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 27b436982fSEric Blake #define MAX_IO_BYTES (1 << 20) /* 1 Mb */ 28b436982fSEric Blake #define DEFAULT_MIRROR_BUF_SIZE (MAX_IN_FLIGHT * MAX_IO_BYTES) 29402a4741SPaolo Bonzini 30402a4741SPaolo Bonzini /* The mirroring buffer is a list of granularity-sized chunks. 31402a4741SPaolo Bonzini * Free chunks are organized in a list. 32402a4741SPaolo Bonzini */ 33402a4741SPaolo Bonzini typedef struct MirrorBuffer { 34402a4741SPaolo Bonzini QSIMPLEQ_ENTRY(MirrorBuffer) next; 35402a4741SPaolo Bonzini } MirrorBuffer; 36893f7ebaSPaolo Bonzini 37893f7ebaSPaolo Bonzini typedef struct MirrorBlockJob { 38893f7ebaSPaolo Bonzini BlockJob common; 39893f7ebaSPaolo Bonzini RateLimit limit; 40e253f4b8SKevin Wolf BlockBackend *target; 414ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs; 424ef85a9cSKevin Wolf BlockDriverState *source; 435bc361b8SFam Zheng BlockDriverState *base; 444ef85a9cSKevin Wolf 4509158f00SBenoît Canet /* The name of the graph node to replace */ 4609158f00SBenoît Canet char *replaces; 4709158f00SBenoît Canet /* The BDS to replace */ 4809158f00SBenoît Canet BlockDriverState *to_replace; 4909158f00SBenoît Canet /* Used to block operations on the drive-mirror-replace target */ 5009158f00SBenoît Canet Error *replace_blocker; 5103544a6eSFam Zheng bool is_none_mode; 52274fcceeSMax Reitz BlockMirrorBackingMode backing_mode; 53b952b558SPaolo Bonzini BlockdevOnError on_source_error, on_target_error; 54d63ffd87SPaolo Bonzini bool synced; 55d63ffd87SPaolo Bonzini bool should_complete; 56eee13dfeSPaolo Bonzini int64_t granularity; 57b812f671SPaolo Bonzini size_t buf_size; 58b21c7652SMax Reitz int64_t bdev_length; 59b812f671SPaolo Bonzini unsigned long *cow_bitmap; 60e4654d2dSFam Zheng BdrvDirtyBitmap *dirty_bitmap; 61dc162c8eSFam Zheng BdrvDirtyBitmapIter *dbi; 62893f7ebaSPaolo Bonzini uint8_t *buf; 63402a4741SPaolo Bonzini QSIMPLEQ_HEAD(, MirrorBuffer) buf_free; 64402a4741SPaolo Bonzini int buf_free_count; 65bd48bde8SPaolo Bonzini 6649efb1f5SDenis V. Lunev uint64_t last_pause_ns; 67402a4741SPaolo Bonzini unsigned long *in_flight_bitmap; 68bd48bde8SPaolo Bonzini int in_flight; 69b436982fSEric Blake int64_t bytes_in_flight; 70bd48bde8SPaolo Bonzini int ret; 710fc9f8eaSFam Zheng bool unmap; 72e424aff5SKevin Wolf bool waiting_for_io; 73b436982fSEric Blake int target_cluster_size; 74e5b43573SFam Zheng int max_iov; 7590ab48ebSAnton Nefedov bool initial_zeroing_ongoing; 76893f7ebaSPaolo Bonzini } MirrorBlockJob; 77893f7ebaSPaolo Bonzini 78bd48bde8SPaolo Bonzini typedef struct MirrorOp { 79bd48bde8SPaolo Bonzini MirrorBlockJob *s; 80bd48bde8SPaolo Bonzini QEMUIOVector qiov; 81b436982fSEric Blake int64_t offset; 82b436982fSEric Blake uint64_t bytes; 83bd48bde8SPaolo Bonzini } MirrorOp; 84bd48bde8SPaolo Bonzini 85b952b558SPaolo Bonzini static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read, 86b952b558SPaolo Bonzini int error) 87b952b558SPaolo Bonzini { 88b952b558SPaolo Bonzini s->synced = false; 89b952b558SPaolo Bonzini if (read) { 9081e254dcSKevin Wolf return block_job_error_action(&s->common, s->on_source_error, 9181e254dcSKevin Wolf true, error); 92b952b558SPaolo Bonzini } else { 9381e254dcSKevin Wolf return block_job_error_action(&s->common, s->on_target_error, 9481e254dcSKevin Wolf false, error); 95b952b558SPaolo Bonzini } 96b952b558SPaolo Bonzini } 97b952b558SPaolo Bonzini 98bd48bde8SPaolo Bonzini static void mirror_iteration_done(MirrorOp *op, int ret) 99bd48bde8SPaolo Bonzini { 100bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 101402a4741SPaolo Bonzini struct iovec *iov; 102bd48bde8SPaolo Bonzini int64_t chunk_num; 103b436982fSEric Blake int i, nb_chunks; 104bd48bde8SPaolo Bonzini 105b436982fSEric Blake trace_mirror_iteration_done(s, op->offset, op->bytes, ret); 106bd48bde8SPaolo Bonzini 107bd48bde8SPaolo Bonzini s->in_flight--; 108b436982fSEric Blake s->bytes_in_flight -= op->bytes; 109402a4741SPaolo Bonzini iov = op->qiov.iov; 110402a4741SPaolo Bonzini for (i = 0; i < op->qiov.niov; i++) { 111402a4741SPaolo Bonzini MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base; 112402a4741SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next); 113402a4741SPaolo Bonzini s->buf_free_count++; 114402a4741SPaolo Bonzini } 115402a4741SPaolo Bonzini 116b436982fSEric Blake chunk_num = op->offset / s->granularity; 117b436982fSEric Blake nb_chunks = DIV_ROUND_UP(op->bytes, s->granularity); 118402a4741SPaolo Bonzini bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks); 119b21c7652SMax Reitz if (ret >= 0) { 120b21c7652SMax Reitz if (s->cow_bitmap) { 121bd48bde8SPaolo Bonzini bitmap_set(s->cow_bitmap, chunk_num, nb_chunks); 122bd48bde8SPaolo Bonzini } 12390ab48ebSAnton Nefedov if (!s->initial_zeroing_ongoing) { 124b436982fSEric Blake s->common.offset += op->bytes; 125b21c7652SMax Reitz } 12690ab48ebSAnton Nefedov } 1276df3bf8eSZhang Min qemu_iovec_destroy(&op->qiov); 128c84b3192SPaolo Bonzini g_free(op); 1297b770c72SStefan Hajnoczi 130e424aff5SKevin Wolf if (s->waiting_for_io) { 1310b8b8753SPaolo Bonzini qemu_coroutine_enter(s->common.co); 132bd48bde8SPaolo Bonzini } 1337b770c72SStefan Hajnoczi } 134bd48bde8SPaolo Bonzini 135bd48bde8SPaolo Bonzini static void mirror_write_complete(void *opaque, int ret) 136bd48bde8SPaolo Bonzini { 137bd48bde8SPaolo Bonzini MirrorOp *op = opaque; 138bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 139b9e413ddSPaolo Bonzini 140b9e413ddSPaolo Bonzini aio_context_acquire(blk_get_aio_context(s->common.blk)); 141bd48bde8SPaolo Bonzini if (ret < 0) { 142bd48bde8SPaolo Bonzini BlockErrorAction action; 143bd48bde8SPaolo Bonzini 144e0d7f73eSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); 145bd48bde8SPaolo Bonzini action = mirror_error_action(s, false, -ret); 146a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 147bd48bde8SPaolo Bonzini s->ret = ret; 148bd48bde8SPaolo Bonzini } 149bd48bde8SPaolo Bonzini } 150bd48bde8SPaolo Bonzini mirror_iteration_done(op, ret); 151b9e413ddSPaolo Bonzini aio_context_release(blk_get_aio_context(s->common.blk)); 152bd48bde8SPaolo Bonzini } 153bd48bde8SPaolo Bonzini 154bd48bde8SPaolo Bonzini static void mirror_read_complete(void *opaque, int ret) 155bd48bde8SPaolo Bonzini { 156bd48bde8SPaolo Bonzini MirrorOp *op = opaque; 157bd48bde8SPaolo Bonzini MirrorBlockJob *s = op->s; 158b9e413ddSPaolo Bonzini 159b9e413ddSPaolo Bonzini aio_context_acquire(blk_get_aio_context(s->common.blk)); 160bd48bde8SPaolo Bonzini if (ret < 0) { 161bd48bde8SPaolo Bonzini BlockErrorAction action; 162bd48bde8SPaolo Bonzini 163e0d7f73eSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, op->offset, op->bytes); 164bd48bde8SPaolo Bonzini action = mirror_error_action(s, true, -ret); 165a589569fSWenchao Xia if (action == BLOCK_ERROR_ACTION_REPORT && s->ret >= 0) { 166bd48bde8SPaolo Bonzini s->ret = ret; 167bd48bde8SPaolo Bonzini } 168bd48bde8SPaolo Bonzini 169bd48bde8SPaolo Bonzini mirror_iteration_done(op, ret); 170b9e413ddSPaolo Bonzini } else { 171b436982fSEric Blake blk_aio_pwritev(s->target, op->offset, &op->qiov, 17273698c30SEric Blake 0, mirror_write_complete, op); 173bd48bde8SPaolo Bonzini } 174b9e413ddSPaolo Bonzini aio_context_release(blk_get_aio_context(s->common.blk)); 175b9e413ddSPaolo Bonzini } 176bd48bde8SPaolo Bonzini 177782d97efSEric Blake /* Clip bytes relative to offset to not exceed end-of-file */ 178782d97efSEric Blake static inline int64_t mirror_clip_bytes(MirrorBlockJob *s, 179782d97efSEric Blake int64_t offset, 180782d97efSEric Blake int64_t bytes) 181782d97efSEric Blake { 182782d97efSEric Blake return MIN(bytes, s->bdev_length - offset); 183782d97efSEric Blake } 184782d97efSEric Blake 185782d97efSEric Blake /* Round offset and/or bytes to target cluster if COW is needed, and 186782d97efSEric Blake * return the offset of the adjusted tail against original. */ 187782d97efSEric Blake static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset, 188ae4cc877SEric Blake uint64_t *bytes) 189893f7ebaSPaolo Bonzini { 190e5b43573SFam Zheng bool need_cow; 191e5b43573SFam Zheng int ret = 0; 192782d97efSEric Blake int64_t align_offset = *offset; 1937cfd5275SEric Blake int64_t align_bytes = *bytes; 194782d97efSEric Blake int max_bytes = s->granularity * s->max_iov; 195893f7ebaSPaolo Bonzini 196782d97efSEric Blake need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap); 197782d97efSEric Blake need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity, 198e5b43573SFam Zheng s->cow_bitmap); 199e5b43573SFam Zheng if (need_cow) { 200782d97efSEric Blake bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes, 201782d97efSEric Blake &align_offset, &align_bytes); 2028f0720ecSPaolo Bonzini } 2038f0720ecSPaolo Bonzini 204782d97efSEric Blake if (align_bytes > max_bytes) { 205782d97efSEric Blake align_bytes = max_bytes; 206e5b43573SFam Zheng if (need_cow) { 207782d97efSEric Blake align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size); 208e5b43573SFam Zheng } 209e5b43573SFam Zheng } 210782d97efSEric Blake /* Clipping may result in align_bytes unaligned to chunk boundary, but 2114150ae60SFam Zheng * that doesn't matter because it's already the end of source image. */ 212782d97efSEric Blake align_bytes = mirror_clip_bytes(s, align_offset, align_bytes); 213402a4741SPaolo Bonzini 214782d97efSEric Blake ret = align_offset + align_bytes - (*offset + *bytes); 215782d97efSEric Blake *offset = align_offset; 216782d97efSEric Blake *bytes = align_bytes; 217e5b43573SFam Zheng assert(ret >= 0); 218e5b43573SFam Zheng return ret; 219e5b43573SFam Zheng } 220e5b43573SFam Zheng 22121cd917fSFam Zheng static inline void mirror_wait_for_io(MirrorBlockJob *s) 22221cd917fSFam Zheng { 22321cd917fSFam Zheng assert(!s->waiting_for_io); 22421cd917fSFam Zheng s->waiting_for_io = true; 22521cd917fSFam Zheng qemu_coroutine_yield(); 22621cd917fSFam Zheng s->waiting_for_io = false; 22721cd917fSFam Zheng } 22821cd917fSFam Zheng 229e5b43573SFam Zheng /* Submit async read while handling COW. 230ae4cc877SEric Blake * Returns: The number of bytes copied after and including offset, 231ae4cc877SEric Blake * excluding any bytes copied prior to offset due to alignment. 232ae4cc877SEric Blake * This will be @bytes if no alignment is necessary, or 233ae4cc877SEric Blake * (new_end - offset) if tail is rounded up or down due to 234e5b43573SFam Zheng * alignment or buffer limit. 235402a4741SPaolo Bonzini */ 236ae4cc877SEric Blake static uint64_t mirror_do_read(MirrorBlockJob *s, int64_t offset, 237ae4cc877SEric Blake uint64_t bytes) 238e5b43573SFam Zheng { 239e253f4b8SKevin Wolf BlockBackend *source = s->common.blk; 240ae4cc877SEric Blake int nb_chunks; 241ae4cc877SEric Blake uint64_t ret; 242e5b43573SFam Zheng MirrorOp *op; 243ae4cc877SEric Blake uint64_t max_bytes; 244402a4741SPaolo Bonzini 245ae4cc877SEric Blake max_bytes = s->granularity * s->max_iov; 246e5b43573SFam Zheng 247e5b43573SFam Zheng /* We can only handle as much as buf_size at a time. */ 248ae4cc877SEric Blake bytes = MIN(s->buf_size, MIN(max_bytes, bytes)); 249ae4cc877SEric Blake assert(bytes); 250ae4cc877SEric Blake assert(bytes < BDRV_REQUEST_MAX_BYTES); 251ae4cc877SEric Blake ret = bytes; 252e5b43573SFam Zheng 253e5b43573SFam Zheng if (s->cow_bitmap) { 254ae4cc877SEric Blake ret += mirror_cow_align(s, &offset, &bytes); 255e5b43573SFam Zheng } 256ae4cc877SEric Blake assert(bytes <= s->buf_size); 257ae4cc877SEric Blake /* The offset is granularity-aligned because: 258e5b43573SFam Zheng * 1) Caller passes in aligned values; 259e5b43573SFam Zheng * 2) mirror_cow_align is used only when target cluster is larger. */ 260ae4cc877SEric Blake assert(QEMU_IS_ALIGNED(offset, s->granularity)); 261ae4cc877SEric Blake /* The range is sector-aligned, since bdrv_getlength() rounds up. */ 262ae4cc877SEric Blake assert(QEMU_IS_ALIGNED(bytes, BDRV_SECTOR_SIZE)); 263ae4cc877SEric Blake nb_chunks = DIV_ROUND_UP(bytes, s->granularity); 264e5b43573SFam Zheng 265e5b43573SFam Zheng while (s->buf_free_count < nb_chunks) { 266ae4cc877SEric Blake trace_mirror_yield_in_flight(s, offset, s->in_flight); 26721cd917fSFam Zheng mirror_wait_for_io(s); 268b812f671SPaolo Bonzini } 269b812f671SPaolo Bonzini 270bd48bde8SPaolo Bonzini /* Allocate a MirrorOp that is used as an AIO callback. */ 271c84b3192SPaolo Bonzini op = g_new(MirrorOp, 1); 272bd48bde8SPaolo Bonzini op->s = s; 273ae4cc877SEric Blake op->offset = offset; 274ae4cc877SEric Blake op->bytes = bytes; 275402a4741SPaolo Bonzini 276402a4741SPaolo Bonzini /* Now make a QEMUIOVector taking enough granularity-sized chunks 277402a4741SPaolo Bonzini * from s->buf_free. 278402a4741SPaolo Bonzini */ 279402a4741SPaolo Bonzini qemu_iovec_init(&op->qiov, nb_chunks); 280402a4741SPaolo Bonzini while (nb_chunks-- > 0) { 281402a4741SPaolo Bonzini MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free); 282ae4cc877SEric Blake size_t remaining = bytes - op->qiov.size; 2835a0f6fd5SKevin Wolf 284402a4741SPaolo Bonzini QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next); 285402a4741SPaolo Bonzini s->buf_free_count--; 2865a0f6fd5SKevin Wolf qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining)); 287402a4741SPaolo Bonzini } 288402a4741SPaolo Bonzini 289893f7ebaSPaolo Bonzini /* Copy the dirty cluster. */ 290bd48bde8SPaolo Bonzini s->in_flight++; 291ae4cc877SEric Blake s->bytes_in_flight += bytes; 292ae4cc877SEric Blake trace_mirror_one_iteration(s, offset, bytes); 293dcfb3bebSFam Zheng 294ae4cc877SEric Blake blk_aio_preadv(source, offset, &op->qiov, 0, mirror_read_complete, op); 295e5b43573SFam Zheng return ret; 296e5b43573SFam Zheng } 297e5b43573SFam Zheng 298e5b43573SFam Zheng static void mirror_do_zero_or_discard(MirrorBlockJob *s, 299e6f24193SEric Blake int64_t offset, 300e6f24193SEric Blake uint64_t bytes, 301e5b43573SFam Zheng bool is_discard) 302e5b43573SFam Zheng { 303e5b43573SFam Zheng MirrorOp *op; 304e5b43573SFam Zheng 305e5b43573SFam Zheng /* Allocate a MirrorOp that is used as an AIO callback. The qiov is zeroed 306e5b43573SFam Zheng * so the freeing in mirror_iteration_done is nop. */ 307e5b43573SFam Zheng op = g_new0(MirrorOp, 1); 308e5b43573SFam Zheng op->s = s; 309e6f24193SEric Blake op->offset = offset; 310e6f24193SEric Blake op->bytes = bytes; 311e5b43573SFam Zheng 312e5b43573SFam Zheng s->in_flight++; 313e6f24193SEric Blake s->bytes_in_flight += bytes; 314e5b43573SFam Zheng if (is_discard) { 315e6f24193SEric Blake blk_aio_pdiscard(s->target, offset, 316b436982fSEric Blake op->bytes, mirror_write_complete, op); 317e5b43573SFam Zheng } else { 318e6f24193SEric Blake blk_aio_pwrite_zeroes(s->target, offset, 319b436982fSEric Blake op->bytes, s->unmap ? BDRV_REQ_MAY_UNMAP : 0, 320dcfb3bebSFam Zheng mirror_write_complete, op); 321e5b43573SFam Zheng } 322e5b43573SFam Zheng } 323e5b43573SFam Zheng 324e5b43573SFam Zheng static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s) 325e5b43573SFam Zheng { 3264ef85a9cSKevin Wolf BlockDriverState *source = s->source; 327fb2ef791SEric Blake int64_t offset, first_chunk; 328e5b43573SFam Zheng uint64_t delay_ns = 0; 329e5b43573SFam Zheng /* At least the first dirty chunk is mirrored in one iteration. */ 330e5b43573SFam Zheng int nb_chunks = 1; 3314b5004d9SDenis V. Lunev bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target)); 332b436982fSEric Blake int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES); 333e5b43573SFam Zheng 334b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_lock(s->dirty_bitmap); 335f798184cSEric Blake offset = bdrv_dirty_iter_next(s->dbi); 336fb2ef791SEric Blake if (offset < 0) { 337dc162c8eSFam Zheng bdrv_set_dirty_iter(s->dbi, 0); 338f798184cSEric Blake offset = bdrv_dirty_iter_next(s->dbi); 3399a46dba7SEric Blake trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap)); 340fb2ef791SEric Blake assert(offset >= 0); 341e5b43573SFam Zheng } 342b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_unlock(s->dirty_bitmap); 343e5b43573SFam Zheng 344fb2ef791SEric Blake first_chunk = offset / s->granularity; 3459c83625bSMax Reitz while (test_bit(first_chunk, s->in_flight_bitmap)) { 346fb2ef791SEric Blake trace_mirror_yield_in_flight(s, offset, s->in_flight); 3479c83625bSMax Reitz mirror_wait_for_io(s); 3489c83625bSMax Reitz } 3499c83625bSMax Reitz 350565ac01fSStefan Hajnoczi block_job_pause_point(&s->common); 351565ac01fSStefan Hajnoczi 352e5b43573SFam Zheng /* Find the number of consective dirty chunks following the first dirty 353e5b43573SFam Zheng * one, and wait for in flight requests in them. */ 354b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_lock(s->dirty_bitmap); 355fb2ef791SEric Blake while (nb_chunks * s->granularity < s->buf_size) { 356dc162c8eSFam Zheng int64_t next_dirty; 357fb2ef791SEric Blake int64_t next_offset = offset + nb_chunks * s->granularity; 358fb2ef791SEric Blake int64_t next_chunk = next_offset / s->granularity; 359fb2ef791SEric Blake if (next_offset >= s->bdev_length || 3603b5d4df0SEric Blake !bdrv_get_dirty_locked(source, s->dirty_bitmap, next_offset)) { 361e5b43573SFam Zheng break; 362e5b43573SFam Zheng } 363e5b43573SFam Zheng if (test_bit(next_chunk, s->in_flight_bitmap)) { 364e5b43573SFam Zheng break; 365e5b43573SFam Zheng } 3669c83625bSMax Reitz 367f798184cSEric Blake next_dirty = bdrv_dirty_iter_next(s->dbi); 368fb2ef791SEric Blake if (next_dirty > next_offset || next_dirty < 0) { 369f27a2742SMax Reitz /* The bitmap iterator's cache is stale, refresh it */ 370715a74d8SEric Blake bdrv_set_dirty_iter(s->dbi, next_offset); 371f798184cSEric Blake next_dirty = bdrv_dirty_iter_next(s->dbi); 372f27a2742SMax Reitz } 373fb2ef791SEric Blake assert(next_dirty == next_offset); 374e5b43573SFam Zheng nb_chunks++; 375e5b43573SFam Zheng } 376e5b43573SFam Zheng 377e5b43573SFam Zheng /* Clear dirty bits before querying the block status, because 37831826642SEric Blake * calling bdrv_block_status_above could yield - if some blocks are 379e5b43573SFam Zheng * marked dirty in this window, we need to know. 380e5b43573SFam Zheng */ 381e0d7f73eSEric Blake bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset, 382e0d7f73eSEric Blake nb_chunks * s->granularity); 383b64bd51eSPaolo Bonzini bdrv_dirty_bitmap_unlock(s->dirty_bitmap); 384b64bd51eSPaolo Bonzini 385fb2ef791SEric Blake bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks); 386fb2ef791SEric Blake while (nb_chunks > 0 && offset < s->bdev_length) { 38731826642SEric Blake int ret; 3887cfd5275SEric Blake int64_t io_bytes; 389f3e4ce4aSEric Blake int64_t io_bytes_acct; 390e5b43573SFam Zheng enum MirrorMethod { 391e5b43573SFam Zheng MIRROR_METHOD_COPY, 392e5b43573SFam Zheng MIRROR_METHOD_ZERO, 393e5b43573SFam Zheng MIRROR_METHOD_DISCARD 394e5b43573SFam Zheng } mirror_method = MIRROR_METHOD_COPY; 395e5b43573SFam Zheng 396fb2ef791SEric Blake assert(!(offset % s->granularity)); 39731826642SEric Blake ret = bdrv_block_status_above(source, NULL, offset, 39831826642SEric Blake nb_chunks * s->granularity, 39931826642SEric Blake &io_bytes, NULL, NULL); 400e5b43573SFam Zheng if (ret < 0) { 401fb2ef791SEric Blake io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes); 4020965a41eSVladimir Sementsov-Ogievskiy } else if (ret & BDRV_BLOCK_DATA) { 403fb2ef791SEric Blake io_bytes = MIN(io_bytes, max_io_bytes); 404e5b43573SFam Zheng } 405e5b43573SFam Zheng 406fb2ef791SEric Blake io_bytes -= io_bytes % s->granularity; 407fb2ef791SEric Blake if (io_bytes < s->granularity) { 408fb2ef791SEric Blake io_bytes = s->granularity; 409e5b43573SFam Zheng } else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) { 410fb2ef791SEric Blake int64_t target_offset; 4117cfd5275SEric Blake int64_t target_bytes; 412fb2ef791SEric Blake bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes, 413fb2ef791SEric Blake &target_offset, &target_bytes); 414fb2ef791SEric Blake if (target_offset == offset && 415fb2ef791SEric Blake target_bytes == io_bytes) { 416e5b43573SFam Zheng mirror_method = ret & BDRV_BLOCK_ZERO ? 417e5b43573SFam Zheng MIRROR_METHOD_ZERO : 418e5b43573SFam Zheng MIRROR_METHOD_DISCARD; 419e5b43573SFam Zheng } 420e5b43573SFam Zheng } 421e5b43573SFam Zheng 422cf56a3c6SDenis V. Lunev while (s->in_flight >= MAX_IN_FLIGHT) { 423fb2ef791SEric Blake trace_mirror_yield_in_flight(s, offset, s->in_flight); 424cf56a3c6SDenis V. Lunev mirror_wait_for_io(s); 425cf56a3c6SDenis V. Lunev } 426cf56a3c6SDenis V. Lunev 427dbaa7b57SVladimir Sementsov-Ogievskiy if (s->ret < 0) { 428dbaa7b57SVladimir Sementsov-Ogievskiy return 0; 429dbaa7b57SVladimir Sementsov-Ogievskiy } 430dbaa7b57SVladimir Sementsov-Ogievskiy 431fb2ef791SEric Blake io_bytes = mirror_clip_bytes(s, offset, io_bytes); 432e5b43573SFam Zheng switch (mirror_method) { 433e5b43573SFam Zheng case MIRROR_METHOD_COPY: 434fb2ef791SEric Blake io_bytes = io_bytes_acct = mirror_do_read(s, offset, io_bytes); 435e5b43573SFam Zheng break; 436e5b43573SFam Zheng case MIRROR_METHOD_ZERO: 437e5b43573SFam Zheng case MIRROR_METHOD_DISCARD: 438fb2ef791SEric Blake mirror_do_zero_or_discard(s, offset, io_bytes, 4394b5004d9SDenis V. Lunev mirror_method == MIRROR_METHOD_DISCARD); 4404b5004d9SDenis V. Lunev if (write_zeroes_ok) { 441f3e4ce4aSEric Blake io_bytes_acct = 0; 4424b5004d9SDenis V. Lunev } else { 443fb2ef791SEric Blake io_bytes_acct = io_bytes; 4444b5004d9SDenis V. Lunev } 445e5b43573SFam Zheng break; 446e5b43573SFam Zheng default: 447e5b43573SFam Zheng abort(); 448e5b43573SFam Zheng } 449fb2ef791SEric Blake assert(io_bytes); 450fb2ef791SEric Blake offset += io_bytes; 451fb2ef791SEric Blake nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity); 452f14a39ccSSascha Silbe if (s->common.speed) { 453f3e4ce4aSEric Blake delay_ns = ratelimit_calculate_delay(&s->limit, io_bytes_acct); 454f14a39ccSSascha Silbe } 455dcfb3bebSFam Zheng } 456cc8c9d6cSPaolo Bonzini return delay_ns; 457893f7ebaSPaolo Bonzini } 458b952b558SPaolo Bonzini 459402a4741SPaolo Bonzini static void mirror_free_init(MirrorBlockJob *s) 460402a4741SPaolo Bonzini { 461402a4741SPaolo Bonzini int granularity = s->granularity; 462402a4741SPaolo Bonzini size_t buf_size = s->buf_size; 463402a4741SPaolo Bonzini uint8_t *buf = s->buf; 464402a4741SPaolo Bonzini 465402a4741SPaolo Bonzini assert(s->buf_free_count == 0); 466402a4741SPaolo Bonzini QSIMPLEQ_INIT(&s->buf_free); 467402a4741SPaolo Bonzini while (buf_size != 0) { 468402a4741SPaolo Bonzini MirrorBuffer *cur = (MirrorBuffer *)buf; 469402a4741SPaolo Bonzini QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next); 470402a4741SPaolo Bonzini s->buf_free_count++; 471402a4741SPaolo Bonzini buf_size -= granularity; 472402a4741SPaolo Bonzini buf += granularity; 473402a4741SPaolo Bonzini } 474402a4741SPaolo Bonzini } 475402a4741SPaolo Bonzini 476bae8196dSPaolo Bonzini /* This is also used for the .pause callback. There is no matching 477bae8196dSPaolo Bonzini * mirror_resume() because mirror_run() will begin iterating again 478bae8196dSPaolo Bonzini * when the job is resumed. 479bae8196dSPaolo Bonzini */ 480bae8196dSPaolo Bonzini static void mirror_wait_for_all_io(MirrorBlockJob *s) 481bd48bde8SPaolo Bonzini { 482bd48bde8SPaolo Bonzini while (s->in_flight > 0) { 48321cd917fSFam Zheng mirror_wait_for_io(s); 484bd48bde8SPaolo Bonzini } 485893f7ebaSPaolo Bonzini } 486893f7ebaSPaolo Bonzini 4875a7e7a0bSStefan Hajnoczi typedef struct { 4885a7e7a0bSStefan Hajnoczi int ret; 4895a7e7a0bSStefan Hajnoczi } MirrorExitData; 4905a7e7a0bSStefan Hajnoczi 4915a7e7a0bSStefan Hajnoczi static void mirror_exit(BlockJob *job, void *opaque) 4925a7e7a0bSStefan Hajnoczi { 4935a7e7a0bSStefan Hajnoczi MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 4945a7e7a0bSStefan Hajnoczi MirrorExitData *data = opaque; 4955a7e7a0bSStefan Hajnoczi AioContext *replace_aio_context = NULL; 4964ef85a9cSKevin Wolf BlockDriverState *src = s->source; 497e253f4b8SKevin Wolf BlockDriverState *target_bs = blk_bs(s->target); 4984ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs = s->mirror_top_bs; 49912fa4af6SKevin Wolf Error *local_err = NULL; 5003f09bfbcSKevin Wolf 5012119882cSPaolo Bonzini bdrv_release_dirty_bitmap(src, s->dirty_bitmap); 5022119882cSPaolo Bonzini 5033f09bfbcSKevin Wolf /* Make sure that the source BDS doesn't go away before we called 5043f09bfbcSKevin Wolf * block_job_completed(). */ 5053f09bfbcSKevin Wolf bdrv_ref(src); 5064ef85a9cSKevin Wolf bdrv_ref(mirror_top_bs); 5077d9fcb39SKevin Wolf bdrv_ref(target_bs); 5087d9fcb39SKevin Wolf 5097d9fcb39SKevin Wolf /* Remove target parent that still uses BLK_PERM_WRITE/RESIZE before 5107d9fcb39SKevin Wolf * inserting target_bs at s->to_replace, where we might not be able to get 51163c8ef28SKevin Wolf * these permissions. 51263c8ef28SKevin Wolf * 51363c8ef28SKevin Wolf * Note that blk_unref() alone doesn't necessarily drop permissions because 51463c8ef28SKevin Wolf * we might be running nested inside mirror_drain(), which takes an extra 51563c8ef28SKevin Wolf * reference, so use an explicit blk_set_perm() first. */ 51663c8ef28SKevin Wolf blk_set_perm(s->target, 0, BLK_PERM_ALL, &error_abort); 5177d9fcb39SKevin Wolf blk_unref(s->target); 5187d9fcb39SKevin Wolf s->target = NULL; 5194ef85a9cSKevin Wolf 5204ef85a9cSKevin Wolf /* We don't access the source any more. Dropping any WRITE/RESIZE is 5214ef85a9cSKevin Wolf * required before it could become a backing file of target_bs. */ 5224ef85a9cSKevin Wolf bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, 5234ef85a9cSKevin Wolf &error_abort); 5244ef85a9cSKevin Wolf if (s->backing_mode == MIRROR_SOURCE_BACKING_CHAIN) { 5254ef85a9cSKevin Wolf BlockDriverState *backing = s->is_none_mode ? src : s->base; 5264ef85a9cSKevin Wolf if (backing_bs(target_bs) != backing) { 52712fa4af6SKevin Wolf bdrv_set_backing_hd(target_bs, backing, &local_err); 52812fa4af6SKevin Wolf if (local_err) { 52912fa4af6SKevin Wolf error_report_err(local_err); 53012fa4af6SKevin Wolf data->ret = -EPERM; 53112fa4af6SKevin Wolf } 5324ef85a9cSKevin Wolf } 5334ef85a9cSKevin Wolf } 5345a7e7a0bSStefan Hajnoczi 5355a7e7a0bSStefan Hajnoczi if (s->to_replace) { 5365a7e7a0bSStefan Hajnoczi replace_aio_context = bdrv_get_aio_context(s->to_replace); 5375a7e7a0bSStefan Hajnoczi aio_context_acquire(replace_aio_context); 5385a7e7a0bSStefan Hajnoczi } 5395a7e7a0bSStefan Hajnoczi 5405a7e7a0bSStefan Hajnoczi if (s->should_complete && data->ret == 0) { 541e253f4b8SKevin Wolf BlockDriverState *to_replace = src; 5425a7e7a0bSStefan Hajnoczi if (s->to_replace) { 5435a7e7a0bSStefan Hajnoczi to_replace = s->to_replace; 5445a7e7a0bSStefan Hajnoczi } 54540365552SKevin Wolf 546e253f4b8SKevin Wolf if (bdrv_get_flags(target_bs) != bdrv_get_flags(to_replace)) { 547e253f4b8SKevin Wolf bdrv_reopen(target_bs, bdrv_get_flags(to_replace), NULL); 5485a7e7a0bSStefan Hajnoczi } 549b8804815SKevin Wolf 550b8804815SKevin Wolf /* The mirror job has no requests in flight any more, but we need to 551b8804815SKevin Wolf * drain potential other users of the BDS before changing the graph. */ 552e253f4b8SKevin Wolf bdrv_drained_begin(target_bs); 5535fe31c25SKevin Wolf bdrv_replace_node(to_replace, target_bs, &local_err); 554e253f4b8SKevin Wolf bdrv_drained_end(target_bs); 5555fe31c25SKevin Wolf if (local_err) { 5565fe31c25SKevin Wolf error_report_err(local_err); 5575fe31c25SKevin Wolf data->ret = -EPERM; 5585fe31c25SKevin Wolf } 5595a7e7a0bSStefan Hajnoczi } 5605a7e7a0bSStefan Hajnoczi if (s->to_replace) { 5615a7e7a0bSStefan Hajnoczi bdrv_op_unblock_all(s->to_replace, s->replace_blocker); 5625a7e7a0bSStefan Hajnoczi error_free(s->replace_blocker); 5635a7e7a0bSStefan Hajnoczi bdrv_unref(s->to_replace); 5645a7e7a0bSStefan Hajnoczi } 5655a7e7a0bSStefan Hajnoczi if (replace_aio_context) { 5665a7e7a0bSStefan Hajnoczi aio_context_release(replace_aio_context); 5675a7e7a0bSStefan Hajnoczi } 5685a7e7a0bSStefan Hajnoczi g_free(s->replaces); 5697d9fcb39SKevin Wolf bdrv_unref(target_bs); 5704ef85a9cSKevin Wolf 5714ef85a9cSKevin Wolf /* Remove the mirror filter driver from the graph. Before this, get rid of 5724ef85a9cSKevin Wolf * the blockers on the intermediate nodes so that the resulting state is 5730bf74767SKevin Wolf * valid. Also give up permissions on mirror_top_bs->backing, which might 5740bf74767SKevin Wolf * block the removal. */ 5754ef85a9cSKevin Wolf block_job_remove_all_bdrv(job); 576c1cef672SFam Zheng bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, 577c1cef672SFam Zheng &error_abort); 5785fe31c25SKevin Wolf bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); 5794ef85a9cSKevin Wolf 5804ef85a9cSKevin Wolf /* We just changed the BDS the job BB refers to (with either or both of the 5815fe31c25SKevin Wolf * bdrv_replace_node() calls), so switch the BB back so the cleanup does 5825fe31c25SKevin Wolf * the right thing. We don't need any permissions any more now. */ 5834ef85a9cSKevin Wolf blk_remove_bs(job->blk); 5844ef85a9cSKevin Wolf blk_set_perm(job->blk, 0, BLK_PERM_ALL, &error_abort); 5854ef85a9cSKevin Wolf blk_insert_bs(job->blk, mirror_top_bs, &error_abort); 5864ef85a9cSKevin Wolf 5875a7e7a0bSStefan Hajnoczi block_job_completed(&s->common, data->ret); 5884ef85a9cSKevin Wolf 5895a7e7a0bSStefan Hajnoczi g_free(data); 590176c3699SFam Zheng bdrv_drained_end(src); 5914ef85a9cSKevin Wolf bdrv_unref(mirror_top_bs); 5923f09bfbcSKevin Wolf bdrv_unref(src); 5935a7e7a0bSStefan Hajnoczi } 5945a7e7a0bSStefan Hajnoczi 59549efb1f5SDenis V. Lunev static void mirror_throttle(MirrorBlockJob *s) 59649efb1f5SDenis V. Lunev { 59749efb1f5SDenis V. Lunev int64_t now = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 59849efb1f5SDenis V. Lunev 59949efb1f5SDenis V. Lunev if (now - s->last_pause_ns > SLICE_TIME) { 60049efb1f5SDenis V. Lunev s->last_pause_ns = now; 601*5bf1d5a7SPaolo Bonzini block_job_sleep_ns(&s->common, 0); 60249efb1f5SDenis V. Lunev } else { 60349efb1f5SDenis V. Lunev block_job_pause_point(&s->common); 60449efb1f5SDenis V. Lunev } 60549efb1f5SDenis V. Lunev } 60649efb1f5SDenis V. Lunev 607c0b363adSDenis V. Lunev static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s) 608c0b363adSDenis V. Lunev { 60923ca459aSEric Blake int64_t offset; 610c0b363adSDenis V. Lunev BlockDriverState *base = s->base; 6114ef85a9cSKevin Wolf BlockDriverState *bs = s->source; 612c0b363adSDenis V. Lunev BlockDriverState *target_bs = blk_bs(s->target); 61323ca459aSEric Blake int ret; 61451b0a488SEric Blake int64_t count; 615c0b363adSDenis V. Lunev 616b7d5062cSDenis V. Lunev if (base == NULL && !bdrv_has_zero_init(target_bs)) { 617c7c2769cSDenis V. Lunev if (!bdrv_can_write_zeroes_with_unmap(target_bs)) { 618e0d7f73eSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, 0, s->bdev_length); 619b7d5062cSDenis V. Lunev return 0; 620b7d5062cSDenis V. Lunev } 621b7d5062cSDenis V. Lunev 62290ab48ebSAnton Nefedov s->initial_zeroing_ongoing = true; 62323ca459aSEric Blake for (offset = 0; offset < s->bdev_length; ) { 62423ca459aSEric Blake int bytes = MIN(s->bdev_length - offset, 62523ca459aSEric Blake QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); 626c7c2769cSDenis V. Lunev 627c7c2769cSDenis V. Lunev mirror_throttle(s); 628c7c2769cSDenis V. Lunev 629c7c2769cSDenis V. Lunev if (block_job_is_cancelled(&s->common)) { 63090ab48ebSAnton Nefedov s->initial_zeroing_ongoing = false; 631c7c2769cSDenis V. Lunev return 0; 632c7c2769cSDenis V. Lunev } 633c7c2769cSDenis V. Lunev 634c7c2769cSDenis V. Lunev if (s->in_flight >= MAX_IN_FLIGHT) { 63567adf4b3SEric Blake trace_mirror_yield(s, UINT64_MAX, s->buf_free_count, 63667adf4b3SEric Blake s->in_flight); 637c7c2769cSDenis V. Lunev mirror_wait_for_io(s); 638c7c2769cSDenis V. Lunev continue; 639c7c2769cSDenis V. Lunev } 640c7c2769cSDenis V. Lunev 64123ca459aSEric Blake mirror_do_zero_or_discard(s, offset, bytes, false); 64223ca459aSEric Blake offset += bytes; 643c7c2769cSDenis V. Lunev } 644c7c2769cSDenis V. Lunev 645bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 64690ab48ebSAnton Nefedov s->initial_zeroing_ongoing = false; 647c7c2769cSDenis V. Lunev } 648c7c2769cSDenis V. Lunev 649c0b363adSDenis V. Lunev /* First part, loop on the sectors and initialize the dirty bitmap. */ 65023ca459aSEric Blake for (offset = 0; offset < s->bdev_length; ) { 651c0b363adSDenis V. Lunev /* Just to make sure we are not exceeding int limit. */ 65223ca459aSEric Blake int bytes = MIN(s->bdev_length - offset, 65323ca459aSEric Blake QEMU_ALIGN_DOWN(INT_MAX, s->granularity)); 654c0b363adSDenis V. Lunev 655c0b363adSDenis V. Lunev mirror_throttle(s); 656c0b363adSDenis V. Lunev 657c0b363adSDenis V. Lunev if (block_job_is_cancelled(&s->common)) { 658c0b363adSDenis V. Lunev return 0; 659c0b363adSDenis V. Lunev } 660c0b363adSDenis V. Lunev 66123ca459aSEric Blake ret = bdrv_is_allocated_above(bs, base, offset, bytes, &count); 662c0b363adSDenis V. Lunev if (ret < 0) { 663c0b363adSDenis V. Lunev return ret; 664c0b363adSDenis V. Lunev } 665c0b363adSDenis V. Lunev 66623ca459aSEric Blake assert(count); 667b7d5062cSDenis V. Lunev if (ret == 1) { 66823ca459aSEric Blake bdrv_set_dirty_bitmap(s->dirty_bitmap, offset, count); 669c0b363adSDenis V. Lunev } 67023ca459aSEric Blake offset += count; 671c0b363adSDenis V. Lunev } 672c0b363adSDenis V. Lunev return 0; 673c0b363adSDenis V. Lunev } 674c0b363adSDenis V. Lunev 675bdffb31dSPaolo Bonzini /* Called when going out of the streaming phase to flush the bulk of the 676bdffb31dSPaolo Bonzini * data to the medium, or just before completing. 677bdffb31dSPaolo Bonzini */ 678bdffb31dSPaolo Bonzini static int mirror_flush(MirrorBlockJob *s) 679bdffb31dSPaolo Bonzini { 680bdffb31dSPaolo Bonzini int ret = blk_flush(s->target); 681bdffb31dSPaolo Bonzini if (ret < 0) { 682bdffb31dSPaolo Bonzini if (mirror_error_action(s, false, -ret) == BLOCK_ERROR_ACTION_REPORT) { 683bdffb31dSPaolo Bonzini s->ret = ret; 684bdffb31dSPaolo Bonzini } 685bdffb31dSPaolo Bonzini } 686bdffb31dSPaolo Bonzini return ret; 687bdffb31dSPaolo Bonzini } 688bdffb31dSPaolo Bonzini 689893f7ebaSPaolo Bonzini static void coroutine_fn mirror_run(void *opaque) 690893f7ebaSPaolo Bonzini { 691893f7ebaSPaolo Bonzini MirrorBlockJob *s = opaque; 6925a7e7a0bSStefan Hajnoczi MirrorExitData *data; 6934ef85a9cSKevin Wolf BlockDriverState *bs = s->source; 694e253f4b8SKevin Wolf BlockDriverState *target_bs = blk_bs(s->target); 6959a0cec66SPaolo Bonzini bool need_drain = true; 696c0b363adSDenis V. Lunev int64_t length; 697b812f671SPaolo Bonzini BlockDriverInfo bdi; 6981d33936eSJeff Cody char backing_filename[2]; /* we only need 2 characters because we are only 6991d33936eSJeff Cody checking for a NULL string */ 700893f7ebaSPaolo Bonzini int ret = 0; 701893f7ebaSPaolo Bonzini 702893f7ebaSPaolo Bonzini if (block_job_is_cancelled(&s->common)) { 703893f7ebaSPaolo Bonzini goto immediate_exit; 704893f7ebaSPaolo Bonzini } 705893f7ebaSPaolo Bonzini 706b21c7652SMax Reitz s->bdev_length = bdrv_getlength(bs); 707b21c7652SMax Reitz if (s->bdev_length < 0) { 708b21c7652SMax Reitz ret = s->bdev_length; 709373df5b1SFam Zheng goto immediate_exit; 710becc347eSKevin Wolf } 711becc347eSKevin Wolf 712becc347eSKevin Wolf /* Active commit must resize the base image if its size differs from the 713becc347eSKevin Wolf * active layer. */ 714becc347eSKevin Wolf if (s->base == blk_bs(s->target)) { 715becc347eSKevin Wolf int64_t base_length; 716becc347eSKevin Wolf 717becc347eSKevin Wolf base_length = blk_getlength(s->target); 718becc347eSKevin Wolf if (base_length < 0) { 719becc347eSKevin Wolf ret = base_length; 720becc347eSKevin Wolf goto immediate_exit; 721becc347eSKevin Wolf } 722becc347eSKevin Wolf 723becc347eSKevin Wolf if (s->bdev_length > base_length) { 7243a691c50SMax Reitz ret = blk_truncate(s->target, s->bdev_length, PREALLOC_MODE_OFF, 7253a691c50SMax Reitz NULL); 726becc347eSKevin Wolf if (ret < 0) { 727becc347eSKevin Wolf goto immediate_exit; 728becc347eSKevin Wolf } 729becc347eSKevin Wolf } 730becc347eSKevin Wolf } 731becc347eSKevin Wolf 732becc347eSKevin Wolf if (s->bdev_length == 0) { 7339e48b025SFam Zheng /* Report BLOCK_JOB_READY and wait for complete. */ 7349e48b025SFam Zheng block_job_event_ready(&s->common); 7359e48b025SFam Zheng s->synced = true; 7369e48b025SFam Zheng while (!block_job_is_cancelled(&s->common) && !s->should_complete) { 7379e48b025SFam Zheng block_job_yield(&s->common); 7389e48b025SFam Zheng } 7399e48b025SFam Zheng s->common.cancelled = false; 7409e48b025SFam Zheng goto immediate_exit; 741893f7ebaSPaolo Bonzini } 742893f7ebaSPaolo Bonzini 743b21c7652SMax Reitz length = DIV_ROUND_UP(s->bdev_length, s->granularity); 744402a4741SPaolo Bonzini s->in_flight_bitmap = bitmap_new(length); 745402a4741SPaolo Bonzini 746b812f671SPaolo Bonzini /* If we have no backing file yet in the destination, we cannot let 747b812f671SPaolo Bonzini * the destination do COW. Instead, we copy sectors around the 748b812f671SPaolo Bonzini * dirty data if needed. We need a bitmap to do that. 749b812f671SPaolo Bonzini */ 750e253f4b8SKevin Wolf bdrv_get_backing_filename(target_bs, backing_filename, 751b812f671SPaolo Bonzini sizeof(backing_filename)); 752e253f4b8SKevin Wolf if (!bdrv_get_info(target_bs, &bdi) && bdi.cluster_size) { 753b436982fSEric Blake s->target_cluster_size = bdi.cluster_size; 754b436982fSEric Blake } else { 755b436982fSEric Blake s->target_cluster_size = BDRV_SECTOR_SIZE; 756c3cc95bdSFam Zheng } 757b436982fSEric Blake if (backing_filename[0] && !target_bs->backing && 758b436982fSEric Blake s->granularity < s->target_cluster_size) { 759b436982fSEric Blake s->buf_size = MAX(s->buf_size, s->target_cluster_size); 760b812f671SPaolo Bonzini s->cow_bitmap = bitmap_new(length); 761b812f671SPaolo Bonzini } 762e253f4b8SKevin Wolf s->max_iov = MIN(bs->bl.max_iov, target_bs->bl.max_iov); 763b812f671SPaolo Bonzini 7647504edf4SKevin Wolf s->buf = qemu_try_blockalign(bs, s->buf_size); 7657504edf4SKevin Wolf if (s->buf == NULL) { 7667504edf4SKevin Wolf ret = -ENOMEM; 7677504edf4SKevin Wolf goto immediate_exit; 7687504edf4SKevin Wolf } 7697504edf4SKevin Wolf 770402a4741SPaolo Bonzini mirror_free_init(s); 771893f7ebaSPaolo Bonzini 77249efb1f5SDenis V. Lunev s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 77303544a6eSFam Zheng if (!s->is_none_mode) { 774c0b363adSDenis V. Lunev ret = mirror_dirty_init(s); 775c0b363adSDenis V. Lunev if (ret < 0 || block_job_is_cancelled(&s->common)) { 7764c0cbd6fSFam Zheng goto immediate_exit; 7774c0cbd6fSFam Zheng } 778893f7ebaSPaolo Bonzini } 779893f7ebaSPaolo Bonzini 780dc162c8eSFam Zheng assert(!s->dbi); 781715a74d8SEric Blake s->dbi = bdrv_dirty_iter_new(s->dirty_bitmap); 782893f7ebaSPaolo Bonzini for (;;) { 783cc8c9d6cSPaolo Bonzini uint64_t delay_ns = 0; 78449efb1f5SDenis V. Lunev int64_t cnt, delta; 785893f7ebaSPaolo Bonzini bool should_complete; 786893f7ebaSPaolo Bonzini 787bd48bde8SPaolo Bonzini if (s->ret < 0) { 788bd48bde8SPaolo Bonzini ret = s->ret; 789893f7ebaSPaolo Bonzini goto immediate_exit; 790893f7ebaSPaolo Bonzini } 791bd48bde8SPaolo Bonzini 792565ac01fSStefan Hajnoczi block_job_pause_point(&s->common); 793565ac01fSStefan Hajnoczi 79420dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 795b21c7652SMax Reitz /* s->common.offset contains the number of bytes already processed so 7969a46dba7SEric Blake * far, cnt is the number of dirty bytes remaining and 797b436982fSEric Blake * s->bytes_in_flight is the number of bytes currently being 798b21c7652SMax Reitz * processed; together those are the current total operation length */ 7999a46dba7SEric Blake s->common.len = s->common.offset + s->bytes_in_flight + cnt; 800bd48bde8SPaolo Bonzini 801bd48bde8SPaolo Bonzini /* Note that even when no rate limit is applied we need to yield 802a7282330SFam Zheng * periodically with no pending I/O so that bdrv_drain_all() returns. 803bd48bde8SPaolo Bonzini * We do so every SLICE_TIME nanoseconds, or when there is an error, 804bd48bde8SPaolo Bonzini * or when the source is clean, whichever comes first. 805bd48bde8SPaolo Bonzini */ 80649efb1f5SDenis V. Lunev delta = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - s->last_pause_ns; 80749efb1f5SDenis V. Lunev if (delta < SLICE_TIME && 808bd48bde8SPaolo Bonzini s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 809cf56a3c6SDenis V. Lunev if (s->in_flight >= MAX_IN_FLIGHT || s->buf_free_count == 0 || 810402a4741SPaolo Bonzini (cnt == 0 && s->in_flight > 0)) { 8119a46dba7SEric Blake trace_mirror_yield(s, cnt, s->buf_free_count, s->in_flight); 81221cd917fSFam Zheng mirror_wait_for_io(s); 813bd48bde8SPaolo Bonzini continue; 814bd48bde8SPaolo Bonzini } else if (cnt != 0) { 815cc8c9d6cSPaolo Bonzini delay_ns = mirror_iteration(s); 816893f7ebaSPaolo Bonzini } 817cc8c9d6cSPaolo Bonzini } 818893f7ebaSPaolo Bonzini 819893f7ebaSPaolo Bonzini should_complete = false; 820bd48bde8SPaolo Bonzini if (s->in_flight == 0 && cnt == 0) { 821893f7ebaSPaolo Bonzini trace_mirror_before_flush(s); 822bdffb31dSPaolo Bonzini if (!s->synced) { 823bdffb31dSPaolo Bonzini if (mirror_flush(s) < 0) { 824bdffb31dSPaolo Bonzini /* Go check s->ret. */ 825bdffb31dSPaolo Bonzini continue; 826893f7ebaSPaolo Bonzini } 827893f7ebaSPaolo Bonzini /* We're out of the streaming phase. From now on, if the job 828893f7ebaSPaolo Bonzini * is cancelled we will actually complete all pending I/O and 829893f7ebaSPaolo Bonzini * report completion. This way, block-job-cancel will leave 830893f7ebaSPaolo Bonzini * the target in a consistent state. 831893f7ebaSPaolo Bonzini */ 832bcada37bSWenchao Xia block_job_event_ready(&s->common); 833d63ffd87SPaolo Bonzini s->synced = true; 834d63ffd87SPaolo Bonzini } 835d63ffd87SPaolo Bonzini 836d63ffd87SPaolo Bonzini should_complete = s->should_complete || 837d63ffd87SPaolo Bonzini block_job_is_cancelled(&s->common); 83820dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 839893f7ebaSPaolo Bonzini } 840893f7ebaSPaolo Bonzini 841893f7ebaSPaolo Bonzini if (cnt == 0 && should_complete) { 842893f7ebaSPaolo Bonzini /* The dirty bitmap is not updated while operations are pending. 843893f7ebaSPaolo Bonzini * If we're about to exit, wait for pending operations before 844893f7ebaSPaolo Bonzini * calling bdrv_get_dirty_count(bs), or we may exit while the 845893f7ebaSPaolo Bonzini * source has dirty data to copy! 846893f7ebaSPaolo Bonzini * 847893f7ebaSPaolo Bonzini * Note that I/O can be submitted by the guest while 8489a0cec66SPaolo Bonzini * mirror_populate runs, so pause it now. Before deciding 8499a0cec66SPaolo Bonzini * whether to switch to target check one last time if I/O has 8509a0cec66SPaolo Bonzini * come in the meanwhile, and if not flush the data to disk. 851893f7ebaSPaolo Bonzini */ 8529a46dba7SEric Blake trace_mirror_before_drain(s, cnt); 8539a0cec66SPaolo Bonzini 8549a0cec66SPaolo Bonzini bdrv_drained_begin(bs); 85520dca810SJohn Snow cnt = bdrv_get_dirty_count(s->dirty_bitmap); 856bdffb31dSPaolo Bonzini if (cnt > 0 || mirror_flush(s) < 0) { 8579a0cec66SPaolo Bonzini bdrv_drained_end(bs); 8589a0cec66SPaolo Bonzini continue; 8599a0cec66SPaolo Bonzini } 8609a0cec66SPaolo Bonzini 8619a0cec66SPaolo Bonzini /* The two disks are in sync. Exit and report successful 8629a0cec66SPaolo Bonzini * completion. 8639a0cec66SPaolo Bonzini */ 8649a0cec66SPaolo Bonzini assert(QLIST_EMPTY(&bs->tracked_requests)); 8659a0cec66SPaolo Bonzini s->common.cancelled = false; 8669a0cec66SPaolo Bonzini need_drain = false; 8679a0cec66SPaolo Bonzini break; 868893f7ebaSPaolo Bonzini } 869893f7ebaSPaolo Bonzini 870893f7ebaSPaolo Bonzini ret = 0; 8719a46dba7SEric Blake trace_mirror_before_sleep(s, cnt, s->synced, delay_ns); 872d63ffd87SPaolo Bonzini if (!s->synced) { 873*5bf1d5a7SPaolo Bonzini block_job_sleep_ns(&s->common, delay_ns); 874893f7ebaSPaolo Bonzini if (block_job_is_cancelled(&s->common)) { 875893f7ebaSPaolo Bonzini break; 876893f7ebaSPaolo Bonzini } 877893f7ebaSPaolo Bonzini } else if (!should_complete) { 878bd48bde8SPaolo Bonzini delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0); 879*5bf1d5a7SPaolo Bonzini block_job_sleep_ns(&s->common, delay_ns); 880893f7ebaSPaolo Bonzini } 88149efb1f5SDenis V. Lunev s->last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME); 882893f7ebaSPaolo Bonzini } 883893f7ebaSPaolo Bonzini 884893f7ebaSPaolo Bonzini immediate_exit: 885bd48bde8SPaolo Bonzini if (s->in_flight > 0) { 886bd48bde8SPaolo Bonzini /* We get here only if something went wrong. Either the job failed, 887bd48bde8SPaolo Bonzini * or it was cancelled prematurely so that we do not guarantee that 888bd48bde8SPaolo Bonzini * the target is a copy of the source. 889bd48bde8SPaolo Bonzini */ 890bd48bde8SPaolo Bonzini assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common))); 8919a0cec66SPaolo Bonzini assert(need_drain); 892bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 893bd48bde8SPaolo Bonzini } 894bd48bde8SPaolo Bonzini 895bd48bde8SPaolo Bonzini assert(s->in_flight == 0); 8967191bf31SMarkus Armbruster qemu_vfree(s->buf); 897b812f671SPaolo Bonzini g_free(s->cow_bitmap); 898402a4741SPaolo Bonzini g_free(s->in_flight_bitmap); 899dc162c8eSFam Zheng bdrv_dirty_iter_free(s->dbi); 9005a7e7a0bSStefan Hajnoczi 9015a7e7a0bSStefan Hajnoczi data = g_malloc(sizeof(*data)); 9025a7e7a0bSStefan Hajnoczi data->ret = ret; 9039a0cec66SPaolo Bonzini 9049a0cec66SPaolo Bonzini if (need_drain) { 905e253f4b8SKevin Wolf bdrv_drained_begin(bs); 9069a0cec66SPaolo Bonzini } 9075a7e7a0bSStefan Hajnoczi block_job_defer_to_main_loop(&s->common, mirror_exit, data); 908893f7ebaSPaolo Bonzini } 909893f7ebaSPaolo Bonzini 910893f7ebaSPaolo Bonzini static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp) 911893f7ebaSPaolo Bonzini { 912893f7ebaSPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 913893f7ebaSPaolo Bonzini 914893f7ebaSPaolo Bonzini if (speed < 0) { 915c6bd8c70SMarkus Armbruster error_setg(errp, QERR_INVALID_PARAMETER, "speed"); 916893f7ebaSPaolo Bonzini return; 917893f7ebaSPaolo Bonzini } 918f3e4ce4aSEric Blake ratelimit_set_speed(&s->limit, speed, SLICE_TIME); 919893f7ebaSPaolo Bonzini } 920893f7ebaSPaolo Bonzini 921d63ffd87SPaolo Bonzini static void mirror_complete(BlockJob *job, Error **errp) 922d63ffd87SPaolo Bonzini { 923d63ffd87SPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 9244ef85a9cSKevin Wolf BlockDriverState *target; 925d63ffd87SPaolo Bonzini 926274fcceeSMax Reitz target = blk_bs(s->target); 927274fcceeSMax Reitz 928d63ffd87SPaolo Bonzini if (!s->synced) { 9299df229c3SAlberto Garcia error_setg(errp, "The active block job '%s' cannot be completed", 9309df229c3SAlberto Garcia job->id); 931d63ffd87SPaolo Bonzini return; 932d63ffd87SPaolo Bonzini } 933d63ffd87SPaolo Bonzini 934274fcceeSMax Reitz if (s->backing_mode == MIRROR_OPEN_BACKING_CHAIN) { 935274fcceeSMax Reitz int ret; 936274fcceeSMax Reitz 937274fcceeSMax Reitz assert(!target->backing); 938274fcceeSMax Reitz ret = bdrv_open_backing_file(target, NULL, "backing", errp); 939274fcceeSMax Reitz if (ret < 0) { 940274fcceeSMax Reitz return; 941274fcceeSMax Reitz } 942274fcceeSMax Reitz } 943274fcceeSMax Reitz 94415d67298SChanglong Xie /* block all operations on to_replace bs */ 94509158f00SBenoît Canet if (s->replaces) { 9465a7e7a0bSStefan Hajnoczi AioContext *replace_aio_context; 9475a7e7a0bSStefan Hajnoczi 948e12f3784SWen Congyang s->to_replace = bdrv_find_node(s->replaces); 94909158f00SBenoît Canet if (!s->to_replace) { 950e12f3784SWen Congyang error_setg(errp, "Node name '%s' not found", s->replaces); 95109158f00SBenoît Canet return; 95209158f00SBenoît Canet } 95309158f00SBenoît Canet 9545a7e7a0bSStefan Hajnoczi replace_aio_context = bdrv_get_aio_context(s->to_replace); 9555a7e7a0bSStefan Hajnoczi aio_context_acquire(replace_aio_context); 9565a7e7a0bSStefan Hajnoczi 9574ef85a9cSKevin Wolf /* TODO Translate this into permission system. Current definition of 9584ef85a9cSKevin Wolf * GRAPH_MOD would require to request it for the parents; they might 9594ef85a9cSKevin Wolf * not even be BlockDriverStates, however, so a BdrvChild can't address 9604ef85a9cSKevin Wolf * them. May need redefinition of GRAPH_MOD. */ 96109158f00SBenoît Canet error_setg(&s->replace_blocker, 96209158f00SBenoît Canet "block device is in use by block-job-complete"); 96309158f00SBenoît Canet bdrv_op_block_all(s->to_replace, s->replace_blocker); 96409158f00SBenoît Canet bdrv_ref(s->to_replace); 9655a7e7a0bSStefan Hajnoczi 9665a7e7a0bSStefan Hajnoczi aio_context_release(replace_aio_context); 96709158f00SBenoît Canet } 96809158f00SBenoît Canet 969d63ffd87SPaolo Bonzini s->should_complete = true; 970751ebd76SFam Zheng block_job_enter(&s->common); 971d63ffd87SPaolo Bonzini } 972d63ffd87SPaolo Bonzini 973bae8196dSPaolo Bonzini static void mirror_pause(BlockJob *job) 974565ac01fSStefan Hajnoczi { 975565ac01fSStefan Hajnoczi MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 976565ac01fSStefan Hajnoczi 977bae8196dSPaolo Bonzini mirror_wait_for_all_io(s); 978565ac01fSStefan Hajnoczi } 979565ac01fSStefan Hajnoczi 980565ac01fSStefan Hajnoczi static void mirror_attached_aio_context(BlockJob *job, AioContext *new_context) 981565ac01fSStefan Hajnoczi { 982565ac01fSStefan Hajnoczi MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 983565ac01fSStefan Hajnoczi 984565ac01fSStefan Hajnoczi blk_set_aio_context(s->target, new_context); 985565ac01fSStefan Hajnoczi } 986565ac01fSStefan Hajnoczi 987bae8196dSPaolo Bonzini static void mirror_drain(BlockJob *job) 988bae8196dSPaolo Bonzini { 989bae8196dSPaolo Bonzini MirrorBlockJob *s = container_of(job, MirrorBlockJob, common); 990bae8196dSPaolo Bonzini 991bae8196dSPaolo Bonzini /* Need to keep a reference in case blk_drain triggers execution 992bae8196dSPaolo Bonzini * of mirror_complete... 993bae8196dSPaolo Bonzini */ 994bae8196dSPaolo Bonzini if (s->target) { 995bae8196dSPaolo Bonzini BlockBackend *target = s->target; 996bae8196dSPaolo Bonzini blk_ref(target); 997bae8196dSPaolo Bonzini blk_drain(target); 998bae8196dSPaolo Bonzini blk_unref(target); 999bae8196dSPaolo Bonzini } 1000bae8196dSPaolo Bonzini } 1001bae8196dSPaolo Bonzini 10023fc4b10aSFam Zheng static const BlockJobDriver mirror_job_driver = { 1003893f7ebaSPaolo Bonzini .instance_size = sizeof(MirrorBlockJob), 100479e14bf7SFam Zheng .job_type = BLOCK_JOB_TYPE_MIRROR, 1005893f7ebaSPaolo Bonzini .set_speed = mirror_set_speed, 1006a7815a76SJohn Snow .start = mirror_run, 1007d63ffd87SPaolo Bonzini .complete = mirror_complete, 1008565ac01fSStefan Hajnoczi .pause = mirror_pause, 1009565ac01fSStefan Hajnoczi .attached_aio_context = mirror_attached_aio_context, 1010bae8196dSPaolo Bonzini .drain = mirror_drain, 1011893f7ebaSPaolo Bonzini }; 1012893f7ebaSPaolo Bonzini 101303544a6eSFam Zheng static const BlockJobDriver commit_active_job_driver = { 101403544a6eSFam Zheng .instance_size = sizeof(MirrorBlockJob), 101503544a6eSFam Zheng .job_type = BLOCK_JOB_TYPE_COMMIT, 101603544a6eSFam Zheng .set_speed = mirror_set_speed, 1017a7815a76SJohn Snow .start = mirror_run, 101803544a6eSFam Zheng .complete = mirror_complete, 1019565ac01fSStefan Hajnoczi .pause = mirror_pause, 1020565ac01fSStefan Hajnoczi .attached_aio_context = mirror_attached_aio_context, 1021bae8196dSPaolo Bonzini .drain = mirror_drain, 102203544a6eSFam Zheng }; 102303544a6eSFam Zheng 10244ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_preadv(BlockDriverState *bs, 10254ef85a9cSKevin Wolf uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) 10264ef85a9cSKevin Wolf { 10274ef85a9cSKevin Wolf return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags); 10284ef85a9cSKevin Wolf } 10294ef85a9cSKevin Wolf 10304ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pwritev(BlockDriverState *bs, 10314ef85a9cSKevin Wolf uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags) 10324ef85a9cSKevin Wolf { 10334ef85a9cSKevin Wolf return bdrv_co_pwritev(bs->backing, offset, bytes, qiov, flags); 10344ef85a9cSKevin Wolf } 10354ef85a9cSKevin Wolf 10364ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_flush(BlockDriverState *bs) 10374ef85a9cSKevin Wolf { 1038ce960aa9SVladimir Sementsov-Ogievskiy if (bs->backing == NULL) { 1039ce960aa9SVladimir Sementsov-Ogievskiy /* we can be here after failed bdrv_append in mirror_start_job */ 1040ce960aa9SVladimir Sementsov-Ogievskiy return 0; 1041ce960aa9SVladimir Sementsov-Ogievskiy } 10424ef85a9cSKevin Wolf return bdrv_co_flush(bs->backing->bs); 10434ef85a9cSKevin Wolf } 10444ef85a9cSKevin Wolf 10454ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs, 1046f5a5ca79SManos Pitsidianakis int64_t offset, int bytes, BdrvRequestFlags flags) 10474ef85a9cSKevin Wolf { 1048f5a5ca79SManos Pitsidianakis return bdrv_co_pwrite_zeroes(bs->backing, offset, bytes, flags); 10494ef85a9cSKevin Wolf } 10504ef85a9cSKevin Wolf 10514ef85a9cSKevin Wolf static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs, 1052f5a5ca79SManos Pitsidianakis int64_t offset, int bytes) 10534ef85a9cSKevin Wolf { 1054f5a5ca79SManos Pitsidianakis return bdrv_co_pdiscard(bs->backing->bs, offset, bytes); 10554ef85a9cSKevin Wolf } 10564ef85a9cSKevin Wolf 1057fd4a6493SKevin Wolf static void bdrv_mirror_top_refresh_filename(BlockDriverState *bs, QDict *opts) 1058fd4a6493SKevin Wolf { 105918775ff3SVladimir Sementsov-Ogievskiy if (bs->backing == NULL) { 106018775ff3SVladimir Sementsov-Ogievskiy /* we can be here after failed bdrv_attach_child in 106118775ff3SVladimir Sementsov-Ogievskiy * bdrv_set_backing_hd */ 106218775ff3SVladimir Sementsov-Ogievskiy return; 106318775ff3SVladimir Sementsov-Ogievskiy } 1064fd4a6493SKevin Wolf bdrv_refresh_filename(bs->backing->bs); 1065fd4a6493SKevin Wolf pstrcpy(bs->exact_filename, sizeof(bs->exact_filename), 1066fd4a6493SKevin Wolf bs->backing->bs->filename); 1067fd4a6493SKevin Wolf } 1068fd4a6493SKevin Wolf 10694ef85a9cSKevin Wolf static void bdrv_mirror_top_close(BlockDriverState *bs) 10704ef85a9cSKevin Wolf { 10714ef85a9cSKevin Wolf } 10724ef85a9cSKevin Wolf 10734ef85a9cSKevin Wolf static void bdrv_mirror_top_child_perm(BlockDriverState *bs, BdrvChild *c, 10744ef85a9cSKevin Wolf const BdrvChildRole *role, 1075e0995dc3SKevin Wolf BlockReopenQueue *reopen_queue, 10764ef85a9cSKevin Wolf uint64_t perm, uint64_t shared, 10774ef85a9cSKevin Wolf uint64_t *nperm, uint64_t *nshared) 10784ef85a9cSKevin Wolf { 10794ef85a9cSKevin Wolf /* Must be able to forward guest writes to the real image */ 10804ef85a9cSKevin Wolf *nperm = 0; 10814ef85a9cSKevin Wolf if (perm & BLK_PERM_WRITE) { 10824ef85a9cSKevin Wolf *nperm |= BLK_PERM_WRITE; 10834ef85a9cSKevin Wolf } 10844ef85a9cSKevin Wolf 10854ef85a9cSKevin Wolf *nshared = BLK_PERM_ALL; 10864ef85a9cSKevin Wolf } 10874ef85a9cSKevin Wolf 10884ef85a9cSKevin Wolf /* Dummy node that provides consistent read to its users without requiring it 10894ef85a9cSKevin Wolf * from its backing file and that allows writes on the backing file chain. */ 10904ef85a9cSKevin Wolf static BlockDriver bdrv_mirror_top = { 10914ef85a9cSKevin Wolf .format_name = "mirror_top", 10924ef85a9cSKevin Wolf .bdrv_co_preadv = bdrv_mirror_top_preadv, 10934ef85a9cSKevin Wolf .bdrv_co_pwritev = bdrv_mirror_top_pwritev, 10944ef85a9cSKevin Wolf .bdrv_co_pwrite_zeroes = bdrv_mirror_top_pwrite_zeroes, 10954ef85a9cSKevin Wolf .bdrv_co_pdiscard = bdrv_mirror_top_pdiscard, 10964ef85a9cSKevin Wolf .bdrv_co_flush = bdrv_mirror_top_flush, 1097f7cc69b3SManos Pitsidianakis .bdrv_co_get_block_status = bdrv_co_get_block_status_from_backing, 1098fd4a6493SKevin Wolf .bdrv_refresh_filename = bdrv_mirror_top_refresh_filename, 10994ef85a9cSKevin Wolf .bdrv_close = bdrv_mirror_top_close, 11004ef85a9cSKevin Wolf .bdrv_child_perm = bdrv_mirror_top_child_perm, 11014ef85a9cSKevin Wolf }; 11024ef85a9cSKevin Wolf 110371aa9867SAlberto Garcia static void mirror_start_job(const char *job_id, BlockDriverState *bs, 110447970dfbSJohn Snow int creation_flags, BlockDriverState *target, 110547970dfbSJohn Snow const char *replaces, int64_t speed, 110647970dfbSJohn Snow uint32_t granularity, int64_t buf_size, 1107274fcceeSMax Reitz BlockMirrorBackingMode backing_mode, 110803544a6eSFam Zheng BlockdevOnError on_source_error, 1109b952b558SPaolo Bonzini BlockdevOnError on_target_error, 11100fc9f8eaSFam Zheng bool unmap, 1111097310b5SMarkus Armbruster BlockCompletionFunc *cb, 111251ccfa2dSFam Zheng void *opaque, 111303544a6eSFam Zheng const BlockJobDriver *driver, 1114b49f7eadSWen Congyang bool is_none_mode, BlockDriverState *base, 111551ccfa2dSFam Zheng bool auto_complete, const char *filter_node_name, 1116045a2f82SFam Zheng bool is_mirror, 111751ccfa2dSFam Zheng Error **errp) 1118893f7ebaSPaolo Bonzini { 1119893f7ebaSPaolo Bonzini MirrorBlockJob *s; 11204ef85a9cSKevin Wolf BlockDriverState *mirror_top_bs; 11214ef85a9cSKevin Wolf bool target_graph_mod; 11224ef85a9cSKevin Wolf bool target_is_backing; 1123b2c2832cSKevin Wolf Error *local_err = NULL; 1124d7086422SKevin Wolf int ret; 1125893f7ebaSPaolo Bonzini 1126eee13dfeSPaolo Bonzini if (granularity == 0) { 1127341ebc2fSJohn Snow granularity = bdrv_get_default_bitmap_granularity(target); 1128eee13dfeSPaolo Bonzini } 1129eee13dfeSPaolo Bonzini 113031826642SEric Blake assert(is_power_of_2(granularity)); 1131eee13dfeSPaolo Bonzini 113248ac0a4dSWen Congyang if (buf_size < 0) { 113348ac0a4dSWen Congyang error_setg(errp, "Invalid parameter 'buf-size'"); 113448ac0a4dSWen Congyang return; 113548ac0a4dSWen Congyang } 113648ac0a4dSWen Congyang 113748ac0a4dSWen Congyang if (buf_size == 0) { 113848ac0a4dSWen Congyang buf_size = DEFAULT_MIRROR_BUF_SIZE; 113948ac0a4dSWen Congyang } 11405bc361b8SFam Zheng 11414ef85a9cSKevin Wolf /* In the case of active commit, add dummy driver to provide consistent 11424ef85a9cSKevin Wolf * reads on the top, while disabling it in the intermediate nodes, and make 11434ef85a9cSKevin Wolf * the backing chain writable. */ 11446cdbceb1SKevin Wolf mirror_top_bs = bdrv_new_open_driver(&bdrv_mirror_top, filter_node_name, 11456cdbceb1SKevin Wolf BDRV_O_RDWR, errp); 11464ef85a9cSKevin Wolf if (mirror_top_bs == NULL) { 1147893f7ebaSPaolo Bonzini return; 1148893f7ebaSPaolo Bonzini } 1149d3c8c674SKevin Wolf if (!filter_node_name) { 1150d3c8c674SKevin Wolf mirror_top_bs->implicit = true; 1151d3c8c674SKevin Wolf } 11524ef85a9cSKevin Wolf mirror_top_bs->total_sectors = bs->total_sectors; 115319dd29e8SFam Zheng bdrv_set_aio_context(mirror_top_bs, bdrv_get_aio_context(bs)); 1154893f7ebaSPaolo Bonzini 11554ef85a9cSKevin Wolf /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep 11567a25fcd0SMax Reitz * it alive until block_job_create() succeeds even if bs has no parent. */ 11574ef85a9cSKevin Wolf bdrv_ref(mirror_top_bs); 11584ef85a9cSKevin Wolf bdrv_drained_begin(bs); 1159b2c2832cSKevin Wolf bdrv_append(mirror_top_bs, bs, &local_err); 11604ef85a9cSKevin Wolf bdrv_drained_end(bs); 11614ef85a9cSKevin Wolf 1162b2c2832cSKevin Wolf if (local_err) { 1163b2c2832cSKevin Wolf bdrv_unref(mirror_top_bs); 1164b2c2832cSKevin Wolf error_propagate(errp, local_err); 1165b2c2832cSKevin Wolf return; 1166b2c2832cSKevin Wolf } 1167b2c2832cSKevin Wolf 11684ef85a9cSKevin Wolf /* Make sure that the source is not resized while the job is running */ 11694ef85a9cSKevin Wolf s = block_job_create(job_id, driver, mirror_top_bs, 11704ef85a9cSKevin Wolf BLK_PERM_CONSISTENT_READ, 11714ef85a9cSKevin Wolf BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED | 11724ef85a9cSKevin Wolf BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD, speed, 11734ef85a9cSKevin Wolf creation_flags, cb, opaque, errp); 11744ef85a9cSKevin Wolf if (!s) { 11754ef85a9cSKevin Wolf goto fail; 11764ef85a9cSKevin Wolf } 11777a25fcd0SMax Reitz /* The block job now has a reference to this node */ 11787a25fcd0SMax Reitz bdrv_unref(mirror_top_bs); 11797a25fcd0SMax Reitz 11804ef85a9cSKevin Wolf s->source = bs; 11814ef85a9cSKevin Wolf s->mirror_top_bs = mirror_top_bs; 11824ef85a9cSKevin Wolf 11834ef85a9cSKevin Wolf /* No resize for the target either; while the mirror is still running, a 11844ef85a9cSKevin Wolf * consistent read isn't necessarily possible. We could possibly allow 11854ef85a9cSKevin Wolf * writes and graph modifications, though it would likely defeat the 11864ef85a9cSKevin Wolf * purpose of a mirror, so leave them blocked for now. 11874ef85a9cSKevin Wolf * 11884ef85a9cSKevin Wolf * In the case of active commit, things look a bit different, though, 11894ef85a9cSKevin Wolf * because the target is an already populated backing file in active use. 11904ef85a9cSKevin Wolf * We can allow anything except resize there.*/ 11914ef85a9cSKevin Wolf target_is_backing = bdrv_chain_contains(bs, target); 11924ef85a9cSKevin Wolf target_graph_mod = (backing_mode != MIRROR_LEAVE_BACKING_CHAIN); 11934ef85a9cSKevin Wolf s->target = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE | 11944ef85a9cSKevin Wolf (target_graph_mod ? BLK_PERM_GRAPH_MOD : 0), 11954ef85a9cSKevin Wolf BLK_PERM_WRITE_UNCHANGED | 11964ef85a9cSKevin Wolf (target_is_backing ? BLK_PERM_CONSISTENT_READ | 11974ef85a9cSKevin Wolf BLK_PERM_WRITE | 11984ef85a9cSKevin Wolf BLK_PERM_GRAPH_MOD : 0)); 1199d7086422SKevin Wolf ret = blk_insert_bs(s->target, target, errp); 1200d7086422SKevin Wolf if (ret < 0) { 12014ef85a9cSKevin Wolf goto fail; 1202d7086422SKevin Wolf } 1203045a2f82SFam Zheng if (is_mirror) { 1204045a2f82SFam Zheng /* XXX: Mirror target could be a NBD server of target QEMU in the case 1205045a2f82SFam Zheng * of non-shared block migration. To allow migration completion, we 1206045a2f82SFam Zheng * have to allow "inactivate" of the target BB. When that happens, we 1207045a2f82SFam Zheng * know the job is drained, and the vcpus are stopped, so no write 1208045a2f82SFam Zheng * operation will be performed. Block layer already has assertions to 1209045a2f82SFam Zheng * ensure that. */ 1210045a2f82SFam Zheng blk_set_force_allow_inactivate(s->target); 1211045a2f82SFam Zheng } 1212e253f4b8SKevin Wolf 121309158f00SBenoît Canet s->replaces = g_strdup(replaces); 1214b952b558SPaolo Bonzini s->on_source_error = on_source_error; 1215b952b558SPaolo Bonzini s->on_target_error = on_target_error; 121603544a6eSFam Zheng s->is_none_mode = is_none_mode; 1217274fcceeSMax Reitz s->backing_mode = backing_mode; 12185bc361b8SFam Zheng s->base = base; 1219eee13dfeSPaolo Bonzini s->granularity = granularity; 122048ac0a4dSWen Congyang s->buf_size = ROUND_UP(buf_size, granularity); 12210fc9f8eaSFam Zheng s->unmap = unmap; 1222b49f7eadSWen Congyang if (auto_complete) { 1223b49f7eadSWen Congyang s->should_complete = true; 1224b49f7eadSWen Congyang } 1225b812f671SPaolo Bonzini 12260db6e54aSFam Zheng s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity, NULL, errp); 1227b8afb520SFam Zheng if (!s->dirty_bitmap) { 122888f9d1b3SKevin Wolf goto fail; 1229b8afb520SFam Zheng } 123010f3cd15SAlberto Garcia 12314ef85a9cSKevin Wolf /* Required permissions are already taken with blk_new() */ 123276d554e2SKevin Wolf block_job_add_bdrv(&s->common, "target", target, 0, BLK_PERM_ALL, 123376d554e2SKevin Wolf &error_abort); 123476d554e2SKevin Wolf 1235f3ede4b0SAlberto Garcia /* In commit_active_start() all intermediate nodes disappear, so 1236f3ede4b0SAlberto Garcia * any jobs in them must be blocked */ 12374ef85a9cSKevin Wolf if (target_is_backing) { 1238f3ede4b0SAlberto Garcia BlockDriverState *iter; 1239f3ede4b0SAlberto Garcia for (iter = backing_bs(bs); iter != target; iter = backing_bs(iter)) { 12404ef85a9cSKevin Wolf /* XXX BLK_PERM_WRITE needs to be allowed so we don't block 12414ef85a9cSKevin Wolf * ourselves at s->base (if writes are blocked for a node, they are 12424ef85a9cSKevin Wolf * also blocked for its backing file). The other options would be a 12434ef85a9cSKevin Wolf * second filter driver above s->base (== target). */ 12444ef85a9cSKevin Wolf ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0, 12454ef85a9cSKevin Wolf BLK_PERM_WRITE_UNCHANGED | BLK_PERM_WRITE, 12464ef85a9cSKevin Wolf errp); 12474ef85a9cSKevin Wolf if (ret < 0) { 12484ef85a9cSKevin Wolf goto fail; 12494ef85a9cSKevin Wolf } 1250f3ede4b0SAlberto Garcia } 1251f3ede4b0SAlberto Garcia } 125210f3cd15SAlberto Garcia 12535ccac6f1SJohn Snow trace_mirror_start(bs, s, opaque); 12545ccac6f1SJohn Snow block_job_start(&s->common); 12554ef85a9cSKevin Wolf return; 12564ef85a9cSKevin Wolf 12574ef85a9cSKevin Wolf fail: 12584ef85a9cSKevin Wolf if (s) { 12597a25fcd0SMax Reitz /* Make sure this BDS does not go away until we have completed the graph 12607a25fcd0SMax Reitz * changes below */ 12617a25fcd0SMax Reitz bdrv_ref(mirror_top_bs); 12627a25fcd0SMax Reitz 12634ef85a9cSKevin Wolf g_free(s->replaces); 12644ef85a9cSKevin Wolf blk_unref(s->target); 126505b0d8e3SPaolo Bonzini block_job_early_fail(&s->common); 12664ef85a9cSKevin Wolf } 12674ef85a9cSKevin Wolf 1268c1cef672SFam Zheng bdrv_child_try_set_perm(mirror_top_bs->backing, 0, BLK_PERM_ALL, 1269c1cef672SFam Zheng &error_abort); 12705fe31c25SKevin Wolf bdrv_replace_node(mirror_top_bs, backing_bs(mirror_top_bs), &error_abort); 12717a25fcd0SMax Reitz 12727a25fcd0SMax Reitz bdrv_unref(mirror_top_bs); 1273893f7ebaSPaolo Bonzini } 127403544a6eSFam Zheng 127571aa9867SAlberto Garcia void mirror_start(const char *job_id, BlockDriverState *bs, 127671aa9867SAlberto Garcia BlockDriverState *target, const char *replaces, 12775fba6c0eSJohn Snow int64_t speed, uint32_t granularity, int64_t buf_size, 1278274fcceeSMax Reitz MirrorSyncMode mode, BlockMirrorBackingMode backing_mode, 1279274fcceeSMax Reitz BlockdevOnError on_source_error, 128003544a6eSFam Zheng BlockdevOnError on_target_error, 12816cdbceb1SKevin Wolf bool unmap, const char *filter_node_name, Error **errp) 128203544a6eSFam Zheng { 128303544a6eSFam Zheng bool is_none_mode; 128403544a6eSFam Zheng BlockDriverState *base; 128503544a6eSFam Zheng 12864b80ab2bSJohn Snow if (mode == MIRROR_SYNC_MODE_INCREMENTAL) { 12874b80ab2bSJohn Snow error_setg(errp, "Sync mode 'incremental' not supported"); 1288d58d8453SJohn Snow return; 1289d58d8453SJohn Snow } 129003544a6eSFam Zheng is_none_mode = mode == MIRROR_SYNC_MODE_NONE; 1291760e0063SKevin Wolf base = mode == MIRROR_SYNC_MODE_TOP ? backing_bs(bs) : NULL; 129247970dfbSJohn Snow mirror_start_job(job_id, bs, BLOCK_JOB_DEFAULT, target, replaces, 1293274fcceeSMax Reitz speed, granularity, buf_size, backing_mode, 129451ccfa2dSFam Zheng on_source_error, on_target_error, unmap, NULL, NULL, 12956cdbceb1SKevin Wolf &mirror_job_driver, is_none_mode, base, false, 1296045a2f82SFam Zheng filter_node_name, true, errp); 129703544a6eSFam Zheng } 129803544a6eSFam Zheng 1299fd62c609SAlberto Garcia void commit_active_start(const char *job_id, BlockDriverState *bs, 130047970dfbSJohn Snow BlockDriverState *base, int creation_flags, 130147970dfbSJohn Snow int64_t speed, BlockdevOnError on_error, 13020db832f4SKevin Wolf const char *filter_node_name, 130378bbd910SFam Zheng BlockCompletionFunc *cb, void *opaque, 130478bbd910SFam Zheng bool auto_complete, Error **errp) 130503544a6eSFam Zheng { 13064da83585SJeff Cody int orig_base_flags; 1307cc67f4d1SJeff Cody Error *local_err = NULL; 13084da83585SJeff Cody 13094da83585SJeff Cody orig_base_flags = bdrv_get_flags(base); 13104da83585SJeff Cody 131120a63d2cSFam Zheng if (bdrv_reopen(base, bs->open_flags, errp)) { 131220a63d2cSFam Zheng return; 131320a63d2cSFam Zheng } 13144da83585SJeff Cody 131547970dfbSJohn Snow mirror_start_job(job_id, bs, creation_flags, base, NULL, speed, 0, 0, 131671aa9867SAlberto Garcia MIRROR_LEAVE_BACKING_CHAIN, 131751ccfa2dSFam Zheng on_error, on_error, true, cb, opaque, 13186cdbceb1SKevin Wolf &commit_active_job_driver, false, base, auto_complete, 1319045a2f82SFam Zheng filter_node_name, false, &local_err); 13200fb6395cSMarkus Armbruster if (local_err) { 1321cc67f4d1SJeff Cody error_propagate(errp, local_err); 13224da83585SJeff Cody goto error_restore_flags; 13234da83585SJeff Cody } 13244da83585SJeff Cody 13254da83585SJeff Cody return; 13264da83585SJeff Cody 13274da83585SJeff Cody error_restore_flags: 13284da83585SJeff Cody /* ignore error and errp for bdrv_reopen, because we want to propagate 13294da83585SJeff Cody * the original error */ 13304da83585SJeff Cody bdrv_reopen(base, orig_base_flags, NULL); 13314da83585SJeff Cody return; 133203544a6eSFam Zheng } 1333