]>
Commit | Line | Data |
---|---|---|
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_int.h" | |
18 | #include "hw/hw.h" | |
19 | #include "qemu-queue.h" | |
20 | #include "qemu-timer.h" | |
21 | #include "block-migration.h" | |
22 | #include "migration.h" | |
23 | #include "blockdev.h" | |
24 | #include <assert.h> | |
25 | ||
26 | #define BLOCK_SIZE (BDRV_SECTORS_PER_DIRTY_CHUNK << BDRV_SECTOR_BITS) | |
27 | ||
28 | #define BLK_MIG_FLAG_DEVICE_BLOCK 0x01 | |
29 | #define BLK_MIG_FLAG_EOS 0x02 | |
30 | #define BLK_MIG_FLAG_PROGRESS 0x04 | |
31 | ||
32 | #define MAX_IS_ALLOCATED_SEARCH 65536 | |
33 | ||
34 | //#define DEBUG_BLK_MIGRATION | |
35 | ||
36 | #ifdef DEBUG_BLK_MIGRATION | |
37 | #define DPRINTF(fmt, ...) \ | |
38 | do { printf("blk_migration: " fmt, ## __VA_ARGS__); } while (0) | |
39 | #else | |
40 | #define DPRINTF(fmt, ...) \ | |
41 | do { } while (0) | |
42 | #endif | |
43 | ||
44 | typedef struct BlkMigDevState { | |
45 | BlockDriverState *bs; | |
46 | int bulk_completed; | |
47 | int shared_base; | |
48 | int64_t cur_sector; | |
49 | int64_t cur_dirty; | |
50 | int64_t completed_sectors; | |
51 | int64_t total_sectors; | |
52 | int64_t dirty; | |
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 | long double total_time; | |
81 | long double prev_time_offset; | |
82 | int reads; | |
83 | } BlkMigState; | |
84 | ||
85 | static BlkMigState block_mig_state; | |
86 | ||
87 | static void blk_send(QEMUFile *f, BlkMigBlock * blk) | |
88 | { | |
89 | int len; | |
90 | ||
91 | /* sector number and flags */ | |
92 | qemu_put_be64(f, (blk->sector << BDRV_SECTOR_BITS) | |
93 | | BLK_MIG_FLAG_DEVICE_BLOCK); | |
94 | ||
95 | /* device name */ | |
96 | len = strlen(blk->bmds->bs->device_name); | |
97 | qemu_put_byte(f, len); | |
98 | qemu_put_buffer(f, (uint8_t *)blk->bmds->bs->device_name, len); | |
99 | ||
100 | qemu_put_buffer(f, blk->buf, BLOCK_SIZE); | |
101 | } | |
102 | ||
103 | int blk_mig_active(void) | |
104 | { | |
105 | return !QSIMPLEQ_EMPTY(&block_mig_state.bmds_list); | |
106 | } | |
107 | ||
108 | uint64_t blk_mig_bytes_transferred(void) | |
109 | { | |
110 | BlkMigDevState *bmds; | |
111 | uint64_t sum = 0; | |
112 | ||
113 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
114 | sum += bmds->completed_sectors; | |
115 | } | |
116 | return sum << BDRV_SECTOR_BITS; | |
117 | } | |
118 | ||
119 | uint64_t blk_mig_bytes_remaining(void) | |
120 | { | |
121 | return blk_mig_bytes_total() - blk_mig_bytes_transferred(); | |
122 | } | |
123 | ||
124 | uint64_t blk_mig_bytes_total(void) | |
125 | { | |
126 | BlkMigDevState *bmds; | |
127 | uint64_t sum = 0; | |
128 | ||
129 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
130 | sum += bmds->total_sectors; | |
131 | } | |
132 | return sum << BDRV_SECTOR_BITS; | |
133 | } | |
134 | ||
135 | static inline long double compute_read_bwidth(void) | |
136 | { | |
137 | assert(block_mig_state.total_time != 0); | |
138 | return (block_mig_state.reads / block_mig_state.total_time) * BLOCK_SIZE; | |
139 | } | |
140 | ||
141 | static int bmds_aio_inflight(BlkMigDevState *bmds, int64_t sector) | |
142 | { | |
143 | int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK; | |
144 | ||
145 | if ((sector << BDRV_SECTOR_BITS) < bdrv_getlength(bmds->bs)) { | |
146 | return !!(bmds->aio_bitmap[chunk / (sizeof(unsigned long) * 8)] & | |
147 | (1UL << (chunk % (sizeof(unsigned long) * 8)))); | |
148 | } else { | |
149 | return 0; | |
150 | } | |
151 | } | |
152 | ||
153 | static void bmds_set_aio_inflight(BlkMigDevState *bmds, int64_t sector_num, | |
154 | int nb_sectors, int set) | |
155 | { | |
156 | int64_t start, end; | |
157 | unsigned long val, idx, bit; | |
158 | ||
159 | start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK; | |
160 | end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK; | |
161 | ||
162 | for (; start <= end; start++) { | |
163 | idx = start / (sizeof(unsigned long) * 8); | |
164 | bit = start % (sizeof(unsigned long) * 8); | |
165 | val = bmds->aio_bitmap[idx]; | |
166 | if (set) { | |
167 | val |= 1UL << bit; | |
168 | } else { | |
169 | val &= ~(1UL << bit); | |
170 | } | |
171 | bmds->aio_bitmap[idx] = val; | |
172 | } | |
173 | } | |
174 | ||
175 | static void alloc_aio_bitmap(BlkMigDevState *bmds) | |
176 | { | |
177 | BlockDriverState *bs = bmds->bs; | |
178 | int64_t bitmap_size; | |
179 | ||
180 | bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) + | |
181 | BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1; | |
182 | bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8; | |
183 | ||
184 | bmds->aio_bitmap = g_malloc0(bitmap_size); | |
185 | } | |
186 | ||
187 | static void blk_mig_read_cb(void *opaque, int ret) | |
188 | { | |
189 | long double curr_time = qemu_get_clock_ns(rt_clock); | |
190 | BlkMigBlock *blk = opaque; | |
191 | ||
192 | blk->ret = ret; | |
193 | ||
194 | block_mig_state.reads++; | |
195 | block_mig_state.total_time += (curr_time - block_mig_state.prev_time_offset); | |
196 | block_mig_state.prev_time_offset = curr_time; | |
197 | ||
198 | QSIMPLEQ_INSERT_TAIL(&block_mig_state.blk_list, blk, entry); | |
199 | bmds_set_aio_inflight(blk->bmds, blk->sector, blk->nr_sectors, 0); | |
200 | ||
201 | block_mig_state.submitted--; | |
202 | block_mig_state.read_done++; | |
203 | assert(block_mig_state.submitted >= 0); | |
204 | } | |
205 | ||
206 | static int mig_save_device_bulk(QEMUFile *f, BlkMigDevState *bmds) | |
207 | { | |
208 | int64_t total_sectors = bmds->total_sectors; | |
209 | int64_t cur_sector = bmds->cur_sector; | |
210 | BlockDriverState *bs = bmds->bs; | |
211 | BlkMigBlock *blk; | |
212 | int nr_sectors; | |
213 | ||
214 | if (bmds->shared_base) { | |
215 | while (cur_sector < total_sectors && | |
216 | !bdrv_is_allocated(bs, cur_sector, MAX_IS_ALLOCATED_SEARCH, | |
217 | &nr_sectors)) { | |
218 | cur_sector += nr_sectors; | |
219 | } | |
220 | } | |
221 | ||
222 | if (cur_sector >= total_sectors) { | |
223 | bmds->cur_sector = bmds->completed_sectors = total_sectors; | |
224 | return 1; | |
225 | } | |
226 | ||
227 | bmds->completed_sectors = cur_sector; | |
228 | ||
229 | cur_sector &= ~((int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK - 1); | |
230 | ||
231 | /* we are going to transfer a full block even if it is not allocated */ | |
232 | nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK; | |
233 | ||
234 | if (total_sectors - cur_sector < BDRV_SECTORS_PER_DIRTY_CHUNK) { | |
235 | nr_sectors = total_sectors - cur_sector; | |
236 | } | |
237 | ||
238 | blk = g_malloc(sizeof(BlkMigBlock)); | |
239 | blk->buf = g_malloc(BLOCK_SIZE); | |
240 | blk->bmds = bmds; | |
241 | blk->sector = cur_sector; | |
242 | blk->nr_sectors = nr_sectors; | |
243 | ||
244 | blk->iov.iov_base = blk->buf; | |
245 | blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE; | |
246 | qemu_iovec_init_external(&blk->qiov, &blk->iov, 1); | |
247 | ||
248 | if (block_mig_state.submitted == 0) { | |
249 | block_mig_state.prev_time_offset = qemu_get_clock_ns(rt_clock); | |
250 | } | |
251 | ||
252 | blk->aiocb = bdrv_aio_readv(bs, cur_sector, &blk->qiov, | |
253 | nr_sectors, blk_mig_read_cb, blk); | |
254 | block_mig_state.submitted++; | |
255 | ||
256 | bdrv_reset_dirty(bs, cur_sector, nr_sectors); | |
257 | bmds->cur_sector = cur_sector + nr_sectors; | |
258 | ||
259 | return (bmds->cur_sector >= total_sectors); | |
260 | } | |
261 | ||
262 | static void set_dirty_tracking(int enable) | |
263 | { | |
264 | BlkMigDevState *bmds; | |
265 | ||
266 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
267 | bdrv_set_dirty_tracking(bmds->bs, enable); | |
268 | } | |
269 | } | |
270 | ||
271 | static void init_blk_migration_it(void *opaque, BlockDriverState *bs) | |
272 | { | |
273 | BlkMigDevState *bmds; | |
274 | int64_t sectors; | |
275 | ||
276 | if (!bdrv_is_read_only(bs)) { | |
277 | sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS; | |
278 | if (sectors <= 0) { | |
279 | return; | |
280 | } | |
281 | ||
282 | bmds = g_malloc0(sizeof(BlkMigDevState)); | |
283 | bmds->bs = bs; | |
284 | bmds->bulk_completed = 0; | |
285 | bmds->total_sectors = sectors; | |
286 | bmds->completed_sectors = 0; | |
287 | bmds->shared_base = block_mig_state.shared_base; | |
288 | alloc_aio_bitmap(bmds); | |
289 | drive_get_ref(drive_get_by_blockdev(bs)); | |
290 | bdrv_set_in_use(bs, 1); | |
291 | ||
292 | block_mig_state.total_sector_sum += sectors; | |
293 | ||
294 | if (bmds->shared_base) { | |
295 | DPRINTF("Start migration for %s with shared base image\n", | |
296 | bs->device_name); | |
297 | } else { | |
298 | DPRINTF("Start full migration for %s\n", bs->device_name); | |
299 | } | |
300 | ||
301 | QSIMPLEQ_INSERT_TAIL(&block_mig_state.bmds_list, bmds, entry); | |
302 | } | |
303 | } | |
304 | ||
305 | static void init_blk_migration(QEMUFile *f) | |
306 | { | |
307 | block_mig_state.submitted = 0; | |
308 | block_mig_state.read_done = 0; | |
309 | block_mig_state.transferred = 0; | |
310 | block_mig_state.total_sector_sum = 0; | |
311 | block_mig_state.prev_progress = -1; | |
312 | block_mig_state.bulk_completed = 0; | |
313 | block_mig_state.total_time = 0; | |
314 | block_mig_state.reads = 0; | |
315 | ||
316 | bdrv_iterate(init_blk_migration_it, NULL); | |
317 | } | |
318 | ||
319 | static int blk_mig_save_bulked_block(QEMUFile *f) | |
320 | { | |
321 | int64_t completed_sector_sum = 0; | |
322 | BlkMigDevState *bmds; | |
323 | int progress; | |
324 | int ret = 0; | |
325 | ||
326 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
327 | if (bmds->bulk_completed == 0) { | |
328 | if (mig_save_device_bulk(f, bmds) == 1) { | |
329 | /* completed bulk section for this device */ | |
330 | bmds->bulk_completed = 1; | |
331 | } | |
332 | completed_sector_sum += bmds->completed_sectors; | |
333 | ret = 1; | |
334 | break; | |
335 | } else { | |
336 | completed_sector_sum += bmds->completed_sectors; | |
337 | } | |
338 | } | |
339 | ||
340 | if (block_mig_state.total_sector_sum != 0) { | |
341 | progress = completed_sector_sum * 100 / | |
342 | block_mig_state.total_sector_sum; | |
343 | } else { | |
344 | progress = 100; | |
345 | } | |
346 | if (progress != block_mig_state.prev_progress) { | |
347 | block_mig_state.prev_progress = progress; | |
348 | qemu_put_be64(f, (progress << BDRV_SECTOR_BITS) | |
349 | | BLK_MIG_FLAG_PROGRESS); | |
350 | DPRINTF("Completed %d %%\r", progress); | |
351 | } | |
352 | ||
353 | return ret; | |
354 | } | |
355 | ||
356 | static void blk_mig_reset_dirty_cursor(void) | |
357 | { | |
358 | BlkMigDevState *bmds; | |
359 | ||
360 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
361 | bmds->cur_dirty = 0; | |
362 | } | |
363 | } | |
364 | ||
365 | static int mig_save_device_dirty(QEMUFile *f, BlkMigDevState *bmds, | |
366 | int is_async) | |
367 | { | |
368 | BlkMigBlock *blk; | |
369 | int64_t total_sectors = bmds->total_sectors; | |
370 | int64_t sector; | |
371 | int nr_sectors; | |
372 | int ret = -EIO; | |
373 | ||
374 | for (sector = bmds->cur_dirty; sector < bmds->total_sectors;) { | |
375 | if (bmds_aio_inflight(bmds, sector)) { | |
376 | bdrv_drain_all(); | |
377 | } | |
378 | if (bdrv_get_dirty(bmds->bs, sector)) { | |
379 | ||
380 | if (total_sectors - sector < BDRV_SECTORS_PER_DIRTY_CHUNK) { | |
381 | nr_sectors = total_sectors - sector; | |
382 | } else { | |
383 | nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK; | |
384 | } | |
385 | blk = g_malloc(sizeof(BlkMigBlock)); | |
386 | blk->buf = g_malloc(BLOCK_SIZE); | |
387 | blk->bmds = bmds; | |
388 | blk->sector = sector; | |
389 | blk->nr_sectors = nr_sectors; | |
390 | ||
391 | if (is_async) { | |
392 | blk->iov.iov_base = blk->buf; | |
393 | blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE; | |
394 | qemu_iovec_init_external(&blk->qiov, &blk->iov, 1); | |
395 | ||
396 | if (block_mig_state.submitted == 0) { | |
397 | block_mig_state.prev_time_offset = qemu_get_clock_ns(rt_clock); | |
398 | } | |
399 | ||
400 | blk->aiocb = bdrv_aio_readv(bmds->bs, sector, &blk->qiov, | |
401 | nr_sectors, blk_mig_read_cb, blk); | |
402 | block_mig_state.submitted++; | |
403 | bmds_set_aio_inflight(bmds, sector, nr_sectors, 1); | |
404 | } else { | |
405 | ret = bdrv_read(bmds->bs, sector, blk->buf, nr_sectors); | |
406 | if (ret < 0) { | |
407 | goto error; | |
408 | } | |
409 | blk_send(f, blk); | |
410 | ||
411 | g_free(blk->buf); | |
412 | g_free(blk); | |
413 | } | |
414 | ||
415 | bdrv_reset_dirty(bmds->bs, sector, nr_sectors); | |
416 | break; | |
417 | } | |
418 | sector += BDRV_SECTORS_PER_DIRTY_CHUNK; | |
419 | bmds->cur_dirty = sector; | |
420 | } | |
421 | ||
422 | return (bmds->cur_dirty >= bmds->total_sectors); | |
423 | ||
424 | error: | |
425 | DPRINTF("Error reading sector %" PRId64 "\n", sector); | |
426 | g_free(blk->buf); | |
427 | g_free(blk); | |
428 | return ret; | |
429 | } | |
430 | ||
431 | /* return value: | |
432 | * 0: too much data for max_downtime | |
433 | * 1: few enough data for max_downtime | |
434 | */ | |
435 | static int blk_mig_save_dirty_block(QEMUFile *f, int is_async) | |
436 | { | |
437 | BlkMigDevState *bmds; | |
438 | int ret = 1; | |
439 | ||
440 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
441 | ret = mig_save_device_dirty(f, bmds, is_async); | |
442 | if (ret <= 0) { | |
443 | break; | |
444 | } | |
445 | } | |
446 | ||
447 | return ret; | |
448 | } | |
449 | ||
450 | static int flush_blks(QEMUFile *f) | |
451 | { | |
452 | BlkMigBlock *blk; | |
453 | int ret = 0; | |
454 | ||
455 | DPRINTF("%s Enter submitted %d read_done %d transferred %d\n", | |
456 | __FUNCTION__, block_mig_state.submitted, block_mig_state.read_done, | |
457 | block_mig_state.transferred); | |
458 | ||
459 | while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) { | |
460 | if (qemu_file_rate_limit(f)) { | |
461 | break; | |
462 | } | |
463 | if (blk->ret < 0) { | |
464 | ret = blk->ret; | |
465 | break; | |
466 | } | |
467 | blk_send(f, blk); | |
468 | ||
469 | QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry); | |
470 | g_free(blk->buf); | |
471 | g_free(blk); | |
472 | ||
473 | block_mig_state.read_done--; | |
474 | block_mig_state.transferred++; | |
475 | assert(block_mig_state.read_done >= 0); | |
476 | } | |
477 | ||
478 | DPRINTF("%s Exit submitted %d read_done %d transferred %d\n", __FUNCTION__, | |
479 | block_mig_state.submitted, block_mig_state.read_done, | |
480 | block_mig_state.transferred); | |
481 | return ret; | |
482 | } | |
483 | ||
484 | static int64_t get_remaining_dirty(void) | |
485 | { | |
486 | BlkMigDevState *bmds; | |
487 | int64_t dirty = 0; | |
488 | ||
489 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
490 | dirty += bdrv_get_dirty_count(bmds->bs); | |
491 | } | |
492 | ||
493 | return dirty * BLOCK_SIZE; | |
494 | } | |
495 | ||
496 | static int is_stage2_completed(void) | |
497 | { | |
498 | int64_t remaining_dirty; | |
499 | long double bwidth; | |
500 | ||
501 | if (block_mig_state.bulk_completed == 1) { | |
502 | ||
503 | remaining_dirty = get_remaining_dirty(); | |
504 | if (remaining_dirty == 0) { | |
505 | return 1; | |
506 | } | |
507 | ||
508 | bwidth = compute_read_bwidth(); | |
509 | ||
510 | if ((remaining_dirty / bwidth) <= | |
511 | migrate_max_downtime()) { | |
512 | /* finish stage2 because we think that we can finish remaining work | |
513 | below max_downtime */ | |
514 | ||
515 | return 1; | |
516 | } | |
517 | } | |
518 | ||
519 | return 0; | |
520 | } | |
521 | ||
522 | static void blk_mig_cleanup(void) | |
523 | { | |
524 | BlkMigDevState *bmds; | |
525 | BlkMigBlock *blk; | |
526 | ||
527 | bdrv_drain_all(); | |
528 | ||
529 | set_dirty_tracking(0); | |
530 | ||
531 | while ((bmds = QSIMPLEQ_FIRST(&block_mig_state.bmds_list)) != NULL) { | |
532 | QSIMPLEQ_REMOVE_HEAD(&block_mig_state.bmds_list, entry); | |
533 | bdrv_set_in_use(bmds->bs, 0); | |
534 | drive_put_ref(drive_get_by_blockdev(bmds->bs)); | |
535 | g_free(bmds->aio_bitmap); | |
536 | g_free(bmds); | |
537 | } | |
538 | ||
539 | while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) { | |
540 | QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry); | |
541 | g_free(blk->buf); | |
542 | g_free(blk); | |
543 | } | |
544 | } | |
545 | ||
546 | static void block_migration_cancel(void *opaque) | |
547 | { | |
548 | blk_mig_cleanup(); | |
549 | } | |
550 | ||
551 | static int block_save_setup(QEMUFile *f, void *opaque) | |
552 | { | |
553 | int ret; | |
554 | ||
555 | DPRINTF("Enter save live setup submitted %d transferred %d\n", | |
556 | block_mig_state.submitted, block_mig_state.transferred); | |
557 | ||
558 | init_blk_migration(f); | |
559 | ||
560 | /* start track dirty blocks */ | |
561 | set_dirty_tracking(1); | |
562 | ||
563 | ret = flush_blks(f); | |
564 | if (ret) { | |
565 | blk_mig_cleanup(); | |
566 | return ret; | |
567 | } | |
568 | ||
569 | blk_mig_reset_dirty_cursor(); | |
570 | ||
571 | qemu_put_be64(f, BLK_MIG_FLAG_EOS); | |
572 | ||
573 | return 0; | |
574 | } | |
575 | ||
576 | static int block_save_iterate(QEMUFile *f, void *opaque) | |
577 | { | |
578 | int ret; | |
579 | ||
580 | DPRINTF("Enter save live iterate submitted %d transferred %d\n", | |
581 | block_mig_state.submitted, block_mig_state.transferred); | |
582 | ||
583 | ret = flush_blks(f); | |
584 | if (ret) { | |
585 | blk_mig_cleanup(); | |
586 | return ret; | |
587 | } | |
588 | ||
589 | blk_mig_reset_dirty_cursor(); | |
590 | ||
591 | /* control the rate of transfer */ | |
592 | while ((block_mig_state.submitted + | |
593 | block_mig_state.read_done) * BLOCK_SIZE < | |
594 | qemu_file_get_rate_limit(f)) { | |
595 | if (block_mig_state.bulk_completed == 0) { | |
596 | /* first finish the bulk phase */ | |
597 | if (blk_mig_save_bulked_block(f) == 0) { | |
598 | /* finished saving bulk on all devices */ | |
599 | block_mig_state.bulk_completed = 1; | |
600 | } | |
601 | } else { | |
602 | ret = blk_mig_save_dirty_block(f, 1); | |
603 | if (ret != 0) { | |
604 | /* no more dirty blocks */ | |
605 | break; | |
606 | } | |
607 | } | |
608 | } | |
609 | if (ret) { | |
610 | blk_mig_cleanup(); | |
611 | return ret; | |
612 | } | |
613 | ||
614 | ret = flush_blks(f); | |
615 | if (ret) { | |
616 | blk_mig_cleanup(); | |
617 | return ret; | |
618 | } | |
619 | ||
620 | qemu_put_be64(f, BLK_MIG_FLAG_EOS); | |
621 | ||
622 | return is_stage2_completed(); | |
623 | } | |
624 | ||
625 | static int block_save_complete(QEMUFile *f, void *opaque) | |
626 | { | |
627 | int ret; | |
628 | ||
629 | DPRINTF("Enter save live complete submitted %d transferred %d\n", | |
630 | block_mig_state.submitted, block_mig_state.transferred); | |
631 | ||
632 | ret = flush_blks(f); | |
633 | if (ret) { | |
634 | blk_mig_cleanup(); | |
635 | return ret; | |
636 | } | |
637 | ||
638 | blk_mig_reset_dirty_cursor(); | |
639 | ||
640 | /* we know for sure that save bulk is completed and | |
641 | all async read completed */ | |
642 | assert(block_mig_state.submitted == 0); | |
643 | ||
644 | do { | |
645 | ret = blk_mig_save_dirty_block(f, 0); | |
646 | } while (ret == 0); | |
647 | ||
648 | blk_mig_cleanup(); | |
649 | if (ret) { | |
650 | return ret; | |
651 | } | |
652 | /* report completion */ | |
653 | qemu_put_be64(f, (100 << BDRV_SECTOR_BITS) | BLK_MIG_FLAG_PROGRESS); | |
654 | ||
655 | DPRINTF("Block migration completed\n"); | |
656 | ||
657 | qemu_put_be64(f, BLK_MIG_FLAG_EOS); | |
658 | ||
659 | return 0; | |
660 | } | |
661 | ||
662 | static int block_load(QEMUFile *f, void *opaque, int version_id) | |
663 | { | |
664 | static int banner_printed; | |
665 | int len, flags; | |
666 | char device_name[256]; | |
667 | int64_t addr; | |
668 | BlockDriverState *bs, *bs_prev = NULL; | |
669 | uint8_t *buf; | |
670 | int64_t total_sectors = 0; | |
671 | int nr_sectors; | |
672 | int ret; | |
673 | ||
674 | do { | |
675 | addr = qemu_get_be64(f); | |
676 | ||
677 | flags = addr & ~BDRV_SECTOR_MASK; | |
678 | addr >>= BDRV_SECTOR_BITS; | |
679 | ||
680 | if (flags & BLK_MIG_FLAG_DEVICE_BLOCK) { | |
681 | /* get device name */ | |
682 | len = qemu_get_byte(f); | |
683 | qemu_get_buffer(f, (uint8_t *)device_name, len); | |
684 | device_name[len] = '\0'; | |
685 | ||
686 | bs = bdrv_find(device_name); | |
687 | if (!bs) { | |
688 | fprintf(stderr, "Error unknown block device %s\n", | |
689 | device_name); | |
690 | return -EINVAL; | |
691 | } | |
692 | ||
693 | if (bs != bs_prev) { | |
694 | bs_prev = bs; | |
695 | total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS; | |
696 | if (total_sectors <= 0) { | |
697 | error_report("Error getting length of block device %s", | |
698 | device_name); | |
699 | return -EINVAL; | |
700 | } | |
701 | } | |
702 | ||
703 | if (total_sectors - addr < BDRV_SECTORS_PER_DIRTY_CHUNK) { | |
704 | nr_sectors = total_sectors - addr; | |
705 | } else { | |
706 | nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK; | |
707 | } | |
708 | ||
709 | buf = g_malloc(BLOCK_SIZE); | |
710 | ||
711 | qemu_get_buffer(f, buf, BLOCK_SIZE); | |
712 | ret = bdrv_write(bs, addr, buf, nr_sectors); | |
713 | ||
714 | g_free(buf); | |
715 | if (ret < 0) { | |
716 | return ret; | |
717 | } | |
718 | } else if (flags & BLK_MIG_FLAG_PROGRESS) { | |
719 | if (!banner_printed) { | |
720 | printf("Receiving block device images\n"); | |
721 | banner_printed = 1; | |
722 | } | |
723 | printf("Completed %d %%%c", (int)addr, | |
724 | (addr == 100) ? '\n' : '\r'); | |
725 | fflush(stdout); | |
726 | } else if (!(flags & BLK_MIG_FLAG_EOS)) { | |
727 | fprintf(stderr, "Unknown flags\n"); | |
728 | return -EINVAL; | |
729 | } | |
730 | ret = qemu_file_get_error(f); | |
731 | if (ret != 0) { | |
732 | return ret; | |
733 | } | |
734 | } while (!(flags & BLK_MIG_FLAG_EOS)); | |
735 | ||
736 | return 0; | |
737 | } | |
738 | ||
739 | static void block_set_params(const MigrationParams *params, void *opaque) | |
740 | { | |
741 | block_mig_state.blk_enable = params->blk; | |
742 | block_mig_state.shared_base = params->shared; | |
743 | ||
744 | /* shared base means that blk_enable = 1 */ | |
745 | block_mig_state.blk_enable |= params->shared; | |
746 | } | |
747 | ||
748 | static bool block_is_active(void *opaque) | |
749 | { | |
750 | return block_mig_state.blk_enable == 1; | |
751 | } | |
752 | ||
753 | SaveVMHandlers savevm_block_handlers = { | |
754 | .set_params = block_set_params, | |
755 | .save_live_setup = block_save_setup, | |
756 | .save_live_iterate = block_save_iterate, | |
757 | .save_live_complete = block_save_complete, | |
758 | .load_state = block_load, | |
759 | .cancel = block_migration_cancel, | |
760 | .is_active = block_is_active, | |
761 | }; | |
762 | ||
763 | void blk_mig_init(void) | |
764 | { | |
765 | QSIMPLEQ_INIT(&block_mig_state.bmds_list); | |
766 | QSIMPLEQ_INIT(&block_mig_state.blk_list); | |
767 | ||
768 | register_savevm_live(NULL, "block", 0, 1, &savevm_block_handlers, | |
769 | &block_mig_state); | |
770 | } |