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 |
6a27b656 | 18 | #define VQ_NAME_LEN 16 |
e467cde2 | 19 | |
5087a50e MT |
20 | static int major; |
21 | static DEFINE_IDA(vd_index_ida); | |
22 | ||
2a647bfe | 23 | static struct workqueue_struct *virtblk_wq; |
4f3bf19c | 24 | |
6a27b656 ML |
25 | struct virtio_blk_vq { |
26 | struct virtqueue *vq; | |
27 | spinlock_t lock; | |
28 | char name[VQ_NAME_LEN]; | |
29 | } ____cacheline_aligned_in_smp; | |
30 | ||
e467cde2 RR |
31 | struct virtio_blk |
32 | { | |
e467cde2 | 33 | struct virtio_device *vdev; |
e467cde2 RR |
34 | |
35 | /* The disk structure for the kernel. */ | |
36 | struct gendisk *disk; | |
37 | ||
24d2f903 CH |
38 | /* Block layer tags. */ |
39 | struct blk_mq_tag_set tag_set; | |
40 | ||
7a7c924c CH |
41 | /* Process context for config space updates */ |
42 | struct work_struct config_work; | |
43 | ||
4678d6f9 MT |
44 | /* Lock for config space updates */ |
45 | struct mutex config_lock; | |
46 | ||
47 | /* enable config space updates */ | |
48 | bool config_enable; | |
49 | ||
0864b79a RR |
50 | /* What host tells us, plus 2 for header & tailer. */ |
51 | unsigned int sg_elems; | |
52 | ||
5087a50e MT |
53 | /* Ida index - used to track minor number allocations. */ |
54 | int index; | |
6a27b656 ML |
55 | |
56 | /* num of vqs */ | |
57 | int num_vqs; | |
58 | struct virtio_blk_vq *vqs; | |
e467cde2 RR |
59 | }; |
60 | ||
61 | struct virtblk_req | |
62 | { | |
e467cde2 RR |
63 | struct request *req; |
64 | struct virtio_blk_outhdr out_hdr; | |
1cde26f9 | 65 | struct virtio_scsi_inhdr in_hdr; |
cb38fa23 | 66 | u8 status; |
a98755c5 | 67 | struct scatterlist sg[]; |
e467cde2 RR |
68 | }; |
69 | ||
a98755c5 AH |
70 | static inline int virtblk_result(struct virtblk_req *vbr) |
71 | { | |
72 | switch (vbr->status) { | |
73 | case VIRTIO_BLK_S_OK: | |
74 | return 0; | |
75 | case VIRTIO_BLK_S_UNSUPP: | |
76 | return -ENOTTY; | |
77 | default: | |
78 | return -EIO; | |
79 | } | |
80 | } | |
81 | ||
8f39db9d | 82 | static int __virtblk_add_req(struct virtqueue *vq, |
20af3cfd PB |
83 | struct virtblk_req *vbr, |
84 | struct scatterlist *data_sg, | |
0a11cc36 | 85 | bool have_data) |
c85a1f91 | 86 | { |
20af3cfd | 87 | struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6]; |
8f39db9d | 88 | unsigned int num_out = 0, num_in = 0; |
20af3cfd | 89 | int type = vbr->out_hdr.type & ~VIRTIO_BLK_T_OUT; |
8f39db9d PB |
90 | |
91 | sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr)); | |
92 | sgs[num_out++] = &hdr; | |
93 | ||
20af3cfd PB |
94 | /* |
95 | * If this is a packet command we need a couple of additional headers. | |
96 | * Behind the normal outhdr we put a segment with the scsi command | |
97 | * block, and before the normal inhdr we put the sense data and the | |
98 | * inhdr with additional status information. | |
99 | */ | |
100 | if (type == VIRTIO_BLK_T_SCSI_CMD) { | |
101 | sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len); | |
102 | sgs[num_out++] = &cmd; | |
103 | } | |
104 | ||
0a11cc36 | 105 | if (have_data) { |
8f39db9d | 106 | if (vbr->out_hdr.type & VIRTIO_BLK_T_OUT) |
20af3cfd | 107 | sgs[num_out++] = data_sg; |
8f39db9d | 108 | else |
20af3cfd PB |
109 | sgs[num_out + num_in++] = data_sg; |
110 | } | |
111 | ||
112 | if (type == VIRTIO_BLK_T_SCSI_CMD) { | |
113 | sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE); | |
114 | sgs[num_out + num_in++] = &sense; | |
115 | sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr)); | |
116 | sgs[num_out + num_in++] = &inhdr; | |
8f39db9d PB |
117 | } |
118 | ||
119 | sg_init_one(&status, &vbr->status, sizeof(vbr->status)); | |
120 | sgs[num_out + num_in++] = &status; | |
121 | ||
122 | return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC); | |
5ee21a52 PB |
123 | } |
124 | ||
5124c285 | 125 | static inline void virtblk_request_done(struct request *req) |
a98755c5 | 126 | { |
9d74e257 | 127 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); |
a98755c5 AH |
128 | int error = virtblk_result(vbr); |
129 | ||
130 | if (req->cmd_type == REQ_TYPE_BLOCK_PC) { | |
131 | req->resid_len = vbr->in_hdr.residual; | |
132 | req->sense_len = vbr->in_hdr.sense_len; | |
133 | req->errors = vbr->in_hdr.errors; | |
134 | } else if (req->cmd_type == REQ_TYPE_SPECIAL) { | |
135 | req->errors = (error != 0); | |
136 | } | |
137 | ||
1cf7e9c6 | 138 | blk_mq_end_io(req, error); |
a98755c5 AH |
139 | } |
140 | ||
141 | static void virtblk_done(struct virtqueue *vq) | |
e467cde2 RR |
142 | { |
143 | struct virtio_blk *vblk = vq->vdev->priv; | |
1cf7e9c6 | 144 | bool req_done = false; |
6a27b656 | 145 | int qid = vq->index; |
e467cde2 | 146 | struct virtblk_req *vbr; |
e467cde2 | 147 | unsigned long flags; |
a98755c5 | 148 | unsigned int len; |
e467cde2 | 149 | |
6a27b656 | 150 | spin_lock_irqsave(&vblk->vqs[qid].lock, flags); |
bb811108 AH |
151 | do { |
152 | virtqueue_disable_cb(vq); | |
6a27b656 | 153 | while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) { |
5124c285 | 154 | blk_mq_complete_request(vbr->req); |
1cf7e9c6 | 155 | req_done = true; |
33659ebb | 156 | } |
7f03b17d HG |
157 | if (unlikely(virtqueue_is_broken(vq))) |
158 | break; | |
bb811108 | 159 | } while (!virtqueue_enable_cb(vq)); |
1cf7e9c6 | 160 | |
e467cde2 | 161 | /* In case queue is stopped waiting for more buffers. */ |
a98755c5 | 162 | if (req_done) |
1b4a3258 | 163 | blk_mq_start_stopped_hw_queues(vblk->disk->queue, true); |
6a27b656 | 164 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); |
a98755c5 AH |
165 | } |
166 | ||
bf572297 CH |
167 | static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req, |
168 | bool last) | |
e467cde2 | 169 | { |
1cf7e9c6 | 170 | struct virtio_blk *vblk = hctx->queue->queuedata; |
9d74e257 | 171 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); |
1cf7e9c6 | 172 | unsigned long flags; |
20af3cfd | 173 | unsigned int num; |
6a27b656 | 174 | int qid = hctx->queue_num; |
5261b85e | 175 | int err; |
e8edca6f | 176 | bool notify = false; |
e467cde2 | 177 | |
1cf7e9c6 | 178 | BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); |
e467cde2 RR |
179 | |
180 | vbr->req = req; | |
dd40e456 FT |
181 | if (req->cmd_flags & REQ_FLUSH) { |
182 | vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH; | |
4cb2ea28 | 183 | vbr->out_hdr.sector = 0; |
184 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
dd40e456 FT |
185 | } else { |
186 | switch (req->cmd_type) { | |
187 | case REQ_TYPE_FS: | |
188 | vbr->out_hdr.type = 0; | |
189 | vbr->out_hdr.sector = blk_rq_pos(vbr->req); | |
190 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
191 | break; | |
192 | case REQ_TYPE_BLOCK_PC: | |
193 | vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD; | |
f1b0ef06 CH |
194 | vbr->out_hdr.sector = 0; |
195 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
196 | break; | |
dd40e456 FT |
197 | case REQ_TYPE_SPECIAL: |
198 | vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID; | |
199 | vbr->out_hdr.sector = 0; | |
200 | vbr->out_hdr.ioprio = req_get_ioprio(vbr->req); | |
201 | break; | |
202 | default: | |
203 | /* We don't put anything else in the queue. */ | |
204 | BUG(); | |
f1b0ef06 | 205 | } |
e467cde2 RR |
206 | } |
207 | ||
e2490073 CH |
208 | blk_mq_start_request(req); |
209 | ||
1cf7e9c6 | 210 | num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg); |
1cde26f9 | 211 | if (num) { |
20af3cfd | 212 | if (rq_data_dir(vbr->req) == WRITE) |
1cde26f9 | 213 | vbr->out_hdr.type |= VIRTIO_BLK_T_OUT; |
20af3cfd | 214 | else |
1cde26f9 | 215 | vbr->out_hdr.type |= VIRTIO_BLK_T_IN; |
e467cde2 RR |
216 | } |
217 | ||
6a27b656 ML |
218 | spin_lock_irqsave(&vblk->vqs[qid].lock, flags); |
219 | err = __virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num); | |
5261b85e | 220 | if (err) { |
6a27b656 | 221 | virtqueue_kick(vblk->vqs[qid].vq); |
1cf7e9c6 | 222 | blk_mq_stop_hw_queue(hctx); |
6a27b656 | 223 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); |
5261b85e RR |
224 | /* Out of mem doesn't actually happen, since we fall back |
225 | * to direct descriptors */ | |
226 | if (err == -ENOMEM || err == -ENOSPC) | |
227 | return BLK_MQ_RQ_QUEUE_BUSY; | |
228 | return BLK_MQ_RQ_QUEUE_ERROR; | |
a98755c5 AH |
229 | } |
230 | ||
6a27b656 | 231 | if (last && virtqueue_kick_prepare(vblk->vqs[qid].vq)) |
e8edca6f | 232 | notify = true; |
6a27b656 | 233 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); |
e8edca6f ML |
234 | |
235 | if (notify) | |
6a27b656 | 236 | virtqueue_notify(vblk->vqs[qid].vq); |
1cf7e9c6 | 237 | return BLK_MQ_RQ_QUEUE_OK; |
a98755c5 AH |
238 | } |
239 | ||
4cb2ea28 | 240 | /* return id (s/n) string for *disk to *id_str |
241 | */ | |
242 | static int virtblk_get_id(struct gendisk *disk, char *id_str) | |
243 | { | |
244 | struct virtio_blk *vblk = disk->private_data; | |
245 | struct request *req; | |
246 | struct bio *bio; | |
e4c4776d | 247 | int err; |
4cb2ea28 | 248 | |
249 | bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES, | |
250 | GFP_KERNEL); | |
251 | if (IS_ERR(bio)) | |
252 | return PTR_ERR(bio); | |
253 | ||
254 | req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL); | |
255 | if (IS_ERR(req)) { | |
256 | bio_put(bio); | |
257 | return PTR_ERR(req); | |
258 | } | |
259 | ||
260 | req->cmd_type = REQ_TYPE_SPECIAL; | |
e4c4776d MS |
261 | err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); |
262 | blk_put_request(req); | |
263 | ||
264 | return err; | |
4cb2ea28 | 265 | } |
266 | ||
fe5a50a1 CH |
267 | static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, |
268 | unsigned int cmd, unsigned long data) | |
e467cde2 | 269 | { |
1cde26f9 HR |
270 | struct gendisk *disk = bdev->bd_disk; |
271 | struct virtio_blk *vblk = disk->private_data; | |
272 | ||
273 | /* | |
274 | * Only allow the generic SCSI ioctls if the host can support it. | |
275 | */ | |
276 | if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) | |
d9ecdea7 | 277 | return -ENOTTY; |
1cde26f9 | 278 | |
577ebb37 PB |
279 | return scsi_cmd_blk_ioctl(bdev, mode, cmd, |
280 | (void __user *)data); | |
e467cde2 RR |
281 | } |
282 | ||
135da0b0 CB |
283 | /* We provide getgeo only to please some old bootloader/partitioning tools */ |
284 | static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
285 | { | |
48e4043d | 286 | struct virtio_blk *vblk = bd->bd_disk->private_data; |
48e4043d RH |
287 | |
288 | /* see if the host passed in geometry config */ | |
855e0c52 RR |
289 | if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) { |
290 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
291 | geometry.cylinders, &geo->cylinders); | |
292 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
293 | geometry.heads, &geo->heads); | |
294 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
295 | geometry.sectors, &geo->sectors); | |
48e4043d RH |
296 | } else { |
297 | /* some standard values, similar to sd */ | |
298 | geo->heads = 1 << 6; | |
299 | geo->sectors = 1 << 5; | |
300 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
301 | } | |
135da0b0 CB |
302 | return 0; |
303 | } | |
304 | ||
83d5cde4 | 305 | static const struct block_device_operations virtblk_fops = { |
8a6cfeb6 | 306 | .ioctl = virtblk_ioctl, |
135da0b0 CB |
307 | .owner = THIS_MODULE, |
308 | .getgeo = virtblk_getgeo, | |
e467cde2 RR |
309 | }; |
310 | ||
d50ed907 CB |
311 | static int index_to_minor(int index) |
312 | { | |
313 | return index << PART_BITS; | |
314 | } | |
315 | ||
5087a50e MT |
316 | static int minor_to_index(int minor) |
317 | { | |
318 | return minor >> PART_BITS; | |
319 | } | |
320 | ||
a5eb9e4f RH |
321 | static ssize_t virtblk_serial_show(struct device *dev, |
322 | struct device_attribute *attr, char *buf) | |
323 | { | |
324 | struct gendisk *disk = dev_to_disk(dev); | |
325 | int err; | |
326 | ||
327 | /* sysfs gives us a PAGE_SIZE buffer */ | |
328 | BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); | |
329 | ||
330 | buf[VIRTIO_BLK_ID_BYTES] = '\0'; | |
331 | err = virtblk_get_id(disk, buf); | |
332 | if (!err) | |
333 | return strlen(buf); | |
334 | ||
335 | if (err == -EIO) /* Unsupported? Make it empty. */ | |
336 | return 0; | |
337 | ||
338 | return err; | |
339 | } | |
340 | DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL); | |
341 | ||
7a7c924c CH |
342 | static void virtblk_config_changed_work(struct work_struct *work) |
343 | { | |
344 | struct virtio_blk *vblk = | |
345 | container_of(work, struct virtio_blk, config_work); | |
346 | struct virtio_device *vdev = vblk->vdev; | |
347 | struct request_queue *q = vblk->disk->queue; | |
348 | char cap_str_2[10], cap_str_10[10]; | |
9d9598b8 | 349 | char *envp[] = { "RESIZE=1", NULL }; |
7a7c924c CH |
350 | u64 capacity, size; |
351 | ||
4678d6f9 MT |
352 | mutex_lock(&vblk->config_lock); |
353 | if (!vblk->config_enable) | |
354 | goto done; | |
355 | ||
7a7c924c | 356 | /* Host must always specify the capacity. */ |
855e0c52 | 357 | virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity); |
7a7c924c CH |
358 | |
359 | /* If capacity is too big, truncate with warning. */ | |
360 | if ((sector_t)capacity != capacity) { | |
361 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
362 | (unsigned long long)capacity); | |
363 | capacity = (sector_t)-1; | |
364 | } | |
365 | ||
366 | size = capacity * queue_logical_block_size(q); | |
367 | string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); | |
368 | string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); | |
369 | ||
370 | dev_notice(&vdev->dev, | |
371 | "new size: %llu %d-byte logical blocks (%s/%s)\n", | |
372 | (unsigned long long)capacity, | |
373 | queue_logical_block_size(q), | |
374 | cap_str_10, cap_str_2); | |
375 | ||
376 | set_capacity(vblk->disk, capacity); | |
e9986f30 | 377 | revalidate_disk(vblk->disk); |
9d9598b8 | 378 | kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp); |
4678d6f9 MT |
379 | done: |
380 | mutex_unlock(&vblk->config_lock); | |
7a7c924c CH |
381 | } |
382 | ||
383 | static void virtblk_config_changed(struct virtio_device *vdev) | |
384 | { | |
385 | struct virtio_blk *vblk = vdev->priv; | |
386 | ||
387 | queue_work(virtblk_wq, &vblk->config_work); | |
388 | } | |
389 | ||
6abd6e5a AS |
390 | static int init_vq(struct virtio_blk *vblk) |
391 | { | |
392 | int err = 0; | |
6a27b656 ML |
393 | int i; |
394 | vq_callback_t **callbacks; | |
395 | const char **names; | |
396 | struct virtqueue **vqs; | |
397 | unsigned short num_vqs; | |
398 | struct virtio_device *vdev = vblk->vdev; | |
399 | ||
400 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ, | |
401 | struct virtio_blk_config, num_queues, | |
402 | &num_vqs); | |
403 | if (err) | |
404 | num_vqs = 1; | |
405 | ||
406 | vblk->vqs = kmalloc(sizeof(*vblk->vqs) * num_vqs, GFP_KERNEL); | |
407 | if (!vblk->vqs) { | |
408 | err = -ENOMEM; | |
409 | goto out; | |
410 | } | |
411 | ||
412 | names = kmalloc(sizeof(*names) * num_vqs, GFP_KERNEL); | |
413 | if (!names) | |
414 | goto err_names; | |
415 | ||
416 | callbacks = kmalloc(sizeof(*callbacks) * num_vqs, GFP_KERNEL); | |
417 | if (!callbacks) | |
418 | goto err_callbacks; | |
419 | ||
420 | vqs = kmalloc(sizeof(*vqs) * num_vqs, GFP_KERNEL); | |
421 | if (!vqs) | |
422 | goto err_vqs; | |
6abd6e5a | 423 | |
6a27b656 ML |
424 | for (i = 0; i < num_vqs; i++) { |
425 | callbacks[i] = virtblk_done; | |
426 | snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i); | |
427 | names[i] = vblk->vqs[i].name; | |
428 | } | |
429 | ||
430 | /* Discover virtqueues and write information to configuration. */ | |
431 | err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names); | |
432 | if (err) | |
433 | goto err_find_vqs; | |
6abd6e5a | 434 | |
6a27b656 ML |
435 | for (i = 0; i < num_vqs; i++) { |
436 | spin_lock_init(&vblk->vqs[i].lock); | |
437 | vblk->vqs[i].vq = vqs[i]; | |
438 | } | |
439 | vblk->num_vqs = num_vqs; | |
440 | ||
441 | err_find_vqs: | |
442 | kfree(vqs); | |
443 | err_vqs: | |
444 | kfree(callbacks); | |
445 | err_callbacks: | |
446 | kfree(names); | |
447 | err_names: | |
448 | if (err) | |
449 | kfree(vblk->vqs); | |
450 | out: | |
6abd6e5a AS |
451 | return err; |
452 | } | |
453 | ||
c0aa3e09 RM |
454 | /* |
455 | * Legacy naming scheme used for virtio devices. We are stuck with it for | |
456 | * virtio blk but don't ever use it for any new driver. | |
457 | */ | |
458 | static int virtblk_name_format(char *prefix, int index, char *buf, int buflen) | |
459 | { | |
460 | const int base = 'z' - 'a' + 1; | |
461 | char *begin = buf + strlen(prefix); | |
462 | char *end = buf + buflen; | |
463 | char *p; | |
464 | int unit; | |
465 | ||
466 | p = end - 1; | |
467 | *p = '\0'; | |
468 | unit = base; | |
469 | do { | |
470 | if (p == begin) | |
471 | return -EINVAL; | |
472 | *--p = 'a' + (index % unit); | |
473 | index = (index / unit) - 1; | |
474 | } while (index >= 0); | |
475 | ||
476 | memmove(begin, p, end - p); | |
477 | memcpy(buf, prefix, strlen(prefix)); | |
478 | ||
479 | return 0; | |
480 | } | |
481 | ||
cd5d5038 PB |
482 | static int virtblk_get_cache_mode(struct virtio_device *vdev) |
483 | { | |
484 | u8 writeback; | |
485 | int err; | |
486 | ||
855e0c52 RR |
487 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE, |
488 | struct virtio_blk_config, wce, | |
489 | &writeback); | |
cd5d5038 PB |
490 | if (err) |
491 | writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE); | |
492 | ||
493 | return writeback; | |
494 | } | |
495 | ||
496 | static void virtblk_update_cache_mode(struct virtio_device *vdev) | |
497 | { | |
498 | u8 writeback = virtblk_get_cache_mode(vdev); | |
499 | struct virtio_blk *vblk = vdev->priv; | |
500 | ||
c85a1f91 | 501 | if (writeback) |
cd5d5038 PB |
502 | blk_queue_flush(vblk->disk->queue, REQ_FLUSH); |
503 | else | |
504 | blk_queue_flush(vblk->disk->queue, 0); | |
505 | ||
506 | revalidate_disk(vblk->disk); | |
507 | } | |
508 | ||
509 | static const char *const virtblk_cache_types[] = { | |
510 | "write through", "write back" | |
511 | }; | |
512 | ||
513 | static ssize_t | |
514 | virtblk_cache_type_store(struct device *dev, struct device_attribute *attr, | |
515 | const char *buf, size_t count) | |
516 | { | |
517 | struct gendisk *disk = dev_to_disk(dev); | |
518 | struct virtio_blk *vblk = disk->private_data; | |
519 | struct virtio_device *vdev = vblk->vdev; | |
520 | int i; | |
cd5d5038 PB |
521 | |
522 | BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE)); | |
523 | for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; ) | |
524 | if (sysfs_streq(buf, virtblk_cache_types[i])) | |
525 | break; | |
526 | ||
527 | if (i < 0) | |
528 | return -EINVAL; | |
529 | ||
855e0c52 | 530 | virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i); |
cd5d5038 PB |
531 | virtblk_update_cache_mode(vdev); |
532 | return count; | |
533 | } | |
534 | ||
535 | static ssize_t | |
536 | virtblk_cache_type_show(struct device *dev, struct device_attribute *attr, | |
537 | char *buf) | |
538 | { | |
539 | struct gendisk *disk = dev_to_disk(dev); | |
540 | struct virtio_blk *vblk = disk->private_data; | |
541 | u8 writeback = virtblk_get_cache_mode(vblk->vdev); | |
542 | ||
543 | BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types)); | |
544 | return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]); | |
545 | } | |
546 | ||
547 | static const struct device_attribute dev_attr_cache_type_ro = | |
548 | __ATTR(cache_type, S_IRUGO, | |
549 | virtblk_cache_type_show, NULL); | |
550 | static const struct device_attribute dev_attr_cache_type_rw = | |
551 | __ATTR(cache_type, S_IRUGO|S_IWUSR, | |
552 | virtblk_cache_type_show, virtblk_cache_type_store); | |
553 | ||
24d2f903 CH |
554 | static int virtblk_init_request(void *data, struct request *rq, |
555 | unsigned int hctx_idx, unsigned int request_idx, | |
556 | unsigned int numa_node) | |
e9b267d9 CH |
557 | { |
558 | struct virtio_blk *vblk = data; | |
559 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq); | |
560 | ||
561 | sg_init_table(vbr->sg, vblk->sg_elems); | |
562 | return 0; | |
563 | } | |
564 | ||
1cf7e9c6 JA |
565 | static struct blk_mq_ops virtio_mq_ops = { |
566 | .queue_rq = virtio_queue_rq, | |
567 | .map_queue = blk_mq_map_queue, | |
5124c285 | 568 | .complete = virtblk_request_done, |
24d2f903 | 569 | .init_request = virtblk_init_request, |
1cf7e9c6 JA |
570 | }; |
571 | ||
24d2f903 CH |
572 | static unsigned int virtblk_queue_depth; |
573 | module_param_named(queue_depth, virtblk_queue_depth, uint, 0444); | |
1cf7e9c6 | 574 | |
8d85fce7 | 575 | static int virtblk_probe(struct virtio_device *vdev) |
e467cde2 RR |
576 | { |
577 | struct virtio_blk *vblk; | |
69740c8b | 578 | struct request_queue *q; |
5087a50e | 579 | int err, index; |
a98755c5 | 580 | |
e467cde2 | 581 | u64 cap; |
69740c8b CH |
582 | u32 v, blk_size, sg_elems, opt_io_size; |
583 | u16 min_io_size; | |
584 | u8 physical_block_exp, alignment_offset; | |
e467cde2 | 585 | |
5087a50e MT |
586 | err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS), |
587 | GFP_KERNEL); | |
588 | if (err < 0) | |
589 | goto out; | |
590 | index = err; | |
4f3bf19c | 591 | |
0864b79a | 592 | /* We need to know how many segments before we allocate. */ |
855e0c52 RR |
593 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX, |
594 | struct virtio_blk_config, seg_max, | |
595 | &sg_elems); | |
a5b365a6 CH |
596 | |
597 | /* We need at least one SG element, whatever they say. */ | |
598 | if (err || !sg_elems) | |
0864b79a RR |
599 | sg_elems = 1; |
600 | ||
601 | /* We need an extra sg elements at head and tail. */ | |
602 | sg_elems += 2; | |
1cf7e9c6 | 603 | vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL); |
e467cde2 RR |
604 | if (!vblk) { |
605 | err = -ENOMEM; | |
5087a50e | 606 | goto out_free_index; |
e467cde2 RR |
607 | } |
608 | ||
e467cde2 | 609 | vblk->vdev = vdev; |
0864b79a | 610 | vblk->sg_elems = sg_elems; |
4678d6f9 | 611 | mutex_init(&vblk->config_lock); |
a98755c5 | 612 | |
7a7c924c | 613 | INIT_WORK(&vblk->config_work, virtblk_config_changed_work); |
4678d6f9 | 614 | vblk->config_enable = true; |
e467cde2 | 615 | |
6abd6e5a AS |
616 | err = init_vq(vblk); |
617 | if (err) | |
e467cde2 | 618 | goto out_free_vblk; |
e467cde2 | 619 | |
e467cde2 | 620 | /* FIXME: How many partitions? How long is a piece of string? */ |
4f3bf19c | 621 | vblk->disk = alloc_disk(1 << PART_BITS); |
e467cde2 RR |
622 | if (!vblk->disk) { |
623 | err = -ENOMEM; | |
1cf7e9c6 | 624 | goto out_free_vq; |
e467cde2 RR |
625 | } |
626 | ||
fc4324b4 | 627 | /* Default queue sizing is to fill the ring. */ |
24d2f903 | 628 | if (!virtblk_queue_depth) { |
6a27b656 | 629 | virtblk_queue_depth = vblk->vqs[0].vq->num_free; |
fc4324b4 RR |
630 | /* ... but without indirect descs, we use 2 descs per req */ |
631 | if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC)) | |
24d2f903 | 632 | virtblk_queue_depth /= 2; |
fc4324b4 | 633 | } |
24d2f903 CH |
634 | |
635 | memset(&vblk->tag_set, 0, sizeof(vblk->tag_set)); | |
636 | vblk->tag_set.ops = &virtio_mq_ops; | |
24d2f903 CH |
637 | vblk->tag_set.queue_depth = virtblk_queue_depth; |
638 | vblk->tag_set.numa_node = NUMA_NO_NODE; | |
639 | vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; | |
640 | vblk->tag_set.cmd_size = | |
1cf7e9c6 JA |
641 | sizeof(struct virtblk_req) + |
642 | sizeof(struct scatterlist) * sg_elems; | |
24d2f903 | 643 | vblk->tag_set.driver_data = vblk; |
6a27b656 | 644 | vblk->tag_set.nr_hw_queues = vblk->num_vqs; |
1cf7e9c6 | 645 | |
24d2f903 CH |
646 | err = blk_mq_alloc_tag_set(&vblk->tag_set); |
647 | if (err) | |
648 | goto out_put_disk; | |
649 | ||
650 | q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set); | |
69740c8b | 651 | if (!q) { |
e467cde2 | 652 | err = -ENOMEM; |
24d2f903 | 653 | goto out_free_tags; |
e467cde2 RR |
654 | } |
655 | ||
69740c8b | 656 | q->queuedata = vblk; |
7d116b62 | 657 | |
c0aa3e09 | 658 | virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN); |
d50ed907 | 659 | |
e467cde2 | 660 | vblk->disk->major = major; |
d50ed907 | 661 | vblk->disk->first_minor = index_to_minor(index); |
e467cde2 RR |
662 | vblk->disk->private_data = vblk; |
663 | vblk->disk->fops = &virtblk_fops; | |
c4839346 | 664 | vblk->disk->driverfs_dev = &vdev->dev; |
5087a50e | 665 | vblk->index = index; |
4f3bf19c | 666 | |
02c42b7a | 667 | /* configure queue flush support */ |
cd5d5038 | 668 | virtblk_update_cache_mode(vdev); |
e467cde2 | 669 | |
3ef53609 CB |
670 | /* If disk is read-only in the host, the guest should obey */ |
671 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) | |
672 | set_disk_ro(vblk->disk, 1); | |
673 | ||
a586d4f6 | 674 | /* Host must always specify the capacity. */ |
855e0c52 | 675 | virtio_cread(vdev, struct virtio_blk_config, capacity, &cap); |
e467cde2 RR |
676 | |
677 | /* If capacity is too big, truncate with warning. */ | |
678 | if ((sector_t)cap != cap) { | |
679 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
680 | (unsigned long long)cap); | |
681 | cap = (sector_t)-1; | |
682 | } | |
683 | set_capacity(vblk->disk, cap); | |
684 | ||
0864b79a | 685 | /* We can handle whatever the host told us to handle. */ |
ee714f2d | 686 | blk_queue_max_segments(q, vblk->sg_elems-2); |
0864b79a | 687 | |
4eff3cae | 688 | /* No need to bounce any requests */ |
69740c8b | 689 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
4eff3cae | 690 | |
4b7f7e20 | 691 | /* No real sector limit. */ |
ee714f2d | 692 | blk_queue_max_hw_sectors(q, -1U); |
4b7f7e20 | 693 | |
a586d4f6 RR |
694 | /* Host can optionally specify maximum segment size and number of |
695 | * segments. */ | |
855e0c52 RR |
696 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX, |
697 | struct virtio_blk_config, size_max, &v); | |
e467cde2 | 698 | if (!err) |
69740c8b | 699 | blk_queue_max_segment_size(q, v); |
4b7f7e20 | 700 | else |
69740c8b | 701 | blk_queue_max_segment_size(q, -1U); |
e467cde2 | 702 | |
066f4d82 | 703 | /* Host can optionally specify the block size of the device */ |
855e0c52 RR |
704 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE, |
705 | struct virtio_blk_config, blk_size, | |
706 | &blk_size); | |
066f4d82 | 707 | if (!err) |
69740c8b CH |
708 | blk_queue_logical_block_size(q, blk_size); |
709 | else | |
710 | blk_size = queue_logical_block_size(q); | |
711 | ||
712 | /* Use topology information if available */ | |
855e0c52 RR |
713 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
714 | struct virtio_blk_config, physical_block_exp, | |
715 | &physical_block_exp); | |
69740c8b CH |
716 | if (!err && physical_block_exp) |
717 | blk_queue_physical_block_size(q, | |
718 | blk_size * (1 << physical_block_exp)); | |
719 | ||
855e0c52 RR |
720 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
721 | struct virtio_blk_config, alignment_offset, | |
722 | &alignment_offset); | |
69740c8b CH |
723 | if (!err && alignment_offset) |
724 | blk_queue_alignment_offset(q, blk_size * alignment_offset); | |
725 | ||
855e0c52 RR |
726 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
727 | struct virtio_blk_config, min_io_size, | |
728 | &min_io_size); | |
69740c8b CH |
729 | if (!err && min_io_size) |
730 | blk_queue_io_min(q, blk_size * min_io_size); | |
731 | ||
855e0c52 RR |
732 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
733 | struct virtio_blk_config, opt_io_size, | |
734 | &opt_io_size); | |
69740c8b CH |
735 | if (!err && opt_io_size) |
736 | blk_queue_io_opt(q, blk_size * opt_io_size); | |
737 | ||
e467cde2 | 738 | add_disk(vblk->disk); |
a5eb9e4f RH |
739 | err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial); |
740 | if (err) | |
741 | goto out_del_disk; | |
742 | ||
cd5d5038 PB |
743 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) |
744 | err = device_create_file(disk_to_dev(vblk->disk), | |
745 | &dev_attr_cache_type_rw); | |
746 | else | |
747 | err = device_create_file(disk_to_dev(vblk->disk), | |
748 | &dev_attr_cache_type_ro); | |
749 | if (err) | |
750 | goto out_del_disk; | |
e467cde2 RR |
751 | return 0; |
752 | ||
a5eb9e4f RH |
753 | out_del_disk: |
754 | del_gendisk(vblk->disk); | |
755 | blk_cleanup_queue(vblk->disk->queue); | |
24d2f903 CH |
756 | out_free_tags: |
757 | blk_mq_free_tag_set(&vblk->tag_set); | |
e467cde2 RR |
758 | out_put_disk: |
759 | put_disk(vblk->disk); | |
e467cde2 | 760 | out_free_vq: |
d2a7ddda | 761 | vdev->config->del_vqs(vdev); |
e467cde2 RR |
762 | out_free_vblk: |
763 | kfree(vblk); | |
5087a50e MT |
764 | out_free_index: |
765 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
766 | out: |
767 | return err; | |
768 | } | |
769 | ||
8d85fce7 | 770 | static void virtblk_remove(struct virtio_device *vdev) |
e467cde2 RR |
771 | { |
772 | struct virtio_blk *vblk = vdev->priv; | |
5087a50e | 773 | int index = vblk->index; |
f4953fe6 | 774 | int refc; |
e467cde2 | 775 | |
4678d6f9 MT |
776 | /* Prevent config work handler from accessing the device. */ |
777 | mutex_lock(&vblk->config_lock); | |
778 | vblk->config_enable = false; | |
779 | mutex_unlock(&vblk->config_lock); | |
7a7c924c | 780 | |
02e2b124 | 781 | del_gendisk(vblk->disk); |
483001c7 | 782 | blk_cleanup_queue(vblk->disk->queue); |
02e2b124 | 783 | |
24d2f903 CH |
784 | blk_mq_free_tag_set(&vblk->tag_set); |
785 | ||
6e5aa7ef RR |
786 | /* Stop all the virtqueues. */ |
787 | vdev->config->reset(vdev); | |
788 | ||
4678d6f9 MT |
789 | flush_work(&vblk->config_work); |
790 | ||
f4953fe6 | 791 | refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount); |
e467cde2 | 792 | put_disk(vblk->disk); |
d2a7ddda | 793 | vdev->config->del_vqs(vdev); |
6a27b656 | 794 | kfree(vblk->vqs); |
e467cde2 | 795 | kfree(vblk); |
f4953fe6 AG |
796 | |
797 | /* Only free device id if we don't have any users */ | |
798 | if (refc == 1) | |
799 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
800 | } |
801 | ||
89107000 | 802 | #ifdef CONFIG_PM_SLEEP |
f8fb5bc2 AS |
803 | static int virtblk_freeze(struct virtio_device *vdev) |
804 | { | |
805 | struct virtio_blk *vblk = vdev->priv; | |
806 | ||
807 | /* Ensure we don't receive any more interrupts */ | |
808 | vdev->config->reset(vdev); | |
809 | ||
810 | /* Prevent config work handler from accessing the device. */ | |
811 | mutex_lock(&vblk->config_lock); | |
812 | vblk->config_enable = false; | |
813 | mutex_unlock(&vblk->config_lock); | |
814 | ||
815 | flush_work(&vblk->config_work); | |
816 | ||
1cf7e9c6 | 817 | blk_mq_stop_hw_queues(vblk->disk->queue); |
f8fb5bc2 AS |
818 | |
819 | vdev->config->del_vqs(vdev); | |
820 | return 0; | |
821 | } | |
822 | ||
823 | static int virtblk_restore(struct virtio_device *vdev) | |
824 | { | |
825 | struct virtio_blk *vblk = vdev->priv; | |
826 | int ret; | |
827 | ||
828 | vblk->config_enable = true; | |
829 | ret = init_vq(vdev->priv); | |
1cf7e9c6 | 830 | if (!ret) |
1b4a3258 | 831 | blk_mq_start_stopped_hw_queues(vblk->disk->queue, true); |
1cf7e9c6 | 832 | |
f8fb5bc2 AS |
833 | return ret; |
834 | } | |
835 | #endif | |
836 | ||
47483e25 | 837 | static const struct virtio_device_id id_table[] = { |
e467cde2 RR |
838 | { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, |
839 | { 0 }, | |
840 | }; | |
841 | ||
c45a6816 | 842 | static unsigned int features[] = { |
02c42b7a TH |
843 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, |
844 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI, | |
6a27b656 ML |
845 | VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE, |
846 | VIRTIO_BLK_F_MQ, | |
c45a6816 RR |
847 | }; |
848 | ||
8d85fce7 | 849 | static struct virtio_driver virtio_blk = { |
7a7c924c CH |
850 | .feature_table = features, |
851 | .feature_table_size = ARRAY_SIZE(features), | |
852 | .driver.name = KBUILD_MODNAME, | |
853 | .driver.owner = THIS_MODULE, | |
854 | .id_table = id_table, | |
855 | .probe = virtblk_probe, | |
8d85fce7 | 856 | .remove = virtblk_remove, |
7a7c924c | 857 | .config_changed = virtblk_config_changed, |
89107000 | 858 | #ifdef CONFIG_PM_SLEEP |
f8fb5bc2 AS |
859 | .freeze = virtblk_freeze, |
860 | .restore = virtblk_restore, | |
861 | #endif | |
e467cde2 RR |
862 | }; |
863 | ||
864 | static int __init init(void) | |
865 | { | |
7a7c924c CH |
866 | int error; |
867 | ||
868 | virtblk_wq = alloc_workqueue("virtio-blk", 0, 0); | |
869 | if (!virtblk_wq) | |
870 | return -ENOMEM; | |
871 | ||
4f3bf19c | 872 | major = register_blkdev(0, "virtblk"); |
7a7c924c CH |
873 | if (major < 0) { |
874 | error = major; | |
875 | goto out_destroy_workqueue; | |
876 | } | |
877 | ||
878 | error = register_virtio_driver(&virtio_blk); | |
879 | if (error) | |
880 | goto out_unregister_blkdev; | |
881 | return 0; | |
882 | ||
883 | out_unregister_blkdev: | |
884 | unregister_blkdev(major, "virtblk"); | |
885 | out_destroy_workqueue: | |
886 | destroy_workqueue(virtblk_wq); | |
887 | return error; | |
e467cde2 RR |
888 | } |
889 | ||
890 | static void __exit fini(void) | |
891 | { | |
4f3bf19c | 892 | unregister_blkdev(major, "virtblk"); |
e467cde2 | 893 | unregister_virtio_driver(&virtio_blk); |
7a7c924c | 894 | destroy_workqueue(virtblk_wq); |
e467cde2 RR |
895 | } |
896 | module_init(init); | |
897 | module_exit(fini); | |
898 | ||
899 | MODULE_DEVICE_TABLE(virtio, id_table); | |
900 | MODULE_DESCRIPTION("Virtio block driver"); | |
901 | MODULE_LICENSE("GPL"); |