]>
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 | ||
24d2f903 CH |
33 | /* Block layer tags. */ |
34 | struct blk_mq_tag_set tag_set; | |
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; | |
e467cde2 RR |
50 | }; |
51 | ||
52 | struct virtblk_req | |
53 | { | |
e467cde2 RR |
54 | struct request *req; |
55 | struct virtio_blk_outhdr out_hdr; | |
1cde26f9 | 56 | struct virtio_scsi_inhdr in_hdr; |
cb38fa23 | 57 | u8 status; |
a98755c5 | 58 | struct scatterlist sg[]; |
e467cde2 RR |
59 | }; |
60 | ||
a98755c5 AH |
61 | static inline int virtblk_result(struct virtblk_req *vbr) |
62 | { | |
63 | switch (vbr->status) { | |
64 | case VIRTIO_BLK_S_OK: | |
65 | return 0; | |
66 | case VIRTIO_BLK_S_UNSUPP: | |
67 | return -ENOTTY; | |
68 | default: | |
69 | return -EIO; | |
70 | } | |
71 | } | |
72 | ||
8f39db9d | 73 | static int __virtblk_add_req(struct virtqueue *vq, |
20af3cfd PB |
74 | struct virtblk_req *vbr, |
75 | struct scatterlist *data_sg, | |
0a11cc36 | 76 | bool have_data) |
c85a1f91 | 77 | { |
20af3cfd | 78 | struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6]; |
8f39db9d | 79 | unsigned int num_out = 0, num_in = 0; |
20af3cfd | 80 | int type = vbr->out_hdr.type & ~VIRTIO_BLK_T_OUT; |
8f39db9d PB |
81 | |
82 | sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr)); | |
83 | sgs[num_out++] = &hdr; | |
84 | ||
20af3cfd PB |
85 | /* |
86 | * If this is a packet command we need a couple of additional headers. | |
87 | * Behind the normal outhdr we put a segment with the scsi command | |
88 | * block, and before the normal inhdr we put the sense data and the | |
89 | * inhdr with additional status information. | |
90 | */ | |
91 | if (type == VIRTIO_BLK_T_SCSI_CMD) { | |
92 | sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len); | |
93 | sgs[num_out++] = &cmd; | |
94 | } | |
95 | ||
0a11cc36 | 96 | if (have_data) { |
8f39db9d | 97 | if (vbr->out_hdr.type & VIRTIO_BLK_T_OUT) |
20af3cfd | 98 | sgs[num_out++] = data_sg; |
8f39db9d | 99 | else |
20af3cfd PB |
100 | sgs[num_out + num_in++] = data_sg; |
101 | } | |
102 | ||
103 | if (type == VIRTIO_BLK_T_SCSI_CMD) { | |
104 | sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE); | |
105 | sgs[num_out + num_in++] = &sense; | |
106 | sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr)); | |
107 | sgs[num_out + num_in++] = &inhdr; | |
8f39db9d PB |
108 | } |
109 | ||
110 | sg_init_one(&status, &vbr->status, sizeof(vbr->status)); | |
111 | sgs[num_out + num_in++] = &status; | |
112 | ||
113 | return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC); | |
5ee21a52 PB |
114 | } |
115 | ||
5124c285 | 116 | static inline void virtblk_request_done(struct request *req) |
a98755c5 | 117 | { |
9d74e257 | 118 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); |
a98755c5 AH |
119 | int error = virtblk_result(vbr); |
120 | ||
121 | if (req->cmd_type == REQ_TYPE_BLOCK_PC) { | |
122 | req->resid_len = vbr->in_hdr.residual; | |
123 | req->sense_len = vbr->in_hdr.sense_len; | |
124 | req->errors = vbr->in_hdr.errors; | |
125 | } else if (req->cmd_type == REQ_TYPE_SPECIAL) { | |
126 | req->errors = (error != 0); | |
127 | } | |
128 | ||
1cf7e9c6 | 129 | blk_mq_end_io(req, error); |
a98755c5 AH |
130 | } |
131 | ||
132 | static void virtblk_done(struct virtqueue *vq) | |
e467cde2 RR |
133 | { |
134 | struct virtio_blk *vblk = vq->vdev->priv; | |
1cf7e9c6 | 135 | bool req_done = false; |
e467cde2 | 136 | struct virtblk_req *vbr; |
e467cde2 | 137 | unsigned long flags; |
a98755c5 | 138 | unsigned int len; |
e467cde2 | 139 | |
1cf7e9c6 | 140 | spin_lock_irqsave(&vblk->vq_lock, flags); |
bb811108 AH |
141 | do { |
142 | virtqueue_disable_cb(vq); | |
143 | while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) { | |
5124c285 | 144 | blk_mq_complete_request(vbr->req); |
1cf7e9c6 | 145 | req_done = true; |
33659ebb | 146 | } |
7f03b17d HG |
147 | if (unlikely(virtqueue_is_broken(vq))) |
148 | break; | |
bb811108 | 149 | } while (!virtqueue_enable_cb(vq)); |
1cf7e9c6 | 150 | |
e467cde2 | 151 | /* In case queue is stopped waiting for more buffers. */ |
a98755c5 | 152 | if (req_done) |
1b4a3258 | 153 | blk_mq_start_stopped_hw_queues(vblk->disk->queue, true); |
0c29e93e | 154 | spin_unlock_irqrestore(&vblk->vq_lock, flags); |
a98755c5 AH |
155 | } |
156 | ||
1cf7e9c6 | 157 | static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req) |
e467cde2 | 158 | { |
1cf7e9c6 | 159 | struct virtio_blk *vblk = hctx->queue->queuedata; |
9d74e257 | 160 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); |
1cf7e9c6 | 161 | unsigned long flags; |
20af3cfd | 162 | unsigned int num; |
1cf7e9c6 | 163 | const bool last = (req->cmd_flags & REQ_END) != 0; |
5261b85e | 164 | int err; |
e8edca6f | 165 | bool notify = false; |
e467cde2 | 166 | |
1cf7e9c6 | 167 | BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); |
e467cde2 RR |
168 | |
169 | vbr->req = req; | |
dd40e456 FT |
170 | if (req->cmd_flags & REQ_FLUSH) { |
171 | vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH; | |
4cb2ea28 | 172 | vbr->out_hdr.sector = 0; |
173 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
dd40e456 FT |
174 | } else { |
175 | switch (req->cmd_type) { | |
176 | case REQ_TYPE_FS: | |
177 | vbr->out_hdr.type = 0; | |
178 | vbr->out_hdr.sector = blk_rq_pos(vbr->req); | |
179 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
180 | break; | |
181 | case REQ_TYPE_BLOCK_PC: | |
182 | vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; | |
f1b0ef06 CH |
183 | vbr->out_hdr.sector = 0; |
184 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
185 | break; | |
dd40e456 FT |
186 | case REQ_TYPE_SPECIAL: |
187 | vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID; | |
188 | vbr->out_hdr.sector = 0; | |
189 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
190 | break; | |
191 | default: | |
192 | /* We don't put anything else in the queue. */ | |
193 | BUG(); | |
f1b0ef06 | 194 | } |
e467cde2 RR |
195 | } |
196 | ||
1cf7e9c6 | 197 | num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg); |
1cde26f9 | 198 | if (num) { |
20af3cfd | 199 | if (rq_data_dir(vbr->req) == WRITE) |
1cde26f9 | 200 | vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; |
20af3cfd | 201 | else |
1cde26f9 | 202 | vbr->out_hdr.type |= VIRTIO_BLK_T_IN; |
e467cde2 RR |
203 | } |
204 | ||
1cf7e9c6 | 205 | spin_lock_irqsave(&vblk->vq_lock, flags); |
5261b85e RR |
206 | err = __virtblk_add_req(vblk->vq, vbr, vbr->sg, num); |
207 | if (err) { | |
f02b9ac3 | 208 | virtqueue_kick(vblk->vq); |
1cf7e9c6 | 209 | blk_mq_stop_hw_queue(hctx); |
0c29e93e | 210 | spin_unlock_irqrestore(&vblk->vq_lock, flags); |
5261b85e RR |
211 | /* Out of mem doesn't actually happen, since we fall back |
212 | * to direct descriptors */ | |
213 | if (err == -ENOMEM || err == -ENOSPC) | |
214 | return BLK_MQ_RQ_QUEUE_BUSY; | |
215 | return BLK_MQ_RQ_QUEUE_ERROR; | |
a98755c5 AH |
216 | } |
217 | ||
e8edca6f ML |
218 | if (last && virtqueue_kick_prepare(vblk->vq)) |
219 | notify = true; | |
f02b9ac3 | 220 | spin_unlock_irqrestore(&vblk->vq_lock, flags); |
e8edca6f ML |
221 | |
222 | if (notify) | |
223 | virtqueue_notify(vblk->vq); | |
1cf7e9c6 | 224 | return BLK_MQ_RQ_QUEUE_OK; |
a98755c5 AH |
225 | } |
226 | ||
4cb2ea28 | 227 | /* return id (s/n) string for *disk to *id_str |
228 | */ | |
229 | static int virtblk_get_id(struct gendisk *disk, char *id_str) | |
230 | { | |
231 | struct virtio_blk *vblk = disk->private_data; | |
232 | struct request *req; | |
233 | struct bio *bio; | |
e4c4776d | 234 | int err; |
4cb2ea28 | 235 | |
236 | bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES, | |
237 | GFP_KERNEL); | |
238 | if (IS_ERR(bio)) | |
239 | return PTR_ERR(bio); | |
240 | ||
241 | req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL); | |
242 | if (IS_ERR(req)) { | |
243 | bio_put(bio); | |
244 | return PTR_ERR(req); | |
245 | } | |
246 | ||
247 | req->cmd_type = REQ_TYPE_SPECIAL; | |
e4c4776d MS |
248 | err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); |
249 | blk_put_request(req); | |
250 | ||
251 | return err; | |
4cb2ea28 | 252 | } |
253 | ||
fe5a50a1 CH |
254 | static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, |
255 | unsigned int cmd, unsigned long data) | |
e467cde2 | 256 | { |
1cde26f9 HR |
257 | struct gendisk *disk = bdev->bd_disk; |
258 | struct virtio_blk *vblk = disk->private_data; | |
259 | ||
260 | /* | |
261 | * Only allow the generic SCSI ioctls if the host can support it. | |
262 | */ | |
263 | if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) | |
d9ecdea7 | 264 | return -ENOTTY; |
1cde26f9 | 265 | |
577ebb37 PB |
266 | return scsi_cmd_blk_ioctl(bdev, mode, cmd, |
267 | (void __user *)data); | |
e467cde2 RR |
268 | } |
269 | ||
135da0b0 CB |
270 | /* We provide getgeo only to please some old bootloader/partitioning tools */ |
271 | static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
272 | { | |
48e4043d | 273 | struct virtio_blk *vblk = bd->bd_disk->private_data; |
48e4043d RH |
274 | |
275 | /* see if the host passed in geometry config */ | |
855e0c52 RR |
276 | if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) { |
277 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
278 | geometry.cylinders, &geo->cylinders); | |
279 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
280 | geometry.heads, &geo->heads); | |
281 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
282 | geometry.sectors, &geo->sectors); | |
48e4043d RH |
283 | } else { |
284 | /* some standard values, similar to sd */ | |
285 | geo->heads = 1 << 6; | |
286 | geo->sectors = 1 << 5; | |
287 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
288 | } | |
135da0b0 CB |
289 | return 0; |
290 | } | |
291 | ||
83d5cde4 | 292 | static const struct block_device_operations virtblk_fops = { |
8a6cfeb6 | 293 | .ioctl = virtblk_ioctl, |
135da0b0 CB |
294 | .owner = THIS_MODULE, |
295 | .getgeo = virtblk_getgeo, | |
e467cde2 RR |
296 | }; |
297 | ||
d50ed907 CB |
298 | static int index_to_minor(int index) |
299 | { | |
300 | return index << PART_BITS; | |
301 | } | |
302 | ||
5087a50e MT |
303 | static int minor_to_index(int minor) |
304 | { | |
305 | return minor >> PART_BITS; | |
306 | } | |
307 | ||
a5eb9e4f RH |
308 | static ssize_t virtblk_serial_show(struct device *dev, |
309 | struct device_attribute *attr, char *buf) | |
310 | { | |
311 | struct gendisk *disk = dev_to_disk(dev); | |
312 | int err; | |
313 | ||
314 | /* sysfs gives us a PAGE_SIZE buffer */ | |
315 | BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); | |
316 | ||
317 | buf[VIRTIO_BLK_ID_BYTES] = '\0'; | |
318 | err = virtblk_get_id(disk, buf); | |
319 | if (!err) | |
320 | return strlen(buf); | |
321 | ||
322 | if (err == -EIO) /* Unsupported? Make it empty. */ | |
323 | return 0; | |
324 | ||
325 | return err; | |
326 | } | |
327 | DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL); | |
328 | ||
7a7c924c CH |
329 | static void virtblk_config_changed_work(struct work_struct *work) |
330 | { | |
331 | struct virtio_blk *vblk = | |
332 | container_of(work, struct virtio_blk, config_work); | |
333 | struct virtio_device *vdev = vblk->vdev; | |
334 | struct request_queue *q = vblk->disk->queue; | |
335 | char cap_str_2[10], cap_str_10[10]; | |
9d9598b8 | 336 | char *envp[] = { "RESIZE=1", NULL }; |
7a7c924c CH |
337 | u64 capacity, size; |
338 | ||
4678d6f9 MT |
339 | mutex_lock(&vblk->config_lock); |
340 | if (!vblk->config_enable) | |
341 | goto done; | |
342 | ||
7a7c924c | 343 | /* Host must always specify the capacity. */ |
855e0c52 | 344 | virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity); |
7a7c924c CH |
345 | |
346 | /* If capacity is too big, truncate with warning. */ | |
347 | if ((sector_t)capacity != capacity) { | |
348 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
349 | (unsigned long long)capacity); | |
350 | capacity = (sector_t)-1; | |
351 | } | |
352 | ||
353 | size = capacity * queue_logical_block_size(q); | |
354 | string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); | |
355 | string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); | |
356 | ||
357 | dev_notice(&vdev->dev, | |
358 | "new size: %llu %d-byte logical blocks (%s/%s)\n", | |
359 | (unsigned long long)capacity, | |
360 | queue_logical_block_size(q), | |
361 | cap_str_10, cap_str_2); | |
362 | ||
363 | set_capacity(vblk->disk, capacity); | |
e9986f30 | 364 | revalidate_disk(vblk->disk); |
9d9598b8 | 365 | kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp); |
4678d6f9 MT |
366 | done: |
367 | mutex_unlock(&vblk->config_lock); | |
7a7c924c CH |
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 | ||
6abd6e5a AS |
377 | static int init_vq(struct virtio_blk *vblk) |
378 | { | |
379 | int err = 0; | |
380 | ||
381 | /* We expect one virtqueue, for output. */ | |
a98755c5 | 382 | vblk->vq = virtio_find_single_vq(vblk->vdev, virtblk_done, "requests"); |
6abd6e5a AS |
383 | if (IS_ERR(vblk->vq)) |
384 | err = PTR_ERR(vblk->vq); | |
385 | ||
386 | return err; | |
387 | } | |
388 | ||
c0aa3e09 RM |
389 | /* |
390 | * Legacy naming scheme used for virtio devices. We are stuck with it for | |
391 | * virtio blk but don't ever use it for any new driver. | |
392 | */ | |
393 | static int virtblk_name_format(char *prefix, int index, char *buf, int buflen) | |
394 | { | |
395 | const int base = 'z' - 'a' + 1; | |
396 | char *begin = buf + strlen(prefix); | |
397 | char *end = buf + buflen; | |
398 | char *p; | |
399 | int unit; | |
400 | ||
401 | p = end - 1; | |
402 | *p = '\0'; | |
403 | unit = base; | |
404 | do { | |
405 | if (p == begin) | |
406 | return -EINVAL; | |
407 | *--p = 'a' + (index % unit); | |
408 | index = (index / unit) - 1; | |
409 | } while (index >= 0); | |
410 | ||
411 | memmove(begin, p, end - p); | |
412 | memcpy(buf, prefix, strlen(prefix)); | |
413 | ||
414 | return 0; | |
415 | } | |
416 | ||
cd5d5038 PB |
417 | static int virtblk_get_cache_mode(struct virtio_device *vdev) |
418 | { | |
419 | u8 writeback; | |
420 | int err; | |
421 | ||
855e0c52 RR |
422 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE, |
423 | struct virtio_blk_config, wce, | |
424 | &writeback); | |
cd5d5038 PB |
425 | if (err) |
426 | writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE); | |
427 | ||
428 | return writeback; | |
429 | } | |
430 | ||
431 | static void virtblk_update_cache_mode(struct virtio_device *vdev) | |
432 | { | |
433 | u8 writeback = virtblk_get_cache_mode(vdev); | |
434 | struct virtio_blk *vblk = vdev->priv; | |
435 | ||
c85a1f91 | 436 | if (writeback) |
cd5d5038 PB |
437 | blk_queue_flush(vblk->disk->queue, REQ_FLUSH); |
438 | else | |
439 | blk_queue_flush(vblk->disk->queue, 0); | |
440 | ||
441 | revalidate_disk(vblk->disk); | |
442 | } | |
443 | ||
444 | static const char *const virtblk_cache_types[] = { | |
445 | "write through", "write back" | |
446 | }; | |
447 | ||
448 | static ssize_t | |
449 | virtblk_cache_type_store(struct device *dev, struct device_attribute *attr, | |
450 | const char *buf, size_t count) | |
451 | { | |
452 | struct gendisk *disk = dev_to_disk(dev); | |
453 | struct virtio_blk *vblk = disk->private_data; | |
454 | struct virtio_device *vdev = vblk->vdev; | |
455 | int i; | |
cd5d5038 PB |
456 | |
457 | BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE)); | |
458 | for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; ) | |
459 | if (sysfs_streq(buf, virtblk_cache_types[i])) | |
460 | break; | |
461 | ||
462 | if (i < 0) | |
463 | return -EINVAL; | |
464 | ||
855e0c52 | 465 | virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i); |
cd5d5038 PB |
466 | virtblk_update_cache_mode(vdev); |
467 | return count; | |
468 | } | |
469 | ||
470 | static ssize_t | |
471 | virtblk_cache_type_show(struct device *dev, struct device_attribute *attr, | |
472 | char *buf) | |
473 | { | |
474 | struct gendisk *disk = dev_to_disk(dev); | |
475 | struct virtio_blk *vblk = disk->private_data; | |
476 | u8 writeback = virtblk_get_cache_mode(vblk->vdev); | |
477 | ||
478 | BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types)); | |
479 | return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]); | |
480 | } | |
481 | ||
482 | static const struct device_attribute dev_attr_cache_type_ro = | |
483 | __ATTR(cache_type, S_IRUGO, | |
484 | virtblk_cache_type_show, NULL); | |
485 | static const struct device_attribute dev_attr_cache_type_rw = | |
486 | __ATTR(cache_type, S_IRUGO|S_IWUSR, | |
487 | virtblk_cache_type_show, virtblk_cache_type_store); | |
488 | ||
24d2f903 CH |
489 | static int virtblk_init_request(void *data, struct request *rq, |
490 | unsigned int hctx_idx, unsigned int request_idx, | |
491 | unsigned int numa_node) | |
e9b267d9 CH |
492 | { |
493 | struct virtio_blk *vblk = data; | |
494 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq); | |
495 | ||
496 | sg_init_table(vbr->sg, vblk->sg_elems); | |
497 | return 0; | |
498 | } | |
499 | ||
1cf7e9c6 JA |
500 | static struct blk_mq_ops virtio_mq_ops = { |
501 | .queue_rq = virtio_queue_rq, | |
502 | .map_queue = blk_mq_map_queue, | |
5124c285 | 503 | .complete = virtblk_request_done, |
24d2f903 | 504 | .init_request = virtblk_init_request, |
1cf7e9c6 JA |
505 | }; |
506 | ||
24d2f903 CH |
507 | static unsigned int virtblk_queue_depth; |
508 | module_param_named(queue_depth, virtblk_queue_depth, uint, 0444); | |
1cf7e9c6 | 509 | |
8d85fce7 | 510 | static int virtblk_probe(struct virtio_device *vdev) |
e467cde2 RR |
511 | { |
512 | struct virtio_blk *vblk; | |
69740c8b | 513 | struct request_queue *q; |
5087a50e | 514 | int err, index; |
a98755c5 | 515 | |
e467cde2 | 516 | u64 cap; |
69740c8b CH |
517 | u32 v, blk_size, sg_elems, opt_io_size; |
518 | u16 min_io_size; | |
519 | u8 physical_block_exp, alignment_offset; | |
e467cde2 | 520 | |
5087a50e MT |
521 | err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS), |
522 | GFP_KERNEL); | |
523 | if (err < 0) | |
524 | goto out; | |
525 | index = err; | |
4f3bf19c | 526 | |
0864b79a | 527 | /* We need to know how many segments before we allocate. */ |
855e0c52 RR |
528 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX, |
529 | struct virtio_blk_config, seg_max, | |
530 | &sg_elems); | |
a5b365a6 CH |
531 | |
532 | /* We need at least one SG element, whatever they say. */ | |
533 | if (err || !sg_elems) | |
0864b79a RR |
534 | sg_elems = 1; |
535 | ||
536 | /* We need an extra sg elements at head and tail. */ | |
537 | sg_elems += 2; | |
1cf7e9c6 | 538 | vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL); |
e467cde2 RR |
539 | if (!vblk) { |
540 | err = -ENOMEM; | |
5087a50e | 541 | goto out_free_index; |
e467cde2 RR |
542 | } |
543 | ||
e467cde2 | 544 | vblk->vdev = vdev; |
0864b79a | 545 | vblk->sg_elems = sg_elems; |
4678d6f9 | 546 | mutex_init(&vblk->config_lock); |
a98755c5 | 547 | |
7a7c924c | 548 | INIT_WORK(&vblk->config_work, virtblk_config_changed_work); |
4678d6f9 | 549 | vblk->config_enable = true; |
e467cde2 | 550 | |
6abd6e5a AS |
551 | err = init_vq(vblk); |
552 | if (err) | |
e467cde2 | 553 | goto out_free_vblk; |
1cf7e9c6 | 554 | spin_lock_init(&vblk->vq_lock); |
e467cde2 | 555 | |
e467cde2 | 556 | /* FIXME: How many partitions? How long is a piece of string? */ |
4f3bf19c | 557 | vblk->disk = alloc_disk(1 << PART_BITS); |
e467cde2 RR |
558 | if (!vblk->disk) { |
559 | err = -ENOMEM; | |
1cf7e9c6 | 560 | goto out_free_vq; |
e467cde2 RR |
561 | } |
562 | ||
fc4324b4 | 563 | /* Default queue sizing is to fill the ring. */ |
24d2f903 CH |
564 | if (!virtblk_queue_depth) { |
565 | virtblk_queue_depth = vblk->vq->num_free; | |
fc4324b4 RR |
566 | /* ... but without indirect descs, we use 2 descs per req */ |
567 | if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC)) | |
24d2f903 | 568 | virtblk_queue_depth /= 2; |
fc4324b4 | 569 | } |
24d2f903 CH |
570 | |
571 | memset(&vblk->tag_set, 0, sizeof(vblk->tag_set)); | |
572 | vblk->tag_set.ops = &virtio_mq_ops; | |
573 | vblk->tag_set.nr_hw_queues = 1; | |
574 | vblk->tag_set.queue_depth = virtblk_queue_depth; | |
575 | vblk->tag_set.numa_node = NUMA_NO_NODE; | |
576 | vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; | |
577 | vblk->tag_set.cmd_size = | |
1cf7e9c6 JA |
578 | sizeof(struct virtblk_req) + |
579 | sizeof(struct scatterlist) * sg_elems; | |
24d2f903 | 580 | vblk->tag_set.driver_data = vblk; |
1cf7e9c6 | 581 | |
24d2f903 CH |
582 | err = blk_mq_alloc_tag_set(&vblk->tag_set); |
583 | if (err) | |
584 | goto out_put_disk; | |
585 | ||
586 | q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set); | |
69740c8b | 587 | if (!q) { |
e467cde2 | 588 | err = -ENOMEM; |
24d2f903 | 589 | goto out_free_tags; |
e467cde2 RR |
590 | } |
591 | ||
69740c8b | 592 | q->queuedata = vblk; |
7d116b62 | 593 | |
c0aa3e09 | 594 | virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN); |
d50ed907 | 595 | |
e467cde2 | 596 | vblk->disk->major = major; |
d50ed907 | 597 | vblk->disk->first_minor = index_to_minor(index); |
e467cde2 RR |
598 | vblk->disk->private_data = vblk; |
599 | vblk->disk->fops = &virtblk_fops; | |
c4839346 | 600 | vblk->disk->driverfs_dev = &vdev->dev; |
5087a50e | 601 | vblk->index = index; |
4f3bf19c | 602 | |
02c42b7a | 603 | /* configure queue flush support */ |
cd5d5038 | 604 | virtblk_update_cache_mode(vdev); |
e467cde2 | 605 | |
3ef53609 CB |
606 | /* If disk is read-only in the host, the guest should obey */ |
607 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) | |
608 | set_disk_ro(vblk->disk, 1); | |
609 | ||
a586d4f6 | 610 | /* Host must always specify the capacity. */ |
855e0c52 | 611 | virtio_cread(vdev, struct virtio_blk_config, capacity, &cap); |
e467cde2 RR |
612 | |
613 | /* If capacity is too big, truncate with warning. */ | |
614 | if ((sector_t)cap != cap) { | |
615 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
616 | (unsigned long long)cap); | |
617 | cap = (sector_t)-1; | |
618 | } | |
619 | set_capacity(vblk->disk, cap); | |
620 | ||
0864b79a | 621 | /* We can handle whatever the host told us to handle. */ |
ee714f2d | 622 | blk_queue_max_segments(q, vblk->sg_elems-2); |
0864b79a | 623 | |
4eff3cae | 624 | /* No need to bounce any requests */ |
69740c8b | 625 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
4eff3cae | 626 | |
4b7f7e20 | 627 | /* No real sector limit. */ |
ee714f2d | 628 | blk_queue_max_hw_sectors(q, -1U); |
4b7f7e20 | 629 | |
a586d4f6 RR |
630 | /* Host can optionally specify maximum segment size and number of |
631 | * segments. */ | |
855e0c52 RR |
632 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX, |
633 | struct virtio_blk_config, size_max, &v); | |
e467cde2 | 634 | if (!err) |
69740c8b | 635 | blk_queue_max_segment_size(q, v); |
4b7f7e20 | 636 | else |
69740c8b | 637 | blk_queue_max_segment_size(q, -1U); |
e467cde2 | 638 | |
066f4d82 | 639 | /* Host can optionally specify the block size of the device */ |
855e0c52 RR |
640 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE, |
641 | struct virtio_blk_config, blk_size, | |
642 | &blk_size); | |
066f4d82 | 643 | if (!err) |
69740c8b CH |
644 | blk_queue_logical_block_size(q, blk_size); |
645 | else | |
646 | blk_size = queue_logical_block_size(q); | |
647 | ||
648 | /* Use topology information if available */ | |
855e0c52 RR |
649 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
650 | struct virtio_blk_config, physical_block_exp, | |
651 | &physical_block_exp); | |
69740c8b CH |
652 | if (!err && physical_block_exp) |
653 | blk_queue_physical_block_size(q, | |
654 | blk_size * (1 << physical_block_exp)); | |
655 | ||
855e0c52 RR |
656 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
657 | struct virtio_blk_config, alignment_offset, | |
658 | &alignment_offset); | |
69740c8b CH |
659 | if (!err && alignment_offset) |
660 | blk_queue_alignment_offset(q, blk_size * alignment_offset); | |
661 | ||
855e0c52 RR |
662 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
663 | struct virtio_blk_config, min_io_size, | |
664 | &min_io_size); | |
69740c8b CH |
665 | if (!err && min_io_size) |
666 | blk_queue_io_min(q, blk_size * min_io_size); | |
667 | ||
855e0c52 RR |
668 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
669 | struct virtio_blk_config, opt_io_size, | |
670 | &opt_io_size); | |
69740c8b CH |
671 | if (!err && opt_io_size) |
672 | blk_queue_io_opt(q, blk_size * opt_io_size); | |
673 | ||
e467cde2 | 674 | add_disk(vblk->disk); |
a5eb9e4f RH |
675 | err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial); |
676 | if (err) | |
677 | goto out_del_disk; | |
678 | ||
cd5d5038 PB |
679 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) |
680 | err = device_create_file(disk_to_dev(vblk->disk), | |
681 | &dev_attr_cache_type_rw); | |
682 | else | |
683 | err = device_create_file(disk_to_dev(vblk->disk), | |
684 | &dev_attr_cache_type_ro); | |
685 | if (err) | |
686 | goto out_del_disk; | |
e467cde2 RR |
687 | return 0; |
688 | ||
a5eb9e4f RH |
689 | out_del_disk: |
690 | del_gendisk(vblk->disk); | |
691 | blk_cleanup_queue(vblk->disk->queue); | |
24d2f903 CH |
692 | out_free_tags: |
693 | blk_mq_free_tag_set(&vblk->tag_set); | |
e467cde2 RR |
694 | out_put_disk: |
695 | put_disk(vblk->disk); | |
e467cde2 | 696 | out_free_vq: |
d2a7ddda | 697 | vdev->config->del_vqs(vdev); |
e467cde2 RR |
698 | out_free_vblk: |
699 | kfree(vblk); | |
5087a50e MT |
700 | out_free_index: |
701 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
702 | out: |
703 | return err; | |
704 | } | |
705 | ||
8d85fce7 | 706 | static void virtblk_remove(struct virtio_device *vdev) |
e467cde2 RR |
707 | { |
708 | struct virtio_blk *vblk = vdev->priv; | |
5087a50e | 709 | int index = vblk->index; |
f4953fe6 | 710 | int refc; |
e467cde2 | 711 | |
4678d6f9 MT |
712 | /* Prevent config work handler from accessing the device. */ |
713 | mutex_lock(&vblk->config_lock); | |
714 | vblk->config_enable = false; | |
715 | mutex_unlock(&vblk->config_lock); | |
7a7c924c | 716 | |
02e2b124 | 717 | del_gendisk(vblk->disk); |
483001c7 | 718 | blk_cleanup_queue(vblk->disk->queue); |
02e2b124 | 719 | |
24d2f903 CH |
720 | blk_mq_free_tag_set(&vblk->tag_set); |
721 | ||
6e5aa7ef RR |
722 | /* Stop all the virtqueues. */ |
723 | vdev->config->reset(vdev); | |
724 | ||
4678d6f9 MT |
725 | flush_work(&vblk->config_work); |
726 | ||
f4953fe6 | 727 | refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount); |
e467cde2 | 728 | put_disk(vblk->disk); |
d2a7ddda | 729 | vdev->config->del_vqs(vdev); |
e467cde2 | 730 | kfree(vblk); |
f4953fe6 AG |
731 | |
732 | /* Only free device id if we don't have any users */ | |
733 | if (refc == 1) | |
734 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
735 | } |
736 | ||
89107000 | 737 | #ifdef CONFIG_PM_SLEEP |
f8fb5bc2 AS |
738 | static int virtblk_freeze(struct virtio_device *vdev) |
739 | { | |
740 | struct virtio_blk *vblk = vdev->priv; | |
741 | ||
742 | /* Ensure we don't receive any more interrupts */ | |
743 | vdev->config->reset(vdev); | |
744 | ||
745 | /* Prevent config work handler from accessing the device. */ | |
746 | mutex_lock(&vblk->config_lock); | |
747 | vblk->config_enable = false; | |
748 | mutex_unlock(&vblk->config_lock); | |
749 | ||
750 | flush_work(&vblk->config_work); | |
751 | ||
1cf7e9c6 | 752 | blk_mq_stop_hw_queues(vblk->disk->queue); |
f8fb5bc2 AS |
753 | |
754 | vdev->config->del_vqs(vdev); | |
755 | return 0; | |
756 | } | |
757 | ||
758 | static int virtblk_restore(struct virtio_device *vdev) | |
759 | { | |
760 | struct virtio_blk *vblk = vdev->priv; | |
761 | int ret; | |
762 | ||
763 | vblk->config_enable = true; | |
764 | ret = init_vq(vdev->priv); | |
1cf7e9c6 | 765 | if (!ret) |
1b4a3258 | 766 | blk_mq_start_stopped_hw_queues(vblk->disk->queue, true); |
1cf7e9c6 | 767 | |
f8fb5bc2 AS |
768 | return ret; |
769 | } | |
770 | #endif | |
771 | ||
47483e25 | 772 | static const struct virtio_device_id id_table[] = { |
e467cde2 RR |
773 | { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, |
774 | { 0 }, | |
775 | }; | |
776 | ||
c45a6816 | 777 | static unsigned int features[] = { |
02c42b7a TH |
778 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, |
779 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI, | |
cd5d5038 | 780 | VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE |
c45a6816 RR |
781 | }; |
782 | ||
8d85fce7 | 783 | static struct virtio_driver virtio_blk = { |
7a7c924c CH |
784 | .feature_table = features, |
785 | .feature_table_size = ARRAY_SIZE(features), | |
786 | .driver.name = KBUILD_MODNAME, | |
787 | .driver.owner = THIS_MODULE, | |
788 | .id_table = id_table, | |
789 | .probe = virtblk_probe, | |
8d85fce7 | 790 | .remove = virtblk_remove, |
7a7c924c | 791 | .config_changed = virtblk_config_changed, |
89107000 | 792 | #ifdef CONFIG_PM_SLEEP |
f8fb5bc2 AS |
793 | .freeze = virtblk_freeze, |
794 | .restore = virtblk_restore, | |
795 | #endif | |
e467cde2 RR |
796 | }; |
797 | ||
798 | static int __init init(void) | |
799 | { | |
7a7c924c CH |
800 | int error; |
801 | ||
802 | virtblk_wq = alloc_workqueue("virtio-blk", 0, 0); | |
803 | if (!virtblk_wq) | |
804 | return -ENOMEM; | |
805 | ||
4f3bf19c | 806 | major = register_blkdev(0, "virtblk"); |
7a7c924c CH |
807 | if (major < 0) { |
808 | error = major; | |
809 | goto out_destroy_workqueue; | |
810 | } | |
811 | ||
812 | error = register_virtio_driver(&virtio_blk); | |
813 | if (error) | |
814 | goto out_unregister_blkdev; | |
815 | return 0; | |
816 | ||
817 | out_unregister_blkdev: | |
818 | unregister_blkdev(major, "virtblk"); | |
819 | out_destroy_workqueue: | |
820 | destroy_workqueue(virtblk_wq); | |
821 | return error; | |
e467cde2 RR |
822 | } |
823 | ||
824 | static void __exit fini(void) | |
825 | { | |
4f3bf19c | 826 | unregister_blkdev(major, "virtblk"); |
e467cde2 | 827 | unregister_virtio_driver(&virtio_blk); |
7a7c924c | 828 | destroy_workqueue(virtblk_wq); |
e467cde2 RR |
829 | } |
830 | module_init(init); | |
831 | module_exit(fini); | |
832 | ||
833 | MODULE_DEVICE_TABLE(virtio, id_table); | |
834 | MODULE_DESCRIPTION("Virtio block driver"); | |
835 | MODULE_LICENSE("GPL"); |