]>
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" |
2446333c | 17 | #include "blockdev.h" |
6e02c38d | 18 | #include "virtio-blk.h" |
5bb23927 | 19 | #include "scsi-defs.h" |
1063b8b1 CH |
20 | #ifdef __linux__ |
21 | # include <scsi/sg.h> | |
22 | #endif | |
6e02c38d AL |
23 | |
24 | typedef struct VirtIOBlock | |
25 | { | |
26 | VirtIODevice vdev; | |
27 | BlockDriverState *bs; | |
28 | VirtQueue *vq; | |
869a5c6d | 29 | void *rq; |
213189ab | 30 | QEMUBH *bh; |
9752c371 | 31 | BlockConf *conf; |
12c5674b | 32 | VirtIOBlkConf *blk; |
8cfacf07 | 33 | unsigned short sector_mask; |
9d0d3138 | 34 | DeviceState *qdev; |
6e02c38d AL |
35 | } VirtIOBlock; |
36 | ||
37 | static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev) | |
38 | { | |
39 | return (VirtIOBlock *)vdev; | |
40 | } | |
41 | ||
42 | typedef struct VirtIOBlockReq | |
43 | { | |
44 | VirtIOBlock *dev; | |
45 | VirtQueueElement elem; | |
46 | struct virtio_blk_inhdr *in; | |
47 | struct virtio_blk_outhdr *out; | |
1063b8b1 | 48 | struct virtio_scsi_inhdr *scsi; |
d28a1b6e | 49 | QEMUIOVector qiov; |
869a5c6d | 50 | struct VirtIOBlockReq *next; |
a597e79c | 51 | BlockAcctCookie acct; |
6e02c38d AL |
52 | } VirtIOBlockReq; |
53 | ||
869a5c6d AL |
54 | static void virtio_blk_req_complete(VirtIOBlockReq *req, int status) |
55 | { | |
56 | VirtIOBlock *s = req->dev; | |
57 | ||
6d519a5f SH |
58 | trace_virtio_blk_req_complete(req, status); |
59 | ||
92e3c2a3 | 60 | stb_p(&req->in->status, status); |
d28a1b6e | 61 | virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in)); |
869a5c6d | 62 | virtio_notify(&s->vdev, s->vq); |
869a5c6d AL |
63 | } |
64 | ||
f35d68f0 KW |
65 | static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error, |
66 | int is_read) | |
869a5c6d | 67 | { |
abd7f68d | 68 | BlockErrorAction action = bdrv_get_on_error(req->dev->bs, is_read); |
869a5c6d AL |
69 | VirtIOBlock *s = req->dev; |
70 | ||
eaa6c85f | 71 | if (action == BLOCK_ERR_IGNORE) { |
329c0a48 | 72 | bdrv_emit_qmp_error_event(s->bs, BDRV_ACTION_IGNORE, is_read); |
869a5c6d | 73 | return 0; |
eaa6c85f | 74 | } |
869a5c6d AL |
75 | |
76 | if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC) | |
77 | || action == BLOCK_ERR_STOP_ANY) { | |
78 | req->next = s->rq; | |
79 | s->rq = req; | |
329c0a48 | 80 | bdrv_emit_qmp_error_event(s->bs, BDRV_ACTION_STOP, is_read); |
0461d5a6 | 81 | vm_stop(RUN_STATE_IO_ERROR); |
af239a62 | 82 | bdrv_iostatus_set_err(s->bs, error); |
869a5c6d AL |
83 | } else { |
84 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); | |
a597e79c CH |
85 | bdrv_acct_done(s->bs, &req->acct); |
86 | g_free(req); | |
329c0a48 | 87 | bdrv_emit_qmp_error_event(s->bs, BDRV_ACTION_REPORT, is_read); |
869a5c6d AL |
88 | } |
89 | ||
90 | return 1; | |
91 | } | |
92 | ||
6e02c38d AL |
93 | static void virtio_blk_rw_complete(void *opaque, int ret) |
94 | { | |
95 | VirtIOBlockReq *req = opaque; | |
6e02c38d | 96 | |
6d519a5f SH |
97 | trace_virtio_blk_rw_complete(req, ret); |
98 | ||
f35d68f0 | 99 | if (ret) { |
92e3c2a3 | 100 | int is_read = !(ldl_p(&req->out->type) & VIRTIO_BLK_T_OUT); |
f35d68f0 | 101 | if (virtio_blk_handle_rw_error(req, -ret, is_read)) |
869a5c6d | 102 | return; |
6e02c38d AL |
103 | } |
104 | ||
f35d68f0 | 105 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); |
a597e79c CH |
106 | bdrv_acct_done(req->dev->bs, &req->acct); |
107 | g_free(req); | |
869a5c6d | 108 | } |
6e02c38d | 109 | |
aa659be3 CH |
110 | static void virtio_blk_flush_complete(void *opaque, int ret) |
111 | { | |
112 | VirtIOBlockReq *req = opaque; | |
113 | ||
8c269b54 KW |
114 | if (ret) { |
115 | if (virtio_blk_handle_rw_error(req, -ret, 0)) { | |
116 | return; | |
117 | } | |
118 | } | |
119 | ||
120 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); | |
a597e79c CH |
121 | bdrv_acct_done(req->dev->bs, &req->acct); |
122 | g_free(req); | |
aa659be3 CH |
123 | } |
124 | ||
869a5c6d AL |
125 | static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s) |
126 | { | |
7267c094 | 127 | VirtIOBlockReq *req = g_malloc(sizeof(*req)); |
487414f1 | 128 | req->dev = s; |
de6c8042 SH |
129 | req->qiov.size = 0; |
130 | req->next = NULL; | |
869a5c6d | 131 | return req; |
6e02c38d AL |
132 | } |
133 | ||
134 | static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s) | |
135 | { | |
869a5c6d | 136 | VirtIOBlockReq *req = virtio_blk_alloc_request(s); |
6e02c38d | 137 | |
869a5c6d AL |
138 | if (req != NULL) { |
139 | if (!virtqueue_pop(s->vq, &req->elem)) { | |
7267c094 | 140 | g_free(req); |
869a5c6d AL |
141 | return NULL; |
142 | } | |
6e02c38d AL |
143 | } |
144 | ||
145 | return req; | |
146 | } | |
147 | ||
1063b8b1 CH |
148 | static void virtio_blk_handle_scsi(VirtIOBlockReq *req) |
149 | { | |
47ce9ef7 | 150 | #ifdef __linux__ |
4277906d | 151 | int ret; |
1063b8b1 | 152 | int i; |
47ce9ef7 SW |
153 | #endif |
154 | int status = VIRTIO_BLK_S_OK; | |
1063b8b1 CH |
155 | |
156 | /* | |
157 | * We require at least one output segment each for the virtio_blk_outhdr | |
158 | * and the SCSI command block. | |
159 | * | |
160 | * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr | |
161 | * and the sense buffer pointer in the input segments. | |
162 | */ | |
163 | if (req->elem.out_num < 2 || req->elem.in_num < 3) { | |
164 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); | |
a597e79c | 165 | g_free(req); |
1063b8b1 CH |
166 | return; |
167 | } | |
168 | ||
169 | /* | |
f34e73cd PB |
170 | * The scsi inhdr is placed in the second-to-last input segment, just |
171 | * before the regular inhdr. | |
1063b8b1 | 172 | */ |
f34e73cd PB |
173 | req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base; |
174 | ||
a6c5c84a | 175 | if (!req->dev->blk->scsi) { |
f34e73cd PB |
176 | status = VIRTIO_BLK_S_UNSUPP; |
177 | goto fail; | |
1063b8b1 CH |
178 | } |
179 | ||
180 | /* | |
f34e73cd | 181 | * No support for bidirection commands yet. |
1063b8b1 | 182 | */ |
f34e73cd PB |
183 | if (req->elem.out_num > 2 && req->elem.in_num > 3) { |
184 | status = VIRTIO_BLK_S_UNSUPP; | |
185 | goto fail; | |
186 | } | |
1063b8b1 | 187 | |
f34e73cd PB |
188 | #ifdef __linux__ |
189 | struct sg_io_hdr hdr; | |
1063b8b1 CH |
190 | memset(&hdr, 0, sizeof(struct sg_io_hdr)); |
191 | hdr.interface_id = 'S'; | |
192 | hdr.cmd_len = req->elem.out_sg[1].iov_len; | |
193 | hdr.cmdp = req->elem.out_sg[1].iov_base; | |
194 | hdr.dxfer_len = 0; | |
195 | ||
196 | if (req->elem.out_num > 2) { | |
197 | /* | |
198 | * If there are more than the minimally required 2 output segments | |
199 | * there is write payload starting from the third iovec. | |
200 | */ | |
201 | hdr.dxfer_direction = SG_DXFER_TO_DEV; | |
202 | hdr.iovec_count = req->elem.out_num - 2; | |
203 | ||
204 | for (i = 0; i < hdr.iovec_count; i++) | |
205 | hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len; | |
206 | ||
207 | hdr.dxferp = req->elem.out_sg + 2; | |
208 | ||
209 | } else if (req->elem.in_num > 3) { | |
210 | /* | |
211 | * If we have more than 3 input segments the guest wants to actually | |
212 | * read data. | |
213 | */ | |
214 | hdr.dxfer_direction = SG_DXFER_FROM_DEV; | |
215 | hdr.iovec_count = req->elem.in_num - 3; | |
216 | for (i = 0; i < hdr.iovec_count; i++) | |
217 | hdr.dxfer_len += req->elem.in_sg[i].iov_len; | |
218 | ||
219 | hdr.dxferp = req->elem.in_sg; | |
1063b8b1 CH |
220 | } else { |
221 | /* | |
222 | * Some SCSI commands don't actually transfer any data. | |
223 | */ | |
224 | hdr.dxfer_direction = SG_DXFER_NONE; | |
225 | } | |
226 | ||
227 | hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base; | |
228 | hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len; | |
1063b8b1 CH |
229 | |
230 | ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr); | |
231 | if (ret) { | |
232 | status = VIRTIO_BLK_S_UNSUPP; | |
f34e73cd | 233 | goto fail; |
1063b8b1 CH |
234 | } |
235 | ||
5bb23927 PB |
236 | /* |
237 | * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi) | |
238 | * clear the masked_status field [hence status gets cleared too, see | |
239 | * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED | |
240 | * status has occurred. However they do set DRIVER_SENSE in driver_status | |
241 | * field. Also a (sb_len_wr > 0) indicates there is a sense buffer. | |
242 | */ | |
243 | if (hdr.status == 0 && hdr.sb_len_wr > 0) { | |
244 | hdr.status = CHECK_CONDITION; | |
245 | } | |
246 | ||
247 | stl_p(&req->scsi->errors, | |
248 | hdr.status | (hdr.msg_status << 8) | | |
249 | (hdr.host_status << 16) | (hdr.driver_status << 24)); | |
92e3c2a3 AJ |
250 | stl_p(&req->scsi->residual, hdr.resid); |
251 | stl_p(&req->scsi->sense_len, hdr.sb_len_wr); | |
252 | stl_p(&req->scsi->data_len, hdr.dxfer_len); | |
1063b8b1 CH |
253 | |
254 | virtio_blk_req_complete(req, status); | |
a597e79c | 255 | g_free(req); |
1063b8b1 | 256 | #else |
f34e73cd PB |
257 | abort(); |
258 | #endif | |
259 | ||
260 | fail: | |
261 | /* Just put anything nonzero so that the ioctl fails in the guest. */ | |
262 | stl_p(&req->scsi->errors, 255); | |
263 | virtio_blk_req_complete(req, status); | |
a597e79c | 264 | g_free(req); |
1063b8b1 | 265 | } |
1063b8b1 | 266 | |
c20fd872 CH |
267 | typedef struct MultiReqBuffer { |
268 | BlockRequest blkreq[32]; | |
269 | unsigned int num_writes; | |
270 | } MultiReqBuffer; | |
271 | ||
272 | static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb) | |
869a5c6d | 273 | { |
91553dcc | 274 | int i, ret; |
91553dcc | 275 | |
c20fd872 CH |
276 | if (!mrb->num_writes) { |
277 | return; | |
278 | } | |
279 | ||
280 | ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes); | |
91553dcc | 281 | if (ret != 0) { |
c20fd872 CH |
282 | for (i = 0; i < mrb->num_writes; i++) { |
283 | if (mrb->blkreq[i].error) { | |
284 | virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO); | |
91553dcc KW |
285 | } |
286 | } | |
287 | } | |
c20fd872 CH |
288 | |
289 | mrb->num_writes = 0; | |
91553dcc | 290 | } |
87b245db | 291 | |
c20fd872 | 292 | static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
aa659be3 | 293 | { |
a597e79c CH |
294 | bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH); |
295 | ||
618fbb84 CH |
296 | /* |
297 | * Make sure all outstanding writes are posted to the backing device. | |
298 | */ | |
c20fd872 | 299 | virtio_submit_multiwrite(req->dev->bs, mrb); |
ad54ae80 | 300 | bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req); |
aa659be3 CH |
301 | } |
302 | ||
c20fd872 | 303 | static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
91553dcc | 304 | { |
c20fd872 | 305 | BlockRequest *blkreq; |
92e3c2a3 | 306 | uint64_t sector; |
c20fd872 | 307 | |
92e3c2a3 | 308 | sector = ldq_p(&req->out->sector); |
6d519a5f | 309 | |
a597e79c CH |
310 | bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE); |
311 | ||
92e3c2a3 AJ |
312 | trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512); |
313 | ||
314 | if (sector & req->dev->sector_mask) { | |
8cfacf07 CH |
315 | virtio_blk_rw_complete(req, -EIO); |
316 | return; | |
317 | } | |
52c05023 CH |
318 | if (req->qiov.size % req->dev->conf->logical_block_size) { |
319 | virtio_blk_rw_complete(req, -EIO); | |
320 | return; | |
321 | } | |
8cfacf07 | 322 | |
c20fd872 CH |
323 | if (mrb->num_writes == 32) { |
324 | virtio_submit_multiwrite(req->dev->bs, mrb); | |
87b245db | 325 | } |
91553dcc | 326 | |
c20fd872 | 327 | blkreq = &mrb->blkreq[mrb->num_writes]; |
92e3c2a3 | 328 | blkreq->sector = sector; |
c20fd872 CH |
329 | blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE; |
330 | blkreq->qiov = &req->qiov; | |
331 | blkreq->cb = virtio_blk_rw_complete; | |
332 | blkreq->opaque = req; | |
333 | blkreq->error = 0; | |
91553dcc | 334 | |
c20fd872 | 335 | mrb->num_writes++; |
d28a1b6e | 336 | } |
869a5c6d | 337 | |
d28a1b6e AL |
338 | static void virtio_blk_handle_read(VirtIOBlockReq *req) |
339 | { | |
92e3c2a3 AJ |
340 | uint64_t sector; |
341 | ||
342 | sector = ldq_p(&req->out->sector); | |
87b245db | 343 | |
a597e79c CH |
344 | bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ); |
345 | ||
81b6b9fa SH |
346 | trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512); |
347 | ||
92e3c2a3 | 348 | if (sector & req->dev->sector_mask) { |
8cfacf07 CH |
349 | virtio_blk_rw_complete(req, -EIO); |
350 | return; | |
351 | } | |
52c05023 CH |
352 | if (req->qiov.size % req->dev->conf->logical_block_size) { |
353 | virtio_blk_rw_complete(req, -EIO); | |
354 | return; | |
355 | } | |
ad54ae80 PB |
356 | bdrv_aio_readv(req->dev->bs, sector, &req->qiov, |
357 | req->qiov.size / BDRV_SECTOR_SIZE, | |
358 | virtio_blk_rw_complete, req); | |
869a5c6d AL |
359 | } |
360 | ||
bc6694d4 KW |
361 | static void virtio_blk_handle_request(VirtIOBlockReq *req, |
362 | MultiReqBuffer *mrb) | |
363 | { | |
92e3c2a3 AJ |
364 | uint32_t type; |
365 | ||
bc6694d4 | 366 | if (req->elem.out_num < 1 || req->elem.in_num < 1) { |
870cef1d | 367 | error_report("virtio-blk missing headers"); |
bc6694d4 KW |
368 | exit(1); |
369 | } | |
370 | ||
371 | if (req->elem.out_sg[0].iov_len < sizeof(*req->out) || | |
372 | req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) { | |
870cef1d | 373 | error_report("virtio-blk header not in correct element"); |
bc6694d4 KW |
374 | exit(1); |
375 | } | |
376 | ||
377 | req->out = (void *)req->elem.out_sg[0].iov_base; | |
378 | req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base; | |
379 | ||
92e3c2a3 AJ |
380 | type = ldl_p(&req->out->type); |
381 | ||
382 | if (type & VIRTIO_BLK_T_FLUSH) { | |
c20fd872 | 383 | virtio_blk_handle_flush(req, mrb); |
92e3c2a3 | 384 | } else if (type & VIRTIO_BLK_T_SCSI_CMD) { |
bc6694d4 | 385 | virtio_blk_handle_scsi(req); |
92e3c2a3 | 386 | } else if (type & VIRTIO_BLK_T_GET_ID) { |
2930b313 | 387 | VirtIOBlock *s = req->dev; |
388 | ||
a8686a9b MA |
389 | /* |
390 | * NB: per existing s/n string convention the string is | |
391 | * terminated by '\0' only when shorter than buffer. | |
392 | */ | |
393 | strncpy(req->elem.in_sg[0].iov_base, | |
12c5674b | 394 | s->blk->serial ? s->blk->serial : "", |
a8686a9b | 395 | MIN(req->elem.in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES)); |
2930b313 | 396 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); |
a597e79c | 397 | g_free(req); |
92e3c2a3 | 398 | } else if (type & VIRTIO_BLK_T_OUT) { |
bc6694d4 KW |
399 | qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1], |
400 | req->elem.out_num - 1); | |
c20fd872 | 401 | virtio_blk_handle_write(req, mrb); |
bc6694d4 KW |
402 | } else { |
403 | qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0], | |
404 | req->elem.in_num - 1); | |
405 | virtio_blk_handle_read(req); | |
406 | } | |
407 | } | |
408 | ||
6e02c38d AL |
409 | static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq) |
410 | { | |
411 | VirtIOBlock *s = to_virtio_blk(vdev); | |
412 | VirtIOBlockReq *req; | |
bc6694d4 KW |
413 | MultiReqBuffer mrb = { |
414 | .num_writes = 0, | |
bc6694d4 | 415 | }; |
6e02c38d AL |
416 | |
417 | while ((req = virtio_blk_get_request(s))) { | |
bc6694d4 | 418 | virtio_blk_handle_request(req, &mrb); |
6e02c38d | 419 | } |
91553dcc | 420 | |
c20fd872 | 421 | virtio_submit_multiwrite(s->bs, &mrb); |
91553dcc | 422 | |
6e02c38d AL |
423 | /* |
424 | * FIXME: Want to check for completions before returning to guest mode, | |
425 | * so cached reads and writes are reported as quickly as possible. But | |
426 | * that should be done in the generic block layer. | |
427 | */ | |
428 | } | |
429 | ||
213189ab | 430 | static void virtio_blk_dma_restart_bh(void *opaque) |
869a5c6d AL |
431 | { |
432 | VirtIOBlock *s = opaque; | |
433 | VirtIOBlockReq *req = s->rq; | |
f1b52868 KW |
434 | MultiReqBuffer mrb = { |
435 | .num_writes = 0, | |
f1b52868 | 436 | }; |
869a5c6d | 437 | |
213189ab MA |
438 | qemu_bh_delete(s->bh); |
439 | s->bh = NULL; | |
869a5c6d AL |
440 | |
441 | s->rq = NULL; | |
442 | ||
443 | while (req) { | |
f1b52868 | 444 | virtio_blk_handle_request(req, &mrb); |
869a5c6d AL |
445 | req = req->next; |
446 | } | |
f1b52868 | 447 | |
c20fd872 | 448 | virtio_submit_multiwrite(s->bs, &mrb); |
869a5c6d AL |
449 | } |
450 | ||
1dfb4dd9 LC |
451 | static void virtio_blk_dma_restart_cb(void *opaque, int running, |
452 | RunState state) | |
213189ab MA |
453 | { |
454 | VirtIOBlock *s = opaque; | |
455 | ||
456 | if (!running) | |
457 | return; | |
458 | ||
459 | if (!s->bh) { | |
460 | s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s); | |
461 | qemu_bh_schedule(s->bh); | |
462 | } | |
463 | } | |
464 | ||
6e02c38d AL |
465 | static void virtio_blk_reset(VirtIODevice *vdev) |
466 | { | |
467 | /* | |
468 | * This should cancel pending requests, but can't do nicely until there | |
469 | * are per-device request lists. | |
470 | */ | |
922453bc | 471 | bdrv_drain_all(); |
6e02c38d AL |
472 | } |
473 | ||
bf011293 | 474 | /* coalesce internal state, copy to pci i/o region 0 |
475 | */ | |
6e02c38d AL |
476 | static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config) |
477 | { | |
478 | VirtIOBlock *s = to_virtio_blk(vdev); | |
479 | struct virtio_blk_config blkcfg; | |
480 | uint64_t capacity; | |
481 | int cylinders, heads, secs; | |
3a395142 | 482 | int blk_size = s->conf->logical_block_size; |
6e02c38d AL |
483 | |
484 | bdrv_get_geometry(s->bs, &capacity); | |
485 | bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs); | |
5c5dafdc | 486 | memset(&blkcfg, 0, sizeof(blkcfg)); |
6e02c38d AL |
487 | stq_raw(&blkcfg.capacity, capacity); |
488 | stl_raw(&blkcfg.seg_max, 128 - 2); | |
489 | stw_raw(&blkcfg.cylinders, cylinders); | |
3a395142 PB |
490 | stl_raw(&blkcfg.blk_size, blk_size); |
491 | stw_raw(&blkcfg.min_io_size, s->conf->min_io_size / blk_size); | |
492 | stw_raw(&blkcfg.opt_io_size, s->conf->opt_io_size / blk_size); | |
6e02c38d | 493 | blkcfg.heads = heads; |
8cfacf07 | 494 | blkcfg.sectors = secs & ~s->sector_mask; |
c7085da7 | 495 | blkcfg.size_max = 0; |
9752c371 CH |
496 | blkcfg.physical_block_exp = get_physical_block_exp(s->conf); |
497 | blkcfg.alignment_offset = 0; | |
37d5ddd6 | 498 | memcpy(config, &blkcfg, sizeof(struct virtio_blk_config)); |
6e02c38d AL |
499 | } |
500 | ||
8172539d | 501 | static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features) |
6e02c38d | 502 | { |
bf011293 | 503 | VirtIOBlock *s = to_virtio_blk(vdev); |
1063b8b1 CH |
504 | |
505 | features |= (1 << VIRTIO_BLK_F_SEG_MAX); | |
506 | features |= (1 << VIRTIO_BLK_F_GEOMETRY); | |
9752c371 | 507 | features |= (1 << VIRTIO_BLK_F_TOPOLOGY); |
8cfacf07 | 508 | features |= (1 << VIRTIO_BLK_F_BLK_SIZE); |
a6c5c84a | 509 | features |= (1 << VIRTIO_BLK_F_SCSI); |
aa659be3 CH |
510 | |
511 | if (bdrv_enable_write_cache(s->bs)) | |
512 | features |= (1 << VIRTIO_BLK_F_WCACHE); | |
c79662f7 NS |
513 | |
514 | if (bdrv_is_read_only(s->bs)) | |
515 | features |= 1 << VIRTIO_BLK_F_RO; | |
1063b8b1 CH |
516 | |
517 | return features; | |
6e02c38d AL |
518 | } |
519 | ||
520 | static void virtio_blk_save(QEMUFile *f, void *opaque) | |
521 | { | |
522 | VirtIOBlock *s = opaque; | |
869a5c6d AL |
523 | VirtIOBlockReq *req = s->rq; |
524 | ||
6e02c38d | 525 | virtio_save(&s->vdev, f); |
869a5c6d AL |
526 | |
527 | while (req) { | |
528 | qemu_put_sbyte(f, 1); | |
529 | qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem)); | |
530 | req = req->next; | |
531 | } | |
532 | qemu_put_sbyte(f, 0); | |
6e02c38d AL |
533 | } |
534 | ||
535 | static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id) | |
536 | { | |
537 | VirtIOBlock *s = opaque; | |
2a633c46 | 538 | int ret; |
6e02c38d | 539 | |
869a5c6d | 540 | if (version_id != 2) |
6e02c38d AL |
541 | return -EINVAL; |
542 | ||
2a633c46 OW |
543 | ret = virtio_load(&s->vdev, f); |
544 | if (ret) { | |
545 | return ret; | |
546 | } | |
547 | ||
869a5c6d AL |
548 | while (qemu_get_sbyte(f)) { |
549 | VirtIOBlockReq *req = virtio_blk_alloc_request(s); | |
550 | qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem)); | |
551 | req->next = s->rq; | |
20a81e4d | 552 | s->rq = req; |
b6a4805b KW |
553 | |
554 | virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr, | |
555 | req->elem.in_num, 1); | |
556 | virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr, | |
557 | req->elem.out_num, 0); | |
869a5c6d | 558 | } |
6e02c38d AL |
559 | |
560 | return 0; | |
561 | } | |
562 | ||
145feb17 | 563 | static void virtio_blk_resize(void *opaque) |
e5051fc7 CH |
564 | { |
565 | VirtIOBlock *s = opaque; | |
566 | ||
145feb17 | 567 | virtio_notify_config(&s->vdev); |
e5051fc7 CH |
568 | } |
569 | ||
0e49de52 | 570 | static const BlockDevOps virtio_block_ops = { |
145feb17 | 571 | .resize_cb = virtio_blk_resize, |
0e49de52 MA |
572 | }; |
573 | ||
12c5674b | 574 | VirtIODevice *virtio_blk_init(DeviceState *dev, VirtIOBlkConf *blk) |
6e02c38d AL |
575 | { |
576 | VirtIOBlock *s; | |
577 | int cylinders, heads, secs; | |
578 | static int virtio_blk_id; | |
2930b313 | 579 | DriveInfo *dinfo; |
cf21e106 | 580 | |
12c5674b | 581 | if (!blk->conf.bs) { |
6a84cb1f | 582 | error_report("drive property not set"); |
d75d25e3 MA |
583 | return NULL; |
584 | } | |
12c5674b | 585 | if (!bdrv_is_inserted(blk->conf.bs)) { |
98f28ad7 MA |
586 | error_report("Device needs media, but drive is empty"); |
587 | return NULL; | |
588 | } | |
d75d25e3 | 589 | |
12c5674b | 590 | if (!blk->serial) { |
a8686a9b | 591 | /* try to fall back to value set with legacy -drive serial=... */ |
12c5674b | 592 | dinfo = drive_get_by_blockdev(blk->conf.bs); |
a8686a9b | 593 | if (*dinfo->serial) { |
12c5674b | 594 | blk->serial = strdup(dinfo->serial); |
a8686a9b MA |
595 | } |
596 | } | |
597 | ||
53c25cea | 598 | s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK, |
37d5ddd6 | 599 | sizeof(struct virtio_blk_config), |
53c25cea | 600 | sizeof(VirtIOBlock)); |
6e02c38d AL |
601 | |
602 | s->vdev.get_config = virtio_blk_update_config; | |
603 | s->vdev.get_features = virtio_blk_get_features; | |
604 | s->vdev.reset = virtio_blk_reset; | |
12c5674b PB |
605 | s->bs = blk->conf.bs; |
606 | s->conf = &blk->conf; | |
607 | s->blk = blk; | |
869a5c6d | 608 | s->rq = NULL; |
1573a35d | 609 | s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1; |
6e02c38d | 610 | bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs); |
6e02c38d AL |
611 | |
612 | s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output); | |
613 | ||
869a5c6d | 614 | qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s); |
9d0d3138 | 615 | s->qdev = dev; |
0be71e32 | 616 | register_savevm(dev, "virtio-blk", virtio_blk_id++, 2, |
6e02c38d | 617 | virtio_blk_save, virtio_blk_load, s); |
0e49de52 | 618 | bdrv_set_dev_ops(s->bs, &virtio_block_ops, s); |
12c5674b | 619 | bdrv_set_buffer_alignment(s->bs, s->conf->logical_block_size); |
6e02c38d | 620 | |
af239a62 | 621 | bdrv_iostatus_enable(s->bs); |
12c5674b | 622 | add_boot_device_path(s->conf->bootindex, dev, "/disk@0,0"); |
1ca4d09a | 623 | |
53c25cea | 624 | return &s->vdev; |
6e02c38d | 625 | } |
9d0d3138 AW |
626 | |
627 | void virtio_blk_exit(VirtIODevice *vdev) | |
628 | { | |
629 | VirtIOBlock *s = to_virtio_blk(vdev); | |
630 | unregister_savevm(s->qdev, "virtio-blk", s); | |
0e47931b | 631 | blockdev_mark_auto_del(s->bs); |
d92551f2 | 632 | virtio_cleanup(vdev); |
9d0d3138 | 633 | } |