root / block-migration.c @ a55ce1c8
History | View | Annotate | Download (19.4 kB)
1 |
/*
|
---|---|
2 |
* QEMU live block migration
|
3 |
*
|
4 |
* Copyright IBM, Corp. 2009
|
5 |
*
|
6 |
* Authors:
|
7 |
* Liran Schour <lirans@il.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 |
* Contributions after 2012-01-13 are licensed under the terms of the
|
13 |
* GNU GPL, version 2 or (at your option) any later version.
|
14 |
*/
|
15 |
|
16 |
#include "qemu-common.h" |
17 |
#include "block/block_int.h" |
18 |
#include "hw/hw.h" |
19 |
#include "qemu/queue.h" |
20 |
#include "qemu/timer.h" |
21 |
#include "migration/block.h" |
22 |
#include "migration/migration.h" |
23 |
#include "sysemu/blockdev.h" |
24 |
#include <assert.h> |
25 |
|
26 |
#define BLOCK_SIZE (1 << 20) |
27 |
#define BDRV_SECTORS_PER_DIRTY_CHUNK (BLOCK_SIZE >> BDRV_SECTOR_BITS)
|
28 |
|
29 |
#define BLK_MIG_FLAG_DEVICE_BLOCK 0x01 |
30 |
#define BLK_MIG_FLAG_EOS 0x02 |
31 |
#define BLK_MIG_FLAG_PROGRESS 0x04 |
32 |
|
33 |
#define MAX_IS_ALLOCATED_SEARCH 65536 |
34 |
|
35 |
//#define DEBUG_BLK_MIGRATION
|
36 |
|
37 |
#ifdef DEBUG_BLK_MIGRATION
|
38 |
#define DPRINTF(fmt, ...) \
|
39 |
do { printf("blk_migration: " fmt, ## __VA_ARGS__); } while (0) |
40 |
#else
|
41 |
#define DPRINTF(fmt, ...) \
|
42 |
do { } while (0) |
43 |
#endif
|
44 |
|
45 |
typedef struct BlkMigDevState { |
46 |
BlockDriverState *bs; |
47 |
int bulk_completed;
|
48 |
int shared_base;
|
49 |
int64_t cur_sector; |
50 |
int64_t cur_dirty; |
51 |
int64_t completed_sectors; |
52 |
int64_t total_sectors; |
53 |
QSIMPLEQ_ENTRY(BlkMigDevState) entry; |
54 |
unsigned long *aio_bitmap; |
55 |
} BlkMigDevState; |
56 |
|
57 |
typedef struct BlkMigBlock { |
58 |
uint8_t *buf; |
59 |
BlkMigDevState *bmds; |
60 |
int64_t sector; |
61 |
int nr_sectors;
|
62 |
struct iovec iov;
|
63 |
QEMUIOVector qiov; |
64 |
BlockDriverAIOCB *aiocb; |
65 |
int ret;
|
66 |
QSIMPLEQ_ENTRY(BlkMigBlock) entry; |
67 |
} BlkMigBlock; |
68 |
|
69 |
typedef struct BlkMigState { |
70 |
int blk_enable;
|
71 |
int shared_base;
|
72 |
QSIMPLEQ_HEAD(bmds_list, BlkMigDevState) bmds_list; |
73 |
QSIMPLEQ_HEAD(blk_list, BlkMigBlock) blk_list; |
74 |
int submitted;
|
75 |
int read_done;
|
76 |
int transferred;
|
77 |
int64_t total_sector_sum; |
78 |
int prev_progress;
|
79 |
int bulk_completed;
|
80 |
} BlkMigState; |
81 |
|
82 |
static BlkMigState block_mig_state;
|
83 |
|
84 |
static void blk_send(QEMUFile *f, BlkMigBlock * blk) |
85 |
{ |
86 |
int len;
|
87 |
|
88 |
/* sector number and flags */
|
89 |
qemu_put_be64(f, (blk->sector << BDRV_SECTOR_BITS) |
90 |
| BLK_MIG_FLAG_DEVICE_BLOCK); |
91 |
|
92 |
/* device name */
|
93 |
len = strlen(blk->bmds->bs->device_name); |
94 |
qemu_put_byte(f, len); |
95 |
qemu_put_buffer(f, (uint8_t *)blk->bmds->bs->device_name, len); |
96 |
|
97 |
qemu_put_buffer(f, blk->buf, BLOCK_SIZE); |
98 |
} |
99 |
|
100 |
int blk_mig_active(void) |
101 |
{ |
102 |
return !QSIMPLEQ_EMPTY(&block_mig_state.bmds_list);
|
103 |
} |
104 |
|
105 |
uint64_t blk_mig_bytes_transferred(void)
|
106 |
{ |
107 |
BlkMigDevState *bmds; |
108 |
uint64_t sum = 0;
|
109 |
|
110 |
QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { |
111 |
sum += bmds->completed_sectors; |
112 |
} |
113 |
return sum << BDRV_SECTOR_BITS;
|
114 |
} |
115 |
|
116 |
uint64_t blk_mig_bytes_remaining(void)
|
117 |
{ |
118 |
return blk_mig_bytes_total() - blk_mig_bytes_transferred();
|
119 |
} |
120 |
|
121 |
uint64_t blk_mig_bytes_total(void)
|
122 |
{ |
123 |
BlkMigDevState *bmds; |
124 |
uint64_t sum = 0;
|
125 |
|
126 |
QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { |
127 |
sum += bmds->total_sectors; |
128 |
} |
129 |
return sum << BDRV_SECTOR_BITS;
|
130 |
} |
131 |
|
132 |
static int bmds_aio_inflight(BlkMigDevState *bmds, int64_t sector) |
133 |
{ |
134 |
int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK; |
135 |
|
136 |
if ((sector << BDRV_SECTOR_BITS) < bdrv_getlength(bmds->bs)) {
|
137 |
return !!(bmds->aio_bitmap[chunk / (sizeof(unsigned long) * 8)] & |
138 |
(1UL << (chunk % (sizeof(unsigned long) * 8)))); |
139 |
} else {
|
140 |
return 0; |
141 |
} |
142 |
} |
143 |
|
144 |
static void bmds_set_aio_inflight(BlkMigDevState *bmds, int64_t sector_num, |
145 |
int nb_sectors, int set) |
146 |
{ |
147 |
int64_t start, end; |
148 |
unsigned long val, idx, bit; |
149 |
|
150 |
start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK; |
151 |
end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
|
152 |
|
153 |
for (; start <= end; start++) {
|
154 |
idx = start / (sizeof(unsigned long) * 8); |
155 |
bit = start % (sizeof(unsigned long) * 8); |
156 |
val = bmds->aio_bitmap[idx]; |
157 |
if (set) {
|
158 |
val |= 1UL << bit;
|
159 |
} else {
|
160 |
val &= ~(1UL << bit);
|
161 |
} |
162 |
bmds->aio_bitmap[idx] = val; |
163 |
} |
164 |
} |
165 |
|
166 |
static void alloc_aio_bitmap(BlkMigDevState *bmds) |
167 |
{ |
168 |
BlockDriverState *bs = bmds->bs; |
169 |
int64_t bitmap_size; |
170 |
|
171 |
bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) + |
172 |
BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1; |
173 |
bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
|
174 |
|
175 |
bmds->aio_bitmap = g_malloc0(bitmap_size); |
176 |
} |
177 |
|
178 |
static void blk_mig_read_cb(void *opaque, int ret) |
179 |
{ |
180 |
BlkMigBlock *blk = opaque; |
181 |
|
182 |
blk->ret = ret; |
183 |
|
184 |
QSIMPLEQ_INSERT_TAIL(&block_mig_state.blk_list, blk, entry); |
185 |
bmds_set_aio_inflight(blk->bmds, blk->sector, blk->nr_sectors, 0);
|
186 |
|
187 |
block_mig_state.submitted--; |
188 |
block_mig_state.read_done++; |
189 |
assert(block_mig_state.submitted >= 0);
|
190 |
} |
191 |
|
192 |
static int mig_save_device_bulk(QEMUFile *f, BlkMigDevState *bmds) |
193 |
{ |
194 |
int64_t total_sectors = bmds->total_sectors; |
195 |
int64_t cur_sector = bmds->cur_sector; |
196 |
BlockDriverState *bs = bmds->bs; |
197 |
BlkMigBlock *blk; |
198 |
int nr_sectors;
|
199 |
|
200 |
if (bmds->shared_base) {
|
201 |
while (cur_sector < total_sectors &&
|
202 |
!bdrv_is_allocated(bs, cur_sector, MAX_IS_ALLOCATED_SEARCH, |
203 |
&nr_sectors)) { |
204 |
cur_sector += nr_sectors; |
205 |
} |
206 |
} |
207 |
|
208 |
if (cur_sector >= total_sectors) {
|
209 |
bmds->cur_sector = bmds->completed_sectors = total_sectors; |
210 |
return 1; |
211 |
} |
212 |
|
213 |
bmds->completed_sectors = cur_sector; |
214 |
|
215 |
cur_sector &= ~((int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK - 1);
|
216 |
|
217 |
/* we are going to transfer a full block even if it is not allocated */
|
218 |
nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK; |
219 |
|
220 |
if (total_sectors - cur_sector < BDRV_SECTORS_PER_DIRTY_CHUNK) {
|
221 |
nr_sectors = total_sectors - cur_sector; |
222 |
} |
223 |
|
224 |
blk = g_malloc(sizeof(BlkMigBlock));
|
225 |
blk->buf = g_malloc(BLOCK_SIZE); |
226 |
blk->bmds = bmds; |
227 |
blk->sector = cur_sector; |
228 |
blk->nr_sectors = nr_sectors; |
229 |
|
230 |
blk->iov.iov_base = blk->buf; |
231 |
blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE; |
232 |
qemu_iovec_init_external(&blk->qiov, &blk->iov, 1);
|
233 |
|
234 |
blk->aiocb = bdrv_aio_readv(bs, cur_sector, &blk->qiov, |
235 |
nr_sectors, blk_mig_read_cb, blk); |
236 |
block_mig_state.submitted++; |
237 |
|
238 |
bdrv_reset_dirty(bs, cur_sector, nr_sectors); |
239 |
bmds->cur_sector = cur_sector + nr_sectors; |
240 |
|
241 |
return (bmds->cur_sector >= total_sectors);
|
242 |
} |
243 |
|
244 |
static void set_dirty_tracking(int enable) |
245 |
{ |
246 |
BlkMigDevState *bmds; |
247 |
|
248 |
QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { |
249 |
bdrv_set_dirty_tracking(bmds->bs, enable ? BLOCK_SIZE : 0);
|
250 |
} |
251 |
} |
252 |
|
253 |
static void init_blk_migration_it(void *opaque, BlockDriverState *bs) |
254 |
{ |
255 |
BlkMigDevState *bmds; |
256 |
int64_t sectors; |
257 |
|
258 |
if (!bdrv_is_read_only(bs)) {
|
259 |
sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS; |
260 |
if (sectors <= 0) { |
261 |
return;
|
262 |
} |
263 |
|
264 |
bmds = g_malloc0(sizeof(BlkMigDevState));
|
265 |
bmds->bs = bs; |
266 |
bmds->bulk_completed = 0;
|
267 |
bmds->total_sectors = sectors; |
268 |
bmds->completed_sectors = 0;
|
269 |
bmds->shared_base = block_mig_state.shared_base; |
270 |
alloc_aio_bitmap(bmds); |
271 |
drive_get_ref(drive_get_by_blockdev(bs)); |
272 |
bdrv_set_in_use(bs, 1);
|
273 |
|
274 |
block_mig_state.total_sector_sum += sectors; |
275 |
|
276 |
if (bmds->shared_base) {
|
277 |
DPRINTF("Start migration for %s with shared base image\n",
|
278 |
bs->device_name); |
279 |
} else {
|
280 |
DPRINTF("Start full migration for %s\n", bs->device_name);
|
281 |
} |
282 |
|
283 |
QSIMPLEQ_INSERT_TAIL(&block_mig_state.bmds_list, bmds, entry); |
284 |
} |
285 |
} |
286 |
|
287 |
static void init_blk_migration(QEMUFile *f) |
288 |
{ |
289 |
block_mig_state.submitted = 0;
|
290 |
block_mig_state.read_done = 0;
|
291 |
block_mig_state.transferred = 0;
|
292 |
block_mig_state.total_sector_sum = 0;
|
293 |
block_mig_state.prev_progress = -1;
|
294 |
block_mig_state.bulk_completed = 0;
|
295 |
|
296 |
bdrv_iterate(init_blk_migration_it, NULL);
|
297 |
} |
298 |
|
299 |
static int blk_mig_save_bulked_block(QEMUFile *f) |
300 |
{ |
301 |
int64_t completed_sector_sum = 0;
|
302 |
BlkMigDevState *bmds; |
303 |
int progress;
|
304 |
int ret = 0; |
305 |
|
306 |
QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { |
307 |
if (bmds->bulk_completed == 0) { |
308 |
if (mig_save_device_bulk(f, bmds) == 1) { |
309 |
/* completed bulk section for this device */
|
310 |
bmds->bulk_completed = 1;
|
311 |
} |
312 |
completed_sector_sum += bmds->completed_sectors; |
313 |
ret = 1;
|
314 |
break;
|
315 |
} else {
|
316 |
completed_sector_sum += bmds->completed_sectors; |
317 |
} |
318 |
} |
319 |
|
320 |
if (block_mig_state.total_sector_sum != 0) { |
321 |
progress = completed_sector_sum * 100 /
|
322 |
block_mig_state.total_sector_sum; |
323 |
} else {
|
324 |
progress = 100;
|
325 |
} |
326 |
if (progress != block_mig_state.prev_progress) {
|
327 |
block_mig_state.prev_progress = progress; |
328 |
qemu_put_be64(f, (progress << BDRV_SECTOR_BITS) |
329 |
| BLK_MIG_FLAG_PROGRESS); |
330 |
DPRINTF("Completed %d %%\r", progress);
|
331 |
} |
332 |
|
333 |
return ret;
|
334 |
} |
335 |
|
336 |
static void blk_mig_reset_dirty_cursor(void) |
337 |
{ |
338 |
BlkMigDevState *bmds; |
339 |
|
340 |
QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { |
341 |
bmds->cur_dirty = 0;
|
342 |
} |
343 |
} |
344 |
|
345 |
static int mig_save_device_dirty(QEMUFile *f, BlkMigDevState *bmds, |
346 |
int is_async)
|
347 |
{ |
348 |
BlkMigBlock *blk; |
349 |
int64_t total_sectors = bmds->total_sectors; |
350 |
int64_t sector; |
351 |
int nr_sectors;
|
352 |
int ret = -EIO;
|
353 |
|
354 |
for (sector = bmds->cur_dirty; sector < bmds->total_sectors;) {
|
355 |
if (bmds_aio_inflight(bmds, sector)) {
|
356 |
bdrv_drain_all(); |
357 |
} |
358 |
if (bdrv_get_dirty(bmds->bs, sector)) {
|
359 |
|
360 |
if (total_sectors - sector < BDRV_SECTORS_PER_DIRTY_CHUNK) {
|
361 |
nr_sectors = total_sectors - sector; |
362 |
} else {
|
363 |
nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK; |
364 |
} |
365 |
blk = g_malloc(sizeof(BlkMigBlock));
|
366 |
blk->buf = g_malloc(BLOCK_SIZE); |
367 |
blk->bmds = bmds; |
368 |
blk->sector = sector; |
369 |
blk->nr_sectors = nr_sectors; |
370 |
|
371 |
if (is_async) {
|
372 |
blk->iov.iov_base = blk->buf; |
373 |
blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE; |
374 |
qemu_iovec_init_external(&blk->qiov, &blk->iov, 1);
|
375 |
|
376 |
blk->aiocb = bdrv_aio_readv(bmds->bs, sector, &blk->qiov, |
377 |
nr_sectors, blk_mig_read_cb, blk); |
378 |
block_mig_state.submitted++; |
379 |
bmds_set_aio_inflight(bmds, sector, nr_sectors, 1);
|
380 |
} else {
|
381 |
ret = bdrv_read(bmds->bs, sector, blk->buf, nr_sectors); |
382 |
if (ret < 0) { |
383 |
goto error;
|
384 |
} |
385 |
blk_send(f, blk); |
386 |
|
387 |
g_free(blk->buf); |
388 |
g_free(blk); |
389 |
} |
390 |
|
391 |
bdrv_reset_dirty(bmds->bs, sector, nr_sectors); |
392 |
break;
|
393 |
} |
394 |
sector += BDRV_SECTORS_PER_DIRTY_CHUNK; |
395 |
bmds->cur_dirty = sector; |
396 |
} |
397 |
|
398 |
return (bmds->cur_dirty >= bmds->total_sectors);
|
399 |
|
400 |
error:
|
401 |
DPRINTF("Error reading sector %" PRId64 "\n", sector); |
402 |
g_free(blk->buf); |
403 |
g_free(blk); |
404 |
return ret;
|
405 |
} |
406 |
|
407 |
/* return value:
|
408 |
* 0: too much data for max_downtime
|
409 |
* 1: few enough data for max_downtime
|
410 |
*/
|
411 |
static int blk_mig_save_dirty_block(QEMUFile *f, int is_async) |
412 |
{ |
413 |
BlkMigDevState *bmds; |
414 |
int ret = 1; |
415 |
|
416 |
QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { |
417 |
ret = mig_save_device_dirty(f, bmds, is_async); |
418 |
if (ret <= 0) { |
419 |
break;
|
420 |
} |
421 |
} |
422 |
|
423 |
return ret;
|
424 |
} |
425 |
|
426 |
static int flush_blks(QEMUFile *f) |
427 |
{ |
428 |
BlkMigBlock *blk; |
429 |
int ret = 0; |
430 |
|
431 |
DPRINTF("%s Enter submitted %d read_done %d transferred %d\n",
|
432 |
__FUNCTION__, block_mig_state.submitted, block_mig_state.read_done, |
433 |
block_mig_state.transferred); |
434 |
|
435 |
while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) { |
436 |
if (qemu_file_rate_limit(f)) {
|
437 |
break;
|
438 |
} |
439 |
if (blk->ret < 0) { |
440 |
ret = blk->ret; |
441 |
break;
|
442 |
} |
443 |
blk_send(f, blk); |
444 |
|
445 |
QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry); |
446 |
g_free(blk->buf); |
447 |
g_free(blk); |
448 |
|
449 |
block_mig_state.read_done--; |
450 |
block_mig_state.transferred++; |
451 |
assert(block_mig_state.read_done >= 0);
|
452 |
} |
453 |
|
454 |
DPRINTF("%s Exit submitted %d read_done %d transferred %d\n", __FUNCTION__,
|
455 |
block_mig_state.submitted, block_mig_state.read_done, |
456 |
block_mig_state.transferred); |
457 |
return ret;
|
458 |
} |
459 |
|
460 |
static int64_t get_remaining_dirty(void) |
461 |
{ |
462 |
BlkMigDevState *bmds; |
463 |
int64_t dirty = 0;
|
464 |
|
465 |
QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { |
466 |
dirty += bdrv_get_dirty_count(bmds->bs); |
467 |
} |
468 |
|
469 |
return dirty << BDRV_SECTOR_BITS;
|
470 |
} |
471 |
|
472 |
static void blk_mig_cleanup(void) |
473 |
{ |
474 |
BlkMigDevState *bmds; |
475 |
BlkMigBlock *blk; |
476 |
|
477 |
bdrv_drain_all(); |
478 |
|
479 |
set_dirty_tracking(0);
|
480 |
|
481 |
while ((bmds = QSIMPLEQ_FIRST(&block_mig_state.bmds_list)) != NULL) { |
482 |
QSIMPLEQ_REMOVE_HEAD(&block_mig_state.bmds_list, entry); |
483 |
bdrv_set_in_use(bmds->bs, 0);
|
484 |
drive_put_ref(drive_get_by_blockdev(bmds->bs)); |
485 |
g_free(bmds->aio_bitmap); |
486 |
g_free(bmds); |
487 |
} |
488 |
|
489 |
while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) { |
490 |
QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry); |
491 |
g_free(blk->buf); |
492 |
g_free(blk); |
493 |
} |
494 |
} |
495 |
|
496 |
static void block_migration_cancel(void *opaque) |
497 |
{ |
498 |
blk_mig_cleanup(); |
499 |
} |
500 |
|
501 |
static int block_save_setup(QEMUFile *f, void *opaque) |
502 |
{ |
503 |
int ret;
|
504 |
|
505 |
DPRINTF("Enter save live setup submitted %d transferred %d\n",
|
506 |
block_mig_state.submitted, block_mig_state.transferred); |
507 |
|
508 |
init_blk_migration(f); |
509 |
|
510 |
/* start track dirty blocks */
|
511 |
set_dirty_tracking(1);
|
512 |
|
513 |
ret = flush_blks(f); |
514 |
blk_mig_reset_dirty_cursor(); |
515 |
qemu_put_be64(f, BLK_MIG_FLAG_EOS); |
516 |
|
517 |
return ret;
|
518 |
} |
519 |
|
520 |
static int block_save_iterate(QEMUFile *f, void *opaque) |
521 |
{ |
522 |
int ret;
|
523 |
int64_t last_ftell = qemu_ftell(f); |
524 |
|
525 |
DPRINTF("Enter save live iterate submitted %d transferred %d\n",
|
526 |
block_mig_state.submitted, block_mig_state.transferred); |
527 |
|
528 |
ret = flush_blks(f); |
529 |
if (ret) {
|
530 |
return ret;
|
531 |
} |
532 |
|
533 |
blk_mig_reset_dirty_cursor(); |
534 |
|
535 |
/* control the rate of transfer */
|
536 |
while ((block_mig_state.submitted +
|
537 |
block_mig_state.read_done) * BLOCK_SIZE < |
538 |
qemu_file_get_rate_limit(f)) { |
539 |
if (block_mig_state.bulk_completed == 0) { |
540 |
/* first finish the bulk phase */
|
541 |
if (blk_mig_save_bulked_block(f) == 0) { |
542 |
/* finished saving bulk on all devices */
|
543 |
block_mig_state.bulk_completed = 1;
|
544 |
} |
545 |
} else {
|
546 |
ret = blk_mig_save_dirty_block(f, 1);
|
547 |
if (ret < 0) { |
548 |
return ret;
|
549 |
} |
550 |
if (ret != 0) { |
551 |
/* no more dirty blocks */
|
552 |
break;
|
553 |
} |
554 |
} |
555 |
} |
556 |
|
557 |
ret = flush_blks(f); |
558 |
if (ret) {
|
559 |
return ret;
|
560 |
} |
561 |
|
562 |
qemu_put_be64(f, BLK_MIG_FLAG_EOS); |
563 |
|
564 |
return qemu_ftell(f) - last_ftell;
|
565 |
} |
566 |
|
567 |
static int block_save_complete(QEMUFile *f, void *opaque) |
568 |
{ |
569 |
int ret;
|
570 |
|
571 |
DPRINTF("Enter save live complete submitted %d transferred %d\n",
|
572 |
block_mig_state.submitted, block_mig_state.transferred); |
573 |
|
574 |
ret = flush_blks(f); |
575 |
if (ret) {
|
576 |
return ret;
|
577 |
} |
578 |
|
579 |
blk_mig_reset_dirty_cursor(); |
580 |
|
581 |
/* we know for sure that save bulk is completed and
|
582 |
all async read completed */
|
583 |
assert(block_mig_state.submitted == 0);
|
584 |
|
585 |
do {
|
586 |
ret = blk_mig_save_dirty_block(f, 0);
|
587 |
if (ret < 0) { |
588 |
return ret;
|
589 |
} |
590 |
} while (ret == 0); |
591 |
|
592 |
/* report completion */
|
593 |
qemu_put_be64(f, (100 << BDRV_SECTOR_BITS) | BLK_MIG_FLAG_PROGRESS);
|
594 |
|
595 |
DPRINTF("Block migration completed\n");
|
596 |
|
597 |
qemu_put_be64(f, BLK_MIG_FLAG_EOS); |
598 |
|
599 |
blk_mig_cleanup(); |
600 |
return 0; |
601 |
} |
602 |
|
603 |
static uint64_t block_save_pending(QEMUFile *f, void *opaque, uint64_t max_size) |
604 |
{ |
605 |
/* Estimate pending number of bytes to send */
|
606 |
uint64_t pending = get_remaining_dirty() + |
607 |
block_mig_state.submitted * BLOCK_SIZE + |
608 |
block_mig_state.read_done * BLOCK_SIZE; |
609 |
|
610 |
/* Report at least one block pending during bulk phase */
|
611 |
if (pending == 0 && !block_mig_state.bulk_completed) { |
612 |
pending = BLOCK_SIZE; |
613 |
} |
614 |
|
615 |
DPRINTF("Enter save live pending %" PRIu64 "\n", pending); |
616 |
return pending;
|
617 |
} |
618 |
|
619 |
static int block_load(QEMUFile *f, void *opaque, int version_id) |
620 |
{ |
621 |
static int banner_printed; |
622 |
int len, flags;
|
623 |
char device_name[256]; |
624 |
int64_t addr; |
625 |
BlockDriverState *bs, *bs_prev = NULL;
|
626 |
uint8_t *buf; |
627 |
int64_t total_sectors = 0;
|
628 |
int nr_sectors;
|
629 |
int ret;
|
630 |
|
631 |
do {
|
632 |
addr = qemu_get_be64(f); |
633 |
|
634 |
flags = addr & ~BDRV_SECTOR_MASK; |
635 |
addr >>= BDRV_SECTOR_BITS; |
636 |
|
637 |
if (flags & BLK_MIG_FLAG_DEVICE_BLOCK) {
|
638 |
/* get device name */
|
639 |
len = qemu_get_byte(f); |
640 |
qemu_get_buffer(f, (uint8_t *)device_name, len); |
641 |
device_name[len] = '\0';
|
642 |
|
643 |
bs = bdrv_find(device_name); |
644 |
if (!bs) {
|
645 |
fprintf(stderr, "Error unknown block device %s\n",
|
646 |
device_name); |
647 |
return -EINVAL;
|
648 |
} |
649 |
|
650 |
if (bs != bs_prev) {
|
651 |
bs_prev = bs; |
652 |
total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS; |
653 |
if (total_sectors <= 0) { |
654 |
error_report("Error getting length of block device %s",
|
655 |
device_name); |
656 |
return -EINVAL;
|
657 |
} |
658 |
} |
659 |
|
660 |
if (total_sectors - addr < BDRV_SECTORS_PER_DIRTY_CHUNK) {
|
661 |
nr_sectors = total_sectors - addr; |
662 |
} else {
|
663 |
nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK; |
664 |
} |
665 |
|
666 |
buf = g_malloc(BLOCK_SIZE); |
667 |
|
668 |
qemu_get_buffer(f, buf, BLOCK_SIZE); |
669 |
ret = bdrv_write(bs, addr, buf, nr_sectors); |
670 |
|
671 |
g_free(buf); |
672 |
if (ret < 0) { |
673 |
return ret;
|
674 |
} |
675 |
} else if (flags & BLK_MIG_FLAG_PROGRESS) { |
676 |
if (!banner_printed) {
|
677 |
printf("Receiving block device images\n");
|
678 |
banner_printed = 1;
|
679 |
} |
680 |
printf("Completed %d %%%c", (int)addr, |
681 |
(addr == 100) ? '\n' : '\r'); |
682 |
fflush(stdout); |
683 |
} else if (!(flags & BLK_MIG_FLAG_EOS)) { |
684 |
fprintf(stderr, "Unknown block migration flags: %#x\n", flags);
|
685 |
return -EINVAL;
|
686 |
} |
687 |
ret = qemu_file_get_error(f); |
688 |
if (ret != 0) { |
689 |
return ret;
|
690 |
} |
691 |
} while (!(flags & BLK_MIG_FLAG_EOS));
|
692 |
|
693 |
return 0; |
694 |
} |
695 |
|
696 |
static void block_set_params(const MigrationParams *params, void *opaque) |
697 |
{ |
698 |
block_mig_state.blk_enable = params->blk; |
699 |
block_mig_state.shared_base = params->shared; |
700 |
|
701 |
/* shared base means that blk_enable = 1 */
|
702 |
block_mig_state.blk_enable |= params->shared; |
703 |
} |
704 |
|
705 |
static bool block_is_active(void *opaque) |
706 |
{ |
707 |
return block_mig_state.blk_enable == 1; |
708 |
} |
709 |
|
710 |
SaveVMHandlers savevm_block_handlers = { |
711 |
.set_params = block_set_params, |
712 |
.save_live_setup = block_save_setup, |
713 |
.save_live_iterate = block_save_iterate, |
714 |
.save_live_complete = block_save_complete, |
715 |
.save_live_pending = block_save_pending, |
716 |
.load_state = block_load, |
717 |
.cancel = block_migration_cancel, |
718 |
.is_active = block_is_active, |
719 |
}; |
720 |
|
721 |
void blk_mig_init(void) |
722 |
{ |
723 |
QSIMPLEQ_INIT(&block_mig_state.bmds_list); |
724 |
QSIMPLEQ_INIT(&block_mig_state.blk_list); |
725 |
|
726 |
register_savevm_live(NULL, "block", 0, 1, &savevm_block_handlers, |
727 |
&block_mig_state); |
728 |
} |