]>
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 | ||
869a5c6d AL |
14 | #include <qemu-common.h> |
15 | #include <sysemu.h> | |
6e02c38d AL |
16 | #include "virtio-blk.h" |
17 | #include "block_int.h" | |
1063b8b1 CH |
18 | #ifdef __linux__ |
19 | # include <scsi/sg.h> | |
20 | #endif | |
6e02c38d AL |
21 | |
22 | typedef struct VirtIOBlock | |
23 | { | |
24 | VirtIODevice vdev; | |
25 | BlockDriverState *bs; | |
26 | VirtQueue *vq; | |
869a5c6d | 27 | void *rq; |
bf011293 | 28 | char serial_str[BLOCK_SERIAL_STRLEN + 1]; |
6e02c38d AL |
29 | } VirtIOBlock; |
30 | ||
31 | static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev) | |
32 | { | |
33 | return (VirtIOBlock *)vdev; | |
34 | } | |
35 | ||
bf011293 | 36 | /* store identify data in little endian format |
37 | */ | |
38 | static inline void put_le16(uint16_t *p, unsigned int v) | |
39 | { | |
40 | *p = cpu_to_le16(v); | |
41 | } | |
42 | ||
43 | /* copy to *dst from *src, nul pad dst tail as needed to len bytes | |
44 | */ | |
45 | static inline void padstr(char *dst, const char *src, int len) | |
46 | { | |
47 | while (len--) | |
48 | *dst++ = *src ? *src++ : '\0'; | |
49 | } | |
50 | ||
51 | /* setup simulated identify data as appropriate for virtio block device | |
52 | * | |
53 | * ref: AT Attachment 8 - ATA/ATAPI Command Set (ATA8-ACS) | |
54 | */ | |
55 | static inline void virtio_identify_template(struct virtio_blk_config *bc) | |
56 | { | |
57 | uint16_t *p = &bc->identify[0]; | |
58 | uint64_t lba_sectors = bc->capacity; | |
59 | ||
60 | memset(p, 0, sizeof(bc->identify)); | |
61 | put_le16(p + 0, 0x0); /* ATA device */ | |
62 | padstr((char *)(p + 23), QEMU_VERSION, 8); /* firmware revision */ | |
63 | padstr((char *)(p + 27), "QEMU VIRT_BLK", 40); /* model# */ | |
64 | put_le16(p + 47, 0x80ff); /* max xfer 255 sectors */ | |
65 | put_le16(p + 49, 0x0b00); /* support IORDY/LBA/DMA */ | |
66 | put_le16(p + 59, 0x1ff); /* cur xfer 255 sectors */ | |
67 | put_le16(p + 80, 0x1f0); /* support ATA8/7/6/5/4 */ | |
68 | put_le16(p + 81, 0x16); | |
69 | put_le16(p + 82, 0x400); | |
70 | put_le16(p + 83, 0x400); | |
71 | put_le16(p + 100, lba_sectors); | |
72 | put_le16(p + 101, lba_sectors >> 16); | |
73 | put_le16(p + 102, lba_sectors >> 32); | |
74 | put_le16(p + 103, lba_sectors >> 48); | |
75 | } | |
76 | ||
6e02c38d AL |
77 | typedef struct VirtIOBlockReq |
78 | { | |
79 | VirtIOBlock *dev; | |
80 | VirtQueueElement elem; | |
81 | struct virtio_blk_inhdr *in; | |
82 | struct virtio_blk_outhdr *out; | |
1063b8b1 | 83 | struct virtio_scsi_inhdr *scsi; |
d28a1b6e | 84 | QEMUIOVector qiov; |
869a5c6d | 85 | struct VirtIOBlockReq *next; |
6e02c38d AL |
86 | } VirtIOBlockReq; |
87 | ||
869a5c6d AL |
88 | static void virtio_blk_req_complete(VirtIOBlockReq *req, int status) |
89 | { | |
90 | VirtIOBlock *s = req->dev; | |
91 | ||
92 | req->in->status = status; | |
d28a1b6e | 93 | virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in)); |
869a5c6d AL |
94 | virtio_notify(&s->vdev, s->vq); |
95 | ||
869a5c6d AL |
96 | qemu_free(req); |
97 | } | |
98 | ||
99 | static int virtio_blk_handle_write_error(VirtIOBlockReq *req, int error) | |
100 | { | |
101 | BlockInterfaceErrorAction action = drive_get_onerror(req->dev->bs); | |
102 | VirtIOBlock *s = req->dev; | |
103 | ||
104 | if (action == BLOCK_ERR_IGNORE) | |
105 | return 0; | |
106 | ||
107 | if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC) | |
108 | || action == BLOCK_ERR_STOP_ANY) { | |
109 | req->next = s->rq; | |
110 | s->rq = req; | |
111 | vm_stop(0); | |
112 | } else { | |
113 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); | |
114 | } | |
115 | ||
116 | return 1; | |
117 | } | |
118 | ||
6e02c38d AL |
119 | static void virtio_blk_rw_complete(void *opaque, int ret) |
120 | { | |
121 | VirtIOBlockReq *req = opaque; | |
6e02c38d | 122 | |
d28a1b6e | 123 | if (ret && (req->out->type & VIRTIO_BLK_T_OUT)) { |
869a5c6d AL |
124 | if (virtio_blk_handle_write_error(req, -ret)) |
125 | return; | |
6e02c38d AL |
126 | } |
127 | ||
869a5c6d AL |
128 | virtio_blk_req_complete(req, VIRTIO_BLK_S_OK); |
129 | } | |
6e02c38d | 130 | |
869a5c6d AL |
131 | static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s) |
132 | { | |
133 | VirtIOBlockReq *req = qemu_mallocz(sizeof(*req)); | |
487414f1 | 134 | req->dev = s; |
869a5c6d | 135 | return req; |
6e02c38d AL |
136 | } |
137 | ||
138 | static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s) | |
139 | { | |
869a5c6d | 140 | VirtIOBlockReq *req = virtio_blk_alloc_request(s); |
6e02c38d | 141 | |
869a5c6d AL |
142 | if (req != NULL) { |
143 | if (!virtqueue_pop(s->vq, &req->elem)) { | |
144 | qemu_free(req); | |
145 | return NULL; | |
146 | } | |
6e02c38d AL |
147 | } |
148 | ||
149 | return req; | |
150 | } | |
151 | ||
1063b8b1 CH |
152 | #ifdef __linux__ |
153 | static void virtio_blk_handle_scsi(VirtIOBlockReq *req) | |
154 | { | |
155 | struct sg_io_hdr hdr; | |
156 | int ret, size = 0; | |
157 | int status; | |
158 | int i; | |
159 | ||
160 | /* | |
161 | * We require at least one output segment each for the virtio_blk_outhdr | |
162 | * and the SCSI command block. | |
163 | * | |
164 | * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr | |
165 | * and the sense buffer pointer in the input segments. | |
166 | */ | |
167 | if (req->elem.out_num < 2 || req->elem.in_num < 3) { | |
168 | virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR); | |
169 | return; | |
170 | } | |
171 | ||
172 | /* | |
173 | * No support for bidirection commands yet. | |
174 | */ | |
175 | if (req->elem.out_num > 2 && req->elem.in_num > 3) { | |
176 | virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); | |
177 | return; | |
178 | } | |
179 | ||
180 | /* | |
181 | * The scsi inhdr is placed in the second-to-last input segment, just | |
182 | * before the regular inhdr. | |
183 | */ | |
184 | req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base; | |
185 | size = sizeof(*req->in) + sizeof(*req->scsi); | |
186 | ||
187 | memset(&hdr, 0, sizeof(struct sg_io_hdr)); | |
188 | hdr.interface_id = 'S'; | |
189 | hdr.cmd_len = req->elem.out_sg[1].iov_len; | |
190 | hdr.cmdp = req->elem.out_sg[1].iov_base; | |
191 | hdr.dxfer_len = 0; | |
192 | ||
193 | if (req->elem.out_num > 2) { | |
194 | /* | |
195 | * If there are more than the minimally required 2 output segments | |
196 | * there is write payload starting from the third iovec. | |
197 | */ | |
198 | hdr.dxfer_direction = SG_DXFER_TO_DEV; | |
199 | hdr.iovec_count = req->elem.out_num - 2; | |
200 | ||
201 | for (i = 0; i < hdr.iovec_count; i++) | |
202 | hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len; | |
203 | ||
204 | hdr.dxferp = req->elem.out_sg + 2; | |
205 | ||
206 | } else if (req->elem.in_num > 3) { | |
207 | /* | |
208 | * If we have more than 3 input segments the guest wants to actually | |
209 | * read data. | |
210 | */ | |
211 | hdr.dxfer_direction = SG_DXFER_FROM_DEV; | |
212 | hdr.iovec_count = req->elem.in_num - 3; | |
213 | for (i = 0; i < hdr.iovec_count; i++) | |
214 | hdr.dxfer_len += req->elem.in_sg[i].iov_len; | |
215 | ||
216 | hdr.dxferp = req->elem.in_sg; | |
217 | size += hdr.dxfer_len; | |
218 | } else { | |
219 | /* | |
220 | * Some SCSI commands don't actually transfer any data. | |
221 | */ | |
222 | hdr.dxfer_direction = SG_DXFER_NONE; | |
223 | } | |
224 | ||
225 | hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base; | |
226 | hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len; | |
227 | size += hdr.mx_sb_len; | |
228 | ||
229 | ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr); | |
230 | if (ret) { | |
231 | status = VIRTIO_BLK_S_UNSUPP; | |
232 | hdr.status = ret; | |
233 | hdr.resid = hdr.dxfer_len; | |
234 | } else if (hdr.status) { | |
235 | status = VIRTIO_BLK_S_IOERR; | |
236 | } else { | |
237 | status = VIRTIO_BLK_S_OK; | |
238 | } | |
239 | ||
240 | req->scsi->errors = hdr.status; | |
241 | req->scsi->residual = hdr.resid; | |
242 | req->scsi->sense_len = hdr.sb_len_wr; | |
243 | req->scsi->data_len = hdr.dxfer_len; | |
244 | ||
245 | virtio_blk_req_complete(req, status); | |
246 | } | |
247 | #else | |
248 | static void virtio_blk_handle_scsi(VirtIOBlockReq *req) | |
249 | { | |
250 | virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP); | |
251 | } | |
252 | #endif /* __linux__ */ | |
253 | ||
d28a1b6e | 254 | static void virtio_blk_handle_write(VirtIOBlockReq *req) |
869a5c6d | 255 | { |
d28a1b6e AL |
256 | bdrv_aio_writev(req->dev->bs, req->out->sector, &req->qiov, |
257 | req->qiov.size / 512, virtio_blk_rw_complete, req); | |
258 | } | |
869a5c6d | 259 | |
d28a1b6e AL |
260 | static void virtio_blk_handle_read(VirtIOBlockReq *req) |
261 | { | |
262 | bdrv_aio_readv(req->dev->bs, req->out->sector, &req->qiov, | |
263 | req->qiov.size / 512, virtio_blk_rw_complete, req); | |
869a5c6d AL |
264 | } |
265 | ||
6e02c38d AL |
266 | static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq) |
267 | { | |
268 | VirtIOBlock *s = to_virtio_blk(vdev); | |
269 | VirtIOBlockReq *req; | |
270 | ||
271 | while ((req = virtio_blk_get_request(s))) { | |
6e02c38d AL |
272 | if (req->elem.out_num < 1 || req->elem.in_num < 1) { |
273 | fprintf(stderr, "virtio-blk missing headers\n"); | |
274 | exit(1); | |
275 | } | |
276 | ||
277 | if (req->elem.out_sg[0].iov_len < sizeof(*req->out) || | |
278 | req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) { | |
279 | fprintf(stderr, "virtio-blk header not in correct element\n"); | |
280 | exit(1); | |
281 | } | |
282 | ||
283 | req->out = (void *)req->elem.out_sg[0].iov_base; | |
284 | req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base; | |
285 | ||
286 | if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) { | |
1063b8b1 | 287 | virtio_blk_handle_scsi(req); |
6e02c38d | 288 | } else if (req->out->type & VIRTIO_BLK_T_OUT) { |
d28a1b6e AL |
289 | qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1], |
290 | req->elem.out_num - 1); | |
291 | virtio_blk_handle_write(req); | |
6e02c38d | 292 | } else { |
d28a1b6e AL |
293 | qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0], |
294 | req->elem.in_num - 1); | |
295 | virtio_blk_handle_read(req); | |
6e02c38d AL |
296 | } |
297 | } | |
298 | /* | |
299 | * FIXME: Want to check for completions before returning to guest mode, | |
300 | * so cached reads and writes are reported as quickly as possible. But | |
301 | * that should be done in the generic block layer. | |
302 | */ | |
303 | } | |
304 | ||
869a5c6d AL |
305 | static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason) |
306 | { | |
307 | VirtIOBlock *s = opaque; | |
308 | VirtIOBlockReq *req = s->rq; | |
309 | ||
310 | if (!running) | |
311 | return; | |
312 | ||
313 | s->rq = NULL; | |
314 | ||
315 | while (req) { | |
316 | virtio_blk_handle_write(req); | |
317 | req = req->next; | |
318 | } | |
319 | } | |
320 | ||
6e02c38d AL |
321 | static void virtio_blk_reset(VirtIODevice *vdev) |
322 | { | |
323 | /* | |
324 | * This should cancel pending requests, but can't do nicely until there | |
325 | * are per-device request lists. | |
326 | */ | |
327 | qemu_aio_flush(); | |
328 | } | |
329 | ||
bf011293 | 330 | /* coalesce internal state, copy to pci i/o region 0 |
331 | */ | |
6e02c38d AL |
332 | static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config) |
333 | { | |
334 | VirtIOBlock *s = to_virtio_blk(vdev); | |
335 | struct virtio_blk_config blkcfg; | |
336 | uint64_t capacity; | |
337 | int cylinders, heads, secs; | |
338 | ||
339 | bdrv_get_geometry(s->bs, &capacity); | |
340 | bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs); | |
5c5dafdc | 341 | memset(&blkcfg, 0, sizeof(blkcfg)); |
6e02c38d AL |
342 | stq_raw(&blkcfg.capacity, capacity); |
343 | stl_raw(&blkcfg.seg_max, 128 - 2); | |
344 | stw_raw(&blkcfg.cylinders, cylinders); | |
345 | blkcfg.heads = heads; | |
346 | blkcfg.sectors = secs; | |
c7085da7 | 347 | blkcfg.size_max = 0; |
bf011293 | 348 | virtio_identify_template(&blkcfg); |
349 | memcpy(&blkcfg.identify[VIRTIO_BLK_ID_SN], s->serial_str, | |
350 | VIRTIO_BLK_ID_SN_BYTES); | |
6e02c38d AL |
351 | memcpy(config, &blkcfg, sizeof(blkcfg)); |
352 | } | |
353 | ||
354 | static uint32_t virtio_blk_get_features(VirtIODevice *vdev) | |
355 | { | |
bf011293 | 356 | VirtIOBlock *s = to_virtio_blk(vdev); |
1063b8b1 CH |
357 | uint32_t features = 0; |
358 | ||
359 | features |= (1 << VIRTIO_BLK_F_SEG_MAX); | |
360 | features |= (1 << VIRTIO_BLK_F_GEOMETRY); | |
361 | #ifdef __linux__ | |
362 | features |= (1 << VIRTIO_BLK_F_SCSI); | |
363 | #endif | |
bf011293 | 364 | if (strcmp(s->serial_str, "0")) |
365 | features |= 1 << VIRTIO_BLK_F_IDENTIFY; | |
1063b8b1 CH |
366 | |
367 | return features; | |
6e02c38d AL |
368 | } |
369 | ||
370 | static void virtio_blk_save(QEMUFile *f, void *opaque) | |
371 | { | |
372 | VirtIOBlock *s = opaque; | |
869a5c6d AL |
373 | VirtIOBlockReq *req = s->rq; |
374 | ||
6e02c38d | 375 | virtio_save(&s->vdev, f); |
869a5c6d AL |
376 | |
377 | while (req) { | |
378 | qemu_put_sbyte(f, 1); | |
379 | qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem)); | |
380 | req = req->next; | |
381 | } | |
382 | qemu_put_sbyte(f, 0); | |
6e02c38d AL |
383 | } |
384 | ||
385 | static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id) | |
386 | { | |
387 | VirtIOBlock *s = opaque; | |
388 | ||
869a5c6d | 389 | if (version_id != 2) |
6e02c38d AL |
390 | return -EINVAL; |
391 | ||
392 | virtio_load(&s->vdev, f); | |
869a5c6d AL |
393 | while (qemu_get_sbyte(f)) { |
394 | VirtIOBlockReq *req = virtio_blk_alloc_request(s); | |
395 | qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem)); | |
396 | req->next = s->rq; | |
397 | s->rq = req->next; | |
398 | } | |
6e02c38d AL |
399 | |
400 | return 0; | |
401 | } | |
402 | ||
53c25cea | 403 | VirtIODevice *virtio_blk_init(DeviceState *dev) |
6e02c38d AL |
404 | { |
405 | VirtIOBlock *s; | |
406 | int cylinders, heads, secs; | |
407 | static int virtio_blk_id; | |
07e3af9a | 408 | BlockDriverState *bs; |
bf011293 | 409 | char *ps; |
cf21e106 | 410 | |
53c25cea PB |
411 | s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK, |
412 | sizeof(struct virtio_blk_config), | |
413 | sizeof(VirtIOBlock)); | |
6e02c38d | 414 | |
53c25cea | 415 | bs = qdev_init_bdrv(dev, IF_VIRTIO); |
6e02c38d AL |
416 | s->vdev.get_config = virtio_blk_update_config; |
417 | s->vdev.get_features = virtio_blk_get_features; | |
418 | s->vdev.reset = virtio_blk_reset; | |
419 | s->bs = bs; | |
869a5c6d | 420 | s->rq = NULL; |
bf011293 | 421 | if (strlen(ps = (char *)drive_get_serial(bs))) |
422 | strncpy(s->serial_str, ps, sizeof(s->serial_str)); | |
423 | else | |
424 | snprintf(s->serial_str, sizeof(s->serial_str), "0"); | |
53c25cea | 425 | bs->private = dev; |
6e02c38d AL |
426 | bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs); |
427 | bdrv_set_geometry_hint(s->bs, cylinders, heads, secs); | |
428 | ||
429 | s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output); | |
430 | ||
869a5c6d AL |
431 | qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s); |
432 | register_savevm("virtio-blk", virtio_blk_id++, 2, | |
6e02c38d AL |
433 | virtio_blk_save, virtio_blk_load, s); |
434 | ||
53c25cea | 435 | return &s->vdev; |
6e02c38d | 436 | } |