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 | ||
5124c285 | 113 | static inline void virtblk_request_done(struct request *req) |
a98755c5 | 114 | { |
9d74e257 | 115 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); |
a98755c5 AH |
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) { | |
5124c285 | 141 | blk_mq_complete_request(vbr->req); |
1cf7e9c6 | 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 | 156 | struct virtio_blk *vblk = hctx->queue->queuedata; |
9d74e257 | 157 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); |
1cf7e9c6 | 158 | unsigned long flags; |
20af3cfd | 159 | unsigned int num; |
1cf7e9c6 | 160 | const bool last = (req->cmd_flags & REQ_END) != 0; |
5261b85e | 161 | int err; |
e467cde2 | 162 | |
1cf7e9c6 | 163 | BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); |
e467cde2 RR |
164 | |
165 | vbr->req = req; | |
dd40e456 FT |
166 | if (req->cmd_flags & REQ_FLUSH) { |
167 | vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH; | |
4cb2ea28 | 168 | vbr->out_hdr.sector = 0; |
169 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
dd40e456 FT |
170 | } else { |
171 | switch (req->cmd_type) { | |
172 | case REQ_TYPE_FS: | |
173 | vbr->out_hdr.type = 0; | |
174 | vbr->out_hdr.sector = blk_rq_pos(vbr->req); | |
175 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
176 | break; | |
177 | case REQ_TYPE_BLOCK_PC: | |
178 | vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; | |
f1b0ef06 CH |
179 | vbr->out_hdr.sector = 0; |
180 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
181 | break; | |
dd40e456 FT |
182 | case REQ_TYPE_SPECIAL: |
183 | vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID; | |
184 | vbr->out_hdr.sector = 0; | |
185 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
186 | break; | |
187 | default: | |
188 | /* We don't put anything else in the queue. */ | |
189 | BUG(); | |
f1b0ef06 | 190 | } |
e467cde2 RR |
191 | } |
192 | ||
1cf7e9c6 | 193 | num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg); |
1cde26f9 | 194 | if (num) { |
20af3cfd | 195 | if (rq_data_dir(vbr->req) == WRITE) |
1cde26f9 | 196 | vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; |
20af3cfd | 197 | else |
1cde26f9 | 198 | vbr->out_hdr.type |= VIRTIO_BLK_T_IN; |
e467cde2 RR |
199 | } |
200 | ||
1cf7e9c6 | 201 | spin_lock_irqsave(&vblk->vq_lock, flags); |
5261b85e RR |
202 | err = __virtblk_add_req(vblk->vq, vbr, vbr->sg, num); |
203 | if (err) { | |
f02b9ac3 | 204 | virtqueue_kick(vblk->vq); |
1cf7e9c6 JA |
205 | spin_unlock_irqrestore(&vblk->vq_lock, flags); |
206 | blk_mq_stop_hw_queue(hctx); | |
5261b85e RR |
207 | /* Out of mem doesn't actually happen, since we fall back |
208 | * to direct descriptors */ | |
209 | if (err == -ENOMEM || err == -ENOSPC) | |
210 | return BLK_MQ_RQ_QUEUE_BUSY; | |
211 | return BLK_MQ_RQ_QUEUE_ERROR; | |
a98755c5 AH |
212 | } |
213 | ||
1cf7e9c6 JA |
214 | if (last) |
215 | virtqueue_kick(vblk->vq); | |
f02b9ac3 SL |
216 | |
217 | spin_unlock_irqrestore(&vblk->vq_lock, flags); | |
1cf7e9c6 | 218 | return BLK_MQ_RQ_QUEUE_OK; |
a98755c5 AH |
219 | } |
220 | ||
4cb2ea28 | 221 | /* return id (s/n) string for *disk to *id_str |
222 | */ | |
223 | static int virtblk_get_id(struct gendisk *disk, char *id_str) | |
224 | { | |
225 | struct virtio_blk *vblk = disk->private_data; | |
226 | struct request *req; | |
227 | struct bio *bio; | |
e4c4776d | 228 | int err; |
4cb2ea28 | 229 | |
230 | bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES, | |
231 | GFP_KERNEL); | |
232 | if (IS_ERR(bio)) | |
233 | return PTR_ERR(bio); | |
234 | ||
235 | req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL); | |
236 | if (IS_ERR(req)) { | |
237 | bio_put(bio); | |
238 | return PTR_ERR(req); | |
239 | } | |
240 | ||
241 | req->cmd_type = REQ_TYPE_SPECIAL; | |
e4c4776d MS |
242 | err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); |
243 | blk_put_request(req); | |
244 | ||
245 | return err; | |
4cb2ea28 | 246 | } |
247 | ||
fe5a50a1 CH |
248 | static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, |
249 | unsigned int cmd, unsigned long data) | |
e467cde2 | 250 | { |
1cde26f9 HR |
251 | struct gendisk *disk = bdev->bd_disk; |
252 | struct virtio_blk *vblk = disk->private_data; | |
253 | ||
254 | /* | |
255 | * Only allow the generic SCSI ioctls if the host can support it. | |
256 | */ | |
257 | if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) | |
d9ecdea7 | 258 | return -ENOTTY; |
1cde26f9 | 259 | |
577ebb37 PB |
260 | return scsi_cmd_blk_ioctl(bdev, mode, cmd, |
261 | (void __user *)data); | |
e467cde2 RR |
262 | } |
263 | ||
135da0b0 CB |
264 | /* We provide getgeo only to please some old bootloader/partitioning tools */ |
265 | static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
266 | { | |
48e4043d | 267 | struct virtio_blk *vblk = bd->bd_disk->private_data; |
48e4043d RH |
268 | |
269 | /* see if the host passed in geometry config */ | |
855e0c52 RR |
270 | if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) { |
271 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
272 | geometry.cylinders, &geo->cylinders); | |
273 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
274 | geometry.heads, &geo->heads); | |
275 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
276 | geometry.sectors, &geo->sectors); | |
48e4043d RH |
277 | } else { |
278 | /* some standard values, similar to sd */ | |
279 | geo->heads = 1 << 6; | |
280 | geo->sectors = 1 << 5; | |
281 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
282 | } | |
135da0b0 CB |
283 | return 0; |
284 | } | |
285 | ||
83d5cde4 | 286 | static const struct block_device_operations virtblk_fops = { |
8a6cfeb6 | 287 | .ioctl = virtblk_ioctl, |
135da0b0 CB |
288 | .owner = THIS_MODULE, |
289 | .getgeo = virtblk_getgeo, | |
e467cde2 RR |
290 | }; |
291 | ||
d50ed907 CB |
292 | static int index_to_minor(int index) |
293 | { | |
294 | return index << PART_BITS; | |
295 | } | |
296 | ||
5087a50e MT |
297 | static int minor_to_index(int minor) |
298 | { | |
299 | return minor >> PART_BITS; | |
300 | } | |
301 | ||
a5eb9e4f RH |
302 | static ssize_t virtblk_serial_show(struct device *dev, |
303 | struct device_attribute *attr, char *buf) | |
304 | { | |
305 | struct gendisk *disk = dev_to_disk(dev); | |
306 | int err; | |
307 | ||
308 | /* sysfs gives us a PAGE_SIZE buffer */ | |
309 | BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); | |
310 | ||
311 | buf[VIRTIO_BLK_ID_BYTES] = '\0'; | |
312 | err = virtblk_get_id(disk, buf); | |
313 | if (!err) | |
314 | return strlen(buf); | |
315 | ||
316 | if (err == -EIO) /* Unsupported? Make it empty. */ | |
317 | return 0; | |
318 | ||
319 | return err; | |
320 | } | |
321 | DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL); | |
322 | ||
7a7c924c CH |
323 | static void virtblk_config_changed_work(struct work_struct *work) |
324 | { | |
325 | struct virtio_blk *vblk = | |
326 | container_of(work, struct virtio_blk, config_work); | |
327 | struct virtio_device *vdev = vblk->vdev; | |
328 | struct request_queue *q = vblk->disk->queue; | |
329 | char cap_str_2[10], cap_str_10[10]; | |
9d9598b8 | 330 | char *envp[] = { "RESIZE=1", NULL }; |
7a7c924c CH |
331 | u64 capacity, size; |
332 | ||
4678d6f9 MT |
333 | mutex_lock(&vblk->config_lock); |
334 | if (!vblk->config_enable) | |
335 | goto done; | |
336 | ||
7a7c924c | 337 | /* Host must always specify the capacity. */ |
855e0c52 | 338 | virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity); |
7a7c924c CH |
339 | |
340 | /* If capacity is too big, truncate with warning. */ | |
341 | if ((sector_t)capacity != capacity) { | |
342 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
343 | (unsigned long long)capacity); | |
344 | capacity = (sector_t)-1; | |
345 | } | |
346 | ||
347 | size = capacity * queue_logical_block_size(q); | |
348 | string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); | |
349 | string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); | |
350 | ||
351 | dev_notice(&vdev->dev, | |
352 | "new size: %llu %d-byte logical blocks (%s/%s)\n", | |
353 | (unsigned long long)capacity, | |
354 | queue_logical_block_size(q), | |
355 | cap_str_10, cap_str_2); | |
356 | ||
357 | set_capacity(vblk->disk, capacity); | |
e9986f30 | 358 | revalidate_disk(vblk->disk); |
9d9598b8 | 359 | kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp); |
4678d6f9 MT |
360 | done: |
361 | mutex_unlock(&vblk->config_lock); | |
7a7c924c CH |
362 | } |
363 | ||
364 | static void virtblk_config_changed(struct virtio_device *vdev) | |
365 | { | |
366 | struct virtio_blk *vblk = vdev->priv; | |
367 | ||
368 | queue_work(virtblk_wq, &vblk->config_work); | |
369 | } | |
370 | ||
6abd6e5a AS |
371 | static int init_vq(struct virtio_blk *vblk) |
372 | { | |
373 | int err = 0; | |
374 | ||
375 | /* We expect one virtqueue, for output. */ | |
a98755c5 | 376 | vblk->vq = virtio_find_single_vq(vblk->vdev, virtblk_done, "requests"); |
6abd6e5a AS |
377 | if (IS_ERR(vblk->vq)) |
378 | err = PTR_ERR(vblk->vq); | |
379 | ||
380 | return err; | |
381 | } | |
382 | ||
c0aa3e09 RM |
383 | /* |
384 | * Legacy naming scheme used for virtio devices. We are stuck with it for | |
385 | * virtio blk but don't ever use it for any new driver. | |
386 | */ | |
387 | static int virtblk_name_format(char *prefix, int index, char *buf, int buflen) | |
388 | { | |
389 | const int base = 'z' - 'a' + 1; | |
390 | char *begin = buf + strlen(prefix); | |
391 | char *end = buf + buflen; | |
392 | char *p; | |
393 | int unit; | |
394 | ||
395 | p = end - 1; | |
396 | *p = '\0'; | |
397 | unit = base; | |
398 | do { | |
399 | if (p == begin) | |
400 | return -EINVAL; | |
401 | *--p = 'a' + (index % unit); | |
402 | index = (index / unit) - 1; | |
403 | } while (index >= 0); | |
404 | ||
405 | memmove(begin, p, end - p); | |
406 | memcpy(buf, prefix, strlen(prefix)); | |
407 | ||
408 | return 0; | |
409 | } | |
410 | ||
cd5d5038 PB |
411 | static int virtblk_get_cache_mode(struct virtio_device *vdev) |
412 | { | |
413 | u8 writeback; | |
414 | int err; | |
415 | ||
855e0c52 RR |
416 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE, |
417 | struct virtio_blk_config, wce, | |
418 | &writeback); | |
cd5d5038 PB |
419 | if (err) |
420 | writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE); | |
421 | ||
422 | return writeback; | |
423 | } | |
424 | ||
425 | static void virtblk_update_cache_mode(struct virtio_device *vdev) | |
426 | { | |
427 | u8 writeback = virtblk_get_cache_mode(vdev); | |
428 | struct virtio_blk *vblk = vdev->priv; | |
429 | ||
c85a1f91 | 430 | if (writeback) |
cd5d5038 PB |
431 | blk_queue_flush(vblk->disk->queue, REQ_FLUSH); |
432 | else | |
433 | blk_queue_flush(vblk->disk->queue, 0); | |
434 | ||
435 | revalidate_disk(vblk->disk); | |
436 | } | |
437 | ||
438 | static const char *const virtblk_cache_types[] = { | |
439 | "write through", "write back" | |
440 | }; | |
441 | ||
442 | static ssize_t | |
443 | virtblk_cache_type_store(struct device *dev, struct device_attribute *attr, | |
444 | const char *buf, size_t count) | |
445 | { | |
446 | struct gendisk *disk = dev_to_disk(dev); | |
447 | struct virtio_blk *vblk = disk->private_data; | |
448 | struct virtio_device *vdev = vblk->vdev; | |
449 | int i; | |
cd5d5038 PB |
450 | |
451 | BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE)); | |
452 | for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; ) | |
453 | if (sysfs_streq(buf, virtblk_cache_types[i])) | |
454 | break; | |
455 | ||
456 | if (i < 0) | |
457 | return -EINVAL; | |
458 | ||
855e0c52 | 459 | virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i); |
cd5d5038 PB |
460 | virtblk_update_cache_mode(vdev); |
461 | return count; | |
462 | } | |
463 | ||
464 | static ssize_t | |
465 | virtblk_cache_type_show(struct device *dev, struct device_attribute *attr, | |
466 | char *buf) | |
467 | { | |
468 | struct gendisk *disk = dev_to_disk(dev); | |
469 | struct virtio_blk *vblk = disk->private_data; | |
470 | u8 writeback = virtblk_get_cache_mode(vblk->vdev); | |
471 | ||
472 | BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types)); | |
473 | return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]); | |
474 | } | |
475 | ||
476 | static const struct device_attribute dev_attr_cache_type_ro = | |
477 | __ATTR(cache_type, S_IRUGO, | |
478 | virtblk_cache_type_show, NULL); | |
479 | static const struct device_attribute dev_attr_cache_type_rw = | |
480 | __ATTR(cache_type, S_IRUGO|S_IWUSR, | |
481 | virtblk_cache_type_show, virtblk_cache_type_store); | |
482 | ||
e9b267d9 CH |
483 | static int virtblk_init_request(void *data, struct blk_mq_hw_ctx *hctx, |
484 | struct request *rq, unsigned int nr) | |
485 | { | |
486 | struct virtio_blk *vblk = data; | |
487 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq); | |
488 | ||
489 | sg_init_table(vbr->sg, vblk->sg_elems); | |
490 | return 0; | |
491 | } | |
492 | ||
1cf7e9c6 JA |
493 | static struct blk_mq_ops virtio_mq_ops = { |
494 | .queue_rq = virtio_queue_rq, | |
495 | .map_queue = blk_mq_map_queue, | |
496 | .alloc_hctx = blk_mq_alloc_single_hw_queue, | |
497 | .free_hctx = blk_mq_free_single_hw_queue, | |
e9b267d9 | 498 | .init_request = virtblk_init_request, |
5124c285 | 499 | .complete = virtblk_request_done, |
1cf7e9c6 JA |
500 | }; |
501 | ||
502 | static struct blk_mq_reg virtio_mq_reg = { | |
503 | .ops = &virtio_mq_ops, | |
504 | .nr_hw_queues = 1, | |
fc4324b4 | 505 | .queue_depth = 0, /* Set in virtblk_probe */ |
1cf7e9c6 JA |
506 | .numa_node = NUMA_NO_NODE, |
507 | .flags = BLK_MQ_F_SHOULD_MERGE, | |
508 | }; | |
fc4324b4 | 509 | module_param_named(queue_depth, virtio_mq_reg.queue_depth, uint, 0444); |
1cf7e9c6 | 510 | |
8d85fce7 | 511 | static int virtblk_probe(struct virtio_device *vdev) |
e467cde2 RR |
512 | { |
513 | struct virtio_blk *vblk; | |
69740c8b | 514 | struct request_queue *q; |
5087a50e | 515 | int err, index; |
a98755c5 | 516 | |
e467cde2 | 517 | u64 cap; |
69740c8b CH |
518 | u32 v, blk_size, sg_elems, opt_io_size; |
519 | u16 min_io_size; | |
520 | u8 physical_block_exp, alignment_offset; | |
e467cde2 | 521 | |
5087a50e MT |
522 | err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS), |
523 | GFP_KERNEL); | |
524 | if (err < 0) | |
525 | goto out; | |
526 | index = err; | |
4f3bf19c | 527 | |
0864b79a | 528 | /* We need to know how many segments before we allocate. */ |
855e0c52 RR |
529 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX, |
530 | struct virtio_blk_config, seg_max, | |
531 | &sg_elems); | |
a5b365a6 CH |
532 | |
533 | /* We need at least one SG element, whatever they say. */ | |
534 | if (err || !sg_elems) | |
0864b79a RR |
535 | sg_elems = 1; |
536 | ||
537 | /* We need an extra sg elements at head and tail. */ | |
538 | sg_elems += 2; | |
1cf7e9c6 | 539 | vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL); |
e467cde2 RR |
540 | if (!vblk) { |
541 | err = -ENOMEM; | |
5087a50e | 542 | goto out_free_index; |
e467cde2 RR |
543 | } |
544 | ||
e467cde2 | 545 | vblk->vdev = vdev; |
0864b79a | 546 | vblk->sg_elems = sg_elems; |
4678d6f9 | 547 | mutex_init(&vblk->config_lock); |
a98755c5 | 548 | |
7a7c924c | 549 | INIT_WORK(&vblk->config_work, virtblk_config_changed_work); |
4678d6f9 | 550 | vblk->config_enable = true; |
e467cde2 | 551 | |
6abd6e5a AS |
552 | err = init_vq(vblk); |
553 | if (err) | |
e467cde2 | 554 | goto out_free_vblk; |
1cf7e9c6 | 555 | spin_lock_init(&vblk->vq_lock); |
e467cde2 | 556 | |
e467cde2 | 557 | /* FIXME: How many partitions? How long is a piece of string? */ |
4f3bf19c | 558 | vblk->disk = alloc_disk(1 << PART_BITS); |
e467cde2 RR |
559 | if (!vblk->disk) { |
560 | err = -ENOMEM; | |
1cf7e9c6 | 561 | goto out_free_vq; |
e467cde2 RR |
562 | } |
563 | ||
fc4324b4 RR |
564 | /* Default queue sizing is to fill the ring. */ |
565 | if (!virtio_mq_reg.queue_depth) { | |
566 | virtio_mq_reg.queue_depth = vblk->vq->num_free; | |
567 | /* ... but without indirect descs, we use 2 descs per req */ | |
568 | if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC)) | |
569 | virtio_mq_reg.queue_depth /= 2; | |
570 | } | |
1cf7e9c6 JA |
571 | virtio_mq_reg.cmd_size = |
572 | sizeof(struct virtblk_req) + | |
573 | sizeof(struct scatterlist) * sg_elems; | |
574 | ||
575 | q = vblk->disk->queue = blk_mq_init_queue(&virtio_mq_reg, vblk); | |
69740c8b | 576 | if (!q) { |
e467cde2 RR |
577 | err = -ENOMEM; |
578 | goto out_put_disk; | |
579 | } | |
580 | ||
69740c8b | 581 | q->queuedata = vblk; |
7d116b62 | 582 | |
c0aa3e09 | 583 | virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN); |
d50ed907 | 584 | |
e467cde2 | 585 | vblk->disk->major = major; |
d50ed907 | 586 | vblk->disk->first_minor = index_to_minor(index); |
e467cde2 RR |
587 | vblk->disk->private_data = vblk; |
588 | vblk->disk->fops = &virtblk_fops; | |
c4839346 | 589 | vblk->disk->driverfs_dev = &vdev->dev; |
5087a50e | 590 | vblk->index = index; |
4f3bf19c | 591 | |
02c42b7a | 592 | /* configure queue flush support */ |
cd5d5038 | 593 | virtblk_update_cache_mode(vdev); |
e467cde2 | 594 | |
3ef53609 CB |
595 | /* If disk is read-only in the host, the guest should obey */ |
596 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) | |
597 | set_disk_ro(vblk->disk, 1); | |
598 | ||
a586d4f6 | 599 | /* Host must always specify the capacity. */ |
855e0c52 | 600 | virtio_cread(vdev, struct virtio_blk_config, capacity, &cap); |
e467cde2 RR |
601 | |
602 | /* If capacity is too big, truncate with warning. */ | |
603 | if ((sector_t)cap != cap) { | |
604 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
605 | (unsigned long long)cap); | |
606 | cap = (sector_t)-1; | |
607 | } | |
608 | set_capacity(vblk->disk, cap); | |
609 | ||
0864b79a | 610 | /* We can handle whatever the host told us to handle. */ |
ee714f2d | 611 | blk_queue_max_segments(q, vblk->sg_elems-2); |
0864b79a | 612 | |
4eff3cae | 613 | /* No need to bounce any requests */ |
69740c8b | 614 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
4eff3cae | 615 | |
4b7f7e20 | 616 | /* No real sector limit. */ |
ee714f2d | 617 | blk_queue_max_hw_sectors(q, -1U); |
4b7f7e20 | 618 | |
a586d4f6 RR |
619 | /* Host can optionally specify maximum segment size and number of |
620 | * segments. */ | |
855e0c52 RR |
621 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX, |
622 | struct virtio_blk_config, size_max, &v); | |
e467cde2 | 623 | if (!err) |
69740c8b | 624 | blk_queue_max_segment_size(q, v); |
4b7f7e20 | 625 | else |
69740c8b | 626 | blk_queue_max_segment_size(q, -1U); |
e467cde2 | 627 | |
066f4d82 | 628 | /* Host can optionally specify the block size of the device */ |
855e0c52 RR |
629 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE, |
630 | struct virtio_blk_config, blk_size, | |
631 | &blk_size); | |
066f4d82 | 632 | if (!err) |
69740c8b CH |
633 | blk_queue_logical_block_size(q, blk_size); |
634 | else | |
635 | blk_size = queue_logical_block_size(q); | |
636 | ||
637 | /* Use topology information if available */ | |
855e0c52 RR |
638 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
639 | struct virtio_blk_config, physical_block_exp, | |
640 | &physical_block_exp); | |
69740c8b CH |
641 | if (!err && physical_block_exp) |
642 | blk_queue_physical_block_size(q, | |
643 | blk_size * (1 << physical_block_exp)); | |
644 | ||
855e0c52 RR |
645 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
646 | struct virtio_blk_config, alignment_offset, | |
647 | &alignment_offset); | |
69740c8b CH |
648 | if (!err && alignment_offset) |
649 | blk_queue_alignment_offset(q, blk_size * alignment_offset); | |
650 | ||
855e0c52 RR |
651 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
652 | struct virtio_blk_config, min_io_size, | |
653 | &min_io_size); | |
69740c8b CH |
654 | if (!err && min_io_size) |
655 | blk_queue_io_min(q, blk_size * min_io_size); | |
656 | ||
855e0c52 RR |
657 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
658 | struct virtio_blk_config, opt_io_size, | |
659 | &opt_io_size); | |
69740c8b CH |
660 | if (!err && opt_io_size) |
661 | blk_queue_io_opt(q, blk_size * opt_io_size); | |
662 | ||
e467cde2 | 663 | add_disk(vblk->disk); |
a5eb9e4f RH |
664 | err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial); |
665 | if (err) | |
666 | goto out_del_disk; | |
667 | ||
cd5d5038 PB |
668 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) |
669 | err = device_create_file(disk_to_dev(vblk->disk), | |
670 | &dev_attr_cache_type_rw); | |
671 | else | |
672 | err = device_create_file(disk_to_dev(vblk->disk), | |
673 | &dev_attr_cache_type_ro); | |
674 | if (err) | |
675 | goto out_del_disk; | |
e467cde2 RR |
676 | return 0; |
677 | ||
a5eb9e4f RH |
678 | out_del_disk: |
679 | del_gendisk(vblk->disk); | |
680 | blk_cleanup_queue(vblk->disk->queue); | |
e467cde2 RR |
681 | out_put_disk: |
682 | put_disk(vblk->disk); | |
e467cde2 | 683 | out_free_vq: |
d2a7ddda | 684 | vdev->config->del_vqs(vdev); |
e467cde2 RR |
685 | out_free_vblk: |
686 | kfree(vblk); | |
5087a50e MT |
687 | out_free_index: |
688 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
689 | out: |
690 | return err; | |
691 | } | |
692 | ||
8d85fce7 | 693 | static void virtblk_remove(struct virtio_device *vdev) |
e467cde2 RR |
694 | { |
695 | struct virtio_blk *vblk = vdev->priv; | |
5087a50e | 696 | int index = vblk->index; |
f4953fe6 | 697 | int refc; |
e467cde2 | 698 | |
4678d6f9 MT |
699 | /* Prevent config work handler from accessing the device. */ |
700 | mutex_lock(&vblk->config_lock); | |
701 | vblk->config_enable = false; | |
702 | mutex_unlock(&vblk->config_lock); | |
7a7c924c | 703 | |
02e2b124 | 704 | del_gendisk(vblk->disk); |
483001c7 | 705 | blk_cleanup_queue(vblk->disk->queue); |
02e2b124 | 706 | |
6e5aa7ef RR |
707 | /* Stop all the virtqueues. */ |
708 | vdev->config->reset(vdev); | |
709 | ||
4678d6f9 MT |
710 | flush_work(&vblk->config_work); |
711 | ||
f4953fe6 | 712 | refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount); |
e467cde2 | 713 | put_disk(vblk->disk); |
d2a7ddda | 714 | vdev->config->del_vqs(vdev); |
e467cde2 | 715 | kfree(vblk); |
f4953fe6 AG |
716 | |
717 | /* Only free device id if we don't have any users */ | |
718 | if (refc == 1) | |
719 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
720 | } |
721 | ||
89107000 | 722 | #ifdef CONFIG_PM_SLEEP |
f8fb5bc2 AS |
723 | static int virtblk_freeze(struct virtio_device *vdev) |
724 | { | |
725 | struct virtio_blk *vblk = vdev->priv; | |
726 | ||
727 | /* Ensure we don't receive any more interrupts */ | |
728 | vdev->config->reset(vdev); | |
729 | ||
730 | /* Prevent config work handler from accessing the device. */ | |
731 | mutex_lock(&vblk->config_lock); | |
732 | vblk->config_enable = false; | |
733 | mutex_unlock(&vblk->config_lock); | |
734 | ||
735 | flush_work(&vblk->config_work); | |
736 | ||
1cf7e9c6 | 737 | blk_mq_stop_hw_queues(vblk->disk->queue); |
f8fb5bc2 AS |
738 | |
739 | vdev->config->del_vqs(vdev); | |
740 | return 0; | |
741 | } | |
742 | ||
743 | static int virtblk_restore(struct virtio_device *vdev) | |
744 | { | |
745 | struct virtio_blk *vblk = vdev->priv; | |
746 | int ret; | |
747 | ||
748 | vblk->config_enable = true; | |
749 | ret = init_vq(vdev->priv); | |
1cf7e9c6 JA |
750 | if (!ret) |
751 | blk_mq_start_stopped_hw_queues(vblk->disk->queue); | |
752 | ||
f8fb5bc2 AS |
753 | return ret; |
754 | } | |
755 | #endif | |
756 | ||
47483e25 | 757 | static const struct virtio_device_id id_table[] = { |
e467cde2 RR |
758 | { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, |
759 | { 0 }, | |
760 | }; | |
761 | ||
c45a6816 | 762 | static unsigned int features[] = { |
02c42b7a TH |
763 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, |
764 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI, | |
cd5d5038 | 765 | VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE |
c45a6816 RR |
766 | }; |
767 | ||
8d85fce7 | 768 | static struct virtio_driver virtio_blk = { |
7a7c924c CH |
769 | .feature_table = features, |
770 | .feature_table_size = ARRAY_SIZE(features), | |
771 | .driver.name = KBUILD_MODNAME, | |
772 | .driver.owner = THIS_MODULE, | |
773 | .id_table = id_table, | |
774 | .probe = virtblk_probe, | |
8d85fce7 | 775 | .remove = virtblk_remove, |
7a7c924c | 776 | .config_changed = virtblk_config_changed, |
89107000 | 777 | #ifdef CONFIG_PM_SLEEP |
f8fb5bc2 AS |
778 | .freeze = virtblk_freeze, |
779 | .restore = virtblk_restore, | |
780 | #endif | |
e467cde2 RR |
781 | }; |
782 | ||
783 | static int __init init(void) | |
784 | { | |
7a7c924c CH |
785 | int error; |
786 | ||
787 | virtblk_wq = alloc_workqueue("virtio-blk", 0, 0); | |
788 | if (!virtblk_wq) | |
789 | return -ENOMEM; | |
790 | ||
4f3bf19c | 791 | major = register_blkdev(0, "virtblk"); |
7a7c924c CH |
792 | if (major < 0) { |
793 | error = major; | |
794 | goto out_destroy_workqueue; | |
795 | } | |
796 | ||
797 | error = register_virtio_driver(&virtio_blk); | |
798 | if (error) | |
799 | goto out_unregister_blkdev; | |
800 | return 0; | |
801 | ||
802 | out_unregister_blkdev: | |
803 | unregister_blkdev(major, "virtblk"); | |
804 | out_destroy_workqueue: | |
805 | destroy_workqueue(virtblk_wq); | |
806 | return error; | |
e467cde2 RR |
807 | } |
808 | ||
809 | static void __exit fini(void) | |
810 | { | |
4f3bf19c | 811 | unregister_blkdev(major, "virtblk"); |
e467cde2 | 812 | unregister_virtio_driver(&virtio_blk); |
7a7c924c | 813 | destroy_workqueue(virtblk_wq); |
e467cde2 RR |
814 | } |
815 | module_init(init); | |
816 | module_exit(fini); | |
817 | ||
818 | MODULE_DEVICE_TABLE(virtio, id_table); | |
819 | MODULE_DESCRIPTION("Virtio block driver"); | |
820 | MODULE_LICENSE("GPL"); |