]>
Commit | Line | Data |
---|---|---|
1 | //#define DEBUG | |
2 | #include <linux/spinlock.h> | |
3 | #include <linux/slab.h> | |
4 | #include <linux/blkdev.h> | |
5 | #include <linux/hdreg.h> | |
6 | #include <linux/module.h> | |
7 | #include <linux/mutex.h> | |
8 | #include <linux/virtio.h> | |
9 | #include <linux/virtio_blk.h> | |
10 | #include <linux/scatterlist.h> | |
11 | #include <linux/string_helpers.h> | |
12 | #include <scsi/scsi_cmnd.h> | |
13 | #include <linux/idr.h> | |
14 | #include <linux/blk-mq.h> | |
15 | #include <linux/numa.h> | |
16 | ||
17 | #define PART_BITS 4 | |
18 | #define VQ_NAME_LEN 16 | |
19 | ||
20 | static int major; | |
21 | static DEFINE_IDA(vd_index_ida); | |
22 | ||
23 | static struct workqueue_struct *virtblk_wq; | |
24 | ||
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 | ||
31 | struct virtio_blk { | |
32 | struct virtio_device *vdev; | |
33 | ||
34 | /* The disk structure for the kernel. */ | |
35 | struct gendisk *disk; | |
36 | ||
37 | /* Block layer tags. */ | |
38 | struct blk_mq_tag_set tag_set; | |
39 | ||
40 | /* Process context for config space updates */ | |
41 | struct work_struct config_work; | |
42 | ||
43 | /* What host tells us, plus 2 for header & tailer. */ | |
44 | unsigned int sg_elems; | |
45 | ||
46 | /* Ida index - used to track minor number allocations. */ | |
47 | int index; | |
48 | ||
49 | /* num of vqs */ | |
50 | int num_vqs; | |
51 | struct virtio_blk_vq *vqs; | |
52 | }; | |
53 | ||
54 | struct virtblk_req { | |
55 | struct request *req; | |
56 | struct virtio_blk_outhdr out_hdr; | |
57 | struct virtio_scsi_inhdr in_hdr; | |
58 | u8 status; | |
59 | struct scatterlist sg[]; | |
60 | }; | |
61 | ||
62 | static inline int virtblk_result(struct virtblk_req *vbr) | |
63 | { | |
64 | switch (vbr->status) { | |
65 | case VIRTIO_BLK_S_OK: | |
66 | return 0; | |
67 | case VIRTIO_BLK_S_UNSUPP: | |
68 | return -ENOTTY; | |
69 | default: | |
70 | return -EIO; | |
71 | } | |
72 | } | |
73 | ||
74 | static int __virtblk_add_req(struct virtqueue *vq, | |
75 | struct virtblk_req *vbr, | |
76 | struct scatterlist *data_sg, | |
77 | bool have_data) | |
78 | { | |
79 | struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6]; | |
80 | unsigned int num_out = 0, num_in = 0; | |
81 | __virtio32 type = vbr->out_hdr.type & ~cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT); | |
82 | ||
83 | sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr)); | |
84 | sgs[num_out++] = &hdr; | |
85 | ||
86 | /* | |
87 | * If this is a packet command we need a couple of additional headers. | |
88 | * Behind the normal outhdr we put a segment with the scsi command | |
89 | * block, and before the normal inhdr we put the sense data and the | |
90 | * inhdr with additional status information. | |
91 | */ | |
92 | if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) { | |
93 | sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len); | |
94 | sgs[num_out++] = &cmd; | |
95 | } | |
96 | ||
97 | if (have_data) { | |
98 | if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT)) | |
99 | sgs[num_out++] = data_sg; | |
100 | else | |
101 | sgs[num_out + num_in++] = data_sg; | |
102 | } | |
103 | ||
104 | if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) { | |
105 | sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE); | |
106 | sgs[num_out + num_in++] = &sense; | |
107 | sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr)); | |
108 | sgs[num_out + num_in++] = &inhdr; | |
109 | } | |
110 | ||
111 | sg_init_one(&status, &vbr->status, sizeof(vbr->status)); | |
112 | sgs[num_out + num_in++] = &status; | |
113 | ||
114 | return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC); | |
115 | } | |
116 | ||
117 | static inline void virtblk_request_done(struct request *req) | |
118 | { | |
119 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); | |
120 | struct virtio_blk *vblk = req->q->queuedata; | |
121 | int error = virtblk_result(vbr); | |
122 | ||
123 | if (req->cmd_type == REQ_TYPE_BLOCK_PC) { | |
124 | req->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual); | |
125 | req->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len); | |
126 | req->errors = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors); | |
127 | } else if (req->cmd_type == REQ_TYPE_DRV_PRIV) { | |
128 | req->errors = (error != 0); | |
129 | } | |
130 | ||
131 | blk_mq_end_request(req, error); | |
132 | } | |
133 | ||
134 | static void virtblk_done(struct virtqueue *vq) | |
135 | { | |
136 | struct virtio_blk *vblk = vq->vdev->priv; | |
137 | bool req_done = false; | |
138 | int qid = vq->index; | |
139 | struct virtblk_req *vbr; | |
140 | unsigned long flags; | |
141 | unsigned int len; | |
142 | ||
143 | spin_lock_irqsave(&vblk->vqs[qid].lock, flags); | |
144 | do { | |
145 | virtqueue_disable_cb(vq); | |
146 | while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) { | |
147 | blk_mq_complete_request(vbr->req, vbr->req->errors); | |
148 | req_done = true; | |
149 | } | |
150 | if (unlikely(virtqueue_is_broken(vq))) | |
151 | break; | |
152 | } while (!virtqueue_enable_cb(vq)); | |
153 | ||
154 | /* In case queue is stopped waiting for more buffers. */ | |
155 | if (req_done) | |
156 | blk_mq_start_stopped_hw_queues(vblk->disk->queue, true); | |
157 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); | |
158 | } | |
159 | ||
160 | static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, | |
161 | const struct blk_mq_queue_data *bd) | |
162 | { | |
163 | struct virtio_blk *vblk = hctx->queue->queuedata; | |
164 | struct request *req = bd->rq; | |
165 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); | |
166 | unsigned long flags; | |
167 | unsigned int num; | |
168 | int qid = hctx->queue_num; | |
169 | int err; | |
170 | bool notify = false; | |
171 | ||
172 | BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); | |
173 | ||
174 | vbr->req = req; | |
175 | if (req_op(req) == REQ_OP_FLUSH) { | |
176 | vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_FLUSH); | |
177 | vbr->out_hdr.sector = 0; | |
178 | vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req)); | |
179 | } else { | |
180 | switch (req->cmd_type) { | |
181 | case REQ_TYPE_FS: | |
182 | vbr->out_hdr.type = 0; | |
183 | vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, blk_rq_pos(vbr->req)); | |
184 | vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req)); | |
185 | break; | |
186 | case REQ_TYPE_BLOCK_PC: | |
187 | vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_SCSI_CMD); | |
188 | vbr->out_hdr.sector = 0; | |
189 | vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req)); | |
190 | break; | |
191 | case REQ_TYPE_DRV_PRIV: | |
192 | vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID); | |
193 | vbr->out_hdr.sector = 0; | |
194 | vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req)); | |
195 | break; | |
196 | default: | |
197 | /* We don't put anything else in the queue. */ | |
198 | BUG(); | |
199 | } | |
200 | } | |
201 | ||
202 | blk_mq_start_request(req); | |
203 | ||
204 | num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg); | |
205 | if (num) { | |
206 | if (rq_data_dir(vbr->req) == WRITE) | |
207 | vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT); | |
208 | else | |
209 | vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN); | |
210 | } | |
211 | ||
212 | spin_lock_irqsave(&vblk->vqs[qid].lock, flags); | |
213 | err = __virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num); | |
214 | if (err) { | |
215 | virtqueue_kick(vblk->vqs[qid].vq); | |
216 | blk_mq_stop_hw_queue(hctx); | |
217 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); | |
218 | /* Out of mem doesn't actually happen, since we fall back | |
219 | * to direct descriptors */ | |
220 | if (err == -ENOMEM || err == -ENOSPC) | |
221 | return BLK_MQ_RQ_QUEUE_BUSY; | |
222 | return BLK_MQ_RQ_QUEUE_ERROR; | |
223 | } | |
224 | ||
225 | if (bd->last && virtqueue_kick_prepare(vblk->vqs[qid].vq)) | |
226 | notify = true; | |
227 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); | |
228 | ||
229 | if (notify) | |
230 | virtqueue_notify(vblk->vqs[qid].vq); | |
231 | return BLK_MQ_RQ_QUEUE_OK; | |
232 | } | |
233 | ||
234 | /* return id (s/n) string for *disk to *id_str | |
235 | */ | |
236 | static int virtblk_get_id(struct gendisk *disk, char *id_str) | |
237 | { | |
238 | struct virtio_blk *vblk = disk->private_data; | |
239 | struct request_queue *q = vblk->disk->queue; | |
240 | struct request *req; | |
241 | int err; | |
242 | ||
243 | req = blk_get_request(q, READ, GFP_KERNEL); | |
244 | if (IS_ERR(req)) | |
245 | return PTR_ERR(req); | |
246 | req->cmd_type = REQ_TYPE_DRV_PRIV; | |
247 | ||
248 | err = blk_rq_map_kern(q, req, id_str, VIRTIO_BLK_ID_BYTES, GFP_KERNEL); | |
249 | if (err) | |
250 | goto out; | |
251 | ||
252 | err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); | |
253 | out: | |
254 | blk_put_request(req); | |
255 | return err; | |
256 | } | |
257 | ||
258 | static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, | |
259 | unsigned int cmd, unsigned long data) | |
260 | { | |
261 | struct gendisk *disk = bdev->bd_disk; | |
262 | struct virtio_blk *vblk = disk->private_data; | |
263 | ||
264 | /* | |
265 | * Only allow the generic SCSI ioctls if the host can support it. | |
266 | */ | |
267 | if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) | |
268 | return -ENOTTY; | |
269 | ||
270 | return scsi_cmd_blk_ioctl(bdev, mode, cmd, | |
271 | (void __user *)data); | |
272 | } | |
273 | ||
274 | /* We provide getgeo only to please some old bootloader/partitioning tools */ | |
275 | static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
276 | { | |
277 | struct virtio_blk *vblk = bd->bd_disk->private_data; | |
278 | ||
279 | /* see if the host passed in geometry config */ | |
280 | if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) { | |
281 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
282 | geometry.cylinders, &geo->cylinders); | |
283 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
284 | geometry.heads, &geo->heads); | |
285 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
286 | geometry.sectors, &geo->sectors); | |
287 | } else { | |
288 | /* some standard values, similar to sd */ | |
289 | geo->heads = 1 << 6; | |
290 | geo->sectors = 1 << 5; | |
291 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
292 | } | |
293 | return 0; | |
294 | } | |
295 | ||
296 | static const struct block_device_operations virtblk_fops = { | |
297 | .ioctl = virtblk_ioctl, | |
298 | .owner = THIS_MODULE, | |
299 | .getgeo = virtblk_getgeo, | |
300 | }; | |
301 | ||
302 | static int index_to_minor(int index) | |
303 | { | |
304 | return index << PART_BITS; | |
305 | } | |
306 | ||
307 | static int minor_to_index(int minor) | |
308 | { | |
309 | return minor >> PART_BITS; | |
310 | } | |
311 | ||
312 | static ssize_t virtblk_serial_show(struct device *dev, | |
313 | struct device_attribute *attr, char *buf) | |
314 | { | |
315 | struct gendisk *disk = dev_to_disk(dev); | |
316 | int err; | |
317 | ||
318 | /* sysfs gives us a PAGE_SIZE buffer */ | |
319 | BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); | |
320 | ||
321 | buf[VIRTIO_BLK_ID_BYTES] = '\0'; | |
322 | err = virtblk_get_id(disk, buf); | |
323 | if (!err) | |
324 | return strlen(buf); | |
325 | ||
326 | if (err == -EIO) /* Unsupported? Make it empty. */ | |
327 | return 0; | |
328 | ||
329 | return err; | |
330 | } | |
331 | ||
332 | static DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL); | |
333 | ||
334 | static void virtblk_config_changed_work(struct work_struct *work) | |
335 | { | |
336 | struct virtio_blk *vblk = | |
337 | container_of(work, struct virtio_blk, config_work); | |
338 | struct virtio_device *vdev = vblk->vdev; | |
339 | struct request_queue *q = vblk->disk->queue; | |
340 | char cap_str_2[10], cap_str_10[10]; | |
341 | char *envp[] = { "RESIZE=1", NULL }; | |
342 | u64 capacity; | |
343 | ||
344 | /* Host must always specify the capacity. */ | |
345 | virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity); | |
346 | ||
347 | /* If capacity is too big, truncate with warning. */ | |
348 | if ((sector_t)capacity != capacity) { | |
349 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
350 | (unsigned long long)capacity); | |
351 | capacity = (sector_t)-1; | |
352 | } | |
353 | ||
354 | string_get_size(capacity, queue_logical_block_size(q), | |
355 | STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); | |
356 | string_get_size(capacity, queue_logical_block_size(q), | |
357 | STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); | |
358 | ||
359 | dev_notice(&vdev->dev, | |
360 | "new size: %llu %d-byte logical blocks (%s/%s)\n", | |
361 | (unsigned long long)capacity, | |
362 | queue_logical_block_size(q), | |
363 | cap_str_10, cap_str_2); | |
364 | ||
365 | set_capacity(vblk->disk, capacity); | |
366 | revalidate_disk(vblk->disk); | |
367 | kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp); | |
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 | ||
377 | static int init_vq(struct virtio_blk *vblk) | |
378 | { | |
379 | int err; | |
380 | int i; | |
381 | vq_callback_t **callbacks; | |
382 | const char **names; | |
383 | struct virtqueue **vqs; | |
384 | unsigned short num_vqs; | |
385 | struct virtio_device *vdev = vblk->vdev; | |
386 | ||
387 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ, | |
388 | struct virtio_blk_config, num_queues, | |
389 | &num_vqs); | |
390 | if (err) | |
391 | num_vqs = 1; | |
392 | ||
393 | vblk->vqs = kmalloc_array(num_vqs, sizeof(*vblk->vqs), GFP_KERNEL); | |
394 | if (!vblk->vqs) | |
395 | return -ENOMEM; | |
396 | ||
397 | names = kmalloc_array(num_vqs, sizeof(*names), GFP_KERNEL); | |
398 | callbacks = kmalloc_array(num_vqs, sizeof(*callbacks), GFP_KERNEL); | |
399 | vqs = kmalloc_array(num_vqs, sizeof(*vqs), GFP_KERNEL); | |
400 | if (!names || !callbacks || !vqs) { | |
401 | err = -ENOMEM; | |
402 | goto out; | |
403 | } | |
404 | ||
405 | for (i = 0; i < num_vqs; i++) { | |
406 | callbacks[i] = virtblk_done; | |
407 | snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i); | |
408 | names[i] = vblk->vqs[i].name; | |
409 | } | |
410 | ||
411 | /* Discover virtqueues and write information to configuration. */ | |
412 | err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names); | |
413 | if (err) | |
414 | goto out; | |
415 | ||
416 | for (i = 0; i < num_vqs; i++) { | |
417 | spin_lock_init(&vblk->vqs[i].lock); | |
418 | vblk->vqs[i].vq = vqs[i]; | |
419 | } | |
420 | vblk->num_vqs = num_vqs; | |
421 | ||
422 | out: | |
423 | kfree(vqs); | |
424 | kfree(callbacks); | |
425 | kfree(names); | |
426 | if (err) | |
427 | kfree(vblk->vqs); | |
428 | return err; | |
429 | } | |
430 | ||
431 | /* | |
432 | * Legacy naming scheme used for virtio devices. We are stuck with it for | |
433 | * virtio blk but don't ever use it for any new driver. | |
434 | */ | |
435 | static int virtblk_name_format(char *prefix, int index, char *buf, int buflen) | |
436 | { | |
437 | const int base = 'z' - 'a' + 1; | |
438 | char *begin = buf + strlen(prefix); | |
439 | char *end = buf + buflen; | |
440 | char *p; | |
441 | int unit; | |
442 | ||
443 | p = end - 1; | |
444 | *p = '\0'; | |
445 | unit = base; | |
446 | do { | |
447 | if (p == begin) | |
448 | return -EINVAL; | |
449 | *--p = 'a' + (index % unit); | |
450 | index = (index / unit) - 1; | |
451 | } while (index >= 0); | |
452 | ||
453 | memmove(begin, p, end - p); | |
454 | memcpy(buf, prefix, strlen(prefix)); | |
455 | ||
456 | return 0; | |
457 | } | |
458 | ||
459 | static int virtblk_get_cache_mode(struct virtio_device *vdev) | |
460 | { | |
461 | u8 writeback; | |
462 | int err; | |
463 | ||
464 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE, | |
465 | struct virtio_blk_config, wce, | |
466 | &writeback); | |
467 | ||
468 | /* | |
469 | * If WCE is not configurable and flush is not available, | |
470 | * assume no writeback cache is in use. | |
471 | */ | |
472 | if (err) | |
473 | writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH); | |
474 | ||
475 | return writeback; | |
476 | } | |
477 | ||
478 | static void virtblk_update_cache_mode(struct virtio_device *vdev) | |
479 | { | |
480 | u8 writeback = virtblk_get_cache_mode(vdev); | |
481 | struct virtio_blk *vblk = vdev->priv; | |
482 | ||
483 | blk_queue_write_cache(vblk->disk->queue, writeback, false); | |
484 | revalidate_disk(vblk->disk); | |
485 | } | |
486 | ||
487 | static const char *const virtblk_cache_types[] = { | |
488 | "write through", "write back" | |
489 | }; | |
490 | ||
491 | static ssize_t | |
492 | virtblk_cache_type_store(struct device *dev, struct device_attribute *attr, | |
493 | const char *buf, size_t count) | |
494 | { | |
495 | struct gendisk *disk = dev_to_disk(dev); | |
496 | struct virtio_blk *vblk = disk->private_data; | |
497 | struct virtio_device *vdev = vblk->vdev; | |
498 | int i; | |
499 | ||
500 | BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE)); | |
501 | for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; ) | |
502 | if (sysfs_streq(buf, virtblk_cache_types[i])) | |
503 | break; | |
504 | ||
505 | if (i < 0) | |
506 | return -EINVAL; | |
507 | ||
508 | virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i); | |
509 | virtblk_update_cache_mode(vdev); | |
510 | return count; | |
511 | } | |
512 | ||
513 | static ssize_t | |
514 | virtblk_cache_type_show(struct device *dev, struct device_attribute *attr, | |
515 | char *buf) | |
516 | { | |
517 | struct gendisk *disk = dev_to_disk(dev); | |
518 | struct virtio_blk *vblk = disk->private_data; | |
519 | u8 writeback = virtblk_get_cache_mode(vblk->vdev); | |
520 | ||
521 | BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types)); | |
522 | return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]); | |
523 | } | |
524 | ||
525 | static const struct device_attribute dev_attr_cache_type_ro = | |
526 | __ATTR(cache_type, S_IRUGO, | |
527 | virtblk_cache_type_show, NULL); | |
528 | static const struct device_attribute dev_attr_cache_type_rw = | |
529 | __ATTR(cache_type, S_IRUGO|S_IWUSR, | |
530 | virtblk_cache_type_show, virtblk_cache_type_store); | |
531 | ||
532 | static int virtblk_init_request(void *data, struct request *rq, | |
533 | unsigned int hctx_idx, unsigned int request_idx, | |
534 | unsigned int numa_node) | |
535 | { | |
536 | struct virtio_blk *vblk = data; | |
537 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq); | |
538 | ||
539 | sg_init_table(vbr->sg, vblk->sg_elems); | |
540 | return 0; | |
541 | } | |
542 | ||
543 | static struct blk_mq_ops virtio_mq_ops = { | |
544 | .queue_rq = virtio_queue_rq, | |
545 | .complete = virtblk_request_done, | |
546 | .init_request = virtblk_init_request, | |
547 | }; | |
548 | ||
549 | static unsigned int virtblk_queue_depth; | |
550 | module_param_named(queue_depth, virtblk_queue_depth, uint, 0444); | |
551 | ||
552 | static int virtblk_probe(struct virtio_device *vdev) | |
553 | { | |
554 | struct virtio_blk *vblk; | |
555 | struct request_queue *q; | |
556 | int err, index; | |
557 | ||
558 | u64 cap; | |
559 | u32 v, blk_size, sg_elems, opt_io_size; | |
560 | u16 min_io_size; | |
561 | u8 physical_block_exp, alignment_offset; | |
562 | ||
563 | if (!vdev->config->get) { | |
564 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
565 | __func__); | |
566 | return -EINVAL; | |
567 | } | |
568 | ||
569 | err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS), | |
570 | GFP_KERNEL); | |
571 | if (err < 0) | |
572 | goto out; | |
573 | index = err; | |
574 | ||
575 | /* We need to know how many segments before we allocate. */ | |
576 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX, | |
577 | struct virtio_blk_config, seg_max, | |
578 | &sg_elems); | |
579 | ||
580 | /* We need at least one SG element, whatever they say. */ | |
581 | if (err || !sg_elems) | |
582 | sg_elems = 1; | |
583 | ||
584 | /* We need an extra sg elements at head and tail. */ | |
585 | sg_elems += 2; | |
586 | vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL); | |
587 | if (!vblk) { | |
588 | err = -ENOMEM; | |
589 | goto out_free_index; | |
590 | } | |
591 | ||
592 | vblk->vdev = vdev; | |
593 | vblk->sg_elems = sg_elems; | |
594 | ||
595 | INIT_WORK(&vblk->config_work, virtblk_config_changed_work); | |
596 | ||
597 | err = init_vq(vblk); | |
598 | if (err) | |
599 | goto out_free_vblk; | |
600 | ||
601 | /* FIXME: How many partitions? How long is a piece of string? */ | |
602 | vblk->disk = alloc_disk(1 << PART_BITS); | |
603 | if (!vblk->disk) { | |
604 | err = -ENOMEM; | |
605 | goto out_free_vq; | |
606 | } | |
607 | ||
608 | /* Default queue sizing is to fill the ring. */ | |
609 | if (!virtblk_queue_depth) { | |
610 | virtblk_queue_depth = vblk->vqs[0].vq->num_free; | |
611 | /* ... but without indirect descs, we use 2 descs per req */ | |
612 | if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC)) | |
613 | virtblk_queue_depth /= 2; | |
614 | } | |
615 | ||
616 | memset(&vblk->tag_set, 0, sizeof(vblk->tag_set)); | |
617 | vblk->tag_set.ops = &virtio_mq_ops; | |
618 | vblk->tag_set.queue_depth = virtblk_queue_depth; | |
619 | vblk->tag_set.numa_node = NUMA_NO_NODE; | |
620 | vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; | |
621 | vblk->tag_set.cmd_size = | |
622 | sizeof(struct virtblk_req) + | |
623 | sizeof(struct scatterlist) * sg_elems; | |
624 | vblk->tag_set.driver_data = vblk; | |
625 | vblk->tag_set.nr_hw_queues = vblk->num_vqs; | |
626 | ||
627 | err = blk_mq_alloc_tag_set(&vblk->tag_set); | |
628 | if (err) | |
629 | goto out_put_disk; | |
630 | ||
631 | q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set); | |
632 | if (IS_ERR(q)) { | |
633 | err = -ENOMEM; | |
634 | goto out_free_tags; | |
635 | } | |
636 | ||
637 | q->queuedata = vblk; | |
638 | ||
639 | virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN); | |
640 | ||
641 | vblk->disk->major = major; | |
642 | vblk->disk->first_minor = index_to_minor(index); | |
643 | vblk->disk->private_data = vblk; | |
644 | vblk->disk->fops = &virtblk_fops; | |
645 | vblk->disk->flags |= GENHD_FL_EXT_DEVT; | |
646 | vblk->index = index; | |
647 | ||
648 | /* configure queue flush support */ | |
649 | virtblk_update_cache_mode(vdev); | |
650 | ||
651 | /* If disk is read-only in the host, the guest should obey */ | |
652 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) | |
653 | set_disk_ro(vblk->disk, 1); | |
654 | ||
655 | /* Host must always specify the capacity. */ | |
656 | virtio_cread(vdev, struct virtio_blk_config, capacity, &cap); | |
657 | ||
658 | /* If capacity is too big, truncate with warning. */ | |
659 | if ((sector_t)cap != cap) { | |
660 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
661 | (unsigned long long)cap); | |
662 | cap = (sector_t)-1; | |
663 | } | |
664 | set_capacity(vblk->disk, cap); | |
665 | ||
666 | /* We can handle whatever the host told us to handle. */ | |
667 | blk_queue_max_segments(q, vblk->sg_elems-2); | |
668 | ||
669 | /* No need to bounce any requests */ | |
670 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); | |
671 | ||
672 | /* No real sector limit. */ | |
673 | blk_queue_max_hw_sectors(q, -1U); | |
674 | ||
675 | /* Host can optionally specify maximum segment size and number of | |
676 | * segments. */ | |
677 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX, | |
678 | struct virtio_blk_config, size_max, &v); | |
679 | if (!err) | |
680 | blk_queue_max_segment_size(q, v); | |
681 | else | |
682 | blk_queue_max_segment_size(q, -1U); | |
683 | ||
684 | /* Host can optionally specify the block size of the device */ | |
685 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE, | |
686 | struct virtio_blk_config, blk_size, | |
687 | &blk_size); | |
688 | if (!err) | |
689 | blk_queue_logical_block_size(q, blk_size); | |
690 | else | |
691 | blk_size = queue_logical_block_size(q); | |
692 | ||
693 | /* Use topology information if available */ | |
694 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, | |
695 | struct virtio_blk_config, physical_block_exp, | |
696 | &physical_block_exp); | |
697 | if (!err && physical_block_exp) | |
698 | blk_queue_physical_block_size(q, | |
699 | blk_size * (1 << physical_block_exp)); | |
700 | ||
701 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, | |
702 | struct virtio_blk_config, alignment_offset, | |
703 | &alignment_offset); | |
704 | if (!err && alignment_offset) | |
705 | blk_queue_alignment_offset(q, blk_size * alignment_offset); | |
706 | ||
707 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, | |
708 | struct virtio_blk_config, min_io_size, | |
709 | &min_io_size); | |
710 | if (!err && min_io_size) | |
711 | blk_queue_io_min(q, blk_size * min_io_size); | |
712 | ||
713 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, | |
714 | struct virtio_blk_config, opt_io_size, | |
715 | &opt_io_size); | |
716 | if (!err && opt_io_size) | |
717 | blk_queue_io_opt(q, blk_size * opt_io_size); | |
718 | ||
719 | virtio_device_ready(vdev); | |
720 | ||
721 | device_add_disk(&vdev->dev, vblk->disk); | |
722 | err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial); | |
723 | if (err) | |
724 | goto out_del_disk; | |
725 | ||
726 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) | |
727 | err = device_create_file(disk_to_dev(vblk->disk), | |
728 | &dev_attr_cache_type_rw); | |
729 | else | |
730 | err = device_create_file(disk_to_dev(vblk->disk), | |
731 | &dev_attr_cache_type_ro); | |
732 | if (err) | |
733 | goto out_del_disk; | |
734 | return 0; | |
735 | ||
736 | out_del_disk: | |
737 | del_gendisk(vblk->disk); | |
738 | blk_cleanup_queue(vblk->disk->queue); | |
739 | out_free_tags: | |
740 | blk_mq_free_tag_set(&vblk->tag_set); | |
741 | out_put_disk: | |
742 | put_disk(vblk->disk); | |
743 | out_free_vq: | |
744 | vdev->config->del_vqs(vdev); | |
745 | out_free_vblk: | |
746 | kfree(vblk); | |
747 | out_free_index: | |
748 | ida_simple_remove(&vd_index_ida, index); | |
749 | out: | |
750 | return err; | |
751 | } | |
752 | ||
753 | static void virtblk_remove(struct virtio_device *vdev) | |
754 | { | |
755 | struct virtio_blk *vblk = vdev->priv; | |
756 | int index = vblk->index; | |
757 | int refc; | |
758 | ||
759 | /* Make sure no work handler is accessing the device. */ | |
760 | flush_work(&vblk->config_work); | |
761 | ||
762 | del_gendisk(vblk->disk); | |
763 | blk_cleanup_queue(vblk->disk->queue); | |
764 | ||
765 | blk_mq_free_tag_set(&vblk->tag_set); | |
766 | ||
767 | /* Stop all the virtqueues. */ | |
768 | vdev->config->reset(vdev); | |
769 | ||
770 | refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount); | |
771 | put_disk(vblk->disk); | |
772 | vdev->config->del_vqs(vdev); | |
773 | kfree(vblk->vqs); | |
774 | kfree(vblk); | |
775 | ||
776 | /* Only free device id if we don't have any users */ | |
777 | if (refc == 1) | |
778 | ida_simple_remove(&vd_index_ida, index); | |
779 | } | |
780 | ||
781 | #ifdef CONFIG_PM_SLEEP | |
782 | static int virtblk_freeze(struct virtio_device *vdev) | |
783 | { | |
784 | struct virtio_blk *vblk = vdev->priv; | |
785 | ||
786 | /* Ensure we don't receive any more interrupts */ | |
787 | vdev->config->reset(vdev); | |
788 | ||
789 | /* Make sure no work handler is accessing the device. */ | |
790 | flush_work(&vblk->config_work); | |
791 | ||
792 | blk_mq_stop_hw_queues(vblk->disk->queue); | |
793 | ||
794 | vdev->config->del_vqs(vdev); | |
795 | return 0; | |
796 | } | |
797 | ||
798 | static int virtblk_restore(struct virtio_device *vdev) | |
799 | { | |
800 | struct virtio_blk *vblk = vdev->priv; | |
801 | int ret; | |
802 | ||
803 | ret = init_vq(vdev->priv); | |
804 | if (ret) | |
805 | return ret; | |
806 | ||
807 | virtio_device_ready(vdev); | |
808 | ||
809 | blk_mq_start_stopped_hw_queues(vblk->disk->queue, true); | |
810 | return 0; | |
811 | } | |
812 | #endif | |
813 | ||
814 | static const struct virtio_device_id id_table[] = { | |
815 | { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, | |
816 | { 0 }, | |
817 | }; | |
818 | ||
819 | static unsigned int features_legacy[] = { | |
820 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, | |
821 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI, | |
822 | VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE, | |
823 | VIRTIO_BLK_F_MQ, | |
824 | } | |
825 | ; | |
826 | static unsigned int features[] = { | |
827 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, | |
828 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, | |
829 | VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE, | |
830 | VIRTIO_BLK_F_MQ, | |
831 | }; | |
832 | ||
833 | static struct virtio_driver virtio_blk = { | |
834 | .feature_table = features, | |
835 | .feature_table_size = ARRAY_SIZE(features), | |
836 | .feature_table_legacy = features_legacy, | |
837 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
838 | .driver.name = KBUILD_MODNAME, | |
839 | .driver.owner = THIS_MODULE, | |
840 | .id_table = id_table, | |
841 | .probe = virtblk_probe, | |
842 | .remove = virtblk_remove, | |
843 | .config_changed = virtblk_config_changed, | |
844 | #ifdef CONFIG_PM_SLEEP | |
845 | .freeze = virtblk_freeze, | |
846 | .restore = virtblk_restore, | |
847 | #endif | |
848 | }; | |
849 | ||
850 | static int __init init(void) | |
851 | { | |
852 | int error; | |
853 | ||
854 | virtblk_wq = alloc_workqueue("virtio-blk", 0, 0); | |
855 | if (!virtblk_wq) | |
856 | return -ENOMEM; | |
857 | ||
858 | major = register_blkdev(0, "virtblk"); | |
859 | if (major < 0) { | |
860 | error = major; | |
861 | goto out_destroy_workqueue; | |
862 | } | |
863 | ||
864 | error = register_virtio_driver(&virtio_blk); | |
865 | if (error) | |
866 | goto out_unregister_blkdev; | |
867 | return 0; | |
868 | ||
869 | out_unregister_blkdev: | |
870 | unregister_blkdev(major, "virtblk"); | |
871 | out_destroy_workqueue: | |
872 | destroy_workqueue(virtblk_wq); | |
873 | return error; | |
874 | } | |
875 | ||
876 | static void __exit fini(void) | |
877 | { | |
878 | unregister_virtio_driver(&virtio_blk); | |
879 | unregister_blkdev(major, "virtblk"); | |
880 | destroy_workqueue(virtblk_wq); | |
881 | } | |
882 | module_init(init); | |
883 | module_exit(fini); | |
884 | ||
885 | MODULE_DEVICE_TABLE(virtio, id_table); | |
886 | MODULE_DESCRIPTION("Virtio block driver"); | |
887 | MODULE_LICENSE("GPL"); |