]>
Commit | Line | Data |
---|---|---|
6e02c38d AL |
1 | /* |
2 | * Virtio Block Device | |
3 | * | |
4 | * Copyright IBM, Corp. 2007 | |
5 | * | |
6 | * Authors: | |
7 | * Anthony Liguori <aliguori@us.ibm.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
10 | * the COPYING file in the top-level directory. | |
11 | * | |
12 | */ | |
13 | ||
5a61cb60 | 14 | #include "qemu-common.h" |
d75d25e3 | 15 | #include "qemu-error.h" |
6d519a5f | 16 | #include "trace.h" |
9db1c0f7 | 17 | #include "hw/block-common.h" |
2446333c | 18 | #include "blockdev.h" |
6e02c38d | 19 | #include "virtio-blk.h" |
5bb23927 | 20 | #include "scsi-defs.h" |
1063b8b1 CH |
21 | #ifdef __linux__ |
22 | # include <scsi/sg.h> | |
23 | #endif | |
6e02c38d AL |
24 | |
25 | typedef struct VirtIOBlock | |
26 | { | |
27 | VirtIODevice vdev; | |
28 | BlockDriverState *bs; | |
29 | VirtQueue *vq; | |
869a5c6d | 30 | void *rq; |
213189ab | 31 | QEMUBH *bh; |
9752c371 | 32 | BlockConf *conf; |
12c5674b | 33 | VirtIOBlkConf *blk; |
8cfacf07 | 34 | unsigned short sector_mask; |
9d0d3138 | 35 | DeviceState *qdev; |
6e02c38d AL |
36 | } VirtIOBlock; |
37 | ||
38 | static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev) | |
39 | { | |
40 | return (VirtIOBlock *)vdev; | |
41 | } | |
42 | ||
43 | typedef struct VirtIOBlockReq | |
44 | { | |
45 | VirtIOBlock *dev; | |
46 | VirtQueueElement elem; | |
47 | struct virtio_blk_inhdr *in; | |
48 | struct virtio_blk_outhdr *out; | |
1063b8b1 | 49 | struct virtio_scsi_inhdr *scsi; |
d28a1b6e | 50 | QEMUIOVector qiov; |
869a5c6d | 51 | struct VirtIOBlockReq *next; |
a597e79c | 52 | BlockAcctCookie acct; |
6e02c38d AL |
53 | } VirtIOBlockReq; |
54 | ||
869a5c6d AL |
55 | static void virtio_blk_req_complete(VirtIOBlockReq *req, int status) |
56 | { | |
57 | VirtIOBlock *s = req->dev; | |
58 | ||
6d519a5f SH |
59 | trace_virtio_blk_req_complete(req, status); |
60 | ||
92e3c2a3 | 61 | stb_p(&req->in->status, status); |
d28a1b6e | 62 | virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in)); |
869a5c6d | 63 | virtio_notify(&s->vdev, s->vq); |
869a5c6d AL |
64 | } |
65 | ||
f35d68f0 | 66 | static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error, |
1ceee0d5 | 67 | bool is_read) |
869a5c6d | 68 | { |
3e1caa5f | 69 | BlockErrorAction action = bdrv_get_error_action(req->dev->bs, is_read, error); |
869a5c6d AL |
70 | VirtIOBlock *s = req->dev; |
71 | ||
3e1caa5f | 72 | if (action == BDRV_ACTION_STOP) { |
869a5c6d AL |
73 | req->next = s->rq; |
74 | s->rq = req; | |
3e1caa5f | 75 | } else if (action == BDRV_ACTION_REPORT) { |
869a5c6d | 76 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); |
a597e79c CH |
77 | bdrv_acct_done(s->bs, &req->acct); |
78 | g_free(req); | |
869a5c6d AL |
79 | } |
80 | ||
3e1caa5f PB |
81 | bdrv_error_action(s->bs, action, is_read, error); |
82 | return action != BDRV_ACTION_IGNORE; | |
869a5c6d AL |
83 | } |
84 | ||
6e02c38d AL |
85 | static void virtio_blk_rw_complete(void *opaque, int ret) |
86 | { | |
87 | VirtIOBlockReq *req = opaque; | |
6e02c38d | 88 | |
6d519a5f SH |
89 | trace_virtio_blk_rw_complete(req, ret); |
90 | ||
f35d68f0 | 91 | if (ret) { |
1ceee0d5 | 92 | bool is_read = !(ldl_p(&req->out->type) & VIRTIO_BLK_T_OUT); |
f35d68f0 | 93 | if (virtio_blk_handle_rw_error(req, -ret, is_read)) |
869a5c6d | 94 | return; |
6e02c38d AL |
95 | } |
96 | ||
f35d68f0 | 97 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); |
a597e79c CH |
98 | bdrv_acct_done(req->dev->bs, &req->acct); |
99 | g_free(req); | |
869a5c6d | 100 | } |
6e02c38d | 101 | |
aa659be3 CH |
102 | static void virtio_blk_flush_complete(void *opaque, int ret) |
103 | { | |
104 | VirtIOBlockReq *req = opaque; | |
105 | ||
8c269b54 KW |
106 | if (ret) { |
107 | if (virtio_blk_handle_rw_error(req, -ret, 0)) { | |
108 | return; | |
109 | } | |
110 | } | |
111 | ||
112 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); | |
a597e79c CH |
113 | bdrv_acct_done(req->dev->bs, &req->acct); |
114 | g_free(req); | |
aa659be3 CH |
115 | } |
116 | ||
869a5c6d AL |
117 | static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s) |
118 | { | |
7267c094 | 119 | VirtIOBlockReq *req = g_malloc(sizeof(*req)); |
487414f1 | 120 | req->dev = s; |
de6c8042 SH |
121 | req->qiov.size = 0; |
122 | req->next = NULL; | |
869a5c6d | 123 | return req; |
6e02c38d AL |
124 | } |
125 | ||
126 | static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s) | |
127 | { | |
869a5c6d | 128 | VirtIOBlockReq *req = virtio_blk_alloc_request(s); |
6e02c38d | 129 | |
869a5c6d AL |
130 | if (req != NULL) { |
131 | if (!virtqueue_pop(s->vq, &req->elem)) { | |
7267c094 | 132 | g_free(req); |
869a5c6d AL |
133 | return NULL; |
134 | } | |
6e02c38d AL |
135 | } |
136 | ||
137 | return req; | |
138 | } | |
139 | ||
1063b8b1 CH |
140 | static void virtio_blk_handle_scsi(VirtIOBlockReq *req) |
141 | { | |
47ce9ef7 | 142 | #ifdef __linux__ |
4277906d | 143 | int ret; |
1063b8b1 | 144 | int i; |
47ce9ef7 SW |
145 | #endif |
146 | int status = VIRTIO_BLK_S_OK; | |
1063b8b1 CH |
147 | |
148 | /* | |
149 | * We require at least one output segment each for the virtio_blk_outhdr | |
150 | * and the SCSI command block. | |
151 | * | |
152 | * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr | |
153 | * and the sense buffer pointer in the input segments. | |
154 | */ | |
155 | if (req->elem.out_num < 2 || req->elem.in_num < 3) { | |
156 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); | |
a597e79c | 157 | g_free(req); |
1063b8b1 CH |
158 | return; |
159 | } | |
160 | ||
161 | /* | |
f34e73cd PB |
162 | * The scsi inhdr is placed in the second-to-last input segment, just |
163 | * before the regular inhdr. | |
1063b8b1 | 164 | */ |
f34e73cd PB |
165 | req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base; |
166 | ||
a6c5c84a | 167 | if (!req->dev->blk->scsi) { |
f34e73cd PB |
168 | status = VIRTIO_BLK_S_UNSUPP; |
169 | goto fail; | |
1063b8b1 CH |
170 | } |
171 | ||
172 | /* | |
f34e73cd | 173 | * No support for bidirection commands yet. |
1063b8b1 | 174 | */ |
f34e73cd PB |
175 | if (req->elem.out_num > 2 && req->elem.in_num > 3) { |
176 | status = VIRTIO_BLK_S_UNSUPP; | |
177 | goto fail; | |
178 | } | |
1063b8b1 | 179 | |
f34e73cd PB |
180 | #ifdef __linux__ |
181 | struct sg_io_hdr hdr; | |
1063b8b1 CH |
182 | memset(&hdr, 0, sizeof(struct sg_io_hdr)); |
183 | hdr.interface_id = 'S'; | |
184 | hdr.cmd_len = req->elem.out_sg[1].iov_len; | |
185 | hdr.cmdp = req->elem.out_sg[1].iov_base; | |
186 | hdr.dxfer_len = 0; | |
187 | ||
188 | if (req->elem.out_num > 2) { | |
189 | /* | |
190 | * If there are more than the minimally required 2 output segments | |
191 | * there is write payload starting from the third iovec. | |
192 | */ | |
193 | hdr.dxfer_direction = SG_DXFER_TO_DEV; | |
194 | hdr.iovec_count = req->elem.out_num - 2; | |
195 | ||
196 | for (i = 0; i < hdr.iovec_count; i++) | |
197 | hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len; | |
198 | ||
199 | hdr.dxferp = req->elem.out_sg + 2; | |
200 | ||
201 | } else if (req->elem.in_num > 3) { | |
202 | /* | |
203 | * If we have more than 3 input segments the guest wants to actually | |
204 | * read data. | |
205 | */ | |
206 | hdr.dxfer_direction = SG_DXFER_FROM_DEV; | |
207 | hdr.iovec_count = req->elem.in_num - 3; | |
208 | for (i = 0; i < hdr.iovec_count; i++) | |
209 | hdr.dxfer_len += req->elem.in_sg[i].iov_len; | |
210 | ||
211 | hdr.dxferp = req->elem.in_sg; | |
1063b8b1 CH |
212 | } else { |
213 | /* | |
214 | * Some SCSI commands don't actually transfer any data. | |
215 | */ | |
216 | hdr.dxfer_direction = SG_DXFER_NONE; | |
217 | } | |
218 | ||
219 | hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base; | |
220 | hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len; | |
1063b8b1 CH |
221 | |
222 | ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr); | |
223 | if (ret) { | |
224 | status = VIRTIO_BLK_S_UNSUPP; | |
f34e73cd | 225 | goto fail; |
1063b8b1 CH |
226 | } |
227 | ||
5bb23927 PB |
228 | /* |
229 | * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi) | |
230 | * clear the masked_status field [hence status gets cleared too, see | |
231 | * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED | |
232 | * status has occurred. However they do set DRIVER_SENSE in driver_status | |
233 | * field. Also a (sb_len_wr > 0) indicates there is a sense buffer. | |
234 | */ | |
235 | if (hdr.status == 0 && hdr.sb_len_wr > 0) { | |
236 | hdr.status = CHECK_CONDITION; | |
237 | } | |
238 | ||
239 | stl_p(&req->scsi->errors, | |
240 | hdr.status | (hdr.msg_status << 8) | | |
241 | (hdr.host_status << 16) | (hdr.driver_status << 24)); | |
92e3c2a3 AJ |
242 | stl_p(&req->scsi->residual, hdr.resid); |
243 | stl_p(&req->scsi->sense_len, hdr.sb_len_wr); | |
244 | stl_p(&req->scsi->data_len, hdr.dxfer_len); | |
1063b8b1 CH |
245 | |
246 | virtio_blk_req_complete(req, status); | |
a597e79c | 247 | g_free(req); |
730a9c53 | 248 | return; |
1063b8b1 | 249 | #else |
f34e73cd PB |
250 | abort(); |
251 | #endif | |
252 | ||
253 | fail: | |
254 | /* Just put anything nonzero so that the ioctl fails in the guest. */ | |
255 | stl_p(&req->scsi->errors, 255); | |
256 | virtio_blk_req_complete(req, status); | |
a597e79c | 257 | g_free(req); |
1063b8b1 | 258 | } |
1063b8b1 | 259 | |
c20fd872 CH |
260 | typedef struct MultiReqBuffer { |
261 | BlockRequest blkreq[32]; | |
262 | unsigned int num_writes; | |
263 | } MultiReqBuffer; | |
264 | ||
265 | static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb) | |
869a5c6d | 266 | { |
91553dcc | 267 | int i, ret; |
91553dcc | 268 | |
c20fd872 CH |
269 | if (!mrb->num_writes) { |
270 | return; | |
271 | } | |
272 | ||
273 | ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes); | |
91553dcc | 274 | if (ret != 0) { |
c20fd872 CH |
275 | for (i = 0; i < mrb->num_writes; i++) { |
276 | if (mrb->blkreq[i].error) { | |
277 | virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO); | |
91553dcc KW |
278 | } |
279 | } | |
280 | } | |
c20fd872 CH |
281 | |
282 | mrb->num_writes = 0; | |
91553dcc | 283 | } |
87b245db | 284 | |
c20fd872 | 285 | static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
aa659be3 | 286 | { |
a597e79c CH |
287 | bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH); |
288 | ||
618fbb84 CH |
289 | /* |
290 | * Make sure all outstanding writes are posted to the backing device. | |
291 | */ | |
c20fd872 | 292 | virtio_submit_multiwrite(req->dev->bs, mrb); |
ad54ae80 | 293 | bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req); |
aa659be3 CH |
294 | } |
295 | ||
c20fd872 | 296 | static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
91553dcc | 297 | { |
c20fd872 | 298 | BlockRequest *blkreq; |
92e3c2a3 | 299 | uint64_t sector; |
c20fd872 | 300 | |
92e3c2a3 | 301 | sector = ldq_p(&req->out->sector); |
6d519a5f | 302 | |
a597e79c CH |
303 | bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE); |
304 | ||
92e3c2a3 AJ |
305 | trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512); |
306 | ||
307 | if (sector & req->dev->sector_mask) { | |
8cfacf07 CH |
308 | virtio_blk_rw_complete(req, -EIO); |
309 | return; | |
310 | } | |
52c05023 CH |
311 | if (req->qiov.size % req->dev->conf->logical_block_size) { |
312 | virtio_blk_rw_complete(req, -EIO); | |
313 | return; | |
314 | } | |
8cfacf07 | 315 | |
c20fd872 CH |
316 | if (mrb->num_writes == 32) { |
317 | virtio_submit_multiwrite(req->dev->bs, mrb); | |
87b245db | 318 | } |
91553dcc | 319 | |
c20fd872 | 320 | blkreq = &mrb->blkreq[mrb->num_writes]; |
92e3c2a3 | 321 | blkreq->sector = sector; |
c20fd872 CH |
322 | blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE; |
323 | blkreq->qiov = &req->qiov; | |
324 | blkreq->cb = virtio_blk_rw_complete; | |
325 | blkreq->opaque = req; | |
326 | blkreq->error = 0; | |
91553dcc | 327 | |
c20fd872 | 328 | mrb->num_writes++; |
d28a1b6e | 329 | } |
869a5c6d | 330 | |
d28a1b6e AL |
331 | static void virtio_blk_handle_read(VirtIOBlockReq *req) |
332 | { | |
92e3c2a3 AJ |
333 | uint64_t sector; |
334 | ||
335 | sector = ldq_p(&req->out->sector); | |
87b245db | 336 | |
a597e79c CH |
337 | bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ); |
338 | ||
81b6b9fa SH |
339 | trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512); |
340 | ||
92e3c2a3 | 341 | if (sector & req->dev->sector_mask) { |
8cfacf07 CH |
342 | virtio_blk_rw_complete(req, -EIO); |
343 | return; | |
344 | } | |
52c05023 CH |
345 | if (req->qiov.size % req->dev->conf->logical_block_size) { |
346 | virtio_blk_rw_complete(req, -EIO); | |
347 | return; | |
348 | } | |
ad54ae80 PB |
349 | bdrv_aio_readv(req->dev->bs, sector, &req->qiov, |
350 | req->qiov.size / BDRV_SECTOR_SIZE, | |
351 | virtio_blk_rw_complete, req); | |
869a5c6d AL |
352 | } |
353 | ||
bc6694d4 KW |
354 | static void virtio_blk_handle_request(VirtIOBlockReq *req, |
355 | MultiReqBuffer *mrb) | |
356 | { | |
92e3c2a3 AJ |
357 | uint32_t type; |
358 | ||
bc6694d4 | 359 | if (req->elem.out_num < 1 || req->elem.in_num < 1) { |
870cef1d | 360 | error_report("virtio-blk missing headers"); |
bc6694d4 KW |
361 | exit(1); |
362 | } | |
363 | ||
364 | if (req->elem.out_sg[0].iov_len < sizeof(*req->out) || | |
365 | req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) { | |
870cef1d | 366 | error_report("virtio-blk header not in correct element"); |
bc6694d4 KW |
367 | exit(1); |
368 | } | |
369 | ||
370 | req->out = (void *)req->elem.out_sg[0].iov_base; | |
371 | req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base; | |
372 | ||
92e3c2a3 AJ |
373 | type = ldl_p(&req->out->type); |
374 | ||
375 | if (type & VIRTIO_BLK_T_FLUSH) { | |
c20fd872 | 376 | virtio_blk_handle_flush(req, mrb); |
92e3c2a3 | 377 | } else if (type & VIRTIO_BLK_T_SCSI_CMD) { |
bc6694d4 | 378 | virtio_blk_handle_scsi(req); |
92e3c2a3 | 379 | } else if (type & VIRTIO_BLK_T_GET_ID) { |
2930b313 | 380 | VirtIOBlock *s = req->dev; |
381 | ||
a8686a9b MA |
382 | /* |
383 | * NB: per existing s/n string convention the string is | |
384 | * terminated by '\0' only when shorter than buffer. | |
385 | */ | |
386 | strncpy(req->elem.in_sg[0].iov_base, | |
12c5674b | 387 | s->blk->serial ? s->blk->serial : "", |
a8686a9b | 388 | MIN(req->elem.in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES)); |
2930b313 | 389 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); |
a597e79c | 390 | g_free(req); |
92e3c2a3 | 391 | } else if (type & VIRTIO_BLK_T_OUT) { |
bc6694d4 KW |
392 | qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1], |
393 | req->elem.out_num - 1); | |
c20fd872 | 394 | virtio_blk_handle_write(req, mrb); |
bc6694d4 KW |
395 | } else { |
396 | qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0], | |
397 | req->elem.in_num - 1); | |
398 | virtio_blk_handle_read(req); | |
399 | } | |
400 | } | |
401 | ||
6e02c38d AL |
402 | static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq) |
403 | { | |
404 | VirtIOBlock *s = to_virtio_blk(vdev); | |
405 | VirtIOBlockReq *req; | |
bc6694d4 KW |
406 | MultiReqBuffer mrb = { |
407 | .num_writes = 0, | |
bc6694d4 | 408 | }; |
6e02c38d AL |
409 | |
410 | while ((req = virtio_blk_get_request(s))) { | |
bc6694d4 | 411 | virtio_blk_handle_request(req, &mrb); |
6e02c38d | 412 | } |
91553dcc | 413 | |
c20fd872 | 414 | virtio_submit_multiwrite(s->bs, &mrb); |
91553dcc | 415 | |
6e02c38d AL |
416 | /* |
417 | * FIXME: Want to check for completions before returning to guest mode, | |
418 | * so cached reads and writes are reported as quickly as possible. But | |
419 | * that should be done in the generic block layer. | |
420 | */ | |
421 | } | |
422 | ||
213189ab | 423 | static void virtio_blk_dma_restart_bh(void *opaque) |
869a5c6d AL |
424 | { |
425 | VirtIOBlock *s = opaque; | |
426 | VirtIOBlockReq *req = s->rq; | |
f1b52868 KW |
427 | MultiReqBuffer mrb = { |
428 | .num_writes = 0, | |
f1b52868 | 429 | }; |
869a5c6d | 430 | |
213189ab MA |
431 | qemu_bh_delete(s->bh); |
432 | s->bh = NULL; | |
869a5c6d AL |
433 | |
434 | s->rq = NULL; | |
435 | ||
436 | while (req) { | |
f1b52868 | 437 | virtio_blk_handle_request(req, &mrb); |
869a5c6d AL |
438 | req = req->next; |
439 | } | |
f1b52868 | 440 | |
c20fd872 | 441 | virtio_submit_multiwrite(s->bs, &mrb); |
869a5c6d AL |
442 | } |
443 | ||
1dfb4dd9 LC |
444 | static void virtio_blk_dma_restart_cb(void *opaque, int running, |
445 | RunState state) | |
213189ab MA |
446 | { |
447 | VirtIOBlock *s = opaque; | |
448 | ||
449 | if (!running) | |
450 | return; | |
451 | ||
452 | if (!s->bh) { | |
453 | s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s); | |
454 | qemu_bh_schedule(s->bh); | |
455 | } | |
456 | } | |
457 | ||
6e02c38d AL |
458 | static void virtio_blk_reset(VirtIODevice *vdev) |
459 | { | |
460 | /* | |
461 | * This should cancel pending requests, but can't do nicely until there | |
462 | * are per-device request lists. | |
463 | */ | |
922453bc | 464 | bdrv_drain_all(); |
6e02c38d AL |
465 | } |
466 | ||
bf011293 | 467 | /* coalesce internal state, copy to pci i/o region 0 |
468 | */ | |
6e02c38d AL |
469 | static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config) |
470 | { | |
471 | VirtIOBlock *s = to_virtio_blk(vdev); | |
472 | struct virtio_blk_config blkcfg; | |
473 | uint64_t capacity; | |
3a395142 | 474 | int blk_size = s->conf->logical_block_size; |
6e02c38d AL |
475 | |
476 | bdrv_get_geometry(s->bs, &capacity); | |
5c5dafdc | 477 | memset(&blkcfg, 0, sizeof(blkcfg)); |
6e02c38d AL |
478 | stq_raw(&blkcfg.capacity, capacity); |
479 | stl_raw(&blkcfg.seg_max, 128 - 2); | |
e63e7fde | 480 | stw_raw(&blkcfg.cylinders, s->conf->cyls); |
3a395142 PB |
481 | stl_raw(&blkcfg.blk_size, blk_size); |
482 | stw_raw(&blkcfg.min_io_size, s->conf->min_io_size / blk_size); | |
483 | stw_raw(&blkcfg.opt_io_size, s->conf->opt_io_size / blk_size); | |
e63e7fde | 484 | blkcfg.heads = s->conf->heads; |
136be99e CB |
485 | /* |
486 | * We must ensure that the block device capacity is a multiple of | |
487 | * the logical block size. If that is not the case, lets use | |
488 | * sector_mask to adopt the geometry to have a correct picture. | |
489 | * For those devices where the capacity is ok for the given geometry | |
490 | * we dont touch the sector value of the geometry, since some devices | |
491 | * (like s390 dasd) need a specific value. Here the capacity is already | |
492 | * cyls*heads*secs*blk_size and the sector value is not block size | |
493 | * divided by 512 - instead it is the amount of blk_size blocks | |
494 | * per track (cylinder). | |
495 | */ | |
e63e7fde MA |
496 | if (bdrv_getlength(s->bs) / s->conf->heads / s->conf->secs % blk_size) { |
497 | blkcfg.sectors = s->conf->secs & ~s->sector_mask; | |
136be99e | 498 | } else { |
e63e7fde | 499 | blkcfg.sectors = s->conf->secs; |
136be99e | 500 | } |
c7085da7 | 501 | blkcfg.size_max = 0; |
9752c371 CH |
502 | blkcfg.physical_block_exp = get_physical_block_exp(s->conf); |
503 | blkcfg.alignment_offset = 0; | |
13e3dce0 | 504 | blkcfg.wce = bdrv_enable_write_cache(s->bs); |
37d5ddd6 | 505 | memcpy(config, &blkcfg, sizeof(struct virtio_blk_config)); |
6e02c38d AL |
506 | } |
507 | ||
13e3dce0 PB |
508 | static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config) |
509 | { | |
510 | VirtIOBlock *s = to_virtio_blk(vdev); | |
511 | struct virtio_blk_config blkcfg; | |
512 | ||
513 | memcpy(&blkcfg, config, sizeof(blkcfg)); | |
514 | bdrv_set_enable_write_cache(s->bs, blkcfg.wce != 0); | |
515 | } | |
516 | ||
8172539d | 517 | static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features) |
6e02c38d | 518 | { |
bf011293 | 519 | VirtIOBlock *s = to_virtio_blk(vdev); |
1063b8b1 CH |
520 | |
521 | features |= (1 << VIRTIO_BLK_F_SEG_MAX); | |
522 | features |= (1 << VIRTIO_BLK_F_GEOMETRY); | |
9752c371 | 523 | features |= (1 << VIRTIO_BLK_F_TOPOLOGY); |
8cfacf07 | 524 | features |= (1 << VIRTIO_BLK_F_BLK_SIZE); |
a6c5c84a | 525 | features |= (1 << VIRTIO_BLK_F_SCSI); |
aa659be3 CH |
526 | |
527 | if (bdrv_enable_write_cache(s->bs)) | |
13e3dce0 PB |
528 | features |= (1 << VIRTIO_BLK_F_WCE); |
529 | ||
c79662f7 NS |
530 | if (bdrv_is_read_only(s->bs)) |
531 | features |= 1 << VIRTIO_BLK_F_RO; | |
1063b8b1 CH |
532 | |
533 | return features; | |
6e02c38d AL |
534 | } |
535 | ||
9315cbfd PB |
536 | static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status) |
537 | { | |
538 | VirtIOBlock *s = to_virtio_blk(vdev); | |
539 | uint32_t features; | |
540 | ||
541 | if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) { | |
542 | return; | |
543 | } | |
544 | ||
545 | features = vdev->guest_features; | |
546 | bdrv_set_enable_write_cache(s->bs, !!(features & (1 << VIRTIO_BLK_F_WCE))); | |
547 | } | |
548 | ||
6e02c38d AL |
549 | static void virtio_blk_save(QEMUFile *f, void *opaque) |
550 | { | |
551 | VirtIOBlock *s = opaque; | |
869a5c6d AL |
552 | VirtIOBlockReq *req = s->rq; |
553 | ||
6e02c38d | 554 | virtio_save(&s->vdev, f); |
869a5c6d AL |
555 | |
556 | while (req) { | |
557 | qemu_put_sbyte(f, 1); | |
558 | qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem)); | |
559 | req = req->next; | |
560 | } | |
561 | qemu_put_sbyte(f, 0); | |
6e02c38d AL |
562 | } |
563 | ||
564 | static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id) | |
565 | { | |
566 | VirtIOBlock *s = opaque; | |
2a633c46 | 567 | int ret; |
6e02c38d | 568 | |
869a5c6d | 569 | if (version_id != 2) |
6e02c38d AL |
570 | return -EINVAL; |
571 | ||
2a633c46 OW |
572 | ret = virtio_load(&s->vdev, f); |
573 | if (ret) { | |
574 | return ret; | |
575 | } | |
576 | ||
869a5c6d AL |
577 | while (qemu_get_sbyte(f)) { |
578 | VirtIOBlockReq *req = virtio_blk_alloc_request(s); | |
579 | qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem)); | |
580 | req->next = s->rq; | |
20a81e4d | 581 | s->rq = req; |
b6a4805b KW |
582 | |
583 | virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr, | |
584 | req->elem.in_num, 1); | |
585 | virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr, | |
586 | req->elem.out_num, 0); | |
869a5c6d | 587 | } |
6e02c38d AL |
588 | |
589 | return 0; | |
590 | } | |
591 | ||
145feb17 | 592 | static void virtio_blk_resize(void *opaque) |
e5051fc7 CH |
593 | { |
594 | VirtIOBlock *s = opaque; | |
595 | ||
145feb17 | 596 | virtio_notify_config(&s->vdev); |
e5051fc7 CH |
597 | } |
598 | ||
0e49de52 | 599 | static const BlockDevOps virtio_block_ops = { |
145feb17 | 600 | .resize_cb = virtio_blk_resize, |
0e49de52 MA |
601 | }; |
602 | ||
12c5674b | 603 | VirtIODevice *virtio_blk_init(DeviceState *dev, VirtIOBlkConf *blk) |
6e02c38d AL |
604 | { |
605 | VirtIOBlock *s; | |
6e02c38d | 606 | static int virtio_blk_id; |
cf21e106 | 607 | |
12c5674b | 608 | if (!blk->conf.bs) { |
6a84cb1f | 609 | error_report("drive property not set"); |
d75d25e3 MA |
610 | return NULL; |
611 | } | |
12c5674b | 612 | if (!bdrv_is_inserted(blk->conf.bs)) { |
98f28ad7 MA |
613 | error_report("Device needs media, but drive is empty"); |
614 | return NULL; | |
615 | } | |
d75d25e3 | 616 | |
911525db | 617 | blkconf_serial(&blk->conf, &blk->serial); |
b7eb0c9f MA |
618 | if (blkconf_geometry(&blk->conf, NULL, 65535, 255, 255) < 0) { |
619 | return NULL; | |
620 | } | |
a8686a9b | 621 | |
53c25cea | 622 | s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK, |
37d5ddd6 | 623 | sizeof(struct virtio_blk_config), |
53c25cea | 624 | sizeof(VirtIOBlock)); |
6e02c38d AL |
625 | |
626 | s->vdev.get_config = virtio_blk_update_config; | |
13e3dce0 | 627 | s->vdev.set_config = virtio_blk_set_config; |
6e02c38d | 628 | s->vdev.get_features = virtio_blk_get_features; |
9315cbfd | 629 | s->vdev.set_status = virtio_blk_set_status; |
6e02c38d | 630 | s->vdev.reset = virtio_blk_reset; |
12c5674b PB |
631 | s->bs = blk->conf.bs; |
632 | s->conf = &blk->conf; | |
633 | s->blk = blk; | |
869a5c6d | 634 | s->rq = NULL; |
1573a35d | 635 | s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1; |
e63e7fde | 636 | |
6e02c38d AL |
637 | s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output); |
638 | ||
869a5c6d | 639 | qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s); |
9d0d3138 | 640 | s->qdev = dev; |
0be71e32 | 641 | register_savevm(dev, "virtio-blk", virtio_blk_id++, 2, |
6e02c38d | 642 | virtio_blk_save, virtio_blk_load, s); |
0e49de52 | 643 | bdrv_set_dev_ops(s->bs, &virtio_block_ops, s); |
12c5674b | 644 | bdrv_set_buffer_alignment(s->bs, s->conf->logical_block_size); |
6e02c38d | 645 | |
af239a62 | 646 | bdrv_iostatus_enable(s->bs); |
12c5674b | 647 | add_boot_device_path(s->conf->bootindex, dev, "/disk@0,0"); |
1ca4d09a | 648 | |
53c25cea | 649 | return &s->vdev; |
6e02c38d | 650 | } |
9d0d3138 AW |
651 | |
652 | void virtio_blk_exit(VirtIODevice *vdev) | |
653 | { | |
654 | VirtIOBlock *s = to_virtio_blk(vdev); | |
655 | unregister_savevm(s->qdev, "virtio-blk", s); | |
0e47931b | 656 | blockdev_mark_auto_del(s->bs); |
d92551f2 | 657 | virtio_cleanup(vdev); |
9d0d3138 | 658 | } |