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 | ||
0864b79a RR |
44 | /* What host tells us, plus 2 for header & tailer. */ |
45 | unsigned int sg_elems; | |
46 | ||
5087a50e MT |
47 | /* Ida index - used to track minor number allocations. */ |
48 | int index; | |
6a27b656 ML |
49 | |
50 | /* num of vqs */ | |
51 | int num_vqs; | |
52 | struct virtio_blk_vq *vqs; | |
e467cde2 RR |
53 | }; |
54 | ||
55 | struct virtblk_req | |
56 | { | |
e467cde2 RR |
57 | struct request *req; |
58 | struct virtio_blk_outhdr out_hdr; | |
1cde26f9 | 59 | struct virtio_scsi_inhdr in_hdr; |
cb38fa23 | 60 | u8 status; |
a98755c5 | 61 | struct scatterlist sg[]; |
e467cde2 RR |
62 | }; |
63 | ||
a98755c5 AH |
64 | static inline int virtblk_result(struct virtblk_req *vbr) |
65 | { | |
66 | switch (vbr->status) { | |
67 | case VIRTIO_BLK_S_OK: | |
68 | return 0; | |
69 | case VIRTIO_BLK_S_UNSUPP: | |
70 | return -ENOTTY; | |
71 | default: | |
72 | return -EIO; | |
73 | } | |
74 | } | |
75 | ||
8f39db9d | 76 | static int __virtblk_add_req(struct virtqueue *vq, |
20af3cfd PB |
77 | struct virtblk_req *vbr, |
78 | struct scatterlist *data_sg, | |
0a11cc36 | 79 | bool have_data) |
c85a1f91 | 80 | { |
20af3cfd | 81 | struct scatterlist hdr, status, cmd, sense, inhdr, *sgs[6]; |
8f39db9d | 82 | unsigned int num_out = 0, num_in = 0; |
19c1c5a6 | 83 | __virtio32 type = vbr->out_hdr.type & ~cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT); |
8f39db9d PB |
84 | |
85 | sg_init_one(&hdr, &vbr->out_hdr, sizeof(vbr->out_hdr)); | |
86 | sgs[num_out++] = &hdr; | |
87 | ||
20af3cfd PB |
88 | /* |
89 | * If this is a packet command we need a couple of additional headers. | |
90 | * Behind the normal outhdr we put a segment with the scsi command | |
91 | * block, and before the normal inhdr we put the sense data and the | |
92 | * inhdr with additional status information. | |
93 | */ | |
19c1c5a6 | 94 | if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) { |
20af3cfd PB |
95 | sg_init_one(&cmd, vbr->req->cmd, vbr->req->cmd_len); |
96 | sgs[num_out++] = &cmd; | |
97 | } | |
98 | ||
0a11cc36 | 99 | if (have_data) { |
19c1c5a6 | 100 | if (vbr->out_hdr.type & cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_OUT)) |
20af3cfd | 101 | sgs[num_out++] = data_sg; |
8f39db9d | 102 | else |
20af3cfd PB |
103 | sgs[num_out + num_in++] = data_sg; |
104 | } | |
105 | ||
19c1c5a6 | 106 | if (type == cpu_to_virtio32(vq->vdev, VIRTIO_BLK_T_SCSI_CMD)) { |
20af3cfd PB |
107 | sg_init_one(&sense, vbr->req->sense, SCSI_SENSE_BUFFERSIZE); |
108 | sgs[num_out + num_in++] = &sense; | |
109 | sg_init_one(&inhdr, &vbr->in_hdr, sizeof(vbr->in_hdr)); | |
110 | sgs[num_out + num_in++] = &inhdr; | |
8f39db9d PB |
111 | } |
112 | ||
113 | sg_init_one(&status, &vbr->status, sizeof(vbr->status)); | |
114 | sgs[num_out + num_in++] = &status; | |
115 | ||
116 | return virtqueue_add_sgs(vq, sgs, num_out, num_in, vbr, GFP_ATOMIC); | |
5ee21a52 PB |
117 | } |
118 | ||
5124c285 | 119 | static inline void virtblk_request_done(struct request *req) |
a98755c5 | 120 | { |
9d74e257 | 121 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); |
19c1c5a6 | 122 | struct virtio_blk *vblk = req->q->queuedata; |
a98755c5 AH |
123 | int error = virtblk_result(vbr); |
124 | ||
125 | if (req->cmd_type == REQ_TYPE_BLOCK_PC) { | |
19c1c5a6 MT |
126 | req->resid_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.residual); |
127 | req->sense_len = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.sense_len); | |
128 | req->errors = virtio32_to_cpu(vblk->vdev, vbr->in_hdr.errors); | |
a98755c5 AH |
129 | } else if (req->cmd_type == REQ_TYPE_SPECIAL) { |
130 | req->errors = (error != 0); | |
131 | } | |
132 | ||
c8a446ad | 133 | blk_mq_end_request(req, error); |
a98755c5 AH |
134 | } |
135 | ||
136 | static void virtblk_done(struct virtqueue *vq) | |
e467cde2 RR |
137 | { |
138 | struct virtio_blk *vblk = vq->vdev->priv; | |
1cf7e9c6 | 139 | bool req_done = false; |
6a27b656 | 140 | int qid = vq->index; |
e467cde2 | 141 | struct virtblk_req *vbr; |
e467cde2 | 142 | unsigned long flags; |
a98755c5 | 143 | unsigned int len; |
e467cde2 | 144 | |
6a27b656 | 145 | spin_lock_irqsave(&vblk->vqs[qid].lock, flags); |
bb811108 AH |
146 | do { |
147 | virtqueue_disable_cb(vq); | |
6a27b656 | 148 | while ((vbr = virtqueue_get_buf(vblk->vqs[qid].vq, &len)) != NULL) { |
5124c285 | 149 | blk_mq_complete_request(vbr->req); |
1cf7e9c6 | 150 | req_done = true; |
33659ebb | 151 | } |
7f03b17d HG |
152 | if (unlikely(virtqueue_is_broken(vq))) |
153 | break; | |
bb811108 | 154 | } while (!virtqueue_enable_cb(vq)); |
1cf7e9c6 | 155 | |
e467cde2 | 156 | /* In case queue is stopped waiting for more buffers. */ |
a98755c5 | 157 | if (req_done) |
1b4a3258 | 158 | blk_mq_start_stopped_hw_queues(vblk->disk->queue, true); |
6a27b656 | 159 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); |
a98755c5 AH |
160 | } |
161 | ||
bf572297 CH |
162 | static int virtio_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *req, |
163 | bool last) | |
e467cde2 | 164 | { |
1cf7e9c6 | 165 | struct virtio_blk *vblk = hctx->queue->queuedata; |
9d74e257 | 166 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(req); |
1cf7e9c6 | 167 | unsigned long flags; |
20af3cfd | 168 | unsigned int num; |
6a27b656 | 169 | int qid = hctx->queue_num; |
5261b85e | 170 | int err; |
e8edca6f | 171 | bool notify = false; |
e467cde2 | 172 | |
1cf7e9c6 | 173 | BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems); |
e467cde2 RR |
174 | |
175 | vbr->req = req; | |
dd40e456 | 176 | if (req->cmd_flags & REQ_FLUSH) { |
19c1c5a6 | 177 | vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_FLUSH); |
4cb2ea28 | 178 | vbr->out_hdr.sector = 0; |
19c1c5a6 | 179 | vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req)); |
dd40e456 FT |
180 | } else { |
181 | switch (req->cmd_type) { | |
182 | case REQ_TYPE_FS: | |
183 | vbr->out_hdr.type = 0; | |
19c1c5a6 MT |
184 | vbr->out_hdr.sector = cpu_to_virtio64(vblk->vdev, blk_rq_pos(vbr->req)); |
185 | vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req)); | |
dd40e456 FT |
186 | break; |
187 | case REQ_TYPE_BLOCK_PC: | |
19c1c5a6 | 188 | vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_SCSI_CMD); |
f1b0ef06 | 189 | vbr->out_hdr.sector = 0; |
19c1c5a6 | 190 | vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req)); |
f1b0ef06 | 191 | break; |
dd40e456 | 192 | case REQ_TYPE_SPECIAL: |
19c1c5a6 | 193 | vbr->out_hdr.type = cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_GET_ID); |
dd40e456 | 194 | vbr->out_hdr.sector = 0; |
19c1c5a6 | 195 | vbr->out_hdr.ioprio = cpu_to_virtio32(vblk->vdev, req_get_ioprio(vbr->req)); |
dd40e456 FT |
196 | break; |
197 | default: | |
198 | /* We don't put anything else in the queue. */ | |
199 | BUG(); | |
f1b0ef06 | 200 | } |
e467cde2 RR |
201 | } |
202 | ||
e2490073 CH |
203 | blk_mq_start_request(req); |
204 | ||
1cf7e9c6 | 205 | num = blk_rq_map_sg(hctx->queue, vbr->req, vbr->sg); |
1cde26f9 | 206 | if (num) { |
20af3cfd | 207 | if (rq_data_dir(vbr->req) == WRITE) |
19c1c5a6 | 208 | vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_OUT); |
20af3cfd | 209 | else |
19c1c5a6 | 210 | vbr->out_hdr.type |= cpu_to_virtio32(vblk->vdev, VIRTIO_BLK_T_IN); |
e467cde2 RR |
211 | } |
212 | ||
6a27b656 ML |
213 | spin_lock_irqsave(&vblk->vqs[qid].lock, flags); |
214 | err = __virtblk_add_req(vblk->vqs[qid].vq, vbr, vbr->sg, num); | |
5261b85e | 215 | if (err) { |
6a27b656 | 216 | virtqueue_kick(vblk->vqs[qid].vq); |
1cf7e9c6 | 217 | blk_mq_stop_hw_queue(hctx); |
6a27b656 | 218 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); |
5261b85e RR |
219 | /* Out of mem doesn't actually happen, since we fall back |
220 | * to direct descriptors */ | |
221 | if (err == -ENOMEM || err == -ENOSPC) | |
222 | return BLK_MQ_RQ_QUEUE_BUSY; | |
223 | return BLK_MQ_RQ_QUEUE_ERROR; | |
a98755c5 AH |
224 | } |
225 | ||
6a27b656 | 226 | if (last && virtqueue_kick_prepare(vblk->vqs[qid].vq)) |
e8edca6f | 227 | notify = true; |
6a27b656 | 228 | spin_unlock_irqrestore(&vblk->vqs[qid].lock, flags); |
e8edca6f ML |
229 | |
230 | if (notify) | |
6a27b656 | 231 | virtqueue_notify(vblk->vqs[qid].vq); |
1cf7e9c6 | 232 | return BLK_MQ_RQ_QUEUE_OK; |
a98755c5 AH |
233 | } |
234 | ||
4cb2ea28 | 235 | /* return id (s/n) string for *disk to *id_str |
236 | */ | |
237 | static int virtblk_get_id(struct gendisk *disk, char *id_str) | |
238 | { | |
239 | struct virtio_blk *vblk = disk->private_data; | |
240 | struct request *req; | |
241 | struct bio *bio; | |
e4c4776d | 242 | int err; |
4cb2ea28 | 243 | |
244 | bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES, | |
245 | GFP_KERNEL); | |
246 | if (IS_ERR(bio)) | |
247 | return PTR_ERR(bio); | |
248 | ||
249 | req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL); | |
250 | if (IS_ERR(req)) { | |
251 | bio_put(bio); | |
252 | return PTR_ERR(req); | |
253 | } | |
254 | ||
255 | req->cmd_type = REQ_TYPE_SPECIAL; | |
e4c4776d MS |
256 | err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false); |
257 | blk_put_request(req); | |
258 | ||
259 | return err; | |
4cb2ea28 | 260 | } |
261 | ||
fe5a50a1 CH |
262 | static int virtblk_ioctl(struct block_device *bdev, fmode_t mode, |
263 | unsigned int cmd, unsigned long data) | |
e467cde2 | 264 | { |
1cde26f9 HR |
265 | struct gendisk *disk = bdev->bd_disk; |
266 | struct virtio_blk *vblk = disk->private_data; | |
267 | ||
268 | /* | |
269 | * Only allow the generic SCSI ioctls if the host can support it. | |
270 | */ | |
271 | if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI)) | |
d9ecdea7 | 272 | return -ENOTTY; |
1cde26f9 | 273 | |
577ebb37 PB |
274 | return scsi_cmd_blk_ioctl(bdev, mode, cmd, |
275 | (void __user *)data); | |
e467cde2 RR |
276 | } |
277 | ||
135da0b0 CB |
278 | /* We provide getgeo only to please some old bootloader/partitioning tools */ |
279 | static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo) | |
280 | { | |
48e4043d | 281 | struct virtio_blk *vblk = bd->bd_disk->private_data; |
48e4043d RH |
282 | |
283 | /* see if the host passed in geometry config */ | |
855e0c52 RR |
284 | if (virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_GEOMETRY)) { |
285 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
286 | geometry.cylinders, &geo->cylinders); | |
287 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
288 | geometry.heads, &geo->heads); | |
289 | virtio_cread(vblk->vdev, struct virtio_blk_config, | |
290 | geometry.sectors, &geo->sectors); | |
48e4043d RH |
291 | } else { |
292 | /* some standard values, similar to sd */ | |
293 | geo->heads = 1 << 6; | |
294 | geo->sectors = 1 << 5; | |
295 | geo->cylinders = get_capacity(bd->bd_disk) >> 11; | |
296 | } | |
135da0b0 CB |
297 | return 0; |
298 | } | |
299 | ||
83d5cde4 | 300 | static const struct block_device_operations virtblk_fops = { |
8a6cfeb6 | 301 | .ioctl = virtblk_ioctl, |
135da0b0 CB |
302 | .owner = THIS_MODULE, |
303 | .getgeo = virtblk_getgeo, | |
e467cde2 RR |
304 | }; |
305 | ||
d50ed907 CB |
306 | static int index_to_minor(int index) |
307 | { | |
308 | return index << PART_BITS; | |
309 | } | |
310 | ||
5087a50e MT |
311 | static int minor_to_index(int minor) |
312 | { | |
313 | return minor >> PART_BITS; | |
314 | } | |
315 | ||
a5eb9e4f RH |
316 | static ssize_t virtblk_serial_show(struct device *dev, |
317 | struct device_attribute *attr, char *buf) | |
318 | { | |
319 | struct gendisk *disk = dev_to_disk(dev); | |
320 | int err; | |
321 | ||
322 | /* sysfs gives us a PAGE_SIZE buffer */ | |
323 | BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES); | |
324 | ||
325 | buf[VIRTIO_BLK_ID_BYTES] = '\0'; | |
326 | err = virtblk_get_id(disk, buf); | |
327 | if (!err) | |
328 | return strlen(buf); | |
329 | ||
330 | if (err == -EIO) /* Unsupported? Make it empty. */ | |
331 | return 0; | |
332 | ||
333 | return err; | |
334 | } | |
393c525b MT |
335 | |
336 | static DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL); | |
a5eb9e4f | 337 | |
7a7c924c CH |
338 | static void virtblk_config_changed_work(struct work_struct *work) |
339 | { | |
340 | struct virtio_blk *vblk = | |
341 | container_of(work, struct virtio_blk, config_work); | |
342 | struct virtio_device *vdev = vblk->vdev; | |
343 | struct request_queue *q = vblk->disk->queue; | |
344 | char cap_str_2[10], cap_str_10[10]; | |
9d9598b8 | 345 | char *envp[] = { "RESIZE=1", NULL }; |
7a7c924c CH |
346 | u64 capacity, size; |
347 | ||
348 | /* Host must always specify the capacity. */ | |
855e0c52 | 349 | virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity); |
7a7c924c CH |
350 | |
351 | /* If capacity is too big, truncate with warning. */ | |
352 | if ((sector_t)capacity != capacity) { | |
353 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
354 | (unsigned long long)capacity); | |
355 | capacity = (sector_t)-1; | |
356 | } | |
357 | ||
358 | size = capacity * queue_logical_block_size(q); | |
359 | string_get_size(size, STRING_UNITS_2, cap_str_2, sizeof(cap_str_2)); | |
360 | string_get_size(size, STRING_UNITS_10, cap_str_10, sizeof(cap_str_10)); | |
361 | ||
362 | dev_notice(&vdev->dev, | |
363 | "new size: %llu %d-byte logical blocks (%s/%s)\n", | |
364 | (unsigned long long)capacity, | |
365 | queue_logical_block_size(q), | |
366 | cap_str_10, cap_str_2); | |
367 | ||
368 | set_capacity(vblk->disk, capacity); | |
e9986f30 | 369 | revalidate_disk(vblk->disk); |
9d9598b8 | 370 | kobject_uevent_env(&disk_to_dev(vblk->disk)->kobj, KOBJ_CHANGE, envp); |
7a7c924c CH |
371 | } |
372 | ||
373 | static void virtblk_config_changed(struct virtio_device *vdev) | |
374 | { | |
375 | struct virtio_blk *vblk = vdev->priv; | |
376 | ||
377 | queue_work(virtblk_wq, &vblk->config_work); | |
378 | } | |
379 | ||
6abd6e5a AS |
380 | static int init_vq(struct virtio_blk *vblk) |
381 | { | |
382 | int err = 0; | |
6a27b656 ML |
383 | int i; |
384 | vq_callback_t **callbacks; | |
385 | const char **names; | |
386 | struct virtqueue **vqs; | |
387 | unsigned short num_vqs; | |
388 | struct virtio_device *vdev = vblk->vdev; | |
389 | ||
390 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_MQ, | |
391 | struct virtio_blk_config, num_queues, | |
392 | &num_vqs); | |
393 | if (err) | |
394 | num_vqs = 1; | |
395 | ||
396 | vblk->vqs = kmalloc(sizeof(*vblk->vqs) * num_vqs, GFP_KERNEL); | |
397 | if (!vblk->vqs) { | |
398 | err = -ENOMEM; | |
399 | goto out; | |
400 | } | |
401 | ||
402 | names = kmalloc(sizeof(*names) * num_vqs, GFP_KERNEL); | |
403 | if (!names) | |
404 | goto err_names; | |
405 | ||
406 | callbacks = kmalloc(sizeof(*callbacks) * num_vqs, GFP_KERNEL); | |
407 | if (!callbacks) | |
408 | goto err_callbacks; | |
409 | ||
410 | vqs = kmalloc(sizeof(*vqs) * num_vqs, GFP_KERNEL); | |
411 | if (!vqs) | |
412 | goto err_vqs; | |
6abd6e5a | 413 | |
6a27b656 ML |
414 | for (i = 0; i < num_vqs; i++) { |
415 | callbacks[i] = virtblk_done; | |
416 | snprintf(vblk->vqs[i].name, VQ_NAME_LEN, "req.%d", i); | |
417 | names[i] = vblk->vqs[i].name; | |
418 | } | |
419 | ||
420 | /* Discover virtqueues and write information to configuration. */ | |
421 | err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names); | |
422 | if (err) | |
423 | goto err_find_vqs; | |
6abd6e5a | 424 | |
6a27b656 ML |
425 | for (i = 0; i < num_vqs; i++) { |
426 | spin_lock_init(&vblk->vqs[i].lock); | |
427 | vblk->vqs[i].vq = vqs[i]; | |
428 | } | |
429 | vblk->num_vqs = num_vqs; | |
430 | ||
431 | err_find_vqs: | |
432 | kfree(vqs); | |
433 | err_vqs: | |
434 | kfree(callbacks); | |
435 | err_callbacks: | |
436 | kfree(names); | |
437 | err_names: | |
438 | if (err) | |
439 | kfree(vblk->vqs); | |
440 | out: | |
6abd6e5a AS |
441 | return err; |
442 | } | |
443 | ||
c0aa3e09 RM |
444 | /* |
445 | * Legacy naming scheme used for virtio devices. We are stuck with it for | |
446 | * virtio blk but don't ever use it for any new driver. | |
447 | */ | |
448 | static int virtblk_name_format(char *prefix, int index, char *buf, int buflen) | |
449 | { | |
450 | const int base = 'z' - 'a' + 1; | |
451 | char *begin = buf + strlen(prefix); | |
452 | char *end = buf + buflen; | |
453 | char *p; | |
454 | int unit; | |
455 | ||
456 | p = end - 1; | |
457 | *p = '\0'; | |
458 | unit = base; | |
459 | do { | |
460 | if (p == begin) | |
461 | return -EINVAL; | |
462 | *--p = 'a' + (index % unit); | |
463 | index = (index / unit) - 1; | |
464 | } while (index >= 0); | |
465 | ||
466 | memmove(begin, p, end - p); | |
467 | memcpy(buf, prefix, strlen(prefix)); | |
468 | ||
469 | return 0; | |
470 | } | |
471 | ||
cd5d5038 PB |
472 | static int virtblk_get_cache_mode(struct virtio_device *vdev) |
473 | { | |
474 | u8 writeback; | |
475 | int err; | |
476 | ||
855e0c52 RR |
477 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE, |
478 | struct virtio_blk_config, wce, | |
479 | &writeback); | |
cd5d5038 | 480 | if (err) |
19c1c5a6 MT |
481 | writeback = virtio_has_feature(vdev, VIRTIO_BLK_F_WCE) || |
482 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1); | |
cd5d5038 PB |
483 | |
484 | return writeback; | |
485 | } | |
486 | ||
487 | static void virtblk_update_cache_mode(struct virtio_device *vdev) | |
488 | { | |
489 | u8 writeback = virtblk_get_cache_mode(vdev); | |
490 | struct virtio_blk *vblk = vdev->priv; | |
491 | ||
c85a1f91 | 492 | if (writeback) |
cd5d5038 PB |
493 | blk_queue_flush(vblk->disk->queue, REQ_FLUSH); |
494 | else | |
495 | blk_queue_flush(vblk->disk->queue, 0); | |
496 | ||
497 | revalidate_disk(vblk->disk); | |
498 | } | |
499 | ||
500 | static const char *const virtblk_cache_types[] = { | |
501 | "write through", "write back" | |
502 | }; | |
503 | ||
504 | static ssize_t | |
505 | virtblk_cache_type_store(struct device *dev, struct device_attribute *attr, | |
506 | const char *buf, size_t count) | |
507 | { | |
508 | struct gendisk *disk = dev_to_disk(dev); | |
509 | struct virtio_blk *vblk = disk->private_data; | |
510 | struct virtio_device *vdev = vblk->vdev; | |
511 | int i; | |
cd5d5038 PB |
512 | |
513 | BUG_ON(!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_CONFIG_WCE)); | |
514 | for (i = ARRAY_SIZE(virtblk_cache_types); --i >= 0; ) | |
515 | if (sysfs_streq(buf, virtblk_cache_types[i])) | |
516 | break; | |
517 | ||
518 | if (i < 0) | |
519 | return -EINVAL; | |
520 | ||
855e0c52 | 521 | virtio_cwrite8(vdev, offsetof(struct virtio_blk_config, wce), i); |
cd5d5038 PB |
522 | virtblk_update_cache_mode(vdev); |
523 | return count; | |
524 | } | |
525 | ||
526 | static ssize_t | |
527 | virtblk_cache_type_show(struct device *dev, struct device_attribute *attr, | |
528 | char *buf) | |
529 | { | |
530 | struct gendisk *disk = dev_to_disk(dev); | |
531 | struct virtio_blk *vblk = disk->private_data; | |
532 | u8 writeback = virtblk_get_cache_mode(vblk->vdev); | |
533 | ||
534 | BUG_ON(writeback >= ARRAY_SIZE(virtblk_cache_types)); | |
535 | return snprintf(buf, 40, "%s\n", virtblk_cache_types[writeback]); | |
536 | } | |
537 | ||
538 | static const struct device_attribute dev_attr_cache_type_ro = | |
539 | __ATTR(cache_type, S_IRUGO, | |
540 | virtblk_cache_type_show, NULL); | |
541 | static const struct device_attribute dev_attr_cache_type_rw = | |
542 | __ATTR(cache_type, S_IRUGO|S_IWUSR, | |
543 | virtblk_cache_type_show, virtblk_cache_type_store); | |
544 | ||
24d2f903 CH |
545 | static int virtblk_init_request(void *data, struct request *rq, |
546 | unsigned int hctx_idx, unsigned int request_idx, | |
547 | unsigned int numa_node) | |
e9b267d9 CH |
548 | { |
549 | struct virtio_blk *vblk = data; | |
550 | struct virtblk_req *vbr = blk_mq_rq_to_pdu(rq); | |
551 | ||
552 | sg_init_table(vbr->sg, vblk->sg_elems); | |
553 | return 0; | |
554 | } | |
555 | ||
1cf7e9c6 JA |
556 | static struct blk_mq_ops virtio_mq_ops = { |
557 | .queue_rq = virtio_queue_rq, | |
558 | .map_queue = blk_mq_map_queue, | |
5124c285 | 559 | .complete = virtblk_request_done, |
24d2f903 | 560 | .init_request = virtblk_init_request, |
1cf7e9c6 JA |
561 | }; |
562 | ||
24d2f903 CH |
563 | static unsigned int virtblk_queue_depth; |
564 | module_param_named(queue_depth, virtblk_queue_depth, uint, 0444); | |
1cf7e9c6 | 565 | |
8d85fce7 | 566 | static int virtblk_probe(struct virtio_device *vdev) |
e467cde2 RR |
567 | { |
568 | struct virtio_blk *vblk; | |
69740c8b | 569 | struct request_queue *q; |
5087a50e | 570 | int err, index; |
a98755c5 | 571 | |
e467cde2 | 572 | u64 cap; |
69740c8b CH |
573 | u32 v, blk_size, sg_elems, opt_io_size; |
574 | u16 min_io_size; | |
575 | u8 physical_block_exp, alignment_offset; | |
e467cde2 | 576 | |
5087a50e MT |
577 | err = ida_simple_get(&vd_index_ida, 0, minor_to_index(1 << MINORBITS), |
578 | GFP_KERNEL); | |
579 | if (err < 0) | |
580 | goto out; | |
581 | index = err; | |
4f3bf19c | 582 | |
0864b79a | 583 | /* We need to know how many segments before we allocate. */ |
855e0c52 RR |
584 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SEG_MAX, |
585 | struct virtio_blk_config, seg_max, | |
586 | &sg_elems); | |
a5b365a6 CH |
587 | |
588 | /* We need at least one SG element, whatever they say. */ | |
589 | if (err || !sg_elems) | |
0864b79a RR |
590 | sg_elems = 1; |
591 | ||
592 | /* We need an extra sg elements at head and tail. */ | |
593 | sg_elems += 2; | |
1cf7e9c6 | 594 | vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL); |
e467cde2 RR |
595 | if (!vblk) { |
596 | err = -ENOMEM; | |
5087a50e | 597 | goto out_free_index; |
e467cde2 RR |
598 | } |
599 | ||
e467cde2 | 600 | vblk->vdev = vdev; |
0864b79a | 601 | vblk->sg_elems = sg_elems; |
a98755c5 | 602 | |
7a7c924c | 603 | INIT_WORK(&vblk->config_work, virtblk_config_changed_work); |
e467cde2 | 604 | |
6abd6e5a AS |
605 | err = init_vq(vblk); |
606 | if (err) | |
e467cde2 | 607 | goto out_free_vblk; |
e467cde2 | 608 | |
e467cde2 | 609 | /* FIXME: How many partitions? How long is a piece of string? */ |
4f3bf19c | 610 | vblk->disk = alloc_disk(1 << PART_BITS); |
e467cde2 RR |
611 | if (!vblk->disk) { |
612 | err = -ENOMEM; | |
1cf7e9c6 | 613 | goto out_free_vq; |
e467cde2 RR |
614 | } |
615 | ||
fc4324b4 | 616 | /* Default queue sizing is to fill the ring. */ |
24d2f903 | 617 | if (!virtblk_queue_depth) { |
6a27b656 | 618 | virtblk_queue_depth = vblk->vqs[0].vq->num_free; |
fc4324b4 RR |
619 | /* ... but without indirect descs, we use 2 descs per req */ |
620 | if (!virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC)) | |
24d2f903 | 621 | virtblk_queue_depth /= 2; |
fc4324b4 | 622 | } |
24d2f903 CH |
623 | |
624 | memset(&vblk->tag_set, 0, sizeof(vblk->tag_set)); | |
625 | vblk->tag_set.ops = &virtio_mq_ops; | |
24d2f903 CH |
626 | vblk->tag_set.queue_depth = virtblk_queue_depth; |
627 | vblk->tag_set.numa_node = NUMA_NO_NODE; | |
628 | vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; | |
629 | vblk->tag_set.cmd_size = | |
1cf7e9c6 JA |
630 | sizeof(struct virtblk_req) + |
631 | sizeof(struct scatterlist) * sg_elems; | |
24d2f903 | 632 | vblk->tag_set.driver_data = vblk; |
6a27b656 | 633 | vblk->tag_set.nr_hw_queues = vblk->num_vqs; |
1cf7e9c6 | 634 | |
24d2f903 CH |
635 | err = blk_mq_alloc_tag_set(&vblk->tag_set); |
636 | if (err) | |
637 | goto out_put_disk; | |
638 | ||
639 | q = vblk->disk->queue = blk_mq_init_queue(&vblk->tag_set); | |
69740c8b | 640 | if (!q) { |
e467cde2 | 641 | err = -ENOMEM; |
24d2f903 | 642 | goto out_free_tags; |
e467cde2 RR |
643 | } |
644 | ||
69740c8b | 645 | q->queuedata = vblk; |
7d116b62 | 646 | |
c0aa3e09 | 647 | virtblk_name_format("vd", index, vblk->disk->disk_name, DISK_NAME_LEN); |
d50ed907 | 648 | |
e467cde2 | 649 | vblk->disk->major = major; |
d50ed907 | 650 | vblk->disk->first_minor = index_to_minor(index); |
e467cde2 RR |
651 | vblk->disk->private_data = vblk; |
652 | vblk->disk->fops = &virtblk_fops; | |
c4839346 | 653 | vblk->disk->driverfs_dev = &vdev->dev; |
5087a50e | 654 | vblk->index = index; |
4f3bf19c | 655 | |
02c42b7a | 656 | /* configure queue flush support */ |
cd5d5038 | 657 | virtblk_update_cache_mode(vdev); |
e467cde2 | 658 | |
3ef53609 CB |
659 | /* If disk is read-only in the host, the guest should obey */ |
660 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO)) | |
661 | set_disk_ro(vblk->disk, 1); | |
662 | ||
a586d4f6 | 663 | /* Host must always specify the capacity. */ |
855e0c52 | 664 | virtio_cread(vdev, struct virtio_blk_config, capacity, &cap); |
e467cde2 RR |
665 | |
666 | /* If capacity is too big, truncate with warning. */ | |
667 | if ((sector_t)cap != cap) { | |
668 | dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n", | |
669 | (unsigned long long)cap); | |
670 | cap = (sector_t)-1; | |
671 | } | |
672 | set_capacity(vblk->disk, cap); | |
673 | ||
0864b79a | 674 | /* We can handle whatever the host told us to handle. */ |
ee714f2d | 675 | blk_queue_max_segments(q, vblk->sg_elems-2); |
0864b79a | 676 | |
4eff3cae | 677 | /* No need to bounce any requests */ |
69740c8b | 678 | blk_queue_bounce_limit(q, BLK_BOUNCE_ANY); |
4eff3cae | 679 | |
4b7f7e20 | 680 | /* No real sector limit. */ |
ee714f2d | 681 | blk_queue_max_hw_sectors(q, -1U); |
4b7f7e20 | 682 | |
a586d4f6 RR |
683 | /* Host can optionally specify maximum segment size and number of |
684 | * segments. */ | |
855e0c52 RR |
685 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_SIZE_MAX, |
686 | struct virtio_blk_config, size_max, &v); | |
e467cde2 | 687 | if (!err) |
69740c8b | 688 | blk_queue_max_segment_size(q, v); |
4b7f7e20 | 689 | else |
69740c8b | 690 | blk_queue_max_segment_size(q, -1U); |
e467cde2 | 691 | |
066f4d82 | 692 | /* Host can optionally specify the block size of the device */ |
855e0c52 RR |
693 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_BLK_SIZE, |
694 | struct virtio_blk_config, blk_size, | |
695 | &blk_size); | |
066f4d82 | 696 | if (!err) |
69740c8b CH |
697 | blk_queue_logical_block_size(q, blk_size); |
698 | else | |
699 | blk_size = queue_logical_block_size(q); | |
700 | ||
701 | /* Use topology information if available */ | |
855e0c52 RR |
702 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
703 | struct virtio_blk_config, physical_block_exp, | |
704 | &physical_block_exp); | |
69740c8b CH |
705 | if (!err && physical_block_exp) |
706 | blk_queue_physical_block_size(q, | |
707 | blk_size * (1 << physical_block_exp)); | |
708 | ||
855e0c52 RR |
709 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
710 | struct virtio_blk_config, alignment_offset, | |
711 | &alignment_offset); | |
69740c8b CH |
712 | if (!err && alignment_offset) |
713 | blk_queue_alignment_offset(q, blk_size * alignment_offset); | |
714 | ||
855e0c52 RR |
715 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
716 | struct virtio_blk_config, min_io_size, | |
717 | &min_io_size); | |
69740c8b CH |
718 | if (!err && min_io_size) |
719 | blk_queue_io_min(q, blk_size * min_io_size); | |
720 | ||
855e0c52 RR |
721 | err = virtio_cread_feature(vdev, VIRTIO_BLK_F_TOPOLOGY, |
722 | struct virtio_blk_config, opt_io_size, | |
723 | &opt_io_size); | |
69740c8b CH |
724 | if (!err && opt_io_size) |
725 | blk_queue_io_opt(q, blk_size * opt_io_size); | |
726 | ||
7a11370e MT |
727 | virtio_device_ready(vdev); |
728 | ||
e467cde2 | 729 | add_disk(vblk->disk); |
a5eb9e4f RH |
730 | err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial); |
731 | if (err) | |
732 | goto out_del_disk; | |
733 | ||
cd5d5038 PB |
734 | if (virtio_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) |
735 | err = device_create_file(disk_to_dev(vblk->disk), | |
736 | &dev_attr_cache_type_rw); | |
737 | else | |
738 | err = device_create_file(disk_to_dev(vblk->disk), | |
739 | &dev_attr_cache_type_ro); | |
740 | if (err) | |
741 | goto out_del_disk; | |
e467cde2 RR |
742 | return 0; |
743 | ||
a5eb9e4f RH |
744 | out_del_disk: |
745 | del_gendisk(vblk->disk); | |
746 | blk_cleanup_queue(vblk->disk->queue); | |
24d2f903 CH |
747 | out_free_tags: |
748 | blk_mq_free_tag_set(&vblk->tag_set); | |
e467cde2 RR |
749 | out_put_disk: |
750 | put_disk(vblk->disk); | |
e467cde2 | 751 | out_free_vq: |
d2a7ddda | 752 | vdev->config->del_vqs(vdev); |
e467cde2 RR |
753 | out_free_vblk: |
754 | kfree(vblk); | |
5087a50e MT |
755 | out_free_index: |
756 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
757 | out: |
758 | return err; | |
759 | } | |
760 | ||
8d85fce7 | 761 | static void virtblk_remove(struct virtio_device *vdev) |
e467cde2 RR |
762 | { |
763 | struct virtio_blk *vblk = vdev->priv; | |
5087a50e | 764 | int index = vblk->index; |
f4953fe6 | 765 | int refc; |
e467cde2 | 766 | |
cc74f719 MT |
767 | /* Make sure no work handler is accessing the device. */ |
768 | flush_work(&vblk->config_work); | |
7a7c924c | 769 | |
02e2b124 | 770 | del_gendisk(vblk->disk); |
483001c7 | 771 | blk_cleanup_queue(vblk->disk->queue); |
02e2b124 | 772 | |
24d2f903 CH |
773 | blk_mq_free_tag_set(&vblk->tag_set); |
774 | ||
6e5aa7ef RR |
775 | /* Stop all the virtqueues. */ |
776 | vdev->config->reset(vdev); | |
777 | ||
f4953fe6 | 778 | refc = atomic_read(&disk_to_dev(vblk->disk)->kobj.kref.refcount); |
e467cde2 | 779 | put_disk(vblk->disk); |
d2a7ddda | 780 | vdev->config->del_vqs(vdev); |
6a27b656 | 781 | kfree(vblk->vqs); |
e467cde2 | 782 | kfree(vblk); |
f4953fe6 AG |
783 | |
784 | /* Only free device id if we don't have any users */ | |
785 | if (refc == 1) | |
786 | ida_simple_remove(&vd_index_ida, index); | |
e467cde2 RR |
787 | } |
788 | ||
89107000 | 789 | #ifdef CONFIG_PM_SLEEP |
f8fb5bc2 AS |
790 | static int virtblk_freeze(struct virtio_device *vdev) |
791 | { | |
792 | struct virtio_blk *vblk = vdev->priv; | |
793 | ||
794 | /* Ensure we don't receive any more interrupts */ | |
795 | vdev->config->reset(vdev); | |
796 | ||
cc74f719 | 797 | /* Make sure no work handler is accessing the device. */ |
f8fb5bc2 AS |
798 | flush_work(&vblk->config_work); |
799 | ||
1cf7e9c6 | 800 | blk_mq_stop_hw_queues(vblk->disk->queue); |
f8fb5bc2 AS |
801 | |
802 | vdev->config->del_vqs(vdev); | |
803 | return 0; | |
804 | } | |
805 | ||
806 | static int virtblk_restore(struct virtio_device *vdev) | |
807 | { | |
808 | struct virtio_blk *vblk = vdev->priv; | |
809 | int ret; | |
810 | ||
f8fb5bc2 | 811 | ret = init_vq(vdev->priv); |
6d62c37f MT |
812 | if (ret) |
813 | return ret; | |
814 | ||
815 | virtio_device_ready(vdev); | |
1cf7e9c6 | 816 | |
6d62c37f MT |
817 | blk_mq_start_stopped_hw_queues(vblk->disk->queue, true); |
818 | return 0; | |
f8fb5bc2 AS |
819 | } |
820 | #endif | |
821 | ||
47483e25 | 822 | static const struct virtio_device_id id_table[] = { |
e467cde2 RR |
823 | { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID }, |
824 | { 0 }, | |
825 | }; | |
826 | ||
19c1c5a6 | 827 | static unsigned int features_legacy[] = { |
02c42b7a TH |
828 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, |
829 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, VIRTIO_BLK_F_SCSI, | |
6a27b656 ML |
830 | VIRTIO_BLK_F_WCE, VIRTIO_BLK_F_TOPOLOGY, VIRTIO_BLK_F_CONFIG_WCE, |
831 | VIRTIO_BLK_F_MQ, | |
19c1c5a6 MT |
832 | } |
833 | ; | |
834 | static unsigned int features[] = { | |
835 | VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX, VIRTIO_BLK_F_GEOMETRY, | |
836 | VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE, | |
837 | VIRTIO_BLK_F_TOPOLOGY, | |
838 | VIRTIO_BLK_F_MQ, | |
839 | VIRTIO_F_VERSION_1, | |
c45a6816 RR |
840 | }; |
841 | ||
8d85fce7 | 842 | static struct virtio_driver virtio_blk = { |
19c1c5a6 MT |
843 | .feature_table = features, |
844 | .feature_table_size = ARRAY_SIZE(features), | |
845 | .feature_table_legacy = features_legacy, | |
846 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
847 | .driver.name = KBUILD_MODNAME, | |
848 | .driver.owner = THIS_MODULE, | |
849 | .id_table = id_table, | |
850 | .probe = virtblk_probe, | |
851 | .remove = virtblk_remove, | |
852 | .config_changed = virtblk_config_changed, | |
89107000 | 853 | #ifdef CONFIG_PM_SLEEP |
19c1c5a6 MT |
854 | .freeze = virtblk_freeze, |
855 | .restore = virtblk_restore, | |
f8fb5bc2 | 856 | #endif |
e467cde2 RR |
857 | }; |
858 | ||
859 | static int __init init(void) | |
860 | { | |
7a7c924c CH |
861 | int error; |
862 | ||
863 | virtblk_wq = alloc_workqueue("virtio-blk", 0, 0); | |
864 | if (!virtblk_wq) | |
865 | return -ENOMEM; | |
866 | ||
4f3bf19c | 867 | major = register_blkdev(0, "virtblk"); |
7a7c924c CH |
868 | if (major < 0) { |
869 | error = major; | |
870 | goto out_destroy_workqueue; | |
871 | } | |
872 | ||
873 | error = register_virtio_driver(&virtio_blk); | |
874 | if (error) | |
875 | goto out_unregister_blkdev; | |
876 | return 0; | |
877 | ||
878 | out_unregister_blkdev: | |
879 | unregister_blkdev(major, "virtblk"); | |
880 | out_destroy_workqueue: | |
881 | destroy_workqueue(virtblk_wq); | |
882 | return error; | |
e467cde2 RR |
883 | } |
884 | ||
885 | static void __exit fini(void) | |
886 | { | |
4f3bf19c | 887 | unregister_blkdev(major, "virtblk"); |
e467cde2 | 888 | unregister_virtio_driver(&virtio_blk); |
7a7c924c | 889 | destroy_workqueue(virtblk_wq); |
e467cde2 RR |
890 | } |
891 | module_init(init); | |
892 | module_exit(fini); | |
893 | ||
894 | MODULE_DEVICE_TABLE(virtio, id_table); | |
895 | MODULE_DESCRIPTION("Virtio block driver"); | |
896 | MODULE_LICENSE("GPL"); |