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" 15893f7ebaSPaolo Bonzini #include "trace.h" 16737e150eSPaolo Bonzini #include "block/blockjob.h" 17737e150eSPaolo Bonzini #include "block/block_int.h" 18373340b2SMax Reitz #include "sysemu/block-backend.h" 19da34e65cSMarkus Armbruster #include "qapi/error.h" 20cc7a8ea7SMarkus Armbruster #include "qapi/qmp/qerror.h" 21893f7ebaSPaolo Bonzini #include "qemu/ratelimit.h" 22b812f671SPaolo Bonzini #include "qemu/bitmap.h" 2340365552SKevin Wolf #include "qemu/error-report.h" 24893f7ebaSPaolo Bonzini 25893f7ebaSPaolo Bonzini #define SLICE_TIME 100000000ULL /* ns */ 26402a4741SPaolo Bonzini #define MAX_IN_FLIGHT 16 2748ac0a4dSWen Congyang #define DEFAULT_MIRROR_BUF_SIZE (10 << 20) 28402a4741SPaolo Bonzini 29402a4741SPaolo Bonzini /* The mirroring buffer is a list of granularity-sized chunks. 30402a4741SPaolo Bonzini * Free chunks are organized in a list. 31402a4741SPaolo Bonzini */ 32402a4741SPaolo Bonzini typedef struct MirrorBuffer { 33402a4741SPaolo Bonzini QSIMPLEQ_ENTRY(MirrorBuffer) next; 34402a4741SPaolo Bonzini } MirrorBuffer; 35893f7ebaSPaolo Bonzini 36893f7ebaSPaolo Bonzini typedef struct MirrorBlockJob { 37893f7ebaSPaolo Bonzini BlockJob common; 38893f7ebaSPaolo Bonzini RateLimit limit; 39893f7ebaSPaolo Bonzini BlockDriverState *target; 405bc361b8SFam Zheng BlockDriverState *base; 4109158f00SBenoît Canet /* The name of the graph node to replace */ 4209158f00SBenoît Canet char *replaces; 4309158f00SBenoît Canet /* The BDS to replace */ 4409158f00SBenoît Canet BlockDriverState *to_replace; 4509158f00SBenoît Canet /* Used to block operations on the drive-mirror-replace target */ 4609158f00SBenoît Canet Error *replace_blocker; 4703544a6eSFam Zheng bool is_none_mode; 48b952b558SPaolo Bonzini BlockdevOnError on_source_error, on_target_error; 49d63ffd87SPaolo Bonzini bool synced; 50d63ffd87SPaolo Bonzini bool should_complete; 51eee13dfeSPaolo Bonzini int64_t granularity; 52b812f671SPaolo Bonzini size_t buf_size; 53b21c7652SMax Reitz int64_t bdev_length; 54b812f671SPaolo Bonzini unsigned long *cow_bitmap; 55e4654d2dSFam Zheng BdrvDirtyBitmap *dirty_bitmap; 568f0720ecSPaolo Bonzini HBitmapIter hbi; 57893f7ebaSPaolo Bonzini uint8_t *buf; 58402a4741SPaolo Bonzini QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; 59402a4741SPaolo Bonzini int buf_free_count; 60bd48bde8SPaolo Bonzini 61402a4741SPaolo Bonzini unsigned long *in_flight_bitmap; 62bd48bde8SPaolo Bonzini int in_flight; 63b21c7652SMax Reitz int sectors_in_flight; 64bd48bde8SPaolo Bonzini int ret; 650fc9f8eaSFam Zheng bool unmap; 66e424aff5SKevin Wolf bool waiting_for_io; 67e5b43573SFam Zheng int target_cluster_sectors; 68e5b43573SFam Zheng int max_iov; 69893f7ebaSPaolo Bonzini } MirrorBlockJob; 70893f7ebaSPaolo Bonzini 71bd48bde8SPaolo Bonzini typedef struct MirrorOp { 72bd48bde8SPaolo Bonzini MirrorBlockJob *s; 73bd48bde8SPaolo Bonzini QEMUIOVector qiov; 74bd48bde8SPaolo Bonzini int64_t sector_num; 75bd48bde8SPaolo Bonzini int nb_sectors; 76bd48bde8SPaolo Bonzini } MirrorOp; 77bd48bde8SPaolo Bonzini 78b952b558SPaolo Bonzini static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, 79b952b558SPaolo Bonzini int error) 80b952b558SPaolo Bonzini { 81b952b558SPaolo Bonzini s->synced = false; 82b952b558SPaolo Bonzini if (read) { 83b952b558SPaolo Bonzini return block_job_error_action(&s->common, s->common.bs, 84b952b558SPaolo Bonzini s->on_source_error, true, error); 85b952b558SPaolo Bonzini } else { 86b952b558SPaolo Bonzini return block_job_error_action(&s->common, s->target, 87b952b558SPaolo Bonzini s->on_target_error, false, error); 88b952b558SPaolo Bonzini } 89b952b558SPaolo Bonzini } 90b952b558SPaolo Bonzini 91bd48bde8SPaolo Bonzini static void mirror_iteration_done(MirrorOp *op, int ret) 92bd48bde8SPaolo Bonzini { 93bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 94402a4741SPaolo Bonzini struct iovec *iov; 95bd48bde8SPaolo Bonzini int64_t chunk_num; 96402a4741SPaolo Bonzini int i, nb_chunks, sectors_per_chunk; 97bd48bde8SPaolo Bonzini 98bd48bde8SPaolo Bonzini trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret); 99bd48bde8SPaolo Bonzini 100bd48bde8SPaolo Bonzini s->in_flight--; 101b21c7652SMax Reitz s->sectors_in_flight -= op->nb_sectors; 102402a4741SPaolo Bonzini iov = op->qiov.iov; 103402a4741SPaolo Bonzini for (i = 0; i < op->qiov.niov; i++) { 104402a4741SPaolo Bonzini MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; 105402a4741SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); 106402a4741SPaolo Bonzini s->buf_free_count++; 107402a4741SPaolo Bonzini } 108402a4741SPaolo Bonzini 109bd48bde8SPaolo Bonzini sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; 110bd48bde8SPaolo Bonzini chunk_num = op->sector_num / sectors_per_chunk; 1114150ae60SFam Zheng nb_chunks = DIV_ROUND_UP(op->nb_sectors, sectors_per_chunk); 112402a4741SPaolo Bonzini bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); 113b21c7652SMax Reitz if (ret >= 0) { 114b21c7652SMax Reitz if (s->cow_bitmap) { 115bd48bde8SPaolo Bonzini bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); 116bd48bde8SPaolo Bonzini } 117b21c7652SMax Reitz s->common.offset += (uint64_t)op->nb_sectors * BDRV_SECTOR_SIZE; 118b21c7652SMax Reitz } 119bd48bde8SPaolo Bonzini 1206df3bf8eSZhang Min qemu_iovec_destroy(&op->qiov); 121c84b3192SPaolo Bonzini g_free(op); 1227b770c72SStefan Hajnoczi 123e424aff5SKevin Wolf if (s->waiting_for_io) { 124bd48bde8SPaolo Bonzini qemu_coroutine_enter(s->common.co, NULL); 125bd48bde8SPaolo Bonzini } 1267b770c72SStefan Hajnoczi } 127bd48bde8SPaolo Bonzini 128bd48bde8SPaolo Bonzini static void mirror_write_complete(void *opaque, int ret) 129bd48bde8SPaolo Bonzini { 130bd48bde8SPaolo Bonzini MirrorOp *op = opaque; 131bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 132bd48bde8SPaolo Bonzini if (ret < 0) { 133bd48bde8SPaolo Bonzini BlockErrorAction action; 134bd48bde8SPaolo Bonzini 13520dca810SJohn Snow bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); 136bd48bde8SPaolo Bonzini action = mirror_error_action(s, false, -ret); 137a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 138bd48bde8SPaolo Bonzini s->ret = ret; 139bd48bde8SPaolo Bonzini } 140bd48bde8SPaolo Bonzini } 141bd48bde8SPaolo Bonzini mirror_iteration_done(op, ret); 142bd48bde8SPaolo Bonzini } 143bd48bde8SPaolo Bonzini 144bd48bde8SPaolo Bonzini static void mirror_read_complete(void *opaque, int ret) 145bd48bde8SPaolo Bonzini { 146bd48bde8SPaolo Bonzini MirrorOp *op = opaque; 147bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 148bd48bde8SPaolo Bonzini if (ret < 0) { 149bd48bde8SPaolo Bonzini BlockErrorAction action; 150bd48bde8SPaolo Bonzini 15120dca810SJohn Snow bdrv_set_dirty_bitmap(s->dirty_bitmap, op->sector_num, op->nb_sectors); 152bd48bde8SPaolo Bonzini action = mirror_error_action(s, true, -ret); 153a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 154bd48bde8SPaolo Bonzini s->ret = ret; 155bd48bde8SPaolo Bonzini } 156bd48bde8SPaolo Bonzini 157bd48bde8SPaolo Bonzini mirror_iteration_done(op, ret); 158bd48bde8SPaolo Bonzini return; 159bd48bde8SPaolo Bonzini } 160bd48bde8SPaolo Bonzini bdrv_aio_writev(s->target, op->sector_num, &op->qiov, op->nb_sectors, 161bd48bde8SPaolo Bonzini mirror_write_complete, op); 162bd48bde8SPaolo Bonzini } 163bd48bde8SPaolo Bonzini 1644150ae60SFam Zheng static inline void mirror_clip_sectors(MirrorBlockJob *s, 1654150ae60SFam Zheng int64_t sector_num, 1664150ae60SFam Zheng int *nb_sectors) 1674150ae60SFam Zheng { 1684150ae60SFam Zheng *nb_sectors = MIN(*nb_sectors, 1694150ae60SFam Zheng s->bdev_length / BDRV_SECTOR_SIZE - sector_num); 1704150ae60SFam Zheng } 1714150ae60SFam Zheng 172e5b43573SFam Zheng /* Round sector_num and/or nb_sectors to target cluster if COW is needed, and 173e5b43573SFam Zheng * return the offset of the adjusted tail sector against original. */ 174e5b43573SFam Zheng static int mirror_cow_align(MirrorBlockJob *s, 175e5b43573SFam Zheng int64_t *sector_num, 176e5b43573SFam Zheng int *nb_sectors) 177893f7ebaSPaolo Bonzini { 178e5b43573SFam Zheng bool need_cow; 179e5b43573SFam Zheng int ret = 0; 180e5b43573SFam Zheng int chunk_sectors = s->granularity >> BDRV_SECTOR_BITS; 181e5b43573SFam Zheng int64_t align_sector_num = *sector_num; 182e5b43573SFam Zheng int align_nb_sectors = *nb_sectors; 183e5b43573SFam Zheng int max_sectors = chunk_sectors * s->max_iov; 184893f7ebaSPaolo Bonzini 185e5b43573SFam Zheng need_cow = !test_bit(*sector_num / chunk_sectors, s->cow_bitmap); 186e5b43573SFam Zheng need_cow |= !test_bit((*sector_num + *nb_sectors - 1) / chunk_sectors, 187e5b43573SFam Zheng s->cow_bitmap); 188e5b43573SFam Zheng if (need_cow) { 189e5b43573SFam Zheng bdrv_round_to_clusters(s->target, *sector_num, *nb_sectors, 190e5b43573SFam Zheng &align_sector_num, &align_nb_sectors); 1918f0720ecSPaolo Bonzini } 1928f0720ecSPaolo Bonzini 193e5b43573SFam Zheng if (align_nb_sectors > max_sectors) { 194e5b43573SFam Zheng align_nb_sectors = max_sectors; 195e5b43573SFam Zheng if (need_cow) { 196e5b43573SFam Zheng align_nb_sectors = QEMU_ALIGN_DOWN(align_nb_sectors, 197e5b43573SFam Zheng s->target_cluster_sectors); 198e5b43573SFam Zheng } 199e5b43573SFam Zheng } 2004150ae60SFam Zheng /* Clipping may result in align_nb_sectors unaligned to chunk boundary, but 2014150ae60SFam Zheng * that doesn't matter because it's already the end of source image. */ 2024150ae60SFam Zheng mirror_clip_sectors(s, align_sector_num, &align_nb_sectors); 203402a4741SPaolo Bonzini 204e5b43573SFam Zheng ret = align_sector_num + align_nb_sectors - (*sector_num + *nb_sectors); 205e5b43573SFam Zheng *sector_num = align_sector_num; 206e5b43573SFam Zheng *nb_sectors = align_nb_sectors; 207e5b43573SFam Zheng assert(ret >= 0); 208e5b43573SFam Zheng return ret; 209e5b43573SFam Zheng } 210e5b43573SFam Zheng 21121cd917fSFam Zheng static inline void mirror_wait_for_io(MirrorBlockJob *s) 21221cd917fSFam Zheng { 21321cd917fSFam Zheng assert(!s->waiting_for_io); 21421cd917fSFam Zheng s->waiting_for_io = true; 21521cd917fSFam Zheng qemu_coroutine_yield(); 21621cd917fSFam Zheng s->waiting_for_io = false; 21721cd917fSFam Zheng } 21821cd917fSFam Zheng 219e5b43573SFam Zheng /* Submit async read while handling COW. 220e5b43573SFam Zheng * Returns: nb_sectors if no alignment is necessary, or 221e5b43573SFam Zheng * (new_end - sector_num) if tail is rounded up or down due to 222e5b43573SFam Zheng * alignment or buffer limit. 223402a4741SPaolo Bonzini */ 224e5b43573SFam Zheng static int mirror_do_read(MirrorBlockJob *s, int64_t sector_num, 225e5b43573SFam Zheng int nb_sectors) 226e5b43573SFam Zheng { 227e5b43573SFam Zheng BlockDriverState *source = s->common.bs; 228e5b43573SFam Zheng int sectors_per_chunk, nb_chunks; 229e5b43573SFam Zheng int ret = nb_sectors; 230e5b43573SFam Zheng MirrorOp *op; 231402a4741SPaolo Bonzini 232e5b43573SFam Zheng sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; 233e5b43573SFam Zheng 234e5b43573SFam Zheng /* We can only handle as much as buf_size at a time. */ 235e5b43573SFam Zheng nb_sectors = MIN(s->buf_size >> BDRV_SECTOR_BITS, nb_sectors); 236e5b43573SFam Zheng assert(nb_sectors); 237e5b43573SFam Zheng 238e5b43573SFam Zheng if (s->cow_bitmap) { 239e5b43573SFam Zheng ret += mirror_cow_align(s, §or_num, &nb_sectors); 240e5b43573SFam Zheng } 241e5b43573SFam Zheng assert(nb_sectors << BDRV_SECTOR_BITS <= s->buf_size); 242e5b43573SFam Zheng /* The sector range must meet granularity because: 243e5b43573SFam Zheng * 1) Caller passes in aligned values; 244e5b43573SFam Zheng * 2) mirror_cow_align is used only when target cluster is larger. */ 245e5b43573SFam Zheng assert(!(sector_num % sectors_per_chunk)); 2464150ae60SFam Zheng nb_chunks = DIV_ROUND_UP(nb_sectors, sectors_per_chunk); 247e5b43573SFam Zheng 248e5b43573SFam Zheng while (s->buf_free_count < nb_chunks) { 249402a4741SPaolo Bonzini trace_mirror_yield_in_flight(s, sector_num, s->in_flight); 25021cd917fSFam Zheng mirror_wait_for_io(s); 251b812f671SPaolo Bonzini } 252b812f671SPaolo Bonzini 253bd48bde8SPaolo Bonzini /* Allocate a MirrorOp that is used as an AIO callback. */ 254c84b3192SPaolo Bonzini op = g_new(MirrorOp, 1); 255bd48bde8SPaolo Bonzini op->s = s; 256bd48bde8SPaolo Bonzini op->sector_num = sector_num; 257bd48bde8SPaolo Bonzini op->nb_sectors = nb_sectors; 258402a4741SPaolo Bonzini 259402a4741SPaolo Bonzini /* Now make a QEMUIOVector taking enough granularity-sized chunks 260402a4741SPaolo Bonzini * from s->buf_free. 261402a4741SPaolo Bonzini */ 262402a4741SPaolo Bonzini qemu_iovec_init(&op->qiov, nb_chunks); 263402a4741SPaolo Bonzini while (nb_chunks-- > 0) { 264402a4741SPaolo Bonzini MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); 265e5b43573SFam Zheng size_t remaining = nb_sectors * BDRV_SECTOR_SIZE - op->qiov.size; 2665a0f6fd5SKevin Wolf 267402a4741SPaolo Bonzini QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); 268402a4741SPaolo Bonzini s->buf_free_count--; 2695a0f6fd5SKevin Wolf qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); 270402a4741SPaolo Bonzini } 271402a4741SPaolo Bonzini 272893f7ebaSPaolo Bonzini /* Copy the dirty cluster. */ 273bd48bde8SPaolo Bonzini s->in_flight++; 274b21c7652SMax Reitz s->sectors_in_flight += nb_sectors; 275b812f671SPaolo Bonzini trace_mirror_one_iteration(s, sector_num, nb_sectors); 276dcfb3bebSFam Zheng 277bd48bde8SPaolo Bonzini bdrv_aio_readv(source, sector_num, &op->qiov, nb_sectors, 278bd48bde8SPaolo Bonzini mirror_read_complete, op); 279e5b43573SFam Zheng return ret; 280e5b43573SFam Zheng } 281e5b43573SFam Zheng 282e5b43573SFam Zheng static void mirror_do_zero_or_discard(MirrorBlockJob *s, 283e5b43573SFam Zheng int64_t sector_num, 284e5b43573SFam Zheng int nb_sectors, 285e5b43573SFam Zheng bool is_discard) 286e5b43573SFam Zheng { 287e5b43573SFam Zheng MirrorOp *op; 288e5b43573SFam Zheng 289e5b43573SFam Zheng /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed 290e5b43573SFam Zheng * so the freeing in mirror_iteration_done is nop. */ 291e5b43573SFam Zheng op = g_new0(MirrorOp, 1); 292e5b43573SFam Zheng op->s = s; 293e5b43573SFam Zheng op->sector_num = sector_num; 294e5b43573SFam Zheng op->nb_sectors = nb_sectors; 295e5b43573SFam Zheng 296e5b43573SFam Zheng s->in_flight++; 297e5b43573SFam Zheng s->sectors_in_flight += nb_sectors; 298e5b43573SFam Zheng if (is_discard) { 299e5b43573SFam Zheng bdrv_aio_discard(s->target, sector_num, op->nb_sectors, 300e5b43573SFam Zheng mirror_write_complete, op); 301e5b43573SFam Zheng } else { 302dcfb3bebSFam Zheng bdrv_aio_write_zeroes(s->target, sector_num, op->nb_sectors, 303dcfb3bebSFam Zheng s->unmap ? BDRV_REQ_MAY_UNMAP : 0, 304dcfb3bebSFam Zheng mirror_write_complete, op); 305e5b43573SFam Zheng } 306e5b43573SFam Zheng } 307e5b43573SFam Zheng 308e5b43573SFam Zheng static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) 309e5b43573SFam Zheng { 310e5b43573SFam Zheng BlockDriverState *source = s->common.bs; 3119c83625bSMax Reitz int64_t sector_num, first_chunk; 312e5b43573SFam Zheng uint64_t delay_ns = 0; 313e5b43573SFam Zheng /* At least the first dirty chunk is mirrored in one iteration. */ 314e5b43573SFam Zheng int nb_chunks = 1; 315e5b43573SFam Zheng int64_t end = s->bdev_length / BDRV_SECTOR_SIZE; 316e5b43573SFam Zheng int sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS; 317e5b43573SFam Zheng 318e5b43573SFam Zheng sector_num = hbitmap_iter_next(&s->hbi); 319e5b43573SFam Zheng if (sector_num < 0) { 320e5b43573SFam Zheng bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi); 321e5b43573SFam Zheng sector_num = hbitmap_iter_next(&s->hbi); 322e5b43573SFam Zheng trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); 323e5b43573SFam Zheng assert(sector_num >= 0); 324e5b43573SFam Zheng } 325e5b43573SFam Zheng 3269c83625bSMax Reitz first_chunk = sector_num / sectors_per_chunk; 3279c83625bSMax Reitz while (test_bit(first_chunk, s->in_flight_bitmap)) { 3289c83625bSMax Reitz trace_mirror_yield_in_flight(s, first_chunk, s->in_flight); 3299c83625bSMax Reitz mirror_wait_for_io(s); 3309c83625bSMax Reitz } 3319c83625bSMax Reitz 332e5b43573SFam Zheng /* Find the number of consective dirty chunks following the first dirty 333e5b43573SFam Zheng * one, and wait for in flight requests in them. */ 334e5b43573SFam Zheng while (nb_chunks * sectors_per_chunk < (s->buf_size >> BDRV_SECTOR_BITS)) { 335e5b43573SFam Zheng int64_t hbitmap_next; 336e5b43573SFam Zheng int64_t next_sector = sector_num + nb_chunks * sectors_per_chunk; 337e5b43573SFam Zheng int64_t next_chunk = next_sector / sectors_per_chunk; 338e5b43573SFam Zheng if (next_sector >= end || 339e5b43573SFam Zheng !bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) { 340e5b43573SFam Zheng break; 341e5b43573SFam Zheng } 342e5b43573SFam Zheng if (test_bit(next_chunk, s->in_flight_bitmap)) { 343e5b43573SFam Zheng break; 344e5b43573SFam Zheng } 3459c83625bSMax Reitz 346e5b43573SFam Zheng hbitmap_next = hbitmap_iter_next(&s->hbi); 347f27a2742SMax Reitz if (hbitmap_next > next_sector || hbitmap_next < 0) { 348f27a2742SMax Reitz /* The bitmap iterator's cache is stale, refresh it */ 349f27a2742SMax Reitz bdrv_set_dirty_iter(&s->hbi, next_sector); 350f27a2742SMax Reitz hbitmap_next = hbitmap_iter_next(&s->hbi); 351f27a2742SMax Reitz } 352e5b43573SFam Zheng assert(hbitmap_next == next_sector); 353e5b43573SFam Zheng nb_chunks++; 354e5b43573SFam Zheng } 355e5b43573SFam Zheng 356e5b43573SFam Zheng /* Clear dirty bits before querying the block status, because 357e5b43573SFam Zheng * calling bdrv_get_block_status_above could yield - if some blocks are 358e5b43573SFam Zheng * marked dirty in this window, we need to know. 359e5b43573SFam Zheng */ 360e5b43573SFam Zheng bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, 361e5b43573SFam Zheng nb_chunks * sectors_per_chunk); 362e5b43573SFam Zheng bitmap_set(s->in_flight_bitmap, sector_num / sectors_per_chunk, nb_chunks); 363e5b43573SFam Zheng while (nb_chunks > 0 && sector_num < end) { 364e5b43573SFam Zheng int ret; 365e5b43573SFam Zheng int io_sectors; 366e5b43573SFam Zheng BlockDriverState *file; 367e5b43573SFam Zheng enum MirrorMethod { 368e5b43573SFam Zheng MIRROR_METHOD_COPY, 369e5b43573SFam Zheng MIRROR_METHOD_ZERO, 370e5b43573SFam Zheng MIRROR_METHOD_DISCARD 371e5b43573SFam Zheng } mirror_method = MIRROR_METHOD_COPY; 372e5b43573SFam Zheng 373e5b43573SFam Zheng assert(!(sector_num % sectors_per_chunk)); 374e5b43573SFam Zheng ret = bdrv_get_block_status_above(source, NULL, sector_num, 375e5b43573SFam Zheng nb_chunks * sectors_per_chunk, 376e5b43573SFam Zheng &io_sectors, &file); 377e5b43573SFam Zheng if (ret < 0) { 378e5b43573SFam Zheng io_sectors = nb_chunks * sectors_per_chunk; 379e5b43573SFam Zheng } 380e5b43573SFam Zheng 381e5b43573SFam Zheng io_sectors -= io_sectors % sectors_per_chunk; 382e5b43573SFam Zheng if (io_sectors < sectors_per_chunk) { 383e5b43573SFam Zheng io_sectors = sectors_per_chunk; 384e5b43573SFam Zheng } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { 385e5b43573SFam Zheng int64_t target_sector_num; 386e5b43573SFam Zheng int target_nb_sectors; 387e5b43573SFam Zheng bdrv_round_to_clusters(s->target, sector_num, io_sectors, 388e5b43573SFam Zheng &target_sector_num, &target_nb_sectors); 389e5b43573SFam Zheng if (target_sector_num == sector_num && 390e5b43573SFam Zheng target_nb_sectors == io_sectors) { 391e5b43573SFam Zheng mirror_method = ret & BDRV_BLOCK_ZERO ? 392e5b43573SFam Zheng MIRROR_METHOD_ZERO : 393e5b43573SFam Zheng MIRROR_METHOD_DISCARD; 394e5b43573SFam Zheng } 395e5b43573SFam Zheng } 396e5b43573SFam Zheng 3974150ae60SFam Zheng mirror_clip_sectors(s, sector_num, &io_sectors); 398e5b43573SFam Zheng switch (mirror_method) { 399e5b43573SFam Zheng case MIRROR_METHOD_COPY: 400e5b43573SFam Zheng io_sectors = mirror_do_read(s, sector_num, io_sectors); 401e5b43573SFam Zheng break; 402e5b43573SFam Zheng case MIRROR_METHOD_ZERO: 403e5b43573SFam Zheng mirror_do_zero_or_discard(s, sector_num, io_sectors, false); 404e5b43573SFam Zheng break; 405e5b43573SFam Zheng case MIRROR_METHOD_DISCARD: 406e5b43573SFam Zheng mirror_do_zero_or_discard(s, sector_num, io_sectors, true); 407e5b43573SFam Zheng break; 408e5b43573SFam Zheng default: 409e5b43573SFam Zheng abort(); 410e5b43573SFam Zheng } 411e5b43573SFam Zheng assert(io_sectors); 412e5b43573SFam Zheng sector_num += io_sectors; 4134150ae60SFam Zheng nb_chunks -= DIV_ROUND_UP(io_sectors, sectors_per_chunk); 414e5b43573SFam Zheng delay_ns += ratelimit_calculate_delay(&s->limit, io_sectors); 415dcfb3bebSFam Zheng } 416cc8c9d6cSPaolo Bonzini return delay_ns; 417893f7ebaSPaolo Bonzini } 418b952b558SPaolo Bonzini 419402a4741SPaolo Bonzini static void mirror_free_init(MirrorBlockJob *s) 420402a4741SPaolo Bonzini { 421402a4741SPaolo Bonzini int granularity = s->granularity; 422402a4741SPaolo Bonzini size_t buf_size = s->buf_size; 423402a4741SPaolo Bonzini uint8_t *buf = s->buf; 424402a4741SPaolo Bonzini 425402a4741SPaolo Bonzini assert(s->buf_free_count == 0); 426402a4741SPaolo Bonzini QSIMPLEQ_INIT(&s->buf_free); 427402a4741SPaolo Bonzini while (buf_size != 0) { 428402a4741SPaolo Bonzini MirrorBuffer *cur = (MirrorBuffer *)buf; 429402a4741SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); 430402a4741SPaolo Bonzini s->buf_free_count++; 431402a4741SPaolo Bonzini buf_size -= granularity; 432402a4741SPaolo Bonzini buf += granularity; 433402a4741SPaolo Bonzini } 434402a4741SPaolo Bonzini } 435402a4741SPaolo Bonzini 436bd48bde8SPaolo Bonzini static void mirror_drain(MirrorBlockJob *s) 437bd48bde8SPaolo Bonzini { 438bd48bde8SPaolo Bonzini while (s->in_flight > 0) { 43921cd917fSFam Zheng mirror_wait_for_io(s); 440bd48bde8SPaolo Bonzini } 441893f7ebaSPaolo Bonzini } 442893f7ebaSPaolo Bonzini 4435a7e7a0bSStefan Hajnoczi typedef struct { 4445a7e7a0bSStefan Hajnoczi int ret; 4455a7e7a0bSStefan Hajnoczi } MirrorExitData; 4465a7e7a0bSStefan Hajnoczi 4475a7e7a0bSStefan Hajnoczi static void mirror_exit(BlockJob *job, void *opaque) 4485a7e7a0bSStefan Hajnoczi { 4495a7e7a0bSStefan Hajnoczi MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 4505a7e7a0bSStefan Hajnoczi MirrorExitData *data = opaque; 4515a7e7a0bSStefan Hajnoczi AioContext *replace_aio_context = NULL; 4523f09bfbcSKevin Wolf BlockDriverState *src = s->common.bs; 4533f09bfbcSKevin Wolf 4543f09bfbcSKevin Wolf /* Make sure that the source BDS doesn't go away before we called 4553f09bfbcSKevin Wolf * block_job_completed(). */ 4563f09bfbcSKevin Wolf bdrv_ref(src); 4575a7e7a0bSStefan Hajnoczi 4585a7e7a0bSStefan Hajnoczi if (s->to_replace) { 4595a7e7a0bSStefan Hajnoczi replace_aio_context = bdrv_get_aio_context(s->to_replace); 4605a7e7a0bSStefan Hajnoczi aio_context_acquire(replace_aio_context); 4615a7e7a0bSStefan Hajnoczi } 4625a7e7a0bSStefan Hajnoczi 4635a7e7a0bSStefan Hajnoczi if (s->should_complete && data->ret == 0) { 4645a7e7a0bSStefan Hajnoczi BlockDriverState *to_replace = s->common.bs; 4655a7e7a0bSStefan Hajnoczi if (s->to_replace) { 4665a7e7a0bSStefan Hajnoczi to_replace = s->to_replace; 4675a7e7a0bSStefan Hajnoczi } 46840365552SKevin Wolf 46940365552SKevin Wolf /* This was checked in mirror_start_job(), but meanwhile one of the 47040365552SKevin Wolf * nodes could have been newly attached to a BlockBackend. */ 47140365552SKevin Wolf if (to_replace->blk && s->target->blk) { 47240365552SKevin Wolf error_report("block job: Can't create node with two BlockBackends"); 47340365552SKevin Wolf data->ret = -EINVAL; 47440365552SKevin Wolf goto out; 47540365552SKevin Wolf } 47640365552SKevin Wolf 4775a7e7a0bSStefan Hajnoczi if (bdrv_get_flags(s->target) != bdrv_get_flags(to_replace)) { 4785a7e7a0bSStefan Hajnoczi bdrv_reopen(s->target, bdrv_get_flags(to_replace), NULL); 4795a7e7a0bSStefan Hajnoczi } 4803f09bfbcSKevin Wolf bdrv_replace_in_backing_chain(to_replace, s->target); 4815a7e7a0bSStefan Hajnoczi } 48240365552SKevin Wolf 48340365552SKevin Wolf out: 4845a7e7a0bSStefan Hajnoczi if (s->to_replace) { 4855a7e7a0bSStefan Hajnoczi bdrv_op_unblock_all(s->to_replace, s->replace_blocker); 4865a7e7a0bSStefan Hajnoczi error_free(s->replace_blocker); 4875a7e7a0bSStefan Hajnoczi bdrv_unref(s->to_replace); 4885a7e7a0bSStefan Hajnoczi } 4895a7e7a0bSStefan Hajnoczi if (replace_aio_context) { 4905a7e7a0bSStefan Hajnoczi aio_context_release(replace_aio_context); 4915a7e7a0bSStefan Hajnoczi } 4925a7e7a0bSStefan Hajnoczi g_free(s->replaces); 49310f3cd15SAlberto Garcia bdrv_op_unblock_all(s->target, s->common.blocker); 4945a7e7a0bSStefan Hajnoczi bdrv_unref(s->target); 4955a7e7a0bSStefan Hajnoczi block_job_completed(&s->common, data->ret); 4965a7e7a0bSStefan Hajnoczi g_free(data); 497176c3699SFam Zheng bdrv_drained_end(src); 498*ab27c3b5SFam Zheng if (qemu_get_aio_context() == bdrv_get_aio_context(src)) { 499*ab27c3b5SFam Zheng aio_enable_external(iohandler_get_aio_context()); 500*ab27c3b5SFam Zheng } 5013f09bfbcSKevin Wolf bdrv_unref(src); 5025a7e7a0bSStefan Hajnoczi } 5035a7e7a0bSStefan Hajnoczi 504893f7ebaSPaolo Bonzini static void coroutine_fn mirror_run(void *opaque) 505893f7ebaSPaolo Bonzini { 506893f7ebaSPaolo Bonzini MirrorBlockJob *s = opaque; 5075a7e7a0bSStefan Hajnoczi MirrorExitData *data; 508893f7ebaSPaolo Bonzini BlockDriverState *bs = s->common.bs; 50999900697SFam Zheng int64_t sector_num, end, length; 510bd48bde8SPaolo Bonzini uint64_t last_pause_ns; 511b812f671SPaolo Bonzini BlockDriverInfo bdi; 5121d33936eSJeff Cody char backing_filename[2]; /* we only need 2 characters because we are only 5131d33936eSJeff Cody checking for a NULL string */ 514893f7ebaSPaolo Bonzini int ret = 0; 515893f7ebaSPaolo Bonzini int n; 516e5b43573SFam Zheng int target_cluster_size = BDRV_SECTOR_SIZE; 517893f7ebaSPaolo Bonzini 518893f7ebaSPaolo Bonzini if (block_job_is_cancelled(&s->common)) { 519893f7ebaSPaolo Bonzini goto immediate_exit; 520893f7ebaSPaolo Bonzini } 521893f7ebaSPaolo Bonzini 522b21c7652SMax Reitz s->bdev_length = bdrv_getlength(bs); 523b21c7652SMax Reitz if (s->bdev_length < 0) { 524b21c7652SMax Reitz ret = s->bdev_length; 525373df5b1SFam Zheng goto immediate_exit; 526b21c7652SMax Reitz } else if (s->bdev_length == 0) { 5279e48b025SFam Zheng /* Report BLOCK_JOB_READY and wait for complete. */ 5289e48b025SFam Zheng block_job_event_ready(&s->common); 5299e48b025SFam Zheng s->synced = true; 5309e48b025SFam Zheng while (!block_job_is_cancelled(&s->common) && !s->should_complete) { 5319e48b025SFam Zheng block_job_yield(&s->common); 5329e48b025SFam Zheng } 5339e48b025SFam Zheng s->common.cancelled = false; 5349e48b025SFam Zheng goto immediate_exit; 535893f7ebaSPaolo Bonzini } 536893f7ebaSPaolo Bonzini 537b21c7652SMax Reitz length = DIV_ROUND_UP(s->bdev_length, s->granularity); 538402a4741SPaolo Bonzini s->in_flight_bitmap = bitmap_new(length); 539402a4741SPaolo Bonzini 540b812f671SPaolo Bonzini /* If we have no backing file yet in the destination, we cannot let 541b812f671SPaolo Bonzini * the destination do COW. Instead, we copy sectors around the 542b812f671SPaolo Bonzini * dirty data if needed. We need a bitmap to do that. 543b812f671SPaolo Bonzini */ 544b812f671SPaolo Bonzini bdrv_get_backing_filename(s->target, backing_filename, 545b812f671SPaolo Bonzini sizeof(backing_filename)); 546e5b43573SFam Zheng if (!bdrv_get_info(s->target, &bdi) && bdi.cluster_size) { 547e5b43573SFam Zheng target_cluster_size = bdi.cluster_size; 548c3cc95bdSFam Zheng } 549e5b43573SFam Zheng if (backing_filename[0] && !s->target->backing 550e5b43573SFam Zheng && s->granularity < target_cluster_size) { 551e5b43573SFam Zheng s->buf_size = MAX(s->buf_size, target_cluster_size); 552b812f671SPaolo Bonzini s->cow_bitmap = bitmap_new(length); 553b812f671SPaolo Bonzini } 554e5b43573SFam Zheng s->target_cluster_sectors = target_cluster_size >> BDRV_SECTOR_BITS; 555e5b43573SFam Zheng s->max_iov = MIN(s->common.bs->bl.max_iov, s->target->bl.max_iov); 556b812f671SPaolo Bonzini 557b21c7652SMax Reitz end = s->bdev_length / BDRV_SECTOR_SIZE; 5587504edf4SKevin Wolf s->buf = qemu_try_blockalign(bs, s->buf_size); 5597504edf4SKevin Wolf if (s->buf == NULL) { 5607504edf4SKevin Wolf ret = -ENOMEM; 5617504edf4SKevin Wolf goto immediate_exit; 5627504edf4SKevin Wolf } 5637504edf4SKevin Wolf 564402a4741SPaolo Bonzini mirror_free_init(s); 565893f7ebaSPaolo Bonzini 5664c0cbd6fSFam Zheng last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 56703544a6eSFam Zheng if (!s->is_none_mode) { 568893f7ebaSPaolo Bonzini /* First part, loop on the sectors and initialize the dirty bitmap. */ 5695bc361b8SFam Zheng BlockDriverState *base = s->base; 5705279efebSJeff Cody bool mark_all_dirty = s->base == NULL && !bdrv_has_zero_init(s->target); 5715279efebSJeff Cody 572893f7ebaSPaolo Bonzini for (sector_num = 0; sector_num < end; ) { 57399900697SFam Zheng /* Just to make sure we are not exceeding int limit. */ 57499900697SFam Zheng int nb_sectors = MIN(INT_MAX >> BDRV_SECTOR_BITS, 57599900697SFam Zheng end - sector_num); 5764c0cbd6fSFam Zheng int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 5774c0cbd6fSFam Zheng 5784c0cbd6fSFam Zheng if (now - last_pause_ns > SLICE_TIME) { 5794c0cbd6fSFam Zheng last_pause_ns = now; 5804c0cbd6fSFam Zheng block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, 0); 5814c0cbd6fSFam Zheng } 5824c0cbd6fSFam Zheng 5834c0cbd6fSFam Zheng if (block_job_is_cancelled(&s->common)) { 5844c0cbd6fSFam Zheng goto immediate_exit; 5854c0cbd6fSFam Zheng } 5864c0cbd6fSFam Zheng 58799900697SFam Zheng ret = bdrv_is_allocated_above(bs, base, sector_num, nb_sectors, &n); 588893f7ebaSPaolo Bonzini 589893f7ebaSPaolo Bonzini if (ret < 0) { 590893f7ebaSPaolo Bonzini goto immediate_exit; 591893f7ebaSPaolo Bonzini } 592893f7ebaSPaolo Bonzini 593893f7ebaSPaolo Bonzini assert(n > 0); 5945279efebSJeff Cody if (ret == 1 || mark_all_dirty) { 59520dca810SJohn Snow bdrv_set_dirty_bitmap(s->dirty_bitmap, sector_num, n); 596893f7ebaSPaolo Bonzini } 59799900697SFam Zheng sector_num += n; 598893f7ebaSPaolo Bonzini } 599893f7ebaSPaolo Bonzini } 600893f7ebaSPaolo Bonzini 60120dca810SJohn Snow bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi); 602893f7ebaSPaolo Bonzini for (;;) { 603cc8c9d6cSPaolo Bonzini uint64_t delay_ns = 0; 604893f7ebaSPaolo Bonzini int64_t cnt; 605893f7ebaSPaolo Bonzini bool should_complete; 606893f7ebaSPaolo Bonzini 607bd48bde8SPaolo Bonzini if (s->ret < 0) { 608bd48bde8SPaolo Bonzini ret = s->ret; 609893f7ebaSPaolo Bonzini goto immediate_exit; 610893f7ebaSPaolo Bonzini } 611bd48bde8SPaolo Bonzini 61220dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 613b21c7652SMax Reitz /* s->common.offset contains the number of bytes already processed so 614b21c7652SMax Reitz * far, cnt is the number of dirty sectors remaining and 615b21c7652SMax Reitz * s->sectors_in_flight is the number of sectors currently being 616b21c7652SMax Reitz * processed; together those are the current total operation length */ 617b21c7652SMax Reitz s->common.len = s->common.offset + 618b21c7652SMax Reitz (cnt + s->sectors_in_flight) * BDRV_SECTOR_SIZE; 619bd48bde8SPaolo Bonzini 620bd48bde8SPaolo Bonzini /* Note that even when no rate limit is applied we need to yield 621a7282330SFam Zheng * periodically with no pending I/O so that bdrv_drain_all() returns. 622bd48bde8SPaolo Bonzini * We do so every SLICE_TIME nanoseconds, or when there is an error, 623bd48bde8SPaolo Bonzini * or when the source is clean, whichever comes first. 624bd48bde8SPaolo Bonzini */ 625bc72ad67SAlex Bligh if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - last_pause_ns < SLICE_TIME && 626bd48bde8SPaolo Bonzini s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 627402a4741SPaolo Bonzini if (s->in_flight == MAX_IN_FLIGHT || s->buf_free_count == 0 || 628402a4741SPaolo Bonzini (cnt == 0 && s->in_flight > 0)) { 629402a4741SPaolo Bonzini trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt); 63021cd917fSFam Zheng mirror_wait_for_io(s); 631bd48bde8SPaolo Bonzini continue; 632bd48bde8SPaolo Bonzini } else if (cnt != 0) { 633cc8c9d6cSPaolo Bonzini delay_ns = mirror_iteration(s); 634893f7ebaSPaolo Bonzini } 635cc8c9d6cSPaolo Bonzini } 636893f7ebaSPaolo Bonzini 637893f7ebaSPaolo Bonzini should_complete = false; 638bd48bde8SPaolo Bonzini if (s->in_flight == 0 && cnt == 0) { 639893f7ebaSPaolo Bonzini trace_mirror_before_flush(s); 640893f7ebaSPaolo Bonzini ret = bdrv_flush(s->target); 641893f7ebaSPaolo Bonzini if (ret < 0) { 642a589569fSWenchao Xia if (mirror_error_action(s, false, -ret) == 643a589569fSWenchao Xia BLOCK_ERROR_ACTION_REPORT) { 644893f7ebaSPaolo Bonzini goto immediate_exit; 645893f7ebaSPaolo Bonzini } 646b952b558SPaolo Bonzini } else { 647893f7ebaSPaolo Bonzini /* We're out of the streaming phase. From now on, if the job 648893f7ebaSPaolo Bonzini * is cancelled we will actually complete all pending I/O and 649893f7ebaSPaolo Bonzini * report completion. This way, block-job-cancel will leave 650893f7ebaSPaolo Bonzini * the target in a consistent state. 651893f7ebaSPaolo Bonzini */ 652d63ffd87SPaolo Bonzini if (!s->synced) { 653bcada37bSWenchao Xia block_job_event_ready(&s->common); 654d63ffd87SPaolo Bonzini s->synced = true; 655d63ffd87SPaolo Bonzini } 656d63ffd87SPaolo Bonzini 657d63ffd87SPaolo Bonzini should_complete = s->should_complete || 658d63ffd87SPaolo Bonzini block_job_is_cancelled(&s->common); 65920dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 660893f7ebaSPaolo Bonzini } 661b952b558SPaolo Bonzini } 662893f7ebaSPaolo Bonzini 663893f7ebaSPaolo Bonzini if (cnt == 0 && should_complete) { 664893f7ebaSPaolo Bonzini /* The dirty bitmap is not updated while operations are pending. 665893f7ebaSPaolo Bonzini * If we're about to exit, wait for pending operations before 666893f7ebaSPaolo Bonzini * calling bdrv_get_dirty_count(bs), or we may exit while the 667893f7ebaSPaolo Bonzini * source has dirty data to copy! 668893f7ebaSPaolo Bonzini * 669893f7ebaSPaolo Bonzini * Note that I/O can be submitted by the guest while 670893f7ebaSPaolo Bonzini * mirror_populate runs. 671893f7ebaSPaolo Bonzini */ 672893f7ebaSPaolo Bonzini trace_mirror_before_drain(s, cnt); 67339bf92ddSFam Zheng bdrv_co_drain(bs); 67420dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 675893f7ebaSPaolo Bonzini } 676893f7ebaSPaolo Bonzini 677893f7ebaSPaolo Bonzini ret = 0; 678cc8c9d6cSPaolo Bonzini trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); 679d63ffd87SPaolo Bonzini if (!s->synced) { 6807483d1e5SAlex Bligh block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); 681893f7ebaSPaolo Bonzini if (block_job_is_cancelled(&s->common)) { 682893f7ebaSPaolo Bonzini break; 683893f7ebaSPaolo Bonzini } 684893f7ebaSPaolo Bonzini } else if (!should_complete) { 685bd48bde8SPaolo Bonzini delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); 6867483d1e5SAlex Bligh block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns); 687893f7ebaSPaolo Bonzini } else if (cnt == 0) { 688893f7ebaSPaolo Bonzini /* The two disks are in sync. Exit and report successful 689893f7ebaSPaolo Bonzini * completion. 690893f7ebaSPaolo Bonzini */ 691893f7ebaSPaolo Bonzini assert(QLIST_EMPTY(&bs->tracked_requests)); 692893f7ebaSPaolo Bonzini s->common.cancelled = false; 693893f7ebaSPaolo Bonzini break; 694893f7ebaSPaolo Bonzini } 695bc72ad67SAlex Bligh last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 696893f7ebaSPaolo Bonzini } 697893f7ebaSPaolo Bonzini 698893f7ebaSPaolo Bonzini immediate_exit: 699bd48bde8SPaolo Bonzini if (s->in_flight > 0) { 700bd48bde8SPaolo Bonzini /* We get here only if something went wrong. Either the job failed, 701bd48bde8SPaolo Bonzini * or it was cancelled prematurely so that we do not guarantee that 702bd48bde8SPaolo Bonzini * the target is a copy of the source. 703bd48bde8SPaolo Bonzini */ 704bd48bde8SPaolo Bonzini assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); 705bd48bde8SPaolo Bonzini mirror_drain(s); 706bd48bde8SPaolo Bonzini } 707bd48bde8SPaolo Bonzini 708bd48bde8SPaolo Bonzini assert(s->in_flight == 0); 7097191bf31SMarkus Armbruster qemu_vfree(s->buf); 710b812f671SPaolo Bonzini g_free(s->cow_bitmap); 711402a4741SPaolo Bonzini g_free(s->in_flight_bitmap); 712e4654d2dSFam Zheng bdrv_release_dirty_bitmap(bs, s->dirty_bitmap); 713373340b2SMax Reitz if (s->target->blk) { 714373340b2SMax Reitz blk_iostatus_disable(s->target->blk); 715373340b2SMax Reitz } 7165a7e7a0bSStefan Hajnoczi 7175a7e7a0bSStefan Hajnoczi data = g_malloc(sizeof(*data)); 7185a7e7a0bSStefan Hajnoczi data->ret = ret; 719176c3699SFam Zheng /* Before we switch to target in mirror_exit, make sure data doesn't 720176c3699SFam Zheng * change. */ 721176c3699SFam Zheng bdrv_drained_begin(s->common.bs); 722*ab27c3b5SFam Zheng if (qemu_get_aio_context() == bdrv_get_aio_context(bs)) { 723*ab27c3b5SFam Zheng /* FIXME: virtio host notifiers run on iohandler_ctx, therefore the 724*ab27c3b5SFam Zheng * above bdrv_drained_end isn't enough to quiesce it. This is ugly, we 725*ab27c3b5SFam Zheng * need a block layer API change to achieve this. */ 726*ab27c3b5SFam Zheng aio_disable_external(iohandler_get_aio_context()); 727*ab27c3b5SFam Zheng } 7285a7e7a0bSStefan Hajnoczi block_job_defer_to_main_loop(&s->common, mirror_exit, data); 729893f7ebaSPaolo Bonzini } 730893f7ebaSPaolo Bonzini 731893f7ebaSPaolo Bonzini static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) 732893f7ebaSPaolo Bonzini { 733893f7ebaSPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 734893f7ebaSPaolo Bonzini 735893f7ebaSPaolo Bonzini if (speed < 0) { 736c6bd8c70SMarkus Armbruster error_setg(errp, QERR_INVALID_PARAMETER, "speed"); 737893f7ebaSPaolo Bonzini return; 738893f7ebaSPaolo Bonzini } 739893f7ebaSPaolo Bonzini ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME); 740893f7ebaSPaolo Bonzini } 741893f7ebaSPaolo Bonzini 742b952b558SPaolo Bonzini static void mirror_iostatus_reset(BlockJob *job) 743b952b558SPaolo Bonzini { 744b952b558SPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 745b952b558SPaolo Bonzini 746373340b2SMax Reitz if (s->target->blk) { 747373340b2SMax Reitz blk_iostatus_reset(s->target->blk); 748373340b2SMax Reitz } 749b952b558SPaolo Bonzini } 750b952b558SPaolo Bonzini 751d63ffd87SPaolo Bonzini static void mirror_complete(BlockJob *job, Error **errp) 752d63ffd87SPaolo Bonzini { 753d63ffd87SPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 75434b5d2c6SMax Reitz Error *local_err = NULL; 755d63ffd87SPaolo Bonzini int ret; 756d63ffd87SPaolo Bonzini 757d9b7b057SKevin Wolf ret = bdrv_open_backing_file(s->target, NULL, "backing", &local_err); 758d63ffd87SPaolo Bonzini if (ret < 0) { 75934b5d2c6SMax Reitz error_propagate(errp, local_err); 760d63ffd87SPaolo Bonzini return; 761d63ffd87SPaolo Bonzini } 762d63ffd87SPaolo Bonzini if (!s->synced) { 7638ccb9569SKevin Wolf error_setg(errp, QERR_BLOCK_JOB_NOT_READY, job->id); 764d63ffd87SPaolo Bonzini return; 765d63ffd87SPaolo Bonzini } 766d63ffd87SPaolo Bonzini 76709158f00SBenoît Canet /* check the target bs is not blocked and block all operations on it */ 76809158f00SBenoît Canet if (s->replaces) { 7695a7e7a0bSStefan Hajnoczi AioContext *replace_aio_context; 7705a7e7a0bSStefan Hajnoczi 771e12f3784SWen Congyang s->to_replace = bdrv_find_node(s->replaces); 77209158f00SBenoît Canet if (!s->to_replace) { 773e12f3784SWen Congyang error_setg(errp, "Node name '%s' not found", s->replaces); 77409158f00SBenoît Canet return; 77509158f00SBenoît Canet } 77609158f00SBenoît Canet 7775a7e7a0bSStefan Hajnoczi replace_aio_context = bdrv_get_aio_context(s->to_replace); 7785a7e7a0bSStefan Hajnoczi aio_context_acquire(replace_aio_context); 7795a7e7a0bSStefan Hajnoczi 78009158f00SBenoît Canet error_setg(&s->replace_blocker, 78109158f00SBenoît Canet "block device is in use by block-job-complete"); 78209158f00SBenoît Canet bdrv_op_block_all(s->to_replace, s->replace_blocker); 78309158f00SBenoît Canet bdrv_ref(s->to_replace); 7845a7e7a0bSStefan Hajnoczi 7855a7e7a0bSStefan Hajnoczi aio_context_release(replace_aio_context); 78609158f00SBenoît Canet } 78709158f00SBenoît Canet 788d63ffd87SPaolo Bonzini s->should_complete = true; 789751ebd76SFam Zheng block_job_enter(&s->common); 790d63ffd87SPaolo Bonzini } 791d63ffd87SPaolo Bonzini 7923fc4b10aSFam Zheng static const BlockJobDriver mirror_job_driver = { 793893f7ebaSPaolo Bonzini .instance_size = sizeof(MirrorBlockJob), 79479e14bf7SFam Zheng .job_type = BLOCK_JOB_TYPE_MIRROR, 795893f7ebaSPaolo Bonzini .set_speed = mirror_set_speed, 796b952b558SPaolo Bonzini .iostatus_reset= mirror_iostatus_reset, 797d63ffd87SPaolo Bonzini .complete = mirror_complete, 798893f7ebaSPaolo Bonzini }; 799893f7ebaSPaolo Bonzini 80003544a6eSFam Zheng static const BlockJobDriver commit_active_job_driver = { 80103544a6eSFam Zheng .instance_size = sizeof(MirrorBlockJob), 80203544a6eSFam Zheng .job_type = BLOCK_JOB_TYPE_COMMIT, 80303544a6eSFam Zheng .set_speed = mirror_set_speed, 80403544a6eSFam Zheng .iostatus_reset 80503544a6eSFam Zheng = mirror_iostatus_reset, 80603544a6eSFam Zheng .complete = mirror_complete, 80703544a6eSFam Zheng }; 80803544a6eSFam Zheng 80903544a6eSFam Zheng static void mirror_start_job(BlockDriverState *bs, BlockDriverState *target, 81009158f00SBenoît Canet const char *replaces, 8115fba6c0eSJohn Snow int64_t speed, uint32_t granularity, 81203544a6eSFam Zheng int64_t buf_size, 81303544a6eSFam Zheng BlockdevOnError on_source_error, 814b952b558SPaolo Bonzini BlockdevOnError on_target_error, 8150fc9f8eaSFam Zheng bool unmap, 816097310b5SMarkus Armbruster BlockCompletionFunc *cb, 81703544a6eSFam Zheng void *opaque, Error **errp, 81803544a6eSFam Zheng const BlockJobDriver *driver, 81903544a6eSFam Zheng bool is_none_mode, BlockDriverState *base) 820893f7ebaSPaolo Bonzini { 821893f7ebaSPaolo Bonzini MirrorBlockJob *s; 82240365552SKevin Wolf BlockDriverState *replaced_bs; 823893f7ebaSPaolo Bonzini 824eee13dfeSPaolo Bonzini if (granularity == 0) { 825341ebc2fSJohn Snow granularity = bdrv_get_default_bitmap_granularity(target); 826eee13dfeSPaolo Bonzini } 827eee13dfeSPaolo Bonzini 828eee13dfeSPaolo Bonzini assert ((granularity & (granularity - 1)) == 0); 829eee13dfeSPaolo Bonzini 830b952b558SPaolo Bonzini if ((on_source_error == BLOCKDEV_ON_ERROR_STOP || 831b952b558SPaolo Bonzini on_source_error == BLOCKDEV_ON_ERROR_ENOSPC) && 832373340b2SMax Reitz (!bs->blk || !blk_iostatus_is_enabled(bs->blk))) { 833c6bd8c70SMarkus Armbruster error_setg(errp, QERR_INVALID_PARAMETER, "on-source-error"); 834b952b558SPaolo Bonzini return; 835b952b558SPaolo Bonzini } 836b952b558SPaolo Bonzini 83748ac0a4dSWen Congyang if (buf_size < 0) { 83848ac0a4dSWen Congyang error_setg(errp, "Invalid parameter 'buf-size'"); 83948ac0a4dSWen Congyang return; 84048ac0a4dSWen Congyang } 84148ac0a4dSWen Congyang 84248ac0a4dSWen Congyang if (buf_size == 0) { 84348ac0a4dSWen Congyang buf_size = DEFAULT_MIRROR_BUF_SIZE; 84448ac0a4dSWen Congyang } 8455bc361b8SFam Zheng 84640365552SKevin Wolf /* We can't support this case as long as the block layer can't handle 84740365552SKevin Wolf * multiple BlockBackends per BlockDriverState. */ 84840365552SKevin Wolf if (replaces) { 84940365552SKevin Wolf replaced_bs = bdrv_lookup_bs(replaces, replaces, errp); 85040365552SKevin Wolf if (replaced_bs == NULL) { 85140365552SKevin Wolf return; 85240365552SKevin Wolf } 85340365552SKevin Wolf } else { 85440365552SKevin Wolf replaced_bs = bs; 85540365552SKevin Wolf } 85640365552SKevin Wolf if (replaced_bs->blk && target->blk) { 85740365552SKevin Wolf error_setg(errp, "Can't create node with two BlockBackends"); 85840365552SKevin Wolf return; 85940365552SKevin Wolf } 86040365552SKevin Wolf 86103544a6eSFam Zheng s = block_job_create(driver, bs, speed, cb, opaque, errp); 862893f7ebaSPaolo Bonzini if (!s) { 863893f7ebaSPaolo Bonzini return; 864893f7ebaSPaolo Bonzini } 865893f7ebaSPaolo Bonzini 86609158f00SBenoît Canet s->replaces = g_strdup(replaces); 867b952b558SPaolo Bonzini s->on_source_error = on_source_error; 868b952b558SPaolo Bonzini s->on_target_error = on_target_error; 869893f7ebaSPaolo Bonzini s->target = target; 87003544a6eSFam Zheng s->is_none_mode = is_none_mode; 8715bc361b8SFam Zheng s->base = base; 872eee13dfeSPaolo Bonzini s->granularity = granularity; 87348ac0a4dSWen Congyang s->buf_size = ROUND_UP(buf_size, granularity); 8740fc9f8eaSFam Zheng s->unmap = unmap; 875b812f671SPaolo Bonzini 8760db6e54aSFam Zheng s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); 877b8afb520SFam Zheng if (!s->dirty_bitmap) { 87897031164STing Wang g_free(s->replaces); 87918930ba3SFam Zheng block_job_unref(&s->common); 880b8afb520SFam Zheng return; 881b8afb520SFam Zheng } 88210f3cd15SAlberto Garcia 88310f3cd15SAlberto Garcia bdrv_op_block_all(s->target, s->common.blocker); 88410f3cd15SAlberto Garcia 885373340b2SMax Reitz if (s->target->blk) { 886373340b2SMax Reitz blk_set_on_error(s->target->blk, on_target_error, on_target_error); 887373340b2SMax Reitz blk_iostatus_enable(s->target->blk); 888373340b2SMax Reitz } 889893f7ebaSPaolo Bonzini s->common.co = qemu_coroutine_create(mirror_run); 890893f7ebaSPaolo Bonzini trace_mirror_start(bs, s, s->common.co, opaque); 891893f7ebaSPaolo Bonzini qemu_coroutine_enter(s->common.co, s); 892893f7ebaSPaolo Bonzini } 89303544a6eSFam Zheng 89403544a6eSFam Zheng void mirror_start(BlockDriverState *bs, BlockDriverState *target, 89509158f00SBenoît Canet const char *replaces, 8965fba6c0eSJohn Snow int64_t speed, uint32_t granularity, int64_t buf_size, 89703544a6eSFam Zheng MirrorSyncMode mode, BlockdevOnError on_source_error, 89803544a6eSFam Zheng BlockdevOnError on_target_error, 8990fc9f8eaSFam Zheng bool unmap, 900097310b5SMarkus Armbruster BlockCompletionFunc *cb, 90103544a6eSFam Zheng void *opaque, Error **errp) 90203544a6eSFam Zheng { 90303544a6eSFam Zheng bool is_none_mode; 90403544a6eSFam Zheng BlockDriverState *base; 90503544a6eSFam Zheng 9064b80ab2bSJohn Snow if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { 9074b80ab2bSJohn Snow error_setg(errp, "Sync mode 'incremental' not supported"); 908d58d8453SJohn Snow return; 909d58d8453SJohn Snow } 91003544a6eSFam Zheng is_none_mode = mode == MIRROR_SYNC_MODE_NONE; 911760e0063SKevin Wolf base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; 91209158f00SBenoît Canet mirror_start_job(bs, target, replaces, 91309158f00SBenoît Canet speed, granularity, buf_size, 9140fc9f8eaSFam Zheng on_source_error, on_target_error, unmap, cb, opaque, errp, 91503544a6eSFam Zheng &mirror_job_driver, is_none_mode, base); 91603544a6eSFam Zheng } 91703544a6eSFam Zheng 91803544a6eSFam Zheng void commit_active_start(BlockDriverState *bs, BlockDriverState *base, 91903544a6eSFam Zheng int64_t speed, 92003544a6eSFam Zheng BlockdevOnError on_error, 921097310b5SMarkus Armbruster BlockCompletionFunc *cb, 92203544a6eSFam Zheng void *opaque, Error **errp) 92303544a6eSFam Zheng { 9244da83585SJeff Cody int64_t length, base_length; 9254da83585SJeff Cody int orig_base_flags; 92639a611a3SJeff Cody int ret; 927cc67f4d1SJeff Cody Error *local_err = NULL; 9284da83585SJeff Cody 9294da83585SJeff Cody orig_base_flags = bdrv_get_flags(base); 9304da83585SJeff Cody 93120a63d2cSFam Zheng if (bdrv_reopen(base, bs->open_flags, errp)) { 93220a63d2cSFam Zheng return; 93320a63d2cSFam Zheng } 9344da83585SJeff Cody 9354da83585SJeff Cody length = bdrv_getlength(bs); 9364da83585SJeff Cody if (length < 0) { 93739a611a3SJeff Cody error_setg_errno(errp, -length, 93839a611a3SJeff Cody "Unable to determine length of %s", bs->filename); 9394da83585SJeff Cody goto error_restore_flags; 9404da83585SJeff Cody } 9414da83585SJeff Cody 9424da83585SJeff Cody base_length = bdrv_getlength(base); 9434da83585SJeff Cody if (base_length < 0) { 94439a611a3SJeff Cody error_setg_errno(errp, -base_length, 94539a611a3SJeff Cody "Unable to determine length of %s", base->filename); 9464da83585SJeff Cody goto error_restore_flags; 9474da83585SJeff Cody } 9484da83585SJeff Cody 9494da83585SJeff Cody if (length > base_length) { 95039a611a3SJeff Cody ret = bdrv_truncate(base, length); 95139a611a3SJeff Cody if (ret < 0) { 95239a611a3SJeff Cody error_setg_errno(errp, -ret, 95339a611a3SJeff Cody "Top image %s is larger than base image %s, and " 9544da83585SJeff Cody "resize of base image failed", 9554da83585SJeff Cody bs->filename, base->filename); 9564da83585SJeff Cody goto error_restore_flags; 9574da83585SJeff Cody } 9584da83585SJeff Cody } 9594da83585SJeff Cody 96020a63d2cSFam Zheng bdrv_ref(base); 96109158f00SBenoît Canet mirror_start_job(bs, base, NULL, speed, 0, 0, 9620fc9f8eaSFam Zheng on_error, on_error, false, cb, opaque, &local_err, 96303544a6eSFam Zheng &commit_active_job_driver, false, base); 9640fb6395cSMarkus Armbruster if (local_err) { 965cc67f4d1SJeff Cody error_propagate(errp, local_err); 9664da83585SJeff Cody goto error_restore_flags; 9674da83585SJeff Cody } 9684da83585SJeff Cody 9694da83585SJeff Cody return; 9704da83585SJeff Cody 9714da83585SJeff Cody error_restore_flags: 9724da83585SJeff Cody /* ignore error and errp for bdrv_reopen, because we want to propagate 9734da83585SJeff Cody * the original error */ 9744da83585SJeff Cody bdrv_reopen(base, orig_base_flags, NULL); 9754da83585SJeff Cody return; 97603544a6eSFam Zheng } 977