, &geo->cylinders); | |
267 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
268 | geometry.heads, &geo->heads); | |
269 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
270 | geometry.sectors, &geo->sectors); | |
48e4043d RH |
271 | } else { |
272 | /* some standard values, similar to sd */ | |
273 | geo->heads = 1 << 6; | |
274 | geo->sectors = 1 << 5; | |
275 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
276 | } | |
135da0b0 CB |
277 | return 0; |
278 | } | |
279 | ||
83d5cde4 | 280 | static const struct block_device_operations virtblk_fops = { |
8a6cfeb6 | 281 | .ioctl = virtblk_ioctl, |
135da0b0 CB |
282 | .owner = THIS_MODULE, |
283 | .getgeo = virtblk_getgeo, | |
e467cde2 RR |
284 | }; |
285 | ||
d50ed907 CB |
286 | static int index_to_minor(int index) |
287 | { | |
288 | return index << PART_BITS; | |
289 | } | |
290 | ||
5087a50e MT |
291 | static int minor_to_index(int minor) |
292 | { | |
293 | return minor >> PART_BITS; | |
294 | } | |
295 | ||
a5eb9e4f RH |
296 | static ssize_t virtblk_serial_show(struct device *dev, |
297 | struct device_attribute *attr, char *buf) | |
298 | { | |
299 | struct gendisk *disk = dev_to_disk(dev); | |
300 | int err; | |
301 | ||
302 | /* sysfs gives us a PAGE_SIZE buffer */ | |
303 | BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); | |
304 | ||
305 | buf[VIRTIO_BLK_ID_BYTES] = '\0'; | |
306 | err = virtblk_get_id(disk, buf); | |
307 | if (!err) | |
308 | return strlen(buf); | |
309 | ||
310 | if (err == -EIO) /* Unsupported? Make it empty. */ | |
311 | return 0; | |
312 | ||
313 | return err; | |
314 | } | |
315 | DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL); | |
316 | ||
7a7c924c CH |
317 | static void virtblk_config_changed_work(struct work_struct *work) |
318 | { | |
319 | struct virtio_blk *vblk = | |
320 | container_of(work, struct virtio_blk, config_work); | |
321 | struct virtio_device *vdev = vblk->vdev; | |
322 | struct request_queue *q = vblk->disk->queue; | |
323 | char cap_str_2[10], cap_str_10[10]; | |
9d9598b8 | 324 | char *envp[] = { "RESIZE=1", NULL }; |
7a7c924c CH |
325 | u64 capacity, size; |
326 | ||
4678d6f9 MT |
327 | mutex_lock(&vblk->config_lock); |
328 | if (!vblk->config_enable) | |
329 | goto done; | |
330 | ||
7a7c924c | 331 | /* Host must always specify the capacity. */ |
855e0c52 | 332 | virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity); |
7a7c924c CH |
333 | |
334 | /* If capacity is too big, truncate with warning. */ | |
335 | if ((sector_t)capacity != capacity) { | |
336 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
337 | (unsigned long long)capacity); | |
338 | capacity = (sector_t)-1; | |
339 | } | |
340 | ||
341 | size = capacity * queue_logical_block_size(q); | |
342 | string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); | |
343 | string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); | |
344 | ||
345 | dev_notice(&vdev->dev, | |
346 | "new size: %llu %d-byte logical blocks (%s/%s)\n", | |
347 | (unsigned long long)capacity, | |
348 | queue_logical_block_size(q), | |
349 | cap_str_10, cap_str_2); | |
350 | ||
351 | set_capacity(vblk->disk, capacity); | |
e9986f30 | 352 | revalidate_disk(vblk->disk); |
9d9598b8 | 353 | kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp); |
4678d6f9 MT |
354 | done: |
355 | mutex_unlock(&vblk->config_lock); | |
7a7c924c CH |
356 | } |
357 | ||
358 | static void virtblk_config_changed(struct virtio_device *vdev) | |
359 | { | |
360 | struct virtio_blk *vblk = vdev->priv; | |
361 | ||
362 | queue_work(virtblk_wq, &vblk->config_work); | |
363 | } | |
364 | ||
6abd6e5a AS |
365 | static int init_vq(struct virtio_blk *vblk) |
366 | { | |
367 | int err = 0; | |
368 | ||
369 | /* We expect one virtqueue, for output. */ | |
a98755c5 | 370 | vblk->vq = virtio_find_single_vq(vblk->vdev, virtblk_done, "requests"); |
6abd6e5a AS |
371 | if (IS_ERR(vblk->vq)) |
372 | err = PTR_ERR(vblk->vq); | |
373 | ||
374 | return err; | |
375 | } | |
376 | ||
c0aa3e09 RM |
377 | /* |
378 | * Legacy naming scheme used for virtio devices. We are stuck with it for | |
379 | * virtio blk but don't ever use it for any new driver. | |
380 | */ | |
381 | static int virtblk_name_format(char *prefix, int index, char *buf, int buflen) | |
382 | { | |
383 | const int base = 'z' - 'a' + 1; | |
384 | char *begin = buf + strlen(prefix); | |
385 | char *end = buf + buflen; | |
386 | char *p; | |
387 | int unit; | |
388 | ||
389 | p = end - 1; | |
390 | *p = '\0'; | |
391 | unit = base; | |
392 | do { | |
393 | if (p == begin) | |
394 | return -EINVAL; | |
395 | *--p = 'a' + (index % unit); | |
396 | index = (index / unit) - 1; | |
397 | } while (index >= 0); | |
398 | ||
399 | memmove(begin, p, end - p); | |
400 | memcpy(buf, prefix, strlen(prefix)); | |
401 | ||
402 | return 0; | |
403 | } | |
404 | ||
cd5d5038 PB |
405 | static int virtblk_get_cache_mode(struct virtio_device *vdev) |
406 | { | |
407 | u8 writeback; | |
408 | int err; | |
409 | ||
855e0c52 RR |
410 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE, |
411 | struct virtio_blk_config, wce, | |
412 | &writeback); | |
cd5d5038 PB |
413 | if (err) |
414 | writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE); | |
415 | ||
416 | return writeback; | |
417 | } | |
418 | ||
419 | static void virtblk_update_cache_mode(struct virtio_device *vdev) | |
420 | { | |
421 | u8 writeback = virtblk_get_cache_mode(vdev); | |
422 | struct virtio_blk *vblk = vdev->priv; | |
423 | ||
c85a1f91 | 424 | if (writeback) |
cd5d5038 PB |
425 | blk_queue_flush(vblk->disk->queue, REQ_FLUSH); |
426 | else | |
427 | blk_queue_flush(vblk->disk->queue, 0); | |
428 | ||
429 | revalidate_disk(vblk->disk); | |
430 | } | |
431 | ||
432 | static const char *const virtblk_cache_types[] = { | |
433 | "write through", "write back" | |
434 | }; | |
435 | ||
436 | static ssize_t | |
437 | virtblk_cache_type_store(struct device *dev, struct device_attribute *attr, | |
438 | const char *buf, size_t count) | |
439 | { | |
440 | struct gendisk *disk = dev_to_disk(dev); | |
441 | struct virtio_blk *vblk = disk->private_data; | |
442 | struct virtio_device *vdev = vblk->vdev; | |
443 | int i; | |
cd5d5038 PB |
444 | |
445 | BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE)); | |
446 | for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; ) | |
447 | if (sysfs_streq(buf, virtblk_cache_types[i])) | |
448 | break; | |
449 | ||
450 | if (i < 0) | |
451 | return -EINVAL; | |
452 | ||
855e0c52 | 453 | virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i); |
cd5d5038 PB |
454 | virtblk_update_cache_mode(vdev); |
455 | return count; | |
456 | } | |
457 | ||
458 | static ssize_t | |
459 | virtblk_cache_type_show(struct device *dev, struct device_attribute *attr, | |
460 | char *buf) | |
461 | { | |
462 | struct gendisk *disk = dev_to_disk(dev); | |
463 | struct virtio_blk *vblk = disk->private_data; | |
464 | u8 writeback = virtblk_get_cache_mode(vblk->vdev); | |
465 | ||
466 | BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types)); | |
467 | return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]); | |
468 | } | |
469 | ||
470 | static const struct device_attribute dev_attr_cache_type_ro = | |
471 | __ATTR(cache_type, S_IRUGO, | |
472 | virtblk_cache_type_show, NULL); | |
473 | static const struct device_attribute dev_attr_cache_type_rw = | |
474 | __ATTR(cache_type, S_IRUGO|S_IWUSR, | |
475 | virtblk_cache_type_show, virtblk_cache_type_store); | |
476 | ||
1cf7e9c6 JA |
477 | static struct blk_mq_ops virtio_mq_ops = { |
478 | .queue_rq = virtio_queue_rq, | |
479 | .map_queue = blk_mq_map_queue, | |
480 | .alloc_hctx = blk_mq_alloc_single_hw_queue, | |
481 | .free_hctx = blk_mq_free_single_hw_queue, | |
482 | }; | |
483 | ||
484 | static struct blk_mq_reg virtio_mq_reg = { | |
485 | .ops = &virtio_mq_ops, | |
486 | .nr_hw_queues = 1, | |
487 | .queue_depth = 64, | |
488 | .numa_node = NUMA_NO_NODE, | |
489 | .flags = BLK_MQ_F_SHOULD_MERGE, | |
490 | }; | |
491 | ||
492 | static void virtblk_init_vbr(void *data, struct blk_mq_hw_ctx *hctx, | |
493 | struct request *rq, unsigned int nr) | |
494 | { | |
495 | struct virtio_blk *vblk = data; | |
496 | struct virtblk_req *vbr = rq->special; | |
497 | ||
498 | sg_init_table(vbr->sg, vblk->sg_elems); | |
499 | } | |
500 | ||
8d85fce7 | 501 | static int virtblk_probe(struct virtio_device *vdev) |
e467cde2 RR |
502 | { |
503 | struct virtio_blk *vblk; | |
69740c8b | 504 | struct request_queue *q; |
5087a50e | 505 | int err, index; |
a98755c5 | 506 | |
e467cde2 | 507 | u64 cap; |
69740c8b CH |
508 | u32 v, blk_size, sg_elems, opt_io_size; |
509 | u16 min_io_size; | |
510 | u8 physical_block_exp, alignment_offset; | |
e467cde2 | 511 | |
5087a50e MT |
512 | err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS), |
513 | GFP_KERNEL); | |
514 | if (err < 0) | |
515 | goto out; | |
516 | index = err; | |
4f3bf19c | 517 | |
0864b79a | 518 | /* We need to know how many segments before we allocate. */ |
855e0c52 RR |
519 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX, |
520 | struct virtio_blk_config, seg_max, | |
521 | &sg_elems); | |
a5b365a6 CH |
522 | |
523 | /* We need at least one SG element, whatever they say. */ | |
524 | if (err || !sg_elems) | |
0864b79a RR |
525 | sg_elems = 1; |
526 | ||
527 | /* We need an extra sg elements at head and tail. */ | |
528 | sg_elems += 2; | |
1cf7e9c6 | 529 | vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL); |
e467cde2 RR |
530 | if (!vblk) { |
531 | err = -ENOMEM; | |
5087a50e | 532 | goto out_free_index; |
e467cde2 RR |
533 | } |
534 | ||
e467cde2 | 535 | vblk->vdev = vdev; |
0864b79a | 536 | vblk->sg_elems = sg_elems; |
4678d6f9 | 537 | mutex_init(&vblk->config_lock); |
a98755c5 | 538 | |
7a7c924c | 539 | INIT_WORK(&vblk->config_work, virtblk_config_changed_work); |
4678d6f9 | 540 | vblk->config_enable = true; |
e467cde2 | 541 | |
6abd6e5a AS |
542 | err = init_vq(vblk); |
543 | if (err) | |
e467cde2 | 544 | goto out_free_vblk; |
1cf7e9c6 | 545 | spin_lock_init(&vblk->vq_lock); |
e467cde2 | 546 | |
e467cde2 | 547 | /* FIXME: How many partitions? How long is a piece of string? */ |
4f3bf19c | 548 | vblk->disk = alloc_disk(1 << PART_BITS); |
e467cde2 RR |
549 | if (!vblk->disk) { |
550 | err = -ENOMEM; | |
1cf7e9c6 | 551 | goto out_free_vq; |
e467cde2 RR |
552 | } |
553 | ||
1cf7e9c6 JA |
554 | virtio_mq_reg.cmd_size = |
555 | sizeof(struct virtblk_req) + | |
556 | sizeof(struct scatterlist) * sg_elems; | |
557 | ||
558 | q = vblk->disk->queue = blk_mq_init_queue(&virtio_mq_reg, vblk); | |
69740c8b | 559 | if (!q) { |
e467cde2 RR |
560 | err = -ENOMEM; |
561 | goto out_put_disk; | |
562 | } | |
563 | ||
1cf7e9c6 JA |
564 | blk_mq_init_commands(q, virtblk_init_vbr, vblk); |
565 | ||
69740c8b | 566 | q->queuedata = vblk; |
7d116b62 | 567 | |
c0aa3e09 | 568 | virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN); |
d50ed907 | 569 | |
e467cde2 | 570 | vblk->disk->major = major; |
d50ed907 | 571 | vblk->disk->first_minor = index_to_minor(index); |
e467cde2 RR |
572 | vblk->disk->private_data = vblk; |
573 | vblk->disk->fops = &virtblk_fops; | |
c4839346 | 574 | vblk->disk->driverfs_dev = &vdev->dev; |
5087a50e | 575 | vblk->index = index; |
4f3bf19c | 576 | |
02c42b7a | 577 | /* configure queue flush support */ |
cd5d5038 | 578 | virtblk_update_cache_mode(vdev); |
e467cde2 | 579 | |
3ef53609 CB |
580 | /* If disk is read-only in the host, the guest should obey */ |
581 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) | |
582 | set_disk_ro(vblk->disk, 1); | |
583 | ||
a586d4f6 | 584 | /* Host must always specify the capacity. */ |
855e0c52 | 585 | virtio_cread(vdev, struct virtio_blk_config, capacity, &cap); |
e467cde2 RR |
586 | |
587 | /* If capacity is too big, truncate with warning. */ | |
588 | if ((sector_t)cap != cap) { | |
589 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
590 | (unsigned long long)cap); | |
591 | cap = (sector_t)-1; | |
592 | } | |
593 | set_capacity(vblk->disk, cap); | |
594 | ||
0864b79a | 595 | /* We can handle whatever the host told us to handle. */ |
ee714f2d | 596 | blk_queue_max_segments(q, vblk->sg_elems-2); |
0864b79a | 597 | |
4eff3cae | 598 | /* No need to bounce any requests */ |
69740c8b | 599 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
4eff3cae | 600 | |
4b7f7e20 | 601 | /* No real sector limit. */ |
ee714f2d | 602 | blk_queue_max_hw_sectors(q, -1U); |
4b7f7e20 | 603 | |
a586d4f6 RR |
604 | /* Host can optionally specify maximum segment size and number of |
605 | * segments. */ | |
855e0c52 RR |
606 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX, |
607 | struct virtio_blk_config, size_max, &v); | |
e467cde2 | 608 | if (!err) |
69740c8b | 609 | blk_queue_max_segment_size(q, v); |
4b7f7e20 | 610 | else |
69740c8b | 611 | blk_queue_max_segment_size(q, -1U); |
e467cde2 | 612 | |
066f4d82 | 613 | /* Host can optionally specify the block size of the device */ |
855e0c52 RR |
614 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE, |
615 | struct virtio_blk_config, blk_size, | |
616 | &blk_size); | |
066f4d82 | 617 | if (!err) |
69740c8b CH |
618 | blk_queue_logical_block_size(q, blk_size); |
619 | else | |
620 | blk_size = queue_logical_block_size(q); | |
621 | ||
622 | /* Use topology information if available */ | |
855e0c52 RR |
623 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
624 | struct virtio_blk_config, physical_block_exp, | |
625 | &physical_block_exp); | |
69740c8b CH |
626 | if (!err && physical_block_exp) |
627 | blk_queue_physical_block_size(q, | |
628 | blk_size * (1 << physical_block_exp)); | |
629 | ||
855e0c52 RR |
630 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
631 | struct virtio_blk_config, alignment_offset, | |
632 | &alignment_offset); | |
69740c8b CH |
633 | if (!err && alignment_offset) |
634 | blk_queue_alignment_offset(q, blk_size * alignment_offset); | |
635 | ||
855e0c52 RR |
636 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
637 | struct virtio_blk_config, min_io_size, | |
638 | &min_io_size); | |
69740c8b CH |
639 | if (!err && min_io_size) |
640 | blk_queue_io_min(q, blk_size * min_io_size); | |
641 | ||
855e0c52 RR |
642 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
643 | struct virtio_blk_config, opt_io_size, | |
644 | &opt_io_size); | |
69740c8b CH |
645 | if (!err && opt_io_size) |
646 | blk_queue_io_opt(q, blk_size * opt_io_size); | |
647 | ||
e467cde2 | 648 | add_disk(vblk->disk); |
a5eb9e4f RH |
649 | err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial); |
650 | if (err) | |
651 | goto out_del_disk; | |
652 | ||
cd5d5038 PB |
653 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) |
654 | err = device_create_file(disk_to_dev(vblk->disk), | |
655 | &dev_attr_cache_type_rw); | |
656 | else | |
657 | err = device_create_file(disk_to_dev(vblk->disk), | |
658 | &dev_attr_cache_type_ro); | |
659 | if (err) | |
660 | goto out_del_disk; | |
e467cde2 RR |
661 | return 0; |
662 | ||
a5eb9e4f RH |
663 | out_del_disk: |
664 | del_gendisk(vblk->disk); | |
665 | blk_cleanup_queue(vblk->disk->queue); | |
e467cde2 RR |
666 | out_put_disk: |
667 | put_disk(vblk->disk); | |
e467cde2 | 668 | out_free_vq: |
d2a7ddda | 669 | vdev->config->del_vqs(vdev); |
e467cde2 RR |
670 | out_free_vblk: |
671 | kfree(vblk); | |
5087a50e MT |
672 | out_free_index: |
673 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
674 | out: |
675 | return err; | |
676 | } | |
677 | ||
8d85fce7 | 678 | static void virtblk_remove(struct virtio_device *vdev) |
e467cde2 RR |
679 | { |
680 | struct virtio_blk *vblk = vdev->priv; | |
5087a50e | 681 | int index = vblk->index; |
f4953fe6 | 682 | int refc; |
e467cde2 | 683 | |
4678d6f9 MT |
684 | /* Prevent config work handler from accessing the device. */ |
685 | mutex_lock(&vblk->config_lock); | |
686 | vblk->config_enable = false; | |
687 | mutex_unlock(&vblk->config_lock); | |
7a7c924c | 688 | |
02e2b124 | 689 | del_gendisk(vblk->disk); |
483001c7 | 690 | blk_cleanup_queue(vblk->disk->queue); |
02e2b124 | 691 | |
6e5aa7ef RR |
692 | /* Stop all the virtqueues. */ |
693 | vdev->config->reset(vdev); | |
694 | ||
4678d6f9 MT |
695 | flush_work(&vblk->config_work); |
696 | ||
f4953fe6 | 697 | refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount); |
e467cde2 | 698 | put_disk(vblk->disk); |
d2a7ddda | 699 | vdev->config->del_vqs(vdev); |
e467cde2 | 700 | kfree(vblk); |
f4953fe6 AG |
701 | |
702 | /* Only free device id if we don't have any users */ | |
703 | if (refc == 1) | |
704 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
705 | } |
706 | ||
89107000 | 707 | #ifdef CONFIG_PM_SLEEP |
f8fb5bc2 AS |
708 | static int virtblk_freeze(struct virtio_device *vdev) |
709 | { | |
710 | struct virtio_blk *vblk = vdev->priv; | |
711 | ||
712 | /* Ensure we don't receive any more interrupts */ | |
713 | vdev->config->reset(vdev); | |
714 | ||
715 | /* Prevent config work handler from accessing the device. */ | |
716 | mutex_lock(&vblk->config_lock); | |
717 | vblk->config_enable = false; | |
718 | mutex_unlock(&vblk->config_lock); | |
719 | ||
720 | flush_work(&vblk->config_work); | |
721 | ||
1cf7e9c6 | 722 | blk_mq_stop_hw_queues(vblk->disk->queue); |
f8fb5bc2 AS |
723 | |
724 | vdev->config->del_vqs(vdev); | |
725 | return 0; | |
726 | } | |
727 | ||
728 | static int virtblk_restore(struct virtio_device *vdev) | |
729 | { | |
730 | struct virtio_blk *vblk = vdev->priv; | |
731 | int ret; | |
732 | ||
733 | vblk->config_enable = true; | |
734 | ret = init_vq(vdev->priv); | |
1cf7e9c6 JA |
735 | if (!ret) |
736 | blk_mq_start_stopped_hw_queues(vblk->disk->queue); | |
737 | ||
f8fb5bc2 AS |
738 | return ret; |
739 | } | |
740 | #endif | |
741 | ||
47483e25 | 742 | static const struct virtio_device_id id_table[] = { |
e467cde2 RR |
743 | { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, |
744 | { 0 }, | |
745 | }; | |
746 | ||
c45a6816 | 747 | static unsigned int features[] = { |
02c42b7a TH |
748 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, |
749 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI, | |
cd5d5038 | 750 | VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE |
c45a6816 RR |
751 | }; |
752 | ||
8d85fce7 | 753 | static struct virtio_driver virtio_blk = { |
7a7c924c CH |
754 | .feature_table = features, |
755 | .feature_table_size = ARRAY_SIZE(features), | |
756 | .driver.name = KBUILD_MODNAME, | |
757 | .driver.owner = THIS_MODULE, | |
758 | .id_table = id_table, | |
759 | .probe = virtblk_probe, | |
8d85fce7 | 760 | .remove = virtblk_remove, |
7a7c924c | 761 | .config_changed = virtblk_config_changed, |
89107000 | 762 | #ifdef CONFIG_PM_SLEEP |
f8fb5bc2 AS |
763 | .freeze = virtblk_freeze, |
764 | .restore = virtblk_restore, | |
765 | #endif | |
e467cde2 RR |
766 | }; |
767 | ||
768 | static int __init init(void) | |
769 | { | |
7a7c924c CH |
770 | int error; |
771 | ||
772 | virtblk_wq = alloc_workqueue("virtio-blk", 0, 0); | |
773 | if (!virtblk_wq) | |
774 | return -ENOMEM; | |
775 | ||
4f3bf19c | 776 | major = register_blkdev(0, "virtblk"); |
7a7c924c CH |
777 | if (major < 0) { |
778 | error = major; | |
779 | goto out_destroy_workqueue; | |
780 | } | |
781 | ||
782 | error = register_virtio_driver(&virtio_blk); | |
783 | if (error) | |
784 | goto out_unregister_blkdev; | |
785 | return 0; | |
786 | ||
787 | out_unregister_blkdev: | |
788 | unregister_blkdev(major, "virtblk"); | |
789 | out_destroy_workqueue: | |
790 | destroy_workqueue(virtblk_wq); | |
791 | return error; | |
e467cde2 RR |
792 | } |
793 | ||
794 | static void __exit fini(void) | |
795 | { | |
4f3bf19c | 796 | unregister_blkdev(major, "virtblk"); |
e467cde2 | 797 | unregister_virtio_driver(&virtio_blk); |
7a7c924c | 798 | destroy_workqueue(virtblk_wq); |
e467cde2 RR |
799 | } |
800 | module_init(init); | |
801 | module_exit(fini); | |
802 | ||
803 | MODULE_DEVICE_TABLE(virtio, id_table); | |
804 | MODULE_DESCRIPTION("Virtio block driver"); | |
805 | MODULE_LICENSE("GPL"); |