blob: 25a03889f471c0a81dec66d59519137c9f7fd13d [file] [log] [blame]
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +02001/*
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 *
Paolo Bonzini6b620ca2012-01-13 17:44:23 +010012 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020014 */
15
16#include "qemu-common.h"
Paolo Bonzini737e1502012-12-17 18:19:44 +010017#include "block/block_int.h"
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020018#include "hw/hw.h"
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010019#include "qemu/queue.h"
20#include "qemu/timer.h"
Paolo Bonzinicaf71f82012-12-17 18:19:50 +010021#include "migration/block.h"
22#include "migration/migration.h"
Paolo Bonzini9c17d612012-12-17 18:20:04 +010023#include "sysemu/blockdev.h"
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020024#include <assert.h>
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020025
Paolo Bonzini50717e92013-01-21 17:09:45 +010026#define BLOCK_SIZE (1 << 20)
27#define BDRV_SECTORS_PER_DIRTY_CHUNK (BLOCK_SIZE >> BDRV_SECTOR_BITS)
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020028
29#define BLK_MIG_FLAG_DEVICE_BLOCK 0x01
30#define BLK_MIG_FLAG_EOS 0x02
Jan Kiszka01e61e22009-12-01 15:20:17 +010031#define BLK_MIG_FLAG_PROGRESS 0x04
Peter Lieven323004a2013-07-18 09:48:50 +020032#define BLK_MIG_FLAG_ZERO_BLOCK 0x08
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020033
34#define MAX_IS_ALLOCATED_SEARCH 65536
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020035
36//#define DEBUG_BLK_MIGRATION
37
38#ifdef DEBUG_BLK_MIGRATION
malcd0f2c4c2010-02-07 02:03:50 +030039#define DPRINTF(fmt, ...) \
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020040 do { printf("blk_migration: " fmt, ## __VA_ARGS__); } while (0)
41#else
malcd0f2c4c2010-02-07 02:03:50 +030042#define DPRINTF(fmt, ...) \
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020043 do { } while (0)
44#endif
45
Jan Kiszkaa55eb922009-11-30 18:21:19 +010046typedef struct BlkMigDevState {
Paolo Bonzini323920c2013-02-22 17:36:24 +010047 /* Written during setup phase. Can be read without a lock. */
Jan Kiszkaa55eb922009-11-30 18:21:19 +010048 BlockDriverState *bs;
Jan Kiszkaa55eb922009-11-30 18:21:19 +010049 int shared_base;
Jan Kiszkaa55eb922009-11-30 18:21:19 +010050 int64_t total_sectors;
Jan Kiszka5e5328b2009-11-30 18:21:20 +010051 QSIMPLEQ_ENTRY(BlkMigDevState) entry;
Paolo Bonzini323920c2013-02-22 17:36:24 +010052
53 /* Only used by migration thread. Does not need a lock. */
54 int bulk_completed;
55 int64_t cur_sector;
56 int64_t cur_dirty;
57
Paolo Bonzini52e850d2013-02-22 17:36:25 +010058 /* Protected by block migration lock. */
Marcelo Tosatti33656af2010-11-08 17:02:56 -020059 unsigned long *aio_bitmap;
Paolo Bonzini323920c2013-02-22 17:36:24 +010060 int64_t completed_sectors;
Fam Zhenge4654d22013-11-13 18:29:43 +080061 BdrvDirtyBitmap *dirty_bitmap;
Fam Zheng3718d8a2014-05-23 21:29:43 +080062 Error *blocker;
Jan Kiszkaa55eb922009-11-30 18:21:19 +010063} BlkMigDevState;
64
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020065typedef struct BlkMigBlock {
Paolo Bonzini323920c2013-02-22 17:36:24 +010066 /* Only used by migration thread. */
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020067 uint8_t *buf;
68 BlkMigDevState *bmds;
69 int64_t sector;
Marcelo Tosatti33656af2010-11-08 17:02:56 -020070 int nr_sectors;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020071 struct iovec iov;
72 QEMUIOVector qiov;
73 BlockDriverAIOCB *aiocb;
Paolo Bonzini323920c2013-02-22 17:36:24 +010074
Paolo Bonzini52e850d2013-02-22 17:36:25 +010075 /* Protected by block migration lock. */
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020076 int ret;
Jan Kiszka5e5328b2009-11-30 18:21:20 +010077 QSIMPLEQ_ENTRY(BlkMigBlock) entry;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020078} BlkMigBlock;
79
80typedef struct BlkMigState {
Paolo Bonzini323920c2013-02-22 17:36:24 +010081 /* Written during setup phase. Can be read without a lock. */
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020082 int blk_enable;
83 int shared_base;
Jan Kiszka5e5328b2009-11-30 18:21:20 +010084 QSIMPLEQ_HEAD(bmds_list, BlkMigDevState) bmds_list;
Paolo Bonzini323920c2013-02-22 17:36:24 +010085 int64_t total_sector_sum;
Peter Lieven323004a2013-07-18 09:48:50 +020086 bool zero_blocks;
Paolo Bonzini323920c2013-02-22 17:36:24 +010087
Paolo Bonzini52e850d2013-02-22 17:36:25 +010088 /* Protected by lock. */
Jan Kiszka5e5328b2009-11-30 18:21:20 +010089 QSIMPLEQ_HEAD(blk_list, BlkMigBlock) blk_list;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020090 int submitted;
91 int read_done;
Paolo Bonzini323920c2013-02-22 17:36:24 +010092
93 /* Only used by migration thread. Does not need a lock. */
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +020094 int transferred;
Jan Kiszka01e61e22009-12-01 15:20:17 +010095 int prev_progress;
Liran Schoure970ec02010-01-26 10:31:45 +020096 int bulk_completed;
Paolo Bonzini52e850d2013-02-22 17:36:25 +010097
98 /* Lock must be taken _inside_ the iothread lock. */
99 QemuMutex lock;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200100} BlkMigState;
101
Jan Kiszkad11ecd32009-11-30 18:21:20 +0100102static BlkMigState block_mig_state;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200103
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100104static void blk_mig_lock(void)
105{
106 qemu_mutex_lock(&block_mig_state.lock);
107}
108
109static void blk_mig_unlock(void)
110{
111 qemu_mutex_unlock(&block_mig_state.lock);
112}
113
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100114/* Must run outside of the iothread lock during the bulk phase,
115 * or the VM will stall.
116 */
117
Jan Kiszka13f0b672009-11-30 18:21:21 +0100118static void blk_send(QEMUFile *f, BlkMigBlock * blk)
119{
120 int len;
Peter Lieven323004a2013-07-18 09:48:50 +0200121 uint64_t flags = BLK_MIG_FLAG_DEVICE_BLOCK;
122
123 if (block_mig_state.zero_blocks &&
124 buffer_is_zero(blk->buf, BLOCK_SIZE)) {
125 flags |= BLK_MIG_FLAG_ZERO_BLOCK;
126 }
Jan Kiszka13f0b672009-11-30 18:21:21 +0100127
128 /* sector number and flags */
129 qemu_put_be64(f, (blk->sector << BDRV_SECTOR_BITS)
Peter Lieven323004a2013-07-18 09:48:50 +0200130 | flags);
Jan Kiszka13f0b672009-11-30 18:21:21 +0100131
132 /* device name */
133 len = strlen(blk->bmds->bs->device_name);
134 qemu_put_byte(f, len);
135 qemu_put_buffer(f, (uint8_t *)blk->bmds->bs->device_name, len);
136
Peter Lieven323004a2013-07-18 09:48:50 +0200137 /* if a block is zero we need to flush here since the network
138 * bandwidth is now a lot higher than the storage device bandwidth.
139 * thus if we queue zero blocks we slow down the migration */
140 if (flags & BLK_MIG_FLAG_ZERO_BLOCK) {
141 qemu_fflush(f);
142 return;
143 }
144
Jan Kiszka13f0b672009-11-30 18:21:21 +0100145 qemu_put_buffer(f, blk->buf, BLOCK_SIZE);
146}
147
Jan Kiszka25f23642009-11-30 18:21:21 +0100148int blk_mig_active(void)
149{
150 return !QSIMPLEQ_EMPTY(&block_mig_state.bmds_list);
151}
152
153uint64_t blk_mig_bytes_transferred(void)
154{
155 BlkMigDevState *bmds;
156 uint64_t sum = 0;
157
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100158 blk_mig_lock();
Jan Kiszka25f23642009-11-30 18:21:21 +0100159 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
160 sum += bmds->completed_sectors;
161 }
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100162 blk_mig_unlock();
Jan Kiszka25f23642009-11-30 18:21:21 +0100163 return sum << BDRV_SECTOR_BITS;
164}
165
166uint64_t blk_mig_bytes_remaining(void)
167{
168 return blk_mig_bytes_total() - blk_mig_bytes_transferred();
169}
170
171uint64_t blk_mig_bytes_total(void)
172{
173 BlkMigDevState *bmds;
174 uint64_t sum = 0;
175
176 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
177 sum += bmds->total_sectors;
178 }
179 return sum << BDRV_SECTOR_BITS;
180}
181
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100182
183/* Called with migration lock held. */
184
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200185static int bmds_aio_inflight(BlkMigDevState *bmds, int64_t sector)
186{
187 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
188
Marcelo Tosatti62155e22010-11-12 16:07:50 -0200189 if ((sector << BDRV_SECTOR_BITS) < bdrv_getlength(bmds->bs)) {
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200190 return !!(bmds->aio_bitmap[chunk / (sizeof(unsigned long) * 8)] &
191 (1UL << (chunk % (sizeof(unsigned long) * 8))));
192 } else {
193 return 0;
194 }
195}
196
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100197/* Called with migration lock held. */
198
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200199static void bmds_set_aio_inflight(BlkMigDevState *bmds, int64_t sector_num,
200 int nb_sectors, int set)
201{
202 int64_t start, end;
203 unsigned long val, idx, bit;
204
205 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
206 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
207
208 for (; start <= end; start++) {
209 idx = start / (sizeof(unsigned long) * 8);
210 bit = start % (sizeof(unsigned long) * 8);
211 val = bmds->aio_bitmap[idx];
212 if (set) {
Marcelo Tosatti62155e22010-11-12 16:07:50 -0200213 val |= 1UL << bit;
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200214 } else {
Marcelo Tosatti62155e22010-11-12 16:07:50 -0200215 val &= ~(1UL << bit);
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200216 }
217 bmds->aio_bitmap[idx] = val;
218 }
219}
220
221static void alloc_aio_bitmap(BlkMigDevState *bmds)
222{
223 BlockDriverState *bs = bmds->bs;
224 int64_t bitmap_size;
225
226 bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
227 BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
228 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
229
Anthony Liguori7267c092011-08-20 22:09:37 -0500230 bmds->aio_bitmap = g_malloc0(bitmap_size);
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200231}
232
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100233/* Never hold migration lock when yielding to the main loop! */
234
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200235static void blk_mig_read_cb(void *opaque, int ret)
236{
237 BlkMigBlock *blk = opaque;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100238
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100239 blk_mig_lock();
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200240 blk->ret = ret;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100241
Jan Kiszka5e5328b2009-11-30 18:21:20 +0100242 QSIMPLEQ_INSERT_TAIL(&block_mig_state.blk_list, blk, entry);
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200243 bmds_set_aio_inflight(blk->bmds, blk->sector, blk->nr_sectors, 0);
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100244
Jan Kiszkad11ecd32009-11-30 18:21:20 +0100245 block_mig_state.submitted--;
246 block_mig_state.read_done++;
247 assert(block_mig_state.submitted >= 0);
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100248 blk_mig_unlock();
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200249}
250
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100251/* Called with no lock taken. */
252
Luiz Capitulino539de122011-12-05 14:06:56 -0200253static int mig_save_device_bulk(QEMUFile *f, BlkMigDevState *bmds)
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100254{
Jan Kiszka57cce122009-11-30 18:21:20 +0100255 int64_t total_sectors = bmds->total_sectors;
256 int64_t cur_sector = bmds->cur_sector;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200257 BlockDriverState *bs = bmds->bs;
Jan Kiszka57cce122009-11-30 18:21:20 +0100258 BlkMigBlock *blk;
Jan Kiszka13f0b672009-11-30 18:21:21 +0100259 int nr_sectors;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100260
261 if (bmds->shared_base) {
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100262 qemu_mutex_lock_iothread();
Jan Kiszkab1d10852009-11-30 18:21:20 +0100263 while (cur_sector < total_sectors &&
Jan Kiszka57cce122009-11-30 18:21:20 +0100264 !bdrv_is_allocated(bs, cur_sector, MAX_IS_ALLOCATED_SEARCH,
265 &nr_sectors)) {
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200266 cur_sector += nr_sectors;
267 }
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100268 qemu_mutex_unlock_iothread();
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200269 }
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100270
271 if (cur_sector >= total_sectors) {
Jan Kiszka82801d82009-11-30 18:21:21 +0100272 bmds->cur_sector = bmds->completed_sectors = total_sectors;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200273 return 1;
274 }
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100275
Jan Kiszka82801d82009-11-30 18:21:21 +0100276 bmds->completed_sectors = cur_sector;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100277
Jan Kiszka6ea44302009-11-30 18:21:19 +0100278 cur_sector &= ~((int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK - 1);
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100279
Jan Kiszka6ea44302009-11-30 18:21:19 +0100280 /* we are going to transfer a full block even if it is not allocated */
281 nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100282
Jan Kiszka6ea44302009-11-30 18:21:19 +0100283 if (total_sectors - cur_sector < BDRV_SECTORS_PER_DIRTY_CHUNK) {
Jan Kiszka57cce122009-11-30 18:21:20 +0100284 nr_sectors = total_sectors - cur_sector;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200285 }
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100286
Anthony Liguori7267c092011-08-20 22:09:37 -0500287 blk = g_malloc(sizeof(BlkMigBlock));
288 blk->buf = g_malloc(BLOCK_SIZE);
Jan Kiszka13f0b672009-11-30 18:21:21 +0100289 blk->bmds = bmds;
290 blk->sector = cur_sector;
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200291 blk->nr_sectors = nr_sectors;
Jan Kiszka13f0b672009-11-30 18:21:21 +0100292
Liran Schoure970ec02010-01-26 10:31:45 +0200293 blk->iov.iov_base = blk->buf;
294 blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE;
295 qemu_iovec_init_external(&blk->qiov, &blk->iov, 1);
Jan Kiszka57cce122009-11-30 18:21:20 +0100296
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100297 blk_mig_lock();
Paolo Bonzini13197e32013-02-22 17:36:23 +0100298 block_mig_state.submitted++;
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100299 blk_mig_unlock();
Paolo Bonzini13197e32013-02-22 17:36:23 +0100300
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100301 qemu_mutex_lock_iothread();
Liran Schoure970ec02010-01-26 10:31:45 +0200302 blk->aiocb = bdrv_aio_readv(bs, cur_sector, &blk->qiov,
303 nr_sectors, blk_mig_read_cb, blk);
Liran Schourd76cac72010-01-26 14:04:11 +0200304
Jan Kiszka13f0b672009-11-30 18:21:21 +0100305 bdrv_reset_dirty(bs, cur_sector, nr_sectors);
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100306 qemu_mutex_unlock_iothread();
Jan Kiszka13f0b672009-11-30 18:21:21 +0100307
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100308 bmds->cur_sector = cur_sector + nr_sectors;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200309 return (bmds->cur_sector >= total_sectors);
310}
311
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100312/* Called with iothread lock taken. */
313
Fam Zhengb8afb522014-04-16 09:34:30 +0800314static int set_dirty_tracking(void)
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200315{
316 BlkMigDevState *bmds;
Fam Zhengb8afb522014-04-16 09:34:30 +0800317 int ret;
Jan Kiszka5e5328b2009-11-30 18:21:20 +0100318
319 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
Fam Zhengb8afb522014-04-16 09:34:30 +0800320 bmds->dirty_bitmap = bdrv_create_dirty_bitmap(bmds->bs, BLOCK_SIZE,
321 NULL);
322 if (!bmds->dirty_bitmap) {
323 ret = -errno;
324 goto fail;
325 }
Fam Zhenge4654d22013-11-13 18:29:43 +0800326 }
Fam Zhengb8afb522014-04-16 09:34:30 +0800327 return 0;
328
329fail:
330 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
331 if (bmds->dirty_bitmap) {
332 bdrv_release_dirty_bitmap(bmds->bs, bmds->dirty_bitmap);
333 }
334 }
335 return ret;
Fam Zhenge4654d22013-11-13 18:29:43 +0800336}
337
338static void unset_dirty_tracking(void)
339{
340 BlkMigDevState *bmds;
341
342 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
343 bdrv_release_dirty_bitmap(bmds->bs, bmds->dirty_bitmap);
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200344 }
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200345}
346
Stefan Hajnoczib66460e2010-04-09 15:22:13 +0100347static void init_blk_migration_it(void *opaque, BlockDriverState *bs)
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200348{
Jan Kiszka5e5328b2009-11-30 18:21:20 +0100349 BlkMigDevState *bmds;
Jan Kiszka792773b2009-11-30 20:34:55 +0100350 int64_t sectors;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100351
Markus Armbrusterd2466732010-06-28 10:45:02 +0200352 if (!bdrv_is_read_only(bs)) {
Stefan Hajnoczib66460e2010-04-09 15:22:13 +0100353 sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
Shahar Havivi31f54f22010-07-10 18:59:06 +0300354 if (sectors <= 0) {
Stefan Hajnoczib66460e2010-04-09 15:22:13 +0100355 return;
356 }
357
Anthony Liguori7267c092011-08-20 22:09:37 -0500358 bmds = g_malloc0(sizeof(BlkMigDevState));
Stefan Hajnoczib66460e2010-04-09 15:22:13 +0100359 bmds->bs = bs;
360 bmds->bulk_completed = 0;
361 bmds->total_sectors = sectors;
362 bmds->completed_sectors = 0;
363 bmds->shared_base = block_mig_state.shared_base;
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200364 alloc_aio_bitmap(bmds);
Fam Zheng3718d8a2014-05-23 21:29:43 +0800365 error_setg(&bmds->blocker, "block device is in use by migration");
366 bdrv_op_block_all(bs, bmds->blocker);
Fam Zheng8442cfd2013-08-23 09:14:48 +0800367 bdrv_ref(bs);
Stefan Hajnoczib66460e2010-04-09 15:22:13 +0100368
369 block_mig_state.total_sector_sum += sectors;
370
371 if (bmds->shared_base) {
Luiz Capitulino539de122011-12-05 14:06:56 -0200372 DPRINTF("Start migration for %s with shared base image\n",
373 bs->device_name);
Stefan Hajnoczib66460e2010-04-09 15:22:13 +0100374 } else {
Luiz Capitulino539de122011-12-05 14:06:56 -0200375 DPRINTF("Start full migration for %s\n", bs->device_name);
Stefan Hajnoczib66460e2010-04-09 15:22:13 +0100376 }
377
378 QSIMPLEQ_INSERT_TAIL(&block_mig_state.bmds_list, bmds, entry);
379 }
380}
381
Luiz Capitulino539de122011-12-05 14:06:56 -0200382static void init_blk_migration(QEMUFile *f)
Stefan Hajnoczib66460e2010-04-09 15:22:13 +0100383{
Jan Kiszka69d63a92009-11-30 18:21:20 +0100384 block_mig_state.submitted = 0;
385 block_mig_state.read_done = 0;
386 block_mig_state.transferred = 0;
Jan Kiszka82801d82009-11-30 18:21:21 +0100387 block_mig_state.total_sector_sum = 0;
Jan Kiszka01e61e22009-12-01 15:20:17 +0100388 block_mig_state.prev_progress = -1;
Liran Schoure970ec02010-01-26 10:31:45 +0200389 block_mig_state.bulk_completed = 0;
Peter Lieven323004a2013-07-18 09:48:50 +0200390 block_mig_state.zero_blocks = migrate_zero_blocks();
Jan Kiszka69d63a92009-11-30 18:21:20 +0100391
Luiz Capitulino539de122011-12-05 14:06:56 -0200392 bdrv_iterate(init_blk_migration_it, NULL);
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200393}
394
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100395/* Called with no lock taken. */
396
Luiz Capitulino539de122011-12-05 14:06:56 -0200397static int blk_mig_save_bulked_block(QEMUFile *f)
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200398{
Jan Kiszka82801d82009-11-30 18:21:21 +0100399 int64_t completed_sector_sum = 0;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200400 BlkMigDevState *bmds;
Jan Kiszka01e61e22009-12-01 15:20:17 +0100401 int progress;
Jan Kiszka82801d82009-11-30 18:21:21 +0100402 int ret = 0;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200403
Jan Kiszka5e5328b2009-11-30 18:21:20 +0100404 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100405 if (bmds->bulk_completed == 0) {
Luiz Capitulino539de122011-12-05 14:06:56 -0200406 if (mig_save_device_bulk(f, bmds) == 1) {
Jan Kiszka57cce122009-11-30 18:21:20 +0100407 /* completed bulk section for this device */
408 bmds->bulk_completed = 1;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200409 }
Jan Kiszka82801d82009-11-30 18:21:21 +0100410 completed_sector_sum += bmds->completed_sectors;
411 ret = 1;
412 break;
413 } else {
414 completed_sector_sum += bmds->completed_sectors;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200415 }
416 }
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100417
Pierre Riteau8b6b2af2011-01-12 14:41:00 +0100418 if (block_mig_state.total_sector_sum != 0) {
419 progress = completed_sector_sum * 100 /
420 block_mig_state.total_sector_sum;
421 } else {
422 progress = 100;
423 }
Jan Kiszka01e61e22009-12-01 15:20:17 +0100424 if (progress != block_mig_state.prev_progress) {
425 block_mig_state.prev_progress = progress;
426 qemu_put_be64(f, (progress << BDRV_SECTOR_BITS)
427 | BLK_MIG_FLAG_PROGRESS);
Luiz Capitulino539de122011-12-05 14:06:56 -0200428 DPRINTF("Completed %d %%\r", progress);
Jan Kiszka82801d82009-11-30 18:21:21 +0100429 }
430
431 return ret;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200432}
433
Liran Schourd76cac72010-01-26 14:04:11 +0200434static void blk_mig_reset_dirty_cursor(void)
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200435{
436 BlkMigDevState *bmds;
Jan Kiszka575a58d2009-11-30 18:21:20 +0100437
Jan Kiszka5e5328b2009-11-30 18:21:20 +0100438 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
Liran Schourd76cac72010-01-26 14:04:11 +0200439 bmds->cur_dirty = 0;
440 }
441}
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100442
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100443/* Called with iothread lock taken. */
444
Luiz Capitulino539de122011-12-05 14:06:56 -0200445static int mig_save_device_dirty(QEMUFile *f, BlkMigDevState *bmds,
446 int is_async)
Liran Schourd76cac72010-01-26 14:04:11 +0200447{
448 BlkMigBlock *blk;
449 int64_t total_sectors = bmds->total_sectors;
450 int64_t sector;
451 int nr_sectors;
Juan Quinteladcd1d222011-09-21 23:01:54 +0200452 int ret = -EIO;
Liran Schourd76cac72010-01-26 14:04:11 +0200453
454 for (sector = bmds->cur_dirty; sector < bmds->total_sectors;) {
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100455 blk_mig_lock();
Marcelo Tosatti62155e22010-11-12 16:07:50 -0200456 if (bmds_aio_inflight(bmds, sector)) {
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100457 blk_mig_unlock();
Stefan Hajnoczi922453b2011-11-30 12:23:43 +0000458 bdrv_drain_all();
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100459 } else {
460 blk_mig_unlock();
Marcelo Tosatti62155e22010-11-12 16:07:50 -0200461 }
Fam Zhenge4654d22013-11-13 18:29:43 +0800462 if (bdrv_get_dirty(bmds->bs, bmds->dirty_bitmap, sector)) {
Liran Schourd76cac72010-01-26 14:04:11 +0200463
464 if (total_sectors - sector < BDRV_SECTORS_PER_DIRTY_CHUNK) {
465 nr_sectors = total_sectors - sector;
466 } else {
467 nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100468 }
Anthony Liguori7267c092011-08-20 22:09:37 -0500469 blk = g_malloc(sizeof(BlkMigBlock));
470 blk->buf = g_malloc(BLOCK_SIZE);
Liran Schourd76cac72010-01-26 14:04:11 +0200471 blk->bmds = bmds;
472 blk->sector = sector;
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200473 blk->nr_sectors = nr_sectors;
Liran Schourd76cac72010-01-26 14:04:11 +0200474
Liran Schour889ae392010-01-26 10:31:49 +0200475 if (is_async) {
Liran Schourd76cac72010-01-26 14:04:11 +0200476 blk->iov.iov_base = blk->buf;
477 blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE;
478 qemu_iovec_init_external(&blk->qiov, &blk->iov, 1);
479
480 blk->aiocb = bdrv_aio_readv(bmds->bs, sector, &blk->qiov,
481 nr_sectors, blk_mig_read_cb, blk);
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100482
483 blk_mig_lock();
Liran Schourd76cac72010-01-26 14:04:11 +0200484 block_mig_state.submitted++;
Marcelo Tosatti33656af2010-11-08 17:02:56 -0200485 bmds_set_aio_inflight(bmds, sector, nr_sectors, 1);
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100486 blk_mig_unlock();
Liran Schourd76cac72010-01-26 14:04:11 +0200487 } else {
Juan Quinteladcd1d222011-09-21 23:01:54 +0200488 ret = bdrv_read(bmds->bs, sector, blk->buf, nr_sectors);
489 if (ret < 0) {
Liran Schourd76cac72010-01-26 14:04:11 +0200490 goto error;
491 }
492 blk_send(f, blk);
493
Anthony Liguori7267c092011-08-20 22:09:37 -0500494 g_free(blk->buf);
495 g_free(blk);
Liran Schourd76cac72010-01-26 14:04:11 +0200496 }
497
498 bdrv_reset_dirty(bmds->bs, sector, nr_sectors);
499 break;
500 }
501 sector += BDRV_SECTORS_PER_DIRTY_CHUNK;
502 bmds->cur_dirty = sector;
503 }
504
505 return (bmds->cur_dirty >= bmds->total_sectors);
506
Liran Schour889ae392010-01-26 10:31:49 +0200507error:
Luiz Capitulino539de122011-12-05 14:06:56 -0200508 DPRINTF("Error reading sector %" PRId64 "\n", sector);
Anthony Liguori7267c092011-08-20 22:09:37 -0500509 g_free(blk->buf);
510 g_free(blk);
Juan Quintela43be3a22012-08-29 21:59:22 +0200511 return ret;
Liran Schourd76cac72010-01-26 14:04:11 +0200512}
513
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100514/* Called with iothread lock taken.
515 *
516 * return value:
Juan Quintelaceb2bd02012-08-29 21:37:14 +0200517 * 0: too much data for max_downtime
518 * 1: few enough data for max_downtime
519*/
Luiz Capitulino539de122011-12-05 14:06:56 -0200520static int blk_mig_save_dirty_block(QEMUFile *f, int is_async)
Liran Schourd76cac72010-01-26 14:04:11 +0200521{
522 BlkMigDevState *bmds;
Juan Quintelaceb2bd02012-08-29 21:37:14 +0200523 int ret = 1;
Liran Schourd76cac72010-01-26 14:04:11 +0200524
525 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
Juan Quintelaceb2bd02012-08-29 21:37:14 +0200526 ret = mig_save_device_dirty(f, bmds, is_async);
Juan Quintela43be3a22012-08-29 21:59:22 +0200527 if (ret <= 0) {
Liran Schourd76cac72010-01-26 14:04:11 +0200528 break;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200529 }
530 }
Jan Kiszka575a58d2009-11-30 18:21:20 +0100531
Liran Schourd76cac72010-01-26 14:04:11 +0200532 return ret;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200533}
534
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100535/* Called with no locks taken. */
536
Juan Quintela59feec42012-08-29 20:17:13 +0200537static int flush_blks(QEMUFile *f)
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200538{
Jan Kiszka5e5328b2009-11-30 18:21:20 +0100539 BlkMigBlock *blk;
Juan Quintela59feec42012-08-29 20:17:13 +0200540 int ret = 0;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100541
malcd0f2c4c2010-02-07 02:03:50 +0300542 DPRINTF("%s Enter submitted %d read_done %d transferred %d\n",
Jan Kiszkad11ecd32009-11-30 18:21:20 +0100543 __FUNCTION__, block_mig_state.submitted, block_mig_state.read_done,
544 block_mig_state.transferred);
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100545
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100546 blk_mig_lock();
Jan Kiszka5e5328b2009-11-30 18:21:20 +0100547 while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) {
548 if (qemu_file_rate_limit(f)) {
549 break;
550 }
Jan Kiszka4b640362009-11-30 18:21:21 +0100551 if (blk->ret < 0) {
Juan Quintela59feec42012-08-29 20:17:13 +0200552 ret = blk->ret;
Jan Kiszka4b640362009-11-30 18:21:21 +0100553 break;
554 }
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100555
Jan Kiszka5e5328b2009-11-30 18:21:20 +0100556 QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry);
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100557 blk_mig_unlock();
Paolo Bonzini13197e32013-02-22 17:36:23 +0100558 blk_send(f, blk);
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100559 blk_mig_lock();
Paolo Bonzini13197e32013-02-22 17:36:23 +0100560
Anthony Liguori7267c092011-08-20 22:09:37 -0500561 g_free(blk->buf);
562 g_free(blk);
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100563
Jan Kiszkad11ecd32009-11-30 18:21:20 +0100564 block_mig_state.read_done--;
565 block_mig_state.transferred++;
566 assert(block_mig_state.read_done >= 0);
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200567 }
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100568 blk_mig_unlock();
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200569
malcd0f2c4c2010-02-07 02:03:50 +0300570 DPRINTF("%s Exit submitted %d read_done %d transferred %d\n", __FUNCTION__,
Jan Kiszkad11ecd32009-11-30 18:21:20 +0100571 block_mig_state.submitted, block_mig_state.read_done,
572 block_mig_state.transferred);
Juan Quintela59feec42012-08-29 20:17:13 +0200573 return ret;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200574}
575
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100576/* Called with iothread lock taken. */
577
Liran Schour889ae392010-01-26 10:31:49 +0200578static int64_t get_remaining_dirty(void)
579{
580 BlkMigDevState *bmds;
581 int64_t dirty = 0;
582
583 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
Fam Zhenge4654d22013-11-13 18:29:43 +0800584 dirty += bdrv_get_dirty_count(bmds->bs, bmds->dirty_bitmap);
Liran Schour889ae392010-01-26 10:31:49 +0200585 }
586
Paolo Bonziniacc906c2013-01-21 17:09:44 +0100587 return dirty << BDRV_SECTOR_BITS;
Liran Schour889ae392010-01-26 10:31:49 +0200588}
589
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100590/* Called with iothread lock taken. */
591
Luiz Capitulino539de122011-12-05 14:06:56 -0200592static void blk_mig_cleanup(void)
Jan Kiszka4ec7fcc2009-11-30 18:21:21 +0100593{
Jan Kiszka82801d82009-11-30 18:21:21 +0100594 BlkMigDevState *bmds;
595 BlkMigBlock *blk;
Jan Kiszka4ec7fcc2009-11-30 18:21:21 +0100596
Kevin Wolf946d58b2012-09-25 15:47:36 +0200597 bdrv_drain_all();
598
Fam Zhenge4654d22013-11-13 18:29:43 +0800599 unset_dirty_tracking();
Marcelo Tosatti8f794c52011-01-26 12:12:31 -0200600
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100601 blk_mig_lock();
Jan Kiszka82801d82009-11-30 18:21:21 +0100602 while ((bmds = QSIMPLEQ_FIRST(&block_mig_state.bmds_list)) != NULL) {
603 QSIMPLEQ_REMOVE_HEAD(&block_mig_state.bmds_list, entry);
Fam Zheng3718d8a2014-05-23 21:29:43 +0800604 bdrv_op_unblock_all(bmds->bs, bmds->blocker);
605 error_free(bmds->blocker);
Fam Zheng8442cfd2013-08-23 09:14:48 +0800606 bdrv_unref(bmds->bs);
Anthony Liguori7267c092011-08-20 22:09:37 -0500607 g_free(bmds->aio_bitmap);
608 g_free(bmds);
Jan Kiszka4ec7fcc2009-11-30 18:21:21 +0100609 }
610
Jan Kiszka82801d82009-11-30 18:21:21 +0100611 while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) {
612 QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry);
Anthony Liguori7267c092011-08-20 22:09:37 -0500613 g_free(blk->buf);
614 g_free(blk);
Jan Kiszka4ec7fcc2009-11-30 18:21:21 +0100615 }
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100616 blk_mig_unlock();
Jan Kiszka4ec7fcc2009-11-30 18:21:21 +0100617}
618
Juan Quintela9b5bfab2012-06-26 19:26:41 +0200619static void block_migration_cancel(void *opaque)
620{
621 blk_mig_cleanup();
622}
623
Juan Quintelad1315aa2012-06-28 15:11:57 +0200624static int block_save_setup(QEMUFile *f, void *opaque)
625{
626 int ret;
627
628 DPRINTF("Enter save live setup submitted %d transferred %d\n",
629 block_mig_state.submitted, block_mig_state.transferred);
630
Paolo Bonzini9b095032013-02-22 17:36:28 +0100631 qemu_mutex_lock_iothread();
chai wen1ac362c2014-06-04 11:47:37 +0800632 init_blk_migration(f);
Juan Quintelad1315aa2012-06-28 15:11:57 +0200633
634 /* start track dirty blocks */
Fam Zhengb8afb522014-04-16 09:34:30 +0800635 ret = set_dirty_tracking();
636
637 if (ret) {
638 qemu_mutex_unlock_iothread();
639 return ret;
640 }
641
Paolo Bonzini9b095032013-02-22 17:36:28 +0100642 qemu_mutex_unlock_iothread();
Juan Quintelad1315aa2012-06-28 15:11:57 +0200643
Juan Quintela59feec42012-08-29 20:17:13 +0200644 ret = flush_blks(f);
Juan Quintelad1315aa2012-06-28 15:11:57 +0200645 blk_mig_reset_dirty_cursor();
Juan Quintelad1315aa2012-06-28 15:11:57 +0200646 qemu_put_be64(f, BLK_MIG_FLAG_EOS);
647
Paolo Bonzinid418cf52013-02-22 17:36:11 +0100648 return ret;
Juan Quintelad1315aa2012-06-28 15:11:57 +0200649}
650
Juan Quintela16310a32012-06-28 15:31:37 +0200651static int block_save_iterate(QEMUFile *f, void *opaque)
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200652{
Juan Quintela29757252011-10-19 15:22:18 +0200653 int ret;
Stefan Hajnoczi6aaa9da2013-02-12 10:37:15 +0100654 int64_t last_ftell = qemu_ftell(f);
Juan Quintela29757252011-10-19 15:22:18 +0200655
Juan Quintela16310a32012-06-28 15:31:37 +0200656 DPRINTF("Enter save live iterate submitted %d transferred %d\n",
657 block_mig_state.submitted, block_mig_state.transferred);
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100658
Juan Quintela59feec42012-08-29 20:17:13 +0200659 ret = flush_blks(f);
Juan Quintela29757252011-10-19 15:22:18 +0200660 if (ret) {
Juan Quintela29757252011-10-19 15:22:18 +0200661 return ret;
Jan Kiszka4b640362009-11-30 18:21:21 +0100662 }
663
Liran Schourd76cac72010-01-26 14:04:11 +0200664 blk_mig_reset_dirty_cursor();
665
Juan Quintela16310a32012-06-28 15:31:37 +0200666 /* control the rate of transfer */
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100667 blk_mig_lock();
Juan Quintela16310a32012-06-28 15:31:37 +0200668 while ((block_mig_state.submitted +
669 block_mig_state.read_done) * BLOCK_SIZE <
670 qemu_file_get_rate_limit(f)) {
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100671 blk_mig_unlock();
Juan Quintela16310a32012-06-28 15:31:37 +0200672 if (block_mig_state.bulk_completed == 0) {
673 /* first finish the bulk phase */
674 if (blk_mig_save_bulked_block(f) == 0) {
675 /* finished saving bulk on all devices */
676 block_mig_state.bulk_completed = 1;
Liran Schourd76cac72010-01-26 14:04:11 +0200677 }
Paolo Bonzini13197e32013-02-22 17:36:23 +0100678 ret = 0;
Juan Quintela16310a32012-06-28 15:31:37 +0200679 } else {
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100680 /* Always called with iothread lock taken for
681 * simplicity, block_save_complete also calls it.
682 */
683 qemu_mutex_lock_iothread();
Juan Quintela43be3a22012-08-29 21:59:22 +0200684 ret = blk_mig_save_dirty_block(f, 1);
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100685 qemu_mutex_unlock_iothread();
Paolo Bonzini13197e32013-02-22 17:36:23 +0100686 }
687 if (ret < 0) {
688 return ret;
689 }
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100690 blk_mig_lock();
Paolo Bonzini13197e32013-02-22 17:36:23 +0100691 if (ret != 0) {
692 /* no more dirty blocks */
693 break;
Liran Schourd76cac72010-01-26 14:04:11 +0200694 }
Jan Kiszka4b640362009-11-30 18:21:21 +0100695 }
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100696 blk_mig_unlock();
Jan Kiszka4b640362009-11-30 18:21:21 +0100697
Juan Quintela59feec42012-08-29 20:17:13 +0200698 ret = flush_blks(f);
Juan Quintela16310a32012-06-28 15:31:37 +0200699 if (ret) {
Juan Quintela16310a32012-06-28 15:31:37 +0200700 return ret;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200701 }
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100702
703 qemu_put_be64(f, BLK_MIG_FLAG_EOS);
Stefan Hajnoczi6aaa9da2013-02-12 10:37:15 +0100704 return qemu_ftell(f) - last_ftell;
Juan Quintela16310a32012-06-28 15:31:37 +0200705}
706
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100707/* Called with iothread lock taken. */
708
Juan Quintela16310a32012-06-28 15:31:37 +0200709static int block_save_complete(QEMUFile *f, void *opaque)
710{
711 int ret;
712
713 DPRINTF("Enter save live complete submitted %d transferred %d\n",
714 block_mig_state.submitted, block_mig_state.transferred);
715
Juan Quintela59feec42012-08-29 20:17:13 +0200716 ret = flush_blks(f);
Juan Quintela16310a32012-06-28 15:31:37 +0200717 if (ret) {
Juan Quintela16310a32012-06-28 15:31:37 +0200718 return ret;
719 }
720
721 blk_mig_reset_dirty_cursor();
722
723 /* we know for sure that save bulk is completed and
724 all async read completed */
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100725 blk_mig_lock();
Juan Quintela16310a32012-06-28 15:31:37 +0200726 assert(block_mig_state.submitted == 0);
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100727 blk_mig_unlock();
Juan Quintela16310a32012-06-28 15:31:37 +0200728
Juan Quintela43be3a22012-08-29 21:59:22 +0200729 do {
730 ret = blk_mig_save_dirty_block(f, 0);
Paolo Bonzinid418cf52013-02-22 17:36:11 +0100731 if (ret < 0) {
732 return ret;
733 }
Juan Quintela43be3a22012-08-29 21:59:22 +0200734 } while (ret == 0);
735
Juan Quintela43be3a22012-08-29 21:59:22 +0200736 /* report completion */
737 qemu_put_be64(f, (100 << BDRV_SECTOR_BITS) | BLK_MIG_FLAG_PROGRESS);
Juan Quintela16310a32012-06-28 15:31:37 +0200738
739 DPRINTF("Block migration completed\n");
740
741 qemu_put_be64(f, BLK_MIG_FLAG_EOS);
742
Paolo Bonzinid418cf52013-02-22 17:36:11 +0100743 blk_mig_cleanup();
Juan Quintela16310a32012-06-28 15:31:37 +0200744 return 0;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200745}
746
Juan Quintelae4ed1542012-09-21 11:18:18 +0200747static uint64_t block_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
748{
Stefan Hajnoczi6aaa9da2013-02-12 10:37:15 +0100749 /* Estimate pending number of bytes to send */
Paolo Bonzini13197e32013-02-22 17:36:23 +0100750 uint64_t pending;
751
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100752 qemu_mutex_lock_iothread();
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100753 blk_mig_lock();
Paolo Bonzini13197e32013-02-22 17:36:23 +0100754 pending = get_remaining_dirty() +
Stefan Hajnoczi6aaa9da2013-02-12 10:37:15 +0100755 block_mig_state.submitted * BLOCK_SIZE +
756 block_mig_state.read_done * BLOCK_SIZE;
Juan Quintelae4ed1542012-09-21 11:18:18 +0200757
Stefan Hajnoczi6aaa9da2013-02-12 10:37:15 +0100758 /* Report at least one block pending during bulk phase */
759 if (pending == 0 && !block_mig_state.bulk_completed) {
760 pending = BLOCK_SIZE;
761 }
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100762 blk_mig_unlock();
Paolo Bonzini32c835b2013-02-22 17:36:27 +0100763 qemu_mutex_unlock_iothread();
Juan Quintelae4ed1542012-09-21 11:18:18 +0200764
Stefan Hajnoczi6aaa9da2013-02-12 10:37:15 +0100765 DPRINTF("Enter save live pending %" PRIu64 "\n", pending);
766 return pending;
Juan Quintelae4ed1542012-09-21 11:18:18 +0200767}
768
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200769static int block_load(QEMUFile *f, void *opaque, int version_id)
770{
Jan Kiszka01e61e22009-12-01 15:20:17 +0100771 static int banner_printed;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200772 int len, flags;
773 char device_name[256];
774 int64_t addr;
Pierre Riteau77358b52011-01-21 12:42:30 +0100775 BlockDriverState *bs, *bs_prev = NULL;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200776 uint8_t *buf;
Pierre Riteau77358b52011-01-21 12:42:30 +0100777 int64_t total_sectors = 0;
778 int nr_sectors;
Juan Quintela42802d42011-10-05 01:14:46 +0200779 int ret;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100780
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200781 do {
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200782 addr = qemu_get_be64(f);
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100783
Jan Kiszka6ea44302009-11-30 18:21:19 +0100784 flags = addr & ~BDRV_SECTOR_MASK;
785 addr >>= BDRV_SECTOR_BITS;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100786
787 if (flags & BLK_MIG_FLAG_DEVICE_BLOCK) {
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200788 /* get device name */
789 len = qemu_get_byte(f);
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200790 qemu_get_buffer(f, (uint8_t *)device_name, len);
791 device_name[len] = '\0';
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100792
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200793 bs = bdrv_find(device_name);
Jan Kiszka4b640362009-11-30 18:21:21 +0100794 if (!bs) {
795 fprintf(stderr, "Error unknown block device %s\n",
796 device_name);
797 return -EINVAL;
798 }
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100799
Pierre Riteau77358b52011-01-21 12:42:30 +0100800 if (bs != bs_prev) {
801 bs_prev = bs;
802 total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
803 if (total_sectors <= 0) {
Markus Armbruster6daf1942011-06-22 14:03:54 +0200804 error_report("Error getting length of block device %s",
Pierre Riteau77358b52011-01-21 12:42:30 +0100805 device_name);
806 return -EINVAL;
807 }
808 }
809
810 if (total_sectors - addr < BDRV_SECTORS_PER_DIRTY_CHUNK) {
811 nr_sectors = total_sectors - addr;
812 } else {
813 nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
814 }
815
Peter Lieven323004a2013-07-18 09:48:50 +0200816 if (flags & BLK_MIG_FLAG_ZERO_BLOCK) {
Peter Lievend32f35c2013-10-24 12:06:52 +0200817 ret = bdrv_write_zeroes(bs, addr, nr_sectors,
818 BDRV_REQ_MAY_UNMAP);
Peter Lieven323004a2013-07-18 09:48:50 +0200819 } else {
820 buf = g_malloc(BLOCK_SIZE);
821 qemu_get_buffer(f, buf, BLOCK_SIZE);
822 ret = bdrv_write(bs, addr, buf, nr_sectors);
823 g_free(buf);
824 }
Jan Kiszka575a58d2009-11-30 18:21:20 +0100825
Yoshiaki Tamurab02bea32010-07-20 18:19:00 +0900826 if (ret < 0) {
827 return ret;
828 }
Jan Kiszka01e61e22009-12-01 15:20:17 +0100829 } else if (flags & BLK_MIG_FLAG_PROGRESS) {
830 if (!banner_printed) {
831 printf("Receiving block device images\n");
832 banner_printed = 1;
833 }
834 printf("Completed %d %%%c", (int)addr,
835 (addr == 100) ? '\n' : '\r');
836 fflush(stdout);
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100837 } else if (!(flags & BLK_MIG_FLAG_EOS)) {
Stefan Hajnoczid5f1f282013-02-10 23:12:44 +0100838 fprintf(stderr, "Unknown block migration flags: %#x\n", flags);
Jan Kiszka4b640362009-11-30 18:21:21 +0100839 return -EINVAL;
840 }
Juan Quintela42802d42011-10-05 01:14:46 +0200841 ret = qemu_file_get_error(f);
842 if (ret != 0) {
843 return ret;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200844 }
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100845 } while (!(flags & BLK_MIG_FLAG_EOS));
846
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200847 return 0;
848}
849
Isaku Yamahata6607ae22012-06-19 18:43:09 +0300850static void block_set_params(const MigrationParams *params, void *opaque)
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200851{
Isaku Yamahata6607ae22012-06-19 18:43:09 +0300852 block_mig_state.blk_enable = params->blk;
853 block_mig_state.shared_base = params->shared;
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100854
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200855 /* shared base means that blk_enable = 1 */
Isaku Yamahata6607ae22012-06-19 18:43:09 +0300856 block_mig_state.blk_enable |= params->shared;
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200857}
858
Juan Quintela6bd68782012-06-27 10:59:15 +0200859static bool block_is_active(void *opaque)
860{
861 return block_mig_state.blk_enable == 1;
862}
863
Juan Quintela7908c782012-06-26 18:46:10 +0200864SaveVMHandlers savevm_block_handlers = {
865 .set_params = block_set_params,
Juan Quintelad1315aa2012-06-28 15:11:57 +0200866 .save_live_setup = block_save_setup,
Juan Quintela16310a32012-06-28 15:31:37 +0200867 .save_live_iterate = block_save_iterate,
868 .save_live_complete = block_save_complete,
Juan Quintelae4ed1542012-09-21 11:18:18 +0200869 .save_live_pending = block_save_pending,
Juan Quintela7908c782012-06-26 18:46:10 +0200870 .load_state = block_load,
Juan Quintela9b5bfab2012-06-26 19:26:41 +0200871 .cancel = block_migration_cancel,
Juan Quintela6bd68782012-06-27 10:59:15 +0200872 .is_active = block_is_active,
Juan Quintela7908c782012-06-26 18:46:10 +0200873};
874
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200875void blk_mig_init(void)
Jan Kiszkaa55eb922009-11-30 18:21:19 +0100876{
Jan Kiszka5e5328b2009-11-30 18:21:20 +0100877 QSIMPLEQ_INIT(&block_mig_state.bmds_list);
878 QSIMPLEQ_INIT(&block_mig_state.blk_list);
Paolo Bonzini52e850d2013-02-22 17:36:25 +0100879 qemu_mutex_init(&block_mig_state.lock);
Jan Kiszka5e5328b2009-11-30 18:21:20 +0100880
Juan Quintela7908c782012-06-26 18:46:10 +0200881 register_savevm_live(NULL, "block", 0, 1, &savevm_block_handlers,
882 &block_mig_state);
lirans@il.ibm.comc163b5c2009-11-02 15:40:58 +0200883}