]>
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" |
827805a2 | 15 | #include "qemu/iov.h" |
1de7afc9 | 16 | #include "qemu/error-report.h" |
6d519a5f | 17 | #include "trace.h" |
0d09e41a | 18 | #include "hw/block/block.h" |
9c17d615 | 19 | #include "sysemu/blockdev.h" |
0d09e41a PB |
20 | #include "hw/virtio/virtio-blk.h" |
21 | #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE | |
22 | # include "dataplane/virtio-blk.h" | |
84db52d0 | 23 | # include "migration/migration.h" |
0d09e41a PB |
24 | #endif |
25 | #include "block/scsi.h" | |
1063b8b1 CH |
26 | #ifdef __linux__ |
27 | # include <scsi/sg.h> | |
28 | #endif | |
0d09e41a | 29 | #include "hw/virtio/virtio-bus.h" |
783d1897 | 30 | #include "hw/virtio/virtio-access.h" |
6e02c38d | 31 | |
f897bf75 | 32 | VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s) |
671ec3f0 | 33 | { |
869d66af | 34 | VirtIOBlockReq *req = g_slice_new(VirtIOBlockReq); |
671ec3f0 | 35 | req->dev = s; |
869d66af SH |
36 | req->qiov.size = 0; |
37 | req->next = NULL; | |
671ec3f0 FZ |
38 | return req; |
39 | } | |
40 | ||
f897bf75 | 41 | void virtio_blk_free_request(VirtIOBlockReq *req) |
671ec3f0 FZ |
42 | { |
43 | if (req) { | |
671ec3f0 FZ |
44 | g_slice_free(VirtIOBlockReq, req); |
45 | } | |
46 | } | |
47 | ||
bf4bd461 FZ |
48 | static void virtio_blk_complete_request(VirtIOBlockReq *req, |
49 | unsigned char status) | |
869a5c6d AL |
50 | { |
51 | VirtIOBlock *s = req->dev; | |
1cc91b7d | 52 | VirtIODevice *vdev = VIRTIO_DEVICE(s); |
869a5c6d | 53 | |
6d519a5f SH |
54 | trace_virtio_blk_req_complete(req, status); |
55 | ||
92e3c2a3 | 56 | stb_p(&req->in->status, status); |
f897bf75 | 57 | virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in)); |
1cc91b7d | 58 | virtio_notify(vdev, s->vq); |
869a5c6d AL |
59 | } |
60 | ||
bf4bd461 FZ |
61 | static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status) |
62 | { | |
63 | req->dev->complete_request(req, status); | |
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 | ||
a589569f | 72 | if (action == BLOCK_ERROR_ACTION_STOP) { |
869a5c6d AL |
73 | req->next = s->rq; |
74 | s->rq = req; | |
a589569f | 75 | } else if (action == BLOCK_ERROR_ACTION_REPORT) { |
869a5c6d | 76 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); |
a597e79c | 77 | bdrv_acct_done(s->bs, &req->acct); |
671ec3f0 | 78 | virtio_blk_free_request(req); |
869a5c6d AL |
79 | } |
80 | ||
3e1caa5f | 81 | bdrv_error_action(s->bs, action, is_read, error); |
a589569f | 82 | return action != BLOCK_ERROR_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) { |
783d1897 RR |
92 | int p = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type); |
93 | bool is_read = !(p & VIRTIO_BLK_T_OUT); | |
f35d68f0 | 94 | if (virtio_blk_handle_rw_error(req, -ret, is_read)) |
869a5c6d | 95 | return; |
6e02c38d AL |
96 | } |
97 | ||
f35d68f0 | 98 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); |
a597e79c | 99 | bdrv_acct_done(req->dev->bs, &req->acct); |
671ec3f0 | 100 | virtio_blk_free_request(req); |
869a5c6d | 101 | } |
6e02c38d | 102 | |
aa659be3 CH |
103 | static void virtio_blk_flush_complete(void *opaque, int ret) |
104 | { | |
105 | VirtIOBlockReq *req = opaque; | |
106 | ||
8c269b54 KW |
107 | if (ret) { |
108 | if (virtio_blk_handle_rw_error(req, -ret, 0)) { | |
109 | return; | |
110 | } | |
111 | } | |
112 | ||
113 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); | |
a597e79c | 114 | bdrv_acct_done(req->dev->bs, &req->acct); |
671ec3f0 | 115 | virtio_blk_free_request(req); |
6e02c38d AL |
116 | } |
117 | ||
118 | static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s) | |
119 | { | |
869a5c6d | 120 | VirtIOBlockReq *req = virtio_blk_alloc_request(s); |
6e02c38d | 121 | |
f897bf75 | 122 | if (!virtqueue_pop(s->vq, &req->elem)) { |
671ec3f0 FZ |
123 | virtio_blk_free_request(req); |
124 | return NULL; | |
6e02c38d AL |
125 | } |
126 | ||
127 | return req; | |
128 | } | |
129 | ||
5a05cbee FZ |
130 | int virtio_blk_handle_scsi_req(VirtIOBlock *blk, |
131 | VirtQueueElement *elem) | |
1063b8b1 | 132 | { |
5a05cbee FZ |
133 | int status = VIRTIO_BLK_S_OK; |
134 | struct virtio_scsi_inhdr *scsi = NULL; | |
783d1897 RR |
135 | VirtIODevice *vdev = VIRTIO_DEVICE(blk); |
136 | ||
47ce9ef7 | 137 | #ifdef __linux__ |
1063b8b1 | 138 | int i; |
5a05cbee | 139 | struct sg_io_hdr hdr; |
47ce9ef7 | 140 | #endif |
1063b8b1 CH |
141 | |
142 | /* | |
143 | * We require at least one output segment each for the virtio_blk_outhdr | |
144 | * and the SCSI command block. | |
145 | * | |
146 | * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr | |
147 | * and the sense buffer pointer in the input segments. | |
148 | */ | |
5a05cbee FZ |
149 | if (elem->out_num < 2 || elem->in_num < 3) { |
150 | status = VIRTIO_BLK_S_IOERR; | |
151 | goto fail; | |
1063b8b1 CH |
152 | } |
153 | ||
154 | /* | |
f34e73cd PB |
155 | * The scsi inhdr is placed in the second-to-last input segment, just |
156 | * before the regular inhdr. | |
1063b8b1 | 157 | */ |
5a05cbee | 158 | scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base; |
f34e73cd | 159 | |
5a05cbee | 160 | if (!blk->blk.scsi) { |
f34e73cd PB |
161 | status = VIRTIO_BLK_S_UNSUPP; |
162 | goto fail; | |
1063b8b1 CH |
163 | } |
164 | ||
165 | /* | |
f34e73cd | 166 | * No support for bidirection commands yet. |
1063b8b1 | 167 | */ |
5a05cbee | 168 | if (elem->out_num > 2 && elem->in_num > 3) { |
f34e73cd PB |
169 | status = VIRTIO_BLK_S_UNSUPP; |
170 | goto fail; | |
171 | } | |
1063b8b1 | 172 | |
f34e73cd | 173 | #ifdef __linux__ |
1063b8b1 CH |
174 | memset(&hdr, 0, sizeof(struct sg_io_hdr)); |
175 | hdr.interface_id = 'S'; | |
5a05cbee FZ |
176 | hdr.cmd_len = elem->out_sg[1].iov_len; |
177 | hdr.cmdp = elem->out_sg[1].iov_base; | |
1063b8b1 CH |
178 | hdr.dxfer_len = 0; |
179 | ||
5a05cbee | 180 | if (elem->out_num > 2) { |
1063b8b1 CH |
181 | /* |
182 | * If there are more than the minimally required 2 output segments | |
183 | * there is write payload starting from the third iovec. | |
184 | */ | |
185 | hdr.dxfer_direction = SG_DXFER_TO_DEV; | |
5a05cbee | 186 | hdr.iovec_count = elem->out_num - 2; |
1063b8b1 CH |
187 | |
188 | for (i = 0; i < hdr.iovec_count; i++) | |
5a05cbee | 189 | hdr.dxfer_len += elem->out_sg[i + 2].iov_len; |
1063b8b1 | 190 | |
5a05cbee | 191 | hdr.dxferp = elem->out_sg + 2; |
1063b8b1 | 192 | |
5a05cbee | 193 | } else if (elem->in_num > 3) { |
1063b8b1 CH |
194 | /* |
195 | * If we have more than 3 input segments the guest wants to actually | |
196 | * read data. | |
197 | */ | |
198 | hdr.dxfer_direction = SG_DXFER_FROM_DEV; | |
5a05cbee | 199 | hdr.iovec_count = elem->in_num - 3; |
1063b8b1 | 200 | for (i = 0; i < hdr.iovec_count; i++) |
5a05cbee | 201 | hdr.dxfer_len += elem->in_sg[i].iov_len; |
1063b8b1 | 202 | |
5a05cbee | 203 | hdr.dxferp = elem->in_sg; |
1063b8b1 CH |
204 | } else { |
205 | /* | |
206 | * Some SCSI commands don't actually transfer any data. | |
207 | */ | |
208 | hdr.dxfer_direction = SG_DXFER_NONE; | |
209 | } | |
210 | ||
5a05cbee FZ |
211 | hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base; |
212 | hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len; | |
1063b8b1 | 213 | |
5a05cbee FZ |
214 | status = bdrv_ioctl(blk->bs, SG_IO, &hdr); |
215 | if (status) { | |
1063b8b1 | 216 | status = VIRTIO_BLK_S_UNSUPP; |
f34e73cd | 217 | goto fail; |
1063b8b1 CH |
218 | } |
219 | ||
5bb23927 PB |
220 | /* |
221 | * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi) | |
222 | * clear the masked_status field [hence status gets cleared too, see | |
223 | * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED | |
224 | * status has occurred. However they do set DRIVER_SENSE in driver_status | |
225 | * field. Also a (sb_len_wr > 0) indicates there is a sense buffer. | |
226 | */ | |
227 | if (hdr.status == 0 && hdr.sb_len_wr > 0) { | |
228 | hdr.status = CHECK_CONDITION; | |
229 | } | |
230 | ||
783d1897 RR |
231 | virtio_stl_p(vdev, &scsi->errors, |
232 | hdr.status | (hdr.msg_status << 8) | | |
233 | (hdr.host_status << 16) | (hdr.driver_status << 24)); | |
234 | virtio_stl_p(vdev, &scsi->residual, hdr.resid); | |
235 | virtio_stl_p(vdev, &scsi->sense_len, hdr.sb_len_wr); | |
236 | virtio_stl_p(vdev, &scsi->data_len, hdr.dxfer_len); | |
1063b8b1 | 237 | |
5a05cbee | 238 | return status; |
1063b8b1 | 239 | #else |
f34e73cd PB |
240 | abort(); |
241 | #endif | |
242 | ||
243 | fail: | |
244 | /* Just put anything nonzero so that the ioctl fails in the guest. */ | |
5a05cbee | 245 | if (scsi) { |
783d1897 | 246 | virtio_stl_p(vdev, &scsi->errors, 255); |
5a05cbee FZ |
247 | } |
248 | return status; | |
249 | } | |
250 | ||
251 | static void virtio_blk_handle_scsi(VirtIOBlockReq *req) | |
252 | { | |
253 | int status; | |
254 | ||
f897bf75 | 255 | status = virtio_blk_handle_scsi_req(req->dev, &req->elem); |
f34e73cd | 256 | virtio_blk_req_complete(req, status); |
671ec3f0 | 257 | virtio_blk_free_request(req); |
1063b8b1 | 258 | } |
1063b8b1 | 259 | |
fee65db7 | 260 | void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb) |
869a5c6d | 261 | { |
91553dcc | 262 | int i, ret; |
91553dcc | 263 | |
c20fd872 CH |
264 | if (!mrb->num_writes) { |
265 | return; | |
266 | } | |
267 | ||
268 | ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes); | |
91553dcc | 269 | if (ret != 0) { |
c20fd872 CH |
270 | for (i = 0; i < mrb->num_writes; i++) { |
271 | if (mrb->blkreq[i].error) { | |
272 | virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO); | |
91553dcc KW |
273 | } |
274 | } | |
275 | } | |
c20fd872 CH |
276 | |
277 | mrb->num_writes = 0; | |
91553dcc | 278 | } |
87b245db | 279 | |
c20fd872 | 280 | static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
aa659be3 | 281 | { |
a597e79c CH |
282 | bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH); |
283 | ||
618fbb84 CH |
284 | /* |
285 | * Make sure all outstanding writes are posted to the backing device. | |
286 | */ | |
c20fd872 | 287 | virtio_submit_multiwrite(req->dev->bs, mrb); |
ad54ae80 | 288 | bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req); |
aa659be3 CH |
289 | } |
290 | ||
d0e14376 MA |
291 | static bool virtio_blk_sect_range_ok(VirtIOBlock *dev, |
292 | uint64_t sector, size_t size) | |
293 | { | |
3c2daac0 MA |
294 | uint64_t nb_sectors = size >> BDRV_SECTOR_BITS; |
295 | uint64_t total_sectors; | |
296 | ||
d0e14376 MA |
297 | if (sector & dev->sector_mask) { |
298 | return false; | |
299 | } | |
300 | if (size % dev->conf->logical_block_size) { | |
301 | return false; | |
302 | } | |
3c2daac0 MA |
303 | bdrv_get_geometry(dev->bs, &total_sectors); |
304 | if (sector > total_sectors || nb_sectors > total_sectors - sector) { | |
305 | return false; | |
306 | } | |
d0e14376 MA |
307 | return true; |
308 | } | |
309 | ||
c20fd872 | 310 | static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
91553dcc | 311 | { |
c20fd872 | 312 | BlockRequest *blkreq; |
92e3c2a3 | 313 | uint64_t sector; |
c20fd872 | 314 | |
783d1897 | 315 | sector = virtio_ldq_p(VIRTIO_DEVICE(req->dev), &req->out.sector); |
6d519a5f | 316 | |
92e3c2a3 AJ |
317 | trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512); |
318 | ||
d0e14376 | 319 | if (!virtio_blk_sect_range_ok(req->dev, sector, req->qiov.size)) { |
42e38c1f MA |
320 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); |
321 | virtio_blk_free_request(req); | |
52c05023 CH |
322 | return; |
323 | } | |
8cfacf07 | 324 | |
42e38c1f MA |
325 | bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE); |
326 | ||
c20fd872 CH |
327 | if (mrb->num_writes == 32) { |
328 | virtio_submit_multiwrite(req->dev->bs, mrb); | |
87b245db | 329 | } |
91553dcc | 330 | |
c20fd872 | 331 | blkreq = &mrb->blkreq[mrb->num_writes]; |
92e3c2a3 | 332 | blkreq->sector = sector; |
c20fd872 CH |
333 | blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE; |
334 | blkreq->qiov = &req->qiov; | |
335 | blkreq->cb = virtio_blk_rw_complete; | |
336 | blkreq->opaque = req; | |
337 | blkreq->error = 0; | |
91553dcc | 338 | |
c20fd872 | 339 | mrb->num_writes++; |
d28a1b6e | 340 | } |
869a5c6d | 341 | |
d28a1b6e AL |
342 | static void virtio_blk_handle_read(VirtIOBlockReq *req) |
343 | { | |
92e3c2a3 AJ |
344 | uint64_t sector; |
345 | ||
783d1897 | 346 | sector = virtio_ldq_p(VIRTIO_DEVICE(req->dev), &req->out.sector); |
87b245db | 347 | |
81b6b9fa SH |
348 | trace_virtio_blk_handle_read(req, sector, req->qiov.size / 512); |
349 | ||
d0e14376 | 350 | if (!virtio_blk_sect_range_ok(req->dev, sector, req->qiov.size)) { |
42e38c1f MA |
351 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); |
352 | virtio_blk_free_request(req); | |
52c05023 CH |
353 | return; |
354 | } | |
42e38c1f MA |
355 | |
356 | bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ); | |
ad54ae80 PB |
357 | bdrv_aio_readv(req->dev->bs, sector, &req->qiov, |
358 | req->qiov.size / BDRV_SECTOR_SIZE, | |
359 | virtio_blk_rw_complete, req); | |
869a5c6d AL |
360 | } |
361 | ||
fee65db7 | 362 | void virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
bc6694d4 | 363 | { |
92e3c2a3 | 364 | uint32_t type; |
f897bf75 SH |
365 | struct iovec *in_iov = req->elem.in_sg; |
366 | struct iovec *iov = req->elem.out_sg; | |
367 | unsigned in_num = req->elem.in_num; | |
368 | unsigned out_num = req->elem.out_num; | |
92e3c2a3 | 369 | |
f897bf75 | 370 | if (req->elem.out_num < 1 || req->elem.in_num < 1) { |
870cef1d | 371 | error_report("virtio-blk missing headers"); |
bc6694d4 KW |
372 | exit(1); |
373 | } | |
374 | ||
827805a2 FZ |
375 | if (unlikely(iov_to_buf(iov, out_num, 0, &req->out, |
376 | sizeof(req->out)) != sizeof(req->out))) { | |
377 | error_report("virtio-blk request outhdr too short"); | |
378 | exit(1); | |
379 | } | |
ee17e848 | 380 | |
827805a2 | 381 | iov_discard_front(&iov, &out_num, sizeof(req->out)); |
ee17e848 FZ |
382 | |
383 | if (in_num < 1 || | |
384 | in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) { | |
385 | error_report("virtio-blk request inhdr too short"); | |
386 | exit(1); | |
387 | } | |
388 | ||
389 | req->in = (void *)in_iov[in_num - 1].iov_base | |
390 | + in_iov[in_num - 1].iov_len | |
391 | - sizeof(struct virtio_blk_inhdr); | |
392 | iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr)); | |
bc6694d4 | 393 | |
783d1897 | 394 | type = virtio_ldl_p(VIRTIO_DEVICE(req->dev), &req->out.type); |
92e3c2a3 AJ |
395 | |
396 | if (type & VIRTIO_BLK_T_FLUSH) { | |
c20fd872 | 397 | virtio_blk_handle_flush(req, mrb); |
92e3c2a3 | 398 | } else if (type & VIRTIO_BLK_T_SCSI_CMD) { |
bc6694d4 | 399 | virtio_blk_handle_scsi(req); |
92e3c2a3 | 400 | } else if (type & VIRTIO_BLK_T_GET_ID) { |
2930b313 | 401 | VirtIOBlock *s = req->dev; |
402 | ||
a8686a9b MA |
403 | /* |
404 | * NB: per existing s/n string convention the string is | |
405 | * terminated by '\0' only when shorter than buffer. | |
406 | */ | |
a83ceea8 MM |
407 | const char *serial = s->blk.serial ? s->blk.serial : ""; |
408 | size_t size = MIN(strlen(serial) + 1, | |
409 | MIN(iov_size(in_iov, in_num), | |
410 | VIRTIO_BLK_ID_BYTES)); | |
411 | iov_from_buf(in_iov, in_num, 0, serial, size); | |
2930b313 | 412 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); |
671ec3f0 | 413 | virtio_blk_free_request(req); |
92e3c2a3 | 414 | } else if (type & VIRTIO_BLK_T_OUT) { |
a83ceea8 | 415 | qemu_iovec_init_external(&req->qiov, iov, out_num); |
c20fd872 | 416 | virtio_blk_handle_write(req, mrb); |
9e72c450 AZ |
417 | } else if (type == VIRTIO_BLK_T_IN || type == VIRTIO_BLK_T_BARRIER) { |
418 | /* VIRTIO_BLK_T_IN is 0, so we can't just & it. */ | |
a83ceea8 | 419 | qemu_iovec_init_external(&req->qiov, in_iov, in_num); |
bc6694d4 | 420 | virtio_blk_handle_read(req); |
9e72c450 AZ |
421 | } else { |
422 | virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); | |
671ec3f0 | 423 | virtio_blk_free_request(req); |
bc6694d4 KW |
424 | } |
425 | } | |
426 | ||
6e02c38d AL |
427 | static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq) |
428 | { | |
1cc91b7d | 429 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
6e02c38d | 430 | VirtIOBlockReq *req; |
bc6694d4 KW |
431 | MultiReqBuffer mrb = { |
432 | .num_writes = 0, | |
bc6694d4 | 433 | }; |
6e02c38d | 434 | |
392808b4 SH |
435 | #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE |
436 | /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start | |
437 | * dataplane here instead of waiting for .set_status(). | |
438 | */ | |
439 | if (s->dataplane) { | |
440 | virtio_blk_data_plane_start(s->dataplane); | |
441 | return; | |
442 | } | |
443 | #endif | |
444 | ||
6e02c38d | 445 | while ((req = virtio_blk_get_request(s))) { |
bc6694d4 | 446 | virtio_blk_handle_request(req, &mrb); |
6e02c38d | 447 | } |
91553dcc | 448 | |
c20fd872 | 449 | virtio_submit_multiwrite(s->bs, &mrb); |
91553dcc | 450 | |
6e02c38d AL |
451 | /* |
452 | * FIXME: Want to check for completions before returning to guest mode, | |
453 | * so cached reads and writes are reported as quickly as possible. But | |
454 | * that should be done in the generic block layer. | |
455 | */ | |
456 | } | |
457 | ||
213189ab | 458 | static void virtio_blk_dma_restart_bh(void *opaque) |
869a5c6d AL |
459 | { |
460 | VirtIOBlock *s = opaque; | |
461 | VirtIOBlockReq *req = s->rq; | |
f1b52868 KW |
462 | MultiReqBuffer mrb = { |
463 | .num_writes = 0, | |
f1b52868 | 464 | }; |
869a5c6d | 465 | |
213189ab MA |
466 | qemu_bh_delete(s->bh); |
467 | s->bh = NULL; | |
869a5c6d AL |
468 | |
469 | s->rq = NULL; | |
470 | ||
471 | while (req) { | |
f1b52868 | 472 | virtio_blk_handle_request(req, &mrb); |
869a5c6d AL |
473 | req = req->next; |
474 | } | |
f1b52868 | 475 | |
c20fd872 | 476 | virtio_submit_multiwrite(s->bs, &mrb); |
869a5c6d AL |
477 | } |
478 | ||
1dfb4dd9 LC |
479 | static void virtio_blk_dma_restart_cb(void *opaque, int running, |
480 | RunState state) | |
213189ab MA |
481 | { |
482 | VirtIOBlock *s = opaque; | |
483 | ||
392808b4 | 484 | if (!running) { |
213189ab | 485 | return; |
392808b4 | 486 | } |
213189ab MA |
487 | |
488 | if (!s->bh) { | |
4407c1c5 FZ |
489 | s->bh = aio_bh_new(bdrv_get_aio_context(s->blk.conf.bs), |
490 | virtio_blk_dma_restart_bh, s); | |
213189ab MA |
491 | qemu_bh_schedule(s->bh); |
492 | } | |
493 | } | |
494 | ||
6e02c38d AL |
495 | static void virtio_blk_reset(VirtIODevice *vdev) |
496 | { | |
1cc91b7d | 497 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
392808b4 | 498 | |
ef5bc962 | 499 | #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE |
392808b4 SH |
500 | if (s->dataplane) { |
501 | virtio_blk_data_plane_stop(s->dataplane); | |
502 | } | |
503 | #endif | |
504 | ||
6e02c38d AL |
505 | /* |
506 | * This should cancel pending requests, but can't do nicely until there | |
507 | * are per-device request lists. | |
508 | */ | |
922453bc | 509 | bdrv_drain_all(); |
ef5bc962 | 510 | bdrv_set_enable_write_cache(s->bs, s->original_wce); |
6e02c38d AL |
511 | } |
512 | ||
bf011293 | 513 | /* coalesce internal state, copy to pci i/o region 0 |
514 | */ | |
6e02c38d AL |
515 | static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config) |
516 | { | |
1cc91b7d | 517 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
6e02c38d AL |
518 | struct virtio_blk_config blkcfg; |
519 | uint64_t capacity; | |
3a395142 | 520 | int blk_size = s->conf->logical_block_size; |
6e02c38d AL |
521 | |
522 | bdrv_get_geometry(s->bs, &capacity); | |
5c5dafdc | 523 | memset(&blkcfg, 0, sizeof(blkcfg)); |
783d1897 RR |
524 | virtio_stq_p(vdev, &blkcfg.capacity, capacity); |
525 | virtio_stl_p(vdev, &blkcfg.seg_max, 128 - 2); | |
526 | virtio_stw_p(vdev, &blkcfg.cylinders, s->conf->cyls); | |
527 | virtio_stl_p(vdev, &blkcfg.blk_size, blk_size); | |
528 | virtio_stw_p(vdev, &blkcfg.min_io_size, s->conf->min_io_size / blk_size); | |
529 | virtio_stw_p(vdev, &blkcfg.opt_io_size, s->conf->opt_io_size / blk_size); | |
e63e7fde | 530 | blkcfg.heads = s->conf->heads; |
136be99e CB |
531 | /* |
532 | * We must ensure that the block device capacity is a multiple of | |
e03ba136 | 533 | * the logical block size. If that is not the case, let's use |
136be99e CB |
534 | * sector_mask to adopt the geometry to have a correct picture. |
535 | * For those devices where the capacity is ok for the given geometry | |
e03ba136 | 536 | * we don't touch the sector value of the geometry, since some devices |
136be99e CB |
537 | * (like s390 dasd) need a specific value. Here the capacity is already |
538 | * cyls*heads*secs*blk_size and the sector value is not block size | |
539 | * divided by 512 - instead it is the amount of blk_size blocks | |
540 | * per track (cylinder). | |
541 | */ | |
e63e7fde MA |
542 | if (bdrv_getlength(s->bs) / s->conf->heads / s->conf->secs % blk_size) { |
543 | blkcfg.sectors = s->conf->secs & ~s->sector_mask; | |
136be99e | 544 | } else { |
e63e7fde | 545 | blkcfg.sectors = s->conf->secs; |
136be99e | 546 | } |
c7085da7 | 547 | blkcfg.size_max = 0; |
9752c371 CH |
548 | blkcfg.physical_block_exp = get_physical_block_exp(s->conf); |
549 | blkcfg.alignment_offset = 0; | |
13e3dce0 | 550 | blkcfg.wce = bdrv_enable_write_cache(s->bs); |
37d5ddd6 | 551 | memcpy(config, &blkcfg, sizeof(struct virtio_blk_config)); |
6e02c38d AL |
552 | } |
553 | ||
13e3dce0 PB |
554 | static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config) |
555 | { | |
1cc91b7d | 556 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
13e3dce0 PB |
557 | struct virtio_blk_config blkcfg; |
558 | ||
559 | memcpy(&blkcfg, config, sizeof(blkcfg)); | |
6d7e73d6 FZ |
560 | |
561 | aio_context_acquire(bdrv_get_aio_context(s->bs)); | |
13e3dce0 | 562 | bdrv_set_enable_write_cache(s->bs, blkcfg.wce != 0); |
6d7e73d6 | 563 | aio_context_release(bdrv_get_aio_context(s->bs)); |
13e3dce0 PB |
564 | } |
565 | ||
8172539d | 566 | static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features) |
6e02c38d | 567 | { |
1cc91b7d | 568 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
1063b8b1 CH |
569 | |
570 | features |= (1 << VIRTIO_BLK_F_SEG_MAX); | |
571 | features |= (1 << VIRTIO_BLK_F_GEOMETRY); | |
9752c371 | 572 | features |= (1 << VIRTIO_BLK_F_TOPOLOGY); |
8cfacf07 | 573 | features |= (1 << VIRTIO_BLK_F_BLK_SIZE); |
a6c5c84a | 574 | features |= (1 << VIRTIO_BLK_F_SCSI); |
aa659be3 | 575 | |
da3dcefa | 576 | if (s->blk.config_wce) { |
8a873ba7 SH |
577 | features |= (1 << VIRTIO_BLK_F_CONFIG_WCE); |
578 | } | |
aa659be3 | 579 | if (bdrv_enable_write_cache(s->bs)) |
13e3dce0 PB |
580 | features |= (1 << VIRTIO_BLK_F_WCE); |
581 | ||
c79662f7 NS |
582 | if (bdrv_is_read_only(s->bs)) |
583 | features |= 1 << VIRTIO_BLK_F_RO; | |
1063b8b1 CH |
584 | |
585 | return features; | |
6e02c38d AL |
586 | } |
587 | ||
9315cbfd PB |
588 | static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status) |
589 | { | |
1cc91b7d | 590 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
9315cbfd PB |
591 | uint32_t features; |
592 | ||
392808b4 | 593 | #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE |
cf139388 SH |
594 | if (s->dataplane && !(status & (VIRTIO_CONFIG_S_DRIVER | |
595 | VIRTIO_CONFIG_S_DRIVER_OK))) { | |
392808b4 SH |
596 | virtio_blk_data_plane_stop(s->dataplane); |
597 | } | |
598 | #endif | |
599 | ||
9315cbfd PB |
600 | if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) { |
601 | return; | |
602 | } | |
603 | ||
604 | features = vdev->guest_features; | |
ef5bc962 PB |
605 | |
606 | /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send | |
607 | * cache flushes. Thus, the "auto writethrough" behavior is never | |
608 | * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature. | |
609 | * Leaving it enabled would break the following sequence: | |
610 | * | |
611 | * Guest started with "-drive cache=writethrough" | |
612 | * Guest sets status to 0 | |
613 | * Guest sets DRIVER bit in status field | |
614 | * Guest reads host features (WCE=0, CONFIG_WCE=1) | |
615 | * Guest writes guest features (WCE=0, CONFIG_WCE=1) | |
616 | * Guest writes 1 to the WCE configuration field (writeback mode) | |
617 | * Guest sets DRIVER_OK bit in status field | |
618 | * | |
619 | * s->bs would erroneously be placed in writethrough mode. | |
620 | */ | |
621 | if (!(features & (1 << VIRTIO_BLK_F_CONFIG_WCE))) { | |
6d7e73d6 FZ |
622 | aio_context_acquire(bdrv_get_aio_context(s->bs)); |
623 | bdrv_set_enable_write_cache(s->bs, | |
624 | !!(features & (1 << VIRTIO_BLK_F_WCE))); | |
625 | aio_context_release(bdrv_get_aio_context(s->bs)); | |
ef5bc962 | 626 | } |
9315cbfd PB |
627 | } |
628 | ||
6e02c38d AL |
629 | static void virtio_blk_save(QEMUFile *f, void *opaque) |
630 | { | |
b2b295a7 | 631 | VirtIODevice *vdev = VIRTIO_DEVICE(opaque); |
869a5c6d | 632 | |
1cc91b7d | 633 | virtio_save(vdev, f); |
b2b295a7 | 634 | } |
869a5c6d | 635 | |
b2b295a7 GK |
636 | static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f) |
637 | { | |
638 | VirtIOBlock *s = VIRTIO_BLK(vdev); | |
639 | VirtIOBlockReq *req = s->rq; | |
640 | ||
869a5c6d AL |
641 | while (req) { |
642 | qemu_put_sbyte(f, 1); | |
f897bf75 | 643 | qemu_put_buffer(f, (unsigned char *)&req->elem, |
671ec3f0 | 644 | sizeof(VirtQueueElement)); |
869a5c6d AL |
645 | req = req->next; |
646 | } | |
647 | qemu_put_sbyte(f, 0); | |
6e02c38d AL |
648 | } |
649 | ||
650 | static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id) | |
651 | { | |
652 | VirtIOBlock *s = opaque; | |
1cc91b7d | 653 | VirtIODevice *vdev = VIRTIO_DEVICE(s); |
6e02c38d | 654 | |
869a5c6d | 655 | if (version_id != 2) |
6e02c38d AL |
656 | return -EINVAL; |
657 | ||
b2b295a7 GK |
658 | return virtio_load(vdev, f, version_id); |
659 | } | |
660 | ||
661 | static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f, | |
662 | int version_id) | |
663 | { | |
664 | VirtIOBlock *s = VIRTIO_BLK(vdev); | |
2a633c46 | 665 | |
869a5c6d AL |
666 | while (qemu_get_sbyte(f)) { |
667 | VirtIOBlockReq *req = virtio_blk_alloc_request(s); | |
f897bf75 | 668 | qemu_get_buffer(f, (unsigned char *)&req->elem, |
671ec3f0 | 669 | sizeof(VirtQueueElement)); |
869a5c6d | 670 | req->next = s->rq; |
20a81e4d | 671 | s->rq = req; |
b6a4805b | 672 | |
f897bf75 SH |
673 | virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr, |
674 | req->elem.in_num, 1); | |
675 | virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr, | |
676 | req->elem.out_num, 0); | |
869a5c6d | 677 | } |
6e02c38d AL |
678 | |
679 | return 0; | |
680 | } | |
681 | ||
145feb17 | 682 | static void virtio_blk_resize(void *opaque) |
e5051fc7 | 683 | { |
1cc91b7d | 684 | VirtIODevice *vdev = VIRTIO_DEVICE(opaque); |
e5051fc7 | 685 | |
1cc91b7d | 686 | virtio_notify_config(vdev); |
e5051fc7 CH |
687 | } |
688 | ||
0e49de52 | 689 | static const BlockDevOps virtio_block_ops = { |
145feb17 | 690 | .resize_cb = virtio_blk_resize, |
0e49de52 MA |
691 | }; |
692 | ||
84db52d0 SH |
693 | #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE |
694 | /* Disable dataplane thread during live migration since it does not | |
695 | * update the dirty memory bitmap yet. | |
696 | */ | |
697 | static void virtio_blk_migration_state_changed(Notifier *notifier, void *data) | |
698 | { | |
699 | VirtIOBlock *s = container_of(notifier, VirtIOBlock, | |
700 | migration_state_notifier); | |
701 | MigrationState *mig = data; | |
3ffeeef7 | 702 | Error *err = NULL; |
84db52d0 SH |
703 | |
704 | if (migration_in_setup(mig)) { | |
705 | if (!s->dataplane) { | |
706 | return; | |
707 | } | |
708 | virtio_blk_data_plane_destroy(s->dataplane); | |
709 | s->dataplane = NULL; | |
710 | } else if (migration_has_finished(mig) || | |
711 | migration_has_failed(mig)) { | |
712 | if (s->dataplane) { | |
713 | return; | |
714 | } | |
715 | bdrv_drain_all(); /* complete in-flight non-dataplane requests */ | |
716 | virtio_blk_data_plane_create(VIRTIO_DEVICE(s), &s->blk, | |
3ffeeef7 AF |
717 | &s->dataplane, &err); |
718 | if (err != NULL) { | |
719 | error_report("%s", error_get_pretty(err)); | |
720 | error_free(err); | |
721 | } | |
84db52d0 SH |
722 | } |
723 | } | |
724 | #endif /* CONFIG_VIRTIO_BLK_DATA_PLANE */ | |
725 | ||
75884afd | 726 | static void virtio_blk_device_realize(DeviceState *dev, Error **errp) |
1c028ddf | 727 | { |
75884afd | 728 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
179b417e | 729 | VirtIOBlock *s = VIRTIO_BLK(dev); |
05ff6865 | 730 | VirtIOBlkConf *blk = &(s->blk); |
3ffeeef7 AF |
731 | #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE |
732 | Error *err = NULL; | |
733 | #endif | |
6e02c38d | 734 | static int virtio_blk_id; |
cf21e106 | 735 | |
12c5674b | 736 | if (!blk->conf.bs) { |
75884afd AF |
737 | error_setg(errp, "drive property not set"); |
738 | return; | |
d75d25e3 | 739 | } |
12c5674b | 740 | if (!bdrv_is_inserted(blk->conf.bs)) { |
75884afd AF |
741 | error_setg(errp, "Device needs media, but drive is empty"); |
742 | return; | |
98f28ad7 | 743 | } |
d75d25e3 | 744 | |
911525db | 745 | blkconf_serial(&blk->conf, &blk->serial); |
ef5bc962 | 746 | s->original_wce = bdrv_enable_write_cache(blk->conf.bs); |
b7eb0c9f | 747 | if (blkconf_geometry(&blk->conf, NULL, 65535, 255, 255) < 0) { |
75884afd AF |
748 | error_setg(errp, "Error setting geometry"); |
749 | return; | |
b7eb0c9f | 750 | } |
a8686a9b | 751 | |
05ff6865 FK |
752 | virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK, |
753 | sizeof(struct virtio_blk_config)); | |
6e02c38d | 754 | |
12c5674b PB |
755 | s->bs = blk->conf.bs; |
756 | s->conf = &blk->conf; | |
869a5c6d | 757 | s->rq = NULL; |
1573a35d | 758 | s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1; |
e63e7fde | 759 | |
05ff6865 | 760 | s->vq = virtio_add_queue(vdev, 128, virtio_blk_handle_output); |
bf4bd461 | 761 | s->complete_request = virtio_blk_complete_request; |
392808b4 | 762 | #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE |
3ffeeef7 AF |
763 | virtio_blk_data_plane_create(vdev, blk, &s->dataplane, &err); |
764 | if (err != NULL) { | |
75884afd | 765 | error_propagate(errp, err); |
6a1a8cc7 | 766 | virtio_cleanup(vdev); |
75884afd | 767 | return; |
392808b4 | 768 | } |
84db52d0 SH |
769 | s->migration_state_notifier.notify = virtio_blk_migration_state_changed; |
770 | add_migration_state_change_notifier(&s->migration_state_notifier); | |
392808b4 | 771 | #endif |
6e02c38d | 772 | |
69b302b2 | 773 | s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s); |
179b417e | 774 | register_savevm(dev, "virtio-blk", virtio_blk_id++, 2, |
6e02c38d | 775 | virtio_blk_save, virtio_blk_load, s); |
0e49de52 | 776 | bdrv_set_dev_ops(s->bs, &virtio_block_ops, s); |
1b7fd729 | 777 | bdrv_set_guest_block_size(s->bs, s->conf->logical_block_size); |
6e02c38d | 778 | |
af239a62 | 779 | bdrv_iostatus_enable(s->bs); |
1ca4d09a | 780 | |
179b417e | 781 | add_boot_device_path(s->conf->bootindex, dev, "/disk@0,0"); |
1c028ddf FK |
782 | } |
783 | ||
306ec6c3 | 784 | static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp) |
1c028ddf | 785 | { |
306ec6c3 AF |
786 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
787 | VirtIOBlock *s = VIRTIO_BLK(dev); | |
788 | ||
1c028ddf | 789 | #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE |
84db52d0 | 790 | remove_migration_state_change_notifier(&s->migration_state_notifier); |
1c028ddf FK |
791 | virtio_blk_data_plane_destroy(s->dataplane); |
792 | s->dataplane = NULL; | |
793 | #endif | |
794 | qemu_del_vm_change_state_handler(s->change); | |
306ec6c3 | 795 | unregister_savevm(dev, "virtio-blk", s); |
1c028ddf | 796 | blockdev_mark_auto_del(s->bs); |
6a1a8cc7 | 797 | virtio_cleanup(vdev); |
1c028ddf FK |
798 | } |
799 | ||
467b3f33 SH |
800 | static void virtio_blk_instance_init(Object *obj) |
801 | { | |
802 | VirtIOBlock *s = VIRTIO_BLK(obj); | |
803 | ||
804 | object_property_add_link(obj, "iothread", TYPE_IOTHREAD, | |
805 | (Object **)&s->blk.iothread, | |
806 | qdev_prop_allow_set_link_before_realize, | |
807 | OBJ_PROP_LINK_UNREF_ON_RELEASE, NULL); | |
808 | } | |
809 | ||
1c028ddf | 810 | static Property virtio_blk_properties[] = { |
32a877e4 SH |
811 | DEFINE_BLOCK_PROPERTIES(VirtIOBlock, blk.conf), |
812 | DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, blk.conf), | |
813 | DEFINE_PROP_STRING("serial", VirtIOBlock, blk.serial), | |
814 | DEFINE_PROP_BIT("config-wce", VirtIOBlock, blk.config_wce, 0, true), | |
32a877e4 SH |
815 | #ifdef __linux__ |
816 | DEFINE_PROP_BIT("scsi", VirtIOBlock, blk.scsi, 0, true), | |
817 | #endif | |
818 | #ifdef CONFIG_VIRTIO_BLK_DATA_PLANE | |
819 | DEFINE_PROP_BIT("x-data-plane", VirtIOBlock, blk.data_plane, 0, false), | |
820 | #endif | |
1c028ddf FK |
821 | DEFINE_PROP_END_OF_LIST(), |
822 | }; | |
823 | ||
824 | static void virtio_blk_class_init(ObjectClass *klass, void *data) | |
825 | { | |
826 | DeviceClass *dc = DEVICE_CLASS(klass); | |
827 | VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); | |
75884afd | 828 | |
1c028ddf | 829 | dc->props = virtio_blk_properties; |
125ee0ed | 830 | set_bit(DEVICE_CATEGORY_STORAGE, dc->categories); |
75884afd | 831 | vdc->realize = virtio_blk_device_realize; |
306ec6c3 | 832 | vdc->unrealize = virtio_blk_device_unrealize; |
1c028ddf FK |
833 | vdc->get_config = virtio_blk_update_config; |
834 | vdc->set_config = virtio_blk_set_config; | |
835 | vdc->get_features = virtio_blk_get_features; | |
836 | vdc->set_status = virtio_blk_set_status; | |
837 | vdc->reset = virtio_blk_reset; | |
b2b295a7 GK |
838 | vdc->save = virtio_blk_save_device; |
839 | vdc->load = virtio_blk_load_device; | |
1c028ddf FK |
840 | } |
841 | ||
842 | static const TypeInfo virtio_device_info = { | |
843 | .name = TYPE_VIRTIO_BLK, | |
844 | .parent = TYPE_VIRTIO_DEVICE, | |
845 | .instance_size = sizeof(VirtIOBlock), | |
467b3f33 | 846 | .instance_init = virtio_blk_instance_init, |
1c028ddf FK |
847 | .class_init = virtio_blk_class_init, |
848 | }; | |
849 | ||
850 | static void virtio_register_types(void) | |
851 | { | |
852 | type_register_static(&virtio_device_info); | |
853 | } | |
854 | ||
855 | type_init(virtio_register_types) |