1893f7ebaSPaolo Bonzini /* 2893f7ebaSPaolo Bonzini * Image mirroring 3893f7ebaSPaolo Bonzini * 4893f7ebaSPaolo Bonzini * Copyright Red Hat, Inc. 2012 5893f7ebaSPaolo Bonzini * 6893f7ebaSPaolo Bonzini * Authors: 7893f7ebaSPaolo Bonzini * Paolo Bonzini <pbonzini@redhat.com> 8893f7ebaSPaolo Bonzini * 9893f7ebaSPaolo Bonzini * This work is licensed under the terms of the GNU LGPL, version 2 or later. 10893f7ebaSPaolo Bonzini * See the COPYING.LIB file in the top-level directory. 11893f7ebaSPaolo Bonzini * 12893f7ebaSPaolo Bonzini */ 13893f7ebaSPaolo Bonzini 1480c71a24SPeter Maydell #include "qemu/osdep.h" 15fd4a6493SKevin Wolf #include "qemu/cutils.h" 16893f7ebaSPaolo Bonzini #include "trace.h" 17c87621eaSJohn Snow #include "block/blockjob_int.h" 18737e150eSPaolo Bonzini #include "block/block_int.h" 19373340b2SMax Reitz #include "sysemu/block-backend.h" 20da34e65cSMarkus Armbruster #include "qapi/error.h" 21cc7a8ea7SMarkus Armbruster #include "qapi/qmp/qerror.h" 22893f7ebaSPaolo Bonzini #include "qemu/ratelimit.h" 23b812f671SPaolo Bonzini #include "qemu/bitmap.h" 24893f7ebaSPaolo Bonzini 25893f7ebaSPaolo Bonzini #define SLICE_TIME 100000000ULL /* ns */ 26402a4741SPaolo Bonzini #define MAX_IN_FLIGHT 16 270965a41eSVladimir Sementsov-Ogievskiy #define MAX_IO_SECTORS ((1 << 20) >> BDRV_SECTOR_BITS) /* 1 Mb */ 280965a41eSVladimir Sementsov-Ogievskiy #define DEFAULT_MIRROR_BUF_SIZE \ 290965a41eSVladimir Sementsov-Ogievskiy (MAX_IN_FLIGHT * MAX_IO_SECTORS * BDRV_SECTOR_SIZE) 30402a4741SPaolo Bonzini 31402a4741SPaolo Bonzini /* The mirroring buffer is a list of granularity-sized chunks. 32402a4741SPaolo Bonzini * Free chunks are organized in a list. 33402a4741SPaolo Bonzini */ 34402a4741SPaolo Bonzini typedef struct MirrorBuffer { 35402a4741SPaolo Bonzini QSIMPLEQ_ENTRY(MirrorBuffer) next; 36402a4741SPaolo Bonzini } MirrorBuffer; 37893f7ebaSPaolo Bonzini 38893f7ebaSPaolo Bonzini typedef struct MirrorBlockJob { 39893f7ebaSPaolo Bonzini BlockJob common; 40893f7ebaSPaolo Bonzini RateLimit limit; 41e253f4b8SKevin Wolf BlockBackend *target; 424ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs; 434ef85a9cSKevin Wolf BlockDriverState *source; 445bc361b8SFam Zheng BlockDriverState *base; 454ef85a9cSKevin Wolf 4609158f00SBenoît Canet /* The name of the graph node to replace */ 4709158f00SBenoît Canet char *replaces; 4809158f00SBenoît Canet /* The BDS to replace */ 4909158f00SBenoît Canet BlockDriverState *to_replace; 5009158f00SBenoît Canet /* Used to block operations on the drive-mirror-replace target */ 5109158f00SBenoît Canet Error *replace_blocker; 5203544a6eSFam Zheng bool is_none_mode; 53274fcceeSMax Reitz BlockMirrorBackingMode backing_mode; 54b952b558SPaolo Bonzini BlockdevOnError on_source_error, on_target_error; 55d63ffd87SPaolo Bonzini bool synced; 56d63ffd87SPaolo Bonzini bool should_complete; 57eee13dfeSPaolo Bonzini int64_t granularity; 58b812f671SPaolo Bonzini size_t buf_size; 59b21c7652SMax Reitz int64_t bdev_length; 60b812f671SPaolo Bonzini unsigned long *cow_bitmap; 61e4654d2dSFam Zheng BdrvDirtyBitmap *dirty_bitmap; 62dc162c8eSFam Zheng BdrvDirtyBitmapIter *dbi; 63893f7ebaSPaolo Bonzini uint8_t *buf; 64402a4741SPaolo Bonzini QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; 65402a4741SPaolo Bonzini int buf_free_count; 66bd48bde8SPaolo Bonzini 6749efb1f5SDenis V. Lunev uint64_t last_pause_ns; 68402a4741SPaolo Bonzini unsigned long *in_flight_bitmap; 69bd48bde8SPaolo Bonzini int in_flight; 70531509baSDenis V. Lunev int64_t sectors_in_flight; 71bd48bde8SPaolo Bonzini int ret; 720fc9f8eaSFam Zheng bool unmap; 73e424aff5SKevin Wolf bool waiting_for_io; 74e5b43573SFam Zheng int target_cluster_sectors; 75e5b43573SFam Zheng int max_iov; 7690ab48ebSAnton Nefedov bool initial_zeroing_ongoing; 77893f7ebaSPaolo Bonzini } MirrorBlockJob; 78893f7ebaSPaolo Bonzini 79bd48bde8SPaolo Bonzini typedef struct MirrorOp { 80bd48bde8SPaolo Bonzini MirrorBlockJob *s; 81bd48bde8SPaolo Bonzini QEMUIOVector qiov; 82bd48bde8SPaolo Bonzini int64_t sector_num; 83bd48bde8SPaolo Bonzini int nb_sectors; 84bd48bde8SPaolo Bonzini } MirrorOp; 85bd48bde8SPaolo Bonzini 86b952b558SPaolo Bonzini static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, 87b952b558SPaolo Bonzini int error) 88b952b558SPaolo Bonzini { 89b952b558SPaolo Bonzini s->synced = false; 90b952b558SPaolo Bonzini if (read) { 9181e254dcSKevin Wolf return block_job_error_action(&s->common, s->on_source_error, 9281e254dcSKevin Wolf true, error); 93b952b558SPaolo Bonzini } else { 9481e254dcSKevin Wolf return block_job_error_action(&s->common, s->on_target_error, 9581e254dcSKevin Wolf false, error); 96b952b558SPaolo Bonzini } 97b952b558SPaolo Bonzini } 98b952b558SPaolo Bonzini 99bd48bde8SPaolo Bonzini static void mirror_iteration_done(MirrorOp *op, int ret) 100bd48bde8SPaolo Bonzini { 101bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 102402a4741SPaolo Bonzini struct iovec *iov; 103bd48bde8SPaolo Bonzini int64_t chunk_num; 104402a4741SPaolo Bonzini int i, nb_chunks, sectors_per_chunk; 105bd48bde8SPaolo Bonzini 106bd48bde8SPaolo Bonzini trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret); 107bd48bde8SPaolo Bonzini 108bd48bde8SPaolo Bonzini s->in_flight--; 109b21c7652SMax Reitz s->sectors_in_flight -= op->nb_sectors; 110402a4741SPaolo Bonzini iov = op->qiov.iov; 111402a4741SPaolo Bonzini for (i = 0; i < op->qiov.niov; i++) { 112402a4741SPaolo Bonzini MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; 113402a4741SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); 114402a4741SPaolo Bonzini s->buf_free_count++; 115402a4741SPaolo Bonzini } 116402a4741SPaolo Bonzini 117bd48bde8SPaolo Bonzini sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; 118bd48bde8SPaolo Bonzini chunk_num = op->sector_num / sectors_per_chunk; 1194150ae60SFam Zheng nb_chunks = DIV_ROUND_UP(op->nb_sectors, sectors_per_chunk); 120402a4741SPaolo Bonzini bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); 121b21c7652SMax Reitz if (ret >= 0) { 122b21c7652SMax Reitz if (s->cow_bitmap) { 123bd48bde8SPaolo Bonzini bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); 124bd48bde8SPaolo Bonzini } 12590ab48ebSAnton Nefedov if (!s->initial_zeroing_ongoing) { 126b21c7652SMax Reitz s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE; 127b21c7652SMax Reitz } 12890ab48ebSAnton Nefedov } 1296df3bf8eSZhang Min qemu_iovec_destroy(&op->qiov); 130c84b3192SPaolo Bonzini g_free(op); 1317b770c72SStefan Hajnoczi 132e424aff5SKevin Wolf if (s->waiting_for_io) { 1330b8b8753SPaolo Bonzini qemu_coroutine_enter(s->common.co); 134bd48bde8SPaolo Bonzini } 1357b770c72SStefan Hajnoczi } 136bd48bde8SPaolo Bonzini 137bd48bde8SPaolo Bonzini static void mirror_write_complete(void *opaque, int ret) 138bd48bde8SPaolo Bonzini { 139bd48bde8SPaolo Bonzini MirrorOp *op = opaque; 140bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 141b9e413ddSPaolo Bonzini 142b9e413ddSPaolo Bonzini aio_context_acquire(blk_get_aio_context(s->common.blk)); 143bd48bde8SPaolo Bonzini if (ret < 0) { 144bd48bde8SPaolo Bonzini BlockErrorAction action; 145bd48bde8SPaolo Bonzini 14620dca810SJohn Snow bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); 147bd48bde8SPaolo Bonzini action = mirror_error_action(s, false, -ret); 148a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 149bd48bde8SPaolo Bonzini s->ret = ret; 150bd48bde8SPaolo Bonzini } 151bd48bde8SPaolo Bonzini } 152bd48bde8SPaolo Bonzini mirror_iteration_done(op, ret); 153b9e413ddSPaolo Bonzini aio_context_release(blk_get_aio_context(s->common.blk)); 154bd48bde8SPaolo Bonzini } 155bd48bde8SPaolo Bonzini 156bd48bde8SPaolo Bonzini static void mirror_read_complete(void *opaque, int ret) 157bd48bde8SPaolo Bonzini { 158bd48bde8SPaolo Bonzini MirrorOp *op = opaque; 159bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 160b9e413ddSPaolo Bonzini 161b9e413ddSPaolo Bonzini aio_context_acquire(blk_get_aio_context(s->common.blk)); 162bd48bde8SPaolo Bonzini if (ret < 0) { 163bd48bde8SPaolo Bonzini BlockErrorAction action; 164bd48bde8SPaolo Bonzini 16520dca810SJohn Snow bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); 166bd48bde8SPaolo Bonzini action = mirror_error_action(s, true, -ret); 167a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 168bd48bde8SPaolo Bonzini s->ret = ret; 169bd48bde8SPaolo Bonzini } 170bd48bde8SPaolo Bonzini 171bd48bde8SPaolo Bonzini mirror_iteration_done(op, ret); 172b9e413ddSPaolo Bonzini } else { 173e253f4b8SKevin Wolf blk_aio_pwritev(s->target, op->sector_num * BDRV_SECTOR_SIZE, &op->qiov, 17473698c30SEric Blake 0, mirror_write_complete, op); 175bd48bde8SPaolo Bonzini } 176b9e413ddSPaolo Bonzini aio_context_release(blk_get_aio_context(s->common.blk)); 177b9e413ddSPaolo Bonzini } 178bd48bde8SPaolo Bonzini 1794150ae60SFam Zheng static inline void mirror_clip_sectors(MirrorBlockJob *s, 1804150ae60SFam Zheng int64_t sector_num, 1814150ae60SFam Zheng int *nb_sectors) 1824150ae60SFam Zheng { 1834150ae60SFam Zheng *nb_sectors = MIN(*nb_sectors, 1844150ae60SFam Zheng s->bdev_length / BDRV_SECTOR_SIZE - sector_num); 1854150ae60SFam Zheng } 1864150ae60SFam Zheng 187e5b43573SFam Zheng /* Round sector_num and/or nb_sectors to target cluster if COW is needed, and 188e5b43573SFam Zheng * return the offset of the adjusted tail sector against original. */ 189e5b43573SFam Zheng static int mirror_cow_align(MirrorBlockJob *s, 190e5b43573SFam Zheng int64_t *sector_num, 191e5b43573SFam Zheng int *nb_sectors) 192893f7ebaSPaolo Bonzini { 193e5b43573SFam Zheng bool need_cow; 194e5b43573SFam Zheng int ret = 0; 195e5b43573SFam Zheng int chunk_sectors = s->granularity >> BDRV_SECTOR_BITS; 196e5b43573SFam Zheng int64_t align_sector_num = *sector_num; 197e5b43573SFam Zheng int align_nb_sectors = *nb_sectors; 198e5b43573SFam Zheng int max_sectors = chunk_sectors * s->max_iov; 199893f7ebaSPaolo Bonzini 200e5b43573SFam Zheng need_cow = !test_bit(*sector_num / chunk_sectors, s->cow_bitmap); 201e5b43573SFam Zheng need_cow |= !test_bit((*sector_num + *nb_sectors - 1) / chunk_sectors, 202e5b43573SFam Zheng s->cow_bitmap); 203e5b43573SFam Zheng if (need_cow) { 204244483e6SKevin Wolf bdrv_round_sectors_to_clusters(blk_bs(s->target), *sector_num, 205244483e6SKevin Wolf *nb_sectors, &align_sector_num, 206244483e6SKevin Wolf &align_nb_sectors); 2078f0720ecSPaolo Bonzini } 2088f0720ecSPaolo Bonzini 209e5b43573SFam Zheng if (align_nb_sectors > max_sectors) { 210e5b43573SFam Zheng align_nb_sectors = max_sectors; 211e5b43573SFam Zheng if (need_cow) { 212e5b43573SFam Zheng align_nb_sectors = QEMU_ALIGN_DOWN(align_nb_sectors, 213e5b43573SFam Zheng s->target_cluster_sectors); 214e5b43573SFam Zheng } 215e5b43573SFam Zheng } 2164150ae60SFam Zheng /* Clipping may result in align_nb_sectors unaligned to chunk boundary, but 2174150ae60SFam Zheng * that doesn't matter because it's already the end of source image. */ 2184150ae60SFam Zheng mirror_clip_sectors(s, align_sector_num, &align_nb_sectors); 219402a4741SPaolo Bonzini 220e5b43573SFam Zheng ret = align_sector_num + align_nb_sectors - (*sector_num + *nb_sectors); 221e5b43573SFam Zheng *sector_num = align_sector_num; 222e5b43573SFam Zheng *nb_sectors = align_nb_sectors; 223e5b43573SFam Zheng assert(ret >= 0); 224e5b43573SFam Zheng return ret; 225e5b43573SFam Zheng } 226e5b43573SFam Zheng 22721cd917fSFam Zheng static inline void mirror_wait_for_io(MirrorBlockJob *s) 22821cd917fSFam Zheng { 22921cd917fSFam Zheng assert(!s->waiting_for_io); 23021cd917fSFam Zheng s->waiting_for_io = true; 23121cd917fSFam Zheng qemu_coroutine_yield(); 23221cd917fSFam Zheng s->waiting_for_io = false; 23321cd917fSFam Zheng } 23421cd917fSFam Zheng 235e5b43573SFam Zheng /* Submit async read while handling COW. 23617612955SJohn Snow * Returns: The number of sectors copied after and including sector_num, 23717612955SJohn Snow * excluding any sectors copied prior to sector_num due to alignment. 23817612955SJohn Snow * This will be nb_sectors if no alignment is necessary, or 239e5b43573SFam Zheng * (new_end - sector_num) if tail is rounded up or down due to 240e5b43573SFam Zheng * alignment or buffer limit. 241402a4741SPaolo Bonzini */ 242e5b43573SFam Zheng static int mirror_do_read(MirrorBlockJob *s, int64_t sector_num, 243e5b43573SFam Zheng int nb_sectors) 244e5b43573SFam Zheng { 245e253f4b8SKevin Wolf BlockBackend *source = s->common.blk; 246e5b43573SFam Zheng int sectors_per_chunk, nb_chunks; 24717612955SJohn Snow int ret; 248e5b43573SFam Zheng MirrorOp *op; 249e4808881SJohn Snow int max_sectors; 250402a4741SPaolo Bonzini 251e5b43573SFam Zheng sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; 252e4808881SJohn Snow max_sectors = sectors_per_chunk * s->max_iov; 253e5b43573SFam Zheng 254e5b43573SFam Zheng /* We can only handle as much as buf_size at a time. */ 255e5b43573SFam Zheng nb_sectors = MIN(s->buf_size >> BDRV_SECTOR_BITS, nb_sectors); 256e4808881SJohn Snow nb_sectors = MIN(max_sectors, nb_sectors); 257e5b43573SFam Zheng assert(nb_sectors); 25817612955SJohn Snow ret = nb_sectors; 259e5b43573SFam Zheng 260e5b43573SFam Zheng if (s->cow_bitmap) { 261e5b43573SFam Zheng ret += mirror_cow_align(s, §or_num, &nb_sectors); 262e5b43573SFam Zheng } 263e5b43573SFam Zheng assert(nb_sectors << BDRV_SECTOR_BITS <= s->buf_size); 264e5b43573SFam Zheng /* The sector range must meet granularity because: 265e5b43573SFam Zheng * 1) Caller passes in aligned values; 266e5b43573SFam Zheng * 2) mirror_cow_align is used only when target cluster is larger. */ 267e5b43573SFam Zheng assert(!(sector_num % sectors_per_chunk)); 2684150ae60SFam Zheng nb_chunks = DIV_ROUND_UP(nb_sectors, sectors_per_chunk); 269e5b43573SFam Zheng 270e5b43573SFam Zheng while (s->buf_free_count < nb_chunks) { 271402a4741SPaolo Bonzini trace_mirror_yield_in_flight(s, sector_num, s->in_flight); 27221cd917fSFam Zheng mirror_wait_for_io(s); 273b812f671SPaolo Bonzini } 274b812f671SPaolo Bonzini 275bd48bde8SPaolo Bonzini /* Allocate a MirrorOp that is used as an AIO callback. */ 276c84b3192SPaolo Bonzini op = g_new(MirrorOp, 1); 277bd48bde8SPaolo Bonzini op->s = s; 278bd48bde8SPaolo Bonzini op->sector_num = sector_num; 279bd48bde8SPaolo Bonzini op->nb_sectors = nb_sectors; 280402a4741SPaolo Bonzini 281402a4741SPaolo Bonzini /* Now make a QEMUIOVector taking enough granularity-sized chunks 282402a4741SPaolo Bonzini * from s->buf_free. 283402a4741SPaolo Bonzini */ 284402a4741SPaolo Bonzini qemu_iovec_init(&op->qiov, nb_chunks); 285402a4741SPaolo Bonzini while (nb_chunks-- > 0) { 286402a4741SPaolo Bonzini MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); 287e5b43573SFam Zheng size_t remaining = nb_sectors * BDRV_SECTOR_SIZE - op->qiov.size; 2885a0f6fd5SKevin Wolf 289402a4741SPaolo Bonzini QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); 290402a4741SPaolo Bonzini s->buf_free_count--; 2915a0f6fd5SKevin Wolf qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); 292402a4741SPaolo Bonzini } 293402a4741SPaolo Bonzini 294893f7ebaSPaolo Bonzini /* Copy the dirty cluster. */ 295bd48bde8SPaolo Bonzini s->in_flight++; 296b21c7652SMax Reitz s->sectors_in_flight += nb_sectors; 297b812f671SPaolo Bonzini trace_mirror_one_iteration(s, sector_num, nb_sectors); 298dcfb3bebSFam Zheng 29973698c30SEric Blake blk_aio_preadv(source, sector_num * BDRV_SECTOR_SIZE, &op->qiov, 0, 300bd48bde8SPaolo Bonzini mirror_read_complete, op); 301e5b43573SFam Zheng return ret; 302e5b43573SFam Zheng } 303e5b43573SFam Zheng 304e5b43573SFam Zheng static void mirror_do_zero_or_discard(MirrorBlockJob *s, 305e5b43573SFam Zheng int64_t sector_num, 306e5b43573SFam Zheng int nb_sectors, 307e5b43573SFam Zheng bool is_discard) 308e5b43573SFam Zheng { 309e5b43573SFam Zheng MirrorOp *op; 310e5b43573SFam Zheng 311e5b43573SFam Zheng /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed 312e5b43573SFam Zheng * so the freeing in mirror_iteration_done is nop. */ 313e5b43573SFam Zheng op = g_new0(MirrorOp, 1); 314e5b43573SFam Zheng op->s = s; 315e5b43573SFam Zheng op->sector_num = sector_num; 316e5b43573SFam Zheng op->nb_sectors = nb_sectors; 317e5b43573SFam Zheng 318e5b43573SFam Zheng s->in_flight++; 319e5b43573SFam Zheng s->sectors_in_flight += nb_sectors; 320e5b43573SFam Zheng if (is_discard) { 3211c6c4bb7SEric Blake blk_aio_pdiscard(s->target, sector_num << BDRV_SECTOR_BITS, 3221c6c4bb7SEric Blake op->nb_sectors << BDRV_SECTOR_BITS, 323e5b43573SFam Zheng mirror_write_complete, op); 324e5b43573SFam Zheng } else { 325e253f4b8SKevin Wolf blk_aio_pwrite_zeroes(s->target, sector_num * BDRV_SECTOR_SIZE, 326e253f4b8SKevin Wolf op->nb_sectors * BDRV_SECTOR_SIZE, 327dcfb3bebSFam Zheng s->unmap ? BDRV_REQ_MAY_UNMAP : 0, 328dcfb3bebSFam Zheng mirror_write_complete, op); 329e5b43573SFam Zheng } 330e5b43573SFam Zheng } 331e5b43573SFam Zheng 332e5b43573SFam Zheng static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) 333e5b43573SFam Zheng { 3344ef85a9cSKevin Wolf BlockDriverState *source = s->source; 3359c83625bSMax Reitz int64_t sector_num, first_chunk; 336e5b43573SFam Zheng uint64_t delay_ns = 0; 337e5b43573SFam Zheng /* At least the first dirty chunk is mirrored in one iteration. */ 338e5b43573SFam Zheng int nb_chunks = 1; 339e5b43573SFam Zheng int64_t end = s->bdev_length / BDRV_SECTOR_SIZE; 340e5b43573SFam Zheng int sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; 3414b5004d9SDenis V. Lunev bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); 3420965a41eSVladimir Sementsov-Ogievskiy int max_io_sectors = MAX((s->buf_size >> BDRV_SECTOR_BITS) / MAX_IN_FLIGHT, 3430965a41eSVladimir Sementsov-Ogievskiy MAX_IO_SECTORS); 344e5b43573SFam Zheng 345dc162c8eSFam Zheng sector_num = bdrv_dirty_iter_next(s->dbi); 346e5b43573SFam Zheng if (sector_num < 0) { 347dc162c8eSFam Zheng bdrv_set_dirty_iter(s->dbi, 0); 348dc162c8eSFam Zheng sector_num = bdrv_dirty_iter_next(s->dbi); 349e5b43573SFam Zheng trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); 350e5b43573SFam Zheng assert(sector_num >= 0); 351e5b43573SFam Zheng } 352e5b43573SFam Zheng 3539c83625bSMax Reitz first_chunk = sector_num / sectors_per_chunk; 3549c83625bSMax Reitz while (test_bit(first_chunk, s->in_flight_bitmap)) { 355ff04198bSDenis V. Lunev trace_mirror_yield_in_flight(s, sector_num, s->in_flight); 3569c83625bSMax Reitz mirror_wait_for_io(s); 3579c83625bSMax Reitz } 3589c83625bSMax Reitz 359565ac01fSStefan Hajnoczi block_job_pause_point(&s->common); 360565ac01fSStefan Hajnoczi 361e5b43573SFam Zheng /* Find the number of consective dirty chunks following the first dirty 362e5b43573SFam Zheng * one, and wait for in flight requests in them. */ 363e5b43573SFam Zheng while (nb_chunks * sectors_per_chunk < (s->buf_size >> BDRV_SECTOR_BITS)) { 364dc162c8eSFam Zheng int64_t next_dirty; 365e5b43573SFam Zheng int64_t next_sector = sector_num + nb_chunks * sectors_per_chunk; 366e5b43573SFam Zheng int64_t next_chunk = next_sector / sectors_per_chunk; 367e5b43573SFam Zheng if (next_sector >= end || 368e5b43573SFam Zheng !bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) { 369e5b43573SFam Zheng break; 370e5b43573SFam Zheng } 371e5b43573SFam Zheng if (test_bit(next_chunk, s->in_flight_bitmap)) { 372e5b43573SFam Zheng break; 373e5b43573SFam Zheng } 3749c83625bSMax Reitz 375dc162c8eSFam Zheng next_dirty = bdrv_dirty_iter_next(s->dbi); 376dc162c8eSFam Zheng if (next_dirty > next_sector || next_dirty < 0) { 377f27a2742SMax Reitz /* The bitmap iterator's cache is stale, refresh it */ 378dc162c8eSFam Zheng bdrv_set_dirty_iter(s->dbi, next_sector); 379dc162c8eSFam Zheng next_dirty = bdrv_dirty_iter_next(s->dbi); 380f27a2742SMax Reitz } 381dc162c8eSFam Zheng assert(next_dirty == next_sector); 382e5b43573SFam Zheng nb_chunks++; 383e5b43573SFam Zheng } 384e5b43573SFam Zheng 385e5b43573SFam Zheng /* Clear dirty bits before querying the block status, because 386e5b43573SFam Zheng * calling bdrv_get_block_status_above could yield - if some blocks are 387e5b43573SFam Zheng * marked dirty in this window, we need to know. 388e5b43573SFam Zheng */ 389e5b43573SFam Zheng bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, 390e5b43573SFam Zheng nb_chunks * sectors_per_chunk); 391e5b43573SFam Zheng bitmap_set(s->in_flight_bitmap, sector_num / sectors_per_chunk, nb_chunks); 392e5b43573SFam Zheng while (nb_chunks > 0 && sector_num < end) { 39339c11580SJohn Snow int64_t ret; 3944b5004d9SDenis V. Lunev int io_sectors, io_sectors_acct; 395e5b43573SFam Zheng BlockDriverState *file; 396e5b43573SFam Zheng enum MirrorMethod { 397e5b43573SFam Zheng MIRROR_METHOD_COPY, 398e5b43573SFam Zheng MIRROR_METHOD_ZERO, 399e5b43573SFam Zheng MIRROR_METHOD_DISCARD 400e5b43573SFam Zheng } mirror_method = MIRROR_METHOD_COPY; 401e5b43573SFam Zheng 402e5b43573SFam Zheng assert(!(sector_num % sectors_per_chunk)); 403e5b43573SFam Zheng ret = bdrv_get_block_status_above(source, NULL, sector_num, 404e5b43573SFam Zheng nb_chunks * sectors_per_chunk, 405e5b43573SFam Zheng &io_sectors, &file); 406e5b43573SFam Zheng if (ret < 0) { 4070965a41eSVladimir Sementsov-Ogievskiy io_sectors = MIN(nb_chunks * sectors_per_chunk, max_io_sectors); 4080965a41eSVladimir Sementsov-Ogievskiy } else if (ret & BDRV_BLOCK_DATA) { 4090965a41eSVladimir Sementsov-Ogievskiy io_sectors = MIN(io_sectors, max_io_sectors); 410e5b43573SFam Zheng } 411e5b43573SFam Zheng 412e5b43573SFam Zheng io_sectors -= io_sectors % sectors_per_chunk; 413e5b43573SFam Zheng if (io_sectors < sectors_per_chunk) { 414e5b43573SFam Zheng io_sectors = sectors_per_chunk; 415e5b43573SFam Zheng } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { 416e5b43573SFam Zheng int64_t target_sector_num; 417e5b43573SFam Zheng int target_nb_sectors; 418244483e6SKevin Wolf bdrv_round_sectors_to_clusters(blk_bs(s->target), sector_num, 419244483e6SKevin Wolf io_sectors, &target_sector_num, 420244483e6SKevin Wolf &target_nb_sectors); 421e5b43573SFam Zheng if (target_sector_num == sector_num && 422e5b43573SFam Zheng target_nb_sectors == io_sectors) { 423e5b43573SFam Zheng mirror_method = ret & BDRV_BLOCK_ZERO ? 424e5b43573SFam Zheng MIRROR_METHOD_ZERO : 425e5b43573SFam Zheng MIRROR_METHOD_DISCARD; 426e5b43573SFam Zheng } 427e5b43573SFam Zheng } 428e5b43573SFam Zheng 429cf56a3c6SDenis V. Lunev while (s->in_flight >= MAX_IN_FLIGHT) { 430cf56a3c6SDenis V. Lunev trace_mirror_yield_in_flight(s, sector_num, s->in_flight); 431cf56a3c6SDenis V. Lunev mirror_wait_for_io(s); 432cf56a3c6SDenis V. Lunev } 433cf56a3c6SDenis V. Lunev 434dbaa7b57SVladimir Sementsov-Ogievskiy if (s->ret < 0) { 435dbaa7b57SVladimir Sementsov-Ogievskiy return 0; 436dbaa7b57SVladimir Sementsov-Ogievskiy } 437dbaa7b57SVladimir Sementsov-Ogievskiy 4384150ae60SFam Zheng mirror_clip_sectors(s, sector_num, &io_sectors); 439e5b43573SFam Zheng switch (mirror_method) { 440e5b43573SFam Zheng case MIRROR_METHOD_COPY: 441e5b43573SFam Zheng io_sectors = mirror_do_read(s, sector_num, io_sectors); 4424b5004d9SDenis V. Lunev io_sectors_acct = io_sectors; 443e5b43573SFam Zheng break; 444e5b43573SFam Zheng case MIRROR_METHOD_ZERO: 445e5b43573SFam Zheng case MIRROR_METHOD_DISCARD: 4464b5004d9SDenis V. Lunev mirror_do_zero_or_discard(s, sector_num, io_sectors, 4474b5004d9SDenis V. Lunev mirror_method == MIRROR_METHOD_DISCARD); 4484b5004d9SDenis V. Lunev if (write_zeroes_ok) { 4494b5004d9SDenis V. Lunev io_sectors_acct = 0; 4504b5004d9SDenis V. Lunev } else { 4514b5004d9SDenis V. Lunev io_sectors_acct = io_sectors; 4524b5004d9SDenis V. Lunev } 453e5b43573SFam Zheng break; 454e5b43573SFam Zheng default: 455e5b43573SFam Zheng abort(); 456e5b43573SFam Zheng } 457e5b43573SFam Zheng assert(io_sectors); 458e5b43573SFam Zheng sector_num += io_sectors; 4594150ae60SFam Zheng nb_chunks -= DIV_ROUND_UP(io_sectors, sectors_per_chunk); 460f14a39ccSSascha Silbe if (s->common.speed) { 4614b5004d9SDenis V. Lunev delay_ns = ratelimit_calculate_delay(&s->limit, io_sectors_acct); 462f14a39ccSSascha Silbe } 463dcfb3bebSFam Zheng } 464cc8c9d6cSPaolo Bonzini return delay_ns; 465893f7ebaSPaolo Bonzini } 466b952b558SPaolo Bonzini 467402a4741SPaolo Bonzini static void mirror_free_init(MirrorBlockJob *s) 468402a4741SPaolo Bonzini { 469402a4741SPaolo Bonzini int granularity = s->granularity; 470402a4741SPaolo Bonzini size_t buf_size = s->buf_size; 471402a4741SPaolo Bonzini uint8_t *buf = s->buf; 472402a4741SPaolo Bonzini 473402a4741SPaolo Bonzini assert(s->buf_free_count == 0); 474402a4741SPaolo Bonzini QSIMPLEQ_INIT(&s->buf_free); 475402a4741SPaolo Bonzini while (buf_size != 0) { 476402a4741SPaolo Bonzini MirrorBuffer *cur = (MirrorBuffer *)buf; 477402a4741SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); 478402a4741SPaolo Bonzini s->buf_free_count++; 479402a4741SPaolo Bonzini buf_size -= granularity; 480402a4741SPaolo Bonzini buf += granularity; 481402a4741SPaolo Bonzini } 482402a4741SPaolo Bonzini } 483402a4741SPaolo Bonzini 484bae8196dSPaolo Bonzini /* This is also used for the .pause callback. There is no matching 485bae8196dSPaolo Bonzini * mirror_resume() because mirror_run() will begin iterating again 486bae8196dSPaolo Bonzini * when the job is resumed. 487bae8196dSPaolo Bonzini */ 488bae8196dSPaolo Bonzini static void mirror_wait_for_all_io(MirrorBlockJob *s) 489bd48bde8SPaolo Bonzini { 490bd48bde8SPaolo Bonzini while (s->in_flight > 0) { 49121cd917fSFam Zheng mirror_wait_for_io(s); 492bd48bde8SPaolo Bonzini } 493893f7ebaSPaolo Bonzini } 494893f7ebaSPaolo Bonzini 4955a7e7a0bSStefan Hajnoczi typedef struct { 4965a7e7a0bSStefan Hajnoczi int ret; 4975a7e7a0bSStefan Hajnoczi } MirrorExitData; 4985a7e7a0bSStefan Hajnoczi 4995a7e7a0bSStefan Hajnoczi static void mirror_exit(BlockJob *job, void *opaque) 5005a7e7a0bSStefan Hajnoczi { 5015a7e7a0bSStefan Hajnoczi MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 5025a7e7a0bSStefan Hajnoczi MirrorExitData *data = opaque; 5035a7e7a0bSStefan Hajnoczi AioContext *replace_aio_context = NULL; 5044ef85a9cSKevin Wolf BlockDriverState *src = s->source; 505e253f4b8SKevin Wolf BlockDriverState *target_bs = blk_bs(s->target); 5064ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs = s->mirror_top_bs; 50712fa4af6SKevin Wolf Error *local_err = NULL; 5083f09bfbcSKevin Wolf 5093f09bfbcSKevin Wolf /* Make sure that the source BDS doesn't go away before we called 5103f09bfbcSKevin Wolf * block_job_completed(). */ 5113f09bfbcSKevin Wolf bdrv_ref(src); 5124ef85a9cSKevin Wolf bdrv_ref(mirror_top_bs); 5137d9fcb39SKevin Wolf bdrv_ref(target_bs); 5147d9fcb39SKevin Wolf 5157d9fcb39SKevin Wolf /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before 5167d9fcb39SKevin Wolf * inserting target_bs at s->to_replace, where we might not be able to get 5177d9fcb39SKevin Wolf * these permissions. */ 5187d9fcb39SKevin Wolf blk_unref(s->target); 5197d9fcb39SKevin Wolf s->target = NULL; 5204ef85a9cSKevin Wolf 5214ef85a9cSKevin Wolf /* We don't access the source any more. Dropping any WRITE/RESIZE is 5224ef85a9cSKevin Wolf * required before it could become a backing file of target_bs. */ 5234ef85a9cSKevin Wolf bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, 5244ef85a9cSKevin Wolf &error_abort); 5254ef85a9cSKevin Wolf if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { 5264ef85a9cSKevin Wolf BlockDriverState *backing = s->is_none_mode ? src : s->base; 5274ef85a9cSKevin Wolf if (backing_bs(target_bs) != backing) { 52812fa4af6SKevin Wolf bdrv_set_backing_hd(target_bs, backing, &local_err); 52912fa4af6SKevin Wolf if (local_err) { 53012fa4af6SKevin Wolf error_report_err(local_err); 53112fa4af6SKevin Wolf data->ret = -EPERM; 53212fa4af6SKevin Wolf } 5334ef85a9cSKevin Wolf } 5344ef85a9cSKevin Wolf } 5355a7e7a0bSStefan Hajnoczi 5365a7e7a0bSStefan Hajnoczi if (s->to_replace) { 5375a7e7a0bSStefan Hajnoczi replace_aio_context = bdrv_get_aio_context(s->to_replace); 5385a7e7a0bSStefan Hajnoczi aio_context_acquire(replace_aio_context); 5395a7e7a0bSStefan Hajnoczi } 5405a7e7a0bSStefan Hajnoczi 5415a7e7a0bSStefan Hajnoczi if (s->should_complete && data->ret == 0) { 542e253f4b8SKevin Wolf BlockDriverState *to_replace = src; 5435a7e7a0bSStefan Hajnoczi if (s->to_replace) { 5445a7e7a0bSStefan Hajnoczi to_replace = s->to_replace; 5455a7e7a0bSStefan Hajnoczi } 54640365552SKevin Wolf 547e253f4b8SKevin Wolf if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { 548e253f4b8SKevin Wolf bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); 5495a7e7a0bSStefan Hajnoczi } 550b8804815SKevin Wolf 551b8804815SKevin Wolf /* The mirror job has no requests in flight any more, but we need to 552b8804815SKevin Wolf * drain potential other users of the BDS before changing the graph. */ 553e253f4b8SKevin Wolf bdrv_drained_begin(target_bs); 5545fe31c25SKevin Wolf bdrv_replace_node(to_replace, target_bs, &local_err); 555e253f4b8SKevin Wolf bdrv_drained_end(target_bs); 5565fe31c25SKevin Wolf if (local_err) { 5575fe31c25SKevin Wolf error_report_err(local_err); 5585fe31c25SKevin Wolf data->ret = -EPERM; 5595fe31c25SKevin Wolf } 5605a7e7a0bSStefan Hajnoczi } 5615a7e7a0bSStefan Hajnoczi if (s->to_replace) { 5625a7e7a0bSStefan Hajnoczi bdrv_op_unblock_all(s->to_replace, s->replace_blocker); 5635a7e7a0bSStefan Hajnoczi error_free(s->replace_blocker); 5645a7e7a0bSStefan Hajnoczi bdrv_unref(s->to_replace); 5655a7e7a0bSStefan Hajnoczi } 5665a7e7a0bSStefan Hajnoczi if (replace_aio_context) { 5675a7e7a0bSStefan Hajnoczi aio_context_release(replace_aio_context); 5685a7e7a0bSStefan Hajnoczi } 5695a7e7a0bSStefan Hajnoczi g_free(s->replaces); 5707d9fcb39SKevin Wolf bdrv_unref(target_bs); 5714ef85a9cSKevin Wolf 5724ef85a9cSKevin Wolf /* Remove the mirror filter driver from the graph. Before this, get rid of 5734ef85a9cSKevin Wolf * the blockers on the intermediate nodes so that the resulting state is 5740bf74767SKevin Wolf * valid. Also give up permissions on mirror_top_bs->backing, which might 5750bf74767SKevin Wolf * block the removal. */ 5764ef85a9cSKevin Wolf block_job_remove_all_bdrv(job); 577c1cef672SFam Zheng bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, 578c1cef672SFam Zheng &error_abort); 5795fe31c25SKevin Wolf bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); 5804ef85a9cSKevin Wolf 5814ef85a9cSKevin Wolf /* We just changed the BDS the job BB refers to (with either or both of the 5825fe31c25SKevin Wolf * bdrv_replace_node() calls), so switch the BB back so the cleanup does 5835fe31c25SKevin Wolf * the right thing. We don't need any permissions any more now. */ 5844ef85a9cSKevin Wolf blk_remove_bs(job->blk); 5854ef85a9cSKevin Wolf blk_set_perm(job->blk, 0, BLK_PERM_ALL, &error_abort); 5864ef85a9cSKevin Wolf blk_insert_bs(job->blk, mirror_top_bs, &error_abort); 5874ef85a9cSKevin Wolf 5885a7e7a0bSStefan Hajnoczi block_job_completed(&s->common, data->ret); 5894ef85a9cSKevin Wolf 5905a7e7a0bSStefan Hajnoczi g_free(data); 591176c3699SFam Zheng bdrv_drained_end(src); 5924ef85a9cSKevin Wolf bdrv_unref(mirror_top_bs); 5933f09bfbcSKevin Wolf bdrv_unref(src); 5945a7e7a0bSStefan Hajnoczi } 5955a7e7a0bSStefan Hajnoczi 59649efb1f5SDenis V. Lunev static void mirror_throttle(MirrorBlockJob *s) 59749efb1f5SDenis V. Lunev { 59849efb1f5SDenis V. Lunev int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 59949efb1f5SDenis V. Lunev 60049efb1f5SDenis V. Lunev if (now - s->last_pause_ns > SLICE_TIME) { 60149efb1f5SDenis V. Lunev s->last_pause_ns = now; 60249efb1f5SDenis V. Lunev block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0); 60349efb1f5SDenis V. Lunev } else { 60449efb1f5SDenis V. Lunev block_job_pause_point(&s->common); 60549efb1f5SDenis V. Lunev } 60649efb1f5SDenis V. Lunev } 60749efb1f5SDenis V. Lunev 608c0b363adSDenis V. Lunev static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) 609c0b363adSDenis V. Lunev { 610c0b363adSDenis V. Lunev int64_t sector_num, end; 611c0b363adSDenis V. Lunev BlockDriverState *base = s->base; 6124ef85a9cSKevin Wolf BlockDriverState *bs = s->source; 613c0b363adSDenis V. Lunev BlockDriverState *target_bs = blk_bs(s->target); 614c0b363adSDenis V. Lunev int ret, n; 615c0b363adSDenis V. Lunev 616c0b363adSDenis V. Lunev end = s->bdev_length / BDRV_SECTOR_SIZE; 617c0b363adSDenis V. Lunev 618b7d5062cSDenis V. Lunev if (base == NULL && !bdrv_has_zero_init(target_bs)) { 619c7c2769cSDenis V. Lunev if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { 620b7d5062cSDenis V. Lunev bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, end); 621b7d5062cSDenis V. Lunev return 0; 622b7d5062cSDenis V. Lunev } 623b7d5062cSDenis V. Lunev 62490ab48ebSAnton Nefedov s->initial_zeroing_ongoing = true; 625c7c2769cSDenis V. Lunev for (sector_num = 0; sector_num < end; ) { 626c7c2769cSDenis V. Lunev int nb_sectors = MIN(end - sector_num, 627c7c2769cSDenis V. Lunev QEMU_ALIGN_DOWN(INT_MAX, s->granularity) >> BDRV_SECTOR_BITS); 628c7c2769cSDenis V. Lunev 629c7c2769cSDenis V. Lunev mirror_throttle(s); 630c7c2769cSDenis V. Lunev 631c7c2769cSDenis V. Lunev if (block_job_is_cancelled(&s->common)) { 63290ab48ebSAnton Nefedov s->initial_zeroing_ongoing = false; 633c7c2769cSDenis V. Lunev return 0; 634c7c2769cSDenis V. Lunev } 635c7c2769cSDenis V. Lunev 636c7c2769cSDenis V. Lunev if (s->in_flight >= MAX_IN_FLIGHT) { 63767adf4b3SEric Blake trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, 63867adf4b3SEric Blake s->in_flight); 639c7c2769cSDenis V. Lunev mirror_wait_for_io(s); 640c7c2769cSDenis V. Lunev continue; 641c7c2769cSDenis V. Lunev } 642c7c2769cSDenis V. Lunev 643c7c2769cSDenis V. Lunev mirror_do_zero_or_discard(s, sector_num, nb_sectors, false); 644c7c2769cSDenis V. Lunev sector_num += nb_sectors; 645c7c2769cSDenis V. Lunev } 646c7c2769cSDenis V. Lunev 647bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 64890ab48ebSAnton Nefedov s->initial_zeroing_ongoing = false; 649c7c2769cSDenis V. Lunev } 650c7c2769cSDenis V. Lunev 651c0b363adSDenis V. Lunev /* First part, loop on the sectors and initialize the dirty bitmap. */ 652c0b363adSDenis V. Lunev for (sector_num = 0; sector_num < end; ) { 653c0b363adSDenis V. Lunev /* Just to make sure we are not exceeding int limit. */ 654c0b363adSDenis V. Lunev int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS, 655c0b363adSDenis V. Lunev end - sector_num); 656c0b363adSDenis V. Lunev 657c0b363adSDenis V. Lunev mirror_throttle(s); 658c0b363adSDenis V. Lunev 659c0b363adSDenis V. Lunev if (block_job_is_cancelled(&s->common)) { 660c0b363adSDenis V. Lunev return 0; 661c0b363adSDenis V. Lunev } 662c0b363adSDenis V. Lunev 663c0b363adSDenis V. Lunev ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n); 664c0b363adSDenis V. Lunev if (ret < 0) { 665c0b363adSDenis V. Lunev return ret; 666c0b363adSDenis V. Lunev } 667c0b363adSDenis V. Lunev 668c0b363adSDenis V. Lunev assert(n > 0); 669b7d5062cSDenis V. Lunev if (ret == 1) { 670c0b363adSDenis V. Lunev bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n); 671c0b363adSDenis V. Lunev } 672c0b363adSDenis V. Lunev sector_num += n; 673c0b363adSDenis V. Lunev } 674c0b363adSDenis V. Lunev return 0; 675c0b363adSDenis V. Lunev } 676c0b363adSDenis V. Lunev 677bdffb31dSPaolo Bonzini /* Called when going out of the streaming phase to flush the bulk of the 678bdffb31dSPaolo Bonzini * data to the medium, or just before completing. 679bdffb31dSPaolo Bonzini */ 680bdffb31dSPaolo Bonzini static int mirror_flush(MirrorBlockJob *s) 681bdffb31dSPaolo Bonzini { 682bdffb31dSPaolo Bonzini int ret = blk_flush(s->target); 683bdffb31dSPaolo Bonzini if (ret < 0) { 684bdffb31dSPaolo Bonzini if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { 685bdffb31dSPaolo Bonzini s->ret = ret; 686bdffb31dSPaolo Bonzini } 687bdffb31dSPaolo Bonzini } 688bdffb31dSPaolo Bonzini return ret; 689bdffb31dSPaolo Bonzini } 690bdffb31dSPaolo Bonzini 691893f7ebaSPaolo Bonzini static void coroutine_fn mirror_run(void *opaque) 692893f7ebaSPaolo Bonzini { 693893f7ebaSPaolo Bonzini MirrorBlockJob *s = opaque; 6945a7e7a0bSStefan Hajnoczi MirrorExitData *data; 6954ef85a9cSKevin Wolf BlockDriverState *bs = s->source; 696e253f4b8SKevin Wolf BlockDriverState *target_bs = blk_bs(s->target); 6979a0cec66SPaolo Bonzini bool need_drain = true; 698c0b363adSDenis V. Lunev int64_t length; 699b812f671SPaolo Bonzini BlockDriverInfo bdi; 7001d33936eSJeff Cody char backing_filename[2]; /* we only need 2 characters because we are only 7011d33936eSJeff Cody checking for a NULL string */ 702893f7ebaSPaolo Bonzini int ret = 0; 703e5b43573SFam Zheng int target_cluster_size = BDRV_SECTOR_SIZE; 704893f7ebaSPaolo Bonzini 705893f7ebaSPaolo Bonzini if (block_job_is_cancelled(&s->common)) { 706893f7ebaSPaolo Bonzini goto immediate_exit; 707893f7ebaSPaolo Bonzini } 708893f7ebaSPaolo Bonzini 709b21c7652SMax Reitz s->bdev_length = bdrv_getlength(bs); 710b21c7652SMax Reitz if (s->bdev_length < 0) { 711b21c7652SMax Reitz ret = s->bdev_length; 712373df5b1SFam Zheng goto immediate_exit; 713becc347eSKevin Wolf } 714becc347eSKevin Wolf 715becc347eSKevin Wolf /* Active commit must resize the base image if its size differs from the 716becc347eSKevin Wolf * active layer. */ 717becc347eSKevin Wolf if (s->base == blk_bs(s->target)) { 718becc347eSKevin Wolf int64_t base_length; 719becc347eSKevin Wolf 720becc347eSKevin Wolf base_length = blk_getlength(s->target); 721becc347eSKevin Wolf if (base_length < 0) { 722becc347eSKevin Wolf ret = base_length; 723becc347eSKevin Wolf goto immediate_exit; 724becc347eSKevin Wolf } 725becc347eSKevin Wolf 726becc347eSKevin Wolf if (s->bdev_length > base_length) { 727becc347eSKevin Wolf ret = blk_truncate(s->target, s->bdev_length); 728becc347eSKevin Wolf if (ret < 0) { 729becc347eSKevin Wolf goto immediate_exit; 730becc347eSKevin Wolf } 731becc347eSKevin Wolf } 732becc347eSKevin Wolf } 733becc347eSKevin Wolf 734becc347eSKevin Wolf if (s->bdev_length == 0) { 7359e48b025SFam Zheng /* Report BLOCK_JOB_READY and wait for complete. */ 7369e48b025SFam Zheng block_job_event_ready(&s->common); 7379e48b025SFam Zheng s->synced = true; 7389e48b025SFam Zheng while (!block_job_is_cancelled(&s->common) && !s->should_complete) { 7399e48b025SFam Zheng block_job_yield(&s->common); 7409e48b025SFam Zheng } 7419e48b025SFam Zheng s->common.cancelled = false; 7429e48b025SFam Zheng goto immediate_exit; 743893f7ebaSPaolo Bonzini } 744893f7ebaSPaolo Bonzini 745b21c7652SMax Reitz length = DIV_ROUND_UP(s->bdev_length, s->granularity); 746402a4741SPaolo Bonzini s->in_flight_bitmap = bitmap_new(length); 747402a4741SPaolo Bonzini 748b812f671SPaolo Bonzini /* If we have no backing file yet in the destination, we cannot let 749b812f671SPaolo Bonzini * the destination do COW. Instead, we copy sectors around the 750b812f671SPaolo Bonzini * dirty data if needed. We need a bitmap to do that. 751b812f671SPaolo Bonzini */ 752e253f4b8SKevin Wolf bdrv_get_backing_filename(target_bs, backing_filename, 753b812f671SPaolo Bonzini sizeof(backing_filename)); 754e253f4b8SKevin Wolf if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { 755e5b43573SFam Zheng target_cluster_size = bdi.cluster_size; 756c3cc95bdSFam Zheng } 757e253f4b8SKevin Wolf if (backing_filename[0] && !target_bs->backing 758e5b43573SFam Zheng && s->granularity < target_cluster_size) { 759e5b43573SFam Zheng s->buf_size = MAX(s->buf_size, target_cluster_size); 760b812f671SPaolo Bonzini s->cow_bitmap = bitmap_new(length); 761b812f671SPaolo Bonzini } 762e5b43573SFam Zheng s->target_cluster_sectors = target_cluster_size >> BDRV_SECTOR_BITS; 763e253f4b8SKevin Wolf s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); 764b812f671SPaolo Bonzini 7657504edf4SKevin Wolf s->buf = qemu_try_blockalign(bs, s->buf_size); 7667504edf4SKevin Wolf if (s->buf == NULL) { 7677504edf4SKevin Wolf ret = -ENOMEM; 7687504edf4SKevin Wolf goto immediate_exit; 7697504edf4SKevin Wolf } 7707504edf4SKevin Wolf 771402a4741SPaolo Bonzini mirror_free_init(s); 772893f7ebaSPaolo Bonzini 77349efb1f5SDenis V. Lunev s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 77403544a6eSFam Zheng if (!s->is_none_mode) { 775c0b363adSDenis V. Lunev ret = mirror_dirty_init(s); 776c0b363adSDenis V. Lunev if (ret < 0 || block_job_is_cancelled(&s->common)) { 7774c0cbd6fSFam Zheng goto immediate_exit; 7784c0cbd6fSFam Zheng } 779893f7ebaSPaolo Bonzini } 780893f7ebaSPaolo Bonzini 781dc162c8eSFam Zheng assert(!s->dbi); 782dc162c8eSFam Zheng s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap, 0); 783893f7ebaSPaolo Bonzini for (;;) { 784cc8c9d6cSPaolo Bonzini uint64_t delay_ns = 0; 78549efb1f5SDenis V. Lunev int64_t cnt, delta; 786893f7ebaSPaolo Bonzini bool should_complete; 787893f7ebaSPaolo Bonzini 788bd48bde8SPaolo Bonzini if (s->ret < 0) { 789bd48bde8SPaolo Bonzini ret = s->ret; 790893f7ebaSPaolo Bonzini goto immediate_exit; 791893f7ebaSPaolo Bonzini } 792bd48bde8SPaolo Bonzini 793565ac01fSStefan Hajnoczi block_job_pause_point(&s->common); 794565ac01fSStefan Hajnoczi 79520dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 796b21c7652SMax Reitz /* s->common.offset contains the number of bytes already processed so 797b21c7652SMax Reitz * far, cnt is the number of dirty sectors remaining and 798b21c7652SMax Reitz * s->sectors_in_flight is the number of sectors currently being 799b21c7652SMax Reitz * processed; together those are the current total operation length */ 800b21c7652SMax Reitz s->common.len = s->common.offset + 801b21c7652SMax Reitz (cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE; 802bd48bde8SPaolo Bonzini 803bd48bde8SPaolo Bonzini /* Note that even when no rate limit is applied we need to yield 804a7282330SFam Zheng * periodically with no pending I/O so that bdrv_drain_all() returns. 805bd48bde8SPaolo Bonzini * We do so every SLICE_TIME nanoseconds, or when there is an error, 806bd48bde8SPaolo Bonzini * or when the source is clean, whichever comes first. 807bd48bde8SPaolo Bonzini */ 80849efb1f5SDenis V. Lunev delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; 80949efb1f5SDenis V. Lunev if (delta < SLICE_TIME && 810bd48bde8SPaolo Bonzini s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 811cf56a3c6SDenis V. Lunev if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || 812402a4741SPaolo Bonzini (cnt == 0 && s->in_flight > 0)) { 81367adf4b3SEric Blake trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); 81421cd917fSFam Zheng mirror_wait_for_io(s); 815bd48bde8SPaolo Bonzini continue; 816bd48bde8SPaolo Bonzini } else if (cnt != 0) { 817cc8c9d6cSPaolo Bonzini delay_ns = mirror_iteration(s); 818893f7ebaSPaolo Bonzini } 819cc8c9d6cSPaolo Bonzini } 820893f7ebaSPaolo Bonzini 821893f7ebaSPaolo Bonzini should_complete = false; 822bd48bde8SPaolo Bonzini if (s->in_flight == 0 && cnt == 0) { 823893f7ebaSPaolo Bonzini trace_mirror_before_flush(s); 824bdffb31dSPaolo Bonzini if (!s->synced) { 825bdffb31dSPaolo Bonzini if (mirror_flush(s) < 0) { 826bdffb31dSPaolo Bonzini /* Go check s->ret. */ 827bdffb31dSPaolo Bonzini continue; 828893f7ebaSPaolo Bonzini } 829893f7ebaSPaolo Bonzini /* We're out of the streaming phase. From now on, if the job 830893f7ebaSPaolo Bonzini * is cancelled we will actually complete all pending I/O and 831893f7ebaSPaolo Bonzini * report completion. This way, block-job-cancel will leave 832893f7ebaSPaolo Bonzini * the target in a consistent state. 833893f7ebaSPaolo Bonzini */ 834bcada37bSWenchao Xia block_job_event_ready(&s->common); 835d63ffd87SPaolo Bonzini s->synced = true; 836d63ffd87SPaolo Bonzini } 837d63ffd87SPaolo Bonzini 838d63ffd87SPaolo Bonzini should_complete = s->should_complete || 839d63ffd87SPaolo Bonzini block_job_is_cancelled(&s->common); 84020dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 841893f7ebaSPaolo Bonzini } 842893f7ebaSPaolo Bonzini 843893f7ebaSPaolo Bonzini if (cnt == 0 && should_complete) { 844893f7ebaSPaolo Bonzini /* The dirty bitmap is not updated while operations are pending. 845893f7ebaSPaolo Bonzini * If we're about to exit, wait for pending operations before 846893f7ebaSPaolo Bonzini * calling bdrv_get_dirty_count(bs), or we may exit while the 847893f7ebaSPaolo Bonzini * source has dirty data to copy! 848893f7ebaSPaolo Bonzini * 849893f7ebaSPaolo Bonzini * Note that I/O can be submitted by the guest while 8509a0cec66SPaolo Bonzini * mirror_populate runs, so pause it now. Before deciding 8519a0cec66SPaolo Bonzini * whether to switch to target check one last time if I/O has 8529a0cec66SPaolo Bonzini * come in the meanwhile, and if not flush the data to disk. 853893f7ebaSPaolo Bonzini */ 854893f7ebaSPaolo Bonzini trace_mirror_before_drain(s, cnt); 8559a0cec66SPaolo Bonzini 8569a0cec66SPaolo Bonzini bdrv_drained_begin(bs); 85720dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 858bdffb31dSPaolo Bonzini if (cnt > 0 || mirror_flush(s) < 0) { 8599a0cec66SPaolo Bonzini bdrv_drained_end(bs); 8609a0cec66SPaolo Bonzini continue; 8619a0cec66SPaolo Bonzini } 8629a0cec66SPaolo Bonzini 8639a0cec66SPaolo Bonzini /* The two disks are in sync. Exit and report successful 8649a0cec66SPaolo Bonzini * completion. 8659a0cec66SPaolo Bonzini */ 8669a0cec66SPaolo Bonzini assert(QLIST_EMPTY(&bs->tracked_requests)); 8679a0cec66SPaolo Bonzini s->common.cancelled = false; 8689a0cec66SPaolo Bonzini need_drain = false; 8699a0cec66SPaolo Bonzini break; 870893f7ebaSPaolo Bonzini } 871893f7ebaSPaolo Bonzini 872893f7ebaSPaolo Bonzini ret = 0; 873cc8c9d6cSPaolo Bonzini trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); 874d63ffd87SPaolo Bonzini if (!s->synced) { 8757483d1e5SAlex Bligh block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); 876893f7ebaSPaolo Bonzini if (block_job_is_cancelled(&s->common)) { 877893f7ebaSPaolo Bonzini break; 878893f7ebaSPaolo Bonzini } 879893f7ebaSPaolo Bonzini } else if (!should_complete) { 880bd48bde8SPaolo Bonzini delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); 8817483d1e5SAlex Bligh block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); 882893f7ebaSPaolo Bonzini } 88349efb1f5SDenis V. Lunev s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 884893f7ebaSPaolo Bonzini } 885893f7ebaSPaolo Bonzini 886893f7ebaSPaolo Bonzini immediate_exit: 887bd48bde8SPaolo Bonzini if (s->in_flight > 0) { 888bd48bde8SPaolo Bonzini /* We get here only if something went wrong. Either the job failed, 889bd48bde8SPaolo Bonzini * or it was cancelled prematurely so that we do not guarantee that 890bd48bde8SPaolo Bonzini * the target is a copy of the source. 891bd48bde8SPaolo Bonzini */ 892bd48bde8SPaolo Bonzini assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); 8939a0cec66SPaolo Bonzini assert(need_drain); 894bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 895bd48bde8SPaolo Bonzini } 896bd48bde8SPaolo Bonzini 897bd48bde8SPaolo Bonzini assert(s->in_flight == 0); 8987191bf31SMarkus Armbruster qemu_vfree(s->buf); 899b812f671SPaolo Bonzini g_free(s->cow_bitmap); 900402a4741SPaolo Bonzini g_free(s->in_flight_bitmap); 901dc162c8eSFam Zheng bdrv_dirty_iter_free(s->dbi); 902e4654d2dSFam Zheng bdrv_release_dirty_bitmap(bs, s->dirty_bitmap); 9035a7e7a0bSStefan Hajnoczi 9045a7e7a0bSStefan Hajnoczi data = g_malloc(sizeof(*data)); 9055a7e7a0bSStefan Hajnoczi data->ret = ret; 9069a0cec66SPaolo Bonzini 9079a0cec66SPaolo Bonzini if (need_drain) { 908e253f4b8SKevin Wolf bdrv_drained_begin(bs); 9099a0cec66SPaolo Bonzini } 9105a7e7a0bSStefan Hajnoczi block_job_defer_to_main_loop(&s->common, mirror_exit, data); 911893f7ebaSPaolo Bonzini } 912893f7ebaSPaolo Bonzini 913893f7ebaSPaolo Bonzini static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) 914893f7ebaSPaolo Bonzini { 915893f7ebaSPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 916893f7ebaSPaolo Bonzini 917893f7ebaSPaolo Bonzini if (speed < 0) { 918c6bd8c70SMarkus Armbruster error_setg(errp, QERR_INVALID_PARAMETER, "speed"); 919893f7ebaSPaolo Bonzini return; 920893f7ebaSPaolo Bonzini } 921893f7ebaSPaolo Bonzini ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); 922893f7ebaSPaolo Bonzini } 923893f7ebaSPaolo Bonzini 924d63ffd87SPaolo Bonzini static void mirror_complete(BlockJob *job, Error **errp) 925d63ffd87SPaolo Bonzini { 926d63ffd87SPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 9274ef85a9cSKevin Wolf BlockDriverState *target; 928d63ffd87SPaolo Bonzini 929274fcceeSMax Reitz target = blk_bs(s->target); 930274fcceeSMax Reitz 931d63ffd87SPaolo Bonzini if (!s->synced) { 9329df229c3SAlberto Garcia error_setg(errp, "The active block job '%s' cannot be completed", 9339df229c3SAlberto Garcia job->id); 934d63ffd87SPaolo Bonzini return; 935d63ffd87SPaolo Bonzini } 936d63ffd87SPaolo Bonzini 937274fcceeSMax Reitz if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { 938274fcceeSMax Reitz int ret; 939274fcceeSMax Reitz 940274fcceeSMax Reitz assert(!target->backing); 941274fcceeSMax Reitz ret = bdrv_open_backing_file(target, NULL, "backing", errp); 942274fcceeSMax Reitz if (ret < 0) { 943274fcceeSMax Reitz return; 944274fcceeSMax Reitz } 945274fcceeSMax Reitz } 946274fcceeSMax Reitz 94715d67298SChanglong Xie /* block all operations on to_replace bs */ 94809158f00SBenoît Canet if (s->replaces) { 9495a7e7a0bSStefan Hajnoczi AioContext *replace_aio_context; 9505a7e7a0bSStefan Hajnoczi 951e12f3784SWen Congyang s->to_replace = bdrv_find_node(s->replaces); 95209158f00SBenoît Canet if (!s->to_replace) { 953e12f3784SWen Congyang error_setg(errp, "Node name '%s' not found", s->replaces); 95409158f00SBenoît Canet return; 95509158f00SBenoît Canet } 95609158f00SBenoît Canet 9575a7e7a0bSStefan Hajnoczi replace_aio_context = bdrv_get_aio_context(s->to_replace); 9585a7e7a0bSStefan Hajnoczi aio_context_acquire(replace_aio_context); 9595a7e7a0bSStefan Hajnoczi 9604ef85a9cSKevin Wolf /* TODO Translate this into permission system. Current definition of 9614ef85a9cSKevin Wolf * GRAPH_MOD would require to request it for the parents; they might 9624ef85a9cSKevin Wolf * not even be BlockDriverStates, however, so a BdrvChild can't address 9634ef85a9cSKevin Wolf * them. May need redefinition of GRAPH_MOD. */ 96409158f00SBenoît Canet error_setg(&s->replace_blocker, 96509158f00SBenoît Canet "block device is in use by block-job-complete"); 96609158f00SBenoît Canet bdrv_op_block_all(s->to_replace, s->replace_blocker); 96709158f00SBenoît Canet bdrv_ref(s->to_replace); 9685a7e7a0bSStefan Hajnoczi 9695a7e7a0bSStefan Hajnoczi aio_context_release(replace_aio_context); 97009158f00SBenoît Canet } 97109158f00SBenoît Canet 972d63ffd87SPaolo Bonzini s->should_complete = true; 973751ebd76SFam Zheng block_job_enter(&s->common); 974d63ffd87SPaolo Bonzini } 975d63ffd87SPaolo Bonzini 976bae8196dSPaolo Bonzini static void mirror_pause(BlockJob *job) 977565ac01fSStefan Hajnoczi { 978565ac01fSStefan Hajnoczi MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 979565ac01fSStefan Hajnoczi 980bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 981565ac01fSStefan Hajnoczi } 982565ac01fSStefan Hajnoczi 983565ac01fSStefan Hajnoczi static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) 984565ac01fSStefan Hajnoczi { 985565ac01fSStefan Hajnoczi MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 986565ac01fSStefan Hajnoczi 987565ac01fSStefan Hajnoczi blk_set_aio_context(s->target, new_context); 988565ac01fSStefan Hajnoczi } 989565ac01fSStefan Hajnoczi 990bae8196dSPaolo Bonzini static void mirror_drain(BlockJob *job) 991bae8196dSPaolo Bonzini { 992bae8196dSPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 993bae8196dSPaolo Bonzini 994bae8196dSPaolo Bonzini /* Need to keep a reference in case blk_drain triggers execution 995bae8196dSPaolo Bonzini * of mirror_complete... 996bae8196dSPaolo Bonzini */ 997bae8196dSPaolo Bonzini if (s->target) { 998bae8196dSPaolo Bonzini BlockBackend *target = s->target; 999bae8196dSPaolo Bonzini blk_ref(target); 1000bae8196dSPaolo Bonzini blk_drain(target); 1001bae8196dSPaolo Bonzini blk_unref(target); 1002bae8196dSPaolo Bonzini } 1003bae8196dSPaolo Bonzini } 1004bae8196dSPaolo Bonzini 10053fc4b10aSFam Zheng static const BlockJobDriver mirror_job_driver = { 1006893f7ebaSPaolo Bonzini .instance_size = sizeof(MirrorBlockJob), 100779e14bf7SFam Zheng .job_type = BLOCK_JOB_TYPE_MIRROR, 1008893f7ebaSPaolo Bonzini .set_speed = mirror_set_speed, 1009a7815a76SJohn Snow .start = mirror_run, 1010d63ffd87SPaolo Bonzini .complete = mirror_complete, 1011565ac01fSStefan Hajnoczi .pause = mirror_pause, 1012565ac01fSStefan Hajnoczi .attached_aio_context = mirror_attached_aio_context, 1013bae8196dSPaolo Bonzini .drain = mirror_drain, 1014893f7ebaSPaolo Bonzini }; 1015893f7ebaSPaolo Bonzini 101603544a6eSFam Zheng static const BlockJobDriver commit_active_job_driver = { 101703544a6eSFam Zheng .instance_size = sizeof(MirrorBlockJob), 101803544a6eSFam Zheng .job_type = BLOCK_JOB_TYPE_COMMIT, 101903544a6eSFam Zheng .set_speed = mirror_set_speed, 1020a7815a76SJohn Snow .start = mirror_run, 102103544a6eSFam Zheng .complete = mirror_complete, 1022565ac01fSStefan Hajnoczi .pause = mirror_pause, 1023565ac01fSStefan Hajnoczi .attached_aio_context = mirror_attached_aio_context, 1024bae8196dSPaolo Bonzini .drain = mirror_drain, 102503544a6eSFam Zheng }; 102603544a6eSFam Zheng 10274ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, 10284ef85a9cSKevin Wolf uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) 10294ef85a9cSKevin Wolf { 10304ef85a9cSKevin Wolf return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); 10314ef85a9cSKevin Wolf } 10324ef85a9cSKevin Wolf 10334ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, 10344ef85a9cSKevin Wolf uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) 10354ef85a9cSKevin Wolf { 10364ef85a9cSKevin Wolf return bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); 10374ef85a9cSKevin Wolf } 10384ef85a9cSKevin Wolf 10394ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) 10404ef85a9cSKevin Wolf { 10414ef85a9cSKevin Wolf return bdrv_co_flush(bs->backing->bs); 10424ef85a9cSKevin Wolf } 10434ef85a9cSKevin Wolf 10444ef85a9cSKevin Wolf static int64_t coroutine_fn bdrv_mirror_top_get_block_status( 10454ef85a9cSKevin Wolf BlockDriverState *bs, int64_t sector_num, int nb_sectors, int *pnum, 10464ef85a9cSKevin Wolf BlockDriverState **file) 10474ef85a9cSKevin Wolf { 10484ef85a9cSKevin Wolf *pnum = nb_sectors; 10494ef85a9cSKevin Wolf *file = bs->backing->bs; 10504ef85a9cSKevin Wolf return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID | BDRV_BLOCK_DATA | 10514ef85a9cSKevin Wolf (sector_num << BDRV_SECTOR_BITS); 10524ef85a9cSKevin Wolf } 10534ef85a9cSKevin Wolf 10544ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, 10554ef85a9cSKevin Wolf int64_t offset, int count, BdrvRequestFlags flags) 10564ef85a9cSKevin Wolf { 10574ef85a9cSKevin Wolf return bdrv_co_pwrite_zeroes(bs->backing, offset, count, flags); 10584ef85a9cSKevin Wolf } 10594ef85a9cSKevin Wolf 10604ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, 10614ef85a9cSKevin Wolf int64_t offset, int count) 10624ef85a9cSKevin Wolf { 10634ef85a9cSKevin Wolf return bdrv_co_pdiscard(bs->backing->bs, offset, count); 10644ef85a9cSKevin Wolf } 10654ef85a9cSKevin Wolf 1066fd4a6493SKevin Wolf static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs, QDict *opts) 1067fd4a6493SKevin Wolf { 1068fd4a6493SKevin Wolf bdrv_refresh_filename(bs->backing->bs); 1069fd4a6493SKevin Wolf pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), 1070fd4a6493SKevin Wolf bs->backing->bs->filename); 1071fd4a6493SKevin Wolf } 1072fd4a6493SKevin Wolf 10734ef85a9cSKevin Wolf static void bdrv_mirror_top_close(BlockDriverState *bs) 10744ef85a9cSKevin Wolf { 10754ef85a9cSKevin Wolf } 10764ef85a9cSKevin Wolf 10774ef85a9cSKevin Wolf static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, 10784ef85a9cSKevin Wolf const BdrvChildRole *role, 10794ef85a9cSKevin Wolf uint64_t perm, uint64_t shared, 10804ef85a9cSKevin Wolf uint64_t *nperm, uint64_t *nshared) 10814ef85a9cSKevin Wolf { 10824ef85a9cSKevin Wolf /* Must be able to forward guest writes to the real image */ 10834ef85a9cSKevin Wolf *nperm = 0; 10844ef85a9cSKevin Wolf if (perm & BLK_PERM_WRITE) { 10854ef85a9cSKevin Wolf *nperm |= BLK_PERM_WRITE; 10864ef85a9cSKevin Wolf } 10874ef85a9cSKevin Wolf 10884ef85a9cSKevin Wolf *nshared = BLK_PERM_ALL; 10894ef85a9cSKevin Wolf } 10904ef85a9cSKevin Wolf 10914ef85a9cSKevin Wolf /* Dummy node that provides consistent read to its users without requiring it 10924ef85a9cSKevin Wolf * from its backing file and that allows writes on the backing file chain. */ 10934ef85a9cSKevin Wolf static BlockDriver bdrv_mirror_top = { 10944ef85a9cSKevin Wolf .format_name = "mirror_top", 10954ef85a9cSKevin Wolf .bdrv_co_preadv = bdrv_mirror_top_preadv, 10964ef85a9cSKevin Wolf .bdrv_co_pwritev = bdrv_mirror_top_pwritev, 10974ef85a9cSKevin Wolf .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, 10984ef85a9cSKevin Wolf .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, 10994ef85a9cSKevin Wolf .bdrv_co_flush = bdrv_mirror_top_flush, 11004ef85a9cSKevin Wolf .bdrv_co_get_block_status = bdrv_mirror_top_get_block_status, 1101fd4a6493SKevin Wolf .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, 11024ef85a9cSKevin Wolf .bdrv_close = bdrv_mirror_top_close, 11034ef85a9cSKevin Wolf .bdrv_child_perm = bdrv_mirror_top_child_perm, 11044ef85a9cSKevin Wolf }; 11054ef85a9cSKevin Wolf 110671aa9867SAlberto Garcia static void mirror_start_job(const char *job_id, BlockDriverState *bs, 110747970dfbSJohn Snow int creation_flags, BlockDriverState *target, 110847970dfbSJohn Snow const char *replaces, int64_t speed, 110947970dfbSJohn Snow uint32_t granularity, int64_t buf_size, 1110274fcceeSMax Reitz BlockMirrorBackingMode backing_mode, 111103544a6eSFam Zheng BlockdevOnError on_source_error, 1112b952b558SPaolo Bonzini BlockdevOnError on_target_error, 11130fc9f8eaSFam Zheng bool unmap, 1114097310b5SMarkus Armbruster BlockCompletionFunc *cb, 111503544a6eSFam Zheng void *opaque, Error **errp, 111603544a6eSFam Zheng const BlockJobDriver *driver, 1117b49f7eadSWen Congyang bool is_none_mode, BlockDriverState *base, 11186cdbceb1SKevin Wolf bool auto_complete, const char *filter_node_name) 1119893f7ebaSPaolo Bonzini { 1120893f7ebaSPaolo Bonzini MirrorBlockJob *s; 11214ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs; 11224ef85a9cSKevin Wolf bool target_graph_mod; 11234ef85a9cSKevin Wolf bool target_is_backing; 1124b2c2832cSKevin Wolf Error *local_err = NULL; 1125d7086422SKevin Wolf int ret; 1126893f7ebaSPaolo Bonzini 1127eee13dfeSPaolo Bonzini if (granularity == 0) { 1128341ebc2fSJohn Snow granularity = bdrv_get_default_bitmap_granularity(target); 1129eee13dfeSPaolo Bonzini } 1130eee13dfeSPaolo Bonzini 1131eee13dfeSPaolo Bonzini assert ((granularity & (granularity - 1)) == 0); 1132eee13dfeSPaolo Bonzini 113348ac0a4dSWen Congyang if (buf_size < 0) { 113448ac0a4dSWen Congyang error_setg(errp, "Invalid parameter 'buf-size'"); 113548ac0a4dSWen Congyang return; 113648ac0a4dSWen Congyang } 113748ac0a4dSWen Congyang 113848ac0a4dSWen Congyang if (buf_size == 0) { 113948ac0a4dSWen Congyang buf_size = DEFAULT_MIRROR_BUF_SIZE; 114048ac0a4dSWen Congyang } 11415bc361b8SFam Zheng 11424ef85a9cSKevin Wolf /* In the case of active commit, add dummy driver to provide consistent 11434ef85a9cSKevin Wolf * reads on the top, while disabling it in the intermediate nodes, and make 11444ef85a9cSKevin Wolf * the backing chain writable. */ 11456cdbceb1SKevin Wolf mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, 11466cdbceb1SKevin Wolf BDRV_O_RDWR, errp); 11474ef85a9cSKevin Wolf if (mirror_top_bs == NULL) { 1148893f7ebaSPaolo Bonzini return; 1149893f7ebaSPaolo Bonzini } 11504ef85a9cSKevin Wolf mirror_top_bs->total_sectors = bs->total_sectors; 1151*19dd29e8SFam Zheng bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs)); 1152893f7ebaSPaolo Bonzini 11534ef85a9cSKevin Wolf /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep 11547a25fcd0SMax Reitz * it alive until block_job_create() succeeds even if bs has no parent. */ 11554ef85a9cSKevin Wolf bdrv_ref(mirror_top_bs); 11564ef85a9cSKevin Wolf bdrv_drained_begin(bs); 1157b2c2832cSKevin Wolf bdrv_append(mirror_top_bs, bs, &local_err); 11584ef85a9cSKevin Wolf bdrv_drained_end(bs); 11594ef85a9cSKevin Wolf 1160b2c2832cSKevin Wolf if (local_err) { 1161b2c2832cSKevin Wolf bdrv_unref(mirror_top_bs); 1162b2c2832cSKevin Wolf error_propagate(errp, local_err); 1163b2c2832cSKevin Wolf return; 1164b2c2832cSKevin Wolf } 1165b2c2832cSKevin Wolf 11664ef85a9cSKevin Wolf /* Make sure that the source is not resized while the job is running */ 11674ef85a9cSKevin Wolf s = block_job_create(job_id, driver, mirror_top_bs, 11684ef85a9cSKevin Wolf BLK_PERM_CONSISTENT_READ, 11694ef85a9cSKevin Wolf BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | 11704ef85a9cSKevin Wolf BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, 11714ef85a9cSKevin Wolf creation_flags, cb, opaque, errp); 11724ef85a9cSKevin Wolf if (!s) { 11734ef85a9cSKevin Wolf goto fail; 11744ef85a9cSKevin Wolf } 11757a25fcd0SMax Reitz /* The block job now has a reference to this node */ 11767a25fcd0SMax Reitz bdrv_unref(mirror_top_bs); 11777a25fcd0SMax Reitz 11784ef85a9cSKevin Wolf s->source = bs; 11794ef85a9cSKevin Wolf s->mirror_top_bs = mirror_top_bs; 11804ef85a9cSKevin Wolf 11814ef85a9cSKevin Wolf /* No resize for the target either; while the mirror is still running, a 11824ef85a9cSKevin Wolf * consistent read isn't necessarily possible. We could possibly allow 11834ef85a9cSKevin Wolf * writes and graph modifications, though it would likely defeat the 11844ef85a9cSKevin Wolf * purpose of a mirror, so leave them blocked for now. 11854ef85a9cSKevin Wolf * 11864ef85a9cSKevin Wolf * In the case of active commit, things look a bit different, though, 11874ef85a9cSKevin Wolf * because the target is an already populated backing file in active use. 11884ef85a9cSKevin Wolf * We can allow anything except resize there.*/ 11894ef85a9cSKevin Wolf target_is_backing = bdrv_chain_contains(bs, target); 11904ef85a9cSKevin Wolf target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); 11914ef85a9cSKevin Wolf s->target = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE | 11924ef85a9cSKevin Wolf (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), 11934ef85a9cSKevin Wolf BLK_PERM_WRITE_UNCHANGED | 11944ef85a9cSKevin Wolf (target_is_backing ? BLK_PERM_CONSISTENT_READ | 11954ef85a9cSKevin Wolf BLK_PERM_WRITE | 11964ef85a9cSKevin Wolf BLK_PERM_GRAPH_MOD : 0)); 1197d7086422SKevin Wolf ret = blk_insert_bs(s->target, target, errp); 1198d7086422SKevin Wolf if (ret < 0) { 11994ef85a9cSKevin Wolf goto fail; 1200d7086422SKevin Wolf } 1201e253f4b8SKevin Wolf 120209158f00SBenoît Canet s->replaces = g_strdup(replaces); 1203b952b558SPaolo Bonzini s->on_source_error = on_source_error; 1204b952b558SPaolo Bonzini s->on_target_error = on_target_error; 120503544a6eSFam Zheng s->is_none_mode = is_none_mode; 1206274fcceeSMax Reitz s->backing_mode = backing_mode; 12075bc361b8SFam Zheng s->base = base; 1208eee13dfeSPaolo Bonzini s->granularity = granularity; 120948ac0a4dSWen Congyang s->buf_size = ROUND_UP(buf_size, granularity); 12100fc9f8eaSFam Zheng s->unmap = unmap; 1211b49f7eadSWen Congyang if (auto_complete) { 1212b49f7eadSWen Congyang s->should_complete = true; 1213b49f7eadSWen Congyang } 1214b812f671SPaolo Bonzini 12150db6e54aSFam Zheng s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); 1216b8afb520SFam Zheng if (!s->dirty_bitmap) { 121788f9d1b3SKevin Wolf goto fail; 1218b8afb520SFam Zheng } 121910f3cd15SAlberto Garcia 12204ef85a9cSKevin Wolf /* Required permissions are already taken with blk_new() */ 122176d554e2SKevin Wolf block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, 122276d554e2SKevin Wolf &error_abort); 122376d554e2SKevin Wolf 1224f3ede4b0SAlberto Garcia /* In commit_active_start() all intermediate nodes disappear, so 1225f3ede4b0SAlberto Garcia * any jobs in them must be blocked */ 12264ef85a9cSKevin Wolf if (target_is_backing) { 1227f3ede4b0SAlberto Garcia BlockDriverState *iter; 1228f3ede4b0SAlberto Garcia for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { 12294ef85a9cSKevin Wolf /* XXX BLK_PERM_WRITE needs to be allowed so we don't block 12304ef85a9cSKevin Wolf * ourselves at s->base (if writes are blocked for a node, they are 12314ef85a9cSKevin Wolf * also blocked for its backing file). The other options would be a 12324ef85a9cSKevin Wolf * second filter driver above s->base (== target). */ 12334ef85a9cSKevin Wolf ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, 12344ef85a9cSKevin Wolf BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, 12354ef85a9cSKevin Wolf errp); 12364ef85a9cSKevin Wolf if (ret < 0) { 12374ef85a9cSKevin Wolf goto fail; 12384ef85a9cSKevin Wolf } 1239f3ede4b0SAlberto Garcia } 1240f3ede4b0SAlberto Garcia } 124110f3cd15SAlberto Garcia 12425ccac6f1SJohn Snow trace_mirror_start(bs, s, opaque); 12435ccac6f1SJohn Snow block_job_start(&s->common); 12444ef85a9cSKevin Wolf return; 12454ef85a9cSKevin Wolf 12464ef85a9cSKevin Wolf fail: 12474ef85a9cSKevin Wolf if (s) { 12487a25fcd0SMax Reitz /* Make sure this BDS does not go away until we have completed the graph 12497a25fcd0SMax Reitz * changes below */ 12507a25fcd0SMax Reitz bdrv_ref(mirror_top_bs); 12517a25fcd0SMax Reitz 12524ef85a9cSKevin Wolf g_free(s->replaces); 12534ef85a9cSKevin Wolf blk_unref(s->target); 12544ef85a9cSKevin Wolf block_job_unref(&s->common); 12554ef85a9cSKevin Wolf } 12564ef85a9cSKevin Wolf 1257c1cef672SFam Zheng bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, 1258c1cef672SFam Zheng &error_abort); 12595fe31c25SKevin Wolf bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); 12607a25fcd0SMax Reitz 12617a25fcd0SMax Reitz bdrv_unref(mirror_top_bs); 1262893f7ebaSPaolo Bonzini } 126303544a6eSFam Zheng 126471aa9867SAlberto Garcia void mirror_start(const char *job_id, BlockDriverState *bs, 126571aa9867SAlberto Garcia BlockDriverState *target, const char *replaces, 12665fba6c0eSJohn Snow int64_t speed, uint32_t granularity, int64_t buf_size, 1267274fcceeSMax Reitz MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, 1268274fcceeSMax Reitz BlockdevOnError on_source_error, 126903544a6eSFam Zheng BlockdevOnError on_target_error, 12706cdbceb1SKevin Wolf bool unmap, const char *filter_node_name, Error **errp) 127103544a6eSFam Zheng { 127203544a6eSFam Zheng bool is_none_mode; 127303544a6eSFam Zheng BlockDriverState *base; 127403544a6eSFam Zheng 12754b80ab2bSJohn Snow if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { 12764b80ab2bSJohn Snow error_setg(errp, "Sync mode 'incremental' not supported"); 1277d58d8453SJohn Snow return; 1278d58d8453SJohn Snow } 127903544a6eSFam Zheng is_none_mode = mode == MIRROR_SYNC_MODE_NONE; 1280760e0063SKevin Wolf base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; 128147970dfbSJohn Snow mirror_start_job(job_id, bs, BLOCK_JOB_DEFAULT, target, replaces, 1282274fcceeSMax Reitz speed, granularity, buf_size, backing_mode, 12838254b6d9SJohn Snow on_source_error, on_target_error, unmap, NULL, NULL, errp, 12846cdbceb1SKevin Wolf &mirror_job_driver, is_none_mode, base, false, 12856cdbceb1SKevin Wolf filter_node_name); 128603544a6eSFam Zheng } 128703544a6eSFam Zheng 1288fd62c609SAlberto Garcia void commit_active_start(const char *job_id, BlockDriverState *bs, 128947970dfbSJohn Snow BlockDriverState *base, int creation_flags, 129047970dfbSJohn Snow int64_t speed, BlockdevOnError on_error, 12910db832f4SKevin Wolf const char *filter_node_name, 129247970dfbSJohn Snow BlockCompletionFunc *cb, void *opaque, Error **errp, 1293b49f7eadSWen Congyang bool auto_complete) 129403544a6eSFam Zheng { 12954da83585SJeff Cody int orig_base_flags; 1296cc67f4d1SJeff Cody Error *local_err = NULL; 12974da83585SJeff Cody 12984da83585SJeff Cody orig_base_flags = bdrv_get_flags(base); 12994da83585SJeff Cody 130020a63d2cSFam Zheng if (bdrv_reopen(base, bs->open_flags, errp)) { 130120a63d2cSFam Zheng return; 130220a63d2cSFam Zheng } 13034da83585SJeff Cody 130447970dfbSJohn Snow mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, 130571aa9867SAlberto Garcia MIRROR_LEAVE_BACKING_CHAIN, 13066f13acf9SFam Zheng on_error, on_error, true, cb, opaque, &local_err, 13076cdbceb1SKevin Wolf &commit_active_job_driver, false, base, auto_complete, 13080db832f4SKevin Wolf filter_node_name); 13090fb6395cSMarkus Armbruster if (local_err) { 1310cc67f4d1SJeff Cody error_propagate(errp, local_err); 13114da83585SJeff Cody goto error_restore_flags; 13124da83585SJeff Cody } 13134da83585SJeff Cody 13144da83585SJeff Cody return; 13154da83585SJeff Cody 13164da83585SJeff Cody error_restore_flags: 13174da83585SJeff Cody /* ignore error and errp for bdrv_reopen, because we want to propagate 13184da83585SJeff Cody * the original error */ 13194da83585SJeff Cody bdrv_reopen(base, orig_base_flags, NULL); 13204da83585SJeff Cody return; 132103544a6eSFam Zheng } 1322