]>
Commit | Line | Data |
---|---|---|
c163b5ca LS |
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 | * | |
6b620ca3 PB |
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. | |
c163b5ca LS |
14 | */ |
15 | ||
1393a485 | 16 | #include "qemu/osdep.h" |
da34e65c | 17 | #include "qapi/error.h" |
bfb197e0 | 18 | #include "qemu/error-report.h" |
f348b6d1 | 19 | #include "qemu/cutils.h" |
1de7afc9 | 20 | #include "qemu/queue.h" |
2c9e6fec JQ |
21 | #include "block.h" |
22 | #include "migration/misc.h" | |
6666c96a | 23 | #include "migration.h" |
f2a8f0a6 | 24 | #include "migration/register.h" |
08a0aee1 | 25 | #include "qemu-file.h" |
987772d9 | 26 | #include "migration/vmstate.h" |
c9ebaf74 | 27 | #include "sysemu/block-backend.h" |
c163b5ca | 28 | |
50717e94 PB |
29 | #define BLOCK_SIZE (1 << 20) |
30 | #define BDRV_SECTORS_PER_DIRTY_CHUNK (BLOCK_SIZE >> BDRV_SECTOR_BITS) | |
c163b5ca LS |
31 | |
32 | #define BLK_MIG_FLAG_DEVICE_BLOCK 0x01 | |
33 | #define BLK_MIG_FLAG_EOS 0x02 | |
01e61e2d | 34 | #define BLK_MIG_FLAG_PROGRESS 0x04 |
323004a3 | 35 | #define BLK_MIG_FLAG_ZERO_BLOCK 0x08 |
c163b5ca | 36 | |
d6a644bb | 37 | #define MAX_IS_ALLOCATED_SEARCH (65536 * BDRV_SECTOR_SIZE) |
c163b5ca | 38 | |
f77dcdbc WC |
39 | #define MAX_INFLIGHT_IO 512 |
40 | ||
c163b5ca LS |
41 | //#define DEBUG_BLK_MIGRATION |
42 | ||
43 | #ifdef DEBUG_BLK_MIGRATION | |
d0f2c4c6 | 44 | #define DPRINTF(fmt, ...) \ |
c163b5ca LS |
45 | do { printf("blk_migration: " fmt, ## __VA_ARGS__); } while (0) |
46 | #else | |
d0f2c4c6 | 47 | #define DPRINTF(fmt, ...) \ |
c163b5ca LS |
48 | do { } while (0) |
49 | #endif | |
50 | ||
a55eb92c | 51 | typedef struct BlkMigDevState { |
323920c4 | 52 | /* Written during setup phase. Can be read without a lock. */ |
ebd2f9e7 KW |
53 | BlockBackend *blk; |
54 | char *blk_name; | |
a55eb92c | 55 | int shared_base; |
a55eb92c | 56 | int64_t total_sectors; |
5e5328be | 57 | QSIMPLEQ_ENTRY(BlkMigDevState) entry; |
ef0716df | 58 | Error *blocker; |
323920c4 PB |
59 | |
60 | /* Only used by migration thread. Does not need a lock. */ | |
61 | int bulk_completed; | |
62 | int64_t cur_sector; | |
63 | int64_t cur_dirty; | |
64 | ||
ef0716df PB |
65 | /* Data in the aio_bitmap is protected by block migration lock. |
66 | * Allocation and free happen during setup and cleanup respectively. | |
67 | */ | |
33656af7 | 68 | unsigned long *aio_bitmap; |
ef0716df PB |
69 | |
70 | /* Protected by block migration lock. */ | |
323920c4 | 71 | int64_t completed_sectors; |
ef0716df PB |
72 | |
73 | /* During migration this is protected by iothread lock / AioContext. | |
74 | * Allocation and free happen during setup and cleanup respectively. | |
75 | */ | |
e4654d2d | 76 | BdrvDirtyBitmap *dirty_bitmap; |
a55eb92c JK |
77 | } BlkMigDevState; |
78 | ||
c163b5ca | 79 | typedef struct BlkMigBlock { |
323920c4 | 80 | /* Only used by migration thread. */ |
c163b5ca LS |
81 | uint8_t *buf; |
82 | BlkMigDevState *bmds; | |
83 | int64_t sector; | |
33656af7 | 84 | int nr_sectors; |
c163b5ca LS |
85 | struct iovec iov; |
86 | QEMUIOVector qiov; | |
7c84b1b8 | 87 | BlockAIOCB *aiocb; |
323920c4 | 88 | |
52e850de | 89 | /* Protected by block migration lock. */ |
c163b5ca | 90 | int ret; |
5e5328be | 91 | QSIMPLEQ_ENTRY(BlkMigBlock) entry; |
c163b5ca LS |
92 | } BlkMigBlock; |
93 | ||
94 | typedef struct BlkMigState { | |
5e5328be | 95 | QSIMPLEQ_HEAD(bmds_list, BlkMigDevState) bmds_list; |
323920c4 | 96 | int64_t total_sector_sum; |
323004a3 | 97 | bool zero_blocks; |
323920c4 | 98 | |
52e850de | 99 | /* Protected by lock. */ |
5e5328be | 100 | QSIMPLEQ_HEAD(blk_list, BlkMigBlock) blk_list; |
c163b5ca LS |
101 | int submitted; |
102 | int read_done; | |
323920c4 PB |
103 | |
104 | /* Only used by migration thread. Does not need a lock. */ | |
c163b5ca | 105 | int transferred; |
01e61e2d | 106 | int prev_progress; |
e970ec0b | 107 | int bulk_completed; |
52e850de | 108 | |
ef0716df | 109 | /* Lock must be taken _inside_ the iothread lock and any AioContexts. */ |
52e850de | 110 | QemuMutex lock; |
c163b5ca LS |
111 | } BlkMigState; |
112 | ||
d11ecd3d | 113 | static BlkMigState block_mig_state; |
c163b5ca | 114 | |
52e850de PB |
115 | static void blk_mig_lock(void) |
116 | { | |
117 | qemu_mutex_lock(&block_mig_state.lock); | |
118 | } | |
119 | ||
120 | static void blk_mig_unlock(void) | |
121 | { | |
122 | qemu_mutex_unlock(&block_mig_state.lock); | |
123 | } | |
124 | ||
32c835ba PB |
125 | /* Must run outside of the iothread lock during the bulk phase, |
126 | * or the VM will stall. | |
127 | */ | |
128 | ||
13f0b67f JK |
129 | static void blk_send(QEMUFile *f, BlkMigBlock * blk) |
130 | { | |
131 | int len; | |
323004a3 PL |
132 | uint64_t flags = BLK_MIG_FLAG_DEVICE_BLOCK; |
133 | ||
134 | if (block_mig_state.zero_blocks && | |
135 | buffer_is_zero(blk->buf, BLOCK_SIZE)) { | |
136 | flags |= BLK_MIG_FLAG_ZERO_BLOCK; | |
137 | } | |
13f0b67f JK |
138 | |
139 | /* sector number and flags */ | |
140 | qemu_put_be64(f, (blk->sector << BDRV_SECTOR_BITS) | |
323004a3 | 141 | | flags); |
13f0b67f JK |
142 | |
143 | /* device name */ | |
ebd2f9e7 | 144 | len = strlen(blk->bmds->blk_name); |
13f0b67f | 145 | qemu_put_byte(f, len); |
ebd2f9e7 | 146 | qemu_put_buffer(f, (uint8_t *) blk->bmds->blk_name, len); |
13f0b67f | 147 | |
323004a3 PL |
148 | /* if a block is zero we need to flush here since the network |
149 | * bandwidth is now a lot higher than the storage device bandwidth. | |
150 | * thus if we queue zero blocks we slow down the migration */ | |
151 | if (flags & BLK_MIG_FLAG_ZERO_BLOCK) { | |
152 | qemu_fflush(f); | |
153 | return; | |
154 | } | |
155 | ||
13f0b67f JK |
156 | qemu_put_buffer(f, blk->buf, BLOCK_SIZE); |
157 | } | |
158 | ||
25f23643 JK |
159 | int blk_mig_active(void) |
160 | { | |
161 | return !QSIMPLEQ_EMPTY(&block_mig_state.bmds_list); | |
162 | } | |
163 | ||
9ac78b61 PL |
164 | int blk_mig_bulk_active(void) |
165 | { | |
166 | return blk_mig_active() && !block_mig_state.bulk_completed; | |
167 | } | |
168 | ||
25f23643 JK |
169 | uint64_t blk_mig_bytes_transferred(void) |
170 | { | |
171 | BlkMigDevState *bmds; | |
172 | uint64_t sum = 0; | |
173 | ||
52e850de | 174 | blk_mig_lock(); |
25f23643 JK |
175 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { |
176 | sum += bmds->completed_sectors; | |
177 | } | |
52e850de | 178 | blk_mig_unlock(); |
25f23643 JK |
179 | return sum << BDRV_SECTOR_BITS; |
180 | } | |
181 | ||
182 | uint64_t blk_mig_bytes_remaining(void) | |
183 | { | |
184 | return blk_mig_bytes_total() - blk_mig_bytes_transferred(); | |
185 | } | |
186 | ||
187 | uint64_t blk_mig_bytes_total(void) | |
188 | { | |
189 | BlkMigDevState *bmds; | |
190 | uint64_t sum = 0; | |
191 | ||
192 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
193 | sum += bmds->total_sectors; | |
194 | } | |
195 | return sum << BDRV_SECTOR_BITS; | |
196 | } | |
197 | ||
52e850de PB |
198 | |
199 | /* Called with migration lock held. */ | |
200 | ||
33656af7 MT |
201 | static int bmds_aio_inflight(BlkMigDevState *bmds, int64_t sector) |
202 | { | |
203 | int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK; | |
204 | ||
ebd2f9e7 | 205 | if (sector < blk_nb_sectors(bmds->blk)) { |
33656af7 MT |
206 | return !!(bmds->aio_bitmap[chunk / (sizeof(unsigned long) * 8)] & |
207 | (1UL << (chunk % (sizeof(unsigned long) * 8)))); | |
208 | } else { | |
209 | return 0; | |
210 | } | |
211 | } | |
212 | ||
52e850de PB |
213 | /* Called with migration lock held. */ |
214 | ||
33656af7 MT |
215 | static void bmds_set_aio_inflight(BlkMigDevState *bmds, int64_t sector_num, |
216 | int nb_sectors, int set) | |
217 | { | |
218 | int64_t start, end; | |
219 | unsigned long val, idx, bit; | |
220 | ||
221 | start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK; | |
222 | end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK; | |
223 | ||
224 | for (; start <= end; start++) { | |
225 | idx = start / (sizeof(unsigned long) * 8); | |
226 | bit = start % (sizeof(unsigned long) * 8); | |
227 | val = bmds->aio_bitmap[idx]; | |
228 | if (set) { | |
62155e2b | 229 | val |= 1UL << bit; |
33656af7 | 230 | } else { |
62155e2b | 231 | val &= ~(1UL << bit); |
33656af7 MT |
232 | } |
233 | bmds->aio_bitmap[idx] = val; | |
234 | } | |
235 | } | |
236 | ||
237 | static void alloc_aio_bitmap(BlkMigDevState *bmds) | |
238 | { | |
ebd2f9e7 | 239 | BlockBackend *bb = bmds->blk; |
33656af7 MT |
240 | int64_t bitmap_size; |
241 | ||
ebd2f9e7 | 242 | bitmap_size = blk_nb_sectors(bb) + BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1; |
33656af7 MT |
243 | bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8; |
244 | ||
7267c094 | 245 | bmds->aio_bitmap = g_malloc0(bitmap_size); |
33656af7 MT |
246 | } |
247 | ||
52e850de PB |
248 | /* Never hold migration lock when yielding to the main loop! */ |
249 | ||
c163b5ca LS |
250 | static void blk_mig_read_cb(void *opaque, int ret) |
251 | { | |
252 | BlkMigBlock *blk = opaque; | |
a55eb92c | 253 | |
52e850de | 254 | blk_mig_lock(); |
c163b5ca | 255 | blk->ret = ret; |
a55eb92c | 256 | |
5e5328be | 257 | QSIMPLEQ_INSERT_TAIL(&block_mig_state.blk_list, blk, entry); |
33656af7 | 258 | bmds_set_aio_inflight(blk->bmds, blk->sector, blk->nr_sectors, 0); |
a55eb92c | 259 | |
d11ecd3d JK |
260 | block_mig_state.submitted--; |
261 | block_mig_state.read_done++; | |
262 | assert(block_mig_state.submitted >= 0); | |
52e850de | 263 | blk_mig_unlock(); |
c163b5ca LS |
264 | } |
265 | ||
32c835ba PB |
266 | /* Called with no lock taken. */ |
267 | ||
539de124 | 268 | static int mig_save_device_bulk(QEMUFile *f, BlkMigDevState *bmds) |
a55eb92c | 269 | { |
57cce12d JK |
270 | int64_t total_sectors = bmds->total_sectors; |
271 | int64_t cur_sector = bmds->cur_sector; | |
ebd2f9e7 | 272 | BlockBackend *bb = bmds->blk; |
c163b5ca | 273 | BlkMigBlock *blk; |
13f0b67f | 274 | int nr_sectors; |
d6a644bb | 275 | int64_t count; |
a55eb92c | 276 | |
57cce12d | 277 | if (bmds->shared_base) { |
32c835ba | 278 | qemu_mutex_lock_iothread(); |
ebd2f9e7 | 279 | aio_context_acquire(blk_get_aio_context(bb)); |
d6a644bb EB |
280 | /* Skip unallocated sectors; intentionally treats failure or |
281 | * partial sector as an allocated sector */ | |
b1d10856 | 282 | while (cur_sector < total_sectors && |
d6a644bb EB |
283 | !bdrv_is_allocated(blk_bs(bb), cur_sector * BDRV_SECTOR_SIZE, |
284 | MAX_IS_ALLOCATED_SEARCH, &count)) { | |
285 | if (count < BDRV_SECTOR_SIZE) { | |
286 | break; | |
287 | } | |
288 | cur_sector += count >> BDRV_SECTOR_BITS; | |
c163b5ca | 289 | } |
ebd2f9e7 | 290 | aio_context_release(blk_get_aio_context(bb)); |
32c835ba | 291 | qemu_mutex_unlock_iothread(); |
c163b5ca | 292 | } |
a55eb92c JK |
293 | |
294 | if (cur_sector >= total_sectors) { | |
82801d8f | 295 | bmds->cur_sector = bmds->completed_sectors = total_sectors; |
c163b5ca LS |
296 | return 1; |
297 | } | |
a55eb92c | 298 | |
82801d8f | 299 | bmds->completed_sectors = cur_sector; |
a55eb92c | 300 | |
57cce12d JK |
301 | cur_sector &= ~((int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK - 1); |
302 | ||
6ea44308 JK |
303 | /* we are going to transfer a full block even if it is not allocated */ |
304 | nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK; | |
c163b5ca | 305 | |
6ea44308 | 306 | if (total_sectors - cur_sector < BDRV_SECTORS_PER_DIRTY_CHUNK) { |
57cce12d | 307 | nr_sectors = total_sectors - cur_sector; |
c163b5ca | 308 | } |
a55eb92c | 309 | |
5839e53b | 310 | blk = g_new(BlkMigBlock, 1); |
7267c094 | 311 | blk->buf = g_malloc(BLOCK_SIZE); |
13f0b67f JK |
312 | blk->bmds = bmds; |
313 | blk->sector = cur_sector; | |
33656af7 | 314 | blk->nr_sectors = nr_sectors; |
a55eb92c | 315 | |
e970ec0b LS |
316 | blk->iov.iov_base = blk->buf; |
317 | blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE; | |
318 | qemu_iovec_init_external(&blk->qiov, &blk->iov, 1); | |
a55eb92c | 319 | |
52e850de | 320 | blk_mig_lock(); |
13197e3c | 321 | block_mig_state.submitted++; |
52e850de | 322 | blk_mig_unlock(); |
13197e3c | 323 | |
ef0716df PB |
324 | /* We do not know if bs is under the main thread (and thus does |
325 | * not acquire the AioContext when doing AIO) or rather under | |
326 | * dataplane. Thus acquire both the iothread mutex and the | |
327 | * AioContext. | |
328 | * | |
329 | * This is ugly and will disappear when we make bdrv_* thread-safe, | |
330 | * without the need to acquire the AioContext. | |
331 | */ | |
32c835ba | 332 | qemu_mutex_lock_iothread(); |
ebd2f9e7 KW |
333 | aio_context_acquire(blk_get_aio_context(bmds->blk)); |
334 | blk->aiocb = blk_aio_preadv(bb, cur_sector * BDRV_SECTOR_SIZE, &blk->qiov, | |
335 | 0, blk_mig_read_cb, blk); | |
d76cac7d | 336 | |
e0d7f73e EB |
337 | bdrv_reset_dirty_bitmap(bmds->dirty_bitmap, cur_sector * BDRV_SECTOR_SIZE, |
338 | nr_sectors * BDRV_SECTOR_SIZE); | |
ebd2f9e7 | 339 | aio_context_release(blk_get_aio_context(bmds->blk)); |
32c835ba | 340 | qemu_mutex_unlock_iothread(); |
a55eb92c | 341 | |
32c835ba | 342 | bmds->cur_sector = cur_sector + nr_sectors; |
13f0b67f | 343 | return (bmds->cur_sector >= total_sectors); |
c163b5ca LS |
344 | } |
345 | ||
32c835ba PB |
346 | /* Called with iothread lock taken. */ |
347 | ||
b8afb520 | 348 | static int set_dirty_tracking(void) |
c163b5ca LS |
349 | { |
350 | BlkMigDevState *bmds; | |
b8afb520 FZ |
351 | int ret; |
352 | ||
353 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
ebd2f9e7 KW |
354 | bmds->dirty_bitmap = bdrv_create_dirty_bitmap(blk_bs(bmds->blk), |
355 | BLOCK_SIZE, NULL, NULL); | |
b8afb520 FZ |
356 | if (!bmds->dirty_bitmap) { |
357 | ret = -errno; | |
358 | goto fail; | |
359 | } | |
360 | } | |
361 | return 0; | |
5e5328be | 362 | |
b8afb520 | 363 | fail: |
5e5328be | 364 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { |
b8afb520 | 365 | if (bmds->dirty_bitmap) { |
ebd2f9e7 | 366 | bdrv_release_dirty_bitmap(blk_bs(bmds->blk), bmds->dirty_bitmap); |
b8afb520 | 367 | } |
e4654d2d | 368 | } |
b8afb520 | 369 | return ret; |
e4654d2d FZ |
370 | } |
371 | ||
ef0716df PB |
372 | /* Called with iothread lock taken. */ |
373 | ||
e4654d2d FZ |
374 | static void unset_dirty_tracking(void) |
375 | { | |
376 | BlkMigDevState *bmds; | |
377 | ||
378 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
ebd2f9e7 | 379 | bdrv_release_dirty_bitmap(blk_bs(bmds->blk), bmds->dirty_bitmap); |
c163b5ca | 380 | } |
c163b5ca LS |
381 | } |
382 | ||
6f5ef23a | 383 | static int init_blk_migration(QEMUFile *f) |
c163b5ca | 384 | { |
fea68bb6 | 385 | BlockDriverState *bs; |
5e5328be | 386 | BlkMigDevState *bmds; |
792773b2 | 387 | int64_t sectors; |
88be7b4b | 388 | BdrvNextIterator it; |
ebd2f9e7 KW |
389 | int i, num_bs = 0; |
390 | struct { | |
391 | BlkMigDevState *bmds; | |
392 | BlockDriverState *bs; | |
393 | } *bmds_bs; | |
6f5ef23a KW |
394 | Error *local_err = NULL; |
395 | int ret; | |
a55eb92c | 396 | |
fea68bb6 MA |
397 | block_mig_state.submitted = 0; |
398 | block_mig_state.read_done = 0; | |
399 | block_mig_state.transferred = 0; | |
400 | block_mig_state.total_sector_sum = 0; | |
401 | block_mig_state.prev_progress = -1; | |
402 | block_mig_state.bulk_completed = 0; | |
403 | block_mig_state.zero_blocks = migrate_zero_blocks(); | |
404 | ||
88be7b4b | 405 | for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) { |
ebd2f9e7 KW |
406 | num_bs++; |
407 | } | |
408 | bmds_bs = g_malloc0(num_bs * sizeof(*bmds_bs)); | |
409 | ||
410 | for (i = 0, bs = bdrv_first(&it); bs; bs = bdrv_next(&it), i++) { | |
fea68bb6 MA |
411 | if (bdrv_is_read_only(bs)) { |
412 | continue; | |
413 | } | |
414 | ||
57322b78 | 415 | sectors = bdrv_nb_sectors(bs); |
31f54f24 | 416 | if (sectors <= 0) { |
6f5ef23a | 417 | ret = sectors; |
5e003f17 | 418 | bdrv_next_cleanup(&it); |
ebd2f9e7 | 419 | goto out; |
b66460e4 SH |
420 | } |
421 | ||
5839e53b | 422 | bmds = g_new0(BlkMigDevState, 1); |
6f5ef23a | 423 | bmds->blk = blk_new(BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL); |
ebd2f9e7 | 424 | bmds->blk_name = g_strdup(bdrv_get_device_name(bs)); |
b66460e4 SH |
425 | bmds->bulk_completed = 0; |
426 | bmds->total_sectors = sectors; | |
427 | bmds->completed_sectors = 0; | |
ce7c817c | 428 | bmds->shared_base = migrate_use_block_incremental(); |
ebd2f9e7 KW |
429 | |
430 | assert(i < num_bs); | |
431 | bmds_bs[i].bmds = bmds; | |
432 | bmds_bs[i].bs = bs; | |
b66460e4 SH |
433 | |
434 | block_mig_state.total_sector_sum += sectors; | |
435 | ||
436 | if (bmds->shared_base) { | |
539de124 | 437 | DPRINTF("Start migration for %s with shared base image\n", |
bfb197e0 | 438 | bdrv_get_device_name(bs)); |
b66460e4 | 439 | } else { |
bfb197e0 | 440 | DPRINTF("Start full migration for %s\n", bdrv_get_device_name(bs)); |
b66460e4 SH |
441 | } |
442 | ||
443 | QSIMPLEQ_INSERT_TAIL(&block_mig_state.bmds_list, bmds, entry); | |
444 | } | |
ebd2f9e7 KW |
445 | |
446 | /* Can only insert new BDSes now because doing so while iterating block | |
447 | * devices may end up in a deadlock (iterating the new BDSes, too). */ | |
448 | for (i = 0; i < num_bs; i++) { | |
449 | BlkMigDevState *bmds = bmds_bs[i].bmds; | |
450 | BlockDriverState *bs = bmds_bs[i].bs; | |
451 | ||
452 | if (bmds) { | |
6f5ef23a KW |
453 | ret = blk_insert_bs(bmds->blk, bs, &local_err); |
454 | if (ret < 0) { | |
455 | error_report_err(local_err); | |
456 | goto out; | |
457 | } | |
ebd2f9e7 KW |
458 | |
459 | alloc_aio_bitmap(bmds); | |
460 | error_setg(&bmds->blocker, "block device is in use by migration"); | |
461 | bdrv_op_block_all(bs, bmds->blocker); | |
462 | } | |
463 | } | |
464 | ||
6f5ef23a | 465 | ret = 0; |
ebd2f9e7 KW |
466 | out: |
467 | g_free(bmds_bs); | |
6f5ef23a | 468 | return ret; |
b66460e4 SH |
469 | } |
470 | ||
32c835ba PB |
471 | /* Called with no lock taken. */ |
472 | ||
539de124 | 473 | static int blk_mig_save_bulked_block(QEMUFile *f) |
c163b5ca | 474 | { |
82801d8f | 475 | int64_t completed_sector_sum = 0; |
c163b5ca | 476 | BlkMigDevState *bmds; |
01e61e2d | 477 | int progress; |
82801d8f | 478 | int ret = 0; |
c163b5ca | 479 | |
5e5328be | 480 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { |
a55eb92c | 481 | if (bmds->bulk_completed == 0) { |
539de124 | 482 | if (mig_save_device_bulk(f, bmds) == 1) { |
57cce12d JK |
483 | /* completed bulk section for this device */ |
484 | bmds->bulk_completed = 1; | |
c163b5ca | 485 | } |
82801d8f JK |
486 | completed_sector_sum += bmds->completed_sectors; |
487 | ret = 1; | |
488 | break; | |
489 | } else { | |
490 | completed_sector_sum += bmds->completed_sectors; | |
c163b5ca LS |
491 | } |
492 | } | |
a55eb92c | 493 | |
8b6b2afc PR |
494 | if (block_mig_state.total_sector_sum != 0) { |
495 | progress = completed_sector_sum * 100 / | |
496 | block_mig_state.total_sector_sum; | |
497 | } else { | |
498 | progress = 100; | |
499 | } | |
01e61e2d JK |
500 | if (progress != block_mig_state.prev_progress) { |
501 | block_mig_state.prev_progress = progress; | |
502 | qemu_put_be64(f, (progress << BDRV_SECTOR_BITS) | |
503 | | BLK_MIG_FLAG_PROGRESS); | |
539de124 | 504 | DPRINTF("Completed %d %%\r", progress); |
82801d8f JK |
505 | } |
506 | ||
507 | return ret; | |
c163b5ca LS |
508 | } |
509 | ||
d76cac7d | 510 | static void blk_mig_reset_dirty_cursor(void) |
c163b5ca LS |
511 | { |
512 | BlkMigDevState *bmds; | |
d76cac7d LS |
513 | |
514 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
515 | bmds->cur_dirty = 0; | |
516 | } | |
517 | } | |
518 | ||
ef0716df | 519 | /* Called with iothread lock and AioContext taken. */ |
32c835ba | 520 | |
539de124 LC |
521 | static int mig_save_device_dirty(QEMUFile *f, BlkMigDevState *bmds, |
522 | int is_async) | |
d76cac7d LS |
523 | { |
524 | BlkMigBlock *blk; | |
ebd2f9e7 | 525 | BlockDriverState *bs = blk_bs(bmds->blk); |
d76cac7d | 526 | int64_t total_sectors = bmds->total_sectors; |
c163b5ca | 527 | int64_t sector; |
d76cac7d | 528 | int nr_sectors; |
dcd1d224 | 529 | int ret = -EIO; |
a55eb92c | 530 | |
d76cac7d | 531 | for (sector = bmds->cur_dirty; sector < bmds->total_sectors;) { |
52e850de | 532 | blk_mig_lock(); |
62155e2b | 533 | if (bmds_aio_inflight(bmds, sector)) { |
52e850de | 534 | blk_mig_unlock(); |
ebd2f9e7 | 535 | blk_drain(bmds->blk); |
52e850de PB |
536 | } else { |
537 | blk_mig_unlock(); | |
62155e2b | 538 | } |
b64bd51e | 539 | bdrv_dirty_bitmap_lock(bmds->dirty_bitmap); |
3b5d4df0 EB |
540 | if (bdrv_get_dirty_locked(bs, bmds->dirty_bitmap, |
541 | sector * BDRV_SECTOR_SIZE)) { | |
d76cac7d LS |
542 | if (total_sectors - sector < BDRV_SECTORS_PER_DIRTY_CHUNK) { |
543 | nr_sectors = total_sectors - sector; | |
544 | } else { | |
545 | nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK; | |
546 | } | |
e0d7f73e EB |
547 | bdrv_reset_dirty_bitmap_locked(bmds->dirty_bitmap, |
548 | sector * BDRV_SECTOR_SIZE, | |
549 | nr_sectors * BDRV_SECTOR_SIZE); | |
b64bd51e | 550 | bdrv_dirty_bitmap_unlock(bmds->dirty_bitmap); |
c0bad499 | 551 | |
5839e53b | 552 | blk = g_new(BlkMigBlock, 1); |
7267c094 | 553 | blk->buf = g_malloc(BLOCK_SIZE); |
d76cac7d LS |
554 | blk->bmds = bmds; |
555 | blk->sector = sector; | |
33656af7 | 556 | blk->nr_sectors = nr_sectors; |
d76cac7d | 557 | |
889ae39c | 558 | if (is_async) { |
d76cac7d LS |
559 | blk->iov.iov_base = blk->buf; |
560 | blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE; | |
561 | qemu_iovec_init_external(&blk->qiov, &blk->iov, 1); | |
562 | ||
ebd2f9e7 KW |
563 | blk->aiocb = blk_aio_preadv(bmds->blk, |
564 | sector * BDRV_SECTOR_SIZE, | |
565 | &blk->qiov, 0, blk_mig_read_cb, | |
566 | blk); | |
52e850de PB |
567 | |
568 | blk_mig_lock(); | |
d76cac7d | 569 | block_mig_state.submitted++; |
33656af7 | 570 | bmds_set_aio_inflight(bmds, sector, nr_sectors, 1); |
52e850de | 571 | blk_mig_unlock(); |
d76cac7d | 572 | } else { |
ebd2f9e7 KW |
573 | ret = blk_pread(bmds->blk, sector * BDRV_SECTOR_SIZE, blk->buf, |
574 | nr_sectors * BDRV_SECTOR_SIZE); | |
dcd1d224 | 575 | if (ret < 0) { |
d76cac7d | 576 | goto error; |
c163b5ca | 577 | } |
d76cac7d | 578 | blk_send(f, blk); |
a55eb92c | 579 | |
7267c094 AL |
580 | g_free(blk->buf); |
581 | g_free(blk); | |
a55eb92c | 582 | } |
d76cac7d | 583 | |
1cf6aa74 LC |
584 | sector += nr_sectors; |
585 | bmds->cur_dirty = sector; | |
d76cac7d | 586 | break; |
c163b5ca | 587 | } |
b64bd51e PB |
588 | |
589 | bdrv_dirty_bitmap_unlock(bmds->dirty_bitmap); | |
d76cac7d LS |
590 | sector += BDRV_SECTORS_PER_DIRTY_CHUNK; |
591 | bmds->cur_dirty = sector; | |
c163b5ca | 592 | } |
575a58d7 | 593 | |
d76cac7d LS |
594 | return (bmds->cur_dirty >= bmds->total_sectors); |
595 | ||
889ae39c | 596 | error: |
539de124 | 597 | DPRINTF("Error reading sector %" PRId64 "\n", sector); |
7267c094 AL |
598 | g_free(blk->buf); |
599 | g_free(blk); | |
43be3a25 | 600 | return ret; |
d76cac7d LS |
601 | } |
602 | ||
32c835ba PB |
603 | /* Called with iothread lock taken. |
604 | * | |
605 | * return value: | |
ceb2bd09 JQ |
606 | * 0: too much data for max_downtime |
607 | * 1: few enough data for max_downtime | |
608 | */ | |
539de124 | 609 | static int blk_mig_save_dirty_block(QEMUFile *f, int is_async) |
d76cac7d LS |
610 | { |
611 | BlkMigDevState *bmds; | |
ceb2bd09 | 612 | int ret = 1; |
d76cac7d LS |
613 | |
614 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
ebd2f9e7 | 615 | aio_context_acquire(blk_get_aio_context(bmds->blk)); |
ceb2bd09 | 616 | ret = mig_save_device_dirty(f, bmds, is_async); |
ebd2f9e7 | 617 | aio_context_release(blk_get_aio_context(bmds->blk)); |
43be3a25 | 618 | if (ret <= 0) { |
d76cac7d LS |
619 | break; |
620 | } | |
621 | } | |
622 | ||
623 | return ret; | |
c163b5ca LS |
624 | } |
625 | ||
32c835ba PB |
626 | /* Called with no locks taken. */ |
627 | ||
59feec42 | 628 | static int flush_blks(QEMUFile *f) |
c163b5ca | 629 | { |
5e5328be | 630 | BlkMigBlock *blk; |
59feec42 | 631 | int ret = 0; |
a55eb92c | 632 | |
d0f2c4c6 | 633 | DPRINTF("%s Enter submitted %d read_done %d transferred %d\n", |
d11ecd3d JK |
634 | __FUNCTION__, block_mig_state.submitted, block_mig_state.read_done, |
635 | block_mig_state.transferred); | |
a55eb92c | 636 | |
52e850de | 637 | blk_mig_lock(); |
5e5328be JK |
638 | while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) { |
639 | if (qemu_file_rate_limit(f)) { | |
640 | break; | |
641 | } | |
4b640365 | 642 | if (blk->ret < 0) { |
59feec42 | 643 | ret = blk->ret; |
4b640365 JK |
644 | break; |
645 | } | |
a55eb92c | 646 | |
5e5328be | 647 | QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry); |
52e850de | 648 | blk_mig_unlock(); |
13197e3c | 649 | blk_send(f, blk); |
52e850de | 650 | blk_mig_lock(); |
13197e3c | 651 | |
7267c094 AL |
652 | g_free(blk->buf); |
653 | g_free(blk); | |
a55eb92c | 654 | |
d11ecd3d JK |
655 | block_mig_state.read_done--; |
656 | block_mig_state.transferred++; | |
657 | assert(block_mig_state.read_done >= 0); | |
c163b5ca | 658 | } |
52e850de | 659 | blk_mig_unlock(); |
c163b5ca | 660 | |
d0f2c4c6 | 661 | DPRINTF("%s Exit submitted %d read_done %d transferred %d\n", __FUNCTION__, |
d11ecd3d JK |
662 | block_mig_state.submitted, block_mig_state.read_done, |
663 | block_mig_state.transferred); | |
59feec42 | 664 | return ret; |
c163b5ca LS |
665 | } |
666 | ||
32c835ba PB |
667 | /* Called with iothread lock taken. */ |
668 | ||
889ae39c LS |
669 | static int64_t get_remaining_dirty(void) |
670 | { | |
671 | BlkMigDevState *bmds; | |
672 | int64_t dirty = 0; | |
673 | ||
674 | QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) { | |
ebd2f9e7 | 675 | aio_context_acquire(blk_get_aio_context(bmds->blk)); |
20dca810 | 676 | dirty += bdrv_get_dirty_count(bmds->dirty_bitmap); |
ebd2f9e7 | 677 | aio_context_release(blk_get_aio_context(bmds->blk)); |
889ae39c LS |
678 | } |
679 | ||
9a46dba7 | 680 | return dirty; |
889ae39c LS |
681 | } |
682 | ||
32c835ba | 683 | |
362fdf17 KW |
684 | |
685 | /* Called with iothread lock taken. */ | |
686 | static void block_migration_cleanup_bmds(void) | |
4ec7fcc7 | 687 | { |
82801d8f | 688 | BlkMigDevState *bmds; |
ef0716df | 689 | AioContext *ctx; |
4ec7fcc7 | 690 | |
e4654d2d | 691 | unset_dirty_tracking(); |
8f794c55 | 692 | |
82801d8f JK |
693 | while ((bmds = QSIMPLEQ_FIRST(&block_mig_state.bmds_list)) != NULL) { |
694 | QSIMPLEQ_REMOVE_HEAD(&block_mig_state.bmds_list, entry); | |
ebd2f9e7 | 695 | bdrv_op_unblock_all(blk_bs(bmds->blk), bmds->blocker); |
3718d8ab | 696 | error_free(bmds->blocker); |
ef0716df | 697 | |
ebd2f9e7 KW |
698 | /* Save ctx, because bmds->blk can disappear during blk_unref. */ |
699 | ctx = blk_get_aio_context(bmds->blk); | |
ef0716df | 700 | aio_context_acquire(ctx); |
ebd2f9e7 | 701 | blk_unref(bmds->blk); |
ef0716df PB |
702 | aio_context_release(ctx); |
703 | ||
ebd2f9e7 | 704 | g_free(bmds->blk_name); |
7267c094 AL |
705 | g_free(bmds->aio_bitmap); |
706 | g_free(bmds); | |
4ec7fcc7 | 707 | } |
362fdf17 KW |
708 | } |
709 | ||
710 | /* Called with iothread lock taken. */ | |
711 | static void block_migration_cleanup(void *opaque) | |
712 | { | |
713 | BlkMigBlock *blk; | |
714 | ||
715 | bdrv_drain_all(); | |
716 | ||
717 | block_migration_cleanup_bmds(); | |
4ec7fcc7 | 718 | |
ef0716df | 719 | blk_mig_lock(); |
82801d8f JK |
720 | while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) { |
721 | QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry); | |
7267c094 AL |
722 | g_free(blk->buf); |
723 | g_free(blk); | |
4ec7fcc7 | 724 | } |
52e850de | 725 | blk_mig_unlock(); |
4ec7fcc7 JK |
726 | } |
727 | ||
d1315aac | 728 | static int block_save_setup(QEMUFile *f, void *opaque) |
c163b5ca | 729 | { |
2975725f JQ |
730 | int ret; |
731 | ||
d1315aac JQ |
732 | DPRINTF("Enter save live setup submitted %d transferred %d\n", |
733 | block_mig_state.submitted, block_mig_state.transferred); | |
a55eb92c | 734 | |
9b095037 | 735 | qemu_mutex_lock_iothread(); |
6f5ef23a KW |
736 | ret = init_blk_migration(f); |
737 | if (ret < 0) { | |
738 | qemu_mutex_unlock_iothread(); | |
739 | return ret; | |
740 | } | |
d1315aac JQ |
741 | |
742 | /* start track dirty blocks */ | |
b8afb520 FZ |
743 | ret = set_dirty_tracking(); |
744 | ||
ef0716df PB |
745 | qemu_mutex_unlock_iothread(); |
746 | ||
b8afb520 | 747 | if (ret) { |
b8afb520 FZ |
748 | return ret; |
749 | } | |
750 | ||
59feec42 | 751 | ret = flush_blks(f); |
d1315aac | 752 | blk_mig_reset_dirty_cursor(); |
d1315aac JQ |
753 | qemu_put_be64(f, BLK_MIG_FLAG_EOS); |
754 | ||
d418cf57 | 755 | return ret; |
d1315aac JQ |
756 | } |
757 | ||
16310a3c | 758 | static int block_save_iterate(QEMUFile *f, void *opaque) |
d1315aac JQ |
759 | { |
760 | int ret; | |
6aaa9dae | 761 | int64_t last_ftell = qemu_ftell(f); |
ebd9fbd7 | 762 | int64_t delta_ftell; |
d1315aac | 763 | |
16310a3c JQ |
764 | DPRINTF("Enter save live iterate submitted %d transferred %d\n", |
765 | block_mig_state.submitted, block_mig_state.transferred); | |
d1315aac | 766 | |
59feec42 | 767 | ret = flush_blks(f); |
2975725f | 768 | if (ret) { |
2975725f | 769 | return ret; |
4b640365 JK |
770 | } |
771 | ||
d76cac7d LS |
772 | blk_mig_reset_dirty_cursor(); |
773 | ||
16310a3c | 774 | /* control the rate of transfer */ |
52e850de | 775 | blk_mig_lock(); |
16310a3c JQ |
776 | while ((block_mig_state.submitted + |
777 | block_mig_state.read_done) * BLOCK_SIZE < | |
f77dcdbc WC |
778 | qemu_file_get_rate_limit(f) && |
779 | (block_mig_state.submitted + | |
780 | block_mig_state.read_done) < | |
781 | MAX_INFLIGHT_IO) { | |
52e850de | 782 | blk_mig_unlock(); |
16310a3c JQ |
783 | if (block_mig_state.bulk_completed == 0) { |
784 | /* first finish the bulk phase */ | |
785 | if (blk_mig_save_bulked_block(f) == 0) { | |
786 | /* finished saving bulk on all devices */ | |
787 | block_mig_state.bulk_completed = 1; | |
788 | } | |
13197e3c | 789 | ret = 0; |
16310a3c | 790 | } else { |
32c835ba PB |
791 | /* Always called with iothread lock taken for |
792 | * simplicity, block_save_complete also calls it. | |
793 | */ | |
794 | qemu_mutex_lock_iothread(); | |
43be3a25 | 795 | ret = blk_mig_save_dirty_block(f, 1); |
32c835ba | 796 | qemu_mutex_unlock_iothread(); |
13197e3c PB |
797 | } |
798 | if (ret < 0) { | |
799 | return ret; | |
800 | } | |
52e850de | 801 | blk_mig_lock(); |
13197e3c PB |
802 | if (ret != 0) { |
803 | /* no more dirty blocks */ | |
804 | break; | |
a55eb92c | 805 | } |
16310a3c | 806 | } |
52e850de | 807 | blk_mig_unlock(); |
a55eb92c | 808 | |
59feec42 | 809 | ret = flush_blks(f); |
16310a3c | 810 | if (ret) { |
16310a3c | 811 | return ret; |
4b640365 JK |
812 | } |
813 | ||
16310a3c | 814 | qemu_put_be64(f, BLK_MIG_FLAG_EOS); |
ebd9fbd7 GH |
815 | delta_ftell = qemu_ftell(f) - last_ftell; |
816 | if (delta_ftell > 0) { | |
817 | return 1; | |
818 | } else if (delta_ftell < 0) { | |
819 | return -1; | |
820 | } else { | |
821 | return 0; | |
822 | } | |
16310a3c JQ |
823 | } |
824 | ||
32c835ba PB |
825 | /* Called with iothread lock taken. */ |
826 | ||
16310a3c JQ |
827 | static int block_save_complete(QEMUFile *f, void *opaque) |
828 | { | |
829 | int ret; | |
830 | ||
831 | DPRINTF("Enter save live complete submitted %d transferred %d\n", | |
832 | block_mig_state.submitted, block_mig_state.transferred); | |
833 | ||
59feec42 | 834 | ret = flush_blks(f); |
16310a3c | 835 | if (ret) { |
16310a3c JQ |
836 | return ret; |
837 | } | |
a55eb92c | 838 | |
16310a3c | 839 | blk_mig_reset_dirty_cursor(); |
01e61e2d | 840 | |
16310a3c JQ |
841 | /* we know for sure that save bulk is completed and |
842 | all async read completed */ | |
52e850de | 843 | blk_mig_lock(); |
16310a3c | 844 | assert(block_mig_state.submitted == 0); |
52e850de | 845 | blk_mig_unlock(); |
16310a3c | 846 | |
43be3a25 JQ |
847 | do { |
848 | ret = blk_mig_save_dirty_block(f, 0); | |
d418cf57 PB |
849 | if (ret < 0) { |
850 | return ret; | |
851 | } | |
43be3a25 | 852 | } while (ret == 0); |
4b640365 | 853 | |
43be3a25 JQ |
854 | /* report completion */ |
855 | qemu_put_be64(f, (100 << BDRV_SECTOR_BITS) | BLK_MIG_FLAG_PROGRESS); | |
a55eb92c | 856 | |
16310a3c JQ |
857 | DPRINTF("Block migration completed\n"); |
858 | ||
a55eb92c JK |
859 | qemu_put_be64(f, BLK_MIG_FLAG_EOS); |
860 | ||
362fdf17 KW |
861 | /* Make sure that our BlockBackends are gone, so that the block driver |
862 | * nodes can be inactivated. */ | |
863 | block_migration_cleanup_bmds(); | |
864 | ||
16310a3c | 865 | return 0; |
c163b5ca LS |
866 | } |
867 | ||
c31b098f DDAG |
868 | static void block_save_pending(QEMUFile *f, void *opaque, uint64_t max_size, |
869 | uint64_t *non_postcopiable_pending, | |
870 | uint64_t *postcopiable_pending) | |
e4ed1541 | 871 | { |
6aaa9dae | 872 | /* Estimate pending number of bytes to send */ |
13197e3c PB |
873 | uint64_t pending; |
874 | ||
32c835ba | 875 | qemu_mutex_lock_iothread(); |
ef0716df PB |
876 | pending = get_remaining_dirty(); |
877 | qemu_mutex_unlock_iothread(); | |
878 | ||
52e850de | 879 | blk_mig_lock(); |
ef0716df PB |
880 | pending += block_mig_state.submitted * BLOCK_SIZE + |
881 | block_mig_state.read_done * BLOCK_SIZE; | |
882 | blk_mig_unlock(); | |
6aaa9dae SH |
883 | |
884 | /* Report at least one block pending during bulk phase */ | |
04636dc4 VSO |
885 | if (pending <= max_size && !block_mig_state.bulk_completed) { |
886 | pending = max_size + BLOCK_SIZE; | |
6aaa9dae | 887 | } |
e4ed1541 | 888 | |
6aaa9dae | 889 | DPRINTF("Enter save live pending %" PRIu64 "\n", pending); |
c31b098f DDAG |
890 | /* We don't do postcopy */ |
891 | *non_postcopiable_pending += pending; | |
e4ed1541 JQ |
892 | } |
893 | ||
c163b5ca LS |
894 | static int block_load(QEMUFile *f, void *opaque, int version_id) |
895 | { | |
01e61e2d | 896 | static int banner_printed; |
c163b5ca LS |
897 | int len, flags; |
898 | char device_name[256]; | |
899 | int64_t addr; | |
3c254ab8 | 900 | BlockBackend *blk, *blk_prev = NULL; |
9bd9c7f5 | 901 | Error *local_err = NULL; |
c163b5ca | 902 | uint8_t *buf; |
77358b59 PR |
903 | int64_t total_sectors = 0; |
904 | int nr_sectors; | |
42802d47 | 905 | int ret; |
3928d50b LC |
906 | BlockDriverInfo bdi; |
907 | int cluster_size = BLOCK_SIZE; | |
a55eb92c | 908 | |
c163b5ca | 909 | do { |
c163b5ca | 910 | addr = qemu_get_be64(f); |
a55eb92c | 911 | |
6ea44308 JK |
912 | flags = addr & ~BDRV_SECTOR_MASK; |
913 | addr >>= BDRV_SECTOR_BITS; | |
a55eb92c JK |
914 | |
915 | if (flags & BLK_MIG_FLAG_DEVICE_BLOCK) { | |
c163b5ca LS |
916 | /* get device name */ |
917 | len = qemu_get_byte(f); | |
c163b5ca LS |
918 | qemu_get_buffer(f, (uint8_t *)device_name, len); |
919 | device_name[len] = '\0'; | |
a55eb92c | 920 | |
c9ebaf74 FZ |
921 | blk = blk_by_name(device_name); |
922 | if (!blk) { | |
4b640365 JK |
923 | fprintf(stderr, "Error unknown block device %s\n", |
924 | device_name); | |
925 | return -EINVAL; | |
926 | } | |
a55eb92c | 927 | |
ad2964b4 KW |
928 | if (blk != blk_prev) { |
929 | blk_prev = blk; | |
930 | total_sectors = blk_nb_sectors(blk); | |
77358b59 | 931 | if (total_sectors <= 0) { |
6daf194d | 932 | error_report("Error getting length of block device %s", |
77358b59 PR |
933 | device_name); |
934 | return -EINVAL; | |
935 | } | |
9bd9c7f5 | 936 | |
ad2964b4 | 937 | blk_invalidate_cache(blk, &local_err); |
9bd9c7f5 KW |
938 | if (local_err) { |
939 | error_report_err(local_err); | |
940 | return -EINVAL; | |
941 | } | |
3928d50b LC |
942 | |
943 | ret = bdrv_get_info(blk_bs(blk), &bdi); | |
944 | if (ret == 0 && bdi.cluster_size > 0 && | |
945 | bdi.cluster_size <= BLOCK_SIZE && | |
946 | BLOCK_SIZE % bdi.cluster_size == 0) { | |
947 | cluster_size = bdi.cluster_size; | |
948 | } else { | |
949 | cluster_size = BLOCK_SIZE; | |
950 | } | |
77358b59 PR |
951 | } |
952 | ||
953 | if (total_sectors - addr < BDRV_SECTORS_PER_DIRTY_CHUNK) { | |
954 | nr_sectors = total_sectors - addr; | |
955 | } else { | |
956 | nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK; | |
957 | } | |
958 | ||
323004a3 | 959 | if (flags & BLK_MIG_FLAG_ZERO_BLOCK) { |
ad2964b4 KW |
960 | ret = blk_pwrite_zeroes(blk, addr * BDRV_SECTOR_SIZE, |
961 | nr_sectors * BDRV_SECTOR_SIZE, | |
962 | BDRV_REQ_MAY_UNMAP); | |
323004a3 | 963 | } else { |
3928d50b LC |
964 | int i; |
965 | int64_t cur_addr; | |
966 | uint8_t *cur_buf; | |
967 | ||
323004a3 PL |
968 | buf = g_malloc(BLOCK_SIZE); |
969 | qemu_get_buffer(f, buf, BLOCK_SIZE); | |
3928d50b LC |
970 | for (i = 0; i < BLOCK_SIZE / cluster_size; i++) { |
971 | cur_addr = addr * BDRV_SECTOR_SIZE + i * cluster_size; | |
972 | cur_buf = buf + i * cluster_size; | |
973 | ||
974 | if ((!block_mig_state.zero_blocks || | |
975 | cluster_size < BLOCK_SIZE) && | |
976 | buffer_is_zero(cur_buf, cluster_size)) { | |
977 | ret = blk_pwrite_zeroes(blk, cur_addr, | |
978 | cluster_size, | |
979 | BDRV_REQ_MAY_UNMAP); | |
980 | } else { | |
981 | ret = blk_pwrite(blk, cur_addr, cur_buf, | |
982 | cluster_size, 0); | |
983 | } | |
984 | if (ret < 0) { | |
985 | break; | |
986 | } | |
987 | } | |
323004a3 PL |
988 | g_free(buf); |
989 | } | |
575a58d7 | 990 | |
b02bea3a YT |
991 | if (ret < 0) { |
992 | return ret; | |
993 | } | |
01e61e2d JK |
994 | } else if (flags & BLK_MIG_FLAG_PROGRESS) { |
995 | if (!banner_printed) { | |
996 | printf("Receiving block device images\n"); | |
997 | banner_printed = 1; | |
998 | } | |
999 | printf("Completed %d %%%c", (int)addr, | |
1000 | (addr == 100) ? '\n' : '\r'); | |
1001 | fflush(stdout); | |
a55eb92c | 1002 | } else if (!(flags & BLK_MIG_FLAG_EOS)) { |
d5f1f286 | 1003 | fprintf(stderr, "Unknown block migration flags: %#x\n", flags); |
4b640365 JK |
1004 | return -EINVAL; |
1005 | } | |
42802d47 JQ |
1006 | ret = qemu_file_get_error(f); |
1007 | if (ret != 0) { | |
1008 | return ret; | |
c163b5ca | 1009 | } |
a55eb92c JK |
1010 | } while (!(flags & BLK_MIG_FLAG_EOS)); |
1011 | ||
c163b5ca LS |
1012 | return 0; |
1013 | } | |
1014 | ||
6bd68781 JQ |
1015 | static bool block_is_active(void *opaque) |
1016 | { | |
ce7c817c | 1017 | return migrate_use_block(); |
6bd68781 JQ |
1018 | } |
1019 | ||
7a46d042 | 1020 | static SaveVMHandlers savevm_block_handlers = { |
9907e842 | 1021 | .save_setup = block_save_setup, |
16310a3c | 1022 | .save_live_iterate = block_save_iterate, |
a3e06c3d | 1023 | .save_live_complete_precopy = block_save_complete, |
e4ed1541 | 1024 | .save_live_pending = block_save_pending, |
7908c78d | 1025 | .load_state = block_load, |
70f794fc | 1026 | .save_cleanup = block_migration_cleanup, |
6bd68781 | 1027 | .is_active = block_is_active, |
7908c78d JQ |
1028 | }; |
1029 | ||
c163b5ca | 1030 | void blk_mig_init(void) |
a55eb92c | 1031 | { |
5e5328be JK |
1032 | QSIMPLEQ_INIT(&block_mig_state.bmds_list); |
1033 | QSIMPLEQ_INIT(&block_mig_state.blk_list); | |
52e850de | 1034 | qemu_mutex_init(&block_mig_state.lock); |
5e5328be | 1035 | |
7908c78d JQ |
1036 | register_savevm_live(NULL, "block", 0, 1, &savevm_block_handlers, |
1037 | &block_mig_state); | |
c163b5ca | 1038 | } |