]>
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 | ||
80c71a24 | 14 | #include "qemu/osdep.h" |
da34e65c | 15 | #include "qapi/error.h" |
827805a2 | 16 | #include "qemu/iov.h" |
0b8fa32f | 17 | #include "qemu/module.h" |
1de7afc9 | 18 | #include "qemu/error-report.h" |
6d519a5f | 19 | #include "trace.h" |
0d09e41a | 20 | #include "hw/block/block.h" |
9c17d615 | 21 | #include "sysemu/blockdev.h" |
0d09e41a | 22 | #include "hw/virtio/virtio-blk.h" |
52b53c04 | 23 | #include "dataplane/virtio-blk.h" |
08e2c9f1 | 24 | #include "scsi/constants.h" |
1063b8b1 CH |
25 | #ifdef __linux__ |
26 | # include <scsi/sg.h> | |
27 | #endif | |
0d09e41a | 28 | #include "hw/virtio/virtio-bus.h" |
ca77ee28 | 29 | #include "migration/qemu-file-types.h" |
783d1897 | 30 | #include "hw/virtio/virtio-access.h" |
6e02c38d | 31 | |
20764be0 | 32 | /* Config size before the discard support (hide associated config fields) */ |
42824b4d CL |
33 | #define VIRTIO_BLK_CFG_SIZE offsetof(struct virtio_blk_config, \ |
34 | max_discard_sectors) | |
20764be0 SG |
35 | /* |
36 | * Starting from the discard feature, we can use this array to properly | |
37 | * set the config size depending on the features enabled. | |
38 | */ | |
39 | static VirtIOFeature feature_sizes[] = { | |
40 | {.flags = 1ULL << VIRTIO_BLK_F_DISCARD, | |
41 | .end = virtio_endof(struct virtio_blk_config, discard_sector_alignment)}, | |
42 | {.flags = 1ULL << VIRTIO_BLK_F_WRITE_ZEROES, | |
43 | .end = virtio_endof(struct virtio_blk_config, write_zeroes_may_unmap)}, | |
44 | {} | |
45 | }; | |
46 | ||
47 | static void virtio_blk_set_config_size(VirtIOBlock *s, uint64_t host_features) | |
48 | { | |
49 | s->config_size = MAX(VIRTIO_BLK_CFG_SIZE, | |
50 | virtio_feature_get_config_size(feature_sizes, host_features)); | |
51 | ||
52 | assert(s->config_size <= sizeof(struct virtio_blk_config)); | |
53 | } | |
42824b4d | 54 | |
d14dde5e GK |
55 | static void virtio_blk_init_request(VirtIOBlock *s, VirtQueue *vq, |
56 | VirtIOBlockReq *req) | |
671ec3f0 | 57 | { |
671ec3f0 | 58 | req->dev = s; |
edaffd9f | 59 | req->vq = vq; |
869d66af | 60 | req->qiov.size = 0; |
2a6cdd6d | 61 | req->in_len = 0; |
869d66af | 62 | req->next = NULL; |
95f7142a | 63 | req->mr_next = NULL; |
671ec3f0 FZ |
64 | } |
65 | ||
d14dde5e | 66 | static void virtio_blk_free_request(VirtIOBlockReq *req) |
671ec3f0 | 67 | { |
1d29b5b0 | 68 | g_free(req); |
671ec3f0 FZ |
69 | } |
70 | ||
03de2f52 | 71 | static void virtio_blk_req_complete(VirtIOBlockReq *req, unsigned char status) |
869a5c6d AL |
72 | { |
73 | VirtIOBlock *s = req->dev; | |
1cc91b7d | 74 | VirtIODevice *vdev = VIRTIO_DEVICE(s); |
869a5c6d | 75 | |
a576ceac | 76 | trace_virtio_blk_req_complete(vdev, req, status); |
6d519a5f | 77 | |
92e3c2a3 | 78 | stb_p(&req->in->status, status); |
edaffd9f | 79 | virtqueue_push(req->vq, &req->elem, req->in_len); |
eb41cf78 | 80 | if (s->dataplane_started && !s->dataplane_disabled) { |
edaffd9f | 81 | virtio_blk_data_plane_notify(s->dataplane, req->vq); |
03de2f52 | 82 | } else { |
edaffd9f | 83 | virtio_notify(vdev, req->vq); |
03de2f52 | 84 | } |
bf4bd461 FZ |
85 | } |
86 | ||
f35d68f0 | 87 | static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error, |
00f639fb | 88 | bool is_read, bool acct_failed) |
869a5c6d | 89 | { |
869a5c6d | 90 | VirtIOBlock *s = req->dev; |
9a6719d5 | 91 | BlockErrorAction action = blk_get_error_action(s->blk, is_read, error); |
869a5c6d | 92 | |
a589569f | 93 | if (action == BLOCK_ERROR_ACTION_STOP) { |
466138dc FZ |
94 | /* Break the link as the next request is going to be parsed from the |
95 | * ring again. Otherwise we may end up doing a double completion! */ | |
96 | req->mr_next = NULL; | |
869a5c6d AL |
97 | req->next = s->rq; |
98 | s->rq = req; | |
a589569f | 99 | } else if (action == BLOCK_ERROR_ACTION_REPORT) { |
869a5c6d | 100 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); |
00f639fb SG |
101 | if (acct_failed) { |
102 | block_acct_failed(blk_get_stats(s->blk), &req->acct); | |
103 | } | |
671ec3f0 | 104 | virtio_blk_free_request(req); |
869a5c6d AL |
105 | } |
106 | ||
4be74634 | 107 | blk_error_action(s->blk, action, is_read, error); |
a589569f | 108 | return action != BLOCK_ERROR_ACTION_IGNORE; |
869a5c6d AL |
109 | } |
110 | ||
6e02c38d AL |
111 | static void virtio_blk_rw_complete(void *opaque, int ret) |
112 | { | |
95f7142a | 113 | VirtIOBlockReq *next = opaque; |
b9e413dd | 114 | VirtIOBlock *s = next->dev; |
a576ceac | 115 | VirtIODevice *vdev = VIRTIO_DEVICE(s); |
95f7142a | 116 | |
b9e413dd | 117 | aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); |
95f7142a PL |
118 | while (next) { |
119 | VirtIOBlockReq *req = next; | |
120 | next = req->mr_next; | |
a576ceac | 121 | trace_virtio_blk_rw_complete(vdev, req, ret); |
95f7142a PL |
122 | |
123 | if (req->qiov.nalloc != -1) { | |
e61809ed | 124 | /* If nalloc is != -1 req->qiov is a local copy of the original |
9bb192a4 YB |
125 | * external iovec. It was allocated in submit_requests to be |
126 | * able to merge requests. */ | |
95f7142a PL |
127 | qemu_iovec_destroy(&req->qiov); |
128 | } | |
6e02c38d | 129 | |
95f7142a | 130 | if (ret) { |
bf4069fb | 131 | int p = virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type); |
95f7142a | 132 | bool is_read = !(p & VIRTIO_BLK_T_OUT); |
2a6cdd6d PB |
133 | /* Note that memory may be dirtied on read failure. If the |
134 | * virtio request is not completed here, as is the case for | |
135 | * BLOCK_ERROR_ACTION_STOP, the memory may not be copied | |
136 | * correctly during live migration. While this is ugly, | |
137 | * it is acceptable because the device is free to write to | |
138 | * the memory until the request is completed (which will | |
139 | * happen on the other side of the migration). | |
140 | */ | |
00f639fb | 141 | if (virtio_blk_handle_rw_error(req, -ret, is_read, true)) { |
95f7142a PL |
142 | continue; |
143 | } | |
144 | } | |
6d519a5f | 145 | |
95f7142a | 146 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); |
bf4069fb | 147 | block_acct_done(blk_get_stats(s->blk), &req->acct); |
95f7142a | 148 | virtio_blk_free_request(req); |
6e02c38d | 149 | } |
b9e413dd | 150 | aio_context_release(blk_get_aio_context(s->conf.conf.blk)); |
869a5c6d | 151 | } |
6e02c38d | 152 | |
aa659be3 CH |
153 | static void virtio_blk_flush_complete(void *opaque, int ret) |
154 | { | |
155 | VirtIOBlockReq *req = opaque; | |
b9e413dd | 156 | VirtIOBlock *s = req->dev; |
aa659be3 | 157 | |
b9e413dd | 158 | aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); |
8c269b54 | 159 | if (ret) { |
00f639fb | 160 | if (virtio_blk_handle_rw_error(req, -ret, 0, true)) { |
b9e413dd | 161 | goto out; |
8c269b54 KW |
162 | } |
163 | } | |
164 | ||
165 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); | |
9a6719d5 | 166 | block_acct_done(blk_get_stats(s->blk), &req->acct); |
671ec3f0 | 167 | virtio_blk_free_request(req); |
b9e413dd PB |
168 | |
169 | out: | |
170 | aio_context_release(blk_get_aio_context(s->conf.conf.blk)); | |
6e02c38d AL |
171 | } |
172 | ||
37b06f8d SG |
173 | static void virtio_blk_discard_write_zeroes_complete(void *opaque, int ret) |
174 | { | |
175 | VirtIOBlockReq *req = opaque; | |
176 | VirtIOBlock *s = req->dev; | |
177 | bool is_write_zeroes = (virtio_ldl_p(VIRTIO_DEVICE(s), &req->out.type) & | |
178 | ~VIRTIO_BLK_T_BARRIER) == VIRTIO_BLK_T_WRITE_ZEROES; | |
179 | ||
180 | aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); | |
181 | if (ret) { | |
182 | if (virtio_blk_handle_rw_error(req, -ret, false, is_write_zeroes)) { | |
183 | goto out; | |
184 | } | |
185 | } | |
186 | ||
187 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); | |
188 | if (is_write_zeroes) { | |
189 | block_acct_done(blk_get_stats(s->blk), &req->acct); | |
190 | } | |
191 | virtio_blk_free_request(req); | |
192 | ||
193 | out: | |
194 | aio_context_release(blk_get_aio_context(s->conf.conf.blk)); | |
195 | } | |
196 | ||
1dc936aa FZ |
197 | #ifdef __linux__ |
198 | ||
199 | typedef struct { | |
200 | VirtIOBlockReq *req; | |
201 | struct sg_io_hdr hdr; | |
202 | } VirtIOBlockIoctlReq; | |
203 | ||
204 | static void virtio_blk_ioctl_complete(void *opaque, int status) | |
205 | { | |
206 | VirtIOBlockIoctlReq *ioctl_req = opaque; | |
207 | VirtIOBlockReq *req = ioctl_req->req; | |
9d456654 PB |
208 | VirtIOBlock *s = req->dev; |
209 | VirtIODevice *vdev = VIRTIO_DEVICE(s); | |
1dc936aa FZ |
210 | struct virtio_scsi_inhdr *scsi; |
211 | struct sg_io_hdr *hdr; | |
212 | ||
213 | scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base; | |
214 | ||
215 | if (status) { | |
216 | status = VIRTIO_BLK_S_UNSUPP; | |
217 | virtio_stl_p(vdev, &scsi->errors, 255); | |
218 | goto out; | |
219 | } | |
220 | ||
221 | hdr = &ioctl_req->hdr; | |
222 | /* | |
223 | * From SCSI-Generic-HOWTO: "Some lower level drivers (e.g. ide-scsi) | |
224 | * clear the masked_status field [hence status gets cleared too, see | |
225 | * block/scsi_ioctl.c] even when a CHECK_CONDITION or COMMAND_TERMINATED | |
226 | * status has occurred. However they do set DRIVER_SENSE in driver_status | |
227 | * field. Also a (sb_len_wr > 0) indicates there is a sense buffer. | |
228 | */ | |
229 | if (hdr->status == 0 && hdr->sb_len_wr > 0) { | |
230 | hdr->status = CHECK_CONDITION; | |
231 | } | |
232 | ||
233 | virtio_stl_p(vdev, &scsi->errors, | |
234 | hdr->status | (hdr->msg_status << 8) | | |
235 | (hdr->host_status << 16) | (hdr->driver_status << 24)); | |
236 | virtio_stl_p(vdev, &scsi->residual, hdr->resid); | |
237 | virtio_stl_p(vdev, &scsi->sense_len, hdr->sb_len_wr); | |
238 | virtio_stl_p(vdev, &scsi->data_len, hdr->dxfer_len); | |
239 | ||
240 | out: | |
b9e413dd | 241 | aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); |
1dc936aa FZ |
242 | virtio_blk_req_complete(req, status); |
243 | virtio_blk_free_request(req); | |
b9e413dd | 244 | aio_context_release(blk_get_aio_context(s->conf.conf.blk)); |
1dc936aa FZ |
245 | g_free(ioctl_req); |
246 | } | |
247 | ||
248 | #endif | |
249 | ||
edaffd9f | 250 | static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s, VirtQueue *vq) |
6e02c38d | 251 | { |
edaffd9f | 252 | VirtIOBlockReq *req = virtqueue_pop(vq, sizeof(VirtIOBlockReq)); |
6e02c38d | 253 | |
51b19ebe | 254 | if (req) { |
edaffd9f | 255 | virtio_blk_init_request(s, vq, req); |
6e02c38d | 256 | } |
6e02c38d AL |
257 | return req; |
258 | } | |
259 | ||
75344fa4 | 260 | static int virtio_blk_handle_scsi_req(VirtIOBlockReq *req) |
1063b8b1 | 261 | { |
5a05cbee FZ |
262 | int status = VIRTIO_BLK_S_OK; |
263 | struct virtio_scsi_inhdr *scsi = NULL; | |
75344fa4 | 264 | VirtIOBlock *blk = req->dev; |
bf4069fb AR |
265 | VirtIODevice *vdev = VIRTIO_DEVICE(blk); |
266 | VirtQueueElement *elem = &req->elem; | |
783d1897 | 267 | |
47ce9ef7 | 268 | #ifdef __linux__ |
1063b8b1 | 269 | int i; |
1dc936aa | 270 | VirtIOBlockIoctlReq *ioctl_req; |
a209f461 | 271 | BlockAIOCB *acb; |
47ce9ef7 | 272 | #endif |
1063b8b1 CH |
273 | |
274 | /* | |
275 | * We require at least one output segment each for the virtio_blk_outhdr | |
276 | * and the SCSI command block. | |
277 | * | |
278 | * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr | |
279 | * and the sense buffer pointer in the input segments. | |
280 | */ | |
5a05cbee FZ |
281 | if (elem->out_num < 2 || elem->in_num < 3) { |
282 | status = VIRTIO_BLK_S_IOERR; | |
283 | goto fail; | |
1063b8b1 CH |
284 | } |
285 | ||
286 | /* | |
f34e73cd PB |
287 | * The scsi inhdr is placed in the second-to-last input segment, just |
288 | * before the regular inhdr. | |
1063b8b1 | 289 | */ |
5a05cbee | 290 | scsi = (void *)elem->in_sg[elem->in_num - 2].iov_base; |
f34e73cd | 291 | |
bbe8bd4d | 292 | if (!virtio_has_feature(blk->host_features, VIRTIO_BLK_F_SCSI)) { |
f34e73cd PB |
293 | status = VIRTIO_BLK_S_UNSUPP; |
294 | goto fail; | |
1063b8b1 CH |
295 | } |
296 | ||
297 | /* | |
f34e73cd | 298 | * No support for bidirection commands yet. |
1063b8b1 | 299 | */ |
5a05cbee | 300 | if (elem->out_num > 2 && elem->in_num > 3) { |
f34e73cd PB |
301 | status = VIRTIO_BLK_S_UNSUPP; |
302 | goto fail; | |
303 | } | |
1063b8b1 | 304 | |
f34e73cd | 305 | #ifdef __linux__ |
1dc936aa FZ |
306 | ioctl_req = g_new0(VirtIOBlockIoctlReq, 1); |
307 | ioctl_req->req = req; | |
308 | ioctl_req->hdr.interface_id = 'S'; | |
309 | ioctl_req->hdr.cmd_len = elem->out_sg[1].iov_len; | |
310 | ioctl_req->hdr.cmdp = elem->out_sg[1].iov_base; | |
311 | ioctl_req->hdr.dxfer_len = 0; | |
1063b8b1 | 312 | |
5a05cbee | 313 | if (elem->out_num > 2) { |
1063b8b1 CH |
314 | /* |
315 | * If there are more than the minimally required 2 output segments | |
316 | * there is write payload starting from the third iovec. | |
317 | */ | |
1dc936aa FZ |
318 | ioctl_req->hdr.dxfer_direction = SG_DXFER_TO_DEV; |
319 | ioctl_req->hdr.iovec_count = elem->out_num - 2; | |
1063b8b1 | 320 | |
1dc936aa FZ |
321 | for (i = 0; i < ioctl_req->hdr.iovec_count; i++) { |
322 | ioctl_req->hdr.dxfer_len += elem->out_sg[i + 2].iov_len; | |
323 | } | |
1063b8b1 | 324 | |
1dc936aa | 325 | ioctl_req->hdr.dxferp = elem->out_sg + 2; |
1063b8b1 | 326 | |
5a05cbee | 327 | } else if (elem->in_num > 3) { |
1063b8b1 CH |
328 | /* |
329 | * If we have more than 3 input segments the guest wants to actually | |
330 | * read data. | |
331 | */ | |
1dc936aa FZ |
332 | ioctl_req->hdr.dxfer_direction = SG_DXFER_FROM_DEV; |
333 | ioctl_req->hdr.iovec_count = elem->in_num - 3; | |
334 | for (i = 0; i < ioctl_req->hdr.iovec_count; i++) { | |
335 | ioctl_req->hdr.dxfer_len += elem->in_sg[i].iov_len; | |
336 | } | |
1063b8b1 | 337 | |
1dc936aa | 338 | ioctl_req->hdr.dxferp = elem->in_sg; |
1063b8b1 CH |
339 | } else { |
340 | /* | |
341 | * Some SCSI commands don't actually transfer any data. | |
342 | */ | |
1dc936aa | 343 | ioctl_req->hdr.dxfer_direction = SG_DXFER_NONE; |
1063b8b1 CH |
344 | } |
345 | ||
1dc936aa FZ |
346 | ioctl_req->hdr.sbp = elem->in_sg[elem->in_num - 3].iov_base; |
347 | ioctl_req->hdr.mx_sb_len = elem->in_sg[elem->in_num - 3].iov_len; | |
1063b8b1 | 348 | |
a209f461 FZ |
349 | acb = blk_aio_ioctl(blk->blk, SG_IO, &ioctl_req->hdr, |
350 | virtio_blk_ioctl_complete, ioctl_req); | |
351 | if (!acb) { | |
352 | g_free(ioctl_req); | |
353 | status = VIRTIO_BLK_S_UNSUPP; | |
354 | goto fail; | |
355 | } | |
1dc936aa | 356 | return -EINPROGRESS; |
1063b8b1 | 357 | #else |
f34e73cd PB |
358 | abort(); |
359 | #endif | |
360 | ||
361 | fail: | |
362 | /* Just put anything nonzero so that the ioctl fails in the guest. */ | |
5a05cbee | 363 | if (scsi) { |
783d1897 | 364 | virtio_stl_p(vdev, &scsi->errors, 255); |
5a05cbee FZ |
365 | } |
366 | return status; | |
367 | } | |
368 | ||
369 | static void virtio_blk_handle_scsi(VirtIOBlockReq *req) | |
370 | { | |
371 | int status; | |
372 | ||
75344fa4 | 373 | status = virtio_blk_handle_scsi_req(req); |
1dc936aa FZ |
374 | if (status != -EINPROGRESS) { |
375 | virtio_blk_req_complete(req, status); | |
376 | virtio_blk_free_request(req); | |
377 | } | |
1063b8b1 | 378 | } |
1063b8b1 | 379 | |
95f7142a PL |
380 | static inline void submit_requests(BlockBackend *blk, MultiReqBuffer *mrb, |
381 | int start, int num_reqs, int niov) | |
869a5c6d | 382 | { |
95f7142a PL |
383 | QEMUIOVector *qiov = &mrb->reqs[start]->qiov; |
384 | int64_t sector_num = mrb->reqs[start]->sector_num; | |
95f7142a PL |
385 | bool is_write = mrb->is_write; |
386 | ||
387 | if (num_reqs > 1) { | |
388 | int i; | |
389 | struct iovec *tmp_iov = qiov->iov; | |
390 | int tmp_niov = qiov->niov; | |
391 | ||
392 | /* mrb->reqs[start]->qiov was initialized from external so we can't | |
b5772fdd | 393 | * modify it here. We need to initialize it locally and then add the |
95f7142a PL |
394 | * external iovecs. */ |
395 | qemu_iovec_init(qiov, niov); | |
396 | ||
397 | for (i = 0; i < tmp_niov; i++) { | |
398 | qemu_iovec_add(qiov, tmp_iov[i].iov_base, tmp_iov[i].iov_len); | |
399 | } | |
400 | ||
401 | for (i = start + 1; i < start + num_reqs; i++) { | |
402 | qemu_iovec_concat(qiov, &mrb->reqs[i]->qiov, 0, | |
403 | mrb->reqs[i]->qiov.size); | |
404 | mrb->reqs[i - 1]->mr_next = mrb->reqs[i]; | |
95f7142a | 405 | } |
95f7142a | 406 | |
a576ceac SH |
407 | trace_virtio_blk_submit_multireq(VIRTIO_DEVICE(mrb->reqs[start]->dev), |
408 | mrb, start, num_reqs, | |
b5772fdd EB |
409 | sector_num << BDRV_SECTOR_BITS, |
410 | qiov->size, is_write); | |
95f7142a PL |
411 | block_acct_merge_done(blk_get_stats(blk), |
412 | is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ, | |
413 | num_reqs - 1); | |
414 | } | |
91553dcc | 415 | |
95f7142a | 416 | if (is_write) { |
b5772fdd EB |
417 | blk_aio_pwritev(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0, |
418 | virtio_blk_rw_complete, mrb->reqs[start]); | |
95f7142a | 419 | } else { |
b5772fdd EB |
420 | blk_aio_preadv(blk, sector_num << BDRV_SECTOR_BITS, qiov, 0, |
421 | virtio_blk_rw_complete, mrb->reqs[start]); | |
95f7142a PL |
422 | } |
423 | } | |
424 | ||
425 | static int multireq_compare(const void *a, const void *b) | |
426 | { | |
427 | const VirtIOBlockReq *req1 = *(VirtIOBlockReq **)a, | |
428 | *req2 = *(VirtIOBlockReq **)b; | |
429 | ||
430 | /* | |
431 | * Note that we can't simply subtract sector_num1 from sector_num2 | |
432 | * here as that could overflow the return value. | |
433 | */ | |
434 | if (req1->sector_num > req2->sector_num) { | |
435 | return 1; | |
436 | } else if (req1->sector_num < req2->sector_num) { | |
437 | return -1; | |
438 | } else { | |
439 | return 0; | |
440 | } | |
441 | } | |
442 | ||
d14dde5e | 443 | static void virtio_blk_submit_multireq(BlockBackend *blk, MultiReqBuffer *mrb) |
95f7142a PL |
444 | { |
445 | int i = 0, start = 0, num_reqs = 0, niov = 0, nb_sectors = 0; | |
5def6b80 | 446 | uint32_t max_transfer; |
95f7142a PL |
447 | int64_t sector_num = 0; |
448 | ||
449 | if (mrb->num_reqs == 1) { | |
450 | submit_requests(blk, mrb, 0, 1, -1); | |
451 | mrb->num_reqs = 0; | |
c20fd872 CH |
452 | return; |
453 | } | |
454 | ||
5def6b80 | 455 | max_transfer = blk_get_max_transfer(mrb->reqs[0]->dev->blk); |
95f7142a PL |
456 | |
457 | qsort(mrb->reqs, mrb->num_reqs, sizeof(*mrb->reqs), | |
458 | &multireq_compare); | |
459 | ||
460 | for (i = 0; i < mrb->num_reqs; i++) { | |
461 | VirtIOBlockReq *req = mrb->reqs[i]; | |
462 | if (num_reqs > 0) { | |
49cffbc6 GA |
463 | /* |
464 | * NOTE: We cannot merge the requests in below situations: | |
465 | * 1. requests are not sequential | |
466 | * 2. merge would exceed maximum number of IOVs | |
467 | * 3. merge would exceed maximum transfer length of backend device | |
468 | */ | |
469 | if (sector_num + nb_sectors != req->sector_num || | |
222565f6 | 470 | niov > blk_get_max_iov(blk) - req->qiov.niov || |
5def6b80 EB |
471 | req->qiov.size > max_transfer || |
472 | nb_sectors > (max_transfer - | |
473 | req->qiov.size) / BDRV_SECTOR_SIZE) { | |
95f7142a PL |
474 | submit_requests(blk, mrb, start, num_reqs, niov); |
475 | num_reqs = 0; | |
91553dcc KW |
476 | } |
477 | } | |
95f7142a PL |
478 | |
479 | if (num_reqs == 0) { | |
480 | sector_num = req->sector_num; | |
481 | nb_sectors = niov = 0; | |
482 | start = i; | |
483 | } | |
484 | ||
485 | nb_sectors += req->qiov.size / BDRV_SECTOR_SIZE; | |
486 | niov += req->qiov.niov; | |
487 | num_reqs++; | |
91553dcc | 488 | } |
c20fd872 | 489 | |
95f7142a PL |
490 | submit_requests(blk, mrb, start, num_reqs, niov); |
491 | mrb->num_reqs = 0; | |
91553dcc | 492 | } |
87b245db | 493 | |
c20fd872 | 494 | static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
aa659be3 | 495 | { |
bf4069fb AR |
496 | VirtIOBlock *s = req->dev; |
497 | ||
498 | block_acct_start(blk_get_stats(s->blk), &req->acct, 0, | |
5366d0c8 | 499 | BLOCK_ACCT_FLUSH); |
a597e79c | 500 | |
618fbb84 CH |
501 | /* |
502 | * Make sure all outstanding writes are posted to the backing device. | |
503 | */ | |
95f7142a | 504 | if (mrb->is_write && mrb->num_reqs > 0) { |
bf4069fb | 505 | virtio_blk_submit_multireq(s->blk, mrb); |
95f7142a | 506 | } |
bf4069fb | 507 | blk_aio_flush(s->blk, virtio_blk_flush_complete, req); |
aa659be3 CH |
508 | } |
509 | ||
d0e14376 MA |
510 | static bool virtio_blk_sect_range_ok(VirtIOBlock *dev, |
511 | uint64_t sector, size_t size) | |
512 | { | |
3c2daac0 MA |
513 | uint64_t nb_sectors = size >> BDRV_SECTOR_BITS; |
514 | uint64_t total_sectors; | |
515 | ||
75af1f34 | 516 | if (nb_sectors > BDRV_REQUEST_MAX_SECTORS) { |
95f7142a PL |
517 | return false; |
518 | } | |
d0e14376 MA |
519 | if (sector & dev->sector_mask) { |
520 | return false; | |
521 | } | |
2a30307f | 522 | if (size % dev->conf.conf.logical_block_size) { |
d0e14376 MA |
523 | return false; |
524 | } | |
4be74634 | 525 | blk_get_geometry(dev->blk, &total_sectors); |
3c2daac0 MA |
526 | if (sector > total_sectors || nb_sectors > total_sectors - sector) { |
527 | return false; | |
528 | } | |
d0e14376 MA |
529 | return true; |
530 | } | |
531 | ||
37b06f8d SG |
532 | static uint8_t virtio_blk_handle_discard_write_zeroes(VirtIOBlockReq *req, |
533 | struct virtio_blk_discard_write_zeroes *dwz_hdr, bool is_write_zeroes) | |
534 | { | |
535 | VirtIOBlock *s = req->dev; | |
536 | VirtIODevice *vdev = VIRTIO_DEVICE(s); | |
537 | uint64_t sector; | |
538 | uint32_t num_sectors, flags, max_sectors; | |
539 | uint8_t err_status; | |
540 | int bytes; | |
541 | ||
542 | sector = virtio_ldq_p(vdev, &dwz_hdr->sector); | |
543 | num_sectors = virtio_ldl_p(vdev, &dwz_hdr->num_sectors); | |
544 | flags = virtio_ldl_p(vdev, &dwz_hdr->flags); | |
545 | max_sectors = is_write_zeroes ? s->conf.max_write_zeroes_sectors : | |
546 | s->conf.max_discard_sectors; | |
547 | ||
548 | /* | |
549 | * max_sectors is at most BDRV_REQUEST_MAX_SECTORS, this check | |
550 | * make us sure that "num_sectors << BDRV_SECTOR_BITS" can fit in | |
551 | * the integer variable. | |
552 | */ | |
553 | if (unlikely(num_sectors > max_sectors)) { | |
554 | err_status = VIRTIO_BLK_S_IOERR; | |
555 | goto err; | |
556 | } | |
557 | ||
558 | bytes = num_sectors << BDRV_SECTOR_BITS; | |
559 | ||
560 | if (unlikely(!virtio_blk_sect_range_ok(s, sector, bytes))) { | |
561 | err_status = VIRTIO_BLK_S_IOERR; | |
562 | goto err; | |
563 | } | |
564 | ||
565 | /* | |
566 | * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for discard | |
567 | * and write zeroes commands if any unknown flag is set. | |
568 | */ | |
569 | if (unlikely(flags & ~VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { | |
570 | err_status = VIRTIO_BLK_S_UNSUPP; | |
571 | goto err; | |
572 | } | |
573 | ||
574 | if (is_write_zeroes) { /* VIRTIO_BLK_T_WRITE_ZEROES */ | |
575 | int blk_aio_flags = 0; | |
576 | ||
577 | if (flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP) { | |
578 | blk_aio_flags |= BDRV_REQ_MAY_UNMAP; | |
579 | } | |
580 | ||
581 | block_acct_start(blk_get_stats(s->blk), &req->acct, bytes, | |
582 | BLOCK_ACCT_WRITE); | |
583 | ||
584 | blk_aio_pwrite_zeroes(s->blk, sector << BDRV_SECTOR_BITS, | |
585 | bytes, blk_aio_flags, | |
586 | virtio_blk_discard_write_zeroes_complete, req); | |
587 | } else { /* VIRTIO_BLK_T_DISCARD */ | |
588 | /* | |
589 | * The device MUST set the status byte to VIRTIO_BLK_S_UNSUPP for | |
590 | * discard commands if the unmap flag is set. | |
591 | */ | |
592 | if (unlikely(flags & VIRTIO_BLK_WRITE_ZEROES_FLAG_UNMAP)) { | |
593 | err_status = VIRTIO_BLK_S_UNSUPP; | |
594 | goto err; | |
595 | } | |
596 | ||
597 | blk_aio_pdiscard(s->blk, sector << BDRV_SECTOR_BITS, bytes, | |
598 | virtio_blk_discard_write_zeroes_complete, req); | |
599 | } | |
600 | ||
601 | return VIRTIO_BLK_S_OK; | |
602 | ||
603 | err: | |
604 | if (is_write_zeroes) { | |
605 | block_acct_invalid(blk_get_stats(s->blk), BLOCK_ACCT_WRITE); | |
606 | } | |
607 | return err_status; | |
608 | } | |
609 | ||
20ea686a | 610 | static int virtio_blk_handle_request(VirtIOBlockReq *req, MultiReqBuffer *mrb) |
bc6694d4 | 611 | { |
92e3c2a3 | 612 | uint32_t type; |
f897bf75 | 613 | struct iovec *in_iov = req->elem.in_sg; |
5636da76 | 614 | struct iovec *out_iov = req->elem.out_sg; |
f897bf75 SH |
615 | unsigned in_num = req->elem.in_num; |
616 | unsigned out_num = req->elem.out_num; | |
20ea686a GK |
617 | VirtIOBlock *s = req->dev; |
618 | VirtIODevice *vdev = VIRTIO_DEVICE(s); | |
92e3c2a3 | 619 | |
f897bf75 | 620 | if (req->elem.out_num < 1 || req->elem.in_num < 1) { |
20ea686a GK |
621 | virtio_error(vdev, "virtio-blk missing headers"); |
622 | return -1; | |
bc6694d4 KW |
623 | } |
624 | ||
5636da76 | 625 | if (unlikely(iov_to_buf(out_iov, out_num, 0, &req->out, |
827805a2 | 626 | sizeof(req->out)) != sizeof(req->out))) { |
20ea686a GK |
627 | virtio_error(vdev, "virtio-blk request outhdr too short"); |
628 | return -1; | |
827805a2 | 629 | } |
ee17e848 | 630 | |
5636da76 | 631 | iov_discard_front(&out_iov, &out_num, sizeof(req->out)); |
ee17e848 | 632 | |
12048545 | 633 | if (in_iov[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) { |
20ea686a GK |
634 | virtio_error(vdev, "virtio-blk request inhdr too short"); |
635 | return -1; | |
ee17e848 FZ |
636 | } |
637 | ||
2a6cdd6d PB |
638 | /* We always touch the last byte, so just see how big in_iov is. */ |
639 | req->in_len = iov_size(in_iov, in_num); | |
ee17e848 FZ |
640 | req->in = (void *)in_iov[in_num - 1].iov_base |
641 | + in_iov[in_num - 1].iov_len | |
642 | - sizeof(struct virtio_blk_inhdr); | |
643 | iov_discard_back(in_iov, &in_num, sizeof(struct virtio_blk_inhdr)); | |
bc6694d4 | 644 | |
9a6719d5 | 645 | type = virtio_ldl_p(vdev, &req->out.type); |
92e3c2a3 | 646 | |
95f7142a | 647 | /* VIRTIO_BLK_T_OUT defines the command direction. VIRTIO_BLK_T_BARRIER |
631b22ea | 648 | * is an optional flag. Although a guest should not send this flag if |
95f7142a PL |
649 | * not negotiated we ignored it in the past. So keep ignoring it. */ |
650 | switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) { | |
651 | case VIRTIO_BLK_T_IN: | |
652 | { | |
653 | bool is_write = type & VIRTIO_BLK_T_OUT; | |
9a6719d5 | 654 | req->sector_num = virtio_ldq_p(vdev, &req->out.sector); |
95f7142a PL |
655 | |
656 | if (is_write) { | |
5636da76 | 657 | qemu_iovec_init_external(&req->qiov, out_iov, out_num); |
a576ceac | 658 | trace_virtio_blk_handle_write(vdev, req, req->sector_num, |
95f7142a PL |
659 | req->qiov.size / BDRV_SECTOR_SIZE); |
660 | } else { | |
661 | qemu_iovec_init_external(&req->qiov, in_iov, in_num); | |
a576ceac | 662 | trace_virtio_blk_handle_read(vdev, req, req->sector_num, |
95f7142a PL |
663 | req->qiov.size / BDRV_SECTOR_SIZE); |
664 | } | |
665 | ||
9a6719d5 | 666 | if (!virtio_blk_sect_range_ok(s, req->sector_num, req->qiov.size)) { |
95f7142a | 667 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); |
9a6719d5 | 668 | block_acct_invalid(blk_get_stats(s->blk), |
01762e03 | 669 | is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ); |
95f7142a | 670 | virtio_blk_free_request(req); |
20ea686a | 671 | return 0; |
95f7142a PL |
672 | } |
673 | ||
9a6719d5 | 674 | block_acct_start(blk_get_stats(s->blk), &req->acct, req->qiov.size, |
95f7142a PL |
675 | is_write ? BLOCK_ACCT_WRITE : BLOCK_ACCT_READ); |
676 | ||
677 | /* merge would exceed maximum number of requests or IO direction | |
678 | * changes */ | |
679 | if (mrb->num_reqs > 0 && (mrb->num_reqs == VIRTIO_BLK_MAX_MERGE_REQS || | |
c99495ac | 680 | is_write != mrb->is_write || |
9a6719d5 SG |
681 | !s->conf.request_merging)) { |
682 | virtio_blk_submit_multireq(s->blk, mrb); | |
95f7142a PL |
683 | } |
684 | ||
685 | assert(mrb->num_reqs < VIRTIO_BLK_MAX_MERGE_REQS); | |
686 | mrb->reqs[mrb->num_reqs++] = req; | |
687 | mrb->is_write = is_write; | |
688 | break; | |
689 | } | |
690 | case VIRTIO_BLK_T_FLUSH: | |
c20fd872 | 691 | virtio_blk_handle_flush(req, mrb); |
95f7142a PL |
692 | break; |
693 | case VIRTIO_BLK_T_SCSI_CMD: | |
bc6694d4 | 694 | virtio_blk_handle_scsi(req); |
95f7142a PL |
695 | break; |
696 | case VIRTIO_BLK_T_GET_ID: | |
697 | { | |
a8686a9b MA |
698 | /* |
699 | * NB: per existing s/n string convention the string is | |
700 | * terminated by '\0' only when shorter than buffer. | |
701 | */ | |
2a30307f | 702 | const char *serial = s->conf.serial ? s->conf.serial : ""; |
a83ceea8 MM |
703 | size_t size = MIN(strlen(serial) + 1, |
704 | MIN(iov_size(in_iov, in_num), | |
705 | VIRTIO_BLK_ID_BYTES)); | |
706 | iov_from_buf(in_iov, in_num, 0, serial, size); | |
2930b313 | 707 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); |
671ec3f0 | 708 | virtio_blk_free_request(req); |
95f7142a PL |
709 | break; |
710 | } | |
37b06f8d SG |
711 | /* |
712 | * VIRTIO_BLK_T_DISCARD and VIRTIO_BLK_T_WRITE_ZEROES are defined with | |
713 | * VIRTIO_BLK_T_OUT flag set. We masked this flag in the switch statement, | |
714 | * so we must mask it for these requests, then we will check if it is set. | |
715 | */ | |
716 | case VIRTIO_BLK_T_DISCARD & ~VIRTIO_BLK_T_OUT: | |
717 | case VIRTIO_BLK_T_WRITE_ZEROES & ~VIRTIO_BLK_T_OUT: | |
718 | { | |
719 | struct virtio_blk_discard_write_zeroes dwz_hdr; | |
720 | size_t out_len = iov_size(out_iov, out_num); | |
721 | bool is_write_zeroes = (type & ~VIRTIO_BLK_T_BARRIER) == | |
722 | VIRTIO_BLK_T_WRITE_ZEROES; | |
723 | uint8_t err_status; | |
724 | ||
725 | /* | |
726 | * Unsupported if VIRTIO_BLK_T_OUT is not set or the request contains | |
727 | * more than one segment. | |
728 | */ | |
729 | if (unlikely(!(type & VIRTIO_BLK_T_OUT) || | |
730 | out_len > sizeof(dwz_hdr))) { | |
731 | virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); | |
732 | virtio_blk_free_request(req); | |
733 | return 0; | |
734 | } | |
735 | ||
736 | if (unlikely(iov_to_buf(out_iov, out_num, 0, &dwz_hdr, | |
737 | sizeof(dwz_hdr)) != sizeof(dwz_hdr))) { | |
738 | virtio_error(vdev, "virtio-blk discard/write_zeroes header" | |
739 | " too short"); | |
740 | return -1; | |
741 | } | |
742 | ||
743 | err_status = virtio_blk_handle_discard_write_zeroes(req, &dwz_hdr, | |
744 | is_write_zeroes); | |
745 | if (err_status != VIRTIO_BLK_S_OK) { | |
746 | virtio_blk_req_complete(req, err_status); | |
747 | virtio_blk_free_request(req); | |
748 | } | |
749 | ||
750 | break; | |
751 | } | |
95f7142a | 752 | default: |
9e72c450 | 753 | virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); |
671ec3f0 | 754 | virtio_blk_free_request(req); |
bc6694d4 | 755 | } |
20ea686a | 756 | return 0; |
bc6694d4 KW |
757 | } |
758 | ||
07931698 | 759 | bool virtio_blk_handle_vq(VirtIOBlock *s, VirtQueue *vq) |
6e02c38d | 760 | { |
6e02c38d | 761 | VirtIOBlockReq *req; |
95f7142a | 762 | MultiReqBuffer mrb = {}; |
07931698 | 763 | bool progress = false; |
6e02c38d | 764 | |
9d456654 | 765 | aio_context_acquire(blk_get_aio_context(s->blk)); |
fc73548e SH |
766 | blk_io_plug(s->blk); |
767 | ||
9ef9d402 SH |
768 | do { |
769 | virtio_queue_set_notification(vq, 0); | |
770 | ||
771 | while ((req = virtio_blk_get_request(s, vq))) { | |
07931698 | 772 | progress = true; |
9ef9d402 SH |
773 | if (virtio_blk_handle_request(req, &mrb)) { |
774 | virtqueue_detach_element(req->vq, &req->elem, 0); | |
775 | virtio_blk_free_request(req); | |
776 | break; | |
777 | } | |
20ea686a | 778 | } |
9ef9d402 SH |
779 | |
780 | virtio_queue_set_notification(vq, 1); | |
781 | } while (!virtio_queue_empty(vq)); | |
91553dcc | 782 | |
95f7142a PL |
783 | if (mrb.num_reqs) { |
784 | virtio_blk_submit_multireq(s->blk, &mrb); | |
785 | } | |
fc73548e SH |
786 | |
787 | blk_io_unplug(s->blk); | |
9d456654 | 788 | aio_context_release(blk_get_aio_context(s->blk)); |
07931698 FZ |
789 | return progress; |
790 | } | |
791 | ||
792 | static void virtio_blk_handle_output_do(VirtIOBlock *s, VirtQueue *vq) | |
793 | { | |
794 | virtio_blk_handle_vq(s, vq); | |
6e02c38d AL |
795 | } |
796 | ||
8a2fad57 MT |
797 | static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq) |
798 | { | |
799 | VirtIOBlock *s = (VirtIOBlock *)vdev; | |
800 | ||
801 | if (s->dataplane) { | |
802 | /* Some guests kick before setting VIRTIO_CONFIG_S_DRIVER_OK so start | |
803 | * dataplane here instead of waiting for .set_status(). | |
804 | */ | |
9ffe337c | 805 | virtio_device_start_ioeventfd(vdev); |
8a2fad57 MT |
806 | if (!s->dataplane_disabled) { |
807 | return; | |
808 | } | |
809 | } | |
07931698 | 810 | virtio_blk_handle_output_do(s, vq); |
8a2fad57 MT |
811 | } |
812 | ||
213189ab | 813 | static void virtio_blk_dma_restart_bh(void *opaque) |
869a5c6d AL |
814 | { |
815 | VirtIOBlock *s = opaque; | |
816 | VirtIOBlockReq *req = s->rq; | |
95f7142a | 817 | MultiReqBuffer mrb = {}; |
869a5c6d | 818 | |
213189ab MA |
819 | qemu_bh_delete(s->bh); |
820 | s->bh = NULL; | |
869a5c6d AL |
821 | |
822 | s->rq = NULL; | |
823 | ||
1919631e | 824 | aio_context_acquire(blk_get_aio_context(s->conf.conf.blk)); |
869a5c6d | 825 | while (req) { |
1bdb176a | 826 | VirtIOBlockReq *next = req->next; |
20ea686a GK |
827 | if (virtio_blk_handle_request(req, &mrb)) { |
828 | /* Device is now broken and won't do any processing until it gets | |
829 | * reset. Already queued requests will be lost: let's purge them. | |
830 | */ | |
831 | while (req) { | |
832 | next = req->next; | |
833 | virtqueue_detach_element(req->vq, &req->elem, 0); | |
834 | virtio_blk_free_request(req); | |
835 | req = next; | |
836 | } | |
837 | break; | |
838 | } | |
1bdb176a | 839 | req = next; |
869a5c6d | 840 | } |
f1b52868 | 841 | |
95f7142a PL |
842 | if (mrb.num_reqs) { |
843 | virtio_blk_submit_multireq(s->blk, &mrb); | |
844 | } | |
680f2002 | 845 | blk_dec_in_flight(s->conf.conf.blk); |
1919631e | 846 | aio_context_release(blk_get_aio_context(s->conf.conf.blk)); |
869a5c6d AL |
847 | } |
848 | ||
1dfb4dd9 LC |
849 | static void virtio_blk_dma_restart_cb(void *opaque, int running, |
850 | RunState state) | |
213189ab MA |
851 | { |
852 | VirtIOBlock *s = opaque; | |
853 | ||
392808b4 | 854 | if (!running) { |
213189ab | 855 | return; |
392808b4 | 856 | } |
213189ab MA |
857 | |
858 | if (!s->bh) { | |
680f2002 KW |
859 | /* FIXME The data plane is not started yet, so these requests are |
860 | * processed in the main thread. */ | |
4be74634 | 861 | s->bh = aio_bh_new(blk_get_aio_context(s->conf.conf.blk), |
4407c1c5 | 862 | virtio_blk_dma_restart_bh, s); |
680f2002 | 863 | blk_inc_in_flight(s->conf.conf.blk); |
213189ab MA |
864 | qemu_bh_schedule(s->bh); |
865 | } | |
866 | } | |
867 | ||
6e02c38d AL |
868 | static void virtio_blk_reset(VirtIODevice *vdev) |
869 | { | |
1cc91b7d | 870 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
6e40b3bf | 871 | AioContext *ctx; |
26307f6a | 872 | VirtIOBlockReq *req; |
392808b4 | 873 | |
6e40b3bf AY |
874 | ctx = blk_get_aio_context(s->blk); |
875 | aio_context_acquire(ctx); | |
876 | blk_drain(s->blk); | |
877 | ||
26307f6a FZ |
878 | /* We drop queued requests after blk_drain() because blk_drain() itself can |
879 | * produce them. */ | |
880 | while (s->rq) { | |
881 | req = s->rq; | |
882 | s->rq = req->next; | |
97b93c8a | 883 | virtqueue_detach_element(req->vq, &req->elem, 0); |
26307f6a FZ |
884 | virtio_blk_free_request(req); |
885 | } | |
886 | ||
6e40b3bf AY |
887 | aio_context_release(ctx); |
888 | ||
9ffe337c | 889 | assert(!s->dataplane_started); |
4be74634 | 890 | blk_set_enable_write_cache(s->blk, s->original_wce); |
6e02c38d AL |
891 | } |
892 | ||
bf011293 | 893 | /* coalesce internal state, copy to pci i/o region 0 |
894 | */ | |
6e02c38d AL |
895 | static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config) |
896 | { | |
1cc91b7d | 897 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
2a30307f | 898 | BlockConf *conf = &s->conf.conf; |
6e02c38d AL |
899 | struct virtio_blk_config blkcfg; |
900 | uint64_t capacity; | |
17d0bc01 | 901 | int64_t length; |
f7516731 | 902 | int blk_size = conf->logical_block_size; |
6e02c38d | 903 | |
4be74634 | 904 | blk_get_geometry(s->blk, &capacity); |
5c5dafdc | 905 | memset(&blkcfg, 0, sizeof(blkcfg)); |
783d1897 RR |
906 | virtio_stq_p(vdev, &blkcfg.capacity, capacity); |
907 | virtio_stl_p(vdev, &blkcfg.seg_max, 128 - 2); | |
907eb3e5 | 908 | virtio_stw_p(vdev, &blkcfg.geometry.cylinders, conf->cyls); |
783d1897 | 909 | virtio_stl_p(vdev, &blkcfg.blk_size, blk_size); |
f7516731 MA |
910 | virtio_stw_p(vdev, &blkcfg.min_io_size, conf->min_io_size / blk_size); |
911 | virtio_stw_p(vdev, &blkcfg.opt_io_size, conf->opt_io_size / blk_size); | |
907eb3e5 | 912 | blkcfg.geometry.heads = conf->heads; |
136be99e CB |
913 | /* |
914 | * We must ensure that the block device capacity is a multiple of | |
e03ba136 | 915 | * the logical block size. If that is not the case, let's use |
136be99e CB |
916 | * sector_mask to adopt the geometry to have a correct picture. |
917 | * For those devices where the capacity is ok for the given geometry | |
e03ba136 | 918 | * we don't touch the sector value of the geometry, since some devices |
136be99e CB |
919 | * (like s390 dasd) need a specific value. Here the capacity is already |
920 | * cyls*heads*secs*blk_size and the sector value is not block size | |
921 | * divided by 512 - instead it is the amount of blk_size blocks | |
922 | * per track (cylinder). | |
923 | */ | |
17d0bc01 SH |
924 | length = blk_getlength(s->blk); |
925 | if (length > 0 && length / conf->heads / conf->secs % blk_size) { | |
907eb3e5 | 926 | blkcfg.geometry.sectors = conf->secs & ~s->sector_mask; |
136be99e | 927 | } else { |
907eb3e5 | 928 | blkcfg.geometry.sectors = conf->secs; |
136be99e | 929 | } |
c7085da7 | 930 | blkcfg.size_max = 0; |
f7516731 | 931 | blkcfg.physical_block_exp = get_physical_block_exp(conf); |
9752c371 | 932 | blkcfg.alignment_offset = 0; |
4be74634 | 933 | blkcfg.wce = blk_enable_write_cache(s->blk); |
2f270590 | 934 | virtio_stw_p(vdev, &blkcfg.num_queues, s->conf.num_queues); |
37b06f8d SG |
935 | if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD)) { |
936 | virtio_stl_p(vdev, &blkcfg.max_discard_sectors, | |
937 | s->conf.max_discard_sectors); | |
938 | virtio_stl_p(vdev, &blkcfg.discard_sector_alignment, | |
939 | blk_size >> BDRV_SECTOR_BITS); | |
940 | /* | |
941 | * We support only one segment per request since multiple segments | |
942 | * are not widely used and there are no userspace APIs that allow | |
943 | * applications to submit multiple segments in a single call. | |
944 | */ | |
945 | virtio_stl_p(vdev, &blkcfg.max_discard_seg, 1); | |
946 | } | |
947 | if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES)) { | |
948 | virtio_stl_p(vdev, &blkcfg.max_write_zeroes_sectors, | |
949 | s->conf.max_write_zeroes_sectors); | |
950 | blkcfg.write_zeroes_may_unmap = 1; | |
951 | virtio_stl_p(vdev, &blkcfg.max_write_zeroes_seg, 1); | |
952 | } | |
20764be0 | 953 | memcpy(config, &blkcfg, s->config_size); |
6e02c38d AL |
954 | } |
955 | ||
13e3dce0 PB |
956 | static void virtio_blk_set_config(VirtIODevice *vdev, const uint8_t *config) |
957 | { | |
1cc91b7d | 958 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
13e3dce0 PB |
959 | struct virtio_blk_config blkcfg; |
960 | ||
20764be0 | 961 | memcpy(&blkcfg, config, s->config_size); |
6d7e73d6 | 962 | |
4be74634 MA |
963 | aio_context_acquire(blk_get_aio_context(s->blk)); |
964 | blk_set_enable_write_cache(s->blk, blkcfg.wce != 0); | |
965 | aio_context_release(blk_get_aio_context(s->blk)); | |
13e3dce0 PB |
966 | } |
967 | ||
9d5b731d JW |
968 | static uint64_t virtio_blk_get_features(VirtIODevice *vdev, uint64_t features, |
969 | Error **errp) | |
6e02c38d | 970 | { |
1cc91b7d | 971 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
1063b8b1 | 972 | |
bbe8bd4d SG |
973 | /* Firstly sync all virtio-blk possible supported features */ |
974 | features |= s->host_features; | |
975 | ||
0cd09c3a CH |
976 | virtio_add_feature(&features, VIRTIO_BLK_F_SEG_MAX); |
977 | virtio_add_feature(&features, VIRTIO_BLK_F_GEOMETRY); | |
978 | virtio_add_feature(&features, VIRTIO_BLK_F_TOPOLOGY); | |
979 | virtio_add_feature(&features, VIRTIO_BLK_F_BLK_SIZE); | |
95129d6f | 980 | if (virtio_has_feature(features, VIRTIO_F_VERSION_1)) { |
bbe8bd4d | 981 | if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_SCSI)) { |
efb8206c JW |
982 | error_setg(errp, "Please set scsi=off for virtio-blk devices in order to use virtio 1.0"); |
983 | return 0; | |
984 | } | |
efb8206c | 985 | } else { |
c9b11f97 | 986 | virtio_clear_feature(&features, VIRTIO_F_ANY_LAYOUT); |
efb8206c JW |
987 | virtio_add_feature(&features, VIRTIO_BLK_F_SCSI); |
988 | } | |
aa659be3 | 989 | |
4be74634 | 990 | if (blk_enable_write_cache(s->blk)) { |
0cd09c3a | 991 | virtio_add_feature(&features, VIRTIO_BLK_F_WCE); |
4be74634 MA |
992 | } |
993 | if (blk_is_read_only(s->blk)) { | |
0cd09c3a | 994 | virtio_add_feature(&features, VIRTIO_BLK_F_RO); |
4be74634 | 995 | } |
2f270590 SH |
996 | if (s->conf.num_queues > 1) { |
997 | virtio_add_feature(&features, VIRTIO_BLK_F_MQ); | |
998 | } | |
1063b8b1 CH |
999 | |
1000 | return features; | |
6e02c38d AL |
1001 | } |
1002 | ||
9315cbfd PB |
1003 | static void virtio_blk_set_status(VirtIODevice *vdev, uint8_t status) |
1004 | { | |
1cc91b7d | 1005 | VirtIOBlock *s = VIRTIO_BLK(vdev); |
9315cbfd | 1006 | |
9ffe337c PB |
1007 | if (!(status & (VIRTIO_CONFIG_S_DRIVER | VIRTIO_CONFIG_S_DRIVER_OK))) { |
1008 | assert(!s->dataplane_started); | |
392808b4 | 1009 | } |
392808b4 | 1010 | |
9315cbfd PB |
1011 | if (!(status & VIRTIO_CONFIG_S_DRIVER_OK)) { |
1012 | return; | |
1013 | } | |
1014 | ||
ef5bc962 PB |
1015 | /* A guest that supports VIRTIO_BLK_F_CONFIG_WCE must be able to send |
1016 | * cache flushes. Thus, the "auto writethrough" behavior is never | |
1017 | * necessary for guests that support the VIRTIO_BLK_F_CONFIG_WCE feature. | |
1018 | * Leaving it enabled would break the following sequence: | |
1019 | * | |
1020 | * Guest started with "-drive cache=writethrough" | |
1021 | * Guest sets status to 0 | |
1022 | * Guest sets DRIVER bit in status field | |
1023 | * Guest reads host features (WCE=0, CONFIG_WCE=1) | |
1024 | * Guest writes guest features (WCE=0, CONFIG_WCE=1) | |
1025 | * Guest writes 1 to the WCE configuration field (writeback mode) | |
1026 | * Guest sets DRIVER_OK bit in status field | |
1027 | * | |
4be74634 | 1028 | * s->blk would erroneously be placed in writethrough mode. |
ef5bc962 | 1029 | */ |
95129d6f | 1030 | if (!virtio_vdev_has_feature(vdev, VIRTIO_BLK_F_CONFIG_WCE)) { |
4be74634 MA |
1031 | aio_context_acquire(blk_get_aio_context(s->blk)); |
1032 | blk_set_enable_write_cache(s->blk, | |
95129d6f CH |
1033 | virtio_vdev_has_feature(vdev, |
1034 | VIRTIO_BLK_F_WCE)); | |
4be74634 | 1035 | aio_context_release(blk_get_aio_context(s->blk)); |
ef5bc962 | 1036 | } |
9315cbfd PB |
1037 | } |
1038 | ||
b2b295a7 GK |
1039 | static void virtio_blk_save_device(VirtIODevice *vdev, QEMUFile *f) |
1040 | { | |
1041 | VirtIOBlock *s = VIRTIO_BLK(vdev); | |
1042 | VirtIOBlockReq *req = s->rq; | |
1043 | ||
869a5c6d AL |
1044 | while (req) { |
1045 | qemu_put_sbyte(f, 1); | |
30d8bf6d SH |
1046 | |
1047 | if (s->conf.num_queues > 1) { | |
1048 | qemu_put_be32(f, virtio_get_queue_index(req->vq)); | |
1049 | } | |
1050 | ||
ab281c17 | 1051 | qemu_put_virtqueue_element(f, &req->elem); |
869a5c6d AL |
1052 | req = req->next; |
1053 | } | |
1054 | qemu_put_sbyte(f, 0); | |
6e02c38d AL |
1055 | } |
1056 | ||
b2b295a7 GK |
1057 | static int virtio_blk_load_device(VirtIODevice *vdev, QEMUFile *f, |
1058 | int version_id) | |
1059 | { | |
1060 | VirtIOBlock *s = VIRTIO_BLK(vdev); | |
2a633c46 | 1061 | |
869a5c6d | 1062 | while (qemu_get_sbyte(f)) { |
30d8bf6d SH |
1063 | unsigned nvqs = s->conf.num_queues; |
1064 | unsigned vq_idx = 0; | |
ab281c17 | 1065 | VirtIOBlockReq *req; |
30d8bf6d SH |
1066 | |
1067 | if (nvqs > 1) { | |
1068 | vq_idx = qemu_get_be32(f); | |
1069 | ||
1070 | if (vq_idx >= nvqs) { | |
1071 | error_report("Invalid virtqueue index in request list: %#x", | |
1072 | vq_idx); | |
1073 | return -EINVAL; | |
1074 | } | |
1075 | } | |
1076 | ||
8607f5c3 | 1077 | req = qemu_get_virtqueue_element(vdev, f, sizeof(VirtIOBlockReq)); |
30d8bf6d | 1078 | virtio_blk_init_request(s, virtio_get_queue(vdev, vq_idx), req); |
869a5c6d | 1079 | req->next = s->rq; |
20a81e4d | 1080 | s->rq = req; |
869a5c6d | 1081 | } |
6e02c38d AL |
1082 | |
1083 | return 0; | |
1084 | } | |
1085 | ||
145feb17 | 1086 | static void virtio_blk_resize(void *opaque) |
e5051fc7 | 1087 | { |
1cc91b7d | 1088 | VirtIODevice *vdev = VIRTIO_DEVICE(opaque); |
e5051fc7 | 1089 | |
1cc91b7d | 1090 | virtio_notify_config(vdev); |
e5051fc7 CH |
1091 | } |
1092 | ||
0e49de52 | 1093 | static const BlockDevOps virtio_block_ops = { |
145feb17 | 1094 | .resize_cb = virtio_blk_resize, |
0e49de52 MA |
1095 | }; |
1096 | ||
75884afd | 1097 | static void virtio_blk_device_realize(DeviceState *dev, Error **errp) |
1c028ddf | 1098 | { |
75884afd | 1099 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
179b417e | 1100 | VirtIOBlock *s = VIRTIO_BLK(dev); |
2a30307f | 1101 | VirtIOBlkConf *conf = &s->conf; |
3ffeeef7 | 1102 | Error *err = NULL; |
2f270590 | 1103 | unsigned i; |
cf21e106 | 1104 | |
4be74634 | 1105 | if (!conf->conf.blk) { |
75884afd AF |
1106 | error_setg(errp, "drive property not set"); |
1107 | return; | |
d75d25e3 | 1108 | } |
4be74634 | 1109 | if (!blk_is_inserted(conf->conf.blk)) { |
75884afd AF |
1110 | error_setg(errp, "Device needs media, but drive is empty"); |
1111 | return; | |
98f28ad7 | 1112 | } |
2f270590 SH |
1113 | if (!conf->num_queues) { |
1114 | error_setg(errp, "num-queues property must be larger than 0"); | |
1115 | return; | |
1116 | } | |
6040aedd MK |
1117 | if (!is_power_of_2(conf->queue_size) || |
1118 | conf->queue_size > VIRTQUEUE_MAX_SIZE) { | |
1119 | error_setg(errp, "invalid queue-size property (%" PRIu16 "), " | |
1120 | "must be a power of 2 (max %d)", | |
1121 | conf->queue_size, VIRTQUEUE_MAX_SIZE); | |
1122 | return; | |
1123 | } | |
d75d25e3 | 1124 | |
ceff3e1f MZ |
1125 | if (!blkconf_apply_backend_options(&conf->conf, |
1126 | blk_is_read_only(conf->conf.blk), true, | |
1127 | errp)) { | |
a17c17a2 KW |
1128 | return; |
1129 | } | |
4be74634 | 1130 | s->original_wce = blk_enable_write_cache(conf->conf.blk); |
ceff3e1f | 1131 | if (!blkconf_geometry(&conf->conf, NULL, 65535, 255, 255, errp)) { |
75884afd | 1132 | return; |
b7eb0c9f | 1133 | } |
ceff3e1f | 1134 | |
0eb28a42 | 1135 | blkconf_blocksizes(&conf->conf); |
a8686a9b | 1136 | |
0a75b60c MK |
1137 | if (conf->conf.logical_block_size > |
1138 | conf->conf.physical_block_size) { | |
1139 | error_setg(errp, | |
1140 | "logical_block_size > physical_block_size not supported"); | |
1141 | return; | |
1142 | } | |
1143 | ||
37b06f8d SG |
1144 | if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_DISCARD) && |
1145 | (!conf->max_discard_sectors || | |
1146 | conf->max_discard_sectors > BDRV_REQUEST_MAX_SECTORS)) { | |
1147 | error_setg(errp, "invalid max-discard-sectors property (%" PRIu32 ")" | |
1148 | ", must be between 1 and %d", | |
1149 | conf->max_discard_sectors, (int)BDRV_REQUEST_MAX_SECTORS); | |
1150 | return; | |
1151 | } | |
1152 | ||
1153 | if (virtio_has_feature(s->host_features, VIRTIO_BLK_F_WRITE_ZEROES) && | |
1154 | (!conf->max_write_zeroes_sectors || | |
1155 | conf->max_write_zeroes_sectors > BDRV_REQUEST_MAX_SECTORS)) { | |
1156 | error_setg(errp, "invalid max-write-zeroes-sectors property (%" PRIu32 | |
1157 | "), must be between 1 and %d", | |
1158 | conf->max_write_zeroes_sectors, | |
1159 | (int)BDRV_REQUEST_MAX_SECTORS); | |
1160 | return; | |
1161 | } | |
1162 | ||
20764be0 SG |
1163 | virtio_blk_set_config_size(s, s->host_features); |
1164 | ||
1165 | virtio_init(vdev, "virtio-blk", VIRTIO_ID_BLOCK, s->config_size); | |
6e02c38d | 1166 | |
4be74634 | 1167 | s->blk = conf->conf.blk; |
869a5c6d | 1168 | s->rq = NULL; |
2a30307f | 1169 | s->sector_mask = (s->conf.conf.logical_block_size / BDRV_SECTOR_SIZE) - 1; |
e63e7fde | 1170 | |
2f270590 | 1171 | for (i = 0; i < conf->num_queues; i++) { |
6040aedd | 1172 | virtio_add_queue(vdev, conf->queue_size, virtio_blk_handle_output); |
2f270590 | 1173 | } |
2a30307f | 1174 | virtio_blk_data_plane_create(vdev, conf, &s->dataplane, &err); |
3ffeeef7 | 1175 | if (err != NULL) { |
75884afd | 1176 | error_propagate(errp, err); |
6a1a8cc7 | 1177 | virtio_cleanup(vdev); |
75884afd | 1178 | return; |
392808b4 | 1179 | } |
6e02c38d | 1180 | |
69b302b2 | 1181 | s->change = qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s); |
4be74634 MA |
1182 | blk_set_dev_ops(s->blk, &virtio_block_ops, s); |
1183 | blk_set_guest_block_size(s->blk, s->conf.conf.logical_block_size); | |
6e02c38d | 1184 | |
4be74634 | 1185 | blk_iostatus_enable(s->blk); |
1c028ddf FK |
1186 | } |
1187 | ||
306ec6c3 | 1188 | static void virtio_blk_device_unrealize(DeviceState *dev, Error **errp) |
1c028ddf | 1189 | { |
306ec6c3 AF |
1190 | VirtIODevice *vdev = VIRTIO_DEVICE(dev); |
1191 | VirtIOBlock *s = VIRTIO_BLK(dev); | |
1192 | ||
1c028ddf FK |
1193 | virtio_blk_data_plane_destroy(s->dataplane); |
1194 | s->dataplane = NULL; | |
1c028ddf | 1195 | qemu_del_vm_change_state_handler(s->change); |
4be74634 | 1196 | blockdev_mark_auto_del(s->blk); |
6a1a8cc7 | 1197 | virtio_cleanup(vdev); |
1c028ddf FK |
1198 | } |
1199 | ||
467b3f33 SH |
1200 | static void virtio_blk_instance_init(Object *obj) |
1201 | { | |
1202 | VirtIOBlock *s = VIRTIO_BLK(obj); | |
1203 | ||
2a30307f | 1204 | device_add_bootindex_property(obj, &s->conf.conf.bootindex, |
3342ec32 GA |
1205 | "bootindex", "/disk@0,0", |
1206 | DEVICE(obj), NULL); | |
467b3f33 SH |
1207 | } |
1208 | ||
977a117f HP |
1209 | static const VMStateDescription vmstate_virtio_blk = { |
1210 | .name = "virtio-blk", | |
1211 | .minimum_version_id = 2, | |
1212 | .version_id = 2, | |
1213 | .fields = (VMStateField[]) { | |
1214 | VMSTATE_VIRTIO_DEVICE, | |
1215 | VMSTATE_END_OF_LIST() | |
1216 | }, | |
1217 | }; | |
bbded32c | 1218 | |
1c028ddf | 1219 | static Property virtio_blk_properties[] = { |
2a30307f | 1220 | DEFINE_BLOCK_PROPERTIES(VirtIOBlock, conf.conf), |
8c398252 | 1221 | DEFINE_BLOCK_ERROR_PROPERTIES(VirtIOBlock, conf.conf), |
2a30307f MA |
1222 | DEFINE_BLOCK_CHS_PROPERTIES(VirtIOBlock, conf.conf), |
1223 | DEFINE_PROP_STRING("serial", VirtIOBlock, conf.serial), | |
bbe8bd4d SG |
1224 | DEFINE_PROP_BIT64("config-wce", VirtIOBlock, host_features, |
1225 | VIRTIO_BLK_F_CONFIG_WCE, true), | |
32a877e4 | 1226 | #ifdef __linux__ |
bbe8bd4d SG |
1227 | DEFINE_PROP_BIT64("scsi", VirtIOBlock, host_features, |
1228 | VIRTIO_BLK_F_SCSI, false), | |
32a877e4 | 1229 | #endif |
c99495ac PL |
1230 | DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0, |
1231 | true), | |
2f270590 | 1232 | DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 1), |
6040aedd | 1233 | DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 128), |
d679ac09 FZ |
1234 | DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD, |
1235 | IOThread *), | |
5c81161f SG |
1236 | DEFINE_PROP_BIT64("discard", VirtIOBlock, host_features, |
1237 | VIRTIO_BLK_F_DISCARD, true), | |
1238 | DEFINE_PROP_BIT64("write-zeroes", VirtIOBlock, host_features, | |
1239 | VIRTIO_BLK_F_WRITE_ZEROES, true), | |
37b06f8d SG |
1240 | DEFINE_PROP_UINT32("max-discard-sectors", VirtIOBlock, |
1241 | conf.max_discard_sectors, BDRV_REQUEST_MAX_SECTORS), | |
1242 | DEFINE_PROP_UINT32("max-write-zeroes-sectors", VirtIOBlock, | |
1243 | conf.max_write_zeroes_sectors, BDRV_REQUEST_MAX_SECTORS), | |
1c028ddf FK |
1244 | DEFINE_PROP_END_OF_LIST(), |
1245 | }; | |
1246 | ||
1247 | static void virtio_blk_class_init(ObjectClass *klass, void *data) | |
1248 | { | |
1249 | DeviceClass *dc = DEVICE_CLASS(klass); | |
1250 | VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass); | |
75884afd | 1251 | |
1c028ddf | 1252 | dc->props = virtio_blk_properties; |
bbded32c | 1253 | dc->vmsd = &vmstate_virtio_blk; |
125ee0ed | 1254 | set_bit(DEVICE_CATEGORY_STORAGE, dc->categories); |
75884afd | 1255 | vdc->realize = virtio_blk_device_realize; |
306ec6c3 | 1256 | vdc->unrealize = virtio_blk_device_unrealize; |
1c028ddf FK |
1257 | vdc->get_config = virtio_blk_update_config; |
1258 | vdc->set_config = virtio_blk_set_config; | |
1259 | vdc->get_features = virtio_blk_get_features; | |
1260 | vdc->set_status = virtio_blk_set_status; | |
1261 | vdc->reset = virtio_blk_reset; | |
b2b295a7 GK |
1262 | vdc->save = virtio_blk_save_device; |
1263 | vdc->load = virtio_blk_load_device; | |
9ffe337c PB |
1264 | vdc->start_ioeventfd = virtio_blk_data_plane_start; |
1265 | vdc->stop_ioeventfd = virtio_blk_data_plane_stop; | |
1c028ddf FK |
1266 | } |
1267 | ||
b5c7ceaf | 1268 | static const TypeInfo virtio_blk_info = { |
1c028ddf FK |
1269 | .name = TYPE_VIRTIO_BLK, |
1270 | .parent = TYPE_VIRTIO_DEVICE, | |
1271 | .instance_size = sizeof(VirtIOBlock), | |
467b3f33 | 1272 | .instance_init = virtio_blk_instance_init, |
1c028ddf FK |
1273 | .class_init = virtio_blk_class_init, |
1274 | }; | |
1275 | ||
1276 | static void virtio_register_types(void) | |
1277 | { | |
b5c7ceaf | 1278 | type_register_static(&virtio_blk_info); |
1c028ddf FK |
1279 | } |
1280 | ||
1281 | type_init(virtio_register_types) |